Commit | Line | Data |
---|---|---|
b3b94faa DT |
1 | /* |
2 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
3a8a9a10 | 3 | * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved. |
b3b94faa DT |
4 | * |
5 | * This copyrighted material is made available to anyone wishing to use, | |
6 | * modify, copy, or redistribute it subject to the terms and conditions | |
e9fc2aa0 | 7 | * of the GNU General Public License version 2. |
b3b94faa DT |
8 | */ |
9 | ||
10 | /* | |
61e085a8 SW |
11 | * Implements Extendible Hashing as described in: |
12 | * "Extendible Hashing" by Fagin, et al in | |
13 | * __ACM Trans. on Database Systems__, Sept 1979. | |
14 | * | |
15 | * | |
16 | * Here's the layout of dirents which is essentially the same as that of ext2 | |
17 | * within a single block. The field de_name_len is the number of bytes | |
18 | * actually required for the name (no null terminator). The field de_rec_len | |
19 | * is the number of bytes allocated to the dirent. The offset of the next | |
20 | * dirent in the block is (dirent + dirent->de_rec_len). When a dirent is | |
21 | * deleted, the preceding dirent inherits its allocated space, ie | |
22 | * prev->de_rec_len += deleted->de_rec_len. Since the next dirent is obtained | |
23 | * by adding de_rec_len to the current dirent, this essentially causes the | |
24 | * deleted dirent to get jumped over when iterating through all the dirents. | |
25 | * | |
26 | * When deleting the first dirent in a block, there is no previous dirent so | |
27 | * the field de_ino is set to zero to designate it as deleted. When allocating | |
28 | * a dirent, gfs2_dirent_alloc iterates through the dirents in a block. If the | |
29 | * first dirent has (de_ino == 0) and de_rec_len is large enough, this first | |
30 | * dirent is allocated. Otherwise it must go through all the 'used' dirents | |
31 | * searching for one in which the amount of total space minus the amount of | |
32 | * used space will provide enough space for the new dirent. | |
33 | * | |
34 | * There are two types of blocks in which dirents reside. In a stuffed dinode, | |
35 | * the dirents begin at offset sizeof(struct gfs2_dinode) from the beginning of | |
36 | * the block. In leaves, they begin at offset sizeof(struct gfs2_leaf) from the | |
37 | * beginning of the leaf block. The dirents reside in leaves when | |
38 | * | |
383f01fb | 39 | * dip->i_diskflags & GFS2_DIF_EXHASH is true |
61e085a8 SW |
40 | * |
41 | * Otherwise, the dirents are "linear", within a single stuffed dinode block. | |
42 | * | |
43 | * When the dirents are in leaves, the actual contents of the directory file are | |
44 | * used as an array of 64-bit block pointers pointing to the leaf blocks. The | |
45 | * dirents are NOT in the directory file itself. There can be more than one | |
46 | * block pointer in the array that points to the same leaf. In fact, when a | |
47 | * directory is first converted from linear to exhash, all of the pointers | |
48 | * point to the same leaf. | |
49 | * | |
50 | * When a leaf is completely full, the size of the hash table can be | |
51 | * doubled unless it is already at the maximum size which is hard coded into | |
52 | * GFS2_DIR_MAX_DEPTH. After that, leaves are chained together in a linked list, | |
53 | * but never before the maximum hash table size has been reached. | |
54 | */ | |
b3b94faa | 55 | |
d77d1b58 JP |
56 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
57 | ||
b3b94faa DT |
58 | #include <linux/slab.h> |
59 | #include <linux/spinlock.h> | |
b3b94faa DT |
60 | #include <linux/buffer_head.h> |
61 | #include <linux/sort.h> | |
5c676f6d | 62 | #include <linux/gfs2_ondisk.h> |
71b86f56 | 63 | #include <linux/crc32.h> |
fe1bdedc | 64 | #include <linux/vmalloc.h> |
b3b94faa DT |
65 | |
66 | #include "gfs2.h" | |
5c676f6d | 67 | #include "incore.h" |
b3b94faa DT |
68 | #include "dir.h" |
69 | #include "glock.h" | |
70 | #include "inode.h" | |
b3b94faa DT |
71 | #include "meta_io.h" |
72 | #include "quota.h" | |
73 | #include "rgrp.h" | |
74 | #include "trans.h" | |
e13940ba | 75 | #include "bmap.h" |
5c676f6d | 76 | #include "util.h" |
b3b94faa DT |
77 | |
78 | #define IS_LEAF 1 /* Hashed (leaf) directory */ | |
79 | #define IS_DINODE 2 /* Linear (stuffed dinode block) directory */ | |
80 | ||
dfe4d34b BP |
81 | #define MAX_RA_BLOCKS 32 /* max read-ahead blocks */ |
82 | ||
cd915493 SW |
83 | #define gfs2_disk_hash2offset(h) (((u64)(h)) >> 1) |
84 | #define gfs2_dir_offset2hash(p) ((u32)(((u64)(p)) << 1)) | |
b3b94faa | 85 | |
8d123585 SW |
86 | struct qstr gfs2_qdot __read_mostly; |
87 | struct qstr gfs2_qdotdot __read_mostly; | |
88 | ||
2bdbc5d7 SW |
89 | typedef int (*gfs2_dscan_t)(const struct gfs2_dirent *dent, |
90 | const struct qstr *name, void *opaque); | |
b3b94faa | 91 | |
cd915493 | 92 | int gfs2_dir_get_new_buffer(struct gfs2_inode *ip, u64 block, |
61e085a8 | 93 | struct buffer_head **bhp) |
e13940ba SW |
94 | { |
95 | struct buffer_head *bh; | |
e13940ba | 96 | |
61e085a8 | 97 | bh = gfs2_meta_new(ip->i_gl, block); |
350a9b0a | 98 | gfs2_trans_add_meta(ip->i_gl, bh); |
61e085a8 SW |
99 | gfs2_metatype_set(bh, GFS2_METATYPE_JD, GFS2_FORMAT_JD); |
100 | gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header)); | |
e13940ba SW |
101 | *bhp = bh; |
102 | return 0; | |
103 | } | |
104 | ||
cd915493 | 105 | static int gfs2_dir_get_existing_buffer(struct gfs2_inode *ip, u64 block, |
61e085a8 SW |
106 | struct buffer_head **bhp) |
107 | { | |
108 | struct buffer_head *bh; | |
109 | int error; | |
e13940ba | 110 | |
7276b3b0 | 111 | error = gfs2_meta_read(ip->i_gl, block, DIO_WAIT, &bh); |
61e085a8 SW |
112 | if (error) |
113 | return error; | |
feaa7bba | 114 | if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), bh, GFS2_METATYPE_JD)) { |
61e085a8 SW |
115 | brelse(bh); |
116 | return -EIO; | |
117 | } | |
118 | *bhp = bh; | |
119 | return 0; | |
120 | } | |
e13940ba SW |
121 | |
122 | static int gfs2_dir_write_stuffed(struct gfs2_inode *ip, const char *buf, | |
123 | unsigned int offset, unsigned int size) | |
e13940ba SW |
124 | { |
125 | struct buffer_head *dibh; | |
126 | int error; | |
127 | ||
128 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
129 | if (error) | |
130 | return error; | |
131 | ||
350a9b0a | 132 | gfs2_trans_add_meta(ip->i_gl, dibh); |
c752666c | 133 | memcpy(dibh->b_data + offset + sizeof(struct gfs2_dinode), buf, size); |
a2e0f799 SW |
134 | if (ip->i_inode.i_size < offset + size) |
135 | i_size_write(&ip->i_inode, offset + size); | |
4bd91ba1 | 136 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME; |
539e5d6b | 137 | gfs2_dinode_out(ip, dibh->b_data); |
e13940ba SW |
138 | |
139 | brelse(dibh); | |
140 | ||
141 | return size; | |
142 | } | |
143 | ||
144 | ||
145 | ||
146 | /** | |
147 | * gfs2_dir_write_data - Write directory information to the inode | |
148 | * @ip: The GFS2 inode | |
149 | * @buf: The buffer containing information to be written | |
150 | * @offset: The file offset to start writing at | |
151 | * @size: The amount of data to write | |
152 | * | |
153 | * Returns: The number of bytes correctly written or error code | |
154 | */ | |
155 | static int gfs2_dir_write_data(struct gfs2_inode *ip, const char *buf, | |
cd915493 | 156 | u64 offset, unsigned int size) |
e13940ba | 157 | { |
feaa7bba | 158 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
e13940ba | 159 | struct buffer_head *dibh; |
cd915493 SW |
160 | u64 lblock, dblock; |
161 | u32 extlen = 0; | |
e13940ba SW |
162 | unsigned int o; |
163 | int copied = 0; | |
164 | int error = 0; | |
9b8c81d1 | 165 | int new = 0; |
e13940ba SW |
166 | |
167 | if (!size) | |
168 | return 0; | |
169 | ||
170 | if (gfs2_is_stuffed(ip) && | |
171 | offset + size <= sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) | |
568f4c96 SW |
172 | return gfs2_dir_write_stuffed(ip, buf, (unsigned int)offset, |
173 | size); | |
e13940ba SW |
174 | |
175 | if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip))) | |
176 | return -EINVAL; | |
177 | ||
178 | if (gfs2_is_stuffed(ip)) { | |
f25ef0c1 | 179 | error = gfs2_unstuff_dinode(ip, NULL); |
e13940ba | 180 | if (error) |
c752666c | 181 | return error; |
e13940ba SW |
182 | } |
183 | ||
184 | lblock = offset; | |
185 | o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header); | |
186 | ||
187 | while (copied < size) { | |
188 | unsigned int amount; | |
189 | struct buffer_head *bh; | |
e13940ba SW |
190 | |
191 | amount = size - copied; | |
192 | if (amount > sdp->sd_sb.sb_bsize - o) | |
193 | amount = sdp->sd_sb.sb_bsize - o; | |
194 | ||
195 | if (!extlen) { | |
196 | new = 1; | |
feaa7bba | 197 | error = gfs2_extent_map(&ip->i_inode, lblock, &new, |
fd88de56 | 198 | &dblock, &extlen); |
e13940ba SW |
199 | if (error) |
200 | goto fail; | |
201 | error = -EIO; | |
202 | if (gfs2_assert_withdraw(sdp, dblock)) | |
203 | goto fail; | |
204 | } | |
205 | ||
61e085a8 SW |
206 | if (amount == sdp->sd_jbsize || new) |
207 | error = gfs2_dir_get_new_buffer(ip, dblock, &bh); | |
208 | else | |
209 | error = gfs2_dir_get_existing_buffer(ip, dblock, &bh); | |
210 | ||
e13940ba SW |
211 | if (error) |
212 | goto fail; | |
213 | ||
350a9b0a | 214 | gfs2_trans_add_meta(ip->i_gl, bh); |
e13940ba SW |
215 | memcpy(bh->b_data + o, buf, amount); |
216 | brelse(bh); | |
e13940ba | 217 | |
899bb264 | 218 | buf += amount; |
e13940ba SW |
219 | copied += amount; |
220 | lblock++; | |
221 | dblock++; | |
222 | extlen--; | |
223 | ||
224 | o = sizeof(struct gfs2_meta_header); | |
225 | } | |
226 | ||
227 | out: | |
228 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
229 | if (error) | |
230 | return error; | |
231 | ||
a2e0f799 SW |
232 | if (ip->i_inode.i_size < offset + copied) |
233 | i_size_write(&ip->i_inode, offset + copied); | |
4bd91ba1 | 234 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME; |
e13940ba | 235 | |
350a9b0a | 236 | gfs2_trans_add_meta(ip->i_gl, dibh); |
539e5d6b | 237 | gfs2_dinode_out(ip, dibh->b_data); |
e13940ba SW |
238 | brelse(dibh); |
239 | ||
240 | return copied; | |
241 | fail: | |
242 | if (copied) | |
243 | goto out; | |
244 | return error; | |
245 | } | |
246 | ||
4c28d338 SW |
247 | static int gfs2_dir_read_stuffed(struct gfs2_inode *ip, __be64 *buf, |
248 | unsigned int size) | |
e13940ba SW |
249 | { |
250 | struct buffer_head *dibh; | |
251 | int error; | |
252 | ||
253 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
254 | if (!error) { | |
4c28d338 | 255 | memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size); |
e13940ba SW |
256 | brelse(dibh); |
257 | } | |
258 | ||
259 | return (error) ? error : size; | |
260 | } | |
261 | ||
262 | ||
263 | /** | |
264 | * gfs2_dir_read_data - Read a data from a directory inode | |
265 | * @ip: The GFS2 Inode | |
266 | * @buf: The buffer to place result into | |
e13940ba SW |
267 | * @size: Amount of data to transfer |
268 | * | |
269 | * Returns: The amount of data actually copied or the error | |
270 | */ | |
4c28d338 SW |
271 | static int gfs2_dir_read_data(struct gfs2_inode *ip, __be64 *buf, |
272 | unsigned int size) | |
e13940ba | 273 | { |
feaa7bba | 274 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
cd915493 SW |
275 | u64 lblock, dblock; |
276 | u32 extlen = 0; | |
e13940ba SW |
277 | unsigned int o; |
278 | int copied = 0; | |
279 | int error = 0; | |
e13940ba SW |
280 | |
281 | if (gfs2_is_stuffed(ip)) | |
4c28d338 | 282 | return gfs2_dir_read_stuffed(ip, buf, size); |
e13940ba SW |
283 | |
284 | if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip))) | |
285 | return -EINVAL; | |
286 | ||
4c28d338 | 287 | lblock = 0; |
e13940ba SW |
288 | o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header); |
289 | ||
290 | while (copied < size) { | |
291 | unsigned int amount; | |
292 | struct buffer_head *bh; | |
293 | int new; | |
294 | ||
295 | amount = size - copied; | |
296 | if (amount > sdp->sd_sb.sb_bsize - o) | |
297 | amount = sdp->sd_sb.sb_bsize - o; | |
298 | ||
299 | if (!extlen) { | |
300 | new = 0; | |
feaa7bba | 301 | error = gfs2_extent_map(&ip->i_inode, lblock, &new, |
fd88de56 | 302 | &dblock, &extlen); |
7276b3b0 | 303 | if (error || !dblock) |
e13940ba | 304 | goto fail; |
7276b3b0 | 305 | BUG_ON(extlen < 1); |
7276b3b0 | 306 | bh = gfs2_meta_ra(ip->i_gl, dblock, extlen); |
b7d8ac3e | 307 | } else { |
7276b3b0 | 308 | error = gfs2_meta_read(ip->i_gl, dblock, DIO_WAIT, &bh); |
e13940ba SW |
309 | if (error) |
310 | goto fail; | |
7276b3b0 SW |
311 | } |
312 | error = gfs2_metatype_check(sdp, bh, GFS2_METATYPE_JD); | |
313 | if (error) { | |
314 | brelse(bh); | |
315 | goto fail; | |
316 | } | |
317 | dblock++; | |
318 | extlen--; | |
e13940ba SW |
319 | memcpy(buf, bh->b_data + o, amount); |
320 | brelse(bh); | |
4c28d338 | 321 | buf += (amount/sizeof(__be64)); |
e13940ba SW |
322 | copied += amount; |
323 | lblock++; | |
e13940ba SW |
324 | o = sizeof(struct gfs2_meta_header); |
325 | } | |
326 | ||
327 | return copied; | |
328 | fail: | |
329 | return (copied) ? copied : error; | |
330 | } | |
331 | ||
17d539f0 SW |
332 | /** |
333 | * gfs2_dir_get_hash_table - Get pointer to the dir hash table | |
334 | * @ip: The inode in question | |
335 | * | |
336 | * Returns: The hash table or an error | |
337 | */ | |
338 | ||
339 | static __be64 *gfs2_dir_get_hash_table(struct gfs2_inode *ip) | |
340 | { | |
341 | struct inode *inode = &ip->i_inode; | |
342 | int ret; | |
343 | u32 hsize; | |
344 | __be64 *hc; | |
345 | ||
346 | BUG_ON(!(ip->i_diskflags & GFS2_DIF_EXHASH)); | |
347 | ||
348 | hc = ip->i_hash_cache; | |
349 | if (hc) | |
350 | return hc; | |
351 | ||
352 | hsize = 1 << ip->i_depth; | |
353 | hsize *= sizeof(__be64); | |
354 | if (hsize != i_size_read(&ip->i_inode)) { | |
355 | gfs2_consist_inode(ip); | |
356 | return ERR_PTR(-EIO); | |
357 | } | |
358 | ||
e8830d88 BP |
359 | hc = kmalloc(hsize, GFP_NOFS | __GFP_NOWARN); |
360 | if (hc == NULL) | |
361 | hc = __vmalloc(hsize, GFP_NOFS, PAGE_KERNEL); | |
362 | ||
17d539f0 SW |
363 | if (hc == NULL) |
364 | return ERR_PTR(-ENOMEM); | |
365 | ||
4c28d338 | 366 | ret = gfs2_dir_read_data(ip, hc, hsize); |
17d539f0 | 367 | if (ret < 0) { |
e8830d88 BP |
368 | if (is_vmalloc_addr(hc)) |
369 | vfree(hc); | |
370 | else | |
371 | kfree(hc); | |
17d539f0 SW |
372 | return ERR_PTR(ret); |
373 | } | |
374 | ||
375 | spin_lock(&inode->i_lock); | |
e8830d88 BP |
376 | if (ip->i_hash_cache) { |
377 | if (is_vmalloc_addr(hc)) | |
378 | vfree(hc); | |
379 | else | |
380 | kfree(hc); | |
381 | } else { | |
17d539f0 | 382 | ip->i_hash_cache = hc; |
e8830d88 | 383 | } |
17d539f0 SW |
384 | spin_unlock(&inode->i_lock); |
385 | ||
386 | return ip->i_hash_cache; | |
387 | } | |
388 | ||
389 | /** | |
390 | * gfs2_dir_hash_inval - Invalidate dir hash | |
391 | * @ip: The directory inode | |
392 | * | |
393 | * Must be called with an exclusive glock, or during glock invalidation. | |
394 | */ | |
395 | void gfs2_dir_hash_inval(struct gfs2_inode *ip) | |
396 | { | |
397 | __be64 *hc = ip->i_hash_cache; | |
398 | ip->i_hash_cache = NULL; | |
e8830d88 BP |
399 | if (is_vmalloc_addr(hc)) |
400 | vfree(hc); | |
401 | else | |
402 | kfree(hc); | |
17d539f0 SW |
403 | } |
404 | ||
5e7d65cd SW |
405 | static inline int gfs2_dirent_sentinel(const struct gfs2_dirent *dent) |
406 | { | |
407 | return dent->de_inum.no_addr == 0 || dent->de_inum.no_formal_ino == 0; | |
408 | } | |
409 | ||
c752666c SW |
410 | static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent, |
411 | const struct qstr *name, int ret) | |
412 | { | |
5e7d65cd | 413 | if (!gfs2_dirent_sentinel(dent) && |
c752666c SW |
414 | be32_to_cpu(dent->de_hash) == name->hash && |
415 | be16_to_cpu(dent->de_name_len) == name->len && | |
2bdbc5d7 | 416 | memcmp(dent+1, name->name, name->len) == 0) |
c752666c SW |
417 | return ret; |
418 | return 0; | |
419 | } | |
420 | ||
421 | static int gfs2_dirent_find(const struct gfs2_dirent *dent, | |
71b86f56 SW |
422 | const struct qstr *name, |
423 | void *opaque) | |
c752666c SW |
424 | { |
425 | return __gfs2_dirent_find(dent, name, 1); | |
426 | } | |
427 | ||
428 | static int gfs2_dirent_prev(const struct gfs2_dirent *dent, | |
71b86f56 SW |
429 | const struct qstr *name, |
430 | void *opaque) | |
c752666c SW |
431 | { |
432 | return __gfs2_dirent_find(dent, name, 2); | |
433 | } | |
434 | ||
435 | /* | |
436 | * name->name holds ptr to start of block. | |
437 | * name->len holds size of block. | |
b3b94faa | 438 | */ |
c752666c | 439 | static int gfs2_dirent_last(const struct gfs2_dirent *dent, |
71b86f56 SW |
440 | const struct qstr *name, |
441 | void *opaque) | |
c752666c SW |
442 | { |
443 | const char *start = name->name; | |
444 | const char *end = (const char *)dent + be16_to_cpu(dent->de_rec_len); | |
445 | if (name->len == (end - start)) | |
446 | return 1; | |
447 | return 0; | |
448 | } | |
b3b94faa | 449 | |
c752666c | 450 | static int gfs2_dirent_find_space(const struct gfs2_dirent *dent, |
71b86f56 SW |
451 | const struct qstr *name, |
452 | void *opaque) | |
b3b94faa | 453 | { |
c752666c SW |
454 | unsigned required = GFS2_DIRENT_SIZE(name->len); |
455 | unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len)); | |
456 | unsigned totlen = be16_to_cpu(dent->de_rec_len); | |
457 | ||
5e7d65cd | 458 | if (gfs2_dirent_sentinel(dent)) |
728a756b | 459 | actual = 0; |
c5392124 | 460 | if (totlen - actual >= required) |
c752666c SW |
461 | return 1; |
462 | return 0; | |
463 | } | |
464 | ||
71b86f56 SW |
465 | struct dirent_gather { |
466 | const struct gfs2_dirent **pdent; | |
467 | unsigned offset; | |
468 | }; | |
469 | ||
470 | static int gfs2_dirent_gather(const struct gfs2_dirent *dent, | |
471 | const struct qstr *name, | |
472 | void *opaque) | |
473 | { | |
474 | struct dirent_gather *g = opaque; | |
5e7d65cd | 475 | if (!gfs2_dirent_sentinel(dent)) { |
71b86f56 SW |
476 | g->pdent[g->offset++] = dent; |
477 | } | |
478 | return 0; | |
479 | } | |
480 | ||
c752666c SW |
481 | /* |
482 | * Other possible things to check: | |
483 | * - Inode located within filesystem size (and on valid block) | |
484 | * - Valid directory entry type | |
485 | * Not sure how heavy-weight we want to make this... could also check | |
486 | * hash is correct for example, but that would take a lot of extra time. | |
487 | * For now the most important thing is to check that the various sizes | |
488 | * are correct. | |
489 | */ | |
490 | static int gfs2_check_dirent(struct gfs2_dirent *dent, unsigned int offset, | |
491 | unsigned int size, unsigned int len, int first) | |
492 | { | |
493 | const char *msg = "gfs2_dirent too small"; | |
494 | if (unlikely(size < sizeof(struct gfs2_dirent))) | |
495 | goto error; | |
496 | msg = "gfs2_dirent misaligned"; | |
497 | if (unlikely(offset & 0x7)) | |
498 | goto error; | |
499 | msg = "gfs2_dirent points beyond end of block"; | |
500 | if (unlikely(offset + size > len)) | |
501 | goto error; | |
502 | msg = "zero inode number"; | |
5e7d65cd | 503 | if (unlikely(!first && gfs2_dirent_sentinel(dent))) |
c752666c SW |
504 | goto error; |
505 | msg = "name length is greater than space in dirent"; | |
5e7d65cd | 506 | if (!gfs2_dirent_sentinel(dent) && |
c752666c SW |
507 | unlikely(sizeof(struct gfs2_dirent)+be16_to_cpu(dent->de_name_len) > |
508 | size)) | |
509 | goto error; | |
510 | return 0; | |
511 | error: | |
d77d1b58 JP |
512 | pr_warn("%s: %s (%s)\n", |
513 | __func__, msg, first ? "first in block" : "not first in block"); | |
c752666c | 514 | return -EIO; |
b3b94faa DT |
515 | } |
516 | ||
71b86f56 | 517 | static int gfs2_dirent_offset(const void *buf) |
c752666c | 518 | { |
71b86f56 SW |
519 | const struct gfs2_meta_header *h = buf; |
520 | int offset; | |
c752666c SW |
521 | |
522 | BUG_ON(buf == NULL); | |
c752666c | 523 | |
e3167ded | 524 | switch(be32_to_cpu(h->mh_type)) { |
c752666c SW |
525 | case GFS2_METATYPE_LF: |
526 | offset = sizeof(struct gfs2_leaf); | |
527 | break; | |
528 | case GFS2_METATYPE_DI: | |
529 | offset = sizeof(struct gfs2_dinode); | |
530 | break; | |
531 | default: | |
532 | goto wrong_type; | |
533 | } | |
71b86f56 SW |
534 | return offset; |
535 | wrong_type: | |
d77d1b58 | 536 | pr_warn("%s: wrong block type %u\n", __func__, be32_to_cpu(h->mh_type)); |
71b86f56 SW |
537 | return -1; |
538 | } | |
539 | ||
2bdbc5d7 | 540 | static struct gfs2_dirent *gfs2_dirent_scan(struct inode *inode, void *buf, |
71b86f56 SW |
541 | unsigned int len, gfs2_dscan_t scan, |
542 | const struct qstr *name, | |
543 | void *opaque) | |
544 | { | |
545 | struct gfs2_dirent *dent, *prev; | |
546 | unsigned offset; | |
547 | unsigned size; | |
548 | int ret = 0; | |
c752666c | 549 | |
71b86f56 SW |
550 | ret = gfs2_dirent_offset(buf); |
551 | if (ret < 0) | |
552 | goto consist_inode; | |
553 | ||
554 | offset = ret; | |
c752666c | 555 | prev = NULL; |
2bdbc5d7 | 556 | dent = buf + offset; |
c752666c SW |
557 | size = be16_to_cpu(dent->de_rec_len); |
558 | if (gfs2_check_dirent(dent, offset, size, len, 1)) | |
559 | goto consist_inode; | |
560 | do { | |
71b86f56 | 561 | ret = scan(dent, name, opaque); |
c752666c SW |
562 | if (ret) |
563 | break; | |
564 | offset += size; | |
565 | if (offset == len) | |
566 | break; | |
567 | prev = dent; | |
2bdbc5d7 | 568 | dent = buf + offset; |
c752666c SW |
569 | size = be16_to_cpu(dent->de_rec_len); |
570 | if (gfs2_check_dirent(dent, offset, size, len, 0)) | |
571 | goto consist_inode; | |
572 | } while(1); | |
573 | ||
574 | switch(ret) { | |
575 | case 0: | |
576 | return NULL; | |
577 | case 1: | |
578 | return dent; | |
579 | case 2: | |
580 | return prev ? prev : dent; | |
581 | default: | |
582 | BUG_ON(ret > 0); | |
583 | return ERR_PTR(ret); | |
584 | } | |
585 | ||
c752666c | 586 | consist_inode: |
feaa7bba | 587 | gfs2_consist_inode(GFS2_I(inode)); |
c752666c SW |
588 | return ERR_PTR(-EIO); |
589 | } | |
590 | ||
2bdbc5d7 SW |
591 | static int dirent_check_reclen(struct gfs2_inode *dip, |
592 | const struct gfs2_dirent *d, const void *end_p) | |
593 | { | |
594 | const void *ptr = d; | |
595 | u16 rec_len = be16_to_cpu(d->de_rec_len); | |
596 | ||
597 | if (unlikely(rec_len < sizeof(struct gfs2_dirent))) | |
598 | goto broken; | |
599 | ptr += rec_len; | |
600 | if (ptr < end_p) | |
601 | return rec_len; | |
602 | if (ptr == end_p) | |
603 | return -ENOENT; | |
604 | broken: | |
605 | gfs2_consist_inode(dip); | |
606 | return -EIO; | |
607 | } | |
608 | ||
b3b94faa DT |
609 | /** |
610 | * dirent_next - Next dirent | |
611 | * @dip: the directory | |
612 | * @bh: The buffer | |
613 | * @dent: Pointer to list of dirents | |
614 | * | |
615 | * Returns: 0 on success, error code otherwise | |
616 | */ | |
617 | ||
618 | static int dirent_next(struct gfs2_inode *dip, struct buffer_head *bh, | |
619 | struct gfs2_dirent **dent) | |
620 | { | |
2bdbc5d7 SW |
621 | struct gfs2_dirent *cur = *dent, *tmp; |
622 | char *bh_end = bh->b_data + bh->b_size; | |
623 | int ret; | |
b3b94faa | 624 | |
2bdbc5d7 SW |
625 | ret = dirent_check_reclen(dip, cur, bh_end); |
626 | if (ret < 0) | |
627 | return ret; | |
4dd651ad | 628 | |
2bdbc5d7 SW |
629 | tmp = (void *)cur + ret; |
630 | ret = dirent_check_reclen(dip, tmp, bh_end); | |
631 | if (ret == -EIO) | |
632 | return ret; | |
4dd651ad | 633 | |
b3b94faa | 634 | /* Only the first dent could ever have de_inum.no_addr == 0 */ |
5e7d65cd | 635 | if (gfs2_dirent_sentinel(tmp)) { |
b3b94faa DT |
636 | gfs2_consist_inode(dip); |
637 | return -EIO; | |
638 | } | |
639 | ||
640 | *dent = tmp; | |
b3b94faa DT |
641 | return 0; |
642 | } | |
643 | ||
644 | /** | |
645 | * dirent_del - Delete a dirent | |
646 | * @dip: The GFS2 inode | |
647 | * @bh: The buffer | |
648 | * @prev: The previous dirent | |
649 | * @cur: The current dirent | |
650 | * | |
651 | */ | |
652 | ||
653 | static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh, | |
654 | struct gfs2_dirent *prev, struct gfs2_dirent *cur) | |
655 | { | |
cd915493 | 656 | u16 cur_rec_len, prev_rec_len; |
b3b94faa | 657 | |
5e7d65cd | 658 | if (gfs2_dirent_sentinel(cur)) { |
b3b94faa DT |
659 | gfs2_consist_inode(dip); |
660 | return; | |
661 | } | |
662 | ||
350a9b0a | 663 | gfs2_trans_add_meta(dip->i_gl, bh); |
b3b94faa DT |
664 | |
665 | /* If there is no prev entry, this is the first entry in the block. | |
666 | The de_rec_len is already as big as it needs to be. Just zero | |
667 | out the inode number and return. */ | |
668 | ||
669 | if (!prev) { | |
5e7d65cd SW |
670 | cur->de_inum.no_addr = 0; |
671 | cur->de_inum.no_formal_ino = 0; | |
b3b94faa DT |
672 | return; |
673 | } | |
674 | ||
675 | /* Combine this dentry with the previous one. */ | |
676 | ||
fc69d0d3 SW |
677 | prev_rec_len = be16_to_cpu(prev->de_rec_len); |
678 | cur_rec_len = be16_to_cpu(cur->de_rec_len); | |
b3b94faa DT |
679 | |
680 | if ((char *)prev + prev_rec_len != (char *)cur) | |
681 | gfs2_consist_inode(dip); | |
682 | if ((char *)cur + cur_rec_len > bh->b_data + bh->b_size) | |
683 | gfs2_consist_inode(dip); | |
684 | ||
685 | prev_rec_len += cur_rec_len; | |
fc69d0d3 | 686 | prev->de_rec_len = cpu_to_be16(prev_rec_len); |
b3b94faa DT |
687 | } |
688 | ||
c752666c SW |
689 | /* |
690 | * Takes a dent from which to grab space as an argument. Returns the | |
691 | * newly created dent. | |
b3b94faa | 692 | */ |
08bc2dbc AB |
693 | static struct gfs2_dirent *gfs2_init_dirent(struct inode *inode, |
694 | struct gfs2_dirent *dent, | |
695 | const struct qstr *name, | |
696 | struct buffer_head *bh) | |
b3b94faa | 697 | { |
feaa7bba | 698 | struct gfs2_inode *ip = GFS2_I(inode); |
c752666c SW |
699 | struct gfs2_dirent *ndent; |
700 | unsigned offset = 0, totlen; | |
701 | ||
5e7d65cd | 702 | if (!gfs2_dirent_sentinel(dent)) |
c752666c SW |
703 | offset = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len)); |
704 | totlen = be16_to_cpu(dent->de_rec_len); | |
705 | BUG_ON(offset + name->len > totlen); | |
350a9b0a | 706 | gfs2_trans_add_meta(ip->i_gl, bh); |
c752666c SW |
707 | ndent = (struct gfs2_dirent *)((char *)dent + offset); |
708 | dent->de_rec_len = cpu_to_be16(offset); | |
709 | gfs2_qstr2dirent(name, totlen - offset, ndent); | |
710 | return ndent; | |
b3b94faa DT |
711 | } |
712 | ||
c752666c SW |
713 | static struct gfs2_dirent *gfs2_dirent_alloc(struct inode *inode, |
714 | struct buffer_head *bh, | |
715 | const struct qstr *name) | |
b3b94faa DT |
716 | { |
717 | struct gfs2_dirent *dent; | |
907b9bce | 718 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, |
71b86f56 | 719 | gfs2_dirent_find_space, name, NULL); |
c752666c SW |
720 | if (!dent || IS_ERR(dent)) |
721 | return dent; | |
722 | return gfs2_init_dirent(inode, dent, name, bh); | |
b3b94faa DT |
723 | } |
724 | ||
cd915493 | 725 | static int get_leaf(struct gfs2_inode *dip, u64 leaf_no, |
b3b94faa DT |
726 | struct buffer_head **bhp) |
727 | { | |
728 | int error; | |
729 | ||
7276b3b0 | 730 | error = gfs2_meta_read(dip->i_gl, leaf_no, DIO_WAIT, bhp); |
feaa7bba | 731 | if (!error && gfs2_metatype_check(GFS2_SB(&dip->i_inode), *bhp, GFS2_METATYPE_LF)) { |
d77d1b58 | 732 | /* pr_info("block num=%llu\n", leaf_no); */ |
b3b94faa | 733 | error = -EIO; |
feaa7bba | 734 | } |
b3b94faa DT |
735 | |
736 | return error; | |
737 | } | |
738 | ||
739 | /** | |
740 | * get_leaf_nr - Get a leaf number associated with the index | |
741 | * @dip: The GFS2 inode | |
742 | * @index: | |
743 | * @leaf_out: | |
744 | * | |
745 | * Returns: 0 on success, error code otherwise | |
746 | */ | |
747 | ||
cd915493 SW |
748 | static int get_leaf_nr(struct gfs2_inode *dip, u32 index, |
749 | u64 *leaf_out) | |
b3b94faa | 750 | { |
17d539f0 | 751 | __be64 *hash; |
b3b94faa | 752 | |
17d539f0 SW |
753 | hash = gfs2_dir_get_hash_table(dip); |
754 | if (IS_ERR(hash)) | |
755 | return PTR_ERR(hash); | |
756 | *leaf_out = be64_to_cpu(*(hash + index)); | |
b3b94faa DT |
757 | return 0; |
758 | } | |
759 | ||
cd915493 | 760 | static int get_first_leaf(struct gfs2_inode *dip, u32 index, |
b3b94faa DT |
761 | struct buffer_head **bh_out) |
762 | { | |
cd915493 | 763 | u64 leaf_no; |
b3b94faa DT |
764 | int error; |
765 | ||
766 | error = get_leaf_nr(dip, index, &leaf_no); | |
767 | if (!error) | |
768 | error = get_leaf(dip, leaf_no, bh_out); | |
769 | ||
770 | return error; | |
771 | } | |
772 | ||
c752666c SW |
773 | static struct gfs2_dirent *gfs2_dirent_search(struct inode *inode, |
774 | const struct qstr *name, | |
775 | gfs2_dscan_t scan, | |
776 | struct buffer_head **pbh) | |
b3b94faa | 777 | { |
c752666c SW |
778 | struct buffer_head *bh; |
779 | struct gfs2_dirent *dent; | |
feaa7bba | 780 | struct gfs2_inode *ip = GFS2_I(inode); |
b3b94faa DT |
781 | int error; |
782 | ||
383f01fb | 783 | if (ip->i_diskflags & GFS2_DIF_EXHASH) { |
c752666c | 784 | struct gfs2_leaf *leaf; |
9a004508 | 785 | unsigned hsize = 1 << ip->i_depth; |
c752666c SW |
786 | unsigned index; |
787 | u64 ln; | |
a2e0f799 | 788 | if (hsize * sizeof(u64) != i_size_read(inode)) { |
c752666c SW |
789 | gfs2_consist_inode(ip); |
790 | return ERR_PTR(-EIO); | |
791 | } | |
907b9bce | 792 | |
9a004508 | 793 | index = name->hash >> (32 - ip->i_depth); |
c752666c SW |
794 | error = get_first_leaf(ip, index, &bh); |
795 | if (error) | |
796 | return ERR_PTR(error); | |
797 | do { | |
798 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, | |
71b86f56 | 799 | scan, name, NULL); |
c752666c SW |
800 | if (dent) |
801 | goto got_dent; | |
802 | leaf = (struct gfs2_leaf *)bh->b_data; | |
803 | ln = be64_to_cpu(leaf->lf_next); | |
f4154ea0 | 804 | brelse(bh); |
c752666c SW |
805 | if (!ln) |
806 | break; | |
907b9bce | 807 | |
c752666c SW |
808 | error = get_leaf(ip, ln, &bh); |
809 | } while(!error); | |
b3b94faa | 810 | |
c752666c | 811 | return error ? ERR_PTR(error) : NULL; |
b3b94faa | 812 | } |
b3b94faa | 813 | |
907b9bce | 814 | |
c752666c SW |
815 | error = gfs2_meta_inode_buffer(ip, &bh); |
816 | if (error) | |
817 | return ERR_PTR(error); | |
71b86f56 | 818 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, scan, name, NULL); |
c752666c | 819 | got_dent: |
f4154ea0 | 820 | if (unlikely(dent == NULL || IS_ERR(dent))) { |
ed386507 SW |
821 | brelse(bh); |
822 | bh = NULL; | |
823 | } | |
c752666c SW |
824 | *pbh = bh; |
825 | return dent; | |
826 | } | |
b3b94faa | 827 | |
c752666c SW |
828 | static struct gfs2_leaf *new_leaf(struct inode *inode, struct buffer_head **pbh, u16 depth) |
829 | { | |
feaa7bba | 830 | struct gfs2_inode *ip = GFS2_I(inode); |
b45e41d7 | 831 | unsigned int n = 1; |
09010978 SW |
832 | u64 bn; |
833 | int error; | |
834 | struct buffer_head *bh; | |
c752666c SW |
835 | struct gfs2_leaf *leaf; |
836 | struct gfs2_dirent *dent; | |
26fe5750 | 837 | struct qstr name = { .name = "" }; |
01bcb0de | 838 | struct timespec tv = CURRENT_TIME; |
09010978 | 839 | |
6e87ed0f | 840 | error = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL); |
09010978 SW |
841 | if (error) |
842 | return NULL; | |
843 | bh = gfs2_meta_new(ip->i_gl, bn); | |
c752666c SW |
844 | if (!bh) |
845 | return NULL; | |
09010978 | 846 | |
5731be53 | 847 | gfs2_trans_add_unrevoke(GFS2_SB(inode), bn, 1); |
350a9b0a | 848 | gfs2_trans_add_meta(ip->i_gl, bh); |
c752666c SW |
849 | gfs2_metatype_set(bh, GFS2_METATYPE_LF, GFS2_FORMAT_LF); |
850 | leaf = (struct gfs2_leaf *)bh->b_data; | |
851 | leaf->lf_depth = cpu_to_be16(depth); | |
2bdbc5d7 | 852 | leaf->lf_entries = 0; |
a2d7d021 | 853 | leaf->lf_dirent_format = cpu_to_be32(GFS2_FORMAT_DE); |
2bdbc5d7 | 854 | leaf->lf_next = 0; |
01bcb0de SW |
855 | leaf->lf_inode = cpu_to_be64(ip->i_no_addr); |
856 | leaf->lf_dist = cpu_to_be32(1); | |
857 | leaf->lf_nsec = cpu_to_be32(tv.tv_nsec); | |
858 | leaf->lf_sec = cpu_to_be64(tv.tv_sec); | |
859 | memset(leaf->lf_reserved2, 0, sizeof(leaf->lf_reserved2)); | |
c752666c | 860 | dent = (struct gfs2_dirent *)(leaf+1); |
71b86f56 | 861 | gfs2_qstr2dirent(&name, bh->b_size - sizeof(struct gfs2_leaf), dent); |
c752666c SW |
862 | *pbh = bh; |
863 | return leaf; | |
b3b94faa DT |
864 | } |
865 | ||
866 | /** | |
867 | * dir_make_exhash - Convert a stuffed directory into an ExHash directory | |
868 | * @dip: The GFS2 inode | |
869 | * | |
870 | * Returns: 0 on success, error code otherwise | |
871 | */ | |
872 | ||
c752666c | 873 | static int dir_make_exhash(struct inode *inode) |
b3b94faa | 874 | { |
feaa7bba SW |
875 | struct gfs2_inode *dip = GFS2_I(inode); |
876 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
b3b94faa | 877 | struct gfs2_dirent *dent; |
c752666c | 878 | struct qstr args; |
b3b94faa DT |
879 | struct buffer_head *bh, *dibh; |
880 | struct gfs2_leaf *leaf; | |
881 | int y; | |
cd915493 | 882 | u32 x; |
b44b84d7 AV |
883 | __be64 *lp; |
884 | u64 bn; | |
b3b94faa DT |
885 | int error; |
886 | ||
887 | error = gfs2_meta_inode_buffer(dip, &dibh); | |
888 | if (error) | |
889 | return error; | |
890 | ||
b3b94faa DT |
891 | /* Turn over a new leaf */ |
892 | ||
c752666c SW |
893 | leaf = new_leaf(inode, &bh, 0); |
894 | if (!leaf) | |
895 | return -ENOSPC; | |
896 | bn = bh->b_blocknr; | |
b3b94faa | 897 | |
ad6203f2 SW |
898 | gfs2_assert(sdp, dip->i_entries < (1 << 16)); |
899 | leaf->lf_entries = cpu_to_be16(dip->i_entries); | |
b3b94faa DT |
900 | |
901 | /* Copy dirents */ | |
902 | ||
903 | gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_leaf), dibh, | |
904 | sizeof(struct gfs2_dinode)); | |
905 | ||
906 | /* Find last entry */ | |
907 | ||
908 | x = 0; | |
c752666c SW |
909 | args.len = bh->b_size - sizeof(struct gfs2_dinode) + |
910 | sizeof(struct gfs2_leaf); | |
911 | args.name = bh->b_data; | |
feaa7bba | 912 | dent = gfs2_dirent_scan(&dip->i_inode, bh->b_data, bh->b_size, |
71b86f56 | 913 | gfs2_dirent_last, &args, NULL); |
c752666c SW |
914 | if (!dent) { |
915 | brelse(bh); | |
916 | brelse(dibh); | |
917 | return -EIO; | |
918 | } | |
919 | if (IS_ERR(dent)) { | |
920 | brelse(bh); | |
921 | brelse(dibh); | |
922 | return PTR_ERR(dent); | |
b3b94faa | 923 | } |
b3b94faa DT |
924 | |
925 | /* Adjust the last dirent's record length | |
926 | (Remember that dent still points to the last entry.) */ | |
927 | ||
4dd651ad | 928 | dent->de_rec_len = cpu_to_be16(be16_to_cpu(dent->de_rec_len) + |
b3b94faa | 929 | sizeof(struct gfs2_dinode) - |
4dd651ad | 930 | sizeof(struct gfs2_leaf)); |
b3b94faa DT |
931 | |
932 | brelse(bh); | |
933 | ||
934 | /* We're done with the new leaf block, now setup the new | |
935 | hash table. */ | |
936 | ||
350a9b0a | 937 | gfs2_trans_add_meta(dip->i_gl, dibh); |
b3b94faa DT |
938 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); |
939 | ||
b44b84d7 | 940 | lp = (__be64 *)(dibh->b_data + sizeof(struct gfs2_dinode)); |
b3b94faa DT |
941 | |
942 | for (x = sdp->sd_hash_ptrs; x--; lp++) | |
943 | *lp = cpu_to_be64(bn); | |
944 | ||
a2e0f799 | 945 | i_size_write(inode, sdp->sd_sb.sb_bsize / 2); |
77658aad | 946 | gfs2_add_inode_blocks(&dip->i_inode, 1); |
383f01fb | 947 | dip->i_diskflags |= GFS2_DIF_EXHASH; |
b3b94faa DT |
948 | |
949 | for (x = sdp->sd_hash_ptrs, y = -1; x; x >>= 1, y++) ; | |
9a004508 | 950 | dip->i_depth = y; |
b3b94faa | 951 | |
539e5d6b | 952 | gfs2_dinode_out(dip, dibh->b_data); |
b3b94faa DT |
953 | |
954 | brelse(dibh); | |
955 | ||
956 | return 0; | |
957 | } | |
958 | ||
959 | /** | |
960 | * dir_split_leaf - Split a leaf block into two | |
961 | * @dip: The GFS2 inode | |
962 | * @index: | |
963 | * @leaf_no: | |
964 | * | |
965 | * Returns: 0 on success, error code on failure | |
966 | */ | |
967 | ||
c752666c | 968 | static int dir_split_leaf(struct inode *inode, const struct qstr *name) |
b3b94faa | 969 | { |
feaa7bba | 970 | struct gfs2_inode *dip = GFS2_I(inode); |
b3b94faa DT |
971 | struct buffer_head *nbh, *obh, *dibh; |
972 | struct gfs2_leaf *nleaf, *oleaf; | |
4da3c646 | 973 | struct gfs2_dirent *dent = NULL, *prev = NULL, *next = NULL, *new; |
cd915493 | 974 | u32 start, len, half_len, divider; |
b44b84d7 AV |
975 | u64 bn, leaf_no; |
976 | __be64 *lp; | |
cd915493 | 977 | u32 index; |
b3b94faa DT |
978 | int x, moved = 0; |
979 | int error; | |
980 | ||
9a004508 | 981 | index = name->hash >> (32 - dip->i_depth); |
c752666c SW |
982 | error = get_leaf_nr(dip, index, &leaf_no); |
983 | if (error) | |
984 | return error; | |
b3b94faa DT |
985 | |
986 | /* Get the old leaf block */ | |
b3b94faa DT |
987 | error = get_leaf(dip, leaf_no, &obh); |
988 | if (error) | |
e90deff5 | 989 | return error; |
b3b94faa | 990 | |
b3b94faa | 991 | oleaf = (struct gfs2_leaf *)obh->b_data; |
9a004508 | 992 | if (dip->i_depth == be16_to_cpu(oleaf->lf_depth)) { |
e90deff5 SW |
993 | brelse(obh); |
994 | return 1; /* can't split */ | |
995 | } | |
996 | ||
350a9b0a | 997 | gfs2_trans_add_meta(dip->i_gl, obh); |
b3b94faa | 998 | |
c752666c SW |
999 | nleaf = new_leaf(inode, &nbh, be16_to_cpu(oleaf->lf_depth) + 1); |
1000 | if (!nleaf) { | |
1001 | brelse(obh); | |
1002 | return -ENOSPC; | |
1003 | } | |
1004 | bn = nbh->b_blocknr; | |
b3b94faa | 1005 | |
c752666c | 1006 | /* Compute the start and len of leaf pointers in the hash table. */ |
9a004508 | 1007 | len = 1 << (dip->i_depth - be16_to_cpu(oleaf->lf_depth)); |
b3b94faa DT |
1008 | half_len = len >> 1; |
1009 | if (!half_len) { | |
d77d1b58 JP |
1010 | pr_warn("i_depth %u lf_depth %u index %u\n", |
1011 | dip->i_depth, be16_to_cpu(oleaf->lf_depth), index); | |
b3b94faa DT |
1012 | gfs2_consist_inode(dip); |
1013 | error = -EIO; | |
1014 | goto fail_brelse; | |
1015 | } | |
1016 | ||
1017 | start = (index & ~(len - 1)); | |
1018 | ||
1019 | /* Change the pointers. | |
1020 | Don't bother distinguishing stuffed from non-stuffed. | |
1021 | This code is complicated enough already. */ | |
4244b52e DR |
1022 | lp = kmalloc(half_len * sizeof(__be64), GFP_NOFS); |
1023 | if (!lp) { | |
1024 | error = -ENOMEM; | |
1025 | goto fail_brelse; | |
1026 | } | |
1027 | ||
b3b94faa | 1028 | /* Change the pointers */ |
b3b94faa DT |
1029 | for (x = 0; x < half_len; x++) |
1030 | lp[x] = cpu_to_be64(bn); | |
1031 | ||
17d539f0 SW |
1032 | gfs2_dir_hash_inval(dip); |
1033 | ||
cd915493 SW |
1034 | error = gfs2_dir_write_data(dip, (char *)lp, start * sizeof(u64), |
1035 | half_len * sizeof(u64)); | |
1036 | if (error != half_len * sizeof(u64)) { | |
b3b94faa DT |
1037 | if (error >= 0) |
1038 | error = -EIO; | |
1039 | goto fail_lpfree; | |
1040 | } | |
1041 | ||
1042 | kfree(lp); | |
1043 | ||
1044 | /* Compute the divider */ | |
9a004508 | 1045 | divider = (start + half_len) << (32 - dip->i_depth); |
b3b94faa DT |
1046 | |
1047 | /* Copy the entries */ | |
1579343a | 1048 | dent = (struct gfs2_dirent *)(obh->b_data + sizeof(struct gfs2_leaf)); |
b3b94faa DT |
1049 | |
1050 | do { | |
1051 | next = dent; | |
1052 | if (dirent_next(dip, obh, &next)) | |
1053 | next = NULL; | |
1054 | ||
5e7d65cd | 1055 | if (!gfs2_dirent_sentinel(dent) && |
b3b94faa | 1056 | be32_to_cpu(dent->de_hash) < divider) { |
c752666c SW |
1057 | struct qstr str; |
1058 | str.name = (char*)(dent+1); | |
1059 | str.len = be16_to_cpu(dent->de_name_len); | |
1060 | str.hash = be32_to_cpu(dent->de_hash); | |
71b86f56 | 1061 | new = gfs2_dirent_alloc(inode, nbh, &str); |
c752666c SW |
1062 | if (IS_ERR(new)) { |
1063 | error = PTR_ERR(new); | |
1064 | break; | |
1065 | } | |
b3b94faa DT |
1066 | |
1067 | new->de_inum = dent->de_inum; /* No endian worries */ | |
b3b94faa | 1068 | new->de_type = dent->de_type; /* No endian worries */ |
bb16b342 | 1069 | be16_add_cpu(&nleaf->lf_entries, 1); |
b3b94faa DT |
1070 | |
1071 | dirent_del(dip, obh, prev, dent); | |
1072 | ||
1073 | if (!oleaf->lf_entries) | |
1074 | gfs2_consist_inode(dip); | |
bb16b342 | 1075 | be16_add_cpu(&oleaf->lf_entries, -1); |
b3b94faa DT |
1076 | |
1077 | if (!prev) | |
1078 | prev = dent; | |
1079 | ||
1080 | moved = 1; | |
c752666c | 1081 | } else { |
b3b94faa | 1082 | prev = dent; |
c752666c | 1083 | } |
b3b94faa | 1084 | dent = next; |
c752666c | 1085 | } while (dent); |
b3b94faa | 1086 | |
c752666c | 1087 | oleaf->lf_depth = nleaf->lf_depth; |
b3b94faa DT |
1088 | |
1089 | error = gfs2_meta_inode_buffer(dip, &dibh); | |
feaa7bba | 1090 | if (!gfs2_assert_withdraw(GFS2_SB(&dip->i_inode), !error)) { |
350a9b0a | 1091 | gfs2_trans_add_meta(dip->i_gl, dibh); |
77658aad | 1092 | gfs2_add_inode_blocks(&dip->i_inode, 1); |
539e5d6b | 1093 | gfs2_dinode_out(dip, dibh->b_data); |
b3b94faa DT |
1094 | brelse(dibh); |
1095 | } | |
1096 | ||
1097 | brelse(obh); | |
1098 | brelse(nbh); | |
1099 | ||
1100 | return error; | |
1101 | ||
e90deff5 | 1102 | fail_lpfree: |
b3b94faa DT |
1103 | kfree(lp); |
1104 | ||
e90deff5 | 1105 | fail_brelse: |
b3b94faa | 1106 | brelse(obh); |
b3b94faa DT |
1107 | brelse(nbh); |
1108 | return error; | |
1109 | } | |
1110 | ||
1111 | /** | |
1112 | * dir_double_exhash - Double size of ExHash table | |
1113 | * @dip: The GFS2 dinode | |
1114 | * | |
1115 | * Returns: 0 on success, error code on failure | |
1116 | */ | |
1117 | ||
1118 | static int dir_double_exhash(struct gfs2_inode *dip) | |
1119 | { | |
b3b94faa | 1120 | struct buffer_head *dibh; |
cd915493 | 1121 | u32 hsize; |
17d539f0 SW |
1122 | u32 hsize_bytes; |
1123 | __be64 *hc; | |
1124 | __be64 *hc2, *h; | |
b3b94faa DT |
1125 | int x; |
1126 | int error = 0; | |
1127 | ||
9a004508 | 1128 | hsize = 1 << dip->i_depth; |
17d539f0 | 1129 | hsize_bytes = hsize * sizeof(__be64); |
b3b94faa | 1130 | |
17d539f0 SW |
1131 | hc = gfs2_dir_get_hash_table(dip); |
1132 | if (IS_ERR(hc)) | |
1133 | return PTR_ERR(hc); | |
b3b94faa | 1134 | |
512cbf02 | 1135 | hc2 = kmalloc(hsize_bytes * 2, GFP_NOFS | __GFP_NOWARN); |
e8830d88 BP |
1136 | if (hc2 == NULL) |
1137 | hc2 = __vmalloc(hsize_bytes * 2, GFP_NOFS, PAGE_KERNEL); | |
1138 | ||
17d539f0 | 1139 | if (!hc2) |
4244b52e | 1140 | return -ENOMEM; |
b3b94faa | 1141 | |
512cbf02 | 1142 | h = hc2; |
17d539f0 SW |
1143 | error = gfs2_meta_inode_buffer(dip, &dibh); |
1144 | if (error) | |
1145 | goto out_kfree; | |
b3b94faa | 1146 | |
17d539f0 SW |
1147 | for (x = 0; x < hsize; x++) { |
1148 | *h++ = *hc; | |
1149 | *h++ = *hc; | |
1150 | hc++; | |
b3b94faa DT |
1151 | } |
1152 | ||
17d539f0 SW |
1153 | error = gfs2_dir_write_data(dip, (char *)hc2, 0, hsize_bytes * 2); |
1154 | if (error != (hsize_bytes * 2)) | |
1155 | goto fail; | |
b3b94faa | 1156 | |
17d539f0 SW |
1157 | gfs2_dir_hash_inval(dip); |
1158 | dip->i_hash_cache = hc2; | |
1159 | dip->i_depth++; | |
1160 | gfs2_dinode_out(dip, dibh->b_data); | |
1161 | brelse(dibh); | |
1162 | return 0; | |
b3b94faa | 1163 | |
a91ea69f | 1164 | fail: |
17d539f0 SW |
1165 | /* Replace original hash table & size */ |
1166 | gfs2_dir_write_data(dip, (char *)hc, 0, hsize_bytes); | |
1167 | i_size_write(&dip->i_inode, hsize_bytes); | |
1168 | gfs2_dinode_out(dip, dibh->b_data); | |
1169 | brelse(dibh); | |
1170 | out_kfree: | |
e8830d88 BP |
1171 | if (is_vmalloc_addr(hc2)) |
1172 | vfree(hc2); | |
1173 | else | |
1174 | kfree(hc2); | |
b3b94faa DT |
1175 | return error; |
1176 | } | |
1177 | ||
1178 | /** | |
1179 | * compare_dents - compare directory entries by hash value | |
1180 | * @a: first dent | |
1181 | * @b: second dent | |
1182 | * | |
1183 | * When comparing the hash entries of @a to @b: | |
1184 | * gt: returns 1 | |
1185 | * lt: returns -1 | |
1186 | * eq: returns 0 | |
1187 | */ | |
1188 | ||
1189 | static int compare_dents(const void *a, const void *b) | |
1190 | { | |
2bdbc5d7 | 1191 | const struct gfs2_dirent *dent_a, *dent_b; |
cd915493 | 1192 | u32 hash_a, hash_b; |
b3b94faa DT |
1193 | int ret = 0; |
1194 | ||
2bdbc5d7 | 1195 | dent_a = *(const struct gfs2_dirent **)a; |
c752666c | 1196 | hash_a = be32_to_cpu(dent_a->de_hash); |
b3b94faa | 1197 | |
2bdbc5d7 | 1198 | dent_b = *(const struct gfs2_dirent **)b; |
c752666c | 1199 | hash_b = be32_to_cpu(dent_b->de_hash); |
b3b94faa DT |
1200 | |
1201 | if (hash_a > hash_b) | |
1202 | ret = 1; | |
1203 | else if (hash_a < hash_b) | |
1204 | ret = -1; | |
1205 | else { | |
4dd651ad SW |
1206 | unsigned int len_a = be16_to_cpu(dent_a->de_name_len); |
1207 | unsigned int len_b = be16_to_cpu(dent_b->de_name_len); | |
b3b94faa DT |
1208 | |
1209 | if (len_a > len_b) | |
1210 | ret = 1; | |
1211 | else if (len_a < len_b) | |
1212 | ret = -1; | |
1213 | else | |
2bdbc5d7 | 1214 | ret = memcmp(dent_a + 1, dent_b + 1, len_a); |
b3b94faa DT |
1215 | } |
1216 | ||
1217 | return ret; | |
1218 | } | |
1219 | ||
1220 | /** | |
1221 | * do_filldir_main - read out directory entries | |
1222 | * @dip: The GFS2 inode | |
d81a8ef5 | 1223 | * @ctx: what to feed the entries to |
b3b94faa DT |
1224 | * @darr: an array of struct gfs2_dirent pointers to read |
1225 | * @entries: the number of entries in darr | |
1226 | * @copied: pointer to int that's non-zero if a entry has been copied out | |
1227 | * | |
1228 | * Jump through some hoops to make sure that if there are hash collsions, | |
1229 | * they are read out at the beginning of a buffer. We want to minimize | |
1230 | * the possibility that they will fall into different readdir buffers or | |
1231 | * that someone will want to seek to that location. | |
1232 | * | |
d81a8ef5 | 1233 | * Returns: errno, >0 if the actor tells you to stop |
b3b94faa DT |
1234 | */ |
1235 | ||
d81a8ef5 | 1236 | static int do_filldir_main(struct gfs2_inode *dip, struct dir_context *ctx, |
cd915493 | 1237 | const struct gfs2_dirent **darr, u32 entries, |
b3b94faa DT |
1238 | int *copied) |
1239 | { | |
71b86f56 | 1240 | const struct gfs2_dirent *dent, *dent_next; |
cd915493 | 1241 | u64 off, off_next; |
b3b94faa DT |
1242 | unsigned int x, y; |
1243 | int run = 0; | |
b3b94faa DT |
1244 | |
1245 | sort(darr, entries, sizeof(struct gfs2_dirent *), compare_dents, NULL); | |
1246 | ||
1247 | dent_next = darr[0]; | |
1248 | off_next = be32_to_cpu(dent_next->de_hash); | |
1249 | off_next = gfs2_disk_hash2offset(off_next); | |
1250 | ||
1251 | for (x = 0, y = 1; x < entries; x++, y++) { | |
1252 | dent = dent_next; | |
1253 | off = off_next; | |
1254 | ||
1255 | if (y < entries) { | |
1256 | dent_next = darr[y]; | |
1257 | off_next = be32_to_cpu(dent_next->de_hash); | |
1258 | off_next = gfs2_disk_hash2offset(off_next); | |
1259 | ||
d81a8ef5 | 1260 | if (off < ctx->pos) |
b3b94faa | 1261 | continue; |
d81a8ef5 | 1262 | ctx->pos = off; |
b3b94faa DT |
1263 | |
1264 | if (off_next == off) { | |
1265 | if (*copied && !run) | |
1266 | return 1; | |
1267 | run = 1; | |
1268 | } else | |
1269 | run = 0; | |
1270 | } else { | |
d81a8ef5 | 1271 | if (off < ctx->pos) |
b3b94faa | 1272 | continue; |
d81a8ef5 | 1273 | ctx->pos = off; |
b3b94faa DT |
1274 | } |
1275 | ||
d81a8ef5 | 1276 | if (!dir_emit(ctx, (const char *)(dent + 1), |
4dd651ad | 1277 | be16_to_cpu(dent->de_name_len), |
d81a8ef5 AV |
1278 | be64_to_cpu(dent->de_inum.no_addr), |
1279 | be16_to_cpu(dent->de_type))) | |
b3b94faa DT |
1280 | return 1; |
1281 | ||
1282 | *copied = 1; | |
1283 | } | |
1284 | ||
d81a8ef5 | 1285 | /* Increment the ctx->pos by one, so the next time we come into the |
b3b94faa DT |
1286 | do_filldir fxn, we get the next entry instead of the last one in the |
1287 | current leaf */ | |
1288 | ||
d81a8ef5 | 1289 | ctx->pos++; |
b3b94faa DT |
1290 | |
1291 | return 0; | |
1292 | } | |
1293 | ||
d2a97a4e SW |
1294 | static void *gfs2_alloc_sort_buffer(unsigned size) |
1295 | { | |
1296 | void *ptr = NULL; | |
1297 | ||
1298 | if (size < KMALLOC_MAX_SIZE) | |
1299 | ptr = kmalloc(size, GFP_NOFS | __GFP_NOWARN); | |
1300 | if (!ptr) | |
1301 | ptr = __vmalloc(size, GFP_NOFS, PAGE_KERNEL); | |
1302 | return ptr; | |
1303 | } | |
1304 | ||
1305 | static void gfs2_free_sort_buffer(void *ptr) | |
1306 | { | |
1307 | if (is_vmalloc_addr(ptr)) | |
1308 | vfree(ptr); | |
1309 | else | |
1310 | kfree(ptr); | |
1311 | } | |
1312 | ||
d81a8ef5 AV |
1313 | static int gfs2_dir_read_leaf(struct inode *inode, struct dir_context *ctx, |
1314 | int *copied, unsigned *depth, | |
3699e3a4 | 1315 | u64 leaf_no) |
b3b94faa | 1316 | { |
feaa7bba | 1317 | struct gfs2_inode *ip = GFS2_I(inode); |
bdd19a22 | 1318 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
71b86f56 SW |
1319 | struct buffer_head *bh; |
1320 | struct gfs2_leaf *lf; | |
bdd19a22 | 1321 | unsigned entries = 0, entries2 = 0; |
71b86f56 SW |
1322 | unsigned leaves = 0; |
1323 | const struct gfs2_dirent **darr, *dent; | |
1324 | struct dirent_gather g; | |
1325 | struct buffer_head **larr; | |
1326 | int leaf = 0; | |
1327 | int error, i; | |
1328 | u64 lfn = leaf_no; | |
b3b94faa | 1329 | |
b3b94faa | 1330 | do { |
71b86f56 | 1331 | error = get_leaf(ip, lfn, &bh); |
b3b94faa | 1332 | if (error) |
71b86f56 SW |
1333 | goto out; |
1334 | lf = (struct gfs2_leaf *)bh->b_data; | |
1335 | if (leaves == 0) | |
1336 | *depth = be16_to_cpu(lf->lf_depth); | |
1337 | entries += be16_to_cpu(lf->lf_entries); | |
1338 | leaves++; | |
1339 | lfn = be64_to_cpu(lf->lf_next); | |
1340 | brelse(bh); | |
1341 | } while(lfn); | |
b3b94faa DT |
1342 | |
1343 | if (!entries) | |
1344 | return 0; | |
1345 | ||
71b86f56 | 1346 | error = -ENOMEM; |
bdd19a22 SW |
1347 | /* |
1348 | * The extra 99 entries are not normally used, but are a buffer | |
1349 | * zone in case the number of entries in the leaf is corrupt. | |
1350 | * 99 is the maximum number of entries that can fit in a single | |
1351 | * leaf block. | |
1352 | */ | |
d2a97a4e | 1353 | larr = gfs2_alloc_sort_buffer((leaves + entries + 99) * sizeof(void *)); |
71b86f56 SW |
1354 | if (!larr) |
1355 | goto out; | |
1356 | darr = (const struct gfs2_dirent **)(larr + leaves); | |
1357 | g.pdent = darr; | |
1358 | g.offset = 0; | |
1359 | lfn = leaf_no; | |
b3b94faa | 1360 | |
71b86f56 SW |
1361 | do { |
1362 | error = get_leaf(ip, lfn, &bh); | |
b3b94faa | 1363 | if (error) |
d2a97a4e | 1364 | goto out_free; |
71b86f56 SW |
1365 | lf = (struct gfs2_leaf *)bh->b_data; |
1366 | lfn = be64_to_cpu(lf->lf_next); | |
1367 | if (lf->lf_entries) { | |
bdd19a22 | 1368 | entries2 += be16_to_cpu(lf->lf_entries); |
71b86f56 SW |
1369 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, |
1370 | gfs2_dirent_gather, NULL, &g); | |
1371 | error = PTR_ERR(dent); | |
bdd19a22 | 1372 | if (IS_ERR(dent)) |
d2a97a4e | 1373 | goto out_free; |
bdd19a22 | 1374 | if (entries2 != g.offset) { |
f391a4ea | 1375 | fs_warn(sdp, "Number of entries corrupt in dir " |
1376 | "leaf %llu, entries2 (%u) != " | |
1377 | "g.offset (%u)\n", | |
1378 | (unsigned long long)bh->b_blocknr, | |
1379 | entries2, g.offset); | |
bdd19a22 SW |
1380 | |
1381 | error = -EIO; | |
d2a97a4e | 1382 | goto out_free; |
71b86f56 SW |
1383 | } |
1384 | error = 0; | |
1385 | larr[leaf++] = bh; | |
b3b94faa | 1386 | } else { |
71b86f56 | 1387 | brelse(bh); |
b3b94faa | 1388 | } |
71b86f56 | 1389 | } while(lfn); |
b3b94faa | 1390 | |
bdd19a22 | 1391 | BUG_ON(entries2 != entries); |
d81a8ef5 | 1392 | error = do_filldir_main(ip, ctx, darr, entries, copied); |
d2a97a4e | 1393 | out_free: |
71b86f56 SW |
1394 | for(i = 0; i < leaf; i++) |
1395 | brelse(larr[i]); | |
d2a97a4e | 1396 | gfs2_free_sort_buffer(larr); |
71b86f56 | 1397 | out: |
b3b94faa DT |
1398 | return error; |
1399 | } | |
1400 | ||
79c4c379 SW |
1401 | /** |
1402 | * gfs2_dir_readahead - Issue read-ahead requests for leaf blocks. | |
dfe4d34b BP |
1403 | * |
1404 | * Note: we can't calculate each index like dir_e_read can because we don't | |
1405 | * have the leaf, and therefore we don't have the depth, and therefore we | |
1406 | * don't have the length. So we have to just read enough ahead to make up | |
79c4c379 SW |
1407 | * for the loss of information. |
1408 | */ | |
dfe4d34b BP |
1409 | static void gfs2_dir_readahead(struct inode *inode, unsigned hsize, u32 index, |
1410 | struct file_ra_state *f_ra) | |
1411 | { | |
1412 | struct gfs2_inode *ip = GFS2_I(inode); | |
1413 | struct gfs2_glock *gl = ip->i_gl; | |
1414 | struct buffer_head *bh; | |
1415 | u64 blocknr = 0, last; | |
1416 | unsigned count; | |
1417 | ||
1418 | /* First check if we've already read-ahead for the whole range. */ | |
79c4c379 | 1419 | if (index + MAX_RA_BLOCKS < f_ra->start) |
dfe4d34b BP |
1420 | return; |
1421 | ||
1422 | f_ra->start = max((pgoff_t)index, f_ra->start); | |
1423 | for (count = 0; count < MAX_RA_BLOCKS; count++) { | |
1424 | if (f_ra->start >= hsize) /* if exceeded the hash table */ | |
1425 | break; | |
1426 | ||
1427 | last = blocknr; | |
1428 | blocknr = be64_to_cpu(ip->i_hash_cache[f_ra->start]); | |
1429 | f_ra->start++; | |
1430 | if (blocknr == last) | |
1431 | continue; | |
1432 | ||
1433 | bh = gfs2_getbuf(gl, blocknr, 1); | |
1434 | if (trylock_buffer(bh)) { | |
1435 | if (buffer_uptodate(bh)) { | |
1436 | unlock_buffer(bh); | |
1437 | brelse(bh); | |
1438 | continue; | |
1439 | } | |
1440 | bh->b_end_io = end_buffer_read_sync; | |
1441 | submit_bh(READA | REQ_META, bh); | |
1442 | continue; | |
1443 | } | |
1444 | brelse(bh); | |
1445 | } | |
1446 | } | |
17d539f0 | 1447 | |
b3b94faa | 1448 | /** |
c752666c SW |
1449 | * dir_e_read - Reads the entries from a directory into a filldir buffer |
1450 | * @dip: dinode pointer | |
d81a8ef5 | 1451 | * @ctx: actor to feed the entries to |
b3b94faa | 1452 | * |
c752666c | 1453 | * Returns: errno |
b3b94faa DT |
1454 | */ |
1455 | ||
d81a8ef5 AV |
1456 | static int dir_e_read(struct inode *inode, struct dir_context *ctx, |
1457 | struct file_ra_state *f_ra) | |
b3b94faa | 1458 | { |
feaa7bba | 1459 | struct gfs2_inode *dip = GFS2_I(inode); |
cd915493 | 1460 | u32 hsize, len = 0; |
cd915493 | 1461 | u32 hash, index; |
b44b84d7 | 1462 | __be64 *lp; |
c752666c SW |
1463 | int copied = 0; |
1464 | int error = 0; | |
4da3c646 | 1465 | unsigned depth = 0; |
b3b94faa | 1466 | |
9a004508 | 1467 | hsize = 1 << dip->i_depth; |
d81a8ef5 | 1468 | hash = gfs2_dir_offset2hash(ctx->pos); |
9a004508 | 1469 | index = hash >> (32 - dip->i_depth); |
b3b94faa | 1470 | |
79c4c379 | 1471 | if (dip->i_hash_cache == NULL) |
dfe4d34b | 1472 | f_ra->start = 0; |
17d539f0 SW |
1473 | lp = gfs2_dir_get_hash_table(dip); |
1474 | if (IS_ERR(lp)) | |
1475 | return PTR_ERR(lp); | |
b3b94faa | 1476 | |
dfe4d34b BP |
1477 | gfs2_dir_readahead(inode, hsize, index, f_ra); |
1478 | ||
b3b94faa | 1479 | while (index < hsize) { |
d81a8ef5 | 1480 | error = gfs2_dir_read_leaf(inode, ctx, |
71b86f56 | 1481 | &copied, &depth, |
17d539f0 | 1482 | be64_to_cpu(lp[index])); |
b3b94faa | 1483 | if (error) |
71b86f56 | 1484 | break; |
b3b94faa | 1485 | |
9a004508 | 1486 | len = 1 << (dip->i_depth - depth); |
b3b94faa DT |
1487 | index = (index & ~(len - 1)) + len; |
1488 | } | |
1489 | ||
71b86f56 SW |
1490 | if (error > 0) |
1491 | error = 0; | |
b3b94faa DT |
1492 | return error; |
1493 | } | |
1494 | ||
d81a8ef5 AV |
1495 | int gfs2_dir_read(struct inode *inode, struct dir_context *ctx, |
1496 | struct file_ra_state *f_ra) | |
b3b94faa | 1497 | { |
feaa7bba | 1498 | struct gfs2_inode *dip = GFS2_I(inode); |
bdd19a22 | 1499 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
71b86f56 SW |
1500 | struct dirent_gather g; |
1501 | const struct gfs2_dirent **darr, *dent; | |
b3b94faa DT |
1502 | struct buffer_head *dibh; |
1503 | int copied = 0; | |
1504 | int error; | |
1505 | ||
ad6203f2 | 1506 | if (!dip->i_entries) |
71b86f56 SW |
1507 | return 0; |
1508 | ||
383f01fb | 1509 | if (dip->i_diskflags & GFS2_DIF_EXHASH) |
d81a8ef5 | 1510 | return dir_e_read(inode, ctx, f_ra); |
71b86f56 | 1511 | |
b3b94faa DT |
1512 | if (!gfs2_is_stuffed(dip)) { |
1513 | gfs2_consist_inode(dip); | |
1514 | return -EIO; | |
1515 | } | |
1516 | ||
b3b94faa DT |
1517 | error = gfs2_meta_inode_buffer(dip, &dibh); |
1518 | if (error) | |
1519 | return error; | |
1520 | ||
71b86f56 | 1521 | error = -ENOMEM; |
bdd19a22 | 1522 | /* 96 is max number of dirents which can be stuffed into an inode */ |
16c5f06f | 1523 | darr = kmalloc(96 * sizeof(struct gfs2_dirent *), GFP_NOFS); |
71b86f56 SW |
1524 | if (darr) { |
1525 | g.pdent = darr; | |
1526 | g.offset = 0; | |
1527 | dent = gfs2_dirent_scan(inode, dibh->b_data, dibh->b_size, | |
1528 | gfs2_dirent_gather, NULL, &g); | |
1529 | if (IS_ERR(dent)) { | |
1530 | error = PTR_ERR(dent); | |
1531 | goto out; | |
1532 | } | |
ad6203f2 | 1533 | if (dip->i_entries != g.offset) { |
bdd19a22 | 1534 | fs_warn(sdp, "Number of entries corrupt in dir %llu, " |
ad6203f2 | 1535 | "ip->i_entries (%u) != g.offset (%u)\n", |
dbb7cae2 | 1536 | (unsigned long long)dip->i_no_addr, |
ad6203f2 | 1537 | dip->i_entries, |
bdd19a22 SW |
1538 | g.offset); |
1539 | error = -EIO; | |
1540 | goto out; | |
1541 | } | |
d81a8ef5 | 1542 | error = do_filldir_main(dip, ctx, darr, |
ad6203f2 | 1543 | dip->i_entries, &copied); |
71b86f56 SW |
1544 | out: |
1545 | kfree(darr); | |
1546 | } | |
1547 | ||
b3b94faa DT |
1548 | if (error > 0) |
1549 | error = 0; | |
1550 | ||
1551 | brelse(dibh); | |
1552 | ||
1553 | return error; | |
1554 | } | |
1555 | ||
b3b94faa DT |
1556 | /** |
1557 | * gfs2_dir_search - Search a directory | |
5a00f3cc SW |
1558 | * @dip: The GFS2 dir inode |
1559 | * @name: The name we are looking up | |
1560 | * @fail_on_exist: Fail if the name exists rather than looking it up | |
b3b94faa DT |
1561 | * |
1562 | * This routine searches a directory for a file or another directory. | |
1563 | * Assumes a glock is held on dip. | |
1564 | * | |
1565 | * Returns: errno | |
1566 | */ | |
1567 | ||
5a00f3cc SW |
1568 | struct inode *gfs2_dir_search(struct inode *dir, const struct qstr *name, |
1569 | bool fail_on_exist) | |
b3b94faa | 1570 | { |
c752666c SW |
1571 | struct buffer_head *bh; |
1572 | struct gfs2_dirent *dent; | |
5a00f3cc SW |
1573 | u64 addr, formal_ino; |
1574 | u16 dtype; | |
dbb7cae2 SW |
1575 | |
1576 | dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh); | |
1577 | if (dent) { | |
1578 | if (IS_ERR(dent)) | |
e231c2ee | 1579 | return ERR_CAST(dent); |
5a00f3cc SW |
1580 | dtype = be16_to_cpu(dent->de_type); |
1581 | addr = be64_to_cpu(dent->de_inum.no_addr); | |
1582 | formal_ino = be64_to_cpu(dent->de_inum.no_formal_ino); | |
dbb7cae2 | 1583 | brelse(bh); |
5a00f3cc SW |
1584 | if (fail_on_exist) |
1585 | return ERR_PTR(-EEXIST); | |
1586 | return gfs2_inode_lookup(dir->i_sb, dtype, addr, formal_ino, 0); | |
dbb7cae2 SW |
1587 | } |
1588 | return ERR_PTR(-ENOENT); | |
1589 | } | |
1590 | ||
1591 | int gfs2_dir_check(struct inode *dir, const struct qstr *name, | |
1592 | const struct gfs2_inode *ip) | |
1593 | { | |
1594 | struct buffer_head *bh; | |
1595 | struct gfs2_dirent *dent; | |
1596 | int ret = -ENOENT; | |
c752666c SW |
1597 | |
1598 | dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh); | |
1599 | if (dent) { | |
1600 | if (IS_ERR(dent)) | |
1601 | return PTR_ERR(dent); | |
dbb7cae2 SW |
1602 | if (ip) { |
1603 | if (be64_to_cpu(dent->de_inum.no_addr) != ip->i_no_addr) | |
1604 | goto out; | |
1605 | if (be64_to_cpu(dent->de_inum.no_formal_ino) != | |
1606 | ip->i_no_formal_ino) | |
1607 | goto out; | |
1608 | if (unlikely(IF2DT(ip->i_inode.i_mode) != | |
1609 | be16_to_cpu(dent->de_type))) { | |
1610 | gfs2_consist_inode(GFS2_I(dir)); | |
1611 | ret = -EIO; | |
1612 | goto out; | |
1613 | } | |
1614 | } | |
1615 | ret = 0; | |
1616 | out: | |
c752666c | 1617 | brelse(bh); |
c752666c | 1618 | } |
dbb7cae2 | 1619 | return ret; |
c752666c SW |
1620 | } |
1621 | ||
01bcb0de SW |
1622 | /** |
1623 | * dir_new_leaf - Add a new leaf onto hash chain | |
1624 | * @inode: The directory | |
1625 | * @name: The name we are adding | |
1626 | * | |
1627 | * This adds a new dir leaf onto an existing leaf when there is not | |
1628 | * enough space to add a new dir entry. This is a last resort after | |
1629 | * we've expanded the hash table to max size and also split existing | |
1630 | * leaf blocks, so it will only occur for very large directories. | |
1631 | * | |
1632 | * The dist parameter is set to 1 for leaf blocks directly attached | |
1633 | * to the hash table, 2 for one layer of indirection, 3 for two layers | |
1634 | * etc. We are thus able to tell the difference between an old leaf | |
1635 | * with dist set to zero (i.e. "don't know") and a new one where we | |
1636 | * set this information for debug/fsck purposes. | |
1637 | * | |
1638 | * Returns: 0 on success, or -ve on error | |
1639 | */ | |
1640 | ||
c752666c SW |
1641 | static int dir_new_leaf(struct inode *inode, const struct qstr *name) |
1642 | { | |
1643 | struct buffer_head *bh, *obh; | |
feaa7bba | 1644 | struct gfs2_inode *ip = GFS2_I(inode); |
c752666c | 1645 | struct gfs2_leaf *leaf, *oleaf; |
01bcb0de | 1646 | u32 dist = 1; |
b3b94faa | 1647 | int error; |
c752666c SW |
1648 | u32 index; |
1649 | u64 bn; | |
b3b94faa | 1650 | |
9a004508 | 1651 | index = name->hash >> (32 - ip->i_depth); |
c752666c SW |
1652 | error = get_first_leaf(ip, index, &obh); |
1653 | if (error) | |
1654 | return error; | |
1655 | do { | |
01bcb0de | 1656 | dist++; |
c752666c SW |
1657 | oleaf = (struct gfs2_leaf *)obh->b_data; |
1658 | bn = be64_to_cpu(oleaf->lf_next); | |
1659 | if (!bn) | |
1660 | break; | |
1661 | brelse(obh); | |
1662 | error = get_leaf(ip, bn, &obh); | |
1663 | if (error) | |
1664 | return error; | |
1665 | } while(1); | |
b3b94faa | 1666 | |
350a9b0a | 1667 | gfs2_trans_add_meta(ip->i_gl, obh); |
c752666c SW |
1668 | |
1669 | leaf = new_leaf(inode, &bh, be16_to_cpu(oleaf->lf_depth)); | |
1670 | if (!leaf) { | |
1671 | brelse(obh); | |
1672 | return -ENOSPC; | |
1673 | } | |
01bcb0de | 1674 | leaf->lf_dist = cpu_to_be32(dist); |
4d8012b6 | 1675 | oleaf->lf_next = cpu_to_be64(bh->b_blocknr); |
c752666c SW |
1676 | brelse(bh); |
1677 | brelse(obh); | |
1678 | ||
1679 | error = gfs2_meta_inode_buffer(ip, &bh); | |
1680 | if (error) | |
1681 | return error; | |
350a9b0a | 1682 | gfs2_trans_add_meta(ip->i_gl, bh); |
77658aad | 1683 | gfs2_add_inode_blocks(&ip->i_inode, 1); |
539e5d6b | 1684 | gfs2_dinode_out(ip, bh->b_data); |
c752666c SW |
1685 | brelse(bh); |
1686 | return 0; | |
b3b94faa DT |
1687 | } |
1688 | ||
44aaada9 SW |
1689 | static u16 gfs2_inode_ra_len(const struct gfs2_inode *ip) |
1690 | { | |
1691 | u64 where = ip->i_no_addr + 1; | |
1692 | if (ip->i_eattr == where) | |
1693 | return 1; | |
1694 | return 0; | |
1695 | } | |
1696 | ||
b3b94faa DT |
1697 | /** |
1698 | * gfs2_dir_add - Add new filename into directory | |
2b47dad8 SW |
1699 | * @inode: The directory inode |
1700 | * @name: The new name | |
1701 | * @nip: The GFS2 inode to be linked in to the directory | |
1702 | * @da: The directory addition info | |
1703 | * | |
1704 | * If the call to gfs2_diradd_alloc_required resulted in there being | |
1705 | * no need to allocate any new directory blocks, then it will contain | |
1706 | * a pointer to the directory entry and the bh in which it resides. We | |
1707 | * can use that without having to repeat the search. If there was no | |
1708 | * free space, then we must now create more space. | |
b3b94faa DT |
1709 | * |
1710 | * Returns: 0 on success, error code on failure | |
1711 | */ | |
1712 | ||
c752666c | 1713 | int gfs2_dir_add(struct inode *inode, const struct qstr *name, |
2b47dad8 | 1714 | const struct gfs2_inode *nip, struct gfs2_diradd *da) |
b3b94faa | 1715 | { |
feaa7bba | 1716 | struct gfs2_inode *ip = GFS2_I(inode); |
2b47dad8 SW |
1717 | struct buffer_head *bh = da->bh; |
1718 | struct gfs2_dirent *dent = da->dent; | |
01bcb0de | 1719 | struct timespec tv; |
c752666c | 1720 | struct gfs2_leaf *leaf; |
b3b94faa DT |
1721 | int error; |
1722 | ||
c752666c | 1723 | while(1) { |
2b47dad8 SW |
1724 | if (da->bh == NULL) { |
1725 | dent = gfs2_dirent_search(inode, name, | |
1726 | gfs2_dirent_find_space, &bh); | |
1727 | } | |
c752666c SW |
1728 | if (dent) { |
1729 | if (IS_ERR(dent)) | |
1730 | return PTR_ERR(dent); | |
1731 | dent = gfs2_init_dirent(inode, dent, name, bh); | |
dbb7cae2 | 1732 | gfs2_inum_out(nip, dent); |
3d6ecb7d | 1733 | dent->de_type = cpu_to_be16(IF2DT(nip->i_inode.i_mode)); |
44aaada9 | 1734 | dent->de_rahead = cpu_to_be16(gfs2_inode_ra_len(nip)); |
01bcb0de | 1735 | tv = CURRENT_TIME; |
383f01fb | 1736 | if (ip->i_diskflags & GFS2_DIF_EXHASH) { |
c752666c | 1737 | leaf = (struct gfs2_leaf *)bh->b_data; |
bb16b342 | 1738 | be16_add_cpu(&leaf->lf_entries, 1); |
01bcb0de SW |
1739 | leaf->lf_nsec = cpu_to_be32(tv.tv_nsec); |
1740 | leaf->lf_sec = cpu_to_be64(tv.tv_sec); | |
c752666c | 1741 | } |
2b47dad8 SW |
1742 | da->dent = NULL; |
1743 | da->bh = NULL; | |
c752666c | 1744 | brelse(bh); |
ad6203f2 | 1745 | ip->i_entries++; |
01bcb0de | 1746 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = tv; |
3d6ecb7d SW |
1747 | if (S_ISDIR(nip->i_inode.i_mode)) |
1748 | inc_nlink(&ip->i_inode); | |
343cd8f0 | 1749 | mark_inode_dirty(inode); |
c752666c SW |
1750 | error = 0; |
1751 | break; | |
1752 | } | |
383f01fb | 1753 | if (!(ip->i_diskflags & GFS2_DIF_EXHASH)) { |
c752666c SW |
1754 | error = dir_make_exhash(inode); |
1755 | if (error) | |
1756 | break; | |
1757 | continue; | |
1758 | } | |
1759 | error = dir_split_leaf(inode, name); | |
1760 | if (error == 0) | |
1761 | continue; | |
e90deff5 | 1762 | if (error < 0) |
c752666c | 1763 | break; |
9a004508 | 1764 | if (ip->i_depth < GFS2_DIR_MAX_DEPTH) { |
c752666c SW |
1765 | error = dir_double_exhash(ip); |
1766 | if (error) | |
1767 | break; | |
1768 | error = dir_split_leaf(inode, name); | |
e90deff5 | 1769 | if (error < 0) |
c752666c | 1770 | break; |
e90deff5 SW |
1771 | if (error == 0) |
1772 | continue; | |
c752666c SW |
1773 | } |
1774 | error = dir_new_leaf(inode, name); | |
1775 | if (!error) | |
1776 | continue; | |
1777 | error = -ENOSPC; | |
1778 | break; | |
1779 | } | |
b3b94faa DT |
1780 | return error; |
1781 | } | |
1782 | ||
c752666c | 1783 | |
b3b94faa DT |
1784 | /** |
1785 | * gfs2_dir_del - Delete a directory entry | |
1786 | * @dip: The GFS2 inode | |
1787 | * @filename: The filename | |
1788 | * | |
1789 | * Returns: 0 on success, error code on failure | |
1790 | */ | |
1791 | ||
855d23ce | 1792 | int gfs2_dir_del(struct gfs2_inode *dip, const struct dentry *dentry) |
b3b94faa | 1793 | { |
855d23ce | 1794 | const struct qstr *name = &dentry->d_name; |
c752666c SW |
1795 | struct gfs2_dirent *dent, *prev = NULL; |
1796 | struct buffer_head *bh; | |
01bcb0de | 1797 | struct timespec tv = CURRENT_TIME; |
b3b94faa | 1798 | |
c752666c SW |
1799 | /* Returns _either_ the entry (if its first in block) or the |
1800 | previous entry otherwise */ | |
feaa7bba | 1801 | dent = gfs2_dirent_search(&dip->i_inode, name, gfs2_dirent_prev, &bh); |
c752666c SW |
1802 | if (!dent) { |
1803 | gfs2_consist_inode(dip); | |
1804 | return -EIO; | |
1805 | } | |
1806 | if (IS_ERR(dent)) { | |
1807 | gfs2_consist_inode(dip); | |
1808 | return PTR_ERR(dent); | |
1809 | } | |
1810 | /* If not first in block, adjust pointers accordingly */ | |
71b86f56 | 1811 | if (gfs2_dirent_find(dent, name, NULL) == 0) { |
c752666c SW |
1812 | prev = dent; |
1813 | dent = (struct gfs2_dirent *)((char *)dent + be16_to_cpu(prev->de_rec_len)); | |
1814 | } | |
1815 | ||
1816 | dirent_del(dip, bh, prev, dent); | |
383f01fb | 1817 | if (dip->i_diskflags & GFS2_DIF_EXHASH) { |
c752666c SW |
1818 | struct gfs2_leaf *leaf = (struct gfs2_leaf *)bh->b_data; |
1819 | u16 entries = be16_to_cpu(leaf->lf_entries); | |
1820 | if (!entries) | |
1821 | gfs2_consist_inode(dip); | |
1822 | leaf->lf_entries = cpu_to_be16(--entries); | |
01bcb0de SW |
1823 | leaf->lf_nsec = cpu_to_be32(tv.tv_nsec); |
1824 | leaf->lf_sec = cpu_to_be64(tv.tv_sec); | |
c752666c | 1825 | } |
ed386507 | 1826 | brelse(bh); |
c752666c | 1827 | |
ad6203f2 | 1828 | if (!dip->i_entries) |
c752666c | 1829 | gfs2_consist_inode(dip); |
ad6203f2 | 1830 | dip->i_entries--; |
01bcb0de | 1831 | dip->i_inode.i_mtime = dip->i_inode.i_ctime = tv; |
855d23ce SW |
1832 | if (S_ISDIR(dentry->d_inode->i_mode)) |
1833 | drop_nlink(&dip->i_inode); | |
feaa7bba | 1834 | mark_inode_dirty(&dip->i_inode); |
b3b94faa | 1835 | |
ab9bbda0 | 1836 | return 0; |
b3b94faa DT |
1837 | } |
1838 | ||
b3b94faa DT |
1839 | /** |
1840 | * gfs2_dir_mvino - Change inode number of directory entry | |
1841 | * @dip: The GFS2 inode | |
1842 | * @filename: | |
1843 | * @new_inode: | |
1844 | * | |
1845 | * This routine changes the inode number of a directory entry. It's used | |
1846 | * by rename to change ".." when a directory is moved. | |
1847 | * Assumes a glock is held on dvp. | |
1848 | * | |
1849 | * Returns: errno | |
1850 | */ | |
1851 | ||
c752666c | 1852 | int gfs2_dir_mvino(struct gfs2_inode *dip, const struct qstr *filename, |
dbb7cae2 | 1853 | const struct gfs2_inode *nip, unsigned int new_type) |
b3b94faa | 1854 | { |
c752666c SW |
1855 | struct buffer_head *bh; |
1856 | struct gfs2_dirent *dent; | |
b3b94faa DT |
1857 | int error; |
1858 | ||
feaa7bba | 1859 | dent = gfs2_dirent_search(&dip->i_inode, filename, gfs2_dirent_find, &bh); |
c752666c SW |
1860 | if (!dent) { |
1861 | gfs2_consist_inode(dip); | |
1862 | return -EIO; | |
1863 | } | |
1864 | if (IS_ERR(dent)) | |
1865 | return PTR_ERR(dent); | |
b3b94faa | 1866 | |
350a9b0a | 1867 | gfs2_trans_add_meta(dip->i_gl, bh); |
dbb7cae2 | 1868 | gfs2_inum_out(nip, dent); |
c752666c SW |
1869 | dent->de_type = cpu_to_be16(new_type); |
1870 | ||
383f01fb | 1871 | if (dip->i_diskflags & GFS2_DIF_EXHASH) { |
c752666c SW |
1872 | brelse(bh); |
1873 | error = gfs2_meta_inode_buffer(dip, &bh); | |
1874 | if (error) | |
1875 | return error; | |
350a9b0a | 1876 | gfs2_trans_add_meta(dip->i_gl, bh); |
c752666c SW |
1877 | } |
1878 | ||
4bd91ba1 | 1879 | dip->i_inode.i_mtime = dip->i_inode.i_ctime = CURRENT_TIME; |
539e5d6b | 1880 | gfs2_dinode_out(dip, bh->b_data); |
c752666c SW |
1881 | brelse(bh); |
1882 | return 0; | |
b3b94faa DT |
1883 | } |
1884 | ||
b3b94faa DT |
1885 | /** |
1886 | * leaf_dealloc - Deallocate a directory leaf | |
1887 | * @dip: the directory | |
1888 | * @index: the hash table offset in the directory | |
1889 | * @len: the number of pointers to this leaf | |
1890 | * @leaf_no: the leaf number | |
ec038c82 | 1891 | * @leaf_bh: buffer_head for the starting leaf |
d24a7a43 | 1892 | * last_dealloc: 1 if this is the final dealloc for the leaf, else 0 |
b3b94faa DT |
1893 | * |
1894 | * Returns: errno | |
1895 | */ | |
1896 | ||
cd915493 | 1897 | static int leaf_dealloc(struct gfs2_inode *dip, u32 index, u32 len, |
ec038c82 BP |
1898 | u64 leaf_no, struct buffer_head *leaf_bh, |
1899 | int last_dealloc) | |
b3b94faa | 1900 | { |
feaa7bba | 1901 | struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); |
c752666c | 1902 | struct gfs2_leaf *tmp_leaf; |
b3b94faa DT |
1903 | struct gfs2_rgrp_list rlist; |
1904 | struct buffer_head *bh, *dibh; | |
cd915493 | 1905 | u64 blk, nblk; |
b3b94faa DT |
1906 | unsigned int rg_blocks = 0, l_blocks = 0; |
1907 | char *ht; | |
cd915493 | 1908 | unsigned int x, size = len * sizeof(u64); |
b3b94faa DT |
1909 | int error; |
1910 | ||
5e2f7d61 BP |
1911 | error = gfs2_rindex_update(sdp); |
1912 | if (error) | |
1913 | return error; | |
1914 | ||
b3b94faa DT |
1915 | memset(&rlist, 0, sizeof(struct gfs2_rgrp_list)); |
1916 | ||
8be04b93 | 1917 | ht = kzalloc(size, GFP_NOFS | __GFP_NOWARN); |
e8830d88 BP |
1918 | if (ht == NULL) |
1919 | ht = vzalloc(size); | |
b3b94faa DT |
1920 | if (!ht) |
1921 | return -ENOMEM; | |
1922 | ||
f4108a60 | 1923 | error = gfs2_quota_hold(dip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE); |
b3b94faa | 1924 | if (error) |
5407e242 | 1925 | goto out; |
b3b94faa | 1926 | |
b3b94faa | 1927 | /* Count the number of leaves */ |
ec038c82 | 1928 | bh = leaf_bh; |
b3b94faa | 1929 | |
c752666c | 1930 | for (blk = leaf_no; blk; blk = nblk) { |
ec038c82 BP |
1931 | if (blk != leaf_no) { |
1932 | error = get_leaf(dip, blk, &bh); | |
1933 | if (error) | |
1934 | goto out_rlist; | |
1935 | } | |
c752666c SW |
1936 | tmp_leaf = (struct gfs2_leaf *)bh->b_data; |
1937 | nblk = be64_to_cpu(tmp_leaf->lf_next); | |
ec038c82 BP |
1938 | if (blk != leaf_no) |
1939 | brelse(bh); | |
b3b94faa | 1940 | |
70b0c365 | 1941 | gfs2_rlist_add(dip, &rlist, blk); |
b3b94faa DT |
1942 | l_blocks++; |
1943 | } | |
1944 | ||
fe6c991c | 1945 | gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE); |
b3b94faa DT |
1946 | |
1947 | for (x = 0; x < rlist.rl_rgrps; x++) { | |
1948 | struct gfs2_rgrpd *rgd; | |
5c676f6d | 1949 | rgd = rlist.rl_ghs[x].gh_gl->gl_object; |
bb8d8a6f | 1950 | rg_blocks += rgd->rd_length; |
b3b94faa DT |
1951 | } |
1952 | ||
1953 | error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs); | |
1954 | if (error) | |
1955 | goto out_rlist; | |
1956 | ||
1957 | error = gfs2_trans_begin(sdp, | |
5c676f6d | 1958 | rg_blocks + (DIV_ROUND_UP(size, sdp->sd_jbsize) + 1) + |
b3b94faa DT |
1959 | RES_DINODE + RES_STATFS + RES_QUOTA, l_blocks); |
1960 | if (error) | |
1961 | goto out_rg_gunlock; | |
1962 | ||
ec038c82 BP |
1963 | bh = leaf_bh; |
1964 | ||
c752666c | 1965 | for (blk = leaf_no; blk; blk = nblk) { |
ec038c82 BP |
1966 | if (blk != leaf_no) { |
1967 | error = get_leaf(dip, blk, &bh); | |
1968 | if (error) | |
1969 | goto out_end_trans; | |
1970 | } | |
c752666c SW |
1971 | tmp_leaf = (struct gfs2_leaf *)bh->b_data; |
1972 | nblk = be64_to_cpu(tmp_leaf->lf_next); | |
ec038c82 BP |
1973 | if (blk != leaf_no) |
1974 | brelse(bh); | |
b3b94faa DT |
1975 | |
1976 | gfs2_free_meta(dip, blk, 1); | |
77658aad | 1977 | gfs2_add_inode_blocks(&dip->i_inode, -1); |
b3b94faa DT |
1978 | } |
1979 | ||
cd915493 | 1980 | error = gfs2_dir_write_data(dip, ht, index * sizeof(u64), size); |
b3b94faa DT |
1981 | if (error != size) { |
1982 | if (error >= 0) | |
1983 | error = -EIO; | |
1984 | goto out_end_trans; | |
1985 | } | |
1986 | ||
1987 | error = gfs2_meta_inode_buffer(dip, &dibh); | |
1988 | if (error) | |
1989 | goto out_end_trans; | |
1990 | ||
350a9b0a | 1991 | gfs2_trans_add_meta(dip->i_gl, dibh); |
d24a7a43 BP |
1992 | /* On the last dealloc, make this a regular file in case we crash. |
1993 | (We don't want to free these blocks a second time.) */ | |
1994 | if (last_dealloc) | |
1995 | dip->i_inode.i_mode = S_IFREG; | |
539e5d6b | 1996 | gfs2_dinode_out(dip, dibh->b_data); |
b3b94faa DT |
1997 | brelse(dibh); |
1998 | ||
a91ea69f | 1999 | out_end_trans: |
b3b94faa | 2000 | gfs2_trans_end(sdp); |
a91ea69f | 2001 | out_rg_gunlock: |
b3b94faa | 2002 | gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs); |
a91ea69f | 2003 | out_rlist: |
b3b94faa | 2004 | gfs2_rlist_free(&rlist); |
b3b94faa | 2005 | gfs2_quota_unhold(dip); |
182fe5ab | 2006 | out: |
e8830d88 BP |
2007 | if (is_vmalloc_addr(ht)) |
2008 | vfree(ht); | |
2009 | else | |
2010 | kfree(ht); | |
b3b94faa DT |
2011 | return error; |
2012 | } | |
2013 | ||
2014 | /** | |
2015 | * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory | |
2016 | * @dip: the directory | |
2017 | * | |
2018 | * Dealloc all on-disk directory leaves to FREEMETA state | |
2019 | * Change on-disk inode type to "regular file" | |
2020 | * | |
2021 | * Returns: errno | |
2022 | */ | |
2023 | ||
2024 | int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip) | |
2025 | { | |
556bb179 BP |
2026 | struct buffer_head *bh; |
2027 | struct gfs2_leaf *leaf; | |
2028 | u32 hsize, len; | |
556bb179 BP |
2029 | u32 index = 0, next_index; |
2030 | __be64 *lp; | |
2031 | u64 leaf_no; | |
2032 | int error = 0, last; | |
2033 | ||
2034 | hsize = 1 << dip->i_depth; | |
556bb179 | 2035 | |
17d539f0 SW |
2036 | lp = gfs2_dir_get_hash_table(dip); |
2037 | if (IS_ERR(lp)) | |
2038 | return PTR_ERR(lp); | |
556bb179 BP |
2039 | |
2040 | while (index < hsize) { | |
17d539f0 | 2041 | leaf_no = be64_to_cpu(lp[index]); |
556bb179 BP |
2042 | if (leaf_no) { |
2043 | error = get_leaf(dip, leaf_no, &bh); | |
2044 | if (error) | |
2045 | goto out; | |
2046 | leaf = (struct gfs2_leaf *)bh->b_data; | |
2047 | len = 1 << (dip->i_depth - be16_to_cpu(leaf->lf_depth)); | |
2048 | ||
2049 | next_index = (index & ~(len - 1)) + len; | |
2050 | last = ((next_index >= hsize) ? 1 : 0); | |
2051 | error = leaf_dealloc(dip, index, len, leaf_no, bh, | |
2052 | last); | |
2053 | brelse(bh); | |
2054 | if (error) | |
2055 | goto out; | |
2056 | index = next_index; | |
2057 | } else | |
2058 | index++; | |
2059 | } | |
2060 | ||
2061 | if (index != hsize) { | |
2062 | gfs2_consist_inode(dip); | |
2063 | error = -EIO; | |
2064 | } | |
2065 | ||
2066 | out: | |
556bb179 BP |
2067 | |
2068 | return error; | |
b3b94faa DT |
2069 | } |
2070 | ||
2071 | /** | |
2072 | * gfs2_diradd_alloc_required - find if adding entry will require an allocation | |
2073 | * @ip: the file being written to | |
2074 | * @filname: the filename that's going to be added | |
3c1c0ae1 | 2075 | * @da: The structure to return dir alloc info |
b3b94faa | 2076 | * |
3c1c0ae1 | 2077 | * Returns: 0 if ok, -ve on error |
b3b94faa DT |
2078 | */ |
2079 | ||
3c1c0ae1 SW |
2080 | int gfs2_diradd_alloc_required(struct inode *inode, const struct qstr *name, |
2081 | struct gfs2_diradd *da) | |
b3b94faa | 2082 | { |
22b5a6c0 | 2083 | struct gfs2_inode *ip = GFS2_I(inode); |
3c1c0ae1 | 2084 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
22b5a6c0 | 2085 | const unsigned int extra = sizeof(struct gfs2_dinode) - sizeof(struct gfs2_leaf); |
c752666c SW |
2086 | struct gfs2_dirent *dent; |
2087 | struct buffer_head *bh; | |
b3b94faa | 2088 | |
3c1c0ae1 | 2089 | da->nr_blocks = 0; |
2b47dad8 SW |
2090 | da->bh = NULL; |
2091 | da->dent = NULL; | |
3c1c0ae1 | 2092 | |
c752666c | 2093 | dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, &bh); |
ed386507 | 2094 | if (!dent) { |
3c1c0ae1 | 2095 | da->nr_blocks = sdp->sd_max_dirres; |
22b5a6c0 SW |
2096 | if (!(ip->i_diskflags & GFS2_DIF_EXHASH) && |
2097 | (GFS2_DIRENT_SIZE(name->len) < extra)) | |
2098 | da->nr_blocks = 1; | |
3c1c0ae1 | 2099 | return 0; |
ed386507 | 2100 | } |
c752666c SW |
2101 | if (IS_ERR(dent)) |
2102 | return PTR_ERR(dent); | |
19aeb5a6 BP |
2103 | |
2104 | if (da->save_loc) { | |
2105 | da->bh = bh; | |
2106 | da->dent = dent; | |
2107 | } else { | |
2108 | brelse(bh); | |
2109 | } | |
c752666c | 2110 | return 0; |
b3b94faa DT |
2111 | } |
2112 |