Commit | Line | Data |
---|---|---|
ccd979bd MF |
1 | /* -*- mode: c; c-basic-offset: 8; -*- |
2 | * vim: noexpandtab sw=8 ts=8 sts=0: | |
3 | * | |
4 | * dir.c | |
5 | * | |
6 | * Creates, reads, walks and deletes directory-nodes | |
7 | * | |
8 | * Copyright (C) 2002, 2004 Oracle. All rights reserved. | |
9 | * | |
10 | * Portions of this code from linux/fs/ext3/dir.c | |
11 | * | |
12 | * Copyright (C) 1992, 1993, 1994, 1995 | |
13 | * Remy Card (card@masi.ibp.fr) | |
14 | * Laboratoire MASI - Institut Blaise pascal | |
15 | * Universite Pierre et Marie Curie (Paris VI) | |
16 | * | |
17 | * from | |
18 | * | |
19 | * linux/fs/minix/dir.c | |
20 | * | |
21 | * Copyright (C) 1991, 1992 Linux Torvalds | |
22 | * | |
23 | * This program is free software; you can redistribute it and/or | |
24 | * modify it under the terms of the GNU General Public | |
25 | * License as published by the Free Software Foundation; either | |
26 | * version 2 of the License, or (at your option) any later version. | |
27 | * | |
28 | * This program is distributed in the hope that it will be useful, | |
29 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
30 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
31 | * General Public License for more details. | |
32 | * | |
33 | * You should have received a copy of the GNU General Public | |
34 | * License along with this program; if not, write to the | |
35 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
36 | * Boston, MA 021110-1307, USA. | |
37 | */ | |
38 | ||
39 | #include <linux/fs.h> | |
40 | #include <linux/types.h> | |
41 | #include <linux/slab.h> | |
42 | #include <linux/highmem.h> | |
43 | ||
44 | #define MLOG_MASK_PREFIX ML_NAMEI | |
45 | #include <cluster/masklog.h> | |
46 | ||
47 | #include "ocfs2.h" | |
48 | ||
49 | #include "alloc.h" | |
50 | #include "dir.h" | |
51 | #include "dlmglue.h" | |
52 | #include "extent_map.h" | |
53 | #include "file.h" | |
54 | #include "inode.h" | |
55 | #include "journal.h" | |
56 | #include "namei.h" | |
57 | #include "suballoc.h" | |
316f4b9f | 58 | #include "super.h" |
ccd979bd MF |
59 | #include "uptodate.h" |
60 | ||
61 | #include "buffer_head_io.h" | |
62 | ||
316f4b9f MF |
63 | #define NAMEI_RA_CHUNKS 2 |
64 | #define NAMEI_RA_BLOCKS 4 | |
65 | #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS) | |
66 | #define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b)) | |
67 | ||
ccd979bd MF |
68 | static unsigned char ocfs2_filetype_table[] = { |
69 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK | |
70 | }; | |
71 | ||
72 | static int ocfs2_extend_dir(struct ocfs2_super *osb, | |
73 | struct inode *dir, | |
74 | struct buffer_head *parent_fe_bh, | |
5b6a3a2b | 75 | unsigned int blocks_wanted, |
ccd979bd | 76 | struct buffer_head **new_de_bh); |
316f4b9f MF |
77 | static int ocfs2_do_extend_dir(struct super_block *sb, |
78 | handle_t *handle, | |
79 | struct inode *dir, | |
80 | struct buffer_head *parent_fe_bh, | |
81 | struct ocfs2_alloc_context *data_ac, | |
82 | struct ocfs2_alloc_context *meta_ac, | |
83 | struct buffer_head **new_bh); | |
84 | ||
07446dc7 JB |
85 | static struct buffer_head *ocfs2_bread(struct inode *inode, |
86 | int block, int *err, int reada) | |
87 | { | |
88 | struct buffer_head *bh = NULL; | |
89 | int tmperr; | |
90 | u64 p_blkno; | |
91 | int readflags = OCFS2_BH_CACHED; | |
92 | ||
93 | if (reada) | |
94 | readflags |= OCFS2_BH_READAHEAD; | |
95 | ||
96 | if (((u64)block << inode->i_sb->s_blocksize_bits) >= | |
97 | i_size_read(inode)) { | |
98 | BUG_ON(!reada); | |
99 | return NULL; | |
100 | } | |
101 | ||
102 | down_read(&OCFS2_I(inode)->ip_alloc_sem); | |
103 | tmperr = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL, | |
104 | NULL); | |
105 | up_read(&OCFS2_I(inode)->ip_alloc_sem); | |
106 | if (tmperr < 0) { | |
107 | mlog_errno(tmperr); | |
108 | goto fail; | |
109 | } | |
110 | ||
111 | tmperr = ocfs2_read_blocks(inode, p_blkno, 1, &bh, readflags); | |
112 | if (tmperr < 0) | |
113 | goto fail; | |
114 | ||
115 | tmperr = 0; | |
116 | ||
117 | *err = 0; | |
118 | return bh; | |
119 | ||
120 | fail: | |
121 | brelse(bh); | |
122 | bh = NULL; | |
123 | ||
124 | *err = -EIO; | |
125 | return NULL; | |
126 | } | |
127 | ||
23193e51 MF |
128 | /* |
129 | * bh passed here can be an inode block or a dir data block, depending | |
130 | * on the inode inline data flag. | |
131 | */ | |
5eae5b96 MF |
132 | static int ocfs2_check_dir_entry(struct inode * dir, |
133 | struct ocfs2_dir_entry * de, | |
134 | struct buffer_head * bh, | |
135 | unsigned long offset) | |
316f4b9f MF |
136 | { |
137 | const char *error_msg = NULL; | |
138 | const int rlen = le16_to_cpu(de->rec_len); | |
139 | ||
140 | if (rlen < OCFS2_DIR_REC_LEN(1)) | |
141 | error_msg = "rec_len is smaller than minimal"; | |
142 | else if (rlen % 4 != 0) | |
143 | error_msg = "rec_len % 4 != 0"; | |
144 | else if (rlen < OCFS2_DIR_REC_LEN(de->name_len)) | |
145 | error_msg = "rec_len is too small for name_len"; | |
146 | else if (((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize) | |
147 | error_msg = "directory entry across blocks"; | |
148 | ||
149 | if (error_msg != NULL) | |
150 | mlog(ML_ERROR, "bad entry in directory #%llu: %s - " | |
151 | "offset=%lu, inode=%llu, rec_len=%d, name_len=%d\n", | |
152 | (unsigned long long)OCFS2_I(dir)->ip_blkno, error_msg, | |
153 | offset, (unsigned long long)le64_to_cpu(de->inode), rlen, | |
154 | de->name_len); | |
155 | return error_msg == NULL ? 1 : 0; | |
156 | } | |
157 | ||
158 | static inline int ocfs2_match(int len, | |
159 | const char * const name, | |
160 | struct ocfs2_dir_entry *de) | |
161 | { | |
162 | if (len != de->name_len) | |
163 | return 0; | |
164 | if (!de->inode) | |
165 | return 0; | |
166 | return !memcmp(name, de->name, len); | |
167 | } | |
168 | ||
169 | /* | |
170 | * Returns 0 if not found, -1 on failure, and 1 on success | |
171 | */ | |
172 | static int inline ocfs2_search_dirblock(struct buffer_head *bh, | |
173 | struct inode *dir, | |
174 | const char *name, int namelen, | |
175 | unsigned long offset, | |
23193e51 MF |
176 | char *first_de, |
177 | unsigned int bytes, | |
316f4b9f MF |
178 | struct ocfs2_dir_entry **res_dir) |
179 | { | |
180 | struct ocfs2_dir_entry *de; | |
181 | char *dlimit, *de_buf; | |
182 | int de_len; | |
183 | int ret = 0; | |
184 | ||
185 | mlog_entry_void(); | |
186 | ||
23193e51 MF |
187 | de_buf = first_de; |
188 | dlimit = de_buf + bytes; | |
316f4b9f MF |
189 | |
190 | while (de_buf < dlimit) { | |
191 | /* this code is executed quadratically often */ | |
192 | /* do minimal checking `by hand' */ | |
193 | ||
194 | de = (struct ocfs2_dir_entry *) de_buf; | |
195 | ||
196 | if (de_buf + namelen <= dlimit && | |
197 | ocfs2_match(namelen, name, de)) { | |
198 | /* found a match - just to be sure, do a full check */ | |
199 | if (!ocfs2_check_dir_entry(dir, de, bh, offset)) { | |
200 | ret = -1; | |
201 | goto bail; | |
202 | } | |
203 | *res_dir = de; | |
204 | ret = 1; | |
205 | goto bail; | |
206 | } | |
207 | ||
208 | /* prevent looping on a bad block */ | |
209 | de_len = le16_to_cpu(de->rec_len); | |
210 | if (de_len <= 0) { | |
211 | ret = -1; | |
212 | goto bail; | |
213 | } | |
214 | ||
215 | de_buf += de_len; | |
216 | offset += de_len; | |
217 | } | |
218 | ||
219 | bail: | |
220 | mlog_exit(ret); | |
221 | return ret; | |
222 | } | |
223 | ||
23193e51 MF |
224 | static struct buffer_head *ocfs2_find_entry_id(const char *name, |
225 | int namelen, | |
226 | struct inode *dir, | |
227 | struct ocfs2_dir_entry **res_dir) | |
228 | { | |
229 | int ret, found; | |
230 | struct buffer_head *di_bh = NULL; | |
231 | struct ocfs2_dinode *di; | |
232 | struct ocfs2_inline_data *data; | |
233 | ||
0fcaa56a | 234 | ret = ocfs2_read_block(dir, OCFS2_I(dir)->ip_blkno, &di_bh); |
23193e51 MF |
235 | if (ret) { |
236 | mlog_errno(ret); | |
237 | goto out; | |
238 | } | |
239 | ||
240 | di = (struct ocfs2_dinode *)di_bh->b_data; | |
241 | data = &di->id2.i_data; | |
242 | ||
243 | found = ocfs2_search_dirblock(di_bh, dir, name, namelen, 0, | |
244 | data->id_data, i_size_read(dir), res_dir); | |
245 | if (found == 1) | |
246 | return di_bh; | |
247 | ||
248 | brelse(di_bh); | |
249 | out: | |
250 | return NULL; | |
251 | } | |
252 | ||
0af4bd38 AB |
253 | static struct buffer_head *ocfs2_find_entry_el(const char *name, int namelen, |
254 | struct inode *dir, | |
255 | struct ocfs2_dir_entry **res_dir) | |
316f4b9f MF |
256 | { |
257 | struct super_block *sb; | |
258 | struct buffer_head *bh_use[NAMEI_RA_SIZE]; | |
259 | struct buffer_head *bh, *ret = NULL; | |
260 | unsigned long start, block, b; | |
261 | int ra_max = 0; /* Number of bh's in the readahead | |
262 | buffer, bh_use[] */ | |
263 | int ra_ptr = 0; /* Current index into readahead | |
264 | buffer */ | |
265 | int num = 0; | |
266 | int nblocks, i, err; | |
267 | ||
268 | mlog_entry_void(); | |
269 | ||
316f4b9f MF |
270 | sb = dir->i_sb; |
271 | ||
272 | nblocks = i_size_read(dir) >> sb->s_blocksize_bits; | |
273 | start = OCFS2_I(dir)->ip_dir_start_lookup; | |
274 | if (start >= nblocks) | |
275 | start = 0; | |
276 | block = start; | |
277 | ||
278 | restart: | |
279 | do { | |
280 | /* | |
281 | * We deal with the read-ahead logic here. | |
282 | */ | |
283 | if (ra_ptr >= ra_max) { | |
284 | /* Refill the readahead buffer */ | |
285 | ra_ptr = 0; | |
286 | b = block; | |
287 | for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) { | |
288 | /* | |
289 | * Terminate if we reach the end of the | |
290 | * directory and must wrap, or if our | |
291 | * search has finished at this block. | |
292 | */ | |
293 | if (b >= nblocks || (num && block == start)) { | |
294 | bh_use[ra_max] = NULL; | |
295 | break; | |
296 | } | |
297 | num++; | |
298 | ||
299 | bh = ocfs2_bread(dir, b++, &err, 1); | |
300 | bh_use[ra_max] = bh; | |
301 | } | |
302 | } | |
303 | if ((bh = bh_use[ra_ptr++]) == NULL) | |
304 | goto next; | |
305 | wait_on_buffer(bh); | |
306 | if (!buffer_uptodate(bh)) { | |
307 | /* read error, skip block & hope for the best */ | |
308 | ocfs2_error(dir->i_sb, "reading directory %llu, " | |
309 | "offset %lu\n", | |
310 | (unsigned long long)OCFS2_I(dir)->ip_blkno, | |
311 | block); | |
312 | brelse(bh); | |
313 | goto next; | |
314 | } | |
315 | i = ocfs2_search_dirblock(bh, dir, name, namelen, | |
316 | block << sb->s_blocksize_bits, | |
23193e51 | 317 | bh->b_data, sb->s_blocksize, |
316f4b9f MF |
318 | res_dir); |
319 | if (i == 1) { | |
320 | OCFS2_I(dir)->ip_dir_start_lookup = block; | |
321 | ret = bh; | |
322 | goto cleanup_and_exit; | |
323 | } else { | |
324 | brelse(bh); | |
325 | if (i < 0) | |
326 | goto cleanup_and_exit; | |
327 | } | |
328 | next: | |
329 | if (++block >= nblocks) | |
330 | block = 0; | |
331 | } while (block != start); | |
332 | ||
333 | /* | |
334 | * If the directory has grown while we were searching, then | |
335 | * search the last part of the directory before giving up. | |
336 | */ | |
337 | block = nblocks; | |
338 | nblocks = i_size_read(dir) >> sb->s_blocksize_bits; | |
339 | if (block < nblocks) { | |
340 | start = 0; | |
341 | goto restart; | |
342 | } | |
343 | ||
344 | cleanup_and_exit: | |
345 | /* Clean up the read-ahead blocks */ | |
346 | for (; ra_ptr < ra_max; ra_ptr++) | |
347 | brelse(bh_use[ra_ptr]); | |
348 | ||
349 | mlog_exit_ptr(ret); | |
350 | return ret; | |
351 | } | |
352 | ||
23193e51 MF |
353 | /* |
354 | * Try to find an entry of the provided name within 'dir'. | |
355 | * | |
356 | * If nothing was found, NULL is returned. Otherwise, a buffer_head | |
357 | * and pointer to the dir entry are passed back. | |
358 | * | |
359 | * Caller can NOT assume anything about the contents of the | |
360 | * buffer_head - it is passed back only so that it can be passed into | |
361 | * any one of the manipulation functions (add entry, delete entry, | |
362 | * etc). As an example, bh in the extent directory case is a data | |
363 | * block, in the inline-data case it actually points to an inode. | |
364 | */ | |
365 | struct buffer_head *ocfs2_find_entry(const char *name, int namelen, | |
366 | struct inode *dir, | |
367 | struct ocfs2_dir_entry **res_dir) | |
368 | { | |
369 | *res_dir = NULL; | |
370 | ||
371 | if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) | |
372 | return ocfs2_find_entry_id(name, namelen, dir, res_dir); | |
373 | ||
374 | return ocfs2_find_entry_el(name, namelen, dir, res_dir); | |
375 | } | |
376 | ||
5b6a3a2b MF |
377 | /* |
378 | * Update inode number and type of a previously found directory entry. | |
379 | */ | |
38760e24 MF |
380 | int ocfs2_update_entry(struct inode *dir, handle_t *handle, |
381 | struct buffer_head *de_bh, struct ocfs2_dir_entry *de, | |
382 | struct inode *new_entry_inode) | |
383 | { | |
384 | int ret; | |
385 | ||
5b6a3a2b MF |
386 | /* |
387 | * The same code works fine for both inline-data and extent | |
388 | * based directories, so no need to split this up. | |
389 | */ | |
390 | ||
38760e24 MF |
391 | ret = ocfs2_journal_access(handle, dir, de_bh, |
392 | OCFS2_JOURNAL_ACCESS_WRITE); | |
393 | if (ret) { | |
394 | mlog_errno(ret); | |
395 | goto out; | |
396 | } | |
397 | ||
398 | de->inode = cpu_to_le64(OCFS2_I(new_entry_inode)->ip_blkno); | |
399 | ocfs2_set_de_type(de, new_entry_inode->i_mode); | |
400 | ||
401 | ocfs2_journal_dirty(handle, de_bh); | |
402 | ||
403 | out: | |
404 | return ret; | |
405 | } | |
406 | ||
5b6a3a2b MF |
407 | static int __ocfs2_delete_entry(handle_t *handle, struct inode *dir, |
408 | struct ocfs2_dir_entry *de_del, | |
409 | struct buffer_head *bh, char *first_de, | |
410 | unsigned int bytes) | |
316f4b9f MF |
411 | { |
412 | struct ocfs2_dir_entry *de, *pde; | |
413 | int i, status = -ENOENT; | |
414 | ||
415 | mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p)\n", handle, dir, de_del, bh); | |
416 | ||
417 | i = 0; | |
418 | pde = NULL; | |
5b6a3a2b MF |
419 | de = (struct ocfs2_dir_entry *) first_de; |
420 | while (i < bytes) { | |
316f4b9f MF |
421 | if (!ocfs2_check_dir_entry(dir, de, bh, i)) { |
422 | status = -EIO; | |
423 | mlog_errno(status); | |
424 | goto bail; | |
425 | } | |
426 | if (de == de_del) { | |
427 | status = ocfs2_journal_access(handle, dir, bh, | |
428 | OCFS2_JOURNAL_ACCESS_WRITE); | |
429 | if (status < 0) { | |
430 | status = -EIO; | |
431 | mlog_errno(status); | |
432 | goto bail; | |
433 | } | |
434 | if (pde) | |
0dd3256e MS |
435 | le16_add_cpu(&pde->rec_len, |
436 | le16_to_cpu(de->rec_len)); | |
316f4b9f MF |
437 | else |
438 | de->inode = 0; | |
439 | dir->i_version++; | |
440 | status = ocfs2_journal_dirty(handle, bh); | |
441 | goto bail; | |
442 | } | |
443 | i += le16_to_cpu(de->rec_len); | |
444 | pde = de; | |
445 | de = (struct ocfs2_dir_entry *)((char *)de + le16_to_cpu(de->rec_len)); | |
446 | } | |
447 | bail: | |
448 | mlog_exit(status); | |
449 | return status; | |
450 | } | |
451 | ||
5b6a3a2b MF |
452 | static inline int ocfs2_delete_entry_id(handle_t *handle, |
453 | struct inode *dir, | |
454 | struct ocfs2_dir_entry *de_del, | |
455 | struct buffer_head *bh) | |
456 | { | |
457 | int ret; | |
458 | struct buffer_head *di_bh = NULL; | |
459 | struct ocfs2_dinode *di; | |
460 | struct ocfs2_inline_data *data; | |
461 | ||
0fcaa56a | 462 | ret = ocfs2_read_block(dir, OCFS2_I(dir)->ip_blkno, &di_bh); |
5b6a3a2b MF |
463 | if (ret) { |
464 | mlog_errno(ret); | |
465 | goto out; | |
466 | } | |
467 | ||
468 | di = (struct ocfs2_dinode *)di_bh->b_data; | |
469 | data = &di->id2.i_data; | |
470 | ||
471 | ret = __ocfs2_delete_entry(handle, dir, de_del, bh, data->id_data, | |
472 | i_size_read(dir)); | |
473 | ||
474 | brelse(di_bh); | |
475 | out: | |
476 | return ret; | |
477 | } | |
478 | ||
479 | static inline int ocfs2_delete_entry_el(handle_t *handle, | |
480 | struct inode *dir, | |
481 | struct ocfs2_dir_entry *de_del, | |
482 | struct buffer_head *bh) | |
483 | { | |
484 | return __ocfs2_delete_entry(handle, dir, de_del, bh, bh->b_data, | |
485 | bh->b_size); | |
486 | } | |
487 | ||
488 | /* | |
489 | * ocfs2_delete_entry deletes a directory entry by merging it with the | |
490 | * previous entry | |
491 | */ | |
492 | int ocfs2_delete_entry(handle_t *handle, | |
493 | struct inode *dir, | |
494 | struct ocfs2_dir_entry *de_del, | |
495 | struct buffer_head *bh) | |
496 | { | |
497 | if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) | |
498 | return ocfs2_delete_entry_id(handle, dir, de_del, bh); | |
499 | ||
500 | return ocfs2_delete_entry_el(handle, dir, de_del, bh); | |
501 | } | |
502 | ||
8553cf4f MF |
503 | /* |
504 | * Check whether 'de' has enough room to hold an entry of | |
505 | * 'new_rec_len' bytes. | |
506 | */ | |
507 | static inline int ocfs2_dirent_would_fit(struct ocfs2_dir_entry *de, | |
508 | unsigned int new_rec_len) | |
509 | { | |
510 | unsigned int de_really_used; | |
511 | ||
512 | /* Check whether this is an empty record with enough space */ | |
513 | if (le64_to_cpu(de->inode) == 0 && | |
514 | le16_to_cpu(de->rec_len) >= new_rec_len) | |
515 | return 1; | |
516 | ||
517 | /* | |
518 | * Record might have free space at the end which we can | |
519 | * use. | |
520 | */ | |
521 | de_really_used = OCFS2_DIR_REC_LEN(de->name_len); | |
522 | if (le16_to_cpu(de->rec_len) >= (de_really_used + new_rec_len)) | |
523 | return 1; | |
524 | ||
525 | return 0; | |
526 | } | |
527 | ||
316f4b9f MF |
528 | /* we don't always have a dentry for what we want to add, so people |
529 | * like orphan dir can call this instead. | |
530 | * | |
531 | * If you pass me insert_bh, I'll skip the search of the other dir | |
532 | * blocks and put the record in there. | |
533 | */ | |
534 | int __ocfs2_add_entry(handle_t *handle, | |
535 | struct inode *dir, | |
536 | const char *name, int namelen, | |
537 | struct inode *inode, u64 blkno, | |
538 | struct buffer_head *parent_fe_bh, | |
539 | struct buffer_head *insert_bh) | |
540 | { | |
541 | unsigned long offset; | |
542 | unsigned short rec_len; | |
543 | struct ocfs2_dir_entry *de, *de1; | |
5b6a3a2b MF |
544 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_fe_bh->b_data; |
545 | struct super_block *sb = dir->i_sb; | |
316f4b9f | 546 | int retval, status; |
5b6a3a2b MF |
547 | unsigned int size = sb->s_blocksize; |
548 | char *data_start = insert_bh->b_data; | |
316f4b9f MF |
549 | |
550 | mlog_entry_void(); | |
551 | ||
316f4b9f MF |
552 | if (!namelen) |
553 | return -EINVAL; | |
554 | ||
5b6a3a2b MF |
555 | if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
556 | data_start = di->id2.i_data.id_data; | |
557 | size = i_size_read(dir); | |
558 | ||
559 | BUG_ON(insert_bh != parent_fe_bh); | |
560 | } | |
561 | ||
316f4b9f MF |
562 | rec_len = OCFS2_DIR_REC_LEN(namelen); |
563 | offset = 0; | |
5b6a3a2b | 564 | de = (struct ocfs2_dir_entry *) data_start; |
316f4b9f | 565 | while (1) { |
5b6a3a2b MF |
566 | BUG_ON((char *)de >= (size + data_start)); |
567 | ||
316f4b9f MF |
568 | /* These checks should've already been passed by the |
569 | * prepare function, but I guess we can leave them | |
570 | * here anyway. */ | |
571 | if (!ocfs2_check_dir_entry(dir, de, insert_bh, offset)) { | |
572 | retval = -ENOENT; | |
573 | goto bail; | |
574 | } | |
575 | if (ocfs2_match(namelen, name, de)) { | |
576 | retval = -EEXIST; | |
577 | goto bail; | |
578 | } | |
8553cf4f MF |
579 | |
580 | if (ocfs2_dirent_would_fit(de, rec_len)) { | |
316f4b9f MF |
581 | dir->i_mtime = dir->i_ctime = CURRENT_TIME; |
582 | retval = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh); | |
583 | if (retval < 0) { | |
584 | mlog_errno(retval); | |
585 | goto bail; | |
586 | } | |
587 | ||
588 | status = ocfs2_journal_access(handle, dir, insert_bh, | |
589 | OCFS2_JOURNAL_ACCESS_WRITE); | |
590 | /* By now the buffer is marked for journaling */ | |
591 | offset += le16_to_cpu(de->rec_len); | |
592 | if (le64_to_cpu(de->inode)) { | |
593 | de1 = (struct ocfs2_dir_entry *)((char *) de + | |
594 | OCFS2_DIR_REC_LEN(de->name_len)); | |
595 | de1->rec_len = | |
596 | cpu_to_le16(le16_to_cpu(de->rec_len) - | |
597 | OCFS2_DIR_REC_LEN(de->name_len)); | |
598 | de->rec_len = cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len)); | |
599 | de = de1; | |
600 | } | |
601 | de->file_type = OCFS2_FT_UNKNOWN; | |
602 | if (blkno) { | |
603 | de->inode = cpu_to_le64(blkno); | |
604 | ocfs2_set_de_type(de, inode->i_mode); | |
605 | } else | |
606 | de->inode = 0; | |
607 | de->name_len = namelen; | |
608 | memcpy(de->name, name, namelen); | |
609 | ||
610 | dir->i_version++; | |
611 | status = ocfs2_journal_dirty(handle, insert_bh); | |
612 | retval = 0; | |
613 | goto bail; | |
614 | } | |
615 | offset += le16_to_cpu(de->rec_len); | |
616 | de = (struct ocfs2_dir_entry *) ((char *) de + le16_to_cpu(de->rec_len)); | |
617 | } | |
618 | ||
619 | /* when you think about it, the assert above should prevent us | |
620 | * from ever getting here. */ | |
621 | retval = -ENOSPC; | |
622 | bail: | |
623 | ||
624 | mlog_exit(retval); | |
625 | return retval; | |
626 | } | |
627 | ||
23193e51 | 628 | static int ocfs2_dir_foreach_blk_id(struct inode *inode, |
2b47c361 | 629 | u64 *f_version, |
23193e51 | 630 | loff_t *f_pos, void *priv, |
e7b34019 | 631 | filldir_t filldir, int *filldir_err) |
23193e51 MF |
632 | { |
633 | int ret, i, filldir_ret; | |
634 | unsigned long offset = *f_pos; | |
635 | struct buffer_head *di_bh = NULL; | |
636 | struct ocfs2_dinode *di; | |
637 | struct ocfs2_inline_data *data; | |
638 | struct ocfs2_dir_entry *de; | |
639 | ||
0fcaa56a | 640 | ret = ocfs2_read_block(inode, OCFS2_I(inode)->ip_blkno, &di_bh); |
23193e51 MF |
641 | if (ret) { |
642 | mlog(ML_ERROR, "Unable to read inode block for dir %llu\n", | |
643 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
644 | goto out; | |
645 | } | |
646 | ||
647 | di = (struct ocfs2_dinode *)di_bh->b_data; | |
648 | data = &di->id2.i_data; | |
649 | ||
650 | while (*f_pos < i_size_read(inode)) { | |
651 | revalidate: | |
652 | /* If the dir block has changed since the last call to | |
653 | * readdir(2), then we might be pointing to an invalid | |
654 | * dirent right now. Scan from the start of the block | |
655 | * to make sure. */ | |
656 | if (*f_version != inode->i_version) { | |
657 | for (i = 0; i < i_size_read(inode) && i < offset; ) { | |
658 | de = (struct ocfs2_dir_entry *) | |
659 | (data->id_data + i); | |
660 | /* It's too expensive to do a full | |
661 | * dirent test each time round this | |
662 | * loop, but we do have to test at | |
663 | * least that it is non-zero. A | |
664 | * failure will be detected in the | |
665 | * dirent test below. */ | |
666 | if (le16_to_cpu(de->rec_len) < | |
667 | OCFS2_DIR_REC_LEN(1)) | |
668 | break; | |
669 | i += le16_to_cpu(de->rec_len); | |
670 | } | |
671 | *f_pos = offset = i; | |
672 | *f_version = inode->i_version; | |
673 | } | |
674 | ||
675 | de = (struct ocfs2_dir_entry *) (data->id_data + *f_pos); | |
676 | if (!ocfs2_check_dir_entry(inode, de, di_bh, *f_pos)) { | |
677 | /* On error, skip the f_pos to the end. */ | |
678 | *f_pos = i_size_read(inode); | |
679 | goto out; | |
680 | } | |
681 | offset += le16_to_cpu(de->rec_len); | |
682 | if (le64_to_cpu(de->inode)) { | |
683 | /* We might block in the next section | |
684 | * if the data destination is | |
685 | * currently swapped out. So, use a | |
686 | * version stamp to detect whether or | |
687 | * not the directory has been modified | |
688 | * during the copy operation. | |
689 | */ | |
2b47c361 | 690 | u64 version = *f_version; |
23193e51 MF |
691 | unsigned char d_type = DT_UNKNOWN; |
692 | ||
693 | if (de->file_type < OCFS2_FT_MAX) | |
694 | d_type = ocfs2_filetype_table[de->file_type]; | |
695 | ||
696 | filldir_ret = filldir(priv, de->name, | |
697 | de->name_len, | |
698 | *f_pos, | |
699 | le64_to_cpu(de->inode), | |
700 | d_type); | |
e7b34019 MF |
701 | if (filldir_ret) { |
702 | if (filldir_err) | |
703 | *filldir_err = filldir_ret; | |
23193e51 | 704 | break; |
e7b34019 | 705 | } |
23193e51 MF |
706 | if (version != *f_version) |
707 | goto revalidate; | |
708 | } | |
709 | *f_pos += le16_to_cpu(de->rec_len); | |
710 | } | |
711 | ||
712 | out: | |
713 | brelse(di_bh); | |
714 | ||
715 | return 0; | |
716 | } | |
717 | ||
718 | static int ocfs2_dir_foreach_blk_el(struct inode *inode, | |
2b47c361 | 719 | u64 *f_version, |
23193e51 | 720 | loff_t *f_pos, void *priv, |
e7b34019 | 721 | filldir_t filldir, int *filldir_err) |
ccd979bd MF |
722 | { |
723 | int error = 0; | |
aa958874 MF |
724 | unsigned long offset, blk, last_ra_blk = 0; |
725 | int i, stored; | |
ccd979bd MF |
726 | struct buffer_head * bh, * tmp; |
727 | struct ocfs2_dir_entry * de; | |
728 | int err; | |
ccd979bd | 729 | struct super_block * sb = inode->i_sb; |
aa958874 | 730 | unsigned int ra_sectors = 16; |
ccd979bd MF |
731 | |
732 | stored = 0; | |
733 | bh = NULL; | |
734 | ||
b8bc5f4f | 735 | offset = (*f_pos) & (sb->s_blocksize - 1); |
ccd979bd | 736 | |
b8bc5f4f MF |
737 | while (!error && !stored && *f_pos < i_size_read(inode)) { |
738 | blk = (*f_pos) >> sb->s_blocksize_bits; | |
ccd979bd MF |
739 | bh = ocfs2_bread(inode, blk, &err, 0); |
740 | if (!bh) { | |
b0697053 MF |
741 | mlog(ML_ERROR, |
742 | "directory #%llu contains a hole at offset %lld\n", | |
743 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
b8bc5f4f MF |
744 | *f_pos); |
745 | *f_pos += sb->s_blocksize - offset; | |
ccd979bd MF |
746 | continue; |
747 | } | |
748 | ||
aa958874 MF |
749 | /* The idea here is to begin with 8k read-ahead and to stay |
750 | * 4k ahead of our current position. | |
751 | * | |
752 | * TODO: Use the pagecache for this. We just need to | |
753 | * make sure it's cluster-safe... */ | |
754 | if (!last_ra_blk | |
755 | || (((last_ra_blk - blk) << 9) <= (ra_sectors / 2))) { | |
756 | for (i = ra_sectors >> (sb->s_blocksize_bits - 9); | |
ccd979bd MF |
757 | i > 0; i--) { |
758 | tmp = ocfs2_bread(inode, ++blk, &err, 1); | |
a81cb88b | 759 | brelse(tmp); |
ccd979bd | 760 | } |
aa958874 MF |
761 | last_ra_blk = blk; |
762 | ra_sectors = 8; | |
ccd979bd MF |
763 | } |
764 | ||
765 | revalidate: | |
766 | /* If the dir block has changed since the last call to | |
767 | * readdir(2), then we might be pointing to an invalid | |
768 | * dirent right now. Scan from the start of the block | |
769 | * to make sure. */ | |
b8bc5f4f | 770 | if (*f_version != inode->i_version) { |
ccd979bd MF |
771 | for (i = 0; i < sb->s_blocksize && i < offset; ) { |
772 | de = (struct ocfs2_dir_entry *) (bh->b_data + i); | |
773 | /* It's too expensive to do a full | |
774 | * dirent test each time round this | |
775 | * loop, but we do have to test at | |
776 | * least that it is non-zero. A | |
777 | * failure will be detected in the | |
778 | * dirent test below. */ | |
779 | if (le16_to_cpu(de->rec_len) < | |
780 | OCFS2_DIR_REC_LEN(1)) | |
781 | break; | |
782 | i += le16_to_cpu(de->rec_len); | |
783 | } | |
784 | offset = i; | |
b8bc5f4f | 785 | *f_pos = ((*f_pos) & ~(sb->s_blocksize - 1)) |
ccd979bd | 786 | | offset; |
b8bc5f4f | 787 | *f_version = inode->i_version; |
ccd979bd MF |
788 | } |
789 | ||
b8bc5f4f | 790 | while (!error && *f_pos < i_size_read(inode) |
ccd979bd MF |
791 | && offset < sb->s_blocksize) { |
792 | de = (struct ocfs2_dir_entry *) (bh->b_data + offset); | |
793 | if (!ocfs2_check_dir_entry(inode, de, bh, offset)) { | |
794 | /* On error, skip the f_pos to the | |
795 | next block. */ | |
b8bc5f4f | 796 | *f_pos = ((*f_pos) | (sb->s_blocksize - 1)) + 1; |
ccd979bd | 797 | brelse(bh); |
b8bc5f4f | 798 | goto out; |
ccd979bd MF |
799 | } |
800 | offset += le16_to_cpu(de->rec_len); | |
801 | if (le64_to_cpu(de->inode)) { | |
802 | /* We might block in the next section | |
803 | * if the data destination is | |
804 | * currently swapped out. So, use a | |
805 | * version stamp to detect whether or | |
806 | * not the directory has been modified | |
807 | * during the copy operation. | |
808 | */ | |
b8bc5f4f | 809 | unsigned long version = *f_version; |
ccd979bd MF |
810 | unsigned char d_type = DT_UNKNOWN; |
811 | ||
812 | if (de->file_type < OCFS2_FT_MAX) | |
813 | d_type = ocfs2_filetype_table[de->file_type]; | |
b8bc5f4f | 814 | error = filldir(priv, de->name, |
ccd979bd | 815 | de->name_len, |
b8bc5f4f | 816 | *f_pos, |
7e853679 | 817 | le64_to_cpu(de->inode), |
ccd979bd | 818 | d_type); |
e7b34019 MF |
819 | if (error) { |
820 | if (filldir_err) | |
821 | *filldir_err = error; | |
ccd979bd | 822 | break; |
e7b34019 | 823 | } |
b8bc5f4f | 824 | if (version != *f_version) |
ccd979bd MF |
825 | goto revalidate; |
826 | stored ++; | |
827 | } | |
b8bc5f4f | 828 | *f_pos += le16_to_cpu(de->rec_len); |
ccd979bd MF |
829 | } |
830 | offset = 0; | |
831 | brelse(bh); | |
832 | } | |
833 | ||
834 | stored = 0; | |
b8bc5f4f MF |
835 | out: |
836 | return stored; | |
837 | } | |
838 | ||
2b47c361 | 839 | static int ocfs2_dir_foreach_blk(struct inode *inode, u64 *f_version, |
e7b34019 MF |
840 | loff_t *f_pos, void *priv, filldir_t filldir, |
841 | int *filldir_err) | |
23193e51 MF |
842 | { |
843 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) | |
844 | return ocfs2_dir_foreach_blk_id(inode, f_version, f_pos, priv, | |
e7b34019 | 845 | filldir, filldir_err); |
23193e51 | 846 | |
e7b34019 MF |
847 | return ocfs2_dir_foreach_blk_el(inode, f_version, f_pos, priv, filldir, |
848 | filldir_err); | |
23193e51 MF |
849 | } |
850 | ||
5eae5b96 MF |
851 | /* |
852 | * This is intended to be called from inside other kernel functions, | |
853 | * so we fake some arguments. | |
854 | */ | |
855 | int ocfs2_dir_foreach(struct inode *inode, loff_t *f_pos, void *priv, | |
856 | filldir_t filldir) | |
857 | { | |
e7b34019 | 858 | int ret = 0, filldir_err = 0; |
2b47c361 | 859 | u64 version = inode->i_version; |
5eae5b96 MF |
860 | |
861 | while (*f_pos < i_size_read(inode)) { | |
862 | ret = ocfs2_dir_foreach_blk(inode, &version, f_pos, priv, | |
e7b34019 MF |
863 | filldir, &filldir_err); |
864 | if (ret || filldir_err) | |
5eae5b96 MF |
865 | break; |
866 | } | |
867 | ||
e7b34019 MF |
868 | if (ret > 0) |
869 | ret = -EIO; | |
870 | ||
5eae5b96 MF |
871 | return 0; |
872 | } | |
873 | ||
b8bc5f4f MF |
874 | /* |
875 | * ocfs2_readdir() | |
876 | * | |
877 | */ | |
878 | int ocfs2_readdir(struct file * filp, void * dirent, filldir_t filldir) | |
879 | { | |
880 | int error = 0; | |
881 | struct inode *inode = filp->f_path.dentry->d_inode; | |
882 | int lock_level = 0; | |
883 | ||
884 | mlog_entry("dirino=%llu\n", | |
885 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
886 | ||
e63aecb6 | 887 | error = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level); |
b8bc5f4f MF |
888 | if (lock_level && error >= 0) { |
889 | /* We release EX lock which used to update atime | |
890 | * and get PR lock again to reduce contention | |
891 | * on commonly accessed directories. */ | |
e63aecb6 | 892 | ocfs2_inode_unlock(inode, 1); |
b8bc5f4f | 893 | lock_level = 0; |
e63aecb6 | 894 | error = ocfs2_inode_lock(inode, NULL, 0); |
b8bc5f4f MF |
895 | } |
896 | if (error < 0) { | |
897 | if (error != -ENOENT) | |
898 | mlog_errno(error); | |
899 | /* we haven't got any yet, so propagate the error. */ | |
900 | goto bail_nolock; | |
901 | } | |
902 | ||
903 | error = ocfs2_dir_foreach_blk(inode, &filp->f_version, &filp->f_pos, | |
e7b34019 | 904 | dirent, filldir, NULL); |
b8bc5f4f | 905 | |
e63aecb6 | 906 | ocfs2_inode_unlock(inode, lock_level); |
ccd979bd | 907 | |
aa958874 | 908 | bail_nolock: |
b8bc5f4f | 909 | mlog_exit(error); |
ccd979bd | 910 | |
b8bc5f4f | 911 | return error; |
ccd979bd MF |
912 | } |
913 | ||
914 | /* | |
1b1dcc1b | 915 | * NOTE: this should always be called with parent dir i_mutex taken. |
ccd979bd MF |
916 | */ |
917 | int ocfs2_find_files_on_disk(const char *name, | |
918 | int namelen, | |
919 | u64 *blkno, | |
920 | struct inode *inode, | |
921 | struct buffer_head **dirent_bh, | |
922 | struct ocfs2_dir_entry **dirent) | |
923 | { | |
924 | int status = -ENOENT; | |
ccd979bd | 925 | |
2b388c67 JB |
926 | mlog_entry("(name=%.*s, blkno=%p, inode=%p, dirent_bh=%p, dirent=%p)\n", |
927 | namelen, name, blkno, inode, dirent_bh, dirent); | |
ccd979bd MF |
928 | |
929 | *dirent_bh = ocfs2_find_entry(name, namelen, inode, dirent); | |
930 | if (!*dirent_bh || !*dirent) { | |
931 | status = -ENOENT; | |
932 | goto leave; | |
933 | } | |
934 | ||
935 | *blkno = le64_to_cpu((*dirent)->inode); | |
936 | ||
937 | status = 0; | |
938 | leave: | |
939 | if (status < 0) { | |
940 | *dirent = NULL; | |
a81cb88b MF |
941 | brelse(*dirent_bh); |
942 | *dirent_bh = NULL; | |
ccd979bd MF |
943 | } |
944 | ||
945 | mlog_exit(status); | |
946 | return status; | |
947 | } | |
948 | ||
be94d117 MF |
949 | /* |
950 | * Convenience function for callers which just want the block number | |
951 | * mapped to a name and don't require the full dirent info, etc. | |
952 | */ | |
953 | int ocfs2_lookup_ino_from_name(struct inode *dir, const char *name, | |
954 | int namelen, u64 *blkno) | |
955 | { | |
956 | int ret; | |
957 | struct buffer_head *bh = NULL; | |
958 | struct ocfs2_dir_entry *dirent = NULL; | |
959 | ||
960 | ret = ocfs2_find_files_on_disk(name, namelen, blkno, dir, &bh, &dirent); | |
961 | brelse(bh); | |
962 | ||
963 | return ret; | |
964 | } | |
965 | ||
ccd979bd MF |
966 | /* Check for a name within a directory. |
967 | * | |
968 | * Return 0 if the name does not exist | |
969 | * Return -EEXIST if the directory contains the name | |
970 | * | |
1b1dcc1b | 971 | * Callers should have i_mutex + a cluster lock on dir |
ccd979bd MF |
972 | */ |
973 | int ocfs2_check_dir_for_entry(struct inode *dir, | |
974 | const char *name, | |
975 | int namelen) | |
976 | { | |
977 | int ret; | |
978 | struct buffer_head *dirent_bh = NULL; | |
979 | struct ocfs2_dir_entry *dirent = NULL; | |
980 | ||
b0697053 MF |
981 | mlog_entry("dir %llu, name '%.*s'\n", |
982 | (unsigned long long)OCFS2_I(dir)->ip_blkno, namelen, name); | |
ccd979bd MF |
983 | |
984 | ret = -EEXIST; | |
985 | dirent_bh = ocfs2_find_entry(name, namelen, dir, &dirent); | |
986 | if (dirent_bh) | |
987 | goto bail; | |
988 | ||
989 | ret = 0; | |
990 | bail: | |
a81cb88b | 991 | brelse(dirent_bh); |
ccd979bd MF |
992 | |
993 | mlog_exit(ret); | |
994 | return ret; | |
995 | } | |
996 | ||
0bfbbf62 MF |
997 | struct ocfs2_empty_dir_priv { |
998 | unsigned seen_dot; | |
999 | unsigned seen_dot_dot; | |
1000 | unsigned seen_other; | |
1001 | }; | |
1002 | static int ocfs2_empty_dir_filldir(void *priv, const char *name, int name_len, | |
1003 | loff_t pos, u64 ino, unsigned type) | |
1004 | { | |
1005 | struct ocfs2_empty_dir_priv *p = priv; | |
1006 | ||
1007 | /* | |
1008 | * Check the positions of "." and ".." records to be sure | |
1009 | * they're in the correct place. | |
1010 | */ | |
1011 | if (name_len == 1 && !strncmp(".", name, 1) && pos == 0) { | |
1012 | p->seen_dot = 1; | |
1013 | return 0; | |
1014 | } | |
1015 | ||
1016 | if (name_len == 2 && !strncmp("..", name, 2) && | |
1017 | pos == OCFS2_DIR_REC_LEN(1)) { | |
1018 | p->seen_dot_dot = 1; | |
1019 | return 0; | |
1020 | } | |
1021 | ||
1022 | p->seen_other = 1; | |
1023 | return 1; | |
1024 | } | |
ccd979bd MF |
1025 | /* |
1026 | * routine to check that the specified directory is empty (for rmdir) | |
0bfbbf62 MF |
1027 | * |
1028 | * Returns 1 if dir is empty, zero otherwise. | |
ccd979bd MF |
1029 | */ |
1030 | int ocfs2_empty_dir(struct inode *inode) | |
1031 | { | |
0bfbbf62 MF |
1032 | int ret; |
1033 | loff_t start = 0; | |
1034 | struct ocfs2_empty_dir_priv priv; | |
ccd979bd | 1035 | |
0bfbbf62 | 1036 | memset(&priv, 0, sizeof(priv)); |
ccd979bd | 1037 | |
0bfbbf62 MF |
1038 | ret = ocfs2_dir_foreach(inode, &start, &priv, ocfs2_empty_dir_filldir); |
1039 | if (ret) | |
1040 | mlog_errno(ret); | |
1041 | ||
1042 | if (!priv.seen_dot || !priv.seen_dot_dot) { | |
1043 | mlog(ML_ERROR, "bad directory (dir #%llu) - no `.' or `..'\n", | |
b0697053 | 1044 | (unsigned long long)OCFS2_I(inode)->ip_blkno); |
0bfbbf62 MF |
1045 | /* |
1046 | * XXX: Is it really safe to allow an unlink to continue? | |
1047 | */ | |
ccd979bd MF |
1048 | return 1; |
1049 | } | |
0bfbbf62 MF |
1050 | |
1051 | return !priv.seen_other; | |
ccd979bd MF |
1052 | } |
1053 | ||
5b6a3a2b MF |
1054 | static void ocfs2_fill_initial_dirents(struct inode *inode, |
1055 | struct inode *parent, | |
1056 | char *start, unsigned int size) | |
1057 | { | |
1058 | struct ocfs2_dir_entry *de = (struct ocfs2_dir_entry *)start; | |
1059 | ||
1060 | de->inode = cpu_to_le64(OCFS2_I(inode)->ip_blkno); | |
1061 | de->name_len = 1; | |
1062 | de->rec_len = | |
1063 | cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len)); | |
1064 | strcpy(de->name, "."); | |
1065 | ocfs2_set_de_type(de, S_IFDIR); | |
1066 | ||
1067 | de = (struct ocfs2_dir_entry *) ((char *)de + le16_to_cpu(de->rec_len)); | |
1068 | de->inode = cpu_to_le64(OCFS2_I(parent)->ip_blkno); | |
1069 | de->rec_len = cpu_to_le16(size - OCFS2_DIR_REC_LEN(1)); | |
1070 | de->name_len = 2; | |
1071 | strcpy(de->name, ".."); | |
1072 | ocfs2_set_de_type(de, S_IFDIR); | |
1073 | } | |
1074 | ||
1075 | /* | |
1076 | * This works together with code in ocfs2_mknod_locked() which sets | |
1077 | * the inline-data flag and initializes the inline-data section. | |
1078 | */ | |
1079 | static int ocfs2_fill_new_dir_id(struct ocfs2_super *osb, | |
1080 | handle_t *handle, | |
1081 | struct inode *parent, | |
1082 | struct inode *inode, | |
1083 | struct buffer_head *di_bh) | |
1084 | { | |
1085 | int ret; | |
1086 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
1087 | struct ocfs2_inline_data *data = &di->id2.i_data; | |
1088 | unsigned int size = le16_to_cpu(data->id_count); | |
1089 | ||
1090 | ret = ocfs2_journal_access(handle, inode, di_bh, | |
1091 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1092 | if (ret) { | |
1093 | mlog_errno(ret); | |
1094 | goto out; | |
1095 | } | |
1096 | ||
1097 | ocfs2_fill_initial_dirents(inode, parent, data->id_data, size); | |
1098 | ||
1099 | ocfs2_journal_dirty(handle, di_bh); | |
1100 | if (ret) { | |
1101 | mlog_errno(ret); | |
1102 | goto out; | |
1103 | } | |
1104 | ||
1105 | i_size_write(inode, size); | |
1106 | inode->i_nlink = 2; | |
1107 | inode->i_blocks = ocfs2_inode_sector_count(inode); | |
1108 | ||
1109 | ret = ocfs2_mark_inode_dirty(handle, inode, di_bh); | |
1110 | if (ret < 0) | |
1111 | mlog_errno(ret); | |
1112 | ||
1113 | out: | |
1114 | return ret; | |
1115 | } | |
1116 | ||
1117 | static int ocfs2_fill_new_dir_el(struct ocfs2_super *osb, | |
1118 | handle_t *handle, | |
1119 | struct inode *parent, | |
1120 | struct inode *inode, | |
1121 | struct buffer_head *fe_bh, | |
1122 | struct ocfs2_alloc_context *data_ac) | |
316f4b9f MF |
1123 | { |
1124 | int status; | |
1125 | struct buffer_head *new_bh = NULL; | |
316f4b9f MF |
1126 | |
1127 | mlog_entry_void(); | |
1128 | ||
1129 | status = ocfs2_do_extend_dir(osb->sb, handle, inode, fe_bh, | |
1130 | data_ac, NULL, &new_bh); | |
1131 | if (status < 0) { | |
1132 | mlog_errno(status); | |
1133 | goto bail; | |
1134 | } | |
1135 | ||
1136 | ocfs2_set_new_buffer_uptodate(inode, new_bh); | |
1137 | ||
1138 | status = ocfs2_journal_access(handle, inode, new_bh, | |
1139 | OCFS2_JOURNAL_ACCESS_CREATE); | |
1140 | if (status < 0) { | |
1141 | mlog_errno(status); | |
1142 | goto bail; | |
1143 | } | |
1144 | memset(new_bh->b_data, 0, osb->sb->s_blocksize); | |
1145 | ||
5b6a3a2b MF |
1146 | ocfs2_fill_initial_dirents(inode, parent, new_bh->b_data, |
1147 | osb->sb->s_blocksize); | |
316f4b9f MF |
1148 | |
1149 | status = ocfs2_journal_dirty(handle, new_bh); | |
1150 | if (status < 0) { | |
1151 | mlog_errno(status); | |
1152 | goto bail; | |
1153 | } | |
1154 | ||
1155 | i_size_write(inode, inode->i_sb->s_blocksize); | |
1156 | inode->i_nlink = 2; | |
1157 | inode->i_blocks = ocfs2_inode_sector_count(inode); | |
1158 | status = ocfs2_mark_inode_dirty(handle, inode, fe_bh); | |
1159 | if (status < 0) { | |
1160 | mlog_errno(status); | |
1161 | goto bail; | |
1162 | } | |
1163 | ||
1164 | status = 0; | |
1165 | bail: | |
a81cb88b | 1166 | brelse(new_bh); |
316f4b9f MF |
1167 | |
1168 | mlog_exit(status); | |
1169 | return status; | |
1170 | } | |
1171 | ||
5b6a3a2b MF |
1172 | int ocfs2_fill_new_dir(struct ocfs2_super *osb, |
1173 | handle_t *handle, | |
1174 | struct inode *parent, | |
1175 | struct inode *inode, | |
1176 | struct buffer_head *fe_bh, | |
1177 | struct ocfs2_alloc_context *data_ac) | |
1178 | { | |
1179 | BUG_ON(!ocfs2_supports_inline_data(osb) && data_ac == NULL); | |
1180 | ||
1181 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) | |
1182 | return ocfs2_fill_new_dir_id(osb, handle, parent, inode, fe_bh); | |
1183 | ||
1184 | return ocfs2_fill_new_dir_el(osb, handle, parent, inode, fe_bh, | |
1185 | data_ac); | |
1186 | } | |
1187 | ||
1188 | static void ocfs2_expand_last_dirent(char *start, unsigned int old_size, | |
1189 | unsigned int new_size) | |
1190 | { | |
1191 | struct ocfs2_dir_entry *de; | |
1192 | struct ocfs2_dir_entry *prev_de; | |
1193 | char *de_buf, *limit; | |
1194 | unsigned int bytes = new_size - old_size; | |
1195 | ||
1196 | limit = start + old_size; | |
1197 | de_buf = start; | |
1198 | de = (struct ocfs2_dir_entry *)de_buf; | |
1199 | do { | |
1200 | prev_de = de; | |
1201 | de_buf += le16_to_cpu(de->rec_len); | |
1202 | de = (struct ocfs2_dir_entry *)de_buf; | |
1203 | } while (de_buf < limit); | |
1204 | ||
1205 | le16_add_cpu(&prev_de->rec_len, bytes); | |
1206 | } | |
1207 | ||
1208 | /* | |
1209 | * We allocate enough clusters to fulfill "blocks_wanted", but set | |
1210 | * i_size to exactly one block. Ocfs2_extend_dir() will handle the | |
1211 | * rest automatically for us. | |
1212 | * | |
1213 | * *first_block_bh is a pointer to the 1st data block allocated to the | |
1214 | * directory. | |
1215 | */ | |
1216 | static int ocfs2_expand_inline_dir(struct inode *dir, struct buffer_head *di_bh, | |
1217 | unsigned int blocks_wanted, | |
1218 | struct buffer_head **first_block_bh) | |
1219 | { | |
1220 | int ret, credits = OCFS2_INLINE_TO_EXTENTS_CREDITS; | |
1221 | u32 alloc, bit_off, len; | |
1222 | struct super_block *sb = dir->i_sb; | |
1223 | u64 blkno, bytes = blocks_wanted << sb->s_blocksize_bits; | |
1224 | struct ocfs2_super *osb = OCFS2_SB(dir->i_sb); | |
1225 | struct ocfs2_inode_info *oi = OCFS2_I(dir); | |
1226 | struct ocfs2_alloc_context *data_ac; | |
1227 | struct buffer_head *dirdata_bh = NULL; | |
1228 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
1229 | handle_t *handle; | |
f99b9b7c JB |
1230 | struct ocfs2_extent_tree et; |
1231 | ||
8d6220d6 | 1232 | ocfs2_init_dinode_extent_tree(&et, dir, di_bh); |
5b6a3a2b MF |
1233 | |
1234 | alloc = ocfs2_clusters_for_bytes(sb, bytes); | |
1235 | ||
1236 | /* | |
1237 | * We should never need more than 2 clusters for this - | |
1238 | * maximum dirent size is far less than one block. In fact, | |
1239 | * the only time we'd need more than one cluster is if | |
1240 | * blocksize == clustersize and the dirent won't fit in the | |
1241 | * extra space that the expansion to a single block gives. As | |
1242 | * of today, that only happens on 4k/4k file systems. | |
1243 | */ | |
1244 | BUG_ON(alloc > 2); | |
1245 | ||
1246 | ret = ocfs2_reserve_clusters(osb, alloc, &data_ac); | |
1247 | if (ret) { | |
1248 | mlog_errno(ret); | |
1249 | goto out; | |
1250 | } | |
1251 | ||
1252 | down_write(&oi->ip_alloc_sem); | |
1253 | ||
1254 | /* | |
c78bad11 | 1255 | * Prepare for worst case allocation scenario of two separate |
5b6a3a2b MF |
1256 | * extents. |
1257 | */ | |
1258 | if (alloc == 2) | |
1259 | credits += OCFS2_SUBALLOC_ALLOC; | |
1260 | ||
1261 | handle = ocfs2_start_trans(osb, credits); | |
1262 | if (IS_ERR(handle)) { | |
1263 | ret = PTR_ERR(handle); | |
1264 | mlog_errno(ret); | |
1265 | goto out_sem; | |
1266 | } | |
1267 | ||
1268 | /* | |
1269 | * Try to claim as many clusters as the bitmap can give though | |
1270 | * if we only get one now, that's enough to continue. The rest | |
1271 | * will be claimed after the conversion to extents. | |
1272 | */ | |
1273 | ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off, &len); | |
1274 | if (ret) { | |
1275 | mlog_errno(ret); | |
1276 | goto out_commit; | |
1277 | } | |
1278 | ||
1279 | /* | |
1280 | * Operations are carefully ordered so that we set up the new | |
1281 | * data block first. The conversion from inline data to | |
1282 | * extents follows. | |
1283 | */ | |
1284 | blkno = ocfs2_clusters_to_blocks(dir->i_sb, bit_off); | |
1285 | dirdata_bh = sb_getblk(sb, blkno); | |
1286 | if (!dirdata_bh) { | |
1287 | ret = -EIO; | |
1288 | mlog_errno(ret); | |
1289 | goto out_commit; | |
1290 | } | |
1291 | ||
1292 | ocfs2_set_new_buffer_uptodate(dir, dirdata_bh); | |
1293 | ||
1294 | ret = ocfs2_journal_access(handle, dir, dirdata_bh, | |
1295 | OCFS2_JOURNAL_ACCESS_CREATE); | |
1296 | if (ret) { | |
1297 | mlog_errno(ret); | |
1298 | goto out_commit; | |
1299 | } | |
1300 | ||
1301 | memcpy(dirdata_bh->b_data, di->id2.i_data.id_data, i_size_read(dir)); | |
1302 | memset(dirdata_bh->b_data + i_size_read(dir), 0, | |
1303 | sb->s_blocksize - i_size_read(dir)); | |
1304 | ocfs2_expand_last_dirent(dirdata_bh->b_data, i_size_read(dir), | |
1305 | sb->s_blocksize); | |
1306 | ||
1307 | ret = ocfs2_journal_dirty(handle, dirdata_bh); | |
1308 | if (ret) { | |
1309 | mlog_errno(ret); | |
1310 | goto out_commit; | |
1311 | } | |
1312 | ||
1313 | /* | |
1314 | * Set extent, i_size, etc on the directory. After this, the | |
1315 | * inode should contain the same exact dirents as before and | |
1316 | * be fully accessible from system calls. | |
1317 | * | |
1318 | * We let the later dirent insert modify c/mtime - to the user | |
1319 | * the data hasn't changed. | |
1320 | */ | |
1321 | ret = ocfs2_journal_access(handle, dir, di_bh, | |
1322 | OCFS2_JOURNAL_ACCESS_CREATE); | |
1323 | if (ret) { | |
1324 | mlog_errno(ret); | |
1325 | goto out_commit; | |
1326 | } | |
1327 | ||
1328 | spin_lock(&oi->ip_lock); | |
1329 | oi->ip_dyn_features &= ~OCFS2_INLINE_DATA_FL; | |
1330 | di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features); | |
1331 | spin_unlock(&oi->ip_lock); | |
1332 | ||
1333 | ocfs2_dinode_new_extent_list(dir, di); | |
1334 | ||
1335 | i_size_write(dir, sb->s_blocksize); | |
1336 | dir->i_mtime = dir->i_ctime = CURRENT_TIME; | |
1337 | ||
1338 | di->i_size = cpu_to_le64(sb->s_blocksize); | |
1339 | di->i_ctime = di->i_mtime = cpu_to_le64(dir->i_ctime.tv_sec); | |
1340 | di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(dir->i_ctime.tv_nsec); | |
5b6a3a2b MF |
1341 | |
1342 | /* | |
1343 | * This should never fail as our extent list is empty and all | |
1344 | * related blocks have been journaled already. | |
1345 | */ | |
f99b9b7c JB |
1346 | ret = ocfs2_insert_extent(osb, handle, dir, &et, 0, blkno, len, |
1347 | 0, NULL); | |
5b6a3a2b MF |
1348 | if (ret) { |
1349 | mlog_errno(ret); | |
83cab533 | 1350 | goto out_commit; |
5b6a3a2b MF |
1351 | } |
1352 | ||
9780eb6c MF |
1353 | /* |
1354 | * Set i_blocks after the extent insert for the most up to | |
1355 | * date ip_clusters value. | |
1356 | */ | |
1357 | dir->i_blocks = ocfs2_inode_sector_count(dir); | |
1358 | ||
5b6a3a2b MF |
1359 | ret = ocfs2_journal_dirty(handle, di_bh); |
1360 | if (ret) { | |
1361 | mlog_errno(ret); | |
1362 | goto out_commit; | |
1363 | } | |
1364 | ||
1365 | /* | |
1366 | * We asked for two clusters, but only got one in the 1st | |
1367 | * pass. Claim the 2nd cluster as a separate extent. | |
1368 | */ | |
1369 | if (alloc > len) { | |
1370 | ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off, | |
1371 | &len); | |
1372 | if (ret) { | |
1373 | mlog_errno(ret); | |
1374 | goto out_commit; | |
1375 | } | |
1376 | blkno = ocfs2_clusters_to_blocks(dir->i_sb, bit_off); | |
1377 | ||
f99b9b7c JB |
1378 | ret = ocfs2_insert_extent(osb, handle, dir, &et, 1, |
1379 | blkno, len, 0, NULL); | |
5b6a3a2b MF |
1380 | if (ret) { |
1381 | mlog_errno(ret); | |
83cab533 | 1382 | goto out_commit; |
5b6a3a2b MF |
1383 | } |
1384 | } | |
1385 | ||
1386 | *first_block_bh = dirdata_bh; | |
1387 | dirdata_bh = NULL; | |
1388 | ||
1389 | out_commit: | |
1390 | ocfs2_commit_trans(osb, handle); | |
1391 | ||
1392 | out_sem: | |
1393 | up_write(&oi->ip_alloc_sem); | |
1394 | ||
1395 | out: | |
1396 | if (data_ac) | |
1397 | ocfs2_free_alloc_context(data_ac); | |
1398 | ||
1399 | brelse(dirdata_bh); | |
1400 | ||
1401 | return ret; | |
1402 | } | |
1403 | ||
ccd979bd | 1404 | /* returns a bh of the 1st new block in the allocation. */ |
316f4b9f MF |
1405 | static int ocfs2_do_extend_dir(struct super_block *sb, |
1406 | handle_t *handle, | |
1407 | struct inode *dir, | |
1408 | struct buffer_head *parent_fe_bh, | |
1409 | struct ocfs2_alloc_context *data_ac, | |
1410 | struct ocfs2_alloc_context *meta_ac, | |
1411 | struct buffer_head **new_bh) | |
ccd979bd MF |
1412 | { |
1413 | int status; | |
1414 | int extend; | |
8110b073 | 1415 | u64 p_blkno, v_blkno; |
ccd979bd MF |
1416 | |
1417 | spin_lock(&OCFS2_I(dir)->ip_lock); | |
1418 | extend = (i_size_read(dir) == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters)); | |
1419 | spin_unlock(&OCFS2_I(dir)->ip_lock); | |
1420 | ||
1421 | if (extend) { | |
dcd0538f MF |
1422 | u32 offset = OCFS2_I(dir)->ip_clusters; |
1423 | ||
0eb8d47e TM |
1424 | status = ocfs2_add_inode_data(OCFS2_SB(sb), dir, &offset, |
1425 | 1, 0, parent_fe_bh, handle, | |
1426 | data_ac, meta_ac, NULL); | |
ccd979bd MF |
1427 | BUG_ON(status == -EAGAIN); |
1428 | if (status < 0) { | |
1429 | mlog_errno(status); | |
1430 | goto bail; | |
1431 | } | |
1432 | } | |
1433 | ||
8110b073 MF |
1434 | v_blkno = ocfs2_blocks_for_bytes(sb, i_size_read(dir)); |
1435 | status = ocfs2_extent_map_get_blocks(dir, v_blkno, &p_blkno, NULL, NULL); | |
ccd979bd MF |
1436 | if (status < 0) { |
1437 | mlog_errno(status); | |
1438 | goto bail; | |
1439 | } | |
1440 | ||
1441 | *new_bh = sb_getblk(sb, p_blkno); | |
1442 | if (!*new_bh) { | |
1443 | status = -EIO; | |
1444 | mlog_errno(status); | |
1445 | goto bail; | |
1446 | } | |
1447 | status = 0; | |
1448 | bail: | |
1449 | mlog_exit(status); | |
1450 | return status; | |
1451 | } | |
1452 | ||
5b6a3a2b MF |
1453 | /* |
1454 | * Assumes you already have a cluster lock on the directory. | |
1455 | * | |
1456 | * 'blocks_wanted' is only used if we have an inline directory which | |
1457 | * is to be turned into an extent based one. The size of the dirent to | |
1458 | * insert might be larger than the space gained by growing to just one | |
1459 | * block, so we may have to grow the inode by two blocks in that case. | |
1460 | */ | |
ccd979bd MF |
1461 | static int ocfs2_extend_dir(struct ocfs2_super *osb, |
1462 | struct inode *dir, | |
1463 | struct buffer_head *parent_fe_bh, | |
5b6a3a2b | 1464 | unsigned int blocks_wanted, |
ccd979bd MF |
1465 | struct buffer_head **new_de_bh) |
1466 | { | |
1467 | int status = 0; | |
ee19a779 | 1468 | int credits, num_free_extents, drop_alloc_sem = 0; |
ccd979bd MF |
1469 | loff_t dir_i_size; |
1470 | struct ocfs2_dinode *fe = (struct ocfs2_dinode *) parent_fe_bh->b_data; | |
811f933d | 1471 | struct ocfs2_extent_list *el = &fe->id2.i_list; |
ccd979bd MF |
1472 | struct ocfs2_alloc_context *data_ac = NULL; |
1473 | struct ocfs2_alloc_context *meta_ac = NULL; | |
1fabe148 | 1474 | handle_t *handle = NULL; |
ccd979bd MF |
1475 | struct buffer_head *new_bh = NULL; |
1476 | struct ocfs2_dir_entry * de; | |
1477 | struct super_block *sb = osb->sb; | |
f99b9b7c | 1478 | struct ocfs2_extent_tree et; |
ccd979bd MF |
1479 | |
1480 | mlog_entry_void(); | |
1481 | ||
5b6a3a2b MF |
1482 | if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
1483 | status = ocfs2_expand_inline_dir(dir, parent_fe_bh, | |
1484 | blocks_wanted, &new_bh); | |
1485 | if (status) { | |
1486 | mlog_errno(status); | |
1487 | goto bail; | |
1488 | } | |
1489 | ||
1490 | if (blocks_wanted == 1) { | |
1491 | /* | |
1492 | * If the new dirent will fit inside the space | |
1493 | * created by pushing out to one block, then | |
1494 | * we can complete the operation | |
1495 | * here. Otherwise we have to expand i_size | |
1496 | * and format the 2nd block below. | |
1497 | */ | |
1498 | BUG_ON(new_bh == NULL); | |
1499 | goto bail_bh; | |
1500 | } | |
1501 | ||
1502 | /* | |
1503 | * Get rid of 'new_bh' - we want to format the 2nd | |
1504 | * data block and return that instead. | |
1505 | */ | |
1506 | brelse(new_bh); | |
1507 | new_bh = NULL; | |
1508 | ||
1509 | dir_i_size = i_size_read(dir); | |
1510 | credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS; | |
1511 | goto do_extend; | |
1512 | } | |
1513 | ||
ccd979bd | 1514 | dir_i_size = i_size_read(dir); |
b0697053 MF |
1515 | mlog(0, "extending dir %llu (i_size = %lld)\n", |
1516 | (unsigned long long)OCFS2_I(dir)->ip_blkno, dir_i_size); | |
ccd979bd | 1517 | |
ccd979bd MF |
1518 | /* dir->i_size is always block aligned. */ |
1519 | spin_lock(&OCFS2_I(dir)->ip_lock); | |
1520 | if (dir_i_size == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters)) { | |
1521 | spin_unlock(&OCFS2_I(dir)->ip_lock); | |
8d6220d6 | 1522 | ocfs2_init_dinode_extent_tree(&et, dir, parent_fe_bh); |
f99b9b7c | 1523 | num_free_extents = ocfs2_num_free_extents(osb, dir, &et); |
ccd979bd MF |
1524 | if (num_free_extents < 0) { |
1525 | status = num_free_extents; | |
1526 | mlog_errno(status); | |
1527 | goto bail; | |
1528 | } | |
1529 | ||
1530 | if (!num_free_extents) { | |
811f933d | 1531 | status = ocfs2_reserve_new_metadata(osb, el, &meta_ac); |
ccd979bd MF |
1532 | if (status < 0) { |
1533 | if (status != -ENOSPC) | |
1534 | mlog_errno(status); | |
1535 | goto bail; | |
1536 | } | |
1537 | } | |
1538 | ||
da5cbf2f | 1539 | status = ocfs2_reserve_clusters(osb, 1, &data_ac); |
ccd979bd MF |
1540 | if (status < 0) { |
1541 | if (status != -ENOSPC) | |
1542 | mlog_errno(status); | |
1543 | goto bail; | |
1544 | } | |
1545 | ||
811f933d | 1546 | credits = ocfs2_calc_extend_credits(sb, el, 1); |
ccd979bd MF |
1547 | } else { |
1548 | spin_unlock(&OCFS2_I(dir)->ip_lock); | |
1549 | credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS; | |
1550 | } | |
1551 | ||
5b6a3a2b | 1552 | do_extend: |
ee19a779 JB |
1553 | down_write(&OCFS2_I(dir)->ip_alloc_sem); |
1554 | drop_alloc_sem = 1; | |
1555 | ||
65eff9cc | 1556 | handle = ocfs2_start_trans(osb, credits); |
ccd979bd MF |
1557 | if (IS_ERR(handle)) { |
1558 | status = PTR_ERR(handle); | |
1559 | handle = NULL; | |
1560 | mlog_errno(status); | |
1561 | goto bail; | |
1562 | } | |
1563 | ||
1564 | status = ocfs2_do_extend_dir(osb->sb, handle, dir, parent_fe_bh, | |
1565 | data_ac, meta_ac, &new_bh); | |
1566 | if (status < 0) { | |
1567 | mlog_errno(status); | |
1568 | goto bail; | |
1569 | } | |
1570 | ||
1571 | ocfs2_set_new_buffer_uptodate(dir, new_bh); | |
1572 | ||
1573 | status = ocfs2_journal_access(handle, dir, new_bh, | |
1574 | OCFS2_JOURNAL_ACCESS_CREATE); | |
1575 | if (status < 0) { | |
1576 | mlog_errno(status); | |
1577 | goto bail; | |
1578 | } | |
1579 | memset(new_bh->b_data, 0, sb->s_blocksize); | |
1580 | de = (struct ocfs2_dir_entry *) new_bh->b_data; | |
1581 | de->inode = 0; | |
1582 | de->rec_len = cpu_to_le16(sb->s_blocksize); | |
1583 | status = ocfs2_journal_dirty(handle, new_bh); | |
1584 | if (status < 0) { | |
1585 | mlog_errno(status); | |
1586 | goto bail; | |
1587 | } | |
1588 | ||
1589 | dir_i_size += dir->i_sb->s_blocksize; | |
1590 | i_size_write(dir, dir_i_size); | |
8110b073 | 1591 | dir->i_blocks = ocfs2_inode_sector_count(dir); |
ccd979bd MF |
1592 | status = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh); |
1593 | if (status < 0) { | |
1594 | mlog_errno(status); | |
1595 | goto bail; | |
1596 | } | |
1597 | ||
5b6a3a2b | 1598 | bail_bh: |
ccd979bd MF |
1599 | *new_de_bh = new_bh; |
1600 | get_bh(*new_de_bh); | |
1601 | bail: | |
ee19a779 JB |
1602 | if (drop_alloc_sem) |
1603 | up_write(&OCFS2_I(dir)->ip_alloc_sem); | |
ccd979bd | 1604 | if (handle) |
02dc1af4 | 1605 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
1606 | |
1607 | if (data_ac) | |
1608 | ocfs2_free_alloc_context(data_ac); | |
1609 | if (meta_ac) | |
1610 | ocfs2_free_alloc_context(meta_ac); | |
1611 | ||
a81cb88b | 1612 | brelse(new_bh); |
ccd979bd MF |
1613 | |
1614 | mlog_exit(status); | |
1615 | return status; | |
1616 | } | |
1617 | ||
5b6a3a2b MF |
1618 | static int ocfs2_find_dir_space_id(struct inode *dir, struct buffer_head *di_bh, |
1619 | const char *name, int namelen, | |
1620 | struct buffer_head **ret_de_bh, | |
1621 | unsigned int *blocks_wanted) | |
ccd979bd | 1622 | { |
5b6a3a2b MF |
1623 | int ret; |
1624 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
1625 | struct ocfs2_dir_entry *de, *last_de = NULL; | |
1626 | char *de_buf, *limit; | |
1627 | unsigned long offset = 0; | |
1628 | unsigned int rec_len, new_rec_len; | |
1629 | ||
1630 | de_buf = di->id2.i_data.id_data; | |
1631 | limit = de_buf + i_size_read(dir); | |
1632 | rec_len = OCFS2_DIR_REC_LEN(namelen); | |
ccd979bd | 1633 | |
5b6a3a2b MF |
1634 | while (de_buf < limit) { |
1635 | de = (struct ocfs2_dir_entry *)de_buf; | |
ccd979bd | 1636 | |
5b6a3a2b MF |
1637 | if (!ocfs2_check_dir_entry(dir, de, di_bh, offset)) { |
1638 | ret = -ENOENT; | |
1639 | goto out; | |
1640 | } | |
1641 | if (ocfs2_match(namelen, name, de)) { | |
1642 | ret = -EEXIST; | |
1643 | goto out; | |
1644 | } | |
1645 | if (ocfs2_dirent_would_fit(de, rec_len)) { | |
1646 | /* Ok, we found a spot. Return this bh and let | |
1647 | * the caller actually fill it in. */ | |
1648 | *ret_de_bh = di_bh; | |
1649 | get_bh(*ret_de_bh); | |
1650 | ret = 0; | |
1651 | goto out; | |
1652 | } | |
ccd979bd | 1653 | |
5b6a3a2b MF |
1654 | last_de = de; |
1655 | de_buf += le16_to_cpu(de->rec_len); | |
1656 | offset += le16_to_cpu(de->rec_len); | |
1657 | } | |
ccd979bd | 1658 | |
5b6a3a2b MF |
1659 | /* |
1660 | * We're going to require expansion of the directory - figure | |
1661 | * out how many blocks we'll need so that a place for the | |
1662 | * dirent can be found. | |
1663 | */ | |
1664 | *blocks_wanted = 1; | |
1665 | new_rec_len = le16_to_cpu(last_de->rec_len) + (dir->i_sb->s_blocksize - i_size_read(dir)); | |
1666 | if (new_rec_len < (rec_len + OCFS2_DIR_REC_LEN(last_de->name_len))) | |
1667 | *blocks_wanted = 2; | |
ccd979bd | 1668 | |
5b6a3a2b MF |
1669 | ret = -ENOSPC; |
1670 | out: | |
1671 | return ret; | |
1672 | } | |
1673 | ||
1674 | static int ocfs2_find_dir_space_el(struct inode *dir, const char *name, | |
1675 | int namelen, struct buffer_head **ret_de_bh) | |
1676 | { | |
1677 | unsigned long offset; | |
1678 | struct buffer_head *bh = NULL; | |
1679 | unsigned short rec_len; | |
1680 | struct ocfs2_dir_entry *de; | |
1681 | struct super_block *sb = dir->i_sb; | |
1682 | int status; | |
ccd979bd MF |
1683 | |
1684 | bh = ocfs2_bread(dir, 0, &status, 0); | |
1685 | if (!bh) { | |
1686 | mlog_errno(status); | |
1687 | goto bail; | |
1688 | } | |
1689 | ||
1690 | rec_len = OCFS2_DIR_REC_LEN(namelen); | |
1691 | offset = 0; | |
1692 | de = (struct ocfs2_dir_entry *) bh->b_data; | |
1693 | while (1) { | |
1694 | if ((char *)de >= sb->s_blocksize + bh->b_data) { | |
1695 | brelse(bh); | |
1696 | bh = NULL; | |
1697 | ||
1698 | if (i_size_read(dir) <= offset) { | |
5b6a3a2b MF |
1699 | /* |
1700 | * Caller will have to expand this | |
1701 | * directory. | |
1702 | */ | |
1703 | status = -ENOSPC; | |
ccd979bd MF |
1704 | goto bail; |
1705 | } | |
1706 | bh = ocfs2_bread(dir, | |
1707 | offset >> sb->s_blocksize_bits, | |
1708 | &status, | |
1709 | 0); | |
1710 | if (!bh) { | |
1711 | mlog_errno(status); | |
1712 | goto bail; | |
1713 | } | |
1714 | /* move to next block */ | |
1715 | de = (struct ocfs2_dir_entry *) bh->b_data; | |
1716 | } | |
1717 | if (!ocfs2_check_dir_entry(dir, de, bh, offset)) { | |
1718 | status = -ENOENT; | |
1719 | goto bail; | |
1720 | } | |
1721 | if (ocfs2_match(namelen, name, de)) { | |
1722 | status = -EEXIST; | |
1723 | goto bail; | |
1724 | } | |
8553cf4f | 1725 | if (ocfs2_dirent_would_fit(de, rec_len)) { |
ccd979bd MF |
1726 | /* Ok, we found a spot. Return this bh and let |
1727 | * the caller actually fill it in. */ | |
1728 | *ret_de_bh = bh; | |
1729 | get_bh(*ret_de_bh); | |
1730 | status = 0; | |
1731 | goto bail; | |
1732 | } | |
1733 | offset += le16_to_cpu(de->rec_len); | |
1734 | de = (struct ocfs2_dir_entry *)((char *) de + le16_to_cpu(de->rec_len)); | |
1735 | } | |
1736 | ||
1737 | status = 0; | |
1738 | bail: | |
a81cb88b | 1739 | brelse(bh); |
ccd979bd MF |
1740 | |
1741 | mlog_exit(status); | |
1742 | return status; | |
1743 | } | |
5b6a3a2b MF |
1744 | |
1745 | int ocfs2_prepare_dir_for_insert(struct ocfs2_super *osb, | |
1746 | struct inode *dir, | |
1747 | struct buffer_head *parent_fe_bh, | |
1748 | const char *name, | |
1749 | int namelen, | |
1750 | struct buffer_head **ret_de_bh) | |
1751 | { | |
1752 | int ret; | |
1753 | unsigned int blocks_wanted = 1; | |
1754 | struct buffer_head *bh = NULL; | |
1755 | ||
1756 | mlog(0, "getting ready to insert namelen %d into dir %llu\n", | |
1757 | namelen, (unsigned long long)OCFS2_I(dir)->ip_blkno); | |
1758 | ||
1759 | *ret_de_bh = NULL; | |
1760 | ||
1761 | if (!namelen) { | |
1762 | ret = -EINVAL; | |
1763 | mlog_errno(ret); | |
1764 | goto out; | |
1765 | } | |
1766 | ||
1767 | if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { | |
1768 | ret = ocfs2_find_dir_space_id(dir, parent_fe_bh, name, | |
1769 | namelen, &bh, &blocks_wanted); | |
1770 | } else | |
1771 | ret = ocfs2_find_dir_space_el(dir, name, namelen, &bh); | |
1772 | ||
1773 | if (ret && ret != -ENOSPC) { | |
1774 | mlog_errno(ret); | |
1775 | goto out; | |
1776 | } | |
1777 | ||
1778 | if (ret == -ENOSPC) { | |
1779 | /* | |
1780 | * We have to expand the directory to add this name. | |
1781 | */ | |
1782 | BUG_ON(bh); | |
1783 | ||
1784 | ret = ocfs2_extend_dir(osb, dir, parent_fe_bh, blocks_wanted, | |
1785 | &bh); | |
1786 | if (ret) { | |
1787 | if (ret != -ENOSPC) | |
1788 | mlog_errno(ret); | |
1789 | goto out; | |
1790 | } | |
1791 | ||
1792 | BUG_ON(!bh); | |
1793 | } | |
1794 | ||
1795 | *ret_de_bh = bh; | |
1796 | bh = NULL; | |
1797 | out: | |
a81cb88b | 1798 | brelse(bh); |
5b6a3a2b MF |
1799 | return ret; |
1800 | } |