?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'v6v7' into devel
[linux/fpc-iii.git] / fs / gfs2 / dir.c
blob5c356d09c321c10133afc7cf93aba2eddd1cb3c1
1 /*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
4 *
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
8 */
9
10 /*
11 * Implements Extendible Hashing as described in:
12 * "Extendible Hashing" by Fagin, et al in
13 * __ACM Trans. on Database Systems__, Sept 1979.
14 *
15 *
16 * Here's the layout of dirents which is essentially the same as that of ext2
17 * within a single block. The field de_name_len is the number of bytes
18 * actually required for the name (no null terminator). The field de_rec_len
19 * is the number of bytes allocated to the dirent. The offset of the next
20 * dirent in the block is (dirent + dirent->de_rec_len). When a dirent is
21 * deleted, the preceding dirent inherits its allocated space, ie
22 * prev->de_rec_len += deleted->de_rec_len. Since the next dirent is obtained
23 * by adding de_rec_len to the current dirent, this essentially causes the
24 * deleted dirent to get jumped over when iterating through all the dirents.
25 *
26 * When deleting the first dirent in a block, there is no previous dirent so
27 * the field de_ino is set to zero to designate it as deleted. When allocating
28 * a dirent, gfs2_dirent_alloc iterates through the dirents in a block. If the
29 * first dirent has (de_ino == 0) and de_rec_len is large enough, this first
30 * dirent is allocated. Otherwise it must go through all the 'used' dirents
31 * searching for one in which the amount of total space minus the amount of
32 * used space will provide enough space for the new dirent.
33 *
34 * There are two types of blocks in which dirents reside. In a stuffed dinode,
35 * the dirents begin at offset sizeof(struct gfs2_dinode) from the beginning of
36 * the block. In leaves, they begin at offset sizeof(struct gfs2_leaf) from the
37 * beginning of the leaf block. The dirents reside in leaves when
38 *
39 * dip->i_diskflags & GFS2_DIF_EXHASH is true
40 *
41 * Otherwise, the dirents are "linear", within a single stuffed dinode block.
42 *
43 * When the dirents are in leaves, the actual contents of the directory file are
44 * used as an array of 64-bit block pointers pointing to the leaf blocks. The
45 * dirents are NOT in the directory file itself. There can be more than one
46 * block pointer in the array that points to the same leaf. In fact, when a
47 * directory is first converted from linear to exhash, all of the pointers
48 * point to the same leaf.
49 *
50 * When a leaf is completely full, the size of the hash table can be
51 * doubled unless it is already at the maximum size which is hard coded into
52 * GFS2_DIR_MAX_DEPTH. After that, leaves are chained together in a linked list,
53 * but never before the maximum hash table size has been reached.
54 */
55
56 #include <linux/slab.h>
57 #include <linux/spinlock.h>
58 #include <linux/buffer_head.h>
59 #include <linux/sort.h>
60 #include <linux/gfs2_ondisk.h>
61 #include <linux/crc32.h>
62 #include <linux/vmalloc.h>
63
64 #include "gfs2.h"
65 #include "incore.h"
66 #include "dir.h"
67 #include "glock.h"
68 #include "inode.h"
69 #include "meta_io.h"
70 #include "quota.h"
71 #include "rgrp.h"
72 #include "trans.h"
73 #include "bmap.h"
74 #include "util.h"
75
76 #define IS_LEAF 1 /* Hashed (leaf) directory */
77 #define IS_DINODE 2 /* Linear (stuffed dinode block) directory */
78
79 #define gfs2_disk_hash2offset(h) (((u64)(h)) >> 1)
80 #define gfs2_dir_offset2hash(p) ((u32)(((u64)(p)) << 1))
81
82 struct qstr gfs2_qdot __read_mostly;
83 struct qstr gfs2_qdotdot __read_mostly;
84
85 typedef int (*leaf_call_t) (struct gfs2_inode *dip, u32 index, u32 len,
86 u64 leaf_no, void *data);
87 typedef int (*gfs2_dscan_t)(const struct gfs2_dirent *dent,
88 const struct qstr *name, void *opaque);
89
90
91 int gfs2_dir_get_new_buffer(struct gfs2_inode *ip, u64 block,
92 struct buffer_head **bhp)
93 {
94 struct buffer_head *bh;
95
96 bh = gfs2_meta_new(ip->i_gl, block);
97 gfs2_trans_add_bh(ip->i_gl, bh, 1);
98 gfs2_metatype_set(bh, GFS2_METATYPE_JD, GFS2_FORMAT_JD);
99 gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
100 *bhp = bh;
101 return 0;
102 }
103
104 static int gfs2_dir_get_existing_buffer(struct gfs2_inode *ip, u64 block,
105 struct buffer_head **bhp)
106 {
107 struct buffer_head *bh;
108 int error;
109
110 error = gfs2_meta_read(ip->i_gl, block, DIO_WAIT, &bh);
111 if (error)
112 return error;
113 if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), bh, GFS2_METATYPE_JD)) {
114 brelse(bh);
115 return -EIO;
116 }
117 *bhp = bh;
118 return 0;
119 }
120
121 static int gfs2_dir_write_stuffed(struct gfs2_inode *ip, const char *buf,
122 unsigned int offset, unsigned int size)
123 {
124 struct buffer_head *dibh;
125 int error;
126
127 error = gfs2_meta_inode_buffer(ip, &dibh);
128 if (error)
129 return error;
130
131 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
132 memcpy(dibh->b_data + offset + sizeof(struct gfs2_dinode), buf, size);
133 if (ip->i_inode.i_size < offset + size)
134 i_size_write(&ip->i_inode, offset + size);
135 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
136 gfs2_dinode_out(ip, dibh->b_data);
137
138 brelse(dibh);
139
140 return size;
141 }
142
143
144
145 /**
146 * gfs2_dir_write_data - Write directory information to the inode
147 * @ip: The GFS2 inode
148 * @buf: The buffer containing information to be written
149 * @offset: The file offset to start writing at
150 * @size: The amount of data to write
151 *
152 * Returns: The number of bytes correctly written or error code
153 */
154 static int gfs2_dir_write_data(struct gfs2_inode *ip, const char *buf,
155 u64 offset, unsigned int size)
156 {
157 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
158 struct buffer_head *dibh;
159 u64 lblock, dblock;
160 u32 extlen = 0;
161 unsigned int o;
162 int copied = 0;
163 int error = 0;
164 int new = 0;
165
166 if (!size)
167 return 0;
168
169 if (gfs2_is_stuffed(ip) &&
170 offset + size <= sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode))
171 return gfs2_dir_write_stuffed(ip, buf, (unsigned int)offset,
172 size);
173
174 if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
175 return -EINVAL;
176
177 if (gfs2_is_stuffed(ip)) {
178 error = gfs2_unstuff_dinode(ip, NULL);
179 if (error)
180 return error;
181 }
182
183 lblock = offset;
184 o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
185
186 while (copied < size) {
187 unsigned int amount;
188 struct buffer_head *bh;
189
190 amount = size - copied;
191 if (amount > sdp->sd_sb.sb_bsize - o)
192 amount = sdp->sd_sb.sb_bsize - o;
193
194 if (!extlen) {
195 new = 1;
196 error = gfs2_extent_map(&ip->i_inode, lblock, &new,
197 &dblock, &extlen);
198 if (error)
199 goto fail;
200 error = -EIO;
201 if (gfs2_assert_withdraw(sdp, dblock))
202 goto fail;
203 }
204
205 if (amount == sdp->sd_jbsize || new)
206 error = gfs2_dir_get_new_buffer(ip, dblock, &bh);
207 else
208 error = gfs2_dir_get_existing_buffer(ip, dblock, &bh);
209
210 if (error)
211 goto fail;
212
213 gfs2_trans_add_bh(ip->i_gl, bh, 1);
214 memcpy(bh->b_data + o, buf, amount);
215 brelse(bh);
216
217 buf += amount;
218 copied += amount;
219 lblock++;
220 dblock++;
221 extlen--;
222
223 o = sizeof(struct gfs2_meta_header);
224 }
225
226 out:
227 error = gfs2_meta_inode_buffer(ip, &dibh);
228 if (error)
229 return error;
230
231 if (ip->i_inode.i_size < offset + copied)
232 i_size_write(&ip->i_inode, offset + copied);
233 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
234
235 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
236 gfs2_dinode_out(ip, dibh->b_data);
237 brelse(dibh);
238
239 return copied;
240 fail:
241 if (copied)
242 goto out;
243 return error;
244 }
245
246 static int gfs2_dir_read_stuffed(struct gfs2_inode *ip, char *buf,
247 u64 offset, unsigned int size)
248 {
249 struct buffer_head *dibh;
250 int error;
251
252 error = gfs2_meta_inode_buffer(ip, &dibh);
253 if (!error) {
254 offset += sizeof(struct gfs2_dinode);
255 memcpy(buf, dibh->b_data + offset, size);
256 brelse(dibh);
257 }
258
259 return (error) ? error : size;
260 }
261
262
263 /**
264 * gfs2_dir_read_data - Read a data from a directory inode
265 * @ip: The GFS2 Inode
266 * @buf: The buffer to place result into
267 * @offset: File offset to begin jdata_readng from
268 * @size: Amount of data to transfer
269 *
270 * Returns: The amount of data actually copied or the error
271 */
272 static int gfs2_dir_read_data(struct gfs2_inode *ip, char *buf, u64 offset,
273 unsigned int size, unsigned ra)
274 {
275 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
276 u64 lblock, dblock;
277 u32 extlen = 0;
278 unsigned int o;
279 int copied = 0;
280 int error = 0;
281 u64 disksize = i_size_read(&ip->i_inode);
282
283 if (offset >= disksize)
284 return 0;
285
286 if (offset + size > disksize)
287 size = disksize - offset;
288
289 if (!size)
290 return 0;
291
292 if (gfs2_is_stuffed(ip))
293 return gfs2_dir_read_stuffed(ip, buf, offset, size);
294
295 if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
296 return -EINVAL;
297
298 lblock = offset;
299 o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
300
301 while (copied < size) {
302 unsigned int amount;
303 struct buffer_head *bh;
304 int new;
305
306 amount = size - copied;
307 if (amount > sdp->sd_sb.sb_bsize - o)
308 amount = sdp->sd_sb.sb_bsize - o;
309
310 if (!extlen) {
311 new = 0;
312 error = gfs2_extent_map(&ip->i_inode, lblock, &new,
313 &dblock, &extlen);
314 if (error || !dblock)
315 goto fail;
316 BUG_ON(extlen < 1);
317 if (!ra)
318 extlen = 1;
319 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
320 } else {
321 error = gfs2_meta_read(ip->i_gl, dblock, DIO_WAIT, &bh);
322 if (error)
323 goto fail;
324 }
325 error = gfs2_metatype_check(sdp, bh, GFS2_METATYPE_JD);
326 if (error) {
327 brelse(bh);
328 goto fail;
329 }
330 dblock++;
331 extlen--;
332 memcpy(buf, bh->b_data + o, amount);
333 brelse(bh);
334 buf += amount;
335 copied += amount;
336 lblock++;
337 o = sizeof(struct gfs2_meta_header);
338 }
339
340 return copied;
341 fail:
342 return (copied) ? copied : error;
343 }
344
345 static inline int gfs2_dirent_sentinel(const struct gfs2_dirent *dent)
346 {
347 return dent->de_inum.no_addr == 0 || dent->de_inum.no_formal_ino == 0;
348 }
349
350 static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent,
351 const struct qstr *name, int ret)
352 {
353 if (!gfs2_dirent_sentinel(dent) &&
354 be32_to_cpu(dent->de_hash) == name->hash &&
355 be16_to_cpu(dent->de_name_len) == name->len &&
356 memcmp(dent+1, name->name, name->len) == 0)
357 return ret;
358 return 0;
359 }
360
361 static int gfs2_dirent_find(const struct gfs2_dirent *dent,
362 const struct qstr *name,
363 void *opaque)
364 {
365 return __gfs2_dirent_find(dent, name, 1);
366 }
367
368 static int gfs2_dirent_prev(const struct gfs2_dirent *dent,
369 const struct qstr *name,
370 void *opaque)
371 {
372 return __gfs2_dirent_find(dent, name, 2);
373 }
374
375 /*
376 * name->name holds ptr to start of block.
377 * name->len holds size of block.
378 */
379 static int gfs2_dirent_last(const struct gfs2_dirent *dent,
380 const struct qstr *name,
381 void *opaque)
382 {
383 const char *start = name->name;
384 const char *end = (const char *)dent + be16_to_cpu(dent->de_rec_len);
385 if (name->len == (end - start))
386 return 1;
387 return 0;
388 }
389
390 static int gfs2_dirent_find_space(const struct gfs2_dirent *dent,
391 const struct qstr *name,
392 void *opaque)
393 {
394 unsigned required = GFS2_DIRENT_SIZE(name->len);
395 unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
396 unsigned totlen = be16_to_cpu(dent->de_rec_len);
397
398 if (gfs2_dirent_sentinel(dent))
399 actual = 0;
400 if (totlen - actual >= required)
401 return 1;
402 return 0;
403 }
404
405 struct dirent_gather {
406 const struct gfs2_dirent **pdent;
407 unsigned offset;
408 };
409
410 static int gfs2_dirent_gather(const struct gfs2_dirent *dent,
411 const struct qstr *name,
412 void *opaque)
413 {
414 struct dirent_gather *g = opaque;
415 if (!gfs2_dirent_sentinel(dent)) {
416 g->pdent[g->offset++] = dent;
417 }
418 return 0;
419 }
420
421 /*
422 * Other possible things to check:
423 * - Inode located within filesystem size (and on valid block)
424 * - Valid directory entry type
425 * Not sure how heavy-weight we want to make this... could also check
426 * hash is correct for example, but that would take a lot of extra time.
427 * For now the most important thing is to check that the various sizes
428 * are correct.
429 */
430 static int gfs2_check_dirent(struct gfs2_dirent *dent, unsigned int offset,
431 unsigned int size, unsigned int len, int first)
432 {
433 const char *msg = "gfs2_dirent too small";
434 if (unlikely(size < sizeof(struct gfs2_dirent)))
435 goto error;
436 msg = "gfs2_dirent misaligned";
437 if (unlikely(offset & 0x7))
438 goto error;
439 msg = "gfs2_dirent points beyond end of block";
440 if (unlikely(offset + size > len))
441 goto error;
442 msg = "zero inode number";
443 if (unlikely(!first && gfs2_dirent_sentinel(dent)))
444 goto error;
445 msg = "name length is greater than space in dirent";
446 if (!gfs2_dirent_sentinel(dent) &&
447 unlikely(sizeof(struct gfs2_dirent)+be16_to_cpu(dent->de_name_len) >
448 size))
449 goto error;
450 return 0;
451 error:
452 printk(KERN_WARNING "gfs2_check_dirent: %s (%s)\n", msg,
453 first ? "first in block" : "not first in block");
454 return -EIO;
455 }
456
457 static int gfs2_dirent_offset(const void *buf)
458 {
459 const struct gfs2_meta_header *h = buf;
460 int offset;
461
462 BUG_ON(buf == NULL);
463
464 switch(be32_to_cpu(h->mh_type)) {
465 case GFS2_METATYPE_LF:
466 offset = sizeof(struct gfs2_leaf);
467 break;
468 case GFS2_METATYPE_DI:
469 offset = sizeof(struct gfs2_dinode);
470 break;
471 default:
472 goto wrong_type;
473 }
474 return offset;
475 wrong_type:
476 printk(KERN_WARNING "gfs2_scan_dirent: wrong block type %u\n",
477 be32_to_cpu(h->mh_type));
478 return -1;
479 }
480
481 static struct gfs2_dirent *gfs2_dirent_scan(struct inode *inode, void *buf,
482 unsigned int len, gfs2_dscan_t scan,
483 const struct qstr *name,
484 void *opaque)
485 {
486 struct gfs2_dirent *dent, *prev;
487 unsigned offset;
488 unsigned size;
489 int ret = 0;
490
491 ret = gfs2_dirent_offset(buf);
492 if (ret < 0)
493 goto consist_inode;
494
495 offset = ret;
496 prev = NULL;
497 dent = buf + offset;
498 size = be16_to_cpu(dent->de_rec_len);
499 if (gfs2_check_dirent(dent, offset, size, len, 1))
500 goto consist_inode;
501 do {
502 ret = scan(dent, name, opaque);
503 if (ret)
504 break;
505 offset += size;
506 if (offset == len)
507 break;
508 prev = dent;
509 dent = buf + offset;
510 size = be16_to_cpu(dent->de_rec_len);
511 if (gfs2_check_dirent(dent, offset, size, len, 0))
512 goto consist_inode;
513 } while(1);
514
515 switch(ret) {
516 case 0:
517 return NULL;
518 case 1:
519 return dent;
520 case 2:
521 return prev ? prev : dent;
522 default:
523 BUG_ON(ret > 0);
524 return ERR_PTR(ret);
525 }
526
527 consist_inode:
528 gfs2_consist_inode(GFS2_I(inode));
529 return ERR_PTR(-EIO);
530 }
531
532 static int dirent_check_reclen(struct gfs2_inode *dip,
533 const struct gfs2_dirent *d, const void *end_p)
534 {
535 const void *ptr = d;
536 u16 rec_len = be16_to_cpu(d->de_rec_len);
537
538 if (unlikely(rec_len < sizeof(struct gfs2_dirent)))
539 goto broken;
540 ptr += rec_len;
541 if (ptr < end_p)
542 return rec_len;
543 if (ptr == end_p)
544 return -ENOENT;
545 broken:
546 gfs2_consist_inode(dip);
547 return -EIO;
548 }
549
550 /**
551 * dirent_next - Next dirent
552 * @dip: the directory
553 * @bh: The buffer
554 * @dent: Pointer to list of dirents
555 *
556 * Returns: 0 on success, error code otherwise
557 */
558
559 static int dirent_next(struct gfs2_inode *dip, struct buffer_head *bh,
560 struct gfs2_dirent **dent)
561 {
562 struct gfs2_dirent *cur = *dent, *tmp;
563 char *bh_end = bh->b_data + bh->b_size;
564 int ret;
565
566 ret = dirent_check_reclen(dip, cur, bh_end);
567 if (ret < 0)
568 return ret;
569
570 tmp = (void *)cur + ret;
571 ret = dirent_check_reclen(dip, tmp, bh_end);
572 if (ret == -EIO)
573 return ret;
574
575 /* Only the first dent could ever have de_inum.no_addr == 0 */
576 if (gfs2_dirent_sentinel(tmp)) {
577 gfs2_consist_inode(dip);
578 return -EIO;
579 }
580
581 *dent = tmp;
582 return 0;
583 }
584
585 /**
586 * dirent_del - Delete a dirent
587 * @dip: The GFS2 inode
588 * @bh: The buffer
589 * @prev: The previous dirent
590 * @cur: The current dirent
591 *
592 */
593
594 static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh,
595 struct gfs2_dirent *prev, struct gfs2_dirent *cur)
596 {
597 u16 cur_rec_len, prev_rec_len;
598
599 if (gfs2_dirent_sentinel(cur)) {
600 gfs2_consist_inode(dip);
601 return;
602 }
603
604 gfs2_trans_add_bh(dip->i_gl, bh, 1);
605
606 /* If there is no prev entry, this is the first entry in the block.
607 The de_rec_len is already as big as it needs to be. Just zero
608 out the inode number and return. */
609
610 if (!prev) {
611 cur->de_inum.no_addr = 0;
612 cur->de_inum.no_formal_ino = 0;
613 return;
614 }
615
616 /* Combine this dentry with the previous one. */
617
618 prev_rec_len = be16_to_cpu(prev->de_rec_len);
619 cur_rec_len = be16_to_cpu(cur->de_rec_len);
620
621 if ((char *)prev + prev_rec_len != (char *)cur)
622 gfs2_consist_inode(dip);
623 if ((char *)cur + cur_rec_len > bh->b_data + bh->b_size)
624 gfs2_consist_inode(dip);
625
626 prev_rec_len += cur_rec_len;
627 prev->de_rec_len = cpu_to_be16(prev_rec_len);
628 }
629
630 /*
631 * Takes a dent from which to grab space as an argument. Returns the
632 * newly created dent.
633 */
634 static struct gfs2_dirent *gfs2_init_dirent(struct inode *inode,
635 struct gfs2_dirent *dent,
636 const struct qstr *name,
637 struct buffer_head *bh)
638 {
639 struct gfs2_inode *ip = GFS2_I(inode);
640 struct gfs2_dirent *ndent;
641 unsigned offset = 0, totlen;
642
643 if (!gfs2_dirent_sentinel(dent))
644 offset = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
645 totlen = be16_to_cpu(dent->de_rec_len);
646 BUG_ON(offset + name->len > totlen);
647 gfs2_trans_add_bh(ip->i_gl, bh, 1);
648 ndent = (struct gfs2_dirent *)((char *)dent + offset);
649 dent->de_rec_len = cpu_to_be16(offset);
650 gfs2_qstr2dirent(name, totlen - offset, ndent);
651 return ndent;
652 }
653
654 static struct gfs2_dirent *gfs2_dirent_alloc(struct inode *inode,
655 struct buffer_head *bh,
656 const struct qstr *name)
657 {
658 struct gfs2_dirent *dent;
659 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
660 gfs2_dirent_find_space, name, NULL);
661 if (!dent || IS_ERR(dent))
662 return dent;
663 return gfs2_init_dirent(inode, dent, name, bh);
664 }
665
666 static int get_leaf(struct gfs2_inode *dip, u64 leaf_no,
667 struct buffer_head **bhp)
668 {
669 int error;
670
671 error = gfs2_meta_read(dip->i_gl, leaf_no, DIO_WAIT, bhp);
672 if (!error && gfs2_metatype_check(GFS2_SB(&dip->i_inode), *bhp, GFS2_METATYPE_LF)) {
673 /* printk(KERN_INFO "block num=%llu\n", leaf_no); */
674 error = -EIO;
675 }
676
677 return error;
678 }
679
680 /**
681 * get_leaf_nr - Get a leaf number associated with the index
682 * @dip: The GFS2 inode
683 * @index:
684 * @leaf_out:
685 *
686 * Returns: 0 on success, error code otherwise
687 */
688
689 static int get_leaf_nr(struct gfs2_inode *dip, u32 index,
690 u64 *leaf_out)
691 {
692 __be64 leaf_no;
693 int error;
694
695 error = gfs2_dir_read_data(dip, (char *)&leaf_no,
696 index * sizeof(__be64),
697 sizeof(__be64), 0);
698 if (error != sizeof(u64))
699 return (error < 0) ? error : -EIO;
700
701 *leaf_out = be64_to_cpu(leaf_no);
702
703 return 0;
704 }
705
706 static int get_first_leaf(struct gfs2_inode *dip, u32 index,
707 struct buffer_head **bh_out)
708 {
709 u64 leaf_no;
710 int error;
711
712 error = get_leaf_nr(dip, index, &leaf_no);
713 if (!error)
714 error = get_leaf(dip, leaf_no, bh_out);
715
716 return error;
717 }
718
719 static struct gfs2_dirent *gfs2_dirent_search(struct inode *inode,
720 const struct qstr *name,
721 gfs2_dscan_t scan,
722 struct buffer_head **pbh)
723 {
724 struct buffer_head *bh;
725 struct gfs2_dirent *dent;
726 struct gfs2_inode *ip = GFS2_I(inode);
727 int error;
728
729 if (ip->i_diskflags & GFS2_DIF_EXHASH) {
730 struct gfs2_leaf *leaf;
731 unsigned hsize = 1 << ip->i_depth;
732 unsigned index;
733 u64 ln;
734 if (hsize * sizeof(u64) != i_size_read(inode)) {
735 gfs2_consist_inode(ip);
736 return ERR_PTR(-EIO);
737 }
738
739 index = name->hash >> (32 - ip->i_depth);
740 error = get_first_leaf(ip, index, &bh);
741 if (error)
742 return ERR_PTR(error);
743 do {
744 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
745 scan, name, NULL);
746 if (dent)
747 goto got_dent;
748 leaf = (struct gfs2_leaf *)bh->b_data;
749 ln = be64_to_cpu(leaf->lf_next);
750 brelse(bh);
751 if (!ln)
752 break;
753
754 error = get_leaf(ip, ln, &bh);
755 } while(!error);
756
757 return error ? ERR_PTR(error) : NULL;
758 }
759
760
761 error = gfs2_meta_inode_buffer(ip, &bh);
762 if (error)
763 return ERR_PTR(error);
764 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, scan, name, NULL);
765 got_dent:
766 if (unlikely(dent == NULL || IS_ERR(dent))) {
767 brelse(bh);
768 bh = NULL;
769 }
770 *pbh = bh;
771 return dent;
772 }
773
774 static struct gfs2_leaf *new_leaf(struct inode *inode, struct buffer_head **pbh, u16 depth)
775 {
776 struct gfs2_inode *ip = GFS2_I(inode);
777 unsigned int n = 1;
778 u64 bn;
779 int error;
780 struct buffer_head *bh;
781 struct gfs2_leaf *leaf;
782 struct gfs2_dirent *dent;
783 struct qstr name = { .name = "", .len = 0, .hash = 0 };
784
785 error = gfs2_alloc_block(ip, &bn, &n);
786 if (error)
787 return NULL;
788 bh = gfs2_meta_new(ip->i_gl, bn);
789 if (!bh)
790 return NULL;
791
792 gfs2_trans_add_unrevoke(GFS2_SB(inode), bn, 1);
793 gfs2_trans_add_bh(ip->i_gl, bh, 1);
794 gfs2_metatype_set(bh, GFS2_METATYPE_LF, GFS2_FORMAT_LF);
795 leaf = (struct gfs2_leaf *)bh->b_data;
796 leaf->lf_depth = cpu_to_be16(depth);
797 leaf->lf_entries = 0;
798 leaf->lf_dirent_format = cpu_to_be32(GFS2_FORMAT_DE);
799 leaf->lf_next = 0;
800 memset(leaf->lf_reserved, 0, sizeof(leaf->lf_reserved));
801 dent = (struct gfs2_dirent *)(leaf+1);
802 gfs2_qstr2dirent(&name, bh->b_size - sizeof(struct gfs2_leaf), dent);
803 *pbh = bh;
804 return leaf;
805 }
806
807 /**
808 * dir_make_exhash - Convert a stuffed directory into an ExHash directory
809 * @dip: The GFS2 inode
810 *
811 * Returns: 0 on success, error code otherwise
812 */
813
814 static int dir_make_exhash(struct inode *inode)
815 {
816 struct gfs2_inode *dip = GFS2_I(inode);
817 struct gfs2_sbd *sdp = GFS2_SB(inode);
818 struct gfs2_dirent *dent;
819 struct qstr args;
820 struct buffer_head *bh, *dibh;
821 struct gfs2_leaf *leaf;
822 int y;
823 u32 x;
824 __be64 *lp;
825 u64 bn;
826 int error;
827
828 error = gfs2_meta_inode_buffer(dip, &dibh);
829 if (error)
830 return error;
831
832 /* Turn over a new leaf */
833
834 leaf = new_leaf(inode, &bh, 0);
835 if (!leaf)
836 return -ENOSPC;
837 bn = bh->b_blocknr;
838
839 gfs2_assert(sdp, dip->i_entries < (1 << 16));
840 leaf->lf_entries = cpu_to_be16(dip->i_entries);
841
842 /* Copy dirents */
843
844 gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_leaf), dibh,
845 sizeof(struct gfs2_dinode));
846
847 /* Find last entry */
848
849 x = 0;
850 args.len = bh->b_size - sizeof(struct gfs2_dinode) +
851 sizeof(struct gfs2_leaf);
852 args.name = bh->b_data;
853 dent = gfs2_dirent_scan(&dip->i_inode, bh->b_data, bh->b_size,
854 gfs2_dirent_last, &args, NULL);
855 if (!dent) {
856 brelse(bh);
857 brelse(dibh);
858 return -EIO;
859 }
860 if (IS_ERR(dent)) {
861 brelse(bh);
862 brelse(dibh);
863 return PTR_ERR(dent);
864 }
865
866 /* Adjust the last dirent's record length
867 (Remember that dent still points to the last entry.) */
868
869 dent->de_rec_len = cpu_to_be16(be16_to_cpu(dent->de_rec_len) +
870 sizeof(struct gfs2_dinode) -
871 sizeof(struct gfs2_leaf));
872
873 brelse(bh);
874
875 /* We're done with the new leaf block, now setup the new
876 hash table. */
877
878 gfs2_trans_add_bh(dip->i_gl, dibh, 1);
879 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
880
881 lp = (__be64 *)(dibh->b_data + sizeof(struct gfs2_dinode));
882
883 for (x = sdp->sd_hash_ptrs; x--; lp++)
884 *lp = cpu_to_be64(bn);
885
886 i_size_write(inode, sdp->sd_sb.sb_bsize / 2);
887 gfs2_add_inode_blocks(&dip->i_inode, 1);
888 dip->i_diskflags |= GFS2_DIF_EXHASH;
889
890 for (x = sdp->sd_hash_ptrs, y = -1; x; x >>= 1, y++) ;
891 dip->i_depth = y;
892
893 gfs2_dinode_out(dip, dibh->b_data);
894
895 brelse(dibh);
896
897 return 0;
898 }
899
900 /**
901 * dir_split_leaf - Split a leaf block into two
902 * @dip: The GFS2 inode
903 * @index:
904 * @leaf_no:
905 *
906 * Returns: 0 on success, error code on failure
907 */
908
909 static int dir_split_leaf(struct inode *inode, const struct qstr *name)
910 {
911 struct gfs2_inode *dip = GFS2_I(inode);
912 struct buffer_head *nbh, *obh, *dibh;
913 struct gfs2_leaf *nleaf, *oleaf;
914 struct gfs2_dirent *dent = NULL, *prev = NULL, *next = NULL, *new;
915 u32 start, len, half_len, divider;
916 u64 bn, leaf_no;
917 __be64 *lp;
918 u32 index;
919 int x, moved = 0;
920 int error;
921
922 index = name->hash >> (32 - dip->i_depth);
923 error = get_leaf_nr(dip, index, &leaf_no);
924 if (error)
925 return error;
926
927 /* Get the old leaf block */
928 error = get_leaf(dip, leaf_no, &obh);
929 if (error)
930 return error;
931
932 oleaf = (struct gfs2_leaf *)obh->b_data;
933 if (dip->i_depth == be16_to_cpu(oleaf->lf_depth)) {
934 brelse(obh);
935 return 1; /* can't split */
936 }
937
938 gfs2_trans_add_bh(dip->i_gl, obh, 1);
939
940 nleaf = new_leaf(inode, &nbh, be16_to_cpu(oleaf->lf_depth) + 1);
941 if (!nleaf) {
942 brelse(obh);
943 return -ENOSPC;
944 }
945 bn = nbh->b_blocknr;
946
947 /* Compute the start and len of leaf pointers in the hash table. */
948 len = 1 << (dip->i_depth - be16_to_cpu(oleaf->lf_depth));
949 half_len = len >> 1;
950 if (!half_len) {
951 printk(KERN_WARNING "i_depth %u lf_depth %u index %u\n", dip->i_depth, be16_to_cpu(oleaf->lf_depth), index);
952 gfs2_consist_inode(dip);
953 error = -EIO;
954 goto fail_brelse;
955 }
956
957 start = (index & ~(len - 1));
958
959 /* Change the pointers.
960 Don't bother distinguishing stuffed from non-stuffed.
961 This code is complicated enough already. */
962 lp = kmalloc(half_len * sizeof(__be64), GFP_NOFS);
963 if (!lp) {
964 error = -ENOMEM;
965 goto fail_brelse;
966 }
967
968 /* Change the pointers */
969 for (x = 0; x < half_len; x++)
970 lp[x] = cpu_to_be64(bn);
971
972 error = gfs2_dir_write_data(dip, (char *)lp, start * sizeof(u64),
973 half_len * sizeof(u64));
974 if (error != half_len * sizeof(u64)) {
975 if (error >= 0)
976 error = -EIO;
977 goto fail_lpfree;
978 }
979
980 kfree(lp);
981
982 /* Compute the divider */
983 divider = (start + half_len) << (32 - dip->i_depth);
984
985 /* Copy the entries */
986 dent = (struct gfs2_dirent *)(obh->b_data + sizeof(struct gfs2_leaf));
987
988 do {
989 next = dent;
990 if (dirent_next(dip, obh, &next))
991 next = NULL;
992
993 if (!gfs2_dirent_sentinel(dent) &&
994 be32_to_cpu(dent->de_hash) < divider) {
995 struct qstr str;
996 str.name = (char*)(dent+1);
997 str.len = be16_to_cpu(dent->de_name_len);
998 str.hash = be32_to_cpu(dent->de_hash);
999 new = gfs2_dirent_alloc(inode, nbh, &str);
1000 if (IS_ERR(new)) {
1001 error = PTR_ERR(new);
1002 break;
1003 }
1004
1005 new->de_inum = dent->de_inum; /* No endian worries */
1006 new->de_type = dent->de_type; /* No endian worries */
1007 be16_add_cpu(&nleaf->lf_entries, 1);
1008
1009 dirent_del(dip, obh, prev, dent);
1010
1011 if (!oleaf->lf_entries)
1012 gfs2_consist_inode(dip);
1013 be16_add_cpu(&oleaf->lf_entries, -1);
1014
1015 if (!prev)
1016 prev = dent;
1017
1018 moved = 1;
1019 } else {
1020 prev = dent;
1021 }
1022 dent = next;
1023 } while (dent);
1024
1025 oleaf->lf_depth = nleaf->lf_depth;
1026
1027 error = gfs2_meta_inode_buffer(dip, &dibh);
1028 if (!gfs2_assert_withdraw(GFS2_SB(&dip->i_inode), !error)) {
1029 gfs2_trans_add_bh(dip->i_gl, dibh, 1);
1030 gfs2_add_inode_blocks(&dip->i_inode, 1);
1031 gfs2_dinode_out(dip, dibh->b_data);
1032 brelse(dibh);
1033 }
1034
1035 brelse(obh);
1036 brelse(nbh);
1037
1038 return error;
1039
1040 fail_lpfree:
1041 kfree(lp);
1042
1043 fail_brelse:
1044 brelse(obh);
1045 brelse(nbh);
1046 return error;
1047 }
1048
1049 /**
1050 * dir_double_exhash - Double size of ExHash table
1051 * @dip: The GFS2 dinode
1052 *
1053 * Returns: 0 on success, error code on failure
1054 */
1055
1056 static int dir_double_exhash(struct gfs2_inode *dip)
1057 {
1058 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1059 struct buffer_head *dibh;
1060 u32 hsize;
1061 u64 *buf;
1062 u64 *from, *to;
1063 u64 block;
1064 u64 disksize = i_size_read(&dip->i_inode);
1065 int x;
1066 int error = 0;
1067
1068 hsize = 1 << dip->i_depth;
1069 if (hsize * sizeof(u64) != disksize) {
1070 gfs2_consist_inode(dip);
1071 return -EIO;
1072 }
1073
1074 /* Allocate both the "from" and "to" buffers in one big chunk */
1075
1076 buf = kcalloc(3, sdp->sd_hash_bsize, GFP_NOFS);
1077 if (!buf)
1078 return -ENOMEM;
1079
1080 for (block = disksize >> sdp->sd_hash_bsize_shift; block--;) {
1081 error = gfs2_dir_read_data(dip, (char *)buf,
1082 block * sdp->sd_hash_bsize,
1083 sdp->sd_hash_bsize, 1);
1084 if (error != sdp->sd_hash_bsize) {
1085 if (error >= 0)
1086 error = -EIO;
1087 goto fail;
1088 }
1089
1090 from = buf;
1091 to = (u64 *)((char *)buf + sdp->sd_hash_bsize);
1092
1093 for (x = sdp->sd_hash_ptrs; x--; from++) {
1094 *to++ = *from; /* No endianess worries */
1095 *to++ = *from;
1096 }
1097
1098 error = gfs2_dir_write_data(dip,
1099 (char *)buf + sdp->sd_hash_bsize,
1100 block * sdp->sd_sb.sb_bsize,
1101 sdp->sd_sb.sb_bsize);
1102 if (error != sdp->sd_sb.sb_bsize) {
1103 if (error >= 0)
1104 error = -EIO;
1105 goto fail;
1106 }
1107 }
1108
1109 kfree(buf);
1110
1111 error = gfs2_meta_inode_buffer(dip, &dibh);
1112 if (!gfs2_assert_withdraw(sdp, !error)) {
1113 dip->i_depth++;
1114 gfs2_dinode_out(dip, dibh->b_data);
1115 brelse(dibh);
1116 }
1117
1118 return error;
1119
1120 fail:
1121 kfree(buf);
1122 return error;
1123 }
1124
1125 /**
1126 * compare_dents - compare directory entries by hash value
1127 * @a: first dent
1128 * @b: second dent
1129 *
1130 * When comparing the hash entries of @a to @b:
1131 * gt: returns 1
1132 * lt: returns -1
1133 * eq: returns 0
1134 */
1135
1136 static int compare_dents(const void *a, const void *b)
1137 {
1138 const struct gfs2_dirent *dent_a, *dent_b;
1139 u32 hash_a, hash_b;
1140 int ret = 0;
1141
1142 dent_a = *(const struct gfs2_dirent **)a;
1143 hash_a = be32_to_cpu(dent_a->de_hash);
1144
1145 dent_b = *(const struct gfs2_dirent **)b;
1146 hash_b = be32_to_cpu(dent_b->de_hash);
1147
1148 if (hash_a > hash_b)
1149 ret = 1;
1150 else if (hash_a < hash_b)
1151 ret = -1;
1152 else {
1153 unsigned int len_a = be16_to_cpu(dent_a->de_name_len);
1154 unsigned int len_b = be16_to_cpu(dent_b->de_name_len);
1155
1156 if (len_a > len_b)
1157 ret = 1;
1158 else if (len_a < len_b)
1159 ret = -1;
1160 else
1161 ret = memcmp(dent_a + 1, dent_b + 1, len_a);
1162 }
1163
1164 return ret;
1165 }
1166
1167 /**
1168 * do_filldir_main - read out directory entries
1169 * @dip: The GFS2 inode
1170 * @offset: The offset in the file to read from
1171 * @opaque: opaque data to pass to filldir
1172 * @filldir: The function to pass entries to
1173 * @darr: an array of struct gfs2_dirent pointers to read
1174 * @entries: the number of entries in darr
1175 * @copied: pointer to int that's non-zero if a entry has been copied out
1176 *
1177 * Jump through some hoops to make sure that if there are hash collsions,
1178 * they are read out at the beginning of a buffer. We want to minimize
1179 * the possibility that they will fall into different readdir buffers or
1180 * that someone will want to seek to that location.
1181 *
1182 * Returns: errno, >0 on exception from filldir
1183 */
1184
1185 static int do_filldir_main(struct gfs2_inode *dip, u64 *offset,
1186 void *opaque, filldir_t filldir,
1187 const struct gfs2_dirent **darr, u32 entries,
1188 int *copied)
1189 {
1190 const struct gfs2_dirent *dent, *dent_next;
1191 u64 off, off_next;
1192 unsigned int x, y;
1193 int run = 0;
1194 int error = 0;
1195
1196 sort(darr, entries, sizeof(struct gfs2_dirent *), compare_dents, NULL);
1197
1198 dent_next = darr[0];
1199 off_next = be32_to_cpu(dent_next->de_hash);
1200 off_next = gfs2_disk_hash2offset(off_next);
1201
1202 for (x = 0, y = 1; x < entries; x++, y++) {
1203 dent = dent_next;
1204 off = off_next;
1205
1206 if (y < entries) {
1207 dent_next = darr[y];
1208 off_next = be32_to_cpu(dent_next->de_hash);
1209 off_next = gfs2_disk_hash2offset(off_next);
1210
1211 if (off < *offset)
1212 continue;
1213 *offset = off;
1214
1215 if (off_next == off) {
1216 if (*copied && !run)
1217 return 1;
1218 run = 1;
1219 } else
1220 run = 0;
1221 } else {
1222 if (off < *offset)
1223 continue;
1224 *offset = off;
1225 }
1226
1227 error = filldir(opaque, (const char *)(dent + 1),
1228 be16_to_cpu(dent->de_name_len),
1229 off, be64_to_cpu(dent->de_inum.no_addr),
1230 be16_to_cpu(dent->de_type));
1231 if (error)
1232 return 1;
1233
1234 *copied = 1;
1235 }
1236
1237 /* Increment the *offset by one, so the next time we come into the
1238 do_filldir fxn, we get the next entry instead of the last one in the
1239 current leaf */
1240
1241 (*offset)++;
1242
1243 return 0;
1244 }
1245
1246 static void *gfs2_alloc_sort_buffer(unsigned size)
1247 {
1248 void *ptr = NULL;
1249
1250 if (size < KMALLOC_MAX_SIZE)
1251 ptr = kmalloc(size, GFP_NOFS | __GFP_NOWARN);
1252 if (!ptr)
1253 ptr = __vmalloc(size, GFP_NOFS, PAGE_KERNEL);
1254 return ptr;
1255 }
1256
1257 static void gfs2_free_sort_buffer(void *ptr)
1258 {
1259 if (is_vmalloc_addr(ptr))
1260 vfree(ptr);
1261 else
1262 kfree(ptr);
1263 }
1264
1265 static int gfs2_dir_read_leaf(struct inode *inode, u64 *offset, void *opaque,
1266 filldir_t filldir, int *copied, unsigned *depth,
1267 u64 leaf_no)
1268 {
1269 struct gfs2_inode *ip = GFS2_I(inode);
1270 struct gfs2_sbd *sdp = GFS2_SB(inode);
1271 struct buffer_head *bh;
1272 struct gfs2_leaf *lf;
1273 unsigned entries = 0, entries2 = 0;
1274 unsigned leaves = 0;
1275 const struct gfs2_dirent **darr, *dent;
1276 struct dirent_gather g;
1277 struct buffer_head **larr;
1278 int leaf = 0;
1279 int error, i;
1280 u64 lfn = leaf_no;
1281
1282 do {
1283 error = get_leaf(ip, lfn, &bh);
1284 if (error)
1285 goto out;
1286 lf = (struct gfs2_leaf *)bh->b_data;
1287 if (leaves == 0)
1288 *depth = be16_to_cpu(lf->lf_depth);
1289 entries += be16_to_cpu(lf->lf_entries);
1290 leaves++;
1291 lfn = be64_to_cpu(lf->lf_next);
1292 brelse(bh);
1293 } while(lfn);
1294
1295 if (!entries)
1296 return 0;
1297
1298 error = -ENOMEM;
1299 /*
1300 * The extra 99 entries are not normally used, but are a buffer
1301 * zone in case the number of entries in the leaf is corrupt.
1302 * 99 is the maximum number of entries that can fit in a single
1303 * leaf block.
1304 */
1305 larr = gfs2_alloc_sort_buffer((leaves + entries + 99) * sizeof(void *));
1306 if (!larr)
1307 goto out;
1308 darr = (const struct gfs2_dirent **)(larr + leaves);
1309 g.pdent = darr;
1310 g.offset = 0;
1311 lfn = leaf_no;
1312
1313 do {
1314 error = get_leaf(ip, lfn, &bh);
1315 if (error)
1316 goto out_free;
1317 lf = (struct gfs2_leaf *)bh->b_data;
1318 lfn = be64_to_cpu(lf->lf_next);
1319 if (lf->lf_entries) {
1320 entries2 += be16_to_cpu(lf->lf_entries);
1321 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
1322 gfs2_dirent_gather, NULL, &g);
1323 error = PTR_ERR(dent);
1324 if (IS_ERR(dent))
1325 goto out_free;
1326 if (entries2 != g.offset) {
1327 fs_warn(sdp, "Number of entries corrupt in dir "
1328 "leaf %llu, entries2 (%u) != "
1329 "g.offset (%u)\n",
1330 (unsigned long long)bh->b_blocknr,
1331 entries2, g.offset);
1332
1333 error = -EIO;
1334 goto out_free;
1335 }
1336 error = 0;
1337 larr[leaf++] = bh;
1338 } else {
1339 brelse(bh);
1340 }
1341 } while(lfn);
1342
1343 BUG_ON(entries2 != entries);
1344 error = do_filldir_main(ip, offset, opaque, filldir, darr,
1345 entries, copied);
1346 out_free:
1347 for(i = 0; i < leaf; i++)
1348 brelse(larr[i]);
1349 gfs2_free_sort_buffer(larr);
1350 out:
1351 return error;
1352 }
1353
1354 /**
1355 * dir_e_read - Reads the entries from a directory into a filldir buffer
1356 * @dip: dinode pointer
1357 * @offset: the hash of the last entry read shifted to the right once
1358 * @opaque: buffer for the filldir function to fill
1359 * @filldir: points to the filldir function to use
1360 *
1361 * Returns: errno
1362 */
1363
1364 static int dir_e_read(struct inode *inode, u64 *offset, void *opaque,
1365 filldir_t filldir)
1366 {
1367 struct gfs2_inode *dip = GFS2_I(inode);
1368 struct gfs2_sbd *sdp = GFS2_SB(inode);
1369 u32 hsize, len = 0;
1370 u32 ht_offset, lp_offset, ht_offset_cur = -1;
1371 u32 hash, index;
1372 __be64 *lp;
1373 int copied = 0;
1374 int error = 0;
1375 unsigned depth = 0;
1376
1377 hsize = 1 << dip->i_depth;
1378 if (hsize * sizeof(u64) != i_size_read(inode)) {
1379 gfs2_consist_inode(dip);
1380 return -EIO;
1381 }
1382
1383 hash = gfs2_dir_offset2hash(*offset);
1384 index = hash >> (32 - dip->i_depth);
1385
1386 lp = kmalloc(sdp->sd_hash_bsize, GFP_NOFS);
1387 if (!lp)
1388 return -ENOMEM;
1389
1390 while (index < hsize) {
1391 lp_offset = index & (sdp->sd_hash_ptrs - 1);
1392 ht_offset = index - lp_offset;
1393
1394 if (ht_offset_cur != ht_offset) {
1395 error = gfs2_dir_read_data(dip, (char *)lp,
1396 ht_offset * sizeof(__be64),
1397 sdp->sd_hash_bsize, 1);
1398 if (error != sdp->sd_hash_bsize) {
1399 if (error >= 0)
1400 error = -EIO;
1401 goto out;
1402 }
1403 ht_offset_cur = ht_offset;
1404 }
1405
1406 error = gfs2_dir_read_leaf(inode, offset, opaque, filldir,
1407 &copied, &depth,
1408 be64_to_cpu(lp[lp_offset]));
1409 if (error)
1410 break;
1411
1412 len = 1 << (dip->i_depth - depth);
1413 index = (index & ~(len - 1)) + len;
1414 }
1415
1416 out:
1417 kfree(lp);
1418 if (error > 0)
1419 error = 0;
1420 return error;
1421 }
1422
1423 int gfs2_dir_read(struct inode *inode, u64 *offset, void *opaque,
1424 filldir_t filldir)
1425 {
1426 struct gfs2_inode *dip = GFS2_I(inode);
1427 struct gfs2_sbd *sdp = GFS2_SB(inode);
1428 struct dirent_gather g;
1429 const struct gfs2_dirent **darr, *dent;
1430 struct buffer_head *dibh;
1431 int copied = 0;
1432 int error;
1433
1434 if (!dip->i_entries)
1435 return 0;
1436
1437 if (dip->i_diskflags & GFS2_DIF_EXHASH)
1438 return dir_e_read(inode, offset, opaque, filldir);
1439
1440 if (!gfs2_is_stuffed(dip)) {
1441 gfs2_consist_inode(dip);
1442 return -EIO;
1443 }
1444
1445 error = gfs2_meta_inode_buffer(dip, &dibh);
1446 if (error)
1447 return error;
1448
1449 error = -ENOMEM;
1450 /* 96 is max number of dirents which can be stuffed into an inode */
1451 darr = kmalloc(96 * sizeof(struct gfs2_dirent *), GFP_NOFS);
1452 if (darr) {
1453 g.pdent = darr;
1454 g.offset = 0;
1455 dent = gfs2_dirent_scan(inode, dibh->b_data, dibh->b_size,
1456 gfs2_dirent_gather, NULL, &g);
1457 if (IS_ERR(dent)) {
1458 error = PTR_ERR(dent);
1459 goto out;
1460 }
1461 if (dip->i_entries != g.offset) {
1462 fs_warn(sdp, "Number of entries corrupt in dir %llu, "
1463 "ip->i_entries (%u) != g.offset (%u)\n",
1464 (unsigned long long)dip->i_no_addr,
1465 dip->i_entries,
1466 g.offset);
1467 error = -EIO;
1468 goto out;
1469 }
1470 error = do_filldir_main(dip, offset, opaque, filldir, darr,
1471 dip->i_entries, &copied);
1472 out:
1473 kfree(darr);
1474 }
1475
1476 if (error > 0)
1477 error = 0;
1478
1479 brelse(dibh);
1480
1481 return error;
1482 }
1483
1484 /**
1485 * gfs2_dir_search - Search a directory
1486 * @dip: The GFS2 inode
1487 * @filename:
1488 * @inode:
1489 *
1490 * This routine searches a directory for a file or another directory.
1491 * Assumes a glock is held on dip.
1492 *
1493 * Returns: errno
1494 */
1495
1496 struct inode *gfs2_dir_search(struct inode *dir, const struct qstr *name)
1497 {
1498 struct buffer_head *bh;
1499 struct gfs2_dirent *dent;
1500 struct inode *inode;
1501
1502 dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
1503 if (dent) {
1504 if (IS_ERR(dent))
1505 return ERR_CAST(dent);
1506 inode = gfs2_inode_lookup(dir->i_sb,
1507 be16_to_cpu(dent->de_type),
1508 be64_to_cpu(dent->de_inum.no_addr),
1509 be64_to_cpu(dent->de_inum.no_formal_ino));
1510 brelse(bh);
1511 return inode;
1512 }
1513 return ERR_PTR(-ENOENT);
1514 }
1515
1516 int gfs2_dir_check(struct inode *dir, const struct qstr *name,
1517 const struct gfs2_inode *ip)
1518 {
1519 struct buffer_head *bh;
1520 struct gfs2_dirent *dent;
1521 int ret = -ENOENT;
1522
1523 dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
1524 if (dent) {
1525 if (IS_ERR(dent))
1526 return PTR_ERR(dent);
1527 if (ip) {
1528 if (be64_to_cpu(dent->de_inum.no_addr) != ip->i_no_addr)
1529 goto out;
1530 if (be64_to_cpu(dent->de_inum.no_formal_ino) !=
1531 ip->i_no_formal_ino)
1532 goto out;
1533 if (unlikely(IF2DT(ip->i_inode.i_mode) !=
1534 be16_to_cpu(dent->de_type))) {
1535 gfs2_consist_inode(GFS2_I(dir));
1536 ret = -EIO;
1537 goto out;
1538 }
1539 }
1540 ret = 0;
1541 out:
1542 brelse(bh);
1543 }
1544 return ret;
1545 }
1546
1547 static int dir_new_leaf(struct inode *inode, const struct qstr *name)
1548 {
1549 struct buffer_head *bh, *obh;
1550 struct gfs2_inode *ip = GFS2_I(inode);
1551 struct gfs2_leaf *leaf, *oleaf;
1552 int error;
1553 u32 index;
1554 u64 bn;
1555
1556 index = name->hash >> (32 - ip->i_depth);
1557 error = get_first_leaf(ip, index, &obh);
1558 if (error)
1559 return error;
1560 do {
1561 oleaf = (struct gfs2_leaf *)obh->b_data;
1562 bn = be64_to_cpu(oleaf->lf_next);
1563 if (!bn)
1564 break;
1565 brelse(obh);
1566 error = get_leaf(ip, bn, &obh);
1567 if (error)
1568 return error;
1569 } while(1);
1570
1571 gfs2_trans_add_bh(ip->i_gl, obh, 1);
1572
1573 leaf = new_leaf(inode, &bh, be16_to_cpu(oleaf->lf_depth));
1574 if (!leaf) {
1575 brelse(obh);
1576 return -ENOSPC;
1577 }
1578 oleaf->lf_next = cpu_to_be64(bh->b_blocknr);
1579 brelse(bh);
1580 brelse(obh);
1581
1582 error = gfs2_meta_inode_buffer(ip, &bh);
1583 if (error)
1584 return error;
1585 gfs2_trans_add_bh(ip->i_gl, bh, 1);
1586 gfs2_add_inode_blocks(&ip->i_inode, 1);
1587 gfs2_dinode_out(ip, bh->b_data);
1588 brelse(bh);
1589 return 0;
1590 }
1591
1592 /**
1593 * gfs2_dir_add - Add new filename into directory
1594 * @dip: The GFS2 inode
1595 * @filename: The new name
1596 * @inode: The inode number of the entry
1597 * @type: The type of the entry
1598 *
1599 * Returns: 0 on success, error code on failure
1600 */
1601
1602 int gfs2_dir_add(struct inode *inode, const struct qstr *name,
1603 const struct gfs2_inode *nip, unsigned type)
1604 {
1605 struct gfs2_inode *ip = GFS2_I(inode);
1606 struct buffer_head *bh;
1607 struct gfs2_dirent *dent;
1608 struct gfs2_leaf *leaf;
1609 int error;
1610
1611 while(1) {
1612 dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space,
1613 &bh);
1614 if (dent) {
1615 if (IS_ERR(dent))
1616 return PTR_ERR(dent);
1617 dent = gfs2_init_dirent(inode, dent, name, bh);
1618 gfs2_inum_out(nip, dent);
1619 dent->de_type = cpu_to_be16(type);
1620 if (ip->i_diskflags & GFS2_DIF_EXHASH) {
1621 leaf = (struct gfs2_leaf *)bh->b_data;
1622 be16_add_cpu(&leaf->lf_entries, 1);
1623 }
1624 brelse(bh);
1625 error = gfs2_meta_inode_buffer(ip, &bh);
1626 if (error)
1627 break;
1628 gfs2_trans_add_bh(ip->i_gl, bh, 1);
1629 ip->i_entries++;
1630 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
1631 gfs2_dinode_out(ip, bh->b_data);
1632 brelse(bh);
1633 error = 0;
1634 break;
1635 }
1636 if (!(ip->i_diskflags & GFS2_DIF_EXHASH)) {
1637 error = dir_make_exhash(inode);
1638 if (error)
1639 break;
1640 continue;
1641 }
1642 error = dir_split_leaf(inode, name);
1643 if (error == 0)
1644 continue;
1645 if (error < 0)
1646 break;
1647 if (ip->i_depth < GFS2_DIR_MAX_DEPTH) {
1648 error = dir_double_exhash(ip);
1649 if (error)
1650 break;
1651 error = dir_split_leaf(inode, name);
1652 if (error < 0)
1653 break;
1654 if (error == 0)
1655 continue;
1656 }
1657 error = dir_new_leaf(inode, name);
1658 if (!error)
1659 continue;
1660 error = -ENOSPC;
1661 break;
1662 }
1663 return error;
1664 }
1665
1666
1667 /**
1668 * gfs2_dir_del - Delete a directory entry
1669 * @dip: The GFS2 inode
1670 * @filename: The filename
1671 *
1672 * Returns: 0 on success, error code on failure
1673 */
1674
1675 int gfs2_dir_del(struct gfs2_inode *dip, const struct qstr *name)
1676 {
1677 struct gfs2_dirent *dent, *prev = NULL;
1678 struct buffer_head *bh;
1679 int error;
1680
1681 /* Returns _either_ the entry (if its first in block) or the
1682 previous entry otherwise */
1683 dent = gfs2_dirent_search(&dip->i_inode, name, gfs2_dirent_prev, &bh);
1684 if (!dent) {
1685 gfs2_consist_inode(dip);
1686 return -EIO;
1687 }
1688 if (IS_ERR(dent)) {
1689 gfs2_consist_inode(dip);
1690 return PTR_ERR(dent);
1691 }
1692 /* If not first in block, adjust pointers accordingly */
1693 if (gfs2_dirent_find(dent, name, NULL) == 0) {
1694 prev = dent;
1695 dent = (struct gfs2_dirent *)((char *)dent + be16_to_cpu(prev->de_rec_len));
1696 }
1697
1698 dirent_del(dip, bh, prev, dent);
1699 if (dip->i_diskflags & GFS2_DIF_EXHASH) {
1700 struct gfs2_leaf *leaf = (struct gfs2_leaf *)bh->b_data;
1701 u16 entries = be16_to_cpu(leaf->lf_entries);
1702 if (!entries)
1703 gfs2_consist_inode(dip);
1704 leaf->lf_entries = cpu_to_be16(--entries);
1705 }
1706 brelse(bh);
1707
1708 error = gfs2_meta_inode_buffer(dip, &bh);
1709 if (error)
1710 return error;
1711
1712 if (!dip->i_entries)
1713 gfs2_consist_inode(dip);
1714 gfs2_trans_add_bh(dip->i_gl, bh, 1);
1715 dip->i_entries--;
1716 dip->i_inode.i_mtime = dip->i_inode.i_ctime = CURRENT_TIME;
1717 gfs2_dinode_out(dip, bh->b_data);
1718 brelse(bh);
1719 mark_inode_dirty(&dip->i_inode);
1720
1721 return error;
1722 }
1723
1724 /**
1725 * gfs2_dir_mvino - Change inode number of directory entry
1726 * @dip: The GFS2 inode
1727 * @filename:
1728 * @new_inode:
1729 *
1730 * This routine changes the inode number of a directory entry. It's used
1731 * by rename to change ".." when a directory is moved.
1732 * Assumes a glock is held on dvp.
1733 *
1734 * Returns: errno
1735 */
1736
1737 int gfs2_dir_mvino(struct gfs2_inode *dip, const struct qstr *filename,
1738 const struct gfs2_inode *nip, unsigned int new_type)
1739 {
1740 struct buffer_head *bh;
1741 struct gfs2_dirent *dent;
1742 int error;
1743
1744 dent = gfs2_dirent_search(&dip->i_inode, filename, gfs2_dirent_find, &bh);
1745 if (!dent) {
1746 gfs2_consist_inode(dip);
1747 return -EIO;
1748 }
1749 if (IS_ERR(dent))
1750 return PTR_ERR(dent);
1751
1752 gfs2_trans_add_bh(dip->i_gl, bh, 1);
1753 gfs2_inum_out(nip, dent);
1754 dent->de_type = cpu_to_be16(new_type);
1755
1756 if (dip->i_diskflags & GFS2_DIF_EXHASH) {
1757 brelse(bh);
1758 error = gfs2_meta_inode_buffer(dip, &bh);
1759 if (error)
1760 return error;
1761 gfs2_trans_add_bh(dip->i_gl, bh, 1);
1762 }
1763
1764 dip->i_inode.i_mtime = dip->i_inode.i_ctime = CURRENT_TIME;
1765 gfs2_dinode_out(dip, bh->b_data);
1766 brelse(bh);
1767 return 0;
1768 }
1769
1770 /**
1771 * foreach_leaf - call a function for each leaf in a directory
1772 * @dip: the directory
1773 * @lc: the function to call for each each
1774 * @data: private data to pass to it
1775 *
1776 * Returns: errno
1777 */
1778
1779 static int foreach_leaf(struct gfs2_inode *dip, leaf_call_t lc, void *data)
1780 {
1781 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1782 struct buffer_head *bh;
1783 struct gfs2_leaf *leaf;
1784 u32 hsize, len;
1785 u32 ht_offset, lp_offset, ht_offset_cur = -1;
1786 u32 index = 0;
1787 __be64 *lp;
1788 u64 leaf_no;
1789 int error = 0;
1790
1791 hsize = 1 << dip->i_depth;
1792 if (hsize * sizeof(u64) != i_size_read(&dip->i_inode)) {
1793 gfs2_consist_inode(dip);
1794 return -EIO;
1795 }
1796
1797 lp = kmalloc(sdp->sd_hash_bsize, GFP_NOFS);
1798 if (!lp)
1799 return -ENOMEM;
1800
1801 while (index < hsize) {
1802 lp_offset = index & (sdp->sd_hash_ptrs - 1);
1803 ht_offset = index - lp_offset;
1804
1805 if (ht_offset_cur != ht_offset) {
1806 error = gfs2_dir_read_data(dip, (char *)lp,
1807 ht_offset * sizeof(__be64),
1808 sdp->sd_hash_bsize, 1);
1809 if (error != sdp->sd_hash_bsize) {
1810 if (error >= 0)
1811 error = -EIO;
1812 goto out;
1813 }
1814 ht_offset_cur = ht_offset;
1815 }
1816
1817 leaf_no = be64_to_cpu(lp[lp_offset]);
1818 if (leaf_no) {
1819 error = get_leaf(dip, leaf_no, &bh);
1820 if (error)
1821 goto out;
1822 leaf = (struct gfs2_leaf *)bh->b_data;
1823 len = 1 << (dip->i_depth - be16_to_cpu(leaf->lf_depth));
1824 brelse(bh);
1825
1826 error = lc(dip, index, len, leaf_no, data);
1827 if (error)
1828 goto out;
1829
1830 index = (index & ~(len - 1)) + len;
1831 } else
1832 index++;
1833 }
1834
1835 if (index != hsize) {
1836 gfs2_consist_inode(dip);
1837 error = -EIO;
1838 }
1839
1840 out:
1841 kfree(lp);
1842
1843 return error;
1844 }
1845
1846 /**
1847 * leaf_dealloc - Deallocate a directory leaf
1848 * @dip: the directory
1849 * @index: the hash table offset in the directory
1850 * @len: the number of pointers to this leaf
1851 * @leaf_no: the leaf number
1852 * @data: not used
1853 *
1854 * Returns: errno
1855 */
1856
1857 static int leaf_dealloc(struct gfs2_inode *dip, u32 index, u32 len,
1858 u64 leaf_no, void *data)
1859 {
1860 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1861 struct gfs2_leaf *tmp_leaf;
1862 struct gfs2_rgrp_list rlist;
1863 struct buffer_head *bh, *dibh;
1864 u64 blk, nblk;
1865 unsigned int rg_blocks = 0, l_blocks = 0;
1866 char *ht;
1867 unsigned int x, size = len * sizeof(u64);
1868 int error;
1869
1870 memset(&rlist, 0, sizeof(struct gfs2_rgrp_list));
1871
1872 ht = kzalloc(size, GFP_NOFS);
1873 if (!ht)
1874 return -ENOMEM;
1875
1876 if (!gfs2_alloc_get(dip)) {
1877 error = -ENOMEM;
1878 goto out;
1879 }
1880
1881 error = gfs2_quota_hold(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
1882 if (error)
1883 goto out_put;
1884
1885 error = gfs2_rindex_hold(sdp, &dip->i_alloc->al_ri_gh);
1886 if (error)
1887 goto out_qs;
1888
1889 /* Count the number of leaves */
1890
1891 for (blk = leaf_no; blk; blk = nblk) {
1892 error = get_leaf(dip, blk, &bh);
1893 if (error)
1894 goto out_rlist;
1895 tmp_leaf = (struct gfs2_leaf *)bh->b_data;
1896 nblk = be64_to_cpu(tmp_leaf->lf_next);
1897 brelse(bh);
1898
1899 gfs2_rlist_add(sdp, &rlist, blk);
1900 l_blocks++;
1901 }
1902
1903 gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE);
1904
1905 for (x = 0; x < rlist.rl_rgrps; x++) {
1906 struct gfs2_rgrpd *rgd;
1907 rgd = rlist.rl_ghs[x].gh_gl->gl_object;
1908 rg_blocks += rgd->rd_length;
1909 }
1910
1911 error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs);
1912 if (error)
1913 goto out_rlist;
1914
1915 error = gfs2_trans_begin(sdp,
1916 rg_blocks + (DIV_ROUND_UP(size, sdp->sd_jbsize) + 1) +
1917 RES_DINODE + RES_STATFS + RES_QUOTA, l_blocks);
1918 if (error)
1919 goto out_rg_gunlock;
1920
1921 for (blk = leaf_no; blk; blk = nblk) {
1922 error = get_leaf(dip, blk, &bh);
1923 if (error)
1924 goto out_end_trans;
1925 tmp_leaf = (struct gfs2_leaf *)bh->b_data;
1926 nblk = be64_to_cpu(tmp_leaf->lf_next);
1927 brelse(bh);
1928
1929 gfs2_free_meta(dip, blk, 1);
1930 gfs2_add_inode_blocks(&dip->i_inode, -1);
1931 }
1932
1933 error = gfs2_dir_write_data(dip, ht, index * sizeof(u64), size);
1934 if (error != size) {
1935 if (error >= 0)
1936 error = -EIO;
1937 goto out_end_trans;
1938 }
1939
1940 error = gfs2_meta_inode_buffer(dip, &dibh);
1941 if (error)
1942 goto out_end_trans;
1943
1944 gfs2_trans_add_bh(dip->i_gl, dibh, 1);
1945 gfs2_dinode_out(dip, dibh->b_data);
1946 brelse(dibh);
1947
1948 out_end_trans:
1949 gfs2_trans_end(sdp);
1950 out_rg_gunlock:
1951 gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
1952 out_rlist:
1953 gfs2_rlist_free(&rlist);
1954 gfs2_glock_dq_uninit(&dip->i_alloc->al_ri_gh);
1955 out_qs:
1956 gfs2_quota_unhold(dip);
1957 out_put:
1958 gfs2_alloc_put(dip);
1959 out:
1960 kfree(ht);
1961 return error;
1962 }
1963
1964 /**
1965 * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory
1966 * @dip: the directory
1967 *
1968 * Dealloc all on-disk directory leaves to FREEMETA state
1969 * Change on-disk inode type to "regular file"
1970 *
1971 * Returns: errno
1972 */
1973
1974 int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip)
1975 {
1976 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1977 struct buffer_head *bh;
1978 int error;
1979
1980 /* Dealloc on-disk leaves to FREEMETA state */
1981 error = foreach_leaf(dip, leaf_dealloc, NULL);
1982 if (error)
1983 return error;
1984
1985 /* Make this a regular file in case we crash.
1986 (We don't want to free these blocks a second time.) */
1987
1988 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1989 if (error)
1990 return error;
1991
1992 error = gfs2_meta_inode_buffer(dip, &bh);
1993 if (!error) {
1994 gfs2_trans_add_bh(dip->i_gl, bh, 1);
1995 ((struct gfs2_dinode *)bh->b_data)->di_mode =
1996 cpu_to_be32(S_IFREG);
1997 brelse(bh);
1998 }
1999
2000 gfs2_trans_end(sdp);
2001
2002 return error;
2003 }
2004
2005 /**
2006 * gfs2_diradd_alloc_required - find if adding entry will require an allocation
2007 * @ip: the file being written to
2008 * @filname: the filename that's going to be added
2009 *
2010 * Returns: 1 if alloc required, 0 if not, -ve on error
2011 */
2012
2013 int gfs2_diradd_alloc_required(struct inode *inode, const struct qstr *name)
2014 {
2015 struct gfs2_dirent *dent;
2016 struct buffer_head *bh;
2017
2018 dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, &bh);
2019 if (!dent) {
2020 return 1;
2021 }
2022 if (IS_ERR(dent))
2023 return PTR_ERR(dent);
2024 brelse(bh);
2025 return 0;
2026 }
2027
Linux with added FPC-III support