Commit | Line | Data |
---|---|---|
ccd979bd MF |
1 | /* -*- mode: c; c-basic-offset: 8; -*- |
2 | * vim: noexpandtab sw=8 ts=8 sts=0: | |
3 | * | |
4 | * alloc.c | |
5 | * | |
6 | * Extent allocs and frees | |
7 | * | |
8 | * Copyright (C) 2002, 2004 Oracle. All rights reserved. | |
9 | * | |
10 | * This program is free software; you can redistribute it and/or | |
11 | * modify it under the terms of the GNU General Public | |
12 | * License as published by the Free Software Foundation; either | |
13 | * version 2 of the License, or (at your option) any later version. | |
14 | * | |
15 | * This program is distributed in the hope that it will be useful, | |
16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
18 | * General Public License for more details. | |
19 | * | |
20 | * You should have received a copy of the GNU General Public | |
21 | * License along with this program; if not, write to the | |
22 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
23 | * Boston, MA 021110-1307, USA. | |
24 | */ | |
25 | ||
26 | #include <linux/fs.h> | |
27 | #include <linux/types.h> | |
28 | #include <linux/slab.h> | |
29 | #include <linux/highmem.h> | |
60b11392 | 30 | #include <linux/swap.h> |
a90714c1 | 31 | #include <linux/quotaops.h> |
ccd979bd MF |
32 | |
33 | #define MLOG_MASK_PREFIX ML_DISK_ALLOC | |
34 | #include <cluster/masklog.h> | |
35 | ||
36 | #include "ocfs2.h" | |
37 | ||
38 | #include "alloc.h" | |
60b11392 | 39 | #include "aops.h" |
d6b32bbb | 40 | #include "blockcheck.h" |
ccd979bd MF |
41 | #include "dlmglue.h" |
42 | #include "extent_map.h" | |
43 | #include "inode.h" | |
44 | #include "journal.h" | |
45 | #include "localalloc.h" | |
46 | #include "suballoc.h" | |
47 | #include "sysfile.h" | |
48 | #include "file.h" | |
49 | #include "super.h" | |
50 | #include "uptodate.h" | |
2a50a743 | 51 | #include "xattr.h" |
ccd979bd MF |
52 | |
53 | #include "buffer_head_io.h" | |
54 | ||
e7d4cb6b | 55 | |
1625f8ac JB |
56 | /* |
57 | * Operations for a specific extent tree type. | |
58 | * | |
59 | * To implement an on-disk btree (extent tree) type in ocfs2, add | |
60 | * an ocfs2_extent_tree_operations structure and the matching | |
8d6220d6 | 61 | * ocfs2_init_<thingy>_extent_tree() function. That's pretty much it |
1625f8ac JB |
62 | * for the allocation portion of the extent tree. |
63 | */ | |
e7d4cb6b | 64 | struct ocfs2_extent_tree_operations { |
1625f8ac JB |
65 | /* |
66 | * last_eb_blk is the block number of the right most leaf extent | |
67 | * block. Most on-disk structures containing an extent tree store | |
68 | * this value for fast access. The ->eo_set_last_eb_blk() and | |
69 | * ->eo_get_last_eb_blk() operations access this value. They are | |
70 | * both required. | |
71 | */ | |
35dc0aa3 JB |
72 | void (*eo_set_last_eb_blk)(struct ocfs2_extent_tree *et, |
73 | u64 blkno); | |
74 | u64 (*eo_get_last_eb_blk)(struct ocfs2_extent_tree *et); | |
1625f8ac JB |
75 | |
76 | /* | |
77 | * The on-disk structure usually keeps track of how many total | |
78 | * clusters are stored in this extent tree. This function updates | |
79 | * that value. new_clusters is the delta, and must be | |
80 | * added to the total. Required. | |
81 | */ | |
6136ca5f | 82 | void (*eo_update_clusters)(struct ocfs2_extent_tree *et, |
35dc0aa3 | 83 | u32 new_clusters); |
1625f8ac JB |
84 | |
85 | /* | |
86 | * If ->eo_insert_check() exists, it is called before rec is | |
87 | * inserted into the extent tree. It is optional. | |
88 | */ | |
6136ca5f | 89 | int (*eo_insert_check)(struct ocfs2_extent_tree *et, |
1e61ee79 | 90 | struct ocfs2_extent_rec *rec); |
6136ca5f | 91 | int (*eo_sanity_check)(struct ocfs2_extent_tree *et); |
0ce1010f | 92 | |
1625f8ac JB |
93 | /* |
94 | * -------------------------------------------------------------- | |
95 | * The remaining are internal to ocfs2_extent_tree and don't have | |
96 | * accessor functions | |
97 | */ | |
98 | ||
99 | /* | |
100 | * ->eo_fill_root_el() takes et->et_object and sets et->et_root_el. | |
101 | * It is required. | |
102 | */ | |
0ce1010f | 103 | void (*eo_fill_root_el)(struct ocfs2_extent_tree *et); |
1625f8ac JB |
104 | |
105 | /* | |
106 | * ->eo_fill_max_leaf_clusters sets et->et_max_leaf_clusters if | |
107 | * it exists. If it does not, et->et_max_leaf_clusters is set | |
108 | * to 0 (unlimited). Optional. | |
109 | */ | |
6136ca5f | 110 | void (*eo_fill_max_leaf_clusters)(struct ocfs2_extent_tree *et); |
e7d4cb6b TM |
111 | }; |
112 | ||
e7d4cb6b | 113 | |
f99b9b7c JB |
114 | /* |
115 | * Pre-declare ocfs2_dinode_et_ops so we can use it as a sanity check | |
116 | * in the methods. | |
117 | */ | |
118 | static u64 ocfs2_dinode_get_last_eb_blk(struct ocfs2_extent_tree *et); | |
119 | static void ocfs2_dinode_set_last_eb_blk(struct ocfs2_extent_tree *et, | |
120 | u64 blkno); | |
6136ca5f | 121 | static void ocfs2_dinode_update_clusters(struct ocfs2_extent_tree *et, |
f99b9b7c | 122 | u32 clusters); |
6136ca5f | 123 | static int ocfs2_dinode_insert_check(struct ocfs2_extent_tree *et, |
f99b9b7c | 124 | struct ocfs2_extent_rec *rec); |
6136ca5f | 125 | static int ocfs2_dinode_sanity_check(struct ocfs2_extent_tree *et); |
f99b9b7c JB |
126 | static void ocfs2_dinode_fill_root_el(struct ocfs2_extent_tree *et); |
127 | static struct ocfs2_extent_tree_operations ocfs2_dinode_et_ops = { | |
128 | .eo_set_last_eb_blk = ocfs2_dinode_set_last_eb_blk, | |
129 | .eo_get_last_eb_blk = ocfs2_dinode_get_last_eb_blk, | |
130 | .eo_update_clusters = ocfs2_dinode_update_clusters, | |
131 | .eo_insert_check = ocfs2_dinode_insert_check, | |
132 | .eo_sanity_check = ocfs2_dinode_sanity_check, | |
133 | .eo_fill_root_el = ocfs2_dinode_fill_root_el, | |
134 | }; | |
0ce1010f | 135 | |
e7d4cb6b TM |
136 | static void ocfs2_dinode_set_last_eb_blk(struct ocfs2_extent_tree *et, |
137 | u64 blkno) | |
138 | { | |
ea5efa15 | 139 | struct ocfs2_dinode *di = et->et_object; |
e7d4cb6b | 140 | |
f99b9b7c | 141 | BUG_ON(et->et_ops != &ocfs2_dinode_et_ops); |
e7d4cb6b TM |
142 | di->i_last_eb_blk = cpu_to_le64(blkno); |
143 | } | |
144 | ||
145 | static u64 ocfs2_dinode_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
146 | { | |
ea5efa15 | 147 | struct ocfs2_dinode *di = et->et_object; |
e7d4cb6b | 148 | |
f99b9b7c | 149 | BUG_ON(et->et_ops != &ocfs2_dinode_et_ops); |
e7d4cb6b TM |
150 | return le64_to_cpu(di->i_last_eb_blk); |
151 | } | |
152 | ||
6136ca5f | 153 | static void ocfs2_dinode_update_clusters(struct ocfs2_extent_tree *et, |
e7d4cb6b TM |
154 | u32 clusters) |
155 | { | |
6136ca5f | 156 | struct ocfs2_inode_info *oi = cache_info_to_inode(et->et_ci); |
ea5efa15 | 157 | struct ocfs2_dinode *di = et->et_object; |
e7d4cb6b TM |
158 | |
159 | le32_add_cpu(&di->i_clusters, clusters); | |
6136ca5f JB |
160 | spin_lock(&oi->ip_lock); |
161 | oi->ip_clusters = le32_to_cpu(di->i_clusters); | |
162 | spin_unlock(&oi->ip_lock); | |
e7d4cb6b TM |
163 | } |
164 | ||
6136ca5f | 165 | static int ocfs2_dinode_insert_check(struct ocfs2_extent_tree *et, |
1e61ee79 JB |
166 | struct ocfs2_extent_rec *rec) |
167 | { | |
6136ca5f JB |
168 | struct ocfs2_inode_info *oi = cache_info_to_inode(et->et_ci); |
169 | struct ocfs2_super *osb = OCFS2_SB(oi->vfs_inode.i_sb); | |
1e61ee79 | 170 | |
6136ca5f | 171 | BUG_ON(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL); |
1e61ee79 | 172 | mlog_bug_on_msg(!ocfs2_sparse_alloc(osb) && |
6136ca5f | 173 | (oi->ip_clusters != le32_to_cpu(rec->e_cpos)), |
1e61ee79 JB |
174 | "Device %s, asking for sparse allocation: inode %llu, " |
175 | "cpos %u, clusters %u\n", | |
176 | osb->dev_str, | |
6136ca5f JB |
177 | (unsigned long long)oi->ip_blkno, |
178 | rec->e_cpos, oi->ip_clusters); | |
1e61ee79 JB |
179 | |
180 | return 0; | |
181 | } | |
182 | ||
6136ca5f | 183 | static int ocfs2_dinode_sanity_check(struct ocfs2_extent_tree *et) |
e7d4cb6b | 184 | { |
10995aa2 | 185 | struct ocfs2_dinode *di = et->et_object; |
e7d4cb6b | 186 | |
f99b9b7c | 187 | BUG_ON(et->et_ops != &ocfs2_dinode_et_ops); |
10995aa2 | 188 | BUG_ON(!OCFS2_IS_VALID_DINODE(di)); |
e7d4cb6b | 189 | |
10995aa2 | 190 | return 0; |
e7d4cb6b TM |
191 | } |
192 | ||
f99b9b7c JB |
193 | static void ocfs2_dinode_fill_root_el(struct ocfs2_extent_tree *et) |
194 | { | |
195 | struct ocfs2_dinode *di = et->et_object; | |
196 | ||
197 | et->et_root_el = &di->id2.i_list; | |
198 | } | |
199 | ||
e7d4cb6b | 200 | |
0ce1010f JB |
201 | static void ocfs2_xattr_value_fill_root_el(struct ocfs2_extent_tree *et) |
202 | { | |
2a50a743 | 203 | struct ocfs2_xattr_value_buf *vb = et->et_object; |
0ce1010f | 204 | |
2a50a743 | 205 | et->et_root_el = &vb->vb_xv->xr_list; |
0ce1010f JB |
206 | } |
207 | ||
f56654c4 TM |
208 | static void ocfs2_xattr_value_set_last_eb_blk(struct ocfs2_extent_tree *et, |
209 | u64 blkno) | |
210 | { | |
2a50a743 | 211 | struct ocfs2_xattr_value_buf *vb = et->et_object; |
f56654c4 | 212 | |
2a50a743 | 213 | vb->vb_xv->xr_last_eb_blk = cpu_to_le64(blkno); |
f56654c4 TM |
214 | } |
215 | ||
216 | static u64 ocfs2_xattr_value_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
217 | { | |
2a50a743 | 218 | struct ocfs2_xattr_value_buf *vb = et->et_object; |
f56654c4 | 219 | |
2a50a743 | 220 | return le64_to_cpu(vb->vb_xv->xr_last_eb_blk); |
f56654c4 TM |
221 | } |
222 | ||
6136ca5f | 223 | static void ocfs2_xattr_value_update_clusters(struct ocfs2_extent_tree *et, |
f56654c4 TM |
224 | u32 clusters) |
225 | { | |
2a50a743 | 226 | struct ocfs2_xattr_value_buf *vb = et->et_object; |
f56654c4 | 227 | |
2a50a743 | 228 | le32_add_cpu(&vb->vb_xv->xr_clusters, clusters); |
f56654c4 TM |
229 | } |
230 | ||
1a09f556 | 231 | static struct ocfs2_extent_tree_operations ocfs2_xattr_value_et_ops = { |
35dc0aa3 JB |
232 | .eo_set_last_eb_blk = ocfs2_xattr_value_set_last_eb_blk, |
233 | .eo_get_last_eb_blk = ocfs2_xattr_value_get_last_eb_blk, | |
234 | .eo_update_clusters = ocfs2_xattr_value_update_clusters, | |
0ce1010f | 235 | .eo_fill_root_el = ocfs2_xattr_value_fill_root_el, |
f56654c4 TM |
236 | }; |
237 | ||
0ce1010f JB |
238 | static void ocfs2_xattr_tree_fill_root_el(struct ocfs2_extent_tree *et) |
239 | { | |
240 | struct ocfs2_xattr_block *xb = et->et_object; | |
241 | ||
242 | et->et_root_el = &xb->xb_attrs.xb_root.xt_list; | |
243 | } | |
244 | ||
6136ca5f | 245 | static void ocfs2_xattr_tree_fill_max_leaf_clusters(struct ocfs2_extent_tree *et) |
943cced3 | 246 | { |
6136ca5f | 247 | struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci); |
943cced3 | 248 | et->et_max_leaf_clusters = |
6136ca5f | 249 | ocfs2_clusters_for_bytes(sb, OCFS2_MAX_XATTR_TREE_LEAF_SIZE); |
943cced3 JB |
250 | } |
251 | ||
ba492615 TM |
252 | static void ocfs2_xattr_tree_set_last_eb_blk(struct ocfs2_extent_tree *et, |
253 | u64 blkno) | |
254 | { | |
ea5efa15 | 255 | struct ocfs2_xattr_block *xb = et->et_object; |
ba492615 TM |
256 | struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root; |
257 | ||
258 | xt->xt_last_eb_blk = cpu_to_le64(blkno); | |
259 | } | |
260 | ||
261 | static u64 ocfs2_xattr_tree_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
262 | { | |
ea5efa15 | 263 | struct ocfs2_xattr_block *xb = et->et_object; |
ba492615 TM |
264 | struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root; |
265 | ||
266 | return le64_to_cpu(xt->xt_last_eb_blk); | |
267 | } | |
268 | ||
6136ca5f | 269 | static void ocfs2_xattr_tree_update_clusters(struct ocfs2_extent_tree *et, |
ba492615 TM |
270 | u32 clusters) |
271 | { | |
ea5efa15 | 272 | struct ocfs2_xattr_block *xb = et->et_object; |
ba492615 TM |
273 | |
274 | le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, clusters); | |
275 | } | |
276 | ||
ba492615 | 277 | static struct ocfs2_extent_tree_operations ocfs2_xattr_tree_et_ops = { |
35dc0aa3 JB |
278 | .eo_set_last_eb_blk = ocfs2_xattr_tree_set_last_eb_blk, |
279 | .eo_get_last_eb_blk = ocfs2_xattr_tree_get_last_eb_blk, | |
280 | .eo_update_clusters = ocfs2_xattr_tree_update_clusters, | |
0ce1010f | 281 | .eo_fill_root_el = ocfs2_xattr_tree_fill_root_el, |
943cced3 | 282 | .eo_fill_max_leaf_clusters = ocfs2_xattr_tree_fill_max_leaf_clusters, |
ba492615 TM |
283 | }; |
284 | ||
9b7895ef MF |
285 | static void ocfs2_dx_root_set_last_eb_blk(struct ocfs2_extent_tree *et, |
286 | u64 blkno) | |
287 | { | |
288 | struct ocfs2_dx_root_block *dx_root = et->et_object; | |
289 | ||
290 | dx_root->dr_last_eb_blk = cpu_to_le64(blkno); | |
291 | } | |
292 | ||
293 | static u64 ocfs2_dx_root_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
294 | { | |
295 | struct ocfs2_dx_root_block *dx_root = et->et_object; | |
296 | ||
297 | return le64_to_cpu(dx_root->dr_last_eb_blk); | |
298 | } | |
299 | ||
6136ca5f | 300 | static void ocfs2_dx_root_update_clusters(struct ocfs2_extent_tree *et, |
9b7895ef MF |
301 | u32 clusters) |
302 | { | |
303 | struct ocfs2_dx_root_block *dx_root = et->et_object; | |
304 | ||
305 | le32_add_cpu(&dx_root->dr_clusters, clusters); | |
306 | } | |
307 | ||
6136ca5f | 308 | static int ocfs2_dx_root_sanity_check(struct ocfs2_extent_tree *et) |
9b7895ef MF |
309 | { |
310 | struct ocfs2_dx_root_block *dx_root = et->et_object; | |
311 | ||
312 | BUG_ON(!OCFS2_IS_VALID_DX_ROOT(dx_root)); | |
313 | ||
314 | return 0; | |
315 | } | |
316 | ||
317 | static void ocfs2_dx_root_fill_root_el(struct ocfs2_extent_tree *et) | |
318 | { | |
319 | struct ocfs2_dx_root_block *dx_root = et->et_object; | |
320 | ||
321 | et->et_root_el = &dx_root->dr_list; | |
322 | } | |
323 | ||
324 | static struct ocfs2_extent_tree_operations ocfs2_dx_root_et_ops = { | |
325 | .eo_set_last_eb_blk = ocfs2_dx_root_set_last_eb_blk, | |
326 | .eo_get_last_eb_blk = ocfs2_dx_root_get_last_eb_blk, | |
327 | .eo_update_clusters = ocfs2_dx_root_update_clusters, | |
328 | .eo_sanity_check = ocfs2_dx_root_sanity_check, | |
329 | .eo_fill_root_el = ocfs2_dx_root_fill_root_el, | |
330 | }; | |
331 | ||
8d6220d6 JB |
332 | static void __ocfs2_init_extent_tree(struct ocfs2_extent_tree *et, |
333 | struct inode *inode, | |
334 | struct buffer_head *bh, | |
13723d00 | 335 | ocfs2_journal_access_func access, |
8d6220d6 JB |
336 | void *obj, |
337 | struct ocfs2_extent_tree_operations *ops) | |
e7d4cb6b | 338 | { |
1a09f556 | 339 | et->et_ops = ops; |
ce1d9ea6 | 340 | et->et_root_bh = bh; |
d9a0a1f8 | 341 | et->et_ci = INODE_CACHE(inode); |
13723d00 | 342 | et->et_root_journal_access = access; |
ea5efa15 JB |
343 | if (!obj) |
344 | obj = (void *)bh->b_data; | |
345 | et->et_object = obj; | |
e7d4cb6b | 346 | |
0ce1010f | 347 | et->et_ops->eo_fill_root_el(et); |
943cced3 JB |
348 | if (!et->et_ops->eo_fill_max_leaf_clusters) |
349 | et->et_max_leaf_clusters = 0; | |
350 | else | |
6136ca5f | 351 | et->et_ops->eo_fill_max_leaf_clusters(et); |
e7d4cb6b TM |
352 | } |
353 | ||
8d6220d6 JB |
354 | void ocfs2_init_dinode_extent_tree(struct ocfs2_extent_tree *et, |
355 | struct inode *inode, | |
356 | struct buffer_head *bh) | |
1a09f556 | 357 | { |
13723d00 JB |
358 | __ocfs2_init_extent_tree(et, inode, bh, ocfs2_journal_access_di, |
359 | NULL, &ocfs2_dinode_et_ops); | |
1a09f556 JB |
360 | } |
361 | ||
8d6220d6 | 362 | void ocfs2_init_xattr_tree_extent_tree(struct ocfs2_extent_tree *et, |
f99b9b7c | 363 | struct inode *inode, |
8d6220d6 | 364 | struct buffer_head *bh) |
1a09f556 | 365 | { |
13723d00 JB |
366 | __ocfs2_init_extent_tree(et, inode, bh, ocfs2_journal_access_xb, |
367 | NULL, &ocfs2_xattr_tree_et_ops); | |
1a09f556 JB |
368 | } |
369 | ||
8d6220d6 JB |
370 | void ocfs2_init_xattr_value_extent_tree(struct ocfs2_extent_tree *et, |
371 | struct inode *inode, | |
2a50a743 | 372 | struct ocfs2_xattr_value_buf *vb) |
e7d4cb6b | 373 | { |
2a50a743 | 374 | __ocfs2_init_extent_tree(et, inode, vb->vb_bh, vb->vb_access, vb, |
8d6220d6 | 375 | &ocfs2_xattr_value_et_ops); |
e7d4cb6b TM |
376 | } |
377 | ||
9b7895ef MF |
378 | void ocfs2_init_dx_root_extent_tree(struct ocfs2_extent_tree *et, |
379 | struct inode *inode, | |
380 | struct buffer_head *bh) | |
381 | { | |
382 | __ocfs2_init_extent_tree(et, inode, bh, ocfs2_journal_access_dr, | |
383 | NULL, &ocfs2_dx_root_et_ops); | |
384 | } | |
385 | ||
35dc0aa3 JB |
386 | static inline void ocfs2_et_set_last_eb_blk(struct ocfs2_extent_tree *et, |
387 | u64 new_last_eb_blk) | |
e7d4cb6b | 388 | { |
ce1d9ea6 | 389 | et->et_ops->eo_set_last_eb_blk(et, new_last_eb_blk); |
e7d4cb6b TM |
390 | } |
391 | ||
35dc0aa3 | 392 | static inline u64 ocfs2_et_get_last_eb_blk(struct ocfs2_extent_tree *et) |
e7d4cb6b | 393 | { |
ce1d9ea6 | 394 | return et->et_ops->eo_get_last_eb_blk(et); |
e7d4cb6b TM |
395 | } |
396 | ||
6136ca5f | 397 | static inline void ocfs2_et_update_clusters(struct ocfs2_extent_tree *et, |
35dc0aa3 JB |
398 | u32 clusters) |
399 | { | |
6136ca5f | 400 | et->et_ops->eo_update_clusters(et, clusters); |
35dc0aa3 JB |
401 | } |
402 | ||
13723d00 | 403 | static inline int ocfs2_et_root_journal_access(handle_t *handle, |
13723d00 JB |
404 | struct ocfs2_extent_tree *et, |
405 | int type) | |
406 | { | |
d9a0a1f8 | 407 | return et->et_root_journal_access(handle, et->et_ci, et->et_root_bh, |
13723d00 JB |
408 | type); |
409 | } | |
410 | ||
6136ca5f | 411 | static inline int ocfs2_et_insert_check(struct ocfs2_extent_tree *et, |
1e61ee79 JB |
412 | struct ocfs2_extent_rec *rec) |
413 | { | |
414 | int ret = 0; | |
415 | ||
416 | if (et->et_ops->eo_insert_check) | |
6136ca5f | 417 | ret = et->et_ops->eo_insert_check(et, rec); |
1e61ee79 JB |
418 | return ret; |
419 | } | |
420 | ||
6136ca5f | 421 | static inline int ocfs2_et_sanity_check(struct ocfs2_extent_tree *et) |
e7d4cb6b | 422 | { |
1e61ee79 JB |
423 | int ret = 0; |
424 | ||
425 | if (et->et_ops->eo_sanity_check) | |
6136ca5f | 426 | ret = et->et_ops->eo_sanity_check(et); |
1e61ee79 | 427 | return ret; |
e7d4cb6b TM |
428 | } |
429 | ||
dcd0538f | 430 | static void ocfs2_free_truncate_context(struct ocfs2_truncate_context *tc); |
59a5e416 MF |
431 | static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt, |
432 | struct ocfs2_extent_block *eb); | |
ccd979bd | 433 | |
dcd0538f MF |
434 | /* |
435 | * Structures which describe a path through a btree, and functions to | |
436 | * manipulate them. | |
437 | * | |
438 | * The idea here is to be as generic as possible with the tree | |
439 | * manipulation code. | |
440 | */ | |
441 | struct ocfs2_path_item { | |
442 | struct buffer_head *bh; | |
443 | struct ocfs2_extent_list *el; | |
444 | }; | |
ccd979bd | 445 | |
dcd0538f | 446 | #define OCFS2_MAX_PATH_DEPTH 5 |
ccd979bd | 447 | |
dcd0538f | 448 | struct ocfs2_path { |
13723d00 JB |
449 | int p_tree_depth; |
450 | ocfs2_journal_access_func p_root_access; | |
451 | struct ocfs2_path_item p_node[OCFS2_MAX_PATH_DEPTH]; | |
dcd0538f | 452 | }; |
ccd979bd | 453 | |
dcd0538f MF |
454 | #define path_root_bh(_path) ((_path)->p_node[0].bh) |
455 | #define path_root_el(_path) ((_path)->p_node[0].el) | |
13723d00 | 456 | #define path_root_access(_path)((_path)->p_root_access) |
dcd0538f MF |
457 | #define path_leaf_bh(_path) ((_path)->p_node[(_path)->p_tree_depth].bh) |
458 | #define path_leaf_el(_path) ((_path)->p_node[(_path)->p_tree_depth].el) | |
459 | #define path_num_items(_path) ((_path)->p_tree_depth + 1) | |
ccd979bd | 460 | |
facdb77f JB |
461 | static int ocfs2_find_path(struct ocfs2_caching_info *ci, |
462 | struct ocfs2_path *path, u32 cpos); | |
d401dc12 JB |
463 | static void ocfs2_adjust_rightmost_records(handle_t *handle, |
464 | struct ocfs2_extent_tree *et, | |
6b791bcc TM |
465 | struct ocfs2_path *path, |
466 | struct ocfs2_extent_rec *insert_rec); | |
dcd0538f MF |
467 | /* |
468 | * Reset the actual path elements so that we can re-use the structure | |
469 | * to build another path. Generally, this involves freeing the buffer | |
470 | * heads. | |
471 | */ | |
472 | static void ocfs2_reinit_path(struct ocfs2_path *path, int keep_root) | |
473 | { | |
474 | int i, start = 0, depth = 0; | |
475 | struct ocfs2_path_item *node; | |
ccd979bd | 476 | |
dcd0538f MF |
477 | if (keep_root) |
478 | start = 1; | |
ccd979bd | 479 | |
dcd0538f MF |
480 | for(i = start; i < path_num_items(path); i++) { |
481 | node = &path->p_node[i]; | |
482 | ||
483 | brelse(node->bh); | |
484 | node->bh = NULL; | |
485 | node->el = NULL; | |
486 | } | |
487 | ||
488 | /* | |
489 | * Tree depth may change during truncate, or insert. If we're | |
490 | * keeping the root extent list, then make sure that our path | |
491 | * structure reflects the proper depth. | |
492 | */ | |
493 | if (keep_root) | |
494 | depth = le16_to_cpu(path_root_el(path)->l_tree_depth); | |
13723d00 JB |
495 | else |
496 | path_root_access(path) = NULL; | |
dcd0538f MF |
497 | |
498 | path->p_tree_depth = depth; | |
499 | } | |
500 | ||
501 | static void ocfs2_free_path(struct ocfs2_path *path) | |
502 | { | |
503 | if (path) { | |
504 | ocfs2_reinit_path(path, 0); | |
505 | kfree(path); | |
506 | } | |
507 | } | |
508 | ||
328d5752 MF |
509 | /* |
510 | * All the elements of src into dest. After this call, src could be freed | |
511 | * without affecting dest. | |
512 | * | |
513 | * Both paths should have the same root. Any non-root elements of dest | |
514 | * will be freed. | |
515 | */ | |
516 | static void ocfs2_cp_path(struct ocfs2_path *dest, struct ocfs2_path *src) | |
517 | { | |
518 | int i; | |
519 | ||
520 | BUG_ON(path_root_bh(dest) != path_root_bh(src)); | |
521 | BUG_ON(path_root_el(dest) != path_root_el(src)); | |
13723d00 | 522 | BUG_ON(path_root_access(dest) != path_root_access(src)); |
328d5752 MF |
523 | |
524 | ocfs2_reinit_path(dest, 1); | |
525 | ||
526 | for(i = 1; i < OCFS2_MAX_PATH_DEPTH; i++) { | |
527 | dest->p_node[i].bh = src->p_node[i].bh; | |
528 | dest->p_node[i].el = src->p_node[i].el; | |
529 | ||
530 | if (dest->p_node[i].bh) | |
531 | get_bh(dest->p_node[i].bh); | |
532 | } | |
533 | } | |
534 | ||
dcd0538f MF |
535 | /* |
536 | * Make the *dest path the same as src and re-initialize src path to | |
537 | * have a root only. | |
538 | */ | |
539 | static void ocfs2_mv_path(struct ocfs2_path *dest, struct ocfs2_path *src) | |
540 | { | |
541 | int i; | |
542 | ||
543 | BUG_ON(path_root_bh(dest) != path_root_bh(src)); | |
13723d00 | 544 | BUG_ON(path_root_access(dest) != path_root_access(src)); |
dcd0538f MF |
545 | |
546 | for(i = 1; i < OCFS2_MAX_PATH_DEPTH; i++) { | |
547 | brelse(dest->p_node[i].bh); | |
548 | ||
549 | dest->p_node[i].bh = src->p_node[i].bh; | |
550 | dest->p_node[i].el = src->p_node[i].el; | |
551 | ||
552 | src->p_node[i].bh = NULL; | |
553 | src->p_node[i].el = NULL; | |
554 | } | |
555 | } | |
556 | ||
557 | /* | |
558 | * Insert an extent block at given index. | |
559 | * | |
560 | * This will not take an additional reference on eb_bh. | |
561 | */ | |
562 | static inline void ocfs2_path_insert_eb(struct ocfs2_path *path, int index, | |
563 | struct buffer_head *eb_bh) | |
564 | { | |
565 | struct ocfs2_extent_block *eb = (struct ocfs2_extent_block *)eb_bh->b_data; | |
566 | ||
567 | /* | |
568 | * Right now, no root bh is an extent block, so this helps | |
569 | * catch code errors with dinode trees. The assertion can be | |
570 | * safely removed if we ever need to insert extent block | |
571 | * structures at the root. | |
572 | */ | |
573 | BUG_ON(index == 0); | |
574 | ||
575 | path->p_node[index].bh = eb_bh; | |
576 | path->p_node[index].el = &eb->h_list; | |
577 | } | |
578 | ||
579 | static struct ocfs2_path *ocfs2_new_path(struct buffer_head *root_bh, | |
13723d00 JB |
580 | struct ocfs2_extent_list *root_el, |
581 | ocfs2_journal_access_func access) | |
dcd0538f MF |
582 | { |
583 | struct ocfs2_path *path; | |
ccd979bd | 584 | |
dcd0538f MF |
585 | BUG_ON(le16_to_cpu(root_el->l_tree_depth) >= OCFS2_MAX_PATH_DEPTH); |
586 | ||
587 | path = kzalloc(sizeof(*path), GFP_NOFS); | |
588 | if (path) { | |
589 | path->p_tree_depth = le16_to_cpu(root_el->l_tree_depth); | |
590 | get_bh(root_bh); | |
591 | path_root_bh(path) = root_bh; | |
592 | path_root_el(path) = root_el; | |
13723d00 | 593 | path_root_access(path) = access; |
dcd0538f MF |
594 | } |
595 | ||
596 | return path; | |
597 | } | |
598 | ||
ffdd7a54 JB |
599 | static struct ocfs2_path *ocfs2_new_path_from_path(struct ocfs2_path *path) |
600 | { | |
13723d00 JB |
601 | return ocfs2_new_path(path_root_bh(path), path_root_el(path), |
602 | path_root_access(path)); | |
ffdd7a54 JB |
603 | } |
604 | ||
605 | static struct ocfs2_path *ocfs2_new_path_from_et(struct ocfs2_extent_tree *et) | |
606 | { | |
13723d00 JB |
607 | return ocfs2_new_path(et->et_root_bh, et->et_root_el, |
608 | et->et_root_journal_access); | |
609 | } | |
610 | ||
611 | /* | |
612 | * Journal the buffer at depth idx. All idx>0 are extent_blocks, | |
613 | * otherwise it's the root_access function. | |
614 | * | |
615 | * I don't like the way this function's name looks next to | |
616 | * ocfs2_journal_access_path(), but I don't have a better one. | |
617 | */ | |
618 | static int ocfs2_path_bh_journal_access(handle_t *handle, | |
0cf2f763 | 619 | struct ocfs2_caching_info *ci, |
13723d00 JB |
620 | struct ocfs2_path *path, |
621 | int idx) | |
622 | { | |
623 | ocfs2_journal_access_func access = path_root_access(path); | |
624 | ||
625 | if (!access) | |
626 | access = ocfs2_journal_access; | |
627 | ||
628 | if (idx) | |
629 | access = ocfs2_journal_access_eb; | |
630 | ||
0cf2f763 | 631 | return access(handle, ci, path->p_node[idx].bh, |
13723d00 | 632 | OCFS2_JOURNAL_ACCESS_WRITE); |
ffdd7a54 JB |
633 | } |
634 | ||
dcd0538f MF |
635 | /* |
636 | * Convenience function to journal all components in a path. | |
637 | */ | |
0cf2f763 JB |
638 | static int ocfs2_journal_access_path(struct ocfs2_caching_info *ci, |
639 | handle_t *handle, | |
dcd0538f MF |
640 | struct ocfs2_path *path) |
641 | { | |
642 | int i, ret = 0; | |
643 | ||
644 | if (!path) | |
645 | goto out; | |
646 | ||
647 | for(i = 0; i < path_num_items(path); i++) { | |
0cf2f763 | 648 | ret = ocfs2_path_bh_journal_access(handle, ci, path, i); |
dcd0538f MF |
649 | if (ret < 0) { |
650 | mlog_errno(ret); | |
651 | goto out; | |
652 | } | |
653 | } | |
654 | ||
655 | out: | |
656 | return ret; | |
657 | } | |
658 | ||
328d5752 MF |
659 | /* |
660 | * Return the index of the extent record which contains cluster #v_cluster. | |
661 | * -1 is returned if it was not found. | |
662 | * | |
663 | * Should work fine on interior and exterior nodes. | |
664 | */ | |
665 | int ocfs2_search_extent_list(struct ocfs2_extent_list *el, u32 v_cluster) | |
666 | { | |
667 | int ret = -1; | |
668 | int i; | |
669 | struct ocfs2_extent_rec *rec; | |
670 | u32 rec_end, rec_start, clusters; | |
671 | ||
672 | for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) { | |
673 | rec = &el->l_recs[i]; | |
674 | ||
675 | rec_start = le32_to_cpu(rec->e_cpos); | |
676 | clusters = ocfs2_rec_clusters(el, rec); | |
677 | ||
678 | rec_end = rec_start + clusters; | |
679 | ||
680 | if (v_cluster >= rec_start && v_cluster < rec_end) { | |
681 | ret = i; | |
682 | break; | |
683 | } | |
684 | } | |
685 | ||
686 | return ret; | |
687 | } | |
688 | ||
dcd0538f MF |
689 | enum ocfs2_contig_type { |
690 | CONTIG_NONE = 0, | |
691 | CONTIG_LEFT, | |
328d5752 MF |
692 | CONTIG_RIGHT, |
693 | CONTIG_LEFTRIGHT, | |
dcd0538f MF |
694 | }; |
695 | ||
e48edee2 MF |
696 | |
697 | /* | |
698 | * NOTE: ocfs2_block_extent_contig(), ocfs2_extents_adjacent() and | |
699 | * ocfs2_extent_contig only work properly against leaf nodes! | |
700 | */ | |
dcd0538f MF |
701 | static int ocfs2_block_extent_contig(struct super_block *sb, |
702 | struct ocfs2_extent_rec *ext, | |
703 | u64 blkno) | |
ccd979bd | 704 | { |
e48edee2 MF |
705 | u64 blk_end = le64_to_cpu(ext->e_blkno); |
706 | ||
707 | blk_end += ocfs2_clusters_to_blocks(sb, | |
708 | le16_to_cpu(ext->e_leaf_clusters)); | |
709 | ||
710 | return blkno == blk_end; | |
ccd979bd MF |
711 | } |
712 | ||
dcd0538f MF |
713 | static int ocfs2_extents_adjacent(struct ocfs2_extent_rec *left, |
714 | struct ocfs2_extent_rec *right) | |
715 | { | |
e48edee2 MF |
716 | u32 left_range; |
717 | ||
718 | left_range = le32_to_cpu(left->e_cpos) + | |
719 | le16_to_cpu(left->e_leaf_clusters); | |
720 | ||
721 | return (left_range == le32_to_cpu(right->e_cpos)); | |
dcd0538f MF |
722 | } |
723 | ||
724 | static enum ocfs2_contig_type | |
725 | ocfs2_extent_contig(struct inode *inode, | |
726 | struct ocfs2_extent_rec *ext, | |
727 | struct ocfs2_extent_rec *insert_rec) | |
728 | { | |
729 | u64 blkno = le64_to_cpu(insert_rec->e_blkno); | |
730 | ||
328d5752 MF |
731 | /* |
732 | * Refuse to coalesce extent records with different flag | |
733 | * fields - we don't want to mix unwritten extents with user | |
734 | * data. | |
735 | */ | |
736 | if (ext->e_flags != insert_rec->e_flags) | |
737 | return CONTIG_NONE; | |
738 | ||
dcd0538f MF |
739 | if (ocfs2_extents_adjacent(ext, insert_rec) && |
740 | ocfs2_block_extent_contig(inode->i_sb, ext, blkno)) | |
741 | return CONTIG_RIGHT; | |
742 | ||
743 | blkno = le64_to_cpu(ext->e_blkno); | |
744 | if (ocfs2_extents_adjacent(insert_rec, ext) && | |
745 | ocfs2_block_extent_contig(inode->i_sb, insert_rec, blkno)) | |
746 | return CONTIG_LEFT; | |
747 | ||
748 | return CONTIG_NONE; | |
749 | } | |
750 | ||
751 | /* | |
752 | * NOTE: We can have pretty much any combination of contiguousness and | |
753 | * appending. | |
754 | * | |
755 | * The usefulness of APPEND_TAIL is more in that it lets us know that | |
756 | * we'll have to update the path to that leaf. | |
757 | */ | |
758 | enum ocfs2_append_type { | |
759 | APPEND_NONE = 0, | |
760 | APPEND_TAIL, | |
761 | }; | |
762 | ||
328d5752 MF |
763 | enum ocfs2_split_type { |
764 | SPLIT_NONE = 0, | |
765 | SPLIT_LEFT, | |
766 | SPLIT_RIGHT, | |
767 | }; | |
768 | ||
dcd0538f | 769 | struct ocfs2_insert_type { |
328d5752 | 770 | enum ocfs2_split_type ins_split; |
dcd0538f MF |
771 | enum ocfs2_append_type ins_appending; |
772 | enum ocfs2_contig_type ins_contig; | |
773 | int ins_contig_index; | |
dcd0538f MF |
774 | int ins_tree_depth; |
775 | }; | |
776 | ||
328d5752 MF |
777 | struct ocfs2_merge_ctxt { |
778 | enum ocfs2_contig_type c_contig_type; | |
779 | int c_has_empty_extent; | |
780 | int c_split_covers_rec; | |
328d5752 MF |
781 | }; |
782 | ||
5e96581a JB |
783 | static int ocfs2_validate_extent_block(struct super_block *sb, |
784 | struct buffer_head *bh) | |
785 | { | |
d6b32bbb | 786 | int rc; |
5e96581a JB |
787 | struct ocfs2_extent_block *eb = |
788 | (struct ocfs2_extent_block *)bh->b_data; | |
789 | ||
970e4936 JB |
790 | mlog(0, "Validating extent block %llu\n", |
791 | (unsigned long long)bh->b_blocknr); | |
792 | ||
d6b32bbb JB |
793 | BUG_ON(!buffer_uptodate(bh)); |
794 | ||
795 | /* | |
796 | * If the ecc fails, we return the error but otherwise | |
797 | * leave the filesystem running. We know any error is | |
798 | * local to this block. | |
799 | */ | |
800 | rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &eb->h_check); | |
13723d00 JB |
801 | if (rc) { |
802 | mlog(ML_ERROR, "Checksum failed for extent block %llu\n", | |
803 | (unsigned long long)bh->b_blocknr); | |
d6b32bbb | 804 | return rc; |
13723d00 | 805 | } |
d6b32bbb JB |
806 | |
807 | /* | |
808 | * Errors after here are fatal. | |
809 | */ | |
810 | ||
5e96581a JB |
811 | if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) { |
812 | ocfs2_error(sb, | |
813 | "Extent block #%llu has bad signature %.*s", | |
814 | (unsigned long long)bh->b_blocknr, 7, | |
815 | eb->h_signature); | |
816 | return -EINVAL; | |
817 | } | |
818 | ||
819 | if (le64_to_cpu(eb->h_blkno) != bh->b_blocknr) { | |
820 | ocfs2_error(sb, | |
821 | "Extent block #%llu has an invalid h_blkno " | |
822 | "of %llu", | |
823 | (unsigned long long)bh->b_blocknr, | |
824 | (unsigned long long)le64_to_cpu(eb->h_blkno)); | |
825 | return -EINVAL; | |
826 | } | |
827 | ||
828 | if (le32_to_cpu(eb->h_fs_generation) != OCFS2_SB(sb)->fs_generation) { | |
829 | ocfs2_error(sb, | |
830 | "Extent block #%llu has an invalid " | |
831 | "h_fs_generation of #%u", | |
832 | (unsigned long long)bh->b_blocknr, | |
833 | le32_to_cpu(eb->h_fs_generation)); | |
834 | return -EINVAL; | |
835 | } | |
836 | ||
837 | return 0; | |
838 | } | |
839 | ||
3d03a305 | 840 | int ocfs2_read_extent_block(struct ocfs2_caching_info *ci, u64 eb_blkno, |
5e96581a JB |
841 | struct buffer_head **bh) |
842 | { | |
843 | int rc; | |
844 | struct buffer_head *tmp = *bh; | |
845 | ||
3d03a305 | 846 | rc = ocfs2_read_block(ci, eb_blkno, &tmp, |
970e4936 | 847 | ocfs2_validate_extent_block); |
5e96581a JB |
848 | |
849 | /* If ocfs2_read_block() got us a new bh, pass it up. */ | |
970e4936 | 850 | if (!rc && !*bh) |
5e96581a JB |
851 | *bh = tmp; |
852 | ||
5e96581a JB |
853 | return rc; |
854 | } | |
855 | ||
856 | ||
ccd979bd MF |
857 | /* |
858 | * How many free extents have we got before we need more meta data? | |
859 | */ | |
860 | int ocfs2_num_free_extents(struct ocfs2_super *osb, | |
f99b9b7c | 861 | struct ocfs2_extent_tree *et) |
ccd979bd MF |
862 | { |
863 | int retval; | |
e7d4cb6b | 864 | struct ocfs2_extent_list *el = NULL; |
ccd979bd MF |
865 | struct ocfs2_extent_block *eb; |
866 | struct buffer_head *eb_bh = NULL; | |
e7d4cb6b | 867 | u64 last_eb_blk = 0; |
ccd979bd MF |
868 | |
869 | mlog_entry_void(); | |
870 | ||
f99b9b7c JB |
871 | el = et->et_root_el; |
872 | last_eb_blk = ocfs2_et_get_last_eb_blk(et); | |
ccd979bd | 873 | |
e7d4cb6b | 874 | if (last_eb_blk) { |
3d03a305 JB |
875 | retval = ocfs2_read_extent_block(et->et_ci, last_eb_blk, |
876 | &eb_bh); | |
ccd979bd MF |
877 | if (retval < 0) { |
878 | mlog_errno(retval); | |
879 | goto bail; | |
880 | } | |
881 | eb = (struct ocfs2_extent_block *) eb_bh->b_data; | |
882 | el = &eb->h_list; | |
e7d4cb6b | 883 | } |
ccd979bd MF |
884 | |
885 | BUG_ON(el->l_tree_depth != 0); | |
886 | ||
887 | retval = le16_to_cpu(el->l_count) - le16_to_cpu(el->l_next_free_rec); | |
888 | bail: | |
a81cb88b | 889 | brelse(eb_bh); |
ccd979bd MF |
890 | |
891 | mlog_exit(retval); | |
892 | return retval; | |
893 | } | |
894 | ||
895 | /* expects array to already be allocated | |
896 | * | |
897 | * sets h_signature, h_blkno, h_suballoc_bit, h_suballoc_slot, and | |
898 | * l_count for you | |
899 | */ | |
42a5a7a9 JB |
900 | static int ocfs2_create_new_meta_bhs(handle_t *handle, |
901 | struct ocfs2_extent_tree *et, | |
ccd979bd MF |
902 | int wanted, |
903 | struct ocfs2_alloc_context *meta_ac, | |
904 | struct buffer_head *bhs[]) | |
905 | { | |
906 | int count, status, i; | |
907 | u16 suballoc_bit_start; | |
908 | u32 num_got; | |
909 | u64 first_blkno; | |
42a5a7a9 JB |
910 | struct ocfs2_super *osb = |
911 | OCFS2_SB(ocfs2_metadata_cache_get_super(et->et_ci)); | |
ccd979bd MF |
912 | struct ocfs2_extent_block *eb; |
913 | ||
914 | mlog_entry_void(); | |
915 | ||
916 | count = 0; | |
917 | while (count < wanted) { | |
918 | status = ocfs2_claim_metadata(osb, | |
919 | handle, | |
920 | meta_ac, | |
921 | wanted - count, | |
922 | &suballoc_bit_start, | |
923 | &num_got, | |
924 | &first_blkno); | |
925 | if (status < 0) { | |
926 | mlog_errno(status); | |
927 | goto bail; | |
928 | } | |
929 | ||
930 | for(i = count; i < (num_got + count); i++) { | |
931 | bhs[i] = sb_getblk(osb->sb, first_blkno); | |
932 | if (bhs[i] == NULL) { | |
933 | status = -EIO; | |
934 | mlog_errno(status); | |
935 | goto bail; | |
936 | } | |
42a5a7a9 | 937 | ocfs2_set_new_buffer_uptodate(et->et_ci, bhs[i]); |
ccd979bd | 938 | |
42a5a7a9 JB |
939 | status = ocfs2_journal_access_eb(handle, et->et_ci, |
940 | bhs[i], | |
13723d00 | 941 | OCFS2_JOURNAL_ACCESS_CREATE); |
ccd979bd MF |
942 | if (status < 0) { |
943 | mlog_errno(status); | |
944 | goto bail; | |
945 | } | |
946 | ||
947 | memset(bhs[i]->b_data, 0, osb->sb->s_blocksize); | |
948 | eb = (struct ocfs2_extent_block *) bhs[i]->b_data; | |
949 | /* Ok, setup the minimal stuff here. */ | |
950 | strcpy(eb->h_signature, OCFS2_EXTENT_BLOCK_SIGNATURE); | |
951 | eb->h_blkno = cpu_to_le64(first_blkno); | |
952 | eb->h_fs_generation = cpu_to_le32(osb->fs_generation); | |
ccd979bd | 953 | eb->h_suballoc_slot = cpu_to_le16(osb->slot_num); |
ccd979bd MF |
954 | eb->h_suballoc_bit = cpu_to_le16(suballoc_bit_start); |
955 | eb->h_list.l_count = | |
956 | cpu_to_le16(ocfs2_extent_recs_per_eb(osb->sb)); | |
957 | ||
958 | suballoc_bit_start++; | |
959 | first_blkno++; | |
960 | ||
961 | /* We'll also be dirtied by the caller, so | |
962 | * this isn't absolutely necessary. */ | |
963 | status = ocfs2_journal_dirty(handle, bhs[i]); | |
964 | if (status < 0) { | |
965 | mlog_errno(status); | |
966 | goto bail; | |
967 | } | |
968 | } | |
969 | ||
970 | count += num_got; | |
971 | } | |
972 | ||
973 | status = 0; | |
974 | bail: | |
975 | if (status < 0) { | |
976 | for(i = 0; i < wanted; i++) { | |
a81cb88b | 977 | brelse(bhs[i]); |
ccd979bd MF |
978 | bhs[i] = NULL; |
979 | } | |
980 | } | |
981 | mlog_exit(status); | |
982 | return status; | |
983 | } | |
984 | ||
dcd0538f MF |
985 | /* |
986 | * Helper function for ocfs2_add_branch() and ocfs2_shift_tree_depth(). | |
987 | * | |
988 | * Returns the sum of the rightmost extent rec logical offset and | |
989 | * cluster count. | |
990 | * | |
991 | * ocfs2_add_branch() uses this to determine what logical cluster | |
992 | * value should be populated into the leftmost new branch records. | |
993 | * | |
994 | * ocfs2_shift_tree_depth() uses this to determine the # clusters | |
995 | * value for the new topmost tree record. | |
996 | */ | |
997 | static inline u32 ocfs2_sum_rightmost_rec(struct ocfs2_extent_list *el) | |
998 | { | |
999 | int i; | |
1000 | ||
1001 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
1002 | ||
1003 | return le32_to_cpu(el->l_recs[i].e_cpos) + | |
e48edee2 | 1004 | ocfs2_rec_clusters(el, &el->l_recs[i]); |
dcd0538f MF |
1005 | } |
1006 | ||
6b791bcc TM |
1007 | /* |
1008 | * Change range of the branches in the right most path according to the leaf | |
1009 | * extent block's rightmost record. | |
1010 | */ | |
1011 | static int ocfs2_adjust_rightmost_branch(handle_t *handle, | |
6b791bcc TM |
1012 | struct ocfs2_extent_tree *et) |
1013 | { | |
1014 | int status; | |
1015 | struct ocfs2_path *path = NULL; | |
1016 | struct ocfs2_extent_list *el; | |
1017 | struct ocfs2_extent_rec *rec; | |
1018 | ||
1019 | path = ocfs2_new_path_from_et(et); | |
1020 | if (!path) { | |
1021 | status = -ENOMEM; | |
1022 | return status; | |
1023 | } | |
1024 | ||
facdb77f | 1025 | status = ocfs2_find_path(et->et_ci, path, UINT_MAX); |
6b791bcc TM |
1026 | if (status < 0) { |
1027 | mlog_errno(status); | |
1028 | goto out; | |
1029 | } | |
1030 | ||
1031 | status = ocfs2_extend_trans(handle, path_num_items(path) + | |
1032 | handle->h_buffer_credits); | |
1033 | if (status < 0) { | |
1034 | mlog_errno(status); | |
1035 | goto out; | |
1036 | } | |
1037 | ||
d401dc12 | 1038 | status = ocfs2_journal_access_path(et->et_ci, handle, path); |
6b791bcc TM |
1039 | if (status < 0) { |
1040 | mlog_errno(status); | |
1041 | goto out; | |
1042 | } | |
1043 | ||
1044 | el = path_leaf_el(path); | |
1045 | rec = &el->l_recs[le32_to_cpu(el->l_next_free_rec) - 1]; | |
1046 | ||
d401dc12 | 1047 | ocfs2_adjust_rightmost_records(handle, et, path, rec); |
6b791bcc TM |
1048 | |
1049 | out: | |
1050 | ocfs2_free_path(path); | |
1051 | return status; | |
1052 | } | |
1053 | ||
ccd979bd MF |
1054 | /* |
1055 | * Add an entire tree branch to our inode. eb_bh is the extent block | |
d401dc12 | 1056 | * to start at, if we don't want to start the branch at the root |
ccd979bd MF |
1057 | * structure. |
1058 | * | |
1059 | * last_eb_bh is required as we have to update it's next_leaf pointer | |
1060 | * for the new last extent block. | |
1061 | * | |
1062 | * the new branch will be 'empty' in the sense that every block will | |
e48edee2 | 1063 | * contain a single record with cluster count == 0. |
ccd979bd | 1064 | */ |
d401dc12 | 1065 | static int ocfs2_add_branch(handle_t *handle, |
e7d4cb6b | 1066 | struct ocfs2_extent_tree *et, |
ccd979bd | 1067 | struct buffer_head *eb_bh, |
328d5752 | 1068 | struct buffer_head **last_eb_bh, |
ccd979bd MF |
1069 | struct ocfs2_alloc_context *meta_ac) |
1070 | { | |
1071 | int status, new_blocks, i; | |
1072 | u64 next_blkno, new_last_eb_blk; | |
1073 | struct buffer_head *bh; | |
1074 | struct buffer_head **new_eb_bhs = NULL; | |
ccd979bd MF |
1075 | struct ocfs2_extent_block *eb; |
1076 | struct ocfs2_extent_list *eb_el; | |
1077 | struct ocfs2_extent_list *el; | |
6b791bcc | 1078 | u32 new_cpos, root_end; |
ccd979bd MF |
1079 | |
1080 | mlog_entry_void(); | |
1081 | ||
328d5752 | 1082 | BUG_ON(!last_eb_bh || !*last_eb_bh); |
ccd979bd | 1083 | |
ccd979bd MF |
1084 | if (eb_bh) { |
1085 | eb = (struct ocfs2_extent_block *) eb_bh->b_data; | |
1086 | el = &eb->h_list; | |
1087 | } else | |
ce1d9ea6 | 1088 | el = et->et_root_el; |
ccd979bd MF |
1089 | |
1090 | /* we never add a branch to a leaf. */ | |
1091 | BUG_ON(!el->l_tree_depth); | |
1092 | ||
1093 | new_blocks = le16_to_cpu(el->l_tree_depth); | |
1094 | ||
6b791bcc TM |
1095 | eb = (struct ocfs2_extent_block *)(*last_eb_bh)->b_data; |
1096 | new_cpos = ocfs2_sum_rightmost_rec(&eb->h_list); | |
1097 | root_end = ocfs2_sum_rightmost_rec(et->et_root_el); | |
1098 | ||
1099 | /* | |
1100 | * If there is a gap before the root end and the real end | |
1101 | * of the righmost leaf block, we need to remove the gap | |
1102 | * between new_cpos and root_end first so that the tree | |
1103 | * is consistent after we add a new branch(it will start | |
1104 | * from new_cpos). | |
1105 | */ | |
1106 | if (root_end > new_cpos) { | |
1107 | mlog(0, "adjust the cluster end from %u to %u\n", | |
1108 | root_end, new_cpos); | |
d401dc12 | 1109 | status = ocfs2_adjust_rightmost_branch(handle, et); |
6b791bcc TM |
1110 | if (status) { |
1111 | mlog_errno(status); | |
1112 | goto bail; | |
1113 | } | |
1114 | } | |
1115 | ||
ccd979bd MF |
1116 | /* allocate the number of new eb blocks we need */ |
1117 | new_eb_bhs = kcalloc(new_blocks, sizeof(struct buffer_head *), | |
1118 | GFP_KERNEL); | |
1119 | if (!new_eb_bhs) { | |
1120 | status = -ENOMEM; | |
1121 | mlog_errno(status); | |
1122 | goto bail; | |
1123 | } | |
1124 | ||
42a5a7a9 | 1125 | status = ocfs2_create_new_meta_bhs(handle, et, new_blocks, |
ccd979bd MF |
1126 | meta_ac, new_eb_bhs); |
1127 | if (status < 0) { | |
1128 | mlog_errno(status); | |
1129 | goto bail; | |
1130 | } | |
1131 | ||
1132 | /* Note: new_eb_bhs[new_blocks - 1] is the guy which will be | |
1133 | * linked with the rest of the tree. | |
1134 | * conversly, new_eb_bhs[0] is the new bottommost leaf. | |
1135 | * | |
1136 | * when we leave the loop, new_last_eb_blk will point to the | |
1137 | * newest leaf, and next_blkno will point to the topmost extent | |
1138 | * block. */ | |
1139 | next_blkno = new_last_eb_blk = 0; | |
1140 | for(i = 0; i < new_blocks; i++) { | |
1141 | bh = new_eb_bhs[i]; | |
1142 | eb = (struct ocfs2_extent_block *) bh->b_data; | |
5e96581a JB |
1143 | /* ocfs2_create_new_meta_bhs() should create it right! */ |
1144 | BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb)); | |
ccd979bd MF |
1145 | eb_el = &eb->h_list; |
1146 | ||
d401dc12 | 1147 | status = ocfs2_journal_access_eb(handle, et->et_ci, bh, |
13723d00 | 1148 | OCFS2_JOURNAL_ACCESS_CREATE); |
ccd979bd MF |
1149 | if (status < 0) { |
1150 | mlog_errno(status); | |
1151 | goto bail; | |
1152 | } | |
1153 | ||
1154 | eb->h_next_leaf_blk = 0; | |
1155 | eb_el->l_tree_depth = cpu_to_le16(i); | |
1156 | eb_el->l_next_free_rec = cpu_to_le16(1); | |
dcd0538f MF |
1157 | /* |
1158 | * This actually counts as an empty extent as | |
1159 | * c_clusters == 0 | |
1160 | */ | |
1161 | eb_el->l_recs[0].e_cpos = cpu_to_le32(new_cpos); | |
ccd979bd | 1162 | eb_el->l_recs[0].e_blkno = cpu_to_le64(next_blkno); |
e48edee2 MF |
1163 | /* |
1164 | * eb_el isn't always an interior node, but even leaf | |
1165 | * nodes want a zero'd flags and reserved field so | |
1166 | * this gets the whole 32 bits regardless of use. | |
1167 | */ | |
1168 | eb_el->l_recs[0].e_int_clusters = cpu_to_le32(0); | |
ccd979bd MF |
1169 | if (!eb_el->l_tree_depth) |
1170 | new_last_eb_blk = le64_to_cpu(eb->h_blkno); | |
1171 | ||
1172 | status = ocfs2_journal_dirty(handle, bh); | |
1173 | if (status < 0) { | |
1174 | mlog_errno(status); | |
1175 | goto bail; | |
1176 | } | |
1177 | ||
1178 | next_blkno = le64_to_cpu(eb->h_blkno); | |
1179 | } | |
1180 | ||
1181 | /* This is a bit hairy. We want to update up to three blocks | |
1182 | * here without leaving any of them in an inconsistent state | |
1183 | * in case of error. We don't have to worry about | |
1184 | * journal_dirty erroring as it won't unless we've aborted the | |
1185 | * handle (in which case we would never be here) so reserving | |
1186 | * the write with journal_access is all we need to do. */ | |
d401dc12 | 1187 | status = ocfs2_journal_access_eb(handle, et->et_ci, *last_eb_bh, |
13723d00 | 1188 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
1189 | if (status < 0) { |
1190 | mlog_errno(status); | |
1191 | goto bail; | |
1192 | } | |
d9a0a1f8 | 1193 | status = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 1194 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
1195 | if (status < 0) { |
1196 | mlog_errno(status); | |
1197 | goto bail; | |
1198 | } | |
1199 | if (eb_bh) { | |
d401dc12 | 1200 | status = ocfs2_journal_access_eb(handle, et->et_ci, eb_bh, |
13723d00 | 1201 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
1202 | if (status < 0) { |
1203 | mlog_errno(status); | |
1204 | goto bail; | |
1205 | } | |
1206 | } | |
1207 | ||
1208 | /* Link the new branch into the rest of the tree (el will | |
e7d4cb6b | 1209 | * either be on the root_bh, or the extent block passed in. */ |
ccd979bd MF |
1210 | i = le16_to_cpu(el->l_next_free_rec); |
1211 | el->l_recs[i].e_blkno = cpu_to_le64(next_blkno); | |
dcd0538f | 1212 | el->l_recs[i].e_cpos = cpu_to_le32(new_cpos); |
e48edee2 | 1213 | el->l_recs[i].e_int_clusters = 0; |
ccd979bd MF |
1214 | le16_add_cpu(&el->l_next_free_rec, 1); |
1215 | ||
1216 | /* fe needs a new last extent block pointer, as does the | |
1217 | * next_leaf on the previously last-extent-block. */ | |
35dc0aa3 | 1218 | ocfs2_et_set_last_eb_blk(et, new_last_eb_blk); |
ccd979bd | 1219 | |
328d5752 | 1220 | eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data; |
ccd979bd MF |
1221 | eb->h_next_leaf_blk = cpu_to_le64(new_last_eb_blk); |
1222 | ||
328d5752 | 1223 | status = ocfs2_journal_dirty(handle, *last_eb_bh); |
ccd979bd MF |
1224 | if (status < 0) |
1225 | mlog_errno(status); | |
ce1d9ea6 | 1226 | status = ocfs2_journal_dirty(handle, et->et_root_bh); |
ccd979bd MF |
1227 | if (status < 0) |
1228 | mlog_errno(status); | |
1229 | if (eb_bh) { | |
1230 | status = ocfs2_journal_dirty(handle, eb_bh); | |
1231 | if (status < 0) | |
1232 | mlog_errno(status); | |
1233 | } | |
1234 | ||
328d5752 MF |
1235 | /* |
1236 | * Some callers want to track the rightmost leaf so pass it | |
1237 | * back here. | |
1238 | */ | |
1239 | brelse(*last_eb_bh); | |
1240 | get_bh(new_eb_bhs[0]); | |
1241 | *last_eb_bh = new_eb_bhs[0]; | |
1242 | ||
ccd979bd MF |
1243 | status = 0; |
1244 | bail: | |
1245 | if (new_eb_bhs) { | |
1246 | for (i = 0; i < new_blocks; i++) | |
a81cb88b | 1247 | brelse(new_eb_bhs[i]); |
ccd979bd MF |
1248 | kfree(new_eb_bhs); |
1249 | } | |
1250 | ||
1251 | mlog_exit(status); | |
1252 | return status; | |
1253 | } | |
1254 | ||
1255 | /* | |
1256 | * adds another level to the allocation tree. | |
1257 | * returns back the new extent block so you can add a branch to it | |
1258 | * after this call. | |
1259 | */ | |
d401dc12 | 1260 | static int ocfs2_shift_tree_depth(handle_t *handle, |
e7d4cb6b | 1261 | struct ocfs2_extent_tree *et, |
ccd979bd MF |
1262 | struct ocfs2_alloc_context *meta_ac, |
1263 | struct buffer_head **ret_new_eb_bh) | |
1264 | { | |
1265 | int status, i; | |
dcd0538f | 1266 | u32 new_clusters; |
ccd979bd | 1267 | struct buffer_head *new_eb_bh = NULL; |
ccd979bd | 1268 | struct ocfs2_extent_block *eb; |
e7d4cb6b | 1269 | struct ocfs2_extent_list *root_el; |
ccd979bd MF |
1270 | struct ocfs2_extent_list *eb_el; |
1271 | ||
1272 | mlog_entry_void(); | |
1273 | ||
42a5a7a9 | 1274 | status = ocfs2_create_new_meta_bhs(handle, et, 1, meta_ac, |
ccd979bd MF |
1275 | &new_eb_bh); |
1276 | if (status < 0) { | |
1277 | mlog_errno(status); | |
1278 | goto bail; | |
1279 | } | |
1280 | ||
1281 | eb = (struct ocfs2_extent_block *) new_eb_bh->b_data; | |
5e96581a JB |
1282 | /* ocfs2_create_new_meta_bhs() should create it right! */ |
1283 | BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb)); | |
ccd979bd MF |
1284 | |
1285 | eb_el = &eb->h_list; | |
ce1d9ea6 | 1286 | root_el = et->et_root_el; |
ccd979bd | 1287 | |
d401dc12 | 1288 | status = ocfs2_journal_access_eb(handle, et->et_ci, new_eb_bh, |
13723d00 | 1289 | OCFS2_JOURNAL_ACCESS_CREATE); |
ccd979bd MF |
1290 | if (status < 0) { |
1291 | mlog_errno(status); | |
1292 | goto bail; | |
1293 | } | |
1294 | ||
e7d4cb6b TM |
1295 | /* copy the root extent list data into the new extent block */ |
1296 | eb_el->l_tree_depth = root_el->l_tree_depth; | |
1297 | eb_el->l_next_free_rec = root_el->l_next_free_rec; | |
1298 | for (i = 0; i < le16_to_cpu(root_el->l_next_free_rec); i++) | |
1299 | eb_el->l_recs[i] = root_el->l_recs[i]; | |
ccd979bd MF |
1300 | |
1301 | status = ocfs2_journal_dirty(handle, new_eb_bh); | |
1302 | if (status < 0) { | |
1303 | mlog_errno(status); | |
1304 | goto bail; | |
1305 | } | |
1306 | ||
d9a0a1f8 | 1307 | status = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 1308 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
1309 | if (status < 0) { |
1310 | mlog_errno(status); | |
1311 | goto bail; | |
1312 | } | |
1313 | ||
dcd0538f MF |
1314 | new_clusters = ocfs2_sum_rightmost_rec(eb_el); |
1315 | ||
e7d4cb6b TM |
1316 | /* update root_bh now */ |
1317 | le16_add_cpu(&root_el->l_tree_depth, 1); | |
1318 | root_el->l_recs[0].e_cpos = 0; | |
1319 | root_el->l_recs[0].e_blkno = eb->h_blkno; | |
1320 | root_el->l_recs[0].e_int_clusters = cpu_to_le32(new_clusters); | |
1321 | for (i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++) | |
1322 | memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec)); | |
1323 | root_el->l_next_free_rec = cpu_to_le16(1); | |
ccd979bd MF |
1324 | |
1325 | /* If this is our 1st tree depth shift, then last_eb_blk | |
1326 | * becomes the allocated extent block */ | |
e7d4cb6b | 1327 | if (root_el->l_tree_depth == cpu_to_le16(1)) |
35dc0aa3 | 1328 | ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno)); |
ccd979bd | 1329 | |
ce1d9ea6 | 1330 | status = ocfs2_journal_dirty(handle, et->et_root_bh); |
ccd979bd MF |
1331 | if (status < 0) { |
1332 | mlog_errno(status); | |
1333 | goto bail; | |
1334 | } | |
1335 | ||
1336 | *ret_new_eb_bh = new_eb_bh; | |
1337 | new_eb_bh = NULL; | |
1338 | status = 0; | |
1339 | bail: | |
a81cb88b | 1340 | brelse(new_eb_bh); |
ccd979bd MF |
1341 | |
1342 | mlog_exit(status); | |
1343 | return status; | |
1344 | } | |
1345 | ||
ccd979bd MF |
1346 | /* |
1347 | * Should only be called when there is no space left in any of the | |
1348 | * leaf nodes. What we want to do is find the lowest tree depth | |
1349 | * non-leaf extent block with room for new records. There are three | |
1350 | * valid results of this search: | |
1351 | * | |
1352 | * 1) a lowest extent block is found, then we pass it back in | |
1353 | * *lowest_eb_bh and return '0' | |
1354 | * | |
e7d4cb6b | 1355 | * 2) the search fails to find anything, but the root_el has room. We |
ccd979bd MF |
1356 | * pass NULL back in *lowest_eb_bh, but still return '0' |
1357 | * | |
e7d4cb6b | 1358 | * 3) the search fails to find anything AND the root_el is full, in |
ccd979bd MF |
1359 | * which case we return > 0 |
1360 | * | |
1361 | * return status < 0 indicates an error. | |
1362 | */ | |
d401dc12 | 1363 | static int ocfs2_find_branch_target(struct ocfs2_extent_tree *et, |
ccd979bd MF |
1364 | struct buffer_head **target_bh) |
1365 | { | |
1366 | int status = 0, i; | |
1367 | u64 blkno; | |
ccd979bd MF |
1368 | struct ocfs2_extent_block *eb; |
1369 | struct ocfs2_extent_list *el; | |
1370 | struct buffer_head *bh = NULL; | |
1371 | struct buffer_head *lowest_bh = NULL; | |
1372 | ||
1373 | mlog_entry_void(); | |
1374 | ||
1375 | *target_bh = NULL; | |
1376 | ||
ce1d9ea6 | 1377 | el = et->et_root_el; |
ccd979bd MF |
1378 | |
1379 | while(le16_to_cpu(el->l_tree_depth) > 1) { | |
1380 | if (le16_to_cpu(el->l_next_free_rec) == 0) { | |
3d03a305 JB |
1381 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
1382 | "Owner %llu has empty " | |
ccd979bd | 1383 | "extent list (next_free_rec == 0)", |
3d03a305 | 1384 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci)); |
ccd979bd MF |
1385 | status = -EIO; |
1386 | goto bail; | |
1387 | } | |
1388 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
1389 | blkno = le64_to_cpu(el->l_recs[i].e_blkno); | |
1390 | if (!blkno) { | |
3d03a305 JB |
1391 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
1392 | "Owner %llu has extent " | |
ccd979bd MF |
1393 | "list where extent # %d has no physical " |
1394 | "block start", | |
3d03a305 | 1395 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), i); |
ccd979bd MF |
1396 | status = -EIO; |
1397 | goto bail; | |
1398 | } | |
1399 | ||
a81cb88b MF |
1400 | brelse(bh); |
1401 | bh = NULL; | |
ccd979bd | 1402 | |
3d03a305 | 1403 | status = ocfs2_read_extent_block(et->et_ci, blkno, &bh); |
ccd979bd MF |
1404 | if (status < 0) { |
1405 | mlog_errno(status); | |
1406 | goto bail; | |
1407 | } | |
dcd0538f MF |
1408 | |
1409 | eb = (struct ocfs2_extent_block *) bh->b_data; | |
dcd0538f MF |
1410 | el = &eb->h_list; |
1411 | ||
1412 | if (le16_to_cpu(el->l_next_free_rec) < | |
1413 | le16_to_cpu(el->l_count)) { | |
a81cb88b | 1414 | brelse(lowest_bh); |
dcd0538f MF |
1415 | lowest_bh = bh; |
1416 | get_bh(lowest_bh); | |
1417 | } | |
1418 | } | |
1419 | ||
1420 | /* If we didn't find one and the fe doesn't have any room, | |
1421 | * then return '1' */ | |
ce1d9ea6 | 1422 | el = et->et_root_el; |
e7d4cb6b | 1423 | if (!lowest_bh && (el->l_next_free_rec == el->l_count)) |
dcd0538f MF |
1424 | status = 1; |
1425 | ||
1426 | *target_bh = lowest_bh; | |
1427 | bail: | |
a81cb88b | 1428 | brelse(bh); |
dcd0538f MF |
1429 | |
1430 | mlog_exit(status); | |
1431 | return status; | |
1432 | } | |
1433 | ||
c3afcbb3 MF |
1434 | /* |
1435 | * Grow a b-tree so that it has more records. | |
1436 | * | |
1437 | * We might shift the tree depth in which case existing paths should | |
1438 | * be considered invalid. | |
1439 | * | |
1440 | * Tree depth after the grow is returned via *final_depth. | |
328d5752 MF |
1441 | * |
1442 | * *last_eb_bh will be updated by ocfs2_add_branch(). | |
c3afcbb3 | 1443 | */ |
d401dc12 JB |
1444 | static int ocfs2_grow_tree(handle_t *handle, struct ocfs2_extent_tree *et, |
1445 | int *final_depth, struct buffer_head **last_eb_bh, | |
c3afcbb3 MF |
1446 | struct ocfs2_alloc_context *meta_ac) |
1447 | { | |
1448 | int ret, shift; | |
ce1d9ea6 | 1449 | struct ocfs2_extent_list *el = et->et_root_el; |
e7d4cb6b | 1450 | int depth = le16_to_cpu(el->l_tree_depth); |
c3afcbb3 MF |
1451 | struct buffer_head *bh = NULL; |
1452 | ||
1453 | BUG_ON(meta_ac == NULL); | |
1454 | ||
d401dc12 | 1455 | shift = ocfs2_find_branch_target(et, &bh); |
c3afcbb3 MF |
1456 | if (shift < 0) { |
1457 | ret = shift; | |
1458 | mlog_errno(ret); | |
1459 | goto out; | |
1460 | } | |
1461 | ||
1462 | /* We traveled all the way to the bottom of the allocation tree | |
1463 | * and didn't find room for any more extents - we need to add | |
1464 | * another tree level */ | |
1465 | if (shift) { | |
1466 | BUG_ON(bh); | |
1467 | mlog(0, "need to shift tree depth (current = %d)\n", depth); | |
1468 | ||
1469 | /* ocfs2_shift_tree_depth will return us a buffer with | |
1470 | * the new extent block (so we can pass that to | |
1471 | * ocfs2_add_branch). */ | |
d401dc12 | 1472 | ret = ocfs2_shift_tree_depth(handle, et, meta_ac, &bh); |
c3afcbb3 MF |
1473 | if (ret < 0) { |
1474 | mlog_errno(ret); | |
1475 | goto out; | |
1476 | } | |
1477 | depth++; | |
328d5752 MF |
1478 | if (depth == 1) { |
1479 | /* | |
1480 | * Special case: we have room now if we shifted from | |
1481 | * tree_depth 0, so no more work needs to be done. | |
1482 | * | |
1483 | * We won't be calling add_branch, so pass | |
1484 | * back *last_eb_bh as the new leaf. At depth | |
1485 | * zero, it should always be null so there's | |
1486 | * no reason to brelse. | |
1487 | */ | |
1488 | BUG_ON(*last_eb_bh); | |
1489 | get_bh(bh); | |
1490 | *last_eb_bh = bh; | |
c3afcbb3 | 1491 | goto out; |
328d5752 | 1492 | } |
c3afcbb3 MF |
1493 | } |
1494 | ||
1495 | /* call ocfs2_add_branch to add the final part of the tree with | |
1496 | * the new data. */ | |
1497 | mlog(0, "add branch. bh = %p\n", bh); | |
d401dc12 | 1498 | ret = ocfs2_add_branch(handle, et, bh, last_eb_bh, |
c3afcbb3 MF |
1499 | meta_ac); |
1500 | if (ret < 0) { | |
1501 | mlog_errno(ret); | |
1502 | goto out; | |
1503 | } | |
1504 | ||
1505 | out: | |
1506 | if (final_depth) | |
1507 | *final_depth = depth; | |
1508 | brelse(bh); | |
1509 | return ret; | |
1510 | } | |
1511 | ||
dcd0538f MF |
1512 | /* |
1513 | * This function will discard the rightmost extent record. | |
1514 | */ | |
1515 | static void ocfs2_shift_records_right(struct ocfs2_extent_list *el) | |
1516 | { | |
1517 | int next_free = le16_to_cpu(el->l_next_free_rec); | |
1518 | int count = le16_to_cpu(el->l_count); | |
1519 | unsigned int num_bytes; | |
1520 | ||
1521 | BUG_ON(!next_free); | |
1522 | /* This will cause us to go off the end of our extent list. */ | |
1523 | BUG_ON(next_free >= count); | |
1524 | ||
1525 | num_bytes = sizeof(struct ocfs2_extent_rec) * next_free; | |
1526 | ||
1527 | memmove(&el->l_recs[1], &el->l_recs[0], num_bytes); | |
1528 | } | |
1529 | ||
1530 | static void ocfs2_rotate_leaf(struct ocfs2_extent_list *el, | |
1531 | struct ocfs2_extent_rec *insert_rec) | |
1532 | { | |
1533 | int i, insert_index, next_free, has_empty, num_bytes; | |
1534 | u32 insert_cpos = le32_to_cpu(insert_rec->e_cpos); | |
1535 | struct ocfs2_extent_rec *rec; | |
1536 | ||
1537 | next_free = le16_to_cpu(el->l_next_free_rec); | |
1538 | has_empty = ocfs2_is_empty_extent(&el->l_recs[0]); | |
1539 | ||
1540 | BUG_ON(!next_free); | |
1541 | ||
1542 | /* The tree code before us didn't allow enough room in the leaf. */ | |
b1f3550f | 1543 | BUG_ON(el->l_next_free_rec == el->l_count && !has_empty); |
dcd0538f MF |
1544 | |
1545 | /* | |
1546 | * The easiest way to approach this is to just remove the | |
1547 | * empty extent and temporarily decrement next_free. | |
1548 | */ | |
1549 | if (has_empty) { | |
1550 | /* | |
1551 | * If next_free was 1 (only an empty extent), this | |
1552 | * loop won't execute, which is fine. We still want | |
1553 | * the decrement above to happen. | |
1554 | */ | |
1555 | for(i = 0; i < (next_free - 1); i++) | |
1556 | el->l_recs[i] = el->l_recs[i+1]; | |
1557 | ||
1558 | next_free--; | |
1559 | } | |
1560 | ||
1561 | /* | |
1562 | * Figure out what the new record index should be. | |
1563 | */ | |
1564 | for(i = 0; i < next_free; i++) { | |
1565 | rec = &el->l_recs[i]; | |
1566 | ||
1567 | if (insert_cpos < le32_to_cpu(rec->e_cpos)) | |
1568 | break; | |
1569 | } | |
1570 | insert_index = i; | |
1571 | ||
1572 | mlog(0, "ins %u: index %d, has_empty %d, next_free %d, count %d\n", | |
1573 | insert_cpos, insert_index, has_empty, next_free, le16_to_cpu(el->l_count)); | |
1574 | ||
1575 | BUG_ON(insert_index < 0); | |
1576 | BUG_ON(insert_index >= le16_to_cpu(el->l_count)); | |
1577 | BUG_ON(insert_index > next_free); | |
1578 | ||
1579 | /* | |
1580 | * No need to memmove if we're just adding to the tail. | |
1581 | */ | |
1582 | if (insert_index != next_free) { | |
1583 | BUG_ON(next_free >= le16_to_cpu(el->l_count)); | |
1584 | ||
1585 | num_bytes = next_free - insert_index; | |
1586 | num_bytes *= sizeof(struct ocfs2_extent_rec); | |
1587 | memmove(&el->l_recs[insert_index + 1], | |
1588 | &el->l_recs[insert_index], | |
1589 | num_bytes); | |
1590 | } | |
1591 | ||
1592 | /* | |
1593 | * Either we had an empty extent, and need to re-increment or | |
1594 | * there was no empty extent on a non full rightmost leaf node, | |
1595 | * in which case we still need to increment. | |
1596 | */ | |
1597 | next_free++; | |
1598 | el->l_next_free_rec = cpu_to_le16(next_free); | |
1599 | /* | |
1600 | * Make sure none of the math above just messed up our tree. | |
1601 | */ | |
1602 | BUG_ON(le16_to_cpu(el->l_next_free_rec) > le16_to_cpu(el->l_count)); | |
1603 | ||
1604 | el->l_recs[insert_index] = *insert_rec; | |
1605 | ||
1606 | } | |
1607 | ||
328d5752 MF |
1608 | static void ocfs2_remove_empty_extent(struct ocfs2_extent_list *el) |
1609 | { | |
1610 | int size, num_recs = le16_to_cpu(el->l_next_free_rec); | |
1611 | ||
1612 | BUG_ON(num_recs == 0); | |
1613 | ||
1614 | if (ocfs2_is_empty_extent(&el->l_recs[0])) { | |
1615 | num_recs--; | |
1616 | size = num_recs * sizeof(struct ocfs2_extent_rec); | |
1617 | memmove(&el->l_recs[0], &el->l_recs[1], size); | |
1618 | memset(&el->l_recs[num_recs], 0, | |
1619 | sizeof(struct ocfs2_extent_rec)); | |
1620 | el->l_next_free_rec = cpu_to_le16(num_recs); | |
1621 | } | |
1622 | } | |
1623 | ||
dcd0538f MF |
1624 | /* |
1625 | * Create an empty extent record . | |
1626 | * | |
1627 | * l_next_free_rec may be updated. | |
1628 | * | |
1629 | * If an empty extent already exists do nothing. | |
1630 | */ | |
1631 | static void ocfs2_create_empty_extent(struct ocfs2_extent_list *el) | |
1632 | { | |
1633 | int next_free = le16_to_cpu(el->l_next_free_rec); | |
1634 | ||
e48edee2 MF |
1635 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); |
1636 | ||
dcd0538f MF |
1637 | if (next_free == 0) |
1638 | goto set_and_inc; | |
1639 | ||
1640 | if (ocfs2_is_empty_extent(&el->l_recs[0])) | |
1641 | return; | |
1642 | ||
1643 | mlog_bug_on_msg(el->l_count == el->l_next_free_rec, | |
1644 | "Asked to create an empty extent in a full list:\n" | |
1645 | "count = %u, tree depth = %u", | |
1646 | le16_to_cpu(el->l_count), | |
1647 | le16_to_cpu(el->l_tree_depth)); | |
1648 | ||
1649 | ocfs2_shift_records_right(el); | |
1650 | ||
1651 | set_and_inc: | |
1652 | le16_add_cpu(&el->l_next_free_rec, 1); | |
1653 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
1654 | } | |
1655 | ||
1656 | /* | |
1657 | * For a rotation which involves two leaf nodes, the "root node" is | |
1658 | * the lowest level tree node which contains a path to both leafs. This | |
1659 | * resulting set of information can be used to form a complete "subtree" | |
1660 | * | |
1661 | * This function is passed two full paths from the dinode down to a | |
1662 | * pair of adjacent leaves. It's task is to figure out which path | |
1663 | * index contains the subtree root - this can be the root index itself | |
1664 | * in a worst-case rotation. | |
1665 | * | |
1666 | * The array index of the subtree root is passed back. | |
1667 | */ | |
7dc02805 | 1668 | static int ocfs2_find_subtree_root(struct ocfs2_extent_tree *et, |
dcd0538f MF |
1669 | struct ocfs2_path *left, |
1670 | struct ocfs2_path *right) | |
1671 | { | |
1672 | int i = 0; | |
1673 | ||
1674 | /* | |
1675 | * Check that the caller passed in two paths from the same tree. | |
1676 | */ | |
1677 | BUG_ON(path_root_bh(left) != path_root_bh(right)); | |
1678 | ||
1679 | do { | |
1680 | i++; | |
1681 | ||
1682 | /* | |
1683 | * The caller didn't pass two adjacent paths. | |
1684 | */ | |
1685 | mlog_bug_on_msg(i > left->p_tree_depth, | |
7dc02805 | 1686 | "Owner %llu, left depth %u, right depth %u\n" |
dcd0538f | 1687 | "left leaf blk %llu, right leaf blk %llu\n", |
7dc02805 JB |
1688 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
1689 | left->p_tree_depth, right->p_tree_depth, | |
dcd0538f MF |
1690 | (unsigned long long)path_leaf_bh(left)->b_blocknr, |
1691 | (unsigned long long)path_leaf_bh(right)->b_blocknr); | |
1692 | } while (left->p_node[i].bh->b_blocknr == | |
1693 | right->p_node[i].bh->b_blocknr); | |
1694 | ||
1695 | return i - 1; | |
1696 | } | |
1697 | ||
1698 | typedef void (path_insert_t)(void *, struct buffer_head *); | |
1699 | ||
1700 | /* | |
1701 | * Traverse a btree path in search of cpos, starting at root_el. | |
1702 | * | |
1703 | * This code can be called with a cpos larger than the tree, in which | |
1704 | * case it will return the rightmost path. | |
1705 | */ | |
facdb77f | 1706 | static int __ocfs2_find_path(struct ocfs2_caching_info *ci, |
dcd0538f MF |
1707 | struct ocfs2_extent_list *root_el, u32 cpos, |
1708 | path_insert_t *func, void *data) | |
1709 | { | |
1710 | int i, ret = 0; | |
1711 | u32 range; | |
1712 | u64 blkno; | |
1713 | struct buffer_head *bh = NULL; | |
1714 | struct ocfs2_extent_block *eb; | |
1715 | struct ocfs2_extent_list *el; | |
1716 | struct ocfs2_extent_rec *rec; | |
dcd0538f MF |
1717 | |
1718 | el = root_el; | |
1719 | while (el->l_tree_depth) { | |
1720 | if (le16_to_cpu(el->l_next_free_rec) == 0) { | |
facdb77f JB |
1721 | ocfs2_error(ocfs2_metadata_cache_get_super(ci), |
1722 | "Owner %llu has empty extent list at " | |
dcd0538f | 1723 | "depth %u\n", |
facdb77f | 1724 | (unsigned long long)ocfs2_metadata_cache_owner(ci), |
dcd0538f MF |
1725 | le16_to_cpu(el->l_tree_depth)); |
1726 | ret = -EROFS; | |
1727 | goto out; | |
1728 | ||
1729 | } | |
1730 | ||
1731 | for(i = 0; i < le16_to_cpu(el->l_next_free_rec) - 1; i++) { | |
1732 | rec = &el->l_recs[i]; | |
1733 | ||
1734 | /* | |
1735 | * In the case that cpos is off the allocation | |
1736 | * tree, this should just wind up returning the | |
1737 | * rightmost record. | |
1738 | */ | |
1739 | range = le32_to_cpu(rec->e_cpos) + | |
e48edee2 | 1740 | ocfs2_rec_clusters(el, rec); |
dcd0538f MF |
1741 | if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range) |
1742 | break; | |
1743 | } | |
1744 | ||
1745 | blkno = le64_to_cpu(el->l_recs[i].e_blkno); | |
1746 | if (blkno == 0) { | |
facdb77f JB |
1747 | ocfs2_error(ocfs2_metadata_cache_get_super(ci), |
1748 | "Owner %llu has bad blkno in extent list " | |
dcd0538f | 1749 | "at depth %u (index %d)\n", |
facdb77f | 1750 | (unsigned long long)ocfs2_metadata_cache_owner(ci), |
dcd0538f MF |
1751 | le16_to_cpu(el->l_tree_depth), i); |
1752 | ret = -EROFS; | |
1753 | goto out; | |
1754 | } | |
1755 | ||
1756 | brelse(bh); | |
1757 | bh = NULL; | |
facdb77f | 1758 | ret = ocfs2_read_extent_block(ci, blkno, &bh); |
dcd0538f MF |
1759 | if (ret) { |
1760 | mlog_errno(ret); | |
1761 | goto out; | |
1762 | } | |
1763 | ||
1764 | eb = (struct ocfs2_extent_block *) bh->b_data; | |
1765 | el = &eb->h_list; | |
dcd0538f MF |
1766 | |
1767 | if (le16_to_cpu(el->l_next_free_rec) > | |
1768 | le16_to_cpu(el->l_count)) { | |
facdb77f JB |
1769 | ocfs2_error(ocfs2_metadata_cache_get_super(ci), |
1770 | "Owner %llu has bad count in extent list " | |
dcd0538f | 1771 | "at block %llu (next free=%u, count=%u)\n", |
facdb77f | 1772 | (unsigned long long)ocfs2_metadata_cache_owner(ci), |
dcd0538f MF |
1773 | (unsigned long long)bh->b_blocknr, |
1774 | le16_to_cpu(el->l_next_free_rec), | |
1775 | le16_to_cpu(el->l_count)); | |
1776 | ret = -EROFS; | |
1777 | goto out; | |
1778 | } | |
1779 | ||
1780 | if (func) | |
1781 | func(data, bh); | |
1782 | } | |
1783 | ||
1784 | out: | |
1785 | /* | |
1786 | * Catch any trailing bh that the loop didn't handle. | |
1787 | */ | |
1788 | brelse(bh); | |
1789 | ||
1790 | return ret; | |
1791 | } | |
1792 | ||
1793 | /* | |
1794 | * Given an initialized path (that is, it has a valid root extent | |
1795 | * list), this function will traverse the btree in search of the path | |
1796 | * which would contain cpos. | |
1797 | * | |
1798 | * The path traveled is recorded in the path structure. | |
1799 | * | |
1800 | * Note that this will not do any comparisons on leaf node extent | |
1801 | * records, so it will work fine in the case that we just added a tree | |
1802 | * branch. | |
1803 | */ | |
1804 | struct find_path_data { | |
1805 | int index; | |
1806 | struct ocfs2_path *path; | |
1807 | }; | |
1808 | static void find_path_ins(void *data, struct buffer_head *bh) | |
1809 | { | |
1810 | struct find_path_data *fp = data; | |
1811 | ||
1812 | get_bh(bh); | |
1813 | ocfs2_path_insert_eb(fp->path, fp->index, bh); | |
1814 | fp->index++; | |
1815 | } | |
facdb77f JB |
1816 | static int ocfs2_find_path(struct ocfs2_caching_info *ci, |
1817 | struct ocfs2_path *path, u32 cpos) | |
dcd0538f MF |
1818 | { |
1819 | struct find_path_data data; | |
1820 | ||
1821 | data.index = 1; | |
1822 | data.path = path; | |
facdb77f | 1823 | return __ocfs2_find_path(ci, path_root_el(path), cpos, |
dcd0538f MF |
1824 | find_path_ins, &data); |
1825 | } | |
1826 | ||
1827 | static void find_leaf_ins(void *data, struct buffer_head *bh) | |
1828 | { | |
1829 | struct ocfs2_extent_block *eb =(struct ocfs2_extent_block *)bh->b_data; | |
1830 | struct ocfs2_extent_list *el = &eb->h_list; | |
1831 | struct buffer_head **ret = data; | |
1832 | ||
1833 | /* We want to retain only the leaf block. */ | |
1834 | if (le16_to_cpu(el->l_tree_depth) == 0) { | |
1835 | get_bh(bh); | |
1836 | *ret = bh; | |
1837 | } | |
1838 | } | |
1839 | /* | |
1840 | * Find the leaf block in the tree which would contain cpos. No | |
1841 | * checking of the actual leaf is done. | |
1842 | * | |
1843 | * Some paths want to call this instead of allocating a path structure | |
1844 | * and calling ocfs2_find_path(). | |
1845 | * | |
1846 | * This function doesn't handle non btree extent lists. | |
1847 | */ | |
facdb77f JB |
1848 | int ocfs2_find_leaf(struct ocfs2_caching_info *ci, |
1849 | struct ocfs2_extent_list *root_el, u32 cpos, | |
1850 | struct buffer_head **leaf_bh) | |
dcd0538f MF |
1851 | { |
1852 | int ret; | |
1853 | struct buffer_head *bh = NULL; | |
1854 | ||
facdb77f | 1855 | ret = __ocfs2_find_path(ci, root_el, cpos, find_leaf_ins, &bh); |
dcd0538f MF |
1856 | if (ret) { |
1857 | mlog_errno(ret); | |
1858 | goto out; | |
1859 | } | |
1860 | ||
1861 | *leaf_bh = bh; | |
1862 | out: | |
1863 | return ret; | |
1864 | } | |
1865 | ||
1866 | /* | |
1867 | * Adjust the adjacent records (left_rec, right_rec) involved in a rotation. | |
1868 | * | |
1869 | * Basically, we've moved stuff around at the bottom of the tree and | |
1870 | * we need to fix up the extent records above the changes to reflect | |
1871 | * the new changes. | |
1872 | * | |
1873 | * left_rec: the record on the left. | |
1874 | * left_child_el: is the child list pointed to by left_rec | |
1875 | * right_rec: the record to the right of left_rec | |
1876 | * right_child_el: is the child list pointed to by right_rec | |
1877 | * | |
1878 | * By definition, this only works on interior nodes. | |
1879 | */ | |
1880 | static void ocfs2_adjust_adjacent_records(struct ocfs2_extent_rec *left_rec, | |
1881 | struct ocfs2_extent_list *left_child_el, | |
1882 | struct ocfs2_extent_rec *right_rec, | |
1883 | struct ocfs2_extent_list *right_child_el) | |
1884 | { | |
1885 | u32 left_clusters, right_end; | |
1886 | ||
1887 | /* | |
1888 | * Interior nodes never have holes. Their cpos is the cpos of | |
1889 | * the leftmost record in their child list. Their cluster | |
1890 | * count covers the full theoretical range of their child list | |
1891 | * - the range between their cpos and the cpos of the record | |
1892 | * immediately to their right. | |
1893 | */ | |
1894 | left_clusters = le32_to_cpu(right_child_el->l_recs[0].e_cpos); | |
82e12644 TM |
1895 | if (!ocfs2_rec_clusters(right_child_el, &right_child_el->l_recs[0])) { |
1896 | BUG_ON(right_child_el->l_tree_depth); | |
328d5752 MF |
1897 | BUG_ON(le16_to_cpu(right_child_el->l_next_free_rec) <= 1); |
1898 | left_clusters = le32_to_cpu(right_child_el->l_recs[1].e_cpos); | |
1899 | } | |
dcd0538f | 1900 | left_clusters -= le32_to_cpu(left_rec->e_cpos); |
e48edee2 | 1901 | left_rec->e_int_clusters = cpu_to_le32(left_clusters); |
dcd0538f MF |
1902 | |
1903 | /* | |
1904 | * Calculate the rightmost cluster count boundary before | |
e48edee2 | 1905 | * moving cpos - we will need to adjust clusters after |
dcd0538f MF |
1906 | * updating e_cpos to keep the same highest cluster count. |
1907 | */ | |
1908 | right_end = le32_to_cpu(right_rec->e_cpos); | |
e48edee2 | 1909 | right_end += le32_to_cpu(right_rec->e_int_clusters); |
dcd0538f MF |
1910 | |
1911 | right_rec->e_cpos = left_rec->e_cpos; | |
1912 | le32_add_cpu(&right_rec->e_cpos, left_clusters); | |
1913 | ||
1914 | right_end -= le32_to_cpu(right_rec->e_cpos); | |
e48edee2 | 1915 | right_rec->e_int_clusters = cpu_to_le32(right_end); |
dcd0538f MF |
1916 | } |
1917 | ||
1918 | /* | |
1919 | * Adjust the adjacent root node records involved in a | |
1920 | * rotation. left_el_blkno is passed in as a key so that we can easily | |
1921 | * find it's index in the root list. | |
1922 | */ | |
1923 | static void ocfs2_adjust_root_records(struct ocfs2_extent_list *root_el, | |
1924 | struct ocfs2_extent_list *left_el, | |
1925 | struct ocfs2_extent_list *right_el, | |
1926 | u64 left_el_blkno) | |
1927 | { | |
1928 | int i; | |
1929 | ||
1930 | BUG_ON(le16_to_cpu(root_el->l_tree_depth) <= | |
1931 | le16_to_cpu(left_el->l_tree_depth)); | |
1932 | ||
1933 | for(i = 0; i < le16_to_cpu(root_el->l_next_free_rec) - 1; i++) { | |
1934 | if (le64_to_cpu(root_el->l_recs[i].e_blkno) == left_el_blkno) | |
1935 | break; | |
1936 | } | |
1937 | ||
1938 | /* | |
1939 | * The path walking code should have never returned a root and | |
1940 | * two paths which are not adjacent. | |
1941 | */ | |
1942 | BUG_ON(i >= (le16_to_cpu(root_el->l_next_free_rec) - 1)); | |
1943 | ||
1944 | ocfs2_adjust_adjacent_records(&root_el->l_recs[i], left_el, | |
1945 | &root_el->l_recs[i + 1], right_el); | |
1946 | } | |
1947 | ||
1948 | /* | |
1949 | * We've changed a leaf block (in right_path) and need to reflect that | |
1950 | * change back up the subtree. | |
1951 | * | |
1952 | * This happens in multiple places: | |
1953 | * - When we've moved an extent record from the left path leaf to the right | |
1954 | * path leaf to make room for an empty extent in the left path leaf. | |
1955 | * - When our insert into the right path leaf is at the leftmost edge | |
1956 | * and requires an update of the path immediately to it's left. This | |
1957 | * can occur at the end of some types of rotation and appending inserts. | |
677b9752 TM |
1958 | * - When we've adjusted the last extent record in the left path leaf and the |
1959 | * 1st extent record in the right path leaf during cross extent block merge. | |
dcd0538f | 1960 | */ |
4619c73e | 1961 | static void ocfs2_complete_edge_insert(handle_t *handle, |
dcd0538f MF |
1962 | struct ocfs2_path *left_path, |
1963 | struct ocfs2_path *right_path, | |
1964 | int subtree_index) | |
1965 | { | |
1966 | int ret, i, idx; | |
1967 | struct ocfs2_extent_list *el, *left_el, *right_el; | |
1968 | struct ocfs2_extent_rec *left_rec, *right_rec; | |
1969 | struct buffer_head *root_bh = left_path->p_node[subtree_index].bh; | |
1970 | ||
1971 | /* | |
1972 | * Update the counts and position values within all the | |
1973 | * interior nodes to reflect the leaf rotation we just did. | |
1974 | * | |
1975 | * The root node is handled below the loop. | |
1976 | * | |
1977 | * We begin the loop with right_el and left_el pointing to the | |
1978 | * leaf lists and work our way up. | |
1979 | * | |
1980 | * NOTE: within this loop, left_el and right_el always refer | |
1981 | * to the *child* lists. | |
1982 | */ | |
1983 | left_el = path_leaf_el(left_path); | |
1984 | right_el = path_leaf_el(right_path); | |
1985 | for(i = left_path->p_tree_depth - 1; i > subtree_index; i--) { | |
1986 | mlog(0, "Adjust records at index %u\n", i); | |
1987 | ||
1988 | /* | |
1989 | * One nice property of knowing that all of these | |
1990 | * nodes are below the root is that we only deal with | |
1991 | * the leftmost right node record and the rightmost | |
1992 | * left node record. | |
1993 | */ | |
1994 | el = left_path->p_node[i].el; | |
1995 | idx = le16_to_cpu(left_el->l_next_free_rec) - 1; | |
1996 | left_rec = &el->l_recs[idx]; | |
1997 | ||
1998 | el = right_path->p_node[i].el; | |
1999 | right_rec = &el->l_recs[0]; | |
2000 | ||
2001 | ocfs2_adjust_adjacent_records(left_rec, left_el, right_rec, | |
2002 | right_el); | |
2003 | ||
2004 | ret = ocfs2_journal_dirty(handle, left_path->p_node[i].bh); | |
2005 | if (ret) | |
2006 | mlog_errno(ret); | |
2007 | ||
2008 | ret = ocfs2_journal_dirty(handle, right_path->p_node[i].bh); | |
2009 | if (ret) | |
2010 | mlog_errno(ret); | |
2011 | ||
2012 | /* | |
2013 | * Setup our list pointers now so that the current | |
2014 | * parents become children in the next iteration. | |
2015 | */ | |
2016 | left_el = left_path->p_node[i].el; | |
2017 | right_el = right_path->p_node[i].el; | |
2018 | } | |
2019 | ||
2020 | /* | |
2021 | * At the root node, adjust the two adjacent records which | |
2022 | * begin our path to the leaves. | |
2023 | */ | |
2024 | ||
2025 | el = left_path->p_node[subtree_index].el; | |
2026 | left_el = left_path->p_node[subtree_index + 1].el; | |
2027 | right_el = right_path->p_node[subtree_index + 1].el; | |
2028 | ||
2029 | ocfs2_adjust_root_records(el, left_el, right_el, | |
2030 | left_path->p_node[subtree_index + 1].bh->b_blocknr); | |
2031 | ||
2032 | root_bh = left_path->p_node[subtree_index].bh; | |
2033 | ||
2034 | ret = ocfs2_journal_dirty(handle, root_bh); | |
2035 | if (ret) | |
2036 | mlog_errno(ret); | |
2037 | } | |
2038 | ||
5c601aba JB |
2039 | static int ocfs2_rotate_subtree_right(handle_t *handle, |
2040 | struct ocfs2_extent_tree *et, | |
dcd0538f MF |
2041 | struct ocfs2_path *left_path, |
2042 | struct ocfs2_path *right_path, | |
2043 | int subtree_index) | |
2044 | { | |
2045 | int ret, i; | |
2046 | struct buffer_head *right_leaf_bh; | |
2047 | struct buffer_head *left_leaf_bh = NULL; | |
2048 | struct buffer_head *root_bh; | |
2049 | struct ocfs2_extent_list *right_el, *left_el; | |
2050 | struct ocfs2_extent_rec move_rec; | |
2051 | ||
2052 | left_leaf_bh = path_leaf_bh(left_path); | |
2053 | left_el = path_leaf_el(left_path); | |
2054 | ||
2055 | if (left_el->l_next_free_rec != left_el->l_count) { | |
5c601aba | 2056 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
dcd0538f MF |
2057 | "Inode %llu has non-full interior leaf node %llu" |
2058 | "(next free = %u)", | |
5c601aba | 2059 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
dcd0538f MF |
2060 | (unsigned long long)left_leaf_bh->b_blocknr, |
2061 | le16_to_cpu(left_el->l_next_free_rec)); | |
2062 | return -EROFS; | |
2063 | } | |
2064 | ||
2065 | /* | |
2066 | * This extent block may already have an empty record, so we | |
2067 | * return early if so. | |
2068 | */ | |
2069 | if (ocfs2_is_empty_extent(&left_el->l_recs[0])) | |
2070 | return 0; | |
2071 | ||
2072 | root_bh = left_path->p_node[subtree_index].bh; | |
2073 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
2074 | ||
5c601aba | 2075 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path, |
13723d00 | 2076 | subtree_index); |
dcd0538f MF |
2077 | if (ret) { |
2078 | mlog_errno(ret); | |
2079 | goto out; | |
2080 | } | |
2081 | ||
2082 | for(i = subtree_index + 1; i < path_num_items(right_path); i++) { | |
5c601aba | 2083 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 2084 | right_path, i); |
dcd0538f MF |
2085 | if (ret) { |
2086 | mlog_errno(ret); | |
2087 | goto out; | |
2088 | } | |
2089 | ||
5c601aba | 2090 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 2091 | left_path, i); |
dcd0538f MF |
2092 | if (ret) { |
2093 | mlog_errno(ret); | |
2094 | goto out; | |
2095 | } | |
2096 | } | |
2097 | ||
2098 | right_leaf_bh = path_leaf_bh(right_path); | |
2099 | right_el = path_leaf_el(right_path); | |
2100 | ||
2101 | /* This is a code error, not a disk corruption. */ | |
2102 | mlog_bug_on_msg(!right_el->l_next_free_rec, "Inode %llu: Rotate fails " | |
2103 | "because rightmost leaf block %llu is empty\n", | |
5c601aba | 2104 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
dcd0538f MF |
2105 | (unsigned long long)right_leaf_bh->b_blocknr); |
2106 | ||
2107 | ocfs2_create_empty_extent(right_el); | |
2108 | ||
2109 | ret = ocfs2_journal_dirty(handle, right_leaf_bh); | |
2110 | if (ret) { | |
2111 | mlog_errno(ret); | |
2112 | goto out; | |
2113 | } | |
2114 | ||
2115 | /* Do the copy now. */ | |
2116 | i = le16_to_cpu(left_el->l_next_free_rec) - 1; | |
2117 | move_rec = left_el->l_recs[i]; | |
2118 | right_el->l_recs[0] = move_rec; | |
2119 | ||
2120 | /* | |
2121 | * Clear out the record we just copied and shift everything | |
2122 | * over, leaving an empty extent in the left leaf. | |
2123 | * | |
2124 | * We temporarily subtract from next_free_rec so that the | |
2125 | * shift will lose the tail record (which is now defunct). | |
2126 | */ | |
2127 | le16_add_cpu(&left_el->l_next_free_rec, -1); | |
2128 | ocfs2_shift_records_right(left_el); | |
2129 | memset(&left_el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
2130 | le16_add_cpu(&left_el->l_next_free_rec, 1); | |
2131 | ||
2132 | ret = ocfs2_journal_dirty(handle, left_leaf_bh); | |
2133 | if (ret) { | |
2134 | mlog_errno(ret); | |
2135 | goto out; | |
2136 | } | |
2137 | ||
4619c73e JB |
2138 | ocfs2_complete_edge_insert(handle, left_path, right_path, |
2139 | subtree_index); | |
dcd0538f MF |
2140 | |
2141 | out: | |
2142 | return ret; | |
2143 | } | |
2144 | ||
2145 | /* | |
2146 | * Given a full path, determine what cpos value would return us a path | |
2147 | * containing the leaf immediately to the left of the current one. | |
2148 | * | |
2149 | * Will return zero if the path passed in is already the leftmost path. | |
2150 | */ | |
2151 | static int ocfs2_find_cpos_for_left_leaf(struct super_block *sb, | |
2152 | struct ocfs2_path *path, u32 *cpos) | |
2153 | { | |
2154 | int i, j, ret = 0; | |
2155 | u64 blkno; | |
2156 | struct ocfs2_extent_list *el; | |
2157 | ||
e48edee2 MF |
2158 | BUG_ON(path->p_tree_depth == 0); |
2159 | ||
dcd0538f MF |
2160 | *cpos = 0; |
2161 | ||
2162 | blkno = path_leaf_bh(path)->b_blocknr; | |
2163 | ||
2164 | /* Start at the tree node just above the leaf and work our way up. */ | |
2165 | i = path->p_tree_depth - 1; | |
2166 | while (i >= 0) { | |
2167 | el = path->p_node[i].el; | |
2168 | ||
2169 | /* | |
2170 | * Find the extent record just before the one in our | |
2171 | * path. | |
2172 | */ | |
2173 | for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) { | |
2174 | if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) { | |
2175 | if (j == 0) { | |
2176 | if (i == 0) { | |
2177 | /* | |
2178 | * We've determined that the | |
2179 | * path specified is already | |
2180 | * the leftmost one - return a | |
2181 | * cpos of zero. | |
2182 | */ | |
2183 | goto out; | |
2184 | } | |
2185 | /* | |
2186 | * The leftmost record points to our | |
2187 | * leaf - we need to travel up the | |
2188 | * tree one level. | |
2189 | */ | |
2190 | goto next_node; | |
2191 | } | |
2192 | ||
2193 | *cpos = le32_to_cpu(el->l_recs[j - 1].e_cpos); | |
e48edee2 MF |
2194 | *cpos = *cpos + ocfs2_rec_clusters(el, |
2195 | &el->l_recs[j - 1]); | |
2196 | *cpos = *cpos - 1; | |
dcd0538f MF |
2197 | goto out; |
2198 | } | |
2199 | } | |
2200 | ||
2201 | /* | |
2202 | * If we got here, we never found a valid node where | |
2203 | * the tree indicated one should be. | |
2204 | */ | |
2205 | ocfs2_error(sb, | |
2206 | "Invalid extent tree at extent block %llu\n", | |
2207 | (unsigned long long)blkno); | |
2208 | ret = -EROFS; | |
2209 | goto out; | |
2210 | ||
2211 | next_node: | |
2212 | blkno = path->p_node[i].bh->b_blocknr; | |
2213 | i--; | |
2214 | } | |
2215 | ||
2216 | out: | |
2217 | return ret; | |
2218 | } | |
2219 | ||
328d5752 MF |
2220 | /* |
2221 | * Extend the transaction by enough credits to complete the rotation, | |
2222 | * and still leave at least the original number of credits allocated | |
2223 | * to this transaction. | |
2224 | */ | |
dcd0538f | 2225 | static int ocfs2_extend_rotate_transaction(handle_t *handle, int subtree_depth, |
328d5752 | 2226 | int op_credits, |
dcd0538f MF |
2227 | struct ocfs2_path *path) |
2228 | { | |
328d5752 | 2229 | int credits = (path->p_tree_depth - subtree_depth) * 2 + 1 + op_credits; |
dcd0538f MF |
2230 | |
2231 | if (handle->h_buffer_credits < credits) | |
2232 | return ocfs2_extend_trans(handle, credits); | |
2233 | ||
2234 | return 0; | |
2235 | } | |
2236 | ||
2237 | /* | |
2238 | * Trap the case where we're inserting into the theoretical range past | |
2239 | * the _actual_ left leaf range. Otherwise, we'll rotate a record | |
2240 | * whose cpos is less than ours into the right leaf. | |
2241 | * | |
2242 | * It's only necessary to look at the rightmost record of the left | |
2243 | * leaf because the logic that calls us should ensure that the | |
2244 | * theoretical ranges in the path components above the leaves are | |
2245 | * correct. | |
2246 | */ | |
2247 | static int ocfs2_rotate_requires_path_adjustment(struct ocfs2_path *left_path, | |
2248 | u32 insert_cpos) | |
2249 | { | |
2250 | struct ocfs2_extent_list *left_el; | |
2251 | struct ocfs2_extent_rec *rec; | |
2252 | int next_free; | |
2253 | ||
2254 | left_el = path_leaf_el(left_path); | |
2255 | next_free = le16_to_cpu(left_el->l_next_free_rec); | |
2256 | rec = &left_el->l_recs[next_free - 1]; | |
2257 | ||
2258 | if (insert_cpos > le32_to_cpu(rec->e_cpos)) | |
2259 | return 1; | |
2260 | return 0; | |
2261 | } | |
2262 | ||
328d5752 MF |
2263 | static int ocfs2_leftmost_rec_contains(struct ocfs2_extent_list *el, u32 cpos) |
2264 | { | |
2265 | int next_free = le16_to_cpu(el->l_next_free_rec); | |
2266 | unsigned int range; | |
2267 | struct ocfs2_extent_rec *rec; | |
2268 | ||
2269 | if (next_free == 0) | |
2270 | return 0; | |
2271 | ||
2272 | rec = &el->l_recs[0]; | |
2273 | if (ocfs2_is_empty_extent(rec)) { | |
2274 | /* Empty list. */ | |
2275 | if (next_free == 1) | |
2276 | return 0; | |
2277 | rec = &el->l_recs[1]; | |
2278 | } | |
2279 | ||
2280 | range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
2281 | if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range) | |
2282 | return 1; | |
2283 | return 0; | |
2284 | } | |
2285 | ||
dcd0538f MF |
2286 | /* |
2287 | * Rotate all the records in a btree right one record, starting at insert_cpos. | |
2288 | * | |
2289 | * The path to the rightmost leaf should be passed in. | |
2290 | * | |
2291 | * The array is assumed to be large enough to hold an entire path (tree depth). | |
2292 | * | |
2293 | * Upon succesful return from this function: | |
2294 | * | |
2295 | * - The 'right_path' array will contain a path to the leaf block | |
2296 | * whose range contains e_cpos. | |
2297 | * - That leaf block will have a single empty extent in list index 0. | |
2298 | * - In the case that the rotation requires a post-insert update, | |
2299 | * *ret_left_path will contain a valid path which can be passed to | |
2300 | * ocfs2_insert_path(). | |
2301 | */ | |
1bbf0b8d | 2302 | static int ocfs2_rotate_tree_right(handle_t *handle, |
5c601aba | 2303 | struct ocfs2_extent_tree *et, |
328d5752 | 2304 | enum ocfs2_split_type split, |
dcd0538f MF |
2305 | u32 insert_cpos, |
2306 | struct ocfs2_path *right_path, | |
2307 | struct ocfs2_path **ret_left_path) | |
2308 | { | |
328d5752 | 2309 | int ret, start, orig_credits = handle->h_buffer_credits; |
dcd0538f MF |
2310 | u32 cpos; |
2311 | struct ocfs2_path *left_path = NULL; | |
5c601aba | 2312 | struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci); |
dcd0538f MF |
2313 | |
2314 | *ret_left_path = NULL; | |
2315 | ||
ffdd7a54 | 2316 | left_path = ocfs2_new_path_from_path(right_path); |
dcd0538f MF |
2317 | if (!left_path) { |
2318 | ret = -ENOMEM; | |
2319 | mlog_errno(ret); | |
2320 | goto out; | |
2321 | } | |
2322 | ||
5c601aba | 2323 | ret = ocfs2_find_cpos_for_left_leaf(sb, right_path, &cpos); |
dcd0538f MF |
2324 | if (ret) { |
2325 | mlog_errno(ret); | |
2326 | goto out; | |
2327 | } | |
2328 | ||
2329 | mlog(0, "Insert: %u, first left path cpos: %u\n", insert_cpos, cpos); | |
2330 | ||
2331 | /* | |
2332 | * What we want to do here is: | |
2333 | * | |
2334 | * 1) Start with the rightmost path. | |
2335 | * | |
2336 | * 2) Determine a path to the leaf block directly to the left | |
2337 | * of that leaf. | |
2338 | * | |
2339 | * 3) Determine the 'subtree root' - the lowest level tree node | |
2340 | * which contains a path to both leaves. | |
2341 | * | |
2342 | * 4) Rotate the subtree. | |
2343 | * | |
2344 | * 5) Find the next subtree by considering the left path to be | |
2345 | * the new right path. | |
2346 | * | |
2347 | * The check at the top of this while loop also accepts | |
2348 | * insert_cpos == cpos because cpos is only a _theoretical_ | |
2349 | * value to get us the left path - insert_cpos might very well | |
2350 | * be filling that hole. | |
2351 | * | |
2352 | * Stop at a cpos of '0' because we either started at the | |
2353 | * leftmost branch (i.e., a tree with one branch and a | |
2354 | * rotation inside of it), or we've gone as far as we can in | |
2355 | * rotating subtrees. | |
2356 | */ | |
2357 | while (cpos && insert_cpos <= cpos) { | |
2358 | mlog(0, "Rotating a tree: ins. cpos: %u, left path cpos: %u\n", | |
2359 | insert_cpos, cpos); | |
2360 | ||
5c601aba | 2361 | ret = ocfs2_find_path(et->et_ci, left_path, cpos); |
dcd0538f MF |
2362 | if (ret) { |
2363 | mlog_errno(ret); | |
2364 | goto out; | |
2365 | } | |
2366 | ||
2367 | mlog_bug_on_msg(path_leaf_bh(left_path) == | |
2368 | path_leaf_bh(right_path), | |
5c601aba | 2369 | "Owner %llu: error during insert of %u " |
dcd0538f MF |
2370 | "(left path cpos %u) results in two identical " |
2371 | "paths ending at %llu\n", | |
5c601aba JB |
2372 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
2373 | insert_cpos, cpos, | |
dcd0538f MF |
2374 | (unsigned long long) |
2375 | path_leaf_bh(left_path)->b_blocknr); | |
2376 | ||
328d5752 MF |
2377 | if (split == SPLIT_NONE && |
2378 | ocfs2_rotate_requires_path_adjustment(left_path, | |
dcd0538f | 2379 | insert_cpos)) { |
dcd0538f MF |
2380 | |
2381 | /* | |
2382 | * We've rotated the tree as much as we | |
2383 | * should. The rest is up to | |
2384 | * ocfs2_insert_path() to complete, after the | |
2385 | * record insertion. We indicate this | |
2386 | * situation by returning the left path. | |
2387 | * | |
2388 | * The reason we don't adjust the records here | |
2389 | * before the record insert is that an error | |
2390 | * later might break the rule where a parent | |
2391 | * record e_cpos will reflect the actual | |
2392 | * e_cpos of the 1st nonempty record of the | |
2393 | * child list. | |
2394 | */ | |
2395 | *ret_left_path = left_path; | |
2396 | goto out_ret_path; | |
2397 | } | |
2398 | ||
7dc02805 | 2399 | start = ocfs2_find_subtree_root(et, left_path, right_path); |
dcd0538f MF |
2400 | |
2401 | mlog(0, "Subtree root at index %d (blk %llu, depth %d)\n", | |
2402 | start, | |
2403 | (unsigned long long) right_path->p_node[start].bh->b_blocknr, | |
2404 | right_path->p_tree_depth); | |
2405 | ||
2406 | ret = ocfs2_extend_rotate_transaction(handle, start, | |
328d5752 | 2407 | orig_credits, right_path); |
dcd0538f MF |
2408 | if (ret) { |
2409 | mlog_errno(ret); | |
2410 | goto out; | |
2411 | } | |
2412 | ||
5c601aba | 2413 | ret = ocfs2_rotate_subtree_right(handle, et, left_path, |
dcd0538f MF |
2414 | right_path, start); |
2415 | if (ret) { | |
2416 | mlog_errno(ret); | |
2417 | goto out; | |
2418 | } | |
2419 | ||
328d5752 MF |
2420 | if (split != SPLIT_NONE && |
2421 | ocfs2_leftmost_rec_contains(path_leaf_el(right_path), | |
2422 | insert_cpos)) { | |
2423 | /* | |
2424 | * A rotate moves the rightmost left leaf | |
2425 | * record over to the leftmost right leaf | |
2426 | * slot. If we're doing an extent split | |
2427 | * instead of a real insert, then we have to | |
2428 | * check that the extent to be split wasn't | |
2429 | * just moved over. If it was, then we can | |
2430 | * exit here, passing left_path back - | |
2431 | * ocfs2_split_extent() is smart enough to | |
2432 | * search both leaves. | |
2433 | */ | |
2434 | *ret_left_path = left_path; | |
2435 | goto out_ret_path; | |
2436 | } | |
2437 | ||
dcd0538f MF |
2438 | /* |
2439 | * There is no need to re-read the next right path | |
2440 | * as we know that it'll be our current left | |
2441 | * path. Optimize by copying values instead. | |
2442 | */ | |
2443 | ocfs2_mv_path(right_path, left_path); | |
2444 | ||
5c601aba | 2445 | ret = ocfs2_find_cpos_for_left_leaf(sb, right_path, &cpos); |
dcd0538f MF |
2446 | if (ret) { |
2447 | mlog_errno(ret); | |
2448 | goto out; | |
2449 | } | |
2450 | } | |
2451 | ||
2452 | out: | |
2453 | ocfs2_free_path(left_path); | |
2454 | ||
2455 | out_ret_path: | |
2456 | return ret; | |
2457 | } | |
2458 | ||
09106bae JB |
2459 | static int ocfs2_update_edge_lengths(handle_t *handle, |
2460 | struct ocfs2_extent_tree *et, | |
3c5e1068 | 2461 | int subtree_index, struct ocfs2_path *path) |
dcd0538f | 2462 | { |
3c5e1068 | 2463 | int i, idx, ret; |
dcd0538f | 2464 | struct ocfs2_extent_rec *rec; |
328d5752 MF |
2465 | struct ocfs2_extent_list *el; |
2466 | struct ocfs2_extent_block *eb; | |
2467 | u32 range; | |
dcd0538f | 2468 | |
3c5e1068 TM |
2469 | /* |
2470 | * In normal tree rotation process, we will never touch the | |
2471 | * tree branch above subtree_index and ocfs2_extend_rotate_transaction | |
2472 | * doesn't reserve the credits for them either. | |
2473 | * | |
2474 | * But we do have a special case here which will update the rightmost | |
2475 | * records for all the bh in the path. | |
2476 | * So we have to allocate extra credits and access them. | |
2477 | */ | |
2478 | ret = ocfs2_extend_trans(handle, | |
2479 | handle->h_buffer_credits + subtree_index); | |
2480 | if (ret) { | |
2481 | mlog_errno(ret); | |
2482 | goto out; | |
2483 | } | |
2484 | ||
09106bae | 2485 | ret = ocfs2_journal_access_path(et->et_ci, handle, path); |
3c5e1068 TM |
2486 | if (ret) { |
2487 | mlog_errno(ret); | |
2488 | goto out; | |
2489 | } | |
2490 | ||
328d5752 MF |
2491 | /* Path should always be rightmost. */ |
2492 | eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data; | |
2493 | BUG_ON(eb->h_next_leaf_blk != 0ULL); | |
dcd0538f | 2494 | |
328d5752 MF |
2495 | el = &eb->h_list; |
2496 | BUG_ON(le16_to_cpu(el->l_next_free_rec) == 0); | |
2497 | idx = le16_to_cpu(el->l_next_free_rec) - 1; | |
2498 | rec = &el->l_recs[idx]; | |
2499 | range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
dcd0538f | 2500 | |
328d5752 MF |
2501 | for (i = 0; i < path->p_tree_depth; i++) { |
2502 | el = path->p_node[i].el; | |
2503 | idx = le16_to_cpu(el->l_next_free_rec) - 1; | |
2504 | rec = &el->l_recs[idx]; | |
dcd0538f | 2505 | |
328d5752 MF |
2506 | rec->e_int_clusters = cpu_to_le32(range); |
2507 | le32_add_cpu(&rec->e_int_clusters, -le32_to_cpu(rec->e_cpos)); | |
dcd0538f | 2508 | |
328d5752 | 2509 | ocfs2_journal_dirty(handle, path->p_node[i].bh); |
dcd0538f | 2510 | } |
3c5e1068 TM |
2511 | out: |
2512 | return ret; | |
dcd0538f MF |
2513 | } |
2514 | ||
6641b0ce JB |
2515 | static void ocfs2_unlink_path(handle_t *handle, |
2516 | struct ocfs2_extent_tree *et, | |
328d5752 MF |
2517 | struct ocfs2_cached_dealloc_ctxt *dealloc, |
2518 | struct ocfs2_path *path, int unlink_start) | |
dcd0538f | 2519 | { |
328d5752 MF |
2520 | int ret, i; |
2521 | struct ocfs2_extent_block *eb; | |
2522 | struct ocfs2_extent_list *el; | |
2523 | struct buffer_head *bh; | |
2524 | ||
2525 | for(i = unlink_start; i < path_num_items(path); i++) { | |
2526 | bh = path->p_node[i].bh; | |
2527 | ||
2528 | eb = (struct ocfs2_extent_block *)bh->b_data; | |
2529 | /* | |
2530 | * Not all nodes might have had their final count | |
2531 | * decremented by the caller - handle this here. | |
2532 | */ | |
2533 | el = &eb->h_list; | |
2534 | if (le16_to_cpu(el->l_next_free_rec) > 1) { | |
2535 | mlog(ML_ERROR, | |
2536 | "Inode %llu, attempted to remove extent block " | |
2537 | "%llu with %u records\n", | |
6641b0ce | 2538 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
328d5752 MF |
2539 | (unsigned long long)le64_to_cpu(eb->h_blkno), |
2540 | le16_to_cpu(el->l_next_free_rec)); | |
2541 | ||
2542 | ocfs2_journal_dirty(handle, bh); | |
6641b0ce | 2543 | ocfs2_remove_from_cache(et->et_ci, bh); |
328d5752 MF |
2544 | continue; |
2545 | } | |
2546 | ||
2547 | el->l_next_free_rec = 0; | |
2548 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
2549 | ||
2550 | ocfs2_journal_dirty(handle, bh); | |
2551 | ||
2552 | ret = ocfs2_cache_extent_block_free(dealloc, eb); | |
2553 | if (ret) | |
2554 | mlog_errno(ret); | |
2555 | ||
6641b0ce | 2556 | ocfs2_remove_from_cache(et->et_ci, bh); |
328d5752 | 2557 | } |
dcd0538f MF |
2558 | } |
2559 | ||
6641b0ce JB |
2560 | static void ocfs2_unlink_subtree(handle_t *handle, |
2561 | struct ocfs2_extent_tree *et, | |
328d5752 MF |
2562 | struct ocfs2_path *left_path, |
2563 | struct ocfs2_path *right_path, | |
2564 | int subtree_index, | |
2565 | struct ocfs2_cached_dealloc_ctxt *dealloc) | |
dcd0538f | 2566 | { |
328d5752 MF |
2567 | int i; |
2568 | struct buffer_head *root_bh = left_path->p_node[subtree_index].bh; | |
2569 | struct ocfs2_extent_list *root_el = left_path->p_node[subtree_index].el; | |
dcd0538f | 2570 | struct ocfs2_extent_list *el; |
328d5752 | 2571 | struct ocfs2_extent_block *eb; |
dcd0538f | 2572 | |
328d5752 | 2573 | el = path_leaf_el(left_path); |
dcd0538f | 2574 | |
328d5752 | 2575 | eb = (struct ocfs2_extent_block *)right_path->p_node[subtree_index + 1].bh->b_data; |
e48edee2 | 2576 | |
328d5752 MF |
2577 | for(i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++) |
2578 | if (root_el->l_recs[i].e_blkno == eb->h_blkno) | |
2579 | break; | |
dcd0538f | 2580 | |
328d5752 | 2581 | BUG_ON(i >= le16_to_cpu(root_el->l_next_free_rec)); |
dcd0538f | 2582 | |
328d5752 MF |
2583 | memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec)); |
2584 | le16_add_cpu(&root_el->l_next_free_rec, -1); | |
2585 | ||
2586 | eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data; | |
2587 | eb->h_next_leaf_blk = 0; | |
2588 | ||
2589 | ocfs2_journal_dirty(handle, root_bh); | |
2590 | ocfs2_journal_dirty(handle, path_leaf_bh(left_path)); | |
2591 | ||
6641b0ce | 2592 | ocfs2_unlink_path(handle, et, dealloc, right_path, |
328d5752 MF |
2593 | subtree_index + 1); |
2594 | } | |
2595 | ||
1e2dd63f JB |
2596 | static int ocfs2_rotate_subtree_left(handle_t *handle, |
2597 | struct ocfs2_extent_tree *et, | |
328d5752 MF |
2598 | struct ocfs2_path *left_path, |
2599 | struct ocfs2_path *right_path, | |
2600 | int subtree_index, | |
2601 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
1e2dd63f | 2602 | int *deleted) |
328d5752 MF |
2603 | { |
2604 | int ret, i, del_right_subtree = 0, right_has_empty = 0; | |
e7d4cb6b | 2605 | struct buffer_head *root_bh, *et_root_bh = path_root_bh(right_path); |
328d5752 MF |
2606 | struct ocfs2_extent_list *right_leaf_el, *left_leaf_el; |
2607 | struct ocfs2_extent_block *eb; | |
2608 | ||
2609 | *deleted = 0; | |
2610 | ||
2611 | right_leaf_el = path_leaf_el(right_path); | |
2612 | left_leaf_el = path_leaf_el(left_path); | |
2613 | root_bh = left_path->p_node[subtree_index].bh; | |
2614 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
2615 | ||
2616 | if (!ocfs2_is_empty_extent(&left_leaf_el->l_recs[0])) | |
2617 | return 0; | |
dcd0538f | 2618 | |
328d5752 MF |
2619 | eb = (struct ocfs2_extent_block *)path_leaf_bh(right_path)->b_data; |
2620 | if (ocfs2_is_empty_extent(&right_leaf_el->l_recs[0])) { | |
dcd0538f | 2621 | /* |
328d5752 MF |
2622 | * It's legal for us to proceed if the right leaf is |
2623 | * the rightmost one and it has an empty extent. There | |
2624 | * are two cases to handle - whether the leaf will be | |
2625 | * empty after removal or not. If the leaf isn't empty | |
2626 | * then just remove the empty extent up front. The | |
2627 | * next block will handle empty leaves by flagging | |
2628 | * them for unlink. | |
2629 | * | |
2630 | * Non rightmost leaves will throw -EAGAIN and the | |
2631 | * caller can manually move the subtree and retry. | |
dcd0538f | 2632 | */ |
dcd0538f | 2633 | |
328d5752 MF |
2634 | if (eb->h_next_leaf_blk != 0ULL) |
2635 | return -EAGAIN; | |
2636 | ||
2637 | if (le16_to_cpu(right_leaf_el->l_next_free_rec) > 1) { | |
1e2dd63f | 2638 | ret = ocfs2_journal_access_eb(handle, et->et_ci, |
13723d00 JB |
2639 | path_leaf_bh(right_path), |
2640 | OCFS2_JOURNAL_ACCESS_WRITE); | |
dcd0538f MF |
2641 | if (ret) { |
2642 | mlog_errno(ret); | |
2643 | goto out; | |
2644 | } | |
2645 | ||
328d5752 MF |
2646 | ocfs2_remove_empty_extent(right_leaf_el); |
2647 | } else | |
2648 | right_has_empty = 1; | |
dcd0538f MF |
2649 | } |
2650 | ||
328d5752 MF |
2651 | if (eb->h_next_leaf_blk == 0ULL && |
2652 | le16_to_cpu(right_leaf_el->l_next_free_rec) == 1) { | |
2653 | /* | |
2654 | * We have to update i_last_eb_blk during the meta | |
2655 | * data delete. | |
2656 | */ | |
d9a0a1f8 | 2657 | ret = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 2658 | OCFS2_JOURNAL_ACCESS_WRITE); |
328d5752 MF |
2659 | if (ret) { |
2660 | mlog_errno(ret); | |
2661 | goto out; | |
2662 | } | |
2663 | ||
2664 | del_right_subtree = 1; | |
2665 | } | |
2666 | ||
2667 | /* | |
2668 | * Getting here with an empty extent in the right path implies | |
2669 | * that it's the rightmost path and will be deleted. | |
2670 | */ | |
2671 | BUG_ON(right_has_empty && !del_right_subtree); | |
2672 | ||
1e2dd63f | 2673 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path, |
13723d00 | 2674 | subtree_index); |
328d5752 MF |
2675 | if (ret) { |
2676 | mlog_errno(ret); | |
2677 | goto out; | |
2678 | } | |
2679 | ||
2680 | for(i = subtree_index + 1; i < path_num_items(right_path); i++) { | |
1e2dd63f | 2681 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 2682 | right_path, i); |
328d5752 MF |
2683 | if (ret) { |
2684 | mlog_errno(ret); | |
2685 | goto out; | |
2686 | } | |
2687 | ||
1e2dd63f | 2688 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 2689 | left_path, i); |
328d5752 MF |
2690 | if (ret) { |
2691 | mlog_errno(ret); | |
2692 | goto out; | |
2693 | } | |
2694 | } | |
2695 | ||
2696 | if (!right_has_empty) { | |
2697 | /* | |
2698 | * Only do this if we're moving a real | |
2699 | * record. Otherwise, the action is delayed until | |
2700 | * after removal of the right path in which case we | |
2701 | * can do a simple shift to remove the empty extent. | |
2702 | */ | |
2703 | ocfs2_rotate_leaf(left_leaf_el, &right_leaf_el->l_recs[0]); | |
2704 | memset(&right_leaf_el->l_recs[0], 0, | |
2705 | sizeof(struct ocfs2_extent_rec)); | |
2706 | } | |
2707 | if (eb->h_next_leaf_blk == 0ULL) { | |
2708 | /* | |
2709 | * Move recs over to get rid of empty extent, decrease | |
2710 | * next_free. This is allowed to remove the last | |
2711 | * extent in our leaf (setting l_next_free_rec to | |
2712 | * zero) - the delete code below won't care. | |
2713 | */ | |
2714 | ocfs2_remove_empty_extent(right_leaf_el); | |
2715 | } | |
2716 | ||
2717 | ret = ocfs2_journal_dirty(handle, path_leaf_bh(left_path)); | |
2718 | if (ret) | |
2719 | mlog_errno(ret); | |
2720 | ret = ocfs2_journal_dirty(handle, path_leaf_bh(right_path)); | |
2721 | if (ret) | |
2722 | mlog_errno(ret); | |
2723 | ||
2724 | if (del_right_subtree) { | |
6641b0ce | 2725 | ocfs2_unlink_subtree(handle, et, left_path, right_path, |
328d5752 | 2726 | subtree_index, dealloc); |
09106bae | 2727 | ret = ocfs2_update_edge_lengths(handle, et, subtree_index, |
3c5e1068 TM |
2728 | left_path); |
2729 | if (ret) { | |
2730 | mlog_errno(ret); | |
2731 | goto out; | |
2732 | } | |
328d5752 MF |
2733 | |
2734 | eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data; | |
35dc0aa3 | 2735 | ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno)); |
328d5752 MF |
2736 | |
2737 | /* | |
2738 | * Removal of the extent in the left leaf was skipped | |
2739 | * above so we could delete the right path | |
2740 | * 1st. | |
2741 | */ | |
2742 | if (right_has_empty) | |
2743 | ocfs2_remove_empty_extent(left_leaf_el); | |
2744 | ||
e7d4cb6b | 2745 | ret = ocfs2_journal_dirty(handle, et_root_bh); |
328d5752 MF |
2746 | if (ret) |
2747 | mlog_errno(ret); | |
2748 | ||
2749 | *deleted = 1; | |
2750 | } else | |
4619c73e | 2751 | ocfs2_complete_edge_insert(handle, left_path, right_path, |
328d5752 MF |
2752 | subtree_index); |
2753 | ||
2754 | out: | |
2755 | return ret; | |
2756 | } | |
2757 | ||
2758 | /* | |
2759 | * Given a full path, determine what cpos value would return us a path | |
2760 | * containing the leaf immediately to the right of the current one. | |
2761 | * | |
2762 | * Will return zero if the path passed in is already the rightmost path. | |
2763 | * | |
2764 | * This looks similar, but is subtly different to | |
2765 | * ocfs2_find_cpos_for_left_leaf(). | |
2766 | */ | |
2767 | static int ocfs2_find_cpos_for_right_leaf(struct super_block *sb, | |
2768 | struct ocfs2_path *path, u32 *cpos) | |
2769 | { | |
2770 | int i, j, ret = 0; | |
2771 | u64 blkno; | |
2772 | struct ocfs2_extent_list *el; | |
2773 | ||
2774 | *cpos = 0; | |
2775 | ||
2776 | if (path->p_tree_depth == 0) | |
2777 | return 0; | |
2778 | ||
2779 | blkno = path_leaf_bh(path)->b_blocknr; | |
2780 | ||
2781 | /* Start at the tree node just above the leaf and work our way up. */ | |
2782 | i = path->p_tree_depth - 1; | |
2783 | while (i >= 0) { | |
2784 | int next_free; | |
2785 | ||
2786 | el = path->p_node[i].el; | |
2787 | ||
2788 | /* | |
2789 | * Find the extent record just after the one in our | |
2790 | * path. | |
2791 | */ | |
2792 | next_free = le16_to_cpu(el->l_next_free_rec); | |
2793 | for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) { | |
2794 | if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) { | |
2795 | if (j == (next_free - 1)) { | |
2796 | if (i == 0) { | |
2797 | /* | |
2798 | * We've determined that the | |
2799 | * path specified is already | |
2800 | * the rightmost one - return a | |
2801 | * cpos of zero. | |
2802 | */ | |
2803 | goto out; | |
2804 | } | |
2805 | /* | |
2806 | * The rightmost record points to our | |
2807 | * leaf - we need to travel up the | |
2808 | * tree one level. | |
2809 | */ | |
2810 | goto next_node; | |
2811 | } | |
2812 | ||
2813 | *cpos = le32_to_cpu(el->l_recs[j + 1].e_cpos); | |
2814 | goto out; | |
2815 | } | |
2816 | } | |
2817 | ||
2818 | /* | |
2819 | * If we got here, we never found a valid node where | |
2820 | * the tree indicated one should be. | |
2821 | */ | |
2822 | ocfs2_error(sb, | |
2823 | "Invalid extent tree at extent block %llu\n", | |
2824 | (unsigned long long)blkno); | |
2825 | ret = -EROFS; | |
2826 | goto out; | |
2827 | ||
2828 | next_node: | |
2829 | blkno = path->p_node[i].bh->b_blocknr; | |
2830 | i--; | |
2831 | } | |
2832 | ||
2833 | out: | |
2834 | return ret; | |
2835 | } | |
2836 | ||
70f18c08 JB |
2837 | static int ocfs2_rotate_rightmost_leaf_left(handle_t *handle, |
2838 | struct ocfs2_extent_tree *et, | |
13723d00 | 2839 | struct ocfs2_path *path) |
328d5752 MF |
2840 | { |
2841 | int ret; | |
13723d00 JB |
2842 | struct buffer_head *bh = path_leaf_bh(path); |
2843 | struct ocfs2_extent_list *el = path_leaf_el(path); | |
328d5752 MF |
2844 | |
2845 | if (!ocfs2_is_empty_extent(&el->l_recs[0])) | |
2846 | return 0; | |
2847 | ||
70f18c08 | 2848 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, path, |
13723d00 | 2849 | path_num_items(path) - 1); |
328d5752 MF |
2850 | if (ret) { |
2851 | mlog_errno(ret); | |
2852 | goto out; | |
2853 | } | |
2854 | ||
2855 | ocfs2_remove_empty_extent(el); | |
2856 | ||
2857 | ret = ocfs2_journal_dirty(handle, bh); | |
2858 | if (ret) | |
2859 | mlog_errno(ret); | |
2860 | ||
2861 | out: | |
2862 | return ret; | |
2863 | } | |
2864 | ||
e46f74dc JB |
2865 | static int __ocfs2_rotate_tree_left(handle_t *handle, |
2866 | struct ocfs2_extent_tree *et, | |
2867 | int orig_credits, | |
328d5752 MF |
2868 | struct ocfs2_path *path, |
2869 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
e46f74dc | 2870 | struct ocfs2_path **empty_extent_path) |
328d5752 MF |
2871 | { |
2872 | int ret, subtree_root, deleted; | |
2873 | u32 right_cpos; | |
2874 | struct ocfs2_path *left_path = NULL; | |
2875 | struct ocfs2_path *right_path = NULL; | |
e46f74dc | 2876 | struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci); |
328d5752 MF |
2877 | |
2878 | BUG_ON(!ocfs2_is_empty_extent(&(path_leaf_el(path)->l_recs[0]))); | |
2879 | ||
2880 | *empty_extent_path = NULL; | |
2881 | ||
e46f74dc | 2882 | ret = ocfs2_find_cpos_for_right_leaf(sb, path, &right_cpos); |
328d5752 MF |
2883 | if (ret) { |
2884 | mlog_errno(ret); | |
2885 | goto out; | |
2886 | } | |
2887 | ||
ffdd7a54 | 2888 | left_path = ocfs2_new_path_from_path(path); |
328d5752 MF |
2889 | if (!left_path) { |
2890 | ret = -ENOMEM; | |
2891 | mlog_errno(ret); | |
2892 | goto out; | |
2893 | } | |
2894 | ||
2895 | ocfs2_cp_path(left_path, path); | |
2896 | ||
ffdd7a54 | 2897 | right_path = ocfs2_new_path_from_path(path); |
328d5752 MF |
2898 | if (!right_path) { |
2899 | ret = -ENOMEM; | |
2900 | mlog_errno(ret); | |
2901 | goto out; | |
2902 | } | |
2903 | ||
2904 | while (right_cpos) { | |
facdb77f | 2905 | ret = ocfs2_find_path(et->et_ci, right_path, right_cpos); |
328d5752 MF |
2906 | if (ret) { |
2907 | mlog_errno(ret); | |
2908 | goto out; | |
2909 | } | |
2910 | ||
7dc02805 | 2911 | subtree_root = ocfs2_find_subtree_root(et, left_path, |
328d5752 MF |
2912 | right_path); |
2913 | ||
2914 | mlog(0, "Subtree root at index %d (blk %llu, depth %d)\n", | |
2915 | subtree_root, | |
2916 | (unsigned long long) | |
2917 | right_path->p_node[subtree_root].bh->b_blocknr, | |
2918 | right_path->p_tree_depth); | |
2919 | ||
2920 | ret = ocfs2_extend_rotate_transaction(handle, subtree_root, | |
2921 | orig_credits, left_path); | |
2922 | if (ret) { | |
2923 | mlog_errno(ret); | |
2924 | goto out; | |
2925 | } | |
2926 | ||
e8aed345 MF |
2927 | /* |
2928 | * Caller might still want to make changes to the | |
2929 | * tree root, so re-add it to the journal here. | |
2930 | */ | |
e46f74dc | 2931 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 2932 | left_path, 0); |
e8aed345 MF |
2933 | if (ret) { |
2934 | mlog_errno(ret); | |
2935 | goto out; | |
2936 | } | |
2937 | ||
1e2dd63f | 2938 | ret = ocfs2_rotate_subtree_left(handle, et, left_path, |
328d5752 | 2939 | right_path, subtree_root, |
1e2dd63f | 2940 | dealloc, &deleted); |
328d5752 MF |
2941 | if (ret == -EAGAIN) { |
2942 | /* | |
2943 | * The rotation has to temporarily stop due to | |
2944 | * the right subtree having an empty | |
2945 | * extent. Pass it back to the caller for a | |
2946 | * fixup. | |
2947 | */ | |
2948 | *empty_extent_path = right_path; | |
2949 | right_path = NULL; | |
2950 | goto out; | |
2951 | } | |
2952 | if (ret) { | |
2953 | mlog_errno(ret); | |
2954 | goto out; | |
2955 | } | |
2956 | ||
2957 | /* | |
2958 | * The subtree rotate might have removed records on | |
2959 | * the rightmost edge. If so, then rotation is | |
2960 | * complete. | |
2961 | */ | |
2962 | if (deleted) | |
2963 | break; | |
2964 | ||
2965 | ocfs2_mv_path(left_path, right_path); | |
2966 | ||
e46f74dc | 2967 | ret = ocfs2_find_cpos_for_right_leaf(sb, left_path, |
328d5752 MF |
2968 | &right_cpos); |
2969 | if (ret) { | |
2970 | mlog_errno(ret); | |
2971 | goto out; | |
2972 | } | |
2973 | } | |
2974 | ||
2975 | out: | |
2976 | ocfs2_free_path(right_path); | |
2977 | ocfs2_free_path(left_path); | |
2978 | ||
2979 | return ret; | |
2980 | } | |
2981 | ||
70f18c08 JB |
2982 | static int ocfs2_remove_rightmost_path(handle_t *handle, |
2983 | struct ocfs2_extent_tree *et, | |
e7d4cb6b | 2984 | struct ocfs2_path *path, |
70f18c08 | 2985 | struct ocfs2_cached_dealloc_ctxt *dealloc) |
328d5752 MF |
2986 | { |
2987 | int ret, subtree_index; | |
2988 | u32 cpos; | |
2989 | struct ocfs2_path *left_path = NULL; | |
328d5752 MF |
2990 | struct ocfs2_extent_block *eb; |
2991 | struct ocfs2_extent_list *el; | |
2992 | ||
328d5752 | 2993 | |
6136ca5f | 2994 | ret = ocfs2_et_sanity_check(et); |
e7d4cb6b TM |
2995 | if (ret) |
2996 | goto out; | |
328d5752 MF |
2997 | /* |
2998 | * There's two ways we handle this depending on | |
2999 | * whether path is the only existing one. | |
3000 | */ | |
3001 | ret = ocfs2_extend_rotate_transaction(handle, 0, | |
3002 | handle->h_buffer_credits, | |
3003 | path); | |
3004 | if (ret) { | |
3005 | mlog_errno(ret); | |
3006 | goto out; | |
3007 | } | |
3008 | ||
d9a0a1f8 | 3009 | ret = ocfs2_journal_access_path(et->et_ci, handle, path); |
328d5752 MF |
3010 | if (ret) { |
3011 | mlog_errno(ret); | |
3012 | goto out; | |
3013 | } | |
3014 | ||
3d03a305 JB |
3015 | ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci), |
3016 | path, &cpos); | |
328d5752 MF |
3017 | if (ret) { |
3018 | mlog_errno(ret); | |
3019 | goto out; | |
3020 | } | |
3021 | ||
3022 | if (cpos) { | |
3023 | /* | |
3024 | * We have a path to the left of this one - it needs | |
3025 | * an update too. | |
3026 | */ | |
ffdd7a54 | 3027 | left_path = ocfs2_new_path_from_path(path); |
328d5752 MF |
3028 | if (!left_path) { |
3029 | ret = -ENOMEM; | |
3030 | mlog_errno(ret); | |
3031 | goto out; | |
3032 | } | |
3033 | ||
facdb77f | 3034 | ret = ocfs2_find_path(et->et_ci, left_path, cpos); |
328d5752 MF |
3035 | if (ret) { |
3036 | mlog_errno(ret); | |
3037 | goto out; | |
3038 | } | |
3039 | ||
d9a0a1f8 | 3040 | ret = ocfs2_journal_access_path(et->et_ci, handle, left_path); |
328d5752 MF |
3041 | if (ret) { |
3042 | mlog_errno(ret); | |
3043 | goto out; | |
3044 | } | |
3045 | ||
7dc02805 | 3046 | subtree_index = ocfs2_find_subtree_root(et, left_path, path); |
328d5752 | 3047 | |
6641b0ce | 3048 | ocfs2_unlink_subtree(handle, et, left_path, path, |
328d5752 | 3049 | subtree_index, dealloc); |
09106bae | 3050 | ret = ocfs2_update_edge_lengths(handle, et, subtree_index, |
3c5e1068 TM |
3051 | left_path); |
3052 | if (ret) { | |
3053 | mlog_errno(ret); | |
3054 | goto out; | |
3055 | } | |
328d5752 MF |
3056 | |
3057 | eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data; | |
35dc0aa3 | 3058 | ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno)); |
328d5752 MF |
3059 | } else { |
3060 | /* | |
3061 | * 'path' is also the leftmost path which | |
3062 | * means it must be the only one. This gets | |
3063 | * handled differently because we want to | |
70f18c08 | 3064 | * revert the root back to having extents |
328d5752 MF |
3065 | * in-line. |
3066 | */ | |
6641b0ce | 3067 | ocfs2_unlink_path(handle, et, dealloc, path, 1); |
328d5752 | 3068 | |
ce1d9ea6 | 3069 | el = et->et_root_el; |
328d5752 MF |
3070 | el->l_tree_depth = 0; |
3071 | el->l_next_free_rec = 0; | |
3072 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
3073 | ||
35dc0aa3 | 3074 | ocfs2_et_set_last_eb_blk(et, 0); |
328d5752 MF |
3075 | } |
3076 | ||
3077 | ocfs2_journal_dirty(handle, path_root_bh(path)); | |
3078 | ||
3079 | out: | |
3080 | ocfs2_free_path(left_path); | |
3081 | return ret; | |
3082 | } | |
3083 | ||
3084 | /* | |
3085 | * Left rotation of btree records. | |
3086 | * | |
3087 | * In many ways, this is (unsurprisingly) the opposite of right | |
3088 | * rotation. We start at some non-rightmost path containing an empty | |
3089 | * extent in the leaf block. The code works its way to the rightmost | |
3090 | * path by rotating records to the left in every subtree. | |
3091 | * | |
3092 | * This is used by any code which reduces the number of extent records | |
3093 | * in a leaf. After removal, an empty record should be placed in the | |
3094 | * leftmost list position. | |
3095 | * | |
3096 | * This won't handle a length update of the rightmost path records if | |
3097 | * the rightmost tree leaf record is removed so the caller is | |
3098 | * responsible for detecting and correcting that. | |
3099 | */ | |
70f18c08 JB |
3100 | static int ocfs2_rotate_tree_left(handle_t *handle, |
3101 | struct ocfs2_extent_tree *et, | |
328d5752 | 3102 | struct ocfs2_path *path, |
70f18c08 | 3103 | struct ocfs2_cached_dealloc_ctxt *dealloc) |
328d5752 MF |
3104 | { |
3105 | int ret, orig_credits = handle->h_buffer_credits; | |
3106 | struct ocfs2_path *tmp_path = NULL, *restart_path = NULL; | |
3107 | struct ocfs2_extent_block *eb; | |
3108 | struct ocfs2_extent_list *el; | |
3109 | ||
3110 | el = path_leaf_el(path); | |
3111 | if (!ocfs2_is_empty_extent(&el->l_recs[0])) | |
3112 | return 0; | |
3113 | ||
3114 | if (path->p_tree_depth == 0) { | |
3115 | rightmost_no_delete: | |
3116 | /* | |
e7d4cb6b | 3117 | * Inline extents. This is trivially handled, so do |
328d5752 MF |
3118 | * it up front. |
3119 | */ | |
70f18c08 | 3120 | ret = ocfs2_rotate_rightmost_leaf_left(handle, et, path); |
328d5752 MF |
3121 | if (ret) |
3122 | mlog_errno(ret); | |
3123 | goto out; | |
3124 | } | |
3125 | ||
3126 | /* | |
3127 | * Handle rightmost branch now. There's several cases: | |
3128 | * 1) simple rotation leaving records in there. That's trivial. | |
3129 | * 2) rotation requiring a branch delete - there's no more | |
3130 | * records left. Two cases of this: | |
3131 | * a) There are branches to the left. | |
3132 | * b) This is also the leftmost (the only) branch. | |
3133 | * | |
3134 | * 1) is handled via ocfs2_rotate_rightmost_leaf_left() | |
3135 | * 2a) we need the left branch so that we can update it with the unlink | |
70f18c08 | 3136 | * 2b) we need to bring the root back to inline extents. |
328d5752 MF |
3137 | */ |
3138 | ||
3139 | eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data; | |
3140 | el = &eb->h_list; | |
3141 | if (eb->h_next_leaf_blk == 0) { | |
3142 | /* | |
3143 | * This gets a bit tricky if we're going to delete the | |
3144 | * rightmost path. Get the other cases out of the way | |
3145 | * 1st. | |
3146 | */ | |
3147 | if (le16_to_cpu(el->l_next_free_rec) > 1) | |
3148 | goto rightmost_no_delete; | |
3149 | ||
3150 | if (le16_to_cpu(el->l_next_free_rec) == 0) { | |
3151 | ret = -EIO; | |
70f18c08 JB |
3152 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
3153 | "Owner %llu has empty extent block at %llu", | |
3154 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), | |
328d5752 MF |
3155 | (unsigned long long)le64_to_cpu(eb->h_blkno)); |
3156 | goto out; | |
3157 | } | |
3158 | ||
3159 | /* | |
3160 | * XXX: The caller can not trust "path" any more after | |
3161 | * this as it will have been deleted. What do we do? | |
3162 | * | |
3163 | * In theory the rotate-for-merge code will never get | |
3164 | * here because it'll always ask for a rotate in a | |
3165 | * nonempty list. | |
3166 | */ | |
3167 | ||
70f18c08 JB |
3168 | ret = ocfs2_remove_rightmost_path(handle, et, path, |
3169 | dealloc); | |
328d5752 MF |
3170 | if (ret) |
3171 | mlog_errno(ret); | |
3172 | goto out; | |
3173 | } | |
3174 | ||
3175 | /* | |
3176 | * Now we can loop, remembering the path we get from -EAGAIN | |
3177 | * and restarting from there. | |
3178 | */ | |
3179 | try_rotate: | |
e46f74dc JB |
3180 | ret = __ocfs2_rotate_tree_left(handle, et, orig_credits, path, |
3181 | dealloc, &restart_path); | |
328d5752 MF |
3182 | if (ret && ret != -EAGAIN) { |
3183 | mlog_errno(ret); | |
3184 | goto out; | |
3185 | } | |
3186 | ||
3187 | while (ret == -EAGAIN) { | |
3188 | tmp_path = restart_path; | |
3189 | restart_path = NULL; | |
3190 | ||
e46f74dc | 3191 | ret = __ocfs2_rotate_tree_left(handle, et, orig_credits, |
328d5752 | 3192 | tmp_path, dealloc, |
e46f74dc | 3193 | &restart_path); |
328d5752 MF |
3194 | if (ret && ret != -EAGAIN) { |
3195 | mlog_errno(ret); | |
3196 | goto out; | |
3197 | } | |
3198 | ||
3199 | ocfs2_free_path(tmp_path); | |
3200 | tmp_path = NULL; | |
3201 | ||
3202 | if (ret == 0) | |
3203 | goto try_rotate; | |
3204 | } | |
3205 | ||
3206 | out: | |
3207 | ocfs2_free_path(tmp_path); | |
3208 | ocfs2_free_path(restart_path); | |
3209 | return ret; | |
3210 | } | |
3211 | ||
3212 | static void ocfs2_cleanup_merge(struct ocfs2_extent_list *el, | |
3213 | int index) | |
3214 | { | |
3215 | struct ocfs2_extent_rec *rec = &el->l_recs[index]; | |
3216 | unsigned int size; | |
3217 | ||
3218 | if (rec->e_leaf_clusters == 0) { | |
3219 | /* | |
3220 | * We consumed all of the merged-from record. An empty | |
3221 | * extent cannot exist anywhere but the 1st array | |
3222 | * position, so move things over if the merged-from | |
3223 | * record doesn't occupy that position. | |
3224 | * | |
3225 | * This creates a new empty extent so the caller | |
3226 | * should be smart enough to have removed any existing | |
3227 | * ones. | |
3228 | */ | |
3229 | if (index > 0) { | |
3230 | BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0])); | |
3231 | size = index * sizeof(struct ocfs2_extent_rec); | |
3232 | memmove(&el->l_recs[1], &el->l_recs[0], size); | |
3233 | } | |
3234 | ||
3235 | /* | |
3236 | * Always memset - the caller doesn't check whether it | |
3237 | * created an empty extent, so there could be junk in | |
3238 | * the other fields. | |
3239 | */ | |
3240 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
3241 | } | |
3242 | } | |
3243 | ||
4fe82c31 | 3244 | static int ocfs2_get_right_path(struct ocfs2_extent_tree *et, |
677b9752 TM |
3245 | struct ocfs2_path *left_path, |
3246 | struct ocfs2_path **ret_right_path) | |
3247 | { | |
3248 | int ret; | |
3249 | u32 right_cpos; | |
3250 | struct ocfs2_path *right_path = NULL; | |
3251 | struct ocfs2_extent_list *left_el; | |
3252 | ||
3253 | *ret_right_path = NULL; | |
3254 | ||
3255 | /* This function shouldn't be called for non-trees. */ | |
3256 | BUG_ON(left_path->p_tree_depth == 0); | |
3257 | ||
3258 | left_el = path_leaf_el(left_path); | |
3259 | BUG_ON(left_el->l_next_free_rec != left_el->l_count); | |
3260 | ||
4fe82c31 JB |
3261 | ret = ocfs2_find_cpos_for_right_leaf(ocfs2_metadata_cache_get_super(et->et_ci), |
3262 | left_path, &right_cpos); | |
677b9752 TM |
3263 | if (ret) { |
3264 | mlog_errno(ret); | |
3265 | goto out; | |
3266 | } | |
3267 | ||
3268 | /* This function shouldn't be called for the rightmost leaf. */ | |
3269 | BUG_ON(right_cpos == 0); | |
3270 | ||
ffdd7a54 | 3271 | right_path = ocfs2_new_path_from_path(left_path); |
677b9752 TM |
3272 | if (!right_path) { |
3273 | ret = -ENOMEM; | |
3274 | mlog_errno(ret); | |
3275 | goto out; | |
3276 | } | |
3277 | ||
4fe82c31 | 3278 | ret = ocfs2_find_path(et->et_ci, right_path, right_cpos); |
677b9752 TM |
3279 | if (ret) { |
3280 | mlog_errno(ret); | |
3281 | goto out; | |
3282 | } | |
3283 | ||
3284 | *ret_right_path = right_path; | |
3285 | out: | |
3286 | if (ret) | |
3287 | ocfs2_free_path(right_path); | |
3288 | return ret; | |
3289 | } | |
3290 | ||
328d5752 MF |
3291 | /* |
3292 | * Remove split_rec clusters from the record at index and merge them | |
677b9752 TM |
3293 | * onto the beginning of the record "next" to it. |
3294 | * For index < l_count - 1, the next means the extent rec at index + 1. | |
3295 | * For index == l_count - 1, the "next" means the 1st extent rec of the | |
3296 | * next extent block. | |
328d5752 | 3297 | */ |
4fe82c31 | 3298 | static int ocfs2_merge_rec_right(struct ocfs2_path *left_path, |
677b9752 | 3299 | handle_t *handle, |
7dc02805 | 3300 | struct ocfs2_extent_tree *et, |
677b9752 TM |
3301 | struct ocfs2_extent_rec *split_rec, |
3302 | int index) | |
328d5752 | 3303 | { |
677b9752 | 3304 | int ret, next_free, i; |
328d5752 MF |
3305 | unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters); |
3306 | struct ocfs2_extent_rec *left_rec; | |
3307 | struct ocfs2_extent_rec *right_rec; | |
677b9752 TM |
3308 | struct ocfs2_extent_list *right_el; |
3309 | struct ocfs2_path *right_path = NULL; | |
3310 | int subtree_index = 0; | |
3311 | struct ocfs2_extent_list *el = path_leaf_el(left_path); | |
3312 | struct buffer_head *bh = path_leaf_bh(left_path); | |
3313 | struct buffer_head *root_bh = NULL; | |
328d5752 MF |
3314 | |
3315 | BUG_ON(index >= le16_to_cpu(el->l_next_free_rec)); | |
328d5752 | 3316 | left_rec = &el->l_recs[index]; |
677b9752 | 3317 | |
9d8df6aa | 3318 | if (index == le16_to_cpu(el->l_next_free_rec) - 1 && |
677b9752 TM |
3319 | le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count)) { |
3320 | /* we meet with a cross extent block merge. */ | |
4fe82c31 | 3321 | ret = ocfs2_get_right_path(et, left_path, &right_path); |
677b9752 TM |
3322 | if (ret) { |
3323 | mlog_errno(ret); | |
3324 | goto out; | |
3325 | } | |
3326 | ||
3327 | right_el = path_leaf_el(right_path); | |
3328 | next_free = le16_to_cpu(right_el->l_next_free_rec); | |
3329 | BUG_ON(next_free <= 0); | |
3330 | right_rec = &right_el->l_recs[0]; | |
3331 | if (ocfs2_is_empty_extent(right_rec)) { | |
9d8df6aa | 3332 | BUG_ON(next_free <= 1); |
677b9752 TM |
3333 | right_rec = &right_el->l_recs[1]; |
3334 | } | |
3335 | ||
3336 | BUG_ON(le32_to_cpu(left_rec->e_cpos) + | |
3337 | le16_to_cpu(left_rec->e_leaf_clusters) != | |
3338 | le32_to_cpu(right_rec->e_cpos)); | |
3339 | ||
7dc02805 JB |
3340 | subtree_index = ocfs2_find_subtree_root(et, left_path, |
3341 | right_path); | |
677b9752 TM |
3342 | |
3343 | ret = ocfs2_extend_rotate_transaction(handle, subtree_index, | |
3344 | handle->h_buffer_credits, | |
3345 | right_path); | |
3346 | if (ret) { | |
3347 | mlog_errno(ret); | |
3348 | goto out; | |
3349 | } | |
3350 | ||
3351 | root_bh = left_path->p_node[subtree_index].bh; | |
3352 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
3353 | ||
7dc02805 | 3354 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path, |
13723d00 | 3355 | subtree_index); |
677b9752 TM |
3356 | if (ret) { |
3357 | mlog_errno(ret); | |
3358 | goto out; | |
3359 | } | |
3360 | ||
3361 | for (i = subtree_index + 1; | |
3362 | i < path_num_items(right_path); i++) { | |
7dc02805 | 3363 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 3364 | right_path, i); |
677b9752 TM |
3365 | if (ret) { |
3366 | mlog_errno(ret); | |
3367 | goto out; | |
3368 | } | |
3369 | ||
7dc02805 | 3370 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 3371 | left_path, i); |
677b9752 TM |
3372 | if (ret) { |
3373 | mlog_errno(ret); | |
3374 | goto out; | |
3375 | } | |
3376 | } | |
3377 | ||
3378 | } else { | |
3379 | BUG_ON(index == le16_to_cpu(el->l_next_free_rec) - 1); | |
3380 | right_rec = &el->l_recs[index + 1]; | |
3381 | } | |
328d5752 | 3382 | |
7dc02805 | 3383 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, left_path, |
13723d00 | 3384 | path_num_items(left_path) - 1); |
328d5752 MF |
3385 | if (ret) { |
3386 | mlog_errno(ret); | |
3387 | goto out; | |
3388 | } | |
3389 | ||
3390 | le16_add_cpu(&left_rec->e_leaf_clusters, -split_clusters); | |
3391 | ||
3392 | le32_add_cpu(&right_rec->e_cpos, -split_clusters); | |
3393 | le64_add_cpu(&right_rec->e_blkno, | |
7dc02805 JB |
3394 | -ocfs2_clusters_to_blocks(ocfs2_metadata_cache_get_super(et->et_ci), |
3395 | split_clusters)); | |
328d5752 MF |
3396 | le16_add_cpu(&right_rec->e_leaf_clusters, split_clusters); |
3397 | ||
3398 | ocfs2_cleanup_merge(el, index); | |
3399 | ||
3400 | ret = ocfs2_journal_dirty(handle, bh); | |
3401 | if (ret) | |
3402 | mlog_errno(ret); | |
3403 | ||
677b9752 TM |
3404 | if (right_path) { |
3405 | ret = ocfs2_journal_dirty(handle, path_leaf_bh(right_path)); | |
3406 | if (ret) | |
3407 | mlog_errno(ret); | |
3408 | ||
4619c73e JB |
3409 | ocfs2_complete_edge_insert(handle, left_path, right_path, |
3410 | subtree_index); | |
677b9752 TM |
3411 | } |
3412 | out: | |
3413 | if (right_path) | |
3414 | ocfs2_free_path(right_path); | |
3415 | return ret; | |
3416 | } | |
3417 | ||
4fe82c31 | 3418 | static int ocfs2_get_left_path(struct ocfs2_extent_tree *et, |
677b9752 TM |
3419 | struct ocfs2_path *right_path, |
3420 | struct ocfs2_path **ret_left_path) | |
3421 | { | |
3422 | int ret; | |
3423 | u32 left_cpos; | |
3424 | struct ocfs2_path *left_path = NULL; | |
3425 | ||
3426 | *ret_left_path = NULL; | |
3427 | ||
3428 | /* This function shouldn't be called for non-trees. */ | |
3429 | BUG_ON(right_path->p_tree_depth == 0); | |
3430 | ||
4fe82c31 | 3431 | ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci), |
677b9752 TM |
3432 | right_path, &left_cpos); |
3433 | if (ret) { | |
3434 | mlog_errno(ret); | |
3435 | goto out; | |
3436 | } | |
3437 | ||
3438 | /* This function shouldn't be called for the leftmost leaf. */ | |
3439 | BUG_ON(left_cpos == 0); | |
3440 | ||
ffdd7a54 | 3441 | left_path = ocfs2_new_path_from_path(right_path); |
677b9752 TM |
3442 | if (!left_path) { |
3443 | ret = -ENOMEM; | |
3444 | mlog_errno(ret); | |
3445 | goto out; | |
3446 | } | |
3447 | ||
4fe82c31 | 3448 | ret = ocfs2_find_path(et->et_ci, left_path, left_cpos); |
677b9752 TM |
3449 | if (ret) { |
3450 | mlog_errno(ret); | |
3451 | goto out; | |
3452 | } | |
3453 | ||
3454 | *ret_left_path = left_path; | |
328d5752 | 3455 | out: |
677b9752 TM |
3456 | if (ret) |
3457 | ocfs2_free_path(left_path); | |
328d5752 MF |
3458 | return ret; |
3459 | } | |
3460 | ||
3461 | /* | |
3462 | * Remove split_rec clusters from the record at index and merge them | |
677b9752 TM |
3463 | * onto the tail of the record "before" it. |
3464 | * For index > 0, the "before" means the extent rec at index - 1. | |
3465 | * | |
3466 | * For index == 0, the "before" means the last record of the previous | |
3467 | * extent block. And there is also a situation that we may need to | |
3468 | * remove the rightmost leaf extent block in the right_path and change | |
3469 | * the right path to indicate the new rightmost path. | |
328d5752 | 3470 | */ |
4fe82c31 | 3471 | static int ocfs2_merge_rec_left(struct ocfs2_path *right_path, |
328d5752 | 3472 | handle_t *handle, |
4fe82c31 | 3473 | struct ocfs2_extent_tree *et, |
328d5752 | 3474 | struct ocfs2_extent_rec *split_rec, |
677b9752 TM |
3475 | struct ocfs2_cached_dealloc_ctxt *dealloc, |
3476 | int index) | |
328d5752 | 3477 | { |
677b9752 | 3478 | int ret, i, subtree_index = 0, has_empty_extent = 0; |
328d5752 MF |
3479 | unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters); |
3480 | struct ocfs2_extent_rec *left_rec; | |
3481 | struct ocfs2_extent_rec *right_rec; | |
677b9752 TM |
3482 | struct ocfs2_extent_list *el = path_leaf_el(right_path); |
3483 | struct buffer_head *bh = path_leaf_bh(right_path); | |
3484 | struct buffer_head *root_bh = NULL; | |
3485 | struct ocfs2_path *left_path = NULL; | |
3486 | struct ocfs2_extent_list *left_el; | |
328d5752 | 3487 | |
677b9752 | 3488 | BUG_ON(index < 0); |
328d5752 | 3489 | |
328d5752 | 3490 | right_rec = &el->l_recs[index]; |
677b9752 TM |
3491 | if (index == 0) { |
3492 | /* we meet with a cross extent block merge. */ | |
4fe82c31 | 3493 | ret = ocfs2_get_left_path(et, right_path, &left_path); |
677b9752 TM |
3494 | if (ret) { |
3495 | mlog_errno(ret); | |
3496 | goto out; | |
3497 | } | |
3498 | ||
3499 | left_el = path_leaf_el(left_path); | |
3500 | BUG_ON(le16_to_cpu(left_el->l_next_free_rec) != | |
3501 | le16_to_cpu(left_el->l_count)); | |
3502 | ||
3503 | left_rec = &left_el->l_recs[ | |
3504 | le16_to_cpu(left_el->l_next_free_rec) - 1]; | |
3505 | BUG_ON(le32_to_cpu(left_rec->e_cpos) + | |
3506 | le16_to_cpu(left_rec->e_leaf_clusters) != | |
3507 | le32_to_cpu(split_rec->e_cpos)); | |
3508 | ||
7dc02805 JB |
3509 | subtree_index = ocfs2_find_subtree_root(et, left_path, |
3510 | right_path); | |
677b9752 TM |
3511 | |
3512 | ret = ocfs2_extend_rotate_transaction(handle, subtree_index, | |
3513 | handle->h_buffer_credits, | |
3514 | left_path); | |
3515 | if (ret) { | |
3516 | mlog_errno(ret); | |
3517 | goto out; | |
3518 | } | |
3519 | ||
3520 | root_bh = left_path->p_node[subtree_index].bh; | |
3521 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
3522 | ||
4fe82c31 | 3523 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path, |
13723d00 | 3524 | subtree_index); |
677b9752 TM |
3525 | if (ret) { |
3526 | mlog_errno(ret); | |
3527 | goto out; | |
3528 | } | |
3529 | ||
3530 | for (i = subtree_index + 1; | |
3531 | i < path_num_items(right_path); i++) { | |
4fe82c31 | 3532 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 3533 | right_path, i); |
677b9752 TM |
3534 | if (ret) { |
3535 | mlog_errno(ret); | |
3536 | goto out; | |
3537 | } | |
3538 | ||
4fe82c31 | 3539 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 3540 | left_path, i); |
677b9752 TM |
3541 | if (ret) { |
3542 | mlog_errno(ret); | |
3543 | goto out; | |
3544 | } | |
3545 | } | |
3546 | } else { | |
3547 | left_rec = &el->l_recs[index - 1]; | |
3548 | if (ocfs2_is_empty_extent(&el->l_recs[0])) | |
3549 | has_empty_extent = 1; | |
3550 | } | |
328d5752 | 3551 | |
4fe82c31 | 3552 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path, |
9047beab | 3553 | path_num_items(right_path) - 1); |
328d5752 MF |
3554 | if (ret) { |
3555 | mlog_errno(ret); | |
3556 | goto out; | |
3557 | } | |
3558 | ||
3559 | if (has_empty_extent && index == 1) { | |
3560 | /* | |
3561 | * The easy case - we can just plop the record right in. | |
3562 | */ | |
3563 | *left_rec = *split_rec; | |
3564 | ||
3565 | has_empty_extent = 0; | |
677b9752 | 3566 | } else |
328d5752 | 3567 | le16_add_cpu(&left_rec->e_leaf_clusters, split_clusters); |
328d5752 MF |
3568 | |
3569 | le32_add_cpu(&right_rec->e_cpos, split_clusters); | |
3570 | le64_add_cpu(&right_rec->e_blkno, | |
4fe82c31 JB |
3571 | ocfs2_clusters_to_blocks(ocfs2_metadata_cache_get_super(et->et_ci), |
3572 | split_clusters)); | |
328d5752 MF |
3573 | le16_add_cpu(&right_rec->e_leaf_clusters, -split_clusters); |
3574 | ||
3575 | ocfs2_cleanup_merge(el, index); | |
3576 | ||
3577 | ret = ocfs2_journal_dirty(handle, bh); | |
3578 | if (ret) | |
3579 | mlog_errno(ret); | |
3580 | ||
677b9752 TM |
3581 | if (left_path) { |
3582 | ret = ocfs2_journal_dirty(handle, path_leaf_bh(left_path)); | |
3583 | if (ret) | |
3584 | mlog_errno(ret); | |
3585 | ||
3586 | /* | |
3587 | * In the situation that the right_rec is empty and the extent | |
3588 | * block is empty also, ocfs2_complete_edge_insert can't handle | |
3589 | * it and we need to delete the right extent block. | |
3590 | */ | |
3591 | if (le16_to_cpu(right_rec->e_leaf_clusters) == 0 && | |
3592 | le16_to_cpu(el->l_next_free_rec) == 1) { | |
3593 | ||
70f18c08 | 3594 | ret = ocfs2_remove_rightmost_path(handle, et, |
e7d4cb6b | 3595 | right_path, |
70f18c08 | 3596 | dealloc); |
677b9752 TM |
3597 | if (ret) { |
3598 | mlog_errno(ret); | |
3599 | goto out; | |
3600 | } | |
3601 | ||
3602 | /* Now the rightmost extent block has been deleted. | |
3603 | * So we use the new rightmost path. | |
3604 | */ | |
3605 | ocfs2_mv_path(right_path, left_path); | |
3606 | left_path = NULL; | |
3607 | } else | |
4619c73e | 3608 | ocfs2_complete_edge_insert(handle, left_path, |
677b9752 TM |
3609 | right_path, subtree_index); |
3610 | } | |
328d5752 | 3611 | out: |
677b9752 TM |
3612 | if (left_path) |
3613 | ocfs2_free_path(left_path); | |
328d5752 MF |
3614 | return ret; |
3615 | } | |
3616 | ||
c495dd24 JB |
3617 | static int ocfs2_try_to_merge_extent(handle_t *handle, |
3618 | struct ocfs2_extent_tree *et, | |
677b9752 | 3619 | struct ocfs2_path *path, |
328d5752 MF |
3620 | int split_index, |
3621 | struct ocfs2_extent_rec *split_rec, | |
3622 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
c495dd24 | 3623 | struct ocfs2_merge_ctxt *ctxt) |
328d5752 | 3624 | { |
518d7269 | 3625 | int ret = 0; |
677b9752 | 3626 | struct ocfs2_extent_list *el = path_leaf_el(path); |
328d5752 MF |
3627 | struct ocfs2_extent_rec *rec = &el->l_recs[split_index]; |
3628 | ||
3629 | BUG_ON(ctxt->c_contig_type == CONTIG_NONE); | |
3630 | ||
518d7269 TM |
3631 | if (ctxt->c_split_covers_rec && ctxt->c_has_empty_extent) { |
3632 | /* | |
3633 | * The merge code will need to create an empty | |
3634 | * extent to take the place of the newly | |
3635 | * emptied slot. Remove any pre-existing empty | |
3636 | * extents - having more than one in a leaf is | |
3637 | * illegal. | |
3638 | */ | |
70f18c08 | 3639 | ret = ocfs2_rotate_tree_left(handle, et, path, dealloc); |
518d7269 TM |
3640 | if (ret) { |
3641 | mlog_errno(ret); | |
3642 | goto out; | |
328d5752 | 3643 | } |
518d7269 TM |
3644 | split_index--; |
3645 | rec = &el->l_recs[split_index]; | |
328d5752 MF |
3646 | } |
3647 | ||
3648 | if (ctxt->c_contig_type == CONTIG_LEFTRIGHT) { | |
3649 | /* | |
3650 | * Left-right contig implies this. | |
3651 | */ | |
3652 | BUG_ON(!ctxt->c_split_covers_rec); | |
328d5752 MF |
3653 | |
3654 | /* | |
3655 | * Since the leftright insert always covers the entire | |
3656 | * extent, this call will delete the insert record | |
3657 | * entirely, resulting in an empty extent record added to | |
3658 | * the extent block. | |
3659 | * | |
3660 | * Since the adding of an empty extent shifts | |
3661 | * everything back to the right, there's no need to | |
3662 | * update split_index here. | |
677b9752 TM |
3663 | * |
3664 | * When the split_index is zero, we need to merge it to the | |
3665 | * prevoius extent block. It is more efficient and easier | |
3666 | * if we do merge_right first and merge_left later. | |
328d5752 | 3667 | */ |
4fe82c31 | 3668 | ret = ocfs2_merge_rec_right(path, handle, et, split_rec, |
677b9752 | 3669 | split_index); |
328d5752 MF |
3670 | if (ret) { |
3671 | mlog_errno(ret); | |
3672 | goto out; | |
3673 | } | |
3674 | ||
3675 | /* | |
3676 | * We can only get this from logic error above. | |
3677 | */ | |
3678 | BUG_ON(!ocfs2_is_empty_extent(&el->l_recs[0])); | |
3679 | ||
677b9752 | 3680 | /* The merge left us with an empty extent, remove it. */ |
70f18c08 | 3681 | ret = ocfs2_rotate_tree_left(handle, et, path, dealloc); |
328d5752 MF |
3682 | if (ret) { |
3683 | mlog_errno(ret); | |
3684 | goto out; | |
3685 | } | |
677b9752 | 3686 | |
328d5752 MF |
3687 | rec = &el->l_recs[split_index]; |
3688 | ||
3689 | /* | |
3690 | * Note that we don't pass split_rec here on purpose - | |
677b9752 | 3691 | * we've merged it into the rec already. |
328d5752 | 3692 | */ |
4fe82c31 JB |
3693 | ret = ocfs2_merge_rec_left(path, handle, et, rec, |
3694 | dealloc, split_index); | |
677b9752 | 3695 | |
328d5752 MF |
3696 | if (ret) { |
3697 | mlog_errno(ret); | |
3698 | goto out; | |
3699 | } | |
3700 | ||
70f18c08 | 3701 | ret = ocfs2_rotate_tree_left(handle, et, path, dealloc); |
328d5752 MF |
3702 | /* |
3703 | * Error from this last rotate is not critical, so | |
3704 | * print but don't bubble it up. | |
3705 | */ | |
3706 | if (ret) | |
3707 | mlog_errno(ret); | |
3708 | ret = 0; | |
3709 | } else { | |
3710 | /* | |
3711 | * Merge a record to the left or right. | |
3712 | * | |
3713 | * 'contig_type' is relative to the existing record, | |
3714 | * so for example, if we're "right contig", it's to | |
3715 | * the record on the left (hence the left merge). | |
3716 | */ | |
3717 | if (ctxt->c_contig_type == CONTIG_RIGHT) { | |
4fe82c31 JB |
3718 | ret = ocfs2_merge_rec_left(path, handle, et, |
3719 | split_rec, dealloc, | |
328d5752 MF |
3720 | split_index); |
3721 | if (ret) { | |
3722 | mlog_errno(ret); | |
3723 | goto out; | |
3724 | } | |
3725 | } else { | |
4fe82c31 | 3726 | ret = ocfs2_merge_rec_right(path, handle, |
7dc02805 | 3727 | et, split_rec, |
328d5752 MF |
3728 | split_index); |
3729 | if (ret) { | |
3730 | mlog_errno(ret); | |
3731 | goto out; | |
3732 | } | |
3733 | } | |
3734 | ||
3735 | if (ctxt->c_split_covers_rec) { | |
3736 | /* | |
3737 | * The merge may have left an empty extent in | |
3738 | * our leaf. Try to rotate it away. | |
3739 | */ | |
70f18c08 JB |
3740 | ret = ocfs2_rotate_tree_left(handle, et, path, |
3741 | dealloc); | |
328d5752 MF |
3742 | if (ret) |
3743 | mlog_errno(ret); | |
3744 | ret = 0; | |
3745 | } | |
3746 | } | |
3747 | ||
3748 | out: | |
3749 | return ret; | |
3750 | } | |
3751 | ||
3752 | static void ocfs2_subtract_from_rec(struct super_block *sb, | |
3753 | enum ocfs2_split_type split, | |
3754 | struct ocfs2_extent_rec *rec, | |
3755 | struct ocfs2_extent_rec *split_rec) | |
3756 | { | |
3757 | u64 len_blocks; | |
3758 | ||
3759 | len_blocks = ocfs2_clusters_to_blocks(sb, | |
3760 | le16_to_cpu(split_rec->e_leaf_clusters)); | |
3761 | ||
3762 | if (split == SPLIT_LEFT) { | |
3763 | /* | |
3764 | * Region is on the left edge of the existing | |
3765 | * record. | |
3766 | */ | |
3767 | le32_add_cpu(&rec->e_cpos, | |
3768 | le16_to_cpu(split_rec->e_leaf_clusters)); | |
3769 | le64_add_cpu(&rec->e_blkno, len_blocks); | |
3770 | le16_add_cpu(&rec->e_leaf_clusters, | |
3771 | -le16_to_cpu(split_rec->e_leaf_clusters)); | |
3772 | } else { | |
3773 | /* | |
3774 | * Region is on the right edge of the existing | |
3775 | * record. | |
3776 | */ | |
3777 | le16_add_cpu(&rec->e_leaf_clusters, | |
3778 | -le16_to_cpu(split_rec->e_leaf_clusters)); | |
3779 | } | |
3780 | } | |
3781 | ||
3782 | /* | |
3783 | * Do the final bits of extent record insertion at the target leaf | |
3784 | * list. If this leaf is part of an allocation tree, it is assumed | |
3785 | * that the tree above has been prepared. | |
3786 | */ | |
3787 | static void ocfs2_insert_at_leaf(struct ocfs2_extent_rec *insert_rec, | |
3788 | struct ocfs2_extent_list *el, | |
3789 | struct ocfs2_insert_type *insert, | |
3790 | struct inode *inode) | |
3791 | { | |
3792 | int i = insert->ins_contig_index; | |
3793 | unsigned int range; | |
3794 | struct ocfs2_extent_rec *rec; | |
3795 | ||
3796 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); | |
3797 | ||
3798 | if (insert->ins_split != SPLIT_NONE) { | |
3799 | i = ocfs2_search_extent_list(el, le32_to_cpu(insert_rec->e_cpos)); | |
3800 | BUG_ON(i == -1); | |
3801 | rec = &el->l_recs[i]; | |
3802 | ocfs2_subtract_from_rec(inode->i_sb, insert->ins_split, rec, | |
3803 | insert_rec); | |
3804 | goto rotate; | |
3805 | } | |
3806 | ||
3807 | /* | |
3808 | * Contiguous insert - either left or right. | |
3809 | */ | |
3810 | if (insert->ins_contig != CONTIG_NONE) { | |
3811 | rec = &el->l_recs[i]; | |
3812 | if (insert->ins_contig == CONTIG_LEFT) { | |
3813 | rec->e_blkno = insert_rec->e_blkno; | |
3814 | rec->e_cpos = insert_rec->e_cpos; | |
3815 | } | |
3816 | le16_add_cpu(&rec->e_leaf_clusters, | |
3817 | le16_to_cpu(insert_rec->e_leaf_clusters)); | |
3818 | return; | |
3819 | } | |
3820 | ||
3821 | /* | |
3822 | * Handle insert into an empty leaf. | |
3823 | */ | |
3824 | if (le16_to_cpu(el->l_next_free_rec) == 0 || | |
3825 | ((le16_to_cpu(el->l_next_free_rec) == 1) && | |
3826 | ocfs2_is_empty_extent(&el->l_recs[0]))) { | |
3827 | el->l_recs[0] = *insert_rec; | |
3828 | el->l_next_free_rec = cpu_to_le16(1); | |
3829 | return; | |
3830 | } | |
3831 | ||
3832 | /* | |
3833 | * Appending insert. | |
3834 | */ | |
3835 | if (insert->ins_appending == APPEND_TAIL) { | |
3836 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
3837 | rec = &el->l_recs[i]; | |
3838 | range = le32_to_cpu(rec->e_cpos) | |
3839 | + le16_to_cpu(rec->e_leaf_clusters); | |
3840 | BUG_ON(le32_to_cpu(insert_rec->e_cpos) < range); | |
3841 | ||
3842 | mlog_bug_on_msg(le16_to_cpu(el->l_next_free_rec) >= | |
3843 | le16_to_cpu(el->l_count), | |
3844 | "inode %lu, depth %u, count %u, next free %u, " | |
3845 | "rec.cpos %u, rec.clusters %u, " | |
3846 | "insert.cpos %u, insert.clusters %u\n", | |
3847 | inode->i_ino, | |
3848 | le16_to_cpu(el->l_tree_depth), | |
3849 | le16_to_cpu(el->l_count), | |
3850 | le16_to_cpu(el->l_next_free_rec), | |
3851 | le32_to_cpu(el->l_recs[i].e_cpos), | |
3852 | le16_to_cpu(el->l_recs[i].e_leaf_clusters), | |
3853 | le32_to_cpu(insert_rec->e_cpos), | |
3854 | le16_to_cpu(insert_rec->e_leaf_clusters)); | |
3855 | i++; | |
3856 | el->l_recs[i] = *insert_rec; | |
3857 | le16_add_cpu(&el->l_next_free_rec, 1); | |
3858 | return; | |
3859 | } | |
3860 | ||
3861 | rotate: | |
3862 | /* | |
3863 | * Ok, we have to rotate. | |
3864 | * | |
3865 | * At this point, it is safe to assume that inserting into an | |
3866 | * empty leaf and appending to a leaf have both been handled | |
3867 | * above. | |
3868 | * | |
3869 | * This leaf needs to have space, either by the empty 1st | |
3870 | * extent record, or by virtue of an l_next_rec < l_count. | |
3871 | */ | |
3872 | ocfs2_rotate_leaf(el, insert_rec); | |
3873 | } | |
3874 | ||
d401dc12 JB |
3875 | static void ocfs2_adjust_rightmost_records(handle_t *handle, |
3876 | struct ocfs2_extent_tree *et, | |
328d5752 MF |
3877 | struct ocfs2_path *path, |
3878 | struct ocfs2_extent_rec *insert_rec) | |
3879 | { | |
3880 | int ret, i, next_free; | |
3881 | struct buffer_head *bh; | |
3882 | struct ocfs2_extent_list *el; | |
3883 | struct ocfs2_extent_rec *rec; | |
3884 | ||
3885 | /* | |
3886 | * Update everything except the leaf block. | |
3887 | */ | |
3888 | for (i = 0; i < path->p_tree_depth; i++) { | |
3889 | bh = path->p_node[i].bh; | |
3890 | el = path->p_node[i].el; | |
3891 | ||
dcd0538f MF |
3892 | next_free = le16_to_cpu(el->l_next_free_rec); |
3893 | if (next_free == 0) { | |
d401dc12 JB |
3894 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
3895 | "Owner %llu has a bad extent list", | |
3896 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci)); | |
dcd0538f | 3897 | ret = -EIO; |
328d5752 MF |
3898 | return; |
3899 | } | |
3900 | ||
3901 | rec = &el->l_recs[next_free - 1]; | |
3902 | ||
3903 | rec->e_int_clusters = insert_rec->e_cpos; | |
3904 | le32_add_cpu(&rec->e_int_clusters, | |
3905 | le16_to_cpu(insert_rec->e_leaf_clusters)); | |
3906 | le32_add_cpu(&rec->e_int_clusters, | |
3907 | -le32_to_cpu(rec->e_cpos)); | |
3908 | ||
3909 | ret = ocfs2_journal_dirty(handle, bh); | |
3910 | if (ret) | |
3911 | mlog_errno(ret); | |
3912 | ||
3913 | } | |
3914 | } | |
3915 | ||
d401dc12 JB |
3916 | static int ocfs2_append_rec_to_path(handle_t *handle, |
3917 | struct ocfs2_extent_tree *et, | |
328d5752 MF |
3918 | struct ocfs2_extent_rec *insert_rec, |
3919 | struct ocfs2_path *right_path, | |
3920 | struct ocfs2_path **ret_left_path) | |
3921 | { | |
3922 | int ret, next_free; | |
3923 | struct ocfs2_extent_list *el; | |
3924 | struct ocfs2_path *left_path = NULL; | |
3925 | ||
3926 | *ret_left_path = NULL; | |
3927 | ||
3928 | /* | |
3929 | * This shouldn't happen for non-trees. The extent rec cluster | |
3930 | * count manipulation below only works for interior nodes. | |
3931 | */ | |
3932 | BUG_ON(right_path->p_tree_depth == 0); | |
3933 | ||
3934 | /* | |
3935 | * If our appending insert is at the leftmost edge of a leaf, | |
3936 | * then we might need to update the rightmost records of the | |
3937 | * neighboring path. | |
3938 | */ | |
3939 | el = path_leaf_el(right_path); | |
3940 | next_free = le16_to_cpu(el->l_next_free_rec); | |
3941 | if (next_free == 0 || | |
3942 | (next_free == 1 && ocfs2_is_empty_extent(&el->l_recs[0]))) { | |
3943 | u32 left_cpos; | |
3944 | ||
d401dc12 JB |
3945 | ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci), |
3946 | right_path, &left_cpos); | |
328d5752 MF |
3947 | if (ret) { |
3948 | mlog_errno(ret); | |
dcd0538f MF |
3949 | goto out; |
3950 | } | |
3951 | ||
328d5752 MF |
3952 | mlog(0, "Append may need a left path update. cpos: %u, " |
3953 | "left_cpos: %u\n", le32_to_cpu(insert_rec->e_cpos), | |
3954 | left_cpos); | |
e48edee2 | 3955 | |
328d5752 MF |
3956 | /* |
3957 | * No need to worry if the append is already in the | |
3958 | * leftmost leaf. | |
3959 | */ | |
3960 | if (left_cpos) { | |
ffdd7a54 | 3961 | left_path = ocfs2_new_path_from_path(right_path); |
328d5752 MF |
3962 | if (!left_path) { |
3963 | ret = -ENOMEM; | |
3964 | mlog_errno(ret); | |
3965 | goto out; | |
3966 | } | |
dcd0538f | 3967 | |
d401dc12 | 3968 | ret = ocfs2_find_path(et->et_ci, left_path, |
facdb77f | 3969 | left_cpos); |
328d5752 MF |
3970 | if (ret) { |
3971 | mlog_errno(ret); | |
3972 | goto out; | |
3973 | } | |
dcd0538f | 3974 | |
328d5752 MF |
3975 | /* |
3976 | * ocfs2_insert_path() will pass the left_path to the | |
3977 | * journal for us. | |
3978 | */ | |
3979 | } | |
3980 | } | |
dcd0538f | 3981 | |
d401dc12 | 3982 | ret = ocfs2_journal_access_path(et->et_ci, handle, right_path); |
328d5752 MF |
3983 | if (ret) { |
3984 | mlog_errno(ret); | |
3985 | goto out; | |
dcd0538f MF |
3986 | } |
3987 | ||
d401dc12 | 3988 | ocfs2_adjust_rightmost_records(handle, et, right_path, insert_rec); |
328d5752 | 3989 | |
dcd0538f MF |
3990 | *ret_left_path = left_path; |
3991 | ret = 0; | |
3992 | out: | |
3993 | if (ret != 0) | |
3994 | ocfs2_free_path(left_path); | |
3995 | ||
3996 | return ret; | |
3997 | } | |
3998 | ||
328d5752 MF |
3999 | static void ocfs2_split_record(struct inode *inode, |
4000 | struct ocfs2_path *left_path, | |
4001 | struct ocfs2_path *right_path, | |
4002 | struct ocfs2_extent_rec *split_rec, | |
4003 | enum ocfs2_split_type split) | |
4004 | { | |
4005 | int index; | |
4006 | u32 cpos = le32_to_cpu(split_rec->e_cpos); | |
4007 | struct ocfs2_extent_list *left_el = NULL, *right_el, *insert_el, *el; | |
4008 | struct ocfs2_extent_rec *rec, *tmprec; | |
4009 | ||
c19a28e1 | 4010 | right_el = path_leaf_el(right_path); |
328d5752 MF |
4011 | if (left_path) |
4012 | left_el = path_leaf_el(left_path); | |
4013 | ||
4014 | el = right_el; | |
4015 | insert_el = right_el; | |
4016 | index = ocfs2_search_extent_list(el, cpos); | |
4017 | if (index != -1) { | |
4018 | if (index == 0 && left_path) { | |
4019 | BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0])); | |
4020 | ||
4021 | /* | |
4022 | * This typically means that the record | |
4023 | * started in the left path but moved to the | |
4024 | * right as a result of rotation. We either | |
4025 | * move the existing record to the left, or we | |
4026 | * do the later insert there. | |
4027 | * | |
4028 | * In this case, the left path should always | |
4029 | * exist as the rotate code will have passed | |
4030 | * it back for a post-insert update. | |
4031 | */ | |
4032 | ||
4033 | if (split == SPLIT_LEFT) { | |
4034 | /* | |
4035 | * It's a left split. Since we know | |
4036 | * that the rotate code gave us an | |
4037 | * empty extent in the left path, we | |
4038 | * can just do the insert there. | |
4039 | */ | |
4040 | insert_el = left_el; | |
4041 | } else { | |
4042 | /* | |
4043 | * Right split - we have to move the | |
4044 | * existing record over to the left | |
4045 | * leaf. The insert will be into the | |
4046 | * newly created empty extent in the | |
4047 | * right leaf. | |
4048 | */ | |
4049 | tmprec = &right_el->l_recs[index]; | |
4050 | ocfs2_rotate_leaf(left_el, tmprec); | |
4051 | el = left_el; | |
4052 | ||
4053 | memset(tmprec, 0, sizeof(*tmprec)); | |
4054 | index = ocfs2_search_extent_list(left_el, cpos); | |
4055 | BUG_ON(index == -1); | |
4056 | } | |
4057 | } | |
4058 | } else { | |
4059 | BUG_ON(!left_path); | |
4060 | BUG_ON(!ocfs2_is_empty_extent(&left_el->l_recs[0])); | |
4061 | /* | |
4062 | * Left path is easy - we can just allow the insert to | |
4063 | * happen. | |
4064 | */ | |
4065 | el = left_el; | |
4066 | insert_el = left_el; | |
4067 | index = ocfs2_search_extent_list(el, cpos); | |
4068 | BUG_ON(index == -1); | |
4069 | } | |
4070 | ||
4071 | rec = &el->l_recs[index]; | |
4072 | ocfs2_subtract_from_rec(inode->i_sb, split, rec, split_rec); | |
4073 | ocfs2_rotate_leaf(insert_el, split_rec); | |
4074 | } | |
4075 | ||
dcd0538f | 4076 | /* |
e7d4cb6b TM |
4077 | * This function only does inserts on an allocation b-tree. For tree |
4078 | * depth = 0, ocfs2_insert_at_leaf() is called directly. | |
dcd0538f MF |
4079 | * |
4080 | * right_path is the path we want to do the actual insert | |
4081 | * in. left_path should only be passed in if we need to update that | |
4082 | * portion of the tree after an edge insert. | |
4083 | */ | |
4084 | static int ocfs2_insert_path(struct inode *inode, | |
4085 | handle_t *handle, | |
7dc02805 | 4086 | struct ocfs2_extent_tree *et, |
dcd0538f MF |
4087 | struct ocfs2_path *left_path, |
4088 | struct ocfs2_path *right_path, | |
4089 | struct ocfs2_extent_rec *insert_rec, | |
4090 | struct ocfs2_insert_type *insert) | |
4091 | { | |
4092 | int ret, subtree_index; | |
4093 | struct buffer_head *leaf_bh = path_leaf_bh(right_path); | |
dcd0538f | 4094 | |
dcd0538f MF |
4095 | if (left_path) { |
4096 | int credits = handle->h_buffer_credits; | |
4097 | ||
4098 | /* | |
4099 | * There's a chance that left_path got passed back to | |
4100 | * us without being accounted for in the | |
4101 | * journal. Extend our transaction here to be sure we | |
4102 | * can change those blocks. | |
4103 | */ | |
4104 | credits += left_path->p_tree_depth; | |
4105 | ||
4106 | ret = ocfs2_extend_trans(handle, credits); | |
4107 | if (ret < 0) { | |
4108 | mlog_errno(ret); | |
4109 | goto out; | |
4110 | } | |
4111 | ||
7dc02805 | 4112 | ret = ocfs2_journal_access_path(et->et_ci, handle, left_path); |
dcd0538f MF |
4113 | if (ret < 0) { |
4114 | mlog_errno(ret); | |
4115 | goto out; | |
4116 | } | |
4117 | } | |
4118 | ||
e8aed345 MF |
4119 | /* |
4120 | * Pass both paths to the journal. The majority of inserts | |
4121 | * will be touching all components anyway. | |
4122 | */ | |
7dc02805 | 4123 | ret = ocfs2_journal_access_path(et->et_ci, handle, right_path); |
e8aed345 MF |
4124 | if (ret < 0) { |
4125 | mlog_errno(ret); | |
4126 | goto out; | |
4127 | } | |
4128 | ||
328d5752 MF |
4129 | if (insert->ins_split != SPLIT_NONE) { |
4130 | /* | |
4131 | * We could call ocfs2_insert_at_leaf() for some types | |
c78bad11 | 4132 | * of splits, but it's easier to just let one separate |
328d5752 MF |
4133 | * function sort it all out. |
4134 | */ | |
4135 | ocfs2_split_record(inode, left_path, right_path, | |
4136 | insert_rec, insert->ins_split); | |
e8aed345 MF |
4137 | |
4138 | /* | |
4139 | * Split might have modified either leaf and we don't | |
4140 | * have a guarantee that the later edge insert will | |
4141 | * dirty this for us. | |
4142 | */ | |
4143 | if (left_path) | |
4144 | ret = ocfs2_journal_dirty(handle, | |
4145 | path_leaf_bh(left_path)); | |
4146 | if (ret) | |
4147 | mlog_errno(ret); | |
328d5752 MF |
4148 | } else |
4149 | ocfs2_insert_at_leaf(insert_rec, path_leaf_el(right_path), | |
4150 | insert, inode); | |
dcd0538f | 4151 | |
dcd0538f MF |
4152 | ret = ocfs2_journal_dirty(handle, leaf_bh); |
4153 | if (ret) | |
4154 | mlog_errno(ret); | |
4155 | ||
4156 | if (left_path) { | |
4157 | /* | |
4158 | * The rotate code has indicated that we need to fix | |
4159 | * up portions of the tree after the insert. | |
4160 | * | |
4161 | * XXX: Should we extend the transaction here? | |
4162 | */ | |
7dc02805 | 4163 | subtree_index = ocfs2_find_subtree_root(et, left_path, |
dcd0538f | 4164 | right_path); |
4619c73e JB |
4165 | ocfs2_complete_edge_insert(handle, left_path, right_path, |
4166 | subtree_index); | |
dcd0538f MF |
4167 | } |
4168 | ||
4169 | ret = 0; | |
4170 | out: | |
4171 | return ret; | |
4172 | } | |
4173 | ||
4174 | static int ocfs2_do_insert_extent(struct inode *inode, | |
4175 | handle_t *handle, | |
e7d4cb6b | 4176 | struct ocfs2_extent_tree *et, |
dcd0538f MF |
4177 | struct ocfs2_extent_rec *insert_rec, |
4178 | struct ocfs2_insert_type *type) | |
4179 | { | |
4180 | int ret, rotate = 0; | |
4181 | u32 cpos; | |
4182 | struct ocfs2_path *right_path = NULL; | |
4183 | struct ocfs2_path *left_path = NULL; | |
dcd0538f MF |
4184 | struct ocfs2_extent_list *el; |
4185 | ||
ce1d9ea6 | 4186 | el = et->et_root_el; |
dcd0538f | 4187 | |
d9a0a1f8 | 4188 | ret = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 4189 | OCFS2_JOURNAL_ACCESS_WRITE); |
dcd0538f MF |
4190 | if (ret) { |
4191 | mlog_errno(ret); | |
4192 | goto out; | |
4193 | } | |
4194 | ||
4195 | if (le16_to_cpu(el->l_tree_depth) == 0) { | |
4196 | ocfs2_insert_at_leaf(insert_rec, el, type, inode); | |
4197 | goto out_update_clusters; | |
4198 | } | |
4199 | ||
ffdd7a54 | 4200 | right_path = ocfs2_new_path_from_et(et); |
dcd0538f MF |
4201 | if (!right_path) { |
4202 | ret = -ENOMEM; | |
4203 | mlog_errno(ret); | |
4204 | goto out; | |
4205 | } | |
4206 | ||
4207 | /* | |
4208 | * Determine the path to start with. Rotations need the | |
4209 | * rightmost path, everything else can go directly to the | |
4210 | * target leaf. | |
4211 | */ | |
4212 | cpos = le32_to_cpu(insert_rec->e_cpos); | |
4213 | if (type->ins_appending == APPEND_NONE && | |
4214 | type->ins_contig == CONTIG_NONE) { | |
4215 | rotate = 1; | |
4216 | cpos = UINT_MAX; | |
4217 | } | |
4218 | ||
facdb77f | 4219 | ret = ocfs2_find_path(et->et_ci, right_path, cpos); |
dcd0538f MF |
4220 | if (ret) { |
4221 | mlog_errno(ret); | |
4222 | goto out; | |
4223 | } | |
4224 | ||
4225 | /* | |
4226 | * Rotations and appends need special treatment - they modify | |
4227 | * parts of the tree's above them. | |
4228 | * | |
4229 | * Both might pass back a path immediate to the left of the | |
4230 | * one being inserted to. This will be cause | |
4231 | * ocfs2_insert_path() to modify the rightmost records of | |
4232 | * left_path to account for an edge insert. | |
4233 | * | |
4234 | * XXX: When modifying this code, keep in mind that an insert | |
4235 | * can wind up skipping both of these two special cases... | |
4236 | */ | |
4237 | if (rotate) { | |
1bbf0b8d | 4238 | ret = ocfs2_rotate_tree_right(handle, et, type->ins_split, |
dcd0538f MF |
4239 | le32_to_cpu(insert_rec->e_cpos), |
4240 | right_path, &left_path); | |
4241 | if (ret) { | |
4242 | mlog_errno(ret); | |
4243 | goto out; | |
4244 | } | |
e8aed345 MF |
4245 | |
4246 | /* | |
4247 | * ocfs2_rotate_tree_right() might have extended the | |
4248 | * transaction without re-journaling our tree root. | |
4249 | */ | |
d9a0a1f8 | 4250 | ret = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 4251 | OCFS2_JOURNAL_ACCESS_WRITE); |
e8aed345 MF |
4252 | if (ret) { |
4253 | mlog_errno(ret); | |
4254 | goto out; | |
4255 | } | |
dcd0538f MF |
4256 | } else if (type->ins_appending == APPEND_TAIL |
4257 | && type->ins_contig != CONTIG_LEFT) { | |
d401dc12 | 4258 | ret = ocfs2_append_rec_to_path(handle, et, insert_rec, |
dcd0538f MF |
4259 | right_path, &left_path); |
4260 | if (ret) { | |
4261 | mlog_errno(ret); | |
4262 | goto out; | |
4263 | } | |
4264 | } | |
4265 | ||
7dc02805 | 4266 | ret = ocfs2_insert_path(inode, handle, et, left_path, right_path, |
dcd0538f MF |
4267 | insert_rec, type); |
4268 | if (ret) { | |
4269 | mlog_errno(ret); | |
4270 | goto out; | |
4271 | } | |
4272 | ||
4273 | out_update_clusters: | |
328d5752 | 4274 | if (type->ins_split == SPLIT_NONE) |
6136ca5f | 4275 | ocfs2_et_update_clusters(et, |
35dc0aa3 | 4276 | le16_to_cpu(insert_rec->e_leaf_clusters)); |
dcd0538f | 4277 | |
ce1d9ea6 | 4278 | ret = ocfs2_journal_dirty(handle, et->et_root_bh); |
dcd0538f MF |
4279 | if (ret) |
4280 | mlog_errno(ret); | |
4281 | ||
4282 | out: | |
4283 | ocfs2_free_path(left_path); | |
4284 | ocfs2_free_path(right_path); | |
4285 | ||
4286 | return ret; | |
4287 | } | |
4288 | ||
328d5752 | 4289 | static enum ocfs2_contig_type |
ad5a4d70 | 4290 | ocfs2_figure_merge_contig_type(struct inode *inode, struct ocfs2_path *path, |
328d5752 MF |
4291 | struct ocfs2_extent_list *el, int index, |
4292 | struct ocfs2_extent_rec *split_rec) | |
4293 | { | |
ad5a4d70 | 4294 | int status; |
328d5752 | 4295 | enum ocfs2_contig_type ret = CONTIG_NONE; |
ad5a4d70 TM |
4296 | u32 left_cpos, right_cpos; |
4297 | struct ocfs2_extent_rec *rec = NULL; | |
4298 | struct ocfs2_extent_list *new_el; | |
4299 | struct ocfs2_path *left_path = NULL, *right_path = NULL; | |
4300 | struct buffer_head *bh; | |
4301 | struct ocfs2_extent_block *eb; | |
4302 | ||
4303 | if (index > 0) { | |
4304 | rec = &el->l_recs[index - 1]; | |
4305 | } else if (path->p_tree_depth > 0) { | |
4306 | status = ocfs2_find_cpos_for_left_leaf(inode->i_sb, | |
4307 | path, &left_cpos); | |
4308 | if (status) | |
4309 | goto out; | |
4310 | ||
4311 | if (left_cpos != 0) { | |
ffdd7a54 | 4312 | left_path = ocfs2_new_path_from_path(path); |
ad5a4d70 TM |
4313 | if (!left_path) |
4314 | goto out; | |
4315 | ||
facdb77f JB |
4316 | status = ocfs2_find_path(INODE_CACHE(inode), |
4317 | left_path, left_cpos); | |
ad5a4d70 TM |
4318 | if (status) |
4319 | goto out; | |
4320 | ||
4321 | new_el = path_leaf_el(left_path); | |
4322 | ||
4323 | if (le16_to_cpu(new_el->l_next_free_rec) != | |
4324 | le16_to_cpu(new_el->l_count)) { | |
4325 | bh = path_leaf_bh(left_path); | |
4326 | eb = (struct ocfs2_extent_block *)bh->b_data; | |
5e96581a JB |
4327 | ocfs2_error(inode->i_sb, |
4328 | "Extent block #%llu has an " | |
4329 | "invalid l_next_free_rec of " | |
4330 | "%d. It should have " | |
4331 | "matched the l_count of %d", | |
4332 | (unsigned long long)le64_to_cpu(eb->h_blkno), | |
4333 | le16_to_cpu(new_el->l_next_free_rec), | |
4334 | le16_to_cpu(new_el->l_count)); | |
4335 | status = -EINVAL; | |
ad5a4d70 TM |
4336 | goto out; |
4337 | } | |
4338 | rec = &new_el->l_recs[ | |
4339 | le16_to_cpu(new_el->l_next_free_rec) - 1]; | |
4340 | } | |
4341 | } | |
328d5752 MF |
4342 | |
4343 | /* | |
4344 | * We're careful to check for an empty extent record here - | |
4345 | * the merge code will know what to do if it sees one. | |
4346 | */ | |
ad5a4d70 | 4347 | if (rec) { |
328d5752 MF |
4348 | if (index == 1 && ocfs2_is_empty_extent(rec)) { |
4349 | if (split_rec->e_cpos == el->l_recs[index].e_cpos) | |
4350 | ret = CONTIG_RIGHT; | |
4351 | } else { | |
4352 | ret = ocfs2_extent_contig(inode, rec, split_rec); | |
4353 | } | |
4354 | } | |
4355 | ||
ad5a4d70 TM |
4356 | rec = NULL; |
4357 | if (index < (le16_to_cpu(el->l_next_free_rec) - 1)) | |
4358 | rec = &el->l_recs[index + 1]; | |
4359 | else if (le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count) && | |
4360 | path->p_tree_depth > 0) { | |
4361 | status = ocfs2_find_cpos_for_right_leaf(inode->i_sb, | |
4362 | path, &right_cpos); | |
4363 | if (status) | |
4364 | goto out; | |
4365 | ||
4366 | if (right_cpos == 0) | |
4367 | goto out; | |
4368 | ||
ffdd7a54 | 4369 | right_path = ocfs2_new_path_from_path(path); |
ad5a4d70 TM |
4370 | if (!right_path) |
4371 | goto out; | |
4372 | ||
facdb77f | 4373 | status = ocfs2_find_path(INODE_CACHE(inode), right_path, right_cpos); |
ad5a4d70 TM |
4374 | if (status) |
4375 | goto out; | |
4376 | ||
4377 | new_el = path_leaf_el(right_path); | |
4378 | rec = &new_el->l_recs[0]; | |
4379 | if (ocfs2_is_empty_extent(rec)) { | |
4380 | if (le16_to_cpu(new_el->l_next_free_rec) <= 1) { | |
4381 | bh = path_leaf_bh(right_path); | |
4382 | eb = (struct ocfs2_extent_block *)bh->b_data; | |
5e96581a JB |
4383 | ocfs2_error(inode->i_sb, |
4384 | "Extent block #%llu has an " | |
4385 | "invalid l_next_free_rec of %d", | |
4386 | (unsigned long long)le64_to_cpu(eb->h_blkno), | |
4387 | le16_to_cpu(new_el->l_next_free_rec)); | |
4388 | status = -EINVAL; | |
ad5a4d70 TM |
4389 | goto out; |
4390 | } | |
4391 | rec = &new_el->l_recs[1]; | |
4392 | } | |
4393 | } | |
4394 | ||
4395 | if (rec) { | |
328d5752 MF |
4396 | enum ocfs2_contig_type contig_type; |
4397 | ||
328d5752 MF |
4398 | contig_type = ocfs2_extent_contig(inode, rec, split_rec); |
4399 | ||
4400 | if (contig_type == CONTIG_LEFT && ret == CONTIG_RIGHT) | |
4401 | ret = CONTIG_LEFTRIGHT; | |
4402 | else if (ret == CONTIG_NONE) | |
4403 | ret = contig_type; | |
4404 | } | |
4405 | ||
ad5a4d70 TM |
4406 | out: |
4407 | if (left_path) | |
4408 | ocfs2_free_path(left_path); | |
4409 | if (right_path) | |
4410 | ocfs2_free_path(right_path); | |
4411 | ||
328d5752 MF |
4412 | return ret; |
4413 | } | |
4414 | ||
dcd0538f MF |
4415 | static void ocfs2_figure_contig_type(struct inode *inode, |
4416 | struct ocfs2_insert_type *insert, | |
4417 | struct ocfs2_extent_list *el, | |
ca12b7c4 TM |
4418 | struct ocfs2_extent_rec *insert_rec, |
4419 | struct ocfs2_extent_tree *et) | |
dcd0538f MF |
4420 | { |
4421 | int i; | |
4422 | enum ocfs2_contig_type contig_type = CONTIG_NONE; | |
4423 | ||
e48edee2 MF |
4424 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); |
4425 | ||
dcd0538f MF |
4426 | for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) { |
4427 | contig_type = ocfs2_extent_contig(inode, &el->l_recs[i], | |
4428 | insert_rec); | |
4429 | if (contig_type != CONTIG_NONE) { | |
4430 | insert->ins_contig_index = i; | |
4431 | break; | |
4432 | } | |
4433 | } | |
4434 | insert->ins_contig = contig_type; | |
ca12b7c4 TM |
4435 | |
4436 | if (insert->ins_contig != CONTIG_NONE) { | |
4437 | struct ocfs2_extent_rec *rec = | |
4438 | &el->l_recs[insert->ins_contig_index]; | |
4439 | unsigned int len = le16_to_cpu(rec->e_leaf_clusters) + | |
4440 | le16_to_cpu(insert_rec->e_leaf_clusters); | |
4441 | ||
4442 | /* | |
4443 | * Caller might want us to limit the size of extents, don't | |
4444 | * calculate contiguousness if we might exceed that limit. | |
4445 | */ | |
ce1d9ea6 JB |
4446 | if (et->et_max_leaf_clusters && |
4447 | (len > et->et_max_leaf_clusters)) | |
ca12b7c4 TM |
4448 | insert->ins_contig = CONTIG_NONE; |
4449 | } | |
dcd0538f MF |
4450 | } |
4451 | ||
4452 | /* | |
4453 | * This should only be called against the righmost leaf extent list. | |
4454 | * | |
4455 | * ocfs2_figure_appending_type() will figure out whether we'll have to | |
4456 | * insert at the tail of the rightmost leaf. | |
4457 | * | |
e7d4cb6b TM |
4458 | * This should also work against the root extent list for tree's with 0 |
4459 | * depth. If we consider the root extent list to be the rightmost leaf node | |
dcd0538f MF |
4460 | * then the logic here makes sense. |
4461 | */ | |
4462 | static void ocfs2_figure_appending_type(struct ocfs2_insert_type *insert, | |
4463 | struct ocfs2_extent_list *el, | |
4464 | struct ocfs2_extent_rec *insert_rec) | |
4465 | { | |
4466 | int i; | |
4467 | u32 cpos = le32_to_cpu(insert_rec->e_cpos); | |
4468 | struct ocfs2_extent_rec *rec; | |
4469 | ||
4470 | insert->ins_appending = APPEND_NONE; | |
4471 | ||
e48edee2 | 4472 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); |
dcd0538f MF |
4473 | |
4474 | if (!el->l_next_free_rec) | |
4475 | goto set_tail_append; | |
4476 | ||
4477 | if (ocfs2_is_empty_extent(&el->l_recs[0])) { | |
4478 | /* Were all records empty? */ | |
4479 | if (le16_to_cpu(el->l_next_free_rec) == 1) | |
4480 | goto set_tail_append; | |
4481 | } | |
4482 | ||
4483 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
4484 | rec = &el->l_recs[i]; | |
4485 | ||
e48edee2 MF |
4486 | if (cpos >= |
4487 | (le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters))) | |
dcd0538f MF |
4488 | goto set_tail_append; |
4489 | ||
4490 | return; | |
4491 | ||
4492 | set_tail_append: | |
4493 | insert->ins_appending = APPEND_TAIL; | |
4494 | } | |
4495 | ||
4496 | /* | |
4497 | * Helper function called at the begining of an insert. | |
4498 | * | |
4499 | * This computes a few things that are commonly used in the process of | |
4500 | * inserting into the btree: | |
4501 | * - Whether the new extent is contiguous with an existing one. | |
4502 | * - The current tree depth. | |
4503 | * - Whether the insert is an appending one. | |
4504 | * - The total # of free records in the tree. | |
4505 | * | |
4506 | * All of the information is stored on the ocfs2_insert_type | |
4507 | * structure. | |
4508 | */ | |
4509 | static int ocfs2_figure_insert_type(struct inode *inode, | |
e7d4cb6b | 4510 | struct ocfs2_extent_tree *et, |
dcd0538f MF |
4511 | struct buffer_head **last_eb_bh, |
4512 | struct ocfs2_extent_rec *insert_rec, | |
c77534f6 | 4513 | int *free_records, |
dcd0538f MF |
4514 | struct ocfs2_insert_type *insert) |
4515 | { | |
4516 | int ret; | |
dcd0538f MF |
4517 | struct ocfs2_extent_block *eb; |
4518 | struct ocfs2_extent_list *el; | |
4519 | struct ocfs2_path *path = NULL; | |
4520 | struct buffer_head *bh = NULL; | |
4521 | ||
328d5752 MF |
4522 | insert->ins_split = SPLIT_NONE; |
4523 | ||
ce1d9ea6 | 4524 | el = et->et_root_el; |
dcd0538f MF |
4525 | insert->ins_tree_depth = le16_to_cpu(el->l_tree_depth); |
4526 | ||
4527 | if (el->l_tree_depth) { | |
4528 | /* | |
4529 | * If we have tree depth, we read in the | |
4530 | * rightmost extent block ahead of time as | |
4531 | * ocfs2_figure_insert_type() and ocfs2_add_branch() | |
4532 | * may want it later. | |
4533 | */ | |
3d03a305 | 4534 | ret = ocfs2_read_extent_block(et->et_ci, |
5e96581a JB |
4535 | ocfs2_et_get_last_eb_blk(et), |
4536 | &bh); | |
dcd0538f MF |
4537 | if (ret) { |
4538 | mlog_exit(ret); | |
4539 | goto out; | |
4540 | } | |
ccd979bd | 4541 | eb = (struct ocfs2_extent_block *) bh->b_data; |
ccd979bd | 4542 | el = &eb->h_list; |
dcd0538f | 4543 | } |
ccd979bd | 4544 | |
dcd0538f MF |
4545 | /* |
4546 | * Unless we have a contiguous insert, we'll need to know if | |
4547 | * there is room left in our allocation tree for another | |
4548 | * extent record. | |
4549 | * | |
4550 | * XXX: This test is simplistic, we can search for empty | |
4551 | * extent records too. | |
4552 | */ | |
c77534f6 | 4553 | *free_records = le16_to_cpu(el->l_count) - |
dcd0538f MF |
4554 | le16_to_cpu(el->l_next_free_rec); |
4555 | ||
4556 | if (!insert->ins_tree_depth) { | |
ca12b7c4 | 4557 | ocfs2_figure_contig_type(inode, insert, el, insert_rec, et); |
dcd0538f MF |
4558 | ocfs2_figure_appending_type(insert, el, insert_rec); |
4559 | return 0; | |
ccd979bd MF |
4560 | } |
4561 | ||
ffdd7a54 | 4562 | path = ocfs2_new_path_from_et(et); |
dcd0538f MF |
4563 | if (!path) { |
4564 | ret = -ENOMEM; | |
4565 | mlog_errno(ret); | |
4566 | goto out; | |
4567 | } | |
ccd979bd | 4568 | |
dcd0538f MF |
4569 | /* |
4570 | * In the case that we're inserting past what the tree | |
4571 | * currently accounts for, ocfs2_find_path() will return for | |
4572 | * us the rightmost tree path. This is accounted for below in | |
4573 | * the appending code. | |
4574 | */ | |
facdb77f | 4575 | ret = ocfs2_find_path(et->et_ci, path, le32_to_cpu(insert_rec->e_cpos)); |
dcd0538f MF |
4576 | if (ret) { |
4577 | mlog_errno(ret); | |
4578 | goto out; | |
4579 | } | |
ccd979bd | 4580 | |
dcd0538f MF |
4581 | el = path_leaf_el(path); |
4582 | ||
4583 | /* | |
4584 | * Now that we have the path, there's two things we want to determine: | |
4585 | * 1) Contiguousness (also set contig_index if this is so) | |
4586 | * | |
4587 | * 2) Are we doing an append? We can trivially break this up | |
4588 | * into two types of appends: simple record append, or a | |
4589 | * rotate inside the tail leaf. | |
4590 | */ | |
ca12b7c4 | 4591 | ocfs2_figure_contig_type(inode, insert, el, insert_rec, et); |
dcd0538f MF |
4592 | |
4593 | /* | |
4594 | * The insert code isn't quite ready to deal with all cases of | |
4595 | * left contiguousness. Specifically, if it's an insert into | |
4596 | * the 1st record in a leaf, it will require the adjustment of | |
e48edee2 | 4597 | * cluster count on the last record of the path directly to it's |
dcd0538f MF |
4598 | * left. For now, just catch that case and fool the layers |
4599 | * above us. This works just fine for tree_depth == 0, which | |
4600 | * is why we allow that above. | |
4601 | */ | |
4602 | if (insert->ins_contig == CONTIG_LEFT && | |
4603 | insert->ins_contig_index == 0) | |
4604 | insert->ins_contig = CONTIG_NONE; | |
4605 | ||
4606 | /* | |
4607 | * Ok, so we can simply compare against last_eb to figure out | |
4608 | * whether the path doesn't exist. This will only happen in | |
4609 | * the case that we're doing a tail append, so maybe we can | |
4610 | * take advantage of that information somehow. | |
4611 | */ | |
35dc0aa3 | 4612 | if (ocfs2_et_get_last_eb_blk(et) == |
e7d4cb6b | 4613 | path_leaf_bh(path)->b_blocknr) { |
dcd0538f MF |
4614 | /* |
4615 | * Ok, ocfs2_find_path() returned us the rightmost | |
4616 | * tree path. This might be an appending insert. There are | |
4617 | * two cases: | |
4618 | * 1) We're doing a true append at the tail: | |
4619 | * -This might even be off the end of the leaf | |
4620 | * 2) We're "appending" by rotating in the tail | |
4621 | */ | |
4622 | ocfs2_figure_appending_type(insert, el, insert_rec); | |
4623 | } | |
4624 | ||
4625 | out: | |
4626 | ocfs2_free_path(path); | |
4627 | ||
4628 | if (ret == 0) | |
4629 | *last_eb_bh = bh; | |
4630 | else | |
4631 | brelse(bh); | |
4632 | return ret; | |
ccd979bd MF |
4633 | } |
4634 | ||
dcd0538f MF |
4635 | /* |
4636 | * Insert an extent into an inode btree. | |
4637 | * | |
4638 | * The caller needs to update fe->i_clusters | |
4639 | */ | |
f99b9b7c JB |
4640 | int ocfs2_insert_extent(struct ocfs2_super *osb, |
4641 | handle_t *handle, | |
4642 | struct inode *inode, | |
4643 | struct ocfs2_extent_tree *et, | |
4644 | u32 cpos, | |
4645 | u64 start_blk, | |
4646 | u32 new_clusters, | |
4647 | u8 flags, | |
4648 | struct ocfs2_alloc_context *meta_ac) | |
ccd979bd | 4649 | { |
c3afcbb3 | 4650 | int status; |
c77534f6 | 4651 | int uninitialized_var(free_records); |
ccd979bd | 4652 | struct buffer_head *last_eb_bh = NULL; |
dcd0538f MF |
4653 | struct ocfs2_insert_type insert = {0, }; |
4654 | struct ocfs2_extent_rec rec; | |
4655 | ||
4656 | mlog(0, "add %u clusters at position %u to inode %llu\n", | |
4657 | new_clusters, cpos, (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
4658 | ||
e48edee2 | 4659 | memset(&rec, 0, sizeof(rec)); |
dcd0538f MF |
4660 | rec.e_cpos = cpu_to_le32(cpos); |
4661 | rec.e_blkno = cpu_to_le64(start_blk); | |
e48edee2 | 4662 | rec.e_leaf_clusters = cpu_to_le16(new_clusters); |
2ae99a60 | 4663 | rec.e_flags = flags; |
6136ca5f | 4664 | status = ocfs2_et_insert_check(et, &rec); |
1e61ee79 JB |
4665 | if (status) { |
4666 | mlog_errno(status); | |
4667 | goto bail; | |
4668 | } | |
dcd0538f | 4669 | |
e7d4cb6b | 4670 | status = ocfs2_figure_insert_type(inode, et, &last_eb_bh, &rec, |
c77534f6 | 4671 | &free_records, &insert); |
dcd0538f MF |
4672 | if (status < 0) { |
4673 | mlog_errno(status); | |
4674 | goto bail; | |
ccd979bd MF |
4675 | } |
4676 | ||
dcd0538f MF |
4677 | mlog(0, "Insert.appending: %u, Insert.Contig: %u, " |
4678 | "Insert.contig_index: %d, Insert.free_records: %d, " | |
4679 | "Insert.tree_depth: %d\n", | |
4680 | insert.ins_appending, insert.ins_contig, insert.ins_contig_index, | |
c77534f6 | 4681 | free_records, insert.ins_tree_depth); |
ccd979bd | 4682 | |
c77534f6 | 4683 | if (insert.ins_contig == CONTIG_NONE && free_records == 0) { |
d401dc12 | 4684 | status = ocfs2_grow_tree(handle, et, |
328d5752 | 4685 | &insert.ins_tree_depth, &last_eb_bh, |
c3afcbb3 MF |
4686 | meta_ac); |
4687 | if (status) { | |
ccd979bd MF |
4688 | mlog_errno(status); |
4689 | goto bail; | |
4690 | } | |
ccd979bd MF |
4691 | } |
4692 | ||
dcd0538f | 4693 | /* Finally, we can add clusters. This might rotate the tree for us. */ |
e7d4cb6b | 4694 | status = ocfs2_do_insert_extent(inode, handle, et, &rec, &insert); |
ccd979bd MF |
4695 | if (status < 0) |
4696 | mlog_errno(status); | |
f99b9b7c | 4697 | else if (et->et_ops == &ocfs2_dinode_et_ops) |
83418978 | 4698 | ocfs2_extent_map_insert_rec(inode, &rec); |
ccd979bd MF |
4699 | |
4700 | bail: | |
a81cb88b | 4701 | brelse(last_eb_bh); |
ccd979bd | 4702 | |
f56654c4 TM |
4703 | mlog_exit(status); |
4704 | return status; | |
4705 | } | |
4706 | ||
0eb8d47e TM |
4707 | /* |
4708 | * Allcate and add clusters into the extent b-tree. | |
4709 | * The new clusters(clusters_to_add) will be inserted at logical_offset. | |
f99b9b7c | 4710 | * The extent b-tree's root is specified by et, and |
0eb8d47e TM |
4711 | * it is not limited to the file storage. Any extent tree can use this |
4712 | * function if it implements the proper ocfs2_extent_tree. | |
4713 | */ | |
4714 | int ocfs2_add_clusters_in_btree(struct ocfs2_super *osb, | |
4715 | struct inode *inode, | |
4716 | u32 *logical_offset, | |
4717 | u32 clusters_to_add, | |
4718 | int mark_unwritten, | |
f99b9b7c | 4719 | struct ocfs2_extent_tree *et, |
0eb8d47e TM |
4720 | handle_t *handle, |
4721 | struct ocfs2_alloc_context *data_ac, | |
4722 | struct ocfs2_alloc_context *meta_ac, | |
f99b9b7c | 4723 | enum ocfs2_alloc_restarted *reason_ret) |
0eb8d47e TM |
4724 | { |
4725 | int status = 0; | |
4726 | int free_extents; | |
4727 | enum ocfs2_alloc_restarted reason = RESTART_NONE; | |
4728 | u32 bit_off, num_bits; | |
4729 | u64 block; | |
4730 | u8 flags = 0; | |
4731 | ||
4732 | BUG_ON(!clusters_to_add); | |
4733 | ||
4734 | if (mark_unwritten) | |
4735 | flags = OCFS2_EXT_UNWRITTEN; | |
4736 | ||
3d03a305 | 4737 | free_extents = ocfs2_num_free_extents(osb, et); |
0eb8d47e TM |
4738 | if (free_extents < 0) { |
4739 | status = free_extents; | |
4740 | mlog_errno(status); | |
4741 | goto leave; | |
4742 | } | |
4743 | ||
4744 | /* there are two cases which could cause us to EAGAIN in the | |
4745 | * we-need-more-metadata case: | |
4746 | * 1) we haven't reserved *any* | |
4747 | * 2) we are so fragmented, we've needed to add metadata too | |
4748 | * many times. */ | |
4749 | if (!free_extents && !meta_ac) { | |
4750 | mlog(0, "we haven't reserved any metadata!\n"); | |
4751 | status = -EAGAIN; | |
4752 | reason = RESTART_META; | |
4753 | goto leave; | |
4754 | } else if ((!free_extents) | |
4755 | && (ocfs2_alloc_context_bits_left(meta_ac) | |
f99b9b7c | 4756 | < ocfs2_extend_meta_needed(et->et_root_el))) { |
0eb8d47e TM |
4757 | mlog(0, "filesystem is really fragmented...\n"); |
4758 | status = -EAGAIN; | |
4759 | reason = RESTART_META; | |
4760 | goto leave; | |
4761 | } | |
4762 | ||
4763 | status = __ocfs2_claim_clusters(osb, handle, data_ac, 1, | |
4764 | clusters_to_add, &bit_off, &num_bits); | |
4765 | if (status < 0) { | |
4766 | if (status != -ENOSPC) | |
4767 | mlog_errno(status); | |
4768 | goto leave; | |
4769 | } | |
4770 | ||
4771 | BUG_ON(num_bits > clusters_to_add); | |
4772 | ||
13723d00 | 4773 | /* reserve our write early -- insert_extent may update the tree root */ |
d9a0a1f8 | 4774 | status = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 4775 | OCFS2_JOURNAL_ACCESS_WRITE); |
0eb8d47e TM |
4776 | if (status < 0) { |
4777 | mlog_errno(status); | |
4778 | goto leave; | |
4779 | } | |
4780 | ||
4781 | block = ocfs2_clusters_to_blocks(osb->sb, bit_off); | |
4782 | mlog(0, "Allocating %u clusters at block %u for inode %llu\n", | |
4783 | num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
f99b9b7c JB |
4784 | status = ocfs2_insert_extent(osb, handle, inode, et, |
4785 | *logical_offset, block, | |
4786 | num_bits, flags, meta_ac); | |
0eb8d47e TM |
4787 | if (status < 0) { |
4788 | mlog_errno(status); | |
4789 | goto leave; | |
4790 | } | |
4791 | ||
f99b9b7c | 4792 | status = ocfs2_journal_dirty(handle, et->et_root_bh); |
0eb8d47e TM |
4793 | if (status < 0) { |
4794 | mlog_errno(status); | |
4795 | goto leave; | |
4796 | } | |
4797 | ||
4798 | clusters_to_add -= num_bits; | |
4799 | *logical_offset += num_bits; | |
4800 | ||
4801 | if (clusters_to_add) { | |
4802 | mlog(0, "need to alloc once more, wanted = %u\n", | |
4803 | clusters_to_add); | |
4804 | status = -EAGAIN; | |
4805 | reason = RESTART_TRANS; | |
4806 | } | |
4807 | ||
4808 | leave: | |
4809 | mlog_exit(status); | |
4810 | if (reason_ret) | |
4811 | *reason_ret = reason; | |
4812 | return status; | |
4813 | } | |
4814 | ||
328d5752 MF |
4815 | static void ocfs2_make_right_split_rec(struct super_block *sb, |
4816 | struct ocfs2_extent_rec *split_rec, | |
4817 | u32 cpos, | |
4818 | struct ocfs2_extent_rec *rec) | |
4819 | { | |
4820 | u32 rec_cpos = le32_to_cpu(rec->e_cpos); | |
4821 | u32 rec_range = rec_cpos + le16_to_cpu(rec->e_leaf_clusters); | |
4822 | ||
4823 | memset(split_rec, 0, sizeof(struct ocfs2_extent_rec)); | |
4824 | ||
4825 | split_rec->e_cpos = cpu_to_le32(cpos); | |
4826 | split_rec->e_leaf_clusters = cpu_to_le16(rec_range - cpos); | |
4827 | ||
4828 | split_rec->e_blkno = rec->e_blkno; | |
4829 | le64_add_cpu(&split_rec->e_blkno, | |
4830 | ocfs2_clusters_to_blocks(sb, cpos - rec_cpos)); | |
4831 | ||
4832 | split_rec->e_flags = rec->e_flags; | |
4833 | } | |
4834 | ||
4835 | static int ocfs2_split_and_insert(struct inode *inode, | |
4836 | handle_t *handle, | |
4837 | struct ocfs2_path *path, | |
e7d4cb6b | 4838 | struct ocfs2_extent_tree *et, |
328d5752 MF |
4839 | struct buffer_head **last_eb_bh, |
4840 | int split_index, | |
4841 | struct ocfs2_extent_rec *orig_split_rec, | |
4842 | struct ocfs2_alloc_context *meta_ac) | |
4843 | { | |
4844 | int ret = 0, depth; | |
4845 | unsigned int insert_range, rec_range, do_leftright = 0; | |
4846 | struct ocfs2_extent_rec tmprec; | |
4847 | struct ocfs2_extent_list *rightmost_el; | |
4848 | struct ocfs2_extent_rec rec; | |
4849 | struct ocfs2_extent_rec split_rec = *orig_split_rec; | |
4850 | struct ocfs2_insert_type insert; | |
4851 | struct ocfs2_extent_block *eb; | |
328d5752 MF |
4852 | |
4853 | leftright: | |
4854 | /* | |
4855 | * Store a copy of the record on the stack - it might move | |
4856 | * around as the tree is manipulated below. | |
4857 | */ | |
4858 | rec = path_leaf_el(path)->l_recs[split_index]; | |
4859 | ||
ce1d9ea6 | 4860 | rightmost_el = et->et_root_el; |
328d5752 MF |
4861 | |
4862 | depth = le16_to_cpu(rightmost_el->l_tree_depth); | |
4863 | if (depth) { | |
4864 | BUG_ON(!(*last_eb_bh)); | |
4865 | eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data; | |
4866 | rightmost_el = &eb->h_list; | |
4867 | } | |
4868 | ||
4869 | if (le16_to_cpu(rightmost_el->l_next_free_rec) == | |
4870 | le16_to_cpu(rightmost_el->l_count)) { | |
d401dc12 | 4871 | ret = ocfs2_grow_tree(handle, et, |
e7d4cb6b | 4872 | &depth, last_eb_bh, meta_ac); |
328d5752 MF |
4873 | if (ret) { |
4874 | mlog_errno(ret); | |
4875 | goto out; | |
4876 | } | |
328d5752 MF |
4877 | } |
4878 | ||
4879 | memset(&insert, 0, sizeof(struct ocfs2_insert_type)); | |
4880 | insert.ins_appending = APPEND_NONE; | |
4881 | insert.ins_contig = CONTIG_NONE; | |
328d5752 MF |
4882 | insert.ins_tree_depth = depth; |
4883 | ||
4884 | insert_range = le32_to_cpu(split_rec.e_cpos) + | |
4885 | le16_to_cpu(split_rec.e_leaf_clusters); | |
4886 | rec_range = le32_to_cpu(rec.e_cpos) + | |
4887 | le16_to_cpu(rec.e_leaf_clusters); | |
4888 | ||
4889 | if (split_rec.e_cpos == rec.e_cpos) { | |
4890 | insert.ins_split = SPLIT_LEFT; | |
4891 | } else if (insert_range == rec_range) { | |
4892 | insert.ins_split = SPLIT_RIGHT; | |
4893 | } else { | |
4894 | /* | |
4895 | * Left/right split. We fake this as a right split | |
4896 | * first and then make a second pass as a left split. | |
4897 | */ | |
4898 | insert.ins_split = SPLIT_RIGHT; | |
4899 | ||
4900 | ocfs2_make_right_split_rec(inode->i_sb, &tmprec, insert_range, | |
4901 | &rec); | |
4902 | ||
4903 | split_rec = tmprec; | |
4904 | ||
4905 | BUG_ON(do_leftright); | |
4906 | do_leftright = 1; | |
4907 | } | |
4908 | ||
e7d4cb6b | 4909 | ret = ocfs2_do_insert_extent(inode, handle, et, &split_rec, &insert); |
328d5752 MF |
4910 | if (ret) { |
4911 | mlog_errno(ret); | |
4912 | goto out; | |
4913 | } | |
4914 | ||
4915 | if (do_leftright == 1) { | |
4916 | u32 cpos; | |
4917 | struct ocfs2_extent_list *el; | |
4918 | ||
4919 | do_leftright++; | |
4920 | split_rec = *orig_split_rec; | |
4921 | ||
4922 | ocfs2_reinit_path(path, 1); | |
4923 | ||
4924 | cpos = le32_to_cpu(split_rec.e_cpos); | |
facdb77f | 4925 | ret = ocfs2_find_path(et->et_ci, path, cpos); |
328d5752 MF |
4926 | if (ret) { |
4927 | mlog_errno(ret); | |
4928 | goto out; | |
4929 | } | |
4930 | ||
4931 | el = path_leaf_el(path); | |
4932 | split_index = ocfs2_search_extent_list(el, cpos); | |
4933 | goto leftright; | |
4934 | } | |
4935 | out: | |
4936 | ||
4937 | return ret; | |
4938 | } | |
4939 | ||
47be12e4 TM |
4940 | static int ocfs2_replace_extent_rec(struct inode *inode, |
4941 | handle_t *handle, | |
4942 | struct ocfs2_path *path, | |
4943 | struct ocfs2_extent_list *el, | |
4944 | int split_index, | |
4945 | struct ocfs2_extent_rec *split_rec) | |
4946 | { | |
4947 | int ret; | |
4948 | ||
0cf2f763 | 4949 | ret = ocfs2_path_bh_journal_access(handle, INODE_CACHE(inode), path, |
47be12e4 TM |
4950 | path_num_items(path) - 1); |
4951 | if (ret) { | |
4952 | mlog_errno(ret); | |
4953 | goto out; | |
4954 | } | |
4955 | ||
4956 | el->l_recs[split_index] = *split_rec; | |
4957 | ||
4958 | ocfs2_journal_dirty(handle, path_leaf_bh(path)); | |
4959 | out: | |
4960 | return ret; | |
4961 | } | |
4962 | ||
328d5752 MF |
4963 | /* |
4964 | * Mark part or all of the extent record at split_index in the leaf | |
4965 | * pointed to by path as written. This removes the unwritten | |
4966 | * extent flag. | |
4967 | * | |
4968 | * Care is taken to handle contiguousness so as to not grow the tree. | |
4969 | * | |
4970 | * meta_ac is not strictly necessary - we only truly need it if growth | |
4971 | * of the tree is required. All other cases will degrade into a less | |
4972 | * optimal tree layout. | |
4973 | * | |
e7d4cb6b TM |
4974 | * last_eb_bh should be the rightmost leaf block for any extent |
4975 | * btree. Since a split may grow the tree or a merge might shrink it, | |
4976 | * the caller cannot trust the contents of that buffer after this call. | |
328d5752 MF |
4977 | * |
4978 | * This code is optimized for readability - several passes might be | |
4979 | * made over certain portions of the tree. All of those blocks will | |
4980 | * have been brought into cache (and pinned via the journal), so the | |
4981 | * extra overhead is not expressed in terms of disk reads. | |
4982 | */ | |
4983 | static int __ocfs2_mark_extent_written(struct inode *inode, | |
e7d4cb6b | 4984 | struct ocfs2_extent_tree *et, |
328d5752 MF |
4985 | handle_t *handle, |
4986 | struct ocfs2_path *path, | |
4987 | int split_index, | |
4988 | struct ocfs2_extent_rec *split_rec, | |
4989 | struct ocfs2_alloc_context *meta_ac, | |
4990 | struct ocfs2_cached_dealloc_ctxt *dealloc) | |
4991 | { | |
4992 | int ret = 0; | |
4993 | struct ocfs2_extent_list *el = path_leaf_el(path); | |
e8aed345 | 4994 | struct buffer_head *last_eb_bh = NULL; |
328d5752 MF |
4995 | struct ocfs2_extent_rec *rec = &el->l_recs[split_index]; |
4996 | struct ocfs2_merge_ctxt ctxt; | |
4997 | struct ocfs2_extent_list *rightmost_el; | |
4998 | ||
3cf0c507 | 4999 | if (!(rec->e_flags & OCFS2_EXT_UNWRITTEN)) { |
328d5752 MF |
5000 | ret = -EIO; |
5001 | mlog_errno(ret); | |
5002 | goto out; | |
5003 | } | |
5004 | ||
5005 | if (le32_to_cpu(rec->e_cpos) > le32_to_cpu(split_rec->e_cpos) || | |
5006 | ((le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters)) < | |
5007 | (le32_to_cpu(split_rec->e_cpos) + le16_to_cpu(split_rec->e_leaf_clusters)))) { | |
5008 | ret = -EIO; | |
5009 | mlog_errno(ret); | |
5010 | goto out; | |
5011 | } | |
5012 | ||
ad5a4d70 | 5013 | ctxt.c_contig_type = ocfs2_figure_merge_contig_type(inode, path, el, |
328d5752 MF |
5014 | split_index, |
5015 | split_rec); | |
5016 | ||
5017 | /* | |
5018 | * The core merge / split code wants to know how much room is | |
5019 | * left in this inodes allocation tree, so we pass the | |
5020 | * rightmost extent list. | |
5021 | */ | |
5022 | if (path->p_tree_depth) { | |
5023 | struct ocfs2_extent_block *eb; | |
328d5752 | 5024 | |
3d03a305 | 5025 | ret = ocfs2_read_extent_block(et->et_ci, |
5e96581a JB |
5026 | ocfs2_et_get_last_eb_blk(et), |
5027 | &last_eb_bh); | |
328d5752 MF |
5028 | if (ret) { |
5029 | mlog_exit(ret); | |
5030 | goto out; | |
5031 | } | |
5032 | ||
5033 | eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; | |
328d5752 MF |
5034 | rightmost_el = &eb->h_list; |
5035 | } else | |
5036 | rightmost_el = path_root_el(path); | |
5037 | ||
328d5752 MF |
5038 | if (rec->e_cpos == split_rec->e_cpos && |
5039 | rec->e_leaf_clusters == split_rec->e_leaf_clusters) | |
5040 | ctxt.c_split_covers_rec = 1; | |
5041 | else | |
5042 | ctxt.c_split_covers_rec = 0; | |
5043 | ||
5044 | ctxt.c_has_empty_extent = ocfs2_is_empty_extent(&el->l_recs[0]); | |
5045 | ||
015452b1 MF |
5046 | mlog(0, "index: %d, contig: %u, has_empty: %u, split_covers: %u\n", |
5047 | split_index, ctxt.c_contig_type, ctxt.c_has_empty_extent, | |
5048 | ctxt.c_split_covers_rec); | |
328d5752 MF |
5049 | |
5050 | if (ctxt.c_contig_type == CONTIG_NONE) { | |
5051 | if (ctxt.c_split_covers_rec) | |
47be12e4 TM |
5052 | ret = ocfs2_replace_extent_rec(inode, handle, |
5053 | path, el, | |
5054 | split_index, split_rec); | |
328d5752 | 5055 | else |
e7d4cb6b | 5056 | ret = ocfs2_split_and_insert(inode, handle, path, et, |
328d5752 MF |
5057 | &last_eb_bh, split_index, |
5058 | split_rec, meta_ac); | |
5059 | if (ret) | |
5060 | mlog_errno(ret); | |
5061 | } else { | |
c495dd24 | 5062 | ret = ocfs2_try_to_merge_extent(handle, et, path, |
328d5752 | 5063 | split_index, split_rec, |
c495dd24 | 5064 | dealloc, &ctxt); |
328d5752 MF |
5065 | if (ret) |
5066 | mlog_errno(ret); | |
5067 | } | |
5068 | ||
328d5752 MF |
5069 | out: |
5070 | brelse(last_eb_bh); | |
5071 | return ret; | |
5072 | } | |
5073 | ||
5074 | /* | |
5075 | * Mark the already-existing extent at cpos as written for len clusters. | |
5076 | * | |
5077 | * If the existing extent is larger than the request, initiate a | |
5078 | * split. An attempt will be made at merging with adjacent extents. | |
5079 | * | |
5080 | * The caller is responsible for passing down meta_ac if we'll need it. | |
5081 | */ | |
f99b9b7c JB |
5082 | int ocfs2_mark_extent_written(struct inode *inode, |
5083 | struct ocfs2_extent_tree *et, | |
328d5752 MF |
5084 | handle_t *handle, u32 cpos, u32 len, u32 phys, |
5085 | struct ocfs2_alloc_context *meta_ac, | |
f99b9b7c | 5086 | struct ocfs2_cached_dealloc_ctxt *dealloc) |
328d5752 MF |
5087 | { |
5088 | int ret, index; | |
5089 | u64 start_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys); | |
5090 | struct ocfs2_extent_rec split_rec; | |
5091 | struct ocfs2_path *left_path = NULL; | |
5092 | struct ocfs2_extent_list *el; | |
5093 | ||
5094 | mlog(0, "Inode %lu cpos %u, len %u, phys %u (%llu)\n", | |
5095 | inode->i_ino, cpos, len, phys, (unsigned long long)start_blkno); | |
5096 | ||
5097 | if (!ocfs2_writes_unwritten_extents(OCFS2_SB(inode->i_sb))) { | |
5098 | ocfs2_error(inode->i_sb, "Inode %llu has unwritten extents " | |
5099 | "that are being written to, but the feature bit " | |
5100 | "is not set in the super block.", | |
5101 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
5102 | ret = -EROFS; | |
5103 | goto out; | |
5104 | } | |
5105 | ||
5106 | /* | |
5107 | * XXX: This should be fixed up so that we just re-insert the | |
5108 | * next extent records. | |
f99b9b7c JB |
5109 | * |
5110 | * XXX: This is a hack on the extent tree, maybe it should be | |
5111 | * an op? | |
328d5752 | 5112 | */ |
f99b9b7c | 5113 | if (et->et_ops == &ocfs2_dinode_et_ops) |
e7d4cb6b | 5114 | ocfs2_extent_map_trunc(inode, 0); |
328d5752 | 5115 | |
ffdd7a54 | 5116 | left_path = ocfs2_new_path_from_et(et); |
328d5752 MF |
5117 | if (!left_path) { |
5118 | ret = -ENOMEM; | |
5119 | mlog_errno(ret); | |
5120 | goto out; | |
5121 | } | |
5122 | ||
facdb77f | 5123 | ret = ocfs2_find_path(et->et_ci, left_path, cpos); |
328d5752 MF |
5124 | if (ret) { |
5125 | mlog_errno(ret); | |
5126 | goto out; | |
5127 | } | |
5128 | el = path_leaf_el(left_path); | |
5129 | ||
5130 | index = ocfs2_search_extent_list(el, cpos); | |
5131 | if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) { | |
5132 | ocfs2_error(inode->i_sb, | |
5133 | "Inode %llu has an extent at cpos %u which can no " | |
5134 | "longer be found.\n", | |
5135 | (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos); | |
5136 | ret = -EROFS; | |
5137 | goto out; | |
5138 | } | |
5139 | ||
5140 | memset(&split_rec, 0, sizeof(struct ocfs2_extent_rec)); | |
5141 | split_rec.e_cpos = cpu_to_le32(cpos); | |
5142 | split_rec.e_leaf_clusters = cpu_to_le16(len); | |
5143 | split_rec.e_blkno = cpu_to_le64(start_blkno); | |
5144 | split_rec.e_flags = path_leaf_el(left_path)->l_recs[index].e_flags; | |
5145 | split_rec.e_flags &= ~OCFS2_EXT_UNWRITTEN; | |
5146 | ||
f99b9b7c | 5147 | ret = __ocfs2_mark_extent_written(inode, et, handle, left_path, |
e7d4cb6b TM |
5148 | index, &split_rec, meta_ac, |
5149 | dealloc); | |
328d5752 MF |
5150 | if (ret) |
5151 | mlog_errno(ret); | |
5152 | ||
5153 | out: | |
5154 | ocfs2_free_path(left_path); | |
5155 | return ret; | |
5156 | } | |
5157 | ||
e7d4cb6b | 5158 | static int ocfs2_split_tree(struct inode *inode, struct ocfs2_extent_tree *et, |
d0c7d708 MF |
5159 | handle_t *handle, struct ocfs2_path *path, |
5160 | int index, u32 new_range, | |
5161 | struct ocfs2_alloc_context *meta_ac) | |
5162 | { | |
5163 | int ret, depth, credits = handle->h_buffer_credits; | |
d0c7d708 MF |
5164 | struct buffer_head *last_eb_bh = NULL; |
5165 | struct ocfs2_extent_block *eb; | |
5166 | struct ocfs2_extent_list *rightmost_el, *el; | |
5167 | struct ocfs2_extent_rec split_rec; | |
5168 | struct ocfs2_extent_rec *rec; | |
5169 | struct ocfs2_insert_type insert; | |
5170 | ||
5171 | /* | |
5172 | * Setup the record to split before we grow the tree. | |
5173 | */ | |
5174 | el = path_leaf_el(path); | |
5175 | rec = &el->l_recs[index]; | |
5176 | ocfs2_make_right_split_rec(inode->i_sb, &split_rec, new_range, rec); | |
5177 | ||
5178 | depth = path->p_tree_depth; | |
5179 | if (depth > 0) { | |
3d03a305 | 5180 | ret = ocfs2_read_extent_block(et->et_ci, |
5e96581a JB |
5181 | ocfs2_et_get_last_eb_blk(et), |
5182 | &last_eb_bh); | |
d0c7d708 MF |
5183 | if (ret < 0) { |
5184 | mlog_errno(ret); | |
5185 | goto out; | |
5186 | } | |
5187 | ||
5188 | eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; | |
5189 | rightmost_el = &eb->h_list; | |
5190 | } else | |
5191 | rightmost_el = path_leaf_el(path); | |
5192 | ||
811f933d | 5193 | credits += path->p_tree_depth + |
ce1d9ea6 | 5194 | ocfs2_extend_meta_needed(et->et_root_el); |
d0c7d708 MF |
5195 | ret = ocfs2_extend_trans(handle, credits); |
5196 | if (ret) { | |
5197 | mlog_errno(ret); | |
5198 | goto out; | |
5199 | } | |
5200 | ||
5201 | if (le16_to_cpu(rightmost_el->l_next_free_rec) == | |
5202 | le16_to_cpu(rightmost_el->l_count)) { | |
d401dc12 | 5203 | ret = ocfs2_grow_tree(handle, et, &depth, &last_eb_bh, |
d0c7d708 MF |
5204 | meta_ac); |
5205 | if (ret) { | |
5206 | mlog_errno(ret); | |
5207 | goto out; | |
5208 | } | |
d0c7d708 MF |
5209 | } |
5210 | ||
5211 | memset(&insert, 0, sizeof(struct ocfs2_insert_type)); | |
5212 | insert.ins_appending = APPEND_NONE; | |
5213 | insert.ins_contig = CONTIG_NONE; | |
5214 | insert.ins_split = SPLIT_RIGHT; | |
d0c7d708 MF |
5215 | insert.ins_tree_depth = depth; |
5216 | ||
e7d4cb6b | 5217 | ret = ocfs2_do_insert_extent(inode, handle, et, &split_rec, &insert); |
d0c7d708 MF |
5218 | if (ret) |
5219 | mlog_errno(ret); | |
5220 | ||
5221 | out: | |
5222 | brelse(last_eb_bh); | |
5223 | return ret; | |
5224 | } | |
5225 | ||
5226 | static int ocfs2_truncate_rec(struct inode *inode, handle_t *handle, | |
5227 | struct ocfs2_path *path, int index, | |
5228 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
e7d4cb6b TM |
5229 | u32 cpos, u32 len, |
5230 | struct ocfs2_extent_tree *et) | |
d0c7d708 MF |
5231 | { |
5232 | int ret; | |
5233 | u32 left_cpos, rec_range, trunc_range; | |
5234 | int wants_rotate = 0, is_rightmost_tree_rec = 0; | |
5235 | struct super_block *sb = inode->i_sb; | |
5236 | struct ocfs2_path *left_path = NULL; | |
5237 | struct ocfs2_extent_list *el = path_leaf_el(path); | |
5238 | struct ocfs2_extent_rec *rec; | |
5239 | struct ocfs2_extent_block *eb; | |
5240 | ||
5241 | if (ocfs2_is_empty_extent(&el->l_recs[0]) && index > 0) { | |
70f18c08 | 5242 | ret = ocfs2_rotate_tree_left(handle, et, path, dealloc); |
d0c7d708 MF |
5243 | if (ret) { |
5244 | mlog_errno(ret); | |
5245 | goto out; | |
5246 | } | |
5247 | ||
5248 | index--; | |
5249 | } | |
5250 | ||
5251 | if (index == (le16_to_cpu(el->l_next_free_rec) - 1) && | |
5252 | path->p_tree_depth) { | |
5253 | /* | |
5254 | * Check whether this is the rightmost tree record. If | |
5255 | * we remove all of this record or part of its right | |
5256 | * edge then an update of the record lengths above it | |
5257 | * will be required. | |
5258 | */ | |
5259 | eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data; | |
5260 | if (eb->h_next_leaf_blk == 0) | |
5261 | is_rightmost_tree_rec = 1; | |
5262 | } | |
5263 | ||
5264 | rec = &el->l_recs[index]; | |
5265 | if (index == 0 && path->p_tree_depth && | |
5266 | le32_to_cpu(rec->e_cpos) == cpos) { | |
5267 | /* | |
5268 | * Changing the leftmost offset (via partial or whole | |
5269 | * record truncate) of an interior (or rightmost) path | |
5270 | * means we have to update the subtree that is formed | |
5271 | * by this leaf and the one to it's left. | |
5272 | * | |
5273 | * There are two cases we can skip: | |
5274 | * 1) Path is the leftmost one in our inode tree. | |
5275 | * 2) The leaf is rightmost and will be empty after | |
5276 | * we remove the extent record - the rotate code | |
5277 | * knows how to update the newly formed edge. | |
5278 | */ | |
5279 | ||
5280 | ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, path, | |
5281 | &left_cpos); | |
5282 | if (ret) { | |
5283 | mlog_errno(ret); | |
5284 | goto out; | |
5285 | } | |
5286 | ||
5287 | if (left_cpos && le16_to_cpu(el->l_next_free_rec) > 1) { | |
ffdd7a54 | 5288 | left_path = ocfs2_new_path_from_path(path); |
d0c7d708 MF |
5289 | if (!left_path) { |
5290 | ret = -ENOMEM; | |
5291 | mlog_errno(ret); | |
5292 | goto out; | |
5293 | } | |
5294 | ||
facdb77f JB |
5295 | ret = ocfs2_find_path(et->et_ci, left_path, |
5296 | left_cpos); | |
d0c7d708 MF |
5297 | if (ret) { |
5298 | mlog_errno(ret); | |
5299 | goto out; | |
5300 | } | |
5301 | } | |
5302 | } | |
5303 | ||
5304 | ret = ocfs2_extend_rotate_transaction(handle, 0, | |
5305 | handle->h_buffer_credits, | |
5306 | path); | |
5307 | if (ret) { | |
5308 | mlog_errno(ret); | |
5309 | goto out; | |
5310 | } | |
5311 | ||
d9a0a1f8 | 5312 | ret = ocfs2_journal_access_path(et->et_ci, handle, path); |
d0c7d708 MF |
5313 | if (ret) { |
5314 | mlog_errno(ret); | |
5315 | goto out; | |
5316 | } | |
5317 | ||
d9a0a1f8 | 5318 | ret = ocfs2_journal_access_path(et->et_ci, handle, left_path); |
d0c7d708 MF |
5319 | if (ret) { |
5320 | mlog_errno(ret); | |
5321 | goto out; | |
5322 | } | |
5323 | ||
5324 | rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
5325 | trunc_range = cpos + len; | |
5326 | ||
5327 | if (le32_to_cpu(rec->e_cpos) == cpos && rec_range == trunc_range) { | |
5328 | int next_free; | |
5329 | ||
5330 | memset(rec, 0, sizeof(*rec)); | |
5331 | ocfs2_cleanup_merge(el, index); | |
5332 | wants_rotate = 1; | |
5333 | ||
5334 | next_free = le16_to_cpu(el->l_next_free_rec); | |
5335 | if (is_rightmost_tree_rec && next_free > 1) { | |
5336 | /* | |
5337 | * We skip the edge update if this path will | |
5338 | * be deleted by the rotate code. | |
5339 | */ | |
5340 | rec = &el->l_recs[next_free - 1]; | |
d401dc12 | 5341 | ocfs2_adjust_rightmost_records(handle, et, path, |
d0c7d708 MF |
5342 | rec); |
5343 | } | |
5344 | } else if (le32_to_cpu(rec->e_cpos) == cpos) { | |
5345 | /* Remove leftmost portion of the record. */ | |
5346 | le32_add_cpu(&rec->e_cpos, len); | |
5347 | le64_add_cpu(&rec->e_blkno, ocfs2_clusters_to_blocks(sb, len)); | |
5348 | le16_add_cpu(&rec->e_leaf_clusters, -len); | |
5349 | } else if (rec_range == trunc_range) { | |
5350 | /* Remove rightmost portion of the record */ | |
5351 | le16_add_cpu(&rec->e_leaf_clusters, -len); | |
5352 | if (is_rightmost_tree_rec) | |
d401dc12 | 5353 | ocfs2_adjust_rightmost_records(handle, et, path, rec); |
d0c7d708 MF |
5354 | } else { |
5355 | /* Caller should have trapped this. */ | |
5356 | mlog(ML_ERROR, "Inode %llu: Invalid record truncate: (%u, %u) " | |
5357 | "(%u, %u)\n", (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
5358 | le32_to_cpu(rec->e_cpos), | |
5359 | le16_to_cpu(rec->e_leaf_clusters), cpos, len); | |
5360 | BUG(); | |
5361 | } | |
5362 | ||
5363 | if (left_path) { | |
5364 | int subtree_index; | |
5365 | ||
7dc02805 | 5366 | subtree_index = ocfs2_find_subtree_root(et, left_path, path); |
4619c73e | 5367 | ocfs2_complete_edge_insert(handle, left_path, path, |
d0c7d708 MF |
5368 | subtree_index); |
5369 | } | |
5370 | ||
5371 | ocfs2_journal_dirty(handle, path_leaf_bh(path)); | |
5372 | ||
70f18c08 | 5373 | ret = ocfs2_rotate_tree_left(handle, et, path, dealloc); |
d0c7d708 MF |
5374 | if (ret) { |
5375 | mlog_errno(ret); | |
5376 | goto out; | |
5377 | } | |
5378 | ||
5379 | out: | |
5380 | ocfs2_free_path(left_path); | |
5381 | return ret; | |
5382 | } | |
5383 | ||
f99b9b7c JB |
5384 | int ocfs2_remove_extent(struct inode *inode, |
5385 | struct ocfs2_extent_tree *et, | |
063c4561 MF |
5386 | u32 cpos, u32 len, handle_t *handle, |
5387 | struct ocfs2_alloc_context *meta_ac, | |
f99b9b7c | 5388 | struct ocfs2_cached_dealloc_ctxt *dealloc) |
d0c7d708 MF |
5389 | { |
5390 | int ret, index; | |
5391 | u32 rec_range, trunc_range; | |
5392 | struct ocfs2_extent_rec *rec; | |
5393 | struct ocfs2_extent_list *el; | |
e7d4cb6b | 5394 | struct ocfs2_path *path = NULL; |
d0c7d708 MF |
5395 | |
5396 | ocfs2_extent_map_trunc(inode, 0); | |
5397 | ||
ffdd7a54 | 5398 | path = ocfs2_new_path_from_et(et); |
d0c7d708 MF |
5399 | if (!path) { |
5400 | ret = -ENOMEM; | |
5401 | mlog_errno(ret); | |
5402 | goto out; | |
5403 | } | |
5404 | ||
facdb77f | 5405 | ret = ocfs2_find_path(et->et_ci, path, cpos); |
d0c7d708 MF |
5406 | if (ret) { |
5407 | mlog_errno(ret); | |
5408 | goto out; | |
5409 | } | |
5410 | ||
5411 | el = path_leaf_el(path); | |
5412 | index = ocfs2_search_extent_list(el, cpos); | |
5413 | if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) { | |
5414 | ocfs2_error(inode->i_sb, | |
5415 | "Inode %llu has an extent at cpos %u which can no " | |
5416 | "longer be found.\n", | |
5417 | (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos); | |
5418 | ret = -EROFS; | |
5419 | goto out; | |
5420 | } | |
5421 | ||
5422 | /* | |
5423 | * We have 3 cases of extent removal: | |
5424 | * 1) Range covers the entire extent rec | |
5425 | * 2) Range begins or ends on one edge of the extent rec | |
5426 | * 3) Range is in the middle of the extent rec (no shared edges) | |
5427 | * | |
5428 | * For case 1 we remove the extent rec and left rotate to | |
5429 | * fill the hole. | |
5430 | * | |
5431 | * For case 2 we just shrink the existing extent rec, with a | |
5432 | * tree update if the shrinking edge is also the edge of an | |
5433 | * extent block. | |
5434 | * | |
5435 | * For case 3 we do a right split to turn the extent rec into | |
5436 | * something case 2 can handle. | |
5437 | */ | |
5438 | rec = &el->l_recs[index]; | |
5439 | rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
5440 | trunc_range = cpos + len; | |
5441 | ||
5442 | BUG_ON(cpos < le32_to_cpu(rec->e_cpos) || trunc_range > rec_range); | |
5443 | ||
5444 | mlog(0, "Inode %llu, remove (cpos %u, len %u). Existing index %d " | |
5445 | "(cpos %u, len %u)\n", | |
5446 | (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos, len, index, | |
5447 | le32_to_cpu(rec->e_cpos), ocfs2_rec_clusters(el, rec)); | |
5448 | ||
5449 | if (le32_to_cpu(rec->e_cpos) == cpos || rec_range == trunc_range) { | |
5450 | ret = ocfs2_truncate_rec(inode, handle, path, index, dealloc, | |
f99b9b7c | 5451 | cpos, len, et); |
d0c7d708 MF |
5452 | if (ret) { |
5453 | mlog_errno(ret); | |
5454 | goto out; | |
5455 | } | |
5456 | } else { | |
f99b9b7c | 5457 | ret = ocfs2_split_tree(inode, et, handle, path, index, |
d0c7d708 MF |
5458 | trunc_range, meta_ac); |
5459 | if (ret) { | |
5460 | mlog_errno(ret); | |
5461 | goto out; | |
5462 | } | |
5463 | ||
5464 | /* | |
5465 | * The split could have manipulated the tree enough to | |
5466 | * move the record location, so we have to look for it again. | |
5467 | */ | |
5468 | ocfs2_reinit_path(path, 1); | |
5469 | ||
facdb77f | 5470 | ret = ocfs2_find_path(et->et_ci, path, cpos); |
d0c7d708 MF |
5471 | if (ret) { |
5472 | mlog_errno(ret); | |
5473 | goto out; | |
5474 | } | |
5475 | ||
5476 | el = path_leaf_el(path); | |
5477 | index = ocfs2_search_extent_list(el, cpos); | |
5478 | if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) { | |
5479 | ocfs2_error(inode->i_sb, | |
5480 | "Inode %llu: split at cpos %u lost record.", | |
5481 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
5482 | cpos); | |
5483 | ret = -EROFS; | |
5484 | goto out; | |
5485 | } | |
5486 | ||
5487 | /* | |
5488 | * Double check our values here. If anything is fishy, | |
5489 | * it's easier to catch it at the top level. | |
5490 | */ | |
5491 | rec = &el->l_recs[index]; | |
5492 | rec_range = le32_to_cpu(rec->e_cpos) + | |
5493 | ocfs2_rec_clusters(el, rec); | |
5494 | if (rec_range != trunc_range) { | |
5495 | ocfs2_error(inode->i_sb, | |
5496 | "Inode %llu: error after split at cpos %u" | |
5497 | "trunc len %u, existing record is (%u,%u)", | |
5498 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
5499 | cpos, len, le32_to_cpu(rec->e_cpos), | |
5500 | ocfs2_rec_clusters(el, rec)); | |
5501 | ret = -EROFS; | |
5502 | goto out; | |
5503 | } | |
5504 | ||
5505 | ret = ocfs2_truncate_rec(inode, handle, path, index, dealloc, | |
f99b9b7c | 5506 | cpos, len, et); |
d0c7d708 MF |
5507 | if (ret) { |
5508 | mlog_errno(ret); | |
5509 | goto out; | |
5510 | } | |
5511 | } | |
5512 | ||
5513 | out: | |
5514 | ocfs2_free_path(path); | |
5515 | return ret; | |
5516 | } | |
5517 | ||
fecc0112 MF |
5518 | int ocfs2_remove_btree_range(struct inode *inode, |
5519 | struct ocfs2_extent_tree *et, | |
5520 | u32 cpos, u32 phys_cpos, u32 len, | |
5521 | struct ocfs2_cached_dealloc_ctxt *dealloc) | |
5522 | { | |
5523 | int ret; | |
5524 | u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos); | |
5525 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
5526 | struct inode *tl_inode = osb->osb_tl_inode; | |
5527 | handle_t *handle; | |
5528 | struct ocfs2_alloc_context *meta_ac = NULL; | |
5529 | ||
5530 | ret = ocfs2_lock_allocators(inode, et, 0, 1, NULL, &meta_ac); | |
5531 | if (ret) { | |
5532 | mlog_errno(ret); | |
5533 | return ret; | |
5534 | } | |
5535 | ||
5536 | mutex_lock(&tl_inode->i_mutex); | |
5537 | ||
5538 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
5539 | ret = __ocfs2_flush_truncate_log(osb); | |
5540 | if (ret < 0) { | |
5541 | mlog_errno(ret); | |
5542 | goto out; | |
5543 | } | |
5544 | } | |
5545 | ||
a90714c1 | 5546 | handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb)); |
fecc0112 MF |
5547 | if (IS_ERR(handle)) { |
5548 | ret = PTR_ERR(handle); | |
5549 | mlog_errno(ret); | |
5550 | goto out; | |
5551 | } | |
5552 | ||
d9a0a1f8 | 5553 | ret = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 5554 | OCFS2_JOURNAL_ACCESS_WRITE); |
fecc0112 MF |
5555 | if (ret) { |
5556 | mlog_errno(ret); | |
5557 | goto out; | |
5558 | } | |
5559 | ||
fd4ef231 MF |
5560 | vfs_dq_free_space_nodirty(inode, |
5561 | ocfs2_clusters_to_bytes(inode->i_sb, len)); | |
5562 | ||
fecc0112 MF |
5563 | ret = ocfs2_remove_extent(inode, et, cpos, len, handle, meta_ac, |
5564 | dealloc); | |
5565 | if (ret) { | |
5566 | mlog_errno(ret); | |
5567 | goto out_commit; | |
5568 | } | |
5569 | ||
6136ca5f | 5570 | ocfs2_et_update_clusters(et, -len); |
fecc0112 MF |
5571 | |
5572 | ret = ocfs2_journal_dirty(handle, et->et_root_bh); | |
5573 | if (ret) { | |
5574 | mlog_errno(ret); | |
5575 | goto out_commit; | |
5576 | } | |
5577 | ||
5578 | ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len); | |
5579 | if (ret) | |
5580 | mlog_errno(ret); | |
5581 | ||
5582 | out_commit: | |
5583 | ocfs2_commit_trans(osb, handle); | |
5584 | out: | |
5585 | mutex_unlock(&tl_inode->i_mutex); | |
5586 | ||
5587 | if (meta_ac) | |
5588 | ocfs2_free_alloc_context(meta_ac); | |
5589 | ||
5590 | return ret; | |
5591 | } | |
5592 | ||
063c4561 | 5593 | int ocfs2_truncate_log_needs_flush(struct ocfs2_super *osb) |
ccd979bd MF |
5594 | { |
5595 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5596 | struct ocfs2_dinode *di; | |
5597 | struct ocfs2_truncate_log *tl; | |
5598 | ||
5599 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
5600 | tl = &di->id2.i_dealloc; | |
5601 | ||
5602 | mlog_bug_on_msg(le16_to_cpu(tl->tl_used) > le16_to_cpu(tl->tl_count), | |
5603 | "slot %d, invalid truncate log parameters: used = " | |
5604 | "%u, count = %u\n", osb->slot_num, | |
5605 | le16_to_cpu(tl->tl_used), le16_to_cpu(tl->tl_count)); | |
5606 | return le16_to_cpu(tl->tl_used) == le16_to_cpu(tl->tl_count); | |
5607 | } | |
5608 | ||
5609 | static int ocfs2_truncate_log_can_coalesce(struct ocfs2_truncate_log *tl, | |
5610 | unsigned int new_start) | |
5611 | { | |
5612 | unsigned int tail_index; | |
5613 | unsigned int current_tail; | |
5614 | ||
5615 | /* No records, nothing to coalesce */ | |
5616 | if (!le16_to_cpu(tl->tl_used)) | |
5617 | return 0; | |
5618 | ||
5619 | tail_index = le16_to_cpu(tl->tl_used) - 1; | |
5620 | current_tail = le32_to_cpu(tl->tl_recs[tail_index].t_start); | |
5621 | current_tail += le32_to_cpu(tl->tl_recs[tail_index].t_clusters); | |
5622 | ||
5623 | return current_tail == new_start; | |
5624 | } | |
5625 | ||
063c4561 MF |
5626 | int ocfs2_truncate_log_append(struct ocfs2_super *osb, |
5627 | handle_t *handle, | |
5628 | u64 start_blk, | |
5629 | unsigned int num_clusters) | |
ccd979bd MF |
5630 | { |
5631 | int status, index; | |
5632 | unsigned int start_cluster, tl_count; | |
5633 | struct inode *tl_inode = osb->osb_tl_inode; | |
5634 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5635 | struct ocfs2_dinode *di; | |
5636 | struct ocfs2_truncate_log *tl; | |
5637 | ||
b0697053 MF |
5638 | mlog_entry("start_blk = %llu, num_clusters = %u\n", |
5639 | (unsigned long long)start_blk, num_clusters); | |
ccd979bd | 5640 | |
1b1dcc1b | 5641 | BUG_ON(mutex_trylock(&tl_inode->i_mutex)); |
ccd979bd MF |
5642 | |
5643 | start_cluster = ocfs2_blocks_to_clusters(osb->sb, start_blk); | |
5644 | ||
5645 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
ccd979bd | 5646 | |
10995aa2 JB |
5647 | /* tl_bh is loaded from ocfs2_truncate_log_init(). It's validated |
5648 | * by the underlying call to ocfs2_read_inode_block(), so any | |
5649 | * corruption is a code bug */ | |
5650 | BUG_ON(!OCFS2_IS_VALID_DINODE(di)); | |
5651 | ||
5652 | tl = &di->id2.i_dealloc; | |
ccd979bd MF |
5653 | tl_count = le16_to_cpu(tl->tl_count); |
5654 | mlog_bug_on_msg(tl_count > ocfs2_truncate_recs_per_inode(osb->sb) || | |
5655 | tl_count == 0, | |
b0697053 MF |
5656 | "Truncate record count on #%llu invalid " |
5657 | "wanted %u, actual %u\n", | |
5658 | (unsigned long long)OCFS2_I(tl_inode)->ip_blkno, | |
ccd979bd MF |
5659 | ocfs2_truncate_recs_per_inode(osb->sb), |
5660 | le16_to_cpu(tl->tl_count)); | |
5661 | ||
5662 | /* Caller should have known to flush before calling us. */ | |
5663 | index = le16_to_cpu(tl->tl_used); | |
5664 | if (index >= tl_count) { | |
5665 | status = -ENOSPC; | |
5666 | mlog_errno(status); | |
5667 | goto bail; | |
5668 | } | |
5669 | ||
0cf2f763 | 5670 | status = ocfs2_journal_access_di(handle, INODE_CACHE(tl_inode), tl_bh, |
13723d00 | 5671 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
5672 | if (status < 0) { |
5673 | mlog_errno(status); | |
5674 | goto bail; | |
5675 | } | |
5676 | ||
5677 | mlog(0, "Log truncate of %u clusters starting at cluster %u to " | |
b0697053 MF |
5678 | "%llu (index = %d)\n", num_clusters, start_cluster, |
5679 | (unsigned long long)OCFS2_I(tl_inode)->ip_blkno, index); | |
ccd979bd MF |
5680 | |
5681 | if (ocfs2_truncate_log_can_coalesce(tl, start_cluster)) { | |
5682 | /* | |
5683 | * Move index back to the record we are coalescing with. | |
5684 | * ocfs2_truncate_log_can_coalesce() guarantees nonzero | |
5685 | */ | |
5686 | index--; | |
5687 | ||
5688 | num_clusters += le32_to_cpu(tl->tl_recs[index].t_clusters); | |
5689 | mlog(0, "Coalesce with index %u (start = %u, clusters = %u)\n", | |
5690 | index, le32_to_cpu(tl->tl_recs[index].t_start), | |
5691 | num_clusters); | |
5692 | } else { | |
5693 | tl->tl_recs[index].t_start = cpu_to_le32(start_cluster); | |
5694 | tl->tl_used = cpu_to_le16(index + 1); | |
5695 | } | |
5696 | tl->tl_recs[index].t_clusters = cpu_to_le32(num_clusters); | |
5697 | ||
5698 | status = ocfs2_journal_dirty(handle, tl_bh); | |
5699 | if (status < 0) { | |
5700 | mlog_errno(status); | |
5701 | goto bail; | |
5702 | } | |
5703 | ||
5704 | bail: | |
5705 | mlog_exit(status); | |
5706 | return status; | |
5707 | } | |
5708 | ||
5709 | static int ocfs2_replay_truncate_records(struct ocfs2_super *osb, | |
1fabe148 | 5710 | handle_t *handle, |
ccd979bd MF |
5711 | struct inode *data_alloc_inode, |
5712 | struct buffer_head *data_alloc_bh) | |
5713 | { | |
5714 | int status = 0; | |
5715 | int i; | |
5716 | unsigned int num_clusters; | |
5717 | u64 start_blk; | |
5718 | struct ocfs2_truncate_rec rec; | |
5719 | struct ocfs2_dinode *di; | |
5720 | struct ocfs2_truncate_log *tl; | |
5721 | struct inode *tl_inode = osb->osb_tl_inode; | |
5722 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5723 | ||
5724 | mlog_entry_void(); | |
5725 | ||
5726 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
5727 | tl = &di->id2.i_dealloc; | |
5728 | i = le16_to_cpu(tl->tl_used) - 1; | |
5729 | while (i >= 0) { | |
5730 | /* Caller has given us at least enough credits to | |
5731 | * update the truncate log dinode */ | |
0cf2f763 | 5732 | status = ocfs2_journal_access_di(handle, INODE_CACHE(tl_inode), tl_bh, |
13723d00 | 5733 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
5734 | if (status < 0) { |
5735 | mlog_errno(status); | |
5736 | goto bail; | |
5737 | } | |
5738 | ||
5739 | tl->tl_used = cpu_to_le16(i); | |
5740 | ||
5741 | status = ocfs2_journal_dirty(handle, tl_bh); | |
5742 | if (status < 0) { | |
5743 | mlog_errno(status); | |
5744 | goto bail; | |
5745 | } | |
5746 | ||
5747 | /* TODO: Perhaps we can calculate the bulk of the | |
5748 | * credits up front rather than extending like | |
5749 | * this. */ | |
5750 | status = ocfs2_extend_trans(handle, | |
5751 | OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC); | |
5752 | if (status < 0) { | |
5753 | mlog_errno(status); | |
5754 | goto bail; | |
5755 | } | |
5756 | ||
5757 | rec = tl->tl_recs[i]; | |
5758 | start_blk = ocfs2_clusters_to_blocks(data_alloc_inode->i_sb, | |
5759 | le32_to_cpu(rec.t_start)); | |
5760 | num_clusters = le32_to_cpu(rec.t_clusters); | |
5761 | ||
5762 | /* if start_blk is not set, we ignore the record as | |
5763 | * invalid. */ | |
5764 | if (start_blk) { | |
5765 | mlog(0, "free record %d, start = %u, clusters = %u\n", | |
5766 | i, le32_to_cpu(rec.t_start), num_clusters); | |
5767 | ||
5768 | status = ocfs2_free_clusters(handle, data_alloc_inode, | |
5769 | data_alloc_bh, start_blk, | |
5770 | num_clusters); | |
5771 | if (status < 0) { | |
5772 | mlog_errno(status); | |
5773 | goto bail; | |
5774 | } | |
5775 | } | |
5776 | i--; | |
5777 | } | |
5778 | ||
5779 | bail: | |
5780 | mlog_exit(status); | |
5781 | return status; | |
5782 | } | |
5783 | ||
1b1dcc1b | 5784 | /* Expects you to already be holding tl_inode->i_mutex */ |
063c4561 | 5785 | int __ocfs2_flush_truncate_log(struct ocfs2_super *osb) |
ccd979bd MF |
5786 | { |
5787 | int status; | |
5788 | unsigned int num_to_flush; | |
1fabe148 | 5789 | handle_t *handle; |
ccd979bd MF |
5790 | struct inode *tl_inode = osb->osb_tl_inode; |
5791 | struct inode *data_alloc_inode = NULL; | |
5792 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5793 | struct buffer_head *data_alloc_bh = NULL; | |
5794 | struct ocfs2_dinode *di; | |
5795 | struct ocfs2_truncate_log *tl; | |
5796 | ||
5797 | mlog_entry_void(); | |
5798 | ||
1b1dcc1b | 5799 | BUG_ON(mutex_trylock(&tl_inode->i_mutex)); |
ccd979bd MF |
5800 | |
5801 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
ccd979bd | 5802 | |
10995aa2 JB |
5803 | /* tl_bh is loaded from ocfs2_truncate_log_init(). It's validated |
5804 | * by the underlying call to ocfs2_read_inode_block(), so any | |
5805 | * corruption is a code bug */ | |
5806 | BUG_ON(!OCFS2_IS_VALID_DINODE(di)); | |
5807 | ||
5808 | tl = &di->id2.i_dealloc; | |
ccd979bd | 5809 | num_to_flush = le16_to_cpu(tl->tl_used); |
b0697053 MF |
5810 | mlog(0, "Flush %u records from truncate log #%llu\n", |
5811 | num_to_flush, (unsigned long long)OCFS2_I(tl_inode)->ip_blkno); | |
ccd979bd MF |
5812 | if (!num_to_flush) { |
5813 | status = 0; | |
e08dc8b9 | 5814 | goto out; |
ccd979bd MF |
5815 | } |
5816 | ||
5817 | data_alloc_inode = ocfs2_get_system_file_inode(osb, | |
5818 | GLOBAL_BITMAP_SYSTEM_INODE, | |
5819 | OCFS2_INVALID_SLOT); | |
5820 | if (!data_alloc_inode) { | |
5821 | status = -EINVAL; | |
5822 | mlog(ML_ERROR, "Could not get bitmap inode!\n"); | |
e08dc8b9 | 5823 | goto out; |
ccd979bd MF |
5824 | } |
5825 | ||
e08dc8b9 MF |
5826 | mutex_lock(&data_alloc_inode->i_mutex); |
5827 | ||
e63aecb6 | 5828 | status = ocfs2_inode_lock(data_alloc_inode, &data_alloc_bh, 1); |
ccd979bd MF |
5829 | if (status < 0) { |
5830 | mlog_errno(status); | |
e08dc8b9 | 5831 | goto out_mutex; |
ccd979bd MF |
5832 | } |
5833 | ||
65eff9cc | 5834 | handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE); |
ccd979bd MF |
5835 | if (IS_ERR(handle)) { |
5836 | status = PTR_ERR(handle); | |
ccd979bd | 5837 | mlog_errno(status); |
e08dc8b9 | 5838 | goto out_unlock; |
ccd979bd MF |
5839 | } |
5840 | ||
5841 | status = ocfs2_replay_truncate_records(osb, handle, data_alloc_inode, | |
5842 | data_alloc_bh); | |
e08dc8b9 | 5843 | if (status < 0) |
ccd979bd | 5844 | mlog_errno(status); |
ccd979bd | 5845 | |
02dc1af4 | 5846 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 5847 | |
e08dc8b9 MF |
5848 | out_unlock: |
5849 | brelse(data_alloc_bh); | |
e63aecb6 | 5850 | ocfs2_inode_unlock(data_alloc_inode, 1); |
ccd979bd | 5851 | |
e08dc8b9 MF |
5852 | out_mutex: |
5853 | mutex_unlock(&data_alloc_inode->i_mutex); | |
5854 | iput(data_alloc_inode); | |
ccd979bd | 5855 | |
e08dc8b9 | 5856 | out: |
ccd979bd MF |
5857 | mlog_exit(status); |
5858 | return status; | |
5859 | } | |
5860 | ||
5861 | int ocfs2_flush_truncate_log(struct ocfs2_super *osb) | |
5862 | { | |
5863 | int status; | |
5864 | struct inode *tl_inode = osb->osb_tl_inode; | |
5865 | ||
1b1dcc1b | 5866 | mutex_lock(&tl_inode->i_mutex); |
ccd979bd | 5867 | status = __ocfs2_flush_truncate_log(osb); |
1b1dcc1b | 5868 | mutex_unlock(&tl_inode->i_mutex); |
ccd979bd MF |
5869 | |
5870 | return status; | |
5871 | } | |
5872 | ||
c4028958 | 5873 | static void ocfs2_truncate_log_worker(struct work_struct *work) |
ccd979bd MF |
5874 | { |
5875 | int status; | |
c4028958 DH |
5876 | struct ocfs2_super *osb = |
5877 | container_of(work, struct ocfs2_super, | |
5878 | osb_truncate_log_wq.work); | |
ccd979bd MF |
5879 | |
5880 | mlog_entry_void(); | |
5881 | ||
5882 | status = ocfs2_flush_truncate_log(osb); | |
5883 | if (status < 0) | |
5884 | mlog_errno(status); | |
4d0ddb2c TM |
5885 | else |
5886 | ocfs2_init_inode_steal_slot(osb); | |
ccd979bd MF |
5887 | |
5888 | mlog_exit(status); | |
5889 | } | |
5890 | ||
5891 | #define OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL (2 * HZ) | |
5892 | void ocfs2_schedule_truncate_log_flush(struct ocfs2_super *osb, | |
5893 | int cancel) | |
5894 | { | |
5895 | if (osb->osb_tl_inode) { | |
5896 | /* We want to push off log flushes while truncates are | |
5897 | * still running. */ | |
5898 | if (cancel) | |
5899 | cancel_delayed_work(&osb->osb_truncate_log_wq); | |
5900 | ||
5901 | queue_delayed_work(ocfs2_wq, &osb->osb_truncate_log_wq, | |
5902 | OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL); | |
5903 | } | |
5904 | } | |
5905 | ||
5906 | static int ocfs2_get_truncate_log_info(struct ocfs2_super *osb, | |
5907 | int slot_num, | |
5908 | struct inode **tl_inode, | |
5909 | struct buffer_head **tl_bh) | |
5910 | { | |
5911 | int status; | |
5912 | struct inode *inode = NULL; | |
5913 | struct buffer_head *bh = NULL; | |
5914 | ||
5915 | inode = ocfs2_get_system_file_inode(osb, | |
5916 | TRUNCATE_LOG_SYSTEM_INODE, | |
5917 | slot_num); | |
5918 | if (!inode) { | |
5919 | status = -EINVAL; | |
5920 | mlog(ML_ERROR, "Could not get load truncate log inode!\n"); | |
5921 | goto bail; | |
5922 | } | |
5923 | ||
b657c95c | 5924 | status = ocfs2_read_inode_block(inode, &bh); |
ccd979bd MF |
5925 | if (status < 0) { |
5926 | iput(inode); | |
5927 | mlog_errno(status); | |
5928 | goto bail; | |
5929 | } | |
5930 | ||
5931 | *tl_inode = inode; | |
5932 | *tl_bh = bh; | |
5933 | bail: | |
5934 | mlog_exit(status); | |
5935 | return status; | |
5936 | } | |
5937 | ||
5938 | /* called during the 1st stage of node recovery. we stamp a clean | |
5939 | * truncate log and pass back a copy for processing later. if the | |
5940 | * truncate log does not require processing, a *tl_copy is set to | |
5941 | * NULL. */ | |
5942 | int ocfs2_begin_truncate_log_recovery(struct ocfs2_super *osb, | |
5943 | int slot_num, | |
5944 | struct ocfs2_dinode **tl_copy) | |
5945 | { | |
5946 | int status; | |
5947 | struct inode *tl_inode = NULL; | |
5948 | struct buffer_head *tl_bh = NULL; | |
5949 | struct ocfs2_dinode *di; | |
5950 | struct ocfs2_truncate_log *tl; | |
5951 | ||
5952 | *tl_copy = NULL; | |
5953 | ||
5954 | mlog(0, "recover truncate log from slot %d\n", slot_num); | |
5955 | ||
5956 | status = ocfs2_get_truncate_log_info(osb, slot_num, &tl_inode, &tl_bh); | |
5957 | if (status < 0) { | |
5958 | mlog_errno(status); | |
5959 | goto bail; | |
5960 | } | |
5961 | ||
5962 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
ccd979bd | 5963 | |
10995aa2 JB |
5964 | /* tl_bh is loaded from ocfs2_get_truncate_log_info(). It's |
5965 | * validated by the underlying call to ocfs2_read_inode_block(), | |
5966 | * so any corruption is a code bug */ | |
5967 | BUG_ON(!OCFS2_IS_VALID_DINODE(di)); | |
5968 | ||
5969 | tl = &di->id2.i_dealloc; | |
ccd979bd MF |
5970 | if (le16_to_cpu(tl->tl_used)) { |
5971 | mlog(0, "We'll have %u logs to recover\n", | |
5972 | le16_to_cpu(tl->tl_used)); | |
5973 | ||
5974 | *tl_copy = kmalloc(tl_bh->b_size, GFP_KERNEL); | |
5975 | if (!(*tl_copy)) { | |
5976 | status = -ENOMEM; | |
5977 | mlog_errno(status); | |
5978 | goto bail; | |
5979 | } | |
5980 | ||
5981 | /* Assuming the write-out below goes well, this copy | |
5982 | * will be passed back to recovery for processing. */ | |
5983 | memcpy(*tl_copy, tl_bh->b_data, tl_bh->b_size); | |
5984 | ||
5985 | /* All we need to do to clear the truncate log is set | |
5986 | * tl_used. */ | |
5987 | tl->tl_used = 0; | |
5988 | ||
13723d00 | 5989 | ocfs2_compute_meta_ecc(osb->sb, tl_bh->b_data, &di->i_check); |
8cb471e8 | 5990 | status = ocfs2_write_block(osb, tl_bh, INODE_CACHE(tl_inode)); |
ccd979bd MF |
5991 | if (status < 0) { |
5992 | mlog_errno(status); | |
5993 | goto bail; | |
5994 | } | |
5995 | } | |
5996 | ||
5997 | bail: | |
5998 | if (tl_inode) | |
5999 | iput(tl_inode); | |
a81cb88b | 6000 | brelse(tl_bh); |
ccd979bd MF |
6001 | |
6002 | if (status < 0 && (*tl_copy)) { | |
6003 | kfree(*tl_copy); | |
6004 | *tl_copy = NULL; | |
6005 | } | |
6006 | ||
6007 | mlog_exit(status); | |
6008 | return status; | |
6009 | } | |
6010 | ||
6011 | int ocfs2_complete_truncate_log_recovery(struct ocfs2_super *osb, | |
6012 | struct ocfs2_dinode *tl_copy) | |
6013 | { | |
6014 | int status = 0; | |
6015 | int i; | |
6016 | unsigned int clusters, num_recs, start_cluster; | |
6017 | u64 start_blk; | |
1fabe148 | 6018 | handle_t *handle; |
ccd979bd MF |
6019 | struct inode *tl_inode = osb->osb_tl_inode; |
6020 | struct ocfs2_truncate_log *tl; | |
6021 | ||
6022 | mlog_entry_void(); | |
6023 | ||
6024 | if (OCFS2_I(tl_inode)->ip_blkno == le64_to_cpu(tl_copy->i_blkno)) { | |
6025 | mlog(ML_ERROR, "Asked to recover my own truncate log!\n"); | |
6026 | return -EINVAL; | |
6027 | } | |
6028 | ||
6029 | tl = &tl_copy->id2.i_dealloc; | |
6030 | num_recs = le16_to_cpu(tl->tl_used); | |
b0697053 | 6031 | mlog(0, "cleanup %u records from %llu\n", num_recs, |
1ca1a111 | 6032 | (unsigned long long)le64_to_cpu(tl_copy->i_blkno)); |
ccd979bd | 6033 | |
1b1dcc1b | 6034 | mutex_lock(&tl_inode->i_mutex); |
ccd979bd MF |
6035 | for(i = 0; i < num_recs; i++) { |
6036 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
6037 | status = __ocfs2_flush_truncate_log(osb); | |
6038 | if (status < 0) { | |
6039 | mlog_errno(status); | |
6040 | goto bail_up; | |
6041 | } | |
6042 | } | |
6043 | ||
65eff9cc | 6044 | handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE); |
ccd979bd MF |
6045 | if (IS_ERR(handle)) { |
6046 | status = PTR_ERR(handle); | |
6047 | mlog_errno(status); | |
6048 | goto bail_up; | |
6049 | } | |
6050 | ||
6051 | clusters = le32_to_cpu(tl->tl_recs[i].t_clusters); | |
6052 | start_cluster = le32_to_cpu(tl->tl_recs[i].t_start); | |
6053 | start_blk = ocfs2_clusters_to_blocks(osb->sb, start_cluster); | |
6054 | ||
6055 | status = ocfs2_truncate_log_append(osb, handle, | |
6056 | start_blk, clusters); | |
02dc1af4 | 6057 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
6058 | if (status < 0) { |
6059 | mlog_errno(status); | |
6060 | goto bail_up; | |
6061 | } | |
6062 | } | |
6063 | ||
6064 | bail_up: | |
1b1dcc1b | 6065 | mutex_unlock(&tl_inode->i_mutex); |
ccd979bd MF |
6066 | |
6067 | mlog_exit(status); | |
6068 | return status; | |
6069 | } | |
6070 | ||
6071 | void ocfs2_truncate_log_shutdown(struct ocfs2_super *osb) | |
6072 | { | |
6073 | int status; | |
6074 | struct inode *tl_inode = osb->osb_tl_inode; | |
6075 | ||
6076 | mlog_entry_void(); | |
6077 | ||
6078 | if (tl_inode) { | |
6079 | cancel_delayed_work(&osb->osb_truncate_log_wq); | |
6080 | flush_workqueue(ocfs2_wq); | |
6081 | ||
6082 | status = ocfs2_flush_truncate_log(osb); | |
6083 | if (status < 0) | |
6084 | mlog_errno(status); | |
6085 | ||
6086 | brelse(osb->osb_tl_bh); | |
6087 | iput(osb->osb_tl_inode); | |
6088 | } | |
6089 | ||
6090 | mlog_exit_void(); | |
6091 | } | |
6092 | ||
6093 | int ocfs2_truncate_log_init(struct ocfs2_super *osb) | |
6094 | { | |
6095 | int status; | |
6096 | struct inode *tl_inode = NULL; | |
6097 | struct buffer_head *tl_bh = NULL; | |
6098 | ||
6099 | mlog_entry_void(); | |
6100 | ||
6101 | status = ocfs2_get_truncate_log_info(osb, | |
6102 | osb->slot_num, | |
6103 | &tl_inode, | |
6104 | &tl_bh); | |
6105 | if (status < 0) | |
6106 | mlog_errno(status); | |
6107 | ||
6108 | /* ocfs2_truncate_log_shutdown keys on the existence of | |
6109 | * osb->osb_tl_inode so we don't set any of the osb variables | |
6110 | * until we're sure all is well. */ | |
c4028958 DH |
6111 | INIT_DELAYED_WORK(&osb->osb_truncate_log_wq, |
6112 | ocfs2_truncate_log_worker); | |
ccd979bd MF |
6113 | osb->osb_tl_bh = tl_bh; |
6114 | osb->osb_tl_inode = tl_inode; | |
6115 | ||
6116 | mlog_exit(status); | |
6117 | return status; | |
6118 | } | |
6119 | ||
2b604351 MF |
6120 | /* |
6121 | * Delayed de-allocation of suballocator blocks. | |
6122 | * | |
6123 | * Some sets of block de-allocations might involve multiple suballocator inodes. | |
6124 | * | |
6125 | * The locking for this can get extremely complicated, especially when | |
6126 | * the suballocator inodes to delete from aren't known until deep | |
6127 | * within an unrelated codepath. | |
6128 | * | |
6129 | * ocfs2_extent_block structures are a good example of this - an inode | |
6130 | * btree could have been grown by any number of nodes each allocating | |
6131 | * out of their own suballoc inode. | |
6132 | * | |
6133 | * These structures allow the delay of block de-allocation until a | |
6134 | * later time, when locking of multiple cluster inodes won't cause | |
6135 | * deadlock. | |
6136 | */ | |
6137 | ||
6138 | /* | |
2891d290 TM |
6139 | * Describe a single bit freed from a suballocator. For the block |
6140 | * suballocators, it represents one block. For the global cluster | |
6141 | * allocator, it represents some clusters and free_bit indicates | |
6142 | * clusters number. | |
2b604351 MF |
6143 | */ |
6144 | struct ocfs2_cached_block_free { | |
6145 | struct ocfs2_cached_block_free *free_next; | |
6146 | u64 free_blk; | |
6147 | unsigned int free_bit; | |
6148 | }; | |
6149 | ||
6150 | struct ocfs2_per_slot_free_list { | |
6151 | struct ocfs2_per_slot_free_list *f_next_suballocator; | |
6152 | int f_inode_type; | |
6153 | int f_slot; | |
6154 | struct ocfs2_cached_block_free *f_first; | |
6155 | }; | |
6156 | ||
2891d290 TM |
6157 | static int ocfs2_free_cached_blocks(struct ocfs2_super *osb, |
6158 | int sysfile_type, | |
6159 | int slot, | |
6160 | struct ocfs2_cached_block_free *head) | |
2b604351 MF |
6161 | { |
6162 | int ret; | |
6163 | u64 bg_blkno; | |
6164 | handle_t *handle; | |
6165 | struct inode *inode; | |
6166 | struct buffer_head *di_bh = NULL; | |
6167 | struct ocfs2_cached_block_free *tmp; | |
6168 | ||
6169 | inode = ocfs2_get_system_file_inode(osb, sysfile_type, slot); | |
6170 | if (!inode) { | |
6171 | ret = -EINVAL; | |
6172 | mlog_errno(ret); | |
6173 | goto out; | |
6174 | } | |
6175 | ||
6176 | mutex_lock(&inode->i_mutex); | |
6177 | ||
e63aecb6 | 6178 | ret = ocfs2_inode_lock(inode, &di_bh, 1); |
2b604351 MF |
6179 | if (ret) { |
6180 | mlog_errno(ret); | |
6181 | goto out_mutex; | |
6182 | } | |
6183 | ||
6184 | handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE); | |
6185 | if (IS_ERR(handle)) { | |
6186 | ret = PTR_ERR(handle); | |
6187 | mlog_errno(ret); | |
6188 | goto out_unlock; | |
6189 | } | |
6190 | ||
6191 | while (head) { | |
6192 | bg_blkno = ocfs2_which_suballoc_group(head->free_blk, | |
6193 | head->free_bit); | |
6194 | mlog(0, "Free bit: (bit %u, blkno %llu)\n", | |
6195 | head->free_bit, (unsigned long long)head->free_blk); | |
6196 | ||
6197 | ret = ocfs2_free_suballoc_bits(handle, inode, di_bh, | |
6198 | head->free_bit, bg_blkno, 1); | |
6199 | if (ret) { | |
6200 | mlog_errno(ret); | |
6201 | goto out_journal; | |
6202 | } | |
6203 | ||
6204 | ret = ocfs2_extend_trans(handle, OCFS2_SUBALLOC_FREE); | |
6205 | if (ret) { | |
6206 | mlog_errno(ret); | |
6207 | goto out_journal; | |
6208 | } | |
6209 | ||
6210 | tmp = head; | |
6211 | head = head->free_next; | |
6212 | kfree(tmp); | |
6213 | } | |
6214 | ||
6215 | out_journal: | |
6216 | ocfs2_commit_trans(osb, handle); | |
6217 | ||
6218 | out_unlock: | |
e63aecb6 | 6219 | ocfs2_inode_unlock(inode, 1); |
2b604351 MF |
6220 | brelse(di_bh); |
6221 | out_mutex: | |
6222 | mutex_unlock(&inode->i_mutex); | |
6223 | iput(inode); | |
6224 | out: | |
6225 | while(head) { | |
6226 | /* Premature exit may have left some dangling items. */ | |
6227 | tmp = head; | |
6228 | head = head->free_next; | |
6229 | kfree(tmp); | |
6230 | } | |
6231 | ||
6232 | return ret; | |
6233 | } | |
6234 | ||
2891d290 TM |
6235 | int ocfs2_cache_cluster_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt, |
6236 | u64 blkno, unsigned int bit) | |
6237 | { | |
6238 | int ret = 0; | |
6239 | struct ocfs2_cached_block_free *item; | |
6240 | ||
6241 | item = kmalloc(sizeof(*item), GFP_NOFS); | |
6242 | if (item == NULL) { | |
6243 | ret = -ENOMEM; | |
6244 | mlog_errno(ret); | |
6245 | return ret; | |
6246 | } | |
6247 | ||
6248 | mlog(0, "Insert clusters: (bit %u, blk %llu)\n", | |
6249 | bit, (unsigned long long)blkno); | |
6250 | ||
6251 | item->free_blk = blkno; | |
6252 | item->free_bit = bit; | |
6253 | item->free_next = ctxt->c_global_allocator; | |
6254 | ||
6255 | ctxt->c_global_allocator = item; | |
6256 | return ret; | |
6257 | } | |
6258 | ||
6259 | static int ocfs2_free_cached_clusters(struct ocfs2_super *osb, | |
6260 | struct ocfs2_cached_block_free *head) | |
6261 | { | |
6262 | struct ocfs2_cached_block_free *tmp; | |
6263 | struct inode *tl_inode = osb->osb_tl_inode; | |
6264 | handle_t *handle; | |
6265 | int ret = 0; | |
6266 | ||
6267 | mutex_lock(&tl_inode->i_mutex); | |
6268 | ||
6269 | while (head) { | |
6270 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
6271 | ret = __ocfs2_flush_truncate_log(osb); | |
6272 | if (ret < 0) { | |
6273 | mlog_errno(ret); | |
6274 | break; | |
6275 | } | |
6276 | } | |
6277 | ||
6278 | handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE); | |
6279 | if (IS_ERR(handle)) { | |
6280 | ret = PTR_ERR(handle); | |
6281 | mlog_errno(ret); | |
6282 | break; | |
6283 | } | |
6284 | ||
6285 | ret = ocfs2_truncate_log_append(osb, handle, head->free_blk, | |
6286 | head->free_bit); | |
6287 | ||
6288 | ocfs2_commit_trans(osb, handle); | |
6289 | tmp = head; | |
6290 | head = head->free_next; | |
6291 | kfree(tmp); | |
6292 | ||
6293 | if (ret < 0) { | |
6294 | mlog_errno(ret); | |
6295 | break; | |
6296 | } | |
6297 | } | |
6298 | ||
6299 | mutex_unlock(&tl_inode->i_mutex); | |
6300 | ||
6301 | while (head) { | |
6302 | /* Premature exit may have left some dangling items. */ | |
6303 | tmp = head; | |
6304 | head = head->free_next; | |
6305 | kfree(tmp); | |
6306 | } | |
6307 | ||
6308 | return ret; | |
6309 | } | |
6310 | ||
2b604351 MF |
6311 | int ocfs2_run_deallocs(struct ocfs2_super *osb, |
6312 | struct ocfs2_cached_dealloc_ctxt *ctxt) | |
6313 | { | |
6314 | int ret = 0, ret2; | |
6315 | struct ocfs2_per_slot_free_list *fl; | |
6316 | ||
6317 | if (!ctxt) | |
6318 | return 0; | |
6319 | ||
6320 | while (ctxt->c_first_suballocator) { | |
6321 | fl = ctxt->c_first_suballocator; | |
6322 | ||
6323 | if (fl->f_first) { | |
6324 | mlog(0, "Free items: (type %u, slot %d)\n", | |
6325 | fl->f_inode_type, fl->f_slot); | |
2891d290 TM |
6326 | ret2 = ocfs2_free_cached_blocks(osb, |
6327 | fl->f_inode_type, | |
6328 | fl->f_slot, | |
6329 | fl->f_first); | |
2b604351 MF |
6330 | if (ret2) |
6331 | mlog_errno(ret2); | |
6332 | if (!ret) | |
6333 | ret = ret2; | |
6334 | } | |
6335 | ||
6336 | ctxt->c_first_suballocator = fl->f_next_suballocator; | |
6337 | kfree(fl); | |
6338 | } | |
6339 | ||
2891d290 TM |
6340 | if (ctxt->c_global_allocator) { |
6341 | ret2 = ocfs2_free_cached_clusters(osb, | |
6342 | ctxt->c_global_allocator); | |
6343 | if (ret2) | |
6344 | mlog_errno(ret2); | |
6345 | if (!ret) | |
6346 | ret = ret2; | |
6347 | ||
6348 | ctxt->c_global_allocator = NULL; | |
6349 | } | |
6350 | ||
2b604351 MF |
6351 | return ret; |
6352 | } | |
6353 | ||
6354 | static struct ocfs2_per_slot_free_list * | |
6355 | ocfs2_find_per_slot_free_list(int type, | |
6356 | int slot, | |
6357 | struct ocfs2_cached_dealloc_ctxt *ctxt) | |
6358 | { | |
6359 | struct ocfs2_per_slot_free_list *fl = ctxt->c_first_suballocator; | |
6360 | ||
6361 | while (fl) { | |
6362 | if (fl->f_inode_type == type && fl->f_slot == slot) | |
6363 | return fl; | |
6364 | ||
6365 | fl = fl->f_next_suballocator; | |
6366 | } | |
6367 | ||
6368 | fl = kmalloc(sizeof(*fl), GFP_NOFS); | |
6369 | if (fl) { | |
6370 | fl->f_inode_type = type; | |
6371 | fl->f_slot = slot; | |
6372 | fl->f_first = NULL; | |
6373 | fl->f_next_suballocator = ctxt->c_first_suballocator; | |
6374 | ||
6375 | ctxt->c_first_suballocator = fl; | |
6376 | } | |
6377 | return fl; | |
6378 | } | |
6379 | ||
6380 | static int ocfs2_cache_block_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt, | |
6381 | int type, int slot, u64 blkno, | |
6382 | unsigned int bit) | |
6383 | { | |
6384 | int ret; | |
6385 | struct ocfs2_per_slot_free_list *fl; | |
6386 | struct ocfs2_cached_block_free *item; | |
6387 | ||
6388 | fl = ocfs2_find_per_slot_free_list(type, slot, ctxt); | |
6389 | if (fl == NULL) { | |
6390 | ret = -ENOMEM; | |
6391 | mlog_errno(ret); | |
6392 | goto out; | |
6393 | } | |
6394 | ||
6395 | item = kmalloc(sizeof(*item), GFP_NOFS); | |
6396 | if (item == NULL) { | |
6397 | ret = -ENOMEM; | |
6398 | mlog_errno(ret); | |
6399 | goto out; | |
6400 | } | |
6401 | ||
6402 | mlog(0, "Insert: (type %d, slot %u, bit %u, blk %llu)\n", | |
6403 | type, slot, bit, (unsigned long long)blkno); | |
6404 | ||
6405 | item->free_blk = blkno; | |
6406 | item->free_bit = bit; | |
6407 | item->free_next = fl->f_first; | |
6408 | ||
6409 | fl->f_first = item; | |
6410 | ||
6411 | ret = 0; | |
6412 | out: | |
6413 | return ret; | |
6414 | } | |
6415 | ||
59a5e416 MF |
6416 | static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt, |
6417 | struct ocfs2_extent_block *eb) | |
6418 | { | |
6419 | return ocfs2_cache_block_dealloc(ctxt, EXTENT_ALLOC_SYSTEM_INODE, | |
6420 | le16_to_cpu(eb->h_suballoc_slot), | |
6421 | le64_to_cpu(eb->h_blkno), | |
6422 | le16_to_cpu(eb->h_suballoc_bit)); | |
6423 | } | |
6424 | ||
ccd979bd MF |
6425 | /* This function will figure out whether the currently last extent |
6426 | * block will be deleted, and if it will, what the new last extent | |
6427 | * block will be so we can update his h_next_leaf_blk field, as well | |
6428 | * as the dinodes i_last_eb_blk */ | |
dcd0538f | 6429 | static int ocfs2_find_new_last_ext_blk(struct inode *inode, |
3a0782d0 | 6430 | unsigned int clusters_to_del, |
dcd0538f | 6431 | struct ocfs2_path *path, |
ccd979bd MF |
6432 | struct buffer_head **new_last_eb) |
6433 | { | |
3a0782d0 | 6434 | int next_free, ret = 0; |
dcd0538f | 6435 | u32 cpos; |
3a0782d0 | 6436 | struct ocfs2_extent_rec *rec; |
ccd979bd MF |
6437 | struct ocfs2_extent_block *eb; |
6438 | struct ocfs2_extent_list *el; | |
6439 | struct buffer_head *bh = NULL; | |
6440 | ||
6441 | *new_last_eb = NULL; | |
6442 | ||
ccd979bd | 6443 | /* we have no tree, so of course, no last_eb. */ |
dcd0538f MF |
6444 | if (!path->p_tree_depth) |
6445 | goto out; | |
ccd979bd MF |
6446 | |
6447 | /* trunc to zero special case - this makes tree_depth = 0 | |
6448 | * regardless of what it is. */ | |
3a0782d0 | 6449 | if (OCFS2_I(inode)->ip_clusters == clusters_to_del) |
dcd0538f | 6450 | goto out; |
ccd979bd | 6451 | |
dcd0538f | 6452 | el = path_leaf_el(path); |
ccd979bd MF |
6453 | BUG_ON(!el->l_next_free_rec); |
6454 | ||
3a0782d0 MF |
6455 | /* |
6456 | * Make sure that this extent list will actually be empty | |
6457 | * after we clear away the data. We can shortcut out if | |
6458 | * there's more than one non-empty extent in the | |
6459 | * list. Otherwise, a check of the remaining extent is | |
6460 | * necessary. | |
6461 | */ | |
6462 | next_free = le16_to_cpu(el->l_next_free_rec); | |
6463 | rec = NULL; | |
dcd0538f | 6464 | if (ocfs2_is_empty_extent(&el->l_recs[0])) { |
3a0782d0 | 6465 | if (next_free > 2) |
dcd0538f | 6466 | goto out; |
3a0782d0 MF |
6467 | |
6468 | /* We may have a valid extent in index 1, check it. */ | |
6469 | if (next_free == 2) | |
6470 | rec = &el->l_recs[1]; | |
6471 | ||
6472 | /* | |
6473 | * Fall through - no more nonempty extents, so we want | |
6474 | * to delete this leaf. | |
6475 | */ | |
6476 | } else { | |
6477 | if (next_free > 1) | |
6478 | goto out; | |
6479 | ||
6480 | rec = &el->l_recs[0]; | |
6481 | } | |
6482 | ||
6483 | if (rec) { | |
6484 | /* | |
6485 | * Check it we'll only be trimming off the end of this | |
6486 | * cluster. | |
6487 | */ | |
e48edee2 | 6488 | if (le16_to_cpu(rec->e_leaf_clusters) > clusters_to_del) |
3a0782d0 MF |
6489 | goto out; |
6490 | } | |
ccd979bd | 6491 | |
dcd0538f MF |
6492 | ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, path, &cpos); |
6493 | if (ret) { | |
6494 | mlog_errno(ret); | |
6495 | goto out; | |
6496 | } | |
ccd979bd | 6497 | |
facdb77f | 6498 | ret = ocfs2_find_leaf(INODE_CACHE(inode), path_root_el(path), cpos, &bh); |
dcd0538f MF |
6499 | if (ret) { |
6500 | mlog_errno(ret); | |
6501 | goto out; | |
6502 | } | |
ccd979bd | 6503 | |
dcd0538f MF |
6504 | eb = (struct ocfs2_extent_block *) bh->b_data; |
6505 | el = &eb->h_list; | |
5e96581a JB |
6506 | |
6507 | /* ocfs2_find_leaf() gets the eb from ocfs2_read_extent_block(). | |
6508 | * Any corruption is a code bug. */ | |
6509 | BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb)); | |
ccd979bd MF |
6510 | |
6511 | *new_last_eb = bh; | |
6512 | get_bh(*new_last_eb); | |
dcd0538f MF |
6513 | mlog(0, "returning block %llu, (cpos: %u)\n", |
6514 | (unsigned long long)le64_to_cpu(eb->h_blkno), cpos); | |
6515 | out: | |
6516 | brelse(bh); | |
ccd979bd | 6517 | |
dcd0538f | 6518 | return ret; |
ccd979bd MF |
6519 | } |
6520 | ||
3a0782d0 MF |
6521 | /* |
6522 | * Trim some clusters off the rightmost edge of a tree. Only called | |
6523 | * during truncate. | |
6524 | * | |
6525 | * The caller needs to: | |
6526 | * - start journaling of each path component. | |
6527 | * - compute and fully set up any new last ext block | |
6528 | */ | |
6529 | static int ocfs2_trim_tree(struct inode *inode, struct ocfs2_path *path, | |
6530 | handle_t *handle, struct ocfs2_truncate_context *tc, | |
6531 | u32 clusters_to_del, u64 *delete_start) | |
6532 | { | |
6533 | int ret, i, index = path->p_tree_depth; | |
6534 | u32 new_edge = 0; | |
6535 | u64 deleted_eb = 0; | |
6536 | struct buffer_head *bh; | |
6537 | struct ocfs2_extent_list *el; | |
6538 | struct ocfs2_extent_rec *rec; | |
6539 | ||
6540 | *delete_start = 0; | |
6541 | ||
6542 | while (index >= 0) { | |
6543 | bh = path->p_node[index].bh; | |
6544 | el = path->p_node[index].el; | |
6545 | ||
6546 | mlog(0, "traveling tree (index = %d, block = %llu)\n", | |
6547 | index, (unsigned long long)bh->b_blocknr); | |
6548 | ||
6549 | BUG_ON(le16_to_cpu(el->l_next_free_rec) == 0); | |
6550 | ||
6551 | if (index != | |
6552 | (path->p_tree_depth - le16_to_cpu(el->l_tree_depth))) { | |
6553 | ocfs2_error(inode->i_sb, | |
6554 | "Inode %lu has invalid ext. block %llu", | |
6555 | inode->i_ino, | |
6556 | (unsigned long long)bh->b_blocknr); | |
6557 | ret = -EROFS; | |
6558 | goto out; | |
6559 | } | |
6560 | ||
6561 | find_tail_record: | |
6562 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
6563 | rec = &el->l_recs[i]; | |
6564 | ||
6565 | mlog(0, "Extent list before: record %d: (%u, %u, %llu), " | |
6566 | "next = %u\n", i, le32_to_cpu(rec->e_cpos), | |
e48edee2 | 6567 | ocfs2_rec_clusters(el, rec), |
3a0782d0 MF |
6568 | (unsigned long long)le64_to_cpu(rec->e_blkno), |
6569 | le16_to_cpu(el->l_next_free_rec)); | |
6570 | ||
e48edee2 | 6571 | BUG_ON(ocfs2_rec_clusters(el, rec) < clusters_to_del); |
3a0782d0 MF |
6572 | |
6573 | if (le16_to_cpu(el->l_tree_depth) == 0) { | |
6574 | /* | |
6575 | * If the leaf block contains a single empty | |
6576 | * extent and no records, we can just remove | |
6577 | * the block. | |
6578 | */ | |
6579 | if (i == 0 && ocfs2_is_empty_extent(rec)) { | |
6580 | memset(rec, 0, | |
6581 | sizeof(struct ocfs2_extent_rec)); | |
6582 | el->l_next_free_rec = cpu_to_le16(0); | |
6583 | ||
6584 | goto delete; | |
6585 | } | |
6586 | ||
6587 | /* | |
6588 | * Remove any empty extents by shifting things | |
6589 | * left. That should make life much easier on | |
6590 | * the code below. This condition is rare | |
6591 | * enough that we shouldn't see a performance | |
6592 | * hit. | |
6593 | */ | |
6594 | if (ocfs2_is_empty_extent(&el->l_recs[0])) { | |
6595 | le16_add_cpu(&el->l_next_free_rec, -1); | |
6596 | ||
6597 | for(i = 0; | |
6598 | i < le16_to_cpu(el->l_next_free_rec); i++) | |
6599 | el->l_recs[i] = el->l_recs[i + 1]; | |
6600 | ||
6601 | memset(&el->l_recs[i], 0, | |
6602 | sizeof(struct ocfs2_extent_rec)); | |
6603 | ||
6604 | /* | |
6605 | * We've modified our extent list. The | |
6606 | * simplest way to handle this change | |
6607 | * is to being the search from the | |
6608 | * start again. | |
6609 | */ | |
6610 | goto find_tail_record; | |
6611 | } | |
6612 | ||
e48edee2 | 6613 | le16_add_cpu(&rec->e_leaf_clusters, -clusters_to_del); |
3a0782d0 MF |
6614 | |
6615 | /* | |
6616 | * We'll use "new_edge" on our way back up the | |
6617 | * tree to know what our rightmost cpos is. | |
6618 | */ | |
e48edee2 | 6619 | new_edge = le16_to_cpu(rec->e_leaf_clusters); |
3a0782d0 MF |
6620 | new_edge += le32_to_cpu(rec->e_cpos); |
6621 | ||
6622 | /* | |
6623 | * The caller will use this to delete data blocks. | |
6624 | */ | |
6625 | *delete_start = le64_to_cpu(rec->e_blkno) | |
6626 | + ocfs2_clusters_to_blocks(inode->i_sb, | |
e48edee2 | 6627 | le16_to_cpu(rec->e_leaf_clusters)); |
3a0782d0 MF |
6628 | |
6629 | /* | |
6630 | * If it's now empty, remove this record. | |
6631 | */ | |
e48edee2 | 6632 | if (le16_to_cpu(rec->e_leaf_clusters) == 0) { |
3a0782d0 MF |
6633 | memset(rec, 0, |
6634 | sizeof(struct ocfs2_extent_rec)); | |
6635 | le16_add_cpu(&el->l_next_free_rec, -1); | |
6636 | } | |
6637 | } else { | |
6638 | if (le64_to_cpu(rec->e_blkno) == deleted_eb) { | |
6639 | memset(rec, 0, | |
6640 | sizeof(struct ocfs2_extent_rec)); | |
6641 | le16_add_cpu(&el->l_next_free_rec, -1); | |
6642 | ||
6643 | goto delete; | |
6644 | } | |
6645 | ||
6646 | /* Can this actually happen? */ | |
6647 | if (le16_to_cpu(el->l_next_free_rec) == 0) | |
6648 | goto delete; | |
6649 | ||
6650 | /* | |
6651 | * We never actually deleted any clusters | |
6652 | * because our leaf was empty. There's no | |
6653 | * reason to adjust the rightmost edge then. | |
6654 | */ | |
6655 | if (new_edge == 0) | |
6656 | goto delete; | |
6657 | ||
e48edee2 MF |
6658 | rec->e_int_clusters = cpu_to_le32(new_edge); |
6659 | le32_add_cpu(&rec->e_int_clusters, | |
3a0782d0 MF |
6660 | -le32_to_cpu(rec->e_cpos)); |
6661 | ||
6662 | /* | |
6663 | * A deleted child record should have been | |
6664 | * caught above. | |
6665 | */ | |
e48edee2 | 6666 | BUG_ON(le32_to_cpu(rec->e_int_clusters) == 0); |
3a0782d0 MF |
6667 | } |
6668 | ||
6669 | delete: | |
6670 | ret = ocfs2_journal_dirty(handle, bh); | |
6671 | if (ret) { | |
6672 | mlog_errno(ret); | |
6673 | goto out; | |
6674 | } | |
6675 | ||
6676 | mlog(0, "extent list container %llu, after: record %d: " | |
6677 | "(%u, %u, %llu), next = %u.\n", | |
6678 | (unsigned long long)bh->b_blocknr, i, | |
e48edee2 | 6679 | le32_to_cpu(rec->e_cpos), ocfs2_rec_clusters(el, rec), |
3a0782d0 MF |
6680 | (unsigned long long)le64_to_cpu(rec->e_blkno), |
6681 | le16_to_cpu(el->l_next_free_rec)); | |
6682 | ||
6683 | /* | |
6684 | * We must be careful to only attempt delete of an | |
6685 | * extent block (and not the root inode block). | |
6686 | */ | |
6687 | if (index > 0 && le16_to_cpu(el->l_next_free_rec) == 0) { | |
6688 | struct ocfs2_extent_block *eb = | |
6689 | (struct ocfs2_extent_block *)bh->b_data; | |
6690 | ||
6691 | /* | |
6692 | * Save this for use when processing the | |
6693 | * parent block. | |
6694 | */ | |
6695 | deleted_eb = le64_to_cpu(eb->h_blkno); | |
6696 | ||
6697 | mlog(0, "deleting this extent block.\n"); | |
6698 | ||
8cb471e8 | 6699 | ocfs2_remove_from_cache(INODE_CACHE(inode), bh); |
3a0782d0 | 6700 | |
e48edee2 | 6701 | BUG_ON(ocfs2_rec_clusters(el, &el->l_recs[0])); |
3a0782d0 MF |
6702 | BUG_ON(le32_to_cpu(el->l_recs[0].e_cpos)); |
6703 | BUG_ON(le64_to_cpu(el->l_recs[0].e_blkno)); | |
6704 | ||
59a5e416 MF |
6705 | ret = ocfs2_cache_extent_block_free(&tc->tc_dealloc, eb); |
6706 | /* An error here is not fatal. */ | |
6707 | if (ret < 0) | |
6708 | mlog_errno(ret); | |
3a0782d0 MF |
6709 | } else { |
6710 | deleted_eb = 0; | |
6711 | } | |
6712 | ||
6713 | index--; | |
6714 | } | |
6715 | ||
6716 | ret = 0; | |
6717 | out: | |
6718 | return ret; | |
6719 | } | |
6720 | ||
ccd979bd MF |
6721 | static int ocfs2_do_truncate(struct ocfs2_super *osb, |
6722 | unsigned int clusters_to_del, | |
6723 | struct inode *inode, | |
6724 | struct buffer_head *fe_bh, | |
1fabe148 | 6725 | handle_t *handle, |
dcd0538f MF |
6726 | struct ocfs2_truncate_context *tc, |
6727 | struct ocfs2_path *path) | |
ccd979bd | 6728 | { |
3a0782d0 | 6729 | int status; |
ccd979bd | 6730 | struct ocfs2_dinode *fe; |
ccd979bd MF |
6731 | struct ocfs2_extent_block *last_eb = NULL; |
6732 | struct ocfs2_extent_list *el; | |
ccd979bd | 6733 | struct buffer_head *last_eb_bh = NULL; |
ccd979bd MF |
6734 | u64 delete_blk = 0; |
6735 | ||
6736 | fe = (struct ocfs2_dinode *) fe_bh->b_data; | |
6737 | ||
3a0782d0 | 6738 | status = ocfs2_find_new_last_ext_blk(inode, clusters_to_del, |
dcd0538f | 6739 | path, &last_eb_bh); |
ccd979bd MF |
6740 | if (status < 0) { |
6741 | mlog_errno(status); | |
6742 | goto bail; | |
6743 | } | |
dcd0538f MF |
6744 | |
6745 | /* | |
6746 | * Each component will be touched, so we might as well journal | |
6747 | * here to avoid having to handle errors later. | |
6748 | */ | |
0cf2f763 | 6749 | status = ocfs2_journal_access_path(INODE_CACHE(inode), handle, path); |
3a0782d0 MF |
6750 | if (status < 0) { |
6751 | mlog_errno(status); | |
6752 | goto bail; | |
dcd0538f MF |
6753 | } |
6754 | ||
6755 | if (last_eb_bh) { | |
0cf2f763 | 6756 | status = ocfs2_journal_access_eb(handle, INODE_CACHE(inode), last_eb_bh, |
13723d00 | 6757 | OCFS2_JOURNAL_ACCESS_WRITE); |
dcd0538f MF |
6758 | if (status < 0) { |
6759 | mlog_errno(status); | |
6760 | goto bail; | |
6761 | } | |
6762 | ||
ccd979bd | 6763 | last_eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; |
dcd0538f | 6764 | } |
ccd979bd | 6765 | |
dcd0538f MF |
6766 | el = &(fe->id2.i_list); |
6767 | ||
6768 | /* | |
6769 | * Lower levels depend on this never happening, but it's best | |
6770 | * to check it up here before changing the tree. | |
6771 | */ | |
e48edee2 | 6772 | if (el->l_tree_depth && el->l_recs[0].e_int_clusters == 0) { |
dcd0538f MF |
6773 | ocfs2_error(inode->i_sb, |
6774 | "Inode %lu has an empty extent record, depth %u\n", | |
6775 | inode->i_ino, le16_to_cpu(el->l_tree_depth)); | |
3a0782d0 | 6776 | status = -EROFS; |
ccd979bd MF |
6777 | goto bail; |
6778 | } | |
ccd979bd | 6779 | |
a90714c1 JK |
6780 | vfs_dq_free_space_nodirty(inode, |
6781 | ocfs2_clusters_to_bytes(osb->sb, clusters_to_del)); | |
ccd979bd MF |
6782 | spin_lock(&OCFS2_I(inode)->ip_lock); |
6783 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters) - | |
6784 | clusters_to_del; | |
6785 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
6786 | le32_add_cpu(&fe->i_clusters, -clusters_to_del); | |
e535e2ef | 6787 | inode->i_blocks = ocfs2_inode_sector_count(inode); |
ccd979bd | 6788 | |
3a0782d0 MF |
6789 | status = ocfs2_trim_tree(inode, path, handle, tc, |
6790 | clusters_to_del, &delete_blk); | |
6791 | if (status) { | |
6792 | mlog_errno(status); | |
6793 | goto bail; | |
ccd979bd MF |
6794 | } |
6795 | ||
dcd0538f | 6796 | if (le32_to_cpu(fe->i_clusters) == 0) { |
ccd979bd MF |
6797 | /* trunc to zero is a special case. */ |
6798 | el->l_tree_depth = 0; | |
6799 | fe->i_last_eb_blk = 0; | |
6800 | } else if (last_eb) | |
6801 | fe->i_last_eb_blk = last_eb->h_blkno; | |
6802 | ||
6803 | status = ocfs2_journal_dirty(handle, fe_bh); | |
6804 | if (status < 0) { | |
6805 | mlog_errno(status); | |
6806 | goto bail; | |
6807 | } | |
6808 | ||
6809 | if (last_eb) { | |
6810 | /* If there will be a new last extent block, then by | |
6811 | * definition, there cannot be any leaves to the right of | |
6812 | * him. */ | |
ccd979bd MF |
6813 | last_eb->h_next_leaf_blk = 0; |
6814 | status = ocfs2_journal_dirty(handle, last_eb_bh); | |
6815 | if (status < 0) { | |
6816 | mlog_errno(status); | |
6817 | goto bail; | |
6818 | } | |
6819 | } | |
6820 | ||
3a0782d0 MF |
6821 | if (delete_blk) { |
6822 | status = ocfs2_truncate_log_append(osb, handle, delete_blk, | |
6823 | clusters_to_del); | |
ccd979bd MF |
6824 | if (status < 0) { |
6825 | mlog_errno(status); | |
6826 | goto bail; | |
6827 | } | |
ccd979bd MF |
6828 | } |
6829 | status = 0; | |
6830 | bail: | |
60e2ec48 | 6831 | brelse(last_eb_bh); |
ccd979bd MF |
6832 | mlog_exit(status); |
6833 | return status; | |
6834 | } | |
6835 | ||
2b4e30fb | 6836 | static int ocfs2_zero_func(handle_t *handle, struct buffer_head *bh) |
60b11392 MF |
6837 | { |
6838 | set_buffer_uptodate(bh); | |
6839 | mark_buffer_dirty(bh); | |
6840 | return 0; | |
6841 | } | |
6842 | ||
1d410a6e MF |
6843 | static void ocfs2_map_and_dirty_page(struct inode *inode, handle_t *handle, |
6844 | unsigned int from, unsigned int to, | |
6845 | struct page *page, int zero, u64 *phys) | |
6846 | { | |
6847 | int ret, partial = 0; | |
6848 | ||
6849 | ret = ocfs2_map_page_blocks(page, phys, inode, from, to, 0); | |
6850 | if (ret) | |
6851 | mlog_errno(ret); | |
6852 | ||
6853 | if (zero) | |
eebd2aa3 | 6854 | zero_user_segment(page, from, to); |
1d410a6e MF |
6855 | |
6856 | /* | |
6857 | * Need to set the buffers we zero'd into uptodate | |
6858 | * here if they aren't - ocfs2_map_page_blocks() | |
6859 | * might've skipped some | |
6860 | */ | |
2b4e30fb JB |
6861 | ret = walk_page_buffers(handle, page_buffers(page), |
6862 | from, to, &partial, | |
6863 | ocfs2_zero_func); | |
6864 | if (ret < 0) | |
6865 | mlog_errno(ret); | |
6866 | else if (ocfs2_should_order_data(inode)) { | |
6867 | ret = ocfs2_jbd2_file_inode(handle, inode); | |
1d410a6e MF |
6868 | if (ret < 0) |
6869 | mlog_errno(ret); | |
6870 | } | |
6871 | ||
6872 | if (!partial) | |
6873 | SetPageUptodate(page); | |
6874 | ||
6875 | flush_dcache_page(page); | |
6876 | } | |
6877 | ||
35edec1d MF |
6878 | static void ocfs2_zero_cluster_pages(struct inode *inode, loff_t start, |
6879 | loff_t end, struct page **pages, | |
6880 | int numpages, u64 phys, handle_t *handle) | |
60b11392 | 6881 | { |
1d410a6e | 6882 | int i; |
60b11392 MF |
6883 | struct page *page; |
6884 | unsigned int from, to = PAGE_CACHE_SIZE; | |
6885 | struct super_block *sb = inode->i_sb; | |
6886 | ||
6887 | BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(sb))); | |
6888 | ||
6889 | if (numpages == 0) | |
6890 | goto out; | |
6891 | ||
35edec1d | 6892 | to = PAGE_CACHE_SIZE; |
60b11392 MF |
6893 | for(i = 0; i < numpages; i++) { |
6894 | page = pages[i]; | |
6895 | ||
35edec1d MF |
6896 | from = start & (PAGE_CACHE_SIZE - 1); |
6897 | if ((end >> PAGE_CACHE_SHIFT) == page->index) | |
6898 | to = end & (PAGE_CACHE_SIZE - 1); | |
6899 | ||
60b11392 MF |
6900 | BUG_ON(from > PAGE_CACHE_SIZE); |
6901 | BUG_ON(to > PAGE_CACHE_SIZE); | |
6902 | ||
1d410a6e MF |
6903 | ocfs2_map_and_dirty_page(inode, handle, from, to, page, 1, |
6904 | &phys); | |
60b11392 | 6905 | |
35edec1d | 6906 | start = (page->index + 1) << PAGE_CACHE_SHIFT; |
60b11392 MF |
6907 | } |
6908 | out: | |
1d410a6e MF |
6909 | if (pages) |
6910 | ocfs2_unlock_and_free_pages(pages, numpages); | |
60b11392 MF |
6911 | } |
6912 | ||
35edec1d | 6913 | static int ocfs2_grab_eof_pages(struct inode *inode, loff_t start, loff_t end, |
1d410a6e | 6914 | struct page **pages, int *num) |
60b11392 | 6915 | { |
1d410a6e | 6916 | int numpages, ret = 0; |
60b11392 MF |
6917 | struct super_block *sb = inode->i_sb; |
6918 | struct address_space *mapping = inode->i_mapping; | |
6919 | unsigned long index; | |
35edec1d | 6920 | loff_t last_page_bytes; |
60b11392 | 6921 | |
35edec1d | 6922 | BUG_ON(start > end); |
60b11392 | 6923 | |
35edec1d MF |
6924 | BUG_ON(start >> OCFS2_SB(sb)->s_clustersize_bits != |
6925 | (end - 1) >> OCFS2_SB(sb)->s_clustersize_bits); | |
6926 | ||
1d410a6e | 6927 | numpages = 0; |
35edec1d MF |
6928 | last_page_bytes = PAGE_ALIGN(end); |
6929 | index = start >> PAGE_CACHE_SHIFT; | |
60b11392 MF |
6930 | do { |
6931 | pages[numpages] = grab_cache_page(mapping, index); | |
6932 | if (!pages[numpages]) { | |
6933 | ret = -ENOMEM; | |
6934 | mlog_errno(ret); | |
6935 | goto out; | |
6936 | } | |
6937 | ||
6938 | numpages++; | |
6939 | index++; | |
35edec1d | 6940 | } while (index < (last_page_bytes >> PAGE_CACHE_SHIFT)); |
60b11392 MF |
6941 | |
6942 | out: | |
6943 | if (ret != 0) { | |
1d410a6e MF |
6944 | if (pages) |
6945 | ocfs2_unlock_and_free_pages(pages, numpages); | |
60b11392 MF |
6946 | numpages = 0; |
6947 | } | |
6948 | ||
6949 | *num = numpages; | |
6950 | ||
6951 | return ret; | |
6952 | } | |
6953 | ||
6954 | /* | |
6955 | * Zero the area past i_size but still within an allocated | |
6956 | * cluster. This avoids exposing nonzero data on subsequent file | |
6957 | * extends. | |
6958 | * | |
6959 | * We need to call this before i_size is updated on the inode because | |
6960 | * otherwise block_write_full_page() will skip writeout of pages past | |
6961 | * i_size. The new_i_size parameter is passed for this reason. | |
6962 | */ | |
35edec1d MF |
6963 | int ocfs2_zero_range_for_truncate(struct inode *inode, handle_t *handle, |
6964 | u64 range_start, u64 range_end) | |
60b11392 | 6965 | { |
1d410a6e | 6966 | int ret = 0, numpages; |
60b11392 MF |
6967 | struct page **pages = NULL; |
6968 | u64 phys; | |
1d410a6e MF |
6969 | unsigned int ext_flags; |
6970 | struct super_block *sb = inode->i_sb; | |
60b11392 MF |
6971 | |
6972 | /* | |
6973 | * File systems which don't support sparse files zero on every | |
6974 | * extend. | |
6975 | */ | |
1d410a6e | 6976 | if (!ocfs2_sparse_alloc(OCFS2_SB(sb))) |
60b11392 MF |
6977 | return 0; |
6978 | ||
1d410a6e | 6979 | pages = kcalloc(ocfs2_pages_per_cluster(sb), |
60b11392 MF |
6980 | sizeof(struct page *), GFP_NOFS); |
6981 | if (pages == NULL) { | |
6982 | ret = -ENOMEM; | |
6983 | mlog_errno(ret); | |
6984 | goto out; | |
6985 | } | |
6986 | ||
1d410a6e MF |
6987 | if (range_start == range_end) |
6988 | goto out; | |
6989 | ||
6990 | ret = ocfs2_extent_map_get_blocks(inode, | |
6991 | range_start >> sb->s_blocksize_bits, | |
6992 | &phys, NULL, &ext_flags); | |
60b11392 MF |
6993 | if (ret) { |
6994 | mlog_errno(ret); | |
6995 | goto out; | |
6996 | } | |
6997 | ||
1d410a6e MF |
6998 | /* |
6999 | * Tail is a hole, or is marked unwritten. In either case, we | |
7000 | * can count on read and write to return/push zero's. | |
7001 | */ | |
7002 | if (phys == 0 || ext_flags & OCFS2_EXT_UNWRITTEN) | |
60b11392 MF |
7003 | goto out; |
7004 | ||
1d410a6e MF |
7005 | ret = ocfs2_grab_eof_pages(inode, range_start, range_end, pages, |
7006 | &numpages); | |
7007 | if (ret) { | |
7008 | mlog_errno(ret); | |
7009 | goto out; | |
7010 | } | |
7011 | ||
35edec1d MF |
7012 | ocfs2_zero_cluster_pages(inode, range_start, range_end, pages, |
7013 | numpages, phys, handle); | |
60b11392 MF |
7014 | |
7015 | /* | |
7016 | * Initiate writeout of the pages we zero'd here. We don't | |
7017 | * wait on them - the truncate_inode_pages() call later will | |
7018 | * do that for us. | |
7019 | */ | |
35edec1d MF |
7020 | ret = do_sync_mapping_range(inode->i_mapping, range_start, |
7021 | range_end - 1, SYNC_FILE_RANGE_WRITE); | |
60b11392 MF |
7022 | if (ret) |
7023 | mlog_errno(ret); | |
7024 | ||
7025 | out: | |
7026 | if (pages) | |
7027 | kfree(pages); | |
7028 | ||
7029 | return ret; | |
7030 | } | |
7031 | ||
fdd77704 TY |
7032 | static void ocfs2_zero_dinode_id2_with_xattr(struct inode *inode, |
7033 | struct ocfs2_dinode *di) | |
1afc32b9 MF |
7034 | { |
7035 | unsigned int blocksize = 1 << inode->i_sb->s_blocksize_bits; | |
fdd77704 | 7036 | unsigned int xattrsize = le16_to_cpu(di->i_xattr_inline_size); |
1afc32b9 | 7037 | |
fdd77704 TY |
7038 | if (le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_XATTR_FL) |
7039 | memset(&di->id2, 0, blocksize - | |
7040 | offsetof(struct ocfs2_dinode, id2) - | |
7041 | xattrsize); | |
7042 | else | |
7043 | memset(&di->id2, 0, blocksize - | |
7044 | offsetof(struct ocfs2_dinode, id2)); | |
1afc32b9 MF |
7045 | } |
7046 | ||
5b6a3a2b MF |
7047 | void ocfs2_dinode_new_extent_list(struct inode *inode, |
7048 | struct ocfs2_dinode *di) | |
7049 | { | |
fdd77704 | 7050 | ocfs2_zero_dinode_id2_with_xattr(inode, di); |
5b6a3a2b MF |
7051 | di->id2.i_list.l_tree_depth = 0; |
7052 | di->id2.i_list.l_next_free_rec = 0; | |
fdd77704 TY |
7053 | di->id2.i_list.l_count = cpu_to_le16( |
7054 | ocfs2_extent_recs_per_inode_with_xattr(inode->i_sb, di)); | |
5b6a3a2b MF |
7055 | } |
7056 | ||
1afc32b9 MF |
7057 | void ocfs2_set_inode_data_inline(struct inode *inode, struct ocfs2_dinode *di) |
7058 | { | |
7059 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
7060 | struct ocfs2_inline_data *idata = &di->id2.i_data; | |
7061 | ||
7062 | spin_lock(&oi->ip_lock); | |
7063 | oi->ip_dyn_features |= OCFS2_INLINE_DATA_FL; | |
7064 | di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features); | |
7065 | spin_unlock(&oi->ip_lock); | |
7066 | ||
7067 | /* | |
7068 | * We clear the entire i_data structure here so that all | |
7069 | * fields can be properly initialized. | |
7070 | */ | |
fdd77704 | 7071 | ocfs2_zero_dinode_id2_with_xattr(inode, di); |
1afc32b9 | 7072 | |
fdd77704 TY |
7073 | idata->id_count = cpu_to_le16( |
7074 | ocfs2_max_inline_data_with_xattr(inode->i_sb, di)); | |
1afc32b9 MF |
7075 | } |
7076 | ||
7077 | int ocfs2_convert_inline_data_to_extents(struct inode *inode, | |
7078 | struct buffer_head *di_bh) | |
7079 | { | |
7080 | int ret, i, has_data, num_pages = 0; | |
7081 | handle_t *handle; | |
7082 | u64 uninitialized_var(block); | |
7083 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
7084 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
7085 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
1afc32b9 MF |
7086 | struct ocfs2_alloc_context *data_ac = NULL; |
7087 | struct page **pages = NULL; | |
7088 | loff_t end = osb->s_clustersize; | |
f99b9b7c | 7089 | struct ocfs2_extent_tree et; |
a90714c1 | 7090 | int did_quota = 0; |
1afc32b9 MF |
7091 | |
7092 | has_data = i_size_read(inode) ? 1 : 0; | |
7093 | ||
7094 | if (has_data) { | |
7095 | pages = kcalloc(ocfs2_pages_per_cluster(osb->sb), | |
7096 | sizeof(struct page *), GFP_NOFS); | |
7097 | if (pages == NULL) { | |
7098 | ret = -ENOMEM; | |
7099 | mlog_errno(ret); | |
7100 | goto out; | |
7101 | } | |
7102 | ||
7103 | ret = ocfs2_reserve_clusters(osb, 1, &data_ac); | |
7104 | if (ret) { | |
7105 | mlog_errno(ret); | |
7106 | goto out; | |
7107 | } | |
7108 | } | |
7109 | ||
a90714c1 JK |
7110 | handle = ocfs2_start_trans(osb, |
7111 | ocfs2_inline_to_extents_credits(osb->sb)); | |
1afc32b9 MF |
7112 | if (IS_ERR(handle)) { |
7113 | ret = PTR_ERR(handle); | |
7114 | mlog_errno(ret); | |
7115 | goto out_unlock; | |
7116 | } | |
7117 | ||
0cf2f763 | 7118 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh, |
13723d00 | 7119 | OCFS2_JOURNAL_ACCESS_WRITE); |
1afc32b9 MF |
7120 | if (ret) { |
7121 | mlog_errno(ret); | |
7122 | goto out_commit; | |
7123 | } | |
7124 | ||
7125 | if (has_data) { | |
7126 | u32 bit_off, num; | |
7127 | unsigned int page_end; | |
7128 | u64 phys; | |
7129 | ||
a90714c1 JK |
7130 | if (vfs_dq_alloc_space_nodirty(inode, |
7131 | ocfs2_clusters_to_bytes(osb->sb, 1))) { | |
7132 | ret = -EDQUOT; | |
7133 | goto out_commit; | |
7134 | } | |
7135 | did_quota = 1; | |
7136 | ||
1afc32b9 MF |
7137 | ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off, |
7138 | &num); | |
7139 | if (ret) { | |
7140 | mlog_errno(ret); | |
7141 | goto out_commit; | |
7142 | } | |
7143 | ||
7144 | /* | |
7145 | * Save two copies, one for insert, and one that can | |
7146 | * be changed by ocfs2_map_and_dirty_page() below. | |
7147 | */ | |
7148 | block = phys = ocfs2_clusters_to_blocks(inode->i_sb, bit_off); | |
7149 | ||
7150 | /* | |
7151 | * Non sparse file systems zero on extend, so no need | |
7152 | * to do that now. | |
7153 | */ | |
7154 | if (!ocfs2_sparse_alloc(osb) && | |
7155 | PAGE_CACHE_SIZE < osb->s_clustersize) | |
7156 | end = PAGE_CACHE_SIZE; | |
7157 | ||
7158 | ret = ocfs2_grab_eof_pages(inode, 0, end, pages, &num_pages); | |
7159 | if (ret) { | |
7160 | mlog_errno(ret); | |
7161 | goto out_commit; | |
7162 | } | |
7163 | ||
7164 | /* | |
7165 | * This should populate the 1st page for us and mark | |
7166 | * it up to date. | |
7167 | */ | |
7168 | ret = ocfs2_read_inline_data(inode, pages[0], di_bh); | |
7169 | if (ret) { | |
7170 | mlog_errno(ret); | |
7171 | goto out_commit; | |
7172 | } | |
7173 | ||
7174 | page_end = PAGE_CACHE_SIZE; | |
7175 | if (PAGE_CACHE_SIZE > osb->s_clustersize) | |
7176 | page_end = osb->s_clustersize; | |
7177 | ||
7178 | for (i = 0; i < num_pages; i++) | |
7179 | ocfs2_map_and_dirty_page(inode, handle, 0, page_end, | |
7180 | pages[i], i > 0, &phys); | |
7181 | } | |
7182 | ||
7183 | spin_lock(&oi->ip_lock); | |
7184 | oi->ip_dyn_features &= ~OCFS2_INLINE_DATA_FL; | |
7185 | di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features); | |
7186 | spin_unlock(&oi->ip_lock); | |
7187 | ||
5b6a3a2b | 7188 | ocfs2_dinode_new_extent_list(inode, di); |
1afc32b9 MF |
7189 | |
7190 | ocfs2_journal_dirty(handle, di_bh); | |
7191 | ||
7192 | if (has_data) { | |
7193 | /* | |
7194 | * An error at this point should be extremely rare. If | |
7195 | * this proves to be false, we could always re-build | |
7196 | * the in-inode data from our pages. | |
7197 | */ | |
8d6220d6 | 7198 | ocfs2_init_dinode_extent_tree(&et, inode, di_bh); |
f99b9b7c JB |
7199 | ret = ocfs2_insert_extent(osb, handle, inode, &et, |
7200 | 0, block, 1, 0, NULL); | |
1afc32b9 MF |
7201 | if (ret) { |
7202 | mlog_errno(ret); | |
7203 | goto out_commit; | |
7204 | } | |
7205 | ||
7206 | inode->i_blocks = ocfs2_inode_sector_count(inode); | |
7207 | } | |
7208 | ||
7209 | out_commit: | |
a90714c1 JK |
7210 | if (ret < 0 && did_quota) |
7211 | vfs_dq_free_space_nodirty(inode, | |
7212 | ocfs2_clusters_to_bytes(osb->sb, 1)); | |
7213 | ||
1afc32b9 MF |
7214 | ocfs2_commit_trans(osb, handle); |
7215 | ||
7216 | out_unlock: | |
7217 | if (data_ac) | |
7218 | ocfs2_free_alloc_context(data_ac); | |
7219 | ||
7220 | out: | |
7221 | if (pages) { | |
7222 | ocfs2_unlock_and_free_pages(pages, num_pages); | |
7223 | kfree(pages); | |
7224 | } | |
7225 | ||
7226 | return ret; | |
7227 | } | |
7228 | ||
ccd979bd MF |
7229 | /* |
7230 | * It is expected, that by the time you call this function, | |
7231 | * inode->i_size and fe->i_size have been adjusted. | |
7232 | * | |
7233 | * WARNING: This will kfree the truncate context | |
7234 | */ | |
7235 | int ocfs2_commit_truncate(struct ocfs2_super *osb, | |
7236 | struct inode *inode, | |
7237 | struct buffer_head *fe_bh, | |
7238 | struct ocfs2_truncate_context *tc) | |
7239 | { | |
7240 | int status, i, credits, tl_sem = 0; | |
dcd0538f | 7241 | u32 clusters_to_del, new_highest_cpos, range; |
ccd979bd | 7242 | struct ocfs2_extent_list *el; |
1fabe148 | 7243 | handle_t *handle = NULL; |
ccd979bd | 7244 | struct inode *tl_inode = osb->osb_tl_inode; |
dcd0538f | 7245 | struct ocfs2_path *path = NULL; |
e7d4cb6b | 7246 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data; |
ccd979bd MF |
7247 | |
7248 | mlog_entry_void(); | |
7249 | ||
dcd0538f | 7250 | new_highest_cpos = ocfs2_clusters_for_bytes(osb->sb, |
ccd979bd MF |
7251 | i_size_read(inode)); |
7252 | ||
13723d00 JB |
7253 | path = ocfs2_new_path(fe_bh, &di->id2.i_list, |
7254 | ocfs2_journal_access_di); | |
dcd0538f MF |
7255 | if (!path) { |
7256 | status = -ENOMEM; | |
7257 | mlog_errno(status); | |
7258 | goto bail; | |
7259 | } | |
83418978 MF |
7260 | |
7261 | ocfs2_extent_map_trunc(inode, new_highest_cpos); | |
7262 | ||
ccd979bd | 7263 | start: |
3a0782d0 MF |
7264 | /* |
7265 | * Check that we still have allocation to delete. | |
7266 | */ | |
7267 | if (OCFS2_I(inode)->ip_clusters == 0) { | |
7268 | status = 0; | |
7269 | goto bail; | |
7270 | } | |
7271 | ||
dcd0538f MF |
7272 | /* |
7273 | * Truncate always works against the rightmost tree branch. | |
7274 | */ | |
facdb77f | 7275 | status = ocfs2_find_path(INODE_CACHE(inode), path, UINT_MAX); |
dcd0538f MF |
7276 | if (status) { |
7277 | mlog_errno(status); | |
7278 | goto bail; | |
ccd979bd MF |
7279 | } |
7280 | ||
dcd0538f MF |
7281 | mlog(0, "inode->ip_clusters = %u, tree_depth = %u\n", |
7282 | OCFS2_I(inode)->ip_clusters, path->p_tree_depth); | |
7283 | ||
7284 | /* | |
7285 | * By now, el will point to the extent list on the bottom most | |
7286 | * portion of this tree. Only the tail record is considered in | |
7287 | * each pass. | |
7288 | * | |
7289 | * We handle the following cases, in order: | |
7290 | * - empty extent: delete the remaining branch | |
7291 | * - remove the entire record | |
7292 | * - remove a partial record | |
7293 | * - no record needs to be removed (truncate has completed) | |
7294 | */ | |
7295 | el = path_leaf_el(path); | |
3a0782d0 MF |
7296 | if (le16_to_cpu(el->l_next_free_rec) == 0) { |
7297 | ocfs2_error(inode->i_sb, | |
7298 | "Inode %llu has empty extent block at %llu\n", | |
7299 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
7300 | (unsigned long long)path_leaf_bh(path)->b_blocknr); | |
7301 | status = -EROFS; | |
7302 | goto bail; | |
7303 | } | |
7304 | ||
ccd979bd | 7305 | i = le16_to_cpu(el->l_next_free_rec) - 1; |
dcd0538f | 7306 | range = le32_to_cpu(el->l_recs[i].e_cpos) + |
e48edee2 | 7307 | ocfs2_rec_clusters(el, &el->l_recs[i]); |
dcd0538f MF |
7308 | if (i == 0 && ocfs2_is_empty_extent(&el->l_recs[i])) { |
7309 | clusters_to_del = 0; | |
7310 | } else if (le32_to_cpu(el->l_recs[i].e_cpos) >= new_highest_cpos) { | |
e48edee2 | 7311 | clusters_to_del = ocfs2_rec_clusters(el, &el->l_recs[i]); |
dcd0538f | 7312 | } else if (range > new_highest_cpos) { |
e48edee2 | 7313 | clusters_to_del = (ocfs2_rec_clusters(el, &el->l_recs[i]) + |
ccd979bd | 7314 | le32_to_cpu(el->l_recs[i].e_cpos)) - |
dcd0538f MF |
7315 | new_highest_cpos; |
7316 | } else { | |
7317 | status = 0; | |
7318 | goto bail; | |
7319 | } | |
ccd979bd | 7320 | |
dcd0538f MF |
7321 | mlog(0, "clusters_to_del = %u in this pass, tail blk=%llu\n", |
7322 | clusters_to_del, (unsigned long long)path_leaf_bh(path)->b_blocknr); | |
7323 | ||
1b1dcc1b | 7324 | mutex_lock(&tl_inode->i_mutex); |
ccd979bd MF |
7325 | tl_sem = 1; |
7326 | /* ocfs2_truncate_log_needs_flush guarantees us at least one | |
7327 | * record is free for use. If there isn't any, we flush to get | |
7328 | * an empty truncate log. */ | |
7329 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
7330 | status = __ocfs2_flush_truncate_log(osb); | |
7331 | if (status < 0) { | |
7332 | mlog_errno(status); | |
7333 | goto bail; | |
7334 | } | |
7335 | } | |
7336 | ||
7337 | credits = ocfs2_calc_tree_trunc_credits(osb->sb, clusters_to_del, | |
dcd0538f MF |
7338 | (struct ocfs2_dinode *)fe_bh->b_data, |
7339 | el); | |
65eff9cc | 7340 | handle = ocfs2_start_trans(osb, credits); |
ccd979bd MF |
7341 | if (IS_ERR(handle)) { |
7342 | status = PTR_ERR(handle); | |
7343 | handle = NULL; | |
7344 | mlog_errno(status); | |
7345 | goto bail; | |
7346 | } | |
7347 | ||
dcd0538f MF |
7348 | status = ocfs2_do_truncate(osb, clusters_to_del, inode, fe_bh, handle, |
7349 | tc, path); | |
ccd979bd MF |
7350 | if (status < 0) { |
7351 | mlog_errno(status); | |
7352 | goto bail; | |
7353 | } | |
7354 | ||
1b1dcc1b | 7355 | mutex_unlock(&tl_inode->i_mutex); |
ccd979bd MF |
7356 | tl_sem = 0; |
7357 | ||
02dc1af4 | 7358 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
7359 | handle = NULL; |
7360 | ||
dcd0538f MF |
7361 | ocfs2_reinit_path(path, 1); |
7362 | ||
7363 | /* | |
3a0782d0 MF |
7364 | * The check above will catch the case where we've truncated |
7365 | * away all allocation. | |
dcd0538f | 7366 | */ |
3a0782d0 MF |
7367 | goto start; |
7368 | ||
ccd979bd | 7369 | bail: |
ccd979bd MF |
7370 | |
7371 | ocfs2_schedule_truncate_log_flush(osb, 1); | |
7372 | ||
7373 | if (tl_sem) | |
1b1dcc1b | 7374 | mutex_unlock(&tl_inode->i_mutex); |
ccd979bd MF |
7375 | |
7376 | if (handle) | |
02dc1af4 | 7377 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 7378 | |
59a5e416 MF |
7379 | ocfs2_run_deallocs(osb, &tc->tc_dealloc); |
7380 | ||
dcd0538f | 7381 | ocfs2_free_path(path); |
ccd979bd MF |
7382 | |
7383 | /* This will drop the ext_alloc cluster lock for us */ | |
7384 | ocfs2_free_truncate_context(tc); | |
7385 | ||
7386 | mlog_exit(status); | |
7387 | return status; | |
7388 | } | |
7389 | ||
ccd979bd | 7390 | /* |
59a5e416 | 7391 | * Expects the inode to already be locked. |
ccd979bd MF |
7392 | */ |
7393 | int ocfs2_prepare_truncate(struct ocfs2_super *osb, | |
7394 | struct inode *inode, | |
7395 | struct buffer_head *fe_bh, | |
7396 | struct ocfs2_truncate_context **tc) | |
7397 | { | |
59a5e416 | 7398 | int status; |
ccd979bd MF |
7399 | unsigned int new_i_clusters; |
7400 | struct ocfs2_dinode *fe; | |
7401 | struct ocfs2_extent_block *eb; | |
ccd979bd | 7402 | struct buffer_head *last_eb_bh = NULL; |
ccd979bd MF |
7403 | |
7404 | mlog_entry_void(); | |
7405 | ||
7406 | *tc = NULL; | |
7407 | ||
7408 | new_i_clusters = ocfs2_clusters_for_bytes(osb->sb, | |
7409 | i_size_read(inode)); | |
7410 | fe = (struct ocfs2_dinode *) fe_bh->b_data; | |
7411 | ||
7412 | mlog(0, "fe->i_clusters = %u, new_i_clusters = %u, fe->i_size =" | |
1ca1a111 MF |
7413 | "%llu\n", le32_to_cpu(fe->i_clusters), new_i_clusters, |
7414 | (unsigned long long)le64_to_cpu(fe->i_size)); | |
ccd979bd | 7415 | |
cd861280 | 7416 | *tc = kzalloc(sizeof(struct ocfs2_truncate_context), GFP_KERNEL); |
ccd979bd MF |
7417 | if (!(*tc)) { |
7418 | status = -ENOMEM; | |
7419 | mlog_errno(status); | |
7420 | goto bail; | |
7421 | } | |
59a5e416 | 7422 | ocfs2_init_dealloc_ctxt(&(*tc)->tc_dealloc); |
ccd979bd | 7423 | |
ccd979bd | 7424 | if (fe->id2.i_list.l_tree_depth) { |
3d03a305 | 7425 | status = ocfs2_read_extent_block(INODE_CACHE(inode), |
5e96581a JB |
7426 | le64_to_cpu(fe->i_last_eb_blk), |
7427 | &last_eb_bh); | |
ccd979bd MF |
7428 | if (status < 0) { |
7429 | mlog_errno(status); | |
7430 | goto bail; | |
7431 | } | |
7432 | eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; | |
ccd979bd MF |
7433 | } |
7434 | ||
7435 | (*tc)->tc_last_eb_bh = last_eb_bh; | |
7436 | ||
ccd979bd MF |
7437 | status = 0; |
7438 | bail: | |
7439 | if (status < 0) { | |
7440 | if (*tc) | |
7441 | ocfs2_free_truncate_context(*tc); | |
7442 | *tc = NULL; | |
7443 | } | |
7444 | mlog_exit_void(); | |
7445 | return status; | |
7446 | } | |
7447 | ||
1afc32b9 MF |
7448 | /* |
7449 | * 'start' is inclusive, 'end' is not. | |
7450 | */ | |
7451 | int ocfs2_truncate_inline(struct inode *inode, struct buffer_head *di_bh, | |
7452 | unsigned int start, unsigned int end, int trunc) | |
7453 | { | |
7454 | int ret; | |
7455 | unsigned int numbytes; | |
7456 | handle_t *handle; | |
7457 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
7458 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
7459 | struct ocfs2_inline_data *idata = &di->id2.i_data; | |
7460 | ||
7461 | if (end > i_size_read(inode)) | |
7462 | end = i_size_read(inode); | |
7463 | ||
7464 | BUG_ON(start >= end); | |
7465 | ||
7466 | if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) || | |
7467 | !(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL) || | |
7468 | !ocfs2_supports_inline_data(osb)) { | |
7469 | ocfs2_error(inode->i_sb, | |
7470 | "Inline data flags for inode %llu don't agree! " | |
7471 | "Disk: 0x%x, Memory: 0x%x, Superblock: 0x%x\n", | |
7472 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
7473 | le16_to_cpu(di->i_dyn_features), | |
7474 | OCFS2_I(inode)->ip_dyn_features, | |
7475 | osb->s_feature_incompat); | |
7476 | ret = -EROFS; | |
7477 | goto out; | |
7478 | } | |
7479 | ||
7480 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
7481 | if (IS_ERR(handle)) { | |
7482 | ret = PTR_ERR(handle); | |
7483 | mlog_errno(ret); | |
7484 | goto out; | |
7485 | } | |
7486 | ||
0cf2f763 | 7487 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh, |
13723d00 | 7488 | OCFS2_JOURNAL_ACCESS_WRITE); |
1afc32b9 MF |
7489 | if (ret) { |
7490 | mlog_errno(ret); | |
7491 | goto out_commit; | |
7492 | } | |
7493 | ||
7494 | numbytes = end - start; | |
7495 | memset(idata->id_data + start, 0, numbytes); | |
7496 | ||
7497 | /* | |
7498 | * No need to worry about the data page here - it's been | |
7499 | * truncated already and inline data doesn't need it for | |
7500 | * pushing zero's to disk, so we'll let readpage pick it up | |
7501 | * later. | |
7502 | */ | |
7503 | if (trunc) { | |
7504 | i_size_write(inode, start); | |
7505 | di->i_size = cpu_to_le64(start); | |
7506 | } | |
7507 | ||
7508 | inode->i_blocks = ocfs2_inode_sector_count(inode); | |
7509 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; | |
7510 | ||
7511 | di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec); | |
7512 | di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); | |
7513 | ||
7514 | ocfs2_journal_dirty(handle, di_bh); | |
7515 | ||
7516 | out_commit: | |
7517 | ocfs2_commit_trans(osb, handle); | |
7518 | ||
7519 | out: | |
7520 | return ret; | |
7521 | } | |
7522 | ||
ccd979bd MF |
7523 | static void ocfs2_free_truncate_context(struct ocfs2_truncate_context *tc) |
7524 | { | |
59a5e416 MF |
7525 | /* |
7526 | * The caller is responsible for completing deallocation | |
7527 | * before freeing the context. | |
7528 | */ | |
7529 | if (tc->tc_dealloc.c_first_suballocator != NULL) | |
7530 | mlog(ML_NOTICE, | |
7531 | "Truncate completion has non-empty dealloc context\n"); | |
ccd979bd | 7532 | |
a81cb88b | 7533 | brelse(tc->tc_last_eb_bh); |
ccd979bd MF |
7534 | |
7535 | kfree(tc); | |
7536 | } |