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 | |
b3b94faa DT |
56 | #include <linux/slab.h> |
57 | #include <linux/spinlock.h> | |
b3b94faa DT |
58 | #include <linux/buffer_head.h> |
59 | #include <linux/sort.h> | |
5c676f6d | 60 | #include <linux/gfs2_ondisk.h> |
71b86f56 | 61 | #include <linux/crc32.h> |
fe1bdedc | 62 | #include <linux/vmalloc.h> |
b3b94faa DT |
63 | |
64 | #include "gfs2.h" | |
5c676f6d | 65 | #include "incore.h" |
b3b94faa DT |
66 | #include "dir.h" |
67 | #include "glock.h" | |
68 | #include "inode.h" | |
b3b94faa DT |
69 | #include "meta_io.h" |
70 | #include "quota.h" | |
71 | #include "rgrp.h" | |
72 | #include "trans.h" | |
e13940ba | 73 | #include "bmap.h" |
5c676f6d | 74 | #include "util.h" |
b3b94faa DT |
75 | |
76 | #define IS_LEAF 1 /* Hashed (leaf) directory */ | |
77 | #define IS_DINODE 2 /* Linear (stuffed dinode block) directory */ | |
78 | ||
dfe4d34b BP |
79 | #define MAX_RA_BLOCKS 32 /* max read-ahead blocks */ |
80 | ||
cd915493 SW |
81 | #define gfs2_disk_hash2offset(h) (((u64)(h)) >> 1) |
82 | #define gfs2_dir_offset2hash(p) ((u32)(((u64)(p)) << 1)) | |
b3b94faa | 83 | |
8d123585 SW |
84 | struct qstr gfs2_qdot __read_mostly; |
85 | struct qstr gfs2_qdotdot __read_mostly; | |
86 | ||
2bdbc5d7 SW |
87 | typedef int (*gfs2_dscan_t)(const struct gfs2_dirent *dent, |
88 | const struct qstr *name, void *opaque); | |
b3b94faa | 89 | |
cd915493 | 90 | int gfs2_dir_get_new_buffer(struct gfs2_inode *ip, u64 block, |
61e085a8 | 91 | struct buffer_head **bhp) |
e13940ba SW |
92 | { |
93 | struct buffer_head *bh; | |
e13940ba | 94 | |
61e085a8 | 95 | bh = gfs2_meta_new(ip->i_gl, block); |
350a9b0a | 96 | gfs2_trans_add_meta(ip->i_gl, bh); |
61e085a8 SW |
97 | gfs2_metatype_set(bh, GFS2_METATYPE_JD, GFS2_FORMAT_JD); |
98 | gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header)); | |
e13940ba SW |
99 | *bhp = bh; |
100 | return 0; | |
101 | } | |
102 | ||
cd915493 | 103 | static int gfs2_dir_get_existing_buffer(struct gfs2_inode *ip, u64 block, |
61e085a8 SW |
104 | struct buffer_head **bhp) |
105 | { | |
106 | struct buffer_head *bh; | |
107 | int error; | |
e13940ba | 108 | |
7276b3b0 | 109 | error = gfs2_meta_read(ip->i_gl, block, DIO_WAIT, &bh); |
61e085a8 SW |
110 | if (error) |
111 | return error; | |
feaa7bba | 112 | if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), bh, GFS2_METATYPE_JD)) { |
61e085a8 SW |
113 | brelse(bh); |
114 | return -EIO; | |
115 | } | |
116 | *bhp = bh; | |
117 | return 0; | |
118 | } | |
e13940ba SW |
119 | |
120 | static int gfs2_dir_write_stuffed(struct gfs2_inode *ip, const char *buf, | |
121 | unsigned int offset, unsigned int size) | |
e13940ba SW |
122 | { |
123 | struct buffer_head *dibh; | |
124 | int error; | |
125 | ||
126 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
127 | if (error) | |
128 | return error; | |
129 | ||
350a9b0a | 130 | gfs2_trans_add_meta(ip->i_gl, dibh); |
c752666c | 131 | memcpy(dibh->b_data + offset + sizeof(struct gfs2_dinode), buf, size); |
a2e0f799 SW |
132 | if (ip->i_inode.i_size < offset + size) |
133 | i_size_write(&ip->i_inode, offset + size); | |
4bd91ba1 | 134 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME; |
539e5d6b | 135 | gfs2_dinode_out(ip, dibh->b_data); |
e13940ba SW |
136 | |
137 | brelse(dibh); | |
138 | ||
139 | return size; | |
140 | } | |
141 | ||
142 | ||
143 | ||
144 | /** | |
145 | * gfs2_dir_write_data - Write directory information to the inode | |
146 | * @ip: The GFS2 inode | |
147 | * @buf: The buffer containing information to be written | |
148 | * @offset: The file offset to start writing at | |
149 | * @size: The amount of data to write | |
150 | * | |
151 | * Returns: The number of bytes correctly written or error code | |
152 | */ | |
153 | static int gfs2_dir_write_data(struct gfs2_inode *ip, const char *buf, | |
cd915493 | 154 | u64 offset, unsigned int size) |
e13940ba | 155 | { |
feaa7bba | 156 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
e13940ba | 157 | struct buffer_head *dibh; |
cd915493 SW |
158 | u64 lblock, dblock; |
159 | u32 extlen = 0; | |
e13940ba SW |
160 | unsigned int o; |
161 | int copied = 0; | |
162 | int error = 0; | |
9b8c81d1 | 163 | int new = 0; |
e13940ba SW |
164 | |
165 | if (!size) | |
166 | return 0; | |
167 | ||
168 | if (gfs2_is_stuffed(ip) && | |
169 | offset + size <= sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) | |
568f4c96 SW |
170 | return gfs2_dir_write_stuffed(ip, buf, (unsigned int)offset, |
171 | size); | |
e13940ba SW |
172 | |
173 | if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip))) | |
174 | return -EINVAL; | |
175 | ||
176 | if (gfs2_is_stuffed(ip)) { | |
f25ef0c1 | 177 | error = gfs2_unstuff_dinode(ip, NULL); |
e13940ba | 178 | if (error) |
c752666c | 179 | return error; |
e13940ba SW |
180 | } |
181 | ||
182 | lblock = offset; | |
183 | o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header); | |
184 | ||
185 | while (copied < size) { | |
186 | unsigned int amount; | |
187 | struct buffer_head *bh; | |
e13940ba SW |
188 | |
189 | amount = size - copied; | |
190 | if (amount > sdp->sd_sb.sb_bsize - o) | |
191 | amount = sdp->sd_sb.sb_bsize - o; | |
192 | ||
193 | if (!extlen) { | |
194 | new = 1; | |
feaa7bba | 195 | error = gfs2_extent_map(&ip->i_inode, lblock, &new, |
fd88de56 | 196 | &dblock, &extlen); |
e13940ba SW |
197 | if (error) |
198 | goto fail; | |
199 | error = -EIO; | |
200 | if (gfs2_assert_withdraw(sdp, dblock)) | |
201 | goto fail; | |
202 | } | |
203 | ||
61e085a8 SW |
204 | if (amount == sdp->sd_jbsize || new) |
205 | error = gfs2_dir_get_new_buffer(ip, dblock, &bh); | |
206 | else | |
207 | error = gfs2_dir_get_existing_buffer(ip, dblock, &bh); | |
208 | ||
e13940ba SW |
209 | if (error) |
210 | goto fail; | |
211 | ||
350a9b0a | 212 | gfs2_trans_add_meta(ip->i_gl, bh); |
e13940ba SW |
213 | memcpy(bh->b_data + o, buf, amount); |
214 | brelse(bh); | |
e13940ba | 215 | |
899bb264 | 216 | buf += amount; |
e13940ba SW |
217 | copied += amount; |
218 | lblock++; | |
219 | dblock++; | |
220 | extlen--; | |
221 | ||
222 | o = sizeof(struct gfs2_meta_header); | |
223 | } | |
224 | ||
225 | out: | |
226 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
227 | if (error) | |
228 | return error; | |
229 | ||
a2e0f799 SW |
230 | if (ip->i_inode.i_size < offset + copied) |
231 | i_size_write(&ip->i_inode, offset + copied); | |
4bd91ba1 | 232 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME; |
e13940ba | 233 | |
350a9b0a | 234 | gfs2_trans_add_meta(ip->i_gl, dibh); |
539e5d6b | 235 | gfs2_dinode_out(ip, dibh->b_data); |
e13940ba SW |
236 | brelse(dibh); |
237 | ||
238 | return copied; | |
239 | fail: | |
240 | if (copied) | |
241 | goto out; | |
242 | return error; | |
243 | } | |
244 | ||
4c28d338 SW |
245 | static int gfs2_dir_read_stuffed(struct gfs2_inode *ip, __be64 *buf, |
246 | unsigned int size) | |
e13940ba SW |
247 | { |
248 | struct buffer_head *dibh; | |
249 | int error; | |
250 | ||
251 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
252 | if (!error) { | |
4c28d338 | 253 | memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size); |
e13940ba SW |
254 | brelse(dibh); |
255 | } | |
256 | ||
257 | return (error) ? error : size; | |
258 | } | |
259 | ||
260 | ||
261 | /** | |
262 | * gfs2_dir_read_data - Read a data from a directory inode | |
263 | * @ip: The GFS2 Inode | |
264 | * @buf: The buffer to place result into | |
e13940ba SW |
265 | * @size: Amount of data to transfer |
266 | * | |
267 | * Returns: The amount of data actually copied or the error | |
268 | */ | |
4c28d338 SW |
269 | static int gfs2_dir_read_data(struct gfs2_inode *ip, __be64 *buf, |
270 | unsigned int size) | |
e13940ba | 271 | { |
feaa7bba | 272 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
cd915493 SW |
273 | u64 lblock, dblock; |
274 | u32 extlen = 0; | |
e13940ba SW |
275 | unsigned int o; |
276 | int copied = 0; | |
277 | int error = 0; | |
e13940ba SW |
278 | |
279 | if (gfs2_is_stuffed(ip)) | |
4c28d338 | 280 | return gfs2_dir_read_stuffed(ip, buf, size); |
e13940ba SW |
281 | |
282 | if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip))) | |
283 | return -EINVAL; | |
284 | ||
4c28d338 | 285 | lblock = 0; |
e13940ba SW |
286 | o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header); |
287 | ||
288 | while (copied < size) { | |
289 | unsigned int amount; | |
290 | struct buffer_head *bh; | |
291 | int new; | |
292 | ||
293 | amount = size - copied; | |
294 | if (amount > sdp->sd_sb.sb_bsize - o) | |
295 | amount = sdp->sd_sb.sb_bsize - o; | |
296 | ||
297 | if (!extlen) { | |
298 | new = 0; | |
feaa7bba | 299 | error = gfs2_extent_map(&ip->i_inode, lblock, &new, |
fd88de56 | 300 | &dblock, &extlen); |
7276b3b0 | 301 | if (error || !dblock) |
e13940ba | 302 | goto fail; |
7276b3b0 | 303 | BUG_ON(extlen < 1); |
7276b3b0 | 304 | bh = gfs2_meta_ra(ip->i_gl, dblock, extlen); |
b7d8ac3e | 305 | } else { |
7276b3b0 | 306 | error = gfs2_meta_read(ip->i_gl, dblock, DIO_WAIT, &bh); |
e13940ba SW |
307 | if (error) |
308 | goto fail; | |
7276b3b0 SW |
309 | } |
310 | error = gfs2_metatype_check(sdp, bh, GFS2_METATYPE_JD); | |
311 | if (error) { | |
312 | brelse(bh); | |
313 | goto fail; | |
314 | } | |
315 | dblock++; | |
316 | extlen--; | |
e13940ba SW |
317 | memcpy(buf, bh->b_data + o, amount); |
318 | brelse(bh); | |
4c28d338 | 319 | buf += (amount/sizeof(__be64)); |
e13940ba SW |
320 | copied += amount; |
321 | lblock++; | |
e13940ba SW |
322 | o = sizeof(struct gfs2_meta_header); |
323 | } | |
324 | ||
325 | return copied; | |
326 | fail: | |
327 | return (copied) ? copied : error; | |
328 | } | |
329 | ||
17d539f0 SW |
330 | /** |
331 | * gfs2_dir_get_hash_table - Get pointer to the dir hash table | |
332 | * @ip: The inode in question | |
333 | * | |
334 | * Returns: The hash table or an error | |
335 | */ | |
336 | ||
337 | static __be64 *gfs2_dir_get_hash_table(struct gfs2_inode *ip) | |
338 | { | |
339 | struct inode *inode = &ip->i_inode; | |
340 | int ret; | |
341 | u32 hsize; | |
342 | __be64 *hc; | |
343 | ||
344 | BUG_ON(!(ip->i_diskflags & GFS2_DIF_EXHASH)); | |
345 | ||
346 | hc = ip->i_hash_cache; | |
347 | if (hc) | |
348 | return hc; | |
349 | ||
350 | hsize = 1 << ip->i_depth; | |
351 | hsize *= sizeof(__be64); | |
352 | if (hsize != i_size_read(&ip->i_inode)) { | |
353 | gfs2_consist_inode(ip); | |
354 | return ERR_PTR(-EIO); | |
355 | } | |
356 | ||
e8830d88 BP |
357 | hc = kmalloc(hsize, GFP_NOFS | __GFP_NOWARN); |
358 | if (hc == NULL) | |
359 | hc = __vmalloc(hsize, GFP_NOFS, PAGE_KERNEL); | |
360 | ||
17d539f0 SW |
361 | if (hc == NULL) |
362 | return ERR_PTR(-ENOMEM); | |
363 | ||
4c28d338 | 364 | ret = gfs2_dir_read_data(ip, hc, hsize); |
17d539f0 | 365 | if (ret < 0) { |
e8830d88 BP |
366 | if (is_vmalloc_addr(hc)) |
367 | vfree(hc); | |
368 | else | |
369 | kfree(hc); | |
17d539f0 SW |
370 | return ERR_PTR(ret); |
371 | } | |
372 | ||
373 | spin_lock(&inode->i_lock); | |
e8830d88 BP |
374 | if (ip->i_hash_cache) { |
375 | if (is_vmalloc_addr(hc)) | |
376 | vfree(hc); | |
377 | else | |
378 | kfree(hc); | |
379 | } else { | |
17d539f0 | 380 | ip->i_hash_cache = hc; |
e8830d88 | 381 | } |
17d539f0 SW |
382 | spin_unlock(&inode->i_lock); |
383 | ||
384 | return ip->i_hash_cache; | |
385 | } | |
386 | ||
387 | /** | |
388 | * gfs2_dir_hash_inval - Invalidate dir hash | |
389 | * @ip: The directory inode | |
390 | * | |
391 | * Must be called with an exclusive glock, or during glock invalidation. | |
392 | */ | |
393 | void gfs2_dir_hash_inval(struct gfs2_inode *ip) | |
394 | { | |
395 | __be64 *hc = ip->i_hash_cache; | |
396 | ip->i_hash_cache = NULL; | |
e8830d88 BP |
397 | if (is_vmalloc_addr(hc)) |
398 | vfree(hc); | |
399 | else | |
400 | kfree(hc); | |
17d539f0 SW |
401 | } |
402 | ||
5e7d65cd SW |
403 | static inline int gfs2_dirent_sentinel(const struct gfs2_dirent *dent) |
404 | { | |
405 | return dent->de_inum.no_addr == 0 || dent->de_inum.no_formal_ino == 0; | |
406 | } | |
407 | ||
c752666c SW |
408 | static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent, |
409 | const struct qstr *name, int ret) | |
410 | { | |
5e7d65cd | 411 | if (!gfs2_dirent_sentinel(dent) && |
c752666c SW |
412 | be32_to_cpu(dent->de_hash) == name->hash && |
413 | be16_to_cpu(dent->de_name_len) == name->len && | |
2bdbc5d7 | 414 | memcmp(dent+1, name->name, name->len) == 0) |
c752666c SW |
415 | return ret; |
416 | return 0; | |
417 | } | |
418 | ||
419 | static int gfs2_dirent_find(const struct gfs2_dirent *dent, | |
71b86f56 SW |
420 | const struct qstr *name, |
421 | void *opaque) | |
c752666c SW |
422 | { |
423 | return __gfs2_dirent_find(dent, name, 1); | |
424 | } | |
425 | ||
426 | static int gfs2_dirent_prev(const struct gfs2_dirent *dent, | |
71b86f56 SW |
427 | const struct qstr *name, |
428 | void *opaque) | |
c752666c SW |
429 | { |
430 | return __gfs2_dirent_find(dent, name, 2); | |
431 | } | |
432 | ||
433 | /* | |
434 | * name->name holds ptr to start of block. | |
435 | * name->len holds size of block. | |
b3b94faa | 436 | */ |
c752666c | 437 | static int gfs2_dirent_last(const struct gfs2_dirent *dent, |
71b86f56 SW |
438 | const struct qstr *name, |
439 | void *opaque) | |
c752666c SW |
440 | { |
441 | const char *start = name->name; | |
442 | const char *end = (const char *)dent + be16_to_cpu(dent->de_rec_len); | |
443 | if (name->len == (end - start)) | |
444 | return 1; | |
445 | return 0; | |
446 | } | |
b3b94faa | 447 | |
c752666c | 448 | static int gfs2_dirent_find_space(const struct gfs2_dirent *dent, |
71b86f56 SW |
449 | const struct qstr *name, |
450 | void *opaque) | |
b3b94faa | 451 | { |
c752666c SW |
452 | unsigned required = GFS2_DIRENT_SIZE(name->len); |
453 | unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len)); | |
454 | unsigned totlen = be16_to_cpu(dent->de_rec_len); | |
455 | ||
5e7d65cd | 456 | if (gfs2_dirent_sentinel(dent)) |
728a756b | 457 | actual = 0; |
c5392124 | 458 | if (totlen - actual >= required) |
c752666c SW |
459 | return 1; |
460 | return 0; | |
461 | } | |
462 | ||
71b86f56 SW |
463 | struct dirent_gather { |
464 | const struct gfs2_dirent **pdent; | |
465 | unsigned offset; | |
466 | }; | |
467 | ||
468 | static int gfs2_dirent_gather(const struct gfs2_dirent *dent, | |
469 | const struct qstr *name, | |
470 | void *opaque) | |
471 | { | |
472 | struct dirent_gather *g = opaque; | |
5e7d65cd | 473 | if (!gfs2_dirent_sentinel(dent)) { |
71b86f56 SW |
474 | g->pdent[g->offset++] = dent; |
475 | } | |
476 | return 0; | |
477 | } | |
478 | ||
c752666c SW |
479 | /* |
480 | * Other possible things to check: | |
481 | * - Inode located within filesystem size (and on valid block) | |
482 | * - Valid directory entry type | |
483 | * Not sure how heavy-weight we want to make this... could also check | |
484 | * hash is correct for example, but that would take a lot of extra time. | |
485 | * For now the most important thing is to check that the various sizes | |
486 | * are correct. | |
487 | */ | |
488 | static int gfs2_check_dirent(struct gfs2_dirent *dent, unsigned int offset, | |
489 | unsigned int size, unsigned int len, int first) | |
490 | { | |
491 | const char *msg = "gfs2_dirent too small"; | |
492 | if (unlikely(size < sizeof(struct gfs2_dirent))) | |
493 | goto error; | |
494 | msg = "gfs2_dirent misaligned"; | |
495 | if (unlikely(offset & 0x7)) | |
496 | goto error; | |
497 | msg = "gfs2_dirent points beyond end of block"; | |
498 | if (unlikely(offset + size > len)) | |
499 | goto error; | |
500 | msg = "zero inode number"; | |
5e7d65cd | 501 | if (unlikely(!first && gfs2_dirent_sentinel(dent))) |
c752666c SW |
502 | goto error; |
503 | msg = "name length is greater than space in dirent"; | |
5e7d65cd | 504 | if (!gfs2_dirent_sentinel(dent) && |
c752666c SW |
505 | unlikely(sizeof(struct gfs2_dirent)+be16_to_cpu(dent->de_name_len) > |
506 | size)) | |
507 | goto error; | |
508 | return 0; | |
509 | error: | |
510 | printk(KERN_WARNING "gfs2_check_dirent: %s (%s)\n", msg, | |
511 | first ? "first in block" : "not first in block"); | |
512 | return -EIO; | |
b3b94faa DT |
513 | } |
514 | ||
71b86f56 | 515 | static int gfs2_dirent_offset(const void *buf) |
c752666c | 516 | { |
71b86f56 SW |
517 | const struct gfs2_meta_header *h = buf; |
518 | int offset; | |
c752666c SW |
519 | |
520 | BUG_ON(buf == NULL); | |
c752666c | 521 | |
e3167ded | 522 | switch(be32_to_cpu(h->mh_type)) { |
c752666c SW |
523 | case GFS2_METATYPE_LF: |
524 | offset = sizeof(struct gfs2_leaf); | |
525 | break; | |
526 | case GFS2_METATYPE_DI: | |
527 | offset = sizeof(struct gfs2_dinode); | |
528 | break; | |
529 | default: | |
530 | goto wrong_type; | |
531 | } | |
71b86f56 SW |
532 | return offset; |
533 | wrong_type: | |
534 | printk(KERN_WARNING "gfs2_scan_dirent: wrong block type %u\n", | |
e3167ded | 535 | be32_to_cpu(h->mh_type)); |
71b86f56 SW |
536 | return -1; |
537 | } | |
538 | ||
2bdbc5d7 | 539 | static struct gfs2_dirent *gfs2_dirent_scan(struct inode *inode, void *buf, |
71b86f56 SW |
540 | unsigned int len, gfs2_dscan_t scan, |
541 | const struct qstr *name, | |
542 | void *opaque) | |
543 | { | |
544 | struct gfs2_dirent *dent, *prev; | |
545 | unsigned offset; | |
546 | unsigned size; | |
547 | int ret = 0; | |
c752666c | 548 | |
71b86f56 SW |
549 | ret = gfs2_dirent_offset(buf); |
550 | if (ret < 0) | |
551 | goto consist_inode; | |
552 | ||
553 | offset = ret; | |
c752666c | 554 | prev = NULL; |
2bdbc5d7 | 555 | dent = buf + offset; |
c752666c SW |
556 | size = be16_to_cpu(dent->de_rec_len); |
557 | if (gfs2_check_dirent(dent, offset, size, len, 1)) | |
558 | goto consist_inode; | |
559 | do { | |
71b86f56 | 560 | ret = scan(dent, name, opaque); |
c752666c SW |
561 | if (ret) |
562 | break; | |
563 | offset += size; | |
564 | if (offset == len) | |
565 | break; | |
566 | prev = dent; | |
2bdbc5d7 | 567 | dent = buf + offset; |
c752666c SW |
568 | size = be16_to_cpu(dent->de_rec_len); |
569 | if (gfs2_check_dirent(dent, offset, size, len, 0)) | |
570 | goto consist_inode; | |
571 | } while(1); | |
572 | ||
573 | switch(ret) { | |
574 | case 0: | |
575 | return NULL; | |
576 | case 1: | |
577 | return dent; | |
578 | case 2: | |
579 | return prev ? prev : dent; | |
580 | default: | |
581 | BUG_ON(ret > 0); | |
582 | return ERR_PTR(ret); | |
583 | } | |
584 | ||
c752666c | 585 | consist_inode: |
feaa7bba | 586 | gfs2_consist_inode(GFS2_I(inode)); |
c752666c SW |
587 | return ERR_PTR(-EIO); |
588 | } | |
589 | ||
2bdbc5d7 SW |
590 | static int dirent_check_reclen(struct gfs2_inode *dip, |
591 | const struct gfs2_dirent *d, const void *end_p) | |
592 | { | |
593 | const void *ptr = d; | |
594 | u16 rec_len = be16_to_cpu(d->de_rec_len); | |
595 | ||
596 | if (unlikely(rec_len < sizeof(struct gfs2_dirent))) | |
597 | goto broken; | |
598 | ptr += rec_len; | |
599 | if (ptr < end_p) | |
600 | return rec_len; | |
601 | if (ptr == end_p) | |
602 | return -ENOENT; | |
603 | broken: | |
604 | gfs2_consist_inode(dip); | |
605 | return -EIO; | |
606 | } | |
607 | ||
b3b94faa DT |
608 | /** |
609 | * dirent_next - Next dirent | |
610 | * @dip: the directory | |
611 | * @bh: The buffer | |
612 | * @dent: Pointer to list of dirents | |
613 | * | |
614 | * Returns: 0 on success, error code otherwise | |
615 | */ | |
616 | ||
617 | static int dirent_next(struct gfs2_inode *dip, struct buffer_head *bh, | |
618 | struct gfs2_dirent **dent) | |
619 | { | |
2bdbc5d7 SW |
620 | struct gfs2_dirent *cur = *dent, *tmp; |
621 | char *bh_end = bh->b_data + bh->b_size; | |
622 | int ret; | |
b3b94faa | 623 | |
2bdbc5d7 SW |
624 | ret = dirent_check_reclen(dip, cur, bh_end); |
625 | if (ret < 0) | |
626 | return ret; | |
4dd651ad | 627 | |
2bdbc5d7 SW |
628 | tmp = (void *)cur + ret; |
629 | ret = dirent_check_reclen(dip, tmp, bh_end); | |
630 | if (ret == -EIO) | |
631 | return ret; | |
4dd651ad | 632 | |
b3b94faa | 633 | /* Only the first dent could ever have de_inum.no_addr == 0 */ |
5e7d65cd | 634 | if (gfs2_dirent_sentinel(tmp)) { |
b3b94faa DT |
635 | gfs2_consist_inode(dip); |
636 | return -EIO; | |
637 | } | |
638 | ||
639 | *dent = tmp; | |
b3b94faa DT |
640 | return 0; |
641 | } | |
642 | ||
643 | /** | |
644 | * dirent_del - Delete a dirent | |
645 | * @dip: The GFS2 inode | |
646 | * @bh: The buffer | |
647 | * @prev: The previous dirent | |
648 | * @cur: The current dirent | |
649 | * | |
650 | */ | |
651 | ||
652 | static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh, | |
653 | struct gfs2_dirent *prev, struct gfs2_dirent *cur) | |
654 | { | |
cd915493 | 655 | u16 cur_rec_len, prev_rec_len; |
b3b94faa | 656 | |
5e7d65cd | 657 | if (gfs2_dirent_sentinel(cur)) { |
b3b94faa DT |
658 | gfs2_consist_inode(dip); |
659 | return; | |
660 | } | |
661 | ||
350a9b0a | 662 | gfs2_trans_add_meta(dip->i_gl, bh); |
b3b94faa DT |
663 | |
664 | /* If there is no prev entry, this is the first entry in the block. | |
665 | The de_rec_len is already as big as it needs to be. Just zero | |
666 | out the inode number and return. */ | |
667 | ||
668 | if (!prev) { | |
5e7d65cd SW |
669 | cur->de_inum.no_addr = 0; |
670 | cur->de_inum.no_formal_ino = 0; | |
b3b94faa DT |
671 | return; |
672 | } | |
673 | ||
674 | /* Combine this dentry with the previous one. */ | |
675 | ||
fc69d0d3 SW |
676 | prev_rec_len = be16_to_cpu(prev->de_rec_len); |
677 | cur_rec_len = be16_to_cpu(cur->de_rec_len); | |
b3b94faa DT |
678 | |
679 | if ((char *)prev + prev_rec_len != (char *)cur) | |
680 | gfs2_consist_inode(dip); | |
681 | if ((char *)cur + cur_rec_len > bh->b_data + bh->b_size) | |
682 | gfs2_consist_inode(dip); | |
683 | ||
684 | prev_rec_len += cur_rec_len; | |
fc69d0d3 | 685 | prev->de_rec_len = cpu_to_be16(prev_rec_len); |
b3b94faa DT |
686 | } |
687 | ||
c752666c SW |
688 | /* |
689 | * Takes a dent from which to grab space as an argument. Returns the | |
690 | * newly created dent. | |
b3b94faa | 691 | */ |
08bc2dbc AB |
692 | static struct gfs2_dirent *gfs2_init_dirent(struct inode *inode, |
693 | struct gfs2_dirent *dent, | |
694 | const struct qstr *name, | |
695 | struct buffer_head *bh) | |
b3b94faa | 696 | { |
feaa7bba | 697 | struct gfs2_inode *ip = GFS2_I(inode); |
c752666c SW |
698 | struct gfs2_dirent *ndent; |
699 | unsigned offset = 0, totlen; | |
700 | ||
5e7d65cd | 701 | if (!gfs2_dirent_sentinel(dent)) |
c752666c SW |
702 | offset = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len)); |
703 | totlen = be16_to_cpu(dent->de_rec_len); | |
704 | BUG_ON(offset + name->len > totlen); | |
350a9b0a | 705 | gfs2_trans_add_meta(ip->i_gl, bh); |
c752666c SW |
706 | ndent = (struct gfs2_dirent *)((char *)dent + offset); |
707 | dent->de_rec_len = cpu_to_be16(offset); | |
708 | gfs2_qstr2dirent(name, totlen - offset, ndent); | |
709 | return ndent; | |
b3b94faa DT |
710 | } |
711 | ||
c752666c SW |
712 | static struct gfs2_dirent *gfs2_dirent_alloc(struct inode *inode, |
713 | struct buffer_head *bh, | |
714 | const struct qstr *name) | |
b3b94faa DT |
715 | { |
716 | struct gfs2_dirent *dent; | |
907b9bce | 717 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, |
71b86f56 | 718 | gfs2_dirent_find_space, name, NULL); |
c752666c SW |
719 | if (!dent || IS_ERR(dent)) |
720 | return dent; | |
721 | return gfs2_init_dirent(inode, dent, name, bh); | |
b3b94faa DT |
722 | } |
723 | ||
cd915493 | 724 | static int get_leaf(struct gfs2_inode *dip, u64 leaf_no, |
b3b94faa DT |
725 | struct buffer_head **bhp) |
726 | { | |
727 | int error; | |
728 | ||
7276b3b0 | 729 | error = gfs2_meta_read(dip->i_gl, leaf_no, DIO_WAIT, bhp); |
feaa7bba SW |
730 | if (!error && gfs2_metatype_check(GFS2_SB(&dip->i_inode), *bhp, GFS2_METATYPE_LF)) { |
731 | /* printk(KERN_INFO "block num=%llu\n", leaf_no); */ | |
b3b94faa | 732 | error = -EIO; |
feaa7bba | 733 | } |
b3b94faa DT |
734 | |
735 | return error; | |
736 | } | |
737 | ||
738 | /** | |
739 | * get_leaf_nr - Get a leaf number associated with the index | |
740 | * @dip: The GFS2 inode | |
741 | * @index: | |
742 | * @leaf_out: | |
743 | * | |
744 | * Returns: 0 on success, error code otherwise | |
745 | */ | |
746 | ||
cd915493 SW |
747 | static int get_leaf_nr(struct gfs2_inode *dip, u32 index, |
748 | u64 *leaf_out) | |
b3b94faa | 749 | { |
17d539f0 | 750 | __be64 *hash; |
b3b94faa | 751 | |
17d539f0 SW |
752 | hash = gfs2_dir_get_hash_table(dip); |
753 | if (IS_ERR(hash)) | |
754 | return PTR_ERR(hash); | |
755 | *leaf_out = be64_to_cpu(*(hash + index)); | |
b3b94faa DT |
756 | return 0; |
757 | } | |
758 | ||
cd915493 | 759 | static int get_first_leaf(struct gfs2_inode *dip, u32 index, |
b3b94faa DT |
760 | struct buffer_head **bh_out) |
761 | { | |
cd915493 | 762 | u64 leaf_no; |
b3b94faa DT |
763 | int error; |
764 | ||
765 | error = get_leaf_nr(dip, index, &leaf_no); | |
766 | if (!error) | |
767 | error = get_leaf(dip, leaf_no, bh_out); | |
768 | ||
769 | return error; | |
770 | } | |
771 | ||
c752666c SW |
772 | static struct gfs2_dirent *gfs2_dirent_search(struct inode *inode, |
773 | const struct qstr *name, | |
774 | gfs2_dscan_t scan, | |
775 | struct buffer_head **pbh) | |
b3b94faa | 776 | { |
c752666c SW |
777 | struct buffer_head *bh; |
778 | struct gfs2_dirent *dent; | |
feaa7bba | 779 | struct gfs2_inode *ip = GFS2_I(inode); |
b3b94faa DT |
780 | int error; |
781 | ||
383f01fb | 782 | if (ip->i_diskflags & GFS2_DIF_EXHASH) { |
c752666c | 783 | struct gfs2_leaf *leaf; |
9a004508 | 784 | unsigned hsize = 1 << ip->i_depth; |
c752666c SW |
785 | unsigned index; |
786 | u64 ln; | |
a2e0f799 | 787 | if (hsize * sizeof(u64) != i_size_read(inode)) { |
c752666c SW |
788 | gfs2_consist_inode(ip); |
789 | return ERR_PTR(-EIO); | |
790 | } | |
907b9bce | 791 | |
9a004508 | 792 | index = name->hash >> (32 - ip->i_depth); |
c752666c SW |
793 | error = get_first_leaf(ip, index, &bh); |
794 | if (error) | |
795 | return ERR_PTR(error); | |
796 | do { | |
797 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, | |
71b86f56 | 798 | scan, name, NULL); |
c752666c SW |
799 | if (dent) |
800 | goto got_dent; | |
801 | leaf = (struct gfs2_leaf *)bh->b_data; | |
802 | ln = be64_to_cpu(leaf->lf_next); | |
f4154ea0 | 803 | brelse(bh); |
c752666c SW |
804 | if (!ln) |
805 | break; | |
907b9bce | 806 | |
c752666c SW |
807 | error = get_leaf(ip, ln, &bh); |
808 | } while(!error); | |
b3b94faa | 809 | |
c752666c | 810 | return error ? ERR_PTR(error) : NULL; |
b3b94faa | 811 | } |
b3b94faa | 812 | |
907b9bce | 813 | |
c752666c SW |
814 | error = gfs2_meta_inode_buffer(ip, &bh); |
815 | if (error) | |
816 | return ERR_PTR(error); | |
71b86f56 | 817 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, scan, name, NULL); |
c752666c | 818 | got_dent: |
f4154ea0 | 819 | if (unlikely(dent == NULL || IS_ERR(dent))) { |
ed386507 SW |
820 | brelse(bh); |
821 | bh = NULL; | |
822 | } | |
c752666c SW |
823 | *pbh = bh; |
824 | return dent; | |
825 | } | |
b3b94faa | 826 | |
c752666c SW |
827 | static struct gfs2_leaf *new_leaf(struct inode *inode, struct buffer_head **pbh, u16 depth) |
828 | { | |
feaa7bba | 829 | struct gfs2_inode *ip = GFS2_I(inode); |
b45e41d7 | 830 | unsigned int n = 1; |
09010978 SW |
831 | u64 bn; |
832 | int error; | |
833 | struct buffer_head *bh; | |
c752666c SW |
834 | struct gfs2_leaf *leaf; |
835 | struct gfs2_dirent *dent; | |
26fe5750 | 836 | struct qstr name = { .name = "" }; |
01bcb0de | 837 | struct timespec tv = CURRENT_TIME; |
09010978 | 838 | |
6e87ed0f | 839 | error = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL); |
09010978 SW |
840 | if (error) |
841 | return NULL; | |
842 | bh = gfs2_meta_new(ip->i_gl, bn); | |
c752666c SW |
843 | if (!bh) |
844 | return NULL; | |
09010978 | 845 | |
5731be53 | 846 | gfs2_trans_add_unrevoke(GFS2_SB(inode), bn, 1); |
350a9b0a | 847 | gfs2_trans_add_meta(ip->i_gl, bh); |
c752666c SW |
848 | gfs2_metatype_set(bh, GFS2_METATYPE_LF, GFS2_FORMAT_LF); |
849 | leaf = (struct gfs2_leaf *)bh->b_data; | |
850 | leaf->lf_depth = cpu_to_be16(depth); | |
2bdbc5d7 | 851 | leaf->lf_entries = 0; |
a2d7d021 | 852 | leaf->lf_dirent_format = cpu_to_be32(GFS2_FORMAT_DE); |
2bdbc5d7 | 853 | leaf->lf_next = 0; |
01bcb0de SW |
854 | leaf->lf_inode = cpu_to_be64(ip->i_no_addr); |
855 | leaf->lf_dist = cpu_to_be32(1); | |
856 | leaf->lf_nsec = cpu_to_be32(tv.tv_nsec); | |
857 | leaf->lf_sec = cpu_to_be64(tv.tv_sec); | |
858 | memset(leaf->lf_reserved2, 0, sizeof(leaf->lf_reserved2)); | |
c752666c | 859 | dent = (struct gfs2_dirent *)(leaf+1); |
71b86f56 | 860 | gfs2_qstr2dirent(&name, bh->b_size - sizeof(struct gfs2_leaf), dent); |
c752666c SW |
861 | *pbh = bh; |
862 | return leaf; | |
b3b94faa DT |
863 | } |
864 | ||
865 | /** | |
866 | * dir_make_exhash - Convert a stuffed directory into an ExHash directory | |
867 | * @dip: The GFS2 inode | |
868 | * | |
869 | * Returns: 0 on success, error code otherwise | |
870 | */ | |
871 | ||
c752666c | 872 | static int dir_make_exhash(struct inode *inode) |
b3b94faa | 873 | { |
feaa7bba SW |
874 | struct gfs2_inode *dip = GFS2_I(inode); |
875 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
b3b94faa | 876 | struct gfs2_dirent *dent; |
c752666c | 877 | struct qstr args; |
b3b94faa DT |
878 | struct buffer_head *bh, *dibh; |
879 | struct gfs2_leaf *leaf; | |
880 | int y; | |
cd915493 | 881 | u32 x; |
b44b84d7 AV |
882 | __be64 *lp; |
883 | u64 bn; | |
b3b94faa DT |
884 | int error; |
885 | ||
886 | error = gfs2_meta_inode_buffer(dip, &dibh); | |
887 | if (error) | |
888 | return error; | |
889 | ||
b3b94faa DT |
890 | /* Turn over a new leaf */ |
891 | ||
c752666c SW |
892 | leaf = new_leaf(inode, &bh, 0); |
893 | if (!leaf) | |
894 | return -ENOSPC; | |
895 | bn = bh->b_blocknr; | |
b3b94faa | 896 | |
ad6203f2 SW |
897 | gfs2_assert(sdp, dip->i_entries < (1 << 16)); |
898 | leaf->lf_entries = cpu_to_be16(dip->i_entries); | |
b3b94faa DT |
899 | |
900 | /* Copy dirents */ | |
901 | ||
902 | gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_leaf), dibh, | |
903 | sizeof(struct gfs2_dinode)); | |
904 | ||
905 | /* Find last entry */ | |
906 | ||
907 | x = 0; | |
c752666c SW |
908 | args.len = bh->b_size - sizeof(struct gfs2_dinode) + |
909 | sizeof(struct gfs2_leaf); | |
910 | args.name = bh->b_data; | |
feaa7bba | 911 | dent = gfs2_dirent_scan(&dip->i_inode, bh->b_data, bh->b_size, |
71b86f56 | 912 | gfs2_dirent_last, &args, NULL); |
c752666c SW |
913 | if (!dent) { |
914 | brelse(bh); | |
915 | brelse(dibh); | |
916 | return -EIO; | |
917 | } | |
918 | if (IS_ERR(dent)) { | |
919 | brelse(bh); | |
920 | brelse(dibh); | |
921 | return PTR_ERR(dent); | |
b3b94faa | 922 | } |
b3b94faa DT |
923 | |
924 | /* Adjust the last dirent's record length | |
925 | (Remember that dent still points to the last entry.) */ | |
926 | ||
4dd651ad | 927 | dent->de_rec_len = cpu_to_be16(be16_to_cpu(dent->de_rec_len) + |
b3b94faa | 928 | sizeof(struct gfs2_dinode) - |
4dd651ad | 929 | sizeof(struct gfs2_leaf)); |
b3b94faa DT |
930 | |
931 | brelse(bh); | |
932 | ||
933 | /* We're done with the new leaf block, now setup the new | |
934 | hash table. */ | |
935 | ||
350a9b0a | 936 | gfs2_trans_add_meta(dip->i_gl, dibh); |
b3b94faa DT |
937 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); |
938 | ||
b44b84d7 | 939 | lp = (__be64 *)(dibh->b_data + sizeof(struct gfs2_dinode)); |
b3b94faa DT |
940 | |
941 | for (x = sdp->sd_hash_ptrs; x--; lp++) | |
942 | *lp = cpu_to_be64(bn); | |
943 | ||
a2e0f799 | 944 | i_size_write(inode, sdp->sd_sb.sb_bsize / 2); |
77658aad | 945 | gfs2_add_inode_blocks(&dip->i_inode, 1); |
383f01fb | 946 | dip->i_diskflags |= GFS2_DIF_EXHASH; |
b3b94faa DT |
947 | |
948 | for (x = sdp->sd_hash_ptrs, y = -1; x; x >>= 1, y++) ; | |
9a004508 | 949 | dip->i_depth = y; |
b3b94faa | 950 | |
539e5d6b | 951 | gfs2_dinode_out(dip, dibh->b_data); |
b3b94faa DT |
952 | |
953 | brelse(dibh); | |
954 | ||
955 | return 0; | |
956 | } | |
957 | ||
958 | /** | |
959 | * dir_split_leaf - Split a leaf block into two | |
960 | * @dip: The GFS2 inode | |
961 | * @index: | |
962 | * @leaf_no: | |
963 | * | |
964 | * Returns: 0 on success, error code on failure | |
965 | */ | |
966 | ||
c752666c | 967 | static int dir_split_leaf(struct inode *inode, const struct qstr *name) |
b3b94faa | 968 | { |
feaa7bba | 969 | struct gfs2_inode *dip = GFS2_I(inode); |
b3b94faa DT |
970 | struct buffer_head *nbh, *obh, *dibh; |
971 | struct gfs2_leaf *nleaf, *oleaf; | |
4da3c646 | 972 | struct gfs2_dirent *dent = NULL, *prev = NULL, *next = NULL, *new; |
cd915493 | 973 | u32 start, len, half_len, divider; |
b44b84d7 AV |
974 | u64 bn, leaf_no; |
975 | __be64 *lp; | |
cd915493 | 976 | u32 index; |
b3b94faa DT |
977 | int x, moved = 0; |
978 | int error; | |
979 | ||
9a004508 | 980 | index = name->hash >> (32 - dip->i_depth); |
c752666c SW |
981 | error = get_leaf_nr(dip, index, &leaf_no); |
982 | if (error) | |
983 | return error; | |
b3b94faa DT |
984 | |
985 | /* Get the old leaf block */ | |
b3b94faa DT |
986 | error = get_leaf(dip, leaf_no, &obh); |
987 | if (error) | |
e90deff5 | 988 | return error; |
b3b94faa | 989 | |
b3b94faa | 990 | oleaf = (struct gfs2_leaf *)obh->b_data; |
9a004508 | 991 | if (dip->i_depth == be16_to_cpu(oleaf->lf_depth)) { |
e90deff5 SW |
992 | brelse(obh); |
993 | return 1; /* can't split */ | |
994 | } | |
995 | ||
350a9b0a | 996 | gfs2_trans_add_meta(dip->i_gl, obh); |
b3b94faa | 997 | |
c752666c SW |
998 | nleaf = new_leaf(inode, &nbh, be16_to_cpu(oleaf->lf_depth) + 1); |
999 | if (!nleaf) { | |
1000 | brelse(obh); | |
1001 | return -ENOSPC; | |
1002 | } | |
1003 | bn = nbh->b_blocknr; | |
b3b94faa | 1004 | |
c752666c | 1005 | /* Compute the start and len of leaf pointers in the hash table. */ |
9a004508 | 1006 | len = 1 << (dip->i_depth - be16_to_cpu(oleaf->lf_depth)); |
b3b94faa DT |
1007 | half_len = len >> 1; |
1008 | if (!half_len) { | |
9a004508 | 1009 | printk(KERN_WARNING "i_depth %u lf_depth %u index %u\n", dip->i_depth, be16_to_cpu(oleaf->lf_depth), index); |
b3b94faa DT |
1010 | gfs2_consist_inode(dip); |
1011 | error = -EIO; | |
1012 | goto fail_brelse; | |
1013 | } | |
1014 | ||
1015 | start = (index & ~(len - 1)); | |
1016 | ||
1017 | /* Change the pointers. | |
1018 | Don't bother distinguishing stuffed from non-stuffed. | |
1019 | This code is complicated enough already. */ | |
4244b52e DR |
1020 | lp = kmalloc(half_len * sizeof(__be64), GFP_NOFS); |
1021 | if (!lp) { | |
1022 | error = -ENOMEM; | |
1023 | goto fail_brelse; | |
1024 | } | |
1025 | ||
b3b94faa | 1026 | /* Change the pointers */ |
b3b94faa DT |
1027 | for (x = 0; x < half_len; x++) |
1028 | lp[x] = cpu_to_be64(bn); | |
1029 | ||
17d539f0 SW |
1030 | gfs2_dir_hash_inval(dip); |
1031 | ||
cd915493 SW |
1032 | error = gfs2_dir_write_data(dip, (char *)lp, start * sizeof(u64), |
1033 | half_len * sizeof(u64)); | |
1034 | if (error != half_len * sizeof(u64)) { | |
b3b94faa DT |
1035 | if (error >= 0) |
1036 | error = -EIO; | |
1037 | goto fail_lpfree; | |
1038 | } | |
1039 | ||
1040 | kfree(lp); | |
1041 | ||
1042 | /* Compute the divider */ | |
9a004508 | 1043 | divider = (start + half_len) << (32 - dip->i_depth); |
b3b94faa DT |
1044 | |
1045 | /* Copy the entries */ | |
1579343a | 1046 | dent = (struct gfs2_dirent *)(obh->b_data + sizeof(struct gfs2_leaf)); |
b3b94faa DT |
1047 | |
1048 | do { | |
1049 | next = dent; | |
1050 | if (dirent_next(dip, obh, &next)) | |
1051 | next = NULL; | |
1052 | ||
5e7d65cd | 1053 | if (!gfs2_dirent_sentinel(dent) && |
b3b94faa | 1054 | be32_to_cpu(dent->de_hash) < divider) { |
c752666c SW |
1055 | struct qstr str; |
1056 | str.name = (char*)(dent+1); | |
1057 | str.len = be16_to_cpu(dent->de_name_len); | |
1058 | str.hash = be32_to_cpu(dent->de_hash); | |
71b86f56 | 1059 | new = gfs2_dirent_alloc(inode, nbh, &str); |
c752666c SW |
1060 | if (IS_ERR(new)) { |
1061 | error = PTR_ERR(new); | |
1062 | break; | |
1063 | } | |
b3b94faa DT |
1064 | |
1065 | new->de_inum = dent->de_inum; /* No endian worries */ | |
b3b94faa | 1066 | new->de_type = dent->de_type; /* No endian worries */ |
bb16b342 | 1067 | be16_add_cpu(&nleaf->lf_entries, 1); |
b3b94faa DT |
1068 | |
1069 | dirent_del(dip, obh, prev, dent); | |
1070 | ||
1071 | if (!oleaf->lf_entries) | |
1072 | gfs2_consist_inode(dip); | |
bb16b342 | 1073 | be16_add_cpu(&oleaf->lf_entries, -1); |
b3b94faa DT |
1074 | |
1075 | if (!prev) | |
1076 | prev = dent; | |
1077 | ||
1078 | moved = 1; | |
c752666c | 1079 | } else { |
b3b94faa | 1080 | prev = dent; |
c752666c | 1081 | } |
b3b94faa | 1082 | dent = next; |
c752666c | 1083 | } while (dent); |
b3b94faa | 1084 | |
c752666c | 1085 | oleaf->lf_depth = nleaf->lf_depth; |
b3b94faa DT |
1086 | |
1087 | error = gfs2_meta_inode_buffer(dip, &dibh); | |
feaa7bba | 1088 | if (!gfs2_assert_withdraw(GFS2_SB(&dip->i_inode), !error)) { |
350a9b0a | 1089 | gfs2_trans_add_meta(dip->i_gl, dibh); |
77658aad | 1090 | gfs2_add_inode_blocks(&dip->i_inode, 1); |
539e5d6b | 1091 | gfs2_dinode_out(dip, dibh->b_data); |
b3b94faa DT |
1092 | brelse(dibh); |
1093 | } | |
1094 | ||
1095 | brelse(obh); | |
1096 | brelse(nbh); | |
1097 | ||
1098 | return error; | |
1099 | ||
e90deff5 | 1100 | fail_lpfree: |
b3b94faa DT |
1101 | kfree(lp); |
1102 | ||
e90deff5 | 1103 | fail_brelse: |
b3b94faa | 1104 | brelse(obh); |
b3b94faa DT |
1105 | brelse(nbh); |
1106 | return error; | |
1107 | } | |
1108 | ||
1109 | /** | |
1110 | * dir_double_exhash - Double size of ExHash table | |
1111 | * @dip: The GFS2 dinode | |
1112 | * | |
1113 | * Returns: 0 on success, error code on failure | |
1114 | */ | |
1115 | ||
1116 | static int dir_double_exhash(struct gfs2_inode *dip) | |
1117 | { | |
b3b94faa | 1118 | struct buffer_head *dibh; |
cd915493 | 1119 | u32 hsize; |
17d539f0 SW |
1120 | u32 hsize_bytes; |
1121 | __be64 *hc; | |
1122 | __be64 *hc2, *h; | |
b3b94faa DT |
1123 | int x; |
1124 | int error = 0; | |
1125 | ||
9a004508 | 1126 | hsize = 1 << dip->i_depth; |
17d539f0 | 1127 | hsize_bytes = hsize * sizeof(__be64); |
b3b94faa | 1128 | |
17d539f0 SW |
1129 | hc = gfs2_dir_get_hash_table(dip); |
1130 | if (IS_ERR(hc)) | |
1131 | return PTR_ERR(hc); | |
b3b94faa | 1132 | |
512cbf02 | 1133 | hc2 = kmalloc(hsize_bytes * 2, GFP_NOFS | __GFP_NOWARN); |
e8830d88 BP |
1134 | if (hc2 == NULL) |
1135 | hc2 = __vmalloc(hsize_bytes * 2, GFP_NOFS, PAGE_KERNEL); | |
1136 | ||
17d539f0 | 1137 | if (!hc2) |
4244b52e | 1138 | return -ENOMEM; |
b3b94faa | 1139 | |
512cbf02 | 1140 | h = hc2; |
17d539f0 SW |
1141 | error = gfs2_meta_inode_buffer(dip, &dibh); |
1142 | if (error) | |
1143 | goto out_kfree; | |
b3b94faa | 1144 | |
17d539f0 SW |
1145 | for (x = 0; x < hsize; x++) { |
1146 | *h++ = *hc; | |
1147 | *h++ = *hc; | |
1148 | hc++; | |
b3b94faa DT |
1149 | } |
1150 | ||
17d539f0 SW |
1151 | error = gfs2_dir_write_data(dip, (char *)hc2, 0, hsize_bytes * 2); |
1152 | if (error != (hsize_bytes * 2)) | |
1153 | goto fail; | |
b3b94faa | 1154 | |
17d539f0 SW |
1155 | gfs2_dir_hash_inval(dip); |
1156 | dip->i_hash_cache = hc2; | |
1157 | dip->i_depth++; | |
1158 | gfs2_dinode_out(dip, dibh->b_data); | |
1159 | brelse(dibh); | |
1160 | return 0; | |
b3b94faa | 1161 | |
a91ea69f | 1162 | fail: |
17d539f0 SW |
1163 | /* Replace original hash table & size */ |
1164 | gfs2_dir_write_data(dip, (char *)hc, 0, hsize_bytes); | |
1165 | i_size_write(&dip->i_inode, hsize_bytes); | |
1166 | gfs2_dinode_out(dip, dibh->b_data); | |
1167 | brelse(dibh); | |
1168 | out_kfree: | |
e8830d88 BP |
1169 | if (is_vmalloc_addr(hc2)) |
1170 | vfree(hc2); | |
1171 | else | |
1172 | kfree(hc2); | |
b3b94faa DT |
1173 | return error; |
1174 | } | |
1175 | ||
1176 | /** | |
1177 | * compare_dents - compare directory entries by hash value | |
1178 | * @a: first dent | |
1179 | * @b: second dent | |
1180 | * | |
1181 | * When comparing the hash entries of @a to @b: | |
1182 | * gt: returns 1 | |
1183 | * lt: returns -1 | |
1184 | * eq: returns 0 | |
1185 | */ | |
1186 | ||
1187 | static int compare_dents(const void *a, const void *b) | |
1188 | { | |
2bdbc5d7 | 1189 | const struct gfs2_dirent *dent_a, *dent_b; |
cd915493 | 1190 | u32 hash_a, hash_b; |
b3b94faa DT |
1191 | int ret = 0; |
1192 | ||
2bdbc5d7 | 1193 | dent_a = *(const struct gfs2_dirent **)a; |
c752666c | 1194 | hash_a = be32_to_cpu(dent_a->de_hash); |
b3b94faa | 1195 | |
2bdbc5d7 | 1196 | dent_b = *(const struct gfs2_dirent **)b; |
c752666c | 1197 | hash_b = be32_to_cpu(dent_b->de_hash); |
b3b94faa DT |
1198 | |
1199 | if (hash_a > hash_b) | |
1200 | ret = 1; | |
1201 | else if (hash_a < hash_b) | |
1202 | ret = -1; | |
1203 | else { | |
4dd651ad SW |
1204 | unsigned int len_a = be16_to_cpu(dent_a->de_name_len); |
1205 | unsigned int len_b = be16_to_cpu(dent_b->de_name_len); | |
b3b94faa DT |
1206 | |
1207 | if (len_a > len_b) | |
1208 | ret = 1; | |
1209 | else if (len_a < len_b) | |
1210 | ret = -1; | |
1211 | else | |
2bdbc5d7 | 1212 | ret = memcmp(dent_a + 1, dent_b + 1, len_a); |
b3b94faa DT |
1213 | } |
1214 | ||
1215 | return ret; | |
1216 | } | |
1217 | ||
1218 | /** | |
1219 | * do_filldir_main - read out directory entries | |
1220 | * @dip: The GFS2 inode | |
d81a8ef5 | 1221 | * @ctx: what to feed the entries to |
b3b94faa DT |
1222 | * @darr: an array of struct gfs2_dirent pointers to read |
1223 | * @entries: the number of entries in darr | |
1224 | * @copied: pointer to int that's non-zero if a entry has been copied out | |
1225 | * | |
1226 | * Jump through some hoops to make sure that if there are hash collsions, | |
1227 | * they are read out at the beginning of a buffer. We want to minimize | |
1228 | * the possibility that they will fall into different readdir buffers or | |
1229 | * that someone will want to seek to that location. | |
1230 | * | |
d81a8ef5 | 1231 | * Returns: errno, >0 if the actor tells you to stop |
b3b94faa DT |
1232 | */ |
1233 | ||
d81a8ef5 | 1234 | static int do_filldir_main(struct gfs2_inode *dip, struct dir_context *ctx, |
cd915493 | 1235 | const struct gfs2_dirent **darr, u32 entries, |
b3b94faa DT |
1236 | int *copied) |
1237 | { | |
71b86f56 | 1238 | const struct gfs2_dirent *dent, *dent_next; |
cd915493 | 1239 | u64 off, off_next; |
b3b94faa DT |
1240 | unsigned int x, y; |
1241 | int run = 0; | |
b3b94faa DT |
1242 | |
1243 | sort(darr, entries, sizeof(struct gfs2_dirent *), compare_dents, NULL); | |
1244 | ||
1245 | dent_next = darr[0]; | |
1246 | off_next = be32_to_cpu(dent_next->de_hash); | |
1247 | off_next = gfs2_disk_hash2offset(off_next); | |
1248 | ||
1249 | for (x = 0, y = 1; x < entries; x++, y++) { | |
1250 | dent = dent_next; | |
1251 | off = off_next; | |
1252 | ||
1253 | if (y < entries) { | |
1254 | dent_next = darr[y]; | |
1255 | off_next = be32_to_cpu(dent_next->de_hash); | |
1256 | off_next = gfs2_disk_hash2offset(off_next); | |
1257 | ||
d81a8ef5 | 1258 | if (off < ctx->pos) |
b3b94faa | 1259 | continue; |
d81a8ef5 | 1260 | ctx->pos = off; |
b3b94faa DT |
1261 | |
1262 | if (off_next == off) { | |
1263 | if (*copied && !run) | |
1264 | return 1; | |
1265 | run = 1; | |
1266 | } else | |
1267 | run = 0; | |
1268 | } else { | |
d81a8ef5 | 1269 | if (off < ctx->pos) |
b3b94faa | 1270 | continue; |
d81a8ef5 | 1271 | ctx->pos = off; |
b3b94faa DT |
1272 | } |
1273 | ||
d81a8ef5 | 1274 | if (!dir_emit(ctx, (const char *)(dent + 1), |
4dd651ad | 1275 | be16_to_cpu(dent->de_name_len), |
d81a8ef5 AV |
1276 | be64_to_cpu(dent->de_inum.no_addr), |
1277 | be16_to_cpu(dent->de_type))) | |
b3b94faa DT |
1278 | return 1; |
1279 | ||
1280 | *copied = 1; | |
1281 | } | |
1282 | ||
d81a8ef5 | 1283 | /* Increment the ctx->pos by one, so the next time we come into the |
b3b94faa DT |
1284 | do_filldir fxn, we get the next entry instead of the last one in the |
1285 | current leaf */ | |
1286 | ||
d81a8ef5 | 1287 | ctx->pos++; |
b3b94faa DT |
1288 | |
1289 | return 0; | |
1290 | } | |
1291 | ||
d2a97a4e SW |
1292 | static void *gfs2_alloc_sort_buffer(unsigned size) |
1293 | { | |
1294 | void *ptr = NULL; | |
1295 | ||
1296 | if (size < KMALLOC_MAX_SIZE) | |
1297 | ptr = kmalloc(size, GFP_NOFS | __GFP_NOWARN); | |
1298 | if (!ptr) | |
1299 | ptr = __vmalloc(size, GFP_NOFS, PAGE_KERNEL); | |
1300 | return ptr; | |
1301 | } | |
1302 | ||
1303 | static void gfs2_free_sort_buffer(void *ptr) | |
1304 | { | |
1305 | if (is_vmalloc_addr(ptr)) | |
1306 | vfree(ptr); | |
1307 | else | |
1308 | kfree(ptr); | |
1309 | } | |
1310 | ||
d81a8ef5 AV |
1311 | static int gfs2_dir_read_leaf(struct inode *inode, struct dir_context *ctx, |
1312 | int *copied, unsigned *depth, | |
3699e3a4 | 1313 | u64 leaf_no) |
b3b94faa | 1314 | { |
feaa7bba | 1315 | struct gfs2_inode *ip = GFS2_I(inode); |
bdd19a22 | 1316 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
71b86f56 SW |
1317 | struct buffer_head *bh; |
1318 | struct gfs2_leaf *lf; | |
bdd19a22 | 1319 | unsigned entries = 0, entries2 = 0; |
71b86f56 SW |
1320 | unsigned leaves = 0; |
1321 | const struct gfs2_dirent **darr, *dent; | |
1322 | struct dirent_gather g; | |
1323 | struct buffer_head **larr; | |
1324 | int leaf = 0; | |
1325 | int error, i; | |
1326 | u64 lfn = leaf_no; | |
b3b94faa | 1327 | |
b3b94faa | 1328 | do { |
71b86f56 | 1329 | error = get_leaf(ip, lfn, &bh); |
b3b94faa | 1330 | if (error) |
71b86f56 SW |
1331 | goto out; |
1332 | lf = (struct gfs2_leaf *)bh->b_data; | |
1333 | if (leaves == 0) | |
1334 | *depth = be16_to_cpu(lf->lf_depth); | |
1335 | entries += be16_to_cpu(lf->lf_entries); | |
1336 | leaves++; | |
1337 | lfn = be64_to_cpu(lf->lf_next); | |
1338 | brelse(bh); | |
1339 | } while(lfn); | |
b3b94faa DT |
1340 | |
1341 | if (!entries) | |
1342 | return 0; | |
1343 | ||
71b86f56 | 1344 | error = -ENOMEM; |
bdd19a22 SW |
1345 | /* |
1346 | * The extra 99 entries are not normally used, but are a buffer | |
1347 | * zone in case the number of entries in the leaf is corrupt. | |
1348 | * 99 is the maximum number of entries that can fit in a single | |
1349 | * leaf block. | |
1350 | */ | |
d2a97a4e | 1351 | larr = gfs2_alloc_sort_buffer((leaves + entries + 99) * sizeof(void *)); |
71b86f56 SW |
1352 | if (!larr) |
1353 | goto out; | |
1354 | darr = (const struct gfs2_dirent **)(larr + leaves); | |
1355 | g.pdent = darr; | |
1356 | g.offset = 0; | |
1357 | lfn = leaf_no; | |
b3b94faa | 1358 | |
71b86f56 SW |
1359 | do { |
1360 | error = get_leaf(ip, lfn, &bh); | |
b3b94faa | 1361 | if (error) |
d2a97a4e | 1362 | goto out_free; |
71b86f56 SW |
1363 | lf = (struct gfs2_leaf *)bh->b_data; |
1364 | lfn = be64_to_cpu(lf->lf_next); | |
1365 | if (lf->lf_entries) { | |
bdd19a22 | 1366 | entries2 += be16_to_cpu(lf->lf_entries); |
71b86f56 SW |
1367 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, |
1368 | gfs2_dirent_gather, NULL, &g); | |
1369 | error = PTR_ERR(dent); | |
bdd19a22 | 1370 | if (IS_ERR(dent)) |
d2a97a4e | 1371 | goto out_free; |
bdd19a22 | 1372 | if (entries2 != g.offset) { |
f391a4ea | 1373 | fs_warn(sdp, "Number of entries corrupt in dir " |
1374 | "leaf %llu, entries2 (%u) != " | |
1375 | "g.offset (%u)\n", | |
1376 | (unsigned long long)bh->b_blocknr, | |
1377 | entries2, g.offset); | |
bdd19a22 SW |
1378 | |
1379 | error = -EIO; | |
d2a97a4e | 1380 | goto out_free; |
71b86f56 SW |
1381 | } |
1382 | error = 0; | |
1383 | larr[leaf++] = bh; | |
b3b94faa | 1384 | } else { |
71b86f56 | 1385 | brelse(bh); |
b3b94faa | 1386 | } |
71b86f56 | 1387 | } while(lfn); |
b3b94faa | 1388 | |
bdd19a22 | 1389 | BUG_ON(entries2 != entries); |
d81a8ef5 | 1390 | error = do_filldir_main(ip, ctx, darr, entries, copied); |
d2a97a4e | 1391 | out_free: |
71b86f56 SW |
1392 | for(i = 0; i < leaf; i++) |
1393 | brelse(larr[i]); | |
d2a97a4e | 1394 | gfs2_free_sort_buffer(larr); |
71b86f56 | 1395 | out: |
b3b94faa DT |
1396 | return error; |
1397 | } | |
1398 | ||
79c4c379 SW |
1399 | /** |
1400 | * gfs2_dir_readahead - Issue read-ahead requests for leaf blocks. | |
dfe4d34b BP |
1401 | * |
1402 | * Note: we can't calculate each index like dir_e_read can because we don't | |
1403 | * have the leaf, and therefore we don't have the depth, and therefore we | |
1404 | * don't have the length. So we have to just read enough ahead to make up | |
79c4c379 SW |
1405 | * for the loss of information. |
1406 | */ | |
dfe4d34b BP |
1407 | static void gfs2_dir_readahead(struct inode *inode, unsigned hsize, u32 index, |
1408 | struct file_ra_state *f_ra) | |
1409 | { | |
1410 | struct gfs2_inode *ip = GFS2_I(inode); | |
1411 | struct gfs2_glock *gl = ip->i_gl; | |
1412 | struct buffer_head *bh; | |
1413 | u64 blocknr = 0, last; | |
1414 | unsigned count; | |
1415 | ||
1416 | /* First check if we've already read-ahead for the whole range. */ | |
79c4c379 | 1417 | if (index + MAX_RA_BLOCKS < f_ra->start) |
dfe4d34b BP |
1418 | return; |
1419 | ||
1420 | f_ra->start = max((pgoff_t)index, f_ra->start); | |
1421 | for (count = 0; count < MAX_RA_BLOCKS; count++) { | |
1422 | if (f_ra->start >= hsize) /* if exceeded the hash table */ | |
1423 | break; | |
1424 | ||
1425 | last = blocknr; | |
1426 | blocknr = be64_to_cpu(ip->i_hash_cache[f_ra->start]); | |
1427 | f_ra->start++; | |
1428 | if (blocknr == last) | |
1429 | continue; | |
1430 | ||
1431 | bh = gfs2_getbuf(gl, blocknr, 1); | |
1432 | if (trylock_buffer(bh)) { | |
1433 | if (buffer_uptodate(bh)) { | |
1434 | unlock_buffer(bh); | |
1435 | brelse(bh); | |
1436 | continue; | |
1437 | } | |
1438 | bh->b_end_io = end_buffer_read_sync; | |
1439 | submit_bh(READA | REQ_META, bh); | |
1440 | continue; | |
1441 | } | |
1442 | brelse(bh); | |
1443 | } | |
1444 | } | |
17d539f0 | 1445 | |
b3b94faa | 1446 | /** |
c752666c SW |
1447 | * dir_e_read - Reads the entries from a directory into a filldir buffer |
1448 | * @dip: dinode pointer | |
d81a8ef5 | 1449 | * @ctx: actor to feed the entries to |
b3b94faa | 1450 | * |
c752666c | 1451 | * Returns: errno |
b3b94faa DT |
1452 | */ |
1453 | ||
d81a8ef5 AV |
1454 | static int dir_e_read(struct inode *inode, struct dir_context *ctx, |
1455 | struct file_ra_state *f_ra) | |
b3b94faa | 1456 | { |
feaa7bba | 1457 | struct gfs2_inode *dip = GFS2_I(inode); |
cd915493 | 1458 | u32 hsize, len = 0; |
cd915493 | 1459 | u32 hash, index; |
b44b84d7 | 1460 | __be64 *lp; |
c752666c SW |
1461 | int copied = 0; |
1462 | int error = 0; | |
4da3c646 | 1463 | unsigned depth = 0; |
b3b94faa | 1464 | |
9a004508 | 1465 | hsize = 1 << dip->i_depth; |
d81a8ef5 | 1466 | hash = gfs2_dir_offset2hash(ctx->pos); |
9a004508 | 1467 | index = hash >> (32 - dip->i_depth); |
b3b94faa | 1468 | |
79c4c379 | 1469 | if (dip->i_hash_cache == NULL) |
dfe4d34b | 1470 | f_ra->start = 0; |
17d539f0 SW |
1471 | lp = gfs2_dir_get_hash_table(dip); |
1472 | if (IS_ERR(lp)) | |
1473 | return PTR_ERR(lp); | |
b3b94faa | 1474 | |
dfe4d34b BP |
1475 | gfs2_dir_readahead(inode, hsize, index, f_ra); |
1476 | ||
b3b94faa | 1477 | while (index < hsize) { |
d81a8ef5 | 1478 | error = gfs2_dir_read_leaf(inode, ctx, |
71b86f56 | 1479 | &copied, &depth, |
17d539f0 | 1480 | be64_to_cpu(lp[index])); |
b3b94faa | 1481 | if (error) |
71b86f56 | 1482 | break; |
b3b94faa | 1483 | |
9a004508 | 1484 | len = 1 << (dip->i_depth - depth); |
b3b94faa DT |
1485 | index = (index & ~(len - 1)) + len; |
1486 | } | |
1487 | ||
71b86f56 SW |
1488 | if (error > 0) |
1489 | error = 0; | |
b3b94faa DT |
1490 | return error; |
1491 | } | |
1492 | ||
d81a8ef5 AV |
1493 | int gfs2_dir_read(struct inode *inode, struct dir_context *ctx, |
1494 | struct file_ra_state *f_ra) | |
b3b94faa | 1495 | { |
feaa7bba | 1496 | struct gfs2_inode *dip = GFS2_I(inode); |
bdd19a22 | 1497 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
71b86f56 SW |
1498 | struct dirent_gather g; |
1499 | const struct gfs2_dirent **darr, *dent; | |
b3b94faa DT |
1500 | struct buffer_head *dibh; |
1501 | int copied = 0; | |
1502 | int error; | |
1503 | ||
ad6203f2 | 1504 | if (!dip->i_entries) |
71b86f56 SW |
1505 | return 0; |
1506 | ||
383f01fb | 1507 | if (dip->i_diskflags & GFS2_DIF_EXHASH) |
d81a8ef5 | 1508 | return dir_e_read(inode, ctx, f_ra); |
71b86f56 | 1509 | |
b3b94faa DT |
1510 | if (!gfs2_is_stuffed(dip)) { |
1511 | gfs2_consist_inode(dip); | |
1512 | return -EIO; | |
1513 | } | |
1514 | ||
b3b94faa DT |
1515 | error = gfs2_meta_inode_buffer(dip, &dibh); |
1516 | if (error) | |
1517 | return error; | |
1518 | ||
71b86f56 | 1519 | error = -ENOMEM; |
bdd19a22 | 1520 | /* 96 is max number of dirents which can be stuffed into an inode */ |
16c5f06f | 1521 | darr = kmalloc(96 * sizeof(struct gfs2_dirent *), GFP_NOFS); |
71b86f56 SW |
1522 | if (darr) { |
1523 | g.pdent = darr; | |
1524 | g.offset = 0; | |
1525 | dent = gfs2_dirent_scan(inode, dibh->b_data, dibh->b_size, | |
1526 | gfs2_dirent_gather, NULL, &g); | |
1527 | if (IS_ERR(dent)) { | |
1528 | error = PTR_ERR(dent); | |
1529 | goto out; | |
1530 | } | |
ad6203f2 | 1531 | if (dip->i_entries != g.offset) { |
bdd19a22 | 1532 | fs_warn(sdp, "Number of entries corrupt in dir %llu, " |
ad6203f2 | 1533 | "ip->i_entries (%u) != g.offset (%u)\n", |
dbb7cae2 | 1534 | (unsigned long long)dip->i_no_addr, |
ad6203f2 | 1535 | dip->i_entries, |
bdd19a22 SW |
1536 | g.offset); |
1537 | error = -EIO; | |
1538 | goto out; | |
1539 | } | |
d81a8ef5 | 1540 | error = do_filldir_main(dip, ctx, darr, |
ad6203f2 | 1541 | dip->i_entries, &copied); |
71b86f56 SW |
1542 | out: |
1543 | kfree(darr); | |
1544 | } | |
1545 | ||
b3b94faa DT |
1546 | if (error > 0) |
1547 | error = 0; | |
1548 | ||
1549 | brelse(dibh); | |
1550 | ||
1551 | return error; | |
1552 | } | |
1553 | ||
b3b94faa DT |
1554 | /** |
1555 | * gfs2_dir_search - Search a directory | |
5a00f3cc SW |
1556 | * @dip: The GFS2 dir inode |
1557 | * @name: The name we are looking up | |
1558 | * @fail_on_exist: Fail if the name exists rather than looking it up | |
b3b94faa DT |
1559 | * |
1560 | * This routine searches a directory for a file or another directory. | |
1561 | * Assumes a glock is held on dip. | |
1562 | * | |
1563 | * Returns: errno | |
1564 | */ | |
1565 | ||
5a00f3cc SW |
1566 | struct inode *gfs2_dir_search(struct inode *dir, const struct qstr *name, |
1567 | bool fail_on_exist) | |
b3b94faa | 1568 | { |
c752666c SW |
1569 | struct buffer_head *bh; |
1570 | struct gfs2_dirent *dent; | |
5a00f3cc SW |
1571 | u64 addr, formal_ino; |
1572 | u16 dtype; | |
dbb7cae2 SW |
1573 | |
1574 | dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh); | |
1575 | if (dent) { | |
1576 | if (IS_ERR(dent)) | |
e231c2ee | 1577 | return ERR_CAST(dent); |
5a00f3cc SW |
1578 | dtype = be16_to_cpu(dent->de_type); |
1579 | addr = be64_to_cpu(dent->de_inum.no_addr); | |
1580 | formal_ino = be64_to_cpu(dent->de_inum.no_formal_ino); | |
dbb7cae2 | 1581 | brelse(bh); |
5a00f3cc SW |
1582 | if (fail_on_exist) |
1583 | return ERR_PTR(-EEXIST); | |
1584 | return gfs2_inode_lookup(dir->i_sb, dtype, addr, formal_ino, 0); | |
dbb7cae2 SW |
1585 | } |
1586 | return ERR_PTR(-ENOENT); | |
1587 | } | |
1588 | ||
1589 | int gfs2_dir_check(struct inode *dir, const struct qstr *name, | |
1590 | const struct gfs2_inode *ip) | |
1591 | { | |
1592 | struct buffer_head *bh; | |
1593 | struct gfs2_dirent *dent; | |
1594 | int ret = -ENOENT; | |
c752666c SW |
1595 | |
1596 | dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh); | |
1597 | if (dent) { | |
1598 | if (IS_ERR(dent)) | |
1599 | return PTR_ERR(dent); | |
dbb7cae2 SW |
1600 | if (ip) { |
1601 | if (be64_to_cpu(dent->de_inum.no_addr) != ip->i_no_addr) | |
1602 | goto out; | |
1603 | if (be64_to_cpu(dent->de_inum.no_formal_ino) != | |
1604 | ip->i_no_formal_ino) | |
1605 | goto out; | |
1606 | if (unlikely(IF2DT(ip->i_inode.i_mode) != | |
1607 | be16_to_cpu(dent->de_type))) { | |
1608 | gfs2_consist_inode(GFS2_I(dir)); | |
1609 | ret = -EIO; | |
1610 | goto out; | |
1611 | } | |
1612 | } | |
1613 | ret = 0; | |
1614 | out: | |
c752666c | 1615 | brelse(bh); |
c752666c | 1616 | } |
dbb7cae2 | 1617 | return ret; |
c752666c SW |
1618 | } |
1619 | ||
01bcb0de SW |
1620 | /** |
1621 | * dir_new_leaf - Add a new leaf onto hash chain | |
1622 | * @inode: The directory | |
1623 | * @name: The name we are adding | |
1624 | * | |
1625 | * This adds a new dir leaf onto an existing leaf when there is not | |
1626 | * enough space to add a new dir entry. This is a last resort after | |
1627 | * we've expanded the hash table to max size and also split existing | |
1628 | * leaf blocks, so it will only occur for very large directories. | |
1629 | * | |
1630 | * The dist parameter is set to 1 for leaf blocks directly attached | |
1631 | * to the hash table, 2 for one layer of indirection, 3 for two layers | |
1632 | * etc. We are thus able to tell the difference between an old leaf | |
1633 | * with dist set to zero (i.e. "don't know") and a new one where we | |
1634 | * set this information for debug/fsck purposes. | |
1635 | * | |
1636 | * Returns: 0 on success, or -ve on error | |
1637 | */ | |
1638 | ||
c752666c SW |
1639 | static int dir_new_leaf(struct inode *inode, const struct qstr *name) |
1640 | { | |
1641 | struct buffer_head *bh, *obh; | |
feaa7bba | 1642 | struct gfs2_inode *ip = GFS2_I(inode); |
c752666c | 1643 | struct gfs2_leaf *leaf, *oleaf; |
01bcb0de | 1644 | u32 dist = 1; |
b3b94faa | 1645 | int error; |
c752666c SW |
1646 | u32 index; |
1647 | u64 bn; | |
b3b94faa | 1648 | |
9a004508 | 1649 | index = name->hash >> (32 - ip->i_depth); |
c752666c SW |
1650 | error = get_first_leaf(ip, index, &obh); |
1651 | if (error) | |
1652 | return error; | |
1653 | do { | |
01bcb0de | 1654 | dist++; |
c752666c SW |
1655 | oleaf = (struct gfs2_leaf *)obh->b_data; |
1656 | bn = be64_to_cpu(oleaf->lf_next); | |
1657 | if (!bn) | |
1658 | break; | |
1659 | brelse(obh); | |
1660 | error = get_leaf(ip, bn, &obh); | |
1661 | if (error) | |
1662 | return error; | |
1663 | } while(1); | |
b3b94faa | 1664 | |
350a9b0a | 1665 | gfs2_trans_add_meta(ip->i_gl, obh); |
c752666c SW |
1666 | |
1667 | leaf = new_leaf(inode, &bh, be16_to_cpu(oleaf->lf_depth)); | |
1668 | if (!leaf) { | |
1669 | brelse(obh); | |
1670 | return -ENOSPC; | |
1671 | } | |
01bcb0de | 1672 | leaf->lf_dist = cpu_to_be32(dist); |
4d8012b6 | 1673 | oleaf->lf_next = cpu_to_be64(bh->b_blocknr); |
c752666c SW |
1674 | brelse(bh); |
1675 | brelse(obh); | |
1676 | ||
1677 | error = gfs2_meta_inode_buffer(ip, &bh); | |
1678 | if (error) | |
1679 | return error; | |
350a9b0a | 1680 | gfs2_trans_add_meta(ip->i_gl, bh); |
77658aad | 1681 | gfs2_add_inode_blocks(&ip->i_inode, 1); |
539e5d6b | 1682 | gfs2_dinode_out(ip, bh->b_data); |
c752666c SW |
1683 | brelse(bh); |
1684 | return 0; | |
b3b94faa DT |
1685 | } |
1686 | ||
1687 | /** | |
1688 | * gfs2_dir_add - Add new filename into directory | |
2b47dad8 SW |
1689 | * @inode: The directory inode |
1690 | * @name: The new name | |
1691 | * @nip: The GFS2 inode to be linked in to the directory | |
1692 | * @da: The directory addition info | |
1693 | * | |
1694 | * If the call to gfs2_diradd_alloc_required resulted in there being | |
1695 | * no need to allocate any new directory blocks, then it will contain | |
1696 | * a pointer to the directory entry and the bh in which it resides. We | |
1697 | * can use that without having to repeat the search. If there was no | |
1698 | * free space, then we must now create more space. | |
b3b94faa DT |
1699 | * |
1700 | * Returns: 0 on success, error code on failure | |
1701 | */ | |
1702 | ||
c752666c | 1703 | int gfs2_dir_add(struct inode *inode, const struct qstr *name, |
2b47dad8 | 1704 | const struct gfs2_inode *nip, struct gfs2_diradd *da) |
b3b94faa | 1705 | { |
feaa7bba | 1706 | struct gfs2_inode *ip = GFS2_I(inode); |
2b47dad8 SW |
1707 | struct buffer_head *bh = da->bh; |
1708 | struct gfs2_dirent *dent = da->dent; | |
01bcb0de | 1709 | struct timespec tv; |
c752666c | 1710 | struct gfs2_leaf *leaf; |
b3b94faa DT |
1711 | int error; |
1712 | ||
c752666c | 1713 | while(1) { |
2b47dad8 SW |
1714 | if (da->bh == NULL) { |
1715 | dent = gfs2_dirent_search(inode, name, | |
1716 | gfs2_dirent_find_space, &bh); | |
1717 | } | |
c752666c SW |
1718 | if (dent) { |
1719 | if (IS_ERR(dent)) | |
1720 | return PTR_ERR(dent); | |
1721 | dent = gfs2_init_dirent(inode, dent, name, bh); | |
dbb7cae2 | 1722 | gfs2_inum_out(nip, dent); |
3d6ecb7d | 1723 | dent->de_type = cpu_to_be16(IF2DT(nip->i_inode.i_mode)); |
01bcb0de | 1724 | tv = CURRENT_TIME; |
383f01fb | 1725 | if (ip->i_diskflags & GFS2_DIF_EXHASH) { |
c752666c | 1726 | leaf = (struct gfs2_leaf *)bh->b_data; |
bb16b342 | 1727 | be16_add_cpu(&leaf->lf_entries, 1); |
01bcb0de SW |
1728 | leaf->lf_nsec = cpu_to_be32(tv.tv_nsec); |
1729 | leaf->lf_sec = cpu_to_be64(tv.tv_sec); | |
c752666c | 1730 | } |
2b47dad8 SW |
1731 | da->dent = NULL; |
1732 | da->bh = NULL; | |
c752666c | 1733 | brelse(bh); |
ad6203f2 | 1734 | ip->i_entries++; |
01bcb0de | 1735 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = tv; |
3d6ecb7d SW |
1736 | if (S_ISDIR(nip->i_inode.i_mode)) |
1737 | inc_nlink(&ip->i_inode); | |
343cd8f0 | 1738 | mark_inode_dirty(inode); |
c752666c SW |
1739 | error = 0; |
1740 | break; | |
1741 | } | |
383f01fb | 1742 | if (!(ip->i_diskflags & GFS2_DIF_EXHASH)) { |
c752666c SW |
1743 | error = dir_make_exhash(inode); |
1744 | if (error) | |
1745 | break; | |
1746 | continue; | |
1747 | } | |
1748 | error = dir_split_leaf(inode, name); | |
1749 | if (error == 0) | |
1750 | continue; | |
e90deff5 | 1751 | if (error < 0) |
c752666c | 1752 | break; |
9a004508 | 1753 | if (ip->i_depth < GFS2_DIR_MAX_DEPTH) { |
c752666c SW |
1754 | error = dir_double_exhash(ip); |
1755 | if (error) | |
1756 | break; | |
1757 | error = dir_split_leaf(inode, name); | |
e90deff5 | 1758 | if (error < 0) |
c752666c | 1759 | break; |
e90deff5 SW |
1760 | if (error == 0) |
1761 | continue; | |
c752666c SW |
1762 | } |
1763 | error = dir_new_leaf(inode, name); | |
1764 | if (!error) | |
1765 | continue; | |
1766 | error = -ENOSPC; | |
1767 | break; | |
1768 | } | |
b3b94faa DT |
1769 | return error; |
1770 | } | |
1771 | ||
c752666c | 1772 | |
b3b94faa DT |
1773 | /** |
1774 | * gfs2_dir_del - Delete a directory entry | |
1775 | * @dip: The GFS2 inode | |
1776 | * @filename: The filename | |
1777 | * | |
1778 | * Returns: 0 on success, error code on failure | |
1779 | */ | |
1780 | ||
855d23ce | 1781 | int gfs2_dir_del(struct gfs2_inode *dip, const struct dentry *dentry) |
b3b94faa | 1782 | { |
855d23ce | 1783 | const struct qstr *name = &dentry->d_name; |
c752666c SW |
1784 | struct gfs2_dirent *dent, *prev = NULL; |
1785 | struct buffer_head *bh; | |
01bcb0de | 1786 | struct timespec tv = CURRENT_TIME; |
b3b94faa | 1787 | |
c752666c SW |
1788 | /* Returns _either_ the entry (if its first in block) or the |
1789 | previous entry otherwise */ | |
feaa7bba | 1790 | dent = gfs2_dirent_search(&dip->i_inode, name, gfs2_dirent_prev, &bh); |
c752666c SW |
1791 | if (!dent) { |
1792 | gfs2_consist_inode(dip); | |
1793 | return -EIO; | |
1794 | } | |
1795 | if (IS_ERR(dent)) { | |
1796 | gfs2_consist_inode(dip); | |
1797 | return PTR_ERR(dent); | |
1798 | } | |
1799 | /* If not first in block, adjust pointers accordingly */ | |
71b86f56 | 1800 | if (gfs2_dirent_find(dent, name, NULL) == 0) { |
c752666c SW |
1801 | prev = dent; |
1802 | dent = (struct gfs2_dirent *)((char *)dent + be16_to_cpu(prev->de_rec_len)); | |
1803 | } | |
1804 | ||
1805 | dirent_del(dip, bh, prev, dent); | |
383f01fb | 1806 | if (dip->i_diskflags & GFS2_DIF_EXHASH) { |
c752666c SW |
1807 | struct gfs2_leaf *leaf = (struct gfs2_leaf *)bh->b_data; |
1808 | u16 entries = be16_to_cpu(leaf->lf_entries); | |
1809 | if (!entries) | |
1810 | gfs2_consist_inode(dip); | |
1811 | leaf->lf_entries = cpu_to_be16(--entries); | |
01bcb0de SW |
1812 | leaf->lf_nsec = cpu_to_be32(tv.tv_nsec); |
1813 | leaf->lf_sec = cpu_to_be64(tv.tv_sec); | |
c752666c | 1814 | } |
ed386507 | 1815 | brelse(bh); |
c752666c | 1816 | |
ad6203f2 | 1817 | if (!dip->i_entries) |
c752666c | 1818 | gfs2_consist_inode(dip); |
ad6203f2 | 1819 | dip->i_entries--; |
01bcb0de | 1820 | dip->i_inode.i_mtime = dip->i_inode.i_ctime = tv; |
855d23ce SW |
1821 | if (S_ISDIR(dentry->d_inode->i_mode)) |
1822 | drop_nlink(&dip->i_inode); | |
feaa7bba | 1823 | mark_inode_dirty(&dip->i_inode); |
b3b94faa | 1824 | |
ab9bbda0 | 1825 | return 0; |
b3b94faa DT |
1826 | } |
1827 | ||
b3b94faa DT |
1828 | /** |
1829 | * gfs2_dir_mvino - Change inode number of directory entry | |
1830 | * @dip: The GFS2 inode | |
1831 | * @filename: | |
1832 | * @new_inode: | |
1833 | * | |
1834 | * This routine changes the inode number of a directory entry. It's used | |
1835 | * by rename to change ".." when a directory is moved. | |
1836 | * Assumes a glock is held on dvp. | |
1837 | * | |
1838 | * Returns: errno | |
1839 | */ | |
1840 | ||
c752666c | 1841 | int gfs2_dir_mvino(struct gfs2_inode *dip, const struct qstr *filename, |
dbb7cae2 | 1842 | const struct gfs2_inode *nip, unsigned int new_type) |
b3b94faa | 1843 | { |
c752666c SW |
1844 | struct buffer_head *bh; |
1845 | struct gfs2_dirent *dent; | |
b3b94faa DT |
1846 | int error; |
1847 | ||
feaa7bba | 1848 | dent = gfs2_dirent_search(&dip->i_inode, filename, gfs2_dirent_find, &bh); |
c752666c SW |
1849 | if (!dent) { |
1850 | gfs2_consist_inode(dip); | |
1851 | return -EIO; | |
1852 | } | |
1853 | if (IS_ERR(dent)) | |
1854 | return PTR_ERR(dent); | |
b3b94faa | 1855 | |
350a9b0a | 1856 | gfs2_trans_add_meta(dip->i_gl, bh); |
dbb7cae2 | 1857 | gfs2_inum_out(nip, dent); |
c752666c SW |
1858 | dent->de_type = cpu_to_be16(new_type); |
1859 | ||
383f01fb | 1860 | if (dip->i_diskflags & GFS2_DIF_EXHASH) { |
c752666c SW |
1861 | brelse(bh); |
1862 | error = gfs2_meta_inode_buffer(dip, &bh); | |
1863 | if (error) | |
1864 | return error; | |
350a9b0a | 1865 | gfs2_trans_add_meta(dip->i_gl, bh); |
c752666c SW |
1866 | } |
1867 | ||
4bd91ba1 | 1868 | dip->i_inode.i_mtime = dip->i_inode.i_ctime = CURRENT_TIME; |
539e5d6b | 1869 | gfs2_dinode_out(dip, bh->b_data); |
c752666c SW |
1870 | brelse(bh); |
1871 | return 0; | |
b3b94faa DT |
1872 | } |
1873 | ||
b3b94faa DT |
1874 | /** |
1875 | * leaf_dealloc - Deallocate a directory leaf | |
1876 | * @dip: the directory | |
1877 | * @index: the hash table offset in the directory | |
1878 | * @len: the number of pointers to this leaf | |
1879 | * @leaf_no: the leaf number | |
ec038c82 | 1880 | * @leaf_bh: buffer_head for the starting leaf |
d24a7a43 | 1881 | * last_dealloc: 1 if this is the final dealloc for the leaf, else 0 |
b3b94faa DT |
1882 | * |
1883 | * Returns: errno | |
1884 | */ | |
1885 | ||
cd915493 | 1886 | static int leaf_dealloc(struct gfs2_inode *dip, u32 index, u32 len, |
ec038c82 BP |
1887 | u64 leaf_no, struct buffer_head *leaf_bh, |
1888 | int last_dealloc) | |
b3b94faa | 1889 | { |
feaa7bba | 1890 | struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); |
c752666c | 1891 | struct gfs2_leaf *tmp_leaf; |
b3b94faa DT |
1892 | struct gfs2_rgrp_list rlist; |
1893 | struct buffer_head *bh, *dibh; | |
cd915493 | 1894 | u64 blk, nblk; |
b3b94faa DT |
1895 | unsigned int rg_blocks = 0, l_blocks = 0; |
1896 | char *ht; | |
cd915493 | 1897 | unsigned int x, size = len * sizeof(u64); |
b3b94faa DT |
1898 | int error; |
1899 | ||
5e2f7d61 BP |
1900 | error = gfs2_rindex_update(sdp); |
1901 | if (error) | |
1902 | return error; | |
1903 | ||
b3b94faa DT |
1904 | memset(&rlist, 0, sizeof(struct gfs2_rgrp_list)); |
1905 | ||
8be04b93 | 1906 | ht = kzalloc(size, GFP_NOFS | __GFP_NOWARN); |
e8830d88 BP |
1907 | if (ht == NULL) |
1908 | ht = vzalloc(size); | |
b3b94faa DT |
1909 | if (!ht) |
1910 | return -ENOMEM; | |
1911 | ||
f4108a60 | 1912 | error = gfs2_quota_hold(dip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE); |
b3b94faa | 1913 | if (error) |
5407e242 | 1914 | goto out; |
b3b94faa | 1915 | |
b3b94faa | 1916 | /* Count the number of leaves */ |
ec038c82 | 1917 | bh = leaf_bh; |
b3b94faa | 1918 | |
c752666c | 1919 | for (blk = leaf_no; blk; blk = nblk) { |
ec038c82 BP |
1920 | if (blk != leaf_no) { |
1921 | error = get_leaf(dip, blk, &bh); | |
1922 | if (error) | |
1923 | goto out_rlist; | |
1924 | } | |
c752666c SW |
1925 | tmp_leaf = (struct gfs2_leaf *)bh->b_data; |
1926 | nblk = be64_to_cpu(tmp_leaf->lf_next); | |
ec038c82 BP |
1927 | if (blk != leaf_no) |
1928 | brelse(bh); | |
b3b94faa | 1929 | |
70b0c365 | 1930 | gfs2_rlist_add(dip, &rlist, blk); |
b3b94faa DT |
1931 | l_blocks++; |
1932 | } | |
1933 | ||
fe6c991c | 1934 | gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE); |
b3b94faa DT |
1935 | |
1936 | for (x = 0; x < rlist.rl_rgrps; x++) { | |
1937 | struct gfs2_rgrpd *rgd; | |
5c676f6d | 1938 | rgd = rlist.rl_ghs[x].gh_gl->gl_object; |
bb8d8a6f | 1939 | rg_blocks += rgd->rd_length; |
b3b94faa DT |
1940 | } |
1941 | ||
1942 | error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs); | |
1943 | if (error) | |
1944 | goto out_rlist; | |
1945 | ||
1946 | error = gfs2_trans_begin(sdp, | |
5c676f6d | 1947 | rg_blocks + (DIV_ROUND_UP(size, sdp->sd_jbsize) + 1) + |
b3b94faa DT |
1948 | RES_DINODE + RES_STATFS + RES_QUOTA, l_blocks); |
1949 | if (error) | |
1950 | goto out_rg_gunlock; | |
1951 | ||
ec038c82 BP |
1952 | bh = leaf_bh; |
1953 | ||
c752666c | 1954 | for (blk = leaf_no; blk; blk = nblk) { |
ec038c82 BP |
1955 | if (blk != leaf_no) { |
1956 | error = get_leaf(dip, blk, &bh); | |
1957 | if (error) | |
1958 | goto out_end_trans; | |
1959 | } | |
c752666c SW |
1960 | tmp_leaf = (struct gfs2_leaf *)bh->b_data; |
1961 | nblk = be64_to_cpu(tmp_leaf->lf_next); | |
ec038c82 BP |
1962 | if (blk != leaf_no) |
1963 | brelse(bh); | |
b3b94faa DT |
1964 | |
1965 | gfs2_free_meta(dip, blk, 1); | |
77658aad | 1966 | gfs2_add_inode_blocks(&dip->i_inode, -1); |
b3b94faa DT |
1967 | } |
1968 | ||
cd915493 | 1969 | error = gfs2_dir_write_data(dip, ht, index * sizeof(u64), size); |
b3b94faa DT |
1970 | if (error != size) { |
1971 | if (error >= 0) | |
1972 | error = -EIO; | |
1973 | goto out_end_trans; | |
1974 | } | |
1975 | ||
1976 | error = gfs2_meta_inode_buffer(dip, &dibh); | |
1977 | if (error) | |
1978 | goto out_end_trans; | |
1979 | ||
350a9b0a | 1980 | gfs2_trans_add_meta(dip->i_gl, dibh); |
d24a7a43 BP |
1981 | /* On the last dealloc, make this a regular file in case we crash. |
1982 | (We don't want to free these blocks a second time.) */ | |
1983 | if (last_dealloc) | |
1984 | dip->i_inode.i_mode = S_IFREG; | |
539e5d6b | 1985 | gfs2_dinode_out(dip, dibh->b_data); |
b3b94faa DT |
1986 | brelse(dibh); |
1987 | ||
a91ea69f | 1988 | out_end_trans: |
b3b94faa | 1989 | gfs2_trans_end(sdp); |
a91ea69f | 1990 | out_rg_gunlock: |
b3b94faa | 1991 | gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs); |
a91ea69f | 1992 | out_rlist: |
b3b94faa | 1993 | gfs2_rlist_free(&rlist); |
b3b94faa | 1994 | gfs2_quota_unhold(dip); |
182fe5ab | 1995 | out: |
e8830d88 BP |
1996 | if (is_vmalloc_addr(ht)) |
1997 | vfree(ht); | |
1998 | else | |
1999 | kfree(ht); | |
b3b94faa DT |
2000 | return error; |
2001 | } | |
2002 | ||
2003 | /** | |
2004 | * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory | |
2005 | * @dip: the directory | |
2006 | * | |
2007 | * Dealloc all on-disk directory leaves to FREEMETA state | |
2008 | * Change on-disk inode type to "regular file" | |
2009 | * | |
2010 | * Returns: errno | |
2011 | */ | |
2012 | ||
2013 | int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip) | |
2014 | { | |
556bb179 BP |
2015 | struct buffer_head *bh; |
2016 | struct gfs2_leaf *leaf; | |
2017 | u32 hsize, len; | |
556bb179 BP |
2018 | u32 index = 0, next_index; |
2019 | __be64 *lp; | |
2020 | u64 leaf_no; | |
2021 | int error = 0, last; | |
2022 | ||
2023 | hsize = 1 << dip->i_depth; | |
556bb179 | 2024 | |
17d539f0 SW |
2025 | lp = gfs2_dir_get_hash_table(dip); |
2026 | if (IS_ERR(lp)) | |
2027 | return PTR_ERR(lp); | |
556bb179 BP |
2028 | |
2029 | while (index < hsize) { | |
17d539f0 | 2030 | leaf_no = be64_to_cpu(lp[index]); |
556bb179 BP |
2031 | if (leaf_no) { |
2032 | error = get_leaf(dip, leaf_no, &bh); | |
2033 | if (error) | |
2034 | goto out; | |
2035 | leaf = (struct gfs2_leaf *)bh->b_data; | |
2036 | len = 1 << (dip->i_depth - be16_to_cpu(leaf->lf_depth)); | |
2037 | ||
2038 | next_index = (index & ~(len - 1)) + len; | |
2039 | last = ((next_index >= hsize) ? 1 : 0); | |
2040 | error = leaf_dealloc(dip, index, len, leaf_no, bh, | |
2041 | last); | |
2042 | brelse(bh); | |
2043 | if (error) | |
2044 | goto out; | |
2045 | index = next_index; | |
2046 | } else | |
2047 | index++; | |
2048 | } | |
2049 | ||
2050 | if (index != hsize) { | |
2051 | gfs2_consist_inode(dip); | |
2052 | error = -EIO; | |
2053 | } | |
2054 | ||
2055 | out: | |
556bb179 BP |
2056 | |
2057 | return error; | |
b3b94faa DT |
2058 | } |
2059 | ||
2060 | /** | |
2061 | * gfs2_diradd_alloc_required - find if adding entry will require an allocation | |
2062 | * @ip: the file being written to | |
2063 | * @filname: the filename that's going to be added | |
3c1c0ae1 | 2064 | * @da: The structure to return dir alloc info |
b3b94faa | 2065 | * |
3c1c0ae1 | 2066 | * Returns: 0 if ok, -ve on error |
b3b94faa DT |
2067 | */ |
2068 | ||
3c1c0ae1 SW |
2069 | int gfs2_diradd_alloc_required(struct inode *inode, const struct qstr *name, |
2070 | struct gfs2_diradd *da) | |
b3b94faa | 2071 | { |
22b5a6c0 | 2072 | struct gfs2_inode *ip = GFS2_I(inode); |
3c1c0ae1 | 2073 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
22b5a6c0 | 2074 | const unsigned int extra = sizeof(struct gfs2_dinode) - sizeof(struct gfs2_leaf); |
c752666c SW |
2075 | struct gfs2_dirent *dent; |
2076 | struct buffer_head *bh; | |
b3b94faa | 2077 | |
3c1c0ae1 | 2078 | da->nr_blocks = 0; |
2b47dad8 SW |
2079 | da->bh = NULL; |
2080 | da->dent = NULL; | |
3c1c0ae1 | 2081 | |
c752666c | 2082 | dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, &bh); |
ed386507 | 2083 | if (!dent) { |
3c1c0ae1 | 2084 | da->nr_blocks = sdp->sd_max_dirres; |
22b5a6c0 SW |
2085 | if (!(ip->i_diskflags & GFS2_DIF_EXHASH) && |
2086 | (GFS2_DIRENT_SIZE(name->len) < extra)) | |
2087 | da->nr_blocks = 1; | |
3c1c0ae1 | 2088 | return 0; |
ed386507 | 2089 | } |
c752666c SW |
2090 | if (IS_ERR(dent)) |
2091 | return PTR_ERR(dent); | |
2b47dad8 SW |
2092 | da->bh = bh; |
2093 | da->dent = dent; | |
c752666c | 2094 | return 0; |
b3b94faa DT |
2095 | } |
2096 |