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_CACHE(inode), el, major_hash,
815 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
818 if (el->l_tree_depth) {
819 ocfs2_error(inode->i_sb,
820 "Inode %lu has non zero tree depth in "
821 "btree tree block %llu\n", inode->i_ino,
822 (unsigned long long)eb_bh->b_blocknr);
829 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
830 rec = &el->l_recs[i];
832 if (le32_to_cpu(rec->e_cpos) <= major_hash) {
839 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
840 "record (%u, %u, 0) in btree", inode->i_ino,
841 le32_to_cpu(rec->e_cpos),
842 ocfs2_rec_clusters(el, rec));
848 *ret_phys_blkno = le64_to_cpu(rec->e_blkno);
850 *ret_cpos = le32_to_cpu(rec->e_cpos);
852 *ret_clen = le16_to_cpu(rec->e_leaf_clusters);
860 * Returns the block index, from the start of the cluster which this
863 static inline unsigned int __ocfs2_dx_dir_hash_idx(struct ocfs2_super *osb,
866 return minor_hash & osb->osb_dx_mask;
869 static inline unsigned int ocfs2_dx_dir_hash_idx(struct ocfs2_super *osb,
870 struct ocfs2_dx_hinfo *hinfo)
872 return __ocfs2_dx_dir_hash_idx(osb, hinfo->minor_hash);
875 static int ocfs2_dx_dir_lookup(struct inode *inode,
876 struct ocfs2_extent_list *el,
877 struct ocfs2_dx_hinfo *hinfo,
882 unsigned int cend, uninitialized_var(clen);
883 u32 uninitialized_var(cpos);
884 u64 uninitialized_var(blkno);
885 u32 name_hash = hinfo->major_hash;
887 ret = ocfs2_dx_dir_lookup_rec(inode, el, name_hash, &cpos, &blkno,
895 if (name_hash >= cend) {
896 /* We want the last cluster */
897 blkno += ocfs2_clusters_to_blocks(inode->i_sb, clen - 1);
900 blkno += ocfs2_clusters_to_blocks(inode->i_sb,
906 * We now have the cluster which should hold our entry. To
907 * find the exact block from the start of the cluster to
908 * search, we take the lower bits of the hash.
910 blkno += ocfs2_dx_dir_hash_idx(OCFS2_SB(inode->i_sb), hinfo);
913 *ret_phys_blkno = blkno;
922 static int ocfs2_dx_dir_search(const char *name, int namelen,
924 struct ocfs2_dx_root_block *dx_root,
925 struct ocfs2_dir_lookup_result *res)
928 u64 uninitialized_var(phys);
929 struct buffer_head *dx_leaf_bh = NULL;
930 struct ocfs2_dx_leaf *dx_leaf;
931 struct ocfs2_dx_entry *dx_entry = NULL;
932 struct buffer_head *dir_ent_bh = NULL;
933 struct ocfs2_dir_entry *dir_ent = NULL;
934 struct ocfs2_dx_hinfo *hinfo = &res->dl_hinfo;
935 struct ocfs2_extent_list *dr_el;
936 struct ocfs2_dx_entry_list *entry_list;
938 ocfs2_dx_dir_name_hash(dir, name, namelen, &res->dl_hinfo);
940 if (ocfs2_dx_root_inline(dx_root)) {
941 entry_list = &dx_root->dr_entries;
945 dr_el = &dx_root->dr_list;
947 ret = ocfs2_dx_dir_lookup(dir, dr_el, hinfo, NULL, &phys);
953 mlog(0, "Dir %llu: name: \"%.*s\", lookup of hash: %u.0x%x "
955 (unsigned long long)OCFS2_I(dir)->ip_blkno,
956 namelen, name, hinfo->major_hash, hinfo->minor_hash,
957 (unsigned long long)phys);
959 ret = ocfs2_read_dx_leaf(dir, phys, &dx_leaf_bh);
965 dx_leaf = (struct ocfs2_dx_leaf *) dx_leaf_bh->b_data;
967 mlog(0, "leaf info: num_used: %d, count: %d\n",
968 le16_to_cpu(dx_leaf->dl_list.de_num_used),
969 le16_to_cpu(dx_leaf->dl_list.de_count));
971 entry_list = &dx_leaf->dl_list;
975 * Empty leaf is legal, so no need to check for that.
978 for (i = 0; i < le16_to_cpu(entry_list->de_num_used); i++) {
979 dx_entry = &entry_list->de_entries[i];
981 if (hinfo->major_hash != le32_to_cpu(dx_entry->dx_major_hash)
982 || hinfo->minor_hash != le32_to_cpu(dx_entry->dx_minor_hash))
986 * Search unindexed leaf block now. We're not
987 * guaranteed to find anything.
989 ret = ocfs2_read_dir_block_direct(dir,
990 le64_to_cpu(dx_entry->dx_dirent_blk),
998 * XXX: We should check the unindexed block here,
1002 found = ocfs2_search_dirblock(dir_ent_bh, dir, name, namelen,
1003 0, dir_ent_bh->b_data,
1004 dir->i_sb->s_blocksize, &dir_ent);
1009 /* This means we found a bad directory entry. */
1024 res->dl_leaf_bh = dir_ent_bh;
1025 res->dl_entry = dir_ent;
1026 res->dl_dx_leaf_bh = dx_leaf_bh;
1027 res->dl_dx_entry = dx_entry;
1038 static int ocfs2_find_entry_dx(const char *name, int namelen,
1040 struct ocfs2_dir_lookup_result *lookup)
1043 struct buffer_head *di_bh = NULL;
1044 struct ocfs2_dinode *di;
1045 struct buffer_head *dx_root_bh = NULL;
1046 struct ocfs2_dx_root_block *dx_root;
1048 ret = ocfs2_read_inode_block(dir, &di_bh);
1054 di = (struct ocfs2_dinode *)di_bh->b_data;
1056 ret = ocfs2_read_dx_root(dir, di, &dx_root_bh);
1061 dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data;
1063 ret = ocfs2_dx_dir_search(name, namelen, dir, dx_root, lookup);
1070 lookup->dl_dx_root_bh = dx_root_bh;
1079 * Try to find an entry of the provided name within 'dir'.
1081 * If nothing was found, -ENOENT is returned. Otherwise, zero is
1082 * returned and the struct 'res' will contain information useful to
1083 * other directory manipulation functions.
1085 * Caller can NOT assume anything about the contents of the
1086 * buffer_heads - they are passed back only so that it can be passed
1087 * into any one of the manipulation functions (add entry, delete
1088 * entry, etc). As an example, bh in the extent directory case is a
1089 * data block, in the inline-data case it actually points to an inode,
1090 * in the indexed directory case, multiple buffers are involved.
1092 int ocfs2_find_entry(const char *name, int namelen,
1093 struct inode *dir, struct ocfs2_dir_lookup_result *lookup)
1095 struct buffer_head *bh;
1096 struct ocfs2_dir_entry *res_dir = NULL;
1098 if (ocfs2_dir_indexed(dir))
1099 return ocfs2_find_entry_dx(name, namelen, dir, lookup);
1102 * The unindexed dir code only uses part of the lookup
1103 * structure, so there's no reason to push it down further
1106 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
1107 bh = ocfs2_find_entry_id(name, namelen, dir, &res_dir);
1109 bh = ocfs2_find_entry_el(name, namelen, dir, &res_dir);
1114 lookup->dl_leaf_bh = bh;
1115 lookup->dl_entry = res_dir;
1120 * Update inode number and type of a previously found directory entry.
1122 int ocfs2_update_entry(struct inode *dir, handle_t *handle,
1123 struct ocfs2_dir_lookup_result *res,
1124 struct inode *new_entry_inode)
1127 ocfs2_journal_access_func access = ocfs2_journal_access_db;
1128 struct ocfs2_dir_entry *de = res->dl_entry;
1129 struct buffer_head *de_bh = res->dl_leaf_bh;
1132 * The same code works fine for both inline-data and extent
1133 * based directories, so no need to split this up. The only
1134 * difference is the journal_access function.
1137 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
1138 access = ocfs2_journal_access_di;
1140 ret = access(handle, INODE_CACHE(dir), de_bh,
1141 OCFS2_JOURNAL_ACCESS_WRITE);
1147 de->inode = cpu_to_le64(OCFS2_I(new_entry_inode)->ip_blkno);
1148 ocfs2_set_de_type(de, new_entry_inode->i_mode);
1150 ocfs2_journal_dirty(handle, de_bh);
1157 * __ocfs2_delete_entry deletes a directory entry by merging it with the
1160 static int __ocfs2_delete_entry(handle_t *handle, struct inode *dir,
1161 struct ocfs2_dir_entry *de_del,
1162 struct buffer_head *bh, char *first_de,
1165 struct ocfs2_dir_entry *de, *pde;
1166 int i, status = -ENOENT;
1167 ocfs2_journal_access_func access = ocfs2_journal_access_db;
1169 mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p)\n", handle, dir, de_del, bh);
1171 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
1172 access = ocfs2_journal_access_di;
1176 de = (struct ocfs2_dir_entry *) first_de;
1178 if (!ocfs2_check_dir_entry(dir, de, bh, i)) {
1184 status = access(handle, INODE_CACHE(dir), bh,
1185 OCFS2_JOURNAL_ACCESS_WRITE);
1192 le16_add_cpu(&pde->rec_len,
1193 le16_to_cpu(de->rec_len));
1197 ocfs2_journal_dirty(handle, bh);
1200 i += le16_to_cpu(de->rec_len);
1202 de = (struct ocfs2_dir_entry *)((char *)de + le16_to_cpu(de->rec_len));
1209 static unsigned int ocfs2_figure_dirent_hole(struct ocfs2_dir_entry *de)
1213 if (le64_to_cpu(de->inode) == 0)
1214 hole = le16_to_cpu(de->rec_len);
1216 hole = le16_to_cpu(de->rec_len) -
1217 OCFS2_DIR_REC_LEN(de->name_len);
1222 static int ocfs2_find_max_rec_len(struct super_block *sb,
1223 struct buffer_head *dirblock_bh)
1225 int size, this_hole, largest_hole = 0;
1226 char *trailer, *de_buf, *limit, *start = dirblock_bh->b_data;
1227 struct ocfs2_dir_entry *de;
1229 trailer = (char *)ocfs2_trailer_from_bh(dirblock_bh, sb);
1230 size = ocfs2_dir_trailer_blk_off(sb);
1231 limit = start + size;
1233 de = (struct ocfs2_dir_entry *)de_buf;
1235 if (de_buf != trailer) {
1236 this_hole = ocfs2_figure_dirent_hole(de);
1237 if (this_hole > largest_hole)
1238 largest_hole = this_hole;
1241 de_buf += le16_to_cpu(de->rec_len);
1242 de = (struct ocfs2_dir_entry *)de_buf;
1243 } while (de_buf < limit);
1245 if (largest_hole >= OCFS2_DIR_MIN_REC_LEN)
1246 return largest_hole;
1250 static void ocfs2_dx_list_remove_entry(struct ocfs2_dx_entry_list *entry_list,
1253 int num_used = le16_to_cpu(entry_list->de_num_used);
1255 if (num_used == 1 || index == (num_used - 1))
1258 memmove(&entry_list->de_entries[index],
1259 &entry_list->de_entries[index + 1],
1260 (num_used - index - 1)*sizeof(struct ocfs2_dx_entry));
1263 memset(&entry_list->de_entries[num_used], 0,
1264 sizeof(struct ocfs2_dx_entry));
1265 entry_list->de_num_used = cpu_to_le16(num_used);
1268 static int ocfs2_delete_entry_dx(handle_t *handle, struct inode *dir,
1269 struct ocfs2_dir_lookup_result *lookup)
1271 int ret, index, max_rec_len, add_to_free_list = 0;
1272 struct buffer_head *dx_root_bh = lookup->dl_dx_root_bh;
1273 struct buffer_head *leaf_bh = lookup->dl_leaf_bh;
1274 struct ocfs2_dx_leaf *dx_leaf;
1275 struct ocfs2_dx_entry *dx_entry = lookup->dl_dx_entry;
1276 struct ocfs2_dir_block_trailer *trailer;
1277 struct ocfs2_dx_root_block *dx_root;
1278 struct ocfs2_dx_entry_list *entry_list;
1281 * This function gets a bit messy because we might have to
1282 * modify the root block, regardless of whether the indexed
1283 * entries are stored inline.
1287 * *Only* set 'entry_list' here, based on where we're looking
1288 * for the indexed entries. Later, we might still want to
1289 * journal both blocks, based on free list state.
1291 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
1292 if (ocfs2_dx_root_inline(dx_root)) {
1293 entry_list = &dx_root->dr_entries;
1295 dx_leaf = (struct ocfs2_dx_leaf *) lookup->dl_dx_leaf_bh->b_data;
1296 entry_list = &dx_leaf->dl_list;
1299 /* Neither of these are a disk corruption - that should have
1300 * been caught by lookup, before we got here. */
1301 BUG_ON(le16_to_cpu(entry_list->de_count) <= 0);
1302 BUG_ON(le16_to_cpu(entry_list->de_num_used) <= 0);
1304 index = (char *)dx_entry - (char *)entry_list->de_entries;
1305 index /= sizeof(*dx_entry);
1307 if (index >= le16_to_cpu(entry_list->de_num_used)) {
1308 mlog(ML_ERROR, "Dir %llu: Bad dx_entry ptr idx %d, (%p, %p)\n",
1309 (unsigned long long)OCFS2_I(dir)->ip_blkno, index,
1310 entry_list, dx_entry);
1315 * We know that removal of this dirent will leave enough room
1316 * for a new one, so add this block to the free list if it
1317 * isn't already there.
1319 trailer = ocfs2_trailer_from_bh(leaf_bh, dir->i_sb);
1320 if (trailer->db_free_rec_len == 0)
1321 add_to_free_list = 1;
1324 * Add the block holding our index into the journal before
1325 * removing the unindexed entry. If we get an error return
1326 * from __ocfs2_delete_entry(), then it hasn't removed the
1327 * entry yet. Likewise, successful return means we *must*
1328 * remove the indexed entry.
1330 * We're also careful to journal the root tree block here as
1331 * the entry count needs to be updated. Also, we might be
1332 * adding to the start of the free list.
1334 ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
1335 OCFS2_JOURNAL_ACCESS_WRITE);
1341 if (!ocfs2_dx_root_inline(dx_root)) {
1342 ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir),
1343 lookup->dl_dx_leaf_bh,
1344 OCFS2_JOURNAL_ACCESS_WRITE);
1351 mlog(0, "Dir %llu: delete entry at index: %d\n",
1352 (unsigned long long)OCFS2_I(dir)->ip_blkno, index);
1354 ret = __ocfs2_delete_entry(handle, dir, lookup->dl_entry,
1355 leaf_bh, leaf_bh->b_data, leaf_bh->b_size);
1361 max_rec_len = ocfs2_find_max_rec_len(dir->i_sb, leaf_bh);
1362 trailer->db_free_rec_len = cpu_to_le16(max_rec_len);
1363 if (add_to_free_list) {
1364 trailer->db_free_next = dx_root->dr_free_blk;
1365 dx_root->dr_free_blk = cpu_to_le64(leaf_bh->b_blocknr);
1366 ocfs2_journal_dirty(handle, dx_root_bh);
1369 /* leaf_bh was journal_accessed for us in __ocfs2_delete_entry */
1370 ocfs2_journal_dirty(handle, leaf_bh);
1372 le32_add_cpu(&dx_root->dr_num_entries, -1);
1373 ocfs2_journal_dirty(handle, dx_root_bh);
1375 ocfs2_dx_list_remove_entry(entry_list, index);
1377 if (!ocfs2_dx_root_inline(dx_root))
1378 ocfs2_journal_dirty(handle, lookup->dl_dx_leaf_bh);
1384 static inline int ocfs2_delete_entry_id(handle_t *handle,
1386 struct ocfs2_dir_entry *de_del,
1387 struct buffer_head *bh)
1390 struct buffer_head *di_bh = NULL;
1391 struct ocfs2_dinode *di;
1392 struct ocfs2_inline_data *data;
1394 ret = ocfs2_read_inode_block(dir, &di_bh);
1400 di = (struct ocfs2_dinode *)di_bh->b_data;
1401 data = &di->id2.i_data;
1403 ret = __ocfs2_delete_entry(handle, dir, de_del, bh, data->id_data,
1411 static inline int ocfs2_delete_entry_el(handle_t *handle,
1413 struct ocfs2_dir_entry *de_del,
1414 struct buffer_head *bh)
1416 return __ocfs2_delete_entry(handle, dir, de_del, bh, bh->b_data,
1421 * Delete a directory entry. Hide the details of directory
1422 * implementation from the caller.
1424 int ocfs2_delete_entry(handle_t *handle,
1426 struct ocfs2_dir_lookup_result *res)
1428 if (ocfs2_dir_indexed(dir))
1429 return ocfs2_delete_entry_dx(handle, dir, res);
1431 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
1432 return ocfs2_delete_entry_id(handle, dir, res->dl_entry,
1435 return ocfs2_delete_entry_el(handle, dir, res->dl_entry,
1440 * Check whether 'de' has enough room to hold an entry of
1441 * 'new_rec_len' bytes.
1443 static inline int ocfs2_dirent_would_fit(struct ocfs2_dir_entry *de,
1444 unsigned int new_rec_len)
1446 unsigned int de_really_used;
1448 /* Check whether this is an empty record with enough space */
1449 if (le64_to_cpu(de->inode) == 0 &&
1450 le16_to_cpu(de->rec_len) >= new_rec_len)
1454 * Record might have free space at the end which we can
1457 de_really_used = OCFS2_DIR_REC_LEN(de->name_len);
1458 if (le16_to_cpu(de->rec_len) >= (de_really_used + new_rec_len))
1464 static void ocfs2_dx_dir_leaf_insert_tail(struct ocfs2_dx_leaf *dx_leaf,
1465 struct ocfs2_dx_entry *dx_new_entry)
1469 i = le16_to_cpu(dx_leaf->dl_list.de_num_used);
1470 dx_leaf->dl_list.de_entries[i] = *dx_new_entry;
1472 le16_add_cpu(&dx_leaf->dl_list.de_num_used, 1);
1475 static void ocfs2_dx_entry_list_insert(struct ocfs2_dx_entry_list *entry_list,
1476 struct ocfs2_dx_hinfo *hinfo,
1480 struct ocfs2_dx_entry *dx_entry;
1482 i = le16_to_cpu(entry_list->de_num_used);
1483 dx_entry = &entry_list->de_entries[i];
1485 memset(dx_entry, 0, sizeof(*dx_entry));
1486 dx_entry->dx_major_hash = cpu_to_le32(hinfo->major_hash);
1487 dx_entry->dx_minor_hash = cpu_to_le32(hinfo->minor_hash);
1488 dx_entry->dx_dirent_blk = cpu_to_le64(dirent_blk);
1490 le16_add_cpu(&entry_list->de_num_used, 1);
1493 static int __ocfs2_dx_dir_leaf_insert(struct inode *dir, handle_t *handle,
1494 struct ocfs2_dx_hinfo *hinfo,
1496 struct buffer_head *dx_leaf_bh)
1499 struct ocfs2_dx_leaf *dx_leaf;
1501 ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir), dx_leaf_bh,
1502 OCFS2_JOURNAL_ACCESS_WRITE);
1508 dx_leaf = (struct ocfs2_dx_leaf *)dx_leaf_bh->b_data;
1509 ocfs2_dx_entry_list_insert(&dx_leaf->dl_list, hinfo, dirent_blk);
1510 ocfs2_journal_dirty(handle, dx_leaf_bh);
1516 static void ocfs2_dx_inline_root_insert(struct inode *dir, handle_t *handle,
1517 struct ocfs2_dx_hinfo *hinfo,
1519 struct ocfs2_dx_root_block *dx_root)
1521 ocfs2_dx_entry_list_insert(&dx_root->dr_entries, hinfo, dirent_blk);
1524 static int ocfs2_dx_dir_insert(struct inode *dir, handle_t *handle,
1525 struct ocfs2_dir_lookup_result *lookup)
1528 struct ocfs2_dx_root_block *dx_root;
1529 struct buffer_head *dx_root_bh = lookup->dl_dx_root_bh;
1531 ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
1532 OCFS2_JOURNAL_ACCESS_WRITE);
1538 dx_root = (struct ocfs2_dx_root_block *)lookup->dl_dx_root_bh->b_data;
1539 if (ocfs2_dx_root_inline(dx_root)) {
1540 ocfs2_dx_inline_root_insert(dir, handle,
1542 lookup->dl_leaf_bh->b_blocknr,
1545 ret = __ocfs2_dx_dir_leaf_insert(dir, handle, &lookup->dl_hinfo,
1546 lookup->dl_leaf_bh->b_blocknr,
1547 lookup->dl_dx_leaf_bh);
1552 le32_add_cpu(&dx_root->dr_num_entries, 1);
1553 ocfs2_journal_dirty(handle, dx_root_bh);
1559 static void ocfs2_remove_block_from_free_list(struct inode *dir,
1561 struct ocfs2_dir_lookup_result *lookup)
1563 struct ocfs2_dir_block_trailer *trailer, *prev;
1564 struct ocfs2_dx_root_block *dx_root;
1565 struct buffer_head *bh;
1567 trailer = ocfs2_trailer_from_bh(lookup->dl_leaf_bh, dir->i_sb);
1569 if (ocfs2_free_list_at_root(lookup)) {
1570 bh = lookup->dl_dx_root_bh;
1571 dx_root = (struct ocfs2_dx_root_block *)bh->b_data;
1572 dx_root->dr_free_blk = trailer->db_free_next;
1574 bh = lookup->dl_prev_leaf_bh;
1575 prev = ocfs2_trailer_from_bh(bh, dir->i_sb);
1576 prev->db_free_next = trailer->db_free_next;
1579 trailer->db_free_rec_len = cpu_to_le16(0);
1580 trailer->db_free_next = cpu_to_le64(0);
1582 ocfs2_journal_dirty(handle, bh);
1583 ocfs2_journal_dirty(handle, lookup->dl_leaf_bh);
1587 * This expects that a journal write has been reserved on
1588 * lookup->dl_prev_leaf_bh or lookup->dl_dx_root_bh
1590 static void ocfs2_recalc_free_list(struct inode *dir, handle_t *handle,
1591 struct ocfs2_dir_lookup_result *lookup)
1594 struct ocfs2_dir_block_trailer *trailer;
1596 /* Walk dl_leaf_bh to figure out what the new free rec_len is. */
1597 max_rec_len = ocfs2_find_max_rec_len(dir->i_sb, lookup->dl_leaf_bh);
1600 * There's still room in this block, so no need to remove it
1601 * from the free list. In this case, we just want to update
1602 * the rec len accounting.
1604 trailer = ocfs2_trailer_from_bh(lookup->dl_leaf_bh, dir->i_sb);
1605 trailer->db_free_rec_len = cpu_to_le16(max_rec_len);
1606 ocfs2_journal_dirty(handle, lookup->dl_leaf_bh);
1608 ocfs2_remove_block_from_free_list(dir, handle, lookup);
1612 /* we don't always have a dentry for what we want to add, so people
1613 * like orphan dir can call this instead.
1615 * The lookup context must have been filled from
1616 * ocfs2_prepare_dir_for_insert.
1618 int __ocfs2_add_entry(handle_t *handle,
1620 const char *name, int namelen,
1621 struct inode *inode, u64 blkno,
1622 struct buffer_head *parent_fe_bh,
1623 struct ocfs2_dir_lookup_result *lookup)
1625 unsigned long offset;
1626 unsigned short rec_len;
1627 struct ocfs2_dir_entry *de, *de1;
1628 struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_fe_bh->b_data;
1629 struct super_block *sb = dir->i_sb;
1631 unsigned int size = sb->s_blocksize;
1632 struct buffer_head *insert_bh = lookup->dl_leaf_bh;
1633 char *data_start = insert_bh->b_data;
1640 if (ocfs2_dir_indexed(dir)) {
1641 struct buffer_head *bh;
1644 * An indexed dir may require that we update the free space
1645 * list. Reserve a write to the previous node in the list so
1646 * that we don't fail later.
1648 * XXX: This can be either a dx_root_block, or an unindexed
1649 * directory tree leaf block.
1651 if (ocfs2_free_list_at_root(lookup)) {
1652 bh = lookup->dl_dx_root_bh;
1653 retval = ocfs2_journal_access_dr(handle,
1654 INODE_CACHE(dir), bh,
1655 OCFS2_JOURNAL_ACCESS_WRITE);
1657 bh = lookup->dl_prev_leaf_bh;
1658 retval = ocfs2_journal_access_db(handle,
1659 INODE_CACHE(dir), bh,
1660 OCFS2_JOURNAL_ACCESS_WRITE);
1666 } else if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1667 data_start = di->id2.i_data.id_data;
1668 size = i_size_read(dir);
1670 BUG_ON(insert_bh != parent_fe_bh);
1673 rec_len = OCFS2_DIR_REC_LEN(namelen);
1675 de = (struct ocfs2_dir_entry *) data_start;
1677 BUG_ON((char *)de >= (size + data_start));
1679 /* These checks should've already been passed by the
1680 * prepare function, but I guess we can leave them
1682 if (!ocfs2_check_dir_entry(dir, de, insert_bh, offset)) {
1686 if (ocfs2_match(namelen, name, de)) {
1691 /* We're guaranteed that we should have space, so we
1692 * can't possibly have hit the trailer...right? */
1693 mlog_bug_on_msg(ocfs2_skip_dir_trailer(dir, de, offset, size),
1694 "Hit dir trailer trying to insert %.*s "
1695 "(namelen %d) into directory %llu. "
1696 "offset is %lu, trailer offset is %d\n",
1697 namelen, name, namelen,
1698 (unsigned long long)parent_fe_bh->b_blocknr,
1699 offset, ocfs2_dir_trailer_blk_off(dir->i_sb));
1701 if (ocfs2_dirent_would_fit(de, rec_len)) {
1702 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1703 retval = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
1709 if (insert_bh == parent_fe_bh)
1710 status = ocfs2_journal_access_di(handle,
1713 OCFS2_JOURNAL_ACCESS_WRITE);
1715 status = ocfs2_journal_access_db(handle,
1718 OCFS2_JOURNAL_ACCESS_WRITE);
1720 if (ocfs2_dir_indexed(dir)) {
1721 status = ocfs2_dx_dir_insert(dir,
1731 /* By now the buffer is marked for journaling */
1732 offset += le16_to_cpu(de->rec_len);
1733 if (le64_to_cpu(de->inode)) {
1734 de1 = (struct ocfs2_dir_entry *)((char *) de +
1735 OCFS2_DIR_REC_LEN(de->name_len));
1737 cpu_to_le16(le16_to_cpu(de->rec_len) -
1738 OCFS2_DIR_REC_LEN(de->name_len));
1739 de->rec_len = cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
1742 de->file_type = OCFS2_FT_UNKNOWN;
1744 de->inode = cpu_to_le64(blkno);
1745 ocfs2_set_de_type(de, inode->i_mode);
1748 de->name_len = namelen;
1749 memcpy(de->name, name, namelen);
1751 if (ocfs2_dir_indexed(dir))
1752 ocfs2_recalc_free_list(dir, handle, lookup);
1755 ocfs2_journal_dirty(handle, insert_bh);
1760 offset += le16_to_cpu(de->rec_len);
1761 de = (struct ocfs2_dir_entry *) ((char *) de + le16_to_cpu(de->rec_len));
1764 /* when you think about it, the assert above should prevent us
1765 * from ever getting here. */
1773 static int ocfs2_dir_foreach_blk_id(struct inode *inode,
1775 loff_t *f_pos, void *priv,
1776 filldir_t filldir, int *filldir_err)
1778 int ret, i, filldir_ret;
1779 unsigned long offset = *f_pos;
1780 struct buffer_head *di_bh = NULL;
1781 struct ocfs2_dinode *di;
1782 struct ocfs2_inline_data *data;
1783 struct ocfs2_dir_entry *de;
1785 ret = ocfs2_read_inode_block(inode, &di_bh);
1787 mlog(ML_ERROR, "Unable to read inode block for dir %llu\n",
1788 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1792 di = (struct ocfs2_dinode *)di_bh->b_data;
1793 data = &di->id2.i_data;
1795 while (*f_pos < i_size_read(inode)) {
1797 /* If the dir block has changed since the last call to
1798 * readdir(2), then we might be pointing to an invalid
1799 * dirent right now. Scan from the start of the block
1801 if (*f_version != inode->i_version) {
1802 for (i = 0; i < i_size_read(inode) && i < offset; ) {
1803 de = (struct ocfs2_dir_entry *)
1804 (data->id_data + i);
1805 /* It's too expensive to do a full
1806 * dirent test each time round this
1807 * loop, but we do have to test at
1808 * least that it is non-zero. A
1809 * failure will be detected in the
1810 * dirent test below. */
1811 if (le16_to_cpu(de->rec_len) <
1812 OCFS2_DIR_REC_LEN(1))
1814 i += le16_to_cpu(de->rec_len);
1816 *f_pos = offset = i;
1817 *f_version = inode->i_version;
1820 de = (struct ocfs2_dir_entry *) (data->id_data + *f_pos);
1821 if (!ocfs2_check_dir_entry(inode, de, di_bh, *f_pos)) {
1822 /* On error, skip the f_pos to the end. */
1823 *f_pos = i_size_read(inode);
1826 offset += le16_to_cpu(de->rec_len);
1827 if (le64_to_cpu(de->inode)) {
1828 /* We might block in the next section
1829 * if the data destination is
1830 * currently swapped out. So, use a
1831 * version stamp to detect whether or
1832 * not the directory has been modified
1833 * during the copy operation.
1835 u64 version = *f_version;
1836 unsigned char d_type = DT_UNKNOWN;
1838 if (de->file_type < OCFS2_FT_MAX)
1839 d_type = ocfs2_filetype_table[de->file_type];
1841 filldir_ret = filldir(priv, de->name,
1844 le64_to_cpu(de->inode),
1848 *filldir_err = filldir_ret;
1851 if (version != *f_version)
1854 *f_pos += le16_to_cpu(de->rec_len);
1864 * NOTE: This function can be called against unindexed directories,
1867 static int ocfs2_dir_foreach_blk_el(struct inode *inode,
1869 loff_t *f_pos, void *priv,
1870 filldir_t filldir, int *filldir_err)
1873 unsigned long offset, blk, last_ra_blk = 0;
1875 struct buffer_head * bh, * tmp;
1876 struct ocfs2_dir_entry * de;
1877 struct super_block * sb = inode->i_sb;
1878 unsigned int ra_sectors = 16;
1883 offset = (*f_pos) & (sb->s_blocksize - 1);
1885 while (!error && !stored && *f_pos < i_size_read(inode)) {
1886 blk = (*f_pos) >> sb->s_blocksize_bits;
1887 if (ocfs2_read_dir_block(inode, blk, &bh, 0)) {
1888 /* Skip the corrupt dirblock and keep trying */
1889 *f_pos += sb->s_blocksize - offset;
1893 /* The idea here is to begin with 8k read-ahead and to stay
1894 * 4k ahead of our current position.
1896 * TODO: Use the pagecache for this. We just need to
1897 * make sure it's cluster-safe... */
1899 || (((last_ra_blk - blk) << 9) <= (ra_sectors / 2))) {
1900 for (i = ra_sectors >> (sb->s_blocksize_bits - 9);
1903 if (!ocfs2_read_dir_block(inode, ++blk, &tmp,
1904 OCFS2_BH_READAHEAD))
1912 /* If the dir block has changed since the last call to
1913 * readdir(2), then we might be pointing to an invalid
1914 * dirent right now. Scan from the start of the block
1916 if (*f_version != inode->i_version) {
1917 for (i = 0; i < sb->s_blocksize && i < offset; ) {
1918 de = (struct ocfs2_dir_entry *) (bh->b_data + i);
1919 /* It's too expensive to do a full
1920 * dirent test each time round this
1921 * loop, but we do have to test at
1922 * least that it is non-zero. A
1923 * failure will be detected in the
1924 * dirent test below. */
1925 if (le16_to_cpu(de->rec_len) <
1926 OCFS2_DIR_REC_LEN(1))
1928 i += le16_to_cpu(de->rec_len);
1931 *f_pos = ((*f_pos) & ~(sb->s_blocksize - 1))
1933 *f_version = inode->i_version;
1936 while (!error && *f_pos < i_size_read(inode)
1937 && offset < sb->s_blocksize) {
1938 de = (struct ocfs2_dir_entry *) (bh->b_data + offset);
1939 if (!ocfs2_check_dir_entry(inode, de, bh, offset)) {
1940 /* On error, skip the f_pos to the
1942 *f_pos = ((*f_pos) | (sb->s_blocksize - 1)) + 1;
1946 offset += le16_to_cpu(de->rec_len);
1947 if (le64_to_cpu(de->inode)) {
1948 /* We might block in the next section
1949 * if the data destination is
1950 * currently swapped out. So, use a
1951 * version stamp to detect whether or
1952 * not the directory has been modified
1953 * during the copy operation.
1955 unsigned long version = *f_version;
1956 unsigned char d_type = DT_UNKNOWN;
1958 if (de->file_type < OCFS2_FT_MAX)
1959 d_type = ocfs2_filetype_table[de->file_type];
1960 error = filldir(priv, de->name,
1963 le64_to_cpu(de->inode),
1967 *filldir_err = error;
1970 if (version != *f_version)
1974 *f_pos += le16_to_cpu(de->rec_len);
1986 static int ocfs2_dir_foreach_blk(struct inode *inode, u64 *f_version,
1987 loff_t *f_pos, void *priv, filldir_t filldir,
1990 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
1991 return ocfs2_dir_foreach_blk_id(inode, f_version, f_pos, priv,
1992 filldir, filldir_err);
1994 return ocfs2_dir_foreach_blk_el(inode, f_version, f_pos, priv, filldir,
1999 * This is intended to be called from inside other kernel functions,
2000 * so we fake some arguments.
2002 int ocfs2_dir_foreach(struct inode *inode, loff_t *f_pos, void *priv,
2005 int ret = 0, filldir_err = 0;
2006 u64 version = inode->i_version;
2008 while (*f_pos < i_size_read(inode)) {
2009 ret = ocfs2_dir_foreach_blk(inode, &version, f_pos, priv,
2010 filldir, &filldir_err);
2011 if (ret || filldir_err)
2025 int ocfs2_readdir(struct file * filp, void * dirent, filldir_t filldir)
2028 struct inode *inode = filp->f_path.dentry->d_inode;
2031 mlog_entry("dirino=%llu\n",
2032 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2034 error = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level);
2035 if (lock_level && error >= 0) {
2036 /* We release EX lock which used to update atime
2037 * and get PR lock again to reduce contention
2038 * on commonly accessed directories. */
2039 ocfs2_inode_unlock(inode, 1);
2041 error = ocfs2_inode_lock(inode, NULL, 0);
2044 if (error != -ENOENT)
2046 /* we haven't got any yet, so propagate the error. */
2050 error = ocfs2_dir_foreach_blk(inode, &filp->f_version, &filp->f_pos,
2051 dirent, filldir, NULL);
2053 ocfs2_inode_unlock(inode, lock_level);
2062 * NOTE: this should always be called with parent dir i_mutex taken.
2064 int ocfs2_find_files_on_disk(const char *name,
2067 struct inode *inode,
2068 struct ocfs2_dir_lookup_result *lookup)
2070 int status = -ENOENT;
2072 mlog(0, "name=%.*s, blkno=%p, inode=%llu\n", namelen, name, blkno,
2073 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2075 status = ocfs2_find_entry(name, namelen, inode, lookup);
2079 *blkno = le64_to_cpu(lookup->dl_entry->inode);
2088 * Convenience function for callers which just want the block number
2089 * mapped to a name and don't require the full dirent info, etc.
2091 int ocfs2_lookup_ino_from_name(struct inode *dir, const char *name,
2092 int namelen, u64 *blkno)
2095 struct ocfs2_dir_lookup_result lookup = { NULL, };
2097 ret = ocfs2_find_files_on_disk(name, namelen, blkno, dir, &lookup);
2098 ocfs2_free_dir_lookup_result(&lookup);
2103 /* Check for a name within a directory.
2105 * Return 0 if the name does not exist
2106 * Return -EEXIST if the directory contains the name
2108 * Callers should have i_mutex + a cluster lock on dir
2110 int ocfs2_check_dir_for_entry(struct inode *dir,
2115 struct ocfs2_dir_lookup_result lookup = { NULL, };
2117 mlog_entry("dir %llu, name '%.*s'\n",
2118 (unsigned long long)OCFS2_I(dir)->ip_blkno, namelen, name);
2121 if (ocfs2_find_entry(name, namelen, dir, &lookup) == 0)
2126 ocfs2_free_dir_lookup_result(&lookup);
2132 struct ocfs2_empty_dir_priv {
2134 unsigned seen_dot_dot;
2135 unsigned seen_other;
2138 static int ocfs2_empty_dir_filldir(void *priv, const char *name, int name_len,
2139 loff_t pos, u64 ino, unsigned type)
2141 struct ocfs2_empty_dir_priv *p = priv;
2144 * Check the positions of "." and ".." records to be sure
2145 * they're in the correct place.
2147 * Indexed directories don't need to proceed past the first
2148 * two entries, so we end the scan after seeing '..'. Despite
2149 * that, we allow the scan to proceed In the event that we
2150 * have a corrupted indexed directory (no dot or dot dot
2151 * entries). This allows us to double check for existing
2152 * entries which might not have been found in the index.
2154 if (name_len == 1 && !strncmp(".", name, 1) && pos == 0) {
2159 if (name_len == 2 && !strncmp("..", name, 2) &&
2160 pos == OCFS2_DIR_REC_LEN(1)) {
2161 p->seen_dot_dot = 1;
2163 if (p->dx_dir && p->seen_dot)
2173 static int ocfs2_empty_dir_dx(struct inode *inode,
2174 struct ocfs2_empty_dir_priv *priv)
2177 struct buffer_head *di_bh = NULL;
2178 struct buffer_head *dx_root_bh = NULL;
2179 struct ocfs2_dinode *di;
2180 struct ocfs2_dx_root_block *dx_root;
2184 ret = ocfs2_read_inode_block(inode, &di_bh);
2189 di = (struct ocfs2_dinode *)di_bh->b_data;
2191 ret = ocfs2_read_dx_root(inode, di, &dx_root_bh);
2196 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
2198 if (le32_to_cpu(dx_root->dr_num_entries) != 2)
2199 priv->seen_other = 1;
2208 * routine to check that the specified directory is empty (for rmdir)
2210 * Returns 1 if dir is empty, zero otherwise.
2212 * XXX: This is a performance problem for unindexed directories.
2214 int ocfs2_empty_dir(struct inode *inode)
2218 struct ocfs2_empty_dir_priv priv;
2220 memset(&priv, 0, sizeof(priv));
2222 if (ocfs2_dir_indexed(inode)) {
2223 ret = ocfs2_empty_dir_dx(inode, &priv);
2227 * We still run ocfs2_dir_foreach to get the checks
2232 ret = ocfs2_dir_foreach(inode, &start, &priv, ocfs2_empty_dir_filldir);
2236 if (!priv.seen_dot || !priv.seen_dot_dot) {
2237 mlog(ML_ERROR, "bad directory (dir #%llu) - no `.' or `..'\n",
2238 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2240 * XXX: Is it really safe to allow an unlink to continue?
2245 return !priv.seen_other;
2249 * Fills "." and ".." dirents in a new directory block. Returns dirent for
2250 * "..", which might be used during creation of a directory with a trailing
2251 * header. It is otherwise safe to ignore the return code.
2253 static struct ocfs2_dir_entry *ocfs2_fill_initial_dirents(struct inode *inode,
2254 struct inode *parent,
2258 struct ocfs2_dir_entry *de = (struct ocfs2_dir_entry *)start;
2260 de->inode = cpu_to_le64(OCFS2_I(inode)->ip_blkno);
2263 cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
2264 strcpy(de->name, ".");
2265 ocfs2_set_de_type(de, S_IFDIR);
2267 de = (struct ocfs2_dir_entry *) ((char *)de + le16_to_cpu(de->rec_len));
2268 de->inode = cpu_to_le64(OCFS2_I(parent)->ip_blkno);
2269 de->rec_len = cpu_to_le16(size - OCFS2_DIR_REC_LEN(1));
2271 strcpy(de->name, "..");
2272 ocfs2_set_de_type(de, S_IFDIR);
2278 * This works together with code in ocfs2_mknod_locked() which sets
2279 * the inline-data flag and initializes the inline-data section.
2281 static int ocfs2_fill_new_dir_id(struct ocfs2_super *osb,
2283 struct inode *parent,
2284 struct inode *inode,
2285 struct buffer_head *di_bh)
2288 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2289 struct ocfs2_inline_data *data = &di->id2.i_data;
2290 unsigned int size = le16_to_cpu(data->id_count);
2292 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2293 OCFS2_JOURNAL_ACCESS_WRITE);
2299 ocfs2_fill_initial_dirents(inode, parent, data->id_data, size);
2300 ocfs2_journal_dirty(handle, di_bh);
2302 i_size_write(inode, size);
2304 inode->i_blocks = ocfs2_inode_sector_count(inode);
2306 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
2314 static int ocfs2_fill_new_dir_el(struct ocfs2_super *osb,
2316 struct inode *parent,
2317 struct inode *inode,
2318 struct buffer_head *fe_bh,
2319 struct ocfs2_alloc_context *data_ac,
2320 struct buffer_head **ret_new_bh)
2323 unsigned int size = osb->sb->s_blocksize;
2324 struct buffer_head *new_bh = NULL;
2325 struct ocfs2_dir_entry *de;
2329 if (ocfs2_new_dir_wants_trailer(inode))
2330 size = ocfs2_dir_trailer_blk_off(parent->i_sb);
2332 status = ocfs2_do_extend_dir(osb->sb, handle, inode, fe_bh,
2333 data_ac, NULL, &new_bh);
2339 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2341 status = ocfs2_journal_access_db(handle, INODE_CACHE(inode), new_bh,
2342 OCFS2_JOURNAL_ACCESS_CREATE);
2347 memset(new_bh->b_data, 0, osb->sb->s_blocksize);
2349 de = ocfs2_fill_initial_dirents(inode, parent, new_bh->b_data, size);
2350 if (ocfs2_new_dir_wants_trailer(inode)) {
2351 int size = le16_to_cpu(de->rec_len);
2354 * Figure out the size of the hole left over after
2355 * insertion of '.' and '..'. The trailer wants this
2358 size -= OCFS2_DIR_REC_LEN(2);
2359 size -= sizeof(struct ocfs2_dir_block_trailer);
2361 ocfs2_init_dir_trailer(inode, new_bh, size);
2364 ocfs2_journal_dirty(handle, new_bh);
2366 i_size_write(inode, inode->i_sb->s_blocksize);
2368 inode->i_blocks = ocfs2_inode_sector_count(inode);
2369 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
2377 *ret_new_bh = new_bh;
2387 static int ocfs2_dx_dir_attach_index(struct ocfs2_super *osb,
2388 handle_t *handle, struct inode *dir,
2389 struct buffer_head *di_bh,
2390 struct buffer_head *dirdata_bh,
2391 struct ocfs2_alloc_context *meta_ac,
2392 int dx_inline, u32 num_entries,
2393 struct buffer_head **ret_dx_root_bh)
2396 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
2397 u16 dr_suballoc_bit;
2399 unsigned int num_bits;
2400 struct buffer_head *dx_root_bh = NULL;
2401 struct ocfs2_dx_root_block *dx_root;
2402 struct ocfs2_dir_block_trailer *trailer =
2403 ocfs2_trailer_from_bh(dirdata_bh, dir->i_sb);
2405 ret = ocfs2_claim_metadata(osb, handle, meta_ac, 1, &dr_suballoc_bit,
2406 &num_bits, &dr_blkno);
2412 mlog(0, "Dir %llu, attach new index block: %llu\n",
2413 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2414 (unsigned long long)dr_blkno);
2416 dx_root_bh = sb_getblk(osb->sb, dr_blkno);
2417 if (dx_root_bh == NULL) {
2421 ocfs2_set_new_buffer_uptodate(INODE_CACHE(dir), dx_root_bh);
2423 ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
2424 OCFS2_JOURNAL_ACCESS_CREATE);
2430 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
2431 memset(dx_root, 0, osb->sb->s_blocksize);
2432 strcpy(dx_root->dr_signature, OCFS2_DX_ROOT_SIGNATURE);
2433 dx_root->dr_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
2434 dx_root->dr_suballoc_bit = cpu_to_le16(dr_suballoc_bit);
2435 dx_root->dr_fs_generation = cpu_to_le32(osb->fs_generation);
2436 dx_root->dr_blkno = cpu_to_le64(dr_blkno);
2437 dx_root->dr_dir_blkno = cpu_to_le64(OCFS2_I(dir)->ip_blkno);
2438 dx_root->dr_num_entries = cpu_to_le32(num_entries);
2439 if (le16_to_cpu(trailer->db_free_rec_len))
2440 dx_root->dr_free_blk = cpu_to_le64(dirdata_bh->b_blocknr);
2442 dx_root->dr_free_blk = cpu_to_le64(0);
2445 dx_root->dr_flags |= OCFS2_DX_FLAG_INLINE;
2446 dx_root->dr_entries.de_count =
2447 cpu_to_le16(ocfs2_dx_entries_per_root(osb->sb));
2449 dx_root->dr_list.l_count =
2450 cpu_to_le16(ocfs2_extent_recs_per_dx_root(osb->sb));
2452 ocfs2_journal_dirty(handle, dx_root_bh);
2454 ret = ocfs2_journal_access_di(handle, INODE_CACHE(dir), di_bh,
2455 OCFS2_JOURNAL_ACCESS_CREATE);
2461 di->i_dx_root = cpu_to_le64(dr_blkno);
2463 OCFS2_I(dir)->ip_dyn_features |= OCFS2_INDEXED_DIR_FL;
2464 di->i_dyn_features = cpu_to_le16(OCFS2_I(dir)->ip_dyn_features);
2466 ocfs2_journal_dirty(handle, di_bh);
2468 *ret_dx_root_bh = dx_root_bh;
2476 static int ocfs2_dx_dir_format_cluster(struct ocfs2_super *osb,
2477 handle_t *handle, struct inode *dir,
2478 struct buffer_head **dx_leaves,
2479 int num_dx_leaves, u64 start_blk)
2482 struct ocfs2_dx_leaf *dx_leaf;
2483 struct buffer_head *bh;
2485 for (i = 0; i < num_dx_leaves; i++) {
2486 bh = sb_getblk(osb->sb, start_blk + i);
2493 ocfs2_set_new_buffer_uptodate(INODE_CACHE(dir), bh);
2495 ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir), bh,
2496 OCFS2_JOURNAL_ACCESS_CREATE);
2502 dx_leaf = (struct ocfs2_dx_leaf *) bh->b_data;
2504 memset(dx_leaf, 0, osb->sb->s_blocksize);
2505 strcpy(dx_leaf->dl_signature, OCFS2_DX_LEAF_SIGNATURE);
2506 dx_leaf->dl_fs_generation = cpu_to_le32(osb->fs_generation);
2507 dx_leaf->dl_blkno = cpu_to_le64(bh->b_blocknr);
2508 dx_leaf->dl_list.de_count =
2509 cpu_to_le16(ocfs2_dx_entries_per_leaf(osb->sb));
2512 "Dir %llu, format dx_leaf: %llu, entry count: %u\n",
2513 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2514 (unsigned long long)bh->b_blocknr,
2515 le16_to_cpu(dx_leaf->dl_list.de_count));
2517 ocfs2_journal_dirty(handle, bh);
2526 * Allocates and formats a new cluster for use in an indexed dir
2527 * leaf. This version will not do the extent insert, so that it can be
2528 * used by operations which need careful ordering.
2530 static int __ocfs2_dx_dir_new_cluster(struct inode *dir,
2531 u32 cpos, handle_t *handle,
2532 struct ocfs2_alloc_context *data_ac,
2533 struct buffer_head **dx_leaves,
2534 int num_dx_leaves, u64 *ret_phys_blkno)
2539 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2542 * XXX: For create, this should claim cluster for the index
2543 * *before* the unindexed insert so that we have a better
2544 * chance of contiguousness as the directory grows in number
2547 ret = __ocfs2_claim_clusters(osb, handle, data_ac, 1, 1, &phys, &num);
2554 * Format the new cluster first. That way, we're inserting
2557 phys_blkno = ocfs2_clusters_to_blocks(osb->sb, phys);
2558 ret = ocfs2_dx_dir_format_cluster(osb, handle, dir, dx_leaves,
2559 num_dx_leaves, phys_blkno);
2565 *ret_phys_blkno = phys_blkno;
2570 static int ocfs2_dx_dir_new_cluster(struct inode *dir,
2571 struct ocfs2_extent_tree *et,
2572 u32 cpos, handle_t *handle,
2573 struct ocfs2_alloc_context *data_ac,
2574 struct ocfs2_alloc_context *meta_ac,
2575 struct buffer_head **dx_leaves,
2581 ret = __ocfs2_dx_dir_new_cluster(dir, cpos, handle, data_ac, dx_leaves,
2582 num_dx_leaves, &phys_blkno);
2588 ret = ocfs2_insert_extent(handle, et, cpos, phys_blkno, 1, 0,
2596 static struct buffer_head **ocfs2_dx_dir_kmalloc_leaves(struct super_block *sb,
2597 int *ret_num_leaves)
2599 int num_dx_leaves = ocfs2_clusters_to_blocks(sb, 1);
2600 struct buffer_head **dx_leaves;
2602 dx_leaves = kcalloc(num_dx_leaves, sizeof(struct buffer_head *),
2604 if (dx_leaves && ret_num_leaves)
2605 *ret_num_leaves = num_dx_leaves;
2610 static int ocfs2_fill_new_dir_dx(struct ocfs2_super *osb,
2612 struct inode *parent,
2613 struct inode *inode,
2614 struct buffer_head *di_bh,
2615 struct ocfs2_alloc_context *data_ac,
2616 struct ocfs2_alloc_context *meta_ac)
2619 struct buffer_head *leaf_bh = NULL;
2620 struct buffer_head *dx_root_bh = NULL;
2621 struct ocfs2_dx_hinfo hinfo;
2622 struct ocfs2_dx_root_block *dx_root;
2623 struct ocfs2_dx_entry_list *entry_list;
2626 * Our strategy is to create the directory as though it were
2627 * unindexed, then add the index block. This works with very
2628 * little complication since the state of a new directory is a
2629 * very well known quantity.
2631 * Essentially, we have two dirents ("." and ".."), in the 1st
2632 * block which need indexing. These are easily inserted into
2636 ret = ocfs2_fill_new_dir_el(osb, handle, parent, inode, di_bh,
2643 ret = ocfs2_dx_dir_attach_index(osb, handle, inode, di_bh, leaf_bh,
2644 meta_ac, 1, 2, &dx_root_bh);
2649 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
2650 entry_list = &dx_root->dr_entries;
2652 /* Buffer has been journaled for us by ocfs2_dx_dir_attach_index */
2653 ocfs2_dx_dir_name_hash(inode, ".", 1, &hinfo);
2654 ocfs2_dx_entry_list_insert(entry_list, &hinfo, leaf_bh->b_blocknr);
2656 ocfs2_dx_dir_name_hash(inode, "..", 2, &hinfo);
2657 ocfs2_dx_entry_list_insert(entry_list, &hinfo, leaf_bh->b_blocknr);
2665 int ocfs2_fill_new_dir(struct ocfs2_super *osb,
2667 struct inode *parent,
2668 struct inode *inode,
2669 struct buffer_head *fe_bh,
2670 struct ocfs2_alloc_context *data_ac,
2671 struct ocfs2_alloc_context *meta_ac)
2674 BUG_ON(!ocfs2_supports_inline_data(osb) && data_ac == NULL);
2676 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2677 return ocfs2_fill_new_dir_id(osb, handle, parent, inode, fe_bh);
2679 if (ocfs2_supports_indexed_dirs(osb))
2680 return ocfs2_fill_new_dir_dx(osb, handle, parent, inode, fe_bh,
2683 return ocfs2_fill_new_dir_el(osb, handle, parent, inode, fe_bh,
2687 static int ocfs2_dx_dir_index_block(struct inode *dir,
2689 struct buffer_head **dx_leaves,
2691 u32 *num_dx_entries,
2692 struct buffer_head *dirent_bh)
2694 int ret = 0, namelen, i;
2695 char *de_buf, *limit;
2696 struct ocfs2_dir_entry *de;
2697 struct buffer_head *dx_leaf_bh;
2698 struct ocfs2_dx_hinfo hinfo;
2699 u64 dirent_blk = dirent_bh->b_blocknr;
2701 de_buf = dirent_bh->b_data;
2702 limit = de_buf + dir->i_sb->s_blocksize;
2704 while (de_buf < limit) {
2705 de = (struct ocfs2_dir_entry *)de_buf;
2707 namelen = de->name_len;
2708 if (!namelen || !de->inode)
2711 ocfs2_dx_dir_name_hash(dir, de->name, namelen, &hinfo);
2713 i = ocfs2_dx_dir_hash_idx(OCFS2_SB(dir->i_sb), &hinfo);
2714 dx_leaf_bh = dx_leaves[i];
2716 ret = __ocfs2_dx_dir_leaf_insert(dir, handle, &hinfo,
2717 dirent_blk, dx_leaf_bh);
2723 *num_dx_entries = *num_dx_entries + 1;
2726 de_buf += le16_to_cpu(de->rec_len);
2734 * XXX: This expects dx_root_bh to already be part of the transaction.
2736 static void ocfs2_dx_dir_index_root_block(struct inode *dir,
2737 struct buffer_head *dx_root_bh,
2738 struct buffer_head *dirent_bh)
2740 char *de_buf, *limit;
2741 struct ocfs2_dx_root_block *dx_root;
2742 struct ocfs2_dir_entry *de;
2743 struct ocfs2_dx_hinfo hinfo;
2744 u64 dirent_blk = dirent_bh->b_blocknr;
2746 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
2748 de_buf = dirent_bh->b_data;
2749 limit = de_buf + dir->i_sb->s_blocksize;
2751 while (de_buf < limit) {
2752 de = (struct ocfs2_dir_entry *)de_buf;
2754 if (!de->name_len || !de->inode)
2757 ocfs2_dx_dir_name_hash(dir, de->name, de->name_len, &hinfo);
2760 "dir: %llu, major: 0x%x minor: 0x%x, index: %u, name: %.*s\n",
2761 (unsigned long long)dir->i_ino, hinfo.major_hash,
2763 le16_to_cpu(dx_root->dr_entries.de_num_used),
2764 de->name_len, de->name);
2766 ocfs2_dx_entry_list_insert(&dx_root->dr_entries, &hinfo,
2769 le32_add_cpu(&dx_root->dr_num_entries, 1);
2771 de_buf += le16_to_cpu(de->rec_len);
2776 * Count the number of inline directory entries in di_bh and compare
2777 * them against the number of entries we can hold in an inline dx root
2780 static int ocfs2_new_dx_should_be_inline(struct inode *dir,
2781 struct buffer_head *di_bh)
2783 int dirent_count = 0;
2784 char *de_buf, *limit;
2785 struct ocfs2_dir_entry *de;
2786 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2788 de_buf = di->id2.i_data.id_data;
2789 limit = de_buf + i_size_read(dir);
2791 while (de_buf < limit) {
2792 de = (struct ocfs2_dir_entry *)de_buf;
2794 if (de->name_len && de->inode)
2797 de_buf += le16_to_cpu(de->rec_len);
2800 /* We are careful to leave room for one extra record. */
2801 return dirent_count < ocfs2_dx_entries_per_root(dir->i_sb);
2805 * Expand rec_len of the rightmost dirent in a directory block so that it
2806 * contains the end of our valid space for dirents. We do this during
2807 * expansion from an inline directory to one with extents. The first dir block
2808 * in that case is taken from the inline data portion of the inode block.
2810 * This will also return the largest amount of contiguous space for a dirent
2811 * in the block. That value is *not* necessarily the last dirent, even after
2812 * expansion. The directory indexing code wants this value for free space
2813 * accounting. We do this here since we're already walking the entire dir
2816 * We add the dir trailer if this filesystem wants it.
2818 static unsigned int ocfs2_expand_last_dirent(char *start, unsigned int old_size,
2821 struct super_block *sb = dir->i_sb;
2822 struct ocfs2_dir_entry *de;
2823 struct ocfs2_dir_entry *prev_de;
2824 char *de_buf, *limit;
2825 unsigned int new_size = sb->s_blocksize;
2826 unsigned int bytes, this_hole;
2827 unsigned int largest_hole = 0;
2829 if (ocfs2_new_dir_wants_trailer(dir))
2830 new_size = ocfs2_dir_trailer_blk_off(sb);
2832 bytes = new_size - old_size;
2834 limit = start + old_size;
2836 de = (struct ocfs2_dir_entry *)de_buf;
2838 this_hole = ocfs2_figure_dirent_hole(de);
2839 if (this_hole > largest_hole)
2840 largest_hole = this_hole;
2843 de_buf += le16_to_cpu(de->rec_len);
2844 de = (struct ocfs2_dir_entry *)de_buf;
2845 } while (de_buf < limit);
2847 le16_add_cpu(&prev_de->rec_len, bytes);
2849 /* We need to double check this after modification of the final
2851 this_hole = ocfs2_figure_dirent_hole(prev_de);
2852 if (this_hole > largest_hole)
2853 largest_hole = this_hole;
2855 if (largest_hole >= OCFS2_DIR_MIN_REC_LEN)
2856 return largest_hole;
2861 * We allocate enough clusters to fulfill "blocks_wanted", but set
2862 * i_size to exactly one block. Ocfs2_extend_dir() will handle the
2863 * rest automatically for us.
2865 * *first_block_bh is a pointer to the 1st data block allocated to the
2868 static int ocfs2_expand_inline_dir(struct inode *dir, struct buffer_head *di_bh,
2869 unsigned int blocks_wanted,
2870 struct ocfs2_dir_lookup_result *lookup,
2871 struct buffer_head **first_block_bh)
2873 u32 alloc, dx_alloc, bit_off, len, num_dx_entries = 0;
2874 struct super_block *sb = dir->i_sb;
2875 int ret, i, num_dx_leaves = 0, dx_inline = 0,
2876 credits = ocfs2_inline_to_extents_credits(sb);
2877 u64 dx_insert_blkno, blkno,
2878 bytes = blocks_wanted << sb->s_blocksize_bits;
2879 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2880 struct ocfs2_inode_info *oi = OCFS2_I(dir);
2881 struct ocfs2_alloc_context *data_ac;
2882 struct ocfs2_alloc_context *meta_ac = NULL;
2883 struct buffer_head *dirdata_bh = NULL;
2884 struct buffer_head *dx_root_bh = NULL;
2885 struct buffer_head **dx_leaves = NULL;
2886 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2888 struct ocfs2_extent_tree et;
2889 struct ocfs2_extent_tree dx_et;
2890 int did_quota = 0, bytes_allocated = 0;
2892 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(dir), di_bh);
2894 alloc = ocfs2_clusters_for_bytes(sb, bytes);
2897 down_write(&oi->ip_alloc_sem);
2899 if (ocfs2_supports_indexed_dirs(osb)) {
2900 credits += ocfs2_add_dir_index_credits(sb);
2902 dx_inline = ocfs2_new_dx_should_be_inline(dir, di_bh);
2904 /* Add one more cluster for an index leaf */
2906 dx_leaves = ocfs2_dx_dir_kmalloc_leaves(sb,
2915 /* This gets us the dx_root */
2916 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
2924 * We should never need more than 2 clusters for the unindexed
2925 * tree - maximum dirent size is far less than one block. In
2926 * fact, the only time we'd need more than one cluster is if
2927 * blocksize == clustersize and the dirent won't fit in the
2928 * extra space that the expansion to a single block gives. As
2929 * of today, that only happens on 4k/4k file systems.
2933 ret = ocfs2_reserve_clusters(osb, alloc + dx_alloc, &data_ac);
2940 * Prepare for worst case allocation scenario of two separate
2941 * extents in the unindexed tree.
2944 credits += OCFS2_SUBALLOC_ALLOC;
2946 handle = ocfs2_start_trans(osb, credits);
2947 if (IS_ERR(handle)) {
2948 ret = PTR_ERR(handle);
2953 ret = dquot_alloc_space_nodirty(dir,
2954 ocfs2_clusters_to_bytes(osb->sb, alloc + dx_alloc));
2959 if (ocfs2_supports_indexed_dirs(osb) && !dx_inline) {
2961 * Allocate our index cluster first, to maximize the
2962 * possibility that unindexed leaves grow
2965 ret = __ocfs2_dx_dir_new_cluster(dir, 0, handle, data_ac,
2966 dx_leaves, num_dx_leaves,
2972 bytes_allocated += ocfs2_clusters_to_bytes(dir->i_sb, 1);
2976 * Try to claim as many clusters as the bitmap can give though
2977 * if we only get one now, that's enough to continue. The rest
2978 * will be claimed after the conversion to extents.
2980 if (ocfs2_dir_resv_allowed(osb))
2981 data_ac->ac_resv = &oi->ip_la_data_resv;
2982 ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off, &len);
2987 bytes_allocated += ocfs2_clusters_to_bytes(dir->i_sb, 1);
2990 * Operations are carefully ordered so that we set up the new
2991 * data block first. The conversion from inline data to
2994 blkno = ocfs2_clusters_to_blocks(dir->i_sb, bit_off);
2995 dirdata_bh = sb_getblk(sb, blkno);
3002 ocfs2_set_new_buffer_uptodate(INODE_CACHE(dir), dirdata_bh);
3004 ret = ocfs2_journal_access_db(handle, INODE_CACHE(dir), dirdata_bh,
3005 OCFS2_JOURNAL_ACCESS_CREATE);
3011 memcpy(dirdata_bh->b_data, di->id2.i_data.id_data, i_size_read(dir));
3012 memset(dirdata_bh->b_data + i_size_read(dir), 0,
3013 sb->s_blocksize - i_size_read(dir));
3014 i = ocfs2_expand_last_dirent(dirdata_bh->b_data, i_size_read(dir), dir);
3015 if (ocfs2_new_dir_wants_trailer(dir)) {
3017 * Prepare the dir trailer up front. It will otherwise look
3018 * like a valid dirent. Even if inserting the index fails
3019 * (unlikely), then all we'll have done is given first dir
3020 * block a small amount of fragmentation.
3022 ocfs2_init_dir_trailer(dir, dirdata_bh, i);
3025 ocfs2_journal_dirty(handle, dirdata_bh);
3027 if (ocfs2_supports_indexed_dirs(osb) && !dx_inline) {
3029 * Dx dirs with an external cluster need to do this up
3030 * front. Inline dx root's get handled later, after
3031 * we've allocated our root block. We get passed back
3032 * a total number of items so that dr_num_entries can
3033 * be correctly set once the dx_root has been
3036 ret = ocfs2_dx_dir_index_block(dir, handle, dx_leaves,
3037 num_dx_leaves, &num_dx_entries,
3046 * Set extent, i_size, etc on the directory. After this, the
3047 * inode should contain the same exact dirents as before and
3048 * be fully accessible from system calls.
3050 * We let the later dirent insert modify c/mtime - to the user
3051 * the data hasn't changed.
3053 ret = ocfs2_journal_access_di(handle, INODE_CACHE(dir), di_bh,
3054 OCFS2_JOURNAL_ACCESS_CREATE);
3060 spin_lock(&oi->ip_lock);
3061 oi->ip_dyn_features &= ~OCFS2_INLINE_DATA_FL;
3062 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
3063 spin_unlock(&oi->ip_lock);
3065 ocfs2_dinode_new_extent_list(dir, di);
3067 i_size_write(dir, sb->s_blocksize);
3068 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3070 di->i_size = cpu_to_le64(sb->s_blocksize);
3071 di->i_ctime = di->i_mtime = cpu_to_le64(dir->i_ctime.tv_sec);
3072 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(dir->i_ctime.tv_nsec);
3075 * This should never fail as our extent list is empty and all
3076 * related blocks have been journaled already.
3078 ret = ocfs2_insert_extent(handle, &et, 0, blkno, len,
3086 * Set i_blocks after the extent insert for the most up to
3087 * date ip_clusters value.
3089 dir->i_blocks = ocfs2_inode_sector_count(dir);
3091 ocfs2_journal_dirty(handle, di_bh);
3093 if (ocfs2_supports_indexed_dirs(osb)) {
3094 ret = ocfs2_dx_dir_attach_index(osb, handle, dir, di_bh,
3095 dirdata_bh, meta_ac, dx_inline,
3096 num_dx_entries, &dx_root_bh);
3103 ocfs2_dx_dir_index_root_block(dir, dx_root_bh,
3106 ocfs2_init_dx_root_extent_tree(&dx_et,
3109 ret = ocfs2_insert_extent(handle, &dx_et, 0,
3110 dx_insert_blkno, 1, 0, NULL);
3117 * We asked for two clusters, but only got one in the 1st
3118 * pass. Claim the 2nd cluster as a separate extent.
3121 ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off,
3127 blkno = ocfs2_clusters_to_blocks(dir->i_sb, bit_off);
3129 ret = ocfs2_insert_extent(handle, &et, 1,
3130 blkno, len, 0, NULL);
3135 bytes_allocated += ocfs2_clusters_to_bytes(dir->i_sb, 1);
3138 *first_block_bh = dirdata_bh;
3140 if (ocfs2_supports_indexed_dirs(osb)) {
3145 * We need to return the correct block within the
3146 * cluster which should hold our entry.
3148 off = ocfs2_dx_dir_hash_idx(OCFS2_SB(dir->i_sb),
3150 get_bh(dx_leaves[off]);
3151 lookup->dl_dx_leaf_bh = dx_leaves[off];
3153 lookup->dl_dx_root_bh = dx_root_bh;
3158 if (ret < 0 && did_quota)
3159 dquot_free_space_nodirty(dir, bytes_allocated);
3161 ocfs2_commit_trans(osb, handle);
3164 up_write(&oi->ip_alloc_sem);
3166 ocfs2_free_alloc_context(data_ac);
3168 ocfs2_free_alloc_context(meta_ac);
3171 for (i = 0; i < num_dx_leaves; i++)
3172 brelse(dx_leaves[i]);
3182 /* returns a bh of the 1st new block in the allocation. */
3183 static int ocfs2_do_extend_dir(struct super_block *sb,
3186 struct buffer_head *parent_fe_bh,
3187 struct ocfs2_alloc_context *data_ac,
3188 struct ocfs2_alloc_context *meta_ac,
3189 struct buffer_head **new_bh)
3192 int extend, did_quota = 0;
3193 u64 p_blkno, v_blkno;
3195 spin_lock(&OCFS2_I(dir)->ip_lock);
3196 extend = (i_size_read(dir) == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters));
3197 spin_unlock(&OCFS2_I(dir)->ip_lock);
3200 u32 offset = OCFS2_I(dir)->ip_clusters;
3202 status = dquot_alloc_space_nodirty(dir,
3203 ocfs2_clusters_to_bytes(sb, 1));
3208 status = ocfs2_add_inode_data(OCFS2_SB(sb), dir, &offset,
3209 1, 0, parent_fe_bh, handle,
3210 data_ac, meta_ac, NULL);
3211 BUG_ON(status == -EAGAIN);
3218 v_blkno = ocfs2_blocks_for_bytes(sb, i_size_read(dir));
3219 status = ocfs2_extent_map_get_blocks(dir, v_blkno, &p_blkno, NULL, NULL);
3225 *new_bh = sb_getblk(sb, p_blkno);
3233 if (did_quota && status < 0)
3234 dquot_free_space_nodirty(dir, ocfs2_clusters_to_bytes(sb, 1));
3240 * Assumes you already have a cluster lock on the directory.
3242 * 'blocks_wanted' is only used if we have an inline directory which
3243 * is to be turned into an extent based one. The size of the dirent to
3244 * insert might be larger than the space gained by growing to just one
3245 * block, so we may have to grow the inode by two blocks in that case.
3247 * If the directory is already indexed, dx_root_bh must be provided.
3249 static int ocfs2_extend_dir(struct ocfs2_super *osb,
3251 struct buffer_head *parent_fe_bh,
3252 unsigned int blocks_wanted,
3253 struct ocfs2_dir_lookup_result *lookup,
3254 struct buffer_head **new_de_bh)
3257 int credits, num_free_extents, drop_alloc_sem = 0;
3259 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
3260 struct ocfs2_extent_list *el = &fe->id2.i_list;
3261 struct ocfs2_alloc_context *data_ac = NULL;
3262 struct ocfs2_alloc_context *meta_ac = NULL;
3263 handle_t *handle = NULL;
3264 struct buffer_head *new_bh = NULL;
3265 struct ocfs2_dir_entry * de;
3266 struct super_block *sb = osb->sb;
3267 struct ocfs2_extent_tree et;
3268 struct buffer_head *dx_root_bh = lookup->dl_dx_root_bh;
3272 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
3274 * This would be a code error as an inline directory should
3275 * never have an index root.
3279 status = ocfs2_expand_inline_dir(dir, parent_fe_bh,
3280 blocks_wanted, lookup,
3287 /* Expansion from inline to an indexed directory will
3288 * have given us this. */
3289 dx_root_bh = lookup->dl_dx_root_bh;
3291 if (blocks_wanted == 1) {
3293 * If the new dirent will fit inside the space
3294 * created by pushing out to one block, then
3295 * we can complete the operation
3296 * here. Otherwise we have to expand i_size
3297 * and format the 2nd block below.
3299 BUG_ON(new_bh == NULL);
3304 * Get rid of 'new_bh' - we want to format the 2nd
3305 * data block and return that instead.
3310 down_write(&OCFS2_I(dir)->ip_alloc_sem);
3312 dir_i_size = i_size_read(dir);
3313 credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS;
3317 down_write(&OCFS2_I(dir)->ip_alloc_sem);
3319 dir_i_size = i_size_read(dir);
3320 mlog(0, "extending dir %llu (i_size = %lld)\n",
3321 (unsigned long long)OCFS2_I(dir)->ip_blkno, dir_i_size);
3323 /* dir->i_size is always block aligned. */
3324 spin_lock(&OCFS2_I(dir)->ip_lock);
3325 if (dir_i_size == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters)) {
3326 spin_unlock(&OCFS2_I(dir)->ip_lock);
3327 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(dir),
3329 num_free_extents = ocfs2_num_free_extents(osb, &et);
3330 if (num_free_extents < 0) {
3331 status = num_free_extents;
3336 if (!num_free_extents) {
3337 status = ocfs2_reserve_new_metadata(osb, el, &meta_ac);
3339 if (status != -ENOSPC)
3345 status = ocfs2_reserve_clusters(osb, 1, &data_ac);
3347 if (status != -ENOSPC)
3352 if (ocfs2_dir_resv_allowed(osb))
3353 data_ac->ac_resv = &OCFS2_I(dir)->ip_la_data_resv;
3355 credits = ocfs2_calc_extend_credits(sb, el, 1);
3357 spin_unlock(&OCFS2_I(dir)->ip_lock);
3358 credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS;
3362 if (ocfs2_dir_indexed(dir))
3363 credits++; /* For attaching the new dirent block to the
3366 handle = ocfs2_start_trans(osb, credits);
3367 if (IS_ERR(handle)) {
3368 status = PTR_ERR(handle);
3374 status = ocfs2_do_extend_dir(osb->sb, handle, dir, parent_fe_bh,
3375 data_ac, meta_ac, &new_bh);
3381 ocfs2_set_new_buffer_uptodate(INODE_CACHE(dir), new_bh);
3383 status = ocfs2_journal_access_db(handle, INODE_CACHE(dir), new_bh,
3384 OCFS2_JOURNAL_ACCESS_CREATE);
3389 memset(new_bh->b_data, 0, sb->s_blocksize);
3391 de = (struct ocfs2_dir_entry *) new_bh->b_data;
3393 if (ocfs2_supports_dir_trailer(dir)) {
3394 de->rec_len = cpu_to_le16(ocfs2_dir_trailer_blk_off(sb));
3396 ocfs2_init_dir_trailer(dir, new_bh, le16_to_cpu(de->rec_len));
3398 if (ocfs2_dir_indexed(dir)) {
3399 status = ocfs2_dx_dir_link_trailer(dir, handle,
3400 dx_root_bh, new_bh);
3407 de->rec_len = cpu_to_le16(sb->s_blocksize);
3409 ocfs2_journal_dirty(handle, new_bh);
3411 dir_i_size += dir->i_sb->s_blocksize;
3412 i_size_write(dir, dir_i_size);
3413 dir->i_blocks = ocfs2_inode_sector_count(dir);
3414 status = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
3421 *new_de_bh = new_bh;
3425 ocfs2_commit_trans(osb, handle);
3427 up_write(&OCFS2_I(dir)->ip_alloc_sem);
3430 ocfs2_free_alloc_context(data_ac);
3432 ocfs2_free_alloc_context(meta_ac);
3440 static int ocfs2_find_dir_space_id(struct inode *dir, struct buffer_head *di_bh,
3441 const char *name, int namelen,
3442 struct buffer_head **ret_de_bh,
3443 unsigned int *blocks_wanted)
3446 struct super_block *sb = dir->i_sb;
3447 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3448 struct ocfs2_dir_entry *de, *last_de = NULL;
3449 char *de_buf, *limit;
3450 unsigned long offset = 0;
3451 unsigned int rec_len, new_rec_len, free_space = dir->i_sb->s_blocksize;
3454 * This calculates how many free bytes we'd have in block zero, should
3455 * this function force expansion to an extent tree.
3457 if (ocfs2_new_dir_wants_trailer(dir))
3458 free_space = ocfs2_dir_trailer_blk_off(sb) - i_size_read(dir);
3460 free_space = dir->i_sb->s_blocksize - i_size_read(dir);
3462 de_buf = di->id2.i_data.id_data;
3463 limit = de_buf + i_size_read(dir);
3464 rec_len = OCFS2_DIR_REC_LEN(namelen);
3466 while (de_buf < limit) {
3467 de = (struct ocfs2_dir_entry *)de_buf;
3469 if (!ocfs2_check_dir_entry(dir, de, di_bh, offset)) {
3473 if (ocfs2_match(namelen, name, de)) {
3478 * No need to check for a trailing dirent record here as
3479 * they're not used for inline dirs.
3482 if (ocfs2_dirent_would_fit(de, rec_len)) {
3483 /* Ok, we found a spot. Return this bh and let
3484 * the caller actually fill it in. */
3492 de_buf += le16_to_cpu(de->rec_len);
3493 offset += le16_to_cpu(de->rec_len);
3497 * We're going to require expansion of the directory - figure
3498 * out how many blocks we'll need so that a place for the
3499 * dirent can be found.
3502 new_rec_len = le16_to_cpu(last_de->rec_len) + free_space;
3503 if (new_rec_len < (rec_len + OCFS2_DIR_REC_LEN(last_de->name_len)))
3511 static int ocfs2_find_dir_space_el(struct inode *dir, const char *name,
3512 int namelen, struct buffer_head **ret_de_bh)
3514 unsigned long offset;
3515 struct buffer_head *bh = NULL;
3516 unsigned short rec_len;
3517 struct ocfs2_dir_entry *de;
3518 struct super_block *sb = dir->i_sb;
3520 int blocksize = dir->i_sb->s_blocksize;
3522 status = ocfs2_read_dir_block(dir, 0, &bh, 0);
3528 rec_len = OCFS2_DIR_REC_LEN(namelen);
3530 de = (struct ocfs2_dir_entry *) bh->b_data;
3532 if ((char *)de >= sb->s_blocksize + bh->b_data) {
3536 if (i_size_read(dir) <= offset) {
3538 * Caller will have to expand this
3544 status = ocfs2_read_dir_block(dir,
3545 offset >> sb->s_blocksize_bits,
3551 /* move to next block */
3552 de = (struct ocfs2_dir_entry *) bh->b_data;
3554 if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
3558 if (ocfs2_match(namelen, name, de)) {
3563 if (ocfs2_skip_dir_trailer(dir, de, offset % blocksize,
3567 if (ocfs2_dirent_would_fit(de, rec_len)) {
3568 /* Ok, we found a spot. Return this bh and let
3569 * the caller actually fill it in. */
3576 offset += le16_to_cpu(de->rec_len);
3577 de = (struct ocfs2_dir_entry *)((char *) de + le16_to_cpu(de->rec_len));
3588 static int dx_leaf_sort_cmp(const void *a, const void *b)
3590 const struct ocfs2_dx_entry *entry1 = a;
3591 const struct ocfs2_dx_entry *entry2 = b;
3592 u32 major_hash1 = le32_to_cpu(entry1->dx_major_hash);
3593 u32 major_hash2 = le32_to_cpu(entry2->dx_major_hash);
3594 u32 minor_hash1 = le32_to_cpu(entry1->dx_minor_hash);
3595 u32 minor_hash2 = le32_to_cpu(entry2->dx_minor_hash);
3597 if (major_hash1 > major_hash2)
3599 if (major_hash1 < major_hash2)
3603 * It is not strictly necessary to sort by minor
3605 if (minor_hash1 > minor_hash2)
3607 if (minor_hash1 < minor_hash2)
3612 static void dx_leaf_sort_swap(void *a, void *b, int size)
3614 struct ocfs2_dx_entry *entry1 = a;
3615 struct ocfs2_dx_entry *entry2 = b;
3616 struct ocfs2_dx_entry tmp;
3618 BUG_ON(size != sizeof(*entry1));
3625 static int ocfs2_dx_leaf_same_major(struct ocfs2_dx_leaf *dx_leaf)
3627 struct ocfs2_dx_entry_list *dl_list = &dx_leaf->dl_list;
3628 int i, num = le16_to_cpu(dl_list->de_num_used);
3630 for (i = 0; i < (num - 1); i++) {
3631 if (le32_to_cpu(dl_list->de_entries[i].dx_major_hash) !=
3632 le32_to_cpu(dl_list->de_entries[i + 1].dx_major_hash))
3640 * Find the optimal value to split this leaf on. This expects the leaf
3641 * entries to be in sorted order.
3643 * leaf_cpos is the cpos of the leaf we're splitting. insert_hash is
3644 * the hash we want to insert.
3646 * This function is only concerned with the major hash - that which
3647 * determines which cluster an item belongs to.
3649 static int ocfs2_dx_dir_find_leaf_split(struct ocfs2_dx_leaf *dx_leaf,
3650 u32 leaf_cpos, u32 insert_hash,
3653 struct ocfs2_dx_entry_list *dl_list = &dx_leaf->dl_list;
3654 int i, num_used = le16_to_cpu(dl_list->de_num_used);
3658 * There's a couple rare, but nasty corner cases we have to
3659 * check for here. All of them involve a leaf where all value
3660 * have the same hash, which is what we look for first.
3662 * Most of the time, all of the above is false, and we simply
3663 * pick the median value for a split.
3665 allsame = ocfs2_dx_leaf_same_major(dx_leaf);
3667 u32 val = le32_to_cpu(dl_list->de_entries[0].dx_major_hash);
3669 if (val == insert_hash) {
3671 * No matter where we would choose to split,
3672 * the new entry would want to occupy the same
3673 * block as these. Since there's no space left
3674 * in their existing block, we know there
3675 * won't be space after the split.
3680 if (val == leaf_cpos) {
3682 * Because val is the same as leaf_cpos (which
3683 * is the smallest value this leaf can have),
3684 * yet is not equal to insert_hash, then we
3685 * know that insert_hash *must* be larger than
3686 * val (and leaf_cpos). At least cpos+1 in value.
3688 * We also know then, that there cannot be an
3689 * adjacent extent (otherwise we'd be looking
3690 * at it). Choosing this value gives us a
3691 * chance to get some contiguousness.
3693 *split_hash = leaf_cpos + 1;
3697 if (val > insert_hash) {
3699 * val can not be the same as insert hash, and
3700 * also must be larger than leaf_cpos. Also,
3701 * we know that there can't be a leaf between
3702 * cpos and val, otherwise the entries with
3703 * hash 'val' would be there.
3709 *split_hash = insert_hash;
3714 * Since the records are sorted and the checks above
3715 * guaranteed that not all records in this block are the same,
3716 * we simple travel forward, from the median, and pick the 1st
3717 * record whose value is larger than leaf_cpos.
3719 for (i = (num_used / 2); i < num_used; i++)
3720 if (le32_to_cpu(dl_list->de_entries[i].dx_major_hash) >
3724 BUG_ON(i == num_used); /* Should be impossible */
3725 *split_hash = le32_to_cpu(dl_list->de_entries[i].dx_major_hash);
3730 * Transfer all entries in orig_dx_leaves whose major hash is equal to or
3731 * larger than split_hash into new_dx_leaves. We use a temporary
3732 * buffer (tmp_dx_leaf) to make the changes to the original leaf blocks.
3734 * Since the block offset inside a leaf (cluster) is a constant mask
3735 * of minor_hash, we can optimize - an item at block offset X within
3736 * the original cluster, will be at offset X within the new cluster.
3738 static void ocfs2_dx_dir_transfer_leaf(struct inode *dir, u32 split_hash,
3740 struct ocfs2_dx_leaf *tmp_dx_leaf,
3741 struct buffer_head **orig_dx_leaves,
3742 struct buffer_head **new_dx_leaves,
3747 struct ocfs2_dx_leaf *orig_dx_leaf, *new_dx_leaf;
3748 struct ocfs2_dx_entry_list *orig_list, *new_list, *tmp_list;
3749 struct ocfs2_dx_entry *dx_entry;
3751 tmp_list = &tmp_dx_leaf->dl_list;
3753 for (i = 0; i < num_dx_leaves; i++) {
3754 orig_dx_leaf = (struct ocfs2_dx_leaf *) orig_dx_leaves[i]->b_data;
3755 orig_list = &orig_dx_leaf->dl_list;
3756 new_dx_leaf = (struct ocfs2_dx_leaf *) new_dx_leaves[i]->b_data;
3757 new_list = &new_dx_leaf->dl_list;
3759 num_used = le16_to_cpu(orig_list->de_num_used);
3761 memcpy(tmp_dx_leaf, orig_dx_leaf, dir->i_sb->s_blocksize);
3762 tmp_list->de_num_used = cpu_to_le16(0);
3763 memset(&tmp_list->de_entries, 0, sizeof(*dx_entry)*num_used);
3765 for (j = 0; j < num_used; j++) {
3766 dx_entry = &orig_list->de_entries[j];
3767 major_hash = le32_to_cpu(dx_entry->dx_major_hash);
3768 if (major_hash >= split_hash)
3769 ocfs2_dx_dir_leaf_insert_tail(new_dx_leaf,
3772 ocfs2_dx_dir_leaf_insert_tail(tmp_dx_leaf,
3775 memcpy(orig_dx_leaf, tmp_dx_leaf, dir->i_sb->s_blocksize);
3777 ocfs2_journal_dirty(handle, orig_dx_leaves[i]);
3778 ocfs2_journal_dirty(handle, new_dx_leaves[i]);
3782 static int ocfs2_dx_dir_rebalance_credits(struct ocfs2_super *osb,
3783 struct ocfs2_dx_root_block *dx_root)
3785 int credits = ocfs2_clusters_to_blocks(osb->sb, 2);
3787 credits += ocfs2_calc_extend_credits(osb->sb, &dx_root->dr_list, 1);
3788 credits += ocfs2_quota_trans_credits(osb->sb);
3793 * Find the median value in dx_leaf_bh and allocate a new leaf to move
3794 * half our entries into.
3796 static int ocfs2_dx_dir_rebalance(struct ocfs2_super *osb, struct inode *dir,
3797 struct buffer_head *dx_root_bh,
3798 struct buffer_head *dx_leaf_bh,
3799 struct ocfs2_dx_hinfo *hinfo, u32 leaf_cpos,
3802 struct ocfs2_dx_leaf *dx_leaf = (struct ocfs2_dx_leaf *)dx_leaf_bh->b_data;
3803 int credits, ret, i, num_used, did_quota = 0;
3804 u32 cpos, split_hash, insert_hash = hinfo->major_hash;
3805 u64 orig_leaves_start;
3807 struct buffer_head **orig_dx_leaves = NULL;
3808 struct buffer_head **new_dx_leaves = NULL;
3809 struct ocfs2_alloc_context *data_ac = NULL, *meta_ac = NULL;
3810 struct ocfs2_extent_tree et;
3811 handle_t *handle = NULL;
3812 struct ocfs2_dx_root_block *dx_root;
3813 struct ocfs2_dx_leaf *tmp_dx_leaf = NULL;
3815 mlog(0, "DX Dir: %llu, rebalance leaf leaf_blkno: %llu insert: %u\n",
3816 (unsigned long long)OCFS2_I(dir)->ip_blkno,
3817 (unsigned long long)leaf_blkno, insert_hash);
3819 ocfs2_init_dx_root_extent_tree(&et, INODE_CACHE(dir), dx_root_bh);
3821 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
3823 * XXX: This is a rather large limit. We should use a more
3826 if (le32_to_cpu(dx_root->dr_clusters) == UINT_MAX)
3829 num_used = le16_to_cpu(dx_leaf->dl_list.de_num_used);
3830 if (num_used < le16_to_cpu(dx_leaf->dl_list.de_count)) {
3831 mlog(ML_ERROR, "DX Dir: %llu, Asked to rebalance empty leaf: "
3832 "%llu, %d\n", (unsigned long long)OCFS2_I(dir)->ip_blkno,
3833 (unsigned long long)leaf_blkno, num_used);
3838 orig_dx_leaves = ocfs2_dx_dir_kmalloc_leaves(osb->sb, &num_dx_leaves);
3839 if (!orig_dx_leaves) {
3845 new_dx_leaves = ocfs2_dx_dir_kmalloc_leaves(osb->sb, NULL);
3846 if (!new_dx_leaves) {
3852 ret = ocfs2_lock_allocators(dir, &et, 1, 0, &data_ac, &meta_ac);
3859 credits = ocfs2_dx_dir_rebalance_credits(osb, dx_root);
3860 handle = ocfs2_start_trans(osb, credits);
3861 if (IS_ERR(handle)) {
3862 ret = PTR_ERR(handle);
3868 ret = dquot_alloc_space_nodirty(dir,
3869 ocfs2_clusters_to_bytes(dir->i_sb, 1));
3874 ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir), dx_leaf_bh,
3875 OCFS2_JOURNAL_ACCESS_WRITE);
3882 * This block is changing anyway, so we can sort it in place.
3884 sort(dx_leaf->dl_list.de_entries, num_used,
3885 sizeof(struct ocfs2_dx_entry), dx_leaf_sort_cmp,
3888 ocfs2_journal_dirty(handle, dx_leaf_bh);
3890 ret = ocfs2_dx_dir_find_leaf_split(dx_leaf, leaf_cpos, insert_hash,
3897 mlog(0, "Split leaf (%u) at %u, insert major hash is %u\n",
3898 leaf_cpos, split_hash, insert_hash);
3901 * We have to carefully order operations here. There are items
3902 * which want to be in the new cluster before insert, but in
3903 * order to put those items in the new cluster, we alter the
3904 * old cluster. A failure to insert gets nasty.
3906 * So, start by reserving writes to the old
3907 * cluster. ocfs2_dx_dir_new_cluster will reserve writes on
3908 * the new cluster for us, before inserting it. The insert
3909 * won't happen if there's an error before that. Once the
3910 * insert is done then, we can transfer from one leaf into the
3911 * other without fear of hitting any error.
3915 * The leaf transfer wants some scratch space so that we don't
3916 * wind up doing a bunch of expensive memmove().
3918 tmp_dx_leaf = kmalloc(osb->sb->s_blocksize, GFP_NOFS);
3925 orig_leaves_start = ocfs2_block_to_cluster_start(dir->i_sb, leaf_blkno);
3926 ret = ocfs2_read_dx_leaves(dir, orig_leaves_start, num_dx_leaves,
3933 for (i = 0; i < num_dx_leaves; i++) {
3934 ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir),
3936 OCFS2_JOURNAL_ACCESS_WRITE);
3944 ret = ocfs2_dx_dir_new_cluster(dir, &et, cpos, handle,
3945 data_ac, meta_ac, new_dx_leaves,
3952 ocfs2_dx_dir_transfer_leaf(dir, split_hash, handle, tmp_dx_leaf,
3953 orig_dx_leaves, new_dx_leaves, num_dx_leaves);
3956 if (ret < 0 && did_quota)
3957 dquot_free_space_nodirty(dir,
3958 ocfs2_clusters_to_bytes(dir->i_sb, 1));
3960 ocfs2_commit_trans(osb, handle);
3963 if (orig_dx_leaves || new_dx_leaves) {
3964 for (i = 0; i < num_dx_leaves; i++) {
3966 brelse(orig_dx_leaves[i]);
3968 brelse(new_dx_leaves[i]);
3970 kfree(orig_dx_leaves);
3971 kfree(new_dx_leaves);
3975 ocfs2_free_alloc_context(meta_ac);
3977 ocfs2_free_alloc_context(data_ac);
3983 static int ocfs2_find_dir_space_dx(struct ocfs2_super *osb, struct inode *dir,
3984 struct buffer_head *di_bh,
3985 struct buffer_head *dx_root_bh,
3986 const char *name, int namelen,
3987 struct ocfs2_dir_lookup_result *lookup)
3989 int ret, rebalanced = 0;
3990 struct ocfs2_dx_root_block *dx_root;
3991 struct buffer_head *dx_leaf_bh = NULL;
3992 struct ocfs2_dx_leaf *dx_leaf;
3996 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
3999 ret = ocfs2_dx_dir_lookup(dir, &dx_root->dr_list, &lookup->dl_hinfo,
4000 &leaf_cpos, &blkno);
4006 ret = ocfs2_read_dx_leaf(dir, blkno, &dx_leaf_bh);
4012 dx_leaf = (struct ocfs2_dx_leaf *)dx_leaf_bh->b_data;
4014 if (le16_to_cpu(dx_leaf->dl_list.de_num_used) >=
4015 le16_to_cpu(dx_leaf->dl_list.de_count)) {
4018 * Rebalancing should have provided us with
4019 * space in an appropriate leaf.
4021 * XXX: Is this an abnormal condition then?
4022 * Should we print a message here?
4028 ret = ocfs2_dx_dir_rebalance(osb, dir, dx_root_bh, dx_leaf_bh,
4029 &lookup->dl_hinfo, leaf_cpos,
4038 * Restart the lookup. The rebalance might have
4039 * changed which block our item fits into. Mark our
4040 * progress, so we only execute this once.
4045 goto restart_search;
4048 lookup->dl_dx_leaf_bh = dx_leaf_bh;
4056 static int ocfs2_search_dx_free_list(struct inode *dir,
4057 struct buffer_head *dx_root_bh,
4059 struct ocfs2_dir_lookup_result *lookup)
4062 struct buffer_head *leaf_bh = NULL, *prev_leaf_bh = NULL;
4063 struct ocfs2_dir_block_trailer *db;
4065 int rec_len = OCFS2_DIR_REC_LEN(namelen);
4066 struct ocfs2_dx_root_block *dx_root;
4068 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
4069 next_block = le64_to_cpu(dx_root->dr_free_blk);
4071 while (next_block) {
4072 brelse(prev_leaf_bh);
4073 prev_leaf_bh = leaf_bh;
4076 ret = ocfs2_read_dir_block_direct(dir, next_block, &leaf_bh);
4082 db = ocfs2_trailer_from_bh(leaf_bh, dir->i_sb);
4083 if (rec_len <= le16_to_cpu(db->db_free_rec_len)) {
4084 lookup->dl_leaf_bh = leaf_bh;
4085 lookup->dl_prev_leaf_bh = prev_leaf_bh;
4087 prev_leaf_bh = NULL;
4091 next_block = le64_to_cpu(db->db_free_next);
4100 brelse(prev_leaf_bh);
4104 static int ocfs2_expand_inline_dx_root(struct inode *dir,
4105 struct buffer_head *dx_root_bh)
4107 int ret, num_dx_leaves, i, j, did_quota = 0;
4108 struct buffer_head **dx_leaves = NULL;
4109 struct ocfs2_extent_tree et;
4111 struct ocfs2_alloc_context *data_ac = NULL;
4112 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
4113 handle_t *handle = NULL;
4114 struct ocfs2_dx_root_block *dx_root;
4115 struct ocfs2_dx_entry_list *entry_list;
4116 struct ocfs2_dx_entry *dx_entry;
4117 struct ocfs2_dx_leaf *target_leaf;
4119 ret = ocfs2_reserve_clusters(osb, 1, &data_ac);
4125 dx_leaves = ocfs2_dx_dir_kmalloc_leaves(osb->sb, &num_dx_leaves);
4132 handle = ocfs2_start_trans(osb, ocfs2_calc_dxi_expand_credits(osb->sb));
4133 if (IS_ERR(handle)) {
4134 ret = PTR_ERR(handle);
4139 ret = dquot_alloc_space_nodirty(dir,
4140 ocfs2_clusters_to_bytes(osb->sb, 1));
4146 * We do this up front, before the allocation, so that a
4147 * failure to add the dx_root_bh to the journal won't result
4148 * us losing clusters.
4150 ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
4151 OCFS2_JOURNAL_ACCESS_WRITE);
4157 ret = __ocfs2_dx_dir_new_cluster(dir, 0, handle, data_ac, dx_leaves,
4158 num_dx_leaves, &insert_blkno);
4165 * Transfer the entries from our dx_root into the appropriate
4168 dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data;
4169 entry_list = &dx_root->dr_entries;
4171 for (i = 0; i < le16_to_cpu(entry_list->de_num_used); i++) {
4172 dx_entry = &entry_list->de_entries[i];
4174 j = __ocfs2_dx_dir_hash_idx(osb,
4175 le32_to_cpu(dx_entry->dx_minor_hash));
4176 target_leaf = (struct ocfs2_dx_leaf *)dx_leaves[j]->b_data;
4178 ocfs2_dx_dir_leaf_insert_tail(target_leaf, dx_entry);
4180 /* Each leaf has been passed to the journal already
4181 * via __ocfs2_dx_dir_new_cluster() */
4184 dx_root->dr_flags &= ~OCFS2_DX_FLAG_INLINE;
4185 memset(&dx_root->dr_list, 0, osb->sb->s_blocksize -
4186 offsetof(struct ocfs2_dx_root_block, dr_list));
4187 dx_root->dr_list.l_count =
4188 cpu_to_le16(ocfs2_extent_recs_per_dx_root(osb->sb));
4190 /* This should never fail considering we start with an empty
4192 ocfs2_init_dx_root_extent_tree(&et, INODE_CACHE(dir), dx_root_bh);
4193 ret = ocfs2_insert_extent(handle, &et, 0, insert_blkno, 1, 0, NULL);
4198 ocfs2_journal_dirty(handle, dx_root_bh);
4201 if (ret < 0 && did_quota)
4202 dquot_free_space_nodirty(dir,
4203 ocfs2_clusters_to_bytes(dir->i_sb, 1));
4205 ocfs2_commit_trans(osb, handle);
4209 ocfs2_free_alloc_context(data_ac);
4212 for (i = 0; i < num_dx_leaves; i++)
4213 brelse(dx_leaves[i]);
4219 static int ocfs2_inline_dx_has_space(struct buffer_head *dx_root_bh)
4221 struct ocfs2_dx_root_block *dx_root;
4222 struct ocfs2_dx_entry_list *entry_list;
4224 dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data;
4225 entry_list = &dx_root->dr_entries;
4227 if (le16_to_cpu(entry_list->de_num_used) >=
4228 le16_to_cpu(entry_list->de_count))
4234 static int ocfs2_prepare_dx_dir_for_insert(struct inode *dir,
4235 struct buffer_head *di_bh,
4238 struct ocfs2_dir_lookup_result *lookup)
4240 int ret, free_dx_root = 1;
4241 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
4242 struct buffer_head *dx_root_bh = NULL;
4243 struct buffer_head *leaf_bh = NULL;
4244 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
4245 struct ocfs2_dx_root_block *dx_root;
4247 ret = ocfs2_read_dx_root(dir, di, &dx_root_bh);
4253 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
4254 if (le32_to_cpu(dx_root->dr_num_entries) == OCFS2_DX_ENTRIES_MAX) {
4260 if (ocfs2_dx_root_inline(dx_root)) {
4261 ret = ocfs2_inline_dx_has_space(dx_root_bh);
4267 * We ran out of room in the root block. Expand it to
4268 * an extent, then allow ocfs2_find_dir_space_dx to do
4271 ret = ocfs2_expand_inline_dx_root(dir, dx_root_bh);
4279 * Insert preparation for an indexed directory is split into two
4280 * steps. The call to find_dir_space_dx reserves room in the index for
4281 * an additional item. If we run out of space there, it's a real error
4282 * we can't continue on.
4284 ret = ocfs2_find_dir_space_dx(osb, dir, di_bh, dx_root_bh, name,
4293 * Next, we need to find space in the unindexed tree. This call
4294 * searches using the free space linked list. If the unindexed tree
4295 * lacks sufficient space, we'll expand it below. The expansion code
4296 * is smart enough to add any new blocks to the free space list.
4298 ret = ocfs2_search_dx_free_list(dir, dx_root_bh, namelen, lookup);
4299 if (ret && ret != -ENOSPC) {
4304 /* Do this up here - ocfs2_extend_dir might need the dx_root */
4305 lookup->dl_dx_root_bh = dx_root_bh;
4308 if (ret == -ENOSPC) {
4309 ret = ocfs2_extend_dir(osb, dir, di_bh, 1, lookup, &leaf_bh);
4317 * We make the assumption here that new leaf blocks are added
4318 * to the front of our free list.
4320 lookup->dl_prev_leaf_bh = NULL;
4321 lookup->dl_leaf_bh = leaf_bh;
4331 * Get a directory ready for insert. Any directory allocation required
4332 * happens here. Success returns zero, and enough context in the dir
4333 * lookup result that ocfs2_add_entry() will be able complete the task
4334 * with minimal performance impact.
4336 int ocfs2_prepare_dir_for_insert(struct ocfs2_super *osb,
4338 struct buffer_head *parent_fe_bh,
4341 struct ocfs2_dir_lookup_result *lookup)
4344 unsigned int blocks_wanted = 1;
4345 struct buffer_head *bh = NULL;
4347 mlog(0, "getting ready to insert namelen %d into dir %llu\n",
4348 namelen, (unsigned long long)OCFS2_I(dir)->ip_blkno);
4357 * Do this up front to reduce confusion.
4359 * The directory might start inline, then be turned into an
4360 * indexed one, in which case we'd need to hash deep inside
4361 * ocfs2_find_dir_space_id(). Since
4362 * ocfs2_prepare_dx_dir_for_insert() also needs this hash
4363 * done, there seems no point in spreading out the calls. We
4364 * can optimize away the case where the file system doesn't
4367 if (ocfs2_supports_indexed_dirs(osb))
4368 ocfs2_dx_dir_name_hash(dir, name, namelen, &lookup->dl_hinfo);
4370 if (ocfs2_dir_indexed(dir)) {
4371 ret = ocfs2_prepare_dx_dir_for_insert(dir, parent_fe_bh,
4372 name, namelen, lookup);
4378 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
4379 ret = ocfs2_find_dir_space_id(dir, parent_fe_bh, name,
4380 namelen, &bh, &blocks_wanted);
4382 ret = ocfs2_find_dir_space_el(dir, name, namelen, &bh);
4384 if (ret && ret != -ENOSPC) {
4389 if (ret == -ENOSPC) {
4391 * We have to expand the directory to add this name.
4395 ret = ocfs2_extend_dir(osb, dir, parent_fe_bh, blocks_wanted,
4406 lookup->dl_leaf_bh = bh;
4413 static int ocfs2_dx_dir_remove_index(struct inode *dir,
4414 struct buffer_head *di_bh,
4415 struct buffer_head *dx_root_bh)
4418 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
4419 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
4420 struct ocfs2_dx_root_block *dx_root;
4421 struct inode *dx_alloc_inode = NULL;
4422 struct buffer_head *dx_alloc_bh = NULL;
4428 dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data;
4430 dx_alloc_inode = ocfs2_get_system_file_inode(osb,
4431 EXTENT_ALLOC_SYSTEM_INODE,
4432 le16_to_cpu(dx_root->dr_suballoc_slot));
4433 if (!dx_alloc_inode) {
4438 mutex_lock(&dx_alloc_inode->i_mutex);
4440 ret = ocfs2_inode_lock(dx_alloc_inode, &dx_alloc_bh, 1);
4446 handle = ocfs2_start_trans(osb, OCFS2_DX_ROOT_REMOVE_CREDITS);
4447 if (IS_ERR(handle)) {
4448 ret = PTR_ERR(handle);
4453 ret = ocfs2_journal_access_di(handle, INODE_CACHE(dir), di_bh,
4454 OCFS2_JOURNAL_ACCESS_WRITE);
4460 OCFS2_I(dir)->ip_dyn_features &= ~OCFS2_INDEXED_DIR_FL;
4461 di->i_dyn_features = cpu_to_le16(OCFS2_I(dir)->ip_dyn_features);
4462 di->i_dx_root = cpu_to_le64(0ULL);
4464 ocfs2_journal_dirty(handle, di_bh);
4466 blk = le64_to_cpu(dx_root->dr_blkno);
4467 bit = le16_to_cpu(dx_root->dr_suballoc_bit);
4468 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
4469 ret = ocfs2_free_suballoc_bits(handle, dx_alloc_inode, dx_alloc_bh,
4475 ocfs2_commit_trans(osb, handle);
4478 ocfs2_inode_unlock(dx_alloc_inode, 1);
4481 mutex_unlock(&dx_alloc_inode->i_mutex);
4482 brelse(dx_alloc_bh);
4484 iput(dx_alloc_inode);
4488 int ocfs2_dx_dir_truncate(struct inode *dir, struct buffer_head *di_bh)
4491 unsigned int uninitialized_var(clen);
4492 u32 major_hash = UINT_MAX, p_cpos, uninitialized_var(cpos);
4493 u64 uninitialized_var(blkno);
4494 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
4495 struct buffer_head *dx_root_bh = NULL;
4496 struct ocfs2_dx_root_block *dx_root;
4497 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
4498 struct ocfs2_cached_dealloc_ctxt dealloc;
4499 struct ocfs2_extent_tree et;
4501 ocfs2_init_dealloc_ctxt(&dealloc);
4503 if (!ocfs2_dir_indexed(dir))
4506 ret = ocfs2_read_dx_root(dir, di, &dx_root_bh);
4511 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
4513 if (ocfs2_dx_root_inline(dx_root))
4516 ocfs2_init_dx_root_extent_tree(&et, INODE_CACHE(dir), dx_root_bh);
4518 /* XXX: What if dr_clusters is too large? */
4519 while (le32_to_cpu(dx_root->dr_clusters)) {
4520 ret = ocfs2_dx_dir_lookup_rec(dir, &dx_root->dr_list,
4521 major_hash, &cpos, &blkno, &clen);
4527 p_cpos = ocfs2_blocks_to_clusters(dir->i_sb, blkno);
4529 ret = ocfs2_remove_btree_range(dir, &et, cpos, p_cpos, clen, 0,
4539 major_hash = cpos - 1;
4543 ret = ocfs2_dx_dir_remove_index(dir, di_bh, dx_root_bh);
4549 ocfs2_remove_from_cache(INODE_CACHE(dir), dx_root_bh);
4551 ocfs2_schedule_truncate_log_flush(osb, 1);
4552 ocfs2_run_deallocs(osb, &dealloc);