aboutsummaryrefslogtreecommitdiffstats
path: root/fs/gfs2/dir.c
diff options
context:
space:
mode:
Diffstat (limited to 'fs/gfs2/dir.c')
-rw-r--r--fs/gfs2/dir.c1961
1 files changed, 1961 insertions, 0 deletions
diff --git a/fs/gfs2/dir.c b/fs/gfs2/dir.c
new file mode 100644
index 000000000000..459498cac93b
--- /dev/null
+++ b/fs/gfs2/dir.c
@@ -0,0 +1,1961 @@
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/sched.h>
57#include <linux/slab.h>
58#include <linux/spinlock.h>
59#include <linux/buffer_head.h>
60#include <linux/sort.h>
61#include <linux/gfs2_ondisk.h>
62#include <linux/crc32.h>
63#include <linux/vmalloc.h>
64#include <linux/lm_interface.h>
65
66#include "gfs2.h"
67#include "incore.h"
68#include "dir.h"
69#include "glock.h"
70#include "inode.h"
71#include "meta_io.h"
72#include "quota.h"
73#include "rgrp.h"
74#include "trans.h"
75#include "bmap.h"
76#include "util.h"
77
78#define IS_LEAF 1 /* Hashed (leaf) directory */
79#define IS_DINODE 2 /* Linear (stuffed dinode block) directory */
80
81#define gfs2_disk_hash2offset(h) (((u64)(h)) >> 1)
82#define gfs2_dir_offset2hash(p) ((u32)(((u64)(p)) << 1))
83
84typedef int (*leaf_call_t) (struct gfs2_inode *dip, u32 index, u32 len,
85 u64 leaf_no, void *data);
86typedef int (*gfs2_dscan_t)(const struct gfs2_dirent *dent,
87 const struct qstr *name, void *opaque);
88
89
90int gfs2_dir_get_new_buffer(struct gfs2_inode *ip, u64 block,
91 struct buffer_head **bhp)
92{
93 struct buffer_head *bh;
94
95 bh = gfs2_meta_new(ip->i_gl, block);
96 gfs2_trans_add_bh(ip->i_gl, bh, 1);
97 gfs2_metatype_set(bh, GFS2_METATYPE_JD, GFS2_FORMAT_JD);
98 gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
99 *bhp = bh;
100 return 0;
101}
102
103static int gfs2_dir_get_existing_buffer(struct gfs2_inode *ip, u64 block,
104 struct buffer_head **bhp)
105{
106 struct buffer_head *bh;
107 int error;
108
109 error = gfs2_meta_read(ip->i_gl, block, DIO_WAIT, &bh);
110 if (error)
111 return error;
112 if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), bh, GFS2_METATYPE_JD)) {
113 brelse(bh);
114 return -EIO;
115 }
116 *bhp = bh;
117 return 0;
118}
119
120static int gfs2_dir_write_stuffed(struct gfs2_inode *ip, const char *buf,
121 unsigned int offset, unsigned int size)
122{
123 struct buffer_head *dibh;
124 int error;
125
126 error = gfs2_meta_inode_buffer(ip, &dibh);
127 if (error)
128 return error;
129
130 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
131 memcpy(dibh->b_data + offset + sizeof(struct gfs2_dinode), buf, size);
132 if (ip->i_di.di_size < offset + size)
133 ip->i_di.di_size = offset + size;
134 ip->i_di.di_mtime = ip->i_di.di_ctime = get_seconds();
135 gfs2_dinode_out(&ip->i_di, dibh->b_data);
136
137 brelse(dibh);
138
139 return size;
140}
141
142
143
144/**
145 * gfs2_dir_write_data - Write directory information to the inode
146 * @ip: The GFS2 inode
147 * @buf: The buffer containing information to be written
148 * @offset: The file offset to start writing at
149 * @size: The amount of data to write
150 *
151 * Returns: The number of bytes correctly written or error code
152 */
153static int gfs2_dir_write_data(struct gfs2_inode *ip, const char *buf,
154 u64 offset, unsigned int size)
155{
156 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
157 struct buffer_head *dibh;
158 u64 lblock, dblock;
159 u32 extlen = 0;
160 unsigned int o;
161 int copied = 0;
162 int error = 0;
163
164 if (!size)
165 return 0;
166
167 if (gfs2_is_stuffed(ip) &&
168 offset + size <= sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode))
169 return gfs2_dir_write_stuffed(ip, buf, (unsigned int)offset,
170 size);
171
172 if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
173 return -EINVAL;
174
175 if (gfs2_is_stuffed(ip)) {
176 error = gfs2_unstuff_dinode(ip, NULL);
177 if (error)
178 return error;
179 }
180
181 lblock = offset;
182 o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
183
184 while (copied < size) {
185 unsigned int amount;
186 struct buffer_head *bh;
187 int new;
188
189 amount = size - copied;
190 if (amount > sdp->sd_sb.sb_bsize - o)
191 amount = sdp->sd_sb.sb_bsize - o;
192
193 if (!extlen) {
194 new = 1;
195 error = gfs2_extent_map(&ip->i_inode, lblock, &new,
196 &dblock, &extlen);
197 if (error)
198 goto fail;
199 error = -EIO;
200 if (gfs2_assert_withdraw(sdp, dblock))
201 goto fail;
202 }
203
204 if (amount == sdp->sd_jbsize || new)
205 error = gfs2_dir_get_new_buffer(ip, dblock, &bh);
206 else
207 error = gfs2_dir_get_existing_buffer(ip, dblock, &bh);
208
209 if (error)
210 goto fail;
211
212 gfs2_trans_add_bh(ip->i_gl, bh, 1);
213 memcpy(bh->b_data + o, buf, amount);
214 brelse(bh);
215 if (error)
216 goto fail;
217
218 buf += amount;
219 copied += amount;
220 lblock++;
221 dblock++;
222 extlen--;
223
224 o = sizeof(struct gfs2_meta_header);
225 }
226
227out:
228 error = gfs2_meta_inode_buffer(ip, &dibh);
229 if (error)
230 return error;
231
232 if (ip->i_di.di_size < offset + copied)
233 ip->i_di.di_size = offset + copied;
234 ip->i_di.di_mtime = ip->i_di.di_ctime = get_seconds();
235
236 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
237 gfs2_dinode_out(&ip->i_di, dibh->b_data);
238 brelse(dibh);
239
240 return copied;
241fail:
242 if (copied)
243 goto out;
244 return error;
245}
246
247static int gfs2_dir_read_stuffed(struct gfs2_inode *ip, char *buf,
248 u64 offset, unsigned int size)
249{
250 struct buffer_head *dibh;
251 int error;
252
253 error = gfs2_meta_inode_buffer(ip, &dibh);
254 if (!error) {
255 offset += sizeof(struct gfs2_dinode);
256 memcpy(buf, dibh->b_data + offset, size);
257 brelse(dibh);
258 }
259
260 return (error) ? error : size;
261}
262
263
264/**
265 * gfs2_dir_read_data - Read a data from a directory inode
266 * @ip: The GFS2 Inode
267 * @buf: The buffer to place result into
268 * @offset: File offset to begin jdata_readng from
269 * @size: Amount of data to transfer
270 *
271 * Returns: The amount of data actually copied or the error
272 */
273static int gfs2_dir_read_data(struct gfs2_inode *ip, char *buf, u64 offset,
274 unsigned int size, unsigned ra)
275{
276 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
277 u64 lblock, dblock;
278 u32 extlen = 0;
279 unsigned int o;
280 int copied = 0;
281 int error = 0;
282
283 if (offset >= ip->i_di.di_size)
284 return 0;
285
286 if (offset + size > ip->i_di.di_size)
287 size = ip->i_di.di_size - offset;
288
289 if (!size)
290 return 0;
291
292 if (gfs2_is_stuffed(ip))
293 return gfs2_dir_read_stuffed(ip, buf, offset, size);
294
295 if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
296 return -EINVAL;
297
298 lblock = offset;
299 o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
300
301 while (copied < size) {
302 unsigned int amount;
303 struct buffer_head *bh;
304 int new;
305
306 amount = size - copied;
307 if (amount > sdp->sd_sb.sb_bsize - o)
308 amount = sdp->sd_sb.sb_bsize - o;
309
310 if (!extlen) {
311 new = 0;
312 error = gfs2_extent_map(&ip->i_inode, lblock, &new,
313 &dblock, &extlen);
314 if (error || !dblock)
315 goto fail;
316 BUG_ON(extlen < 1);
317 if (!ra)
318 extlen = 1;
319 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
320 }
321 if (!bh) {
322 error = gfs2_meta_read(ip->i_gl, dblock, DIO_WAIT, &bh);
323 if (error)
324 goto fail;
325 }
326 error = gfs2_metatype_check(sdp, bh, GFS2_METATYPE_JD);
327 if (error) {
328 brelse(bh);
329 goto fail;
330 }
331 dblock++;
332 extlen--;
333 memcpy(buf, bh->b_data + o, amount);
334 brelse(bh);
335 bh = NULL;
336 buf += amount;
337 copied += amount;
338 lblock++;
339 o = sizeof(struct gfs2_meta_header);
340 }
341
342 return copied;
343fail:
344 return (copied) ? copied : error;
345}
346
347static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent,
348 const struct qstr *name, int ret)
349{
350 if (dent->de_inum.no_addr != 0 &&
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
358static 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
365static 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 */
376static 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
387static 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 (!dent->de_inum.no_addr)
396 actual = GFS2_DIRENT_SIZE(0);
397 if (totlen - actual >= required)
398 return 1;
399 return 0;
400}
401
402struct dirent_gather {
403 const struct gfs2_dirent **pdent;
404 unsigned offset;
405};
406
407static 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 (dent->de_inum.no_addr) {
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 */
427static 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 && !dent->de_inum.no_addr))
441 goto error;
442 msg = "name length is greater than space in dirent";
443 if (dent->de_inum.no_addr &&
444 unlikely(sizeof(struct gfs2_dirent)+be16_to_cpu(dent->de_name_len) >
445 size))
446 goto error;
447 return 0;
448error:
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
454static 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;
472wrong_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
478static 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
524consist_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
541static 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
561static 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;
574broken:
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
588static 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 (!tmp->de_inum.no_addr) {
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
623static 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 (!cur->de_inum.no_addr) {
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; /* No endianess worries */
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 */
662static 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 (dent->de_inum.no_addr)
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
682static 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
694static 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
717static int get_leaf_nr(struct gfs2_inode *dip, u32 index,
718 u64 *leaf_out)
719{
720 u64 leaf_no;
721 int error;
722
723 error = gfs2_dir_read_data(dip, (char *)&leaf_no,
724 index * sizeof(u64),
725 sizeof(u64), 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
734static 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
747static 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_di.di_flags & GFS2_DIF_EXHASH) {
758 struct gfs2_leaf *leaf;
759 unsigned hsize = 1 << ip->i_di.di_depth;
760 unsigned index;
761 u64 ln;
762 if (hsize * sizeof(u64) != ip->i_di.di_size) {
763 gfs2_consist_inode(ip);
764 return ERR_PTR(-EIO);
765 }
766
767 index = name->hash >> (32 - ip->i_di.di_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);
793got_dent:
794 if (unlikely(dent == NULL || IS_ERR(dent))) {
795 brelse(bh);
796 bh = NULL;
797 }
798 *pbh = bh;
799 return dent;
800}
801
802static struct gfs2_leaf *new_leaf(struct inode *inode, struct buffer_head **pbh, u16 depth)
803{
804 struct gfs2_inode *ip = GFS2_I(inode);
805 u64 bn = gfs2_alloc_meta(ip);
806 struct buffer_head *bh = gfs2_meta_new(ip->i_gl, bn);
807 struct gfs2_leaf *leaf;
808 struct gfs2_dirent *dent;
809 struct qstr name = { .name = "", .len = 0, .hash = 0 };
810 if (!bh)
811 return NULL;
812
813 gfs2_trans_add_bh(ip->i_gl, bh, 1);
814 gfs2_metatype_set(bh, GFS2_METATYPE_LF, GFS2_FORMAT_LF);
815 leaf = (struct gfs2_leaf *)bh->b_data;
816 leaf->lf_depth = cpu_to_be16(depth);
817 leaf->lf_entries = 0;
818 leaf->lf_dirent_format = cpu_to_be16(GFS2_FORMAT_DE);
819 leaf->lf_next = 0;
820 memset(leaf->lf_reserved, 0, sizeof(leaf->lf_reserved));
821 dent = (struct gfs2_dirent *)(leaf+1);
822 gfs2_qstr2dirent(&name, bh->b_size - sizeof(struct gfs2_leaf), dent);
823 *pbh = bh;
824 return leaf;
825}
826
827/**
828 * dir_make_exhash - Convert a stuffed directory into an ExHash directory
829 * @dip: The GFS2 inode
830 *
831 * Returns: 0 on success, error code otherwise
832 */
833
834static int dir_make_exhash(struct inode *inode)
835{
836 struct gfs2_inode *dip = GFS2_I(inode);
837 struct gfs2_sbd *sdp = GFS2_SB(inode);
838 struct gfs2_dirent *dent;
839 struct qstr args;
840 struct buffer_head *bh, *dibh;
841 struct gfs2_leaf *leaf;
842 int y;
843 u32 x;
844 u64 *lp, bn;
845 int error;
846
847 error = gfs2_meta_inode_buffer(dip, &dibh);
848 if (error)
849 return error;
850
851 /* Turn over a new leaf */
852
853 leaf = new_leaf(inode, &bh, 0);
854 if (!leaf)
855 return -ENOSPC;
856 bn = bh->b_blocknr;
857
858 gfs2_assert(sdp, dip->i_di.di_entries < (1 << 16));
859 leaf->lf_entries = cpu_to_be16(dip->i_di.di_entries);
860
861 /* Copy dirents */
862
863 gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_leaf), dibh,
864 sizeof(struct gfs2_dinode));
865
866 /* Find last entry */
867
868 x = 0;
869 args.len = bh->b_size - sizeof(struct gfs2_dinode) +
870 sizeof(struct gfs2_leaf);
871 args.name = bh->b_data;
872 dent = gfs2_dirent_scan(&dip->i_inode, bh->b_data, bh->b_size,
873 gfs2_dirent_last, &args, NULL);
874 if (!dent) {
875 brelse(bh);
876 brelse(dibh);
877 return -EIO;
878 }
879 if (IS_ERR(dent)) {
880 brelse(bh);
881 brelse(dibh);
882 return PTR_ERR(dent);
883 }
884
885 /* Adjust the last dirent's record length
886 (Remember that dent still points to the last entry.) */
887
888 dent->de_rec_len = cpu_to_be16(be16_to_cpu(dent->de_rec_len) +
889 sizeof(struct gfs2_dinode) -
890 sizeof(struct gfs2_leaf));
891
892 brelse(bh);
893
894 /* We're done with the new leaf block, now setup the new
895 hash table. */
896
897 gfs2_trans_add_bh(dip->i_gl, dibh, 1);
898 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
899
900 lp = (u64 *)(dibh->b_data + sizeof(struct gfs2_dinode));
901
902 for (x = sdp->sd_hash_ptrs; x--; lp++)
903 *lp = cpu_to_be64(bn);
904
905 dip->i_di.di_size = sdp->sd_sb.sb_bsize / 2;
906 dip->i_di.di_blocks++;
907 dip->i_di.di_flags |= GFS2_DIF_EXHASH;
908 dip->i_di.di_payload_format = 0;
909
910 for (x = sdp->sd_hash_ptrs, y = -1; x; x >>= 1, y++) ;
911 dip->i_di.di_depth = y;
912
913 gfs2_dinode_out(&dip->i_di, dibh->b_data);
914
915 brelse(dibh);
916
917 return 0;
918}
919
920/**
921 * dir_split_leaf - Split a leaf block into two
922 * @dip: The GFS2 inode
923 * @index:
924 * @leaf_no:
925 *
926 * Returns: 0 on success, error code on failure
927 */
928
929static int dir_split_leaf(struct inode *inode, const struct qstr *name)
930{
931 struct gfs2_inode *dip = GFS2_I(inode);
932 struct buffer_head *nbh, *obh, *dibh;
933 struct gfs2_leaf *nleaf, *oleaf;
934 struct gfs2_dirent *dent = NULL, *prev = NULL, *next = NULL, *new;
935 u32 start, len, half_len, divider;
936 u64 bn, *lp, leaf_no;
937 u32 index;
938 int x, moved = 0;
939 int error;
940
941 index = name->hash >> (32 - dip->i_di.di_depth);
942 error = get_leaf_nr(dip, index, &leaf_no);
943 if (error)
944 return error;
945
946 /* Get the old leaf block */
947 error = get_leaf(dip, leaf_no, &obh);
948 if (error)
949 return error;
950
951 oleaf = (struct gfs2_leaf *)obh->b_data;
952 if (dip->i_di.di_depth == be16_to_cpu(oleaf->lf_depth)) {
953 brelse(obh);
954 return 1; /* can't split */
955 }
956
957 gfs2_trans_add_bh(dip->i_gl, obh, 1);
958
959 nleaf = new_leaf(inode, &nbh, be16_to_cpu(oleaf->lf_depth) + 1);
960 if (!nleaf) {
961 brelse(obh);
962 return -ENOSPC;
963 }
964 bn = nbh->b_blocknr;
965
966 /* Compute the start and len of leaf pointers in the hash table. */
967 len = 1 << (dip->i_di.di_depth - be16_to_cpu(oleaf->lf_depth));
968 half_len = len >> 1;
969 if (!half_len) {
970 printk(KERN_WARNING "di_depth %u lf_depth %u index %u\n", dip->i_di.di_depth, be16_to_cpu(oleaf->lf_depth), index);
971 gfs2_consist_inode(dip);
972 error = -EIO;
973 goto fail_brelse;
974 }
975
976 start = (index & ~(len - 1));
977
978 /* Change the pointers.
979 Don't bother distinguishing stuffed from non-stuffed.
980 This code is complicated enough already. */
981 lp = kmalloc(half_len * sizeof(u64), GFP_NOFS | __GFP_NOFAIL);
982 /* Change the pointers */
983 for (x = 0; x < half_len; x++)
984 lp[x] = cpu_to_be64(bn);
985
986 error = gfs2_dir_write_data(dip, (char *)lp, start * sizeof(u64),
987 half_len * sizeof(u64));
988 if (error != half_len * sizeof(u64)) {
989 if (error >= 0)
990 error = -EIO;
991 goto fail_lpfree;
992 }
993
994 kfree(lp);
995
996 /* Compute the divider */
997 divider = (start + half_len) << (32 - dip->i_di.di_depth);
998
999 /* Copy the entries */
1000 dirent_first(dip, obh, &dent);
1001
1002 do {
1003 next = dent;
1004 if (dirent_next(dip, obh, &next))
1005 next = NULL;
1006
1007 if (dent->de_inum.no_addr &&
1008 be32_to_cpu(dent->de_hash) < divider) {
1009 struct qstr str;
1010 str.name = (char*)(dent+1);
1011 str.len = be16_to_cpu(dent->de_name_len);
1012 str.hash = be32_to_cpu(dent->de_hash);
1013 new = gfs2_dirent_alloc(inode, nbh, &str);
1014 if (IS_ERR(new)) {
1015 error = PTR_ERR(new);
1016 break;
1017 }
1018
1019 new->de_inum = dent->de_inum; /* No endian worries */
1020 new->de_type = dent->de_type; /* No endian worries */
1021 nleaf->lf_entries = cpu_to_be16(be16_to_cpu(nleaf->lf_entries)+1);
1022
1023 dirent_del(dip, obh, prev, dent);
1024
1025 if (!oleaf->lf_entries)
1026 gfs2_consist_inode(dip);
1027 oleaf->lf_entries = cpu_to_be16(be16_to_cpu(oleaf->lf_entries)-1);
1028
1029 if (!prev)
1030 prev = dent;
1031
1032 moved = 1;
1033 } else {
1034 prev = dent;
1035 }
1036 dent = next;
1037 } while (dent);
1038
1039 oleaf->lf_depth = nleaf->lf_depth;
1040
1041 error = gfs2_meta_inode_buffer(dip, &dibh);
1042 if (!gfs2_assert_withdraw(GFS2_SB(&dip->i_inode), !error)) {
1043 dip->i_di.di_blocks++;
1044 gfs2_dinode_out(&dip->i_di, dibh->b_data);
1045 brelse(dibh);
1046 }
1047
1048 brelse(obh);
1049 brelse(nbh);
1050
1051 return error;
1052
1053fail_lpfree:
1054 kfree(lp);
1055
1056fail_brelse:
1057 brelse(obh);
1058 brelse(nbh);
1059 return error;
1060}
1061
1062/**
1063 * dir_double_exhash - Double size of ExHash table
1064 * @dip: The GFS2 dinode
1065 *
1066 * Returns: 0 on success, error code on failure
1067 */
1068
1069static int dir_double_exhash(struct gfs2_inode *dip)
1070{
1071 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1072 struct buffer_head *dibh;
1073 u32 hsize;
1074 u64 *buf;
1075 u64 *from, *to;
1076 u64 block;
1077 int x;
1078 int error = 0;
1079
1080 hsize = 1 << dip->i_di.di_depth;
1081 if (hsize * sizeof(u64) != dip->i_di.di_size) {
1082 gfs2_consist_inode(dip);
1083 return -EIO;
1084 }
1085
1086 /* Allocate both the "from" and "to" buffers in one big chunk */
1087
1088 buf = kcalloc(3, sdp->sd_hash_bsize, GFP_KERNEL | __GFP_NOFAIL);
1089
1090 for (block = dip->i_di.di_size >> sdp->sd_hash_bsize_shift; block--;) {
1091 error = gfs2_dir_read_data(dip, (char *)buf,
1092 block * sdp->sd_hash_bsize,
1093 sdp->sd_hash_bsize, 1);
1094 if (error != sdp->sd_hash_bsize) {
1095 if (error >= 0)
1096 error = -EIO;
1097 goto fail;
1098 }
1099
1100 from = buf;
1101 to = (u64 *)((char *)buf + sdp->sd_hash_bsize);
1102
1103 for (x = sdp->sd_hash_ptrs; x--; from++) {
1104 *to++ = *from; /* No endianess worries */
1105 *to++ = *from;
1106 }
1107
1108 error = gfs2_dir_write_data(dip,
1109 (char *)buf + sdp->sd_hash_bsize,
1110 block * sdp->sd_sb.sb_bsize,
1111 sdp->sd_sb.sb_bsize);
1112 if (error != sdp->sd_sb.sb_bsize) {
1113 if (error >= 0)
1114 error = -EIO;
1115 goto fail;
1116 }
1117 }
1118
1119 kfree(buf);
1120
1121 error = gfs2_meta_inode_buffer(dip, &dibh);
1122 if (!gfs2_assert_withdraw(sdp, !error)) {
1123 dip->i_di.di_depth++;
1124 gfs2_dinode_out(&dip->i_di, dibh->b_data);
1125 brelse(dibh);
1126 }
1127
1128 return error;
1129
1130fail:
1131 kfree(buf);
1132 return error;
1133}
1134
1135/**
1136 * compare_dents - compare directory entries by hash value
1137 * @a: first dent
1138 * @b: second dent
1139 *
1140 * When comparing the hash entries of @a to @b:
1141 * gt: returns 1
1142 * lt: returns -1
1143 * eq: returns 0
1144 */
1145
1146static int compare_dents(const void *a, const void *b)
1147{
1148 const struct gfs2_dirent *dent_a, *dent_b;
1149 u32 hash_a, hash_b;
1150 int ret = 0;
1151
1152 dent_a = *(const struct gfs2_dirent **)a;
1153 hash_a = be32_to_cpu(dent_a->de_hash);
1154
1155 dent_b = *(const struct gfs2_dirent **)b;
1156 hash_b = be32_to_cpu(dent_b->de_hash);
1157
1158 if (hash_a > hash_b)
1159 ret = 1;
1160 else if (hash_a < hash_b)
1161 ret = -1;
1162 else {
1163 unsigned int len_a = be16_to_cpu(dent_a->de_name_len);
1164 unsigned int len_b = be16_to_cpu(dent_b->de_name_len);
1165
1166 if (len_a > len_b)
1167 ret = 1;
1168 else if (len_a < len_b)
1169 ret = -1;
1170 else
1171 ret = memcmp(dent_a + 1, dent_b + 1, len_a);
1172 }
1173
1174 return ret;
1175}
1176
1177/**
1178 * do_filldir_main - read out directory entries
1179 * @dip: The GFS2 inode
1180 * @offset: The offset in the file to read from
1181 * @opaque: opaque data to pass to filldir
1182 * @filldir: The function to pass entries to
1183 * @darr: an array of struct gfs2_dirent pointers to read
1184 * @entries: the number of entries in darr
1185 * @copied: pointer to int that's non-zero if a entry has been copied out
1186 *
1187 * Jump through some hoops to make sure that if there are hash collsions,
1188 * they are read out at the beginning of a buffer. We want to minimize
1189 * the possibility that they will fall into different readdir buffers or
1190 * that someone will want to seek to that location.
1191 *
1192 * Returns: errno, >0 on exception from filldir
1193 */
1194
1195static int do_filldir_main(struct gfs2_inode *dip, u64 *offset,
1196 void *opaque, gfs2_filldir_t filldir,
1197 const struct gfs2_dirent **darr, u32 entries,
1198 int *copied)
1199{
1200 const struct gfs2_dirent *dent, *dent_next;
1201 struct gfs2_inum inum;
1202 u64 off, off_next;
1203 unsigned int x, y;
1204 int run = 0;
1205 int error = 0;
1206
1207 sort(darr, entries, sizeof(struct gfs2_dirent *), compare_dents, NULL);
1208
1209 dent_next = darr[0];
1210 off_next = be32_to_cpu(dent_next->de_hash);
1211 off_next = gfs2_disk_hash2offset(off_next);
1212
1213 for (x = 0, y = 1; x < entries; x++, y++) {
1214 dent = dent_next;
1215 off = off_next;
1216
1217 if (y < entries) {
1218 dent_next = darr[y];
1219 off_next = be32_to_cpu(dent_next->de_hash);
1220 off_next = gfs2_disk_hash2offset(off_next);
1221
1222 if (off < *offset)
1223 continue;
1224 *offset = off;
1225
1226 if (off_next == off) {
1227 if (*copied && !run)
1228 return 1;
1229 run = 1;
1230 } else
1231 run = 0;
1232 } else {
1233 if (off < *offset)
1234 continue;
1235 *offset = off;
1236 }
1237
1238 gfs2_inum_in(&inum, (char *)&dent->de_inum);
1239
1240 error = filldir(opaque, (const char *)(dent + 1),
1241 be16_to_cpu(dent->de_name_len),
1242 off, &inum,
1243 be16_to_cpu(dent->de_type));
1244 if (error)
1245 return 1;
1246
1247 *copied = 1;
1248 }
1249
1250 /* Increment the *offset by one, so the next time we come into the
1251 do_filldir fxn, we get the next entry instead of the last one in the
1252 current leaf */
1253
1254 (*offset)++;
1255
1256 return 0;
1257}
1258
1259static int gfs2_dir_read_leaf(struct inode *inode, u64 *offset, void *opaque,
1260 gfs2_filldir_t filldir, int *copied,
1261 unsigned *depth, u64 leaf_no)
1262{
1263 struct gfs2_inode *ip = GFS2_I(inode);
1264 struct buffer_head *bh;
1265 struct gfs2_leaf *lf;
1266 unsigned entries = 0;
1267 unsigned leaves = 0;
1268 const struct gfs2_dirent **darr, *dent;
1269 struct dirent_gather g;
1270 struct buffer_head **larr;
1271 int leaf = 0;
1272 int error, i;
1273 u64 lfn = leaf_no;
1274
1275 do {
1276 error = get_leaf(ip, lfn, &bh);
1277 if (error)
1278 goto out;
1279 lf = (struct gfs2_leaf *)bh->b_data;
1280 if (leaves == 0)
1281 *depth = be16_to_cpu(lf->lf_depth);
1282 entries += be16_to_cpu(lf->lf_entries);
1283 leaves++;
1284 lfn = be64_to_cpu(lf->lf_next);
1285 brelse(bh);
1286 } while(lfn);
1287
1288 if (!entries)
1289 return 0;
1290
1291 error = -ENOMEM;
1292 larr = vmalloc((leaves + entries) * sizeof(void *));
1293 if (!larr)
1294 goto out;
1295 darr = (const struct gfs2_dirent **)(larr + leaves);
1296 g.pdent = darr;
1297 g.offset = 0;
1298 lfn = leaf_no;
1299
1300 do {
1301 error = get_leaf(ip, lfn, &bh);
1302 if (error)
1303 goto out_kfree;
1304 lf = (struct gfs2_leaf *)bh->b_data;
1305 lfn = be64_to_cpu(lf->lf_next);
1306 if (lf->lf_entries) {
1307 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
1308 gfs2_dirent_gather, NULL, &g);
1309 error = PTR_ERR(dent);
1310 if (IS_ERR(dent)) {
1311 goto out_kfree;
1312 }
1313 error = 0;
1314 larr[leaf++] = bh;
1315 } else {
1316 brelse(bh);
1317 }
1318 } while(lfn);
1319
1320 error = do_filldir_main(ip, offset, opaque, filldir, darr,
1321 entries, copied);
1322out_kfree:
1323 for(i = 0; i < leaf; i++)
1324 brelse(larr[i]);
1325 vfree(larr);
1326out:
1327 return error;
1328}
1329
1330/**
1331 * dir_e_read - Reads the entries from a directory into a filldir buffer
1332 * @dip: dinode pointer
1333 * @offset: the hash of the last entry read shifted to the right once
1334 * @opaque: buffer for the filldir function to fill
1335 * @filldir: points to the filldir function to use
1336 *
1337 * Returns: errno
1338 */
1339
1340static int dir_e_read(struct inode *inode, u64 *offset, void *opaque,
1341 gfs2_filldir_t filldir)
1342{
1343 struct gfs2_inode *dip = GFS2_I(inode);
1344 struct gfs2_sbd *sdp = GFS2_SB(inode);
1345 u32 hsize, len = 0;
1346 u32 ht_offset, lp_offset, ht_offset_cur = -1;
1347 u32 hash, index;
1348 u64 *lp;
1349 int copied = 0;
1350 int error = 0;
1351 unsigned depth = 0;
1352
1353 hsize = 1 << dip->i_di.di_depth;
1354 if (hsize * sizeof(u64) != dip->i_di.di_size) {
1355 gfs2_consist_inode(dip);
1356 return -EIO;
1357 }
1358
1359 hash = gfs2_dir_offset2hash(*offset);
1360 index = hash >> (32 - dip->i_di.di_depth);
1361
1362 lp = kmalloc(sdp->sd_hash_bsize, GFP_KERNEL);
1363 if (!lp)
1364 return -ENOMEM;
1365
1366 while (index < hsize) {
1367 lp_offset = index & (sdp->sd_hash_ptrs - 1);
1368 ht_offset = index - lp_offset;
1369
1370 if (ht_offset_cur != ht_offset) {
1371 error = gfs2_dir_read_data(dip, (char *)lp,
1372 ht_offset * sizeof(u64),
1373 sdp->sd_hash_bsize, 1);
1374 if (error != sdp->sd_hash_bsize) {
1375 if (error >= 0)
1376 error = -EIO;
1377 goto out;
1378 }
1379 ht_offset_cur = ht_offset;
1380 }
1381
1382 error = gfs2_dir_read_leaf(inode, offset, opaque, filldir,
1383 &copied, &depth,
1384 be64_to_cpu(lp[lp_offset]));
1385 if (error)
1386 break;
1387
1388 len = 1 << (dip->i_di.di_depth - depth);
1389 index = (index & ~(len - 1)) + len;
1390 }
1391
1392out:
1393 kfree(lp);
1394 if (error > 0)
1395 error = 0;
1396 return error;
1397}
1398
1399int gfs2_dir_read(struct inode *inode, u64 *offset, void *opaque,
1400 gfs2_filldir_t filldir)
1401{
1402 struct gfs2_inode *dip = GFS2_I(inode);
1403 struct dirent_gather g;
1404 const struct gfs2_dirent **darr, *dent;
1405 struct buffer_head *dibh;
1406 int copied = 0;
1407 int error;
1408
1409 if (!dip->i_di.di_entries)
1410 return 0;
1411
1412 if (dip->i_di.di_flags & GFS2_DIF_EXHASH)
1413 return dir_e_read(inode, offset, opaque, filldir);
1414
1415 if (!gfs2_is_stuffed(dip)) {
1416 gfs2_consist_inode(dip);
1417 return -EIO;
1418 }
1419
1420 error = gfs2_meta_inode_buffer(dip, &dibh);
1421 if (error)
1422 return error;
1423
1424 error = -ENOMEM;
1425 darr = kmalloc(dip->i_di.di_entries * sizeof(struct gfs2_dirent *),
1426 GFP_KERNEL);
1427 if (darr) {
1428 g.pdent = darr;
1429 g.offset = 0;
1430 dent = gfs2_dirent_scan(inode, dibh->b_data, dibh->b_size,
1431 gfs2_dirent_gather, NULL, &g);
1432 if (IS_ERR(dent)) {
1433 error = PTR_ERR(dent);
1434 goto out;
1435 }
1436 error = do_filldir_main(dip, offset, opaque, filldir, darr,
1437 dip->i_di.di_entries, &copied);
1438out:
1439 kfree(darr);
1440 }
1441
1442 if (error > 0)
1443 error = 0;
1444
1445 brelse(dibh);
1446
1447 return error;
1448}
1449
1450/**
1451 * gfs2_dir_search - Search a directory
1452 * @dip: The GFS2 inode
1453 * @filename:
1454 * @inode:
1455 *
1456 * This routine searches a directory for a file or another directory.
1457 * Assumes a glock is held on dip.
1458 *
1459 * Returns: errno
1460 */
1461
1462int gfs2_dir_search(struct inode *dir, const struct qstr *name,
1463 struct gfs2_inum *inum, unsigned int *type)
1464{
1465 struct buffer_head *bh;
1466 struct gfs2_dirent *dent;
1467
1468 dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
1469 if (dent) {
1470 if (IS_ERR(dent))
1471 return PTR_ERR(dent);
1472 if (inum)
1473 gfs2_inum_in(inum, (char *)&dent->de_inum);
1474 if (type)
1475 *type = be16_to_cpu(dent->de_type);
1476 brelse(bh);
1477 return 0;
1478 }
1479 return -ENOENT;
1480}
1481
1482static int dir_new_leaf(struct inode *inode, const struct qstr *name)
1483{
1484 struct buffer_head *bh, *obh;
1485 struct gfs2_inode *ip = GFS2_I(inode);
1486 struct gfs2_leaf *leaf, *oleaf;
1487 int error;
1488 u32 index;
1489 u64 bn;
1490
1491 index = name->hash >> (32 - ip->i_di.di_depth);
1492 error = get_first_leaf(ip, index, &obh);
1493 if (error)
1494 return error;
1495 do {
1496 oleaf = (struct gfs2_leaf *)obh->b_data;
1497 bn = be64_to_cpu(oleaf->lf_next);
1498 if (!bn)
1499 break;
1500 brelse(obh);
1501 error = get_leaf(ip, bn, &obh);
1502 if (error)
1503 return error;
1504 } while(1);
1505
1506 gfs2_trans_add_bh(ip->i_gl, obh, 1);
1507
1508 leaf = new_leaf(inode, &bh, be16_to_cpu(oleaf->lf_depth));
1509 if (!leaf) {
1510 brelse(obh);
1511 return -ENOSPC;
1512 }
1513 oleaf->lf_next = cpu_to_be64(bh->b_blocknr);
1514 brelse(bh);
1515 brelse(obh);
1516
1517 error = gfs2_meta_inode_buffer(ip, &bh);
1518 if (error)
1519 return error;
1520 gfs2_trans_add_bh(ip->i_gl, bh, 1);
1521 ip->i_di.di_blocks++;
1522 gfs2_dinode_out(&ip->i_di, bh->b_data);
1523 brelse(bh);
1524 return 0;
1525}
1526
1527/**
1528 * gfs2_dir_add - Add new filename into directory
1529 * @dip: The GFS2 inode
1530 * @filename: The new name
1531 * @inode: The inode number of the entry
1532 * @type: The type of the entry
1533 *
1534 * Returns: 0 on success, error code on failure
1535 */
1536
1537int gfs2_dir_add(struct inode *inode, const struct qstr *name,
1538 const struct gfs2_inum *inum, unsigned type)
1539{
1540 struct gfs2_inode *ip = GFS2_I(inode);
1541 struct buffer_head *bh;
1542 struct gfs2_dirent *dent;
1543 struct gfs2_leaf *leaf;
1544 int error;
1545
1546 while(1) {
1547 dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space,
1548 &bh);
1549 if (dent) {
1550 if (IS_ERR(dent))
1551 return PTR_ERR(dent);
1552 dent = gfs2_init_dirent(inode, dent, name, bh);
1553 gfs2_inum_out(inum, (char *)&dent->de_inum);
1554 dent->de_type = cpu_to_be16(type);
1555 if (ip->i_di.di_flags & GFS2_DIF_EXHASH) {
1556 leaf = (struct gfs2_leaf *)bh->b_data;
1557 leaf->lf_entries = cpu_to_be16(be16_to_cpu(leaf->lf_entries) + 1);
1558 }
1559 brelse(bh);
1560 error = gfs2_meta_inode_buffer(ip, &bh);
1561 if (error)
1562 break;
1563 gfs2_trans_add_bh(ip->i_gl, bh, 1);
1564 ip->i_di.di_entries++;
1565 ip->i_di.di_mtime = ip->i_di.di_ctime = get_seconds();
1566 gfs2_dinode_out(&ip->i_di, bh->b_data);
1567 brelse(bh);
1568 error = 0;
1569 break;
1570 }
1571 if (!(ip->i_di.di_flags & GFS2_DIF_EXHASH)) {
1572 error = dir_make_exhash(inode);
1573 if (error)
1574 break;
1575 continue;
1576 }
1577 error = dir_split_leaf(inode, name);
1578 if (error == 0)
1579 continue;
1580 if (error < 0)
1581 break;
1582 if (ip->i_di.di_depth < GFS2_DIR_MAX_DEPTH) {
1583 error = dir_double_exhash(ip);
1584 if (error)
1585 break;
1586 error = dir_split_leaf(inode, name);
1587 if (error < 0)
1588 break;
1589 if (error == 0)
1590 continue;
1591 }
1592 error = dir_new_leaf(inode, name);
1593 if (!error)
1594 continue;
1595 error = -ENOSPC;
1596 break;
1597 }
1598 return error;
1599}
1600
1601
1602/**
1603 * gfs2_dir_del - Delete a directory entry
1604 * @dip: The GFS2 inode
1605 * @filename: The filename
1606 *
1607 * Returns: 0 on success, error code on failure
1608 */
1609
1610int gfs2_dir_del(struct gfs2_inode *dip, const struct qstr *name)
1611{
1612 struct gfs2_dirent *dent, *prev = NULL;
1613 struct buffer_head *bh;
1614 int error;
1615
1616 /* Returns _either_ the entry (if its first in block) or the
1617 previous entry otherwise */
1618 dent = gfs2_dirent_search(&dip->i_inode, name, gfs2_dirent_prev, &bh);
1619 if (!dent) {
1620 gfs2_consist_inode(dip);
1621 return -EIO;
1622 }
1623 if (IS_ERR(dent)) {
1624 gfs2_consist_inode(dip);
1625 return PTR_ERR(dent);
1626 }
1627 /* If not first in block, adjust pointers accordingly */
1628 if (gfs2_dirent_find(dent, name, NULL) == 0) {
1629 prev = dent;
1630 dent = (struct gfs2_dirent *)((char *)dent + be16_to_cpu(prev->de_rec_len));
1631 }
1632
1633 dirent_del(dip, bh, prev, dent);
1634 if (dip->i_di.di_flags & GFS2_DIF_EXHASH) {
1635 struct gfs2_leaf *leaf = (struct gfs2_leaf *)bh->b_data;
1636 u16 entries = be16_to_cpu(leaf->lf_entries);
1637 if (!entries)
1638 gfs2_consist_inode(dip);
1639 leaf->lf_entries = cpu_to_be16(--entries);
1640 }
1641 brelse(bh);
1642
1643 error = gfs2_meta_inode_buffer(dip, &bh);
1644 if (error)
1645 return error;
1646
1647 if (!dip->i_di.di_entries)
1648 gfs2_consist_inode(dip);
1649 gfs2_trans_add_bh(dip->i_gl, bh, 1);
1650 dip->i_di.di_entries--;
1651 dip->i_di.di_mtime = dip->i_di.di_ctime = get_seconds();
1652 gfs2_dinode_out(&dip->i_di, bh->b_data);
1653 brelse(bh);
1654 mark_inode_dirty(&dip->i_inode);
1655
1656 return error;
1657}
1658
1659/**
1660 * gfs2_dir_mvino - Change inode number of directory entry
1661 * @dip: The GFS2 inode
1662 * @filename:
1663 * @new_inode:
1664 *
1665 * This routine changes the inode number of a directory entry. It's used
1666 * by rename to change ".." when a directory is moved.
1667 * Assumes a glock is held on dvp.
1668 *
1669 * Returns: errno
1670 */
1671
1672int gfs2_dir_mvino(struct gfs2_inode *dip, const struct qstr *filename,
1673 struct gfs2_inum *inum, unsigned int new_type)
1674{
1675 struct buffer_head *bh;
1676 struct gfs2_dirent *dent;
1677 int error;
1678
1679 dent = gfs2_dirent_search(&dip->i_inode, filename, gfs2_dirent_find, &bh);
1680 if (!dent) {
1681 gfs2_consist_inode(dip);
1682 return -EIO;
1683 }
1684 if (IS_ERR(dent))
1685 return PTR_ERR(dent);
1686
1687 gfs2_trans_add_bh(dip->i_gl, bh, 1);
1688 gfs2_inum_out(inum, (char *)&dent->de_inum);
1689 dent->de_type = cpu_to_be16(new_type);
1690
1691 if (dip->i_di.di_flags & GFS2_DIF_EXHASH) {
1692 brelse(bh);
1693 error = gfs2_meta_inode_buffer(dip, &bh);
1694 if (error)
1695 return error;
1696 gfs2_trans_add_bh(dip->i_gl, bh, 1);
1697 }
1698
1699 dip->i_di.di_mtime = dip->i_di.di_ctime = get_seconds();
1700 gfs2_dinode_out(&dip->i_di, bh->b_data);
1701 brelse(bh);
1702 return 0;
1703}
1704
1705/**
1706 * foreach_leaf - call a function for each leaf in a directory
1707 * @dip: the directory
1708 * @lc: the function to call for each each
1709 * @data: private data to pass to it
1710 *
1711 * Returns: errno
1712 */
1713
1714static int foreach_leaf(struct gfs2_inode *dip, leaf_call_t lc, void *data)
1715{
1716 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1717 struct buffer_head *bh;
1718 struct gfs2_leaf *leaf;
1719 u32 hsize, len;
1720 u32 ht_offset, lp_offset, ht_offset_cur = -1;
1721 u32 index = 0;
1722 u64 *lp;
1723 u64 leaf_no;
1724 int error = 0;
1725
1726 hsize = 1 << dip->i_di.di_depth;
1727 if (hsize * sizeof(u64) != dip->i_di.di_size) {
1728 gfs2_consist_inode(dip);
1729 return -EIO;
1730 }
1731
1732 lp = kmalloc(sdp->sd_hash_bsize, GFP_KERNEL);
1733 if (!lp)
1734 return -ENOMEM;
1735
1736 while (index < hsize) {
1737 lp_offset = index & (sdp->sd_hash_ptrs - 1);
1738 ht_offset = index - lp_offset;
1739
1740 if (ht_offset_cur != ht_offset) {
1741 error = gfs2_dir_read_data(dip, (char *)lp,
1742 ht_offset * sizeof(u64),
1743 sdp->sd_hash_bsize, 1);
1744 if (error != sdp->sd_hash_bsize) {
1745 if (error >= 0)
1746 error = -EIO;
1747 goto out;
1748 }
1749 ht_offset_cur = ht_offset;
1750 }
1751
1752 leaf_no = be64_to_cpu(lp[lp_offset]);
1753 if (leaf_no) {
1754 error = get_leaf(dip, leaf_no, &bh);
1755 if (error)
1756 goto out;
1757 leaf = (struct gfs2_leaf *)bh->b_data;
1758 len = 1 << (dip->i_di.di_depth - be16_to_cpu(leaf->lf_depth));
1759 brelse(bh);
1760
1761 error = lc(dip, index, len, leaf_no, data);
1762 if (error)
1763 goto out;
1764
1765 index = (index & ~(len - 1)) + len;
1766 } else
1767 index++;
1768 }
1769
1770 if (index != hsize) {
1771 gfs2_consist_inode(dip);
1772 error = -EIO;
1773 }
1774
1775out:
1776 kfree(lp);
1777
1778 return error;
1779}
1780
1781/**
1782 * leaf_dealloc - Deallocate a directory leaf
1783 * @dip: the directory
1784 * @index: the hash table offset in the directory
1785 * @len: the number of pointers to this leaf
1786 * @leaf_no: the leaf number
1787 * @data: not used
1788 *
1789 * Returns: errno
1790 */
1791
1792static int leaf_dealloc(struct gfs2_inode *dip, u32 index, u32 len,
1793 u64 leaf_no, void *data)
1794{
1795 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1796 struct gfs2_leaf *tmp_leaf;
1797 struct gfs2_rgrp_list rlist;
1798 struct buffer_head *bh, *dibh;
1799 u64 blk, nblk;
1800 unsigned int rg_blocks = 0, l_blocks = 0;
1801 char *ht;
1802 unsigned int x, size = len * sizeof(u64);
1803 int error;
1804
1805 memset(&rlist, 0, sizeof(struct gfs2_rgrp_list));
1806
1807 ht = kzalloc(size, GFP_KERNEL);
1808 if (!ht)
1809 return -ENOMEM;
1810
1811 gfs2_alloc_get(dip);
1812
1813 error = gfs2_quota_hold(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
1814 if (error)
1815 goto out;
1816
1817 error = gfs2_rindex_hold(sdp, &dip->i_alloc.al_ri_gh);
1818 if (error)
1819 goto out_qs;
1820
1821 /* Count the number of leaves */
1822
1823 for (blk = leaf_no; blk; blk = nblk) {
1824 error = get_leaf(dip, blk, &bh);
1825 if (error)
1826 goto out_rlist;
1827 tmp_leaf = (struct gfs2_leaf *)bh->b_data;
1828 nblk = be64_to_cpu(tmp_leaf->lf_next);
1829 brelse(bh);
1830
1831 gfs2_rlist_add(sdp, &rlist, blk);
1832 l_blocks++;
1833 }
1834
1835 gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE, 0);
1836
1837 for (x = 0; x < rlist.rl_rgrps; x++) {
1838 struct gfs2_rgrpd *rgd;
1839 rgd = rlist.rl_ghs[x].gh_gl->gl_object;
1840 rg_blocks += rgd->rd_ri.ri_length;
1841 }
1842
1843 error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs);
1844 if (error)
1845 goto out_rlist;
1846
1847 error = gfs2_trans_begin(sdp,
1848 rg_blocks + (DIV_ROUND_UP(size, sdp->sd_jbsize) + 1) +
1849 RES_DINODE + RES_STATFS + RES_QUOTA, l_blocks);
1850 if (error)
1851 goto out_rg_gunlock;
1852
1853 for (blk = leaf_no; blk; blk = nblk) {
1854 error = get_leaf(dip, blk, &bh);
1855 if (error)
1856 goto out_end_trans;
1857 tmp_leaf = (struct gfs2_leaf *)bh->b_data;
1858 nblk = be64_to_cpu(tmp_leaf->lf_next);
1859 brelse(bh);
1860
1861 gfs2_free_meta(dip, blk, 1);
1862
1863 if (!dip->i_di.di_blocks)
1864 gfs2_consist_inode(dip);
1865 dip->i_di.di_blocks--;
1866 }
1867
1868 error = gfs2_dir_write_data(dip, ht, index * sizeof(u64), size);
1869 if (error != size) {
1870 if (error >= 0)
1871 error = -EIO;
1872 goto out_end_trans;
1873 }
1874
1875 error = gfs2_meta_inode_buffer(dip, &dibh);
1876 if (error)
1877 goto out_end_trans;
1878
1879 gfs2_trans_add_bh(dip->i_gl, dibh, 1);
1880 gfs2_dinode_out(&dip->i_di, dibh->b_data);
1881 brelse(dibh);
1882
1883out_end_trans:
1884 gfs2_trans_end(sdp);
1885out_rg_gunlock:
1886 gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
1887out_rlist:
1888 gfs2_rlist_free(&rlist);
1889 gfs2_glock_dq_uninit(&dip->i_alloc.al_ri_gh);
1890out_qs:
1891 gfs2_quota_unhold(dip);
1892out:
1893 gfs2_alloc_put(dip);
1894 kfree(ht);
1895 return error;
1896}
1897
1898/**
1899 * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory
1900 * @dip: the directory
1901 *
1902 * Dealloc all on-disk directory leaves to FREEMETA state
1903 * Change on-disk inode type to "regular file"
1904 *
1905 * Returns: errno
1906 */
1907
1908int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip)
1909{
1910 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1911 struct buffer_head *bh;
1912 int error;
1913
1914 /* Dealloc on-disk leaves to FREEMETA state */
1915 error = foreach_leaf(dip, leaf_dealloc, NULL);
1916 if (error)
1917 return error;
1918
1919 /* Make this a regular file in case we crash.
1920 (We don't want to free these blocks a second time.) */
1921
1922 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1923 if (error)
1924 return error;
1925
1926 error = gfs2_meta_inode_buffer(dip, &bh);
1927 if (!error) {
1928 gfs2_trans_add_bh(dip->i_gl, bh, 1);
1929 ((struct gfs2_dinode *)bh->b_data)->di_mode =
1930 cpu_to_be32(S_IFREG);
1931 brelse(bh);
1932 }
1933
1934 gfs2_trans_end(sdp);
1935
1936 return error;
1937}
1938
1939/**
1940 * gfs2_diradd_alloc_required - find if adding entry will require an allocation
1941 * @ip: the file being written to
1942 * @filname: the filename that's going to be added
1943 *
1944 * Returns: 1 if alloc required, 0 if not, -ve on error
1945 */
1946
1947int gfs2_diradd_alloc_required(struct inode *inode, const struct qstr *name)
1948{
1949 struct gfs2_dirent *dent;
1950 struct buffer_head *bh;
1951
1952 dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, &bh);
1953 if (!dent) {
1954 return 1;
1955 }
1956 if (IS_ERR(dent))
1957 return PTR_ERR(dent);
1958 brelse(bh);
1959 return 0;
1960}
1961