]>
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) { |
3cdcf63e | 368 | kvfree(hc); |
17d539f0 SW |
369 | return ERR_PTR(ret); |
370 | } | |
371 | ||
372 | spin_lock(&inode->i_lock); | |
3cdcf63e AV |
373 | if (ip->i_hash_cache) |
374 | kvfree(hc); | |
375 | else | |
17d539f0 SW |
376 | ip->i_hash_cache = hc; |
377 | spin_unlock(&inode->i_lock); | |
378 | ||
379 | return ip->i_hash_cache; | |
380 | } | |
381 | ||
382 | /** | |
383 | * gfs2_dir_hash_inval - Invalidate dir hash | |
384 | * @ip: The directory inode | |
385 | * | |
386 | * Must be called with an exclusive glock, or during glock invalidation. | |
387 | */ | |
388 | void gfs2_dir_hash_inval(struct gfs2_inode *ip) | |
389 | { | |
390 | __be64 *hc = ip->i_hash_cache; | |
391 | ip->i_hash_cache = NULL; | |
3cdcf63e | 392 | kvfree(hc); |
17d539f0 SW |
393 | } |
394 | ||
5e7d65cd SW |
395 | static inline int gfs2_dirent_sentinel(const struct gfs2_dirent *dent) |
396 | { | |
397 | return dent->de_inum.no_addr == 0 || dent->de_inum.no_formal_ino == 0; | |
398 | } | |
399 | ||
c752666c SW |
400 | static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent, |
401 | const struct qstr *name, int ret) | |
402 | { | |
5e7d65cd | 403 | if (!gfs2_dirent_sentinel(dent) && |
c752666c SW |
404 | be32_to_cpu(dent->de_hash) == name->hash && |
405 | be16_to_cpu(dent->de_name_len) == name->len && | |
2bdbc5d7 | 406 | memcmp(dent+1, name->name, name->len) == 0) |
c752666c SW |
407 | return ret; |
408 | return 0; | |
409 | } | |
410 | ||
411 | static int gfs2_dirent_find(const struct gfs2_dirent *dent, | |
71b86f56 SW |
412 | const struct qstr *name, |
413 | void *opaque) | |
c752666c SW |
414 | { |
415 | return __gfs2_dirent_find(dent, name, 1); | |
416 | } | |
417 | ||
418 | static int gfs2_dirent_prev(const struct gfs2_dirent *dent, | |
71b86f56 SW |
419 | const struct qstr *name, |
420 | void *opaque) | |
c752666c SW |
421 | { |
422 | return __gfs2_dirent_find(dent, name, 2); | |
423 | } | |
424 | ||
425 | /* | |
426 | * name->name holds ptr to start of block. | |
427 | * name->len holds size of block. | |
b3b94faa | 428 | */ |
c752666c | 429 | static int gfs2_dirent_last(const struct gfs2_dirent *dent, |
71b86f56 SW |
430 | const struct qstr *name, |
431 | void *opaque) | |
c752666c SW |
432 | { |
433 | const char *start = name->name; | |
434 | const char *end = (const char *)dent + be16_to_cpu(dent->de_rec_len); | |
435 | if (name->len == (end - start)) | |
436 | return 1; | |
437 | return 0; | |
438 | } | |
b3b94faa | 439 | |
c752666c | 440 | static int gfs2_dirent_find_space(const struct gfs2_dirent *dent, |
71b86f56 SW |
441 | const struct qstr *name, |
442 | void *opaque) | |
b3b94faa | 443 | { |
c752666c SW |
444 | unsigned required = GFS2_DIRENT_SIZE(name->len); |
445 | unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len)); | |
446 | unsigned totlen = be16_to_cpu(dent->de_rec_len); | |
447 | ||
5e7d65cd | 448 | if (gfs2_dirent_sentinel(dent)) |
728a756b | 449 | actual = 0; |
c5392124 | 450 | if (totlen - actual >= required) |
c752666c SW |
451 | return 1; |
452 | return 0; | |
453 | } | |
454 | ||
71b86f56 SW |
455 | struct dirent_gather { |
456 | const struct gfs2_dirent **pdent; | |
457 | unsigned offset; | |
458 | }; | |
459 | ||
460 | static int gfs2_dirent_gather(const struct gfs2_dirent *dent, | |
461 | const struct qstr *name, | |
462 | void *opaque) | |
463 | { | |
464 | struct dirent_gather *g = opaque; | |
5e7d65cd | 465 | if (!gfs2_dirent_sentinel(dent)) { |
71b86f56 SW |
466 | g->pdent[g->offset++] = dent; |
467 | } | |
468 | return 0; | |
469 | } | |
470 | ||
c752666c SW |
471 | /* |
472 | * Other possible things to check: | |
473 | * - Inode located within filesystem size (and on valid block) | |
474 | * - Valid directory entry type | |
475 | * Not sure how heavy-weight we want to make this... could also check | |
476 | * hash is correct for example, but that would take a lot of extra time. | |
477 | * For now the most important thing is to check that the various sizes | |
478 | * are correct. | |
479 | */ | |
480 | static int gfs2_check_dirent(struct gfs2_dirent *dent, unsigned int offset, | |
481 | unsigned int size, unsigned int len, int first) | |
482 | { | |
483 | const char *msg = "gfs2_dirent too small"; | |
484 | if (unlikely(size < sizeof(struct gfs2_dirent))) | |
485 | goto error; | |
486 | msg = "gfs2_dirent misaligned"; | |
487 | if (unlikely(offset & 0x7)) | |
488 | goto error; | |
489 | msg = "gfs2_dirent points beyond end of block"; | |
490 | if (unlikely(offset + size > len)) | |
491 | goto error; | |
492 | msg = "zero inode number"; | |
5e7d65cd | 493 | if (unlikely(!first && gfs2_dirent_sentinel(dent))) |
c752666c SW |
494 | goto error; |
495 | msg = "name length is greater than space in dirent"; | |
5e7d65cd | 496 | if (!gfs2_dirent_sentinel(dent) && |
c752666c SW |
497 | unlikely(sizeof(struct gfs2_dirent)+be16_to_cpu(dent->de_name_len) > |
498 | size)) | |
499 | goto error; | |
500 | return 0; | |
501 | error: | |
d77d1b58 JP |
502 | pr_warn("%s: %s (%s)\n", |
503 | __func__, msg, first ? "first in block" : "not first in block"); | |
c752666c | 504 | return -EIO; |
b3b94faa DT |
505 | } |
506 | ||
71b86f56 | 507 | static int gfs2_dirent_offset(const void *buf) |
c752666c | 508 | { |
71b86f56 SW |
509 | const struct gfs2_meta_header *h = buf; |
510 | int offset; | |
c752666c SW |
511 | |
512 | BUG_ON(buf == NULL); | |
c752666c | 513 | |
e3167ded | 514 | switch(be32_to_cpu(h->mh_type)) { |
c752666c SW |
515 | case GFS2_METATYPE_LF: |
516 | offset = sizeof(struct gfs2_leaf); | |
517 | break; | |
518 | case GFS2_METATYPE_DI: | |
519 | offset = sizeof(struct gfs2_dinode); | |
520 | break; | |
521 | default: | |
522 | goto wrong_type; | |
523 | } | |
71b86f56 SW |
524 | return offset; |
525 | wrong_type: | |
d77d1b58 | 526 | pr_warn("%s: wrong block type %u\n", __func__, be32_to_cpu(h->mh_type)); |
71b86f56 SW |
527 | return -1; |
528 | } | |
529 | ||
2bdbc5d7 | 530 | static struct gfs2_dirent *gfs2_dirent_scan(struct inode *inode, void *buf, |
71b86f56 SW |
531 | unsigned int len, gfs2_dscan_t scan, |
532 | const struct qstr *name, | |
533 | void *opaque) | |
534 | { | |
535 | struct gfs2_dirent *dent, *prev; | |
536 | unsigned offset; | |
537 | unsigned size; | |
538 | int ret = 0; | |
c752666c | 539 | |
71b86f56 SW |
540 | ret = gfs2_dirent_offset(buf); |
541 | if (ret < 0) | |
542 | goto consist_inode; | |
543 | ||
544 | offset = ret; | |
c752666c | 545 | prev = NULL; |
2bdbc5d7 | 546 | dent = buf + offset; |
c752666c SW |
547 | size = be16_to_cpu(dent->de_rec_len); |
548 | if (gfs2_check_dirent(dent, offset, size, len, 1)) | |
549 | goto consist_inode; | |
550 | do { | |
71b86f56 | 551 | ret = scan(dent, name, opaque); |
c752666c SW |
552 | if (ret) |
553 | break; | |
554 | offset += size; | |
555 | if (offset == len) | |
556 | break; | |
557 | prev = dent; | |
2bdbc5d7 | 558 | dent = buf + offset; |
c752666c SW |
559 | size = be16_to_cpu(dent->de_rec_len); |
560 | if (gfs2_check_dirent(dent, offset, size, len, 0)) | |
561 | goto consist_inode; | |
562 | } while(1); | |
563 | ||
564 | switch(ret) { | |
565 | case 0: | |
566 | return NULL; | |
567 | case 1: | |
568 | return dent; | |
569 | case 2: | |
570 | return prev ? prev : dent; | |
571 | default: | |
572 | BUG_ON(ret > 0); | |
573 | return ERR_PTR(ret); | |
574 | } | |
575 | ||
c752666c | 576 | consist_inode: |
feaa7bba | 577 | gfs2_consist_inode(GFS2_I(inode)); |
c752666c SW |
578 | return ERR_PTR(-EIO); |
579 | } | |
580 | ||
2bdbc5d7 SW |
581 | static int dirent_check_reclen(struct gfs2_inode *dip, |
582 | const struct gfs2_dirent *d, const void *end_p) | |
583 | { | |
584 | const void *ptr = d; | |
585 | u16 rec_len = be16_to_cpu(d->de_rec_len); | |
586 | ||
587 | if (unlikely(rec_len < sizeof(struct gfs2_dirent))) | |
588 | goto broken; | |
589 | ptr += rec_len; | |
590 | if (ptr < end_p) | |
591 | return rec_len; | |
592 | if (ptr == end_p) | |
593 | return -ENOENT; | |
594 | broken: | |
595 | gfs2_consist_inode(dip); | |
596 | return -EIO; | |
597 | } | |
598 | ||
b3b94faa DT |
599 | /** |
600 | * dirent_next - Next dirent | |
601 | * @dip: the directory | |
602 | * @bh: The buffer | |
603 | * @dent: Pointer to list of dirents | |
604 | * | |
605 | * Returns: 0 on success, error code otherwise | |
606 | */ | |
607 | ||
608 | static int dirent_next(struct gfs2_inode *dip, struct buffer_head *bh, | |
609 | struct gfs2_dirent **dent) | |
610 | { | |
2bdbc5d7 SW |
611 | struct gfs2_dirent *cur = *dent, *tmp; |
612 | char *bh_end = bh->b_data + bh->b_size; | |
613 | int ret; | |
b3b94faa | 614 | |
2bdbc5d7 SW |
615 | ret = dirent_check_reclen(dip, cur, bh_end); |
616 | if (ret < 0) | |
617 | return ret; | |
4dd651ad | 618 | |
2bdbc5d7 SW |
619 | tmp = (void *)cur + ret; |
620 | ret = dirent_check_reclen(dip, tmp, bh_end); | |
621 | if (ret == -EIO) | |
622 | return ret; | |
4dd651ad | 623 | |
b3b94faa | 624 | /* Only the first dent could ever have de_inum.no_addr == 0 */ |
5e7d65cd | 625 | if (gfs2_dirent_sentinel(tmp)) { |
b3b94faa DT |
626 | gfs2_consist_inode(dip); |
627 | return -EIO; | |
628 | } | |
629 | ||
630 | *dent = tmp; | |
b3b94faa DT |
631 | return 0; |
632 | } | |
633 | ||
634 | /** | |
635 | * dirent_del - Delete a dirent | |
636 | * @dip: The GFS2 inode | |
637 | * @bh: The buffer | |
638 | * @prev: The previous dirent | |
639 | * @cur: The current dirent | |
640 | * | |
641 | */ | |
642 | ||
643 | static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh, | |
644 | struct gfs2_dirent *prev, struct gfs2_dirent *cur) | |
645 | { | |
cd915493 | 646 | u16 cur_rec_len, prev_rec_len; |
b3b94faa | 647 | |
5e7d65cd | 648 | if (gfs2_dirent_sentinel(cur)) { |
b3b94faa DT |
649 | gfs2_consist_inode(dip); |
650 | return; | |
651 | } | |
652 | ||
350a9b0a | 653 | gfs2_trans_add_meta(dip->i_gl, bh); |
b3b94faa DT |
654 | |
655 | /* If there is no prev entry, this is the first entry in the block. | |
656 | The de_rec_len is already as big as it needs to be. Just zero | |
657 | out the inode number and return. */ | |
658 | ||
659 | if (!prev) { | |
5e7d65cd SW |
660 | cur->de_inum.no_addr = 0; |
661 | cur->de_inum.no_formal_ino = 0; | |
b3b94faa DT |
662 | return; |
663 | } | |
664 | ||
665 | /* Combine this dentry with the previous one. */ | |
666 | ||
fc69d0d3 SW |
667 | prev_rec_len = be16_to_cpu(prev->de_rec_len); |
668 | cur_rec_len = be16_to_cpu(cur->de_rec_len); | |
b3b94faa DT |
669 | |
670 | if ((char *)prev + prev_rec_len != (char *)cur) | |
671 | gfs2_consist_inode(dip); | |
672 | if ((char *)cur + cur_rec_len > bh->b_data + bh->b_size) | |
673 | gfs2_consist_inode(dip); | |
674 | ||
675 | prev_rec_len += cur_rec_len; | |
fc69d0d3 | 676 | prev->de_rec_len = cpu_to_be16(prev_rec_len); |
b3b94faa DT |
677 | } |
678 | ||
c752666c SW |
679 | /* |
680 | * Takes a dent from which to grab space as an argument. Returns the | |
681 | * newly created dent. | |
b3b94faa | 682 | */ |
08bc2dbc AB |
683 | static struct gfs2_dirent *gfs2_init_dirent(struct inode *inode, |
684 | struct gfs2_dirent *dent, | |
685 | const struct qstr *name, | |
686 | struct buffer_head *bh) | |
b3b94faa | 687 | { |
feaa7bba | 688 | struct gfs2_inode *ip = GFS2_I(inode); |
c752666c SW |
689 | struct gfs2_dirent *ndent; |
690 | unsigned offset = 0, totlen; | |
691 | ||
5e7d65cd | 692 | if (!gfs2_dirent_sentinel(dent)) |
c752666c SW |
693 | offset = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len)); |
694 | totlen = be16_to_cpu(dent->de_rec_len); | |
695 | BUG_ON(offset + name->len > totlen); | |
350a9b0a | 696 | gfs2_trans_add_meta(ip->i_gl, bh); |
c752666c SW |
697 | ndent = (struct gfs2_dirent *)((char *)dent + offset); |
698 | dent->de_rec_len = cpu_to_be16(offset); | |
699 | gfs2_qstr2dirent(name, totlen - offset, ndent); | |
700 | return ndent; | |
b3b94faa DT |
701 | } |
702 | ||
c752666c SW |
703 | static struct gfs2_dirent *gfs2_dirent_alloc(struct inode *inode, |
704 | struct buffer_head *bh, | |
705 | const struct qstr *name) | |
b3b94faa DT |
706 | { |
707 | struct gfs2_dirent *dent; | |
907b9bce | 708 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, |
71b86f56 | 709 | gfs2_dirent_find_space, name, NULL); |
c752666c SW |
710 | if (!dent || IS_ERR(dent)) |
711 | return dent; | |
712 | return gfs2_init_dirent(inode, dent, name, bh); | |
b3b94faa DT |
713 | } |
714 | ||
cd915493 | 715 | static int get_leaf(struct gfs2_inode *dip, u64 leaf_no, |
b3b94faa DT |
716 | struct buffer_head **bhp) |
717 | { | |
718 | int error; | |
719 | ||
7276b3b0 | 720 | error = gfs2_meta_read(dip->i_gl, leaf_no, DIO_WAIT, bhp); |
feaa7bba | 721 | if (!error && gfs2_metatype_check(GFS2_SB(&dip->i_inode), *bhp, GFS2_METATYPE_LF)) { |
d77d1b58 | 722 | /* pr_info("block num=%llu\n", leaf_no); */ |
b3b94faa | 723 | error = -EIO; |
feaa7bba | 724 | } |
b3b94faa DT |
725 | |
726 | return error; | |
727 | } | |
728 | ||
729 | /** | |
730 | * get_leaf_nr - Get a leaf number associated with the index | |
731 | * @dip: The GFS2 inode | |
732 | * @index: | |
733 | * @leaf_out: | |
734 | * | |
735 | * Returns: 0 on success, error code otherwise | |
736 | */ | |
737 | ||
cd915493 SW |
738 | static int get_leaf_nr(struct gfs2_inode *dip, u32 index, |
739 | u64 *leaf_out) | |
b3b94faa | 740 | { |
17d539f0 | 741 | __be64 *hash; |
b3b94faa | 742 | |
17d539f0 SW |
743 | hash = gfs2_dir_get_hash_table(dip); |
744 | if (IS_ERR(hash)) | |
745 | return PTR_ERR(hash); | |
746 | *leaf_out = be64_to_cpu(*(hash + index)); | |
b3b94faa DT |
747 | return 0; |
748 | } | |
749 | ||
cd915493 | 750 | static int get_first_leaf(struct gfs2_inode *dip, u32 index, |
b3b94faa DT |
751 | struct buffer_head **bh_out) |
752 | { | |
cd915493 | 753 | u64 leaf_no; |
b3b94faa DT |
754 | int error; |
755 | ||
756 | error = get_leaf_nr(dip, index, &leaf_no); | |
757 | if (!error) | |
758 | error = get_leaf(dip, leaf_no, bh_out); | |
759 | ||
760 | return error; | |
761 | } | |
762 | ||
c752666c SW |
763 | static struct gfs2_dirent *gfs2_dirent_search(struct inode *inode, |
764 | const struct qstr *name, | |
765 | gfs2_dscan_t scan, | |
766 | struct buffer_head **pbh) | |
b3b94faa | 767 | { |
c752666c SW |
768 | struct buffer_head *bh; |
769 | struct gfs2_dirent *dent; | |
feaa7bba | 770 | struct gfs2_inode *ip = GFS2_I(inode); |
b3b94faa DT |
771 | int error; |
772 | ||
383f01fb | 773 | if (ip->i_diskflags & GFS2_DIF_EXHASH) { |
c752666c | 774 | struct gfs2_leaf *leaf; |
9a004508 | 775 | unsigned hsize = 1 << ip->i_depth; |
c752666c SW |
776 | unsigned index; |
777 | u64 ln; | |
a2e0f799 | 778 | if (hsize * sizeof(u64) != i_size_read(inode)) { |
c752666c SW |
779 | gfs2_consist_inode(ip); |
780 | return ERR_PTR(-EIO); | |
781 | } | |
907b9bce | 782 | |
9a004508 | 783 | index = name->hash >> (32 - ip->i_depth); |
c752666c SW |
784 | error = get_first_leaf(ip, index, &bh); |
785 | if (error) | |
786 | return ERR_PTR(error); | |
787 | do { | |
788 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, | |
71b86f56 | 789 | scan, name, NULL); |
c752666c SW |
790 | if (dent) |
791 | goto got_dent; | |
792 | leaf = (struct gfs2_leaf *)bh->b_data; | |
793 | ln = be64_to_cpu(leaf->lf_next); | |
f4154ea0 | 794 | brelse(bh); |
c752666c SW |
795 | if (!ln) |
796 | break; | |
907b9bce | 797 | |
c752666c SW |
798 | error = get_leaf(ip, ln, &bh); |
799 | } while(!error); | |
b3b94faa | 800 | |
c752666c | 801 | return error ? ERR_PTR(error) : NULL; |
b3b94faa | 802 | } |
b3b94faa | 803 | |
907b9bce | 804 | |
c752666c SW |
805 | error = gfs2_meta_inode_buffer(ip, &bh); |
806 | if (error) | |
807 | return ERR_PTR(error); | |
71b86f56 | 808 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, scan, name, NULL); |
c752666c | 809 | got_dent: |
f4154ea0 | 810 | if (unlikely(dent == NULL || IS_ERR(dent))) { |
ed386507 SW |
811 | brelse(bh); |
812 | bh = NULL; | |
813 | } | |
c752666c SW |
814 | *pbh = bh; |
815 | return dent; | |
816 | } | |
b3b94faa | 817 | |
c752666c SW |
818 | static struct gfs2_leaf *new_leaf(struct inode *inode, struct buffer_head **pbh, u16 depth) |
819 | { | |
feaa7bba | 820 | struct gfs2_inode *ip = GFS2_I(inode); |
b45e41d7 | 821 | unsigned int n = 1; |
09010978 SW |
822 | u64 bn; |
823 | int error; | |
824 | struct buffer_head *bh; | |
c752666c SW |
825 | struct gfs2_leaf *leaf; |
826 | struct gfs2_dirent *dent; | |
26fe5750 | 827 | struct qstr name = { .name = "" }; |
01bcb0de | 828 | struct timespec tv = CURRENT_TIME; |
09010978 | 829 | |
6e87ed0f | 830 | error = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL); |
09010978 SW |
831 | if (error) |
832 | return NULL; | |
833 | bh = gfs2_meta_new(ip->i_gl, bn); | |
c752666c SW |
834 | if (!bh) |
835 | return NULL; | |
09010978 | 836 | |
5731be53 | 837 | gfs2_trans_add_unrevoke(GFS2_SB(inode), bn, 1); |
350a9b0a | 838 | gfs2_trans_add_meta(ip->i_gl, bh); |
c752666c SW |
839 | gfs2_metatype_set(bh, GFS2_METATYPE_LF, GFS2_FORMAT_LF); |
840 | leaf = (struct gfs2_leaf *)bh->b_data; | |
841 | leaf->lf_depth = cpu_to_be16(depth); | |
2bdbc5d7 | 842 | leaf->lf_entries = 0; |
a2d7d021 | 843 | leaf->lf_dirent_format = cpu_to_be32(GFS2_FORMAT_DE); |
2bdbc5d7 | 844 | leaf->lf_next = 0; |
01bcb0de SW |
845 | leaf->lf_inode = cpu_to_be64(ip->i_no_addr); |
846 | leaf->lf_dist = cpu_to_be32(1); | |
847 | leaf->lf_nsec = cpu_to_be32(tv.tv_nsec); | |
848 | leaf->lf_sec = cpu_to_be64(tv.tv_sec); | |
849 | memset(leaf->lf_reserved2, 0, sizeof(leaf->lf_reserved2)); | |
c752666c | 850 | dent = (struct gfs2_dirent *)(leaf+1); |
71b86f56 | 851 | gfs2_qstr2dirent(&name, bh->b_size - sizeof(struct gfs2_leaf), dent); |
c752666c SW |
852 | *pbh = bh; |
853 | return leaf; | |
b3b94faa DT |
854 | } |
855 | ||
856 | /** | |
857 | * dir_make_exhash - Convert a stuffed directory into an ExHash directory | |
858 | * @dip: The GFS2 inode | |
859 | * | |
860 | * Returns: 0 on success, error code otherwise | |
861 | */ | |
862 | ||
c752666c | 863 | static int dir_make_exhash(struct inode *inode) |
b3b94faa | 864 | { |
feaa7bba SW |
865 | struct gfs2_inode *dip = GFS2_I(inode); |
866 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
b3b94faa | 867 | struct gfs2_dirent *dent; |
c752666c | 868 | struct qstr args; |
b3b94faa DT |
869 | struct buffer_head *bh, *dibh; |
870 | struct gfs2_leaf *leaf; | |
871 | int y; | |
cd915493 | 872 | u32 x; |
b44b84d7 AV |
873 | __be64 *lp; |
874 | u64 bn; | |
b3b94faa DT |
875 | int error; |
876 | ||
877 | error = gfs2_meta_inode_buffer(dip, &dibh); | |
878 | if (error) | |
879 | return error; | |
880 | ||
b3b94faa DT |
881 | /* Turn over a new leaf */ |
882 | ||
c752666c SW |
883 | leaf = new_leaf(inode, &bh, 0); |
884 | if (!leaf) | |
885 | return -ENOSPC; | |
886 | bn = bh->b_blocknr; | |
b3b94faa | 887 | |
ad6203f2 SW |
888 | gfs2_assert(sdp, dip->i_entries < (1 << 16)); |
889 | leaf->lf_entries = cpu_to_be16(dip->i_entries); | |
b3b94faa DT |
890 | |
891 | /* Copy dirents */ | |
892 | ||
893 | gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_leaf), dibh, | |
894 | sizeof(struct gfs2_dinode)); | |
895 | ||
896 | /* Find last entry */ | |
897 | ||
898 | x = 0; | |
c752666c SW |
899 | args.len = bh->b_size - sizeof(struct gfs2_dinode) + |
900 | sizeof(struct gfs2_leaf); | |
901 | args.name = bh->b_data; | |
feaa7bba | 902 | dent = gfs2_dirent_scan(&dip->i_inode, bh->b_data, bh->b_size, |
71b86f56 | 903 | gfs2_dirent_last, &args, NULL); |
c752666c SW |
904 | if (!dent) { |
905 | brelse(bh); | |
906 | brelse(dibh); | |
907 | return -EIO; | |
908 | } | |
909 | if (IS_ERR(dent)) { | |
910 | brelse(bh); | |
911 | brelse(dibh); | |
912 | return PTR_ERR(dent); | |
b3b94faa | 913 | } |
b3b94faa DT |
914 | |
915 | /* Adjust the last dirent's record length | |
916 | (Remember that dent still points to the last entry.) */ | |
917 | ||
4dd651ad | 918 | dent->de_rec_len = cpu_to_be16(be16_to_cpu(dent->de_rec_len) + |
b3b94faa | 919 | sizeof(struct gfs2_dinode) - |
4dd651ad | 920 | sizeof(struct gfs2_leaf)); |
b3b94faa DT |
921 | |
922 | brelse(bh); | |
923 | ||
924 | /* We're done with the new leaf block, now setup the new | |
925 | hash table. */ | |
926 | ||
350a9b0a | 927 | gfs2_trans_add_meta(dip->i_gl, dibh); |
b3b94faa DT |
928 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); |
929 | ||
b44b84d7 | 930 | lp = (__be64 *)(dibh->b_data + sizeof(struct gfs2_dinode)); |
b3b94faa DT |
931 | |
932 | for (x = sdp->sd_hash_ptrs; x--; lp++) | |
933 | *lp = cpu_to_be64(bn); | |
934 | ||
a2e0f799 | 935 | i_size_write(inode, sdp->sd_sb.sb_bsize / 2); |
77658aad | 936 | gfs2_add_inode_blocks(&dip->i_inode, 1); |
383f01fb | 937 | dip->i_diskflags |= GFS2_DIF_EXHASH; |
b3b94faa DT |
938 | |
939 | for (x = sdp->sd_hash_ptrs, y = -1; x; x >>= 1, y++) ; | |
9a004508 | 940 | dip->i_depth = y; |
b3b94faa | 941 | |
539e5d6b | 942 | gfs2_dinode_out(dip, dibh->b_data); |
b3b94faa DT |
943 | |
944 | brelse(dibh); | |
945 | ||
946 | return 0; | |
947 | } | |
948 | ||
949 | /** | |
950 | * dir_split_leaf - Split a leaf block into two | |
951 | * @dip: The GFS2 inode | |
952 | * @index: | |
953 | * @leaf_no: | |
954 | * | |
955 | * Returns: 0 on success, error code on failure | |
956 | */ | |
957 | ||
c752666c | 958 | static int dir_split_leaf(struct inode *inode, const struct qstr *name) |
b3b94faa | 959 | { |
feaa7bba | 960 | struct gfs2_inode *dip = GFS2_I(inode); |
b3b94faa DT |
961 | struct buffer_head *nbh, *obh, *dibh; |
962 | struct gfs2_leaf *nleaf, *oleaf; | |
4da3c646 | 963 | struct gfs2_dirent *dent = NULL, *prev = NULL, *next = NULL, *new; |
cd915493 | 964 | u32 start, len, half_len, divider; |
b44b84d7 AV |
965 | u64 bn, leaf_no; |
966 | __be64 *lp; | |
cd915493 | 967 | u32 index; |
b3b94faa DT |
968 | int x, moved = 0; |
969 | int error; | |
970 | ||
9a004508 | 971 | index = name->hash >> (32 - dip->i_depth); |
c752666c SW |
972 | error = get_leaf_nr(dip, index, &leaf_no); |
973 | if (error) | |
974 | return error; | |
b3b94faa DT |
975 | |
976 | /* Get the old leaf block */ | |
b3b94faa DT |
977 | error = get_leaf(dip, leaf_no, &obh); |
978 | if (error) | |
e90deff5 | 979 | return error; |
b3b94faa | 980 | |
b3b94faa | 981 | oleaf = (struct gfs2_leaf *)obh->b_data; |
9a004508 | 982 | if (dip->i_depth == be16_to_cpu(oleaf->lf_depth)) { |
e90deff5 SW |
983 | brelse(obh); |
984 | return 1; /* can't split */ | |
985 | } | |
986 | ||
350a9b0a | 987 | gfs2_trans_add_meta(dip->i_gl, obh); |
b3b94faa | 988 | |
c752666c SW |
989 | nleaf = new_leaf(inode, &nbh, be16_to_cpu(oleaf->lf_depth) + 1); |
990 | if (!nleaf) { | |
991 | brelse(obh); | |
992 | return -ENOSPC; | |
993 | } | |
994 | bn = nbh->b_blocknr; | |
b3b94faa | 995 | |
c752666c | 996 | /* Compute the start and len of leaf pointers in the hash table. */ |
9a004508 | 997 | len = 1 << (dip->i_depth - be16_to_cpu(oleaf->lf_depth)); |
b3b94faa DT |
998 | half_len = len >> 1; |
999 | if (!half_len) { | |
d77d1b58 JP |
1000 | pr_warn("i_depth %u lf_depth %u index %u\n", |
1001 | dip->i_depth, be16_to_cpu(oleaf->lf_depth), index); | |
b3b94faa DT |
1002 | gfs2_consist_inode(dip); |
1003 | error = -EIO; | |
1004 | goto fail_brelse; | |
1005 | } | |
1006 | ||
1007 | start = (index & ~(len - 1)); | |
1008 | ||
1009 | /* Change the pointers. | |
1010 | Don't bother distinguishing stuffed from non-stuffed. | |
1011 | This code is complicated enough already. */ | |
4244b52e DR |
1012 | lp = kmalloc(half_len * sizeof(__be64), GFP_NOFS); |
1013 | if (!lp) { | |
1014 | error = -ENOMEM; | |
1015 | goto fail_brelse; | |
1016 | } | |
1017 | ||
b3b94faa | 1018 | /* Change the pointers */ |
b3b94faa DT |
1019 | for (x = 0; x < half_len; x++) |
1020 | lp[x] = cpu_to_be64(bn); | |
1021 | ||
17d539f0 SW |
1022 | gfs2_dir_hash_inval(dip); |
1023 | ||
cd915493 SW |
1024 | error = gfs2_dir_write_data(dip, (char *)lp, start * sizeof(u64), |
1025 | half_len * sizeof(u64)); | |
1026 | if (error != half_len * sizeof(u64)) { | |
b3b94faa DT |
1027 | if (error >= 0) |
1028 | error = -EIO; | |
1029 | goto fail_lpfree; | |
1030 | } | |
1031 | ||
1032 | kfree(lp); | |
1033 | ||
1034 | /* Compute the divider */ | |
9a004508 | 1035 | divider = (start + half_len) << (32 - dip->i_depth); |
b3b94faa DT |
1036 | |
1037 | /* Copy the entries */ | |
1579343a | 1038 | dent = (struct gfs2_dirent *)(obh->b_data + sizeof(struct gfs2_leaf)); |
b3b94faa DT |
1039 | |
1040 | do { | |
1041 | next = dent; | |
1042 | if (dirent_next(dip, obh, &next)) | |
1043 | next = NULL; | |
1044 | ||
5e7d65cd | 1045 | if (!gfs2_dirent_sentinel(dent) && |
b3b94faa | 1046 | be32_to_cpu(dent->de_hash) < divider) { |
c752666c SW |
1047 | struct qstr str; |
1048 | str.name = (char*)(dent+1); | |
1049 | str.len = be16_to_cpu(dent->de_name_len); | |
1050 | str.hash = be32_to_cpu(dent->de_hash); | |
71b86f56 | 1051 | new = gfs2_dirent_alloc(inode, nbh, &str); |
c752666c SW |
1052 | if (IS_ERR(new)) { |
1053 | error = PTR_ERR(new); | |
1054 | break; | |
1055 | } | |
b3b94faa DT |
1056 | |
1057 | new->de_inum = dent->de_inum; /* No endian worries */ | |
b3b94faa | 1058 | new->de_type = dent->de_type; /* No endian worries */ |
bb16b342 | 1059 | be16_add_cpu(&nleaf->lf_entries, 1); |
b3b94faa DT |
1060 | |
1061 | dirent_del(dip, obh, prev, dent); | |
1062 | ||
1063 | if (!oleaf->lf_entries) | |
1064 | gfs2_consist_inode(dip); | |
bb16b342 | 1065 | be16_add_cpu(&oleaf->lf_entries, -1); |
b3b94faa DT |
1066 | |
1067 | if (!prev) | |
1068 | prev = dent; | |
1069 | ||
1070 | moved = 1; | |
c752666c | 1071 | } else { |
b3b94faa | 1072 | prev = dent; |
c752666c | 1073 | } |
b3b94faa | 1074 | dent = next; |
c752666c | 1075 | } while (dent); |
b3b94faa | 1076 | |
c752666c | 1077 | oleaf->lf_depth = nleaf->lf_depth; |
b3b94faa DT |
1078 | |
1079 | error = gfs2_meta_inode_buffer(dip, &dibh); | |
feaa7bba | 1080 | if (!gfs2_assert_withdraw(GFS2_SB(&dip->i_inode), !error)) { |
350a9b0a | 1081 | gfs2_trans_add_meta(dip->i_gl, dibh); |
77658aad | 1082 | gfs2_add_inode_blocks(&dip->i_inode, 1); |
539e5d6b | 1083 | gfs2_dinode_out(dip, dibh->b_data); |
b3b94faa DT |
1084 | brelse(dibh); |
1085 | } | |
1086 | ||
1087 | brelse(obh); | |
1088 | brelse(nbh); | |
1089 | ||
1090 | return error; | |
1091 | ||
e90deff5 | 1092 | fail_lpfree: |
b3b94faa DT |
1093 | kfree(lp); |
1094 | ||
e90deff5 | 1095 | fail_brelse: |
b3b94faa | 1096 | brelse(obh); |
b3b94faa DT |
1097 | brelse(nbh); |
1098 | return error; | |
1099 | } | |
1100 | ||
1101 | /** | |
1102 | * dir_double_exhash - Double size of ExHash table | |
1103 | * @dip: The GFS2 dinode | |
1104 | * | |
1105 | * Returns: 0 on success, error code on failure | |
1106 | */ | |
1107 | ||
1108 | static int dir_double_exhash(struct gfs2_inode *dip) | |
1109 | { | |
b3b94faa | 1110 | struct buffer_head *dibh; |
cd915493 | 1111 | u32 hsize; |
17d539f0 SW |
1112 | u32 hsize_bytes; |
1113 | __be64 *hc; | |
1114 | __be64 *hc2, *h; | |
b3b94faa DT |
1115 | int x; |
1116 | int error = 0; | |
1117 | ||
9a004508 | 1118 | hsize = 1 << dip->i_depth; |
17d539f0 | 1119 | hsize_bytes = hsize * sizeof(__be64); |
b3b94faa | 1120 | |
17d539f0 SW |
1121 | hc = gfs2_dir_get_hash_table(dip); |
1122 | if (IS_ERR(hc)) | |
1123 | return PTR_ERR(hc); | |
b3b94faa | 1124 | |
512cbf02 | 1125 | hc2 = kmalloc(hsize_bytes * 2, GFP_NOFS | __GFP_NOWARN); |
e8830d88 BP |
1126 | if (hc2 == NULL) |
1127 | hc2 = __vmalloc(hsize_bytes * 2, GFP_NOFS, PAGE_KERNEL); | |
1128 | ||
17d539f0 | 1129 | if (!hc2) |
4244b52e | 1130 | return -ENOMEM; |
b3b94faa | 1131 | |
512cbf02 | 1132 | h = hc2; |
17d539f0 SW |
1133 | error = gfs2_meta_inode_buffer(dip, &dibh); |
1134 | if (error) | |
1135 | goto out_kfree; | |
b3b94faa | 1136 | |
17d539f0 SW |
1137 | for (x = 0; x < hsize; x++) { |
1138 | *h++ = *hc; | |
1139 | *h++ = *hc; | |
1140 | hc++; | |
b3b94faa DT |
1141 | } |
1142 | ||
17d539f0 SW |
1143 | error = gfs2_dir_write_data(dip, (char *)hc2, 0, hsize_bytes * 2); |
1144 | if (error != (hsize_bytes * 2)) | |
1145 | goto fail; | |
b3b94faa | 1146 | |
17d539f0 SW |
1147 | gfs2_dir_hash_inval(dip); |
1148 | dip->i_hash_cache = hc2; | |
1149 | dip->i_depth++; | |
1150 | gfs2_dinode_out(dip, dibh->b_data); | |
1151 | brelse(dibh); | |
1152 | return 0; | |
b3b94faa | 1153 | |
a91ea69f | 1154 | fail: |
17d539f0 SW |
1155 | /* Replace original hash table & size */ |
1156 | gfs2_dir_write_data(dip, (char *)hc, 0, hsize_bytes); | |
1157 | i_size_write(&dip->i_inode, hsize_bytes); | |
1158 | gfs2_dinode_out(dip, dibh->b_data); | |
1159 | brelse(dibh); | |
1160 | out_kfree: | |
3cdcf63e | 1161 | kvfree(hc2); |
b3b94faa DT |
1162 | return error; |
1163 | } | |
1164 | ||
1165 | /** | |
1166 | * compare_dents - compare directory entries by hash value | |
1167 | * @a: first dent | |
1168 | * @b: second dent | |
1169 | * | |
1170 | * When comparing the hash entries of @a to @b: | |
1171 | * gt: returns 1 | |
1172 | * lt: returns -1 | |
1173 | * eq: returns 0 | |
1174 | */ | |
1175 | ||
1176 | static int compare_dents(const void *a, const void *b) | |
1177 | { | |
2bdbc5d7 | 1178 | const struct gfs2_dirent *dent_a, *dent_b; |
cd915493 | 1179 | u32 hash_a, hash_b; |
b3b94faa DT |
1180 | int ret = 0; |
1181 | ||
2bdbc5d7 | 1182 | dent_a = *(const struct gfs2_dirent **)a; |
c752666c | 1183 | hash_a = be32_to_cpu(dent_a->de_hash); |
b3b94faa | 1184 | |
2bdbc5d7 | 1185 | dent_b = *(const struct gfs2_dirent **)b; |
c752666c | 1186 | hash_b = be32_to_cpu(dent_b->de_hash); |
b3b94faa DT |
1187 | |
1188 | if (hash_a > hash_b) | |
1189 | ret = 1; | |
1190 | else if (hash_a < hash_b) | |
1191 | ret = -1; | |
1192 | else { | |
4dd651ad SW |
1193 | unsigned int len_a = be16_to_cpu(dent_a->de_name_len); |
1194 | unsigned int len_b = be16_to_cpu(dent_b->de_name_len); | |
b3b94faa DT |
1195 | |
1196 | if (len_a > len_b) | |
1197 | ret = 1; | |
1198 | else if (len_a < len_b) | |
1199 | ret = -1; | |
1200 | else | |
2bdbc5d7 | 1201 | ret = memcmp(dent_a + 1, dent_b + 1, len_a); |
b3b94faa DT |
1202 | } |
1203 | ||
1204 | return ret; | |
1205 | } | |
1206 | ||
1207 | /** | |
1208 | * do_filldir_main - read out directory entries | |
1209 | * @dip: The GFS2 inode | |
d81a8ef5 | 1210 | * @ctx: what to feed the entries to |
b3b94faa DT |
1211 | * @darr: an array of struct gfs2_dirent pointers to read |
1212 | * @entries: the number of entries in darr | |
1213 | * @copied: pointer to int that's non-zero if a entry has been copied out | |
1214 | * | |
1215 | * Jump through some hoops to make sure that if there are hash collsions, | |
1216 | * they are read out at the beginning of a buffer. We want to minimize | |
1217 | * the possibility that they will fall into different readdir buffers or | |
1218 | * that someone will want to seek to that location. | |
1219 | * | |
d81a8ef5 | 1220 | * Returns: errno, >0 if the actor tells you to stop |
b3b94faa DT |
1221 | */ |
1222 | ||
d81a8ef5 | 1223 | static int do_filldir_main(struct gfs2_inode *dip, struct dir_context *ctx, |
cd915493 | 1224 | const struct gfs2_dirent **darr, u32 entries, |
b3b94faa DT |
1225 | int *copied) |
1226 | { | |
71b86f56 | 1227 | const struct gfs2_dirent *dent, *dent_next; |
cd915493 | 1228 | u64 off, off_next; |
b3b94faa DT |
1229 | unsigned int x, y; |
1230 | int run = 0; | |
b3b94faa DT |
1231 | |
1232 | sort(darr, entries, sizeof(struct gfs2_dirent *), compare_dents, NULL); | |
1233 | ||
1234 | dent_next = darr[0]; | |
1235 | off_next = be32_to_cpu(dent_next->de_hash); | |
1236 | off_next = gfs2_disk_hash2offset(off_next); | |
1237 | ||
1238 | for (x = 0, y = 1; x < entries; x++, y++) { | |
1239 | dent = dent_next; | |
1240 | off = off_next; | |
1241 | ||
1242 | if (y < entries) { | |
1243 | dent_next = darr[y]; | |
1244 | off_next = be32_to_cpu(dent_next->de_hash); | |
1245 | off_next = gfs2_disk_hash2offset(off_next); | |
1246 | ||
d81a8ef5 | 1247 | if (off < ctx->pos) |
b3b94faa | 1248 | continue; |
d81a8ef5 | 1249 | ctx->pos = off; |
b3b94faa DT |
1250 | |
1251 | if (off_next == off) { | |
1252 | if (*copied && !run) | |
1253 | return 1; | |
1254 | run = 1; | |
1255 | } else | |
1256 | run = 0; | |
1257 | } else { | |
d81a8ef5 | 1258 | if (off < ctx->pos) |
b3b94faa | 1259 | continue; |
d81a8ef5 | 1260 | ctx->pos = off; |
b3b94faa DT |
1261 | } |
1262 | ||
d81a8ef5 | 1263 | if (!dir_emit(ctx, (const char *)(dent + 1), |
4dd651ad | 1264 | be16_to_cpu(dent->de_name_len), |
d81a8ef5 AV |
1265 | be64_to_cpu(dent->de_inum.no_addr), |
1266 | be16_to_cpu(dent->de_type))) | |
b3b94faa DT |
1267 | return 1; |
1268 | ||
1269 | *copied = 1; | |
1270 | } | |
1271 | ||
d81a8ef5 | 1272 | /* Increment the ctx->pos by one, so the next time we come into the |
b3b94faa DT |
1273 | do_filldir fxn, we get the next entry instead of the last one in the |
1274 | current leaf */ | |
1275 | ||
d81a8ef5 | 1276 | ctx->pos++; |
b3b94faa DT |
1277 | |
1278 | return 0; | |
1279 | } | |
1280 | ||
d2a97a4e SW |
1281 | static void *gfs2_alloc_sort_buffer(unsigned size) |
1282 | { | |
1283 | void *ptr = NULL; | |
1284 | ||
1285 | if (size < KMALLOC_MAX_SIZE) | |
1286 | ptr = kmalloc(size, GFP_NOFS | __GFP_NOWARN); | |
1287 | if (!ptr) | |
1288 | ptr = __vmalloc(size, GFP_NOFS, PAGE_KERNEL); | |
1289 | return ptr; | |
1290 | } | |
1291 | ||
d81a8ef5 AV |
1292 | static int gfs2_dir_read_leaf(struct inode *inode, struct dir_context *ctx, |
1293 | int *copied, unsigned *depth, | |
3699e3a4 | 1294 | u64 leaf_no) |
b3b94faa | 1295 | { |
feaa7bba | 1296 | struct gfs2_inode *ip = GFS2_I(inode); |
bdd19a22 | 1297 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
71b86f56 SW |
1298 | struct buffer_head *bh; |
1299 | struct gfs2_leaf *lf; | |
bdd19a22 | 1300 | unsigned entries = 0, entries2 = 0; |
71b86f56 SW |
1301 | unsigned leaves = 0; |
1302 | const struct gfs2_dirent **darr, *dent; | |
1303 | struct dirent_gather g; | |
1304 | struct buffer_head **larr; | |
1305 | int leaf = 0; | |
1306 | int error, i; | |
1307 | u64 lfn = leaf_no; | |
b3b94faa | 1308 | |
b3b94faa | 1309 | do { |
71b86f56 | 1310 | error = get_leaf(ip, lfn, &bh); |
b3b94faa | 1311 | if (error) |
71b86f56 SW |
1312 | goto out; |
1313 | lf = (struct gfs2_leaf *)bh->b_data; | |
1314 | if (leaves == 0) | |
1315 | *depth = be16_to_cpu(lf->lf_depth); | |
1316 | entries += be16_to_cpu(lf->lf_entries); | |
1317 | leaves++; | |
1318 | lfn = be64_to_cpu(lf->lf_next); | |
1319 | brelse(bh); | |
1320 | } while(lfn); | |
b3b94faa DT |
1321 | |
1322 | if (!entries) | |
1323 | return 0; | |
1324 | ||
71b86f56 | 1325 | error = -ENOMEM; |
bdd19a22 SW |
1326 | /* |
1327 | * The extra 99 entries are not normally used, but are a buffer | |
1328 | * zone in case the number of entries in the leaf is corrupt. | |
1329 | * 99 is the maximum number of entries that can fit in a single | |
1330 | * leaf block. | |
1331 | */ | |
d2a97a4e | 1332 | larr = gfs2_alloc_sort_buffer((leaves + entries + 99) * sizeof(void *)); |
71b86f56 SW |
1333 | if (!larr) |
1334 | goto out; | |
1335 | darr = (const struct gfs2_dirent **)(larr + leaves); | |
1336 | g.pdent = darr; | |
1337 | g.offset = 0; | |
1338 | lfn = leaf_no; | |
b3b94faa | 1339 | |
71b86f56 SW |
1340 | do { |
1341 | error = get_leaf(ip, lfn, &bh); | |
b3b94faa | 1342 | if (error) |
d2a97a4e | 1343 | goto out_free; |
71b86f56 SW |
1344 | lf = (struct gfs2_leaf *)bh->b_data; |
1345 | lfn = be64_to_cpu(lf->lf_next); | |
1346 | if (lf->lf_entries) { | |
bdd19a22 | 1347 | entries2 += be16_to_cpu(lf->lf_entries); |
71b86f56 SW |
1348 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, |
1349 | gfs2_dirent_gather, NULL, &g); | |
1350 | error = PTR_ERR(dent); | |
bdd19a22 | 1351 | if (IS_ERR(dent)) |
d2a97a4e | 1352 | goto out_free; |
bdd19a22 | 1353 | if (entries2 != g.offset) { |
f391a4ea AM |
1354 | fs_warn(sdp, "Number of entries corrupt in dir " |
1355 | "leaf %llu, entries2 (%u) != " | |
1356 | "g.offset (%u)\n", | |
1357 | (unsigned long long)bh->b_blocknr, | |
1358 | entries2, g.offset); | |
bdd19a22 SW |
1359 | |
1360 | error = -EIO; | |
d2a97a4e | 1361 | goto out_free; |
71b86f56 SW |
1362 | } |
1363 | error = 0; | |
1364 | larr[leaf++] = bh; | |
b3b94faa | 1365 | } else { |
71b86f56 | 1366 | brelse(bh); |
b3b94faa | 1367 | } |
71b86f56 | 1368 | } while(lfn); |
b3b94faa | 1369 | |
bdd19a22 | 1370 | BUG_ON(entries2 != entries); |
d81a8ef5 | 1371 | error = do_filldir_main(ip, ctx, darr, entries, copied); |
d2a97a4e | 1372 | out_free: |
71b86f56 SW |
1373 | for(i = 0; i < leaf; i++) |
1374 | brelse(larr[i]); | |
3cdcf63e | 1375 | kvfree(larr); |
71b86f56 | 1376 | out: |
b3b94faa DT |
1377 | return error; |
1378 | } | |
1379 | ||
79c4c379 SW |
1380 | /** |
1381 | * gfs2_dir_readahead - Issue read-ahead requests for leaf blocks. | |
dfe4d34b BP |
1382 | * |
1383 | * Note: we can't calculate each index like dir_e_read can because we don't | |
1384 | * have the leaf, and therefore we don't have the depth, and therefore we | |
1385 | * don't have the length. So we have to just read enough ahead to make up | |
79c4c379 SW |
1386 | * for the loss of information. |
1387 | */ | |
dfe4d34b BP |
1388 | static void gfs2_dir_readahead(struct inode *inode, unsigned hsize, u32 index, |
1389 | struct file_ra_state *f_ra) | |
1390 | { | |
1391 | struct gfs2_inode *ip = GFS2_I(inode); | |
1392 | struct gfs2_glock *gl = ip->i_gl; | |
1393 | struct buffer_head *bh; | |
1394 | u64 blocknr = 0, last; | |
1395 | unsigned count; | |
1396 | ||
1397 | /* First check if we've already read-ahead for the whole range. */ | |
79c4c379 | 1398 | if (index + MAX_RA_BLOCKS < f_ra->start) |
dfe4d34b BP |
1399 | return; |
1400 | ||
1401 | f_ra->start = max((pgoff_t)index, f_ra->start); | |
1402 | for (count = 0; count < MAX_RA_BLOCKS; count++) { | |
1403 | if (f_ra->start >= hsize) /* if exceeded the hash table */ | |
1404 | break; | |
1405 | ||
1406 | last = blocknr; | |
1407 | blocknr = be64_to_cpu(ip->i_hash_cache[f_ra->start]); | |
1408 | f_ra->start++; | |
1409 | if (blocknr == last) | |
1410 | continue; | |
1411 | ||
1412 | bh = gfs2_getbuf(gl, blocknr, 1); | |
1413 | if (trylock_buffer(bh)) { | |
1414 | if (buffer_uptodate(bh)) { | |
1415 | unlock_buffer(bh); | |
1416 | brelse(bh); | |
1417 | continue; | |
1418 | } | |
1419 | bh->b_end_io = end_buffer_read_sync; | |
1420 | submit_bh(READA | REQ_META, bh); | |
1421 | continue; | |
1422 | } | |
1423 | brelse(bh); | |
1424 | } | |
1425 | } | |
17d539f0 | 1426 | |
b3b94faa | 1427 | /** |
c752666c SW |
1428 | * dir_e_read - Reads the entries from a directory into a filldir buffer |
1429 | * @dip: dinode pointer | |
d81a8ef5 | 1430 | * @ctx: actor to feed the entries to |
b3b94faa | 1431 | * |
c752666c | 1432 | * Returns: errno |
b3b94faa DT |
1433 | */ |
1434 | ||
d81a8ef5 AV |
1435 | static int dir_e_read(struct inode *inode, struct dir_context *ctx, |
1436 | struct file_ra_state *f_ra) | |
b3b94faa | 1437 | { |
feaa7bba | 1438 | struct gfs2_inode *dip = GFS2_I(inode); |
cd915493 | 1439 | u32 hsize, len = 0; |
cd915493 | 1440 | u32 hash, index; |
b44b84d7 | 1441 | __be64 *lp; |
c752666c SW |
1442 | int copied = 0; |
1443 | int error = 0; | |
4da3c646 | 1444 | unsigned depth = 0; |
b3b94faa | 1445 | |
9a004508 | 1446 | hsize = 1 << dip->i_depth; |
d81a8ef5 | 1447 | hash = gfs2_dir_offset2hash(ctx->pos); |
9a004508 | 1448 | index = hash >> (32 - dip->i_depth); |
b3b94faa | 1449 | |
79c4c379 | 1450 | if (dip->i_hash_cache == NULL) |
dfe4d34b | 1451 | f_ra->start = 0; |
17d539f0 SW |
1452 | lp = gfs2_dir_get_hash_table(dip); |
1453 | if (IS_ERR(lp)) | |
1454 | return PTR_ERR(lp); | |
b3b94faa | 1455 | |
dfe4d34b BP |
1456 | gfs2_dir_readahead(inode, hsize, index, f_ra); |
1457 | ||
b3b94faa | 1458 | while (index < hsize) { |
d81a8ef5 | 1459 | error = gfs2_dir_read_leaf(inode, ctx, |
71b86f56 | 1460 | &copied, &depth, |
17d539f0 | 1461 | be64_to_cpu(lp[index])); |
b3b94faa | 1462 | if (error) |
71b86f56 | 1463 | break; |
b3b94faa | 1464 | |
9a004508 | 1465 | len = 1 << (dip->i_depth - depth); |
b3b94faa DT |
1466 | index = (index & ~(len - 1)) + len; |
1467 | } | |
1468 | ||
71b86f56 SW |
1469 | if (error > 0) |
1470 | error = 0; | |
b3b94faa DT |
1471 | return error; |
1472 | } | |
1473 | ||
d81a8ef5 AV |
1474 | int gfs2_dir_read(struct inode *inode, struct dir_context *ctx, |
1475 | struct file_ra_state *f_ra) | |
b3b94faa | 1476 | { |
feaa7bba | 1477 | struct gfs2_inode *dip = GFS2_I(inode); |
bdd19a22 | 1478 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
71b86f56 SW |
1479 | struct dirent_gather g; |
1480 | const struct gfs2_dirent **darr, *dent; | |
b3b94faa DT |
1481 | struct buffer_head *dibh; |
1482 | int copied = 0; | |
1483 | int error; | |
1484 | ||
ad6203f2 | 1485 | if (!dip->i_entries) |
71b86f56 SW |
1486 | return 0; |
1487 | ||
383f01fb | 1488 | if (dip->i_diskflags & GFS2_DIF_EXHASH) |
d81a8ef5 | 1489 | return dir_e_read(inode, ctx, f_ra); |
71b86f56 | 1490 | |
b3b94faa DT |
1491 | if (!gfs2_is_stuffed(dip)) { |
1492 | gfs2_consist_inode(dip); | |
1493 | return -EIO; | |
1494 | } | |
1495 | ||
b3b94faa DT |
1496 | error = gfs2_meta_inode_buffer(dip, &dibh); |
1497 | if (error) | |
1498 | return error; | |
1499 | ||
71b86f56 | 1500 | error = -ENOMEM; |
bdd19a22 | 1501 | /* 96 is max number of dirents which can be stuffed into an inode */ |
16c5f06f | 1502 | darr = kmalloc(96 * sizeof(struct gfs2_dirent *), GFP_NOFS); |
71b86f56 SW |
1503 | if (darr) { |
1504 | g.pdent = darr; | |
1505 | g.offset = 0; | |
1506 | dent = gfs2_dirent_scan(inode, dibh->b_data, dibh->b_size, | |
1507 | gfs2_dirent_gather, NULL, &g); | |
1508 | if (IS_ERR(dent)) { | |
1509 | error = PTR_ERR(dent); | |
1510 | goto out; | |
1511 | } | |
ad6203f2 | 1512 | if (dip->i_entries != g.offset) { |
bdd19a22 | 1513 | fs_warn(sdp, "Number of entries corrupt in dir %llu, " |
ad6203f2 | 1514 | "ip->i_entries (%u) != g.offset (%u)\n", |
dbb7cae2 | 1515 | (unsigned long long)dip->i_no_addr, |
ad6203f2 | 1516 | dip->i_entries, |
bdd19a22 SW |
1517 | g.offset); |
1518 | error = -EIO; | |
1519 | goto out; | |
1520 | } | |
d81a8ef5 | 1521 | error = do_filldir_main(dip, ctx, darr, |
ad6203f2 | 1522 | dip->i_entries, &copied); |
71b86f56 SW |
1523 | out: |
1524 | kfree(darr); | |
1525 | } | |
1526 | ||
b3b94faa DT |
1527 | if (error > 0) |
1528 | error = 0; | |
1529 | ||
1530 | brelse(dibh); | |
1531 | ||
1532 | return error; | |
1533 | } | |
1534 | ||
b3b94faa DT |
1535 | /** |
1536 | * gfs2_dir_search - Search a directory | |
5a00f3cc SW |
1537 | * @dip: The GFS2 dir inode |
1538 | * @name: The name we are looking up | |
1539 | * @fail_on_exist: Fail if the name exists rather than looking it up | |
b3b94faa DT |
1540 | * |
1541 | * This routine searches a directory for a file or another directory. | |
1542 | * Assumes a glock is held on dip. | |
1543 | * | |
1544 | * Returns: errno | |
1545 | */ | |
1546 | ||
5a00f3cc SW |
1547 | struct inode *gfs2_dir_search(struct inode *dir, const struct qstr *name, |
1548 | bool fail_on_exist) | |
b3b94faa | 1549 | { |
c752666c SW |
1550 | struct buffer_head *bh; |
1551 | struct gfs2_dirent *dent; | |
5a00f3cc SW |
1552 | u64 addr, formal_ino; |
1553 | u16 dtype; | |
dbb7cae2 SW |
1554 | |
1555 | dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh); | |
1556 | if (dent) { | |
1557 | if (IS_ERR(dent)) | |
e231c2ee | 1558 | return ERR_CAST(dent); |
5a00f3cc SW |
1559 | dtype = be16_to_cpu(dent->de_type); |
1560 | addr = be64_to_cpu(dent->de_inum.no_addr); | |
1561 | formal_ino = be64_to_cpu(dent->de_inum.no_formal_ino); | |
dbb7cae2 | 1562 | brelse(bh); |
5a00f3cc SW |
1563 | if (fail_on_exist) |
1564 | return ERR_PTR(-EEXIST); | |
1565 | return gfs2_inode_lookup(dir->i_sb, dtype, addr, formal_ino, 0); | |
dbb7cae2 SW |
1566 | } |
1567 | return ERR_PTR(-ENOENT); | |
1568 | } | |
1569 | ||
1570 | int gfs2_dir_check(struct inode *dir, const struct qstr *name, | |
1571 | const struct gfs2_inode *ip) | |
1572 | { | |
1573 | struct buffer_head *bh; | |
1574 | struct gfs2_dirent *dent; | |
1575 | int ret = -ENOENT; | |
c752666c SW |
1576 | |
1577 | dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh); | |
1578 | if (dent) { | |
1579 | if (IS_ERR(dent)) | |
1580 | return PTR_ERR(dent); | |
dbb7cae2 SW |
1581 | if (ip) { |
1582 | if (be64_to_cpu(dent->de_inum.no_addr) != ip->i_no_addr) | |
1583 | goto out; | |
1584 | if (be64_to_cpu(dent->de_inum.no_formal_ino) != | |
1585 | ip->i_no_formal_ino) | |
1586 | goto out; | |
1587 | if (unlikely(IF2DT(ip->i_inode.i_mode) != | |
1588 | be16_to_cpu(dent->de_type))) { | |
1589 | gfs2_consist_inode(GFS2_I(dir)); | |
1590 | ret = -EIO; | |
1591 | goto out; | |
1592 | } | |
1593 | } | |
1594 | ret = 0; | |
1595 | out: | |
c752666c | 1596 | brelse(bh); |
c752666c | 1597 | } |
dbb7cae2 | 1598 | return ret; |
c752666c SW |
1599 | } |
1600 | ||
01bcb0de SW |
1601 | /** |
1602 | * dir_new_leaf - Add a new leaf onto hash chain | |
1603 | * @inode: The directory | |
1604 | * @name: The name we are adding | |
1605 | * | |
1606 | * This adds a new dir leaf onto an existing leaf when there is not | |
1607 | * enough space to add a new dir entry. This is a last resort after | |
1608 | * we've expanded the hash table to max size and also split existing | |
1609 | * leaf blocks, so it will only occur for very large directories. | |
1610 | * | |
1611 | * The dist parameter is set to 1 for leaf blocks directly attached | |
1612 | * to the hash table, 2 for one layer of indirection, 3 for two layers | |
1613 | * etc. We are thus able to tell the difference between an old leaf | |
1614 | * with dist set to zero (i.e. "don't know") and a new one where we | |
1615 | * set this information for debug/fsck purposes. | |
1616 | * | |
1617 | * Returns: 0 on success, or -ve on error | |
1618 | */ | |
1619 | ||
c752666c SW |
1620 | static int dir_new_leaf(struct inode *inode, const struct qstr *name) |
1621 | { | |
1622 | struct buffer_head *bh, *obh; | |
feaa7bba | 1623 | struct gfs2_inode *ip = GFS2_I(inode); |
c752666c | 1624 | struct gfs2_leaf *leaf, *oleaf; |
01bcb0de | 1625 | u32 dist = 1; |
b3b94faa | 1626 | int error; |
c752666c SW |
1627 | u32 index; |
1628 | u64 bn; | |
b3b94faa | 1629 | |
9a004508 | 1630 | index = name->hash >> (32 - ip->i_depth); |
c752666c SW |
1631 | error = get_first_leaf(ip, index, &obh); |
1632 | if (error) | |
1633 | return error; | |
1634 | do { | |
01bcb0de | 1635 | dist++; |
c752666c SW |
1636 | oleaf = (struct gfs2_leaf *)obh->b_data; |
1637 | bn = be64_to_cpu(oleaf->lf_next); | |
1638 | if (!bn) | |
1639 | break; | |
1640 | brelse(obh); | |
1641 | error = get_leaf(ip, bn, &obh); | |
1642 | if (error) | |
1643 | return error; | |
1644 | } while(1); | |
b3b94faa | 1645 | |
350a9b0a | 1646 | gfs2_trans_add_meta(ip->i_gl, obh); |
c752666c SW |
1647 | |
1648 | leaf = new_leaf(inode, &bh, be16_to_cpu(oleaf->lf_depth)); | |
1649 | if (!leaf) { | |
1650 | brelse(obh); | |
1651 | return -ENOSPC; | |
1652 | } | |
01bcb0de | 1653 | leaf->lf_dist = cpu_to_be32(dist); |
4d8012b6 | 1654 | oleaf->lf_next = cpu_to_be64(bh->b_blocknr); |
c752666c SW |
1655 | brelse(bh); |
1656 | brelse(obh); | |
1657 | ||
1658 | error = gfs2_meta_inode_buffer(ip, &bh); | |
1659 | if (error) | |
1660 | return error; | |
350a9b0a | 1661 | gfs2_trans_add_meta(ip->i_gl, bh); |
77658aad | 1662 | gfs2_add_inode_blocks(&ip->i_inode, 1); |
539e5d6b | 1663 | gfs2_dinode_out(ip, bh->b_data); |
c752666c SW |
1664 | brelse(bh); |
1665 | return 0; | |
b3b94faa DT |
1666 | } |
1667 | ||
44aaada9 SW |
1668 | static u16 gfs2_inode_ra_len(const struct gfs2_inode *ip) |
1669 | { | |
1670 | u64 where = ip->i_no_addr + 1; | |
1671 | if (ip->i_eattr == where) | |
1672 | return 1; | |
1673 | return 0; | |
1674 | } | |
1675 | ||
b3b94faa DT |
1676 | /** |
1677 | * gfs2_dir_add - Add new filename into directory | |
2b47dad8 SW |
1678 | * @inode: The directory inode |
1679 | * @name: The new name | |
1680 | * @nip: The GFS2 inode to be linked in to the directory | |
1681 | * @da: The directory addition info | |
1682 | * | |
1683 | * If the call to gfs2_diradd_alloc_required resulted in there being | |
1684 | * no need to allocate any new directory blocks, then it will contain | |
1685 | * a pointer to the directory entry and the bh in which it resides. We | |
1686 | * can use that without having to repeat the search. If there was no | |
1687 | * free space, then we must now create more space. | |
b3b94faa DT |
1688 | * |
1689 | * Returns: 0 on success, error code on failure | |
1690 | */ | |
1691 | ||
c752666c | 1692 | int gfs2_dir_add(struct inode *inode, const struct qstr *name, |
2b47dad8 | 1693 | const struct gfs2_inode *nip, struct gfs2_diradd *da) |
b3b94faa | 1694 | { |
feaa7bba | 1695 | struct gfs2_inode *ip = GFS2_I(inode); |
2b47dad8 SW |
1696 | struct buffer_head *bh = da->bh; |
1697 | struct gfs2_dirent *dent = da->dent; | |
01bcb0de | 1698 | struct timespec tv; |
c752666c | 1699 | struct gfs2_leaf *leaf; |
b3b94faa DT |
1700 | int error; |
1701 | ||
c752666c | 1702 | while(1) { |
2b47dad8 SW |
1703 | if (da->bh == NULL) { |
1704 | dent = gfs2_dirent_search(inode, name, | |
1705 | gfs2_dirent_find_space, &bh); | |
1706 | } | |
c752666c SW |
1707 | if (dent) { |
1708 | if (IS_ERR(dent)) | |
1709 | return PTR_ERR(dent); | |
1710 | dent = gfs2_init_dirent(inode, dent, name, bh); | |
dbb7cae2 | 1711 | gfs2_inum_out(nip, dent); |
3d6ecb7d | 1712 | dent->de_type = cpu_to_be16(IF2DT(nip->i_inode.i_mode)); |
44aaada9 | 1713 | dent->de_rahead = cpu_to_be16(gfs2_inode_ra_len(nip)); |
01bcb0de | 1714 | tv = CURRENT_TIME; |
383f01fb | 1715 | if (ip->i_diskflags & GFS2_DIF_EXHASH) { |
c752666c | 1716 | leaf = (struct gfs2_leaf *)bh->b_data; |
bb16b342 | 1717 | be16_add_cpu(&leaf->lf_entries, 1); |
01bcb0de SW |
1718 | leaf->lf_nsec = cpu_to_be32(tv.tv_nsec); |
1719 | leaf->lf_sec = cpu_to_be64(tv.tv_sec); | |
c752666c | 1720 | } |
2b47dad8 SW |
1721 | da->dent = NULL; |
1722 | da->bh = NULL; | |
c752666c | 1723 | brelse(bh); |
ad6203f2 | 1724 | ip->i_entries++; |
01bcb0de | 1725 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = tv; |
3d6ecb7d SW |
1726 | if (S_ISDIR(nip->i_inode.i_mode)) |
1727 | inc_nlink(&ip->i_inode); | |
343cd8f0 | 1728 | mark_inode_dirty(inode); |
c752666c SW |
1729 | error = 0; |
1730 | break; | |
1731 | } | |
383f01fb | 1732 | if (!(ip->i_diskflags & GFS2_DIF_EXHASH)) { |
c752666c SW |
1733 | error = dir_make_exhash(inode); |
1734 | if (error) | |
1735 | break; | |
1736 | continue; | |
1737 | } | |
1738 | error = dir_split_leaf(inode, name); | |
1739 | if (error == 0) | |
1740 | continue; | |
e90deff5 | 1741 | if (error < 0) |
c752666c | 1742 | break; |
9a004508 | 1743 | if (ip->i_depth < GFS2_DIR_MAX_DEPTH) { |
c752666c SW |
1744 | error = dir_double_exhash(ip); |
1745 | if (error) | |
1746 | break; | |
1747 | error = dir_split_leaf(inode, name); | |
e90deff5 | 1748 | if (error < 0) |
c752666c | 1749 | break; |
e90deff5 SW |
1750 | if (error == 0) |
1751 | continue; | |
c752666c SW |
1752 | } |
1753 | error = dir_new_leaf(inode, name); | |
1754 | if (!error) | |
1755 | continue; | |
1756 | error = -ENOSPC; | |
1757 | break; | |
1758 | } | |
b3b94faa DT |
1759 | return error; |
1760 | } | |
1761 | ||
c752666c | 1762 | |
b3b94faa DT |
1763 | /** |
1764 | * gfs2_dir_del - Delete a directory entry | |
1765 | * @dip: The GFS2 inode | |
1766 | * @filename: The filename | |
1767 | * | |
1768 | * Returns: 0 on success, error code on failure | |
1769 | */ | |
1770 | ||
855d23ce | 1771 | int gfs2_dir_del(struct gfs2_inode *dip, const struct dentry *dentry) |
b3b94faa | 1772 | { |
855d23ce | 1773 | const struct qstr *name = &dentry->d_name; |
c752666c SW |
1774 | struct gfs2_dirent *dent, *prev = NULL; |
1775 | struct buffer_head *bh; | |
01bcb0de | 1776 | struct timespec tv = CURRENT_TIME; |
b3b94faa | 1777 | |
c752666c SW |
1778 | /* Returns _either_ the entry (if its first in block) or the |
1779 | previous entry otherwise */ | |
feaa7bba | 1780 | dent = gfs2_dirent_search(&dip->i_inode, name, gfs2_dirent_prev, &bh); |
c752666c SW |
1781 | if (!dent) { |
1782 | gfs2_consist_inode(dip); | |
1783 | return -EIO; | |
1784 | } | |
1785 | if (IS_ERR(dent)) { | |
1786 | gfs2_consist_inode(dip); | |
1787 | return PTR_ERR(dent); | |
1788 | } | |
1789 | /* If not first in block, adjust pointers accordingly */ | |
71b86f56 | 1790 | if (gfs2_dirent_find(dent, name, NULL) == 0) { |
c752666c SW |
1791 | prev = dent; |
1792 | dent = (struct gfs2_dirent *)((char *)dent + be16_to_cpu(prev->de_rec_len)); | |
1793 | } | |
1794 | ||
1795 | dirent_del(dip, bh, prev, dent); | |
383f01fb | 1796 | if (dip->i_diskflags & GFS2_DIF_EXHASH) { |
c752666c SW |
1797 | struct gfs2_leaf *leaf = (struct gfs2_leaf *)bh->b_data; |
1798 | u16 entries = be16_to_cpu(leaf->lf_entries); | |
1799 | if (!entries) | |
1800 | gfs2_consist_inode(dip); | |
1801 | leaf->lf_entries = cpu_to_be16(--entries); | |
01bcb0de SW |
1802 | leaf->lf_nsec = cpu_to_be32(tv.tv_nsec); |
1803 | leaf->lf_sec = cpu_to_be64(tv.tv_sec); | |
c752666c | 1804 | } |
ed386507 | 1805 | brelse(bh); |
c752666c | 1806 | |
ad6203f2 | 1807 | if (!dip->i_entries) |
c752666c | 1808 | gfs2_consist_inode(dip); |
ad6203f2 | 1809 | dip->i_entries--; |
01bcb0de | 1810 | dip->i_inode.i_mtime = dip->i_inode.i_ctime = tv; |
855d23ce SW |
1811 | if (S_ISDIR(dentry->d_inode->i_mode)) |
1812 | drop_nlink(&dip->i_inode); | |
feaa7bba | 1813 | mark_inode_dirty(&dip->i_inode); |
b3b94faa | 1814 | |
ab9bbda0 | 1815 | return 0; |
b3b94faa DT |
1816 | } |
1817 | ||
b3b94faa DT |
1818 | /** |
1819 | * gfs2_dir_mvino - Change inode number of directory entry | |
1820 | * @dip: The GFS2 inode | |
1821 | * @filename: | |
1822 | * @new_inode: | |
1823 | * | |
1824 | * This routine changes the inode number of a directory entry. It's used | |
1825 | * by rename to change ".." when a directory is moved. | |
1826 | * Assumes a glock is held on dvp. | |
1827 | * | |
1828 | * Returns: errno | |
1829 | */ | |
1830 | ||
c752666c | 1831 | int gfs2_dir_mvino(struct gfs2_inode *dip, const struct qstr *filename, |
dbb7cae2 | 1832 | const struct gfs2_inode *nip, unsigned int new_type) |
b3b94faa | 1833 | { |
c752666c SW |
1834 | struct buffer_head *bh; |
1835 | struct gfs2_dirent *dent; | |
b3b94faa DT |
1836 | int error; |
1837 | ||
feaa7bba | 1838 | dent = gfs2_dirent_search(&dip->i_inode, filename, gfs2_dirent_find, &bh); |
c752666c SW |
1839 | if (!dent) { |
1840 | gfs2_consist_inode(dip); | |
1841 | return -EIO; | |
1842 | } | |
1843 | if (IS_ERR(dent)) | |
1844 | return PTR_ERR(dent); | |
b3b94faa | 1845 | |
350a9b0a | 1846 | gfs2_trans_add_meta(dip->i_gl, bh); |
dbb7cae2 | 1847 | gfs2_inum_out(nip, dent); |
c752666c SW |
1848 | dent->de_type = cpu_to_be16(new_type); |
1849 | ||
383f01fb | 1850 | if (dip->i_diskflags & GFS2_DIF_EXHASH) { |
c752666c SW |
1851 | brelse(bh); |
1852 | error = gfs2_meta_inode_buffer(dip, &bh); | |
1853 | if (error) | |
1854 | return error; | |
350a9b0a | 1855 | gfs2_trans_add_meta(dip->i_gl, bh); |
c752666c SW |
1856 | } |
1857 | ||
4bd91ba1 | 1858 | dip->i_inode.i_mtime = dip->i_inode.i_ctime = CURRENT_TIME; |
539e5d6b | 1859 | gfs2_dinode_out(dip, bh->b_data); |
c752666c SW |
1860 | brelse(bh); |
1861 | return 0; | |
b3b94faa DT |
1862 | } |
1863 | ||
b3b94faa DT |
1864 | /** |
1865 | * leaf_dealloc - Deallocate a directory leaf | |
1866 | * @dip: the directory | |
1867 | * @index: the hash table offset in the directory | |
1868 | * @len: the number of pointers to this leaf | |
1869 | * @leaf_no: the leaf number | |
ec038c82 | 1870 | * @leaf_bh: buffer_head for the starting leaf |
d24a7a43 | 1871 | * last_dealloc: 1 if this is the final dealloc for the leaf, else 0 |
b3b94faa DT |
1872 | * |
1873 | * Returns: errno | |
1874 | */ | |
1875 | ||
cd915493 | 1876 | static int leaf_dealloc(struct gfs2_inode *dip, u32 index, u32 len, |
ec038c82 BP |
1877 | u64 leaf_no, struct buffer_head *leaf_bh, |
1878 | int last_dealloc) | |
b3b94faa | 1879 | { |
feaa7bba | 1880 | struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); |
c752666c | 1881 | struct gfs2_leaf *tmp_leaf; |
b3b94faa DT |
1882 | struct gfs2_rgrp_list rlist; |
1883 | struct buffer_head *bh, *dibh; | |
cd915493 | 1884 | u64 blk, nblk; |
b3b94faa DT |
1885 | unsigned int rg_blocks = 0, l_blocks = 0; |
1886 | char *ht; | |
cd915493 | 1887 | unsigned int x, size = len * sizeof(u64); |
b3b94faa DT |
1888 | int error; |
1889 | ||
5e2f7d61 BP |
1890 | error = gfs2_rindex_update(sdp); |
1891 | if (error) | |
1892 | return error; | |
1893 | ||
b3b94faa DT |
1894 | memset(&rlist, 0, sizeof(struct gfs2_rgrp_list)); |
1895 | ||
8be04b93 | 1896 | ht = kzalloc(size, GFP_NOFS | __GFP_NOWARN); |
e8830d88 BP |
1897 | if (ht == NULL) |
1898 | ht = vzalloc(size); | |
b3b94faa DT |
1899 | if (!ht) |
1900 | return -ENOMEM; | |
1901 | ||
f4108a60 | 1902 | error = gfs2_quota_hold(dip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE); |
b3b94faa | 1903 | if (error) |
5407e242 | 1904 | goto out; |
b3b94faa | 1905 | |
b3b94faa | 1906 | /* Count the number of leaves */ |
ec038c82 | 1907 | bh = leaf_bh; |
b3b94faa | 1908 | |
c752666c | 1909 | for (blk = leaf_no; blk; blk = nblk) { |
ec038c82 BP |
1910 | if (blk != leaf_no) { |
1911 | error = get_leaf(dip, blk, &bh); | |
1912 | if (error) | |
1913 | goto out_rlist; | |
1914 | } | |
c752666c SW |
1915 | tmp_leaf = (struct gfs2_leaf *)bh->b_data; |
1916 | nblk = be64_to_cpu(tmp_leaf->lf_next); | |
ec038c82 BP |
1917 | if (blk != leaf_no) |
1918 | brelse(bh); | |
b3b94faa | 1919 | |
70b0c365 | 1920 | gfs2_rlist_add(dip, &rlist, blk); |
b3b94faa DT |
1921 | l_blocks++; |
1922 | } | |
1923 | ||
fe6c991c | 1924 | gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE); |
b3b94faa DT |
1925 | |
1926 | for (x = 0; x < rlist.rl_rgrps; x++) { | |
1927 | struct gfs2_rgrpd *rgd; | |
5c676f6d | 1928 | rgd = rlist.rl_ghs[x].gh_gl->gl_object; |
bb8d8a6f | 1929 | rg_blocks += rgd->rd_length; |
b3b94faa DT |
1930 | } |
1931 | ||
1932 | error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs); | |
1933 | if (error) | |
1934 | goto out_rlist; | |
1935 | ||
1936 | error = gfs2_trans_begin(sdp, | |
5c676f6d | 1937 | rg_blocks + (DIV_ROUND_UP(size, sdp->sd_jbsize) + 1) + |
b3b94faa DT |
1938 | RES_DINODE + RES_STATFS + RES_QUOTA, l_blocks); |
1939 | if (error) | |
1940 | goto out_rg_gunlock; | |
1941 | ||
ec038c82 BP |
1942 | bh = leaf_bh; |
1943 | ||
c752666c | 1944 | for (blk = leaf_no; blk; blk = nblk) { |
ec038c82 BP |
1945 | if (blk != leaf_no) { |
1946 | error = get_leaf(dip, blk, &bh); | |
1947 | if (error) | |
1948 | goto out_end_trans; | |
1949 | } | |
c752666c SW |
1950 | tmp_leaf = (struct gfs2_leaf *)bh->b_data; |
1951 | nblk = be64_to_cpu(tmp_leaf->lf_next); | |
ec038c82 BP |
1952 | if (blk != leaf_no) |
1953 | brelse(bh); | |
b3b94faa DT |
1954 | |
1955 | gfs2_free_meta(dip, blk, 1); | |
77658aad | 1956 | gfs2_add_inode_blocks(&dip->i_inode, -1); |
b3b94faa DT |
1957 | } |
1958 | ||
cd915493 | 1959 | error = gfs2_dir_write_data(dip, ht, index * sizeof(u64), size); |
b3b94faa DT |
1960 | if (error != size) { |
1961 | if (error >= 0) | |
1962 | error = -EIO; | |
1963 | goto out_end_trans; | |
1964 | } | |
1965 | ||
1966 | error = gfs2_meta_inode_buffer(dip, &dibh); | |
1967 | if (error) | |
1968 | goto out_end_trans; | |
1969 | ||
350a9b0a | 1970 | gfs2_trans_add_meta(dip->i_gl, dibh); |
d24a7a43 BP |
1971 | /* On the last dealloc, make this a regular file in case we crash. |
1972 | (We don't want to free these blocks a second time.) */ | |
1973 | if (last_dealloc) | |
1974 | dip->i_inode.i_mode = S_IFREG; | |
539e5d6b | 1975 | gfs2_dinode_out(dip, dibh->b_data); |
b3b94faa DT |
1976 | brelse(dibh); |
1977 | ||
a91ea69f | 1978 | out_end_trans: |
b3b94faa | 1979 | gfs2_trans_end(sdp); |
a91ea69f | 1980 | out_rg_gunlock: |
b3b94faa | 1981 | gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs); |
a91ea69f | 1982 | out_rlist: |
b3b94faa | 1983 | gfs2_rlist_free(&rlist); |
b3b94faa | 1984 | gfs2_quota_unhold(dip); |
182fe5ab | 1985 | out: |
3cdcf63e | 1986 | kvfree(ht); |
b3b94faa DT |
1987 | return error; |
1988 | } | |
1989 | ||
1990 | /** | |
1991 | * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory | |
1992 | * @dip: the directory | |
1993 | * | |
1994 | * Dealloc all on-disk directory leaves to FREEMETA state | |
1995 | * Change on-disk inode type to "regular file" | |
1996 | * | |
1997 | * Returns: errno | |
1998 | */ | |
1999 | ||
2000 | int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip) | |
2001 | { | |
556bb179 BP |
2002 | struct buffer_head *bh; |
2003 | struct gfs2_leaf *leaf; | |
2004 | u32 hsize, len; | |
556bb179 BP |
2005 | u32 index = 0, next_index; |
2006 | __be64 *lp; | |
2007 | u64 leaf_no; | |
2008 | int error = 0, last; | |
2009 | ||
2010 | hsize = 1 << dip->i_depth; | |
556bb179 | 2011 | |
17d539f0 SW |
2012 | lp = gfs2_dir_get_hash_table(dip); |
2013 | if (IS_ERR(lp)) | |
2014 | return PTR_ERR(lp); | |
556bb179 BP |
2015 | |
2016 | while (index < hsize) { | |
17d539f0 | 2017 | leaf_no = be64_to_cpu(lp[index]); |
556bb179 BP |
2018 | if (leaf_no) { |
2019 | error = get_leaf(dip, leaf_no, &bh); | |
2020 | if (error) | |
2021 | goto out; | |
2022 | leaf = (struct gfs2_leaf *)bh->b_data; | |
2023 | len = 1 << (dip->i_depth - be16_to_cpu(leaf->lf_depth)); | |
2024 | ||
2025 | next_index = (index & ~(len - 1)) + len; | |
2026 | last = ((next_index >= hsize) ? 1 : 0); | |
2027 | error = leaf_dealloc(dip, index, len, leaf_no, bh, | |
2028 | last); | |
2029 | brelse(bh); | |
2030 | if (error) | |
2031 | goto out; | |
2032 | index = next_index; | |
2033 | } else | |
2034 | index++; | |
2035 | } | |
2036 | ||
2037 | if (index != hsize) { | |
2038 | gfs2_consist_inode(dip); | |
2039 | error = -EIO; | |
2040 | } | |
2041 | ||
2042 | out: | |
556bb179 BP |
2043 | |
2044 | return error; | |
b3b94faa DT |
2045 | } |
2046 | ||
2047 | /** | |
2048 | * gfs2_diradd_alloc_required - find if adding entry will require an allocation | |
2049 | * @ip: the file being written to | |
2050 | * @filname: the filename that's going to be added | |
3c1c0ae1 | 2051 | * @da: The structure to return dir alloc info |
b3b94faa | 2052 | * |
3c1c0ae1 | 2053 | * Returns: 0 if ok, -ve on error |
b3b94faa DT |
2054 | */ |
2055 | ||
3c1c0ae1 SW |
2056 | int gfs2_diradd_alloc_required(struct inode *inode, const struct qstr *name, |
2057 | struct gfs2_diradd *da) | |
b3b94faa | 2058 | { |
22b5a6c0 | 2059 | struct gfs2_inode *ip = GFS2_I(inode); |
3c1c0ae1 | 2060 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
22b5a6c0 | 2061 | const unsigned int extra = sizeof(struct gfs2_dinode) - sizeof(struct gfs2_leaf); |
c752666c SW |
2062 | struct gfs2_dirent *dent; |
2063 | struct buffer_head *bh; | |
b3b94faa | 2064 | |
3c1c0ae1 | 2065 | da->nr_blocks = 0; |
2b47dad8 SW |
2066 | da->bh = NULL; |
2067 | da->dent = NULL; | |
3c1c0ae1 | 2068 | |
c752666c | 2069 | dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, &bh); |
ed386507 | 2070 | if (!dent) { |
3c1c0ae1 | 2071 | da->nr_blocks = sdp->sd_max_dirres; |
22b5a6c0 SW |
2072 | if (!(ip->i_diskflags & GFS2_DIF_EXHASH) && |
2073 | (GFS2_DIRENT_SIZE(name->len) < extra)) | |
2074 | da->nr_blocks = 1; | |
3c1c0ae1 | 2075 | return 0; |
ed386507 | 2076 | } |
c752666c SW |
2077 | if (IS_ERR(dent)) |
2078 | return PTR_ERR(dent); | |
19aeb5a6 BP |
2079 | |
2080 | if (da->save_loc) { | |
2081 | da->bh = bh; | |
2082 | da->dent = dent; | |
2083 | } else { | |
2084 | brelse(bh); | |
2085 | } | |
c752666c | 2086 | return 0; |
b3b94faa DT |
2087 | } |
2088 |