]>
Commit | Line | Data |
---|---|---|
b3b94faa DT |
1 | /* |
2 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
3 | * Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved. | |
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 | |
7 | * of the GNU General Public License v.2. | |
8 | */ | |
9 | ||
10 | /* | |
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 | * | |
39 | * dip->i_di.di_flags & GFS2_DIF_EXHASH is true | |
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 block | |
46 | * pointer in the array that points to the same leaf. In fact, when a directory | |
47 | * is first converted from linear to exhash, all of the pointers point to the | |
48 | * 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 | */ | |
55 | ||
56 | #include <linux/sched.h> | |
57 | #include <linux/slab.h> | |
58 | #include <linux/spinlock.h> | |
59 | #include <linux/completion.h> | |
60 | #include <linux/buffer_head.h> | |
61 | #include <linux/sort.h> | |
5c676f6d | 62 | #include <linux/gfs2_ondisk.h> |
b3b94faa DT |
63 | #include <asm/semaphore.h> |
64 | ||
65 | #include "gfs2.h" | |
5c676f6d SW |
66 | #include "lm_interface.h" |
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 | ||
81 | #if 1 | |
82 | #define gfs2_disk_hash2offset(h) (((uint64_t)(h)) >> 1) | |
83 | #define gfs2_dir_offset2hash(p) ((uint32_t)(((uint64_t)(p)) << 1)) | |
84 | #else | |
85 | #define gfs2_disk_hash2offset(h) (((uint64_t)(h))) | |
86 | #define gfs2_dir_offset2hash(p) ((uint32_t)(((uint64_t)(p)))) | |
87 | #endif | |
88 | ||
89 | typedef int (*leaf_call_t) (struct gfs2_inode *dip, | |
90 | uint32_t index, uint32_t len, uint64_t leaf_no, | |
91 | void *data); | |
92 | ||
18ec7d5c SW |
93 | int gfs2_dir_get_buffer(struct gfs2_inode *ip, uint64_t block, int new, |
94 | struct buffer_head **bhp) | |
e13940ba SW |
95 | { |
96 | struct buffer_head *bh; | |
97 | int error = 0; | |
98 | ||
99 | if (new) { | |
100 | bh = gfs2_meta_new(ip->i_gl, block); | |
101 | gfs2_trans_add_bh(ip->i_gl, bh, 1); | |
102 | gfs2_metatype_set(bh, GFS2_METATYPE_JD, GFS2_FORMAT_JD); | |
103 | gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header)); | |
104 | } else { | |
568f4c96 SW |
105 | error = gfs2_meta_read(ip->i_gl, block, DIO_START | DIO_WAIT, |
106 | &bh); | |
e13940ba SW |
107 | if (error) |
108 | return error; | |
109 | if (gfs2_metatype_check(ip->i_sbd, bh, GFS2_METATYPE_JD)) { | |
110 | brelse(bh); | |
111 | return -EIO; | |
112 | } | |
113 | } | |
114 | ||
115 | *bhp = bh; | |
116 | return 0; | |
117 | } | |
118 | ||
119 | ||
120 | ||
121 | static int gfs2_dir_write_stuffed(struct gfs2_inode *ip, const char *buf, | |
122 | unsigned int offset, unsigned int size) | |
123 | ||
124 | { | |
125 | struct buffer_head *dibh; | |
126 | int error; | |
127 | ||
128 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
129 | if (error) | |
130 | return error; | |
131 | ||
132 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); | |
c752666c | 133 | memcpy(dibh->b_data + offset + sizeof(struct gfs2_dinode), buf, size); |
e13940ba SW |
134 | if (ip->i_di.di_size < offset + size) |
135 | ip->i_di.di_size = offset + size; | |
136 | ip->i_di.di_mtime = ip->i_di.di_ctime = get_seconds(); | |
137 | gfs2_dinode_out(&ip->i_di, dibh->b_data); | |
138 | ||
139 | brelse(dibh); | |
140 | ||
141 | return size; | |
142 | } | |
143 | ||
144 | ||
145 | ||
146 | /** | |
147 | * gfs2_dir_write_data - Write directory information to the inode | |
148 | * @ip: The GFS2 inode | |
149 | * @buf: The buffer containing information to be written | |
150 | * @offset: The file offset to start writing at | |
151 | * @size: The amount of data to write | |
152 | * | |
153 | * Returns: The number of bytes correctly written or error code | |
154 | */ | |
155 | static int gfs2_dir_write_data(struct gfs2_inode *ip, const char *buf, | |
156 | uint64_t offset, unsigned int size) | |
157 | { | |
158 | struct gfs2_sbd *sdp = ip->i_sbd; | |
159 | struct buffer_head *dibh; | |
160 | uint64_t lblock, dblock; | |
161 | uint32_t extlen = 0; | |
162 | unsigned int o; | |
163 | int copied = 0; | |
164 | int error = 0; | |
165 | ||
166 | if (!size) | |
167 | return 0; | |
168 | ||
169 | if (gfs2_is_stuffed(ip) && | |
170 | offset + size <= sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) | |
568f4c96 SW |
171 | return gfs2_dir_write_stuffed(ip, buf, (unsigned int)offset, |
172 | size); | |
e13940ba SW |
173 | |
174 | if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip))) | |
175 | return -EINVAL; | |
176 | ||
177 | if (gfs2_is_stuffed(ip)) { | |
178 | error = gfs2_unstuff_dinode(ip, NULL, NULL); | |
179 | if (error) | |
c752666c | 180 | return error; |
e13940ba SW |
181 | } |
182 | ||
183 | lblock = offset; | |
184 | o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header); | |
185 | ||
186 | while (copied < size) { | |
187 | unsigned int amount; | |
188 | struct buffer_head *bh; | |
189 | int new; | |
190 | ||
191 | amount = size - copied; | |
192 | if (amount > sdp->sd_sb.sb_bsize - o) | |
193 | amount = sdp->sd_sb.sb_bsize - o; | |
194 | ||
195 | if (!extlen) { | |
196 | new = 1; | |
568f4c96 SW |
197 | error = gfs2_block_map(ip, lblock, &new, &dblock, |
198 | &extlen); | |
e13940ba SW |
199 | if (error) |
200 | goto fail; | |
201 | error = -EIO; | |
202 | if (gfs2_assert_withdraw(sdp, dblock)) | |
203 | goto fail; | |
204 | } | |
205 | ||
568f4c96 SW |
206 | error = gfs2_dir_get_buffer(ip, dblock, |
207 | (amount == sdp->sd_jbsize) ? | |
208 | 1 : new, &bh); | |
e13940ba SW |
209 | if (error) |
210 | goto fail; | |
211 | ||
212 | gfs2_trans_add_bh(ip->i_gl, bh, 1); | |
213 | memcpy(bh->b_data + o, buf, amount); | |
214 | brelse(bh); | |
215 | if (error) | |
216 | goto fail; | |
217 | ||
218 | copied += amount; | |
219 | lblock++; | |
220 | dblock++; | |
221 | extlen--; | |
222 | ||
223 | o = sizeof(struct gfs2_meta_header); | |
224 | } | |
225 | ||
226 | out: | |
227 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
228 | if (error) | |
229 | return error; | |
230 | ||
231 | if (ip->i_di.di_size < offset + copied) | |
232 | ip->i_di.di_size = offset + copied; | |
233 | ip->i_di.di_mtime = ip->i_di.di_ctime = get_seconds(); | |
234 | ||
235 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); | |
236 | gfs2_dinode_out(&ip->i_di, dibh->b_data); | |
237 | brelse(dibh); | |
238 | ||
239 | return copied; | |
240 | fail: | |
241 | if (copied) | |
242 | goto out; | |
243 | return error; | |
244 | } | |
245 | ||
246 | static int gfs2_dir_read_stuffed(struct gfs2_inode *ip, char *buf, | |
c752666c | 247 | unsigned int offset, unsigned int size) |
e13940ba SW |
248 | { |
249 | struct buffer_head *dibh; | |
250 | int error; | |
251 | ||
252 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
253 | if (!error) { | |
254 | offset += sizeof(struct gfs2_dinode); | |
255 | memcpy(buf, dibh->b_data + offset, size); | |
256 | brelse(dibh); | |
257 | } | |
258 | ||
259 | return (error) ? error : size; | |
260 | } | |
261 | ||
262 | ||
263 | /** | |
264 | * gfs2_dir_read_data - Read a data from a directory inode | |
265 | * @ip: The GFS2 Inode | |
266 | * @buf: The buffer to place result into | |
267 | * @offset: File offset to begin jdata_readng from | |
268 | * @size: Amount of data to transfer | |
269 | * | |
270 | * Returns: The amount of data actually copied or the error | |
271 | */ | |
272 | static int gfs2_dir_read_data(struct gfs2_inode *ip, char *buf, | |
273 | uint64_t offset, unsigned int size) | |
274 | { | |
275 | struct gfs2_sbd *sdp = ip->i_sbd; | |
276 | uint64_t lblock, dblock; | |
277 | uint32_t extlen = 0; | |
278 | unsigned int o; | |
279 | int copied = 0; | |
280 | int error = 0; | |
281 | ||
282 | if (offset >= ip->i_di.di_size) | |
283 | return 0; | |
284 | ||
285 | if ((offset + size) > ip->i_di.di_size) | |
286 | size = ip->i_di.di_size - offset; | |
287 | ||
288 | if (!size) | |
289 | return 0; | |
290 | ||
291 | if (gfs2_is_stuffed(ip)) | |
568f4c96 SW |
292 | return gfs2_dir_read_stuffed(ip, buf, (unsigned int)offset, |
293 | size); | |
e13940ba SW |
294 | |
295 | if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip))) | |
296 | return -EINVAL; | |
297 | ||
298 | lblock = offset; | |
299 | o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header); | |
300 | ||
301 | while (copied < size) { | |
302 | unsigned int amount; | |
303 | struct buffer_head *bh; | |
304 | int new; | |
305 | ||
306 | amount = size - copied; | |
307 | if (amount > sdp->sd_sb.sb_bsize - o) | |
308 | amount = sdp->sd_sb.sb_bsize - o; | |
309 | ||
310 | if (!extlen) { | |
311 | new = 0; | |
568f4c96 SW |
312 | error = gfs2_block_map(ip, lblock, &new, &dblock, |
313 | &extlen); | |
e13940ba SW |
314 | if (error) |
315 | goto fail; | |
316 | } | |
317 | ||
318 | if (extlen > 1) | |
319 | gfs2_meta_ra(ip->i_gl, dblock, extlen); | |
320 | ||
321 | if (dblock) { | |
322 | error = gfs2_dir_get_buffer(ip, dblock, new, &bh); | |
323 | if (error) | |
324 | goto fail; | |
325 | dblock++; | |
326 | extlen--; | |
327 | } else | |
328 | bh = NULL; | |
329 | ||
330 | memcpy(buf, bh->b_data + o, amount); | |
331 | brelse(bh); | |
332 | if (error) | |
333 | goto fail; | |
334 | ||
335 | copied += amount; | |
336 | lblock++; | |
337 | ||
338 | o = sizeof(struct gfs2_meta_header); | |
339 | } | |
340 | ||
341 | return copied; | |
342 | fail: | |
343 | return (copied) ? copied : error; | |
344 | } | |
345 | ||
c752666c SW |
346 | typedef int (*gfs2_dscan_t)(const struct gfs2_dirent *dent, |
347 | const struct qstr *name); | |
348 | ||
349 | static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent, | |
350 | const struct qstr *name, int ret) | |
351 | { | |
352 | if (dent->de_inum.no_addr != 0 && | |
353 | be32_to_cpu(dent->de_hash) == name->hash && | |
354 | be16_to_cpu(dent->de_name_len) == name->len && | |
355 | memcmp((char *)(dent+1), name->name, name->len) == 0) | |
356 | return ret; | |
357 | return 0; | |
358 | } | |
359 | ||
360 | static int gfs2_dirent_find(const struct gfs2_dirent *dent, | |
361 | const struct qstr *name) | |
362 | { | |
363 | return __gfs2_dirent_find(dent, name, 1); | |
364 | } | |
365 | ||
366 | static int gfs2_dirent_prev(const struct gfs2_dirent *dent, | |
367 | const struct qstr *name) | |
368 | { | |
369 | return __gfs2_dirent_find(dent, name, 2); | |
370 | } | |
371 | ||
372 | /* | |
373 | * name->name holds ptr to start of block. | |
374 | * name->len holds size of block. | |
b3b94faa | 375 | */ |
c752666c SW |
376 | static int gfs2_dirent_last(const struct gfs2_dirent *dent, |
377 | const struct qstr *name) | |
378 | { | |
379 | const char *start = name->name; | |
380 | const char *end = (const char *)dent + be16_to_cpu(dent->de_rec_len); | |
381 | if (name->len == (end - start)) | |
382 | return 1; | |
383 | return 0; | |
384 | } | |
b3b94faa | 385 | |
c752666c SW |
386 | static int gfs2_dirent_find_space(const struct gfs2_dirent *dent, |
387 | const struct qstr *name) | |
b3b94faa | 388 | { |
c752666c SW |
389 | unsigned required = GFS2_DIRENT_SIZE(name->len); |
390 | unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len)); | |
391 | unsigned totlen = be16_to_cpu(dent->de_rec_len); | |
392 | ||
393 | if ((totlen - actual) >= required) | |
394 | return 1; | |
395 | return 0; | |
396 | } | |
397 | ||
398 | /* | |
399 | * Other possible things to check: | |
400 | * - Inode located within filesystem size (and on valid block) | |
401 | * - Valid directory entry type | |
402 | * Not sure how heavy-weight we want to make this... could also check | |
403 | * hash is correct for example, but that would take a lot of extra time. | |
404 | * For now the most important thing is to check that the various sizes | |
405 | * are correct. | |
406 | */ | |
407 | static int gfs2_check_dirent(struct gfs2_dirent *dent, unsigned int offset, | |
408 | unsigned int size, unsigned int len, int first) | |
409 | { | |
410 | const char *msg = "gfs2_dirent too small"; | |
411 | if (unlikely(size < sizeof(struct gfs2_dirent))) | |
412 | goto error; | |
413 | msg = "gfs2_dirent misaligned"; | |
414 | if (unlikely(offset & 0x7)) | |
415 | goto error; | |
416 | msg = "gfs2_dirent points beyond end of block"; | |
417 | if (unlikely(offset + size > len)) | |
418 | goto error; | |
419 | msg = "zero inode number"; | |
420 | if (unlikely(!first && !dent->de_inum.no_addr)) | |
421 | goto error; | |
422 | msg = "name length is greater than space in dirent"; | |
423 | if (dent->de_inum.no_addr && | |
424 | unlikely(sizeof(struct gfs2_dirent)+be16_to_cpu(dent->de_name_len) > | |
425 | size)) | |
426 | goto error; | |
427 | return 0; | |
428 | error: | |
429 | printk(KERN_WARNING "gfs2_check_dirent: %s (%s)\n", msg, | |
430 | first ? "first in block" : "not first in block"); | |
431 | return -EIO; | |
b3b94faa DT |
432 | } |
433 | ||
c752666c SW |
434 | static struct gfs2_dirent *gfs2_dirent_scan(struct inode *inode, |
435 | void *buf, | |
436 | unsigned int len, gfs2_dscan_t scan, | |
437 | const struct qstr *name) | |
438 | { | |
439 | struct gfs2_meta_header *h = buf; | |
440 | struct gfs2_dirent *dent, *prev; | |
441 | unsigned offset; | |
442 | unsigned size; | |
443 | int ret = 0; | |
444 | ||
445 | BUG_ON(buf == NULL); | |
446 | BUG_ON(name == NULL); | |
447 | ||
448 | switch(be16_to_cpu(h->mh_type)) { | |
449 | case GFS2_METATYPE_LF: | |
450 | offset = sizeof(struct gfs2_leaf); | |
451 | break; | |
452 | case GFS2_METATYPE_DI: | |
453 | offset = sizeof(struct gfs2_dinode); | |
454 | break; | |
455 | default: | |
456 | goto wrong_type; | |
457 | } | |
458 | ||
459 | prev = NULL; | |
460 | dent = (struct gfs2_dirent *)(buf + offset); | |
461 | size = be16_to_cpu(dent->de_rec_len); | |
462 | if (gfs2_check_dirent(dent, offset, size, len, 1)) | |
463 | goto consist_inode; | |
464 | do { | |
465 | ret = scan(dent, name); | |
466 | if (ret) | |
467 | break; | |
468 | offset += size; | |
469 | if (offset == len) | |
470 | break; | |
471 | prev = dent; | |
472 | dent = (struct gfs2_dirent *)(buf + offset); | |
473 | size = be16_to_cpu(dent->de_rec_len); | |
474 | if (gfs2_check_dirent(dent, offset, size, len, 0)) | |
475 | goto consist_inode; | |
476 | } while(1); | |
477 | ||
478 | switch(ret) { | |
479 | case 0: | |
480 | return NULL; | |
481 | case 1: | |
482 | return dent; | |
483 | case 2: | |
484 | return prev ? prev : dent; | |
485 | default: | |
486 | BUG_ON(ret > 0); | |
487 | return ERR_PTR(ret); | |
488 | } | |
489 | ||
490 | wrong_type: | |
491 | printk(KERN_WARNING "gfs2_scan_dirent: %p wrong block type %u\n", scan, | |
492 | be16_to_cpu(h->mh_type)); | |
493 | consist_inode: | |
494 | gfs2_consist_inode(inode->u.generic_ip); | |
495 | return ERR_PTR(-EIO); | |
496 | } | |
497 | ||
498 | ||
b3b94faa DT |
499 | /** |
500 | * dirent_first - Return the first dirent | |
501 | * @dip: the directory | |
502 | * @bh: The buffer | |
503 | * @dent: Pointer to list of dirents | |
504 | * | |
505 | * return first dirent whether bh points to leaf or stuffed dinode | |
506 | * | |
507 | * Returns: IS_LEAF, IS_DINODE, or -errno | |
508 | */ | |
509 | ||
510 | static int dirent_first(struct gfs2_inode *dip, struct buffer_head *bh, | |
511 | struct gfs2_dirent **dent) | |
512 | { | |
513 | struct gfs2_meta_header *h = (struct gfs2_meta_header *)bh->b_data; | |
514 | ||
515 | if (be16_to_cpu(h->mh_type) == GFS2_METATYPE_LF) { | |
516 | if (gfs2_meta_check(dip->i_sbd, bh)) | |
517 | return -EIO; | |
518 | *dent = (struct gfs2_dirent *)(bh->b_data + | |
519 | sizeof(struct gfs2_leaf)); | |
520 | return IS_LEAF; | |
521 | } else { | |
522 | if (gfs2_metatype_check(dip->i_sbd, bh, GFS2_METATYPE_DI)) | |
523 | return -EIO; | |
524 | *dent = (struct gfs2_dirent *)(bh->b_data + | |
525 | sizeof(struct gfs2_dinode)); | |
526 | return IS_DINODE; | |
527 | } | |
528 | } | |
529 | ||
530 | /** | |
531 | * dirent_next - Next dirent | |
532 | * @dip: the directory | |
533 | * @bh: The buffer | |
534 | * @dent: Pointer to list of dirents | |
535 | * | |
536 | * Returns: 0 on success, error code otherwise | |
537 | */ | |
538 | ||
539 | static int dirent_next(struct gfs2_inode *dip, struct buffer_head *bh, | |
540 | struct gfs2_dirent **dent) | |
541 | { | |
542 | struct gfs2_dirent *tmp, *cur; | |
543 | char *bh_end; | |
fc69d0d3 | 544 | uint16_t cur_rec_len; |
b3b94faa DT |
545 | |
546 | cur = *dent; | |
547 | bh_end = bh->b_data + bh->b_size; | |
fc69d0d3 | 548 | cur_rec_len = be16_to_cpu(cur->de_rec_len); |
b3b94faa DT |
549 | |
550 | if ((char *)cur + cur_rec_len >= bh_end) { | |
551 | if ((char *)cur + cur_rec_len > bh_end) { | |
552 | gfs2_consist_inode(dip); | |
553 | return -EIO; | |
554 | } | |
555 | return -ENOENT; | |
556 | } | |
557 | ||
558 | tmp = (struct gfs2_dirent *)((char *)cur + cur_rec_len); | |
559 | ||
fc69d0d3 | 560 | if ((char *)tmp + be16_to_cpu(tmp->de_rec_len) > bh_end) { |
b3b94faa DT |
561 | gfs2_consist_inode(dip); |
562 | return -EIO; | |
563 | } | |
4dd651ad SW |
564 | |
565 | if (cur_rec_len == 0) { | |
566 | gfs2_consist_inode(dip); | |
567 | return -EIO; | |
568 | } | |
569 | ||
b3b94faa DT |
570 | /* Only the first dent could ever have de_inum.no_addr == 0 */ |
571 | if (!tmp->de_inum.no_addr) { | |
572 | gfs2_consist_inode(dip); | |
573 | return -EIO; | |
574 | } | |
575 | ||
576 | *dent = tmp; | |
577 | ||
578 | return 0; | |
579 | } | |
580 | ||
581 | /** | |
582 | * dirent_del - Delete a dirent | |
583 | * @dip: The GFS2 inode | |
584 | * @bh: The buffer | |
585 | * @prev: The previous dirent | |
586 | * @cur: The current dirent | |
587 | * | |
588 | */ | |
589 | ||
590 | static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh, | |
591 | struct gfs2_dirent *prev, struct gfs2_dirent *cur) | |
592 | { | |
fc69d0d3 | 593 | uint16_t cur_rec_len, prev_rec_len; |
b3b94faa DT |
594 | |
595 | if (!cur->de_inum.no_addr) { | |
596 | gfs2_consist_inode(dip); | |
597 | return; | |
598 | } | |
599 | ||
d4e9c4c3 | 600 | gfs2_trans_add_bh(dip->i_gl, bh, 1); |
b3b94faa DT |
601 | |
602 | /* If there is no prev entry, this is the first entry in the block. | |
603 | The de_rec_len is already as big as it needs to be. Just zero | |
604 | out the inode number and return. */ | |
605 | ||
606 | if (!prev) { | |
607 | cur->de_inum.no_addr = 0; /* No endianess worries */ | |
608 | return; | |
609 | } | |
610 | ||
611 | /* Combine this dentry with the previous one. */ | |
612 | ||
fc69d0d3 SW |
613 | prev_rec_len = be16_to_cpu(prev->de_rec_len); |
614 | cur_rec_len = be16_to_cpu(cur->de_rec_len); | |
b3b94faa DT |
615 | |
616 | if ((char *)prev + prev_rec_len != (char *)cur) | |
617 | gfs2_consist_inode(dip); | |
618 | if ((char *)cur + cur_rec_len > bh->b_data + bh->b_size) | |
619 | gfs2_consist_inode(dip); | |
620 | ||
621 | prev_rec_len += cur_rec_len; | |
fc69d0d3 | 622 | prev->de_rec_len = cpu_to_be16(prev_rec_len); |
b3b94faa DT |
623 | } |
624 | ||
c752666c SW |
625 | /* |
626 | * Takes a dent from which to grab space as an argument. Returns the | |
627 | * newly created dent. | |
b3b94faa | 628 | */ |
c752666c SW |
629 | struct gfs2_dirent *gfs2_init_dirent(struct inode *inode, |
630 | struct gfs2_dirent *dent, | |
631 | const struct qstr *name, | |
632 | struct buffer_head *bh) | |
b3b94faa | 633 | { |
c752666c SW |
634 | struct gfs2_inode *ip = inode->u.generic_ip; |
635 | struct gfs2_dirent *ndent; | |
636 | unsigned offset = 0, totlen; | |
637 | ||
638 | if (dent->de_inum.no_addr) | |
639 | offset = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len)); | |
640 | totlen = be16_to_cpu(dent->de_rec_len); | |
641 | BUG_ON(offset + name->len > totlen); | |
642 | gfs2_trans_add_bh(ip->i_gl, bh, 1); | |
643 | ndent = (struct gfs2_dirent *)((char *)dent + offset); | |
644 | dent->de_rec_len = cpu_to_be16(offset); | |
645 | gfs2_qstr2dirent(name, totlen - offset, ndent); | |
646 | return ndent; | |
b3b94faa DT |
647 | } |
648 | ||
c752666c SW |
649 | static struct gfs2_dirent *gfs2_dirent_alloc(struct inode *inode, |
650 | struct buffer_head *bh, | |
651 | const struct qstr *name) | |
b3b94faa DT |
652 | { |
653 | struct gfs2_dirent *dent; | |
c752666c SW |
654 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, gfs2_dirent_find_space, name); |
655 | if (!dent || IS_ERR(dent)) | |
656 | return dent; | |
657 | return gfs2_init_dirent(inode, dent, name, bh); | |
b3b94faa DT |
658 | } |
659 | ||
660 | static int get_leaf(struct gfs2_inode *dip, uint64_t leaf_no, | |
661 | struct buffer_head **bhp) | |
662 | { | |
663 | int error; | |
664 | ||
665 | error = gfs2_meta_read(dip->i_gl, leaf_no, DIO_START | DIO_WAIT, bhp); | |
666 | if (!error && gfs2_metatype_check(dip->i_sbd, *bhp, GFS2_METATYPE_LF)) | |
667 | error = -EIO; | |
668 | ||
669 | return error; | |
670 | } | |
671 | ||
672 | /** | |
673 | * get_leaf_nr - Get a leaf number associated with the index | |
674 | * @dip: The GFS2 inode | |
675 | * @index: | |
676 | * @leaf_out: | |
677 | * | |
678 | * Returns: 0 on success, error code otherwise | |
679 | */ | |
680 | ||
681 | static int get_leaf_nr(struct gfs2_inode *dip, uint32_t index, | |
682 | uint64_t *leaf_out) | |
683 | { | |
684 | uint64_t leaf_no; | |
685 | int error; | |
686 | ||
e13940ba | 687 | error = gfs2_dir_read_data(dip, (char *)&leaf_no, |
b3b94faa DT |
688 | index * sizeof(uint64_t), |
689 | sizeof(uint64_t)); | |
690 | if (error != sizeof(uint64_t)) | |
691 | return (error < 0) ? error : -EIO; | |
692 | ||
693 | *leaf_out = be64_to_cpu(leaf_no); | |
694 | ||
695 | return 0; | |
696 | } | |
697 | ||
698 | static int get_first_leaf(struct gfs2_inode *dip, uint32_t index, | |
699 | struct buffer_head **bh_out) | |
700 | { | |
701 | uint64_t leaf_no; | |
702 | int error; | |
703 | ||
704 | error = get_leaf_nr(dip, index, &leaf_no); | |
705 | if (!error) | |
706 | error = get_leaf(dip, leaf_no, bh_out); | |
707 | ||
708 | return error; | |
709 | } | |
710 | ||
c752666c SW |
711 | static struct gfs2_dirent *gfs2_dirent_search(struct inode *inode, |
712 | const struct qstr *name, | |
713 | gfs2_dscan_t scan, | |
714 | struct buffer_head **pbh) | |
b3b94faa | 715 | { |
c752666c SW |
716 | struct buffer_head *bh; |
717 | struct gfs2_dirent *dent; | |
718 | struct gfs2_inode *ip = inode->u.generic_ip; | |
b3b94faa DT |
719 | int error; |
720 | ||
c752666c SW |
721 | if (ip->i_di.di_flags & GFS2_DIF_EXHASH) { |
722 | struct gfs2_leaf *leaf; | |
723 | unsigned hsize = 1 << ip->i_di.di_depth; | |
724 | unsigned index; | |
725 | u64 ln; | |
726 | if (hsize * sizeof(u64) != ip->i_di.di_size) { | |
727 | gfs2_consist_inode(ip); | |
728 | return ERR_PTR(-EIO); | |
729 | } | |
b3b94faa | 730 | |
c752666c SW |
731 | index = name->hash >> (32 - ip->i_di.di_depth); |
732 | error = get_first_leaf(ip, index, &bh); | |
733 | if (error) | |
734 | return ERR_PTR(error); | |
735 | do { | |
736 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, | |
737 | scan, name); | |
738 | if (dent) | |
739 | goto got_dent; | |
740 | leaf = (struct gfs2_leaf *)bh->b_data; | |
741 | ln = be64_to_cpu(leaf->lf_next); | |
b3b94faa | 742 | brelse(bh); |
c752666c SW |
743 | if (!ln) |
744 | break; | |
745 | error = get_leaf(ip, ln, &bh); | |
746 | } while(!error); | |
b3b94faa | 747 | |
c752666c | 748 | return error ? ERR_PTR(error) : NULL; |
b3b94faa | 749 | } |
b3b94faa | 750 | |
c752666c SW |
751 | error = gfs2_meta_inode_buffer(ip, &bh); |
752 | if (error) | |
753 | return ERR_PTR(error); | |
754 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, scan, name); | |
755 | got_dent: | |
756 | *pbh = bh; | |
757 | return dent; | |
758 | } | |
b3b94faa | 759 | |
c752666c SW |
760 | static struct gfs2_leaf *new_leaf(struct inode *inode, struct buffer_head **pbh, u16 depth) |
761 | { | |
762 | struct gfs2_inode *ip = inode->u.generic_ip; | |
763 | u64 bn = gfs2_alloc_meta(ip); | |
764 | struct buffer_head *bh = gfs2_meta_new(ip->i_gl, bn); | |
765 | struct gfs2_leaf *leaf; | |
766 | struct gfs2_dirent *dent; | |
767 | if (!bh) | |
768 | return NULL; | |
769 | gfs2_trans_add_bh(ip->i_gl, bh, 1); | |
770 | gfs2_metatype_set(bh, GFS2_METATYPE_LF, GFS2_FORMAT_LF); | |
771 | leaf = (struct gfs2_leaf *)bh->b_data; | |
772 | leaf->lf_depth = cpu_to_be16(depth); | |
773 | leaf->lf_entries = cpu_to_be16(0); | |
774 | leaf->lf_dirent_format = cpu_to_be16(GFS2_FORMAT_DE); | |
775 | leaf->lf_next = cpu_to_be64(0); | |
776 | memset(leaf->lf_reserved, 0, sizeof(leaf->lf_reserved)); | |
777 | dent = (struct gfs2_dirent *)(leaf+1); | |
778 | dent->de_inum.no_formal_ino = cpu_to_be64(0); | |
779 | dent->de_inum.no_addr = cpu_to_be64(0); | |
780 | dent->de_hash = cpu_to_be32(0); | |
781 | dent->de_rec_len = cpu_to_be16(bh->b_size - sizeof(struct gfs2_leaf)); | |
782 | dent->de_name_len = cpu_to_be16(0); | |
783 | dent->de_type = cpu_to_be16(0); | |
784 | *pbh = bh; | |
785 | return leaf; | |
b3b94faa DT |
786 | } |
787 | ||
788 | /** | |
789 | * dir_make_exhash - Convert a stuffed directory into an ExHash directory | |
790 | * @dip: The GFS2 inode | |
791 | * | |
792 | * Returns: 0 on success, error code otherwise | |
793 | */ | |
794 | ||
c752666c | 795 | static int dir_make_exhash(struct inode *inode) |
b3b94faa | 796 | { |
c752666c | 797 | struct gfs2_inode *dip = inode->u.generic_ip; |
b3b94faa DT |
798 | struct gfs2_sbd *sdp = dip->i_sbd; |
799 | struct gfs2_dirent *dent; | |
c752666c | 800 | struct qstr args; |
b3b94faa DT |
801 | struct buffer_head *bh, *dibh; |
802 | struct gfs2_leaf *leaf; | |
803 | int y; | |
804 | uint32_t x; | |
805 | uint64_t *lp, bn; | |
806 | int error; | |
807 | ||
808 | error = gfs2_meta_inode_buffer(dip, &dibh); | |
809 | if (error) | |
810 | return error; | |
811 | ||
b3b94faa DT |
812 | /* Turn over a new leaf */ |
813 | ||
c752666c SW |
814 | leaf = new_leaf(inode, &bh, 0); |
815 | if (!leaf) | |
816 | return -ENOSPC; | |
817 | bn = bh->b_blocknr; | |
b3b94faa DT |
818 | |
819 | gfs2_assert(sdp, dip->i_di.di_entries < (1 << 16)); | |
b3b94faa DT |
820 | leaf->lf_entries = cpu_to_be16(dip->i_di.di_entries); |
821 | ||
822 | /* Copy dirents */ | |
823 | ||
824 | gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_leaf), dibh, | |
825 | sizeof(struct gfs2_dinode)); | |
826 | ||
827 | /* Find last entry */ | |
828 | ||
829 | x = 0; | |
c752666c SW |
830 | args.len = bh->b_size - sizeof(struct gfs2_dinode) + |
831 | sizeof(struct gfs2_leaf); | |
832 | args.name = bh->b_data; | |
833 | dent = gfs2_dirent_scan(dip->i_vnode, bh->b_data, bh->b_size, | |
834 | gfs2_dirent_last, &args); | |
835 | if (!dent) { | |
836 | brelse(bh); | |
837 | brelse(dibh); | |
838 | return -EIO; | |
839 | } | |
840 | if (IS_ERR(dent)) { | |
841 | brelse(bh); | |
842 | brelse(dibh); | |
843 | return PTR_ERR(dent); | |
b3b94faa | 844 | } |
b3b94faa DT |
845 | |
846 | /* Adjust the last dirent's record length | |
847 | (Remember that dent still points to the last entry.) */ | |
848 | ||
4dd651ad | 849 | dent->de_rec_len = cpu_to_be16(be16_to_cpu(dent->de_rec_len) + |
b3b94faa | 850 | sizeof(struct gfs2_dinode) - |
4dd651ad | 851 | sizeof(struct gfs2_leaf)); |
b3b94faa DT |
852 | |
853 | brelse(bh); | |
854 | ||
855 | /* We're done with the new leaf block, now setup the new | |
856 | hash table. */ | |
857 | ||
d4e9c4c3 | 858 | gfs2_trans_add_bh(dip->i_gl, dibh, 1); |
b3b94faa DT |
859 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); |
860 | ||
861 | lp = (uint64_t *)(dibh->b_data + sizeof(struct gfs2_dinode)); | |
862 | ||
863 | for (x = sdp->sd_hash_ptrs; x--; lp++) | |
864 | *lp = cpu_to_be64(bn); | |
865 | ||
866 | dip->i_di.di_size = sdp->sd_sb.sb_bsize / 2; | |
867 | dip->i_di.di_blocks++; | |
868 | dip->i_di.di_flags |= GFS2_DIF_EXHASH; | |
869 | dip->i_di.di_payload_format = 0; | |
870 | ||
871 | for (x = sdp->sd_hash_ptrs, y = -1; x; x >>= 1, y++) ; | |
872 | dip->i_di.di_depth = y; | |
873 | ||
874 | gfs2_dinode_out(&dip->i_di, dibh->b_data); | |
875 | ||
876 | brelse(dibh); | |
877 | ||
878 | return 0; | |
879 | } | |
880 | ||
881 | /** | |
882 | * dir_split_leaf - Split a leaf block into two | |
883 | * @dip: The GFS2 inode | |
884 | * @index: | |
885 | * @leaf_no: | |
886 | * | |
887 | * Returns: 0 on success, error code on failure | |
888 | */ | |
889 | ||
c752666c | 890 | static int dir_split_leaf(struct inode *inode, const struct qstr *name) |
b3b94faa | 891 | { |
c752666c | 892 | struct gfs2_inode *dip = inode->u.generic_ip; |
b3b94faa DT |
893 | struct buffer_head *nbh, *obh, *dibh; |
894 | struct gfs2_leaf *nleaf, *oleaf; | |
895 | struct gfs2_dirent *dent, *prev = NULL, *next = NULL, *new; | |
896 | uint32_t start, len, half_len, divider; | |
c752666c SW |
897 | uint64_t bn, *lp, leaf_no; |
898 | uint32_t index; | |
b3b94faa DT |
899 | int x, moved = 0; |
900 | int error; | |
901 | ||
c752666c SW |
902 | index = name->hash >> (32 - dip->i_di.di_depth); |
903 | error = get_leaf_nr(dip, index, &leaf_no); | |
904 | if (error) | |
905 | return error; | |
b3b94faa DT |
906 | |
907 | /* Get the old leaf block */ | |
b3b94faa DT |
908 | error = get_leaf(dip, leaf_no, &obh); |
909 | if (error) | |
910 | goto fail; | |
911 | ||
d4e9c4c3 | 912 | gfs2_trans_add_bh(dip->i_gl, obh, 1); |
b3b94faa DT |
913 | oleaf = (struct gfs2_leaf *)obh->b_data; |
914 | ||
c752666c SW |
915 | nleaf = new_leaf(inode, &nbh, be16_to_cpu(oleaf->lf_depth) + 1); |
916 | if (!nleaf) { | |
917 | brelse(obh); | |
918 | return -ENOSPC; | |
919 | } | |
920 | bn = nbh->b_blocknr; | |
b3b94faa | 921 | |
c752666c | 922 | /* Compute the start and len of leaf pointers in the hash table. */ |
b3b94faa DT |
923 | len = 1 << (dip->i_di.di_depth - be16_to_cpu(oleaf->lf_depth)); |
924 | half_len = len >> 1; | |
925 | if (!half_len) { | |
926 | gfs2_consist_inode(dip); | |
927 | error = -EIO; | |
928 | goto fail_brelse; | |
929 | } | |
930 | ||
931 | start = (index & ~(len - 1)); | |
932 | ||
933 | /* Change the pointers. | |
934 | Don't bother distinguishing stuffed from non-stuffed. | |
935 | This code is complicated enough already. */ | |
c752666c | 936 | lp = kmalloc(half_len * sizeof(uint64_t), GFP_NOFS | __GFP_NOFAIL); |
b3b94faa | 937 | /* Change the pointers */ |
b3b94faa DT |
938 | for (x = 0; x < half_len; x++) |
939 | lp[x] = cpu_to_be64(bn); | |
940 | ||
e13940ba | 941 | error = gfs2_dir_write_data(dip, (char *)lp, start * sizeof(uint64_t), |
b3b94faa DT |
942 | half_len * sizeof(uint64_t)); |
943 | if (error != half_len * sizeof(uint64_t)) { | |
944 | if (error >= 0) | |
945 | error = -EIO; | |
946 | goto fail_lpfree; | |
947 | } | |
948 | ||
949 | kfree(lp); | |
950 | ||
951 | /* Compute the divider */ | |
b3b94faa DT |
952 | divider = (start + half_len) << (32 - dip->i_di.di_depth); |
953 | ||
954 | /* Copy the entries */ | |
b3b94faa DT |
955 | dirent_first(dip, obh, &dent); |
956 | ||
957 | do { | |
958 | next = dent; | |
959 | if (dirent_next(dip, obh, &next)) | |
960 | next = NULL; | |
961 | ||
962 | if (dent->de_inum.no_addr && | |
963 | be32_to_cpu(dent->de_hash) < divider) { | |
c752666c SW |
964 | struct qstr str; |
965 | str.name = (char*)(dent+1); | |
966 | str.len = be16_to_cpu(dent->de_name_len); | |
967 | str.hash = be32_to_cpu(dent->de_hash); | |
968 | new = gfs2_dirent_alloc(dip->i_vnode, nbh, &str); | |
969 | if (IS_ERR(new)) { | |
970 | error = PTR_ERR(new); | |
971 | break; | |
972 | } | |
b3b94faa DT |
973 | |
974 | new->de_inum = dent->de_inum; /* No endian worries */ | |
b3b94faa | 975 | new->de_type = dent->de_type; /* No endian worries */ |
c752666c | 976 | nleaf->lf_entries = cpu_to_be16(be16_to_cpu(nleaf->lf_entries)+1); |
b3b94faa DT |
977 | |
978 | dirent_del(dip, obh, prev, dent); | |
979 | ||
980 | if (!oleaf->lf_entries) | |
981 | gfs2_consist_inode(dip); | |
c752666c | 982 | oleaf->lf_entries = cpu_to_be16(be16_to_cpu(oleaf->lf_entries)-1); |
b3b94faa DT |
983 | |
984 | if (!prev) | |
985 | prev = dent; | |
986 | ||
987 | moved = 1; | |
c752666c | 988 | } else { |
b3b94faa | 989 | prev = dent; |
c752666c | 990 | } |
b3b94faa | 991 | dent = next; |
c752666c | 992 | } while (dent); |
b3b94faa | 993 | |
c752666c | 994 | oleaf->lf_depth = nleaf->lf_depth; |
b3b94faa DT |
995 | |
996 | error = gfs2_meta_inode_buffer(dip, &dibh); | |
997 | if (!gfs2_assert_withdraw(dip->i_sbd, !error)) { | |
998 | dip->i_di.di_blocks++; | |
999 | gfs2_dinode_out(&dip->i_di, dibh->b_data); | |
1000 | brelse(dibh); | |
1001 | } | |
1002 | ||
1003 | brelse(obh); | |
1004 | brelse(nbh); | |
1005 | ||
1006 | return error; | |
1007 | ||
1008 | fail_lpfree: | |
1009 | kfree(lp); | |
1010 | ||
1011 | fail_brelse: | |
1012 | brelse(obh); | |
1013 | ||
1014 | fail: | |
1015 | brelse(nbh); | |
1016 | return error; | |
1017 | } | |
1018 | ||
1019 | /** | |
1020 | * dir_double_exhash - Double size of ExHash table | |
1021 | * @dip: The GFS2 dinode | |
1022 | * | |
1023 | * Returns: 0 on success, error code on failure | |
1024 | */ | |
1025 | ||
1026 | static int dir_double_exhash(struct gfs2_inode *dip) | |
1027 | { | |
1028 | struct gfs2_sbd *sdp = dip->i_sbd; | |
1029 | struct buffer_head *dibh; | |
1030 | uint32_t hsize; | |
1031 | uint64_t *buf; | |
1032 | uint64_t *from, *to; | |
1033 | uint64_t block; | |
1034 | int x; | |
1035 | int error = 0; | |
1036 | ||
1037 | hsize = 1 << dip->i_di.di_depth; | |
1038 | if (hsize * sizeof(uint64_t) != dip->i_di.di_size) { | |
1039 | gfs2_consist_inode(dip); | |
1040 | return -EIO; | |
1041 | } | |
1042 | ||
1043 | /* Allocate both the "from" and "to" buffers in one big chunk */ | |
1044 | ||
1045 | buf = kcalloc(3, sdp->sd_hash_bsize, GFP_KERNEL | __GFP_NOFAIL); | |
1046 | ||
1047 | for (block = dip->i_di.di_size >> sdp->sd_hash_bsize_shift; block--;) { | |
e13940ba | 1048 | error = gfs2_dir_read_data(dip, (char *)buf, |
b3b94faa DT |
1049 | block * sdp->sd_hash_bsize, |
1050 | sdp->sd_hash_bsize); | |
1051 | if (error != sdp->sd_hash_bsize) { | |
1052 | if (error >= 0) | |
1053 | error = -EIO; | |
1054 | goto fail; | |
1055 | } | |
1056 | ||
1057 | from = buf; | |
1058 | to = (uint64_t *)((char *)buf + sdp->sd_hash_bsize); | |
1059 | ||
1060 | for (x = sdp->sd_hash_ptrs; x--; from++) { | |
1061 | *to++ = *from; /* No endianess worries */ | |
1062 | *to++ = *from; | |
1063 | } | |
1064 | ||
e13940ba | 1065 | error = gfs2_dir_write_data(dip, |
b3b94faa DT |
1066 | (char *)buf + sdp->sd_hash_bsize, |
1067 | block * sdp->sd_sb.sb_bsize, | |
1068 | sdp->sd_sb.sb_bsize); | |
1069 | if (error != sdp->sd_sb.sb_bsize) { | |
1070 | if (error >= 0) | |
1071 | error = -EIO; | |
1072 | goto fail; | |
1073 | } | |
1074 | } | |
1075 | ||
1076 | kfree(buf); | |
1077 | ||
1078 | error = gfs2_meta_inode_buffer(dip, &dibh); | |
1079 | if (!gfs2_assert_withdraw(sdp, !error)) { | |
1080 | dip->i_di.di_depth++; | |
1081 | gfs2_dinode_out(&dip->i_di, dibh->b_data); | |
1082 | brelse(dibh); | |
1083 | } | |
1084 | ||
1085 | return error; | |
1086 | ||
1087 | fail: | |
1088 | kfree(buf); | |
1089 | ||
1090 | return error; | |
1091 | } | |
1092 | ||
1093 | /** | |
1094 | * compare_dents - compare directory entries by hash value | |
1095 | * @a: first dent | |
1096 | * @b: second dent | |
1097 | * | |
1098 | * When comparing the hash entries of @a to @b: | |
1099 | * gt: returns 1 | |
1100 | * lt: returns -1 | |
1101 | * eq: returns 0 | |
1102 | */ | |
1103 | ||
1104 | static int compare_dents(const void *a, const void *b) | |
1105 | { | |
1106 | struct gfs2_dirent *dent_a, *dent_b; | |
1107 | uint32_t hash_a, hash_b; | |
1108 | int ret = 0; | |
1109 | ||
1110 | dent_a = *(struct gfs2_dirent **)a; | |
c752666c | 1111 | hash_a = be32_to_cpu(dent_a->de_hash); |
b3b94faa DT |
1112 | |
1113 | dent_b = *(struct gfs2_dirent **)b; | |
c752666c | 1114 | hash_b = be32_to_cpu(dent_b->de_hash); |
b3b94faa DT |
1115 | |
1116 | if (hash_a > hash_b) | |
1117 | ret = 1; | |
1118 | else if (hash_a < hash_b) | |
1119 | ret = -1; | |
1120 | else { | |
4dd651ad SW |
1121 | unsigned int len_a = be16_to_cpu(dent_a->de_name_len); |
1122 | unsigned int len_b = be16_to_cpu(dent_b->de_name_len); | |
b3b94faa DT |
1123 | |
1124 | if (len_a > len_b) | |
1125 | ret = 1; | |
1126 | else if (len_a < len_b) | |
1127 | ret = -1; | |
1128 | else | |
1129 | ret = memcmp((char *)(dent_a + 1), | |
1130 | (char *)(dent_b + 1), | |
1131 | len_a); | |
1132 | } | |
1133 | ||
1134 | return ret; | |
1135 | } | |
1136 | ||
1137 | /** | |
1138 | * do_filldir_main - read out directory entries | |
1139 | * @dip: The GFS2 inode | |
1140 | * @offset: The offset in the file to read from | |
1141 | * @opaque: opaque data to pass to filldir | |
1142 | * @filldir: The function to pass entries to | |
1143 | * @darr: an array of struct gfs2_dirent pointers to read | |
1144 | * @entries: the number of entries in darr | |
1145 | * @copied: pointer to int that's non-zero if a entry has been copied out | |
1146 | * | |
1147 | * Jump through some hoops to make sure that if there are hash collsions, | |
1148 | * they are read out at the beginning of a buffer. We want to minimize | |
1149 | * the possibility that they will fall into different readdir buffers or | |
1150 | * that someone will want to seek to that location. | |
1151 | * | |
1152 | * Returns: errno, >0 on exception from filldir | |
1153 | */ | |
1154 | ||
1155 | static int do_filldir_main(struct gfs2_inode *dip, uint64_t *offset, | |
1156 | void *opaque, gfs2_filldir_t filldir, | |
1157 | struct gfs2_dirent **darr, uint32_t entries, | |
1158 | int *copied) | |
1159 | { | |
1160 | struct gfs2_dirent *dent, *dent_next; | |
1161 | struct gfs2_inum inum; | |
1162 | uint64_t off, off_next; | |
1163 | unsigned int x, y; | |
1164 | int run = 0; | |
1165 | int error = 0; | |
1166 | ||
1167 | sort(darr, entries, sizeof(struct gfs2_dirent *), compare_dents, NULL); | |
1168 | ||
1169 | dent_next = darr[0]; | |
1170 | off_next = be32_to_cpu(dent_next->de_hash); | |
1171 | off_next = gfs2_disk_hash2offset(off_next); | |
1172 | ||
1173 | for (x = 0, y = 1; x < entries; x++, y++) { | |
1174 | dent = dent_next; | |
1175 | off = off_next; | |
1176 | ||
1177 | if (y < entries) { | |
1178 | dent_next = darr[y]; | |
1179 | off_next = be32_to_cpu(dent_next->de_hash); | |
1180 | off_next = gfs2_disk_hash2offset(off_next); | |
1181 | ||
1182 | if (off < *offset) | |
1183 | continue; | |
1184 | *offset = off; | |
1185 | ||
1186 | if (off_next == off) { | |
1187 | if (*copied && !run) | |
1188 | return 1; | |
1189 | run = 1; | |
1190 | } else | |
1191 | run = 0; | |
1192 | } else { | |
1193 | if (off < *offset) | |
1194 | continue; | |
1195 | *offset = off; | |
1196 | } | |
1197 | ||
1198 | gfs2_inum_in(&inum, (char *)&dent->de_inum); | |
1199 | ||
1200 | error = filldir(opaque, (char *)(dent + 1), | |
4dd651ad | 1201 | be16_to_cpu(dent->de_name_len), |
b3b94faa | 1202 | off, &inum, |
4dd651ad | 1203 | be16_to_cpu(dent->de_type)); |
b3b94faa DT |
1204 | if (error) |
1205 | return 1; | |
1206 | ||
1207 | *copied = 1; | |
1208 | } | |
1209 | ||
1210 | /* Increment the *offset by one, so the next time we come into the | |
1211 | do_filldir fxn, we get the next entry instead of the last one in the | |
1212 | current leaf */ | |
1213 | ||
1214 | (*offset)++; | |
1215 | ||
1216 | return 0; | |
1217 | } | |
1218 | ||
1219 | /** | |
1220 | * do_filldir_single - Read directory entries out of a single block | |
1221 | * @dip: The GFS2 inode | |
1222 | * @offset: The offset in the file to read from | |
1223 | * @opaque: opaque data to pass to filldir | |
1224 | * @filldir: The function to pass entries to | |
1225 | * @bh: the block | |
1226 | * @entries: the number of entries in the block | |
1227 | * @copied: pointer to int that's non-zero if a entry has been copied out | |
1228 | * | |
1229 | * Returns: errno, >0 on exception from filldir | |
1230 | */ | |
1231 | ||
1232 | static int do_filldir_single(struct gfs2_inode *dip, uint64_t *offset, | |
1233 | void *opaque, gfs2_filldir_t filldir, | |
1234 | struct buffer_head *bh, uint32_t entries, | |
1235 | int *copied) | |
1236 | { | |
1237 | struct gfs2_dirent **darr; | |
1238 | struct gfs2_dirent *de; | |
1239 | unsigned int e = 0; | |
1240 | int error; | |
1241 | ||
1242 | if (!entries) | |
1243 | return 0; | |
1244 | ||
1245 | darr = kcalloc(entries, sizeof(struct gfs2_dirent *), GFP_KERNEL); | |
1246 | if (!darr) | |
1247 | return -ENOMEM; | |
1248 | ||
1249 | dirent_first(dip, bh, &de); | |
1250 | do { | |
1251 | if (!de->de_inum.no_addr) | |
1252 | continue; | |
1253 | if (e >= entries) { | |
1254 | gfs2_consist_inode(dip); | |
1255 | error = -EIO; | |
1256 | goto out; | |
1257 | } | |
1258 | darr[e++] = de; | |
c752666c | 1259 | } while (dirent_next(dip, bh, &de) == 0); |
b3b94faa DT |
1260 | |
1261 | if (e != entries) { | |
1262 | gfs2_consist_inode(dip); | |
1263 | error = -EIO; | |
1264 | goto out; | |
1265 | } | |
1266 | ||
1267 | error = do_filldir_main(dip, offset, opaque, filldir, darr, | |
1268 | entries, copied); | |
1269 | ||
1270 | out: | |
1271 | kfree(darr); | |
1272 | ||
1273 | return error; | |
1274 | } | |
1275 | ||
1276 | /** | |
1277 | * do_filldir_multi - Read directory entries out of a linked leaf list | |
1278 | * @dip: The GFS2 inode | |
1279 | * @offset: The offset in the file to read from | |
1280 | * @opaque: opaque data to pass to filldir | |
1281 | * @filldir: The function to pass entries to | |
1282 | * @bh: the first leaf in the list | |
1283 | * @copied: pointer to int that's non-zero if a entry has been copied out | |
1284 | * | |
1285 | * Returns: errno, >0 on exception from filldir | |
1286 | */ | |
1287 | ||
1288 | static int do_filldir_multi(struct gfs2_inode *dip, uint64_t *offset, | |
1289 | void *opaque, gfs2_filldir_t filldir, | |
1290 | struct buffer_head *bh, int *copied) | |
1291 | { | |
1292 | struct buffer_head **larr = NULL; | |
1293 | struct gfs2_dirent **darr; | |
1294 | struct gfs2_leaf *leaf; | |
1295 | struct buffer_head *tmp_bh; | |
1296 | struct gfs2_dirent *de; | |
1297 | unsigned int entries, e = 0; | |
1298 | unsigned int leaves = 0, l = 0; | |
1299 | unsigned int x; | |
1300 | uint64_t ln; | |
1301 | int error = 0; | |
1302 | ||
1303 | /* Count leaves and entries */ | |
1304 | ||
1305 | leaf = (struct gfs2_leaf *)bh->b_data; | |
1306 | entries = be16_to_cpu(leaf->lf_entries); | |
c752666c | 1307 | ln = be64_to_cpu(leaf->lf_next); |
b3b94faa DT |
1308 | |
1309 | while (ln) { | |
b3b94faa DT |
1310 | error = get_leaf(dip, ln, &tmp_bh); |
1311 | if (error) | |
1312 | return error; | |
1313 | ||
1314 | leaf = (struct gfs2_leaf *)tmp_bh->b_data; | |
1315 | if (leaf->lf_entries) { | |
1316 | entries += be16_to_cpu(leaf->lf_entries); | |
1317 | leaves++; | |
1318 | } | |
c752666c | 1319 | ln = be64_to_cpu(leaf->lf_next); |
b3b94faa DT |
1320 | |
1321 | brelse(tmp_bh); | |
1322 | } | |
1323 | ||
1324 | if (!entries) | |
1325 | return 0; | |
1326 | ||
1327 | if (leaves) { | |
1328 | larr = kcalloc(leaves, sizeof(struct buffer_head *),GFP_KERNEL); | |
1329 | if (!larr) | |
1330 | return -ENOMEM; | |
1331 | } | |
1332 | ||
1333 | darr = kcalloc(entries, sizeof(struct gfs2_dirent *), GFP_KERNEL); | |
1334 | if (!darr) { | |
1335 | kfree(larr); | |
1336 | return -ENOMEM; | |
1337 | } | |
1338 | ||
1339 | leaf = (struct gfs2_leaf *)bh->b_data; | |
1340 | if (leaf->lf_entries) { | |
1341 | dirent_first(dip, bh, &de); | |
1342 | do { | |
1343 | if (!de->de_inum.no_addr) | |
1344 | continue; | |
1345 | if (e >= entries) { | |
1346 | gfs2_consist_inode(dip); | |
1347 | error = -EIO; | |
1348 | goto out; | |
1349 | } | |
1350 | darr[e++] = de; | |
c752666c | 1351 | } while (dirent_next(dip, bh, &de) == 0); |
b3b94faa | 1352 | } |
c752666c | 1353 | ln = be64_to_cpu(leaf->lf_next); |
b3b94faa DT |
1354 | |
1355 | while (ln) { | |
b3b94faa DT |
1356 | error = get_leaf(dip, ln, &tmp_bh); |
1357 | if (error) | |
1358 | goto out; | |
1359 | ||
1360 | leaf = (struct gfs2_leaf *)tmp_bh->b_data; | |
1361 | if (leaf->lf_entries) { | |
1362 | dirent_first(dip, tmp_bh, &de); | |
1363 | do { | |
1364 | if (!de->de_inum.no_addr) | |
1365 | continue; | |
1366 | if (e >= entries) { | |
1367 | gfs2_consist_inode(dip); | |
1368 | error = -EIO; | |
1369 | goto out; | |
1370 | } | |
1371 | darr[e++] = de; | |
c752666c | 1372 | } while (dirent_next(dip, tmp_bh, &de) == 0); |
b3b94faa DT |
1373 | |
1374 | larr[l++] = tmp_bh; | |
1375 | ||
c752666c | 1376 | ln = be64_to_cpu(leaf->lf_next); |
b3b94faa | 1377 | } else { |
c752666c | 1378 | ln = be64_to_cpu(leaf->lf_next); |
b3b94faa DT |
1379 | brelse(tmp_bh); |
1380 | } | |
1381 | } | |
1382 | ||
1383 | if (gfs2_assert_withdraw(dip->i_sbd, l == leaves)) { | |
1384 | error = -EIO; | |
1385 | goto out; | |
1386 | } | |
1387 | if (e != entries) { | |
1388 | gfs2_consist_inode(dip); | |
1389 | error = -EIO; | |
1390 | goto out; | |
1391 | } | |
1392 | ||
1393 | error = do_filldir_main(dip, offset, opaque, filldir, darr, | |
1394 | entries, copied); | |
1395 | ||
1396 | out: | |
1397 | kfree(darr); | |
1398 | for (x = 0; x < l; x++) | |
1399 | brelse(larr[x]); | |
1400 | kfree(larr); | |
1401 | ||
1402 | return error; | |
1403 | } | |
1404 | ||
1405 | /** | |
c752666c SW |
1406 | * dir_e_read - Reads the entries from a directory into a filldir buffer |
1407 | * @dip: dinode pointer | |
1408 | * @offset: the hash of the last entry read shifted to the right once | |
1409 | * @opaque: buffer for the filldir function to fill | |
1410 | * @filldir: points to the filldir function to use | |
b3b94faa | 1411 | * |
c752666c | 1412 | * Returns: errno |
b3b94faa DT |
1413 | */ |
1414 | ||
c752666c SW |
1415 | static int dir_e_read(struct gfs2_inode *dip, uint64_t *offset, void *opaque, |
1416 | gfs2_filldir_t filldir) | |
b3b94faa | 1417 | { |
c752666c | 1418 | struct gfs2_sbd *sdp = dip->i_sbd; |
b3b94faa | 1419 | struct buffer_head *bh; |
c752666c SW |
1420 | struct gfs2_leaf *leaf; |
1421 | uint32_t hsize, len; | |
1422 | uint32_t ht_offset, lp_offset, ht_offset_cur = -1; | |
1423 | uint32_t hash, index; | |
1424 | uint64_t *lp; | |
1425 | int copied = 0; | |
1426 | int error = 0; | |
b3b94faa DT |
1427 | |
1428 | hsize = 1 << dip->i_di.di_depth; | |
1429 | if (hsize * sizeof(uint64_t) != dip->i_di.di_size) { | |
1430 | gfs2_consist_inode(dip); | |
1431 | return -EIO; | |
1432 | } | |
1433 | ||
1434 | hash = gfs2_dir_offset2hash(*offset); | |
1435 | index = hash >> (32 - dip->i_di.di_depth); | |
1436 | ||
1437 | lp = kmalloc(sdp->sd_hash_bsize, GFP_KERNEL); | |
1438 | if (!lp) | |
1439 | return -ENOMEM; | |
1440 | ||
1441 | while (index < hsize) { | |
1442 | lp_offset = index & (sdp->sd_hash_ptrs - 1); | |
1443 | ht_offset = index - lp_offset; | |
1444 | ||
1445 | if (ht_offset_cur != ht_offset) { | |
e13940ba | 1446 | error = gfs2_dir_read_data(dip, (char *)lp, |
b3b94faa DT |
1447 | ht_offset * sizeof(uint64_t), |
1448 | sdp->sd_hash_bsize); | |
1449 | if (error != sdp->sd_hash_bsize) { | |
1450 | if (error >= 0) | |
1451 | error = -EIO; | |
1452 | goto out; | |
1453 | } | |
1454 | ht_offset_cur = ht_offset; | |
1455 | } | |
1456 | ||
1457 | error = get_leaf(dip, be64_to_cpu(lp[lp_offset]), &bh); | |
1458 | if (error) | |
1459 | goto out; | |
1460 | ||
c752666c SW |
1461 | leaf = (struct gfs2_leaf *)bh->b_data; |
1462 | if (leaf->lf_next) | |
b3b94faa DT |
1463 | error = do_filldir_multi(dip, offset, opaque, filldir, |
1464 | bh, &copied); | |
1465 | else | |
1466 | error = do_filldir_single(dip, offset, opaque, filldir, | |
c752666c SW |
1467 | bh, |
1468 | be16_to_cpu(leaf->lf_entries), | |
1469 | &copied); | |
b3b94faa DT |
1470 | |
1471 | brelse(bh); | |
1472 | ||
1473 | if (error) { | |
1474 | if (error > 0) | |
1475 | error = 0; | |
1476 | goto out; | |
1477 | } | |
1478 | ||
c752666c | 1479 | len = 1 << (dip->i_di.di_depth - be16_to_cpu(leaf->lf_depth)); |
b3b94faa DT |
1480 | index = (index & ~(len - 1)) + len; |
1481 | } | |
1482 | ||
1483 | out: | |
1484 | kfree(lp); | |
1485 | ||
1486 | return error; | |
1487 | } | |
1488 | ||
b3b94faa DT |
1489 | static int dir_l_read(struct gfs2_inode *dip, uint64_t *offset, void *opaque, |
1490 | gfs2_filldir_t filldir) | |
1491 | { | |
1492 | struct buffer_head *dibh; | |
1493 | int copied = 0; | |
1494 | int error; | |
1495 | ||
1496 | if (!gfs2_is_stuffed(dip)) { | |
1497 | gfs2_consist_inode(dip); | |
1498 | return -EIO; | |
1499 | } | |
1500 | ||
1501 | if (!dip->i_di.di_entries) | |
1502 | return 0; | |
1503 | ||
1504 | error = gfs2_meta_inode_buffer(dip, &dibh); | |
1505 | if (error) | |
1506 | return error; | |
1507 | ||
1508 | error = do_filldir_single(dip, offset, | |
1509 | opaque, filldir, | |
1510 | dibh, dip->i_di.di_entries, | |
1511 | &copied); | |
1512 | if (error > 0) | |
1513 | error = 0; | |
1514 | ||
1515 | brelse(dibh); | |
1516 | ||
1517 | return error; | |
1518 | } | |
1519 | ||
b3b94faa DT |
1520 | /** |
1521 | * gfs2_dir_search - Search a directory | |
1522 | * @dip: The GFS2 inode | |
1523 | * @filename: | |
1524 | * @inode: | |
1525 | * | |
1526 | * This routine searches a directory for a file or another directory. | |
1527 | * Assumes a glock is held on dip. | |
1528 | * | |
1529 | * Returns: errno | |
1530 | */ | |
1531 | ||
c752666c | 1532 | int gfs2_dir_search(struct inode *dir, const struct qstr *name, |
b3b94faa DT |
1533 | struct gfs2_inum *inum, unsigned int *type) |
1534 | { | |
c752666c SW |
1535 | struct buffer_head *bh; |
1536 | struct gfs2_dirent *dent; | |
1537 | ||
1538 | dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh); | |
1539 | if (dent) { | |
1540 | if (IS_ERR(dent)) | |
1541 | return PTR_ERR(dent); | |
1542 | if (inum) | |
1543 | gfs2_inum_in(inum, (char *)&dent->de_inum); | |
1544 | if (type) | |
1545 | *type = be16_to_cpu(dent->de_type); | |
1546 | brelse(bh); | |
1547 | return 0; | |
1548 | } | |
1549 | return -ENOENT; | |
1550 | } | |
1551 | ||
1552 | static int dir_new_leaf(struct inode *inode, const struct qstr *name) | |
1553 | { | |
1554 | struct buffer_head *bh, *obh; | |
1555 | struct gfs2_inode *ip = inode->u.generic_ip; | |
1556 | struct gfs2_leaf *leaf, *oleaf; | |
b3b94faa | 1557 | int error; |
c752666c SW |
1558 | u32 index; |
1559 | u64 bn; | |
b3b94faa | 1560 | |
c752666c SW |
1561 | index = name->hash >> (32 - ip->i_di.di_depth); |
1562 | error = get_first_leaf(ip, index, &obh); | |
1563 | if (error) | |
1564 | return error; | |
1565 | do { | |
1566 | oleaf = (struct gfs2_leaf *)obh->b_data; | |
1567 | bn = be64_to_cpu(oleaf->lf_next); | |
1568 | if (!bn) | |
1569 | break; | |
1570 | brelse(obh); | |
1571 | error = get_leaf(ip, bn, &obh); | |
1572 | if (error) | |
1573 | return error; | |
1574 | } while(1); | |
b3b94faa | 1575 | |
c752666c SW |
1576 | gfs2_trans_add_bh(ip->i_gl, obh, 1); |
1577 | ||
1578 | leaf = new_leaf(inode, &bh, be16_to_cpu(oleaf->lf_depth)); | |
1579 | if (!leaf) { | |
1580 | brelse(obh); | |
1581 | return -ENOSPC; | |
1582 | } | |
1583 | oleaf->lf_next = cpu_to_be64(bn); | |
1584 | brelse(bh); | |
1585 | brelse(obh); | |
1586 | ||
1587 | error = gfs2_meta_inode_buffer(ip, &bh); | |
1588 | if (error) | |
1589 | return error; | |
1590 | gfs2_trans_add_bh(ip->i_gl, bh, 1); | |
1591 | ip->i_di.di_blocks++; | |
1592 | gfs2_dinode_out(&ip->i_di, bh->b_data); | |
1593 | brelse(bh); | |
1594 | return 0; | |
b3b94faa DT |
1595 | } |
1596 | ||
1597 | /** | |
1598 | * gfs2_dir_add - Add new filename into directory | |
1599 | * @dip: The GFS2 inode | |
1600 | * @filename: The new name | |
1601 | * @inode: The inode number of the entry | |
1602 | * @type: The type of the entry | |
1603 | * | |
1604 | * Returns: 0 on success, error code on failure | |
1605 | */ | |
1606 | ||
c752666c SW |
1607 | int gfs2_dir_add(struct inode *inode, const struct qstr *name, |
1608 | const struct gfs2_inum *inum, unsigned type) | |
b3b94faa | 1609 | { |
c752666c SW |
1610 | struct gfs2_inode *ip = inode->u.generic_ip; |
1611 | struct buffer_head *bh; | |
1612 | struct gfs2_dirent *dent; | |
1613 | struct gfs2_leaf *leaf; | |
b3b94faa DT |
1614 | int error; |
1615 | ||
c752666c SW |
1616 | while(1) { |
1617 | dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, | |
1618 | &bh); | |
1619 | if (dent) { | |
1620 | if (IS_ERR(dent)) | |
1621 | return PTR_ERR(dent); | |
1622 | dent = gfs2_init_dirent(inode, dent, name, bh); | |
1623 | gfs2_inum_out(inum, (char *)&dent->de_inum); | |
1624 | dent->de_type = cpu_to_be16(type); | |
1625 | if (ip->i_di.di_flags & GFS2_DIF_EXHASH) { | |
1626 | leaf = (struct gfs2_leaf *)bh->b_data; | |
1627 | leaf->lf_entries = cpu_to_be16(be16_to_cpu(leaf->lf_entries) + 1); | |
1628 | } | |
1629 | brelse(bh); | |
1630 | error = gfs2_meta_inode_buffer(ip, &bh); | |
1631 | if (error) | |
1632 | break; | |
1633 | gfs2_trans_add_bh(ip->i_gl, bh, 1); | |
1634 | ip->i_di.di_entries++; | |
1635 | ip->i_di.di_mtime = ip->i_di.di_ctime = get_seconds(); | |
1636 | gfs2_dinode_out(&ip->i_di, bh->b_data); | |
1637 | brelse(bh); | |
1638 | error = 0; | |
1639 | break; | |
1640 | } | |
1641 | if (!(ip->i_di.di_flags & GFS2_DIF_EXHASH)) { | |
1642 | error = dir_make_exhash(inode); | |
1643 | if (error) | |
1644 | break; | |
1645 | continue; | |
1646 | } | |
1647 | error = dir_split_leaf(inode, name); | |
1648 | if (error == 0) | |
1649 | continue; | |
1650 | if (error != -ENOSPC) | |
1651 | break; | |
1652 | if (ip->i_di.di_depth < GFS2_DIR_MAX_DEPTH) { | |
1653 | error = dir_double_exhash(ip); | |
1654 | if (error) | |
1655 | break; | |
1656 | error = dir_split_leaf(inode, name); | |
1657 | if (error) | |
1658 | break; | |
1659 | continue; | |
1660 | } | |
1661 | error = dir_new_leaf(inode, name); | |
1662 | if (!error) | |
1663 | continue; | |
1664 | error = -ENOSPC; | |
1665 | break; | |
1666 | } | |
b3b94faa DT |
1667 | return error; |
1668 | } | |
1669 | ||
c752666c | 1670 | |
b3b94faa DT |
1671 | /** |
1672 | * gfs2_dir_del - Delete a directory entry | |
1673 | * @dip: The GFS2 inode | |
1674 | * @filename: The filename | |
1675 | * | |
1676 | * Returns: 0 on success, error code on failure | |
1677 | */ | |
1678 | ||
c752666c | 1679 | int gfs2_dir_del(struct gfs2_inode *dip, const struct qstr *name) |
b3b94faa | 1680 | { |
c752666c SW |
1681 | struct gfs2_dirent *dent, *prev = NULL; |
1682 | struct buffer_head *bh; | |
b3b94faa DT |
1683 | int error; |
1684 | ||
c752666c SW |
1685 | /* Returns _either_ the entry (if its first in block) or the |
1686 | previous entry otherwise */ | |
1687 | dent = gfs2_dirent_search(dip->i_vnode, name, gfs2_dirent_prev, &bh); | |
1688 | if (!dent) { | |
1689 | gfs2_consist_inode(dip); | |
1690 | return -EIO; | |
1691 | } | |
1692 | if (IS_ERR(dent)) { | |
1693 | gfs2_consist_inode(dip); | |
1694 | return PTR_ERR(dent); | |
1695 | } | |
1696 | /* If not first in block, adjust pointers accordingly */ | |
1697 | if (gfs2_dirent_find(dent, name) == 0) { | |
1698 | prev = dent; | |
1699 | dent = (struct gfs2_dirent *)((char *)dent + be16_to_cpu(prev->de_rec_len)); | |
1700 | } | |
1701 | ||
1702 | dirent_del(dip, bh, prev, dent); | |
1703 | if (dip->i_di.di_flags & GFS2_DIF_EXHASH) { | |
1704 | struct gfs2_leaf *leaf = (struct gfs2_leaf *)bh->b_data; | |
1705 | u16 entries = be16_to_cpu(leaf->lf_entries); | |
1706 | if (!entries) | |
1707 | gfs2_consist_inode(dip); | |
1708 | leaf->lf_entries = cpu_to_be16(--entries); | |
1709 | brelse(bh); | |
1710 | } | |
1711 | ||
1712 | error = gfs2_meta_inode_buffer(dip, &bh); | |
1713 | if (error) | |
1714 | return error; | |
1715 | ||
1716 | if (!dip->i_di.di_entries) | |
1717 | gfs2_consist_inode(dip); | |
1718 | gfs2_trans_add_bh(dip->i_gl, bh, 1); | |
1719 | dip->i_di.di_entries--; | |
1720 | dip->i_di.di_mtime = dip->i_di.di_ctime = get_seconds(); | |
1721 | gfs2_dinode_out(&dip->i_di, bh->b_data); | |
1722 | brelse(bh); | |
b3b94faa DT |
1723 | |
1724 | return error; | |
1725 | } | |
1726 | ||
1727 | int gfs2_dir_read(struct gfs2_inode *dip, uint64_t *offset, void *opaque, | |
1728 | gfs2_filldir_t filldir) | |
1729 | { | |
1730 | int error; | |
1731 | ||
1732 | if (dip->i_di.di_flags & GFS2_DIF_EXHASH) | |
1733 | error = dir_e_read(dip, offset, opaque, filldir); | |
1734 | else | |
1735 | error = dir_l_read(dip, offset, opaque, filldir); | |
1736 | ||
1737 | return error; | |
1738 | } | |
1739 | ||
1740 | /** | |
1741 | * gfs2_dir_mvino - Change inode number of directory entry | |
1742 | * @dip: The GFS2 inode | |
1743 | * @filename: | |
1744 | * @new_inode: | |
1745 | * | |
1746 | * This routine changes the inode number of a directory entry. It's used | |
1747 | * by rename to change ".." when a directory is moved. | |
1748 | * Assumes a glock is held on dvp. | |
1749 | * | |
1750 | * Returns: errno | |
1751 | */ | |
1752 | ||
c752666c | 1753 | int gfs2_dir_mvino(struct gfs2_inode *dip, const struct qstr *filename, |
b3b94faa DT |
1754 | struct gfs2_inum *inum, unsigned int new_type) |
1755 | { | |
c752666c SW |
1756 | struct buffer_head *bh; |
1757 | struct gfs2_dirent *dent; | |
b3b94faa DT |
1758 | int error; |
1759 | ||
c752666c SW |
1760 | dent = gfs2_dirent_search(dip->i_vnode, filename, gfs2_dirent_find, &bh); |
1761 | if (!dent) { | |
1762 | gfs2_consist_inode(dip); | |
1763 | return -EIO; | |
1764 | } | |
1765 | if (IS_ERR(dent)) | |
1766 | return PTR_ERR(dent); | |
b3b94faa | 1767 | |
c752666c SW |
1768 | gfs2_trans_add_bh(dip->i_gl, bh, 1); |
1769 | gfs2_inum_out(inum, (char *)&dent->de_inum); | |
1770 | dent->de_type = cpu_to_be16(new_type); | |
1771 | ||
1772 | if (dip->i_di.di_flags & GFS2_DIF_EXHASH) { | |
1773 | brelse(bh); | |
1774 | error = gfs2_meta_inode_buffer(dip, &bh); | |
1775 | if (error) | |
1776 | return error; | |
1777 | gfs2_trans_add_bh(dip->i_gl, bh, 1); | |
1778 | } | |
1779 | ||
1780 | dip->i_di.di_mtime = dip->i_di.di_ctime = get_seconds(); | |
1781 | gfs2_dinode_out(&dip->i_di, bh->b_data); | |
1782 | brelse(bh); | |
1783 | return 0; | |
b3b94faa DT |
1784 | } |
1785 | ||
1786 | /** | |
1787 | * foreach_leaf - call a function for each leaf in a directory | |
1788 | * @dip: the directory | |
1789 | * @lc: the function to call for each each | |
1790 | * @data: private data to pass to it | |
1791 | * | |
1792 | * Returns: errno | |
1793 | */ | |
1794 | ||
1795 | static int foreach_leaf(struct gfs2_inode *dip, leaf_call_t lc, void *data) | |
1796 | { | |
1797 | struct gfs2_sbd *sdp = dip->i_sbd; | |
1798 | struct buffer_head *bh; | |
c752666c | 1799 | struct gfs2_leaf *leaf; |
b3b94faa DT |
1800 | uint32_t hsize, len; |
1801 | uint32_t ht_offset, lp_offset, ht_offset_cur = -1; | |
1802 | uint32_t index = 0; | |
1803 | uint64_t *lp; | |
1804 | uint64_t leaf_no; | |
1805 | int error = 0; | |
1806 | ||
1807 | hsize = 1 << dip->i_di.di_depth; | |
1808 | if (hsize * sizeof(uint64_t) != dip->i_di.di_size) { | |
1809 | gfs2_consist_inode(dip); | |
1810 | return -EIO; | |
1811 | } | |
1812 | ||
1813 | lp = kmalloc(sdp->sd_hash_bsize, GFP_KERNEL); | |
1814 | if (!lp) | |
1815 | return -ENOMEM; | |
1816 | ||
1817 | while (index < hsize) { | |
1818 | lp_offset = index & (sdp->sd_hash_ptrs - 1); | |
1819 | ht_offset = index - lp_offset; | |
1820 | ||
1821 | if (ht_offset_cur != ht_offset) { | |
e13940ba | 1822 | error = gfs2_dir_read_data(dip, (char *)lp, |
b3b94faa DT |
1823 | ht_offset * sizeof(uint64_t), |
1824 | sdp->sd_hash_bsize); | |
1825 | if (error != sdp->sd_hash_bsize) { | |
1826 | if (error >= 0) | |
1827 | error = -EIO; | |
1828 | goto out; | |
1829 | } | |
1830 | ht_offset_cur = ht_offset; | |
1831 | } | |
1832 | ||
1833 | leaf_no = be64_to_cpu(lp[lp_offset]); | |
1834 | if (leaf_no) { | |
1835 | error = get_leaf(dip, leaf_no, &bh); | |
1836 | if (error) | |
1837 | goto out; | |
c752666c | 1838 | leaf = (struct gfs2_leaf *)bh->b_data; |
b3b94faa DT |
1839 | brelse(bh); |
1840 | ||
c752666c | 1841 | len = 1 << (dip->i_di.di_depth - be16_to_cpu(leaf->lf_depth)); |
b3b94faa DT |
1842 | |
1843 | error = lc(dip, index, len, leaf_no, data); | |
1844 | if (error) | |
1845 | goto out; | |
1846 | ||
1847 | index = (index & ~(len - 1)) + len; | |
1848 | } else | |
1849 | index++; | |
1850 | } | |
1851 | ||
1852 | if (index != hsize) { | |
1853 | gfs2_consist_inode(dip); | |
1854 | error = -EIO; | |
1855 | } | |
1856 | ||
1857 | out: | |
1858 | kfree(lp); | |
1859 | ||
1860 | return error; | |
1861 | } | |
1862 | ||
1863 | /** | |
1864 | * leaf_dealloc - Deallocate a directory leaf | |
1865 | * @dip: the directory | |
1866 | * @index: the hash table offset in the directory | |
1867 | * @len: the number of pointers to this leaf | |
1868 | * @leaf_no: the leaf number | |
1869 | * @data: not used | |
1870 | * | |
1871 | * Returns: errno | |
1872 | */ | |
1873 | ||
1874 | static int leaf_dealloc(struct gfs2_inode *dip, uint32_t index, uint32_t len, | |
1875 | uint64_t leaf_no, void *data) | |
1876 | { | |
1877 | struct gfs2_sbd *sdp = dip->i_sbd; | |
c752666c | 1878 | struct gfs2_leaf *tmp_leaf; |
b3b94faa DT |
1879 | struct gfs2_rgrp_list rlist; |
1880 | struct buffer_head *bh, *dibh; | |
c752666c | 1881 | uint64_t blk, nblk; |
b3b94faa DT |
1882 | unsigned int rg_blocks = 0, l_blocks = 0; |
1883 | char *ht; | |
1884 | unsigned int x, size = len * sizeof(uint64_t); | |
1885 | int error; | |
1886 | ||
1887 | memset(&rlist, 0, sizeof(struct gfs2_rgrp_list)); | |
1888 | ||
1889 | ht = kzalloc(size, GFP_KERNEL); | |
1890 | if (!ht) | |
1891 | return -ENOMEM; | |
1892 | ||
1893 | gfs2_alloc_get(dip); | |
1894 | ||
1895 | error = gfs2_quota_hold(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); | |
1896 | if (error) | |
1897 | goto out; | |
1898 | ||
1899 | error = gfs2_rindex_hold(sdp, &dip->i_alloc.al_ri_gh); | |
1900 | if (error) | |
1901 | goto out_qs; | |
1902 | ||
1903 | /* Count the number of leaves */ | |
1904 | ||
c752666c | 1905 | for (blk = leaf_no; blk; blk = nblk) { |
b3b94faa DT |
1906 | error = get_leaf(dip, blk, &bh); |
1907 | if (error) | |
1908 | goto out_rlist; | |
c752666c SW |
1909 | tmp_leaf = (struct gfs2_leaf *)bh->b_data; |
1910 | nblk = be64_to_cpu(tmp_leaf->lf_next); | |
b3b94faa DT |
1911 | brelse(bh); |
1912 | ||
1913 | gfs2_rlist_add(sdp, &rlist, blk); | |
1914 | l_blocks++; | |
1915 | } | |
1916 | ||
1917 | gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE, 0); | |
1918 | ||
1919 | for (x = 0; x < rlist.rl_rgrps; x++) { | |
1920 | struct gfs2_rgrpd *rgd; | |
5c676f6d | 1921 | rgd = rlist.rl_ghs[x].gh_gl->gl_object; |
b3b94faa DT |
1922 | rg_blocks += rgd->rd_ri.ri_length; |
1923 | } | |
1924 | ||
1925 | error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs); | |
1926 | if (error) | |
1927 | goto out_rlist; | |
1928 | ||
1929 | error = gfs2_trans_begin(sdp, | |
5c676f6d | 1930 | rg_blocks + (DIV_ROUND_UP(size, sdp->sd_jbsize) + 1) + |
b3b94faa DT |
1931 | RES_DINODE + RES_STATFS + RES_QUOTA, l_blocks); |
1932 | if (error) | |
1933 | goto out_rg_gunlock; | |
1934 | ||
c752666c | 1935 | for (blk = leaf_no; blk; blk = nblk) { |
b3b94faa DT |
1936 | error = get_leaf(dip, blk, &bh); |
1937 | if (error) | |
1938 | goto out_end_trans; | |
c752666c SW |
1939 | tmp_leaf = (struct gfs2_leaf *)bh->b_data; |
1940 | nblk = be64_to_cpu(tmp_leaf->lf_next); | |
b3b94faa DT |
1941 | brelse(bh); |
1942 | ||
1943 | gfs2_free_meta(dip, blk, 1); | |
1944 | ||
1945 | if (!dip->i_di.di_blocks) | |
1946 | gfs2_consist_inode(dip); | |
1947 | dip->i_di.di_blocks--; | |
1948 | } | |
1949 | ||
e13940ba | 1950 | error = gfs2_dir_write_data(dip, ht, index * sizeof(uint64_t), size); |
b3b94faa DT |
1951 | if (error != size) { |
1952 | if (error >= 0) | |
1953 | error = -EIO; | |
1954 | goto out_end_trans; | |
1955 | } | |
1956 | ||
1957 | error = gfs2_meta_inode_buffer(dip, &dibh); | |
1958 | if (error) | |
1959 | goto out_end_trans; | |
1960 | ||
d4e9c4c3 | 1961 | gfs2_trans_add_bh(dip->i_gl, dibh, 1); |
b3b94faa DT |
1962 | gfs2_dinode_out(&dip->i_di, dibh->b_data); |
1963 | brelse(dibh); | |
1964 | ||
1965 | out_end_trans: | |
1966 | gfs2_trans_end(sdp); | |
1967 | ||
1968 | out_rg_gunlock: | |
1969 | gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs); | |
1970 | ||
1971 | out_rlist: | |
1972 | gfs2_rlist_free(&rlist); | |
1973 | gfs2_glock_dq_uninit(&dip->i_alloc.al_ri_gh); | |
1974 | ||
1975 | out_qs: | |
1976 | gfs2_quota_unhold(dip); | |
1977 | ||
1978 | out: | |
1979 | gfs2_alloc_put(dip); | |
1980 | kfree(ht); | |
1981 | ||
1982 | return error; | |
1983 | } | |
1984 | ||
1985 | /** | |
1986 | * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory | |
1987 | * @dip: the directory | |
1988 | * | |
1989 | * Dealloc all on-disk directory leaves to FREEMETA state | |
1990 | * Change on-disk inode type to "regular file" | |
1991 | * | |
1992 | * Returns: errno | |
1993 | */ | |
1994 | ||
1995 | int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip) | |
1996 | { | |
1997 | struct gfs2_sbd *sdp = dip->i_sbd; | |
1998 | struct buffer_head *bh; | |
1999 | int error; | |
2000 | ||
2001 | /* Dealloc on-disk leaves to FREEMETA state */ | |
2002 | error = foreach_leaf(dip, leaf_dealloc, NULL); | |
2003 | if (error) | |
2004 | return error; | |
2005 | ||
2006 | /* Make this a regular file in case we crash. | |
2007 | (We don't want to free these blocks a second time.) */ | |
2008 | ||
2009 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); | |
2010 | if (error) | |
2011 | return error; | |
2012 | ||
2013 | error = gfs2_meta_inode_buffer(dip, &bh); | |
2014 | if (!error) { | |
d4e9c4c3 | 2015 | gfs2_trans_add_bh(dip->i_gl, bh, 1); |
568f4c96 SW |
2016 | ((struct gfs2_dinode *)bh->b_data)->di_mode = |
2017 | cpu_to_be32(S_IFREG); | |
b3b94faa DT |
2018 | brelse(bh); |
2019 | } | |
2020 | ||
2021 | gfs2_trans_end(sdp); | |
2022 | ||
2023 | return error; | |
2024 | } | |
2025 | ||
2026 | /** | |
2027 | * gfs2_diradd_alloc_required - find if adding entry will require an allocation | |
2028 | * @ip: the file being written to | |
2029 | * @filname: the filename that's going to be added | |
b3b94faa | 2030 | * |
c752666c | 2031 | * Returns: 1 if alloc required, 0 if not, -ve on error |
b3b94faa DT |
2032 | */ |
2033 | ||
c752666c SW |
2034 | int gfs2_diradd_alloc_required(struct inode *inode, |
2035 | const struct qstr *name) | |
b3b94faa | 2036 | { |
c752666c SW |
2037 | struct gfs2_dirent *dent; |
2038 | struct buffer_head *bh; | |
b3b94faa | 2039 | |
c752666c SW |
2040 | dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, &bh); |
2041 | if (!dent) | |
2042 | return 1; | |
2043 | if (IS_ERR(dent)) | |
2044 | return PTR_ERR(dent); | |
2045 | brelse(bh); | |
2046 | return 0; | |
b3b94faa DT |
2047 | } |
2048 |