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.
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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
58 #include <linux/slab.h>
59 #include <linux/spinlock.h>
60 #include <linux/buffer_head.h>
61 #include <linux/sort.h>
62 #include <linux/gfs2_ondisk.h>
63 #include <linux/crc32.h>
64 #include <linux/vmalloc.h>
65 #include <linux/bio.h>
79 #define IS_LEAF 1 /* Hashed (leaf) directory */
80 #define IS_DINODE 2 /* Linear (stuffed dinode block) directory */
82 #define MAX_RA_BLOCKS 32 /* max read-ahead blocks */
84 #define gfs2_disk_hash2offset(h) (((u64)(h)) >> 1)
85 #define gfs2_dir_offset2hash(p) ((u32)(((u64)(p)) << 1))
86 #define GFS2_HASH_INDEX_MASK 0xffffc000
87 #define GFS2_USE_HASH_FLAG 0x2000
89 struct qstr gfs2_qdot __read_mostly;
90 struct qstr gfs2_qdotdot __read_mostly;
92 typedef int (*gfs2_dscan_t)(const struct gfs2_dirent *dent,
93 const struct qstr *name, void *opaque);
95 int gfs2_dir_get_new_buffer(struct gfs2_inode *ip, u64 block,
96 struct buffer_head **bhp)
98 struct buffer_head *bh;
100 bh = gfs2_meta_new(ip->i_gl, block);
101 gfs2_trans_add_meta(ip->i_gl, bh);
102 gfs2_metatype_set(bh, GFS2_METATYPE_JD, GFS2_FORMAT_JD);
103 gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
108 static int gfs2_dir_get_existing_buffer(struct gfs2_inode *ip, u64 block,
109 struct buffer_head **bhp)
111 struct buffer_head *bh;
114 error = gfs2_meta_read(ip->i_gl, block, DIO_WAIT, 0, &bh);
117 if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), bh, GFS2_METATYPE_JD)) {
125 static int gfs2_dir_write_stuffed(struct gfs2_inode *ip, const char *buf,
126 unsigned int offset, unsigned int size)
128 struct buffer_head *dibh;
131 error = gfs2_meta_inode_buffer(ip, &dibh);
135 gfs2_trans_add_meta(ip->i_gl, dibh);
136 memcpy(dibh->b_data + offset + sizeof(struct gfs2_dinode), buf, size);
137 if (ip->i_inode.i_size < offset + size)
138 i_size_write(&ip->i_inode, offset + size);
139 ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode);
140 gfs2_dinode_out(ip, dibh->b_data);
150 * gfs2_dir_write_data - Write directory information to the inode
151 * @ip: The GFS2 inode
152 * @buf: The buffer containing information to be written
153 * @offset: The file offset to start writing at
154 * @size: The amount of data to write
156 * Returns: The number of bytes correctly written or error code
158 static int gfs2_dir_write_data(struct gfs2_inode *ip, const char *buf,
159 u64 offset, unsigned int size)
161 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
162 struct buffer_head *dibh;
173 if (gfs2_is_stuffed(ip) &&
174 offset + size <= sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode))
175 return gfs2_dir_write_stuffed(ip, buf, (unsigned int)offset,
178 if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
181 if (gfs2_is_stuffed(ip)) {
182 error = gfs2_unstuff_dinode(ip, NULL);
188 o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
190 while (copied < size) {
192 struct buffer_head *bh;
194 amount = size - copied;
195 if (amount > sdp->sd_sb.sb_bsize - o)
196 amount = sdp->sd_sb.sb_bsize - o;
200 error = gfs2_extent_map(&ip->i_inode, lblock, &new,
205 if (gfs2_assert_withdraw(sdp, dblock))
209 if (amount == sdp->sd_jbsize || new)
210 error = gfs2_dir_get_new_buffer(ip, dblock, &bh);
212 error = gfs2_dir_get_existing_buffer(ip, dblock, &bh);
217 gfs2_trans_add_meta(ip->i_gl, bh);
218 memcpy(bh->b_data + o, buf, amount);
227 o = sizeof(struct gfs2_meta_header);
231 error = gfs2_meta_inode_buffer(ip, &dibh);
235 if (ip->i_inode.i_size < offset + copied)
236 i_size_write(&ip->i_inode, offset + copied);
237 ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode);
239 gfs2_trans_add_meta(ip->i_gl, dibh);
240 gfs2_dinode_out(ip, dibh->b_data);
250 static int gfs2_dir_read_stuffed(struct gfs2_inode *ip, __be64 *buf,
253 struct buffer_head *dibh;
256 error = gfs2_meta_inode_buffer(ip, &dibh);
258 memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size);
262 return (error) ? error : size;
267 * gfs2_dir_read_data - Read a data from a directory inode
268 * @ip: The GFS2 Inode
269 * @buf: The buffer to place result into
270 * @size: Amount of data to transfer
272 * Returns: The amount of data actually copied or the error
274 static int gfs2_dir_read_data(struct gfs2_inode *ip, __be64 *buf,
277 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
284 if (gfs2_is_stuffed(ip))
285 return gfs2_dir_read_stuffed(ip, buf, size);
287 if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
291 o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
293 while (copied < size) {
295 struct buffer_head *bh;
298 amount = size - copied;
299 if (amount > sdp->sd_sb.sb_bsize - o)
300 amount = sdp->sd_sb.sb_bsize - o;
304 error = gfs2_extent_map(&ip->i_inode, lblock, &new,
306 if (error || !dblock)
309 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
311 error = gfs2_meta_read(ip->i_gl, dblock, DIO_WAIT, 0, &bh);
315 error = gfs2_metatype_check(sdp, bh, GFS2_METATYPE_JD);
322 memcpy(buf, bh->b_data + o, amount);
324 buf += (amount/sizeof(__be64));
327 o = sizeof(struct gfs2_meta_header);
332 return (copied) ? copied : error;
336 * gfs2_dir_get_hash_table - Get pointer to the dir hash table
337 * @ip: The inode in question
339 * Returns: The hash table or an error
342 static __be64 *gfs2_dir_get_hash_table(struct gfs2_inode *ip)
344 struct inode *inode = &ip->i_inode;
349 BUG_ON(!(ip->i_diskflags & GFS2_DIF_EXHASH));
351 hc = ip->i_hash_cache;
355 hsize = BIT(ip->i_depth);
356 hsize *= sizeof(__be64);
357 if (hsize != i_size_read(&ip->i_inode)) {
358 gfs2_consist_inode(ip);
359 return ERR_PTR(-EIO);
362 hc = kmalloc(hsize, GFP_NOFS | __GFP_NOWARN);
364 hc = __vmalloc(hsize, GFP_NOFS, PAGE_KERNEL);
367 return ERR_PTR(-ENOMEM);
369 ret = gfs2_dir_read_data(ip, hc, hsize);
375 spin_lock(&inode->i_lock);
376 if (likely(!ip->i_hash_cache)) {
377 ip->i_hash_cache = hc;
380 spin_unlock(&inode->i_lock);
383 return ip->i_hash_cache;
387 * gfs2_dir_hash_inval - Invalidate dir hash
388 * @ip: The directory inode
390 * Must be called with an exclusive glock, or during glock invalidation.
392 void gfs2_dir_hash_inval(struct gfs2_inode *ip)
396 spin_lock(&ip->i_inode.i_lock);
397 hc = ip->i_hash_cache;
398 ip->i_hash_cache = NULL;
399 spin_unlock(&ip->i_inode.i_lock);
404 static inline int gfs2_dirent_sentinel(const struct gfs2_dirent *dent)
406 return dent->de_inum.no_addr == 0 || dent->de_inum.no_formal_ino == 0;
409 static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent,
410 const struct qstr *name, int ret)
412 if (!gfs2_dirent_sentinel(dent) &&
413 be32_to_cpu(dent->de_hash) == name->hash &&
414 be16_to_cpu(dent->de_name_len) == name->len &&
415 memcmp(dent+1, name->name, name->len) == 0)
420 static int gfs2_dirent_find(const struct gfs2_dirent *dent,
421 const struct qstr *name,
424 return __gfs2_dirent_find(dent, name, 1);
427 static int gfs2_dirent_prev(const struct gfs2_dirent *dent,
428 const struct qstr *name,
431 return __gfs2_dirent_find(dent, name, 2);
435 * name->name holds ptr to start of block.
436 * name->len holds size of block.
438 static int gfs2_dirent_last(const struct gfs2_dirent *dent,
439 const struct qstr *name,
442 const char *start = name->name;
443 const char *end = (const char *)dent + be16_to_cpu(dent->de_rec_len);
444 if (name->len == (end - start))
449 /* Look for the dirent that contains the offset specified in data. Once we
450 * find that dirent, there must be space available there for the new dirent */
451 static int gfs2_dirent_find_offset(const struct gfs2_dirent *dent,
452 const struct qstr *name,
455 unsigned required = GFS2_DIRENT_SIZE(name->len);
456 unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
457 unsigned totlen = be16_to_cpu(dent->de_rec_len);
459 if (ptr < (void *)dent || ptr >= (void *)dent + totlen)
461 if (gfs2_dirent_sentinel(dent))
463 if (ptr < (void *)dent + actual)
465 if ((void *)dent + totlen >= ptr + required)
470 static int gfs2_dirent_find_space(const struct gfs2_dirent *dent,
471 const struct qstr *name,
474 unsigned required = GFS2_DIRENT_SIZE(name->len);
475 unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
476 unsigned totlen = be16_to_cpu(dent->de_rec_len);
478 if (gfs2_dirent_sentinel(dent))
480 if (totlen - actual >= required)
485 struct dirent_gather {
486 const struct gfs2_dirent **pdent;
490 static int gfs2_dirent_gather(const struct gfs2_dirent *dent,
491 const struct qstr *name,
494 struct dirent_gather *g = opaque;
495 if (!gfs2_dirent_sentinel(dent)) {
496 g->pdent[g->offset++] = dent;
502 * Other possible things to check:
503 * - Inode located within filesystem size (and on valid block)
504 * - Valid directory entry type
505 * Not sure how heavy-weight we want to make this... could also check
506 * hash is correct for example, but that would take a lot of extra time.
507 * For now the most important thing is to check that the various sizes
510 static int gfs2_check_dirent(struct gfs2_dirent *dent, unsigned int offset,
511 unsigned int size, unsigned int len, int first)
513 const char *msg = "gfs2_dirent too small";
514 if (unlikely(size < sizeof(struct gfs2_dirent)))
516 msg = "gfs2_dirent misaligned";
517 if (unlikely(offset & 0x7))
519 msg = "gfs2_dirent points beyond end of block";
520 if (unlikely(offset + size > len))
522 msg = "zero inode number";
523 if (unlikely(!first && gfs2_dirent_sentinel(dent)))
525 msg = "name length is greater than space in dirent";
526 if (!gfs2_dirent_sentinel(dent) &&
527 unlikely(sizeof(struct gfs2_dirent)+be16_to_cpu(dent->de_name_len) >
532 pr_warn("%s: %s (%s)\n",
533 __func__, msg, first ? "first in block" : "not first in block");
537 static int gfs2_dirent_offset(const void *buf)
539 const struct gfs2_meta_header *h = buf;
544 switch(be32_to_cpu(h->mh_type)) {
545 case GFS2_METATYPE_LF:
546 offset = sizeof(struct gfs2_leaf);
548 case GFS2_METATYPE_DI:
549 offset = sizeof(struct gfs2_dinode);
556 pr_warn("%s: wrong block type %u\n", __func__, be32_to_cpu(h->mh_type));
560 static struct gfs2_dirent *gfs2_dirent_scan(struct inode *inode, void *buf,
561 unsigned int len, gfs2_dscan_t scan,
562 const struct qstr *name,
565 struct gfs2_dirent *dent, *prev;
570 ret = gfs2_dirent_offset(buf);
577 size = be16_to_cpu(dent->de_rec_len);
578 if (gfs2_check_dirent(dent, offset, size, len, 1))
581 ret = scan(dent, name, opaque);
589 size = be16_to_cpu(dent->de_rec_len);
590 if (gfs2_check_dirent(dent, offset, size, len, 0))
600 return prev ? prev : dent;
607 gfs2_consist_inode(GFS2_I(inode));
608 return ERR_PTR(-EIO);
611 static int dirent_check_reclen(struct gfs2_inode *dip,
612 const struct gfs2_dirent *d, const void *end_p)
615 u16 rec_len = be16_to_cpu(d->de_rec_len);
617 if (unlikely(rec_len < sizeof(struct gfs2_dirent)))
625 gfs2_consist_inode(dip);
630 * dirent_next - Next dirent
631 * @dip: the directory
633 * @dent: Pointer to list of dirents
635 * Returns: 0 on success, error code otherwise
638 static int dirent_next(struct gfs2_inode *dip, struct buffer_head *bh,
639 struct gfs2_dirent **dent)
641 struct gfs2_dirent *cur = *dent, *tmp;
642 char *bh_end = bh->b_data + bh->b_size;
645 ret = dirent_check_reclen(dip, cur, bh_end);
649 tmp = (void *)cur + ret;
650 ret = dirent_check_reclen(dip, tmp, bh_end);
654 /* Only the first dent could ever have de_inum.no_addr == 0 */
655 if (gfs2_dirent_sentinel(tmp)) {
656 gfs2_consist_inode(dip);
665 * dirent_del - Delete a dirent
666 * @dip: The GFS2 inode
668 * @prev: The previous dirent
669 * @cur: The current dirent
673 static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh,
674 struct gfs2_dirent *prev, struct gfs2_dirent *cur)
676 u16 cur_rec_len, prev_rec_len;
678 if (gfs2_dirent_sentinel(cur)) {
679 gfs2_consist_inode(dip);
683 gfs2_trans_add_meta(dip->i_gl, bh);
685 /* If there is no prev entry, this is the first entry in the block.
686 The de_rec_len is already as big as it needs to be. Just zero
687 out the inode number and return. */
690 cur->de_inum.no_addr = 0;
691 cur->de_inum.no_formal_ino = 0;
695 /* Combine this dentry with the previous one. */
697 prev_rec_len = be16_to_cpu(prev->de_rec_len);
698 cur_rec_len = be16_to_cpu(cur->de_rec_len);
700 if ((char *)prev + prev_rec_len != (char *)cur)
701 gfs2_consist_inode(dip);
702 if ((char *)cur + cur_rec_len > bh->b_data + bh->b_size)
703 gfs2_consist_inode(dip);
705 prev_rec_len += cur_rec_len;
706 prev->de_rec_len = cpu_to_be16(prev_rec_len);
710 static struct gfs2_dirent *do_init_dirent(struct inode *inode,
711 struct gfs2_dirent *dent,
712 const struct qstr *name,
713 struct buffer_head *bh,
716 struct gfs2_inode *ip = GFS2_I(inode);
717 struct gfs2_dirent *ndent;
720 totlen = be16_to_cpu(dent->de_rec_len);
721 BUG_ON(offset + name->len > totlen);
722 gfs2_trans_add_meta(ip->i_gl, bh);
723 ndent = (struct gfs2_dirent *)((char *)dent + offset);
724 dent->de_rec_len = cpu_to_be16(offset);
725 gfs2_qstr2dirent(name, totlen - offset, ndent);
731 * Takes a dent from which to grab space as an argument. Returns the
732 * newly created dent.
734 static struct gfs2_dirent *gfs2_init_dirent(struct inode *inode,
735 struct gfs2_dirent *dent,
736 const struct qstr *name,
737 struct buffer_head *bh)
741 if (!gfs2_dirent_sentinel(dent))
742 offset = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
743 return do_init_dirent(inode, dent, name, bh, offset);
746 static struct gfs2_dirent *gfs2_dirent_split_alloc(struct inode *inode,
747 struct buffer_head *bh,
748 const struct qstr *name,
751 struct gfs2_dirent *dent;
752 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
753 gfs2_dirent_find_offset, name, ptr);
754 if (!dent || IS_ERR(dent))
756 return do_init_dirent(inode, dent, name, bh,
757 (unsigned)(ptr - (void *)dent));
760 static int get_leaf(struct gfs2_inode *dip, u64 leaf_no,
761 struct buffer_head **bhp)
765 error = gfs2_meta_read(dip->i_gl, leaf_no, DIO_WAIT, 0, bhp);
766 if (!error && gfs2_metatype_check(GFS2_SB(&dip->i_inode), *bhp, GFS2_METATYPE_LF)) {
767 /* pr_info("block num=%llu\n", leaf_no); */
775 * get_leaf_nr - Get a leaf number associated with the index
776 * @dip: The GFS2 inode
780 * Returns: 0 on success, error code otherwise
783 static int get_leaf_nr(struct gfs2_inode *dip, u32 index,
789 hash = gfs2_dir_get_hash_table(dip);
790 error = PTR_ERR_OR_ZERO(hash);
793 *leaf_out = be64_to_cpu(*(hash + index));
798 static int get_first_leaf(struct gfs2_inode *dip, u32 index,
799 struct buffer_head **bh_out)
804 error = get_leaf_nr(dip, index, &leaf_no);
806 error = get_leaf(dip, leaf_no, bh_out);
811 static struct gfs2_dirent *gfs2_dirent_search(struct inode *inode,
812 const struct qstr *name,
814 struct buffer_head **pbh)
816 struct buffer_head *bh;
817 struct gfs2_dirent *dent;
818 struct gfs2_inode *ip = GFS2_I(inode);
821 if (ip->i_diskflags & GFS2_DIF_EXHASH) {
822 struct gfs2_leaf *leaf;
823 unsigned int hsize = BIT(ip->i_depth);
826 if (hsize * sizeof(u64) != i_size_read(inode)) {
827 gfs2_consist_inode(ip);
828 return ERR_PTR(-EIO);
831 index = name->hash >> (32 - ip->i_depth);
832 error = get_first_leaf(ip, index, &bh);
834 return ERR_PTR(error);
836 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
840 leaf = (struct gfs2_leaf *)bh->b_data;
841 ln = be64_to_cpu(leaf->lf_next);
846 error = get_leaf(ip, ln, &bh);
849 return error ? ERR_PTR(error) : NULL;
853 error = gfs2_meta_inode_buffer(ip, &bh);
855 return ERR_PTR(error);
856 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, scan, name, NULL);
858 if (unlikely(dent == NULL || IS_ERR(dent))) {
866 static struct gfs2_leaf *new_leaf(struct inode *inode, struct buffer_head **pbh, u16 depth)
868 struct gfs2_inode *ip = GFS2_I(inode);
872 struct buffer_head *bh;
873 struct gfs2_leaf *leaf;
874 struct gfs2_dirent *dent;
875 struct qstr name = { .name = "" };
876 struct timespec tv = current_time(inode);
878 error = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL);
881 bh = gfs2_meta_new(ip->i_gl, bn);
885 gfs2_trans_add_unrevoke(GFS2_SB(inode), bn, 1);
886 gfs2_trans_add_meta(ip->i_gl, bh);
887 gfs2_metatype_set(bh, GFS2_METATYPE_LF, GFS2_FORMAT_LF);
888 leaf = (struct gfs2_leaf *)bh->b_data;
889 leaf->lf_depth = cpu_to_be16(depth);
890 leaf->lf_entries = 0;
891 leaf->lf_dirent_format = cpu_to_be32(GFS2_FORMAT_DE);
893 leaf->lf_inode = cpu_to_be64(ip->i_no_addr);
894 leaf->lf_dist = cpu_to_be32(1);
895 leaf->lf_nsec = cpu_to_be32(tv.tv_nsec);
896 leaf->lf_sec = cpu_to_be64(tv.tv_sec);
897 memset(leaf->lf_reserved2, 0, sizeof(leaf->lf_reserved2));
898 dent = (struct gfs2_dirent *)(leaf+1);
899 gfs2_qstr2dirent(&name, bh->b_size - sizeof(struct gfs2_leaf), dent);
905 * dir_make_exhash - Convert a stuffed directory into an ExHash directory
906 * @dip: The GFS2 inode
908 * Returns: 0 on success, error code otherwise
911 static int dir_make_exhash(struct inode *inode)
913 struct gfs2_inode *dip = GFS2_I(inode);
914 struct gfs2_sbd *sdp = GFS2_SB(inode);
915 struct gfs2_dirent *dent;
917 struct buffer_head *bh, *dibh;
918 struct gfs2_leaf *leaf;
925 error = gfs2_meta_inode_buffer(dip, &dibh);
929 /* Turn over a new leaf */
931 leaf = new_leaf(inode, &bh, 0);
936 gfs2_assert(sdp, dip->i_entries < BIT(16));
937 leaf->lf_entries = cpu_to_be16(dip->i_entries);
941 gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_leaf), dibh,
942 sizeof(struct gfs2_dinode));
944 /* Find last entry */
947 args.len = bh->b_size - sizeof(struct gfs2_dinode) +
948 sizeof(struct gfs2_leaf);
949 args.name = bh->b_data;
950 dent = gfs2_dirent_scan(&dip->i_inode, bh->b_data, bh->b_size,
951 gfs2_dirent_last, &args, NULL);
960 return PTR_ERR(dent);
963 /* Adjust the last dirent's record length
964 (Remember that dent still points to the last entry.) */
966 dent->de_rec_len = cpu_to_be16(be16_to_cpu(dent->de_rec_len) +
967 sizeof(struct gfs2_dinode) -
968 sizeof(struct gfs2_leaf));
972 /* We're done with the new leaf block, now setup the new
975 gfs2_trans_add_meta(dip->i_gl, dibh);
976 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
978 lp = (__be64 *)(dibh->b_data + sizeof(struct gfs2_dinode));
980 for (x = sdp->sd_hash_ptrs; x--; lp++)
981 *lp = cpu_to_be64(bn);
983 i_size_write(inode, sdp->sd_sb.sb_bsize / 2);
984 gfs2_add_inode_blocks(&dip->i_inode, 1);
985 dip->i_diskflags |= GFS2_DIF_EXHASH;
987 for (x = sdp->sd_hash_ptrs, y = -1; x; x >>= 1, y++) ;
990 gfs2_dinode_out(dip, dibh->b_data);
998 * dir_split_leaf - Split a leaf block into two
999 * @dip: The GFS2 inode
1003 * Returns: 0 on success, error code on failure
1006 static int dir_split_leaf(struct inode *inode, const struct qstr *name)
1008 struct gfs2_inode *dip = GFS2_I(inode);
1009 struct buffer_head *nbh, *obh, *dibh;
1010 struct gfs2_leaf *nleaf, *oleaf;
1011 struct gfs2_dirent *dent = NULL, *prev = NULL, *next = NULL, *new;
1012 u32 start, len, half_len, divider;
1019 index = name->hash >> (32 - dip->i_depth);
1020 error = get_leaf_nr(dip, index, &leaf_no);
1024 /* Get the old leaf block */
1025 error = get_leaf(dip, leaf_no, &obh);
1029 oleaf = (struct gfs2_leaf *)obh->b_data;
1030 if (dip->i_depth == be16_to_cpu(oleaf->lf_depth)) {
1032 return 1; /* can't split */
1035 gfs2_trans_add_meta(dip->i_gl, obh);
1037 nleaf = new_leaf(inode, &nbh, be16_to_cpu(oleaf->lf_depth) + 1);
1042 bn = nbh->b_blocknr;
1044 /* Compute the start and len of leaf pointers in the hash table. */
1045 len = BIT(dip->i_depth - be16_to_cpu(oleaf->lf_depth));
1046 half_len = len >> 1;
1048 pr_warn("i_depth %u lf_depth %u index %u\n",
1049 dip->i_depth, be16_to_cpu(oleaf->lf_depth), index);
1050 gfs2_consist_inode(dip);
1055 start = (index & ~(len - 1));
1057 /* Change the pointers.
1058 Don't bother distinguishing stuffed from non-stuffed.
1059 This code is complicated enough already. */
1060 lp = kmalloc(half_len * sizeof(__be64), GFP_NOFS);
1066 /* Change the pointers */
1067 for (x = 0; x < half_len; x++)
1068 lp[x] = cpu_to_be64(bn);
1070 gfs2_dir_hash_inval(dip);
1072 error = gfs2_dir_write_data(dip, (char *)lp, start * sizeof(u64),
1073 half_len * sizeof(u64));
1074 if (error != half_len * sizeof(u64)) {
1082 /* Compute the divider */
1083 divider = (start + half_len) << (32 - dip->i_depth);
1085 /* Copy the entries */
1086 dent = (struct gfs2_dirent *)(obh->b_data + sizeof(struct gfs2_leaf));
1090 if (dirent_next(dip, obh, &next))
1093 if (!gfs2_dirent_sentinel(dent) &&
1094 be32_to_cpu(dent->de_hash) < divider) {
1096 void *ptr = ((char *)dent - obh->b_data) + nbh->b_data;
1097 str.name = (char*)(dent+1);
1098 str.len = be16_to_cpu(dent->de_name_len);
1099 str.hash = be32_to_cpu(dent->de_hash);
1100 new = gfs2_dirent_split_alloc(inode, nbh, &str, ptr);
1102 error = PTR_ERR(new);
1106 new->de_inum = dent->de_inum; /* No endian worries */
1107 new->de_type = dent->de_type; /* No endian worries */
1108 be16_add_cpu(&nleaf->lf_entries, 1);
1110 dirent_del(dip, obh, prev, dent);
1112 if (!oleaf->lf_entries)
1113 gfs2_consist_inode(dip);
1114 be16_add_cpu(&oleaf->lf_entries, -1);
1126 oleaf->lf_depth = nleaf->lf_depth;
1128 error = gfs2_meta_inode_buffer(dip, &dibh);
1129 if (!gfs2_assert_withdraw(GFS2_SB(&dip->i_inode), !error)) {
1130 gfs2_trans_add_meta(dip->i_gl, dibh);
1131 gfs2_add_inode_blocks(&dip->i_inode, 1);
1132 gfs2_dinode_out(dip, dibh->b_data);
1151 * dir_double_exhash - Double size of ExHash table
1152 * @dip: The GFS2 dinode
1154 * Returns: 0 on success, error code on failure
1157 static int dir_double_exhash(struct gfs2_inode *dip)
1159 struct buffer_head *dibh;
1167 hsize = BIT(dip->i_depth);
1168 hsize_bytes = hsize * sizeof(__be64);
1170 hc = gfs2_dir_get_hash_table(dip);
1174 hc2 = kmalloc(hsize_bytes * 2, GFP_NOFS | __GFP_NOWARN);
1176 hc2 = __vmalloc(hsize_bytes * 2, GFP_NOFS, PAGE_KERNEL);
1182 error = gfs2_meta_inode_buffer(dip, &dibh);
1186 for (x = 0; x < hsize; x++) {
1192 error = gfs2_dir_write_data(dip, (char *)hc2, 0, hsize_bytes * 2);
1193 if (error != (hsize_bytes * 2))
1196 gfs2_dir_hash_inval(dip);
1197 dip->i_hash_cache = hc2;
1199 gfs2_dinode_out(dip, dibh->b_data);
1204 /* Replace original hash table & size */
1205 gfs2_dir_write_data(dip, (char *)hc, 0, hsize_bytes);
1206 i_size_write(&dip->i_inode, hsize_bytes);
1207 gfs2_dinode_out(dip, dibh->b_data);
1215 * compare_dents - compare directory entries by hash value
1219 * When comparing the hash entries of @a to @b:
1225 static int compare_dents(const void *a, const void *b)
1227 const struct gfs2_dirent *dent_a, *dent_b;
1231 dent_a = *(const struct gfs2_dirent **)a;
1232 hash_a = dent_a->de_cookie;
1234 dent_b = *(const struct gfs2_dirent **)b;
1235 hash_b = dent_b->de_cookie;
1237 if (hash_a > hash_b)
1239 else if (hash_a < hash_b)
1242 unsigned int len_a = be16_to_cpu(dent_a->de_name_len);
1243 unsigned int len_b = be16_to_cpu(dent_b->de_name_len);
1247 else if (len_a < len_b)
1250 ret = memcmp(dent_a + 1, dent_b + 1, len_a);
1257 * do_filldir_main - read out directory entries
1258 * @dip: The GFS2 inode
1259 * @ctx: what to feed the entries to
1260 * @darr: an array of struct gfs2_dirent pointers to read
1261 * @entries: the number of entries in darr
1262 * @copied: pointer to int that's non-zero if a entry has been copied out
1264 * Jump through some hoops to make sure that if there are hash collsions,
1265 * they are read out at the beginning of a buffer. We want to minimize
1266 * the possibility that they will fall into different readdir buffers or
1267 * that someone will want to seek to that location.
1269 * Returns: errno, >0 if the actor tells you to stop
1272 static int do_filldir_main(struct gfs2_inode *dip, struct dir_context *ctx,
1273 struct gfs2_dirent **darr, u32 entries,
1274 u32 sort_start, int *copied)
1276 const struct gfs2_dirent *dent, *dent_next;
1281 if (sort_start < entries)
1282 sort(&darr[sort_start], entries - sort_start,
1283 sizeof(struct gfs2_dirent *), compare_dents, NULL);
1285 dent_next = darr[0];
1286 off_next = dent_next->de_cookie;
1288 for (x = 0, y = 1; x < entries; x++, y++) {
1293 dent_next = darr[y];
1294 off_next = dent_next->de_cookie;
1300 if (off_next == off) {
1301 if (*copied && !run)
1312 if (!dir_emit(ctx, (const char *)(dent + 1),
1313 be16_to_cpu(dent->de_name_len),
1314 be64_to_cpu(dent->de_inum.no_addr),
1315 be16_to_cpu(dent->de_type)))
1321 /* Increment the ctx->pos by one, so the next time we come into the
1322 do_filldir fxn, we get the next entry instead of the last one in the
1330 static void *gfs2_alloc_sort_buffer(unsigned size)
1334 if (size < KMALLOC_MAX_SIZE)
1335 ptr = kmalloc(size, GFP_NOFS | __GFP_NOWARN);
1337 ptr = __vmalloc(size, GFP_NOFS, PAGE_KERNEL);
1342 static int gfs2_set_cookies(struct gfs2_sbd *sdp, struct buffer_head *bh,
1343 unsigned leaf_nr, struct gfs2_dirent **darr,
1349 for (i = 0; i < entries; i++) {
1352 darr[i]->de_cookie = be32_to_cpu(darr[i]->de_hash);
1353 darr[i]->de_cookie = gfs2_disk_hash2offset(darr[i]->de_cookie);
1355 if (!sdp->sd_args.ar_loccookie)
1357 offset = (char *)(darr[i]) -
1358 (bh->b_data + gfs2_dirent_offset(bh->b_data));
1359 offset /= GFS2_MIN_DIRENT_SIZE;
1360 offset += leaf_nr * sdp->sd_max_dents_per_leaf;
1361 if (offset >= GFS2_USE_HASH_FLAG ||
1362 leaf_nr >= GFS2_USE_HASH_FLAG) {
1363 darr[i]->de_cookie |= GFS2_USE_HASH_FLAG;
1368 darr[i]->de_cookie &= GFS2_HASH_INDEX_MASK;
1369 darr[i]->de_cookie |= offset;
1375 static int gfs2_dir_read_leaf(struct inode *inode, struct dir_context *ctx,
1376 int *copied, unsigned *depth,
1379 struct gfs2_inode *ip = GFS2_I(inode);
1380 struct gfs2_sbd *sdp = GFS2_SB(inode);
1381 struct buffer_head *bh;
1382 struct gfs2_leaf *lf;
1383 unsigned entries = 0, entries2 = 0;
1384 unsigned leaves = 0, leaf = 0, offset, sort_offset;
1385 struct gfs2_dirent **darr, *dent;
1386 struct dirent_gather g;
1387 struct buffer_head **larr;
1388 int error, i, need_sort = 0, sort_id;
1392 error = get_leaf(ip, lfn, &bh);
1395 lf = (struct gfs2_leaf *)bh->b_data;
1397 *depth = be16_to_cpu(lf->lf_depth);
1398 entries += be16_to_cpu(lf->lf_entries);
1400 lfn = be64_to_cpu(lf->lf_next);
1404 if (*depth < GFS2_DIR_MAX_DEPTH || !sdp->sd_args.ar_loccookie) {
1414 * The extra 99 entries are not normally used, but are a buffer
1415 * zone in case the number of entries in the leaf is corrupt.
1416 * 99 is the maximum number of entries that can fit in a single
1419 larr = gfs2_alloc_sort_buffer((leaves + entries + 99) * sizeof(void *));
1422 darr = (struct gfs2_dirent **)(larr + leaves);
1423 g.pdent = (const struct gfs2_dirent **)darr;
1428 error = get_leaf(ip, lfn, &bh);
1431 lf = (struct gfs2_leaf *)bh->b_data;
1432 lfn = be64_to_cpu(lf->lf_next);
1433 if (lf->lf_entries) {
1435 entries2 += be16_to_cpu(lf->lf_entries);
1436 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
1437 gfs2_dirent_gather, NULL, &g);
1438 error = PTR_ERR(dent);
1441 if (entries2 != g.offset) {
1442 fs_warn(sdp, "Number of entries corrupt in dir "
1443 "leaf %llu, entries2 (%u) != "
1445 (unsigned long long)bh->b_blocknr,
1446 entries2, g.offset);
1452 sort_id = gfs2_set_cookies(sdp, bh, leaf, &darr[offset],
1453 be16_to_cpu(lf->lf_entries));
1454 if (!need_sort && sort_id >= 0) {
1456 sort_offset = offset + sort_id;
1460 larr[leaf++] = NULL;
1465 BUG_ON(entries2 != entries);
1466 error = do_filldir_main(ip, ctx, darr, entries, need_sort ?
1467 sort_offset : entries, copied);
1469 for(i = 0; i < leaf; i++)
1478 * gfs2_dir_readahead - Issue read-ahead requests for leaf blocks.
1480 * Note: we can't calculate each index like dir_e_read can because we don't
1481 * have the leaf, and therefore we don't have the depth, and therefore we
1482 * don't have the length. So we have to just read enough ahead to make up
1483 * for the loss of information.
1485 static void gfs2_dir_readahead(struct inode *inode, unsigned hsize, u32 index,
1486 struct file_ra_state *f_ra)
1488 struct gfs2_inode *ip = GFS2_I(inode);
1489 struct gfs2_glock *gl = ip->i_gl;
1490 struct buffer_head *bh;
1491 u64 blocknr = 0, last;
1494 /* First check if we've already read-ahead for the whole range. */
1495 if (index + MAX_RA_BLOCKS < f_ra->start)
1498 f_ra->start = max((pgoff_t)index, f_ra->start);
1499 for (count = 0; count < MAX_RA_BLOCKS; count++) {
1500 if (f_ra->start >= hsize) /* if exceeded the hash table */
1504 blocknr = be64_to_cpu(ip->i_hash_cache[f_ra->start]);
1506 if (blocknr == last)
1509 bh = gfs2_getbuf(gl, blocknr, 1);
1510 if (trylock_buffer(bh)) {
1511 if (buffer_uptodate(bh)) {
1516 bh->b_end_io = end_buffer_read_sync;
1517 submit_bh(REQ_OP_READ, REQ_RAHEAD | REQ_META, bh);
1525 * dir_e_read - Reads the entries from a directory into a filldir buffer
1526 * @dip: dinode pointer
1527 * @ctx: actor to feed the entries to
1532 static int dir_e_read(struct inode *inode, struct dir_context *ctx,
1533 struct file_ra_state *f_ra)
1535 struct gfs2_inode *dip = GFS2_I(inode);
1543 hsize = BIT(dip->i_depth);
1544 hash = gfs2_dir_offset2hash(ctx->pos);
1545 index = hash >> (32 - dip->i_depth);
1547 if (dip->i_hash_cache == NULL)
1549 lp = gfs2_dir_get_hash_table(dip);
1553 gfs2_dir_readahead(inode, hsize, index, f_ra);
1555 while (index < hsize) {
1556 error = gfs2_dir_read_leaf(inode, ctx,
1558 be64_to_cpu(lp[index]));
1562 len = BIT(dip->i_depth - depth);
1563 index = (index & ~(len - 1)) + len;
1571 int gfs2_dir_read(struct inode *inode, struct dir_context *ctx,
1572 struct file_ra_state *f_ra)
1574 struct gfs2_inode *dip = GFS2_I(inode);
1575 struct gfs2_sbd *sdp = GFS2_SB(inode);
1576 struct dirent_gather g;
1577 struct gfs2_dirent **darr, *dent;
1578 struct buffer_head *dibh;
1582 if (!dip->i_entries)
1585 if (dip->i_diskflags & GFS2_DIF_EXHASH)
1586 return dir_e_read(inode, ctx, f_ra);
1588 if (!gfs2_is_stuffed(dip)) {
1589 gfs2_consist_inode(dip);
1593 error = gfs2_meta_inode_buffer(dip, &dibh);
1598 /* 96 is max number of dirents which can be stuffed into an inode */
1599 darr = kmalloc(96 * sizeof(struct gfs2_dirent *), GFP_NOFS);
1601 g.pdent = (const struct gfs2_dirent **)darr;
1603 dent = gfs2_dirent_scan(inode, dibh->b_data, dibh->b_size,
1604 gfs2_dirent_gather, NULL, &g);
1606 error = PTR_ERR(dent);
1609 if (dip->i_entries != g.offset) {
1610 fs_warn(sdp, "Number of entries corrupt in dir %llu, "
1611 "ip->i_entries (%u) != g.offset (%u)\n",
1612 (unsigned long long)dip->i_no_addr,
1618 gfs2_set_cookies(sdp, dibh, 0, darr, dip->i_entries);
1619 error = do_filldir_main(dip, ctx, darr,
1620 dip->i_entries, 0, &copied);
1634 * gfs2_dir_search - Search a directory
1635 * @dip: The GFS2 dir inode
1636 * @name: The name we are looking up
1637 * @fail_on_exist: Fail if the name exists rather than looking it up
1639 * This routine searches a directory for a file or another directory.
1640 * Assumes a glock is held on dip.
1645 struct inode *gfs2_dir_search(struct inode *dir, const struct qstr *name,
1648 struct buffer_head *bh;
1649 struct gfs2_dirent *dent;
1650 u64 addr, formal_ino;
1653 dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
1655 struct inode *inode;
1659 return ERR_CAST(dent);
1660 dtype = be16_to_cpu(dent->de_type);
1661 rahead = be16_to_cpu(dent->de_rahead);
1662 addr = be64_to_cpu(dent->de_inum.no_addr);
1663 formal_ino = be64_to_cpu(dent->de_inum.no_formal_ino);
1666 return ERR_PTR(-EEXIST);
1667 inode = gfs2_inode_lookup(dir->i_sb, dtype, addr, formal_ino,
1668 GFS2_BLKST_FREE /* ignore */);
1670 GFS2_I(inode)->i_rahead = rahead;
1673 return ERR_PTR(-ENOENT);
1676 int gfs2_dir_check(struct inode *dir, const struct qstr *name,
1677 const struct gfs2_inode *ip)
1679 struct buffer_head *bh;
1680 struct gfs2_dirent *dent;
1683 dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
1686 return PTR_ERR(dent);
1688 if (be64_to_cpu(dent->de_inum.no_addr) != ip->i_no_addr)
1690 if (be64_to_cpu(dent->de_inum.no_formal_ino) !=
1691 ip->i_no_formal_ino)
1693 if (unlikely(IF2DT(ip->i_inode.i_mode) !=
1694 be16_to_cpu(dent->de_type))) {
1695 gfs2_consist_inode(GFS2_I(dir));
1708 * dir_new_leaf - Add a new leaf onto hash chain
1709 * @inode: The directory
1710 * @name: The name we are adding
1712 * This adds a new dir leaf onto an existing leaf when there is not
1713 * enough space to add a new dir entry. This is a last resort after
1714 * we've expanded the hash table to max size and also split existing
1715 * leaf blocks, so it will only occur for very large directories.
1717 * The dist parameter is set to 1 for leaf blocks directly attached
1718 * to the hash table, 2 for one layer of indirection, 3 for two layers
1719 * etc. We are thus able to tell the difference between an old leaf
1720 * with dist set to zero (i.e. "don't know") and a new one where we
1721 * set this information for debug/fsck purposes.
1723 * Returns: 0 on success, or -ve on error
1726 static int dir_new_leaf(struct inode *inode, const struct qstr *name)
1728 struct buffer_head *bh, *obh;
1729 struct gfs2_inode *ip = GFS2_I(inode);
1730 struct gfs2_leaf *leaf, *oleaf;
1736 index = name->hash >> (32 - ip->i_depth);
1737 error = get_first_leaf(ip, index, &obh);
1742 oleaf = (struct gfs2_leaf *)obh->b_data;
1743 bn = be64_to_cpu(oleaf->lf_next);
1747 error = get_leaf(ip, bn, &obh);
1752 gfs2_trans_add_meta(ip->i_gl, obh);
1754 leaf = new_leaf(inode, &bh, be16_to_cpu(oleaf->lf_depth));
1759 leaf->lf_dist = cpu_to_be32(dist);
1760 oleaf->lf_next = cpu_to_be64(bh->b_blocknr);
1764 error = gfs2_meta_inode_buffer(ip, &bh);
1767 gfs2_trans_add_meta(ip->i_gl, bh);
1768 gfs2_add_inode_blocks(&ip->i_inode, 1);
1769 gfs2_dinode_out(ip, bh->b_data);
1774 static u16 gfs2_inode_ra_len(const struct gfs2_inode *ip)
1776 u64 where = ip->i_no_addr + 1;
1777 if (ip->i_eattr == where)
1783 * gfs2_dir_add - Add new filename into directory
1784 * @inode: The directory inode
1785 * @name: The new name
1786 * @nip: The GFS2 inode to be linked in to the directory
1787 * @da: The directory addition info
1789 * If the call to gfs2_diradd_alloc_required resulted in there being
1790 * no need to allocate any new directory blocks, then it will contain
1791 * a pointer to the directory entry and the bh in which it resides. We
1792 * can use that without having to repeat the search. If there was no
1793 * free space, then we must now create more space.
1795 * Returns: 0 on success, error code on failure
1798 int gfs2_dir_add(struct inode *inode, const struct qstr *name,
1799 const struct gfs2_inode *nip, struct gfs2_diradd *da)
1801 struct gfs2_inode *ip = GFS2_I(inode);
1802 struct buffer_head *bh = da->bh;
1803 struct gfs2_dirent *dent = da->dent;
1805 struct gfs2_leaf *leaf;
1809 if (da->bh == NULL) {
1810 dent = gfs2_dirent_search(inode, name,
1811 gfs2_dirent_find_space, &bh);
1815 return PTR_ERR(dent);
1816 dent = gfs2_init_dirent(inode, dent, name, bh);
1817 gfs2_inum_out(nip, dent);
1818 dent->de_type = cpu_to_be16(IF2DT(nip->i_inode.i_mode));
1819 dent->de_rahead = cpu_to_be16(gfs2_inode_ra_len(nip));
1820 tv = current_time(&ip->i_inode);
1821 if (ip->i_diskflags & GFS2_DIF_EXHASH) {
1822 leaf = (struct gfs2_leaf *)bh->b_data;
1823 be16_add_cpu(&leaf->lf_entries, 1);
1824 leaf->lf_nsec = cpu_to_be32(tv.tv_nsec);
1825 leaf->lf_sec = cpu_to_be64(tv.tv_sec);
1831 ip->i_inode.i_mtime = ip->i_inode.i_ctime = tv;
1832 if (S_ISDIR(nip->i_inode.i_mode))
1833 inc_nlink(&ip->i_inode);
1834 mark_inode_dirty(inode);
1838 if (!(ip->i_diskflags & GFS2_DIF_EXHASH)) {
1839 error = dir_make_exhash(inode);
1844 error = dir_split_leaf(inode, name);
1849 if (ip->i_depth < GFS2_DIR_MAX_DEPTH) {
1850 error = dir_double_exhash(ip);
1853 error = dir_split_leaf(inode, name);
1859 error = dir_new_leaf(inode, name);
1870 * gfs2_dir_del - Delete a directory entry
1871 * @dip: The GFS2 inode
1872 * @filename: The filename
1874 * Returns: 0 on success, error code on failure
1877 int gfs2_dir_del(struct gfs2_inode *dip, const struct dentry *dentry)
1879 const struct qstr *name = &dentry->d_name;
1880 struct gfs2_dirent *dent, *prev = NULL;
1881 struct buffer_head *bh;
1882 struct timespec tv = current_time(&dip->i_inode);
1884 /* Returns _either_ the entry (if its first in block) or the
1885 previous entry otherwise */
1886 dent = gfs2_dirent_search(&dip->i_inode, name, gfs2_dirent_prev, &bh);
1888 gfs2_consist_inode(dip);
1892 gfs2_consist_inode(dip);
1893 return PTR_ERR(dent);
1895 /* If not first in block, adjust pointers accordingly */
1896 if (gfs2_dirent_find(dent, name, NULL) == 0) {
1898 dent = (struct gfs2_dirent *)((char *)dent + be16_to_cpu(prev->de_rec_len));
1901 dirent_del(dip, bh, prev, dent);
1902 if (dip->i_diskflags & GFS2_DIF_EXHASH) {
1903 struct gfs2_leaf *leaf = (struct gfs2_leaf *)bh->b_data;
1904 u16 entries = be16_to_cpu(leaf->lf_entries);
1906 gfs2_consist_inode(dip);
1907 leaf->lf_entries = cpu_to_be16(--entries);
1908 leaf->lf_nsec = cpu_to_be32(tv.tv_nsec);
1909 leaf->lf_sec = cpu_to_be64(tv.tv_sec);
1913 if (!dip->i_entries)
1914 gfs2_consist_inode(dip);
1916 dip->i_inode.i_mtime = dip->i_inode.i_ctime = tv;
1917 if (d_is_dir(dentry))
1918 drop_nlink(&dip->i_inode);
1919 mark_inode_dirty(&dip->i_inode);
1925 * gfs2_dir_mvino - Change inode number of directory entry
1926 * @dip: The GFS2 inode
1930 * This routine changes the inode number of a directory entry. It's used
1931 * by rename to change ".." when a directory is moved.
1932 * Assumes a glock is held on dvp.
1937 int gfs2_dir_mvino(struct gfs2_inode *dip, const struct qstr *filename,
1938 const struct gfs2_inode *nip, unsigned int new_type)
1940 struct buffer_head *bh;
1941 struct gfs2_dirent *dent;
1944 dent = gfs2_dirent_search(&dip->i_inode, filename, gfs2_dirent_find, &bh);
1946 gfs2_consist_inode(dip);
1950 return PTR_ERR(dent);
1952 gfs2_trans_add_meta(dip->i_gl, bh);
1953 gfs2_inum_out(nip, dent);
1954 dent->de_type = cpu_to_be16(new_type);
1956 if (dip->i_diskflags & GFS2_DIF_EXHASH) {
1958 error = gfs2_meta_inode_buffer(dip, &bh);
1961 gfs2_trans_add_meta(dip->i_gl, bh);
1964 dip->i_inode.i_mtime = dip->i_inode.i_ctime = current_time(&dip->i_inode);
1965 gfs2_dinode_out(dip, bh->b_data);
1971 * leaf_dealloc - Deallocate a directory leaf
1972 * @dip: the directory
1973 * @index: the hash table offset in the directory
1974 * @len: the number of pointers to this leaf
1975 * @leaf_no: the leaf number
1976 * @leaf_bh: buffer_head for the starting leaf
1977 * last_dealloc: 1 if this is the final dealloc for the leaf, else 0
1982 static int leaf_dealloc(struct gfs2_inode *dip, u32 index, u32 len,
1983 u64 leaf_no, struct buffer_head *leaf_bh,
1986 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1987 struct gfs2_leaf *tmp_leaf;
1988 struct gfs2_rgrp_list rlist;
1989 struct buffer_head *bh, *dibh;
1991 unsigned int rg_blocks = 0, l_blocks = 0;
1993 unsigned int x, size = len * sizeof(u64);
1996 error = gfs2_rindex_update(sdp);
2000 memset(&rlist, 0, sizeof(struct gfs2_rgrp_list));
2002 ht = kzalloc(size, GFP_NOFS | __GFP_NOWARN);
2004 ht = __vmalloc(size, GFP_NOFS | __GFP_NOWARN | __GFP_ZERO,
2009 error = gfs2_quota_hold(dip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
2013 /* Count the number of leaves */
2016 for (blk = leaf_no; blk; blk = nblk) {
2017 if (blk != leaf_no) {
2018 error = get_leaf(dip, blk, &bh);
2022 tmp_leaf = (struct gfs2_leaf *)bh->b_data;
2023 nblk = be64_to_cpu(tmp_leaf->lf_next);
2027 gfs2_rlist_add(dip, &rlist, blk);
2031 gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE);
2033 for (x = 0; x < rlist.rl_rgrps; x++) {
2034 struct gfs2_rgrpd *rgd;
2035 rgd = rlist.rl_ghs[x].gh_gl->gl_object;
2036 rg_blocks += rgd->rd_length;
2039 error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs);
2043 error = gfs2_trans_begin(sdp,
2044 rg_blocks + (DIV_ROUND_UP(size, sdp->sd_jbsize) + 1) +
2045 RES_DINODE + RES_STATFS + RES_QUOTA, l_blocks);
2047 goto out_rg_gunlock;
2051 for (blk = leaf_no; blk; blk = nblk) {
2052 if (blk != leaf_no) {
2053 error = get_leaf(dip, blk, &bh);
2057 tmp_leaf = (struct gfs2_leaf *)bh->b_data;
2058 nblk = be64_to_cpu(tmp_leaf->lf_next);
2062 gfs2_free_meta(dip, blk, 1);
2063 gfs2_add_inode_blocks(&dip->i_inode, -1);
2066 error = gfs2_dir_write_data(dip, ht, index * sizeof(u64), size);
2067 if (error != size) {
2073 error = gfs2_meta_inode_buffer(dip, &dibh);
2077 gfs2_trans_add_meta(dip->i_gl, dibh);
2078 /* On the last dealloc, make this a regular file in case we crash.
2079 (We don't want to free these blocks a second time.) */
2081 dip->i_inode.i_mode = S_IFREG;
2082 gfs2_dinode_out(dip, dibh->b_data);
2086 gfs2_trans_end(sdp);
2088 gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
2090 gfs2_rlist_free(&rlist);
2091 gfs2_quota_unhold(dip);
2098 * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory
2099 * @dip: the directory
2101 * Dealloc all on-disk directory leaves to FREEMETA state
2102 * Change on-disk inode type to "regular file"
2107 int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip)
2109 struct buffer_head *bh;
2110 struct gfs2_leaf *leaf;
2112 u32 index = 0, next_index;
2115 int error = 0, last;
2117 hsize = BIT(dip->i_depth);
2119 lp = gfs2_dir_get_hash_table(dip);
2123 while (index < hsize) {
2124 leaf_no = be64_to_cpu(lp[index]);
2126 error = get_leaf(dip, leaf_no, &bh);
2129 leaf = (struct gfs2_leaf *)bh->b_data;
2130 len = BIT(dip->i_depth - be16_to_cpu(leaf->lf_depth));
2132 next_index = (index & ~(len - 1)) + len;
2133 last = ((next_index >= hsize) ? 1 : 0);
2134 error = leaf_dealloc(dip, index, len, leaf_no, bh,
2144 if (index != hsize) {
2145 gfs2_consist_inode(dip);
2155 * gfs2_diradd_alloc_required - find if adding entry will require an allocation
2156 * @ip: the file being written to
2157 * @filname: the filename that's going to be added
2158 * @da: The structure to return dir alloc info
2160 * Returns: 0 if ok, -ve on error
2163 int gfs2_diradd_alloc_required(struct inode *inode, const struct qstr *name,
2164 struct gfs2_diradd *da)
2166 struct gfs2_inode *ip = GFS2_I(inode);
2167 struct gfs2_sbd *sdp = GFS2_SB(inode);
2168 const unsigned int extra = sizeof(struct gfs2_dinode) - sizeof(struct gfs2_leaf);
2169 struct gfs2_dirent *dent;
2170 struct buffer_head *bh;
2176 dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, &bh);
2178 da->nr_blocks = sdp->sd_max_dirres;
2179 if (!(ip->i_diskflags & GFS2_DIF_EXHASH) &&
2180 (GFS2_DIRENT_SIZE(name->len) < extra))
2185 return PTR_ERR(dent);