[GFS2] Fix a page lock / glock deadlock
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / fs / gfs2 / dir.c
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_di.di_flags & 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 #include <linux/lm_interface.h>
64
65 #include "gfs2.h"
66 #include "incore.h"
67 #include "dir.h"
68 #include "glock.h"
69 #include "inode.h"
70 #include "meta_io.h"
71 #include "quota.h"
72 #include "rgrp.h"
73 #include "trans.h"
74 #include "bmap.h"
75 #include "util.h"
76
77 #define IS_LEAF 1 /* Hashed (leaf) directory */
78 #define IS_DINODE 2 /* Linear (stuffed dinode block) directory */
79
80 #define gfs2_disk_hash2offset(h) (((u64)(h)) >> 1)
81 #define gfs2_dir_offset2hash(p) ((u32)(((u64)(p)) << 1))
82
83 typedef int (*leaf_call_t) (struct gfs2_inode *dip, u32 index, u32 len,
84 u64 leaf_no, void *data);
85 typedef int (*gfs2_dscan_t)(const struct gfs2_dirent *dent,
86 const struct qstr *name, void *opaque);
87
88
89 int gfs2_dir_get_new_buffer(struct gfs2_inode *ip, u64 block,
90 struct buffer_head **bhp)
91 {
92 struct buffer_head *bh;
93
94 bh = gfs2_meta_new(ip->i_gl, block);
95 gfs2_trans_add_bh(ip->i_gl, bh, 1);
96 gfs2_metatype_set(bh, GFS2_METATYPE_JD, GFS2_FORMAT_JD);
97 gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
98 *bhp = bh;
99 return 0;
100 }
101
102 static int gfs2_dir_get_existing_buffer(struct gfs2_inode *ip, u64 block,
103 struct buffer_head **bhp)
104 {
105 struct buffer_head *bh;
106 int error;
107
108 error = gfs2_meta_read(ip->i_gl, block, DIO_WAIT, &bh);
109 if (error)
110 return error;
111 if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), bh, GFS2_METATYPE_JD)) {
112 brelse(bh);
113 return -EIO;
114 }
115 *bhp = bh;
116 return 0;
117 }
118
119 static int gfs2_dir_write_stuffed(struct gfs2_inode *ip, const char *buf,
120 unsigned int offset, unsigned int size)
121 {
122 struct buffer_head *dibh;
123 int error;
124
125 error = gfs2_meta_inode_buffer(ip, &dibh);
126 if (error)
127 return error;
128
129 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
130 memcpy(dibh->b_data + offset + sizeof(struct gfs2_dinode), buf, size);
131 if (ip->i_di.di_size < offset + size)
132 ip->i_di.di_size = offset + size;
133 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
134 gfs2_dinode_out(ip, dibh->b_data);
135
136 brelse(dibh);
137
138 return size;
139 }
140
141
142
143 /**
144 * gfs2_dir_write_data - Write directory information to the inode
145 * @ip: The GFS2 inode
146 * @buf: The buffer containing information to be written
147 * @offset: The file offset to start writing at
148 * @size: The amount of data to write
149 *
150 * Returns: The number of bytes correctly written or error code
151 */
152 static int gfs2_dir_write_data(struct gfs2_inode *ip, const char *buf,
153 u64 offset, unsigned int size)
154 {
155 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
156 struct buffer_head *dibh;
157 u64 lblock, dblock;
158 u32 extlen = 0;
159 unsigned int o;
160 int copied = 0;
161 int error = 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 int new = 0;
187
188 amount = size - copied;
189 if (amount > sdp->sd_sb.sb_bsize - o)
190 amount = sdp->sd_sb.sb_bsize - o;
191
192 if (!extlen) {
193 new = 1;
194 error = gfs2_extent_map(&ip->i_inode, lblock, &new,
195 &dblock, &extlen);
196 if (error)
197 goto fail;
198 error = -EIO;
199 if (gfs2_assert_withdraw(sdp, dblock))
200 goto fail;
201 }
202
203 if (amount == sdp->sd_jbsize || new)
204 error = gfs2_dir_get_new_buffer(ip, dblock, &bh);
205 else
206 error = gfs2_dir_get_existing_buffer(ip, dblock, &bh);
207
208 if (error)
209 goto fail;
210
211 gfs2_trans_add_bh(ip->i_gl, bh, 1);
212 memcpy(bh->b_data + o, buf, amount);
213 brelse(bh);
214
215 buf += amount;
216 copied += amount;
217 lblock++;
218 dblock++;
219 extlen--;
220
221 o = sizeof(struct gfs2_meta_header);
222 }
223
224 out:
225 error = gfs2_meta_inode_buffer(ip, &dibh);
226 if (error)
227 return error;
228
229 if (ip->i_di.di_size < offset + copied)
230 ip->i_di.di_size = offset + copied;
231 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
232
233 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
234 gfs2_dinode_out(ip, dibh->b_data);
235 brelse(dibh);
236
237 return copied;
238 fail:
239 if (copied)
240 goto out;
241 return error;
242 }
243
244 static int gfs2_dir_read_stuffed(struct gfs2_inode *ip, char *buf,
245 u64 offset, unsigned int size)
246 {
247 struct buffer_head *dibh;
248 int error;
249
250 error = gfs2_meta_inode_buffer(ip, &dibh);
251 if (!error) {
252 offset += sizeof(struct gfs2_dinode);
253 memcpy(buf, dibh->b_data + offset, size);
254 brelse(dibh);
255 }
256
257 return (error) ? error : size;
258 }
259
260
261 /**
262 * gfs2_dir_read_data - Read a data from a directory inode
263 * @ip: The GFS2 Inode
264 * @buf: The buffer to place result into
265 * @offset: File offset to begin jdata_readng from
266 * @size: Amount of data to transfer
267 *
268 * Returns: The amount of data actually copied or the error
269 */
270 static int gfs2_dir_read_data(struct gfs2_inode *ip, char *buf, u64 offset,
271 unsigned int size, unsigned ra)
272 {
273 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
274 u64 lblock, dblock;
275 u32 extlen = 0;
276 unsigned int o;
277 int copied = 0;
278 int error = 0;
279
280 if (offset >= ip->i_di.di_size)
281 return 0;
282
283 if (offset + size > ip->i_di.di_size)
284 size = ip->i_di.di_size - offset;
285
286 if (!size)
287 return 0;
288
289 if (gfs2_is_stuffed(ip))
290 return gfs2_dir_read_stuffed(ip, buf, offset, size);
291
292 if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
293 return -EINVAL;
294
295 lblock = offset;
296 o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
297
298 while (copied < size) {
299 unsigned int amount;
300 struct buffer_head *bh;
301 int new;
302
303 amount = size - copied;
304 if (amount > sdp->sd_sb.sb_bsize - o)
305 amount = sdp->sd_sb.sb_bsize - o;
306
307 if (!extlen) {
308 new = 0;
309 error = gfs2_extent_map(&ip->i_inode, lblock, &new,
310 &dblock, &extlen);
311 if (error || !dblock)
312 goto fail;
313 BUG_ON(extlen < 1);
314 if (!ra)
315 extlen = 1;
316 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
317 } else {
318 error = gfs2_meta_read(ip->i_gl, dblock, DIO_WAIT, &bh);
319 if (error)
320 goto fail;
321 }
322 error = gfs2_metatype_check(sdp, bh, GFS2_METATYPE_JD);
323 if (error) {
324 brelse(bh);
325 goto fail;
326 }
327 dblock++;
328 extlen--;
329 memcpy(buf, bh->b_data + o, amount);
330 brelse(bh);
331 buf += amount;
332 copied += amount;
333 lblock++;
334 o = sizeof(struct gfs2_meta_header);
335 }
336
337 return copied;
338 fail:
339 return (copied) ? copied : error;
340 }
341
342 static inline int gfs2_dirent_sentinel(const struct gfs2_dirent *dent)
343 {
344 return dent->de_inum.no_addr == 0 || dent->de_inum.no_formal_ino == 0;
345 }
346
347 static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent,
348 const struct qstr *name, int ret)
349 {
350 if (!gfs2_dirent_sentinel(dent) &&
351 be32_to_cpu(dent->de_hash) == name->hash &&
352 be16_to_cpu(dent->de_name_len) == name->len &&
353 memcmp(dent+1, name->name, name->len) == 0)
354 return ret;
355 return 0;
356 }
357
358 static int gfs2_dirent_find(const struct gfs2_dirent *dent,
359 const struct qstr *name,
360 void *opaque)
361 {
362 return __gfs2_dirent_find(dent, name, 1);
363 }
364
365 static int gfs2_dirent_prev(const struct gfs2_dirent *dent,
366 const struct qstr *name,
367 void *opaque)
368 {
369 return __gfs2_dirent_find(dent, name, 2);
370 }
371
372 /*
373 * name->name holds ptr to start of block.
374 * name->len holds size of block.
375 */
376 static int gfs2_dirent_last(const struct gfs2_dirent *dent,
377 const struct qstr *name,
378 void *opaque)
379 {
380 const char *start = name->name;
381 const char *end = (const char *)dent + be16_to_cpu(dent->de_rec_len);
382 if (name->len == (end - start))
383 return 1;
384 return 0;
385 }
386
387 static int gfs2_dirent_find_space(const struct gfs2_dirent *dent,
388 const struct qstr *name,
389 void *opaque)
390 {
391 unsigned required = GFS2_DIRENT_SIZE(name->len);
392 unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
393 unsigned totlen = be16_to_cpu(dent->de_rec_len);
394
395 if (gfs2_dirent_sentinel(dent))
396 actual = GFS2_DIRENT_SIZE(0);
397 if (totlen - actual >= required)
398 return 1;
399 return 0;
400 }
401
402 struct dirent_gather {
403 const struct gfs2_dirent **pdent;
404 unsigned offset;
405 };
406
407 static int gfs2_dirent_gather(const struct gfs2_dirent *dent,
408 const struct qstr *name,
409 void *opaque)
410 {
411 struct dirent_gather *g = opaque;
412 if (!gfs2_dirent_sentinel(dent)) {
413 g->pdent[g->offset++] = dent;
414 }
415 return 0;
416 }
417
418 /*
419 * Other possible things to check:
420 * - Inode located within filesystem size (and on valid block)
421 * - Valid directory entry type
422 * Not sure how heavy-weight we want to make this... could also check
423 * hash is correct for example, but that would take a lot of extra time.
424 * For now the most important thing is to check that the various sizes
425 * are correct.
426 */
427 static int gfs2_check_dirent(struct gfs2_dirent *dent, unsigned int offset,
428 unsigned int size, unsigned int len, int first)
429 {
430 const char *msg = "gfs2_dirent too small";
431 if (unlikely(size < sizeof(struct gfs2_dirent)))
432 goto error;
433 msg = "gfs2_dirent misaligned";
434 if (unlikely(offset & 0x7))
435 goto error;
436 msg = "gfs2_dirent points beyond end of block";
437 if (unlikely(offset + size > len))
438 goto error;
439 msg = "zero inode number";
440 if (unlikely(!first && gfs2_dirent_sentinel(dent)))
441 goto error;
442 msg = "name length is greater than space in dirent";
443 if (!gfs2_dirent_sentinel(dent) &&
444 unlikely(sizeof(struct gfs2_dirent)+be16_to_cpu(dent->de_name_len) >
445 size))
446 goto error;
447 return 0;
448 error:
449 printk(KERN_WARNING "gfs2_check_dirent: %s (%s)\n", msg,
450 first ? "first in block" : "not first in block");
451 return -EIO;
452 }
453
454 static int gfs2_dirent_offset(const void *buf)
455 {
456 const struct gfs2_meta_header *h = buf;
457 int offset;
458
459 BUG_ON(buf == NULL);
460
461 switch(be32_to_cpu(h->mh_type)) {
462 case GFS2_METATYPE_LF:
463 offset = sizeof(struct gfs2_leaf);
464 break;
465 case GFS2_METATYPE_DI:
466 offset = sizeof(struct gfs2_dinode);
467 break;
468 default:
469 goto wrong_type;
470 }
471 return offset;
472 wrong_type:
473 printk(KERN_WARNING "gfs2_scan_dirent: wrong block type %u\n",
474 be32_to_cpu(h->mh_type));
475 return -1;
476 }
477
478 static struct gfs2_dirent *gfs2_dirent_scan(struct inode *inode, void *buf,
479 unsigned int len, gfs2_dscan_t scan,
480 const struct qstr *name,
481 void *opaque)
482 {
483 struct gfs2_dirent *dent, *prev;
484 unsigned offset;
485 unsigned size;
486 int ret = 0;
487
488 ret = gfs2_dirent_offset(buf);
489 if (ret < 0)
490 goto consist_inode;
491
492 offset = ret;
493 prev = NULL;
494 dent = buf + offset;
495 size = be16_to_cpu(dent->de_rec_len);
496 if (gfs2_check_dirent(dent, offset, size, len, 1))
497 goto consist_inode;
498 do {
499 ret = scan(dent, name, opaque);
500 if (ret)
501 break;
502 offset += size;
503 if (offset == len)
504 break;
505 prev = dent;
506 dent = buf + offset;
507 size = be16_to_cpu(dent->de_rec_len);
508 if (gfs2_check_dirent(dent, offset, size, len, 0))
509 goto consist_inode;
510 } while(1);
511
512 switch(ret) {
513 case 0:
514 return NULL;
515 case 1:
516 return dent;
517 case 2:
518 return prev ? prev : dent;
519 default:
520 BUG_ON(ret > 0);
521 return ERR_PTR(ret);
522 }
523
524 consist_inode:
525 gfs2_consist_inode(GFS2_I(inode));
526 return ERR_PTR(-EIO);
527 }
528
529
530 /**
531 * dirent_first - Return the first dirent
532 * @dip: the directory
533 * @bh: The buffer
534 * @dent: Pointer to list of dirents
535 *
536 * return first dirent whether bh points to leaf or stuffed dinode
537 *
538 * Returns: IS_LEAF, IS_DINODE, or -errno
539 */
540
541 static int dirent_first(struct gfs2_inode *dip, struct buffer_head *bh,
542 struct gfs2_dirent **dent)
543 {
544 struct gfs2_meta_header *h = (struct gfs2_meta_header *)bh->b_data;
545
546 if (be32_to_cpu(h->mh_type) == GFS2_METATYPE_LF) {
547 if (gfs2_meta_check(GFS2_SB(&dip->i_inode), bh))
548 return -EIO;
549 *dent = (struct gfs2_dirent *)(bh->b_data +
550 sizeof(struct gfs2_leaf));
551 return IS_LEAF;
552 } else {
553 if (gfs2_metatype_check(GFS2_SB(&dip->i_inode), bh, GFS2_METATYPE_DI))
554 return -EIO;
555 *dent = (struct gfs2_dirent *)(bh->b_data +
556 sizeof(struct gfs2_dinode));
557 return IS_DINODE;
558 }
559 }
560
561 static int dirent_check_reclen(struct gfs2_inode *dip,
562 const struct gfs2_dirent *d, const void *end_p)
563 {
564 const void *ptr = d;
565 u16 rec_len = be16_to_cpu(d->de_rec_len);
566
567 if (unlikely(rec_len < sizeof(struct gfs2_dirent)))
568 goto broken;
569 ptr += rec_len;
570 if (ptr < end_p)
571 return rec_len;
572 if (ptr == end_p)
573 return -ENOENT;
574 broken:
575 gfs2_consist_inode(dip);
576 return -EIO;
577 }
578
579 /**
580 * dirent_next - Next dirent
581 * @dip: the directory
582 * @bh: The buffer
583 * @dent: Pointer to list of dirents
584 *
585 * Returns: 0 on success, error code otherwise
586 */
587
588 static int dirent_next(struct gfs2_inode *dip, struct buffer_head *bh,
589 struct gfs2_dirent **dent)
590 {
591 struct gfs2_dirent *cur = *dent, *tmp;
592 char *bh_end = bh->b_data + bh->b_size;
593 int ret;
594
595 ret = dirent_check_reclen(dip, cur, bh_end);
596 if (ret < 0)
597 return ret;
598
599 tmp = (void *)cur + ret;
600 ret = dirent_check_reclen(dip, tmp, bh_end);
601 if (ret == -EIO)
602 return ret;
603
604 /* Only the first dent could ever have de_inum.no_addr == 0 */
605 if (gfs2_dirent_sentinel(tmp)) {
606 gfs2_consist_inode(dip);
607 return -EIO;
608 }
609
610 *dent = tmp;
611 return 0;
612 }
613
614 /**
615 * dirent_del - Delete a dirent
616 * @dip: The GFS2 inode
617 * @bh: The buffer
618 * @prev: The previous dirent
619 * @cur: The current dirent
620 *
621 */
622
623 static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh,
624 struct gfs2_dirent *prev, struct gfs2_dirent *cur)
625 {
626 u16 cur_rec_len, prev_rec_len;
627
628 if (gfs2_dirent_sentinel(cur)) {
629 gfs2_consist_inode(dip);
630 return;
631 }
632
633 gfs2_trans_add_bh(dip->i_gl, bh, 1);
634
635 /* If there is no prev entry, this is the first entry in the block.
636 The de_rec_len is already as big as it needs to be. Just zero
637 out the inode number and return. */
638
639 if (!prev) {
640 cur->de_inum.no_addr = 0;
641 cur->de_inum.no_formal_ino = 0;
642 return;
643 }
644
645 /* Combine this dentry with the previous one. */
646
647 prev_rec_len = be16_to_cpu(prev->de_rec_len);
648 cur_rec_len = be16_to_cpu(cur->de_rec_len);
649
650 if ((char *)prev + prev_rec_len != (char *)cur)
651 gfs2_consist_inode(dip);
652 if ((char *)cur + cur_rec_len > bh->b_data + bh->b_size)
653 gfs2_consist_inode(dip);
654
655 prev_rec_len += cur_rec_len;
656 prev->de_rec_len = cpu_to_be16(prev_rec_len);
657 }
658
659 /*
660 * Takes a dent from which to grab space as an argument. Returns the
661 * newly created dent.
662 */
663 static struct gfs2_dirent *gfs2_init_dirent(struct inode *inode,
664 struct gfs2_dirent *dent,
665 const struct qstr *name,
666 struct buffer_head *bh)
667 {
668 struct gfs2_inode *ip = GFS2_I(inode);
669 struct gfs2_dirent *ndent;
670 unsigned offset = 0, totlen;
671
672 if (!gfs2_dirent_sentinel(dent))
673 offset = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
674 totlen = be16_to_cpu(dent->de_rec_len);
675 BUG_ON(offset + name->len > totlen);
676 gfs2_trans_add_bh(ip->i_gl, bh, 1);
677 ndent = (struct gfs2_dirent *)((char *)dent + offset);
678 dent->de_rec_len = cpu_to_be16(offset);
679 gfs2_qstr2dirent(name, totlen - offset, ndent);
680 return ndent;
681 }
682
683 static struct gfs2_dirent *gfs2_dirent_alloc(struct inode *inode,
684 struct buffer_head *bh,
685 const struct qstr *name)
686 {
687 struct gfs2_dirent *dent;
688 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
689 gfs2_dirent_find_space, name, NULL);
690 if (!dent || IS_ERR(dent))
691 return dent;
692 return gfs2_init_dirent(inode, dent, name, bh);
693 }
694
695 static int get_leaf(struct gfs2_inode *dip, u64 leaf_no,
696 struct buffer_head **bhp)
697 {
698 int error;
699
700 error = gfs2_meta_read(dip->i_gl, leaf_no, DIO_WAIT, bhp);
701 if (!error && gfs2_metatype_check(GFS2_SB(&dip->i_inode), *bhp, GFS2_METATYPE_LF)) {
702 /* printk(KERN_INFO "block num=%llu\n", leaf_no); */
703 error = -EIO;
704 }
705
706 return error;
707 }
708
709 /**
710 * get_leaf_nr - Get a leaf number associated with the index
711 * @dip: The GFS2 inode
712 * @index:
713 * @leaf_out:
714 *
715 * Returns: 0 on success, error code otherwise
716 */
717
718 static int get_leaf_nr(struct gfs2_inode *dip, u32 index,
719 u64 *leaf_out)
720 {
721 __be64 leaf_no;
722 int error;
723
724 error = gfs2_dir_read_data(dip, (char *)&leaf_no,
725 index * sizeof(__be64),
726 sizeof(__be64), 0);
727 if (error != sizeof(u64))
728 return (error < 0) ? error : -EIO;
729
730 *leaf_out = be64_to_cpu(leaf_no);
731
732 return 0;
733 }
734
735 static int get_first_leaf(struct gfs2_inode *dip, u32 index,
736 struct buffer_head **bh_out)
737 {
738 u64 leaf_no;
739 int error;
740
741 error = get_leaf_nr(dip, index, &leaf_no);
742 if (!error)
743 error = get_leaf(dip, leaf_no, bh_out);
744
745 return error;
746 }
747
748 static struct gfs2_dirent *gfs2_dirent_search(struct inode *inode,
749 const struct qstr *name,
750 gfs2_dscan_t scan,
751 struct buffer_head **pbh)
752 {
753 struct buffer_head *bh;
754 struct gfs2_dirent *dent;
755 struct gfs2_inode *ip = GFS2_I(inode);
756 int error;
757
758 if (ip->i_di.di_flags & GFS2_DIF_EXHASH) {
759 struct gfs2_leaf *leaf;
760 unsigned hsize = 1 << ip->i_depth;
761 unsigned index;
762 u64 ln;
763 if (hsize * sizeof(u64) != ip->i_di.di_size) {
764 gfs2_consist_inode(ip);
765 return ERR_PTR(-EIO);
766 }
767
768 index = name->hash >> (32 - ip->i_depth);
769 error = get_first_leaf(ip, index, &bh);
770 if (error)
771 return ERR_PTR(error);
772 do {
773 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
774 scan, name, NULL);
775 if (dent)
776 goto got_dent;
777 leaf = (struct gfs2_leaf *)bh->b_data;
778 ln = be64_to_cpu(leaf->lf_next);
779 brelse(bh);
780 if (!ln)
781 break;
782
783 error = get_leaf(ip, ln, &bh);
784 } while(!error);
785
786 return error ? ERR_PTR(error) : NULL;
787 }
788
789
790 error = gfs2_meta_inode_buffer(ip, &bh);
791 if (error)
792 return ERR_PTR(error);
793 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, scan, name, NULL);
794 got_dent:
795 if (unlikely(dent == NULL || IS_ERR(dent))) {
796 brelse(bh);
797 bh = NULL;
798 }
799 *pbh = bh;
800 return dent;
801 }
802
803 static struct gfs2_leaf *new_leaf(struct inode *inode, struct buffer_head **pbh, u16 depth)
804 {
805 struct gfs2_inode *ip = GFS2_I(inode);
806 unsigned int n = 1;
807 u64 bn = gfs2_alloc_block(ip, &n);
808 struct buffer_head *bh = gfs2_meta_new(ip->i_gl, bn);
809 struct gfs2_leaf *leaf;
810 struct gfs2_dirent *dent;
811 struct qstr name = { .name = "", .len = 0, .hash = 0 };
812 if (!bh)
813 return NULL;
814 gfs2_trans_add_unrevoke(GFS2_SB(inode), bn, 1);
815 gfs2_trans_add_bh(ip->i_gl, bh, 1);
816 gfs2_metatype_set(bh, GFS2_METATYPE_LF, GFS2_FORMAT_LF);
817 leaf = (struct gfs2_leaf *)bh->b_data;
818 leaf->lf_depth = cpu_to_be16(depth);
819 leaf->lf_entries = 0;
820 leaf->lf_dirent_format = cpu_to_be32(GFS2_FORMAT_DE);
821 leaf->lf_next = 0;
822 memset(leaf->lf_reserved, 0, sizeof(leaf->lf_reserved));
823 dent = (struct gfs2_dirent *)(leaf+1);
824 gfs2_qstr2dirent(&name, bh->b_size - sizeof(struct gfs2_leaf), dent);
825 *pbh = bh;
826 return leaf;
827 }
828
829 /**
830 * dir_make_exhash - Convert a stuffed directory into an ExHash directory
831 * @dip: The GFS2 inode
832 *
833 * Returns: 0 on success, error code otherwise
834 */
835
836 static int dir_make_exhash(struct inode *inode)
837 {
838 struct gfs2_inode *dip = GFS2_I(inode);
839 struct gfs2_sbd *sdp = GFS2_SB(inode);
840 struct gfs2_dirent *dent;
841 struct qstr args;
842 struct buffer_head *bh, *dibh;
843 struct gfs2_leaf *leaf;
844 int y;
845 u32 x;
846 __be64 *lp;
847 u64 bn;
848 int error;
849
850 error = gfs2_meta_inode_buffer(dip, &dibh);
851 if (error)
852 return error;
853
854 /* Turn over a new leaf */
855
856 leaf = new_leaf(inode, &bh, 0);
857 if (!leaf)
858 return -ENOSPC;
859 bn = bh->b_blocknr;
860
861 gfs2_assert(sdp, dip->i_di.di_entries < (1 << 16));
862 leaf->lf_entries = cpu_to_be16(dip->i_di.di_entries);
863
864 /* Copy dirents */
865
866 gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_leaf), dibh,
867 sizeof(struct gfs2_dinode));
868
869 /* Find last entry */
870
871 x = 0;
872 args.len = bh->b_size - sizeof(struct gfs2_dinode) +
873 sizeof(struct gfs2_leaf);
874 args.name = bh->b_data;
875 dent = gfs2_dirent_scan(&dip->i_inode, bh->b_data, bh->b_size,
876 gfs2_dirent_last, &args, NULL);
877 if (!dent) {
878 brelse(bh);
879 brelse(dibh);
880 return -EIO;
881 }
882 if (IS_ERR(dent)) {
883 brelse(bh);
884 brelse(dibh);
885 return PTR_ERR(dent);
886 }
887
888 /* Adjust the last dirent's record length
889 (Remember that dent still points to the last entry.) */
890
891 dent->de_rec_len = cpu_to_be16(be16_to_cpu(dent->de_rec_len) +
892 sizeof(struct gfs2_dinode) -
893 sizeof(struct gfs2_leaf));
894
895 brelse(bh);
896
897 /* We're done with the new leaf block, now setup the new
898 hash table. */
899
900 gfs2_trans_add_bh(dip->i_gl, dibh, 1);
901 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
902
903 lp = (__be64 *)(dibh->b_data + sizeof(struct gfs2_dinode));
904
905 for (x = sdp->sd_hash_ptrs; x--; lp++)
906 *lp = cpu_to_be64(bn);
907
908 dip->i_di.di_size = sdp->sd_sb.sb_bsize / 2;
909 gfs2_add_inode_blocks(&dip->i_inode, 1);
910 dip->i_di.di_flags |= GFS2_DIF_EXHASH;
911
912 for (x = sdp->sd_hash_ptrs, y = -1; x; x >>= 1, y++) ;
913 dip->i_depth = y;
914
915 gfs2_dinode_out(dip, dibh->b_data);
916
917 brelse(dibh);
918
919 return 0;
920 }
921
922 /**
923 * dir_split_leaf - Split a leaf block into two
924 * @dip: The GFS2 inode
925 * @index:
926 * @leaf_no:
927 *
928 * Returns: 0 on success, error code on failure
929 */
930
931 static int dir_split_leaf(struct inode *inode, const struct qstr *name)
932 {
933 struct gfs2_inode *dip = GFS2_I(inode);
934 struct buffer_head *nbh, *obh, *dibh;
935 struct gfs2_leaf *nleaf, *oleaf;
936 struct gfs2_dirent *dent = NULL, *prev = NULL, *next = NULL, *new;
937 u32 start, len, half_len, divider;
938 u64 bn, leaf_no;
939 __be64 *lp;
940 u32 index;
941 int x, moved = 0;
942 int error;
943
944 index = name->hash >> (32 - dip->i_depth);
945 error = get_leaf_nr(dip, index, &leaf_no);
946 if (error)
947 return error;
948
949 /* Get the old leaf block */
950 error = get_leaf(dip, leaf_no, &obh);
951 if (error)
952 return error;
953
954 oleaf = (struct gfs2_leaf *)obh->b_data;
955 if (dip->i_depth == be16_to_cpu(oleaf->lf_depth)) {
956 brelse(obh);
957 return 1; /* can't split */
958 }
959
960 gfs2_trans_add_bh(dip->i_gl, obh, 1);
961
962 nleaf = new_leaf(inode, &nbh, be16_to_cpu(oleaf->lf_depth) + 1);
963 if (!nleaf) {
964 brelse(obh);
965 return -ENOSPC;
966 }
967 bn = nbh->b_blocknr;
968
969 /* Compute the start and len of leaf pointers in the hash table. */
970 len = 1 << (dip->i_depth - be16_to_cpu(oleaf->lf_depth));
971 half_len = len >> 1;
972 if (!half_len) {
973 printk(KERN_WARNING "i_depth %u lf_depth %u index %u\n", dip->i_depth, be16_to_cpu(oleaf->lf_depth), index);
974 gfs2_consist_inode(dip);
975 error = -EIO;
976 goto fail_brelse;
977 }
978
979 start = (index & ~(len - 1));
980
981 /* Change the pointers.
982 Don't bother distinguishing stuffed from non-stuffed.
983 This code is complicated enough already. */
984 lp = kmalloc(half_len * sizeof(__be64), GFP_NOFS | __GFP_NOFAIL);
985 /* Change the pointers */
986 for (x = 0; x < half_len; x++)
987 lp[x] = cpu_to_be64(bn);
988
989 error = gfs2_dir_write_data(dip, (char *)lp, start * sizeof(u64),
990 half_len * sizeof(u64));
991 if (error != half_len * sizeof(u64)) {
992 if (error >= 0)
993 error = -EIO;
994 goto fail_lpfree;
995 }
996
997 kfree(lp);
998
999 /* Compute the divider */
1000 divider = (start + half_len) << (32 - dip->i_depth);
1001
1002 /* Copy the entries */
1003 dirent_first(dip, obh, &dent);
1004
1005 do {
1006 next = dent;
1007 if (dirent_next(dip, obh, &next))
1008 next = NULL;
1009
1010 if (!gfs2_dirent_sentinel(dent) &&
1011 be32_to_cpu(dent->de_hash) < divider) {
1012 struct qstr str;
1013 str.name = (char*)(dent+1);
1014 str.len = be16_to_cpu(dent->de_name_len);
1015 str.hash = be32_to_cpu(dent->de_hash);
1016 new = gfs2_dirent_alloc(inode, nbh, &str);
1017 if (IS_ERR(new)) {
1018 error = PTR_ERR(new);
1019 break;
1020 }
1021
1022 new->de_inum = dent->de_inum; /* No endian worries */
1023 new->de_type = dent->de_type; /* No endian worries */
1024 be16_add_cpu(&nleaf->lf_entries, 1);
1025
1026 dirent_del(dip, obh, prev, dent);
1027
1028 if (!oleaf->lf_entries)
1029 gfs2_consist_inode(dip);
1030 be16_add_cpu(&oleaf->lf_entries, -1);
1031
1032 if (!prev)
1033 prev = dent;
1034
1035 moved = 1;
1036 } else {
1037 prev = dent;
1038 }
1039 dent = next;
1040 } while (dent);
1041
1042 oleaf->lf_depth = nleaf->lf_depth;
1043
1044 error = gfs2_meta_inode_buffer(dip, &dibh);
1045 if (!gfs2_assert_withdraw(GFS2_SB(&dip->i_inode), !error)) {
1046 gfs2_trans_add_bh(dip->i_gl, dibh, 1);
1047 gfs2_add_inode_blocks(&dip->i_inode, 1);
1048 gfs2_dinode_out(dip, dibh->b_data);
1049 brelse(dibh);
1050 }
1051
1052 brelse(obh);
1053 brelse(nbh);
1054
1055 return error;
1056
1057 fail_lpfree:
1058 kfree(lp);
1059
1060 fail_brelse:
1061 brelse(obh);
1062 brelse(nbh);
1063 return error;
1064 }
1065
1066 /**
1067 * dir_double_exhash - Double size of ExHash table
1068 * @dip: The GFS2 dinode
1069 *
1070 * Returns: 0 on success, error code on failure
1071 */
1072
1073 static int dir_double_exhash(struct gfs2_inode *dip)
1074 {
1075 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1076 struct buffer_head *dibh;
1077 u32 hsize;
1078 u64 *buf;
1079 u64 *from, *to;
1080 u64 block;
1081 int x;
1082 int error = 0;
1083
1084 hsize = 1 << dip->i_depth;
1085 if (hsize * sizeof(u64) != dip->i_di.di_size) {
1086 gfs2_consist_inode(dip);
1087 return -EIO;
1088 }
1089
1090 /* Allocate both the "from" and "to" buffers in one big chunk */
1091
1092 buf = kcalloc(3, sdp->sd_hash_bsize, GFP_KERNEL | __GFP_NOFAIL);
1093
1094 for (block = dip->i_di.di_size >> sdp->sd_hash_bsize_shift; block--;) {
1095 error = gfs2_dir_read_data(dip, (char *)buf,
1096 block * sdp->sd_hash_bsize,
1097 sdp->sd_hash_bsize, 1);
1098 if (error != sdp->sd_hash_bsize) {
1099 if (error >= 0)
1100 error = -EIO;
1101 goto fail;
1102 }
1103
1104 from = buf;
1105 to = (u64 *)((char *)buf + sdp->sd_hash_bsize);
1106
1107 for (x = sdp->sd_hash_ptrs; x--; from++) {
1108 *to++ = *from; /* No endianess worries */
1109 *to++ = *from;
1110 }
1111
1112 error = gfs2_dir_write_data(dip,
1113 (char *)buf + sdp->sd_hash_bsize,
1114 block * sdp->sd_sb.sb_bsize,
1115 sdp->sd_sb.sb_bsize);
1116 if (error != sdp->sd_sb.sb_bsize) {
1117 if (error >= 0)
1118 error = -EIO;
1119 goto fail;
1120 }
1121 }
1122
1123 kfree(buf);
1124
1125 error = gfs2_meta_inode_buffer(dip, &dibh);
1126 if (!gfs2_assert_withdraw(sdp, !error)) {
1127 dip->i_depth++;
1128 gfs2_dinode_out(dip, dibh->b_data);
1129 brelse(dibh);
1130 }
1131
1132 return error;
1133
1134 fail:
1135 kfree(buf);
1136 return error;
1137 }
1138
1139 /**
1140 * compare_dents - compare directory entries by hash value
1141 * @a: first dent
1142 * @b: second dent
1143 *
1144 * When comparing the hash entries of @a to @b:
1145 * gt: returns 1
1146 * lt: returns -1
1147 * eq: returns 0
1148 */
1149
1150 static int compare_dents(const void *a, const void *b)
1151 {
1152 const struct gfs2_dirent *dent_a, *dent_b;
1153 u32 hash_a, hash_b;
1154 int ret = 0;
1155
1156 dent_a = *(const struct gfs2_dirent **)a;
1157 hash_a = be32_to_cpu(dent_a->de_hash);
1158
1159 dent_b = *(const struct gfs2_dirent **)b;
1160 hash_b = be32_to_cpu(dent_b->de_hash);
1161
1162 if (hash_a > hash_b)
1163 ret = 1;
1164 else if (hash_a < hash_b)
1165 ret = -1;
1166 else {
1167 unsigned int len_a = be16_to_cpu(dent_a->de_name_len);
1168 unsigned int len_b = be16_to_cpu(dent_b->de_name_len);
1169
1170 if (len_a > len_b)
1171 ret = 1;
1172 else if (len_a < len_b)
1173 ret = -1;
1174 else
1175 ret = memcmp(dent_a + 1, dent_b + 1, len_a);
1176 }
1177
1178 return ret;
1179 }
1180
1181 /**
1182 * do_filldir_main - read out directory entries
1183 * @dip: The GFS2 inode
1184 * @offset: The offset in the file to read from
1185 * @opaque: opaque data to pass to filldir
1186 * @filldir: The function to pass entries to
1187 * @darr: an array of struct gfs2_dirent pointers to read
1188 * @entries: the number of entries in darr
1189 * @copied: pointer to int that's non-zero if a entry has been copied out
1190 *
1191 * Jump through some hoops to make sure that if there are hash collsions,
1192 * they are read out at the beginning of a buffer. We want to minimize
1193 * the possibility that they will fall into different readdir buffers or
1194 * that someone will want to seek to that location.
1195 *
1196 * Returns: errno, >0 on exception from filldir
1197 */
1198
1199 static int do_filldir_main(struct gfs2_inode *dip, u64 *offset,
1200 void *opaque, filldir_t filldir,
1201 const struct gfs2_dirent **darr, u32 entries,
1202 int *copied)
1203 {
1204 const struct gfs2_dirent *dent, *dent_next;
1205 u64 off, off_next;
1206 unsigned int x, y;
1207 int run = 0;
1208 int error = 0;
1209
1210 sort(darr, entries, sizeof(struct gfs2_dirent *), compare_dents, NULL);
1211
1212 dent_next = darr[0];
1213 off_next = be32_to_cpu(dent_next->de_hash);
1214 off_next = gfs2_disk_hash2offset(off_next);
1215
1216 for (x = 0, y = 1; x < entries; x++, y++) {
1217 dent = dent_next;
1218 off = off_next;
1219
1220 if (y < entries) {
1221 dent_next = darr[y];
1222 off_next = be32_to_cpu(dent_next->de_hash);
1223 off_next = gfs2_disk_hash2offset(off_next);
1224
1225 if (off < *offset)
1226 continue;
1227 *offset = off;
1228
1229 if (off_next == off) {
1230 if (*copied && !run)
1231 return 1;
1232 run = 1;
1233 } else
1234 run = 0;
1235 } else {
1236 if (off < *offset)
1237 continue;
1238 *offset = off;
1239 }
1240
1241 error = filldir(opaque, (const char *)(dent + 1),
1242 be16_to_cpu(dent->de_name_len),
1243 off, be64_to_cpu(dent->de_inum.no_addr),
1244 be16_to_cpu(dent->de_type));
1245 if (error)
1246 return 1;
1247
1248 *copied = 1;
1249 }
1250
1251 /* Increment the *offset by one, so the next time we come into the
1252 do_filldir fxn, we get the next entry instead of the last one in the
1253 current leaf */
1254
1255 (*offset)++;
1256
1257 return 0;
1258 }
1259
1260 static int gfs2_dir_read_leaf(struct inode *inode, u64 *offset, void *opaque,
1261 filldir_t filldir, int *copied, unsigned *depth,
1262 u64 leaf_no)
1263 {
1264 struct gfs2_inode *ip = GFS2_I(inode);
1265 struct gfs2_sbd *sdp = GFS2_SB(inode);
1266 struct buffer_head *bh;
1267 struct gfs2_leaf *lf;
1268 unsigned entries = 0, entries2 = 0;
1269 unsigned leaves = 0;
1270 const struct gfs2_dirent **darr, *dent;
1271 struct dirent_gather g;
1272 struct buffer_head **larr;
1273 int leaf = 0;
1274 int error, i;
1275 u64 lfn = leaf_no;
1276
1277 do {
1278 error = get_leaf(ip, lfn, &bh);
1279 if (error)
1280 goto out;
1281 lf = (struct gfs2_leaf *)bh->b_data;
1282 if (leaves == 0)
1283 *depth = be16_to_cpu(lf->lf_depth);
1284 entries += be16_to_cpu(lf->lf_entries);
1285 leaves++;
1286 lfn = be64_to_cpu(lf->lf_next);
1287 brelse(bh);
1288 } while(lfn);
1289
1290 if (!entries)
1291 return 0;
1292
1293 error = -ENOMEM;
1294 /*
1295 * The extra 99 entries are not normally used, but are a buffer
1296 * zone in case the number of entries in the leaf is corrupt.
1297 * 99 is the maximum number of entries that can fit in a single
1298 * leaf block.
1299 */
1300 larr = vmalloc((leaves + entries + 99) * sizeof(void *));
1301 if (!larr)
1302 goto out;
1303 darr = (const struct gfs2_dirent **)(larr + leaves);
1304 g.pdent = darr;
1305 g.offset = 0;
1306 lfn = leaf_no;
1307
1308 do {
1309 error = get_leaf(ip, lfn, &bh);
1310 if (error)
1311 goto out_kfree;
1312 lf = (struct gfs2_leaf *)bh->b_data;
1313 lfn = be64_to_cpu(lf->lf_next);
1314 if (lf->lf_entries) {
1315 entries2 += be16_to_cpu(lf->lf_entries);
1316 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
1317 gfs2_dirent_gather, NULL, &g);
1318 error = PTR_ERR(dent);
1319 if (IS_ERR(dent))
1320 goto out_kfree;
1321 if (entries2 != g.offset) {
1322 fs_warn(sdp, "Number of entries corrupt in dir "
1323 "leaf %llu, entries2 (%u) != "
1324 "g.offset (%u)\n",
1325 (unsigned long long)bh->b_blocknr,
1326 entries2, g.offset);
1327
1328 error = -EIO;
1329 goto out_kfree;
1330 }
1331 error = 0;
1332 larr[leaf++] = bh;
1333 } else {
1334 brelse(bh);
1335 }
1336 } while(lfn);
1337
1338 BUG_ON(entries2 != entries);
1339 error = do_filldir_main(ip, offset, opaque, filldir, darr,
1340 entries, copied);
1341 out_kfree:
1342 for(i = 0; i < leaf; i++)
1343 brelse(larr[i]);
1344 vfree(larr);
1345 out:
1346 return error;
1347 }
1348
1349 /**
1350 * dir_e_read - Reads the entries from a directory into a filldir buffer
1351 * @dip: dinode pointer
1352 * @offset: the hash of the last entry read shifted to the right once
1353 * @opaque: buffer for the filldir function to fill
1354 * @filldir: points to the filldir function to use
1355 *
1356 * Returns: errno
1357 */
1358
1359 static int dir_e_read(struct inode *inode, u64 *offset, void *opaque,
1360 filldir_t filldir)
1361 {
1362 struct gfs2_inode *dip = GFS2_I(inode);
1363 struct gfs2_sbd *sdp = GFS2_SB(inode);
1364 u32 hsize, len = 0;
1365 u32 ht_offset, lp_offset, ht_offset_cur = -1;
1366 u32 hash, index;
1367 __be64 *lp;
1368 int copied = 0;
1369 int error = 0;
1370 unsigned depth = 0;
1371
1372 hsize = 1 << dip->i_depth;
1373 if (hsize * sizeof(u64) != dip->i_di.di_size) {
1374 gfs2_consist_inode(dip);
1375 return -EIO;
1376 }
1377
1378 hash = gfs2_dir_offset2hash(*offset);
1379 index = hash >> (32 - dip->i_depth);
1380
1381 lp = kmalloc(sdp->sd_hash_bsize, GFP_KERNEL);
1382 if (!lp)
1383 return -ENOMEM;
1384
1385 while (index < hsize) {
1386 lp_offset = index & (sdp->sd_hash_ptrs - 1);
1387 ht_offset = index - lp_offset;
1388
1389 if (ht_offset_cur != ht_offset) {
1390 error = gfs2_dir_read_data(dip, (char *)lp,
1391 ht_offset * sizeof(__be64),
1392 sdp->sd_hash_bsize, 1);
1393 if (error != sdp->sd_hash_bsize) {
1394 if (error >= 0)
1395 error = -EIO;
1396 goto out;
1397 }
1398 ht_offset_cur = ht_offset;
1399 }
1400
1401 error = gfs2_dir_read_leaf(inode, offset, opaque, filldir,
1402 &copied, &depth,
1403 be64_to_cpu(lp[lp_offset]));
1404 if (error)
1405 break;
1406
1407 len = 1 << (dip->i_depth - depth);
1408 index = (index & ~(len - 1)) + len;
1409 }
1410
1411 out:
1412 kfree(lp);
1413 if (error > 0)
1414 error = 0;
1415 return error;
1416 }
1417
1418 int gfs2_dir_read(struct inode *inode, u64 *offset, void *opaque,
1419 filldir_t filldir)
1420 {
1421 struct gfs2_inode *dip = GFS2_I(inode);
1422 struct gfs2_sbd *sdp = GFS2_SB(inode);
1423 struct dirent_gather g;
1424 const struct gfs2_dirent **darr, *dent;
1425 struct buffer_head *dibh;
1426 int copied = 0;
1427 int error;
1428
1429 if (!dip->i_di.di_entries)
1430 return 0;
1431
1432 if (dip->i_di.di_flags & GFS2_DIF_EXHASH)
1433 return dir_e_read(inode, offset, opaque, filldir);
1434
1435 if (!gfs2_is_stuffed(dip)) {
1436 gfs2_consist_inode(dip);
1437 return -EIO;
1438 }
1439
1440 error = gfs2_meta_inode_buffer(dip, &dibh);
1441 if (error)
1442 return error;
1443
1444 error = -ENOMEM;
1445 /* 96 is max number of dirents which can be stuffed into an inode */
1446 darr = kmalloc(96 * sizeof(struct gfs2_dirent *), GFP_KERNEL);
1447 if (darr) {
1448 g.pdent = darr;
1449 g.offset = 0;
1450 dent = gfs2_dirent_scan(inode, dibh->b_data, dibh->b_size,
1451 gfs2_dirent_gather, NULL, &g);
1452 if (IS_ERR(dent)) {
1453 error = PTR_ERR(dent);
1454 goto out;
1455 }
1456 if (dip->i_di.di_entries != g.offset) {
1457 fs_warn(sdp, "Number of entries corrupt in dir %llu, "
1458 "ip->i_di.di_entries (%u) != g.offset (%u)\n",
1459 (unsigned long long)dip->i_no_addr,
1460 dip->i_di.di_entries,
1461 g.offset);
1462 error = -EIO;
1463 goto out;
1464 }
1465 error = do_filldir_main(dip, offset, opaque, filldir, darr,
1466 dip->i_di.di_entries, &copied);
1467 out:
1468 kfree(darr);
1469 }
1470
1471 if (error > 0)
1472 error = 0;
1473
1474 brelse(dibh);
1475
1476 return error;
1477 }
1478
1479 /**
1480 * gfs2_dir_search - Search a directory
1481 * @dip: The GFS2 inode
1482 * @filename:
1483 * @inode:
1484 *
1485 * This routine searches a directory for a file or another directory.
1486 * Assumes a glock is held on dip.
1487 *
1488 * Returns: errno
1489 */
1490
1491 struct inode *gfs2_dir_search(struct inode *dir, const struct qstr *name)
1492 {
1493 struct buffer_head *bh;
1494 struct gfs2_dirent *dent;
1495 struct inode *inode;
1496
1497 dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
1498 if (dent) {
1499 if (IS_ERR(dent))
1500 return ERR_CAST(dent);
1501 inode = gfs2_inode_lookup(dir->i_sb,
1502 be16_to_cpu(dent->de_type),
1503 be64_to_cpu(dent->de_inum.no_addr),
1504 be64_to_cpu(dent->de_inum.no_formal_ino), 0);
1505 brelse(bh);
1506 return inode;
1507 }
1508 return ERR_PTR(-ENOENT);
1509 }
1510
1511 int gfs2_dir_check(struct inode *dir, const struct qstr *name,
1512 const struct gfs2_inode *ip)
1513 {
1514 struct buffer_head *bh;
1515 struct gfs2_dirent *dent;
1516 int ret = -ENOENT;
1517
1518 dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
1519 if (dent) {
1520 if (IS_ERR(dent))
1521 return PTR_ERR(dent);
1522 if (ip) {
1523 if (be64_to_cpu(dent->de_inum.no_addr) != ip->i_no_addr)
1524 goto out;
1525 if (be64_to_cpu(dent->de_inum.no_formal_ino) !=
1526 ip->i_no_formal_ino)
1527 goto out;
1528 if (unlikely(IF2DT(ip->i_inode.i_mode) !=
1529 be16_to_cpu(dent->de_type))) {
1530 gfs2_consist_inode(GFS2_I(dir));
1531 ret = -EIO;
1532 goto out;
1533 }
1534 }
1535 ret = 0;
1536 out:
1537 brelse(bh);
1538 }
1539 return ret;
1540 }
1541
1542 static int dir_new_leaf(struct inode *inode, const struct qstr *name)
1543 {
1544 struct buffer_head *bh, *obh;
1545 struct gfs2_inode *ip = GFS2_I(inode);
1546 struct gfs2_leaf *leaf, *oleaf;
1547 int error;
1548 u32 index;
1549 u64 bn;
1550
1551 index = name->hash >> (32 - ip->i_depth);
1552 error = get_first_leaf(ip, index, &obh);
1553 if (error)
1554 return error;
1555 do {
1556 oleaf = (struct gfs2_leaf *)obh->b_data;
1557 bn = be64_to_cpu(oleaf->lf_next);
1558 if (!bn)
1559 break;
1560 brelse(obh);
1561 error = get_leaf(ip, bn, &obh);
1562 if (error)
1563 return error;
1564 } while(1);
1565
1566 gfs2_trans_add_bh(ip->i_gl, obh, 1);
1567
1568 leaf = new_leaf(inode, &bh, be16_to_cpu(oleaf->lf_depth));
1569 if (!leaf) {
1570 brelse(obh);
1571 return -ENOSPC;
1572 }
1573 oleaf->lf_next = cpu_to_be64(bh->b_blocknr);
1574 brelse(bh);
1575 brelse(obh);
1576
1577 error = gfs2_meta_inode_buffer(ip, &bh);
1578 if (error)
1579 return error;
1580 gfs2_trans_add_bh(ip->i_gl, bh, 1);
1581 gfs2_add_inode_blocks(&ip->i_inode, 1);
1582 gfs2_dinode_out(ip, bh->b_data);
1583 brelse(bh);
1584 return 0;
1585 }
1586
1587 /**
1588 * gfs2_dir_add - Add new filename into directory
1589 * @dip: The GFS2 inode
1590 * @filename: The new name
1591 * @inode: The inode number of the entry
1592 * @type: The type of the entry
1593 *
1594 * Returns: 0 on success, error code on failure
1595 */
1596
1597 int gfs2_dir_add(struct inode *inode, const struct qstr *name,
1598 const struct gfs2_inode *nip, unsigned type)
1599 {
1600 struct gfs2_inode *ip = GFS2_I(inode);
1601 struct buffer_head *bh;
1602 struct gfs2_dirent *dent;
1603 struct gfs2_leaf *leaf;
1604 int error;
1605
1606 while(1) {
1607 dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space,
1608 &bh);
1609 if (dent) {
1610 if (IS_ERR(dent))
1611 return PTR_ERR(dent);
1612 dent = gfs2_init_dirent(inode, dent, name, bh);
1613 gfs2_inum_out(nip, dent);
1614 dent->de_type = cpu_to_be16(type);
1615 if (ip->i_di.di_flags & GFS2_DIF_EXHASH) {
1616 leaf = (struct gfs2_leaf *)bh->b_data;
1617 be16_add_cpu(&leaf->lf_entries, 1);
1618 }
1619 brelse(bh);
1620 error = gfs2_meta_inode_buffer(ip, &bh);
1621 if (error)
1622 break;
1623 gfs2_trans_add_bh(ip->i_gl, bh, 1);
1624 ip->i_di.di_entries++;
1625 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
1626 gfs2_dinode_out(ip, bh->b_data);
1627 brelse(bh);
1628 error = 0;
1629 break;
1630 }
1631 if (!(ip->i_di.di_flags & GFS2_DIF_EXHASH)) {
1632 error = dir_make_exhash(inode);
1633 if (error)
1634 break;
1635 continue;
1636 }
1637 error = dir_split_leaf(inode, name);
1638 if (error == 0)
1639 continue;
1640 if (error < 0)
1641 break;
1642 if (ip->i_depth < GFS2_DIR_MAX_DEPTH) {
1643 error = dir_double_exhash(ip);
1644 if (error)
1645 break;
1646 error = dir_split_leaf(inode, name);
1647 if (error < 0)
1648 break;
1649 if (error == 0)
1650 continue;
1651 }
1652 error = dir_new_leaf(inode, name);
1653 if (!error)
1654 continue;
1655 error = -ENOSPC;
1656 break;
1657 }
1658 return error;
1659 }
1660
1661
1662 /**
1663 * gfs2_dir_del - Delete a directory entry
1664 * @dip: The GFS2 inode
1665 * @filename: The filename
1666 *
1667 * Returns: 0 on success, error code on failure
1668 */
1669
1670 int gfs2_dir_del(struct gfs2_inode *dip, const struct qstr *name)
1671 {
1672 struct gfs2_dirent *dent, *prev = NULL;
1673 struct buffer_head *bh;
1674 int error;
1675
1676 /* Returns _either_ the entry (if its first in block) or the
1677 previous entry otherwise */
1678 dent = gfs2_dirent_search(&dip->i_inode, name, gfs2_dirent_prev, &bh);
1679 if (!dent) {
1680 gfs2_consist_inode(dip);
1681 return -EIO;
1682 }
1683 if (IS_ERR(dent)) {
1684 gfs2_consist_inode(dip);
1685 return PTR_ERR(dent);
1686 }
1687 /* If not first in block, adjust pointers accordingly */
1688 if (gfs2_dirent_find(dent, name, NULL) == 0) {
1689 prev = dent;
1690 dent = (struct gfs2_dirent *)((char *)dent + be16_to_cpu(prev->de_rec_len));
1691 }
1692
1693 dirent_del(dip, bh, prev, dent);
1694 if (dip->i_di.di_flags & GFS2_DIF_EXHASH) {
1695 struct gfs2_leaf *leaf = (struct gfs2_leaf *)bh->b_data;
1696 u16 entries = be16_to_cpu(leaf->lf_entries);
1697 if (!entries)
1698 gfs2_consist_inode(dip);
1699 leaf->lf_entries = cpu_to_be16(--entries);
1700 }
1701 brelse(bh);
1702
1703 error = gfs2_meta_inode_buffer(dip, &bh);
1704 if (error)
1705 return error;
1706
1707 if (!dip->i_di.di_entries)
1708 gfs2_consist_inode(dip);
1709 gfs2_trans_add_bh(dip->i_gl, bh, 1);
1710 dip->i_di.di_entries--;
1711 dip->i_inode.i_mtime = dip->i_inode.i_ctime = CURRENT_TIME;
1712 gfs2_dinode_out(dip, bh->b_data);
1713 brelse(bh);
1714 mark_inode_dirty(&dip->i_inode);
1715
1716 return error;
1717 }
1718
1719 /**
1720 * gfs2_dir_mvino - Change inode number of directory entry
1721 * @dip: The GFS2 inode
1722 * @filename:
1723 * @new_inode:
1724 *
1725 * This routine changes the inode number of a directory entry. It's used
1726 * by rename to change ".." when a directory is moved.
1727 * Assumes a glock is held on dvp.
1728 *
1729 * Returns: errno
1730 */
1731
1732 int gfs2_dir_mvino(struct gfs2_inode *dip, const struct qstr *filename,
1733 const struct gfs2_inode *nip, unsigned int new_type)
1734 {
1735 struct buffer_head *bh;
1736 struct gfs2_dirent *dent;
1737 int error;
1738
1739 dent = gfs2_dirent_search(&dip->i_inode, filename, gfs2_dirent_find, &bh);
1740 if (!dent) {
1741 gfs2_consist_inode(dip);
1742 return -EIO;
1743 }
1744 if (IS_ERR(dent))
1745 return PTR_ERR(dent);
1746
1747 gfs2_trans_add_bh(dip->i_gl, bh, 1);
1748 gfs2_inum_out(nip, dent);
1749 dent->de_type = cpu_to_be16(new_type);
1750
1751 if (dip->i_di.di_flags & GFS2_DIF_EXHASH) {
1752 brelse(bh);
1753 error = gfs2_meta_inode_buffer(dip, &bh);
1754 if (error)
1755 return error;
1756 gfs2_trans_add_bh(dip->i_gl, bh, 1);
1757 }
1758
1759 dip->i_inode.i_mtime = dip->i_inode.i_ctime = CURRENT_TIME;
1760 gfs2_dinode_out(dip, bh->b_data);
1761 brelse(bh);
1762 return 0;
1763 }
1764
1765 /**
1766 * foreach_leaf - call a function for each leaf in a directory
1767 * @dip: the directory
1768 * @lc: the function to call for each each
1769 * @data: private data to pass to it
1770 *
1771 * Returns: errno
1772 */
1773
1774 static int foreach_leaf(struct gfs2_inode *dip, leaf_call_t lc, void *data)
1775 {
1776 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1777 struct buffer_head *bh;
1778 struct gfs2_leaf *leaf;
1779 u32 hsize, len;
1780 u32 ht_offset, lp_offset, ht_offset_cur = -1;
1781 u32 index = 0;
1782 __be64 *lp;
1783 u64 leaf_no;
1784 int error = 0;
1785
1786 hsize = 1 << dip->i_depth;
1787 if (hsize * sizeof(u64) != dip->i_di.di_size) {
1788 gfs2_consist_inode(dip);
1789 return -EIO;
1790 }
1791
1792 lp = kmalloc(sdp->sd_hash_bsize, GFP_KERNEL);
1793 if (!lp)
1794 return -ENOMEM;
1795
1796 while (index < hsize) {
1797 lp_offset = index & (sdp->sd_hash_ptrs - 1);
1798 ht_offset = index - lp_offset;
1799
1800 if (ht_offset_cur != ht_offset) {
1801 error = gfs2_dir_read_data(dip, (char *)lp,
1802 ht_offset * sizeof(__be64),
1803 sdp->sd_hash_bsize, 1);
1804 if (error != sdp->sd_hash_bsize) {
1805 if (error >= 0)
1806 error = -EIO;
1807 goto out;
1808 }
1809 ht_offset_cur = ht_offset;
1810 }
1811
1812 leaf_no = be64_to_cpu(lp[lp_offset]);
1813 if (leaf_no) {
1814 error = get_leaf(dip, leaf_no, &bh);
1815 if (error)
1816 goto out;
1817 leaf = (struct gfs2_leaf *)bh->b_data;
1818 len = 1 << (dip->i_depth - be16_to_cpu(leaf->lf_depth));
1819 brelse(bh);
1820
1821 error = lc(dip, index, len, leaf_no, data);
1822 if (error)
1823 goto out;
1824
1825 index = (index & ~(len - 1)) + len;
1826 } else
1827 index++;
1828 }
1829
1830 if (index != hsize) {
1831 gfs2_consist_inode(dip);
1832 error = -EIO;
1833 }
1834
1835 out:
1836 kfree(lp);
1837
1838 return error;
1839 }
1840
1841 /**
1842 * leaf_dealloc - Deallocate a directory leaf
1843 * @dip: the directory
1844 * @index: the hash table offset in the directory
1845 * @len: the number of pointers to this leaf
1846 * @leaf_no: the leaf number
1847 * @data: not used
1848 *
1849 * Returns: errno
1850 */
1851
1852 static int leaf_dealloc(struct gfs2_inode *dip, u32 index, u32 len,
1853 u64 leaf_no, void *data)
1854 {
1855 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1856 struct gfs2_leaf *tmp_leaf;
1857 struct gfs2_rgrp_list rlist;
1858 struct buffer_head *bh, *dibh;
1859 u64 blk, nblk;
1860 unsigned int rg_blocks = 0, l_blocks = 0;
1861 char *ht;
1862 unsigned int x, size = len * sizeof(u64);
1863 int error;
1864
1865 memset(&rlist, 0, sizeof(struct gfs2_rgrp_list));
1866
1867 ht = kzalloc(size, GFP_KERNEL);
1868 if (!ht)
1869 return -ENOMEM;
1870
1871 gfs2_alloc_get(dip);
1872
1873 error = gfs2_quota_hold(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
1874 if (error)
1875 goto out;
1876
1877 error = gfs2_rindex_hold(sdp, &dip->i_alloc->al_ri_gh);
1878 if (error)
1879 goto out_qs;
1880
1881 /* Count the number of leaves */
1882
1883 for (blk = leaf_no; blk; blk = nblk) {
1884 error = get_leaf(dip, blk, &bh);
1885 if (error)
1886 goto out_rlist;
1887 tmp_leaf = (struct gfs2_leaf *)bh->b_data;
1888 nblk = be64_to_cpu(tmp_leaf->lf_next);
1889 brelse(bh);
1890
1891 gfs2_rlist_add(sdp, &rlist, blk);
1892 l_blocks++;
1893 }
1894
1895 gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE);
1896
1897 for (x = 0; x < rlist.rl_rgrps; x++) {
1898 struct gfs2_rgrpd *rgd;
1899 rgd = rlist.rl_ghs[x].gh_gl->gl_object;
1900 rg_blocks += rgd->rd_length;
1901 }
1902
1903 error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs);
1904 if (error)
1905 goto out_rlist;
1906
1907 error = gfs2_trans_begin(sdp,
1908 rg_blocks + (DIV_ROUND_UP(size, sdp->sd_jbsize) + 1) +
1909 RES_DINODE + RES_STATFS + RES_QUOTA, l_blocks);
1910 if (error)
1911 goto out_rg_gunlock;
1912
1913 for (blk = leaf_no; blk; blk = nblk) {
1914 error = get_leaf(dip, blk, &bh);
1915 if (error)
1916 goto out_end_trans;
1917 tmp_leaf = (struct gfs2_leaf *)bh->b_data;
1918 nblk = be64_to_cpu(tmp_leaf->lf_next);
1919 brelse(bh);
1920
1921 gfs2_free_meta(dip, blk, 1);
1922 gfs2_add_inode_blocks(&dip->i_inode, -1);
1923 }
1924
1925 error = gfs2_dir_write_data(dip, ht, index * sizeof(u64), size);
1926 if (error != size) {
1927 if (error >= 0)
1928 error = -EIO;
1929 goto out_end_trans;
1930 }
1931
1932 error = gfs2_meta_inode_buffer(dip, &dibh);
1933 if (error)
1934 goto out_end_trans;
1935
1936 gfs2_trans_add_bh(dip->i_gl, dibh, 1);
1937 gfs2_dinode_out(dip, dibh->b_data);
1938 brelse(dibh);
1939
1940 out_end_trans:
1941 gfs2_trans_end(sdp);
1942 out_rg_gunlock:
1943 gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
1944 out_rlist:
1945 gfs2_rlist_free(&rlist);
1946 gfs2_glock_dq_uninit(&dip->i_alloc->al_ri_gh);
1947 out_qs:
1948 gfs2_quota_unhold(dip);
1949 out:
1950 gfs2_alloc_put(dip);
1951 kfree(ht);
1952 return error;
1953 }
1954
1955 /**
1956 * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory
1957 * @dip: the directory
1958 *
1959 * Dealloc all on-disk directory leaves to FREEMETA state
1960 * Change on-disk inode type to "regular file"
1961 *
1962 * Returns: errno
1963 */
1964
1965 int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip)
1966 {
1967 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1968 struct buffer_head *bh;
1969 int error;
1970
1971 /* Dealloc on-disk leaves to FREEMETA state */
1972 error = foreach_leaf(dip, leaf_dealloc, NULL);
1973 if (error)
1974 return error;
1975
1976 /* Make this a regular file in case we crash.
1977 (We don't want to free these blocks a second time.) */
1978
1979 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1980 if (error)
1981 return error;
1982
1983 error = gfs2_meta_inode_buffer(dip, &bh);
1984 if (!error) {
1985 gfs2_trans_add_bh(dip->i_gl, bh, 1);
1986 ((struct gfs2_dinode *)bh->b_data)->di_mode =
1987 cpu_to_be32(S_IFREG);
1988 brelse(bh);
1989 }
1990
1991 gfs2_trans_end(sdp);
1992
1993 return error;
1994 }
1995
1996 /**
1997 * gfs2_diradd_alloc_required - find if adding entry will require an allocation
1998 * @ip: the file being written to
1999 * @filname: the filename that's going to be added
2000 *
2001 * Returns: 1 if alloc required, 0 if not, -ve on error
2002 */
2003
2004 int gfs2_diradd_alloc_required(struct inode *inode, const struct qstr *name)
2005 {
2006 struct gfs2_dirent *dent;
2007 struct buffer_head *bh;
2008
2009 dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, &bh);
2010 if (!dent) {
2011 return 1;
2012 }
2013 if (IS_ERR(dent))
2014 return PTR_ERR(dent);
2015 brelse(bh);
2016 return 0;
2017 }
2018