1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
8 * Implements Extendible Hashing as described in:
9 * "Extendible Hashing" by Fagin, et al in
10 * __ACM Trans. on Database Systems__, Sept 1979.
13 * Here's the layout of dirents which is essentially the same as that of ext2
14 * within a single block. The field de_name_len is the number of bytes
15 * actually required for the name (no null terminator). The field de_rec_len
16 * is the number of bytes allocated to the dirent. The offset of the next
17 * dirent in the block is (dirent + dirent->de_rec_len). When a dirent is
18 * deleted, the preceding dirent inherits its allocated space, ie
19 * prev->de_rec_len += deleted->de_rec_len. Since the next dirent is obtained
20 * by adding de_rec_len to the current dirent, this essentially causes the
21 * deleted dirent to get jumped over when iterating through all the dirents.
23 * When deleting the first dirent in a block, there is no previous dirent so
24 * the field de_ino is set to zero to designate it as deleted. When allocating
25 * a dirent, gfs2_dirent_alloc iterates through the dirents in a block. If the
26 * first dirent has (de_ino == 0) and de_rec_len is large enough, this first
27 * dirent is allocated. Otherwise it must go through all the 'used' dirents
28 * searching for one in which the amount of total space minus the amount of
29 * used space will provide enough space for the new dirent.
31 * There are two types of blocks in which dirents reside. In a stuffed dinode,
32 * the dirents begin at offset sizeof(struct gfs2_dinode) from the beginning of
33 * the block. In leaves, they begin at offset sizeof(struct gfs2_leaf) from the
34 * beginning of the leaf block. The dirents reside in leaves when
36 * dip->i_diskflags & GFS2_DIF_EXHASH is true
38 * Otherwise, the dirents are "linear", within a single stuffed dinode block.
40 * When the dirents are in leaves, the actual contents of the directory file are
41 * used as an array of 64-bit block pointers pointing to the leaf blocks. The
42 * dirents are NOT in the directory file itself. There can be more than one
43 * block pointer in the array that points to the same leaf. In fact, when a
44 * directory is first converted from linear to exhash, all of the pointers
45 * point to the same leaf.
47 * When a leaf is completely full, the size of the hash table can be
48 * doubled unless it is already at the maximum size which is hard coded into
49 * GFS2_DIR_MAX_DEPTH. After that, leaves are chained together in a linked list,
50 * but never before the maximum hash table size has been reached.
53 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
55 #include <linux/slab.h>
56 #include <linux/spinlock.h>
57 #include <linux/buffer_head.h>
58 #include <linux/sort.h>
59 #include <linux/gfs2_ondisk.h>
60 #include <linux/crc32.h>
61 #include <linux/vmalloc.h>
62 #include <linux/bio.h>
76 #define MAX_RA_BLOCKS 32 /* max read-ahead blocks */
78 #define gfs2_disk_hash2offset(h) (((u64)(h)) >> 1)
79 #define gfs2_dir_offset2hash(p) ((u32)(((u64)(p)) << 1))
80 #define GFS2_HASH_INDEX_MASK 0xffffc000
81 #define GFS2_USE_HASH_FLAG 0x2000
83 struct qstr gfs2_qdot __read_mostly;
84 struct qstr gfs2_qdotdot __read_mostly;
86 typedef int (*gfs2_dscan_t)(const struct gfs2_dirent *dent,
87 const struct qstr *name, void *opaque);
89 int gfs2_dir_get_new_buffer(struct gfs2_inode *ip, u64 block,
90 struct buffer_head **bhp)
92 struct buffer_head *bh;
94 bh = gfs2_meta_new(ip->i_gl, block);
95 gfs2_trans_add_meta(ip->i_gl, bh);
96 gfs2_metatype_set(bh, GFS2_METATYPE_JD, GFS2_FORMAT_JD);
97 gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
102 static int gfs2_dir_get_existing_buffer(struct gfs2_inode *ip, u64 block,
103 struct buffer_head **bhp)
105 struct buffer_head *bh;
108 error = gfs2_meta_read(ip->i_gl, block, DIO_WAIT, 0, &bh);
111 if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), bh, GFS2_METATYPE_JD)) {
119 static int gfs2_dir_write_stuffed(struct gfs2_inode *ip, const char *buf,
120 unsigned int offset, unsigned int size)
122 struct buffer_head *dibh;
125 error = gfs2_meta_inode_buffer(ip, &dibh);
129 gfs2_trans_add_meta(ip->i_gl, dibh);
130 memcpy(dibh->b_data + offset + sizeof(struct gfs2_dinode), buf, size);
131 if (ip->i_inode.i_size < offset + size)
132 i_size_write(&ip->i_inode, offset + size);
133 ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode);
134 gfs2_dinode_out(ip, dibh->b_data);
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
150 * Returns: The number of bytes correctly written or error code
152 static int gfs2_dir_write_data(struct gfs2_inode *ip, const char *buf,
153 u64 offset, unsigned int size)
155 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
156 struct buffer_head *dibh;
167 if (gfs2_is_stuffed(ip) && offset + size <= gfs2_max_stuffed_size(ip))
168 return gfs2_dir_write_stuffed(ip, buf, (unsigned int)offset,
171 if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
174 if (gfs2_is_stuffed(ip)) {
175 error = gfs2_unstuff_dinode(ip, NULL);
181 o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
183 while (copied < size) {
185 struct buffer_head *bh;
187 amount = size - copied;
188 if (amount > sdp->sd_sb.sb_bsize - o)
189 amount = sdp->sd_sb.sb_bsize - o;
193 error = gfs2_extent_map(&ip->i_inode, lblock, &new,
198 if (gfs2_assert_withdraw(sdp, dblock))
202 if (amount == sdp->sd_jbsize || new)
203 error = gfs2_dir_get_new_buffer(ip, dblock, &bh);
205 error = gfs2_dir_get_existing_buffer(ip, dblock, &bh);
210 gfs2_trans_add_meta(ip->i_gl, bh);
211 memcpy(bh->b_data + o, buf, amount);
220 o = sizeof(struct gfs2_meta_header);
224 error = gfs2_meta_inode_buffer(ip, &dibh);
228 if (ip->i_inode.i_size < offset + copied)
229 i_size_write(&ip->i_inode, offset + copied);
230 ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode);
232 gfs2_trans_add_meta(ip->i_gl, dibh);
233 gfs2_dinode_out(ip, dibh->b_data);
243 static int gfs2_dir_read_stuffed(struct gfs2_inode *ip, __be64 *buf,
246 struct buffer_head *dibh;
249 error = gfs2_meta_inode_buffer(ip, &dibh);
251 memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size);
255 return (error) ? error : size;
260 * gfs2_dir_read_data - Read a data from a directory inode
261 * @ip: The GFS2 Inode
262 * @buf: The buffer to place result into
263 * @size: Amount of data to transfer
265 * Returns: The amount of data actually copied or the error
267 static int gfs2_dir_read_data(struct gfs2_inode *ip, __be64 *buf,
270 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
277 if (gfs2_is_stuffed(ip))
278 return gfs2_dir_read_stuffed(ip, buf, size);
280 if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
284 o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
286 while (copied < size) {
288 struct buffer_head *bh;
291 amount = size - copied;
292 if (amount > sdp->sd_sb.sb_bsize - o)
293 amount = sdp->sd_sb.sb_bsize - o;
297 error = gfs2_extent_map(&ip->i_inode, lblock, &new,
299 if (error || !dblock)
302 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
304 error = gfs2_meta_read(ip->i_gl, dblock, DIO_WAIT, 0, &bh);
308 error = gfs2_metatype_check(sdp, bh, GFS2_METATYPE_JD);
315 memcpy(buf, bh->b_data + o, amount);
317 buf += (amount/sizeof(__be64));
320 o = sizeof(struct gfs2_meta_header);
325 return (copied) ? copied : error;
329 * gfs2_dir_get_hash_table - Get pointer to the dir hash table
330 * @ip: The inode in question
332 * Returns: The hash table or an error
335 static __be64 *gfs2_dir_get_hash_table(struct gfs2_inode *ip)
337 struct inode *inode = &ip->i_inode;
342 BUG_ON(!(ip->i_diskflags & GFS2_DIF_EXHASH));
344 hc = ip->i_hash_cache;
348 hsize = BIT(ip->i_depth);
349 hsize *= sizeof(__be64);
350 if (hsize != i_size_read(&ip->i_inode)) {
351 gfs2_consist_inode(ip);
352 return ERR_PTR(-EIO);
355 hc = kmalloc(hsize, GFP_NOFS | __GFP_NOWARN);
357 hc = __vmalloc(hsize, GFP_NOFS, PAGE_KERNEL);
360 return ERR_PTR(-ENOMEM);
362 ret = gfs2_dir_read_data(ip, hc, hsize);
368 spin_lock(&inode->i_lock);
369 if (likely(!ip->i_hash_cache)) {
370 ip->i_hash_cache = hc;
373 spin_unlock(&inode->i_lock);
376 return ip->i_hash_cache;
380 * gfs2_dir_hash_inval - Invalidate dir hash
381 * @ip: The directory inode
383 * Must be called with an exclusive glock, or during glock invalidation.
385 void gfs2_dir_hash_inval(struct gfs2_inode *ip)
389 spin_lock(&ip->i_inode.i_lock);
390 hc = ip->i_hash_cache;
391 ip->i_hash_cache = NULL;
392 spin_unlock(&ip->i_inode.i_lock);
397 static inline int gfs2_dirent_sentinel(const struct gfs2_dirent *dent)
399 return dent->de_inum.no_addr == 0 || dent->de_inum.no_formal_ino == 0;
402 static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent,
403 const struct qstr *name, int ret)
405 if (!gfs2_dirent_sentinel(dent) &&
406 be32_to_cpu(dent->de_hash) == name->hash &&
407 be16_to_cpu(dent->de_name_len) == name->len &&
408 memcmp(dent+1, name->name, name->len) == 0)
413 static int gfs2_dirent_find(const struct gfs2_dirent *dent,
414 const struct qstr *name,
417 return __gfs2_dirent_find(dent, name, 1);
420 static int gfs2_dirent_prev(const struct gfs2_dirent *dent,
421 const struct qstr *name,
424 return __gfs2_dirent_find(dent, name, 2);
428 * name->name holds ptr to start of block.
429 * name->len holds size of block.
431 static int gfs2_dirent_last(const struct gfs2_dirent *dent,
432 const struct qstr *name,
435 const char *start = name->name;
436 const char *end = (const char *)dent + be16_to_cpu(dent->de_rec_len);
437 if (name->len == (end - start))
442 /* Look for the dirent that contains the offset specified in data. Once we
443 * find that dirent, there must be space available there for the new dirent */
444 static int gfs2_dirent_find_offset(const struct gfs2_dirent *dent,
445 const struct qstr *name,
448 unsigned required = GFS2_DIRENT_SIZE(name->len);
449 unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
450 unsigned totlen = be16_to_cpu(dent->de_rec_len);
452 if (ptr < (void *)dent || ptr >= (void *)dent + totlen)
454 if (gfs2_dirent_sentinel(dent))
456 if (ptr < (void *)dent + actual)
458 if ((void *)dent + totlen >= ptr + required)
463 static int gfs2_dirent_find_space(const struct gfs2_dirent *dent,
464 const struct qstr *name,
467 unsigned required = GFS2_DIRENT_SIZE(name->len);
468 unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
469 unsigned totlen = be16_to_cpu(dent->de_rec_len);
471 if (gfs2_dirent_sentinel(dent))
473 if (totlen - actual >= required)
478 struct dirent_gather {
479 const struct gfs2_dirent **pdent;
483 static int gfs2_dirent_gather(const struct gfs2_dirent *dent,
484 const struct qstr *name,
487 struct dirent_gather *g = opaque;
488 if (!gfs2_dirent_sentinel(dent)) {
489 g->pdent[g->offset++] = dent;
495 * Other possible things to check:
496 * - Inode located within filesystem size (and on valid block)
497 * - Valid directory entry type
498 * Not sure how heavy-weight we want to make this... could also check
499 * hash is correct for example, but that would take a lot of extra time.
500 * For now the most important thing is to check that the various sizes
503 static int gfs2_check_dirent(struct gfs2_sbd *sdp,
504 struct gfs2_dirent *dent, unsigned int offset,
505 unsigned int size, unsigned int len, int first)
507 const char *msg = "gfs2_dirent too small";
508 if (unlikely(size < sizeof(struct gfs2_dirent)))
510 msg = "gfs2_dirent misaligned";
511 if (unlikely(offset & 0x7))
513 msg = "gfs2_dirent points beyond end of block";
514 if (unlikely(offset + size > len))
516 msg = "zero inode number";
517 if (unlikely(!first && gfs2_dirent_sentinel(dent)))
519 msg = "name length is greater than space in dirent";
520 if (!gfs2_dirent_sentinel(dent) &&
521 unlikely(sizeof(struct gfs2_dirent)+be16_to_cpu(dent->de_name_len) >
526 fs_warn(sdp, "%s: %s (%s)\n",
527 __func__, msg, first ? "first in block" : "not first in block");
531 static int gfs2_dirent_offset(struct gfs2_sbd *sdp, const void *buf)
533 const struct gfs2_meta_header *h = buf;
538 switch(be32_to_cpu(h->mh_type)) {
539 case GFS2_METATYPE_LF:
540 offset = sizeof(struct gfs2_leaf);
542 case GFS2_METATYPE_DI:
543 offset = sizeof(struct gfs2_dinode);
550 fs_warn(sdp, "%s: wrong block type %u\n", __func__,
551 be32_to_cpu(h->mh_type));
555 static struct gfs2_dirent *gfs2_dirent_scan(struct inode *inode, void *buf,
556 unsigned int len, gfs2_dscan_t scan,
557 const struct qstr *name,
560 struct gfs2_dirent *dent, *prev;
565 ret = gfs2_dirent_offset(GFS2_SB(inode), buf);
572 size = be16_to_cpu(dent->de_rec_len);
573 if (gfs2_check_dirent(GFS2_SB(inode), dent, offset, size, len, 1))
576 ret = scan(dent, name, opaque);
584 size = be16_to_cpu(dent->de_rec_len);
585 if (gfs2_check_dirent(GFS2_SB(inode), dent, offset, size,
596 return prev ? prev : dent;
603 gfs2_consist_inode(GFS2_I(inode));
604 return ERR_PTR(-EIO);
607 static int dirent_check_reclen(struct gfs2_inode *dip,
608 const struct gfs2_dirent *d, const void *end_p)
611 u16 rec_len = be16_to_cpu(d->de_rec_len);
613 if (unlikely(rec_len < sizeof(struct gfs2_dirent)))
621 gfs2_consist_inode(dip);
626 * dirent_next - Next dirent
627 * @dip: the directory
629 * @dent: Pointer to list of dirents
631 * Returns: 0 on success, error code otherwise
634 static int dirent_next(struct gfs2_inode *dip, struct buffer_head *bh,
635 struct gfs2_dirent **dent)
637 struct gfs2_dirent *cur = *dent, *tmp;
638 char *bh_end = bh->b_data + bh->b_size;
641 ret = dirent_check_reclen(dip, cur, bh_end);
645 tmp = (void *)cur + ret;
646 ret = dirent_check_reclen(dip, tmp, bh_end);
650 /* Only the first dent could ever have de_inum.no_addr == 0 */
651 if (gfs2_dirent_sentinel(tmp)) {
652 gfs2_consist_inode(dip);
661 * dirent_del - Delete a dirent
662 * @dip: The GFS2 inode
664 * @prev: The previous dirent
665 * @cur: The current dirent
669 static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh,
670 struct gfs2_dirent *prev, struct gfs2_dirent *cur)
672 u16 cur_rec_len, prev_rec_len;
674 if (gfs2_dirent_sentinel(cur)) {
675 gfs2_consist_inode(dip);
679 gfs2_trans_add_meta(dip->i_gl, bh);
681 /* If there is no prev entry, this is the first entry in the block.
682 The de_rec_len is already as big as it needs to be. Just zero
683 out the inode number and return. */
686 cur->de_inum.no_addr = 0;
687 cur->de_inum.no_formal_ino = 0;
691 /* Combine this dentry with the previous one. */
693 prev_rec_len = be16_to_cpu(prev->de_rec_len);
694 cur_rec_len = be16_to_cpu(cur->de_rec_len);
696 if ((char *)prev + prev_rec_len != (char *)cur)
697 gfs2_consist_inode(dip);
698 if ((char *)cur + cur_rec_len > bh->b_data + bh->b_size)
699 gfs2_consist_inode(dip);
701 prev_rec_len += cur_rec_len;
702 prev->de_rec_len = cpu_to_be16(prev_rec_len);
706 static struct gfs2_dirent *do_init_dirent(struct inode *inode,
707 struct gfs2_dirent *dent,
708 const struct qstr *name,
709 struct buffer_head *bh,
712 struct gfs2_inode *ip = GFS2_I(inode);
713 struct gfs2_dirent *ndent;
716 totlen = be16_to_cpu(dent->de_rec_len);
717 BUG_ON(offset + name->len > totlen);
718 gfs2_trans_add_meta(ip->i_gl, bh);
719 ndent = (struct gfs2_dirent *)((char *)dent + offset);
720 dent->de_rec_len = cpu_to_be16(offset);
721 gfs2_qstr2dirent(name, totlen - offset, ndent);
727 * Takes a dent from which to grab space as an argument. Returns the
728 * newly created dent.
730 static struct gfs2_dirent *gfs2_init_dirent(struct inode *inode,
731 struct gfs2_dirent *dent,
732 const struct qstr *name,
733 struct buffer_head *bh)
737 if (!gfs2_dirent_sentinel(dent))
738 offset = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
739 return do_init_dirent(inode, dent, name, bh, offset);
742 static struct gfs2_dirent *gfs2_dirent_split_alloc(struct inode *inode,
743 struct buffer_head *bh,
744 const struct qstr *name,
747 struct gfs2_dirent *dent;
748 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
749 gfs2_dirent_find_offset, name, ptr);
750 if (IS_ERR_OR_NULL(dent))
752 return do_init_dirent(inode, dent, name, bh,
753 (unsigned)(ptr - (void *)dent));
756 static int get_leaf(struct gfs2_inode *dip, u64 leaf_no,
757 struct buffer_head **bhp)
761 error = gfs2_meta_read(dip->i_gl, leaf_no, DIO_WAIT, 0, bhp);
762 if (!error && gfs2_metatype_check(GFS2_SB(&dip->i_inode), *bhp, GFS2_METATYPE_LF)) {
763 /* pr_info("block num=%llu\n", leaf_no); */
771 * get_leaf_nr - Get a leaf number associated with the index
772 * @dip: The GFS2 inode
776 * Returns: 0 on success, error code otherwise
779 static int get_leaf_nr(struct gfs2_inode *dip, u32 index,
785 hash = gfs2_dir_get_hash_table(dip);
786 error = PTR_ERR_OR_ZERO(hash);
789 *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 int hsize = BIT(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 (IS_ERR_OR_NULL(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 timespec64 tv = current_time(inode);
873 error = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL);
876 bh = gfs2_meta_new(ip->i_gl, bn);
880 gfs2_trans_remove_revoke(GFS2_SB(inode), bn, 1);
881 gfs2_trans_add_meta(ip->i_gl, bh);
882 gfs2_metatype_set(bh, GFS2_METATYPE_LF, GFS2_FORMAT_LF);
883 leaf = (struct gfs2_leaf *)bh->b_data;
884 leaf->lf_depth = cpu_to_be16(depth);
885 leaf->lf_entries = 0;
886 leaf->lf_dirent_format = cpu_to_be32(GFS2_FORMAT_DE);
888 leaf->lf_inode = cpu_to_be64(ip->i_no_addr);
889 leaf->lf_dist = cpu_to_be32(1);
890 leaf->lf_nsec = cpu_to_be32(tv.tv_nsec);
891 leaf->lf_sec = cpu_to_be64(tv.tv_sec);
892 memset(leaf->lf_reserved2, 0, sizeof(leaf->lf_reserved2));
893 dent = (struct gfs2_dirent *)(leaf+1);
894 gfs2_qstr2dirent(&empty_name, bh->b_size - sizeof(struct gfs2_leaf), dent);
900 * dir_make_exhash - Convert a stuffed directory into an ExHash directory
901 * @dip: The GFS2 inode
903 * Returns: 0 on success, error code otherwise
906 static int dir_make_exhash(struct inode *inode)
908 struct gfs2_inode *dip = GFS2_I(inode);
909 struct gfs2_sbd *sdp = GFS2_SB(inode);
910 struct gfs2_dirent *dent;
912 struct buffer_head *bh, *dibh;
913 struct gfs2_leaf *leaf;
920 error = gfs2_meta_inode_buffer(dip, &dibh);
924 /* Turn over a new leaf */
926 leaf = new_leaf(inode, &bh, 0);
931 gfs2_assert(sdp, dip->i_entries < BIT(16));
932 leaf->lf_entries = cpu_to_be16(dip->i_entries);
936 gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_leaf), dibh,
937 sizeof(struct gfs2_dinode));
939 /* Find last entry */
942 args.len = bh->b_size - sizeof(struct gfs2_dinode) +
943 sizeof(struct gfs2_leaf);
944 args.name = bh->b_data;
945 dent = gfs2_dirent_scan(&dip->i_inode, bh->b_data, bh->b_size,
946 gfs2_dirent_last, &args, NULL);
955 return PTR_ERR(dent);
958 /* Adjust the last dirent's record length
959 (Remember that dent still points to the last entry.) */
961 dent->de_rec_len = cpu_to_be16(be16_to_cpu(dent->de_rec_len) +
962 sizeof(struct gfs2_dinode) -
963 sizeof(struct gfs2_leaf));
967 /* We're done with the new leaf block, now setup the new
970 gfs2_trans_add_meta(dip->i_gl, dibh);
971 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
973 lp = (__be64 *)(dibh->b_data + sizeof(struct gfs2_dinode));
975 for (x = sdp->sd_hash_ptrs; x--; lp++)
976 *lp = cpu_to_be64(bn);
978 i_size_write(inode, sdp->sd_sb.sb_bsize / 2);
979 gfs2_add_inode_blocks(&dip->i_inode, 1);
980 dip->i_diskflags |= GFS2_DIF_EXHASH;
982 for (x = sdp->sd_hash_ptrs, y = -1; x; x >>= 1, y++) ;
985 gfs2_dinode_out(dip, dibh->b_data);
993 * dir_split_leaf - Split a leaf block into two
994 * @dip: The GFS2 inode
998 * Returns: 0 on success, error code on failure
1001 static int dir_split_leaf(struct inode *inode, const struct qstr *name)
1003 struct gfs2_inode *dip = GFS2_I(inode);
1004 struct buffer_head *nbh, *obh, *dibh;
1005 struct gfs2_leaf *nleaf, *oleaf;
1006 struct gfs2_dirent *dent = NULL, *prev = NULL, *next = NULL, *new;
1007 u32 start, len, half_len, divider;
1014 index = name->hash >> (32 - dip->i_depth);
1015 error = get_leaf_nr(dip, index, &leaf_no);
1019 /* Get the old leaf block */
1020 error = get_leaf(dip, leaf_no, &obh);
1024 oleaf = (struct gfs2_leaf *)obh->b_data;
1025 if (dip->i_depth == be16_to_cpu(oleaf->lf_depth)) {
1027 return 1; /* can't split */
1030 gfs2_trans_add_meta(dip->i_gl, obh);
1032 nleaf = new_leaf(inode, &nbh, be16_to_cpu(oleaf->lf_depth) + 1);
1037 bn = nbh->b_blocknr;
1039 /* Compute the start and len of leaf pointers in the hash table. */
1040 len = BIT(dip->i_depth - be16_to_cpu(oleaf->lf_depth));
1041 half_len = len >> 1;
1043 fs_warn(GFS2_SB(inode), "i_depth %u lf_depth %u index %u\n",
1044 dip->i_depth, be16_to_cpu(oleaf->lf_depth), index);
1045 gfs2_consist_inode(dip);
1050 start = (index & ~(len - 1));
1052 /* Change the pointers.
1053 Don't bother distinguishing stuffed from non-stuffed.
1054 This code is complicated enough already. */
1055 lp = kmalloc_array(half_len, sizeof(__be64), GFP_NOFS);
1061 /* Change the pointers */
1062 for (x = 0; x < half_len; x++)
1063 lp[x] = cpu_to_be64(bn);
1065 gfs2_dir_hash_inval(dip);
1067 error = gfs2_dir_write_data(dip, (char *)lp, start * sizeof(u64),
1068 half_len * sizeof(u64));
1069 if (error != half_len * sizeof(u64)) {
1077 /* Compute the divider */
1078 divider = (start + half_len) << (32 - dip->i_depth);
1080 /* Copy the entries */
1081 dent = (struct gfs2_dirent *)(obh->b_data + sizeof(struct gfs2_leaf));
1085 if (dirent_next(dip, obh, &next))
1088 if (!gfs2_dirent_sentinel(dent) &&
1089 be32_to_cpu(dent->de_hash) < divider) {
1091 void *ptr = ((char *)dent - obh->b_data) + nbh->b_data;
1092 str.name = (char*)(dent+1);
1093 str.len = be16_to_cpu(dent->de_name_len);
1094 str.hash = be32_to_cpu(dent->de_hash);
1095 new = gfs2_dirent_split_alloc(inode, nbh, &str, ptr);
1097 error = PTR_ERR(new);
1101 new->de_inum = dent->de_inum; /* No endian worries */
1102 new->de_type = dent->de_type; /* No endian worries */
1103 be16_add_cpu(&nleaf->lf_entries, 1);
1105 dirent_del(dip, obh, prev, dent);
1107 if (!oleaf->lf_entries)
1108 gfs2_consist_inode(dip);
1109 be16_add_cpu(&oleaf->lf_entries, -1);
1119 oleaf->lf_depth = nleaf->lf_depth;
1121 error = gfs2_meta_inode_buffer(dip, &dibh);
1122 if (!gfs2_assert_withdraw(GFS2_SB(&dip->i_inode), !error)) {
1123 gfs2_trans_add_meta(dip->i_gl, dibh);
1124 gfs2_add_inode_blocks(&dip->i_inode, 1);
1125 gfs2_dinode_out(dip, dibh->b_data);
1144 * dir_double_exhash - Double size of ExHash table
1145 * @dip: The GFS2 dinode
1147 * Returns: 0 on success, error code on failure
1150 static int dir_double_exhash(struct gfs2_inode *dip)
1152 struct buffer_head *dibh;
1160 hsize = BIT(dip->i_depth);
1161 hsize_bytes = hsize * sizeof(__be64);
1163 hc = gfs2_dir_get_hash_table(dip);
1167 hc2 = kmalloc_array(hsize_bytes, 2, GFP_NOFS | __GFP_NOWARN);
1169 hc2 = __vmalloc(hsize_bytes * 2, GFP_NOFS, PAGE_KERNEL);
1175 error = gfs2_meta_inode_buffer(dip, &dibh);
1179 for (x = 0; x < hsize; x++) {
1185 error = gfs2_dir_write_data(dip, (char *)hc2, 0, hsize_bytes * 2);
1186 if (error != (hsize_bytes * 2))
1189 gfs2_dir_hash_inval(dip);
1190 dip->i_hash_cache = hc2;
1192 gfs2_dinode_out(dip, dibh->b_data);
1197 /* Replace original hash table & size */
1198 gfs2_dir_write_data(dip, (char *)hc, 0, hsize_bytes);
1199 i_size_write(&dip->i_inode, hsize_bytes);
1200 gfs2_dinode_out(dip, dibh->b_data);
1208 * compare_dents - compare directory entries by hash value
1212 * When comparing the hash entries of @a to @b:
1218 static int compare_dents(const void *a, const void *b)
1220 const struct gfs2_dirent *dent_a, *dent_b;
1224 dent_a = *(const struct gfs2_dirent **)a;
1225 hash_a = dent_a->de_cookie;
1227 dent_b = *(const struct gfs2_dirent **)b;
1228 hash_b = dent_b->de_cookie;
1230 if (hash_a > hash_b)
1232 else if (hash_a < hash_b)
1235 unsigned int len_a = be16_to_cpu(dent_a->de_name_len);
1236 unsigned int len_b = be16_to_cpu(dent_b->de_name_len);
1240 else if (len_a < len_b)
1243 ret = memcmp(dent_a + 1, dent_b + 1, len_a);
1250 * do_filldir_main - read out directory entries
1251 * @dip: The GFS2 inode
1252 * @ctx: what to feed the entries to
1253 * @darr: an array of struct gfs2_dirent pointers to read
1254 * @entries: the number of entries in darr
1255 * @copied: pointer to int that's non-zero if a entry has been copied out
1257 * Jump through some hoops to make sure that if there are hash collsions,
1258 * they are read out at the beginning of a buffer. We want to minimize
1259 * the possibility that they will fall into different readdir buffers or
1260 * that someone will want to seek to that location.
1262 * Returns: errno, >0 if the actor tells you to stop
1265 static int do_filldir_main(struct gfs2_inode *dip, struct dir_context *ctx,
1266 struct gfs2_dirent **darr, u32 entries,
1267 u32 sort_start, int *copied)
1269 const struct gfs2_dirent *dent, *dent_next;
1274 if (sort_start < entries)
1275 sort(&darr[sort_start], entries - sort_start,
1276 sizeof(struct gfs2_dirent *), compare_dents, NULL);
1278 dent_next = darr[0];
1279 off_next = dent_next->de_cookie;
1281 for (x = 0, y = 1; x < entries; x++, y++) {
1286 dent_next = darr[y];
1287 off_next = dent_next->de_cookie;
1293 if (off_next == off) {
1294 if (*copied && !run)
1305 if (!dir_emit(ctx, (const char *)(dent + 1),
1306 be16_to_cpu(dent->de_name_len),
1307 be64_to_cpu(dent->de_inum.no_addr),
1308 be16_to_cpu(dent->de_type)))
1314 /* Increment the ctx->pos by one, so the next time we come into the
1315 do_filldir fxn, we get the next entry instead of the last one in the
1323 static void *gfs2_alloc_sort_buffer(unsigned size)
1327 if (size < KMALLOC_MAX_SIZE)
1328 ptr = kmalloc(size, GFP_NOFS | __GFP_NOWARN);
1330 ptr = __vmalloc(size, GFP_NOFS, PAGE_KERNEL);
1335 static int gfs2_set_cookies(struct gfs2_sbd *sdp, struct buffer_head *bh,
1336 unsigned leaf_nr, struct gfs2_dirent **darr,
1342 for (i = 0; i < entries; i++) {
1345 darr[i]->de_cookie = be32_to_cpu(darr[i]->de_hash);
1346 darr[i]->de_cookie = gfs2_disk_hash2offset(darr[i]->de_cookie);
1348 if (!sdp->sd_args.ar_loccookie)
1350 offset = (char *)(darr[i]) -
1351 (bh->b_data + gfs2_dirent_offset(sdp, bh->b_data));
1352 offset /= GFS2_MIN_DIRENT_SIZE;
1353 offset += leaf_nr * sdp->sd_max_dents_per_leaf;
1354 if (offset >= GFS2_USE_HASH_FLAG ||
1355 leaf_nr >= GFS2_USE_HASH_FLAG) {
1356 darr[i]->de_cookie |= GFS2_USE_HASH_FLAG;
1361 darr[i]->de_cookie &= GFS2_HASH_INDEX_MASK;
1362 darr[i]->de_cookie |= offset;
1368 static int gfs2_dir_read_leaf(struct inode *inode, struct dir_context *ctx,
1369 int *copied, unsigned *depth,
1372 struct gfs2_inode *ip = GFS2_I(inode);
1373 struct gfs2_sbd *sdp = GFS2_SB(inode);
1374 struct buffer_head *bh;
1375 struct gfs2_leaf *lf;
1376 unsigned entries = 0, entries2 = 0;
1377 unsigned leaves = 0, leaf = 0, offset, sort_offset;
1378 struct gfs2_dirent **darr, *dent;
1379 struct dirent_gather g;
1380 struct buffer_head **larr;
1381 int error, i, need_sort = 0, sort_id;
1385 error = get_leaf(ip, lfn, &bh);
1388 lf = (struct gfs2_leaf *)bh->b_data;
1390 *depth = be16_to_cpu(lf->lf_depth);
1391 entries += be16_to_cpu(lf->lf_entries);
1393 lfn = be64_to_cpu(lf->lf_next);
1397 if (*depth < GFS2_DIR_MAX_DEPTH || !sdp->sd_args.ar_loccookie) {
1407 * The extra 99 entries are not normally used, but are a buffer
1408 * zone in case the number of entries in the leaf is corrupt.
1409 * 99 is the maximum number of entries that can fit in a single
1412 larr = gfs2_alloc_sort_buffer((leaves + entries + 99) * sizeof(void *));
1415 darr = (struct gfs2_dirent **)(larr + leaves);
1416 g.pdent = (const struct gfs2_dirent **)darr;
1421 error = get_leaf(ip, lfn, &bh);
1424 lf = (struct gfs2_leaf *)bh->b_data;
1425 lfn = be64_to_cpu(lf->lf_next);
1426 if (lf->lf_entries) {
1428 entries2 += be16_to_cpu(lf->lf_entries);
1429 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
1430 gfs2_dirent_gather, NULL, &g);
1431 error = PTR_ERR(dent);
1434 if (entries2 != g.offset) {
1435 fs_warn(sdp, "Number of entries corrupt in dir "
1436 "leaf %llu, entries2 (%u) != "
1438 (unsigned long long)bh->b_blocknr,
1439 entries2, g.offset);
1440 gfs2_consist_inode(ip);
1445 sort_id = gfs2_set_cookies(sdp, bh, leaf, &darr[offset],
1446 be16_to_cpu(lf->lf_entries));
1447 if (!need_sort && sort_id >= 0) {
1449 sort_offset = offset + sort_id;
1453 larr[leaf++] = NULL;
1458 BUG_ON(entries2 != entries);
1459 error = do_filldir_main(ip, ctx, darr, entries, need_sort ?
1460 sort_offset : entries, copied);
1462 for(i = 0; i < leaf; i++)
1470 * gfs2_dir_readahead - Issue read-ahead requests for leaf blocks.
1472 * Note: we can't calculate each index like dir_e_read can because we don't
1473 * have the leaf, and therefore we don't have the depth, and therefore we
1474 * don't have the length. So we have to just read enough ahead to make up
1475 * for the loss of information.
1477 static void gfs2_dir_readahead(struct inode *inode, unsigned hsize, u32 index,
1478 struct file_ra_state *f_ra)
1480 struct gfs2_inode *ip = GFS2_I(inode);
1481 struct gfs2_glock *gl = ip->i_gl;
1482 struct buffer_head *bh;
1483 u64 blocknr = 0, last;
1486 /* First check if we've already read-ahead for the whole range. */
1487 if (index + MAX_RA_BLOCKS < f_ra->start)
1490 f_ra->start = max((pgoff_t)index, f_ra->start);
1491 for (count = 0; count < MAX_RA_BLOCKS; count++) {
1492 if (f_ra->start >= hsize) /* if exceeded the hash table */
1496 blocknr = be64_to_cpu(ip->i_hash_cache[f_ra->start]);
1498 if (blocknr == last)
1501 bh = gfs2_getbuf(gl, blocknr, 1);
1502 if (trylock_buffer(bh)) {
1503 if (buffer_uptodate(bh)) {
1508 bh->b_end_io = end_buffer_read_sync;
1509 submit_bh(REQ_OP_READ,
1510 REQ_RAHEAD | REQ_META | REQ_PRIO,
1519 * dir_e_read - Reads the entries from a directory into a filldir buffer
1520 * @dip: dinode pointer
1521 * @ctx: actor to feed the entries to
1526 static int dir_e_read(struct inode *inode, struct dir_context *ctx,
1527 struct file_ra_state *f_ra)
1529 struct gfs2_inode *dip = GFS2_I(inode);
1537 hsize = BIT(dip->i_depth);
1538 hash = gfs2_dir_offset2hash(ctx->pos);
1539 index = hash >> (32 - dip->i_depth);
1541 if (dip->i_hash_cache == NULL)
1543 lp = gfs2_dir_get_hash_table(dip);
1547 gfs2_dir_readahead(inode, hsize, index, f_ra);
1549 while (index < hsize) {
1550 error = gfs2_dir_read_leaf(inode, ctx,
1552 be64_to_cpu(lp[index]));
1556 len = BIT(dip->i_depth - depth);
1557 index = (index & ~(len - 1)) + len;
1565 int gfs2_dir_read(struct inode *inode, struct dir_context *ctx,
1566 struct file_ra_state *f_ra)
1568 struct gfs2_inode *dip = GFS2_I(inode);
1569 struct gfs2_sbd *sdp = GFS2_SB(inode);
1570 struct dirent_gather g;
1571 struct gfs2_dirent **darr, *dent;
1572 struct buffer_head *dibh;
1576 if (!dip->i_entries)
1579 if (dip->i_diskflags & GFS2_DIF_EXHASH)
1580 return dir_e_read(inode, ctx, f_ra);
1582 if (!gfs2_is_stuffed(dip)) {
1583 gfs2_consist_inode(dip);
1587 error = gfs2_meta_inode_buffer(dip, &dibh);
1592 /* 96 is max number of dirents which can be stuffed into an inode */
1593 darr = kmalloc_array(96, sizeof(struct gfs2_dirent *), GFP_NOFS);
1595 g.pdent = (const struct gfs2_dirent **)darr;
1597 dent = gfs2_dirent_scan(inode, dibh->b_data, dibh->b_size,
1598 gfs2_dirent_gather, NULL, &g);
1600 error = PTR_ERR(dent);
1603 if (dip->i_entries != g.offset) {
1604 fs_warn(sdp, "Number of entries corrupt in dir %llu, "
1605 "ip->i_entries (%u) != g.offset (%u)\n",
1606 (unsigned long long)dip->i_no_addr,
1609 gfs2_consist_inode(dip);
1613 gfs2_set_cookies(sdp, dibh, 0, darr, dip->i_entries);
1614 error = do_filldir_main(dip, ctx, darr,
1615 dip->i_entries, 0, &copied);
1629 * gfs2_dir_search - Search a directory
1630 * @dip: The GFS2 dir inode
1631 * @name: The name we are looking up
1632 * @fail_on_exist: Fail if the name exists rather than looking it up
1634 * This routine searches a directory for a file or another directory.
1635 * Assumes a glock is held on dip.
1640 struct inode *gfs2_dir_search(struct inode *dir, const struct qstr *name,
1643 struct buffer_head *bh;
1644 struct gfs2_dirent *dent;
1645 u64 addr, formal_ino;
1648 dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
1650 struct inode *inode;
1654 return ERR_CAST(dent);
1655 dtype = be16_to_cpu(dent->de_type);
1656 rahead = be16_to_cpu(dent->de_rahead);
1657 addr = be64_to_cpu(dent->de_inum.no_addr);
1658 formal_ino = be64_to_cpu(dent->de_inum.no_formal_ino);
1661 return ERR_PTR(-EEXIST);
1662 inode = gfs2_inode_lookup(dir->i_sb, dtype, addr, formal_ino,
1663 GFS2_BLKST_FREE /* ignore */);
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;
1799 struct timespec64 tv;
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));
1815 tv = current_time(&ip->i_inode);
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 timespec64 tv = current_time(&dip->i_inode);
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;
1938 dent = gfs2_dirent_search(&dip->i_inode, filename, gfs2_dirent_find, &bh);
1940 gfs2_consist_inode(dip);
1944 return PTR_ERR(dent);
1946 gfs2_trans_add_meta(dip->i_gl, bh);
1947 gfs2_inum_out(nip, dent);
1948 dent->de_type = cpu_to_be16(new_type);
1951 dip->i_inode.i_mtime = dip->i_inode.i_ctime = current_time(&dip->i_inode);
1952 mark_inode_dirty_sync(&dip->i_inode);
1957 * leaf_dealloc - Deallocate a directory leaf
1958 * @dip: the directory
1959 * @index: the hash table offset in the directory
1960 * @len: the number of pointers to this leaf
1961 * @leaf_no: the leaf number
1962 * @leaf_bh: buffer_head for the starting leaf
1963 * last_dealloc: 1 if this is the final dealloc for the leaf, else 0
1968 static int leaf_dealloc(struct gfs2_inode *dip, u32 index, u32 len,
1969 u64 leaf_no, struct buffer_head *leaf_bh,
1972 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1973 struct gfs2_leaf *tmp_leaf;
1974 struct gfs2_rgrp_list rlist;
1975 struct buffer_head *bh, *dibh;
1977 unsigned int rg_blocks = 0, l_blocks = 0;
1979 unsigned int x, size = len * sizeof(u64);
1982 error = gfs2_rindex_update(sdp);
1986 memset(&rlist, 0, sizeof(struct gfs2_rgrp_list));
1988 ht = kzalloc(size, GFP_NOFS | __GFP_NOWARN);
1990 ht = __vmalloc(size, GFP_NOFS | __GFP_NOWARN | __GFP_ZERO,
1995 error = gfs2_quota_hold(dip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1999 /* Count the number of leaves */
2002 for (blk = leaf_no; blk; blk = nblk) {
2003 if (blk != leaf_no) {
2004 error = get_leaf(dip, blk, &bh);
2008 tmp_leaf = (struct gfs2_leaf *)bh->b_data;
2009 nblk = be64_to_cpu(tmp_leaf->lf_next);
2013 gfs2_rlist_add(dip, &rlist, blk);
2017 gfs2_rlist_alloc(&rlist);
2019 for (x = 0; x < rlist.rl_rgrps; x++) {
2020 struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(rlist.rl_ghs[x].gh_gl);
2022 rg_blocks += rgd->rd_length;
2025 error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs);
2029 error = gfs2_trans_begin(sdp,
2030 rg_blocks + (DIV_ROUND_UP(size, sdp->sd_jbsize) + 1) +
2031 RES_DINODE + RES_STATFS + RES_QUOTA, l_blocks);
2033 goto out_rg_gunlock;
2037 for (blk = leaf_no; blk; blk = nblk) {
2038 struct gfs2_rgrpd *rgd;
2040 if (blk != leaf_no) {
2041 error = get_leaf(dip, blk, &bh);
2045 tmp_leaf = (struct gfs2_leaf *)bh->b_data;
2046 nblk = be64_to_cpu(tmp_leaf->lf_next);
2050 rgd = gfs2_blk2rgrpd(sdp, blk, true);
2051 gfs2_free_meta(dip, rgd, blk, 1);
2052 gfs2_add_inode_blocks(&dip->i_inode, -1);
2055 error = gfs2_dir_write_data(dip, ht, index * sizeof(u64), size);
2056 if (error != size) {
2062 error = gfs2_meta_inode_buffer(dip, &dibh);
2066 gfs2_trans_add_meta(dip->i_gl, dibh);
2067 /* On the last dealloc, make this a regular file in case we crash.
2068 (We don't want to free these blocks a second time.) */
2070 dip->i_inode.i_mode = S_IFREG;
2071 gfs2_dinode_out(dip, dibh->b_data);
2075 gfs2_trans_end(sdp);
2077 gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
2079 gfs2_rlist_free(&rlist);
2080 gfs2_quota_unhold(dip);
2087 * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory
2088 * @dip: the directory
2090 * Dealloc all on-disk directory leaves to FREEMETA state
2091 * Change on-disk inode type to "regular file"
2096 int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip)
2098 struct buffer_head *bh;
2099 struct gfs2_leaf *leaf;
2101 u32 index = 0, next_index;
2104 int error = 0, last;
2106 hsize = BIT(dip->i_depth);
2108 lp = gfs2_dir_get_hash_table(dip);
2112 while (index < hsize) {
2113 leaf_no = be64_to_cpu(lp[index]);
2115 error = get_leaf(dip, leaf_no, &bh);
2118 leaf = (struct gfs2_leaf *)bh->b_data;
2119 len = BIT(dip->i_depth - be16_to_cpu(leaf->lf_depth));
2121 next_index = (index & ~(len - 1)) + len;
2122 last = ((next_index >= hsize) ? 1 : 0);
2123 error = leaf_dealloc(dip, index, len, leaf_no, bh,
2133 if (index != hsize) {
2134 gfs2_consist_inode(dip);
2144 * gfs2_diradd_alloc_required - find if adding entry will require an allocation
2145 * @ip: the file being written to
2146 * @filname: the filename that's going to be added
2147 * @da: The structure to return dir alloc info
2149 * Returns: 0 if ok, -ve on error
2152 int gfs2_diradd_alloc_required(struct inode *inode, const struct qstr *name,
2153 struct gfs2_diradd *da)
2155 struct gfs2_inode *ip = GFS2_I(inode);
2156 struct gfs2_sbd *sdp = GFS2_SB(inode);
2157 const unsigned int extra = sizeof(struct gfs2_dinode) - sizeof(struct gfs2_leaf);
2158 struct gfs2_dirent *dent;
2159 struct buffer_head *bh;
2165 dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, &bh);
2167 da->nr_blocks = sdp->sd_max_dirres;
2168 if (!(ip->i_diskflags & GFS2_DIF_EXHASH) &&
2169 (GFS2_DIRENT_SIZE(name->len) < extra))
2174 return PTR_ERR(dent);