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>
78 #define IS_LEAF 1 /* Hashed (leaf) directory */
79 #define IS_DINODE 2 /* Linear (stuffed dinode block) directory */
81 #define MAX_RA_BLOCKS 32 /* max read-ahead blocks */
83 #define gfs2_disk_hash2offset(h) (((u64)(h)) >> 1)
84 #define gfs2_dir_offset2hash(p) ((u32)(((u64)(p)) << 1))
85 #define GFS2_HASH_INDEX_MASK 0xffffc000
86 #define GFS2_USE_HASH_FLAG 0x2000
88 struct qstr gfs2_qdot __read_mostly;
89 struct qstr gfs2_qdotdot __read_mostly;
91 typedef int (*gfs2_dscan_t)(const struct gfs2_dirent *dent,
92 const struct qstr *name, void *opaque);
94 int gfs2_dir_get_new_buffer(struct gfs2_inode *ip, u64 block,
95 struct buffer_head **bhp)
97 struct buffer_head *bh;
99 bh = gfs2_meta_new(ip->i_gl, block);
100 gfs2_trans_add_meta(ip->i_gl, bh);
101 gfs2_metatype_set(bh, GFS2_METATYPE_JD, GFS2_FORMAT_JD);
102 gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
107 static int gfs2_dir_get_existing_buffer(struct gfs2_inode *ip, u64 block,
108 struct buffer_head **bhp)
110 struct buffer_head *bh;
113 error = gfs2_meta_read(ip->i_gl, block, DIO_WAIT, 0, &bh);
116 if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), bh, GFS2_METATYPE_JD)) {
124 static int gfs2_dir_write_stuffed(struct gfs2_inode *ip, const char *buf,
125 unsigned int offset, unsigned int size)
127 struct buffer_head *dibh;
130 error = gfs2_meta_inode_buffer(ip, &dibh);
134 gfs2_trans_add_meta(ip->i_gl, dibh);
135 memcpy(dibh->b_data + offset + sizeof(struct gfs2_dinode), buf, size);
136 if (ip->i_inode.i_size < offset + size)
137 i_size_write(&ip->i_inode, offset + size);
138 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
139 gfs2_dinode_out(ip, dibh->b_data);
149 * gfs2_dir_write_data - Write directory information to the inode
150 * @ip: The GFS2 inode
151 * @buf: The buffer containing information to be written
152 * @offset: The file offset to start writing at
153 * @size: The amount of data to write
155 * Returns: The number of bytes correctly written or error code
157 static int gfs2_dir_write_data(struct gfs2_inode *ip, const char *buf,
158 u64 offset, unsigned int size)
160 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
161 struct buffer_head *dibh;
172 if (gfs2_is_stuffed(ip) &&
173 offset + size <= sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode))
174 return gfs2_dir_write_stuffed(ip, buf, (unsigned int)offset,
177 if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
180 if (gfs2_is_stuffed(ip)) {
181 error = gfs2_unstuff_dinode(ip, NULL);
187 o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
189 while (copied < size) {
191 struct buffer_head *bh;
193 amount = size - copied;
194 if (amount > sdp->sd_sb.sb_bsize - o)
195 amount = sdp->sd_sb.sb_bsize - o;
199 error = gfs2_extent_map(&ip->i_inode, lblock, &new,
204 if (gfs2_assert_withdraw(sdp, dblock))
208 if (amount == sdp->sd_jbsize || new)
209 error = gfs2_dir_get_new_buffer(ip, dblock, &bh);
211 error = gfs2_dir_get_existing_buffer(ip, dblock, &bh);
216 gfs2_trans_add_meta(ip->i_gl, bh);
217 memcpy(bh->b_data + o, buf, amount);
226 o = sizeof(struct gfs2_meta_header);
230 error = gfs2_meta_inode_buffer(ip, &dibh);
234 if (ip->i_inode.i_size < offset + copied)
235 i_size_write(&ip->i_inode, offset + copied);
236 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
238 gfs2_trans_add_meta(ip->i_gl, dibh);
239 gfs2_dinode_out(ip, dibh->b_data);
249 static int gfs2_dir_read_stuffed(struct gfs2_inode *ip, __be64 *buf,
252 struct buffer_head *dibh;
255 error = gfs2_meta_inode_buffer(ip, &dibh);
257 memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size);
261 return (error) ? error : size;
266 * gfs2_dir_read_data - Read a data from a directory inode
267 * @ip: The GFS2 Inode
268 * @buf: The buffer to place result into
269 * @size: Amount of data to transfer
271 * Returns: The amount of data actually copied or the error
273 static int gfs2_dir_read_data(struct gfs2_inode *ip, __be64 *buf,
276 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
283 if (gfs2_is_stuffed(ip))
284 return gfs2_dir_read_stuffed(ip, buf, size);
286 if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
290 o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
292 while (copied < size) {
294 struct buffer_head *bh;
297 amount = size - copied;
298 if (amount > sdp->sd_sb.sb_bsize - o)
299 amount = sdp->sd_sb.sb_bsize - o;
303 error = gfs2_extent_map(&ip->i_inode, lblock, &new,
305 if (error || !dblock)
308 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
310 error = gfs2_meta_read(ip->i_gl, dblock, DIO_WAIT, 0, &bh);
314 error = gfs2_metatype_check(sdp, bh, GFS2_METATYPE_JD);
321 memcpy(buf, bh->b_data + o, amount);
323 buf += (amount/sizeof(__be64));
326 o = sizeof(struct gfs2_meta_header);
331 return (copied) ? copied : error;
335 * gfs2_dir_get_hash_table - Get pointer to the dir hash table
336 * @ip: The inode in question
338 * Returns: The hash table or an error
341 static __be64 *gfs2_dir_get_hash_table(struct gfs2_inode *ip)
343 struct inode *inode = &ip->i_inode;
348 BUG_ON(!(ip->i_diskflags & GFS2_DIF_EXHASH));
350 hc = ip->i_hash_cache;
354 hsize = 1 << ip->i_depth;
355 hsize *= sizeof(__be64);
356 if (hsize != i_size_read(&ip->i_inode)) {
357 gfs2_consist_inode(ip);
358 return ERR_PTR(-EIO);
361 hc = kmalloc(hsize, GFP_NOFS | __GFP_NOWARN);
363 hc = __vmalloc(hsize, GFP_NOFS, PAGE_KERNEL);
366 return ERR_PTR(-ENOMEM);
368 ret = gfs2_dir_read_data(ip, hc, hsize);
374 spin_lock(&inode->i_lock);
375 if (likely(!ip->i_hash_cache)) {
376 ip->i_hash_cache = hc;
379 spin_unlock(&inode->i_lock);
382 return ip->i_hash_cache;
386 * gfs2_dir_hash_inval - Invalidate dir hash
387 * @ip: The directory inode
389 * Must be called with an exclusive glock, or during glock invalidation.
391 void gfs2_dir_hash_inval(struct gfs2_inode *ip)
395 spin_lock(&ip->i_inode.i_lock);
396 hc = ip->i_hash_cache;
397 ip->i_hash_cache = NULL;
398 spin_unlock(&ip->i_inode.i_lock);
403 static inline int gfs2_dirent_sentinel(const struct gfs2_dirent *dent)
405 return dent->de_inum.no_addr == 0 || dent->de_inum.no_formal_ino == 0;
408 static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent,
409 const struct qstr *name, int ret)
411 if (!gfs2_dirent_sentinel(dent) &&
412 be32_to_cpu(dent->de_hash) == name->hash &&
413 be16_to_cpu(dent->de_name_len) == name->len &&
414 memcmp(dent+1, name->name, name->len) == 0)
419 static int gfs2_dirent_find(const struct gfs2_dirent *dent,
420 const struct qstr *name,
423 return __gfs2_dirent_find(dent, name, 1);
426 static int gfs2_dirent_prev(const struct gfs2_dirent *dent,
427 const struct qstr *name,
430 return __gfs2_dirent_find(dent, name, 2);
434 * name->name holds ptr to start of block.
435 * name->len holds size of block.
437 static int gfs2_dirent_last(const struct gfs2_dirent *dent,
438 const struct qstr *name,
441 const char *start = name->name;
442 const char *end = (const char *)dent + be16_to_cpu(dent->de_rec_len);
443 if (name->len == (end - start))
448 /* Look for the dirent that contains the offset specified in data. Once we
449 * find that dirent, there must be space available there for the new dirent */
450 static int gfs2_dirent_find_offset(const struct gfs2_dirent *dent,
451 const struct qstr *name,
454 unsigned required = GFS2_DIRENT_SIZE(name->len);
455 unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
456 unsigned totlen = be16_to_cpu(dent->de_rec_len);
458 if (ptr < (void *)dent || ptr >= (void *)dent + totlen)
460 if (gfs2_dirent_sentinel(dent))
462 if (ptr < (void *)dent + actual)
464 if ((void *)dent + totlen >= ptr + required)
469 static int gfs2_dirent_find_space(const struct gfs2_dirent *dent,
470 const struct qstr *name,
473 unsigned required = GFS2_DIRENT_SIZE(name->len);
474 unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
475 unsigned totlen = be16_to_cpu(dent->de_rec_len);
477 if (gfs2_dirent_sentinel(dent))
479 if (totlen - actual >= required)
484 struct dirent_gather {
485 const struct gfs2_dirent **pdent;
489 static int gfs2_dirent_gather(const struct gfs2_dirent *dent,
490 const struct qstr *name,
493 struct dirent_gather *g = opaque;
494 if (!gfs2_dirent_sentinel(dent)) {
495 g->pdent[g->offset++] = dent;
501 * Other possible things to check:
502 * - Inode located within filesystem size (and on valid block)
503 * - Valid directory entry type
504 * Not sure how heavy-weight we want to make this... could also check
505 * hash is correct for example, but that would take a lot of extra time.
506 * For now the most important thing is to check that the various sizes
509 static int gfs2_check_dirent(struct gfs2_dirent *dent, unsigned int offset,
510 unsigned int size, unsigned int len, int first)
512 const char *msg = "gfs2_dirent too small";
513 if (unlikely(size < sizeof(struct gfs2_dirent)))
515 msg = "gfs2_dirent misaligned";
516 if (unlikely(offset & 0x7))
518 msg = "gfs2_dirent points beyond end of block";
519 if (unlikely(offset + size > len))
521 msg = "zero inode number";
522 if (unlikely(!first && gfs2_dirent_sentinel(dent)))
524 msg = "name length is greater than space in dirent";
525 if (!gfs2_dirent_sentinel(dent) &&
526 unlikely(sizeof(struct gfs2_dirent)+be16_to_cpu(dent->de_name_len) >
531 pr_warn("%s: %s (%s)\n",
532 __func__, msg, first ? "first in block" : "not first in block");
536 static int gfs2_dirent_offset(const void *buf)
538 const struct gfs2_meta_header *h = buf;
543 switch(be32_to_cpu(h->mh_type)) {
544 case GFS2_METATYPE_LF:
545 offset = sizeof(struct gfs2_leaf);
547 case GFS2_METATYPE_DI:
548 offset = sizeof(struct gfs2_dinode);
555 pr_warn("%s: wrong block type %u\n", __func__, be32_to_cpu(h->mh_type));
559 static struct gfs2_dirent *gfs2_dirent_scan(struct inode *inode, void *buf,
560 unsigned int len, gfs2_dscan_t scan,
561 const struct qstr *name,
564 struct gfs2_dirent *dent, *prev;
569 ret = gfs2_dirent_offset(buf);
576 size = be16_to_cpu(dent->de_rec_len);
577 if (gfs2_check_dirent(dent, offset, size, len, 1))
580 ret = scan(dent, name, opaque);
588 size = be16_to_cpu(dent->de_rec_len);
589 if (gfs2_check_dirent(dent, offset, size, len, 0))
599 return prev ? prev : dent;
606 gfs2_consist_inode(GFS2_I(inode));
607 return ERR_PTR(-EIO);
610 static int dirent_check_reclen(struct gfs2_inode *dip,
611 const struct gfs2_dirent *d, const void *end_p)
614 u16 rec_len = be16_to_cpu(d->de_rec_len);
616 if (unlikely(rec_len < sizeof(struct gfs2_dirent)))
624 gfs2_consist_inode(dip);
629 * dirent_next - Next dirent
630 * @dip: the directory
632 * @dent: Pointer to list of dirents
634 * Returns: 0 on success, error code otherwise
637 static int dirent_next(struct gfs2_inode *dip, struct buffer_head *bh,
638 struct gfs2_dirent **dent)
640 struct gfs2_dirent *cur = *dent, *tmp;
641 char *bh_end = bh->b_data + bh->b_size;
644 ret = dirent_check_reclen(dip, cur, bh_end);
648 tmp = (void *)cur + ret;
649 ret = dirent_check_reclen(dip, tmp, bh_end);
653 /* Only the first dent could ever have de_inum.no_addr == 0 */
654 if (gfs2_dirent_sentinel(tmp)) {
655 gfs2_consist_inode(dip);
664 * dirent_del - Delete a dirent
665 * @dip: The GFS2 inode
667 * @prev: The previous dirent
668 * @cur: The current dirent
672 static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh,
673 struct gfs2_dirent *prev, struct gfs2_dirent *cur)
675 u16 cur_rec_len, prev_rec_len;
677 if (gfs2_dirent_sentinel(cur)) {
678 gfs2_consist_inode(dip);
682 gfs2_trans_add_meta(dip->i_gl, bh);
684 /* If there is no prev entry, this is the first entry in the block.
685 The de_rec_len is already as big as it needs to be. Just zero
686 out the inode number and return. */
689 cur->de_inum.no_addr = 0;
690 cur->de_inum.no_formal_ino = 0;
694 /* Combine this dentry with the previous one. */
696 prev_rec_len = be16_to_cpu(prev->de_rec_len);
697 cur_rec_len = be16_to_cpu(cur->de_rec_len);
699 if ((char *)prev + prev_rec_len != (char *)cur)
700 gfs2_consist_inode(dip);
701 if ((char *)cur + cur_rec_len > bh->b_data + bh->b_size)
702 gfs2_consist_inode(dip);
704 prev_rec_len += cur_rec_len;
705 prev->de_rec_len = cpu_to_be16(prev_rec_len);
709 static struct gfs2_dirent *do_init_dirent(struct inode *inode,
710 struct gfs2_dirent *dent,
711 const struct qstr *name,
712 struct buffer_head *bh,
715 struct gfs2_inode *ip = GFS2_I(inode);
716 struct gfs2_dirent *ndent;
719 totlen = be16_to_cpu(dent->de_rec_len);
720 BUG_ON(offset + name->len > totlen);
721 gfs2_trans_add_meta(ip->i_gl, bh);
722 ndent = (struct gfs2_dirent *)((char *)dent + offset);
723 dent->de_rec_len = cpu_to_be16(offset);
724 gfs2_qstr2dirent(name, totlen - offset, ndent);
730 * Takes a dent from which to grab space as an argument. Returns the
731 * newly created dent.
733 static struct gfs2_dirent *gfs2_init_dirent(struct inode *inode,
734 struct gfs2_dirent *dent,
735 const struct qstr *name,
736 struct buffer_head *bh)
740 if (!gfs2_dirent_sentinel(dent))
741 offset = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
742 return do_init_dirent(inode, dent, name, bh, offset);
745 static struct gfs2_dirent *gfs2_dirent_split_alloc(struct inode *inode,
746 struct buffer_head *bh,
747 const struct qstr *name,
750 struct gfs2_dirent *dent;
751 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
752 gfs2_dirent_find_offset, name, ptr);
753 if (!dent || IS_ERR(dent))
755 return do_init_dirent(inode, dent, name, bh,
756 (unsigned)(ptr - (void *)dent));
759 static int get_leaf(struct gfs2_inode *dip, u64 leaf_no,
760 struct buffer_head **bhp)
764 error = gfs2_meta_read(dip->i_gl, leaf_no, DIO_WAIT, 0, bhp);
765 if (!error && gfs2_metatype_check(GFS2_SB(&dip->i_inode), *bhp, GFS2_METATYPE_LF)) {
766 /* pr_info("block num=%llu\n", leaf_no); */
774 * get_leaf_nr - Get a leaf number associated with the index
775 * @dip: The GFS2 inode
779 * Returns: 0 on success, error code otherwise
782 static int get_leaf_nr(struct gfs2_inode *dip, u32 index,
787 hash = gfs2_dir_get_hash_table(dip);
789 return PTR_ERR(hash);
790 *leaf_out = be64_to_cpu(*(hash + index));
794 static int get_first_leaf(struct gfs2_inode *dip, u32 index,
795 struct buffer_head **bh_out)
800 error = get_leaf_nr(dip, index, &leaf_no);
802 error = get_leaf(dip, leaf_no, bh_out);
807 static struct gfs2_dirent *gfs2_dirent_search(struct inode *inode,
808 const struct qstr *name,
810 struct buffer_head **pbh)
812 struct buffer_head *bh;
813 struct gfs2_dirent *dent;
814 struct gfs2_inode *ip = GFS2_I(inode);
817 if (ip->i_diskflags & GFS2_DIF_EXHASH) {
818 struct gfs2_leaf *leaf;
819 unsigned hsize = 1 << ip->i_depth;
822 if (hsize * sizeof(u64) != i_size_read(inode)) {
823 gfs2_consist_inode(ip);
824 return ERR_PTR(-EIO);
827 index = name->hash >> (32 - ip->i_depth);
828 error = get_first_leaf(ip, index, &bh);
830 return ERR_PTR(error);
832 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
836 leaf = (struct gfs2_leaf *)bh->b_data;
837 ln = be64_to_cpu(leaf->lf_next);
842 error = get_leaf(ip, ln, &bh);
845 return error ? ERR_PTR(error) : NULL;
849 error = gfs2_meta_inode_buffer(ip, &bh);
851 return ERR_PTR(error);
852 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, scan, name, NULL);
854 if (unlikely(dent == NULL || IS_ERR(dent))) {
862 static struct gfs2_leaf *new_leaf(struct inode *inode, struct buffer_head **pbh, u16 depth)
864 struct gfs2_inode *ip = GFS2_I(inode);
868 struct buffer_head *bh;
869 struct gfs2_leaf *leaf;
870 struct gfs2_dirent *dent;
871 struct qstr name = { .name = "" };
872 struct timespec tv = CURRENT_TIME;
874 error = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL);
877 bh = gfs2_meta_new(ip->i_gl, bn);
881 gfs2_trans_add_unrevoke(GFS2_SB(inode), bn, 1);
882 gfs2_trans_add_meta(ip->i_gl, bh);
883 gfs2_metatype_set(bh, GFS2_METATYPE_LF, GFS2_FORMAT_LF);
884 leaf = (struct gfs2_leaf *)bh->b_data;
885 leaf->lf_depth = cpu_to_be16(depth);
886 leaf->lf_entries = 0;
887 leaf->lf_dirent_format = cpu_to_be32(GFS2_FORMAT_DE);
889 leaf->lf_inode = cpu_to_be64(ip->i_no_addr);
890 leaf->lf_dist = cpu_to_be32(1);
891 leaf->lf_nsec = cpu_to_be32(tv.tv_nsec);
892 leaf->lf_sec = cpu_to_be64(tv.tv_sec);
893 memset(leaf->lf_reserved2, 0, sizeof(leaf->lf_reserved2));
894 dent = (struct gfs2_dirent *)(leaf+1);
895 gfs2_qstr2dirent(&name, bh->b_size - sizeof(struct gfs2_leaf), dent);
901 * dir_make_exhash - Convert a stuffed directory into an ExHash directory
902 * @dip: The GFS2 inode
904 * Returns: 0 on success, error code otherwise
907 static int dir_make_exhash(struct inode *inode)
909 struct gfs2_inode *dip = GFS2_I(inode);
910 struct gfs2_sbd *sdp = GFS2_SB(inode);
911 struct gfs2_dirent *dent;
913 struct buffer_head *bh, *dibh;
914 struct gfs2_leaf *leaf;
921 error = gfs2_meta_inode_buffer(dip, &dibh);
925 /* Turn over a new leaf */
927 leaf = new_leaf(inode, &bh, 0);
932 gfs2_assert(sdp, dip->i_entries < (1 << 16));
933 leaf->lf_entries = cpu_to_be16(dip->i_entries);
937 gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_leaf), dibh,
938 sizeof(struct gfs2_dinode));
940 /* Find last entry */
943 args.len = bh->b_size - sizeof(struct gfs2_dinode) +
944 sizeof(struct gfs2_leaf);
945 args.name = bh->b_data;
946 dent = gfs2_dirent_scan(&dip->i_inode, bh->b_data, bh->b_size,
947 gfs2_dirent_last, &args, NULL);
956 return PTR_ERR(dent);
959 /* Adjust the last dirent's record length
960 (Remember that dent still points to the last entry.) */
962 dent->de_rec_len = cpu_to_be16(be16_to_cpu(dent->de_rec_len) +
963 sizeof(struct gfs2_dinode) -
964 sizeof(struct gfs2_leaf));
968 /* We're done with the new leaf block, now setup the new
971 gfs2_trans_add_meta(dip->i_gl, dibh);
972 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
974 lp = (__be64 *)(dibh->b_data + sizeof(struct gfs2_dinode));
976 for (x = sdp->sd_hash_ptrs; x--; lp++)
977 *lp = cpu_to_be64(bn);
979 i_size_write(inode, sdp->sd_sb.sb_bsize / 2);
980 gfs2_add_inode_blocks(&dip->i_inode, 1);
981 dip->i_diskflags |= GFS2_DIF_EXHASH;
983 for (x = sdp->sd_hash_ptrs, y = -1; x; x >>= 1, y++) ;
986 gfs2_dinode_out(dip, dibh->b_data);
994 * dir_split_leaf - Split a leaf block into two
995 * @dip: The GFS2 inode
999 * Returns: 0 on success, error code on failure
1002 static int dir_split_leaf(struct inode *inode, const struct qstr *name)
1004 struct gfs2_inode *dip = GFS2_I(inode);
1005 struct buffer_head *nbh, *obh, *dibh;
1006 struct gfs2_leaf *nleaf, *oleaf;
1007 struct gfs2_dirent *dent = NULL, *prev = NULL, *next = NULL, *new;
1008 u32 start, len, half_len, divider;
1015 index = name->hash >> (32 - dip->i_depth);
1016 error = get_leaf_nr(dip, index, &leaf_no);
1020 /* Get the old leaf block */
1021 error = get_leaf(dip, leaf_no, &obh);
1025 oleaf = (struct gfs2_leaf *)obh->b_data;
1026 if (dip->i_depth == be16_to_cpu(oleaf->lf_depth)) {
1028 return 1; /* can't split */
1031 gfs2_trans_add_meta(dip->i_gl, obh);
1033 nleaf = new_leaf(inode, &nbh, be16_to_cpu(oleaf->lf_depth) + 1);
1038 bn = nbh->b_blocknr;
1040 /* Compute the start and len of leaf pointers in the hash table. */
1041 len = 1 << (dip->i_depth - be16_to_cpu(oleaf->lf_depth));
1042 half_len = len >> 1;
1044 pr_warn("i_depth %u lf_depth %u index %u\n",
1045 dip->i_depth, be16_to_cpu(oleaf->lf_depth), index);
1046 gfs2_consist_inode(dip);
1051 start = (index & ~(len - 1));
1053 /* Change the pointers.
1054 Don't bother distinguishing stuffed from non-stuffed.
1055 This code is complicated enough already. */
1056 lp = kmalloc(half_len * sizeof(__be64), GFP_NOFS);
1062 /* Change the pointers */
1063 for (x = 0; x < half_len; x++)
1064 lp[x] = cpu_to_be64(bn);
1066 gfs2_dir_hash_inval(dip);
1068 error = gfs2_dir_write_data(dip, (char *)lp, start * sizeof(u64),
1069 half_len * sizeof(u64));
1070 if (error != half_len * sizeof(u64)) {
1078 /* Compute the divider */
1079 divider = (start + half_len) << (32 - dip->i_depth);
1081 /* Copy the entries */
1082 dent = (struct gfs2_dirent *)(obh->b_data + sizeof(struct gfs2_leaf));
1086 if (dirent_next(dip, obh, &next))
1089 if (!gfs2_dirent_sentinel(dent) &&
1090 be32_to_cpu(dent->de_hash) < divider) {
1092 void *ptr = ((char *)dent - obh->b_data) + nbh->b_data;
1093 str.name = (char*)(dent+1);
1094 str.len = be16_to_cpu(dent->de_name_len);
1095 str.hash = be32_to_cpu(dent->de_hash);
1096 new = gfs2_dirent_split_alloc(inode, nbh, &str, ptr);
1098 error = PTR_ERR(new);
1102 new->de_inum = dent->de_inum; /* No endian worries */
1103 new->de_type = dent->de_type; /* No endian worries */
1104 be16_add_cpu(&nleaf->lf_entries, 1);
1106 dirent_del(dip, obh, prev, dent);
1108 if (!oleaf->lf_entries)
1109 gfs2_consist_inode(dip);
1110 be16_add_cpu(&oleaf->lf_entries, -1);
1122 oleaf->lf_depth = nleaf->lf_depth;
1124 error = gfs2_meta_inode_buffer(dip, &dibh);
1125 if (!gfs2_assert_withdraw(GFS2_SB(&dip->i_inode), !error)) {
1126 gfs2_trans_add_meta(dip->i_gl, dibh);
1127 gfs2_add_inode_blocks(&dip->i_inode, 1);
1128 gfs2_dinode_out(dip, dibh->b_data);
1147 * dir_double_exhash - Double size of ExHash table
1148 * @dip: The GFS2 dinode
1150 * Returns: 0 on success, error code on failure
1153 static int dir_double_exhash(struct gfs2_inode *dip)
1155 struct buffer_head *dibh;
1163 hsize = 1 << dip->i_depth;
1164 hsize_bytes = hsize * sizeof(__be64);
1166 hc = gfs2_dir_get_hash_table(dip);
1170 hc2 = kmalloc(hsize_bytes * 2, GFP_NOFS | __GFP_NOWARN);
1172 hc2 = __vmalloc(hsize_bytes * 2, GFP_NOFS, PAGE_KERNEL);
1178 error = gfs2_meta_inode_buffer(dip, &dibh);
1182 for (x = 0; x < hsize; x++) {
1188 error = gfs2_dir_write_data(dip, (char *)hc2, 0, hsize_bytes * 2);
1189 if (error != (hsize_bytes * 2))
1192 gfs2_dir_hash_inval(dip);
1193 dip->i_hash_cache = hc2;
1195 gfs2_dinode_out(dip, dibh->b_data);
1200 /* Replace original hash table & size */
1201 gfs2_dir_write_data(dip, (char *)hc, 0, hsize_bytes);
1202 i_size_write(&dip->i_inode, hsize_bytes);
1203 gfs2_dinode_out(dip, dibh->b_data);
1211 * compare_dents - compare directory entries by hash value
1215 * When comparing the hash entries of @a to @b:
1221 static int compare_dents(const void *a, const void *b)
1223 const struct gfs2_dirent *dent_a, *dent_b;
1227 dent_a = *(const struct gfs2_dirent **)a;
1228 hash_a = dent_a->de_cookie;
1230 dent_b = *(const struct gfs2_dirent **)b;
1231 hash_b = dent_b->de_cookie;
1233 if (hash_a > hash_b)
1235 else if (hash_a < hash_b)
1238 unsigned int len_a = be16_to_cpu(dent_a->de_name_len);
1239 unsigned int len_b = be16_to_cpu(dent_b->de_name_len);
1243 else if (len_a < len_b)
1246 ret = memcmp(dent_a + 1, dent_b + 1, len_a);
1253 * do_filldir_main - read out directory entries
1254 * @dip: The GFS2 inode
1255 * @ctx: what to feed the entries to
1256 * @darr: an array of struct gfs2_dirent pointers to read
1257 * @entries: the number of entries in darr
1258 * @copied: pointer to int that's non-zero if a entry has been copied out
1260 * Jump through some hoops to make sure that if there are hash collsions,
1261 * they are read out at the beginning of a buffer. We want to minimize
1262 * the possibility that they will fall into different readdir buffers or
1263 * that someone will want to seek to that location.
1265 * Returns: errno, >0 if the actor tells you to stop
1268 static int do_filldir_main(struct gfs2_inode *dip, struct dir_context *ctx,
1269 struct gfs2_dirent **darr, u32 entries,
1270 u32 sort_start, int *copied)
1272 const struct gfs2_dirent *dent, *dent_next;
1277 if (sort_start < entries)
1278 sort(&darr[sort_start], entries - sort_start,
1279 sizeof(struct gfs2_dirent *), compare_dents, NULL);
1281 dent_next = darr[0];
1282 off_next = dent_next->de_cookie;
1284 for (x = 0, y = 1; x < entries; x++, y++) {
1289 dent_next = darr[y];
1290 off_next = dent_next->de_cookie;
1296 if (off_next == off) {
1297 if (*copied && !run)
1308 if (!dir_emit(ctx, (const char *)(dent + 1),
1309 be16_to_cpu(dent->de_name_len),
1310 be64_to_cpu(dent->de_inum.no_addr),
1311 be16_to_cpu(dent->de_type)))
1317 /* Increment the ctx->pos by one, so the next time we come into the
1318 do_filldir fxn, we get the next entry instead of the last one in the
1326 static void *gfs2_alloc_sort_buffer(unsigned size)
1330 if (size < KMALLOC_MAX_SIZE)
1331 ptr = kmalloc(size, GFP_NOFS | __GFP_NOWARN);
1333 ptr = __vmalloc(size, GFP_NOFS, PAGE_KERNEL);
1338 static int gfs2_set_cookies(struct gfs2_sbd *sdp, struct buffer_head *bh,
1339 unsigned leaf_nr, struct gfs2_dirent **darr,
1345 for (i = 0; i < entries; i++) {
1348 darr[i]->de_cookie = be32_to_cpu(darr[i]->de_hash);
1349 darr[i]->de_cookie = gfs2_disk_hash2offset(darr[i]->de_cookie);
1351 if (!sdp->sd_args.ar_loccookie)
1353 offset = (char *)(darr[i]) -
1354 (bh->b_data + gfs2_dirent_offset(bh->b_data));
1355 offset /= GFS2_MIN_DIRENT_SIZE;
1356 offset += leaf_nr * sdp->sd_max_dents_per_leaf;
1357 if (offset >= GFS2_USE_HASH_FLAG ||
1358 leaf_nr >= GFS2_USE_HASH_FLAG) {
1359 darr[i]->de_cookie |= GFS2_USE_HASH_FLAG;
1364 darr[i]->de_cookie &= GFS2_HASH_INDEX_MASK;
1365 darr[i]->de_cookie |= offset;
1371 static int gfs2_dir_read_leaf(struct inode *inode, struct dir_context *ctx,
1372 int *copied, unsigned *depth,
1375 struct gfs2_inode *ip = GFS2_I(inode);
1376 struct gfs2_sbd *sdp = GFS2_SB(inode);
1377 struct buffer_head *bh;
1378 struct gfs2_leaf *lf;
1379 unsigned entries = 0, entries2 = 0;
1380 unsigned leaves = 0, leaf = 0, offset, sort_offset;
1381 struct gfs2_dirent **darr, *dent;
1382 struct dirent_gather g;
1383 struct buffer_head **larr;
1384 int error, i, need_sort = 0, sort_id;
1388 error = get_leaf(ip, lfn, &bh);
1391 lf = (struct gfs2_leaf *)bh->b_data;
1393 *depth = be16_to_cpu(lf->lf_depth);
1394 entries += be16_to_cpu(lf->lf_entries);
1396 lfn = be64_to_cpu(lf->lf_next);
1400 if (*depth < GFS2_DIR_MAX_DEPTH || !sdp->sd_args.ar_loccookie) {
1410 * The extra 99 entries are not normally used, but are a buffer
1411 * zone in case the number of entries in the leaf is corrupt.
1412 * 99 is the maximum number of entries that can fit in a single
1415 larr = gfs2_alloc_sort_buffer((leaves + entries + 99) * sizeof(void *));
1418 darr = (struct gfs2_dirent **)(larr + leaves);
1419 g.pdent = (const struct gfs2_dirent **)darr;
1424 error = get_leaf(ip, lfn, &bh);
1427 lf = (struct gfs2_leaf *)bh->b_data;
1428 lfn = be64_to_cpu(lf->lf_next);
1429 if (lf->lf_entries) {
1431 entries2 += be16_to_cpu(lf->lf_entries);
1432 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
1433 gfs2_dirent_gather, NULL, &g);
1434 error = PTR_ERR(dent);
1437 if (entries2 != g.offset) {
1438 fs_warn(sdp, "Number of entries corrupt in dir "
1439 "leaf %llu, entries2 (%u) != "
1441 (unsigned long long)bh->b_blocknr,
1442 entries2, g.offset);
1448 sort_id = gfs2_set_cookies(sdp, bh, leaf, &darr[offset],
1449 be16_to_cpu(lf->lf_entries));
1450 if (!need_sort && sort_id >= 0) {
1452 sort_offset = offset + sort_id;
1456 larr[leaf++] = NULL;
1461 BUG_ON(entries2 != entries);
1462 error = do_filldir_main(ip, ctx, darr, entries, need_sort ?
1463 sort_offset : entries, copied);
1465 for(i = 0; i < leaf; i++)
1474 * gfs2_dir_readahead - Issue read-ahead requests for leaf blocks.
1476 * Note: we can't calculate each index like dir_e_read can because we don't
1477 * have the leaf, and therefore we don't have the depth, and therefore we
1478 * don't have the length. So we have to just read enough ahead to make up
1479 * for the loss of information.
1481 static void gfs2_dir_readahead(struct inode *inode, unsigned hsize, u32 index,
1482 struct file_ra_state *f_ra)
1484 struct gfs2_inode *ip = GFS2_I(inode);
1485 struct gfs2_glock *gl = ip->i_gl;
1486 struct buffer_head *bh;
1487 u64 blocknr = 0, last;
1490 /* First check if we've already read-ahead for the whole range. */
1491 if (index + MAX_RA_BLOCKS < f_ra->start)
1494 f_ra->start = max((pgoff_t)index, f_ra->start);
1495 for (count = 0; count < MAX_RA_BLOCKS; count++) {
1496 if (f_ra->start >= hsize) /* if exceeded the hash table */
1500 blocknr = be64_to_cpu(ip->i_hash_cache[f_ra->start]);
1502 if (blocknr == last)
1505 bh = gfs2_getbuf(gl, blocknr, 1);
1506 if (trylock_buffer(bh)) {
1507 if (buffer_uptodate(bh)) {
1512 bh->b_end_io = end_buffer_read_sync;
1513 submit_bh(READA | REQ_META, bh);
1521 * dir_e_read - Reads the entries from a directory into a filldir buffer
1522 * @dip: dinode pointer
1523 * @ctx: actor to feed the entries to
1528 static int dir_e_read(struct inode *inode, struct dir_context *ctx,
1529 struct file_ra_state *f_ra)
1531 struct gfs2_inode *dip = GFS2_I(inode);
1539 hsize = 1 << dip->i_depth;
1540 hash = gfs2_dir_offset2hash(ctx->pos);
1541 index = hash >> (32 - dip->i_depth);
1543 if (dip->i_hash_cache == NULL)
1545 lp = gfs2_dir_get_hash_table(dip);
1549 gfs2_dir_readahead(inode, hsize, index, f_ra);
1551 while (index < hsize) {
1552 error = gfs2_dir_read_leaf(inode, ctx,
1554 be64_to_cpu(lp[index]));
1558 len = 1 << (dip->i_depth - depth);
1559 index = (index & ~(len - 1)) + len;
1567 int gfs2_dir_read(struct inode *inode, struct dir_context *ctx,
1568 struct file_ra_state *f_ra)
1570 struct gfs2_inode *dip = GFS2_I(inode);
1571 struct gfs2_sbd *sdp = GFS2_SB(inode);
1572 struct dirent_gather g;
1573 struct gfs2_dirent **darr, *dent;
1574 struct buffer_head *dibh;
1578 if (!dip->i_entries)
1581 if (dip->i_diskflags & GFS2_DIF_EXHASH)
1582 return dir_e_read(inode, ctx, f_ra);
1584 if (!gfs2_is_stuffed(dip)) {
1585 gfs2_consist_inode(dip);
1589 error = gfs2_meta_inode_buffer(dip, &dibh);
1594 /* 96 is max number of dirents which can be stuffed into an inode */
1595 darr = kmalloc(96 * sizeof(struct gfs2_dirent *), GFP_NOFS);
1597 g.pdent = (const struct gfs2_dirent **)darr;
1599 dent = gfs2_dirent_scan(inode, dibh->b_data, dibh->b_size,
1600 gfs2_dirent_gather, NULL, &g);
1602 error = PTR_ERR(dent);
1605 if (dip->i_entries != g.offset) {
1606 fs_warn(sdp, "Number of entries corrupt in dir %llu, "
1607 "ip->i_entries (%u) != g.offset (%u)\n",
1608 (unsigned long long)dip->i_no_addr,
1614 gfs2_set_cookies(sdp, dibh, 0, darr, dip->i_entries);
1615 error = do_filldir_main(dip, ctx, darr,
1616 dip->i_entries, 0, &copied);
1630 * gfs2_dir_search - Search a directory
1631 * @dip: The GFS2 dir inode
1632 * @name: The name we are looking up
1633 * @fail_on_exist: Fail if the name exists rather than looking it up
1635 * This routine searches a directory for a file or another directory.
1636 * Assumes a glock is held on dip.
1641 struct inode *gfs2_dir_search(struct inode *dir, const struct qstr *name,
1644 struct buffer_head *bh;
1645 struct gfs2_dirent *dent;
1646 u64 addr, formal_ino;
1649 dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
1651 struct inode *inode;
1655 return ERR_CAST(dent);
1656 dtype = be16_to_cpu(dent->de_type);
1657 rahead = be16_to_cpu(dent->de_rahead);
1658 addr = be64_to_cpu(dent->de_inum.no_addr);
1659 formal_ino = be64_to_cpu(dent->de_inum.no_formal_ino);
1662 return ERR_PTR(-EEXIST);
1663 inode = gfs2_inode_lookup(dir->i_sb, dtype, addr, formal_ino, 0);
1665 GFS2_I(inode)->i_rahead = rahead;
1668 return ERR_PTR(-ENOENT);
1671 int gfs2_dir_check(struct inode *dir, const struct qstr *name,
1672 const struct gfs2_inode *ip)
1674 struct buffer_head *bh;
1675 struct gfs2_dirent *dent;
1678 dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
1681 return PTR_ERR(dent);
1683 if (be64_to_cpu(dent->de_inum.no_addr) != ip->i_no_addr)
1685 if (be64_to_cpu(dent->de_inum.no_formal_ino) !=
1686 ip->i_no_formal_ino)
1688 if (unlikely(IF2DT(ip->i_inode.i_mode) !=
1689 be16_to_cpu(dent->de_type))) {
1690 gfs2_consist_inode(GFS2_I(dir));
1703 * dir_new_leaf - Add a new leaf onto hash chain
1704 * @inode: The directory
1705 * @name: The name we are adding
1707 * This adds a new dir leaf onto an existing leaf when there is not
1708 * enough space to add a new dir entry. This is a last resort after
1709 * we've expanded the hash table to max size and also split existing
1710 * leaf blocks, so it will only occur for very large directories.
1712 * The dist parameter is set to 1 for leaf blocks directly attached
1713 * to the hash table, 2 for one layer of indirection, 3 for two layers
1714 * etc. We are thus able to tell the difference between an old leaf
1715 * with dist set to zero (i.e. "don't know") and a new one where we
1716 * set this information for debug/fsck purposes.
1718 * Returns: 0 on success, or -ve on error
1721 static int dir_new_leaf(struct inode *inode, const struct qstr *name)
1723 struct buffer_head *bh, *obh;
1724 struct gfs2_inode *ip = GFS2_I(inode);
1725 struct gfs2_leaf *leaf, *oleaf;
1731 index = name->hash >> (32 - ip->i_depth);
1732 error = get_first_leaf(ip, index, &obh);
1737 oleaf = (struct gfs2_leaf *)obh->b_data;
1738 bn = be64_to_cpu(oleaf->lf_next);
1742 error = get_leaf(ip, bn, &obh);
1747 gfs2_trans_add_meta(ip->i_gl, obh);
1749 leaf = new_leaf(inode, &bh, be16_to_cpu(oleaf->lf_depth));
1754 leaf->lf_dist = cpu_to_be32(dist);
1755 oleaf->lf_next = cpu_to_be64(bh->b_blocknr);
1759 error = gfs2_meta_inode_buffer(ip, &bh);
1762 gfs2_trans_add_meta(ip->i_gl, bh);
1763 gfs2_add_inode_blocks(&ip->i_inode, 1);
1764 gfs2_dinode_out(ip, bh->b_data);
1769 static u16 gfs2_inode_ra_len(const struct gfs2_inode *ip)
1771 u64 where = ip->i_no_addr + 1;
1772 if (ip->i_eattr == where)
1778 * gfs2_dir_add - Add new filename into directory
1779 * @inode: The directory inode
1780 * @name: The new name
1781 * @nip: The GFS2 inode to be linked in to the directory
1782 * @da: The directory addition info
1784 * If the call to gfs2_diradd_alloc_required resulted in there being
1785 * no need to allocate any new directory blocks, then it will contain
1786 * a pointer to the directory entry and the bh in which it resides. We
1787 * can use that without having to repeat the search. If there was no
1788 * free space, then we must now create more space.
1790 * Returns: 0 on success, error code on failure
1793 int gfs2_dir_add(struct inode *inode, const struct qstr *name,
1794 const struct gfs2_inode *nip, struct gfs2_diradd *da)
1796 struct gfs2_inode *ip = GFS2_I(inode);
1797 struct buffer_head *bh = da->bh;
1798 struct gfs2_dirent *dent = da->dent;
1800 struct gfs2_leaf *leaf;
1804 if (da->bh == NULL) {
1805 dent = gfs2_dirent_search(inode, name,
1806 gfs2_dirent_find_space, &bh);
1810 return PTR_ERR(dent);
1811 dent = gfs2_init_dirent(inode, dent, name, bh);
1812 gfs2_inum_out(nip, dent);
1813 dent->de_type = cpu_to_be16(IF2DT(nip->i_inode.i_mode));
1814 dent->de_rahead = cpu_to_be16(gfs2_inode_ra_len(nip));
1816 if (ip->i_diskflags & GFS2_DIF_EXHASH) {
1817 leaf = (struct gfs2_leaf *)bh->b_data;
1818 be16_add_cpu(&leaf->lf_entries, 1);
1819 leaf->lf_nsec = cpu_to_be32(tv.tv_nsec);
1820 leaf->lf_sec = cpu_to_be64(tv.tv_sec);
1826 ip->i_inode.i_mtime = ip->i_inode.i_ctime = tv;
1827 if (S_ISDIR(nip->i_inode.i_mode))
1828 inc_nlink(&ip->i_inode);
1829 mark_inode_dirty(inode);
1833 if (!(ip->i_diskflags & GFS2_DIF_EXHASH)) {
1834 error = dir_make_exhash(inode);
1839 error = dir_split_leaf(inode, name);
1844 if (ip->i_depth < GFS2_DIR_MAX_DEPTH) {
1845 error = dir_double_exhash(ip);
1848 error = dir_split_leaf(inode, name);
1854 error = dir_new_leaf(inode, name);
1865 * gfs2_dir_del - Delete a directory entry
1866 * @dip: The GFS2 inode
1867 * @filename: The filename
1869 * Returns: 0 on success, error code on failure
1872 int gfs2_dir_del(struct gfs2_inode *dip, const struct dentry *dentry)
1874 const struct qstr *name = &dentry->d_name;
1875 struct gfs2_dirent *dent, *prev = NULL;
1876 struct buffer_head *bh;
1877 struct timespec tv = CURRENT_TIME;
1879 /* Returns _either_ the entry (if its first in block) or the
1880 previous entry otherwise */
1881 dent = gfs2_dirent_search(&dip->i_inode, name, gfs2_dirent_prev, &bh);
1883 gfs2_consist_inode(dip);
1887 gfs2_consist_inode(dip);
1888 return PTR_ERR(dent);
1890 /* If not first in block, adjust pointers accordingly */
1891 if (gfs2_dirent_find(dent, name, NULL) == 0) {
1893 dent = (struct gfs2_dirent *)((char *)dent + be16_to_cpu(prev->de_rec_len));
1896 dirent_del(dip, bh, prev, dent);
1897 if (dip->i_diskflags & GFS2_DIF_EXHASH) {
1898 struct gfs2_leaf *leaf = (struct gfs2_leaf *)bh->b_data;
1899 u16 entries = be16_to_cpu(leaf->lf_entries);
1901 gfs2_consist_inode(dip);
1902 leaf->lf_entries = cpu_to_be16(--entries);
1903 leaf->lf_nsec = cpu_to_be32(tv.tv_nsec);
1904 leaf->lf_sec = cpu_to_be64(tv.tv_sec);
1908 if (!dip->i_entries)
1909 gfs2_consist_inode(dip);
1911 dip->i_inode.i_mtime = dip->i_inode.i_ctime = tv;
1912 if (d_is_dir(dentry))
1913 drop_nlink(&dip->i_inode);
1914 mark_inode_dirty(&dip->i_inode);
1920 * gfs2_dir_mvino - Change inode number of directory entry
1921 * @dip: The GFS2 inode
1925 * This routine changes the inode number of a directory entry. It's used
1926 * by rename to change ".." when a directory is moved.
1927 * Assumes a glock is held on dvp.
1932 int gfs2_dir_mvino(struct gfs2_inode *dip, const struct qstr *filename,
1933 const struct gfs2_inode *nip, unsigned int new_type)
1935 struct buffer_head *bh;
1936 struct gfs2_dirent *dent;
1939 dent = gfs2_dirent_search(&dip->i_inode, filename, gfs2_dirent_find, &bh);
1941 gfs2_consist_inode(dip);
1945 return PTR_ERR(dent);
1947 gfs2_trans_add_meta(dip->i_gl, bh);
1948 gfs2_inum_out(nip, dent);
1949 dent->de_type = cpu_to_be16(new_type);
1951 if (dip->i_diskflags & GFS2_DIF_EXHASH) {
1953 error = gfs2_meta_inode_buffer(dip, &bh);
1956 gfs2_trans_add_meta(dip->i_gl, bh);
1959 dip->i_inode.i_mtime = dip->i_inode.i_ctime = CURRENT_TIME;
1960 gfs2_dinode_out(dip, bh->b_data);
1966 * leaf_dealloc - Deallocate a directory leaf
1967 * @dip: the directory
1968 * @index: the hash table offset in the directory
1969 * @len: the number of pointers to this leaf
1970 * @leaf_no: the leaf number
1971 * @leaf_bh: buffer_head for the starting leaf
1972 * last_dealloc: 1 if this is the final dealloc for the leaf, else 0
1977 static int leaf_dealloc(struct gfs2_inode *dip, u32 index, u32 len,
1978 u64 leaf_no, struct buffer_head *leaf_bh,
1981 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1982 struct gfs2_leaf *tmp_leaf;
1983 struct gfs2_rgrp_list rlist;
1984 struct buffer_head *bh, *dibh;
1986 unsigned int rg_blocks = 0, l_blocks = 0;
1988 unsigned int x, size = len * sizeof(u64);
1991 error = gfs2_rindex_update(sdp);
1995 memset(&rlist, 0, sizeof(struct gfs2_rgrp_list));
1997 ht = kzalloc(size, GFP_NOFS | __GFP_NOWARN);
1999 ht = __vmalloc(size, GFP_NOFS | __GFP_NOWARN | __GFP_ZERO,
2004 error = gfs2_quota_hold(dip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
2008 /* Count the number of leaves */
2011 for (blk = leaf_no; blk; blk = nblk) {
2012 if (blk != leaf_no) {
2013 error = get_leaf(dip, blk, &bh);
2017 tmp_leaf = (struct gfs2_leaf *)bh->b_data;
2018 nblk = be64_to_cpu(tmp_leaf->lf_next);
2022 gfs2_rlist_add(dip, &rlist, blk);
2026 gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE);
2028 for (x = 0; x < rlist.rl_rgrps; x++) {
2029 struct gfs2_rgrpd *rgd;
2030 rgd = rlist.rl_ghs[x].gh_gl->gl_object;
2031 rg_blocks += rgd->rd_length;
2034 error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs);
2038 error = gfs2_trans_begin(sdp,
2039 rg_blocks + (DIV_ROUND_UP(size, sdp->sd_jbsize) + 1) +
2040 RES_DINODE + RES_STATFS + RES_QUOTA, l_blocks);
2042 goto out_rg_gunlock;
2046 for (blk = leaf_no; blk; blk = nblk) {
2047 if (blk != leaf_no) {
2048 error = get_leaf(dip, blk, &bh);
2052 tmp_leaf = (struct gfs2_leaf *)bh->b_data;
2053 nblk = be64_to_cpu(tmp_leaf->lf_next);
2057 gfs2_free_meta(dip, blk, 1);
2058 gfs2_add_inode_blocks(&dip->i_inode, -1);
2061 error = gfs2_dir_write_data(dip, ht, index * sizeof(u64), size);
2062 if (error != size) {
2068 error = gfs2_meta_inode_buffer(dip, &dibh);
2072 gfs2_trans_add_meta(dip->i_gl, dibh);
2073 /* On the last dealloc, make this a regular file in case we crash.
2074 (We don't want to free these blocks a second time.) */
2076 dip->i_inode.i_mode = S_IFREG;
2077 gfs2_dinode_out(dip, dibh->b_data);
2081 gfs2_trans_end(sdp);
2083 gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
2085 gfs2_rlist_free(&rlist);
2086 gfs2_quota_unhold(dip);
2093 * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory
2094 * @dip: the directory
2096 * Dealloc all on-disk directory leaves to FREEMETA state
2097 * Change on-disk inode type to "regular file"
2102 int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip)
2104 struct buffer_head *bh;
2105 struct gfs2_leaf *leaf;
2107 u32 index = 0, next_index;
2110 int error = 0, last;
2112 hsize = 1 << dip->i_depth;
2114 lp = gfs2_dir_get_hash_table(dip);
2118 while (index < hsize) {
2119 leaf_no = be64_to_cpu(lp[index]);
2121 error = get_leaf(dip, leaf_no, &bh);
2124 leaf = (struct gfs2_leaf *)bh->b_data;
2125 len = 1 << (dip->i_depth - be16_to_cpu(leaf->lf_depth));
2127 next_index = (index & ~(len - 1)) + len;
2128 last = ((next_index >= hsize) ? 1 : 0);
2129 error = leaf_dealloc(dip, index, len, leaf_no, bh,
2139 if (index != hsize) {
2140 gfs2_consist_inode(dip);
2150 * gfs2_diradd_alloc_required - find if adding entry will require an allocation
2151 * @ip: the file being written to
2152 * @filname: the filename that's going to be added
2153 * @da: The structure to return dir alloc info
2155 * Returns: 0 if ok, -ve on error
2158 int gfs2_diradd_alloc_required(struct inode *inode, const struct qstr *name,
2159 struct gfs2_diradd *da)
2161 struct gfs2_inode *ip = GFS2_I(inode);
2162 struct gfs2_sbd *sdp = GFS2_SB(inode);
2163 const unsigned int extra = sizeof(struct gfs2_dinode) - sizeof(struct gfs2_leaf);
2164 struct gfs2_dirent *dent;
2165 struct buffer_head *bh;
2171 dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, &bh);
2173 da->nr_blocks = sdp->sd_max_dirres;
2174 if (!(ip->i_diskflags & GFS2_DIF_EXHASH) &&
2175 (GFS2_DIRENT_SIZE(name->len) < extra))
2180 return PTR_ERR(dent);