]>
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 | ||
cd915493 SW |
79 | #define gfs2_disk_hash2offset(h) (((u64)(h)) >> 1) |
80 | #define gfs2_dir_offset2hash(p) ((u32)(((u64)(p)) << 1)) | |
b3b94faa | 81 | |
8d123585 SW |
82 | struct qstr gfs2_qdot __read_mostly; |
83 | struct qstr gfs2_qdotdot __read_mostly; | |
84 | ||
2bdbc5d7 SW |
85 | typedef int (*gfs2_dscan_t)(const struct gfs2_dirent *dent, |
86 | const struct qstr *name, void *opaque); | |
b3b94faa | 87 | |
cd915493 | 88 | int gfs2_dir_get_new_buffer(struct gfs2_inode *ip, u64 block, |
61e085a8 | 89 | struct buffer_head **bhp) |
e13940ba SW |
90 | { |
91 | struct buffer_head *bh; | |
e13940ba | 92 | |
61e085a8 SW |
93 | bh = gfs2_meta_new(ip->i_gl, block); |
94 | gfs2_trans_add_bh(ip->i_gl, bh, 1); | |
95 | gfs2_metatype_set(bh, GFS2_METATYPE_JD, GFS2_FORMAT_JD); | |
96 | gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header)); | |
e13940ba SW |
97 | *bhp = bh; |
98 | return 0; | |
99 | } | |
100 | ||
cd915493 | 101 | static int gfs2_dir_get_existing_buffer(struct gfs2_inode *ip, u64 block, |
61e085a8 SW |
102 | struct buffer_head **bhp) |
103 | { | |
104 | struct buffer_head *bh; | |
105 | int error; | |
e13940ba | 106 | |
7276b3b0 | 107 | error = gfs2_meta_read(ip->i_gl, block, DIO_WAIT, &bh); |
61e085a8 SW |
108 | if (error) |
109 | return error; | |
feaa7bba | 110 | if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), bh, GFS2_METATYPE_JD)) { |
61e085a8 SW |
111 | brelse(bh); |
112 | return -EIO; | |
113 | } | |
114 | *bhp = bh; | |
115 | return 0; | |
116 | } | |
e13940ba SW |
117 | |
118 | static int gfs2_dir_write_stuffed(struct gfs2_inode *ip, const char *buf, | |
119 | unsigned int offset, unsigned int size) | |
e13940ba SW |
120 | { |
121 | struct buffer_head *dibh; | |
122 | int error; | |
123 | ||
124 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
125 | if (error) | |
126 | return error; | |
127 | ||
128 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); | |
c752666c | 129 | memcpy(dibh->b_data + offset + sizeof(struct gfs2_dinode), buf, size); |
a2e0f799 SW |
130 | if (ip->i_inode.i_size < offset + size) |
131 | i_size_write(&ip->i_inode, offset + size); | |
4bd91ba1 | 132 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME; |
539e5d6b | 133 | gfs2_dinode_out(ip, dibh->b_data); |
e13940ba SW |
134 | |
135 | brelse(dibh); | |
136 | ||
137 | return size; | |
138 | } | |
139 | ||
140 | ||
141 | ||
142 | /** | |
143 | * gfs2_dir_write_data - Write directory information to the inode | |
144 | * @ip: The GFS2 inode | |
145 | * @buf: The buffer containing information to be written | |
146 | * @offset: The file offset to start writing at | |
147 | * @size: The amount of data to write | |
148 | * | |
149 | * Returns: The number of bytes correctly written or error code | |
150 | */ | |
151 | static int gfs2_dir_write_data(struct gfs2_inode *ip, const char *buf, | |
cd915493 | 152 | u64 offset, unsigned int size) |
e13940ba | 153 | { |
feaa7bba | 154 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
e13940ba | 155 | struct buffer_head *dibh; |
cd915493 SW |
156 | u64 lblock, dblock; |
157 | u32 extlen = 0; | |
e13940ba SW |
158 | unsigned int o; |
159 | int copied = 0; | |
160 | int error = 0; | |
9b8c81d1 | 161 | int new = 0; |
e13940ba SW |
162 | |
163 | if (!size) | |
164 | return 0; | |
165 | ||
166 | if (gfs2_is_stuffed(ip) && | |
167 | offset + size <= sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) | |
568f4c96 SW |
168 | return gfs2_dir_write_stuffed(ip, buf, (unsigned int)offset, |
169 | size); | |
e13940ba SW |
170 | |
171 | if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip))) | |
172 | return -EINVAL; | |
173 | ||
174 | if (gfs2_is_stuffed(ip)) { | |
f25ef0c1 | 175 | error = gfs2_unstuff_dinode(ip, NULL); |
e13940ba | 176 | if (error) |
c752666c | 177 | return error; |
e13940ba SW |
178 | } |
179 | ||
180 | lblock = offset; | |
181 | o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header); | |
182 | ||
183 | while (copied < size) { | |
184 | unsigned int amount; | |
185 | struct buffer_head *bh; | |
e13940ba SW |
186 | |
187 | amount = size - copied; | |
188 | if (amount > sdp->sd_sb.sb_bsize - o) | |
189 | amount = sdp->sd_sb.sb_bsize - o; | |
190 | ||
191 | if (!extlen) { | |
192 | new = 1; | |
feaa7bba | 193 | error = gfs2_extent_map(&ip->i_inode, lblock, &new, |
fd88de56 | 194 | &dblock, &extlen); |
e13940ba SW |
195 | if (error) |
196 | goto fail; | |
197 | error = -EIO; | |
198 | if (gfs2_assert_withdraw(sdp, dblock)) | |
199 | goto fail; | |
200 | } | |
201 | ||
61e085a8 SW |
202 | if (amount == sdp->sd_jbsize || new) |
203 | error = gfs2_dir_get_new_buffer(ip, dblock, &bh); | |
204 | else | |
205 | error = gfs2_dir_get_existing_buffer(ip, dblock, &bh); | |
206 | ||
e13940ba SW |
207 | if (error) |
208 | goto fail; | |
209 | ||
210 | gfs2_trans_add_bh(ip->i_gl, bh, 1); | |
211 | memcpy(bh->b_data + o, buf, amount); | |
212 | brelse(bh); | |
e13940ba | 213 | |
899bb264 | 214 | buf += amount; |
e13940ba SW |
215 | copied += amount; |
216 | lblock++; | |
217 | dblock++; | |
218 | extlen--; | |
219 | ||
220 | o = sizeof(struct gfs2_meta_header); | |
221 | } | |
222 | ||
223 | out: | |
224 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
225 | if (error) | |
226 | return error; | |
227 | ||
a2e0f799 SW |
228 | if (ip->i_inode.i_size < offset + copied) |
229 | i_size_write(&ip->i_inode, offset + copied); | |
4bd91ba1 | 230 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME; |
e13940ba SW |
231 | |
232 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); | |
539e5d6b | 233 | gfs2_dinode_out(ip, dibh->b_data); |
e13940ba SW |
234 | brelse(dibh); |
235 | ||
236 | return copied; | |
237 | fail: | |
238 | if (copied) | |
239 | goto out; | |
240 | return error; | |
241 | } | |
242 | ||
243 | static int gfs2_dir_read_stuffed(struct gfs2_inode *ip, char *buf, | |
7276b3b0 | 244 | u64 offset, unsigned int size) |
e13940ba SW |
245 | { |
246 | struct buffer_head *dibh; | |
247 | int error; | |
248 | ||
249 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
250 | if (!error) { | |
251 | offset += sizeof(struct gfs2_dinode); | |
252 | memcpy(buf, dibh->b_data + offset, size); | |
253 | brelse(dibh); | |
254 | } | |
255 | ||
256 | return (error) ? error : size; | |
257 | } | |
258 | ||
259 | ||
260 | /** | |
261 | * gfs2_dir_read_data - Read a data from a directory inode | |
262 | * @ip: The GFS2 Inode | |
263 | * @buf: The buffer to place result into | |
264 | * @offset: File offset to begin jdata_readng from | |
265 | * @size: Amount of data to transfer | |
266 | * | |
267 | * Returns: The amount of data actually copied or the error | |
268 | */ | |
7276b3b0 SW |
269 | static int gfs2_dir_read_data(struct gfs2_inode *ip, char *buf, u64 offset, |
270 | unsigned int size, unsigned ra) | |
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; | |
a2e0f799 | 278 | u64 disksize = i_size_read(&ip->i_inode); |
e13940ba | 279 | |
a2e0f799 | 280 | if (offset >= disksize) |
e13940ba SW |
281 | return 0; |
282 | ||
a2e0f799 SW |
283 | if (offset + size > disksize) |
284 | size = disksize - offset; | |
e13940ba SW |
285 | |
286 | if (!size) | |
287 | return 0; | |
288 | ||
289 | if (gfs2_is_stuffed(ip)) | |
7276b3b0 | 290 | return gfs2_dir_read_stuffed(ip, buf, offset, size); |
e13940ba SW |
291 | |
292 | if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip))) | |
293 | return -EINVAL; | |
294 | ||
295 | lblock = offset; | |
296 | o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header); | |
297 | ||
298 | while (copied < size) { | |
299 | unsigned int amount; | |
300 | struct buffer_head *bh; | |
301 | int new; | |
302 | ||
303 | amount = size - copied; | |
304 | if (amount > sdp->sd_sb.sb_bsize - o) | |
305 | amount = sdp->sd_sb.sb_bsize - o; | |
306 | ||
307 | if (!extlen) { | |
308 | new = 0; | |
feaa7bba | 309 | error = gfs2_extent_map(&ip->i_inode, lblock, &new, |
fd88de56 | 310 | &dblock, &extlen); |
7276b3b0 | 311 | if (error || !dblock) |
e13940ba | 312 | goto fail; |
7276b3b0 SW |
313 | BUG_ON(extlen < 1); |
314 | if (!ra) | |
315 | extlen = 1; | |
316 | bh = gfs2_meta_ra(ip->i_gl, dblock, extlen); | |
b7d8ac3e | 317 | } else { |
7276b3b0 | 318 | error = gfs2_meta_read(ip->i_gl, dblock, DIO_WAIT, &bh); |
e13940ba SW |
319 | if (error) |
320 | goto fail; | |
7276b3b0 SW |
321 | } |
322 | error = gfs2_metatype_check(sdp, bh, GFS2_METATYPE_JD); | |
323 | if (error) { | |
324 | brelse(bh); | |
325 | goto fail; | |
326 | } | |
327 | dblock++; | |
328 | extlen--; | |
e13940ba SW |
329 | memcpy(buf, bh->b_data + o, amount); |
330 | brelse(bh); | |
899bb264 | 331 | buf += amount; |
e13940ba SW |
332 | copied += amount; |
333 | lblock++; | |
e13940ba SW |
334 | o = sizeof(struct gfs2_meta_header); |
335 | } | |
336 | ||
337 | return copied; | |
338 | fail: | |
339 | return (copied) ? copied : error; | |
340 | } | |
341 | ||
17d539f0 SW |
342 | /** |
343 | * gfs2_dir_get_hash_table - Get pointer to the dir hash table | |
344 | * @ip: The inode in question | |
345 | * | |
346 | * Returns: The hash table or an error | |
347 | */ | |
348 | ||
349 | static __be64 *gfs2_dir_get_hash_table(struct gfs2_inode *ip) | |
350 | { | |
351 | struct inode *inode = &ip->i_inode; | |
352 | int ret; | |
353 | u32 hsize; | |
354 | __be64 *hc; | |
355 | ||
356 | BUG_ON(!(ip->i_diskflags & GFS2_DIF_EXHASH)); | |
357 | ||
358 | hc = ip->i_hash_cache; | |
359 | if (hc) | |
360 | return hc; | |
361 | ||
362 | hsize = 1 << ip->i_depth; | |
363 | hsize *= sizeof(__be64); | |
364 | if (hsize != i_size_read(&ip->i_inode)) { | |
365 | gfs2_consist_inode(ip); | |
366 | return ERR_PTR(-EIO); | |
367 | } | |
368 | ||
369 | hc = kmalloc(hsize, GFP_NOFS); | |
370 | ret = -ENOMEM; | |
371 | if (hc == NULL) | |
372 | return ERR_PTR(-ENOMEM); | |
373 | ||
374 | ret = gfs2_dir_read_data(ip, (char *)hc, 0, hsize, 1); | |
375 | if (ret < 0) { | |
376 | kfree(hc); | |
377 | return ERR_PTR(ret); | |
378 | } | |
379 | ||
380 | spin_lock(&inode->i_lock); | |
381 | if (ip->i_hash_cache) | |
382 | kfree(hc); | |
383 | else | |
384 | ip->i_hash_cache = hc; | |
385 | spin_unlock(&inode->i_lock); | |
386 | ||
387 | return ip->i_hash_cache; | |
388 | } | |
389 | ||
390 | /** | |
391 | * gfs2_dir_hash_inval - Invalidate dir hash | |
392 | * @ip: The directory inode | |
393 | * | |
394 | * Must be called with an exclusive glock, or during glock invalidation. | |
395 | */ | |
396 | void gfs2_dir_hash_inval(struct gfs2_inode *ip) | |
397 | { | |
398 | __be64 *hc = ip->i_hash_cache; | |
399 | ip->i_hash_cache = NULL; | |
400 | kfree(hc); | |
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 | ||
d4e9c4c3 | 662 | gfs2_trans_add_bh(dip->i_gl, bh, 1); |
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); | |
705 | gfs2_trans_add_bh(ip->i_gl, bh, 1); | |
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; | |
71b86f56 | 836 | struct qstr name = { .name = "", .len = 0, .hash = 0 }; |
09010978 SW |
837 | |
838 | error = gfs2_alloc_block(ip, &bn, &n); | |
839 | if (error) | |
840 | return NULL; | |
841 | bh = gfs2_meta_new(ip->i_gl, bn); | |
c752666c SW |
842 | if (!bh) |
843 | return NULL; | |
09010978 | 844 | |
5731be53 | 845 | gfs2_trans_add_unrevoke(GFS2_SB(inode), bn, 1); |
c752666c SW |
846 | gfs2_trans_add_bh(ip->i_gl, bh, 1); |
847 | gfs2_metatype_set(bh, GFS2_METATYPE_LF, GFS2_FORMAT_LF); | |
848 | leaf = (struct gfs2_leaf *)bh->b_data; | |
849 | leaf->lf_depth = cpu_to_be16(depth); | |
2bdbc5d7 | 850 | leaf->lf_entries = 0; |
a2d7d021 | 851 | leaf->lf_dirent_format = cpu_to_be32(GFS2_FORMAT_DE); |
2bdbc5d7 | 852 | leaf->lf_next = 0; |
c752666c SW |
853 | memset(leaf->lf_reserved, 0, sizeof(leaf->lf_reserved)); |
854 | dent = (struct gfs2_dirent *)(leaf+1); | |
71b86f56 | 855 | gfs2_qstr2dirent(&name, bh->b_size - sizeof(struct gfs2_leaf), dent); |
c752666c SW |
856 | *pbh = bh; |
857 | return leaf; | |
b3b94faa DT |
858 | } |
859 | ||
860 | /** | |
861 | * dir_make_exhash - Convert a stuffed directory into an ExHash directory | |
862 | * @dip: The GFS2 inode | |
863 | * | |
864 | * Returns: 0 on success, error code otherwise | |
865 | */ | |
866 | ||
c752666c | 867 | static int dir_make_exhash(struct inode *inode) |
b3b94faa | 868 | { |
feaa7bba SW |
869 | struct gfs2_inode *dip = GFS2_I(inode); |
870 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
b3b94faa | 871 | struct gfs2_dirent *dent; |
c752666c | 872 | struct qstr args; |
b3b94faa DT |
873 | struct buffer_head *bh, *dibh; |
874 | struct gfs2_leaf *leaf; | |
875 | int y; | |
cd915493 | 876 | u32 x; |
b44b84d7 AV |
877 | __be64 *lp; |
878 | u64 bn; | |
b3b94faa DT |
879 | int error; |
880 | ||
881 | error = gfs2_meta_inode_buffer(dip, &dibh); | |
882 | if (error) | |
883 | return error; | |
884 | ||
b3b94faa DT |
885 | /* Turn over a new leaf */ |
886 | ||
c752666c SW |
887 | leaf = new_leaf(inode, &bh, 0); |
888 | if (!leaf) | |
889 | return -ENOSPC; | |
890 | bn = bh->b_blocknr; | |
b3b94faa | 891 | |
ad6203f2 SW |
892 | gfs2_assert(sdp, dip->i_entries < (1 << 16)); |
893 | leaf->lf_entries = cpu_to_be16(dip->i_entries); | |
b3b94faa DT |
894 | |
895 | /* Copy dirents */ | |
896 | ||
897 | gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_leaf), dibh, | |
898 | sizeof(struct gfs2_dinode)); | |
899 | ||
900 | /* Find last entry */ | |
901 | ||
902 | x = 0; | |
c752666c SW |
903 | args.len = bh->b_size - sizeof(struct gfs2_dinode) + |
904 | sizeof(struct gfs2_leaf); | |
905 | args.name = bh->b_data; | |
feaa7bba | 906 | dent = gfs2_dirent_scan(&dip->i_inode, bh->b_data, bh->b_size, |
71b86f56 | 907 | gfs2_dirent_last, &args, NULL); |
c752666c SW |
908 | if (!dent) { |
909 | brelse(bh); | |
910 | brelse(dibh); | |
911 | return -EIO; | |
912 | } | |
913 | if (IS_ERR(dent)) { | |
914 | brelse(bh); | |
915 | brelse(dibh); | |
916 | return PTR_ERR(dent); | |
b3b94faa | 917 | } |
b3b94faa DT |
918 | |
919 | /* Adjust the last dirent's record length | |
920 | (Remember that dent still points to the last entry.) */ | |
921 | ||
4dd651ad | 922 | dent->de_rec_len = cpu_to_be16(be16_to_cpu(dent->de_rec_len) + |
b3b94faa | 923 | sizeof(struct gfs2_dinode) - |
4dd651ad | 924 | sizeof(struct gfs2_leaf)); |
b3b94faa DT |
925 | |
926 | brelse(bh); | |
927 | ||
928 | /* We're done with the new leaf block, now setup the new | |
929 | hash table. */ | |
930 | ||
d4e9c4c3 | 931 | gfs2_trans_add_bh(dip->i_gl, dibh, 1); |
b3b94faa DT |
932 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); |
933 | ||
b44b84d7 | 934 | lp = (__be64 *)(dibh->b_data + sizeof(struct gfs2_dinode)); |
b3b94faa DT |
935 | |
936 | for (x = sdp->sd_hash_ptrs; x--; lp++) | |
937 | *lp = cpu_to_be64(bn); | |
938 | ||
a2e0f799 | 939 | i_size_write(inode, sdp->sd_sb.sb_bsize / 2); |
77658aad | 940 | gfs2_add_inode_blocks(&dip->i_inode, 1); |
383f01fb | 941 | dip->i_diskflags |= GFS2_DIF_EXHASH; |
b3b94faa DT |
942 | |
943 | for (x = sdp->sd_hash_ptrs, y = -1; x; x >>= 1, y++) ; | |
9a004508 | 944 | dip->i_depth = y; |
b3b94faa | 945 | |
539e5d6b | 946 | gfs2_dinode_out(dip, dibh->b_data); |
b3b94faa DT |
947 | |
948 | brelse(dibh); | |
949 | ||
950 | return 0; | |
951 | } | |
952 | ||
953 | /** | |
954 | * dir_split_leaf - Split a leaf block into two | |
955 | * @dip: The GFS2 inode | |
956 | * @index: | |
957 | * @leaf_no: | |
958 | * | |
959 | * Returns: 0 on success, error code on failure | |
960 | */ | |
961 | ||
c752666c | 962 | static int dir_split_leaf(struct inode *inode, const struct qstr *name) |
b3b94faa | 963 | { |
feaa7bba | 964 | struct gfs2_inode *dip = GFS2_I(inode); |
b3b94faa DT |
965 | struct buffer_head *nbh, *obh, *dibh; |
966 | struct gfs2_leaf *nleaf, *oleaf; | |
4da3c646 | 967 | struct gfs2_dirent *dent = NULL, *prev = NULL, *next = NULL, *new; |
cd915493 | 968 | u32 start, len, half_len, divider; |
b44b84d7 AV |
969 | u64 bn, leaf_no; |
970 | __be64 *lp; | |
cd915493 | 971 | u32 index; |
b3b94faa DT |
972 | int x, moved = 0; |
973 | int error; | |
974 | ||
9a004508 | 975 | index = name->hash >> (32 - dip->i_depth); |
c752666c SW |
976 | error = get_leaf_nr(dip, index, &leaf_no); |
977 | if (error) | |
978 | return error; | |
b3b94faa DT |
979 | |
980 | /* Get the old leaf block */ | |
b3b94faa DT |
981 | error = get_leaf(dip, leaf_no, &obh); |
982 | if (error) | |
e90deff5 | 983 | return error; |
b3b94faa | 984 | |
b3b94faa | 985 | oleaf = (struct gfs2_leaf *)obh->b_data; |
9a004508 | 986 | if (dip->i_depth == be16_to_cpu(oleaf->lf_depth)) { |
e90deff5 SW |
987 | brelse(obh); |
988 | return 1; /* can't split */ | |
989 | } | |
990 | ||
991 | gfs2_trans_add_bh(dip->i_gl, obh, 1); | |
b3b94faa | 992 | |
c752666c SW |
993 | nleaf = new_leaf(inode, &nbh, be16_to_cpu(oleaf->lf_depth) + 1); |
994 | if (!nleaf) { | |
995 | brelse(obh); | |
996 | return -ENOSPC; | |
997 | } | |
998 | bn = nbh->b_blocknr; | |
b3b94faa | 999 | |
c752666c | 1000 | /* Compute the start and len of leaf pointers in the hash table. */ |
9a004508 | 1001 | len = 1 << (dip->i_depth - be16_to_cpu(oleaf->lf_depth)); |
b3b94faa DT |
1002 | half_len = len >> 1; |
1003 | if (!half_len) { | |
9a004508 | 1004 | printk(KERN_WARNING "i_depth %u lf_depth %u index %u\n", dip->i_depth, be16_to_cpu(oleaf->lf_depth), index); |
b3b94faa DT |
1005 | gfs2_consist_inode(dip); |
1006 | error = -EIO; | |
1007 | goto fail_brelse; | |
1008 | } | |
1009 | ||
1010 | start = (index & ~(len - 1)); | |
1011 | ||
1012 | /* Change the pointers. | |
1013 | Don't bother distinguishing stuffed from non-stuffed. | |
1014 | This code is complicated enough already. */ | |
4244b52e DR |
1015 | lp = kmalloc(half_len * sizeof(__be64), GFP_NOFS); |
1016 | if (!lp) { | |
1017 | error = -ENOMEM; | |
1018 | goto fail_brelse; | |
1019 | } | |
1020 | ||
b3b94faa | 1021 | /* Change the pointers */ |
b3b94faa DT |
1022 | for (x = 0; x < half_len; x++) |
1023 | lp[x] = cpu_to_be64(bn); | |
1024 | ||
17d539f0 SW |
1025 | gfs2_dir_hash_inval(dip); |
1026 | ||
cd915493 SW |
1027 | error = gfs2_dir_write_data(dip, (char *)lp, start * sizeof(u64), |
1028 | half_len * sizeof(u64)); | |
1029 | if (error != half_len * sizeof(u64)) { | |
b3b94faa DT |
1030 | if (error >= 0) |
1031 | error = -EIO; | |
1032 | goto fail_lpfree; | |
1033 | } | |
1034 | ||
1035 | kfree(lp); | |
1036 | ||
1037 | /* Compute the divider */ | |
9a004508 | 1038 | divider = (start + half_len) << (32 - dip->i_depth); |
b3b94faa DT |
1039 | |
1040 | /* Copy the entries */ | |
1579343a | 1041 | dent = (struct gfs2_dirent *)(obh->b_data + sizeof(struct gfs2_leaf)); |
b3b94faa DT |
1042 | |
1043 | do { | |
1044 | next = dent; | |
1045 | if (dirent_next(dip, obh, &next)) | |
1046 | next = NULL; | |
1047 | ||
5e7d65cd | 1048 | if (!gfs2_dirent_sentinel(dent) && |
b3b94faa | 1049 | be32_to_cpu(dent->de_hash) < divider) { |
c752666c SW |
1050 | struct qstr str; |
1051 | str.name = (char*)(dent+1); | |
1052 | str.len = be16_to_cpu(dent->de_name_len); | |
1053 | str.hash = be32_to_cpu(dent->de_hash); | |
71b86f56 | 1054 | new = gfs2_dirent_alloc(inode, nbh, &str); |
c752666c SW |
1055 | if (IS_ERR(new)) { |
1056 | error = PTR_ERR(new); | |
1057 | break; | |
1058 | } | |
b3b94faa DT |
1059 | |
1060 | new->de_inum = dent->de_inum; /* No endian worries */ | |
b3b94faa | 1061 | new->de_type = dent->de_type; /* No endian worries */ |
bb16b342 | 1062 | be16_add_cpu(&nleaf->lf_entries, 1); |
b3b94faa DT |
1063 | |
1064 | dirent_del(dip, obh, prev, dent); | |
1065 | ||
1066 | if (!oleaf->lf_entries) | |
1067 | gfs2_consist_inode(dip); | |
bb16b342 | 1068 | be16_add_cpu(&oleaf->lf_entries, -1); |
b3b94faa DT |
1069 | |
1070 | if (!prev) | |
1071 | prev = dent; | |
1072 | ||
1073 | moved = 1; | |
c752666c | 1074 | } else { |
b3b94faa | 1075 | prev = dent; |
c752666c | 1076 | } |
b3b94faa | 1077 | dent = next; |
c752666c | 1078 | } while (dent); |
b3b94faa | 1079 | |
c752666c | 1080 | oleaf->lf_depth = nleaf->lf_depth; |
b3b94faa DT |
1081 | |
1082 | error = gfs2_meta_inode_buffer(dip, &dibh); | |
feaa7bba | 1083 | if (!gfs2_assert_withdraw(GFS2_SB(&dip->i_inode), !error)) { |
382e6e25 | 1084 | gfs2_trans_add_bh(dip->i_gl, dibh, 1); |
77658aad | 1085 | gfs2_add_inode_blocks(&dip->i_inode, 1); |
539e5d6b | 1086 | gfs2_dinode_out(dip, dibh->b_data); |
b3b94faa DT |
1087 | brelse(dibh); |
1088 | } | |
1089 | ||
1090 | brelse(obh); | |
1091 | brelse(nbh); | |
1092 | ||
1093 | return error; | |
1094 | ||
e90deff5 | 1095 | fail_lpfree: |
b3b94faa DT |
1096 | kfree(lp); |
1097 | ||
e90deff5 | 1098 | fail_brelse: |
b3b94faa | 1099 | brelse(obh); |
b3b94faa DT |
1100 | brelse(nbh); |
1101 | return error; | |
1102 | } | |
1103 | ||
1104 | /** | |
1105 | * dir_double_exhash - Double size of ExHash table | |
1106 | * @dip: The GFS2 dinode | |
1107 | * | |
1108 | * Returns: 0 on success, error code on failure | |
1109 | */ | |
1110 | ||
1111 | static int dir_double_exhash(struct gfs2_inode *dip) | |
1112 | { | |
b3b94faa | 1113 | struct buffer_head *dibh; |
cd915493 | 1114 | u32 hsize; |
17d539f0 SW |
1115 | u32 hsize_bytes; |
1116 | __be64 *hc; | |
1117 | __be64 *hc2, *h; | |
b3b94faa DT |
1118 | int x; |
1119 | int error = 0; | |
1120 | ||
9a004508 | 1121 | hsize = 1 << dip->i_depth; |
17d539f0 | 1122 | hsize_bytes = hsize * sizeof(__be64); |
b3b94faa | 1123 | |
17d539f0 SW |
1124 | hc = gfs2_dir_get_hash_table(dip); |
1125 | if (IS_ERR(hc)) | |
1126 | return PTR_ERR(hc); | |
b3b94faa | 1127 | |
17d539f0 SW |
1128 | h = hc2 = kmalloc(hsize_bytes * 2, GFP_NOFS); |
1129 | if (!hc2) | |
4244b52e | 1130 | return -ENOMEM; |
b3b94faa | 1131 | |
17d539f0 SW |
1132 | error = gfs2_meta_inode_buffer(dip, &dibh); |
1133 | if (error) | |
1134 | goto out_kfree; | |
b3b94faa | 1135 | |
17d539f0 SW |
1136 | for (x = 0; x < hsize; x++) { |
1137 | *h++ = *hc; | |
1138 | *h++ = *hc; | |
1139 | hc++; | |
b3b94faa DT |
1140 | } |
1141 | ||
17d539f0 SW |
1142 | error = gfs2_dir_write_data(dip, (char *)hc2, 0, hsize_bytes * 2); |
1143 | if (error != (hsize_bytes * 2)) | |
1144 | goto fail; | |
b3b94faa | 1145 | |
17d539f0 SW |
1146 | gfs2_dir_hash_inval(dip); |
1147 | dip->i_hash_cache = hc2; | |
1148 | dip->i_depth++; | |
1149 | gfs2_dinode_out(dip, dibh->b_data); | |
1150 | brelse(dibh); | |
1151 | return 0; | |
b3b94faa | 1152 | |
a91ea69f | 1153 | fail: |
17d539f0 SW |
1154 | /* Replace original hash table & size */ |
1155 | gfs2_dir_write_data(dip, (char *)hc, 0, hsize_bytes); | |
1156 | i_size_write(&dip->i_inode, hsize_bytes); | |
1157 | gfs2_dinode_out(dip, dibh->b_data); | |
1158 | brelse(dibh); | |
1159 | out_kfree: | |
1160 | kfree(hc2); | |
b3b94faa DT |
1161 | return error; |
1162 | } | |
1163 | ||
1164 | /** | |
1165 | * compare_dents - compare directory entries by hash value | |
1166 | * @a: first dent | |
1167 | * @b: second dent | |
1168 | * | |
1169 | * When comparing the hash entries of @a to @b: | |
1170 | * gt: returns 1 | |
1171 | * lt: returns -1 | |
1172 | * eq: returns 0 | |
1173 | */ | |
1174 | ||
1175 | static int compare_dents(const void *a, const void *b) | |
1176 | { | |
2bdbc5d7 | 1177 | const struct gfs2_dirent *dent_a, *dent_b; |
cd915493 | 1178 | u32 hash_a, hash_b; |
b3b94faa DT |
1179 | int ret = 0; |
1180 | ||
2bdbc5d7 | 1181 | dent_a = *(const struct gfs2_dirent **)a; |
c752666c | 1182 | hash_a = be32_to_cpu(dent_a->de_hash); |
b3b94faa | 1183 | |
2bdbc5d7 | 1184 | dent_b = *(const struct gfs2_dirent **)b; |
c752666c | 1185 | hash_b = be32_to_cpu(dent_b->de_hash); |
b3b94faa DT |
1186 | |
1187 | if (hash_a > hash_b) | |
1188 | ret = 1; | |
1189 | else if (hash_a < hash_b) | |
1190 | ret = -1; | |
1191 | else { | |
4dd651ad SW |
1192 | unsigned int len_a = be16_to_cpu(dent_a->de_name_len); |
1193 | unsigned int len_b = be16_to_cpu(dent_b->de_name_len); | |
b3b94faa DT |
1194 | |
1195 | if (len_a > len_b) | |
1196 | ret = 1; | |
1197 | else if (len_a < len_b) | |
1198 | ret = -1; | |
1199 | else | |
2bdbc5d7 | 1200 | ret = memcmp(dent_a + 1, dent_b + 1, len_a); |
b3b94faa DT |
1201 | } |
1202 | ||
1203 | return ret; | |
1204 | } | |
1205 | ||
1206 | /** | |
1207 | * do_filldir_main - read out directory entries | |
1208 | * @dip: The GFS2 inode | |
1209 | * @offset: The offset in the file to read from | |
1210 | * @opaque: opaque data to pass to filldir | |
1211 | * @filldir: The function to pass entries to | |
1212 | * @darr: an array of struct gfs2_dirent pointers to read | |
1213 | * @entries: the number of entries in darr | |
1214 | * @copied: pointer to int that's non-zero if a entry has been copied out | |
1215 | * | |
1216 | * Jump through some hoops to make sure that if there are hash collsions, | |
1217 | * they are read out at the beginning of a buffer. We want to minimize | |
1218 | * the possibility that they will fall into different readdir buffers or | |
1219 | * that someone will want to seek to that location. | |
1220 | * | |
1221 | * Returns: errno, >0 on exception from filldir | |
1222 | */ | |
1223 | ||
cd915493 | 1224 | static int do_filldir_main(struct gfs2_inode *dip, u64 *offset, |
3699e3a4 | 1225 | void *opaque, filldir_t filldir, |
cd915493 | 1226 | const struct gfs2_dirent **darr, u32 entries, |
b3b94faa DT |
1227 | int *copied) |
1228 | { | |
71b86f56 | 1229 | const struct gfs2_dirent *dent, *dent_next; |
cd915493 | 1230 | u64 off, off_next; |
b3b94faa DT |
1231 | unsigned int x, y; |
1232 | int run = 0; | |
1233 | int error = 0; | |
1234 | ||
1235 | sort(darr, entries, sizeof(struct gfs2_dirent *), compare_dents, NULL); | |
1236 | ||
1237 | dent_next = darr[0]; | |
1238 | off_next = be32_to_cpu(dent_next->de_hash); | |
1239 | off_next = gfs2_disk_hash2offset(off_next); | |
1240 | ||
1241 | for (x = 0, y = 1; x < entries; x++, y++) { | |
1242 | dent = dent_next; | |
1243 | off = off_next; | |
1244 | ||
1245 | if (y < entries) { | |
1246 | dent_next = darr[y]; | |
1247 | off_next = be32_to_cpu(dent_next->de_hash); | |
1248 | off_next = gfs2_disk_hash2offset(off_next); | |
1249 | ||
1250 | if (off < *offset) | |
1251 | continue; | |
1252 | *offset = off; | |
1253 | ||
1254 | if (off_next == off) { | |
1255 | if (*copied && !run) | |
1256 | return 1; | |
1257 | run = 1; | |
1258 | } else | |
1259 | run = 0; | |
1260 | } else { | |
1261 | if (off < *offset) | |
1262 | continue; | |
1263 | *offset = off; | |
1264 | } | |
1265 | ||
2bdbc5d7 | 1266 | error = filldir(opaque, (const char *)(dent + 1), |
4dd651ad | 1267 | be16_to_cpu(dent->de_name_len), |
3699e3a4 | 1268 | off, be64_to_cpu(dent->de_inum.no_addr), |
4dd651ad | 1269 | be16_to_cpu(dent->de_type)); |
b3b94faa DT |
1270 | if (error) |
1271 | return 1; | |
1272 | ||
1273 | *copied = 1; | |
1274 | } | |
1275 | ||
1276 | /* Increment the *offset by one, so the next time we come into the | |
1277 | do_filldir fxn, we get the next entry instead of the last one in the | |
1278 | current leaf */ | |
1279 | ||
1280 | (*offset)++; | |
1281 | ||
1282 | return 0; | |
1283 | } | |
1284 | ||
d2a97a4e SW |
1285 | static void *gfs2_alloc_sort_buffer(unsigned size) |
1286 | { | |
1287 | void *ptr = NULL; | |
1288 | ||
1289 | if (size < KMALLOC_MAX_SIZE) | |
1290 | ptr = kmalloc(size, GFP_NOFS | __GFP_NOWARN); | |
1291 | if (!ptr) | |
1292 | ptr = __vmalloc(size, GFP_NOFS, PAGE_KERNEL); | |
1293 | return ptr; | |
1294 | } | |
1295 | ||
1296 | static void gfs2_free_sort_buffer(void *ptr) | |
1297 | { | |
1298 | if (is_vmalloc_addr(ptr)) | |
1299 | vfree(ptr); | |
1300 | else | |
1301 | kfree(ptr); | |
1302 | } | |
1303 | ||
71b86f56 | 1304 | static int gfs2_dir_read_leaf(struct inode *inode, u64 *offset, void *opaque, |
3699e3a4 SW |
1305 | filldir_t filldir, int *copied, unsigned *depth, |
1306 | u64 leaf_no) | |
b3b94faa | 1307 | { |
feaa7bba | 1308 | struct gfs2_inode *ip = GFS2_I(inode); |
bdd19a22 | 1309 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
71b86f56 SW |
1310 | struct buffer_head *bh; |
1311 | struct gfs2_leaf *lf; | |
bdd19a22 | 1312 | unsigned entries = 0, entries2 = 0; |
71b86f56 SW |
1313 | unsigned leaves = 0; |
1314 | const struct gfs2_dirent **darr, *dent; | |
1315 | struct dirent_gather g; | |
1316 | struct buffer_head **larr; | |
1317 | int leaf = 0; | |
1318 | int error, i; | |
1319 | u64 lfn = leaf_no; | |
b3b94faa | 1320 | |
b3b94faa | 1321 | do { |
71b86f56 | 1322 | error = get_leaf(ip, lfn, &bh); |
b3b94faa | 1323 | if (error) |
71b86f56 SW |
1324 | goto out; |
1325 | lf = (struct gfs2_leaf *)bh->b_data; | |
1326 | if (leaves == 0) | |
1327 | *depth = be16_to_cpu(lf->lf_depth); | |
1328 | entries += be16_to_cpu(lf->lf_entries); | |
1329 | leaves++; | |
1330 | lfn = be64_to_cpu(lf->lf_next); | |
1331 | brelse(bh); | |
1332 | } while(lfn); | |
b3b94faa DT |
1333 | |
1334 | if (!entries) | |
1335 | return 0; | |
1336 | ||
71b86f56 | 1337 | error = -ENOMEM; |
bdd19a22 SW |
1338 | /* |
1339 | * The extra 99 entries are not normally used, but are a buffer | |
1340 | * zone in case the number of entries in the leaf is corrupt. | |
1341 | * 99 is the maximum number of entries that can fit in a single | |
1342 | * leaf block. | |
1343 | */ | |
d2a97a4e | 1344 | larr = gfs2_alloc_sort_buffer((leaves + entries + 99) * sizeof(void *)); |
71b86f56 SW |
1345 | if (!larr) |
1346 | goto out; | |
1347 | darr = (const struct gfs2_dirent **)(larr + leaves); | |
1348 | g.pdent = darr; | |
1349 | g.offset = 0; | |
1350 | lfn = leaf_no; | |
b3b94faa | 1351 | |
71b86f56 SW |
1352 | do { |
1353 | error = get_leaf(ip, lfn, &bh); | |
b3b94faa | 1354 | if (error) |
d2a97a4e | 1355 | goto out_free; |
71b86f56 SW |
1356 | lf = (struct gfs2_leaf *)bh->b_data; |
1357 | lfn = be64_to_cpu(lf->lf_next); | |
1358 | if (lf->lf_entries) { | |
bdd19a22 | 1359 | entries2 += be16_to_cpu(lf->lf_entries); |
71b86f56 SW |
1360 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, |
1361 | gfs2_dirent_gather, NULL, &g); | |
1362 | error = PTR_ERR(dent); | |
bdd19a22 | 1363 | if (IS_ERR(dent)) |
d2a97a4e | 1364 | goto out_free; |
bdd19a22 | 1365 | if (entries2 != g.offset) { |
f391a4ea AM |
1366 | fs_warn(sdp, "Number of entries corrupt in dir " |
1367 | "leaf %llu, entries2 (%u) != " | |
1368 | "g.offset (%u)\n", | |
1369 | (unsigned long long)bh->b_blocknr, | |
1370 | entries2, g.offset); | |
bdd19a22 SW |
1371 | |
1372 | error = -EIO; | |
d2a97a4e | 1373 | goto out_free; |
71b86f56 SW |
1374 | } |
1375 | error = 0; | |
1376 | larr[leaf++] = bh; | |
b3b94faa | 1377 | } else { |
71b86f56 | 1378 | brelse(bh); |
b3b94faa | 1379 | } |
71b86f56 | 1380 | } while(lfn); |
b3b94faa | 1381 | |
bdd19a22 | 1382 | BUG_ON(entries2 != entries); |
71b86f56 | 1383 | error = do_filldir_main(ip, offset, opaque, filldir, darr, |
b3b94faa | 1384 | entries, copied); |
d2a97a4e | 1385 | out_free: |
71b86f56 SW |
1386 | for(i = 0; i < leaf; i++) |
1387 | brelse(larr[i]); | |
d2a97a4e | 1388 | gfs2_free_sort_buffer(larr); |
71b86f56 | 1389 | out: |
b3b94faa DT |
1390 | return error; |
1391 | } | |
1392 | ||
17d539f0 | 1393 | |
b3b94faa | 1394 | /** |
c752666c SW |
1395 | * dir_e_read - Reads the entries from a directory into a filldir buffer |
1396 | * @dip: dinode pointer | |
1397 | * @offset: the hash of the last entry read shifted to the right once | |
1398 | * @opaque: buffer for the filldir function to fill | |
1399 | * @filldir: points to the filldir function to use | |
b3b94faa | 1400 | * |
c752666c | 1401 | * Returns: errno |
b3b94faa DT |
1402 | */ |
1403 | ||
cd915493 | 1404 | static int dir_e_read(struct inode *inode, u64 *offset, void *opaque, |
3699e3a4 | 1405 | filldir_t filldir) |
b3b94faa | 1406 | { |
feaa7bba | 1407 | struct gfs2_inode *dip = GFS2_I(inode); |
cd915493 | 1408 | u32 hsize, len = 0; |
cd915493 | 1409 | u32 hash, index; |
b44b84d7 | 1410 | __be64 *lp; |
c752666c SW |
1411 | int copied = 0; |
1412 | int error = 0; | |
4da3c646 | 1413 | unsigned depth = 0; |
b3b94faa | 1414 | |
9a004508 | 1415 | hsize = 1 << dip->i_depth; |
b3b94faa | 1416 | hash = gfs2_dir_offset2hash(*offset); |
9a004508 | 1417 | index = hash >> (32 - dip->i_depth); |
b3b94faa | 1418 | |
17d539f0 SW |
1419 | lp = gfs2_dir_get_hash_table(dip); |
1420 | if (IS_ERR(lp)) | |
1421 | return PTR_ERR(lp); | |
b3b94faa DT |
1422 | |
1423 | while (index < hsize) { | |
71b86f56 SW |
1424 | error = gfs2_dir_read_leaf(inode, offset, opaque, filldir, |
1425 | &copied, &depth, | |
17d539f0 | 1426 | be64_to_cpu(lp[index])); |
b3b94faa | 1427 | if (error) |
71b86f56 | 1428 | break; |
b3b94faa | 1429 | |
9a004508 | 1430 | len = 1 << (dip->i_depth - depth); |
b3b94faa DT |
1431 | index = (index & ~(len - 1)) + len; |
1432 | } | |
1433 | ||
71b86f56 SW |
1434 | if (error > 0) |
1435 | error = 0; | |
b3b94faa DT |
1436 | return error; |
1437 | } | |
1438 | ||
cd915493 | 1439 | int gfs2_dir_read(struct inode *inode, u64 *offset, void *opaque, |
3699e3a4 | 1440 | filldir_t filldir) |
b3b94faa | 1441 | { |
feaa7bba | 1442 | struct gfs2_inode *dip = GFS2_I(inode); |
bdd19a22 | 1443 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
71b86f56 SW |
1444 | struct dirent_gather g; |
1445 | const struct gfs2_dirent **darr, *dent; | |
b3b94faa DT |
1446 | struct buffer_head *dibh; |
1447 | int copied = 0; | |
1448 | int error; | |
1449 | ||
ad6203f2 | 1450 | if (!dip->i_entries) |
71b86f56 SW |
1451 | return 0; |
1452 | ||
383f01fb | 1453 | if (dip->i_diskflags & GFS2_DIF_EXHASH) |
71b86f56 SW |
1454 | return dir_e_read(inode, offset, opaque, filldir); |
1455 | ||
b3b94faa DT |
1456 | if (!gfs2_is_stuffed(dip)) { |
1457 | gfs2_consist_inode(dip); | |
1458 | return -EIO; | |
1459 | } | |
1460 | ||
b3b94faa DT |
1461 | error = gfs2_meta_inode_buffer(dip, &dibh); |
1462 | if (error) | |
1463 | return error; | |
1464 | ||
71b86f56 | 1465 | error = -ENOMEM; |
bdd19a22 | 1466 | /* 96 is max number of dirents which can be stuffed into an inode */ |
16c5f06f | 1467 | darr = kmalloc(96 * sizeof(struct gfs2_dirent *), GFP_NOFS); |
71b86f56 SW |
1468 | if (darr) { |
1469 | g.pdent = darr; | |
1470 | g.offset = 0; | |
1471 | dent = gfs2_dirent_scan(inode, dibh->b_data, dibh->b_size, | |
1472 | gfs2_dirent_gather, NULL, &g); | |
1473 | if (IS_ERR(dent)) { | |
1474 | error = PTR_ERR(dent); | |
1475 | goto out; | |
1476 | } | |
ad6203f2 | 1477 | if (dip->i_entries != g.offset) { |
bdd19a22 | 1478 | fs_warn(sdp, "Number of entries corrupt in dir %llu, " |
ad6203f2 | 1479 | "ip->i_entries (%u) != g.offset (%u)\n", |
dbb7cae2 | 1480 | (unsigned long long)dip->i_no_addr, |
ad6203f2 | 1481 | dip->i_entries, |
bdd19a22 SW |
1482 | g.offset); |
1483 | error = -EIO; | |
1484 | goto out; | |
1485 | } | |
71b86f56 | 1486 | error = do_filldir_main(dip, offset, opaque, filldir, darr, |
ad6203f2 | 1487 | dip->i_entries, &copied); |
71b86f56 SW |
1488 | out: |
1489 | kfree(darr); | |
1490 | } | |
1491 | ||
b3b94faa DT |
1492 | if (error > 0) |
1493 | error = 0; | |
1494 | ||
1495 | brelse(dibh); | |
1496 | ||
1497 | return error; | |
1498 | } | |
1499 | ||
b3b94faa DT |
1500 | /** |
1501 | * gfs2_dir_search - Search a directory | |
1502 | * @dip: The GFS2 inode | |
1503 | * @filename: | |
1504 | * @inode: | |
1505 | * | |
1506 | * This routine searches a directory for a file or another directory. | |
1507 | * Assumes a glock is held on dip. | |
1508 | * | |
1509 | * Returns: errno | |
1510 | */ | |
1511 | ||
dbb7cae2 | 1512 | struct inode *gfs2_dir_search(struct inode *dir, const struct qstr *name) |
b3b94faa | 1513 | { |
c752666c SW |
1514 | struct buffer_head *bh; |
1515 | struct gfs2_dirent *dent; | |
dbb7cae2 SW |
1516 | struct inode *inode; |
1517 | ||
1518 | dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh); | |
1519 | if (dent) { | |
1520 | if (IS_ERR(dent)) | |
e231c2ee | 1521 | return ERR_CAST(dent); |
bb9bcf06 WC |
1522 | inode = gfs2_inode_lookup(dir->i_sb, |
1523 | be16_to_cpu(dent->de_type), | |
1524 | be64_to_cpu(dent->de_inum.no_addr), | |
44ad37d6 | 1525 | be64_to_cpu(dent->de_inum.no_formal_ino), 0); |
dbb7cae2 SW |
1526 | brelse(bh); |
1527 | return inode; | |
1528 | } | |
1529 | return ERR_PTR(-ENOENT); | |
1530 | } | |
1531 | ||
1532 | int gfs2_dir_check(struct inode *dir, const struct qstr *name, | |
1533 | const struct gfs2_inode *ip) | |
1534 | { | |
1535 | struct buffer_head *bh; | |
1536 | struct gfs2_dirent *dent; | |
1537 | int ret = -ENOENT; | |
c752666c SW |
1538 | |
1539 | dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh); | |
1540 | if (dent) { | |
1541 | if (IS_ERR(dent)) | |
1542 | return PTR_ERR(dent); | |
dbb7cae2 SW |
1543 | if (ip) { |
1544 | if (be64_to_cpu(dent->de_inum.no_addr) != ip->i_no_addr) | |
1545 | goto out; | |
1546 | if (be64_to_cpu(dent->de_inum.no_formal_ino) != | |
1547 | ip->i_no_formal_ino) | |
1548 | goto out; | |
1549 | if (unlikely(IF2DT(ip->i_inode.i_mode) != | |
1550 | be16_to_cpu(dent->de_type))) { | |
1551 | gfs2_consist_inode(GFS2_I(dir)); | |
1552 | ret = -EIO; | |
1553 | goto out; | |
1554 | } | |
1555 | } | |
1556 | ret = 0; | |
1557 | out: | |
c752666c | 1558 | brelse(bh); |
c752666c | 1559 | } |
dbb7cae2 | 1560 | return ret; |
c752666c SW |
1561 | } |
1562 | ||
1563 | static int dir_new_leaf(struct inode *inode, const struct qstr *name) | |
1564 | { | |
1565 | struct buffer_head *bh, *obh; | |
feaa7bba | 1566 | struct gfs2_inode *ip = GFS2_I(inode); |
c752666c | 1567 | struct gfs2_leaf *leaf, *oleaf; |
b3b94faa | 1568 | int error; |
c752666c SW |
1569 | u32 index; |
1570 | u64 bn; | |
b3b94faa | 1571 | |
9a004508 | 1572 | index = name->hash >> (32 - ip->i_depth); |
c752666c SW |
1573 | error = get_first_leaf(ip, index, &obh); |
1574 | if (error) | |
1575 | return error; | |
1576 | do { | |
1577 | oleaf = (struct gfs2_leaf *)obh->b_data; | |
1578 | bn = be64_to_cpu(oleaf->lf_next); | |
1579 | if (!bn) | |
1580 | break; | |
1581 | brelse(obh); | |
1582 | error = get_leaf(ip, bn, &obh); | |
1583 | if (error) | |
1584 | return error; | |
1585 | } while(1); | |
b3b94faa | 1586 | |
c752666c SW |
1587 | gfs2_trans_add_bh(ip->i_gl, obh, 1); |
1588 | ||
1589 | leaf = new_leaf(inode, &bh, be16_to_cpu(oleaf->lf_depth)); | |
1590 | if (!leaf) { | |
1591 | brelse(obh); | |
1592 | return -ENOSPC; | |
1593 | } | |
4d8012b6 | 1594 | oleaf->lf_next = cpu_to_be64(bh->b_blocknr); |
c752666c SW |
1595 | brelse(bh); |
1596 | brelse(obh); | |
1597 | ||
1598 | error = gfs2_meta_inode_buffer(ip, &bh); | |
1599 | if (error) | |
1600 | return error; | |
1601 | gfs2_trans_add_bh(ip->i_gl, bh, 1); | |
77658aad | 1602 | gfs2_add_inode_blocks(&ip->i_inode, 1); |
539e5d6b | 1603 | gfs2_dinode_out(ip, bh->b_data); |
c752666c SW |
1604 | brelse(bh); |
1605 | return 0; | |
b3b94faa DT |
1606 | } |
1607 | ||
1608 | /** | |
1609 | * gfs2_dir_add - Add new filename into directory | |
1610 | * @dip: The GFS2 inode | |
1611 | * @filename: The new name | |
1612 | * @inode: The inode number of the entry | |
1613 | * @type: The type of the entry | |
1614 | * | |
1615 | * Returns: 0 on success, error code on failure | |
1616 | */ | |
1617 | ||
c752666c | 1618 | int gfs2_dir_add(struct inode *inode, const struct qstr *name, |
3d6ecb7d | 1619 | const struct gfs2_inode *nip) |
b3b94faa | 1620 | { |
feaa7bba | 1621 | struct gfs2_inode *ip = GFS2_I(inode); |
c752666c SW |
1622 | struct buffer_head *bh; |
1623 | struct gfs2_dirent *dent; | |
1624 | struct gfs2_leaf *leaf; | |
b3b94faa DT |
1625 | int error; |
1626 | ||
c752666c SW |
1627 | while(1) { |
1628 | dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, | |
1629 | &bh); | |
1630 | if (dent) { | |
1631 | if (IS_ERR(dent)) | |
1632 | return PTR_ERR(dent); | |
1633 | dent = gfs2_init_dirent(inode, dent, name, bh); | |
dbb7cae2 | 1634 | gfs2_inum_out(nip, dent); |
3d6ecb7d | 1635 | dent->de_type = cpu_to_be16(IF2DT(nip->i_inode.i_mode)); |
383f01fb | 1636 | if (ip->i_diskflags & GFS2_DIF_EXHASH) { |
c752666c | 1637 | leaf = (struct gfs2_leaf *)bh->b_data; |
bb16b342 | 1638 | be16_add_cpu(&leaf->lf_entries, 1); |
c752666c SW |
1639 | } |
1640 | brelse(bh); | |
1641 | error = gfs2_meta_inode_buffer(ip, &bh); | |
1642 | if (error) | |
1643 | break; | |
1644 | gfs2_trans_add_bh(ip->i_gl, bh, 1); | |
ad6203f2 | 1645 | ip->i_entries++; |
4bd91ba1 | 1646 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME; |
3d6ecb7d SW |
1647 | if (S_ISDIR(nip->i_inode.i_mode)) |
1648 | inc_nlink(&ip->i_inode); | |
539e5d6b | 1649 | gfs2_dinode_out(ip, bh->b_data); |
c752666c SW |
1650 | brelse(bh); |
1651 | error = 0; | |
1652 | break; | |
1653 | } | |
383f01fb | 1654 | if (!(ip->i_diskflags & GFS2_DIF_EXHASH)) { |
c752666c SW |
1655 | error = dir_make_exhash(inode); |
1656 | if (error) | |
1657 | break; | |
1658 | continue; | |
1659 | } | |
1660 | error = dir_split_leaf(inode, name); | |
1661 | if (error == 0) | |
1662 | continue; | |
e90deff5 | 1663 | if (error < 0) |
c752666c | 1664 | break; |
9a004508 | 1665 | if (ip->i_depth < GFS2_DIR_MAX_DEPTH) { |
c752666c SW |
1666 | error = dir_double_exhash(ip); |
1667 | if (error) | |
1668 | break; | |
1669 | error = dir_split_leaf(inode, name); | |
e90deff5 | 1670 | if (error < 0) |
c752666c | 1671 | break; |
e90deff5 SW |
1672 | if (error == 0) |
1673 | continue; | |
c752666c SW |
1674 | } |
1675 | error = dir_new_leaf(inode, name); | |
1676 | if (!error) | |
1677 | continue; | |
1678 | error = -ENOSPC; | |
1679 | break; | |
1680 | } | |
b3b94faa DT |
1681 | return error; |
1682 | } | |
1683 | ||
c752666c | 1684 | |
b3b94faa DT |
1685 | /** |
1686 | * gfs2_dir_del - Delete a directory entry | |
1687 | * @dip: The GFS2 inode | |
1688 | * @filename: The filename | |
1689 | * | |
1690 | * Returns: 0 on success, error code on failure | |
1691 | */ | |
1692 | ||
855d23ce | 1693 | int gfs2_dir_del(struct gfs2_inode *dip, const struct dentry *dentry) |
b3b94faa | 1694 | { |
855d23ce | 1695 | const struct qstr *name = &dentry->d_name; |
c752666c SW |
1696 | struct gfs2_dirent *dent, *prev = NULL; |
1697 | struct buffer_head *bh; | |
b3b94faa DT |
1698 | int error; |
1699 | ||
c752666c SW |
1700 | /* Returns _either_ the entry (if its first in block) or the |
1701 | previous entry otherwise */ | |
feaa7bba | 1702 | dent = gfs2_dirent_search(&dip->i_inode, name, gfs2_dirent_prev, &bh); |
c752666c SW |
1703 | if (!dent) { |
1704 | gfs2_consist_inode(dip); | |
1705 | return -EIO; | |
1706 | } | |
1707 | if (IS_ERR(dent)) { | |
1708 | gfs2_consist_inode(dip); | |
1709 | return PTR_ERR(dent); | |
1710 | } | |
1711 | /* If not first in block, adjust pointers accordingly */ | |
71b86f56 | 1712 | if (gfs2_dirent_find(dent, name, NULL) == 0) { |
c752666c SW |
1713 | prev = dent; |
1714 | dent = (struct gfs2_dirent *)((char *)dent + be16_to_cpu(prev->de_rec_len)); | |
1715 | } | |
1716 | ||
1717 | dirent_del(dip, bh, prev, dent); | |
383f01fb | 1718 | if (dip->i_diskflags & GFS2_DIF_EXHASH) { |
c752666c SW |
1719 | struct gfs2_leaf *leaf = (struct gfs2_leaf *)bh->b_data; |
1720 | u16 entries = be16_to_cpu(leaf->lf_entries); | |
1721 | if (!entries) | |
1722 | gfs2_consist_inode(dip); | |
1723 | leaf->lf_entries = cpu_to_be16(--entries); | |
c752666c | 1724 | } |
ed386507 | 1725 | brelse(bh); |
c752666c SW |
1726 | |
1727 | error = gfs2_meta_inode_buffer(dip, &bh); | |
1728 | if (error) | |
1729 | return error; | |
1730 | ||
ad6203f2 | 1731 | if (!dip->i_entries) |
c752666c SW |
1732 | gfs2_consist_inode(dip); |
1733 | gfs2_trans_add_bh(dip->i_gl, bh, 1); | |
ad6203f2 | 1734 | dip->i_entries--; |
4bd91ba1 | 1735 | dip->i_inode.i_mtime = dip->i_inode.i_ctime = CURRENT_TIME; |
855d23ce SW |
1736 | if (S_ISDIR(dentry->d_inode->i_mode)) |
1737 | drop_nlink(&dip->i_inode); | |
539e5d6b | 1738 | gfs2_dinode_out(dip, bh->b_data); |
c752666c | 1739 | brelse(bh); |
feaa7bba | 1740 | mark_inode_dirty(&dip->i_inode); |
b3b94faa DT |
1741 | |
1742 | return error; | |
1743 | } | |
1744 | ||
b3b94faa DT |
1745 | /** |
1746 | * gfs2_dir_mvino - Change inode number of directory entry | |
1747 | * @dip: The GFS2 inode | |
1748 | * @filename: | |
1749 | * @new_inode: | |
1750 | * | |
1751 | * This routine changes the inode number of a directory entry. It's used | |
1752 | * by rename to change ".." when a directory is moved. | |
1753 | * Assumes a glock is held on dvp. | |
1754 | * | |
1755 | * Returns: errno | |
1756 | */ | |
1757 | ||
c752666c | 1758 | int gfs2_dir_mvino(struct gfs2_inode *dip, const struct qstr *filename, |
dbb7cae2 | 1759 | const struct gfs2_inode *nip, unsigned int new_type) |
b3b94faa | 1760 | { |
c752666c SW |
1761 | struct buffer_head *bh; |
1762 | struct gfs2_dirent *dent; | |
b3b94faa DT |
1763 | int error; |
1764 | ||
feaa7bba | 1765 | dent = gfs2_dirent_search(&dip->i_inode, filename, gfs2_dirent_find, &bh); |
c752666c SW |
1766 | if (!dent) { |
1767 | gfs2_consist_inode(dip); | |
1768 | return -EIO; | |
1769 | } | |
1770 | if (IS_ERR(dent)) | |
1771 | return PTR_ERR(dent); | |
b3b94faa | 1772 | |
c752666c | 1773 | gfs2_trans_add_bh(dip->i_gl, bh, 1); |
dbb7cae2 | 1774 | gfs2_inum_out(nip, dent); |
c752666c SW |
1775 | dent->de_type = cpu_to_be16(new_type); |
1776 | ||
383f01fb | 1777 | if (dip->i_diskflags & GFS2_DIF_EXHASH) { |
c752666c SW |
1778 | brelse(bh); |
1779 | error = gfs2_meta_inode_buffer(dip, &bh); | |
1780 | if (error) | |
1781 | return error; | |
1782 | gfs2_trans_add_bh(dip->i_gl, bh, 1); | |
1783 | } | |
1784 | ||
4bd91ba1 | 1785 | dip->i_inode.i_mtime = dip->i_inode.i_ctime = CURRENT_TIME; |
539e5d6b | 1786 | gfs2_dinode_out(dip, bh->b_data); |
c752666c SW |
1787 | brelse(bh); |
1788 | return 0; | |
b3b94faa DT |
1789 | } |
1790 | ||
b3b94faa DT |
1791 | /** |
1792 | * leaf_dealloc - Deallocate a directory leaf | |
1793 | * @dip: the directory | |
1794 | * @index: the hash table offset in the directory | |
1795 | * @len: the number of pointers to this leaf | |
1796 | * @leaf_no: the leaf number | |
ec038c82 | 1797 | * @leaf_bh: buffer_head for the starting leaf |
d24a7a43 | 1798 | * last_dealloc: 1 if this is the final dealloc for the leaf, else 0 |
b3b94faa DT |
1799 | * |
1800 | * Returns: errno | |
1801 | */ | |
1802 | ||
cd915493 | 1803 | static int leaf_dealloc(struct gfs2_inode *dip, u32 index, u32 len, |
ec038c82 BP |
1804 | u64 leaf_no, struct buffer_head *leaf_bh, |
1805 | int last_dealloc) | |
b3b94faa | 1806 | { |
feaa7bba | 1807 | struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); |
c752666c | 1808 | struct gfs2_leaf *tmp_leaf; |
b3b94faa DT |
1809 | struct gfs2_rgrp_list rlist; |
1810 | struct buffer_head *bh, *dibh; | |
cd915493 | 1811 | u64 blk, nblk; |
b3b94faa DT |
1812 | unsigned int rg_blocks = 0, l_blocks = 0; |
1813 | char *ht; | |
cd915493 | 1814 | unsigned int x, size = len * sizeof(u64); |
b3b94faa DT |
1815 | int error; |
1816 | ||
1817 | memset(&rlist, 0, sizeof(struct gfs2_rgrp_list)); | |
1818 | ||
16c5f06f | 1819 | ht = kzalloc(size, GFP_NOFS); |
b3b94faa DT |
1820 | if (!ht) |
1821 | return -ENOMEM; | |
1822 | ||
182fe5ab CG |
1823 | if (!gfs2_alloc_get(dip)) { |
1824 | error = -ENOMEM; | |
1825 | goto out; | |
1826 | } | |
b3b94faa DT |
1827 | |
1828 | error = gfs2_quota_hold(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); | |
1829 | if (error) | |
182fe5ab | 1830 | goto out_put; |
b3b94faa | 1831 | |
6dbd8224 | 1832 | error = gfs2_rindex_hold(sdp, &dip->i_alloc->al_ri_gh); |
b3b94faa DT |
1833 | if (error) |
1834 | goto out_qs; | |
1835 | ||
1836 | /* Count the number of leaves */ | |
ec038c82 | 1837 | bh = leaf_bh; |
b3b94faa | 1838 | |
c752666c | 1839 | for (blk = leaf_no; blk; blk = nblk) { |
ec038c82 BP |
1840 | if (blk != leaf_no) { |
1841 | error = get_leaf(dip, blk, &bh); | |
1842 | if (error) | |
1843 | goto out_rlist; | |
1844 | } | |
c752666c SW |
1845 | tmp_leaf = (struct gfs2_leaf *)bh->b_data; |
1846 | nblk = be64_to_cpu(tmp_leaf->lf_next); | |
ec038c82 BP |
1847 | if (blk != leaf_no) |
1848 | brelse(bh); | |
b3b94faa DT |
1849 | |
1850 | gfs2_rlist_add(sdp, &rlist, blk); | |
1851 | l_blocks++; | |
1852 | } | |
1853 | ||
fe6c991c | 1854 | gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE); |
b3b94faa DT |
1855 | |
1856 | for (x = 0; x < rlist.rl_rgrps; x++) { | |
1857 | struct gfs2_rgrpd *rgd; | |
5c676f6d | 1858 | rgd = rlist.rl_ghs[x].gh_gl->gl_object; |
bb8d8a6f | 1859 | rg_blocks += rgd->rd_length; |
b3b94faa DT |
1860 | } |
1861 | ||
1862 | error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs); | |
1863 | if (error) | |
1864 | goto out_rlist; | |
1865 | ||
1866 | error = gfs2_trans_begin(sdp, | |
5c676f6d | 1867 | rg_blocks + (DIV_ROUND_UP(size, sdp->sd_jbsize) + 1) + |
b3b94faa DT |
1868 | RES_DINODE + RES_STATFS + RES_QUOTA, l_blocks); |
1869 | if (error) | |
1870 | goto out_rg_gunlock; | |
1871 | ||
ec038c82 BP |
1872 | bh = leaf_bh; |
1873 | ||
c752666c | 1874 | for (blk = leaf_no; blk; blk = nblk) { |
ec038c82 BP |
1875 | if (blk != leaf_no) { |
1876 | error = get_leaf(dip, blk, &bh); | |
1877 | if (error) | |
1878 | goto out_end_trans; | |
1879 | } | |
c752666c SW |
1880 | tmp_leaf = (struct gfs2_leaf *)bh->b_data; |
1881 | nblk = be64_to_cpu(tmp_leaf->lf_next); | |
ec038c82 BP |
1882 | if (blk != leaf_no) |
1883 | brelse(bh); | |
b3b94faa DT |
1884 | |
1885 | gfs2_free_meta(dip, blk, 1); | |
77658aad | 1886 | gfs2_add_inode_blocks(&dip->i_inode, -1); |
b3b94faa DT |
1887 | } |
1888 | ||
cd915493 | 1889 | error = gfs2_dir_write_data(dip, ht, index * sizeof(u64), size); |
b3b94faa DT |
1890 | if (error != size) { |
1891 | if (error >= 0) | |
1892 | error = -EIO; | |
1893 | goto out_end_trans; | |
1894 | } | |
1895 | ||
1896 | error = gfs2_meta_inode_buffer(dip, &dibh); | |
1897 | if (error) | |
1898 | goto out_end_trans; | |
1899 | ||
d4e9c4c3 | 1900 | gfs2_trans_add_bh(dip->i_gl, dibh, 1); |
d24a7a43 BP |
1901 | /* On the last dealloc, make this a regular file in case we crash. |
1902 | (We don't want to free these blocks a second time.) */ | |
1903 | if (last_dealloc) | |
1904 | dip->i_inode.i_mode = S_IFREG; | |
539e5d6b | 1905 | gfs2_dinode_out(dip, dibh->b_data); |
b3b94faa DT |
1906 | brelse(dibh); |
1907 | ||
a91ea69f | 1908 | out_end_trans: |
b3b94faa | 1909 | gfs2_trans_end(sdp); |
a91ea69f | 1910 | out_rg_gunlock: |
b3b94faa | 1911 | gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs); |
a91ea69f | 1912 | out_rlist: |
b3b94faa | 1913 | gfs2_rlist_free(&rlist); |
6dbd8224 | 1914 | gfs2_glock_dq_uninit(&dip->i_alloc->al_ri_gh); |
a91ea69f | 1915 | out_qs: |
b3b94faa | 1916 | gfs2_quota_unhold(dip); |
182fe5ab | 1917 | out_put: |
b3b94faa | 1918 | gfs2_alloc_put(dip); |
182fe5ab | 1919 | out: |
b3b94faa | 1920 | kfree(ht); |
b3b94faa DT |
1921 | return error; |
1922 | } | |
1923 | ||
1924 | /** | |
1925 | * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory | |
1926 | * @dip: the directory | |
1927 | * | |
1928 | * Dealloc all on-disk directory leaves to FREEMETA state | |
1929 | * Change on-disk inode type to "regular file" | |
1930 | * | |
1931 | * Returns: errno | |
1932 | */ | |
1933 | ||
1934 | int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip) | |
1935 | { | |
556bb179 BP |
1936 | struct buffer_head *bh; |
1937 | struct gfs2_leaf *leaf; | |
1938 | u32 hsize, len; | |
556bb179 BP |
1939 | u32 index = 0, next_index; |
1940 | __be64 *lp; | |
1941 | u64 leaf_no; | |
1942 | int error = 0, last; | |
1943 | ||
1944 | hsize = 1 << dip->i_depth; | |
556bb179 | 1945 | |
17d539f0 SW |
1946 | lp = gfs2_dir_get_hash_table(dip); |
1947 | if (IS_ERR(lp)) | |
1948 | return PTR_ERR(lp); | |
556bb179 BP |
1949 | |
1950 | while (index < hsize) { | |
17d539f0 | 1951 | leaf_no = be64_to_cpu(lp[index]); |
556bb179 BP |
1952 | if (leaf_no) { |
1953 | error = get_leaf(dip, leaf_no, &bh); | |
1954 | if (error) | |
1955 | goto out; | |
1956 | leaf = (struct gfs2_leaf *)bh->b_data; | |
1957 | len = 1 << (dip->i_depth - be16_to_cpu(leaf->lf_depth)); | |
1958 | ||
1959 | next_index = (index & ~(len - 1)) + len; | |
1960 | last = ((next_index >= hsize) ? 1 : 0); | |
1961 | error = leaf_dealloc(dip, index, len, leaf_no, bh, | |
1962 | last); | |
1963 | brelse(bh); | |
1964 | if (error) | |
1965 | goto out; | |
1966 | index = next_index; | |
1967 | } else | |
1968 | index++; | |
1969 | } | |
1970 | ||
1971 | if (index != hsize) { | |
1972 | gfs2_consist_inode(dip); | |
1973 | error = -EIO; | |
1974 | } | |
1975 | ||
1976 | out: | |
556bb179 BP |
1977 | |
1978 | return error; | |
b3b94faa DT |
1979 | } |
1980 | ||
1981 | /** | |
1982 | * gfs2_diradd_alloc_required - find if adding entry will require an allocation | |
1983 | * @ip: the file being written to | |
1984 | * @filname: the filename that's going to be added | |
b3b94faa | 1985 | * |
c752666c | 1986 | * Returns: 1 if alloc required, 0 if not, -ve on error |
b3b94faa DT |
1987 | */ |
1988 | ||
4d8012b6 | 1989 | int gfs2_diradd_alloc_required(struct inode *inode, const struct qstr *name) |
b3b94faa | 1990 | { |
c752666c SW |
1991 | struct gfs2_dirent *dent; |
1992 | struct buffer_head *bh; | |
b3b94faa | 1993 | |
c752666c | 1994 | dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, &bh); |
ed386507 | 1995 | if (!dent) { |
c752666c | 1996 | return 1; |
ed386507 | 1997 | } |
c752666c SW |
1998 | if (IS_ERR(dent)) |
1999 | return PTR_ERR(dent); | |
2000 | brelse(bh); | |
2001 | return 0; | |
b3b94faa DT |
2002 | } |
2003 |