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 #define MLOG_MASK_PREFIX ML_NAMEI
47 #include <cluster/masklog.h>
52 #include "blockcheck.h"
55 #include "extent_map.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)
70 #define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b))
72 static unsigned char ocfs2_filetype_table[] = {
73 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
76 static int ocfs2_do_extend_dir(struct super_block *sb,
79 struct buffer_head *parent_fe_bh,
80 struct ocfs2_alloc_context *data_ac,
81 struct ocfs2_alloc_context *meta_ac,
82 struct buffer_head **new_bh);
83 static int ocfs2_dir_indexed(struct inode *inode);
86 * These are distinct checks because future versions of the file system will
87 * want to have a trailing dirent structure independent of indexing.
89 static int ocfs2_supports_dir_trailer(struct inode *dir)
91 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
93 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
96 return ocfs2_meta_ecc(osb) || ocfs2_dir_indexed(dir);
100 * "new' here refers to the point at which we're creating a new
101 * directory via "mkdir()", but also when we're expanding an inline
102 * directory. In either case, we don't yet have the indexing bit set
103 * on the directory, so the standard checks will fail in when metaecc
104 * is turned off. Only directory-initialization type functions should
105 * use this then. Everything else wants ocfs2_supports_dir_trailer()
107 static int ocfs2_new_dir_wants_trailer(struct inode *dir)
109 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
111 return ocfs2_meta_ecc(osb) ||
112 ocfs2_supports_indexed_dirs(osb);
115 static inline unsigned int ocfs2_dir_trailer_blk_off(struct super_block *sb)
117 return sb->s_blocksize - sizeof(struct ocfs2_dir_block_trailer);
120 #define ocfs2_trailer_from_bh(_bh, _sb) ((struct ocfs2_dir_block_trailer *) ((_bh)->b_data + ocfs2_dir_trailer_blk_off((_sb))))
122 /* XXX ocfs2_block_dqtrailer() is similar but not quite - can we make
123 * them more consistent? */
124 struct ocfs2_dir_block_trailer *ocfs2_dir_trailer_from_size(int blocksize,
129 p += blocksize - sizeof(struct ocfs2_dir_block_trailer);
130 return (struct ocfs2_dir_block_trailer *)p;
134 * XXX: This is executed once on every dirent. We should consider optimizing
137 static int ocfs2_skip_dir_trailer(struct inode *dir,
138 struct ocfs2_dir_entry *de,
139 unsigned long offset,
140 unsigned long blklen)
142 unsigned long toff = blklen - sizeof(struct ocfs2_dir_block_trailer);
144 if (!ocfs2_supports_dir_trailer(dir))
153 static void ocfs2_init_dir_trailer(struct inode *inode,
154 struct buffer_head *bh, u16 rec_len)
156 struct ocfs2_dir_block_trailer *trailer;
158 trailer = ocfs2_trailer_from_bh(bh, inode->i_sb);
159 strcpy(trailer->db_signature, OCFS2_DIR_TRAILER_SIGNATURE);
160 trailer->db_compat_rec_len =
161 cpu_to_le16(sizeof(struct ocfs2_dir_block_trailer));
162 trailer->db_parent_dinode = cpu_to_le64(OCFS2_I(inode)->ip_blkno);
163 trailer->db_blkno = cpu_to_le64(bh->b_blocknr);
164 trailer->db_free_rec_len = cpu_to_le16(rec_len);
167 * Link an unindexed block with a dir trailer structure into the index free
168 * list. This function will modify dirdata_bh, but assumes you've already
169 * passed it to the journal.
171 static int ocfs2_dx_dir_link_trailer(struct inode *dir, handle_t *handle,
172 struct buffer_head *dx_root_bh,
173 struct buffer_head *dirdata_bh)
176 struct ocfs2_dx_root_block *dx_root;
177 struct ocfs2_dir_block_trailer *trailer;
179 ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
180 OCFS2_JOURNAL_ACCESS_WRITE);
185 trailer = ocfs2_trailer_from_bh(dirdata_bh, dir->i_sb);
186 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
188 trailer->db_free_next = dx_root->dr_free_blk;
189 dx_root->dr_free_blk = cpu_to_le64(dirdata_bh->b_blocknr);
191 ocfs2_journal_dirty(handle, dx_root_bh);
197 static int ocfs2_free_list_at_root(struct ocfs2_dir_lookup_result *res)
199 return res->dl_prev_leaf_bh == NULL;
202 void ocfs2_free_dir_lookup_result(struct ocfs2_dir_lookup_result *res)
204 brelse(res->dl_dx_root_bh);
205 brelse(res->dl_leaf_bh);
206 brelse(res->dl_dx_leaf_bh);
207 brelse(res->dl_prev_leaf_bh);
210 static int ocfs2_dir_indexed(struct inode *inode)
212 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INDEXED_DIR_FL)
217 static inline int ocfs2_dx_root_inline(struct ocfs2_dx_root_block *dx_root)
219 return dx_root->dr_flags & OCFS2_DX_FLAG_INLINE;
223 * Hashing code adapted from ext3
225 #define DELTA 0x9E3779B9
227 static void TEA_transform(__u32 buf[4], __u32 const in[])
230 __u32 b0 = buf[0], b1 = buf[1];
231 __u32 a = in[0], b = in[1], c = in[2], d = in[3];
236 b0 += ((b1 << 4)+a) ^ (b1+sum) ^ ((b1 >> 5)+b);
237 b1 += ((b0 << 4)+c) ^ (b0+sum) ^ ((b0 >> 5)+d);
244 static void str2hashbuf(const char *msg, int len, __u32 *buf, int num)
249 pad = (__u32)len | ((__u32)len << 8);
255 for (i = 0; i < len; i++) {
258 val = msg[i] + (val << 8);
271 static void ocfs2_dx_dir_name_hash(struct inode *dir, const char *name, int len,
272 struct ocfs2_dx_hinfo *hinfo)
274 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
279 * XXX: Is this really necessary, if the index is never looked
280 * at by readdir? Is a hash value of '0' a bad idea?
282 if ((len == 1 && !strncmp(".", name, 1)) ||
283 (len == 2 && !strncmp("..", name, 2))) {
288 #ifdef OCFS2_DEBUG_DX_DIRS
290 * This makes it very easy to debug indexing problems. We
291 * should never allow this to be selected without hand editing
294 buf[0] = buf[1] = len;
298 memcpy(buf, osb->osb_dx_seed, sizeof(buf));
302 str2hashbuf(p, len, in, 4);
303 TEA_transform(buf, in);
309 hinfo->major_hash = buf[0];
310 hinfo->minor_hash = buf[1];
314 * bh passed here can be an inode block or a dir data block, depending
315 * on the inode inline data flag.
317 static int ocfs2_check_dir_entry(struct inode * dir,
318 struct ocfs2_dir_entry * de,
319 struct buffer_head * bh,
320 unsigned long offset)
322 const char *error_msg = NULL;
323 const int rlen = le16_to_cpu(de->rec_len);
325 if (rlen < OCFS2_DIR_REC_LEN(1))
326 error_msg = "rec_len is smaller than minimal";
327 else if (rlen % 4 != 0)
328 error_msg = "rec_len % 4 != 0";
329 else if (rlen < OCFS2_DIR_REC_LEN(de->name_len))
330 error_msg = "rec_len is too small for name_len";
331 else if (((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)
332 error_msg = "directory entry across blocks";
334 if (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,
340 return error_msg == NULL ? 1 : 0;
343 static inline int ocfs2_match(int len,
344 const char * const name,
345 struct ocfs2_dir_entry *de)
347 if (len != de->name_len)
351 return !memcmp(name, de->name, len);
355 * Returns 0 if not found, -1 on failure, and 1 on success
357 static int inline ocfs2_search_dirblock(struct buffer_head *bh,
359 const char *name, int namelen,
360 unsigned long offset,
363 struct ocfs2_dir_entry **res_dir)
365 struct ocfs2_dir_entry *de;
366 char *dlimit, *de_buf;
373 dlimit = de_buf + bytes;
375 while (de_buf < dlimit) {
376 /* this code is executed quadratically often */
377 /* do minimal checking `by hand' */
379 de = (struct ocfs2_dir_entry *) de_buf;
381 if (de_buf + namelen <= dlimit &&
382 ocfs2_match(namelen, name, de)) {
383 /* found a match - just to be sure, do a full check */
384 if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
393 /* prevent looping on a bad block */
394 de_len = le16_to_cpu(de->rec_len);
409 static struct buffer_head *ocfs2_find_entry_id(const char *name,
412 struct ocfs2_dir_entry **res_dir)
415 struct buffer_head *di_bh = NULL;
416 struct ocfs2_dinode *di;
417 struct ocfs2_inline_data *data;
419 ret = ocfs2_read_inode_block(dir, &di_bh);
425 di = (struct ocfs2_dinode *)di_bh->b_data;
426 data = &di->id2.i_data;
428 found = ocfs2_search_dirblock(di_bh, dir, name, namelen, 0,
429 data->id_data, i_size_read(dir), res_dir);
438 static int ocfs2_validate_dir_block(struct super_block *sb,
439 struct buffer_head *bh)
442 struct ocfs2_dir_block_trailer *trailer =
443 ocfs2_trailer_from_bh(bh, sb);
447 * We don't validate dirents here, that's handled
448 * in-place when the code walks them.
450 mlog(0, "Validating dirblock %llu\n",
451 (unsigned long long)bh->b_blocknr);
453 BUG_ON(!buffer_uptodate(bh));
456 * If the ecc fails, we return the error but otherwise
457 * leave the filesystem running. We know any error is
458 * local to this block.
460 * Note that we are safe to call this even if the directory
461 * doesn't have a trailer. Filesystems without metaecc will do
462 * nothing, and filesystems with it will have one.
464 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &trailer->db_check);
466 mlog(ML_ERROR, "Checksum failed for dinode %llu\n",
467 (unsigned long long)bh->b_blocknr);
473 * Validate a directory trailer.
475 * We check the trailer here rather than in ocfs2_validate_dir_block()
476 * because that function doesn't have the inode to test.
478 static int ocfs2_check_dir_trailer(struct inode *dir, struct buffer_head *bh)
481 struct ocfs2_dir_block_trailer *trailer;
483 trailer = ocfs2_trailer_from_bh(bh, dir->i_sb);
484 if (!OCFS2_IS_VALID_DIR_TRAILER(trailer)) {
486 ocfs2_error(dir->i_sb,
487 "Invalid dirblock #%llu: "
488 "signature = %.*s\n",
489 (unsigned long long)bh->b_blocknr, 7,
490 trailer->db_signature);
493 if (le64_to_cpu(trailer->db_blkno) != bh->b_blocknr) {
495 ocfs2_error(dir->i_sb,
496 "Directory block #%llu has an invalid "
498 (unsigned long long)bh->b_blocknr,
499 (unsigned long long)le64_to_cpu(trailer->db_blkno));
502 if (le64_to_cpu(trailer->db_parent_dinode) !=
503 OCFS2_I(dir)->ip_blkno) {
505 ocfs2_error(dir->i_sb,
506 "Directory block #%llu on dinode "
507 "#%llu has an invalid parent_dinode "
509 (unsigned long long)bh->b_blocknr,
510 (unsigned long long)OCFS2_I(dir)->ip_blkno,
511 (unsigned long long)le64_to_cpu(trailer->db_blkno));
519 * This function forces all errors to -EIO for consistency with its
520 * predecessor, ocfs2_bread(). We haven't audited what returning the
521 * real error codes would do to callers. We log the real codes with
522 * mlog_errno() before we squash them.
524 static int ocfs2_read_dir_block(struct inode *inode, u64 v_block,
525 struct buffer_head **bh, int flags)
528 struct buffer_head *tmp = *bh;
530 rc = ocfs2_read_virt_blocks(inode, v_block, 1, &tmp, flags,
531 ocfs2_validate_dir_block);
537 if (!(flags & OCFS2_BH_READAHEAD) &&
538 ocfs2_supports_dir_trailer(inode)) {
539 rc = ocfs2_check_dir_trailer(inode, tmp);
548 /* If ocfs2_read_virt_blocks() got us a new bh, pass it up. */
553 return rc ? -EIO : 0;
557 * Read the block at 'phys' which belongs to this directory
558 * inode. This function does no virtual->physical block translation -
559 * what's passed in is assumed to be a valid directory block.
561 static int ocfs2_read_dir_block_direct(struct inode *dir, u64 phys,
562 struct buffer_head **bh)
565 struct buffer_head *tmp = *bh;
567 ret = ocfs2_read_block(INODE_CACHE(dir), phys, &tmp,
568 ocfs2_validate_dir_block);
574 if (ocfs2_supports_dir_trailer(dir)) {
575 ret = ocfs2_check_dir_trailer(dir, tmp);
590 static int ocfs2_validate_dx_root(struct super_block *sb,
591 struct buffer_head *bh)
594 struct ocfs2_dx_root_block *dx_root;
596 BUG_ON(!buffer_uptodate(bh));
598 dx_root = (struct ocfs2_dx_root_block *) bh->b_data;
600 ret = ocfs2_validate_meta_ecc(sb, bh->b_data, &dx_root->dr_check);
603 "Checksum failed for dir index root block %llu\n",
604 (unsigned long long)bh->b_blocknr);
608 if (!OCFS2_IS_VALID_DX_ROOT(dx_root)) {
610 "Dir Index Root # %llu has bad signature %.*s",
611 (unsigned long long)le64_to_cpu(dx_root->dr_blkno),
612 7, dx_root->dr_signature);
619 static int ocfs2_read_dx_root(struct inode *dir, struct ocfs2_dinode *di,
620 struct buffer_head **dx_root_bh)
623 u64 blkno = le64_to_cpu(di->i_dx_root);
624 struct buffer_head *tmp = *dx_root_bh;
626 ret = ocfs2_read_block(INODE_CACHE(dir), blkno, &tmp,
627 ocfs2_validate_dx_root);
629 /* If ocfs2_read_block() got us a new bh, pass it up. */
630 if (!ret && !*dx_root_bh)
636 static int ocfs2_validate_dx_leaf(struct super_block *sb,
637 struct buffer_head *bh)
640 struct ocfs2_dx_leaf *dx_leaf = (struct ocfs2_dx_leaf *)bh->b_data;
642 BUG_ON(!buffer_uptodate(bh));
644 ret = ocfs2_validate_meta_ecc(sb, bh->b_data, &dx_leaf->dl_check);
647 "Checksum failed for dir index leaf block %llu\n",
648 (unsigned long long)bh->b_blocknr);
652 if (!OCFS2_IS_VALID_DX_LEAF(dx_leaf)) {
653 ocfs2_error(sb, "Dir Index Leaf has bad signature %.*s",
654 7, dx_leaf->dl_signature);
661 static int ocfs2_read_dx_leaf(struct inode *dir, u64 blkno,
662 struct buffer_head **dx_leaf_bh)
665 struct buffer_head *tmp = *dx_leaf_bh;
667 ret = ocfs2_read_block(INODE_CACHE(dir), blkno, &tmp,
668 ocfs2_validate_dx_leaf);
670 /* If ocfs2_read_block() got us a new bh, pass it up. */
671 if (!ret && !*dx_leaf_bh)
678 * Read a series of dx_leaf blocks. This expects all buffer_head
679 * pointers to be NULL on function entry.
681 static int ocfs2_read_dx_leaves(struct inode *dir, u64 start, int num,
682 struct buffer_head **dx_leaf_bhs)
686 ret = ocfs2_read_blocks(INODE_CACHE(dir), start, num, dx_leaf_bhs, 0,
687 ocfs2_validate_dx_leaf);
694 static struct buffer_head *ocfs2_find_entry_el(const char *name, int namelen,
696 struct ocfs2_dir_entry **res_dir)
698 struct super_block *sb;
699 struct buffer_head *bh_use[NAMEI_RA_SIZE];
700 struct buffer_head *bh, *ret = NULL;
701 unsigned long start, block, b;
702 int ra_max = 0; /* Number of bh's in the readahead
704 int ra_ptr = 0; /* Current index into readahead
713 nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
714 start = OCFS2_I(dir)->ip_dir_start_lookup;
715 if (start >= nblocks)
722 * We deal with the read-ahead logic here.
724 if (ra_ptr >= ra_max) {
725 /* Refill the readahead buffer */
728 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
730 * Terminate if we reach the end of the
731 * directory and must wrap, or if our
732 * search has finished at this block.
734 if (b >= nblocks || (num && block == start)) {
735 bh_use[ra_max] = NULL;
741 err = ocfs2_read_dir_block(dir, b++, &bh,
746 if ((bh = bh_use[ra_ptr++]) == NULL)
748 if (ocfs2_read_dir_block(dir, block, &bh, 0)) {
749 /* read error, skip block & hope for the best.
750 * ocfs2_read_dir_block() has released the bh. */
751 ocfs2_error(dir->i_sb, "reading directory %llu, "
753 (unsigned long long)OCFS2_I(dir)->ip_blkno,
757 i = ocfs2_search_dirblock(bh, dir, name, namelen,
758 block << sb->s_blocksize_bits,
759 bh->b_data, sb->s_blocksize,
762 OCFS2_I(dir)->ip_dir_start_lookup = block;
764 goto cleanup_and_exit;
768 goto cleanup_and_exit;
771 if (++block >= nblocks)
773 } while (block != start);
776 * If the directory has grown while we were searching, then
777 * search the last part of the directory before giving up.
780 nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
781 if (block < nblocks) {
787 /* Clean up the read-ahead blocks */
788 for (; ra_ptr < ra_max; ra_ptr++)
789 brelse(bh_use[ra_ptr]);
795 static int ocfs2_dx_dir_lookup_rec(struct inode *inode,
796 struct ocfs2_extent_list *el,
800 unsigned int *ret_clen)
802 int ret = 0, i, found;
803 struct buffer_head *eb_bh = NULL;
804 struct ocfs2_extent_block *eb;
805 struct ocfs2_extent_rec *rec = NULL;
807 if (el->l_tree_depth) {
808 ret = ocfs2_find_leaf(inode, el, major_hash, &eb_bh);
814 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
817 if (el->l_tree_depth) {
818 ocfs2_error(inode->i_sb,
819 "Inode %lu has non zero tree depth in "
820 "btree tree block %llu\n", inode->i_ino,
821 (unsigned long long)eb_bh->b_blocknr);
828 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
829 rec = &el->l_recs[i];
831 if (le32_to_cpu(rec->e_cpos) <= major_hash) {
838 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
839 "record (%u, %u, 0) in btree", inode->i_ino,
840 le32_to_cpu(rec->e_cpos),
841 ocfs2_rec_clusters(el, rec));
847 *ret_phys_blkno = le64_to_cpu(rec->e_blkno);
849 *ret_cpos = le32_to_cpu(rec->e_cpos);
851 *ret_clen = le16_to_cpu(rec->e_leaf_clusters);
859 * Returns the block index, from the start of the cluster which this
862 static inline unsigned int __ocfs2_dx_dir_hash_idx(struct ocfs2_super *osb,
865 return minor_hash & osb->osb_dx_mask;
868 static inline unsigned int ocfs2_dx_dir_hash_idx(struct ocfs2_super *osb,
869 struct ocfs2_dx_hinfo *hinfo)
871 return __ocfs2_dx_dir_hash_idx(osb, hinfo->minor_hash);
874 static int ocfs2_dx_dir_lookup(struct inode *inode,
875 struct ocfs2_extent_list *el,
876 struct ocfs2_dx_hinfo *hinfo,
881 unsigned int cend, uninitialized_var(clen);
882 u32 uninitialized_var(cpos);
883 u64 uninitialized_var(blkno);
884 u32 name_hash = hinfo->major_hash;
886 ret = ocfs2_dx_dir_lookup_rec(inode, el, name_hash, &cpos, &blkno,
894 if (name_hash >= cend) {
895 /* We want the last cluster */
896 blkno += ocfs2_clusters_to_blocks(inode->i_sb, clen - 1);
899 blkno += ocfs2_clusters_to_blocks(inode->i_sb,
905 * We now have the cluster which should hold our entry. To
906 * find the exact block from the start of the cluster to
907 * search, we take the lower bits of the hash.
909 blkno += ocfs2_dx_dir_hash_idx(OCFS2_SB(inode->i_sb), hinfo);
912 *ret_phys_blkno = blkno;
921 static int ocfs2_dx_dir_search(const char *name, int namelen,
923 struct ocfs2_dx_root_block *dx_root,
924 struct ocfs2_dir_lookup_result *res)
927 u64 uninitialized_var(phys);
928 struct buffer_head *dx_leaf_bh = NULL;
929 struct ocfs2_dx_leaf *dx_leaf;
930 struct ocfs2_dx_entry *dx_entry = NULL;
931 struct buffer_head *dir_ent_bh = NULL;
932 struct ocfs2_dir_entry *dir_ent = NULL;
933 struct ocfs2_dx_hinfo *hinfo = &res->dl_hinfo;
934 struct ocfs2_extent_list *dr_el;
935 struct ocfs2_dx_entry_list *entry_list;
937 ocfs2_dx_dir_name_hash(dir, name, namelen, &res->dl_hinfo);
939 if (ocfs2_dx_root_inline(dx_root)) {
940 entry_list = &dx_root->dr_entries;
944 dr_el = &dx_root->dr_list;
946 ret = ocfs2_dx_dir_lookup(dir, dr_el, hinfo, NULL, &phys);
952 mlog(0, "Dir %llu: name: \"%.*s\", lookup of hash: %u.0x%x "
954 (unsigned long long)OCFS2_I(dir)->ip_blkno,
955 namelen, name, hinfo->major_hash, hinfo->minor_hash,
956 (unsigned long long)phys);
958 ret = ocfs2_read_dx_leaf(dir, phys, &dx_leaf_bh);
964 dx_leaf = (struct ocfs2_dx_leaf *) dx_leaf_bh->b_data;
966 mlog(0, "leaf info: num_used: %d, count: %d\n",
967 le16_to_cpu(dx_leaf->dl_list.de_num_used),
968 le16_to_cpu(dx_leaf->dl_list.de_count));
970 entry_list = &dx_leaf->dl_list;
974 * Empty leaf is legal, so no need to check for that.
977 for (i = 0; i < le16_to_cpu(entry_list->de_num_used); i++) {
978 dx_entry = &entry_list->de_entries[i];
980 if (hinfo->major_hash != le32_to_cpu(dx_entry->dx_major_hash)
981 || hinfo->minor_hash != le32_to_cpu(dx_entry->dx_minor_hash))
985 * Search unindexed leaf block now. We're not
986 * guaranteed to find anything.
988 ret = ocfs2_read_dir_block_direct(dir,
989 le64_to_cpu(dx_entry->dx_dirent_blk),
997 * XXX: We should check the unindexed block here,
1001 found = ocfs2_search_dirblock(dir_ent_bh, dir, name, namelen,
1002 0, dir_ent_bh->b_data,
1003 dir->i_sb->s_blocksize, &dir_ent);
1008 /* This means we found a bad directory entry. */
1023 res->dl_leaf_bh = dir_ent_bh;
1024 res->dl_entry = dir_ent;
1025 res->dl_dx_leaf_bh = dx_leaf_bh;
1026 res->dl_dx_entry = dx_entry;
1037 static int ocfs2_find_entry_dx(const char *name, int namelen,
1039 struct ocfs2_dir_lookup_result *lookup)
1042 struct buffer_head *di_bh = NULL;
1043 struct ocfs2_dinode *di;
1044 struct buffer_head *dx_root_bh = NULL;
1045 struct ocfs2_dx_root_block *dx_root;
1047 ret = ocfs2_read_inode_block(dir, &di_bh);
1053 di = (struct ocfs2_dinode *)di_bh->b_data;
1055 ret = ocfs2_read_dx_root(dir, di, &dx_root_bh);
1060 dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data;
1062 ret = ocfs2_dx_dir_search(name, namelen, dir, dx_root, lookup);
1069 lookup->dl_dx_root_bh = dx_root_bh;
1078 * Try to find an entry of the provided name within 'dir'.
1080 * If nothing was found, -ENOENT is returned. Otherwise, zero is
1081 * returned and the struct 'res' will contain information useful to
1082 * other directory manipulation functions.
1084 * Caller can NOT assume anything about the contents of the
1085 * buffer_heads - they are passed back only so that it can be passed
1086 * into any one of the manipulation functions (add entry, delete
1087 * entry, etc). As an example, bh in the extent directory case is a
1088 * data block, in the inline-data case it actually points to an inode,
1089 * in the indexed directory case, multiple buffers are involved.
1091 int ocfs2_find_entry(const char *name, int namelen,
1092 struct inode *dir, struct ocfs2_dir_lookup_result *lookup)
1094 struct buffer_head *bh;
1095 struct ocfs2_dir_entry *res_dir = NULL;
1097 if (ocfs2_dir_indexed(dir))
1098 return ocfs2_find_entry_dx(name, namelen, dir, lookup);
1101 * The unindexed dir code only uses part of the lookup
1102 * structure, so there's no reason to push it down further
1105 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
1106 bh = ocfs2_find_entry_id(name, namelen, dir, &res_dir);
1108 bh = ocfs2_find_entry_el(name, namelen, dir, &res_dir);
1113 lookup->dl_leaf_bh = bh;
1114 lookup->dl_entry = res_dir;
1119 * Update inode number and type of a previously found directory entry.
1121 int ocfs2_update_entry(struct inode *dir, handle_t *handle,
1122 struct ocfs2_dir_lookup_result *res,
1123 struct inode *new_entry_inode)
1126 ocfs2_journal_access_func access = ocfs2_journal_access_db;
1127 struct ocfs2_dir_entry *de = res->dl_entry;
1128 struct buffer_head *de_bh = res->dl_leaf_bh;
1131 * The same code works fine for both inline-data and extent
1132 * based directories, so no need to split this up. The only
1133 * difference is the journal_access function.
1136 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
1137 access = ocfs2_journal_access_di;
1139 ret = access(handle, INODE_CACHE(dir), de_bh,
1140 OCFS2_JOURNAL_ACCESS_WRITE);
1146 de->inode = cpu_to_le64(OCFS2_I(new_entry_inode)->ip_blkno);
1147 ocfs2_set_de_type(de, new_entry_inode->i_mode);
1149 ocfs2_journal_dirty(handle, de_bh);
1156 * __ocfs2_delete_entry deletes a directory entry by merging it with the
1159 static int __ocfs2_delete_entry(handle_t *handle, struct inode *dir,
1160 struct ocfs2_dir_entry *de_del,
1161 struct buffer_head *bh, char *first_de,
1164 struct ocfs2_dir_entry *de, *pde;
1165 int i, status = -ENOENT;
1166 ocfs2_journal_access_func access = ocfs2_journal_access_db;
1168 mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p)\n", handle, dir, de_del, bh);
1170 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
1171 access = ocfs2_journal_access_di;
1175 de = (struct ocfs2_dir_entry *) first_de;
1177 if (!ocfs2_check_dir_entry(dir, de, bh, i)) {
1183 status = access(handle, INODE_CACHE(dir), bh,
1184 OCFS2_JOURNAL_ACCESS_WRITE);
1191 le16_add_cpu(&pde->rec_len,
1192 le16_to_cpu(de->rec_len));
1196 status = ocfs2_journal_dirty(handle, bh);
1199 i += le16_to_cpu(de->rec_len);
1201 de = (struct ocfs2_dir_entry *)((char *)de + le16_to_cpu(de->rec_len));
1208 static unsigned int ocfs2_figure_dirent_hole(struct ocfs2_dir_entry *de)
1212 if (le64_to_cpu(de->inode) == 0)
1213 hole = le16_to_cpu(de->rec_len);
1215 hole = le16_to_cpu(de->rec_len) -
1216 OCFS2_DIR_REC_LEN(de->name_len);
1221 static int ocfs2_find_max_rec_len(struct super_block *sb,
1222 struct buffer_head *dirblock_bh)
1224 int size, this_hole, largest_hole = 0;
1225 char *trailer, *de_buf, *limit, *start = dirblock_bh->b_data;
1226 struct ocfs2_dir_entry *de;
1228 trailer = (char *)ocfs2_trailer_from_bh(dirblock_bh, sb);
1229 size = ocfs2_dir_trailer_blk_off(sb);
1230 limit = start + size;
1232 de = (struct ocfs2_dir_entry *)de_buf;
1234 if (de_buf != trailer) {
1235 this_hole = ocfs2_figure_dirent_hole(de);
1236 if (this_hole > largest_hole)
1237 largest_hole = this_hole;
1240 de_buf += le16_to_cpu(de->rec_len);
1241 de = (struct ocfs2_dir_entry *)de_buf;
1242 } while (de_buf < limit);
1244 if (largest_hole >= OCFS2_DIR_MIN_REC_LEN)
1245 return largest_hole;
1249 static void ocfs2_dx_list_remove_entry(struct ocfs2_dx_entry_list *entry_list,
1252 int num_used = le16_to_cpu(entry_list->de_num_used);
1254 if (num_used == 1 || index == (num_used - 1))
1257 memmove(&entry_list->de_entries[index],
1258 &entry_list->de_entries[index + 1],
1259 (num_used - index - 1)*sizeof(struct ocfs2_dx_entry));
1262 memset(&entry_list->de_entries[num_used], 0,
1263 sizeof(struct ocfs2_dx_entry));
1264 entry_list->de_num_used = cpu_to_le16(num_used);
1267 static int ocfs2_delete_entry_dx(handle_t *handle, struct inode *dir,
1268 struct ocfs2_dir_lookup_result *lookup)
1270 int ret, index, max_rec_len, add_to_free_list = 0;
1271 struct buffer_head *dx_root_bh = lookup->dl_dx_root_bh;
1272 struct buffer_head *leaf_bh = lookup->dl_leaf_bh;
1273 struct ocfs2_dx_leaf *dx_leaf;
1274 struct ocfs2_dx_entry *dx_entry = lookup->dl_dx_entry;
1275 struct ocfs2_dir_block_trailer *trailer;
1276 struct ocfs2_dx_root_block *dx_root;
1277 struct ocfs2_dx_entry_list *entry_list;
1280 * This function gets a bit messy because we might have to
1281 * modify the root block, regardless of whether the indexed
1282 * entries are stored inline.
1286 * *Only* set 'entry_list' here, based on where we're looking
1287 * for the indexed entries. Later, we might still want to
1288 * journal both blocks, based on free list state.
1290 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
1291 if (ocfs2_dx_root_inline(dx_root)) {
1292 entry_list = &dx_root->dr_entries;
1294 dx_leaf = (struct ocfs2_dx_leaf *) lookup->dl_dx_leaf_bh->b_data;
1295 entry_list = &dx_leaf->dl_list;
1298 /* Neither of these are a disk corruption - that should have
1299 * been caught by lookup, before we got here. */
1300 BUG_ON(le16_to_cpu(entry_list->de_count) <= 0);
1301 BUG_ON(le16_to_cpu(entry_list->de_num_used) <= 0);
1303 index = (char *)dx_entry - (char *)entry_list->de_entries;
1304 index /= sizeof(*dx_entry);
1306 if (index >= le16_to_cpu(entry_list->de_num_used)) {
1307 mlog(ML_ERROR, "Dir %llu: Bad dx_entry ptr idx %d, (%p, %p)\n",
1308 (unsigned long long)OCFS2_I(dir)->ip_blkno, index,
1309 entry_list, dx_entry);
1314 * We know that removal of this dirent will leave enough room
1315 * for a new one, so add this block to the free list if it
1316 * isn't already there.
1318 trailer = ocfs2_trailer_from_bh(leaf_bh, dir->i_sb);
1319 if (trailer->db_free_rec_len == 0)
1320 add_to_free_list = 1;
1323 * Add the block holding our index into the journal before
1324 * removing the unindexed entry. If we get an error return
1325 * from __ocfs2_delete_entry(), then it hasn't removed the
1326 * entry yet. Likewise, successful return means we *must*
1327 * remove the indexed entry.
1329 * We're also careful to journal the root tree block here as
1330 * the entry count needs to be updated. Also, we might be
1331 * adding to the start of the free list.
1333 ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
1334 OCFS2_JOURNAL_ACCESS_WRITE);
1340 if (!ocfs2_dx_root_inline(dx_root)) {
1341 ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir),
1342 lookup->dl_dx_leaf_bh,
1343 OCFS2_JOURNAL_ACCESS_WRITE);
1350 mlog(0, "Dir %llu: delete entry at index: %d\n",
1351 (unsigned long long)OCFS2_I(dir)->ip_blkno, index);
1353 ret = __ocfs2_delete_entry(handle, dir, lookup->dl_entry,
1354 leaf_bh, leaf_bh->b_data, leaf_bh->b_size);
1360 max_rec_len = ocfs2_find_max_rec_len(dir->i_sb, leaf_bh);
1361 trailer->db_free_rec_len = cpu_to_le16(max_rec_len);
1362 if (add_to_free_list) {
1363 trailer->db_free_next = dx_root->dr_free_blk;
1364 dx_root->dr_free_blk = cpu_to_le64(leaf_bh->b_blocknr);
1365 ocfs2_journal_dirty(handle, dx_root_bh);
1368 /* leaf_bh was journal_accessed for us in __ocfs2_delete_entry */
1369 ocfs2_journal_dirty(handle, leaf_bh);
1371 le32_add_cpu(&dx_root->dr_num_entries, -1);
1372 ocfs2_journal_dirty(handle, dx_root_bh);
1374 ocfs2_dx_list_remove_entry(entry_list, index);
1376 if (!ocfs2_dx_root_inline(dx_root))
1377 ocfs2_journal_dirty(handle, lookup->dl_dx_leaf_bh);
1383 static inline int ocfs2_delete_entry_id(handle_t *handle,
1385 struct ocfs2_dir_entry *de_del,
1386 struct buffer_head *bh)
1389 struct buffer_head *di_bh = NULL;
1390 struct ocfs2_dinode *di;
1391 struct ocfs2_inline_data *data;
1393 ret = ocfs2_read_inode_block(dir, &di_bh);
1399 di = (struct ocfs2_dinode *)di_bh->b_data;
1400 data = &di->id2.i_data;
1402 ret = __ocfs2_delete_entry(handle, dir, de_del, bh, data->id_data,
1410 static inline int ocfs2_delete_entry_el(handle_t *handle,
1412 struct ocfs2_dir_entry *de_del,
1413 struct buffer_head *bh)
1415 return __ocfs2_delete_entry(handle, dir, de_del, bh, bh->b_data,
1420 * Delete a directory entry. Hide the details of directory
1421 * implementation from the caller.
1423 int ocfs2_delete_entry(handle_t *handle,
1425 struct ocfs2_dir_lookup_result *res)
1427 if (ocfs2_dir_indexed(dir))
1428 return ocfs2_delete_entry_dx(handle, dir, res);
1430 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
1431 return ocfs2_delete_entry_id(handle, dir, res->dl_entry,
1434 return ocfs2_delete_entry_el(handle, dir, res->dl_entry,
1439 * Check whether 'de' has enough room to hold an entry of
1440 * 'new_rec_len' bytes.
1442 static inline int ocfs2_dirent_would_fit(struct ocfs2_dir_entry *de,
1443 unsigned int new_rec_len)
1445 unsigned int de_really_used;
1447 /* Check whether this is an empty record with enough space */
1448 if (le64_to_cpu(de->inode) == 0 &&
1449 le16_to_cpu(de->rec_len) >= new_rec_len)
1453 * Record might have free space at the end which we can
1456 de_really_used = OCFS2_DIR_REC_LEN(de->name_len);
1457 if (le16_to_cpu(de->rec_len) >= (de_really_used + new_rec_len))
1463 static void ocfs2_dx_dir_leaf_insert_tail(struct ocfs2_dx_leaf *dx_leaf,
1464 struct ocfs2_dx_entry *dx_new_entry)
1468 i = le16_to_cpu(dx_leaf->dl_list.de_num_used);
1469 dx_leaf->dl_list.de_entries[i] = *dx_new_entry;
1471 le16_add_cpu(&dx_leaf->dl_list.de_num_used, 1);
1474 static void ocfs2_dx_entry_list_insert(struct ocfs2_dx_entry_list *entry_list,
1475 struct ocfs2_dx_hinfo *hinfo,
1479 struct ocfs2_dx_entry *dx_entry;
1481 i = le16_to_cpu(entry_list->de_num_used);
1482 dx_entry = &entry_list->de_entries[i];
1484 memset(dx_entry, 0, sizeof(*dx_entry));
1485 dx_entry->dx_major_hash = cpu_to_le32(hinfo->major_hash);
1486 dx_entry->dx_minor_hash = cpu_to_le32(hinfo->minor_hash);
1487 dx_entry->dx_dirent_blk = cpu_to_le64(dirent_blk);
1489 le16_add_cpu(&entry_list->de_num_used, 1);
1492 static int __ocfs2_dx_dir_leaf_insert(struct inode *dir, handle_t *handle,
1493 struct ocfs2_dx_hinfo *hinfo,
1495 struct buffer_head *dx_leaf_bh)
1498 struct ocfs2_dx_leaf *dx_leaf;
1500 ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir), dx_leaf_bh,
1501 OCFS2_JOURNAL_ACCESS_WRITE);
1507 dx_leaf = (struct ocfs2_dx_leaf *)dx_leaf_bh->b_data;
1508 ocfs2_dx_entry_list_insert(&dx_leaf->dl_list, hinfo, dirent_blk);
1509 ocfs2_journal_dirty(handle, dx_leaf_bh);
1515 static void ocfs2_dx_inline_root_insert(struct inode *dir, handle_t *handle,
1516 struct ocfs2_dx_hinfo *hinfo,
1518 struct ocfs2_dx_root_block *dx_root)
1520 ocfs2_dx_entry_list_insert(&dx_root->dr_entries, hinfo, dirent_blk);
1523 static int ocfs2_dx_dir_insert(struct inode *dir, handle_t *handle,
1524 struct ocfs2_dir_lookup_result *lookup)
1527 struct ocfs2_dx_root_block *dx_root;
1528 struct buffer_head *dx_root_bh = lookup->dl_dx_root_bh;
1530 ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
1531 OCFS2_JOURNAL_ACCESS_WRITE);
1537 dx_root = (struct ocfs2_dx_root_block *)lookup->dl_dx_root_bh->b_data;
1538 if (ocfs2_dx_root_inline(dx_root)) {
1539 ocfs2_dx_inline_root_insert(dir, handle,
1541 lookup->dl_leaf_bh->b_blocknr,
1544 ret = __ocfs2_dx_dir_leaf_insert(dir, handle, &lookup->dl_hinfo,
1545 lookup->dl_leaf_bh->b_blocknr,
1546 lookup->dl_dx_leaf_bh);
1551 le32_add_cpu(&dx_root->dr_num_entries, 1);
1552 ocfs2_journal_dirty(handle, dx_root_bh);
1558 static void ocfs2_remove_block_from_free_list(struct inode *dir,
1560 struct ocfs2_dir_lookup_result *lookup)
1562 struct ocfs2_dir_block_trailer *trailer, *prev;
1563 struct ocfs2_dx_root_block *dx_root;
1564 struct buffer_head *bh;
1566 trailer = ocfs2_trailer_from_bh(lookup->dl_leaf_bh, dir->i_sb);
1568 if (ocfs2_free_list_at_root(lookup)) {
1569 bh = lookup->dl_dx_root_bh;
1570 dx_root = (struct ocfs2_dx_root_block *)bh->b_data;
1571 dx_root->dr_free_blk = trailer->db_free_next;
1573 bh = lookup->dl_prev_leaf_bh;
1574 prev = ocfs2_trailer_from_bh(bh, dir->i_sb);
1575 prev->db_free_next = trailer->db_free_next;
1578 trailer->db_free_rec_len = cpu_to_le16(0);
1579 trailer->db_free_next = cpu_to_le64(0);
1581 ocfs2_journal_dirty(handle, bh);
1582 ocfs2_journal_dirty(handle, lookup->dl_leaf_bh);
1586 * This expects that a journal write has been reserved on
1587 * lookup->dl_prev_leaf_bh or lookup->dl_dx_root_bh
1589 static void ocfs2_recalc_free_list(struct inode *dir, handle_t *handle,
1590 struct ocfs2_dir_lookup_result *lookup)
1593 struct ocfs2_dir_block_trailer *trailer;
1595 /* Walk dl_leaf_bh to figure out what the new free rec_len is. */
1596 max_rec_len = ocfs2_find_max_rec_len(dir->i_sb, lookup->dl_leaf_bh);
1599 * There's still room in this block, so no need to remove it
1600 * from the free list. In this case, we just want to update
1601 * the rec len accounting.
1603 trailer = ocfs2_trailer_from_bh(lookup->dl_leaf_bh, dir->i_sb);
1604 trailer->db_free_rec_len = cpu_to_le16(max_rec_len);
1605 ocfs2_journal_dirty(handle, lookup->dl_leaf_bh);
1607 ocfs2_remove_block_from_free_list(dir, handle, lookup);
1611 /* we don't always have a dentry for what we want to add, so people
1612 * like orphan dir can call this instead.
1614 * The lookup context must have been filled from
1615 * ocfs2_prepare_dir_for_insert.
1617 int __ocfs2_add_entry(handle_t *handle,
1619 const char *name, int namelen,
1620 struct inode *inode, u64 blkno,
1621 struct buffer_head *parent_fe_bh,
1622 struct ocfs2_dir_lookup_result *lookup)
1624 unsigned long offset;
1625 unsigned short rec_len;
1626 struct ocfs2_dir_entry *de, *de1;
1627 struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_fe_bh->b_data;
1628 struct super_block *sb = dir->i_sb;
1630 unsigned int size = sb->s_blocksize;
1631 struct buffer_head *insert_bh = lookup->dl_leaf_bh;
1632 char *data_start = insert_bh->b_data;
1639 if (ocfs2_dir_indexed(dir)) {
1640 struct buffer_head *bh;
1643 * An indexed dir may require that we update the free space
1644 * list. Reserve a write to the previous node in the list so
1645 * that we don't fail later.
1647 * XXX: This can be either a dx_root_block, or an unindexed
1648 * directory tree leaf block.
1650 if (ocfs2_free_list_at_root(lookup)) {
1651 bh = lookup->dl_dx_root_bh;
1652 retval = ocfs2_journal_access_dr(handle,
1653 INODE_CACHE(dir), bh,
1654 OCFS2_JOURNAL_ACCESS_WRITE);
1656 bh = lookup->dl_prev_leaf_bh;
1657 retval = ocfs2_journal_access_db(handle,
1658 INODE_CACHE(dir), bh,
1659 OCFS2_JOURNAL_ACCESS_WRITE);
1665 } else if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1666 data_start = di->id2.i_data.id_data;
1667 size = i_size_read(dir);
1669 BUG_ON(insert_bh != parent_fe_bh);
1672 rec_len = OCFS2_DIR_REC_LEN(namelen);
1674 de = (struct ocfs2_dir_entry *) data_start;
1676 BUG_ON((char *)de >= (size + data_start));
1678 /* These checks should've already been passed by the
1679 * prepare function, but I guess we can leave them
1681 if (!ocfs2_check_dir_entry(dir, de, insert_bh, offset)) {
1685 if (ocfs2_match(namelen, name, de)) {
1690 /* We're guaranteed that we should have space, so we
1691 * can't possibly have hit the trailer...right? */
1692 mlog_bug_on_msg(ocfs2_skip_dir_trailer(dir, de, offset, size),
1693 "Hit dir trailer trying to insert %.*s "
1694 "(namelen %d) into directory %llu. "
1695 "offset is %lu, trailer offset is %d\n",
1696 namelen, name, namelen,
1697 (unsigned long long)parent_fe_bh->b_blocknr,
1698 offset, ocfs2_dir_trailer_blk_off(dir->i_sb));
1700 if (ocfs2_dirent_would_fit(de, rec_len)) {
1701 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1702 retval = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
1708 if (insert_bh == parent_fe_bh)
1709 status = ocfs2_journal_access_di(handle,
1712 OCFS2_JOURNAL_ACCESS_WRITE);
1714 status = ocfs2_journal_access_db(handle,
1717 OCFS2_JOURNAL_ACCESS_WRITE);
1719 if (ocfs2_dir_indexed(dir)) {
1720 status = ocfs2_dx_dir_insert(dir,
1730 /* By now the buffer is marked for journaling */
1731 offset += le16_to_cpu(de->rec_len);
1732 if (le64_to_cpu(de->inode)) {
1733 de1 = (struct ocfs2_dir_entry *)((char *) de +
1734 OCFS2_DIR_REC_LEN(de->name_len));
1736 cpu_to_le16(le16_to_cpu(de->rec_len) -
1737 OCFS2_DIR_REC_LEN(de->name_len));
1738 de->rec_len = cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
1741 de->file_type = OCFS2_FT_UNKNOWN;
1743 de->inode = cpu_to_le64(blkno);
1744 ocfs2_set_de_type(de, inode->i_mode);
1747 de->name_len = namelen;
1748 memcpy(de->name, name, namelen);
1750 if (ocfs2_dir_indexed(dir))
1751 ocfs2_recalc_free_list(dir, handle, lookup);
1754 status = ocfs2_journal_dirty(handle, insert_bh);
1759 offset += le16_to_cpu(de->rec_len);
1760 de = (struct ocfs2_dir_entry *) ((char *) de + le16_to_cpu(de->rec_len));
1763 /* when you think about it, the assert above should prevent us
1764 * from ever getting here. */
1772 static int ocfs2_dir_foreach_blk_id(struct inode *inode,
1774 loff_t *f_pos, void *priv,
1775 filldir_t filldir, int *filldir_err)
1777 int ret, i, filldir_ret;
1778 unsigned long offset = *f_pos;
1779 struct buffer_head *di_bh = NULL;
1780 struct ocfs2_dinode *di;
1781 struct ocfs2_inline_data *data;
1782 struct ocfs2_dir_entry *de;
1784 ret = ocfs2_read_inode_block(inode, &di_bh);
1786 mlog(ML_ERROR, "Unable to read inode block for dir %llu\n",
1787 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1791 di = (struct ocfs2_dinode *)di_bh->b_data;
1792 data = &di->id2.i_data;
1794 while (*f_pos < i_size_read(inode)) {
1796 /* If the dir block has changed since the last call to
1797 * readdir(2), then we might be pointing to an invalid
1798 * dirent right now. Scan from the start of the block
1800 if (*f_version != inode->i_version) {
1801 for (i = 0; i < i_size_read(inode) && i < offset; ) {
1802 de = (struct ocfs2_dir_entry *)
1803 (data->id_data + i);
1804 /* It's too expensive to do a full
1805 * dirent test each time round this
1806 * loop, but we do have to test at
1807 * least that it is non-zero. A
1808 * failure will be detected in the
1809 * dirent test below. */
1810 if (le16_to_cpu(de->rec_len) <
1811 OCFS2_DIR_REC_LEN(1))
1813 i += le16_to_cpu(de->rec_len);
1815 *f_pos = offset = i;
1816 *f_version = inode->i_version;
1819 de = (struct ocfs2_dir_entry *) (data->id_data + *f_pos);
1820 if (!ocfs2_check_dir_entry(inode, de, di_bh, *f_pos)) {
1821 /* On error, skip the f_pos to the end. */
1822 *f_pos = i_size_read(inode);
1825 offset += le16_to_cpu(de->rec_len);
1826 if (le64_to_cpu(de->inode)) {
1827 /* We might block in the next section
1828 * if the data destination is
1829 * currently swapped out. So, use a
1830 * version stamp to detect whether or
1831 * not the directory has been modified
1832 * during the copy operation.
1834 u64 version = *f_version;
1835 unsigned char d_type = DT_UNKNOWN;
1837 if (de->file_type < OCFS2_FT_MAX)
1838 d_type = ocfs2_filetype_table[de->file_type];
1840 filldir_ret = filldir(priv, de->name,
1843 le64_to_cpu(de->inode),
1847 *filldir_err = filldir_ret;
1850 if (version != *f_version)
1853 *f_pos += le16_to_cpu(de->rec_len);
1863 * NOTE: This function can be called against unindexed directories,
1866 static int ocfs2_dir_foreach_blk_el(struct inode *inode,
1868 loff_t *f_pos, void *priv,
1869 filldir_t filldir, int *filldir_err)
1872 unsigned long offset, blk, last_ra_blk = 0;
1874 struct buffer_head * bh, * tmp;
1875 struct ocfs2_dir_entry * de;
1876 struct super_block * sb = inode->i_sb;
1877 unsigned int ra_sectors = 16;
1882 offset = (*f_pos) & (sb->s_blocksize - 1);
1884 while (!error && !stored && *f_pos < i_size_read(inode)) {
1885 blk = (*f_pos) >> sb->s_blocksize_bits;
1886 if (ocfs2_read_dir_block(inode, blk, &bh, 0)) {
1887 /* Skip the corrupt dirblock and keep trying */
1888 *f_pos += sb->s_blocksize - offset;
1892 /* The idea here is to begin with 8k read-ahead and to stay
1893 * 4k ahead of our current position.
1895 * TODO: Use the pagecache for this. We just need to
1896 * make sure it's cluster-safe... */
1898 || (((last_ra_blk - blk) << 9) <= (ra_sectors / 2))) {
1899 for (i = ra_sectors >> (sb->s_blocksize_bits - 9);
1902 if (!ocfs2_read_dir_block(inode, ++blk, &tmp,
1903 OCFS2_BH_READAHEAD))
1911 /* If the dir block has changed since the last call to
1912 * readdir(2), then we might be pointing to an invalid
1913 * dirent right now. Scan from the start of the block
1915 if (*f_version != inode->i_version) {
1916 for (i = 0; i < sb->s_blocksize && i < offset; ) {
1917 de = (struct ocfs2_dir_entry *) (bh->b_data + i);
1918 /* It's too expensive to do a full
1919 * dirent test each time round this
1920 * loop, but we do have to test at
1921 * least that it is non-zero. A
1922 * failure will be detected in the
1923 * dirent test below. */
1924 if (le16_to_cpu(de->rec_len) <
1925 OCFS2_DIR_REC_LEN(1))
1927 i += le16_to_cpu(de->rec_len);
1930 *f_pos = ((*f_pos) & ~(sb->s_blocksize - 1))
1932 *f_version = inode->i_version;
1935 while (!error && *f_pos < i_size_read(inode)
1936 && offset < sb->s_blocksize) {
1937 de = (struct ocfs2_dir_entry *) (bh->b_data + offset);
1938 if (!ocfs2_check_dir_entry(inode, de, bh, offset)) {
1939 /* On error, skip the f_pos to the
1941 *f_pos = ((*f_pos) | (sb->s_blocksize - 1)) + 1;
1945 offset += le16_to_cpu(de->rec_len);
1946 if (le64_to_cpu(de->inode)) {
1947 /* We might block in the next section
1948 * if the data destination is
1949 * currently swapped out. So, use a
1950 * version stamp to detect whether or
1951 * not the directory has been modified
1952 * during the copy operation.
1954 unsigned long version = *f_version;
1955 unsigned char d_type = DT_UNKNOWN;
1957 if (de->file_type < OCFS2_FT_MAX)
1958 d_type = ocfs2_filetype_table[de->file_type];
1959 error = filldir(priv, de->name,
1962 le64_to_cpu(de->inode),
1966 *filldir_err = error;
1969 if (version != *f_version)
1973 *f_pos += le16_to_cpu(de->rec_len);
1985 static int ocfs2_dir_foreach_blk(struct inode *inode, u64 *f_version,
1986 loff_t *f_pos, void *priv, filldir_t filldir,
1989 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
1990 return ocfs2_dir_foreach_blk_id(inode, f_version, f_pos, priv,
1991 filldir, filldir_err);
1993 return ocfs2_dir_foreach_blk_el(inode, f_version, f_pos, priv, filldir,
1998 * This is intended to be called from inside other kernel functions,
1999 * so we fake some arguments.
2001 int ocfs2_dir_foreach(struct inode *inode, loff_t *f_pos, void *priv,
2004 int ret = 0, filldir_err = 0;
2005 u64 version = inode->i_version;
2007 while (*f_pos < i_size_read(inode)) {
2008 ret = ocfs2_dir_foreach_blk(inode, &version, f_pos, priv,
2009 filldir, &filldir_err);
2010 if (ret || filldir_err)
2024 int ocfs2_readdir(struct file * filp, void * dirent, filldir_t filldir)
2027 struct inode *inode = filp->f_path.dentry->d_inode;
2030 mlog_entry("dirino=%llu\n",
2031 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2033 error = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level);
2034 if (lock_level && error >= 0) {
2035 /* We release EX lock which used to update atime
2036 * and get PR lock again to reduce contention
2037 * on commonly accessed directories. */
2038 ocfs2_inode_unlock(inode, 1);
2040 error = ocfs2_inode_lock(inode, NULL, 0);
2043 if (error != -ENOENT)
2045 /* we haven't got any yet, so propagate the error. */
2049 error = ocfs2_dir_foreach_blk(inode, &filp->f_version, &filp->f_pos,
2050 dirent, filldir, NULL);
2052 ocfs2_inode_unlock(inode, lock_level);
2061 * NOTE: this should always be called with parent dir i_mutex taken.
2063 int ocfs2_find_files_on_disk(const char *name,
2066 struct inode *inode,
2067 struct ocfs2_dir_lookup_result *lookup)
2069 int status = -ENOENT;
2071 mlog(0, "name=%.*s, blkno=%p, inode=%llu\n", namelen, name, blkno,
2072 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2074 status = ocfs2_find_entry(name, namelen, inode, lookup);
2078 *blkno = le64_to_cpu(lookup->dl_entry->inode);
2087 * Convenience function for callers which just want the block number
2088 * mapped to a name and don't require the full dirent info, etc.
2090 int ocfs2_lookup_ino_from_name(struct inode *dir, const char *name,
2091 int namelen, u64 *blkno)
2094 struct ocfs2_dir_lookup_result lookup = { NULL, };
2096 ret = ocfs2_find_files_on_disk(name, namelen, blkno, dir, &lookup);
2097 ocfs2_free_dir_lookup_result(&lookup);
2102 /* Check for a name within a directory.
2104 * Return 0 if the name does not exist
2105 * Return -EEXIST if the directory contains the name
2107 * Callers should have i_mutex + a cluster lock on dir
2109 int ocfs2_check_dir_for_entry(struct inode *dir,
2114 struct ocfs2_dir_lookup_result lookup = { NULL, };
2116 mlog_entry("dir %llu, name '%.*s'\n",
2117 (unsigned long long)OCFS2_I(dir)->ip_blkno, namelen, name);
2120 if (ocfs2_find_entry(name, namelen, dir, &lookup) == 0)
2125 ocfs2_free_dir_lookup_result(&lookup);
2131 struct ocfs2_empty_dir_priv {
2133 unsigned seen_dot_dot;
2134 unsigned seen_other;
2137 static int ocfs2_empty_dir_filldir(void *priv, const char *name, int name_len,
2138 loff_t pos, u64 ino, unsigned type)
2140 struct ocfs2_empty_dir_priv *p = priv;
2143 * Check the positions of "." and ".." records to be sure
2144 * they're in the correct place.
2146 * Indexed directories don't need to proceed past the first
2147 * two entries, so we end the scan after seeing '..'. Despite
2148 * that, we allow the scan to proceed In the event that we
2149 * have a corrupted indexed directory (no dot or dot dot
2150 * entries). This allows us to double check for existing
2151 * entries which might not have been found in the index.
2153 if (name_len == 1 && !strncmp(".", name, 1) && pos == 0) {
2158 if (name_len == 2 && !strncmp("..", name, 2) &&
2159 pos == OCFS2_DIR_REC_LEN(1)) {
2160 p->seen_dot_dot = 1;
2162 if (p->dx_dir && p->seen_dot)
2172 static int ocfs2_empty_dir_dx(struct inode *inode,
2173 struct ocfs2_empty_dir_priv *priv)
2176 struct buffer_head *di_bh = NULL;
2177 struct buffer_head *dx_root_bh = NULL;
2178 struct ocfs2_dinode *di;
2179 struct ocfs2_dx_root_block *dx_root;
2183 ret = ocfs2_read_inode_block(inode, &di_bh);
2188 di = (struct ocfs2_dinode *)di_bh->b_data;
2190 ret = ocfs2_read_dx_root(inode, di, &dx_root_bh);
2195 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
2197 if (le32_to_cpu(dx_root->dr_num_entries) != 2)
2198 priv->seen_other = 1;
2207 * routine to check that the specified directory is empty (for rmdir)
2209 * Returns 1 if dir is empty, zero otherwise.
2211 * XXX: This is a performance problem for unindexed directories.
2213 int ocfs2_empty_dir(struct inode *inode)
2217 struct ocfs2_empty_dir_priv priv;
2219 memset(&priv, 0, sizeof(priv));
2221 if (ocfs2_dir_indexed(inode)) {
2222 ret = ocfs2_empty_dir_dx(inode, &priv);
2226 * We still run ocfs2_dir_foreach to get the checks
2231 ret = ocfs2_dir_foreach(inode, &start, &priv, ocfs2_empty_dir_filldir);
2235 if (!priv.seen_dot || !priv.seen_dot_dot) {
2236 mlog(ML_ERROR, "bad directory (dir #%llu) - no `.' or `..'\n",
2237 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2239 * XXX: Is it really safe to allow an unlink to continue?
2244 return !priv.seen_other;
2248 * Fills "." and ".." dirents in a new directory block. Returns dirent for
2249 * "..", which might be used during creation of a directory with a trailing
2250 * header. It is otherwise safe to ignore the return code.
2252 static struct ocfs2_dir_entry *ocfs2_fill_initial_dirents(struct inode *inode,
2253 struct inode *parent,
2257 struct ocfs2_dir_entry *de = (struct ocfs2_dir_entry *)start;
2259 de->inode = cpu_to_le64(OCFS2_I(inode)->ip_blkno);
2262 cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
2263 strcpy(de->name, ".");
2264 ocfs2_set_de_type(de, S_IFDIR);
2266 de = (struct ocfs2_dir_entry *) ((char *)de + le16_to_cpu(de->rec_len));
2267 de->inode = cpu_to_le64(OCFS2_I(parent)->ip_blkno);
2268 de->rec_len = cpu_to_le16(size - OCFS2_DIR_REC_LEN(1));
2270 strcpy(de->name, "..");
2271 ocfs2_set_de_type(de, S_IFDIR);
2277 * This works together with code in ocfs2_mknod_locked() which sets
2278 * the inline-data flag and initializes the inline-data section.
2280 static int ocfs2_fill_new_dir_id(struct ocfs2_super *osb,
2282 struct inode *parent,
2283 struct inode *inode,
2284 struct buffer_head *di_bh)
2287 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2288 struct ocfs2_inline_data *data = &di->id2.i_data;
2289 unsigned int size = le16_to_cpu(data->id_count);
2291 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2292 OCFS2_JOURNAL_ACCESS_WRITE);
2298 ocfs2_fill_initial_dirents(inode, parent, data->id_data, size);
2300 ocfs2_journal_dirty(handle, di_bh);
2306 i_size_write(inode, size);
2308 inode->i_blocks = ocfs2_inode_sector_count(inode);
2310 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
2318 static int ocfs2_fill_new_dir_el(struct ocfs2_super *osb,
2320 struct inode *parent,
2321 struct inode *inode,
2322 struct buffer_head *fe_bh,
2323 struct ocfs2_alloc_context *data_ac,
2324 struct buffer_head **ret_new_bh)
2327 unsigned int size = osb->sb->s_blocksize;
2328 struct buffer_head *new_bh = NULL;
2329 struct ocfs2_dir_entry *de;
2333 if (ocfs2_new_dir_wants_trailer(inode))
2334 size = ocfs2_dir_trailer_blk_off(parent->i_sb);
2336 status = ocfs2_do_extend_dir(osb->sb, handle, inode, fe_bh,
2337 data_ac, NULL, &new_bh);
2343 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2345 status = ocfs2_journal_access_db(handle, INODE_CACHE(inode), new_bh,
2346 OCFS2_JOURNAL_ACCESS_CREATE);
2351 memset(new_bh->b_data, 0, osb->sb->s_blocksize);
2353 de = ocfs2_fill_initial_dirents(inode, parent, new_bh->b_data, size);
2354 if (ocfs2_new_dir_wants_trailer(inode)) {
2355 int size = le16_to_cpu(de->rec_len);
2358 * Figure out the size of the hole left over after
2359 * insertion of '.' and '..'. The trailer wants this
2362 size -= OCFS2_DIR_REC_LEN(2);
2363 size -= sizeof(struct ocfs2_dir_block_trailer);
2365 ocfs2_init_dir_trailer(inode, new_bh, size);
2368 status = ocfs2_journal_dirty(handle, new_bh);
2374 i_size_write(inode, inode->i_sb->s_blocksize);
2376 inode->i_blocks = ocfs2_inode_sector_count(inode);
2377 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
2385 *ret_new_bh = new_bh;
2395 static int ocfs2_dx_dir_attach_index(struct ocfs2_super *osb,
2396 handle_t *handle, struct inode *dir,
2397 struct buffer_head *di_bh,
2398 struct buffer_head *dirdata_bh,
2399 struct ocfs2_alloc_context *meta_ac,
2400 int dx_inline, u32 num_entries,
2401 struct buffer_head **ret_dx_root_bh)
2404 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
2405 u16 dr_suballoc_bit;
2407 unsigned int num_bits;
2408 struct buffer_head *dx_root_bh = NULL;
2409 struct ocfs2_dx_root_block *dx_root;
2410 struct ocfs2_dir_block_trailer *trailer =
2411 ocfs2_trailer_from_bh(dirdata_bh, dir->i_sb);
2413 ret = ocfs2_claim_metadata(osb, handle, meta_ac, 1, &dr_suballoc_bit,
2414 &num_bits, &dr_blkno);
2420 mlog(0, "Dir %llu, attach new index block: %llu\n",
2421 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2422 (unsigned long long)dr_blkno);
2424 dx_root_bh = sb_getblk(osb->sb, dr_blkno);
2425 if (dx_root_bh == NULL) {
2429 ocfs2_set_new_buffer_uptodate(INODE_CACHE(dir), dx_root_bh);
2431 ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
2432 OCFS2_JOURNAL_ACCESS_CREATE);
2438 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
2439 memset(dx_root, 0, osb->sb->s_blocksize);
2440 strcpy(dx_root->dr_signature, OCFS2_DX_ROOT_SIGNATURE);
2441 dx_root->dr_suballoc_slot = cpu_to_le16(osb->slot_num);
2442 dx_root->dr_suballoc_bit = cpu_to_le16(dr_suballoc_bit);
2443 dx_root->dr_fs_generation = cpu_to_le32(osb->fs_generation);
2444 dx_root->dr_blkno = cpu_to_le64(dr_blkno);
2445 dx_root->dr_dir_blkno = cpu_to_le64(OCFS2_I(dir)->ip_blkno);
2446 dx_root->dr_num_entries = cpu_to_le32(num_entries);
2447 if (le16_to_cpu(trailer->db_free_rec_len))
2448 dx_root->dr_free_blk = cpu_to_le64(dirdata_bh->b_blocknr);
2450 dx_root->dr_free_blk = cpu_to_le64(0);
2453 dx_root->dr_flags |= OCFS2_DX_FLAG_INLINE;
2454 dx_root->dr_entries.de_count =
2455 cpu_to_le16(ocfs2_dx_entries_per_root(osb->sb));
2457 dx_root->dr_list.l_count =
2458 cpu_to_le16(ocfs2_extent_recs_per_dx_root(osb->sb));
2461 ret = ocfs2_journal_dirty(handle, dx_root_bh);
2465 ret = ocfs2_journal_access_di(handle, INODE_CACHE(dir), di_bh,
2466 OCFS2_JOURNAL_ACCESS_CREATE);
2472 di->i_dx_root = cpu_to_le64(dr_blkno);
2474 OCFS2_I(dir)->ip_dyn_features |= OCFS2_INDEXED_DIR_FL;
2475 di->i_dyn_features = cpu_to_le16(OCFS2_I(dir)->ip_dyn_features);
2477 ret = ocfs2_journal_dirty(handle, di_bh);
2481 *ret_dx_root_bh = dx_root_bh;
2489 static int ocfs2_dx_dir_format_cluster(struct ocfs2_super *osb,
2490 handle_t *handle, struct inode *dir,
2491 struct buffer_head **dx_leaves,
2492 int num_dx_leaves, u64 start_blk)
2495 struct ocfs2_dx_leaf *dx_leaf;
2496 struct buffer_head *bh;
2498 for (i = 0; i < num_dx_leaves; i++) {
2499 bh = sb_getblk(osb->sb, start_blk + i);
2506 ocfs2_set_new_buffer_uptodate(INODE_CACHE(dir), bh);
2508 ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir), bh,
2509 OCFS2_JOURNAL_ACCESS_CREATE);
2515 dx_leaf = (struct ocfs2_dx_leaf *) bh->b_data;
2517 memset(dx_leaf, 0, osb->sb->s_blocksize);
2518 strcpy(dx_leaf->dl_signature, OCFS2_DX_LEAF_SIGNATURE);
2519 dx_leaf->dl_fs_generation = cpu_to_le32(osb->fs_generation);
2520 dx_leaf->dl_blkno = cpu_to_le64(bh->b_blocknr);
2521 dx_leaf->dl_list.de_count =
2522 cpu_to_le16(ocfs2_dx_entries_per_leaf(osb->sb));
2525 "Dir %llu, format dx_leaf: %llu, entry count: %u\n",
2526 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2527 (unsigned long long)bh->b_blocknr,
2528 le16_to_cpu(dx_leaf->dl_list.de_count));
2530 ocfs2_journal_dirty(handle, bh);
2539 * Allocates and formats a new cluster for use in an indexed dir
2540 * leaf. This version will not do the extent insert, so that it can be
2541 * used by operations which need careful ordering.
2543 static int __ocfs2_dx_dir_new_cluster(struct inode *dir,
2544 u32 cpos, handle_t *handle,
2545 struct ocfs2_alloc_context *data_ac,
2546 struct buffer_head **dx_leaves,
2547 int num_dx_leaves, u64 *ret_phys_blkno)
2552 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2555 * XXX: For create, this should claim cluster for the index
2556 * *before* the unindexed insert so that we have a better
2557 * chance of contiguousness as the directory grows in number
2560 ret = __ocfs2_claim_clusters(osb, handle, data_ac, 1, 1, &phys, &num);
2567 * Format the new cluster first. That way, we're inserting
2570 phys_blkno = ocfs2_clusters_to_blocks(osb->sb, phys);
2571 ret = ocfs2_dx_dir_format_cluster(osb, handle, dir, dx_leaves,
2572 num_dx_leaves, phys_blkno);
2578 *ret_phys_blkno = phys_blkno;
2583 static int ocfs2_dx_dir_new_cluster(struct inode *dir,
2584 struct ocfs2_extent_tree *et,
2585 u32 cpos, handle_t *handle,
2586 struct ocfs2_alloc_context *data_ac,
2587 struct ocfs2_alloc_context *meta_ac,
2588 struct buffer_head **dx_leaves,
2593 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2595 ret = __ocfs2_dx_dir_new_cluster(dir, cpos, handle, data_ac, dx_leaves,
2596 num_dx_leaves, &phys_blkno);
2602 ret = ocfs2_insert_extent(osb, handle, dir, et, cpos, phys_blkno, 1, 0,
2610 static struct buffer_head **ocfs2_dx_dir_kmalloc_leaves(struct super_block *sb,
2611 int *ret_num_leaves)
2613 int num_dx_leaves = ocfs2_clusters_to_blocks(sb, 1);
2614 struct buffer_head **dx_leaves;
2616 dx_leaves = kcalloc(num_dx_leaves, sizeof(struct buffer_head *),
2618 if (dx_leaves && ret_num_leaves)
2619 *ret_num_leaves = num_dx_leaves;
2624 static int ocfs2_fill_new_dir_dx(struct ocfs2_super *osb,
2626 struct inode *parent,
2627 struct inode *inode,
2628 struct buffer_head *di_bh,
2629 struct ocfs2_alloc_context *data_ac,
2630 struct ocfs2_alloc_context *meta_ac)
2633 struct buffer_head *leaf_bh = NULL;
2634 struct buffer_head *dx_root_bh = NULL;
2635 struct ocfs2_dx_hinfo hinfo;
2636 struct ocfs2_dx_root_block *dx_root;
2637 struct ocfs2_dx_entry_list *entry_list;
2640 * Our strategy is to create the directory as though it were
2641 * unindexed, then add the index block. This works with very
2642 * little complication since the state of a new directory is a
2643 * very well known quantity.
2645 * Essentially, we have two dirents ("." and ".."), in the 1st
2646 * block which need indexing. These are easily inserted into
2650 ret = ocfs2_fill_new_dir_el(osb, handle, parent, inode, di_bh,
2657 ret = ocfs2_dx_dir_attach_index(osb, handle, inode, di_bh, leaf_bh,
2658 meta_ac, 1, 2, &dx_root_bh);
2663 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
2664 entry_list = &dx_root->dr_entries;
2666 /* Buffer has been journaled for us by ocfs2_dx_dir_attach_index */
2667 ocfs2_dx_dir_name_hash(inode, ".", 1, &hinfo);
2668 ocfs2_dx_entry_list_insert(entry_list, &hinfo, leaf_bh->b_blocknr);
2670 ocfs2_dx_dir_name_hash(inode, "..", 2, &hinfo);
2671 ocfs2_dx_entry_list_insert(entry_list, &hinfo, leaf_bh->b_blocknr);
2679 int ocfs2_fill_new_dir(struct ocfs2_super *osb,
2681 struct inode *parent,
2682 struct inode *inode,
2683 struct buffer_head *fe_bh,
2684 struct ocfs2_alloc_context *data_ac,
2685 struct ocfs2_alloc_context *meta_ac)
2688 BUG_ON(!ocfs2_supports_inline_data(osb) && data_ac == NULL);
2690 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2691 return ocfs2_fill_new_dir_id(osb, handle, parent, inode, fe_bh);
2693 if (ocfs2_supports_indexed_dirs(osb))
2694 return ocfs2_fill_new_dir_dx(osb, handle, parent, inode, fe_bh,
2697 return ocfs2_fill_new_dir_el(osb, handle, parent, inode, fe_bh,
2701 static int ocfs2_dx_dir_index_block(struct inode *dir,
2703 struct buffer_head **dx_leaves,
2705 u32 *num_dx_entries,
2706 struct buffer_head *dirent_bh)
2708 int ret = 0, namelen, i;
2709 char *de_buf, *limit;
2710 struct ocfs2_dir_entry *de;
2711 struct buffer_head *dx_leaf_bh;
2712 struct ocfs2_dx_hinfo hinfo;
2713 u64 dirent_blk = dirent_bh->b_blocknr;
2715 de_buf = dirent_bh->b_data;
2716 limit = de_buf + dir->i_sb->s_blocksize;
2718 while (de_buf < limit) {
2719 de = (struct ocfs2_dir_entry *)de_buf;
2721 namelen = de->name_len;
2722 if (!namelen || !de->inode)
2725 ocfs2_dx_dir_name_hash(dir, de->name, namelen, &hinfo);
2727 i = ocfs2_dx_dir_hash_idx(OCFS2_SB(dir->i_sb), &hinfo);
2728 dx_leaf_bh = dx_leaves[i];
2730 ret = __ocfs2_dx_dir_leaf_insert(dir, handle, &hinfo,
2731 dirent_blk, dx_leaf_bh);
2737 *num_dx_entries = *num_dx_entries + 1;
2740 de_buf += le16_to_cpu(de->rec_len);
2748 * XXX: This expects dx_root_bh to already be part of the transaction.
2750 static void ocfs2_dx_dir_index_root_block(struct inode *dir,
2751 struct buffer_head *dx_root_bh,
2752 struct buffer_head *dirent_bh)
2754 char *de_buf, *limit;
2755 struct ocfs2_dx_root_block *dx_root;
2756 struct ocfs2_dir_entry *de;
2757 struct ocfs2_dx_hinfo hinfo;
2758 u64 dirent_blk = dirent_bh->b_blocknr;
2760 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
2762 de_buf = dirent_bh->b_data;
2763 limit = de_buf + dir->i_sb->s_blocksize;
2765 while (de_buf < limit) {
2766 de = (struct ocfs2_dir_entry *)de_buf;
2768 if (!de->name_len || !de->inode)
2771 ocfs2_dx_dir_name_hash(dir, de->name, de->name_len, &hinfo);
2774 "dir: %llu, major: 0x%x minor: 0x%x, index: %u, name: %.*s\n",
2775 (unsigned long long)dir->i_ino, hinfo.major_hash,
2777 le16_to_cpu(dx_root->dr_entries.de_num_used),
2778 de->name_len, de->name);
2780 ocfs2_dx_entry_list_insert(&dx_root->dr_entries, &hinfo,
2783 le32_add_cpu(&dx_root->dr_num_entries, 1);
2785 de_buf += le16_to_cpu(de->rec_len);
2790 * Count the number of inline directory entries in di_bh and compare
2791 * them against the number of entries we can hold in an inline dx root
2794 static int ocfs2_new_dx_should_be_inline(struct inode *dir,
2795 struct buffer_head *di_bh)
2797 int dirent_count = 0;
2798 char *de_buf, *limit;
2799 struct ocfs2_dir_entry *de;
2800 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2802 de_buf = di->id2.i_data.id_data;
2803 limit = de_buf + i_size_read(dir);
2805 while (de_buf < limit) {
2806 de = (struct ocfs2_dir_entry *)de_buf;
2808 if (de->name_len && de->inode)
2811 de_buf += le16_to_cpu(de->rec_len);
2814 /* We are careful to leave room for one extra record. */
2815 return dirent_count < ocfs2_dx_entries_per_root(dir->i_sb);
2819 * Expand rec_len of the rightmost dirent in a directory block so that it
2820 * contains the end of our valid space for dirents. We do this during
2821 * expansion from an inline directory to one with extents. The first dir block
2822 * in that case is taken from the inline data portion of the inode block.
2824 * This will also return the largest amount of contiguous space for a dirent
2825 * in the block. That value is *not* necessarily the last dirent, even after
2826 * expansion. The directory indexing code wants this value for free space
2827 * accounting. We do this here since we're already walking the entire dir
2830 * We add the dir trailer if this filesystem wants it.
2832 static unsigned int ocfs2_expand_last_dirent(char *start, unsigned int old_size,
2835 struct super_block *sb = dir->i_sb;
2836 struct ocfs2_dir_entry *de;
2837 struct ocfs2_dir_entry *prev_de;
2838 char *de_buf, *limit;
2839 unsigned int new_size = sb->s_blocksize;
2840 unsigned int bytes, this_hole;
2841 unsigned int largest_hole = 0;
2843 if (ocfs2_new_dir_wants_trailer(dir))
2844 new_size = ocfs2_dir_trailer_blk_off(sb);
2846 bytes = new_size - old_size;
2848 limit = start + old_size;
2850 de = (struct ocfs2_dir_entry *)de_buf;
2852 this_hole = ocfs2_figure_dirent_hole(de);
2853 if (this_hole > largest_hole)
2854 largest_hole = this_hole;
2857 de_buf += le16_to_cpu(de->rec_len);
2858 de = (struct ocfs2_dir_entry *)de_buf;
2859 } while (de_buf < limit);
2861 le16_add_cpu(&prev_de->rec_len, bytes);
2863 /* We need to double check this after modification of the final
2865 this_hole = ocfs2_figure_dirent_hole(prev_de);
2866 if (this_hole > largest_hole)
2867 largest_hole = this_hole;
2869 if (largest_hole >= OCFS2_DIR_MIN_REC_LEN)
2870 return largest_hole;
2875 * We allocate enough clusters to fulfill "blocks_wanted", but set
2876 * i_size to exactly one block. Ocfs2_extend_dir() will handle the
2877 * rest automatically for us.
2879 * *first_block_bh is a pointer to the 1st data block allocated to the
2882 static int ocfs2_expand_inline_dir(struct inode *dir, struct buffer_head *di_bh,
2883 unsigned int blocks_wanted,
2884 struct ocfs2_dir_lookup_result *lookup,
2885 struct buffer_head **first_block_bh)
2887 u32 alloc, dx_alloc, bit_off, len, num_dx_entries = 0;
2888 struct super_block *sb = dir->i_sb;
2889 int ret, i, num_dx_leaves = 0, dx_inline = 0,
2890 credits = ocfs2_inline_to_extents_credits(sb);
2891 u64 dx_insert_blkno, blkno,
2892 bytes = blocks_wanted << sb->s_blocksize_bits;
2893 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2894 struct ocfs2_inode_info *oi = OCFS2_I(dir);
2895 struct ocfs2_alloc_context *data_ac;
2896 struct ocfs2_alloc_context *meta_ac = NULL;
2897 struct buffer_head *dirdata_bh = NULL;
2898 struct buffer_head *dx_root_bh = NULL;
2899 struct buffer_head **dx_leaves = NULL;
2900 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2902 struct ocfs2_extent_tree et;
2903 struct ocfs2_extent_tree dx_et;
2904 int did_quota = 0, bytes_allocated = 0;
2906 ocfs2_init_dinode_extent_tree(&et, dir, di_bh);
2908 alloc = ocfs2_clusters_for_bytes(sb, bytes);
2911 down_write(&oi->ip_alloc_sem);
2913 if (ocfs2_supports_indexed_dirs(osb)) {
2914 credits += ocfs2_add_dir_index_credits(sb);
2916 dx_inline = ocfs2_new_dx_should_be_inline(dir, di_bh);
2918 /* Add one more cluster for an index leaf */
2920 dx_leaves = ocfs2_dx_dir_kmalloc_leaves(sb,
2929 /* This gets us the dx_root */
2930 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
2938 * We should never need more than 2 clusters for the unindexed
2939 * tree - maximum dirent size is far less than one block. In
2940 * fact, the only time we'd need more than one cluster is if
2941 * blocksize == clustersize and the dirent won't fit in the
2942 * extra space that the expansion to a single block gives. As
2943 * of today, that only happens on 4k/4k file systems.
2947 ret = ocfs2_reserve_clusters(osb, alloc + dx_alloc, &data_ac);
2954 * Prepare for worst case allocation scenario of two separate
2955 * extents in the unindexed tree.
2958 credits += OCFS2_SUBALLOC_ALLOC;
2960 handle = ocfs2_start_trans(osb, credits);
2961 if (IS_ERR(handle)) {
2962 ret = PTR_ERR(handle);
2967 if (vfs_dq_alloc_space_nodirty(dir,
2968 ocfs2_clusters_to_bytes(osb->sb,
2969 alloc + dx_alloc))) {
2975 if (ocfs2_supports_indexed_dirs(osb) && !dx_inline) {
2977 * Allocate our index cluster first, to maximize the
2978 * possibility that unindexed leaves grow
2981 ret = __ocfs2_dx_dir_new_cluster(dir, 0, handle, data_ac,
2982 dx_leaves, num_dx_leaves,
2988 bytes_allocated += ocfs2_clusters_to_bytes(dir->i_sb, 1);
2992 * Try to claim as many clusters as the bitmap can give though
2993 * if we only get one now, that's enough to continue. The rest
2994 * will be claimed after the conversion to extents.
2996 ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off, &len);
3001 bytes_allocated += ocfs2_clusters_to_bytes(dir->i_sb, 1);
3004 * Operations are carefully ordered so that we set up the new
3005 * data block first. The conversion from inline data to
3008 blkno = ocfs2_clusters_to_blocks(dir->i_sb, bit_off);
3009 dirdata_bh = sb_getblk(sb, blkno);
3016 ocfs2_set_new_buffer_uptodate(INODE_CACHE(dir), dirdata_bh);
3018 ret = ocfs2_journal_access_db(handle, INODE_CACHE(dir), dirdata_bh,
3019 OCFS2_JOURNAL_ACCESS_CREATE);
3025 memcpy(dirdata_bh->b_data, di->id2.i_data.id_data, i_size_read(dir));
3026 memset(dirdata_bh->b_data + i_size_read(dir), 0,
3027 sb->s_blocksize - i_size_read(dir));
3028 i = ocfs2_expand_last_dirent(dirdata_bh->b_data, i_size_read(dir), dir);
3029 if (ocfs2_new_dir_wants_trailer(dir)) {
3031 * Prepare the dir trailer up front. It will otherwise look
3032 * like a valid dirent. Even if inserting the index fails
3033 * (unlikely), then all we'll have done is given first dir
3034 * block a small amount of fragmentation.
3036 ocfs2_init_dir_trailer(dir, dirdata_bh, i);
3039 ret = ocfs2_journal_dirty(handle, dirdata_bh);
3045 if (ocfs2_supports_indexed_dirs(osb) && !dx_inline) {
3047 * Dx dirs with an external cluster need to do this up
3048 * front. Inline dx root's get handled later, after
3049 * we've allocated our root block. We get passed back
3050 * a total number of items so that dr_num_entries can
3051 * be correctly set once the dx_root has been
3054 ret = ocfs2_dx_dir_index_block(dir, handle, dx_leaves,
3055 num_dx_leaves, &num_dx_entries,
3064 * Set extent, i_size, etc on the directory. After this, the
3065 * inode should contain the same exact dirents as before and
3066 * be fully accessible from system calls.
3068 * We let the later dirent insert modify c/mtime - to the user
3069 * the data hasn't changed.
3071 ret = ocfs2_journal_access_di(handle, INODE_CACHE(dir), di_bh,
3072 OCFS2_JOURNAL_ACCESS_CREATE);
3078 spin_lock(&oi->ip_lock);
3079 oi->ip_dyn_features &= ~OCFS2_INLINE_DATA_FL;
3080 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
3081 spin_unlock(&oi->ip_lock);
3083 ocfs2_dinode_new_extent_list(dir, di);
3085 i_size_write(dir, sb->s_blocksize);
3086 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3088 di->i_size = cpu_to_le64(sb->s_blocksize);
3089 di->i_ctime = di->i_mtime = cpu_to_le64(dir->i_ctime.tv_sec);
3090 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(dir->i_ctime.tv_nsec);
3093 * This should never fail as our extent list is empty and all
3094 * related blocks have been journaled already.
3096 ret = ocfs2_insert_extent(osb, handle, dir, &et, 0, blkno, len,
3104 * Set i_blocks after the extent insert for the most up to
3105 * date ip_clusters value.
3107 dir->i_blocks = ocfs2_inode_sector_count(dir);
3109 ret = ocfs2_journal_dirty(handle, di_bh);
3115 if (ocfs2_supports_indexed_dirs(osb)) {
3116 ret = ocfs2_dx_dir_attach_index(osb, handle, dir, di_bh,
3117 dirdata_bh, meta_ac, dx_inline,
3118 num_dx_entries, &dx_root_bh);
3125 ocfs2_dx_dir_index_root_block(dir, dx_root_bh,
3128 ocfs2_init_dx_root_extent_tree(&dx_et, dir, dx_root_bh);
3129 ret = ocfs2_insert_extent(osb, handle, dir, &dx_et, 0,
3130 dx_insert_blkno, 1, 0, NULL);
3137 * We asked for two clusters, but only got one in the 1st
3138 * pass. Claim the 2nd cluster as a separate extent.
3141 ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off,
3147 blkno = ocfs2_clusters_to_blocks(dir->i_sb, bit_off);
3149 ret = ocfs2_insert_extent(osb, handle, dir, &et, 1,
3150 blkno, len, 0, NULL);
3155 bytes_allocated += ocfs2_clusters_to_bytes(dir->i_sb, 1);
3158 *first_block_bh = dirdata_bh;
3160 if (ocfs2_supports_indexed_dirs(osb)) {
3165 * We need to return the correct block within the
3166 * cluster which should hold our entry.
3168 off = ocfs2_dx_dir_hash_idx(OCFS2_SB(dir->i_sb),
3170 get_bh(dx_leaves[off]);
3171 lookup->dl_dx_leaf_bh = dx_leaves[off];
3173 lookup->dl_dx_root_bh = dx_root_bh;
3178 if (ret < 0 && did_quota)
3179 vfs_dq_free_space_nodirty(dir, bytes_allocated);
3181 ocfs2_commit_trans(osb, handle);
3184 up_write(&oi->ip_alloc_sem);
3186 ocfs2_free_alloc_context(data_ac);
3188 ocfs2_free_alloc_context(meta_ac);
3191 for (i = 0; i < num_dx_leaves; i++)
3192 brelse(dx_leaves[i]);
3202 /* returns a bh of the 1st new block in the allocation. */
3203 static int ocfs2_do_extend_dir(struct super_block *sb,
3206 struct buffer_head *parent_fe_bh,
3207 struct ocfs2_alloc_context *data_ac,
3208 struct ocfs2_alloc_context *meta_ac,
3209 struct buffer_head **new_bh)
3212 int extend, did_quota = 0;
3213 u64 p_blkno, v_blkno;
3215 spin_lock(&OCFS2_I(dir)->ip_lock);
3216 extend = (i_size_read(dir) == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters));
3217 spin_unlock(&OCFS2_I(dir)->ip_lock);
3220 u32 offset = OCFS2_I(dir)->ip_clusters;
3222 if (vfs_dq_alloc_space_nodirty(dir,
3223 ocfs2_clusters_to_bytes(sb, 1))) {
3229 status = ocfs2_add_inode_data(OCFS2_SB(sb), dir, &offset,
3230 1, 0, parent_fe_bh, handle,
3231 data_ac, meta_ac, NULL);
3232 BUG_ON(status == -EAGAIN);
3239 v_blkno = ocfs2_blocks_for_bytes(sb, i_size_read(dir));
3240 status = ocfs2_extent_map_get_blocks(dir, v_blkno, &p_blkno, NULL, NULL);
3246 *new_bh = sb_getblk(sb, p_blkno);
3254 if (did_quota && status < 0)
3255 vfs_dq_free_space_nodirty(dir, ocfs2_clusters_to_bytes(sb, 1));
3261 * Assumes you already have a cluster lock on the directory.
3263 * 'blocks_wanted' is only used if we have an inline directory which
3264 * is to be turned into an extent based one. The size of the dirent to
3265 * insert might be larger than the space gained by growing to just one
3266 * block, so we may have to grow the inode by two blocks in that case.
3268 * If the directory is already indexed, dx_root_bh must be provided.
3270 static int ocfs2_extend_dir(struct ocfs2_super *osb,
3272 struct buffer_head *parent_fe_bh,
3273 unsigned int blocks_wanted,
3274 struct ocfs2_dir_lookup_result *lookup,
3275 struct buffer_head **new_de_bh)
3278 int credits, num_free_extents, drop_alloc_sem = 0;
3280 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
3281 struct ocfs2_extent_list *el = &fe->id2.i_list;
3282 struct ocfs2_alloc_context *data_ac = NULL;
3283 struct ocfs2_alloc_context *meta_ac = NULL;
3284 handle_t *handle = NULL;
3285 struct buffer_head *new_bh = NULL;
3286 struct ocfs2_dir_entry * de;
3287 struct super_block *sb = osb->sb;
3288 struct ocfs2_extent_tree et;
3289 struct buffer_head *dx_root_bh = lookup->dl_dx_root_bh;
3293 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
3295 * This would be a code error as an inline directory should
3296 * never have an index root.
3300 status = ocfs2_expand_inline_dir(dir, parent_fe_bh,
3301 blocks_wanted, lookup,
3308 /* Expansion from inline to an indexed directory will
3309 * have given us this. */
3310 dx_root_bh = lookup->dl_dx_root_bh;
3312 if (blocks_wanted == 1) {
3314 * If the new dirent will fit inside the space
3315 * created by pushing out to one block, then
3316 * we can complete the operation
3317 * here. Otherwise we have to expand i_size
3318 * and format the 2nd block below.
3320 BUG_ON(new_bh == NULL);
3325 * Get rid of 'new_bh' - we want to format the 2nd
3326 * data block and return that instead.
3331 down_write(&OCFS2_I(dir)->ip_alloc_sem);
3333 dir_i_size = i_size_read(dir);
3334 credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS;
3338 down_write(&OCFS2_I(dir)->ip_alloc_sem);
3340 dir_i_size = i_size_read(dir);
3341 mlog(0, "extending dir %llu (i_size = %lld)\n",
3342 (unsigned long long)OCFS2_I(dir)->ip_blkno, dir_i_size);
3344 /* dir->i_size is always block aligned. */
3345 spin_lock(&OCFS2_I(dir)->ip_lock);
3346 if (dir_i_size == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters)) {
3347 spin_unlock(&OCFS2_I(dir)->ip_lock);
3348 ocfs2_init_dinode_extent_tree(&et, dir, parent_fe_bh);
3349 num_free_extents = ocfs2_num_free_extents(osb, dir, &et);
3350 if (num_free_extents < 0) {
3351 status = num_free_extents;
3356 if (!num_free_extents) {
3357 status = ocfs2_reserve_new_metadata(osb, el, &meta_ac);
3359 if (status != -ENOSPC)
3365 status = ocfs2_reserve_clusters(osb, 1, &data_ac);
3367 if (status != -ENOSPC)
3372 credits = ocfs2_calc_extend_credits(sb, el, 1);
3374 spin_unlock(&OCFS2_I(dir)->ip_lock);
3375 credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS;
3379 if (ocfs2_dir_indexed(dir))
3380 credits++; /* For attaching the new dirent block to the
3383 handle = ocfs2_start_trans(osb, credits);
3384 if (IS_ERR(handle)) {
3385 status = PTR_ERR(handle);
3391 status = ocfs2_do_extend_dir(osb->sb, handle, dir, parent_fe_bh,
3392 data_ac, meta_ac, &new_bh);
3398 ocfs2_set_new_buffer_uptodate(INODE_CACHE(dir), new_bh);
3400 status = ocfs2_journal_access_db(handle, INODE_CACHE(dir), new_bh,
3401 OCFS2_JOURNAL_ACCESS_CREATE);
3406 memset(new_bh->b_data, 0, sb->s_blocksize);
3408 de = (struct ocfs2_dir_entry *) new_bh->b_data;
3410 if (ocfs2_supports_dir_trailer(dir)) {
3411 de->rec_len = cpu_to_le16(ocfs2_dir_trailer_blk_off(sb));
3413 ocfs2_init_dir_trailer(dir, new_bh, le16_to_cpu(de->rec_len));
3415 if (ocfs2_dir_indexed(dir)) {
3416 status = ocfs2_dx_dir_link_trailer(dir, handle,
3417 dx_root_bh, new_bh);
3424 de->rec_len = cpu_to_le16(sb->s_blocksize);
3426 status = ocfs2_journal_dirty(handle, new_bh);
3432 dir_i_size += dir->i_sb->s_blocksize;
3433 i_size_write(dir, dir_i_size);
3434 dir->i_blocks = ocfs2_inode_sector_count(dir);
3435 status = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
3442 *new_de_bh = new_bh;
3446 ocfs2_commit_trans(osb, handle);
3448 up_write(&OCFS2_I(dir)->ip_alloc_sem);
3451 ocfs2_free_alloc_context(data_ac);
3453 ocfs2_free_alloc_context(meta_ac);
3461 static int ocfs2_find_dir_space_id(struct inode *dir, struct buffer_head *di_bh,
3462 const char *name, int namelen,
3463 struct buffer_head **ret_de_bh,
3464 unsigned int *blocks_wanted)
3467 struct super_block *sb = dir->i_sb;
3468 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3469 struct ocfs2_dir_entry *de, *last_de = NULL;
3470 char *de_buf, *limit;
3471 unsigned long offset = 0;
3472 unsigned int rec_len, new_rec_len, free_space = dir->i_sb->s_blocksize;
3475 * This calculates how many free bytes we'd have in block zero, should
3476 * this function force expansion to an extent tree.
3478 if (ocfs2_new_dir_wants_trailer(dir))
3479 free_space = ocfs2_dir_trailer_blk_off(sb) - i_size_read(dir);
3481 free_space = dir->i_sb->s_blocksize - i_size_read(dir);
3483 de_buf = di->id2.i_data.id_data;
3484 limit = de_buf + i_size_read(dir);
3485 rec_len = OCFS2_DIR_REC_LEN(namelen);
3487 while (de_buf < limit) {
3488 de = (struct ocfs2_dir_entry *)de_buf;
3490 if (!ocfs2_check_dir_entry(dir, de, di_bh, offset)) {
3494 if (ocfs2_match(namelen, name, de)) {
3499 * No need to check for a trailing dirent record here as
3500 * they're not used for inline dirs.
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. */
3513 de_buf += le16_to_cpu(de->rec_len);
3514 offset += le16_to_cpu(de->rec_len);
3518 * We're going to require expansion of the directory - figure
3519 * out how many blocks we'll need so that a place for the
3520 * dirent can be found.
3523 new_rec_len = le16_to_cpu(last_de->rec_len) + free_space;
3524 if (new_rec_len < (rec_len + OCFS2_DIR_REC_LEN(last_de->name_len)))
3532 static int ocfs2_find_dir_space_el(struct inode *dir, const char *name,
3533 int namelen, struct buffer_head **ret_de_bh)
3535 unsigned long offset;
3536 struct buffer_head *bh = NULL;
3537 unsigned short rec_len;
3538 struct ocfs2_dir_entry *de;
3539 struct super_block *sb = dir->i_sb;
3541 int blocksize = dir->i_sb->s_blocksize;
3543 status = ocfs2_read_dir_block(dir, 0, &bh, 0);
3549 rec_len = OCFS2_DIR_REC_LEN(namelen);
3551 de = (struct ocfs2_dir_entry *) bh->b_data;
3553 if ((char *)de >= sb->s_blocksize + bh->b_data) {
3557 if (i_size_read(dir) <= offset) {
3559 * Caller will have to expand this
3565 status = ocfs2_read_dir_block(dir,
3566 offset >> sb->s_blocksize_bits,
3572 /* move to next block */
3573 de = (struct ocfs2_dir_entry *) bh->b_data;
3575 if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
3579 if (ocfs2_match(namelen, name, de)) {
3584 if (ocfs2_skip_dir_trailer(dir, de, offset % blocksize,
3588 if (ocfs2_dirent_would_fit(de, rec_len)) {
3589 /* Ok, we found a spot. Return this bh and let
3590 * the caller actually fill it in. */
3597 offset += le16_to_cpu(de->rec_len);
3598 de = (struct ocfs2_dir_entry *)((char *) de + le16_to_cpu(de->rec_len));
3609 static int dx_leaf_sort_cmp(const void *a, const void *b)
3611 const struct ocfs2_dx_entry *entry1 = a;
3612 const struct ocfs2_dx_entry *entry2 = b;
3613 u32 major_hash1 = le32_to_cpu(entry1->dx_major_hash);
3614 u32 major_hash2 = le32_to_cpu(entry2->dx_major_hash);
3615 u32 minor_hash1 = le32_to_cpu(entry1->dx_minor_hash);
3616 u32 minor_hash2 = le32_to_cpu(entry2->dx_minor_hash);
3618 if (major_hash1 > major_hash2)
3620 if (major_hash1 < major_hash2)
3624 * It is not strictly necessary to sort by minor
3626 if (minor_hash1 > minor_hash2)
3628 if (minor_hash1 < minor_hash2)
3633 static void dx_leaf_sort_swap(void *a, void *b, int size)
3635 struct ocfs2_dx_entry *entry1 = a;
3636 struct ocfs2_dx_entry *entry2 = b;
3637 struct ocfs2_dx_entry tmp;
3639 BUG_ON(size != sizeof(*entry1));
3646 static int ocfs2_dx_leaf_same_major(struct ocfs2_dx_leaf *dx_leaf)
3648 struct ocfs2_dx_entry_list *dl_list = &dx_leaf->dl_list;
3649 int i, num = le16_to_cpu(dl_list->de_num_used);
3651 for (i = 0; i < (num - 1); i++) {
3652 if (le32_to_cpu(dl_list->de_entries[i].dx_major_hash) !=
3653 le32_to_cpu(dl_list->de_entries[i + 1].dx_major_hash))
3661 * Find the optimal value to split this leaf on. This expects the leaf
3662 * entries to be in sorted order.
3664 * leaf_cpos is the cpos of the leaf we're splitting. insert_hash is
3665 * the hash we want to insert.
3667 * This function is only concerned with the major hash - that which
3668 * determines which cluster an item belongs to.
3670 static int ocfs2_dx_dir_find_leaf_split(struct ocfs2_dx_leaf *dx_leaf,
3671 u32 leaf_cpos, u32 insert_hash,
3674 struct ocfs2_dx_entry_list *dl_list = &dx_leaf->dl_list;
3675 int i, num_used = le16_to_cpu(dl_list->de_num_used);
3679 * There's a couple rare, but nasty corner cases we have to
3680 * check for here. All of them involve a leaf where all value
3681 * have the same hash, which is what we look for first.
3683 * Most of the time, all of the above is false, and we simply
3684 * pick the median value for a split.
3686 allsame = ocfs2_dx_leaf_same_major(dx_leaf);
3688 u32 val = le32_to_cpu(dl_list->de_entries[0].dx_major_hash);
3690 if (val == insert_hash) {
3692 * No matter where we would choose to split,
3693 * the new entry would want to occupy the same
3694 * block as these. Since there's no space left
3695 * in their existing block, we know there
3696 * won't be space after the split.
3701 if (val == leaf_cpos) {
3703 * Because val is the same as leaf_cpos (which
3704 * is the smallest value this leaf can have),
3705 * yet is not equal to insert_hash, then we
3706 * know that insert_hash *must* be larger than
3707 * val (and leaf_cpos). At least cpos+1 in value.
3709 * We also know then, that there cannot be an
3710 * adjacent extent (otherwise we'd be looking
3711 * at it). Choosing this value gives us a
3712 * chance to get some contiguousness.
3714 *split_hash = leaf_cpos + 1;
3718 if (val > insert_hash) {
3720 * val can not be the same as insert hash, and
3721 * also must be larger than leaf_cpos. Also,
3722 * we know that there can't be a leaf between
3723 * cpos and val, otherwise the entries with
3724 * hash 'val' would be there.
3730 *split_hash = insert_hash;
3735 * Since the records are sorted and the checks above
3736 * guaranteed that not all records in this block are the same,
3737 * we simple travel forward, from the median, and pick the 1st
3738 * record whose value is larger than leaf_cpos.
3740 for (i = (num_used / 2); i < num_used; i++)
3741 if (le32_to_cpu(dl_list->de_entries[i].dx_major_hash) >
3745 BUG_ON(i == num_used); /* Should be impossible */
3746 *split_hash = le32_to_cpu(dl_list->de_entries[i].dx_major_hash);
3751 * Transfer all entries in orig_dx_leaves whose major hash is equal to or
3752 * larger than split_hash into new_dx_leaves. We use a temporary
3753 * buffer (tmp_dx_leaf) to make the changes to the original leaf blocks.
3755 * Since the block offset inside a leaf (cluster) is a constant mask
3756 * of minor_hash, we can optimize - an item at block offset X within
3757 * the original cluster, will be at offset X within the new cluster.
3759 static void ocfs2_dx_dir_transfer_leaf(struct inode *dir, u32 split_hash,
3761 struct ocfs2_dx_leaf *tmp_dx_leaf,
3762 struct buffer_head **orig_dx_leaves,
3763 struct buffer_head **new_dx_leaves,
3768 struct ocfs2_dx_leaf *orig_dx_leaf, *new_dx_leaf;
3769 struct ocfs2_dx_entry_list *orig_list, *new_list, *tmp_list;
3770 struct ocfs2_dx_entry *dx_entry;
3772 tmp_list = &tmp_dx_leaf->dl_list;
3774 for (i = 0; i < num_dx_leaves; i++) {
3775 orig_dx_leaf = (struct ocfs2_dx_leaf *) orig_dx_leaves[i]->b_data;
3776 orig_list = &orig_dx_leaf->dl_list;
3777 new_dx_leaf = (struct ocfs2_dx_leaf *) new_dx_leaves[i]->b_data;
3778 new_list = &new_dx_leaf->dl_list;
3780 num_used = le16_to_cpu(orig_list->de_num_used);
3782 memcpy(tmp_dx_leaf, orig_dx_leaf, dir->i_sb->s_blocksize);
3783 tmp_list->de_num_used = cpu_to_le16(0);
3784 memset(&tmp_list->de_entries, 0, sizeof(*dx_entry)*num_used);
3786 for (j = 0; j < num_used; j++) {
3787 dx_entry = &orig_list->de_entries[j];
3788 major_hash = le32_to_cpu(dx_entry->dx_major_hash);
3789 if (major_hash >= split_hash)
3790 ocfs2_dx_dir_leaf_insert_tail(new_dx_leaf,
3793 ocfs2_dx_dir_leaf_insert_tail(tmp_dx_leaf,
3796 memcpy(orig_dx_leaf, tmp_dx_leaf, dir->i_sb->s_blocksize);
3798 ocfs2_journal_dirty(handle, orig_dx_leaves[i]);
3799 ocfs2_journal_dirty(handle, new_dx_leaves[i]);
3803 static int ocfs2_dx_dir_rebalance_credits(struct ocfs2_super *osb,
3804 struct ocfs2_dx_root_block *dx_root)
3806 int credits = ocfs2_clusters_to_blocks(osb->sb, 2);
3808 credits += ocfs2_calc_extend_credits(osb->sb, &dx_root->dr_list, 1);
3809 credits += ocfs2_quota_trans_credits(osb->sb);
3814 * Find the median value in dx_leaf_bh and allocate a new leaf to move
3815 * half our entries into.
3817 static int ocfs2_dx_dir_rebalance(struct ocfs2_super *osb, struct inode *dir,
3818 struct buffer_head *dx_root_bh,
3819 struct buffer_head *dx_leaf_bh,
3820 struct ocfs2_dx_hinfo *hinfo, u32 leaf_cpos,
3823 struct ocfs2_dx_leaf *dx_leaf = (struct ocfs2_dx_leaf *)dx_leaf_bh->b_data;
3824 int credits, ret, i, num_used, did_quota = 0;
3825 u32 cpos, split_hash, insert_hash = hinfo->major_hash;
3826 u64 orig_leaves_start;
3828 struct buffer_head **orig_dx_leaves = NULL;
3829 struct buffer_head **new_dx_leaves = NULL;
3830 struct ocfs2_alloc_context *data_ac = NULL, *meta_ac = NULL;
3831 struct ocfs2_extent_tree et;
3832 handle_t *handle = NULL;
3833 struct ocfs2_dx_root_block *dx_root;
3834 struct ocfs2_dx_leaf *tmp_dx_leaf = NULL;
3836 mlog(0, "DX Dir: %llu, rebalance leaf leaf_blkno: %llu insert: %u\n",
3837 (unsigned long long)OCFS2_I(dir)->ip_blkno,
3838 (unsigned long long)leaf_blkno, insert_hash);
3840 ocfs2_init_dx_root_extent_tree(&et, dir, dx_root_bh);
3842 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
3844 * XXX: This is a rather large limit. We should use a more
3847 if (le32_to_cpu(dx_root->dr_clusters) == UINT_MAX)
3850 num_used = le16_to_cpu(dx_leaf->dl_list.de_num_used);
3851 if (num_used < le16_to_cpu(dx_leaf->dl_list.de_count)) {
3852 mlog(ML_ERROR, "DX Dir: %llu, Asked to rebalance empty leaf: "
3853 "%llu, %d\n", (unsigned long long)OCFS2_I(dir)->ip_blkno,
3854 (unsigned long long)leaf_blkno, num_used);
3859 orig_dx_leaves = ocfs2_dx_dir_kmalloc_leaves(osb->sb, &num_dx_leaves);
3860 if (!orig_dx_leaves) {
3866 new_dx_leaves = ocfs2_dx_dir_kmalloc_leaves(osb->sb, NULL);
3867 if (!new_dx_leaves) {
3873 ret = ocfs2_lock_allocators(dir, &et, 1, 0, &data_ac, &meta_ac);
3880 credits = ocfs2_dx_dir_rebalance_credits(osb, dx_root);
3881 handle = ocfs2_start_trans(osb, credits);
3882 if (IS_ERR(handle)) {
3883 ret = PTR_ERR(handle);
3889 if (vfs_dq_alloc_space_nodirty(dir,
3890 ocfs2_clusters_to_bytes(dir->i_sb, 1))) {
3896 ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir), dx_leaf_bh,
3897 OCFS2_JOURNAL_ACCESS_WRITE);
3904 * This block is changing anyway, so we can sort it in place.
3906 sort(dx_leaf->dl_list.de_entries, num_used,
3907 sizeof(struct ocfs2_dx_entry), dx_leaf_sort_cmp,
3910 ret = ocfs2_journal_dirty(handle, dx_leaf_bh);
3916 ret = ocfs2_dx_dir_find_leaf_split(dx_leaf, leaf_cpos, insert_hash,
3923 mlog(0, "Split leaf (%u) at %u, insert major hash is %u\n",
3924 leaf_cpos, split_hash, insert_hash);
3927 * We have to carefully order operations here. There are items
3928 * which want to be in the new cluster before insert, but in
3929 * order to put those items in the new cluster, we alter the
3930 * old cluster. A failure to insert gets nasty.
3932 * So, start by reserving writes to the old
3933 * cluster. ocfs2_dx_dir_new_cluster will reserve writes on
3934 * the new cluster for us, before inserting it. The insert
3935 * won't happen if there's an error before that. Once the
3936 * insert is done then, we can transfer from one leaf into the
3937 * other without fear of hitting any error.
3941 * The leaf transfer wants some scratch space so that we don't
3942 * wind up doing a bunch of expensive memmove().
3944 tmp_dx_leaf = kmalloc(osb->sb->s_blocksize, GFP_NOFS);
3951 orig_leaves_start = ocfs2_block_to_cluster_start(dir->i_sb, leaf_blkno);
3952 ret = ocfs2_read_dx_leaves(dir, orig_leaves_start, num_dx_leaves,
3959 for (i = 0; i < num_dx_leaves; i++) {
3960 ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir),
3962 OCFS2_JOURNAL_ACCESS_WRITE);
3970 ret = ocfs2_dx_dir_new_cluster(dir, &et, cpos, handle,
3971 data_ac, meta_ac, new_dx_leaves,
3978 ocfs2_dx_dir_transfer_leaf(dir, split_hash, handle, tmp_dx_leaf,
3979 orig_dx_leaves, new_dx_leaves, num_dx_leaves);
3982 if (ret < 0 && did_quota)
3983 vfs_dq_free_space_nodirty(dir,
3984 ocfs2_clusters_to_bytes(dir->i_sb, 1));
3986 ocfs2_commit_trans(osb, handle);
3989 if (orig_dx_leaves || new_dx_leaves) {
3990 for (i = 0; i < num_dx_leaves; i++) {
3992 brelse(orig_dx_leaves[i]);
3994 brelse(new_dx_leaves[i]);
3996 kfree(orig_dx_leaves);
3997 kfree(new_dx_leaves);
4001 ocfs2_free_alloc_context(meta_ac);
4003 ocfs2_free_alloc_context(data_ac);
4009 static int ocfs2_find_dir_space_dx(struct ocfs2_super *osb, struct inode *dir,
4010 struct buffer_head *di_bh,
4011 struct buffer_head *dx_root_bh,
4012 const char *name, int namelen,
4013 struct ocfs2_dir_lookup_result *lookup)
4015 int ret, rebalanced = 0;
4016 struct ocfs2_dx_root_block *dx_root;
4017 struct buffer_head *dx_leaf_bh = NULL;
4018 struct ocfs2_dx_leaf *dx_leaf;
4022 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
4025 ret = ocfs2_dx_dir_lookup(dir, &dx_root->dr_list, &lookup->dl_hinfo,
4026 &leaf_cpos, &blkno);
4032 ret = ocfs2_read_dx_leaf(dir, blkno, &dx_leaf_bh);
4038 dx_leaf = (struct ocfs2_dx_leaf *)dx_leaf_bh->b_data;
4040 if (le16_to_cpu(dx_leaf->dl_list.de_num_used) >=
4041 le16_to_cpu(dx_leaf->dl_list.de_count)) {
4044 * Rebalancing should have provided us with
4045 * space in an appropriate leaf.
4047 * XXX: Is this an abnormal condition then?
4048 * Should we print a message here?
4054 ret = ocfs2_dx_dir_rebalance(osb, dir, dx_root_bh, dx_leaf_bh,
4055 &lookup->dl_hinfo, leaf_cpos,
4064 * Restart the lookup. The rebalance might have
4065 * changed which block our item fits into. Mark our
4066 * progress, so we only execute this once.
4071 goto restart_search;
4074 lookup->dl_dx_leaf_bh = dx_leaf_bh;
4082 static int ocfs2_search_dx_free_list(struct inode *dir,
4083 struct buffer_head *dx_root_bh,
4085 struct ocfs2_dir_lookup_result *lookup)
4088 struct buffer_head *leaf_bh = NULL, *prev_leaf_bh = NULL;
4089 struct ocfs2_dir_block_trailer *db;
4091 int rec_len = OCFS2_DIR_REC_LEN(namelen);
4092 struct ocfs2_dx_root_block *dx_root;
4094 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
4095 next_block = le64_to_cpu(dx_root->dr_free_blk);
4097 while (next_block) {
4098 brelse(prev_leaf_bh);
4099 prev_leaf_bh = leaf_bh;
4102 ret = ocfs2_read_dir_block_direct(dir, next_block, &leaf_bh);
4108 db = ocfs2_trailer_from_bh(leaf_bh, dir->i_sb);
4109 if (rec_len <= le16_to_cpu(db->db_free_rec_len)) {
4110 lookup->dl_leaf_bh = leaf_bh;
4111 lookup->dl_prev_leaf_bh = prev_leaf_bh;
4113 prev_leaf_bh = NULL;
4117 next_block = le64_to_cpu(db->db_free_next);
4126 brelse(prev_leaf_bh);
4130 static int ocfs2_expand_inline_dx_root(struct inode *dir,
4131 struct buffer_head *dx_root_bh)
4133 int ret, num_dx_leaves, i, j, did_quota = 0;
4134 struct buffer_head **dx_leaves = NULL;
4135 struct ocfs2_extent_tree et;
4137 struct ocfs2_alloc_context *data_ac = NULL;
4138 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
4139 handle_t *handle = NULL;
4140 struct ocfs2_dx_root_block *dx_root;
4141 struct ocfs2_dx_entry_list *entry_list;
4142 struct ocfs2_dx_entry *dx_entry;
4143 struct ocfs2_dx_leaf *target_leaf;
4145 ret = ocfs2_reserve_clusters(osb, 1, &data_ac);
4151 dx_leaves = ocfs2_dx_dir_kmalloc_leaves(osb->sb, &num_dx_leaves);
4158 handle = ocfs2_start_trans(osb, ocfs2_calc_dxi_expand_credits(osb->sb));
4159 if (IS_ERR(handle)) {
4160 ret = PTR_ERR(handle);
4165 if (vfs_dq_alloc_space_nodirty(dir,
4166 ocfs2_clusters_to_bytes(osb->sb, 1))) {
4173 * We do this up front, before the allocation, so that a
4174 * failure to add the dx_root_bh to the journal won't result
4175 * us losing clusters.
4177 ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
4178 OCFS2_JOURNAL_ACCESS_WRITE);
4184 ret = __ocfs2_dx_dir_new_cluster(dir, 0, handle, data_ac, dx_leaves,
4185 num_dx_leaves, &insert_blkno);
4192 * Transfer the entries from our dx_root into the appropriate
4195 dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data;
4196 entry_list = &dx_root->dr_entries;
4198 for (i = 0; i < le16_to_cpu(entry_list->de_num_used); i++) {
4199 dx_entry = &entry_list->de_entries[i];
4201 j = __ocfs2_dx_dir_hash_idx(osb,
4202 le32_to_cpu(dx_entry->dx_minor_hash));
4203 target_leaf = (struct ocfs2_dx_leaf *)dx_leaves[j]->b_data;
4205 ocfs2_dx_dir_leaf_insert_tail(target_leaf, dx_entry);
4207 /* Each leaf has been passed to the journal already
4208 * via __ocfs2_dx_dir_new_cluster() */
4211 dx_root->dr_flags &= ~OCFS2_DX_FLAG_INLINE;
4212 memset(&dx_root->dr_list, 0, osb->sb->s_blocksize -
4213 offsetof(struct ocfs2_dx_root_block, dr_list));
4214 dx_root->dr_list.l_count =
4215 cpu_to_le16(ocfs2_extent_recs_per_dx_root(osb->sb));
4217 /* This should never fail considering we start with an empty
4219 ocfs2_init_dx_root_extent_tree(&et, dir, dx_root_bh);
4220 ret = ocfs2_insert_extent(osb, handle, dir, &et, 0,
4221 insert_blkno, 1, 0, NULL);
4226 ocfs2_journal_dirty(handle, dx_root_bh);
4229 if (ret < 0 && did_quota)
4230 vfs_dq_free_space_nodirty(dir,
4231 ocfs2_clusters_to_bytes(dir->i_sb, 1));
4233 ocfs2_commit_trans(osb, handle);
4237 ocfs2_free_alloc_context(data_ac);
4240 for (i = 0; i < num_dx_leaves; i++)
4241 brelse(dx_leaves[i]);
4247 static int ocfs2_inline_dx_has_space(struct buffer_head *dx_root_bh)
4249 struct ocfs2_dx_root_block *dx_root;
4250 struct ocfs2_dx_entry_list *entry_list;
4252 dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data;
4253 entry_list = &dx_root->dr_entries;
4255 if (le16_to_cpu(entry_list->de_num_used) >=
4256 le16_to_cpu(entry_list->de_count))
4262 static int ocfs2_prepare_dx_dir_for_insert(struct inode *dir,
4263 struct buffer_head *di_bh,
4266 struct ocfs2_dir_lookup_result *lookup)
4268 int ret, free_dx_root = 1;
4269 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
4270 struct buffer_head *dx_root_bh = NULL;
4271 struct buffer_head *leaf_bh = NULL;
4272 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
4273 struct ocfs2_dx_root_block *dx_root;
4275 ret = ocfs2_read_dx_root(dir, di, &dx_root_bh);
4281 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
4282 if (le32_to_cpu(dx_root->dr_num_entries) == OCFS2_DX_ENTRIES_MAX) {
4288 if (ocfs2_dx_root_inline(dx_root)) {
4289 ret = ocfs2_inline_dx_has_space(dx_root_bh);
4295 * We ran out of room in the root block. Expand it to
4296 * an extent, then allow ocfs2_find_dir_space_dx to do
4299 ret = ocfs2_expand_inline_dx_root(dir, dx_root_bh);
4307 * Insert preparation for an indexed directory is split into two
4308 * steps. The call to find_dir_space_dx reserves room in the index for
4309 * an additional item. If we run out of space there, it's a real error
4310 * we can't continue on.
4312 ret = ocfs2_find_dir_space_dx(osb, dir, di_bh, dx_root_bh, name,
4321 * Next, we need to find space in the unindexed tree. This call
4322 * searches using the free space linked list. If the unindexed tree
4323 * lacks sufficient space, we'll expand it below. The expansion code
4324 * is smart enough to add any new blocks to the free space list.
4326 ret = ocfs2_search_dx_free_list(dir, dx_root_bh, namelen, lookup);
4327 if (ret && ret != -ENOSPC) {
4332 /* Do this up here - ocfs2_extend_dir might need the dx_root */
4333 lookup->dl_dx_root_bh = dx_root_bh;
4336 if (ret == -ENOSPC) {
4337 ret = ocfs2_extend_dir(osb, dir, di_bh, 1, lookup, &leaf_bh);
4345 * We make the assumption here that new leaf blocks are added
4346 * to the front of our free list.
4348 lookup->dl_prev_leaf_bh = NULL;
4349 lookup->dl_leaf_bh = leaf_bh;
4359 * Get a directory ready for insert. Any directory allocation required
4360 * happens here. Success returns zero, and enough context in the dir
4361 * lookup result that ocfs2_add_entry() will be able complete the task
4362 * with minimal performance impact.
4364 int ocfs2_prepare_dir_for_insert(struct ocfs2_super *osb,
4366 struct buffer_head *parent_fe_bh,
4369 struct ocfs2_dir_lookup_result *lookup)
4372 unsigned int blocks_wanted = 1;
4373 struct buffer_head *bh = NULL;
4375 mlog(0, "getting ready to insert namelen %d into dir %llu\n",
4376 namelen, (unsigned long long)OCFS2_I(dir)->ip_blkno);
4385 * Do this up front to reduce confusion.
4387 * The directory might start inline, then be turned into an
4388 * indexed one, in which case we'd need to hash deep inside
4389 * ocfs2_find_dir_space_id(). Since
4390 * ocfs2_prepare_dx_dir_for_insert() also needs this hash
4391 * done, there seems no point in spreading out the calls. We
4392 * can optimize away the case where the file system doesn't
4395 if (ocfs2_supports_indexed_dirs(osb))
4396 ocfs2_dx_dir_name_hash(dir, name, namelen, &lookup->dl_hinfo);
4398 if (ocfs2_dir_indexed(dir)) {
4399 ret = ocfs2_prepare_dx_dir_for_insert(dir, parent_fe_bh,
4400 name, namelen, lookup);
4406 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
4407 ret = ocfs2_find_dir_space_id(dir, parent_fe_bh, name,
4408 namelen, &bh, &blocks_wanted);
4410 ret = ocfs2_find_dir_space_el(dir, name, namelen, &bh);
4412 if (ret && ret != -ENOSPC) {
4417 if (ret == -ENOSPC) {
4419 * We have to expand the directory to add this name.
4423 ret = ocfs2_extend_dir(osb, dir, parent_fe_bh, blocks_wanted,
4434 lookup->dl_leaf_bh = bh;
4441 static int ocfs2_dx_dir_remove_index(struct inode *dir,
4442 struct buffer_head *di_bh,
4443 struct buffer_head *dx_root_bh)
4446 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
4447 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
4448 struct ocfs2_dx_root_block *dx_root;
4449 struct inode *dx_alloc_inode = NULL;
4450 struct buffer_head *dx_alloc_bh = NULL;
4456 dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data;
4458 dx_alloc_inode = ocfs2_get_system_file_inode(osb,
4459 EXTENT_ALLOC_SYSTEM_INODE,
4460 le16_to_cpu(dx_root->dr_suballoc_slot));
4461 if (!dx_alloc_inode) {
4466 mutex_lock(&dx_alloc_inode->i_mutex);
4468 ret = ocfs2_inode_lock(dx_alloc_inode, &dx_alloc_bh, 1);
4474 handle = ocfs2_start_trans(osb, OCFS2_DX_ROOT_REMOVE_CREDITS);
4475 if (IS_ERR(handle)) {
4476 ret = PTR_ERR(handle);
4481 ret = ocfs2_journal_access_di(handle, INODE_CACHE(dir), di_bh,
4482 OCFS2_JOURNAL_ACCESS_WRITE);
4488 OCFS2_I(dir)->ip_dyn_features &= ~OCFS2_INDEXED_DIR_FL;
4489 di->i_dyn_features = cpu_to_le16(OCFS2_I(dir)->ip_dyn_features);
4490 di->i_dx_root = cpu_to_le64(0ULL);
4492 ocfs2_journal_dirty(handle, di_bh);
4494 blk = le64_to_cpu(dx_root->dr_blkno);
4495 bit = le16_to_cpu(dx_root->dr_suballoc_bit);
4496 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
4497 ret = ocfs2_free_suballoc_bits(handle, dx_alloc_inode, dx_alloc_bh,
4503 ocfs2_commit_trans(osb, handle);
4506 ocfs2_inode_unlock(dx_alloc_inode, 1);
4509 mutex_unlock(&dx_alloc_inode->i_mutex);
4510 brelse(dx_alloc_bh);
4512 iput(dx_alloc_inode);
4516 int ocfs2_dx_dir_truncate(struct inode *dir, struct buffer_head *di_bh)
4519 unsigned int uninitialized_var(clen);
4520 u32 major_hash = UINT_MAX, p_cpos, uninitialized_var(cpos);
4521 u64 uninitialized_var(blkno);
4522 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
4523 struct buffer_head *dx_root_bh = NULL;
4524 struct ocfs2_dx_root_block *dx_root;
4525 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
4526 struct ocfs2_cached_dealloc_ctxt dealloc;
4527 struct ocfs2_extent_tree et;
4529 ocfs2_init_dealloc_ctxt(&dealloc);
4531 if (!ocfs2_dir_indexed(dir))
4534 ret = ocfs2_read_dx_root(dir, di, &dx_root_bh);
4539 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
4541 if (ocfs2_dx_root_inline(dx_root))
4544 ocfs2_init_dx_root_extent_tree(&et, dir, dx_root_bh);
4546 /* XXX: What if dr_clusters is too large? */
4547 while (le32_to_cpu(dx_root->dr_clusters)) {
4548 ret = ocfs2_dx_dir_lookup_rec(dir, &dx_root->dr_list,
4549 major_hash, &cpos, &blkno, &clen);
4555 p_cpos = ocfs2_blocks_to_clusters(dir->i_sb, blkno);
4557 ret = ocfs2_remove_btree_range(dir, &et, cpos, p_cpos, clen,
4567 major_hash = cpos - 1;
4571 ret = ocfs2_dx_dir_remove_index(dir, di_bh, dx_root_bh);
4577 ocfs2_remove_from_cache(INODE_CACHE(dir), dx_root_bh);
4579 ocfs2_schedule_truncate_log_flush(osb, 1);
4580 ocfs2_run_deallocs(osb, &dealloc);