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 inode_set_mtime_to_ts(&ip->i_inode, inode_set_ctime_current(&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);
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_alloc_extent(&ip->i_inode, lblock, &dblock,
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 inode_set_mtime_to_ts(&ip->i_inode, inode_set_ctime_current(&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;
290 amount = size - copied;
291 if (amount > sdp->sd_sb.sb_bsize - o)
292 amount = sdp->sd_sb.sb_bsize - o;
296 error = gfs2_get_extent(&ip->i_inode, lblock,
298 if (error || !dblock)
301 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
303 error = gfs2_meta_read(ip->i_gl, dblock, DIO_WAIT, 0, &bh);
307 error = gfs2_metatype_check(sdp, bh, GFS2_METATYPE_JD);
314 memcpy(buf, bh->b_data + o, amount);
316 buf += (amount/sizeof(__be64));
319 o = sizeof(struct gfs2_meta_header);
324 return (copied) ? copied : error;
328 * gfs2_dir_get_hash_table - Get pointer to the dir hash table
329 * @ip: The inode in question
331 * Returns: The hash table or an error
334 static __be64 *gfs2_dir_get_hash_table(struct gfs2_inode *ip)
336 struct inode *inode = &ip->i_inode;
341 BUG_ON(!(ip->i_diskflags & GFS2_DIF_EXHASH));
343 hc = ip->i_hash_cache;
347 hsize = BIT(ip->i_depth);
348 hsize *= sizeof(__be64);
349 if (hsize != i_size_read(&ip->i_inode)) {
350 gfs2_consist_inode(ip);
351 return ERR_PTR(-EIO);
354 hc = kmalloc(hsize, GFP_NOFS | __GFP_NOWARN);
356 hc = __vmalloc(hsize, GFP_NOFS);
359 return ERR_PTR(-ENOMEM);
361 ret = gfs2_dir_read_data(ip, hc, hsize);
367 spin_lock(&inode->i_lock);
368 if (likely(!ip->i_hash_cache)) {
369 ip->i_hash_cache = hc;
372 spin_unlock(&inode->i_lock);
375 return ip->i_hash_cache;
379 * gfs2_dir_hash_inval - Invalidate dir hash
380 * @ip: The directory inode
382 * Must be called with an exclusive glock, or during glock invalidation.
384 void gfs2_dir_hash_inval(struct gfs2_inode *ip)
388 spin_lock(&ip->i_inode.i_lock);
389 hc = ip->i_hash_cache;
390 ip->i_hash_cache = NULL;
391 spin_unlock(&ip->i_inode.i_lock);
396 static inline int gfs2_dirent_sentinel(const struct gfs2_dirent *dent)
398 return dent->de_inum.no_addr == 0 || dent->de_inum.no_formal_ino == 0;
401 static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent,
402 const struct qstr *name, int ret)
404 if (!gfs2_dirent_sentinel(dent) &&
405 be32_to_cpu(dent->de_hash) == name->hash &&
406 be16_to_cpu(dent->de_name_len) == name->len &&
407 memcmp(dent+1, name->name, name->len) == 0)
412 static int gfs2_dirent_find(const struct gfs2_dirent *dent,
413 const struct qstr *name,
416 return __gfs2_dirent_find(dent, name, 1);
419 static int gfs2_dirent_prev(const struct gfs2_dirent *dent,
420 const struct qstr *name,
423 return __gfs2_dirent_find(dent, name, 2);
427 * name->name holds ptr to start of block.
428 * name->len holds size of block.
430 static int gfs2_dirent_last(const struct gfs2_dirent *dent,
431 const struct qstr *name,
434 const char *start = name->name;
435 const char *end = (const char *)dent + be16_to_cpu(dent->de_rec_len);
436 if (name->len == (end - start))
441 /* Look for the dirent that contains the offset specified in data. Once we
442 * find that dirent, there must be space available there for the new dirent */
443 static int gfs2_dirent_find_offset(const struct gfs2_dirent *dent,
444 const struct qstr *name,
447 unsigned required = GFS2_DIRENT_SIZE(name->len);
448 unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
449 unsigned totlen = be16_to_cpu(dent->de_rec_len);
451 if (ptr < (void *)dent || ptr >= (void *)dent + totlen)
453 if (gfs2_dirent_sentinel(dent))
455 if (ptr < (void *)dent + actual)
457 if ((void *)dent + totlen >= ptr + required)
462 static int gfs2_dirent_find_space(const struct gfs2_dirent *dent,
463 const struct qstr *name,
466 unsigned required = GFS2_DIRENT_SIZE(name->len);
467 unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
468 unsigned totlen = be16_to_cpu(dent->de_rec_len);
470 if (gfs2_dirent_sentinel(dent))
472 if (totlen - actual >= required)
477 struct dirent_gather {
478 const struct gfs2_dirent **pdent;
482 static int gfs2_dirent_gather(const struct gfs2_dirent *dent,
483 const struct qstr *name,
486 struct dirent_gather *g = opaque;
487 if (!gfs2_dirent_sentinel(dent)) {
488 g->pdent[g->offset++] = dent;
494 * Other possible things to check:
495 * - Inode located within filesystem size (and on valid block)
496 * - Valid directory entry type
497 * Not sure how heavy-weight we want to make this... could also check
498 * hash is correct for example, but that would take a lot of extra time.
499 * For now the most important thing is to check that the various sizes
502 static int gfs2_check_dirent(struct gfs2_sbd *sdp,
503 struct gfs2_dirent *dent, unsigned int offset,
504 unsigned int size, unsigned int len, int first)
506 const char *msg = "gfs2_dirent too small";
507 if (unlikely(size < sizeof(struct gfs2_dirent)))
509 msg = "gfs2_dirent misaligned";
510 if (unlikely(offset & 0x7))
512 msg = "gfs2_dirent points beyond end of block";
513 if (unlikely(offset + size > len))
515 msg = "zero inode number";
516 if (unlikely(!first && gfs2_dirent_sentinel(dent)))
518 msg = "name length is greater than space in dirent";
519 if (!gfs2_dirent_sentinel(dent) &&
520 unlikely(sizeof(struct gfs2_dirent)+be16_to_cpu(dent->de_name_len) >
525 fs_warn(sdp, "%s: %s (%s)\n",
526 __func__, msg, first ? "first in block" : "not first in block");
530 static int gfs2_dirent_offset(struct gfs2_sbd *sdp, const void *buf)
532 const struct gfs2_meta_header *h = buf;
537 switch(be32_to_cpu(h->mh_type)) {
538 case GFS2_METATYPE_LF:
539 offset = sizeof(struct gfs2_leaf);
541 case GFS2_METATYPE_DI:
542 offset = sizeof(struct gfs2_dinode);
549 fs_warn(sdp, "%s: wrong block type %u\n", __func__,
550 be32_to_cpu(h->mh_type));
554 static struct gfs2_dirent *gfs2_dirent_scan(struct inode *inode, void *buf,
555 unsigned int len, gfs2_dscan_t scan,
556 const struct qstr *name,
559 struct gfs2_dirent *dent, *prev;
564 ret = gfs2_dirent_offset(GFS2_SB(inode), buf);
566 gfs2_consist_inode(GFS2_I(inode));
567 return ERR_PTR(-EIO);
572 size = be16_to_cpu(dent->de_rec_len);
573 if (gfs2_check_dirent(GFS2_SB(inode), dent, offset, size, len, 1)) {
574 gfs2_consist_inode(GFS2_I(inode));
575 return ERR_PTR(-EIO);
578 ret = scan(dent, name, opaque);
586 size = be16_to_cpu(dent->de_rec_len);
587 if (gfs2_check_dirent(GFS2_SB(inode), dent, offset, size,
589 gfs2_consist_inode(GFS2_I(inode));
590 return ERR_PTR(-EIO);
600 return prev ? prev : dent;
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))) {
614 gfs2_consist_inode(dip);
623 gfs2_consist_inode(dip);
628 * dirent_next - Next dirent
629 * @dip: the directory
631 * @dent: Pointer to list of dirents
633 * Returns: 0 on success, error code otherwise
636 static int dirent_next(struct gfs2_inode *dip, struct buffer_head *bh,
637 struct gfs2_dirent **dent)
639 struct gfs2_dirent *cur = *dent, *tmp;
640 char *bh_end = bh->b_data + bh->b_size;
643 ret = dirent_check_reclen(dip, cur, bh_end);
647 tmp = (void *)cur + ret;
648 ret = dirent_check_reclen(dip, tmp, bh_end);
652 /* Only the first dent could ever have de_inum.no_addr == 0 */
653 if (gfs2_dirent_sentinel(tmp)) {
654 gfs2_consist_inode(dip);
663 * dirent_del - Delete a dirent
664 * @dip: The GFS2 inode
666 * @prev: The previous dirent
667 * @cur: The current dirent
671 static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh,
672 struct gfs2_dirent *prev, struct gfs2_dirent *cur)
674 u16 cur_rec_len, prev_rec_len;
676 if (gfs2_dirent_sentinel(cur)) {
677 gfs2_consist_inode(dip);
681 gfs2_trans_add_meta(dip->i_gl, bh);
683 /* If there is no prev entry, this is the first entry in the block.
684 The de_rec_len is already as big as it needs to be. Just zero
685 out the inode number and return. */
688 cur->de_inum.no_addr = 0;
689 cur->de_inum.no_formal_ino = 0;
693 /* Combine this dentry with the previous one. */
695 prev_rec_len = be16_to_cpu(prev->de_rec_len);
696 cur_rec_len = be16_to_cpu(cur->de_rec_len);
698 if ((char *)prev + prev_rec_len != (char *)cur)
699 gfs2_consist_inode(dip);
700 if ((char *)cur + cur_rec_len > bh->b_data + bh->b_size)
701 gfs2_consist_inode(dip);
703 prev_rec_len += cur_rec_len;
704 prev->de_rec_len = cpu_to_be16(prev_rec_len);
708 static struct gfs2_dirent *do_init_dirent(struct inode *inode,
709 struct gfs2_dirent *dent,
710 const struct qstr *name,
711 struct buffer_head *bh,
714 struct gfs2_inode *ip = GFS2_I(inode);
715 struct gfs2_dirent *ndent;
718 totlen = be16_to_cpu(dent->de_rec_len);
719 BUG_ON(offset + name->len > totlen);
720 gfs2_trans_add_meta(ip->i_gl, bh);
721 ndent = (struct gfs2_dirent *)((char *)dent + offset);
722 dent->de_rec_len = cpu_to_be16(offset);
723 gfs2_qstr2dirent(name, totlen - offset, ndent);
729 * Takes a dent from which to grab space as an argument. Returns the
730 * newly created dent.
732 static struct gfs2_dirent *gfs2_init_dirent(struct inode *inode,
733 struct gfs2_dirent *dent,
734 const struct qstr *name,
735 struct buffer_head *bh)
739 if (!gfs2_dirent_sentinel(dent))
740 offset = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
741 return do_init_dirent(inode, dent, name, bh, offset);
744 static struct gfs2_dirent *gfs2_dirent_split_alloc(struct inode *inode,
745 struct buffer_head *bh,
746 const struct qstr *name,
749 struct gfs2_dirent *dent;
750 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
751 gfs2_dirent_find_offset, name, ptr);
752 if (IS_ERR_OR_NULL(dent))
754 return do_init_dirent(inode, dent, name, bh,
755 (unsigned)(ptr - (void *)dent));
758 static int get_leaf(struct gfs2_inode *dip, u64 leaf_no,
759 struct buffer_head **bhp)
763 error = gfs2_meta_read(dip->i_gl, leaf_no, DIO_WAIT, 0, bhp);
764 if (!error && gfs2_metatype_check(GFS2_SB(&dip->i_inode), *bhp, GFS2_METATYPE_LF)) {
765 /* pr_info("block num=%llu\n", leaf_no); */
773 * get_leaf_nr - Get a leaf number associated with the index
774 * @dip: The GFS2 inode
775 * @index: hash table index of the targeted leaf
776 * @leaf_out: Resulting leaf block number
778 * Returns: 0 on success, error code otherwise
781 static int get_leaf_nr(struct gfs2_inode *dip, u32 index, u64 *leaf_out)
786 hash = gfs2_dir_get_hash_table(dip);
787 error = PTR_ERR_OR_ZERO(hash);
790 *leaf_out = be64_to_cpu(*(hash + index));
795 static int get_first_leaf(struct gfs2_inode *dip, u32 index,
796 struct buffer_head **bh_out)
801 error = get_leaf_nr(dip, index, &leaf_no);
803 error = get_leaf(dip, leaf_no, bh_out);
808 static struct gfs2_dirent *gfs2_dirent_search(struct inode *inode,
809 const struct qstr *name,
811 struct buffer_head **pbh)
813 struct buffer_head *bh;
814 struct gfs2_dirent *dent;
815 struct gfs2_inode *ip = GFS2_I(inode);
818 if (ip->i_diskflags & GFS2_DIF_EXHASH) {
819 struct gfs2_leaf *leaf;
820 unsigned int hsize = BIT(ip->i_depth);
823 if (hsize * sizeof(u64) != i_size_read(inode)) {
824 gfs2_consist_inode(ip);
825 return ERR_PTR(-EIO);
828 index = name->hash >> (32 - ip->i_depth);
829 error = get_first_leaf(ip, index, &bh);
831 return ERR_PTR(error);
833 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
837 leaf = (struct gfs2_leaf *)bh->b_data;
838 ln = be64_to_cpu(leaf->lf_next);
843 error = get_leaf(ip, ln, &bh);
846 return error ? ERR_PTR(error) : NULL;
850 error = gfs2_meta_inode_buffer(ip, &bh);
852 return ERR_PTR(error);
853 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, scan, name, NULL);
855 if (IS_ERR_OR_NULL(dent)) {
863 static struct gfs2_leaf *new_leaf(struct inode *inode, struct buffer_head **pbh, u16 depth)
865 struct gfs2_inode *ip = GFS2_I(inode);
869 struct buffer_head *bh;
870 struct gfs2_leaf *leaf;
871 struct gfs2_dirent *dent;
872 struct timespec64 tv = current_time(inode);
874 error = gfs2_alloc_blocks(ip, &bn, &n, 0);
877 bh = gfs2_meta_new(ip->i_gl, bn);
881 gfs2_trans_remove_revoke(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(&empty_name, bh->b_size - sizeof(struct gfs2_leaf), dent);
901 * dir_make_exhash - Convert a stuffed directory into an ExHash directory
902 * @inode: The directory inode to be converted to exhash
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 < BIT(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 * @inode: The directory inode to be split
996 * @name: name of the dirent we're trying to insert
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);
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 * @sort_start: index of the directory array to start our sort
1256 * @copied: pointer to int that's non-zero if a entry has been copied out
1258 * Jump through some hoops to make sure that if there are hash collsions,
1259 * they are read out at the beginning of a buffer. We want to minimize
1260 * the possibility that they will fall into different readdir buffers or
1261 * that someone will want to seek to that location.
1263 * Returns: errno, >0 if the actor tells you to stop
1266 static int do_filldir_main(struct gfs2_inode *dip, struct dir_context *ctx,
1267 struct gfs2_dirent **darr, u32 entries,
1268 u32 sort_start, int *copied)
1270 const struct gfs2_dirent *dent, *dent_next;
1275 if (sort_start < entries)
1276 sort(&darr[sort_start], entries - sort_start,
1277 sizeof(struct gfs2_dirent *), compare_dents, NULL);
1279 dent_next = darr[0];
1280 off_next = dent_next->de_cookie;
1282 for (x = 0, y = 1; x < entries; x++, y++) {
1287 dent_next = darr[y];
1288 off_next = dent_next->de_cookie;
1294 if (off_next == off) {
1295 if (*copied && !run)
1306 if (!dir_emit(ctx, (const char *)(dent + 1),
1307 be16_to_cpu(dent->de_name_len),
1308 be64_to_cpu(dent->de_inum.no_addr),
1309 be16_to_cpu(dent->de_type)))
1315 /* Increment the ctx->pos by one, so the next time we come into the
1316 do_filldir fxn, we get the next entry instead of the last one in the
1324 static void *gfs2_alloc_sort_buffer(unsigned size)
1328 if (size < KMALLOC_MAX_SIZE)
1329 ptr = kmalloc(size, GFP_NOFS | __GFP_NOWARN);
1331 ptr = __vmalloc(size, GFP_NOFS);
1336 static int gfs2_set_cookies(struct gfs2_sbd *sdp, struct buffer_head *bh,
1337 unsigned leaf_nr, struct gfs2_dirent **darr,
1343 for (i = 0; i < entries; i++) {
1346 darr[i]->de_cookie = be32_to_cpu(darr[i]->de_hash);
1347 darr[i]->de_cookie = gfs2_disk_hash2offset(darr[i]->de_cookie);
1349 if (!sdp->sd_args.ar_loccookie)
1351 offset = (char *)(darr[i]) -
1352 (bh->b_data + gfs2_dirent_offset(sdp, bh->b_data));
1353 offset /= GFS2_MIN_DIRENT_SIZE;
1354 offset += leaf_nr * sdp->sd_max_dents_per_leaf;
1355 if (offset >= GFS2_USE_HASH_FLAG ||
1356 leaf_nr >= GFS2_USE_HASH_FLAG) {
1357 darr[i]->de_cookie |= GFS2_USE_HASH_FLAG;
1362 darr[i]->de_cookie &= GFS2_HASH_INDEX_MASK;
1363 darr[i]->de_cookie |= offset;
1369 static int gfs2_dir_read_leaf(struct inode *inode, struct dir_context *ctx,
1370 int *copied, unsigned *depth,
1373 struct gfs2_inode *ip = GFS2_I(inode);
1374 struct gfs2_sbd *sdp = GFS2_SB(inode);
1375 struct buffer_head *bh;
1376 struct gfs2_leaf *lf;
1377 unsigned entries = 0, entries2 = 0;
1378 unsigned leaves = 0, leaf = 0, offset, sort_offset;
1379 struct gfs2_dirent **darr, *dent;
1380 struct dirent_gather g;
1381 struct buffer_head **larr;
1382 int error, i, need_sort = 0, sort_id;
1386 error = get_leaf(ip, lfn, &bh);
1389 lf = (struct gfs2_leaf *)bh->b_data;
1391 *depth = be16_to_cpu(lf->lf_depth);
1392 entries += be16_to_cpu(lf->lf_entries);
1394 lfn = be64_to_cpu(lf->lf_next);
1398 if (*depth < GFS2_DIR_MAX_DEPTH || !sdp->sd_args.ar_loccookie) {
1408 * The extra 99 entries are not normally used, but are a buffer
1409 * zone in case the number of entries in the leaf is corrupt.
1410 * 99 is the maximum number of entries that can fit in a single
1413 larr = gfs2_alloc_sort_buffer((leaves + entries + 99) * sizeof(void *));
1416 darr = (struct gfs2_dirent **)(larr + leaves);
1417 g.pdent = (const struct gfs2_dirent **)darr;
1422 error = get_leaf(ip, lfn, &bh);
1425 lf = (struct gfs2_leaf *)bh->b_data;
1426 lfn = be64_to_cpu(lf->lf_next);
1427 if (lf->lf_entries) {
1429 entries2 += be16_to_cpu(lf->lf_entries);
1430 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
1431 gfs2_dirent_gather, NULL, &g);
1432 error = PTR_ERR(dent);
1435 if (entries2 != g.offset) {
1436 fs_warn(sdp, "Number of entries corrupt in dir "
1437 "leaf %llu, entries2 (%u) != "
1439 (unsigned long long)bh->b_blocknr,
1440 entries2, g.offset);
1441 gfs2_consist_inode(ip);
1446 sort_id = gfs2_set_cookies(sdp, bh, leaf, &darr[offset],
1447 be16_to_cpu(lf->lf_entries));
1448 if (!need_sort && sort_id >= 0) {
1450 sort_offset = offset + sort_id;
1454 larr[leaf++] = NULL;
1459 BUG_ON(entries2 != entries);
1460 error = do_filldir_main(ip, ctx, darr, entries, need_sort ?
1461 sort_offset : entries, copied);
1463 for(i = 0; i < leaf; i++)
1471 * gfs2_dir_readahead - Issue read-ahead requests for leaf blocks.
1472 * @inode: the directory inode
1473 * @hsize: hash table size
1474 * @index: index into the hash table
1475 * @f_ra: read-ahead parameters
1477 * Note: we can't calculate each index like dir_e_read can because we don't
1478 * have the leaf, and therefore we don't have the depth, and therefore we
1479 * don't have the length. So we have to just read enough ahead to make up
1480 * for the loss of information.
1482 static void gfs2_dir_readahead(struct inode *inode, unsigned hsize, u32 index,
1483 struct file_ra_state *f_ra)
1485 struct gfs2_inode *ip = GFS2_I(inode);
1486 struct gfs2_glock *gl = ip->i_gl;
1487 struct buffer_head *bh;
1488 u64 blocknr = 0, last;
1491 /* First check if we've already read-ahead for the whole range. */
1492 if (index + MAX_RA_BLOCKS < f_ra->start)
1495 f_ra->start = max((pgoff_t)index, f_ra->start);
1496 for (count = 0; count < MAX_RA_BLOCKS; count++) {
1497 if (f_ra->start >= hsize) /* if exceeded the hash table */
1501 blocknr = be64_to_cpu(ip->i_hash_cache[f_ra->start]);
1503 if (blocknr == last)
1506 bh = gfs2_getbuf(gl, blocknr, 1);
1507 if (trylock_buffer(bh)) {
1508 if (buffer_uptodate(bh)) {
1513 bh->b_end_io = end_buffer_read_sync;
1514 submit_bh(REQ_OP_READ | REQ_RAHEAD | REQ_META |
1523 * dir_e_read - Reads the entries from a directory into a filldir buffer
1524 * @inode: the directory inode
1525 * @ctx: actor to feed the entries to
1526 * @f_ra: read-ahead parameters
1531 static int dir_e_read(struct inode *inode, struct dir_context *ctx,
1532 struct file_ra_state *f_ra)
1534 struct gfs2_inode *dip = GFS2_I(inode);
1542 hsize = BIT(dip->i_depth);
1543 hash = gfs2_dir_offset2hash(ctx->pos);
1544 index = hash >> (32 - dip->i_depth);
1546 if (dip->i_hash_cache == NULL)
1548 lp = gfs2_dir_get_hash_table(dip);
1552 gfs2_dir_readahead(inode, hsize, index, f_ra);
1554 while (index < hsize) {
1555 error = gfs2_dir_read_leaf(inode, ctx,
1557 be64_to_cpu(lp[index]));
1561 len = BIT(dip->i_depth - depth);
1562 index = (index & ~(len - 1)) + len;
1570 int gfs2_dir_read(struct inode *inode, struct dir_context *ctx,
1571 struct file_ra_state *f_ra)
1573 struct gfs2_inode *dip = GFS2_I(inode);
1574 struct gfs2_sbd *sdp = GFS2_SB(inode);
1575 struct dirent_gather g;
1576 struct gfs2_dirent **darr, *dent;
1577 struct buffer_head *dibh;
1581 if (!dip->i_entries)
1584 if (dip->i_diskflags & GFS2_DIF_EXHASH)
1585 return dir_e_read(inode, ctx, f_ra);
1587 if (!gfs2_is_stuffed(dip)) {
1588 gfs2_consist_inode(dip);
1592 error = gfs2_meta_inode_buffer(dip, &dibh);
1597 /* 96 is max number of dirents which can be stuffed into an inode */
1598 darr = kmalloc_array(96, sizeof(struct gfs2_dirent *), GFP_NOFS);
1600 g.pdent = (const struct gfs2_dirent **)darr;
1602 dent = gfs2_dirent_scan(inode, dibh->b_data, dibh->b_size,
1603 gfs2_dirent_gather, NULL, &g);
1605 error = PTR_ERR(dent);
1608 if (dip->i_entries != g.offset) {
1609 fs_warn(sdp, "Number of entries corrupt in dir %llu, "
1610 "ip->i_entries (%u) != g.offset (%u)\n",
1611 (unsigned long long)dip->i_no_addr,
1614 gfs2_consist_inode(dip);
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 * @dir: The GFS2 directory 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;
1804 struct timespec64 tv;
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 = inode_set_ctime_current(&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 inode_set_mtime_to_ts(&ip->i_inode, 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 * @dentry: The directory entry we want to delete
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 timespec64 tv;
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 tv = inode_set_ctime_current(&dip->i_inode);
1903 if (dip->i_diskflags & GFS2_DIF_EXHASH) {
1904 struct gfs2_leaf *leaf = (struct gfs2_leaf *)bh->b_data;
1905 u16 entries = be16_to_cpu(leaf->lf_entries);
1907 gfs2_consist_inode(dip);
1908 leaf->lf_entries = cpu_to_be16(--entries);
1909 leaf->lf_nsec = cpu_to_be32(tv.tv_nsec);
1910 leaf->lf_sec = cpu_to_be64(tv.tv_sec);
1914 if (!dip->i_entries)
1915 gfs2_consist_inode(dip);
1917 inode_set_mtime_to_ts(&dip->i_inode, tv);
1918 if (d_is_dir(dentry))
1919 drop_nlink(&dip->i_inode);
1920 mark_inode_dirty(&dip->i_inode);
1926 * gfs2_dir_mvino - Change inode number of directory entry
1927 * @dip: The GFS2 directory inode
1928 * @filename: the filename to be moved
1929 * @nip: the new GFS2 inode
1930 * @new_type: the de_type of the new dirent
1932 * This routine changes the inode number of a directory entry. It's used
1933 * by rename to change ".." when a directory is moved.
1934 * Assumes a glock is held on dvp.
1939 int gfs2_dir_mvino(struct gfs2_inode *dip, const struct qstr *filename,
1940 const struct gfs2_inode *nip, unsigned int new_type)
1942 struct buffer_head *bh;
1943 struct gfs2_dirent *dent;
1945 dent = gfs2_dirent_search(&dip->i_inode, filename, gfs2_dirent_find, &bh);
1947 gfs2_consist_inode(dip);
1951 return PTR_ERR(dent);
1953 gfs2_trans_add_meta(dip->i_gl, bh);
1954 gfs2_inum_out(nip, dent);
1955 dent->de_type = cpu_to_be16(new_type);
1958 inode_set_mtime_to_ts(&dip->i_inode, inode_set_ctime_current(&dip->i_inode));
1959 mark_inode_dirty_sync(&dip->i_inode);
1964 * leaf_dealloc - Deallocate a directory leaf
1965 * @dip: the directory
1966 * @index: the hash table offset in the directory
1967 * @len: the number of pointers to this leaf
1968 * @leaf_no: the leaf number
1969 * @leaf_bh: buffer_head for the starting leaf
1970 * @last_dealloc: 1 if this is the final dealloc for the leaf, else 0
1975 static int leaf_dealloc(struct gfs2_inode *dip, u32 index, u32 len,
1976 u64 leaf_no, struct buffer_head *leaf_bh,
1979 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1980 struct gfs2_leaf *tmp_leaf;
1981 struct gfs2_rgrp_list rlist;
1982 struct buffer_head *bh, *dibh;
1984 unsigned int rg_blocks = 0, l_blocks = 0;
1986 unsigned int x, size = len * sizeof(u64);
1989 error = gfs2_rindex_update(sdp);
1993 memset(&rlist, 0, sizeof(struct gfs2_rgrp_list));
1995 ht = kzalloc(size, GFP_NOFS | __GFP_NOWARN);
1997 ht = __vmalloc(size, GFP_NOFS | __GFP_NOWARN | __GFP_ZERO);
2001 error = gfs2_quota_hold(dip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
2005 /* Count the number of leaves */
2008 for (blk = leaf_no; blk; blk = nblk) {
2009 if (blk != leaf_no) {
2010 error = get_leaf(dip, blk, &bh);
2014 tmp_leaf = (struct gfs2_leaf *)bh->b_data;
2015 nblk = be64_to_cpu(tmp_leaf->lf_next);
2019 gfs2_rlist_add(dip, &rlist, blk);
2023 gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE, LM_FLAG_NODE_SCOPE);
2025 for (x = 0; x < rlist.rl_rgrps; x++) {
2026 struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(rlist.rl_ghs[x].gh_gl);
2028 rg_blocks += rgd->rd_length;
2031 error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs);
2035 error = gfs2_trans_begin(sdp,
2036 rg_blocks + (DIV_ROUND_UP(size, sdp->sd_jbsize) + 1) +
2037 RES_DINODE + RES_STATFS + RES_QUOTA, RES_DINODE +
2040 goto out_rg_gunlock;
2044 for (blk = leaf_no; blk; blk = nblk) {
2045 struct gfs2_rgrpd *rgd;
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 rgd = gfs2_blk2rgrpd(sdp, blk, true);
2058 gfs2_free_meta(dip, rgd, blk, 1);
2059 gfs2_add_inode_blocks(&dip->i_inode, -1);
2062 error = gfs2_dir_write_data(dip, ht, index * sizeof(u64), size);
2063 if (error != size) {
2069 error = gfs2_meta_inode_buffer(dip, &dibh);
2073 gfs2_trans_add_meta(dip->i_gl, dibh);
2074 /* On the last dealloc, make this a regular file in case we crash.
2075 (We don't want to free these blocks a second time.) */
2077 dip->i_inode.i_mode = S_IFREG;
2078 gfs2_dinode_out(dip, dibh->b_data);
2082 gfs2_trans_end(sdp);
2084 gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
2086 gfs2_rlist_free(&rlist);
2087 gfs2_quota_unhold(dip);
2094 * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory
2095 * @dip: the directory
2097 * Dealloc all on-disk directory leaves to FREEMETA state
2098 * Change on-disk inode type to "regular file"
2103 int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip)
2105 struct buffer_head *bh;
2106 struct gfs2_leaf *leaf;
2108 u32 index = 0, next_index;
2111 int error = 0, last;
2113 hsize = BIT(dip->i_depth);
2115 lp = gfs2_dir_get_hash_table(dip);
2119 while (index < hsize) {
2120 leaf_no = be64_to_cpu(lp[index]);
2122 error = get_leaf(dip, leaf_no, &bh);
2125 leaf = (struct gfs2_leaf *)bh->b_data;
2126 len = BIT(dip->i_depth - be16_to_cpu(leaf->lf_depth));
2128 next_index = (index & ~(len - 1)) + len;
2129 last = ((next_index >= hsize) ? 1 : 0);
2130 error = leaf_dealloc(dip, index, len, leaf_no, bh,
2140 if (index != hsize) {
2141 gfs2_consist_inode(dip);
2151 * gfs2_diradd_alloc_required - find if adding entry will require an allocation
2152 * @inode: the directory inode being written to
2153 * @name: the filename that's going to be added
2154 * @da: The structure to return dir alloc info
2156 * Returns: 0 if ok, -ve on error
2159 int gfs2_diradd_alloc_required(struct inode *inode, const struct qstr *name,
2160 struct gfs2_diradd *da)
2162 struct gfs2_inode *ip = GFS2_I(inode);
2163 struct gfs2_sbd *sdp = GFS2_SB(inode);
2164 const unsigned int extra = sizeof(struct gfs2_dinode) - sizeof(struct gfs2_leaf);
2165 struct gfs2_dirent *dent;
2166 struct buffer_head *bh;
2172 dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, &bh);
2174 da->nr_blocks = sdp->sd_max_dirres;
2175 if (!(ip->i_diskflags & GFS2_DIF_EXHASH) &&
2176 (GFS2_DIRENT_SIZE(name->len) < extra))
2181 return PTR_ERR(dent);