Commit | Line | Data |
---|---|---|
7336d0e6 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
b3b94faa DT |
2 | /* |
3 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
3a8a9a10 | 4 | * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved. |
b3b94faa DT |
5 | */ |
6 | ||
7 | /* | |
61e085a8 SW |
8 | * Implements Extendible Hashing as described in: |
9 | * "Extendible Hashing" by Fagin, et al in | |
10 | * __ACM Trans. on Database Systems__, Sept 1979. | |
11 | * | |
12 | * | |
13 | * Here's the layout of dirents which is essentially the same as that of ext2 | |
14 | * within a single block. The field de_name_len is the number of bytes | |
15 | * actually required for the name (no null terminator). The field de_rec_len | |
16 | * is the number of bytes allocated to the dirent. The offset of the next | |
17 | * dirent in the block is (dirent + dirent->de_rec_len). When a dirent is | |
18 | * deleted, the preceding dirent inherits its allocated space, ie | |
19 | * prev->de_rec_len += deleted->de_rec_len. Since the next dirent is obtained | |
20 | * by adding de_rec_len to the current dirent, this essentially causes the | |
21 | * deleted dirent to get jumped over when iterating through all the dirents. | |
22 | * | |
23 | * When deleting the first dirent in a block, there is no previous dirent so | |
24 | * the field de_ino is set to zero to designate it as deleted. When allocating | |
25 | * a dirent, gfs2_dirent_alloc iterates through the dirents in a block. If the | |
26 | * first dirent has (de_ino == 0) and de_rec_len is large enough, this first | |
27 | * dirent is allocated. Otherwise it must go through all the 'used' dirents | |
28 | * searching for one in which the amount of total space minus the amount of | |
29 | * used space will provide enough space for the new dirent. | |
30 | * | |
31 | * There are two types of blocks in which dirents reside. In a stuffed dinode, | |
32 | * the dirents begin at offset sizeof(struct gfs2_dinode) from the beginning of | |
33 | * the block. In leaves, they begin at offset sizeof(struct gfs2_leaf) from the | |
34 | * beginning of the leaf block. The dirents reside in leaves when | |
35 | * | |
383f01fb | 36 | * dip->i_diskflags & GFS2_DIF_EXHASH is true |
61e085a8 SW |
37 | * |
38 | * Otherwise, the dirents are "linear", within a single stuffed dinode block. | |
39 | * | |
40 | * When the dirents are in leaves, the actual contents of the directory file are | |
41 | * used as an array of 64-bit block pointers pointing to the leaf blocks. The | |
42 | * dirents are NOT in the directory file itself. There can be more than one | |
43 | * block pointer in the array that points to the same leaf. In fact, when a | |
44 | * directory is first converted from linear to exhash, all of the pointers | |
45 | * point to the same leaf. | |
46 | * | |
47 | * When a leaf is completely full, the size of the hash table can be | |
48 | * doubled unless it is already at the maximum size which is hard coded into | |
49 | * GFS2_DIR_MAX_DEPTH. After that, leaves are chained together in a linked list, | |
50 | * but never before the maximum hash table size has been reached. | |
51 | */ | |
b3b94faa | 52 | |
d77d1b58 JP |
53 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
54 | ||
b3b94faa DT |
55 | #include <linux/slab.h> |
56 | #include <linux/spinlock.h> | |
b3b94faa DT |
57 | #include <linux/buffer_head.h> |
58 | #include <linux/sort.h> | |
5c676f6d | 59 | #include <linux/gfs2_ondisk.h> |
71b86f56 | 60 | #include <linux/crc32.h> |
fe1bdedc | 61 | #include <linux/vmalloc.h> |
2f8b5444 | 62 | #include <linux/bio.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 | 75 | |
dfe4d34b BP |
76 | #define MAX_RA_BLOCKS 32 /* max read-ahead blocks */ |
77 | ||
cd915493 SW |
78 | #define gfs2_disk_hash2offset(h) (((u64)(h)) >> 1) |
79 | #define gfs2_dir_offset2hash(p) ((u32)(((u64)(p)) << 1)) | |
471f3db2 BM |
80 | #define GFS2_HASH_INDEX_MASK 0xffffc000 |
81 | #define GFS2_USE_HASH_FLAG 0x2000 | |
b3b94faa | 82 | |
8d123585 SW |
83 | struct qstr gfs2_qdot __read_mostly; |
84 | struct qstr gfs2_qdotdot __read_mostly; | |
85 | ||
2bdbc5d7 SW |
86 | typedef int (*gfs2_dscan_t)(const struct gfs2_dirent *dent, |
87 | const struct qstr *name, void *opaque); | |
b3b94faa | 88 | |
cd915493 | 89 | int gfs2_dir_get_new_buffer(struct gfs2_inode *ip, u64 block, |
61e085a8 | 90 | struct buffer_head **bhp) |
e13940ba SW |
91 | { |
92 | struct buffer_head *bh; | |
e13940ba | 93 | |
61e085a8 | 94 | bh = gfs2_meta_new(ip->i_gl, block); |
350a9b0a | 95 | gfs2_trans_add_meta(ip->i_gl, bh); |
61e085a8 SW |
96 | gfs2_metatype_set(bh, GFS2_METATYPE_JD, GFS2_FORMAT_JD); |
97 | gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header)); | |
e13940ba SW |
98 | *bhp = bh; |
99 | return 0; | |
100 | } | |
101 | ||
cd915493 | 102 | static int gfs2_dir_get_existing_buffer(struct gfs2_inode *ip, u64 block, |
61e085a8 SW |
103 | struct buffer_head **bhp) |
104 | { | |
105 | struct buffer_head *bh; | |
106 | int error; | |
e13940ba | 107 | |
c8d57703 | 108 | error = gfs2_meta_read(ip->i_gl, block, DIO_WAIT, 0, &bh); |
61e085a8 SW |
109 | if (error) |
110 | return error; | |
feaa7bba | 111 | if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), bh, GFS2_METATYPE_JD)) { |
61e085a8 SW |
112 | brelse(bh); |
113 | return -EIO; | |
114 | } | |
115 | *bhp = bh; | |
116 | return 0; | |
117 | } | |
e13940ba SW |
118 | |
119 | static int gfs2_dir_write_stuffed(struct gfs2_inode *ip, const char *buf, | |
120 | unsigned int offset, unsigned int size) | |
e13940ba SW |
121 | { |
122 | struct buffer_head *dibh; | |
123 | int error; | |
124 | ||
125 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
126 | if (error) | |
127 | return error; | |
128 | ||
350a9b0a | 129 | gfs2_trans_add_meta(ip->i_gl, dibh); |
c752666c | 130 | memcpy(dibh->b_data + offset + sizeof(struct gfs2_dinode), buf, size); |
a2e0f799 SW |
131 | if (ip->i_inode.i_size < offset + size) |
132 | i_size_write(&ip->i_inode, offset + size); | |
078cd827 | 133 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode); |
539e5d6b | 134 | gfs2_dinode_out(ip, dibh->b_data); |
e13940ba SW |
135 | |
136 | brelse(dibh); | |
137 | ||
138 | return size; | |
139 | } | |
140 | ||
141 | ||
142 | ||
143 | /** | |
144 | * gfs2_dir_write_data - Write directory information to the inode | |
145 | * @ip: The GFS2 inode | |
146 | * @buf: The buffer containing information to be written | |
147 | * @offset: The file offset to start writing at | |
148 | * @size: The amount of data to write | |
149 | * | |
150 | * Returns: The number of bytes correctly written or error code | |
151 | */ | |
152 | static int gfs2_dir_write_data(struct gfs2_inode *ip, const char *buf, | |
cd915493 | 153 | u64 offset, unsigned int size) |
e13940ba | 154 | { |
feaa7bba | 155 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
e13940ba | 156 | struct buffer_head *dibh; |
cd915493 SW |
157 | u64 lblock, dblock; |
158 | u32 extlen = 0; | |
e13940ba SW |
159 | unsigned int o; |
160 | int copied = 0; | |
161 | int error = 0; | |
9b8c81d1 | 162 | int new = 0; |
e13940ba SW |
163 | |
164 | if (!size) | |
165 | return 0; | |
166 | ||
235628c5 | 167 | if (gfs2_is_stuffed(ip) && offset + size <= gfs2_max_stuffed_size(ip)) |
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 | ||
350a9b0a | 210 | gfs2_trans_add_meta(ip->i_gl, bh); |
e13940ba SW |
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); | |
078cd827 | 230 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode); |
e13940ba | 231 | |
350a9b0a | 232 | gfs2_trans_add_meta(ip->i_gl, dibh); |
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 | ||
4c28d338 SW |
243 | static int gfs2_dir_read_stuffed(struct gfs2_inode *ip, __be64 *buf, |
244 | 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) { | |
4c28d338 | 251 | memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size); |
e13940ba SW |
252 | brelse(dibh); |
253 | } | |
254 | ||
255 | return (error) ? error : size; | |
256 | } | |
257 | ||
258 | ||
259 | /** | |
260 | * gfs2_dir_read_data - Read a data from a directory inode | |
261 | * @ip: The GFS2 Inode | |
262 | * @buf: The buffer to place result into | |
e13940ba SW |
263 | * @size: Amount of data to transfer |
264 | * | |
265 | * Returns: The amount of data actually copied or the error | |
266 | */ | |
4c28d338 SW |
267 | static int gfs2_dir_read_data(struct gfs2_inode *ip, __be64 *buf, |
268 | unsigned int size) | |
e13940ba | 269 | { |
feaa7bba | 270 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
cd915493 SW |
271 | u64 lblock, dblock; |
272 | u32 extlen = 0; | |
e13940ba SW |
273 | unsigned int o; |
274 | int copied = 0; | |
275 | int error = 0; | |
e13940ba SW |
276 | |
277 | if (gfs2_is_stuffed(ip)) | |
4c28d338 | 278 | return gfs2_dir_read_stuffed(ip, buf, size); |
e13940ba SW |
279 | |
280 | if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip))) | |
281 | return -EINVAL; | |
282 | ||
4c28d338 | 283 | lblock = 0; |
e13940ba SW |
284 | o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header); |
285 | ||
286 | while (copied < size) { | |
287 | unsigned int amount; | |
288 | struct buffer_head *bh; | |
289 | int new; | |
290 | ||
291 | amount = size - copied; | |
292 | if (amount > sdp->sd_sb.sb_bsize - o) | |
293 | amount = sdp->sd_sb.sb_bsize - o; | |
294 | ||
295 | if (!extlen) { | |
296 | new = 0; | |
feaa7bba | 297 | error = gfs2_extent_map(&ip->i_inode, lblock, &new, |
fd88de56 | 298 | &dblock, &extlen); |
7276b3b0 | 299 | if (error || !dblock) |
e13940ba | 300 | goto fail; |
7276b3b0 | 301 | BUG_ON(extlen < 1); |
7276b3b0 | 302 | bh = gfs2_meta_ra(ip->i_gl, dblock, extlen); |
b7d8ac3e | 303 | } else { |
c8d57703 | 304 | error = gfs2_meta_read(ip->i_gl, dblock, DIO_WAIT, 0, &bh); |
e13940ba SW |
305 | if (error) |
306 | goto fail; | |
7276b3b0 SW |
307 | } |
308 | error = gfs2_metatype_check(sdp, bh, GFS2_METATYPE_JD); | |
309 | if (error) { | |
310 | brelse(bh); | |
311 | goto fail; | |
312 | } | |
313 | dblock++; | |
314 | extlen--; | |
e13940ba SW |
315 | memcpy(buf, bh->b_data + o, amount); |
316 | brelse(bh); | |
4c28d338 | 317 | buf += (amount/sizeof(__be64)); |
e13940ba SW |
318 | copied += amount; |
319 | lblock++; | |
e13940ba SW |
320 | o = sizeof(struct gfs2_meta_header); |
321 | } | |
322 | ||
323 | return copied; | |
324 | fail: | |
325 | return (copied) ? copied : error; | |
326 | } | |
327 | ||
17d539f0 SW |
328 | /** |
329 | * gfs2_dir_get_hash_table - Get pointer to the dir hash table | |
330 | * @ip: The inode in question | |
331 | * | |
332 | * Returns: The hash table or an error | |
333 | */ | |
334 | ||
335 | static __be64 *gfs2_dir_get_hash_table(struct gfs2_inode *ip) | |
336 | { | |
337 | struct inode *inode = &ip->i_inode; | |
338 | int ret; | |
339 | u32 hsize; | |
340 | __be64 *hc; | |
341 | ||
342 | BUG_ON(!(ip->i_diskflags & GFS2_DIF_EXHASH)); | |
343 | ||
344 | hc = ip->i_hash_cache; | |
345 | if (hc) | |
346 | return hc; | |
347 | ||
47a9a527 | 348 | hsize = BIT(ip->i_depth); |
17d539f0 SW |
349 | hsize *= sizeof(__be64); |
350 | if (hsize != i_size_read(&ip->i_inode)) { | |
351 | gfs2_consist_inode(ip); | |
352 | return ERR_PTR(-EIO); | |
353 | } | |
354 | ||
e8830d88 BP |
355 | hc = kmalloc(hsize, GFP_NOFS | __GFP_NOWARN); |
356 | if (hc == NULL) | |
88dca4ca | 357 | hc = __vmalloc(hsize, GFP_NOFS); |
e8830d88 | 358 | |
17d539f0 SW |
359 | if (hc == NULL) |
360 | return ERR_PTR(-ENOMEM); | |
361 | ||
4c28d338 | 362 | ret = gfs2_dir_read_data(ip, hc, hsize); |
17d539f0 | 363 | if (ret < 0) { |
3cdcf63e | 364 | kvfree(hc); |
17d539f0 SW |
365 | return ERR_PTR(ret); |
366 | } | |
367 | ||
368 | spin_lock(&inode->i_lock); | |
9265f1d0 | 369 | if (likely(!ip->i_hash_cache)) { |
17d539f0 | 370 | ip->i_hash_cache = hc; |
9265f1d0 AV |
371 | hc = NULL; |
372 | } | |
17d539f0 | 373 | spin_unlock(&inode->i_lock); |
9265f1d0 | 374 | kvfree(hc); |
17d539f0 SW |
375 | |
376 | return ip->i_hash_cache; | |
377 | } | |
378 | ||
379 | /** | |
380 | * gfs2_dir_hash_inval - Invalidate dir hash | |
381 | * @ip: The directory inode | |
382 | * | |
383 | * Must be called with an exclusive glock, or during glock invalidation. | |
384 | */ | |
385 | void gfs2_dir_hash_inval(struct gfs2_inode *ip) | |
386 | { | |
c36b97e9 BP |
387 | __be64 *hc; |
388 | ||
389 | spin_lock(&ip->i_inode.i_lock); | |
390 | hc = ip->i_hash_cache; | |
17d539f0 | 391 | ip->i_hash_cache = NULL; |
c36b97e9 BP |
392 | spin_unlock(&ip->i_inode.i_lock); |
393 | ||
3cdcf63e | 394 | kvfree(hc); |
17d539f0 SW |
395 | } |
396 | ||
5e7d65cd SW |
397 | static inline int gfs2_dirent_sentinel(const struct gfs2_dirent *dent) |
398 | { | |
399 | return dent->de_inum.no_addr == 0 || dent->de_inum.no_formal_ino == 0; | |
400 | } | |
401 | ||
c752666c SW |
402 | static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent, |
403 | const struct qstr *name, int ret) | |
404 | { | |
5e7d65cd | 405 | if (!gfs2_dirent_sentinel(dent) && |
c752666c SW |
406 | be32_to_cpu(dent->de_hash) == name->hash && |
407 | be16_to_cpu(dent->de_name_len) == name->len && | |
2bdbc5d7 | 408 | memcmp(dent+1, name->name, name->len) == 0) |
c752666c SW |
409 | return ret; |
410 | return 0; | |
411 | } | |
412 | ||
413 | static int gfs2_dirent_find(const struct gfs2_dirent *dent, | |
71b86f56 SW |
414 | const struct qstr *name, |
415 | void *opaque) | |
c752666c SW |
416 | { |
417 | return __gfs2_dirent_find(dent, name, 1); | |
418 | } | |
419 | ||
420 | static int gfs2_dirent_prev(const struct gfs2_dirent *dent, | |
71b86f56 SW |
421 | const struct qstr *name, |
422 | void *opaque) | |
c752666c SW |
423 | { |
424 | return __gfs2_dirent_find(dent, name, 2); | |
425 | } | |
426 | ||
427 | /* | |
428 | * name->name holds ptr to start of block. | |
429 | * name->len holds size of block. | |
b3b94faa | 430 | */ |
c752666c | 431 | static int gfs2_dirent_last(const struct gfs2_dirent *dent, |
71b86f56 SW |
432 | const struct qstr *name, |
433 | void *opaque) | |
c752666c SW |
434 | { |
435 | const char *start = name->name; | |
436 | const char *end = (const char *)dent + be16_to_cpu(dent->de_rec_len); | |
437 | if (name->len == (end - start)) | |
438 | return 1; | |
439 | return 0; | |
440 | } | |
b3b94faa | 441 | |
34017472 BM |
442 | /* Look for the dirent that contains the offset specified in data. Once we |
443 | * find that dirent, there must be space available there for the new dirent */ | |
444 | static int gfs2_dirent_find_offset(const struct gfs2_dirent *dent, | |
445 | const struct qstr *name, | |
446 | void *ptr) | |
447 | { | |
448 | unsigned required = GFS2_DIRENT_SIZE(name->len); | |
449 | unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len)); | |
450 | unsigned totlen = be16_to_cpu(dent->de_rec_len); | |
451 | ||
452 | if (ptr < (void *)dent || ptr >= (void *)dent + totlen) | |
453 | return 0; | |
454 | if (gfs2_dirent_sentinel(dent)) | |
455 | actual = 0; | |
456 | if (ptr < (void *)dent + actual) | |
457 | return -1; | |
458 | if ((void *)dent + totlen >= ptr + required) | |
459 | return 1; | |
460 | return -1; | |
461 | } | |
462 | ||
c752666c | 463 | static int gfs2_dirent_find_space(const struct gfs2_dirent *dent, |
71b86f56 SW |
464 | const struct qstr *name, |
465 | void *opaque) | |
b3b94faa | 466 | { |
c752666c SW |
467 | unsigned required = GFS2_DIRENT_SIZE(name->len); |
468 | unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len)); | |
469 | unsigned totlen = be16_to_cpu(dent->de_rec_len); | |
470 | ||
5e7d65cd | 471 | if (gfs2_dirent_sentinel(dent)) |
728a756b | 472 | actual = 0; |
c5392124 | 473 | if (totlen - actual >= required) |
c752666c SW |
474 | return 1; |
475 | return 0; | |
476 | } | |
477 | ||
71b86f56 SW |
478 | struct dirent_gather { |
479 | const struct gfs2_dirent **pdent; | |
480 | unsigned offset; | |
481 | }; | |
482 | ||
483 | static int gfs2_dirent_gather(const struct gfs2_dirent *dent, | |
484 | const struct qstr *name, | |
485 | void *opaque) | |
486 | { | |
487 | struct dirent_gather *g = opaque; | |
5e7d65cd | 488 | if (!gfs2_dirent_sentinel(dent)) { |
71b86f56 SW |
489 | g->pdent[g->offset++] = dent; |
490 | } | |
491 | return 0; | |
492 | } | |
493 | ||
c752666c SW |
494 | /* |
495 | * Other possible things to check: | |
496 | * - Inode located within filesystem size (and on valid block) | |
497 | * - Valid directory entry type | |
498 | * Not sure how heavy-weight we want to make this... could also check | |
499 | * hash is correct for example, but that would take a lot of extra time. | |
500 | * For now the most important thing is to check that the various sizes | |
501 | * are correct. | |
502 | */ | |
e54c78a2 BP |
503 | static int gfs2_check_dirent(struct gfs2_sbd *sdp, |
504 | struct gfs2_dirent *dent, unsigned int offset, | |
c752666c SW |
505 | unsigned int size, unsigned int len, int first) |
506 | { | |
507 | const char *msg = "gfs2_dirent too small"; | |
508 | if (unlikely(size < sizeof(struct gfs2_dirent))) | |
509 | goto error; | |
510 | msg = "gfs2_dirent misaligned"; | |
511 | if (unlikely(offset & 0x7)) | |
512 | goto error; | |
513 | msg = "gfs2_dirent points beyond end of block"; | |
514 | if (unlikely(offset + size > len)) | |
515 | goto error; | |
516 | msg = "zero inode number"; | |
5e7d65cd | 517 | if (unlikely(!first && gfs2_dirent_sentinel(dent))) |
c752666c SW |
518 | goto error; |
519 | msg = "name length is greater than space in dirent"; | |
5e7d65cd | 520 | if (!gfs2_dirent_sentinel(dent) && |
c752666c SW |
521 | unlikely(sizeof(struct gfs2_dirent)+be16_to_cpu(dent->de_name_len) > |
522 | size)) | |
523 | goto error; | |
524 | return 0; | |
525 | error: | |
e54c78a2 | 526 | fs_warn(sdp, "%s: %s (%s)\n", |
d77d1b58 | 527 | __func__, msg, first ? "first in block" : "not first in block"); |
c752666c | 528 | return -EIO; |
b3b94faa DT |
529 | } |
530 | ||
e54c78a2 | 531 | static int gfs2_dirent_offset(struct gfs2_sbd *sdp, const void *buf) |
c752666c | 532 | { |
71b86f56 SW |
533 | const struct gfs2_meta_header *h = buf; |
534 | int offset; | |
c752666c SW |
535 | |
536 | BUG_ON(buf == NULL); | |
c752666c | 537 | |
e3167ded | 538 | switch(be32_to_cpu(h->mh_type)) { |
c752666c SW |
539 | case GFS2_METATYPE_LF: |
540 | offset = sizeof(struct gfs2_leaf); | |
541 | break; | |
542 | case GFS2_METATYPE_DI: | |
543 | offset = sizeof(struct gfs2_dinode); | |
544 | break; | |
545 | default: | |
546 | goto wrong_type; | |
547 | } | |
71b86f56 SW |
548 | return offset; |
549 | wrong_type: | |
e54c78a2 BP |
550 | fs_warn(sdp, "%s: wrong block type %u\n", __func__, |
551 | be32_to_cpu(h->mh_type)); | |
71b86f56 SW |
552 | return -1; |
553 | } | |
554 | ||
2bdbc5d7 | 555 | static struct gfs2_dirent *gfs2_dirent_scan(struct inode *inode, void *buf, |
71b86f56 SW |
556 | unsigned int len, gfs2_dscan_t scan, |
557 | const struct qstr *name, | |
558 | void *opaque) | |
559 | { | |
560 | struct gfs2_dirent *dent, *prev; | |
561 | unsigned offset; | |
562 | unsigned size; | |
563 | int ret = 0; | |
c752666c | 564 | |
e54c78a2 | 565 | ret = gfs2_dirent_offset(GFS2_SB(inode), buf); |
71b86f56 SW |
566 | if (ret < 0) |
567 | goto consist_inode; | |
568 | ||
569 | offset = ret; | |
c752666c | 570 | prev = NULL; |
2bdbc5d7 | 571 | dent = buf + offset; |
c752666c | 572 | size = be16_to_cpu(dent->de_rec_len); |
e54c78a2 | 573 | if (gfs2_check_dirent(GFS2_SB(inode), dent, offset, size, len, 1)) |
c752666c SW |
574 | goto consist_inode; |
575 | do { | |
71b86f56 | 576 | ret = scan(dent, name, opaque); |
c752666c SW |
577 | if (ret) |
578 | break; | |
579 | offset += size; | |
580 | if (offset == len) | |
581 | break; | |
582 | prev = dent; | |
2bdbc5d7 | 583 | dent = buf + offset; |
c752666c | 584 | size = be16_to_cpu(dent->de_rec_len); |
e54c78a2 BP |
585 | if (gfs2_check_dirent(GFS2_SB(inode), dent, offset, size, |
586 | len, 0)) | |
c752666c SW |
587 | goto consist_inode; |
588 | } while(1); | |
589 | ||
590 | switch(ret) { | |
591 | case 0: | |
592 | return NULL; | |
593 | case 1: | |
594 | return dent; | |
595 | case 2: | |
596 | return prev ? prev : dent; | |
597 | default: | |
598 | BUG_ON(ret > 0); | |
599 | return ERR_PTR(ret); | |
600 | } | |
601 | ||
c752666c | 602 | consist_inode: |
feaa7bba | 603 | gfs2_consist_inode(GFS2_I(inode)); |
c752666c SW |
604 | return ERR_PTR(-EIO); |
605 | } | |
606 | ||
2bdbc5d7 SW |
607 | static int dirent_check_reclen(struct gfs2_inode *dip, |
608 | const struct gfs2_dirent *d, const void *end_p) | |
609 | { | |
610 | const void *ptr = d; | |
611 | u16 rec_len = be16_to_cpu(d->de_rec_len); | |
612 | ||
613 | if (unlikely(rec_len < sizeof(struct gfs2_dirent))) | |
614 | goto broken; | |
615 | ptr += rec_len; | |
616 | if (ptr < end_p) | |
617 | return rec_len; | |
618 | if (ptr == end_p) | |
619 | return -ENOENT; | |
620 | broken: | |
621 | gfs2_consist_inode(dip); | |
622 | return -EIO; | |
623 | } | |
624 | ||
b3b94faa DT |
625 | /** |
626 | * dirent_next - Next dirent | |
627 | * @dip: the directory | |
628 | * @bh: The buffer | |
629 | * @dent: Pointer to list of dirents | |
630 | * | |
631 | * Returns: 0 on success, error code otherwise | |
632 | */ | |
633 | ||
634 | static int dirent_next(struct gfs2_inode *dip, struct buffer_head *bh, | |
635 | struct gfs2_dirent **dent) | |
636 | { | |
2bdbc5d7 SW |
637 | struct gfs2_dirent *cur = *dent, *tmp; |
638 | char *bh_end = bh->b_data + bh->b_size; | |
639 | int ret; | |
b3b94faa | 640 | |
2bdbc5d7 SW |
641 | ret = dirent_check_reclen(dip, cur, bh_end); |
642 | if (ret < 0) | |
643 | return ret; | |
4dd651ad | 644 | |
2bdbc5d7 SW |
645 | tmp = (void *)cur + ret; |
646 | ret = dirent_check_reclen(dip, tmp, bh_end); | |
647 | if (ret == -EIO) | |
648 | return ret; | |
4dd651ad | 649 | |
b3b94faa | 650 | /* Only the first dent could ever have de_inum.no_addr == 0 */ |
5e7d65cd | 651 | if (gfs2_dirent_sentinel(tmp)) { |
b3b94faa DT |
652 | gfs2_consist_inode(dip); |
653 | return -EIO; | |
654 | } | |
655 | ||
656 | *dent = tmp; | |
b3b94faa DT |
657 | return 0; |
658 | } | |
659 | ||
660 | /** | |
661 | * dirent_del - Delete a dirent | |
662 | * @dip: The GFS2 inode | |
663 | * @bh: The buffer | |
664 | * @prev: The previous dirent | |
665 | * @cur: The current dirent | |
666 | * | |
667 | */ | |
668 | ||
669 | static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh, | |
670 | struct gfs2_dirent *prev, struct gfs2_dirent *cur) | |
671 | { | |
cd915493 | 672 | u16 cur_rec_len, prev_rec_len; |
b3b94faa | 673 | |
5e7d65cd | 674 | if (gfs2_dirent_sentinel(cur)) { |
b3b94faa DT |
675 | gfs2_consist_inode(dip); |
676 | return; | |
677 | } | |
678 | ||
350a9b0a | 679 | gfs2_trans_add_meta(dip->i_gl, bh); |
b3b94faa DT |
680 | |
681 | /* If there is no prev entry, this is the first entry in the block. | |
682 | The de_rec_len is already as big as it needs to be. Just zero | |
683 | out the inode number and return. */ | |
684 | ||
685 | if (!prev) { | |
5e7d65cd SW |
686 | cur->de_inum.no_addr = 0; |
687 | cur->de_inum.no_formal_ino = 0; | |
b3b94faa DT |
688 | return; |
689 | } | |
690 | ||
691 | /* Combine this dentry with the previous one. */ | |
692 | ||
fc69d0d3 SW |
693 | prev_rec_len = be16_to_cpu(prev->de_rec_len); |
694 | cur_rec_len = be16_to_cpu(cur->de_rec_len); | |
b3b94faa DT |
695 | |
696 | if ((char *)prev + prev_rec_len != (char *)cur) | |
697 | gfs2_consist_inode(dip); | |
698 | if ((char *)cur + cur_rec_len > bh->b_data + bh->b_size) | |
699 | gfs2_consist_inode(dip); | |
700 | ||
701 | prev_rec_len += cur_rec_len; | |
fc69d0d3 | 702 | prev->de_rec_len = cpu_to_be16(prev_rec_len); |
b3b94faa DT |
703 | } |
704 | ||
34017472 BM |
705 | |
706 | static struct gfs2_dirent *do_init_dirent(struct inode *inode, | |
707 | struct gfs2_dirent *dent, | |
708 | const struct qstr *name, | |
709 | struct buffer_head *bh, | |
710 | unsigned offset) | |
711 | { | |
712 | struct gfs2_inode *ip = GFS2_I(inode); | |
713 | struct gfs2_dirent *ndent; | |
714 | unsigned totlen; | |
715 | ||
716 | totlen = be16_to_cpu(dent->de_rec_len); | |
717 | BUG_ON(offset + name->len > totlen); | |
718 | gfs2_trans_add_meta(ip->i_gl, bh); | |
719 | ndent = (struct gfs2_dirent *)((char *)dent + offset); | |
720 | dent->de_rec_len = cpu_to_be16(offset); | |
721 | gfs2_qstr2dirent(name, totlen - offset, ndent); | |
722 | return ndent; | |
723 | } | |
724 | ||
725 | ||
c752666c SW |
726 | /* |
727 | * Takes a dent from which to grab space as an argument. Returns the | |
728 | * newly created dent. | |
b3b94faa | 729 | */ |
08bc2dbc AB |
730 | static struct gfs2_dirent *gfs2_init_dirent(struct inode *inode, |
731 | struct gfs2_dirent *dent, | |
732 | const struct qstr *name, | |
733 | struct buffer_head *bh) | |
b3b94faa | 734 | { |
34017472 | 735 | unsigned offset = 0; |
c752666c | 736 | |
5e7d65cd | 737 | if (!gfs2_dirent_sentinel(dent)) |
c752666c | 738 | offset = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len)); |
34017472 | 739 | return do_init_dirent(inode, dent, name, bh, offset); |
b3b94faa DT |
740 | } |
741 | ||
34017472 BM |
742 | static struct gfs2_dirent *gfs2_dirent_split_alloc(struct inode *inode, |
743 | struct buffer_head *bh, | |
744 | const struct qstr *name, | |
745 | void *ptr) | |
b3b94faa DT |
746 | { |
747 | struct gfs2_dirent *dent; | |
907b9bce | 748 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, |
34017472 | 749 | gfs2_dirent_find_offset, name, ptr); |
15a798f7 | 750 | if (IS_ERR_OR_NULL(dent)) |
c752666c | 751 | return dent; |
34017472 BM |
752 | return do_init_dirent(inode, dent, name, bh, |
753 | (unsigned)(ptr - (void *)dent)); | |
b3b94faa DT |
754 | } |
755 | ||
cd915493 | 756 | static int get_leaf(struct gfs2_inode *dip, u64 leaf_no, |
b3b94faa DT |
757 | struct buffer_head **bhp) |
758 | { | |
759 | int error; | |
760 | ||
c8d57703 | 761 | error = gfs2_meta_read(dip->i_gl, leaf_no, DIO_WAIT, 0, bhp); |
feaa7bba | 762 | if (!error && gfs2_metatype_check(GFS2_SB(&dip->i_inode), *bhp, GFS2_METATYPE_LF)) { |
d77d1b58 | 763 | /* pr_info("block num=%llu\n", leaf_no); */ |
b3b94faa | 764 | error = -EIO; |
feaa7bba | 765 | } |
b3b94faa DT |
766 | |
767 | return error; | |
768 | } | |
769 | ||
770 | /** | |
771 | * get_leaf_nr - Get a leaf number associated with the index | |
772 | * @dip: The GFS2 inode | |
773 | * @index: | |
774 | * @leaf_out: | |
775 | * | |
776 | * Returns: 0 on success, error code otherwise | |
777 | */ | |
778 | ||
cd915493 SW |
779 | static int get_leaf_nr(struct gfs2_inode *dip, u32 index, |
780 | u64 *leaf_out) | |
b3b94faa | 781 | { |
17d539f0 | 782 | __be64 *hash; |
287980e4 | 783 | int error; |
b3b94faa | 784 | |
17d539f0 | 785 | hash = gfs2_dir_get_hash_table(dip); |
287980e4 AB |
786 | error = PTR_ERR_OR_ZERO(hash); |
787 | ||
788 | if (!error) | |
789 | *leaf_out = be64_to_cpu(*(hash + index)); | |
790 | ||
791 | return error; | |
b3b94faa DT |
792 | } |
793 | ||
cd915493 | 794 | static int get_first_leaf(struct gfs2_inode *dip, u32 index, |
b3b94faa DT |
795 | struct buffer_head **bh_out) |
796 | { | |
cd915493 | 797 | u64 leaf_no; |
b3b94faa DT |
798 | int error; |
799 | ||
800 | error = get_leaf_nr(dip, index, &leaf_no); | |
287980e4 | 801 | if (!error) |
b3b94faa DT |
802 | error = get_leaf(dip, leaf_no, bh_out); |
803 | ||
804 | return error; | |
805 | } | |
806 | ||
c752666c SW |
807 | static struct gfs2_dirent *gfs2_dirent_search(struct inode *inode, |
808 | const struct qstr *name, | |
809 | gfs2_dscan_t scan, | |
810 | struct buffer_head **pbh) | |
b3b94faa | 811 | { |
c752666c SW |
812 | struct buffer_head *bh; |
813 | struct gfs2_dirent *dent; | |
feaa7bba | 814 | struct gfs2_inode *ip = GFS2_I(inode); |
b3b94faa DT |
815 | int error; |
816 | ||
383f01fb | 817 | if (ip->i_diskflags & GFS2_DIF_EXHASH) { |
c752666c | 818 | struct gfs2_leaf *leaf; |
47a9a527 FF |
819 | unsigned int hsize = BIT(ip->i_depth); |
820 | unsigned int index; | |
c752666c | 821 | u64 ln; |
a2e0f799 | 822 | if (hsize * sizeof(u64) != i_size_read(inode)) { |
c752666c SW |
823 | gfs2_consist_inode(ip); |
824 | return ERR_PTR(-EIO); | |
825 | } | |
907b9bce | 826 | |
9a004508 | 827 | index = name->hash >> (32 - ip->i_depth); |
c752666c SW |
828 | error = get_first_leaf(ip, index, &bh); |
829 | if (error) | |
830 | return ERR_PTR(error); | |
831 | do { | |
832 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, | |
71b86f56 | 833 | scan, name, NULL); |
c752666c SW |
834 | if (dent) |
835 | goto got_dent; | |
836 | leaf = (struct gfs2_leaf *)bh->b_data; | |
837 | ln = be64_to_cpu(leaf->lf_next); | |
f4154ea0 | 838 | brelse(bh); |
c752666c SW |
839 | if (!ln) |
840 | break; | |
907b9bce | 841 | |
c752666c SW |
842 | error = get_leaf(ip, ln, &bh); |
843 | } while(!error); | |
b3b94faa | 844 | |
c752666c | 845 | return error ? ERR_PTR(error) : NULL; |
b3b94faa | 846 | } |
b3b94faa | 847 | |
907b9bce | 848 | |
c752666c SW |
849 | error = gfs2_meta_inode_buffer(ip, &bh); |
850 | if (error) | |
851 | return ERR_PTR(error); | |
71b86f56 | 852 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, scan, name, NULL); |
c752666c | 853 | got_dent: |
15a798f7 | 854 | if (IS_ERR_OR_NULL(dent)) { |
ed386507 SW |
855 | brelse(bh); |
856 | bh = NULL; | |
857 | } | |
c752666c SW |
858 | *pbh = bh; |
859 | return dent; | |
860 | } | |
b3b94faa | 861 | |
c752666c SW |
862 | static struct gfs2_leaf *new_leaf(struct inode *inode, struct buffer_head **pbh, u16 depth) |
863 | { | |
feaa7bba | 864 | struct gfs2_inode *ip = GFS2_I(inode); |
b45e41d7 | 865 | unsigned int n = 1; |
09010978 SW |
866 | u64 bn; |
867 | int error; | |
868 | struct buffer_head *bh; | |
c752666c SW |
869 | struct gfs2_leaf *leaf; |
870 | struct gfs2_dirent *dent; | |
95582b00 | 871 | struct timespec64 tv = current_time(inode); |
09010978 | 872 | |
6e87ed0f | 873 | error = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL); |
09010978 SW |
874 | if (error) |
875 | return NULL; | |
876 | bh = gfs2_meta_new(ip->i_gl, bn); | |
c752666c SW |
877 | if (!bh) |
878 | return NULL; | |
09010978 | 879 | |
fbb27873 | 880 | gfs2_trans_remove_revoke(GFS2_SB(inode), bn, 1); |
350a9b0a | 881 | gfs2_trans_add_meta(ip->i_gl, bh); |
c752666c SW |
882 | gfs2_metatype_set(bh, GFS2_METATYPE_LF, GFS2_FORMAT_LF); |
883 | leaf = (struct gfs2_leaf *)bh->b_data; | |
884 | leaf->lf_depth = cpu_to_be16(depth); | |
2bdbc5d7 | 885 | leaf->lf_entries = 0; |
a2d7d021 | 886 | leaf->lf_dirent_format = cpu_to_be32(GFS2_FORMAT_DE); |
2bdbc5d7 | 887 | leaf->lf_next = 0; |
01bcb0de SW |
888 | leaf->lf_inode = cpu_to_be64(ip->i_no_addr); |
889 | leaf->lf_dist = cpu_to_be32(1); | |
890 | leaf->lf_nsec = cpu_to_be32(tv.tv_nsec); | |
891 | leaf->lf_sec = cpu_to_be64(tv.tv_sec); | |
892 | memset(leaf->lf_reserved2, 0, sizeof(leaf->lf_reserved2)); | |
c752666c | 893 | dent = (struct gfs2_dirent *)(leaf+1); |
cdf01226 | 894 | gfs2_qstr2dirent(&empty_name, bh->b_size - sizeof(struct gfs2_leaf), dent); |
c752666c SW |
895 | *pbh = bh; |
896 | return leaf; | |
b3b94faa DT |
897 | } |
898 | ||
899 | /** | |
900 | * dir_make_exhash - Convert a stuffed directory into an ExHash directory | |
901 | * @dip: The GFS2 inode | |
902 | * | |
903 | * Returns: 0 on success, error code otherwise | |
904 | */ | |
905 | ||
c752666c | 906 | static int dir_make_exhash(struct inode *inode) |
b3b94faa | 907 | { |
feaa7bba SW |
908 | struct gfs2_inode *dip = GFS2_I(inode); |
909 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
b3b94faa | 910 | struct gfs2_dirent *dent; |
c752666c | 911 | struct qstr args; |
b3b94faa DT |
912 | struct buffer_head *bh, *dibh; |
913 | struct gfs2_leaf *leaf; | |
914 | int y; | |
cd915493 | 915 | u32 x; |
b44b84d7 AV |
916 | __be64 *lp; |
917 | u64 bn; | |
b3b94faa DT |
918 | int error; |
919 | ||
920 | error = gfs2_meta_inode_buffer(dip, &dibh); | |
921 | if (error) | |
922 | return error; | |
923 | ||
b3b94faa DT |
924 | /* Turn over a new leaf */ |
925 | ||
c752666c SW |
926 | leaf = new_leaf(inode, &bh, 0); |
927 | if (!leaf) | |
928 | return -ENOSPC; | |
929 | bn = bh->b_blocknr; | |
b3b94faa | 930 | |
47a9a527 | 931 | gfs2_assert(sdp, dip->i_entries < BIT(16)); |
ad6203f2 | 932 | leaf->lf_entries = cpu_to_be16(dip->i_entries); |
b3b94faa DT |
933 | |
934 | /* Copy dirents */ | |
935 | ||
936 | gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_leaf), dibh, | |
937 | sizeof(struct gfs2_dinode)); | |
938 | ||
939 | /* Find last entry */ | |
940 | ||
941 | x = 0; | |
c752666c SW |
942 | args.len = bh->b_size - sizeof(struct gfs2_dinode) + |
943 | sizeof(struct gfs2_leaf); | |
944 | args.name = bh->b_data; | |
feaa7bba | 945 | dent = gfs2_dirent_scan(&dip->i_inode, bh->b_data, bh->b_size, |
71b86f56 | 946 | gfs2_dirent_last, &args, NULL); |
c752666c SW |
947 | if (!dent) { |
948 | brelse(bh); | |
949 | brelse(dibh); | |
950 | return -EIO; | |
951 | } | |
952 | if (IS_ERR(dent)) { | |
953 | brelse(bh); | |
954 | brelse(dibh); | |
955 | return PTR_ERR(dent); | |
b3b94faa | 956 | } |
b3b94faa DT |
957 | |
958 | /* Adjust the last dirent's record length | |
959 | (Remember that dent still points to the last entry.) */ | |
960 | ||
4dd651ad | 961 | dent->de_rec_len = cpu_to_be16(be16_to_cpu(dent->de_rec_len) + |
b3b94faa | 962 | sizeof(struct gfs2_dinode) - |
4dd651ad | 963 | sizeof(struct gfs2_leaf)); |
b3b94faa DT |
964 | |
965 | brelse(bh); | |
966 | ||
967 | /* We're done with the new leaf block, now setup the new | |
968 | hash table. */ | |
969 | ||
350a9b0a | 970 | gfs2_trans_add_meta(dip->i_gl, dibh); |
b3b94faa DT |
971 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); |
972 | ||
b44b84d7 | 973 | lp = (__be64 *)(dibh->b_data + sizeof(struct gfs2_dinode)); |
b3b94faa DT |
974 | |
975 | for (x = sdp->sd_hash_ptrs; x--; lp++) | |
976 | *lp = cpu_to_be64(bn); | |
977 | ||
a2e0f799 | 978 | i_size_write(inode, sdp->sd_sb.sb_bsize / 2); |
77658aad | 979 | gfs2_add_inode_blocks(&dip->i_inode, 1); |
383f01fb | 980 | dip->i_diskflags |= GFS2_DIF_EXHASH; |
b3b94faa DT |
981 | |
982 | for (x = sdp->sd_hash_ptrs, y = -1; x; x >>= 1, y++) ; | |
9a004508 | 983 | dip->i_depth = y; |
b3b94faa | 984 | |
539e5d6b | 985 | gfs2_dinode_out(dip, dibh->b_data); |
b3b94faa DT |
986 | |
987 | brelse(dibh); | |
988 | ||
989 | return 0; | |
990 | } | |
991 | ||
992 | /** | |
993 | * dir_split_leaf - Split a leaf block into two | |
994 | * @dip: The GFS2 inode | |
995 | * @index: | |
996 | * @leaf_no: | |
997 | * | |
998 | * Returns: 0 on success, error code on failure | |
999 | */ | |
1000 | ||
c752666c | 1001 | static int dir_split_leaf(struct inode *inode, const struct qstr *name) |
b3b94faa | 1002 | { |
feaa7bba | 1003 | struct gfs2_inode *dip = GFS2_I(inode); |
b3b94faa DT |
1004 | struct buffer_head *nbh, *obh, *dibh; |
1005 | struct gfs2_leaf *nleaf, *oleaf; | |
4da3c646 | 1006 | struct gfs2_dirent *dent = NULL, *prev = NULL, *next = NULL, *new; |
cd915493 | 1007 | u32 start, len, half_len, divider; |
b44b84d7 AV |
1008 | u64 bn, leaf_no; |
1009 | __be64 *lp; | |
cd915493 | 1010 | u32 index; |
d1b0cb93 | 1011 | int x; |
b3b94faa DT |
1012 | int error; |
1013 | ||
9a004508 | 1014 | index = name->hash >> (32 - dip->i_depth); |
c752666c | 1015 | error = get_leaf_nr(dip, index, &leaf_no); |
287980e4 | 1016 | if (error) |
c752666c | 1017 | return error; |
b3b94faa DT |
1018 | |
1019 | /* Get the old leaf block */ | |
b3b94faa DT |
1020 | error = get_leaf(dip, leaf_no, &obh); |
1021 | if (error) | |
e90deff5 | 1022 | return error; |
b3b94faa | 1023 | |
b3b94faa | 1024 | oleaf = (struct gfs2_leaf *)obh->b_data; |
9a004508 | 1025 | if (dip->i_depth == be16_to_cpu(oleaf->lf_depth)) { |
e90deff5 SW |
1026 | brelse(obh); |
1027 | return 1; /* can't split */ | |
1028 | } | |
1029 | ||
350a9b0a | 1030 | gfs2_trans_add_meta(dip->i_gl, obh); |
b3b94faa | 1031 | |
c752666c SW |
1032 | nleaf = new_leaf(inode, &nbh, be16_to_cpu(oleaf->lf_depth) + 1); |
1033 | if (!nleaf) { | |
1034 | brelse(obh); | |
1035 | return -ENOSPC; | |
1036 | } | |
1037 | bn = nbh->b_blocknr; | |
b3b94faa | 1038 | |
c752666c | 1039 | /* Compute the start and len of leaf pointers in the hash table. */ |
47a9a527 | 1040 | len = BIT(dip->i_depth - be16_to_cpu(oleaf->lf_depth)); |
b3b94faa DT |
1041 | half_len = len >> 1; |
1042 | if (!half_len) { | |
e54c78a2 | 1043 | fs_warn(GFS2_SB(inode), "i_depth %u lf_depth %u index %u\n", |
d77d1b58 | 1044 | dip->i_depth, be16_to_cpu(oleaf->lf_depth), index); |
b3b94faa DT |
1045 | gfs2_consist_inode(dip); |
1046 | error = -EIO; | |
1047 | goto fail_brelse; | |
1048 | } | |
1049 | ||
1050 | start = (index & ~(len - 1)); | |
1051 | ||
1052 | /* Change the pointers. | |
1053 | Don't bother distinguishing stuffed from non-stuffed. | |
1054 | This code is complicated enough already. */ | |
6da2ec56 | 1055 | lp = kmalloc_array(half_len, sizeof(__be64), GFP_NOFS); |
4244b52e DR |
1056 | if (!lp) { |
1057 | error = -ENOMEM; | |
1058 | goto fail_brelse; | |
1059 | } | |
1060 | ||
b3b94faa | 1061 | /* Change the pointers */ |
b3b94faa DT |
1062 | for (x = 0; x < half_len; x++) |
1063 | lp[x] = cpu_to_be64(bn); | |
1064 | ||
17d539f0 SW |
1065 | gfs2_dir_hash_inval(dip); |
1066 | ||
cd915493 SW |
1067 | error = gfs2_dir_write_data(dip, (char *)lp, start * sizeof(u64), |
1068 | half_len * sizeof(u64)); | |
1069 | if (error != half_len * sizeof(u64)) { | |
b3b94faa DT |
1070 | if (error >= 0) |
1071 | error = -EIO; | |
1072 | goto fail_lpfree; | |
1073 | } | |
1074 | ||
1075 | kfree(lp); | |
1076 | ||
1077 | /* Compute the divider */ | |
9a004508 | 1078 | divider = (start + half_len) << (32 - dip->i_depth); |
b3b94faa DT |
1079 | |
1080 | /* Copy the entries */ | |
1579343a | 1081 | dent = (struct gfs2_dirent *)(obh->b_data + sizeof(struct gfs2_leaf)); |
b3b94faa DT |
1082 | |
1083 | do { | |
1084 | next = dent; | |
1085 | if (dirent_next(dip, obh, &next)) | |
1086 | next = NULL; | |
1087 | ||
5e7d65cd | 1088 | if (!gfs2_dirent_sentinel(dent) && |
b3b94faa | 1089 | be32_to_cpu(dent->de_hash) < divider) { |
c752666c | 1090 | struct qstr str; |
34017472 | 1091 | void *ptr = ((char *)dent - obh->b_data) + nbh->b_data; |
c752666c SW |
1092 | str.name = (char*)(dent+1); |
1093 | str.len = be16_to_cpu(dent->de_name_len); | |
1094 | str.hash = be32_to_cpu(dent->de_hash); | |
34017472 | 1095 | new = gfs2_dirent_split_alloc(inode, nbh, &str, ptr); |
c752666c SW |
1096 | if (IS_ERR(new)) { |
1097 | error = PTR_ERR(new); | |
1098 | break; | |
1099 | } | |
b3b94faa DT |
1100 | |
1101 | new->de_inum = dent->de_inum; /* No endian worries */ | |
b3b94faa | 1102 | new->de_type = dent->de_type; /* No endian worries */ |
bb16b342 | 1103 | be16_add_cpu(&nleaf->lf_entries, 1); |
b3b94faa DT |
1104 | |
1105 | dirent_del(dip, obh, prev, dent); | |
1106 | ||
1107 | if (!oleaf->lf_entries) | |
1108 | gfs2_consist_inode(dip); | |
bb16b342 | 1109 | be16_add_cpu(&oleaf->lf_entries, -1); |
b3b94faa DT |
1110 | |
1111 | if (!prev) | |
1112 | prev = dent; | |
c752666c | 1113 | } else { |
b3b94faa | 1114 | prev = dent; |
c752666c | 1115 | } |
b3b94faa | 1116 | dent = next; |
c752666c | 1117 | } while (dent); |
b3b94faa | 1118 | |
c752666c | 1119 | oleaf->lf_depth = nleaf->lf_depth; |
b3b94faa DT |
1120 | |
1121 | error = gfs2_meta_inode_buffer(dip, &dibh); | |
feaa7bba | 1122 | if (!gfs2_assert_withdraw(GFS2_SB(&dip->i_inode), !error)) { |
350a9b0a | 1123 | gfs2_trans_add_meta(dip->i_gl, dibh); |
77658aad | 1124 | gfs2_add_inode_blocks(&dip->i_inode, 1); |
539e5d6b | 1125 | gfs2_dinode_out(dip, dibh->b_data); |
b3b94faa DT |
1126 | brelse(dibh); |
1127 | } | |
1128 | ||
1129 | brelse(obh); | |
1130 | brelse(nbh); | |
1131 | ||
1132 | return error; | |
1133 | ||
e90deff5 | 1134 | fail_lpfree: |
b3b94faa DT |
1135 | kfree(lp); |
1136 | ||
e90deff5 | 1137 | fail_brelse: |
b3b94faa | 1138 | brelse(obh); |
b3b94faa DT |
1139 | brelse(nbh); |
1140 | return error; | |
1141 | } | |
1142 | ||
1143 | /** | |
1144 | * dir_double_exhash - Double size of ExHash table | |
1145 | * @dip: The GFS2 dinode | |
1146 | * | |
1147 | * Returns: 0 on success, error code on failure | |
1148 | */ | |
1149 | ||
1150 | static int dir_double_exhash(struct gfs2_inode *dip) | |
1151 | { | |
b3b94faa | 1152 | struct buffer_head *dibh; |
cd915493 | 1153 | u32 hsize; |
17d539f0 SW |
1154 | u32 hsize_bytes; |
1155 | __be64 *hc; | |
1156 | __be64 *hc2, *h; | |
b3b94faa DT |
1157 | int x; |
1158 | int error = 0; | |
1159 | ||
47a9a527 | 1160 | hsize = BIT(dip->i_depth); |
17d539f0 | 1161 | hsize_bytes = hsize * sizeof(__be64); |
b3b94faa | 1162 | |
17d539f0 SW |
1163 | hc = gfs2_dir_get_hash_table(dip); |
1164 | if (IS_ERR(hc)) | |
1165 | return PTR_ERR(hc); | |
b3b94faa | 1166 | |
6da2ec56 | 1167 | hc2 = kmalloc_array(hsize_bytes, 2, GFP_NOFS | __GFP_NOWARN); |
e8830d88 | 1168 | if (hc2 == NULL) |
88dca4ca | 1169 | hc2 = __vmalloc(hsize_bytes * 2, GFP_NOFS); |
e8830d88 | 1170 | |
17d539f0 | 1171 | if (!hc2) |
4244b52e | 1172 | return -ENOMEM; |
b3b94faa | 1173 | |
512cbf02 | 1174 | h = hc2; |
17d539f0 SW |
1175 | error = gfs2_meta_inode_buffer(dip, &dibh); |
1176 | if (error) | |
1177 | goto out_kfree; | |
b3b94faa | 1178 | |
17d539f0 SW |
1179 | for (x = 0; x < hsize; x++) { |
1180 | *h++ = *hc; | |
1181 | *h++ = *hc; | |
1182 | hc++; | |
b3b94faa DT |
1183 | } |
1184 | ||
17d539f0 SW |
1185 | error = gfs2_dir_write_data(dip, (char *)hc2, 0, hsize_bytes * 2); |
1186 | if (error != (hsize_bytes * 2)) | |
1187 | goto fail; | |
b3b94faa | 1188 | |
17d539f0 SW |
1189 | gfs2_dir_hash_inval(dip); |
1190 | dip->i_hash_cache = hc2; | |
1191 | dip->i_depth++; | |
1192 | gfs2_dinode_out(dip, dibh->b_data); | |
1193 | brelse(dibh); | |
1194 | return 0; | |
b3b94faa | 1195 | |
a91ea69f | 1196 | fail: |
17d539f0 SW |
1197 | /* Replace original hash table & size */ |
1198 | gfs2_dir_write_data(dip, (char *)hc, 0, hsize_bytes); | |
1199 | i_size_write(&dip->i_inode, hsize_bytes); | |
1200 | gfs2_dinode_out(dip, dibh->b_data); | |
1201 | brelse(dibh); | |
1202 | out_kfree: | |
3cdcf63e | 1203 | kvfree(hc2); |
b3b94faa DT |
1204 | return error; |
1205 | } | |
1206 | ||
1207 | /** | |
1208 | * compare_dents - compare directory entries by hash value | |
1209 | * @a: first dent | |
1210 | * @b: second dent | |
1211 | * | |
1212 | * When comparing the hash entries of @a to @b: | |
1213 | * gt: returns 1 | |
1214 | * lt: returns -1 | |
1215 | * eq: returns 0 | |
1216 | */ | |
1217 | ||
1218 | static int compare_dents(const void *a, const void *b) | |
1219 | { | |
2bdbc5d7 | 1220 | const struct gfs2_dirent *dent_a, *dent_b; |
cd915493 | 1221 | u32 hash_a, hash_b; |
b3b94faa DT |
1222 | int ret = 0; |
1223 | ||
2bdbc5d7 | 1224 | dent_a = *(const struct gfs2_dirent **)a; |
471f3db2 | 1225 | hash_a = dent_a->de_cookie; |
b3b94faa | 1226 | |
2bdbc5d7 | 1227 | dent_b = *(const struct gfs2_dirent **)b; |
471f3db2 | 1228 | hash_b = dent_b->de_cookie; |
b3b94faa DT |
1229 | |
1230 | if (hash_a > hash_b) | |
1231 | ret = 1; | |
1232 | else if (hash_a < hash_b) | |
1233 | ret = -1; | |
1234 | else { | |
4dd651ad SW |
1235 | unsigned int len_a = be16_to_cpu(dent_a->de_name_len); |
1236 | unsigned int len_b = be16_to_cpu(dent_b->de_name_len); | |
b3b94faa DT |
1237 | |
1238 | if (len_a > len_b) | |
1239 | ret = 1; | |
1240 | else if (len_a < len_b) | |
1241 | ret = -1; | |
1242 | else | |
2bdbc5d7 | 1243 | ret = memcmp(dent_a + 1, dent_b + 1, len_a); |
b3b94faa DT |
1244 | } |
1245 | ||
1246 | return ret; | |
1247 | } | |
1248 | ||
1249 | /** | |
1250 | * do_filldir_main - read out directory entries | |
1251 | * @dip: The GFS2 inode | |
d81a8ef5 | 1252 | * @ctx: what to feed the entries to |
b3b94faa DT |
1253 | * @darr: an array of struct gfs2_dirent pointers to read |
1254 | * @entries: the number of entries in darr | |
1255 | * @copied: pointer to int that's non-zero if a entry has been copied out | |
1256 | * | |
1257 | * Jump through some hoops to make sure that if there are hash collsions, | |
1258 | * they are read out at the beginning of a buffer. We want to minimize | |
1259 | * the possibility that they will fall into different readdir buffers or | |
1260 | * that someone will want to seek to that location. | |
1261 | * | |
d81a8ef5 | 1262 | * Returns: errno, >0 if the actor tells you to stop |
b3b94faa DT |
1263 | */ |
1264 | ||
d81a8ef5 | 1265 | static int do_filldir_main(struct gfs2_inode *dip, struct dir_context *ctx, |
471f3db2 BM |
1266 | struct gfs2_dirent **darr, u32 entries, |
1267 | u32 sort_start, int *copied) | |
b3b94faa | 1268 | { |
71b86f56 | 1269 | const struct gfs2_dirent *dent, *dent_next; |
cd915493 | 1270 | u64 off, off_next; |
b3b94faa DT |
1271 | unsigned int x, y; |
1272 | int run = 0; | |
b3b94faa | 1273 | |
471f3db2 BM |
1274 | if (sort_start < entries) |
1275 | sort(&darr[sort_start], entries - sort_start, | |
1276 | sizeof(struct gfs2_dirent *), compare_dents, NULL); | |
b3b94faa DT |
1277 | |
1278 | dent_next = darr[0]; | |
471f3db2 | 1279 | off_next = dent_next->de_cookie; |
b3b94faa DT |
1280 | |
1281 | for (x = 0, y = 1; x < entries; x++, y++) { | |
1282 | dent = dent_next; | |
1283 | off = off_next; | |
1284 | ||
1285 | if (y < entries) { | |
1286 | dent_next = darr[y]; | |
471f3db2 | 1287 | off_next = dent_next->de_cookie; |
b3b94faa | 1288 | |
d81a8ef5 | 1289 | if (off < ctx->pos) |
b3b94faa | 1290 | continue; |
d81a8ef5 | 1291 | ctx->pos = off; |
b3b94faa DT |
1292 | |
1293 | if (off_next == off) { | |
1294 | if (*copied && !run) | |
1295 | return 1; | |
1296 | run = 1; | |
1297 | } else | |
1298 | run = 0; | |
1299 | } else { | |
d81a8ef5 | 1300 | if (off < ctx->pos) |
b3b94faa | 1301 | continue; |
d81a8ef5 | 1302 | ctx->pos = off; |
b3b94faa DT |
1303 | } |
1304 | ||
d81a8ef5 | 1305 | if (!dir_emit(ctx, (const char *)(dent + 1), |
4dd651ad | 1306 | be16_to_cpu(dent->de_name_len), |
d81a8ef5 AV |
1307 | be64_to_cpu(dent->de_inum.no_addr), |
1308 | be16_to_cpu(dent->de_type))) | |
b3b94faa DT |
1309 | return 1; |
1310 | ||
1311 | *copied = 1; | |
1312 | } | |
1313 | ||
d81a8ef5 | 1314 | /* Increment the ctx->pos by one, so the next time we come into the |
b3b94faa DT |
1315 | do_filldir fxn, we get the next entry instead of the last one in the |
1316 | current leaf */ | |
1317 | ||
d81a8ef5 | 1318 | ctx->pos++; |
b3b94faa DT |
1319 | |
1320 | return 0; | |
1321 | } | |
1322 | ||
d2a97a4e SW |
1323 | static void *gfs2_alloc_sort_buffer(unsigned size) |
1324 | { | |
1325 | void *ptr = NULL; | |
1326 | ||
1327 | if (size < KMALLOC_MAX_SIZE) | |
1328 | ptr = kmalloc(size, GFP_NOFS | __GFP_NOWARN); | |
1329 | if (!ptr) | |
88dca4ca | 1330 | ptr = __vmalloc(size, GFP_NOFS); |
d2a97a4e SW |
1331 | return ptr; |
1332 | } | |
1333 | ||
471f3db2 BM |
1334 | |
1335 | static int gfs2_set_cookies(struct gfs2_sbd *sdp, struct buffer_head *bh, | |
1336 | unsigned leaf_nr, struct gfs2_dirent **darr, | |
1337 | unsigned entries) | |
1338 | { | |
1339 | int sort_id = -1; | |
1340 | int i; | |
1341 | ||
1342 | for (i = 0; i < entries; i++) { | |
1343 | unsigned offset; | |
1344 | ||
1345 | darr[i]->de_cookie = be32_to_cpu(darr[i]->de_hash); | |
1346 | darr[i]->de_cookie = gfs2_disk_hash2offset(darr[i]->de_cookie); | |
1347 | ||
1348 | if (!sdp->sd_args.ar_loccookie) | |
1349 | continue; | |
1350 | offset = (char *)(darr[i]) - | |
e54c78a2 | 1351 | (bh->b_data + gfs2_dirent_offset(sdp, bh->b_data)); |
471f3db2 BM |
1352 | offset /= GFS2_MIN_DIRENT_SIZE; |
1353 | offset += leaf_nr * sdp->sd_max_dents_per_leaf; | |
1354 | if (offset >= GFS2_USE_HASH_FLAG || | |
1355 | leaf_nr >= GFS2_USE_HASH_FLAG) { | |
1356 | darr[i]->de_cookie |= GFS2_USE_HASH_FLAG; | |
1357 | if (sort_id < 0) | |
1358 | sort_id = i; | |
1359 | continue; | |
1360 | } | |
1361 | darr[i]->de_cookie &= GFS2_HASH_INDEX_MASK; | |
1362 | darr[i]->de_cookie |= offset; | |
1363 | } | |
1364 | return sort_id; | |
1365 | } | |
1366 | ||
1367 | ||
d81a8ef5 AV |
1368 | static int gfs2_dir_read_leaf(struct inode *inode, struct dir_context *ctx, |
1369 | int *copied, unsigned *depth, | |
3699e3a4 | 1370 | u64 leaf_no) |
b3b94faa | 1371 | { |
feaa7bba | 1372 | struct gfs2_inode *ip = GFS2_I(inode); |
bdd19a22 | 1373 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
71b86f56 SW |
1374 | struct buffer_head *bh; |
1375 | struct gfs2_leaf *lf; | |
bdd19a22 | 1376 | unsigned entries = 0, entries2 = 0; |
471f3db2 BM |
1377 | unsigned leaves = 0, leaf = 0, offset, sort_offset; |
1378 | struct gfs2_dirent **darr, *dent; | |
71b86f56 SW |
1379 | struct dirent_gather g; |
1380 | struct buffer_head **larr; | |
471f3db2 | 1381 | int error, i, need_sort = 0, sort_id; |
71b86f56 | 1382 | u64 lfn = leaf_no; |
b3b94faa | 1383 | |
b3b94faa | 1384 | do { |
71b86f56 | 1385 | error = get_leaf(ip, lfn, &bh); |
b3b94faa | 1386 | if (error) |
71b86f56 SW |
1387 | goto out; |
1388 | lf = (struct gfs2_leaf *)bh->b_data; | |
1389 | if (leaves == 0) | |
1390 | *depth = be16_to_cpu(lf->lf_depth); | |
1391 | entries += be16_to_cpu(lf->lf_entries); | |
1392 | leaves++; | |
1393 | lfn = be64_to_cpu(lf->lf_next); | |
1394 | brelse(bh); | |
1395 | } while(lfn); | |
b3b94faa | 1396 | |
471f3db2 BM |
1397 | if (*depth < GFS2_DIR_MAX_DEPTH || !sdp->sd_args.ar_loccookie) { |
1398 | need_sort = 1; | |
1399 | sort_offset = 0; | |
1400 | } | |
1401 | ||
b3b94faa DT |
1402 | if (!entries) |
1403 | return 0; | |
1404 | ||
71b86f56 | 1405 | error = -ENOMEM; |
bdd19a22 SW |
1406 | /* |
1407 | * The extra 99 entries are not normally used, but are a buffer | |
1408 | * zone in case the number of entries in the leaf is corrupt. | |
1409 | * 99 is the maximum number of entries that can fit in a single | |
1410 | * leaf block. | |
1411 | */ | |
d2a97a4e | 1412 | larr = gfs2_alloc_sort_buffer((leaves + entries + 99) * sizeof(void *)); |
71b86f56 SW |
1413 | if (!larr) |
1414 | goto out; | |
471f3db2 BM |
1415 | darr = (struct gfs2_dirent **)(larr + leaves); |
1416 | g.pdent = (const struct gfs2_dirent **)darr; | |
71b86f56 SW |
1417 | g.offset = 0; |
1418 | lfn = leaf_no; | |
b3b94faa | 1419 | |
71b86f56 SW |
1420 | do { |
1421 | error = get_leaf(ip, lfn, &bh); | |
b3b94faa | 1422 | if (error) |
d2a97a4e | 1423 | goto out_free; |
71b86f56 SW |
1424 | lf = (struct gfs2_leaf *)bh->b_data; |
1425 | lfn = be64_to_cpu(lf->lf_next); | |
1426 | if (lf->lf_entries) { | |
471f3db2 | 1427 | offset = g.offset; |
bdd19a22 | 1428 | entries2 += be16_to_cpu(lf->lf_entries); |
71b86f56 SW |
1429 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, |
1430 | gfs2_dirent_gather, NULL, &g); | |
1431 | error = PTR_ERR(dent); | |
bdd19a22 | 1432 | if (IS_ERR(dent)) |
d2a97a4e | 1433 | goto out_free; |
bdd19a22 | 1434 | if (entries2 != g.offset) { |
f391a4ea | 1435 | fs_warn(sdp, "Number of entries corrupt in dir " |
1436 | "leaf %llu, entries2 (%u) != " | |
1437 | "g.offset (%u)\n", | |
1438 | (unsigned long long)bh->b_blocknr, | |
1439 | entries2, g.offset); | |
d87d62b7 | 1440 | gfs2_consist_inode(ip); |
bdd19a22 | 1441 | error = -EIO; |
d2a97a4e | 1442 | goto out_free; |
71b86f56 SW |
1443 | } |
1444 | error = 0; | |
471f3db2 BM |
1445 | sort_id = gfs2_set_cookies(sdp, bh, leaf, &darr[offset], |
1446 | be16_to_cpu(lf->lf_entries)); | |
1447 | if (!need_sort && sort_id >= 0) { | |
1448 | need_sort = 1; | |
1449 | sort_offset = offset + sort_id; | |
1450 | } | |
71b86f56 | 1451 | larr[leaf++] = bh; |
b3b94faa | 1452 | } else { |
471f3db2 | 1453 | larr[leaf++] = NULL; |
71b86f56 | 1454 | brelse(bh); |
b3b94faa | 1455 | } |
71b86f56 | 1456 | } while(lfn); |
b3b94faa | 1457 | |
bdd19a22 | 1458 | BUG_ON(entries2 != entries); |
471f3db2 BM |
1459 | error = do_filldir_main(ip, ctx, darr, entries, need_sort ? |
1460 | sort_offset : entries, copied); | |
d2a97a4e | 1461 | out_free: |
71b86f56 | 1462 | for(i = 0; i < leaf; i++) |
bccaef90 | 1463 | brelse(larr[i]); |
3cdcf63e | 1464 | kvfree(larr); |
71b86f56 | 1465 | out: |
b3b94faa DT |
1466 | return error; |
1467 | } | |
1468 | ||
79c4c379 SW |
1469 | /** |
1470 | * gfs2_dir_readahead - Issue read-ahead requests for leaf blocks. | |
dfe4d34b BP |
1471 | * |
1472 | * Note: we can't calculate each index like dir_e_read can because we don't | |
1473 | * have the leaf, and therefore we don't have the depth, and therefore we | |
1474 | * don't have the length. So we have to just read enough ahead to make up | |
79c4c379 SW |
1475 | * for the loss of information. |
1476 | */ | |
dfe4d34b BP |
1477 | static void gfs2_dir_readahead(struct inode *inode, unsigned hsize, u32 index, |
1478 | struct file_ra_state *f_ra) | |
1479 | { | |
1480 | struct gfs2_inode *ip = GFS2_I(inode); | |
1481 | struct gfs2_glock *gl = ip->i_gl; | |
1482 | struct buffer_head *bh; | |
1483 | u64 blocknr = 0, last; | |
1484 | unsigned count; | |
1485 | ||
1486 | /* First check if we've already read-ahead for the whole range. */ | |
79c4c379 | 1487 | if (index + MAX_RA_BLOCKS < f_ra->start) |
dfe4d34b BP |
1488 | return; |
1489 | ||
1490 | f_ra->start = max((pgoff_t)index, f_ra->start); | |
1491 | for (count = 0; count < MAX_RA_BLOCKS; count++) { | |
1492 | if (f_ra->start >= hsize) /* if exceeded the hash table */ | |
1493 | break; | |
1494 | ||
1495 | last = blocknr; | |
1496 | blocknr = be64_to_cpu(ip->i_hash_cache[f_ra->start]); | |
1497 | f_ra->start++; | |
1498 | if (blocknr == last) | |
1499 | continue; | |
1500 | ||
1501 | bh = gfs2_getbuf(gl, blocknr, 1); | |
1502 | if (trylock_buffer(bh)) { | |
1503 | if (buffer_uptodate(bh)) { | |
1504 | unlock_buffer(bh); | |
1505 | brelse(bh); | |
1506 | continue; | |
1507 | } | |
1508 | bh->b_end_io = end_buffer_read_sync; | |
e477b24b CL |
1509 | submit_bh(REQ_OP_READ, |
1510 | REQ_RAHEAD | REQ_META | REQ_PRIO, | |
1511 | bh); | |
dfe4d34b BP |
1512 | continue; |
1513 | } | |
1514 | brelse(bh); | |
1515 | } | |
1516 | } | |
17d539f0 | 1517 | |
b3b94faa | 1518 | /** |
c752666c SW |
1519 | * dir_e_read - Reads the entries from a directory into a filldir buffer |
1520 | * @dip: dinode pointer | |
d81a8ef5 | 1521 | * @ctx: actor to feed the entries to |
b3b94faa | 1522 | * |
c752666c | 1523 | * Returns: errno |
b3b94faa DT |
1524 | */ |
1525 | ||
d81a8ef5 AV |
1526 | static int dir_e_read(struct inode *inode, struct dir_context *ctx, |
1527 | struct file_ra_state *f_ra) | |
b3b94faa | 1528 | { |
feaa7bba | 1529 | struct gfs2_inode *dip = GFS2_I(inode); |
cd915493 | 1530 | u32 hsize, len = 0; |
cd915493 | 1531 | u32 hash, index; |
b44b84d7 | 1532 | __be64 *lp; |
c752666c SW |
1533 | int copied = 0; |
1534 | int error = 0; | |
4da3c646 | 1535 | unsigned depth = 0; |
b3b94faa | 1536 | |
47a9a527 | 1537 | hsize = BIT(dip->i_depth); |
d81a8ef5 | 1538 | hash = gfs2_dir_offset2hash(ctx->pos); |
9a004508 | 1539 | index = hash >> (32 - dip->i_depth); |
b3b94faa | 1540 | |
79c4c379 | 1541 | if (dip->i_hash_cache == NULL) |
dfe4d34b | 1542 | f_ra->start = 0; |
17d539f0 SW |
1543 | lp = gfs2_dir_get_hash_table(dip); |
1544 | if (IS_ERR(lp)) | |
1545 | return PTR_ERR(lp); | |
b3b94faa | 1546 | |
dfe4d34b BP |
1547 | gfs2_dir_readahead(inode, hsize, index, f_ra); |
1548 | ||
b3b94faa | 1549 | while (index < hsize) { |
d81a8ef5 | 1550 | error = gfs2_dir_read_leaf(inode, ctx, |
71b86f56 | 1551 | &copied, &depth, |
17d539f0 | 1552 | be64_to_cpu(lp[index])); |
b3b94faa | 1553 | if (error) |
71b86f56 | 1554 | break; |
b3b94faa | 1555 | |
47a9a527 | 1556 | len = BIT(dip->i_depth - depth); |
b3b94faa DT |
1557 | index = (index & ~(len - 1)) + len; |
1558 | } | |
1559 | ||
71b86f56 SW |
1560 | if (error > 0) |
1561 | error = 0; | |
b3b94faa DT |
1562 | return error; |
1563 | } | |
1564 | ||
d81a8ef5 AV |
1565 | int gfs2_dir_read(struct inode *inode, struct dir_context *ctx, |
1566 | struct file_ra_state *f_ra) | |
b3b94faa | 1567 | { |
feaa7bba | 1568 | struct gfs2_inode *dip = GFS2_I(inode); |
bdd19a22 | 1569 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
71b86f56 | 1570 | struct dirent_gather g; |
471f3db2 | 1571 | struct gfs2_dirent **darr, *dent; |
b3b94faa DT |
1572 | struct buffer_head *dibh; |
1573 | int copied = 0; | |
1574 | int error; | |
1575 | ||
ad6203f2 | 1576 | if (!dip->i_entries) |
71b86f56 SW |
1577 | return 0; |
1578 | ||
383f01fb | 1579 | if (dip->i_diskflags & GFS2_DIF_EXHASH) |
d81a8ef5 | 1580 | return dir_e_read(inode, ctx, f_ra); |
71b86f56 | 1581 | |
b3b94faa DT |
1582 | if (!gfs2_is_stuffed(dip)) { |
1583 | gfs2_consist_inode(dip); | |
1584 | return -EIO; | |
1585 | } | |
1586 | ||
b3b94faa DT |
1587 | error = gfs2_meta_inode_buffer(dip, &dibh); |
1588 | if (error) | |
1589 | return error; | |
1590 | ||
71b86f56 | 1591 | error = -ENOMEM; |
bdd19a22 | 1592 | /* 96 is max number of dirents which can be stuffed into an inode */ |
6da2ec56 | 1593 | darr = kmalloc_array(96, sizeof(struct gfs2_dirent *), GFP_NOFS); |
71b86f56 | 1594 | if (darr) { |
471f3db2 | 1595 | g.pdent = (const struct gfs2_dirent **)darr; |
71b86f56 SW |
1596 | g.offset = 0; |
1597 | dent = gfs2_dirent_scan(inode, dibh->b_data, dibh->b_size, | |
1598 | gfs2_dirent_gather, NULL, &g); | |
1599 | if (IS_ERR(dent)) { | |
1600 | error = PTR_ERR(dent); | |
1601 | goto out; | |
1602 | } | |
ad6203f2 | 1603 | if (dip->i_entries != g.offset) { |
bdd19a22 | 1604 | fs_warn(sdp, "Number of entries corrupt in dir %llu, " |
ad6203f2 | 1605 | "ip->i_entries (%u) != g.offset (%u)\n", |
dbb7cae2 | 1606 | (unsigned long long)dip->i_no_addr, |
ad6203f2 | 1607 | dip->i_entries, |
bdd19a22 | 1608 | g.offset); |
d87d62b7 | 1609 | gfs2_consist_inode(dip); |
bdd19a22 SW |
1610 | error = -EIO; |
1611 | goto out; | |
1612 | } | |
471f3db2 | 1613 | gfs2_set_cookies(sdp, dibh, 0, darr, dip->i_entries); |
d81a8ef5 | 1614 | error = do_filldir_main(dip, ctx, darr, |
471f3db2 | 1615 | dip->i_entries, 0, &copied); |
71b86f56 SW |
1616 | out: |
1617 | kfree(darr); | |
1618 | } | |
1619 | ||
b3b94faa DT |
1620 | if (error > 0) |
1621 | error = 0; | |
1622 | ||
1623 | brelse(dibh); | |
1624 | ||
1625 | return error; | |
1626 | } | |
1627 | ||
b3b94faa DT |
1628 | /** |
1629 | * gfs2_dir_search - Search a directory | |
5a00f3cc SW |
1630 | * @dip: The GFS2 dir inode |
1631 | * @name: The name we are looking up | |
1632 | * @fail_on_exist: Fail if the name exists rather than looking it up | |
b3b94faa DT |
1633 | * |
1634 | * This routine searches a directory for a file or another directory. | |
1635 | * Assumes a glock is held on dip. | |
1636 | * | |
1637 | * Returns: errno | |
1638 | */ | |
1639 | ||
5a00f3cc SW |
1640 | struct inode *gfs2_dir_search(struct inode *dir, const struct qstr *name, |
1641 | bool fail_on_exist) | |
b3b94faa | 1642 | { |
c752666c SW |
1643 | struct buffer_head *bh; |
1644 | struct gfs2_dirent *dent; | |
5a00f3cc SW |
1645 | u64 addr, formal_ino; |
1646 | u16 dtype; | |
dbb7cae2 SW |
1647 | |
1648 | dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh); | |
1649 | if (dent) { | |
c8d57703 AG |
1650 | struct inode *inode; |
1651 | u16 rahead; | |
1652 | ||
dbb7cae2 | 1653 | if (IS_ERR(dent)) |
e231c2ee | 1654 | return ERR_CAST(dent); |
5a00f3cc | 1655 | dtype = be16_to_cpu(dent->de_type); |
c8d57703 | 1656 | rahead = be16_to_cpu(dent->de_rahead); |
5a00f3cc SW |
1657 | addr = be64_to_cpu(dent->de_inum.no_addr); |
1658 | formal_ino = be64_to_cpu(dent->de_inum.no_formal_ino); | |
dbb7cae2 | 1659 | brelse(bh); |
5a00f3cc SW |
1660 | if (fail_on_exist) |
1661 | return ERR_PTR(-EEXIST); | |
3ce37b2c AG |
1662 | inode = gfs2_inode_lookup(dir->i_sb, dtype, addr, formal_ino, |
1663 | GFS2_BLKST_FREE /* ignore */); | |
c8d57703 AG |
1664 | if (!IS_ERR(inode)) |
1665 | GFS2_I(inode)->i_rahead = rahead; | |
1666 | return inode; | |
dbb7cae2 SW |
1667 | } |
1668 | return ERR_PTR(-ENOENT); | |
1669 | } | |
1670 | ||
1671 | int gfs2_dir_check(struct inode *dir, const struct qstr *name, | |
1672 | const struct gfs2_inode *ip) | |
1673 | { | |
1674 | struct buffer_head *bh; | |
1675 | struct gfs2_dirent *dent; | |
1676 | int ret = -ENOENT; | |
c752666c SW |
1677 | |
1678 | dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh); | |
1679 | if (dent) { | |
1680 | if (IS_ERR(dent)) | |
1681 | return PTR_ERR(dent); | |
dbb7cae2 SW |
1682 | if (ip) { |
1683 | if (be64_to_cpu(dent->de_inum.no_addr) != ip->i_no_addr) | |
1684 | goto out; | |
1685 | if (be64_to_cpu(dent->de_inum.no_formal_ino) != | |
1686 | ip->i_no_formal_ino) | |
1687 | goto out; | |
1688 | if (unlikely(IF2DT(ip->i_inode.i_mode) != | |
1689 | be16_to_cpu(dent->de_type))) { | |
1690 | gfs2_consist_inode(GFS2_I(dir)); | |
1691 | ret = -EIO; | |
1692 | goto out; | |
1693 | } | |
1694 | } | |
1695 | ret = 0; | |
1696 | out: | |
c752666c | 1697 | brelse(bh); |
c752666c | 1698 | } |
dbb7cae2 | 1699 | return ret; |
c752666c SW |
1700 | } |
1701 | ||
01bcb0de SW |
1702 | /** |
1703 | * dir_new_leaf - Add a new leaf onto hash chain | |
1704 | * @inode: The directory | |
1705 | * @name: The name we are adding | |
1706 | * | |
1707 | * This adds a new dir leaf onto an existing leaf when there is not | |
1708 | * enough space to add a new dir entry. This is a last resort after | |
1709 | * we've expanded the hash table to max size and also split existing | |
1710 | * leaf blocks, so it will only occur for very large directories. | |
1711 | * | |
1712 | * The dist parameter is set to 1 for leaf blocks directly attached | |
1713 | * to the hash table, 2 for one layer of indirection, 3 for two layers | |
1714 | * etc. We are thus able to tell the difference between an old leaf | |
1715 | * with dist set to zero (i.e. "don't know") and a new one where we | |
1716 | * set this information for debug/fsck purposes. | |
1717 | * | |
1718 | * Returns: 0 on success, or -ve on error | |
1719 | */ | |
1720 | ||
c752666c SW |
1721 | static int dir_new_leaf(struct inode *inode, const struct qstr *name) |
1722 | { | |
1723 | struct buffer_head *bh, *obh; | |
feaa7bba | 1724 | struct gfs2_inode *ip = GFS2_I(inode); |
c752666c | 1725 | struct gfs2_leaf *leaf, *oleaf; |
01bcb0de | 1726 | u32 dist = 1; |
b3b94faa | 1727 | int error; |
c752666c SW |
1728 | u32 index; |
1729 | u64 bn; | |
b3b94faa | 1730 | |
9a004508 | 1731 | index = name->hash >> (32 - ip->i_depth); |
c752666c SW |
1732 | error = get_first_leaf(ip, index, &obh); |
1733 | if (error) | |
1734 | return error; | |
1735 | do { | |
01bcb0de | 1736 | dist++; |
c752666c SW |
1737 | oleaf = (struct gfs2_leaf *)obh->b_data; |
1738 | bn = be64_to_cpu(oleaf->lf_next); | |
1739 | if (!bn) | |
1740 | break; | |
1741 | brelse(obh); | |
1742 | error = get_leaf(ip, bn, &obh); | |
1743 | if (error) | |
1744 | return error; | |
1745 | } while(1); | |
b3b94faa | 1746 | |
350a9b0a | 1747 | gfs2_trans_add_meta(ip->i_gl, obh); |
c752666c SW |
1748 | |
1749 | leaf = new_leaf(inode, &bh, be16_to_cpu(oleaf->lf_depth)); | |
1750 | if (!leaf) { | |
1751 | brelse(obh); | |
1752 | return -ENOSPC; | |
1753 | } | |
01bcb0de | 1754 | leaf->lf_dist = cpu_to_be32(dist); |
4d8012b6 | 1755 | oleaf->lf_next = cpu_to_be64(bh->b_blocknr); |
c752666c SW |
1756 | brelse(bh); |
1757 | brelse(obh); | |
1758 | ||
1759 | error = gfs2_meta_inode_buffer(ip, &bh); | |
1760 | if (error) | |
1761 | return error; | |
350a9b0a | 1762 | gfs2_trans_add_meta(ip->i_gl, bh); |
77658aad | 1763 | gfs2_add_inode_blocks(&ip->i_inode, 1); |
539e5d6b | 1764 | gfs2_dinode_out(ip, bh->b_data); |
c752666c SW |
1765 | brelse(bh); |
1766 | return 0; | |
b3b94faa DT |
1767 | } |
1768 | ||
44aaada9 SW |
1769 | static u16 gfs2_inode_ra_len(const struct gfs2_inode *ip) |
1770 | { | |
1771 | u64 where = ip->i_no_addr + 1; | |
1772 | if (ip->i_eattr == where) | |
1773 | return 1; | |
1774 | return 0; | |
1775 | } | |
1776 | ||
b3b94faa DT |
1777 | /** |
1778 | * gfs2_dir_add - Add new filename into directory | |
2b47dad8 SW |
1779 | * @inode: The directory inode |
1780 | * @name: The new name | |
1781 | * @nip: The GFS2 inode to be linked in to the directory | |
1782 | * @da: The directory addition info | |
1783 | * | |
1784 | * If the call to gfs2_diradd_alloc_required resulted in there being | |
1785 | * no need to allocate any new directory blocks, then it will contain | |
1786 | * a pointer to the directory entry and the bh in which it resides. We | |
1787 | * can use that without having to repeat the search. If there was no | |
1788 | * free space, then we must now create more space. | |
b3b94faa DT |
1789 | * |
1790 | * Returns: 0 on success, error code on failure | |
1791 | */ | |
1792 | ||
c752666c | 1793 | int gfs2_dir_add(struct inode *inode, const struct qstr *name, |
2b47dad8 | 1794 | const struct gfs2_inode *nip, struct gfs2_diradd *da) |
b3b94faa | 1795 | { |
feaa7bba | 1796 | struct gfs2_inode *ip = GFS2_I(inode); |
2b47dad8 SW |
1797 | struct buffer_head *bh = da->bh; |
1798 | struct gfs2_dirent *dent = da->dent; | |
95582b00 | 1799 | struct timespec64 tv; |
c752666c | 1800 | struct gfs2_leaf *leaf; |
b3b94faa DT |
1801 | int error; |
1802 | ||
c752666c | 1803 | while(1) { |
2b47dad8 SW |
1804 | if (da->bh == NULL) { |
1805 | dent = gfs2_dirent_search(inode, name, | |
1806 | gfs2_dirent_find_space, &bh); | |
1807 | } | |
c752666c SW |
1808 | if (dent) { |
1809 | if (IS_ERR(dent)) | |
1810 | return PTR_ERR(dent); | |
1811 | dent = gfs2_init_dirent(inode, dent, name, bh); | |
dbb7cae2 | 1812 | gfs2_inum_out(nip, dent); |
3d6ecb7d | 1813 | dent->de_type = cpu_to_be16(IF2DT(nip->i_inode.i_mode)); |
44aaada9 | 1814 | dent->de_rahead = cpu_to_be16(gfs2_inode_ra_len(nip)); |
078cd827 | 1815 | tv = current_time(&ip->i_inode); |
383f01fb | 1816 | if (ip->i_diskflags & GFS2_DIF_EXHASH) { |
c752666c | 1817 | leaf = (struct gfs2_leaf *)bh->b_data; |
bb16b342 | 1818 | be16_add_cpu(&leaf->lf_entries, 1); |
01bcb0de SW |
1819 | leaf->lf_nsec = cpu_to_be32(tv.tv_nsec); |
1820 | leaf->lf_sec = cpu_to_be64(tv.tv_sec); | |
c752666c | 1821 | } |
2b47dad8 SW |
1822 | da->dent = NULL; |
1823 | da->bh = NULL; | |
c752666c | 1824 | brelse(bh); |
ad6203f2 | 1825 | ip->i_entries++; |
01bcb0de | 1826 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = tv; |
3d6ecb7d SW |
1827 | if (S_ISDIR(nip->i_inode.i_mode)) |
1828 | inc_nlink(&ip->i_inode); | |
343cd8f0 | 1829 | mark_inode_dirty(inode); |
c752666c SW |
1830 | error = 0; |
1831 | break; | |
1832 | } | |
383f01fb | 1833 | if (!(ip->i_diskflags & GFS2_DIF_EXHASH)) { |
c752666c SW |
1834 | error = dir_make_exhash(inode); |
1835 | if (error) | |
1836 | break; | |
1837 | continue; | |
1838 | } | |
1839 | error = dir_split_leaf(inode, name); | |
1840 | if (error == 0) | |
1841 | continue; | |
e90deff5 | 1842 | if (error < 0) |
c752666c | 1843 | break; |
9a004508 | 1844 | if (ip->i_depth < GFS2_DIR_MAX_DEPTH) { |
c752666c SW |
1845 | error = dir_double_exhash(ip); |
1846 | if (error) | |
1847 | break; | |
1848 | error = dir_split_leaf(inode, name); | |
e90deff5 | 1849 | if (error < 0) |
c752666c | 1850 | break; |
e90deff5 SW |
1851 | if (error == 0) |
1852 | continue; | |
c752666c SW |
1853 | } |
1854 | error = dir_new_leaf(inode, name); | |
1855 | if (!error) | |
1856 | continue; | |
1857 | error = -ENOSPC; | |
1858 | break; | |
1859 | } | |
b3b94faa DT |
1860 | return error; |
1861 | } | |
1862 | ||
c752666c | 1863 | |
b3b94faa DT |
1864 | /** |
1865 | * gfs2_dir_del - Delete a directory entry | |
1866 | * @dip: The GFS2 inode | |
1867 | * @filename: The filename | |
1868 | * | |
1869 | * Returns: 0 on success, error code on failure | |
1870 | */ | |
1871 | ||
855d23ce | 1872 | int gfs2_dir_del(struct gfs2_inode *dip, const struct dentry *dentry) |
b3b94faa | 1873 | { |
855d23ce | 1874 | const struct qstr *name = &dentry->d_name; |
c752666c SW |
1875 | struct gfs2_dirent *dent, *prev = NULL; |
1876 | struct buffer_head *bh; | |
95582b00 | 1877 | struct timespec64 tv = current_time(&dip->i_inode); |
b3b94faa | 1878 | |
c752666c SW |
1879 | /* Returns _either_ the entry (if its first in block) or the |
1880 | previous entry otherwise */ | |
feaa7bba | 1881 | dent = gfs2_dirent_search(&dip->i_inode, name, gfs2_dirent_prev, &bh); |
c752666c SW |
1882 | if (!dent) { |
1883 | gfs2_consist_inode(dip); | |
1884 | return -EIO; | |
1885 | } | |
1886 | if (IS_ERR(dent)) { | |
1887 | gfs2_consist_inode(dip); | |
1888 | return PTR_ERR(dent); | |
1889 | } | |
1890 | /* If not first in block, adjust pointers accordingly */ | |
71b86f56 | 1891 | if (gfs2_dirent_find(dent, name, NULL) == 0) { |
c752666c SW |
1892 | prev = dent; |
1893 | dent = (struct gfs2_dirent *)((char *)dent + be16_to_cpu(prev->de_rec_len)); | |
1894 | } | |
1895 | ||
1896 | dirent_del(dip, bh, prev, dent); | |
383f01fb | 1897 | if (dip->i_diskflags & GFS2_DIF_EXHASH) { |
c752666c SW |
1898 | struct gfs2_leaf *leaf = (struct gfs2_leaf *)bh->b_data; |
1899 | u16 entries = be16_to_cpu(leaf->lf_entries); | |
1900 | if (!entries) | |
1901 | gfs2_consist_inode(dip); | |
1902 | leaf->lf_entries = cpu_to_be16(--entries); | |
01bcb0de SW |
1903 | leaf->lf_nsec = cpu_to_be32(tv.tv_nsec); |
1904 | leaf->lf_sec = cpu_to_be64(tv.tv_sec); | |
c752666c | 1905 | } |
ed386507 | 1906 | brelse(bh); |
c752666c | 1907 | |
ad6203f2 | 1908 | if (!dip->i_entries) |
c752666c | 1909 | gfs2_consist_inode(dip); |
ad6203f2 | 1910 | dip->i_entries--; |
01bcb0de | 1911 | dip->i_inode.i_mtime = dip->i_inode.i_ctime = tv; |
e36cb0b8 | 1912 | if (d_is_dir(dentry)) |
855d23ce | 1913 | drop_nlink(&dip->i_inode); |
feaa7bba | 1914 | mark_inode_dirty(&dip->i_inode); |
b3b94faa | 1915 | |
ab9bbda0 | 1916 | return 0; |
b3b94faa DT |
1917 | } |
1918 | ||
b3b94faa DT |
1919 | /** |
1920 | * gfs2_dir_mvino - Change inode number of directory entry | |
1921 | * @dip: The GFS2 inode | |
1922 | * @filename: | |
1923 | * @new_inode: | |
1924 | * | |
1925 | * This routine changes the inode number of a directory entry. It's used | |
1926 | * by rename to change ".." when a directory is moved. | |
1927 | * Assumes a glock is held on dvp. | |
1928 | * | |
1929 | * Returns: errno | |
1930 | */ | |
1931 | ||
c752666c | 1932 | int gfs2_dir_mvino(struct gfs2_inode *dip, const struct qstr *filename, |
dbb7cae2 | 1933 | const struct gfs2_inode *nip, unsigned int new_type) |
b3b94faa | 1934 | { |
c752666c SW |
1935 | struct buffer_head *bh; |
1936 | struct gfs2_dirent *dent; | |
b3b94faa | 1937 | |
feaa7bba | 1938 | dent = gfs2_dirent_search(&dip->i_inode, filename, gfs2_dirent_find, &bh); |
c752666c SW |
1939 | if (!dent) { |
1940 | gfs2_consist_inode(dip); | |
1941 | return -EIO; | |
1942 | } | |
1943 | if (IS_ERR(dent)) | |
1944 | return PTR_ERR(dent); | |
b3b94faa | 1945 | |
350a9b0a | 1946 | gfs2_trans_add_meta(dip->i_gl, bh); |
dbb7cae2 | 1947 | gfs2_inum_out(nip, dent); |
c752666c | 1948 | dent->de_type = cpu_to_be16(new_type); |
83998ccd | 1949 | brelse(bh); |
c752666c | 1950 | |
078cd827 | 1951 | dip->i_inode.i_mtime = dip->i_inode.i_ctime = current_time(&dip->i_inode); |
83998ccd | 1952 | mark_inode_dirty_sync(&dip->i_inode); |
c752666c | 1953 | return 0; |
b3b94faa DT |
1954 | } |
1955 | ||
b3b94faa DT |
1956 | /** |
1957 | * leaf_dealloc - Deallocate a directory leaf | |
1958 | * @dip: the directory | |
1959 | * @index: the hash table offset in the directory | |
1960 | * @len: the number of pointers to this leaf | |
1961 | * @leaf_no: the leaf number | |
ec038c82 | 1962 | * @leaf_bh: buffer_head for the starting leaf |
d24a7a43 | 1963 | * last_dealloc: 1 if this is the final dealloc for the leaf, else 0 |
b3b94faa DT |
1964 | * |
1965 | * Returns: errno | |
1966 | */ | |
1967 | ||
cd915493 | 1968 | static int leaf_dealloc(struct gfs2_inode *dip, u32 index, u32 len, |
ec038c82 BP |
1969 | u64 leaf_no, struct buffer_head *leaf_bh, |
1970 | int last_dealloc) | |
b3b94faa | 1971 | { |
feaa7bba | 1972 | struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); |
c752666c | 1973 | struct gfs2_leaf *tmp_leaf; |
b3b94faa DT |
1974 | struct gfs2_rgrp_list rlist; |
1975 | struct buffer_head *bh, *dibh; | |
cd915493 | 1976 | u64 blk, nblk; |
b3b94faa DT |
1977 | unsigned int rg_blocks = 0, l_blocks = 0; |
1978 | char *ht; | |
cd915493 | 1979 | unsigned int x, size = len * sizeof(u64); |
b3b94faa DT |
1980 | int error; |
1981 | ||
5e2f7d61 BP |
1982 | error = gfs2_rindex_update(sdp); |
1983 | if (error) | |
1984 | return error; | |
1985 | ||
b3b94faa DT |
1986 | memset(&rlist, 0, sizeof(struct gfs2_rgrp_list)); |
1987 | ||
8be04b93 | 1988 | ht = kzalloc(size, GFP_NOFS | __GFP_NOWARN); |
e8830d88 | 1989 | if (ht == NULL) |
88dca4ca | 1990 | ht = __vmalloc(size, GFP_NOFS | __GFP_NOWARN | __GFP_ZERO); |
b3b94faa DT |
1991 | if (!ht) |
1992 | return -ENOMEM; | |
1993 | ||
f4108a60 | 1994 | error = gfs2_quota_hold(dip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE); |
b3b94faa | 1995 | if (error) |
5407e242 | 1996 | goto out; |
b3b94faa | 1997 | |
b3b94faa | 1998 | /* Count the number of leaves */ |
ec038c82 | 1999 | bh = leaf_bh; |
b3b94faa | 2000 | |
c752666c | 2001 | for (blk = leaf_no; blk; blk = nblk) { |
ec038c82 BP |
2002 | if (blk != leaf_no) { |
2003 | error = get_leaf(dip, blk, &bh); | |
2004 | if (error) | |
2005 | goto out_rlist; | |
2006 | } | |
c752666c SW |
2007 | tmp_leaf = (struct gfs2_leaf *)bh->b_data; |
2008 | nblk = be64_to_cpu(tmp_leaf->lf_next); | |
ec038c82 BP |
2009 | if (blk != leaf_no) |
2010 | brelse(bh); | |
b3b94faa | 2011 | |
70b0c365 | 2012 | gfs2_rlist_add(dip, &rlist, blk); |
b3b94faa DT |
2013 | l_blocks++; |
2014 | } | |
2015 | ||
c3abc29e | 2016 | gfs2_rlist_alloc(&rlist); |
b3b94faa DT |
2017 | |
2018 | for (x = 0; x < rlist.rl_rgrps; x++) { | |
6f6597ba AG |
2019 | struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(rlist.rl_ghs[x].gh_gl); |
2020 | ||
bb8d8a6f | 2021 | rg_blocks += rgd->rd_length; |
b3b94faa DT |
2022 | } |
2023 | ||
2024 | error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs); | |
2025 | if (error) | |
2026 | goto out_rlist; | |
2027 | ||
2028 | error = gfs2_trans_begin(sdp, | |
5c676f6d | 2029 | rg_blocks + (DIV_ROUND_UP(size, sdp->sd_jbsize) + 1) + |
cc44457f BP |
2030 | RES_DINODE + RES_STATFS + RES_QUOTA, RES_DINODE + |
2031 | l_blocks); | |
b3b94faa DT |
2032 | if (error) |
2033 | goto out_rg_gunlock; | |
2034 | ||
ec038c82 BP |
2035 | bh = leaf_bh; |
2036 | ||
c752666c | 2037 | for (blk = leaf_no; blk; blk = nblk) { |
0ddeded4 AG |
2038 | struct gfs2_rgrpd *rgd; |
2039 | ||
ec038c82 BP |
2040 | if (blk != leaf_no) { |
2041 | error = get_leaf(dip, blk, &bh); | |
2042 | if (error) | |
2043 | goto out_end_trans; | |
2044 | } | |
c752666c SW |
2045 | tmp_leaf = (struct gfs2_leaf *)bh->b_data; |
2046 | nblk = be64_to_cpu(tmp_leaf->lf_next); | |
ec038c82 BP |
2047 | if (blk != leaf_no) |
2048 | brelse(bh); | |
b3b94faa | 2049 | |
0ddeded4 AG |
2050 | rgd = gfs2_blk2rgrpd(sdp, blk, true); |
2051 | gfs2_free_meta(dip, rgd, blk, 1); | |
77658aad | 2052 | gfs2_add_inode_blocks(&dip->i_inode, -1); |
b3b94faa DT |
2053 | } |
2054 | ||
cd915493 | 2055 | error = gfs2_dir_write_data(dip, ht, index * sizeof(u64), size); |
b3b94faa DT |
2056 | if (error != size) { |
2057 | if (error >= 0) | |
2058 | error = -EIO; | |
2059 | goto out_end_trans; | |
2060 | } | |
2061 | ||
2062 | error = gfs2_meta_inode_buffer(dip, &dibh); | |
2063 | if (error) | |
2064 | goto out_end_trans; | |
2065 | ||
350a9b0a | 2066 | gfs2_trans_add_meta(dip->i_gl, dibh); |
d24a7a43 BP |
2067 | /* On the last dealloc, make this a regular file in case we crash. |
2068 | (We don't want to free these blocks a second time.) */ | |
2069 | if (last_dealloc) | |
2070 | dip->i_inode.i_mode = S_IFREG; | |
539e5d6b | 2071 | gfs2_dinode_out(dip, dibh->b_data); |
b3b94faa DT |
2072 | brelse(dibh); |
2073 | ||
a91ea69f | 2074 | out_end_trans: |
b3b94faa | 2075 | gfs2_trans_end(sdp); |
a91ea69f | 2076 | out_rg_gunlock: |
b3b94faa | 2077 | gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs); |
a91ea69f | 2078 | out_rlist: |
b3b94faa | 2079 | gfs2_rlist_free(&rlist); |
b3b94faa | 2080 | gfs2_quota_unhold(dip); |
182fe5ab | 2081 | out: |
3cdcf63e | 2082 | kvfree(ht); |
b3b94faa DT |
2083 | return error; |
2084 | } | |
2085 | ||
2086 | /** | |
2087 | * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory | |
2088 | * @dip: the directory | |
2089 | * | |
2090 | * Dealloc all on-disk directory leaves to FREEMETA state | |
2091 | * Change on-disk inode type to "regular file" | |
2092 | * | |
2093 | * Returns: errno | |
2094 | */ | |
2095 | ||
2096 | int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip) | |
2097 | { | |
556bb179 BP |
2098 | struct buffer_head *bh; |
2099 | struct gfs2_leaf *leaf; | |
2100 | u32 hsize, len; | |
556bb179 BP |
2101 | u32 index = 0, next_index; |
2102 | __be64 *lp; | |
2103 | u64 leaf_no; | |
2104 | int error = 0, last; | |
2105 | ||
47a9a527 | 2106 | hsize = BIT(dip->i_depth); |
556bb179 | 2107 | |
17d539f0 SW |
2108 | lp = gfs2_dir_get_hash_table(dip); |
2109 | if (IS_ERR(lp)) | |
2110 | return PTR_ERR(lp); | |
556bb179 BP |
2111 | |
2112 | while (index < hsize) { | |
17d539f0 | 2113 | leaf_no = be64_to_cpu(lp[index]); |
556bb179 BP |
2114 | if (leaf_no) { |
2115 | error = get_leaf(dip, leaf_no, &bh); | |
2116 | if (error) | |
2117 | goto out; | |
2118 | leaf = (struct gfs2_leaf *)bh->b_data; | |
47a9a527 | 2119 | len = BIT(dip->i_depth - be16_to_cpu(leaf->lf_depth)); |
556bb179 BP |
2120 | |
2121 | next_index = (index & ~(len - 1)) + len; | |
2122 | last = ((next_index >= hsize) ? 1 : 0); | |
2123 | error = leaf_dealloc(dip, index, len, leaf_no, bh, | |
2124 | last); | |
2125 | brelse(bh); | |
2126 | if (error) | |
2127 | goto out; | |
2128 | index = next_index; | |
2129 | } else | |
2130 | index++; | |
2131 | } | |
2132 | ||
2133 | if (index != hsize) { | |
2134 | gfs2_consist_inode(dip); | |
2135 | error = -EIO; | |
2136 | } | |
2137 | ||
2138 | out: | |
556bb179 BP |
2139 | |
2140 | return error; | |
b3b94faa DT |
2141 | } |
2142 | ||
2143 | /** | |
2144 | * gfs2_diradd_alloc_required - find if adding entry will require an allocation | |
2145 | * @ip: the file being written to | |
2146 | * @filname: the filename that's going to be added | |
3c1c0ae1 | 2147 | * @da: The structure to return dir alloc info |
b3b94faa | 2148 | * |
3c1c0ae1 | 2149 | * Returns: 0 if ok, -ve on error |
b3b94faa DT |
2150 | */ |
2151 | ||
3c1c0ae1 SW |
2152 | int gfs2_diradd_alloc_required(struct inode *inode, const struct qstr *name, |
2153 | struct gfs2_diradd *da) | |
b3b94faa | 2154 | { |
22b5a6c0 | 2155 | struct gfs2_inode *ip = GFS2_I(inode); |
3c1c0ae1 | 2156 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
22b5a6c0 | 2157 | const unsigned int extra = sizeof(struct gfs2_dinode) - sizeof(struct gfs2_leaf); |
c752666c SW |
2158 | struct gfs2_dirent *dent; |
2159 | struct buffer_head *bh; | |
b3b94faa | 2160 | |
3c1c0ae1 | 2161 | da->nr_blocks = 0; |
2b47dad8 SW |
2162 | da->bh = NULL; |
2163 | da->dent = NULL; | |
3c1c0ae1 | 2164 | |
c752666c | 2165 | dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, &bh); |
ed386507 | 2166 | if (!dent) { |
3c1c0ae1 | 2167 | da->nr_blocks = sdp->sd_max_dirres; |
22b5a6c0 SW |
2168 | if (!(ip->i_diskflags & GFS2_DIF_EXHASH) && |
2169 | (GFS2_DIRENT_SIZE(name->len) < extra)) | |
2170 | da->nr_blocks = 1; | |
3c1c0ae1 | 2171 | return 0; |
ed386507 | 2172 | } |
c752666c SW |
2173 | if (IS_ERR(dent)) |
2174 | return PTR_ERR(dent); | |
19aeb5a6 BP |
2175 | |
2176 | if (da->save_loc) { | |
2177 | da->bh = bh; | |
2178 | da->dent = dent; | |
2179 | } else { | |
2180 | brelse(bh); | |
2181 | } | |
c752666c | 2182 | return 0; |
b3b94faa DT |
2183 | } |
2184 |