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" |
bcbbb24a | 52 | #include "refcounttree.h" |
ccd979bd MF |
53 | |
54 | #include "buffer_head_io.h" | |
55 | ||
853a3a14 TM |
56 | enum ocfs2_contig_type { |
57 | CONTIG_NONE = 0, | |
58 | CONTIG_LEFT, | |
59 | CONTIG_RIGHT, | |
60 | CONTIG_LEFTRIGHT, | |
61 | }; | |
e7d4cb6b | 62 | |
853a3a14 TM |
63 | static enum ocfs2_contig_type |
64 | ocfs2_extent_rec_contig(struct super_block *sb, | |
65 | struct ocfs2_extent_rec *ext, | |
66 | struct ocfs2_extent_rec *insert_rec); | |
1625f8ac JB |
67 | /* |
68 | * Operations for a specific extent tree type. | |
69 | * | |
70 | * To implement an on-disk btree (extent tree) type in ocfs2, add | |
71 | * an ocfs2_extent_tree_operations structure and the matching | |
8d6220d6 | 72 | * ocfs2_init_<thingy>_extent_tree() function. That's pretty much it |
1625f8ac JB |
73 | * for the allocation portion of the extent tree. |
74 | */ | |
e7d4cb6b | 75 | struct ocfs2_extent_tree_operations { |
1625f8ac JB |
76 | /* |
77 | * last_eb_blk is the block number of the right most leaf extent | |
78 | * block. Most on-disk structures containing an extent tree store | |
79 | * this value for fast access. The ->eo_set_last_eb_blk() and | |
80 | * ->eo_get_last_eb_blk() operations access this value. They are | |
81 | * both required. | |
82 | */ | |
35dc0aa3 JB |
83 | void (*eo_set_last_eb_blk)(struct ocfs2_extent_tree *et, |
84 | u64 blkno); | |
85 | u64 (*eo_get_last_eb_blk)(struct ocfs2_extent_tree *et); | |
1625f8ac JB |
86 | |
87 | /* | |
88 | * The on-disk structure usually keeps track of how many total | |
89 | * clusters are stored in this extent tree. This function updates | |
90 | * that value. new_clusters is the delta, and must be | |
91 | * added to the total. Required. | |
92 | */ | |
6136ca5f | 93 | void (*eo_update_clusters)(struct ocfs2_extent_tree *et, |
35dc0aa3 | 94 | u32 new_clusters); |
1625f8ac | 95 | |
92ba470c JB |
96 | /* |
97 | * If this extent tree is supported by an extent map, insert | |
98 | * a record into the map. | |
99 | */ | |
100 | void (*eo_extent_map_insert)(struct ocfs2_extent_tree *et, | |
101 | struct ocfs2_extent_rec *rec); | |
102 | ||
4c911eef JB |
103 | /* |
104 | * If this extent tree is supported by an extent map, truncate the | |
105 | * map to clusters, | |
106 | */ | |
107 | void (*eo_extent_map_truncate)(struct ocfs2_extent_tree *et, | |
108 | u32 clusters); | |
109 | ||
1625f8ac JB |
110 | /* |
111 | * If ->eo_insert_check() exists, it is called before rec is | |
112 | * inserted into the extent tree. It is optional. | |
113 | */ | |
6136ca5f | 114 | int (*eo_insert_check)(struct ocfs2_extent_tree *et, |
1e61ee79 | 115 | struct ocfs2_extent_rec *rec); |
6136ca5f | 116 | int (*eo_sanity_check)(struct ocfs2_extent_tree *et); |
0ce1010f | 117 | |
1625f8ac JB |
118 | /* |
119 | * -------------------------------------------------------------- | |
120 | * The remaining are internal to ocfs2_extent_tree and don't have | |
121 | * accessor functions | |
122 | */ | |
123 | ||
124 | /* | |
125 | * ->eo_fill_root_el() takes et->et_object and sets et->et_root_el. | |
126 | * It is required. | |
127 | */ | |
0ce1010f | 128 | void (*eo_fill_root_el)(struct ocfs2_extent_tree *et); |
1625f8ac JB |
129 | |
130 | /* | |
131 | * ->eo_fill_max_leaf_clusters sets et->et_max_leaf_clusters if | |
132 | * it exists. If it does not, et->et_max_leaf_clusters is set | |
133 | * to 0 (unlimited). Optional. | |
134 | */ | |
6136ca5f | 135 | void (*eo_fill_max_leaf_clusters)(struct ocfs2_extent_tree *et); |
853a3a14 TM |
136 | |
137 | /* | |
138 | * ->eo_extent_contig test whether the 2 ocfs2_extent_rec | |
139 | * are contiguous or not. Optional. Don't need to set it if use | |
140 | * ocfs2_extent_rec as the tree leaf. | |
141 | */ | |
142 | enum ocfs2_contig_type | |
143 | (*eo_extent_contig)(struct ocfs2_extent_tree *et, | |
144 | struct ocfs2_extent_rec *ext, | |
145 | struct ocfs2_extent_rec *insert_rec); | |
e7d4cb6b TM |
146 | }; |
147 | ||
e7d4cb6b | 148 | |
f99b9b7c JB |
149 | /* |
150 | * Pre-declare ocfs2_dinode_et_ops so we can use it as a sanity check | |
151 | * in the methods. | |
152 | */ | |
153 | static u64 ocfs2_dinode_get_last_eb_blk(struct ocfs2_extent_tree *et); | |
154 | static void ocfs2_dinode_set_last_eb_blk(struct ocfs2_extent_tree *et, | |
155 | u64 blkno); | |
6136ca5f | 156 | static void ocfs2_dinode_update_clusters(struct ocfs2_extent_tree *et, |
f99b9b7c | 157 | u32 clusters); |
92ba470c JB |
158 | static void ocfs2_dinode_extent_map_insert(struct ocfs2_extent_tree *et, |
159 | struct ocfs2_extent_rec *rec); | |
4c911eef JB |
160 | static void ocfs2_dinode_extent_map_truncate(struct ocfs2_extent_tree *et, |
161 | u32 clusters); | |
6136ca5f | 162 | static int ocfs2_dinode_insert_check(struct ocfs2_extent_tree *et, |
f99b9b7c | 163 | struct ocfs2_extent_rec *rec); |
6136ca5f | 164 | static int ocfs2_dinode_sanity_check(struct ocfs2_extent_tree *et); |
f99b9b7c JB |
165 | static void ocfs2_dinode_fill_root_el(struct ocfs2_extent_tree *et); |
166 | static struct ocfs2_extent_tree_operations ocfs2_dinode_et_ops = { | |
167 | .eo_set_last_eb_blk = ocfs2_dinode_set_last_eb_blk, | |
168 | .eo_get_last_eb_blk = ocfs2_dinode_get_last_eb_blk, | |
169 | .eo_update_clusters = ocfs2_dinode_update_clusters, | |
92ba470c | 170 | .eo_extent_map_insert = ocfs2_dinode_extent_map_insert, |
4c911eef | 171 | .eo_extent_map_truncate = ocfs2_dinode_extent_map_truncate, |
f99b9b7c JB |
172 | .eo_insert_check = ocfs2_dinode_insert_check, |
173 | .eo_sanity_check = ocfs2_dinode_sanity_check, | |
174 | .eo_fill_root_el = ocfs2_dinode_fill_root_el, | |
175 | }; | |
0ce1010f | 176 | |
e7d4cb6b TM |
177 | static void ocfs2_dinode_set_last_eb_blk(struct ocfs2_extent_tree *et, |
178 | u64 blkno) | |
179 | { | |
ea5efa15 | 180 | struct ocfs2_dinode *di = et->et_object; |
e7d4cb6b | 181 | |
f99b9b7c | 182 | BUG_ON(et->et_ops != &ocfs2_dinode_et_ops); |
e7d4cb6b TM |
183 | di->i_last_eb_blk = cpu_to_le64(blkno); |
184 | } | |
185 | ||
186 | static u64 ocfs2_dinode_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
187 | { | |
ea5efa15 | 188 | struct ocfs2_dinode *di = et->et_object; |
e7d4cb6b | 189 | |
f99b9b7c | 190 | BUG_ON(et->et_ops != &ocfs2_dinode_et_ops); |
e7d4cb6b TM |
191 | return le64_to_cpu(di->i_last_eb_blk); |
192 | } | |
193 | ||
6136ca5f | 194 | static void ocfs2_dinode_update_clusters(struct ocfs2_extent_tree *et, |
e7d4cb6b TM |
195 | u32 clusters) |
196 | { | |
6136ca5f | 197 | struct ocfs2_inode_info *oi = cache_info_to_inode(et->et_ci); |
ea5efa15 | 198 | struct ocfs2_dinode *di = et->et_object; |
e7d4cb6b TM |
199 | |
200 | le32_add_cpu(&di->i_clusters, clusters); | |
6136ca5f JB |
201 | spin_lock(&oi->ip_lock); |
202 | oi->ip_clusters = le32_to_cpu(di->i_clusters); | |
203 | spin_unlock(&oi->ip_lock); | |
e7d4cb6b TM |
204 | } |
205 | ||
92ba470c JB |
206 | static void ocfs2_dinode_extent_map_insert(struct ocfs2_extent_tree *et, |
207 | struct ocfs2_extent_rec *rec) | |
208 | { | |
209 | struct inode *inode = &cache_info_to_inode(et->et_ci)->vfs_inode; | |
210 | ||
211 | ocfs2_extent_map_insert_rec(inode, rec); | |
212 | } | |
213 | ||
4c911eef JB |
214 | static void ocfs2_dinode_extent_map_truncate(struct ocfs2_extent_tree *et, |
215 | u32 clusters) | |
216 | { | |
217 | struct inode *inode = &cache_info_to_inode(et->et_ci)->vfs_inode; | |
218 | ||
219 | ocfs2_extent_map_trunc(inode, clusters); | |
220 | } | |
221 | ||
6136ca5f | 222 | static int ocfs2_dinode_insert_check(struct ocfs2_extent_tree *et, |
1e61ee79 JB |
223 | struct ocfs2_extent_rec *rec) |
224 | { | |
6136ca5f JB |
225 | struct ocfs2_inode_info *oi = cache_info_to_inode(et->et_ci); |
226 | struct ocfs2_super *osb = OCFS2_SB(oi->vfs_inode.i_sb); | |
1e61ee79 | 227 | |
6136ca5f | 228 | BUG_ON(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL); |
1e61ee79 | 229 | mlog_bug_on_msg(!ocfs2_sparse_alloc(osb) && |
6136ca5f | 230 | (oi->ip_clusters != le32_to_cpu(rec->e_cpos)), |
1e61ee79 JB |
231 | "Device %s, asking for sparse allocation: inode %llu, " |
232 | "cpos %u, clusters %u\n", | |
233 | osb->dev_str, | |
6136ca5f JB |
234 | (unsigned long long)oi->ip_blkno, |
235 | rec->e_cpos, oi->ip_clusters); | |
1e61ee79 JB |
236 | |
237 | return 0; | |
238 | } | |
239 | ||
6136ca5f | 240 | static int ocfs2_dinode_sanity_check(struct ocfs2_extent_tree *et) |
e7d4cb6b | 241 | { |
10995aa2 | 242 | struct ocfs2_dinode *di = et->et_object; |
e7d4cb6b | 243 | |
f99b9b7c | 244 | BUG_ON(et->et_ops != &ocfs2_dinode_et_ops); |
10995aa2 | 245 | BUG_ON(!OCFS2_IS_VALID_DINODE(di)); |
e7d4cb6b | 246 | |
10995aa2 | 247 | return 0; |
e7d4cb6b TM |
248 | } |
249 | ||
f99b9b7c JB |
250 | static void ocfs2_dinode_fill_root_el(struct ocfs2_extent_tree *et) |
251 | { | |
252 | struct ocfs2_dinode *di = et->et_object; | |
253 | ||
254 | et->et_root_el = &di->id2.i_list; | |
255 | } | |
256 | ||
e7d4cb6b | 257 | |
0ce1010f JB |
258 | static void ocfs2_xattr_value_fill_root_el(struct ocfs2_extent_tree *et) |
259 | { | |
2a50a743 | 260 | struct ocfs2_xattr_value_buf *vb = et->et_object; |
0ce1010f | 261 | |
2a50a743 | 262 | et->et_root_el = &vb->vb_xv->xr_list; |
0ce1010f JB |
263 | } |
264 | ||
f56654c4 TM |
265 | static void ocfs2_xattr_value_set_last_eb_blk(struct ocfs2_extent_tree *et, |
266 | u64 blkno) | |
267 | { | |
2a50a743 | 268 | struct ocfs2_xattr_value_buf *vb = et->et_object; |
f56654c4 | 269 | |
2a50a743 | 270 | vb->vb_xv->xr_last_eb_blk = cpu_to_le64(blkno); |
f56654c4 TM |
271 | } |
272 | ||
273 | static u64 ocfs2_xattr_value_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
274 | { | |
2a50a743 | 275 | struct ocfs2_xattr_value_buf *vb = et->et_object; |
f56654c4 | 276 | |
2a50a743 | 277 | return le64_to_cpu(vb->vb_xv->xr_last_eb_blk); |
f56654c4 TM |
278 | } |
279 | ||
6136ca5f | 280 | static void ocfs2_xattr_value_update_clusters(struct ocfs2_extent_tree *et, |
f56654c4 TM |
281 | u32 clusters) |
282 | { | |
2a50a743 | 283 | struct ocfs2_xattr_value_buf *vb = et->et_object; |
f56654c4 | 284 | |
2a50a743 | 285 | le32_add_cpu(&vb->vb_xv->xr_clusters, clusters); |
f56654c4 TM |
286 | } |
287 | ||
1a09f556 | 288 | static struct ocfs2_extent_tree_operations ocfs2_xattr_value_et_ops = { |
35dc0aa3 JB |
289 | .eo_set_last_eb_blk = ocfs2_xattr_value_set_last_eb_blk, |
290 | .eo_get_last_eb_blk = ocfs2_xattr_value_get_last_eb_blk, | |
291 | .eo_update_clusters = ocfs2_xattr_value_update_clusters, | |
0ce1010f | 292 | .eo_fill_root_el = ocfs2_xattr_value_fill_root_el, |
f56654c4 TM |
293 | }; |
294 | ||
0ce1010f JB |
295 | static void ocfs2_xattr_tree_fill_root_el(struct ocfs2_extent_tree *et) |
296 | { | |
297 | struct ocfs2_xattr_block *xb = et->et_object; | |
298 | ||
299 | et->et_root_el = &xb->xb_attrs.xb_root.xt_list; | |
300 | } | |
301 | ||
6136ca5f | 302 | static void ocfs2_xattr_tree_fill_max_leaf_clusters(struct ocfs2_extent_tree *et) |
943cced3 | 303 | { |
6136ca5f | 304 | struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci); |
943cced3 | 305 | et->et_max_leaf_clusters = |
6136ca5f | 306 | ocfs2_clusters_for_bytes(sb, OCFS2_MAX_XATTR_TREE_LEAF_SIZE); |
943cced3 JB |
307 | } |
308 | ||
ba492615 TM |
309 | static void ocfs2_xattr_tree_set_last_eb_blk(struct ocfs2_extent_tree *et, |
310 | u64 blkno) | |
311 | { | |
ea5efa15 | 312 | struct ocfs2_xattr_block *xb = et->et_object; |
ba492615 TM |
313 | struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root; |
314 | ||
315 | xt->xt_last_eb_blk = cpu_to_le64(blkno); | |
316 | } | |
317 | ||
318 | static u64 ocfs2_xattr_tree_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
319 | { | |
ea5efa15 | 320 | struct ocfs2_xattr_block *xb = et->et_object; |
ba492615 TM |
321 | struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root; |
322 | ||
323 | return le64_to_cpu(xt->xt_last_eb_blk); | |
324 | } | |
325 | ||
6136ca5f | 326 | static void ocfs2_xattr_tree_update_clusters(struct ocfs2_extent_tree *et, |
ba492615 TM |
327 | u32 clusters) |
328 | { | |
ea5efa15 | 329 | struct ocfs2_xattr_block *xb = et->et_object; |
ba492615 TM |
330 | |
331 | le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, clusters); | |
332 | } | |
333 | ||
ba492615 | 334 | static struct ocfs2_extent_tree_operations ocfs2_xattr_tree_et_ops = { |
35dc0aa3 JB |
335 | .eo_set_last_eb_blk = ocfs2_xattr_tree_set_last_eb_blk, |
336 | .eo_get_last_eb_blk = ocfs2_xattr_tree_get_last_eb_blk, | |
337 | .eo_update_clusters = ocfs2_xattr_tree_update_clusters, | |
0ce1010f | 338 | .eo_fill_root_el = ocfs2_xattr_tree_fill_root_el, |
943cced3 | 339 | .eo_fill_max_leaf_clusters = ocfs2_xattr_tree_fill_max_leaf_clusters, |
ba492615 TM |
340 | }; |
341 | ||
9b7895ef MF |
342 | static void ocfs2_dx_root_set_last_eb_blk(struct ocfs2_extent_tree *et, |
343 | u64 blkno) | |
344 | { | |
345 | struct ocfs2_dx_root_block *dx_root = et->et_object; | |
346 | ||
347 | dx_root->dr_last_eb_blk = cpu_to_le64(blkno); | |
348 | } | |
349 | ||
350 | static u64 ocfs2_dx_root_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
351 | { | |
352 | struct ocfs2_dx_root_block *dx_root = et->et_object; | |
353 | ||
354 | return le64_to_cpu(dx_root->dr_last_eb_blk); | |
355 | } | |
356 | ||
6136ca5f | 357 | static void ocfs2_dx_root_update_clusters(struct ocfs2_extent_tree *et, |
9b7895ef MF |
358 | u32 clusters) |
359 | { | |
360 | struct ocfs2_dx_root_block *dx_root = et->et_object; | |
361 | ||
362 | le32_add_cpu(&dx_root->dr_clusters, clusters); | |
363 | } | |
364 | ||
6136ca5f | 365 | static int ocfs2_dx_root_sanity_check(struct ocfs2_extent_tree *et) |
9b7895ef MF |
366 | { |
367 | struct ocfs2_dx_root_block *dx_root = et->et_object; | |
368 | ||
369 | BUG_ON(!OCFS2_IS_VALID_DX_ROOT(dx_root)); | |
370 | ||
371 | return 0; | |
372 | } | |
373 | ||
374 | static void ocfs2_dx_root_fill_root_el(struct ocfs2_extent_tree *et) | |
375 | { | |
376 | struct ocfs2_dx_root_block *dx_root = et->et_object; | |
377 | ||
378 | et->et_root_el = &dx_root->dr_list; | |
379 | } | |
380 | ||
381 | static struct ocfs2_extent_tree_operations ocfs2_dx_root_et_ops = { | |
382 | .eo_set_last_eb_blk = ocfs2_dx_root_set_last_eb_blk, | |
383 | .eo_get_last_eb_blk = ocfs2_dx_root_get_last_eb_blk, | |
384 | .eo_update_clusters = ocfs2_dx_root_update_clusters, | |
385 | .eo_sanity_check = ocfs2_dx_root_sanity_check, | |
386 | .eo_fill_root_el = ocfs2_dx_root_fill_root_el, | |
387 | }; | |
388 | ||
fe924415 TM |
389 | static void ocfs2_refcount_tree_fill_root_el(struct ocfs2_extent_tree *et) |
390 | { | |
391 | struct ocfs2_refcount_block *rb = et->et_object; | |
392 | ||
393 | et->et_root_el = &rb->rf_list; | |
394 | } | |
395 | ||
396 | static void ocfs2_refcount_tree_set_last_eb_blk(struct ocfs2_extent_tree *et, | |
397 | u64 blkno) | |
398 | { | |
399 | struct ocfs2_refcount_block *rb = et->et_object; | |
400 | ||
401 | rb->rf_last_eb_blk = cpu_to_le64(blkno); | |
402 | } | |
403 | ||
404 | static u64 ocfs2_refcount_tree_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
405 | { | |
406 | struct ocfs2_refcount_block *rb = et->et_object; | |
407 | ||
408 | return le64_to_cpu(rb->rf_last_eb_blk); | |
409 | } | |
410 | ||
411 | static void ocfs2_refcount_tree_update_clusters(struct ocfs2_extent_tree *et, | |
412 | u32 clusters) | |
413 | { | |
414 | struct ocfs2_refcount_block *rb = et->et_object; | |
415 | ||
416 | le32_add_cpu(&rb->rf_clusters, clusters); | |
417 | } | |
418 | ||
419 | static enum ocfs2_contig_type | |
420 | ocfs2_refcount_tree_extent_contig(struct ocfs2_extent_tree *et, | |
421 | struct ocfs2_extent_rec *ext, | |
422 | struct ocfs2_extent_rec *insert_rec) | |
423 | { | |
424 | return CONTIG_NONE; | |
425 | } | |
426 | ||
427 | static struct ocfs2_extent_tree_operations ocfs2_refcount_tree_et_ops = { | |
428 | .eo_set_last_eb_blk = ocfs2_refcount_tree_set_last_eb_blk, | |
429 | .eo_get_last_eb_blk = ocfs2_refcount_tree_get_last_eb_blk, | |
430 | .eo_update_clusters = ocfs2_refcount_tree_update_clusters, | |
431 | .eo_fill_root_el = ocfs2_refcount_tree_fill_root_el, | |
432 | .eo_extent_contig = ocfs2_refcount_tree_extent_contig, | |
433 | }; | |
434 | ||
8d6220d6 | 435 | static void __ocfs2_init_extent_tree(struct ocfs2_extent_tree *et, |
5e404e9e | 436 | struct ocfs2_caching_info *ci, |
8d6220d6 | 437 | struct buffer_head *bh, |
13723d00 | 438 | ocfs2_journal_access_func access, |
8d6220d6 JB |
439 | void *obj, |
440 | struct ocfs2_extent_tree_operations *ops) | |
e7d4cb6b | 441 | { |
1a09f556 | 442 | et->et_ops = ops; |
ce1d9ea6 | 443 | et->et_root_bh = bh; |
5e404e9e | 444 | et->et_ci = ci; |
13723d00 | 445 | et->et_root_journal_access = access; |
ea5efa15 JB |
446 | if (!obj) |
447 | obj = (void *)bh->b_data; | |
448 | et->et_object = obj; | |
e7d4cb6b | 449 | |
0ce1010f | 450 | et->et_ops->eo_fill_root_el(et); |
943cced3 JB |
451 | if (!et->et_ops->eo_fill_max_leaf_clusters) |
452 | et->et_max_leaf_clusters = 0; | |
453 | else | |
6136ca5f | 454 | et->et_ops->eo_fill_max_leaf_clusters(et); |
e7d4cb6b TM |
455 | } |
456 | ||
8d6220d6 | 457 | void ocfs2_init_dinode_extent_tree(struct ocfs2_extent_tree *et, |
5e404e9e | 458 | struct ocfs2_caching_info *ci, |
8d6220d6 | 459 | struct buffer_head *bh) |
1a09f556 | 460 | { |
5e404e9e | 461 | __ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_di, |
13723d00 | 462 | NULL, &ocfs2_dinode_et_ops); |
1a09f556 JB |
463 | } |
464 | ||
8d6220d6 | 465 | void ocfs2_init_xattr_tree_extent_tree(struct ocfs2_extent_tree *et, |
5e404e9e | 466 | struct ocfs2_caching_info *ci, |
8d6220d6 | 467 | struct buffer_head *bh) |
1a09f556 | 468 | { |
5e404e9e | 469 | __ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_xb, |
13723d00 | 470 | NULL, &ocfs2_xattr_tree_et_ops); |
1a09f556 JB |
471 | } |
472 | ||
8d6220d6 | 473 | void ocfs2_init_xattr_value_extent_tree(struct ocfs2_extent_tree *et, |
5e404e9e | 474 | struct ocfs2_caching_info *ci, |
2a50a743 | 475 | struct ocfs2_xattr_value_buf *vb) |
e7d4cb6b | 476 | { |
5e404e9e | 477 | __ocfs2_init_extent_tree(et, ci, vb->vb_bh, vb->vb_access, vb, |
8d6220d6 | 478 | &ocfs2_xattr_value_et_ops); |
e7d4cb6b TM |
479 | } |
480 | ||
9b7895ef | 481 | void ocfs2_init_dx_root_extent_tree(struct ocfs2_extent_tree *et, |
5e404e9e | 482 | struct ocfs2_caching_info *ci, |
9b7895ef MF |
483 | struct buffer_head *bh) |
484 | { | |
5e404e9e | 485 | __ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_dr, |
9b7895ef MF |
486 | NULL, &ocfs2_dx_root_et_ops); |
487 | } | |
488 | ||
fe924415 TM |
489 | void ocfs2_init_refcount_extent_tree(struct ocfs2_extent_tree *et, |
490 | struct ocfs2_caching_info *ci, | |
491 | struct buffer_head *bh) | |
492 | { | |
493 | __ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_rb, | |
494 | NULL, &ocfs2_refcount_tree_et_ops); | |
495 | } | |
496 | ||
35dc0aa3 JB |
497 | static inline void ocfs2_et_set_last_eb_blk(struct ocfs2_extent_tree *et, |
498 | u64 new_last_eb_blk) | |
e7d4cb6b | 499 | { |
ce1d9ea6 | 500 | et->et_ops->eo_set_last_eb_blk(et, new_last_eb_blk); |
e7d4cb6b TM |
501 | } |
502 | ||
35dc0aa3 | 503 | static inline u64 ocfs2_et_get_last_eb_blk(struct ocfs2_extent_tree *et) |
e7d4cb6b | 504 | { |
ce1d9ea6 | 505 | return et->et_ops->eo_get_last_eb_blk(et); |
e7d4cb6b TM |
506 | } |
507 | ||
6136ca5f | 508 | static inline void ocfs2_et_update_clusters(struct ocfs2_extent_tree *et, |
35dc0aa3 JB |
509 | u32 clusters) |
510 | { | |
6136ca5f | 511 | et->et_ops->eo_update_clusters(et, clusters); |
35dc0aa3 JB |
512 | } |
513 | ||
92ba470c JB |
514 | static inline void ocfs2_et_extent_map_insert(struct ocfs2_extent_tree *et, |
515 | struct ocfs2_extent_rec *rec) | |
516 | { | |
517 | if (et->et_ops->eo_extent_map_insert) | |
518 | et->et_ops->eo_extent_map_insert(et, rec); | |
519 | } | |
520 | ||
4c911eef JB |
521 | static inline void ocfs2_et_extent_map_truncate(struct ocfs2_extent_tree *et, |
522 | u32 clusters) | |
523 | { | |
524 | if (et->et_ops->eo_extent_map_truncate) | |
525 | et->et_ops->eo_extent_map_truncate(et, clusters); | |
526 | } | |
527 | ||
13723d00 | 528 | static inline int ocfs2_et_root_journal_access(handle_t *handle, |
13723d00 JB |
529 | struct ocfs2_extent_tree *et, |
530 | int type) | |
531 | { | |
d9a0a1f8 | 532 | return et->et_root_journal_access(handle, et->et_ci, et->et_root_bh, |
13723d00 JB |
533 | type); |
534 | } | |
535 | ||
853a3a14 TM |
536 | static inline enum ocfs2_contig_type |
537 | ocfs2_et_extent_contig(struct ocfs2_extent_tree *et, | |
538 | struct ocfs2_extent_rec *rec, | |
539 | struct ocfs2_extent_rec *insert_rec) | |
540 | { | |
541 | if (et->et_ops->eo_extent_contig) | |
542 | return et->et_ops->eo_extent_contig(et, rec, insert_rec); | |
543 | ||
544 | return ocfs2_extent_rec_contig( | |
545 | ocfs2_metadata_cache_get_super(et->et_ci), | |
546 | rec, insert_rec); | |
547 | } | |
548 | ||
6136ca5f | 549 | static inline int ocfs2_et_insert_check(struct ocfs2_extent_tree *et, |
1e61ee79 JB |
550 | struct ocfs2_extent_rec *rec) |
551 | { | |
552 | int ret = 0; | |
553 | ||
554 | if (et->et_ops->eo_insert_check) | |
6136ca5f | 555 | ret = et->et_ops->eo_insert_check(et, rec); |
1e61ee79 JB |
556 | return ret; |
557 | } | |
558 | ||
6136ca5f | 559 | static inline int ocfs2_et_sanity_check(struct ocfs2_extent_tree *et) |
e7d4cb6b | 560 | { |
1e61ee79 JB |
561 | int ret = 0; |
562 | ||
563 | if (et->et_ops->eo_sanity_check) | |
6136ca5f | 564 | ret = et->et_ops->eo_sanity_check(et); |
1e61ee79 | 565 | return ret; |
e7d4cb6b TM |
566 | } |
567 | ||
dcd0538f | 568 | static void ocfs2_free_truncate_context(struct ocfs2_truncate_context *tc); |
59a5e416 MF |
569 | static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt, |
570 | struct ocfs2_extent_block *eb); | |
d401dc12 JB |
571 | static void ocfs2_adjust_rightmost_records(handle_t *handle, |
572 | struct ocfs2_extent_tree *et, | |
6b791bcc TM |
573 | struct ocfs2_path *path, |
574 | struct ocfs2_extent_rec *insert_rec); | |
dcd0538f MF |
575 | /* |
576 | * Reset the actual path elements so that we can re-use the structure | |
577 | * to build another path. Generally, this involves freeing the buffer | |
578 | * heads. | |
579 | */ | |
e2e9f608 | 580 | void ocfs2_reinit_path(struct ocfs2_path *path, int keep_root) |
dcd0538f MF |
581 | { |
582 | int i, start = 0, depth = 0; | |
583 | struct ocfs2_path_item *node; | |
ccd979bd | 584 | |
dcd0538f MF |
585 | if (keep_root) |
586 | start = 1; | |
ccd979bd | 587 | |
dcd0538f MF |
588 | for(i = start; i < path_num_items(path); i++) { |
589 | node = &path->p_node[i]; | |
590 | ||
591 | brelse(node->bh); | |
592 | node->bh = NULL; | |
593 | node->el = NULL; | |
594 | } | |
595 | ||
596 | /* | |
597 | * Tree depth may change during truncate, or insert. If we're | |
598 | * keeping the root extent list, then make sure that our path | |
599 | * structure reflects the proper depth. | |
600 | */ | |
601 | if (keep_root) | |
602 | depth = le16_to_cpu(path_root_el(path)->l_tree_depth); | |
13723d00 JB |
603 | else |
604 | path_root_access(path) = NULL; | |
dcd0538f MF |
605 | |
606 | path->p_tree_depth = depth; | |
607 | } | |
608 | ||
e2e9f608 | 609 | void ocfs2_free_path(struct ocfs2_path *path) |
dcd0538f MF |
610 | { |
611 | if (path) { | |
612 | ocfs2_reinit_path(path, 0); | |
613 | kfree(path); | |
614 | } | |
615 | } | |
616 | ||
328d5752 MF |
617 | /* |
618 | * All the elements of src into dest. After this call, src could be freed | |
619 | * without affecting dest. | |
620 | * | |
621 | * Both paths should have the same root. Any non-root elements of dest | |
622 | * will be freed. | |
623 | */ | |
624 | static void ocfs2_cp_path(struct ocfs2_path *dest, struct ocfs2_path *src) | |
625 | { | |
626 | int i; | |
627 | ||
628 | BUG_ON(path_root_bh(dest) != path_root_bh(src)); | |
629 | BUG_ON(path_root_el(dest) != path_root_el(src)); | |
13723d00 | 630 | BUG_ON(path_root_access(dest) != path_root_access(src)); |
328d5752 MF |
631 | |
632 | ocfs2_reinit_path(dest, 1); | |
633 | ||
634 | for(i = 1; i < OCFS2_MAX_PATH_DEPTH; i++) { | |
635 | dest->p_node[i].bh = src->p_node[i].bh; | |
636 | dest->p_node[i].el = src->p_node[i].el; | |
637 | ||
638 | if (dest->p_node[i].bh) | |
639 | get_bh(dest->p_node[i].bh); | |
640 | } | |
641 | } | |
642 | ||
dcd0538f MF |
643 | /* |
644 | * Make the *dest path the same as src and re-initialize src path to | |
645 | * have a root only. | |
646 | */ | |
647 | static void ocfs2_mv_path(struct ocfs2_path *dest, struct ocfs2_path *src) | |
648 | { | |
649 | int i; | |
650 | ||
651 | BUG_ON(path_root_bh(dest) != path_root_bh(src)); | |
13723d00 | 652 | BUG_ON(path_root_access(dest) != path_root_access(src)); |
dcd0538f MF |
653 | |
654 | for(i = 1; i < OCFS2_MAX_PATH_DEPTH; i++) { | |
655 | brelse(dest->p_node[i].bh); | |
656 | ||
657 | dest->p_node[i].bh = src->p_node[i].bh; | |
658 | dest->p_node[i].el = src->p_node[i].el; | |
659 | ||
660 | src->p_node[i].bh = NULL; | |
661 | src->p_node[i].el = NULL; | |
662 | } | |
663 | } | |
664 | ||
665 | /* | |
666 | * Insert an extent block at given index. | |
667 | * | |
668 | * This will not take an additional reference on eb_bh. | |
669 | */ | |
670 | static inline void ocfs2_path_insert_eb(struct ocfs2_path *path, int index, | |
671 | struct buffer_head *eb_bh) | |
672 | { | |
673 | struct ocfs2_extent_block *eb = (struct ocfs2_extent_block *)eb_bh->b_data; | |
674 | ||
675 | /* | |
676 | * Right now, no root bh is an extent block, so this helps | |
677 | * catch code errors with dinode trees. The assertion can be | |
678 | * safely removed if we ever need to insert extent block | |
679 | * structures at the root. | |
680 | */ | |
681 | BUG_ON(index == 0); | |
682 | ||
683 | path->p_node[index].bh = eb_bh; | |
684 | path->p_node[index].el = &eb->h_list; | |
685 | } | |
686 | ||
687 | static struct ocfs2_path *ocfs2_new_path(struct buffer_head *root_bh, | |
13723d00 JB |
688 | struct ocfs2_extent_list *root_el, |
689 | ocfs2_journal_access_func access) | |
dcd0538f MF |
690 | { |
691 | struct ocfs2_path *path; | |
ccd979bd | 692 | |
dcd0538f MF |
693 | BUG_ON(le16_to_cpu(root_el->l_tree_depth) >= OCFS2_MAX_PATH_DEPTH); |
694 | ||
695 | path = kzalloc(sizeof(*path), GFP_NOFS); | |
696 | if (path) { | |
697 | path->p_tree_depth = le16_to_cpu(root_el->l_tree_depth); | |
698 | get_bh(root_bh); | |
699 | path_root_bh(path) = root_bh; | |
700 | path_root_el(path) = root_el; | |
13723d00 | 701 | path_root_access(path) = access; |
dcd0538f MF |
702 | } |
703 | ||
704 | return path; | |
705 | } | |
706 | ||
e2e9f608 | 707 | struct ocfs2_path *ocfs2_new_path_from_path(struct ocfs2_path *path) |
ffdd7a54 | 708 | { |
13723d00 JB |
709 | return ocfs2_new_path(path_root_bh(path), path_root_el(path), |
710 | path_root_access(path)); | |
ffdd7a54 JB |
711 | } |
712 | ||
e2e9f608 | 713 | struct ocfs2_path *ocfs2_new_path_from_et(struct ocfs2_extent_tree *et) |
ffdd7a54 | 714 | { |
13723d00 JB |
715 | return ocfs2_new_path(et->et_root_bh, et->et_root_el, |
716 | et->et_root_journal_access); | |
717 | } | |
718 | ||
719 | /* | |
720 | * Journal the buffer at depth idx. All idx>0 are extent_blocks, | |
721 | * otherwise it's the root_access function. | |
722 | * | |
723 | * I don't like the way this function's name looks next to | |
724 | * ocfs2_journal_access_path(), but I don't have a better one. | |
725 | */ | |
e2e9f608 TM |
726 | int ocfs2_path_bh_journal_access(handle_t *handle, |
727 | struct ocfs2_caching_info *ci, | |
728 | struct ocfs2_path *path, | |
729 | int idx) | |
13723d00 JB |
730 | { |
731 | ocfs2_journal_access_func access = path_root_access(path); | |
732 | ||
733 | if (!access) | |
734 | access = ocfs2_journal_access; | |
735 | ||
736 | if (idx) | |
737 | access = ocfs2_journal_access_eb; | |
738 | ||
0cf2f763 | 739 | return access(handle, ci, path->p_node[idx].bh, |
13723d00 | 740 | OCFS2_JOURNAL_ACCESS_WRITE); |
ffdd7a54 JB |
741 | } |
742 | ||
dcd0538f MF |
743 | /* |
744 | * Convenience function to journal all components in a path. | |
745 | */ | |
e2e9f608 TM |
746 | int ocfs2_journal_access_path(struct ocfs2_caching_info *ci, |
747 | handle_t *handle, | |
748 | struct ocfs2_path *path) | |
dcd0538f MF |
749 | { |
750 | int i, ret = 0; | |
751 | ||
752 | if (!path) | |
753 | goto out; | |
754 | ||
755 | for(i = 0; i < path_num_items(path); i++) { | |
0cf2f763 | 756 | ret = ocfs2_path_bh_journal_access(handle, ci, path, i); |
dcd0538f MF |
757 | if (ret < 0) { |
758 | mlog_errno(ret); | |
759 | goto out; | |
760 | } | |
761 | } | |
762 | ||
763 | out: | |
764 | return ret; | |
765 | } | |
766 | ||
328d5752 MF |
767 | /* |
768 | * Return the index of the extent record which contains cluster #v_cluster. | |
769 | * -1 is returned if it was not found. | |
770 | * | |
771 | * Should work fine on interior and exterior nodes. | |
772 | */ | |
773 | int ocfs2_search_extent_list(struct ocfs2_extent_list *el, u32 v_cluster) | |
774 | { | |
775 | int ret = -1; | |
776 | int i; | |
777 | struct ocfs2_extent_rec *rec; | |
778 | u32 rec_end, rec_start, clusters; | |
779 | ||
780 | for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) { | |
781 | rec = &el->l_recs[i]; | |
782 | ||
783 | rec_start = le32_to_cpu(rec->e_cpos); | |
784 | clusters = ocfs2_rec_clusters(el, rec); | |
785 | ||
786 | rec_end = rec_start + clusters; | |
787 | ||
788 | if (v_cluster >= rec_start && v_cluster < rec_end) { | |
789 | ret = i; | |
790 | break; | |
791 | } | |
792 | } | |
793 | ||
794 | return ret; | |
795 | } | |
796 | ||
e48edee2 MF |
797 | /* |
798 | * NOTE: ocfs2_block_extent_contig(), ocfs2_extents_adjacent() and | |
853a3a14 | 799 | * ocfs2_extent_rec_contig only work properly against leaf nodes! |
e48edee2 | 800 | */ |
dcd0538f MF |
801 | static int ocfs2_block_extent_contig(struct super_block *sb, |
802 | struct ocfs2_extent_rec *ext, | |
803 | u64 blkno) | |
ccd979bd | 804 | { |
e48edee2 MF |
805 | u64 blk_end = le64_to_cpu(ext->e_blkno); |
806 | ||
807 | blk_end += ocfs2_clusters_to_blocks(sb, | |
808 | le16_to_cpu(ext->e_leaf_clusters)); | |
809 | ||
810 | return blkno == blk_end; | |
ccd979bd MF |
811 | } |
812 | ||
dcd0538f MF |
813 | static int ocfs2_extents_adjacent(struct ocfs2_extent_rec *left, |
814 | struct ocfs2_extent_rec *right) | |
815 | { | |
e48edee2 MF |
816 | u32 left_range; |
817 | ||
818 | left_range = le32_to_cpu(left->e_cpos) + | |
819 | le16_to_cpu(left->e_leaf_clusters); | |
820 | ||
821 | return (left_range == le32_to_cpu(right->e_cpos)); | |
dcd0538f MF |
822 | } |
823 | ||
824 | static enum ocfs2_contig_type | |
853a3a14 TM |
825 | ocfs2_extent_rec_contig(struct super_block *sb, |
826 | struct ocfs2_extent_rec *ext, | |
827 | struct ocfs2_extent_rec *insert_rec) | |
dcd0538f MF |
828 | { |
829 | u64 blkno = le64_to_cpu(insert_rec->e_blkno); | |
830 | ||
328d5752 MF |
831 | /* |
832 | * Refuse to coalesce extent records with different flag | |
833 | * fields - we don't want to mix unwritten extents with user | |
834 | * data. | |
835 | */ | |
836 | if (ext->e_flags != insert_rec->e_flags) | |
837 | return CONTIG_NONE; | |
838 | ||
dcd0538f | 839 | if (ocfs2_extents_adjacent(ext, insert_rec) && |
b4a17651 | 840 | ocfs2_block_extent_contig(sb, ext, blkno)) |
dcd0538f MF |
841 | return CONTIG_RIGHT; |
842 | ||
843 | blkno = le64_to_cpu(ext->e_blkno); | |
844 | if (ocfs2_extents_adjacent(insert_rec, ext) && | |
b4a17651 | 845 | ocfs2_block_extent_contig(sb, insert_rec, blkno)) |
dcd0538f MF |
846 | return CONTIG_LEFT; |
847 | ||
848 | return CONTIG_NONE; | |
849 | } | |
850 | ||
851 | /* | |
852 | * NOTE: We can have pretty much any combination of contiguousness and | |
853 | * appending. | |
854 | * | |
855 | * The usefulness of APPEND_TAIL is more in that it lets us know that | |
856 | * we'll have to update the path to that leaf. | |
857 | */ | |
858 | enum ocfs2_append_type { | |
859 | APPEND_NONE = 0, | |
860 | APPEND_TAIL, | |
861 | }; | |
862 | ||
328d5752 MF |
863 | enum ocfs2_split_type { |
864 | SPLIT_NONE = 0, | |
865 | SPLIT_LEFT, | |
866 | SPLIT_RIGHT, | |
867 | }; | |
868 | ||
dcd0538f | 869 | struct ocfs2_insert_type { |
328d5752 | 870 | enum ocfs2_split_type ins_split; |
dcd0538f MF |
871 | enum ocfs2_append_type ins_appending; |
872 | enum ocfs2_contig_type ins_contig; | |
873 | int ins_contig_index; | |
dcd0538f MF |
874 | int ins_tree_depth; |
875 | }; | |
876 | ||
328d5752 MF |
877 | struct ocfs2_merge_ctxt { |
878 | enum ocfs2_contig_type c_contig_type; | |
879 | int c_has_empty_extent; | |
880 | int c_split_covers_rec; | |
328d5752 MF |
881 | }; |
882 | ||
5e96581a JB |
883 | static int ocfs2_validate_extent_block(struct super_block *sb, |
884 | struct buffer_head *bh) | |
885 | { | |
d6b32bbb | 886 | int rc; |
5e96581a JB |
887 | struct ocfs2_extent_block *eb = |
888 | (struct ocfs2_extent_block *)bh->b_data; | |
889 | ||
970e4936 JB |
890 | mlog(0, "Validating extent block %llu\n", |
891 | (unsigned long long)bh->b_blocknr); | |
892 | ||
d6b32bbb JB |
893 | BUG_ON(!buffer_uptodate(bh)); |
894 | ||
895 | /* | |
896 | * If the ecc fails, we return the error but otherwise | |
897 | * leave the filesystem running. We know any error is | |
898 | * local to this block. | |
899 | */ | |
900 | rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &eb->h_check); | |
13723d00 JB |
901 | if (rc) { |
902 | mlog(ML_ERROR, "Checksum failed for extent block %llu\n", | |
903 | (unsigned long long)bh->b_blocknr); | |
d6b32bbb | 904 | return rc; |
13723d00 | 905 | } |
d6b32bbb JB |
906 | |
907 | /* | |
908 | * Errors after here are fatal. | |
909 | */ | |
910 | ||
5e96581a JB |
911 | if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) { |
912 | ocfs2_error(sb, | |
913 | "Extent block #%llu has bad signature %.*s", | |
914 | (unsigned long long)bh->b_blocknr, 7, | |
915 | eb->h_signature); | |
916 | return -EINVAL; | |
917 | } | |
918 | ||
919 | if (le64_to_cpu(eb->h_blkno) != bh->b_blocknr) { | |
920 | ocfs2_error(sb, | |
921 | "Extent block #%llu has an invalid h_blkno " | |
922 | "of %llu", | |
923 | (unsigned long long)bh->b_blocknr, | |
924 | (unsigned long long)le64_to_cpu(eb->h_blkno)); | |
925 | return -EINVAL; | |
926 | } | |
927 | ||
928 | if (le32_to_cpu(eb->h_fs_generation) != OCFS2_SB(sb)->fs_generation) { | |
929 | ocfs2_error(sb, | |
930 | "Extent block #%llu has an invalid " | |
931 | "h_fs_generation of #%u", | |
932 | (unsigned long long)bh->b_blocknr, | |
933 | le32_to_cpu(eb->h_fs_generation)); | |
934 | return -EINVAL; | |
935 | } | |
936 | ||
937 | return 0; | |
938 | } | |
939 | ||
3d03a305 | 940 | int ocfs2_read_extent_block(struct ocfs2_caching_info *ci, u64 eb_blkno, |
5e96581a JB |
941 | struct buffer_head **bh) |
942 | { | |
943 | int rc; | |
944 | struct buffer_head *tmp = *bh; | |
945 | ||
3d03a305 | 946 | rc = ocfs2_read_block(ci, eb_blkno, &tmp, |
970e4936 | 947 | ocfs2_validate_extent_block); |
5e96581a JB |
948 | |
949 | /* If ocfs2_read_block() got us a new bh, pass it up. */ | |
970e4936 | 950 | if (!rc && !*bh) |
5e96581a JB |
951 | *bh = tmp; |
952 | ||
5e96581a JB |
953 | return rc; |
954 | } | |
955 | ||
956 | ||
ccd979bd MF |
957 | /* |
958 | * How many free extents have we got before we need more meta data? | |
959 | */ | |
960 | int ocfs2_num_free_extents(struct ocfs2_super *osb, | |
f99b9b7c | 961 | struct ocfs2_extent_tree *et) |
ccd979bd MF |
962 | { |
963 | int retval; | |
e7d4cb6b | 964 | struct ocfs2_extent_list *el = NULL; |
ccd979bd MF |
965 | struct ocfs2_extent_block *eb; |
966 | struct buffer_head *eb_bh = NULL; | |
e7d4cb6b | 967 | u64 last_eb_blk = 0; |
ccd979bd MF |
968 | |
969 | mlog_entry_void(); | |
970 | ||
f99b9b7c JB |
971 | el = et->et_root_el; |
972 | last_eb_blk = ocfs2_et_get_last_eb_blk(et); | |
ccd979bd | 973 | |
e7d4cb6b | 974 | if (last_eb_blk) { |
3d03a305 JB |
975 | retval = ocfs2_read_extent_block(et->et_ci, last_eb_blk, |
976 | &eb_bh); | |
ccd979bd MF |
977 | if (retval < 0) { |
978 | mlog_errno(retval); | |
979 | goto bail; | |
980 | } | |
981 | eb = (struct ocfs2_extent_block *) eb_bh->b_data; | |
982 | el = &eb->h_list; | |
e7d4cb6b | 983 | } |
ccd979bd MF |
984 | |
985 | BUG_ON(el->l_tree_depth != 0); | |
986 | ||
987 | retval = le16_to_cpu(el->l_count) - le16_to_cpu(el->l_next_free_rec); | |
988 | bail: | |
a81cb88b | 989 | brelse(eb_bh); |
ccd979bd MF |
990 | |
991 | mlog_exit(retval); | |
992 | return retval; | |
993 | } | |
994 | ||
995 | /* expects array to already be allocated | |
996 | * | |
997 | * sets h_signature, h_blkno, h_suballoc_bit, h_suballoc_slot, and | |
998 | * l_count for you | |
999 | */ | |
42a5a7a9 JB |
1000 | static int ocfs2_create_new_meta_bhs(handle_t *handle, |
1001 | struct ocfs2_extent_tree *et, | |
ccd979bd MF |
1002 | int wanted, |
1003 | struct ocfs2_alloc_context *meta_ac, | |
1004 | struct buffer_head *bhs[]) | |
1005 | { | |
1006 | int count, status, i; | |
1007 | u16 suballoc_bit_start; | |
1008 | u32 num_got; | |
2b6cb576 | 1009 | u64 suballoc_loc, first_blkno; |
42a5a7a9 JB |
1010 | struct ocfs2_super *osb = |
1011 | OCFS2_SB(ocfs2_metadata_cache_get_super(et->et_ci)); | |
ccd979bd MF |
1012 | struct ocfs2_extent_block *eb; |
1013 | ||
1014 | mlog_entry_void(); | |
1015 | ||
1016 | count = 0; | |
1017 | while (count < wanted) { | |
1ed9b777 | 1018 | status = ocfs2_claim_metadata(handle, |
ccd979bd MF |
1019 | meta_ac, |
1020 | wanted - count, | |
2b6cb576 | 1021 | &suballoc_loc, |
ccd979bd MF |
1022 | &suballoc_bit_start, |
1023 | &num_got, | |
1024 | &first_blkno); | |
1025 | if (status < 0) { | |
1026 | mlog_errno(status); | |
1027 | goto bail; | |
1028 | } | |
1029 | ||
1030 | for(i = count; i < (num_got + count); i++) { | |
1031 | bhs[i] = sb_getblk(osb->sb, first_blkno); | |
1032 | if (bhs[i] == NULL) { | |
1033 | status = -EIO; | |
1034 | mlog_errno(status); | |
1035 | goto bail; | |
1036 | } | |
42a5a7a9 | 1037 | ocfs2_set_new_buffer_uptodate(et->et_ci, bhs[i]); |
ccd979bd | 1038 | |
42a5a7a9 JB |
1039 | status = ocfs2_journal_access_eb(handle, et->et_ci, |
1040 | bhs[i], | |
13723d00 | 1041 | OCFS2_JOURNAL_ACCESS_CREATE); |
ccd979bd MF |
1042 | if (status < 0) { |
1043 | mlog_errno(status); | |
1044 | goto bail; | |
1045 | } | |
1046 | ||
1047 | memset(bhs[i]->b_data, 0, osb->sb->s_blocksize); | |
1048 | eb = (struct ocfs2_extent_block *) bhs[i]->b_data; | |
1049 | /* Ok, setup the minimal stuff here. */ | |
1050 | strcpy(eb->h_signature, OCFS2_EXTENT_BLOCK_SIGNATURE); | |
1051 | eb->h_blkno = cpu_to_le64(first_blkno); | |
1052 | eb->h_fs_generation = cpu_to_le32(osb->fs_generation); | |
b89c5428 TY |
1053 | eb->h_suballoc_slot = |
1054 | cpu_to_le16(meta_ac->ac_alloc_slot); | |
2b6cb576 | 1055 | eb->h_suballoc_loc = cpu_to_le64(suballoc_loc); |
ccd979bd MF |
1056 | eb->h_suballoc_bit = cpu_to_le16(suballoc_bit_start); |
1057 | eb->h_list.l_count = | |
1058 | cpu_to_le16(ocfs2_extent_recs_per_eb(osb->sb)); | |
1059 | ||
1060 | suballoc_bit_start++; | |
1061 | first_blkno++; | |
1062 | ||
1063 | /* We'll also be dirtied by the caller, so | |
1064 | * this isn't absolutely necessary. */ | |
ec20cec7 | 1065 | ocfs2_journal_dirty(handle, bhs[i]); |
ccd979bd MF |
1066 | } |
1067 | ||
1068 | count += num_got; | |
1069 | } | |
1070 | ||
1071 | status = 0; | |
1072 | bail: | |
1073 | if (status < 0) { | |
1074 | for(i = 0; i < wanted; i++) { | |
a81cb88b | 1075 | brelse(bhs[i]); |
ccd979bd MF |
1076 | bhs[i] = NULL; |
1077 | } | |
1078 | } | |
1079 | mlog_exit(status); | |
1080 | return status; | |
1081 | } | |
1082 | ||
dcd0538f MF |
1083 | /* |
1084 | * Helper function for ocfs2_add_branch() and ocfs2_shift_tree_depth(). | |
1085 | * | |
1086 | * Returns the sum of the rightmost extent rec logical offset and | |
1087 | * cluster count. | |
1088 | * | |
1089 | * ocfs2_add_branch() uses this to determine what logical cluster | |
1090 | * value should be populated into the leftmost new branch records. | |
1091 | * | |
1092 | * ocfs2_shift_tree_depth() uses this to determine the # clusters | |
1093 | * value for the new topmost tree record. | |
1094 | */ | |
1095 | static inline u32 ocfs2_sum_rightmost_rec(struct ocfs2_extent_list *el) | |
1096 | { | |
1097 | int i; | |
1098 | ||
1099 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
1100 | ||
1101 | return le32_to_cpu(el->l_recs[i].e_cpos) + | |
e48edee2 | 1102 | ocfs2_rec_clusters(el, &el->l_recs[i]); |
dcd0538f MF |
1103 | } |
1104 | ||
6b791bcc TM |
1105 | /* |
1106 | * Change range of the branches in the right most path according to the leaf | |
1107 | * extent block's rightmost record. | |
1108 | */ | |
1109 | static int ocfs2_adjust_rightmost_branch(handle_t *handle, | |
6b791bcc TM |
1110 | struct ocfs2_extent_tree *et) |
1111 | { | |
1112 | int status; | |
1113 | struct ocfs2_path *path = NULL; | |
1114 | struct ocfs2_extent_list *el; | |
1115 | struct ocfs2_extent_rec *rec; | |
1116 | ||
1117 | path = ocfs2_new_path_from_et(et); | |
1118 | if (!path) { | |
1119 | status = -ENOMEM; | |
1120 | return status; | |
1121 | } | |
1122 | ||
facdb77f | 1123 | status = ocfs2_find_path(et->et_ci, path, UINT_MAX); |
6b791bcc TM |
1124 | if (status < 0) { |
1125 | mlog_errno(status); | |
1126 | goto out; | |
1127 | } | |
1128 | ||
c901fb00 | 1129 | status = ocfs2_extend_trans(handle, path_num_items(path)); |
6b791bcc TM |
1130 | if (status < 0) { |
1131 | mlog_errno(status); | |
1132 | goto out; | |
1133 | } | |
1134 | ||
d401dc12 | 1135 | status = ocfs2_journal_access_path(et->et_ci, handle, path); |
6b791bcc TM |
1136 | if (status < 0) { |
1137 | mlog_errno(status); | |
1138 | goto out; | |
1139 | } | |
1140 | ||
1141 | el = path_leaf_el(path); | |
1142 | rec = &el->l_recs[le32_to_cpu(el->l_next_free_rec) - 1]; | |
1143 | ||
d401dc12 | 1144 | ocfs2_adjust_rightmost_records(handle, et, path, rec); |
6b791bcc TM |
1145 | |
1146 | out: | |
1147 | ocfs2_free_path(path); | |
1148 | return status; | |
1149 | } | |
1150 | ||
ccd979bd MF |
1151 | /* |
1152 | * Add an entire tree branch to our inode. eb_bh is the extent block | |
d401dc12 | 1153 | * to start at, if we don't want to start the branch at the root |
ccd979bd MF |
1154 | * structure. |
1155 | * | |
1156 | * last_eb_bh is required as we have to update it's next_leaf pointer | |
1157 | * for the new last extent block. | |
1158 | * | |
1159 | * the new branch will be 'empty' in the sense that every block will | |
e48edee2 | 1160 | * contain a single record with cluster count == 0. |
ccd979bd | 1161 | */ |
d401dc12 | 1162 | static int ocfs2_add_branch(handle_t *handle, |
e7d4cb6b | 1163 | struct ocfs2_extent_tree *et, |
ccd979bd | 1164 | struct buffer_head *eb_bh, |
328d5752 | 1165 | struct buffer_head **last_eb_bh, |
ccd979bd MF |
1166 | struct ocfs2_alloc_context *meta_ac) |
1167 | { | |
1168 | int status, new_blocks, i; | |
1169 | u64 next_blkno, new_last_eb_blk; | |
1170 | struct buffer_head *bh; | |
1171 | struct buffer_head **new_eb_bhs = NULL; | |
ccd979bd MF |
1172 | struct ocfs2_extent_block *eb; |
1173 | struct ocfs2_extent_list *eb_el; | |
1174 | struct ocfs2_extent_list *el; | |
6b791bcc | 1175 | u32 new_cpos, root_end; |
ccd979bd MF |
1176 | |
1177 | mlog_entry_void(); | |
1178 | ||
328d5752 | 1179 | BUG_ON(!last_eb_bh || !*last_eb_bh); |
ccd979bd | 1180 | |
ccd979bd MF |
1181 | if (eb_bh) { |
1182 | eb = (struct ocfs2_extent_block *) eb_bh->b_data; | |
1183 | el = &eb->h_list; | |
1184 | } else | |
ce1d9ea6 | 1185 | el = et->et_root_el; |
ccd979bd MF |
1186 | |
1187 | /* we never add a branch to a leaf. */ | |
1188 | BUG_ON(!el->l_tree_depth); | |
1189 | ||
1190 | new_blocks = le16_to_cpu(el->l_tree_depth); | |
1191 | ||
6b791bcc TM |
1192 | eb = (struct ocfs2_extent_block *)(*last_eb_bh)->b_data; |
1193 | new_cpos = ocfs2_sum_rightmost_rec(&eb->h_list); | |
1194 | root_end = ocfs2_sum_rightmost_rec(et->et_root_el); | |
1195 | ||
1196 | /* | |
1197 | * If there is a gap before the root end and the real end | |
1198 | * of the righmost leaf block, we need to remove the gap | |
1199 | * between new_cpos and root_end first so that the tree | |
1200 | * is consistent after we add a new branch(it will start | |
1201 | * from new_cpos). | |
1202 | */ | |
1203 | if (root_end > new_cpos) { | |
1204 | mlog(0, "adjust the cluster end from %u to %u\n", | |
1205 | root_end, new_cpos); | |
d401dc12 | 1206 | status = ocfs2_adjust_rightmost_branch(handle, et); |
6b791bcc TM |
1207 | if (status) { |
1208 | mlog_errno(status); | |
1209 | goto bail; | |
1210 | } | |
1211 | } | |
1212 | ||
ccd979bd MF |
1213 | /* allocate the number of new eb blocks we need */ |
1214 | new_eb_bhs = kcalloc(new_blocks, sizeof(struct buffer_head *), | |
1215 | GFP_KERNEL); | |
1216 | if (!new_eb_bhs) { | |
1217 | status = -ENOMEM; | |
1218 | mlog_errno(status); | |
1219 | goto bail; | |
1220 | } | |
1221 | ||
42a5a7a9 | 1222 | status = ocfs2_create_new_meta_bhs(handle, et, new_blocks, |
ccd979bd MF |
1223 | meta_ac, new_eb_bhs); |
1224 | if (status < 0) { | |
1225 | mlog_errno(status); | |
1226 | goto bail; | |
1227 | } | |
1228 | ||
1229 | /* Note: new_eb_bhs[new_blocks - 1] is the guy which will be | |
1230 | * linked with the rest of the tree. | |
1231 | * conversly, new_eb_bhs[0] is the new bottommost leaf. | |
1232 | * | |
1233 | * when we leave the loop, new_last_eb_blk will point to the | |
1234 | * newest leaf, and next_blkno will point to the topmost extent | |
1235 | * block. */ | |
1236 | next_blkno = new_last_eb_blk = 0; | |
1237 | for(i = 0; i < new_blocks; i++) { | |
1238 | bh = new_eb_bhs[i]; | |
1239 | eb = (struct ocfs2_extent_block *) bh->b_data; | |
5e96581a JB |
1240 | /* ocfs2_create_new_meta_bhs() should create it right! */ |
1241 | BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb)); | |
ccd979bd MF |
1242 | eb_el = &eb->h_list; |
1243 | ||
d401dc12 | 1244 | status = ocfs2_journal_access_eb(handle, et->et_ci, bh, |
13723d00 | 1245 | OCFS2_JOURNAL_ACCESS_CREATE); |
ccd979bd MF |
1246 | if (status < 0) { |
1247 | mlog_errno(status); | |
1248 | goto bail; | |
1249 | } | |
1250 | ||
1251 | eb->h_next_leaf_blk = 0; | |
1252 | eb_el->l_tree_depth = cpu_to_le16(i); | |
1253 | eb_el->l_next_free_rec = cpu_to_le16(1); | |
dcd0538f MF |
1254 | /* |
1255 | * This actually counts as an empty extent as | |
1256 | * c_clusters == 0 | |
1257 | */ | |
1258 | eb_el->l_recs[0].e_cpos = cpu_to_le32(new_cpos); | |
ccd979bd | 1259 | eb_el->l_recs[0].e_blkno = cpu_to_le64(next_blkno); |
e48edee2 MF |
1260 | /* |
1261 | * eb_el isn't always an interior node, but even leaf | |
1262 | * nodes want a zero'd flags and reserved field so | |
1263 | * this gets the whole 32 bits regardless of use. | |
1264 | */ | |
1265 | eb_el->l_recs[0].e_int_clusters = cpu_to_le32(0); | |
ccd979bd MF |
1266 | if (!eb_el->l_tree_depth) |
1267 | new_last_eb_blk = le64_to_cpu(eb->h_blkno); | |
1268 | ||
ec20cec7 | 1269 | ocfs2_journal_dirty(handle, bh); |
ccd979bd MF |
1270 | next_blkno = le64_to_cpu(eb->h_blkno); |
1271 | } | |
1272 | ||
1273 | /* This is a bit hairy. We want to update up to three blocks | |
1274 | * here without leaving any of them in an inconsistent state | |
1275 | * in case of error. We don't have to worry about | |
1276 | * journal_dirty erroring as it won't unless we've aborted the | |
1277 | * handle (in which case we would never be here) so reserving | |
1278 | * the write with journal_access is all we need to do. */ | |
d401dc12 | 1279 | status = ocfs2_journal_access_eb(handle, et->et_ci, *last_eb_bh, |
13723d00 | 1280 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
1281 | if (status < 0) { |
1282 | mlog_errno(status); | |
1283 | goto bail; | |
1284 | } | |
d9a0a1f8 | 1285 | status = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 1286 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
1287 | if (status < 0) { |
1288 | mlog_errno(status); | |
1289 | goto bail; | |
1290 | } | |
1291 | if (eb_bh) { | |
d401dc12 | 1292 | status = ocfs2_journal_access_eb(handle, et->et_ci, eb_bh, |
13723d00 | 1293 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
1294 | if (status < 0) { |
1295 | mlog_errno(status); | |
1296 | goto bail; | |
1297 | } | |
1298 | } | |
1299 | ||
1300 | /* Link the new branch into the rest of the tree (el will | |
e7d4cb6b | 1301 | * either be on the root_bh, or the extent block passed in. */ |
ccd979bd MF |
1302 | i = le16_to_cpu(el->l_next_free_rec); |
1303 | el->l_recs[i].e_blkno = cpu_to_le64(next_blkno); | |
dcd0538f | 1304 | el->l_recs[i].e_cpos = cpu_to_le32(new_cpos); |
e48edee2 | 1305 | el->l_recs[i].e_int_clusters = 0; |
ccd979bd MF |
1306 | le16_add_cpu(&el->l_next_free_rec, 1); |
1307 | ||
1308 | /* fe needs a new last extent block pointer, as does the | |
1309 | * next_leaf on the previously last-extent-block. */ | |
35dc0aa3 | 1310 | ocfs2_et_set_last_eb_blk(et, new_last_eb_blk); |
ccd979bd | 1311 | |
328d5752 | 1312 | eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data; |
ccd979bd MF |
1313 | eb->h_next_leaf_blk = cpu_to_le64(new_last_eb_blk); |
1314 | ||
ec20cec7 JB |
1315 | ocfs2_journal_dirty(handle, *last_eb_bh); |
1316 | ocfs2_journal_dirty(handle, et->et_root_bh); | |
1317 | if (eb_bh) | |
1318 | ocfs2_journal_dirty(handle, eb_bh); | |
ccd979bd | 1319 | |
328d5752 MF |
1320 | /* |
1321 | * Some callers want to track the rightmost leaf so pass it | |
1322 | * back here. | |
1323 | */ | |
1324 | brelse(*last_eb_bh); | |
1325 | get_bh(new_eb_bhs[0]); | |
1326 | *last_eb_bh = new_eb_bhs[0]; | |
1327 | ||
ccd979bd MF |
1328 | status = 0; |
1329 | bail: | |
1330 | if (new_eb_bhs) { | |
1331 | for (i = 0; i < new_blocks; i++) | |
a81cb88b | 1332 | brelse(new_eb_bhs[i]); |
ccd979bd MF |
1333 | kfree(new_eb_bhs); |
1334 | } | |
1335 | ||
1336 | mlog_exit(status); | |
1337 | return status; | |
1338 | } | |
1339 | ||
1340 | /* | |
1341 | * adds another level to the allocation tree. | |
1342 | * returns back the new extent block so you can add a branch to it | |
1343 | * after this call. | |
1344 | */ | |
d401dc12 | 1345 | static int ocfs2_shift_tree_depth(handle_t *handle, |
e7d4cb6b | 1346 | struct ocfs2_extent_tree *et, |
ccd979bd MF |
1347 | struct ocfs2_alloc_context *meta_ac, |
1348 | struct buffer_head **ret_new_eb_bh) | |
1349 | { | |
1350 | int status, i; | |
dcd0538f | 1351 | u32 new_clusters; |
ccd979bd | 1352 | struct buffer_head *new_eb_bh = NULL; |
ccd979bd | 1353 | struct ocfs2_extent_block *eb; |
e7d4cb6b | 1354 | struct ocfs2_extent_list *root_el; |
ccd979bd MF |
1355 | struct ocfs2_extent_list *eb_el; |
1356 | ||
1357 | mlog_entry_void(); | |
1358 | ||
42a5a7a9 | 1359 | status = ocfs2_create_new_meta_bhs(handle, et, 1, meta_ac, |
ccd979bd MF |
1360 | &new_eb_bh); |
1361 | if (status < 0) { | |
1362 | mlog_errno(status); | |
1363 | goto bail; | |
1364 | } | |
1365 | ||
1366 | eb = (struct ocfs2_extent_block *) new_eb_bh->b_data; | |
5e96581a JB |
1367 | /* ocfs2_create_new_meta_bhs() should create it right! */ |
1368 | BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb)); | |
ccd979bd MF |
1369 | |
1370 | eb_el = &eb->h_list; | |
ce1d9ea6 | 1371 | root_el = et->et_root_el; |
ccd979bd | 1372 | |
d401dc12 | 1373 | status = ocfs2_journal_access_eb(handle, et->et_ci, new_eb_bh, |
13723d00 | 1374 | OCFS2_JOURNAL_ACCESS_CREATE); |
ccd979bd MF |
1375 | if (status < 0) { |
1376 | mlog_errno(status); | |
1377 | goto bail; | |
1378 | } | |
1379 | ||
e7d4cb6b TM |
1380 | /* copy the root extent list data into the new extent block */ |
1381 | eb_el->l_tree_depth = root_el->l_tree_depth; | |
1382 | eb_el->l_next_free_rec = root_el->l_next_free_rec; | |
1383 | for (i = 0; i < le16_to_cpu(root_el->l_next_free_rec); i++) | |
1384 | eb_el->l_recs[i] = root_el->l_recs[i]; | |
ccd979bd | 1385 | |
ec20cec7 | 1386 | ocfs2_journal_dirty(handle, new_eb_bh); |
ccd979bd | 1387 | |
d9a0a1f8 | 1388 | status = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 1389 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
1390 | if (status < 0) { |
1391 | mlog_errno(status); | |
1392 | goto bail; | |
1393 | } | |
1394 | ||
dcd0538f MF |
1395 | new_clusters = ocfs2_sum_rightmost_rec(eb_el); |
1396 | ||
e7d4cb6b TM |
1397 | /* update root_bh now */ |
1398 | le16_add_cpu(&root_el->l_tree_depth, 1); | |
1399 | root_el->l_recs[0].e_cpos = 0; | |
1400 | root_el->l_recs[0].e_blkno = eb->h_blkno; | |
1401 | root_el->l_recs[0].e_int_clusters = cpu_to_le32(new_clusters); | |
1402 | for (i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++) | |
1403 | memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec)); | |
1404 | root_el->l_next_free_rec = cpu_to_le16(1); | |
ccd979bd MF |
1405 | |
1406 | /* If this is our 1st tree depth shift, then last_eb_blk | |
1407 | * becomes the allocated extent block */ | |
e7d4cb6b | 1408 | if (root_el->l_tree_depth == cpu_to_le16(1)) |
35dc0aa3 | 1409 | ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno)); |
ccd979bd | 1410 | |
ec20cec7 | 1411 | ocfs2_journal_dirty(handle, et->et_root_bh); |
ccd979bd MF |
1412 | |
1413 | *ret_new_eb_bh = new_eb_bh; | |
1414 | new_eb_bh = NULL; | |
1415 | status = 0; | |
1416 | bail: | |
a81cb88b | 1417 | brelse(new_eb_bh); |
ccd979bd MF |
1418 | |
1419 | mlog_exit(status); | |
1420 | return status; | |
1421 | } | |
1422 | ||
ccd979bd MF |
1423 | /* |
1424 | * Should only be called when there is no space left in any of the | |
1425 | * leaf nodes. What we want to do is find the lowest tree depth | |
1426 | * non-leaf extent block with room for new records. There are three | |
1427 | * valid results of this search: | |
1428 | * | |
1429 | * 1) a lowest extent block is found, then we pass it back in | |
1430 | * *lowest_eb_bh and return '0' | |
1431 | * | |
e7d4cb6b | 1432 | * 2) the search fails to find anything, but the root_el has room. We |
ccd979bd MF |
1433 | * pass NULL back in *lowest_eb_bh, but still return '0' |
1434 | * | |
e7d4cb6b | 1435 | * 3) the search fails to find anything AND the root_el is full, in |
ccd979bd MF |
1436 | * which case we return > 0 |
1437 | * | |
1438 | * return status < 0 indicates an error. | |
1439 | */ | |
d401dc12 | 1440 | static int ocfs2_find_branch_target(struct ocfs2_extent_tree *et, |
ccd979bd MF |
1441 | struct buffer_head **target_bh) |
1442 | { | |
1443 | int status = 0, i; | |
1444 | u64 blkno; | |
ccd979bd MF |
1445 | struct ocfs2_extent_block *eb; |
1446 | struct ocfs2_extent_list *el; | |
1447 | struct buffer_head *bh = NULL; | |
1448 | struct buffer_head *lowest_bh = NULL; | |
1449 | ||
1450 | mlog_entry_void(); | |
1451 | ||
1452 | *target_bh = NULL; | |
1453 | ||
ce1d9ea6 | 1454 | el = et->et_root_el; |
ccd979bd MF |
1455 | |
1456 | while(le16_to_cpu(el->l_tree_depth) > 1) { | |
1457 | if (le16_to_cpu(el->l_next_free_rec) == 0) { | |
3d03a305 JB |
1458 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
1459 | "Owner %llu has empty " | |
ccd979bd | 1460 | "extent list (next_free_rec == 0)", |
3d03a305 | 1461 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci)); |
ccd979bd MF |
1462 | status = -EIO; |
1463 | goto bail; | |
1464 | } | |
1465 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
1466 | blkno = le64_to_cpu(el->l_recs[i].e_blkno); | |
1467 | if (!blkno) { | |
3d03a305 JB |
1468 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
1469 | "Owner %llu has extent " | |
ccd979bd MF |
1470 | "list where extent # %d has no physical " |
1471 | "block start", | |
3d03a305 | 1472 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), i); |
ccd979bd MF |
1473 | status = -EIO; |
1474 | goto bail; | |
1475 | } | |
1476 | ||
a81cb88b MF |
1477 | brelse(bh); |
1478 | bh = NULL; | |
ccd979bd | 1479 | |
3d03a305 | 1480 | status = ocfs2_read_extent_block(et->et_ci, blkno, &bh); |
ccd979bd MF |
1481 | if (status < 0) { |
1482 | mlog_errno(status); | |
1483 | goto bail; | |
1484 | } | |
dcd0538f MF |
1485 | |
1486 | eb = (struct ocfs2_extent_block *) bh->b_data; | |
dcd0538f MF |
1487 | el = &eb->h_list; |
1488 | ||
1489 | if (le16_to_cpu(el->l_next_free_rec) < | |
1490 | le16_to_cpu(el->l_count)) { | |
a81cb88b | 1491 | brelse(lowest_bh); |
dcd0538f MF |
1492 | lowest_bh = bh; |
1493 | get_bh(lowest_bh); | |
1494 | } | |
1495 | } | |
1496 | ||
1497 | /* If we didn't find one and the fe doesn't have any room, | |
1498 | * then return '1' */ | |
ce1d9ea6 | 1499 | el = et->et_root_el; |
e7d4cb6b | 1500 | if (!lowest_bh && (el->l_next_free_rec == el->l_count)) |
dcd0538f MF |
1501 | status = 1; |
1502 | ||
1503 | *target_bh = lowest_bh; | |
1504 | bail: | |
a81cb88b | 1505 | brelse(bh); |
dcd0538f MF |
1506 | |
1507 | mlog_exit(status); | |
1508 | return status; | |
1509 | } | |
1510 | ||
c3afcbb3 MF |
1511 | /* |
1512 | * Grow a b-tree so that it has more records. | |
1513 | * | |
1514 | * We might shift the tree depth in which case existing paths should | |
1515 | * be considered invalid. | |
1516 | * | |
1517 | * Tree depth after the grow is returned via *final_depth. | |
328d5752 MF |
1518 | * |
1519 | * *last_eb_bh will be updated by ocfs2_add_branch(). | |
c3afcbb3 | 1520 | */ |
d401dc12 JB |
1521 | static int ocfs2_grow_tree(handle_t *handle, struct ocfs2_extent_tree *et, |
1522 | int *final_depth, struct buffer_head **last_eb_bh, | |
c3afcbb3 MF |
1523 | struct ocfs2_alloc_context *meta_ac) |
1524 | { | |
1525 | int ret, shift; | |
ce1d9ea6 | 1526 | struct ocfs2_extent_list *el = et->et_root_el; |
e7d4cb6b | 1527 | int depth = le16_to_cpu(el->l_tree_depth); |
c3afcbb3 MF |
1528 | struct buffer_head *bh = NULL; |
1529 | ||
1530 | BUG_ON(meta_ac == NULL); | |
1531 | ||
d401dc12 | 1532 | shift = ocfs2_find_branch_target(et, &bh); |
c3afcbb3 MF |
1533 | if (shift < 0) { |
1534 | ret = shift; | |
1535 | mlog_errno(ret); | |
1536 | goto out; | |
1537 | } | |
1538 | ||
1539 | /* We traveled all the way to the bottom of the allocation tree | |
1540 | * and didn't find room for any more extents - we need to add | |
1541 | * another tree level */ | |
1542 | if (shift) { | |
1543 | BUG_ON(bh); | |
1544 | mlog(0, "need to shift tree depth (current = %d)\n", depth); | |
1545 | ||
1546 | /* ocfs2_shift_tree_depth will return us a buffer with | |
1547 | * the new extent block (so we can pass that to | |
1548 | * ocfs2_add_branch). */ | |
d401dc12 | 1549 | ret = ocfs2_shift_tree_depth(handle, et, meta_ac, &bh); |
c3afcbb3 MF |
1550 | if (ret < 0) { |
1551 | mlog_errno(ret); | |
1552 | goto out; | |
1553 | } | |
1554 | depth++; | |
328d5752 MF |
1555 | if (depth == 1) { |
1556 | /* | |
1557 | * Special case: we have room now if we shifted from | |
1558 | * tree_depth 0, so no more work needs to be done. | |
1559 | * | |
1560 | * We won't be calling add_branch, so pass | |
1561 | * back *last_eb_bh as the new leaf. At depth | |
1562 | * zero, it should always be null so there's | |
1563 | * no reason to brelse. | |
1564 | */ | |
1565 | BUG_ON(*last_eb_bh); | |
1566 | get_bh(bh); | |
1567 | *last_eb_bh = bh; | |
c3afcbb3 | 1568 | goto out; |
328d5752 | 1569 | } |
c3afcbb3 MF |
1570 | } |
1571 | ||
1572 | /* call ocfs2_add_branch to add the final part of the tree with | |
1573 | * the new data. */ | |
1574 | mlog(0, "add branch. bh = %p\n", bh); | |
d401dc12 | 1575 | ret = ocfs2_add_branch(handle, et, bh, last_eb_bh, |
c3afcbb3 MF |
1576 | meta_ac); |
1577 | if (ret < 0) { | |
1578 | mlog_errno(ret); | |
1579 | goto out; | |
1580 | } | |
1581 | ||
1582 | out: | |
1583 | if (final_depth) | |
1584 | *final_depth = depth; | |
1585 | brelse(bh); | |
1586 | return ret; | |
1587 | } | |
1588 | ||
dcd0538f MF |
1589 | /* |
1590 | * This function will discard the rightmost extent record. | |
1591 | */ | |
1592 | static void ocfs2_shift_records_right(struct ocfs2_extent_list *el) | |
1593 | { | |
1594 | int next_free = le16_to_cpu(el->l_next_free_rec); | |
1595 | int count = le16_to_cpu(el->l_count); | |
1596 | unsigned int num_bytes; | |
1597 | ||
1598 | BUG_ON(!next_free); | |
1599 | /* This will cause us to go off the end of our extent list. */ | |
1600 | BUG_ON(next_free >= count); | |
1601 | ||
1602 | num_bytes = sizeof(struct ocfs2_extent_rec) * next_free; | |
1603 | ||
1604 | memmove(&el->l_recs[1], &el->l_recs[0], num_bytes); | |
1605 | } | |
1606 | ||
1607 | static void ocfs2_rotate_leaf(struct ocfs2_extent_list *el, | |
1608 | struct ocfs2_extent_rec *insert_rec) | |
1609 | { | |
1610 | int i, insert_index, next_free, has_empty, num_bytes; | |
1611 | u32 insert_cpos = le32_to_cpu(insert_rec->e_cpos); | |
1612 | struct ocfs2_extent_rec *rec; | |
1613 | ||
1614 | next_free = le16_to_cpu(el->l_next_free_rec); | |
1615 | has_empty = ocfs2_is_empty_extent(&el->l_recs[0]); | |
1616 | ||
1617 | BUG_ON(!next_free); | |
1618 | ||
1619 | /* The tree code before us didn't allow enough room in the leaf. */ | |
b1f3550f | 1620 | BUG_ON(el->l_next_free_rec == el->l_count && !has_empty); |
dcd0538f MF |
1621 | |
1622 | /* | |
1623 | * The easiest way to approach this is to just remove the | |
1624 | * empty extent and temporarily decrement next_free. | |
1625 | */ | |
1626 | if (has_empty) { | |
1627 | /* | |
1628 | * If next_free was 1 (only an empty extent), this | |
1629 | * loop won't execute, which is fine. We still want | |
1630 | * the decrement above to happen. | |
1631 | */ | |
1632 | for(i = 0; i < (next_free - 1); i++) | |
1633 | el->l_recs[i] = el->l_recs[i+1]; | |
1634 | ||
1635 | next_free--; | |
1636 | } | |
1637 | ||
1638 | /* | |
1639 | * Figure out what the new record index should be. | |
1640 | */ | |
1641 | for(i = 0; i < next_free; i++) { | |
1642 | rec = &el->l_recs[i]; | |
1643 | ||
1644 | if (insert_cpos < le32_to_cpu(rec->e_cpos)) | |
1645 | break; | |
1646 | } | |
1647 | insert_index = i; | |
1648 | ||
1649 | mlog(0, "ins %u: index %d, has_empty %d, next_free %d, count %d\n", | |
1650 | insert_cpos, insert_index, has_empty, next_free, le16_to_cpu(el->l_count)); | |
1651 | ||
1652 | BUG_ON(insert_index < 0); | |
1653 | BUG_ON(insert_index >= le16_to_cpu(el->l_count)); | |
1654 | BUG_ON(insert_index > next_free); | |
1655 | ||
1656 | /* | |
1657 | * No need to memmove if we're just adding to the tail. | |
1658 | */ | |
1659 | if (insert_index != next_free) { | |
1660 | BUG_ON(next_free >= le16_to_cpu(el->l_count)); | |
1661 | ||
1662 | num_bytes = next_free - insert_index; | |
1663 | num_bytes *= sizeof(struct ocfs2_extent_rec); | |
1664 | memmove(&el->l_recs[insert_index + 1], | |
1665 | &el->l_recs[insert_index], | |
1666 | num_bytes); | |
1667 | } | |
1668 | ||
1669 | /* | |
1670 | * Either we had an empty extent, and need to re-increment or | |
1671 | * there was no empty extent on a non full rightmost leaf node, | |
1672 | * in which case we still need to increment. | |
1673 | */ | |
1674 | next_free++; | |
1675 | el->l_next_free_rec = cpu_to_le16(next_free); | |
1676 | /* | |
1677 | * Make sure none of the math above just messed up our tree. | |
1678 | */ | |
1679 | BUG_ON(le16_to_cpu(el->l_next_free_rec) > le16_to_cpu(el->l_count)); | |
1680 | ||
1681 | el->l_recs[insert_index] = *insert_rec; | |
1682 | ||
1683 | } | |
1684 | ||
328d5752 MF |
1685 | static void ocfs2_remove_empty_extent(struct ocfs2_extent_list *el) |
1686 | { | |
1687 | int size, num_recs = le16_to_cpu(el->l_next_free_rec); | |
1688 | ||
1689 | BUG_ON(num_recs == 0); | |
1690 | ||
1691 | if (ocfs2_is_empty_extent(&el->l_recs[0])) { | |
1692 | num_recs--; | |
1693 | size = num_recs * sizeof(struct ocfs2_extent_rec); | |
1694 | memmove(&el->l_recs[0], &el->l_recs[1], size); | |
1695 | memset(&el->l_recs[num_recs], 0, | |
1696 | sizeof(struct ocfs2_extent_rec)); | |
1697 | el->l_next_free_rec = cpu_to_le16(num_recs); | |
1698 | } | |
1699 | } | |
1700 | ||
dcd0538f MF |
1701 | /* |
1702 | * Create an empty extent record . | |
1703 | * | |
1704 | * l_next_free_rec may be updated. | |
1705 | * | |
1706 | * If an empty extent already exists do nothing. | |
1707 | */ | |
1708 | static void ocfs2_create_empty_extent(struct ocfs2_extent_list *el) | |
1709 | { | |
1710 | int next_free = le16_to_cpu(el->l_next_free_rec); | |
1711 | ||
e48edee2 MF |
1712 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); |
1713 | ||
dcd0538f MF |
1714 | if (next_free == 0) |
1715 | goto set_and_inc; | |
1716 | ||
1717 | if (ocfs2_is_empty_extent(&el->l_recs[0])) | |
1718 | return; | |
1719 | ||
1720 | mlog_bug_on_msg(el->l_count == el->l_next_free_rec, | |
1721 | "Asked to create an empty extent in a full list:\n" | |
1722 | "count = %u, tree depth = %u", | |
1723 | le16_to_cpu(el->l_count), | |
1724 | le16_to_cpu(el->l_tree_depth)); | |
1725 | ||
1726 | ocfs2_shift_records_right(el); | |
1727 | ||
1728 | set_and_inc: | |
1729 | le16_add_cpu(&el->l_next_free_rec, 1); | |
1730 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
1731 | } | |
1732 | ||
1733 | /* | |
1734 | * For a rotation which involves two leaf nodes, the "root node" is | |
1735 | * the lowest level tree node which contains a path to both leafs. This | |
1736 | * resulting set of information can be used to form a complete "subtree" | |
1737 | * | |
1738 | * This function is passed two full paths from the dinode down to a | |
1739 | * pair of adjacent leaves. It's task is to figure out which path | |
1740 | * index contains the subtree root - this can be the root index itself | |
1741 | * in a worst-case rotation. | |
1742 | * | |
1743 | * The array index of the subtree root is passed back. | |
1744 | */ | |
38a04e43 TM |
1745 | int ocfs2_find_subtree_root(struct ocfs2_extent_tree *et, |
1746 | struct ocfs2_path *left, | |
1747 | struct ocfs2_path *right) | |
dcd0538f MF |
1748 | { |
1749 | int i = 0; | |
1750 | ||
1751 | /* | |
1752 | * Check that the caller passed in two paths from the same tree. | |
1753 | */ | |
1754 | BUG_ON(path_root_bh(left) != path_root_bh(right)); | |
1755 | ||
1756 | do { | |
1757 | i++; | |
1758 | ||
1759 | /* | |
1760 | * The caller didn't pass two adjacent paths. | |
1761 | */ | |
1762 | mlog_bug_on_msg(i > left->p_tree_depth, | |
7dc02805 | 1763 | "Owner %llu, left depth %u, right depth %u\n" |
dcd0538f | 1764 | "left leaf blk %llu, right leaf blk %llu\n", |
7dc02805 JB |
1765 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
1766 | left->p_tree_depth, right->p_tree_depth, | |
dcd0538f MF |
1767 | (unsigned long long)path_leaf_bh(left)->b_blocknr, |
1768 | (unsigned long long)path_leaf_bh(right)->b_blocknr); | |
1769 | } while (left->p_node[i].bh->b_blocknr == | |
1770 | right->p_node[i].bh->b_blocknr); | |
1771 | ||
1772 | return i - 1; | |
1773 | } | |
1774 | ||
1775 | typedef void (path_insert_t)(void *, struct buffer_head *); | |
1776 | ||
1777 | /* | |
1778 | * Traverse a btree path in search of cpos, starting at root_el. | |
1779 | * | |
1780 | * This code can be called with a cpos larger than the tree, in which | |
1781 | * case it will return the rightmost path. | |
1782 | */ | |
facdb77f | 1783 | static int __ocfs2_find_path(struct ocfs2_caching_info *ci, |
dcd0538f MF |
1784 | struct ocfs2_extent_list *root_el, u32 cpos, |
1785 | path_insert_t *func, void *data) | |
1786 | { | |
1787 | int i, ret = 0; | |
1788 | u32 range; | |
1789 | u64 blkno; | |
1790 | struct buffer_head *bh = NULL; | |
1791 | struct ocfs2_extent_block *eb; | |
1792 | struct ocfs2_extent_list *el; | |
1793 | struct ocfs2_extent_rec *rec; | |
dcd0538f MF |
1794 | |
1795 | el = root_el; | |
1796 | while (el->l_tree_depth) { | |
1797 | if (le16_to_cpu(el->l_next_free_rec) == 0) { | |
facdb77f JB |
1798 | ocfs2_error(ocfs2_metadata_cache_get_super(ci), |
1799 | "Owner %llu has empty extent list at " | |
dcd0538f | 1800 | "depth %u\n", |
facdb77f | 1801 | (unsigned long long)ocfs2_metadata_cache_owner(ci), |
dcd0538f MF |
1802 | le16_to_cpu(el->l_tree_depth)); |
1803 | ret = -EROFS; | |
1804 | goto out; | |
1805 | ||
1806 | } | |
1807 | ||
1808 | for(i = 0; i < le16_to_cpu(el->l_next_free_rec) - 1; i++) { | |
1809 | rec = &el->l_recs[i]; | |
1810 | ||
1811 | /* | |
1812 | * In the case that cpos is off the allocation | |
1813 | * tree, this should just wind up returning the | |
1814 | * rightmost record. | |
1815 | */ | |
1816 | range = le32_to_cpu(rec->e_cpos) + | |
e48edee2 | 1817 | ocfs2_rec_clusters(el, rec); |
dcd0538f MF |
1818 | if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range) |
1819 | break; | |
1820 | } | |
1821 | ||
1822 | blkno = le64_to_cpu(el->l_recs[i].e_blkno); | |
1823 | if (blkno == 0) { | |
facdb77f JB |
1824 | ocfs2_error(ocfs2_metadata_cache_get_super(ci), |
1825 | "Owner %llu has bad blkno in extent list " | |
dcd0538f | 1826 | "at depth %u (index %d)\n", |
facdb77f | 1827 | (unsigned long long)ocfs2_metadata_cache_owner(ci), |
dcd0538f MF |
1828 | le16_to_cpu(el->l_tree_depth), i); |
1829 | ret = -EROFS; | |
1830 | goto out; | |
1831 | } | |
1832 | ||
1833 | brelse(bh); | |
1834 | bh = NULL; | |
facdb77f | 1835 | ret = ocfs2_read_extent_block(ci, blkno, &bh); |
dcd0538f MF |
1836 | if (ret) { |
1837 | mlog_errno(ret); | |
1838 | goto out; | |
1839 | } | |
1840 | ||
1841 | eb = (struct ocfs2_extent_block *) bh->b_data; | |
1842 | el = &eb->h_list; | |
dcd0538f MF |
1843 | |
1844 | if (le16_to_cpu(el->l_next_free_rec) > | |
1845 | le16_to_cpu(el->l_count)) { | |
facdb77f JB |
1846 | ocfs2_error(ocfs2_metadata_cache_get_super(ci), |
1847 | "Owner %llu has bad count in extent list " | |
dcd0538f | 1848 | "at block %llu (next free=%u, count=%u)\n", |
facdb77f | 1849 | (unsigned long long)ocfs2_metadata_cache_owner(ci), |
dcd0538f MF |
1850 | (unsigned long long)bh->b_blocknr, |
1851 | le16_to_cpu(el->l_next_free_rec), | |
1852 | le16_to_cpu(el->l_count)); | |
1853 | ret = -EROFS; | |
1854 | goto out; | |
1855 | } | |
1856 | ||
1857 | if (func) | |
1858 | func(data, bh); | |
1859 | } | |
1860 | ||
1861 | out: | |
1862 | /* | |
1863 | * Catch any trailing bh that the loop didn't handle. | |
1864 | */ | |
1865 | brelse(bh); | |
1866 | ||
1867 | return ret; | |
1868 | } | |
1869 | ||
1870 | /* | |
1871 | * Given an initialized path (that is, it has a valid root extent | |
1872 | * list), this function will traverse the btree in search of the path | |
1873 | * which would contain cpos. | |
1874 | * | |
1875 | * The path traveled is recorded in the path structure. | |
1876 | * | |
1877 | * Note that this will not do any comparisons on leaf node extent | |
1878 | * records, so it will work fine in the case that we just added a tree | |
1879 | * branch. | |
1880 | */ | |
1881 | struct find_path_data { | |
1882 | int index; | |
1883 | struct ocfs2_path *path; | |
1884 | }; | |
1885 | static void find_path_ins(void *data, struct buffer_head *bh) | |
1886 | { | |
1887 | struct find_path_data *fp = data; | |
1888 | ||
1889 | get_bh(bh); | |
1890 | ocfs2_path_insert_eb(fp->path, fp->index, bh); | |
1891 | fp->index++; | |
1892 | } | |
e2e9f608 TM |
1893 | int ocfs2_find_path(struct ocfs2_caching_info *ci, |
1894 | struct ocfs2_path *path, u32 cpos) | |
dcd0538f MF |
1895 | { |
1896 | struct find_path_data data; | |
1897 | ||
1898 | data.index = 1; | |
1899 | data.path = path; | |
facdb77f | 1900 | return __ocfs2_find_path(ci, path_root_el(path), cpos, |
dcd0538f MF |
1901 | find_path_ins, &data); |
1902 | } | |
1903 | ||
1904 | static void find_leaf_ins(void *data, struct buffer_head *bh) | |
1905 | { | |
1906 | struct ocfs2_extent_block *eb =(struct ocfs2_extent_block *)bh->b_data; | |
1907 | struct ocfs2_extent_list *el = &eb->h_list; | |
1908 | struct buffer_head **ret = data; | |
1909 | ||
1910 | /* We want to retain only the leaf block. */ | |
1911 | if (le16_to_cpu(el->l_tree_depth) == 0) { | |
1912 | get_bh(bh); | |
1913 | *ret = bh; | |
1914 | } | |
1915 | } | |
1916 | /* | |
1917 | * Find the leaf block in the tree which would contain cpos. No | |
1918 | * checking of the actual leaf is done. | |
1919 | * | |
1920 | * Some paths want to call this instead of allocating a path structure | |
1921 | * and calling ocfs2_find_path(). | |
1922 | * | |
1923 | * This function doesn't handle non btree extent lists. | |
1924 | */ | |
facdb77f JB |
1925 | int ocfs2_find_leaf(struct ocfs2_caching_info *ci, |
1926 | struct ocfs2_extent_list *root_el, u32 cpos, | |
1927 | struct buffer_head **leaf_bh) | |
dcd0538f MF |
1928 | { |
1929 | int ret; | |
1930 | struct buffer_head *bh = NULL; | |
1931 | ||
facdb77f | 1932 | ret = __ocfs2_find_path(ci, root_el, cpos, find_leaf_ins, &bh); |
dcd0538f MF |
1933 | if (ret) { |
1934 | mlog_errno(ret); | |
1935 | goto out; | |
1936 | } | |
1937 | ||
1938 | *leaf_bh = bh; | |
1939 | out: | |
1940 | return ret; | |
1941 | } | |
1942 | ||
1943 | /* | |
1944 | * Adjust the adjacent records (left_rec, right_rec) involved in a rotation. | |
1945 | * | |
1946 | * Basically, we've moved stuff around at the bottom of the tree and | |
1947 | * we need to fix up the extent records above the changes to reflect | |
1948 | * the new changes. | |
1949 | * | |
1950 | * left_rec: the record on the left. | |
1951 | * left_child_el: is the child list pointed to by left_rec | |
1952 | * right_rec: the record to the right of left_rec | |
1953 | * right_child_el: is the child list pointed to by right_rec | |
1954 | * | |
1955 | * By definition, this only works on interior nodes. | |
1956 | */ | |
1957 | static void ocfs2_adjust_adjacent_records(struct ocfs2_extent_rec *left_rec, | |
1958 | struct ocfs2_extent_list *left_child_el, | |
1959 | struct ocfs2_extent_rec *right_rec, | |
1960 | struct ocfs2_extent_list *right_child_el) | |
1961 | { | |
1962 | u32 left_clusters, right_end; | |
1963 | ||
1964 | /* | |
1965 | * Interior nodes never have holes. Their cpos is the cpos of | |
1966 | * the leftmost record in their child list. Their cluster | |
1967 | * count covers the full theoretical range of their child list | |
1968 | * - the range between their cpos and the cpos of the record | |
1969 | * immediately to their right. | |
1970 | */ | |
1971 | left_clusters = le32_to_cpu(right_child_el->l_recs[0].e_cpos); | |
82e12644 TM |
1972 | if (!ocfs2_rec_clusters(right_child_el, &right_child_el->l_recs[0])) { |
1973 | BUG_ON(right_child_el->l_tree_depth); | |
328d5752 MF |
1974 | BUG_ON(le16_to_cpu(right_child_el->l_next_free_rec) <= 1); |
1975 | left_clusters = le32_to_cpu(right_child_el->l_recs[1].e_cpos); | |
1976 | } | |
dcd0538f | 1977 | left_clusters -= le32_to_cpu(left_rec->e_cpos); |
e48edee2 | 1978 | left_rec->e_int_clusters = cpu_to_le32(left_clusters); |
dcd0538f MF |
1979 | |
1980 | /* | |
1981 | * Calculate the rightmost cluster count boundary before | |
e48edee2 | 1982 | * moving cpos - we will need to adjust clusters after |
dcd0538f MF |
1983 | * updating e_cpos to keep the same highest cluster count. |
1984 | */ | |
1985 | right_end = le32_to_cpu(right_rec->e_cpos); | |
e48edee2 | 1986 | right_end += le32_to_cpu(right_rec->e_int_clusters); |
dcd0538f MF |
1987 | |
1988 | right_rec->e_cpos = left_rec->e_cpos; | |
1989 | le32_add_cpu(&right_rec->e_cpos, left_clusters); | |
1990 | ||
1991 | right_end -= le32_to_cpu(right_rec->e_cpos); | |
e48edee2 | 1992 | right_rec->e_int_clusters = cpu_to_le32(right_end); |
dcd0538f MF |
1993 | } |
1994 | ||
1995 | /* | |
1996 | * Adjust the adjacent root node records involved in a | |
1997 | * rotation. left_el_blkno is passed in as a key so that we can easily | |
1998 | * find it's index in the root list. | |
1999 | */ | |
2000 | static void ocfs2_adjust_root_records(struct ocfs2_extent_list *root_el, | |
2001 | struct ocfs2_extent_list *left_el, | |
2002 | struct ocfs2_extent_list *right_el, | |
2003 | u64 left_el_blkno) | |
2004 | { | |
2005 | int i; | |
2006 | ||
2007 | BUG_ON(le16_to_cpu(root_el->l_tree_depth) <= | |
2008 | le16_to_cpu(left_el->l_tree_depth)); | |
2009 | ||
2010 | for(i = 0; i < le16_to_cpu(root_el->l_next_free_rec) - 1; i++) { | |
2011 | if (le64_to_cpu(root_el->l_recs[i].e_blkno) == left_el_blkno) | |
2012 | break; | |
2013 | } | |
2014 | ||
2015 | /* | |
2016 | * The path walking code should have never returned a root and | |
2017 | * two paths which are not adjacent. | |
2018 | */ | |
2019 | BUG_ON(i >= (le16_to_cpu(root_el->l_next_free_rec) - 1)); | |
2020 | ||
2021 | ocfs2_adjust_adjacent_records(&root_el->l_recs[i], left_el, | |
2022 | &root_el->l_recs[i + 1], right_el); | |
2023 | } | |
2024 | ||
2025 | /* | |
2026 | * We've changed a leaf block (in right_path) and need to reflect that | |
2027 | * change back up the subtree. | |
2028 | * | |
2029 | * This happens in multiple places: | |
2030 | * - When we've moved an extent record from the left path leaf to the right | |
2031 | * path leaf to make room for an empty extent in the left path leaf. | |
2032 | * - When our insert into the right path leaf is at the leftmost edge | |
2033 | * and requires an update of the path immediately to it's left. This | |
2034 | * can occur at the end of some types of rotation and appending inserts. | |
677b9752 TM |
2035 | * - When we've adjusted the last extent record in the left path leaf and the |
2036 | * 1st extent record in the right path leaf during cross extent block merge. | |
dcd0538f | 2037 | */ |
4619c73e | 2038 | static void ocfs2_complete_edge_insert(handle_t *handle, |
dcd0538f MF |
2039 | struct ocfs2_path *left_path, |
2040 | struct ocfs2_path *right_path, | |
2041 | int subtree_index) | |
2042 | { | |
ec20cec7 | 2043 | int i, idx; |
dcd0538f MF |
2044 | struct ocfs2_extent_list *el, *left_el, *right_el; |
2045 | struct ocfs2_extent_rec *left_rec, *right_rec; | |
2046 | struct buffer_head *root_bh = left_path->p_node[subtree_index].bh; | |
2047 | ||
2048 | /* | |
2049 | * Update the counts and position values within all the | |
2050 | * interior nodes to reflect the leaf rotation we just did. | |
2051 | * | |
2052 | * The root node is handled below the loop. | |
2053 | * | |
2054 | * We begin the loop with right_el and left_el pointing to the | |
2055 | * leaf lists and work our way up. | |
2056 | * | |
2057 | * NOTE: within this loop, left_el and right_el always refer | |
2058 | * to the *child* lists. | |
2059 | */ | |
2060 | left_el = path_leaf_el(left_path); | |
2061 | right_el = path_leaf_el(right_path); | |
2062 | for(i = left_path->p_tree_depth - 1; i > subtree_index; i--) { | |
2063 | mlog(0, "Adjust records at index %u\n", i); | |
2064 | ||
2065 | /* | |
2066 | * One nice property of knowing that all of these | |
2067 | * nodes are below the root is that we only deal with | |
2068 | * the leftmost right node record and the rightmost | |
2069 | * left node record. | |
2070 | */ | |
2071 | el = left_path->p_node[i].el; | |
2072 | idx = le16_to_cpu(left_el->l_next_free_rec) - 1; | |
2073 | left_rec = &el->l_recs[idx]; | |
2074 | ||
2075 | el = right_path->p_node[i].el; | |
2076 | right_rec = &el->l_recs[0]; | |
2077 | ||
2078 | ocfs2_adjust_adjacent_records(left_rec, left_el, right_rec, | |
2079 | right_el); | |
2080 | ||
ec20cec7 JB |
2081 | ocfs2_journal_dirty(handle, left_path->p_node[i].bh); |
2082 | ocfs2_journal_dirty(handle, right_path->p_node[i].bh); | |
dcd0538f MF |
2083 | |
2084 | /* | |
2085 | * Setup our list pointers now so that the current | |
2086 | * parents become children in the next iteration. | |
2087 | */ | |
2088 | left_el = left_path->p_node[i].el; | |
2089 | right_el = right_path->p_node[i].el; | |
2090 | } | |
2091 | ||
2092 | /* | |
2093 | * At the root node, adjust the two adjacent records which | |
2094 | * begin our path to the leaves. | |
2095 | */ | |
2096 | ||
2097 | el = left_path->p_node[subtree_index].el; | |
2098 | left_el = left_path->p_node[subtree_index + 1].el; | |
2099 | right_el = right_path->p_node[subtree_index + 1].el; | |
2100 | ||
2101 | ocfs2_adjust_root_records(el, left_el, right_el, | |
2102 | left_path->p_node[subtree_index + 1].bh->b_blocknr); | |
2103 | ||
2104 | root_bh = left_path->p_node[subtree_index].bh; | |
2105 | ||
ec20cec7 | 2106 | ocfs2_journal_dirty(handle, root_bh); |
dcd0538f MF |
2107 | } |
2108 | ||
5c601aba JB |
2109 | static int ocfs2_rotate_subtree_right(handle_t *handle, |
2110 | struct ocfs2_extent_tree *et, | |
dcd0538f MF |
2111 | struct ocfs2_path *left_path, |
2112 | struct ocfs2_path *right_path, | |
2113 | int subtree_index) | |
2114 | { | |
2115 | int ret, i; | |
2116 | struct buffer_head *right_leaf_bh; | |
2117 | struct buffer_head *left_leaf_bh = NULL; | |
2118 | struct buffer_head *root_bh; | |
2119 | struct ocfs2_extent_list *right_el, *left_el; | |
2120 | struct ocfs2_extent_rec move_rec; | |
2121 | ||
2122 | left_leaf_bh = path_leaf_bh(left_path); | |
2123 | left_el = path_leaf_el(left_path); | |
2124 | ||
2125 | if (left_el->l_next_free_rec != left_el->l_count) { | |
5c601aba | 2126 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
dcd0538f MF |
2127 | "Inode %llu has non-full interior leaf node %llu" |
2128 | "(next free = %u)", | |
5c601aba | 2129 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
dcd0538f MF |
2130 | (unsigned long long)left_leaf_bh->b_blocknr, |
2131 | le16_to_cpu(left_el->l_next_free_rec)); | |
2132 | return -EROFS; | |
2133 | } | |
2134 | ||
2135 | /* | |
2136 | * This extent block may already have an empty record, so we | |
2137 | * return early if so. | |
2138 | */ | |
2139 | if (ocfs2_is_empty_extent(&left_el->l_recs[0])) | |
2140 | return 0; | |
2141 | ||
2142 | root_bh = left_path->p_node[subtree_index].bh; | |
2143 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
2144 | ||
5c601aba | 2145 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path, |
13723d00 | 2146 | subtree_index); |
dcd0538f MF |
2147 | if (ret) { |
2148 | mlog_errno(ret); | |
2149 | goto out; | |
2150 | } | |
2151 | ||
2152 | for(i = subtree_index + 1; i < path_num_items(right_path); i++) { | |
5c601aba | 2153 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 2154 | right_path, i); |
dcd0538f MF |
2155 | if (ret) { |
2156 | mlog_errno(ret); | |
2157 | goto out; | |
2158 | } | |
2159 | ||
5c601aba | 2160 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 2161 | left_path, i); |
dcd0538f MF |
2162 | if (ret) { |
2163 | mlog_errno(ret); | |
2164 | goto out; | |
2165 | } | |
2166 | } | |
2167 | ||
2168 | right_leaf_bh = path_leaf_bh(right_path); | |
2169 | right_el = path_leaf_el(right_path); | |
2170 | ||
2171 | /* This is a code error, not a disk corruption. */ | |
2172 | mlog_bug_on_msg(!right_el->l_next_free_rec, "Inode %llu: Rotate fails " | |
2173 | "because rightmost leaf block %llu is empty\n", | |
5c601aba | 2174 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
dcd0538f MF |
2175 | (unsigned long long)right_leaf_bh->b_blocknr); |
2176 | ||
2177 | ocfs2_create_empty_extent(right_el); | |
2178 | ||
ec20cec7 | 2179 | ocfs2_journal_dirty(handle, right_leaf_bh); |
dcd0538f MF |
2180 | |
2181 | /* Do the copy now. */ | |
2182 | i = le16_to_cpu(left_el->l_next_free_rec) - 1; | |
2183 | move_rec = left_el->l_recs[i]; | |
2184 | right_el->l_recs[0] = move_rec; | |
2185 | ||
2186 | /* | |
2187 | * Clear out the record we just copied and shift everything | |
2188 | * over, leaving an empty extent in the left leaf. | |
2189 | * | |
2190 | * We temporarily subtract from next_free_rec so that the | |
2191 | * shift will lose the tail record (which is now defunct). | |
2192 | */ | |
2193 | le16_add_cpu(&left_el->l_next_free_rec, -1); | |
2194 | ocfs2_shift_records_right(left_el); | |
2195 | memset(&left_el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
2196 | le16_add_cpu(&left_el->l_next_free_rec, 1); | |
2197 | ||
ec20cec7 | 2198 | ocfs2_journal_dirty(handle, left_leaf_bh); |
dcd0538f | 2199 | |
4619c73e JB |
2200 | ocfs2_complete_edge_insert(handle, left_path, right_path, |
2201 | subtree_index); | |
dcd0538f MF |
2202 | |
2203 | out: | |
2204 | return ret; | |
2205 | } | |
2206 | ||
2207 | /* | |
2208 | * Given a full path, determine what cpos value would return us a path | |
2209 | * containing the leaf immediately to the left of the current one. | |
2210 | * | |
2211 | * Will return zero if the path passed in is already the leftmost path. | |
2212 | */ | |
ee149a7c TY |
2213 | int ocfs2_find_cpos_for_left_leaf(struct super_block *sb, |
2214 | struct ocfs2_path *path, u32 *cpos) | |
dcd0538f MF |
2215 | { |
2216 | int i, j, ret = 0; | |
2217 | u64 blkno; | |
2218 | struct ocfs2_extent_list *el; | |
2219 | ||
e48edee2 MF |
2220 | BUG_ON(path->p_tree_depth == 0); |
2221 | ||
dcd0538f MF |
2222 | *cpos = 0; |
2223 | ||
2224 | blkno = path_leaf_bh(path)->b_blocknr; | |
2225 | ||
2226 | /* Start at the tree node just above the leaf and work our way up. */ | |
2227 | i = path->p_tree_depth - 1; | |
2228 | while (i >= 0) { | |
2229 | el = path->p_node[i].el; | |
2230 | ||
2231 | /* | |
2232 | * Find the extent record just before the one in our | |
2233 | * path. | |
2234 | */ | |
2235 | for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) { | |
2236 | if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) { | |
2237 | if (j == 0) { | |
2238 | if (i == 0) { | |
2239 | /* | |
2240 | * We've determined that the | |
2241 | * path specified is already | |
2242 | * the leftmost one - return a | |
2243 | * cpos of zero. | |
2244 | */ | |
2245 | goto out; | |
2246 | } | |
2247 | /* | |
2248 | * The leftmost record points to our | |
2249 | * leaf - we need to travel up the | |
2250 | * tree one level. | |
2251 | */ | |
2252 | goto next_node; | |
2253 | } | |
2254 | ||
2255 | *cpos = le32_to_cpu(el->l_recs[j - 1].e_cpos); | |
e48edee2 MF |
2256 | *cpos = *cpos + ocfs2_rec_clusters(el, |
2257 | &el->l_recs[j - 1]); | |
2258 | *cpos = *cpos - 1; | |
dcd0538f MF |
2259 | goto out; |
2260 | } | |
2261 | } | |
2262 | ||
2263 | /* | |
2264 | * If we got here, we never found a valid node where | |
2265 | * the tree indicated one should be. | |
2266 | */ | |
2267 | ocfs2_error(sb, | |
2268 | "Invalid extent tree at extent block %llu\n", | |
2269 | (unsigned long long)blkno); | |
2270 | ret = -EROFS; | |
2271 | goto out; | |
2272 | ||
2273 | next_node: | |
2274 | blkno = path->p_node[i].bh->b_blocknr; | |
2275 | i--; | |
2276 | } | |
2277 | ||
2278 | out: | |
2279 | return ret; | |
2280 | } | |
2281 | ||
328d5752 MF |
2282 | /* |
2283 | * Extend the transaction by enough credits to complete the rotation, | |
2284 | * and still leave at least the original number of credits allocated | |
2285 | * to this transaction. | |
2286 | */ | |
dcd0538f | 2287 | static int ocfs2_extend_rotate_transaction(handle_t *handle, int subtree_depth, |
328d5752 | 2288 | int op_credits, |
dcd0538f MF |
2289 | struct ocfs2_path *path) |
2290 | { | |
c901fb00 | 2291 | int ret = 0; |
328d5752 | 2292 | int credits = (path->p_tree_depth - subtree_depth) * 2 + 1 + op_credits; |
dcd0538f | 2293 | |
c901fb00 | 2294 | if (handle->h_buffer_credits < credits) |
c18b812d TM |
2295 | ret = ocfs2_extend_trans(handle, |
2296 | credits - handle->h_buffer_credits); | |
dcd0538f | 2297 | |
c901fb00 | 2298 | return ret; |
dcd0538f MF |
2299 | } |
2300 | ||
2301 | /* | |
2302 | * Trap the case where we're inserting into the theoretical range past | |
2303 | * the _actual_ left leaf range. Otherwise, we'll rotate a record | |
2304 | * whose cpos is less than ours into the right leaf. | |
2305 | * | |
2306 | * It's only necessary to look at the rightmost record of the left | |
2307 | * leaf because the logic that calls us should ensure that the | |
2308 | * theoretical ranges in the path components above the leaves are | |
2309 | * correct. | |
2310 | */ | |
2311 | static int ocfs2_rotate_requires_path_adjustment(struct ocfs2_path *left_path, | |
2312 | u32 insert_cpos) | |
2313 | { | |
2314 | struct ocfs2_extent_list *left_el; | |
2315 | struct ocfs2_extent_rec *rec; | |
2316 | int next_free; | |
2317 | ||
2318 | left_el = path_leaf_el(left_path); | |
2319 | next_free = le16_to_cpu(left_el->l_next_free_rec); | |
2320 | rec = &left_el->l_recs[next_free - 1]; | |
2321 | ||
2322 | if (insert_cpos > le32_to_cpu(rec->e_cpos)) | |
2323 | return 1; | |
2324 | return 0; | |
2325 | } | |
2326 | ||
328d5752 MF |
2327 | static int ocfs2_leftmost_rec_contains(struct ocfs2_extent_list *el, u32 cpos) |
2328 | { | |
2329 | int next_free = le16_to_cpu(el->l_next_free_rec); | |
2330 | unsigned int range; | |
2331 | struct ocfs2_extent_rec *rec; | |
2332 | ||
2333 | if (next_free == 0) | |
2334 | return 0; | |
2335 | ||
2336 | rec = &el->l_recs[0]; | |
2337 | if (ocfs2_is_empty_extent(rec)) { | |
2338 | /* Empty list. */ | |
2339 | if (next_free == 1) | |
2340 | return 0; | |
2341 | rec = &el->l_recs[1]; | |
2342 | } | |
2343 | ||
2344 | range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
2345 | if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range) | |
2346 | return 1; | |
2347 | return 0; | |
2348 | } | |
2349 | ||
dcd0538f MF |
2350 | /* |
2351 | * Rotate all the records in a btree right one record, starting at insert_cpos. | |
2352 | * | |
2353 | * The path to the rightmost leaf should be passed in. | |
2354 | * | |
2355 | * The array is assumed to be large enough to hold an entire path (tree depth). | |
2356 | * | |
af901ca1 | 2357 | * Upon successful return from this function: |
dcd0538f MF |
2358 | * |
2359 | * - The 'right_path' array will contain a path to the leaf block | |
2360 | * whose range contains e_cpos. | |
2361 | * - That leaf block will have a single empty extent in list index 0. | |
2362 | * - In the case that the rotation requires a post-insert update, | |
2363 | * *ret_left_path will contain a valid path which can be passed to | |
2364 | * ocfs2_insert_path(). | |
2365 | */ | |
1bbf0b8d | 2366 | static int ocfs2_rotate_tree_right(handle_t *handle, |
5c601aba | 2367 | struct ocfs2_extent_tree *et, |
328d5752 | 2368 | enum ocfs2_split_type split, |
dcd0538f MF |
2369 | u32 insert_cpos, |
2370 | struct ocfs2_path *right_path, | |
2371 | struct ocfs2_path **ret_left_path) | |
2372 | { | |
328d5752 | 2373 | int ret, start, orig_credits = handle->h_buffer_credits; |
dcd0538f MF |
2374 | u32 cpos; |
2375 | struct ocfs2_path *left_path = NULL; | |
5c601aba | 2376 | struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci); |
dcd0538f MF |
2377 | |
2378 | *ret_left_path = NULL; | |
2379 | ||
ffdd7a54 | 2380 | left_path = ocfs2_new_path_from_path(right_path); |
dcd0538f MF |
2381 | if (!left_path) { |
2382 | ret = -ENOMEM; | |
2383 | mlog_errno(ret); | |
2384 | goto out; | |
2385 | } | |
2386 | ||
5c601aba | 2387 | ret = ocfs2_find_cpos_for_left_leaf(sb, right_path, &cpos); |
dcd0538f MF |
2388 | if (ret) { |
2389 | mlog_errno(ret); | |
2390 | goto out; | |
2391 | } | |
2392 | ||
2393 | mlog(0, "Insert: %u, first left path cpos: %u\n", insert_cpos, cpos); | |
2394 | ||
2395 | /* | |
2396 | * What we want to do here is: | |
2397 | * | |
2398 | * 1) Start with the rightmost path. | |
2399 | * | |
2400 | * 2) Determine a path to the leaf block directly to the left | |
2401 | * of that leaf. | |
2402 | * | |
2403 | * 3) Determine the 'subtree root' - the lowest level tree node | |
2404 | * which contains a path to both leaves. | |
2405 | * | |
2406 | * 4) Rotate the subtree. | |
2407 | * | |
2408 | * 5) Find the next subtree by considering the left path to be | |
2409 | * the new right path. | |
2410 | * | |
2411 | * The check at the top of this while loop also accepts | |
2412 | * insert_cpos == cpos because cpos is only a _theoretical_ | |
2413 | * value to get us the left path - insert_cpos might very well | |
2414 | * be filling that hole. | |
2415 | * | |
2416 | * Stop at a cpos of '0' because we either started at the | |
2417 | * leftmost branch (i.e., a tree with one branch and a | |
2418 | * rotation inside of it), or we've gone as far as we can in | |
2419 | * rotating subtrees. | |
2420 | */ | |
2421 | while (cpos && insert_cpos <= cpos) { | |
2422 | mlog(0, "Rotating a tree: ins. cpos: %u, left path cpos: %u\n", | |
2423 | insert_cpos, cpos); | |
2424 | ||
5c601aba | 2425 | ret = ocfs2_find_path(et->et_ci, left_path, cpos); |
dcd0538f MF |
2426 | if (ret) { |
2427 | mlog_errno(ret); | |
2428 | goto out; | |
2429 | } | |
2430 | ||
2431 | mlog_bug_on_msg(path_leaf_bh(left_path) == | |
2432 | path_leaf_bh(right_path), | |
5c601aba | 2433 | "Owner %llu: error during insert of %u " |
dcd0538f MF |
2434 | "(left path cpos %u) results in two identical " |
2435 | "paths ending at %llu\n", | |
5c601aba JB |
2436 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
2437 | insert_cpos, cpos, | |
dcd0538f MF |
2438 | (unsigned long long) |
2439 | path_leaf_bh(left_path)->b_blocknr); | |
2440 | ||
328d5752 MF |
2441 | if (split == SPLIT_NONE && |
2442 | ocfs2_rotate_requires_path_adjustment(left_path, | |
dcd0538f | 2443 | insert_cpos)) { |
dcd0538f MF |
2444 | |
2445 | /* | |
2446 | * We've rotated the tree as much as we | |
2447 | * should. The rest is up to | |
2448 | * ocfs2_insert_path() to complete, after the | |
2449 | * record insertion. We indicate this | |
2450 | * situation by returning the left path. | |
2451 | * | |
2452 | * The reason we don't adjust the records here | |
2453 | * before the record insert is that an error | |
2454 | * later might break the rule where a parent | |
2455 | * record e_cpos will reflect the actual | |
2456 | * e_cpos of the 1st nonempty record of the | |
2457 | * child list. | |
2458 | */ | |
2459 | *ret_left_path = left_path; | |
2460 | goto out_ret_path; | |
2461 | } | |
2462 | ||
7dc02805 | 2463 | start = ocfs2_find_subtree_root(et, left_path, right_path); |
dcd0538f MF |
2464 | |
2465 | mlog(0, "Subtree root at index %d (blk %llu, depth %d)\n", | |
2466 | start, | |
2467 | (unsigned long long) right_path->p_node[start].bh->b_blocknr, | |
2468 | right_path->p_tree_depth); | |
2469 | ||
2470 | ret = ocfs2_extend_rotate_transaction(handle, start, | |
328d5752 | 2471 | orig_credits, right_path); |
dcd0538f MF |
2472 | if (ret) { |
2473 | mlog_errno(ret); | |
2474 | goto out; | |
2475 | } | |
2476 | ||
5c601aba | 2477 | ret = ocfs2_rotate_subtree_right(handle, et, left_path, |
dcd0538f MF |
2478 | right_path, start); |
2479 | if (ret) { | |
2480 | mlog_errno(ret); | |
2481 | goto out; | |
2482 | } | |
2483 | ||
328d5752 MF |
2484 | if (split != SPLIT_NONE && |
2485 | ocfs2_leftmost_rec_contains(path_leaf_el(right_path), | |
2486 | insert_cpos)) { | |
2487 | /* | |
2488 | * A rotate moves the rightmost left leaf | |
2489 | * record over to the leftmost right leaf | |
2490 | * slot. If we're doing an extent split | |
2491 | * instead of a real insert, then we have to | |
2492 | * check that the extent to be split wasn't | |
2493 | * just moved over. If it was, then we can | |
2494 | * exit here, passing left_path back - | |
2495 | * ocfs2_split_extent() is smart enough to | |
2496 | * search both leaves. | |
2497 | */ | |
2498 | *ret_left_path = left_path; | |
2499 | goto out_ret_path; | |
2500 | } | |
2501 | ||
dcd0538f MF |
2502 | /* |
2503 | * There is no need to re-read the next right path | |
2504 | * as we know that it'll be our current left | |
2505 | * path. Optimize by copying values instead. | |
2506 | */ | |
2507 | ocfs2_mv_path(right_path, left_path); | |
2508 | ||
5c601aba | 2509 | ret = ocfs2_find_cpos_for_left_leaf(sb, right_path, &cpos); |
dcd0538f MF |
2510 | if (ret) { |
2511 | mlog_errno(ret); | |
2512 | goto out; | |
2513 | } | |
2514 | } | |
2515 | ||
2516 | out: | |
2517 | ocfs2_free_path(left_path); | |
2518 | ||
2519 | out_ret_path: | |
2520 | return ret; | |
2521 | } | |
2522 | ||
09106bae JB |
2523 | static int ocfs2_update_edge_lengths(handle_t *handle, |
2524 | struct ocfs2_extent_tree *et, | |
3c5e1068 | 2525 | int subtree_index, struct ocfs2_path *path) |
dcd0538f | 2526 | { |
3c5e1068 | 2527 | int i, idx, ret; |
dcd0538f | 2528 | struct ocfs2_extent_rec *rec; |
328d5752 MF |
2529 | struct ocfs2_extent_list *el; |
2530 | struct ocfs2_extent_block *eb; | |
2531 | u32 range; | |
dcd0538f | 2532 | |
3c5e1068 TM |
2533 | /* |
2534 | * In normal tree rotation process, we will never touch the | |
2535 | * tree branch above subtree_index and ocfs2_extend_rotate_transaction | |
2536 | * doesn't reserve the credits for them either. | |
2537 | * | |
2538 | * But we do have a special case here which will update the rightmost | |
2539 | * records for all the bh in the path. | |
2540 | * So we have to allocate extra credits and access them. | |
2541 | */ | |
c901fb00 | 2542 | ret = ocfs2_extend_trans(handle, subtree_index); |
3c5e1068 TM |
2543 | if (ret) { |
2544 | mlog_errno(ret); | |
2545 | goto out; | |
2546 | } | |
2547 | ||
09106bae | 2548 | ret = ocfs2_journal_access_path(et->et_ci, handle, path); |
3c5e1068 TM |
2549 | if (ret) { |
2550 | mlog_errno(ret); | |
2551 | goto out; | |
2552 | } | |
2553 | ||
328d5752 MF |
2554 | /* Path should always be rightmost. */ |
2555 | eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data; | |
2556 | BUG_ON(eb->h_next_leaf_blk != 0ULL); | |
dcd0538f | 2557 | |
328d5752 MF |
2558 | el = &eb->h_list; |
2559 | BUG_ON(le16_to_cpu(el->l_next_free_rec) == 0); | |
2560 | idx = le16_to_cpu(el->l_next_free_rec) - 1; | |
2561 | rec = &el->l_recs[idx]; | |
2562 | range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
dcd0538f | 2563 | |
328d5752 MF |
2564 | for (i = 0; i < path->p_tree_depth; i++) { |
2565 | el = path->p_node[i].el; | |
2566 | idx = le16_to_cpu(el->l_next_free_rec) - 1; | |
2567 | rec = &el->l_recs[idx]; | |
dcd0538f | 2568 | |
328d5752 MF |
2569 | rec->e_int_clusters = cpu_to_le32(range); |
2570 | le32_add_cpu(&rec->e_int_clusters, -le32_to_cpu(rec->e_cpos)); | |
dcd0538f | 2571 | |
328d5752 | 2572 | ocfs2_journal_dirty(handle, path->p_node[i].bh); |
dcd0538f | 2573 | } |
3c5e1068 TM |
2574 | out: |
2575 | return ret; | |
dcd0538f MF |
2576 | } |
2577 | ||
6641b0ce JB |
2578 | static void ocfs2_unlink_path(handle_t *handle, |
2579 | struct ocfs2_extent_tree *et, | |
328d5752 MF |
2580 | struct ocfs2_cached_dealloc_ctxt *dealloc, |
2581 | struct ocfs2_path *path, int unlink_start) | |
dcd0538f | 2582 | { |
328d5752 MF |
2583 | int ret, i; |
2584 | struct ocfs2_extent_block *eb; | |
2585 | struct ocfs2_extent_list *el; | |
2586 | struct buffer_head *bh; | |
2587 | ||
2588 | for(i = unlink_start; i < path_num_items(path); i++) { | |
2589 | bh = path->p_node[i].bh; | |
2590 | ||
2591 | eb = (struct ocfs2_extent_block *)bh->b_data; | |
2592 | /* | |
2593 | * Not all nodes might have had their final count | |
2594 | * decremented by the caller - handle this here. | |
2595 | */ | |
2596 | el = &eb->h_list; | |
2597 | if (le16_to_cpu(el->l_next_free_rec) > 1) { | |
2598 | mlog(ML_ERROR, | |
2599 | "Inode %llu, attempted to remove extent block " | |
2600 | "%llu with %u records\n", | |
6641b0ce | 2601 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
328d5752 MF |
2602 | (unsigned long long)le64_to_cpu(eb->h_blkno), |
2603 | le16_to_cpu(el->l_next_free_rec)); | |
2604 | ||
2605 | ocfs2_journal_dirty(handle, bh); | |
6641b0ce | 2606 | ocfs2_remove_from_cache(et->et_ci, bh); |
328d5752 MF |
2607 | continue; |
2608 | } | |
2609 | ||
2610 | el->l_next_free_rec = 0; | |
2611 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
2612 | ||
2613 | ocfs2_journal_dirty(handle, bh); | |
2614 | ||
2615 | ret = ocfs2_cache_extent_block_free(dealloc, eb); | |
2616 | if (ret) | |
2617 | mlog_errno(ret); | |
2618 | ||
6641b0ce | 2619 | ocfs2_remove_from_cache(et->et_ci, bh); |
328d5752 | 2620 | } |
dcd0538f MF |
2621 | } |
2622 | ||
6641b0ce JB |
2623 | static void ocfs2_unlink_subtree(handle_t *handle, |
2624 | struct ocfs2_extent_tree *et, | |
328d5752 MF |
2625 | struct ocfs2_path *left_path, |
2626 | struct ocfs2_path *right_path, | |
2627 | int subtree_index, | |
2628 | struct ocfs2_cached_dealloc_ctxt *dealloc) | |
dcd0538f | 2629 | { |
328d5752 MF |
2630 | int i; |
2631 | struct buffer_head *root_bh = left_path->p_node[subtree_index].bh; | |
2632 | struct ocfs2_extent_list *root_el = left_path->p_node[subtree_index].el; | |
dcd0538f | 2633 | struct ocfs2_extent_list *el; |
328d5752 | 2634 | struct ocfs2_extent_block *eb; |
dcd0538f | 2635 | |
328d5752 | 2636 | el = path_leaf_el(left_path); |
dcd0538f | 2637 | |
328d5752 | 2638 | eb = (struct ocfs2_extent_block *)right_path->p_node[subtree_index + 1].bh->b_data; |
e48edee2 | 2639 | |
328d5752 MF |
2640 | for(i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++) |
2641 | if (root_el->l_recs[i].e_blkno == eb->h_blkno) | |
2642 | break; | |
dcd0538f | 2643 | |
328d5752 | 2644 | BUG_ON(i >= le16_to_cpu(root_el->l_next_free_rec)); |
dcd0538f | 2645 | |
328d5752 MF |
2646 | memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec)); |
2647 | le16_add_cpu(&root_el->l_next_free_rec, -1); | |
2648 | ||
2649 | eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data; | |
2650 | eb->h_next_leaf_blk = 0; | |
2651 | ||
2652 | ocfs2_journal_dirty(handle, root_bh); | |
2653 | ocfs2_journal_dirty(handle, path_leaf_bh(left_path)); | |
2654 | ||
6641b0ce | 2655 | ocfs2_unlink_path(handle, et, dealloc, right_path, |
328d5752 MF |
2656 | subtree_index + 1); |
2657 | } | |
2658 | ||
1e2dd63f JB |
2659 | static int ocfs2_rotate_subtree_left(handle_t *handle, |
2660 | struct ocfs2_extent_tree *et, | |
328d5752 MF |
2661 | struct ocfs2_path *left_path, |
2662 | struct ocfs2_path *right_path, | |
2663 | int subtree_index, | |
2664 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
1e2dd63f | 2665 | int *deleted) |
328d5752 MF |
2666 | { |
2667 | int ret, i, del_right_subtree = 0, right_has_empty = 0; | |
e7d4cb6b | 2668 | struct buffer_head *root_bh, *et_root_bh = path_root_bh(right_path); |
328d5752 MF |
2669 | struct ocfs2_extent_list *right_leaf_el, *left_leaf_el; |
2670 | struct ocfs2_extent_block *eb; | |
2671 | ||
2672 | *deleted = 0; | |
2673 | ||
2674 | right_leaf_el = path_leaf_el(right_path); | |
2675 | left_leaf_el = path_leaf_el(left_path); | |
2676 | root_bh = left_path->p_node[subtree_index].bh; | |
2677 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
2678 | ||
2679 | if (!ocfs2_is_empty_extent(&left_leaf_el->l_recs[0])) | |
2680 | return 0; | |
dcd0538f | 2681 | |
328d5752 MF |
2682 | eb = (struct ocfs2_extent_block *)path_leaf_bh(right_path)->b_data; |
2683 | if (ocfs2_is_empty_extent(&right_leaf_el->l_recs[0])) { | |
dcd0538f | 2684 | /* |
328d5752 MF |
2685 | * It's legal for us to proceed if the right leaf is |
2686 | * the rightmost one and it has an empty extent. There | |
2687 | * are two cases to handle - whether the leaf will be | |
2688 | * empty after removal or not. If the leaf isn't empty | |
2689 | * then just remove the empty extent up front. The | |
2690 | * next block will handle empty leaves by flagging | |
2691 | * them for unlink. | |
2692 | * | |
2693 | * Non rightmost leaves will throw -EAGAIN and the | |
2694 | * caller can manually move the subtree and retry. | |
dcd0538f | 2695 | */ |
dcd0538f | 2696 | |
328d5752 MF |
2697 | if (eb->h_next_leaf_blk != 0ULL) |
2698 | return -EAGAIN; | |
2699 | ||
2700 | if (le16_to_cpu(right_leaf_el->l_next_free_rec) > 1) { | |
1e2dd63f | 2701 | ret = ocfs2_journal_access_eb(handle, et->et_ci, |
13723d00 JB |
2702 | path_leaf_bh(right_path), |
2703 | OCFS2_JOURNAL_ACCESS_WRITE); | |
dcd0538f MF |
2704 | if (ret) { |
2705 | mlog_errno(ret); | |
2706 | goto out; | |
2707 | } | |
2708 | ||
328d5752 MF |
2709 | ocfs2_remove_empty_extent(right_leaf_el); |
2710 | } else | |
2711 | right_has_empty = 1; | |
dcd0538f MF |
2712 | } |
2713 | ||
328d5752 MF |
2714 | if (eb->h_next_leaf_blk == 0ULL && |
2715 | le16_to_cpu(right_leaf_el->l_next_free_rec) == 1) { | |
2716 | /* | |
2717 | * We have to update i_last_eb_blk during the meta | |
2718 | * data delete. | |
2719 | */ | |
d9a0a1f8 | 2720 | ret = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 2721 | OCFS2_JOURNAL_ACCESS_WRITE); |
328d5752 MF |
2722 | if (ret) { |
2723 | mlog_errno(ret); | |
2724 | goto out; | |
2725 | } | |
2726 | ||
2727 | del_right_subtree = 1; | |
2728 | } | |
2729 | ||
2730 | /* | |
2731 | * Getting here with an empty extent in the right path implies | |
2732 | * that it's the rightmost path and will be deleted. | |
2733 | */ | |
2734 | BUG_ON(right_has_empty && !del_right_subtree); | |
2735 | ||
1e2dd63f | 2736 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path, |
13723d00 | 2737 | subtree_index); |
328d5752 MF |
2738 | if (ret) { |
2739 | mlog_errno(ret); | |
2740 | goto out; | |
2741 | } | |
2742 | ||
2743 | for(i = subtree_index + 1; i < path_num_items(right_path); i++) { | |
1e2dd63f | 2744 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 2745 | right_path, i); |
328d5752 MF |
2746 | if (ret) { |
2747 | mlog_errno(ret); | |
2748 | goto out; | |
2749 | } | |
2750 | ||
1e2dd63f | 2751 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 2752 | left_path, i); |
328d5752 MF |
2753 | if (ret) { |
2754 | mlog_errno(ret); | |
2755 | goto out; | |
2756 | } | |
2757 | } | |
2758 | ||
2759 | if (!right_has_empty) { | |
2760 | /* | |
2761 | * Only do this if we're moving a real | |
2762 | * record. Otherwise, the action is delayed until | |
2763 | * after removal of the right path in which case we | |
2764 | * can do a simple shift to remove the empty extent. | |
2765 | */ | |
2766 | ocfs2_rotate_leaf(left_leaf_el, &right_leaf_el->l_recs[0]); | |
2767 | memset(&right_leaf_el->l_recs[0], 0, | |
2768 | sizeof(struct ocfs2_extent_rec)); | |
2769 | } | |
2770 | if (eb->h_next_leaf_blk == 0ULL) { | |
2771 | /* | |
2772 | * Move recs over to get rid of empty extent, decrease | |
2773 | * next_free. This is allowed to remove the last | |
2774 | * extent in our leaf (setting l_next_free_rec to | |
2775 | * zero) - the delete code below won't care. | |
2776 | */ | |
2777 | ocfs2_remove_empty_extent(right_leaf_el); | |
2778 | } | |
2779 | ||
ec20cec7 JB |
2780 | ocfs2_journal_dirty(handle, path_leaf_bh(left_path)); |
2781 | ocfs2_journal_dirty(handle, path_leaf_bh(right_path)); | |
328d5752 MF |
2782 | |
2783 | if (del_right_subtree) { | |
6641b0ce | 2784 | ocfs2_unlink_subtree(handle, et, left_path, right_path, |
328d5752 | 2785 | subtree_index, dealloc); |
09106bae | 2786 | ret = ocfs2_update_edge_lengths(handle, et, subtree_index, |
3c5e1068 TM |
2787 | left_path); |
2788 | if (ret) { | |
2789 | mlog_errno(ret); | |
2790 | goto out; | |
2791 | } | |
328d5752 MF |
2792 | |
2793 | eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data; | |
35dc0aa3 | 2794 | ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno)); |
328d5752 MF |
2795 | |
2796 | /* | |
2797 | * Removal of the extent in the left leaf was skipped | |
2798 | * above so we could delete the right path | |
2799 | * 1st. | |
2800 | */ | |
2801 | if (right_has_empty) | |
2802 | ocfs2_remove_empty_extent(left_leaf_el); | |
2803 | ||
ec20cec7 | 2804 | ocfs2_journal_dirty(handle, et_root_bh); |
328d5752 MF |
2805 | |
2806 | *deleted = 1; | |
2807 | } else | |
4619c73e | 2808 | ocfs2_complete_edge_insert(handle, left_path, right_path, |
328d5752 MF |
2809 | subtree_index); |
2810 | ||
2811 | out: | |
2812 | return ret; | |
2813 | } | |
2814 | ||
2815 | /* | |
2816 | * Given a full path, determine what cpos value would return us a path | |
2817 | * containing the leaf immediately to the right of the current one. | |
2818 | * | |
2819 | * Will return zero if the path passed in is already the rightmost path. | |
2820 | * | |
2821 | * This looks similar, but is subtly different to | |
2822 | * ocfs2_find_cpos_for_left_leaf(). | |
2823 | */ | |
38a04e43 TM |
2824 | int ocfs2_find_cpos_for_right_leaf(struct super_block *sb, |
2825 | struct ocfs2_path *path, u32 *cpos) | |
328d5752 MF |
2826 | { |
2827 | int i, j, ret = 0; | |
2828 | u64 blkno; | |
2829 | struct ocfs2_extent_list *el; | |
2830 | ||
2831 | *cpos = 0; | |
2832 | ||
2833 | if (path->p_tree_depth == 0) | |
2834 | return 0; | |
2835 | ||
2836 | blkno = path_leaf_bh(path)->b_blocknr; | |
2837 | ||
2838 | /* Start at the tree node just above the leaf and work our way up. */ | |
2839 | i = path->p_tree_depth - 1; | |
2840 | while (i >= 0) { | |
2841 | int next_free; | |
2842 | ||
2843 | el = path->p_node[i].el; | |
2844 | ||
2845 | /* | |
2846 | * Find the extent record just after the one in our | |
2847 | * path. | |
2848 | */ | |
2849 | next_free = le16_to_cpu(el->l_next_free_rec); | |
2850 | for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) { | |
2851 | if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) { | |
2852 | if (j == (next_free - 1)) { | |
2853 | if (i == 0) { | |
2854 | /* | |
2855 | * We've determined that the | |
2856 | * path specified is already | |
2857 | * the rightmost one - return a | |
2858 | * cpos of zero. | |
2859 | */ | |
2860 | goto out; | |
2861 | } | |
2862 | /* | |
2863 | * The rightmost record points to our | |
2864 | * leaf - we need to travel up the | |
2865 | * tree one level. | |
2866 | */ | |
2867 | goto next_node; | |
2868 | } | |
2869 | ||
2870 | *cpos = le32_to_cpu(el->l_recs[j + 1].e_cpos); | |
2871 | goto out; | |
2872 | } | |
2873 | } | |
2874 | ||
2875 | /* | |
2876 | * If we got here, we never found a valid node where | |
2877 | * the tree indicated one should be. | |
2878 | */ | |
2879 | ocfs2_error(sb, | |
2880 | "Invalid extent tree at extent block %llu\n", | |
2881 | (unsigned long long)blkno); | |
2882 | ret = -EROFS; | |
2883 | goto out; | |
2884 | ||
2885 | next_node: | |
2886 | blkno = path->p_node[i].bh->b_blocknr; | |
2887 | i--; | |
2888 | } | |
2889 | ||
2890 | out: | |
2891 | return ret; | |
2892 | } | |
2893 | ||
70f18c08 JB |
2894 | static int ocfs2_rotate_rightmost_leaf_left(handle_t *handle, |
2895 | struct ocfs2_extent_tree *et, | |
13723d00 | 2896 | struct ocfs2_path *path) |
328d5752 MF |
2897 | { |
2898 | int ret; | |
13723d00 JB |
2899 | struct buffer_head *bh = path_leaf_bh(path); |
2900 | struct ocfs2_extent_list *el = path_leaf_el(path); | |
328d5752 MF |
2901 | |
2902 | if (!ocfs2_is_empty_extent(&el->l_recs[0])) | |
2903 | return 0; | |
2904 | ||
70f18c08 | 2905 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, path, |
13723d00 | 2906 | path_num_items(path) - 1); |
328d5752 MF |
2907 | if (ret) { |
2908 | mlog_errno(ret); | |
2909 | goto out; | |
2910 | } | |
2911 | ||
2912 | ocfs2_remove_empty_extent(el); | |
ec20cec7 | 2913 | ocfs2_journal_dirty(handle, bh); |
328d5752 MF |
2914 | |
2915 | out: | |
2916 | return ret; | |
2917 | } | |
2918 | ||
e46f74dc JB |
2919 | static int __ocfs2_rotate_tree_left(handle_t *handle, |
2920 | struct ocfs2_extent_tree *et, | |
2921 | int orig_credits, | |
328d5752 MF |
2922 | struct ocfs2_path *path, |
2923 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
e46f74dc | 2924 | struct ocfs2_path **empty_extent_path) |
328d5752 MF |
2925 | { |
2926 | int ret, subtree_root, deleted; | |
2927 | u32 right_cpos; | |
2928 | struct ocfs2_path *left_path = NULL; | |
2929 | struct ocfs2_path *right_path = NULL; | |
e46f74dc | 2930 | struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci); |
328d5752 MF |
2931 | |
2932 | BUG_ON(!ocfs2_is_empty_extent(&(path_leaf_el(path)->l_recs[0]))); | |
2933 | ||
2934 | *empty_extent_path = NULL; | |
2935 | ||
e46f74dc | 2936 | ret = ocfs2_find_cpos_for_right_leaf(sb, path, &right_cpos); |
328d5752 MF |
2937 | if (ret) { |
2938 | mlog_errno(ret); | |
2939 | goto out; | |
2940 | } | |
2941 | ||
ffdd7a54 | 2942 | left_path = ocfs2_new_path_from_path(path); |
328d5752 MF |
2943 | if (!left_path) { |
2944 | ret = -ENOMEM; | |
2945 | mlog_errno(ret); | |
2946 | goto out; | |
2947 | } | |
2948 | ||
2949 | ocfs2_cp_path(left_path, path); | |
2950 | ||
ffdd7a54 | 2951 | right_path = ocfs2_new_path_from_path(path); |
328d5752 MF |
2952 | if (!right_path) { |
2953 | ret = -ENOMEM; | |
2954 | mlog_errno(ret); | |
2955 | goto out; | |
2956 | } | |
2957 | ||
2958 | while (right_cpos) { | |
facdb77f | 2959 | ret = ocfs2_find_path(et->et_ci, right_path, right_cpos); |
328d5752 MF |
2960 | if (ret) { |
2961 | mlog_errno(ret); | |
2962 | goto out; | |
2963 | } | |
2964 | ||
7dc02805 | 2965 | subtree_root = ocfs2_find_subtree_root(et, left_path, |
328d5752 MF |
2966 | right_path); |
2967 | ||
2968 | mlog(0, "Subtree root at index %d (blk %llu, depth %d)\n", | |
2969 | subtree_root, | |
2970 | (unsigned long long) | |
2971 | right_path->p_node[subtree_root].bh->b_blocknr, | |
2972 | right_path->p_tree_depth); | |
2973 | ||
2974 | ret = ocfs2_extend_rotate_transaction(handle, subtree_root, | |
2975 | orig_credits, left_path); | |
2976 | if (ret) { | |
2977 | mlog_errno(ret); | |
2978 | goto out; | |
2979 | } | |
2980 | ||
e8aed345 MF |
2981 | /* |
2982 | * Caller might still want to make changes to the | |
2983 | * tree root, so re-add it to the journal here. | |
2984 | */ | |
e46f74dc | 2985 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 2986 | left_path, 0); |
e8aed345 MF |
2987 | if (ret) { |
2988 | mlog_errno(ret); | |
2989 | goto out; | |
2990 | } | |
2991 | ||
1e2dd63f | 2992 | ret = ocfs2_rotate_subtree_left(handle, et, left_path, |
328d5752 | 2993 | right_path, subtree_root, |
1e2dd63f | 2994 | dealloc, &deleted); |
328d5752 MF |
2995 | if (ret == -EAGAIN) { |
2996 | /* | |
2997 | * The rotation has to temporarily stop due to | |
2998 | * the right subtree having an empty | |
2999 | * extent. Pass it back to the caller for a | |
3000 | * fixup. | |
3001 | */ | |
3002 | *empty_extent_path = right_path; | |
3003 | right_path = NULL; | |
3004 | goto out; | |
3005 | } | |
3006 | if (ret) { | |
3007 | mlog_errno(ret); | |
3008 | goto out; | |
3009 | } | |
3010 | ||
3011 | /* | |
3012 | * The subtree rotate might have removed records on | |
3013 | * the rightmost edge. If so, then rotation is | |
3014 | * complete. | |
3015 | */ | |
3016 | if (deleted) | |
3017 | break; | |
3018 | ||
3019 | ocfs2_mv_path(left_path, right_path); | |
3020 | ||
e46f74dc | 3021 | ret = ocfs2_find_cpos_for_right_leaf(sb, left_path, |
328d5752 MF |
3022 | &right_cpos); |
3023 | if (ret) { | |
3024 | mlog_errno(ret); | |
3025 | goto out; | |
3026 | } | |
3027 | } | |
3028 | ||
3029 | out: | |
3030 | ocfs2_free_path(right_path); | |
3031 | ocfs2_free_path(left_path); | |
3032 | ||
3033 | return ret; | |
3034 | } | |
3035 | ||
70f18c08 JB |
3036 | static int ocfs2_remove_rightmost_path(handle_t *handle, |
3037 | struct ocfs2_extent_tree *et, | |
e7d4cb6b | 3038 | struct ocfs2_path *path, |
70f18c08 | 3039 | struct ocfs2_cached_dealloc_ctxt *dealloc) |
328d5752 MF |
3040 | { |
3041 | int ret, subtree_index; | |
3042 | u32 cpos; | |
3043 | struct ocfs2_path *left_path = NULL; | |
328d5752 MF |
3044 | struct ocfs2_extent_block *eb; |
3045 | struct ocfs2_extent_list *el; | |
3046 | ||
328d5752 | 3047 | |
6136ca5f | 3048 | ret = ocfs2_et_sanity_check(et); |
e7d4cb6b TM |
3049 | if (ret) |
3050 | goto out; | |
328d5752 MF |
3051 | /* |
3052 | * There's two ways we handle this depending on | |
3053 | * whether path is the only existing one. | |
3054 | */ | |
3055 | ret = ocfs2_extend_rotate_transaction(handle, 0, | |
3056 | handle->h_buffer_credits, | |
3057 | path); | |
3058 | if (ret) { | |
3059 | mlog_errno(ret); | |
3060 | goto out; | |
3061 | } | |
3062 | ||
d9a0a1f8 | 3063 | ret = ocfs2_journal_access_path(et->et_ci, handle, path); |
328d5752 MF |
3064 | if (ret) { |
3065 | mlog_errno(ret); | |
3066 | goto out; | |
3067 | } | |
3068 | ||
3d03a305 JB |
3069 | ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci), |
3070 | path, &cpos); | |
328d5752 MF |
3071 | if (ret) { |
3072 | mlog_errno(ret); | |
3073 | goto out; | |
3074 | } | |
3075 | ||
3076 | if (cpos) { | |
3077 | /* | |
3078 | * We have a path to the left of this one - it needs | |
3079 | * an update too. | |
3080 | */ | |
ffdd7a54 | 3081 | left_path = ocfs2_new_path_from_path(path); |
328d5752 MF |
3082 | if (!left_path) { |
3083 | ret = -ENOMEM; | |
3084 | mlog_errno(ret); | |
3085 | goto out; | |
3086 | } | |
3087 | ||
facdb77f | 3088 | ret = ocfs2_find_path(et->et_ci, left_path, cpos); |
328d5752 MF |
3089 | if (ret) { |
3090 | mlog_errno(ret); | |
3091 | goto out; | |
3092 | } | |
3093 | ||
d9a0a1f8 | 3094 | ret = ocfs2_journal_access_path(et->et_ci, handle, left_path); |
328d5752 MF |
3095 | if (ret) { |
3096 | mlog_errno(ret); | |
3097 | goto out; | |
3098 | } | |
3099 | ||
7dc02805 | 3100 | subtree_index = ocfs2_find_subtree_root(et, left_path, path); |
328d5752 | 3101 | |
6641b0ce | 3102 | ocfs2_unlink_subtree(handle, et, left_path, path, |
328d5752 | 3103 | subtree_index, dealloc); |
09106bae | 3104 | ret = ocfs2_update_edge_lengths(handle, et, subtree_index, |
3c5e1068 TM |
3105 | left_path); |
3106 | if (ret) { | |
3107 | mlog_errno(ret); | |
3108 | goto out; | |
3109 | } | |
328d5752 MF |
3110 | |
3111 | eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data; | |
35dc0aa3 | 3112 | ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno)); |
328d5752 MF |
3113 | } else { |
3114 | /* | |
3115 | * 'path' is also the leftmost path which | |
3116 | * means it must be the only one. This gets | |
3117 | * handled differently because we want to | |
70f18c08 | 3118 | * revert the root back to having extents |
328d5752 MF |
3119 | * in-line. |
3120 | */ | |
6641b0ce | 3121 | ocfs2_unlink_path(handle, et, dealloc, path, 1); |
328d5752 | 3122 | |
ce1d9ea6 | 3123 | el = et->et_root_el; |
328d5752 MF |
3124 | el->l_tree_depth = 0; |
3125 | el->l_next_free_rec = 0; | |
3126 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
3127 | ||
35dc0aa3 | 3128 | ocfs2_et_set_last_eb_blk(et, 0); |
328d5752 MF |
3129 | } |
3130 | ||
3131 | ocfs2_journal_dirty(handle, path_root_bh(path)); | |
3132 | ||
3133 | out: | |
3134 | ocfs2_free_path(left_path); | |
3135 | return ret; | |
3136 | } | |
3137 | ||
3138 | /* | |
3139 | * Left rotation of btree records. | |
3140 | * | |
3141 | * In many ways, this is (unsurprisingly) the opposite of right | |
3142 | * rotation. We start at some non-rightmost path containing an empty | |
3143 | * extent in the leaf block. The code works its way to the rightmost | |
3144 | * path by rotating records to the left in every subtree. | |
3145 | * | |
3146 | * This is used by any code which reduces the number of extent records | |
3147 | * in a leaf. After removal, an empty record should be placed in the | |
3148 | * leftmost list position. | |
3149 | * | |
3150 | * This won't handle a length update of the rightmost path records if | |
3151 | * the rightmost tree leaf record is removed so the caller is | |
3152 | * responsible for detecting and correcting that. | |
3153 | */ | |
70f18c08 JB |
3154 | static int ocfs2_rotate_tree_left(handle_t *handle, |
3155 | struct ocfs2_extent_tree *et, | |
328d5752 | 3156 | struct ocfs2_path *path, |
70f18c08 | 3157 | struct ocfs2_cached_dealloc_ctxt *dealloc) |
328d5752 MF |
3158 | { |
3159 | int ret, orig_credits = handle->h_buffer_credits; | |
3160 | struct ocfs2_path *tmp_path = NULL, *restart_path = NULL; | |
3161 | struct ocfs2_extent_block *eb; | |
3162 | struct ocfs2_extent_list *el; | |
3163 | ||
3164 | el = path_leaf_el(path); | |
3165 | if (!ocfs2_is_empty_extent(&el->l_recs[0])) | |
3166 | return 0; | |
3167 | ||
3168 | if (path->p_tree_depth == 0) { | |
3169 | rightmost_no_delete: | |
3170 | /* | |
e7d4cb6b | 3171 | * Inline extents. This is trivially handled, so do |
328d5752 MF |
3172 | * it up front. |
3173 | */ | |
70f18c08 | 3174 | ret = ocfs2_rotate_rightmost_leaf_left(handle, et, path); |
328d5752 MF |
3175 | if (ret) |
3176 | mlog_errno(ret); | |
3177 | goto out; | |
3178 | } | |
3179 | ||
3180 | /* | |
3181 | * Handle rightmost branch now. There's several cases: | |
3182 | * 1) simple rotation leaving records in there. That's trivial. | |
3183 | * 2) rotation requiring a branch delete - there's no more | |
3184 | * records left. Two cases of this: | |
3185 | * a) There are branches to the left. | |
3186 | * b) This is also the leftmost (the only) branch. | |
3187 | * | |
3188 | * 1) is handled via ocfs2_rotate_rightmost_leaf_left() | |
3189 | * 2a) we need the left branch so that we can update it with the unlink | |
70f18c08 | 3190 | * 2b) we need to bring the root back to inline extents. |
328d5752 MF |
3191 | */ |
3192 | ||
3193 | eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data; | |
3194 | el = &eb->h_list; | |
3195 | if (eb->h_next_leaf_blk == 0) { | |
3196 | /* | |
3197 | * This gets a bit tricky if we're going to delete the | |
3198 | * rightmost path. Get the other cases out of the way | |
3199 | * 1st. | |
3200 | */ | |
3201 | if (le16_to_cpu(el->l_next_free_rec) > 1) | |
3202 | goto rightmost_no_delete; | |
3203 | ||
3204 | if (le16_to_cpu(el->l_next_free_rec) == 0) { | |
3205 | ret = -EIO; | |
70f18c08 JB |
3206 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
3207 | "Owner %llu has empty extent block at %llu", | |
3208 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), | |
328d5752 MF |
3209 | (unsigned long long)le64_to_cpu(eb->h_blkno)); |
3210 | goto out; | |
3211 | } | |
3212 | ||
3213 | /* | |
3214 | * XXX: The caller can not trust "path" any more after | |
3215 | * this as it will have been deleted. What do we do? | |
3216 | * | |
3217 | * In theory the rotate-for-merge code will never get | |
3218 | * here because it'll always ask for a rotate in a | |
3219 | * nonempty list. | |
3220 | */ | |
3221 | ||
70f18c08 JB |
3222 | ret = ocfs2_remove_rightmost_path(handle, et, path, |
3223 | dealloc); | |
328d5752 MF |
3224 | if (ret) |
3225 | mlog_errno(ret); | |
3226 | goto out; | |
3227 | } | |
3228 | ||
3229 | /* | |
3230 | * Now we can loop, remembering the path we get from -EAGAIN | |
3231 | * and restarting from there. | |
3232 | */ | |
3233 | try_rotate: | |
e46f74dc JB |
3234 | ret = __ocfs2_rotate_tree_left(handle, et, orig_credits, path, |
3235 | dealloc, &restart_path); | |
328d5752 MF |
3236 | if (ret && ret != -EAGAIN) { |
3237 | mlog_errno(ret); | |
3238 | goto out; | |
3239 | } | |
3240 | ||
3241 | while (ret == -EAGAIN) { | |
3242 | tmp_path = restart_path; | |
3243 | restart_path = NULL; | |
3244 | ||
e46f74dc | 3245 | ret = __ocfs2_rotate_tree_left(handle, et, orig_credits, |
328d5752 | 3246 | tmp_path, dealloc, |
e46f74dc | 3247 | &restart_path); |
328d5752 MF |
3248 | if (ret && ret != -EAGAIN) { |
3249 | mlog_errno(ret); | |
3250 | goto out; | |
3251 | } | |
3252 | ||
3253 | ocfs2_free_path(tmp_path); | |
3254 | tmp_path = NULL; | |
3255 | ||
3256 | if (ret == 0) | |
3257 | goto try_rotate; | |
3258 | } | |
3259 | ||
3260 | out: | |
3261 | ocfs2_free_path(tmp_path); | |
3262 | ocfs2_free_path(restart_path); | |
3263 | return ret; | |
3264 | } | |
3265 | ||
3266 | static void ocfs2_cleanup_merge(struct ocfs2_extent_list *el, | |
3267 | int index) | |
3268 | { | |
3269 | struct ocfs2_extent_rec *rec = &el->l_recs[index]; | |
3270 | unsigned int size; | |
3271 | ||
3272 | if (rec->e_leaf_clusters == 0) { | |
3273 | /* | |
3274 | * We consumed all of the merged-from record. An empty | |
3275 | * extent cannot exist anywhere but the 1st array | |
3276 | * position, so move things over if the merged-from | |
3277 | * record doesn't occupy that position. | |
3278 | * | |
3279 | * This creates a new empty extent so the caller | |
3280 | * should be smart enough to have removed any existing | |
3281 | * ones. | |
3282 | */ | |
3283 | if (index > 0) { | |
3284 | BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0])); | |
3285 | size = index * sizeof(struct ocfs2_extent_rec); | |
3286 | memmove(&el->l_recs[1], &el->l_recs[0], size); | |
3287 | } | |
3288 | ||
3289 | /* | |
3290 | * Always memset - the caller doesn't check whether it | |
3291 | * created an empty extent, so there could be junk in | |
3292 | * the other fields. | |
3293 | */ | |
3294 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
3295 | } | |
3296 | } | |
3297 | ||
4fe82c31 | 3298 | static int ocfs2_get_right_path(struct ocfs2_extent_tree *et, |
677b9752 TM |
3299 | struct ocfs2_path *left_path, |
3300 | struct ocfs2_path **ret_right_path) | |
3301 | { | |
3302 | int ret; | |
3303 | u32 right_cpos; | |
3304 | struct ocfs2_path *right_path = NULL; | |
3305 | struct ocfs2_extent_list *left_el; | |
3306 | ||
3307 | *ret_right_path = NULL; | |
3308 | ||
3309 | /* This function shouldn't be called for non-trees. */ | |
3310 | BUG_ON(left_path->p_tree_depth == 0); | |
3311 | ||
3312 | left_el = path_leaf_el(left_path); | |
3313 | BUG_ON(left_el->l_next_free_rec != left_el->l_count); | |
3314 | ||
4fe82c31 JB |
3315 | ret = ocfs2_find_cpos_for_right_leaf(ocfs2_metadata_cache_get_super(et->et_ci), |
3316 | left_path, &right_cpos); | |
677b9752 TM |
3317 | if (ret) { |
3318 | mlog_errno(ret); | |
3319 | goto out; | |
3320 | } | |
3321 | ||
3322 | /* This function shouldn't be called for the rightmost leaf. */ | |
3323 | BUG_ON(right_cpos == 0); | |
3324 | ||
ffdd7a54 | 3325 | right_path = ocfs2_new_path_from_path(left_path); |
677b9752 TM |
3326 | if (!right_path) { |
3327 | ret = -ENOMEM; | |
3328 | mlog_errno(ret); | |
3329 | goto out; | |
3330 | } | |
3331 | ||
4fe82c31 | 3332 | ret = ocfs2_find_path(et->et_ci, right_path, right_cpos); |
677b9752 TM |
3333 | if (ret) { |
3334 | mlog_errno(ret); | |
3335 | goto out; | |
3336 | } | |
3337 | ||
3338 | *ret_right_path = right_path; | |
3339 | out: | |
3340 | if (ret) | |
3341 | ocfs2_free_path(right_path); | |
3342 | return ret; | |
3343 | } | |
3344 | ||
328d5752 MF |
3345 | /* |
3346 | * Remove split_rec clusters from the record at index and merge them | |
677b9752 TM |
3347 | * onto the beginning of the record "next" to it. |
3348 | * For index < l_count - 1, the next means the extent rec at index + 1. | |
3349 | * For index == l_count - 1, the "next" means the 1st extent rec of the | |
3350 | * next extent block. | |
328d5752 | 3351 | */ |
4fe82c31 | 3352 | static int ocfs2_merge_rec_right(struct ocfs2_path *left_path, |
677b9752 | 3353 | handle_t *handle, |
7dc02805 | 3354 | struct ocfs2_extent_tree *et, |
677b9752 TM |
3355 | struct ocfs2_extent_rec *split_rec, |
3356 | int index) | |
328d5752 | 3357 | { |
677b9752 | 3358 | int ret, next_free, i; |
328d5752 MF |
3359 | unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters); |
3360 | struct ocfs2_extent_rec *left_rec; | |
3361 | struct ocfs2_extent_rec *right_rec; | |
677b9752 TM |
3362 | struct ocfs2_extent_list *right_el; |
3363 | struct ocfs2_path *right_path = NULL; | |
3364 | int subtree_index = 0; | |
3365 | struct ocfs2_extent_list *el = path_leaf_el(left_path); | |
3366 | struct buffer_head *bh = path_leaf_bh(left_path); | |
3367 | struct buffer_head *root_bh = NULL; | |
328d5752 MF |
3368 | |
3369 | BUG_ON(index >= le16_to_cpu(el->l_next_free_rec)); | |
328d5752 | 3370 | left_rec = &el->l_recs[index]; |
677b9752 | 3371 | |
9d8df6aa | 3372 | if (index == le16_to_cpu(el->l_next_free_rec) - 1 && |
677b9752 TM |
3373 | le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count)) { |
3374 | /* we meet with a cross extent block merge. */ | |
4fe82c31 | 3375 | ret = ocfs2_get_right_path(et, left_path, &right_path); |
677b9752 TM |
3376 | if (ret) { |
3377 | mlog_errno(ret); | |
3378 | goto out; | |
3379 | } | |
3380 | ||
3381 | right_el = path_leaf_el(right_path); | |
3382 | next_free = le16_to_cpu(right_el->l_next_free_rec); | |
3383 | BUG_ON(next_free <= 0); | |
3384 | right_rec = &right_el->l_recs[0]; | |
3385 | if (ocfs2_is_empty_extent(right_rec)) { | |
9d8df6aa | 3386 | BUG_ON(next_free <= 1); |
677b9752 TM |
3387 | right_rec = &right_el->l_recs[1]; |
3388 | } | |
3389 | ||
3390 | BUG_ON(le32_to_cpu(left_rec->e_cpos) + | |
3391 | le16_to_cpu(left_rec->e_leaf_clusters) != | |
3392 | le32_to_cpu(right_rec->e_cpos)); | |
3393 | ||
7dc02805 JB |
3394 | subtree_index = ocfs2_find_subtree_root(et, left_path, |
3395 | right_path); | |
677b9752 TM |
3396 | |
3397 | ret = ocfs2_extend_rotate_transaction(handle, subtree_index, | |
3398 | handle->h_buffer_credits, | |
3399 | right_path); | |
3400 | if (ret) { | |
3401 | mlog_errno(ret); | |
3402 | goto out; | |
3403 | } | |
3404 | ||
3405 | root_bh = left_path->p_node[subtree_index].bh; | |
3406 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
3407 | ||
7dc02805 | 3408 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path, |
13723d00 | 3409 | subtree_index); |
677b9752 TM |
3410 | if (ret) { |
3411 | mlog_errno(ret); | |
3412 | goto out; | |
3413 | } | |
3414 | ||
3415 | for (i = subtree_index + 1; | |
3416 | i < path_num_items(right_path); i++) { | |
7dc02805 | 3417 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 3418 | right_path, i); |
677b9752 TM |
3419 | if (ret) { |
3420 | mlog_errno(ret); | |
3421 | goto out; | |
3422 | } | |
3423 | ||
7dc02805 | 3424 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 3425 | left_path, i); |
677b9752 TM |
3426 | if (ret) { |
3427 | mlog_errno(ret); | |
3428 | goto out; | |
3429 | } | |
3430 | } | |
3431 | ||
3432 | } else { | |
3433 | BUG_ON(index == le16_to_cpu(el->l_next_free_rec) - 1); | |
3434 | right_rec = &el->l_recs[index + 1]; | |
3435 | } | |
328d5752 | 3436 | |
7dc02805 | 3437 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, left_path, |
13723d00 | 3438 | path_num_items(left_path) - 1); |
328d5752 MF |
3439 | if (ret) { |
3440 | mlog_errno(ret); | |
3441 | goto out; | |
3442 | } | |
3443 | ||
3444 | le16_add_cpu(&left_rec->e_leaf_clusters, -split_clusters); | |
3445 | ||
3446 | le32_add_cpu(&right_rec->e_cpos, -split_clusters); | |
3447 | le64_add_cpu(&right_rec->e_blkno, | |
7dc02805 JB |
3448 | -ocfs2_clusters_to_blocks(ocfs2_metadata_cache_get_super(et->et_ci), |
3449 | split_clusters)); | |
328d5752 MF |
3450 | le16_add_cpu(&right_rec->e_leaf_clusters, split_clusters); |
3451 | ||
3452 | ocfs2_cleanup_merge(el, index); | |
3453 | ||
ec20cec7 | 3454 | ocfs2_journal_dirty(handle, bh); |
677b9752 | 3455 | if (right_path) { |
ec20cec7 | 3456 | ocfs2_journal_dirty(handle, path_leaf_bh(right_path)); |
4619c73e JB |
3457 | ocfs2_complete_edge_insert(handle, left_path, right_path, |
3458 | subtree_index); | |
677b9752 TM |
3459 | } |
3460 | out: | |
3461 | if (right_path) | |
3462 | ocfs2_free_path(right_path); | |
3463 | return ret; | |
3464 | } | |
3465 | ||
4fe82c31 | 3466 | static int ocfs2_get_left_path(struct ocfs2_extent_tree *et, |
677b9752 TM |
3467 | struct ocfs2_path *right_path, |
3468 | struct ocfs2_path **ret_left_path) | |
3469 | { | |
3470 | int ret; | |
3471 | u32 left_cpos; | |
3472 | struct ocfs2_path *left_path = NULL; | |
3473 | ||
3474 | *ret_left_path = NULL; | |
3475 | ||
3476 | /* This function shouldn't be called for non-trees. */ | |
3477 | BUG_ON(right_path->p_tree_depth == 0); | |
3478 | ||
4fe82c31 | 3479 | ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci), |
677b9752 TM |
3480 | right_path, &left_cpos); |
3481 | if (ret) { | |
3482 | mlog_errno(ret); | |
3483 | goto out; | |
3484 | } | |
3485 | ||
3486 | /* This function shouldn't be called for the leftmost leaf. */ | |
3487 | BUG_ON(left_cpos == 0); | |
3488 | ||
ffdd7a54 | 3489 | left_path = ocfs2_new_path_from_path(right_path); |
677b9752 TM |
3490 | if (!left_path) { |
3491 | ret = -ENOMEM; | |
3492 | mlog_errno(ret); | |
3493 | goto out; | |
3494 | } | |
3495 | ||
4fe82c31 | 3496 | ret = ocfs2_find_path(et->et_ci, left_path, left_cpos); |
677b9752 TM |
3497 | if (ret) { |
3498 | mlog_errno(ret); | |
3499 | goto out; | |
3500 | } | |
3501 | ||
3502 | *ret_left_path = left_path; | |
328d5752 | 3503 | out: |
677b9752 TM |
3504 | if (ret) |
3505 | ocfs2_free_path(left_path); | |
328d5752 MF |
3506 | return ret; |
3507 | } | |
3508 | ||
3509 | /* | |
3510 | * Remove split_rec clusters from the record at index and merge them | |
677b9752 TM |
3511 | * onto the tail of the record "before" it. |
3512 | * For index > 0, the "before" means the extent rec at index - 1. | |
3513 | * | |
3514 | * For index == 0, the "before" means the last record of the previous | |
3515 | * extent block. And there is also a situation that we may need to | |
3516 | * remove the rightmost leaf extent block in the right_path and change | |
3517 | * the right path to indicate the new rightmost path. | |
328d5752 | 3518 | */ |
4fe82c31 | 3519 | static int ocfs2_merge_rec_left(struct ocfs2_path *right_path, |
328d5752 | 3520 | handle_t *handle, |
4fe82c31 | 3521 | struct ocfs2_extent_tree *et, |
328d5752 | 3522 | struct ocfs2_extent_rec *split_rec, |
677b9752 TM |
3523 | struct ocfs2_cached_dealloc_ctxt *dealloc, |
3524 | int index) | |
328d5752 | 3525 | { |
677b9752 | 3526 | int ret, i, subtree_index = 0, has_empty_extent = 0; |
328d5752 MF |
3527 | unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters); |
3528 | struct ocfs2_extent_rec *left_rec; | |
3529 | struct ocfs2_extent_rec *right_rec; | |
677b9752 TM |
3530 | struct ocfs2_extent_list *el = path_leaf_el(right_path); |
3531 | struct buffer_head *bh = path_leaf_bh(right_path); | |
3532 | struct buffer_head *root_bh = NULL; | |
3533 | struct ocfs2_path *left_path = NULL; | |
3534 | struct ocfs2_extent_list *left_el; | |
328d5752 | 3535 | |
677b9752 | 3536 | BUG_ON(index < 0); |
328d5752 | 3537 | |
328d5752 | 3538 | right_rec = &el->l_recs[index]; |
677b9752 TM |
3539 | if (index == 0) { |
3540 | /* we meet with a cross extent block merge. */ | |
4fe82c31 | 3541 | ret = ocfs2_get_left_path(et, right_path, &left_path); |
677b9752 TM |
3542 | if (ret) { |
3543 | mlog_errno(ret); | |
3544 | goto out; | |
3545 | } | |
3546 | ||
3547 | left_el = path_leaf_el(left_path); | |
3548 | BUG_ON(le16_to_cpu(left_el->l_next_free_rec) != | |
3549 | le16_to_cpu(left_el->l_count)); | |
3550 | ||
3551 | left_rec = &left_el->l_recs[ | |
3552 | le16_to_cpu(left_el->l_next_free_rec) - 1]; | |
3553 | BUG_ON(le32_to_cpu(left_rec->e_cpos) + | |
3554 | le16_to_cpu(left_rec->e_leaf_clusters) != | |
3555 | le32_to_cpu(split_rec->e_cpos)); | |
3556 | ||
7dc02805 JB |
3557 | subtree_index = ocfs2_find_subtree_root(et, left_path, |
3558 | right_path); | |
677b9752 TM |
3559 | |
3560 | ret = ocfs2_extend_rotate_transaction(handle, subtree_index, | |
3561 | handle->h_buffer_credits, | |
3562 | left_path); | |
3563 | if (ret) { | |
3564 | mlog_errno(ret); | |
3565 | goto out; | |
3566 | } | |
3567 | ||
3568 | root_bh = left_path->p_node[subtree_index].bh; | |
3569 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
3570 | ||
4fe82c31 | 3571 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path, |
13723d00 | 3572 | subtree_index); |
677b9752 TM |
3573 | if (ret) { |
3574 | mlog_errno(ret); | |
3575 | goto out; | |
3576 | } | |
3577 | ||
3578 | for (i = subtree_index + 1; | |
3579 | i < path_num_items(right_path); i++) { | |
4fe82c31 | 3580 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 3581 | right_path, i); |
677b9752 TM |
3582 | if (ret) { |
3583 | mlog_errno(ret); | |
3584 | goto out; | |
3585 | } | |
3586 | ||
4fe82c31 | 3587 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 3588 | left_path, i); |
677b9752 TM |
3589 | if (ret) { |
3590 | mlog_errno(ret); | |
3591 | goto out; | |
3592 | } | |
3593 | } | |
3594 | } else { | |
3595 | left_rec = &el->l_recs[index - 1]; | |
3596 | if (ocfs2_is_empty_extent(&el->l_recs[0])) | |
3597 | has_empty_extent = 1; | |
3598 | } | |
328d5752 | 3599 | |
4fe82c31 | 3600 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path, |
9047beab | 3601 | path_num_items(right_path) - 1); |
328d5752 MF |
3602 | if (ret) { |
3603 | mlog_errno(ret); | |
3604 | goto out; | |
3605 | } | |
3606 | ||
3607 | if (has_empty_extent && index == 1) { | |
3608 | /* | |
3609 | * The easy case - we can just plop the record right in. | |
3610 | */ | |
3611 | *left_rec = *split_rec; | |
3612 | ||
3613 | has_empty_extent = 0; | |
677b9752 | 3614 | } else |
328d5752 | 3615 | le16_add_cpu(&left_rec->e_leaf_clusters, split_clusters); |
328d5752 MF |
3616 | |
3617 | le32_add_cpu(&right_rec->e_cpos, split_clusters); | |
3618 | le64_add_cpu(&right_rec->e_blkno, | |
4fe82c31 JB |
3619 | ocfs2_clusters_to_blocks(ocfs2_metadata_cache_get_super(et->et_ci), |
3620 | split_clusters)); | |
328d5752 MF |
3621 | le16_add_cpu(&right_rec->e_leaf_clusters, -split_clusters); |
3622 | ||
3623 | ocfs2_cleanup_merge(el, index); | |
3624 | ||
ec20cec7 | 3625 | ocfs2_journal_dirty(handle, bh); |
677b9752 | 3626 | if (left_path) { |
ec20cec7 | 3627 | ocfs2_journal_dirty(handle, path_leaf_bh(left_path)); |
677b9752 TM |
3628 | |
3629 | /* | |
3630 | * In the situation that the right_rec is empty and the extent | |
3631 | * block is empty also, ocfs2_complete_edge_insert can't handle | |
3632 | * it and we need to delete the right extent block. | |
3633 | */ | |
3634 | if (le16_to_cpu(right_rec->e_leaf_clusters) == 0 && | |
3635 | le16_to_cpu(el->l_next_free_rec) == 1) { | |
3636 | ||
70f18c08 | 3637 | ret = ocfs2_remove_rightmost_path(handle, et, |
e7d4cb6b | 3638 | right_path, |
70f18c08 | 3639 | dealloc); |
677b9752 TM |
3640 | if (ret) { |
3641 | mlog_errno(ret); | |
3642 | goto out; | |
3643 | } | |
3644 | ||
3645 | /* Now the rightmost extent block has been deleted. | |
3646 | * So we use the new rightmost path. | |
3647 | */ | |
3648 | ocfs2_mv_path(right_path, left_path); | |
3649 | left_path = NULL; | |
3650 | } else | |
4619c73e | 3651 | ocfs2_complete_edge_insert(handle, left_path, |
677b9752 TM |
3652 | right_path, subtree_index); |
3653 | } | |
328d5752 | 3654 | out: |
677b9752 TM |
3655 | if (left_path) |
3656 | ocfs2_free_path(left_path); | |
328d5752 MF |
3657 | return ret; |
3658 | } | |
3659 | ||
c495dd24 JB |
3660 | static int ocfs2_try_to_merge_extent(handle_t *handle, |
3661 | struct ocfs2_extent_tree *et, | |
677b9752 | 3662 | struct ocfs2_path *path, |
328d5752 MF |
3663 | int split_index, |
3664 | struct ocfs2_extent_rec *split_rec, | |
3665 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
c495dd24 | 3666 | struct ocfs2_merge_ctxt *ctxt) |
328d5752 | 3667 | { |
518d7269 | 3668 | int ret = 0; |
677b9752 | 3669 | struct ocfs2_extent_list *el = path_leaf_el(path); |
328d5752 MF |
3670 | struct ocfs2_extent_rec *rec = &el->l_recs[split_index]; |
3671 | ||
3672 | BUG_ON(ctxt->c_contig_type == CONTIG_NONE); | |
3673 | ||
518d7269 TM |
3674 | if (ctxt->c_split_covers_rec && ctxt->c_has_empty_extent) { |
3675 | /* | |
3676 | * The merge code will need to create an empty | |
3677 | * extent to take the place of the newly | |
3678 | * emptied slot. Remove any pre-existing empty | |
3679 | * extents - having more than one in a leaf is | |
3680 | * illegal. | |
3681 | */ | |
70f18c08 | 3682 | ret = ocfs2_rotate_tree_left(handle, et, path, dealloc); |
518d7269 TM |
3683 | if (ret) { |
3684 | mlog_errno(ret); | |
3685 | goto out; | |
328d5752 | 3686 | } |
518d7269 TM |
3687 | split_index--; |
3688 | rec = &el->l_recs[split_index]; | |
328d5752 MF |
3689 | } |
3690 | ||
3691 | if (ctxt->c_contig_type == CONTIG_LEFTRIGHT) { | |
3692 | /* | |
3693 | * Left-right contig implies this. | |
3694 | */ | |
3695 | BUG_ON(!ctxt->c_split_covers_rec); | |
328d5752 MF |
3696 | |
3697 | /* | |
3698 | * Since the leftright insert always covers the entire | |
3699 | * extent, this call will delete the insert record | |
3700 | * entirely, resulting in an empty extent record added to | |
3701 | * the extent block. | |
3702 | * | |
3703 | * Since the adding of an empty extent shifts | |
3704 | * everything back to the right, there's no need to | |
3705 | * update split_index here. | |
677b9752 TM |
3706 | * |
3707 | * When the split_index is zero, we need to merge it to the | |
3708 | * prevoius extent block. It is more efficient and easier | |
3709 | * if we do merge_right first and merge_left later. | |
328d5752 | 3710 | */ |
4fe82c31 | 3711 | ret = ocfs2_merge_rec_right(path, handle, et, split_rec, |
677b9752 | 3712 | split_index); |
328d5752 MF |
3713 | if (ret) { |
3714 | mlog_errno(ret); | |
3715 | goto out; | |
3716 | } | |
3717 | ||
3718 | /* | |
3719 | * We can only get this from logic error above. | |
3720 | */ | |
3721 | BUG_ON(!ocfs2_is_empty_extent(&el->l_recs[0])); | |
3722 | ||
677b9752 | 3723 | /* The merge left us with an empty extent, remove it. */ |
70f18c08 | 3724 | ret = ocfs2_rotate_tree_left(handle, et, path, dealloc); |
328d5752 MF |
3725 | if (ret) { |
3726 | mlog_errno(ret); | |
3727 | goto out; | |
3728 | } | |
677b9752 | 3729 | |
328d5752 MF |
3730 | rec = &el->l_recs[split_index]; |
3731 | ||
3732 | /* | |
3733 | * Note that we don't pass split_rec here on purpose - | |
677b9752 | 3734 | * we've merged it into the rec already. |
328d5752 | 3735 | */ |
4fe82c31 JB |
3736 | ret = ocfs2_merge_rec_left(path, handle, et, rec, |
3737 | dealloc, split_index); | |
677b9752 | 3738 | |
328d5752 MF |
3739 | if (ret) { |
3740 | mlog_errno(ret); | |
3741 | goto out; | |
3742 | } | |
3743 | ||
70f18c08 | 3744 | ret = ocfs2_rotate_tree_left(handle, et, path, dealloc); |
328d5752 MF |
3745 | /* |
3746 | * Error from this last rotate is not critical, so | |
3747 | * print but don't bubble it up. | |
3748 | */ | |
3749 | if (ret) | |
3750 | mlog_errno(ret); | |
3751 | ret = 0; | |
3752 | } else { | |
3753 | /* | |
3754 | * Merge a record to the left or right. | |
3755 | * | |
3756 | * 'contig_type' is relative to the existing record, | |
3757 | * so for example, if we're "right contig", it's to | |
3758 | * the record on the left (hence the left merge). | |
3759 | */ | |
3760 | if (ctxt->c_contig_type == CONTIG_RIGHT) { | |
4fe82c31 JB |
3761 | ret = ocfs2_merge_rec_left(path, handle, et, |
3762 | split_rec, dealloc, | |
328d5752 MF |
3763 | split_index); |
3764 | if (ret) { | |
3765 | mlog_errno(ret); | |
3766 | goto out; | |
3767 | } | |
3768 | } else { | |
4fe82c31 | 3769 | ret = ocfs2_merge_rec_right(path, handle, |
7dc02805 | 3770 | et, split_rec, |
328d5752 MF |
3771 | split_index); |
3772 | if (ret) { | |
3773 | mlog_errno(ret); | |
3774 | goto out; | |
3775 | } | |
3776 | } | |
3777 | ||
3778 | if (ctxt->c_split_covers_rec) { | |
3779 | /* | |
3780 | * The merge may have left an empty extent in | |
3781 | * our leaf. Try to rotate it away. | |
3782 | */ | |
70f18c08 JB |
3783 | ret = ocfs2_rotate_tree_left(handle, et, path, |
3784 | dealloc); | |
328d5752 MF |
3785 | if (ret) |
3786 | mlog_errno(ret); | |
3787 | ret = 0; | |
3788 | } | |
3789 | } | |
3790 | ||
3791 | out: | |
3792 | return ret; | |
3793 | } | |
3794 | ||
3795 | static void ocfs2_subtract_from_rec(struct super_block *sb, | |
3796 | enum ocfs2_split_type split, | |
3797 | struct ocfs2_extent_rec *rec, | |
3798 | struct ocfs2_extent_rec *split_rec) | |
3799 | { | |
3800 | u64 len_blocks; | |
3801 | ||
3802 | len_blocks = ocfs2_clusters_to_blocks(sb, | |
3803 | le16_to_cpu(split_rec->e_leaf_clusters)); | |
3804 | ||
3805 | if (split == SPLIT_LEFT) { | |
3806 | /* | |
3807 | * Region is on the left edge of the existing | |
3808 | * record. | |
3809 | */ | |
3810 | le32_add_cpu(&rec->e_cpos, | |
3811 | le16_to_cpu(split_rec->e_leaf_clusters)); | |
3812 | le64_add_cpu(&rec->e_blkno, len_blocks); | |
3813 | le16_add_cpu(&rec->e_leaf_clusters, | |
3814 | -le16_to_cpu(split_rec->e_leaf_clusters)); | |
3815 | } else { | |
3816 | /* | |
3817 | * Region is on the right edge of the existing | |
3818 | * record. | |
3819 | */ | |
3820 | le16_add_cpu(&rec->e_leaf_clusters, | |
3821 | -le16_to_cpu(split_rec->e_leaf_clusters)); | |
3822 | } | |
3823 | } | |
3824 | ||
3825 | /* | |
3826 | * Do the final bits of extent record insertion at the target leaf | |
3827 | * list. If this leaf is part of an allocation tree, it is assumed | |
3828 | * that the tree above has been prepared. | |
3829 | */ | |
d5628623 JB |
3830 | static void ocfs2_insert_at_leaf(struct ocfs2_extent_tree *et, |
3831 | struct ocfs2_extent_rec *insert_rec, | |
328d5752 | 3832 | struct ocfs2_extent_list *el, |
d5628623 | 3833 | struct ocfs2_insert_type *insert) |
328d5752 MF |
3834 | { |
3835 | int i = insert->ins_contig_index; | |
3836 | unsigned int range; | |
3837 | struct ocfs2_extent_rec *rec; | |
3838 | ||
3839 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); | |
3840 | ||
3841 | if (insert->ins_split != SPLIT_NONE) { | |
3842 | i = ocfs2_search_extent_list(el, le32_to_cpu(insert_rec->e_cpos)); | |
3843 | BUG_ON(i == -1); | |
3844 | rec = &el->l_recs[i]; | |
d5628623 JB |
3845 | ocfs2_subtract_from_rec(ocfs2_metadata_cache_get_super(et->et_ci), |
3846 | insert->ins_split, rec, | |
328d5752 MF |
3847 | insert_rec); |
3848 | goto rotate; | |
3849 | } | |
3850 | ||
3851 | /* | |
3852 | * Contiguous insert - either left or right. | |
3853 | */ | |
3854 | if (insert->ins_contig != CONTIG_NONE) { | |
3855 | rec = &el->l_recs[i]; | |
3856 | if (insert->ins_contig == CONTIG_LEFT) { | |
3857 | rec->e_blkno = insert_rec->e_blkno; | |
3858 | rec->e_cpos = insert_rec->e_cpos; | |
3859 | } | |
3860 | le16_add_cpu(&rec->e_leaf_clusters, | |
3861 | le16_to_cpu(insert_rec->e_leaf_clusters)); | |
3862 | return; | |
3863 | } | |
3864 | ||
3865 | /* | |
3866 | * Handle insert into an empty leaf. | |
3867 | */ | |
3868 | if (le16_to_cpu(el->l_next_free_rec) == 0 || | |
3869 | ((le16_to_cpu(el->l_next_free_rec) == 1) && | |
3870 | ocfs2_is_empty_extent(&el->l_recs[0]))) { | |
3871 | el->l_recs[0] = *insert_rec; | |
3872 | el->l_next_free_rec = cpu_to_le16(1); | |
3873 | return; | |
3874 | } | |
3875 | ||
3876 | /* | |
3877 | * Appending insert. | |
3878 | */ | |
3879 | if (insert->ins_appending == APPEND_TAIL) { | |
3880 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
3881 | rec = &el->l_recs[i]; | |
3882 | range = le32_to_cpu(rec->e_cpos) | |
3883 | + le16_to_cpu(rec->e_leaf_clusters); | |
3884 | BUG_ON(le32_to_cpu(insert_rec->e_cpos) < range); | |
3885 | ||
3886 | mlog_bug_on_msg(le16_to_cpu(el->l_next_free_rec) >= | |
3887 | le16_to_cpu(el->l_count), | |
d5628623 | 3888 | "owner %llu, depth %u, count %u, next free %u, " |
328d5752 MF |
3889 | "rec.cpos %u, rec.clusters %u, " |
3890 | "insert.cpos %u, insert.clusters %u\n", | |
d5628623 | 3891 | ocfs2_metadata_cache_owner(et->et_ci), |
328d5752 MF |
3892 | le16_to_cpu(el->l_tree_depth), |
3893 | le16_to_cpu(el->l_count), | |
3894 | le16_to_cpu(el->l_next_free_rec), | |
3895 | le32_to_cpu(el->l_recs[i].e_cpos), | |
3896 | le16_to_cpu(el->l_recs[i].e_leaf_clusters), | |
3897 | le32_to_cpu(insert_rec->e_cpos), | |
3898 | le16_to_cpu(insert_rec->e_leaf_clusters)); | |
3899 | i++; | |
3900 | el->l_recs[i] = *insert_rec; | |
3901 | le16_add_cpu(&el->l_next_free_rec, 1); | |
3902 | return; | |
3903 | } | |
3904 | ||
3905 | rotate: | |
3906 | /* | |
3907 | * Ok, we have to rotate. | |
3908 | * | |
3909 | * At this point, it is safe to assume that inserting into an | |
3910 | * empty leaf and appending to a leaf have both been handled | |
3911 | * above. | |
3912 | * | |
3913 | * This leaf needs to have space, either by the empty 1st | |
3914 | * extent record, or by virtue of an l_next_rec < l_count. | |
3915 | */ | |
3916 | ocfs2_rotate_leaf(el, insert_rec); | |
3917 | } | |
3918 | ||
d401dc12 JB |
3919 | static void ocfs2_adjust_rightmost_records(handle_t *handle, |
3920 | struct ocfs2_extent_tree *et, | |
328d5752 MF |
3921 | struct ocfs2_path *path, |
3922 | struct ocfs2_extent_rec *insert_rec) | |
3923 | { | |
3924 | int ret, i, next_free; | |
3925 | struct buffer_head *bh; | |
3926 | struct ocfs2_extent_list *el; | |
3927 | struct ocfs2_extent_rec *rec; | |
3928 | ||
3929 | /* | |
3930 | * Update everything except the leaf block. | |
3931 | */ | |
3932 | for (i = 0; i < path->p_tree_depth; i++) { | |
3933 | bh = path->p_node[i].bh; | |
3934 | el = path->p_node[i].el; | |
3935 | ||
dcd0538f MF |
3936 | next_free = le16_to_cpu(el->l_next_free_rec); |
3937 | if (next_free == 0) { | |
d401dc12 JB |
3938 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
3939 | "Owner %llu has a bad extent list", | |
3940 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci)); | |
dcd0538f | 3941 | ret = -EIO; |
328d5752 MF |
3942 | return; |
3943 | } | |
3944 | ||
3945 | rec = &el->l_recs[next_free - 1]; | |
3946 | ||
3947 | rec->e_int_clusters = insert_rec->e_cpos; | |
3948 | le32_add_cpu(&rec->e_int_clusters, | |
3949 | le16_to_cpu(insert_rec->e_leaf_clusters)); | |
3950 | le32_add_cpu(&rec->e_int_clusters, | |
3951 | -le32_to_cpu(rec->e_cpos)); | |
3952 | ||
ec20cec7 | 3953 | ocfs2_journal_dirty(handle, bh); |
328d5752 MF |
3954 | } |
3955 | } | |
3956 | ||
d401dc12 JB |
3957 | static int ocfs2_append_rec_to_path(handle_t *handle, |
3958 | struct ocfs2_extent_tree *et, | |
328d5752 MF |
3959 | struct ocfs2_extent_rec *insert_rec, |
3960 | struct ocfs2_path *right_path, | |
3961 | struct ocfs2_path **ret_left_path) | |
3962 | { | |
3963 | int ret, next_free; | |
3964 | struct ocfs2_extent_list *el; | |
3965 | struct ocfs2_path *left_path = NULL; | |
3966 | ||
3967 | *ret_left_path = NULL; | |
3968 | ||
3969 | /* | |
3970 | * This shouldn't happen for non-trees. The extent rec cluster | |
3971 | * count manipulation below only works for interior nodes. | |
3972 | */ | |
3973 | BUG_ON(right_path->p_tree_depth == 0); | |
3974 | ||
3975 | /* | |
3976 | * If our appending insert is at the leftmost edge of a leaf, | |
3977 | * then we might need to update the rightmost records of the | |
3978 | * neighboring path. | |
3979 | */ | |
3980 | el = path_leaf_el(right_path); | |
3981 | next_free = le16_to_cpu(el->l_next_free_rec); | |
3982 | if (next_free == 0 || | |
3983 | (next_free == 1 && ocfs2_is_empty_extent(&el->l_recs[0]))) { | |
3984 | u32 left_cpos; | |
3985 | ||
d401dc12 JB |
3986 | ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci), |
3987 | right_path, &left_cpos); | |
328d5752 MF |
3988 | if (ret) { |
3989 | mlog_errno(ret); | |
dcd0538f MF |
3990 | goto out; |
3991 | } | |
3992 | ||
328d5752 MF |
3993 | mlog(0, "Append may need a left path update. cpos: %u, " |
3994 | "left_cpos: %u\n", le32_to_cpu(insert_rec->e_cpos), | |
3995 | left_cpos); | |
e48edee2 | 3996 | |
328d5752 MF |
3997 | /* |
3998 | * No need to worry if the append is already in the | |
3999 | * leftmost leaf. | |
4000 | */ | |
4001 | if (left_cpos) { | |
ffdd7a54 | 4002 | left_path = ocfs2_new_path_from_path(right_path); |
328d5752 MF |
4003 | if (!left_path) { |
4004 | ret = -ENOMEM; | |
4005 | mlog_errno(ret); | |
4006 | goto out; | |
4007 | } | |
dcd0538f | 4008 | |
d401dc12 | 4009 | ret = ocfs2_find_path(et->et_ci, left_path, |
facdb77f | 4010 | left_cpos); |
328d5752 MF |
4011 | if (ret) { |
4012 | mlog_errno(ret); | |
4013 | goto out; | |
4014 | } | |
dcd0538f | 4015 | |
328d5752 MF |
4016 | /* |
4017 | * ocfs2_insert_path() will pass the left_path to the | |
4018 | * journal for us. | |
4019 | */ | |
4020 | } | |
4021 | } | |
dcd0538f | 4022 | |
d401dc12 | 4023 | ret = ocfs2_journal_access_path(et->et_ci, handle, right_path); |
328d5752 MF |
4024 | if (ret) { |
4025 | mlog_errno(ret); | |
4026 | goto out; | |
dcd0538f MF |
4027 | } |
4028 | ||
d401dc12 | 4029 | ocfs2_adjust_rightmost_records(handle, et, right_path, insert_rec); |
328d5752 | 4030 | |
dcd0538f MF |
4031 | *ret_left_path = left_path; |
4032 | ret = 0; | |
4033 | out: | |
4034 | if (ret != 0) | |
4035 | ocfs2_free_path(left_path); | |
4036 | ||
4037 | return ret; | |
4038 | } | |
4039 | ||
c38e52bb | 4040 | static void ocfs2_split_record(struct ocfs2_extent_tree *et, |
328d5752 MF |
4041 | struct ocfs2_path *left_path, |
4042 | struct ocfs2_path *right_path, | |
4043 | struct ocfs2_extent_rec *split_rec, | |
4044 | enum ocfs2_split_type split) | |
4045 | { | |
4046 | int index; | |
4047 | u32 cpos = le32_to_cpu(split_rec->e_cpos); | |
4048 | struct ocfs2_extent_list *left_el = NULL, *right_el, *insert_el, *el; | |
4049 | struct ocfs2_extent_rec *rec, *tmprec; | |
4050 | ||
c19a28e1 | 4051 | right_el = path_leaf_el(right_path); |
328d5752 MF |
4052 | if (left_path) |
4053 | left_el = path_leaf_el(left_path); | |
4054 | ||
4055 | el = right_el; | |
4056 | insert_el = right_el; | |
4057 | index = ocfs2_search_extent_list(el, cpos); | |
4058 | if (index != -1) { | |
4059 | if (index == 0 && left_path) { | |
4060 | BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0])); | |
4061 | ||
4062 | /* | |
4063 | * This typically means that the record | |
4064 | * started in the left path but moved to the | |
4065 | * right as a result of rotation. We either | |
4066 | * move the existing record to the left, or we | |
4067 | * do the later insert there. | |
4068 | * | |
4069 | * In this case, the left path should always | |
4070 | * exist as the rotate code will have passed | |
4071 | * it back for a post-insert update. | |
4072 | */ | |
4073 | ||
4074 | if (split == SPLIT_LEFT) { | |
4075 | /* | |
4076 | * It's a left split. Since we know | |
4077 | * that the rotate code gave us an | |
4078 | * empty extent in the left path, we | |
4079 | * can just do the insert there. | |
4080 | */ | |
4081 | insert_el = left_el; | |
4082 | } else { | |
4083 | /* | |
4084 | * Right split - we have to move the | |
4085 | * existing record over to the left | |
4086 | * leaf. The insert will be into the | |
4087 | * newly created empty extent in the | |
4088 | * right leaf. | |
4089 | */ | |
4090 | tmprec = &right_el->l_recs[index]; | |
4091 | ocfs2_rotate_leaf(left_el, tmprec); | |
4092 | el = left_el; | |
4093 | ||
4094 | memset(tmprec, 0, sizeof(*tmprec)); | |
4095 | index = ocfs2_search_extent_list(left_el, cpos); | |
4096 | BUG_ON(index == -1); | |
4097 | } | |
4098 | } | |
4099 | } else { | |
4100 | BUG_ON(!left_path); | |
4101 | BUG_ON(!ocfs2_is_empty_extent(&left_el->l_recs[0])); | |
4102 | /* | |
4103 | * Left path is easy - we can just allow the insert to | |
4104 | * happen. | |
4105 | */ | |
4106 | el = left_el; | |
4107 | insert_el = left_el; | |
4108 | index = ocfs2_search_extent_list(el, cpos); | |
4109 | BUG_ON(index == -1); | |
4110 | } | |
4111 | ||
4112 | rec = &el->l_recs[index]; | |
c38e52bb JB |
4113 | ocfs2_subtract_from_rec(ocfs2_metadata_cache_get_super(et->et_ci), |
4114 | split, rec, split_rec); | |
328d5752 MF |
4115 | ocfs2_rotate_leaf(insert_el, split_rec); |
4116 | } | |
4117 | ||
dcd0538f | 4118 | /* |
e7d4cb6b TM |
4119 | * This function only does inserts on an allocation b-tree. For tree |
4120 | * depth = 0, ocfs2_insert_at_leaf() is called directly. | |
dcd0538f MF |
4121 | * |
4122 | * right_path is the path we want to do the actual insert | |
4123 | * in. left_path should only be passed in if we need to update that | |
4124 | * portion of the tree after an edge insert. | |
4125 | */ | |
3505bec0 | 4126 | static int ocfs2_insert_path(handle_t *handle, |
7dc02805 | 4127 | struct ocfs2_extent_tree *et, |
dcd0538f MF |
4128 | struct ocfs2_path *left_path, |
4129 | struct ocfs2_path *right_path, | |
4130 | struct ocfs2_extent_rec *insert_rec, | |
4131 | struct ocfs2_insert_type *insert) | |
4132 | { | |
4133 | int ret, subtree_index; | |
4134 | struct buffer_head *leaf_bh = path_leaf_bh(right_path); | |
dcd0538f | 4135 | |
dcd0538f | 4136 | if (left_path) { |
dcd0538f MF |
4137 | /* |
4138 | * There's a chance that left_path got passed back to | |
4139 | * us without being accounted for in the | |
4140 | * journal. Extend our transaction here to be sure we | |
4141 | * can change those blocks. | |
4142 | */ | |
c901fb00 | 4143 | ret = ocfs2_extend_trans(handle, left_path->p_tree_depth); |
dcd0538f MF |
4144 | if (ret < 0) { |
4145 | mlog_errno(ret); | |
4146 | goto out; | |
4147 | } | |
4148 | ||
7dc02805 | 4149 | ret = ocfs2_journal_access_path(et->et_ci, handle, left_path); |
dcd0538f MF |
4150 | if (ret < 0) { |
4151 | mlog_errno(ret); | |
4152 | goto out; | |
4153 | } | |
4154 | } | |
4155 | ||
e8aed345 MF |
4156 | /* |
4157 | * Pass both paths to the journal. The majority of inserts | |
4158 | * will be touching all components anyway. | |
4159 | */ | |
7dc02805 | 4160 | ret = ocfs2_journal_access_path(et->et_ci, handle, right_path); |
e8aed345 MF |
4161 | if (ret < 0) { |
4162 | mlog_errno(ret); | |
4163 | goto out; | |
4164 | } | |
4165 | ||
328d5752 MF |
4166 | if (insert->ins_split != SPLIT_NONE) { |
4167 | /* | |
4168 | * We could call ocfs2_insert_at_leaf() for some types | |
c78bad11 | 4169 | * of splits, but it's easier to just let one separate |
328d5752 MF |
4170 | * function sort it all out. |
4171 | */ | |
c38e52bb | 4172 | ocfs2_split_record(et, left_path, right_path, |
328d5752 | 4173 | insert_rec, insert->ins_split); |
e8aed345 MF |
4174 | |
4175 | /* | |
4176 | * Split might have modified either leaf and we don't | |
4177 | * have a guarantee that the later edge insert will | |
4178 | * dirty this for us. | |
4179 | */ | |
4180 | if (left_path) | |
ec20cec7 JB |
4181 | ocfs2_journal_dirty(handle, |
4182 | path_leaf_bh(left_path)); | |
328d5752 | 4183 | } else |
d5628623 JB |
4184 | ocfs2_insert_at_leaf(et, insert_rec, path_leaf_el(right_path), |
4185 | insert); | |
dcd0538f | 4186 | |
ec20cec7 | 4187 | ocfs2_journal_dirty(handle, leaf_bh); |
dcd0538f MF |
4188 | |
4189 | if (left_path) { | |
4190 | /* | |
4191 | * The rotate code has indicated that we need to fix | |
4192 | * up portions of the tree after the insert. | |
4193 | * | |
4194 | * XXX: Should we extend the transaction here? | |
4195 | */ | |
7dc02805 | 4196 | subtree_index = ocfs2_find_subtree_root(et, left_path, |
dcd0538f | 4197 | right_path); |
4619c73e JB |
4198 | ocfs2_complete_edge_insert(handle, left_path, right_path, |
4199 | subtree_index); | |
dcd0538f MF |
4200 | } |
4201 | ||
4202 | ret = 0; | |
4203 | out: | |
4204 | return ret; | |
4205 | } | |
4206 | ||
3505bec0 | 4207 | static int ocfs2_do_insert_extent(handle_t *handle, |
e7d4cb6b | 4208 | struct ocfs2_extent_tree *et, |
dcd0538f MF |
4209 | struct ocfs2_extent_rec *insert_rec, |
4210 | struct ocfs2_insert_type *type) | |
4211 | { | |
4212 | int ret, rotate = 0; | |
4213 | u32 cpos; | |
4214 | struct ocfs2_path *right_path = NULL; | |
4215 | struct ocfs2_path *left_path = NULL; | |
dcd0538f MF |
4216 | struct ocfs2_extent_list *el; |
4217 | ||
ce1d9ea6 | 4218 | el = et->et_root_el; |
dcd0538f | 4219 | |
d9a0a1f8 | 4220 | ret = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 4221 | OCFS2_JOURNAL_ACCESS_WRITE); |
dcd0538f MF |
4222 | if (ret) { |
4223 | mlog_errno(ret); | |
4224 | goto out; | |
4225 | } | |
4226 | ||
4227 | if (le16_to_cpu(el->l_tree_depth) == 0) { | |
d5628623 | 4228 | ocfs2_insert_at_leaf(et, insert_rec, el, type); |
dcd0538f MF |
4229 | goto out_update_clusters; |
4230 | } | |
4231 | ||
ffdd7a54 | 4232 | right_path = ocfs2_new_path_from_et(et); |
dcd0538f MF |
4233 | if (!right_path) { |
4234 | ret = -ENOMEM; | |
4235 | mlog_errno(ret); | |
4236 | goto out; | |
4237 | } | |
4238 | ||
4239 | /* | |
4240 | * Determine the path to start with. Rotations need the | |
4241 | * rightmost path, everything else can go directly to the | |
4242 | * target leaf. | |
4243 | */ | |
4244 | cpos = le32_to_cpu(insert_rec->e_cpos); | |
4245 | if (type->ins_appending == APPEND_NONE && | |
4246 | type->ins_contig == CONTIG_NONE) { | |
4247 | rotate = 1; | |
4248 | cpos = UINT_MAX; | |
4249 | } | |
4250 | ||
facdb77f | 4251 | ret = ocfs2_find_path(et->et_ci, right_path, cpos); |
dcd0538f MF |
4252 | if (ret) { |
4253 | mlog_errno(ret); | |
4254 | goto out; | |
4255 | } | |
4256 | ||
4257 | /* | |
4258 | * Rotations and appends need special treatment - they modify | |
4259 | * parts of the tree's above them. | |
4260 | * | |
4261 | * Both might pass back a path immediate to the left of the | |
4262 | * one being inserted to. This will be cause | |
4263 | * ocfs2_insert_path() to modify the rightmost records of | |
4264 | * left_path to account for an edge insert. | |
4265 | * | |
4266 | * XXX: When modifying this code, keep in mind that an insert | |
4267 | * can wind up skipping both of these two special cases... | |
4268 | */ | |
4269 | if (rotate) { | |
1bbf0b8d | 4270 | ret = ocfs2_rotate_tree_right(handle, et, type->ins_split, |
dcd0538f MF |
4271 | le32_to_cpu(insert_rec->e_cpos), |
4272 | right_path, &left_path); | |
4273 | if (ret) { | |
4274 | mlog_errno(ret); | |
4275 | goto out; | |
4276 | } | |
e8aed345 MF |
4277 | |
4278 | /* | |
4279 | * ocfs2_rotate_tree_right() might have extended the | |
4280 | * transaction without re-journaling our tree root. | |
4281 | */ | |
d9a0a1f8 | 4282 | ret = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 4283 | OCFS2_JOURNAL_ACCESS_WRITE); |
e8aed345 MF |
4284 | if (ret) { |
4285 | mlog_errno(ret); | |
4286 | goto out; | |
4287 | } | |
dcd0538f MF |
4288 | } else if (type->ins_appending == APPEND_TAIL |
4289 | && type->ins_contig != CONTIG_LEFT) { | |
d401dc12 | 4290 | ret = ocfs2_append_rec_to_path(handle, et, insert_rec, |
dcd0538f MF |
4291 | right_path, &left_path); |
4292 | if (ret) { | |
4293 | mlog_errno(ret); | |
4294 | goto out; | |
4295 | } | |
4296 | } | |
4297 | ||
3505bec0 | 4298 | ret = ocfs2_insert_path(handle, et, left_path, right_path, |
dcd0538f MF |
4299 | insert_rec, type); |
4300 | if (ret) { | |
4301 | mlog_errno(ret); | |
4302 | goto out; | |
4303 | } | |
4304 | ||
4305 | out_update_clusters: | |
328d5752 | 4306 | if (type->ins_split == SPLIT_NONE) |
6136ca5f | 4307 | ocfs2_et_update_clusters(et, |
35dc0aa3 | 4308 | le16_to_cpu(insert_rec->e_leaf_clusters)); |
dcd0538f | 4309 | |
ec20cec7 | 4310 | ocfs2_journal_dirty(handle, et->et_root_bh); |
dcd0538f MF |
4311 | |
4312 | out: | |
4313 | ocfs2_free_path(left_path); | |
4314 | ocfs2_free_path(right_path); | |
4315 | ||
4316 | return ret; | |
4317 | } | |
4318 | ||
328d5752 | 4319 | static enum ocfs2_contig_type |
a2970291 JB |
4320 | ocfs2_figure_merge_contig_type(struct ocfs2_extent_tree *et, |
4321 | struct ocfs2_path *path, | |
328d5752 MF |
4322 | struct ocfs2_extent_list *el, int index, |
4323 | struct ocfs2_extent_rec *split_rec) | |
4324 | { | |
ad5a4d70 | 4325 | int status; |
328d5752 | 4326 | enum ocfs2_contig_type ret = CONTIG_NONE; |
ad5a4d70 TM |
4327 | u32 left_cpos, right_cpos; |
4328 | struct ocfs2_extent_rec *rec = NULL; | |
4329 | struct ocfs2_extent_list *new_el; | |
4330 | struct ocfs2_path *left_path = NULL, *right_path = NULL; | |
4331 | struct buffer_head *bh; | |
4332 | struct ocfs2_extent_block *eb; | |
a2970291 | 4333 | struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci); |
ad5a4d70 TM |
4334 | |
4335 | if (index > 0) { | |
4336 | rec = &el->l_recs[index - 1]; | |
4337 | } else if (path->p_tree_depth > 0) { | |
a2970291 | 4338 | status = ocfs2_find_cpos_for_left_leaf(sb, path, &left_cpos); |
ad5a4d70 TM |
4339 | if (status) |
4340 | goto out; | |
4341 | ||
4342 | if (left_cpos != 0) { | |
ffdd7a54 | 4343 | left_path = ocfs2_new_path_from_path(path); |
ad5a4d70 TM |
4344 | if (!left_path) |
4345 | goto out; | |
4346 | ||
a2970291 JB |
4347 | status = ocfs2_find_path(et->et_ci, left_path, |
4348 | left_cpos); | |
ad5a4d70 TM |
4349 | if (status) |
4350 | goto out; | |
4351 | ||
4352 | new_el = path_leaf_el(left_path); | |
4353 | ||
4354 | if (le16_to_cpu(new_el->l_next_free_rec) != | |
4355 | le16_to_cpu(new_el->l_count)) { | |
4356 | bh = path_leaf_bh(left_path); | |
4357 | eb = (struct ocfs2_extent_block *)bh->b_data; | |
a2970291 | 4358 | ocfs2_error(sb, |
5e96581a JB |
4359 | "Extent block #%llu has an " |
4360 | "invalid l_next_free_rec of " | |
4361 | "%d. It should have " | |
4362 | "matched the l_count of %d", | |
4363 | (unsigned long long)le64_to_cpu(eb->h_blkno), | |
4364 | le16_to_cpu(new_el->l_next_free_rec), | |
4365 | le16_to_cpu(new_el->l_count)); | |
4366 | status = -EINVAL; | |
ad5a4d70 TM |
4367 | goto out; |
4368 | } | |
4369 | rec = &new_el->l_recs[ | |
4370 | le16_to_cpu(new_el->l_next_free_rec) - 1]; | |
4371 | } | |
4372 | } | |
328d5752 MF |
4373 | |
4374 | /* | |
4375 | * We're careful to check for an empty extent record here - | |
4376 | * the merge code will know what to do if it sees one. | |
4377 | */ | |
ad5a4d70 | 4378 | if (rec) { |
328d5752 MF |
4379 | if (index == 1 && ocfs2_is_empty_extent(rec)) { |
4380 | if (split_rec->e_cpos == el->l_recs[index].e_cpos) | |
4381 | ret = CONTIG_RIGHT; | |
4382 | } else { | |
853a3a14 | 4383 | ret = ocfs2_et_extent_contig(et, rec, split_rec); |
328d5752 MF |
4384 | } |
4385 | } | |
4386 | ||
ad5a4d70 TM |
4387 | rec = NULL; |
4388 | if (index < (le16_to_cpu(el->l_next_free_rec) - 1)) | |
4389 | rec = &el->l_recs[index + 1]; | |
4390 | else if (le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count) && | |
4391 | path->p_tree_depth > 0) { | |
a2970291 | 4392 | status = ocfs2_find_cpos_for_right_leaf(sb, path, &right_cpos); |
ad5a4d70 TM |
4393 | if (status) |
4394 | goto out; | |
4395 | ||
4396 | if (right_cpos == 0) | |
4397 | goto out; | |
4398 | ||
ffdd7a54 | 4399 | right_path = ocfs2_new_path_from_path(path); |
ad5a4d70 TM |
4400 | if (!right_path) |
4401 | goto out; | |
4402 | ||
a2970291 | 4403 | status = ocfs2_find_path(et->et_ci, right_path, right_cpos); |
ad5a4d70 TM |
4404 | if (status) |
4405 | goto out; | |
4406 | ||
4407 | new_el = path_leaf_el(right_path); | |
4408 | rec = &new_el->l_recs[0]; | |
4409 | if (ocfs2_is_empty_extent(rec)) { | |
4410 | if (le16_to_cpu(new_el->l_next_free_rec) <= 1) { | |
4411 | bh = path_leaf_bh(right_path); | |
4412 | eb = (struct ocfs2_extent_block *)bh->b_data; | |
a2970291 | 4413 | ocfs2_error(sb, |
5e96581a JB |
4414 | "Extent block #%llu has an " |
4415 | "invalid l_next_free_rec of %d", | |
4416 | (unsigned long long)le64_to_cpu(eb->h_blkno), | |
4417 | le16_to_cpu(new_el->l_next_free_rec)); | |
4418 | status = -EINVAL; | |
ad5a4d70 TM |
4419 | goto out; |
4420 | } | |
4421 | rec = &new_el->l_recs[1]; | |
4422 | } | |
4423 | } | |
4424 | ||
4425 | if (rec) { | |
328d5752 MF |
4426 | enum ocfs2_contig_type contig_type; |
4427 | ||
853a3a14 | 4428 | contig_type = ocfs2_et_extent_contig(et, rec, split_rec); |
328d5752 MF |
4429 | |
4430 | if (contig_type == CONTIG_LEFT && ret == CONTIG_RIGHT) | |
4431 | ret = CONTIG_LEFTRIGHT; | |
4432 | else if (ret == CONTIG_NONE) | |
4433 | ret = contig_type; | |
4434 | } | |
4435 | ||
ad5a4d70 TM |
4436 | out: |
4437 | if (left_path) | |
4438 | ocfs2_free_path(left_path); | |
4439 | if (right_path) | |
4440 | ocfs2_free_path(right_path); | |
4441 | ||
328d5752 MF |
4442 | return ret; |
4443 | } | |
4444 | ||
1ef61b33 | 4445 | static void ocfs2_figure_contig_type(struct ocfs2_extent_tree *et, |
dcd0538f MF |
4446 | struct ocfs2_insert_type *insert, |
4447 | struct ocfs2_extent_list *el, | |
1ef61b33 | 4448 | struct ocfs2_extent_rec *insert_rec) |
dcd0538f MF |
4449 | { |
4450 | int i; | |
4451 | enum ocfs2_contig_type contig_type = CONTIG_NONE; | |
4452 | ||
e48edee2 MF |
4453 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); |
4454 | ||
dcd0538f | 4455 | for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) { |
853a3a14 TM |
4456 | contig_type = ocfs2_et_extent_contig(et, &el->l_recs[i], |
4457 | insert_rec); | |
dcd0538f MF |
4458 | if (contig_type != CONTIG_NONE) { |
4459 | insert->ins_contig_index = i; | |
4460 | break; | |
4461 | } | |
4462 | } | |
4463 | insert->ins_contig = contig_type; | |
ca12b7c4 TM |
4464 | |
4465 | if (insert->ins_contig != CONTIG_NONE) { | |
4466 | struct ocfs2_extent_rec *rec = | |
4467 | &el->l_recs[insert->ins_contig_index]; | |
4468 | unsigned int len = le16_to_cpu(rec->e_leaf_clusters) + | |
4469 | le16_to_cpu(insert_rec->e_leaf_clusters); | |
4470 | ||
4471 | /* | |
4472 | * Caller might want us to limit the size of extents, don't | |
4473 | * calculate contiguousness if we might exceed that limit. | |
4474 | */ | |
ce1d9ea6 JB |
4475 | if (et->et_max_leaf_clusters && |
4476 | (len > et->et_max_leaf_clusters)) | |
ca12b7c4 TM |
4477 | insert->ins_contig = CONTIG_NONE; |
4478 | } | |
dcd0538f MF |
4479 | } |
4480 | ||
4481 | /* | |
4482 | * This should only be called against the righmost leaf extent list. | |
4483 | * | |
4484 | * ocfs2_figure_appending_type() will figure out whether we'll have to | |
4485 | * insert at the tail of the rightmost leaf. | |
4486 | * | |
e7d4cb6b TM |
4487 | * This should also work against the root extent list for tree's with 0 |
4488 | * depth. If we consider the root extent list to be the rightmost leaf node | |
dcd0538f MF |
4489 | * then the logic here makes sense. |
4490 | */ | |
4491 | static void ocfs2_figure_appending_type(struct ocfs2_insert_type *insert, | |
4492 | struct ocfs2_extent_list *el, | |
4493 | struct ocfs2_extent_rec *insert_rec) | |
4494 | { | |
4495 | int i; | |
4496 | u32 cpos = le32_to_cpu(insert_rec->e_cpos); | |
4497 | struct ocfs2_extent_rec *rec; | |
4498 | ||
4499 | insert->ins_appending = APPEND_NONE; | |
4500 | ||
e48edee2 | 4501 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); |
dcd0538f MF |
4502 | |
4503 | if (!el->l_next_free_rec) | |
4504 | goto set_tail_append; | |
4505 | ||
4506 | if (ocfs2_is_empty_extent(&el->l_recs[0])) { | |
4507 | /* Were all records empty? */ | |
4508 | if (le16_to_cpu(el->l_next_free_rec) == 1) | |
4509 | goto set_tail_append; | |
4510 | } | |
4511 | ||
4512 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
4513 | rec = &el->l_recs[i]; | |
4514 | ||
e48edee2 MF |
4515 | if (cpos >= |
4516 | (le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters))) | |
dcd0538f MF |
4517 | goto set_tail_append; |
4518 | ||
4519 | return; | |
4520 | ||
4521 | set_tail_append: | |
4522 | insert->ins_appending = APPEND_TAIL; | |
4523 | } | |
4524 | ||
4525 | /* | |
4526 | * Helper function called at the begining of an insert. | |
4527 | * | |
4528 | * This computes a few things that are commonly used in the process of | |
4529 | * inserting into the btree: | |
4530 | * - Whether the new extent is contiguous with an existing one. | |
4531 | * - The current tree depth. | |
4532 | * - Whether the insert is an appending one. | |
4533 | * - The total # of free records in the tree. | |
4534 | * | |
4535 | * All of the information is stored on the ocfs2_insert_type | |
4536 | * structure. | |
4537 | */ | |
627961b7 | 4538 | static int ocfs2_figure_insert_type(struct ocfs2_extent_tree *et, |
dcd0538f MF |
4539 | struct buffer_head **last_eb_bh, |
4540 | struct ocfs2_extent_rec *insert_rec, | |
c77534f6 | 4541 | int *free_records, |
dcd0538f MF |
4542 | struct ocfs2_insert_type *insert) |
4543 | { | |
4544 | int ret; | |
dcd0538f MF |
4545 | struct ocfs2_extent_block *eb; |
4546 | struct ocfs2_extent_list *el; | |
4547 | struct ocfs2_path *path = NULL; | |
4548 | struct buffer_head *bh = NULL; | |
4549 | ||
328d5752 MF |
4550 | insert->ins_split = SPLIT_NONE; |
4551 | ||
ce1d9ea6 | 4552 | el = et->et_root_el; |
dcd0538f MF |
4553 | insert->ins_tree_depth = le16_to_cpu(el->l_tree_depth); |
4554 | ||
4555 | if (el->l_tree_depth) { | |
4556 | /* | |
4557 | * If we have tree depth, we read in the | |
4558 | * rightmost extent block ahead of time as | |
4559 | * ocfs2_figure_insert_type() and ocfs2_add_branch() | |
4560 | * may want it later. | |
4561 | */ | |
3d03a305 | 4562 | ret = ocfs2_read_extent_block(et->et_ci, |
5e96581a JB |
4563 | ocfs2_et_get_last_eb_blk(et), |
4564 | &bh); | |
dcd0538f MF |
4565 | if (ret) { |
4566 | mlog_exit(ret); | |
4567 | goto out; | |
4568 | } | |
ccd979bd | 4569 | eb = (struct ocfs2_extent_block *) bh->b_data; |
ccd979bd | 4570 | el = &eb->h_list; |
dcd0538f | 4571 | } |
ccd979bd | 4572 | |
dcd0538f MF |
4573 | /* |
4574 | * Unless we have a contiguous insert, we'll need to know if | |
4575 | * there is room left in our allocation tree for another | |
4576 | * extent record. | |
4577 | * | |
4578 | * XXX: This test is simplistic, we can search for empty | |
4579 | * extent records too. | |
4580 | */ | |
c77534f6 | 4581 | *free_records = le16_to_cpu(el->l_count) - |
dcd0538f MF |
4582 | le16_to_cpu(el->l_next_free_rec); |
4583 | ||
4584 | if (!insert->ins_tree_depth) { | |
1ef61b33 | 4585 | ocfs2_figure_contig_type(et, insert, el, insert_rec); |
dcd0538f MF |
4586 | ocfs2_figure_appending_type(insert, el, insert_rec); |
4587 | return 0; | |
ccd979bd MF |
4588 | } |
4589 | ||
ffdd7a54 | 4590 | path = ocfs2_new_path_from_et(et); |
dcd0538f MF |
4591 | if (!path) { |
4592 | ret = -ENOMEM; | |
4593 | mlog_errno(ret); | |
4594 | goto out; | |
4595 | } | |
ccd979bd | 4596 | |
dcd0538f MF |
4597 | /* |
4598 | * In the case that we're inserting past what the tree | |
4599 | * currently accounts for, ocfs2_find_path() will return for | |
4600 | * us the rightmost tree path. This is accounted for below in | |
4601 | * the appending code. | |
4602 | */ | |
facdb77f | 4603 | ret = ocfs2_find_path(et->et_ci, path, le32_to_cpu(insert_rec->e_cpos)); |
dcd0538f MF |
4604 | if (ret) { |
4605 | mlog_errno(ret); | |
4606 | goto out; | |
4607 | } | |
ccd979bd | 4608 | |
dcd0538f MF |
4609 | el = path_leaf_el(path); |
4610 | ||
4611 | /* | |
4612 | * Now that we have the path, there's two things we want to determine: | |
4613 | * 1) Contiguousness (also set contig_index if this is so) | |
4614 | * | |
4615 | * 2) Are we doing an append? We can trivially break this up | |
4616 | * into two types of appends: simple record append, or a | |
4617 | * rotate inside the tail leaf. | |
4618 | */ | |
1ef61b33 | 4619 | ocfs2_figure_contig_type(et, insert, el, insert_rec); |
dcd0538f MF |
4620 | |
4621 | /* | |
4622 | * The insert code isn't quite ready to deal with all cases of | |
4623 | * left contiguousness. Specifically, if it's an insert into | |
4624 | * the 1st record in a leaf, it will require the adjustment of | |
e48edee2 | 4625 | * cluster count on the last record of the path directly to it's |
dcd0538f MF |
4626 | * left. For now, just catch that case and fool the layers |
4627 | * above us. This works just fine for tree_depth == 0, which | |
4628 | * is why we allow that above. | |
4629 | */ | |
4630 | if (insert->ins_contig == CONTIG_LEFT && | |
4631 | insert->ins_contig_index == 0) | |
4632 | insert->ins_contig = CONTIG_NONE; | |
4633 | ||
4634 | /* | |
4635 | * Ok, so we can simply compare against last_eb to figure out | |
4636 | * whether the path doesn't exist. This will only happen in | |
4637 | * the case that we're doing a tail append, so maybe we can | |
4638 | * take advantage of that information somehow. | |
4639 | */ | |
35dc0aa3 | 4640 | if (ocfs2_et_get_last_eb_blk(et) == |
e7d4cb6b | 4641 | path_leaf_bh(path)->b_blocknr) { |
dcd0538f MF |
4642 | /* |
4643 | * Ok, ocfs2_find_path() returned us the rightmost | |
4644 | * tree path. This might be an appending insert. There are | |
4645 | * two cases: | |
4646 | * 1) We're doing a true append at the tail: | |
4647 | * -This might even be off the end of the leaf | |
4648 | * 2) We're "appending" by rotating in the tail | |
4649 | */ | |
4650 | ocfs2_figure_appending_type(insert, el, insert_rec); | |
4651 | } | |
4652 | ||
4653 | out: | |
4654 | ocfs2_free_path(path); | |
4655 | ||
4656 | if (ret == 0) | |
4657 | *last_eb_bh = bh; | |
4658 | else | |
4659 | brelse(bh); | |
4660 | return ret; | |
ccd979bd MF |
4661 | } |
4662 | ||
dcd0538f | 4663 | /* |
cc79d8c1 | 4664 | * Insert an extent into a btree. |
dcd0538f | 4665 | * |
cc79d8c1 | 4666 | * The caller needs to update the owning btree's cluster count. |
dcd0538f | 4667 | */ |
cc79d8c1 | 4668 | int ocfs2_insert_extent(handle_t *handle, |
f99b9b7c JB |
4669 | struct ocfs2_extent_tree *et, |
4670 | u32 cpos, | |
4671 | u64 start_blk, | |
4672 | u32 new_clusters, | |
4673 | u8 flags, | |
4674 | struct ocfs2_alloc_context *meta_ac) | |
ccd979bd | 4675 | { |
c3afcbb3 | 4676 | int status; |
c77534f6 | 4677 | int uninitialized_var(free_records); |
ccd979bd | 4678 | struct buffer_head *last_eb_bh = NULL; |
dcd0538f MF |
4679 | struct ocfs2_insert_type insert = {0, }; |
4680 | struct ocfs2_extent_rec rec; | |
4681 | ||
cc79d8c1 JB |
4682 | mlog(0, "add %u clusters at position %u to owner %llu\n", |
4683 | new_clusters, cpos, | |
4684 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci)); | |
dcd0538f | 4685 | |
e48edee2 | 4686 | memset(&rec, 0, sizeof(rec)); |
dcd0538f MF |
4687 | rec.e_cpos = cpu_to_le32(cpos); |
4688 | rec.e_blkno = cpu_to_le64(start_blk); | |
e48edee2 | 4689 | rec.e_leaf_clusters = cpu_to_le16(new_clusters); |
2ae99a60 | 4690 | rec.e_flags = flags; |
6136ca5f | 4691 | status = ocfs2_et_insert_check(et, &rec); |
1e61ee79 JB |
4692 | if (status) { |
4693 | mlog_errno(status); | |
4694 | goto bail; | |
4695 | } | |
dcd0538f | 4696 | |
627961b7 | 4697 | status = ocfs2_figure_insert_type(et, &last_eb_bh, &rec, |
c77534f6 | 4698 | &free_records, &insert); |
dcd0538f MF |
4699 | if (status < 0) { |
4700 | mlog_errno(status); | |
4701 | goto bail; | |
ccd979bd MF |
4702 | } |
4703 | ||
dcd0538f MF |
4704 | mlog(0, "Insert.appending: %u, Insert.Contig: %u, " |
4705 | "Insert.contig_index: %d, Insert.free_records: %d, " | |
4706 | "Insert.tree_depth: %d\n", | |
4707 | insert.ins_appending, insert.ins_contig, insert.ins_contig_index, | |
c77534f6 | 4708 | free_records, insert.ins_tree_depth); |
ccd979bd | 4709 | |
c77534f6 | 4710 | if (insert.ins_contig == CONTIG_NONE && free_records == 0) { |
d401dc12 | 4711 | status = ocfs2_grow_tree(handle, et, |
328d5752 | 4712 | &insert.ins_tree_depth, &last_eb_bh, |
c3afcbb3 MF |
4713 | meta_ac); |
4714 | if (status) { | |
ccd979bd MF |
4715 | mlog_errno(status); |
4716 | goto bail; | |
4717 | } | |
ccd979bd MF |
4718 | } |
4719 | ||
dcd0538f | 4720 | /* Finally, we can add clusters. This might rotate the tree for us. */ |
3505bec0 | 4721 | status = ocfs2_do_insert_extent(handle, et, &rec, &insert); |
ccd979bd MF |
4722 | if (status < 0) |
4723 | mlog_errno(status); | |
92ba470c JB |
4724 | else |
4725 | ocfs2_et_extent_map_insert(et, &rec); | |
ccd979bd MF |
4726 | |
4727 | bail: | |
a81cb88b | 4728 | brelse(last_eb_bh); |
ccd979bd | 4729 | |
f56654c4 TM |
4730 | mlog_exit(status); |
4731 | return status; | |
4732 | } | |
4733 | ||
0eb8d47e TM |
4734 | /* |
4735 | * Allcate and add clusters into the extent b-tree. | |
4736 | * The new clusters(clusters_to_add) will be inserted at logical_offset. | |
f99b9b7c | 4737 | * The extent b-tree's root is specified by et, and |
0eb8d47e TM |
4738 | * it is not limited to the file storage. Any extent tree can use this |
4739 | * function if it implements the proper ocfs2_extent_tree. | |
4740 | */ | |
cbee7e1a JB |
4741 | int ocfs2_add_clusters_in_btree(handle_t *handle, |
4742 | struct ocfs2_extent_tree *et, | |
0eb8d47e TM |
4743 | u32 *logical_offset, |
4744 | u32 clusters_to_add, | |
4745 | int mark_unwritten, | |
0eb8d47e TM |
4746 | struct ocfs2_alloc_context *data_ac, |
4747 | struct ocfs2_alloc_context *meta_ac, | |
f99b9b7c | 4748 | enum ocfs2_alloc_restarted *reason_ret) |
0eb8d47e TM |
4749 | { |
4750 | int status = 0; | |
4751 | int free_extents; | |
4752 | enum ocfs2_alloc_restarted reason = RESTART_NONE; | |
4753 | u32 bit_off, num_bits; | |
4754 | u64 block; | |
4755 | u8 flags = 0; | |
cbee7e1a JB |
4756 | struct ocfs2_super *osb = |
4757 | OCFS2_SB(ocfs2_metadata_cache_get_super(et->et_ci)); | |
0eb8d47e TM |
4758 | |
4759 | BUG_ON(!clusters_to_add); | |
4760 | ||
4761 | if (mark_unwritten) | |
4762 | flags = OCFS2_EXT_UNWRITTEN; | |
4763 | ||
3d03a305 | 4764 | free_extents = ocfs2_num_free_extents(osb, et); |
0eb8d47e TM |
4765 | if (free_extents < 0) { |
4766 | status = free_extents; | |
4767 | mlog_errno(status); | |
4768 | goto leave; | |
4769 | } | |
4770 | ||
4771 | /* there are two cases which could cause us to EAGAIN in the | |
4772 | * we-need-more-metadata case: | |
4773 | * 1) we haven't reserved *any* | |
4774 | * 2) we are so fragmented, we've needed to add metadata too | |
4775 | * many times. */ | |
4776 | if (!free_extents && !meta_ac) { | |
4777 | mlog(0, "we haven't reserved any metadata!\n"); | |
4778 | status = -EAGAIN; | |
4779 | reason = RESTART_META; | |
4780 | goto leave; | |
4781 | } else if ((!free_extents) | |
4782 | && (ocfs2_alloc_context_bits_left(meta_ac) | |
f99b9b7c | 4783 | < ocfs2_extend_meta_needed(et->et_root_el))) { |
0eb8d47e TM |
4784 | mlog(0, "filesystem is really fragmented...\n"); |
4785 | status = -EAGAIN; | |
4786 | reason = RESTART_META; | |
4787 | goto leave; | |
4788 | } | |
4789 | ||
1ed9b777 | 4790 | status = __ocfs2_claim_clusters(handle, data_ac, 1, |
0eb8d47e TM |
4791 | clusters_to_add, &bit_off, &num_bits); |
4792 | if (status < 0) { | |
4793 | if (status != -ENOSPC) | |
4794 | mlog_errno(status); | |
4795 | goto leave; | |
4796 | } | |
4797 | ||
4798 | BUG_ON(num_bits > clusters_to_add); | |
4799 | ||
13723d00 | 4800 | /* reserve our write early -- insert_extent may update the tree root */ |
d9a0a1f8 | 4801 | status = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 4802 | OCFS2_JOURNAL_ACCESS_WRITE); |
0eb8d47e TM |
4803 | if (status < 0) { |
4804 | mlog_errno(status); | |
4805 | goto leave; | |
4806 | } | |
4807 | ||
4808 | block = ocfs2_clusters_to_blocks(osb->sb, bit_off); | |
cbee7e1a JB |
4809 | mlog(0, "Allocating %u clusters at block %u for owner %llu\n", |
4810 | num_bits, bit_off, | |
4811 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci)); | |
cc79d8c1 | 4812 | status = ocfs2_insert_extent(handle, et, *logical_offset, block, |
f99b9b7c | 4813 | num_bits, flags, meta_ac); |
0eb8d47e TM |
4814 | if (status < 0) { |
4815 | mlog_errno(status); | |
4816 | goto leave; | |
4817 | } | |
4818 | ||
ec20cec7 | 4819 | ocfs2_journal_dirty(handle, et->et_root_bh); |
0eb8d47e TM |
4820 | |
4821 | clusters_to_add -= num_bits; | |
4822 | *logical_offset += num_bits; | |
4823 | ||
4824 | if (clusters_to_add) { | |
4825 | mlog(0, "need to alloc once more, wanted = %u\n", | |
4826 | clusters_to_add); | |
4827 | status = -EAGAIN; | |
4828 | reason = RESTART_TRANS; | |
4829 | } | |
4830 | ||
4831 | leave: | |
4832 | mlog_exit(status); | |
4833 | if (reason_ret) | |
4834 | *reason_ret = reason; | |
4835 | return status; | |
4836 | } | |
4837 | ||
328d5752 MF |
4838 | static void ocfs2_make_right_split_rec(struct super_block *sb, |
4839 | struct ocfs2_extent_rec *split_rec, | |
4840 | u32 cpos, | |
4841 | struct ocfs2_extent_rec *rec) | |
4842 | { | |
4843 | u32 rec_cpos = le32_to_cpu(rec->e_cpos); | |
4844 | u32 rec_range = rec_cpos + le16_to_cpu(rec->e_leaf_clusters); | |
4845 | ||
4846 | memset(split_rec, 0, sizeof(struct ocfs2_extent_rec)); | |
4847 | ||
4848 | split_rec->e_cpos = cpu_to_le32(cpos); | |
4849 | split_rec->e_leaf_clusters = cpu_to_le16(rec_range - cpos); | |
4850 | ||
4851 | split_rec->e_blkno = rec->e_blkno; | |
4852 | le64_add_cpu(&split_rec->e_blkno, | |
4853 | ocfs2_clusters_to_blocks(sb, cpos - rec_cpos)); | |
4854 | ||
4855 | split_rec->e_flags = rec->e_flags; | |
4856 | } | |
4857 | ||
d231129f | 4858 | static int ocfs2_split_and_insert(handle_t *handle, |
e7d4cb6b | 4859 | struct ocfs2_extent_tree *et, |
d231129f | 4860 | struct ocfs2_path *path, |
328d5752 MF |
4861 | struct buffer_head **last_eb_bh, |
4862 | int split_index, | |
4863 | struct ocfs2_extent_rec *orig_split_rec, | |
4864 | struct ocfs2_alloc_context *meta_ac) | |
4865 | { | |
4866 | int ret = 0, depth; | |
4867 | unsigned int insert_range, rec_range, do_leftright = 0; | |
4868 | struct ocfs2_extent_rec tmprec; | |
4869 | struct ocfs2_extent_list *rightmost_el; | |
4870 | struct ocfs2_extent_rec rec; | |
4871 | struct ocfs2_extent_rec split_rec = *orig_split_rec; | |
4872 | struct ocfs2_insert_type insert; | |
4873 | struct ocfs2_extent_block *eb; | |
328d5752 MF |
4874 | |
4875 | leftright: | |
4876 | /* | |
4877 | * Store a copy of the record on the stack - it might move | |
4878 | * around as the tree is manipulated below. | |
4879 | */ | |
4880 | rec = path_leaf_el(path)->l_recs[split_index]; | |
4881 | ||
ce1d9ea6 | 4882 | rightmost_el = et->et_root_el; |
328d5752 MF |
4883 | |
4884 | depth = le16_to_cpu(rightmost_el->l_tree_depth); | |
4885 | if (depth) { | |
4886 | BUG_ON(!(*last_eb_bh)); | |
4887 | eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data; | |
4888 | rightmost_el = &eb->h_list; | |
4889 | } | |
4890 | ||
4891 | if (le16_to_cpu(rightmost_el->l_next_free_rec) == | |
4892 | le16_to_cpu(rightmost_el->l_count)) { | |
d401dc12 | 4893 | ret = ocfs2_grow_tree(handle, et, |
e7d4cb6b | 4894 | &depth, last_eb_bh, meta_ac); |
328d5752 MF |
4895 | if (ret) { |
4896 | mlog_errno(ret); | |
4897 | goto out; | |
4898 | } | |
328d5752 MF |
4899 | } |
4900 | ||
4901 | memset(&insert, 0, sizeof(struct ocfs2_insert_type)); | |
4902 | insert.ins_appending = APPEND_NONE; | |
4903 | insert.ins_contig = CONTIG_NONE; | |
328d5752 MF |
4904 | insert.ins_tree_depth = depth; |
4905 | ||
4906 | insert_range = le32_to_cpu(split_rec.e_cpos) + | |
4907 | le16_to_cpu(split_rec.e_leaf_clusters); | |
4908 | rec_range = le32_to_cpu(rec.e_cpos) + | |
4909 | le16_to_cpu(rec.e_leaf_clusters); | |
4910 | ||
4911 | if (split_rec.e_cpos == rec.e_cpos) { | |
4912 | insert.ins_split = SPLIT_LEFT; | |
4913 | } else if (insert_range == rec_range) { | |
4914 | insert.ins_split = SPLIT_RIGHT; | |
4915 | } else { | |
4916 | /* | |
4917 | * Left/right split. We fake this as a right split | |
4918 | * first and then make a second pass as a left split. | |
4919 | */ | |
4920 | insert.ins_split = SPLIT_RIGHT; | |
4921 | ||
d231129f JB |
4922 | ocfs2_make_right_split_rec(ocfs2_metadata_cache_get_super(et->et_ci), |
4923 | &tmprec, insert_range, &rec); | |
328d5752 MF |
4924 | |
4925 | split_rec = tmprec; | |
4926 | ||
4927 | BUG_ON(do_leftright); | |
4928 | do_leftright = 1; | |
4929 | } | |
4930 | ||
3505bec0 | 4931 | ret = ocfs2_do_insert_extent(handle, et, &split_rec, &insert); |
328d5752 MF |
4932 | if (ret) { |
4933 | mlog_errno(ret); | |
4934 | goto out; | |
4935 | } | |
4936 | ||
4937 | if (do_leftright == 1) { | |
4938 | u32 cpos; | |
4939 | struct ocfs2_extent_list *el; | |
4940 | ||
4941 | do_leftright++; | |
4942 | split_rec = *orig_split_rec; | |
4943 | ||
4944 | ocfs2_reinit_path(path, 1); | |
4945 | ||
4946 | cpos = le32_to_cpu(split_rec.e_cpos); | |
facdb77f | 4947 | ret = ocfs2_find_path(et->et_ci, path, cpos); |
328d5752 MF |
4948 | if (ret) { |
4949 | mlog_errno(ret); | |
4950 | goto out; | |
4951 | } | |
4952 | ||
4953 | el = path_leaf_el(path); | |
4954 | split_index = ocfs2_search_extent_list(el, cpos); | |
4955 | goto leftright; | |
4956 | } | |
4957 | out: | |
4958 | ||
4959 | return ret; | |
4960 | } | |
4961 | ||
f3868d0f JB |
4962 | static int ocfs2_replace_extent_rec(handle_t *handle, |
4963 | struct ocfs2_extent_tree *et, | |
47be12e4 TM |
4964 | struct ocfs2_path *path, |
4965 | struct ocfs2_extent_list *el, | |
4966 | int split_index, | |
4967 | struct ocfs2_extent_rec *split_rec) | |
4968 | { | |
4969 | int ret; | |
4970 | ||
f3868d0f | 4971 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, path, |
47be12e4 TM |
4972 | path_num_items(path) - 1); |
4973 | if (ret) { | |
4974 | mlog_errno(ret); | |
4975 | goto out; | |
4976 | } | |
4977 | ||
4978 | el->l_recs[split_index] = *split_rec; | |
4979 | ||
4980 | ocfs2_journal_dirty(handle, path_leaf_bh(path)); | |
4981 | out: | |
4982 | return ret; | |
4983 | } | |
4984 | ||
328d5752 | 4985 | /* |
555936bf TM |
4986 | * Split part or all of the extent record at split_index in the leaf |
4987 | * pointed to by path. Merge with the contiguous extent record if needed. | |
328d5752 MF |
4988 | * |
4989 | * Care is taken to handle contiguousness so as to not grow the tree. | |
4990 | * | |
4991 | * meta_ac is not strictly necessary - we only truly need it if growth | |
4992 | * of the tree is required. All other cases will degrade into a less | |
4993 | * optimal tree layout. | |
4994 | * | |
e7d4cb6b TM |
4995 | * last_eb_bh should be the rightmost leaf block for any extent |
4996 | * btree. Since a split may grow the tree or a merge might shrink it, | |
4997 | * the caller cannot trust the contents of that buffer after this call. | |
328d5752 MF |
4998 | * |
4999 | * This code is optimized for readability - several passes might be | |
5000 | * made over certain portions of the tree. All of those blocks will | |
5001 | * have been brought into cache (and pinned via the journal), so the | |
5002 | * extra overhead is not expressed in terms of disk reads. | |
5003 | */ | |
e2e9f608 TM |
5004 | int ocfs2_split_extent(handle_t *handle, |
5005 | struct ocfs2_extent_tree *et, | |
5006 | struct ocfs2_path *path, | |
5007 | int split_index, | |
5008 | struct ocfs2_extent_rec *split_rec, | |
5009 | struct ocfs2_alloc_context *meta_ac, | |
5010 | struct ocfs2_cached_dealloc_ctxt *dealloc) | |
328d5752 MF |
5011 | { |
5012 | int ret = 0; | |
5013 | struct ocfs2_extent_list *el = path_leaf_el(path); | |
e8aed345 | 5014 | struct buffer_head *last_eb_bh = NULL; |
328d5752 MF |
5015 | struct ocfs2_extent_rec *rec = &el->l_recs[split_index]; |
5016 | struct ocfs2_merge_ctxt ctxt; | |
5017 | struct ocfs2_extent_list *rightmost_el; | |
5018 | ||
328d5752 MF |
5019 | if (le32_to_cpu(rec->e_cpos) > le32_to_cpu(split_rec->e_cpos) || |
5020 | ((le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters)) < | |
5021 | (le32_to_cpu(split_rec->e_cpos) + le16_to_cpu(split_rec->e_leaf_clusters)))) { | |
5022 | ret = -EIO; | |
5023 | mlog_errno(ret); | |
5024 | goto out; | |
5025 | } | |
5026 | ||
a2970291 | 5027 | ctxt.c_contig_type = ocfs2_figure_merge_contig_type(et, path, el, |
328d5752 MF |
5028 | split_index, |
5029 | split_rec); | |
5030 | ||
5031 | /* | |
5032 | * The core merge / split code wants to know how much room is | |
a1cf076b | 5033 | * left in this allocation tree, so we pass the |
328d5752 MF |
5034 | * rightmost extent list. |
5035 | */ | |
5036 | if (path->p_tree_depth) { | |
5037 | struct ocfs2_extent_block *eb; | |
328d5752 | 5038 | |
3d03a305 | 5039 | ret = ocfs2_read_extent_block(et->et_ci, |
5e96581a JB |
5040 | ocfs2_et_get_last_eb_blk(et), |
5041 | &last_eb_bh); | |
328d5752 MF |
5042 | if (ret) { |
5043 | mlog_exit(ret); | |
5044 | goto out; | |
5045 | } | |
5046 | ||
5047 | eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; | |
328d5752 MF |
5048 | rightmost_el = &eb->h_list; |
5049 | } else | |
5050 | rightmost_el = path_root_el(path); | |
5051 | ||
328d5752 MF |
5052 | if (rec->e_cpos == split_rec->e_cpos && |
5053 | rec->e_leaf_clusters == split_rec->e_leaf_clusters) | |
5054 | ctxt.c_split_covers_rec = 1; | |
5055 | else | |
5056 | ctxt.c_split_covers_rec = 0; | |
5057 | ||
5058 | ctxt.c_has_empty_extent = ocfs2_is_empty_extent(&el->l_recs[0]); | |
5059 | ||
015452b1 MF |
5060 | mlog(0, "index: %d, contig: %u, has_empty: %u, split_covers: %u\n", |
5061 | split_index, ctxt.c_contig_type, ctxt.c_has_empty_extent, | |
5062 | ctxt.c_split_covers_rec); | |
328d5752 MF |
5063 | |
5064 | if (ctxt.c_contig_type == CONTIG_NONE) { | |
5065 | if (ctxt.c_split_covers_rec) | |
f3868d0f | 5066 | ret = ocfs2_replace_extent_rec(handle, et, path, el, |
47be12e4 | 5067 | split_index, split_rec); |
328d5752 | 5068 | else |
d231129f | 5069 | ret = ocfs2_split_and_insert(handle, et, path, |
328d5752 MF |
5070 | &last_eb_bh, split_index, |
5071 | split_rec, meta_ac); | |
5072 | if (ret) | |
5073 | mlog_errno(ret); | |
5074 | } else { | |
c495dd24 | 5075 | ret = ocfs2_try_to_merge_extent(handle, et, path, |
328d5752 | 5076 | split_index, split_rec, |
c495dd24 | 5077 | dealloc, &ctxt); |
328d5752 MF |
5078 | if (ret) |
5079 | mlog_errno(ret); | |
5080 | } | |
5081 | ||
328d5752 MF |
5082 | out: |
5083 | brelse(last_eb_bh); | |
5084 | return ret; | |
5085 | } | |
5086 | ||
5087 | /* | |
555936bf TM |
5088 | * Change the flags of the already-existing extent at cpos for len clusters. |
5089 | * | |
5090 | * new_flags: the flags we want to set. | |
5091 | * clear_flags: the flags we want to clear. | |
5092 | * phys: the new physical offset we want this new extent starts from. | |
328d5752 MF |
5093 | * |
5094 | * If the existing extent is larger than the request, initiate a | |
5095 | * split. An attempt will be made at merging with adjacent extents. | |
5096 | * | |
5097 | * The caller is responsible for passing down meta_ac if we'll need it. | |
5098 | */ | |
1aa75fea TM |
5099 | int ocfs2_change_extent_flag(handle_t *handle, |
5100 | struct ocfs2_extent_tree *et, | |
5101 | u32 cpos, u32 len, u32 phys, | |
5102 | struct ocfs2_alloc_context *meta_ac, | |
5103 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
5104 | int new_flags, int clear_flags) | |
328d5752 MF |
5105 | { |
5106 | int ret, index; | |
555936bf TM |
5107 | struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci); |
5108 | u64 start_blkno = ocfs2_clusters_to_blocks(sb, phys); | |
328d5752 MF |
5109 | struct ocfs2_extent_rec split_rec; |
5110 | struct ocfs2_path *left_path = NULL; | |
5111 | struct ocfs2_extent_list *el; | |
555936bf | 5112 | struct ocfs2_extent_rec *rec; |
328d5752 | 5113 | |
ffdd7a54 | 5114 | left_path = ocfs2_new_path_from_et(et); |
328d5752 MF |
5115 | if (!left_path) { |
5116 | ret = -ENOMEM; | |
5117 | mlog_errno(ret); | |
5118 | goto out; | |
5119 | } | |
5120 | ||
facdb77f | 5121 | ret = ocfs2_find_path(et->et_ci, left_path, cpos); |
328d5752 MF |
5122 | if (ret) { |
5123 | mlog_errno(ret); | |
5124 | goto out; | |
5125 | } | |
5126 | el = path_leaf_el(left_path); | |
5127 | ||
5128 | index = ocfs2_search_extent_list(el, cpos); | |
5129 | if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) { | |
555936bf TM |
5130 | ocfs2_error(sb, |
5131 | "Owner %llu has an extent at cpos %u which can no " | |
328d5752 | 5132 | "longer be found.\n", |
555936bf TM |
5133 | (unsigned long long) |
5134 | ocfs2_metadata_cache_owner(et->et_ci), cpos); | |
328d5752 MF |
5135 | ret = -EROFS; |
5136 | goto out; | |
5137 | } | |
5138 | ||
555936bf TM |
5139 | ret = -EIO; |
5140 | rec = &el->l_recs[index]; | |
5141 | if (new_flags && (rec->e_flags & new_flags)) { | |
5142 | mlog(ML_ERROR, "Owner %llu tried to set %d flags on an " | |
5143 | "extent that already had them", | |
5144 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), | |
5145 | new_flags); | |
5146 | goto out; | |
5147 | } | |
5148 | ||
5149 | if (clear_flags && !(rec->e_flags & clear_flags)) { | |
5150 | mlog(ML_ERROR, "Owner %llu tried to clear %d flags on an " | |
5151 | "extent that didn't have them", | |
5152 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), | |
5153 | clear_flags); | |
5154 | goto out; | |
5155 | } | |
5156 | ||
328d5752 MF |
5157 | memset(&split_rec, 0, sizeof(struct ocfs2_extent_rec)); |
5158 | split_rec.e_cpos = cpu_to_le32(cpos); | |
5159 | split_rec.e_leaf_clusters = cpu_to_le16(len); | |
5160 | split_rec.e_blkno = cpu_to_le64(start_blkno); | |
555936bf TM |
5161 | split_rec.e_flags = rec->e_flags; |
5162 | if (new_flags) | |
5163 | split_rec.e_flags |= new_flags; | |
5164 | if (clear_flags) | |
5165 | split_rec.e_flags &= ~clear_flags; | |
5166 | ||
e2e9f608 TM |
5167 | ret = ocfs2_split_extent(handle, et, left_path, |
5168 | index, &split_rec, meta_ac, | |
5169 | dealloc); | |
328d5752 MF |
5170 | if (ret) |
5171 | mlog_errno(ret); | |
5172 | ||
5173 | out: | |
5174 | ocfs2_free_path(left_path); | |
5175 | return ret; | |
555936bf TM |
5176 | |
5177 | } | |
5178 | ||
5179 | /* | |
5180 | * Mark the already-existing extent at cpos as written for len clusters. | |
5181 | * This removes the unwritten extent flag. | |
5182 | * | |
5183 | * If the existing extent is larger than the request, initiate a | |
5184 | * split. An attempt will be made at merging with adjacent extents. | |
5185 | * | |
5186 | * The caller is responsible for passing down meta_ac if we'll need it. | |
5187 | */ | |
5188 | int ocfs2_mark_extent_written(struct inode *inode, | |
5189 | struct ocfs2_extent_tree *et, | |
5190 | handle_t *handle, u32 cpos, u32 len, u32 phys, | |
5191 | struct ocfs2_alloc_context *meta_ac, | |
5192 | struct ocfs2_cached_dealloc_ctxt *dealloc) | |
5193 | { | |
5194 | int ret; | |
5195 | ||
5196 | mlog(0, "Inode %lu cpos %u, len %u, phys clusters %u\n", | |
5197 | inode->i_ino, cpos, len, phys); | |
5198 | ||
5199 | if (!ocfs2_writes_unwritten_extents(OCFS2_SB(inode->i_sb))) { | |
5200 | ocfs2_error(inode->i_sb, "Inode %llu has unwritten extents " | |
5201 | "that are being written to, but the feature bit " | |
5202 | "is not set in the super block.", | |
5203 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
5204 | ret = -EROFS; | |
5205 | goto out; | |
5206 | } | |
5207 | ||
5208 | /* | |
5209 | * XXX: This should be fixed up so that we just re-insert the | |
5210 | * next extent records. | |
5211 | */ | |
5212 | ocfs2_et_extent_map_truncate(et, 0); | |
5213 | ||
5214 | ret = ocfs2_change_extent_flag(handle, et, cpos, | |
5215 | len, phys, meta_ac, dealloc, | |
5216 | 0, OCFS2_EXT_UNWRITTEN); | |
5217 | if (ret) | |
5218 | mlog_errno(ret); | |
5219 | ||
5220 | out: | |
5221 | return ret; | |
328d5752 MF |
5222 | } |
5223 | ||
dbdcf6a4 JB |
5224 | static int ocfs2_split_tree(handle_t *handle, struct ocfs2_extent_tree *et, |
5225 | struct ocfs2_path *path, | |
d0c7d708 MF |
5226 | int index, u32 new_range, |
5227 | struct ocfs2_alloc_context *meta_ac) | |
5228 | { | |
c901fb00 | 5229 | int ret, depth, credits; |
d0c7d708 MF |
5230 | struct buffer_head *last_eb_bh = NULL; |
5231 | struct ocfs2_extent_block *eb; | |
5232 | struct ocfs2_extent_list *rightmost_el, *el; | |
5233 | struct ocfs2_extent_rec split_rec; | |
5234 | struct ocfs2_extent_rec *rec; | |
5235 | struct ocfs2_insert_type insert; | |
5236 | ||
5237 | /* | |
5238 | * Setup the record to split before we grow the tree. | |
5239 | */ | |
5240 | el = path_leaf_el(path); | |
5241 | rec = &el->l_recs[index]; | |
dbdcf6a4 JB |
5242 | ocfs2_make_right_split_rec(ocfs2_metadata_cache_get_super(et->et_ci), |
5243 | &split_rec, new_range, rec); | |
d0c7d708 MF |
5244 | |
5245 | depth = path->p_tree_depth; | |
5246 | if (depth > 0) { | |
3d03a305 | 5247 | ret = ocfs2_read_extent_block(et->et_ci, |
5e96581a JB |
5248 | ocfs2_et_get_last_eb_blk(et), |
5249 | &last_eb_bh); | |
d0c7d708 MF |
5250 | if (ret < 0) { |
5251 | mlog_errno(ret); | |
5252 | goto out; | |
5253 | } | |
5254 | ||
5255 | eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; | |
5256 | rightmost_el = &eb->h_list; | |
5257 | } else | |
5258 | rightmost_el = path_leaf_el(path); | |
5259 | ||
c901fb00 TM |
5260 | credits = path->p_tree_depth + |
5261 | ocfs2_extend_meta_needed(et->et_root_el); | |
d0c7d708 MF |
5262 | ret = ocfs2_extend_trans(handle, credits); |
5263 | if (ret) { | |
5264 | mlog_errno(ret); | |
5265 | goto out; | |
5266 | } | |
5267 | ||
5268 | if (le16_to_cpu(rightmost_el->l_next_free_rec) == | |
5269 | le16_to_cpu(rightmost_el->l_count)) { | |
d401dc12 | 5270 | ret = ocfs2_grow_tree(handle, et, &depth, &last_eb_bh, |
d0c7d708 MF |
5271 | meta_ac); |
5272 | if (ret) { | |
5273 | mlog_errno(ret); | |
5274 | goto out; | |
5275 | } | |
d0c7d708 MF |
5276 | } |
5277 | ||
5278 | memset(&insert, 0, sizeof(struct ocfs2_insert_type)); | |
5279 | insert.ins_appending = APPEND_NONE; | |
5280 | insert.ins_contig = CONTIG_NONE; | |
5281 | insert.ins_split = SPLIT_RIGHT; | |
d0c7d708 MF |
5282 | insert.ins_tree_depth = depth; |
5283 | ||
3505bec0 | 5284 | ret = ocfs2_do_insert_extent(handle, et, &split_rec, &insert); |
d0c7d708 MF |
5285 | if (ret) |
5286 | mlog_errno(ret); | |
5287 | ||
5288 | out: | |
5289 | brelse(last_eb_bh); | |
5290 | return ret; | |
5291 | } | |
5292 | ||
043beebb JB |
5293 | static int ocfs2_truncate_rec(handle_t *handle, |
5294 | struct ocfs2_extent_tree *et, | |
d0c7d708 MF |
5295 | struct ocfs2_path *path, int index, |
5296 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
043beebb | 5297 | u32 cpos, u32 len) |
d0c7d708 MF |
5298 | { |
5299 | int ret; | |
5300 | u32 left_cpos, rec_range, trunc_range; | |
5301 | int wants_rotate = 0, is_rightmost_tree_rec = 0; | |
043beebb | 5302 | struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci); |
d0c7d708 MF |
5303 | struct ocfs2_path *left_path = NULL; |
5304 | struct ocfs2_extent_list *el = path_leaf_el(path); | |
5305 | struct ocfs2_extent_rec *rec; | |
5306 | struct ocfs2_extent_block *eb; | |
5307 | ||
5308 | if (ocfs2_is_empty_extent(&el->l_recs[0]) && index > 0) { | |
70f18c08 | 5309 | ret = ocfs2_rotate_tree_left(handle, et, path, dealloc); |
d0c7d708 MF |
5310 | if (ret) { |
5311 | mlog_errno(ret); | |
5312 | goto out; | |
5313 | } | |
5314 | ||
5315 | index--; | |
5316 | } | |
5317 | ||
5318 | if (index == (le16_to_cpu(el->l_next_free_rec) - 1) && | |
5319 | path->p_tree_depth) { | |
5320 | /* | |
5321 | * Check whether this is the rightmost tree record. If | |
5322 | * we remove all of this record or part of its right | |
5323 | * edge then an update of the record lengths above it | |
5324 | * will be required. | |
5325 | */ | |
5326 | eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data; | |
5327 | if (eb->h_next_leaf_blk == 0) | |
5328 | is_rightmost_tree_rec = 1; | |
5329 | } | |
5330 | ||
5331 | rec = &el->l_recs[index]; | |
5332 | if (index == 0 && path->p_tree_depth && | |
5333 | le32_to_cpu(rec->e_cpos) == cpos) { | |
5334 | /* | |
5335 | * Changing the leftmost offset (via partial or whole | |
5336 | * record truncate) of an interior (or rightmost) path | |
5337 | * means we have to update the subtree that is formed | |
5338 | * by this leaf and the one to it's left. | |
5339 | * | |
5340 | * There are two cases we can skip: | |
043beebb | 5341 | * 1) Path is the leftmost one in our btree. |
d0c7d708 MF |
5342 | * 2) The leaf is rightmost and will be empty after |
5343 | * we remove the extent record - the rotate code | |
5344 | * knows how to update the newly formed edge. | |
5345 | */ | |
5346 | ||
043beebb | 5347 | ret = ocfs2_find_cpos_for_left_leaf(sb, path, &left_cpos); |
d0c7d708 MF |
5348 | if (ret) { |
5349 | mlog_errno(ret); | |
5350 | goto out; | |
5351 | } | |
5352 | ||
5353 | if (left_cpos && le16_to_cpu(el->l_next_free_rec) > 1) { | |
ffdd7a54 | 5354 | left_path = ocfs2_new_path_from_path(path); |
d0c7d708 MF |
5355 | if (!left_path) { |
5356 | ret = -ENOMEM; | |
5357 | mlog_errno(ret); | |
5358 | goto out; | |
5359 | } | |
5360 | ||
facdb77f JB |
5361 | ret = ocfs2_find_path(et->et_ci, left_path, |
5362 | left_cpos); | |
d0c7d708 MF |
5363 | if (ret) { |
5364 | mlog_errno(ret); | |
5365 | goto out; | |
5366 | } | |
5367 | } | |
5368 | } | |
5369 | ||
5370 | ret = ocfs2_extend_rotate_transaction(handle, 0, | |
5371 | handle->h_buffer_credits, | |
5372 | path); | |
5373 | if (ret) { | |
5374 | mlog_errno(ret); | |
5375 | goto out; | |
5376 | } | |
5377 | ||
d9a0a1f8 | 5378 | ret = ocfs2_journal_access_path(et->et_ci, handle, path); |
d0c7d708 MF |
5379 | if (ret) { |
5380 | mlog_errno(ret); | |
5381 | goto out; | |
5382 | } | |
5383 | ||
d9a0a1f8 | 5384 | ret = ocfs2_journal_access_path(et->et_ci, handle, left_path); |
d0c7d708 MF |
5385 | if (ret) { |
5386 | mlog_errno(ret); | |
5387 | goto out; | |
5388 | } | |
5389 | ||
5390 | rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
5391 | trunc_range = cpos + len; | |
5392 | ||
5393 | if (le32_to_cpu(rec->e_cpos) == cpos && rec_range == trunc_range) { | |
5394 | int next_free; | |
5395 | ||
5396 | memset(rec, 0, sizeof(*rec)); | |
5397 | ocfs2_cleanup_merge(el, index); | |
5398 | wants_rotate = 1; | |
5399 | ||
5400 | next_free = le16_to_cpu(el->l_next_free_rec); | |
5401 | if (is_rightmost_tree_rec && next_free > 1) { | |
5402 | /* | |
5403 | * We skip the edge update if this path will | |
5404 | * be deleted by the rotate code. | |
5405 | */ | |
5406 | rec = &el->l_recs[next_free - 1]; | |
d401dc12 | 5407 | ocfs2_adjust_rightmost_records(handle, et, path, |
d0c7d708 MF |
5408 | rec); |
5409 | } | |
5410 | } else if (le32_to_cpu(rec->e_cpos) == cpos) { | |
5411 | /* Remove leftmost portion of the record. */ | |
5412 | le32_add_cpu(&rec->e_cpos, len); | |
5413 | le64_add_cpu(&rec->e_blkno, ocfs2_clusters_to_blocks(sb, len)); | |
5414 | le16_add_cpu(&rec->e_leaf_clusters, -len); | |
5415 | } else if (rec_range == trunc_range) { | |
5416 | /* Remove rightmost portion of the record */ | |
5417 | le16_add_cpu(&rec->e_leaf_clusters, -len); | |
5418 | if (is_rightmost_tree_rec) | |
d401dc12 | 5419 | ocfs2_adjust_rightmost_records(handle, et, path, rec); |
d0c7d708 MF |
5420 | } else { |
5421 | /* Caller should have trapped this. */ | |
043beebb JB |
5422 | mlog(ML_ERROR, "Owner %llu: Invalid record truncate: (%u, %u) " |
5423 | "(%u, %u)\n", | |
5424 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), | |
d0c7d708 MF |
5425 | le32_to_cpu(rec->e_cpos), |
5426 | le16_to_cpu(rec->e_leaf_clusters), cpos, len); | |
5427 | BUG(); | |
5428 | } | |
5429 | ||
5430 | if (left_path) { | |
5431 | int subtree_index; | |
5432 | ||
7dc02805 | 5433 | subtree_index = ocfs2_find_subtree_root(et, left_path, path); |
4619c73e | 5434 | ocfs2_complete_edge_insert(handle, left_path, path, |
d0c7d708 MF |
5435 | subtree_index); |
5436 | } | |
5437 | ||
5438 | ocfs2_journal_dirty(handle, path_leaf_bh(path)); | |
5439 | ||
70f18c08 | 5440 | ret = ocfs2_rotate_tree_left(handle, et, path, dealloc); |
d0c7d708 MF |
5441 | if (ret) { |
5442 | mlog_errno(ret); | |
5443 | goto out; | |
5444 | } | |
5445 | ||
5446 | out: | |
5447 | ocfs2_free_path(left_path); | |
5448 | return ret; | |
5449 | } | |
5450 | ||
dbdcf6a4 | 5451 | int ocfs2_remove_extent(handle_t *handle, |
f99b9b7c | 5452 | struct ocfs2_extent_tree *et, |
dbdcf6a4 | 5453 | u32 cpos, u32 len, |
063c4561 | 5454 | struct ocfs2_alloc_context *meta_ac, |
f99b9b7c | 5455 | struct ocfs2_cached_dealloc_ctxt *dealloc) |
d0c7d708 MF |
5456 | { |
5457 | int ret, index; | |
5458 | u32 rec_range, trunc_range; | |
5459 | struct ocfs2_extent_rec *rec; | |
5460 | struct ocfs2_extent_list *el; | |
e7d4cb6b | 5461 | struct ocfs2_path *path = NULL; |
d0c7d708 | 5462 | |
4c911eef JB |
5463 | /* |
5464 | * XXX: Why are we truncating to 0 instead of wherever this | |
5465 | * affects us? | |
5466 | */ | |
5467 | ocfs2_et_extent_map_truncate(et, 0); | |
d0c7d708 | 5468 | |
ffdd7a54 | 5469 | path = ocfs2_new_path_from_et(et); |
d0c7d708 MF |
5470 | if (!path) { |
5471 | ret = -ENOMEM; | |
5472 | mlog_errno(ret); | |
5473 | goto out; | |
5474 | } | |
5475 | ||
facdb77f | 5476 | ret = ocfs2_find_path(et->et_ci, path, cpos); |
d0c7d708 MF |
5477 | if (ret) { |
5478 | mlog_errno(ret); | |
5479 | goto out; | |
5480 | } | |
5481 | ||
5482 | el = path_leaf_el(path); | |
5483 | index = ocfs2_search_extent_list(el, cpos); | |
5484 | if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) { | |
dbdcf6a4 JB |
5485 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
5486 | "Owner %llu has an extent at cpos %u which can no " | |
d0c7d708 | 5487 | "longer be found.\n", |
dbdcf6a4 JB |
5488 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
5489 | cpos); | |
d0c7d708 MF |
5490 | ret = -EROFS; |
5491 | goto out; | |
5492 | } | |
5493 | ||
5494 | /* | |
5495 | * We have 3 cases of extent removal: | |
5496 | * 1) Range covers the entire extent rec | |
5497 | * 2) Range begins or ends on one edge of the extent rec | |
5498 | * 3) Range is in the middle of the extent rec (no shared edges) | |
5499 | * | |
5500 | * For case 1 we remove the extent rec and left rotate to | |
5501 | * fill the hole. | |
5502 | * | |
5503 | * For case 2 we just shrink the existing extent rec, with a | |
5504 | * tree update if the shrinking edge is also the edge of an | |
5505 | * extent block. | |
5506 | * | |
5507 | * For case 3 we do a right split to turn the extent rec into | |
5508 | * something case 2 can handle. | |
5509 | */ | |
5510 | rec = &el->l_recs[index]; | |
5511 | rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
5512 | trunc_range = cpos + len; | |
5513 | ||
5514 | BUG_ON(cpos < le32_to_cpu(rec->e_cpos) || trunc_range > rec_range); | |
5515 | ||
dbdcf6a4 | 5516 | mlog(0, "Owner %llu, remove (cpos %u, len %u). Existing index %d " |
d0c7d708 | 5517 | "(cpos %u, len %u)\n", |
dbdcf6a4 JB |
5518 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
5519 | cpos, len, index, | |
d0c7d708 MF |
5520 | le32_to_cpu(rec->e_cpos), ocfs2_rec_clusters(el, rec)); |
5521 | ||
5522 | if (le32_to_cpu(rec->e_cpos) == cpos || rec_range == trunc_range) { | |
043beebb JB |
5523 | ret = ocfs2_truncate_rec(handle, et, path, index, dealloc, |
5524 | cpos, len); | |
d0c7d708 MF |
5525 | if (ret) { |
5526 | mlog_errno(ret); | |
5527 | goto out; | |
5528 | } | |
5529 | } else { | |
dbdcf6a4 | 5530 | ret = ocfs2_split_tree(handle, et, path, index, |
d0c7d708 MF |
5531 | trunc_range, meta_ac); |
5532 | if (ret) { | |
5533 | mlog_errno(ret); | |
5534 | goto out; | |
5535 | } | |
5536 | ||
5537 | /* | |
5538 | * The split could have manipulated the tree enough to | |
5539 | * move the record location, so we have to look for it again. | |
5540 | */ | |
5541 | ocfs2_reinit_path(path, 1); | |
5542 | ||
facdb77f | 5543 | ret = ocfs2_find_path(et->et_ci, path, cpos); |
d0c7d708 MF |
5544 | if (ret) { |
5545 | mlog_errno(ret); | |
5546 | goto out; | |
5547 | } | |
5548 | ||
5549 | el = path_leaf_el(path); | |
5550 | index = ocfs2_search_extent_list(el, cpos); | |
5551 | if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) { | |
dbdcf6a4 JB |
5552 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
5553 | "Owner %llu: split at cpos %u lost record.", | |
5554 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), | |
d0c7d708 MF |
5555 | cpos); |
5556 | ret = -EROFS; | |
5557 | goto out; | |
5558 | } | |
5559 | ||
5560 | /* | |
5561 | * Double check our values here. If anything is fishy, | |
5562 | * it's easier to catch it at the top level. | |
5563 | */ | |
5564 | rec = &el->l_recs[index]; | |
5565 | rec_range = le32_to_cpu(rec->e_cpos) + | |
5566 | ocfs2_rec_clusters(el, rec); | |
5567 | if (rec_range != trunc_range) { | |
dbdcf6a4 JB |
5568 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
5569 | "Owner %llu: error after split at cpos %u" | |
d0c7d708 | 5570 | "trunc len %u, existing record is (%u,%u)", |
dbdcf6a4 | 5571 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
d0c7d708 MF |
5572 | cpos, len, le32_to_cpu(rec->e_cpos), |
5573 | ocfs2_rec_clusters(el, rec)); | |
5574 | ret = -EROFS; | |
5575 | goto out; | |
5576 | } | |
5577 | ||
043beebb JB |
5578 | ret = ocfs2_truncate_rec(handle, et, path, index, dealloc, |
5579 | cpos, len); | |
d0c7d708 MF |
5580 | if (ret) { |
5581 | mlog_errno(ret); | |
5582 | goto out; | |
5583 | } | |
5584 | } | |
5585 | ||
5586 | out: | |
5587 | ocfs2_free_path(path); | |
5588 | return ret; | |
5589 | } | |
5590 | ||
78f94673 TY |
5591 | /* |
5592 | * ocfs2_reserve_blocks_for_rec_trunc() would look basically the | |
5593 | * same as ocfs2_lock_alloctors(), except for it accepts a blocks | |
5594 | * number to reserve some extra blocks, and it only handles meta | |
5595 | * data allocations. | |
5596 | * | |
5597 | * Currently, only ocfs2_remove_btree_range() uses it for truncating | |
5598 | * and punching holes. | |
5599 | */ | |
5600 | static int ocfs2_reserve_blocks_for_rec_trunc(struct inode *inode, | |
5601 | struct ocfs2_extent_tree *et, | |
5602 | u32 extents_to_split, | |
5603 | struct ocfs2_alloc_context **ac, | |
5604 | int extra_blocks) | |
5605 | { | |
5606 | int ret = 0, num_free_extents; | |
5607 | unsigned int max_recs_needed = 2 * extents_to_split; | |
5608 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
5609 | ||
5610 | *ac = NULL; | |
5611 | ||
5612 | num_free_extents = ocfs2_num_free_extents(osb, et); | |
5613 | if (num_free_extents < 0) { | |
5614 | ret = num_free_extents; | |
5615 | mlog_errno(ret); | |
5616 | goto out; | |
5617 | } | |
5618 | ||
5619 | if (!num_free_extents || | |
5620 | (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) | |
5621 | extra_blocks += ocfs2_extend_meta_needed(et->et_root_el); | |
5622 | ||
5623 | if (extra_blocks) { | |
5624 | ret = ocfs2_reserve_new_metadata_blocks(osb, extra_blocks, ac); | |
5625 | if (ret < 0) { | |
5626 | if (ret != -ENOSPC) | |
5627 | mlog_errno(ret); | |
5628 | goto out; | |
5629 | } | |
5630 | } | |
5631 | ||
5632 | out: | |
5633 | if (ret) { | |
5634 | if (*ac) { | |
5635 | ocfs2_free_alloc_context(*ac); | |
5636 | *ac = NULL; | |
5637 | } | |
5638 | } | |
5639 | ||
5640 | return ret; | |
5641 | } | |
5642 | ||
fecc0112 MF |
5643 | int ocfs2_remove_btree_range(struct inode *inode, |
5644 | struct ocfs2_extent_tree *et, | |
78f94673 TY |
5645 | u32 cpos, u32 phys_cpos, u32 len, int flags, |
5646 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
5647 | u64 refcount_loc) | |
fecc0112 | 5648 | { |
78f94673 | 5649 | int ret, credits = 0, extra_blocks = 0; |
fecc0112 MF |
5650 | u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos); |
5651 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
5652 | struct inode *tl_inode = osb->osb_tl_inode; | |
5653 | handle_t *handle; | |
5654 | struct ocfs2_alloc_context *meta_ac = NULL; | |
78f94673 TY |
5655 | struct ocfs2_refcount_tree *ref_tree = NULL; |
5656 | ||
5657 | if ((flags & OCFS2_EXT_REFCOUNTED) && len) { | |
5658 | BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & | |
5659 | OCFS2_HAS_REFCOUNT_FL)); | |
5660 | ||
5661 | ret = ocfs2_lock_refcount_tree(osb, refcount_loc, 1, | |
5662 | &ref_tree, NULL); | |
5663 | if (ret) { | |
5664 | mlog_errno(ret); | |
5665 | goto out; | |
5666 | } | |
5667 | ||
5668 | ret = ocfs2_prepare_refcount_change_for_del(inode, | |
5669 | refcount_loc, | |
5670 | phys_blkno, | |
5671 | len, | |
5672 | &credits, | |
5673 | &extra_blocks); | |
5674 | if (ret < 0) { | |
5675 | mlog_errno(ret); | |
5676 | goto out; | |
5677 | } | |
5678 | } | |
fecc0112 | 5679 | |
78f94673 TY |
5680 | ret = ocfs2_reserve_blocks_for_rec_trunc(inode, et, 1, &meta_ac, |
5681 | extra_blocks); | |
fecc0112 MF |
5682 | if (ret) { |
5683 | mlog_errno(ret); | |
5684 | return ret; | |
5685 | } | |
5686 | ||
5687 | mutex_lock(&tl_inode->i_mutex); | |
5688 | ||
5689 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
5690 | ret = __ocfs2_flush_truncate_log(osb); | |
5691 | if (ret < 0) { | |
5692 | mlog_errno(ret); | |
5693 | goto out; | |
5694 | } | |
5695 | } | |
5696 | ||
78f94673 TY |
5697 | handle = ocfs2_start_trans(osb, |
5698 | ocfs2_remove_extent_credits(osb->sb) + credits); | |
fecc0112 MF |
5699 | if (IS_ERR(handle)) { |
5700 | ret = PTR_ERR(handle); | |
5701 | mlog_errno(ret); | |
5702 | goto out; | |
5703 | } | |
5704 | ||
d9a0a1f8 | 5705 | ret = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 5706 | OCFS2_JOURNAL_ACCESS_WRITE); |
fecc0112 MF |
5707 | if (ret) { |
5708 | mlog_errno(ret); | |
5709 | goto out; | |
5710 | } | |
5711 | ||
5dd4056d | 5712 | dquot_free_space_nodirty(inode, |
fd4ef231 MF |
5713 | ocfs2_clusters_to_bytes(inode->i_sb, len)); |
5714 | ||
dbdcf6a4 | 5715 | ret = ocfs2_remove_extent(handle, et, cpos, len, meta_ac, dealloc); |
fecc0112 MF |
5716 | if (ret) { |
5717 | mlog_errno(ret); | |
5718 | goto out_commit; | |
5719 | } | |
5720 | ||
6136ca5f | 5721 | ocfs2_et_update_clusters(et, -len); |
fecc0112 | 5722 | |
ec20cec7 | 5723 | ocfs2_journal_dirty(handle, et->et_root_bh); |
fecc0112 | 5724 | |
78f94673 TY |
5725 | if (phys_blkno) { |
5726 | if (flags & OCFS2_EXT_REFCOUNTED) | |
5727 | ret = ocfs2_decrease_refcount(inode, handle, | |
5728 | ocfs2_blocks_to_clusters(osb->sb, | |
5729 | phys_blkno), | |
5730 | len, meta_ac, | |
5731 | dealloc, 1); | |
5732 | else | |
5733 | ret = ocfs2_truncate_log_append(osb, handle, | |
5734 | phys_blkno, len); | |
5735 | if (ret) | |
5736 | mlog_errno(ret); | |
5737 | ||
5738 | } | |
fecc0112 MF |
5739 | |
5740 | out_commit: | |
5741 | ocfs2_commit_trans(osb, handle); | |
5742 | out: | |
5743 | mutex_unlock(&tl_inode->i_mutex); | |
5744 | ||
5745 | if (meta_ac) | |
5746 | ocfs2_free_alloc_context(meta_ac); | |
5747 | ||
78f94673 TY |
5748 | if (ref_tree) |
5749 | ocfs2_unlock_refcount_tree(osb, ref_tree, 1); | |
5750 | ||
fecc0112 MF |
5751 | return ret; |
5752 | } | |
5753 | ||
063c4561 | 5754 | int ocfs2_truncate_log_needs_flush(struct ocfs2_super *osb) |
ccd979bd MF |
5755 | { |
5756 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5757 | struct ocfs2_dinode *di; | |
5758 | struct ocfs2_truncate_log *tl; | |
5759 | ||
5760 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
5761 | tl = &di->id2.i_dealloc; | |
5762 | ||
5763 | mlog_bug_on_msg(le16_to_cpu(tl->tl_used) > le16_to_cpu(tl->tl_count), | |
5764 | "slot %d, invalid truncate log parameters: used = " | |
5765 | "%u, count = %u\n", osb->slot_num, | |
5766 | le16_to_cpu(tl->tl_used), le16_to_cpu(tl->tl_count)); | |
5767 | return le16_to_cpu(tl->tl_used) == le16_to_cpu(tl->tl_count); | |
5768 | } | |
5769 | ||
5770 | static int ocfs2_truncate_log_can_coalesce(struct ocfs2_truncate_log *tl, | |
5771 | unsigned int new_start) | |
5772 | { | |
5773 | unsigned int tail_index; | |
5774 | unsigned int current_tail; | |
5775 | ||
5776 | /* No records, nothing to coalesce */ | |
5777 | if (!le16_to_cpu(tl->tl_used)) | |
5778 | return 0; | |
5779 | ||
5780 | tail_index = le16_to_cpu(tl->tl_used) - 1; | |
5781 | current_tail = le32_to_cpu(tl->tl_recs[tail_index].t_start); | |
5782 | current_tail += le32_to_cpu(tl->tl_recs[tail_index].t_clusters); | |
5783 | ||
5784 | return current_tail == new_start; | |
5785 | } | |
5786 | ||
063c4561 MF |
5787 | int ocfs2_truncate_log_append(struct ocfs2_super *osb, |
5788 | handle_t *handle, | |
5789 | u64 start_blk, | |
5790 | unsigned int num_clusters) | |
ccd979bd MF |
5791 | { |
5792 | int status, index; | |
5793 | unsigned int start_cluster, tl_count; | |
5794 | struct inode *tl_inode = osb->osb_tl_inode; | |
5795 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5796 | struct ocfs2_dinode *di; | |
5797 | struct ocfs2_truncate_log *tl; | |
5798 | ||
b0697053 MF |
5799 | mlog_entry("start_blk = %llu, num_clusters = %u\n", |
5800 | (unsigned long long)start_blk, num_clusters); | |
ccd979bd | 5801 | |
1b1dcc1b | 5802 | BUG_ON(mutex_trylock(&tl_inode->i_mutex)); |
ccd979bd MF |
5803 | |
5804 | start_cluster = ocfs2_blocks_to_clusters(osb->sb, start_blk); | |
5805 | ||
5806 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
ccd979bd | 5807 | |
10995aa2 JB |
5808 | /* tl_bh is loaded from ocfs2_truncate_log_init(). It's validated |
5809 | * by the underlying call to ocfs2_read_inode_block(), so any | |
5810 | * corruption is a code bug */ | |
5811 | BUG_ON(!OCFS2_IS_VALID_DINODE(di)); | |
5812 | ||
5813 | tl = &di->id2.i_dealloc; | |
ccd979bd MF |
5814 | tl_count = le16_to_cpu(tl->tl_count); |
5815 | mlog_bug_on_msg(tl_count > ocfs2_truncate_recs_per_inode(osb->sb) || | |
5816 | tl_count == 0, | |
b0697053 MF |
5817 | "Truncate record count on #%llu invalid " |
5818 | "wanted %u, actual %u\n", | |
5819 | (unsigned long long)OCFS2_I(tl_inode)->ip_blkno, | |
ccd979bd MF |
5820 | ocfs2_truncate_recs_per_inode(osb->sb), |
5821 | le16_to_cpu(tl->tl_count)); | |
5822 | ||
5823 | /* Caller should have known to flush before calling us. */ | |
5824 | index = le16_to_cpu(tl->tl_used); | |
5825 | if (index >= tl_count) { | |
5826 | status = -ENOSPC; | |
5827 | mlog_errno(status); | |
5828 | goto bail; | |
5829 | } | |
5830 | ||
0cf2f763 | 5831 | status = ocfs2_journal_access_di(handle, INODE_CACHE(tl_inode), tl_bh, |
13723d00 | 5832 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
5833 | if (status < 0) { |
5834 | mlog_errno(status); | |
5835 | goto bail; | |
5836 | } | |
5837 | ||
5838 | mlog(0, "Log truncate of %u clusters starting at cluster %u to " | |
b0697053 MF |
5839 | "%llu (index = %d)\n", num_clusters, start_cluster, |
5840 | (unsigned long long)OCFS2_I(tl_inode)->ip_blkno, index); | |
ccd979bd MF |
5841 | |
5842 | if (ocfs2_truncate_log_can_coalesce(tl, start_cluster)) { | |
5843 | /* | |
5844 | * Move index back to the record we are coalescing with. | |
5845 | * ocfs2_truncate_log_can_coalesce() guarantees nonzero | |
5846 | */ | |
5847 | index--; | |
5848 | ||
5849 | num_clusters += le32_to_cpu(tl->tl_recs[index].t_clusters); | |
5850 | mlog(0, "Coalesce with index %u (start = %u, clusters = %u)\n", | |
5851 | index, le32_to_cpu(tl->tl_recs[index].t_start), | |
5852 | num_clusters); | |
5853 | } else { | |
5854 | tl->tl_recs[index].t_start = cpu_to_le32(start_cluster); | |
5855 | tl->tl_used = cpu_to_le16(index + 1); | |
5856 | } | |
5857 | tl->tl_recs[index].t_clusters = cpu_to_le32(num_clusters); | |
5858 | ||
ec20cec7 | 5859 | ocfs2_journal_dirty(handle, tl_bh); |
ccd979bd MF |
5860 | |
5861 | bail: | |
5862 | mlog_exit(status); | |
5863 | return status; | |
5864 | } | |
5865 | ||
5866 | static int ocfs2_replay_truncate_records(struct ocfs2_super *osb, | |
1fabe148 | 5867 | handle_t *handle, |
ccd979bd MF |
5868 | struct inode *data_alloc_inode, |
5869 | struct buffer_head *data_alloc_bh) | |
5870 | { | |
5871 | int status = 0; | |
5872 | int i; | |
5873 | unsigned int num_clusters; | |
5874 | u64 start_blk; | |
5875 | struct ocfs2_truncate_rec rec; | |
5876 | struct ocfs2_dinode *di; | |
5877 | struct ocfs2_truncate_log *tl; | |
5878 | struct inode *tl_inode = osb->osb_tl_inode; | |
5879 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5880 | ||
5881 | mlog_entry_void(); | |
5882 | ||
5883 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
5884 | tl = &di->id2.i_dealloc; | |
5885 | i = le16_to_cpu(tl->tl_used) - 1; | |
5886 | while (i >= 0) { | |
5887 | /* Caller has given us at least enough credits to | |
5888 | * update the truncate log dinode */ | |
0cf2f763 | 5889 | status = ocfs2_journal_access_di(handle, INODE_CACHE(tl_inode), tl_bh, |
13723d00 | 5890 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
5891 | if (status < 0) { |
5892 | mlog_errno(status); | |
5893 | goto bail; | |
5894 | } | |
5895 | ||
5896 | tl->tl_used = cpu_to_le16(i); | |
5897 | ||
ec20cec7 | 5898 | ocfs2_journal_dirty(handle, tl_bh); |
ccd979bd MF |
5899 | |
5900 | /* TODO: Perhaps we can calculate the bulk of the | |
5901 | * credits up front rather than extending like | |
5902 | * this. */ | |
5903 | status = ocfs2_extend_trans(handle, | |
5904 | OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC); | |
5905 | if (status < 0) { | |
5906 | mlog_errno(status); | |
5907 | goto bail; | |
5908 | } | |
5909 | ||
5910 | rec = tl->tl_recs[i]; | |
5911 | start_blk = ocfs2_clusters_to_blocks(data_alloc_inode->i_sb, | |
5912 | le32_to_cpu(rec.t_start)); | |
5913 | num_clusters = le32_to_cpu(rec.t_clusters); | |
5914 | ||
5915 | /* if start_blk is not set, we ignore the record as | |
5916 | * invalid. */ | |
5917 | if (start_blk) { | |
5918 | mlog(0, "free record %d, start = %u, clusters = %u\n", | |
5919 | i, le32_to_cpu(rec.t_start), num_clusters); | |
5920 | ||
5921 | status = ocfs2_free_clusters(handle, data_alloc_inode, | |
5922 | data_alloc_bh, start_blk, | |
5923 | num_clusters); | |
5924 | if (status < 0) { | |
5925 | mlog_errno(status); | |
5926 | goto bail; | |
5927 | } | |
5928 | } | |
5929 | i--; | |
5930 | } | |
5931 | ||
5932 | bail: | |
5933 | mlog_exit(status); | |
5934 | return status; | |
5935 | } | |
5936 | ||
1b1dcc1b | 5937 | /* Expects you to already be holding tl_inode->i_mutex */ |
063c4561 | 5938 | int __ocfs2_flush_truncate_log(struct ocfs2_super *osb) |
ccd979bd MF |
5939 | { |
5940 | int status; | |
5941 | unsigned int num_to_flush; | |
1fabe148 | 5942 | handle_t *handle; |
ccd979bd MF |
5943 | struct inode *tl_inode = osb->osb_tl_inode; |
5944 | struct inode *data_alloc_inode = NULL; | |
5945 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5946 | struct buffer_head *data_alloc_bh = NULL; | |
5947 | struct ocfs2_dinode *di; | |
5948 | struct ocfs2_truncate_log *tl; | |
5949 | ||
5950 | mlog_entry_void(); | |
5951 | ||
1b1dcc1b | 5952 | BUG_ON(mutex_trylock(&tl_inode->i_mutex)); |
ccd979bd MF |
5953 | |
5954 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
ccd979bd | 5955 | |
10995aa2 JB |
5956 | /* tl_bh is loaded from ocfs2_truncate_log_init(). It's validated |
5957 | * by the underlying call to ocfs2_read_inode_block(), so any | |
5958 | * corruption is a code bug */ | |
5959 | BUG_ON(!OCFS2_IS_VALID_DINODE(di)); | |
5960 | ||
5961 | tl = &di->id2.i_dealloc; | |
ccd979bd | 5962 | num_to_flush = le16_to_cpu(tl->tl_used); |
b0697053 MF |
5963 | mlog(0, "Flush %u records from truncate log #%llu\n", |
5964 | num_to_flush, (unsigned long long)OCFS2_I(tl_inode)->ip_blkno); | |
ccd979bd MF |
5965 | if (!num_to_flush) { |
5966 | status = 0; | |
e08dc8b9 | 5967 | goto out; |
ccd979bd MF |
5968 | } |
5969 | ||
5970 | data_alloc_inode = ocfs2_get_system_file_inode(osb, | |
5971 | GLOBAL_BITMAP_SYSTEM_INODE, | |
5972 | OCFS2_INVALID_SLOT); | |
5973 | if (!data_alloc_inode) { | |
5974 | status = -EINVAL; | |
5975 | mlog(ML_ERROR, "Could not get bitmap inode!\n"); | |
e08dc8b9 | 5976 | goto out; |
ccd979bd MF |
5977 | } |
5978 | ||
e08dc8b9 MF |
5979 | mutex_lock(&data_alloc_inode->i_mutex); |
5980 | ||
e63aecb6 | 5981 | status = ocfs2_inode_lock(data_alloc_inode, &data_alloc_bh, 1); |
ccd979bd MF |
5982 | if (status < 0) { |
5983 | mlog_errno(status); | |
e08dc8b9 | 5984 | goto out_mutex; |
ccd979bd MF |
5985 | } |
5986 | ||
65eff9cc | 5987 | handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE); |
ccd979bd MF |
5988 | if (IS_ERR(handle)) { |
5989 | status = PTR_ERR(handle); | |
ccd979bd | 5990 | mlog_errno(status); |
e08dc8b9 | 5991 | goto out_unlock; |
ccd979bd MF |
5992 | } |
5993 | ||
5994 | status = ocfs2_replay_truncate_records(osb, handle, data_alloc_inode, | |
5995 | data_alloc_bh); | |
e08dc8b9 | 5996 | if (status < 0) |
ccd979bd | 5997 | mlog_errno(status); |
ccd979bd | 5998 | |
02dc1af4 | 5999 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 6000 | |
e08dc8b9 MF |
6001 | out_unlock: |
6002 | brelse(data_alloc_bh); | |
e63aecb6 | 6003 | ocfs2_inode_unlock(data_alloc_inode, 1); |
ccd979bd | 6004 | |
e08dc8b9 MF |
6005 | out_mutex: |
6006 | mutex_unlock(&data_alloc_inode->i_mutex); | |
6007 | iput(data_alloc_inode); | |
ccd979bd | 6008 | |
e08dc8b9 | 6009 | out: |
ccd979bd MF |
6010 | mlog_exit(status); |
6011 | return status; | |
6012 | } | |
6013 | ||
6014 | int ocfs2_flush_truncate_log(struct ocfs2_super *osb) | |
6015 | { | |
6016 | int status; | |
6017 | struct inode *tl_inode = osb->osb_tl_inode; | |
6018 | ||
1b1dcc1b | 6019 | mutex_lock(&tl_inode->i_mutex); |
ccd979bd | 6020 | status = __ocfs2_flush_truncate_log(osb); |
1b1dcc1b | 6021 | mutex_unlock(&tl_inode->i_mutex); |
ccd979bd MF |
6022 | |
6023 | return status; | |
6024 | } | |
6025 | ||
c4028958 | 6026 | static void ocfs2_truncate_log_worker(struct work_struct *work) |
ccd979bd MF |
6027 | { |
6028 | int status; | |
c4028958 DH |
6029 | struct ocfs2_super *osb = |
6030 | container_of(work, struct ocfs2_super, | |
6031 | osb_truncate_log_wq.work); | |
ccd979bd MF |
6032 | |
6033 | mlog_entry_void(); | |
6034 | ||
6035 | status = ocfs2_flush_truncate_log(osb); | |
6036 | if (status < 0) | |
6037 | mlog_errno(status); | |
4d0ddb2c | 6038 | else |
b89c5428 | 6039 | ocfs2_init_steal_slots(osb); |
ccd979bd MF |
6040 | |
6041 | mlog_exit(status); | |
6042 | } | |
6043 | ||
6044 | #define OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL (2 * HZ) | |
6045 | void ocfs2_schedule_truncate_log_flush(struct ocfs2_super *osb, | |
6046 | int cancel) | |
6047 | { | |
6048 | if (osb->osb_tl_inode) { | |
6049 | /* We want to push off log flushes while truncates are | |
6050 | * still running. */ | |
6051 | if (cancel) | |
6052 | cancel_delayed_work(&osb->osb_truncate_log_wq); | |
6053 | ||
6054 | queue_delayed_work(ocfs2_wq, &osb->osb_truncate_log_wq, | |
6055 | OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL); | |
6056 | } | |
6057 | } | |
6058 | ||
6059 | static int ocfs2_get_truncate_log_info(struct ocfs2_super *osb, | |
6060 | int slot_num, | |
6061 | struct inode **tl_inode, | |
6062 | struct buffer_head **tl_bh) | |
6063 | { | |
6064 | int status; | |
6065 | struct inode *inode = NULL; | |
6066 | struct buffer_head *bh = NULL; | |
6067 | ||
6068 | inode = ocfs2_get_system_file_inode(osb, | |
6069 | TRUNCATE_LOG_SYSTEM_INODE, | |
6070 | slot_num); | |
6071 | if (!inode) { | |
6072 | status = -EINVAL; | |
6073 | mlog(ML_ERROR, "Could not get load truncate log inode!\n"); | |
6074 | goto bail; | |
6075 | } | |
6076 | ||
b657c95c | 6077 | status = ocfs2_read_inode_block(inode, &bh); |
ccd979bd MF |
6078 | if (status < 0) { |
6079 | iput(inode); | |
6080 | mlog_errno(status); | |
6081 | goto bail; | |
6082 | } | |
6083 | ||
6084 | *tl_inode = inode; | |
6085 | *tl_bh = bh; | |
6086 | bail: | |
6087 | mlog_exit(status); | |
6088 | return status; | |
6089 | } | |
6090 | ||
6091 | /* called during the 1st stage of node recovery. we stamp a clean | |
6092 | * truncate log and pass back a copy for processing later. if the | |
6093 | * truncate log does not require processing, a *tl_copy is set to | |
6094 | * NULL. */ | |
6095 | int ocfs2_begin_truncate_log_recovery(struct ocfs2_super *osb, | |
6096 | int slot_num, | |
6097 | struct ocfs2_dinode **tl_copy) | |
6098 | { | |
6099 | int status; | |
6100 | struct inode *tl_inode = NULL; | |
6101 | struct buffer_head *tl_bh = NULL; | |
6102 | struct ocfs2_dinode *di; | |
6103 | struct ocfs2_truncate_log *tl; | |
6104 | ||
6105 | *tl_copy = NULL; | |
6106 | ||
6107 | mlog(0, "recover truncate log from slot %d\n", slot_num); | |
6108 | ||
6109 | status = ocfs2_get_truncate_log_info(osb, slot_num, &tl_inode, &tl_bh); | |
6110 | if (status < 0) { | |
6111 | mlog_errno(status); | |
6112 | goto bail; | |
6113 | } | |
6114 | ||
6115 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
ccd979bd | 6116 | |
10995aa2 JB |
6117 | /* tl_bh is loaded from ocfs2_get_truncate_log_info(). It's |
6118 | * validated by the underlying call to ocfs2_read_inode_block(), | |
6119 | * so any corruption is a code bug */ | |
6120 | BUG_ON(!OCFS2_IS_VALID_DINODE(di)); | |
6121 | ||
6122 | tl = &di->id2.i_dealloc; | |
ccd979bd MF |
6123 | if (le16_to_cpu(tl->tl_used)) { |
6124 | mlog(0, "We'll have %u logs to recover\n", | |
6125 | le16_to_cpu(tl->tl_used)); | |
6126 | ||
6127 | *tl_copy = kmalloc(tl_bh->b_size, GFP_KERNEL); | |
6128 | if (!(*tl_copy)) { | |
6129 | status = -ENOMEM; | |
6130 | mlog_errno(status); | |
6131 | goto bail; | |
6132 | } | |
6133 | ||
6134 | /* Assuming the write-out below goes well, this copy | |
6135 | * will be passed back to recovery for processing. */ | |
6136 | memcpy(*tl_copy, tl_bh->b_data, tl_bh->b_size); | |
6137 | ||
6138 | /* All we need to do to clear the truncate log is set | |
6139 | * tl_used. */ | |
6140 | tl->tl_used = 0; | |
6141 | ||
13723d00 | 6142 | ocfs2_compute_meta_ecc(osb->sb, tl_bh->b_data, &di->i_check); |
8cb471e8 | 6143 | status = ocfs2_write_block(osb, tl_bh, INODE_CACHE(tl_inode)); |
ccd979bd MF |
6144 | if (status < 0) { |
6145 | mlog_errno(status); | |
6146 | goto bail; | |
6147 | } | |
6148 | } | |
6149 | ||
6150 | bail: | |
6151 | if (tl_inode) | |
6152 | iput(tl_inode); | |
a81cb88b | 6153 | brelse(tl_bh); |
ccd979bd MF |
6154 | |
6155 | if (status < 0 && (*tl_copy)) { | |
6156 | kfree(*tl_copy); | |
6157 | *tl_copy = NULL; | |
6158 | } | |
6159 | ||
6160 | mlog_exit(status); | |
6161 | return status; | |
6162 | } | |
6163 | ||
6164 | int ocfs2_complete_truncate_log_recovery(struct ocfs2_super *osb, | |
6165 | struct ocfs2_dinode *tl_copy) | |
6166 | { | |
6167 | int status = 0; | |
6168 | int i; | |
6169 | unsigned int clusters, num_recs, start_cluster; | |
6170 | u64 start_blk; | |
1fabe148 | 6171 | handle_t *handle; |
ccd979bd MF |
6172 | struct inode *tl_inode = osb->osb_tl_inode; |
6173 | struct ocfs2_truncate_log *tl; | |
6174 | ||
6175 | mlog_entry_void(); | |
6176 | ||
6177 | if (OCFS2_I(tl_inode)->ip_blkno == le64_to_cpu(tl_copy->i_blkno)) { | |
6178 | mlog(ML_ERROR, "Asked to recover my own truncate log!\n"); | |
6179 | return -EINVAL; | |
6180 | } | |
6181 | ||
6182 | tl = &tl_copy->id2.i_dealloc; | |
6183 | num_recs = le16_to_cpu(tl->tl_used); | |
b0697053 | 6184 | mlog(0, "cleanup %u records from %llu\n", num_recs, |
1ca1a111 | 6185 | (unsigned long long)le64_to_cpu(tl_copy->i_blkno)); |
ccd979bd | 6186 | |
1b1dcc1b | 6187 | mutex_lock(&tl_inode->i_mutex); |
ccd979bd MF |
6188 | for(i = 0; i < num_recs; i++) { |
6189 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
6190 | status = __ocfs2_flush_truncate_log(osb); | |
6191 | if (status < 0) { | |
6192 | mlog_errno(status); | |
6193 | goto bail_up; | |
6194 | } | |
6195 | } | |
6196 | ||
65eff9cc | 6197 | handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE); |
ccd979bd MF |
6198 | if (IS_ERR(handle)) { |
6199 | status = PTR_ERR(handle); | |
6200 | mlog_errno(status); | |
6201 | goto bail_up; | |
6202 | } | |
6203 | ||
6204 | clusters = le32_to_cpu(tl->tl_recs[i].t_clusters); | |
6205 | start_cluster = le32_to_cpu(tl->tl_recs[i].t_start); | |
6206 | start_blk = ocfs2_clusters_to_blocks(osb->sb, start_cluster); | |
6207 | ||
6208 | status = ocfs2_truncate_log_append(osb, handle, | |
6209 | start_blk, clusters); | |
02dc1af4 | 6210 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
6211 | if (status < 0) { |
6212 | mlog_errno(status); | |
6213 | goto bail_up; | |
6214 | } | |
6215 | } | |
6216 | ||
6217 | bail_up: | |
1b1dcc1b | 6218 | mutex_unlock(&tl_inode->i_mutex); |
ccd979bd MF |
6219 | |
6220 | mlog_exit(status); | |
6221 | return status; | |
6222 | } | |
6223 | ||
6224 | void ocfs2_truncate_log_shutdown(struct ocfs2_super *osb) | |
6225 | { | |
6226 | int status; | |
6227 | struct inode *tl_inode = osb->osb_tl_inode; | |
6228 | ||
6229 | mlog_entry_void(); | |
6230 | ||
6231 | if (tl_inode) { | |
6232 | cancel_delayed_work(&osb->osb_truncate_log_wq); | |
6233 | flush_workqueue(ocfs2_wq); | |
6234 | ||
6235 | status = ocfs2_flush_truncate_log(osb); | |
6236 | if (status < 0) | |
6237 | mlog_errno(status); | |
6238 | ||
6239 | brelse(osb->osb_tl_bh); | |
6240 | iput(osb->osb_tl_inode); | |
6241 | } | |
6242 | ||
6243 | mlog_exit_void(); | |
6244 | } | |
6245 | ||
6246 | int ocfs2_truncate_log_init(struct ocfs2_super *osb) | |
6247 | { | |
6248 | int status; | |
6249 | struct inode *tl_inode = NULL; | |
6250 | struct buffer_head *tl_bh = NULL; | |
6251 | ||
6252 | mlog_entry_void(); | |
6253 | ||
6254 | status = ocfs2_get_truncate_log_info(osb, | |
6255 | osb->slot_num, | |
6256 | &tl_inode, | |
6257 | &tl_bh); | |
6258 | if (status < 0) | |
6259 | mlog_errno(status); | |
6260 | ||
6261 | /* ocfs2_truncate_log_shutdown keys on the existence of | |
6262 | * osb->osb_tl_inode so we don't set any of the osb variables | |
6263 | * until we're sure all is well. */ | |
c4028958 DH |
6264 | INIT_DELAYED_WORK(&osb->osb_truncate_log_wq, |
6265 | ocfs2_truncate_log_worker); | |
ccd979bd MF |
6266 | osb->osb_tl_bh = tl_bh; |
6267 | osb->osb_tl_inode = tl_inode; | |
6268 | ||
6269 | mlog_exit(status); | |
6270 | return status; | |
6271 | } | |
6272 | ||
2b604351 MF |
6273 | /* |
6274 | * Delayed de-allocation of suballocator blocks. | |
6275 | * | |
6276 | * Some sets of block de-allocations might involve multiple suballocator inodes. | |
6277 | * | |
6278 | * The locking for this can get extremely complicated, especially when | |
6279 | * the suballocator inodes to delete from aren't known until deep | |
6280 | * within an unrelated codepath. | |
6281 | * | |
6282 | * ocfs2_extent_block structures are a good example of this - an inode | |
6283 | * btree could have been grown by any number of nodes each allocating | |
6284 | * out of their own suballoc inode. | |
6285 | * | |
6286 | * These structures allow the delay of block de-allocation until a | |
6287 | * later time, when locking of multiple cluster inodes won't cause | |
6288 | * deadlock. | |
6289 | */ | |
6290 | ||
6291 | /* | |
2891d290 TM |
6292 | * Describe a single bit freed from a suballocator. For the block |
6293 | * suballocators, it represents one block. For the global cluster | |
6294 | * allocator, it represents some clusters and free_bit indicates | |
6295 | * clusters number. | |
2b604351 MF |
6296 | */ |
6297 | struct ocfs2_cached_block_free { | |
6298 | struct ocfs2_cached_block_free *free_next; | |
74380c47 | 6299 | u64 free_bg; |
2b604351 MF |
6300 | u64 free_blk; |
6301 | unsigned int free_bit; | |
6302 | }; | |
6303 | ||
6304 | struct ocfs2_per_slot_free_list { | |
6305 | struct ocfs2_per_slot_free_list *f_next_suballocator; | |
6306 | int f_inode_type; | |
6307 | int f_slot; | |
6308 | struct ocfs2_cached_block_free *f_first; | |
6309 | }; | |
6310 | ||
2891d290 TM |
6311 | static int ocfs2_free_cached_blocks(struct ocfs2_super *osb, |
6312 | int sysfile_type, | |
6313 | int slot, | |
6314 | struct ocfs2_cached_block_free *head) | |
2b604351 MF |
6315 | { |
6316 | int ret; | |
6317 | u64 bg_blkno; | |
6318 | handle_t *handle; | |
6319 | struct inode *inode; | |
6320 | struct buffer_head *di_bh = NULL; | |
6321 | struct ocfs2_cached_block_free *tmp; | |
6322 | ||
6323 | inode = ocfs2_get_system_file_inode(osb, sysfile_type, slot); | |
6324 | if (!inode) { | |
6325 | ret = -EINVAL; | |
6326 | mlog_errno(ret); | |
6327 | goto out; | |
6328 | } | |
6329 | ||
6330 | mutex_lock(&inode->i_mutex); | |
6331 | ||
e63aecb6 | 6332 | ret = ocfs2_inode_lock(inode, &di_bh, 1); |
2b604351 MF |
6333 | if (ret) { |
6334 | mlog_errno(ret); | |
6335 | goto out_mutex; | |
6336 | } | |
6337 | ||
6338 | handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE); | |
6339 | if (IS_ERR(handle)) { | |
6340 | ret = PTR_ERR(handle); | |
6341 | mlog_errno(ret); | |
6342 | goto out_unlock; | |
6343 | } | |
6344 | ||
6345 | while (head) { | |
74380c47 TM |
6346 | if (head->free_bg) |
6347 | bg_blkno = head->free_bg; | |
6348 | else | |
6349 | bg_blkno = ocfs2_which_suballoc_group(head->free_blk, | |
6350 | head->free_bit); | |
2b604351 MF |
6351 | mlog(0, "Free bit: (bit %u, blkno %llu)\n", |
6352 | head->free_bit, (unsigned long long)head->free_blk); | |
6353 | ||
6354 | ret = ocfs2_free_suballoc_bits(handle, inode, di_bh, | |
6355 | head->free_bit, bg_blkno, 1); | |
6356 | if (ret) { | |
6357 | mlog_errno(ret); | |
6358 | goto out_journal; | |
6359 | } | |
6360 | ||
6361 | ret = ocfs2_extend_trans(handle, OCFS2_SUBALLOC_FREE); | |
6362 | if (ret) { | |
6363 | mlog_errno(ret); | |
6364 | goto out_journal; | |
6365 | } | |
6366 | ||
6367 | tmp = head; | |
6368 | head = head->free_next; | |
6369 | kfree(tmp); | |
6370 | } | |
6371 | ||
6372 | out_journal: | |
6373 | ocfs2_commit_trans(osb, handle); | |
6374 | ||
6375 | out_unlock: | |
e63aecb6 | 6376 | ocfs2_inode_unlock(inode, 1); |
2b604351 MF |
6377 | brelse(di_bh); |
6378 | out_mutex: | |
6379 | mutex_unlock(&inode->i_mutex); | |
6380 | iput(inode); | |
6381 | out: | |
6382 | while(head) { | |
6383 | /* Premature exit may have left some dangling items. */ | |
6384 | tmp = head; | |
6385 | head = head->free_next; | |
6386 | kfree(tmp); | |
6387 | } | |
6388 | ||
6389 | return ret; | |
6390 | } | |
6391 | ||
2891d290 TM |
6392 | int ocfs2_cache_cluster_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt, |
6393 | u64 blkno, unsigned int bit) | |
6394 | { | |
6395 | int ret = 0; | |
6396 | struct ocfs2_cached_block_free *item; | |
6397 | ||
74380c47 | 6398 | item = kzalloc(sizeof(*item), GFP_NOFS); |
2891d290 TM |
6399 | if (item == NULL) { |
6400 | ret = -ENOMEM; | |
6401 | mlog_errno(ret); | |
6402 | return ret; | |
6403 | } | |
6404 | ||
6405 | mlog(0, "Insert clusters: (bit %u, blk %llu)\n", | |
6406 | bit, (unsigned long long)blkno); | |
6407 | ||
6408 | item->free_blk = blkno; | |
6409 | item->free_bit = bit; | |
6410 | item->free_next = ctxt->c_global_allocator; | |
6411 | ||
6412 | ctxt->c_global_allocator = item; | |
6413 | return ret; | |
6414 | } | |
6415 | ||
6416 | static int ocfs2_free_cached_clusters(struct ocfs2_super *osb, | |
6417 | struct ocfs2_cached_block_free *head) | |
6418 | { | |
6419 | struct ocfs2_cached_block_free *tmp; | |
6420 | struct inode *tl_inode = osb->osb_tl_inode; | |
6421 | handle_t *handle; | |
6422 | int ret = 0; | |
6423 | ||
6424 | mutex_lock(&tl_inode->i_mutex); | |
6425 | ||
6426 | while (head) { | |
6427 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
6428 | ret = __ocfs2_flush_truncate_log(osb); | |
6429 | if (ret < 0) { | |
6430 | mlog_errno(ret); | |
6431 | break; | |
6432 | } | |
6433 | } | |
6434 | ||
6435 | handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE); | |
6436 | if (IS_ERR(handle)) { | |
6437 | ret = PTR_ERR(handle); | |
6438 | mlog_errno(ret); | |
6439 | break; | |
6440 | } | |
6441 | ||
6442 | ret = ocfs2_truncate_log_append(osb, handle, head->free_blk, | |
6443 | head->free_bit); | |
6444 | ||
6445 | ocfs2_commit_trans(osb, handle); | |
6446 | tmp = head; | |
6447 | head = head->free_next; | |
6448 | kfree(tmp); | |
6449 | ||
6450 | if (ret < 0) { | |
6451 | mlog_errno(ret); | |
6452 | break; | |
6453 | } | |
6454 | } | |
6455 | ||
6456 | mutex_unlock(&tl_inode->i_mutex); | |
6457 | ||
6458 | while (head) { | |
6459 | /* Premature exit may have left some dangling items. */ | |
6460 | tmp = head; | |
6461 | head = head->free_next; | |
6462 | kfree(tmp); | |
6463 | } | |
6464 | ||
6465 | return ret; | |
6466 | } | |
6467 | ||
2b604351 MF |
6468 | int ocfs2_run_deallocs(struct ocfs2_super *osb, |
6469 | struct ocfs2_cached_dealloc_ctxt *ctxt) | |
6470 | { | |
6471 | int ret = 0, ret2; | |
6472 | struct ocfs2_per_slot_free_list *fl; | |
6473 | ||
6474 | if (!ctxt) | |
6475 | return 0; | |
6476 | ||
6477 | while (ctxt->c_first_suballocator) { | |
6478 | fl = ctxt->c_first_suballocator; | |
6479 | ||
6480 | if (fl->f_first) { | |
6481 | mlog(0, "Free items: (type %u, slot %d)\n", | |
6482 | fl->f_inode_type, fl->f_slot); | |
2891d290 TM |
6483 | ret2 = ocfs2_free_cached_blocks(osb, |
6484 | fl->f_inode_type, | |
6485 | fl->f_slot, | |
6486 | fl->f_first); | |
2b604351 MF |
6487 | if (ret2) |
6488 | mlog_errno(ret2); | |
6489 | if (!ret) | |
6490 | ret = ret2; | |
6491 | } | |
6492 | ||
6493 | ctxt->c_first_suballocator = fl->f_next_suballocator; | |
6494 | kfree(fl); | |
6495 | } | |
6496 | ||
2891d290 TM |
6497 | if (ctxt->c_global_allocator) { |
6498 | ret2 = ocfs2_free_cached_clusters(osb, | |
6499 | ctxt->c_global_allocator); | |
6500 | if (ret2) | |
6501 | mlog_errno(ret2); | |
6502 | if (!ret) | |
6503 | ret = ret2; | |
6504 | ||
6505 | ctxt->c_global_allocator = NULL; | |
6506 | } | |
6507 | ||
2b604351 MF |
6508 | return ret; |
6509 | } | |
6510 | ||
6511 | static struct ocfs2_per_slot_free_list * | |
6512 | ocfs2_find_per_slot_free_list(int type, | |
6513 | int slot, | |
6514 | struct ocfs2_cached_dealloc_ctxt *ctxt) | |
6515 | { | |
6516 | struct ocfs2_per_slot_free_list *fl = ctxt->c_first_suballocator; | |
6517 | ||
6518 | while (fl) { | |
6519 | if (fl->f_inode_type == type && fl->f_slot == slot) | |
6520 | return fl; | |
6521 | ||
6522 | fl = fl->f_next_suballocator; | |
6523 | } | |
6524 | ||
6525 | fl = kmalloc(sizeof(*fl), GFP_NOFS); | |
6526 | if (fl) { | |
6527 | fl->f_inode_type = type; | |
6528 | fl->f_slot = slot; | |
6529 | fl->f_first = NULL; | |
6530 | fl->f_next_suballocator = ctxt->c_first_suballocator; | |
6531 | ||
6532 | ctxt->c_first_suballocator = fl; | |
6533 | } | |
6534 | return fl; | |
6535 | } | |
6536 | ||
1823cb0b | 6537 | int ocfs2_cache_block_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt, |
74380c47 TM |
6538 | int type, int slot, u64 suballoc, |
6539 | u64 blkno, unsigned int bit) | |
2b604351 MF |
6540 | { |
6541 | int ret; | |
6542 | struct ocfs2_per_slot_free_list *fl; | |
6543 | struct ocfs2_cached_block_free *item; | |
6544 | ||
6545 | fl = ocfs2_find_per_slot_free_list(type, slot, ctxt); | |
6546 | if (fl == NULL) { | |
6547 | ret = -ENOMEM; | |
6548 | mlog_errno(ret); | |
6549 | goto out; | |
6550 | } | |
6551 | ||
74380c47 | 6552 | item = kzalloc(sizeof(*item), GFP_NOFS); |
2b604351 MF |
6553 | if (item == NULL) { |
6554 | ret = -ENOMEM; | |
6555 | mlog_errno(ret); | |
6556 | goto out; | |
6557 | } | |
6558 | ||
6559 | mlog(0, "Insert: (type %d, slot %u, bit %u, blk %llu)\n", | |
6560 | type, slot, bit, (unsigned long long)blkno); | |
6561 | ||
74380c47 | 6562 | item->free_bg = suballoc; |
2b604351 MF |
6563 | item->free_blk = blkno; |
6564 | item->free_bit = bit; | |
6565 | item->free_next = fl->f_first; | |
6566 | ||
6567 | fl->f_first = item; | |
6568 | ||
6569 | ret = 0; | |
6570 | out: | |
6571 | return ret; | |
6572 | } | |
6573 | ||
59a5e416 MF |
6574 | static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt, |
6575 | struct ocfs2_extent_block *eb) | |
6576 | { | |
6577 | return ocfs2_cache_block_dealloc(ctxt, EXTENT_ALLOC_SYSTEM_INODE, | |
6578 | le16_to_cpu(eb->h_suballoc_slot), | |
74380c47 | 6579 | le64_to_cpu(eb->h_suballoc_loc), |
59a5e416 MF |
6580 | le64_to_cpu(eb->h_blkno), |
6581 | le16_to_cpu(eb->h_suballoc_bit)); | |
6582 | } | |
6583 | ||
2b4e30fb | 6584 | static int ocfs2_zero_func(handle_t *handle, struct buffer_head *bh) |
60b11392 MF |
6585 | { |
6586 | set_buffer_uptodate(bh); | |
6587 | mark_buffer_dirty(bh); | |
6588 | return 0; | |
6589 | } | |
6590 | ||
6f70fa51 TM |
6591 | void ocfs2_map_and_dirty_page(struct inode *inode, handle_t *handle, |
6592 | unsigned int from, unsigned int to, | |
6593 | struct page *page, int zero, u64 *phys) | |
1d410a6e MF |
6594 | { |
6595 | int ret, partial = 0; | |
6596 | ||
6597 | ret = ocfs2_map_page_blocks(page, phys, inode, from, to, 0); | |
6598 | if (ret) | |
6599 | mlog_errno(ret); | |
6600 | ||
6601 | if (zero) | |
eebd2aa3 | 6602 | zero_user_segment(page, from, to); |
1d410a6e MF |
6603 | |
6604 | /* | |
6605 | * Need to set the buffers we zero'd into uptodate | |
6606 | * here if they aren't - ocfs2_map_page_blocks() | |
6607 | * might've skipped some | |
6608 | */ | |
2b4e30fb JB |
6609 | ret = walk_page_buffers(handle, page_buffers(page), |
6610 | from, to, &partial, | |
6611 | ocfs2_zero_func); | |
6612 | if (ret < 0) | |
6613 | mlog_errno(ret); | |
6614 | else if (ocfs2_should_order_data(inode)) { | |
6615 | ret = ocfs2_jbd2_file_inode(handle, inode); | |
1d410a6e MF |
6616 | if (ret < 0) |
6617 | mlog_errno(ret); | |
6618 | } | |
6619 | ||
6620 | if (!partial) | |
6621 | SetPageUptodate(page); | |
6622 | ||
6623 | flush_dcache_page(page); | |
6624 | } | |
6625 | ||
35edec1d MF |
6626 | static void ocfs2_zero_cluster_pages(struct inode *inode, loff_t start, |
6627 | loff_t end, struct page **pages, | |
6628 | int numpages, u64 phys, handle_t *handle) | |
60b11392 | 6629 | { |
1d410a6e | 6630 | int i; |
60b11392 MF |
6631 | struct page *page; |
6632 | unsigned int from, to = PAGE_CACHE_SIZE; | |
6633 | struct super_block *sb = inode->i_sb; | |
6634 | ||
6635 | BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(sb))); | |
6636 | ||
6637 | if (numpages == 0) | |
6638 | goto out; | |
6639 | ||
35edec1d | 6640 | to = PAGE_CACHE_SIZE; |
60b11392 MF |
6641 | for(i = 0; i < numpages; i++) { |
6642 | page = pages[i]; | |
6643 | ||
35edec1d MF |
6644 | from = start & (PAGE_CACHE_SIZE - 1); |
6645 | if ((end >> PAGE_CACHE_SHIFT) == page->index) | |
6646 | to = end & (PAGE_CACHE_SIZE - 1); | |
6647 | ||
60b11392 MF |
6648 | BUG_ON(from > PAGE_CACHE_SIZE); |
6649 | BUG_ON(to > PAGE_CACHE_SIZE); | |
6650 | ||
1d410a6e MF |
6651 | ocfs2_map_and_dirty_page(inode, handle, from, to, page, 1, |
6652 | &phys); | |
60b11392 | 6653 | |
35edec1d | 6654 | start = (page->index + 1) << PAGE_CACHE_SHIFT; |
60b11392 MF |
6655 | } |
6656 | out: | |
1d410a6e MF |
6657 | if (pages) |
6658 | ocfs2_unlock_and_free_pages(pages, numpages); | |
60b11392 MF |
6659 | } |
6660 | ||
6f70fa51 TM |
6661 | int ocfs2_grab_pages(struct inode *inode, loff_t start, loff_t end, |
6662 | struct page **pages, int *num) | |
60b11392 | 6663 | { |
1d410a6e | 6664 | int numpages, ret = 0; |
60b11392 MF |
6665 | struct address_space *mapping = inode->i_mapping; |
6666 | unsigned long index; | |
35edec1d | 6667 | loff_t last_page_bytes; |
60b11392 | 6668 | |
35edec1d | 6669 | BUG_ON(start > end); |
60b11392 | 6670 | |
1d410a6e | 6671 | numpages = 0; |
35edec1d MF |
6672 | last_page_bytes = PAGE_ALIGN(end); |
6673 | index = start >> PAGE_CACHE_SHIFT; | |
60b11392 | 6674 | do { |
9b4c0ff3 | 6675 | pages[numpages] = find_or_create_page(mapping, index, GFP_NOFS); |
60b11392 MF |
6676 | if (!pages[numpages]) { |
6677 | ret = -ENOMEM; | |
6678 | mlog_errno(ret); | |
6679 | goto out; | |
6680 | } | |
6681 | ||
6682 | numpages++; | |
6683 | index++; | |
35edec1d | 6684 | } while (index < (last_page_bytes >> PAGE_CACHE_SHIFT)); |
60b11392 MF |
6685 | |
6686 | out: | |
6687 | if (ret != 0) { | |
1d410a6e MF |
6688 | if (pages) |
6689 | ocfs2_unlock_and_free_pages(pages, numpages); | |
60b11392 MF |
6690 | numpages = 0; |
6691 | } | |
6692 | ||
6693 | *num = numpages; | |
6694 | ||
6695 | return ret; | |
6696 | } | |
6697 | ||
6f70fa51 TM |
6698 | static int ocfs2_grab_eof_pages(struct inode *inode, loff_t start, loff_t end, |
6699 | struct page **pages, int *num) | |
6700 | { | |
6701 | struct super_block *sb = inode->i_sb; | |
6702 | ||
6703 | BUG_ON(start >> OCFS2_SB(sb)->s_clustersize_bits != | |
6704 | (end - 1) >> OCFS2_SB(sb)->s_clustersize_bits); | |
6705 | ||
6706 | return ocfs2_grab_pages(inode, start, end, pages, num); | |
6707 | } | |
6708 | ||
60b11392 MF |
6709 | /* |
6710 | * Zero the area past i_size but still within an allocated | |
6711 | * cluster. This avoids exposing nonzero data on subsequent file | |
6712 | * extends. | |
6713 | * | |
6714 | * We need to call this before i_size is updated on the inode because | |
6715 | * otherwise block_write_full_page() will skip writeout of pages past | |
6716 | * i_size. The new_i_size parameter is passed for this reason. | |
6717 | */ | |
35edec1d MF |
6718 | int ocfs2_zero_range_for_truncate(struct inode *inode, handle_t *handle, |
6719 | u64 range_start, u64 range_end) | |
60b11392 | 6720 | { |
1d410a6e | 6721 | int ret = 0, numpages; |
60b11392 MF |
6722 | struct page **pages = NULL; |
6723 | u64 phys; | |
1d410a6e MF |
6724 | unsigned int ext_flags; |
6725 | struct super_block *sb = inode->i_sb; | |
60b11392 MF |
6726 | |
6727 | /* | |
6728 | * File systems which don't support sparse files zero on every | |
6729 | * extend. | |
6730 | */ | |
1d410a6e | 6731 | if (!ocfs2_sparse_alloc(OCFS2_SB(sb))) |
60b11392 MF |
6732 | return 0; |
6733 | ||
1d410a6e | 6734 | pages = kcalloc(ocfs2_pages_per_cluster(sb), |
60b11392 MF |
6735 | sizeof(struct page *), GFP_NOFS); |
6736 | if (pages == NULL) { | |
6737 | ret = -ENOMEM; | |
6738 | mlog_errno(ret); | |
6739 | goto out; | |
6740 | } | |
6741 | ||
1d410a6e MF |
6742 | if (range_start == range_end) |
6743 | goto out; | |
6744 | ||
6745 | ret = ocfs2_extent_map_get_blocks(inode, | |
6746 | range_start >> sb->s_blocksize_bits, | |
6747 | &phys, NULL, &ext_flags); | |
60b11392 MF |
6748 | if (ret) { |
6749 | mlog_errno(ret); | |
6750 | goto out; | |
6751 | } | |
6752 | ||
1d410a6e MF |
6753 | /* |
6754 | * Tail is a hole, or is marked unwritten. In either case, we | |
6755 | * can count on read and write to return/push zero's. | |
6756 | */ | |
6757 | if (phys == 0 || ext_flags & OCFS2_EXT_UNWRITTEN) | |
60b11392 MF |
6758 | goto out; |
6759 | ||
1d410a6e MF |
6760 | ret = ocfs2_grab_eof_pages(inode, range_start, range_end, pages, |
6761 | &numpages); | |
6762 | if (ret) { | |
6763 | mlog_errno(ret); | |
6764 | goto out; | |
6765 | } | |
6766 | ||
35edec1d MF |
6767 | ocfs2_zero_cluster_pages(inode, range_start, range_end, pages, |
6768 | numpages, phys, handle); | |
60b11392 MF |
6769 | |
6770 | /* | |
6771 | * Initiate writeout of the pages we zero'd here. We don't | |
6772 | * wait on them - the truncate_inode_pages() call later will | |
6773 | * do that for us. | |
6774 | */ | |
2cfd30ad CH |
6775 | ret = filemap_fdatawrite_range(inode->i_mapping, range_start, |
6776 | range_end - 1); | |
60b11392 MF |
6777 | if (ret) |
6778 | mlog_errno(ret); | |
6779 | ||
6780 | out: | |
6781 | if (pages) | |
6782 | kfree(pages); | |
6783 | ||
6784 | return ret; | |
6785 | } | |
6786 | ||
fdd77704 TY |
6787 | static void ocfs2_zero_dinode_id2_with_xattr(struct inode *inode, |
6788 | struct ocfs2_dinode *di) | |
1afc32b9 MF |
6789 | { |
6790 | unsigned int blocksize = 1 << inode->i_sb->s_blocksize_bits; | |
fdd77704 | 6791 | unsigned int xattrsize = le16_to_cpu(di->i_xattr_inline_size); |
1afc32b9 | 6792 | |
fdd77704 TY |
6793 | if (le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_XATTR_FL) |
6794 | memset(&di->id2, 0, blocksize - | |
6795 | offsetof(struct ocfs2_dinode, id2) - | |
6796 | xattrsize); | |
6797 | else | |
6798 | memset(&di->id2, 0, blocksize - | |
6799 | offsetof(struct ocfs2_dinode, id2)); | |
1afc32b9 MF |
6800 | } |
6801 | ||
5b6a3a2b MF |
6802 | void ocfs2_dinode_new_extent_list(struct inode *inode, |
6803 | struct ocfs2_dinode *di) | |
6804 | { | |
fdd77704 | 6805 | ocfs2_zero_dinode_id2_with_xattr(inode, di); |
5b6a3a2b MF |
6806 | di->id2.i_list.l_tree_depth = 0; |
6807 | di->id2.i_list.l_next_free_rec = 0; | |
fdd77704 TY |
6808 | di->id2.i_list.l_count = cpu_to_le16( |
6809 | ocfs2_extent_recs_per_inode_with_xattr(inode->i_sb, di)); | |
5b6a3a2b MF |
6810 | } |
6811 | ||
1afc32b9 MF |
6812 | void ocfs2_set_inode_data_inline(struct inode *inode, struct ocfs2_dinode *di) |
6813 | { | |
6814 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
6815 | struct ocfs2_inline_data *idata = &di->id2.i_data; | |
6816 | ||
6817 | spin_lock(&oi->ip_lock); | |
6818 | oi->ip_dyn_features |= OCFS2_INLINE_DATA_FL; | |
6819 | di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features); | |
6820 | spin_unlock(&oi->ip_lock); | |
6821 | ||
6822 | /* | |
6823 | * We clear the entire i_data structure here so that all | |
6824 | * fields can be properly initialized. | |
6825 | */ | |
fdd77704 | 6826 | ocfs2_zero_dinode_id2_with_xattr(inode, di); |
1afc32b9 | 6827 | |
fdd77704 TY |
6828 | idata->id_count = cpu_to_le16( |
6829 | ocfs2_max_inline_data_with_xattr(inode->i_sb, di)); | |
1afc32b9 MF |
6830 | } |
6831 | ||
6832 | int ocfs2_convert_inline_data_to_extents(struct inode *inode, | |
6833 | struct buffer_head *di_bh) | |
6834 | { | |
6835 | int ret, i, has_data, num_pages = 0; | |
6836 | handle_t *handle; | |
6837 | u64 uninitialized_var(block); | |
6838 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
6839 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
6840 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
1afc32b9 MF |
6841 | struct ocfs2_alloc_context *data_ac = NULL; |
6842 | struct page **pages = NULL; | |
6843 | loff_t end = osb->s_clustersize; | |
f99b9b7c | 6844 | struct ocfs2_extent_tree et; |
a90714c1 | 6845 | int did_quota = 0; |
1afc32b9 MF |
6846 | |
6847 | has_data = i_size_read(inode) ? 1 : 0; | |
6848 | ||
6849 | if (has_data) { | |
6850 | pages = kcalloc(ocfs2_pages_per_cluster(osb->sb), | |
6851 | sizeof(struct page *), GFP_NOFS); | |
6852 | if (pages == NULL) { | |
6853 | ret = -ENOMEM; | |
6854 | mlog_errno(ret); | |
6855 | goto out; | |
6856 | } | |
6857 | ||
6858 | ret = ocfs2_reserve_clusters(osb, 1, &data_ac); | |
6859 | if (ret) { | |
6860 | mlog_errno(ret); | |
6861 | goto out; | |
6862 | } | |
6863 | } | |
6864 | ||
a90714c1 JK |
6865 | handle = ocfs2_start_trans(osb, |
6866 | ocfs2_inline_to_extents_credits(osb->sb)); | |
1afc32b9 MF |
6867 | if (IS_ERR(handle)) { |
6868 | ret = PTR_ERR(handle); | |
6869 | mlog_errno(ret); | |
6870 | goto out_unlock; | |
6871 | } | |
6872 | ||
0cf2f763 | 6873 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh, |
13723d00 | 6874 | OCFS2_JOURNAL_ACCESS_WRITE); |
1afc32b9 MF |
6875 | if (ret) { |
6876 | mlog_errno(ret); | |
6877 | goto out_commit; | |
6878 | } | |
6879 | ||
6880 | if (has_data) { | |
6881 | u32 bit_off, num; | |
6882 | unsigned int page_end; | |
6883 | u64 phys; | |
6884 | ||
5dd4056d CH |
6885 | ret = dquot_alloc_space_nodirty(inode, |
6886 | ocfs2_clusters_to_bytes(osb->sb, 1)); | |
6887 | if (ret) | |
a90714c1 | 6888 | goto out_commit; |
a90714c1 JK |
6889 | did_quota = 1; |
6890 | ||
4fe370af MF |
6891 | data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv; |
6892 | ||
1ed9b777 | 6893 | ret = ocfs2_claim_clusters(handle, data_ac, 1, &bit_off, |
1afc32b9 MF |
6894 | &num); |
6895 | if (ret) { | |
6896 | mlog_errno(ret); | |
6897 | goto out_commit; | |
6898 | } | |
6899 | ||
6900 | /* | |
6901 | * Save two copies, one for insert, and one that can | |
6902 | * be changed by ocfs2_map_and_dirty_page() below. | |
6903 | */ | |
6904 | block = phys = ocfs2_clusters_to_blocks(inode->i_sb, bit_off); | |
6905 | ||
6906 | /* | |
6907 | * Non sparse file systems zero on extend, so no need | |
6908 | * to do that now. | |
6909 | */ | |
6910 | if (!ocfs2_sparse_alloc(osb) && | |
6911 | PAGE_CACHE_SIZE < osb->s_clustersize) | |
6912 | end = PAGE_CACHE_SIZE; | |
6913 | ||
6914 | ret = ocfs2_grab_eof_pages(inode, 0, end, pages, &num_pages); | |
6915 | if (ret) { | |
6916 | mlog_errno(ret); | |
6917 | goto out_commit; | |
6918 | } | |
6919 | ||
6920 | /* | |
6921 | * This should populate the 1st page for us and mark | |
6922 | * it up to date. | |
6923 | */ | |
6924 | ret = ocfs2_read_inline_data(inode, pages[0], di_bh); | |
6925 | if (ret) { | |
6926 | mlog_errno(ret); | |
6927 | goto out_commit; | |
6928 | } | |
6929 | ||
6930 | page_end = PAGE_CACHE_SIZE; | |
6931 | if (PAGE_CACHE_SIZE > osb->s_clustersize) | |
6932 | page_end = osb->s_clustersize; | |
6933 | ||
6934 | for (i = 0; i < num_pages; i++) | |
6935 | ocfs2_map_and_dirty_page(inode, handle, 0, page_end, | |
6936 | pages[i], i > 0, &phys); | |
6937 | } | |
6938 | ||
6939 | spin_lock(&oi->ip_lock); | |
6940 | oi->ip_dyn_features &= ~OCFS2_INLINE_DATA_FL; | |
6941 | di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features); | |
6942 | spin_unlock(&oi->ip_lock); | |
6943 | ||
5b6a3a2b | 6944 | ocfs2_dinode_new_extent_list(inode, di); |
1afc32b9 MF |
6945 | |
6946 | ocfs2_journal_dirty(handle, di_bh); | |
6947 | ||
6948 | if (has_data) { | |
6949 | /* | |
6950 | * An error at this point should be extremely rare. If | |
6951 | * this proves to be false, we could always re-build | |
6952 | * the in-inode data from our pages. | |
6953 | */ | |
5e404e9e | 6954 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh); |
cc79d8c1 | 6955 | ret = ocfs2_insert_extent(handle, &et, 0, block, 1, 0, NULL); |
1afc32b9 MF |
6956 | if (ret) { |
6957 | mlog_errno(ret); | |
6958 | goto out_commit; | |
6959 | } | |
6960 | ||
6961 | inode->i_blocks = ocfs2_inode_sector_count(inode); | |
6962 | } | |
6963 | ||
6964 | out_commit: | |
a90714c1 | 6965 | if (ret < 0 && did_quota) |
5dd4056d | 6966 | dquot_free_space_nodirty(inode, |
a90714c1 JK |
6967 | ocfs2_clusters_to_bytes(osb->sb, 1)); |
6968 | ||
1afc32b9 MF |
6969 | ocfs2_commit_trans(osb, handle); |
6970 | ||
6971 | out_unlock: | |
6972 | if (data_ac) | |
6973 | ocfs2_free_alloc_context(data_ac); | |
6974 | ||
6975 | out: | |
6976 | if (pages) { | |
6977 | ocfs2_unlock_and_free_pages(pages, num_pages); | |
6978 | kfree(pages); | |
6979 | } | |
6980 | ||
6981 | return ret; | |
6982 | } | |
6983 | ||
ccd979bd MF |
6984 | /* |
6985 | * It is expected, that by the time you call this function, | |
6986 | * inode->i_size and fe->i_size have been adjusted. | |
6987 | * | |
6988 | * WARNING: This will kfree the truncate context | |
6989 | */ | |
6990 | int ocfs2_commit_truncate(struct ocfs2_super *osb, | |
6991 | struct inode *inode, | |
78f94673 | 6992 | struct buffer_head *di_bh) |
ccd979bd | 6993 | { |
78f94673 TY |
6994 | int status = 0, i, flags = 0; |
6995 | u32 new_highest_cpos, range, trunc_cpos, trunc_len, phys_cpos, coff; | |
bcbbb24a | 6996 | u64 blkno = 0; |
ccd979bd | 6997 | struct ocfs2_extent_list *el; |
78f94673 | 6998 | struct ocfs2_extent_rec *rec; |
dcd0538f | 6999 | struct ocfs2_path *path = NULL; |
78f94673 TY |
7000 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; |
7001 | struct ocfs2_extent_list *root_el = &(di->id2.i_list); | |
7002 | u64 refcount_loc = le64_to_cpu(di->i_refcount_loc); | |
7003 | struct ocfs2_extent_tree et; | |
7004 | struct ocfs2_cached_dealloc_ctxt dealloc; | |
ccd979bd MF |
7005 | |
7006 | mlog_entry_void(); | |
7007 | ||
78f94673 TY |
7008 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh); |
7009 | ocfs2_init_dealloc_ctxt(&dealloc); | |
7010 | ||
dcd0538f | 7011 | new_highest_cpos = ocfs2_clusters_for_bytes(osb->sb, |
ccd979bd MF |
7012 | i_size_read(inode)); |
7013 | ||
78f94673 | 7014 | path = ocfs2_new_path(di_bh, &di->id2.i_list, |
13723d00 | 7015 | ocfs2_journal_access_di); |
dcd0538f MF |
7016 | if (!path) { |
7017 | status = -ENOMEM; | |
7018 | mlog_errno(status); | |
7019 | goto bail; | |
7020 | } | |
83418978 MF |
7021 | |
7022 | ocfs2_extent_map_trunc(inode, new_highest_cpos); | |
7023 | ||
ccd979bd | 7024 | start: |
3a0782d0 MF |
7025 | /* |
7026 | * Check that we still have allocation to delete. | |
7027 | */ | |
7028 | if (OCFS2_I(inode)->ip_clusters == 0) { | |
7029 | status = 0; | |
7030 | goto bail; | |
7031 | } | |
7032 | ||
dcd0538f MF |
7033 | /* |
7034 | * Truncate always works against the rightmost tree branch. | |
7035 | */ | |
facdb77f | 7036 | status = ocfs2_find_path(INODE_CACHE(inode), path, UINT_MAX); |
dcd0538f MF |
7037 | if (status) { |
7038 | mlog_errno(status); | |
7039 | goto bail; | |
ccd979bd MF |
7040 | } |
7041 | ||
dcd0538f MF |
7042 | mlog(0, "inode->ip_clusters = %u, tree_depth = %u\n", |
7043 | OCFS2_I(inode)->ip_clusters, path->p_tree_depth); | |
7044 | ||
7045 | /* | |
7046 | * By now, el will point to the extent list on the bottom most | |
7047 | * portion of this tree. Only the tail record is considered in | |
7048 | * each pass. | |
7049 | * | |
7050 | * We handle the following cases, in order: | |
7051 | * - empty extent: delete the remaining branch | |
7052 | * - remove the entire record | |
7053 | * - remove a partial record | |
7054 | * - no record needs to be removed (truncate has completed) | |
7055 | */ | |
7056 | el = path_leaf_el(path); | |
3a0782d0 MF |
7057 | if (le16_to_cpu(el->l_next_free_rec) == 0) { |
7058 | ocfs2_error(inode->i_sb, | |
7059 | "Inode %llu has empty extent block at %llu\n", | |
7060 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
7061 | (unsigned long long)path_leaf_bh(path)->b_blocknr); | |
7062 | status = -EROFS; | |
7063 | goto bail; | |
7064 | } | |
7065 | ||
ccd979bd | 7066 | i = le16_to_cpu(el->l_next_free_rec) - 1; |
78f94673 TY |
7067 | rec = &el->l_recs[i]; |
7068 | flags = rec->e_flags; | |
7069 | range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
7070 | ||
7071 | if (i == 0 && ocfs2_is_empty_extent(rec)) { | |
7072 | /* | |
7073 | * Lower levels depend on this never happening, but it's best | |
7074 | * to check it up here before changing the tree. | |
7075 | */ | |
7076 | if (root_el->l_tree_depth && rec->e_int_clusters == 0) { | |
7077 | ocfs2_error(inode->i_sb, "Inode %lu has an empty " | |
7078 | "extent record, depth %u\n", inode->i_ino, | |
7079 | le16_to_cpu(root_el->l_tree_depth)); | |
7080 | status = -EROFS; | |
7081 | goto bail; | |
7082 | } | |
7083 | trunc_cpos = le32_to_cpu(rec->e_cpos); | |
7084 | trunc_len = 0; | |
7085 | blkno = 0; | |
7086 | } else if (le32_to_cpu(rec->e_cpos) >= new_highest_cpos) { | |
7087 | /* | |
7088 | * Truncate entire record. | |
7089 | */ | |
7090 | trunc_cpos = le32_to_cpu(rec->e_cpos); | |
7091 | trunc_len = ocfs2_rec_clusters(el, rec); | |
7092 | blkno = le64_to_cpu(rec->e_blkno); | |
dcd0538f | 7093 | } else if (range > new_highest_cpos) { |
78f94673 TY |
7094 | /* |
7095 | * Partial truncate. it also should be | |
7096 | * the last truncate we're doing. | |
7097 | */ | |
7098 | trunc_cpos = new_highest_cpos; | |
7099 | trunc_len = range - new_highest_cpos; | |
7100 | coff = new_highest_cpos - le32_to_cpu(rec->e_cpos); | |
7101 | blkno = le64_to_cpu(rec->e_blkno) + | |
7102 | ocfs2_clusters_to_blocks(inode->i_sb, coff); | |
dcd0538f | 7103 | } else { |
78f94673 TY |
7104 | /* |
7105 | * Truncate completed, leave happily. | |
7106 | */ | |
dcd0538f MF |
7107 | status = 0; |
7108 | goto bail; | |
7109 | } | |
ccd979bd | 7110 | |
78f94673 | 7111 | phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno); |
bcbbb24a | 7112 | |
78f94673 TY |
7113 | status = ocfs2_remove_btree_range(inode, &et, trunc_cpos, |
7114 | phys_cpos, trunc_len, flags, &dealloc, | |
7115 | refcount_loc); | |
ccd979bd MF |
7116 | if (status < 0) { |
7117 | mlog_errno(status); | |
7118 | goto bail; | |
7119 | } | |
7120 | ||
dcd0538f MF |
7121 | ocfs2_reinit_path(path, 1); |
7122 | ||
7123 | /* | |
3a0782d0 MF |
7124 | * The check above will catch the case where we've truncated |
7125 | * away all allocation. | |
dcd0538f | 7126 | */ |
3a0782d0 MF |
7127 | goto start; |
7128 | ||
ccd979bd | 7129 | bail: |
ccd979bd MF |
7130 | |
7131 | ocfs2_schedule_truncate_log_flush(osb, 1); | |
7132 | ||
78f94673 | 7133 | ocfs2_run_deallocs(osb, &dealloc); |
59a5e416 | 7134 | |
dcd0538f | 7135 | ocfs2_free_path(path); |
ccd979bd | 7136 | |
ccd979bd MF |
7137 | mlog_exit(status); |
7138 | return status; | |
7139 | } | |
7140 | ||
ccd979bd | 7141 | /* |
59a5e416 | 7142 | * Expects the inode to already be locked. |
ccd979bd MF |
7143 | */ |
7144 | int ocfs2_prepare_truncate(struct ocfs2_super *osb, | |
7145 | struct inode *inode, | |
7146 | struct buffer_head *fe_bh, | |
7147 | struct ocfs2_truncate_context **tc) | |
7148 | { | |
59a5e416 | 7149 | int status; |
ccd979bd MF |
7150 | unsigned int new_i_clusters; |
7151 | struct ocfs2_dinode *fe; | |
7152 | struct ocfs2_extent_block *eb; | |
ccd979bd | 7153 | struct buffer_head *last_eb_bh = NULL; |
ccd979bd MF |
7154 | |
7155 | mlog_entry_void(); | |
7156 | ||
7157 | *tc = NULL; | |
7158 | ||
7159 | new_i_clusters = ocfs2_clusters_for_bytes(osb->sb, | |
7160 | i_size_read(inode)); | |
7161 | fe = (struct ocfs2_dinode *) fe_bh->b_data; | |
7162 | ||
7163 | mlog(0, "fe->i_clusters = %u, new_i_clusters = %u, fe->i_size =" | |
1ca1a111 MF |
7164 | "%llu\n", le32_to_cpu(fe->i_clusters), new_i_clusters, |
7165 | (unsigned long long)le64_to_cpu(fe->i_size)); | |
ccd979bd | 7166 | |
cd861280 | 7167 | *tc = kzalloc(sizeof(struct ocfs2_truncate_context), GFP_KERNEL); |
ccd979bd MF |
7168 | if (!(*tc)) { |
7169 | status = -ENOMEM; | |
7170 | mlog_errno(status); | |
7171 | goto bail; | |
7172 | } | |
59a5e416 | 7173 | ocfs2_init_dealloc_ctxt(&(*tc)->tc_dealloc); |
ccd979bd | 7174 | |
ccd979bd | 7175 | if (fe->id2.i_list.l_tree_depth) { |
3d03a305 | 7176 | status = ocfs2_read_extent_block(INODE_CACHE(inode), |
5e96581a JB |
7177 | le64_to_cpu(fe->i_last_eb_blk), |
7178 | &last_eb_bh); | |
ccd979bd MF |
7179 | if (status < 0) { |
7180 | mlog_errno(status); | |
7181 | goto bail; | |
7182 | } | |
7183 | eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; | |
ccd979bd MF |
7184 | } |
7185 | ||
7186 | (*tc)->tc_last_eb_bh = last_eb_bh; | |
7187 | ||
ccd979bd MF |
7188 | status = 0; |
7189 | bail: | |
7190 | if (status < 0) { | |
7191 | if (*tc) | |
7192 | ocfs2_free_truncate_context(*tc); | |
7193 | *tc = NULL; | |
7194 | } | |
7195 | mlog_exit_void(); | |
7196 | return status; | |
7197 | } | |
7198 | ||
1afc32b9 MF |
7199 | /* |
7200 | * 'start' is inclusive, 'end' is not. | |
7201 | */ | |
7202 | int ocfs2_truncate_inline(struct inode *inode, struct buffer_head *di_bh, | |
7203 | unsigned int start, unsigned int end, int trunc) | |
7204 | { | |
7205 | int ret; | |
7206 | unsigned int numbytes; | |
7207 | handle_t *handle; | |
7208 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
7209 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
7210 | struct ocfs2_inline_data *idata = &di->id2.i_data; | |
7211 | ||
7212 | if (end > i_size_read(inode)) | |
7213 | end = i_size_read(inode); | |
7214 | ||
7215 | BUG_ON(start >= end); | |
7216 | ||
7217 | if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) || | |
7218 | !(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL) || | |
7219 | !ocfs2_supports_inline_data(osb)) { | |
7220 | ocfs2_error(inode->i_sb, | |
7221 | "Inline data flags for inode %llu don't agree! " | |
7222 | "Disk: 0x%x, Memory: 0x%x, Superblock: 0x%x\n", | |
7223 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
7224 | le16_to_cpu(di->i_dyn_features), | |
7225 | OCFS2_I(inode)->ip_dyn_features, | |
7226 | osb->s_feature_incompat); | |
7227 | ret = -EROFS; | |
7228 | goto out; | |
7229 | } | |
7230 | ||
7231 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
7232 | if (IS_ERR(handle)) { | |
7233 | ret = PTR_ERR(handle); | |
7234 | mlog_errno(ret); | |
7235 | goto out; | |
7236 | } | |
7237 | ||
0cf2f763 | 7238 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh, |
13723d00 | 7239 | OCFS2_JOURNAL_ACCESS_WRITE); |
1afc32b9 MF |
7240 | if (ret) { |
7241 | mlog_errno(ret); | |
7242 | goto out_commit; | |
7243 | } | |
7244 | ||
7245 | numbytes = end - start; | |
7246 | memset(idata->id_data + start, 0, numbytes); | |
7247 | ||
7248 | /* | |
7249 | * No need to worry about the data page here - it's been | |
7250 | * truncated already and inline data doesn't need it for | |
7251 | * pushing zero's to disk, so we'll let readpage pick it up | |
7252 | * later. | |
7253 | */ | |
7254 | if (trunc) { | |
7255 | i_size_write(inode, start); | |
7256 | di->i_size = cpu_to_le64(start); | |
7257 | } | |
7258 | ||
7259 | inode->i_blocks = ocfs2_inode_sector_count(inode); | |
7260 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; | |
7261 | ||
7262 | di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec); | |
7263 | di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); | |
7264 | ||
7265 | ocfs2_journal_dirty(handle, di_bh); | |
7266 | ||
7267 | out_commit: | |
7268 | ocfs2_commit_trans(osb, handle); | |
7269 | ||
7270 | out: | |
7271 | return ret; | |
7272 | } | |
7273 | ||
ccd979bd MF |
7274 | static void ocfs2_free_truncate_context(struct ocfs2_truncate_context *tc) |
7275 | { | |
59a5e416 MF |
7276 | /* |
7277 | * The caller is responsible for completing deallocation | |
7278 | * before freeing the context. | |
7279 | */ | |
7280 | if (tc->tc_dealloc.c_first_suballocator != NULL) | |
7281 | mlog(ML_NOTICE, | |
7282 | "Truncate completion has non-empty dealloc context\n"); | |
ccd979bd | 7283 | |
a81cb88b | 7284 | brelse(tc->tc_last_eb_bh); |
ccd979bd MF |
7285 | |
7286 | kfree(tc); | |
7287 | } |