Commit | Line | Data |
---|---|---|
b3b94faa DT |
1 | /* |
2 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
3a8a9a10 | 3 | * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved. |
b3b94faa DT |
4 | * |
5 | * This copyrighted material is made available to anyone wishing to use, | |
6 | * modify, copy, or redistribute it subject to the terms and conditions | |
e9fc2aa0 | 7 | * of the GNU General Public License version 2. |
b3b94faa DT |
8 | */ |
9 | ||
10 | /* | |
61e085a8 SW |
11 | * Implements Extendible Hashing as described in: |
12 | * "Extendible Hashing" by Fagin, et al in | |
13 | * __ACM Trans. on Database Systems__, Sept 1979. | |
14 | * | |
15 | * | |
16 | * Here's the layout of dirents which is essentially the same as that of ext2 | |
17 | * within a single block. The field de_name_len is the number of bytes | |
18 | * actually required for the name (no null terminator). The field de_rec_len | |
19 | * is the number of bytes allocated to the dirent. The offset of the next | |
20 | * dirent in the block is (dirent + dirent->de_rec_len). When a dirent is | |
21 | * deleted, the preceding dirent inherits its allocated space, ie | |
22 | * prev->de_rec_len += deleted->de_rec_len. Since the next dirent is obtained | |
23 | * by adding de_rec_len to the current dirent, this essentially causes the | |
24 | * deleted dirent to get jumped over when iterating through all the dirents. | |
25 | * | |
26 | * When deleting the first dirent in a block, there is no previous dirent so | |
27 | * the field de_ino is set to zero to designate it as deleted. When allocating | |
28 | * a dirent, gfs2_dirent_alloc iterates through the dirents in a block. If the | |
29 | * first dirent has (de_ino == 0) and de_rec_len is large enough, this first | |
30 | * dirent is allocated. Otherwise it must go through all the 'used' dirents | |
31 | * searching for one in which the amount of total space minus the amount of | |
32 | * used space will provide enough space for the new dirent. | |
33 | * | |
34 | * There are two types of blocks in which dirents reside. In a stuffed dinode, | |
35 | * the dirents begin at offset sizeof(struct gfs2_dinode) from the beginning of | |
36 | * the block. In leaves, they begin at offset sizeof(struct gfs2_leaf) from the | |
37 | * beginning of the leaf block. The dirents reside in leaves when | |
38 | * | |
383f01fb | 39 | * dip->i_diskflags & GFS2_DIF_EXHASH is true |
61e085a8 SW |
40 | * |
41 | * Otherwise, the dirents are "linear", within a single stuffed dinode block. | |
42 | * | |
43 | * When the dirents are in leaves, the actual contents of the directory file are | |
44 | * used as an array of 64-bit block pointers pointing to the leaf blocks. The | |
45 | * dirents are NOT in the directory file itself. There can be more than one | |
46 | * block pointer in the array that points to the same leaf. In fact, when a | |
47 | * directory is first converted from linear to exhash, all of the pointers | |
48 | * point to the same leaf. | |
49 | * | |
50 | * When a leaf is completely full, the size of the hash table can be | |
51 | * doubled unless it is already at the maximum size which is hard coded into | |
52 | * GFS2_DIR_MAX_DEPTH. After that, leaves are chained together in a linked list, | |
53 | * but never before the maximum hash table size has been reached. | |
54 | */ | |
b3b94faa | 55 | |
d77d1b58 JP |
56 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
57 | ||
b3b94faa DT |
58 | #include <linux/slab.h> |
59 | #include <linux/spinlock.h> | |
b3b94faa DT |
60 | #include <linux/buffer_head.h> |
61 | #include <linux/sort.h> | |
5c676f6d | 62 | #include <linux/gfs2_ondisk.h> |
71b86f56 | 63 | #include <linux/crc32.h> |
fe1bdedc | 64 | #include <linux/vmalloc.h> |
b3b94faa DT |
65 | |
66 | #include "gfs2.h" | |
5c676f6d | 67 | #include "incore.h" |
b3b94faa DT |
68 | #include "dir.h" |
69 | #include "glock.h" | |
70 | #include "inode.h" | |
b3b94faa DT |
71 | #include "meta_io.h" |
72 | #include "quota.h" | |
73 | #include "rgrp.h" | |
74 | #include "trans.h" | |
e13940ba | 75 | #include "bmap.h" |
5c676f6d | 76 | #include "util.h" |
b3b94faa DT |
77 | |
78 | #define IS_LEAF 1 /* Hashed (leaf) directory */ | |
79 | #define IS_DINODE 2 /* Linear (stuffed dinode block) directory */ | |
80 | ||
dfe4d34b BP |
81 | #define MAX_RA_BLOCKS 32 /* max read-ahead blocks */ |
82 | ||
cd915493 SW |
83 | #define gfs2_disk_hash2offset(h) (((u64)(h)) >> 1) |
84 | #define gfs2_dir_offset2hash(p) ((u32)(((u64)(p)) << 1)) | |
471f3db2 BM |
85 | #define GFS2_HASH_INDEX_MASK 0xffffc000 |
86 | #define GFS2_USE_HASH_FLAG 0x2000 | |
b3b94faa | 87 | |
8d123585 SW |
88 | struct qstr gfs2_qdot __read_mostly; |
89 | struct qstr gfs2_qdotdot __read_mostly; | |
90 | ||
2bdbc5d7 SW |
91 | typedef int (*gfs2_dscan_t)(const struct gfs2_dirent *dent, |
92 | const struct qstr *name, void *opaque); | |
b3b94faa | 93 | |
cd915493 | 94 | int gfs2_dir_get_new_buffer(struct gfs2_inode *ip, u64 block, |
61e085a8 | 95 | struct buffer_head **bhp) |
e13940ba SW |
96 | { |
97 | struct buffer_head *bh; | |
e13940ba | 98 | |
61e085a8 | 99 | bh = gfs2_meta_new(ip->i_gl, block); |
350a9b0a | 100 | gfs2_trans_add_meta(ip->i_gl, bh); |
61e085a8 SW |
101 | gfs2_metatype_set(bh, GFS2_METATYPE_JD, GFS2_FORMAT_JD); |
102 | gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header)); | |
e13940ba SW |
103 | *bhp = bh; |
104 | return 0; | |
105 | } | |
106 | ||
cd915493 | 107 | static int gfs2_dir_get_existing_buffer(struct gfs2_inode *ip, u64 block, |
61e085a8 SW |
108 | struct buffer_head **bhp) |
109 | { | |
110 | struct buffer_head *bh; | |
111 | int error; | |
e13940ba | 112 | |
c8d57703 | 113 | error = gfs2_meta_read(ip->i_gl, block, DIO_WAIT, 0, &bh); |
61e085a8 SW |
114 | if (error) |
115 | return error; | |
feaa7bba | 116 | if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), bh, GFS2_METATYPE_JD)) { |
61e085a8 SW |
117 | brelse(bh); |
118 | return -EIO; | |
119 | } | |
120 | *bhp = bh; | |
121 | return 0; | |
122 | } | |
e13940ba SW |
123 | |
124 | static int gfs2_dir_write_stuffed(struct gfs2_inode *ip, const char *buf, | |
125 | unsigned int offset, unsigned int size) | |
e13940ba SW |
126 | { |
127 | struct buffer_head *dibh; | |
128 | int error; | |
129 | ||
130 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
131 | if (error) | |
132 | return error; | |
133 | ||
350a9b0a | 134 | gfs2_trans_add_meta(ip->i_gl, dibh); |
c752666c | 135 | memcpy(dibh->b_data + offset + sizeof(struct gfs2_dinode), buf, size); |
a2e0f799 SW |
136 | if (ip->i_inode.i_size < offset + size) |
137 | i_size_write(&ip->i_inode, offset + size); | |
4bd91ba1 | 138 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME; |
539e5d6b | 139 | gfs2_dinode_out(ip, dibh->b_data); |
e13940ba SW |
140 | |
141 | brelse(dibh); | |
142 | ||
143 | return size; | |
144 | } | |
145 | ||
146 | ||
147 | ||
148 | /** | |
149 | * gfs2_dir_write_data - Write directory information to the inode | |
150 | * @ip: The GFS2 inode | |
151 | * @buf: The buffer containing information to be written | |
152 | * @offset: The file offset to start writing at | |
153 | * @size: The amount of data to write | |
154 | * | |
155 | * Returns: The number of bytes correctly written or error code | |
156 | */ | |
157 | static int gfs2_dir_write_data(struct gfs2_inode *ip, const char *buf, | |
cd915493 | 158 | u64 offset, unsigned int size) |
e13940ba | 159 | { |
feaa7bba | 160 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
e13940ba | 161 | struct buffer_head *dibh; |
cd915493 SW |
162 | u64 lblock, dblock; |
163 | u32 extlen = 0; | |
e13940ba SW |
164 | unsigned int o; |
165 | int copied = 0; | |
166 | int error = 0; | |
9b8c81d1 | 167 | int new = 0; |
e13940ba SW |
168 | |
169 | if (!size) | |
170 | return 0; | |
171 | ||
172 | if (gfs2_is_stuffed(ip) && | |
173 | offset + size <= sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) | |
568f4c96 SW |
174 | return gfs2_dir_write_stuffed(ip, buf, (unsigned int)offset, |
175 | size); | |
e13940ba SW |
176 | |
177 | if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip))) | |
178 | return -EINVAL; | |
179 | ||
180 | if (gfs2_is_stuffed(ip)) { | |
f25ef0c1 | 181 | error = gfs2_unstuff_dinode(ip, NULL); |
e13940ba | 182 | if (error) |
c752666c | 183 | return error; |
e13940ba SW |
184 | } |
185 | ||
186 | lblock = offset; | |
187 | o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header); | |
188 | ||
189 | while (copied < size) { | |
190 | unsigned int amount; | |
191 | struct buffer_head *bh; | |
e13940ba SW |
192 | |
193 | amount = size - copied; | |
194 | if (amount > sdp->sd_sb.sb_bsize - o) | |
195 | amount = sdp->sd_sb.sb_bsize - o; | |
196 | ||
197 | if (!extlen) { | |
198 | new = 1; | |
feaa7bba | 199 | error = gfs2_extent_map(&ip->i_inode, lblock, &new, |
fd88de56 | 200 | &dblock, &extlen); |
e13940ba SW |
201 | if (error) |
202 | goto fail; | |
203 | error = -EIO; | |
204 | if (gfs2_assert_withdraw(sdp, dblock)) | |
205 | goto fail; | |
206 | } | |
207 | ||
61e085a8 SW |
208 | if (amount == sdp->sd_jbsize || new) |
209 | error = gfs2_dir_get_new_buffer(ip, dblock, &bh); | |
210 | else | |
211 | error = gfs2_dir_get_existing_buffer(ip, dblock, &bh); | |
212 | ||
e13940ba SW |
213 | if (error) |
214 | goto fail; | |
215 | ||
350a9b0a | 216 | gfs2_trans_add_meta(ip->i_gl, bh); |
e13940ba SW |
217 | memcpy(bh->b_data + o, buf, amount); |
218 | brelse(bh); | |
e13940ba | 219 | |
899bb264 | 220 | buf += amount; |
e13940ba SW |
221 | copied += amount; |
222 | lblock++; | |
223 | dblock++; | |
224 | extlen--; | |
225 | ||
226 | o = sizeof(struct gfs2_meta_header); | |
227 | } | |
228 | ||
229 | out: | |
230 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
231 | if (error) | |
232 | return error; | |
233 | ||
a2e0f799 SW |
234 | if (ip->i_inode.i_size < offset + copied) |
235 | i_size_write(&ip->i_inode, offset + copied); | |
4bd91ba1 | 236 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME; |
e13940ba | 237 | |
350a9b0a | 238 | gfs2_trans_add_meta(ip->i_gl, dibh); |
539e5d6b | 239 | gfs2_dinode_out(ip, dibh->b_data); |
e13940ba SW |
240 | brelse(dibh); |
241 | ||
242 | return copied; | |
243 | fail: | |
244 | if (copied) | |
245 | goto out; | |
246 | return error; | |
247 | } | |
248 | ||
4c28d338 SW |
249 | static int gfs2_dir_read_stuffed(struct gfs2_inode *ip, __be64 *buf, |
250 | unsigned int size) | |
e13940ba SW |
251 | { |
252 | struct buffer_head *dibh; | |
253 | int error; | |
254 | ||
255 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
256 | if (!error) { | |
4c28d338 | 257 | memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size); |
e13940ba SW |
258 | brelse(dibh); |
259 | } | |
260 | ||
261 | return (error) ? error : size; | |
262 | } | |
263 | ||
264 | ||
265 | /** | |
266 | * gfs2_dir_read_data - Read a data from a directory inode | |
267 | * @ip: The GFS2 Inode | |
268 | * @buf: The buffer to place result into | |
e13940ba SW |
269 | * @size: Amount of data to transfer |
270 | * | |
271 | * Returns: The amount of data actually copied or the error | |
272 | */ | |
4c28d338 SW |
273 | static int gfs2_dir_read_data(struct gfs2_inode *ip, __be64 *buf, |
274 | unsigned int size) | |
e13940ba | 275 | { |
feaa7bba | 276 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
cd915493 SW |
277 | u64 lblock, dblock; |
278 | u32 extlen = 0; | |
e13940ba SW |
279 | unsigned int o; |
280 | int copied = 0; | |
281 | int error = 0; | |
e13940ba SW |
282 | |
283 | if (gfs2_is_stuffed(ip)) | |
4c28d338 | 284 | return gfs2_dir_read_stuffed(ip, buf, size); |
e13940ba SW |
285 | |
286 | if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip))) | |
287 | return -EINVAL; | |
288 | ||
4c28d338 | 289 | lblock = 0; |
e13940ba SW |
290 | o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header); |
291 | ||
292 | while (copied < size) { | |
293 | unsigned int amount; | |
294 | struct buffer_head *bh; | |
295 | int new; | |
296 | ||
297 | amount = size - copied; | |
298 | if (amount > sdp->sd_sb.sb_bsize - o) | |
299 | amount = sdp->sd_sb.sb_bsize - o; | |
300 | ||
301 | if (!extlen) { | |
302 | new = 0; | |
feaa7bba | 303 | error = gfs2_extent_map(&ip->i_inode, lblock, &new, |
fd88de56 | 304 | &dblock, &extlen); |
7276b3b0 | 305 | if (error || !dblock) |
e13940ba | 306 | goto fail; |
7276b3b0 | 307 | BUG_ON(extlen < 1); |
7276b3b0 | 308 | bh = gfs2_meta_ra(ip->i_gl, dblock, extlen); |
b7d8ac3e | 309 | } else { |
c8d57703 | 310 | error = gfs2_meta_read(ip->i_gl, dblock, DIO_WAIT, 0, &bh); |
e13940ba SW |
311 | if (error) |
312 | goto fail; | |
7276b3b0 SW |
313 | } |
314 | error = gfs2_metatype_check(sdp, bh, GFS2_METATYPE_JD); | |
315 | if (error) { | |
316 | brelse(bh); | |
317 | goto fail; | |
318 | } | |
319 | dblock++; | |
320 | extlen--; | |
e13940ba SW |
321 | memcpy(buf, bh->b_data + o, amount); |
322 | brelse(bh); | |
4c28d338 | 323 | buf += (amount/sizeof(__be64)); |
e13940ba SW |
324 | copied += amount; |
325 | lblock++; | |
e13940ba SW |
326 | o = sizeof(struct gfs2_meta_header); |
327 | } | |
328 | ||
329 | return copied; | |
330 | fail: | |
331 | return (copied) ? copied : error; | |
332 | } | |
333 | ||
17d539f0 SW |
334 | /** |
335 | * gfs2_dir_get_hash_table - Get pointer to the dir hash table | |
336 | * @ip: The inode in question | |
337 | * | |
338 | * Returns: The hash table or an error | |
339 | */ | |
340 | ||
341 | static __be64 *gfs2_dir_get_hash_table(struct gfs2_inode *ip) | |
342 | { | |
343 | struct inode *inode = &ip->i_inode; | |
344 | int ret; | |
345 | u32 hsize; | |
346 | __be64 *hc; | |
347 | ||
348 | BUG_ON(!(ip->i_diskflags & GFS2_DIF_EXHASH)); | |
349 | ||
350 | hc = ip->i_hash_cache; | |
351 | if (hc) | |
352 | return hc; | |
353 | ||
354 | hsize = 1 << ip->i_depth; | |
355 | hsize *= sizeof(__be64); | |
356 | if (hsize != i_size_read(&ip->i_inode)) { | |
357 | gfs2_consist_inode(ip); | |
358 | return ERR_PTR(-EIO); | |
359 | } | |
360 | ||
e8830d88 BP |
361 | hc = kmalloc(hsize, GFP_NOFS | __GFP_NOWARN); |
362 | if (hc == NULL) | |
363 | hc = __vmalloc(hsize, GFP_NOFS, PAGE_KERNEL); | |
364 | ||
17d539f0 SW |
365 | if (hc == NULL) |
366 | return ERR_PTR(-ENOMEM); | |
367 | ||
4c28d338 | 368 | ret = gfs2_dir_read_data(ip, hc, hsize); |
17d539f0 | 369 | if (ret < 0) { |
3cdcf63e | 370 | kvfree(hc); |
17d539f0 SW |
371 | return ERR_PTR(ret); |
372 | } | |
373 | ||
374 | spin_lock(&inode->i_lock); | |
9265f1d0 | 375 | if (likely(!ip->i_hash_cache)) { |
17d539f0 | 376 | ip->i_hash_cache = hc; |
9265f1d0 AV |
377 | hc = NULL; |
378 | } | |
17d539f0 | 379 | spin_unlock(&inode->i_lock); |
9265f1d0 | 380 | kvfree(hc); |
17d539f0 SW |
381 | |
382 | return ip->i_hash_cache; | |
383 | } | |
384 | ||
385 | /** | |
386 | * gfs2_dir_hash_inval - Invalidate dir hash | |
387 | * @ip: The directory inode | |
388 | * | |
389 | * Must be called with an exclusive glock, or during glock invalidation. | |
390 | */ | |
391 | void gfs2_dir_hash_inval(struct gfs2_inode *ip) | |
392 | { | |
c36b97e9 BP |
393 | __be64 *hc; |
394 | ||
395 | spin_lock(&ip->i_inode.i_lock); | |
396 | hc = ip->i_hash_cache; | |
17d539f0 | 397 | ip->i_hash_cache = NULL; |
c36b97e9 BP |
398 | spin_unlock(&ip->i_inode.i_lock); |
399 | ||
3cdcf63e | 400 | kvfree(hc); |
17d539f0 SW |
401 | } |
402 | ||
5e7d65cd SW |
403 | static inline int gfs2_dirent_sentinel(const struct gfs2_dirent *dent) |
404 | { | |
405 | return dent->de_inum.no_addr == 0 || dent->de_inum.no_formal_ino == 0; | |
406 | } | |
407 | ||
c752666c SW |
408 | static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent, |
409 | const struct qstr *name, int ret) | |
410 | { | |
5e7d65cd | 411 | if (!gfs2_dirent_sentinel(dent) && |
c752666c SW |
412 | be32_to_cpu(dent->de_hash) == name->hash && |
413 | be16_to_cpu(dent->de_name_len) == name->len && | |
2bdbc5d7 | 414 | memcmp(dent+1, name->name, name->len) == 0) |
c752666c SW |
415 | return ret; |
416 | return 0; | |
417 | } | |
418 | ||
419 | static int gfs2_dirent_find(const struct gfs2_dirent *dent, | |
71b86f56 SW |
420 | const struct qstr *name, |
421 | void *opaque) | |
c752666c SW |
422 | { |
423 | return __gfs2_dirent_find(dent, name, 1); | |
424 | } | |
425 | ||
426 | static int gfs2_dirent_prev(const struct gfs2_dirent *dent, | |
71b86f56 SW |
427 | const struct qstr *name, |
428 | void *opaque) | |
c752666c SW |
429 | { |
430 | return __gfs2_dirent_find(dent, name, 2); | |
431 | } | |
432 | ||
433 | /* | |
434 | * name->name holds ptr to start of block. | |
435 | * name->len holds size of block. | |
b3b94faa | 436 | */ |
c752666c | 437 | static int gfs2_dirent_last(const struct gfs2_dirent *dent, |
71b86f56 SW |
438 | const struct qstr *name, |
439 | void *opaque) | |
c752666c SW |
440 | { |
441 | const char *start = name->name; | |
442 | const char *end = (const char *)dent + be16_to_cpu(dent->de_rec_len); | |
443 | if (name->len == (end - start)) | |
444 | return 1; | |
445 | return 0; | |
446 | } | |
b3b94faa | 447 | |
34017472 BM |
448 | /* Look for the dirent that contains the offset specified in data. Once we |
449 | * find that dirent, there must be space available there for the new dirent */ | |
450 | static int gfs2_dirent_find_offset(const struct gfs2_dirent *dent, | |
451 | const struct qstr *name, | |
452 | void *ptr) | |
453 | { | |
454 | unsigned required = GFS2_DIRENT_SIZE(name->len); | |
455 | unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len)); | |
456 | unsigned totlen = be16_to_cpu(dent->de_rec_len); | |
457 | ||
458 | if (ptr < (void *)dent || ptr >= (void *)dent + totlen) | |
459 | return 0; | |
460 | if (gfs2_dirent_sentinel(dent)) | |
461 | actual = 0; | |
462 | if (ptr < (void *)dent + actual) | |
463 | return -1; | |
464 | if ((void *)dent + totlen >= ptr + required) | |
465 | return 1; | |
466 | return -1; | |
467 | } | |
468 | ||
c752666c | 469 | static int gfs2_dirent_find_space(const struct gfs2_dirent *dent, |
71b86f56 SW |
470 | const struct qstr *name, |
471 | void *opaque) | |
b3b94faa | 472 | { |
c752666c SW |
473 | unsigned required = GFS2_DIRENT_SIZE(name->len); |
474 | unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len)); | |
475 | unsigned totlen = be16_to_cpu(dent->de_rec_len); | |
476 | ||
5e7d65cd | 477 | if (gfs2_dirent_sentinel(dent)) |
728a756b | 478 | actual = 0; |
c5392124 | 479 | if (totlen - actual >= required) |
c752666c SW |
480 | return 1; |
481 | return 0; | |
482 | } | |
483 | ||
71b86f56 SW |
484 | struct dirent_gather { |
485 | const struct gfs2_dirent **pdent; | |
486 | unsigned offset; | |
487 | }; | |
488 | ||
489 | static int gfs2_dirent_gather(const struct gfs2_dirent *dent, | |
490 | const struct qstr *name, | |
491 | void *opaque) | |
492 | { | |
493 | struct dirent_gather *g = opaque; | |
5e7d65cd | 494 | if (!gfs2_dirent_sentinel(dent)) { |
71b86f56 SW |
495 | g->pdent[g->offset++] = dent; |
496 | } | |
497 | return 0; | |
498 | } | |
499 | ||
c752666c SW |
500 | /* |
501 | * Other possible things to check: | |
502 | * - Inode located within filesystem size (and on valid block) | |
503 | * - Valid directory entry type | |
504 | * Not sure how heavy-weight we want to make this... could also check | |
505 | * hash is correct for example, but that would take a lot of extra time. | |
506 | * For now the most important thing is to check that the various sizes | |
507 | * are correct. | |
508 | */ | |
509 | static int gfs2_check_dirent(struct gfs2_dirent *dent, unsigned int offset, | |
510 | unsigned int size, unsigned int len, int first) | |
511 | { | |
512 | const char *msg = "gfs2_dirent too small"; | |
513 | if (unlikely(size < sizeof(struct gfs2_dirent))) | |
514 | goto error; | |
515 | msg = "gfs2_dirent misaligned"; | |
516 | if (unlikely(offset & 0x7)) | |
517 | goto error; | |
518 | msg = "gfs2_dirent points beyond end of block"; | |
519 | if (unlikely(offset + size > len)) | |
520 | goto error; | |
521 | msg = "zero inode number"; | |
5e7d65cd | 522 | if (unlikely(!first && gfs2_dirent_sentinel(dent))) |
c752666c SW |
523 | goto error; |
524 | msg = "name length is greater than space in dirent"; | |
5e7d65cd | 525 | if (!gfs2_dirent_sentinel(dent) && |
c752666c SW |
526 | unlikely(sizeof(struct gfs2_dirent)+be16_to_cpu(dent->de_name_len) > |
527 | size)) | |
528 | goto error; | |
529 | return 0; | |
530 | error: | |
d77d1b58 JP |
531 | pr_warn("%s: %s (%s)\n", |
532 | __func__, msg, first ? "first in block" : "not first in block"); | |
c752666c | 533 | return -EIO; |
b3b94faa DT |
534 | } |
535 | ||
71b86f56 | 536 | static int gfs2_dirent_offset(const void *buf) |
c752666c | 537 | { |
71b86f56 SW |
538 | const struct gfs2_meta_header *h = buf; |
539 | int offset; | |
c752666c SW |
540 | |
541 | BUG_ON(buf == NULL); | |
c752666c | 542 | |
e3167ded | 543 | switch(be32_to_cpu(h->mh_type)) { |
c752666c SW |
544 | case GFS2_METATYPE_LF: |
545 | offset = sizeof(struct gfs2_leaf); | |
546 | break; | |
547 | case GFS2_METATYPE_DI: | |
548 | offset = sizeof(struct gfs2_dinode); | |
549 | break; | |
550 | default: | |
551 | goto wrong_type; | |
552 | } | |
71b86f56 SW |
553 | return offset; |
554 | wrong_type: | |
d77d1b58 | 555 | pr_warn("%s: wrong block type %u\n", __func__, be32_to_cpu(h->mh_type)); |
71b86f56 SW |
556 | return -1; |
557 | } | |
558 | ||
2bdbc5d7 | 559 | static struct gfs2_dirent *gfs2_dirent_scan(struct inode *inode, void *buf, |
71b86f56 SW |
560 | unsigned int len, gfs2_dscan_t scan, |
561 | const struct qstr *name, | |
562 | void *opaque) | |
563 | { | |
564 | struct gfs2_dirent *dent, *prev; | |
565 | unsigned offset; | |
566 | unsigned size; | |
567 | int ret = 0; | |
c752666c | 568 | |
71b86f56 SW |
569 | ret = gfs2_dirent_offset(buf); |
570 | if (ret < 0) | |
571 | goto consist_inode; | |
572 | ||
573 | offset = ret; | |
c752666c | 574 | prev = NULL; |
2bdbc5d7 | 575 | dent = buf + offset; |
c752666c SW |
576 | size = be16_to_cpu(dent->de_rec_len); |
577 | if (gfs2_check_dirent(dent, offset, size, len, 1)) | |
578 | goto consist_inode; | |
579 | do { | |
71b86f56 | 580 | ret = scan(dent, name, opaque); |
c752666c SW |
581 | if (ret) |
582 | break; | |
583 | offset += size; | |
584 | if (offset == len) | |
585 | break; | |
586 | prev = dent; | |
2bdbc5d7 | 587 | dent = buf + offset; |
c752666c SW |
588 | size = be16_to_cpu(dent->de_rec_len); |
589 | if (gfs2_check_dirent(dent, offset, size, len, 0)) | |
590 | goto consist_inode; | |
591 | } while(1); | |
592 | ||
593 | switch(ret) { | |
594 | case 0: | |
595 | return NULL; | |
596 | case 1: | |
597 | return dent; | |
598 | case 2: | |
599 | return prev ? prev : dent; | |
600 | default: | |
601 | BUG_ON(ret > 0); | |
602 | return ERR_PTR(ret); | |
603 | } | |
604 | ||
c752666c | 605 | consist_inode: |
feaa7bba | 606 | gfs2_consist_inode(GFS2_I(inode)); |
c752666c SW |
607 | return ERR_PTR(-EIO); |
608 | } | |
609 | ||
2bdbc5d7 SW |
610 | static int dirent_check_reclen(struct gfs2_inode *dip, |
611 | const struct gfs2_dirent *d, const void *end_p) | |
612 | { | |
613 | const void *ptr = d; | |
614 | u16 rec_len = be16_to_cpu(d->de_rec_len); | |
615 | ||
616 | if (unlikely(rec_len < sizeof(struct gfs2_dirent))) | |
617 | goto broken; | |
618 | ptr += rec_len; | |
619 | if (ptr < end_p) | |
620 | return rec_len; | |
621 | if (ptr == end_p) | |
622 | return -ENOENT; | |
623 | broken: | |
624 | gfs2_consist_inode(dip); | |
625 | return -EIO; | |
626 | } | |
627 | ||
b3b94faa DT |
628 | /** |
629 | * dirent_next - Next dirent | |
630 | * @dip: the directory | |
631 | * @bh: The buffer | |
632 | * @dent: Pointer to list of dirents | |
633 | * | |
634 | * Returns: 0 on success, error code otherwise | |
635 | */ | |
636 | ||
637 | static int dirent_next(struct gfs2_inode *dip, struct buffer_head *bh, | |
638 | struct gfs2_dirent **dent) | |
639 | { | |
2bdbc5d7 SW |
640 | struct gfs2_dirent *cur = *dent, *tmp; |
641 | char *bh_end = bh->b_data + bh->b_size; | |
642 | int ret; | |
b3b94faa | 643 | |
2bdbc5d7 SW |
644 | ret = dirent_check_reclen(dip, cur, bh_end); |
645 | if (ret < 0) | |
646 | return ret; | |
4dd651ad | 647 | |
2bdbc5d7 SW |
648 | tmp = (void *)cur + ret; |
649 | ret = dirent_check_reclen(dip, tmp, bh_end); | |
650 | if (ret == -EIO) | |
651 | return ret; | |
4dd651ad | 652 | |
b3b94faa | 653 | /* Only the first dent could ever have de_inum.no_addr == 0 */ |
5e7d65cd | 654 | if (gfs2_dirent_sentinel(tmp)) { |
b3b94faa DT |
655 | gfs2_consist_inode(dip); |
656 | return -EIO; | |
657 | } | |
658 | ||
659 | *dent = tmp; | |
b3b94faa DT |
660 | return 0; |
661 | } | |
662 | ||
663 | /** | |
664 | * dirent_del - Delete a dirent | |
665 | * @dip: The GFS2 inode | |
666 | * @bh: The buffer | |
667 | * @prev: The previous dirent | |
668 | * @cur: The current dirent | |
669 | * | |
670 | */ | |
671 | ||
672 | static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh, | |
673 | struct gfs2_dirent *prev, struct gfs2_dirent *cur) | |
674 | { | |
cd915493 | 675 | u16 cur_rec_len, prev_rec_len; |
b3b94faa | 676 | |
5e7d65cd | 677 | if (gfs2_dirent_sentinel(cur)) { |
b3b94faa DT |
678 | gfs2_consist_inode(dip); |
679 | return; | |
680 | } | |
681 | ||
350a9b0a | 682 | gfs2_trans_add_meta(dip->i_gl, bh); |
b3b94faa DT |
683 | |
684 | /* If there is no prev entry, this is the first entry in the block. | |
685 | The de_rec_len is already as big as it needs to be. Just zero | |
686 | out the inode number and return. */ | |
687 | ||
688 | if (!prev) { | |
5e7d65cd SW |
689 | cur->de_inum.no_addr = 0; |
690 | cur->de_inum.no_formal_ino = 0; | |
b3b94faa DT |
691 | return; |
692 | } | |
693 | ||
694 | /* Combine this dentry with the previous one. */ | |
695 | ||
fc69d0d3 SW |
696 | prev_rec_len = be16_to_cpu(prev->de_rec_len); |
697 | cur_rec_len = be16_to_cpu(cur->de_rec_len); | |
b3b94faa DT |
698 | |
699 | if ((char *)prev + prev_rec_len != (char *)cur) | |
700 | gfs2_consist_inode(dip); | |
701 | if ((char *)cur + cur_rec_len > bh->b_data + bh->b_size) | |
702 | gfs2_consist_inode(dip); | |
703 | ||
704 | prev_rec_len += cur_rec_len; | |
fc69d0d3 | 705 | prev->de_rec_len = cpu_to_be16(prev_rec_len); |
b3b94faa DT |
706 | } |
707 | ||
34017472 BM |
708 | |
709 | static struct gfs2_dirent *do_init_dirent(struct inode *inode, | |
710 | struct gfs2_dirent *dent, | |
711 | const struct qstr *name, | |
712 | struct buffer_head *bh, | |
713 | unsigned offset) | |
714 | { | |
715 | struct gfs2_inode *ip = GFS2_I(inode); | |
716 | struct gfs2_dirent *ndent; | |
717 | unsigned totlen; | |
718 | ||
719 | totlen = be16_to_cpu(dent->de_rec_len); | |
720 | BUG_ON(offset + name->len > totlen); | |
721 | gfs2_trans_add_meta(ip->i_gl, bh); | |
722 | ndent = (struct gfs2_dirent *)((char *)dent + offset); | |
723 | dent->de_rec_len = cpu_to_be16(offset); | |
724 | gfs2_qstr2dirent(name, totlen - offset, ndent); | |
725 | return ndent; | |
726 | } | |
727 | ||
728 | ||
c752666c SW |
729 | /* |
730 | * Takes a dent from which to grab space as an argument. Returns the | |
731 | * newly created dent. | |
b3b94faa | 732 | */ |
08bc2dbc AB |
733 | static struct gfs2_dirent *gfs2_init_dirent(struct inode *inode, |
734 | struct gfs2_dirent *dent, | |
735 | const struct qstr *name, | |
736 | struct buffer_head *bh) | |
b3b94faa | 737 | { |
34017472 | 738 | unsigned offset = 0; |
c752666c | 739 | |
5e7d65cd | 740 | if (!gfs2_dirent_sentinel(dent)) |
c752666c | 741 | offset = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len)); |
34017472 | 742 | return do_init_dirent(inode, dent, name, bh, offset); |
b3b94faa DT |
743 | } |
744 | ||
34017472 BM |
745 | static struct gfs2_dirent *gfs2_dirent_split_alloc(struct inode *inode, |
746 | struct buffer_head *bh, | |
747 | const struct qstr *name, | |
748 | void *ptr) | |
b3b94faa DT |
749 | { |
750 | struct gfs2_dirent *dent; | |
907b9bce | 751 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, |
34017472 | 752 | gfs2_dirent_find_offset, name, ptr); |
c752666c SW |
753 | if (!dent || IS_ERR(dent)) |
754 | return dent; | |
34017472 BM |
755 | return do_init_dirent(inode, dent, name, bh, |
756 | (unsigned)(ptr - (void *)dent)); | |
b3b94faa DT |
757 | } |
758 | ||
cd915493 | 759 | static int get_leaf(struct gfs2_inode *dip, u64 leaf_no, |
b3b94faa DT |
760 | struct buffer_head **bhp) |
761 | { | |
762 | int error; | |
763 | ||
c8d57703 | 764 | error = gfs2_meta_read(dip->i_gl, leaf_no, DIO_WAIT, 0, bhp); |
feaa7bba | 765 | if (!error && gfs2_metatype_check(GFS2_SB(&dip->i_inode), *bhp, GFS2_METATYPE_LF)) { |
d77d1b58 | 766 | /* pr_info("block num=%llu\n", leaf_no); */ |
b3b94faa | 767 | error = -EIO; |
feaa7bba | 768 | } |
b3b94faa DT |
769 | |
770 | return error; | |
771 | } | |
772 | ||
773 | /** | |
774 | * get_leaf_nr - Get a leaf number associated with the index | |
775 | * @dip: The GFS2 inode | |
776 | * @index: | |
777 | * @leaf_out: | |
778 | * | |
779 | * Returns: 0 on success, error code otherwise | |
780 | */ | |
781 | ||
cd915493 SW |
782 | static int get_leaf_nr(struct gfs2_inode *dip, u32 index, |
783 | u64 *leaf_out) | |
b3b94faa | 784 | { |
17d539f0 | 785 | __be64 *hash; |
b3b94faa | 786 | |
17d539f0 SW |
787 | hash = gfs2_dir_get_hash_table(dip); |
788 | if (IS_ERR(hash)) | |
789 | return PTR_ERR(hash); | |
790 | *leaf_out = be64_to_cpu(*(hash + index)); | |
b3b94faa DT |
791 | return 0; |
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); | |
67893f12 | 801 | if (!IS_ERR_VALUE(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; |
9a004508 | 819 | unsigned hsize = 1 << ip->i_depth; |
c752666c SW |
820 | unsigned index; |
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: |
f4154ea0 | 854 | if (unlikely(dent == NULL || IS_ERR(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; | |
26fe5750 | 871 | struct qstr name = { .name = "" }; |
01bcb0de | 872 | struct timespec tv = CURRENT_TIME; |
09010978 | 873 | |
6e87ed0f | 874 | error = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL); |
09010978 SW |
875 | if (error) |
876 | return NULL; | |
877 | bh = gfs2_meta_new(ip->i_gl, bn); | |
c752666c SW |
878 | if (!bh) |
879 | return NULL; | |
09010978 | 880 | |
5731be53 | 881 | gfs2_trans_add_unrevoke(GFS2_SB(inode), bn, 1); |
350a9b0a | 882 | gfs2_trans_add_meta(ip->i_gl, bh); |
c752666c SW |
883 | gfs2_metatype_set(bh, GFS2_METATYPE_LF, GFS2_FORMAT_LF); |
884 | leaf = (struct gfs2_leaf *)bh->b_data; | |
885 | leaf->lf_depth = cpu_to_be16(depth); | |
2bdbc5d7 | 886 | leaf->lf_entries = 0; |
a2d7d021 | 887 | leaf->lf_dirent_format = cpu_to_be32(GFS2_FORMAT_DE); |
2bdbc5d7 | 888 | leaf->lf_next = 0; |
01bcb0de SW |
889 | leaf->lf_inode = cpu_to_be64(ip->i_no_addr); |
890 | leaf->lf_dist = cpu_to_be32(1); | |
891 | leaf->lf_nsec = cpu_to_be32(tv.tv_nsec); | |
892 | leaf->lf_sec = cpu_to_be64(tv.tv_sec); | |
893 | memset(leaf->lf_reserved2, 0, sizeof(leaf->lf_reserved2)); | |
c752666c | 894 | dent = (struct gfs2_dirent *)(leaf+1); |
71b86f56 | 895 | gfs2_qstr2dirent(&name, bh->b_size - sizeof(struct gfs2_leaf), dent); |
c752666c SW |
896 | *pbh = bh; |
897 | return leaf; | |
b3b94faa DT |
898 | } |
899 | ||
900 | /** | |
901 | * dir_make_exhash - Convert a stuffed directory into an ExHash directory | |
902 | * @dip: The GFS2 inode | |
903 | * | |
904 | * Returns: 0 on success, error code otherwise | |
905 | */ | |
906 | ||
c752666c | 907 | static int dir_make_exhash(struct inode *inode) |
b3b94faa | 908 | { |
feaa7bba SW |
909 | struct gfs2_inode *dip = GFS2_I(inode); |
910 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
b3b94faa | 911 | struct gfs2_dirent *dent; |
c752666c | 912 | struct qstr args; |
b3b94faa DT |
913 | struct buffer_head *bh, *dibh; |
914 | struct gfs2_leaf *leaf; | |
915 | int y; | |
cd915493 | 916 | u32 x; |
b44b84d7 AV |
917 | __be64 *lp; |
918 | u64 bn; | |
b3b94faa DT |
919 | int error; |
920 | ||
921 | error = gfs2_meta_inode_buffer(dip, &dibh); | |
922 | if (error) | |
923 | return error; | |
924 | ||
b3b94faa DT |
925 | /* Turn over a new leaf */ |
926 | ||
c752666c SW |
927 | leaf = new_leaf(inode, &bh, 0); |
928 | if (!leaf) | |
929 | return -ENOSPC; | |
930 | bn = bh->b_blocknr; | |
b3b94faa | 931 | |
ad6203f2 SW |
932 | gfs2_assert(sdp, dip->i_entries < (1 << 16)); |
933 | leaf->lf_entries = cpu_to_be16(dip->i_entries); | |
b3b94faa DT |
934 | |
935 | /* Copy dirents */ | |
936 | ||
937 | gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_leaf), dibh, | |
938 | sizeof(struct gfs2_dinode)); | |
939 | ||
940 | /* Find last entry */ | |
941 | ||
942 | x = 0; | |
c752666c SW |
943 | args.len = bh->b_size - sizeof(struct gfs2_dinode) + |
944 | sizeof(struct gfs2_leaf); | |
945 | args.name = bh->b_data; | |
feaa7bba | 946 | dent = gfs2_dirent_scan(&dip->i_inode, bh->b_data, bh->b_size, |
71b86f56 | 947 | gfs2_dirent_last, &args, NULL); |
c752666c SW |
948 | if (!dent) { |
949 | brelse(bh); | |
950 | brelse(dibh); | |
951 | return -EIO; | |
952 | } | |
953 | if (IS_ERR(dent)) { | |
954 | brelse(bh); | |
955 | brelse(dibh); | |
956 | return PTR_ERR(dent); | |
b3b94faa | 957 | } |
b3b94faa DT |
958 | |
959 | /* Adjust the last dirent's record length | |
960 | (Remember that dent still points to the last entry.) */ | |
961 | ||
4dd651ad | 962 | dent->de_rec_len = cpu_to_be16(be16_to_cpu(dent->de_rec_len) + |
b3b94faa | 963 | sizeof(struct gfs2_dinode) - |
4dd651ad | 964 | sizeof(struct gfs2_leaf)); |
b3b94faa DT |
965 | |
966 | brelse(bh); | |
967 | ||
968 | /* We're done with the new leaf block, now setup the new | |
969 | hash table. */ | |
970 | ||
350a9b0a | 971 | gfs2_trans_add_meta(dip->i_gl, dibh); |
b3b94faa DT |
972 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); |
973 | ||
b44b84d7 | 974 | lp = (__be64 *)(dibh->b_data + sizeof(struct gfs2_dinode)); |
b3b94faa DT |
975 | |
976 | for (x = sdp->sd_hash_ptrs; x--; lp++) | |
977 | *lp = cpu_to_be64(bn); | |
978 | ||
a2e0f799 | 979 | i_size_write(inode, sdp->sd_sb.sb_bsize / 2); |
77658aad | 980 | gfs2_add_inode_blocks(&dip->i_inode, 1); |
383f01fb | 981 | dip->i_diskflags |= GFS2_DIF_EXHASH; |
b3b94faa DT |
982 | |
983 | for (x = sdp->sd_hash_ptrs, y = -1; x; x >>= 1, y++) ; | |
9a004508 | 984 | dip->i_depth = y; |
b3b94faa | 985 | |
539e5d6b | 986 | gfs2_dinode_out(dip, dibh->b_data); |
b3b94faa DT |
987 | |
988 | brelse(dibh); | |
989 | ||
990 | return 0; | |
991 | } | |
992 | ||
993 | /** | |
994 | * dir_split_leaf - Split a leaf block into two | |
995 | * @dip: The GFS2 inode | |
996 | * @index: | |
997 | * @leaf_no: | |
998 | * | |
999 | * Returns: 0 on success, error code on failure | |
1000 | */ | |
1001 | ||
c752666c | 1002 | static int dir_split_leaf(struct inode *inode, const struct qstr *name) |
b3b94faa | 1003 | { |
feaa7bba | 1004 | struct gfs2_inode *dip = GFS2_I(inode); |
b3b94faa DT |
1005 | struct buffer_head *nbh, *obh, *dibh; |
1006 | struct gfs2_leaf *nleaf, *oleaf; | |
4da3c646 | 1007 | struct gfs2_dirent *dent = NULL, *prev = NULL, *next = NULL, *new; |
cd915493 | 1008 | u32 start, len, half_len, divider; |
b44b84d7 AV |
1009 | u64 bn, leaf_no; |
1010 | __be64 *lp; | |
cd915493 | 1011 | u32 index; |
b3b94faa DT |
1012 | int x, moved = 0; |
1013 | int error; | |
1014 | ||
9a004508 | 1015 | index = name->hash >> (32 - dip->i_depth); |
c752666c | 1016 | error = get_leaf_nr(dip, index, &leaf_no); |
67893f12 | 1017 | if (IS_ERR_VALUE(error)) |
c752666c | 1018 | return error; |
b3b94faa DT |
1019 | |
1020 | /* Get the old leaf block */ | |
b3b94faa DT |
1021 | error = get_leaf(dip, leaf_no, &obh); |
1022 | if (error) | |
e90deff5 | 1023 | return error; |
b3b94faa | 1024 | |
b3b94faa | 1025 | oleaf = (struct gfs2_leaf *)obh->b_data; |
9a004508 | 1026 | if (dip->i_depth == be16_to_cpu(oleaf->lf_depth)) { |
e90deff5 SW |
1027 | brelse(obh); |
1028 | return 1; /* can't split */ | |
1029 | } | |
1030 | ||
350a9b0a | 1031 | gfs2_trans_add_meta(dip->i_gl, obh); |
b3b94faa | 1032 | |
c752666c SW |
1033 | nleaf = new_leaf(inode, &nbh, be16_to_cpu(oleaf->lf_depth) + 1); |
1034 | if (!nleaf) { | |
1035 | brelse(obh); | |
1036 | return -ENOSPC; | |
1037 | } | |
1038 | bn = nbh->b_blocknr; | |
b3b94faa | 1039 | |
c752666c | 1040 | /* Compute the start and len of leaf pointers in the hash table. */ |
9a004508 | 1041 | len = 1 << (dip->i_depth - be16_to_cpu(oleaf->lf_depth)); |
b3b94faa DT |
1042 | half_len = len >> 1; |
1043 | if (!half_len) { | |
d77d1b58 JP |
1044 | pr_warn("i_depth %u lf_depth %u index %u\n", |
1045 | dip->i_depth, be16_to_cpu(oleaf->lf_depth), index); | |
b3b94faa DT |
1046 | gfs2_consist_inode(dip); |
1047 | error = -EIO; | |
1048 | goto fail_brelse; | |
1049 | } | |
1050 | ||
1051 | start = (index & ~(len - 1)); | |
1052 | ||
1053 | /* Change the pointers. | |
1054 | Don't bother distinguishing stuffed from non-stuffed. | |
1055 | This code is complicated enough already. */ | |
4244b52e DR |
1056 | lp = kmalloc(half_len * sizeof(__be64), GFP_NOFS); |
1057 | if (!lp) { | |
1058 | error = -ENOMEM; | |
1059 | goto fail_brelse; | |
1060 | } | |
1061 | ||
b3b94faa | 1062 | /* Change the pointers */ |
b3b94faa DT |
1063 | for (x = 0; x < half_len; x++) |
1064 | lp[x] = cpu_to_be64(bn); | |
1065 | ||
17d539f0 SW |
1066 | gfs2_dir_hash_inval(dip); |
1067 | ||
cd915493 SW |
1068 | error = gfs2_dir_write_data(dip, (char *)lp, start * sizeof(u64), |
1069 | half_len * sizeof(u64)); | |
1070 | if (error != half_len * sizeof(u64)) { | |
b3b94faa DT |
1071 | if (error >= 0) |
1072 | error = -EIO; | |
1073 | goto fail_lpfree; | |
1074 | } | |
1075 | ||
1076 | kfree(lp); | |
1077 | ||
1078 | /* Compute the divider */ | |
9a004508 | 1079 | divider = (start + half_len) << (32 - dip->i_depth); |
b3b94faa DT |
1080 | |
1081 | /* Copy the entries */ | |
1579343a | 1082 | dent = (struct gfs2_dirent *)(obh->b_data + sizeof(struct gfs2_leaf)); |
b3b94faa DT |
1083 | |
1084 | do { | |
1085 | next = dent; | |
1086 | if (dirent_next(dip, obh, &next)) | |
1087 | next = NULL; | |
1088 | ||
5e7d65cd | 1089 | if (!gfs2_dirent_sentinel(dent) && |
b3b94faa | 1090 | be32_to_cpu(dent->de_hash) < divider) { |
c752666c | 1091 | struct qstr str; |
34017472 | 1092 | void *ptr = ((char *)dent - obh->b_data) + nbh->b_data; |
c752666c SW |
1093 | str.name = (char*)(dent+1); |
1094 | str.len = be16_to_cpu(dent->de_name_len); | |
1095 | str.hash = be32_to_cpu(dent->de_hash); | |
34017472 | 1096 | new = gfs2_dirent_split_alloc(inode, nbh, &str, ptr); |
c752666c SW |
1097 | if (IS_ERR(new)) { |
1098 | error = PTR_ERR(new); | |
1099 | break; | |
1100 | } | |
b3b94faa DT |
1101 | |
1102 | new->de_inum = dent->de_inum; /* No endian worries */ | |
b3b94faa | 1103 | new->de_type = dent->de_type; /* No endian worries */ |
bb16b342 | 1104 | be16_add_cpu(&nleaf->lf_entries, 1); |
b3b94faa DT |
1105 | |
1106 | dirent_del(dip, obh, prev, dent); | |
1107 | ||
1108 | if (!oleaf->lf_entries) | |
1109 | gfs2_consist_inode(dip); | |
bb16b342 | 1110 | be16_add_cpu(&oleaf->lf_entries, -1); |
b3b94faa DT |
1111 | |
1112 | if (!prev) | |
1113 | prev = dent; | |
1114 | ||
1115 | moved = 1; | |
c752666c | 1116 | } else { |
b3b94faa | 1117 | prev = dent; |
c752666c | 1118 | } |
b3b94faa | 1119 | dent = next; |
c752666c | 1120 | } while (dent); |
b3b94faa | 1121 | |
c752666c | 1122 | oleaf->lf_depth = nleaf->lf_depth; |
b3b94faa DT |
1123 | |
1124 | error = gfs2_meta_inode_buffer(dip, &dibh); | |
feaa7bba | 1125 | if (!gfs2_assert_withdraw(GFS2_SB(&dip->i_inode), !error)) { |
350a9b0a | 1126 | gfs2_trans_add_meta(dip->i_gl, dibh); |
77658aad | 1127 | gfs2_add_inode_blocks(&dip->i_inode, 1); |
539e5d6b | 1128 | gfs2_dinode_out(dip, dibh->b_data); |
b3b94faa DT |
1129 | brelse(dibh); |
1130 | } | |
1131 | ||
1132 | brelse(obh); | |
1133 | brelse(nbh); | |
1134 | ||
1135 | return error; | |
1136 | ||
e90deff5 | 1137 | fail_lpfree: |
b3b94faa DT |
1138 | kfree(lp); |
1139 | ||
e90deff5 | 1140 | fail_brelse: |
b3b94faa | 1141 | brelse(obh); |
b3b94faa DT |
1142 | brelse(nbh); |
1143 | return error; | |
1144 | } | |
1145 | ||
1146 | /** | |
1147 | * dir_double_exhash - Double size of ExHash table | |
1148 | * @dip: The GFS2 dinode | |
1149 | * | |
1150 | * Returns: 0 on success, error code on failure | |
1151 | */ | |
1152 | ||
1153 | static int dir_double_exhash(struct gfs2_inode *dip) | |
1154 | { | |
b3b94faa | 1155 | struct buffer_head *dibh; |
cd915493 | 1156 | u32 hsize; |
17d539f0 SW |
1157 | u32 hsize_bytes; |
1158 | __be64 *hc; | |
1159 | __be64 *hc2, *h; | |
b3b94faa DT |
1160 | int x; |
1161 | int error = 0; | |
1162 | ||
9a004508 | 1163 | hsize = 1 << dip->i_depth; |
17d539f0 | 1164 | hsize_bytes = hsize * sizeof(__be64); |
b3b94faa | 1165 | |
17d539f0 SW |
1166 | hc = gfs2_dir_get_hash_table(dip); |
1167 | if (IS_ERR(hc)) | |
1168 | return PTR_ERR(hc); | |
b3b94faa | 1169 | |
512cbf02 | 1170 | hc2 = kmalloc(hsize_bytes * 2, GFP_NOFS | __GFP_NOWARN); |
e8830d88 BP |
1171 | if (hc2 == NULL) |
1172 | hc2 = __vmalloc(hsize_bytes * 2, GFP_NOFS, PAGE_KERNEL); | |
1173 | ||
17d539f0 | 1174 | if (!hc2) |
4244b52e | 1175 | return -ENOMEM; |
b3b94faa | 1176 | |
512cbf02 | 1177 | h = hc2; |
17d539f0 SW |
1178 | error = gfs2_meta_inode_buffer(dip, &dibh); |
1179 | if (error) | |
1180 | goto out_kfree; | |
b3b94faa | 1181 | |
17d539f0 SW |
1182 | for (x = 0; x < hsize; x++) { |
1183 | *h++ = *hc; | |
1184 | *h++ = *hc; | |
1185 | hc++; | |
b3b94faa DT |
1186 | } |
1187 | ||
17d539f0 SW |
1188 | error = gfs2_dir_write_data(dip, (char *)hc2, 0, hsize_bytes * 2); |
1189 | if (error != (hsize_bytes * 2)) | |
1190 | goto fail; | |
b3b94faa | 1191 | |
17d539f0 SW |
1192 | gfs2_dir_hash_inval(dip); |
1193 | dip->i_hash_cache = hc2; | |
1194 | dip->i_depth++; | |
1195 | gfs2_dinode_out(dip, dibh->b_data); | |
1196 | brelse(dibh); | |
1197 | return 0; | |
b3b94faa | 1198 | |
a91ea69f | 1199 | fail: |
17d539f0 SW |
1200 | /* Replace original hash table & size */ |
1201 | gfs2_dir_write_data(dip, (char *)hc, 0, hsize_bytes); | |
1202 | i_size_write(&dip->i_inode, hsize_bytes); | |
1203 | gfs2_dinode_out(dip, dibh->b_data); | |
1204 | brelse(dibh); | |
1205 | out_kfree: | |
3cdcf63e | 1206 | kvfree(hc2); |
b3b94faa DT |
1207 | return error; |
1208 | } | |
1209 | ||
1210 | /** | |
1211 | * compare_dents - compare directory entries by hash value | |
1212 | * @a: first dent | |
1213 | * @b: second dent | |
1214 | * | |
1215 | * When comparing the hash entries of @a to @b: | |
1216 | * gt: returns 1 | |
1217 | * lt: returns -1 | |
1218 | * eq: returns 0 | |
1219 | */ | |
1220 | ||
1221 | static int compare_dents(const void *a, const void *b) | |
1222 | { | |
2bdbc5d7 | 1223 | const struct gfs2_dirent *dent_a, *dent_b; |
cd915493 | 1224 | u32 hash_a, hash_b; |
b3b94faa DT |
1225 | int ret = 0; |
1226 | ||
2bdbc5d7 | 1227 | dent_a = *(const struct gfs2_dirent **)a; |
471f3db2 | 1228 | hash_a = dent_a->de_cookie; |
b3b94faa | 1229 | |
2bdbc5d7 | 1230 | dent_b = *(const struct gfs2_dirent **)b; |
471f3db2 | 1231 | hash_b = dent_b->de_cookie; |
b3b94faa DT |
1232 | |
1233 | if (hash_a > hash_b) | |
1234 | ret = 1; | |
1235 | else if (hash_a < hash_b) | |
1236 | ret = -1; | |
1237 | else { | |
4dd651ad SW |
1238 | unsigned int len_a = be16_to_cpu(dent_a->de_name_len); |
1239 | unsigned int len_b = be16_to_cpu(dent_b->de_name_len); | |
b3b94faa DT |
1240 | |
1241 | if (len_a > len_b) | |
1242 | ret = 1; | |
1243 | else if (len_a < len_b) | |
1244 | ret = -1; | |
1245 | else | |
2bdbc5d7 | 1246 | ret = memcmp(dent_a + 1, dent_b + 1, len_a); |
b3b94faa DT |
1247 | } |
1248 | ||
1249 | return ret; | |
1250 | } | |
1251 | ||
1252 | /** | |
1253 | * do_filldir_main - read out directory entries | |
1254 | * @dip: The GFS2 inode | |
d81a8ef5 | 1255 | * @ctx: what to feed the entries to |
b3b94faa DT |
1256 | * @darr: an array of struct gfs2_dirent pointers to read |
1257 | * @entries: the number of entries in darr | |
1258 | * @copied: pointer to int that's non-zero if a entry has been copied out | |
1259 | * | |
1260 | * Jump through some hoops to make sure that if there are hash collsions, | |
1261 | * they are read out at the beginning of a buffer. We want to minimize | |
1262 | * the possibility that they will fall into different readdir buffers or | |
1263 | * that someone will want to seek to that location. | |
1264 | * | |
d81a8ef5 | 1265 | * Returns: errno, >0 if the actor tells you to stop |
b3b94faa DT |
1266 | */ |
1267 | ||
d81a8ef5 | 1268 | static int do_filldir_main(struct gfs2_inode *dip, struct dir_context *ctx, |
471f3db2 BM |
1269 | struct gfs2_dirent **darr, u32 entries, |
1270 | u32 sort_start, int *copied) | |
b3b94faa | 1271 | { |
71b86f56 | 1272 | const struct gfs2_dirent *dent, *dent_next; |
cd915493 | 1273 | u64 off, off_next; |
b3b94faa DT |
1274 | unsigned int x, y; |
1275 | int run = 0; | |
b3b94faa | 1276 | |
471f3db2 BM |
1277 | if (sort_start < entries) |
1278 | sort(&darr[sort_start], entries - sort_start, | |
1279 | sizeof(struct gfs2_dirent *), compare_dents, NULL); | |
b3b94faa DT |
1280 | |
1281 | dent_next = darr[0]; | |
471f3db2 | 1282 | off_next = dent_next->de_cookie; |
b3b94faa DT |
1283 | |
1284 | for (x = 0, y = 1; x < entries; x++, y++) { | |
1285 | dent = dent_next; | |
1286 | off = off_next; | |
1287 | ||
1288 | if (y < entries) { | |
1289 | dent_next = darr[y]; | |
471f3db2 | 1290 | off_next = dent_next->de_cookie; |
b3b94faa | 1291 | |
d81a8ef5 | 1292 | if (off < ctx->pos) |
b3b94faa | 1293 | continue; |
d81a8ef5 | 1294 | ctx->pos = off; |
b3b94faa DT |
1295 | |
1296 | if (off_next == off) { | |
1297 | if (*copied && !run) | |
1298 | return 1; | |
1299 | run = 1; | |
1300 | } else | |
1301 | run = 0; | |
1302 | } else { | |
d81a8ef5 | 1303 | if (off < ctx->pos) |
b3b94faa | 1304 | continue; |
d81a8ef5 | 1305 | ctx->pos = off; |
b3b94faa DT |
1306 | } |
1307 | ||
d81a8ef5 | 1308 | if (!dir_emit(ctx, (const char *)(dent + 1), |
4dd651ad | 1309 | be16_to_cpu(dent->de_name_len), |
d81a8ef5 AV |
1310 | be64_to_cpu(dent->de_inum.no_addr), |
1311 | be16_to_cpu(dent->de_type))) | |
b3b94faa DT |
1312 | return 1; |
1313 | ||
1314 | *copied = 1; | |
1315 | } | |
1316 | ||
d81a8ef5 | 1317 | /* Increment the ctx->pos by one, so the next time we come into the |
b3b94faa DT |
1318 | do_filldir fxn, we get the next entry instead of the last one in the |
1319 | current leaf */ | |
1320 | ||
d81a8ef5 | 1321 | ctx->pos++; |
b3b94faa DT |
1322 | |
1323 | return 0; | |
1324 | } | |
1325 | ||
d2a97a4e SW |
1326 | static void *gfs2_alloc_sort_buffer(unsigned size) |
1327 | { | |
1328 | void *ptr = NULL; | |
1329 | ||
1330 | if (size < KMALLOC_MAX_SIZE) | |
1331 | ptr = kmalloc(size, GFP_NOFS | __GFP_NOWARN); | |
1332 | if (!ptr) | |
1333 | ptr = __vmalloc(size, GFP_NOFS, PAGE_KERNEL); | |
1334 | return ptr; | |
1335 | } | |
1336 | ||
471f3db2 BM |
1337 | |
1338 | static int gfs2_set_cookies(struct gfs2_sbd *sdp, struct buffer_head *bh, | |
1339 | unsigned leaf_nr, struct gfs2_dirent **darr, | |
1340 | unsigned entries) | |
1341 | { | |
1342 | int sort_id = -1; | |
1343 | int i; | |
1344 | ||
1345 | for (i = 0; i < entries; i++) { | |
1346 | unsigned offset; | |
1347 | ||
1348 | darr[i]->de_cookie = be32_to_cpu(darr[i]->de_hash); | |
1349 | darr[i]->de_cookie = gfs2_disk_hash2offset(darr[i]->de_cookie); | |
1350 | ||
1351 | if (!sdp->sd_args.ar_loccookie) | |
1352 | continue; | |
1353 | offset = (char *)(darr[i]) - | |
1354 | (bh->b_data + gfs2_dirent_offset(bh->b_data)); | |
1355 | offset /= GFS2_MIN_DIRENT_SIZE; | |
1356 | offset += leaf_nr * sdp->sd_max_dents_per_leaf; | |
1357 | if (offset >= GFS2_USE_HASH_FLAG || | |
1358 | leaf_nr >= GFS2_USE_HASH_FLAG) { | |
1359 | darr[i]->de_cookie |= GFS2_USE_HASH_FLAG; | |
1360 | if (sort_id < 0) | |
1361 | sort_id = i; | |
1362 | continue; | |
1363 | } | |
1364 | darr[i]->de_cookie &= GFS2_HASH_INDEX_MASK; | |
1365 | darr[i]->de_cookie |= offset; | |
1366 | } | |
1367 | return sort_id; | |
1368 | } | |
1369 | ||
1370 | ||
d81a8ef5 AV |
1371 | static int gfs2_dir_read_leaf(struct inode *inode, struct dir_context *ctx, |
1372 | int *copied, unsigned *depth, | |
3699e3a4 | 1373 | u64 leaf_no) |
b3b94faa | 1374 | { |
feaa7bba | 1375 | struct gfs2_inode *ip = GFS2_I(inode); |
bdd19a22 | 1376 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
71b86f56 SW |
1377 | struct buffer_head *bh; |
1378 | struct gfs2_leaf *lf; | |
bdd19a22 | 1379 | unsigned entries = 0, entries2 = 0; |
471f3db2 BM |
1380 | unsigned leaves = 0, leaf = 0, offset, sort_offset; |
1381 | struct gfs2_dirent **darr, *dent; | |
71b86f56 SW |
1382 | struct dirent_gather g; |
1383 | struct buffer_head **larr; | |
471f3db2 | 1384 | int error, i, need_sort = 0, sort_id; |
71b86f56 | 1385 | u64 lfn = leaf_no; |
b3b94faa | 1386 | |
b3b94faa | 1387 | do { |
71b86f56 | 1388 | error = get_leaf(ip, lfn, &bh); |
b3b94faa | 1389 | if (error) |
71b86f56 SW |
1390 | goto out; |
1391 | lf = (struct gfs2_leaf *)bh->b_data; | |
1392 | if (leaves == 0) | |
1393 | *depth = be16_to_cpu(lf->lf_depth); | |
1394 | entries += be16_to_cpu(lf->lf_entries); | |
1395 | leaves++; | |
1396 | lfn = be64_to_cpu(lf->lf_next); | |
1397 | brelse(bh); | |
1398 | } while(lfn); | |
b3b94faa | 1399 | |
471f3db2 BM |
1400 | if (*depth < GFS2_DIR_MAX_DEPTH || !sdp->sd_args.ar_loccookie) { |
1401 | need_sort = 1; | |
1402 | sort_offset = 0; | |
1403 | } | |
1404 | ||
b3b94faa DT |
1405 | if (!entries) |
1406 | return 0; | |
1407 | ||
71b86f56 | 1408 | error = -ENOMEM; |
bdd19a22 SW |
1409 | /* |
1410 | * The extra 99 entries are not normally used, but are a buffer | |
1411 | * zone in case the number of entries in the leaf is corrupt. | |
1412 | * 99 is the maximum number of entries that can fit in a single | |
1413 | * leaf block. | |
1414 | */ | |
d2a97a4e | 1415 | larr = gfs2_alloc_sort_buffer((leaves + entries + 99) * sizeof(void *)); |
71b86f56 SW |
1416 | if (!larr) |
1417 | goto out; | |
471f3db2 BM |
1418 | darr = (struct gfs2_dirent **)(larr + leaves); |
1419 | g.pdent = (const struct gfs2_dirent **)darr; | |
71b86f56 SW |
1420 | g.offset = 0; |
1421 | lfn = leaf_no; | |
b3b94faa | 1422 | |
71b86f56 SW |
1423 | do { |
1424 | error = get_leaf(ip, lfn, &bh); | |
b3b94faa | 1425 | if (error) |
d2a97a4e | 1426 | goto out_free; |
71b86f56 SW |
1427 | lf = (struct gfs2_leaf *)bh->b_data; |
1428 | lfn = be64_to_cpu(lf->lf_next); | |
1429 | if (lf->lf_entries) { | |
471f3db2 | 1430 | offset = g.offset; |
bdd19a22 | 1431 | entries2 += be16_to_cpu(lf->lf_entries); |
71b86f56 SW |
1432 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, |
1433 | gfs2_dirent_gather, NULL, &g); | |
1434 | error = PTR_ERR(dent); | |
bdd19a22 | 1435 | if (IS_ERR(dent)) |
d2a97a4e | 1436 | goto out_free; |
bdd19a22 | 1437 | if (entries2 != g.offset) { |
f391a4ea | 1438 | fs_warn(sdp, "Number of entries corrupt in dir " |
1439 | "leaf %llu, entries2 (%u) != " | |
1440 | "g.offset (%u)\n", | |
1441 | (unsigned long long)bh->b_blocknr, | |
1442 | entries2, g.offset); | |
bdd19a22 SW |
1443 | |
1444 | error = -EIO; | |
d2a97a4e | 1445 | goto out_free; |
71b86f56 SW |
1446 | } |
1447 | error = 0; | |
471f3db2 BM |
1448 | sort_id = gfs2_set_cookies(sdp, bh, leaf, &darr[offset], |
1449 | be16_to_cpu(lf->lf_entries)); | |
1450 | if (!need_sort && sort_id >= 0) { | |
1451 | need_sort = 1; | |
1452 | sort_offset = offset + sort_id; | |
1453 | } | |
71b86f56 | 1454 | larr[leaf++] = bh; |
b3b94faa | 1455 | } else { |
471f3db2 | 1456 | larr[leaf++] = NULL; |
71b86f56 | 1457 | brelse(bh); |
b3b94faa | 1458 | } |
71b86f56 | 1459 | } while(lfn); |
b3b94faa | 1460 | |
bdd19a22 | 1461 | BUG_ON(entries2 != entries); |
471f3db2 BM |
1462 | error = do_filldir_main(ip, ctx, darr, entries, need_sort ? |
1463 | sort_offset : entries, copied); | |
d2a97a4e | 1464 | out_free: |
71b86f56 | 1465 | for(i = 0; i < leaf; i++) |
471f3db2 BM |
1466 | if (larr[i]) |
1467 | brelse(larr[i]); | |
3cdcf63e | 1468 | kvfree(larr); |
71b86f56 | 1469 | out: |
b3b94faa DT |
1470 | return error; |
1471 | } | |
1472 | ||
79c4c379 SW |
1473 | /** |
1474 | * gfs2_dir_readahead - Issue read-ahead requests for leaf blocks. | |
dfe4d34b BP |
1475 | * |
1476 | * Note: we can't calculate each index like dir_e_read can because we don't | |
1477 | * have the leaf, and therefore we don't have the depth, and therefore we | |
1478 | * don't have the length. So we have to just read enough ahead to make up | |
79c4c379 SW |
1479 | * for the loss of information. |
1480 | */ | |
dfe4d34b BP |
1481 | static void gfs2_dir_readahead(struct inode *inode, unsigned hsize, u32 index, |
1482 | struct file_ra_state *f_ra) | |
1483 | { | |
1484 | struct gfs2_inode *ip = GFS2_I(inode); | |
1485 | struct gfs2_glock *gl = ip->i_gl; | |
1486 | struct buffer_head *bh; | |
1487 | u64 blocknr = 0, last; | |
1488 | unsigned count; | |
1489 | ||
1490 | /* First check if we've already read-ahead for the whole range. */ | |
79c4c379 | 1491 | if (index + MAX_RA_BLOCKS < f_ra->start) |
dfe4d34b BP |
1492 | return; |
1493 | ||
1494 | f_ra->start = max((pgoff_t)index, f_ra->start); | |
1495 | for (count = 0; count < MAX_RA_BLOCKS; count++) { | |
1496 | if (f_ra->start >= hsize) /* if exceeded the hash table */ | |
1497 | break; | |
1498 | ||
1499 | last = blocknr; | |
1500 | blocknr = be64_to_cpu(ip->i_hash_cache[f_ra->start]); | |
1501 | f_ra->start++; | |
1502 | if (blocknr == last) | |
1503 | continue; | |
1504 | ||
1505 | bh = gfs2_getbuf(gl, blocknr, 1); | |
1506 | if (trylock_buffer(bh)) { | |
1507 | if (buffer_uptodate(bh)) { | |
1508 | unlock_buffer(bh); | |
1509 | brelse(bh); | |
1510 | continue; | |
1511 | } | |
1512 | bh->b_end_io = end_buffer_read_sync; | |
1513 | submit_bh(READA | REQ_META, bh); | |
1514 | continue; | |
1515 | } | |
1516 | brelse(bh); | |
1517 | } | |
1518 | } | |
17d539f0 | 1519 | |
b3b94faa | 1520 | /** |
c752666c SW |
1521 | * dir_e_read - Reads the entries from a directory into a filldir buffer |
1522 | * @dip: dinode pointer | |
d81a8ef5 | 1523 | * @ctx: actor to feed the entries to |
b3b94faa | 1524 | * |
c752666c | 1525 | * Returns: errno |
b3b94faa DT |
1526 | */ |
1527 | ||
d81a8ef5 AV |
1528 | static int dir_e_read(struct inode *inode, struct dir_context *ctx, |
1529 | struct file_ra_state *f_ra) | |
b3b94faa | 1530 | { |
feaa7bba | 1531 | struct gfs2_inode *dip = GFS2_I(inode); |
cd915493 | 1532 | u32 hsize, len = 0; |
cd915493 | 1533 | u32 hash, index; |
b44b84d7 | 1534 | __be64 *lp; |
c752666c SW |
1535 | int copied = 0; |
1536 | int error = 0; | |
4da3c646 | 1537 | unsigned depth = 0; |
b3b94faa | 1538 | |
9a004508 | 1539 | hsize = 1 << dip->i_depth; |
d81a8ef5 | 1540 | hash = gfs2_dir_offset2hash(ctx->pos); |
9a004508 | 1541 | index = hash >> (32 - dip->i_depth); |
b3b94faa | 1542 | |
79c4c379 | 1543 | if (dip->i_hash_cache == NULL) |
dfe4d34b | 1544 | f_ra->start = 0; |
17d539f0 SW |
1545 | lp = gfs2_dir_get_hash_table(dip); |
1546 | if (IS_ERR(lp)) | |
1547 | return PTR_ERR(lp); | |
b3b94faa | 1548 | |
dfe4d34b BP |
1549 | gfs2_dir_readahead(inode, hsize, index, f_ra); |
1550 | ||
b3b94faa | 1551 | while (index < hsize) { |
d81a8ef5 | 1552 | error = gfs2_dir_read_leaf(inode, ctx, |
71b86f56 | 1553 | &copied, &depth, |
17d539f0 | 1554 | be64_to_cpu(lp[index])); |
b3b94faa | 1555 | if (error) |
71b86f56 | 1556 | break; |
b3b94faa | 1557 | |
9a004508 | 1558 | len = 1 << (dip->i_depth - depth); |
b3b94faa DT |
1559 | index = (index & ~(len - 1)) + len; |
1560 | } | |
1561 | ||
71b86f56 SW |
1562 | if (error > 0) |
1563 | error = 0; | |
b3b94faa DT |
1564 | return error; |
1565 | } | |
1566 | ||
d81a8ef5 AV |
1567 | int gfs2_dir_read(struct inode *inode, struct dir_context *ctx, |
1568 | struct file_ra_state *f_ra) | |
b3b94faa | 1569 | { |
feaa7bba | 1570 | struct gfs2_inode *dip = GFS2_I(inode); |
bdd19a22 | 1571 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
71b86f56 | 1572 | struct dirent_gather g; |
471f3db2 | 1573 | struct gfs2_dirent **darr, *dent; |
b3b94faa DT |
1574 | struct buffer_head *dibh; |
1575 | int copied = 0; | |
1576 | int error; | |
1577 | ||
ad6203f2 | 1578 | if (!dip->i_entries) |
71b86f56 SW |
1579 | return 0; |
1580 | ||
383f01fb | 1581 | if (dip->i_diskflags & GFS2_DIF_EXHASH) |
d81a8ef5 | 1582 | return dir_e_read(inode, ctx, f_ra); |
71b86f56 | 1583 | |
b3b94faa DT |
1584 | if (!gfs2_is_stuffed(dip)) { |
1585 | gfs2_consist_inode(dip); | |
1586 | return -EIO; | |
1587 | } | |
1588 | ||
b3b94faa DT |
1589 | error = gfs2_meta_inode_buffer(dip, &dibh); |
1590 | if (error) | |
1591 | return error; | |
1592 | ||
71b86f56 | 1593 | error = -ENOMEM; |
bdd19a22 | 1594 | /* 96 is max number of dirents which can be stuffed into an inode */ |
16c5f06f | 1595 | darr = kmalloc(96 * sizeof(struct gfs2_dirent *), GFP_NOFS); |
71b86f56 | 1596 | if (darr) { |
471f3db2 | 1597 | g.pdent = (const struct gfs2_dirent **)darr; |
71b86f56 SW |
1598 | g.offset = 0; |
1599 | dent = gfs2_dirent_scan(inode, dibh->b_data, dibh->b_size, | |
1600 | gfs2_dirent_gather, NULL, &g); | |
1601 | if (IS_ERR(dent)) { | |
1602 | error = PTR_ERR(dent); | |
1603 | goto out; | |
1604 | } | |
ad6203f2 | 1605 | if (dip->i_entries != g.offset) { |
bdd19a22 | 1606 | fs_warn(sdp, "Number of entries corrupt in dir %llu, " |
ad6203f2 | 1607 | "ip->i_entries (%u) != g.offset (%u)\n", |
dbb7cae2 | 1608 | (unsigned long long)dip->i_no_addr, |
ad6203f2 | 1609 | dip->i_entries, |
bdd19a22 SW |
1610 | g.offset); |
1611 | error = -EIO; | |
1612 | goto out; | |
1613 | } | |
471f3db2 | 1614 | gfs2_set_cookies(sdp, dibh, 0, darr, dip->i_entries); |
d81a8ef5 | 1615 | error = do_filldir_main(dip, ctx, darr, |
471f3db2 | 1616 | dip->i_entries, 0, &copied); |
71b86f56 SW |
1617 | out: |
1618 | kfree(darr); | |
1619 | } | |
1620 | ||
b3b94faa DT |
1621 | if (error > 0) |
1622 | error = 0; | |
1623 | ||
1624 | brelse(dibh); | |
1625 | ||
1626 | return error; | |
1627 | } | |
1628 | ||
b3b94faa DT |
1629 | /** |
1630 | * gfs2_dir_search - Search a directory | |
5a00f3cc SW |
1631 | * @dip: The GFS2 dir inode |
1632 | * @name: The name we are looking up | |
1633 | * @fail_on_exist: Fail if the name exists rather than looking it up | |
b3b94faa DT |
1634 | * |
1635 | * This routine searches a directory for a file or another directory. | |
1636 | * Assumes a glock is held on dip. | |
1637 | * | |
1638 | * Returns: errno | |
1639 | */ | |
1640 | ||
5a00f3cc SW |
1641 | struct inode *gfs2_dir_search(struct inode *dir, const struct qstr *name, |
1642 | bool fail_on_exist) | |
b3b94faa | 1643 | { |
c752666c SW |
1644 | struct buffer_head *bh; |
1645 | struct gfs2_dirent *dent; | |
5a00f3cc SW |
1646 | u64 addr, formal_ino; |
1647 | u16 dtype; | |
dbb7cae2 SW |
1648 | |
1649 | dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh); | |
1650 | if (dent) { | |
c8d57703 AG |
1651 | struct inode *inode; |
1652 | u16 rahead; | |
1653 | ||
dbb7cae2 | 1654 | if (IS_ERR(dent)) |
e231c2ee | 1655 | return ERR_CAST(dent); |
5a00f3cc | 1656 | dtype = be16_to_cpu(dent->de_type); |
c8d57703 | 1657 | rahead = be16_to_cpu(dent->de_rahead); |
5a00f3cc SW |
1658 | addr = be64_to_cpu(dent->de_inum.no_addr); |
1659 | formal_ino = be64_to_cpu(dent->de_inum.no_formal_ino); | |
dbb7cae2 | 1660 | brelse(bh); |
5a00f3cc SW |
1661 | if (fail_on_exist) |
1662 | return ERR_PTR(-EEXIST); | |
73b462d2 | 1663 | inode = gfs2_inode_lookup(dir->i_sb, dtype, addr, formal_ino); |
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; | |
01bcb0de | 1799 | struct timespec 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)); |
01bcb0de | 1815 | tv = CURRENT_TIME; |
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; | |
01bcb0de | 1877 | struct timespec tv = CURRENT_TIME; |
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 DT |
1937 | int error; |
1938 | ||
feaa7bba | 1939 | dent = gfs2_dirent_search(&dip->i_inode, filename, gfs2_dirent_find, &bh); |
c752666c SW |
1940 | if (!dent) { |
1941 | gfs2_consist_inode(dip); | |
1942 | return -EIO; | |
1943 | } | |
1944 | if (IS_ERR(dent)) | |
1945 | return PTR_ERR(dent); | |
b3b94faa | 1946 | |
350a9b0a | 1947 | gfs2_trans_add_meta(dip->i_gl, bh); |
dbb7cae2 | 1948 | gfs2_inum_out(nip, dent); |
c752666c SW |
1949 | dent->de_type = cpu_to_be16(new_type); |
1950 | ||
383f01fb | 1951 | if (dip->i_diskflags & GFS2_DIF_EXHASH) { |
c752666c SW |
1952 | brelse(bh); |
1953 | error = gfs2_meta_inode_buffer(dip, &bh); | |
1954 | if (error) | |
1955 | return error; | |
350a9b0a | 1956 | gfs2_trans_add_meta(dip->i_gl, bh); |
c752666c SW |
1957 | } |
1958 | ||
4bd91ba1 | 1959 | dip->i_inode.i_mtime = dip->i_inode.i_ctime = CURRENT_TIME; |
539e5d6b | 1960 | gfs2_dinode_out(dip, bh->b_data); |
c752666c SW |
1961 | brelse(bh); |
1962 | return 0; | |
b3b94faa DT |
1963 | } |
1964 | ||
b3b94faa DT |
1965 | /** |
1966 | * leaf_dealloc - Deallocate a directory leaf | |
1967 | * @dip: the directory | |
1968 | * @index: the hash table offset in the directory | |
1969 | * @len: the number of pointers to this leaf | |
1970 | * @leaf_no: the leaf number | |
ec038c82 | 1971 | * @leaf_bh: buffer_head for the starting leaf |
d24a7a43 | 1972 | * last_dealloc: 1 if this is the final dealloc for the leaf, else 0 |
b3b94faa DT |
1973 | * |
1974 | * Returns: errno | |
1975 | */ | |
1976 | ||
cd915493 | 1977 | static int leaf_dealloc(struct gfs2_inode *dip, u32 index, u32 len, |
ec038c82 BP |
1978 | u64 leaf_no, struct buffer_head *leaf_bh, |
1979 | int last_dealloc) | |
b3b94faa | 1980 | { |
feaa7bba | 1981 | struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); |
c752666c | 1982 | struct gfs2_leaf *tmp_leaf; |
b3b94faa DT |
1983 | struct gfs2_rgrp_list rlist; |
1984 | struct buffer_head *bh, *dibh; | |
cd915493 | 1985 | u64 blk, nblk; |
b3b94faa DT |
1986 | unsigned int rg_blocks = 0, l_blocks = 0; |
1987 | char *ht; | |
cd915493 | 1988 | unsigned int x, size = len * sizeof(u64); |
b3b94faa DT |
1989 | int error; |
1990 | ||
5e2f7d61 BP |
1991 | error = gfs2_rindex_update(sdp); |
1992 | if (error) | |
1993 | return error; | |
1994 | ||
b3b94faa DT |
1995 | memset(&rlist, 0, sizeof(struct gfs2_rgrp_list)); |
1996 | ||
8be04b93 | 1997 | ht = kzalloc(size, GFP_NOFS | __GFP_NOWARN); |
e8830d88 | 1998 | if (ht == NULL) |
7456a37d OD |
1999 | ht = __vmalloc(size, GFP_NOFS | __GFP_NOWARN | __GFP_ZERO, |
2000 | PAGE_KERNEL); | |
b3b94faa DT |
2001 | if (!ht) |
2002 | return -ENOMEM; | |
2003 | ||
f4108a60 | 2004 | error = gfs2_quota_hold(dip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE); |
b3b94faa | 2005 | if (error) |
5407e242 | 2006 | goto out; |
b3b94faa | 2007 | |
b3b94faa | 2008 | /* Count the number of leaves */ |
ec038c82 | 2009 | bh = leaf_bh; |
b3b94faa | 2010 | |
c752666c | 2011 | for (blk = leaf_no; blk; blk = nblk) { |
ec038c82 BP |
2012 | if (blk != leaf_no) { |
2013 | error = get_leaf(dip, blk, &bh); | |
2014 | if (error) | |
2015 | goto out_rlist; | |
2016 | } | |
c752666c SW |
2017 | tmp_leaf = (struct gfs2_leaf *)bh->b_data; |
2018 | nblk = be64_to_cpu(tmp_leaf->lf_next); | |
ec038c82 BP |
2019 | if (blk != leaf_no) |
2020 | brelse(bh); | |
b3b94faa | 2021 | |
70b0c365 | 2022 | gfs2_rlist_add(dip, &rlist, blk); |
b3b94faa DT |
2023 | l_blocks++; |
2024 | } | |
2025 | ||
fe6c991c | 2026 | gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE); |
b3b94faa DT |
2027 | |
2028 | for (x = 0; x < rlist.rl_rgrps; x++) { | |
2029 | struct gfs2_rgrpd *rgd; | |
5c676f6d | 2030 | rgd = rlist.rl_ghs[x].gh_gl->gl_object; |
bb8d8a6f | 2031 | rg_blocks += rgd->rd_length; |
b3b94faa DT |
2032 | } |
2033 | ||
2034 | error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs); | |
2035 | if (error) | |
2036 | goto out_rlist; | |
2037 | ||
2038 | error = gfs2_trans_begin(sdp, | |
5c676f6d | 2039 | rg_blocks + (DIV_ROUND_UP(size, sdp->sd_jbsize) + 1) + |
b3b94faa DT |
2040 | RES_DINODE + RES_STATFS + RES_QUOTA, l_blocks); |
2041 | if (error) | |
2042 | goto out_rg_gunlock; | |
2043 | ||
ec038c82 BP |
2044 | bh = leaf_bh; |
2045 | ||
c752666c | 2046 | for (blk = leaf_no; blk; blk = nblk) { |
ec038c82 BP |
2047 | if (blk != leaf_no) { |
2048 | error = get_leaf(dip, blk, &bh); | |
2049 | if (error) | |
2050 | goto out_end_trans; | |
2051 | } | |
c752666c SW |
2052 | tmp_leaf = (struct gfs2_leaf *)bh->b_data; |
2053 | nblk = be64_to_cpu(tmp_leaf->lf_next); | |
ec038c82 BP |
2054 | if (blk != leaf_no) |
2055 | brelse(bh); | |
b3b94faa DT |
2056 | |
2057 | gfs2_free_meta(dip, blk, 1); | |
77658aad | 2058 | gfs2_add_inode_blocks(&dip->i_inode, -1); |
b3b94faa DT |
2059 | } |
2060 | ||
cd915493 | 2061 | error = gfs2_dir_write_data(dip, ht, index * sizeof(u64), size); |
b3b94faa DT |
2062 | if (error != size) { |
2063 | if (error >= 0) | |
2064 | error = -EIO; | |
2065 | goto out_end_trans; | |
2066 | } | |
2067 | ||
2068 | error = gfs2_meta_inode_buffer(dip, &dibh); | |
2069 | if (error) | |
2070 | goto out_end_trans; | |
2071 | ||
350a9b0a | 2072 | gfs2_trans_add_meta(dip->i_gl, dibh); |
d24a7a43 BP |
2073 | /* On the last dealloc, make this a regular file in case we crash. |
2074 | (We don't want to free these blocks a second time.) */ | |
2075 | if (last_dealloc) | |
2076 | dip->i_inode.i_mode = S_IFREG; | |
539e5d6b | 2077 | gfs2_dinode_out(dip, dibh->b_data); |
b3b94faa DT |
2078 | brelse(dibh); |
2079 | ||
a91ea69f | 2080 | out_end_trans: |
b3b94faa | 2081 | gfs2_trans_end(sdp); |
a91ea69f | 2082 | out_rg_gunlock: |
b3b94faa | 2083 | gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs); |
a91ea69f | 2084 | out_rlist: |
b3b94faa | 2085 | gfs2_rlist_free(&rlist); |
b3b94faa | 2086 | gfs2_quota_unhold(dip); |
182fe5ab | 2087 | out: |
3cdcf63e | 2088 | kvfree(ht); |
b3b94faa DT |
2089 | return error; |
2090 | } | |
2091 | ||
2092 | /** | |
2093 | * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory | |
2094 | * @dip: the directory | |
2095 | * | |
2096 | * Dealloc all on-disk directory leaves to FREEMETA state | |
2097 | * Change on-disk inode type to "regular file" | |
2098 | * | |
2099 | * Returns: errno | |
2100 | */ | |
2101 | ||
2102 | int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip) | |
2103 | { | |
556bb179 BP |
2104 | struct buffer_head *bh; |
2105 | struct gfs2_leaf *leaf; | |
2106 | u32 hsize, len; | |
556bb179 BP |
2107 | u32 index = 0, next_index; |
2108 | __be64 *lp; | |
2109 | u64 leaf_no; | |
2110 | int error = 0, last; | |
2111 | ||
2112 | hsize = 1 << dip->i_depth; | |
556bb179 | 2113 | |
17d539f0 SW |
2114 | lp = gfs2_dir_get_hash_table(dip); |
2115 | if (IS_ERR(lp)) | |
2116 | return PTR_ERR(lp); | |
556bb179 BP |
2117 | |
2118 | while (index < hsize) { | |
17d539f0 | 2119 | leaf_no = be64_to_cpu(lp[index]); |
556bb179 BP |
2120 | if (leaf_no) { |
2121 | error = get_leaf(dip, leaf_no, &bh); | |
2122 | if (error) | |
2123 | goto out; | |
2124 | leaf = (struct gfs2_leaf *)bh->b_data; | |
2125 | len = 1 << (dip->i_depth - be16_to_cpu(leaf->lf_depth)); | |
2126 | ||
2127 | next_index = (index & ~(len - 1)) + len; | |
2128 | last = ((next_index >= hsize) ? 1 : 0); | |
2129 | error = leaf_dealloc(dip, index, len, leaf_no, bh, | |
2130 | last); | |
2131 | brelse(bh); | |
2132 | if (error) | |
2133 | goto out; | |
2134 | index = next_index; | |
2135 | } else | |
2136 | index++; | |
2137 | } | |
2138 | ||
2139 | if (index != hsize) { | |
2140 | gfs2_consist_inode(dip); | |
2141 | error = -EIO; | |
2142 | } | |
2143 | ||
2144 | out: | |
556bb179 BP |
2145 | |
2146 | return error; | |
b3b94faa DT |
2147 | } |
2148 | ||
2149 | /** | |
2150 | * gfs2_diradd_alloc_required - find if adding entry will require an allocation | |
2151 | * @ip: the file being written to | |
2152 | * @filname: the filename that's going to be added | |
3c1c0ae1 | 2153 | * @da: The structure to return dir alloc info |
b3b94faa | 2154 | * |
3c1c0ae1 | 2155 | * Returns: 0 if ok, -ve on error |
b3b94faa DT |
2156 | */ |
2157 | ||
3c1c0ae1 SW |
2158 | int gfs2_diradd_alloc_required(struct inode *inode, const struct qstr *name, |
2159 | struct gfs2_diradd *da) | |
b3b94faa | 2160 | { |
22b5a6c0 | 2161 | struct gfs2_inode *ip = GFS2_I(inode); |
3c1c0ae1 | 2162 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
22b5a6c0 | 2163 | const unsigned int extra = sizeof(struct gfs2_dinode) - sizeof(struct gfs2_leaf); |
c752666c SW |
2164 | struct gfs2_dirent *dent; |
2165 | struct buffer_head *bh; | |
b3b94faa | 2166 | |
3c1c0ae1 | 2167 | da->nr_blocks = 0; |
2b47dad8 SW |
2168 | da->bh = NULL; |
2169 | da->dent = NULL; | |
3c1c0ae1 | 2170 | |
c752666c | 2171 | dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, &bh); |
ed386507 | 2172 | if (!dent) { |
3c1c0ae1 | 2173 | da->nr_blocks = sdp->sd_max_dirres; |
22b5a6c0 SW |
2174 | if (!(ip->i_diskflags & GFS2_DIF_EXHASH) && |
2175 | (GFS2_DIRENT_SIZE(name->len) < extra)) | |
2176 | da->nr_blocks = 1; | |
3c1c0ae1 | 2177 | return 0; |
ed386507 | 2178 | } |
c752666c SW |
2179 | if (IS_ERR(dent)) |
2180 | return PTR_ERR(dent); | |
19aeb5a6 BP |
2181 | |
2182 | if (da->save_loc) { | |
2183 | da->bh = bh; | |
2184 | da->dent = dent; | |
2185 | } else { | |
2186 | brelse(bh); | |
2187 | } | |
c752666c | 2188 | return 0; |
b3b94faa DT |
2189 | } |
2190 |