[ARM] 5523/2: Updated ep93xx defconfig
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / gfs2 / dir.c
blobaef4d0c067488ccb942065aab8b0c292e8947213
1 /*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
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 */
11 * Implements Extendible Hashing as described in:
12 * "Extendible Hashing" by Fagin, et al in
13 * __ACM Trans. on Database Systems__, Sept 1979.
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.
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.
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
39 * dip->i_diskflags & GFS2_DIF_EXHASH is true
41 * Otherwise, the dirents are "linear", within a single stuffed dinode block.
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.
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.
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>
64 #include "gfs2.h"
65 #include "incore.h"
66 #include "dir.h"
67 #include "glock.h"
68 #include "inode.h"
69 #include "meta_io.h"
70 #include "quota.h"
71 #include "rgrp.h"
72 #include "trans.h"
73 #include "bmap.h"
74 #include "util.h"
76 #define IS_LEAF 1 /* Hashed (leaf) directory */
77 #define IS_DINODE 2 /* Linear (stuffed dinode block) directory */
79 #define gfs2_disk_hash2offset(h) (((u64)(h)) >> 1)
80 #define gfs2_dir_offset2hash(p) ((u32)(((u64)(p)) << 1))
82 typedef int (*leaf_call_t) (struct gfs2_inode *dip, u32 index, u32 len,
83 u64 leaf_no, void *data);
84 typedef int (*gfs2_dscan_t)(const struct gfs2_dirent *dent,
85 const struct qstr *name, void *opaque);
88 int gfs2_dir_get_new_buffer(struct gfs2_inode *ip, u64 block,
89 struct buffer_head **bhp)
91 struct buffer_head *bh;
93 bh = gfs2_meta_new(ip->i_gl, block);
94 gfs2_trans_add_bh(ip->i_gl, bh, 1);
95 gfs2_metatype_set(bh, GFS2_METATYPE_JD, GFS2_FORMAT_JD);
96 gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
97 *bhp = bh;
98 return 0;
101 static int gfs2_dir_get_existing_buffer(struct gfs2_inode *ip, u64 block,
102 struct buffer_head **bhp)
104 struct buffer_head *bh;
105 int error;
107 error = gfs2_meta_read(ip->i_gl, block, DIO_WAIT, &bh);
108 if (error)
109 return error;
110 if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), bh, GFS2_METATYPE_JD)) {
111 brelse(bh);
112 return -EIO;
114 *bhp = bh;
115 return 0;
118 static int gfs2_dir_write_stuffed(struct gfs2_inode *ip, const char *buf,
119 unsigned int offset, unsigned int size)
121 struct buffer_head *dibh;
122 int error;
124 error = gfs2_meta_inode_buffer(ip, &dibh);
125 if (error)
126 return error;
128 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
129 memcpy(dibh->b_data + offset + sizeof(struct gfs2_dinode), buf, size);
130 if (ip->i_disksize < offset + size)
131 ip->i_disksize = offset + size;
132 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
133 gfs2_dinode_out(ip, dibh->b_data);
135 brelse(dibh);
137 return size;
143 * gfs2_dir_write_data - Write directory information to the inode
144 * @ip: The GFS2 inode
145 * @buf: The buffer containing information to be written
146 * @offset: The file offset to start writing at
147 * @size: The amount of data to write
149 * Returns: The number of bytes correctly written or error code
151 static int gfs2_dir_write_data(struct gfs2_inode *ip, const char *buf,
152 u64 offset, unsigned int size)
154 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
155 struct buffer_head *dibh;
156 u64 lblock, dblock;
157 u32 extlen = 0;
158 unsigned int o;
159 int copied = 0;
160 int error = 0;
161 int new = 0;
163 if (!size)
164 return 0;
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);
171 if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
172 return -EINVAL;
174 if (gfs2_is_stuffed(ip)) {
175 error = gfs2_unstuff_dinode(ip, NULL);
176 if (error)
177 return error;
180 lblock = offset;
181 o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
183 while (copied < size) {
184 unsigned int amount;
185 struct buffer_head *bh;
187 amount = size - copied;
188 if (amount > sdp->sd_sb.sb_bsize - o)
189 amount = sdp->sd_sb.sb_bsize - o;
191 if (!extlen) {
192 new = 1;
193 error = gfs2_extent_map(&ip->i_inode, lblock, &new,
194 &dblock, &extlen);
195 if (error)
196 goto fail;
197 error = -EIO;
198 if (gfs2_assert_withdraw(sdp, dblock))
199 goto fail;
202 if (amount == sdp->sd_jbsize || new)
203 error = gfs2_dir_get_new_buffer(ip, dblock, &bh);
204 else
205 error = gfs2_dir_get_existing_buffer(ip, dblock, &bh);
207 if (error)
208 goto fail;
210 gfs2_trans_add_bh(ip->i_gl, bh, 1);
211 memcpy(bh->b_data + o, buf, amount);
212 brelse(bh);
214 buf += amount;
215 copied += amount;
216 lblock++;
217 dblock++;
218 extlen--;
220 o = sizeof(struct gfs2_meta_header);
223 out:
224 error = gfs2_meta_inode_buffer(ip, &dibh);
225 if (error)
226 return error;
228 if (ip->i_disksize < offset + copied)
229 ip->i_disksize = offset + copied;
230 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
232 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
233 gfs2_dinode_out(ip, dibh->b_data);
234 brelse(dibh);
236 return copied;
237 fail:
238 if (copied)
239 goto out;
240 return error;
243 static int gfs2_dir_read_stuffed(struct gfs2_inode *ip, char *buf,
244 u64 offset, unsigned int size)
246 struct buffer_head *dibh;
247 int error;
249 error = gfs2_meta_inode_buffer(ip, &dibh);
250 if (!error) {
251 offset += sizeof(struct gfs2_dinode);
252 memcpy(buf, dibh->b_data + offset, size);
253 brelse(dibh);
256 return (error) ? error : size;
261 * gfs2_dir_read_data - Read a data from a directory inode
262 * @ip: The GFS2 Inode
263 * @buf: The buffer to place result into
264 * @offset: File offset to begin jdata_readng from
265 * @size: Amount of data to transfer
267 * Returns: The amount of data actually copied or the error
269 static int gfs2_dir_read_data(struct gfs2_inode *ip, char *buf, u64 offset,
270 unsigned int size, unsigned ra)
272 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
273 u64 lblock, dblock;
274 u32 extlen = 0;
275 unsigned int o;
276 int copied = 0;
277 int error = 0;
279 if (offset >= ip->i_disksize)
280 return 0;
282 if (offset + size > ip->i_disksize)
283 size = ip->i_disksize - offset;
285 if (!size)
286 return 0;
288 if (gfs2_is_stuffed(ip))
289 return gfs2_dir_read_stuffed(ip, buf, offset, size);
291 if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
292 return -EINVAL;
294 lblock = offset;
295 o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
297 while (copied < size) {
298 unsigned int amount;
299 struct buffer_head *bh;
300 int new;
302 amount = size - copied;
303 if (amount > sdp->sd_sb.sb_bsize - o)
304 amount = sdp->sd_sb.sb_bsize - o;
306 if (!extlen) {
307 new = 0;
308 error = gfs2_extent_map(&ip->i_inode, lblock, &new,
309 &dblock, &extlen);
310 if (error || !dblock)
311 goto fail;
312 BUG_ON(extlen < 1);
313 if (!ra)
314 extlen = 1;
315 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
316 } else {
317 error = gfs2_meta_read(ip->i_gl, dblock, DIO_WAIT, &bh);
318 if (error)
319 goto fail;
321 error = gfs2_metatype_check(sdp, bh, GFS2_METATYPE_JD);
322 if (error) {
323 brelse(bh);
324 goto fail;
326 dblock++;
327 extlen--;
328 memcpy(buf, bh->b_data + o, amount);
329 brelse(bh);
330 buf += amount;
331 copied += amount;
332 lblock++;
333 o = sizeof(struct gfs2_meta_header);
336 return copied;
337 fail:
338 return (copied) ? copied : error;
341 static inline int gfs2_dirent_sentinel(const struct gfs2_dirent *dent)
343 return dent->de_inum.no_addr == 0 || dent->de_inum.no_formal_ino == 0;
346 static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent,
347 const struct qstr *name, int ret)
349 if (!gfs2_dirent_sentinel(dent) &&
350 be32_to_cpu(dent->de_hash) == name->hash &&
351 be16_to_cpu(dent->de_name_len) == name->len &&
352 memcmp(dent+1, name->name, name->len) == 0)
353 return ret;
354 return 0;
357 static int gfs2_dirent_find(const struct gfs2_dirent *dent,
358 const struct qstr *name,
359 void *opaque)
361 return __gfs2_dirent_find(dent, name, 1);
364 static int gfs2_dirent_prev(const struct gfs2_dirent *dent,
365 const struct qstr *name,
366 void *opaque)
368 return __gfs2_dirent_find(dent, name, 2);
372 * name->name holds ptr to start of block.
373 * name->len holds size of block.
375 static int gfs2_dirent_last(const struct gfs2_dirent *dent,
376 const struct qstr *name,
377 void *opaque)
379 const char *start = name->name;
380 const char *end = (const char *)dent + be16_to_cpu(dent->de_rec_len);
381 if (name->len == (end - start))
382 return 1;
383 return 0;
386 static int gfs2_dirent_find_space(const struct gfs2_dirent *dent,
387 const struct qstr *name,
388 void *opaque)
390 unsigned required = GFS2_DIRENT_SIZE(name->len);
391 unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
392 unsigned totlen = be16_to_cpu(dent->de_rec_len);
394 if (gfs2_dirent_sentinel(dent))
395 actual = GFS2_DIRENT_SIZE(0);
396 if (totlen - actual >= required)
397 return 1;
398 return 0;
401 struct dirent_gather {
402 const struct gfs2_dirent **pdent;
403 unsigned offset;
406 static int gfs2_dirent_gather(const struct gfs2_dirent *dent,
407 const struct qstr *name,
408 void *opaque)
410 struct dirent_gather *g = opaque;
411 if (!gfs2_dirent_sentinel(dent)) {
412 g->pdent[g->offset++] = dent;
414 return 0;
418 * Other possible things to check:
419 * - Inode located within filesystem size (and on valid block)
420 * - Valid directory entry type
421 * Not sure how heavy-weight we want to make this... could also check
422 * hash is correct for example, but that would take a lot of extra time.
423 * For now the most important thing is to check that the various sizes
424 * are correct.
426 static int gfs2_check_dirent(struct gfs2_dirent *dent, unsigned int offset,
427 unsigned int size, unsigned int len, int first)
429 const char *msg = "gfs2_dirent too small";
430 if (unlikely(size < sizeof(struct gfs2_dirent)))
431 goto error;
432 msg = "gfs2_dirent misaligned";
433 if (unlikely(offset & 0x7))
434 goto error;
435 msg = "gfs2_dirent points beyond end of block";
436 if (unlikely(offset + size > len))
437 goto error;
438 msg = "zero inode number";
439 if (unlikely(!first && gfs2_dirent_sentinel(dent)))
440 goto error;
441 msg = "name length is greater than space in dirent";
442 if (!gfs2_dirent_sentinel(dent) &&
443 unlikely(sizeof(struct gfs2_dirent)+be16_to_cpu(dent->de_name_len) >
444 size))
445 goto error;
446 return 0;
447 error:
448 printk(KERN_WARNING "gfs2_check_dirent: %s (%s)\n", msg,
449 first ? "first in block" : "not first in block");
450 return -EIO;
453 static int gfs2_dirent_offset(const void *buf)
455 const struct gfs2_meta_header *h = buf;
456 int offset;
458 BUG_ON(buf == NULL);
460 switch(be32_to_cpu(h->mh_type)) {
461 case GFS2_METATYPE_LF:
462 offset = sizeof(struct gfs2_leaf);
463 break;
464 case GFS2_METATYPE_DI:
465 offset = sizeof(struct gfs2_dinode);
466 break;
467 default:
468 goto wrong_type;
470 return offset;
471 wrong_type:
472 printk(KERN_WARNING "gfs2_scan_dirent: wrong block type %u\n",
473 be32_to_cpu(h->mh_type));
474 return -1;
477 static struct gfs2_dirent *gfs2_dirent_scan(struct inode *inode, void *buf,
478 unsigned int len, gfs2_dscan_t scan,
479 const struct qstr *name,
480 void *opaque)
482 struct gfs2_dirent *dent, *prev;
483 unsigned offset;
484 unsigned size;
485 int ret = 0;
487 ret = gfs2_dirent_offset(buf);
488 if (ret < 0)
489 goto consist_inode;
491 offset = ret;
492 prev = NULL;
493 dent = buf + offset;
494 size = be16_to_cpu(dent->de_rec_len);
495 if (gfs2_check_dirent(dent, offset, size, len, 1))
496 goto consist_inode;
497 do {
498 ret = scan(dent, name, opaque);
499 if (ret)
500 break;
501 offset += size;
502 if (offset == len)
503 break;
504 prev = dent;
505 dent = buf + offset;
506 size = be16_to_cpu(dent->de_rec_len);
507 if (gfs2_check_dirent(dent, offset, size, len, 0))
508 goto consist_inode;
509 } while(1);
511 switch(ret) {
512 case 0:
513 return NULL;
514 case 1:
515 return dent;
516 case 2:
517 return prev ? prev : dent;
518 default:
519 BUG_ON(ret > 0);
520 return ERR_PTR(ret);
523 consist_inode:
524 gfs2_consist_inode(GFS2_I(inode));
525 return ERR_PTR(-EIO);
530 * dirent_first - Return the first dirent
531 * @dip: the directory
532 * @bh: The buffer
533 * @dent: Pointer to list of dirents
535 * return first dirent whether bh points to leaf or stuffed dinode
537 * Returns: IS_LEAF, IS_DINODE, or -errno
540 static int dirent_first(struct gfs2_inode *dip, struct buffer_head *bh,
541 struct gfs2_dirent **dent)
543 struct gfs2_meta_header *h = (struct gfs2_meta_header *)bh->b_data;
545 if (be32_to_cpu(h->mh_type) == GFS2_METATYPE_LF) {
546 if (gfs2_meta_check(GFS2_SB(&dip->i_inode), bh))
547 return -EIO;
548 *dent = (struct gfs2_dirent *)(bh->b_data +
549 sizeof(struct gfs2_leaf));
550 return IS_LEAF;
551 } else {
552 if (gfs2_metatype_check(GFS2_SB(&dip->i_inode), bh, GFS2_METATYPE_DI))
553 return -EIO;
554 *dent = (struct gfs2_dirent *)(bh->b_data +
555 sizeof(struct gfs2_dinode));
556 return IS_DINODE;
560 static int dirent_check_reclen(struct gfs2_inode *dip,
561 const struct gfs2_dirent *d, const void *end_p)
563 const void *ptr = d;
564 u16 rec_len = be16_to_cpu(d->de_rec_len);
566 if (unlikely(rec_len < sizeof(struct gfs2_dirent)))
567 goto broken;
568 ptr += rec_len;
569 if (ptr < end_p)
570 return rec_len;
571 if (ptr == end_p)
572 return -ENOENT;
573 broken:
574 gfs2_consist_inode(dip);
575 return -EIO;
579 * dirent_next - Next dirent
580 * @dip: the directory
581 * @bh: The buffer
582 * @dent: Pointer to list of dirents
584 * Returns: 0 on success, error code otherwise
587 static int dirent_next(struct gfs2_inode *dip, struct buffer_head *bh,
588 struct gfs2_dirent **dent)
590 struct gfs2_dirent *cur = *dent, *tmp;
591 char *bh_end = bh->b_data + bh->b_size;
592 int ret;
594 ret = dirent_check_reclen(dip, cur, bh_end);
595 if (ret < 0)
596 return ret;
598 tmp = (void *)cur + ret;
599 ret = dirent_check_reclen(dip, tmp, bh_end);
600 if (ret == -EIO)
601 return ret;
603 /* Only the first dent could ever have de_inum.no_addr == 0 */
604 if (gfs2_dirent_sentinel(tmp)) {
605 gfs2_consist_inode(dip);
606 return -EIO;
609 *dent = tmp;
610 return 0;
614 * dirent_del - Delete a dirent
615 * @dip: The GFS2 inode
616 * @bh: The buffer
617 * @prev: The previous dirent
618 * @cur: The current dirent
622 static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh,
623 struct gfs2_dirent *prev, struct gfs2_dirent *cur)
625 u16 cur_rec_len, prev_rec_len;
627 if (gfs2_dirent_sentinel(cur)) {
628 gfs2_consist_inode(dip);
629 return;
632 gfs2_trans_add_bh(dip->i_gl, bh, 1);
634 /* If there is no prev entry, this is the first entry in the block.
635 The de_rec_len is already as big as it needs to be. Just zero
636 out the inode number and return. */
638 if (!prev) {
639 cur->de_inum.no_addr = 0;
640 cur->de_inum.no_formal_ino = 0;
641 return;
644 /* Combine this dentry with the previous one. */
646 prev_rec_len = be16_to_cpu(prev->de_rec_len);
647 cur_rec_len = be16_to_cpu(cur->de_rec_len);
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);
654 prev_rec_len += cur_rec_len;
655 prev->de_rec_len = cpu_to_be16(prev_rec_len);
659 * Takes a dent from which to grab space as an argument. Returns the
660 * newly created dent.
662 static struct gfs2_dirent *gfs2_init_dirent(struct inode *inode,
663 struct gfs2_dirent *dent,
664 const struct qstr *name,
665 struct buffer_head *bh)
667 struct gfs2_inode *ip = GFS2_I(inode);
668 struct gfs2_dirent *ndent;
669 unsigned offset = 0, totlen;
671 if (!gfs2_dirent_sentinel(dent))
672 offset = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
673 totlen = be16_to_cpu(dent->de_rec_len);
674 BUG_ON(offset + name->len > totlen);
675 gfs2_trans_add_bh(ip->i_gl, bh, 1);
676 ndent = (struct gfs2_dirent *)((char *)dent + offset);
677 dent->de_rec_len = cpu_to_be16(offset);
678 gfs2_qstr2dirent(name, totlen - offset, ndent);
679 return ndent;
682 static struct gfs2_dirent *gfs2_dirent_alloc(struct inode *inode,
683 struct buffer_head *bh,
684 const struct qstr *name)
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);
694 static int get_leaf(struct gfs2_inode *dip, u64 leaf_no,
695 struct buffer_head **bhp)
697 int error;
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;
705 return error;
709 * get_leaf_nr - Get a leaf number associated with the index
710 * @dip: The GFS2 inode
711 * @index:
712 * @leaf_out:
714 * Returns: 0 on success, error code otherwise
717 static int get_leaf_nr(struct gfs2_inode *dip, u32 index,
718 u64 *leaf_out)
720 __be64 leaf_no;
721 int error;
723 error = gfs2_dir_read_data(dip, (char *)&leaf_no,
724 index * sizeof(__be64),
725 sizeof(__be64), 0);
726 if (error != sizeof(u64))
727 return (error < 0) ? error : -EIO;
729 *leaf_out = be64_to_cpu(leaf_no);
731 return 0;
734 static int get_first_leaf(struct gfs2_inode *dip, u32 index,
735 struct buffer_head **bh_out)
737 u64 leaf_no;
738 int error;
740 error = get_leaf_nr(dip, index, &leaf_no);
741 if (!error)
742 error = get_leaf(dip, leaf_no, bh_out);
744 return error;
747 static struct gfs2_dirent *gfs2_dirent_search(struct inode *inode,
748 const struct qstr *name,
749 gfs2_dscan_t scan,
750 struct buffer_head **pbh)
752 struct buffer_head *bh;
753 struct gfs2_dirent *dent;
754 struct gfs2_inode *ip = GFS2_I(inode);
755 int error;
757 if (ip->i_diskflags & GFS2_DIF_EXHASH) {
758 struct gfs2_leaf *leaf;
759 unsigned hsize = 1 << ip->i_depth;
760 unsigned index;
761 u64 ln;
762 if (hsize * sizeof(u64) != ip->i_disksize) {
763 gfs2_consist_inode(ip);
764 return ERR_PTR(-EIO);
767 index = name->hash >> (32 - ip->i_depth);
768 error = get_first_leaf(ip, index, &bh);
769 if (error)
770 return ERR_PTR(error);
771 do {
772 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
773 scan, name, NULL);
774 if (dent)
775 goto got_dent;
776 leaf = (struct gfs2_leaf *)bh->b_data;
777 ln = be64_to_cpu(leaf->lf_next);
778 brelse(bh);
779 if (!ln)
780 break;
782 error = get_leaf(ip, ln, &bh);
783 } while(!error);
785 return error ? ERR_PTR(error) : NULL;
789 error = gfs2_meta_inode_buffer(ip, &bh);
790 if (error)
791 return ERR_PTR(error);
792 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, scan, name, NULL);
793 got_dent:
794 if (unlikely(dent == NULL || IS_ERR(dent))) {
795 brelse(bh);
796 bh = NULL;
798 *pbh = bh;
799 return dent;
802 static struct gfs2_leaf *new_leaf(struct inode *inode, struct buffer_head **pbh, u16 depth)
804 struct gfs2_inode *ip = GFS2_I(inode);
805 unsigned int n = 1;
806 u64 bn = gfs2_alloc_block(ip, &n);
807 struct buffer_head *bh = gfs2_meta_new(ip->i_gl, bn);
808 struct gfs2_leaf *leaf;
809 struct gfs2_dirent *dent;
810 struct qstr name = { .name = "", .len = 0, .hash = 0 };
811 if (!bh)
812 return NULL;
813 gfs2_trans_add_unrevoke(GFS2_SB(inode), bn, 1);
814 gfs2_trans_add_bh(ip->i_gl, bh, 1);
815 gfs2_metatype_set(bh, GFS2_METATYPE_LF, GFS2_FORMAT_LF);
816 leaf = (struct gfs2_leaf *)bh->b_data;
817 leaf->lf_depth = cpu_to_be16(depth);
818 leaf->lf_entries = 0;
819 leaf->lf_dirent_format = cpu_to_be32(GFS2_FORMAT_DE);
820 leaf->lf_next = 0;
821 memset(leaf->lf_reserved, 0, sizeof(leaf->lf_reserved));
822 dent = (struct gfs2_dirent *)(leaf+1);
823 gfs2_qstr2dirent(&name, bh->b_size - sizeof(struct gfs2_leaf), dent);
824 *pbh = bh;
825 return leaf;
829 * dir_make_exhash - Convert a stuffed directory into an ExHash directory
830 * @dip: The GFS2 inode
832 * Returns: 0 on success, error code otherwise
835 static int dir_make_exhash(struct inode *inode)
837 struct gfs2_inode *dip = GFS2_I(inode);
838 struct gfs2_sbd *sdp = GFS2_SB(inode);
839 struct gfs2_dirent *dent;
840 struct qstr args;
841 struct buffer_head *bh, *dibh;
842 struct gfs2_leaf *leaf;
843 int y;
844 u32 x;
845 __be64 *lp;
846 u64 bn;
847 int error;
849 error = gfs2_meta_inode_buffer(dip, &dibh);
850 if (error)
851 return error;
853 /* Turn over a new leaf */
855 leaf = new_leaf(inode, &bh, 0);
856 if (!leaf)
857 return -ENOSPC;
858 bn = bh->b_blocknr;
860 gfs2_assert(sdp, dip->i_entries < (1 << 16));
861 leaf->lf_entries = cpu_to_be16(dip->i_entries);
863 /* Copy dirents */
865 gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_leaf), dibh,
866 sizeof(struct gfs2_dinode));
868 /* Find last entry */
870 x = 0;
871 args.len = bh->b_size - sizeof(struct gfs2_dinode) +
872 sizeof(struct gfs2_leaf);
873 args.name = bh->b_data;
874 dent = gfs2_dirent_scan(&dip->i_inode, bh->b_data, bh->b_size,
875 gfs2_dirent_last, &args, NULL);
876 if (!dent) {
877 brelse(bh);
878 brelse(dibh);
879 return -EIO;
881 if (IS_ERR(dent)) {
882 brelse(bh);
883 brelse(dibh);
884 return PTR_ERR(dent);
887 /* Adjust the last dirent's record length
888 (Remember that dent still points to the last entry.) */
890 dent->de_rec_len = cpu_to_be16(be16_to_cpu(dent->de_rec_len) +
891 sizeof(struct gfs2_dinode) -
892 sizeof(struct gfs2_leaf));
894 brelse(bh);
896 /* We're done with the new leaf block, now setup the new
897 hash table. */
899 gfs2_trans_add_bh(dip->i_gl, dibh, 1);
900 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
902 lp = (__be64 *)(dibh->b_data + sizeof(struct gfs2_dinode));
904 for (x = sdp->sd_hash_ptrs; x--; lp++)
905 *lp = cpu_to_be64(bn);
907 dip->i_disksize = sdp->sd_sb.sb_bsize / 2;
908 gfs2_add_inode_blocks(&dip->i_inode, 1);
909 dip->i_diskflags |= GFS2_DIF_EXHASH;
911 for (x = sdp->sd_hash_ptrs, y = -1; x; x >>= 1, y++) ;
912 dip->i_depth = y;
914 gfs2_dinode_out(dip, dibh->b_data);
916 brelse(dibh);
918 return 0;
922 * dir_split_leaf - Split a leaf block into two
923 * @dip: The GFS2 inode
924 * @index:
925 * @leaf_no:
927 * Returns: 0 on success, error code on failure
930 static int dir_split_leaf(struct inode *inode, const struct qstr *name)
932 struct gfs2_inode *dip = GFS2_I(inode);
933 struct buffer_head *nbh, *obh, *dibh;
934 struct gfs2_leaf *nleaf, *oleaf;
935 struct gfs2_dirent *dent = NULL, *prev = NULL, *next = NULL, *new;
936 u32 start, len, half_len, divider;
937 u64 bn, leaf_no;
938 __be64 *lp;
939 u32 index;
940 int x, moved = 0;
941 int error;
943 index = name->hash >> (32 - dip->i_depth);
944 error = get_leaf_nr(dip, index, &leaf_no);
945 if (error)
946 return error;
948 /* Get the old leaf block */
949 error = get_leaf(dip, leaf_no, &obh);
950 if (error)
951 return error;
953 oleaf = (struct gfs2_leaf *)obh->b_data;
954 if (dip->i_depth == be16_to_cpu(oleaf->lf_depth)) {
955 brelse(obh);
956 return 1; /* can't split */
959 gfs2_trans_add_bh(dip->i_gl, obh, 1);
961 nleaf = new_leaf(inode, &nbh, be16_to_cpu(oleaf->lf_depth) + 1);
962 if (!nleaf) {
963 brelse(obh);
964 return -ENOSPC;
966 bn = nbh->b_blocknr;
968 /* Compute the start and len of leaf pointers in the hash table. */
969 len = 1 << (dip->i_depth - be16_to_cpu(oleaf->lf_depth));
970 half_len = len >> 1;
971 if (!half_len) {
972 printk(KERN_WARNING "i_depth %u lf_depth %u index %u\n", dip->i_depth, be16_to_cpu(oleaf->lf_depth), index);
973 gfs2_consist_inode(dip);
974 error = -EIO;
975 goto fail_brelse;
978 start = (index & ~(len - 1));
980 /* Change the pointers.
981 Don't bother distinguishing stuffed from non-stuffed.
982 This code is complicated enough already. */
983 lp = kmalloc(half_len * sizeof(__be64), GFP_NOFS | __GFP_NOFAIL);
984 /* Change the pointers */
985 for (x = 0; x < half_len; x++)
986 lp[x] = cpu_to_be64(bn);
988 error = gfs2_dir_write_data(dip, (char *)lp, start * sizeof(u64),
989 half_len * sizeof(u64));
990 if (error != half_len * sizeof(u64)) {
991 if (error >= 0)
992 error = -EIO;
993 goto fail_lpfree;
996 kfree(lp);
998 /* Compute the divider */
999 divider = (start + half_len) << (32 - dip->i_depth);
1001 /* Copy the entries */
1002 dirent_first(dip, obh, &dent);
1004 do {
1005 next = dent;
1006 if (dirent_next(dip, obh, &next))
1007 next = NULL;
1009 if (!gfs2_dirent_sentinel(dent) &&
1010 be32_to_cpu(dent->de_hash) < divider) {
1011 struct qstr str;
1012 str.name = (char*)(dent+1);
1013 str.len = be16_to_cpu(dent->de_name_len);
1014 str.hash = be32_to_cpu(dent->de_hash);
1015 new = gfs2_dirent_alloc(inode, nbh, &str);
1016 if (IS_ERR(new)) {
1017 error = PTR_ERR(new);
1018 break;
1021 new->de_inum = dent->de_inum; /* No endian worries */
1022 new->de_type = dent->de_type; /* No endian worries */
1023 be16_add_cpu(&nleaf->lf_entries, 1);
1025 dirent_del(dip, obh, prev, dent);
1027 if (!oleaf->lf_entries)
1028 gfs2_consist_inode(dip);
1029 be16_add_cpu(&oleaf->lf_entries, -1);
1031 if (!prev)
1032 prev = dent;
1034 moved = 1;
1035 } else {
1036 prev = dent;
1038 dent = next;
1039 } while (dent);
1041 oleaf->lf_depth = nleaf->lf_depth;
1043 error = gfs2_meta_inode_buffer(dip, &dibh);
1044 if (!gfs2_assert_withdraw(GFS2_SB(&dip->i_inode), !error)) {
1045 gfs2_trans_add_bh(dip->i_gl, dibh, 1);
1046 gfs2_add_inode_blocks(&dip->i_inode, 1);
1047 gfs2_dinode_out(dip, dibh->b_data);
1048 brelse(dibh);
1051 brelse(obh);
1052 brelse(nbh);
1054 return error;
1056 fail_lpfree:
1057 kfree(lp);
1059 fail_brelse:
1060 brelse(obh);
1061 brelse(nbh);
1062 return error;
1066 * dir_double_exhash - Double size of ExHash table
1067 * @dip: The GFS2 dinode
1069 * Returns: 0 on success, error code on failure
1072 static int dir_double_exhash(struct gfs2_inode *dip)
1074 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1075 struct buffer_head *dibh;
1076 u32 hsize;
1077 u64 *buf;
1078 u64 *from, *to;
1079 u64 block;
1080 int x;
1081 int error = 0;
1083 hsize = 1 << dip->i_depth;
1084 if (hsize * sizeof(u64) != dip->i_disksize) {
1085 gfs2_consist_inode(dip);
1086 return -EIO;
1089 /* Allocate both the "from" and "to" buffers in one big chunk */
1091 buf = kcalloc(3, sdp->sd_hash_bsize, GFP_NOFS | __GFP_NOFAIL);
1093 for (block = dip->i_disksize >> sdp->sd_hash_bsize_shift; block--;) {
1094 error = gfs2_dir_read_data(dip, (char *)buf,
1095 block * sdp->sd_hash_bsize,
1096 sdp->sd_hash_bsize, 1);
1097 if (error != sdp->sd_hash_bsize) {
1098 if (error >= 0)
1099 error = -EIO;
1100 goto fail;
1103 from = buf;
1104 to = (u64 *)((char *)buf + sdp->sd_hash_bsize);
1106 for (x = sdp->sd_hash_ptrs; x--; from++) {
1107 *to++ = *from; /* No endianess worries */
1108 *to++ = *from;
1111 error = gfs2_dir_write_data(dip,
1112 (char *)buf + sdp->sd_hash_bsize,
1113 block * sdp->sd_sb.sb_bsize,
1114 sdp->sd_sb.sb_bsize);
1115 if (error != sdp->sd_sb.sb_bsize) {
1116 if (error >= 0)
1117 error = -EIO;
1118 goto fail;
1122 kfree(buf);
1124 error = gfs2_meta_inode_buffer(dip, &dibh);
1125 if (!gfs2_assert_withdraw(sdp, !error)) {
1126 dip->i_depth++;
1127 gfs2_dinode_out(dip, dibh->b_data);
1128 brelse(dibh);
1131 return error;
1133 fail:
1134 kfree(buf);
1135 return error;
1139 * compare_dents - compare directory entries by hash value
1140 * @a: first dent
1141 * @b: second dent
1143 * When comparing the hash entries of @a to @b:
1144 * gt: returns 1
1145 * lt: returns -1
1146 * eq: returns 0
1149 static int compare_dents(const void *a, const void *b)
1151 const struct gfs2_dirent *dent_a, *dent_b;
1152 u32 hash_a, hash_b;
1153 int ret = 0;
1155 dent_a = *(const struct gfs2_dirent **)a;
1156 hash_a = be32_to_cpu(dent_a->de_hash);
1158 dent_b = *(const struct gfs2_dirent **)b;
1159 hash_b = be32_to_cpu(dent_b->de_hash);
1161 if (hash_a > hash_b)
1162 ret = 1;
1163 else if (hash_a < hash_b)
1164 ret = -1;
1165 else {
1166 unsigned int len_a = be16_to_cpu(dent_a->de_name_len);
1167 unsigned int len_b = be16_to_cpu(dent_b->de_name_len);
1169 if (len_a > len_b)
1170 ret = 1;
1171 else if (len_a < len_b)
1172 ret = -1;
1173 else
1174 ret = memcmp(dent_a + 1, dent_b + 1, len_a);
1177 return ret;
1181 * do_filldir_main - read out directory entries
1182 * @dip: The GFS2 inode
1183 * @offset: The offset in the file to read from
1184 * @opaque: opaque data to pass to filldir
1185 * @filldir: The function to pass entries to
1186 * @darr: an array of struct gfs2_dirent pointers to read
1187 * @entries: the number of entries in darr
1188 * @copied: pointer to int that's non-zero if a entry has been copied out
1190 * Jump through some hoops to make sure that if there are hash collsions,
1191 * they are read out at the beginning of a buffer. We want to minimize
1192 * the possibility that they will fall into different readdir buffers or
1193 * that someone will want to seek to that location.
1195 * Returns: errno, >0 on exception from filldir
1198 static int do_filldir_main(struct gfs2_inode *dip, u64 *offset,
1199 void *opaque, filldir_t filldir,
1200 const struct gfs2_dirent **darr, u32 entries,
1201 int *copied)
1203 const struct gfs2_dirent *dent, *dent_next;
1204 u64 off, off_next;
1205 unsigned int x, y;
1206 int run = 0;
1207 int error = 0;
1209 sort(darr, entries, sizeof(struct gfs2_dirent *), compare_dents, NULL);
1211 dent_next = darr[0];
1212 off_next = be32_to_cpu(dent_next->de_hash);
1213 off_next = gfs2_disk_hash2offset(off_next);
1215 for (x = 0, y = 1; x < entries; x++, y++) {
1216 dent = dent_next;
1217 off = off_next;
1219 if (y < entries) {
1220 dent_next = darr[y];
1221 off_next = be32_to_cpu(dent_next->de_hash);
1222 off_next = gfs2_disk_hash2offset(off_next);
1224 if (off < *offset)
1225 continue;
1226 *offset = off;
1228 if (off_next == off) {
1229 if (*copied && !run)
1230 return 1;
1231 run = 1;
1232 } else
1233 run = 0;
1234 } else {
1235 if (off < *offset)
1236 continue;
1237 *offset = off;
1240 error = filldir(opaque, (const char *)(dent + 1),
1241 be16_to_cpu(dent->de_name_len),
1242 off, be64_to_cpu(dent->de_inum.no_addr),
1243 be16_to_cpu(dent->de_type));
1244 if (error)
1245 return 1;
1247 *copied = 1;
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 */
1254 (*offset)++;
1256 return 0;
1259 static int gfs2_dir_read_leaf(struct inode *inode, u64 *offset, void *opaque,
1260 filldir_t filldir, int *copied, unsigned *depth,
1261 u64 leaf_no)
1263 struct gfs2_inode *ip = GFS2_I(inode);
1264 struct gfs2_sbd *sdp = GFS2_SB(inode);
1265 struct buffer_head *bh;
1266 struct gfs2_leaf *lf;
1267 unsigned entries = 0, entries2 = 0;
1268 unsigned leaves = 0;
1269 const struct gfs2_dirent **darr, *dent;
1270 struct dirent_gather g;
1271 struct buffer_head **larr;
1272 int leaf = 0;
1273 int error, i;
1274 u64 lfn = leaf_no;
1276 do {
1277 error = get_leaf(ip, lfn, &bh);
1278 if (error)
1279 goto out;
1280 lf = (struct gfs2_leaf *)bh->b_data;
1281 if (leaves == 0)
1282 *depth = be16_to_cpu(lf->lf_depth);
1283 entries += be16_to_cpu(lf->lf_entries);
1284 leaves++;
1285 lfn = be64_to_cpu(lf->lf_next);
1286 brelse(bh);
1287 } while(lfn);
1289 if (!entries)
1290 return 0;
1292 error = -ENOMEM;
1294 * The extra 99 entries are not normally used, but are a buffer
1295 * zone in case the number of entries in the leaf is corrupt.
1296 * 99 is the maximum number of entries that can fit in a single
1297 * leaf block.
1299 larr = vmalloc((leaves + entries + 99) * sizeof(void *));
1300 if (!larr)
1301 goto out;
1302 darr = (const struct gfs2_dirent **)(larr + leaves);
1303 g.pdent = darr;
1304 g.offset = 0;
1305 lfn = leaf_no;
1307 do {
1308 error = get_leaf(ip, lfn, &bh);
1309 if (error)
1310 goto out_kfree;
1311 lf = (struct gfs2_leaf *)bh->b_data;
1312 lfn = be64_to_cpu(lf->lf_next);
1313 if (lf->lf_entries) {
1314 entries2 += be16_to_cpu(lf->lf_entries);
1315 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
1316 gfs2_dirent_gather, NULL, &g);
1317 error = PTR_ERR(dent);
1318 if (IS_ERR(dent))
1319 goto out_kfree;
1320 if (entries2 != g.offset) {
1321 fs_warn(sdp, "Number of entries corrupt in dir "
1322 "leaf %llu, entries2 (%u) != "
1323 "g.offset (%u)\n",
1324 (unsigned long long)bh->b_blocknr,
1325 entries2, g.offset);
1327 error = -EIO;
1328 goto out_kfree;
1330 error = 0;
1331 larr[leaf++] = bh;
1332 } else {
1333 brelse(bh);
1335 } while(lfn);
1337 BUG_ON(entries2 != entries);
1338 error = do_filldir_main(ip, offset, opaque, filldir, darr,
1339 entries, copied);
1340 out_kfree:
1341 for(i = 0; i < leaf; i++)
1342 brelse(larr[i]);
1343 vfree(larr);
1344 out:
1345 return error;
1349 * dir_e_read - Reads the entries from a directory into a filldir buffer
1350 * @dip: dinode pointer
1351 * @offset: the hash of the last entry read shifted to the right once
1352 * @opaque: buffer for the filldir function to fill
1353 * @filldir: points to the filldir function to use
1355 * Returns: errno
1358 static int dir_e_read(struct inode *inode, u64 *offset, void *opaque,
1359 filldir_t filldir)
1361 struct gfs2_inode *dip = GFS2_I(inode);
1362 struct gfs2_sbd *sdp = GFS2_SB(inode);
1363 u32 hsize, len = 0;
1364 u32 ht_offset, lp_offset, ht_offset_cur = -1;
1365 u32 hash, index;
1366 __be64 *lp;
1367 int copied = 0;
1368 int error = 0;
1369 unsigned depth = 0;
1371 hsize = 1 << dip->i_depth;
1372 if (hsize * sizeof(u64) != dip->i_disksize) {
1373 gfs2_consist_inode(dip);
1374 return -EIO;
1377 hash = gfs2_dir_offset2hash(*offset);
1378 index = hash >> (32 - dip->i_depth);
1380 lp = kmalloc(sdp->sd_hash_bsize, GFP_NOFS);
1381 if (!lp)
1382 return -ENOMEM;
1384 while (index < hsize) {
1385 lp_offset = index & (sdp->sd_hash_ptrs - 1);
1386 ht_offset = index - lp_offset;
1388 if (ht_offset_cur != ht_offset) {
1389 error = gfs2_dir_read_data(dip, (char *)lp,
1390 ht_offset * sizeof(__be64),
1391 sdp->sd_hash_bsize, 1);
1392 if (error != sdp->sd_hash_bsize) {
1393 if (error >= 0)
1394 error = -EIO;
1395 goto out;
1397 ht_offset_cur = ht_offset;
1400 error = gfs2_dir_read_leaf(inode, offset, opaque, filldir,
1401 &copied, &depth,
1402 be64_to_cpu(lp[lp_offset]));
1403 if (error)
1404 break;
1406 len = 1 << (dip->i_depth - depth);
1407 index = (index & ~(len - 1)) + len;
1410 out:
1411 kfree(lp);
1412 if (error > 0)
1413 error = 0;
1414 return error;
1417 int gfs2_dir_read(struct inode *inode, u64 *offset, void *opaque,
1418 filldir_t filldir)
1420 struct gfs2_inode *dip = GFS2_I(inode);
1421 struct gfs2_sbd *sdp = GFS2_SB(inode);
1422 struct dirent_gather g;
1423 const struct gfs2_dirent **darr, *dent;
1424 struct buffer_head *dibh;
1425 int copied = 0;
1426 int error;
1428 if (!dip->i_entries)
1429 return 0;
1431 if (dip->i_diskflags & GFS2_DIF_EXHASH)
1432 return dir_e_read(inode, offset, opaque, filldir);
1434 if (!gfs2_is_stuffed(dip)) {
1435 gfs2_consist_inode(dip);
1436 return -EIO;
1439 error = gfs2_meta_inode_buffer(dip, &dibh);
1440 if (error)
1441 return error;
1443 error = -ENOMEM;
1444 /* 96 is max number of dirents which can be stuffed into an inode */
1445 darr = kmalloc(96 * sizeof(struct gfs2_dirent *), GFP_NOFS);
1446 if (darr) {
1447 g.pdent = darr;
1448 g.offset = 0;
1449 dent = gfs2_dirent_scan(inode, dibh->b_data, dibh->b_size,
1450 gfs2_dirent_gather, NULL, &g);
1451 if (IS_ERR(dent)) {
1452 error = PTR_ERR(dent);
1453 goto out;
1455 if (dip->i_entries != g.offset) {
1456 fs_warn(sdp, "Number of entries corrupt in dir %llu, "
1457 "ip->i_entries (%u) != g.offset (%u)\n",
1458 (unsigned long long)dip->i_no_addr,
1459 dip->i_entries,
1460 g.offset);
1461 error = -EIO;
1462 goto out;
1464 error = do_filldir_main(dip, offset, opaque, filldir, darr,
1465 dip->i_entries, &copied);
1466 out:
1467 kfree(darr);
1470 if (error > 0)
1471 error = 0;
1473 brelse(dibh);
1475 return error;
1479 * gfs2_dir_search - Search a directory
1480 * @dip: The GFS2 inode
1481 * @filename:
1482 * @inode:
1484 * This routine searches a directory for a file or another directory.
1485 * Assumes a glock is held on dip.
1487 * Returns: errno
1490 struct inode *gfs2_dir_search(struct inode *dir, const struct qstr *name)
1492 struct buffer_head *bh;
1493 struct gfs2_dirent *dent;
1494 struct inode *inode;
1496 dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
1497 if (dent) {
1498 if (IS_ERR(dent))
1499 return ERR_CAST(dent);
1500 inode = gfs2_inode_lookup(dir->i_sb,
1501 be16_to_cpu(dent->de_type),
1502 be64_to_cpu(dent->de_inum.no_addr),
1503 be64_to_cpu(dent->de_inum.no_formal_ino), 0);
1504 brelse(bh);
1505 return inode;
1507 return ERR_PTR(-ENOENT);
1510 int gfs2_dir_check(struct inode *dir, const struct qstr *name,
1511 const struct gfs2_inode *ip)
1513 struct buffer_head *bh;
1514 struct gfs2_dirent *dent;
1515 int ret = -ENOENT;
1517 dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
1518 if (dent) {
1519 if (IS_ERR(dent))
1520 return PTR_ERR(dent);
1521 if (ip) {
1522 if (be64_to_cpu(dent->de_inum.no_addr) != ip->i_no_addr)
1523 goto out;
1524 if (be64_to_cpu(dent->de_inum.no_formal_ino) !=
1525 ip->i_no_formal_ino)
1526 goto out;
1527 if (unlikely(IF2DT(ip->i_inode.i_mode) !=
1528 be16_to_cpu(dent->de_type))) {
1529 gfs2_consist_inode(GFS2_I(dir));
1530 ret = -EIO;
1531 goto out;
1534 ret = 0;
1535 out:
1536 brelse(bh);
1538 return ret;
1541 static int dir_new_leaf(struct inode *inode, const struct qstr *name)
1543 struct buffer_head *bh, *obh;
1544 struct gfs2_inode *ip = GFS2_I(inode);
1545 struct gfs2_leaf *leaf, *oleaf;
1546 int error;
1547 u32 index;
1548 u64 bn;
1550 index = name->hash >> (32 - ip->i_depth);
1551 error = get_first_leaf(ip, index, &obh);
1552 if (error)
1553 return error;
1554 do {
1555 oleaf = (struct gfs2_leaf *)obh->b_data;
1556 bn = be64_to_cpu(oleaf->lf_next);
1557 if (!bn)
1558 break;
1559 brelse(obh);
1560 error = get_leaf(ip, bn, &obh);
1561 if (error)
1562 return error;
1563 } while(1);
1565 gfs2_trans_add_bh(ip->i_gl, obh, 1);
1567 leaf = new_leaf(inode, &bh, be16_to_cpu(oleaf->lf_depth));
1568 if (!leaf) {
1569 brelse(obh);
1570 return -ENOSPC;
1572 oleaf->lf_next = cpu_to_be64(bh->b_blocknr);
1573 brelse(bh);
1574 brelse(obh);
1576 error = gfs2_meta_inode_buffer(ip, &bh);
1577 if (error)
1578 return error;
1579 gfs2_trans_add_bh(ip->i_gl, bh, 1);
1580 gfs2_add_inode_blocks(&ip->i_inode, 1);
1581 gfs2_dinode_out(ip, bh->b_data);
1582 brelse(bh);
1583 return 0;
1587 * gfs2_dir_add - Add new filename into directory
1588 * @dip: The GFS2 inode
1589 * @filename: The new name
1590 * @inode: The inode number of the entry
1591 * @type: The type of the entry
1593 * Returns: 0 on success, error code on failure
1596 int gfs2_dir_add(struct inode *inode, const struct qstr *name,
1597 const struct gfs2_inode *nip, unsigned type)
1599 struct gfs2_inode *ip = GFS2_I(inode);
1600 struct buffer_head *bh;
1601 struct gfs2_dirent *dent;
1602 struct gfs2_leaf *leaf;
1603 int error;
1605 while(1) {
1606 dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space,
1607 &bh);
1608 if (dent) {
1609 if (IS_ERR(dent))
1610 return PTR_ERR(dent);
1611 dent = gfs2_init_dirent(inode, dent, name, bh);
1612 gfs2_inum_out(nip, dent);
1613 dent->de_type = cpu_to_be16(type);
1614 if (ip->i_diskflags & GFS2_DIF_EXHASH) {
1615 leaf = (struct gfs2_leaf *)bh->b_data;
1616 be16_add_cpu(&leaf->lf_entries, 1);
1618 brelse(bh);
1619 error = gfs2_meta_inode_buffer(ip, &bh);
1620 if (error)
1621 break;
1622 gfs2_trans_add_bh(ip->i_gl, bh, 1);
1623 ip->i_entries++;
1624 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
1625 gfs2_dinode_out(ip, bh->b_data);
1626 brelse(bh);
1627 error = 0;
1628 break;
1630 if (!(ip->i_diskflags & GFS2_DIF_EXHASH)) {
1631 error = dir_make_exhash(inode);
1632 if (error)
1633 break;
1634 continue;
1636 error = dir_split_leaf(inode, name);
1637 if (error == 0)
1638 continue;
1639 if (error < 0)
1640 break;
1641 if (ip->i_depth < GFS2_DIR_MAX_DEPTH) {
1642 error = dir_double_exhash(ip);
1643 if (error)
1644 break;
1645 error = dir_split_leaf(inode, name);
1646 if (error < 0)
1647 break;
1648 if (error == 0)
1649 continue;
1651 error = dir_new_leaf(inode, name);
1652 if (!error)
1653 continue;
1654 error = -ENOSPC;
1655 break;
1657 return error;
1662 * gfs2_dir_del - Delete a directory entry
1663 * @dip: The GFS2 inode
1664 * @filename: The filename
1666 * Returns: 0 on success, error code on failure
1669 int gfs2_dir_del(struct gfs2_inode *dip, const struct qstr *name)
1671 struct gfs2_dirent *dent, *prev = NULL;
1672 struct buffer_head *bh;
1673 int error;
1675 /* Returns _either_ the entry (if its first in block) or the
1676 previous entry otherwise */
1677 dent = gfs2_dirent_search(&dip->i_inode, name, gfs2_dirent_prev, &bh);
1678 if (!dent) {
1679 gfs2_consist_inode(dip);
1680 return -EIO;
1682 if (IS_ERR(dent)) {
1683 gfs2_consist_inode(dip);
1684 return PTR_ERR(dent);
1686 /* If not first in block, adjust pointers accordingly */
1687 if (gfs2_dirent_find(dent, name, NULL) == 0) {
1688 prev = dent;
1689 dent = (struct gfs2_dirent *)((char *)dent + be16_to_cpu(prev->de_rec_len));
1692 dirent_del(dip, bh, prev, dent);
1693 if (dip->i_diskflags & GFS2_DIF_EXHASH) {
1694 struct gfs2_leaf *leaf = (struct gfs2_leaf *)bh->b_data;
1695 u16 entries = be16_to_cpu(leaf->lf_entries);
1696 if (!entries)
1697 gfs2_consist_inode(dip);
1698 leaf->lf_entries = cpu_to_be16(--entries);
1700 brelse(bh);
1702 error = gfs2_meta_inode_buffer(dip, &bh);
1703 if (error)
1704 return error;
1706 if (!dip->i_entries)
1707 gfs2_consist_inode(dip);
1708 gfs2_trans_add_bh(dip->i_gl, bh, 1);
1709 dip->i_entries--;
1710 dip->i_inode.i_mtime = dip->i_inode.i_ctime = CURRENT_TIME;
1711 gfs2_dinode_out(dip, bh->b_data);
1712 brelse(bh);
1713 mark_inode_dirty(&dip->i_inode);
1715 return error;
1719 * gfs2_dir_mvino - Change inode number of directory entry
1720 * @dip: The GFS2 inode
1721 * @filename:
1722 * @new_inode:
1724 * This routine changes the inode number of a directory entry. It's used
1725 * by rename to change ".." when a directory is moved.
1726 * Assumes a glock is held on dvp.
1728 * Returns: errno
1731 int gfs2_dir_mvino(struct gfs2_inode *dip, const struct qstr *filename,
1732 const struct gfs2_inode *nip, unsigned int new_type)
1734 struct buffer_head *bh;
1735 struct gfs2_dirent *dent;
1736 int error;
1738 dent = gfs2_dirent_search(&dip->i_inode, filename, gfs2_dirent_find, &bh);
1739 if (!dent) {
1740 gfs2_consist_inode(dip);
1741 return -EIO;
1743 if (IS_ERR(dent))
1744 return PTR_ERR(dent);
1746 gfs2_trans_add_bh(dip->i_gl, bh, 1);
1747 gfs2_inum_out(nip, dent);
1748 dent->de_type = cpu_to_be16(new_type);
1750 if (dip->i_diskflags & GFS2_DIF_EXHASH) {
1751 brelse(bh);
1752 error = gfs2_meta_inode_buffer(dip, &bh);
1753 if (error)
1754 return error;
1755 gfs2_trans_add_bh(dip->i_gl, bh, 1);
1758 dip->i_inode.i_mtime = dip->i_inode.i_ctime = CURRENT_TIME;
1759 gfs2_dinode_out(dip, bh->b_data);
1760 brelse(bh);
1761 return 0;
1765 * foreach_leaf - call a function for each leaf in a directory
1766 * @dip: the directory
1767 * @lc: the function to call for each each
1768 * @data: private data to pass to it
1770 * Returns: errno
1773 static int foreach_leaf(struct gfs2_inode *dip, leaf_call_t lc, void *data)
1775 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1776 struct buffer_head *bh;
1777 struct gfs2_leaf *leaf;
1778 u32 hsize, len;
1779 u32 ht_offset, lp_offset, ht_offset_cur = -1;
1780 u32 index = 0;
1781 __be64 *lp;
1782 u64 leaf_no;
1783 int error = 0;
1785 hsize = 1 << dip->i_depth;
1786 if (hsize * sizeof(u64) != dip->i_disksize) {
1787 gfs2_consist_inode(dip);
1788 return -EIO;
1791 lp = kmalloc(sdp->sd_hash_bsize, GFP_NOFS);
1792 if (!lp)
1793 return -ENOMEM;
1795 while (index < hsize) {
1796 lp_offset = index & (sdp->sd_hash_ptrs - 1);
1797 ht_offset = index - lp_offset;
1799 if (ht_offset_cur != ht_offset) {
1800 error = gfs2_dir_read_data(dip, (char *)lp,
1801 ht_offset * sizeof(__be64),
1802 sdp->sd_hash_bsize, 1);
1803 if (error != sdp->sd_hash_bsize) {
1804 if (error >= 0)
1805 error = -EIO;
1806 goto out;
1808 ht_offset_cur = ht_offset;
1811 leaf_no = be64_to_cpu(lp[lp_offset]);
1812 if (leaf_no) {
1813 error = get_leaf(dip, leaf_no, &bh);
1814 if (error)
1815 goto out;
1816 leaf = (struct gfs2_leaf *)bh->b_data;
1817 len = 1 << (dip->i_depth - be16_to_cpu(leaf->lf_depth));
1818 brelse(bh);
1820 error = lc(dip, index, len, leaf_no, data);
1821 if (error)
1822 goto out;
1824 index = (index & ~(len - 1)) + len;
1825 } else
1826 index++;
1829 if (index != hsize) {
1830 gfs2_consist_inode(dip);
1831 error = -EIO;
1834 out:
1835 kfree(lp);
1837 return error;
1841 * leaf_dealloc - Deallocate a directory leaf
1842 * @dip: the directory
1843 * @index: the hash table offset in the directory
1844 * @len: the number of pointers to this leaf
1845 * @leaf_no: the leaf number
1846 * @data: not used
1848 * Returns: errno
1851 static int leaf_dealloc(struct gfs2_inode *dip, u32 index, u32 len,
1852 u64 leaf_no, void *data)
1854 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1855 struct gfs2_leaf *tmp_leaf;
1856 struct gfs2_rgrp_list rlist;
1857 struct buffer_head *bh, *dibh;
1858 u64 blk, nblk;
1859 unsigned int rg_blocks = 0, l_blocks = 0;
1860 char *ht;
1861 unsigned int x, size = len * sizeof(u64);
1862 int error;
1864 memset(&rlist, 0, sizeof(struct gfs2_rgrp_list));
1866 ht = kzalloc(size, GFP_NOFS);
1867 if (!ht)
1868 return -ENOMEM;
1870 if (!gfs2_alloc_get(dip)) {
1871 error = -ENOMEM;
1872 goto out;
1875 error = gfs2_quota_hold(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
1876 if (error)
1877 goto out_put;
1879 error = gfs2_rindex_hold(sdp, &dip->i_alloc->al_ri_gh);
1880 if (error)
1881 goto out_qs;
1883 /* Count the number of leaves */
1885 for (blk = leaf_no; blk; blk = nblk) {
1886 error = get_leaf(dip, blk, &bh);
1887 if (error)
1888 goto out_rlist;
1889 tmp_leaf = (struct gfs2_leaf *)bh->b_data;
1890 nblk = be64_to_cpu(tmp_leaf->lf_next);
1891 brelse(bh);
1893 gfs2_rlist_add(sdp, &rlist, blk);
1894 l_blocks++;
1897 gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE);
1899 for (x = 0; x < rlist.rl_rgrps; x++) {
1900 struct gfs2_rgrpd *rgd;
1901 rgd = rlist.rl_ghs[x].gh_gl->gl_object;
1902 rg_blocks += rgd->rd_length;
1905 error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs);
1906 if (error)
1907 goto out_rlist;
1909 error = gfs2_trans_begin(sdp,
1910 rg_blocks + (DIV_ROUND_UP(size, sdp->sd_jbsize) + 1) +
1911 RES_DINODE + RES_STATFS + RES_QUOTA, l_blocks);
1912 if (error)
1913 goto out_rg_gunlock;
1915 for (blk = leaf_no; blk; blk = nblk) {
1916 error = get_leaf(dip, blk, &bh);
1917 if (error)
1918 goto out_end_trans;
1919 tmp_leaf = (struct gfs2_leaf *)bh->b_data;
1920 nblk = be64_to_cpu(tmp_leaf->lf_next);
1921 brelse(bh);
1923 gfs2_free_meta(dip, blk, 1);
1924 gfs2_add_inode_blocks(&dip->i_inode, -1);
1927 error = gfs2_dir_write_data(dip, ht, index * sizeof(u64), size);
1928 if (error != size) {
1929 if (error >= 0)
1930 error = -EIO;
1931 goto out_end_trans;
1934 error = gfs2_meta_inode_buffer(dip, &dibh);
1935 if (error)
1936 goto out_end_trans;
1938 gfs2_trans_add_bh(dip->i_gl, dibh, 1);
1939 gfs2_dinode_out(dip, dibh->b_data);
1940 brelse(dibh);
1942 out_end_trans:
1943 gfs2_trans_end(sdp);
1944 out_rg_gunlock:
1945 gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
1946 out_rlist:
1947 gfs2_rlist_free(&rlist);
1948 gfs2_glock_dq_uninit(&dip->i_alloc->al_ri_gh);
1949 out_qs:
1950 gfs2_quota_unhold(dip);
1951 out_put:
1952 gfs2_alloc_put(dip);
1953 out:
1954 kfree(ht);
1955 return error;
1959 * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory
1960 * @dip: the directory
1962 * Dealloc all on-disk directory leaves to FREEMETA state
1963 * Change on-disk inode type to "regular file"
1965 * Returns: errno
1968 int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip)
1970 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1971 struct buffer_head *bh;
1972 int error;
1974 /* Dealloc on-disk leaves to FREEMETA state */
1975 error = foreach_leaf(dip, leaf_dealloc, NULL);
1976 if (error)
1977 return error;
1979 /* Make this a regular file in case we crash.
1980 (We don't want to free these blocks a second time.) */
1982 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1983 if (error)
1984 return error;
1986 error = gfs2_meta_inode_buffer(dip, &bh);
1987 if (!error) {
1988 gfs2_trans_add_bh(dip->i_gl, bh, 1);
1989 ((struct gfs2_dinode *)bh->b_data)->di_mode =
1990 cpu_to_be32(S_IFREG);
1991 brelse(bh);
1994 gfs2_trans_end(sdp);
1996 return error;
2000 * gfs2_diradd_alloc_required - find if adding entry will require an allocation
2001 * @ip: the file being written to
2002 * @filname: the filename that's going to be added
2004 * Returns: 1 if alloc required, 0 if not, -ve on error
2007 int gfs2_diradd_alloc_required(struct inode *inode, const struct qstr *name)
2009 struct gfs2_dirent *dent;
2010 struct buffer_head *bh;
2012 dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, &bh);
2013 if (!dent) {
2014 return 1;
2016 if (IS_ERR(dent))
2017 return PTR_ERR(dent);
2018 brelse(bh);
2019 return 0;