1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * dir.c
4 *
5 * Creates, reads, walks and deletes directory-nodes
6 *
7 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
8 *
9 * Portions of this code from linux/fs/ext3/dir.c
10 *
11 * Copyright (C) 1992, 1993, 1994, 1995
12 * Remy Card (card@masi.ibp.fr)
13 * Laboratoire MASI - Institut Blaise pascal
14 * Universite Pierre et Marie Curie (Paris VI)
15 *
16 * from
17 *
18 * linux/fs/minix/dir.c
19 *
20 * Copyright (C) 1991, 1992 Linus Torvalds
21 */
22
23 #include <linux/fs.h>
24 #include <linux/types.h>
25 #include <linux/slab.h>
26 #include <linux/highmem.h>
27 #include <linux/quotaops.h>
28 #include <linux/sort.h>
29 #include <linux/iversion.h>
30
31 #include <cluster/masklog.h>
32
33 #include "ocfs2.h"
34
35 #include "alloc.h"
36 #include "blockcheck.h"
37 #include "dir.h"
38 #include "dlmglue.h"
39 #include "extent_map.h"
40 #include "file.h"
41 #include "inode.h"
42 #include "journal.h"
43 #include "namei.h"
44 #include "suballoc.h"
45 #include "super.h"
46 #include "sysfile.h"
47 #include "uptodate.h"
48 #include "ocfs2_trace.h"
49
50 #include "buffer_head_io.h"
51
52 #define NAMEI_RA_CHUNKS 2
53 #define NAMEI_RA_BLOCKS 4
54 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
55
56 static int ocfs2_do_extend_dir(struct super_block *sb,
57 handle_t *handle,
58 struct inode *dir,
59 struct buffer_head *parent_fe_bh,
60 struct ocfs2_alloc_context *data_ac,
61 struct ocfs2_alloc_context *meta_ac,
62 struct buffer_head **new_bh);
63 static int ocfs2_dir_indexed(struct inode *inode);
64
65 /*
66 * These are distinct checks because future versions of the file system will
67 * want to have a trailing dirent structure independent of indexing.
68 */
ocfs2_supports_dir_trailer(struct inode * dir)69 static int ocfs2_supports_dir_trailer(struct inode *dir)
70 {
71 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
72
73 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
74 return 0;
75
76 return ocfs2_meta_ecc(osb) || ocfs2_dir_indexed(dir);
77 }
78
79 /*
80 * "new' here refers to the point at which we're creating a new
81 * directory via "mkdir()", but also when we're expanding an inline
82 * directory. In either case, we don't yet have the indexing bit set
83 * on the directory, so the standard checks will fail in when metaecc
84 * is turned off. Only directory-initialization type functions should
85 * use this then. Everything else wants ocfs2_supports_dir_trailer()
86 */
ocfs2_new_dir_wants_trailer(struct inode * dir)87 static int ocfs2_new_dir_wants_trailer(struct inode *dir)
88 {
89 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
90
91 return ocfs2_meta_ecc(osb) ||
92 ocfs2_supports_indexed_dirs(osb);
93 }
94
ocfs2_dir_trailer_blk_off(struct super_block * sb)95 static inline unsigned int ocfs2_dir_trailer_blk_off(struct super_block *sb)
96 {
97 return sb->s_blocksize - sizeof(struct ocfs2_dir_block_trailer);
98 }
99
100 #define ocfs2_trailer_from_bh(_bh, _sb) ((struct ocfs2_dir_block_trailer *) ((_bh)->b_data + ocfs2_dir_trailer_blk_off((_sb))))
101
102 /* XXX ocfs2_block_dqtrailer() is similar but not quite - can we make
103 * them more consistent? */
ocfs2_dir_trailer_from_size(int blocksize,void * data)104 struct ocfs2_dir_block_trailer *ocfs2_dir_trailer_from_size(int blocksize,
105 void *data)
106 {
107 char *p = data;
108
109 p += blocksize - sizeof(struct ocfs2_dir_block_trailer);
110 return (struct ocfs2_dir_block_trailer *)p;
111 }
112
113 /*
114 * XXX: This is executed once on every dirent. We should consider optimizing
115 * it.
116 */
ocfs2_skip_dir_trailer(struct inode * dir,struct ocfs2_dir_entry * de,unsigned long offset,unsigned long blklen)117 static int ocfs2_skip_dir_trailer(struct inode *dir,
118 struct ocfs2_dir_entry *de,
119 unsigned long offset,
120 unsigned long blklen)
121 {
122 unsigned long toff = blklen - sizeof(struct ocfs2_dir_block_trailer);
123
124 if (!ocfs2_supports_dir_trailer(dir))
125 return 0;
126
127 if (offset != toff)
128 return 0;
129
130 return 1;
131 }
132
ocfs2_init_dir_trailer(struct inode * inode,struct buffer_head * bh,u16 rec_len)133 static void ocfs2_init_dir_trailer(struct inode *inode,
134 struct buffer_head *bh, u16 rec_len)
135 {
136 struct ocfs2_dir_block_trailer *trailer;
137
138 trailer = ocfs2_trailer_from_bh(bh, inode->i_sb);
139 strcpy(trailer->db_signature, OCFS2_DIR_TRAILER_SIGNATURE);
140 trailer->db_compat_rec_len =
141 cpu_to_le16(sizeof(struct ocfs2_dir_block_trailer));
142 trailer->db_parent_dinode = cpu_to_le64(OCFS2_I(inode)->ip_blkno);
143 trailer->db_blkno = cpu_to_le64(bh->b_blocknr);
144 trailer->db_free_rec_len = cpu_to_le16(rec_len);
145 }
146 /*
147 * Link an unindexed block with a dir trailer structure into the index free
148 * list. This function will modify dirdata_bh, but assumes you've already
149 * passed it to the journal.
150 */
ocfs2_dx_dir_link_trailer(struct inode * dir,handle_t * handle,struct buffer_head * dx_root_bh,struct buffer_head * dirdata_bh)151 static int ocfs2_dx_dir_link_trailer(struct inode *dir, handle_t *handle,
152 struct buffer_head *dx_root_bh,
153 struct buffer_head *dirdata_bh)
154 {
155 int ret;
156 struct ocfs2_dx_root_block *dx_root;
157 struct ocfs2_dir_block_trailer *trailer;
158
159 ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
160 OCFS2_JOURNAL_ACCESS_WRITE);
161 if (ret) {
162 mlog_errno(ret);
163 goto out;
164 }
165 trailer = ocfs2_trailer_from_bh(dirdata_bh, dir->i_sb);
166 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
167
168 trailer->db_free_next = dx_root->dr_free_blk;
169 dx_root->dr_free_blk = cpu_to_le64(dirdata_bh->b_blocknr);
170
171 ocfs2_journal_dirty(handle, dx_root_bh);
172
173 out:
174 return ret;
175 }
176
ocfs2_free_list_at_root(struct ocfs2_dir_lookup_result * res)177 static int ocfs2_free_list_at_root(struct ocfs2_dir_lookup_result *res)
178 {
179 return res->dl_prev_leaf_bh == NULL;
180 }
181
ocfs2_free_dir_lookup_result(struct ocfs2_dir_lookup_result * res)182 void ocfs2_free_dir_lookup_result(struct ocfs2_dir_lookup_result *res)
183 {
184 brelse(res->dl_dx_root_bh);
185 brelse(res->dl_leaf_bh);
186 brelse(res->dl_dx_leaf_bh);
187 brelse(res->dl_prev_leaf_bh);
188 }
189
ocfs2_dir_indexed(struct inode * inode)190 static int ocfs2_dir_indexed(struct inode *inode)
191 {
192 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INDEXED_DIR_FL)
193 return 1;
194 return 0;
195 }
196
ocfs2_dx_root_inline(struct ocfs2_dx_root_block * dx_root)197 static inline int ocfs2_dx_root_inline(struct ocfs2_dx_root_block *dx_root)
198 {
199 return dx_root->dr_flags & OCFS2_DX_FLAG_INLINE;
200 }
201
202 /*
203 * Hashing code adapted from ext3
204 */
205 #define DELTA 0x9E3779B9
206
TEA_transform(__u32 buf[4],__u32 const in[])207 static void TEA_transform(__u32 buf[4], __u32 const in[])
208 {
209 __u32 sum = 0;
210 __u32 b0 = buf[0], b1 = buf[1];
211 __u32 a = in[0], b = in[1], c = in[2], d = in[3];
212 int n = 16;
213
214 do {
215 sum += DELTA;
216 b0 += ((b1 << 4)+a) ^ (b1+sum) ^ ((b1 >> 5)+b);
217 b1 += ((b0 << 4)+c) ^ (b0+sum) ^ ((b0 >> 5)+d);
218 } while (--n);
219
220 buf[0] += b0;
221 buf[1] += b1;
222 }
223
str2hashbuf(const char * msg,int len,__u32 * buf,int num)224 static void str2hashbuf(const char *msg, int len, __u32 *buf, int num)
225 {
226 __u32 pad, val;
227 int i;
228
229 pad = (__u32)len | ((__u32)len << 8);
230 pad |= pad << 16;
231
232 val = pad;
233 if (len > num*4)
234 len = num * 4;
235 for (i = 0; i < len; i++) {
236 if ((i % 4) == 0)
237 val = pad;
238 val = msg[i] + (val << 8);
239 if ((i % 4) == 3) {
240 *buf++ = val;
241 val = pad;
242 num--;
243 }
244 }
245 if (--num >= 0)
246 *buf++ = val;
247 while (--num >= 0)
248 *buf++ = pad;
249 }
250
ocfs2_dx_dir_name_hash(struct inode * dir,const char * name,int len,struct ocfs2_dx_hinfo * hinfo)251 static void ocfs2_dx_dir_name_hash(struct inode *dir, const char *name, int len,
252 struct ocfs2_dx_hinfo *hinfo)
253 {
254 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
255 const char *p;
256 __u32 in[8], buf[4];
257
258 /*
259 * XXX: Is this really necessary, if the index is never looked
260 * at by readdir? Is a hash value of '0' a bad idea?
261 */
262 if ((len == 1 && !strncmp(".", name, 1)) ||
263 (len == 2 && !strncmp("..", name, 2))) {
264 buf[0] = buf[1] = 0;
265 goto out;
266 }
267
268 #ifdef OCFS2_DEBUG_DX_DIRS
269 /*
270 * This makes it very easy to debug indexing problems. We
271 * should never allow this to be selected without hand editing
272 * this file though.
273 */
274 buf[0] = buf[1] = len;
275 goto out;
276 #endif
277
278 memcpy(buf, osb->osb_dx_seed, sizeof(buf));
279
280 p = name;
281 while (len > 0) {
282 str2hashbuf(p, len, in, 4);
283 TEA_transform(buf, in);
284 len -= 16;
285 p += 16;
286 }
287
288 out:
289 hinfo->major_hash = buf[0];
290 hinfo->minor_hash = buf[1];
291 }
292
293 /*
294 * bh passed here can be an inode block or a dir data block, depending
295 * on the inode inline data flag.
296 */
ocfs2_check_dir_entry(struct inode * dir,struct ocfs2_dir_entry * de,struct buffer_head * bh,char * buf,unsigned int size,unsigned long offset)297 static int ocfs2_check_dir_entry(struct inode *dir,
298 struct ocfs2_dir_entry *de,
299 struct buffer_head *bh,
300 char *buf,
301 unsigned int size,
302 unsigned long offset)
303 {
304 const char *error_msg = NULL;
305 const int rlen = le16_to_cpu(de->rec_len);
306 const unsigned long next_offset = ((char *) de - buf) + rlen;
307
308 if (unlikely(rlen < OCFS2_DIR_REC_LEN(1)))
309 error_msg = "rec_len is smaller than minimal";
310 else if (unlikely(rlen % 4 != 0))
311 error_msg = "rec_len % 4 != 0";
312 else if (unlikely(rlen < OCFS2_DIR_REC_LEN(de->name_len)))
313 error_msg = "rec_len is too small for name_len";
314 else if (unlikely(next_offset > size))
315 error_msg = "directory entry overrun";
316 else if (unlikely(next_offset > size - OCFS2_DIR_REC_LEN(1)) &&
317 next_offset != size)
318 error_msg = "directory entry too close to end";
319
320 if (unlikely(error_msg != NULL))
321 mlog(ML_ERROR, "bad entry in directory #%llu: %s - "
322 "offset=%lu, inode=%llu, rec_len=%d, name_len=%d\n",
323 (unsigned long long)OCFS2_I(dir)->ip_blkno, error_msg,
324 offset, (unsigned long long)le64_to_cpu(de->inode), rlen,
325 de->name_len);
326
327 return error_msg == NULL ? 1 : 0;
328 }
329
ocfs2_match(int len,const char * const name,struct ocfs2_dir_entry * de)330 static inline int ocfs2_match(int len,
331 const char * const name,
332 struct ocfs2_dir_entry *de)
333 {
334 if (len != de->name_len)
335 return 0;
336 if (!de->inode)
337 return 0;
338 return !memcmp(name, de->name, len);
339 }
340
341 /*
342 * Returns 0 if not found, -1 on failure, and 1 on success
343 */
ocfs2_search_dirblock(struct buffer_head * bh,struct inode * dir,const char * name,int namelen,unsigned long offset,char * first_de,unsigned int bytes,struct ocfs2_dir_entry ** res_dir)344 static inline int ocfs2_search_dirblock(struct buffer_head *bh,
345 struct inode *dir,
346 const char *name, int namelen,
347 unsigned long offset,
348 char *first_de,
349 unsigned int bytes,
350 struct ocfs2_dir_entry **res_dir)
351 {
352 struct ocfs2_dir_entry *de;
353 char *dlimit, *de_buf;
354 int de_len;
355 int ret = 0;
356
357 de_buf = first_de;
358 dlimit = de_buf + bytes;
359
360 while (de_buf < dlimit - OCFS2_DIR_MEMBER_LEN) {
361 /* this code is executed quadratically often */
362 /* do minimal checking `by hand' */
363
364 de = (struct ocfs2_dir_entry *) de_buf;
365
366 if (de->name + namelen <= dlimit &&
367 ocfs2_match(namelen, name, de)) {
368 /* found a match - just to be sure, do a full check */
369 if (!ocfs2_check_dir_entry(dir, de, bh, first_de,
370 bytes, offset)) {
371 ret = -1;
372 goto bail;
373 }
374 *res_dir = de;
375 ret = 1;
376 goto bail;
377 }
378
379 /* prevent looping on a bad block */
380 de_len = le16_to_cpu(de->rec_len);
381 if (de_len <= 0) {
382 ret = -1;
383 goto bail;
384 }
385
386 de_buf += de_len;
387 offset += de_len;
388 }
389
390 bail:
391 trace_ocfs2_search_dirblock(ret);
392 return ret;
393 }
394
ocfs2_find_entry_id(const char * name,int namelen,struct inode * dir,struct ocfs2_dir_entry ** res_dir)395 static struct buffer_head *ocfs2_find_entry_id(const char *name,
396 int namelen,
397 struct inode *dir,
398 struct ocfs2_dir_entry **res_dir)
399 {
400 int ret, found;
401 struct buffer_head *di_bh = NULL;
402 struct ocfs2_dinode *di;
403 struct ocfs2_inline_data *data;
404
405 ret = ocfs2_read_inode_block(dir, &di_bh);
406 if (ret) {
407 mlog_errno(ret);
408 goto out;
409 }
410
411 di = (struct ocfs2_dinode *)di_bh->b_data;
412 data = &di->id2.i_data;
413
414 found = ocfs2_search_dirblock(di_bh, dir, name, namelen, 0,
415 data->id_data, i_size_read(dir), res_dir);
416 if (found == 1)
417 return di_bh;
418
419 brelse(di_bh);
420 out:
421 return NULL;
422 }
423
ocfs2_validate_dir_block(struct super_block * sb,struct buffer_head * bh)424 static int ocfs2_validate_dir_block(struct super_block *sb,
425 struct buffer_head *bh)
426 {
427 int rc;
428 struct ocfs2_dir_block_trailer *trailer =
429 ocfs2_trailer_from_bh(bh, sb);
430
431
432 /*
433 * We don't validate dirents here, that's handled
434 * in-place when the code walks them.
435 */
436 trace_ocfs2_validate_dir_block((unsigned long long)bh->b_blocknr);
437
438 BUG_ON(!buffer_uptodate(bh));
439
440 /*
441 * If the ecc fails, we return the error but otherwise
442 * leave the filesystem running. We know any error is
443 * local to this block.
444 *
445 * Note that we are safe to call this even if the directory
446 * doesn't have a trailer. Filesystems without metaecc will do
447 * nothing, and filesystems with it will have one.
448 */
449 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &trailer->db_check);
450 if (rc)
451 mlog(ML_ERROR, "Checksum failed for dinode %llu\n",
452 (unsigned long long)bh->b_blocknr);
453
454 return rc;
455 }
456
457 /*
458 * Validate a directory trailer.
459 *
460 * We check the trailer here rather than in ocfs2_validate_dir_block()
461 * because that function doesn't have the inode to test.
462 */
ocfs2_check_dir_trailer(struct inode * dir,struct buffer_head * bh)463 static int ocfs2_check_dir_trailer(struct inode *dir, struct buffer_head *bh)
464 {
465 int rc = 0;
466 struct ocfs2_dir_block_trailer *trailer;
467
468 trailer = ocfs2_trailer_from_bh(bh, dir->i_sb);
469 if (!OCFS2_IS_VALID_DIR_TRAILER(trailer)) {
470 rc = ocfs2_error(dir->i_sb,
471 "Invalid dirblock #%llu: signature = %.*s\n",
472 (unsigned long long)bh->b_blocknr, 7,
473 trailer->db_signature);
474 goto out;
475 }
476 if (le64_to_cpu(trailer->db_blkno) != bh->b_blocknr) {
477 rc = ocfs2_error(dir->i_sb,
478 "Directory block #%llu has an invalid db_blkno of %llu\n",
479 (unsigned long long)bh->b_blocknr,
480 (unsigned long long)le64_to_cpu(trailer->db_blkno));
481 goto out;
482 }
483 if (le64_to_cpu(trailer->db_parent_dinode) !=
484 OCFS2_I(dir)->ip_blkno) {
485 rc = ocfs2_error(dir->i_sb,
486 "Directory block #%llu on dinode #%llu has an invalid parent_dinode of %llu\n",
487 (unsigned long long)bh->b_blocknr,
488 (unsigned long long)OCFS2_I(dir)->ip_blkno,
489 (unsigned long long)le64_to_cpu(trailer->db_blkno));
490 goto out;
491 }
492 out:
493 return rc;
494 }
495
496 /*
497 * This function forces all errors to -EIO for consistency with its
498 * predecessor, ocfs2_bread(). We haven't audited what returning the
499 * real error codes would do to callers. We log the real codes with
500 * mlog_errno() before we squash them.
501 */
ocfs2_read_dir_block(struct inode * inode,u64 v_block,struct buffer_head ** bh,int flags)502 static int ocfs2_read_dir_block(struct inode *inode, u64 v_block,
503 struct buffer_head **bh, int flags)
504 {
505 int rc = 0;
506 struct buffer_head *tmp = *bh;
507
508 rc = ocfs2_read_virt_blocks(inode, v_block, 1, &tmp, flags,
509 ocfs2_validate_dir_block);
510 if (rc) {
511 mlog_errno(rc);
512 goto out;
513 }
514
515 if (!(flags & OCFS2_BH_READAHEAD) &&
516 ocfs2_supports_dir_trailer(inode)) {
517 rc = ocfs2_check_dir_trailer(inode, tmp);
518 if (rc) {
519 if (!*bh)
520 brelse(tmp);
521 mlog_errno(rc);
522 goto out;
523 }
524 }
525
526 /* If ocfs2_read_virt_blocks() got us a new bh, pass it up. */
527 if (!*bh)
528 *bh = tmp;
529
530 out:
531 return rc ? -EIO : 0;
532 }
533
534 /*
535 * Read the block at 'phys' which belongs to this directory
536 * inode. This function does no virtual->physical block translation -
537 * what's passed in is assumed to be a valid directory block.
538 */
ocfs2_read_dir_block_direct(struct inode * dir,u64 phys,struct buffer_head ** bh)539 static int ocfs2_read_dir_block_direct(struct inode *dir, u64 phys,
540 struct buffer_head **bh)
541 {
542 int ret;
543 struct buffer_head *tmp = *bh;
544
545 ret = ocfs2_read_block(INODE_CACHE(dir), phys, &tmp,
546 ocfs2_validate_dir_block);
547 if (ret) {
548 mlog_errno(ret);
549 goto out;
550 }
551
552 if (ocfs2_supports_dir_trailer(dir)) {
553 ret = ocfs2_check_dir_trailer(dir, tmp);
554 if (ret) {
555 if (!*bh)
556 brelse(tmp);
557 mlog_errno(ret);
558 goto out;
559 }
560 }
561
562 if (!ret && !*bh)
563 *bh = tmp;
564 out:
565 return ret;
566 }
567
ocfs2_validate_dx_root(struct super_block * sb,struct buffer_head * bh)568 static int ocfs2_validate_dx_root(struct super_block *sb,
569 struct buffer_head *bh)
570 {
571 int ret;
572 struct ocfs2_dx_root_block *dx_root;
573
574 BUG_ON(!buffer_uptodate(bh));
575
576 dx_root = (struct ocfs2_dx_root_block *) bh->b_data;
577
578 ret = ocfs2_validate_meta_ecc(sb, bh->b_data, &dx_root->dr_check);
579 if (ret) {
580 mlog(ML_ERROR,
581 "Checksum failed for dir index root block %llu\n",
582 (unsigned long long)bh->b_blocknr);
583 return ret;
584 }
585
586 if (!OCFS2_IS_VALID_DX_ROOT(dx_root)) {
587 ret = ocfs2_error(sb,
588 "Dir Index Root # %llu has bad signature %.*s\n",
589 (unsigned long long)le64_to_cpu(dx_root->dr_blkno),
590 7, dx_root->dr_signature);
591 }
592
593 return ret;
594 }
595
ocfs2_read_dx_root(struct inode * dir,struct ocfs2_dinode * di,struct buffer_head ** dx_root_bh)596 static int ocfs2_read_dx_root(struct inode *dir, struct ocfs2_dinode *di,
597 struct buffer_head **dx_root_bh)
598 {
599 int ret;
600 u64 blkno = le64_to_cpu(di->i_dx_root);
601 struct buffer_head *tmp = *dx_root_bh;
602
603 ret = ocfs2_read_block(INODE_CACHE(dir), blkno, &tmp,
604 ocfs2_validate_dx_root);
605
606 /* If ocfs2_read_block() got us a new bh, pass it up. */
607 if (!ret && !*dx_root_bh)
608 *dx_root_bh = tmp;
609
610 return ret;
611 }
612
ocfs2_validate_dx_leaf(struct super_block * sb,struct buffer_head * bh)613 static int ocfs2_validate_dx_leaf(struct super_block *sb,
614 struct buffer_head *bh)
615 {
616 int ret;
617 struct ocfs2_dx_leaf *dx_leaf = (struct ocfs2_dx_leaf *)bh->b_data;
618
619 BUG_ON(!buffer_uptodate(bh));
620
621 ret = ocfs2_validate_meta_ecc(sb, bh->b_data, &dx_leaf->dl_check);
622 if (ret) {
623 mlog(ML_ERROR,
624 "Checksum failed for dir index leaf block %llu\n",
625 (unsigned long long)bh->b_blocknr);
626 return ret;
627 }
628
629 if (!OCFS2_IS_VALID_DX_LEAF(dx_leaf)) {
630 ret = ocfs2_error(sb, "Dir Index Leaf has bad signature %.*s\n",
631 7, dx_leaf->dl_signature);
632 }
633
634 return ret;
635 }
636
ocfs2_read_dx_leaf(struct inode * dir,u64 blkno,struct buffer_head ** dx_leaf_bh)637 static int ocfs2_read_dx_leaf(struct inode *dir, u64 blkno,
638 struct buffer_head **dx_leaf_bh)
639 {
640 int ret;
641 struct buffer_head *tmp = *dx_leaf_bh;
642
643 ret = ocfs2_read_block(INODE_CACHE(dir), blkno, &tmp,
644 ocfs2_validate_dx_leaf);
645
646 /* If ocfs2_read_block() got us a new bh, pass it up. */
647 if (!ret && !*dx_leaf_bh)
648 *dx_leaf_bh = tmp;
649
650 return ret;
651 }
652
653 /*
654 * Read a series of dx_leaf blocks. This expects all buffer_head
655 * pointers to be NULL on function entry.
656 */
ocfs2_read_dx_leaves(struct inode * dir,u64 start,int num,struct buffer_head ** dx_leaf_bhs)657 static int ocfs2_read_dx_leaves(struct inode *dir, u64 start, int num,
658 struct buffer_head **dx_leaf_bhs)
659 {
660 int ret;
661
662 ret = ocfs2_read_blocks(INODE_CACHE(dir), start, num, dx_leaf_bhs, 0,
663 ocfs2_validate_dx_leaf);
664 if (ret)
665 mlog_errno(ret);
666
667 return ret;
668 }
669
ocfs2_find_entry_el(const char * name,int namelen,struct inode * dir,struct ocfs2_dir_entry ** res_dir)670 static struct buffer_head *ocfs2_find_entry_el(const char *name, int namelen,
671 struct inode *dir,
672 struct ocfs2_dir_entry **res_dir)
673 {
674 struct super_block *sb;
675 struct buffer_head *bh_use[NAMEI_RA_SIZE];
676 struct buffer_head *bh, *ret = NULL;
677 unsigned long start, block, b;
678 int ra_max = 0; /* Number of bh's in the readahead
679 buffer, bh_use[] */
680 int ra_ptr = 0; /* Current index into readahead
681 buffer */
682 int num = 0;
683 int nblocks, i;
684
685 sb = dir->i_sb;
686
687 nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
688 start = OCFS2_I(dir)->ip_dir_start_lookup;
689 if (start >= nblocks)
690 start = 0;
691 block = start;
692
693 restart:
694 do {
695 /*
696 * We deal with the read-ahead logic here.
697 */
698 if (ra_ptr >= ra_max) {
699 /* Refill the readahead buffer */
700 ra_ptr = 0;
701 b = block;
702 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
703 /*
704 * Terminate if we reach the end of the
705 * directory and must wrap, or if our
706 * search has finished at this block.
707 */
708 if (b >= nblocks || (num && block == start)) {
709 bh_use[ra_max] = NULL;
710 break;
711 }
712 num++;
713
714 bh = NULL;
715 ocfs2_read_dir_block(dir, b++, &bh,
716 OCFS2_BH_READAHEAD);
717 bh_use[ra_max] = bh;
718 }
719 }
720 if ((bh = bh_use[ra_ptr++]) == NULL)
721 goto next;
722 if (ocfs2_read_dir_block(dir, block, &bh, 0)) {
723 /* read error, skip block & hope for the best.
724 * ocfs2_read_dir_block() has released the bh. */
725 mlog(ML_ERROR, "reading directory %llu, "
726 "offset %lu\n",
727 (unsigned long long)OCFS2_I(dir)->ip_blkno,
728 block);
729 goto next;
730 }
731 i = ocfs2_search_dirblock(bh, dir, name, namelen,
732 block << sb->s_blocksize_bits,
733 bh->b_data, sb->s_blocksize,
734 res_dir);
735 if (i == 1) {
736 OCFS2_I(dir)->ip_dir_start_lookup = block;
737 ret = bh;
738 goto cleanup_and_exit;
739 } else {
740 brelse(bh);
741 if (i < 0)
742 goto cleanup_and_exit;
743 }
744 next:
745 if (++block >= nblocks)
746 block = 0;
747 } while (block != start);
748
749 /*
750 * If the directory has grown while we were searching, then
751 * search the last part of the directory before giving up.
752 */
753 block = nblocks;
754 nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
755 if (block < nblocks) {
756 start = 0;
757 goto restart;
758 }
759
760 cleanup_and_exit:
761 /* Clean up the read-ahead blocks */
762 for (; ra_ptr < ra_max; ra_ptr++)
763 brelse(bh_use[ra_ptr]);
764
765 trace_ocfs2_find_entry_el(ret);
766 return ret;
767 }
768
ocfs2_dx_dir_lookup_rec(struct inode * inode,struct ocfs2_extent_list * el,u32 major_hash,u32 * ret_cpos,u64 * ret_phys_blkno,unsigned int * ret_clen)769 static int ocfs2_dx_dir_lookup_rec(struct inode *inode,
770 struct ocfs2_extent_list *el,
771 u32 major_hash,
772 u32 *ret_cpos,
773 u64 *ret_phys_blkno,
774 unsigned int *ret_clen)
775 {
776 int ret = 0, i, found;
777 struct buffer_head *eb_bh = NULL;
778 struct ocfs2_extent_block *eb;
779 struct ocfs2_extent_rec *rec = NULL;
780
781 if (el->l_tree_depth) {
782 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, major_hash,
783 &eb_bh);
784 if (ret) {
785 mlog_errno(ret);
786 goto out;
787 }
788
789 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
790 el = &eb->h_list;
791
792 if (el->l_tree_depth) {
793 ret = ocfs2_error(inode->i_sb,
794 "Inode %lu has non zero tree depth in btree tree block %llu\n",
795 inode->i_ino,
796 (unsigned long long)eb_bh->b_blocknr);
797 goto out;
798 }
799 }
800
801 found = 0;
802 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
803 rec = &el->l_recs[i];
804
805 if (le32_to_cpu(rec->e_cpos) <= major_hash) {
806 found = 1;
807 break;
808 }
809 }
810
811 if (!found) {
812 ret = ocfs2_error(inode->i_sb,
813 "Inode %lu has bad extent record (%u, %u, 0) in btree\n",
814 inode->i_ino,
815 le32_to_cpu(rec->e_cpos),
816 ocfs2_rec_clusters(el, rec));
817 goto out;
818 }
819
820 if (ret_phys_blkno)
821 *ret_phys_blkno = le64_to_cpu(rec->e_blkno);
822 if (ret_cpos)
823 *ret_cpos = le32_to_cpu(rec->e_cpos);
824 if (ret_clen)
825 *ret_clen = le16_to_cpu(rec->e_leaf_clusters);
826
827 out:
828 brelse(eb_bh);
829 return ret;
830 }
831
832 /*
833 * Returns the block index, from the start of the cluster which this
834 * hash belongs too.
835 */
__ocfs2_dx_dir_hash_idx(struct ocfs2_super * osb,u32 minor_hash)836 static inline unsigned int __ocfs2_dx_dir_hash_idx(struct ocfs2_super *osb,
837 u32 minor_hash)
838 {
839 return minor_hash & osb->osb_dx_mask;
840 }
841
ocfs2_dx_dir_hash_idx(struct ocfs2_super * osb,struct ocfs2_dx_hinfo * hinfo)842 static inline unsigned int ocfs2_dx_dir_hash_idx(struct ocfs2_super *osb,
843 struct ocfs2_dx_hinfo *hinfo)
844 {
845 return __ocfs2_dx_dir_hash_idx(osb, hinfo->minor_hash);
846 }
847
ocfs2_dx_dir_lookup(struct inode * inode,struct ocfs2_extent_list * el,struct ocfs2_dx_hinfo * hinfo,u32 * ret_cpos,u64 * ret_phys_blkno)848 static int ocfs2_dx_dir_lookup(struct inode *inode,
849 struct ocfs2_extent_list *el,
850 struct ocfs2_dx_hinfo *hinfo,
851 u32 *ret_cpos,
852 u64 *ret_phys_blkno)
853 {
854 int ret = 0;
855 unsigned int cend, clen;
856 u32 cpos;
857 u64 blkno;
858 u32 name_hash = hinfo->major_hash;
859
860 ret = ocfs2_dx_dir_lookup_rec(inode, el, name_hash, &cpos, &blkno,
861 &clen);
862 if (ret) {
863 mlog_errno(ret);
864 goto out;
865 }
866
867 cend = cpos + clen;
868 if (name_hash >= cend) {
869 /* We want the last cluster */
870 blkno += ocfs2_clusters_to_blocks(inode->i_sb, clen - 1);
871 cpos += clen - 1;
872 } else {
873 blkno += ocfs2_clusters_to_blocks(inode->i_sb,
874 name_hash - cpos);
875 cpos = name_hash;
876 }
877
878 /*
879 * We now have the cluster which should hold our entry. To
880 * find the exact block from the start of the cluster to
881 * search, we take the lower bits of the hash.
882 */
883 blkno += ocfs2_dx_dir_hash_idx(OCFS2_SB(inode->i_sb), hinfo);
884
885 if (ret_phys_blkno)
886 *ret_phys_blkno = blkno;
887 if (ret_cpos)
888 *ret_cpos = cpos;
889
890 out:
891
892 return ret;
893 }
894
ocfs2_dx_dir_search(const char * name,int namelen,struct inode * dir,struct ocfs2_dx_root_block * dx_root,struct ocfs2_dir_lookup_result * res)895 static int ocfs2_dx_dir_search(const char *name, int namelen,
896 struct inode *dir,
897 struct ocfs2_dx_root_block *dx_root,
898 struct ocfs2_dir_lookup_result *res)
899 {
900 int ret, i, found;
901 u64 phys;
902 struct buffer_head *dx_leaf_bh = NULL;
903 struct ocfs2_dx_leaf *dx_leaf;
904 struct ocfs2_dx_entry *dx_entry = NULL;
905 struct buffer_head *dir_ent_bh = NULL;
906 struct ocfs2_dir_entry *dir_ent = NULL;
907 struct ocfs2_dx_hinfo *hinfo = &res->dl_hinfo;
908 struct ocfs2_extent_list *dr_el;
909 struct ocfs2_dx_entry_list *entry_list;
910
911 ocfs2_dx_dir_name_hash(dir, name, namelen, &res->dl_hinfo);
912
913 if (ocfs2_dx_root_inline(dx_root)) {
914 entry_list = &dx_root->dr_entries;
915 goto search;
916 }
917
918 dr_el = &dx_root->dr_list;
919
920 ret = ocfs2_dx_dir_lookup(dir, dr_el, hinfo, NULL, &phys);
921 if (ret) {
922 mlog_errno(ret);
923 goto out;
924 }
925
926 trace_ocfs2_dx_dir_search((unsigned long long)OCFS2_I(dir)->ip_blkno,
927 namelen, name, hinfo->major_hash,
928 hinfo->minor_hash, (unsigned long long)phys);
929
930 ret = ocfs2_read_dx_leaf(dir, phys, &dx_leaf_bh);
931 if (ret) {
932 mlog_errno(ret);
933 goto out;
934 }
935
936 dx_leaf = (struct ocfs2_dx_leaf *) dx_leaf_bh->b_data;
937
938 trace_ocfs2_dx_dir_search_leaf_info(
939 le16_to_cpu(dx_leaf->dl_list.de_num_used),
940 le16_to_cpu(dx_leaf->dl_list.de_count));
941
942 entry_list = &dx_leaf->dl_list;
943
944 search:
945 /*
946 * Empty leaf is legal, so no need to check for that.
947 */
948 found = 0;
949 for (i = 0; i < le16_to_cpu(entry_list->de_num_used); i++) {
950 dx_entry = &entry_list->de_entries[i];
951
952 if (hinfo->major_hash != le32_to_cpu(dx_entry->dx_major_hash)
953 || hinfo->minor_hash != le32_to_cpu(dx_entry->dx_minor_hash))
954 continue;
955
956 /*
957 * Search unindexed leaf block now. We're not
958 * guaranteed to find anything.
959 */
960 ret = ocfs2_read_dir_block_direct(dir,
961 le64_to_cpu(dx_entry->dx_dirent_blk),
962 &dir_ent_bh);
963 if (ret) {
964 mlog_errno(ret);
965 goto out;
966 }
967
968 /*
969 * XXX: We should check the unindexed block here,
970 * before using it.
971 */
972
973 found = ocfs2_search_dirblock(dir_ent_bh, dir, name, namelen,
974 0, dir_ent_bh->b_data,
975 dir->i_sb->s_blocksize, &dir_ent);
976 if (found == 1)
977 break;
978
979 if (found == -1) {
980 /* This means we found a bad directory entry. */
981 ret = -EIO;
982 mlog_errno(ret);
983 goto out;
984 }
985
986 brelse(dir_ent_bh);
987 dir_ent_bh = NULL;
988 }
989
990 if (found <= 0) {
991 ret = -ENOENT;
992 goto out;
993 }
994
995 res->dl_leaf_bh = dir_ent_bh;
996 res->dl_entry = dir_ent;
997 res->dl_dx_leaf_bh = dx_leaf_bh;
998 res->dl_dx_entry = dx_entry;
999
1000 ret = 0;
1001 out:
1002 if (ret) {
1003 brelse(dx_leaf_bh);
1004 brelse(dir_ent_bh);
1005 }
1006 return ret;
1007 }
1008
ocfs2_find_entry_dx(const char * name,int namelen,struct inode * dir,struct ocfs2_dir_lookup_result * lookup)1009 static int ocfs2_find_entry_dx(const char *name, int namelen,
1010 struct inode *dir,
1011 struct ocfs2_dir_lookup_result *lookup)
1012 {
1013 int ret;
1014 struct buffer_head *di_bh = NULL;
1015 struct ocfs2_dinode *di;
1016 struct buffer_head *dx_root_bh = NULL;
1017 struct ocfs2_dx_root_block *dx_root;
1018
1019 ret = ocfs2_read_inode_block(dir, &di_bh);
1020 if (ret) {
1021 mlog_errno(ret);
1022 goto out;
1023 }
1024
1025 di = (struct ocfs2_dinode *)di_bh->b_data;
1026
1027 ret = ocfs2_read_dx_root(dir, di, &dx_root_bh);
1028 if (ret) {
1029 mlog_errno(ret);
1030 goto out;
1031 }
1032 dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data;
1033
1034 ret = ocfs2_dx_dir_search(name, namelen, dir, dx_root, lookup);
1035 if (ret) {
1036 if (ret != -ENOENT)
1037 mlog_errno(ret);
1038 goto out;
1039 }
1040
1041 lookup->dl_dx_root_bh = dx_root_bh;
1042 dx_root_bh = NULL;
1043 out:
1044 brelse(di_bh);
1045 brelse(dx_root_bh);
1046 return ret;
1047 }
1048
1049 /*
1050 * Try to find an entry of the provided name within 'dir'.
1051 *
1052 * If nothing was found, -ENOENT is returned. Otherwise, zero is
1053 * returned and the struct 'res' will contain information useful to
1054 * other directory manipulation functions.
1055 *
1056 * Caller can NOT assume anything about the contents of the
1057 * buffer_heads - they are passed back only so that it can be passed
1058 * into any one of the manipulation functions (add entry, delete
1059 * entry, etc). As an example, bh in the extent directory case is a
1060 * data block, in the inline-data case it actually points to an inode,
1061 * in the indexed directory case, multiple buffers are involved.
1062 */
ocfs2_find_entry(const char * name,int namelen,struct inode * dir,struct ocfs2_dir_lookup_result * lookup)1063 int ocfs2_find_entry(const char *name, int namelen,
1064 struct inode *dir, struct ocfs2_dir_lookup_result *lookup)
1065 {
1066 struct buffer_head *bh;
1067 struct ocfs2_dir_entry *res_dir = NULL;
1068
1069 if (ocfs2_dir_indexed(dir))
1070 return ocfs2_find_entry_dx(name, namelen, dir, lookup);
1071
1072 /*
1073 * The unindexed dir code only uses part of the lookup
1074 * structure, so there's no reason to push it down further
1075 * than this.
1076 */
1077 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
1078 bh = ocfs2_find_entry_id(name, namelen, dir, &res_dir);
1079 else
1080 bh = ocfs2_find_entry_el(name, namelen, dir, &res_dir);
1081
1082 if (bh == NULL)
1083 return -ENOENT;
1084
1085 lookup->dl_leaf_bh = bh;
1086 lookup->dl_entry = res_dir;
1087 return 0;
1088 }
1089
1090 /*
1091 * Update inode number and type of a previously found directory entry.
1092 */
ocfs2_update_entry(struct inode * dir,handle_t * handle,struct ocfs2_dir_lookup_result * res,struct inode * new_entry_inode)1093 int ocfs2_update_entry(struct inode *dir, handle_t *handle,
1094 struct ocfs2_dir_lookup_result *res,
1095 struct inode *new_entry_inode)
1096 {
1097 int ret;
1098 ocfs2_journal_access_func access = ocfs2_journal_access_db;
1099 struct ocfs2_dir_entry *de = res->dl_entry;
1100 struct buffer_head *de_bh = res->dl_leaf_bh;
1101
1102 /*
1103 * The same code works fine for both inline-data and extent
1104 * based directories, so no need to split this up. The only
1105 * difference is the journal_access function.
1106 */
1107
1108 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
1109 access = ocfs2_journal_access_di;
1110
1111 ret = access(handle, INODE_CACHE(dir), de_bh,
1112 OCFS2_JOURNAL_ACCESS_WRITE);
1113 if (ret) {
1114 mlog_errno(ret);
1115 goto out;
1116 }
1117
1118 de->inode = cpu_to_le64(OCFS2_I(new_entry_inode)->ip_blkno);
1119 ocfs2_set_de_type(de, new_entry_inode->i_mode);
1120
1121 ocfs2_journal_dirty(handle, de_bh);
1122
1123 out:
1124 return ret;
1125 }
1126
1127 /*
1128 * __ocfs2_delete_entry deletes a directory entry by merging it with the
1129 * previous entry
1130 */
__ocfs2_delete_entry(handle_t * handle,struct inode * dir,struct ocfs2_dir_entry * de_del,struct buffer_head * bh,char * first_de,unsigned int bytes)1131 static int __ocfs2_delete_entry(handle_t *handle, struct inode *dir,
1132 struct ocfs2_dir_entry *de_del,
1133 struct buffer_head *bh, char *first_de,
1134 unsigned int bytes)
1135 {
1136 struct ocfs2_dir_entry *de, *pde;
1137 int i, status = -ENOENT;
1138 ocfs2_journal_access_func access = ocfs2_journal_access_db;
1139
1140 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
1141 access = ocfs2_journal_access_di;
1142
1143 i = 0;
1144 pde = NULL;
1145 de = (struct ocfs2_dir_entry *) first_de;
1146 while (i < bytes) {
1147 if (!ocfs2_check_dir_entry(dir, de, bh, first_de, bytes, i)) {
1148 status = -EIO;
1149 mlog_errno(status);
1150 goto bail;
1151 }
1152 if (de == de_del) {
1153 status = access(handle, INODE_CACHE(dir), bh,
1154 OCFS2_JOURNAL_ACCESS_WRITE);
1155 if (status < 0) {
1156 status = -EIO;
1157 mlog_errno(status);
1158 goto bail;
1159 }
1160 if (pde)
1161 le16_add_cpu(&pde->rec_len,
1162 le16_to_cpu(de->rec_len));
1163 de->inode = 0;
1164 inode_inc_iversion(dir);
1165 ocfs2_journal_dirty(handle, bh);
1166 goto bail;
1167 }
1168 i += le16_to_cpu(de->rec_len);
1169 pde = de;
1170 de = (struct ocfs2_dir_entry *)((char *)de + le16_to_cpu(de->rec_len));
1171 }
1172 bail:
1173 return status;
1174 }
1175
ocfs2_figure_dirent_hole(struct ocfs2_dir_entry * de)1176 static unsigned int ocfs2_figure_dirent_hole(struct ocfs2_dir_entry *de)
1177 {
1178 unsigned int hole;
1179
1180 if (le64_to_cpu(de->inode) == 0)
1181 hole = le16_to_cpu(de->rec_len);
1182 else
1183 hole = le16_to_cpu(de->rec_len) -
1184 OCFS2_DIR_REC_LEN(de->name_len);
1185
1186 return hole;
1187 }
1188
ocfs2_find_max_rec_len(struct super_block * sb,struct buffer_head * dirblock_bh)1189 static int ocfs2_find_max_rec_len(struct super_block *sb,
1190 struct buffer_head *dirblock_bh)
1191 {
1192 int size, this_hole, largest_hole = 0;
1193 char *trailer, *de_buf, *limit, *start = dirblock_bh->b_data;
1194 struct ocfs2_dir_entry *de;
1195
1196 trailer = (char *)ocfs2_trailer_from_bh(dirblock_bh, sb);
1197 size = ocfs2_dir_trailer_blk_off(sb);
1198 limit = start + size;
1199 de_buf = start;
1200 de = (struct ocfs2_dir_entry *)de_buf;
1201 do {
1202 if (de_buf != trailer) {
1203 this_hole = ocfs2_figure_dirent_hole(de);
1204 if (this_hole > largest_hole)
1205 largest_hole = this_hole;
1206 }
1207
1208 de_buf += le16_to_cpu(de->rec_len);
1209 de = (struct ocfs2_dir_entry *)de_buf;
1210 } while (de_buf < limit);
1211
1212 if (largest_hole >= OCFS2_DIR_MIN_REC_LEN)
1213 return largest_hole;
1214 return 0;
1215 }
1216
ocfs2_dx_list_remove_entry(struct ocfs2_dx_entry_list * entry_list,int index)1217 static void ocfs2_dx_list_remove_entry(struct ocfs2_dx_entry_list *entry_list,
1218 int index)
1219 {
1220 int num_used = le16_to_cpu(entry_list->de_num_used);
1221
1222 if (num_used == 1 || index == (num_used - 1))
1223 goto clear;
1224
1225 memmove(&entry_list->de_entries[index],
1226 &entry_list->de_entries[index + 1],
1227 (num_used - index - 1)*sizeof(struct ocfs2_dx_entry));
1228 clear:
1229 num_used--;
1230 memset(&entry_list->de_entries[num_used], 0,
1231 sizeof(struct ocfs2_dx_entry));
1232 entry_list->de_num_used = cpu_to_le16(num_used);
1233 }
1234
ocfs2_delete_entry_dx(handle_t * handle,struct inode * dir,struct ocfs2_dir_lookup_result * lookup)1235 static int ocfs2_delete_entry_dx(handle_t *handle, struct inode *dir,
1236 struct ocfs2_dir_lookup_result *lookup)
1237 {
1238 int ret, index, max_rec_len, add_to_free_list = 0;
1239 struct buffer_head *dx_root_bh = lookup->dl_dx_root_bh;
1240 struct buffer_head *leaf_bh = lookup->dl_leaf_bh;
1241 struct ocfs2_dx_leaf *dx_leaf;
1242 struct ocfs2_dx_entry *dx_entry = lookup->dl_dx_entry;
1243 struct ocfs2_dir_block_trailer *trailer;
1244 struct ocfs2_dx_root_block *dx_root;
1245 struct ocfs2_dx_entry_list *entry_list;
1246
1247 /*
1248 * This function gets a bit messy because we might have to
1249 * modify the root block, regardless of whether the indexed
1250 * entries are stored inline.
1251 */
1252
1253 /*
1254 * *Only* set 'entry_list' here, based on where we're looking
1255 * for the indexed entries. Later, we might still want to
1256 * journal both blocks, based on free list state.
1257 */
1258 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
1259 if (ocfs2_dx_root_inline(dx_root)) {
1260 entry_list = &dx_root->dr_entries;
1261 } else {
1262 dx_leaf = (struct ocfs2_dx_leaf *) lookup->dl_dx_leaf_bh->b_data;
1263 entry_list = &dx_leaf->dl_list;
1264 }
1265
1266 /* Neither of these are a disk corruption - that should have
1267 * been caught by lookup, before we got here. */
1268 BUG_ON(le16_to_cpu(entry_list->de_count) <= 0);
1269 BUG_ON(le16_to_cpu(entry_list->de_num_used) <= 0);
1270
1271 index = (char *)dx_entry - (char *)entry_list->de_entries;
1272 index /= sizeof(*dx_entry);
1273
1274 if (index >= le16_to_cpu(entry_list->de_num_used)) {
1275 mlog(ML_ERROR, "Dir %llu: Bad dx_entry ptr idx %d, (%p, %p)\n",
1276 (unsigned long long)OCFS2_I(dir)->ip_blkno, index,
1277 entry_list, dx_entry);
1278 return -EIO;
1279 }
1280
1281 /*
1282 * We know that removal of this dirent will leave enough room
1283 * for a new one, so add this block to the free list if it
1284 * isn't already there.
1285 */
1286 trailer = ocfs2_trailer_from_bh(leaf_bh, dir->i_sb);
1287 if (trailer->db_free_rec_len == 0)
1288 add_to_free_list = 1;
1289
1290 /*
1291 * Add the block holding our index into the journal before
1292 * removing the unindexed entry. If we get an error return
1293 * from __ocfs2_delete_entry(), then it hasn't removed the
1294 * entry yet. Likewise, successful return means we *must*
1295 * remove the indexed entry.
1296 *
1297 * We're also careful to journal the root tree block here as
1298 * the entry count needs to be updated. Also, we might be
1299 * adding to the start of the free list.
1300 */
1301 ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
1302 OCFS2_JOURNAL_ACCESS_WRITE);
1303 if (ret) {
1304 mlog_errno(ret);
1305 goto out;
1306 }
1307
1308 if (!ocfs2_dx_root_inline(dx_root)) {
1309 ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir),
1310 lookup->dl_dx_leaf_bh,
1311 OCFS2_JOURNAL_ACCESS_WRITE);
1312 if (ret) {
1313 mlog_errno(ret);
1314 goto out;
1315 }
1316 }
1317
1318 trace_ocfs2_delete_entry_dx((unsigned long long)OCFS2_I(dir)->ip_blkno,
1319 index);
1320
1321 ret = __ocfs2_delete_entry(handle, dir, lookup->dl_entry,
1322 leaf_bh, leaf_bh->b_data, leaf_bh->b_size);
1323 if (ret) {
1324 mlog_errno(ret);
1325 goto out;
1326 }
1327
1328 max_rec_len = ocfs2_find_max_rec_len(dir->i_sb, leaf_bh);
1329 trailer->db_free_rec_len = cpu_to_le16(max_rec_len);
1330 if (add_to_free_list) {
1331 trailer->db_free_next = dx_root->dr_free_blk;
1332 dx_root->dr_free_blk = cpu_to_le64(leaf_bh->b_blocknr);
1333 ocfs2_journal_dirty(handle, dx_root_bh);
1334 }
1335
1336 /* leaf_bh was journal_accessed for us in __ocfs2_delete_entry */
1337 ocfs2_journal_dirty(handle, leaf_bh);
1338
1339 le32_add_cpu(&dx_root->dr_num_entries, -1);
1340 ocfs2_journal_dirty(handle, dx_root_bh);
1341
1342 ocfs2_dx_list_remove_entry(entry_list, index);
1343
1344 if (!ocfs2_dx_root_inline(dx_root))
1345 ocfs2_journal_dirty(handle, lookup->dl_dx_leaf_bh);
1346
1347 out:
1348 return ret;
1349 }
1350
ocfs2_delete_entry_id(handle_t * handle,struct inode * dir,struct ocfs2_dir_entry * de_del,struct buffer_head * bh)1351 static inline int ocfs2_delete_entry_id(handle_t *handle,
1352 struct inode *dir,
1353 struct ocfs2_dir_entry *de_del,
1354 struct buffer_head *bh)
1355 {
1356 int ret;
1357 struct buffer_head *di_bh = NULL;
1358 struct ocfs2_dinode *di;
1359 struct ocfs2_inline_data *data;
1360
1361 ret = ocfs2_read_inode_block(dir, &di_bh);
1362 if (ret) {
1363 mlog_errno(ret);
1364 goto out;
1365 }
1366
1367 di = (struct ocfs2_dinode *)di_bh->b_data;
1368 data = &di->id2.i_data;
1369
1370 ret = __ocfs2_delete_entry(handle, dir, de_del, bh, data->id_data,
1371 i_size_read(dir));
1372
1373 brelse(di_bh);
1374 out:
1375 return ret;
1376 }
1377
ocfs2_delete_entry_el(handle_t * handle,struct inode * dir,struct ocfs2_dir_entry * de_del,struct buffer_head * bh)1378 static inline int ocfs2_delete_entry_el(handle_t *handle,
1379 struct inode *dir,
1380 struct ocfs2_dir_entry *de_del,
1381 struct buffer_head *bh)
1382 {
1383 return __ocfs2_delete_entry(handle, dir, de_del, bh, bh->b_data,
1384 bh->b_size);
1385 }
1386
1387 /*
1388 * Delete a directory entry. Hide the details of directory
1389 * implementation from the caller.
1390 */
ocfs2_delete_entry(handle_t * handle,struct inode * dir,struct ocfs2_dir_lookup_result * res)1391 int ocfs2_delete_entry(handle_t *handle,
1392 struct inode *dir,
1393 struct ocfs2_dir_lookup_result *res)
1394 {
1395 if (ocfs2_dir_indexed(dir))
1396 return ocfs2_delete_entry_dx(handle, dir, res);
1397
1398 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
1399 return ocfs2_delete_entry_id(handle, dir, res->dl_entry,
1400 res->dl_leaf_bh);
1401
1402 return ocfs2_delete_entry_el(handle, dir, res->dl_entry,
1403 res->dl_leaf_bh);
1404 }
1405
1406 /*
1407 * Check whether 'de' has enough room to hold an entry of
1408 * 'new_rec_len' bytes.
1409 */
ocfs2_dirent_would_fit(struct ocfs2_dir_entry * de,unsigned int new_rec_len)1410 static inline int ocfs2_dirent_would_fit(struct ocfs2_dir_entry *de,
1411 unsigned int new_rec_len)
1412 {
1413 unsigned int de_really_used;
1414
1415 /* Check whether this is an empty record with enough space */
1416 if (le64_to_cpu(de->inode) == 0 &&
1417 le16_to_cpu(de->rec_len) >= new_rec_len)
1418 return 1;
1419
1420 /*
1421 * Record might have free space at the end which we can
1422 * use.
1423 */
1424 de_really_used = OCFS2_DIR_REC_LEN(de->name_len);
1425 if (le16_to_cpu(de->rec_len) >= (de_really_used + new_rec_len))
1426 return 1;
1427
1428 return 0;
1429 }
1430
ocfs2_dx_dir_leaf_insert_tail(struct ocfs2_dx_leaf * dx_leaf,struct ocfs2_dx_entry * dx_new_entry)1431 static void ocfs2_dx_dir_leaf_insert_tail(struct ocfs2_dx_leaf *dx_leaf,
1432 struct ocfs2_dx_entry *dx_new_entry)
1433 {
1434 int i;
1435
1436 i = le16_to_cpu(dx_leaf->dl_list.de_num_used);
1437 dx_leaf->dl_list.de_entries[i] = *dx_new_entry;
1438
1439 le16_add_cpu(&dx_leaf->dl_list.de_num_used, 1);
1440 }
1441
ocfs2_dx_entry_list_insert(struct ocfs2_dx_entry_list * entry_list,struct ocfs2_dx_hinfo * hinfo,u64 dirent_blk)1442 static void ocfs2_dx_entry_list_insert(struct ocfs2_dx_entry_list *entry_list,
1443 struct ocfs2_dx_hinfo *hinfo,
1444 u64 dirent_blk)
1445 {
1446 int i;
1447 struct ocfs2_dx_entry *dx_entry;
1448
1449 i = le16_to_cpu(entry_list->de_num_used);
1450 dx_entry = &entry_list->de_entries[i];
1451
1452 memset(dx_entry, 0, sizeof(*dx_entry));
1453 dx_entry->dx_major_hash = cpu_to_le32(hinfo->major_hash);
1454 dx_entry->dx_minor_hash = cpu_to_le32(hinfo->minor_hash);
1455 dx_entry->dx_dirent_blk = cpu_to_le64(dirent_blk);
1456
1457 le16_add_cpu(&entry_list->de_num_used, 1);
1458 }
1459
__ocfs2_dx_dir_leaf_insert(struct inode * dir,handle_t * handle,struct ocfs2_dx_hinfo * hinfo,u64 dirent_blk,struct buffer_head * dx_leaf_bh)1460 static int __ocfs2_dx_dir_leaf_insert(struct inode *dir, handle_t *handle,
1461 struct ocfs2_dx_hinfo *hinfo,
1462 u64 dirent_blk,
1463 struct buffer_head *dx_leaf_bh)
1464 {
1465 int ret;
1466 struct ocfs2_dx_leaf *dx_leaf;
1467
1468 ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir), dx_leaf_bh,
1469 OCFS2_JOURNAL_ACCESS_WRITE);
1470 if (ret) {
1471 mlog_errno(ret);
1472 goto out;
1473 }
1474
1475 dx_leaf = (struct ocfs2_dx_leaf *)dx_leaf_bh->b_data;
1476 ocfs2_dx_entry_list_insert(&dx_leaf->dl_list, hinfo, dirent_blk);
1477 ocfs2_journal_dirty(handle, dx_leaf_bh);
1478
1479 out:
1480 return ret;
1481 }
1482
ocfs2_dx_inline_root_insert(struct inode * dir,handle_t * handle,struct ocfs2_dx_hinfo * hinfo,u64 dirent_blk,struct ocfs2_dx_root_block * dx_root)1483 static void ocfs2_dx_inline_root_insert(struct inode *dir, handle_t *handle,
1484 struct ocfs2_dx_hinfo *hinfo,
1485 u64 dirent_blk,
1486 struct ocfs2_dx_root_block *dx_root)
1487 {
1488 ocfs2_dx_entry_list_insert(&dx_root->dr_entries, hinfo, dirent_blk);
1489 }
1490
ocfs2_dx_dir_insert(struct inode * dir,handle_t * handle,struct ocfs2_dir_lookup_result * lookup)1491 static int ocfs2_dx_dir_insert(struct inode *dir, handle_t *handle,
1492 struct ocfs2_dir_lookup_result *lookup)
1493 {
1494 int ret = 0;
1495 struct ocfs2_dx_root_block *dx_root;
1496 struct buffer_head *dx_root_bh = lookup->dl_dx_root_bh;
1497
1498 ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
1499 OCFS2_JOURNAL_ACCESS_WRITE);
1500 if (ret) {
1501 mlog_errno(ret);
1502 goto out;
1503 }
1504
1505 dx_root = (struct ocfs2_dx_root_block *)lookup->dl_dx_root_bh->b_data;
1506 if (ocfs2_dx_root_inline(dx_root)) {
1507 ocfs2_dx_inline_root_insert(dir, handle,
1508 &lookup->dl_hinfo,
1509 lookup->dl_leaf_bh->b_blocknr,
1510 dx_root);
1511 } else {
1512 ret = __ocfs2_dx_dir_leaf_insert(dir, handle, &lookup->dl_hinfo,
1513 lookup->dl_leaf_bh->b_blocknr,
1514 lookup->dl_dx_leaf_bh);
1515 if (ret)
1516 goto out;
1517 }
1518
1519 le32_add_cpu(&dx_root->dr_num_entries, 1);
1520 ocfs2_journal_dirty(handle, dx_root_bh);
1521
1522 out:
1523 return ret;
1524 }
1525
ocfs2_remove_block_from_free_list(struct inode * dir,handle_t * handle,struct ocfs2_dir_lookup_result * lookup)1526 static void ocfs2_remove_block_from_free_list(struct inode *dir,
1527 handle_t *handle,
1528 struct ocfs2_dir_lookup_result *lookup)
1529 {
1530 struct ocfs2_dir_block_trailer *trailer, *prev;
1531 struct ocfs2_dx_root_block *dx_root;
1532 struct buffer_head *bh;
1533
1534 trailer = ocfs2_trailer_from_bh(lookup->dl_leaf_bh, dir->i_sb);
1535
1536 if (ocfs2_free_list_at_root(lookup)) {
1537 bh = lookup->dl_dx_root_bh;
1538 dx_root = (struct ocfs2_dx_root_block *)bh->b_data;
1539 dx_root->dr_free_blk = trailer->db_free_next;
1540 } else {
1541 bh = lookup->dl_prev_leaf_bh;
1542 prev = ocfs2_trailer_from_bh(bh, dir->i_sb);
1543 prev->db_free_next = trailer->db_free_next;
1544 }
1545
1546 trailer->db_free_rec_len = cpu_to_le16(0);
1547 trailer->db_free_next = cpu_to_le64(0);
1548
1549 ocfs2_journal_dirty(handle, bh);
1550 ocfs2_journal_dirty(handle, lookup->dl_leaf_bh);
1551 }
1552
1553 /*
1554 * This expects that a journal write has been reserved on
1555 * lookup->dl_prev_leaf_bh or lookup->dl_dx_root_bh
1556 */
ocfs2_recalc_free_list(struct inode * dir,handle_t * handle,struct ocfs2_dir_lookup_result * lookup)1557 static void ocfs2_recalc_free_list(struct inode *dir, handle_t *handle,
1558 struct ocfs2_dir_lookup_result *lookup)
1559 {
1560 int max_rec_len;
1561 struct ocfs2_dir_block_trailer *trailer;
1562
1563 /* Walk dl_leaf_bh to figure out what the new free rec_len is. */
1564 max_rec_len = ocfs2_find_max_rec_len(dir->i_sb, lookup->dl_leaf_bh);
1565 if (max_rec_len) {
1566 /*
1567 * There's still room in this block, so no need to remove it
1568 * from the free list. In this case, we just want to update
1569 * the rec len accounting.
1570 */
1571 trailer = ocfs2_trailer_from_bh(lookup->dl_leaf_bh, dir->i_sb);
1572 trailer->db_free_rec_len = cpu_to_le16(max_rec_len);
1573 ocfs2_journal_dirty(handle, lookup->dl_leaf_bh);
1574 } else {
1575 ocfs2_remove_block_from_free_list(dir, handle, lookup);
1576 }
1577 }
1578
1579 /* we don't always have a dentry for what we want to add, so people
1580 * like orphan dir can call this instead.
1581 *
1582 * The lookup context must have been filled from
1583 * ocfs2_prepare_dir_for_insert.
1584 */
__ocfs2_add_entry(handle_t * handle,struct inode * dir,const char * name,int namelen,struct inode * inode,u64 blkno,struct buffer_head * parent_fe_bh,struct ocfs2_dir_lookup_result * lookup)1585 int __ocfs2_add_entry(handle_t *handle,
1586 struct inode *dir,
1587 const char *name, int namelen,
1588 struct inode *inode, u64 blkno,
1589 struct buffer_head *parent_fe_bh,
1590 struct ocfs2_dir_lookup_result *lookup)
1591 {
1592 unsigned long offset;
1593 unsigned short rec_len;
1594 struct ocfs2_dir_entry *de, *de1;
1595 struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_fe_bh->b_data;
1596 struct super_block *sb = dir->i_sb;
1597 int retval;
1598 unsigned int size = sb->s_blocksize;
1599 struct buffer_head *insert_bh = lookup->dl_leaf_bh;
1600 char *data_start = insert_bh->b_data;
1601
1602 if (ocfs2_dir_indexed(dir)) {
1603 struct buffer_head *bh;
1604
1605 /*
1606 * An indexed dir may require that we update the free space
1607 * list. Reserve a write to the previous node in the list so
1608 * that we don't fail later.
1609 *
1610 * XXX: This can be either a dx_root_block, or an unindexed
1611 * directory tree leaf block.
1612 */
1613 if (ocfs2_free_list_at_root(lookup)) {
1614 bh = lookup->dl_dx_root_bh;
1615 retval = ocfs2_journal_access_dr(handle,
1616 INODE_CACHE(dir), bh,
1617 OCFS2_JOURNAL_ACCESS_WRITE);
1618 } else {
1619 bh = lookup->dl_prev_leaf_bh;
1620 retval = ocfs2_journal_access_db(handle,
1621 INODE_CACHE(dir), bh,
1622 OCFS2_JOURNAL_ACCESS_WRITE);
1623 }
1624 if (retval) {
1625 mlog_errno(retval);
1626 return retval;
1627 }
1628 } else if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1629 data_start = di->id2.i_data.id_data;
1630 size = i_size_read(dir);
1631
1632 BUG_ON(insert_bh != parent_fe_bh);
1633 }
1634
1635 rec_len = OCFS2_DIR_REC_LEN(namelen);
1636 offset = 0;
1637 de = (struct ocfs2_dir_entry *) data_start;
1638 while (1) {
1639 BUG_ON((char *)de >= (size + data_start));
1640
1641 /* These checks should've already been passed by the
1642 * prepare function, but I guess we can leave them
1643 * here anyway. */
1644 if (!ocfs2_check_dir_entry(dir, de, insert_bh, data_start,
1645 size, offset)) {
1646 retval = -ENOENT;
1647 goto bail;
1648 }
1649 if (ocfs2_match(namelen, name, de)) {
1650 retval = -EEXIST;
1651 goto bail;
1652 }
1653
1654 /* We're guaranteed that we should have space, so we
1655 * can't possibly have hit the trailer...right? */
1656 mlog_bug_on_msg(ocfs2_skip_dir_trailer(dir, de, offset, size),
1657 "Hit dir trailer trying to insert %.*s "
1658 "(namelen %d) into directory %llu. "
1659 "offset is %lu, trailer offset is %d\n",
1660 namelen, name, namelen,
1661 (unsigned long long)parent_fe_bh->b_blocknr,
1662 offset, ocfs2_dir_trailer_blk_off(dir->i_sb));
1663
1664 if (ocfs2_dirent_would_fit(de, rec_len)) {
1665 inode_set_mtime_to_ts(dir,
1666 inode_set_ctime_current(dir));
1667 retval = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
1668 if (retval < 0) {
1669 mlog_errno(retval);
1670 goto bail;
1671 }
1672
1673 if (insert_bh == parent_fe_bh)
1674 retval = ocfs2_journal_access_di(handle,
1675 INODE_CACHE(dir),
1676 insert_bh,
1677 OCFS2_JOURNAL_ACCESS_WRITE);
1678 else {
1679 retval = ocfs2_journal_access_db(handle,
1680 INODE_CACHE(dir),
1681 insert_bh,
1682 OCFS2_JOURNAL_ACCESS_WRITE);
1683
1684 if (!retval && ocfs2_dir_indexed(dir))
1685 retval = ocfs2_dx_dir_insert(dir,
1686 handle,
1687 lookup);
1688 }
1689
1690 if (retval) {
1691 mlog_errno(retval);
1692 goto bail;
1693 }
1694
1695 /* By now the buffer is marked for journaling */
1696 offset += le16_to_cpu(de->rec_len);
1697 if (le64_to_cpu(de->inode)) {
1698 de1 = (struct ocfs2_dir_entry *)((char *) de +
1699 OCFS2_DIR_REC_LEN(de->name_len));
1700 de1->rec_len =
1701 cpu_to_le16(le16_to_cpu(de->rec_len) -
1702 OCFS2_DIR_REC_LEN(de->name_len));
1703 de->rec_len = cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
1704 de = de1;
1705 }
1706 de->file_type = FT_UNKNOWN;
1707 if (blkno) {
1708 de->inode = cpu_to_le64(blkno);
1709 ocfs2_set_de_type(de, inode->i_mode);
1710 } else
1711 de->inode = 0;
1712 de->name_len = namelen;
1713 memcpy(de->name, name, namelen);
1714
1715 if (ocfs2_dir_indexed(dir))
1716 ocfs2_recalc_free_list(dir, handle, lookup);
1717
1718 inode_inc_iversion(dir);
1719 ocfs2_journal_dirty(handle, insert_bh);
1720 retval = 0;
1721 goto bail;
1722 }
1723
1724 offset += le16_to_cpu(de->rec_len);
1725 de = (struct ocfs2_dir_entry *) ((char *) de + le16_to_cpu(de->rec_len));
1726 }
1727
1728 /* when you think about it, the assert above should prevent us
1729 * from ever getting here. */
1730 retval = -ENOSPC;
1731 bail:
1732 if (retval)
1733 mlog_errno(retval);
1734
1735 return retval;
1736 }
1737
ocfs2_dir_foreach_blk_id(struct inode * inode,u64 * f_version,struct dir_context * ctx)1738 static int ocfs2_dir_foreach_blk_id(struct inode *inode,
1739 u64 *f_version,
1740 struct dir_context *ctx)
1741 {
1742 int ret, i;
1743 unsigned long offset = ctx->pos;
1744 struct buffer_head *di_bh = NULL;
1745 struct ocfs2_dinode *di;
1746 struct ocfs2_inline_data *data;
1747 struct ocfs2_dir_entry *de;
1748
1749 ret = ocfs2_read_inode_block(inode, &di_bh);
1750 if (ret) {
1751 mlog(ML_ERROR, "Unable to read inode block for dir %llu\n",
1752 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1753 goto out;
1754 }
1755
1756 di = (struct ocfs2_dinode *)di_bh->b_data;
1757 data = &di->id2.i_data;
1758
1759 while (ctx->pos < i_size_read(inode)) {
1760 /* If the dir block has changed since the last call to
1761 * readdir(2), then we might be pointing to an invalid
1762 * dirent right now. Scan from the start of the block
1763 * to make sure. */
1764 if (!inode_eq_iversion(inode, *f_version)) {
1765 for (i = 0; i < i_size_read(inode) && i < offset; ) {
1766 de = (struct ocfs2_dir_entry *)
1767 (data->id_data + i);
1768 /* It's too expensive to do a full
1769 * dirent test each time round this
1770 * loop, but we do have to test at
1771 * least that it is non-zero. A
1772 * failure will be detected in the
1773 * dirent test below. */
1774 if (le16_to_cpu(de->rec_len) <
1775 OCFS2_DIR_REC_LEN(1))
1776 break;
1777 i += le16_to_cpu(de->rec_len);
1778 }
1779 ctx->pos = offset = i;
1780 *f_version = inode_query_iversion(inode);
1781 }
1782
1783 de = (struct ocfs2_dir_entry *) (data->id_data + ctx->pos);
1784 if (!ocfs2_check_dir_entry(inode, de, di_bh, (char *)data->id_data,
1785 i_size_read(inode), ctx->pos)) {
1786 /* On error, skip the f_pos to the end. */
1787 ctx->pos = i_size_read(inode);
1788 break;
1789 }
1790 offset += le16_to_cpu(de->rec_len);
1791 if (le64_to_cpu(de->inode)) {
1792 if (!dir_emit(ctx, de->name, de->name_len,
1793 le64_to_cpu(de->inode),
1794 fs_ftype_to_dtype(de->file_type)))
1795 goto out;
1796 }
1797 ctx->pos += le16_to_cpu(de->rec_len);
1798 }
1799 out:
1800 brelse(di_bh);
1801 return 0;
1802 }
1803
1804 /*
1805 * NOTE: This function can be called against unindexed directories,
1806 * and indexed ones.
1807 */
ocfs2_dir_foreach_blk_el(struct inode * inode,u64 * f_version,struct dir_context * ctx,bool persist)1808 static int ocfs2_dir_foreach_blk_el(struct inode *inode,
1809 u64 *f_version,
1810 struct dir_context *ctx,
1811 bool persist)
1812 {
1813 unsigned long offset, blk, last_ra_blk = 0;
1814 int i;
1815 struct buffer_head * bh, * tmp;
1816 struct ocfs2_dir_entry * de;
1817 struct super_block * sb = inode->i_sb;
1818 unsigned int ra_sectors = 16;
1819 int stored = 0;
1820
1821 bh = NULL;
1822
1823 offset = ctx->pos & (sb->s_blocksize - 1);
1824
1825 while (ctx->pos < i_size_read(inode)) {
1826 blk = ctx->pos >> sb->s_blocksize_bits;
1827 if (ocfs2_read_dir_block(inode, blk, &bh, 0)) {
1828 /* Skip the corrupt dirblock and keep trying */
1829 ctx->pos += sb->s_blocksize - offset;
1830 continue;
1831 }
1832
1833 /* The idea here is to begin with 8k read-ahead and to stay
1834 * 4k ahead of our current position.
1835 *
1836 * TODO: Use the pagecache for this. We just need to
1837 * make sure it's cluster-safe... */
1838 if (!last_ra_blk
1839 || (((last_ra_blk - blk) << 9) <= (ra_sectors / 2))) {
1840 for (i = ra_sectors >> (sb->s_blocksize_bits - 9);
1841 i > 0; i--) {
1842 tmp = NULL;
1843 if (!ocfs2_read_dir_block(inode, ++blk, &tmp,
1844 OCFS2_BH_READAHEAD))
1845 brelse(tmp);
1846 }
1847 last_ra_blk = blk;
1848 ra_sectors = 8;
1849 }
1850
1851 /* If the dir block has changed since the last call to
1852 * readdir(2), then we might be pointing to an invalid
1853 * dirent right now. Scan from the start of the block
1854 * to make sure. */
1855 if (!inode_eq_iversion(inode, *f_version)) {
1856 for (i = 0; i < sb->s_blocksize && i < offset; ) {
1857 de = (struct ocfs2_dir_entry *) (bh->b_data + i);
1858 /* It's too expensive to do a full
1859 * dirent test each time round this
1860 * loop, but we do have to test at
1861 * least that it is non-zero. A
1862 * failure will be detected in the
1863 * dirent test below. */
1864 if (le16_to_cpu(de->rec_len) <
1865 OCFS2_DIR_REC_LEN(1))
1866 break;
1867 i += le16_to_cpu(de->rec_len);
1868 }
1869 offset = i;
1870 ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1))
1871 | offset;
1872 *f_version = inode_query_iversion(inode);
1873 }
1874
1875 while (ctx->pos < i_size_read(inode)
1876 && offset < sb->s_blocksize) {
1877 de = (struct ocfs2_dir_entry *) (bh->b_data + offset);
1878 if (!ocfs2_check_dir_entry(inode, de, bh, bh->b_data,
1879 sb->s_blocksize, offset)) {
1880 /* On error, skip the f_pos to the
1881 next block. */
1882 ctx->pos = (ctx->pos | (sb->s_blocksize - 1)) + 1;
1883 break;
1884 }
1885 if (le64_to_cpu(de->inode)) {
1886 if (!dir_emit(ctx, de->name,
1887 de->name_len,
1888 le64_to_cpu(de->inode),
1889 fs_ftype_to_dtype(de->file_type))) {
1890 brelse(bh);
1891 return 0;
1892 }
1893 stored++;
1894 }
1895 offset += le16_to_cpu(de->rec_len);
1896 ctx->pos += le16_to_cpu(de->rec_len);
1897 }
1898 offset = 0;
1899 brelse(bh);
1900 bh = NULL;
1901 if (!persist && stored)
1902 break;
1903 }
1904 return 0;
1905 }
1906
ocfs2_dir_foreach_blk(struct inode * inode,u64 * f_version,struct dir_context * ctx,bool persist)1907 static int ocfs2_dir_foreach_blk(struct inode *inode, u64 *f_version,
1908 struct dir_context *ctx,
1909 bool persist)
1910 {
1911 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
1912 return ocfs2_dir_foreach_blk_id(inode, f_version, ctx);
1913 return ocfs2_dir_foreach_blk_el(inode, f_version, ctx, persist);
1914 }
1915
1916 /*
1917 * This is intended to be called from inside other kernel functions,
1918 * so we fake some arguments.
1919 */
ocfs2_dir_foreach(struct inode * inode,struct dir_context * ctx)1920 int ocfs2_dir_foreach(struct inode *inode, struct dir_context *ctx)
1921 {
1922 u64 version = inode_query_iversion(inode);
1923 ocfs2_dir_foreach_blk(inode, &version, ctx, true);
1924 return 0;
1925 }
1926
1927 /*
1928 * ocfs2_readdir()
1929 *
1930 */
ocfs2_readdir(struct file * file,struct dir_context * ctx)1931 int ocfs2_readdir(struct file *file, struct dir_context *ctx)
1932 {
1933 int error = 0;
1934 struct inode *inode = file_inode(file);
1935 struct ocfs2_file_private *fp = file->private_data;
1936 int lock_level = 0;
1937
1938 trace_ocfs2_readdir((unsigned long long)OCFS2_I(inode)->ip_blkno);
1939
1940 error = ocfs2_inode_lock_atime(inode, file->f_path.mnt, &lock_level, 1);
1941 if (lock_level && error >= 0) {
1942 /* We release EX lock which used to update atime
1943 * and get PR lock again to reduce contention
1944 * on commonly accessed directories. */
1945 ocfs2_inode_unlock(inode, 1);
1946 lock_level = 0;
1947 error = ocfs2_inode_lock(inode, NULL, 0);
1948 }
1949 if (error < 0) {
1950 if (error != -ENOENT)
1951 mlog_errno(error);
1952 /* we haven't got any yet, so propagate the error. */
1953 goto bail_nolock;
1954 }
1955
1956 error = ocfs2_dir_foreach_blk(inode, &fp->cookie, ctx, false);
1957
1958 ocfs2_inode_unlock(inode, lock_level);
1959 if (error)
1960 mlog_errno(error);
1961
1962 bail_nolock:
1963
1964 return error;
1965 }
1966
1967 /*
1968 * NOTE: this should always be called with parent dir i_rwsem taken.
1969 */
ocfs2_find_files_on_disk(const char * name,int namelen,u64 * blkno,struct inode * inode,struct ocfs2_dir_lookup_result * lookup)1970 int ocfs2_find_files_on_disk(const char *name,
1971 int namelen,
1972 u64 *blkno,
1973 struct inode *inode,
1974 struct ocfs2_dir_lookup_result *lookup)
1975 {
1976 int status = -ENOENT;
1977
1978 trace_ocfs2_find_files_on_disk(namelen, name, blkno,
1979 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1980
1981 status = ocfs2_find_entry(name, namelen, inode, lookup);
1982 if (status)
1983 goto leave;
1984
1985 *blkno = le64_to_cpu(lookup->dl_entry->inode);
1986
1987 status = 0;
1988 leave:
1989
1990 return status;
1991 }
1992
1993 /*
1994 * Convenience function for callers which just want the block number
1995 * mapped to a name and don't require the full dirent info, etc.
1996 */
ocfs2_lookup_ino_from_name(struct inode * dir,const char * name,int namelen,u64 * blkno)1997 int ocfs2_lookup_ino_from_name(struct inode *dir, const char *name,
1998 int namelen, u64 *blkno)
1999 {
2000 int ret;
2001 struct ocfs2_dir_lookup_result lookup = { NULL, };
2002
2003 ret = ocfs2_find_files_on_disk(name, namelen, blkno, dir, &lookup);
2004 ocfs2_free_dir_lookup_result(&lookup);
2005
2006 return ret;
2007 }
2008
2009 /* Check for a name within a directory.
2010 *
2011 * Return 0 if the name does not exist
2012 * Return -EEXIST if the directory contains the name
2013 *
2014 * Callers should have i_rwsem + a cluster lock on dir
2015 */
ocfs2_check_dir_for_entry(struct inode * dir,const char * name,int namelen)2016 int ocfs2_check_dir_for_entry(struct inode *dir,
2017 const char *name,
2018 int namelen)
2019 {
2020 int ret = 0;
2021 struct ocfs2_dir_lookup_result lookup = { NULL, };
2022
2023 trace_ocfs2_check_dir_for_entry(
2024 (unsigned long long)OCFS2_I(dir)->ip_blkno, namelen, name);
2025
2026 if (ocfs2_find_entry(name, namelen, dir, &lookup) == 0) {
2027 ret = -EEXIST;
2028 mlog_errno(ret);
2029 }
2030
2031 ocfs2_free_dir_lookup_result(&lookup);
2032
2033 return ret;
2034 }
2035
2036 struct ocfs2_empty_dir_priv {
2037 struct dir_context ctx;
2038 unsigned seen_dot;
2039 unsigned seen_dot_dot;
2040 unsigned seen_other;
2041 unsigned dx_dir;
2042 };
ocfs2_empty_dir_filldir(struct dir_context * ctx,const char * name,int name_len,loff_t pos,u64 ino,unsigned type)2043 static bool ocfs2_empty_dir_filldir(struct dir_context *ctx, const char *name,
2044 int name_len, loff_t pos, u64 ino,
2045 unsigned type)
2046 {
2047 struct ocfs2_empty_dir_priv *p =
2048 container_of(ctx, struct ocfs2_empty_dir_priv, ctx);
2049
2050 /*
2051 * Check the positions of "." and ".." records to be sure
2052 * they're in the correct place.
2053 *
2054 * Indexed directories don't need to proceed past the first
2055 * two entries, so we end the scan after seeing '..'. Despite
2056 * that, we allow the scan to proceed In the event that we
2057 * have a corrupted indexed directory (no dot or dot dot
2058 * entries). This allows us to double check for existing
2059 * entries which might not have been found in the index.
2060 */
2061 if (name_len == 1 && !strncmp(".", name, 1) && pos == 0) {
2062 p->seen_dot = 1;
2063 return true;
2064 }
2065
2066 if (name_len == 2 && !strncmp("..", name, 2) &&
2067 pos == OCFS2_DIR_REC_LEN(1)) {
2068 p->seen_dot_dot = 1;
2069
2070 if (p->dx_dir && p->seen_dot)
2071 return false;
2072
2073 return true;
2074 }
2075
2076 p->seen_other = 1;
2077 return false;
2078 }
2079
ocfs2_empty_dir_dx(struct inode * inode,struct ocfs2_empty_dir_priv * priv)2080 static int ocfs2_empty_dir_dx(struct inode *inode,
2081 struct ocfs2_empty_dir_priv *priv)
2082 {
2083 int ret;
2084 struct buffer_head *di_bh = NULL;
2085 struct buffer_head *dx_root_bh = NULL;
2086 struct ocfs2_dinode *di;
2087 struct ocfs2_dx_root_block *dx_root;
2088
2089 priv->dx_dir = 1;
2090
2091 ret = ocfs2_read_inode_block(inode, &di_bh);
2092 if (ret) {
2093 mlog_errno(ret);
2094 goto out;
2095 }
2096 di = (struct ocfs2_dinode *)di_bh->b_data;
2097
2098 ret = ocfs2_read_dx_root(inode, di, &dx_root_bh);
2099 if (ret) {
2100 mlog_errno(ret);
2101 goto out;
2102 }
2103 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
2104
2105 if (le32_to_cpu(dx_root->dr_num_entries) != 2)
2106 priv->seen_other = 1;
2107
2108 out:
2109 brelse(di_bh);
2110 brelse(dx_root_bh);
2111 return ret;
2112 }
2113
2114 /*
2115 * routine to check that the specified directory is empty (for rmdir)
2116 *
2117 * Returns 1 if dir is empty, zero otherwise.
2118 *
2119 * XXX: This is a performance problem for unindexed directories.
2120 */
ocfs2_empty_dir(struct inode * inode)2121 int ocfs2_empty_dir(struct inode *inode)
2122 {
2123 int ret;
2124 struct ocfs2_empty_dir_priv priv = {
2125 .ctx.actor = ocfs2_empty_dir_filldir,
2126 };
2127
2128 if (ocfs2_dir_indexed(inode)) {
2129 ret = ocfs2_empty_dir_dx(inode, &priv);
2130 if (ret)
2131 mlog_errno(ret);
2132 /*
2133 * We still run ocfs2_dir_foreach to get the checks
2134 * for "." and "..".
2135 */
2136 }
2137
2138 ret = ocfs2_dir_foreach(inode, &priv.ctx);
2139 if (ret)
2140 mlog_errno(ret);
2141
2142 if (!priv.seen_dot || !priv.seen_dot_dot) {
2143 mlog(ML_ERROR, "bad directory (dir #%llu) - no `.' or `..'\n",
2144 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2145 /*
2146 * XXX: Is it really safe to allow an unlink to continue?
2147 */
2148 return 1;
2149 }
2150
2151 return !priv.seen_other;
2152 }
2153
2154 /*
2155 * Fills "." and ".." dirents in a new directory block. Returns dirent for
2156 * "..", which might be used during creation of a directory with a trailing
2157 * header. It is otherwise safe to ignore the return code.
2158 */
ocfs2_fill_initial_dirents(struct inode * inode,struct inode * parent,char * start,unsigned int size)2159 static struct ocfs2_dir_entry *ocfs2_fill_initial_dirents(struct inode *inode,
2160 struct inode *parent,
2161 char *start,
2162 unsigned int size)
2163 {
2164 struct ocfs2_dir_entry *de = (struct ocfs2_dir_entry *)start;
2165
2166 de->inode = cpu_to_le64(OCFS2_I(inode)->ip_blkno);
2167 de->name_len = 1;
2168 de->rec_len =
2169 cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
2170 strcpy(de->name, ".");
2171 ocfs2_set_de_type(de, S_IFDIR);
2172
2173 de = (struct ocfs2_dir_entry *) ((char *)de + le16_to_cpu(de->rec_len));
2174 de->inode = cpu_to_le64(OCFS2_I(parent)->ip_blkno);
2175 de->rec_len = cpu_to_le16(size - OCFS2_DIR_REC_LEN(1));
2176 de->name_len = 2;
2177 strcpy(de->name, "..");
2178 ocfs2_set_de_type(de, S_IFDIR);
2179
2180 return de;
2181 }
2182
2183 /*
2184 * This works together with code in ocfs2_mknod_locked() which sets
2185 * the inline-data flag and initializes the inline-data section.
2186 */
ocfs2_fill_new_dir_id(struct ocfs2_super * osb,handle_t * handle,struct inode * parent,struct inode * inode,struct buffer_head * di_bh)2187 static int ocfs2_fill_new_dir_id(struct ocfs2_super *osb,
2188 handle_t *handle,
2189 struct inode *parent,
2190 struct inode *inode,
2191 struct buffer_head *di_bh)
2192 {
2193 int ret;
2194 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2195 struct ocfs2_inline_data *data = &di->id2.i_data;
2196 unsigned int size = le16_to_cpu(data->id_count);
2197
2198 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2199 OCFS2_JOURNAL_ACCESS_WRITE);
2200 if (ret) {
2201 mlog_errno(ret);
2202 goto out;
2203 }
2204
2205 ocfs2_fill_initial_dirents(inode, parent, data->id_data, size);
2206 ocfs2_journal_dirty(handle, di_bh);
2207
2208 i_size_write(inode, size);
2209 set_nlink(inode, 2);
2210 inode->i_blocks = ocfs2_inode_sector_count(inode);
2211
2212 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
2213 if (ret < 0)
2214 mlog_errno(ret);
2215
2216 out:
2217 return ret;
2218 }
2219
ocfs2_fill_new_dir_el(struct ocfs2_super * osb,handle_t * handle,struct inode * parent,struct inode * inode,struct buffer_head * fe_bh,struct ocfs2_alloc_context * data_ac,struct buffer_head ** ret_new_bh)2220 static int ocfs2_fill_new_dir_el(struct ocfs2_super *osb,
2221 handle_t *handle,
2222 struct inode *parent,
2223 struct inode *inode,
2224 struct buffer_head *fe_bh,
2225 struct ocfs2_alloc_context *data_ac,
2226 struct buffer_head **ret_new_bh)
2227 {
2228 int status;
2229 unsigned int size = osb->sb->s_blocksize;
2230 struct buffer_head *new_bh = NULL;
2231 struct ocfs2_dir_entry *de;
2232
2233 if (ocfs2_new_dir_wants_trailer(inode))
2234 size = ocfs2_dir_trailer_blk_off(parent->i_sb);
2235
2236 status = ocfs2_do_extend_dir(osb->sb, handle, inode, fe_bh,
2237 data_ac, NULL, &new_bh);
2238 if (status < 0) {
2239 mlog_errno(status);
2240 goto bail;
2241 }
2242
2243 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2244
2245 status = ocfs2_journal_access_db(handle, INODE_CACHE(inode), new_bh,
2246 OCFS2_JOURNAL_ACCESS_CREATE);
2247 if (status < 0) {
2248 mlog_errno(status);
2249 goto bail;
2250 }
2251 memset(new_bh->b_data, 0, osb->sb->s_blocksize);
2252
2253 de = ocfs2_fill_initial_dirents(inode, parent, new_bh->b_data, size);
2254 if (ocfs2_new_dir_wants_trailer(inode)) {
2255 int size = le16_to_cpu(de->rec_len);
2256
2257 /*
2258 * Figure out the size of the hole left over after
2259 * insertion of '.' and '..'. The trailer wants this
2260 * information.
2261 */
2262 size -= OCFS2_DIR_REC_LEN(2);
2263 size -= sizeof(struct ocfs2_dir_block_trailer);
2264
2265 ocfs2_init_dir_trailer(inode, new_bh, size);
2266 }
2267
2268 ocfs2_journal_dirty(handle, new_bh);
2269
2270 i_size_write(inode, inode->i_sb->s_blocksize);
2271 set_nlink(inode, 2);
2272 inode->i_blocks = ocfs2_inode_sector_count(inode);
2273 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
2274 if (status < 0) {
2275 mlog_errno(status);
2276 goto bail;
2277 }
2278
2279 status = 0;
2280 if (ret_new_bh) {
2281 *ret_new_bh = new_bh;
2282 new_bh = NULL;
2283 }
2284 bail:
2285 brelse(new_bh);
2286
2287 return status;
2288 }
2289
ocfs2_dx_dir_attach_index(struct ocfs2_super * osb,handle_t * handle,struct inode * dir,struct buffer_head * di_bh,struct buffer_head * dirdata_bh,struct ocfs2_alloc_context * meta_ac,int dx_inline,u32 num_entries,struct buffer_head ** ret_dx_root_bh)2290 static int ocfs2_dx_dir_attach_index(struct ocfs2_super *osb,
2291 handle_t *handle, struct inode *dir,
2292 struct buffer_head *di_bh,
2293 struct buffer_head *dirdata_bh,
2294 struct ocfs2_alloc_context *meta_ac,
2295 int dx_inline, u32 num_entries,
2296 struct buffer_head **ret_dx_root_bh)
2297 {
2298 int ret;
2299 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
2300 u16 dr_suballoc_bit;
2301 u64 suballoc_loc, dr_blkno;
2302 unsigned int num_bits;
2303 struct buffer_head *dx_root_bh = NULL;
2304 struct ocfs2_dx_root_block *dx_root;
2305 struct ocfs2_dir_block_trailer *trailer =
2306 ocfs2_trailer_from_bh(dirdata_bh, dir->i_sb);
2307
2308 ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
2309 &dr_suballoc_bit, &num_bits, &dr_blkno);
2310 if (ret) {
2311 mlog_errno(ret);
2312 goto out;
2313 }
2314
2315 trace_ocfs2_dx_dir_attach_index(
2316 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2317 (unsigned long long)dr_blkno);
2318
2319 dx_root_bh = sb_getblk(osb->sb, dr_blkno);
2320 if (dx_root_bh == NULL) {
2321 ret = -ENOMEM;
2322 goto out;
2323 }
2324 ocfs2_set_new_buffer_uptodate(INODE_CACHE(dir), dx_root_bh);
2325
2326 ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
2327 OCFS2_JOURNAL_ACCESS_CREATE);
2328 if (ret < 0) {
2329 mlog_errno(ret);
2330 goto out;
2331 }
2332
2333 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
2334 memset(dx_root, 0, osb->sb->s_blocksize);
2335 strcpy(dx_root->dr_signature, OCFS2_DX_ROOT_SIGNATURE);
2336 dx_root->dr_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
2337 dx_root->dr_suballoc_loc = cpu_to_le64(suballoc_loc);
2338 dx_root->dr_suballoc_bit = cpu_to_le16(dr_suballoc_bit);
2339 dx_root->dr_fs_generation = cpu_to_le32(osb->fs_generation);
2340 dx_root->dr_blkno = cpu_to_le64(dr_blkno);
2341 dx_root->dr_dir_blkno = cpu_to_le64(OCFS2_I(dir)->ip_blkno);
2342 dx_root->dr_num_entries = cpu_to_le32(num_entries);
2343 if (le16_to_cpu(trailer->db_free_rec_len))
2344 dx_root->dr_free_blk = cpu_to_le64(dirdata_bh->b_blocknr);
2345 else
2346 dx_root->dr_free_blk = cpu_to_le64(0);
2347
2348 if (dx_inline) {
2349 dx_root->dr_flags |= OCFS2_DX_FLAG_INLINE;
2350 dx_root->dr_entries.de_count =
2351 cpu_to_le16(ocfs2_dx_entries_per_root(osb->sb));
2352 } else {
2353 dx_root->dr_list.l_count =
2354 cpu_to_le16(ocfs2_extent_recs_per_dx_root(osb->sb));
2355 }
2356 ocfs2_journal_dirty(handle, dx_root_bh);
2357
2358 ret = ocfs2_journal_access_di(handle, INODE_CACHE(dir), di_bh,
2359 OCFS2_JOURNAL_ACCESS_CREATE);
2360 if (ret) {
2361 mlog_errno(ret);
2362 goto out;
2363 }
2364
2365 di->i_dx_root = cpu_to_le64(dr_blkno);
2366
2367 spin_lock(&OCFS2_I(dir)->ip_lock);
2368 OCFS2_I(dir)->ip_dyn_features |= OCFS2_INDEXED_DIR_FL;
2369 di->i_dyn_features = cpu_to_le16(OCFS2_I(dir)->ip_dyn_features);
2370 spin_unlock(&OCFS2_I(dir)->ip_lock);
2371
2372 ocfs2_journal_dirty(handle, di_bh);
2373
2374 *ret_dx_root_bh = dx_root_bh;
2375 dx_root_bh = NULL;
2376
2377 out:
2378 brelse(dx_root_bh);
2379 return ret;
2380 }
2381
ocfs2_dx_dir_format_cluster(struct ocfs2_super * osb,handle_t * handle,struct inode * dir,struct buffer_head ** dx_leaves,int num_dx_leaves,u64 start_blk)2382 static int ocfs2_dx_dir_format_cluster(struct ocfs2_super *osb,
2383 handle_t *handle, struct inode *dir,
2384 struct buffer_head **dx_leaves,
2385 int num_dx_leaves, u64 start_blk)
2386 {
2387 int ret, i;
2388 struct ocfs2_dx_leaf *dx_leaf;
2389 struct buffer_head *bh;
2390
2391 for (i = 0; i < num_dx_leaves; i++) {
2392 bh = sb_getblk(osb->sb, start_blk + i);
2393 if (bh == NULL) {
2394 ret = -ENOMEM;
2395 goto out;
2396 }
2397 dx_leaves[i] = bh;
2398
2399 ocfs2_set_new_buffer_uptodate(INODE_CACHE(dir), bh);
2400
2401 ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir), bh,
2402 OCFS2_JOURNAL_ACCESS_CREATE);
2403 if (ret < 0) {
2404 mlog_errno(ret);
2405 goto out;
2406 }
2407
2408 dx_leaf = (struct ocfs2_dx_leaf *) bh->b_data;
2409
2410 memset(dx_leaf, 0, osb->sb->s_blocksize);
2411 strcpy(dx_leaf->dl_signature, OCFS2_DX_LEAF_SIGNATURE);
2412 dx_leaf->dl_fs_generation = cpu_to_le32(osb->fs_generation);
2413 dx_leaf->dl_blkno = cpu_to_le64(bh->b_blocknr);
2414 dx_leaf->dl_list.de_count =
2415 cpu_to_le16(ocfs2_dx_entries_per_leaf(osb->sb));
2416
2417 trace_ocfs2_dx_dir_format_cluster(
2418 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2419 (unsigned long long)bh->b_blocknr,
2420 le16_to_cpu(dx_leaf->dl_list.de_count));
2421
2422 ocfs2_journal_dirty(handle, bh);
2423 }
2424
2425 ret = 0;
2426 out:
2427 return ret;
2428 }
2429
2430 /*
2431 * Allocates and formats a new cluster for use in an indexed dir
2432 * leaf. This version will not do the extent insert, so that it can be
2433 * used by operations which need careful ordering.
2434 */
__ocfs2_dx_dir_new_cluster(struct inode * dir,u32 cpos,handle_t * handle,struct ocfs2_alloc_context * data_ac,struct buffer_head ** dx_leaves,int num_dx_leaves,u64 * ret_phys_blkno)2435 static int __ocfs2_dx_dir_new_cluster(struct inode *dir,
2436 u32 cpos, handle_t *handle,
2437 struct ocfs2_alloc_context *data_ac,
2438 struct buffer_head **dx_leaves,
2439 int num_dx_leaves, u64 *ret_phys_blkno)
2440 {
2441 int ret;
2442 u32 phys, num;
2443 u64 phys_blkno;
2444 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2445
2446 /*
2447 * XXX: For create, this should claim cluster for the index
2448 * *before* the unindexed insert so that we have a better
2449 * chance of contiguousness as the directory grows in number
2450 * of entries.
2451 */
2452 ret = __ocfs2_claim_clusters(handle, data_ac, 1, 1, &phys, &num);
2453 if (ret) {
2454 mlog_errno(ret);
2455 goto out;
2456 }
2457
2458 /*
2459 * Format the new cluster first. That way, we're inserting
2460 * valid data.
2461 */
2462 phys_blkno = ocfs2_clusters_to_blocks(osb->sb, phys);
2463 ret = ocfs2_dx_dir_format_cluster(osb, handle, dir, dx_leaves,
2464 num_dx_leaves, phys_blkno);
2465 if (ret) {
2466 mlog_errno(ret);
2467 goto out;
2468 }
2469
2470 *ret_phys_blkno = phys_blkno;
2471 out:
2472 return ret;
2473 }
2474
ocfs2_dx_dir_new_cluster(struct inode * dir,struct ocfs2_extent_tree * et,u32 cpos,handle_t * handle,struct ocfs2_alloc_context * data_ac,struct ocfs2_alloc_context * meta_ac,struct buffer_head ** dx_leaves,int num_dx_leaves)2475 static int ocfs2_dx_dir_new_cluster(struct inode *dir,
2476 struct ocfs2_extent_tree *et,
2477 u32 cpos, handle_t *handle,
2478 struct ocfs2_alloc_context *data_ac,
2479 struct ocfs2_alloc_context *meta_ac,
2480 struct buffer_head **dx_leaves,
2481 int num_dx_leaves)
2482 {
2483 int ret;
2484 u64 phys_blkno;
2485
2486 ret = __ocfs2_dx_dir_new_cluster(dir, cpos, handle, data_ac, dx_leaves,
2487 num_dx_leaves, &phys_blkno);
2488 if (ret) {
2489 mlog_errno(ret);
2490 goto out;
2491 }
2492
2493 ret = ocfs2_insert_extent(handle, et, cpos, phys_blkno, 1, 0,
2494 meta_ac);
2495 if (ret)
2496 mlog_errno(ret);
2497 out:
2498 return ret;
2499 }
2500
ocfs2_dx_dir_kmalloc_leaves(struct super_block * sb,int * ret_num_leaves)2501 static struct buffer_head **ocfs2_dx_dir_kmalloc_leaves(struct super_block *sb,
2502 int *ret_num_leaves)
2503 {
2504 int num_dx_leaves = ocfs2_clusters_to_blocks(sb, 1);
2505 struct buffer_head **dx_leaves;
2506
2507 dx_leaves = kcalloc(num_dx_leaves, sizeof(struct buffer_head *),
2508 GFP_NOFS);
2509 if (dx_leaves && ret_num_leaves)
2510 *ret_num_leaves = num_dx_leaves;
2511
2512 return dx_leaves;
2513 }
2514
ocfs2_fill_new_dir_dx(struct ocfs2_super * osb,handle_t * handle,struct inode * parent,struct inode * inode,struct buffer_head * di_bh,struct ocfs2_alloc_context * data_ac,struct ocfs2_alloc_context * meta_ac)2515 static int ocfs2_fill_new_dir_dx(struct ocfs2_super *osb,
2516 handle_t *handle,
2517 struct inode *parent,
2518 struct inode *inode,
2519 struct buffer_head *di_bh,
2520 struct ocfs2_alloc_context *data_ac,
2521 struct ocfs2_alloc_context *meta_ac)
2522 {
2523 int ret;
2524 struct buffer_head *leaf_bh = NULL;
2525 struct buffer_head *dx_root_bh = NULL;
2526 struct ocfs2_dx_hinfo hinfo;
2527 struct ocfs2_dx_root_block *dx_root;
2528 struct ocfs2_dx_entry_list *entry_list;
2529
2530 /*
2531 * Our strategy is to create the directory as though it were
2532 * unindexed, then add the index block. This works with very
2533 * little complication since the state of a new directory is a
2534 * very well known quantity.
2535 *
2536 * Essentially, we have two dirents ("." and ".."), in the 1st
2537 * block which need indexing. These are easily inserted into
2538 * the index block.
2539 */
2540
2541 ret = ocfs2_fill_new_dir_el(osb, handle, parent, inode, di_bh,
2542 data_ac, &leaf_bh);
2543 if (ret) {
2544 mlog_errno(ret);
2545 goto out;
2546 }
2547
2548 ret = ocfs2_dx_dir_attach_index(osb, handle, inode, di_bh, leaf_bh,
2549 meta_ac, 1, 2, &dx_root_bh);
2550 if (ret) {
2551 mlog_errno(ret);
2552 goto out;
2553 }
2554 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
2555 entry_list = &dx_root->dr_entries;
2556
2557 /* Buffer has been journaled for us by ocfs2_dx_dir_attach_index */
2558 ocfs2_dx_dir_name_hash(inode, ".", 1, &hinfo);
2559 ocfs2_dx_entry_list_insert(entry_list, &hinfo, leaf_bh->b_blocknr);
2560
2561 ocfs2_dx_dir_name_hash(inode, "..", 2, &hinfo);
2562 ocfs2_dx_entry_list_insert(entry_list, &hinfo, leaf_bh->b_blocknr);
2563
2564 out:
2565 brelse(dx_root_bh);
2566 brelse(leaf_bh);
2567 return ret;
2568 }
2569
ocfs2_fill_new_dir(struct ocfs2_super * osb,handle_t * handle,struct inode * parent,struct inode * inode,struct buffer_head * fe_bh,struct ocfs2_alloc_context * data_ac,struct ocfs2_alloc_context * meta_ac)2570 int ocfs2_fill_new_dir(struct ocfs2_super *osb,
2571 handle_t *handle,
2572 struct inode *parent,
2573 struct inode *inode,
2574 struct buffer_head *fe_bh,
2575 struct ocfs2_alloc_context *data_ac,
2576 struct ocfs2_alloc_context *meta_ac)
2577
2578 {
2579 BUG_ON(!ocfs2_supports_inline_data(osb) && data_ac == NULL);
2580
2581 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2582 return ocfs2_fill_new_dir_id(osb, handle, parent, inode, fe_bh);
2583
2584 if (ocfs2_supports_indexed_dirs(osb))
2585 return ocfs2_fill_new_dir_dx(osb, handle, parent, inode, fe_bh,
2586 data_ac, meta_ac);
2587
2588 return ocfs2_fill_new_dir_el(osb, handle, parent, inode, fe_bh,
2589 data_ac, NULL);
2590 }
2591
ocfs2_dx_dir_index_block(struct inode * dir,handle_t * handle,struct buffer_head ** dx_leaves,int num_dx_leaves,u32 * num_dx_entries,struct buffer_head * dirent_bh)2592 static int ocfs2_dx_dir_index_block(struct inode *dir,
2593 handle_t *handle,
2594 struct buffer_head **dx_leaves,
2595 int num_dx_leaves,
2596 u32 *num_dx_entries,
2597 struct buffer_head *dirent_bh)
2598 {
2599 int ret = 0, namelen, i;
2600 char *de_buf, *limit;
2601 struct ocfs2_dir_entry *de;
2602 struct buffer_head *dx_leaf_bh;
2603 struct ocfs2_dx_hinfo hinfo;
2604 u64 dirent_blk = dirent_bh->b_blocknr;
2605
2606 de_buf = dirent_bh->b_data;
2607 limit = de_buf + dir->i_sb->s_blocksize;
2608
2609 while (de_buf < limit) {
2610 de = (struct ocfs2_dir_entry *)de_buf;
2611
2612 namelen = de->name_len;
2613 if (!namelen || !de->inode)
2614 goto inc;
2615
2616 ocfs2_dx_dir_name_hash(dir, de->name, namelen, &hinfo);
2617
2618 i = ocfs2_dx_dir_hash_idx(OCFS2_SB(dir->i_sb), &hinfo);
2619 dx_leaf_bh = dx_leaves[i];
2620
2621 ret = __ocfs2_dx_dir_leaf_insert(dir, handle, &hinfo,
2622 dirent_blk, dx_leaf_bh);
2623 if (ret) {
2624 mlog_errno(ret);
2625 goto out;
2626 }
2627
2628 *num_dx_entries = *num_dx_entries + 1;
2629
2630 inc:
2631 de_buf += le16_to_cpu(de->rec_len);
2632 }
2633
2634 out:
2635 return ret;
2636 }
2637
2638 /*
2639 * XXX: This expects dx_root_bh to already be part of the transaction.
2640 */
ocfs2_dx_dir_index_root_block(struct inode * dir,struct buffer_head * dx_root_bh,struct buffer_head * dirent_bh)2641 static void ocfs2_dx_dir_index_root_block(struct inode *dir,
2642 struct buffer_head *dx_root_bh,
2643 struct buffer_head *dirent_bh)
2644 {
2645 char *de_buf, *limit;
2646 struct ocfs2_dx_root_block *dx_root;
2647 struct ocfs2_dir_entry *de;
2648 struct ocfs2_dx_hinfo hinfo;
2649 u64 dirent_blk = dirent_bh->b_blocknr;
2650
2651 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
2652
2653 de_buf = dirent_bh->b_data;
2654 limit = de_buf + dir->i_sb->s_blocksize;
2655
2656 while (de_buf < limit) {
2657 de = (struct ocfs2_dir_entry *)de_buf;
2658
2659 if (!de->name_len || !de->inode)
2660 goto inc;
2661
2662 ocfs2_dx_dir_name_hash(dir, de->name, de->name_len, &hinfo);
2663
2664 trace_ocfs2_dx_dir_index_root_block(
2665 (unsigned long long)dir->i_ino,
2666 hinfo.major_hash, hinfo.minor_hash,
2667 de->name_len, de->name,
2668 le16_to_cpu(dx_root->dr_entries.de_num_used));
2669
2670 ocfs2_dx_entry_list_insert(&dx_root->dr_entries, &hinfo,
2671 dirent_blk);
2672
2673 le32_add_cpu(&dx_root->dr_num_entries, 1);
2674 inc:
2675 de_buf += le16_to_cpu(de->rec_len);
2676 }
2677 }
2678
2679 /*
2680 * Count the number of inline directory entries in di_bh and compare
2681 * them against the number of entries we can hold in an inline dx root
2682 * block.
2683 */
ocfs2_new_dx_should_be_inline(struct inode * dir,struct buffer_head * di_bh)2684 static int ocfs2_new_dx_should_be_inline(struct inode *dir,
2685 struct buffer_head *di_bh)
2686 {
2687 int dirent_count = 0;
2688 char *de_buf, *limit;
2689 struct ocfs2_dir_entry *de;
2690 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2691
2692 de_buf = di->id2.i_data.id_data;
2693 limit = de_buf + i_size_read(dir);
2694
2695 while (de_buf < limit) {
2696 de = (struct ocfs2_dir_entry *)de_buf;
2697
2698 if (de->name_len && de->inode)
2699 dirent_count++;
2700
2701 de_buf += le16_to_cpu(de->rec_len);
2702 }
2703
2704 /* We are careful to leave room for one extra record. */
2705 return dirent_count < ocfs2_dx_entries_per_root(dir->i_sb);
2706 }
2707
2708 /*
2709 * Expand rec_len of the rightmost dirent in a directory block so that it
2710 * contains the end of our valid space for dirents. We do this during
2711 * expansion from an inline directory to one with extents. The first dir block
2712 * in that case is taken from the inline data portion of the inode block.
2713 *
2714 * This will also return the largest amount of contiguous space for a dirent
2715 * in the block. That value is *not* necessarily the last dirent, even after
2716 * expansion. The directory indexing code wants this value for free space
2717 * accounting. We do this here since we're already walking the entire dir
2718 * block.
2719 *
2720 * We add the dir trailer if this filesystem wants it.
2721 */
ocfs2_expand_last_dirent(char * start,unsigned int old_size,struct inode * dir)2722 static unsigned int ocfs2_expand_last_dirent(char *start, unsigned int old_size,
2723 struct inode *dir)
2724 {
2725 struct super_block *sb = dir->i_sb;
2726 struct ocfs2_dir_entry *de;
2727 struct ocfs2_dir_entry *prev_de;
2728 char *de_buf, *limit;
2729 unsigned int new_size = sb->s_blocksize;
2730 unsigned int bytes, this_hole;
2731 unsigned int largest_hole = 0;
2732
2733 if (ocfs2_new_dir_wants_trailer(dir))
2734 new_size = ocfs2_dir_trailer_blk_off(sb);
2735
2736 bytes = new_size - old_size;
2737
2738 limit = start + old_size;
2739 de_buf = start;
2740 de = (struct ocfs2_dir_entry *)de_buf;
2741 do {
2742 this_hole = ocfs2_figure_dirent_hole(de);
2743 if (this_hole > largest_hole)
2744 largest_hole = this_hole;
2745
2746 prev_de = de;
2747 de_buf += le16_to_cpu(de->rec_len);
2748 de = (struct ocfs2_dir_entry *)de_buf;
2749 } while (de_buf < limit);
2750
2751 le16_add_cpu(&prev_de->rec_len, bytes);
2752
2753 /* We need to double check this after modification of the final
2754 * dirent. */
2755 this_hole = ocfs2_figure_dirent_hole(prev_de);
2756 if (this_hole > largest_hole)
2757 largest_hole = this_hole;
2758
2759 if (largest_hole >= OCFS2_DIR_MIN_REC_LEN)
2760 return largest_hole;
2761 return 0;
2762 }
2763
2764 /*
2765 * We allocate enough clusters to fulfill "blocks_wanted", but set
2766 * i_size to exactly one block. Ocfs2_extend_dir() will handle the
2767 * rest automatically for us.
2768 *
2769 * *first_block_bh is a pointer to the 1st data block allocated to the
2770 * directory.
2771 */
ocfs2_expand_inline_dir(struct inode * dir,struct buffer_head * di_bh,unsigned int blocks_wanted,struct ocfs2_dir_lookup_result * lookup,struct buffer_head ** first_block_bh)2772 static int ocfs2_expand_inline_dir(struct inode *dir, struct buffer_head *di_bh,
2773 unsigned int blocks_wanted,
2774 struct ocfs2_dir_lookup_result *lookup,
2775 struct buffer_head **first_block_bh)
2776 {
2777 u32 alloc, dx_alloc, bit_off, len, num_dx_entries = 0;
2778 struct super_block *sb = dir->i_sb;
2779 int ret, i, num_dx_leaves = 0, dx_inline = 0,
2780 credits = ocfs2_inline_to_extents_credits(sb);
2781 u64 dx_insert_blkno, blkno,
2782 bytes = blocks_wanted << sb->s_blocksize_bits;
2783 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2784 struct ocfs2_inode_info *oi = OCFS2_I(dir);
2785 struct ocfs2_alloc_context *data_ac = NULL;
2786 struct ocfs2_alloc_context *meta_ac = NULL;
2787 struct buffer_head *dirdata_bh = NULL;
2788 struct buffer_head *dx_root_bh = NULL;
2789 struct buffer_head **dx_leaves = NULL;
2790 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2791 handle_t *handle;
2792 struct ocfs2_extent_tree et;
2793 struct ocfs2_extent_tree dx_et;
2794 int did_quota = 0, bytes_allocated = 0;
2795
2796 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(dir), di_bh);
2797
2798 alloc = ocfs2_clusters_for_bytes(sb, bytes);
2799 dx_alloc = 0;
2800
2801 down_write(&oi->ip_alloc_sem);
2802
2803 if (ocfs2_supports_indexed_dirs(osb)) {
2804 credits += ocfs2_add_dir_index_credits(sb);
2805
2806 dx_inline = ocfs2_new_dx_should_be_inline(dir, di_bh);
2807 if (!dx_inline) {
2808 /* Add one more cluster for an index leaf */
2809 dx_alloc++;
2810 dx_leaves = ocfs2_dx_dir_kmalloc_leaves(sb,
2811 &num_dx_leaves);
2812 if (!dx_leaves) {
2813 ret = -ENOMEM;
2814 mlog_errno(ret);
2815 goto out;
2816 }
2817 }
2818
2819 /* This gets us the dx_root */
2820 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
2821 if (ret) {
2822 mlog_errno(ret);
2823 goto out;
2824 }
2825 }
2826
2827 /*
2828 * We should never need more than 2 clusters for the unindexed
2829 * tree - maximum dirent size is far less than one block. In
2830 * fact, the only time we'd need more than one cluster is if
2831 * blocksize == clustersize and the dirent won't fit in the
2832 * extra space that the expansion to a single block gives. As
2833 * of today, that only happens on 4k/4k file systems.
2834 */
2835 BUG_ON(alloc > 2);
2836
2837 ret = ocfs2_reserve_clusters(osb, alloc + dx_alloc, &data_ac);
2838 if (ret) {
2839 mlog_errno(ret);
2840 goto out;
2841 }
2842
2843 /*
2844 * Prepare for worst case allocation scenario of two separate
2845 * extents in the unindexed tree.
2846 */
2847 if (alloc == 2)
2848 credits += OCFS2_SUBALLOC_ALLOC;
2849
2850 handle = ocfs2_start_trans(osb, credits);
2851 if (IS_ERR(handle)) {
2852 ret = PTR_ERR(handle);
2853 mlog_errno(ret);
2854 goto out;
2855 }
2856
2857 ret = dquot_alloc_space_nodirty(dir,
2858 ocfs2_clusters_to_bytes(osb->sb, alloc + dx_alloc));
2859 if (ret)
2860 goto out_commit;
2861 did_quota = 1;
2862
2863 if (ocfs2_supports_indexed_dirs(osb) && !dx_inline) {
2864 /*
2865 * Allocate our index cluster first, to maximize the
2866 * possibility that unindexed leaves grow
2867 * contiguously.
2868 */
2869 ret = __ocfs2_dx_dir_new_cluster(dir, 0, handle, data_ac,
2870 dx_leaves, num_dx_leaves,
2871 &dx_insert_blkno);
2872 if (ret) {
2873 mlog_errno(ret);
2874 goto out_commit;
2875 }
2876 bytes_allocated += ocfs2_clusters_to_bytes(dir->i_sb, 1);
2877 }
2878
2879 /*
2880 * Try to claim as many clusters as the bitmap can give though
2881 * if we only get one now, that's enough to continue. The rest
2882 * will be claimed after the conversion to extents.
2883 */
2884 if (ocfs2_dir_resv_allowed(osb))
2885 data_ac->ac_resv = &oi->ip_la_data_resv;
2886 ret = ocfs2_claim_clusters(handle, data_ac, 1, &bit_off, &len);
2887 if (ret) {
2888 mlog_errno(ret);
2889 goto out_commit;
2890 }
2891 bytes_allocated += ocfs2_clusters_to_bytes(dir->i_sb, 1);
2892
2893 /*
2894 * Operations are carefully ordered so that we set up the new
2895 * data block first. The conversion from inline data to
2896 * extents follows.
2897 */
2898 blkno = ocfs2_clusters_to_blocks(dir->i_sb, bit_off);
2899 dirdata_bh = sb_getblk(sb, blkno);
2900 if (!dirdata_bh) {
2901 ret = -ENOMEM;
2902 mlog_errno(ret);
2903 goto out_commit;
2904 }
2905
2906 ocfs2_set_new_buffer_uptodate(INODE_CACHE(dir), dirdata_bh);
2907
2908 ret = ocfs2_journal_access_db(handle, INODE_CACHE(dir), dirdata_bh,
2909 OCFS2_JOURNAL_ACCESS_CREATE);
2910 if (ret) {
2911 mlog_errno(ret);
2912 goto out_commit;
2913 }
2914
2915 memcpy(dirdata_bh->b_data, di->id2.i_data.id_data, i_size_read(dir));
2916 memset(dirdata_bh->b_data + i_size_read(dir), 0,
2917 sb->s_blocksize - i_size_read(dir));
2918 i = ocfs2_expand_last_dirent(dirdata_bh->b_data, i_size_read(dir), dir);
2919 if (ocfs2_new_dir_wants_trailer(dir)) {
2920 /*
2921 * Prepare the dir trailer up front. It will otherwise look
2922 * like a valid dirent. Even if inserting the index fails
2923 * (unlikely), then all we'll have done is given first dir
2924 * block a small amount of fragmentation.
2925 */
2926 ocfs2_init_dir_trailer(dir, dirdata_bh, i);
2927 }
2928
2929 ocfs2_update_inode_fsync_trans(handle, dir, 1);
2930 ocfs2_journal_dirty(handle, dirdata_bh);
2931
2932 if (ocfs2_supports_indexed_dirs(osb) && !dx_inline) {
2933 /*
2934 * Dx dirs with an external cluster need to do this up
2935 * front. Inline dx root's get handled later, after
2936 * we've allocated our root block. We get passed back
2937 * a total number of items so that dr_num_entries can
2938 * be correctly set once the dx_root has been
2939 * allocated.
2940 */
2941 ret = ocfs2_dx_dir_index_block(dir, handle, dx_leaves,
2942 num_dx_leaves, &num_dx_entries,
2943 dirdata_bh);
2944 if (ret) {
2945 mlog_errno(ret);
2946 goto out_commit;
2947 }
2948 }
2949
2950 /*
2951 * Set extent, i_size, etc on the directory. After this, the
2952 * inode should contain the same exact dirents as before and
2953 * be fully accessible from system calls.
2954 *
2955 * We let the later dirent insert modify c/mtime - to the user
2956 * the data hasn't changed.
2957 */
2958 ret = ocfs2_journal_access_di(handle, INODE_CACHE(dir), di_bh,
2959 OCFS2_JOURNAL_ACCESS_CREATE);
2960 if (ret) {
2961 mlog_errno(ret);
2962 goto out_commit;
2963 }
2964
2965 spin_lock(&oi->ip_lock);
2966 oi->ip_dyn_features &= ~OCFS2_INLINE_DATA_FL;
2967 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2968 spin_unlock(&oi->ip_lock);
2969
2970 ocfs2_dinode_new_extent_list(dir, di);
2971
2972 i_size_write(dir, sb->s_blocksize);
2973 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
2974
2975 di->i_size = cpu_to_le64(sb->s_blocksize);
2976 di->i_ctime = di->i_mtime = cpu_to_le64(inode_get_ctime_sec(dir));
2977 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode_get_ctime_nsec(dir));
2978 ocfs2_update_inode_fsync_trans(handle, dir, 1);
2979
2980 /*
2981 * This should never fail as our extent list is empty and all
2982 * related blocks have been journaled already.
2983 */
2984 ret = ocfs2_insert_extent(handle, &et, 0, blkno, len,
2985 0, NULL);
2986 if (ret) {
2987 mlog_errno(ret);
2988 goto out_commit;
2989 }
2990
2991 /*
2992 * Set i_blocks after the extent insert for the most up to
2993 * date ip_clusters value.
2994 */
2995 dir->i_blocks = ocfs2_inode_sector_count(dir);
2996
2997 ocfs2_journal_dirty(handle, di_bh);
2998
2999 if (ocfs2_supports_indexed_dirs(osb)) {
3000 ret = ocfs2_dx_dir_attach_index(osb, handle, dir, di_bh,
3001 dirdata_bh, meta_ac, dx_inline,
3002 num_dx_entries, &dx_root_bh);
3003 if (ret) {
3004 mlog_errno(ret);
3005 goto out_commit;
3006 }
3007
3008 if (dx_inline) {
3009 ocfs2_dx_dir_index_root_block(dir, dx_root_bh,
3010 dirdata_bh);
3011 } else {
3012 ocfs2_init_dx_root_extent_tree(&dx_et,
3013 INODE_CACHE(dir),
3014 dx_root_bh);
3015 ret = ocfs2_insert_extent(handle, &dx_et, 0,
3016 dx_insert_blkno, 1, 0, NULL);
3017 if (ret)
3018 mlog_errno(ret);
3019 }
3020 }
3021
3022 /*
3023 * We asked for two clusters, but only got one in the 1st
3024 * pass. Claim the 2nd cluster as a separate extent.
3025 */
3026 if (alloc > len) {
3027 ret = ocfs2_claim_clusters(handle, data_ac, 1, &bit_off,
3028 &len);
3029 if (ret) {
3030 mlog_errno(ret);
3031 goto out_commit;
3032 }
3033 blkno = ocfs2_clusters_to_blocks(dir->i_sb, bit_off);
3034
3035 ret = ocfs2_insert_extent(handle, &et, 1,
3036 blkno, len, 0, NULL);
3037 if (ret) {
3038 mlog_errno(ret);
3039 goto out_commit;
3040 }
3041 bytes_allocated += ocfs2_clusters_to_bytes(dir->i_sb, 1);
3042 }
3043
3044 *first_block_bh = dirdata_bh;
3045 dirdata_bh = NULL;
3046 if (ocfs2_supports_indexed_dirs(osb)) {
3047 unsigned int off;
3048
3049 if (!dx_inline) {
3050 /*
3051 * We need to return the correct block within the
3052 * cluster which should hold our entry.
3053 */
3054 off = ocfs2_dx_dir_hash_idx(osb,
3055 &lookup->dl_hinfo);
3056 get_bh(dx_leaves[off]);
3057 lookup->dl_dx_leaf_bh = dx_leaves[off];
3058 }
3059 lookup->dl_dx_root_bh = dx_root_bh;
3060 dx_root_bh = NULL;
3061 }
3062
3063 out_commit:
3064 if (ret < 0 && did_quota)
3065 dquot_free_space_nodirty(dir, bytes_allocated);
3066
3067 ocfs2_commit_trans(osb, handle);
3068
3069 out:
3070 up_write(&oi->ip_alloc_sem);
3071 if (data_ac)
3072 ocfs2_free_alloc_context(data_ac);
3073 if (meta_ac)
3074 ocfs2_free_alloc_context(meta_ac);
3075
3076 if (dx_leaves) {
3077 for (i = 0; i < num_dx_leaves; i++)
3078 brelse(dx_leaves[i]);
3079 kfree(dx_leaves);
3080 }
3081
3082 brelse(dirdata_bh);
3083 brelse(dx_root_bh);
3084
3085 return ret;
3086 }
3087
3088 /* returns a bh of the 1st new block in the allocation. */
ocfs2_do_extend_dir(struct super_block * sb,handle_t * handle,struct inode * dir,struct buffer_head * parent_fe_bh,struct ocfs2_alloc_context * data_ac,struct ocfs2_alloc_context * meta_ac,struct buffer_head ** new_bh)3089 static int ocfs2_do_extend_dir(struct super_block *sb,
3090 handle_t *handle,
3091 struct inode *dir,
3092 struct buffer_head *parent_fe_bh,
3093 struct ocfs2_alloc_context *data_ac,
3094 struct ocfs2_alloc_context *meta_ac,
3095 struct buffer_head **new_bh)
3096 {
3097 int status;
3098 int extend, did_quota = 0;
3099 u64 p_blkno, v_blkno;
3100
3101 spin_lock(&OCFS2_I(dir)->ip_lock);
3102 extend = (i_size_read(dir) == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters));
3103 spin_unlock(&OCFS2_I(dir)->ip_lock);
3104
3105 if (extend) {
3106 u32 offset = OCFS2_I(dir)->ip_clusters;
3107
3108 status = dquot_alloc_space_nodirty(dir,
3109 ocfs2_clusters_to_bytes(sb, 1));
3110 if (status)
3111 goto bail;
3112 did_quota = 1;
3113
3114 status = ocfs2_add_inode_data(OCFS2_SB(sb), dir, &offset,
3115 1, 0, parent_fe_bh, handle,
3116 data_ac, meta_ac, NULL);
3117 BUG_ON(status == -EAGAIN);
3118 if (status < 0) {
3119 mlog_errno(status);
3120 goto bail;
3121 }
3122 }
3123
3124 v_blkno = ocfs2_blocks_for_bytes(sb, i_size_read(dir));
3125 status = ocfs2_extent_map_get_blocks(dir, v_blkno, &p_blkno, NULL, NULL);
3126 if (status < 0) {
3127 mlog_errno(status);
3128 goto bail;
3129 }
3130
3131 *new_bh = sb_getblk(sb, p_blkno);
3132 if (!*new_bh) {
3133 status = -ENOMEM;
3134 mlog_errno(status);
3135 goto bail;
3136 }
3137 status = 0;
3138 bail:
3139 if (did_quota && status < 0)
3140 dquot_free_space_nodirty(dir, ocfs2_clusters_to_bytes(sb, 1));
3141 return status;
3142 }
3143
3144 /*
3145 * Assumes you already have a cluster lock on the directory.
3146 *
3147 * 'blocks_wanted' is only used if we have an inline directory which
3148 * is to be turned into an extent based one. The size of the dirent to
3149 * insert might be larger than the space gained by growing to just one
3150 * block, so we may have to grow the inode by two blocks in that case.
3151 *
3152 * If the directory is already indexed, dx_root_bh must be provided.
3153 */
ocfs2_extend_dir(struct ocfs2_super * osb,struct inode * dir,struct buffer_head * parent_fe_bh,unsigned int blocks_wanted,struct ocfs2_dir_lookup_result * lookup,struct buffer_head ** new_de_bh)3154 static int ocfs2_extend_dir(struct ocfs2_super *osb,
3155 struct inode *dir,
3156 struct buffer_head *parent_fe_bh,
3157 unsigned int blocks_wanted,
3158 struct ocfs2_dir_lookup_result *lookup,
3159 struct buffer_head **new_de_bh)
3160 {
3161 int status = 0;
3162 int credits, num_free_extents, drop_alloc_sem = 0;
3163 loff_t dir_i_size;
3164 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
3165 struct ocfs2_extent_list *el = &fe->id2.i_list;
3166 struct ocfs2_alloc_context *data_ac = NULL;
3167 struct ocfs2_alloc_context *meta_ac = NULL;
3168 handle_t *handle = NULL;
3169 struct buffer_head *new_bh = NULL;
3170 struct ocfs2_dir_entry * de;
3171 struct super_block *sb = osb->sb;
3172 struct ocfs2_extent_tree et;
3173 struct buffer_head *dx_root_bh = lookup->dl_dx_root_bh;
3174
3175 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
3176 /*
3177 * This would be a code error as an inline directory should
3178 * never have an index root.
3179 */
3180 BUG_ON(dx_root_bh);
3181
3182 status = ocfs2_expand_inline_dir(dir, parent_fe_bh,
3183 blocks_wanted, lookup,
3184 &new_bh);
3185 if (status) {
3186 mlog_errno(status);
3187 goto bail;
3188 }
3189
3190 /* Expansion from inline to an indexed directory will
3191 * have given us this. */
3192 dx_root_bh = lookup->dl_dx_root_bh;
3193
3194 if (blocks_wanted == 1) {
3195 /*
3196 * If the new dirent will fit inside the space
3197 * created by pushing out to one block, then
3198 * we can complete the operation
3199 * here. Otherwise we have to expand i_size
3200 * and format the 2nd block below.
3201 */
3202 BUG_ON(new_bh == NULL);
3203 goto bail_bh;
3204 }
3205
3206 /*
3207 * Get rid of 'new_bh' - we want to format the 2nd
3208 * data block and return that instead.
3209 */
3210 brelse(new_bh);
3211 new_bh = NULL;
3212
3213 down_write(&OCFS2_I(dir)->ip_alloc_sem);
3214 drop_alloc_sem = 1;
3215 dir_i_size = i_size_read(dir);
3216 credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS;
3217 goto do_extend;
3218 }
3219
3220 down_write(&OCFS2_I(dir)->ip_alloc_sem);
3221 drop_alloc_sem = 1;
3222 dir_i_size = i_size_read(dir);
3223 trace_ocfs2_extend_dir((unsigned long long)OCFS2_I(dir)->ip_blkno,
3224 dir_i_size);
3225
3226 /* dir->i_size is always block aligned. */
3227 spin_lock(&OCFS2_I(dir)->ip_lock);
3228 if (dir_i_size == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters)) {
3229 spin_unlock(&OCFS2_I(dir)->ip_lock);
3230 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(dir),
3231 parent_fe_bh);
3232 num_free_extents = ocfs2_num_free_extents(&et);
3233 if (num_free_extents < 0) {
3234 status = num_free_extents;
3235 mlog_errno(status);
3236 goto bail;
3237 }
3238
3239 if (!num_free_extents) {
3240 status = ocfs2_reserve_new_metadata(osb, el, &meta_ac);
3241 if (status < 0) {
3242 if (status != -ENOSPC)
3243 mlog_errno(status);
3244 goto bail;
3245 }
3246 }
3247
3248 status = ocfs2_reserve_clusters(osb, 1, &data_ac);
3249 if (status < 0) {
3250 if (status != -ENOSPC)
3251 mlog_errno(status);
3252 goto bail;
3253 }
3254
3255 if (ocfs2_dir_resv_allowed(osb))
3256 data_ac->ac_resv = &OCFS2_I(dir)->ip_la_data_resv;
3257
3258 credits = ocfs2_calc_extend_credits(sb, el);
3259 } else {
3260 spin_unlock(&OCFS2_I(dir)->ip_lock);
3261 credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS;
3262 }
3263
3264 do_extend:
3265 if (ocfs2_dir_indexed(dir))
3266 credits++; /* For attaching the new dirent block to the
3267 * dx_root */
3268
3269 handle = ocfs2_start_trans(osb, credits);
3270 if (IS_ERR(handle)) {
3271 status = PTR_ERR(handle);
3272 handle = NULL;
3273 mlog_errno(status);
3274 goto bail;
3275 }
3276
3277 status = ocfs2_do_extend_dir(osb->sb, handle, dir, parent_fe_bh,
3278 data_ac, meta_ac, &new_bh);
3279 if (status < 0) {
3280 mlog_errno(status);
3281 goto bail;
3282 }
3283
3284 ocfs2_set_new_buffer_uptodate(INODE_CACHE(dir), new_bh);
3285
3286 status = ocfs2_journal_access_db(handle, INODE_CACHE(dir), new_bh,
3287 OCFS2_JOURNAL_ACCESS_CREATE);
3288 if (status < 0) {
3289 mlog_errno(status);
3290 goto bail;
3291 }
3292 memset(new_bh->b_data, 0, sb->s_blocksize);
3293
3294 de = (struct ocfs2_dir_entry *) new_bh->b_data;
3295 de->inode = 0;
3296 if (ocfs2_supports_dir_trailer(dir)) {
3297 de->rec_len = cpu_to_le16(ocfs2_dir_trailer_blk_off(sb));
3298
3299 ocfs2_init_dir_trailer(dir, new_bh, le16_to_cpu(de->rec_len));
3300
3301 if (ocfs2_dir_indexed(dir)) {
3302 status = ocfs2_dx_dir_link_trailer(dir, handle,
3303 dx_root_bh, new_bh);
3304 if (status) {
3305 mlog_errno(status);
3306 goto bail;
3307 }
3308 }
3309 } else {
3310 de->rec_len = cpu_to_le16(sb->s_blocksize);
3311 }
3312 ocfs2_update_inode_fsync_trans(handle, dir, 1);
3313 ocfs2_journal_dirty(handle, new_bh);
3314
3315 dir_i_size += dir->i_sb->s_blocksize;
3316 i_size_write(dir, dir_i_size);
3317 dir->i_blocks = ocfs2_inode_sector_count(dir);
3318 status = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
3319 if (status < 0) {
3320 mlog_errno(status);
3321 goto bail;
3322 }
3323
3324 bail_bh:
3325 *new_de_bh = new_bh;
3326 get_bh(*new_de_bh);
3327 bail:
3328 if (handle)
3329 ocfs2_commit_trans(osb, handle);
3330 if (drop_alloc_sem)
3331 up_write(&OCFS2_I(dir)->ip_alloc_sem);
3332
3333 if (data_ac)
3334 ocfs2_free_alloc_context(data_ac);
3335 if (meta_ac)
3336 ocfs2_free_alloc_context(meta_ac);
3337
3338 brelse(new_bh);
3339
3340 return status;
3341 }
3342
ocfs2_find_dir_space_id(struct inode * dir,struct buffer_head * di_bh,const char * name,int namelen,struct buffer_head ** ret_de_bh,unsigned int * blocks_wanted)3343 static int ocfs2_find_dir_space_id(struct inode *dir, struct buffer_head *di_bh,
3344 const char *name, int namelen,
3345 struct buffer_head **ret_de_bh,
3346 unsigned int *blocks_wanted)
3347 {
3348 int ret;
3349 struct super_block *sb = dir->i_sb;
3350 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3351 struct ocfs2_dir_entry *de, *last_de = NULL;
3352 char *first_de, *de_buf, *limit;
3353 unsigned long offset = 0;
3354 unsigned int rec_len, new_rec_len, free_space;
3355
3356 /*
3357 * This calculates how many free bytes we'd have in block zero, should
3358 * this function force expansion to an extent tree.
3359 */
3360 if (ocfs2_new_dir_wants_trailer(dir))
3361 free_space = ocfs2_dir_trailer_blk_off(sb) - i_size_read(dir);
3362 else
3363 free_space = dir->i_sb->s_blocksize - i_size_read(dir);
3364
3365 first_de = di->id2.i_data.id_data;
3366 de_buf = first_de;
3367 limit = de_buf + i_size_read(dir);
3368 rec_len = OCFS2_DIR_REC_LEN(namelen);
3369
3370 while (de_buf < limit) {
3371 de = (struct ocfs2_dir_entry *)de_buf;
3372
3373 if (!ocfs2_check_dir_entry(dir, de, di_bh, first_de,
3374 i_size_read(dir), offset)) {
3375 ret = -ENOENT;
3376 goto out;
3377 }
3378 if (ocfs2_match(namelen, name, de)) {
3379 ret = -EEXIST;
3380 goto out;
3381 }
3382 /*
3383 * No need to check for a trailing dirent record here as
3384 * they're not used for inline dirs.
3385 */
3386
3387 if (ocfs2_dirent_would_fit(de, rec_len)) {
3388 /* Ok, we found a spot. Return this bh and let
3389 * the caller actually fill it in. */
3390 *ret_de_bh = di_bh;
3391 get_bh(*ret_de_bh);
3392 ret = 0;
3393 goto out;
3394 }
3395
3396 last_de = de;
3397 de_buf += le16_to_cpu(de->rec_len);
3398 offset += le16_to_cpu(de->rec_len);
3399 }
3400
3401 /*
3402 * We're going to require expansion of the directory - figure
3403 * out how many blocks we'll need so that a place for the
3404 * dirent can be found.
3405 */
3406 *blocks_wanted = 1;
3407 new_rec_len = le16_to_cpu(last_de->rec_len) + free_space;
3408 if (new_rec_len < (rec_len + OCFS2_DIR_REC_LEN(last_de->name_len)))
3409 *blocks_wanted = 2;
3410
3411 ret = -ENOSPC;
3412 out:
3413 return ret;
3414 }
3415
ocfs2_find_dir_space_el(struct inode * dir,const char * name,int namelen,struct buffer_head ** ret_de_bh)3416 static int ocfs2_find_dir_space_el(struct inode *dir, const char *name,
3417 int namelen, struct buffer_head **ret_de_bh)
3418 {
3419 unsigned long offset;
3420 struct buffer_head *bh = NULL;
3421 unsigned short rec_len;
3422 struct ocfs2_dir_entry *de;
3423 struct super_block *sb = dir->i_sb;
3424 int status;
3425 int blocksize = dir->i_sb->s_blocksize;
3426
3427 status = ocfs2_read_dir_block(dir, 0, &bh, 0);
3428 if (status)
3429 goto bail;
3430
3431 rec_len = OCFS2_DIR_REC_LEN(namelen);
3432 offset = 0;
3433 de = (struct ocfs2_dir_entry *) bh->b_data;
3434 while (1) {
3435 if ((char *)de >= sb->s_blocksize + bh->b_data) {
3436 brelse(bh);
3437 bh = NULL;
3438
3439 if (i_size_read(dir) <= offset) {
3440 /*
3441 * Caller will have to expand this
3442 * directory.
3443 */
3444 status = -ENOSPC;
3445 goto bail;
3446 }
3447 status = ocfs2_read_dir_block(dir,
3448 offset >> sb->s_blocksize_bits,
3449 &bh, 0);
3450 if (status)
3451 goto bail;
3452
3453 /* move to next block */
3454 de = (struct ocfs2_dir_entry *) bh->b_data;
3455 }
3456 if (!ocfs2_check_dir_entry(dir, de, bh, bh->b_data, blocksize,
3457 offset)) {
3458 status = -ENOENT;
3459 goto bail;
3460 }
3461 if (ocfs2_match(namelen, name, de)) {
3462 status = -EEXIST;
3463 goto bail;
3464 }
3465
3466 if (ocfs2_skip_dir_trailer(dir, de, offset % blocksize,
3467 blocksize))
3468 goto next;
3469
3470 if (ocfs2_dirent_would_fit(de, rec_len)) {
3471 /* Ok, we found a spot. Return this bh and let
3472 * the caller actually fill it in. */
3473 *ret_de_bh = bh;
3474 get_bh(*ret_de_bh);
3475 status = 0;
3476 goto bail;
3477 }
3478 next:
3479 offset += le16_to_cpu(de->rec_len);
3480 de = (struct ocfs2_dir_entry *)((char *) de + le16_to_cpu(de->rec_len));
3481 }
3482
3483 bail:
3484 brelse(bh);
3485 if (status)
3486 mlog_errno(status);
3487
3488 return status;
3489 }
3490
dx_leaf_sort_cmp(const void * a,const void * b)3491 static int dx_leaf_sort_cmp(const void *a, const void *b)
3492 {
3493 const struct ocfs2_dx_entry *entry1 = a;
3494 const struct ocfs2_dx_entry *entry2 = b;
3495 u32 major_hash1 = le32_to_cpu(entry1->dx_major_hash);
3496 u32 major_hash2 = le32_to_cpu(entry2->dx_major_hash);
3497 u32 minor_hash1 = le32_to_cpu(entry1->dx_minor_hash);
3498 u32 minor_hash2 = le32_to_cpu(entry2->dx_minor_hash);
3499
3500 if (major_hash1 > major_hash2)
3501 return 1;
3502 if (major_hash1 < major_hash2)
3503 return -1;
3504
3505 /*
3506 * It is not strictly necessary to sort by minor
3507 */
3508 if (minor_hash1 > minor_hash2)
3509 return 1;
3510 if (minor_hash1 < minor_hash2)
3511 return -1;
3512 return 0;
3513 }
3514
ocfs2_dx_leaf_same_major(struct ocfs2_dx_leaf * dx_leaf)3515 static int ocfs2_dx_leaf_same_major(struct ocfs2_dx_leaf *dx_leaf)
3516 {
3517 struct ocfs2_dx_entry_list *dl_list = &dx_leaf->dl_list;
3518 int i, num = le16_to_cpu(dl_list->de_num_used);
3519
3520 for (i = 0; i < (num - 1); i++) {
3521 if (le32_to_cpu(dl_list->de_entries[i].dx_major_hash) !=
3522 le32_to_cpu(dl_list->de_entries[i + 1].dx_major_hash))
3523 return 0;
3524 }
3525
3526 return 1;
3527 }
3528
3529 /*
3530 * Find the optimal value to split this leaf on. This expects the leaf
3531 * entries to be in sorted order.
3532 *
3533 * leaf_cpos is the cpos of the leaf we're splitting. insert_hash is
3534 * the hash we want to insert.
3535 *
3536 * This function is only concerned with the major hash - that which
3537 * determines which cluster an item belongs to.
3538 */
ocfs2_dx_dir_find_leaf_split(struct ocfs2_dx_leaf * dx_leaf,u32 leaf_cpos,u32 insert_hash,u32 * split_hash)3539 static int ocfs2_dx_dir_find_leaf_split(struct ocfs2_dx_leaf *dx_leaf,
3540 u32 leaf_cpos, u32 insert_hash,
3541 u32 *split_hash)
3542 {
3543 struct ocfs2_dx_entry_list *dl_list = &dx_leaf->dl_list;
3544 int i, num_used = le16_to_cpu(dl_list->de_num_used);
3545 int allsame;
3546
3547 /*
3548 * There's a couple rare, but nasty corner cases we have to
3549 * check for here. All of them involve a leaf where all value
3550 * have the same hash, which is what we look for first.
3551 *
3552 * Most of the time, all of the above is false, and we simply
3553 * pick the median value for a split.
3554 */
3555 allsame = ocfs2_dx_leaf_same_major(dx_leaf);
3556 if (allsame) {
3557 u32 val = le32_to_cpu(dl_list->de_entries[0].dx_major_hash);
3558
3559 if (val == insert_hash) {
3560 /*
3561 * No matter where we would choose to split,
3562 * the new entry would want to occupy the same
3563 * block as these. Since there's no space left
3564 * in their existing block, we know there
3565 * won't be space after the split.
3566 */
3567 return -ENOSPC;
3568 }
3569
3570 if (val == leaf_cpos) {
3571 /*
3572 * Because val is the same as leaf_cpos (which
3573 * is the smallest value this leaf can have),
3574 * yet is not equal to insert_hash, then we
3575 * know that insert_hash *must* be larger than
3576 * val (and leaf_cpos). At least cpos+1 in value.
3577 *
3578 * We also know then, that there cannot be an
3579 * adjacent extent (otherwise we'd be looking
3580 * at it). Choosing this value gives us a
3581 * chance to get some contiguousness.
3582 */
3583 *split_hash = leaf_cpos + 1;
3584 return 0;
3585 }
3586
3587 if (val > insert_hash) {
3588 /*
3589 * val can not be the same as insert hash, and
3590 * also must be larger than leaf_cpos. Also,
3591 * we know that there can't be a leaf between
3592 * cpos and val, otherwise the entries with
3593 * hash 'val' would be there.
3594 */
3595 *split_hash = val;
3596 return 0;
3597 }
3598
3599 *split_hash = insert_hash;
3600 return 0;
3601 }
3602
3603 /*
3604 * Since the records are sorted and the checks above
3605 * guaranteed that not all records in this block are the same,
3606 * we simple travel forward, from the median, and pick the 1st
3607 * record whose value is larger than leaf_cpos.
3608 */
3609 for (i = (num_used / 2); i < num_used; i++)
3610 if (le32_to_cpu(dl_list->de_entries[i].dx_major_hash) >
3611 leaf_cpos)
3612 break;
3613
3614 BUG_ON(i == num_used); /* Should be impossible */
3615 *split_hash = le32_to_cpu(dl_list->de_entries[i].dx_major_hash);
3616 return 0;
3617 }
3618
3619 /*
3620 * Transfer all entries in orig_dx_leaves whose major hash is equal to or
3621 * larger than split_hash into new_dx_leaves. We use a temporary
3622 * buffer (tmp_dx_leaf) to make the changes to the original leaf blocks.
3623 *
3624 * Since the block offset inside a leaf (cluster) is a constant mask
3625 * of minor_hash, we can optimize - an item at block offset X within
3626 * the original cluster, will be at offset X within the new cluster.
3627 */
ocfs2_dx_dir_transfer_leaf(struct inode * dir,u32 split_hash,handle_t * handle,struct ocfs2_dx_leaf * tmp_dx_leaf,struct buffer_head ** orig_dx_leaves,struct buffer_head ** new_dx_leaves,int num_dx_leaves)3628 static void ocfs2_dx_dir_transfer_leaf(struct inode *dir, u32 split_hash,
3629 handle_t *handle,
3630 struct ocfs2_dx_leaf *tmp_dx_leaf,
3631 struct buffer_head **orig_dx_leaves,
3632 struct buffer_head **new_dx_leaves,
3633 int num_dx_leaves)
3634 {
3635 int i, j, num_used;
3636 u32 major_hash;
3637 struct ocfs2_dx_leaf *orig_dx_leaf, *new_dx_leaf;
3638 struct ocfs2_dx_entry_list *orig_list, *tmp_list;
3639 struct ocfs2_dx_entry *dx_entry;
3640
3641 tmp_list = &tmp_dx_leaf->dl_list;
3642
3643 for (i = 0; i < num_dx_leaves; i++) {
3644 orig_dx_leaf = (struct ocfs2_dx_leaf *) orig_dx_leaves[i]->b_data;
3645 orig_list = &orig_dx_leaf->dl_list;
3646 new_dx_leaf = (struct ocfs2_dx_leaf *) new_dx_leaves[i]->b_data;
3647
3648 num_used = le16_to_cpu(orig_list->de_num_used);
3649
3650 memcpy(tmp_dx_leaf, orig_dx_leaf, dir->i_sb->s_blocksize);
3651 tmp_list->de_num_used = cpu_to_le16(0);
3652 memset(&tmp_list->de_entries, 0, sizeof(*dx_entry)*num_used);
3653
3654 for (j = 0; j < num_used; j++) {
3655 dx_entry = &orig_list->de_entries[j];
3656 major_hash = le32_to_cpu(dx_entry->dx_major_hash);
3657 if (major_hash >= split_hash)
3658 ocfs2_dx_dir_leaf_insert_tail(new_dx_leaf,
3659 dx_entry);
3660 else
3661 ocfs2_dx_dir_leaf_insert_tail(tmp_dx_leaf,
3662 dx_entry);
3663 }
3664 memcpy(orig_dx_leaf, tmp_dx_leaf, dir->i_sb->s_blocksize);
3665
3666 ocfs2_journal_dirty(handle, orig_dx_leaves[i]);
3667 ocfs2_journal_dirty(handle, new_dx_leaves[i]);
3668 }
3669 }
3670
ocfs2_dx_dir_rebalance_credits(struct ocfs2_super * osb,struct ocfs2_dx_root_block * dx_root)3671 static int ocfs2_dx_dir_rebalance_credits(struct ocfs2_super *osb,
3672 struct ocfs2_dx_root_block *dx_root)
3673 {
3674 int credits = ocfs2_clusters_to_blocks(osb->sb, 3);
3675
3676 credits += ocfs2_calc_extend_credits(osb->sb, &dx_root->dr_list);
3677 credits += ocfs2_quota_trans_credits(osb->sb);
3678 return credits;
3679 }
3680
3681 /*
3682 * Find the median value in dx_leaf_bh and allocate a new leaf to move
3683 * half our entries into.
3684 */
ocfs2_dx_dir_rebalance(struct ocfs2_super * osb,struct inode * dir,struct buffer_head * dx_root_bh,struct buffer_head * dx_leaf_bh,struct ocfs2_dx_hinfo * hinfo,u32 leaf_cpos,u64 leaf_blkno)3685 static int ocfs2_dx_dir_rebalance(struct ocfs2_super *osb, struct inode *dir,
3686 struct buffer_head *dx_root_bh,
3687 struct buffer_head *dx_leaf_bh,
3688 struct ocfs2_dx_hinfo *hinfo, u32 leaf_cpos,
3689 u64 leaf_blkno)
3690 {
3691 struct ocfs2_dx_leaf *dx_leaf = (struct ocfs2_dx_leaf *)dx_leaf_bh->b_data;
3692 int credits, ret, i, num_used, did_quota = 0;
3693 u32 cpos, split_hash, insert_hash = hinfo->major_hash;
3694 u64 orig_leaves_start;
3695 int num_dx_leaves;
3696 struct buffer_head **orig_dx_leaves = NULL;
3697 struct buffer_head **new_dx_leaves = NULL;
3698 struct ocfs2_alloc_context *data_ac = NULL, *meta_ac = NULL;
3699 struct ocfs2_extent_tree et;
3700 handle_t *handle = NULL;
3701 struct ocfs2_dx_root_block *dx_root;
3702 struct ocfs2_dx_leaf *tmp_dx_leaf = NULL;
3703
3704 trace_ocfs2_dx_dir_rebalance((unsigned long long)OCFS2_I(dir)->ip_blkno,
3705 (unsigned long long)leaf_blkno,
3706 insert_hash);
3707
3708 ocfs2_init_dx_root_extent_tree(&et, INODE_CACHE(dir), dx_root_bh);
3709
3710 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
3711 /*
3712 * XXX: This is a rather large limit. We should use a more
3713 * realistic value.
3714 */
3715 if (le32_to_cpu(dx_root->dr_clusters) == UINT_MAX)
3716 return -ENOSPC;
3717
3718 num_used = le16_to_cpu(dx_leaf->dl_list.de_num_used);
3719 if (num_used < le16_to_cpu(dx_leaf->dl_list.de_count)) {
3720 mlog(ML_ERROR, "DX Dir: %llu, Asked to rebalance empty leaf: "
3721 "%llu, %d\n", (unsigned long long)OCFS2_I(dir)->ip_blkno,
3722 (unsigned long long)leaf_blkno, num_used);
3723 ret = -EIO;
3724 goto out;
3725 }
3726
3727 orig_dx_leaves = ocfs2_dx_dir_kmalloc_leaves(osb->sb, &num_dx_leaves);
3728 if (!orig_dx_leaves) {
3729 ret = -ENOMEM;
3730 mlog_errno(ret);
3731 goto out;
3732 }
3733
3734 new_dx_leaves = ocfs2_dx_dir_kmalloc_leaves(osb->sb, NULL);
3735 if (!new_dx_leaves) {
3736 ret = -ENOMEM;
3737 mlog_errno(ret);
3738 goto out;
3739 }
3740
3741 ret = ocfs2_lock_allocators(dir, &et, 1, 0, &data_ac, &meta_ac);
3742 if (ret) {
3743 if (ret != -ENOSPC)
3744 mlog_errno(ret);
3745 goto out;
3746 }
3747
3748 credits = ocfs2_dx_dir_rebalance_credits(osb, dx_root);
3749 handle = ocfs2_start_trans(osb, credits);
3750 if (IS_ERR(handle)) {
3751 ret = PTR_ERR(handle);
3752 handle = NULL;
3753 mlog_errno(ret);
3754 goto out;
3755 }
3756
3757 ret = dquot_alloc_space_nodirty(dir,
3758 ocfs2_clusters_to_bytes(dir->i_sb, 1));
3759 if (ret)
3760 goto out_commit;
3761 did_quota = 1;
3762
3763 ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir), dx_leaf_bh,
3764 OCFS2_JOURNAL_ACCESS_WRITE);
3765 if (ret) {
3766 mlog_errno(ret);
3767 goto out_commit;
3768 }
3769
3770 /*
3771 * This block is changing anyway, so we can sort it in place.
3772 */
3773 sort(dx_leaf->dl_list.de_entries, num_used,
3774 sizeof(struct ocfs2_dx_entry), dx_leaf_sort_cmp,
3775 NULL);
3776
3777 ocfs2_journal_dirty(handle, dx_leaf_bh);
3778
3779 ret = ocfs2_dx_dir_find_leaf_split(dx_leaf, leaf_cpos, insert_hash,
3780 &split_hash);
3781 if (ret) {
3782 mlog_errno(ret);
3783 goto out_commit;
3784 }
3785
3786 trace_ocfs2_dx_dir_rebalance_split(leaf_cpos, split_hash, insert_hash);
3787
3788 /*
3789 * We have to carefully order operations here. There are items
3790 * which want to be in the new cluster before insert, but in
3791 * order to put those items in the new cluster, we alter the
3792 * old cluster. A failure to insert gets nasty.
3793 *
3794 * So, start by reserving writes to the old
3795 * cluster. ocfs2_dx_dir_new_cluster will reserve writes on
3796 * the new cluster for us, before inserting it. The insert
3797 * won't happen if there's an error before that. Once the
3798 * insert is done then, we can transfer from one leaf into the
3799 * other without fear of hitting any error.
3800 */
3801
3802 /*
3803 * The leaf transfer wants some scratch space so that we don't
3804 * wind up doing a bunch of expensive memmove().
3805 */
3806 tmp_dx_leaf = kmalloc(osb->sb->s_blocksize, GFP_NOFS);
3807 if (!tmp_dx_leaf) {
3808 ret = -ENOMEM;
3809 mlog_errno(ret);
3810 goto out_commit;
3811 }
3812
3813 orig_leaves_start = ocfs2_block_to_cluster_start(dir->i_sb, leaf_blkno);
3814 ret = ocfs2_read_dx_leaves(dir, orig_leaves_start, num_dx_leaves,
3815 orig_dx_leaves);
3816 if (ret) {
3817 mlog_errno(ret);
3818 goto out_commit;
3819 }
3820
3821 cpos = split_hash;
3822 ret = ocfs2_dx_dir_new_cluster(dir, &et, cpos, handle,
3823 data_ac, meta_ac, new_dx_leaves,
3824 num_dx_leaves);
3825 if (ret) {
3826 mlog_errno(ret);
3827 goto out_commit;
3828 }
3829
3830 for (i = 0; i < num_dx_leaves; i++) {
3831 ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir),
3832 orig_dx_leaves[i],
3833 OCFS2_JOURNAL_ACCESS_WRITE);
3834 if (ret) {
3835 mlog_errno(ret);
3836 goto out_commit;
3837 }
3838
3839 ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir),
3840 new_dx_leaves[i],
3841 OCFS2_JOURNAL_ACCESS_WRITE);
3842 if (ret) {
3843 mlog_errno(ret);
3844 goto out_commit;
3845 }
3846 }
3847
3848 ocfs2_dx_dir_transfer_leaf(dir, split_hash, handle, tmp_dx_leaf,
3849 orig_dx_leaves, new_dx_leaves, num_dx_leaves);
3850
3851 out_commit:
3852 if (ret < 0 && did_quota)
3853 dquot_free_space_nodirty(dir,
3854 ocfs2_clusters_to_bytes(dir->i_sb, 1));
3855
3856 ocfs2_update_inode_fsync_trans(handle, dir, 1);
3857 ocfs2_commit_trans(osb, handle);
3858
3859 out:
3860 if (orig_dx_leaves || new_dx_leaves) {
3861 for (i = 0; i < num_dx_leaves; i++) {
3862 if (orig_dx_leaves)
3863 brelse(orig_dx_leaves[i]);
3864 if (new_dx_leaves)
3865 brelse(new_dx_leaves[i]);
3866 }
3867 kfree(orig_dx_leaves);
3868 kfree(new_dx_leaves);
3869 }
3870
3871 if (meta_ac)
3872 ocfs2_free_alloc_context(meta_ac);
3873 if (data_ac)
3874 ocfs2_free_alloc_context(data_ac);
3875
3876 kfree(tmp_dx_leaf);
3877 return ret;
3878 }
3879
ocfs2_find_dir_space_dx(struct ocfs2_super * osb,struct inode * dir,struct buffer_head * di_bh,struct buffer_head * dx_root_bh,const char * name,int namelen,struct ocfs2_dir_lookup_result * lookup)3880 static int ocfs2_find_dir_space_dx(struct ocfs2_super *osb, struct inode *dir,
3881 struct buffer_head *di_bh,
3882 struct buffer_head *dx_root_bh,
3883 const char *name, int namelen,
3884 struct ocfs2_dir_lookup_result *lookup)
3885 {
3886 int ret, rebalanced = 0;
3887 struct ocfs2_dx_root_block *dx_root;
3888 struct buffer_head *dx_leaf_bh = NULL;
3889 struct ocfs2_dx_leaf *dx_leaf;
3890 u64 blkno;
3891 u32 leaf_cpos;
3892
3893 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
3894
3895 restart_search:
3896 ret = ocfs2_dx_dir_lookup(dir, &dx_root->dr_list, &lookup->dl_hinfo,
3897 &leaf_cpos, &blkno);
3898 if (ret) {
3899 mlog_errno(ret);
3900 goto out;
3901 }
3902
3903 ret = ocfs2_read_dx_leaf(dir, blkno, &dx_leaf_bh);
3904 if (ret) {
3905 mlog_errno(ret);
3906 goto out;
3907 }
3908
3909 dx_leaf = (struct ocfs2_dx_leaf *)dx_leaf_bh->b_data;
3910
3911 if (le16_to_cpu(dx_leaf->dl_list.de_num_used) >=
3912 le16_to_cpu(dx_leaf->dl_list.de_count)) {
3913 if (rebalanced) {
3914 /*
3915 * Rebalancing should have provided us with
3916 * space in an appropriate leaf.
3917 *
3918 * XXX: Is this an abnormal condition then?
3919 * Should we print a message here?
3920 */
3921 ret = -ENOSPC;
3922 goto out;
3923 }
3924
3925 ret = ocfs2_dx_dir_rebalance(osb, dir, dx_root_bh, dx_leaf_bh,
3926 &lookup->dl_hinfo, leaf_cpos,
3927 blkno);
3928 if (ret) {
3929 if (ret != -ENOSPC)
3930 mlog_errno(ret);
3931 goto out;
3932 }
3933
3934 /*
3935 * Restart the lookup. The rebalance might have
3936 * changed which block our item fits into. Mark our
3937 * progress, so we only execute this once.
3938 */
3939 brelse(dx_leaf_bh);
3940 dx_leaf_bh = NULL;
3941 rebalanced = 1;
3942 goto restart_search;
3943 }
3944
3945 lookup->dl_dx_leaf_bh = dx_leaf_bh;
3946 dx_leaf_bh = NULL;
3947
3948 out:
3949 brelse(dx_leaf_bh);
3950 return ret;
3951 }
3952
ocfs2_search_dx_free_list(struct inode * dir,struct buffer_head * dx_root_bh,int namelen,struct ocfs2_dir_lookup_result * lookup)3953 static int ocfs2_search_dx_free_list(struct inode *dir,
3954 struct buffer_head *dx_root_bh,
3955 int namelen,
3956 struct ocfs2_dir_lookup_result *lookup)
3957 {
3958 int ret = -ENOSPC;
3959 struct buffer_head *leaf_bh = NULL, *prev_leaf_bh = NULL;
3960 struct ocfs2_dir_block_trailer *db;
3961 u64 next_block;
3962 int rec_len = OCFS2_DIR_REC_LEN(namelen);
3963 struct ocfs2_dx_root_block *dx_root;
3964
3965 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
3966 next_block = le64_to_cpu(dx_root->dr_free_blk);
3967
3968 while (next_block) {
3969 brelse(prev_leaf_bh);
3970 prev_leaf_bh = leaf_bh;
3971 leaf_bh = NULL;
3972
3973 ret = ocfs2_read_dir_block_direct(dir, next_block, &leaf_bh);
3974 if (ret) {
3975 mlog_errno(ret);
3976 goto out;
3977 }
3978
3979 db = ocfs2_trailer_from_bh(leaf_bh, dir->i_sb);
3980 if (rec_len <= le16_to_cpu(db->db_free_rec_len)) {
3981 lookup->dl_leaf_bh = leaf_bh;
3982 lookup->dl_prev_leaf_bh = prev_leaf_bh;
3983 leaf_bh = NULL;
3984 prev_leaf_bh = NULL;
3985 break;
3986 }
3987
3988 next_block = le64_to_cpu(db->db_free_next);
3989 }
3990
3991 if (!next_block)
3992 ret = -ENOSPC;
3993
3994 out:
3995
3996 brelse(leaf_bh);
3997 brelse(prev_leaf_bh);
3998 return ret;
3999 }
4000
ocfs2_expand_inline_dx_root(struct inode * dir,struct buffer_head * dx_root_bh)4001 static int ocfs2_expand_inline_dx_root(struct inode *dir,
4002 struct buffer_head *dx_root_bh)
4003 {
4004 int ret, num_dx_leaves, i, j, did_quota = 0;
4005 struct buffer_head **dx_leaves = NULL;
4006 struct ocfs2_extent_tree et;
4007 u64 insert_blkno;
4008 struct ocfs2_alloc_context *data_ac = NULL;
4009 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
4010 handle_t *handle = NULL;
4011 struct ocfs2_dx_root_block *dx_root;
4012 struct ocfs2_dx_entry_list *entry_list;
4013 struct ocfs2_dx_entry *dx_entry;
4014 struct ocfs2_dx_leaf *target_leaf;
4015
4016 ret = ocfs2_reserve_clusters(osb, 1, &data_ac);
4017 if (ret) {
4018 mlog_errno(ret);
4019 goto out;
4020 }
4021
4022 dx_leaves = ocfs2_dx_dir_kmalloc_leaves(osb->sb, &num_dx_leaves);
4023 if (!dx_leaves) {
4024 ret = -ENOMEM;
4025 mlog_errno(ret);
4026 goto out;
4027 }
4028
4029 handle = ocfs2_start_trans(osb, ocfs2_calc_dxi_expand_credits(osb->sb));
4030 if (IS_ERR(handle)) {
4031 ret = PTR_ERR(handle);
4032 mlog_errno(ret);
4033 goto out;
4034 }
4035
4036 ret = dquot_alloc_space_nodirty(dir,
4037 ocfs2_clusters_to_bytes(osb->sb, 1));
4038 if (ret)
4039 goto out_commit;
4040 did_quota = 1;
4041
4042 /*
4043 * We do this up front, before the allocation, so that a
4044 * failure to add the dx_root_bh to the journal won't result
4045 * us losing clusters.
4046 */
4047 ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
4048 OCFS2_JOURNAL_ACCESS_WRITE);
4049 if (ret) {
4050 mlog_errno(ret);
4051 goto out_commit;
4052 }
4053
4054 ret = __ocfs2_dx_dir_new_cluster(dir, 0, handle, data_ac, dx_leaves,
4055 num_dx_leaves, &insert_blkno);
4056 if (ret) {
4057 mlog_errno(ret);
4058 goto out_commit;
4059 }
4060
4061 /*
4062 * Transfer the entries from our dx_root into the appropriate
4063 * block
4064 */
4065 dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data;
4066 entry_list = &dx_root->dr_entries;
4067
4068 for (i = 0; i < le16_to_cpu(entry_list->de_num_used); i++) {
4069 dx_entry = &entry_list->de_entries[i];
4070
4071 j = __ocfs2_dx_dir_hash_idx(osb,
4072 le32_to_cpu(dx_entry->dx_minor_hash));
4073 target_leaf = (struct ocfs2_dx_leaf *)dx_leaves[j]->b_data;
4074
4075 ocfs2_dx_dir_leaf_insert_tail(target_leaf, dx_entry);
4076
4077 /* Each leaf has been passed to the journal already
4078 * via __ocfs2_dx_dir_new_cluster() */
4079 }
4080
4081 dx_root->dr_flags &= ~OCFS2_DX_FLAG_INLINE;
4082 memset(&dx_root->dr_list, 0, osb->sb->s_blocksize -
4083 offsetof(struct ocfs2_dx_root_block, dr_list));
4084 dx_root->dr_list.l_count =
4085 cpu_to_le16(ocfs2_extent_recs_per_dx_root(osb->sb));
4086
4087 /* This should never fail considering we start with an empty
4088 * dx_root. */
4089 ocfs2_init_dx_root_extent_tree(&et, INODE_CACHE(dir), dx_root_bh);
4090 ret = ocfs2_insert_extent(handle, &et, 0, insert_blkno, 1, 0, NULL);
4091 if (ret)
4092 mlog_errno(ret);
4093 did_quota = 0;
4094
4095 ocfs2_update_inode_fsync_trans(handle, dir, 1);
4096 ocfs2_journal_dirty(handle, dx_root_bh);
4097
4098 out_commit:
4099 if (ret < 0 && did_quota)
4100 dquot_free_space_nodirty(dir,
4101 ocfs2_clusters_to_bytes(dir->i_sb, 1));
4102
4103 ocfs2_commit_trans(osb, handle);
4104
4105 out:
4106 if (data_ac)
4107 ocfs2_free_alloc_context(data_ac);
4108
4109 if (dx_leaves) {
4110 for (i = 0; i < num_dx_leaves; i++)
4111 brelse(dx_leaves[i]);
4112 kfree(dx_leaves);
4113 }
4114 return ret;
4115 }
4116
ocfs2_inline_dx_has_space(struct buffer_head * dx_root_bh)4117 static int ocfs2_inline_dx_has_space(struct buffer_head *dx_root_bh)
4118 {
4119 struct ocfs2_dx_root_block *dx_root;
4120 struct ocfs2_dx_entry_list *entry_list;
4121
4122 dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data;
4123 entry_list = &dx_root->dr_entries;
4124
4125 if (le16_to_cpu(entry_list->de_num_used) >=
4126 le16_to_cpu(entry_list->de_count))
4127 return -ENOSPC;
4128
4129 return 0;
4130 }
4131
ocfs2_prepare_dx_dir_for_insert(struct inode * dir,struct buffer_head * di_bh,const char * name,int namelen,struct ocfs2_dir_lookup_result * lookup)4132 static int ocfs2_prepare_dx_dir_for_insert(struct inode *dir,
4133 struct buffer_head *di_bh,
4134 const char *name,
4135 int namelen,
4136 struct ocfs2_dir_lookup_result *lookup)
4137 {
4138 int ret, free_dx_root = 1;
4139 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
4140 struct buffer_head *dx_root_bh = NULL;
4141 struct buffer_head *leaf_bh = NULL;
4142 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
4143 struct ocfs2_dx_root_block *dx_root;
4144
4145 ret = ocfs2_read_dx_root(dir, di, &dx_root_bh);
4146 if (ret) {
4147 mlog_errno(ret);
4148 goto out;
4149 }
4150
4151 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
4152 if (le32_to_cpu(dx_root->dr_num_entries) == OCFS2_DX_ENTRIES_MAX) {
4153 ret = -ENOSPC;
4154 mlog_errno(ret);
4155 goto out;
4156 }
4157
4158 if (ocfs2_dx_root_inline(dx_root)) {
4159 ret = ocfs2_inline_dx_has_space(dx_root_bh);
4160
4161 if (ret == 0)
4162 goto search_el;
4163
4164 /*
4165 * We ran out of room in the root block. Expand it to
4166 * an extent, then allow ocfs2_find_dir_space_dx to do
4167 * the rest.
4168 */
4169 ret = ocfs2_expand_inline_dx_root(dir, dx_root_bh);
4170 if (ret) {
4171 mlog_errno(ret);
4172 goto out;
4173 }
4174 }
4175
4176 /*
4177 * Insert preparation for an indexed directory is split into two
4178 * steps. The call to find_dir_space_dx reserves room in the index for
4179 * an additional item. If we run out of space there, it's a real error
4180 * we can't continue on.
4181 */
4182 ret = ocfs2_find_dir_space_dx(osb, dir, di_bh, dx_root_bh, name,
4183 namelen, lookup);
4184 if (ret) {
4185 mlog_errno(ret);
4186 goto out;
4187 }
4188
4189 search_el:
4190 /*
4191 * Next, we need to find space in the unindexed tree. This call
4192 * searches using the free space linked list. If the unindexed tree
4193 * lacks sufficient space, we'll expand it below. The expansion code
4194 * is smart enough to add any new blocks to the free space list.
4195 */
4196 ret = ocfs2_search_dx_free_list(dir, dx_root_bh, namelen, lookup);
4197 if (ret && ret != -ENOSPC) {
4198 mlog_errno(ret);
4199 goto out;
4200 }
4201
4202 /* Do this up here - ocfs2_extend_dir might need the dx_root */
4203 lookup->dl_dx_root_bh = dx_root_bh;
4204 free_dx_root = 0;
4205
4206 if (ret == -ENOSPC) {
4207 ret = ocfs2_extend_dir(osb, dir, di_bh, 1, lookup, &leaf_bh);
4208
4209 if (ret) {
4210 mlog_errno(ret);
4211 goto out;
4212 }
4213
4214 /*
4215 * We make the assumption here that new leaf blocks are added
4216 * to the front of our free list.
4217 */
4218 lookup->dl_prev_leaf_bh = NULL;
4219 lookup->dl_leaf_bh = leaf_bh;
4220 }
4221
4222 out:
4223 if (free_dx_root)
4224 brelse(dx_root_bh);
4225 return ret;
4226 }
4227
4228 /*
4229 * Get a directory ready for insert. Any directory allocation required
4230 * happens here. Success returns zero, and enough context in the dir
4231 * lookup result that ocfs2_add_entry() will be able complete the task
4232 * with minimal performance impact.
4233 */
ocfs2_prepare_dir_for_insert(struct ocfs2_super * osb,struct inode * dir,struct buffer_head * parent_fe_bh,const char * name,int namelen,struct ocfs2_dir_lookup_result * lookup)4234 int ocfs2_prepare_dir_for_insert(struct ocfs2_super *osb,
4235 struct inode *dir,
4236 struct buffer_head *parent_fe_bh,
4237 const char *name,
4238 int namelen,
4239 struct ocfs2_dir_lookup_result *lookup)
4240 {
4241 int ret;
4242 unsigned int blocks_wanted = 1;
4243 struct buffer_head *bh = NULL;
4244
4245 trace_ocfs2_prepare_dir_for_insert(
4246 (unsigned long long)OCFS2_I(dir)->ip_blkno, namelen);
4247
4248 /*
4249 * Do this up front to reduce confusion.
4250 *
4251 * The directory might start inline, then be turned into an
4252 * indexed one, in which case we'd need to hash deep inside
4253 * ocfs2_find_dir_space_id(). Since
4254 * ocfs2_prepare_dx_dir_for_insert() also needs this hash
4255 * done, there seems no point in spreading out the calls. We
4256 * can optimize away the case where the file system doesn't
4257 * support indexing.
4258 */
4259 if (ocfs2_supports_indexed_dirs(osb))
4260 ocfs2_dx_dir_name_hash(dir, name, namelen, &lookup->dl_hinfo);
4261
4262 if (ocfs2_dir_indexed(dir)) {
4263 ret = ocfs2_prepare_dx_dir_for_insert(dir, parent_fe_bh,
4264 name, namelen, lookup);
4265 if (ret)
4266 mlog_errno(ret);
4267 goto out;
4268 }
4269
4270 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
4271 ret = ocfs2_find_dir_space_id(dir, parent_fe_bh, name,
4272 namelen, &bh, &blocks_wanted);
4273 } else
4274 ret = ocfs2_find_dir_space_el(dir, name, namelen, &bh);
4275
4276 if (ret && ret != -ENOSPC) {
4277 mlog_errno(ret);
4278 goto out;
4279 }
4280
4281 if (ret == -ENOSPC) {
4282 /*
4283 * We have to expand the directory to add this name.
4284 */
4285 BUG_ON(bh);
4286
4287 ret = ocfs2_extend_dir(osb, dir, parent_fe_bh, blocks_wanted,
4288 lookup, &bh);
4289 if (ret) {
4290 if (ret != -ENOSPC)
4291 mlog_errno(ret);
4292 goto out;
4293 }
4294
4295 BUG_ON(!bh);
4296 }
4297
4298 lookup->dl_leaf_bh = bh;
4299 bh = NULL;
4300 out:
4301 brelse(bh);
4302 return ret;
4303 }
4304
ocfs2_dx_dir_remove_index(struct inode * dir,struct buffer_head * di_bh,struct buffer_head * dx_root_bh)4305 static int ocfs2_dx_dir_remove_index(struct inode *dir,
4306 struct buffer_head *di_bh,
4307 struct buffer_head *dx_root_bh)
4308 {
4309 int ret;
4310 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
4311 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
4312 struct ocfs2_dx_root_block *dx_root;
4313 struct inode *dx_alloc_inode = NULL;
4314 struct buffer_head *dx_alloc_bh = NULL;
4315 handle_t *handle;
4316 u64 blk;
4317 u16 bit;
4318 u64 bg_blkno;
4319
4320 dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data;
4321
4322 dx_alloc_inode = ocfs2_get_system_file_inode(osb,
4323 EXTENT_ALLOC_SYSTEM_INODE,
4324 le16_to_cpu(dx_root->dr_suballoc_slot));
4325 if (!dx_alloc_inode) {
4326 ret = -ENOMEM;
4327 mlog_errno(ret);
4328 goto out;
4329 }
4330 inode_lock(dx_alloc_inode);
4331
4332 ret = ocfs2_inode_lock(dx_alloc_inode, &dx_alloc_bh, 1);
4333 if (ret) {
4334 mlog_errno(ret);
4335 goto out_mutex;
4336 }
4337
4338 handle = ocfs2_start_trans(osb, OCFS2_DX_ROOT_REMOVE_CREDITS);
4339 if (IS_ERR(handle)) {
4340 ret = PTR_ERR(handle);
4341 mlog_errno(ret);
4342 goto out_unlock;
4343 }
4344
4345 ret = ocfs2_journal_access_di(handle, INODE_CACHE(dir), di_bh,
4346 OCFS2_JOURNAL_ACCESS_WRITE);
4347 if (ret) {
4348 mlog_errno(ret);
4349 goto out_commit;
4350 }
4351
4352 spin_lock(&OCFS2_I(dir)->ip_lock);
4353 OCFS2_I(dir)->ip_dyn_features &= ~OCFS2_INDEXED_DIR_FL;
4354 di->i_dyn_features = cpu_to_le16(OCFS2_I(dir)->ip_dyn_features);
4355 spin_unlock(&OCFS2_I(dir)->ip_lock);
4356 di->i_dx_root = cpu_to_le64(0ULL);
4357 ocfs2_update_inode_fsync_trans(handle, dir, 1);
4358
4359 ocfs2_journal_dirty(handle, di_bh);
4360
4361 blk = le64_to_cpu(dx_root->dr_blkno);
4362 bit = le16_to_cpu(dx_root->dr_suballoc_bit);
4363 if (dx_root->dr_suballoc_loc)
4364 bg_blkno = le64_to_cpu(dx_root->dr_suballoc_loc);
4365 else
4366 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
4367 ret = ocfs2_free_suballoc_bits(handle, dx_alloc_inode, dx_alloc_bh,
4368 bit, bg_blkno, 1);
4369 if (ret)
4370 mlog_errno(ret);
4371
4372 out_commit:
4373 ocfs2_commit_trans(osb, handle);
4374
4375 out_unlock:
4376 ocfs2_inode_unlock(dx_alloc_inode, 1);
4377
4378 out_mutex:
4379 inode_unlock(dx_alloc_inode);
4380 brelse(dx_alloc_bh);
4381 out:
4382 iput(dx_alloc_inode);
4383 return ret;
4384 }
4385
ocfs2_dx_dir_truncate(struct inode * dir,struct buffer_head * di_bh)4386 int ocfs2_dx_dir_truncate(struct inode *dir, struct buffer_head *di_bh)
4387 {
4388 int ret;
4389 unsigned int clen;
4390 u32 major_hash = UINT_MAX, p_cpos, cpos;
4391 u64 blkno;
4392 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
4393 struct buffer_head *dx_root_bh = NULL;
4394 struct ocfs2_dx_root_block *dx_root;
4395 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
4396 struct ocfs2_cached_dealloc_ctxt dealloc;
4397 struct ocfs2_extent_tree et;
4398
4399 ocfs2_init_dealloc_ctxt(&dealloc);
4400
4401 if (!ocfs2_dir_indexed(dir))
4402 return 0;
4403
4404 ret = ocfs2_read_dx_root(dir, di, &dx_root_bh);
4405 if (ret) {
4406 mlog_errno(ret);
4407 goto out;
4408 }
4409 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
4410
4411 if (ocfs2_dx_root_inline(dx_root))
4412 goto remove_index;
4413
4414 ocfs2_init_dx_root_extent_tree(&et, INODE_CACHE(dir), dx_root_bh);
4415
4416 /* XXX: What if dr_clusters is too large? */
4417 while (le32_to_cpu(dx_root->dr_clusters)) {
4418 ret = ocfs2_dx_dir_lookup_rec(dir, &dx_root->dr_list,
4419 major_hash, &cpos, &blkno, &clen);
4420 if (ret) {
4421 mlog_errno(ret);
4422 goto out;
4423 }
4424
4425 p_cpos = ocfs2_blocks_to_clusters(dir->i_sb, blkno);
4426
4427 ret = ocfs2_remove_btree_range(dir, &et, cpos, p_cpos, clen, 0,
4428 &dealloc, 0, false);
4429 if (ret) {
4430 mlog_errno(ret);
4431 goto out;
4432 }
4433
4434 if (cpos == 0)
4435 break;
4436
4437 major_hash = cpos - 1;
4438 }
4439
4440 remove_index:
4441 ret = ocfs2_dx_dir_remove_index(dir, di_bh, dx_root_bh);
4442 if (ret) {
4443 mlog_errno(ret);
4444 goto out;
4445 }
4446
4447 ocfs2_remove_from_cache(INODE_CACHE(dir), dx_root_bh);
4448 out:
4449 ocfs2_schedule_truncate_log_flush(osb, 1);
4450 ocfs2_run_deallocs(osb, &dealloc);
4451
4452 brelse(dx_root_bh);
4453 return ret;
4454 }
4455