?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Committer: Michael Beasley <mike@snafu.setup>
[mikesnafu-overlay.git] / fs / gfs2 / dir.c
blobc34709512b19053088b79bab935a8a93d2374ae4
1 /*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
4 *
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
8 */
9
10 /*
11 * Implements Extendible Hashing as described in:
12 * "Extendible Hashing" by Fagin, et al in
13 * __ACM Trans. on Database Systems__, Sept 1979.
14 *
15 *
16 * Here's the layout of dirents which is essentially the same as that of ext2
17 * within a single block. The field de_name_len is the number of bytes
18 * actually required for the name (no null terminator). The field de_rec_len
19 * is the number of bytes allocated to the dirent. The offset of the next
20 * dirent in the block is (dirent + dirent->de_rec_len). When a dirent is
21 * deleted, the preceding dirent inherits its allocated space, ie
22 * prev->de_rec_len += deleted->de_rec_len. Since the next dirent is obtained
23 * by adding de_rec_len to the current dirent, this essentially causes the
24 * deleted dirent to get jumped over when iterating through all the dirents.
25 *
26 * When deleting the first dirent in a block, there is no previous dirent so
27 * the field de_ino is set to zero to designate it as deleted. When allocating
28 * a dirent, gfs2_dirent_alloc iterates through the dirents in a block. If the
29 * first dirent has (de_ino == 0) and de_rec_len is large enough, this first
30 * dirent is allocated. Otherwise it must go through all the 'used' dirents
31 * searching for one in which the amount of total space minus the amount of
32 * used space will provide enough space for the new dirent.
33 *
34 * There are two types of blocks in which dirents reside. In a stuffed dinode,
35 * the dirents begin at offset sizeof(struct gfs2_dinode) from the beginning of
36 * the block. In leaves, they begin at offset sizeof(struct gfs2_leaf) from the
37 * beginning of the leaf block. The dirents reside in leaves when
38 *
39 * dip->i_di.di_flags & GFS2_DIF_EXHASH is true
40 *
41 * Otherwise, the dirents are "linear", within a single stuffed dinode block.
42 *
43 * When the dirents are in leaves, the actual contents of the directory file are
44 * used as an array of 64-bit block pointers pointing to the leaf blocks. The
45 * dirents are NOT in the directory file itself. There can be more than one
46 * block pointer in the array that points to the same leaf. In fact, when a
47 * directory is first converted from linear to exhash, all of the pointers
48 * point to the same leaf.
49 *
50 * When a leaf is completely full, the size of the hash table can be
51 * doubled unless it is already at the maximum size which is hard coded into
52 * GFS2_DIR_MAX_DEPTH. After that, leaves are chained together in a linked list,
53 * but never before the maximum hash table size has been reached.
54 */
55
56 #include <linux/slab.h>
57 #include <linux/spinlock.h>
58 #include <linux/buffer_head.h>
59 #include <linux/sort.h>
60 #include <linux/gfs2_ondisk.h>
61 #include <linux/crc32.h>
62 #include <linux/vmalloc.h>
63 #include <linux/lm_interface.h>
64
65 #include "gfs2.h"
66 #include "incore.h"
67 #include "dir.h"
68 #include "glock.h"
69 #include "inode.h"
70 #include "meta_io.h"
71 #include "quota.h"
72 #include "rgrp.h"
73 #include "trans.h"
74 #include "bmap.h"
75 #include "util.h"
76
77 #define IS_LEAF 1 /* Hashed (leaf) directory */
78 #define IS_DINODE 2 /* Linear (stuffed dinode block) directory */
79
80 #define gfs2_disk_hash2offset(h) (((u64)(h)) >> 1)
81 #define gfs2_dir_offset2hash(p) ((u32)(((u64)(p)) << 1))
82
83 typedef int (*leaf_call_t) (struct gfs2_inode *dip, u32 index, u32 len,
84 u64 leaf_no, void *data);
85 typedef int (*gfs2_dscan_t)(const struct gfs2_dirent *dent,
86 const struct qstr *name, void *opaque);
87
88
89 int gfs2_dir_get_new_buffer(struct gfs2_inode *ip, u64 block,
90 struct buffer_head **bhp)
91 {
92 struct buffer_head *bh;
93
94 bh = gfs2_meta_new(ip->i_gl, block);
95 gfs2_trans_add_bh(ip->i_gl, bh, 1);
96 gfs2_metatype_set(bh, GFS2_METATYPE_JD, GFS2_FORMAT_JD);
97 gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
98 *bhp = bh;
99 return 0;
100 }
101
102 static int gfs2_dir_get_existing_buffer(struct gfs2_inode *ip, u64 block,
103 struct buffer_head **bhp)
104 {
105 struct buffer_head *bh;
106 int error;
107
108 error = gfs2_meta_read(ip->i_gl, block, DIO_WAIT, &bh);
109 if (error)
110 return error;
111 if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), bh, GFS2_METATYPE_JD)) {
112 brelse(bh);
113 return -EIO;
114 }
115 *bhp = bh;
116 return 0;
117 }
118
119 static int gfs2_dir_write_stuffed(struct gfs2_inode *ip, const char *buf,
120 unsigned int offset, unsigned int size)
121 {
122 struct buffer_head *dibh;
123 int error;
124
125 error = gfs2_meta_inode_buffer(ip, &dibh);
126 if (error)
127 return error;
128
129 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
130 memcpy(dibh->b_data + offset + sizeof(struct gfs2_dinode), buf, size);
131 if (ip->i_di.di_size < offset + size)
132 ip->i_di.di_size = offset + size;
133 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
134 gfs2_dinode_out(ip, dibh->b_data);
135
136 brelse(dibh);
137
138 return size;
139 }
140
141
142
143 /**
144 * gfs2_dir_write_data - Write directory information to the inode
145 * @ip: The GFS2 inode
146 * @buf: The buffer containing information to be written
147 * @offset: The file offset to start writing at
148 * @size: The amount of data to write
149 *
150 * Returns: The number of bytes correctly written or error code
151 */
152 static int gfs2_dir_write_data(struct gfs2_inode *ip, const char *buf,
153 u64 offset, unsigned int size)
154 {
155 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
156 struct buffer_head *dibh;
157 u64 lblock, dblock;
158 u32 extlen = 0;
159 unsigned int o;
160 int copied = 0;
161 int error = 0;
162
163 if (!size)
164 return 0;
165
166 if (gfs2_is_stuffed(ip) &&
167 offset + size <= sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode))
168 return gfs2_dir_write_stuffed(ip, buf, (unsigned int)offset,
169 size);
170
171 if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
172 return -EINVAL;
173
174 if (gfs2_is_stuffed(ip)) {
175 error = gfs2_unstuff_dinode(ip, NULL);
176 if (error)
177 return error;
178 }
179
180 lblock = offset;
181 o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
182
183 while (copied < size) {
184 unsigned int amount;
185 struct buffer_head *bh;
186 int new = 0;
187
188 amount = size - copied;
189 if (amount > sdp->sd_sb.sb_bsize - o)
190 amount = sdp->sd_sb.sb_bsize - o;
191
192 if (!extlen) {
193 new = 1;
194 error = gfs2_extent_map(&ip->i_inode, lblock, &new,
195 &dblock, &extlen);
196 if (error)
197 goto fail;
198 error = -EIO;
199 if (gfs2_assert_withdraw(sdp, dblock))
200 goto fail;
201 }
202
203 if (amount == sdp->sd_jbsize || new)
204 error = gfs2_dir_get_new_buffer(ip, dblock, &bh);
205 else
206 error = gfs2_dir_get_existing_buffer(ip, dblock, &bh);
207
208 if (error)
209 goto fail;
210
211 gfs2_trans_add_bh(ip->i_gl, bh, 1);
212 memcpy(bh->b_data + o, buf, amount);
213 brelse(bh);
214
215 buf += amount;
216 copied += amount;
217 lblock++;
218 dblock++;
219 extlen--;
220
221 o = sizeof(struct gfs2_meta_header);
222 }
223
224 out:
225 error = gfs2_meta_inode_buffer(ip, &dibh);
226 if (error)
227 return error;
228
229 if (ip->i_di.di_size < offset + copied)
230 ip->i_di.di_size = offset + copied;
231 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
232
233 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
234 gfs2_dinode_out(ip, dibh->b_data);
235 brelse(dibh);
236
237 return copied;
238 fail:
239 if (copied)
240 goto out;
241 return error;
242 }
243
244 static int gfs2_dir_read_stuffed(struct gfs2_inode *ip, char *buf,
245 u64 offset, unsigned int size)
246 {
247 struct buffer_head *dibh;
248 int error;
249
250 error = gfs2_meta_inode_buffer(ip, &dibh);
251 if (!error) {
252 offset += sizeof(struct gfs2_dinode);
253 memcpy(buf, dibh->b_data + offset, size);
254 brelse(dibh);
255 }
256
257 return (error) ? error : size;
258 }
259
260
261 /**
262 * gfs2_dir_read_data - Read a data from a directory inode
263 * @ip: The GFS2 Inode
264 * @buf: The buffer to place result into
265 * @offset: File offset to begin jdata_readng from
266 * @size: Amount of data to transfer
267 *
268 * Returns: The amount of data actually copied or the error
269 */
270 static int gfs2_dir_read_data(struct gfs2_inode *ip, char *buf, u64 offset,
271 unsigned int size, unsigned ra)
272 {
273 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
274 u64 lblock, dblock;
275 u32 extlen = 0;
276 unsigned int o;
277 int copied = 0;
278 int error = 0;
279
280 if (offset >= ip->i_di.di_size)
281 return 0;
282
283 if (offset + size > ip->i_di.di_size)
284 size = ip->i_di.di_size - offset;
285
286 if (!size)
287 return 0;
288
289 if (gfs2_is_stuffed(ip))
290 return gfs2_dir_read_stuffed(ip, buf, offset, size);
291
292 if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
293 return -EINVAL;
294
295 lblock = offset;
296 o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
297
298 while (copied < size) {
299 unsigned int amount;
300 struct buffer_head *bh;
301 int new;
302
303 amount = size - copied;
304 if (amount > sdp->sd_sb.sb_bsize - o)
305 amount = sdp->sd_sb.sb_bsize - o;
306
307 if (!extlen) {
308 new = 0;
309 error = gfs2_extent_map(&ip->i_inode, lblock, &new,
310 &dblock, &extlen);
311 if (error || !dblock)
312 goto fail;
313 BUG_ON(extlen < 1);
314 if (!ra)
315 extlen = 1;
316 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
317 } else {
318 error = gfs2_meta_read(ip->i_gl, dblock, DIO_WAIT, &bh);
319 if (error)
320 goto fail;
321 }
322 error = gfs2_metatype_check(sdp, bh, GFS2_METATYPE_JD);
323 if (error) {
324 brelse(bh);
325 goto fail;
326 }
327 dblock++;
328 extlen--;
329 memcpy(buf, bh->b_data + o, amount);
330 brelse(bh);
331 buf += amount;
332 copied += amount;
333 lblock++;
334 o = sizeof(struct gfs2_meta_header);
335 }
336
337 return copied;
338 fail:
339 return (copied) ? copied : error;
340 }
341
342 static inline int gfs2_dirent_sentinel(const struct gfs2_dirent *dent)
343 {
344 return dent->de_inum.no_addr == 0 || dent->de_inum.no_formal_ino == 0;
345 }
346
347 static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent,
348 const struct qstr *name, int ret)
349 {
350 if (!gfs2_dirent_sentinel(dent) &&
351 be32_to_cpu(dent->de_hash) == name->hash &&
352 be16_to_cpu(dent->de_name_len) == name->len &&
353 memcmp(dent+1, name->name, name->len) == 0)
354 return ret;
355 return 0;
356 }
357
358 static int gfs2_dirent_find(const struct gfs2_dirent *dent,
359 const struct qstr *name,
360 void *opaque)
361 {
362 return __gfs2_dirent_find(dent, name, 1);
363 }
364
365 static int gfs2_dirent_prev(const struct gfs2_dirent *dent,
366 const struct qstr *name,
367 void *opaque)
368 {
369 return __gfs2_dirent_find(dent, name, 2);
370 }
371
372 /*
373 * name->name holds ptr to start of block.
374 * name->len holds size of block.
375 */
376 static int gfs2_dirent_last(const struct gfs2_dirent *dent,
377 const struct qstr *name,
378 void *opaque)
379 {
380 const char *start = name->name;
381 const char *end = (const char *)dent + be16_to_cpu(dent->de_rec_len);
382 if (name->len == (end - start))
383 return 1;
384 return 0;
385 }
386
387 static int gfs2_dirent_find_space(const struct gfs2_dirent *dent,
388 const struct qstr *name,
389 void *opaque)
390 {
391 unsigned required = GFS2_DIRENT_SIZE(name->len);
392 unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
393 unsigned totlen = be16_to_cpu(dent->de_rec_len);
394
395 if (gfs2_dirent_sentinel(dent))
396 actual = GFS2_DIRENT_SIZE(0);
397 if (totlen - actual >= required)
398 return 1;
399 return 0;
400 }
401
402 struct dirent_gather {
403 const struct gfs2_dirent **pdent;
404 unsigned offset;
405 };
406
407 static int gfs2_dirent_gather(const struct gfs2_dirent *dent,
408 const struct qstr *name,
409 void *opaque)
410 {
411 struct dirent_gather *g = opaque;
412 if (!gfs2_dirent_sentinel(dent)) {
413 g->pdent[g->offset++] = dent;
414 }
415 return 0;
416 }
417
418 /*
419 * Other possible things to check:
420 * - Inode located within filesystem size (and on valid block)
421 * - Valid directory entry type
422 * Not sure how heavy-weight we want to make this... could also check
423 * hash is correct for example, but that would take a lot of extra time.
424 * For now the most important thing is to check that the various sizes
425 * are correct.
426 */
427 static int gfs2_check_dirent(struct gfs2_dirent *dent, unsigned int offset,
428 unsigned int size, unsigned int len, int first)
429 {
430 const char *msg = "gfs2_dirent too small";
431 if (unlikely(size < sizeof(struct gfs2_dirent)))
432 goto error;
433 msg = "gfs2_dirent misaligned";
434 if (unlikely(offset & 0x7))
435 goto error;
436 msg = "gfs2_dirent points beyond end of block";
437 if (unlikely(offset + size > len))
438 goto error;
439 msg = "zero inode number";
440 if (unlikely(!first && gfs2_dirent_sentinel(dent)))
441 goto error;
442 msg = "name length is greater than space in dirent";
443 if (!gfs2_dirent_sentinel(dent) &&
444 unlikely(sizeof(struct gfs2_dirent)+be16_to_cpu(dent->de_name_len) >
445 size))
446 goto error;
447 return 0;
448 error:
449 printk(KERN_WARNING "gfs2_check_dirent: %s (%s)\n", msg,
450 first ? "first in block" : "not first in block");
451 return -EIO;
452 }
453
454 static int gfs2_dirent_offset(const void *buf)
455 {
456 const struct gfs2_meta_header *h = buf;
457 int offset;
458
459 BUG_ON(buf == NULL);
460
461 switch(be32_to_cpu(h->mh_type)) {
462 case GFS2_METATYPE_LF:
463 offset = sizeof(struct gfs2_leaf);
464 break;
465 case GFS2_METATYPE_DI:
466 offset = sizeof(struct gfs2_dinode);
467 break;
468 default:
469 goto wrong_type;
470 }
471 return offset;
472 wrong_type:
473 printk(KERN_WARNING "gfs2_scan_dirent: wrong block type %u\n",
474 be32_to_cpu(h->mh_type));
475 return -1;
476 }
477
478 static struct gfs2_dirent *gfs2_dirent_scan(struct inode *inode, void *buf,
479 unsigned int len, gfs2_dscan_t scan,
480 const struct qstr *name,
481 void *opaque)
482 {
483 struct gfs2_dirent *dent, *prev;
484 unsigned offset;
485 unsigned size;
486 int ret = 0;
487
488 ret = gfs2_dirent_offset(buf);
489 if (ret < 0)
490 goto consist_inode;
491
492 offset = ret;
493 prev = NULL;
494 dent = buf + offset;
495 size = be16_to_cpu(dent->de_rec_len);
496 if (gfs2_check_dirent(dent, offset, size, len, 1))
497 goto consist_inode;
498 do {
499 ret = scan(dent, name, opaque);
500 if (ret)
501 break;
502 offset += size;
503 if (offset == len)
504 break;
505 prev = dent;
506 dent = buf + offset;
507 size = be16_to_cpu(dent->de_rec_len);
508 if (gfs2_check_dirent(dent, offset, size, len, 0))
509 goto consist_inode;
510 } while(1);
511
512 switch(ret) {
513 case 0:
514 return NULL;
515 case 1:
516 return dent;
517 case 2:
518 return prev ? prev : dent;
519 default:
520 BUG_ON(ret > 0);
521 return ERR_PTR(ret);
522 }
523
524 consist_inode:
525 gfs2_consist_inode(GFS2_I(inode));
526 return ERR_PTR(-EIO);
527 }
528
529
530 /**
531 * dirent_first - Return the first dirent
532 * @dip: the directory
533 * @bh: The buffer
534 * @dent: Pointer to list of dirents
535 *
536 * return first dirent whether bh points to leaf or stuffed dinode
537 *
538 * Returns: IS_LEAF, IS_DINODE, or -errno
539 */
540
541 static int dirent_first(struct gfs2_inode *dip, struct buffer_head *bh,
542 struct gfs2_dirent **dent)
543 {
544 struct gfs2_meta_header *h = (struct gfs2_meta_header *)bh->b_data;
545
546 if (be32_to_cpu(h->mh_type) == GFS2_METATYPE_LF) {
547 if (gfs2_meta_check(GFS2_SB(&dip->i_inode), bh))
548 return -EIO;
549 *dent = (struct gfs2_dirent *)(bh->b_data +
550 sizeof(struct gfs2_leaf));
551 return IS_LEAF;
552 } else {
553 if (gfs2_metatype_check(GFS2_SB(&dip->i_inode), bh, GFS2_METATYPE_DI))
554 return -EIO;
555 *dent = (struct gfs2_dirent *)(bh->b_data +
556 sizeof(struct gfs2_dinode));
557 return IS_DINODE;
558 }
559 }
560
561 static int dirent_check_reclen(struct gfs2_inode *dip,
562 const struct gfs2_dirent *d, const void *end_p)
563 {
564 const void *ptr = d;
565 u16 rec_len = be16_to_cpu(d->de_rec_len);
566
567 if (unlikely(rec_len < sizeof(struct gfs2_dirent)))
568 goto broken;
569 ptr += rec_len;
570 if (ptr < end_p)
571 return rec_len;
572 if (ptr == end_p)
573 return -ENOENT;
574 broken:
575 gfs2_consist_inode(dip);
576 return -EIO;
577 }
578
579 /**
580 * dirent_next - Next dirent
581 * @dip: the directory
582 * @bh: The buffer
583 * @dent: Pointer to list of dirents
584 *
585 * Returns: 0 on success, error code otherwise
586 */
587
588 static int dirent_next(struct gfs2_inode *dip, struct buffer_head *bh,
589 struct gfs2_dirent **dent)
590 {
591 struct gfs2_dirent *cur = *dent, *tmp;
592 char *bh_end = bh->b_data + bh->b_size;
593 int ret;
594
595 ret = dirent_check_reclen(dip, cur, bh_end);
596 if (ret < 0)
597 return ret;
598
599 tmp = (void *)cur + ret;
600 ret = dirent_check_reclen(dip, tmp, bh_end);
601 if (ret == -EIO)
602 return ret;
603
604 /* Only the first dent could ever have de_inum.no_addr == 0 */
605 if (gfs2_dirent_sentinel(tmp)) {
606 gfs2_consist_inode(dip);
607 return -EIO;
608 }
609
610 *dent = tmp;
611 return 0;
612 }
613
614 /**
615 * dirent_del - Delete a dirent
616 * @dip: The GFS2 inode
617 * @bh: The buffer
618 * @prev: The previous dirent
619 * @cur: The current dirent
620 *
621 */
622
623 static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh,
624 struct gfs2_dirent *prev, struct gfs2_dirent *cur)
625 {
626 u16 cur_rec_len, prev_rec_len;
627
628 if (gfs2_dirent_sentinel(cur)) {
629 gfs2_consist_inode(dip);
630 return;
631 }
632
633 gfs2_trans_add_bh(dip->i_gl, bh, 1);
634
635 /* If there is no prev entry, this is the first entry in the block.
636 The de_rec_len is already as big as it needs to be. Just zero
637 out the inode number and return. */
638
639 if (!prev) {
640 cur->de_inum.no_addr = 0;
641 cur->de_inum.no_formal_ino = 0;
642 return;
643 }
644
645 /* Combine this dentry with the previous one. */
646
647 prev_rec_len = be16_to_cpu(prev->de_rec_len);
648 cur_rec_len = be16_to_cpu(cur->de_rec_len);
649
650 if ((char *)prev + prev_rec_len != (char *)cur)
651 gfs2_consist_inode(dip);
652 if ((char *)cur + cur_rec_len > bh->b_data + bh->b_size)
653 gfs2_consist_inode(dip);
654
655 prev_rec_len += cur_rec_len;
656 prev->de_rec_len = cpu_to_be16(prev_rec_len);
657 }
658
659 /*
660 * Takes a dent from which to grab space as an argument. Returns the
661 * newly created dent.
662 */
663 static struct gfs2_dirent *gfs2_init_dirent(struct inode *inode,
664 struct gfs2_dirent *dent,
665 const struct qstr *name,
666 struct buffer_head *bh)
667 {
668 struct gfs2_inode *ip = GFS2_I(inode);
669 struct gfs2_dirent *ndent;
670 unsigned offset = 0, totlen;
671
672 if (!gfs2_dirent_sentinel(dent))
673 offset = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
674 totlen = be16_to_cpu(dent->de_rec_len);
675 BUG_ON(offset + name->len > totlen);
676 gfs2_trans_add_bh(ip->i_gl, bh, 1);
677 ndent = (struct gfs2_dirent *)((char *)dent + offset);
678 dent->de_rec_len = cpu_to_be16(offset);
679 gfs2_qstr2dirent(name, totlen - offset, ndent);
680 return ndent;
681 }
682
683 static struct gfs2_dirent *gfs2_dirent_alloc(struct inode *inode,
684 struct buffer_head *bh,
685 const struct qstr *name)
686 {
687 struct gfs2_dirent *dent;
688 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
689 gfs2_dirent_find_space, name, NULL);
690 if (!dent || IS_ERR(dent))
691 return dent;
692 return gfs2_init_dirent(inode, dent, name, bh);
693 }
694
695 static int get_leaf(struct gfs2_inode *dip, u64 leaf_no,
696 struct buffer_head **bhp)
697 {
698 int error;
699
700 error = gfs2_meta_read(dip->i_gl, leaf_no, DIO_WAIT, bhp);
701 if (!error && gfs2_metatype_check(GFS2_SB(&dip->i_inode), *bhp, GFS2_METATYPE_LF)) {
702 /* printk(KERN_INFO "block num=%llu\n", leaf_no); */
703 error = -EIO;
704 }
705
706 return error;
707 }
708
709 /**
710 * get_leaf_nr - Get a leaf number associated with the index
711 * @dip: The GFS2 inode
712 * @index:
713 * @leaf_out:
714 *
715 * Returns: 0 on success, error code otherwise
716 */
717
718 static int get_leaf_nr(struct gfs2_inode *dip, u32 index,
719 u64 *leaf_out)
720 {
721 __be64 leaf_no;
722 int error;
723
724 error = gfs2_dir_read_data(dip, (char *)&leaf_no,
725 index * sizeof(__be64),
726 sizeof(__be64), 0);
727 if (error != sizeof(u64))
728 return (error < 0) ? error : -EIO;
729
730 *leaf_out = be64_to_cpu(leaf_no);
731
732 return 0;
733 }
734
735 static int get_first_leaf(struct gfs2_inode *dip, u32 index,
736 struct buffer_head **bh_out)
737 {
738 u64 leaf_no;
739 int error;
740
741 error = get_leaf_nr(dip, index, &leaf_no);
742 if (!error)
743 error = get_leaf(dip, leaf_no, bh_out);
744
745 return error;
746 }
747
748 static struct gfs2_dirent *gfs2_dirent_search(struct inode *inode,
749 const struct qstr *name,
750 gfs2_dscan_t scan,
751 struct buffer_head **pbh)
752 {
753 struct buffer_head *bh;
754 struct gfs2_dirent *dent;
755 struct gfs2_inode *ip = GFS2_I(inode);
756 int error;
757
758 if (ip->i_di.di_flags & GFS2_DIF_EXHASH) {
759 struct gfs2_leaf *leaf;
760 unsigned hsize = 1 << ip->i_di.di_depth;
761 unsigned index;
762 u64 ln;
763 if (hsize * sizeof(u64) != ip->i_di.di_size) {
764 gfs2_consist_inode(ip);
765 return ERR_PTR(-EIO);
766 }
767
768 index = name->hash >> (32 - ip->i_di.di_depth);
769 error = get_first_leaf(ip, index, &bh);
770 if (error)
771 return ERR_PTR(error);
772 do {
773 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
774 scan, name, NULL);
775 if (dent)
776 goto got_dent;
777 leaf = (struct gfs2_leaf *)bh->b_data;
778 ln = be64_to_cpu(leaf->lf_next);
779 brelse(bh);
780 if (!ln)
781 break;
782
783 error = get_leaf(ip, ln, &bh);
784 } while(!error);
785
786 return error ? ERR_PTR(error) : NULL;
787 }
788
789
790 error = gfs2_meta_inode_buffer(ip, &bh);
791 if (error)
792 return ERR_PTR(error);
793 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, scan, name, NULL);
794 got_dent:
795 if (unlikely(dent == NULL || IS_ERR(dent))) {
796 brelse(bh);
797 bh = NULL;
798 }
799 *pbh = bh;
800 return dent;
801 }
802
803 static struct gfs2_leaf *new_leaf(struct inode *inode, struct buffer_head **pbh, u16 depth)
804 {
805 struct gfs2_inode *ip = GFS2_I(inode);
806 u64 bn = gfs2_alloc_meta(ip);
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
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;
826 }
827
828 /**
829 * dir_make_exhash - Convert a stuffed directory into an ExHash directory
830 * @dip: The GFS2 inode
831 *
832 * Returns: 0 on success, error code otherwise
833 */
834
835 static int dir_make_exhash(struct inode *inode)
836 {
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;
848
849 error = gfs2_meta_inode_buffer(dip, &dibh);
850 if (error)
851 return error;
852
853 /* Turn over a new leaf */
854
855 leaf = new_leaf(inode, &bh, 0);
856 if (!leaf)
857 return -ENOSPC;
858 bn = bh->b_blocknr;
859
860 gfs2_assert(sdp, dip->i_di.di_entries < (1 << 16));
861 leaf->lf_entries = cpu_to_be16(dip->i_di.di_entries);
862
863 /* Copy dirents */
864
865 gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_leaf), dibh,
866 sizeof(struct gfs2_dinode));
867
868 /* Find last entry */
869
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;
880 }
881 if (IS_ERR(dent)) {
882 brelse(bh);
883 brelse(dibh);
884 return PTR_ERR(dent);
885 }
886
887 /* Adjust the last dirent's record length
888 (Remember that dent still points to the last entry.) */
889
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));
893
894 brelse(bh);
895
896 /* We're done with the new leaf block, now setup the new
897 hash table. */
898
899 gfs2_trans_add_bh(dip->i_gl, dibh, 1);
900 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
901
902 lp = (__be64 *)(dibh->b_data + sizeof(struct gfs2_dinode));
903
904 for (x = sdp->sd_hash_ptrs; x--; lp++)
905 *lp = cpu_to_be64(bn);
906
907 dip->i_di.di_size = sdp->sd_sb.sb_bsize / 2;
908 dip->i_di.di_blocks++;
909 gfs2_set_inode_blocks(&dip->i_inode);
910 dip->i_di.di_flags |= GFS2_DIF_EXHASH;
911
912 for (x = sdp->sd_hash_ptrs, y = -1; x; x >>= 1, y++) ;
913 dip->i_di.di_depth = y;
914
915 gfs2_dinode_out(dip, dibh->b_data);
916
917 brelse(dibh);
918
919 return 0;
920 }
921
922 /**
923 * dir_split_leaf - Split a leaf block into two
924 * @dip: The GFS2 inode
925 * @index:
926 * @leaf_no:
927 *
928 * Returns: 0 on success, error code on failure
929 */
930
931 static int dir_split_leaf(struct inode *inode, const struct qstr *name)
932 {
933 struct gfs2_inode *dip = GFS2_I(inode);
934 struct buffer_head *nbh, *obh, *dibh;
935 struct gfs2_leaf *nleaf, *oleaf;
936 struct gfs2_dirent *dent = NULL, *prev = NULL, *next = NULL, *new;
937 u32 start, len, half_len, divider;
938 u64 bn, leaf_no;
939 __be64 *lp;
940 u32 index;
941 int x, moved = 0;
942 int error;
943
944 index = name->hash >> (32 - dip->i_di.di_depth);
945 error = get_leaf_nr(dip, index, &leaf_no);
946 if (error)
947 return error;
948
949 /* Get the old leaf block */
950 error = get_leaf(dip, leaf_no, &obh);
951 if (error)
952 return error;
953
954 oleaf = (struct gfs2_leaf *)obh->b_data;
955 if (dip->i_di.di_depth == be16_to_cpu(oleaf->lf_depth)) {
956 brelse(obh);
957 return 1; /* can't split */
958 }
959
960 gfs2_trans_add_bh(dip->i_gl, obh, 1);
961
962 nleaf = new_leaf(inode, &nbh, be16_to_cpu(oleaf->lf_depth) + 1);
963 if (!nleaf) {
964 brelse(obh);
965 return -ENOSPC;
966 }
967 bn = nbh->b_blocknr;
968
969 /* Compute the start and len of leaf pointers in the hash table. */
970 len = 1 << (dip->i_di.di_depth - be16_to_cpu(oleaf->lf_depth));
971 half_len = len >> 1;
972 if (!half_len) {
973 printk(KERN_WARNING "di_depth %u lf_depth %u index %u\n", dip->i_di.di_depth, be16_to_cpu(oleaf->lf_depth), index);
974 gfs2_consist_inode(dip);
975 error = -EIO;
976 goto fail_brelse;
977 }
978
979 start = (index & ~(len - 1));
980
981 /* Change the pointers.
982 Don't bother distinguishing stuffed from non-stuffed.
983 This code is complicated enough already. */
984 lp = kmalloc(half_len * sizeof(__be64), GFP_NOFS | __GFP_NOFAIL);
985 /* Change the pointers */
986 for (x = 0; x < half_len; x++)
987 lp[x] = cpu_to_be64(bn);
988
989 error = gfs2_dir_write_data(dip, (char *)lp, start * sizeof(u64),
990 half_len * sizeof(u64));
991 if (error != half_len * sizeof(u64)) {
992 if (error >= 0)
993 error = -EIO;
994 goto fail_lpfree;
995 }
996
997 kfree(lp);
998
999 /* Compute the divider */
1000 divider = (start + half_len) << (32 - dip->i_di.di_depth);
1001
1002 /* Copy the entries */
1003 dirent_first(dip, obh, &dent);
1004
1005 do {
1006 next = dent;
1007 if (dirent_next(dip, obh, &next))
1008 next = NULL;
1009
1010 if (!gfs2_dirent_sentinel(dent) &&
1011 be32_to_cpu(dent->de_hash) < divider) {
1012 struct qstr str;
1013 str.name = (char*)(dent+1);
1014 str.len = be16_to_cpu(dent->de_name_len);
1015 str.hash = be32_to_cpu(dent->de_hash);
1016 new = gfs2_dirent_alloc(inode, nbh, &str);
1017 if (IS_ERR(new)) {
1018 error = PTR_ERR(new);
1019 break;
1020 }
1021
1022 new->de_inum = dent->de_inum; /* No endian worries */
1023 new->de_type = dent->de_type; /* No endian worries */
1024 nleaf->lf_entries = cpu_to_be16(be16_to_cpu(nleaf->lf_entries)+1);
1025
1026 dirent_del(dip, obh, prev, dent);
1027
1028 if (!oleaf->lf_entries)
1029 gfs2_consist_inode(dip);
1030 oleaf->lf_entries = cpu_to_be16(be16_to_cpu(oleaf->lf_entries)-1);
1031
1032 if (!prev)
1033 prev = dent;
1034
1035 moved = 1;
1036 } else {
1037 prev = dent;
1038 }
1039 dent = next;
1040 } while (dent);
1041
1042 oleaf->lf_depth = nleaf->lf_depth;
1043
1044 error = gfs2_meta_inode_buffer(dip, &dibh);
1045 if (!gfs2_assert_withdraw(GFS2_SB(&dip->i_inode), !error)) {
1046 gfs2_trans_add_bh(dip->i_gl, dibh, 1);
1047 dip->i_di.di_blocks++;
1048 gfs2_set_inode_blocks(&dip->i_inode);
1049 gfs2_dinode_out(dip, dibh->b_data);
1050 brelse(dibh);
1051 }
1052
1053 brelse(obh);
1054 brelse(nbh);
1055
1056 return error;
1057
1058 fail_lpfree:
1059 kfree(lp);
1060
1061 fail_brelse:
1062 brelse(obh);
1063 brelse(nbh);
1064 return error;
1065 }
1066
1067 /**
1068 * dir_double_exhash - Double size of ExHash table
1069 * @dip: The GFS2 dinode
1070 *
1071 * Returns: 0 on success, error code on failure
1072 */
1073
1074 static int dir_double_exhash(struct gfs2_inode *dip)
1075 {
1076 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1077 struct buffer_head *dibh;
1078 u32 hsize;
1079 u64 *buf;
1080 u64 *from, *to;
1081 u64 block;
1082 int x;
1083 int error = 0;
1084
1085 hsize = 1 << dip->i_di.di_depth;
1086 if (hsize * sizeof(u64) != dip->i_di.di_size) {
1087 gfs2_consist_inode(dip);
1088 return -EIO;
1089 }
1090
1091 /* Allocate both the "from" and "to" buffers in one big chunk */
1092
1093 buf = kcalloc(3, sdp->sd_hash_bsize, GFP_KERNEL | __GFP_NOFAIL);
1094
1095 for (block = dip->i_di.di_size >> sdp->sd_hash_bsize_shift; block--;) {
1096 error = gfs2_dir_read_data(dip, (char *)buf,
1097 block * sdp->sd_hash_bsize,
1098 sdp->sd_hash_bsize, 1);
1099 if (error != sdp->sd_hash_bsize) {
1100 if (error >= 0)
1101 error = -EIO;
1102 goto fail;
1103 }
1104
1105 from = buf;
1106 to = (u64 *)((char *)buf + sdp->sd_hash_bsize);
1107
1108 for (x = sdp->sd_hash_ptrs; x--; from++) {
1109 *to++ = *from; /* No endianess worries */
1110 *to++ = *from;
1111 }
1112
1113 error = gfs2_dir_write_data(dip,
1114 (char *)buf + sdp->sd_hash_bsize,
1115 block * sdp->sd_sb.sb_bsize,
1116 sdp->sd_sb.sb_bsize);
1117 if (error != sdp->sd_sb.sb_bsize) {
1118 if (error >= 0)
1119 error = -EIO;
1120 goto fail;
1121 }
1122 }
1123
1124 kfree(buf);
1125
1126 error = gfs2_meta_inode_buffer(dip, &dibh);
1127 if (!gfs2_assert_withdraw(sdp, !error)) {
1128 dip->i_di.di_depth++;
1129 gfs2_dinode_out(dip, dibh->b_data);
1130 brelse(dibh);
1131 }
1132
1133 return error;
1134
1135 fail:
1136 kfree(buf);
1137 return error;
1138 }
1139
1140 /**
1141 * compare_dents - compare directory entries by hash value
1142 * @a: first dent
1143 * @b: second dent
1144 *
1145 * When comparing the hash entries of @a to @b:
1146 * gt: returns 1
1147 * lt: returns -1
1148 * eq: returns 0
1149 */
1150
1151 static int compare_dents(const void *a, const void *b)
1152 {
1153 const struct gfs2_dirent *dent_a, *dent_b;
1154 u32 hash_a, hash_b;
1155 int ret = 0;
1156
1157 dent_a = *(const struct gfs2_dirent **)a;
1158 hash_a = be32_to_cpu(dent_a->de_hash);
1159
1160 dent_b = *(const struct gfs2_dirent **)b;
1161 hash_b = be32_to_cpu(dent_b->de_hash);
1162
1163 if (hash_a > hash_b)
1164 ret = 1;
1165 else if (hash_a < hash_b)
1166 ret = -1;
1167 else {
1168 unsigned int len_a = be16_to_cpu(dent_a->de_name_len);
1169 unsigned int len_b = be16_to_cpu(dent_b->de_name_len);
1170
1171 if (len_a > len_b)
1172 ret = 1;
1173 else if (len_a < len_b)
1174 ret = -1;
1175 else
1176 ret = memcmp(dent_a + 1, dent_b + 1, len_a);
1177 }
1178
1179 return ret;
1180 }
1181
1182 /**
1183 * do_filldir_main - read out directory entries
1184 * @dip: The GFS2 inode
1185 * @offset: The offset in the file to read from
1186 * @opaque: opaque data to pass to filldir
1187 * @filldir: The function to pass entries to
1188 * @darr: an array of struct gfs2_dirent pointers to read
1189 * @entries: the number of entries in darr
1190 * @copied: pointer to int that's non-zero if a entry has been copied out
1191 *
1192 * Jump through some hoops to make sure that if there are hash collsions,
1193 * they are read out at the beginning of a buffer. We want to minimize
1194 * the possibility that they will fall into different readdir buffers or
1195 * that someone will want to seek to that location.
1196 *
1197 * Returns: errno, >0 on exception from filldir
1198 */
1199
1200 static int do_filldir_main(struct gfs2_inode *dip, u64 *offset,
1201 void *opaque, filldir_t filldir,
1202 const struct gfs2_dirent **darr, u32 entries,
1203 int *copied)
1204 {
1205 const struct gfs2_dirent *dent, *dent_next;
1206 u64 off, off_next;
1207 unsigned int x, y;
1208 int run = 0;
1209 int error = 0;
1210
1211 sort(darr, entries, sizeof(struct gfs2_dirent *), compare_dents, NULL);
1212
1213 dent_next = darr[0];
1214 off_next = be32_to_cpu(dent_next->de_hash);
1215 off_next = gfs2_disk_hash2offset(off_next);
1216
1217 for (x = 0, y = 1; x < entries; x++, y++) {
1218 dent = dent_next;
1219 off = off_next;
1220
1221 if (y < entries) {
1222 dent_next = darr[y];
1223 off_next = be32_to_cpu(dent_next->de_hash);
1224 off_next = gfs2_disk_hash2offset(off_next);
1225
1226 if (off < *offset)
1227 continue;
1228 *offset = off;
1229
1230 if (off_next == off) {
1231 if (*copied && !run)
1232 return 1;
1233 run = 1;
1234 } else
1235 run = 0;
1236 } else {
1237 if (off < *offset)
1238 continue;
1239 *offset = off;
1240 }
1241
1242 error = filldir(opaque, (const char *)(dent + 1),
1243 be16_to_cpu(dent->de_name_len),
1244 off, be64_to_cpu(dent->de_inum.no_addr),
1245 be16_to_cpu(dent->de_type));
1246 if (error)
1247 return 1;
1248
1249 *copied = 1;
1250 }
1251
1252 /* Increment the *offset by one, so the next time we come into the
1253 do_filldir fxn, we get the next entry instead of the last one in the
1254 current leaf */
1255
1256 (*offset)++;
1257
1258 return 0;
1259 }
1260
1261 static int gfs2_dir_read_leaf(struct inode *inode, u64 *offset, void *opaque,
1262 filldir_t filldir, int *copied, unsigned *depth,
1263 u64 leaf_no)
1264 {
1265 struct gfs2_inode *ip = GFS2_I(inode);
1266 struct gfs2_sbd *sdp = GFS2_SB(inode);
1267 struct buffer_head *bh;
1268 struct gfs2_leaf *lf;
1269 unsigned entries = 0, entries2 = 0;
1270 unsigned leaves = 0;
1271 const struct gfs2_dirent **darr, *dent;
1272 struct dirent_gather g;
1273 struct buffer_head **larr;
1274 int leaf = 0;
1275 int error, i;
1276 u64 lfn = leaf_no;
1277
1278 do {
1279 error = get_leaf(ip, lfn, &bh);
1280 if (error)
1281 goto out;
1282 lf = (struct gfs2_leaf *)bh->b_data;
1283 if (leaves == 0)
1284 *depth = be16_to_cpu(lf->lf_depth);
1285 entries += be16_to_cpu(lf->lf_entries);
1286 leaves++;
1287 lfn = be64_to_cpu(lf->lf_next);
1288 brelse(bh);
1289 } while(lfn);
1290
1291 if (!entries)
1292 return 0;
1293
1294 error = -ENOMEM;
1295 /*
1296 * The extra 99 entries are not normally used, but are a buffer
1297 * zone in case the number of entries in the leaf is corrupt.
1298 * 99 is the maximum number of entries that can fit in a single
1299 * leaf block.
1300 */
1301 larr = vmalloc((leaves + entries + 99) * sizeof(void *));
1302 if (!larr)
1303 goto out;
1304 darr = (const struct gfs2_dirent **)(larr + leaves);
1305 g.pdent = darr;
1306 g.offset = 0;
1307 lfn = leaf_no;
1308
1309 do {
1310 error = get_leaf(ip, lfn, &bh);
1311 if (error)
1312 goto out_kfree;
1313 lf = (struct gfs2_leaf *)bh->b_data;
1314 lfn = be64_to_cpu(lf->lf_next);
1315 if (lf->lf_entries) {
1316 entries2 += be16_to_cpu(lf->lf_entries);
1317 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
1318 gfs2_dirent_gather, NULL, &g);
1319 error = PTR_ERR(dent);
1320 if (IS_ERR(dent))
1321 goto out_kfree;
1322 if (entries2 != g.offset) {
1323 fs_warn(sdp, "Number of entries corrupt in dir "
1324 "leaf %llu, entries2 (%u) != "
1325 "g.offset (%u)\n",
1326 (unsigned long long)bh->b_blocknr,
1327 entries2, g.offset);
1328
1329 error = -EIO;
1330 goto out_kfree;
1331 }
1332 error = 0;
1333 larr[leaf++] = bh;
1334 } else {
1335 brelse(bh);
1336 }
1337 } while(lfn);
1338
1339 BUG_ON(entries2 != entries);
1340 error = do_filldir_main(ip, offset, opaque, filldir, darr,
1341 entries, copied);
1342 out_kfree:
1343 for(i = 0; i < leaf; i++)
1344 brelse(larr[i]);
1345 vfree(larr);
1346 out:
1347 return error;
1348 }
1349
1350 /**
1351 * dir_e_read - Reads the entries from a directory into a filldir buffer
1352 * @dip: dinode pointer
1353 * @offset: the hash of the last entry read shifted to the right once
1354 * @opaque: buffer for the filldir function to fill
1355 * @filldir: points to the filldir function to use
1356 *
1357 * Returns: errno
1358 */
1359
1360 static int dir_e_read(struct inode *inode, u64 *offset, void *opaque,
1361 filldir_t filldir)
1362 {
1363 struct gfs2_inode *dip = GFS2_I(inode);
1364 struct gfs2_sbd *sdp = GFS2_SB(inode);
1365 u32 hsize, len = 0;
1366 u32 ht_offset, lp_offset, ht_offset_cur = -1;
1367 u32 hash, index;
1368 __be64 *lp;
1369 int copied = 0;
1370 int error = 0;
1371 unsigned depth = 0;
1372
1373 hsize = 1 << dip->i_di.di_depth;
1374 if (hsize * sizeof(u64) != dip->i_di.di_size) {
1375 gfs2_consist_inode(dip);
1376 return -EIO;
1377 }
1378
1379 hash = gfs2_dir_offset2hash(*offset);
1380 index = hash >> (32 - dip->i_di.di_depth);
1381
1382 lp = kmalloc(sdp->sd_hash_bsize, GFP_KERNEL);
1383 if (!lp)
1384 return -ENOMEM;
1385
1386 while (index < hsize) {
1387 lp_offset = index & (sdp->sd_hash_ptrs - 1);
1388 ht_offset = index - lp_offset;
1389
1390 if (ht_offset_cur != ht_offset) {
1391 error = gfs2_dir_read_data(dip, (char *)lp,
1392 ht_offset * sizeof(__be64),
1393 sdp->sd_hash_bsize, 1);
1394 if (error != sdp->sd_hash_bsize) {
1395 if (error >= 0)
1396 error = -EIO;
1397 goto out;
1398 }
1399 ht_offset_cur = ht_offset;
1400 }
1401
1402 error = gfs2_dir_read_leaf(inode, offset, opaque, filldir,
1403 &copied, &depth,
1404 be64_to_cpu(lp[lp_offset]));
1405 if (error)
1406 break;
1407
1408 len = 1 << (dip->i_di.di_depth - depth);
1409 index = (index & ~(len - 1)) + len;
1410 }
1411
1412 out:
1413 kfree(lp);
1414 if (error > 0)
1415 error = 0;
1416 return error;
1417 }
1418
1419 int gfs2_dir_read(struct inode *inode, u64 *offset, void *opaque,
1420 filldir_t filldir)
1421 {
1422 struct gfs2_inode *dip = GFS2_I(inode);
1423 struct gfs2_sbd *sdp = GFS2_SB(inode);
1424 struct dirent_gather g;
1425 const struct gfs2_dirent **darr, *dent;
1426 struct buffer_head *dibh;
1427 int copied = 0;
1428 int error;
1429
1430 if (!dip->i_di.di_entries)
1431 return 0;
1432
1433 if (dip->i_di.di_flags & GFS2_DIF_EXHASH)
1434 return dir_e_read(inode, offset, opaque, filldir);
1435
1436 if (!gfs2_is_stuffed(dip)) {
1437 gfs2_consist_inode(dip);
1438 return -EIO;
1439 }
1440
1441 error = gfs2_meta_inode_buffer(dip, &dibh);
1442 if (error)
1443 return error;
1444
1445 error = -ENOMEM;
1446 /* 96 is max number of dirents which can be stuffed into an inode */
1447 darr = kmalloc(96 * sizeof(struct gfs2_dirent *), GFP_KERNEL);
1448 if (darr) {
1449 g.pdent = darr;
1450 g.offset = 0;
1451 dent = gfs2_dirent_scan(inode, dibh->b_data, dibh->b_size,
1452 gfs2_dirent_gather, NULL, &g);
1453 if (IS_ERR(dent)) {
1454 error = PTR_ERR(dent);
1455 goto out;
1456 }
1457 if (dip->i_di.di_entries != g.offset) {
1458 fs_warn(sdp, "Number of entries corrupt in dir %llu, "
1459 "ip->i_di.di_entries (%u) != g.offset (%u)\n",
1460 (unsigned long long)dip->i_no_addr,
1461 dip->i_di.di_entries,
1462 g.offset);
1463 error = -EIO;
1464 goto out;
1465 }
1466 error = do_filldir_main(dip, offset, opaque, filldir, darr,
1467 dip->i_di.di_entries, &copied);
1468 out:
1469 kfree(darr);
1470 }
1471
1472 if (error > 0)
1473 error = 0;
1474
1475 brelse(dibh);
1476
1477 return error;
1478 }
1479
1480 /**
1481 * gfs2_dir_search - Search a directory
1482 * @dip: The GFS2 inode
1483 * @filename:
1484 * @inode:
1485 *
1486 * This routine searches a directory for a file or another directory.
1487 * Assumes a glock is held on dip.
1488 *
1489 * Returns: errno
1490 */
1491
1492 struct inode *gfs2_dir_search(struct inode *dir, const struct qstr *name)
1493 {
1494 struct buffer_head *bh;
1495 struct gfs2_dirent *dent;
1496 struct inode *inode;
1497
1498 dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
1499 if (dent) {
1500 if (IS_ERR(dent))
1501 return ERR_CAST(dent);
1502 inode = gfs2_inode_lookup(dir->i_sb,
1503 be16_to_cpu(dent->de_type),
1504 be64_to_cpu(dent->de_inum.no_addr),
1505 be64_to_cpu(dent->de_inum.no_formal_ino), 0);
1506 brelse(bh);
1507 return inode;
1508 }
1509 return ERR_PTR(-ENOENT);
1510 }
1511
1512 int gfs2_dir_check(struct inode *dir, const struct qstr *name,
1513 const struct gfs2_inode *ip)
1514 {
1515 struct buffer_head *bh;
1516 struct gfs2_dirent *dent;
1517 int ret = -ENOENT;
1518
1519 dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
1520 if (dent) {
1521 if (IS_ERR(dent))
1522 return PTR_ERR(dent);
1523 if (ip) {
1524 if (be64_to_cpu(dent->de_inum.no_addr) != ip->i_no_addr)
1525 goto out;
1526 if (be64_to_cpu(dent->de_inum.no_formal_ino) !=
1527 ip->i_no_formal_ino)
1528 goto out;
1529 if (unlikely(IF2DT(ip->i_inode.i_mode) !=
1530 be16_to_cpu(dent->de_type))) {
1531 gfs2_consist_inode(GFS2_I(dir));
1532 ret = -EIO;
1533 goto out;
1534 }
1535 }
1536 ret = 0;
1537 out:
1538 brelse(bh);
1539 }
1540 return ret;
1541 }
1542
1543 static int dir_new_leaf(struct inode *inode, const struct qstr *name)
1544 {
1545 struct buffer_head *bh, *obh;
1546 struct gfs2_inode *ip = GFS2_I(inode);
1547 struct gfs2_leaf *leaf, *oleaf;
1548 int error;
1549 u32 index;
1550 u64 bn;
1551
1552 index = name->hash >> (32 - ip->i_di.di_depth);
1553 error = get_first_leaf(ip, index, &obh);
1554 if (error)
1555 return error;
1556 do {
1557 oleaf = (struct gfs2_leaf *)obh->b_data;
1558 bn = be64_to_cpu(oleaf->lf_next);
1559 if (!bn)
1560 break;
1561 brelse(obh);
1562 error = get_leaf(ip, bn, &obh);
1563 if (error)
1564 return error;
1565 } while(1);
1566
1567 gfs2_trans_add_bh(ip->i_gl, obh, 1);
1568
1569 leaf = new_leaf(inode, &bh, be16_to_cpu(oleaf->lf_depth));
1570 if (!leaf) {
1571 brelse(obh);
1572 return -ENOSPC;
1573 }
1574 oleaf->lf_next = cpu_to_be64(bh->b_blocknr);
1575 brelse(bh);
1576 brelse(obh);
1577
1578 error = gfs2_meta_inode_buffer(ip, &bh);
1579 if (error)
1580 return error;
1581 gfs2_trans_add_bh(ip->i_gl, bh, 1);
1582 ip->i_di.di_blocks++;
1583 gfs2_set_inode_blocks(&ip->i_inode);
1584 gfs2_dinode_out(ip, bh->b_data);
1585 brelse(bh);
1586 return 0;
1587 }
1588
1589 /**
1590 * gfs2_dir_add - Add new filename into directory
1591 * @dip: The GFS2 inode
1592 * @filename: The new name
1593 * @inode: The inode number of the entry
1594 * @type: The type of the entry
1595 *
1596 * Returns: 0 on success, error code on failure
1597 */
1598
1599 int gfs2_dir_add(struct inode *inode, const struct qstr *name,
1600 const struct gfs2_inode *nip, unsigned type)
1601 {
1602 struct gfs2_inode *ip = GFS2_I(inode);
1603 struct buffer_head *bh;
1604 struct gfs2_dirent *dent;
1605 struct gfs2_leaf *leaf;
1606 int error;
1607
1608 while(1) {
1609 dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space,
1610 &bh);
1611 if (dent) {
1612 if (IS_ERR(dent))
1613 return PTR_ERR(dent);
1614 dent = gfs2_init_dirent(inode, dent, name, bh);
1615 gfs2_inum_out(nip, dent);
1616 dent->de_type = cpu_to_be16(type);
1617 if (ip->i_di.di_flags & GFS2_DIF_EXHASH) {
1618 leaf = (struct gfs2_leaf *)bh->b_data;
1619 leaf->lf_entries = cpu_to_be16(be16_to_cpu(leaf->lf_entries) + 1);
1620 }
1621 brelse(bh);
1622 error = gfs2_meta_inode_buffer(ip, &bh);
1623 if (error)
1624 break;
1625 gfs2_trans_add_bh(ip->i_gl, bh, 1);
1626 ip->i_di.di_entries++;
1627 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
1628 gfs2_dinode_out(ip, bh->b_data);
1629 brelse(bh);
1630 error = 0;
1631 break;
1632 }
1633 if (!(ip->i_di.di_flags & GFS2_DIF_EXHASH)) {
1634 error = dir_make_exhash(inode);
1635 if (error)
1636 break;
1637 continue;
1638 }
1639 error = dir_split_leaf(inode, name);
1640 if (error == 0)
1641 continue;
1642 if (error < 0)
1643 break;
1644 if (ip->i_di.di_depth < GFS2_DIR_MAX_DEPTH) {
1645 error = dir_double_exhash(ip);
1646 if (error)
1647 break;
1648 error = dir_split_leaf(inode, name);
1649 if (error < 0)
1650 break;
1651 if (error == 0)
1652 continue;
1653 }
1654 error = dir_new_leaf(inode, name);
1655 if (!error)
1656 continue;
1657 error = -ENOSPC;
1658 break;
1659 }
1660 return error;
1661 }
1662
1663
1664 /**
1665 * gfs2_dir_del - Delete a directory entry
1666 * @dip: The GFS2 inode
1667 * @filename: The filename
1668 *
1669 * Returns: 0 on success, error code on failure
1670 */
1671
1672 int gfs2_dir_del(struct gfs2_inode *dip, const struct qstr *name)
1673 {
1674 struct gfs2_dirent *dent, *prev = NULL;
1675 struct buffer_head *bh;
1676 int error;
1677
1678 /* Returns _either_ the entry (if its first in block) or the
1679 previous entry otherwise */
1680 dent = gfs2_dirent_search(&dip->i_inode, name, gfs2_dirent_prev, &bh);
1681 if (!dent) {
1682 gfs2_consist_inode(dip);
1683 return -EIO;
1684 }
1685 if (IS_ERR(dent)) {
1686 gfs2_consist_inode(dip);
1687 return PTR_ERR(dent);
1688 }
1689 /* If not first in block, adjust pointers accordingly */
1690 if (gfs2_dirent_find(dent, name, NULL) == 0) {
1691 prev = dent;
1692 dent = (struct gfs2_dirent *)((char *)dent + be16_to_cpu(prev->de_rec_len));
1693 }
1694
1695 dirent_del(dip, bh, prev, dent);
1696 if (dip->i_di.di_flags & GFS2_DIF_EXHASH) {
1697 struct gfs2_leaf *leaf = (struct gfs2_leaf *)bh->b_data;
1698 u16 entries = be16_to_cpu(leaf->lf_entries);
1699 if (!entries)
1700 gfs2_consist_inode(dip);
1701 leaf->lf_entries = cpu_to_be16(--entries);
1702 }
1703 brelse(bh);
1704
1705 error = gfs2_meta_inode_buffer(dip, &bh);
1706 if (error)
1707 return error;
1708
1709 if (!dip->i_di.di_entries)
1710 gfs2_consist_inode(dip);
1711 gfs2_trans_add_bh(dip->i_gl, bh, 1);
1712 dip->i_di.di_entries--;
1713 dip->i_inode.i_mtime = dip->i_inode.i_ctime = CURRENT_TIME;
1714 gfs2_dinode_out(dip, bh->b_data);
1715 brelse(bh);
1716 mark_inode_dirty(&dip->i_inode);
1717
1718 return error;
1719 }
1720
1721 /**
1722 * gfs2_dir_mvino - Change inode number of directory entry
1723 * @dip: The GFS2 inode
1724 * @filename:
1725 * @new_inode:
1726 *
1727 * This routine changes the inode number of a directory entry. It's used
1728 * by rename to change ".." when a directory is moved.
1729 * Assumes a glock is held on dvp.
1730 *
1731 * Returns: errno
1732 */
1733
1734 int gfs2_dir_mvino(struct gfs2_inode *dip, const struct qstr *filename,
1735 const struct gfs2_inode *nip, unsigned int new_type)
1736 {
1737 struct buffer_head *bh;
1738 struct gfs2_dirent *dent;
1739 int error;
1740
1741 dent = gfs2_dirent_search(&dip->i_inode, filename, gfs2_dirent_find, &bh);
1742 if (!dent) {
1743 gfs2_consist_inode(dip);
1744 return -EIO;
1745 }
1746 if (IS_ERR(dent))
1747 return PTR_ERR(dent);
1748
1749 gfs2_trans_add_bh(dip->i_gl, bh, 1);
1750 gfs2_inum_out(nip, dent);
1751 dent->de_type = cpu_to_be16(new_type);
1752
1753 if (dip->i_di.di_flags & GFS2_DIF_EXHASH) {
1754 brelse(bh);
1755 error = gfs2_meta_inode_buffer(dip, &bh);
1756 if (error)
1757 return error;
1758 gfs2_trans_add_bh(dip->i_gl, bh, 1);
1759 }
1760
1761 dip->i_inode.i_mtime = dip->i_inode.i_ctime = CURRENT_TIME;
1762 gfs2_dinode_out(dip, bh->b_data);
1763 brelse(bh);
1764 return 0;
1765 }
1766
1767 /**
1768 * foreach_leaf - call a function for each leaf in a directory
1769 * @dip: the directory
1770 * @lc: the function to call for each each
1771 * @data: private data to pass to it
1772 *
1773 * Returns: errno
1774 */
1775
1776 static int foreach_leaf(struct gfs2_inode *dip, leaf_call_t lc, void *data)
1777 {
1778 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1779 struct buffer_head *bh;
1780 struct gfs2_leaf *leaf;
1781 u32 hsize, len;
1782 u32 ht_offset, lp_offset, ht_offset_cur = -1;
1783 u32 index = 0;
1784 __be64 *lp;
1785 u64 leaf_no;
1786 int error = 0;
1787
1788 hsize = 1 << dip->i_di.di_depth;
1789 if (hsize * sizeof(u64) != dip->i_di.di_size) {
1790 gfs2_consist_inode(dip);
1791 return -EIO;
1792 }
1793
1794 lp = kmalloc(sdp->sd_hash_bsize, GFP_KERNEL);
1795 if (!lp)
1796 return -ENOMEM;
1797
1798 while (index < hsize) {
1799 lp_offset = index & (sdp->sd_hash_ptrs - 1);
1800 ht_offset = index - lp_offset;
1801
1802 if (ht_offset_cur != ht_offset) {
1803 error = gfs2_dir_read_data(dip, (char *)lp,
1804 ht_offset * sizeof(__be64),
1805 sdp->sd_hash_bsize, 1);
1806 if (error != sdp->sd_hash_bsize) {
1807 if (error >= 0)
1808 error = -EIO;
1809 goto out;
1810 }
1811 ht_offset_cur = ht_offset;
1812 }
1813
1814 leaf_no = be64_to_cpu(lp[lp_offset]);
1815 if (leaf_no) {
1816 error = get_leaf(dip, leaf_no, &bh);
1817 if (error)
1818 goto out;
1819 leaf = (struct gfs2_leaf *)bh->b_data;
1820 len = 1 << (dip->i_di.di_depth - be16_to_cpu(leaf->lf_depth));
1821 brelse(bh);
1822
1823 error = lc(dip, index, len, leaf_no, data);
1824 if (error)
1825 goto out;
1826
1827 index = (index & ~(len - 1)) + len;
1828 } else
1829 index++;
1830 }
1831
1832 if (index != hsize) {
1833 gfs2_consist_inode(dip);
1834 error = -EIO;
1835 }
1836
1837 out:
1838 kfree(lp);
1839
1840 return error;
1841 }
1842
1843 /**
1844 * leaf_dealloc - Deallocate a directory leaf
1845 * @dip: the directory
1846 * @index: the hash table offset in the directory
1847 * @len: the number of pointers to this leaf
1848 * @leaf_no: the leaf number
1849 * @data: not used
1850 *
1851 * Returns: errno
1852 */
1853
1854 static int leaf_dealloc(struct gfs2_inode *dip, u32 index, u32 len,
1855 u64 leaf_no, void *data)
1856 {
1857 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1858 struct gfs2_leaf *tmp_leaf;
1859 struct gfs2_rgrp_list rlist;
1860 struct buffer_head *bh, *dibh;
1861 u64 blk, nblk;
1862 unsigned int rg_blocks = 0, l_blocks = 0;
1863 char *ht;
1864 unsigned int x, size = len * sizeof(u64);
1865 int error;
1866
1867 memset(&rlist, 0, sizeof(struct gfs2_rgrp_list));
1868
1869 ht = kzalloc(size, GFP_KERNEL);
1870 if (!ht)
1871 return -ENOMEM;
1872
1873 gfs2_alloc_get(dip);
1874
1875 error = gfs2_quota_hold(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
1876 if (error)
1877 goto out;
1878
1879 error = gfs2_rindex_hold(sdp, &dip->i_alloc->al_ri_gh);
1880 if (error)
1881 goto out_qs;
1882
1883 /* Count the number of leaves */
1884
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);
1892
1893 gfs2_rlist_add(sdp, &rlist, blk);
1894 l_blocks++;
1895 }
1896
1897 gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE, 0);
1898
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;
1903 }
1904
1905 error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs);
1906 if (error)
1907 goto out_rlist;
1908
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;
1914
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);
1922
1923 gfs2_free_meta(dip, blk, 1);
1924
1925 if (!dip->i_di.di_blocks)
1926 gfs2_consist_inode(dip);
1927 dip->i_di.di_blocks--;
1928 gfs2_set_inode_blocks(&dip->i_inode);
1929 }
1930
1931 error = gfs2_dir_write_data(dip, ht, index * sizeof(u64), size);
1932 if (error != size) {
1933 if (error >= 0)
1934 error = -EIO;
1935 goto out_end_trans;
1936 }
1937
1938 error = gfs2_meta_inode_buffer(dip, &dibh);
1939 if (error)
1940 goto out_end_trans;
1941
1942 gfs2_trans_add_bh(dip->i_gl, dibh, 1);
1943 gfs2_dinode_out(dip, dibh->b_data);
1944 brelse(dibh);
1945
1946 out_end_trans:
1947 gfs2_trans_end(sdp);
1948 out_rg_gunlock:
1949 gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
1950 out_rlist:
1951 gfs2_rlist_free(&rlist);
1952 gfs2_glock_dq_uninit(&dip->i_alloc->al_ri_gh);
1953 out_qs:
1954 gfs2_quota_unhold(dip);
1955 out:
1956 gfs2_alloc_put(dip);
1957 kfree(ht);
1958 return error;
1959 }
1960
1961 /**
1962 * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory
1963 * @dip: the directory
1964 *
1965 * Dealloc all on-disk directory leaves to FREEMETA state
1966 * Change on-disk inode type to "regular file"
1967 *
1968 * Returns: errno
1969 */
1970
1971 int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip)
1972 {
1973 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1974 struct buffer_head *bh;
1975 int error;
1976
1977 /* Dealloc on-disk leaves to FREEMETA state */
1978 error = foreach_leaf(dip, leaf_dealloc, NULL);
1979 if (error)
1980 return error;
1981
1982 /* Make this a regular file in case we crash.
1983 (We don't want to free these blocks a second time.) */
1984
1985 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1986 if (error)
1987 return error;
1988
1989 error = gfs2_meta_inode_buffer(dip, &bh);
1990 if (!error) {
1991 gfs2_trans_add_bh(dip->i_gl, bh, 1);
1992 ((struct gfs2_dinode *)bh->b_data)->di_mode =
1993 cpu_to_be32(S_IFREG);
1994 brelse(bh);
1995 }
1996
1997 gfs2_trans_end(sdp);
1998
1999 return error;
2000 }
2001
2002 /**
2003 * gfs2_diradd_alloc_required - find if adding entry will require an allocation
2004 * @ip: the file being written to
2005 * @filname: the filename that's going to be added
2006 *
2007 * Returns: 1 if alloc required, 0 if not, -ve on error
2008 */
2009
2010 int gfs2_diradd_alloc_required(struct inode *inode, const struct qstr *name)
2011 {
2012 struct gfs2_dirent *dent;
2013 struct buffer_head *bh;
2014
2015 dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, &bh);
2016 if (!dent) {
2017 return 1;
2018 }
2019 if (IS_ERR(dent))
2020 return PTR_ERR(dent);
2021 brelse(bh);
2022 return 0;
2023 }
2024
mah project overlay