]>
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> | |
a90714c1 | 43 | #include <linux/quotaops.h> |
9b7895ef | 44 | #include <linux/sort.h> |
ccd979bd MF |
45 | |
46 | #define MLOG_MASK_PREFIX ML_NAMEI | |
47 | #include <cluster/masklog.h> | |
48 | ||
49 | #include "ocfs2.h" | |
50 | ||
51 | #include "alloc.h" | |
c175a518 | 52 | #include "blockcheck.h" |
ccd979bd MF |
53 | #include "dir.h" |
54 | #include "dlmglue.h" | |
55 | #include "extent_map.h" | |
56 | #include "file.h" | |
57 | #include "inode.h" | |
58 | #include "journal.h" | |
59 | #include "namei.h" | |
60 | #include "suballoc.h" | |
316f4b9f | 61 | #include "super.h" |
9b7895ef | 62 | #include "sysfile.h" |
ccd979bd MF |
63 | #include "uptodate.h" |
64 | ||
65 | #include "buffer_head_io.h" | |
66 | ||
316f4b9f MF |
67 | #define NAMEI_RA_CHUNKS 2 |
68 | #define NAMEI_RA_BLOCKS 4 | |
69 | #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS) | |
70 | #define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b)) | |
71 | ||
ccd979bd MF |
72 | static unsigned char ocfs2_filetype_table[] = { |
73 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK | |
74 | }; | |
75 | ||
316f4b9f MF |
76 | static int ocfs2_do_extend_dir(struct super_block *sb, |
77 | handle_t *handle, | |
78 | struct inode *dir, | |
79 | struct buffer_head *parent_fe_bh, | |
80 | struct ocfs2_alloc_context *data_ac, | |
81 | struct ocfs2_alloc_context *meta_ac, | |
82 | struct buffer_head **new_bh); | |
e7c17e43 | 83 | static int ocfs2_dir_indexed(struct inode *inode); |
316f4b9f | 84 | |
87d35a74 MF |
85 | /* |
86 | * These are distinct checks because future versions of the file system will | |
87 | * want to have a trailing dirent structure independent of indexing. | |
88 | */ | |
e7c17e43 | 89 | static int ocfs2_supports_dir_trailer(struct inode *dir) |
87d35a74 | 90 | { |
e7c17e43 MF |
91 | struct ocfs2_super *osb = OCFS2_SB(dir->i_sb); |
92 | ||
87d35a74 MF |
93 | if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) |
94 | return 0; | |
95 | ||
e7c17e43 | 96 | return ocfs2_meta_ecc(osb) || ocfs2_dir_indexed(dir); |
87d35a74 MF |
97 | } |
98 | ||
e7c17e43 MF |
99 | /* |
100 | * "new' here refers to the point at which we're creating a new | |
101 | * directory via "mkdir()", but also when we're expanding an inline | |
102 | * directory. In either case, we don't yet have the indexing bit set | |
103 | * on the directory, so the standard checks will fail in when metaecc | |
104 | * is turned off. Only directory-initialization type functions should | |
105 | * use this then. Everything else wants ocfs2_supports_dir_trailer() | |
106 | */ | |
107 | static int ocfs2_new_dir_wants_trailer(struct inode *dir) | |
87d35a74 | 108 | { |
e7c17e43 MF |
109 | struct ocfs2_super *osb = OCFS2_SB(dir->i_sb); |
110 | ||
111 | return ocfs2_meta_ecc(osb) || | |
112 | ocfs2_supports_indexed_dirs(osb); | |
87d35a74 MF |
113 | } |
114 | ||
115 | static inline unsigned int ocfs2_dir_trailer_blk_off(struct super_block *sb) | |
116 | { | |
117 | return sb->s_blocksize - sizeof(struct ocfs2_dir_block_trailer); | |
118 | } | |
119 | ||
120 | #define ocfs2_trailer_from_bh(_bh, _sb) ((struct ocfs2_dir_block_trailer *) ((_bh)->b_data + ocfs2_dir_trailer_blk_off((_sb)))) | |
121 | ||
c175a518 JB |
122 | /* XXX ocfs2_block_dqtrailer() is similar but not quite - can we make |
123 | * them more consistent? */ | |
124 | struct ocfs2_dir_block_trailer *ocfs2_dir_trailer_from_size(int blocksize, | |
125 | void *data) | |
126 | { | |
127 | char *p = data; | |
128 | ||
129 | p += blocksize - sizeof(struct ocfs2_dir_block_trailer); | |
130 | return (struct ocfs2_dir_block_trailer *)p; | |
131 | } | |
132 | ||
87d35a74 MF |
133 | /* |
134 | * XXX: This is executed once on every dirent. We should consider optimizing | |
135 | * it. | |
136 | */ | |
137 | static int ocfs2_skip_dir_trailer(struct inode *dir, | |
138 | struct ocfs2_dir_entry *de, | |
139 | unsigned long offset, | |
140 | unsigned long blklen) | |
141 | { | |
142 | unsigned long toff = blklen - sizeof(struct ocfs2_dir_block_trailer); | |
143 | ||
e7c17e43 | 144 | if (!ocfs2_supports_dir_trailer(dir)) |
87d35a74 MF |
145 | return 0; |
146 | ||
147 | if (offset != toff) | |
148 | return 0; | |
149 | ||
150 | return 1; | |
151 | } | |
152 | ||
153 | static void ocfs2_init_dir_trailer(struct inode *inode, | |
e7c17e43 | 154 | struct buffer_head *bh, u16 rec_len) |
87d35a74 MF |
155 | { |
156 | struct ocfs2_dir_block_trailer *trailer; | |
157 | ||
158 | trailer = ocfs2_trailer_from_bh(bh, inode->i_sb); | |
159 | strcpy(trailer->db_signature, OCFS2_DIR_TRAILER_SIGNATURE); | |
160 | trailer->db_compat_rec_len = | |
161 | cpu_to_le16(sizeof(struct ocfs2_dir_block_trailer)); | |
162 | trailer->db_parent_dinode = cpu_to_le64(OCFS2_I(inode)->ip_blkno); | |
163 | trailer->db_blkno = cpu_to_le64(bh->b_blocknr); | |
e7c17e43 MF |
164 | trailer->db_free_rec_len = cpu_to_le16(rec_len); |
165 | } | |
166 | /* | |
167 | * Link an unindexed block with a dir trailer structure into the index free | |
168 | * list. This function will modify dirdata_bh, but assumes you've already | |
169 | * passed it to the journal. | |
170 | */ | |
171 | static int ocfs2_dx_dir_link_trailer(struct inode *dir, handle_t *handle, | |
172 | struct buffer_head *dx_root_bh, | |
173 | struct buffer_head *dirdata_bh) | |
174 | { | |
175 | int ret; | |
176 | struct ocfs2_dx_root_block *dx_root; | |
177 | struct ocfs2_dir_block_trailer *trailer; | |
178 | ||
179 | ret = ocfs2_journal_access_dr(handle, dir, dx_root_bh, | |
180 | OCFS2_JOURNAL_ACCESS_WRITE); | |
181 | if (ret) { | |
182 | mlog_errno(ret); | |
183 | goto out; | |
184 | } | |
185 | trailer = ocfs2_trailer_from_bh(dirdata_bh, dir->i_sb); | |
186 | dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data; | |
187 | ||
188 | trailer->db_free_next = dx_root->dr_free_blk; | |
189 | dx_root->dr_free_blk = cpu_to_le64(dirdata_bh->b_blocknr); | |
190 | ||
191 | ocfs2_journal_dirty(handle, dx_root_bh); | |
192 | ||
193 | out: | |
194 | return ret; | |
195 | } | |
196 | ||
197 | static int ocfs2_free_list_at_root(struct ocfs2_dir_lookup_result *res) | |
198 | { | |
199 | return res->dl_prev_leaf_bh == NULL; | |
87d35a74 MF |
200 | } |
201 | ||
4a12ca3a MF |
202 | void ocfs2_free_dir_lookup_result(struct ocfs2_dir_lookup_result *res) |
203 | { | |
4ed8a6bb | 204 | brelse(res->dl_dx_root_bh); |
4a12ca3a | 205 | brelse(res->dl_leaf_bh); |
9b7895ef | 206 | brelse(res->dl_dx_leaf_bh); |
e7c17e43 | 207 | brelse(res->dl_prev_leaf_bh); |
9b7895ef MF |
208 | } |
209 | ||
210 | static int ocfs2_dir_indexed(struct inode *inode) | |
211 | { | |
212 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INDEXED_DIR_FL) | |
213 | return 1; | |
214 | return 0; | |
215 | } | |
216 | ||
4ed8a6bb MF |
217 | static inline int ocfs2_dx_root_inline(struct ocfs2_dx_root_block *dx_root) |
218 | { | |
219 | return dx_root->dr_flags & OCFS2_DX_FLAG_INLINE; | |
220 | } | |
221 | ||
9b7895ef MF |
222 | /* |
223 | * Hashing code adapted from ext3 | |
224 | */ | |
225 | #define DELTA 0x9E3779B9 | |
226 | ||
227 | static void TEA_transform(__u32 buf[4], __u32 const in[]) | |
228 | { | |
229 | __u32 sum = 0; | |
230 | __u32 b0 = buf[0], b1 = buf[1]; | |
231 | __u32 a = in[0], b = in[1], c = in[2], d = in[3]; | |
232 | int n = 16; | |
233 | ||
234 | do { | |
235 | sum += DELTA; | |
236 | b0 += ((b1 << 4)+a) ^ (b1+sum) ^ ((b1 >> 5)+b); | |
237 | b1 += ((b0 << 4)+c) ^ (b0+sum) ^ ((b0 >> 5)+d); | |
238 | } while (--n); | |
239 | ||
240 | buf[0] += b0; | |
241 | buf[1] += b1; | |
242 | } | |
243 | ||
244 | static void str2hashbuf(const char *msg, int len, __u32 *buf, int num) | |
245 | { | |
246 | __u32 pad, val; | |
247 | int i; | |
248 | ||
249 | pad = (__u32)len | ((__u32)len << 8); | |
250 | pad |= pad << 16; | |
251 | ||
252 | val = pad; | |
253 | if (len > num*4) | |
254 | len = num * 4; | |
255 | for (i = 0; i < len; i++) { | |
256 | if ((i % 4) == 0) | |
257 | val = pad; | |
258 | val = msg[i] + (val << 8); | |
259 | if ((i % 4) == 3) { | |
260 | *buf++ = val; | |
261 | val = pad; | |
262 | num--; | |
263 | } | |
264 | } | |
265 | if (--num >= 0) | |
266 | *buf++ = val; | |
267 | while (--num >= 0) | |
268 | *buf++ = pad; | |
269 | } | |
270 | ||
271 | static void ocfs2_dx_dir_name_hash(struct inode *dir, const char *name, int len, | |
272 | struct ocfs2_dx_hinfo *hinfo) | |
273 | { | |
274 | struct ocfs2_super *osb = OCFS2_SB(dir->i_sb); | |
275 | const char *p; | |
276 | __u32 in[8], buf[4]; | |
277 | ||
278 | /* | |
279 | * XXX: Is this really necessary, if the index is never looked | |
280 | * at by readdir? Is a hash value of '0' a bad idea? | |
281 | */ | |
282 | if ((len == 1 && !strncmp(".", name, 1)) || | |
283 | (len == 2 && !strncmp("..", name, 2))) { | |
284 | buf[0] = buf[1] = 0; | |
285 | goto out; | |
286 | } | |
287 | ||
288 | #ifdef OCFS2_DEBUG_DX_DIRS | |
289 | /* | |
290 | * This makes it very easy to debug indexing problems. We | |
291 | * should never allow this to be selected without hand editing | |
292 | * this file though. | |
293 | */ | |
294 | buf[0] = buf[1] = len; | |
295 | goto out; | |
296 | #endif | |
297 | ||
298 | memcpy(buf, osb->osb_dx_seed, sizeof(buf)); | |
299 | ||
300 | p = name; | |
301 | while (len > 0) { | |
302 | str2hashbuf(p, len, in, 4); | |
303 | TEA_transform(buf, in); | |
304 | len -= 16; | |
305 | p += 16; | |
306 | } | |
307 | ||
308 | out: | |
309 | hinfo->major_hash = buf[0]; | |
310 | hinfo->minor_hash = buf[1]; | |
4a12ca3a MF |
311 | } |
312 | ||
23193e51 MF |
313 | /* |
314 | * bh passed here can be an inode block or a dir data block, depending | |
315 | * on the inode inline data flag. | |
316 | */ | |
5eae5b96 MF |
317 | static int ocfs2_check_dir_entry(struct inode * dir, |
318 | struct ocfs2_dir_entry * de, | |
319 | struct buffer_head * bh, | |
320 | unsigned long offset) | |
316f4b9f MF |
321 | { |
322 | const char *error_msg = NULL; | |
323 | const int rlen = le16_to_cpu(de->rec_len); | |
324 | ||
325 | if (rlen < OCFS2_DIR_REC_LEN(1)) | |
326 | error_msg = "rec_len is smaller than minimal"; | |
327 | else if (rlen % 4 != 0) | |
328 | error_msg = "rec_len % 4 != 0"; | |
329 | else if (rlen < OCFS2_DIR_REC_LEN(de->name_len)) | |
330 | error_msg = "rec_len is too small for name_len"; | |
331 | else if (((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize) | |
332 | error_msg = "directory entry across blocks"; | |
333 | ||
334 | if (error_msg != NULL) | |
335 | mlog(ML_ERROR, "bad entry in directory #%llu: %s - " | |
336 | "offset=%lu, inode=%llu, rec_len=%d, name_len=%d\n", | |
337 | (unsigned long long)OCFS2_I(dir)->ip_blkno, error_msg, | |
338 | offset, (unsigned long long)le64_to_cpu(de->inode), rlen, | |
339 | de->name_len); | |
340 | return error_msg == NULL ? 1 : 0; | |
341 | } | |
342 | ||
343 | static inline int ocfs2_match(int len, | |
344 | const char * const name, | |
345 | struct ocfs2_dir_entry *de) | |
346 | { | |
347 | if (len != de->name_len) | |
348 | return 0; | |
349 | if (!de->inode) | |
350 | return 0; | |
351 | return !memcmp(name, de->name, len); | |
352 | } | |
353 | ||
354 | /* | |
355 | * Returns 0 if not found, -1 on failure, and 1 on success | |
356 | */ | |
357 | static int inline ocfs2_search_dirblock(struct buffer_head *bh, | |
358 | struct inode *dir, | |
359 | const char *name, int namelen, | |
360 | unsigned long offset, | |
23193e51 MF |
361 | char *first_de, |
362 | unsigned int bytes, | |
316f4b9f MF |
363 | struct ocfs2_dir_entry **res_dir) |
364 | { | |
365 | struct ocfs2_dir_entry *de; | |
366 | char *dlimit, *de_buf; | |
367 | int de_len; | |
368 | int ret = 0; | |
369 | ||
370 | mlog_entry_void(); | |
371 | ||
23193e51 MF |
372 | de_buf = first_de; |
373 | dlimit = de_buf + bytes; | |
316f4b9f MF |
374 | |
375 | while (de_buf < dlimit) { | |
376 | /* this code is executed quadratically often */ | |
377 | /* do minimal checking `by hand' */ | |
378 | ||
379 | de = (struct ocfs2_dir_entry *) de_buf; | |
380 | ||
381 | if (de_buf + namelen <= dlimit && | |
382 | ocfs2_match(namelen, name, de)) { | |
383 | /* found a match - just to be sure, do a full check */ | |
384 | if (!ocfs2_check_dir_entry(dir, de, bh, offset)) { | |
385 | ret = -1; | |
386 | goto bail; | |
387 | } | |
388 | *res_dir = de; | |
389 | ret = 1; | |
390 | goto bail; | |
391 | } | |
392 | ||
393 | /* prevent looping on a bad block */ | |
394 | de_len = le16_to_cpu(de->rec_len); | |
395 | if (de_len <= 0) { | |
396 | ret = -1; | |
397 | goto bail; | |
398 | } | |
399 | ||
400 | de_buf += de_len; | |
401 | offset += de_len; | |
402 | } | |
403 | ||
404 | bail: | |
405 | mlog_exit(ret); | |
406 | return ret; | |
407 | } | |
408 | ||
23193e51 MF |
409 | static struct buffer_head *ocfs2_find_entry_id(const char *name, |
410 | int namelen, | |
411 | struct inode *dir, | |
412 | struct ocfs2_dir_entry **res_dir) | |
413 | { | |
414 | int ret, found; | |
415 | struct buffer_head *di_bh = NULL; | |
416 | struct ocfs2_dinode *di; | |
417 | struct ocfs2_inline_data *data; | |
418 | ||
b657c95c | 419 | ret = ocfs2_read_inode_block(dir, &di_bh); |
23193e51 MF |
420 | if (ret) { |
421 | mlog_errno(ret); | |
422 | goto out; | |
423 | } | |
424 | ||
425 | di = (struct ocfs2_dinode *)di_bh->b_data; | |
426 | data = &di->id2.i_data; | |
427 | ||
428 | found = ocfs2_search_dirblock(di_bh, dir, name, namelen, 0, | |
429 | data->id_data, i_size_read(dir), res_dir); | |
430 | if (found == 1) | |
431 | return di_bh; | |
432 | ||
433 | brelse(di_bh); | |
434 | out: | |
435 | return NULL; | |
436 | } | |
437 | ||
a22305cc JB |
438 | static int ocfs2_validate_dir_block(struct super_block *sb, |
439 | struct buffer_head *bh) | |
440 | { | |
c175a518 JB |
441 | int rc; |
442 | struct ocfs2_dir_block_trailer *trailer = | |
443 | ocfs2_trailer_from_bh(bh, sb); | |
444 | ||
445 | ||
a22305cc | 446 | /* |
c175a518 | 447 | * We don't validate dirents here, that's handled |
a22305cc JB |
448 | * in-place when the code walks them. |
449 | */ | |
970e4936 JB |
450 | mlog(0, "Validating dirblock %llu\n", |
451 | (unsigned long long)bh->b_blocknr); | |
a22305cc | 452 | |
c175a518 JB |
453 | BUG_ON(!buffer_uptodate(bh)); |
454 | ||
455 | /* | |
456 | * If the ecc fails, we return the error but otherwise | |
457 | * leave the filesystem running. We know any error is | |
458 | * local to this block. | |
459 | * | |
460 | * Note that we are safe to call this even if the directory | |
461 | * doesn't have a trailer. Filesystems without metaecc will do | |
462 | * nothing, and filesystems with it will have one. | |
463 | */ | |
464 | rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &trailer->db_check); | |
465 | if (rc) | |
466 | mlog(ML_ERROR, "Checksum failed for dinode %llu\n", | |
467 | (unsigned long long)bh->b_blocknr); | |
468 | ||
469 | return rc; | |
a22305cc JB |
470 | } |
471 | ||
9b7895ef MF |
472 | /* |
473 | * Validate a directory trailer. | |
474 | * | |
475 | * We check the trailer here rather than in ocfs2_validate_dir_block() | |
476 | * because that function doesn't have the inode to test. | |
477 | */ | |
478 | static int ocfs2_check_dir_trailer(struct inode *dir, struct buffer_head *bh) | |
479 | { | |
480 | int rc = 0; | |
481 | struct ocfs2_dir_block_trailer *trailer; | |
482 | ||
483 | trailer = ocfs2_trailer_from_bh(bh, dir->i_sb); | |
484 | if (!OCFS2_IS_VALID_DIR_TRAILER(trailer)) { | |
485 | rc = -EINVAL; | |
486 | ocfs2_error(dir->i_sb, | |
487 | "Invalid dirblock #%llu: " | |
488 | "signature = %.*s\n", | |
489 | (unsigned long long)bh->b_blocknr, 7, | |
490 | trailer->db_signature); | |
491 | goto out; | |
492 | } | |
493 | if (le64_to_cpu(trailer->db_blkno) != bh->b_blocknr) { | |
494 | rc = -EINVAL; | |
495 | ocfs2_error(dir->i_sb, | |
496 | "Directory block #%llu has an invalid " | |
497 | "db_blkno of %llu", | |
498 | (unsigned long long)bh->b_blocknr, | |
499 | (unsigned long long)le64_to_cpu(trailer->db_blkno)); | |
500 | goto out; | |
501 | } | |
502 | if (le64_to_cpu(trailer->db_parent_dinode) != | |
503 | OCFS2_I(dir)->ip_blkno) { | |
504 | rc = -EINVAL; | |
505 | ocfs2_error(dir->i_sb, | |
506 | "Directory block #%llu on dinode " | |
507 | "#%llu has an invalid parent_dinode " | |
508 | "of %llu", | |
509 | (unsigned long long)bh->b_blocknr, | |
510 | (unsigned long long)OCFS2_I(dir)->ip_blkno, | |
511 | (unsigned long long)le64_to_cpu(trailer->db_blkno)); | |
512 | goto out; | |
513 | } | |
514 | out: | |
515 | return rc; | |
516 | } | |
517 | ||
a22305cc JB |
518 | /* |
519 | * This function forces all errors to -EIO for consistency with its | |
520 | * predecessor, ocfs2_bread(). We haven't audited what returning the | |
521 | * real error codes would do to callers. We log the real codes with | |
522 | * mlog_errno() before we squash them. | |
523 | */ | |
524 | static int ocfs2_read_dir_block(struct inode *inode, u64 v_block, | |
525 | struct buffer_head **bh, int flags) | |
526 | { | |
527 | int rc = 0; | |
528 | struct buffer_head *tmp = *bh; | |
a22305cc | 529 | |
511308d9 JB |
530 | rc = ocfs2_read_virt_blocks(inode, v_block, 1, &tmp, flags, |
531 | ocfs2_validate_dir_block); | |
87d35a74 | 532 | if (rc) { |
a22305cc | 533 | mlog_errno(rc); |
87d35a74 MF |
534 | goto out; |
535 | } | |
536 | ||
87d35a74 | 537 | if (!(flags & OCFS2_BH_READAHEAD) && |
e7c17e43 | 538 | ocfs2_supports_dir_trailer(inode)) { |
9b7895ef MF |
539 | rc = ocfs2_check_dir_trailer(inode, tmp); |
540 | if (rc) { | |
541 | if (!*bh) | |
542 | brelse(tmp); | |
543 | mlog_errno(rc); | |
87d35a74 MF |
544 | goto out; |
545 | } | |
546 | } | |
a22305cc | 547 | |
511308d9 | 548 | /* If ocfs2_read_virt_blocks() got us a new bh, pass it up. */ |
87d35a74 | 549 | if (!*bh) |
a22305cc JB |
550 | *bh = tmp; |
551 | ||
87d35a74 | 552 | out: |
a22305cc JB |
553 | return rc ? -EIO : 0; |
554 | } | |
555 | ||
9b7895ef MF |
556 | /* |
557 | * Read the block at 'phys' which belongs to this directory | |
558 | * inode. This function does no virtual->physical block translation - | |
559 | * what's passed in is assumed to be a valid directory block. | |
560 | */ | |
561 | static int ocfs2_read_dir_block_direct(struct inode *dir, u64 phys, | |
562 | struct buffer_head **bh) | |
563 | { | |
564 | int ret; | |
565 | struct buffer_head *tmp = *bh; | |
566 | ||
567 | ret = ocfs2_read_block(dir, phys, &tmp, ocfs2_validate_dir_block); | |
568 | if (ret) { | |
569 | mlog_errno(ret); | |
570 | goto out; | |
571 | } | |
572 | ||
573 | if (ocfs2_supports_dir_trailer(dir)) { | |
574 | ret = ocfs2_check_dir_trailer(dir, tmp); | |
575 | if (ret) { | |
576 | if (!*bh) | |
577 | brelse(tmp); | |
578 | mlog_errno(ret); | |
579 | goto out; | |
580 | } | |
581 | } | |
582 | ||
583 | if (!ret && !*bh) | |
584 | *bh = tmp; | |
585 | out: | |
586 | return ret; | |
587 | } | |
588 | ||
589 | static int ocfs2_validate_dx_root(struct super_block *sb, | |
590 | struct buffer_head *bh) | |
591 | { | |
592 | int ret; | |
593 | struct ocfs2_dx_root_block *dx_root; | |
594 | ||
595 | BUG_ON(!buffer_uptodate(bh)); | |
596 | ||
597 | dx_root = (struct ocfs2_dx_root_block *) bh->b_data; | |
598 | ||
599 | ret = ocfs2_validate_meta_ecc(sb, bh->b_data, &dx_root->dr_check); | |
600 | if (ret) { | |
601 | mlog(ML_ERROR, | |
602 | "Checksum failed for dir index root block %llu\n", | |
603 | (unsigned long long)bh->b_blocknr); | |
604 | return ret; | |
605 | } | |
606 | ||
607 | if (!OCFS2_IS_VALID_DX_ROOT(dx_root)) { | |
608 | ocfs2_error(sb, | |
609 | "Dir Index Root # %llu has bad signature %.*s", | |
610 | (unsigned long long)le64_to_cpu(dx_root->dr_blkno), | |
611 | 7, dx_root->dr_signature); | |
612 | return -EINVAL; | |
613 | } | |
614 | ||
615 | return 0; | |
616 | } | |
617 | ||
618 | static int ocfs2_read_dx_root(struct inode *dir, struct ocfs2_dinode *di, | |
619 | struct buffer_head **dx_root_bh) | |
620 | { | |
621 | int ret; | |
622 | u64 blkno = le64_to_cpu(di->i_dx_root); | |
623 | struct buffer_head *tmp = *dx_root_bh; | |
624 | ||
625 | ret = ocfs2_read_block(dir, blkno, &tmp, ocfs2_validate_dx_root); | |
626 | ||
627 | /* If ocfs2_read_block() got us a new bh, pass it up. */ | |
628 | if (!ret && !*dx_root_bh) | |
629 | *dx_root_bh = tmp; | |
630 | ||
631 | return ret; | |
632 | } | |
633 | ||
634 | static int ocfs2_validate_dx_leaf(struct super_block *sb, | |
635 | struct buffer_head *bh) | |
636 | { | |
637 | int ret; | |
638 | struct ocfs2_dx_leaf *dx_leaf = (struct ocfs2_dx_leaf *)bh->b_data; | |
639 | ||
640 | BUG_ON(!buffer_uptodate(bh)); | |
641 | ||
642 | ret = ocfs2_validate_meta_ecc(sb, bh->b_data, &dx_leaf->dl_check); | |
643 | if (ret) { | |
644 | mlog(ML_ERROR, | |
645 | "Checksum failed for dir index leaf block %llu\n", | |
646 | (unsigned long long)bh->b_blocknr); | |
647 | return ret; | |
648 | } | |
649 | ||
650 | if (!OCFS2_IS_VALID_DX_LEAF(dx_leaf)) { | |
651 | ocfs2_error(sb, "Dir Index Leaf has bad signature %.*s", | |
652 | 7, dx_leaf->dl_signature); | |
653 | return -EROFS; | |
654 | } | |
655 | ||
656 | return 0; | |
657 | } | |
658 | ||
659 | static int ocfs2_read_dx_leaf(struct inode *dir, u64 blkno, | |
660 | struct buffer_head **dx_leaf_bh) | |
661 | { | |
662 | int ret; | |
663 | struct buffer_head *tmp = *dx_leaf_bh; | |
664 | ||
665 | ret = ocfs2_read_block(dir, blkno, &tmp, ocfs2_validate_dx_leaf); | |
666 | ||
667 | /* If ocfs2_read_block() got us a new bh, pass it up. */ | |
668 | if (!ret && !*dx_leaf_bh) | |
669 | *dx_leaf_bh = tmp; | |
670 | ||
671 | return ret; | |
672 | } | |
673 | ||
674 | /* | |
675 | * Read a series of dx_leaf blocks. This expects all buffer_head | |
676 | * pointers to be NULL on function entry. | |
677 | */ | |
678 | static int ocfs2_read_dx_leaves(struct inode *dir, u64 start, int num, | |
679 | struct buffer_head **dx_leaf_bhs) | |
680 | { | |
681 | int ret; | |
682 | ||
683 | ret = ocfs2_read_blocks(dir, start, num, dx_leaf_bhs, 0, | |
684 | ocfs2_validate_dx_leaf); | |
685 | if (ret) | |
686 | mlog_errno(ret); | |
687 | ||
688 | return ret; | |
689 | } | |
690 | ||
0af4bd38 AB |
691 | static struct buffer_head *ocfs2_find_entry_el(const char *name, int namelen, |
692 | struct inode *dir, | |
693 | struct ocfs2_dir_entry **res_dir) | |
316f4b9f MF |
694 | { |
695 | struct super_block *sb; | |
696 | struct buffer_head *bh_use[NAMEI_RA_SIZE]; | |
697 | struct buffer_head *bh, *ret = NULL; | |
698 | unsigned long start, block, b; | |
699 | int ra_max = 0; /* Number of bh's in the readahead | |
700 | buffer, bh_use[] */ | |
701 | int ra_ptr = 0; /* Current index into readahead | |
702 | buffer */ | |
703 | int num = 0; | |
704 | int nblocks, i, err; | |
705 | ||
706 | mlog_entry_void(); | |
707 | ||
316f4b9f MF |
708 | sb = dir->i_sb; |
709 | ||
710 | nblocks = i_size_read(dir) >> sb->s_blocksize_bits; | |
711 | start = OCFS2_I(dir)->ip_dir_start_lookup; | |
712 | if (start >= nblocks) | |
713 | start = 0; | |
714 | block = start; | |
715 | ||
716 | restart: | |
717 | do { | |
718 | /* | |
719 | * We deal with the read-ahead logic here. | |
720 | */ | |
721 | if (ra_ptr >= ra_max) { | |
722 | /* Refill the readahead buffer */ | |
723 | ra_ptr = 0; | |
724 | b = block; | |
725 | for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) { | |
726 | /* | |
727 | * Terminate if we reach the end of the | |
728 | * directory and must wrap, or if our | |
729 | * search has finished at this block. | |
730 | */ | |
731 | if (b >= nblocks || (num && block == start)) { | |
732 | bh_use[ra_max] = NULL; | |
733 | break; | |
734 | } | |
735 | num++; | |
736 | ||
a22305cc JB |
737 | bh = NULL; |
738 | err = ocfs2_read_dir_block(dir, b++, &bh, | |
739 | OCFS2_BH_READAHEAD); | |
316f4b9f MF |
740 | bh_use[ra_max] = bh; |
741 | } | |
742 | } | |
743 | if ((bh = bh_use[ra_ptr++]) == NULL) | |
744 | goto next; | |
a22305cc | 745 | if (ocfs2_read_dir_block(dir, block, &bh, 0)) { |
5e0b3dec | 746 | /* read error, skip block & hope for the best. |
a22305cc | 747 | * ocfs2_read_dir_block() has released the bh. */ |
316f4b9f MF |
748 | ocfs2_error(dir->i_sb, "reading directory %llu, " |
749 | "offset %lu\n", | |
750 | (unsigned long long)OCFS2_I(dir)->ip_blkno, | |
751 | block); | |
316f4b9f MF |
752 | goto next; |
753 | } | |
754 | i = ocfs2_search_dirblock(bh, dir, name, namelen, | |
755 | block << sb->s_blocksize_bits, | |
23193e51 | 756 | bh->b_data, sb->s_blocksize, |
316f4b9f MF |
757 | res_dir); |
758 | if (i == 1) { | |
759 | OCFS2_I(dir)->ip_dir_start_lookup = block; | |
760 | ret = bh; | |
761 | goto cleanup_and_exit; | |
762 | } else { | |
763 | brelse(bh); | |
764 | if (i < 0) | |
765 | goto cleanup_and_exit; | |
766 | } | |
767 | next: | |
768 | if (++block >= nblocks) | |
769 | block = 0; | |
770 | } while (block != start); | |
771 | ||
772 | /* | |
773 | * If the directory has grown while we were searching, then | |
774 | * search the last part of the directory before giving up. | |
775 | */ | |
776 | block = nblocks; | |
777 | nblocks = i_size_read(dir) >> sb->s_blocksize_bits; | |
778 | if (block < nblocks) { | |
779 | start = 0; | |
780 | goto restart; | |
781 | } | |
782 | ||
783 | cleanup_and_exit: | |
784 | /* Clean up the read-ahead blocks */ | |
785 | for (; ra_ptr < ra_max; ra_ptr++) | |
786 | brelse(bh_use[ra_ptr]); | |
787 | ||
788 | mlog_exit_ptr(ret); | |
789 | return ret; | |
790 | } | |
791 | ||
9b7895ef MF |
792 | static int ocfs2_dx_dir_lookup_rec(struct inode *inode, |
793 | struct ocfs2_extent_list *el, | |
794 | u32 major_hash, | |
795 | u32 *ret_cpos, | |
796 | u64 *ret_phys_blkno, | |
797 | unsigned int *ret_clen) | |
23193e51 | 798 | { |
9b7895ef MF |
799 | int ret = 0, i, found; |
800 | struct buffer_head *eb_bh = NULL; | |
801 | struct ocfs2_extent_block *eb; | |
802 | struct ocfs2_extent_rec *rec = NULL; | |
23193e51 | 803 | |
9b7895ef MF |
804 | if (el->l_tree_depth) { |
805 | ret = ocfs2_find_leaf(inode, el, major_hash, &eb_bh); | |
806 | if (ret) { | |
807 | mlog_errno(ret); | |
808 | goto out; | |
809 | } | |
23193e51 | 810 | |
9b7895ef MF |
811 | eb = (struct ocfs2_extent_block *) eb_bh->b_data; |
812 | el = &eb->h_list; | |
4a12ca3a | 813 | |
9b7895ef MF |
814 | if (el->l_tree_depth) { |
815 | ocfs2_error(inode->i_sb, | |
816 | "Inode %lu has non zero tree depth in " | |
817 | "btree tree block %llu\n", inode->i_ino, | |
818 | (unsigned long long)eb_bh->b_blocknr); | |
819 | ret = -EROFS; | |
820 | goto out; | |
821 | } | |
822 | } | |
823 | ||
824 | found = 0; | |
825 | for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) { | |
826 | rec = &el->l_recs[i]; | |
827 | ||
828 | if (le32_to_cpu(rec->e_cpos) <= major_hash) { | |
829 | found = 1; | |
830 | break; | |
831 | } | |
832 | } | |
833 | ||
834 | if (!found) { | |
835 | ocfs2_error(inode->i_sb, "Inode %lu has bad extent " | |
836 | "record (%u, %u, 0) in btree", inode->i_ino, | |
837 | le32_to_cpu(rec->e_cpos), | |
838 | ocfs2_rec_clusters(el, rec)); | |
839 | ret = -EROFS; | |
840 | goto out; | |
841 | } | |
842 | ||
843 | if (ret_phys_blkno) | |
844 | *ret_phys_blkno = le64_to_cpu(rec->e_blkno); | |
845 | if (ret_cpos) | |
846 | *ret_cpos = le32_to_cpu(rec->e_cpos); | |
847 | if (ret_clen) | |
848 | *ret_clen = le16_to_cpu(rec->e_leaf_clusters); | |
849 | ||
850 | out: | |
851 | brelse(eb_bh); | |
852 | return ret; | |
23193e51 MF |
853 | } |
854 | ||
5b6a3a2b | 855 | /* |
9b7895ef MF |
856 | * Returns the block index, from the start of the cluster which this |
857 | * hash belongs too. | |
5b6a3a2b | 858 | */ |
4ed8a6bb MF |
859 | static inline unsigned int __ocfs2_dx_dir_hash_idx(struct ocfs2_super *osb, |
860 | u32 minor_hash) | |
38760e24 | 861 | { |
9b7895ef MF |
862 | return minor_hash & osb->osb_dx_mask; |
863 | } | |
5b6a3a2b | 864 | |
4ed8a6bb MF |
865 | static inline unsigned int ocfs2_dx_dir_hash_idx(struct ocfs2_super *osb, |
866 | struct ocfs2_dx_hinfo *hinfo) | |
867 | { | |
868 | return __ocfs2_dx_dir_hash_idx(osb, hinfo->minor_hash); | |
869 | } | |
870 | ||
9b7895ef MF |
871 | static int ocfs2_dx_dir_lookup(struct inode *inode, |
872 | struct ocfs2_extent_list *el, | |
873 | struct ocfs2_dx_hinfo *hinfo, | |
874 | u32 *ret_cpos, | |
875 | u64 *ret_phys_blkno) | |
876 | { | |
877 | int ret = 0; | |
878 | unsigned int cend, uninitialized_var(clen); | |
879 | u32 uninitialized_var(cpos); | |
880 | u64 uninitialized_var(blkno); | |
881 | u32 name_hash = hinfo->major_hash; | |
13723d00 | 882 | |
9b7895ef MF |
883 | ret = ocfs2_dx_dir_lookup_rec(inode, el, name_hash, &cpos, &blkno, |
884 | &clen); | |
38760e24 MF |
885 | if (ret) { |
886 | mlog_errno(ret); | |
887 | goto out; | |
888 | } | |
889 | ||
9b7895ef MF |
890 | cend = cpos + clen; |
891 | if (name_hash >= cend) { | |
892 | /* We want the last cluster */ | |
893 | blkno += ocfs2_clusters_to_blocks(inode->i_sb, clen - 1); | |
894 | cpos += clen - 1; | |
895 | } else { | |
896 | blkno += ocfs2_clusters_to_blocks(inode->i_sb, | |
897 | name_hash - cpos); | |
898 | cpos = name_hash; | |
899 | } | |
38760e24 | 900 | |
9b7895ef MF |
901 | /* |
902 | * We now have the cluster which should hold our entry. To | |
903 | * find the exact block from the start of the cluster to | |
904 | * search, we take the lower bits of the hash. | |
905 | */ | |
906 | blkno += ocfs2_dx_dir_hash_idx(OCFS2_SB(inode->i_sb), hinfo); | |
907 | ||
908 | if (ret_phys_blkno) | |
909 | *ret_phys_blkno = blkno; | |
910 | if (ret_cpos) | |
911 | *ret_cpos = cpos; | |
38760e24 MF |
912 | |
913 | out: | |
9b7895ef | 914 | |
38760e24 MF |
915 | return ret; |
916 | } | |
917 | ||
9b7895ef MF |
918 | static int ocfs2_dx_dir_search(const char *name, int namelen, |
919 | struct inode *dir, | |
4ed8a6bb | 920 | struct ocfs2_dx_root_block *dx_root, |
9b7895ef | 921 | struct ocfs2_dir_lookup_result *res) |
316f4b9f | 922 | { |
9b7895ef MF |
923 | int ret, i, found; |
924 | u64 uninitialized_var(phys); | |
925 | struct buffer_head *dx_leaf_bh = NULL; | |
926 | struct ocfs2_dx_leaf *dx_leaf; | |
927 | struct ocfs2_dx_entry *dx_entry = NULL; | |
928 | struct buffer_head *dir_ent_bh = NULL; | |
929 | struct ocfs2_dir_entry *dir_ent = NULL; | |
930 | struct ocfs2_dx_hinfo *hinfo = &res->dl_hinfo; | |
4ed8a6bb MF |
931 | struct ocfs2_extent_list *dr_el; |
932 | struct ocfs2_dx_entry_list *entry_list; | |
9b7895ef MF |
933 | |
934 | ocfs2_dx_dir_name_hash(dir, name, namelen, &res->dl_hinfo); | |
935 | ||
4ed8a6bb MF |
936 | if (ocfs2_dx_root_inline(dx_root)) { |
937 | entry_list = &dx_root->dr_entries; | |
938 | goto search; | |
939 | } | |
940 | ||
941 | dr_el = &dx_root->dr_list; | |
942 | ||
9b7895ef MF |
943 | ret = ocfs2_dx_dir_lookup(dir, dr_el, hinfo, NULL, &phys); |
944 | if (ret) { | |
945 | mlog_errno(ret); | |
946 | goto out; | |
947 | } | |
316f4b9f | 948 | |
9b7895ef MF |
949 | mlog(0, "Dir %llu: name: \"%.*s\", lookup of hash: %u.0x%x " |
950 | "returns: %llu\n", | |
951 | (unsigned long long)OCFS2_I(dir)->ip_blkno, | |
952 | namelen, name, hinfo->major_hash, hinfo->minor_hash, | |
953 | (unsigned long long)phys); | |
316f4b9f | 954 | |
9b7895ef MF |
955 | ret = ocfs2_read_dx_leaf(dir, phys, &dx_leaf_bh); |
956 | if (ret) { | |
957 | mlog_errno(ret); | |
958 | goto out; | |
959 | } | |
13723d00 | 960 | |
9b7895ef MF |
961 | dx_leaf = (struct ocfs2_dx_leaf *) dx_leaf_bh->b_data; |
962 | ||
963 | mlog(0, "leaf info: num_used: %d, count: %d\n", | |
964 | le16_to_cpu(dx_leaf->dl_list.de_num_used), | |
965 | le16_to_cpu(dx_leaf->dl_list.de_count)); | |
966 | ||
4ed8a6bb MF |
967 | entry_list = &dx_leaf->dl_list; |
968 | ||
969 | search: | |
9b7895ef MF |
970 | /* |
971 | * Empty leaf is legal, so no need to check for that. | |
972 | */ | |
973 | found = 0; | |
4ed8a6bb MF |
974 | for (i = 0; i < le16_to_cpu(entry_list->de_num_used); i++) { |
975 | dx_entry = &entry_list->de_entries[i]; | |
9b7895ef MF |
976 | |
977 | if (hinfo->major_hash != le32_to_cpu(dx_entry->dx_major_hash) | |
978 | || hinfo->minor_hash != le32_to_cpu(dx_entry->dx_minor_hash)) | |
979 | continue; | |
980 | ||
981 | /* | |
982 | * Search unindexed leaf block now. We're not | |
983 | * guaranteed to find anything. | |
984 | */ | |
985 | ret = ocfs2_read_dir_block_direct(dir, | |
986 | le64_to_cpu(dx_entry->dx_dirent_blk), | |
987 | &dir_ent_bh); | |
988 | if (ret) { | |
989 | mlog_errno(ret); | |
990 | goto out; | |
316f4b9f | 991 | } |
9b7895ef MF |
992 | |
993 | /* | |
994 | * XXX: We should check the unindexed block here, | |
995 | * before using it. | |
996 | */ | |
997 | ||
998 | found = ocfs2_search_dirblock(dir_ent_bh, dir, name, namelen, | |
999 | 0, dir_ent_bh->b_data, | |
1000 | dir->i_sb->s_blocksize, &dir_ent); | |
1001 | if (found == 1) | |
1002 | break; | |
1003 | ||
1004 | if (found == -1) { | |
1005 | /* This means we found a bad directory entry. */ | |
1006 | ret = -EIO; | |
1007 | mlog_errno(ret); | |
1008 | goto out; | |
1009 | } | |
1010 | ||
1011 | brelse(dir_ent_bh); | |
1012 | dir_ent_bh = NULL; | |
1013 | } | |
1014 | ||
1015 | if (found <= 0) { | |
1016 | ret = -ENOENT; | |
1017 | goto out; | |
1018 | } | |
1019 | ||
1020 | res->dl_leaf_bh = dir_ent_bh; | |
1021 | res->dl_entry = dir_ent; | |
1022 | res->dl_dx_leaf_bh = dx_leaf_bh; | |
1023 | res->dl_dx_entry = dx_entry; | |
1024 | ||
1025 | ret = 0; | |
1026 | out: | |
1027 | if (ret) { | |
1028 | brelse(dx_leaf_bh); | |
1029 | brelse(dir_ent_bh); | |
1030 | } | |
1031 | return ret; | |
1032 | } | |
1033 | ||
1034 | static int ocfs2_find_entry_dx(const char *name, int namelen, | |
1035 | struct inode *dir, | |
1036 | struct ocfs2_dir_lookup_result *lookup) | |
1037 | { | |
1038 | int ret; | |
1039 | struct buffer_head *di_bh = NULL; | |
1040 | struct ocfs2_dinode *di; | |
1041 | struct buffer_head *dx_root_bh = NULL; | |
1042 | struct ocfs2_dx_root_block *dx_root; | |
1043 | ||
1044 | ret = ocfs2_read_inode_block(dir, &di_bh); | |
1045 | if (ret) { | |
1046 | mlog_errno(ret); | |
1047 | goto out; | |
1048 | } | |
1049 | ||
1050 | di = (struct ocfs2_dinode *)di_bh->b_data; | |
1051 | ||
1052 | ret = ocfs2_read_dx_root(dir, di, &dx_root_bh); | |
1053 | if (ret) { | |
1054 | mlog_errno(ret); | |
1055 | goto out; | |
1056 | } | |
1057 | dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data; | |
1058 | ||
4ed8a6bb | 1059 | ret = ocfs2_dx_dir_search(name, namelen, dir, dx_root, lookup); |
9b7895ef MF |
1060 | if (ret) { |
1061 | if (ret != -ENOENT) | |
1062 | mlog_errno(ret); | |
1063 | goto out; | |
1064 | } | |
1065 | ||
4ed8a6bb MF |
1066 | lookup->dl_dx_root_bh = dx_root_bh; |
1067 | dx_root_bh = NULL; | |
9b7895ef MF |
1068 | out: |
1069 | brelse(di_bh); | |
1070 | brelse(dx_root_bh); | |
1071 | return ret; | |
1072 | } | |
1073 | ||
1074 | /* | |
1075 | * Try to find an entry of the provided name within 'dir'. | |
1076 | * | |
1077 | * If nothing was found, -ENOENT is returned. Otherwise, zero is | |
1078 | * returned and the struct 'res' will contain information useful to | |
1079 | * other directory manipulation functions. | |
1080 | * | |
1081 | * Caller can NOT assume anything about the contents of the | |
1082 | * buffer_heads - they are passed back only so that it can be passed | |
1083 | * into any one of the manipulation functions (add entry, delete | |
1084 | * entry, etc). As an example, bh in the extent directory case is a | |
1085 | * data block, in the inline-data case it actually points to an inode, | |
1086 | * in the indexed directory case, multiple buffers are involved. | |
1087 | */ | |
1088 | int ocfs2_find_entry(const char *name, int namelen, | |
1089 | struct inode *dir, struct ocfs2_dir_lookup_result *lookup) | |
1090 | { | |
1091 | struct buffer_head *bh; | |
1092 | struct ocfs2_dir_entry *res_dir = NULL; | |
1093 | ||
1094 | if (ocfs2_dir_indexed(dir)) | |
1095 | return ocfs2_find_entry_dx(name, namelen, dir, lookup); | |
1096 | ||
1097 | /* | |
1098 | * The unindexed dir code only uses part of the lookup | |
1099 | * structure, so there's no reason to push it down further | |
1100 | * than this. | |
1101 | */ | |
1102 | if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) | |
1103 | bh = ocfs2_find_entry_id(name, namelen, dir, &res_dir); | |
1104 | else | |
1105 | bh = ocfs2_find_entry_el(name, namelen, dir, &res_dir); | |
1106 | ||
1107 | if (bh == NULL) | |
1108 | return -ENOENT; | |
1109 | ||
1110 | lookup->dl_leaf_bh = bh; | |
1111 | lookup->dl_entry = res_dir; | |
1112 | return 0; | |
1113 | } | |
1114 | ||
1115 | /* | |
1116 | * Update inode number and type of a previously found directory entry. | |
1117 | */ | |
1118 | int ocfs2_update_entry(struct inode *dir, handle_t *handle, | |
1119 | struct ocfs2_dir_lookup_result *res, | |
1120 | struct inode *new_entry_inode) | |
1121 | { | |
1122 | int ret; | |
1123 | ocfs2_journal_access_func access = ocfs2_journal_access_db; | |
1124 | struct ocfs2_dir_entry *de = res->dl_entry; | |
1125 | struct buffer_head *de_bh = res->dl_leaf_bh; | |
1126 | ||
1127 | /* | |
1128 | * The same code works fine for both inline-data and extent | |
1129 | * based directories, so no need to split this up. The only | |
1130 | * difference is the journal_access function. | |
1131 | */ | |
1132 | ||
1133 | if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) | |
1134 | access = ocfs2_journal_access_di; | |
1135 | ||
1136 | ret = access(handle, dir, de_bh, OCFS2_JOURNAL_ACCESS_WRITE); | |
1137 | if (ret) { | |
1138 | mlog_errno(ret); | |
1139 | goto out; | |
1140 | } | |
1141 | ||
1142 | de->inode = cpu_to_le64(OCFS2_I(new_entry_inode)->ip_blkno); | |
1143 | ocfs2_set_de_type(de, new_entry_inode->i_mode); | |
1144 | ||
1145 | ocfs2_journal_dirty(handle, de_bh); | |
1146 | ||
1147 | out: | |
1148 | return ret; | |
1149 | } | |
1150 | ||
1151 | /* | |
1152 | * __ocfs2_delete_entry deletes a directory entry by merging it with the | |
1153 | * previous entry | |
1154 | */ | |
1155 | static int __ocfs2_delete_entry(handle_t *handle, struct inode *dir, | |
1156 | struct ocfs2_dir_entry *de_del, | |
1157 | struct buffer_head *bh, char *first_de, | |
1158 | unsigned int bytes) | |
1159 | { | |
1160 | struct ocfs2_dir_entry *de, *pde; | |
1161 | int i, status = -ENOENT; | |
1162 | ocfs2_journal_access_func access = ocfs2_journal_access_db; | |
1163 | ||
1164 | mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p)\n", handle, dir, de_del, bh); | |
1165 | ||
1166 | if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) | |
1167 | access = ocfs2_journal_access_di; | |
1168 | ||
1169 | i = 0; | |
1170 | pde = NULL; | |
1171 | de = (struct ocfs2_dir_entry *) first_de; | |
1172 | while (i < bytes) { | |
1173 | if (!ocfs2_check_dir_entry(dir, de, bh, i)) { | |
1174 | status = -EIO; | |
1175 | mlog_errno(status); | |
1176 | goto bail; | |
1177 | } | |
1178 | if (de == de_del) { | |
1179 | status = access(handle, dir, bh, | |
13723d00 | 1180 | OCFS2_JOURNAL_ACCESS_WRITE); |
316f4b9f MF |
1181 | if (status < 0) { |
1182 | status = -EIO; | |
1183 | mlog_errno(status); | |
1184 | goto bail; | |
1185 | } | |
1186 | if (pde) | |
0dd3256e MS |
1187 | le16_add_cpu(&pde->rec_len, |
1188 | le16_to_cpu(de->rec_len)); | |
316f4b9f MF |
1189 | else |
1190 | de->inode = 0; | |
1191 | dir->i_version++; | |
1192 | status = ocfs2_journal_dirty(handle, bh); | |
1193 | goto bail; | |
1194 | } | |
1195 | i += le16_to_cpu(de->rec_len); | |
1196 | pde = de; | |
1197 | de = (struct ocfs2_dir_entry *)((char *)de + le16_to_cpu(de->rec_len)); | |
1198 | } | |
1199 | bail: | |
1200 | mlog_exit(status); | |
1201 | return status; | |
1202 | } | |
1203 | ||
e7c17e43 MF |
1204 | static unsigned int ocfs2_figure_dirent_hole(struct ocfs2_dir_entry *de) |
1205 | { | |
1206 | unsigned int hole; | |
1207 | ||
1208 | if (le64_to_cpu(de->inode) == 0) | |
1209 | hole = le16_to_cpu(de->rec_len); | |
1210 | else | |
1211 | hole = le16_to_cpu(de->rec_len) - | |
1212 | OCFS2_DIR_REC_LEN(de->name_len); | |
1213 | ||
1214 | return hole; | |
1215 | } | |
1216 | ||
1217 | static int ocfs2_find_max_rec_len(struct super_block *sb, | |
1218 | struct buffer_head *dirblock_bh) | |
1219 | { | |
1220 | int size, this_hole, largest_hole = 0; | |
1221 | char *trailer, *de_buf, *limit, *start = dirblock_bh->b_data; | |
1222 | struct ocfs2_dir_entry *de; | |
1223 | ||
1224 | trailer = (char *)ocfs2_trailer_from_bh(dirblock_bh, sb); | |
1225 | size = ocfs2_dir_trailer_blk_off(sb); | |
1226 | limit = start + size; | |
1227 | de_buf = start; | |
1228 | de = (struct ocfs2_dir_entry *)de_buf; | |
1229 | do { | |
1230 | if (de_buf != trailer) { | |
1231 | this_hole = ocfs2_figure_dirent_hole(de); | |
1232 | if (this_hole > largest_hole) | |
1233 | largest_hole = this_hole; | |
1234 | } | |
1235 | ||
1236 | de_buf += le16_to_cpu(de->rec_len); | |
1237 | de = (struct ocfs2_dir_entry *)de_buf; | |
1238 | } while (de_buf < limit); | |
1239 | ||
1240 | if (largest_hole >= OCFS2_DIR_MIN_REC_LEN) | |
1241 | return largest_hole; | |
1242 | return 0; | |
1243 | } | |
1244 | ||
4ed8a6bb MF |
1245 | static void ocfs2_dx_list_remove_entry(struct ocfs2_dx_entry_list *entry_list, |
1246 | int index) | |
9b7895ef | 1247 | { |
4ed8a6bb | 1248 | int num_used = le16_to_cpu(entry_list->de_num_used); |
9b7895ef MF |
1249 | |
1250 | if (num_used == 1 || index == (num_used - 1)) | |
1251 | goto clear; | |
1252 | ||
4ed8a6bb MF |
1253 | memmove(&entry_list->de_entries[index], |
1254 | &entry_list->de_entries[index + 1], | |
9b7895ef MF |
1255 | (num_used - index - 1)*sizeof(struct ocfs2_dx_entry)); |
1256 | clear: | |
1257 | num_used--; | |
4ed8a6bb | 1258 | memset(&entry_list->de_entries[num_used], 0, |
9b7895ef | 1259 | sizeof(struct ocfs2_dx_entry)); |
4ed8a6bb | 1260 | entry_list->de_num_used = cpu_to_le16(num_used); |
9b7895ef MF |
1261 | } |
1262 | ||
1263 | static int ocfs2_delete_entry_dx(handle_t *handle, struct inode *dir, | |
1264 | struct ocfs2_dir_lookup_result *lookup) | |
1265 | { | |
e7c17e43 | 1266 | int ret, index, max_rec_len, add_to_free_list = 0; |
4ed8a6bb | 1267 | struct buffer_head *dx_root_bh = lookup->dl_dx_root_bh; |
9b7895ef MF |
1268 | struct buffer_head *leaf_bh = lookup->dl_leaf_bh; |
1269 | struct ocfs2_dx_leaf *dx_leaf; | |
1270 | struct ocfs2_dx_entry *dx_entry = lookup->dl_dx_entry; | |
e7c17e43 | 1271 | struct ocfs2_dir_block_trailer *trailer; |
4ed8a6bb MF |
1272 | struct ocfs2_dx_root_block *dx_root; |
1273 | struct ocfs2_dx_entry_list *entry_list; | |
1274 | ||
e7c17e43 MF |
1275 | /* |
1276 | * This function gets a bit messy because we might have to | |
1277 | * modify the root block, regardless of whether the indexed | |
1278 | * entries are stored inline. | |
1279 | */ | |
1280 | ||
1281 | /* | |
1282 | * *Only* set 'entry_list' here, based on where we're looking | |
1283 | * for the indexed entries. Later, we might still want to | |
1284 | * journal both blocks, based on free list state. | |
1285 | */ | |
4ed8a6bb MF |
1286 | dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data; |
1287 | if (ocfs2_dx_root_inline(dx_root)) { | |
1288 | entry_list = &dx_root->dr_entries; | |
1289 | } else { | |
1290 | dx_leaf = (struct ocfs2_dx_leaf *) lookup->dl_dx_leaf_bh->b_data; | |
1291 | entry_list = &dx_leaf->dl_list; | |
1292 | } | |
9b7895ef | 1293 | |
9b7895ef MF |
1294 | /* Neither of these are a disk corruption - that should have |
1295 | * been caught by lookup, before we got here. */ | |
4ed8a6bb MF |
1296 | BUG_ON(le16_to_cpu(entry_list->de_count) <= 0); |
1297 | BUG_ON(le16_to_cpu(entry_list->de_num_used) <= 0); | |
9b7895ef | 1298 | |
4ed8a6bb | 1299 | index = (char *)dx_entry - (char *)entry_list->de_entries; |
9b7895ef MF |
1300 | index /= sizeof(*dx_entry); |
1301 | ||
4ed8a6bb | 1302 | if (index >= le16_to_cpu(entry_list->de_num_used)) { |
9b7895ef | 1303 | mlog(ML_ERROR, "Dir %llu: Bad dx_entry ptr idx %d, (%p, %p)\n", |
4ed8a6bb MF |
1304 | (unsigned long long)OCFS2_I(dir)->ip_blkno, index, |
1305 | entry_list, dx_entry); | |
9b7895ef MF |
1306 | return -EIO; |
1307 | } | |
1308 | ||
e7c17e43 MF |
1309 | /* |
1310 | * We know that removal of this dirent will leave enough room | |
1311 | * for a new one, so add this block to the free list if it | |
1312 | * isn't already there. | |
1313 | */ | |
1314 | trailer = ocfs2_trailer_from_bh(leaf_bh, dir->i_sb); | |
1315 | if (trailer->db_free_rec_len == 0) | |
1316 | add_to_free_list = 1; | |
1317 | ||
9b7895ef | 1318 | /* |
4ed8a6bb MF |
1319 | * Add the block holding our index into the journal before |
1320 | * removing the unindexed entry. If we get an error return | |
1321 | * from __ocfs2_delete_entry(), then it hasn't removed the | |
1322 | * entry yet. Likewise, successful return means we *must* | |
1323 | * remove the indexed entry. | |
1324 | * | |
1325 | * We're also careful to journal the root tree block here if | |
1326 | * we're going to be adding to the start of the free list. | |
9b7895ef | 1327 | */ |
e7c17e43 | 1328 | if (add_to_free_list || ocfs2_dx_root_inline(dx_root)) { |
4ed8a6bb MF |
1329 | ret = ocfs2_journal_access_dr(handle, dir, dx_root_bh, |
1330 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1331 | if (ret) { | |
1332 | mlog_errno(ret); | |
1333 | goto out; | |
1334 | } | |
e7c17e43 MF |
1335 | } |
1336 | ||
1337 | if (!ocfs2_dx_root_inline(dx_root)) { | |
4ed8a6bb MF |
1338 | ret = ocfs2_journal_access_dl(handle, dir, |
1339 | lookup->dl_dx_leaf_bh, | |
1340 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1341 | if (ret) { | |
1342 | mlog_errno(ret); | |
1343 | goto out; | |
1344 | } | |
9b7895ef MF |
1345 | } |
1346 | ||
4ed8a6bb MF |
1347 | mlog(0, "Dir %llu: delete entry at index: %d\n", |
1348 | (unsigned long long)OCFS2_I(dir)->ip_blkno, index); | |
1349 | ||
9b7895ef MF |
1350 | ret = __ocfs2_delete_entry(handle, dir, lookup->dl_entry, |
1351 | leaf_bh, leaf_bh->b_data, leaf_bh->b_size); | |
1352 | if (ret) { | |
1353 | mlog_errno(ret); | |
1354 | goto out; | |
1355 | } | |
1356 | ||
e7c17e43 MF |
1357 | max_rec_len = ocfs2_find_max_rec_len(dir->i_sb, leaf_bh); |
1358 | trailer->db_free_rec_len = cpu_to_le16(max_rec_len); | |
1359 | if (add_to_free_list) { | |
1360 | trailer->db_free_next = dx_root->dr_free_blk; | |
1361 | dx_root->dr_free_blk = cpu_to_le64(leaf_bh->b_blocknr); | |
1362 | ocfs2_journal_dirty(handle, dx_root_bh); | |
1363 | } | |
1364 | ||
1365 | /* leaf_bh was journal_accessed for us in __ocfs2_delete_entry */ | |
1366 | ocfs2_journal_dirty(handle, leaf_bh); | |
1367 | ||
4ed8a6bb | 1368 | ocfs2_dx_list_remove_entry(entry_list, index); |
9b7895ef | 1369 | |
4ed8a6bb MF |
1370 | if (ocfs2_dx_root_inline(dx_root)) |
1371 | ocfs2_journal_dirty(handle, dx_root_bh); | |
1372 | else | |
1373 | ocfs2_journal_dirty(handle, lookup->dl_dx_leaf_bh); | |
9b7895ef MF |
1374 | |
1375 | out: | |
1376 | return ret; | |
1377 | } | |
1378 | ||
5b6a3a2b MF |
1379 | static inline int ocfs2_delete_entry_id(handle_t *handle, |
1380 | struct inode *dir, | |
1381 | struct ocfs2_dir_entry *de_del, | |
1382 | struct buffer_head *bh) | |
1383 | { | |
1384 | int ret; | |
1385 | struct buffer_head *di_bh = NULL; | |
1386 | struct ocfs2_dinode *di; | |
1387 | struct ocfs2_inline_data *data; | |
1388 | ||
b657c95c | 1389 | ret = ocfs2_read_inode_block(dir, &di_bh); |
5b6a3a2b MF |
1390 | if (ret) { |
1391 | mlog_errno(ret); | |
1392 | goto out; | |
1393 | } | |
1394 | ||
1395 | di = (struct ocfs2_dinode *)di_bh->b_data; | |
1396 | data = &di->id2.i_data; | |
1397 | ||
1398 | ret = __ocfs2_delete_entry(handle, dir, de_del, bh, data->id_data, | |
1399 | i_size_read(dir)); | |
1400 | ||
1401 | brelse(di_bh); | |
1402 | out: | |
1403 | return ret; | |
1404 | } | |
1405 | ||
1406 | static inline int ocfs2_delete_entry_el(handle_t *handle, | |
1407 | struct inode *dir, | |
1408 | struct ocfs2_dir_entry *de_del, | |
1409 | struct buffer_head *bh) | |
1410 | { | |
1411 | return __ocfs2_delete_entry(handle, dir, de_del, bh, bh->b_data, | |
1412 | bh->b_size); | |
1413 | } | |
1414 | ||
1415 | /* | |
9b7895ef MF |
1416 | * Delete a directory entry. Hide the details of directory |
1417 | * implementation from the caller. | |
5b6a3a2b MF |
1418 | */ |
1419 | int ocfs2_delete_entry(handle_t *handle, | |
1420 | struct inode *dir, | |
4a12ca3a | 1421 | struct ocfs2_dir_lookup_result *res) |
5b6a3a2b | 1422 | { |
9b7895ef MF |
1423 | if (ocfs2_dir_indexed(dir)) |
1424 | return ocfs2_delete_entry_dx(handle, dir, res); | |
1425 | ||
5b6a3a2b | 1426 | if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) |
4a12ca3a MF |
1427 | return ocfs2_delete_entry_id(handle, dir, res->dl_entry, |
1428 | res->dl_leaf_bh); | |
5b6a3a2b | 1429 | |
4a12ca3a MF |
1430 | return ocfs2_delete_entry_el(handle, dir, res->dl_entry, |
1431 | res->dl_leaf_bh); | |
5b6a3a2b MF |
1432 | } |
1433 | ||
8553cf4f MF |
1434 | /* |
1435 | * Check whether 'de' has enough room to hold an entry of | |
1436 | * 'new_rec_len' bytes. | |
1437 | */ | |
1438 | static inline int ocfs2_dirent_would_fit(struct ocfs2_dir_entry *de, | |
1439 | unsigned int new_rec_len) | |
1440 | { | |
1441 | unsigned int de_really_used; | |
1442 | ||
1443 | /* Check whether this is an empty record with enough space */ | |
1444 | if (le64_to_cpu(de->inode) == 0 && | |
1445 | le16_to_cpu(de->rec_len) >= new_rec_len) | |
1446 | return 1; | |
1447 | ||
1448 | /* | |
1449 | * Record might have free space at the end which we can | |
1450 | * use. | |
1451 | */ | |
1452 | de_really_used = OCFS2_DIR_REC_LEN(de->name_len); | |
1453 | if (le16_to_cpu(de->rec_len) >= (de_really_used + new_rec_len)) | |
1454 | return 1; | |
1455 | ||
1456 | return 0; | |
1457 | } | |
1458 | ||
9b7895ef MF |
1459 | static void ocfs2_dx_dir_leaf_insert_tail(struct ocfs2_dx_leaf *dx_leaf, |
1460 | struct ocfs2_dx_entry *dx_new_entry) | |
1461 | { | |
1462 | int i; | |
1463 | ||
1464 | i = le16_to_cpu(dx_leaf->dl_list.de_num_used); | |
1465 | dx_leaf->dl_list.de_entries[i] = *dx_new_entry; | |
1466 | ||
1467 | le16_add_cpu(&dx_leaf->dl_list.de_num_used, 1); | |
1468 | } | |
1469 | ||
4ed8a6bb MF |
1470 | static void ocfs2_dx_entry_list_insert(struct ocfs2_dx_entry_list *entry_list, |
1471 | struct ocfs2_dx_hinfo *hinfo, | |
1472 | u64 dirent_blk) | |
1473 | { | |
1474 | int i; | |
1475 | struct ocfs2_dx_entry *dx_entry; | |
1476 | ||
1477 | i = le16_to_cpu(entry_list->de_num_used); | |
1478 | dx_entry = &entry_list->de_entries[i]; | |
1479 | ||
1480 | memset(dx_entry, 0, sizeof(*dx_entry)); | |
1481 | dx_entry->dx_major_hash = cpu_to_le32(hinfo->major_hash); | |
1482 | dx_entry->dx_minor_hash = cpu_to_le32(hinfo->minor_hash); | |
1483 | dx_entry->dx_dirent_blk = cpu_to_le64(dirent_blk); | |
1484 | ||
1485 | le16_add_cpu(&entry_list->de_num_used, 1); | |
1486 | } | |
1487 | ||
9b7895ef MF |
1488 | static int __ocfs2_dx_dir_leaf_insert(struct inode *dir, handle_t *handle, |
1489 | struct ocfs2_dx_hinfo *hinfo, | |
1490 | u64 dirent_blk, | |
1491 | struct buffer_head *dx_leaf_bh) | |
1492 | { | |
4ed8a6bb | 1493 | int ret; |
9b7895ef MF |
1494 | struct ocfs2_dx_leaf *dx_leaf; |
1495 | ||
1496 | ret = ocfs2_journal_access_dl(handle, dir, dx_leaf_bh, | |
1497 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1498 | if (ret) { | |
1499 | mlog_errno(ret); | |
1500 | goto out; | |
1501 | } | |
1502 | ||
1503 | dx_leaf = (struct ocfs2_dx_leaf *)dx_leaf_bh->b_data; | |
4ed8a6bb MF |
1504 | ocfs2_dx_entry_list_insert(&dx_leaf->dl_list, hinfo, dirent_blk); |
1505 | ocfs2_journal_dirty(handle, dx_leaf_bh); | |
9b7895ef | 1506 | |
4ed8a6bb MF |
1507 | out: |
1508 | return ret; | |
1509 | } | |
9b7895ef | 1510 | |
4ed8a6bb MF |
1511 | static int ocfs2_dx_inline_root_insert(struct inode *dir, handle_t *handle, |
1512 | struct ocfs2_dx_hinfo *hinfo, | |
1513 | u64 dirent_blk, | |
1514 | struct buffer_head *dx_root_bh) | |
1515 | { | |
1516 | int ret; | |
1517 | struct ocfs2_dx_root_block *dx_root; | |
9b7895ef | 1518 | |
4ed8a6bb MF |
1519 | ret = ocfs2_journal_access_dr(handle, dir, dx_root_bh, |
1520 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1521 | if (ret) { | |
1522 | mlog_errno(ret); | |
1523 | goto out; | |
1524 | } | |
1525 | ||
1526 | dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data; | |
1527 | ocfs2_dx_entry_list_insert(&dx_root->dr_entries, hinfo, dirent_blk); | |
1528 | ocfs2_journal_dirty(handle, dx_root_bh); | |
9b7895ef MF |
1529 | |
1530 | out: | |
1531 | return ret; | |
1532 | } | |
1533 | ||
4ed8a6bb MF |
1534 | static int ocfs2_dx_dir_insert(struct inode *dir, handle_t *handle, |
1535 | struct ocfs2_dir_lookup_result *lookup) | |
9b7895ef | 1536 | { |
4ed8a6bb MF |
1537 | struct ocfs2_dx_root_block *dx_root; |
1538 | ||
1539 | dx_root = (struct ocfs2_dx_root_block *)lookup->dl_dx_root_bh->b_data; | |
1540 | if (ocfs2_dx_root_inline(dx_root)) | |
1541 | return ocfs2_dx_inline_root_insert(dir, handle, | |
1542 | &lookup->dl_hinfo, | |
1543 | lookup->dl_leaf_bh->b_blocknr, | |
1544 | lookup->dl_dx_root_bh); | |
1545 | ||
9b7895ef MF |
1546 | return __ocfs2_dx_dir_leaf_insert(dir, handle, &lookup->dl_hinfo, |
1547 | lookup->dl_leaf_bh->b_blocknr, | |
1548 | lookup->dl_dx_leaf_bh); | |
1549 | } | |
1550 | ||
e7c17e43 MF |
1551 | static void ocfs2_remove_block_from_free_list(struct inode *dir, |
1552 | handle_t *handle, | |
1553 | struct ocfs2_dir_lookup_result *lookup) | |
1554 | { | |
1555 | struct ocfs2_dir_block_trailer *trailer, *prev; | |
1556 | struct ocfs2_dx_root_block *dx_root; | |
1557 | struct buffer_head *bh; | |
1558 | ||
1559 | trailer = ocfs2_trailer_from_bh(lookup->dl_leaf_bh, dir->i_sb); | |
1560 | ||
1561 | if (ocfs2_free_list_at_root(lookup)) { | |
1562 | bh = lookup->dl_dx_root_bh; | |
1563 | dx_root = (struct ocfs2_dx_root_block *)bh->b_data; | |
1564 | dx_root->dr_free_blk = trailer->db_free_next; | |
1565 | } else { | |
1566 | bh = lookup->dl_prev_leaf_bh; | |
1567 | prev = ocfs2_trailer_from_bh(bh, dir->i_sb); | |
1568 | prev->db_free_next = trailer->db_free_next; | |
1569 | } | |
1570 | ||
1571 | trailer->db_free_rec_len = cpu_to_le16(0); | |
1572 | trailer->db_free_next = cpu_to_le64(0); | |
1573 | ||
1574 | ocfs2_journal_dirty(handle, bh); | |
1575 | ocfs2_journal_dirty(handle, lookup->dl_leaf_bh); | |
1576 | } | |
1577 | ||
1578 | /* | |
1579 | * This expects that a journal write has been reserved on | |
1580 | * lookup->dl_prev_leaf_bh or lookup->dl_dx_root_bh | |
1581 | */ | |
1582 | static void ocfs2_recalc_free_list(struct inode *dir, handle_t *handle, | |
1583 | struct ocfs2_dir_lookup_result *lookup) | |
1584 | { | |
1585 | int max_rec_len; | |
1586 | struct ocfs2_dir_block_trailer *trailer; | |
1587 | ||
1588 | /* Walk dl_leaf_bh to figure out what the new free rec_len is. */ | |
1589 | max_rec_len = ocfs2_find_max_rec_len(dir->i_sb, lookup->dl_leaf_bh); | |
1590 | if (max_rec_len) { | |
1591 | /* | |
1592 | * There's still room in this block, so no need to remove it | |
1593 | * from the free list. In this case, we just want to update | |
1594 | * the rec len accounting. | |
1595 | */ | |
1596 | trailer = ocfs2_trailer_from_bh(lookup->dl_leaf_bh, dir->i_sb); | |
1597 | trailer->db_free_rec_len = cpu_to_le16(max_rec_len); | |
1598 | ocfs2_journal_dirty(handle, lookup->dl_leaf_bh); | |
1599 | } else { | |
1600 | ocfs2_remove_block_from_free_list(dir, handle, lookup); | |
1601 | } | |
1602 | } | |
1603 | ||
316f4b9f MF |
1604 | /* we don't always have a dentry for what we want to add, so people |
1605 | * like orphan dir can call this instead. | |
1606 | * | |
4a12ca3a MF |
1607 | * The lookup context must have been filled from |
1608 | * ocfs2_prepare_dir_for_insert. | |
316f4b9f MF |
1609 | */ |
1610 | int __ocfs2_add_entry(handle_t *handle, | |
1611 | struct inode *dir, | |
1612 | const char *name, int namelen, | |
1613 | struct inode *inode, u64 blkno, | |
1614 | struct buffer_head *parent_fe_bh, | |
4a12ca3a | 1615 | struct ocfs2_dir_lookup_result *lookup) |
316f4b9f MF |
1616 | { |
1617 | unsigned long offset; | |
1618 | unsigned short rec_len; | |
1619 | struct ocfs2_dir_entry *de, *de1; | |
5b6a3a2b MF |
1620 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_fe_bh->b_data; |
1621 | struct super_block *sb = dir->i_sb; | |
316f4b9f | 1622 | int retval, status; |
5b6a3a2b | 1623 | unsigned int size = sb->s_blocksize; |
4a12ca3a | 1624 | struct buffer_head *insert_bh = lookup->dl_leaf_bh; |
5b6a3a2b | 1625 | char *data_start = insert_bh->b_data; |
316f4b9f MF |
1626 | |
1627 | mlog_entry_void(); | |
1628 | ||
316f4b9f MF |
1629 | if (!namelen) |
1630 | return -EINVAL; | |
1631 | ||
e7c17e43 MF |
1632 | if (ocfs2_dir_indexed(dir)) { |
1633 | struct buffer_head *bh; | |
1634 | ||
1635 | /* | |
1636 | * An indexed dir may require that we update the free space | |
1637 | * list. Reserve a write to the previous node in the list so | |
1638 | * that we don't fail later. | |
1639 | * | |
1640 | * XXX: This can be either a dx_root_block, or an unindexed | |
1641 | * directory tree leaf block. | |
1642 | */ | |
1643 | if (ocfs2_free_list_at_root(lookup)) { | |
1644 | bh = lookup->dl_dx_root_bh; | |
1645 | retval = ocfs2_journal_access_dr(handle, dir, bh, | |
1646 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1647 | } else { | |
1648 | bh = lookup->dl_prev_leaf_bh; | |
1649 | retval = ocfs2_journal_access_db(handle, dir, bh, | |
1650 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1651 | } | |
1652 | if (retval) { | |
1653 | mlog_errno(retval); | |
1654 | return retval; | |
1655 | } | |
1656 | } else if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { | |
5b6a3a2b MF |
1657 | data_start = di->id2.i_data.id_data; |
1658 | size = i_size_read(dir); | |
1659 | ||
1660 | BUG_ON(insert_bh != parent_fe_bh); | |
1661 | } | |
1662 | ||
316f4b9f MF |
1663 | rec_len = OCFS2_DIR_REC_LEN(namelen); |
1664 | offset = 0; | |
5b6a3a2b | 1665 | de = (struct ocfs2_dir_entry *) data_start; |
316f4b9f | 1666 | while (1) { |
5b6a3a2b MF |
1667 | BUG_ON((char *)de >= (size + data_start)); |
1668 | ||
316f4b9f MF |
1669 | /* These checks should've already been passed by the |
1670 | * prepare function, but I guess we can leave them | |
1671 | * here anyway. */ | |
1672 | if (!ocfs2_check_dir_entry(dir, de, insert_bh, offset)) { | |
1673 | retval = -ENOENT; | |
1674 | goto bail; | |
1675 | } | |
1676 | if (ocfs2_match(namelen, name, de)) { | |
1677 | retval = -EEXIST; | |
1678 | goto bail; | |
1679 | } | |
8553cf4f | 1680 | |
87d35a74 MF |
1681 | /* We're guaranteed that we should have space, so we |
1682 | * can't possibly have hit the trailer...right? */ | |
1683 | mlog_bug_on_msg(ocfs2_skip_dir_trailer(dir, de, offset, size), | |
1684 | "Hit dir trailer trying to insert %.*s " | |
1685 | "(namelen %d) into directory %llu. " | |
1686 | "offset is %lu, trailer offset is %d\n", | |
1687 | namelen, name, namelen, | |
1688 | (unsigned long long)parent_fe_bh->b_blocknr, | |
1689 | offset, ocfs2_dir_trailer_blk_off(dir->i_sb)); | |
1690 | ||
8553cf4f | 1691 | if (ocfs2_dirent_would_fit(de, rec_len)) { |
316f4b9f MF |
1692 | dir->i_mtime = dir->i_ctime = CURRENT_TIME; |
1693 | retval = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh); | |
1694 | if (retval < 0) { | |
1695 | mlog_errno(retval); | |
1696 | goto bail; | |
1697 | } | |
1698 | ||
13723d00 JB |
1699 | if (insert_bh == parent_fe_bh) |
1700 | status = ocfs2_journal_access_di(handle, dir, | |
1701 | insert_bh, | |
1702 | OCFS2_JOURNAL_ACCESS_WRITE); | |
9b7895ef | 1703 | else { |
13723d00 JB |
1704 | status = ocfs2_journal_access_db(handle, dir, |
1705 | insert_bh, | |
4ed8a6bb MF |
1706 | OCFS2_JOURNAL_ACCESS_WRITE); |
1707 | ||
9b7895ef | 1708 | if (ocfs2_dir_indexed(dir)) { |
4ed8a6bb MF |
1709 | status = ocfs2_dx_dir_insert(dir, |
1710 | handle, | |
1711 | lookup); | |
9b7895ef MF |
1712 | if (status) { |
1713 | mlog_errno(status); | |
1714 | goto bail; | |
1715 | } | |
1716 | } | |
1717 | } | |
1718 | ||
316f4b9f MF |
1719 | /* By now the buffer is marked for journaling */ |
1720 | offset += le16_to_cpu(de->rec_len); | |
1721 | if (le64_to_cpu(de->inode)) { | |
1722 | de1 = (struct ocfs2_dir_entry *)((char *) de + | |
1723 | OCFS2_DIR_REC_LEN(de->name_len)); | |
1724 | de1->rec_len = | |
1725 | cpu_to_le16(le16_to_cpu(de->rec_len) - | |
1726 | OCFS2_DIR_REC_LEN(de->name_len)); | |
1727 | de->rec_len = cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len)); | |
1728 | de = de1; | |
1729 | } | |
1730 | de->file_type = OCFS2_FT_UNKNOWN; | |
1731 | if (blkno) { | |
1732 | de->inode = cpu_to_le64(blkno); | |
1733 | ocfs2_set_de_type(de, inode->i_mode); | |
1734 | } else | |
1735 | de->inode = 0; | |
1736 | de->name_len = namelen; | |
1737 | memcpy(de->name, name, namelen); | |
1738 | ||
e7c17e43 MF |
1739 | if (ocfs2_dir_indexed(dir)) |
1740 | ocfs2_recalc_free_list(dir, handle, lookup); | |
1741 | ||
316f4b9f MF |
1742 | dir->i_version++; |
1743 | status = ocfs2_journal_dirty(handle, insert_bh); | |
1744 | retval = 0; | |
1745 | goto bail; | |
1746 | } | |
87d35a74 | 1747 | |
316f4b9f MF |
1748 | offset += le16_to_cpu(de->rec_len); |
1749 | de = (struct ocfs2_dir_entry *) ((char *) de + le16_to_cpu(de->rec_len)); | |
1750 | } | |
1751 | ||
1752 | /* when you think about it, the assert above should prevent us | |
1753 | * from ever getting here. */ | |
1754 | retval = -ENOSPC; | |
1755 | bail: | |
1756 | ||
1757 | mlog_exit(retval); | |
1758 | return retval; | |
1759 | } | |
1760 | ||
23193e51 | 1761 | static int ocfs2_dir_foreach_blk_id(struct inode *inode, |
2b47c361 | 1762 | u64 *f_version, |
23193e51 | 1763 | loff_t *f_pos, void *priv, |
e7b34019 | 1764 | filldir_t filldir, int *filldir_err) |
23193e51 MF |
1765 | { |
1766 | int ret, i, filldir_ret; | |
1767 | unsigned long offset = *f_pos; | |
1768 | struct buffer_head *di_bh = NULL; | |
1769 | struct ocfs2_dinode *di; | |
1770 | struct ocfs2_inline_data *data; | |
1771 | struct ocfs2_dir_entry *de; | |
1772 | ||
b657c95c | 1773 | ret = ocfs2_read_inode_block(inode, &di_bh); |
23193e51 MF |
1774 | if (ret) { |
1775 | mlog(ML_ERROR, "Unable to read inode block for dir %llu\n", | |
1776 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
1777 | goto out; | |
1778 | } | |
1779 | ||
1780 | di = (struct ocfs2_dinode *)di_bh->b_data; | |
1781 | data = &di->id2.i_data; | |
1782 | ||
1783 | while (*f_pos < i_size_read(inode)) { | |
1784 | revalidate: | |
1785 | /* If the dir block has changed since the last call to | |
1786 | * readdir(2), then we might be pointing to an invalid | |
1787 | * dirent right now. Scan from the start of the block | |
1788 | * to make sure. */ | |
1789 | if (*f_version != inode->i_version) { | |
1790 | for (i = 0; i < i_size_read(inode) && i < offset; ) { | |
1791 | de = (struct ocfs2_dir_entry *) | |
1792 | (data->id_data + i); | |
1793 | /* It's too expensive to do a full | |
1794 | * dirent test each time round this | |
1795 | * loop, but we do have to test at | |
1796 | * least that it is non-zero. A | |
1797 | * failure will be detected in the | |
1798 | * dirent test below. */ | |
1799 | if (le16_to_cpu(de->rec_len) < | |
1800 | OCFS2_DIR_REC_LEN(1)) | |
1801 | break; | |
1802 | i += le16_to_cpu(de->rec_len); | |
1803 | } | |
1804 | *f_pos = offset = i; | |
1805 | *f_version = inode->i_version; | |
1806 | } | |
1807 | ||
1808 | de = (struct ocfs2_dir_entry *) (data->id_data + *f_pos); | |
1809 | if (!ocfs2_check_dir_entry(inode, de, di_bh, *f_pos)) { | |
1810 | /* On error, skip the f_pos to the end. */ | |
1811 | *f_pos = i_size_read(inode); | |
1812 | goto out; | |
1813 | } | |
1814 | offset += le16_to_cpu(de->rec_len); | |
1815 | if (le64_to_cpu(de->inode)) { | |
1816 | /* We might block in the next section | |
1817 | * if the data destination is | |
1818 | * currently swapped out. So, use a | |
1819 | * version stamp to detect whether or | |
1820 | * not the directory has been modified | |
1821 | * during the copy operation. | |
1822 | */ | |
2b47c361 | 1823 | u64 version = *f_version; |
23193e51 MF |
1824 | unsigned char d_type = DT_UNKNOWN; |
1825 | ||
1826 | if (de->file_type < OCFS2_FT_MAX) | |
1827 | d_type = ocfs2_filetype_table[de->file_type]; | |
1828 | ||
1829 | filldir_ret = filldir(priv, de->name, | |
1830 | de->name_len, | |
1831 | *f_pos, | |
1832 | le64_to_cpu(de->inode), | |
1833 | d_type); | |
e7b34019 MF |
1834 | if (filldir_ret) { |
1835 | if (filldir_err) | |
1836 | *filldir_err = filldir_ret; | |
23193e51 | 1837 | break; |
e7b34019 | 1838 | } |
23193e51 MF |
1839 | if (version != *f_version) |
1840 | goto revalidate; | |
1841 | } | |
1842 | *f_pos += le16_to_cpu(de->rec_len); | |
1843 | } | |
1844 | ||
1845 | out: | |
1846 | brelse(di_bh); | |
1847 | ||
1848 | return 0; | |
1849 | } | |
1850 | ||
9b7895ef MF |
1851 | /* |
1852 | * NOTE: This function can be called against unindexed directories, | |
1853 | * and indexed ones. | |
1854 | */ | |
23193e51 | 1855 | static int ocfs2_dir_foreach_blk_el(struct inode *inode, |
2b47c361 | 1856 | u64 *f_version, |
23193e51 | 1857 | loff_t *f_pos, void *priv, |
e7b34019 | 1858 | filldir_t filldir, int *filldir_err) |
ccd979bd MF |
1859 | { |
1860 | int error = 0; | |
aa958874 MF |
1861 | unsigned long offset, blk, last_ra_blk = 0; |
1862 | int i, stored; | |
ccd979bd MF |
1863 | struct buffer_head * bh, * tmp; |
1864 | struct ocfs2_dir_entry * de; | |
ccd979bd | 1865 | struct super_block * sb = inode->i_sb; |
aa958874 | 1866 | unsigned int ra_sectors = 16; |
ccd979bd MF |
1867 | |
1868 | stored = 0; | |
1869 | bh = NULL; | |
1870 | ||
b8bc5f4f | 1871 | offset = (*f_pos) & (sb->s_blocksize - 1); |
ccd979bd | 1872 | |
b8bc5f4f MF |
1873 | while (!error && !stored && *f_pos < i_size_read(inode)) { |
1874 | blk = (*f_pos) >> sb->s_blocksize_bits; | |
a22305cc JB |
1875 | if (ocfs2_read_dir_block(inode, blk, &bh, 0)) { |
1876 | /* Skip the corrupt dirblock and keep trying */ | |
b8bc5f4f | 1877 | *f_pos += sb->s_blocksize - offset; |
ccd979bd MF |
1878 | continue; |
1879 | } | |
1880 | ||
aa958874 MF |
1881 | /* The idea here is to begin with 8k read-ahead and to stay |
1882 | * 4k ahead of our current position. | |
1883 | * | |
1884 | * TODO: Use the pagecache for this. We just need to | |
1885 | * make sure it's cluster-safe... */ | |
1886 | if (!last_ra_blk | |
1887 | || (((last_ra_blk - blk) << 9) <= (ra_sectors / 2))) { | |
1888 | for (i = ra_sectors >> (sb->s_blocksize_bits - 9); | |
ccd979bd | 1889 | i > 0; i--) { |
a22305cc JB |
1890 | tmp = NULL; |
1891 | if (!ocfs2_read_dir_block(inode, ++blk, &tmp, | |
1892 | OCFS2_BH_READAHEAD)) | |
1893 | brelse(tmp); | |
ccd979bd | 1894 | } |
aa958874 MF |
1895 | last_ra_blk = blk; |
1896 | ra_sectors = 8; | |
ccd979bd MF |
1897 | } |
1898 | ||
1899 | revalidate: | |
1900 | /* If the dir block has changed since the last call to | |
1901 | * readdir(2), then we might be pointing to an invalid | |
1902 | * dirent right now. Scan from the start of the block | |
1903 | * to make sure. */ | |
b8bc5f4f | 1904 | if (*f_version != inode->i_version) { |
ccd979bd MF |
1905 | for (i = 0; i < sb->s_blocksize && i < offset; ) { |
1906 | de = (struct ocfs2_dir_entry *) (bh->b_data + i); | |
1907 | /* It's too expensive to do a full | |
1908 | * dirent test each time round this | |
1909 | * loop, but we do have to test at | |
1910 | * least that it is non-zero. A | |
1911 | * failure will be detected in the | |
1912 | * dirent test below. */ | |
1913 | if (le16_to_cpu(de->rec_len) < | |
1914 | OCFS2_DIR_REC_LEN(1)) | |
1915 | break; | |
1916 | i += le16_to_cpu(de->rec_len); | |
1917 | } | |
1918 | offset = i; | |
b8bc5f4f | 1919 | *f_pos = ((*f_pos) & ~(sb->s_blocksize - 1)) |
ccd979bd | 1920 | | offset; |
b8bc5f4f | 1921 | *f_version = inode->i_version; |
ccd979bd MF |
1922 | } |
1923 | ||
b8bc5f4f | 1924 | while (!error && *f_pos < i_size_read(inode) |
ccd979bd MF |
1925 | && offset < sb->s_blocksize) { |
1926 | de = (struct ocfs2_dir_entry *) (bh->b_data + offset); | |
1927 | if (!ocfs2_check_dir_entry(inode, de, bh, offset)) { | |
1928 | /* On error, skip the f_pos to the | |
1929 | next block. */ | |
b8bc5f4f | 1930 | *f_pos = ((*f_pos) | (sb->s_blocksize - 1)) + 1; |
ccd979bd | 1931 | brelse(bh); |
b8bc5f4f | 1932 | goto out; |
ccd979bd MF |
1933 | } |
1934 | offset += le16_to_cpu(de->rec_len); | |
1935 | if (le64_to_cpu(de->inode)) { | |
1936 | /* We might block in the next section | |
1937 | * if the data destination is | |
1938 | * currently swapped out. So, use a | |
1939 | * version stamp to detect whether or | |
1940 | * not the directory has been modified | |
1941 | * during the copy operation. | |
1942 | */ | |
b8bc5f4f | 1943 | unsigned long version = *f_version; |
ccd979bd MF |
1944 | unsigned char d_type = DT_UNKNOWN; |
1945 | ||
1946 | if (de->file_type < OCFS2_FT_MAX) | |
1947 | d_type = ocfs2_filetype_table[de->file_type]; | |
b8bc5f4f | 1948 | error = filldir(priv, de->name, |
ccd979bd | 1949 | de->name_len, |
b8bc5f4f | 1950 | *f_pos, |
7e853679 | 1951 | le64_to_cpu(de->inode), |
ccd979bd | 1952 | d_type); |
e7b34019 MF |
1953 | if (error) { |
1954 | if (filldir_err) | |
1955 | *filldir_err = error; | |
ccd979bd | 1956 | break; |
e7b34019 | 1957 | } |
b8bc5f4f | 1958 | if (version != *f_version) |
ccd979bd MF |
1959 | goto revalidate; |
1960 | stored ++; | |
1961 | } | |
b8bc5f4f | 1962 | *f_pos += le16_to_cpu(de->rec_len); |
ccd979bd MF |
1963 | } |
1964 | offset = 0; | |
1965 | brelse(bh); | |
a22305cc | 1966 | bh = NULL; |
ccd979bd MF |
1967 | } |
1968 | ||
1969 | stored = 0; | |
b8bc5f4f MF |
1970 | out: |
1971 | return stored; | |
1972 | } | |
1973 | ||
2b47c361 | 1974 | static int ocfs2_dir_foreach_blk(struct inode *inode, u64 *f_version, |
e7b34019 MF |
1975 | loff_t *f_pos, void *priv, filldir_t filldir, |
1976 | int *filldir_err) | |
23193e51 MF |
1977 | { |
1978 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) | |
1979 | return ocfs2_dir_foreach_blk_id(inode, f_version, f_pos, priv, | |
e7b34019 | 1980 | filldir, filldir_err); |
23193e51 | 1981 | |
e7b34019 MF |
1982 | return ocfs2_dir_foreach_blk_el(inode, f_version, f_pos, priv, filldir, |
1983 | filldir_err); | |
23193e51 MF |
1984 | } |
1985 | ||
5eae5b96 MF |
1986 | /* |
1987 | * This is intended to be called from inside other kernel functions, | |
1988 | * so we fake some arguments. | |
1989 | */ | |
1990 | int ocfs2_dir_foreach(struct inode *inode, loff_t *f_pos, void *priv, | |
1991 | filldir_t filldir) | |
1992 | { | |
e7b34019 | 1993 | int ret = 0, filldir_err = 0; |
2b47c361 | 1994 | u64 version = inode->i_version; |
5eae5b96 MF |
1995 | |
1996 | while (*f_pos < i_size_read(inode)) { | |
1997 | ret = ocfs2_dir_foreach_blk(inode, &version, f_pos, priv, | |
e7b34019 MF |
1998 | filldir, &filldir_err); |
1999 | if (ret || filldir_err) | |
5eae5b96 MF |
2000 | break; |
2001 | } | |
2002 | ||
e7b34019 MF |
2003 | if (ret > 0) |
2004 | ret = -EIO; | |
2005 | ||
5eae5b96 MF |
2006 | return 0; |
2007 | } | |
2008 | ||
b8bc5f4f MF |
2009 | /* |
2010 | * ocfs2_readdir() | |
2011 | * | |
2012 | */ | |
2013 | int ocfs2_readdir(struct file * filp, void * dirent, filldir_t filldir) | |
2014 | { | |
2015 | int error = 0; | |
2016 | struct inode *inode = filp->f_path.dentry->d_inode; | |
2017 | int lock_level = 0; | |
2018 | ||
2019 | mlog_entry("dirino=%llu\n", | |
2020 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
2021 | ||
e63aecb6 | 2022 | error = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level); |
b8bc5f4f MF |
2023 | if (lock_level && error >= 0) { |
2024 | /* We release EX lock which used to update atime | |
2025 | * and get PR lock again to reduce contention | |
2026 | * on commonly accessed directories. */ | |
e63aecb6 | 2027 | ocfs2_inode_unlock(inode, 1); |
b8bc5f4f | 2028 | lock_level = 0; |
e63aecb6 | 2029 | error = ocfs2_inode_lock(inode, NULL, 0); |
b8bc5f4f MF |
2030 | } |
2031 | if (error < 0) { | |
2032 | if (error != -ENOENT) | |
2033 | mlog_errno(error); | |
2034 | /* we haven't got any yet, so propagate the error. */ | |
2035 | goto bail_nolock; | |
2036 | } | |
2037 | ||
2038 | error = ocfs2_dir_foreach_blk(inode, &filp->f_version, &filp->f_pos, | |
e7b34019 | 2039 | dirent, filldir, NULL); |
b8bc5f4f | 2040 | |
e63aecb6 | 2041 | ocfs2_inode_unlock(inode, lock_level); |
ccd979bd | 2042 | |
aa958874 | 2043 | bail_nolock: |
b8bc5f4f | 2044 | mlog_exit(error); |
ccd979bd | 2045 | |
b8bc5f4f | 2046 | return error; |
ccd979bd MF |
2047 | } |
2048 | ||
2049 | /* | |
1b1dcc1b | 2050 | * NOTE: this should always be called with parent dir i_mutex taken. |
ccd979bd MF |
2051 | */ |
2052 | int ocfs2_find_files_on_disk(const char *name, | |
2053 | int namelen, | |
2054 | u64 *blkno, | |
2055 | struct inode *inode, | |
4a12ca3a | 2056 | struct ocfs2_dir_lookup_result *lookup) |
ccd979bd MF |
2057 | { |
2058 | int status = -ENOENT; | |
ccd979bd | 2059 | |
4a12ca3a MF |
2060 | mlog(0, "name=%.*s, blkno=%p, inode=%llu\n", namelen, name, blkno, |
2061 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
ccd979bd | 2062 | |
4a12ca3a MF |
2063 | status = ocfs2_find_entry(name, namelen, inode, lookup); |
2064 | if (status) | |
ccd979bd | 2065 | goto leave; |
ccd979bd | 2066 | |
4a12ca3a | 2067 | *blkno = le64_to_cpu(lookup->dl_entry->inode); |
ccd979bd MF |
2068 | |
2069 | status = 0; | |
2070 | leave: | |
ccd979bd | 2071 | |
ccd979bd MF |
2072 | return status; |
2073 | } | |
2074 | ||
be94d117 MF |
2075 | /* |
2076 | * Convenience function for callers which just want the block number | |
2077 | * mapped to a name and don't require the full dirent info, etc. | |
2078 | */ | |
2079 | int ocfs2_lookup_ino_from_name(struct inode *dir, const char *name, | |
2080 | int namelen, u64 *blkno) | |
2081 | { | |
2082 | int ret; | |
4a12ca3a | 2083 | struct ocfs2_dir_lookup_result lookup = { NULL, }; |
be94d117 | 2084 | |
4a12ca3a MF |
2085 | ret = ocfs2_find_files_on_disk(name, namelen, blkno, dir, &lookup); |
2086 | ocfs2_free_dir_lookup_result(&lookup); | |
be94d117 MF |
2087 | |
2088 | return ret; | |
2089 | } | |
2090 | ||
ccd979bd MF |
2091 | /* Check for a name within a directory. |
2092 | * | |
2093 | * Return 0 if the name does not exist | |
2094 | * Return -EEXIST if the directory contains the name | |
2095 | * | |
1b1dcc1b | 2096 | * Callers should have i_mutex + a cluster lock on dir |
ccd979bd MF |
2097 | */ |
2098 | int ocfs2_check_dir_for_entry(struct inode *dir, | |
2099 | const char *name, | |
2100 | int namelen) | |
2101 | { | |
2102 | int ret; | |
4a12ca3a | 2103 | struct ocfs2_dir_lookup_result lookup = { NULL, }; |
ccd979bd | 2104 | |
b0697053 MF |
2105 | mlog_entry("dir %llu, name '%.*s'\n", |
2106 | (unsigned long long)OCFS2_I(dir)->ip_blkno, namelen, name); | |
ccd979bd MF |
2107 | |
2108 | ret = -EEXIST; | |
4a12ca3a | 2109 | if (ocfs2_find_entry(name, namelen, dir, &lookup) == 0) |
ccd979bd MF |
2110 | goto bail; |
2111 | ||
2112 | ret = 0; | |
2113 | bail: | |
4a12ca3a | 2114 | ocfs2_free_dir_lookup_result(&lookup); |
ccd979bd MF |
2115 | |
2116 | mlog_exit(ret); | |
2117 | return ret; | |
2118 | } | |
2119 | ||
0bfbbf62 MF |
2120 | struct ocfs2_empty_dir_priv { |
2121 | unsigned seen_dot; | |
2122 | unsigned seen_dot_dot; | |
2123 | unsigned seen_other; | |
2124 | }; | |
2125 | static int ocfs2_empty_dir_filldir(void *priv, const char *name, int name_len, | |
2126 | loff_t pos, u64 ino, unsigned type) | |
2127 | { | |
2128 | struct ocfs2_empty_dir_priv *p = priv; | |
2129 | ||
2130 | /* | |
2131 | * Check the positions of "." and ".." records to be sure | |
2132 | * they're in the correct place. | |
2133 | */ | |
2134 | if (name_len == 1 && !strncmp(".", name, 1) && pos == 0) { | |
2135 | p->seen_dot = 1; | |
2136 | return 0; | |
2137 | } | |
2138 | ||
2139 | if (name_len == 2 && !strncmp("..", name, 2) && | |
2140 | pos == OCFS2_DIR_REC_LEN(1)) { | |
2141 | p->seen_dot_dot = 1; | |
2142 | return 0; | |
2143 | } | |
2144 | ||
2145 | p->seen_other = 1; | |
2146 | return 1; | |
2147 | } | |
ccd979bd MF |
2148 | /* |
2149 | * routine to check that the specified directory is empty (for rmdir) | |
0bfbbf62 MF |
2150 | * |
2151 | * Returns 1 if dir is empty, zero otherwise. | |
9b7895ef MF |
2152 | * |
2153 | * XXX: This is a performance problem | |
ccd979bd MF |
2154 | */ |
2155 | int ocfs2_empty_dir(struct inode *inode) | |
2156 | { | |
0bfbbf62 MF |
2157 | int ret; |
2158 | loff_t start = 0; | |
2159 | struct ocfs2_empty_dir_priv priv; | |
ccd979bd | 2160 | |
0bfbbf62 | 2161 | memset(&priv, 0, sizeof(priv)); |
ccd979bd | 2162 | |
0bfbbf62 MF |
2163 | ret = ocfs2_dir_foreach(inode, &start, &priv, ocfs2_empty_dir_filldir); |
2164 | if (ret) | |
2165 | mlog_errno(ret); | |
2166 | ||
2167 | if (!priv.seen_dot || !priv.seen_dot_dot) { | |
2168 | mlog(ML_ERROR, "bad directory (dir #%llu) - no `.' or `..'\n", | |
b0697053 | 2169 | (unsigned long long)OCFS2_I(inode)->ip_blkno); |
0bfbbf62 MF |
2170 | /* |
2171 | * XXX: Is it really safe to allow an unlink to continue? | |
2172 | */ | |
ccd979bd MF |
2173 | return 1; |
2174 | } | |
0bfbbf62 MF |
2175 | |
2176 | return !priv.seen_other; | |
ccd979bd MF |
2177 | } |
2178 | ||
87d35a74 MF |
2179 | /* |
2180 | * Fills "." and ".." dirents in a new directory block. Returns dirent for | |
2181 | * "..", which might be used during creation of a directory with a trailing | |
2182 | * header. It is otherwise safe to ignore the return code. | |
2183 | */ | |
2184 | static struct ocfs2_dir_entry *ocfs2_fill_initial_dirents(struct inode *inode, | |
2185 | struct inode *parent, | |
2186 | char *start, | |
2187 | unsigned int size) | |
5b6a3a2b MF |
2188 | { |
2189 | struct ocfs2_dir_entry *de = (struct ocfs2_dir_entry *)start; | |
2190 | ||
2191 | de->inode = cpu_to_le64(OCFS2_I(inode)->ip_blkno); | |
2192 | de->name_len = 1; | |
2193 | de->rec_len = | |
2194 | cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len)); | |
2195 | strcpy(de->name, "."); | |
2196 | ocfs2_set_de_type(de, S_IFDIR); | |
2197 | ||
2198 | de = (struct ocfs2_dir_entry *) ((char *)de + le16_to_cpu(de->rec_len)); | |
2199 | de->inode = cpu_to_le64(OCFS2_I(parent)->ip_blkno); | |
2200 | de->rec_len = cpu_to_le16(size - OCFS2_DIR_REC_LEN(1)); | |
2201 | de->name_len = 2; | |
2202 | strcpy(de->name, ".."); | |
2203 | ocfs2_set_de_type(de, S_IFDIR); | |
87d35a74 MF |
2204 | |
2205 | return de; | |
5b6a3a2b MF |
2206 | } |
2207 | ||
2208 | /* | |
2209 | * This works together with code in ocfs2_mknod_locked() which sets | |
2210 | * the inline-data flag and initializes the inline-data section. | |
2211 | */ | |
2212 | static int ocfs2_fill_new_dir_id(struct ocfs2_super *osb, | |
2213 | handle_t *handle, | |
2214 | struct inode *parent, | |
2215 | struct inode *inode, | |
2216 | struct buffer_head *di_bh) | |
2217 | { | |
2218 | int ret; | |
2219 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
2220 | struct ocfs2_inline_data *data = &di->id2.i_data; | |
2221 | unsigned int size = le16_to_cpu(data->id_count); | |
2222 | ||
13723d00 JB |
2223 | ret = ocfs2_journal_access_di(handle, inode, di_bh, |
2224 | OCFS2_JOURNAL_ACCESS_WRITE); | |
5b6a3a2b MF |
2225 | if (ret) { |
2226 | mlog_errno(ret); | |
2227 | goto out; | |
2228 | } | |
2229 | ||
2230 | ocfs2_fill_initial_dirents(inode, parent, data->id_data, size); | |
2231 | ||
2232 | ocfs2_journal_dirty(handle, di_bh); | |
2233 | if (ret) { | |
2234 | mlog_errno(ret); | |
2235 | goto out; | |
2236 | } | |
2237 | ||
2238 | i_size_write(inode, size); | |
2239 | inode->i_nlink = 2; | |
2240 | inode->i_blocks = ocfs2_inode_sector_count(inode); | |
2241 | ||
2242 | ret = ocfs2_mark_inode_dirty(handle, inode, di_bh); | |
2243 | if (ret < 0) | |
2244 | mlog_errno(ret); | |
2245 | ||
2246 | out: | |
2247 | return ret; | |
2248 | } | |
2249 | ||
2250 | static int ocfs2_fill_new_dir_el(struct ocfs2_super *osb, | |
2251 | handle_t *handle, | |
2252 | struct inode *parent, | |
2253 | struct inode *inode, | |
2254 | struct buffer_head *fe_bh, | |
9b7895ef MF |
2255 | struct ocfs2_alloc_context *data_ac, |
2256 | struct buffer_head **ret_new_bh) | |
316f4b9f MF |
2257 | { |
2258 | int status; | |
87d35a74 | 2259 | unsigned int size = osb->sb->s_blocksize; |
316f4b9f | 2260 | struct buffer_head *new_bh = NULL; |
87d35a74 | 2261 | struct ocfs2_dir_entry *de; |
316f4b9f MF |
2262 | |
2263 | mlog_entry_void(); | |
2264 | ||
e7c17e43 | 2265 | if (ocfs2_new_dir_wants_trailer(inode)) |
87d35a74 MF |
2266 | size = ocfs2_dir_trailer_blk_off(parent->i_sb); |
2267 | ||
316f4b9f MF |
2268 | status = ocfs2_do_extend_dir(osb->sb, handle, inode, fe_bh, |
2269 | data_ac, NULL, &new_bh); | |
2270 | if (status < 0) { | |
2271 | mlog_errno(status); | |
2272 | goto bail; | |
2273 | } | |
2274 | ||
2275 | ocfs2_set_new_buffer_uptodate(inode, new_bh); | |
2276 | ||
13723d00 JB |
2277 | status = ocfs2_journal_access_db(handle, inode, new_bh, |
2278 | OCFS2_JOURNAL_ACCESS_CREATE); | |
316f4b9f MF |
2279 | if (status < 0) { |
2280 | mlog_errno(status); | |
2281 | goto bail; | |
2282 | } | |
2283 | memset(new_bh->b_data, 0, osb->sb->s_blocksize); | |
2284 | ||
87d35a74 | 2285 | de = ocfs2_fill_initial_dirents(inode, parent, new_bh->b_data, size); |
e7c17e43 MF |
2286 | if (ocfs2_new_dir_wants_trailer(inode)) { |
2287 | int size = le16_to_cpu(de->rec_len); | |
2288 | ||
2289 | /* | |
2290 | * Figure out the size of the hole left over after | |
2291 | * insertion of '.' and '..'. The trailer wants this | |
2292 | * information. | |
2293 | */ | |
2294 | size -= OCFS2_DIR_REC_LEN(2); | |
2295 | size -= sizeof(struct ocfs2_dir_block_trailer); | |
2296 | ||
2297 | ocfs2_init_dir_trailer(inode, new_bh, size); | |
2298 | } | |
316f4b9f MF |
2299 | |
2300 | status = ocfs2_journal_dirty(handle, new_bh); | |
2301 | if (status < 0) { | |
2302 | mlog_errno(status); | |
2303 | goto bail; | |
2304 | } | |
2305 | ||
2306 | i_size_write(inode, inode->i_sb->s_blocksize); | |
2307 | inode->i_nlink = 2; | |
2308 | inode->i_blocks = ocfs2_inode_sector_count(inode); | |
2309 | status = ocfs2_mark_inode_dirty(handle, inode, fe_bh); | |
2310 | if (status < 0) { | |
2311 | mlog_errno(status); | |
2312 | goto bail; | |
2313 | } | |
2314 | ||
2315 | status = 0; | |
9b7895ef MF |
2316 | if (ret_new_bh) { |
2317 | *ret_new_bh = new_bh; | |
2318 | new_bh = NULL; | |
2319 | } | |
316f4b9f | 2320 | bail: |
a81cb88b | 2321 | brelse(new_bh); |
316f4b9f MF |
2322 | |
2323 | mlog_exit(status); | |
2324 | return status; | |
2325 | } | |
2326 | ||
9b7895ef MF |
2327 | static int ocfs2_dx_dir_attach_index(struct ocfs2_super *osb, |
2328 | handle_t *handle, struct inode *dir, | |
2329 | struct buffer_head *di_bh, | |
e7c17e43 | 2330 | struct buffer_head *dirdata_bh, |
9b7895ef | 2331 | struct ocfs2_alloc_context *meta_ac, |
4ed8a6bb | 2332 | int dx_inline, |
9b7895ef | 2333 | struct buffer_head **ret_dx_root_bh) |
5b6a3a2b | 2334 | { |
9b7895ef MF |
2335 | int ret; |
2336 | struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data; | |
2337 | u16 dr_suballoc_bit; | |
2338 | u64 dr_blkno; | |
2339 | unsigned int num_bits; | |
2340 | struct buffer_head *dx_root_bh = NULL; | |
2341 | struct ocfs2_dx_root_block *dx_root; | |
e7c17e43 MF |
2342 | struct ocfs2_dir_block_trailer *trailer = |
2343 | ocfs2_trailer_from_bh(dirdata_bh, dir->i_sb); | |
9b7895ef MF |
2344 | |
2345 | ret = ocfs2_claim_metadata(osb, handle, meta_ac, 1, &dr_suballoc_bit, | |
2346 | &num_bits, &dr_blkno); | |
2347 | if (ret) { | |
2348 | mlog_errno(ret); | |
2349 | goto out; | |
2350 | } | |
5b6a3a2b | 2351 | |
9b7895ef MF |
2352 | mlog(0, "Dir %llu, attach new index block: %llu\n", |
2353 | (unsigned long long)OCFS2_I(dir)->ip_blkno, | |
2354 | (unsigned long long)dr_blkno); | |
5b6a3a2b | 2355 | |
9b7895ef MF |
2356 | dx_root_bh = sb_getblk(osb->sb, dr_blkno); |
2357 | if (dx_root_bh == NULL) { | |
2358 | ret = -EIO; | |
2359 | goto out; | |
2360 | } | |
2361 | ocfs2_set_new_buffer_uptodate(dir, dx_root_bh); | |
87d35a74 | 2362 | |
9b7895ef MF |
2363 | ret = ocfs2_journal_access_dr(handle, dir, dx_root_bh, |
2364 | OCFS2_JOURNAL_ACCESS_CREATE); | |
2365 | if (ret < 0) { | |
2366 | mlog_errno(ret); | |
2367 | goto out; | |
2368 | } | |
87d35a74 | 2369 | |
9b7895ef MF |
2370 | dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data; |
2371 | memset(dx_root, 0, osb->sb->s_blocksize); | |
2372 | strcpy(dx_root->dr_signature, OCFS2_DX_ROOT_SIGNATURE); | |
2373 | dx_root->dr_suballoc_slot = cpu_to_le16(osb->slot_num); | |
2374 | dx_root->dr_suballoc_bit = cpu_to_le16(dr_suballoc_bit); | |
2375 | dx_root->dr_fs_generation = cpu_to_le32(osb->fs_generation); | |
2376 | dx_root->dr_blkno = cpu_to_le64(dr_blkno); | |
2377 | dx_root->dr_dir_blkno = cpu_to_le64(OCFS2_I(dir)->ip_blkno); | |
e7c17e43 MF |
2378 | if (le16_to_cpu(trailer->db_free_rec_len)) |
2379 | dx_root->dr_free_blk = cpu_to_le64(dirdata_bh->b_blocknr); | |
2380 | else | |
2381 | dx_root->dr_free_blk = cpu_to_le64(0); | |
4ed8a6bb MF |
2382 | |
2383 | if (dx_inline) { | |
2384 | dx_root->dr_flags |= OCFS2_DX_FLAG_INLINE; | |
2385 | dx_root->dr_entries.de_count = | |
2386 | cpu_to_le16(ocfs2_dx_entries_per_root(osb->sb)); | |
2387 | } else { | |
2388 | dx_root->dr_list.l_count = | |
2389 | cpu_to_le16(ocfs2_extent_recs_per_dx_root(osb->sb)); | |
2390 | } | |
9b7895ef MF |
2391 | |
2392 | ret = ocfs2_journal_dirty(handle, dx_root_bh); | |
2393 | if (ret) | |
2394 | mlog_errno(ret); | |
5b6a3a2b | 2395 | |
9b7895ef MF |
2396 | ret = ocfs2_journal_access_di(handle, dir, di_bh, |
2397 | OCFS2_JOURNAL_ACCESS_CREATE); | |
2398 | if (ret) { | |
2399 | mlog_errno(ret); | |
2400 | goto out; | |
2401 | } | |
2402 | ||
2403 | di->i_dx_root = cpu_to_le64(dr_blkno); | |
2404 | ||
2405 | OCFS2_I(dir)->ip_dyn_features |= OCFS2_INDEXED_DIR_FL; | |
2406 | di->i_dyn_features = cpu_to_le16(OCFS2_I(dir)->ip_dyn_features); | |
2407 | ||
2408 | ret = ocfs2_journal_dirty(handle, di_bh); | |
2409 | if (ret) | |
2410 | mlog_errno(ret); | |
2411 | ||
2412 | *ret_dx_root_bh = dx_root_bh; | |
2413 | dx_root_bh = NULL; | |
2414 | ||
2415 | out: | |
2416 | brelse(dx_root_bh); | |
2417 | return ret; | |
2418 | } | |
2419 | ||
2420 | static int ocfs2_dx_dir_format_cluster(struct ocfs2_super *osb, | |
2421 | handle_t *handle, struct inode *dir, | |
2422 | struct buffer_head **dx_leaves, | |
2423 | int num_dx_leaves, u64 start_blk) | |
2424 | { | |
2425 | int ret, i; | |
2426 | struct ocfs2_dx_leaf *dx_leaf; | |
2427 | struct buffer_head *bh; | |
2428 | ||
2429 | for (i = 0; i < num_dx_leaves; i++) { | |
2430 | bh = sb_getblk(osb->sb, start_blk + i); | |
2431 | if (bh == NULL) { | |
2432 | ret = -EIO; | |
2433 | goto out; | |
2434 | } | |
2435 | dx_leaves[i] = bh; | |
2436 | ||
2437 | ocfs2_set_new_buffer_uptodate(dir, bh); | |
2438 | ||
2439 | ret = ocfs2_journal_access_dl(handle, dir, bh, | |
2440 | OCFS2_JOURNAL_ACCESS_CREATE); | |
2441 | if (ret < 0) { | |
2442 | mlog_errno(ret); | |
2443 | goto out; | |
2444 | } | |
2445 | ||
2446 | dx_leaf = (struct ocfs2_dx_leaf *) bh->b_data; | |
2447 | ||
2448 | memset(dx_leaf, 0, osb->sb->s_blocksize); | |
2449 | strcpy(dx_leaf->dl_signature, OCFS2_DX_LEAF_SIGNATURE); | |
2450 | dx_leaf->dl_fs_generation = cpu_to_le32(osb->fs_generation); | |
2451 | dx_leaf->dl_blkno = cpu_to_le64(bh->b_blocknr); | |
2452 | dx_leaf->dl_list.de_count = | |
2453 | cpu_to_le16(ocfs2_dx_entries_per_leaf(osb->sb)); | |
2454 | ||
2455 | mlog(0, | |
2456 | "Dir %llu, format dx_leaf: %llu, entry count: %u\n", | |
2457 | (unsigned long long)OCFS2_I(dir)->ip_blkno, | |
2458 | (unsigned long long)bh->b_blocknr, | |
2459 | le16_to_cpu(dx_leaf->dl_list.de_count)); | |
2460 | ||
2461 | ocfs2_journal_dirty(handle, bh); | |
2462 | } | |
2463 | ||
2464 | ret = 0; | |
2465 | out: | |
2466 | return ret; | |
2467 | } | |
2468 | ||
2469 | /* | |
2470 | * Allocates and formats a new cluster for use in an indexed dir | |
2471 | * leaf. This version will not do the extent insert, so that it can be | |
2472 | * used by operations which need careful ordering. | |
2473 | */ | |
2474 | static int __ocfs2_dx_dir_new_cluster(struct inode *dir, | |
2475 | u32 cpos, handle_t *handle, | |
2476 | struct ocfs2_alloc_context *data_ac, | |
2477 | struct buffer_head **dx_leaves, | |
2478 | int num_dx_leaves, u64 *ret_phys_blkno) | |
2479 | { | |
2480 | int ret; | |
2481 | u32 phys, num; | |
2482 | u64 phys_blkno; | |
2483 | struct ocfs2_super *osb = OCFS2_SB(dir->i_sb); | |
2484 | ||
2485 | /* | |
2486 | * XXX: For create, this should claim cluster for the index | |
2487 | * *before* the unindexed insert so that we have a better | |
2488 | * chance of contiguousness as the directory grows in number | |
2489 | * of entries. | |
2490 | */ | |
2491 | ret = __ocfs2_claim_clusters(osb, handle, data_ac, 1, 1, &phys, &num); | |
2492 | if (ret) { | |
2493 | mlog_errno(ret); | |
2494 | goto out; | |
2495 | } | |
2496 | ||
2497 | /* | |
2498 | * Format the new cluster first. That way, we're inserting | |
2499 | * valid data. | |
2500 | */ | |
2501 | phys_blkno = ocfs2_clusters_to_blocks(osb->sb, phys); | |
2502 | ret = ocfs2_dx_dir_format_cluster(osb, handle, dir, dx_leaves, | |
2503 | num_dx_leaves, phys_blkno); | |
2504 | if (ret) { | |
2505 | mlog_errno(ret); | |
2506 | goto out; | |
2507 | } | |
2508 | ||
2509 | *ret_phys_blkno = phys_blkno; | |
2510 | out: | |
2511 | return ret; | |
2512 | } | |
2513 | ||
2514 | static int ocfs2_dx_dir_new_cluster(struct inode *dir, | |
2515 | struct ocfs2_extent_tree *et, | |
2516 | u32 cpos, handle_t *handle, | |
2517 | struct ocfs2_alloc_context *data_ac, | |
2518 | struct ocfs2_alloc_context *meta_ac, | |
2519 | struct buffer_head **dx_leaves, | |
2520 | int num_dx_leaves) | |
2521 | { | |
2522 | int ret; | |
2523 | u64 phys_blkno; | |
2524 | struct ocfs2_super *osb = OCFS2_SB(dir->i_sb); | |
2525 | ||
2526 | ret = __ocfs2_dx_dir_new_cluster(dir, cpos, handle, data_ac, dx_leaves, | |
2527 | num_dx_leaves, &phys_blkno); | |
2528 | if (ret) { | |
2529 | mlog_errno(ret); | |
2530 | goto out; | |
2531 | } | |
2532 | ||
2533 | ret = ocfs2_insert_extent(osb, handle, dir, et, cpos, phys_blkno, 1, 0, | |
2534 | meta_ac); | |
2535 | if (ret) | |
2536 | mlog_errno(ret); | |
2537 | out: | |
2538 | return ret; | |
2539 | } | |
2540 | ||
2541 | static struct buffer_head **ocfs2_dx_dir_kmalloc_leaves(struct super_block *sb, | |
2542 | int *ret_num_leaves) | |
2543 | { | |
2544 | int num_dx_leaves = ocfs2_clusters_to_blocks(sb, 1); | |
2545 | struct buffer_head **dx_leaves; | |
2546 | ||
2547 | dx_leaves = kcalloc(num_dx_leaves, sizeof(struct buffer_head *), | |
2548 | GFP_NOFS); | |
2549 | if (dx_leaves && ret_num_leaves) | |
2550 | *ret_num_leaves = num_dx_leaves; | |
2551 | ||
2552 | return dx_leaves; | |
2553 | } | |
2554 | ||
2555 | static int ocfs2_fill_new_dir_dx(struct ocfs2_super *osb, | |
2556 | handle_t *handle, | |
2557 | struct inode *parent, | |
2558 | struct inode *inode, | |
2559 | struct buffer_head *di_bh, | |
2560 | struct ocfs2_alloc_context *data_ac, | |
2561 | struct ocfs2_alloc_context *meta_ac) | |
2562 | { | |
4ed8a6bb | 2563 | int ret; |
9b7895ef MF |
2564 | struct buffer_head *leaf_bh = NULL; |
2565 | struct buffer_head *dx_root_bh = NULL; | |
9b7895ef | 2566 | struct ocfs2_dx_hinfo hinfo; |
4ed8a6bb MF |
2567 | struct ocfs2_dx_root_block *dx_root; |
2568 | struct ocfs2_dx_entry_list *entry_list; | |
9b7895ef MF |
2569 | |
2570 | /* | |
2571 | * Our strategy is to create the directory as though it were | |
2572 | * unindexed, then add the index block. This works with very | |
2573 | * little complication since the state of a new directory is a | |
2574 | * very well known quantity. | |
2575 | * | |
2576 | * Essentially, we have two dirents ("." and ".."), in the 1st | |
4ed8a6bb MF |
2577 | * block which need indexing. These are easily inserted into |
2578 | * the index block. | |
9b7895ef MF |
2579 | */ |
2580 | ||
2581 | ret = ocfs2_fill_new_dir_el(osb, handle, parent, inode, di_bh, | |
2582 | data_ac, &leaf_bh); | |
2583 | if (ret) { | |
2584 | mlog_errno(ret); | |
2585 | goto out; | |
2586 | } | |
2587 | ||
e7c17e43 | 2588 | ret = ocfs2_dx_dir_attach_index(osb, handle, inode, di_bh, leaf_bh, |
4ed8a6bb | 2589 | meta_ac, 1, &dx_root_bh); |
9b7895ef MF |
2590 | if (ret) { |
2591 | mlog_errno(ret); | |
2592 | goto out; | |
2593 | } | |
4ed8a6bb MF |
2594 | dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data; |
2595 | entry_list = &dx_root->dr_entries; | |
9b7895ef | 2596 | |
4ed8a6bb | 2597 | /* Buffer has been journaled for us by ocfs2_dx_dir_attach_index */ |
e7c17e43 | 2598 | ocfs2_dx_dir_name_hash(inode, ".", 1, &hinfo); |
4ed8a6bb | 2599 | ocfs2_dx_entry_list_insert(entry_list, &hinfo, leaf_bh->b_blocknr); |
9b7895ef MF |
2600 | |
2601 | ocfs2_dx_dir_name_hash(inode, "..", 2, &hinfo); | |
4ed8a6bb | 2602 | ocfs2_dx_entry_list_insert(entry_list, &hinfo, leaf_bh->b_blocknr); |
9b7895ef MF |
2603 | |
2604 | out: | |
9b7895ef MF |
2605 | brelse(dx_root_bh); |
2606 | brelse(leaf_bh); | |
2607 | return ret; | |
2608 | } | |
2609 | ||
2610 | int ocfs2_fill_new_dir(struct ocfs2_super *osb, | |
2611 | handle_t *handle, | |
2612 | struct inode *parent, | |
2613 | struct inode *inode, | |
2614 | struct buffer_head *fe_bh, | |
2615 | struct ocfs2_alloc_context *data_ac, | |
2616 | struct ocfs2_alloc_context *meta_ac) | |
2617 | ||
2618 | { | |
2619 | BUG_ON(!ocfs2_supports_inline_data(osb) && data_ac == NULL); | |
2620 | ||
2621 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) | |
2622 | return ocfs2_fill_new_dir_id(osb, handle, parent, inode, fe_bh); | |
2623 | ||
2624 | if (ocfs2_supports_indexed_dirs(osb)) | |
2625 | return ocfs2_fill_new_dir_dx(osb, handle, parent, inode, fe_bh, | |
2626 | data_ac, meta_ac); | |
2627 | ||
2628 | return ocfs2_fill_new_dir_el(osb, handle, parent, inode, fe_bh, | |
2629 | data_ac, NULL); | |
2630 | } | |
2631 | ||
2632 | static int ocfs2_dx_dir_index_block(struct inode *dir, | |
2633 | handle_t *handle, | |
2634 | struct buffer_head **dx_leaves, | |
2635 | int num_dx_leaves, | |
2636 | struct buffer_head *dirent_bh) | |
2637 | { | |
2638 | int ret, namelen, i; | |
2639 | char *de_buf, *limit; | |
2640 | struct ocfs2_dir_entry *de; | |
2641 | struct buffer_head *dx_leaf_bh; | |
2642 | struct ocfs2_dx_hinfo hinfo; | |
2643 | u64 dirent_blk = dirent_bh->b_blocknr; | |
2644 | ||
2645 | de_buf = dirent_bh->b_data; | |
2646 | limit = de_buf + dir->i_sb->s_blocksize; | |
2647 | ||
2648 | while (de_buf < limit) { | |
2649 | de = (struct ocfs2_dir_entry *)de_buf; | |
2650 | ||
2651 | namelen = de->name_len; | |
2652 | if (!namelen || !de->inode) | |
2653 | goto inc; | |
2654 | ||
2655 | ocfs2_dx_dir_name_hash(dir, de->name, namelen, &hinfo); | |
2656 | ||
2657 | i = ocfs2_dx_dir_hash_idx(OCFS2_SB(dir->i_sb), &hinfo); | |
2658 | dx_leaf_bh = dx_leaves[i]; | |
2659 | ||
2660 | ret = __ocfs2_dx_dir_leaf_insert(dir, handle, &hinfo, | |
2661 | dirent_blk, dx_leaf_bh); | |
2662 | if (ret) { | |
2663 | mlog_errno(ret); | |
2664 | goto out; | |
2665 | } | |
2666 | ||
2667 | inc: | |
2668 | de_buf += le16_to_cpu(de->rec_len); | |
2669 | } | |
2670 | ||
2671 | out: | |
2672 | return ret; | |
2673 | } | |
e7c17e43 MF |
2674 | |
2675 | /* | |
4ed8a6bb MF |
2676 | * XXX: This expects dx_root_bh to already be part of the transaction. |
2677 | */ | |
2678 | static void ocfs2_dx_dir_index_root_block(struct inode *dir, | |
2679 | struct buffer_head *dx_root_bh, | |
2680 | struct buffer_head *dirent_bh) | |
2681 | { | |
2682 | char *de_buf, *limit; | |
2683 | struct ocfs2_dx_root_block *dx_root; | |
2684 | struct ocfs2_dir_entry *de; | |
2685 | struct ocfs2_dx_hinfo hinfo; | |
2686 | u64 dirent_blk = dirent_bh->b_blocknr; | |
2687 | ||
2688 | dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data; | |
2689 | ||
2690 | de_buf = dirent_bh->b_data; | |
2691 | limit = de_buf + dir->i_sb->s_blocksize; | |
2692 | ||
2693 | while (de_buf < limit) { | |
2694 | de = (struct ocfs2_dir_entry *)de_buf; | |
2695 | ||
2696 | if (!de->name_len || !de->inode) | |
2697 | goto inc; | |
2698 | ||
2699 | ocfs2_dx_dir_name_hash(dir, de->name, de->name_len, &hinfo); | |
2700 | ||
2701 | mlog(0, | |
2702 | "dir: %llu, major: 0x%x minor: 0x%x, index: %u, name: %.*s\n", | |
2703 | (unsigned long long)dir->i_ino, hinfo.major_hash, | |
2704 | hinfo.minor_hash, | |
2705 | le16_to_cpu(dx_root->dr_entries.de_num_used), | |
2706 | de->name_len, de->name); | |
2707 | ||
2708 | ocfs2_dx_entry_list_insert(&dx_root->dr_entries, &hinfo, | |
2709 | dirent_blk); | |
2710 | inc: | |
2711 | de_buf += le16_to_cpu(de->rec_len); | |
2712 | } | |
2713 | } | |
2714 | ||
2715 | /* | |
2716 | * Count the number of inline directory entries in di_bh and compare | |
2717 | * them against the number of entries we can hold in an inline dx root | |
2718 | * block. | |
2719 | */ | |
2720 | static int ocfs2_new_dx_should_be_inline(struct inode *dir, | |
2721 | struct buffer_head *di_bh) | |
2722 | { | |
2723 | int dirent_count = 0; | |
2724 | char *de_buf, *limit; | |
2725 | struct ocfs2_dir_entry *de; | |
2726 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
2727 | ||
2728 | de_buf = di->id2.i_data.id_data; | |
2729 | limit = de_buf + i_size_read(dir); | |
2730 | ||
2731 | while (de_buf < limit) { | |
2732 | de = (struct ocfs2_dir_entry *)de_buf; | |
2733 | ||
2734 | if (de->name_len && de->inode) | |
2735 | dirent_count++; | |
2736 | ||
2737 | de_buf += le16_to_cpu(de->rec_len); | |
2738 | } | |
2739 | ||
2740 | /* We are careful to leave room for one extra record. */ | |
2741 | return dirent_count < ocfs2_dx_entries_per_root(dir->i_sb); | |
2742 | } | |
9b7895ef MF |
2743 | |
2744 | /* | |
2745 | * Expand rec_len of the rightmost dirent in a directory block so that it | |
2746 | * contains the end of our valid space for dirents. We do this during | |
2747 | * expansion from an inline directory to one with extents. The first dir block | |
2748 | * in that case is taken from the inline data portion of the inode block. | |
2749 | * | |
e7c17e43 MF |
2750 | * This will also return the largest amount of contiguous space for a dirent |
2751 | * in the block. That value is *not* necessarily the last dirent, even after | |
2752 | * expansion. The directory indexing code wants this value for free space | |
2753 | * accounting. We do this here since we're already walking the entire dir | |
2754 | * block. | |
2755 | * | |
9b7895ef MF |
2756 | * We add the dir trailer if this filesystem wants it. |
2757 | */ | |
e7c17e43 MF |
2758 | static unsigned int ocfs2_expand_last_dirent(char *start, unsigned int old_size, |
2759 | struct inode *dir) | |
9b7895ef | 2760 | { |
e7c17e43 | 2761 | struct super_block *sb = dir->i_sb; |
9b7895ef MF |
2762 | struct ocfs2_dir_entry *de; |
2763 | struct ocfs2_dir_entry *prev_de; | |
2764 | char *de_buf, *limit; | |
2765 | unsigned int new_size = sb->s_blocksize; | |
e7c17e43 MF |
2766 | unsigned int bytes, this_hole; |
2767 | unsigned int largest_hole = 0; | |
9b7895ef | 2768 | |
e7c17e43 | 2769 | if (ocfs2_new_dir_wants_trailer(dir)) |
9b7895ef MF |
2770 | new_size = ocfs2_dir_trailer_blk_off(sb); |
2771 | ||
2772 | bytes = new_size - old_size; | |
2773 | ||
2774 | limit = start + old_size; | |
2775 | de_buf = start; | |
2776 | de = (struct ocfs2_dir_entry *)de_buf; | |
5b6a3a2b | 2777 | do { |
e7c17e43 MF |
2778 | this_hole = ocfs2_figure_dirent_hole(de); |
2779 | if (this_hole > largest_hole) | |
2780 | largest_hole = this_hole; | |
2781 | ||
5b6a3a2b MF |
2782 | prev_de = de; |
2783 | de_buf += le16_to_cpu(de->rec_len); | |
2784 | de = (struct ocfs2_dir_entry *)de_buf; | |
2785 | } while (de_buf < limit); | |
2786 | ||
2787 | le16_add_cpu(&prev_de->rec_len, bytes); | |
e7c17e43 MF |
2788 | |
2789 | /* We need to double check this after modification of the final | |
2790 | * dirent. */ | |
2791 | this_hole = ocfs2_figure_dirent_hole(prev_de); | |
2792 | if (this_hole > largest_hole) | |
2793 | largest_hole = this_hole; | |
2794 | ||
2795 | if (largest_hole >= OCFS2_DIR_MIN_REC_LEN) | |
2796 | return largest_hole; | |
2797 | return 0; | |
5b6a3a2b MF |
2798 | } |
2799 | ||
2800 | /* | |
2801 | * We allocate enough clusters to fulfill "blocks_wanted", but set | |
2802 | * i_size to exactly one block. Ocfs2_extend_dir() will handle the | |
2803 | * rest automatically for us. | |
2804 | * | |
2805 | * *first_block_bh is a pointer to the 1st data block allocated to the | |
2806 | * directory. | |
2807 | */ | |
2808 | static int ocfs2_expand_inline_dir(struct inode *dir, struct buffer_head *di_bh, | |
2809 | unsigned int blocks_wanted, | |
9b7895ef | 2810 | struct ocfs2_dir_lookup_result *lookup, |
5b6a3a2b MF |
2811 | struct buffer_head **first_block_bh) |
2812 | { | |
9b7895ef | 2813 | u32 alloc, dx_alloc, bit_off, len; |
5b6a3a2b | 2814 | struct super_block *sb = dir->i_sb; |
4ed8a6bb | 2815 | int ret, i, num_dx_leaves = 0, dx_inline = 0, |
9b7895ef MF |
2816 | credits = ocfs2_inline_to_extents_credits(sb); |
2817 | u64 dx_insert_blkno, blkno, | |
2818 | bytes = blocks_wanted << sb->s_blocksize_bits; | |
5b6a3a2b MF |
2819 | struct ocfs2_super *osb = OCFS2_SB(dir->i_sb); |
2820 | struct ocfs2_inode_info *oi = OCFS2_I(dir); | |
2821 | struct ocfs2_alloc_context *data_ac; | |
9b7895ef | 2822 | struct ocfs2_alloc_context *meta_ac = NULL; |
5b6a3a2b | 2823 | struct buffer_head *dirdata_bh = NULL; |
9b7895ef MF |
2824 | struct buffer_head *dx_root_bh = NULL; |
2825 | struct buffer_head **dx_leaves = NULL; | |
5b6a3a2b MF |
2826 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; |
2827 | handle_t *handle; | |
f99b9b7c | 2828 | struct ocfs2_extent_tree et; |
9b7895ef MF |
2829 | struct ocfs2_extent_tree dx_et; |
2830 | int did_quota = 0, bytes_allocated = 0; | |
f99b9b7c | 2831 | |
8d6220d6 | 2832 | ocfs2_init_dinode_extent_tree(&et, dir, di_bh); |
5b6a3a2b MF |
2833 | |
2834 | alloc = ocfs2_clusters_for_bytes(sb, bytes); | |
9b7895ef MF |
2835 | dx_alloc = 0; |
2836 | ||
2837 | if (ocfs2_supports_indexed_dirs(osb)) { | |
9b7895ef MF |
2838 | credits += ocfs2_add_dir_index_credits(sb); |
2839 | ||
4ed8a6bb MF |
2840 | dx_inline = ocfs2_new_dx_should_be_inline(dir, di_bh); |
2841 | if (!dx_inline) { | |
2842 | /* Add one more cluster for an index leaf */ | |
2843 | dx_alloc++; | |
2844 | dx_leaves = ocfs2_dx_dir_kmalloc_leaves(sb, | |
2845 | &num_dx_leaves); | |
2846 | if (!dx_leaves) { | |
2847 | ret = -ENOMEM; | |
2848 | mlog_errno(ret); | |
2849 | goto out; | |
2850 | } | |
9b7895ef MF |
2851 | } |
2852 | ||
2853 | /* This gets us the dx_root */ | |
2854 | ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac); | |
2855 | if (ret) { | |
2856 | mlog_errno(ret); | |
2857 | goto out; | |
2858 | } | |
2859 | } | |
5b6a3a2b MF |
2860 | |
2861 | /* | |
9b7895ef MF |
2862 | * We should never need more than 2 clusters for the unindexed |
2863 | * tree - maximum dirent size is far less than one block. In | |
2864 | * fact, the only time we'd need more than one cluster is if | |
5b6a3a2b MF |
2865 | * blocksize == clustersize and the dirent won't fit in the |
2866 | * extra space that the expansion to a single block gives. As | |
2867 | * of today, that only happens on 4k/4k file systems. | |
2868 | */ | |
2869 | BUG_ON(alloc > 2); | |
2870 | ||
2871 | ret = ocfs2_reserve_clusters(osb, alloc, &data_ac); | |
2872 | if (ret) { | |
2873 | mlog_errno(ret); | |
2874 | goto out; | |
2875 | } | |
2876 | ||
2877 | down_write(&oi->ip_alloc_sem); | |
2878 | ||
2879 | /* | |
c78bad11 | 2880 | * Prepare for worst case allocation scenario of two separate |
9b7895ef | 2881 | * extents in the unindexed tree. |
5b6a3a2b MF |
2882 | */ |
2883 | if (alloc == 2) | |
2884 | credits += OCFS2_SUBALLOC_ALLOC; | |
2885 | ||
2886 | handle = ocfs2_start_trans(osb, credits); | |
2887 | if (IS_ERR(handle)) { | |
2888 | ret = PTR_ERR(handle); | |
2889 | mlog_errno(ret); | |
2890 | goto out_sem; | |
2891 | } | |
2892 | ||
a90714c1 | 2893 | if (vfs_dq_alloc_space_nodirty(dir, |
9b7895ef MF |
2894 | ocfs2_clusters_to_bytes(osb->sb, |
2895 | alloc + dx_alloc))) { | |
a90714c1 JK |
2896 | ret = -EDQUOT; |
2897 | goto out_commit; | |
2898 | } | |
2899 | did_quota = 1; | |
9b7895ef | 2900 | |
4ed8a6bb | 2901 | if (ocfs2_supports_indexed_dirs(osb) && !dx_inline) { |
9b7895ef MF |
2902 | /* |
2903 | * Allocate our index cluster first, to maximize the | |
2904 | * possibility that unindexed leaves grow | |
2905 | * contiguously. | |
2906 | */ | |
2907 | ret = __ocfs2_dx_dir_new_cluster(dir, 0, handle, data_ac, | |
2908 | dx_leaves, num_dx_leaves, | |
2909 | &dx_insert_blkno); | |
2910 | if (ret) { | |
2911 | mlog_errno(ret); | |
2912 | goto out_commit; | |
2913 | } | |
2914 | bytes_allocated += ocfs2_clusters_to_bytes(dir->i_sb, 1); | |
2915 | } | |
2916 | ||
5b6a3a2b MF |
2917 | /* |
2918 | * Try to claim as many clusters as the bitmap can give though | |
2919 | * if we only get one now, that's enough to continue. The rest | |
2920 | * will be claimed after the conversion to extents. | |
2921 | */ | |
2922 | ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off, &len); | |
2923 | if (ret) { | |
2924 | mlog_errno(ret); | |
2925 | goto out_commit; | |
2926 | } | |
9b7895ef | 2927 | bytes_allocated += ocfs2_clusters_to_bytes(dir->i_sb, 1); |
5b6a3a2b MF |
2928 | |
2929 | /* | |
2930 | * Operations are carefully ordered so that we set up the new | |
2931 | * data block first. The conversion from inline data to | |
2932 | * extents follows. | |
2933 | */ | |
2934 | blkno = ocfs2_clusters_to_blocks(dir->i_sb, bit_off); | |
2935 | dirdata_bh = sb_getblk(sb, blkno); | |
2936 | if (!dirdata_bh) { | |
2937 | ret = -EIO; | |
2938 | mlog_errno(ret); | |
2939 | goto out_commit; | |
2940 | } | |
2941 | ||
2942 | ocfs2_set_new_buffer_uptodate(dir, dirdata_bh); | |
2943 | ||
13723d00 JB |
2944 | ret = ocfs2_journal_access_db(handle, dir, dirdata_bh, |
2945 | OCFS2_JOURNAL_ACCESS_CREATE); | |
5b6a3a2b MF |
2946 | if (ret) { |
2947 | mlog_errno(ret); | |
2948 | goto out_commit; | |
2949 | } | |
2950 | ||
2951 | memcpy(dirdata_bh->b_data, di->id2.i_data.id_data, i_size_read(dir)); | |
2952 | memset(dirdata_bh->b_data + i_size_read(dir), 0, | |
2953 | sb->s_blocksize - i_size_read(dir)); | |
e7c17e43 MF |
2954 | i = ocfs2_expand_last_dirent(dirdata_bh->b_data, i_size_read(dir), dir); |
2955 | if (ocfs2_new_dir_wants_trailer(dir)) { | |
2956 | /* | |
2957 | * Prepare the dir trailer up front. It will otherwise look | |
2958 | * like a valid dirent. Even if inserting the index fails | |
2959 | * (unlikely), then all we'll have done is given first dir | |
2960 | * block a small amount of fragmentation. | |
2961 | */ | |
2962 | ocfs2_init_dir_trailer(dir, dirdata_bh, i); | |
2963 | } | |
5b6a3a2b MF |
2964 | |
2965 | ret = ocfs2_journal_dirty(handle, dirdata_bh); | |
2966 | if (ret) { | |
2967 | mlog_errno(ret); | |
2968 | goto out_commit; | |
2969 | } | |
2970 | ||
4ed8a6bb MF |
2971 | if (ocfs2_supports_indexed_dirs(osb) && !dx_inline) { |
2972 | /* | |
2973 | * Dx dirs with an external cluster need to do this up | |
2974 | * front. Inline dx root's get handled later, after | |
2975 | * we've allocated our root block. | |
2976 | */ | |
9b7895ef MF |
2977 | ret = ocfs2_dx_dir_index_block(dir, handle, dx_leaves, |
2978 | num_dx_leaves, dirdata_bh); | |
2979 | if (ret) { | |
2980 | mlog_errno(ret); | |
2981 | goto out_commit; | |
2982 | } | |
2983 | } | |
2984 | ||
5b6a3a2b MF |
2985 | /* |
2986 | * Set extent, i_size, etc on the directory. After this, the | |
2987 | * inode should contain the same exact dirents as before and | |
2988 | * be fully accessible from system calls. | |
2989 | * | |
2990 | * We let the later dirent insert modify c/mtime - to the user | |
2991 | * the data hasn't changed. | |
2992 | */ | |
13723d00 JB |
2993 | ret = ocfs2_journal_access_di(handle, dir, di_bh, |
2994 | OCFS2_JOURNAL_ACCESS_CREATE); | |
5b6a3a2b MF |
2995 | if (ret) { |
2996 | mlog_errno(ret); | |
2997 | goto out_commit; | |
2998 | } | |
2999 | ||
3000 | spin_lock(&oi->ip_lock); | |
3001 | oi->ip_dyn_features &= ~OCFS2_INLINE_DATA_FL; | |
3002 | di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features); | |
3003 | spin_unlock(&oi->ip_lock); | |
3004 | ||
3005 | ocfs2_dinode_new_extent_list(dir, di); | |
3006 | ||
3007 | i_size_write(dir, sb->s_blocksize); | |
3008 | dir->i_mtime = dir->i_ctime = CURRENT_TIME; | |
3009 | ||
3010 | di->i_size = cpu_to_le64(sb->s_blocksize); | |
3011 | di->i_ctime = di->i_mtime = cpu_to_le64(dir->i_ctime.tv_sec); | |
3012 | di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(dir->i_ctime.tv_nsec); | |
5b6a3a2b MF |
3013 | |
3014 | /* | |
3015 | * This should never fail as our extent list is empty and all | |
3016 | * related blocks have been journaled already. | |
3017 | */ | |
f99b9b7c JB |
3018 | ret = ocfs2_insert_extent(osb, handle, dir, &et, 0, blkno, len, |
3019 | 0, NULL); | |
5b6a3a2b MF |
3020 | if (ret) { |
3021 | mlog_errno(ret); | |
83cab533 | 3022 | goto out_commit; |
5b6a3a2b MF |
3023 | } |
3024 | ||
9780eb6c MF |
3025 | /* |
3026 | * Set i_blocks after the extent insert for the most up to | |
3027 | * date ip_clusters value. | |
3028 | */ | |
3029 | dir->i_blocks = ocfs2_inode_sector_count(dir); | |
3030 | ||
5b6a3a2b MF |
3031 | ret = ocfs2_journal_dirty(handle, di_bh); |
3032 | if (ret) { | |
3033 | mlog_errno(ret); | |
3034 | goto out_commit; | |
3035 | } | |
3036 | ||
9b7895ef MF |
3037 | if (ocfs2_supports_indexed_dirs(osb)) { |
3038 | ret = ocfs2_dx_dir_attach_index(osb, handle, dir, di_bh, | |
e7c17e43 | 3039 | dirdata_bh, meta_ac, dx_inline, |
4ed8a6bb | 3040 | &dx_root_bh); |
9b7895ef MF |
3041 | if (ret) { |
3042 | mlog_errno(ret); | |
3043 | goto out_commit; | |
3044 | } | |
3045 | ||
4ed8a6bb MF |
3046 | if (dx_inline) { |
3047 | ocfs2_dx_dir_index_root_block(dir, dx_root_bh, | |
3048 | dirdata_bh); | |
3049 | } else { | |
3050 | ocfs2_init_dx_root_extent_tree(&dx_et, dir, dx_root_bh); | |
3051 | ret = ocfs2_insert_extent(osb, handle, dir, &dx_et, 0, | |
3052 | dx_insert_blkno, 1, 0, NULL); | |
3053 | if (ret) | |
3054 | mlog_errno(ret); | |
3055 | } | |
9b7895ef MF |
3056 | } |
3057 | ||
5b6a3a2b MF |
3058 | /* |
3059 | * We asked for two clusters, but only got one in the 1st | |
3060 | * pass. Claim the 2nd cluster as a separate extent. | |
3061 | */ | |
3062 | if (alloc > len) { | |
3063 | ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off, | |
3064 | &len); | |
3065 | if (ret) { | |
3066 | mlog_errno(ret); | |
3067 | goto out_commit; | |
3068 | } | |
3069 | blkno = ocfs2_clusters_to_blocks(dir->i_sb, bit_off); | |
3070 | ||
f99b9b7c JB |
3071 | ret = ocfs2_insert_extent(osb, handle, dir, &et, 1, |
3072 | blkno, len, 0, NULL); | |
5b6a3a2b MF |
3073 | if (ret) { |
3074 | mlog_errno(ret); | |
83cab533 | 3075 | goto out_commit; |
5b6a3a2b | 3076 | } |
9b7895ef | 3077 | bytes_allocated += ocfs2_clusters_to_bytes(dir->i_sb, 1); |
5b6a3a2b MF |
3078 | } |
3079 | ||
3080 | *first_block_bh = dirdata_bh; | |
3081 | dirdata_bh = NULL; | |
9b7895ef MF |
3082 | if (ocfs2_supports_indexed_dirs(osb)) { |
3083 | unsigned int off; | |
3084 | ||
4ed8a6bb MF |
3085 | if (!dx_inline) { |
3086 | /* | |
3087 | * We need to return the correct block within the | |
3088 | * cluster which should hold our entry. | |
3089 | */ | |
3090 | off = ocfs2_dx_dir_hash_idx(OCFS2_SB(dir->i_sb), | |
3091 | &lookup->dl_hinfo); | |
3092 | get_bh(dx_leaves[off]); | |
3093 | lookup->dl_dx_leaf_bh = dx_leaves[off]; | |
3094 | } | |
3095 | lookup->dl_dx_root_bh = dx_root_bh; | |
3096 | dx_root_bh = NULL; | |
9b7895ef | 3097 | } |
5b6a3a2b MF |
3098 | |
3099 | out_commit: | |
a90714c1 | 3100 | if (ret < 0 && did_quota) |
9b7895ef MF |
3101 | vfs_dq_free_space_nodirty(dir, bytes_allocated); |
3102 | ||
5b6a3a2b MF |
3103 | ocfs2_commit_trans(osb, handle); |
3104 | ||
3105 | out_sem: | |
3106 | up_write(&oi->ip_alloc_sem); | |
3107 | ||
3108 | out: | |
3109 | if (data_ac) | |
3110 | ocfs2_free_alloc_context(data_ac); | |
9b7895ef MF |
3111 | if (meta_ac) |
3112 | ocfs2_free_alloc_context(meta_ac); | |
3113 | ||
3114 | if (dx_leaves) { | |
3115 | for (i = 0; i < num_dx_leaves; i++) | |
3116 | brelse(dx_leaves[i]); | |
3117 | kfree(dx_leaves); | |
3118 | } | |
5b6a3a2b MF |
3119 | |
3120 | brelse(dirdata_bh); | |
9b7895ef | 3121 | brelse(dx_root_bh); |
5b6a3a2b MF |
3122 | |
3123 | return ret; | |
3124 | } | |
3125 | ||
ccd979bd | 3126 | /* returns a bh of the 1st new block in the allocation. */ |
316f4b9f MF |
3127 | static int ocfs2_do_extend_dir(struct super_block *sb, |
3128 | handle_t *handle, | |
3129 | struct inode *dir, | |
3130 | struct buffer_head *parent_fe_bh, | |
3131 | struct ocfs2_alloc_context *data_ac, | |
3132 | struct ocfs2_alloc_context *meta_ac, | |
3133 | struct buffer_head **new_bh) | |
ccd979bd MF |
3134 | { |
3135 | int status; | |
a90714c1 | 3136 | int extend, did_quota = 0; |
8110b073 | 3137 | u64 p_blkno, v_blkno; |
ccd979bd MF |
3138 | |
3139 | spin_lock(&OCFS2_I(dir)->ip_lock); | |
3140 | extend = (i_size_read(dir) == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters)); | |
3141 | spin_unlock(&OCFS2_I(dir)->ip_lock); | |
3142 | ||
3143 | if (extend) { | |
dcd0538f MF |
3144 | u32 offset = OCFS2_I(dir)->ip_clusters; |
3145 | ||
a90714c1 JK |
3146 | if (vfs_dq_alloc_space_nodirty(dir, |
3147 | ocfs2_clusters_to_bytes(sb, 1))) { | |
3148 | status = -EDQUOT; | |
3149 | goto bail; | |
3150 | } | |
3151 | did_quota = 1; | |
3152 | ||
0eb8d47e TM |
3153 | status = ocfs2_add_inode_data(OCFS2_SB(sb), dir, &offset, |
3154 | 1, 0, parent_fe_bh, handle, | |
3155 | data_ac, meta_ac, NULL); | |
ccd979bd MF |
3156 | BUG_ON(status == -EAGAIN); |
3157 | if (status < 0) { | |
3158 | mlog_errno(status); | |
3159 | goto bail; | |
3160 | } | |
3161 | } | |
3162 | ||
8110b073 MF |
3163 | v_blkno = ocfs2_blocks_for_bytes(sb, i_size_read(dir)); |
3164 | status = ocfs2_extent_map_get_blocks(dir, v_blkno, &p_blkno, NULL, NULL); | |
ccd979bd MF |
3165 | if (status < 0) { |
3166 | mlog_errno(status); | |
3167 | goto bail; | |
3168 | } | |
3169 | ||
3170 | *new_bh = sb_getblk(sb, p_blkno); | |
3171 | if (!*new_bh) { | |
3172 | status = -EIO; | |
3173 | mlog_errno(status); | |
3174 | goto bail; | |
3175 | } | |
3176 | status = 0; | |
3177 | bail: | |
a90714c1 JK |
3178 | if (did_quota && status < 0) |
3179 | vfs_dq_free_space_nodirty(dir, ocfs2_clusters_to_bytes(sb, 1)); | |
ccd979bd MF |
3180 | mlog_exit(status); |
3181 | return status; | |
3182 | } | |
3183 | ||
5b6a3a2b MF |
3184 | /* |
3185 | * Assumes you already have a cluster lock on the directory. | |
3186 | * | |
3187 | * 'blocks_wanted' is only used if we have an inline directory which | |
3188 | * is to be turned into an extent based one. The size of the dirent to | |
3189 | * insert might be larger than the space gained by growing to just one | |
3190 | * block, so we may have to grow the inode by two blocks in that case. | |
e7c17e43 MF |
3191 | * |
3192 | * If the directory is already indexed, dx_root_bh must be provided. | |
5b6a3a2b | 3193 | */ |
ccd979bd MF |
3194 | static int ocfs2_extend_dir(struct ocfs2_super *osb, |
3195 | struct inode *dir, | |
3196 | struct buffer_head *parent_fe_bh, | |
5b6a3a2b | 3197 | unsigned int blocks_wanted, |
9b7895ef | 3198 | struct ocfs2_dir_lookup_result *lookup, |
ccd979bd MF |
3199 | struct buffer_head **new_de_bh) |
3200 | { | |
3201 | int status = 0; | |
ee19a779 | 3202 | int credits, num_free_extents, drop_alloc_sem = 0; |
ccd979bd MF |
3203 | loff_t dir_i_size; |
3204 | struct ocfs2_dinode *fe = (struct ocfs2_dinode *) parent_fe_bh->b_data; | |
811f933d | 3205 | struct ocfs2_extent_list *el = &fe->id2.i_list; |
ccd979bd MF |
3206 | struct ocfs2_alloc_context *data_ac = NULL; |
3207 | struct ocfs2_alloc_context *meta_ac = NULL; | |
1fabe148 | 3208 | handle_t *handle = NULL; |
ccd979bd MF |
3209 | struct buffer_head *new_bh = NULL; |
3210 | struct ocfs2_dir_entry * de; | |
3211 | struct super_block *sb = osb->sb; | |
f99b9b7c | 3212 | struct ocfs2_extent_tree et; |
e7c17e43 | 3213 | struct buffer_head *dx_root_bh = lookup->dl_dx_root_bh; |
ccd979bd MF |
3214 | |
3215 | mlog_entry_void(); | |
3216 | ||
5b6a3a2b | 3217 | if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
e7c17e43 MF |
3218 | /* |
3219 | * This would be a code error as an inline directory should | |
3220 | * never have an index root. | |
3221 | */ | |
3222 | BUG_ON(dx_root_bh); | |
3223 | ||
5b6a3a2b | 3224 | status = ocfs2_expand_inline_dir(dir, parent_fe_bh, |
9b7895ef MF |
3225 | blocks_wanted, lookup, |
3226 | &new_bh); | |
5b6a3a2b MF |
3227 | if (status) { |
3228 | mlog_errno(status); | |
3229 | goto bail; | |
3230 | } | |
3231 | ||
e7c17e43 MF |
3232 | /* Expansion from inline to an indexed directory will |
3233 | * have given us this. */ | |
3234 | dx_root_bh = lookup->dl_dx_root_bh; | |
3235 | ||
5b6a3a2b MF |
3236 | if (blocks_wanted == 1) { |
3237 | /* | |
3238 | * If the new dirent will fit inside the space | |
3239 | * created by pushing out to one block, then | |
3240 | * we can complete the operation | |
3241 | * here. Otherwise we have to expand i_size | |
3242 | * and format the 2nd block below. | |
3243 | */ | |
3244 | BUG_ON(new_bh == NULL); | |
3245 | goto bail_bh; | |
3246 | } | |
3247 | ||
3248 | /* | |
3249 | * Get rid of 'new_bh' - we want to format the 2nd | |
3250 | * data block and return that instead. | |
3251 | */ | |
3252 | brelse(new_bh); | |
3253 | new_bh = NULL; | |
3254 | ||
3255 | dir_i_size = i_size_read(dir); | |
3256 | credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS; | |
3257 | goto do_extend; | |
3258 | } | |
3259 | ||
ccd979bd | 3260 | dir_i_size = i_size_read(dir); |
b0697053 MF |
3261 | mlog(0, "extending dir %llu (i_size = %lld)\n", |
3262 | (unsigned long long)OCFS2_I(dir)->ip_blkno, dir_i_size); | |
ccd979bd | 3263 | |
ccd979bd MF |
3264 | /* dir->i_size is always block aligned. */ |
3265 | spin_lock(&OCFS2_I(dir)->ip_lock); | |
3266 | if (dir_i_size == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters)) { | |
3267 | spin_unlock(&OCFS2_I(dir)->ip_lock); | |
8d6220d6 | 3268 | ocfs2_init_dinode_extent_tree(&et, dir, parent_fe_bh); |
f99b9b7c | 3269 | num_free_extents = ocfs2_num_free_extents(osb, dir, &et); |
ccd979bd MF |
3270 | if (num_free_extents < 0) { |
3271 | status = num_free_extents; | |
3272 | mlog_errno(status); | |
3273 | goto bail; | |
3274 | } | |
3275 | ||
3276 | if (!num_free_extents) { | |
811f933d | 3277 | status = ocfs2_reserve_new_metadata(osb, el, &meta_ac); |
ccd979bd MF |
3278 | if (status < 0) { |
3279 | if (status != -ENOSPC) | |
3280 | mlog_errno(status); | |
3281 | goto bail; | |
3282 | } | |
3283 | } | |
3284 | ||
da5cbf2f | 3285 | status = ocfs2_reserve_clusters(osb, 1, &data_ac); |
ccd979bd MF |
3286 | if (status < 0) { |
3287 | if (status != -ENOSPC) | |
3288 | mlog_errno(status); | |
3289 | goto bail; | |
3290 | } | |
3291 | ||
811f933d | 3292 | credits = ocfs2_calc_extend_credits(sb, el, 1); |
ccd979bd MF |
3293 | } else { |
3294 | spin_unlock(&OCFS2_I(dir)->ip_lock); | |
3295 | credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS; | |
3296 | } | |
3297 | ||
5b6a3a2b | 3298 | do_extend: |
e7c17e43 MF |
3299 | if (ocfs2_dir_indexed(dir)) |
3300 | credits++; /* For attaching the new dirent block to the | |
3301 | * dx_root */ | |
3302 | ||
ee19a779 JB |
3303 | down_write(&OCFS2_I(dir)->ip_alloc_sem); |
3304 | drop_alloc_sem = 1; | |
3305 | ||
65eff9cc | 3306 | handle = ocfs2_start_trans(osb, credits); |
ccd979bd MF |
3307 | if (IS_ERR(handle)) { |
3308 | status = PTR_ERR(handle); | |
3309 | handle = NULL; | |
3310 | mlog_errno(status); | |
3311 | goto bail; | |
3312 | } | |
3313 | ||
3314 | status = ocfs2_do_extend_dir(osb->sb, handle, dir, parent_fe_bh, | |
3315 | data_ac, meta_ac, &new_bh); | |
3316 | if (status < 0) { | |
3317 | mlog_errno(status); | |
3318 | goto bail; | |
3319 | } | |
3320 | ||
3321 | ocfs2_set_new_buffer_uptodate(dir, new_bh); | |
3322 | ||
13723d00 JB |
3323 | status = ocfs2_journal_access_db(handle, dir, new_bh, |
3324 | OCFS2_JOURNAL_ACCESS_CREATE); | |
ccd979bd MF |
3325 | if (status < 0) { |
3326 | mlog_errno(status); | |
3327 | goto bail; | |
3328 | } | |
3329 | memset(new_bh->b_data, 0, sb->s_blocksize); | |
87d35a74 | 3330 | |
ccd979bd MF |
3331 | de = (struct ocfs2_dir_entry *) new_bh->b_data; |
3332 | de->inode = 0; | |
e7c17e43 | 3333 | if (ocfs2_supports_dir_trailer(dir)) { |
87d35a74 | 3334 | de->rec_len = cpu_to_le16(ocfs2_dir_trailer_blk_off(sb)); |
e7c17e43 MF |
3335 | |
3336 | ocfs2_init_dir_trailer(dir, new_bh, le16_to_cpu(de->rec_len)); | |
3337 | ||
3338 | if (ocfs2_dir_indexed(dir)) { | |
3339 | status = ocfs2_dx_dir_link_trailer(dir, handle, | |
3340 | dx_root_bh, new_bh); | |
3341 | if (status) { | |
3342 | mlog_errno(status); | |
3343 | goto bail; | |
3344 | } | |
3345 | } | |
87d35a74 MF |
3346 | } else { |
3347 | de->rec_len = cpu_to_le16(sb->s_blocksize); | |
3348 | } | |
ccd979bd MF |
3349 | status = ocfs2_journal_dirty(handle, new_bh); |
3350 | if (status < 0) { | |
3351 | mlog_errno(status); | |
3352 | goto bail; | |
3353 | } | |
3354 | ||
3355 | dir_i_size += dir->i_sb->s_blocksize; | |
3356 | i_size_write(dir, dir_i_size); | |
8110b073 | 3357 | dir->i_blocks = ocfs2_inode_sector_count(dir); |
ccd979bd MF |
3358 | status = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh); |
3359 | if (status < 0) { | |
3360 | mlog_errno(status); | |
3361 | goto bail; | |
3362 | } | |
3363 | ||
5b6a3a2b | 3364 | bail_bh: |
ccd979bd MF |
3365 | *new_de_bh = new_bh; |
3366 | get_bh(*new_de_bh); | |
3367 | bail: | |
ee19a779 JB |
3368 | if (drop_alloc_sem) |
3369 | up_write(&OCFS2_I(dir)->ip_alloc_sem); | |
ccd979bd | 3370 | if (handle) |
02dc1af4 | 3371 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
3372 | |
3373 | if (data_ac) | |
3374 | ocfs2_free_alloc_context(data_ac); | |
3375 | if (meta_ac) | |
3376 | ocfs2_free_alloc_context(meta_ac); | |
3377 | ||
a81cb88b | 3378 | brelse(new_bh); |
ccd979bd MF |
3379 | |
3380 | mlog_exit(status); | |
3381 | return status; | |
3382 | } | |
3383 | ||
5b6a3a2b MF |
3384 | static int ocfs2_find_dir_space_id(struct inode *dir, struct buffer_head *di_bh, |
3385 | const char *name, int namelen, | |
3386 | struct buffer_head **ret_de_bh, | |
3387 | unsigned int *blocks_wanted) | |
ccd979bd | 3388 | { |
5b6a3a2b | 3389 | int ret; |
87d35a74 | 3390 | struct super_block *sb = dir->i_sb; |
5b6a3a2b MF |
3391 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; |
3392 | struct ocfs2_dir_entry *de, *last_de = NULL; | |
3393 | char *de_buf, *limit; | |
3394 | unsigned long offset = 0; | |
87d35a74 MF |
3395 | unsigned int rec_len, new_rec_len, free_space = dir->i_sb->s_blocksize; |
3396 | ||
3397 | /* | |
3398 | * This calculates how many free bytes we'd have in block zero, should | |
3399 | * this function force expansion to an extent tree. | |
3400 | */ | |
e7c17e43 | 3401 | if (ocfs2_new_dir_wants_trailer(dir)) |
87d35a74 MF |
3402 | free_space = ocfs2_dir_trailer_blk_off(sb) - i_size_read(dir); |
3403 | else | |
3404 | free_space = dir->i_sb->s_blocksize - i_size_read(dir); | |
5b6a3a2b MF |
3405 | |
3406 | de_buf = di->id2.i_data.id_data; | |
3407 | limit = de_buf + i_size_read(dir); | |
3408 | rec_len = OCFS2_DIR_REC_LEN(namelen); | |
ccd979bd | 3409 | |
5b6a3a2b MF |
3410 | while (de_buf < limit) { |
3411 | de = (struct ocfs2_dir_entry *)de_buf; | |
ccd979bd | 3412 | |
5b6a3a2b MF |
3413 | if (!ocfs2_check_dir_entry(dir, de, di_bh, offset)) { |
3414 | ret = -ENOENT; | |
3415 | goto out; | |
3416 | } | |
3417 | if (ocfs2_match(namelen, name, de)) { | |
3418 | ret = -EEXIST; | |
3419 | goto out; | |
3420 | } | |
87d35a74 MF |
3421 | /* |
3422 | * No need to check for a trailing dirent record here as | |
3423 | * they're not used for inline dirs. | |
3424 | */ | |
3425 | ||
5b6a3a2b MF |
3426 | if (ocfs2_dirent_would_fit(de, rec_len)) { |
3427 | /* Ok, we found a spot. Return this bh and let | |
3428 | * the caller actually fill it in. */ | |
3429 | *ret_de_bh = di_bh; | |
3430 | get_bh(*ret_de_bh); | |
3431 | ret = 0; | |
3432 | goto out; | |
3433 | } | |
ccd979bd | 3434 | |
5b6a3a2b MF |
3435 | last_de = de; |
3436 | de_buf += le16_to_cpu(de->rec_len); | |
3437 | offset += le16_to_cpu(de->rec_len); | |
3438 | } | |
ccd979bd | 3439 | |
5b6a3a2b MF |
3440 | /* |
3441 | * We're going to require expansion of the directory - figure | |
3442 | * out how many blocks we'll need so that a place for the | |
3443 | * dirent can be found. | |
3444 | */ | |
3445 | *blocks_wanted = 1; | |
87d35a74 | 3446 | new_rec_len = le16_to_cpu(last_de->rec_len) + free_space; |
5b6a3a2b MF |
3447 | if (new_rec_len < (rec_len + OCFS2_DIR_REC_LEN(last_de->name_len))) |
3448 | *blocks_wanted = 2; | |
ccd979bd | 3449 | |
5b6a3a2b MF |
3450 | ret = -ENOSPC; |
3451 | out: | |
3452 | return ret; | |
3453 | } | |
3454 | ||
3455 | static int ocfs2_find_dir_space_el(struct inode *dir, const char *name, | |
3456 | int namelen, struct buffer_head **ret_de_bh) | |
3457 | { | |
3458 | unsigned long offset; | |
3459 | struct buffer_head *bh = NULL; | |
3460 | unsigned short rec_len; | |
3461 | struct ocfs2_dir_entry *de; | |
3462 | struct super_block *sb = dir->i_sb; | |
3463 | int status; | |
87d35a74 | 3464 | int blocksize = dir->i_sb->s_blocksize; |
ccd979bd | 3465 | |
a22305cc JB |
3466 | status = ocfs2_read_dir_block(dir, 0, &bh, 0); |
3467 | if (status) { | |
ccd979bd MF |
3468 | mlog_errno(status); |
3469 | goto bail; | |
3470 | } | |
3471 | ||
3472 | rec_len = OCFS2_DIR_REC_LEN(namelen); | |
3473 | offset = 0; | |
3474 | de = (struct ocfs2_dir_entry *) bh->b_data; | |
3475 | while (1) { | |
3476 | if ((char *)de >= sb->s_blocksize + bh->b_data) { | |
3477 | brelse(bh); | |
3478 | bh = NULL; | |
3479 | ||
3480 | if (i_size_read(dir) <= offset) { | |
5b6a3a2b MF |
3481 | /* |
3482 | * Caller will have to expand this | |
3483 | * directory. | |
3484 | */ | |
3485 | status = -ENOSPC; | |
ccd979bd MF |
3486 | goto bail; |
3487 | } | |
a22305cc JB |
3488 | status = ocfs2_read_dir_block(dir, |
3489 | offset >> sb->s_blocksize_bits, | |
3490 | &bh, 0); | |
3491 | if (status) { | |
ccd979bd MF |
3492 | mlog_errno(status); |
3493 | goto bail; | |
3494 | } | |
3495 | /* move to next block */ | |
3496 | de = (struct ocfs2_dir_entry *) bh->b_data; | |
3497 | } | |
3498 | if (!ocfs2_check_dir_entry(dir, de, bh, offset)) { | |
3499 | status = -ENOENT; | |
3500 | goto bail; | |
3501 | } | |
3502 | if (ocfs2_match(namelen, name, de)) { | |
3503 | status = -EEXIST; | |
3504 | goto bail; | |
3505 | } | |
87d35a74 MF |
3506 | |
3507 | if (ocfs2_skip_dir_trailer(dir, de, offset % blocksize, | |
3508 | blocksize)) | |
3509 | goto next; | |
3510 | ||
8553cf4f | 3511 | if (ocfs2_dirent_would_fit(de, rec_len)) { |
ccd979bd MF |
3512 | /* Ok, we found a spot. Return this bh and let |
3513 | * the caller actually fill it in. */ | |
3514 | *ret_de_bh = bh; | |
3515 | get_bh(*ret_de_bh); | |
3516 | status = 0; | |
3517 | goto bail; | |
3518 | } | |
87d35a74 | 3519 | next: |
ccd979bd MF |
3520 | offset += le16_to_cpu(de->rec_len); |
3521 | de = (struct ocfs2_dir_entry *)((char *) de + le16_to_cpu(de->rec_len)); | |
3522 | } | |
3523 | ||
3524 | status = 0; | |
3525 | bail: | |
a81cb88b | 3526 | brelse(bh); |
ccd979bd MF |
3527 | |
3528 | mlog_exit(status); | |
3529 | return status; | |
3530 | } | |
5b6a3a2b | 3531 | |
9b7895ef | 3532 | static int dx_leaf_sort_cmp(const void *a, const void *b) |
5b6a3a2b | 3533 | { |
9b7895ef MF |
3534 | const struct ocfs2_dx_entry *entry1 = a; |
3535 | const struct ocfs2_dx_entry *entry2 = b; | |
3536 | u32 major_hash1 = le32_to_cpu(entry1->dx_major_hash); | |
3537 | u32 major_hash2 = le32_to_cpu(entry2->dx_major_hash); | |
3538 | u32 minor_hash1 = le32_to_cpu(entry1->dx_minor_hash); | |
3539 | u32 minor_hash2 = le32_to_cpu(entry2->dx_minor_hash); | |
3540 | ||
3541 | if (major_hash1 > major_hash2) | |
3542 | return 1; | |
3543 | if (major_hash1 < major_hash2) | |
3544 | return -1; | |
3545 | ||
3546 | /* | |
3547 | * It is not strictly necessary to sort by minor | |
3548 | */ | |
3549 | if (minor_hash1 > minor_hash2) | |
3550 | return 1; | |
3551 | if (minor_hash1 < minor_hash2) | |
3552 | return -1; | |
3553 | return 0; | |
3554 | } | |
3555 | ||
3556 | static void dx_leaf_sort_swap(void *a, void *b, int size) | |
3557 | { | |
3558 | struct ocfs2_dx_entry *entry1 = a; | |
3559 | struct ocfs2_dx_entry *entry2 = b; | |
3560 | struct ocfs2_dx_entry tmp; | |
3561 | ||
3562 | BUG_ON(size != sizeof(*entry1)); | |
3563 | ||
3564 | tmp = *entry1; | |
3565 | *entry1 = *entry2; | |
3566 | *entry2 = tmp; | |
3567 | } | |
3568 | ||
3569 | static int ocfs2_dx_leaf_same_major(struct ocfs2_dx_leaf *dx_leaf) | |
3570 | { | |
3571 | struct ocfs2_dx_entry_list *dl_list = &dx_leaf->dl_list; | |
3572 | int i, num = le16_to_cpu(dl_list->de_num_used); | |
3573 | ||
3574 | for (i = 0; i < (num - 1); i++) { | |
3575 | if (le32_to_cpu(dl_list->de_entries[i].dx_major_hash) != | |
3576 | le32_to_cpu(dl_list->de_entries[i + 1].dx_major_hash)) | |
3577 | return 0; | |
3578 | } | |
3579 | ||
3580 | return 1; | |
3581 | } | |
3582 | ||
3583 | /* | |
3584 | * Find the optimal value to split this leaf on. This expects the leaf | |
3585 | * entries to be in sorted order. | |
3586 | * | |
3587 | * leaf_cpos is the cpos of the leaf we're splitting. insert_hash is | |
3588 | * the hash we want to insert. | |
3589 | * | |
3590 | * This function is only concerned with the major hash - that which | |
3591 | * determines which cluster an item belongs to. | |
3592 | */ | |
3593 | static int ocfs2_dx_dir_find_leaf_split(struct ocfs2_dx_leaf *dx_leaf, | |
3594 | u32 leaf_cpos, u32 insert_hash, | |
3595 | u32 *split_hash) | |
3596 | { | |
3597 | struct ocfs2_dx_entry_list *dl_list = &dx_leaf->dl_list; | |
3598 | int i, num_used = le16_to_cpu(dl_list->de_num_used); | |
3599 | int allsame; | |
3600 | ||
3601 | /* | |
3602 | * There's a couple rare, but nasty corner cases we have to | |
3603 | * check for here. All of them involve a leaf where all value | |
3604 | * have the same hash, which is what we look for first. | |
3605 | * | |
3606 | * Most of the time, all of the above is false, and we simply | |
3607 | * pick the median value for a split. | |
3608 | */ | |
3609 | allsame = ocfs2_dx_leaf_same_major(dx_leaf); | |
3610 | if (allsame) { | |
3611 | u32 val = le32_to_cpu(dl_list->de_entries[0].dx_major_hash); | |
3612 | ||
3613 | if (val == insert_hash) { | |
3614 | /* | |
3615 | * No matter where we would choose to split, | |
3616 | * the new entry would want to occupy the same | |
3617 | * block as these. Since there's no space left | |
3618 | * in their existing block, we know there | |
3619 | * won't be space after the split. | |
3620 | */ | |
3621 | return -ENOSPC; | |
3622 | } | |
3623 | ||
3624 | if (val == leaf_cpos) { | |
3625 | /* | |
3626 | * Because val is the same as leaf_cpos (which | |
3627 | * is the smallest value this leaf can have), | |
3628 | * yet is not equal to insert_hash, then we | |
3629 | * know that insert_hash *must* be larger than | |
3630 | * val (and leaf_cpos). At least cpos+1 in value. | |
3631 | * | |
3632 | * We also know then, that there cannot be an | |
3633 | * adjacent extent (otherwise we'd be looking | |
3634 | * at it). Choosing this value gives us a | |
3635 | * chance to get some contiguousness. | |
3636 | */ | |
3637 | *split_hash = leaf_cpos + 1; | |
3638 | return 0; | |
3639 | } | |
3640 | ||
3641 | if (val > insert_hash) { | |
3642 | /* | |
3643 | * val can not be the same as insert hash, and | |
3644 | * also must be larger than leaf_cpos. Also, | |
3645 | * we know that there can't be a leaf between | |
3646 | * cpos and val, otherwise the entries with | |
3647 | * hash 'val' would be there. | |
3648 | */ | |
3649 | *split_hash = val; | |
3650 | return 0; | |
3651 | } | |
3652 | ||
3653 | *split_hash = insert_hash; | |
3654 | return 0; | |
3655 | } | |
3656 | ||
3657 | /* | |
3658 | * Since the records are sorted and the checks above | |
3659 | * guaranteed that not all records in this block are the same, | |
3660 | * we simple travel forward, from the median, and pick the 1st | |
3661 | * record whose value is larger than leaf_cpos. | |
3662 | */ | |
3663 | for (i = (num_used / 2); i < num_used; i++) | |
3664 | if (le32_to_cpu(dl_list->de_entries[i].dx_major_hash) > | |
3665 | leaf_cpos) | |
3666 | break; | |
3667 | ||
3668 | BUG_ON(i == num_used); /* Should be impossible */ | |
3669 | *split_hash = le32_to_cpu(dl_list->de_entries[i].dx_major_hash); | |
3670 | return 0; | |
3671 | } | |
3672 | ||
3673 | /* | |
3674 | * Transfer all entries in orig_dx_leaves whose major hash is equal to or | |
3675 | * larger than split_hash into new_dx_leaves. We use a temporary | |
3676 | * buffer (tmp_dx_leaf) to make the changes to the original leaf blocks. | |
3677 | * | |
3678 | * Since the block offset inside a leaf (cluster) is a constant mask | |
3679 | * of minor_hash, we can optimize - an item at block offset X within | |
3680 | * the original cluster, will be at offset X within the new cluster. | |
3681 | */ | |
3682 | static void ocfs2_dx_dir_transfer_leaf(struct inode *dir, u32 split_hash, | |
3683 | handle_t *handle, | |
3684 | struct ocfs2_dx_leaf *tmp_dx_leaf, | |
3685 | struct buffer_head **orig_dx_leaves, | |
3686 | struct buffer_head **new_dx_leaves, | |
3687 | int num_dx_leaves) | |
3688 | { | |
3689 | int i, j, num_used; | |
3690 | u32 major_hash; | |
3691 | struct ocfs2_dx_leaf *orig_dx_leaf, *new_dx_leaf; | |
3692 | struct ocfs2_dx_entry_list *orig_list, *new_list, *tmp_list; | |
3693 | struct ocfs2_dx_entry *dx_entry; | |
3694 | ||
3695 | tmp_list = &tmp_dx_leaf->dl_list; | |
3696 | ||
3697 | for (i = 0; i < num_dx_leaves; i++) { | |
3698 | orig_dx_leaf = (struct ocfs2_dx_leaf *) orig_dx_leaves[i]->b_data; | |
3699 | orig_list = &orig_dx_leaf->dl_list; | |
3700 | new_dx_leaf = (struct ocfs2_dx_leaf *) new_dx_leaves[i]->b_data; | |
3701 | new_list = &new_dx_leaf->dl_list; | |
3702 | ||
3703 | num_used = le16_to_cpu(orig_list->de_num_used); | |
3704 | ||
3705 | memcpy(tmp_dx_leaf, orig_dx_leaf, dir->i_sb->s_blocksize); | |
3706 | tmp_list->de_num_used = cpu_to_le16(0); | |
3707 | memset(&tmp_list->de_entries, 0, sizeof(*dx_entry)*num_used); | |
3708 | ||
3709 | for (j = 0; j < num_used; j++) { | |
3710 | dx_entry = &orig_list->de_entries[j]; | |
3711 | major_hash = le32_to_cpu(dx_entry->dx_major_hash); | |
3712 | if (major_hash >= split_hash) | |
3713 | ocfs2_dx_dir_leaf_insert_tail(new_dx_leaf, | |
3714 | dx_entry); | |
3715 | else | |
3716 | ocfs2_dx_dir_leaf_insert_tail(tmp_dx_leaf, | |
3717 | dx_entry); | |
3718 | } | |
3719 | memcpy(orig_dx_leaf, tmp_dx_leaf, dir->i_sb->s_blocksize); | |
3720 | ||
3721 | ocfs2_journal_dirty(handle, orig_dx_leaves[i]); | |
3722 | ocfs2_journal_dirty(handle, new_dx_leaves[i]); | |
3723 | } | |
3724 | } | |
3725 | ||
3726 | static int ocfs2_dx_dir_rebalance_credits(struct ocfs2_super *osb, | |
3727 | struct ocfs2_dx_root_block *dx_root) | |
3728 | { | |
3729 | int credits = ocfs2_clusters_to_blocks(osb->sb, 2); | |
3730 | ||
3731 | credits += ocfs2_calc_extend_credits(osb->sb, &dx_root->dr_list, 1); | |
3732 | credits += ocfs2_quota_trans_credits(osb->sb); | |
3733 | return credits; | |
3734 | } | |
3735 | ||
3736 | /* | |
3737 | * Find the median value in dx_leaf_bh and allocate a new leaf to move | |
3738 | * half our entries into. | |
3739 | */ | |
3740 | static int ocfs2_dx_dir_rebalance(struct ocfs2_super *osb, struct inode *dir, | |
3741 | struct buffer_head *dx_root_bh, | |
3742 | struct buffer_head *dx_leaf_bh, | |
3743 | struct ocfs2_dx_hinfo *hinfo, u32 leaf_cpos, | |
3744 | u64 leaf_blkno) | |
3745 | { | |
3746 | struct ocfs2_dx_leaf *dx_leaf = (struct ocfs2_dx_leaf *)dx_leaf_bh->b_data; | |
3747 | int credits, ret, i, num_used, did_quota = 0; | |
3748 | u32 cpos, split_hash, insert_hash = hinfo->major_hash; | |
3749 | u64 orig_leaves_start; | |
3750 | int num_dx_leaves; | |
3751 | struct buffer_head **orig_dx_leaves = NULL; | |
3752 | struct buffer_head **new_dx_leaves = NULL; | |
3753 | struct ocfs2_alloc_context *data_ac = NULL, *meta_ac = NULL; | |
3754 | struct ocfs2_extent_tree et; | |
3755 | handle_t *handle = NULL; | |
3756 | struct ocfs2_dx_root_block *dx_root; | |
3757 | struct ocfs2_dx_leaf *tmp_dx_leaf = NULL; | |
3758 | ||
3759 | mlog(0, "DX Dir: %llu, rebalance leaf leaf_blkno: %llu insert: %u\n", | |
3760 | (unsigned long long)OCFS2_I(dir)->ip_blkno, | |
3761 | (unsigned long long)leaf_blkno, insert_hash); | |
3762 | ||
3763 | ocfs2_init_dx_root_extent_tree(&et, dir, dx_root_bh); | |
3764 | ||
3765 | dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data; | |
3766 | /* | |
3767 | * XXX: This is a rather large limit. We should use a more | |
3768 | * realistic value. | |
3769 | */ | |
3770 | if (le32_to_cpu(dx_root->dr_clusters) == UINT_MAX) | |
3771 | return -ENOSPC; | |
3772 | ||
3773 | num_used = le16_to_cpu(dx_leaf->dl_list.de_num_used); | |
3774 | if (num_used < le16_to_cpu(dx_leaf->dl_list.de_count)) { | |
3775 | mlog(ML_ERROR, "DX Dir: %llu, Asked to rebalance empty leaf: " | |
3776 | "%llu, %d\n", (unsigned long long)OCFS2_I(dir)->ip_blkno, | |
3777 | (unsigned long long)leaf_blkno, num_used); | |
3778 | ret = -EIO; | |
3779 | goto out; | |
3780 | } | |
3781 | ||
3782 | orig_dx_leaves = ocfs2_dx_dir_kmalloc_leaves(osb->sb, &num_dx_leaves); | |
3783 | if (!orig_dx_leaves) { | |
3784 | ret = -ENOMEM; | |
3785 | mlog_errno(ret); | |
3786 | goto out; | |
3787 | } | |
3788 | ||
3789 | new_dx_leaves = ocfs2_dx_dir_kmalloc_leaves(osb->sb, NULL); | |
3790 | if (!new_dx_leaves) { | |
3791 | ret = -ENOMEM; | |
3792 | mlog_errno(ret); | |
3793 | goto out; | |
3794 | } | |
3795 | ||
3796 | ret = ocfs2_lock_allocators(dir, &et, 1, 0, &data_ac, &meta_ac); | |
3797 | if (ret) { | |
3798 | if (ret != -ENOSPC) | |
3799 | mlog_errno(ret); | |
3800 | goto out; | |
3801 | } | |
3802 | ||
3803 | credits = ocfs2_dx_dir_rebalance_credits(osb, dx_root); | |
3804 | handle = ocfs2_start_trans(osb, credits); | |
3805 | if (IS_ERR(handle)) { | |
3806 | ret = PTR_ERR(handle); | |
3807 | handle = NULL; | |
3808 | mlog_errno(ret); | |
3809 | goto out; | |
3810 | } | |
3811 | ||
3812 | if (vfs_dq_alloc_space_nodirty(dir, | |
3813 | ocfs2_clusters_to_bytes(dir->i_sb, 1))) { | |
3814 | ret = -EDQUOT; | |
3815 | goto out_commit; | |
3816 | } | |
3817 | did_quota = 1; | |
3818 | ||
3819 | ret = ocfs2_journal_access_dl(handle, dir, dx_leaf_bh, | |
3820 | OCFS2_JOURNAL_ACCESS_WRITE); | |
3821 | if (ret) { | |
3822 | mlog_errno(ret); | |
3823 | goto out_commit; | |
3824 | } | |
3825 | ||
3826 | /* | |
3827 | * This block is changing anyway, so we can sort it in place. | |
3828 | */ | |
3829 | sort(dx_leaf->dl_list.de_entries, num_used, | |
3830 | sizeof(struct ocfs2_dx_entry), dx_leaf_sort_cmp, | |
3831 | dx_leaf_sort_swap); | |
3832 | ||
3833 | ret = ocfs2_journal_dirty(handle, dx_leaf_bh); | |
3834 | if (ret) { | |
3835 | mlog_errno(ret); | |
3836 | goto out_commit; | |
3837 | } | |
3838 | ||
3839 | ret = ocfs2_dx_dir_find_leaf_split(dx_leaf, leaf_cpos, insert_hash, | |
3840 | &split_hash); | |
3841 | if (ret) { | |
3842 | mlog_errno(ret); | |
3843 | goto out_commit; | |
3844 | } | |
3845 | ||
3846 | mlog(0, "Split leaf (%u) at %u, insert major hash is %u\n", | |
3847 | leaf_cpos, split_hash, insert_hash); | |
3848 | ||
3849 | /* | |
3850 | * We have to carefully order operations here. There are items | |
3851 | * which want to be in the new cluster before insert, but in | |
3852 | * order to put those items in the new cluster, we alter the | |
3853 | * old cluster. A failure to insert gets nasty. | |
3854 | * | |
3855 | * So, start by reserving writes to the old | |
3856 | * cluster. ocfs2_dx_dir_new_cluster will reserve writes on | |
3857 | * the new cluster for us, before inserting it. The insert | |
3858 | * won't happen if there's an error before that. Once the | |
3859 | * insert is done then, we can transfer from one leaf into the | |
3860 | * other without fear of hitting any error. | |
3861 | */ | |
3862 | ||
3863 | /* | |
3864 | * The leaf transfer wants some scratch space so that we don't | |
3865 | * wind up doing a bunch of expensive memmove(). | |
3866 | */ | |
3867 | tmp_dx_leaf = kmalloc(osb->sb->s_blocksize, GFP_NOFS); | |
3868 | if (!tmp_dx_leaf) { | |
3869 | ret = -ENOMEM; | |
3870 | mlog_errno(ret); | |
3871 | goto out_commit; | |
3872 | } | |
3873 | ||
3874 | orig_leaves_start = leaf_blkno & ~(osb->s_clustersize_bits - | |
3875 | osb->sb->s_blocksize_bits); | |
3876 | ret = ocfs2_read_dx_leaves(dir, orig_leaves_start, num_dx_leaves, | |
3877 | orig_dx_leaves); | |
3878 | if (ret) { | |
3879 | mlog_errno(ret); | |
3880 | goto out_commit; | |
3881 | } | |
3882 | ||
3883 | for (i = 0; i < num_dx_leaves; i++) { | |
3884 | ret = ocfs2_journal_access_dl(handle, dir, orig_dx_leaves[i], | |
3885 | OCFS2_JOURNAL_ACCESS_WRITE); | |
3886 | if (ret) { | |
3887 | mlog_errno(ret); | |
3888 | goto out_commit; | |
3889 | } | |
3890 | } | |
3891 | ||
3892 | cpos = split_hash; | |
3893 | ret = ocfs2_dx_dir_new_cluster(dir, &et, cpos, handle, | |
3894 | data_ac, meta_ac, new_dx_leaves, | |
3895 | num_dx_leaves); | |
3896 | if (ret) { | |
3897 | mlog_errno(ret); | |
3898 | goto out_commit; | |
3899 | } | |
3900 | ||
3901 | ocfs2_dx_dir_transfer_leaf(dir, split_hash, handle, tmp_dx_leaf, | |
3902 | orig_dx_leaves, new_dx_leaves, num_dx_leaves); | |
3903 | ||
3904 | out_commit: | |
3905 | if (ret < 0 && did_quota) | |
3906 | vfs_dq_free_space_nodirty(dir, | |
3907 | ocfs2_clusters_to_bytes(dir->i_sb, 1)); | |
3908 | ||
3909 | ocfs2_commit_trans(osb, handle); | |
3910 | ||
3911 | out: | |
3912 | if (orig_dx_leaves || new_dx_leaves) { | |
3913 | for (i = 0; i < num_dx_leaves; i++) { | |
3914 | if (orig_dx_leaves) | |
3915 | brelse(orig_dx_leaves[i]); | |
3916 | if (new_dx_leaves) | |
3917 | brelse(new_dx_leaves[i]); | |
3918 | } | |
3919 | kfree(orig_dx_leaves); | |
3920 | kfree(new_dx_leaves); | |
3921 | } | |
3922 | ||
3923 | if (meta_ac) | |
3924 | ocfs2_free_alloc_context(meta_ac); | |
3925 | if (data_ac) | |
3926 | ocfs2_free_alloc_context(data_ac); | |
3927 | ||
3928 | kfree(tmp_dx_leaf); | |
3929 | return ret; | |
3930 | } | |
3931 | ||
e7c17e43 MF |
3932 | static int ocfs2_find_dir_space_dx(struct ocfs2_super *osb, struct inode *dir, |
3933 | struct buffer_head *di_bh, | |
3934 | struct buffer_head *dx_root_bh, | |
3935 | const char *name, int namelen, | |
3936 | struct ocfs2_dir_lookup_result *lookup) | |
3937 | { | |
3938 | int ret, rebalanced = 0; | |
3939 | struct ocfs2_dx_root_block *dx_root; | |
3940 | struct buffer_head *dx_leaf_bh = NULL; | |
3941 | struct ocfs2_dx_leaf *dx_leaf; | |
3942 | u64 blkno; | |
3943 | u32 leaf_cpos; | |
3944 | ||
3945 | dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data; | |
3946 | ||
3947 | restart_search: | |
3948 | ret = ocfs2_dx_dir_lookup(dir, &dx_root->dr_list, &lookup->dl_hinfo, | |
3949 | &leaf_cpos, &blkno); | |
3950 | if (ret) { | |
3951 | mlog_errno(ret); | |
3952 | goto out; | |
3953 | } | |
3954 | ||
3955 | ret = ocfs2_read_dx_leaf(dir, blkno, &dx_leaf_bh); | |
3956 | if (ret) { | |
3957 | mlog_errno(ret); | |
3958 | goto out; | |
3959 | } | |
3960 | ||
3961 | dx_leaf = (struct ocfs2_dx_leaf *)dx_leaf_bh->b_data; | |
3962 | ||
3963 | if (le16_to_cpu(dx_leaf->dl_list.de_num_used) >= | |
3964 | le16_to_cpu(dx_leaf->dl_list.de_count)) { | |
3965 | if (rebalanced) { | |
3966 | /* | |
3967 | * Rebalancing should have provided us with | |
3968 | * space in an appropriate leaf. | |
3969 | * | |
3970 | * XXX: Is this an abnormal condition then? | |
3971 | * Should we print a message here? | |
3972 | */ | |
3973 | ret = -ENOSPC; | |
3974 | goto out; | |
3975 | } | |
3976 | ||
3977 | ret = ocfs2_dx_dir_rebalance(osb, dir, dx_root_bh, dx_leaf_bh, | |
3978 | &lookup->dl_hinfo, leaf_cpos, | |
3979 | blkno); | |
3980 | if (ret) { | |
3981 | if (ret != -ENOSPC) | |
3982 | mlog_errno(ret); | |
3983 | goto out; | |
3984 | } | |
3985 | ||
3986 | /* | |
3987 | * Restart the lookup. The rebalance might have | |
3988 | * changed which block our item fits into. Mark our | |
3989 | * progress, so we only execute this once. | |
3990 | */ | |
3991 | brelse(dx_leaf_bh); | |
3992 | dx_leaf_bh = NULL; | |
3993 | rebalanced = 1; | |
3994 | goto restart_search; | |
3995 | } | |
3996 | ||
3997 | lookup->dl_dx_leaf_bh = dx_leaf_bh; | |
3998 | dx_leaf_bh = NULL; | |
3999 | ||
4000 | out: | |
4001 | brelse(dx_leaf_bh); | |
4002 | return ret; | |
4003 | } | |
4004 | ||
4005 | static int ocfs2_search_dx_free_list(struct inode *dir, | |
4006 | struct buffer_head *dx_root_bh, | |
4007 | int namelen, | |
4008 | struct ocfs2_dir_lookup_result *lookup) | |
4009 | { | |
4010 | int ret = -ENOSPC; | |
4011 | struct buffer_head *leaf_bh = NULL, *prev_leaf_bh = NULL; | |
4012 | struct ocfs2_dir_block_trailer *db; | |
4013 | u64 next_block; | |
4014 | int rec_len = OCFS2_DIR_REC_LEN(namelen); | |
4015 | struct ocfs2_dx_root_block *dx_root; | |
4016 | ||
4017 | dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data; | |
4018 | next_block = le64_to_cpu(dx_root->dr_free_blk); | |
4019 | ||
4020 | while (next_block) { | |
4021 | brelse(prev_leaf_bh); | |
4022 | prev_leaf_bh = leaf_bh; | |
4023 | leaf_bh = NULL; | |
4024 | ||
4025 | ret = ocfs2_read_dir_block_direct(dir, next_block, &leaf_bh); | |
4026 | if (ret) { | |
4027 | mlog_errno(ret); | |
4028 | goto out; | |
4029 | } | |
4030 | ||
4031 | db = ocfs2_trailer_from_bh(leaf_bh, dir->i_sb); | |
4032 | if (rec_len <= le16_to_cpu(db->db_free_rec_len)) { | |
4033 | lookup->dl_leaf_bh = leaf_bh; | |
4034 | lookup->dl_prev_leaf_bh = prev_leaf_bh; | |
4035 | leaf_bh = NULL; | |
4036 | prev_leaf_bh = NULL; | |
4037 | break; | |
4038 | } | |
4039 | ||
4040 | next_block = le64_to_cpu(db->db_free_next); | |
4041 | } | |
4042 | ||
4043 | if (!next_block) | |
4044 | ret = -ENOSPC; | |
4045 | ||
4046 | out: | |
4047 | ||
4048 | brelse(leaf_bh); | |
4049 | brelse(prev_leaf_bh); | |
4050 | return ret; | |
4051 | } | |
4052 | ||
4ed8a6bb MF |
4053 | static int ocfs2_expand_inline_dx_root(struct inode *dir, |
4054 | struct buffer_head *dx_root_bh) | |
4055 | { | |
4056 | int ret, num_dx_leaves, i, j, did_quota = 0; | |
4057 | struct buffer_head **dx_leaves = NULL; | |
4058 | struct ocfs2_extent_tree et; | |
4059 | u64 insert_blkno; | |
4060 | struct ocfs2_alloc_context *data_ac = NULL; | |
4061 | struct ocfs2_super *osb = OCFS2_SB(dir->i_sb); | |
4062 | handle_t *handle = NULL; | |
4063 | struct ocfs2_dx_root_block *dx_root; | |
4064 | struct ocfs2_dx_entry_list *entry_list; | |
4065 | struct ocfs2_dx_entry *dx_entry; | |
4066 | struct ocfs2_dx_leaf *target_leaf; | |
4067 | ||
4068 | ret = ocfs2_reserve_clusters(osb, 1, &data_ac); | |
4069 | if (ret) { | |
4070 | mlog_errno(ret); | |
4071 | goto out; | |
4072 | } | |
4073 | ||
4074 | dx_leaves = ocfs2_dx_dir_kmalloc_leaves(osb->sb, &num_dx_leaves); | |
4075 | if (!dx_leaves) { | |
4076 | ret = -ENOMEM; | |
4077 | mlog_errno(ret); | |
4078 | goto out; | |
4079 | } | |
4080 | ||
4081 | handle = ocfs2_start_trans(osb, ocfs2_calc_dxi_expand_credits(osb->sb)); | |
4082 | if (IS_ERR(handle)) { | |
4083 | ret = PTR_ERR(handle); | |
4084 | mlog_errno(ret); | |
4085 | goto out; | |
4086 | } | |
4087 | ||
4088 | if (vfs_dq_alloc_space_nodirty(dir, | |
4089 | ocfs2_clusters_to_bytes(osb->sb, 1))) { | |
4090 | ret = -EDQUOT; | |
4091 | goto out_commit; | |
4092 | } | |
4093 | did_quota = 1; | |
4094 | ||
4095 | /* | |
4096 | * We do this up front, before the allocation, so that a | |
4097 | * failure to add the dx_root_bh to the journal won't result | |
4098 | * us losing clusters. | |
4099 | */ | |
4100 | ret = ocfs2_journal_access_dr(handle, dir, dx_root_bh, | |
4101 | OCFS2_JOURNAL_ACCESS_WRITE); | |
4102 | if (ret) { | |
4103 | mlog_errno(ret); | |
4104 | goto out_commit; | |
4105 | } | |
4106 | ||
4107 | ret = __ocfs2_dx_dir_new_cluster(dir, 0, handle, data_ac, dx_leaves, | |
4108 | num_dx_leaves, &insert_blkno); | |
4109 | if (ret) { | |
4110 | mlog_errno(ret); | |
4111 | goto out_commit; | |
4112 | } | |
4113 | ||
4114 | /* | |
4115 | * Transfer the entries from our dx_root into the appropriate | |
4116 | * block | |
4117 | */ | |
4118 | dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data; | |
4119 | entry_list = &dx_root->dr_entries; | |
4120 | ||
4121 | for (i = 0; i < le16_to_cpu(entry_list->de_num_used); i++) { | |
4122 | dx_entry = &entry_list->de_entries[i]; | |
4123 | ||
4124 | j = __ocfs2_dx_dir_hash_idx(osb, | |
4125 | le32_to_cpu(dx_entry->dx_minor_hash)); | |
4126 | target_leaf = (struct ocfs2_dx_leaf *)dx_leaves[j]->b_data; | |
4127 | ||
4128 | ocfs2_dx_dir_leaf_insert_tail(target_leaf, dx_entry); | |
4129 | ||
4130 | /* Each leaf has been passed to the journal already | |
4131 | * via __ocfs2_dx_dir_new_cluster() */ | |
4132 | } | |
4133 | ||
4134 | dx_root->dr_flags &= ~OCFS2_DX_FLAG_INLINE; | |
4135 | memset(&dx_root->dr_list, 0, osb->sb->s_blocksize - | |
4136 | offsetof(struct ocfs2_dx_root_block, dr_list)); | |
4137 | dx_root->dr_list.l_count = | |
4138 | cpu_to_le16(ocfs2_extent_recs_per_dx_root(osb->sb)); | |
4139 | ||
4140 | /* This should never fail considering we start with an empty | |
4141 | * dx_root. */ | |
4142 | ocfs2_init_dx_root_extent_tree(&et, dir, dx_root_bh); | |
4143 | ret = ocfs2_insert_extent(osb, handle, dir, &et, 0, | |
4144 | insert_blkno, 1, 0, NULL); | |
4145 | if (ret) | |
4146 | mlog_errno(ret); | |
4147 | did_quota = 0; | |
4148 | ||
4149 | ocfs2_journal_dirty(handle, dx_root_bh); | |
4150 | ||
4151 | out_commit: | |
4152 | if (ret < 0 && did_quota) | |
4153 | vfs_dq_free_space_nodirty(dir, | |
4154 | ocfs2_clusters_to_bytes(dir->i_sb, 1)); | |
4155 | ||
4156 | ocfs2_commit_trans(osb, handle); | |
4157 | ||
4158 | out: | |
4159 | if (data_ac) | |
4160 | ocfs2_free_alloc_context(data_ac); | |
4161 | ||
4162 | if (dx_leaves) { | |
4163 | for (i = 0; i < num_dx_leaves; i++) | |
4164 | brelse(dx_leaves[i]); | |
4165 | kfree(dx_leaves); | |
4166 | } | |
4167 | return ret; | |
4168 | } | |
4169 | ||
4170 | static int ocfs2_inline_dx_has_space(struct buffer_head *dx_root_bh) | |
4171 | { | |
4172 | struct ocfs2_dx_root_block *dx_root; | |
4173 | struct ocfs2_dx_entry_list *entry_list; | |
4174 | ||
4175 | dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data; | |
4176 | entry_list = &dx_root->dr_entries; | |
4177 | ||
4178 | if (le16_to_cpu(entry_list->de_num_used) >= | |
4179 | le16_to_cpu(entry_list->de_count)) | |
4180 | return -ENOSPC; | |
4181 | ||
4182 | return 0; | |
4183 | } | |
4184 | ||
e7c17e43 MF |
4185 | static int ocfs2_prepare_dx_dir_for_insert(struct inode *dir, |
4186 | struct buffer_head *di_bh, | |
4187 | const char *name, | |
4188 | int namelen, | |
4189 | struct ocfs2_dir_lookup_result *lookup) | |
9b7895ef | 4190 | { |
e7c17e43 MF |
4191 | int ret, free_dx_root = 1; |
4192 | struct ocfs2_super *osb = OCFS2_SB(dir->i_sb); | |
9b7895ef | 4193 | struct buffer_head *dx_root_bh = NULL; |
e7c17e43 | 4194 | struct buffer_head *leaf_bh = NULL; |
9b7895ef | 4195 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; |
e7c17e43 | 4196 | struct ocfs2_dx_root_block *dx_root; |
9b7895ef MF |
4197 | |
4198 | ret = ocfs2_read_dx_root(dir, di, &dx_root_bh); | |
4199 | if (ret) { | |
4200 | mlog_errno(ret); | |
4201 | goto out; | |
4202 | } | |
4203 | ||
4204 | dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data; | |
4ed8a6bb MF |
4205 | if (ocfs2_dx_root_inline(dx_root)) { |
4206 | ret = ocfs2_inline_dx_has_space(dx_root_bh); | |
4207 | ||
4208 | if (ret == 0) | |
4209 | goto search_el; | |
4210 | ||
4211 | /* | |
4212 | * We ran out of room in the root block. Expand it to | |
4213 | * an extent, then allow ocfs2_find_dir_space_dx to do | |
4214 | * the rest. | |
4215 | */ | |
4216 | ret = ocfs2_expand_inline_dx_root(dir, dx_root_bh); | |
4217 | if (ret) { | |
4218 | mlog_errno(ret); | |
4219 | goto out; | |
4220 | } | |
4221 | } | |
9b7895ef | 4222 | |
e7c17e43 MF |
4223 | /* |
4224 | * Insert preparation for an indexed directory is split into two | |
4225 | * steps. The call to find_dir_space_dx reserves room in the index for | |
4226 | * an additional item. If we run out of space there, it's a real error | |
4227 | * we can't continue on. | |
4228 | */ | |
4229 | ret = ocfs2_find_dir_space_dx(osb, dir, di_bh, dx_root_bh, name, | |
4230 | namelen, lookup); | |
9b7895ef MF |
4231 | if (ret) { |
4232 | mlog_errno(ret); | |
4233 | goto out; | |
4234 | } | |
4235 | ||
e7c17e43 MF |
4236 | search_el: |
4237 | /* | |
4238 | * Next, we need to find space in the unindexed tree. This call | |
4239 | * searches using the free space linked list. If the unindexed tree | |
4240 | * lacks sufficient space, we'll expand it below. The expansion code | |
4241 | * is smart enough to add any new blocks to the free space list. | |
4242 | */ | |
4243 | ret = ocfs2_search_dx_free_list(dir, dx_root_bh, namelen, lookup); | |
4244 | if (ret && ret != -ENOSPC) { | |
9b7895ef MF |
4245 | mlog_errno(ret); |
4246 | goto out; | |
4247 | } | |
4248 | ||
e7c17e43 MF |
4249 | /* Do this up here - ocfs2_extend_dir might need the dx_root */ |
4250 | lookup->dl_dx_root_bh = dx_root_bh; | |
4251 | free_dx_root = 0; | |
9b7895ef | 4252 | |
e7c17e43 MF |
4253 | if (ret == -ENOSPC) { |
4254 | ret = ocfs2_extend_dir(osb, dir, di_bh, 1, lookup, &leaf_bh); | |
9b7895ef | 4255 | |
9b7895ef | 4256 | if (ret) { |
e7c17e43 | 4257 | mlog_errno(ret); |
9b7895ef MF |
4258 | goto out; |
4259 | } | |
4260 | ||
4261 | /* | |
e7c17e43 MF |
4262 | * We make the assumption here that new leaf blocks are added |
4263 | * to the front of our free list. | |
9b7895ef | 4264 | */ |
e7c17e43 MF |
4265 | lookup->dl_prev_leaf_bh = NULL; |
4266 | lookup->dl_leaf_bh = leaf_bh; | |
9b7895ef MF |
4267 | } |
4268 | ||
9b7895ef | 4269 | out: |
e7c17e43 MF |
4270 | if (free_dx_root) |
4271 | brelse(dx_root_bh); | |
9b7895ef MF |
4272 | return ret; |
4273 | } | |
4274 | ||
4275 | /* | |
4276 | * Get a directory ready for insert. Any directory allocation required | |
4277 | * happens here. Success returns zero, and enough context in the dir | |
4278 | * lookup result that ocfs2_add_entry() will be able complete the task | |
4279 | * with minimal performance impact. | |
4280 | */ | |
4281 | int ocfs2_prepare_dir_for_insert(struct ocfs2_super *osb, | |
4282 | struct inode *dir, | |
4283 | struct buffer_head *parent_fe_bh, | |
4284 | const char *name, | |
4285 | int namelen, | |
4286 | struct ocfs2_dir_lookup_result *lookup) | |
4287 | { | |
4288 | int ret; | |
5b6a3a2b MF |
4289 | unsigned int blocks_wanted = 1; |
4290 | struct buffer_head *bh = NULL; | |
4291 | ||
4292 | mlog(0, "getting ready to insert namelen %d into dir %llu\n", | |
4293 | namelen, (unsigned long long)OCFS2_I(dir)->ip_blkno); | |
4294 | ||
5b6a3a2b MF |
4295 | if (!namelen) { |
4296 | ret = -EINVAL; | |
4297 | mlog_errno(ret); | |
4298 | goto out; | |
4299 | } | |
4300 | ||
9b7895ef MF |
4301 | /* |
4302 | * Do this up front to reduce confusion. | |
4303 | * | |
4304 | * The directory might start inline, then be turned into an | |
4305 | * indexed one, in which case we'd need to hash deep inside | |
4306 | * ocfs2_find_dir_space_id(). Since | |
4307 | * ocfs2_prepare_dx_dir_for_insert() also needs this hash | |
4308 | * done, there seems no point in spreading out the calls. We | |
4309 | * can optimize away the case where the file system doesn't | |
4310 | * support indexing. | |
4311 | */ | |
4312 | if (ocfs2_supports_indexed_dirs(osb)) | |
4313 | ocfs2_dx_dir_name_hash(dir, name, namelen, &lookup->dl_hinfo); | |
4314 | ||
4315 | if (ocfs2_dir_indexed(dir)) { | |
e7c17e43 MF |
4316 | ret = ocfs2_prepare_dx_dir_for_insert(dir, parent_fe_bh, |
4317 | name, namelen, lookup); | |
4318 | if (ret) | |
9b7895ef | 4319 | mlog_errno(ret); |
e7c17e43 | 4320 | goto out; |
9b7895ef MF |
4321 | } |
4322 | ||
5b6a3a2b MF |
4323 | if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
4324 | ret = ocfs2_find_dir_space_id(dir, parent_fe_bh, name, | |
4325 | namelen, &bh, &blocks_wanted); | |
4326 | } else | |
4327 | ret = ocfs2_find_dir_space_el(dir, name, namelen, &bh); | |
4328 | ||
4329 | if (ret && ret != -ENOSPC) { | |
4330 | mlog_errno(ret); | |
4331 | goto out; | |
4332 | } | |
4333 | ||
4334 | if (ret == -ENOSPC) { | |
4335 | /* | |
4336 | * We have to expand the directory to add this name. | |
4337 | */ | |
4338 | BUG_ON(bh); | |
4339 | ||
4340 | ret = ocfs2_extend_dir(osb, dir, parent_fe_bh, blocks_wanted, | |
9b7895ef | 4341 | lookup, &bh); |
5b6a3a2b MF |
4342 | if (ret) { |
4343 | if (ret != -ENOSPC) | |
4344 | mlog_errno(ret); | |
4345 | goto out; | |
4346 | } | |
4347 | ||
4348 | BUG_ON(!bh); | |
4349 | } | |
4350 | ||
4a12ca3a | 4351 | lookup->dl_leaf_bh = bh; |
5b6a3a2b MF |
4352 | bh = NULL; |
4353 | out: | |
a81cb88b | 4354 | brelse(bh); |
5b6a3a2b MF |
4355 | return ret; |
4356 | } | |
9b7895ef MF |
4357 | |
4358 | static int ocfs2_dx_dir_remove_index(struct inode *dir, | |
4359 | struct buffer_head *di_bh, | |
4360 | struct buffer_head *dx_root_bh) | |
4361 | { | |
4362 | int ret; | |
4363 | struct ocfs2_super *osb = OCFS2_SB(dir->i_sb); | |
4364 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
4365 | struct ocfs2_dx_root_block *dx_root; | |
4366 | struct inode *dx_alloc_inode = NULL; | |
4367 | struct buffer_head *dx_alloc_bh = NULL; | |
4368 | handle_t *handle; | |
4369 | u64 blk; | |
4370 | u16 bit; | |
4371 | u64 bg_blkno; | |
4372 | ||
4373 | dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data; | |
4374 | ||
4375 | dx_alloc_inode = ocfs2_get_system_file_inode(osb, | |
4376 | EXTENT_ALLOC_SYSTEM_INODE, | |
4377 | le16_to_cpu(dx_root->dr_suballoc_slot)); | |
4378 | if (!dx_alloc_inode) { | |
4379 | ret = -ENOMEM; | |
4380 | mlog_errno(ret); | |
4381 | goto out; | |
4382 | } | |
4383 | mutex_lock(&dx_alloc_inode->i_mutex); | |
4384 | ||
4385 | ret = ocfs2_inode_lock(dx_alloc_inode, &dx_alloc_bh, 1); | |
4386 | if (ret) { | |
4387 | mlog_errno(ret); | |
4388 | goto out_mutex; | |
4389 | } | |
4390 | ||
4391 | handle = ocfs2_start_trans(osb, OCFS2_DX_ROOT_REMOVE_CREDITS); | |
4392 | if (IS_ERR(handle)) { | |
4393 | ret = PTR_ERR(handle); | |
4394 | mlog_errno(ret); | |
4395 | goto out_unlock; | |
4396 | } | |
4397 | ||
4398 | ret = ocfs2_journal_access_di(handle, dir, di_bh, | |
4399 | OCFS2_JOURNAL_ACCESS_WRITE); | |
4400 | if (ret) { | |
4401 | mlog_errno(ret); | |
4402 | goto out_commit; | |
4403 | } | |
4404 | ||
4405 | OCFS2_I(dir)->ip_dyn_features &= ~OCFS2_INDEXED_DIR_FL; | |
4406 | di->i_dyn_features = cpu_to_le16(OCFS2_I(dir)->ip_dyn_features); | |
4407 | di->i_dx_root = cpu_to_le64(0ULL); | |
4408 | ||
4409 | ocfs2_journal_dirty(handle, di_bh); | |
4410 | ||
4411 | blk = le64_to_cpu(dx_root->dr_blkno); | |
4412 | bit = le16_to_cpu(dx_root->dr_suballoc_bit); | |
4413 | bg_blkno = ocfs2_which_suballoc_group(blk, bit); | |
4414 | ret = ocfs2_free_suballoc_bits(handle, dx_alloc_inode, dx_alloc_bh, | |
4415 | bit, bg_blkno, 1); | |
4416 | if (ret) | |
4417 | mlog_errno(ret); | |
4418 | ||
4419 | out_commit: | |
4420 | ocfs2_commit_trans(osb, handle); | |
4421 | ||
4422 | out_unlock: | |
4423 | ocfs2_inode_unlock(dx_alloc_inode, 1); | |
4424 | ||
4425 | out_mutex: | |
4426 | mutex_unlock(&dx_alloc_inode->i_mutex); | |
4427 | brelse(dx_alloc_bh); | |
4428 | out: | |
4429 | iput(dx_alloc_inode); | |
4430 | return ret; | |
4431 | } | |
4432 | ||
4433 | int ocfs2_dx_dir_truncate(struct inode *dir, struct buffer_head *di_bh) | |
4434 | { | |
4435 | int ret; | |
4436 | unsigned int uninitialized_var(clen); | |
4437 | u32 major_hash = UINT_MAX, p_cpos, uninitialized_var(cpos); | |
4438 | u64 uninitialized_var(blkno); | |
4439 | struct ocfs2_super *osb = OCFS2_SB(dir->i_sb); | |
4440 | struct buffer_head *dx_root_bh = NULL; | |
4441 | struct ocfs2_dx_root_block *dx_root; | |
4442 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
4443 | struct ocfs2_cached_dealloc_ctxt dealloc; | |
4444 | struct ocfs2_extent_tree et; | |
4445 | ||
4446 | ocfs2_init_dealloc_ctxt(&dealloc); | |
4447 | ||
4448 | if (!ocfs2_dir_indexed(dir)) | |
4449 | return 0; | |
4450 | ||
4451 | ret = ocfs2_read_dx_root(dir, di, &dx_root_bh); | |
4452 | if (ret) { | |
4453 | mlog_errno(ret); | |
4454 | goto out; | |
4455 | } | |
4ed8a6bb | 4456 | dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data; |
9b7895ef | 4457 | |
4ed8a6bb MF |
4458 | if (ocfs2_dx_root_inline(dx_root)) |
4459 | goto remove_index; | |
9b7895ef | 4460 | |
4ed8a6bb | 4461 | ocfs2_init_dx_root_extent_tree(&et, dir, dx_root_bh); |
9b7895ef MF |
4462 | |
4463 | /* XXX: What if dr_clusters is too large? */ | |
4464 | while (le32_to_cpu(dx_root->dr_clusters)) { | |
4465 | ret = ocfs2_dx_dir_lookup_rec(dir, &dx_root->dr_list, | |
4466 | major_hash, &cpos, &blkno, &clen); | |
4467 | if (ret) { | |
4468 | mlog_errno(ret); | |
4469 | goto out; | |
4470 | } | |
4471 | ||
4472 | p_cpos = ocfs2_blocks_to_clusters(dir->i_sb, blkno); | |
4473 | ||
4474 | ret = ocfs2_remove_btree_range(dir, &et, cpos, p_cpos, clen, | |
4475 | &dealloc); | |
4476 | if (ret) { | |
4477 | mlog_errno(ret); | |
4478 | goto out; | |
4479 | } | |
4480 | ||
4481 | if (cpos == 0) | |
4482 | break; | |
4483 | ||
4484 | major_hash = cpos - 1; | |
4485 | } | |
4486 | ||
4ed8a6bb | 4487 | remove_index: |
9b7895ef MF |
4488 | ret = ocfs2_dx_dir_remove_index(dir, di_bh, dx_root_bh); |
4489 | if (ret) { | |
4490 | mlog_errno(ret); | |
4491 | goto out; | |
4492 | } | |
4493 | ||
4494 | ocfs2_remove_from_cache(dir, dx_root_bh); | |
4495 | out: | |
4496 | ocfs2_schedule_truncate_log_flush(osb, 1); | |
4497 | ocfs2_run_deallocs(osb, &dealloc); | |
4498 | ||
4499 | brelse(dx_root_bh); | |
4500 | return ret; | |
4501 | } |