1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * Creates, reads, walks and deletes directory-nodes
8 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
10 * Portions of this code from linux/fs/ext3/dir.c
12 * Copyright (C) 1992, 1993, 1994, 1995
13 * Remy Card (card@masi.ibp.fr)
14 * Laboratoire MASI - Institut Blaise pascal
15 * Universite Pierre et Marie Curie (Paris VI)
19 * linux/fs/minix/dir.c
21 * Copyright (C) 1991, 1992 Linux Torvalds
23 * This program is free software; you can redistribute it and/or
24 * modify it under the terms of the GNU General Public
25 * License as published by the Free Software Foundation; either
26 * version 2 of the License, or (at your option) any later version.
28 * This program is distributed in the hope that it will be useful,
29 * but WITHOUT ANY WARRANTY; without even the implied warranty of
30 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
31 * General Public License for more details.
33 * You should have received a copy of the GNU General Public
34 * License along with this program; if not, write to the
35 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
36 * Boston, MA 021110-1307, USA.
40 #include <linux/types.h>
41 #include <linux/slab.h>
42 #include <linux/highmem.h>
43 #include <linux/quotaops.h>
44 #include <linux/sort.h>
46 #include <cluster/masklog.h>
51 #include "blockcheck.h"
54 #include "extent_map.h"
63 #include "ocfs2_trace.h"
65 #include "buffer_head_io.h"
67 #define NAMEI_RA_CHUNKS 2
68 #define NAMEI_RA_BLOCKS 4
69 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
71 static unsigned char ocfs2_filetype_table[] = {
72 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
75 static int ocfs2_do_extend_dir(struct super_block *sb,
78 struct buffer_head *parent_fe_bh,
79 struct ocfs2_alloc_context *data_ac,
80 struct ocfs2_alloc_context *meta_ac,
81 struct buffer_head **new_bh);
82 static int ocfs2_dir_indexed(struct inode *inode);
85 * These are distinct checks because future versions of the file system will
86 * want to have a trailing dirent structure independent of indexing.
88 static int ocfs2_supports_dir_trailer(struct inode *dir)
90 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
92 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
95 return ocfs2_meta_ecc(osb) || ocfs2_dir_indexed(dir);
99 * "new' here refers to the point at which we're creating a new
100 * directory via "mkdir()", but also when we're expanding an inline
101 * directory. In either case, we don't yet have the indexing bit set
102 * on the directory, so the standard checks will fail in when metaecc
103 * is turned off. Only directory-initialization type functions should
104 * use this then. Everything else wants ocfs2_supports_dir_trailer()
106 static int ocfs2_new_dir_wants_trailer(struct inode *dir)
108 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
110 return ocfs2_meta_ecc(osb) ||
111 ocfs2_supports_indexed_dirs(osb);
114 static inline unsigned int ocfs2_dir_trailer_blk_off(struct super_block *sb)
116 return sb->s_blocksize - sizeof(struct ocfs2_dir_block_trailer);
119 #define ocfs2_trailer_from_bh(_bh, _sb) ((struct ocfs2_dir_block_trailer *) ((_bh)->b_data + ocfs2_dir_trailer_blk_off((_sb))))
121 /* XXX ocfs2_block_dqtrailer() is similar but not quite - can we make
122 * them more consistent? */
123 struct ocfs2_dir_block_trailer *ocfs2_dir_trailer_from_size(int blocksize,
128 p += blocksize - sizeof(struct ocfs2_dir_block_trailer);
129 return (struct ocfs2_dir_block_trailer *)p;
133 * XXX: This is executed once on every dirent. We should consider optimizing
136 static int ocfs2_skip_dir_trailer(struct inode *dir,
137 struct ocfs2_dir_entry *de,
138 unsigned long offset,
139 unsigned long blklen)
141 unsigned long toff = blklen - sizeof(struct ocfs2_dir_block_trailer);
143 if (!ocfs2_supports_dir_trailer(dir))
152 static void ocfs2_init_dir_trailer(struct inode *inode,
153 struct buffer_head *bh, u16 rec_len)
155 struct ocfs2_dir_block_trailer *trailer;
157 trailer = ocfs2_trailer_from_bh(bh, inode->i_sb);
158 strcpy(trailer->db_signature, OCFS2_DIR_TRAILER_SIGNATURE);
159 trailer->db_compat_rec_len =
160 cpu_to_le16(sizeof(struct ocfs2_dir_block_trailer));
161 trailer->db_parent_dinode = cpu_to_le64(OCFS2_I(inode)->ip_blkno);
162 trailer->db_blkno = cpu_to_le64(bh->b_blocknr);
163 trailer->db_free_rec_len = cpu_to_le16(rec_len);
166 * Link an unindexed block with a dir trailer structure into the index free
167 * list. This function will modify dirdata_bh, but assumes you've already
168 * passed it to the journal.
170 static int ocfs2_dx_dir_link_trailer(struct inode *dir, handle_t *handle,
171 struct buffer_head *dx_root_bh,
172 struct buffer_head *dirdata_bh)
175 struct ocfs2_dx_root_block *dx_root;
176 struct ocfs2_dir_block_trailer *trailer;
178 ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
179 OCFS2_JOURNAL_ACCESS_WRITE);
184 trailer = ocfs2_trailer_from_bh(dirdata_bh, dir->i_sb);
185 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
187 trailer->db_free_next = dx_root->dr_free_blk;
188 dx_root->dr_free_blk = cpu_to_le64(dirdata_bh->b_blocknr);
190 ocfs2_journal_dirty(handle, dx_root_bh);
196 static int ocfs2_free_list_at_root(struct ocfs2_dir_lookup_result *res)
198 return res->dl_prev_leaf_bh == NULL;
201 void ocfs2_free_dir_lookup_result(struct ocfs2_dir_lookup_result *res)
203 brelse(res->dl_dx_root_bh);
204 brelse(res->dl_leaf_bh);
205 brelse(res->dl_dx_leaf_bh);
206 brelse(res->dl_prev_leaf_bh);
209 static int ocfs2_dir_indexed(struct inode *inode)
211 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INDEXED_DIR_FL)
216 static inline int ocfs2_dx_root_inline(struct ocfs2_dx_root_block *dx_root)
218 return dx_root->dr_flags & OCFS2_DX_FLAG_INLINE;
222 * Hashing code adapted from ext3
224 #define DELTA 0x9E3779B9
226 static void TEA_transform(__u32 buf[4], __u32 const in[])
229 __u32 b0 = buf[0], b1 = buf[1];
230 __u32 a = in[0], b = in[1], c = in[2], d = in[3];
235 b0 += ((b1 << 4)+a) ^ (b1+sum) ^ ((b1 >> 5)+b);
236 b1 += ((b0 << 4)+c) ^ (b0+sum) ^ ((b0 >> 5)+d);
243 static void str2hashbuf(const char *msg, int len, __u32 *buf, int num)
248 pad = (__u32)len | ((__u32)len << 8);
254 for (i = 0; i < len; i++) {
257 val = msg[i] + (val << 8);
270 static void ocfs2_dx_dir_name_hash(struct inode *dir, const char *name, int len,
271 struct ocfs2_dx_hinfo *hinfo)
273 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
278 * XXX: Is this really necessary, if the index is never looked
279 * at by readdir? Is a hash value of '0' a bad idea?
281 if ((len == 1 && !strncmp(".", name, 1)) ||
282 (len == 2 && !strncmp("..", name, 2))) {
287 #ifdef OCFS2_DEBUG_DX_DIRS
289 * This makes it very easy to debug indexing problems. We
290 * should never allow this to be selected without hand editing
293 buf[0] = buf[1] = len;
297 memcpy(buf, osb->osb_dx_seed, sizeof(buf));
301 str2hashbuf(p, len, in, 4);
302 TEA_transform(buf, in);
308 hinfo->major_hash = buf[0];
309 hinfo->minor_hash = buf[1];
313 * bh passed here can be an inode block or a dir data block, depending
314 * on the inode inline data flag.
316 static int ocfs2_check_dir_entry(struct inode * dir,
317 struct ocfs2_dir_entry * de,
318 struct buffer_head * bh,
319 unsigned long offset)
321 const char *error_msg = NULL;
322 const int rlen = le16_to_cpu(de->rec_len);
324 if (unlikely(rlen < OCFS2_DIR_REC_LEN(1)))
325 error_msg = "rec_len is smaller than minimal";
326 else if (unlikely(rlen % 4 != 0))
327 error_msg = "rec_len % 4 != 0";
328 else if (unlikely(rlen < OCFS2_DIR_REC_LEN(de->name_len)))
329 error_msg = "rec_len is too small for name_len";
331 ((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize))
332 error_msg = "directory entry across blocks";
334 if (unlikely(error_msg != NULL))
335 mlog(ML_ERROR, "bad entry in directory #%llu: %s - "
336 "offset=%lu, inode=%llu, rec_len=%d, name_len=%d\n",
337 (unsigned long long)OCFS2_I(dir)->ip_blkno, error_msg,
338 offset, (unsigned long long)le64_to_cpu(de->inode), rlen,
341 return error_msg == NULL ? 1 : 0;
344 static inline int ocfs2_match(int len,
345 const char * const name,
346 struct ocfs2_dir_entry *de)
348 if (len != de->name_len)
352 return !memcmp(name, de->name, len);
356 * Returns 0 if not found, -1 on failure, and 1 on success
358 static inline int ocfs2_search_dirblock(struct buffer_head *bh,
360 const char *name, int namelen,
361 unsigned long offset,
364 struct ocfs2_dir_entry **res_dir)
366 struct ocfs2_dir_entry *de;
367 char *dlimit, *de_buf;
372 dlimit = de_buf + bytes;
374 while (de_buf < dlimit) {
375 /* this code is executed quadratically often */
376 /* do minimal checking `by hand' */
378 de = (struct ocfs2_dir_entry *) de_buf;
380 if (de_buf + namelen <= dlimit &&
381 ocfs2_match(namelen, name, de)) {
382 /* found a match - just to be sure, do a full check */
383 if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
392 /* prevent looping on a bad block */
393 de_len = le16_to_cpu(de->rec_len);
404 trace_ocfs2_search_dirblock(ret);
408 static struct buffer_head *ocfs2_find_entry_id(const char *name,
411 struct ocfs2_dir_entry **res_dir)
414 struct buffer_head *di_bh = NULL;
415 struct ocfs2_dinode *di;
416 struct ocfs2_inline_data *data;
418 ret = ocfs2_read_inode_block(dir, &di_bh);
424 di = (struct ocfs2_dinode *)di_bh->b_data;
425 data = &di->id2.i_data;
427 found = ocfs2_search_dirblock(di_bh, dir, name, namelen, 0,
428 data->id_data, i_size_read(dir), res_dir);
437 static int ocfs2_validate_dir_block(struct super_block *sb,
438 struct buffer_head *bh)
441 struct ocfs2_dir_block_trailer *trailer =
442 ocfs2_trailer_from_bh(bh, sb);
446 * We don't validate dirents here, that's handled
447 * in-place when the code walks them.
449 trace_ocfs2_validate_dir_block((unsigned long long)bh->b_blocknr);
451 BUG_ON(!buffer_uptodate(bh));
454 * If the ecc fails, we return the error but otherwise
455 * leave the filesystem running. We know any error is
456 * local to this block.
458 * Note that we are safe to call this even if the directory
459 * doesn't have a trailer. Filesystems without metaecc will do
460 * nothing, and filesystems with it will have one.
462 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &trailer->db_check);
464 mlog(ML_ERROR, "Checksum failed for dinode %llu\n",
465 (unsigned long long)bh->b_blocknr);
471 * Validate a directory trailer.
473 * We check the trailer here rather than in ocfs2_validate_dir_block()
474 * because that function doesn't have the inode to test.
476 static int ocfs2_check_dir_trailer(struct inode *dir, struct buffer_head *bh)
479 struct ocfs2_dir_block_trailer *trailer;
481 trailer = ocfs2_trailer_from_bh(bh, dir->i_sb);
482 if (!OCFS2_IS_VALID_DIR_TRAILER(trailer)) {
484 ocfs2_error(dir->i_sb,
485 "Invalid dirblock #%llu: "
486 "signature = %.*s\n",
487 (unsigned long long)bh->b_blocknr, 7,
488 trailer->db_signature);
491 if (le64_to_cpu(trailer->db_blkno) != bh->b_blocknr) {
493 ocfs2_error(dir->i_sb,
494 "Directory block #%llu has an invalid "
496 (unsigned long long)bh->b_blocknr,
497 (unsigned long long)le64_to_cpu(trailer->db_blkno));
500 if (le64_to_cpu(trailer->db_parent_dinode) !=
501 OCFS2_I(dir)->ip_blkno) {
503 ocfs2_error(dir->i_sb,
504 "Directory block #%llu on dinode "
505 "#%llu has an invalid parent_dinode "
507 (unsigned long long)bh->b_blocknr,
508 (unsigned long long)OCFS2_I(dir)->ip_blkno,
509 (unsigned long long)le64_to_cpu(trailer->db_blkno));
517 * This function forces all errors to -EIO for consistency with its
518 * predecessor, ocfs2_bread(). We haven't audited what returning the
519 * real error codes would do to callers. We log the real codes with
520 * mlog_errno() before we squash them.
522 static int ocfs2_read_dir_block(struct inode *inode, u64 v_block,
523 struct buffer_head **bh, int flags)
526 struct buffer_head *tmp = *bh;
528 rc = ocfs2_read_virt_blocks(inode, v_block, 1, &tmp, flags,
529 ocfs2_validate_dir_block);
535 if (!(flags & OCFS2_BH_READAHEAD) &&
536 ocfs2_supports_dir_trailer(inode)) {
537 rc = ocfs2_check_dir_trailer(inode, tmp);
546 /* If ocfs2_read_virt_blocks() got us a new bh, pass it up. */
551 return rc ? -EIO : 0;
555 * Read the block at 'phys' which belongs to this directory
556 * inode. This function does no virtual->physical block translation -
557 * what's passed in is assumed to be a valid directory block.
559 static int ocfs2_read_dir_block_direct(struct inode *dir, u64 phys,
560 struct buffer_head **bh)
563 struct buffer_head *tmp = *bh;
565 ret = ocfs2_read_block(INODE_CACHE(dir), phys, &tmp,
566 ocfs2_validate_dir_block);
572 if (ocfs2_supports_dir_trailer(dir)) {
573 ret = ocfs2_check_dir_trailer(dir, tmp);
588 static int ocfs2_validate_dx_root(struct super_block *sb,
589 struct buffer_head *bh)
592 struct ocfs2_dx_root_block *dx_root;
594 BUG_ON(!buffer_uptodate(bh));
596 dx_root = (struct ocfs2_dx_root_block *) bh->b_data;
598 ret = ocfs2_validate_meta_ecc(sb, bh->b_data, &dx_root->dr_check);
601 "Checksum failed for dir index root block %llu\n",
602 (unsigned long long)bh->b_blocknr);
606 if (!OCFS2_IS_VALID_DX_ROOT(dx_root)) {
608 "Dir Index Root # %llu has bad signature %.*s",
609 (unsigned long long)le64_to_cpu(dx_root->dr_blkno),
610 7, dx_root->dr_signature);
617 static int ocfs2_read_dx_root(struct inode *dir, struct ocfs2_dinode *di,
618 struct buffer_head **dx_root_bh)
621 u64 blkno = le64_to_cpu(di->i_dx_root);
622 struct buffer_head *tmp = *dx_root_bh;
624 ret = ocfs2_read_block(INODE_CACHE(dir), blkno, &tmp,
625 ocfs2_validate_dx_root);
627 /* If ocfs2_read_block() got us a new bh, pass it up. */
628 if (!ret && !*dx_root_bh)
634 static int ocfs2_validate_dx_leaf(struct super_block *sb,
635 struct buffer_head *bh)
638 struct ocfs2_dx_leaf *dx_leaf = (struct ocfs2_dx_leaf *)bh->b_data;
640 BUG_ON(!buffer_uptodate(bh));
642 ret = ocfs2_validate_meta_ecc(sb, bh->b_data, &dx_leaf->dl_check);
645 "Checksum failed for dir index leaf block %llu\n",
646 (unsigned long long)bh->b_blocknr);
650 if (!OCFS2_IS_VALID_DX_LEAF(dx_leaf)) {
651 ocfs2_error(sb, "Dir Index Leaf has bad signature %.*s",
652 7, dx_leaf->dl_signature);
659 static int ocfs2_read_dx_leaf(struct inode *dir, u64 blkno,
660 struct buffer_head **dx_leaf_bh)
663 struct buffer_head *tmp = *dx_leaf_bh;
665 ret = ocfs2_read_block(INODE_CACHE(dir), blkno, &tmp,
666 ocfs2_validate_dx_leaf);
668 /* If ocfs2_read_block() got us a new bh, pass it up. */
669 if (!ret && !*dx_leaf_bh)
676 * Read a series of dx_leaf blocks. This expects all buffer_head
677 * pointers to be NULL on function entry.
679 static int ocfs2_read_dx_leaves(struct inode *dir, u64 start, int num,
680 struct buffer_head **dx_leaf_bhs)
684 ret = ocfs2_read_blocks(INODE_CACHE(dir), start, num, dx_leaf_bhs, 0,
685 ocfs2_validate_dx_leaf);
692 static struct buffer_head *ocfs2_find_entry_el(const char *name, int namelen,
694 struct ocfs2_dir_entry **res_dir)
696 struct super_block *sb;
697 struct buffer_head *bh_use[NAMEI_RA_SIZE];
698 struct buffer_head *bh, *ret = NULL;
699 unsigned long start, block, b;
700 int ra_max = 0; /* Number of bh's in the readahead
702 int ra_ptr = 0; /* Current index into readahead
709 nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
710 start = OCFS2_I(dir)->ip_dir_start_lookup;
711 if (start >= nblocks)
718 * We deal with the read-ahead logic here.
720 if (ra_ptr >= ra_max) {
721 /* Refill the readahead buffer */
724 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
726 * Terminate if we reach the end of the
727 * directory and must wrap, or if our
728 * search has finished at this block.
730 if (b >= nblocks || (num && block == start)) {
731 bh_use[ra_max] = NULL;
737 err = ocfs2_read_dir_block(dir, b++, &bh,
742 if ((bh = bh_use[ra_ptr++]) == NULL)
744 if (ocfs2_read_dir_block(dir, block, &bh, 0)) {
745 /* read error, skip block & hope for the best.
746 * ocfs2_read_dir_block() has released the bh. */
747 mlog(ML_ERROR, "reading directory %llu, "
749 (unsigned long long)OCFS2_I(dir)->ip_blkno,
753 i = ocfs2_search_dirblock(bh, dir, name, namelen,
754 block << sb->s_blocksize_bits,
755 bh->b_data, sb->s_blocksize,
758 OCFS2_I(dir)->ip_dir_start_lookup = block;
760 goto cleanup_and_exit;
764 goto cleanup_and_exit;
767 if (++block >= nblocks)
769 } while (block != start);
772 * If the directory has grown while we were searching, then
773 * search the last part of the directory before giving up.
776 nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
777 if (block < nblocks) {
783 /* Clean up the read-ahead blocks */
784 for (; ra_ptr < ra_max; ra_ptr++)
785 brelse(bh_use[ra_ptr]);
787 trace_ocfs2_find_entry_el(ret);
791 static int ocfs2_dx_dir_lookup_rec(struct inode *inode,
792 struct ocfs2_extent_list *el,
796 unsigned int *ret_clen)
798 int ret = 0, i, found;
799 struct buffer_head *eb_bh = NULL;
800 struct ocfs2_extent_block *eb;
801 struct ocfs2_extent_rec *rec = NULL;
803 if (el->l_tree_depth) {
804 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, major_hash,
811 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
814 if (el->l_tree_depth) {
815 ocfs2_error(inode->i_sb,
816 "Inode %lu has non zero tree depth in "
817 "btree tree block %llu\n", inode->i_ino,
818 (unsigned long long)eb_bh->b_blocknr);
825 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
826 rec = &el->l_recs[i];
828 if (le32_to_cpu(rec->e_cpos) <= major_hash) {
835 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
836 "record (%u, %u, 0) in btree", inode->i_ino,
837 le32_to_cpu(rec->e_cpos),
838 ocfs2_rec_clusters(el, rec));
844 *ret_phys_blkno = le64_to_cpu(rec->e_blkno);
846 *ret_cpos = le32_to_cpu(rec->e_cpos);
848 *ret_clen = le16_to_cpu(rec->e_leaf_clusters);
856 * Returns the block index, from the start of the cluster which this
859 static inline unsigned int __ocfs2_dx_dir_hash_idx(struct ocfs2_super *osb,
862 return minor_hash & osb->osb_dx_mask;
865 static inline unsigned int ocfs2_dx_dir_hash_idx(struct ocfs2_super *osb,
866 struct ocfs2_dx_hinfo *hinfo)
868 return __ocfs2_dx_dir_hash_idx(osb, hinfo->minor_hash);
871 static int ocfs2_dx_dir_lookup(struct inode *inode,
872 struct ocfs2_extent_list *el,
873 struct ocfs2_dx_hinfo *hinfo,
878 unsigned int cend, uninitialized_var(clen);
879 u32 uninitialized_var(cpos);
880 u64 uninitialized_var(blkno);
881 u32 name_hash = hinfo->major_hash;
883 ret = ocfs2_dx_dir_lookup_rec(inode, el, name_hash, &cpos, &blkno,
891 if (name_hash >= cend) {
892 /* We want the last cluster */
893 blkno += ocfs2_clusters_to_blocks(inode->i_sb, clen - 1);
896 blkno += ocfs2_clusters_to_blocks(inode->i_sb,
902 * We now have the cluster which should hold our entry. To
903 * find the exact block from the start of the cluster to
904 * search, we take the lower bits of the hash.
906 blkno += ocfs2_dx_dir_hash_idx(OCFS2_SB(inode->i_sb), hinfo);
909 *ret_phys_blkno = blkno;
918 static int ocfs2_dx_dir_search(const char *name, int namelen,
920 struct ocfs2_dx_root_block *dx_root,
921 struct ocfs2_dir_lookup_result *res)
924 u64 uninitialized_var(phys);
925 struct buffer_head *dx_leaf_bh = NULL;
926 struct ocfs2_dx_leaf *dx_leaf;
927 struct ocfs2_dx_entry *dx_entry = NULL;
928 struct buffer_head *dir_ent_bh = NULL;
929 struct ocfs2_dir_entry *dir_ent = NULL;
930 struct ocfs2_dx_hinfo *hinfo = &res->dl_hinfo;
931 struct ocfs2_extent_list *dr_el;
932 struct ocfs2_dx_entry_list *entry_list;
934 ocfs2_dx_dir_name_hash(dir, name, namelen, &res->dl_hinfo);
936 if (ocfs2_dx_root_inline(dx_root)) {
937 entry_list = &dx_root->dr_entries;
941 dr_el = &dx_root->dr_list;
943 ret = ocfs2_dx_dir_lookup(dir, dr_el, hinfo, NULL, &phys);
949 trace_ocfs2_dx_dir_search((unsigned long long)OCFS2_I(dir)->ip_blkno,
950 namelen, name, hinfo->major_hash,
951 hinfo->minor_hash, (unsigned long long)phys);
953 ret = ocfs2_read_dx_leaf(dir, phys, &dx_leaf_bh);
959 dx_leaf = (struct ocfs2_dx_leaf *) dx_leaf_bh->b_data;
961 trace_ocfs2_dx_dir_search_leaf_info(
962 le16_to_cpu(dx_leaf->dl_list.de_num_used),
963 le16_to_cpu(dx_leaf->dl_list.de_count));
965 entry_list = &dx_leaf->dl_list;
969 * Empty leaf is legal, so no need to check for that.
972 for (i = 0; i < le16_to_cpu(entry_list->de_num_used); i++) {
973 dx_entry = &entry_list->de_entries[i];
975 if (hinfo->major_hash != le32_to_cpu(dx_entry->dx_major_hash)
976 || hinfo->minor_hash != le32_to_cpu(dx_entry->dx_minor_hash))
980 * Search unindexed leaf block now. We're not
981 * guaranteed to find anything.
983 ret = ocfs2_read_dir_block_direct(dir,
984 le64_to_cpu(dx_entry->dx_dirent_blk),
992 * XXX: We should check the unindexed block here,
996 found = ocfs2_search_dirblock(dir_ent_bh, dir, name, namelen,
997 0, dir_ent_bh->b_data,
998 dir->i_sb->s_blocksize, &dir_ent);
1003 /* This means we found a bad directory entry. */
1018 res->dl_leaf_bh = dir_ent_bh;
1019 res->dl_entry = dir_ent;
1020 res->dl_dx_leaf_bh = dx_leaf_bh;
1021 res->dl_dx_entry = dx_entry;
1032 static int ocfs2_find_entry_dx(const char *name, int namelen,
1034 struct ocfs2_dir_lookup_result *lookup)
1037 struct buffer_head *di_bh = NULL;
1038 struct ocfs2_dinode *di;
1039 struct buffer_head *dx_root_bh = NULL;
1040 struct ocfs2_dx_root_block *dx_root;
1042 ret = ocfs2_read_inode_block(dir, &di_bh);
1048 di = (struct ocfs2_dinode *)di_bh->b_data;
1050 ret = ocfs2_read_dx_root(dir, di, &dx_root_bh);
1055 dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data;
1057 ret = ocfs2_dx_dir_search(name, namelen, dir, dx_root, lookup);
1064 lookup->dl_dx_root_bh = dx_root_bh;
1073 * Try to find an entry of the provided name within 'dir'.
1075 * If nothing was found, -ENOENT is returned. Otherwise, zero is
1076 * returned and the struct 'res' will contain information useful to
1077 * other directory manipulation functions.
1079 * Caller can NOT assume anything about the contents of the
1080 * buffer_heads - they are passed back only so that it can be passed
1081 * into any one of the manipulation functions (add entry, delete
1082 * entry, etc). As an example, bh in the extent directory case is a
1083 * data block, in the inline-data case it actually points to an inode,
1084 * in the indexed directory case, multiple buffers are involved.
1086 int ocfs2_find_entry(const char *name, int namelen,
1087 struct inode *dir, struct ocfs2_dir_lookup_result *lookup)
1089 struct buffer_head *bh;
1090 struct ocfs2_dir_entry *res_dir = NULL;
1092 if (ocfs2_dir_indexed(dir))
1093 return ocfs2_find_entry_dx(name, namelen, dir, lookup);
1096 * The unindexed dir code only uses part of the lookup
1097 * structure, so there's no reason to push it down further
1100 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
1101 bh = ocfs2_find_entry_id(name, namelen, dir, &res_dir);
1103 bh = ocfs2_find_entry_el(name, namelen, dir, &res_dir);
1108 lookup->dl_leaf_bh = bh;
1109 lookup->dl_entry = res_dir;
1114 * Update inode number and type of a previously found directory entry.
1116 int ocfs2_update_entry(struct inode *dir, handle_t *handle,
1117 struct ocfs2_dir_lookup_result *res,
1118 struct inode *new_entry_inode)
1121 ocfs2_journal_access_func access = ocfs2_journal_access_db;
1122 struct ocfs2_dir_entry *de = res->dl_entry;
1123 struct buffer_head *de_bh = res->dl_leaf_bh;
1126 * The same code works fine for both inline-data and extent
1127 * based directories, so no need to split this up. The only
1128 * difference is the journal_access function.
1131 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
1132 access = ocfs2_journal_access_di;
1134 ret = access(handle, INODE_CACHE(dir), de_bh,
1135 OCFS2_JOURNAL_ACCESS_WRITE);
1141 de->inode = cpu_to_le64(OCFS2_I(new_entry_inode)->ip_blkno);
1142 ocfs2_set_de_type(de, new_entry_inode->i_mode);
1144 ocfs2_journal_dirty(handle, de_bh);
1151 * __ocfs2_delete_entry deletes a directory entry by merging it with the
1154 static int __ocfs2_delete_entry(handle_t *handle, struct inode *dir,
1155 struct ocfs2_dir_entry *de_del,
1156 struct buffer_head *bh, char *first_de,
1159 struct ocfs2_dir_entry *de, *pde;
1160 int i, status = -ENOENT;
1161 ocfs2_journal_access_func access = ocfs2_journal_access_db;
1163 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
1164 access = ocfs2_journal_access_di;
1168 de = (struct ocfs2_dir_entry *) first_de;
1170 if (!ocfs2_check_dir_entry(dir, de, bh, i)) {
1176 status = access(handle, INODE_CACHE(dir), bh,
1177 OCFS2_JOURNAL_ACCESS_WRITE);
1184 le16_add_cpu(&pde->rec_len,
1185 le16_to_cpu(de->rec_len));
1188 ocfs2_journal_dirty(handle, bh);
1191 i += le16_to_cpu(de->rec_len);
1193 de = (struct ocfs2_dir_entry *)((char *)de + le16_to_cpu(de->rec_len));
1199 static unsigned int ocfs2_figure_dirent_hole(struct ocfs2_dir_entry *de)
1203 if (le64_to_cpu(de->inode) == 0)
1204 hole = le16_to_cpu(de->rec_len);
1206 hole = le16_to_cpu(de->rec_len) -
1207 OCFS2_DIR_REC_LEN(de->name_len);
1212 static int ocfs2_find_max_rec_len(struct super_block *sb,
1213 struct buffer_head *dirblock_bh)
1215 int size, this_hole, largest_hole = 0;
1216 char *trailer, *de_buf, *limit, *start = dirblock_bh->b_data;
1217 struct ocfs2_dir_entry *de;
1219 trailer = (char *)ocfs2_trailer_from_bh(dirblock_bh, sb);
1220 size = ocfs2_dir_trailer_blk_off(sb);
1221 limit = start + size;
1223 de = (struct ocfs2_dir_entry *)de_buf;
1225 if (de_buf != trailer) {
1226 this_hole = ocfs2_figure_dirent_hole(de);
1227 if (this_hole > largest_hole)
1228 largest_hole = this_hole;
1231 de_buf += le16_to_cpu(de->rec_len);
1232 de = (struct ocfs2_dir_entry *)de_buf;
1233 } while (de_buf < limit);
1235 if (largest_hole >= OCFS2_DIR_MIN_REC_LEN)
1236 return largest_hole;
1240 static void ocfs2_dx_list_remove_entry(struct ocfs2_dx_entry_list *entry_list,
1243 int num_used = le16_to_cpu(entry_list->de_num_used);
1245 if (num_used == 1 || index == (num_used - 1))
1248 memmove(&entry_list->de_entries[index],
1249 &entry_list->de_entries[index + 1],
1250 (num_used - index - 1)*sizeof(struct ocfs2_dx_entry));
1253 memset(&entry_list->de_entries[num_used], 0,
1254 sizeof(struct ocfs2_dx_entry));
1255 entry_list->de_num_used = cpu_to_le16(num_used);
1258 static int ocfs2_delete_entry_dx(handle_t *handle, struct inode *dir,
1259 struct ocfs2_dir_lookup_result *lookup)
1261 int ret, index, max_rec_len, add_to_free_list = 0;
1262 struct buffer_head *dx_root_bh = lookup->dl_dx_root_bh;
1263 struct buffer_head *leaf_bh = lookup->dl_leaf_bh;
1264 struct ocfs2_dx_leaf *dx_leaf;
1265 struct ocfs2_dx_entry *dx_entry = lookup->dl_dx_entry;
1266 struct ocfs2_dir_block_trailer *trailer;
1267 struct ocfs2_dx_root_block *dx_root;
1268 struct ocfs2_dx_entry_list *entry_list;
1271 * This function gets a bit messy because we might have to
1272 * modify the root block, regardless of whether the indexed
1273 * entries are stored inline.
1277 * *Only* set 'entry_list' here, based on where we're looking
1278 * for the indexed entries. Later, we might still want to
1279 * journal both blocks, based on free list state.
1281 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
1282 if (ocfs2_dx_root_inline(dx_root)) {
1283 entry_list = &dx_root->dr_entries;
1285 dx_leaf = (struct ocfs2_dx_leaf *) lookup->dl_dx_leaf_bh->b_data;
1286 entry_list = &dx_leaf->dl_list;
1289 /* Neither of these are a disk corruption - that should have
1290 * been caught by lookup, before we got here. */
1291 BUG_ON(le16_to_cpu(entry_list->de_count) <= 0);
1292 BUG_ON(le16_to_cpu(entry_list->de_num_used) <= 0);
1294 index = (char *)dx_entry - (char *)entry_list->de_entries;
1295 index /= sizeof(*dx_entry);
1297 if (index >= le16_to_cpu(entry_list->de_num_used)) {
1298 mlog(ML_ERROR, "Dir %llu: Bad dx_entry ptr idx %d, (%p, %p)\n",
1299 (unsigned long long)OCFS2_I(dir)->ip_blkno, index,
1300 entry_list, dx_entry);
1305 * We know that removal of this dirent will leave enough room
1306 * for a new one, so add this block to the free list if it
1307 * isn't already there.
1309 trailer = ocfs2_trailer_from_bh(leaf_bh, dir->i_sb);
1310 if (trailer->db_free_rec_len == 0)
1311 add_to_free_list = 1;
1314 * Add the block holding our index into the journal before
1315 * removing the unindexed entry. If we get an error return
1316 * from __ocfs2_delete_entry(), then it hasn't removed the
1317 * entry yet. Likewise, successful return means we *must*
1318 * remove the indexed entry.
1320 * We're also careful to journal the root tree block here as
1321 * the entry count needs to be updated. Also, we might be
1322 * adding to the start of the free list.
1324 ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
1325 OCFS2_JOURNAL_ACCESS_WRITE);
1331 if (!ocfs2_dx_root_inline(dx_root)) {
1332 ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir),
1333 lookup->dl_dx_leaf_bh,
1334 OCFS2_JOURNAL_ACCESS_WRITE);
1341 trace_ocfs2_delete_entry_dx((unsigned long long)OCFS2_I(dir)->ip_blkno,
1344 ret = __ocfs2_delete_entry(handle, dir, lookup->dl_entry,
1345 leaf_bh, leaf_bh->b_data, leaf_bh->b_size);
1351 max_rec_len = ocfs2_find_max_rec_len(dir->i_sb, leaf_bh);
1352 trailer->db_free_rec_len = cpu_to_le16(max_rec_len);
1353 if (add_to_free_list) {
1354 trailer->db_free_next = dx_root->dr_free_blk;
1355 dx_root->dr_free_blk = cpu_to_le64(leaf_bh->b_blocknr);
1356 ocfs2_journal_dirty(handle, dx_root_bh);
1359 /* leaf_bh was journal_accessed for us in __ocfs2_delete_entry */
1360 ocfs2_journal_dirty(handle, leaf_bh);
1362 le32_add_cpu(&dx_root->dr_num_entries, -1);
1363 ocfs2_journal_dirty(handle, dx_root_bh);
1365 ocfs2_dx_list_remove_entry(entry_list, index);
1367 if (!ocfs2_dx_root_inline(dx_root))
1368 ocfs2_journal_dirty(handle, lookup->dl_dx_leaf_bh);
1374 static inline int ocfs2_delete_entry_id(handle_t *handle,
1376 struct ocfs2_dir_entry *de_del,
1377 struct buffer_head *bh)
1380 struct buffer_head *di_bh = NULL;
1381 struct ocfs2_dinode *di;
1382 struct ocfs2_inline_data *data;
1384 ret = ocfs2_read_inode_block(dir, &di_bh);
1390 di = (struct ocfs2_dinode *)di_bh->b_data;
1391 data = &di->id2.i_data;
1393 ret = __ocfs2_delete_entry(handle, dir, de_del, bh, data->id_data,
1401 static inline int ocfs2_delete_entry_el(handle_t *handle,
1403 struct ocfs2_dir_entry *de_del,
1404 struct buffer_head *bh)
1406 return __ocfs2_delete_entry(handle, dir, de_del, bh, bh->b_data,
1411 * Delete a directory entry. Hide the details of directory
1412 * implementation from the caller.
1414 int ocfs2_delete_entry(handle_t *handle,
1416 struct ocfs2_dir_lookup_result *res)
1418 if (ocfs2_dir_indexed(dir))
1419 return ocfs2_delete_entry_dx(handle, dir, res);
1421 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
1422 return ocfs2_delete_entry_id(handle, dir, res->dl_entry,
1425 return ocfs2_delete_entry_el(handle, dir, res->dl_entry,
1430 * Check whether 'de' has enough room to hold an entry of
1431 * 'new_rec_len' bytes.
1433 static inline int ocfs2_dirent_would_fit(struct ocfs2_dir_entry *de,
1434 unsigned int new_rec_len)
1436 unsigned int de_really_used;
1438 /* Check whether this is an empty record with enough space */
1439 if (le64_to_cpu(de->inode) == 0 &&
1440 le16_to_cpu(de->rec_len) >= new_rec_len)
1444 * Record might have free space at the end which we can
1447 de_really_used = OCFS2_DIR_REC_LEN(de->name_len);
1448 if (le16_to_cpu(de->rec_len) >= (de_really_used + new_rec_len))
1454 static void ocfs2_dx_dir_leaf_insert_tail(struct ocfs2_dx_leaf *dx_leaf,
1455 struct ocfs2_dx_entry *dx_new_entry)
1459 i = le16_to_cpu(dx_leaf->dl_list.de_num_used);
1460 dx_leaf->dl_list.de_entries[i] = *dx_new_entry;
1462 le16_add_cpu(&dx_leaf->dl_list.de_num_used, 1);
1465 static void ocfs2_dx_entry_list_insert(struct ocfs2_dx_entry_list *entry_list,
1466 struct ocfs2_dx_hinfo *hinfo,
1470 struct ocfs2_dx_entry *dx_entry;
1472 i = le16_to_cpu(entry_list->de_num_used);
1473 dx_entry = &entry_list->de_entries[i];
1475 memset(dx_entry, 0, sizeof(*dx_entry));
1476 dx_entry->dx_major_hash = cpu_to_le32(hinfo->major_hash);
1477 dx_entry->dx_minor_hash = cpu_to_le32(hinfo->minor_hash);
1478 dx_entry->dx_dirent_blk = cpu_to_le64(dirent_blk);
1480 le16_add_cpu(&entry_list->de_num_used, 1);
1483 static int __ocfs2_dx_dir_leaf_insert(struct inode *dir, handle_t *handle,
1484 struct ocfs2_dx_hinfo *hinfo,
1486 struct buffer_head *dx_leaf_bh)
1489 struct ocfs2_dx_leaf *dx_leaf;
1491 ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir), dx_leaf_bh,
1492 OCFS2_JOURNAL_ACCESS_WRITE);
1498 dx_leaf = (struct ocfs2_dx_leaf *)dx_leaf_bh->b_data;
1499 ocfs2_dx_entry_list_insert(&dx_leaf->dl_list, hinfo, dirent_blk);
1500 ocfs2_journal_dirty(handle, dx_leaf_bh);
1506 static void ocfs2_dx_inline_root_insert(struct inode *dir, handle_t *handle,
1507 struct ocfs2_dx_hinfo *hinfo,
1509 struct ocfs2_dx_root_block *dx_root)
1511 ocfs2_dx_entry_list_insert(&dx_root->dr_entries, hinfo, dirent_blk);
1514 static int ocfs2_dx_dir_insert(struct inode *dir, handle_t *handle,
1515 struct ocfs2_dir_lookup_result *lookup)
1518 struct ocfs2_dx_root_block *dx_root;
1519 struct buffer_head *dx_root_bh = lookup->dl_dx_root_bh;
1521 ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
1522 OCFS2_JOURNAL_ACCESS_WRITE);
1528 dx_root = (struct ocfs2_dx_root_block *)lookup->dl_dx_root_bh->b_data;
1529 if (ocfs2_dx_root_inline(dx_root)) {
1530 ocfs2_dx_inline_root_insert(dir, handle,
1532 lookup->dl_leaf_bh->b_blocknr,
1535 ret = __ocfs2_dx_dir_leaf_insert(dir, handle, &lookup->dl_hinfo,
1536 lookup->dl_leaf_bh->b_blocknr,
1537 lookup->dl_dx_leaf_bh);
1542 le32_add_cpu(&dx_root->dr_num_entries, 1);
1543 ocfs2_journal_dirty(handle, dx_root_bh);
1549 static void ocfs2_remove_block_from_free_list(struct inode *dir,
1551 struct ocfs2_dir_lookup_result *lookup)
1553 struct ocfs2_dir_block_trailer *trailer, *prev;
1554 struct ocfs2_dx_root_block *dx_root;
1555 struct buffer_head *bh;
1557 trailer = ocfs2_trailer_from_bh(lookup->dl_leaf_bh, dir->i_sb);
1559 if (ocfs2_free_list_at_root(lookup)) {
1560 bh = lookup->dl_dx_root_bh;
1561 dx_root = (struct ocfs2_dx_root_block *)bh->b_data;
1562 dx_root->dr_free_blk = trailer->db_free_next;
1564 bh = lookup->dl_prev_leaf_bh;
1565 prev = ocfs2_trailer_from_bh(bh, dir->i_sb);
1566 prev->db_free_next = trailer->db_free_next;
1569 trailer->db_free_rec_len = cpu_to_le16(0);
1570 trailer->db_free_next = cpu_to_le64(0);
1572 ocfs2_journal_dirty(handle, bh);
1573 ocfs2_journal_dirty(handle, lookup->dl_leaf_bh);
1577 * This expects that a journal write has been reserved on
1578 * lookup->dl_prev_leaf_bh or lookup->dl_dx_root_bh
1580 static void ocfs2_recalc_free_list(struct inode *dir, handle_t *handle,
1581 struct ocfs2_dir_lookup_result *lookup)
1584 struct ocfs2_dir_block_trailer *trailer;
1586 /* Walk dl_leaf_bh to figure out what the new free rec_len is. */
1587 max_rec_len = ocfs2_find_max_rec_len(dir->i_sb, lookup->dl_leaf_bh);
1590 * There's still room in this block, so no need to remove it
1591 * from the free list. In this case, we just want to update
1592 * the rec len accounting.
1594 trailer = ocfs2_trailer_from_bh(lookup->dl_leaf_bh, dir->i_sb);
1595 trailer->db_free_rec_len = cpu_to_le16(max_rec_len);
1596 ocfs2_journal_dirty(handle, lookup->dl_leaf_bh);
1598 ocfs2_remove_block_from_free_list(dir, handle, lookup);
1602 /* we don't always have a dentry for what we want to add, so people
1603 * like orphan dir can call this instead.
1605 * The lookup context must have been filled from
1606 * ocfs2_prepare_dir_for_insert.
1608 int __ocfs2_add_entry(handle_t *handle,
1610 const char *name, int namelen,
1611 struct inode *inode, u64 blkno,
1612 struct buffer_head *parent_fe_bh,
1613 struct ocfs2_dir_lookup_result *lookup)
1615 unsigned long offset;
1616 unsigned short rec_len;
1617 struct ocfs2_dir_entry *de, *de1;
1618 struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_fe_bh->b_data;
1619 struct super_block *sb = dir->i_sb;
1621 unsigned int size = sb->s_blocksize;
1622 struct buffer_head *insert_bh = lookup->dl_leaf_bh;
1623 char *data_start = insert_bh->b_data;
1628 if (ocfs2_dir_indexed(dir)) {
1629 struct buffer_head *bh;
1632 * An indexed dir may require that we update the free space
1633 * list. Reserve a write to the previous node in the list so
1634 * that we don't fail later.
1636 * XXX: This can be either a dx_root_block, or an unindexed
1637 * directory tree leaf block.
1639 if (ocfs2_free_list_at_root(lookup)) {
1640 bh = lookup->dl_dx_root_bh;
1641 retval = ocfs2_journal_access_dr(handle,
1642 INODE_CACHE(dir), bh,
1643 OCFS2_JOURNAL_ACCESS_WRITE);
1645 bh = lookup->dl_prev_leaf_bh;
1646 retval = ocfs2_journal_access_db(handle,
1647 INODE_CACHE(dir), bh,
1648 OCFS2_JOURNAL_ACCESS_WRITE);
1654 } else if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1655 data_start = di->id2.i_data.id_data;
1656 size = i_size_read(dir);
1658 BUG_ON(insert_bh != parent_fe_bh);
1661 rec_len = OCFS2_DIR_REC_LEN(namelen);
1663 de = (struct ocfs2_dir_entry *) data_start;
1665 BUG_ON((char *)de >= (size + data_start));
1667 /* These checks should've already been passed by the
1668 * prepare function, but I guess we can leave them
1670 if (!ocfs2_check_dir_entry(dir, de, insert_bh, offset)) {
1674 if (ocfs2_match(namelen, name, de)) {
1679 /* We're guaranteed that we should have space, so we
1680 * can't possibly have hit the trailer...right? */
1681 mlog_bug_on_msg(ocfs2_skip_dir_trailer(dir, de, offset, size),
1682 "Hit dir trailer trying to insert %.*s "
1683 "(namelen %d) into directory %llu. "
1684 "offset is %lu, trailer offset is %d\n",
1685 namelen, name, namelen,
1686 (unsigned long long)parent_fe_bh->b_blocknr,
1687 offset, ocfs2_dir_trailer_blk_off(dir->i_sb));
1689 if (ocfs2_dirent_would_fit(de, rec_len)) {
1690 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1691 retval = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
1697 if (insert_bh == parent_fe_bh)
1698 status = ocfs2_journal_access_di(handle,
1701 OCFS2_JOURNAL_ACCESS_WRITE);
1703 status = ocfs2_journal_access_db(handle,
1706 OCFS2_JOURNAL_ACCESS_WRITE);
1708 if (ocfs2_dir_indexed(dir)) {
1709 status = ocfs2_dx_dir_insert(dir,
1719 /* By now the buffer is marked for journaling */
1720 offset += le16_to_cpu(de->rec_len);
1721 if (le64_to_cpu(de->inode)) {
1722 de1 = (struct ocfs2_dir_entry *)((char *) de +
1723 OCFS2_DIR_REC_LEN(de->name_len));
1725 cpu_to_le16(le16_to_cpu(de->rec_len) -
1726 OCFS2_DIR_REC_LEN(de->name_len));
1727 de->rec_len = cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
1730 de->file_type = OCFS2_FT_UNKNOWN;
1732 de->inode = cpu_to_le64(blkno);
1733 ocfs2_set_de_type(de, inode->i_mode);
1736 de->name_len = namelen;
1737 memcpy(de->name, name, namelen);
1739 if (ocfs2_dir_indexed(dir))
1740 ocfs2_recalc_free_list(dir, handle, lookup);
1743 ocfs2_journal_dirty(handle, insert_bh);
1748 offset += le16_to_cpu(de->rec_len);
1749 de = (struct ocfs2_dir_entry *) ((char *) de + le16_to_cpu(de->rec_len));
1752 /* when you think about it, the assert above should prevent us
1753 * from ever getting here. */
1762 static int ocfs2_dir_foreach_blk_id(struct inode *inode,
1764 struct dir_context *ctx)
1767 unsigned long offset = ctx->pos;
1768 struct buffer_head *di_bh = NULL;
1769 struct ocfs2_dinode *di;
1770 struct ocfs2_inline_data *data;
1771 struct ocfs2_dir_entry *de;
1773 ret = ocfs2_read_inode_block(inode, &di_bh);
1775 mlog(ML_ERROR, "Unable to read inode block for dir %llu\n",
1776 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1780 di = (struct ocfs2_dinode *)di_bh->b_data;
1781 data = &di->id2.i_data;
1783 while (ctx->pos < i_size_read(inode)) {
1784 /* If the dir block has changed since the last call to
1785 * readdir(2), then we might be pointing to an invalid
1786 * dirent right now. Scan from the start of the block
1788 if (*f_version != inode->i_version) {
1789 for (i = 0; i < i_size_read(inode) && i < offset; ) {
1790 de = (struct ocfs2_dir_entry *)
1791 (data->id_data + i);
1792 /* It's too expensive to do a full
1793 * dirent test each time round this
1794 * loop, but we do have to test at
1795 * least that it is non-zero. A
1796 * failure will be detected in the
1797 * dirent test below. */
1798 if (le16_to_cpu(de->rec_len) <
1799 OCFS2_DIR_REC_LEN(1))
1801 i += le16_to_cpu(de->rec_len);
1803 ctx->pos = offset = i;
1804 *f_version = inode->i_version;
1807 de = (struct ocfs2_dir_entry *) (data->id_data + ctx->pos);
1808 if (!ocfs2_check_dir_entry(inode, de, di_bh, ctx->pos)) {
1809 /* On error, skip the f_pos to the end. */
1810 ctx->pos = i_size_read(inode);
1813 offset += le16_to_cpu(de->rec_len);
1814 if (le64_to_cpu(de->inode)) {
1815 unsigned char d_type = DT_UNKNOWN;
1817 if (de->file_type < OCFS2_FT_MAX)
1818 d_type = ocfs2_filetype_table[de->file_type];
1820 if (!dir_emit(ctx, de->name, de->name_len,
1821 le64_to_cpu(de->inode), d_type))
1824 ctx->pos += le16_to_cpu(de->rec_len);
1832 * NOTE: This function can be called against unindexed directories,
1835 static int ocfs2_dir_foreach_blk_el(struct inode *inode,
1837 struct dir_context *ctx,
1840 unsigned long offset, blk, last_ra_blk = 0;
1842 struct buffer_head * bh, * tmp;
1843 struct ocfs2_dir_entry * de;
1844 struct super_block * sb = inode->i_sb;
1845 unsigned int ra_sectors = 16;
1850 offset = ctx->pos & (sb->s_blocksize - 1);
1852 while (ctx->pos < i_size_read(inode)) {
1853 blk = ctx->pos >> sb->s_blocksize_bits;
1854 if (ocfs2_read_dir_block(inode, blk, &bh, 0)) {
1855 /* Skip the corrupt dirblock and keep trying */
1856 ctx->pos += sb->s_blocksize - offset;
1860 /* The idea here is to begin with 8k read-ahead and to stay
1861 * 4k ahead of our current position.
1863 * TODO: Use the pagecache for this. We just need to
1864 * make sure it's cluster-safe... */
1866 || (((last_ra_blk - blk) << 9) <= (ra_sectors / 2))) {
1867 for (i = ra_sectors >> (sb->s_blocksize_bits - 9);
1870 if (!ocfs2_read_dir_block(inode, ++blk, &tmp,
1871 OCFS2_BH_READAHEAD))
1878 /* If the dir block has changed since the last call to
1879 * readdir(2), then we might be pointing to an invalid
1880 * dirent right now. Scan from the start of the block
1882 if (*f_version != inode->i_version) {
1883 for (i = 0; i < sb->s_blocksize && i < offset; ) {
1884 de = (struct ocfs2_dir_entry *) (bh->b_data + i);
1885 /* It's too expensive to do a full
1886 * dirent test each time round this
1887 * loop, but we do have to test at
1888 * least that it is non-zero. A
1889 * failure will be detected in the
1890 * dirent test below. */
1891 if (le16_to_cpu(de->rec_len) <
1892 OCFS2_DIR_REC_LEN(1))
1894 i += le16_to_cpu(de->rec_len);
1897 ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1))
1899 *f_version = inode->i_version;
1902 while (ctx->pos < i_size_read(inode)
1903 && offset < sb->s_blocksize) {
1904 de = (struct ocfs2_dir_entry *) (bh->b_data + offset);
1905 if (!ocfs2_check_dir_entry(inode, de, bh, offset)) {
1906 /* On error, skip the f_pos to the
1908 ctx->pos = (ctx->pos | (sb->s_blocksize - 1)) + 1;
1912 if (le64_to_cpu(de->inode)) {
1913 unsigned char d_type = DT_UNKNOWN;
1915 if (de->file_type < OCFS2_FT_MAX)
1916 d_type = ocfs2_filetype_table[de->file_type];
1917 if (!dir_emit(ctx, de->name,
1919 le64_to_cpu(de->inode),
1926 offset += le16_to_cpu(de->rec_len);
1927 ctx->pos += le16_to_cpu(de->rec_len);
1932 if (!persist && stored)
1938 static int ocfs2_dir_foreach_blk(struct inode *inode, u64 *f_version,
1939 struct dir_context *ctx,
1942 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
1943 return ocfs2_dir_foreach_blk_id(inode, f_version, ctx);
1944 return ocfs2_dir_foreach_blk_el(inode, f_version, ctx, persist);
1948 * This is intended to be called from inside other kernel functions,
1949 * so we fake some arguments.
1951 int ocfs2_dir_foreach(struct inode *inode, struct dir_context *ctx)
1953 u64 version = inode->i_version;
1954 ocfs2_dir_foreach_blk(inode, &version, ctx, true);
1962 int ocfs2_readdir(struct file *file, struct dir_context *ctx)
1965 struct inode *inode = file_inode(file);
1968 trace_ocfs2_readdir((unsigned long long)OCFS2_I(inode)->ip_blkno);
1970 error = ocfs2_inode_lock_atime(inode, file->f_path.mnt, &lock_level);
1971 if (lock_level && error >= 0) {
1972 /* We release EX lock which used to update atime
1973 * and get PR lock again to reduce contention
1974 * on commonly accessed directories. */
1975 ocfs2_inode_unlock(inode, 1);
1977 error = ocfs2_inode_lock(inode, NULL, 0);
1980 if (error != -ENOENT)
1982 /* we haven't got any yet, so propagate the error. */
1986 error = ocfs2_dir_foreach_blk(inode, &file->f_version, ctx, false);
1988 ocfs2_inode_unlock(inode, lock_level);
1998 * NOTE: this should always be called with parent dir i_mutex taken.
2000 int ocfs2_find_files_on_disk(const char *name,
2003 struct inode *inode,
2004 struct ocfs2_dir_lookup_result *lookup)
2006 int status = -ENOENT;
2008 trace_ocfs2_find_files_on_disk(namelen, name, blkno,
2009 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2011 status = ocfs2_find_entry(name, namelen, inode, lookup);
2015 *blkno = le64_to_cpu(lookup->dl_entry->inode);
2024 * Convenience function for callers which just want the block number
2025 * mapped to a name and don't require the full dirent info, etc.
2027 int ocfs2_lookup_ino_from_name(struct inode *dir, const char *name,
2028 int namelen, u64 *blkno)
2031 struct ocfs2_dir_lookup_result lookup = { NULL, };
2033 ret = ocfs2_find_files_on_disk(name, namelen, blkno, dir, &lookup);
2034 ocfs2_free_dir_lookup_result(&lookup);
2039 /* Check for a name within a directory.
2041 * Return 0 if the name does not exist
2042 * Return -EEXIST if the directory contains the name
2044 * Callers should have i_mutex + a cluster lock on dir
2046 int ocfs2_check_dir_for_entry(struct inode *dir,
2051 struct ocfs2_dir_lookup_result lookup = { NULL, };
2053 trace_ocfs2_check_dir_for_entry(
2054 (unsigned long long)OCFS2_I(dir)->ip_blkno, namelen, name);
2057 if (ocfs2_find_entry(name, namelen, dir, &lookup) == 0)
2062 ocfs2_free_dir_lookup_result(&lookup);
2069 struct ocfs2_empty_dir_priv {
2070 struct dir_context ctx;
2072 unsigned seen_dot_dot;
2073 unsigned seen_other;
2076 static int ocfs2_empty_dir_filldir(struct dir_context *ctx, const char *name,
2077 int name_len, loff_t pos, u64 ino,
2080 struct ocfs2_empty_dir_priv *p =
2081 container_of(ctx, struct ocfs2_empty_dir_priv, ctx);
2084 * Check the positions of "." and ".." records to be sure
2085 * they're in the correct place.
2087 * Indexed directories don't need to proceed past the first
2088 * two entries, so we end the scan after seeing '..'. Despite
2089 * that, we allow the scan to proceed In the event that we
2090 * have a corrupted indexed directory (no dot or dot dot
2091 * entries). This allows us to double check for existing
2092 * entries which might not have been found in the index.
2094 if (name_len == 1 && !strncmp(".", name, 1) && pos == 0) {
2099 if (name_len == 2 && !strncmp("..", name, 2) &&
2100 pos == OCFS2_DIR_REC_LEN(1)) {
2101 p->seen_dot_dot = 1;
2103 if (p->dx_dir && p->seen_dot)
2113 static int ocfs2_empty_dir_dx(struct inode *inode,
2114 struct ocfs2_empty_dir_priv *priv)
2117 struct buffer_head *di_bh = NULL;
2118 struct buffer_head *dx_root_bh = NULL;
2119 struct ocfs2_dinode *di;
2120 struct ocfs2_dx_root_block *dx_root;
2124 ret = ocfs2_read_inode_block(inode, &di_bh);
2129 di = (struct ocfs2_dinode *)di_bh->b_data;
2131 ret = ocfs2_read_dx_root(inode, di, &dx_root_bh);
2136 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
2138 if (le32_to_cpu(dx_root->dr_num_entries) != 2)
2139 priv->seen_other = 1;
2148 * routine to check that the specified directory is empty (for rmdir)
2150 * Returns 1 if dir is empty, zero otherwise.
2152 * XXX: This is a performance problem for unindexed directories.
2154 int ocfs2_empty_dir(struct inode *inode)
2157 struct ocfs2_empty_dir_priv priv = {
2158 .ctx.actor = ocfs2_empty_dir_filldir,
2161 if (ocfs2_dir_indexed(inode)) {
2162 ret = ocfs2_empty_dir_dx(inode, &priv);
2166 * We still run ocfs2_dir_foreach to get the checks
2171 ret = ocfs2_dir_foreach(inode, &priv.ctx);
2175 if (!priv.seen_dot || !priv.seen_dot_dot) {
2176 mlog(ML_ERROR, "bad directory (dir #%llu) - no `.' or `..'\n",
2177 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2179 * XXX: Is it really safe to allow an unlink to continue?
2184 return !priv.seen_other;
2188 * Fills "." and ".." dirents in a new directory block. Returns dirent for
2189 * "..", which might be used during creation of a directory with a trailing
2190 * header. It is otherwise safe to ignore the return code.
2192 static struct ocfs2_dir_entry *ocfs2_fill_initial_dirents(struct inode *inode,
2193 struct inode *parent,
2197 struct ocfs2_dir_entry *de = (struct ocfs2_dir_entry *)start;
2199 de->inode = cpu_to_le64(OCFS2_I(inode)->ip_blkno);
2202 cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
2203 strcpy(de->name, ".");
2204 ocfs2_set_de_type(de, S_IFDIR);
2206 de = (struct ocfs2_dir_entry *) ((char *)de + le16_to_cpu(de->rec_len));
2207 de->inode = cpu_to_le64(OCFS2_I(parent)->ip_blkno);
2208 de->rec_len = cpu_to_le16(size - OCFS2_DIR_REC_LEN(1));
2210 strcpy(de->name, "..");
2211 ocfs2_set_de_type(de, S_IFDIR);
2217 * This works together with code in ocfs2_mknod_locked() which sets
2218 * the inline-data flag and initializes the inline-data section.
2220 static int ocfs2_fill_new_dir_id(struct ocfs2_super *osb,
2222 struct inode *parent,
2223 struct inode *inode,
2224 struct buffer_head *di_bh)
2227 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2228 struct ocfs2_inline_data *data = &di->id2.i_data;
2229 unsigned int size = le16_to_cpu(data->id_count);
2231 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2232 OCFS2_JOURNAL_ACCESS_WRITE);
2238 ocfs2_fill_initial_dirents(inode, parent, data->id_data, size);
2239 ocfs2_journal_dirty(handle, di_bh);
2241 i_size_write(inode, size);
2242 set_nlink(inode, 2);
2243 inode->i_blocks = ocfs2_inode_sector_count(inode);
2245 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
2253 static int ocfs2_fill_new_dir_el(struct ocfs2_super *osb,
2255 struct inode *parent,
2256 struct inode *inode,
2257 struct buffer_head *fe_bh,
2258 struct ocfs2_alloc_context *data_ac,
2259 struct buffer_head **ret_new_bh)
2262 unsigned int size = osb->sb->s_blocksize;
2263 struct buffer_head *new_bh = NULL;
2264 struct ocfs2_dir_entry *de;
2266 if (ocfs2_new_dir_wants_trailer(inode))
2267 size = ocfs2_dir_trailer_blk_off(parent->i_sb);
2269 status = ocfs2_do_extend_dir(osb->sb, handle, inode, fe_bh,
2270 data_ac, NULL, &new_bh);
2276 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2278 status = ocfs2_journal_access_db(handle, INODE_CACHE(inode), new_bh,
2279 OCFS2_JOURNAL_ACCESS_CREATE);
2284 memset(new_bh->b_data, 0, osb->sb->s_blocksize);
2286 de = ocfs2_fill_initial_dirents(inode, parent, new_bh->b_data, size);
2287 if (ocfs2_new_dir_wants_trailer(inode)) {
2288 int size = le16_to_cpu(de->rec_len);
2291 * Figure out the size of the hole left over after
2292 * insertion of '.' and '..'. The trailer wants this
2295 size -= OCFS2_DIR_REC_LEN(2);
2296 size -= sizeof(struct ocfs2_dir_block_trailer);
2298 ocfs2_init_dir_trailer(inode, new_bh, size);
2301 ocfs2_journal_dirty(handle, new_bh);
2303 i_size_write(inode, inode->i_sb->s_blocksize);
2304 set_nlink(inode, 2);
2305 inode->i_blocks = ocfs2_inode_sector_count(inode);
2306 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
2314 *ret_new_bh = new_bh;
2323 static int ocfs2_dx_dir_attach_index(struct ocfs2_super *osb,
2324 handle_t *handle, struct inode *dir,
2325 struct buffer_head *di_bh,
2326 struct buffer_head *dirdata_bh,
2327 struct ocfs2_alloc_context *meta_ac,
2328 int dx_inline, u32 num_entries,
2329 struct buffer_head **ret_dx_root_bh)
2332 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
2333 u16 dr_suballoc_bit;
2334 u64 suballoc_loc, dr_blkno;
2335 unsigned int num_bits;
2336 struct buffer_head *dx_root_bh = NULL;
2337 struct ocfs2_dx_root_block *dx_root;
2338 struct ocfs2_dir_block_trailer *trailer =
2339 ocfs2_trailer_from_bh(dirdata_bh, dir->i_sb);
2341 ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
2342 &dr_suballoc_bit, &num_bits, &dr_blkno);
2348 trace_ocfs2_dx_dir_attach_index(
2349 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2350 (unsigned long long)dr_blkno);
2352 dx_root_bh = sb_getblk(osb->sb, dr_blkno);
2353 if (dx_root_bh == NULL) {
2357 ocfs2_set_new_buffer_uptodate(INODE_CACHE(dir), dx_root_bh);
2359 ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
2360 OCFS2_JOURNAL_ACCESS_CREATE);
2366 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
2367 memset(dx_root, 0, osb->sb->s_blocksize);
2368 strcpy(dx_root->dr_signature, OCFS2_DX_ROOT_SIGNATURE);
2369 dx_root->dr_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
2370 dx_root->dr_suballoc_loc = cpu_to_le64(suballoc_loc);
2371 dx_root->dr_suballoc_bit = cpu_to_le16(dr_suballoc_bit);
2372 dx_root->dr_fs_generation = cpu_to_le32(osb->fs_generation);
2373 dx_root->dr_blkno = cpu_to_le64(dr_blkno);
2374 dx_root->dr_dir_blkno = cpu_to_le64(OCFS2_I(dir)->ip_blkno);
2375 dx_root->dr_num_entries = cpu_to_le32(num_entries);
2376 if (le16_to_cpu(trailer->db_free_rec_len))
2377 dx_root->dr_free_blk = cpu_to_le64(dirdata_bh->b_blocknr);
2379 dx_root->dr_free_blk = cpu_to_le64(0);
2382 dx_root->dr_flags |= OCFS2_DX_FLAG_INLINE;
2383 dx_root->dr_entries.de_count =
2384 cpu_to_le16(ocfs2_dx_entries_per_root(osb->sb));
2386 dx_root->dr_list.l_count =
2387 cpu_to_le16(ocfs2_extent_recs_per_dx_root(osb->sb));
2389 ocfs2_journal_dirty(handle, dx_root_bh);
2391 ret = ocfs2_journal_access_di(handle, INODE_CACHE(dir), di_bh,
2392 OCFS2_JOURNAL_ACCESS_CREATE);
2398 di->i_dx_root = cpu_to_le64(dr_blkno);
2400 spin_lock(&OCFS2_I(dir)->ip_lock);
2401 OCFS2_I(dir)->ip_dyn_features |= OCFS2_INDEXED_DIR_FL;
2402 di->i_dyn_features = cpu_to_le16(OCFS2_I(dir)->ip_dyn_features);
2403 spin_unlock(&OCFS2_I(dir)->ip_lock);
2405 ocfs2_journal_dirty(handle, di_bh);
2407 *ret_dx_root_bh = dx_root_bh;
2415 static int ocfs2_dx_dir_format_cluster(struct ocfs2_super *osb,
2416 handle_t *handle, struct inode *dir,
2417 struct buffer_head **dx_leaves,
2418 int num_dx_leaves, u64 start_blk)
2421 struct ocfs2_dx_leaf *dx_leaf;
2422 struct buffer_head *bh;
2424 for (i = 0; i < num_dx_leaves; i++) {
2425 bh = sb_getblk(osb->sb, start_blk + i);
2432 ocfs2_set_new_buffer_uptodate(INODE_CACHE(dir), bh);
2434 ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir), bh,
2435 OCFS2_JOURNAL_ACCESS_CREATE);
2441 dx_leaf = (struct ocfs2_dx_leaf *) bh->b_data;
2443 memset(dx_leaf, 0, osb->sb->s_blocksize);
2444 strcpy(dx_leaf->dl_signature, OCFS2_DX_LEAF_SIGNATURE);
2445 dx_leaf->dl_fs_generation = cpu_to_le32(osb->fs_generation);
2446 dx_leaf->dl_blkno = cpu_to_le64(bh->b_blocknr);
2447 dx_leaf->dl_list.de_count =
2448 cpu_to_le16(ocfs2_dx_entries_per_leaf(osb->sb));
2450 trace_ocfs2_dx_dir_format_cluster(
2451 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2452 (unsigned long long)bh->b_blocknr,
2453 le16_to_cpu(dx_leaf->dl_list.de_count));
2455 ocfs2_journal_dirty(handle, bh);
2464 * Allocates and formats a new cluster for use in an indexed dir
2465 * leaf. This version will not do the extent insert, so that it can be
2466 * used by operations which need careful ordering.
2468 static int __ocfs2_dx_dir_new_cluster(struct inode *dir,
2469 u32 cpos, handle_t *handle,
2470 struct ocfs2_alloc_context *data_ac,
2471 struct buffer_head **dx_leaves,
2472 int num_dx_leaves, u64 *ret_phys_blkno)
2477 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2480 * XXX: For create, this should claim cluster for the index
2481 * *before* the unindexed insert so that we have a better
2482 * chance of contiguousness as the directory grows in number
2485 ret = __ocfs2_claim_clusters(handle, data_ac, 1, 1, &phys, &num);
2492 * Format the new cluster first. That way, we're inserting
2495 phys_blkno = ocfs2_clusters_to_blocks(osb->sb, phys);
2496 ret = ocfs2_dx_dir_format_cluster(osb, handle, dir, dx_leaves,
2497 num_dx_leaves, phys_blkno);
2503 *ret_phys_blkno = phys_blkno;
2508 static int ocfs2_dx_dir_new_cluster(struct inode *dir,
2509 struct ocfs2_extent_tree *et,
2510 u32 cpos, handle_t *handle,
2511 struct ocfs2_alloc_context *data_ac,
2512 struct ocfs2_alloc_context *meta_ac,
2513 struct buffer_head **dx_leaves,
2519 ret = __ocfs2_dx_dir_new_cluster(dir, cpos, handle, data_ac, dx_leaves,
2520 num_dx_leaves, &phys_blkno);
2526 ret = ocfs2_insert_extent(handle, et, cpos, phys_blkno, 1, 0,
2534 static struct buffer_head **ocfs2_dx_dir_kmalloc_leaves(struct super_block *sb,
2535 int *ret_num_leaves)
2537 int num_dx_leaves = ocfs2_clusters_to_blocks(sb, 1);
2538 struct buffer_head **dx_leaves;
2540 dx_leaves = kcalloc(num_dx_leaves, sizeof(struct buffer_head *),
2542 if (dx_leaves && ret_num_leaves)
2543 *ret_num_leaves = num_dx_leaves;
2548 static int ocfs2_fill_new_dir_dx(struct ocfs2_super *osb,
2550 struct inode *parent,
2551 struct inode *inode,
2552 struct buffer_head *di_bh,
2553 struct ocfs2_alloc_context *data_ac,
2554 struct ocfs2_alloc_context *meta_ac)
2557 struct buffer_head *leaf_bh = NULL;
2558 struct buffer_head *dx_root_bh = NULL;
2559 struct ocfs2_dx_hinfo hinfo;
2560 struct ocfs2_dx_root_block *dx_root;
2561 struct ocfs2_dx_entry_list *entry_list;
2564 * Our strategy is to create the directory as though it were
2565 * unindexed, then add the index block. This works with very
2566 * little complication since the state of a new directory is a
2567 * very well known quantity.
2569 * Essentially, we have two dirents ("." and ".."), in the 1st
2570 * block which need indexing. These are easily inserted into
2574 ret = ocfs2_fill_new_dir_el(osb, handle, parent, inode, di_bh,
2581 ret = ocfs2_dx_dir_attach_index(osb, handle, inode, di_bh, leaf_bh,
2582 meta_ac, 1, 2, &dx_root_bh);
2587 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
2588 entry_list = &dx_root->dr_entries;
2590 /* Buffer has been journaled for us by ocfs2_dx_dir_attach_index */
2591 ocfs2_dx_dir_name_hash(inode, ".", 1, &hinfo);
2592 ocfs2_dx_entry_list_insert(entry_list, &hinfo, leaf_bh->b_blocknr);
2594 ocfs2_dx_dir_name_hash(inode, "..", 2, &hinfo);
2595 ocfs2_dx_entry_list_insert(entry_list, &hinfo, leaf_bh->b_blocknr);
2603 int ocfs2_fill_new_dir(struct ocfs2_super *osb,
2605 struct inode *parent,
2606 struct inode *inode,
2607 struct buffer_head *fe_bh,
2608 struct ocfs2_alloc_context *data_ac,
2609 struct ocfs2_alloc_context *meta_ac)
2612 BUG_ON(!ocfs2_supports_inline_data(osb) && data_ac == NULL);
2614 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2615 return ocfs2_fill_new_dir_id(osb, handle, parent, inode, fe_bh);
2617 if (ocfs2_supports_indexed_dirs(osb))
2618 return ocfs2_fill_new_dir_dx(osb, handle, parent, inode, fe_bh,
2621 return ocfs2_fill_new_dir_el(osb, handle, parent, inode, fe_bh,
2625 static int ocfs2_dx_dir_index_block(struct inode *dir,
2627 struct buffer_head **dx_leaves,
2629 u32 *num_dx_entries,
2630 struct buffer_head *dirent_bh)
2632 int ret = 0, namelen, i;
2633 char *de_buf, *limit;
2634 struct ocfs2_dir_entry *de;
2635 struct buffer_head *dx_leaf_bh;
2636 struct ocfs2_dx_hinfo hinfo;
2637 u64 dirent_blk = dirent_bh->b_blocknr;
2639 de_buf = dirent_bh->b_data;
2640 limit = de_buf + dir->i_sb->s_blocksize;
2642 while (de_buf < limit) {
2643 de = (struct ocfs2_dir_entry *)de_buf;
2645 namelen = de->name_len;
2646 if (!namelen || !de->inode)
2649 ocfs2_dx_dir_name_hash(dir, de->name, namelen, &hinfo);
2651 i = ocfs2_dx_dir_hash_idx(OCFS2_SB(dir->i_sb), &hinfo);
2652 dx_leaf_bh = dx_leaves[i];
2654 ret = __ocfs2_dx_dir_leaf_insert(dir, handle, &hinfo,
2655 dirent_blk, dx_leaf_bh);
2661 *num_dx_entries = *num_dx_entries + 1;
2664 de_buf += le16_to_cpu(de->rec_len);
2672 * XXX: This expects dx_root_bh to already be part of the transaction.
2674 static void ocfs2_dx_dir_index_root_block(struct inode *dir,
2675 struct buffer_head *dx_root_bh,
2676 struct buffer_head *dirent_bh)
2678 char *de_buf, *limit;
2679 struct ocfs2_dx_root_block *dx_root;
2680 struct ocfs2_dir_entry *de;
2681 struct ocfs2_dx_hinfo hinfo;
2682 u64 dirent_blk = dirent_bh->b_blocknr;
2684 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
2686 de_buf = dirent_bh->b_data;
2687 limit = de_buf + dir->i_sb->s_blocksize;
2689 while (de_buf < limit) {
2690 de = (struct ocfs2_dir_entry *)de_buf;
2692 if (!de->name_len || !de->inode)
2695 ocfs2_dx_dir_name_hash(dir, de->name, de->name_len, &hinfo);
2697 trace_ocfs2_dx_dir_index_root_block(
2698 (unsigned long long)dir->i_ino,
2699 hinfo.major_hash, hinfo.minor_hash,
2700 de->name_len, de->name,
2701 le16_to_cpu(dx_root->dr_entries.de_num_used));
2703 ocfs2_dx_entry_list_insert(&dx_root->dr_entries, &hinfo,
2706 le32_add_cpu(&dx_root->dr_num_entries, 1);
2708 de_buf += le16_to_cpu(de->rec_len);
2713 * Count the number of inline directory entries in di_bh and compare
2714 * them against the number of entries we can hold in an inline dx root
2717 static int ocfs2_new_dx_should_be_inline(struct inode *dir,
2718 struct buffer_head *di_bh)
2720 int dirent_count = 0;
2721 char *de_buf, *limit;
2722 struct ocfs2_dir_entry *de;
2723 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2725 de_buf = di->id2.i_data.id_data;
2726 limit = de_buf + i_size_read(dir);
2728 while (de_buf < limit) {
2729 de = (struct ocfs2_dir_entry *)de_buf;
2731 if (de->name_len && de->inode)
2734 de_buf += le16_to_cpu(de->rec_len);
2737 /* We are careful to leave room for one extra record. */
2738 return dirent_count < ocfs2_dx_entries_per_root(dir->i_sb);
2742 * Expand rec_len of the rightmost dirent in a directory block so that it
2743 * contains the end of our valid space for dirents. We do this during
2744 * expansion from an inline directory to one with extents. The first dir block
2745 * in that case is taken from the inline data portion of the inode block.
2747 * This will also return the largest amount of contiguous space for a dirent
2748 * in the block. That value is *not* necessarily the last dirent, even after
2749 * expansion. The directory indexing code wants this value for free space
2750 * accounting. We do this here since we're already walking the entire dir
2753 * We add the dir trailer if this filesystem wants it.
2755 static unsigned int ocfs2_expand_last_dirent(char *start, unsigned int old_size,
2758 struct super_block *sb = dir->i_sb;
2759 struct ocfs2_dir_entry *de;
2760 struct ocfs2_dir_entry *prev_de;
2761 char *de_buf, *limit;
2762 unsigned int new_size = sb->s_blocksize;
2763 unsigned int bytes, this_hole;
2764 unsigned int largest_hole = 0;
2766 if (ocfs2_new_dir_wants_trailer(dir))
2767 new_size = ocfs2_dir_trailer_blk_off(sb);
2769 bytes = new_size - old_size;
2771 limit = start + old_size;
2773 de = (struct ocfs2_dir_entry *)de_buf;
2775 this_hole = ocfs2_figure_dirent_hole(de);
2776 if (this_hole > largest_hole)
2777 largest_hole = this_hole;
2780 de_buf += le16_to_cpu(de->rec_len);
2781 de = (struct ocfs2_dir_entry *)de_buf;
2782 } while (de_buf < limit);
2784 le16_add_cpu(&prev_de->rec_len, bytes);
2786 /* We need to double check this after modification of the final
2788 this_hole = ocfs2_figure_dirent_hole(prev_de);
2789 if (this_hole > largest_hole)
2790 largest_hole = this_hole;
2792 if (largest_hole >= OCFS2_DIR_MIN_REC_LEN)
2793 return largest_hole;
2798 * We allocate enough clusters to fulfill "blocks_wanted", but set
2799 * i_size to exactly one block. Ocfs2_extend_dir() will handle the
2800 * rest automatically for us.
2802 * *first_block_bh is a pointer to the 1st data block allocated to the
2805 static int ocfs2_expand_inline_dir(struct inode *dir, struct buffer_head *di_bh,
2806 unsigned int blocks_wanted,
2807 struct ocfs2_dir_lookup_result *lookup,
2808 struct buffer_head **first_block_bh)
2810 u32 alloc, dx_alloc, bit_off, len, num_dx_entries = 0;
2811 struct super_block *sb = dir->i_sb;
2812 int ret, i, num_dx_leaves = 0, dx_inline = 0,
2813 credits = ocfs2_inline_to_extents_credits(sb);
2814 u64 dx_insert_blkno, blkno,
2815 bytes = blocks_wanted << sb->s_blocksize_bits;
2816 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2817 struct ocfs2_inode_info *oi = OCFS2_I(dir);
2818 struct ocfs2_alloc_context *data_ac = NULL;
2819 struct ocfs2_alloc_context *meta_ac = NULL;
2820 struct buffer_head *dirdata_bh = NULL;
2821 struct buffer_head *dx_root_bh = NULL;
2822 struct buffer_head **dx_leaves = NULL;
2823 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2825 struct ocfs2_extent_tree et;
2826 struct ocfs2_extent_tree dx_et;
2827 int did_quota = 0, bytes_allocated = 0;
2829 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(dir), di_bh);
2831 alloc = ocfs2_clusters_for_bytes(sb, bytes);
2834 down_write(&oi->ip_alloc_sem);
2836 if (ocfs2_supports_indexed_dirs(osb)) {
2837 credits += ocfs2_add_dir_index_credits(sb);
2839 dx_inline = ocfs2_new_dx_should_be_inline(dir, di_bh);
2841 /* Add one more cluster for an index leaf */
2843 dx_leaves = ocfs2_dx_dir_kmalloc_leaves(sb,
2852 /* This gets us the dx_root */
2853 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
2861 * We should never need more than 2 clusters for the unindexed
2862 * tree - maximum dirent size is far less than one block. In
2863 * fact, the only time we'd need more than one cluster is if
2864 * blocksize == clustersize and the dirent won't fit in the
2865 * extra space that the expansion to a single block gives. As
2866 * of today, that only happens on 4k/4k file systems.
2870 ret = ocfs2_reserve_clusters(osb, alloc + dx_alloc, &data_ac);
2877 * Prepare for worst case allocation scenario of two separate
2878 * extents in the unindexed tree.
2881 credits += OCFS2_SUBALLOC_ALLOC;
2883 handle = ocfs2_start_trans(osb, credits);
2884 if (IS_ERR(handle)) {
2885 ret = PTR_ERR(handle);
2890 ret = dquot_alloc_space_nodirty(dir,
2891 ocfs2_clusters_to_bytes(osb->sb, alloc + dx_alloc));
2896 if (ocfs2_supports_indexed_dirs(osb) && !dx_inline) {
2898 * Allocate our index cluster first, to maximize the
2899 * possibility that unindexed leaves grow
2902 ret = __ocfs2_dx_dir_new_cluster(dir, 0, handle, data_ac,
2903 dx_leaves, num_dx_leaves,
2909 bytes_allocated += ocfs2_clusters_to_bytes(dir->i_sb, 1);
2913 * Try to claim as many clusters as the bitmap can give though
2914 * if we only get one now, that's enough to continue. The rest
2915 * will be claimed after the conversion to extents.
2917 if (ocfs2_dir_resv_allowed(osb))
2918 data_ac->ac_resv = &oi->ip_la_data_resv;
2919 ret = ocfs2_claim_clusters(handle, data_ac, 1, &bit_off, &len);
2924 bytes_allocated += ocfs2_clusters_to_bytes(dir->i_sb, 1);
2927 * Operations are carefully ordered so that we set up the new
2928 * data block first. The conversion from inline data to
2931 blkno = ocfs2_clusters_to_blocks(dir->i_sb, bit_off);
2932 dirdata_bh = sb_getblk(sb, blkno);
2939 ocfs2_set_new_buffer_uptodate(INODE_CACHE(dir), dirdata_bh);
2941 ret = ocfs2_journal_access_db(handle, INODE_CACHE(dir), dirdata_bh,
2942 OCFS2_JOURNAL_ACCESS_CREATE);
2948 memcpy(dirdata_bh->b_data, di->id2.i_data.id_data, i_size_read(dir));
2949 memset(dirdata_bh->b_data + i_size_read(dir), 0,
2950 sb->s_blocksize - i_size_read(dir));
2951 i = ocfs2_expand_last_dirent(dirdata_bh->b_data, i_size_read(dir), dir);
2952 if (ocfs2_new_dir_wants_trailer(dir)) {
2954 * Prepare the dir trailer up front. It will otherwise look
2955 * like a valid dirent. Even if inserting the index fails
2956 * (unlikely), then all we'll have done is given first dir
2957 * block a small amount of fragmentation.
2959 ocfs2_init_dir_trailer(dir, dirdata_bh, i);
2962 ocfs2_update_inode_fsync_trans(handle, dir, 1);
2963 ocfs2_journal_dirty(handle, dirdata_bh);
2965 if (ocfs2_supports_indexed_dirs(osb) && !dx_inline) {
2967 * Dx dirs with an external cluster need to do this up
2968 * front. Inline dx root's get handled later, after
2969 * we've allocated our root block. We get passed back
2970 * a total number of items so that dr_num_entries can
2971 * be correctly set once the dx_root has been
2974 ret = ocfs2_dx_dir_index_block(dir, handle, dx_leaves,
2975 num_dx_leaves, &num_dx_entries,
2984 * Set extent, i_size, etc on the directory. After this, the
2985 * inode should contain the same exact dirents as before and
2986 * be fully accessible from system calls.
2988 * We let the later dirent insert modify c/mtime - to the user
2989 * the data hasn't changed.
2991 ret = ocfs2_journal_access_di(handle, INODE_CACHE(dir), di_bh,
2992 OCFS2_JOURNAL_ACCESS_CREATE);
2998 spin_lock(&oi->ip_lock);
2999 oi->ip_dyn_features &= ~OCFS2_INLINE_DATA_FL;
3000 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
3001 spin_unlock(&oi->ip_lock);
3003 ocfs2_dinode_new_extent_list(dir, di);
3005 i_size_write(dir, sb->s_blocksize);
3006 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3008 di->i_size = cpu_to_le64(sb->s_blocksize);
3009 di->i_ctime = di->i_mtime = cpu_to_le64(dir->i_ctime.tv_sec);
3010 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(dir->i_ctime.tv_nsec);
3011 ocfs2_update_inode_fsync_trans(handle, dir, 1);
3014 * This should never fail as our extent list is empty and all
3015 * related blocks have been journaled already.
3017 ret = ocfs2_insert_extent(handle, &et, 0, blkno, len,
3025 * Set i_blocks after the extent insert for the most up to
3026 * date ip_clusters value.
3028 dir->i_blocks = ocfs2_inode_sector_count(dir);
3030 ocfs2_journal_dirty(handle, di_bh);
3032 if (ocfs2_supports_indexed_dirs(osb)) {
3033 ret = ocfs2_dx_dir_attach_index(osb, handle, dir, di_bh,
3034 dirdata_bh, meta_ac, dx_inline,
3035 num_dx_entries, &dx_root_bh);
3042 ocfs2_dx_dir_index_root_block(dir, dx_root_bh,
3045 ocfs2_init_dx_root_extent_tree(&dx_et,
3048 ret = ocfs2_insert_extent(handle, &dx_et, 0,
3049 dx_insert_blkno, 1, 0, NULL);
3056 * We asked for two clusters, but only got one in the 1st
3057 * pass. Claim the 2nd cluster as a separate extent.
3060 ret = ocfs2_claim_clusters(handle, data_ac, 1, &bit_off,
3066 blkno = ocfs2_clusters_to_blocks(dir->i_sb, bit_off);
3068 ret = ocfs2_insert_extent(handle, &et, 1,
3069 blkno, len, 0, NULL);
3074 bytes_allocated += ocfs2_clusters_to_bytes(dir->i_sb, 1);
3077 *first_block_bh = dirdata_bh;
3079 if (ocfs2_supports_indexed_dirs(osb)) {
3084 * We need to return the correct block within the
3085 * cluster which should hold our entry.
3087 off = ocfs2_dx_dir_hash_idx(OCFS2_SB(dir->i_sb),
3089 get_bh(dx_leaves[off]);
3090 lookup->dl_dx_leaf_bh = dx_leaves[off];
3092 lookup->dl_dx_root_bh = dx_root_bh;
3097 if (ret < 0 && did_quota)
3098 dquot_free_space_nodirty(dir, bytes_allocated);
3100 ocfs2_commit_trans(osb, handle);
3103 up_write(&oi->ip_alloc_sem);
3105 ocfs2_free_alloc_context(data_ac);
3107 ocfs2_free_alloc_context(meta_ac);
3110 for (i = 0; i < num_dx_leaves; i++)
3111 brelse(dx_leaves[i]);
3121 /* returns a bh of the 1st new block in the allocation. */
3122 static int ocfs2_do_extend_dir(struct super_block *sb,
3125 struct buffer_head *parent_fe_bh,
3126 struct ocfs2_alloc_context *data_ac,
3127 struct ocfs2_alloc_context *meta_ac,
3128 struct buffer_head **new_bh)
3131 int extend, did_quota = 0;
3132 u64 p_blkno, v_blkno;
3134 spin_lock(&OCFS2_I(dir)->ip_lock);
3135 extend = (i_size_read(dir) == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters));
3136 spin_unlock(&OCFS2_I(dir)->ip_lock);
3139 u32 offset = OCFS2_I(dir)->ip_clusters;
3141 status = dquot_alloc_space_nodirty(dir,
3142 ocfs2_clusters_to_bytes(sb, 1));
3147 status = ocfs2_add_inode_data(OCFS2_SB(sb), dir, &offset,
3148 1, 0, parent_fe_bh, handle,
3149 data_ac, meta_ac, NULL);
3150 BUG_ON(status == -EAGAIN);
3157 v_blkno = ocfs2_blocks_for_bytes(sb, i_size_read(dir));
3158 status = ocfs2_extent_map_get_blocks(dir, v_blkno, &p_blkno, NULL, NULL);
3164 *new_bh = sb_getblk(sb, p_blkno);
3172 if (did_quota && status < 0)
3173 dquot_free_space_nodirty(dir, ocfs2_clusters_to_bytes(sb, 1));
3178 * Assumes you already have a cluster lock on the directory.
3180 * 'blocks_wanted' is only used if we have an inline directory which
3181 * is to be turned into an extent based one. The size of the dirent to
3182 * insert might be larger than the space gained by growing to just one
3183 * block, so we may have to grow the inode by two blocks in that case.
3185 * If the directory is already indexed, dx_root_bh must be provided.
3187 static int ocfs2_extend_dir(struct ocfs2_super *osb,
3189 struct buffer_head *parent_fe_bh,
3190 unsigned int blocks_wanted,
3191 struct ocfs2_dir_lookup_result *lookup,
3192 struct buffer_head **new_de_bh)
3195 int credits, num_free_extents, drop_alloc_sem = 0;
3197 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
3198 struct ocfs2_extent_list *el = &fe->id2.i_list;
3199 struct ocfs2_alloc_context *data_ac = NULL;
3200 struct ocfs2_alloc_context *meta_ac = NULL;
3201 handle_t *handle = NULL;
3202 struct buffer_head *new_bh = NULL;
3203 struct ocfs2_dir_entry * de;
3204 struct super_block *sb = osb->sb;
3205 struct ocfs2_extent_tree et;
3206 struct buffer_head *dx_root_bh = lookup->dl_dx_root_bh;
3208 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
3210 * This would be a code error as an inline directory should
3211 * never have an index root.
3215 status = ocfs2_expand_inline_dir(dir, parent_fe_bh,
3216 blocks_wanted, lookup,
3223 /* Expansion from inline to an indexed directory will
3224 * have given us this. */
3225 dx_root_bh = lookup->dl_dx_root_bh;
3227 if (blocks_wanted == 1) {
3229 * If the new dirent will fit inside the space
3230 * created by pushing out to one block, then
3231 * we can complete the operation
3232 * here. Otherwise we have to expand i_size
3233 * and format the 2nd block below.
3235 BUG_ON(new_bh == NULL);
3240 * Get rid of 'new_bh' - we want to format the 2nd
3241 * data block and return that instead.
3246 down_write(&OCFS2_I(dir)->ip_alloc_sem);
3248 dir_i_size = i_size_read(dir);
3249 credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS;
3253 down_write(&OCFS2_I(dir)->ip_alloc_sem);
3255 dir_i_size = i_size_read(dir);
3256 trace_ocfs2_extend_dir((unsigned long long)OCFS2_I(dir)->ip_blkno,
3259 /* dir->i_size is always block aligned. */
3260 spin_lock(&OCFS2_I(dir)->ip_lock);
3261 if (dir_i_size == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters)) {
3262 spin_unlock(&OCFS2_I(dir)->ip_lock);
3263 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(dir),
3265 num_free_extents = ocfs2_num_free_extents(osb, &et);
3266 if (num_free_extents < 0) {
3267 status = num_free_extents;
3272 if (!num_free_extents) {
3273 status = ocfs2_reserve_new_metadata(osb, el, &meta_ac);
3275 if (status != -ENOSPC)
3281 status = ocfs2_reserve_clusters(osb, 1, &data_ac);
3283 if (status != -ENOSPC)
3288 if (ocfs2_dir_resv_allowed(osb))
3289 data_ac->ac_resv = &OCFS2_I(dir)->ip_la_data_resv;
3291 credits = ocfs2_calc_extend_credits(sb, el);
3293 spin_unlock(&OCFS2_I(dir)->ip_lock);
3294 credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS;
3298 if (ocfs2_dir_indexed(dir))
3299 credits++; /* For attaching the new dirent block to the
3302 handle = ocfs2_start_trans(osb, credits);
3303 if (IS_ERR(handle)) {
3304 status = PTR_ERR(handle);
3310 status = ocfs2_do_extend_dir(osb->sb, handle, dir, parent_fe_bh,
3311 data_ac, meta_ac, &new_bh);
3317 ocfs2_set_new_buffer_uptodate(INODE_CACHE(dir), new_bh);
3319 status = ocfs2_journal_access_db(handle, INODE_CACHE(dir), new_bh,
3320 OCFS2_JOURNAL_ACCESS_CREATE);
3325 memset(new_bh->b_data, 0, sb->s_blocksize);
3327 de = (struct ocfs2_dir_entry *) new_bh->b_data;
3329 if (ocfs2_supports_dir_trailer(dir)) {
3330 de->rec_len = cpu_to_le16(ocfs2_dir_trailer_blk_off(sb));
3332 ocfs2_init_dir_trailer(dir, new_bh, le16_to_cpu(de->rec_len));
3334 if (ocfs2_dir_indexed(dir)) {
3335 status = ocfs2_dx_dir_link_trailer(dir, handle,
3336 dx_root_bh, new_bh);
3343 de->rec_len = cpu_to_le16(sb->s_blocksize);
3345 ocfs2_update_inode_fsync_trans(handle, dir, 1);
3346 ocfs2_journal_dirty(handle, new_bh);
3348 dir_i_size += dir->i_sb->s_blocksize;
3349 i_size_write(dir, dir_i_size);
3350 dir->i_blocks = ocfs2_inode_sector_count(dir);
3351 status = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
3358 *new_de_bh = new_bh;
3362 ocfs2_commit_trans(osb, handle);
3364 up_write(&OCFS2_I(dir)->ip_alloc_sem);
3367 ocfs2_free_alloc_context(data_ac);
3369 ocfs2_free_alloc_context(meta_ac);
3376 static int ocfs2_find_dir_space_id(struct inode *dir, struct buffer_head *di_bh,
3377 const char *name, int namelen,
3378 struct buffer_head **ret_de_bh,
3379 unsigned int *blocks_wanted)
3382 struct super_block *sb = dir->i_sb;
3383 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3384 struct ocfs2_dir_entry *de, *last_de = NULL;
3385 char *de_buf, *limit;
3386 unsigned long offset = 0;
3387 unsigned int rec_len, new_rec_len, free_space = dir->i_sb->s_blocksize;
3390 * This calculates how many free bytes we'd have in block zero, should
3391 * this function force expansion to an extent tree.
3393 if (ocfs2_new_dir_wants_trailer(dir))
3394 free_space = ocfs2_dir_trailer_blk_off(sb) - i_size_read(dir);
3396 free_space = dir->i_sb->s_blocksize - i_size_read(dir);
3398 de_buf = di->id2.i_data.id_data;
3399 limit = de_buf + i_size_read(dir);
3400 rec_len = OCFS2_DIR_REC_LEN(namelen);
3402 while (de_buf < limit) {
3403 de = (struct ocfs2_dir_entry *)de_buf;
3405 if (!ocfs2_check_dir_entry(dir, de, di_bh, offset)) {
3409 if (ocfs2_match(namelen, name, de)) {
3414 * No need to check for a trailing dirent record here as
3415 * they're not used for inline dirs.
3418 if (ocfs2_dirent_would_fit(de, rec_len)) {
3419 /* Ok, we found a spot. Return this bh and let
3420 * the caller actually fill it in. */
3428 de_buf += le16_to_cpu(de->rec_len);
3429 offset += le16_to_cpu(de->rec_len);
3433 * We're going to require expansion of the directory - figure
3434 * out how many blocks we'll need so that a place for the
3435 * dirent can be found.
3438 new_rec_len = le16_to_cpu(last_de->rec_len) + free_space;
3439 if (new_rec_len < (rec_len + OCFS2_DIR_REC_LEN(last_de->name_len)))
3447 static int ocfs2_find_dir_space_el(struct inode *dir, const char *name,
3448 int namelen, struct buffer_head **ret_de_bh)
3450 unsigned long offset;
3451 struct buffer_head *bh = NULL;
3452 unsigned short rec_len;
3453 struct ocfs2_dir_entry *de;
3454 struct super_block *sb = dir->i_sb;
3456 int blocksize = dir->i_sb->s_blocksize;
3458 status = ocfs2_read_dir_block(dir, 0, &bh, 0);
3464 rec_len = OCFS2_DIR_REC_LEN(namelen);
3466 de = (struct ocfs2_dir_entry *) bh->b_data;
3468 if ((char *)de >= sb->s_blocksize + bh->b_data) {
3472 if (i_size_read(dir) <= offset) {
3474 * Caller will have to expand this
3480 status = ocfs2_read_dir_block(dir,
3481 offset >> sb->s_blocksize_bits,
3487 /* move to next block */
3488 de = (struct ocfs2_dir_entry *) bh->b_data;
3490 if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
3494 if (ocfs2_match(namelen, name, de)) {
3499 if (ocfs2_skip_dir_trailer(dir, de, offset % blocksize,
3503 if (ocfs2_dirent_would_fit(de, rec_len)) {
3504 /* Ok, we found a spot. Return this bh and let
3505 * the caller actually fill it in. */
3512 offset += le16_to_cpu(de->rec_len);
3513 de = (struct ocfs2_dir_entry *)((char *) de + le16_to_cpu(de->rec_len));
3524 static int dx_leaf_sort_cmp(const void *a, const void *b)
3526 const struct ocfs2_dx_entry *entry1 = a;
3527 const struct ocfs2_dx_entry *entry2 = b;
3528 u32 major_hash1 = le32_to_cpu(entry1->dx_major_hash);
3529 u32 major_hash2 = le32_to_cpu(entry2->dx_major_hash);
3530 u32 minor_hash1 = le32_to_cpu(entry1->dx_minor_hash);
3531 u32 minor_hash2 = le32_to_cpu(entry2->dx_minor_hash);
3533 if (major_hash1 > major_hash2)
3535 if (major_hash1 < major_hash2)
3539 * It is not strictly necessary to sort by minor
3541 if (minor_hash1 > minor_hash2)
3543 if (minor_hash1 < minor_hash2)
3548 static void dx_leaf_sort_swap(void *a, void *b, int size)
3550 struct ocfs2_dx_entry *entry1 = a;
3551 struct ocfs2_dx_entry *entry2 = b;
3552 struct ocfs2_dx_entry tmp;
3554 BUG_ON(size != sizeof(*entry1));
3561 static int ocfs2_dx_leaf_same_major(struct ocfs2_dx_leaf *dx_leaf)
3563 struct ocfs2_dx_entry_list *dl_list = &dx_leaf->dl_list;
3564 int i, num = le16_to_cpu(dl_list->de_num_used);
3566 for (i = 0; i < (num - 1); i++) {
3567 if (le32_to_cpu(dl_list->de_entries[i].dx_major_hash) !=
3568 le32_to_cpu(dl_list->de_entries[i + 1].dx_major_hash))
3576 * Find the optimal value to split this leaf on. This expects the leaf
3577 * entries to be in sorted order.
3579 * leaf_cpos is the cpos of the leaf we're splitting. insert_hash is
3580 * the hash we want to insert.
3582 * This function is only concerned with the major hash - that which
3583 * determines which cluster an item belongs to.
3585 static int ocfs2_dx_dir_find_leaf_split(struct ocfs2_dx_leaf *dx_leaf,
3586 u32 leaf_cpos, u32 insert_hash,
3589 struct ocfs2_dx_entry_list *dl_list = &dx_leaf->dl_list;
3590 int i, num_used = le16_to_cpu(dl_list->de_num_used);
3594 * There's a couple rare, but nasty corner cases we have to
3595 * check for here. All of them involve a leaf where all value
3596 * have the same hash, which is what we look for first.
3598 * Most of the time, all of the above is false, and we simply
3599 * pick the median value for a split.
3601 allsame = ocfs2_dx_leaf_same_major(dx_leaf);
3603 u32 val = le32_to_cpu(dl_list->de_entries[0].dx_major_hash);
3605 if (val == insert_hash) {
3607 * No matter where we would choose to split,
3608 * the new entry would want to occupy the same
3609 * block as these. Since there's no space left
3610 * in their existing block, we know there
3611 * won't be space after the split.
3616 if (val == leaf_cpos) {
3618 * Because val is the same as leaf_cpos (which
3619 * is the smallest value this leaf can have),
3620 * yet is not equal to insert_hash, then we
3621 * know that insert_hash *must* be larger than
3622 * val (and leaf_cpos). At least cpos+1 in value.
3624 * We also know then, that there cannot be an
3625 * adjacent extent (otherwise we'd be looking
3626 * at it). Choosing this value gives us a
3627 * chance to get some contiguousness.
3629 *split_hash = leaf_cpos + 1;
3633 if (val > insert_hash) {
3635 * val can not be the same as insert hash, and
3636 * also must be larger than leaf_cpos. Also,
3637 * we know that there can't be a leaf between
3638 * cpos and val, otherwise the entries with
3639 * hash 'val' would be there.
3645 *split_hash = insert_hash;
3650 * Since the records are sorted and the checks above
3651 * guaranteed that not all records in this block are the same,
3652 * we simple travel forward, from the median, and pick the 1st
3653 * record whose value is larger than leaf_cpos.
3655 for (i = (num_used / 2); i < num_used; i++)
3656 if (le32_to_cpu(dl_list->de_entries[i].dx_major_hash) >
3660 BUG_ON(i == num_used); /* Should be impossible */
3661 *split_hash = le32_to_cpu(dl_list->de_entries[i].dx_major_hash);
3666 * Transfer all entries in orig_dx_leaves whose major hash is equal to or
3667 * larger than split_hash into new_dx_leaves. We use a temporary
3668 * buffer (tmp_dx_leaf) to make the changes to the original leaf blocks.
3670 * Since the block offset inside a leaf (cluster) is a constant mask
3671 * of minor_hash, we can optimize - an item at block offset X within
3672 * the original cluster, will be at offset X within the new cluster.
3674 static void ocfs2_dx_dir_transfer_leaf(struct inode *dir, u32 split_hash,
3676 struct ocfs2_dx_leaf *tmp_dx_leaf,
3677 struct buffer_head **orig_dx_leaves,
3678 struct buffer_head **new_dx_leaves,
3683 struct ocfs2_dx_leaf *orig_dx_leaf, *new_dx_leaf;
3684 struct ocfs2_dx_entry_list *orig_list, *new_list, *tmp_list;
3685 struct ocfs2_dx_entry *dx_entry;
3687 tmp_list = &tmp_dx_leaf->dl_list;
3689 for (i = 0; i < num_dx_leaves; i++) {
3690 orig_dx_leaf = (struct ocfs2_dx_leaf *) orig_dx_leaves[i]->b_data;
3691 orig_list = &orig_dx_leaf->dl_list;
3692 new_dx_leaf = (struct ocfs2_dx_leaf *) new_dx_leaves[i]->b_data;
3693 new_list = &new_dx_leaf->dl_list;
3695 num_used = le16_to_cpu(orig_list->de_num_used);
3697 memcpy(tmp_dx_leaf, orig_dx_leaf, dir->i_sb->s_blocksize);
3698 tmp_list->de_num_used = cpu_to_le16(0);
3699 memset(&tmp_list->de_entries, 0, sizeof(*dx_entry)*num_used);
3701 for (j = 0; j < num_used; j++) {
3702 dx_entry = &orig_list->de_entries[j];
3703 major_hash = le32_to_cpu(dx_entry->dx_major_hash);
3704 if (major_hash >= split_hash)
3705 ocfs2_dx_dir_leaf_insert_tail(new_dx_leaf,
3708 ocfs2_dx_dir_leaf_insert_tail(tmp_dx_leaf,
3711 memcpy(orig_dx_leaf, tmp_dx_leaf, dir->i_sb->s_blocksize);
3713 ocfs2_journal_dirty(handle, orig_dx_leaves[i]);
3714 ocfs2_journal_dirty(handle, new_dx_leaves[i]);
3718 static int ocfs2_dx_dir_rebalance_credits(struct ocfs2_super *osb,
3719 struct ocfs2_dx_root_block *dx_root)
3721 int credits = ocfs2_clusters_to_blocks(osb->sb, 2);
3723 credits += ocfs2_calc_extend_credits(osb->sb, &dx_root->dr_list);
3724 credits += ocfs2_quota_trans_credits(osb->sb);
3729 * Find the median value in dx_leaf_bh and allocate a new leaf to move
3730 * half our entries into.
3732 static int ocfs2_dx_dir_rebalance(struct ocfs2_super *osb, struct inode *dir,
3733 struct buffer_head *dx_root_bh,
3734 struct buffer_head *dx_leaf_bh,
3735 struct ocfs2_dx_hinfo *hinfo, u32 leaf_cpos,
3738 struct ocfs2_dx_leaf *dx_leaf = (struct ocfs2_dx_leaf *)dx_leaf_bh->b_data;
3739 int credits, ret, i, num_used, did_quota = 0;
3740 u32 cpos, split_hash, insert_hash = hinfo->major_hash;
3741 u64 orig_leaves_start;
3743 struct buffer_head **orig_dx_leaves = NULL;
3744 struct buffer_head **new_dx_leaves = NULL;
3745 struct ocfs2_alloc_context *data_ac = NULL, *meta_ac = NULL;
3746 struct ocfs2_extent_tree et;
3747 handle_t *handle = NULL;
3748 struct ocfs2_dx_root_block *dx_root;
3749 struct ocfs2_dx_leaf *tmp_dx_leaf = NULL;
3751 trace_ocfs2_dx_dir_rebalance((unsigned long long)OCFS2_I(dir)->ip_blkno,
3752 (unsigned long long)leaf_blkno,
3755 ocfs2_init_dx_root_extent_tree(&et, INODE_CACHE(dir), dx_root_bh);
3757 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
3759 * XXX: This is a rather large limit. We should use a more
3762 if (le32_to_cpu(dx_root->dr_clusters) == UINT_MAX)
3765 num_used = le16_to_cpu(dx_leaf->dl_list.de_num_used);
3766 if (num_used < le16_to_cpu(dx_leaf->dl_list.de_count)) {
3767 mlog(ML_ERROR, "DX Dir: %llu, Asked to rebalance empty leaf: "
3768 "%llu, %d\n", (unsigned long long)OCFS2_I(dir)->ip_blkno,
3769 (unsigned long long)leaf_blkno, num_used);
3774 orig_dx_leaves = ocfs2_dx_dir_kmalloc_leaves(osb->sb, &num_dx_leaves);
3775 if (!orig_dx_leaves) {
3781 new_dx_leaves = ocfs2_dx_dir_kmalloc_leaves(osb->sb, NULL);
3782 if (!new_dx_leaves) {
3788 ret = ocfs2_lock_allocators(dir, &et, 1, 0, &data_ac, &meta_ac);
3795 credits = ocfs2_dx_dir_rebalance_credits(osb, dx_root);
3796 handle = ocfs2_start_trans(osb, credits);
3797 if (IS_ERR(handle)) {
3798 ret = PTR_ERR(handle);
3804 ret = dquot_alloc_space_nodirty(dir,
3805 ocfs2_clusters_to_bytes(dir->i_sb, 1));
3810 ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir), dx_leaf_bh,
3811 OCFS2_JOURNAL_ACCESS_WRITE);
3818 * This block is changing anyway, so we can sort it in place.
3820 sort(dx_leaf->dl_list.de_entries, num_used,
3821 sizeof(struct ocfs2_dx_entry), dx_leaf_sort_cmp,
3824 ocfs2_journal_dirty(handle, dx_leaf_bh);
3826 ret = ocfs2_dx_dir_find_leaf_split(dx_leaf, leaf_cpos, insert_hash,
3833 trace_ocfs2_dx_dir_rebalance_split(leaf_cpos, split_hash, insert_hash);
3836 * We have to carefully order operations here. There are items
3837 * which want to be in the new cluster before insert, but in
3838 * order to put those items in the new cluster, we alter the
3839 * old cluster. A failure to insert gets nasty.
3841 * So, start by reserving writes to the old
3842 * cluster. ocfs2_dx_dir_new_cluster will reserve writes on
3843 * the new cluster for us, before inserting it. The insert
3844 * won't happen if there's an error before that. Once the
3845 * insert is done then, we can transfer from one leaf into the
3846 * other without fear of hitting any error.
3850 * The leaf transfer wants some scratch space so that we don't
3851 * wind up doing a bunch of expensive memmove().
3853 tmp_dx_leaf = kmalloc(osb->sb->s_blocksize, GFP_NOFS);
3860 orig_leaves_start = ocfs2_block_to_cluster_start(dir->i_sb, leaf_blkno);
3861 ret = ocfs2_read_dx_leaves(dir, orig_leaves_start, num_dx_leaves,
3869 ret = ocfs2_dx_dir_new_cluster(dir, &et, cpos, handle,
3870 data_ac, meta_ac, new_dx_leaves,
3877 for (i = 0; i < num_dx_leaves; i++) {
3878 ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir),
3880 OCFS2_JOURNAL_ACCESS_WRITE);
3886 ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir),
3888 OCFS2_JOURNAL_ACCESS_WRITE);
3895 ocfs2_dx_dir_transfer_leaf(dir, split_hash, handle, tmp_dx_leaf,
3896 orig_dx_leaves, new_dx_leaves, num_dx_leaves);
3899 if (ret < 0 && did_quota)
3900 dquot_free_space_nodirty(dir,
3901 ocfs2_clusters_to_bytes(dir->i_sb, 1));
3903 ocfs2_update_inode_fsync_trans(handle, dir, 1);
3904 ocfs2_commit_trans(osb, handle);
3907 if (orig_dx_leaves || new_dx_leaves) {
3908 for (i = 0; i < num_dx_leaves; i++) {
3910 brelse(orig_dx_leaves[i]);
3912 brelse(new_dx_leaves[i]);
3914 kfree(orig_dx_leaves);
3915 kfree(new_dx_leaves);
3919 ocfs2_free_alloc_context(meta_ac);
3921 ocfs2_free_alloc_context(data_ac);
3927 static int ocfs2_find_dir_space_dx(struct ocfs2_super *osb, struct inode *dir,
3928 struct buffer_head *di_bh,
3929 struct buffer_head *dx_root_bh,
3930 const char *name, int namelen,
3931 struct ocfs2_dir_lookup_result *lookup)
3933 int ret, rebalanced = 0;
3934 struct ocfs2_dx_root_block *dx_root;
3935 struct buffer_head *dx_leaf_bh = NULL;
3936 struct ocfs2_dx_leaf *dx_leaf;
3940 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
3943 ret = ocfs2_dx_dir_lookup(dir, &dx_root->dr_list, &lookup->dl_hinfo,
3944 &leaf_cpos, &blkno);
3950 ret = ocfs2_read_dx_leaf(dir, blkno, &dx_leaf_bh);
3956 dx_leaf = (struct ocfs2_dx_leaf *)dx_leaf_bh->b_data;
3958 if (le16_to_cpu(dx_leaf->dl_list.de_num_used) >=
3959 le16_to_cpu(dx_leaf->dl_list.de_count)) {
3962 * Rebalancing should have provided us with
3963 * space in an appropriate leaf.
3965 * XXX: Is this an abnormal condition then?
3966 * Should we print a message here?
3972 ret = ocfs2_dx_dir_rebalance(osb, dir, dx_root_bh, dx_leaf_bh,
3973 &lookup->dl_hinfo, leaf_cpos,
3982 * Restart the lookup. The rebalance might have
3983 * changed which block our item fits into. Mark our
3984 * progress, so we only execute this once.
3989 goto restart_search;
3992 lookup->dl_dx_leaf_bh = dx_leaf_bh;
4000 static int ocfs2_search_dx_free_list(struct inode *dir,
4001 struct buffer_head *dx_root_bh,
4003 struct ocfs2_dir_lookup_result *lookup)
4006 struct buffer_head *leaf_bh = NULL, *prev_leaf_bh = NULL;
4007 struct ocfs2_dir_block_trailer *db;
4009 int rec_len = OCFS2_DIR_REC_LEN(namelen);
4010 struct ocfs2_dx_root_block *dx_root;
4012 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
4013 next_block = le64_to_cpu(dx_root->dr_free_blk);
4015 while (next_block) {
4016 brelse(prev_leaf_bh);
4017 prev_leaf_bh = leaf_bh;
4020 ret = ocfs2_read_dir_block_direct(dir, next_block, &leaf_bh);
4026 db = ocfs2_trailer_from_bh(leaf_bh, dir->i_sb);
4027 if (rec_len <= le16_to_cpu(db->db_free_rec_len)) {
4028 lookup->dl_leaf_bh = leaf_bh;
4029 lookup->dl_prev_leaf_bh = prev_leaf_bh;
4031 prev_leaf_bh = NULL;
4035 next_block = le64_to_cpu(db->db_free_next);
4044 brelse(prev_leaf_bh);
4048 static int ocfs2_expand_inline_dx_root(struct inode *dir,
4049 struct buffer_head *dx_root_bh)
4051 int ret, num_dx_leaves, i, j, did_quota = 0;
4052 struct buffer_head **dx_leaves = NULL;
4053 struct ocfs2_extent_tree et;
4055 struct ocfs2_alloc_context *data_ac = NULL;
4056 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
4057 handle_t *handle = NULL;
4058 struct ocfs2_dx_root_block *dx_root;
4059 struct ocfs2_dx_entry_list *entry_list;
4060 struct ocfs2_dx_entry *dx_entry;
4061 struct ocfs2_dx_leaf *target_leaf;
4063 ret = ocfs2_reserve_clusters(osb, 1, &data_ac);
4069 dx_leaves = ocfs2_dx_dir_kmalloc_leaves(osb->sb, &num_dx_leaves);
4076 handle = ocfs2_start_trans(osb, ocfs2_calc_dxi_expand_credits(osb->sb));
4077 if (IS_ERR(handle)) {
4078 ret = PTR_ERR(handle);
4083 ret = dquot_alloc_space_nodirty(dir,
4084 ocfs2_clusters_to_bytes(osb->sb, 1));
4090 * We do this up front, before the allocation, so that a
4091 * failure to add the dx_root_bh to the journal won't result
4092 * us losing clusters.
4094 ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
4095 OCFS2_JOURNAL_ACCESS_WRITE);
4101 ret = __ocfs2_dx_dir_new_cluster(dir, 0, handle, data_ac, dx_leaves,
4102 num_dx_leaves, &insert_blkno);
4109 * Transfer the entries from our dx_root into the appropriate
4112 dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data;
4113 entry_list = &dx_root->dr_entries;
4115 for (i = 0; i < le16_to_cpu(entry_list->de_num_used); i++) {
4116 dx_entry = &entry_list->de_entries[i];
4118 j = __ocfs2_dx_dir_hash_idx(osb,
4119 le32_to_cpu(dx_entry->dx_minor_hash));
4120 target_leaf = (struct ocfs2_dx_leaf *)dx_leaves[j]->b_data;
4122 ocfs2_dx_dir_leaf_insert_tail(target_leaf, dx_entry);
4124 /* Each leaf has been passed to the journal already
4125 * via __ocfs2_dx_dir_new_cluster() */
4128 dx_root->dr_flags &= ~OCFS2_DX_FLAG_INLINE;
4129 memset(&dx_root->dr_list, 0, osb->sb->s_blocksize -
4130 offsetof(struct ocfs2_dx_root_block, dr_list));
4131 dx_root->dr_list.l_count =
4132 cpu_to_le16(ocfs2_extent_recs_per_dx_root(osb->sb));
4134 /* This should never fail considering we start with an empty
4136 ocfs2_init_dx_root_extent_tree(&et, INODE_CACHE(dir), dx_root_bh);
4137 ret = ocfs2_insert_extent(handle, &et, 0, insert_blkno, 1, 0, NULL);
4142 ocfs2_update_inode_fsync_trans(handle, dir, 1);
4143 ocfs2_journal_dirty(handle, dx_root_bh);
4146 if (ret < 0 && did_quota)
4147 dquot_free_space_nodirty(dir,
4148 ocfs2_clusters_to_bytes(dir->i_sb, 1));
4150 ocfs2_commit_trans(osb, handle);
4154 ocfs2_free_alloc_context(data_ac);
4157 for (i = 0; i < num_dx_leaves; i++)
4158 brelse(dx_leaves[i]);
4164 static int ocfs2_inline_dx_has_space(struct buffer_head *dx_root_bh)
4166 struct ocfs2_dx_root_block *dx_root;
4167 struct ocfs2_dx_entry_list *entry_list;
4169 dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data;
4170 entry_list = &dx_root->dr_entries;
4172 if (le16_to_cpu(entry_list->de_num_used) >=
4173 le16_to_cpu(entry_list->de_count))
4179 static int ocfs2_prepare_dx_dir_for_insert(struct inode *dir,
4180 struct buffer_head *di_bh,
4183 struct ocfs2_dir_lookup_result *lookup)
4185 int ret, free_dx_root = 1;
4186 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
4187 struct buffer_head *dx_root_bh = NULL;
4188 struct buffer_head *leaf_bh = NULL;
4189 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
4190 struct ocfs2_dx_root_block *dx_root;
4192 ret = ocfs2_read_dx_root(dir, di, &dx_root_bh);
4198 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
4199 if (le32_to_cpu(dx_root->dr_num_entries) == OCFS2_DX_ENTRIES_MAX) {
4205 if (ocfs2_dx_root_inline(dx_root)) {
4206 ret = ocfs2_inline_dx_has_space(dx_root_bh);
4212 * We ran out of room in the root block. Expand it to
4213 * an extent, then allow ocfs2_find_dir_space_dx to do
4216 ret = ocfs2_expand_inline_dx_root(dir, dx_root_bh);
4224 * Insert preparation for an indexed directory is split into two
4225 * steps. The call to find_dir_space_dx reserves room in the index for
4226 * an additional item. If we run out of space there, it's a real error
4227 * we can't continue on.
4229 ret = ocfs2_find_dir_space_dx(osb, dir, di_bh, dx_root_bh, name,
4238 * Next, we need to find space in the unindexed tree. This call
4239 * searches using the free space linked list. If the unindexed tree
4240 * lacks sufficient space, we'll expand it below. The expansion code
4241 * is smart enough to add any new blocks to the free space list.
4243 ret = ocfs2_search_dx_free_list(dir, dx_root_bh, namelen, lookup);
4244 if (ret && ret != -ENOSPC) {
4249 /* Do this up here - ocfs2_extend_dir might need the dx_root */
4250 lookup->dl_dx_root_bh = dx_root_bh;
4253 if (ret == -ENOSPC) {
4254 ret = ocfs2_extend_dir(osb, dir, di_bh, 1, lookup, &leaf_bh);
4262 * We make the assumption here that new leaf blocks are added
4263 * to the front of our free list.
4265 lookup->dl_prev_leaf_bh = NULL;
4266 lookup->dl_leaf_bh = leaf_bh;
4276 * Get a directory ready for insert. Any directory allocation required
4277 * happens here. Success returns zero, and enough context in the dir
4278 * lookup result that ocfs2_add_entry() will be able complete the task
4279 * with minimal performance impact.
4281 int ocfs2_prepare_dir_for_insert(struct ocfs2_super *osb,
4283 struct buffer_head *parent_fe_bh,
4286 struct ocfs2_dir_lookup_result *lookup)
4289 unsigned int blocks_wanted = 1;
4290 struct buffer_head *bh = NULL;
4292 trace_ocfs2_prepare_dir_for_insert(
4293 (unsigned long long)OCFS2_I(dir)->ip_blkno, namelen);
4302 * Do this up front to reduce confusion.
4304 * The directory might start inline, then be turned into an
4305 * indexed one, in which case we'd need to hash deep inside
4306 * ocfs2_find_dir_space_id(). Since
4307 * ocfs2_prepare_dx_dir_for_insert() also needs this hash
4308 * done, there seems no point in spreading out the calls. We
4309 * can optimize away the case where the file system doesn't
4312 if (ocfs2_supports_indexed_dirs(osb))
4313 ocfs2_dx_dir_name_hash(dir, name, namelen, &lookup->dl_hinfo);
4315 if (ocfs2_dir_indexed(dir)) {
4316 ret = ocfs2_prepare_dx_dir_for_insert(dir, parent_fe_bh,
4317 name, namelen, lookup);
4323 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
4324 ret = ocfs2_find_dir_space_id(dir, parent_fe_bh, name,
4325 namelen, &bh, &blocks_wanted);
4327 ret = ocfs2_find_dir_space_el(dir, name, namelen, &bh);
4329 if (ret && ret != -ENOSPC) {
4334 if (ret == -ENOSPC) {
4336 * We have to expand the directory to add this name.
4340 ret = ocfs2_extend_dir(osb, dir, parent_fe_bh, blocks_wanted,
4351 lookup->dl_leaf_bh = bh;
4358 static int ocfs2_dx_dir_remove_index(struct inode *dir,
4359 struct buffer_head *di_bh,
4360 struct buffer_head *dx_root_bh)
4363 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
4364 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
4365 struct ocfs2_dx_root_block *dx_root;
4366 struct inode *dx_alloc_inode = NULL;
4367 struct buffer_head *dx_alloc_bh = NULL;
4373 dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data;
4375 dx_alloc_inode = ocfs2_get_system_file_inode(osb,
4376 EXTENT_ALLOC_SYSTEM_INODE,
4377 le16_to_cpu(dx_root->dr_suballoc_slot));
4378 if (!dx_alloc_inode) {
4383 mutex_lock(&dx_alloc_inode->i_mutex);
4385 ret = ocfs2_inode_lock(dx_alloc_inode, &dx_alloc_bh, 1);
4391 handle = ocfs2_start_trans(osb, OCFS2_DX_ROOT_REMOVE_CREDITS);
4392 if (IS_ERR(handle)) {
4393 ret = PTR_ERR(handle);
4398 ret = ocfs2_journal_access_di(handle, INODE_CACHE(dir), di_bh,
4399 OCFS2_JOURNAL_ACCESS_WRITE);
4405 spin_lock(&OCFS2_I(dir)->ip_lock);
4406 OCFS2_I(dir)->ip_dyn_features &= ~OCFS2_INDEXED_DIR_FL;
4407 di->i_dyn_features = cpu_to_le16(OCFS2_I(dir)->ip_dyn_features);
4408 spin_unlock(&OCFS2_I(dir)->ip_lock);
4409 di->i_dx_root = cpu_to_le64(0ULL);
4410 ocfs2_update_inode_fsync_trans(handle, dir, 1);
4412 ocfs2_journal_dirty(handle, di_bh);
4414 blk = le64_to_cpu(dx_root->dr_blkno);
4415 bit = le16_to_cpu(dx_root->dr_suballoc_bit);
4416 if (dx_root->dr_suballoc_loc)
4417 bg_blkno = le64_to_cpu(dx_root->dr_suballoc_loc);
4419 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
4420 ret = ocfs2_free_suballoc_bits(handle, dx_alloc_inode, dx_alloc_bh,
4426 ocfs2_commit_trans(osb, handle);
4429 ocfs2_inode_unlock(dx_alloc_inode, 1);
4432 mutex_unlock(&dx_alloc_inode->i_mutex);
4433 brelse(dx_alloc_bh);
4435 iput(dx_alloc_inode);
4439 int ocfs2_dx_dir_truncate(struct inode *dir, struct buffer_head *di_bh)
4442 unsigned int uninitialized_var(clen);
4443 u32 major_hash = UINT_MAX, p_cpos, uninitialized_var(cpos);
4444 u64 uninitialized_var(blkno);
4445 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
4446 struct buffer_head *dx_root_bh = NULL;
4447 struct ocfs2_dx_root_block *dx_root;
4448 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
4449 struct ocfs2_cached_dealloc_ctxt dealloc;
4450 struct ocfs2_extent_tree et;
4452 ocfs2_init_dealloc_ctxt(&dealloc);
4454 if (!ocfs2_dir_indexed(dir))
4457 ret = ocfs2_read_dx_root(dir, di, &dx_root_bh);
4462 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
4464 if (ocfs2_dx_root_inline(dx_root))
4467 ocfs2_init_dx_root_extent_tree(&et, INODE_CACHE(dir), dx_root_bh);
4469 /* XXX: What if dr_clusters is too large? */
4470 while (le32_to_cpu(dx_root->dr_clusters)) {
4471 ret = ocfs2_dx_dir_lookup_rec(dir, &dx_root->dr_list,
4472 major_hash, &cpos, &blkno, &clen);
4478 p_cpos = ocfs2_blocks_to_clusters(dir->i_sb, blkno);
4480 ret = ocfs2_remove_btree_range(dir, &et, cpos, p_cpos, clen, 0,
4481 &dealloc, 0, false);
4490 major_hash = cpos - 1;
4494 ret = ocfs2_dx_dir_remove_index(dir, di_bh, dx_root_bh);
4500 ocfs2_remove_from_cache(INODE_CACHE(dir), dx_root_bh);
4502 ocfs2_schedule_truncate_log_flush(osb, 1);
4503 ocfs2_run_deallocs(osb, &dealloc);