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 typedef int (*leaf_call_t) (struct gfs2_inode *dip, u32 index, u32 len,
83 u64 leaf_no, void *data);
84 typedef int (*gfs2_dscan_t)(const struct gfs2_dirent *dent,
85 const struct qstr *name, void *opaque);
86
87
88 int gfs2_dir_get_new_buffer(struct gfs2_inode *ip, u64 block,
89 struct buffer_head **bhp)
90 {
91 struct buffer_head *bh;
92
93 bh = gfs2_meta_new(ip->i_gl, block);
94 gfs2_trans_add_bh(ip->i_gl, bh, 1);
95 gfs2_metatype_set(bh, GFS2_METATYPE_JD, GFS2_FORMAT_JD);
96 gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
97 *bhp = bh;
98 return 0;
99 }
100
101 static int gfs2_dir_get_existing_buffer(struct gfs2_inode *ip, u64 block,
102 struct buffer_head **bhp)
103 {
104 struct buffer_head *bh;
105 int error;
106
107 error = gfs2_meta_read(ip->i_gl, block, DIO_WAIT, &bh);
108 if (error)
109 return error;
110 if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), bh, GFS2_METATYPE_JD)) {
111 brelse(bh);
112 return -EIO;
113 }
114 *bhp = bh;
115 return 0;
116 }
117
118 static int gfs2_dir_write_stuffed(struct gfs2_inode *ip, const char *buf,
119 unsigned int offset, unsigned int size)
120 {
121 struct buffer_head *dibh;
122 int error;
123
124 error = gfs2_meta_inode_buffer(ip, &dibh);
125 if (error)
126 return error;
127
128 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
129 memcpy(dibh->b_data + offset + sizeof(struct gfs2_dinode), buf, size);
130 if (ip->i_disksize < offset + size)
131 ip->i_disksize = offset + size;
132 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
133 gfs2_dinode_out(ip, dibh->b_data);
134
135 brelse(dibh);
136
137 return size;
138 }
139
140
141
142 /**
143 * gfs2_dir_write_data - Write directory information to the inode
144 * @ip: The GFS2 inode
145 * @buf: The buffer containing information to be written
146 * @offset: The file offset to start writing at
147 * @size: The amount of data to write
148 *
149 * Returns: The number of bytes correctly written or error code
150 */
151 static int gfs2_dir_write_data(struct gfs2_inode *ip, const char *buf,
152 u64 offset, unsigned int size)
153 {
154 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
155 struct buffer_head *dibh;
156 u64 lblock, dblock;
157 u32 extlen = 0;
158 unsigned int o;
159 int copied = 0;
160 int error = 0;
161 int new = 0;
162
163 if (!size)
164 return 0;
165
166 if (gfs2_is_stuffed(ip) &&
167 offset + size <= sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode))
168 return gfs2_dir_write_stuffed(ip, buf, (unsigned int)offset,
169 size);
170
171 if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
172 return -EINVAL;
173
174 if (gfs2_is_stuffed(ip)) {
175 error = gfs2_unstuff_dinode(ip, NULL);
176 if (error)
177 return error;
178 }
179
180 lblock = offset;
181 o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
182
183 while (copied < size) {
184 unsigned int amount;
185 struct buffer_head *bh;
186
187 amount = size - copied;
188 if (amount > sdp->sd_sb.sb_bsize - o)
189 amount = sdp->sd_sb.sb_bsize - o;
190
191 if (!extlen) {
192 new = 1;
193 error = gfs2_extent_map(&ip->i_inode, lblock, &new,
194 &dblock, &extlen);
195 if (error)
196 goto fail;
197 error = -EIO;
198 if (gfs2_assert_withdraw(sdp, dblock))
199 goto fail;
200 }
201
202 if (amount == sdp->sd_jbsize || new)
203 error = gfs2_dir_get_new_buffer(ip, dblock, &bh);
204 else
205 error = gfs2_dir_get_existing_buffer(ip, dblock, &bh);
206
207 if (error)
208 goto fail;
209
210 gfs2_trans_add_bh(ip->i_gl, bh, 1);
211 memcpy(bh->b_data + o, buf, amount);
212 brelse(bh);
213
214 buf += amount;
215 copied += amount;
216 lblock++;
217 dblock++;
218 extlen--;
219
220 o = sizeof(struct gfs2_meta_header);
221 }
222
223 out:
224 error = gfs2_meta_inode_buffer(ip, &dibh);
225 if (error)
226 return error;
227
228 if (ip->i_disksize < offset + copied)
229 ip->i_disksize = offset + copied;
230 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
231
232 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
233 gfs2_dinode_out(ip, dibh->b_data);
234 brelse(dibh);
235
236 return copied;
237 fail:
238 if (copied)
239 goto out;
240 return error;
241 }
242
243 static int gfs2_dir_read_stuffed(struct gfs2_inode *ip, char *buf,
244 u64 offset, unsigned int size)
245 {
246 struct buffer_head *dibh;
247 int error;
248
249 error = gfs2_meta_inode_buffer(ip, &dibh);
250 if (!error) {
251 offset += sizeof(struct gfs2_dinode);
252 memcpy(buf, dibh->b_data + offset, size);
253 brelse(dibh);
254 }
255
256 return (error) ? error : size;
257 }
258
259
260 /**
261 * gfs2_dir_read_data - Read a data from a directory inode
262 * @ip: The GFS2 Inode
263 * @buf: The buffer to place result into
264 * @offset: File offset to begin jdata_readng from
265 * @size: Amount of data to transfer
266 *
267 * Returns: The amount of data actually copied or the error
268 */
269 static int gfs2_dir_read_data(struct gfs2_inode *ip, char *buf, u64 offset,
270 unsigned int size, unsigned ra)
271 {
272 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
273 u64 lblock, dblock;
274 u32 extlen = 0;
275 unsigned int o;
276 int copied = 0;
277 int error = 0;
278
279 if (offset >= ip->i_disksize)
280 return 0;
281
282 if (offset + size > ip->i_disksize)
283 size = ip->i_disksize - offset;
284
285 if (!size)
286 return 0;
287
288 if (gfs2_is_stuffed(ip))
289 return gfs2_dir_read_stuffed(ip, buf, offset, size);
290
291 if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
292 return -EINVAL;
293
294 lblock = offset;
295 o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
296
297 while (copied < size) {
298 unsigned int amount;
299 struct buffer_head *bh;
300 int new;
301
302 amount = size - copied;
303 if (amount > sdp->sd_sb.sb_bsize - o)
304 amount = sdp->sd_sb.sb_bsize - o;
305
306 if (!extlen) {
307 new = 0;
308 error = gfs2_extent_map(&ip->i_inode, lblock, &new,
309 &dblock, &extlen);
310 if (error || !dblock)
311 goto fail;
312 BUG_ON(extlen < 1);
313 if (!ra)
314 extlen = 1;
315 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
316 } else {
317 error = gfs2_meta_read(ip->i_gl, dblock, DIO_WAIT, &bh);
318 if (error)
319 goto fail;
320 }
321 error = gfs2_metatype_check(sdp, bh, GFS2_METATYPE_JD);
322 if (error) {
323 brelse(bh);
324 goto fail;
325 }
326 dblock++;
327 extlen--;
328 memcpy(buf, bh->b_data + o, amount);
329 brelse(bh);
330 buf += amount;
331 copied += amount;
332 lblock++;
333 o = sizeof(struct gfs2_meta_header);
334 }
335
336 return copied;
337 fail:
338 return (copied) ? copied : error;
339 }
340
341 static inline int gfs2_dirent_sentinel(const struct gfs2_dirent *dent)
342 {
343 return dent->de_inum.no_addr == 0 || dent->de_inum.no_formal_ino == 0;
344 }
345
346 static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent,
347 const struct qstr *name, int ret)
348 {
349 if (!gfs2_dirent_sentinel(dent) &&
350 be32_to_cpu(dent->de_hash) == name->hash &&
351 be16_to_cpu(dent->de_name_len) == name->len &&
352 memcmp(dent+1, name->name, name->len) == 0)
353 return ret;
354 return 0;
355 }
356
357 static int gfs2_dirent_find(const struct gfs2_dirent *dent,
358 const struct qstr *name,
359 void *opaque)
360 {
361 return __gfs2_dirent_find(dent, name, 1);
362 }
363
364 static int gfs2_dirent_prev(const struct gfs2_dirent *dent,
365 const struct qstr *name,
366 void *opaque)
367 {
368 return __gfs2_dirent_find(dent, name, 2);
369 }
370
371 /*
372 * name->name holds ptr to start of block.
373 * name->len holds size of block.
374 */
375 static int gfs2_dirent_last(const struct gfs2_dirent *dent,
376 const struct qstr *name,
377 void *opaque)
378 {
379 const char *start = name->name;
380 const char *end = (const char *)dent + be16_to_cpu(dent->de_rec_len);
381 if (name->len == (end - start))
382 return 1;
383 return 0;
384 }
385
386 static int gfs2_dirent_find_space(const struct gfs2_dirent *dent,
387 const struct qstr *name,
388 void *opaque)
389 {
390 unsigned required = GFS2_DIRENT_SIZE(name->len);
391 unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
392 unsigned totlen = be16_to_cpu(dent->de_rec_len);
393
394 if (gfs2_dirent_sentinel(dent))
395 actual = GFS2_DIRENT_SIZE(0);
396 if (totlen - actual >= required)
397 return 1;
398 return 0;
399 }
400
401 struct dirent_gather {
402 const struct gfs2_dirent **pdent;
403 unsigned offset;
404 };
405
406 static int gfs2_dirent_gather(const struct gfs2_dirent *dent,
407 const struct qstr *name,
408 void *opaque)
409 {
410 struct dirent_gather *g = opaque;
411 if (!gfs2_dirent_sentinel(dent)) {
412 g->pdent[g->offset++] = dent;
413 }
414 return 0;
415 }
416
417 /*
418 * Other possible things to check:
419 * - Inode located within filesystem size (and on valid block)
420 * - Valid directory entry type
421 * Not sure how heavy-weight we want to make this... could also check
422 * hash is correct for example, but that would take a lot of extra time.
423 * For now the most important thing is to check that the various sizes
424 * are correct.
425 */
426 static int gfs2_check_dirent(struct gfs2_dirent *dent, unsigned int offset,
427 unsigned int size, unsigned int len, int first)
428 {
429 const char *msg = "gfs2_dirent too small";
430 if (unlikely(size < sizeof(struct gfs2_dirent)))
431 goto error;
432 msg = "gfs2_dirent misaligned";
433 if (unlikely(offset & 0x7))
434 goto error;
435 msg = "gfs2_dirent points beyond end of block";
436 if (unlikely(offset + size > len))
437 goto error;
438 msg = "zero inode number";
439 if (unlikely(!first && gfs2_dirent_sentinel(dent)))
440 goto error;
441 msg = "name length is greater than space in dirent";
442 if (!gfs2_dirent_sentinel(dent) &&
443 unlikely(sizeof(struct gfs2_dirent)+be16_to_cpu(dent->de_name_len) >
444 size))
445 goto error;
446 return 0;
447 error:
448 printk(KERN_WARNING "gfs2_check_dirent: %s (%s)\n", msg,
449 first ? "first in block" : "not first in block");
450 return -EIO;
451 }
452
453 static int gfs2_dirent_offset(const void *buf)
454 {
455 const struct gfs2_meta_header *h = buf;
456 int offset;
457
458 BUG_ON(buf == NULL);
459
460 switch(be32_to_cpu(h->mh_type)) {
461 case GFS2_METATYPE_LF:
462 offset = sizeof(struct gfs2_leaf);
463 break;
464 case GFS2_METATYPE_DI:
465 offset = sizeof(struct gfs2_dinode);
466 break;
467 default:
468 goto wrong_type;
469 }
470 return offset;
471 wrong_type:
472 printk(KERN_WARNING "gfs2_scan_dirent: wrong block type %u\n",
473 be32_to_cpu(h->mh_type));
474 return -1;
475 }
476
477 static struct gfs2_dirent *gfs2_dirent_scan(struct inode *inode, void *buf,
478 unsigned int len, gfs2_dscan_t scan,
479 const struct qstr *name,
480 void *opaque)
481 {
482 struct gfs2_dirent *dent, *prev;
483 unsigned offset;
484 unsigned size;
485 int ret = 0;
486
487 ret = gfs2_dirent_offset(buf);
488 if (ret < 0)
489 goto consist_inode;
490
491 offset = ret;
492 prev = NULL;
493 dent = buf + offset;
494 size = be16_to_cpu(dent->de_rec_len);
495 if (gfs2_check_dirent(dent, offset, size, len, 1))
496 goto consist_inode;
497 do {
498 ret = scan(dent, name, opaque);
499 if (ret)
500 break;
501 offset += size;
502 if (offset == len)
503 break;
504 prev = dent;
505 dent = buf + offset;
506 size = be16_to_cpu(dent->de_rec_len);
507 if (gfs2_check_dirent(dent, offset, size, len, 0))
508 goto consist_inode;
509 } while(1);
510
511 switch(ret) {
512 case 0:
513 return NULL;
514 case 1:
515 return dent;
516 case 2:
517 return prev ? prev : dent;
518 default:
519 BUG_ON(ret > 0);
520 return ERR_PTR(ret);
521 }
522
523 consist_inode:
524 gfs2_consist_inode(GFS2_I(inode));
525 return ERR_PTR(-EIO);
526 }
527
528
529 /**
530 * dirent_first - Return the first dirent
531 * @dip: the directory
532 * @bh: The buffer
533 * @dent: Pointer to list of dirents
534 *
535 * return first dirent whether bh points to leaf or stuffed dinode
536 *
537 * Returns: IS_LEAF, IS_DINODE, or -errno
538 */
539
540 static int dirent_first(struct gfs2_inode *dip, struct buffer_head *bh,
541 struct gfs2_dirent **dent)
542 {
543 struct gfs2_meta_header *h = (struct gfs2_meta_header *)bh->b_data;
544
545 if (be32_to_cpu(h->mh_type) == GFS2_METATYPE_LF) {
546 if (gfs2_meta_check(GFS2_SB(&dip->i_inode), bh))
547 return -EIO;
548 *dent = (struct gfs2_dirent *)(bh->b_data +
549 sizeof(struct gfs2_leaf));
550 return IS_LEAF;
551 } else {
552 if (gfs2_metatype_check(GFS2_SB(&dip->i_inode), bh, GFS2_METATYPE_DI))
553 return -EIO;
554 *dent = (struct gfs2_dirent *)(bh->b_data +
555 sizeof(struct gfs2_dinode));
556 return IS_DINODE;
557 }
558 }
559
560 static int dirent_check_reclen(struct gfs2_inode *dip,
561 const struct gfs2_dirent *d, const void *end_p)
562 {
563 const void *ptr = d;
564 u16 rec_len = be16_to_cpu(d->de_rec_len);
565
566 if (unlikely(rec_len < sizeof(struct gfs2_dirent)))
567 goto broken;
568 ptr += rec_len;
569 if (ptr < end_p)
570 return rec_len;
571 if (ptr == end_p)
572 return -ENOENT;
573 broken:
574 gfs2_consist_inode(dip);
575 return -EIO;
576 }
577
578 /**
579 * dirent_next - Next dirent
580 * @dip: the directory
581 * @bh: The buffer
582 * @dent: Pointer to list of dirents
583 *
584 * Returns: 0 on success, error code otherwise
585 */
586
587 static int dirent_next(struct gfs2_inode *dip, struct buffer_head *bh,
588 struct gfs2_dirent **dent)
589 {
590 struct gfs2_dirent *cur = *dent, *tmp;
591 char *bh_end = bh->b_data + bh->b_size;
592 int ret;
593
594 ret = dirent_check_reclen(dip, cur, bh_end);
595 if (ret < 0)
596 return ret;
597
598 tmp = (void *)cur + ret;
599 ret = dirent_check_reclen(dip, tmp, bh_end);
600 if (ret == -EIO)
601 return ret;
602
603 /* Only the first dent could ever have de_inum.no_addr == 0 */
604 if (gfs2_dirent_sentinel(tmp)) {
605 gfs2_consist_inode(dip);
606 return -EIO;
607 }
608
609 *dent = tmp;
610 return 0;
611 }
612
613 /**
614 * dirent_del - Delete a dirent
615 * @dip: The GFS2 inode
616 * @bh: The buffer
617 * @prev: The previous dirent
618 * @cur: The current dirent
619 *
620 */
621
622 static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh,
623 struct gfs2_dirent *prev, struct gfs2_dirent *cur)
624 {
625 u16 cur_rec_len, prev_rec_len;
626
627 if (gfs2_dirent_sentinel(cur)) {
628 gfs2_consist_inode(dip);
629 return;
630 }
631
632 gfs2_trans_add_bh(dip->i_gl, bh, 1);
633
634 /* If there is no prev entry, this is the first entry in the block.
635 The de_rec_len is already as big as it needs to be. Just zero
636 out the inode number and return. */
637
638 if (!prev) {
639 cur->de_inum.no_addr = 0;
640 cur->de_inum.no_formal_ino = 0;
641 return;
642 }
643
644 /* Combine this dentry with the previous one. */
645
646 prev_rec_len = be16_to_cpu(prev->de_rec_len);
647 cur_rec_len = be16_to_cpu(cur->de_rec_len);
648
649 if ((char *)prev + prev_rec_len != (char *)cur)
650 gfs2_consist_inode(dip);
651 if ((char *)cur + cur_rec_len > bh->b_data + bh->b_size)
652 gfs2_consist_inode(dip);
653
654 prev_rec_len += cur_rec_len;
655 prev->de_rec_len = cpu_to_be16(prev_rec_len);
656 }
657
658 /*
659 * Takes a dent from which to grab space as an argument. Returns the
660 * newly created dent.
661 */
662 static struct gfs2_dirent *gfs2_init_dirent(struct inode *inode,
663 struct gfs2_dirent *dent,
664 const struct qstr *name,
665 struct buffer_head *bh)
666 {
667 struct gfs2_inode *ip = GFS2_I(inode);
668 struct gfs2_dirent *ndent;
669 unsigned offset = 0, totlen;
670
671 if (!gfs2_dirent_sentinel(dent))
672 offset = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
673 totlen = be16_to_cpu(dent->de_rec_len);
674 BUG_ON(offset + name->len > totlen);
675 gfs2_trans_add_bh(ip->i_gl, bh, 1);
676 ndent = (struct gfs2_dirent *)((char *)dent + offset);
677 dent->de_rec_len = cpu_to_be16(offset);
678 gfs2_qstr2dirent(name, totlen - offset, ndent);
679 return ndent;
680 }
681
682 static struct gfs2_dirent *gfs2_dirent_alloc(struct inode *inode,
683 struct buffer_head *bh,
684 const struct qstr *name)
685 {
686 struct gfs2_dirent *dent;
687 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
688 gfs2_dirent_find_space, name, NULL);
689 if (!dent || IS_ERR(dent))
690 return dent;
691 return gfs2_init_dirent(inode, dent, name, bh);
692 }
693
694 static int get_leaf(struct gfs2_inode *dip, u64 leaf_no,
695 struct buffer_head **bhp)
696 {
697 int error;
698
699 error = gfs2_meta_read(dip->i_gl, leaf_no, DIO_WAIT, bhp);
700 if (!error && gfs2_metatype_check(GFS2_SB(&dip->i_inode), *bhp, GFS2_METATYPE_LF)) {
701 /* printk(KERN_INFO "block num=%llu\n", leaf_no); */
702 error = -EIO;
703 }
704
705 return error;
706 }
707
708 /**
709 * get_leaf_nr - Get a leaf number associated with the index
710 * @dip: The GFS2 inode
711 * @index:
712 * @leaf_out:
713 *
714 * Returns: 0 on success, error code otherwise
715 */
716
717 static int get_leaf_nr(struct gfs2_inode *dip, u32 index,
718 u64 *leaf_out)
719 {
720 __be64 leaf_no;
721 int error;
722
723 error = gfs2_dir_read_data(dip, (char *)&leaf_no,
724 index * sizeof(__be64),
725 sizeof(__be64), 0);
726 if (error != sizeof(u64))
727 return (error < 0) ? error : -EIO;
728
729 *leaf_out = be64_to_cpu(leaf_no);
730
731 return 0;
732 }
733
734 static int get_first_leaf(struct gfs2_inode *dip, u32 index,
735 struct buffer_head **bh_out)
736 {
737 u64 leaf_no;
738 int error;
739
740 error = get_leaf_nr(dip, index, &leaf_no);
741 if (!error)
742 error = get_leaf(dip, leaf_no, bh_out);
743
744 return error;
745 }
746
747 static struct gfs2_dirent *gfs2_dirent_search(struct inode *inode,
748 const struct qstr *name,
749 gfs2_dscan_t scan,
750 struct buffer_head **pbh)
751 {
752 struct buffer_head *bh;
753 struct gfs2_dirent *dent;
754 struct gfs2_inode *ip = GFS2_I(inode);
755 int error;
756
757 if (ip->i_diskflags & GFS2_DIF_EXHASH) {
758 struct gfs2_leaf *leaf;
759 unsigned hsize = 1 << ip->i_depth;
760 unsigned index;
761 u64 ln;
762 if (hsize * sizeof(u64) != ip->i_disksize) {
763 gfs2_consist_inode(ip);
764 return ERR_PTR(-EIO);
765 }
766
767 index = name->hash >> (32 - ip->i_depth);
768 error = get_first_leaf(ip, index, &bh);
769 if (error)
770 return ERR_PTR(error);
771 do {
772 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
773 scan, name, NULL);
774 if (dent)
775 goto got_dent;
776 leaf = (struct gfs2_leaf *)bh->b_data;
777 ln = be64_to_cpu(leaf->lf_next);
778 brelse(bh);
779 if (!ln)
780 break;
781
782 error = get_leaf(ip, ln, &bh);
783 } while(!error);
784
785 return error ? ERR_PTR(error) : NULL;
786 }
787
788
789 error = gfs2_meta_inode_buffer(ip, &bh);
790 if (error)
791 return ERR_PTR(error);
792 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, scan, name, NULL);
793 got_dent:
794 if (unlikely(dent == NULL || IS_ERR(dent))) {
795 brelse(bh);
796 bh = NULL;
797 }
798 *pbh = bh;
799 return dent;
800 }
801
802 static struct gfs2_leaf *new_leaf(struct inode *inode, struct buffer_head **pbh, u16 depth)
803 {
804 struct gfs2_inode *ip = GFS2_I(inode);
805 unsigned int n = 1;
806 u64 bn;
807 int error;
808 struct buffer_head *bh;
809 struct gfs2_leaf *leaf;
810 struct gfs2_dirent *dent;
811 struct qstr name = { .name = "", .len = 0, .hash = 0 };
812
813 error = gfs2_alloc_block(ip, &bn, &n);
814 if (error)
815 return NULL;
816 bh = gfs2_meta_new(ip->i_gl, bn);
817 if (!bh)
818 return NULL;
819
820 gfs2_trans_add_unrevoke(GFS2_SB(inode), bn, 1);
821 gfs2_trans_add_bh(ip->i_gl, bh, 1);
822 gfs2_metatype_set(bh, GFS2_METATYPE_LF, GFS2_FORMAT_LF);
823 leaf = (struct gfs2_leaf *)bh->b_data;
824 leaf->lf_depth = cpu_to_be16(depth);
825 leaf->lf_entries = 0;
826 leaf->lf_dirent_format = cpu_to_be32(GFS2_FORMAT_DE);
827 leaf->lf_next = 0;
828 memset(leaf->lf_reserved, 0, sizeof(leaf->lf_reserved));
829 dent = (struct gfs2_dirent *)(leaf+1);
830 gfs2_qstr2dirent(&name, bh->b_size - sizeof(struct gfs2_leaf), dent);
831 *pbh = bh;
832 return leaf;
833 }
834
835 /**
836 * dir_make_exhash - Convert a stuffed directory into an ExHash directory
837 * @dip: The GFS2 inode
838 *
839 * Returns: 0 on success, error code otherwise
840 */
841
842 static int dir_make_exhash(struct inode *inode)
843 {
844 struct gfs2_inode *dip = GFS2_I(inode);
845 struct gfs2_sbd *sdp = GFS2_SB(inode);
846 struct gfs2_dirent *dent;
847 struct qstr args;
848 struct buffer_head *bh, *dibh;
849 struct gfs2_leaf *leaf;
850 int y;
851 u32 x;
852 __be64 *lp;
853 u64 bn;
854 int error;
855
856 error = gfs2_meta_inode_buffer(dip, &dibh);
857 if (error)
858 return error;
859
860 /* Turn over a new leaf */
861
862 leaf = new_leaf(inode, &bh, 0);
863 if (!leaf)
864 return -ENOSPC;
865 bn = bh->b_blocknr;
866
867 gfs2_assert(sdp, dip->i_entries < (1 << 16));
868 leaf->lf_entries = cpu_to_be16(dip->i_entries);
869
870 /* Copy dirents */
871
872 gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_leaf), dibh,
873 sizeof(struct gfs2_dinode));
874
875 /* Find last entry */
876
877 x = 0;
878 args.len = bh->b_size - sizeof(struct gfs2_dinode) +
879 sizeof(struct gfs2_leaf);
880 args.name = bh->b_data;
881 dent = gfs2_dirent_scan(&dip->i_inode, bh->b_data, bh->b_size,
882 gfs2_dirent_last, &args, NULL);
883 if (!dent) {
884 brelse(bh);
885 brelse(dibh);
886 return -EIO;
887 }
888 if (IS_ERR(dent)) {
889 brelse(bh);
890 brelse(dibh);
891 return PTR_ERR(dent);
892 }
893
894 /* Adjust the last dirent's record length
895 (Remember that dent still points to the last entry.) */
896
897 dent->de_rec_len = cpu_to_be16(be16_to_cpu(dent->de_rec_len) +
898 sizeof(struct gfs2_dinode) -
899 sizeof(struct gfs2_leaf));
900
901 brelse(bh);
902
903 /* We're done with the new leaf block, now setup the new
904 hash table. */
905
906 gfs2_trans_add_bh(dip->i_gl, dibh, 1);
907 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
908
909 lp = (__be64 *)(dibh->b_data + sizeof(struct gfs2_dinode));
910
911 for (x = sdp->sd_hash_ptrs; x--; lp++)
912 *lp = cpu_to_be64(bn);
913
914 dip->i_disksize = sdp->sd_sb.sb_bsize / 2;
915 gfs2_add_inode_blocks(&dip->i_inode, 1);
916 dip->i_diskflags |= GFS2_DIF_EXHASH;
917
918 for (x = sdp->sd_hash_ptrs, y = -1; x; x >>= 1, y++) ;
919 dip->i_depth = y;
920
921 gfs2_dinode_out(dip, dibh->b_data);
922
923 brelse(dibh);
924
925 return 0;
926 }
927
928 /**
929 * dir_split_leaf - Split a leaf block into two
930 * @dip: The GFS2 inode
931 * @index:
932 * @leaf_no:
933 *
934 * Returns: 0 on success, error code on failure
935 */
936
937 static int dir_split_leaf(struct inode *inode, const struct qstr *name)
938 {
939 struct gfs2_inode *dip = GFS2_I(inode);
940 struct buffer_head *nbh, *obh, *dibh;
941 struct gfs2_leaf *nleaf, *oleaf;
942 struct gfs2_dirent *dent = NULL, *prev = NULL, *next = NULL, *new;
943 u32 start, len, half_len, divider;
944 u64 bn, leaf_no;
945 __be64 *lp;
946 u32 index;
947 int x, moved = 0;
948 int error;
949
950 index = name->hash >> (32 - dip->i_depth);
951 error = get_leaf_nr(dip, index, &leaf_no);
952 if (error)
953 return error;
954
955 /* Get the old leaf block */
956 error = get_leaf(dip, leaf_no, &obh);
957 if (error)
958 return error;
959
960 oleaf = (struct gfs2_leaf *)obh->b_data;
961 if (dip->i_depth == be16_to_cpu(oleaf->lf_depth)) {
962 brelse(obh);
963 return 1; /* can't split */
964 }
965
966 gfs2_trans_add_bh(dip->i_gl, obh, 1);
967
968 nleaf = new_leaf(inode, &nbh, be16_to_cpu(oleaf->lf_depth) + 1);
969 if (!nleaf) {
970 brelse(obh);
971 return -ENOSPC;
972 }
973 bn = nbh->b_blocknr;
974
975 /* Compute the start and len of leaf pointers in the hash table. */
976 len = 1 << (dip->i_depth - be16_to_cpu(oleaf->lf_depth));
977 half_len = len >> 1;
978 if (!half_len) {
979 printk(KERN_WARNING "i_depth %u lf_depth %u index %u\n", dip->i_depth, be16_to_cpu(oleaf->lf_depth), index);
980 gfs2_consist_inode(dip);
981 error = -EIO;
982 goto fail_brelse;
983 }
984
985 start = (index & ~(len - 1));
986
987 /* Change the pointers.
988 Don't bother distinguishing stuffed from non-stuffed.
989 This code is complicated enough already. */
990 lp = kmalloc(half_len * sizeof(__be64), GFP_NOFS | __GFP_NOFAIL);
991 /* Change the pointers */
992 for (x = 0; x < half_len; x++)
993 lp[x] = cpu_to_be64(bn);
994
995 error = gfs2_dir_write_data(dip, (char *)lp, start * sizeof(u64),
996 half_len * sizeof(u64));
997 if (error != half_len * sizeof(u64)) {
998 if (error >= 0)
999 error = -EIO;
1000 goto fail_lpfree;
1001 }
1002
1003 kfree(lp);
1004
1005 /* Compute the divider */
1006 divider = (start + half_len) << (32 - dip->i_depth);
1007
1008 /* Copy the entries */
1009 dirent_first(dip, obh, &dent);
1010
1011 do {
1012 next = dent;
1013 if (dirent_next(dip, obh, &next))
1014 next = NULL;
1015
1016 if (!gfs2_dirent_sentinel(dent) &&
1017 be32_to_cpu(dent->de_hash) < divider) {
1018 struct qstr str;
1019 str.name = (char*)(dent+1);
1020 str.len = be16_to_cpu(dent->de_name_len);
1021 str.hash = be32_to_cpu(dent->de_hash);
1022 new = gfs2_dirent_alloc(inode, nbh, &str);
1023 if (IS_ERR(new)) {
1024 error = PTR_ERR(new);
1025 break;
1026 }
1027
1028 new->de_inum = dent->de_inum; /* No endian worries */
1029 new->de_type = dent->de_type; /* No endian worries */
1030 be16_add_cpu(&nleaf->lf_entries, 1);
1031
1032 dirent_del(dip, obh, prev, dent);
1033
1034 if (!oleaf->lf_entries)
1035 gfs2_consist_inode(dip);
1036 be16_add_cpu(&oleaf->lf_entries, -1);
1037
1038 if (!prev)
1039 prev = dent;
1040
1041 moved = 1;
1042 } else {
1043 prev = dent;
1044 }
1045 dent = next;
1046 } while (dent);
1047
1048 oleaf->lf_depth = nleaf->lf_depth;
1049
1050 error = gfs2_meta_inode_buffer(dip, &dibh);
1051 if (!gfs2_assert_withdraw(GFS2_SB(&dip->i_inode), !error)) {
1052 gfs2_trans_add_bh(dip->i_gl, dibh, 1);
1053 gfs2_add_inode_blocks(&dip->i_inode, 1);
1054 gfs2_dinode_out(dip, dibh->b_data);
1055 brelse(dibh);
1056 }
1057
1058 brelse(obh);
1059 brelse(nbh);
1060
1061 return error;
1062
1063 fail_lpfree:
1064 kfree(lp);
1065
1066 fail_brelse:
1067 brelse(obh);
1068 brelse(nbh);
1069 return error;
1070 }
1071
1072 /**
1073 * dir_double_exhash - Double size of ExHash table
1074 * @dip: The GFS2 dinode
1075 *
1076 * Returns: 0 on success, error code on failure
1077 */
1078
1079 static int dir_double_exhash(struct gfs2_inode *dip)
1080 {
1081 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1082 struct buffer_head *dibh;
1083 u32 hsize;
1084 u64 *buf;
1085 u64 *from, *to;
1086 u64 block;
1087 int x;
1088 int error = 0;
1089
1090 hsize = 1 << dip->i_depth;
1091 if (hsize * sizeof(u64) != dip->i_disksize) {
1092 gfs2_consist_inode(dip);
1093 return -EIO;
1094 }
1095
1096 /* Allocate both the "from" and "to" buffers in one big chunk */
1097
1098 buf = kcalloc(3, sdp->sd_hash_bsize, GFP_NOFS | __GFP_NOFAIL);
1099
1100 for (block = dip->i_disksize >> sdp->sd_hash_bsize_shift; block--;) {
1101 error = gfs2_dir_read_data(dip, (char *)buf,
1102 block * sdp->sd_hash_bsize,
1103 sdp->sd_hash_bsize, 1);
1104 if (error != sdp->sd_hash_bsize) {
1105 if (error >= 0)
1106 error = -EIO;
1107 goto fail;
1108 }
1109
1110 from = buf;
1111 to = (u64 *)((char *)buf + sdp->sd_hash_bsize);
1112
1113 for (x = sdp->sd_hash_ptrs; x--; from++) {
1114 *to++ = *from; /* No endianess worries */
1115 *to++ = *from;
1116 }
1117
1118 error = gfs2_dir_write_data(dip,
1119 (char *)buf + sdp->sd_hash_bsize,
1120 block * sdp->sd_sb.sb_bsize,
1121 sdp->sd_sb.sb_bsize);
1122 if (error != sdp->sd_sb.sb_bsize) {
1123 if (error >= 0)
1124 error = -EIO;
1125 goto fail;
1126 }
1127 }
1128
1129 kfree(buf);
1130
1131 error = gfs2_meta_inode_buffer(dip, &dibh);
1132 if (!gfs2_assert_withdraw(sdp, !error)) {
1133 dip->i_depth++;
1134 gfs2_dinode_out(dip, dibh->b_data);
1135 brelse(dibh);
1136 }
1137
1138 return error;
1139
1140 fail:
1141 kfree(buf);
1142 return error;
1143 }
1144
1145 /**
1146 * compare_dents - compare directory entries by hash value
1147 * @a: first dent
1148 * @b: second dent
1149 *
1150 * When comparing the hash entries of @a to @b:
1151 * gt: returns 1
1152 * lt: returns -1
1153 * eq: returns 0
1154 */
1155
1156 static int compare_dents(const void *a, const void *b)
1157 {
1158 const struct gfs2_dirent *dent_a, *dent_b;
1159 u32 hash_a, hash_b;
1160 int ret = 0;
1161
1162 dent_a = *(const struct gfs2_dirent **)a;
1163 hash_a = be32_to_cpu(dent_a->de_hash);
1164
1165 dent_b = *(const struct gfs2_dirent **)b;
1166 hash_b = be32_to_cpu(dent_b->de_hash);
1167
1168 if (hash_a > hash_b)
1169 ret = 1;
1170 else if (hash_a < hash_b)
1171 ret = -1;
1172 else {
1173 unsigned int len_a = be16_to_cpu(dent_a->de_name_len);
1174 unsigned int len_b = be16_to_cpu(dent_b->de_name_len);
1175
1176 if (len_a > len_b)
1177 ret = 1;
1178 else if (len_a < len_b)
1179 ret = -1;
1180 else
1181 ret = memcmp(dent_a + 1, dent_b + 1, len_a);
1182 }
1183
1184 return ret;
1185 }
1186
1187 /**
1188 * do_filldir_main - read out directory entries
1189 * @dip: The GFS2 inode
1190 * @offset: The offset in the file to read from
1191 * @opaque: opaque data to pass to filldir
1192 * @filldir: The function to pass entries to
1193 * @darr: an array of struct gfs2_dirent pointers to read
1194 * @entries: the number of entries in darr
1195 * @copied: pointer to int that's non-zero if a entry has been copied out
1196 *
1197 * Jump through some hoops to make sure that if there are hash collsions,
1198 * they are read out at the beginning of a buffer. We want to minimize
1199 * the possibility that they will fall into different readdir buffers or
1200 * that someone will want to seek to that location.
1201 *
1202 * Returns: errno, >0 on exception from filldir
1203 */
1204
1205 static int do_filldir_main(struct gfs2_inode *dip, u64 *offset,
1206 void *opaque, filldir_t filldir,
1207 const struct gfs2_dirent **darr, u32 entries,
1208 int *copied)
1209 {
1210 const struct gfs2_dirent *dent, *dent_next;
1211 u64 off, off_next;
1212 unsigned int x, y;
1213 int run = 0;
1214 int error = 0;
1215
1216 sort(darr, entries, sizeof(struct gfs2_dirent *), compare_dents, NULL);
1217
1218 dent_next = darr[0];
1219 off_next = be32_to_cpu(dent_next->de_hash);
1220 off_next = gfs2_disk_hash2offset(off_next);
1221
1222 for (x = 0, y = 1; x < entries; x++, y++) {
1223 dent = dent_next;
1224 off = off_next;
1225
1226 if (y < entries) {
1227 dent_next = darr[y];
1228 off_next = be32_to_cpu(dent_next->de_hash);
1229 off_next = gfs2_disk_hash2offset(off_next);
1230
1231 if (off < *offset)
1232 continue;
1233 *offset = off;
1234
1235 if (off_next == off) {
1236 if (*copied && !run)
1237 return 1;
1238 run = 1;
1239 } else
1240 run = 0;
1241 } else {
1242 if (off < *offset)
1243 continue;
1244 *offset = off;
1245 }
1246
1247 error = filldir(opaque, (const char *)(dent + 1),
1248 be16_to_cpu(dent->de_name_len),
1249 off, be64_to_cpu(dent->de_inum.no_addr),
1250 be16_to_cpu(dent->de_type));
1251 if (error)
1252 return 1;
1253
1254 *copied = 1;
1255 }
1256
1257 /* Increment the *offset by one, so the next time we come into the
1258 do_filldir fxn, we get the next entry instead of the last one in the
1259 current leaf */
1260
1261 (*offset)++;
1262
1263 return 0;
1264 }
1265
1266 static int gfs2_dir_read_leaf(struct inode *inode, u64 *offset, void *opaque,
1267 filldir_t filldir, int *copied, unsigned *depth,
1268 u64 leaf_no)
1269 {
1270 struct gfs2_inode *ip = GFS2_I(inode);
1271 struct gfs2_sbd *sdp = GFS2_SB(inode);
1272 struct buffer_head *bh;
1273 struct gfs2_leaf *lf;
1274 unsigned entries = 0, entries2 = 0;
1275 unsigned leaves = 0;
1276 const struct gfs2_dirent **darr, *dent;
1277 struct dirent_gather g;
1278 struct buffer_head **larr;
1279 int leaf = 0;
1280 int error, i;
1281 u64 lfn = leaf_no;
1282
1283 do {
1284 error = get_leaf(ip, lfn, &bh);
1285 if (error)
1286 goto out;
1287 lf = (struct gfs2_leaf *)bh->b_data;
1288 if (leaves == 0)
1289 *depth = be16_to_cpu(lf->lf_depth);
1290 entries += be16_to_cpu(lf->lf_entries);
1291 leaves++;
1292 lfn = be64_to_cpu(lf->lf_next);
1293 brelse(bh);
1294 } while(lfn);
1295
1296 if (!entries)
1297 return 0;
1298
1299 error = -ENOMEM;
1300 /*
1301 * The extra 99 entries are not normally used, but are a buffer
1302 * zone in case the number of entries in the leaf is corrupt.
1303 * 99 is the maximum number of entries that can fit in a single
1304 * leaf block.
1305 */
1306 larr = vmalloc((leaves + entries + 99) * sizeof(void *));
1307 if (!larr)
1308 goto out;
1309 darr = (const struct gfs2_dirent **)(larr + leaves);
1310 g.pdent = darr;
1311 g.offset = 0;
1312 lfn = leaf_no;
1313
1314 do {
1315 error = get_leaf(ip, lfn, &bh);
1316 if (error)
1317 goto out_kfree;
1318 lf = (struct gfs2_leaf *)bh->b_data;
1319 lfn = be64_to_cpu(lf->lf_next);
1320 if (lf->lf_entries) {
1321 entries2 += be16_to_cpu(lf->lf_entries);
1322 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
1323 gfs2_dirent_gather, NULL, &g);
1324 error = PTR_ERR(dent);
1325 if (IS_ERR(dent))
1326 goto out_kfree;
1327 if (entries2 != g.offset) {
1328 fs_warn(sdp, "Number of entries corrupt in dir "
1329 "leaf %llu, entries2 (%u) != "
1330 "g.offset (%u)\n",
1331 (unsigned long long)bh->b_blocknr,
1332 entries2, g.offset);
1333
1334 error = -EIO;
1335 goto out_kfree;
1336 }
1337 error = 0;
1338 larr[leaf++] = bh;
1339 } else {
1340 brelse(bh);
1341 }
1342 } while(lfn);
1343
1344 BUG_ON(entries2 != entries);
1345 error = do_filldir_main(ip, offset, opaque, filldir, darr,
1346 entries, copied);
1347 out_kfree:
1348 for(i = 0; i < leaf; i++)
1349 brelse(larr[i]);
1350 vfree(larr);
1351 out:
1352 return error;
1353 }
1354
1355 /**
1356 * dir_e_read - Reads the entries from a directory into a filldir buffer
1357 * @dip: dinode pointer
1358 * @offset: the hash of the last entry read shifted to the right once
1359 * @opaque: buffer for the filldir function to fill
1360 * @filldir: points to the filldir function to use
1361 *
1362 * Returns: errno
1363 */
1364
1365 static int dir_e_read(struct inode *inode, u64 *offset, void *opaque,
1366 filldir_t filldir)
1367 {
1368 struct gfs2_inode *dip = GFS2_I(inode);
1369 struct gfs2_sbd *sdp = GFS2_SB(inode);
1370 u32 hsize, len = 0;
1371 u32 ht_offset, lp_offset, ht_offset_cur = -1;
1372 u32 hash, index;
1373 __be64 *lp;
1374 int copied = 0;
1375 int error = 0;
1376 unsigned depth = 0;
1377
1378 hsize = 1 << dip->i_depth;
1379 if (hsize * sizeof(u64) != dip->i_disksize) {
1380 gfs2_consist_inode(dip);
1381 return -EIO;
1382 }
1383
1384 hash = gfs2_dir_offset2hash(*offset);
1385 index = hash >> (32 - dip->i_depth);
1386
1387 lp = kmalloc(sdp->sd_hash_bsize, GFP_NOFS);
1388 if (!lp)
1389 return -ENOMEM;
1390
1391 while (index < hsize) {
1392 lp_offset = index & (sdp->sd_hash_ptrs - 1);
1393 ht_offset = index - lp_offset;
1394
1395 if (ht_offset_cur != ht_offset) {
1396 error = gfs2_dir_read_data(dip, (char *)lp,
1397 ht_offset * sizeof(__be64),
1398 sdp->sd_hash_bsize, 1);
1399 if (error != sdp->sd_hash_bsize) {
1400 if (error >= 0)
1401 error = -EIO;
1402 goto out;
1403 }
1404 ht_offset_cur = ht_offset;
1405 }
1406
1407 error = gfs2_dir_read_leaf(inode, offset, opaque, filldir,
1408 &copied, &depth,
1409 be64_to_cpu(lp[lp_offset]));
1410 if (error)
1411 break;
1412
1413 len = 1 << (dip->i_depth - depth);
1414 index = (index & ~(len - 1)) + len;
1415 }
1416
1417 out:
1418 kfree(lp);
1419 if (error > 0)
1420 error = 0;
1421 return error;
1422 }
1423
1424 int gfs2_dir_read(struct inode *inode, u64 *offset, void *opaque,
1425 filldir_t filldir)
1426 {
1427 struct gfs2_inode *dip = GFS2_I(inode);
1428 struct gfs2_sbd *sdp = GFS2_SB(inode);
1429 struct dirent_gather g;
1430 const struct gfs2_dirent **darr, *dent;
1431 struct buffer_head *dibh;
1432 int copied = 0;
1433 int error;
1434
1435 if (!dip->i_entries)
1436 return 0;
1437
1438 if (dip->i_diskflags & GFS2_DIF_EXHASH)
1439 return dir_e_read(inode, offset, opaque, filldir);
1440
1441 if (!gfs2_is_stuffed(dip)) {
1442 gfs2_consist_inode(dip);
1443 return -EIO;
1444 }
1445
1446 error = gfs2_meta_inode_buffer(dip, &dibh);
1447 if (error)
1448 return error;
1449
1450 error = -ENOMEM;
1451 /* 96 is max number of dirents which can be stuffed into an inode */
1452 darr = kmalloc(96 * sizeof(struct gfs2_dirent *), GFP_NOFS);
1453 if (darr) {
1454 g.pdent = darr;
1455 g.offset = 0;
1456 dent = gfs2_dirent_scan(inode, dibh->b_data, dibh->b_size,
1457 gfs2_dirent_gather, NULL, &g);
1458 if (IS_ERR(dent)) {
1459 error = PTR_ERR(dent);
1460 goto out;
1461 }
1462 if (dip->i_entries != g.offset) {
1463 fs_warn(sdp, "Number of entries corrupt in dir %llu, "
1464 "ip->i_entries (%u) != g.offset (%u)\n",
1465 (unsigned long long)dip->i_no_addr,
1466 dip->i_entries,
1467 g.offset);
1468 error = -EIO;
1469 goto out;
1470 }
1471 error = do_filldir_main(dip, offset, opaque, filldir, darr,
1472 dip->i_entries, &copied);
1473 out:
1474 kfree(darr);
1475 }
1476
1477 if (error > 0)
1478 error = 0;
1479
1480 brelse(dibh);
1481
1482 return error;
1483 }
1484
1485 /**
1486 * gfs2_dir_search - Search a directory
1487 * @dip: The GFS2 inode
1488 * @filename:
1489 * @inode:
1490 *
1491 * This routine searches a directory for a file or another directory.
1492 * Assumes a glock is held on dip.
1493 *
1494 * Returns: errno
1495 */
1496
1497 struct inode *gfs2_dir_search(struct inode *dir, const struct qstr *name)
1498 {
1499 struct buffer_head *bh;
1500 struct gfs2_dirent *dent;
1501 struct inode *inode;
1502
1503 dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
1504 if (dent) {
1505 if (IS_ERR(dent))
1506 return ERR_CAST(dent);
1507 inode = gfs2_inode_lookup(dir->i_sb,
1508 be16_to_cpu(dent->de_type),
1509 be64_to_cpu(dent->de_inum.no_addr),
1510 be64_to_cpu(dent->de_inum.no_formal_ino), 0);
1511 brelse(bh);
1512 return inode;
1513 }
1514 return ERR_PTR(-ENOENT);
1515 }
1516
1517 int gfs2_dir_check(struct inode *dir, const struct qstr *name,
1518 const struct gfs2_inode *ip)
1519 {
1520 struct buffer_head *bh;
1521 struct gfs2_dirent *dent;
1522 int ret = -ENOENT;
1523
1524 dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
1525 if (dent) {
1526 if (IS_ERR(dent))
1527 return PTR_ERR(dent);
1528 if (ip) {
1529 if (be64_to_cpu(dent->de_inum.no_addr) != ip->i_no_addr)
1530 goto out;
1531 if (be64_to_cpu(dent->de_inum.no_formal_ino) !=
1532 ip->i_no_formal_ino)
1533 goto out;
1534 if (unlikely(IF2DT(ip->i_inode.i_mode) !=
1535 be16_to_cpu(dent->de_type))) {
1536 gfs2_consist_inode(GFS2_I(dir));
1537 ret = -EIO;
1538 goto out;
1539 }
1540 }
1541 ret = 0;
1542 out:
1543 brelse(bh);
1544 }
1545 return ret;
1546 }
1547
1548 static int dir_new_leaf(struct inode *inode, const struct qstr *name)
1549 {
1550 struct buffer_head *bh, *obh;
1551 struct gfs2_inode *ip = GFS2_I(inode);
1552 struct gfs2_leaf *leaf, *oleaf;
1553 int error;
1554 u32 index;
1555 u64 bn;
1556
1557 index = name->hash >> (32 - ip->i_depth);
1558 error = get_first_leaf(ip, index, &obh);
1559 if (error)
1560 return error;
1561 do {
1562 oleaf = (struct gfs2_leaf *)obh->b_data;
1563 bn = be64_to_cpu(oleaf->lf_next);
1564 if (!bn)
1565 break;
1566 brelse(obh);
1567 error = get_leaf(ip, bn, &obh);
1568 if (error)
1569 return error;
1570 } while(1);
1571
1572 gfs2_trans_add_bh(ip->i_gl, obh, 1);
1573
1574 leaf = new_leaf(inode, &bh, be16_to_cpu(oleaf->lf_depth));
1575 if (!leaf) {
1576 brelse(obh);
1577 return -ENOSPC;
1578 }
1579 oleaf->lf_next = cpu_to_be64(bh->b_blocknr);
1580 brelse(bh);
1581 brelse(obh);
1582
1583 error = gfs2_meta_inode_buffer(ip, &bh);
1584 if (error)
1585 return error;
1586 gfs2_trans_add_bh(ip->i_gl, bh, 1);
1587 gfs2_add_inode_blocks(&ip->i_inode, 1);
1588 gfs2_dinode_out(ip, bh->b_data);
1589 brelse(bh);
1590 return 0;
1591 }
1592
1593 /**
1594 * gfs2_dir_add - Add new filename into directory
1595 * @dip: The GFS2 inode
1596 * @filename: The new name
1597 * @inode: The inode number of the entry
1598 * @type: The type of the entry
1599 *
1600 * Returns: 0 on success, error code on failure
1601 */
1602
1603 int gfs2_dir_add(struct inode *inode, const struct qstr *name,
1604 const struct gfs2_inode *nip, unsigned type)
1605 {
1606 struct gfs2_inode *ip = GFS2_I(inode);
1607 struct buffer_head *bh;
1608 struct gfs2_dirent *dent;
1609 struct gfs2_leaf *leaf;
1610 int error;
1611
1612 while(1) {
1613 dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space,
1614 &bh);
1615 if (dent) {
1616 if (IS_ERR(dent))
1617 return PTR_ERR(dent);
1618 dent = gfs2_init_dirent(inode, dent, name, bh);
1619 gfs2_inum_out(nip, dent);
1620 dent->de_type = cpu_to_be16(type);
1621 if (ip->i_diskflags & GFS2_DIF_EXHASH) {
1622 leaf = (struct gfs2_leaf *)bh->b_data;
1623 be16_add_cpu(&leaf->lf_entries, 1);
1624 }
1625 brelse(bh);
1626 error = gfs2_meta_inode_buffer(ip, &bh);
1627 if (error)
1628 break;
1629 gfs2_trans_add_bh(ip->i_gl, bh, 1);
1630 ip->i_entries++;
1631 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
1632 gfs2_dinode_out(ip, bh->b_data);
1633 brelse(bh);
1634 error = 0;
1635 break;
1636 }
1637 if (!(ip->i_diskflags & GFS2_DIF_EXHASH)) {
1638 error = dir_make_exhash(inode);
1639 if (error)
1640 break;
1641 continue;
1642 }
1643 error = dir_split_leaf(inode, name);
1644 if (error == 0)
1645 continue;
1646 if (error < 0)
1647 break;
1648 if (ip->i_depth < GFS2_DIR_MAX_DEPTH) {
1649 error = dir_double_exhash(ip);
1650 if (error)
1651 break;
1652 error = dir_split_leaf(inode, name);
1653 if (error < 0)
1654 break;
1655 if (error == 0)
1656 continue;
1657 }
1658 error = dir_new_leaf(inode, name);
1659 if (!error)
1660 continue;
1661 error = -ENOSPC;
1662 break;
1663 }
1664 return error;
1665 }
1666
1667
1668 /**
1669 * gfs2_dir_del - Delete a directory entry
1670 * @dip: The GFS2 inode
1671 * @filename: The filename
1672 *
1673 * Returns: 0 on success, error code on failure
1674 */
1675
1676 int gfs2_dir_del(struct gfs2_inode *dip, const struct qstr *name)
1677 {
1678 struct gfs2_dirent *dent, *prev = NULL;
1679 struct buffer_head *bh;
1680 int error;
1681
1682 /* Returns _either_ the entry (if its first in block) or the
1683 previous entry otherwise */
1684 dent = gfs2_dirent_search(&dip->i_inode, name, gfs2_dirent_prev, &bh);
1685 if (!dent) {
1686 gfs2_consist_inode(dip);
1687 return -EIO;
1688 }
1689 if (IS_ERR(dent)) {
1690 gfs2_consist_inode(dip);
1691 return PTR_ERR(dent);
1692 }
1693 /* If not first in block, adjust pointers accordingly */
1694 if (gfs2_dirent_find(dent, name, NULL) == 0) {
1695 prev = dent;
1696 dent = (struct gfs2_dirent *)((char *)dent + be16_to_cpu(prev->de_rec_len));
1697 }
1698
1699 dirent_del(dip, bh, prev, dent);
1700 if (dip->i_diskflags & GFS2_DIF_EXHASH) {
1701 struct gfs2_leaf *leaf = (struct gfs2_leaf *)bh->b_data;
1702 u16 entries = be16_to_cpu(leaf->lf_entries);
1703 if (!entries)
1704 gfs2_consist_inode(dip);
1705 leaf->lf_entries = cpu_to_be16(--entries);
1706 }
1707 brelse(bh);
1708
1709 error = gfs2_meta_inode_buffer(dip, &bh);
1710 if (error)
1711 return error;
1712
1713 if (!dip->i_entries)
1714 gfs2_consist_inode(dip);
1715 gfs2_trans_add_bh(dip->i_gl, bh, 1);
1716 dip->i_entries--;
1717 dip->i_inode.i_mtime = dip->i_inode.i_ctime = CURRENT_TIME;
1718 gfs2_dinode_out(dip, bh->b_data);
1719 brelse(bh);
1720 mark_inode_dirty(&dip->i_inode);
1721
1722 return error;
1723 }
1724
1725 /**
1726 * gfs2_dir_mvino - Change inode number of directory entry
1727 * @dip: The GFS2 inode
1728 * @filename:
1729 * @new_inode:
1730 *
1731 * This routine changes the inode number of a directory entry. It's used
1732 * by rename to change ".." when a directory is moved.
1733 * Assumes a glock is held on dvp.
1734 *
1735 * Returns: errno
1736 */
1737
1738 int gfs2_dir_mvino(struct gfs2_inode *dip, const struct qstr *filename,
1739 const struct gfs2_inode *nip, unsigned int new_type)
1740 {
1741 struct buffer_head *bh;
1742 struct gfs2_dirent *dent;
1743 int error;
1744
1745 dent = gfs2_dirent_search(&dip->i_inode, filename, gfs2_dirent_find, &bh);
1746 if (!dent) {
1747 gfs2_consist_inode(dip);
1748 return -EIO;
1749 }
1750 if (IS_ERR(dent))
1751 return PTR_ERR(dent);
1752
1753 gfs2_trans_add_bh(dip->i_gl, bh, 1);
1754 gfs2_inum_out(nip, dent);
1755 dent->de_type = cpu_to_be16(new_type);
1756
1757 if (dip->i_diskflags & GFS2_DIF_EXHASH) {
1758 brelse(bh);
1759 error = gfs2_meta_inode_buffer(dip, &bh);
1760 if (error)
1761 return error;
1762 gfs2_trans_add_bh(dip->i_gl, bh, 1);
1763 }
1764
1765 dip->i_inode.i_mtime = dip->i_inode.i_ctime = CURRENT_TIME;
1766 gfs2_dinode_out(dip, bh->b_data);
1767 brelse(bh);
1768 return 0;
1769 }
1770
1771 /**
1772 * foreach_leaf - call a function for each leaf in a directory
1773 * @dip: the directory
1774 * @lc: the function to call for each each
1775 * @data: private data to pass to it
1776 *
1777 * Returns: errno
1778 */
1779
1780 static int foreach_leaf(struct gfs2_inode *dip, leaf_call_t lc, void *data)
1781 {
1782 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1783 struct buffer_head *bh;
1784 struct gfs2_leaf *leaf;
1785 u32 hsize, len;
1786 u32 ht_offset, lp_offset, ht_offset_cur = -1;
1787 u32 index = 0;
1788 __be64 *lp;
1789 u64 leaf_no;
1790 int error = 0;
1791
1792 hsize = 1 << dip->i_depth;
1793 if (hsize * sizeof(u64) != dip->i_disksize) {
1794 gfs2_consist_inode(dip);
1795 return -EIO;
1796 }
1797
1798 lp = kmalloc(sdp->sd_hash_bsize, GFP_NOFS);
1799 if (!lp)
1800 return -ENOMEM;
1801
1802 while (index < hsize) {
1803 lp_offset = index & (sdp->sd_hash_ptrs - 1);
1804 ht_offset = index - lp_offset;
1805
1806 if (ht_offset_cur != ht_offset) {
1807 error = gfs2_dir_read_data(dip, (char *)lp,
1808 ht_offset * sizeof(__be64),
1809 sdp->sd_hash_bsize, 1);
1810 if (error != sdp->sd_hash_bsize) {
1811 if (error >= 0)
1812 error = -EIO;
1813 goto out;
1814 }
1815 ht_offset_cur = ht_offset;
1816 }
1817
1818 leaf_no = be64_to_cpu(lp[lp_offset]);
1819 if (leaf_no) {
1820 error = get_leaf(dip, leaf_no, &bh);
1821 if (error)
1822 goto out;
1823 leaf = (struct gfs2_leaf *)bh->b_data;
1824 len = 1 << (dip->i_depth - be16_to_cpu(leaf->lf_depth));
1825 brelse(bh);
1826
1827 error = lc(dip, index, len, leaf_no, data);
1828 if (error)
1829 goto out;
1830
1831 index = (index & ~(len - 1)) + len;
1832 } else
1833 index++;
1834 }
1835
1836 if (index != hsize) {
1837 gfs2_consist_inode(dip);
1838 error = -EIO;
1839 }
1840
1841 out:
1842 kfree(lp);
1843
1844 return error;
1845 }
1846
1847 /**
1848 * leaf_dealloc - Deallocate a directory leaf
1849 * @dip: the directory
1850 * @index: the hash table offset in the directory
1851 * @len: the number of pointers to this leaf
1852 * @leaf_no: the leaf number
1853 * @data: not used
1854 *
1855 * Returns: errno
1856 */
1857
1858 static int leaf_dealloc(struct gfs2_inode *dip, u32 index, u32 len,
1859 u64 leaf_no, void *data)
1860 {
1861 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1862 struct gfs2_leaf *tmp_leaf;
1863 struct gfs2_rgrp_list rlist;
1864 struct buffer_head *bh, *dibh;
1865 u64 blk, nblk;
1866 unsigned int rg_blocks = 0, l_blocks = 0;
1867 char *ht;
1868 unsigned int x, size = len * sizeof(u64);
1869 int error;
1870
1871 memset(&rlist, 0, sizeof(struct gfs2_rgrp_list));
1872
1873 ht = kzalloc(size, GFP_NOFS);
1874 if (!ht)
1875 return -ENOMEM;
1876
1877 if (!gfs2_alloc_get(dip)) {
1878 error = -ENOMEM;
1879 goto out;
1880 }
1881
1882 error = gfs2_quota_hold(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
1883 if (error)
1884 goto out_put;
1885
1886 error = gfs2_rindex_hold(sdp, &dip->i_alloc->al_ri_gh);
1887 if (error)
1888 goto out_qs;
1889
1890 /* Count the number of leaves */
1891
1892 for (blk = leaf_no; blk; blk = nblk) {
1893 error = get_leaf(dip, blk, &bh);
1894 if (error)
1895 goto out_rlist;
1896 tmp_leaf = (struct gfs2_leaf *)bh->b_data;
1897 nblk = be64_to_cpu(tmp_leaf->lf_next);
1898 brelse(bh);
1899
1900 gfs2_rlist_add(sdp, &rlist, blk);
1901 l_blocks++;
1902 }
1903
1904 gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE);
1905
1906 for (x = 0; x < rlist.rl_rgrps; x++) {
1907 struct gfs2_rgrpd *rgd;
1908 rgd = rlist.rl_ghs[x].gh_gl->gl_object;
1909 rg_blocks += rgd->rd_length;
1910 }
1911
1912 error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs);
1913 if (error)
1914 goto out_rlist;
1915
1916 error = gfs2_trans_begin(sdp,
1917 rg_blocks + (DIV_ROUND_UP(size, sdp->sd_jbsize) + 1) +
1918 RES_DINODE + RES_STATFS + RES_QUOTA, l_blocks);
1919 if (error)
1920 goto out_rg_gunlock;
1921
1922 for (blk = leaf_no; blk; blk = nblk) {
1923 error = get_leaf(dip, blk, &bh);
1924 if (error)
1925 goto out_end_trans;
1926 tmp_leaf = (struct gfs2_leaf *)bh->b_data;
1927 nblk = be64_to_cpu(tmp_leaf->lf_next);
1928 brelse(bh);
1929
1930 gfs2_free_meta(dip, blk, 1);
1931 gfs2_add_inode_blocks(&dip->i_inode, -1);
1932 }
1933
1934 error = gfs2_dir_write_data(dip, ht, index * sizeof(u64), size);
1935 if (error != size) {
1936 if (error >= 0)
1937 error = -EIO;
1938 goto out_end_trans;
1939 }
1940
1941 error = gfs2_meta_inode_buffer(dip, &dibh);
1942 if (error)
1943 goto out_end_trans;
1944
1945 gfs2_trans_add_bh(dip->i_gl, dibh, 1);
1946 gfs2_dinode_out(dip, dibh->b_data);
1947 brelse(dibh);
1948
1949 out_end_trans:
1950 gfs2_trans_end(sdp);
1951 out_rg_gunlock:
1952 gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
1953 out_rlist:
1954 gfs2_rlist_free(&rlist);
1955 gfs2_glock_dq_uninit(&dip->i_alloc->al_ri_gh);
1956 out_qs:
1957 gfs2_quota_unhold(dip);
1958 out_put:
1959 gfs2_alloc_put(dip);
1960 out:
1961 kfree(ht);
1962 return error;
1963 }
1964
1965 /**
1966 * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory
1967 * @dip: the directory
1968 *
1969 * Dealloc all on-disk directory leaves to FREEMETA state
1970 * Change on-disk inode type to "regular file"
1971 *
1972 * Returns: errno
1973 */
1974
1975 int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip)
1976 {
1977 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1978 struct buffer_head *bh;
1979 int error;
1980
1981 /* Dealloc on-disk leaves to FREEMETA state */
1982 error = foreach_leaf(dip, leaf_dealloc, NULL);
1983 if (error)
1984 return error;
1985
1986 /* Make this a regular file in case we crash.
1987 (We don't want to free these blocks a second time.) */
1988
1989 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1990 if (error)
1991 return error;
1992
1993 error = gfs2_meta_inode_buffer(dip, &bh);
1994 if (!error) {
1995 gfs2_trans_add_bh(dip->i_gl, bh, 1);
1996 ((struct gfs2_dinode *)bh->b_data)->di_mode =
1997 cpu_to_be32(S_IFREG);
1998 brelse(bh);
1999 }
2000
2001 gfs2_trans_end(sdp);
2002
2003 return error;
2004 }
2005
2006 /**
2007 * gfs2_diradd_alloc_required - find if adding entry will require an allocation
2008 * @ip: the file being written to
2009 * @filname: the filename that's going to be added
2010 *
2011 * Returns: 1 if alloc required, 0 if not, -ve on error
2012 */
2013
2014 int gfs2_diradd_alloc_required(struct inode *inode, const struct qstr *name)
2015 {
2016 struct gfs2_dirent *dent;
2017 struct buffer_head *bh;
2018
2019 dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, &bh);
2020 if (!dent) {
2021 return 1;
2022 }
2023 if (IS_ERR(dent))
2024 return PTR_ERR(dent);
2025 brelse(bh);
2026 return 0;
2027 }
2028
2029
|
This page was automatically generated by the
LXR engine.
|