blob: 5e6aeb00cb43fa86ee483d5deb806427013263af [file] [log] [blame]
Mark Fashehccd979b2005-12-15 14:31:24 -08001/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * dir.c
5 *
6 * Creates, reads, walks and deletes directory-nodes
7 *
8 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
9 *
10 * Portions of this code from linux/fs/ext3/dir.c
11 *
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)
16 *
17 * from
18 *
19 * linux/fs/minix/dir.c
20 *
21 * Copyright (C) 1991, 1992 Linux Torvalds
22 *
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.
27 *
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.
32 *
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.
37 */
38
39#include <linux/fs.h>
40#include <linux/types.h>
41#include <linux/slab.h>
42#include <linux/highmem.h>
Jan Karaa90714c2008-10-09 19:38:40 +020043#include <linux/quotaops.h>
Mark Fasheh9b7895e2008-11-12 16:27:44 -080044#include <linux/sort.h>
Mark Fashehccd979b2005-12-15 14:31:24 -080045
46#define MLOG_MASK_PREFIX ML_NAMEI
47#include <cluster/masklog.h>
48
49#include "ocfs2.h"
50
51#include "alloc.h"
Joel Beckerc175a512008-12-10 17:58:22 -080052#include "blockcheck.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080053#include "dir.h"
54#include "dlmglue.h"
55#include "extent_map.h"
56#include "file.h"
57#include "inode.h"
58#include "journal.h"
59#include "namei.h"
60#include "suballoc.h"
Mark Fasheh316f4b92007-09-07 18:21:26 -070061#include "super.h"
Mark Fasheh9b7895e2008-11-12 16:27:44 -080062#include "sysfile.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080063#include "uptodate.h"
64
65#include "buffer_head_io.h"
66
Mark Fasheh316f4b92007-09-07 18:21:26 -070067#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))
71
Mark Fashehccd979b2005-12-15 14:31:24 -080072static unsigned char ocfs2_filetype_table[] = {
73 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
74};
75
Mark Fasheh316f4b92007-09-07 18:21:26 -070076static int ocfs2_do_extend_dir(struct super_block *sb,
77 handle_t *handle,
78 struct inode *dir,
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);
Mark Fashehe7c17e42009-01-29 18:17:46 -080083static int ocfs2_dir_indexed(struct inode *inode);
Mark Fasheh316f4b92007-09-07 18:21:26 -070084
Mark Fasheh23193e52007-09-12 13:01:18 -070085/*
Mark Fasheh87d35a72008-12-10 17:36:25 -080086 * These are distinct checks because future versions of the file system will
87 * want to have a trailing dirent structure independent of indexing.
88 */
Mark Fashehe7c17e42009-01-29 18:17:46 -080089static int ocfs2_supports_dir_trailer(struct inode *dir)
Mark Fasheh87d35a72008-12-10 17:36:25 -080090{
Mark Fashehe7c17e42009-01-29 18:17:46 -080091 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
92
Mark Fasheh87d35a72008-12-10 17:36:25 -080093 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
94 return 0;
95
Mark Fashehe7c17e42009-01-29 18:17:46 -080096 return ocfs2_meta_ecc(osb) || ocfs2_dir_indexed(dir);
Mark Fasheh87d35a72008-12-10 17:36:25 -080097}
98
Mark Fashehe7c17e42009-01-29 18:17:46 -080099/*
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()
106 */
107static int ocfs2_new_dir_wants_trailer(struct inode *dir)
Mark Fasheh87d35a72008-12-10 17:36:25 -0800108{
Mark Fashehe7c17e42009-01-29 18:17:46 -0800109 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
110
111 return ocfs2_meta_ecc(osb) ||
112 ocfs2_supports_indexed_dirs(osb);
Mark Fasheh87d35a72008-12-10 17:36:25 -0800113}
114
115static inline unsigned int ocfs2_dir_trailer_blk_off(struct super_block *sb)
116{
117 return sb->s_blocksize - sizeof(struct ocfs2_dir_block_trailer);
118}
119
120#define ocfs2_trailer_from_bh(_bh, _sb) ((struct ocfs2_dir_block_trailer *) ((_bh)->b_data + ocfs2_dir_trailer_blk_off((_sb))))
121
Joel Beckerc175a512008-12-10 17:58:22 -0800122/* XXX ocfs2_block_dqtrailer() is similar but not quite - can we make
123 * them more consistent? */
124struct ocfs2_dir_block_trailer *ocfs2_dir_trailer_from_size(int blocksize,
125 void *data)
126{
127 char *p = data;
128
129 p += blocksize - sizeof(struct ocfs2_dir_block_trailer);
130 return (struct ocfs2_dir_block_trailer *)p;
131}
132
Mark Fasheh87d35a72008-12-10 17:36:25 -0800133/*
134 * XXX: This is executed once on every dirent. We should consider optimizing
135 * it.
136 */
137static int ocfs2_skip_dir_trailer(struct inode *dir,
138 struct ocfs2_dir_entry *de,
139 unsigned long offset,
140 unsigned long blklen)
141{
142 unsigned long toff = blklen - sizeof(struct ocfs2_dir_block_trailer);
143
Mark Fashehe7c17e42009-01-29 18:17:46 -0800144 if (!ocfs2_supports_dir_trailer(dir))
Mark Fasheh87d35a72008-12-10 17:36:25 -0800145 return 0;
146
147 if (offset != toff)
148 return 0;
149
150 return 1;
151}
152
153static void ocfs2_init_dir_trailer(struct inode *inode,
Mark Fashehe7c17e42009-01-29 18:17:46 -0800154 struct buffer_head *bh, u16 rec_len)
Mark Fasheh87d35a72008-12-10 17:36:25 -0800155{
156 struct ocfs2_dir_block_trailer *trailer;
157
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);
Mark Fashehe7c17e42009-01-29 18:17:46 -0800164 trailer->db_free_rec_len = cpu_to_le16(rec_len);
165}
166/*
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.
170 */
171static 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)
174{
175 int ret;
176 struct ocfs2_dx_root_block *dx_root;
177 struct ocfs2_dir_block_trailer *trailer;
178
179 ret = ocfs2_journal_access_dr(handle, dir, dx_root_bh,
180 OCFS2_JOURNAL_ACCESS_WRITE);
181 if (ret) {
182 mlog_errno(ret);
183 goto out;
184 }
185 trailer = ocfs2_trailer_from_bh(dirdata_bh, dir->i_sb);
186 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
187
188 trailer->db_free_next = dx_root->dr_free_blk;
189 dx_root->dr_free_blk = cpu_to_le64(dirdata_bh->b_blocknr);
190
191 ocfs2_journal_dirty(handle, dx_root_bh);
192
193out:
194 return ret;
195}
196
197static int ocfs2_free_list_at_root(struct ocfs2_dir_lookup_result *res)
198{
199 return res->dl_prev_leaf_bh == NULL;
Mark Fasheh87d35a72008-12-10 17:36:25 -0800200}
201
Mark Fasheh4a12ca32008-11-12 15:43:34 -0800202void ocfs2_free_dir_lookup_result(struct ocfs2_dir_lookup_result *res)
203{
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -0800204 brelse(res->dl_dx_root_bh);
Mark Fasheh4a12ca32008-11-12 15:43:34 -0800205 brelse(res->dl_leaf_bh);
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800206 brelse(res->dl_dx_leaf_bh);
Mark Fashehe7c17e42009-01-29 18:17:46 -0800207 brelse(res->dl_prev_leaf_bh);
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800208}
209
210static int ocfs2_dir_indexed(struct inode *inode)
211{
212 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INDEXED_DIR_FL)
213 return 1;
214 return 0;
215}
216
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -0800217static inline int ocfs2_dx_root_inline(struct ocfs2_dx_root_block *dx_root)
218{
219 return dx_root->dr_flags & OCFS2_DX_FLAG_INLINE;
220}
221
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800222/*
223 * Hashing code adapted from ext3
224 */
225#define DELTA 0x9E3779B9
226
227static void TEA_transform(__u32 buf[4], __u32 const in[])
228{
229 __u32 sum = 0;
230 __u32 b0 = buf[0], b1 = buf[1];
231 __u32 a = in[0], b = in[1], c = in[2], d = in[3];
232 int n = 16;
233
234 do {
235 sum += DELTA;
236 b0 += ((b1 << 4)+a) ^ (b1+sum) ^ ((b1 >> 5)+b);
237 b1 += ((b0 << 4)+c) ^ (b0+sum) ^ ((b0 >> 5)+d);
238 } while (--n);
239
240 buf[0] += b0;
241 buf[1] += b1;
242}
243
244static void str2hashbuf(const char *msg, int len, __u32 *buf, int num)
245{
246 __u32 pad, val;
247 int i;
248
249 pad = (__u32)len | ((__u32)len << 8);
250 pad |= pad << 16;
251
252 val = pad;
253 if (len > num*4)
254 len = num * 4;
255 for (i = 0; i < len; i++) {
256 if ((i % 4) == 0)
257 val = pad;
258 val = msg[i] + (val << 8);
259 if ((i % 4) == 3) {
260 *buf++ = val;
261 val = pad;
262 num--;
263 }
264 }
265 if (--num >= 0)
266 *buf++ = val;
267 while (--num >= 0)
268 *buf++ = pad;
269}
270
271static void ocfs2_dx_dir_name_hash(struct inode *dir, const char *name, int len,
272 struct ocfs2_dx_hinfo *hinfo)
273{
274 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
275 const char *p;
276 __u32 in[8], buf[4];
277
278 /*
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?
281 */
282 if ((len == 1 && !strncmp(".", name, 1)) ||
283 (len == 2 && !strncmp("..", name, 2))) {
284 buf[0] = buf[1] = 0;
285 goto out;
286 }
287
288#ifdef OCFS2_DEBUG_DX_DIRS
289 /*
290 * This makes it very easy to debug indexing problems. We
291 * should never allow this to be selected without hand editing
292 * this file though.
293 */
294 buf[0] = buf[1] = len;
295 goto out;
296#endif
297
298 memcpy(buf, osb->osb_dx_seed, sizeof(buf));
299
300 p = name;
301 while (len > 0) {
302 str2hashbuf(p, len, in, 4);
303 TEA_transform(buf, in);
304 len -= 16;
305 p += 16;
306 }
307
308out:
309 hinfo->major_hash = buf[0];
310 hinfo->minor_hash = buf[1];
Mark Fasheh4a12ca32008-11-12 15:43:34 -0800311}
312
Mark Fasheh87d35a72008-12-10 17:36:25 -0800313/*
Mark Fasheh23193e52007-09-12 13:01:18 -0700314 * bh passed here can be an inode block or a dir data block, depending
315 * on the inode inline data flag.
316 */
Mark Fasheh5eae5b92007-09-10 17:50:51 -0700317static int ocfs2_check_dir_entry(struct inode * dir,
318 struct ocfs2_dir_entry * de,
319 struct buffer_head * bh,
320 unsigned long offset)
Mark Fasheh316f4b92007-09-07 18:21:26 -0700321{
322 const char *error_msg = NULL;
323 const int rlen = le16_to_cpu(de->rec_len);
324
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";
333
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,
339 de->name_len);
340 return error_msg == NULL ? 1 : 0;
341}
342
343static inline int ocfs2_match(int len,
344 const char * const name,
345 struct ocfs2_dir_entry *de)
346{
347 if (len != de->name_len)
348 return 0;
349 if (!de->inode)
350 return 0;
351 return !memcmp(name, de->name, len);
352}
353
354/*
355 * Returns 0 if not found, -1 on failure, and 1 on success
356 */
357static int inline ocfs2_search_dirblock(struct buffer_head *bh,
358 struct inode *dir,
359 const char *name, int namelen,
360 unsigned long offset,
Mark Fasheh23193e52007-09-12 13:01:18 -0700361 char *first_de,
362 unsigned int bytes,
Mark Fasheh316f4b92007-09-07 18:21:26 -0700363 struct ocfs2_dir_entry **res_dir)
364{
365 struct ocfs2_dir_entry *de;
366 char *dlimit, *de_buf;
367 int de_len;
368 int ret = 0;
369
370 mlog_entry_void();
371
Mark Fasheh23193e52007-09-12 13:01:18 -0700372 de_buf = first_de;
373 dlimit = de_buf + bytes;
Mark Fasheh316f4b92007-09-07 18:21:26 -0700374
375 while (de_buf < dlimit) {
376 /* this code is executed quadratically often */
377 /* do minimal checking `by hand' */
378
379 de = (struct ocfs2_dir_entry *) de_buf;
380
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)) {
385 ret = -1;
386 goto bail;
387 }
388 *res_dir = de;
389 ret = 1;
390 goto bail;
391 }
392
393 /* prevent looping on a bad block */
394 de_len = le16_to_cpu(de->rec_len);
395 if (de_len <= 0) {
396 ret = -1;
397 goto bail;
398 }
399
400 de_buf += de_len;
401 offset += de_len;
402 }
403
404bail:
405 mlog_exit(ret);
406 return ret;
407}
408
Mark Fasheh23193e52007-09-12 13:01:18 -0700409static struct buffer_head *ocfs2_find_entry_id(const char *name,
410 int namelen,
411 struct inode *dir,
412 struct ocfs2_dir_entry **res_dir)
413{
414 int ret, found;
415 struct buffer_head *di_bh = NULL;
416 struct ocfs2_dinode *di;
417 struct ocfs2_inline_data *data;
418
Joel Beckerb657c952008-11-13 14:49:11 -0800419 ret = ocfs2_read_inode_block(dir, &di_bh);
Mark Fasheh23193e52007-09-12 13:01:18 -0700420 if (ret) {
421 mlog_errno(ret);
422 goto out;
423 }
424
425 di = (struct ocfs2_dinode *)di_bh->b_data;
426 data = &di->id2.i_data;
427
428 found = ocfs2_search_dirblock(di_bh, dir, name, namelen, 0,
429 data->id_data, i_size_read(dir), res_dir);
430 if (found == 1)
431 return di_bh;
432
433 brelse(di_bh);
434out:
435 return NULL;
436}
437
Joel Beckera22305c2008-11-13 14:49:17 -0800438static int ocfs2_validate_dir_block(struct super_block *sb,
439 struct buffer_head *bh)
440{
Joel Beckerc175a512008-12-10 17:58:22 -0800441 int rc;
442 struct ocfs2_dir_block_trailer *trailer =
443 ocfs2_trailer_from_bh(bh, sb);
444
445
Joel Beckera22305c2008-11-13 14:49:17 -0800446 /*
Joel Beckerc175a512008-12-10 17:58:22 -0800447 * We don't validate dirents here, that's handled
Joel Beckera22305c2008-11-13 14:49:17 -0800448 * in-place when the code walks them.
449 */
Joel Becker970e4932008-11-13 14:49:19 -0800450 mlog(0, "Validating dirblock %llu\n",
451 (unsigned long long)bh->b_blocknr);
Joel Beckera22305c2008-11-13 14:49:17 -0800452
Joel Beckerc175a512008-12-10 17:58:22 -0800453 BUG_ON(!buffer_uptodate(bh));
454
455 /*
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.
459 *
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.
463 */
464 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &trailer->db_check);
465 if (rc)
466 mlog(ML_ERROR, "Checksum failed for dinode %llu\n",
467 (unsigned long long)bh->b_blocknr);
468
469 return rc;
Joel Beckera22305c2008-11-13 14:49:17 -0800470}
471
472/*
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800473 * Validate a directory trailer.
474 *
475 * We check the trailer here rather than in ocfs2_validate_dir_block()
476 * because that function doesn't have the inode to test.
477 */
478static int ocfs2_check_dir_trailer(struct inode *dir, struct buffer_head *bh)
479{
480 int rc = 0;
481 struct ocfs2_dir_block_trailer *trailer;
482
483 trailer = ocfs2_trailer_from_bh(bh, dir->i_sb);
484 if (!OCFS2_IS_VALID_DIR_TRAILER(trailer)) {
485 rc = -EINVAL;
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);
491 goto out;
492 }
493 if (le64_to_cpu(trailer->db_blkno) != bh->b_blocknr) {
494 rc = -EINVAL;
495 ocfs2_error(dir->i_sb,
496 "Directory block #%llu has an invalid "
497 "db_blkno of %llu",
498 (unsigned long long)bh->b_blocknr,
499 (unsigned long long)le64_to_cpu(trailer->db_blkno));
500 goto out;
501 }
502 if (le64_to_cpu(trailer->db_parent_dinode) !=
503 OCFS2_I(dir)->ip_blkno) {
504 rc = -EINVAL;
505 ocfs2_error(dir->i_sb,
506 "Directory block #%llu on dinode "
507 "#%llu has an invalid parent_dinode "
508 "of %llu",
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));
512 goto out;
513 }
514out:
515 return rc;
516}
517
518/*
Joel Beckera22305c2008-11-13 14:49:17 -0800519 * 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.
523 */
524static int ocfs2_read_dir_block(struct inode *inode, u64 v_block,
525 struct buffer_head **bh, int flags)
526{
527 int rc = 0;
528 struct buffer_head *tmp = *bh;
Joel Beckera22305c2008-11-13 14:49:17 -0800529
Joel Becker511308d2008-11-13 14:49:21 -0800530 rc = ocfs2_read_virt_blocks(inode, v_block, 1, &tmp, flags,
531 ocfs2_validate_dir_block);
Mark Fasheh87d35a72008-12-10 17:36:25 -0800532 if (rc) {
Joel Beckera22305c2008-11-13 14:49:17 -0800533 mlog_errno(rc);
Mark Fasheh87d35a72008-12-10 17:36:25 -0800534 goto out;
535 }
536
Mark Fasheh87d35a72008-12-10 17:36:25 -0800537 if (!(flags & OCFS2_BH_READAHEAD) &&
Mark Fashehe7c17e42009-01-29 18:17:46 -0800538 ocfs2_supports_dir_trailer(inode)) {
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800539 rc = ocfs2_check_dir_trailer(inode, tmp);
540 if (rc) {
541 if (!*bh)
542 brelse(tmp);
543 mlog_errno(rc);
Mark Fasheh87d35a72008-12-10 17:36:25 -0800544 goto out;
545 }
546 }
Joel Beckera22305c2008-11-13 14:49:17 -0800547
Joel Becker511308d2008-11-13 14:49:21 -0800548 /* If ocfs2_read_virt_blocks() got us a new bh, pass it up. */
Mark Fasheh87d35a72008-12-10 17:36:25 -0800549 if (!*bh)
Joel Beckera22305c2008-11-13 14:49:17 -0800550 *bh = tmp;
551
Mark Fasheh87d35a72008-12-10 17:36:25 -0800552out:
Joel Beckera22305c2008-11-13 14:49:17 -0800553 return rc ? -EIO : 0;
554}
555
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800556/*
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.
560 */
561static int ocfs2_read_dir_block_direct(struct inode *dir, u64 phys,
562 struct buffer_head **bh)
563{
564 int ret;
565 struct buffer_head *tmp = *bh;
566
567 ret = ocfs2_read_block(dir, phys, &tmp, ocfs2_validate_dir_block);
568 if (ret) {
569 mlog_errno(ret);
570 goto out;
571 }
572
573 if (ocfs2_supports_dir_trailer(dir)) {
574 ret = ocfs2_check_dir_trailer(dir, tmp);
575 if (ret) {
576 if (!*bh)
577 brelse(tmp);
578 mlog_errno(ret);
579 goto out;
580 }
581 }
582
583 if (!ret && !*bh)
584 *bh = tmp;
585out:
586 return ret;
587}
588
589static int ocfs2_validate_dx_root(struct super_block *sb,
590 struct buffer_head *bh)
591{
592 int ret;
593 struct ocfs2_dx_root_block *dx_root;
594
595 BUG_ON(!buffer_uptodate(bh));
596
597 dx_root = (struct ocfs2_dx_root_block *) bh->b_data;
598
599 ret = ocfs2_validate_meta_ecc(sb, bh->b_data, &dx_root->dr_check);
600 if (ret) {
601 mlog(ML_ERROR,
602 "Checksum failed for dir index root block %llu\n",
603 (unsigned long long)bh->b_blocknr);
604 return ret;
605 }
606
607 if (!OCFS2_IS_VALID_DX_ROOT(dx_root)) {
608 ocfs2_error(sb,
609 "Dir Index Root # %llu has bad signature %.*s",
610 (unsigned long long)le64_to_cpu(dx_root->dr_blkno),
611 7, dx_root->dr_signature);
612 return -EINVAL;
613 }
614
615 return 0;
616}
617
618static int ocfs2_read_dx_root(struct inode *dir, struct ocfs2_dinode *di,
619 struct buffer_head **dx_root_bh)
620{
621 int ret;
622 u64 blkno = le64_to_cpu(di->i_dx_root);
623 struct buffer_head *tmp = *dx_root_bh;
624
625 ret = ocfs2_read_block(dir, blkno, &tmp, ocfs2_validate_dx_root);
626
627 /* If ocfs2_read_block() got us a new bh, pass it up. */
628 if (!ret && !*dx_root_bh)
629 *dx_root_bh = tmp;
630
631 return ret;
632}
633
634static int ocfs2_validate_dx_leaf(struct super_block *sb,
635 struct buffer_head *bh)
636{
637 int ret;
638 struct ocfs2_dx_leaf *dx_leaf = (struct ocfs2_dx_leaf *)bh->b_data;
639
640 BUG_ON(!buffer_uptodate(bh));
641
642 ret = ocfs2_validate_meta_ecc(sb, bh->b_data, &dx_leaf->dl_check);
643 if (ret) {
644 mlog(ML_ERROR,
645 "Checksum failed for dir index leaf block %llu\n",
646 (unsigned long long)bh->b_blocknr);
647 return ret;
648 }
649
650 if (!OCFS2_IS_VALID_DX_LEAF(dx_leaf)) {
651 ocfs2_error(sb, "Dir Index Leaf has bad signature %.*s",
652 7, dx_leaf->dl_signature);
653 return -EROFS;
654 }
655
656 return 0;
657}
658
659static int ocfs2_read_dx_leaf(struct inode *dir, u64 blkno,
660 struct buffer_head **dx_leaf_bh)
661{
662 int ret;
663 struct buffer_head *tmp = *dx_leaf_bh;
664
665 ret = ocfs2_read_block(dir, blkno, &tmp, ocfs2_validate_dx_leaf);
666
667 /* If ocfs2_read_block() got us a new bh, pass it up. */
668 if (!ret && !*dx_leaf_bh)
669 *dx_leaf_bh = tmp;
670
671 return ret;
672}
673
674/*
675 * Read a series of dx_leaf blocks. This expects all buffer_head
676 * pointers to be NULL on function entry.
677 */
678static int ocfs2_read_dx_leaves(struct inode *dir, u64 start, int num,
679 struct buffer_head **dx_leaf_bhs)
680{
681 int ret;
682
683 ret = ocfs2_read_blocks(dir, start, num, dx_leaf_bhs, 0,
684 ocfs2_validate_dx_leaf);
685 if (ret)
686 mlog_errno(ret);
687
688 return ret;
689}
690
Adrian Bunk0af4bd32007-10-24 18:23:27 +0200691static struct buffer_head *ocfs2_find_entry_el(const char *name, int namelen,
692 struct inode *dir,
693 struct ocfs2_dir_entry **res_dir)
Mark Fasheh316f4b92007-09-07 18:21:26 -0700694{
695 struct super_block *sb;
696 struct buffer_head *bh_use[NAMEI_RA_SIZE];
697 struct buffer_head *bh, *ret = NULL;
698 unsigned long start, block, b;
699 int ra_max = 0; /* Number of bh's in the readahead
700 buffer, bh_use[] */
701 int ra_ptr = 0; /* Current index into readahead
702 buffer */
703 int num = 0;
704 int nblocks, i, err;
705
706 mlog_entry_void();
707
Mark Fasheh316f4b92007-09-07 18:21:26 -0700708 sb = dir->i_sb;
709
710 nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
711 start = OCFS2_I(dir)->ip_dir_start_lookup;
712 if (start >= nblocks)
713 start = 0;
714 block = start;
715
716restart:
717 do {
718 /*
719 * We deal with the read-ahead logic here.
720 */
721 if (ra_ptr >= ra_max) {
722 /* Refill the readahead buffer */
723 ra_ptr = 0;
724 b = block;
725 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
726 /*
727 * Terminate if we reach the end of the
728 * directory and must wrap, or if our
729 * search has finished at this block.
730 */
731 if (b >= nblocks || (num && block == start)) {
732 bh_use[ra_max] = NULL;
733 break;
734 }
735 num++;
736
Joel Beckera22305c2008-11-13 14:49:17 -0800737 bh = NULL;
738 err = ocfs2_read_dir_block(dir, b++, &bh,
739 OCFS2_BH_READAHEAD);
Mark Fasheh316f4b92007-09-07 18:21:26 -0700740 bh_use[ra_max] = bh;
741 }
742 }
743 if ((bh = bh_use[ra_ptr++]) == NULL)
744 goto next;
Joel Beckera22305c2008-11-13 14:49:17 -0800745 if (ocfs2_read_dir_block(dir, block, &bh, 0)) {
Joel Becker5e0b3de2008-10-09 17:20:33 -0700746 /* read error, skip block & hope for the best.
Joel Beckera22305c2008-11-13 14:49:17 -0800747 * ocfs2_read_dir_block() has released the bh. */
Mark Fasheh316f4b92007-09-07 18:21:26 -0700748 ocfs2_error(dir->i_sb, "reading directory %llu, "
749 "offset %lu\n",
750 (unsigned long long)OCFS2_I(dir)->ip_blkno,
751 block);
Mark Fasheh316f4b92007-09-07 18:21:26 -0700752 goto next;
753 }
754 i = ocfs2_search_dirblock(bh, dir, name, namelen,
755 block << sb->s_blocksize_bits,
Mark Fasheh23193e52007-09-12 13:01:18 -0700756 bh->b_data, sb->s_blocksize,
Mark Fasheh316f4b92007-09-07 18:21:26 -0700757 res_dir);
758 if (i == 1) {
759 OCFS2_I(dir)->ip_dir_start_lookup = block;
760 ret = bh;
761 goto cleanup_and_exit;
762 } else {
763 brelse(bh);
764 if (i < 0)
765 goto cleanup_and_exit;
766 }
767 next:
768 if (++block >= nblocks)
769 block = 0;
770 } while (block != start);
771
772 /*
773 * If the directory has grown while we were searching, then
774 * search the last part of the directory before giving up.
775 */
776 block = nblocks;
777 nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
778 if (block < nblocks) {
779 start = 0;
780 goto restart;
781 }
782
783cleanup_and_exit:
784 /* Clean up the read-ahead blocks */
785 for (; ra_ptr < ra_max; ra_ptr++)
786 brelse(bh_use[ra_ptr]);
787
788 mlog_exit_ptr(ret);
789 return ret;
790}
791
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800792static int ocfs2_dx_dir_lookup_rec(struct inode *inode,
793 struct ocfs2_extent_list *el,
794 u32 major_hash,
795 u32 *ret_cpos,
796 u64 *ret_phys_blkno,
797 unsigned int *ret_clen)
798{
799 int ret = 0, i, found;
800 struct buffer_head *eb_bh = NULL;
801 struct ocfs2_extent_block *eb;
802 struct ocfs2_extent_rec *rec = NULL;
803
804 if (el->l_tree_depth) {
805 ret = ocfs2_find_leaf(inode, el, major_hash, &eb_bh);
806 if (ret) {
807 mlog_errno(ret);
808 goto out;
809 }
810
811 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
812 el = &eb->h_list;
813
814 if (el->l_tree_depth) {
815 ocfs2_error(inode->i_sb,
816 "Inode %lu has non zero tree depth in "
817 "btree tree block %llu\n", inode->i_ino,
818 (unsigned long long)eb_bh->b_blocknr);
819 ret = -EROFS;
820 goto out;
821 }
822 }
823
824 found = 0;
825 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
826 rec = &el->l_recs[i];
827
828 if (le32_to_cpu(rec->e_cpos) <= major_hash) {
829 found = 1;
830 break;
831 }
832 }
833
834 if (!found) {
835 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
836 "record (%u, %u, 0) in btree", inode->i_ino,
837 le32_to_cpu(rec->e_cpos),
838 ocfs2_rec_clusters(el, rec));
839 ret = -EROFS;
840 goto out;
841 }
842
843 if (ret_phys_blkno)
844 *ret_phys_blkno = le64_to_cpu(rec->e_blkno);
845 if (ret_cpos)
846 *ret_cpos = le32_to_cpu(rec->e_cpos);
847 if (ret_clen)
848 *ret_clen = le16_to_cpu(rec->e_leaf_clusters);
849
850out:
851 brelse(eb_bh);
852 return ret;
853}
854
855/*
856 * Returns the block index, from the start of the cluster which this
857 * hash belongs too.
858 */
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -0800859static inline unsigned int __ocfs2_dx_dir_hash_idx(struct ocfs2_super *osb,
860 u32 minor_hash)
861{
862 return minor_hash & osb->osb_dx_mask;
863}
864
865static inline unsigned int ocfs2_dx_dir_hash_idx(struct ocfs2_super *osb,
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800866 struct ocfs2_dx_hinfo *hinfo)
867{
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -0800868 return __ocfs2_dx_dir_hash_idx(osb, hinfo->minor_hash);
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800869}
870
871static int ocfs2_dx_dir_lookup(struct inode *inode,
872 struct ocfs2_extent_list *el,
873 struct ocfs2_dx_hinfo *hinfo,
874 u32 *ret_cpos,
875 u64 *ret_phys_blkno)
876{
877 int ret = 0;
878 unsigned int cend, uninitialized_var(clen);
879 u32 uninitialized_var(cpos);
880 u64 uninitialized_var(blkno);
881 u32 name_hash = hinfo->major_hash;
882
883 ret = ocfs2_dx_dir_lookup_rec(inode, el, name_hash, &cpos, &blkno,
884 &clen);
885 if (ret) {
886 mlog_errno(ret);
887 goto out;
888 }
889
890 cend = cpos + clen;
891 if (name_hash >= cend) {
892 /* We want the last cluster */
893 blkno += ocfs2_clusters_to_blocks(inode->i_sb, clen - 1);
894 cpos += clen - 1;
895 } else {
896 blkno += ocfs2_clusters_to_blocks(inode->i_sb,
897 name_hash - cpos);
898 cpos = name_hash;
899 }
900
901 /*
902 * We now have the cluster which should hold our entry. To
903 * find the exact block from the start of the cluster to
904 * search, we take the lower bits of the hash.
905 */
906 blkno += ocfs2_dx_dir_hash_idx(OCFS2_SB(inode->i_sb), hinfo);
907
908 if (ret_phys_blkno)
909 *ret_phys_blkno = blkno;
910 if (ret_cpos)
911 *ret_cpos = cpos;
912
913out:
914
915 return ret;
916}
917
918static int ocfs2_dx_dir_search(const char *name, int namelen,
919 struct inode *dir,
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -0800920 struct ocfs2_dx_root_block *dx_root,
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800921 struct ocfs2_dir_lookup_result *res)
922{
923 int ret, i, found;
924 u64 uninitialized_var(phys);
925 struct buffer_head *dx_leaf_bh = NULL;
926 struct ocfs2_dx_leaf *dx_leaf;
927 struct ocfs2_dx_entry *dx_entry = NULL;
928 struct buffer_head *dir_ent_bh = NULL;
929 struct ocfs2_dir_entry *dir_ent = NULL;
930 struct ocfs2_dx_hinfo *hinfo = &res->dl_hinfo;
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -0800931 struct ocfs2_extent_list *dr_el;
932 struct ocfs2_dx_entry_list *entry_list;
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800933
934 ocfs2_dx_dir_name_hash(dir, name, namelen, &res->dl_hinfo);
935
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -0800936 if (ocfs2_dx_root_inline(dx_root)) {
937 entry_list = &dx_root->dr_entries;
938 goto search;
939 }
940
941 dr_el = &dx_root->dr_list;
942
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800943 ret = ocfs2_dx_dir_lookup(dir, dr_el, hinfo, NULL, &phys);
944 if (ret) {
945 mlog_errno(ret);
946 goto out;
947 }
948
949 mlog(0, "Dir %llu: name: \"%.*s\", lookup of hash: %u.0x%x "
950 "returns: %llu\n",
951 (unsigned long long)OCFS2_I(dir)->ip_blkno,
952 namelen, name, hinfo->major_hash, hinfo->minor_hash,
953 (unsigned long long)phys);
954
955 ret = ocfs2_read_dx_leaf(dir, phys, &dx_leaf_bh);
956 if (ret) {
957 mlog_errno(ret);
958 goto out;
959 }
960
961 dx_leaf = (struct ocfs2_dx_leaf *) dx_leaf_bh->b_data;
962
963 mlog(0, "leaf info: num_used: %d, count: %d\n",
964 le16_to_cpu(dx_leaf->dl_list.de_num_used),
965 le16_to_cpu(dx_leaf->dl_list.de_count));
966
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -0800967 entry_list = &dx_leaf->dl_list;
968
969search:
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800970 /*
971 * Empty leaf is legal, so no need to check for that.
972 */
973 found = 0;
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -0800974 for (i = 0; i < le16_to_cpu(entry_list->de_num_used); i++) {
975 dx_entry = &entry_list->de_entries[i];
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800976
977 if (hinfo->major_hash != le32_to_cpu(dx_entry->dx_major_hash)
978 || hinfo->minor_hash != le32_to_cpu(dx_entry->dx_minor_hash))
979 continue;
980
981 /*
982 * Search unindexed leaf block now. We're not
983 * guaranteed to find anything.
984 */
985 ret = ocfs2_read_dir_block_direct(dir,
986 le64_to_cpu(dx_entry->dx_dirent_blk),
987 &dir_ent_bh);
988 if (ret) {
989 mlog_errno(ret);
990 goto out;
991 }
992
993 /*
994 * XXX: We should check the unindexed block here,
995 * before using it.
996 */
997
998 found = ocfs2_search_dirblock(dir_ent_bh, dir, name, namelen,
999 0, dir_ent_bh->b_data,
1000 dir->i_sb->s_blocksize, &dir_ent);
1001 if (found == 1)
1002 break;
1003
1004 if (found == -1) {
1005 /* This means we found a bad directory entry. */
1006 ret = -EIO;
1007 mlog_errno(ret);
1008 goto out;
1009 }
1010
1011 brelse(dir_ent_bh);
1012 dir_ent_bh = NULL;
1013 }
1014
1015 if (found <= 0) {
1016 ret = -ENOENT;
1017 goto out;
1018 }
1019
1020 res->dl_leaf_bh = dir_ent_bh;
1021 res->dl_entry = dir_ent;
1022 res->dl_dx_leaf_bh = dx_leaf_bh;
1023 res->dl_dx_entry = dx_entry;
1024
1025 ret = 0;
1026out:
1027 if (ret) {
1028 brelse(dx_leaf_bh);
1029 brelse(dir_ent_bh);
1030 }
1031 return ret;
1032}
1033
1034static int ocfs2_find_entry_dx(const char *name, int namelen,
1035 struct inode *dir,
1036 struct ocfs2_dir_lookup_result *lookup)
1037{
1038 int ret;
1039 struct buffer_head *di_bh = NULL;
1040 struct ocfs2_dinode *di;
1041 struct buffer_head *dx_root_bh = NULL;
1042 struct ocfs2_dx_root_block *dx_root;
1043
1044 ret = ocfs2_read_inode_block(dir, &di_bh);
1045 if (ret) {
1046 mlog_errno(ret);
1047 goto out;
1048 }
1049
1050 di = (struct ocfs2_dinode *)di_bh->b_data;
1051
1052 ret = ocfs2_read_dx_root(dir, di, &dx_root_bh);
1053 if (ret) {
1054 mlog_errno(ret);
1055 goto out;
1056 }
1057 dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data;
1058
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001059 ret = ocfs2_dx_dir_search(name, namelen, dir, dx_root, lookup);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001060 if (ret) {
1061 if (ret != -ENOENT)
1062 mlog_errno(ret);
1063 goto out;
1064 }
1065
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001066 lookup->dl_dx_root_bh = dx_root_bh;
1067 dx_root_bh = NULL;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001068out:
1069 brelse(di_bh);
1070 brelse(dx_root_bh);
1071 return ret;
1072}
1073
Mark Fasheh23193e52007-09-12 13:01:18 -07001074/*
1075 * Try to find an entry of the provided name within 'dir'.
1076 *
Mark Fasheh4a12ca32008-11-12 15:43:34 -08001077 * If nothing was found, -ENOENT is returned. Otherwise, zero is
1078 * returned and the struct 'res' will contain information useful to
1079 * other directory manipulation functions.
Mark Fasheh23193e52007-09-12 13:01:18 -07001080 *
1081 * Caller can NOT assume anything about the contents of the
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001082 * buffer_heads - they are passed back only so that it can be passed
1083 * into any one of the manipulation functions (add entry, delete
1084 * entry, etc). As an example, bh in the extent directory case is a
1085 * data block, in the inline-data case it actually points to an inode,
1086 * in the indexed directory case, multiple buffers are involved.
Mark Fasheh23193e52007-09-12 13:01:18 -07001087 */
Mark Fasheh4a12ca32008-11-12 15:43:34 -08001088int ocfs2_find_entry(const char *name, int namelen,
1089 struct inode *dir, struct ocfs2_dir_lookup_result *lookup)
Mark Fasheh23193e52007-09-12 13:01:18 -07001090{
Mark Fasheh4a12ca32008-11-12 15:43:34 -08001091 struct buffer_head *bh;
1092 struct ocfs2_dir_entry *res_dir = NULL;
Mark Fasheh23193e52007-09-12 13:01:18 -07001093
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001094 if (ocfs2_dir_indexed(dir))
1095 return ocfs2_find_entry_dx(name, namelen, dir, lookup);
1096
1097 /*
1098 * The unindexed dir code only uses part of the lookup
1099 * structure, so there's no reason to push it down further
1100 * than this.
1101 */
Mark Fasheh23193e52007-09-12 13:01:18 -07001102 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
Mark Fasheh4a12ca32008-11-12 15:43:34 -08001103 bh = ocfs2_find_entry_id(name, namelen, dir, &res_dir);
1104 else
1105 bh = ocfs2_find_entry_el(name, namelen, dir, &res_dir);
Mark Fasheh23193e52007-09-12 13:01:18 -07001106
Mark Fasheh4a12ca32008-11-12 15:43:34 -08001107 if (bh == NULL)
1108 return -ENOENT;
1109
1110 lookup->dl_leaf_bh = bh;
1111 lookup->dl_entry = res_dir;
1112 return 0;
Mark Fasheh23193e52007-09-12 13:01:18 -07001113}
1114
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001115/*
1116 * Update inode number and type of a previously found directory entry.
1117 */
Mark Fasheh38760e22007-09-11 17:21:56 -07001118int ocfs2_update_entry(struct inode *dir, handle_t *handle,
Mark Fasheh4a12ca32008-11-12 15:43:34 -08001119 struct ocfs2_dir_lookup_result *res,
Mark Fasheh38760e22007-09-11 17:21:56 -07001120 struct inode *new_entry_inode)
1121{
1122 int ret;
Joel Becker13723d02008-10-17 19:25:01 -07001123 ocfs2_journal_access_func access = ocfs2_journal_access_db;
Mark Fasheh4a12ca32008-11-12 15:43:34 -08001124 struct ocfs2_dir_entry *de = res->dl_entry;
1125 struct buffer_head *de_bh = res->dl_leaf_bh;
Mark Fasheh38760e22007-09-11 17:21:56 -07001126
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001127 /*
1128 * The same code works fine for both inline-data and extent
Joel Becker13723d02008-10-17 19:25:01 -07001129 * based directories, so no need to split this up. The only
1130 * difference is the journal_access function.
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001131 */
1132
Joel Becker13723d02008-10-17 19:25:01 -07001133 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
1134 access = ocfs2_journal_access_di;
1135
1136 ret = access(handle, dir, de_bh, OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh38760e22007-09-11 17:21:56 -07001137 if (ret) {
1138 mlog_errno(ret);
1139 goto out;
1140 }
1141
1142 de->inode = cpu_to_le64(OCFS2_I(new_entry_inode)->ip_blkno);
1143 ocfs2_set_de_type(de, new_entry_inode->i_mode);
1144
1145 ocfs2_journal_dirty(handle, de_bh);
1146
1147out:
1148 return ret;
1149}
1150
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001151/*
1152 * __ocfs2_delete_entry deletes a directory entry by merging it with the
1153 * previous entry
1154 */
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001155static int __ocfs2_delete_entry(handle_t *handle, struct inode *dir,
1156 struct ocfs2_dir_entry *de_del,
1157 struct buffer_head *bh, char *first_de,
1158 unsigned int bytes)
Mark Fasheh316f4b92007-09-07 18:21:26 -07001159{
1160 struct ocfs2_dir_entry *de, *pde;
1161 int i, status = -ENOENT;
Joel Becker13723d02008-10-17 19:25:01 -07001162 ocfs2_journal_access_func access = ocfs2_journal_access_db;
Mark Fasheh316f4b92007-09-07 18:21:26 -07001163
1164 mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p)\n", handle, dir, de_del, bh);
1165
Joel Becker13723d02008-10-17 19:25:01 -07001166 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
1167 access = ocfs2_journal_access_di;
1168
Mark Fasheh316f4b92007-09-07 18:21:26 -07001169 i = 0;
1170 pde = NULL;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001171 de = (struct ocfs2_dir_entry *) first_de;
1172 while (i < bytes) {
Mark Fasheh316f4b92007-09-07 18:21:26 -07001173 if (!ocfs2_check_dir_entry(dir, de, bh, i)) {
1174 status = -EIO;
1175 mlog_errno(status);
1176 goto bail;
1177 }
1178 if (de == de_del) {
Joel Becker13723d02008-10-17 19:25:01 -07001179 status = access(handle, dir, bh,
1180 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh316f4b92007-09-07 18:21:26 -07001181 if (status < 0) {
1182 status = -EIO;
1183 mlog_errno(status);
1184 goto bail;
1185 }
1186 if (pde)
Marcin Slusarz0dd32562008-02-13 00:06:18 +01001187 le16_add_cpu(&pde->rec_len,
1188 le16_to_cpu(de->rec_len));
Mark Fasheh316f4b92007-09-07 18:21:26 -07001189 else
1190 de->inode = 0;
1191 dir->i_version++;
1192 status = ocfs2_journal_dirty(handle, bh);
1193 goto bail;
1194 }
1195 i += le16_to_cpu(de->rec_len);
1196 pde = de;
1197 de = (struct ocfs2_dir_entry *)((char *)de + le16_to_cpu(de->rec_len));
1198 }
1199bail:
1200 mlog_exit(status);
1201 return status;
1202}
1203
Mark Fashehe7c17e42009-01-29 18:17:46 -08001204static unsigned int ocfs2_figure_dirent_hole(struct ocfs2_dir_entry *de)
1205{
1206 unsigned int hole;
1207
1208 if (le64_to_cpu(de->inode) == 0)
1209 hole = le16_to_cpu(de->rec_len);
1210 else
1211 hole = le16_to_cpu(de->rec_len) -
1212 OCFS2_DIR_REC_LEN(de->name_len);
1213
1214 return hole;
1215}
1216
1217static int ocfs2_find_max_rec_len(struct super_block *sb,
1218 struct buffer_head *dirblock_bh)
1219{
1220 int size, this_hole, largest_hole = 0;
1221 char *trailer, *de_buf, *limit, *start = dirblock_bh->b_data;
1222 struct ocfs2_dir_entry *de;
1223
1224 trailer = (char *)ocfs2_trailer_from_bh(dirblock_bh, sb);
1225 size = ocfs2_dir_trailer_blk_off(sb);
1226 limit = start + size;
1227 de_buf = start;
1228 de = (struct ocfs2_dir_entry *)de_buf;
1229 do {
1230 if (de_buf != trailer) {
1231 this_hole = ocfs2_figure_dirent_hole(de);
1232 if (this_hole > largest_hole)
1233 largest_hole = this_hole;
1234 }
1235
1236 de_buf += le16_to_cpu(de->rec_len);
1237 de = (struct ocfs2_dir_entry *)de_buf;
1238 } while (de_buf < limit);
1239
1240 if (largest_hole >= OCFS2_DIR_MIN_REC_LEN)
1241 return largest_hole;
1242 return 0;
1243}
1244
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001245static void ocfs2_dx_list_remove_entry(struct ocfs2_dx_entry_list *entry_list,
1246 int index)
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001247{
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001248 int num_used = le16_to_cpu(entry_list->de_num_used);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001249
1250 if (num_used == 1 || index == (num_used - 1))
1251 goto clear;
1252
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001253 memmove(&entry_list->de_entries[index],
1254 &entry_list->de_entries[index + 1],
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001255 (num_used - index - 1)*sizeof(struct ocfs2_dx_entry));
1256clear:
1257 num_used--;
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001258 memset(&entry_list->de_entries[num_used], 0,
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001259 sizeof(struct ocfs2_dx_entry));
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001260 entry_list->de_num_used = cpu_to_le16(num_used);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001261}
1262
1263static int ocfs2_delete_entry_dx(handle_t *handle, struct inode *dir,
1264 struct ocfs2_dir_lookup_result *lookup)
1265{
Mark Fashehe7c17e42009-01-29 18:17:46 -08001266 int ret, index, max_rec_len, add_to_free_list = 0;
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001267 struct buffer_head *dx_root_bh = lookup->dl_dx_root_bh;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001268 struct buffer_head *leaf_bh = lookup->dl_leaf_bh;
1269 struct ocfs2_dx_leaf *dx_leaf;
1270 struct ocfs2_dx_entry *dx_entry = lookup->dl_dx_entry;
Mark Fashehe7c17e42009-01-29 18:17:46 -08001271 struct ocfs2_dir_block_trailer *trailer;
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001272 struct ocfs2_dx_root_block *dx_root;
1273 struct ocfs2_dx_entry_list *entry_list;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001274
Mark Fashehe7c17e42009-01-29 18:17:46 -08001275 /*
1276 * This function gets a bit messy because we might have to
1277 * modify the root block, regardless of whether the indexed
1278 * entries are stored inline.
1279 */
1280
1281 /*
1282 * *Only* set 'entry_list' here, based on where we're looking
1283 * for the indexed entries. Later, we might still want to
1284 * journal both blocks, based on free list state.
1285 */
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001286 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
1287 if (ocfs2_dx_root_inline(dx_root)) {
1288 entry_list = &dx_root->dr_entries;
1289 } else {
1290 dx_leaf = (struct ocfs2_dx_leaf *) lookup->dl_dx_leaf_bh->b_data;
1291 entry_list = &dx_leaf->dl_list;
1292 }
1293
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001294 /* Neither of these are a disk corruption - that should have
1295 * been caught by lookup, before we got here. */
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001296 BUG_ON(le16_to_cpu(entry_list->de_count) <= 0);
1297 BUG_ON(le16_to_cpu(entry_list->de_num_used) <= 0);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001298
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001299 index = (char *)dx_entry - (char *)entry_list->de_entries;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001300 index /= sizeof(*dx_entry);
1301
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001302 if (index >= le16_to_cpu(entry_list->de_num_used)) {
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001303 mlog(ML_ERROR, "Dir %llu: Bad dx_entry ptr idx %d, (%p, %p)\n",
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001304 (unsigned long long)OCFS2_I(dir)->ip_blkno, index,
1305 entry_list, dx_entry);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001306 return -EIO;
1307 }
1308
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001309 /*
Mark Fashehe7c17e42009-01-29 18:17:46 -08001310 * We know that removal of this dirent will leave enough room
1311 * for a new one, so add this block to the free list if it
1312 * isn't already there.
1313 */
1314 trailer = ocfs2_trailer_from_bh(leaf_bh, dir->i_sb);
1315 if (trailer->db_free_rec_len == 0)
1316 add_to_free_list = 1;
1317
1318 /*
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001319 * Add the block holding our index into the journal before
1320 * removing the unindexed entry. If we get an error return
1321 * from __ocfs2_delete_entry(), then it hasn't removed the
1322 * entry yet. Likewise, successful return means we *must*
1323 * remove the indexed entry.
1324 *
Mark Fashehe3a93c22009-02-17 15:29:35 -08001325 * We're also careful to journal the root tree block here as
1326 * the entry count needs to be updated. Also, we might be
1327 * adding to the start of the free list.
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001328 */
Mark Fashehe3a93c22009-02-17 15:29:35 -08001329 ret = ocfs2_journal_access_dr(handle, dir, dx_root_bh,
1330 OCFS2_JOURNAL_ACCESS_WRITE);
1331 if (ret) {
1332 mlog_errno(ret);
1333 goto out;
Mark Fashehe7c17e42009-01-29 18:17:46 -08001334 }
1335
1336 if (!ocfs2_dx_root_inline(dx_root)) {
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001337 ret = ocfs2_journal_access_dl(handle, dir,
1338 lookup->dl_dx_leaf_bh,
1339 OCFS2_JOURNAL_ACCESS_WRITE);
1340 if (ret) {
1341 mlog_errno(ret);
1342 goto out;
1343 }
1344 }
1345
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001346 mlog(0, "Dir %llu: delete entry at index: %d\n",
1347 (unsigned long long)OCFS2_I(dir)->ip_blkno, index);
1348
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001349 ret = __ocfs2_delete_entry(handle, dir, lookup->dl_entry,
1350 leaf_bh, leaf_bh->b_data, leaf_bh->b_size);
1351 if (ret) {
1352 mlog_errno(ret);
1353 goto out;
1354 }
1355
Mark Fashehe7c17e42009-01-29 18:17:46 -08001356 max_rec_len = ocfs2_find_max_rec_len(dir->i_sb, leaf_bh);
1357 trailer->db_free_rec_len = cpu_to_le16(max_rec_len);
1358 if (add_to_free_list) {
1359 trailer->db_free_next = dx_root->dr_free_blk;
1360 dx_root->dr_free_blk = cpu_to_le64(leaf_bh->b_blocknr);
1361 ocfs2_journal_dirty(handle, dx_root_bh);
1362 }
1363
1364 /* leaf_bh was journal_accessed for us in __ocfs2_delete_entry */
1365 ocfs2_journal_dirty(handle, leaf_bh);
1366
Mark Fashehe3a93c22009-02-17 15:29:35 -08001367 le32_add_cpu(&dx_root->dr_num_entries, -1);
1368 ocfs2_journal_dirty(handle, dx_root_bh);
1369
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001370 ocfs2_dx_list_remove_entry(entry_list, index);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001371
Mark Fashehe3a93c22009-02-17 15:29:35 -08001372 if (!ocfs2_dx_root_inline(dx_root))
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001373 ocfs2_journal_dirty(handle, lookup->dl_dx_leaf_bh);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001374
1375out:
1376 return ret;
1377}
1378
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001379static inline int ocfs2_delete_entry_id(handle_t *handle,
1380 struct inode *dir,
1381 struct ocfs2_dir_entry *de_del,
1382 struct buffer_head *bh)
1383{
1384 int ret;
1385 struct buffer_head *di_bh = NULL;
1386 struct ocfs2_dinode *di;
1387 struct ocfs2_inline_data *data;
1388
Joel Beckerb657c952008-11-13 14:49:11 -08001389 ret = ocfs2_read_inode_block(dir, &di_bh);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001390 if (ret) {
1391 mlog_errno(ret);
1392 goto out;
1393 }
1394
1395 di = (struct ocfs2_dinode *)di_bh->b_data;
1396 data = &di->id2.i_data;
1397
1398 ret = __ocfs2_delete_entry(handle, dir, de_del, bh, data->id_data,
1399 i_size_read(dir));
1400
1401 brelse(di_bh);
1402out:
1403 return ret;
1404}
1405
1406static inline int ocfs2_delete_entry_el(handle_t *handle,
1407 struct inode *dir,
1408 struct ocfs2_dir_entry *de_del,
1409 struct buffer_head *bh)
1410{
1411 return __ocfs2_delete_entry(handle, dir, de_del, bh, bh->b_data,
1412 bh->b_size);
1413}
1414
1415/*
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001416 * Delete a directory entry. Hide the details of directory
1417 * implementation from the caller.
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001418 */
1419int ocfs2_delete_entry(handle_t *handle,
1420 struct inode *dir,
Mark Fasheh4a12ca32008-11-12 15:43:34 -08001421 struct ocfs2_dir_lookup_result *res)
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001422{
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001423 if (ocfs2_dir_indexed(dir))
1424 return ocfs2_delete_entry_dx(handle, dir, res);
1425
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001426 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
Mark Fasheh4a12ca32008-11-12 15:43:34 -08001427 return ocfs2_delete_entry_id(handle, dir, res->dl_entry,
1428 res->dl_leaf_bh);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001429
Mark Fasheh4a12ca32008-11-12 15:43:34 -08001430 return ocfs2_delete_entry_el(handle, dir, res->dl_entry,
1431 res->dl_leaf_bh);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001432}
1433
Mark Fasheh8553cf42007-09-13 16:29:01 -07001434/*
1435 * Check whether 'de' has enough room to hold an entry of
1436 * 'new_rec_len' bytes.
1437 */
1438static inline int ocfs2_dirent_would_fit(struct ocfs2_dir_entry *de,
1439 unsigned int new_rec_len)
1440{
1441 unsigned int de_really_used;
1442
1443 /* Check whether this is an empty record with enough space */
1444 if (le64_to_cpu(de->inode) == 0 &&
1445 le16_to_cpu(de->rec_len) >= new_rec_len)
1446 return 1;
1447
1448 /*
1449 * Record might have free space at the end which we can
1450 * use.
1451 */
1452 de_really_used = OCFS2_DIR_REC_LEN(de->name_len);
1453 if (le16_to_cpu(de->rec_len) >= (de_really_used + new_rec_len))
1454 return 1;
1455
1456 return 0;
1457}
1458
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001459static void ocfs2_dx_dir_leaf_insert_tail(struct ocfs2_dx_leaf *dx_leaf,
1460 struct ocfs2_dx_entry *dx_new_entry)
1461{
1462 int i;
1463
1464 i = le16_to_cpu(dx_leaf->dl_list.de_num_used);
1465 dx_leaf->dl_list.de_entries[i] = *dx_new_entry;
1466
1467 le16_add_cpu(&dx_leaf->dl_list.de_num_used, 1);
1468}
1469
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001470static void ocfs2_dx_entry_list_insert(struct ocfs2_dx_entry_list *entry_list,
1471 struct ocfs2_dx_hinfo *hinfo,
1472 u64 dirent_blk)
1473{
1474 int i;
1475 struct ocfs2_dx_entry *dx_entry;
1476
1477 i = le16_to_cpu(entry_list->de_num_used);
1478 dx_entry = &entry_list->de_entries[i];
1479
1480 memset(dx_entry, 0, sizeof(*dx_entry));
1481 dx_entry->dx_major_hash = cpu_to_le32(hinfo->major_hash);
1482 dx_entry->dx_minor_hash = cpu_to_le32(hinfo->minor_hash);
1483 dx_entry->dx_dirent_blk = cpu_to_le64(dirent_blk);
1484
1485 le16_add_cpu(&entry_list->de_num_used, 1);
1486}
1487
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001488static int __ocfs2_dx_dir_leaf_insert(struct inode *dir, handle_t *handle,
1489 struct ocfs2_dx_hinfo *hinfo,
1490 u64 dirent_blk,
1491 struct buffer_head *dx_leaf_bh)
1492{
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001493 int ret;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001494 struct ocfs2_dx_leaf *dx_leaf;
1495
1496 ret = ocfs2_journal_access_dl(handle, dir, dx_leaf_bh,
1497 OCFS2_JOURNAL_ACCESS_WRITE);
1498 if (ret) {
1499 mlog_errno(ret);
1500 goto out;
1501 }
1502
1503 dx_leaf = (struct ocfs2_dx_leaf *)dx_leaf_bh->b_data;
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001504 ocfs2_dx_entry_list_insert(&dx_leaf->dl_list, hinfo, dirent_blk);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001505 ocfs2_journal_dirty(handle, dx_leaf_bh);
1506
1507out:
1508 return ret;
1509}
1510
Mark Fashehe3a93c22009-02-17 15:29:35 -08001511static void ocfs2_dx_inline_root_insert(struct inode *dir, handle_t *handle,
1512 struct ocfs2_dx_hinfo *hinfo,
1513 u64 dirent_blk,
1514 struct ocfs2_dx_root_block *dx_root)
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001515{
Mark Fashehe3a93c22009-02-17 15:29:35 -08001516 ocfs2_dx_entry_list_insert(&dx_root->dr_entries, hinfo, dirent_blk);
1517}
1518
1519static int ocfs2_dx_dir_insert(struct inode *dir, handle_t *handle,
1520 struct ocfs2_dir_lookup_result *lookup)
1521{
1522 int ret = 0;
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001523 struct ocfs2_dx_root_block *dx_root;
Mark Fashehe3a93c22009-02-17 15:29:35 -08001524 struct buffer_head *dx_root_bh = lookup->dl_dx_root_bh;
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001525
1526 ret = ocfs2_journal_access_dr(handle, dir, dx_root_bh,
1527 OCFS2_JOURNAL_ACCESS_WRITE);
1528 if (ret) {
1529 mlog_errno(ret);
1530 goto out;
1531 }
1532
Mark Fashehe3a93c22009-02-17 15:29:35 -08001533 dx_root = (struct ocfs2_dx_root_block *)lookup->dl_dx_root_bh->b_data;
1534 if (ocfs2_dx_root_inline(dx_root)) {
1535 ocfs2_dx_inline_root_insert(dir, handle,
1536 &lookup->dl_hinfo,
1537 lookup->dl_leaf_bh->b_blocknr,
1538 dx_root);
1539 } else {
1540 ret = __ocfs2_dx_dir_leaf_insert(dir, handle, &lookup->dl_hinfo,
1541 lookup->dl_leaf_bh->b_blocknr,
1542 lookup->dl_dx_leaf_bh);
1543 if (ret)
1544 goto out;
1545 }
1546
1547 le32_add_cpu(&dx_root->dr_num_entries, 1);
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001548 ocfs2_journal_dirty(handle, dx_root_bh);
1549
1550out:
1551 return ret;
1552}
1553
Mark Fashehe7c17e42009-01-29 18:17:46 -08001554static void ocfs2_remove_block_from_free_list(struct inode *dir,
1555 handle_t *handle,
1556 struct ocfs2_dir_lookup_result *lookup)
1557{
1558 struct ocfs2_dir_block_trailer *trailer, *prev;
1559 struct ocfs2_dx_root_block *dx_root;
1560 struct buffer_head *bh;
1561
1562 trailer = ocfs2_trailer_from_bh(lookup->dl_leaf_bh, dir->i_sb);
1563
1564 if (ocfs2_free_list_at_root(lookup)) {
1565 bh = lookup->dl_dx_root_bh;
1566 dx_root = (struct ocfs2_dx_root_block *)bh->b_data;
1567 dx_root->dr_free_blk = trailer->db_free_next;
1568 } else {
1569 bh = lookup->dl_prev_leaf_bh;
1570 prev = ocfs2_trailer_from_bh(bh, dir->i_sb);
1571 prev->db_free_next = trailer->db_free_next;
1572 }
1573
1574 trailer->db_free_rec_len = cpu_to_le16(0);
1575 trailer->db_free_next = cpu_to_le64(0);
1576
1577 ocfs2_journal_dirty(handle, bh);
1578 ocfs2_journal_dirty(handle, lookup->dl_leaf_bh);
1579}
1580
1581/*
1582 * This expects that a journal write has been reserved on
1583 * lookup->dl_prev_leaf_bh or lookup->dl_dx_root_bh
1584 */
1585static void ocfs2_recalc_free_list(struct inode *dir, handle_t *handle,
1586 struct ocfs2_dir_lookup_result *lookup)
1587{
1588 int max_rec_len;
1589 struct ocfs2_dir_block_trailer *trailer;
1590
1591 /* Walk dl_leaf_bh to figure out what the new free rec_len is. */
1592 max_rec_len = ocfs2_find_max_rec_len(dir->i_sb, lookup->dl_leaf_bh);
1593 if (max_rec_len) {
1594 /*
1595 * There's still room in this block, so no need to remove it
1596 * from the free list. In this case, we just want to update
1597 * the rec len accounting.
1598 */
1599 trailer = ocfs2_trailer_from_bh(lookup->dl_leaf_bh, dir->i_sb);
1600 trailer->db_free_rec_len = cpu_to_le16(max_rec_len);
1601 ocfs2_journal_dirty(handle, lookup->dl_leaf_bh);
1602 } else {
1603 ocfs2_remove_block_from_free_list(dir, handle, lookup);
1604 }
1605}
1606
Mark Fasheh316f4b92007-09-07 18:21:26 -07001607/* we don't always have a dentry for what we want to add, so people
1608 * like orphan dir can call this instead.
1609 *
Mark Fasheh4a12ca32008-11-12 15:43:34 -08001610 * The lookup context must have been filled from
1611 * ocfs2_prepare_dir_for_insert.
Mark Fasheh316f4b92007-09-07 18:21:26 -07001612 */
1613int __ocfs2_add_entry(handle_t *handle,
1614 struct inode *dir,
1615 const char *name, int namelen,
1616 struct inode *inode, u64 blkno,
1617 struct buffer_head *parent_fe_bh,
Mark Fasheh4a12ca32008-11-12 15:43:34 -08001618 struct ocfs2_dir_lookup_result *lookup)
Mark Fasheh316f4b92007-09-07 18:21:26 -07001619{
1620 unsigned long offset;
1621 unsigned short rec_len;
1622 struct ocfs2_dir_entry *de, *de1;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001623 struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_fe_bh->b_data;
1624 struct super_block *sb = dir->i_sb;
Mark Fasheh316f4b92007-09-07 18:21:26 -07001625 int retval, status;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001626 unsigned int size = sb->s_blocksize;
Mark Fasheh4a12ca32008-11-12 15:43:34 -08001627 struct buffer_head *insert_bh = lookup->dl_leaf_bh;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001628 char *data_start = insert_bh->b_data;
Mark Fasheh316f4b92007-09-07 18:21:26 -07001629
1630 mlog_entry_void();
1631
Mark Fasheh316f4b92007-09-07 18:21:26 -07001632 if (!namelen)
1633 return -EINVAL;
1634
Mark Fashehe7c17e42009-01-29 18:17:46 -08001635 if (ocfs2_dir_indexed(dir)) {
1636 struct buffer_head *bh;
1637
1638 /*
1639 * An indexed dir may require that we update the free space
1640 * list. Reserve a write to the previous node in the list so
1641 * that we don't fail later.
1642 *
1643 * XXX: This can be either a dx_root_block, or an unindexed
1644 * directory tree leaf block.
1645 */
1646 if (ocfs2_free_list_at_root(lookup)) {
1647 bh = lookup->dl_dx_root_bh;
1648 retval = ocfs2_journal_access_dr(handle, dir, bh,
1649 OCFS2_JOURNAL_ACCESS_WRITE);
1650 } else {
1651 bh = lookup->dl_prev_leaf_bh;
1652 retval = ocfs2_journal_access_db(handle, dir, bh,
1653 OCFS2_JOURNAL_ACCESS_WRITE);
1654 }
1655 if (retval) {
1656 mlog_errno(retval);
1657 return retval;
1658 }
1659 } else if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001660 data_start = di->id2.i_data.id_data;
1661 size = i_size_read(dir);
1662
1663 BUG_ON(insert_bh != parent_fe_bh);
1664 }
1665
Mark Fasheh316f4b92007-09-07 18:21:26 -07001666 rec_len = OCFS2_DIR_REC_LEN(namelen);
1667 offset = 0;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001668 de = (struct ocfs2_dir_entry *) data_start;
Mark Fasheh316f4b92007-09-07 18:21:26 -07001669 while (1) {
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001670 BUG_ON((char *)de >= (size + data_start));
1671
Mark Fasheh316f4b92007-09-07 18:21:26 -07001672 /* These checks should've already been passed by the
1673 * prepare function, but I guess we can leave them
1674 * here anyway. */
1675 if (!ocfs2_check_dir_entry(dir, de, insert_bh, offset)) {
1676 retval = -ENOENT;
1677 goto bail;
1678 }
1679 if (ocfs2_match(namelen, name, de)) {
1680 retval = -EEXIST;
1681 goto bail;
1682 }
Mark Fasheh8553cf42007-09-13 16:29:01 -07001683
Mark Fasheh87d35a72008-12-10 17:36:25 -08001684 /* We're guaranteed that we should have space, so we
1685 * can't possibly have hit the trailer...right? */
1686 mlog_bug_on_msg(ocfs2_skip_dir_trailer(dir, de, offset, size),
1687 "Hit dir trailer trying to insert %.*s "
1688 "(namelen %d) into directory %llu. "
1689 "offset is %lu, trailer offset is %d\n",
1690 namelen, name, namelen,
1691 (unsigned long long)parent_fe_bh->b_blocknr,
1692 offset, ocfs2_dir_trailer_blk_off(dir->i_sb));
1693
Mark Fasheh8553cf42007-09-13 16:29:01 -07001694 if (ocfs2_dirent_would_fit(de, rec_len)) {
Mark Fasheh316f4b92007-09-07 18:21:26 -07001695 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1696 retval = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
1697 if (retval < 0) {
1698 mlog_errno(retval);
1699 goto bail;
1700 }
1701
Joel Becker13723d02008-10-17 19:25:01 -07001702 if (insert_bh == parent_fe_bh)
1703 status = ocfs2_journal_access_di(handle, dir,
1704 insert_bh,
1705 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001706 else {
Joel Becker13723d02008-10-17 19:25:01 -07001707 status = ocfs2_journal_access_db(handle, dir,
1708 insert_bh,
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001709 OCFS2_JOURNAL_ACCESS_WRITE);
1710
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001711 if (ocfs2_dir_indexed(dir)) {
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001712 status = ocfs2_dx_dir_insert(dir,
1713 handle,
1714 lookup);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001715 if (status) {
1716 mlog_errno(status);
1717 goto bail;
1718 }
1719 }
1720 }
1721
Mark Fasheh316f4b92007-09-07 18:21:26 -07001722 /* By now the buffer is marked for journaling */
1723 offset += le16_to_cpu(de->rec_len);
1724 if (le64_to_cpu(de->inode)) {
1725 de1 = (struct ocfs2_dir_entry *)((char *) de +
1726 OCFS2_DIR_REC_LEN(de->name_len));
1727 de1->rec_len =
1728 cpu_to_le16(le16_to_cpu(de->rec_len) -
1729 OCFS2_DIR_REC_LEN(de->name_len));
1730 de->rec_len = cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
1731 de = de1;
1732 }
1733 de->file_type = OCFS2_FT_UNKNOWN;
1734 if (blkno) {
1735 de->inode = cpu_to_le64(blkno);
1736 ocfs2_set_de_type(de, inode->i_mode);
1737 } else
1738 de->inode = 0;
1739 de->name_len = namelen;
1740 memcpy(de->name, name, namelen);
1741
Mark Fashehe7c17e42009-01-29 18:17:46 -08001742 if (ocfs2_dir_indexed(dir))
1743 ocfs2_recalc_free_list(dir, handle, lookup);
1744
Mark Fasheh316f4b92007-09-07 18:21:26 -07001745 dir->i_version++;
1746 status = ocfs2_journal_dirty(handle, insert_bh);
1747 retval = 0;
1748 goto bail;
1749 }
Mark Fasheh87d35a72008-12-10 17:36:25 -08001750
Mark Fasheh316f4b92007-09-07 18:21:26 -07001751 offset += le16_to_cpu(de->rec_len);
1752 de = (struct ocfs2_dir_entry *) ((char *) de + le16_to_cpu(de->rec_len));
1753 }
1754
1755 /* when you think about it, the assert above should prevent us
1756 * from ever getting here. */
1757 retval = -ENOSPC;
1758bail:
1759
1760 mlog_exit(retval);
1761 return retval;
1762}
1763
Mark Fasheh23193e52007-09-12 13:01:18 -07001764static int ocfs2_dir_foreach_blk_id(struct inode *inode,
Mathieu Desnoyers2b47c362007-10-16 23:27:21 -07001765 u64 *f_version,
Mark Fasheh23193e52007-09-12 13:01:18 -07001766 loff_t *f_pos, void *priv,
Mark Fashehe7b34012007-09-24 14:25:27 -07001767 filldir_t filldir, int *filldir_err)
Mark Fasheh23193e52007-09-12 13:01:18 -07001768{
1769 int ret, i, filldir_ret;
1770 unsigned long offset = *f_pos;
1771 struct buffer_head *di_bh = NULL;
1772 struct ocfs2_dinode *di;
1773 struct ocfs2_inline_data *data;
1774 struct ocfs2_dir_entry *de;
1775
Joel Beckerb657c952008-11-13 14:49:11 -08001776 ret = ocfs2_read_inode_block(inode, &di_bh);
Mark Fasheh23193e52007-09-12 13:01:18 -07001777 if (ret) {
1778 mlog(ML_ERROR, "Unable to read inode block for dir %llu\n",
1779 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1780 goto out;
1781 }
1782
1783 di = (struct ocfs2_dinode *)di_bh->b_data;
1784 data = &di->id2.i_data;
1785
1786 while (*f_pos < i_size_read(inode)) {
1787revalidate:
1788 /* If the dir block has changed since the last call to
1789 * readdir(2), then we might be pointing to an invalid
1790 * dirent right now. Scan from the start of the block
1791 * to make sure. */
1792 if (*f_version != inode->i_version) {
1793 for (i = 0; i < i_size_read(inode) && i < offset; ) {
1794 de = (struct ocfs2_dir_entry *)
1795 (data->id_data + i);
1796 /* It's too expensive to do a full
1797 * dirent test each time round this
1798 * loop, but we do have to test at
1799 * least that it is non-zero. A
1800 * failure will be detected in the
1801 * dirent test below. */
1802 if (le16_to_cpu(de->rec_len) <
1803 OCFS2_DIR_REC_LEN(1))
1804 break;
1805 i += le16_to_cpu(de->rec_len);
1806 }
1807 *f_pos = offset = i;
1808 *f_version = inode->i_version;
1809 }
1810
1811 de = (struct ocfs2_dir_entry *) (data->id_data + *f_pos);
1812 if (!ocfs2_check_dir_entry(inode, de, di_bh, *f_pos)) {
1813 /* On error, skip the f_pos to the end. */
1814 *f_pos = i_size_read(inode);
1815 goto out;
1816 }
1817 offset += le16_to_cpu(de->rec_len);
1818 if (le64_to_cpu(de->inode)) {
1819 /* We might block in the next section
1820 * if the data destination is
1821 * currently swapped out. So, use a
1822 * version stamp to detect whether or
1823 * not the directory has been modified
1824 * during the copy operation.
1825 */
Mathieu Desnoyers2b47c362007-10-16 23:27:21 -07001826 u64 version = *f_version;
Mark Fasheh23193e52007-09-12 13:01:18 -07001827 unsigned char d_type = DT_UNKNOWN;
1828
1829 if (de->file_type < OCFS2_FT_MAX)
1830 d_type = ocfs2_filetype_table[de->file_type];
1831
1832 filldir_ret = filldir(priv, de->name,
1833 de->name_len,
1834 *f_pos,
1835 le64_to_cpu(de->inode),
1836 d_type);
Mark Fashehe7b34012007-09-24 14:25:27 -07001837 if (filldir_ret) {
1838 if (filldir_err)
1839 *filldir_err = filldir_ret;
Mark Fasheh23193e52007-09-12 13:01:18 -07001840 break;
Mark Fashehe7b34012007-09-24 14:25:27 -07001841 }
Mark Fasheh23193e52007-09-12 13:01:18 -07001842 if (version != *f_version)
1843 goto revalidate;
1844 }
1845 *f_pos += le16_to_cpu(de->rec_len);
1846 }
1847
1848out:
1849 brelse(di_bh);
1850
1851 return 0;
1852}
1853
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001854/*
1855 * NOTE: This function can be called against unindexed directories,
1856 * and indexed ones.
1857 */
Mark Fasheh23193e52007-09-12 13:01:18 -07001858static int ocfs2_dir_foreach_blk_el(struct inode *inode,
Mathieu Desnoyers2b47c362007-10-16 23:27:21 -07001859 u64 *f_version,
Mark Fasheh23193e52007-09-12 13:01:18 -07001860 loff_t *f_pos, void *priv,
Mark Fashehe7b34012007-09-24 14:25:27 -07001861 filldir_t filldir, int *filldir_err)
Mark Fashehccd979b2005-12-15 14:31:24 -08001862{
1863 int error = 0;
Mark Fashehaa958872006-04-21 13:49:02 -07001864 unsigned long offset, blk, last_ra_blk = 0;
1865 int i, stored;
Mark Fashehccd979b2005-12-15 14:31:24 -08001866 struct buffer_head * bh, * tmp;
1867 struct ocfs2_dir_entry * de;
Mark Fashehccd979b2005-12-15 14:31:24 -08001868 struct super_block * sb = inode->i_sb;
Mark Fashehaa958872006-04-21 13:49:02 -07001869 unsigned int ra_sectors = 16;
Mark Fashehccd979b2005-12-15 14:31:24 -08001870
1871 stored = 0;
1872 bh = NULL;
1873
Mark Fashehb8bc5f42007-09-10 17:17:52 -07001874 offset = (*f_pos) & (sb->s_blocksize - 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08001875
Mark Fashehb8bc5f42007-09-10 17:17:52 -07001876 while (!error && !stored && *f_pos < i_size_read(inode)) {
1877 blk = (*f_pos) >> sb->s_blocksize_bits;
Joel Beckera22305c2008-11-13 14:49:17 -08001878 if (ocfs2_read_dir_block(inode, blk, &bh, 0)) {
1879 /* Skip the corrupt dirblock and keep trying */
Mark Fashehb8bc5f42007-09-10 17:17:52 -07001880 *f_pos += sb->s_blocksize - offset;
Mark Fashehccd979b2005-12-15 14:31:24 -08001881 continue;
1882 }
1883
Mark Fashehaa958872006-04-21 13:49:02 -07001884 /* The idea here is to begin with 8k read-ahead and to stay
1885 * 4k ahead of our current position.
1886 *
1887 * TODO: Use the pagecache for this. We just need to
1888 * make sure it's cluster-safe... */
1889 if (!last_ra_blk
1890 || (((last_ra_blk - blk) << 9) <= (ra_sectors / 2))) {
1891 for (i = ra_sectors >> (sb->s_blocksize_bits - 9);
Mark Fashehccd979b2005-12-15 14:31:24 -08001892 i > 0; i--) {
Joel Beckera22305c2008-11-13 14:49:17 -08001893 tmp = NULL;
1894 if (!ocfs2_read_dir_block(inode, ++blk, &tmp,
1895 OCFS2_BH_READAHEAD))
1896 brelse(tmp);
Mark Fashehccd979b2005-12-15 14:31:24 -08001897 }
Mark Fashehaa958872006-04-21 13:49:02 -07001898 last_ra_blk = blk;
1899 ra_sectors = 8;
Mark Fashehccd979b2005-12-15 14:31:24 -08001900 }
1901
1902revalidate:
1903 /* If the dir block has changed since the last call to
1904 * readdir(2), then we might be pointing to an invalid
1905 * dirent right now. Scan from the start of the block
1906 * to make sure. */
Mark Fashehb8bc5f42007-09-10 17:17:52 -07001907 if (*f_version != inode->i_version) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001908 for (i = 0; i < sb->s_blocksize && i < offset; ) {
1909 de = (struct ocfs2_dir_entry *) (bh->b_data + i);
1910 /* It's too expensive to do a full
1911 * dirent test each time round this
1912 * loop, but we do have to test at
1913 * least that it is non-zero. A
1914 * failure will be detected in the
1915 * dirent test below. */
1916 if (le16_to_cpu(de->rec_len) <
1917 OCFS2_DIR_REC_LEN(1))
1918 break;
1919 i += le16_to_cpu(de->rec_len);
1920 }
1921 offset = i;
Mark Fashehb8bc5f42007-09-10 17:17:52 -07001922 *f_pos = ((*f_pos) & ~(sb->s_blocksize - 1))
Mark Fashehccd979b2005-12-15 14:31:24 -08001923 | offset;
Mark Fashehb8bc5f42007-09-10 17:17:52 -07001924 *f_version = inode->i_version;
Mark Fashehccd979b2005-12-15 14:31:24 -08001925 }
1926
Mark Fashehb8bc5f42007-09-10 17:17:52 -07001927 while (!error && *f_pos < i_size_read(inode)
Mark Fashehccd979b2005-12-15 14:31:24 -08001928 && offset < sb->s_blocksize) {
1929 de = (struct ocfs2_dir_entry *) (bh->b_data + offset);
1930 if (!ocfs2_check_dir_entry(inode, de, bh, offset)) {
1931 /* On error, skip the f_pos to the
1932 next block. */
Mark Fashehb8bc5f42007-09-10 17:17:52 -07001933 *f_pos = ((*f_pos) | (sb->s_blocksize - 1)) + 1;
Mark Fashehccd979b2005-12-15 14:31:24 -08001934 brelse(bh);
Mark Fashehb8bc5f42007-09-10 17:17:52 -07001935 goto out;
Mark Fashehccd979b2005-12-15 14:31:24 -08001936 }
1937 offset += le16_to_cpu(de->rec_len);
1938 if (le64_to_cpu(de->inode)) {
1939 /* We might block in the next section
1940 * if the data destination is
1941 * currently swapped out. So, use a
1942 * version stamp to detect whether or
1943 * not the directory has been modified
1944 * during the copy operation.
1945 */
Mark Fashehb8bc5f42007-09-10 17:17:52 -07001946 unsigned long version = *f_version;
Mark Fashehccd979b2005-12-15 14:31:24 -08001947 unsigned char d_type = DT_UNKNOWN;
1948
1949 if (de->file_type < OCFS2_FT_MAX)
1950 d_type = ocfs2_filetype_table[de->file_type];
Mark Fashehb8bc5f42007-09-10 17:17:52 -07001951 error = filldir(priv, de->name,
Mark Fashehccd979b2005-12-15 14:31:24 -08001952 de->name_len,
Mark Fashehb8bc5f42007-09-10 17:17:52 -07001953 *f_pos,
Mark Fasheh7e853672007-09-10 17:30:26 -07001954 le64_to_cpu(de->inode),
Mark Fashehccd979b2005-12-15 14:31:24 -08001955 d_type);
Mark Fashehe7b34012007-09-24 14:25:27 -07001956 if (error) {
1957 if (filldir_err)
1958 *filldir_err = error;
Mark Fashehccd979b2005-12-15 14:31:24 -08001959 break;
Mark Fashehe7b34012007-09-24 14:25:27 -07001960 }
Mark Fashehb8bc5f42007-09-10 17:17:52 -07001961 if (version != *f_version)
Mark Fashehccd979b2005-12-15 14:31:24 -08001962 goto revalidate;
1963 stored ++;
1964 }
Mark Fashehb8bc5f42007-09-10 17:17:52 -07001965 *f_pos += le16_to_cpu(de->rec_len);
Mark Fashehccd979b2005-12-15 14:31:24 -08001966 }
1967 offset = 0;
1968 brelse(bh);
Joel Beckera22305c2008-11-13 14:49:17 -08001969 bh = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08001970 }
1971
1972 stored = 0;
Mark Fashehb8bc5f42007-09-10 17:17:52 -07001973out:
1974 return stored;
1975}
1976
Mathieu Desnoyers2b47c362007-10-16 23:27:21 -07001977static int ocfs2_dir_foreach_blk(struct inode *inode, u64 *f_version,
Mark Fashehe7b34012007-09-24 14:25:27 -07001978 loff_t *f_pos, void *priv, filldir_t filldir,
1979 int *filldir_err)
Mark Fasheh23193e52007-09-12 13:01:18 -07001980{
1981 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
1982 return ocfs2_dir_foreach_blk_id(inode, f_version, f_pos, priv,
Mark Fashehe7b34012007-09-24 14:25:27 -07001983 filldir, filldir_err);
Mark Fasheh23193e52007-09-12 13:01:18 -07001984
Mark Fashehe7b34012007-09-24 14:25:27 -07001985 return ocfs2_dir_foreach_blk_el(inode, f_version, f_pos, priv, filldir,
1986 filldir_err);
Mark Fasheh23193e52007-09-12 13:01:18 -07001987}
1988
Mark Fashehb8bc5f42007-09-10 17:17:52 -07001989/*
Mark Fasheh5eae5b92007-09-10 17:50:51 -07001990 * This is intended to be called from inside other kernel functions,
1991 * so we fake some arguments.
1992 */
1993int ocfs2_dir_foreach(struct inode *inode, loff_t *f_pos, void *priv,
1994 filldir_t filldir)
1995{
Mark Fashehe7b34012007-09-24 14:25:27 -07001996 int ret = 0, filldir_err = 0;
Mathieu Desnoyers2b47c362007-10-16 23:27:21 -07001997 u64 version = inode->i_version;
Mark Fasheh5eae5b92007-09-10 17:50:51 -07001998
1999 while (*f_pos < i_size_read(inode)) {
2000 ret = ocfs2_dir_foreach_blk(inode, &version, f_pos, priv,
Mark Fashehe7b34012007-09-24 14:25:27 -07002001 filldir, &filldir_err);
2002 if (ret || filldir_err)
Mark Fasheh5eae5b92007-09-10 17:50:51 -07002003 break;
2004 }
2005
Mark Fashehe7b34012007-09-24 14:25:27 -07002006 if (ret > 0)
2007 ret = -EIO;
2008
Mark Fasheh5eae5b92007-09-10 17:50:51 -07002009 return 0;
2010}
2011
2012/*
Mark Fashehb8bc5f42007-09-10 17:17:52 -07002013 * ocfs2_readdir()
2014 *
2015 */
2016int ocfs2_readdir(struct file * filp, void * dirent, filldir_t filldir)
2017{
2018 int error = 0;
2019 struct inode *inode = filp->f_path.dentry->d_inode;
2020 int lock_level = 0;
2021
2022 mlog_entry("dirino=%llu\n",
2023 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2024
Mark Fashehe63aecb62007-10-18 15:30:42 -07002025 error = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level);
Mark Fashehb8bc5f42007-09-10 17:17:52 -07002026 if (lock_level && error >= 0) {
2027 /* We release EX lock which used to update atime
2028 * and get PR lock again to reduce contention
2029 * on commonly accessed directories. */
Mark Fashehe63aecb62007-10-18 15:30:42 -07002030 ocfs2_inode_unlock(inode, 1);
Mark Fashehb8bc5f42007-09-10 17:17:52 -07002031 lock_level = 0;
Mark Fashehe63aecb62007-10-18 15:30:42 -07002032 error = ocfs2_inode_lock(inode, NULL, 0);
Mark Fashehb8bc5f42007-09-10 17:17:52 -07002033 }
2034 if (error < 0) {
2035 if (error != -ENOENT)
2036 mlog_errno(error);
2037 /* we haven't got any yet, so propagate the error. */
2038 goto bail_nolock;
2039 }
2040
2041 error = ocfs2_dir_foreach_blk(inode, &filp->f_version, &filp->f_pos,
Mark Fashehe7b34012007-09-24 14:25:27 -07002042 dirent, filldir, NULL);
Mark Fashehb8bc5f42007-09-10 17:17:52 -07002043
Mark Fashehe63aecb62007-10-18 15:30:42 -07002044 ocfs2_inode_unlock(inode, lock_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08002045
Mark Fashehaa958872006-04-21 13:49:02 -07002046bail_nolock:
Mark Fashehb8bc5f42007-09-10 17:17:52 -07002047 mlog_exit(error);
Mark Fashehccd979b2005-12-15 14:31:24 -08002048
Mark Fashehb8bc5f42007-09-10 17:17:52 -07002049 return error;
Mark Fashehccd979b2005-12-15 14:31:24 -08002050}
2051
2052/*
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002053 * NOTE: this should always be called with parent dir i_mutex taken.
Mark Fashehccd979b2005-12-15 14:31:24 -08002054 */
2055int ocfs2_find_files_on_disk(const char *name,
2056 int namelen,
2057 u64 *blkno,
2058 struct inode *inode,
Mark Fasheh4a12ca32008-11-12 15:43:34 -08002059 struct ocfs2_dir_lookup_result *lookup)
Mark Fashehccd979b2005-12-15 14:31:24 -08002060{
2061 int status = -ENOENT;
Mark Fashehccd979b2005-12-15 14:31:24 -08002062
Mark Fasheh4a12ca32008-11-12 15:43:34 -08002063 mlog(0, "name=%.*s, blkno=%p, inode=%llu\n", namelen, name, blkno,
2064 (unsigned long long)OCFS2_I(inode)->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08002065
Mark Fasheh4a12ca32008-11-12 15:43:34 -08002066 status = ocfs2_find_entry(name, namelen, inode, lookup);
2067 if (status)
Mark Fashehccd979b2005-12-15 14:31:24 -08002068 goto leave;
Mark Fashehccd979b2005-12-15 14:31:24 -08002069
Mark Fasheh4a12ca32008-11-12 15:43:34 -08002070 *blkno = le64_to_cpu(lookup->dl_entry->inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08002071
2072 status = 0;
2073leave:
Mark Fashehccd979b2005-12-15 14:31:24 -08002074
Mark Fashehccd979b2005-12-15 14:31:24 -08002075 return status;
2076}
2077
Mark Fashehbe94d112007-09-11 15:22:06 -07002078/*
2079 * Convenience function for callers which just want the block number
2080 * mapped to a name and don't require the full dirent info, etc.
2081 */
2082int ocfs2_lookup_ino_from_name(struct inode *dir, const char *name,
2083 int namelen, u64 *blkno)
2084{
2085 int ret;
Mark Fasheh4a12ca32008-11-12 15:43:34 -08002086 struct ocfs2_dir_lookup_result lookup = { NULL, };
Mark Fashehbe94d112007-09-11 15:22:06 -07002087
Mark Fasheh4a12ca32008-11-12 15:43:34 -08002088 ret = ocfs2_find_files_on_disk(name, namelen, blkno, dir, &lookup);
2089 ocfs2_free_dir_lookup_result(&lookup);
Mark Fashehbe94d112007-09-11 15:22:06 -07002090
2091 return ret;
2092}
2093
Mark Fashehccd979b2005-12-15 14:31:24 -08002094/* Check for a name within a directory.
2095 *
2096 * Return 0 if the name does not exist
2097 * Return -EEXIST if the directory contains the name
2098 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002099 * Callers should have i_mutex + a cluster lock on dir
Mark Fashehccd979b2005-12-15 14:31:24 -08002100 */
2101int ocfs2_check_dir_for_entry(struct inode *dir,
2102 const char *name,
2103 int namelen)
2104{
2105 int ret;
Mark Fasheh4a12ca32008-11-12 15:43:34 -08002106 struct ocfs2_dir_lookup_result lookup = { NULL, };
Mark Fashehccd979b2005-12-15 14:31:24 -08002107
Mark Fashehb06970532006-03-03 10:24:33 -08002108 mlog_entry("dir %llu, name '%.*s'\n",
2109 (unsigned long long)OCFS2_I(dir)->ip_blkno, namelen, name);
Mark Fashehccd979b2005-12-15 14:31:24 -08002110
2111 ret = -EEXIST;
Mark Fasheh4a12ca32008-11-12 15:43:34 -08002112 if (ocfs2_find_entry(name, namelen, dir, &lookup) == 0)
Mark Fashehccd979b2005-12-15 14:31:24 -08002113 goto bail;
2114
2115 ret = 0;
2116bail:
Mark Fasheh4a12ca32008-11-12 15:43:34 -08002117 ocfs2_free_dir_lookup_result(&lookup);
Mark Fashehccd979b2005-12-15 14:31:24 -08002118
2119 mlog_exit(ret);
2120 return ret;
2121}
2122
Mark Fasheh0bfbbf62007-09-12 11:19:00 -07002123struct ocfs2_empty_dir_priv {
2124 unsigned seen_dot;
2125 unsigned seen_dot_dot;
2126 unsigned seen_other;
Mark Fashehe3a93c22009-02-17 15:29:35 -08002127 unsigned dx_dir;
Mark Fasheh0bfbbf62007-09-12 11:19:00 -07002128};
2129static int ocfs2_empty_dir_filldir(void *priv, const char *name, int name_len,
2130 loff_t pos, u64 ino, unsigned type)
2131{
2132 struct ocfs2_empty_dir_priv *p = priv;
2133
2134 /*
2135 * Check the positions of "." and ".." records to be sure
2136 * they're in the correct place.
Mark Fashehe3a93c22009-02-17 15:29:35 -08002137 *
2138 * Indexed directories don't need to proceed past the first
2139 * two entries, so we end the scan after seeing '..'. Despite
2140 * that, we allow the scan to proceed In the event that we
2141 * have a corrupted indexed directory (no dot or dot dot
2142 * entries). This allows us to double check for existing
2143 * entries which might not have been found in the index.
Mark Fasheh0bfbbf62007-09-12 11:19:00 -07002144 */
2145 if (name_len == 1 && !strncmp(".", name, 1) && pos == 0) {
2146 p->seen_dot = 1;
2147 return 0;
2148 }
2149
2150 if (name_len == 2 && !strncmp("..", name, 2) &&
2151 pos == OCFS2_DIR_REC_LEN(1)) {
2152 p->seen_dot_dot = 1;
Mark Fashehe3a93c22009-02-17 15:29:35 -08002153
2154 if (p->dx_dir && p->seen_dot)
2155 return 1;
2156
Mark Fasheh0bfbbf62007-09-12 11:19:00 -07002157 return 0;
2158 }
2159
2160 p->seen_other = 1;
2161 return 1;
2162}
Mark Fashehe3a93c22009-02-17 15:29:35 -08002163
2164static int ocfs2_empty_dir_dx(struct inode *inode,
2165 struct ocfs2_empty_dir_priv *priv)
2166{
2167 int ret;
2168 struct buffer_head *di_bh = NULL;
2169 struct buffer_head *dx_root_bh = NULL;
2170 struct ocfs2_dinode *di;
2171 struct ocfs2_dx_root_block *dx_root;
2172
2173 priv->dx_dir = 1;
2174
2175 ret = ocfs2_read_inode_block(inode, &di_bh);
2176 if (ret) {
2177 mlog_errno(ret);
2178 goto out;
2179 }
2180 di = (struct ocfs2_dinode *)di_bh->b_data;
2181
2182 ret = ocfs2_read_dx_root(inode, di, &dx_root_bh);
2183 if (ret) {
2184 mlog_errno(ret);
2185 goto out;
2186 }
2187 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
2188
2189 if (le32_to_cpu(dx_root->dr_num_entries) != 2)
2190 priv->seen_other = 1;
2191
2192out:
2193 brelse(di_bh);
2194 brelse(dx_root_bh);
2195 return ret;
2196}
2197
Mark Fashehccd979b2005-12-15 14:31:24 -08002198/*
2199 * routine to check that the specified directory is empty (for rmdir)
Mark Fasheh0bfbbf62007-09-12 11:19:00 -07002200 *
2201 * Returns 1 if dir is empty, zero otherwise.
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002202 *
Mark Fashehe3a93c22009-02-17 15:29:35 -08002203 * XXX: This is a performance problem for unindexed directories.
Mark Fashehccd979b2005-12-15 14:31:24 -08002204 */
2205int ocfs2_empty_dir(struct inode *inode)
2206{
Mark Fasheh0bfbbf62007-09-12 11:19:00 -07002207 int ret;
2208 loff_t start = 0;
2209 struct ocfs2_empty_dir_priv priv;
Mark Fashehccd979b2005-12-15 14:31:24 -08002210
Mark Fasheh0bfbbf62007-09-12 11:19:00 -07002211 memset(&priv, 0, sizeof(priv));
2212
Mark Fashehe3a93c22009-02-17 15:29:35 -08002213 if (ocfs2_dir_indexed(inode)) {
2214 ret = ocfs2_empty_dir_dx(inode, &priv);
2215 if (ret)
2216 mlog_errno(ret);
2217 /*
2218 * We still run ocfs2_dir_foreach to get the checks
2219 * for "." and "..".
2220 */
2221 }
2222
Mark Fasheh0bfbbf62007-09-12 11:19:00 -07002223 ret = ocfs2_dir_foreach(inode, &start, &priv, ocfs2_empty_dir_filldir);
2224 if (ret)
2225 mlog_errno(ret);
2226
2227 if (!priv.seen_dot || !priv.seen_dot_dot) {
2228 mlog(ML_ERROR, "bad directory (dir #%llu) - no `.' or `..'\n",
Mark Fashehb06970532006-03-03 10:24:33 -08002229 (unsigned long long)OCFS2_I(inode)->ip_blkno);
Mark Fasheh0bfbbf62007-09-12 11:19:00 -07002230 /*
2231 * XXX: Is it really safe to allow an unlink to continue?
2232 */
Mark Fashehccd979b2005-12-15 14:31:24 -08002233 return 1;
2234 }
2235
Mark Fasheh0bfbbf62007-09-12 11:19:00 -07002236 return !priv.seen_other;
Mark Fashehccd979b2005-12-15 14:31:24 -08002237}
2238
Mark Fasheh87d35a72008-12-10 17:36:25 -08002239/*
2240 * Fills "." and ".." dirents in a new directory block. Returns dirent for
2241 * "..", which might be used during creation of a directory with a trailing
2242 * header. It is otherwise safe to ignore the return code.
2243 */
2244static struct ocfs2_dir_entry *ocfs2_fill_initial_dirents(struct inode *inode,
2245 struct inode *parent,
2246 char *start,
2247 unsigned int size)
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002248{
2249 struct ocfs2_dir_entry *de = (struct ocfs2_dir_entry *)start;
2250
2251 de->inode = cpu_to_le64(OCFS2_I(inode)->ip_blkno);
2252 de->name_len = 1;
2253 de->rec_len =
2254 cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
2255 strcpy(de->name, ".");
2256 ocfs2_set_de_type(de, S_IFDIR);
2257
2258 de = (struct ocfs2_dir_entry *) ((char *)de + le16_to_cpu(de->rec_len));
2259 de->inode = cpu_to_le64(OCFS2_I(parent)->ip_blkno);
2260 de->rec_len = cpu_to_le16(size - OCFS2_DIR_REC_LEN(1));
2261 de->name_len = 2;
2262 strcpy(de->name, "..");
2263 ocfs2_set_de_type(de, S_IFDIR);
Mark Fasheh87d35a72008-12-10 17:36:25 -08002264
2265 return de;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002266}
2267
2268/*
2269 * This works together with code in ocfs2_mknod_locked() which sets
2270 * the inline-data flag and initializes the inline-data section.
2271 */
2272static int ocfs2_fill_new_dir_id(struct ocfs2_super *osb,
2273 handle_t *handle,
2274 struct inode *parent,
2275 struct inode *inode,
2276 struct buffer_head *di_bh)
2277{
2278 int ret;
2279 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2280 struct ocfs2_inline_data *data = &di->id2.i_data;
2281 unsigned int size = le16_to_cpu(data->id_count);
2282
Joel Becker13723d02008-10-17 19:25:01 -07002283 ret = ocfs2_journal_access_di(handle, inode, di_bh,
2284 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002285 if (ret) {
2286 mlog_errno(ret);
2287 goto out;
2288 }
2289
2290 ocfs2_fill_initial_dirents(inode, parent, data->id_data, size);
2291
2292 ocfs2_journal_dirty(handle, di_bh);
2293 if (ret) {
2294 mlog_errno(ret);
2295 goto out;
2296 }
2297
2298 i_size_write(inode, size);
2299 inode->i_nlink = 2;
2300 inode->i_blocks = ocfs2_inode_sector_count(inode);
2301
2302 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
2303 if (ret < 0)
2304 mlog_errno(ret);
2305
2306out:
2307 return ret;
2308}
2309
2310static int ocfs2_fill_new_dir_el(struct ocfs2_super *osb,
2311 handle_t *handle,
2312 struct inode *parent,
2313 struct inode *inode,
2314 struct buffer_head *fe_bh,
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002315 struct ocfs2_alloc_context *data_ac,
2316 struct buffer_head **ret_new_bh)
Mark Fasheh316f4b92007-09-07 18:21:26 -07002317{
2318 int status;
Mark Fasheh87d35a72008-12-10 17:36:25 -08002319 unsigned int size = osb->sb->s_blocksize;
Mark Fasheh316f4b92007-09-07 18:21:26 -07002320 struct buffer_head *new_bh = NULL;
Mark Fasheh87d35a72008-12-10 17:36:25 -08002321 struct ocfs2_dir_entry *de;
Mark Fasheh316f4b92007-09-07 18:21:26 -07002322
2323 mlog_entry_void();
2324
Mark Fashehe7c17e42009-01-29 18:17:46 -08002325 if (ocfs2_new_dir_wants_trailer(inode))
Mark Fasheh87d35a72008-12-10 17:36:25 -08002326 size = ocfs2_dir_trailer_blk_off(parent->i_sb);
2327
Mark Fasheh316f4b92007-09-07 18:21:26 -07002328 status = ocfs2_do_extend_dir(osb->sb, handle, inode, fe_bh,
2329 data_ac, NULL, &new_bh);
2330 if (status < 0) {
2331 mlog_errno(status);
2332 goto bail;
2333 }
2334
2335 ocfs2_set_new_buffer_uptodate(inode, new_bh);
2336
Joel Becker13723d02008-10-17 19:25:01 -07002337 status = ocfs2_journal_access_db(handle, inode, new_bh,
2338 OCFS2_JOURNAL_ACCESS_CREATE);
Mark Fasheh316f4b92007-09-07 18:21:26 -07002339 if (status < 0) {
2340 mlog_errno(status);
2341 goto bail;
2342 }
2343 memset(new_bh->b_data, 0, osb->sb->s_blocksize);
2344
Mark Fasheh87d35a72008-12-10 17:36:25 -08002345 de = ocfs2_fill_initial_dirents(inode, parent, new_bh->b_data, size);
Mark Fashehe7c17e42009-01-29 18:17:46 -08002346 if (ocfs2_new_dir_wants_trailer(inode)) {
2347 int size = le16_to_cpu(de->rec_len);
2348
2349 /*
2350 * Figure out the size of the hole left over after
2351 * insertion of '.' and '..'. The trailer wants this
2352 * information.
2353 */
2354 size -= OCFS2_DIR_REC_LEN(2);
2355 size -= sizeof(struct ocfs2_dir_block_trailer);
2356
2357 ocfs2_init_dir_trailer(inode, new_bh, size);
2358 }
Mark Fasheh316f4b92007-09-07 18:21:26 -07002359
2360 status = ocfs2_journal_dirty(handle, new_bh);
2361 if (status < 0) {
2362 mlog_errno(status);
2363 goto bail;
2364 }
2365
2366 i_size_write(inode, inode->i_sb->s_blocksize);
2367 inode->i_nlink = 2;
2368 inode->i_blocks = ocfs2_inode_sector_count(inode);
2369 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
2370 if (status < 0) {
2371 mlog_errno(status);
2372 goto bail;
2373 }
2374
2375 status = 0;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002376 if (ret_new_bh) {
2377 *ret_new_bh = new_bh;
2378 new_bh = NULL;
2379 }
Mark Fasheh316f4b92007-09-07 18:21:26 -07002380bail:
Mark Fasheha81cb882008-10-07 14:25:16 -07002381 brelse(new_bh);
Mark Fasheh316f4b92007-09-07 18:21:26 -07002382
2383 mlog_exit(status);
2384 return status;
2385}
2386
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002387static int ocfs2_dx_dir_attach_index(struct ocfs2_super *osb,
2388 handle_t *handle, struct inode *dir,
2389 struct buffer_head *di_bh,
Mark Fashehe7c17e42009-01-29 18:17:46 -08002390 struct buffer_head *dirdata_bh,
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002391 struct ocfs2_alloc_context *meta_ac,
Mark Fashehe3a93c22009-02-17 15:29:35 -08002392 int dx_inline, u32 num_entries,
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002393 struct buffer_head **ret_dx_root_bh)
2394{
2395 int ret;
2396 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
2397 u16 dr_suballoc_bit;
2398 u64 dr_blkno;
2399 unsigned int num_bits;
2400 struct buffer_head *dx_root_bh = NULL;
2401 struct ocfs2_dx_root_block *dx_root;
Mark Fashehe7c17e42009-01-29 18:17:46 -08002402 struct ocfs2_dir_block_trailer *trailer =
2403 ocfs2_trailer_from_bh(dirdata_bh, dir->i_sb);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002404
2405 ret = ocfs2_claim_metadata(osb, handle, meta_ac, 1, &dr_suballoc_bit,
2406 &num_bits, &dr_blkno);
2407 if (ret) {
2408 mlog_errno(ret);
2409 goto out;
2410 }
2411
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);
2415
2416 dx_root_bh = sb_getblk(osb->sb, dr_blkno);
2417 if (dx_root_bh == NULL) {
2418 ret = -EIO;
2419 goto out;
2420 }
2421 ocfs2_set_new_buffer_uptodate(dir, dx_root_bh);
2422
2423 ret = ocfs2_journal_access_dr(handle, dir, dx_root_bh,
2424 OCFS2_JOURNAL_ACCESS_CREATE);
2425 if (ret < 0) {
2426 mlog_errno(ret);
2427 goto out;
2428 }
2429
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(osb->slot_num);
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);
Mark Fashehe3a93c22009-02-17 15:29:35 -08002438 dx_root->dr_num_entries = cpu_to_le32(num_entries);
Mark Fashehe7c17e42009-01-29 18:17:46 -08002439 if (le16_to_cpu(trailer->db_free_rec_len))
2440 dx_root->dr_free_blk = cpu_to_le64(dirdata_bh->b_blocknr);
2441 else
2442 dx_root->dr_free_blk = cpu_to_le64(0);
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08002443
2444 if (dx_inline) {
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));
2448 } else {
2449 dx_root->dr_list.l_count =
2450 cpu_to_le16(ocfs2_extent_recs_per_dx_root(osb->sb));
2451 }
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002452
2453 ret = ocfs2_journal_dirty(handle, dx_root_bh);
2454 if (ret)
2455 mlog_errno(ret);
2456
2457 ret = ocfs2_journal_access_di(handle, dir, di_bh,
2458 OCFS2_JOURNAL_ACCESS_CREATE);
2459 if (ret) {
2460 mlog_errno(ret);
2461 goto out;
2462 }
2463
2464 di->i_dx_root = cpu_to_le64(dr_blkno);
2465
2466 OCFS2_I(dir)->ip_dyn_features |= OCFS2_INDEXED_DIR_FL;
2467 di->i_dyn_features = cpu_to_le16(OCFS2_I(dir)->ip_dyn_features);
2468
2469 ret = ocfs2_journal_dirty(handle, di_bh);
2470 if (ret)
2471 mlog_errno(ret);
2472
2473 *ret_dx_root_bh = dx_root_bh;
2474 dx_root_bh = NULL;
2475
2476out:
2477 brelse(dx_root_bh);
2478 return ret;
2479}
2480
2481static int ocfs2_dx_dir_format_cluster(struct ocfs2_super *osb,
2482 handle_t *handle, struct inode *dir,
2483 struct buffer_head **dx_leaves,
2484 int num_dx_leaves, u64 start_blk)
2485{
2486 int ret, i;
2487 struct ocfs2_dx_leaf *dx_leaf;
2488 struct buffer_head *bh;
2489
2490 for (i = 0; i < num_dx_leaves; i++) {
2491 bh = sb_getblk(osb->sb, start_blk + i);
2492 if (bh == NULL) {
2493 ret = -EIO;
2494 goto out;
2495 }
2496 dx_leaves[i] = bh;
2497
2498 ocfs2_set_new_buffer_uptodate(dir, bh);
2499
2500 ret = ocfs2_journal_access_dl(handle, dir, bh,
2501 OCFS2_JOURNAL_ACCESS_CREATE);
2502 if (ret < 0) {
2503 mlog_errno(ret);
2504 goto out;
2505 }
2506
2507 dx_leaf = (struct ocfs2_dx_leaf *) bh->b_data;
2508
2509 memset(dx_leaf, 0, osb->sb->s_blocksize);
2510 strcpy(dx_leaf->dl_signature, OCFS2_DX_LEAF_SIGNATURE);
2511 dx_leaf->dl_fs_generation = cpu_to_le32(osb->fs_generation);
2512 dx_leaf->dl_blkno = cpu_to_le64(bh->b_blocknr);
2513 dx_leaf->dl_list.de_count =
2514 cpu_to_le16(ocfs2_dx_entries_per_leaf(osb->sb));
2515
2516 mlog(0,
2517 "Dir %llu, format dx_leaf: %llu, entry count: %u\n",
2518 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2519 (unsigned long long)bh->b_blocknr,
2520 le16_to_cpu(dx_leaf->dl_list.de_count));
2521
2522 ocfs2_journal_dirty(handle, bh);
2523 }
2524
2525 ret = 0;
2526out:
2527 return ret;
2528}
2529
2530/*
2531 * Allocates and formats a new cluster for use in an indexed dir
2532 * leaf. This version will not do the extent insert, so that it can be
2533 * used by operations which need careful ordering.
2534 */
2535static int __ocfs2_dx_dir_new_cluster(struct inode *dir,
2536 u32 cpos, handle_t *handle,
2537 struct ocfs2_alloc_context *data_ac,
2538 struct buffer_head **dx_leaves,
2539 int num_dx_leaves, u64 *ret_phys_blkno)
2540{
2541 int ret;
2542 u32 phys, num;
2543 u64 phys_blkno;
2544 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2545
2546 /*
2547 * XXX: For create, this should claim cluster for the index
2548 * *before* the unindexed insert so that we have a better
2549 * chance of contiguousness as the directory grows in number
2550 * of entries.
2551 */
2552 ret = __ocfs2_claim_clusters(osb, handle, data_ac, 1, 1, &phys, &num);
2553 if (ret) {
2554 mlog_errno(ret);
2555 goto out;
2556 }
2557
2558 /*
2559 * Format the new cluster first. That way, we're inserting
2560 * valid data.
2561 */
2562 phys_blkno = ocfs2_clusters_to_blocks(osb->sb, phys);
2563 ret = ocfs2_dx_dir_format_cluster(osb, handle, dir, dx_leaves,
2564 num_dx_leaves, phys_blkno);
2565 if (ret) {
2566 mlog_errno(ret);
2567 goto out;
2568 }
2569
2570 *ret_phys_blkno = phys_blkno;
2571out:
2572 return ret;
2573}
2574
2575static int ocfs2_dx_dir_new_cluster(struct inode *dir,
2576 struct ocfs2_extent_tree *et,
2577 u32 cpos, handle_t *handle,
2578 struct ocfs2_alloc_context *data_ac,
2579 struct ocfs2_alloc_context *meta_ac,
2580 struct buffer_head **dx_leaves,
2581 int num_dx_leaves)
2582{
2583 int ret;
2584 u64 phys_blkno;
2585 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2586
2587 ret = __ocfs2_dx_dir_new_cluster(dir, cpos, handle, data_ac, dx_leaves,
2588 num_dx_leaves, &phys_blkno);
2589 if (ret) {
2590 mlog_errno(ret);
2591 goto out;
2592 }
2593
2594 ret = ocfs2_insert_extent(osb, handle, dir, et, cpos, phys_blkno, 1, 0,
2595 meta_ac);
2596 if (ret)
2597 mlog_errno(ret);
2598out:
2599 return ret;
2600}
2601
2602static struct buffer_head **ocfs2_dx_dir_kmalloc_leaves(struct super_block *sb,
2603 int *ret_num_leaves)
2604{
2605 int num_dx_leaves = ocfs2_clusters_to_blocks(sb, 1);
2606 struct buffer_head **dx_leaves;
2607
2608 dx_leaves = kcalloc(num_dx_leaves, sizeof(struct buffer_head *),
2609 GFP_NOFS);
2610 if (dx_leaves && ret_num_leaves)
2611 *ret_num_leaves = num_dx_leaves;
2612
2613 return dx_leaves;
2614}
2615
2616static int ocfs2_fill_new_dir_dx(struct ocfs2_super *osb,
2617 handle_t *handle,
2618 struct inode *parent,
2619 struct inode *inode,
2620 struct buffer_head *di_bh,
2621 struct ocfs2_alloc_context *data_ac,
2622 struct ocfs2_alloc_context *meta_ac)
2623{
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08002624 int ret;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002625 struct buffer_head *leaf_bh = NULL;
2626 struct buffer_head *dx_root_bh = NULL;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002627 struct ocfs2_dx_hinfo hinfo;
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08002628 struct ocfs2_dx_root_block *dx_root;
2629 struct ocfs2_dx_entry_list *entry_list;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002630
2631 /*
2632 * Our strategy is to create the directory as though it were
2633 * unindexed, then add the index block. This works with very
2634 * little complication since the state of a new directory is a
2635 * very well known quantity.
2636 *
2637 * Essentially, we have two dirents ("." and ".."), in the 1st
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08002638 * block which need indexing. These are easily inserted into
2639 * the index block.
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002640 */
2641
2642 ret = ocfs2_fill_new_dir_el(osb, handle, parent, inode, di_bh,
2643 data_ac, &leaf_bh);
2644 if (ret) {
2645 mlog_errno(ret);
2646 goto out;
2647 }
2648
Mark Fashehe7c17e42009-01-29 18:17:46 -08002649 ret = ocfs2_dx_dir_attach_index(osb, handle, inode, di_bh, leaf_bh,
Mark Fashehe3a93c22009-02-17 15:29:35 -08002650 meta_ac, 1, 2, &dx_root_bh);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002651 if (ret) {
2652 mlog_errno(ret);
2653 goto out;
2654 }
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08002655 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
2656 entry_list = &dx_root->dr_entries;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002657
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08002658 /* Buffer has been journaled for us by ocfs2_dx_dir_attach_index */
Mark Fashehe7c17e42009-01-29 18:17:46 -08002659 ocfs2_dx_dir_name_hash(inode, ".", 1, &hinfo);
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08002660 ocfs2_dx_entry_list_insert(entry_list, &hinfo, leaf_bh->b_blocknr);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002661
2662 ocfs2_dx_dir_name_hash(inode, "..", 2, &hinfo);
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08002663 ocfs2_dx_entry_list_insert(entry_list, &hinfo, leaf_bh->b_blocknr);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002664
2665out:
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002666 brelse(dx_root_bh);
2667 brelse(leaf_bh);
2668 return ret;
2669}
2670
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002671int ocfs2_fill_new_dir(struct ocfs2_super *osb,
2672 handle_t *handle,
2673 struct inode *parent,
2674 struct inode *inode,
2675 struct buffer_head *fe_bh,
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002676 struct ocfs2_alloc_context *data_ac,
2677 struct ocfs2_alloc_context *meta_ac)
2678
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002679{
2680 BUG_ON(!ocfs2_supports_inline_data(osb) && data_ac == NULL);
2681
2682 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2683 return ocfs2_fill_new_dir_id(osb, handle, parent, inode, fe_bh);
2684
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002685 if (ocfs2_supports_indexed_dirs(osb))
2686 return ocfs2_fill_new_dir_dx(osb, handle, parent, inode, fe_bh,
2687 data_ac, meta_ac);
2688
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002689 return ocfs2_fill_new_dir_el(osb, handle, parent, inode, fe_bh,
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002690 data_ac, NULL);
2691}
2692
2693static int ocfs2_dx_dir_index_block(struct inode *dir,
2694 handle_t *handle,
2695 struct buffer_head **dx_leaves,
2696 int num_dx_leaves,
Mark Fashehe3a93c22009-02-17 15:29:35 -08002697 u32 *num_dx_entries,
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002698 struct buffer_head *dirent_bh)
2699{
2700 int ret, namelen, i;
2701 char *de_buf, *limit;
2702 struct ocfs2_dir_entry *de;
2703 struct buffer_head *dx_leaf_bh;
2704 struct ocfs2_dx_hinfo hinfo;
2705 u64 dirent_blk = dirent_bh->b_blocknr;
2706
2707 de_buf = dirent_bh->b_data;
2708 limit = de_buf + dir->i_sb->s_blocksize;
2709
2710 while (de_buf < limit) {
2711 de = (struct ocfs2_dir_entry *)de_buf;
2712
2713 namelen = de->name_len;
2714 if (!namelen || !de->inode)
2715 goto inc;
2716
2717 ocfs2_dx_dir_name_hash(dir, de->name, namelen, &hinfo);
2718
2719 i = ocfs2_dx_dir_hash_idx(OCFS2_SB(dir->i_sb), &hinfo);
2720 dx_leaf_bh = dx_leaves[i];
2721
2722 ret = __ocfs2_dx_dir_leaf_insert(dir, handle, &hinfo,
2723 dirent_blk, dx_leaf_bh);
2724 if (ret) {
2725 mlog_errno(ret);
2726 goto out;
2727 }
2728
Mark Fashehe3a93c22009-02-17 15:29:35 -08002729 *num_dx_entries = *num_dx_entries + 1;
2730
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002731inc:
2732 de_buf += le16_to_cpu(de->rec_len);
2733 }
2734
2735out:
2736 return ret;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002737}
Mark Fashehe7c17e42009-01-29 18:17:46 -08002738
2739/*
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08002740 * XXX: This expects dx_root_bh to already be part of the transaction.
2741 */
2742static void ocfs2_dx_dir_index_root_block(struct inode *dir,
2743 struct buffer_head *dx_root_bh,
2744 struct buffer_head *dirent_bh)
2745{
2746 char *de_buf, *limit;
2747 struct ocfs2_dx_root_block *dx_root;
2748 struct ocfs2_dir_entry *de;
2749 struct ocfs2_dx_hinfo hinfo;
2750 u64 dirent_blk = dirent_bh->b_blocknr;
2751
2752 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
2753
2754 de_buf = dirent_bh->b_data;
2755 limit = de_buf + dir->i_sb->s_blocksize;
2756
2757 while (de_buf < limit) {
2758 de = (struct ocfs2_dir_entry *)de_buf;
2759
2760 if (!de->name_len || !de->inode)
2761 goto inc;
2762
2763 ocfs2_dx_dir_name_hash(dir, de->name, de->name_len, &hinfo);
2764
2765 mlog(0,
2766 "dir: %llu, major: 0x%x minor: 0x%x, index: %u, name: %.*s\n",
2767 (unsigned long long)dir->i_ino, hinfo.major_hash,
2768 hinfo.minor_hash,
2769 le16_to_cpu(dx_root->dr_entries.de_num_used),
2770 de->name_len, de->name);
2771
2772 ocfs2_dx_entry_list_insert(&dx_root->dr_entries, &hinfo,
2773 dirent_blk);
Mark Fashehe3a93c22009-02-17 15:29:35 -08002774
2775 le32_add_cpu(&dx_root->dr_num_entries, 1);
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08002776inc:
2777 de_buf += le16_to_cpu(de->rec_len);
2778 }
2779}
2780
2781/*
2782 * Count the number of inline directory entries in di_bh and compare
2783 * them against the number of entries we can hold in an inline dx root
2784 * block.
2785 */
2786static int ocfs2_new_dx_should_be_inline(struct inode *dir,
2787 struct buffer_head *di_bh)
2788{
2789 int dirent_count = 0;
2790 char *de_buf, *limit;
2791 struct ocfs2_dir_entry *de;
2792 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2793
2794 de_buf = di->id2.i_data.id_data;
2795 limit = de_buf + i_size_read(dir);
2796
2797 while (de_buf < limit) {
2798 de = (struct ocfs2_dir_entry *)de_buf;
2799
2800 if (de->name_len && de->inode)
2801 dirent_count++;
2802
2803 de_buf += le16_to_cpu(de->rec_len);
2804 }
2805
2806 /* We are careful to leave room for one extra record. */
2807 return dirent_count < ocfs2_dx_entries_per_root(dir->i_sb);
2808}
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002809
Mark Fasheh87d35a72008-12-10 17:36:25 -08002810/*
2811 * Expand rec_len of the rightmost dirent in a directory block so that it
2812 * contains the end of our valid space for dirents. We do this during
2813 * expansion from an inline directory to one with extents. The first dir block
2814 * in that case is taken from the inline data portion of the inode block.
2815 *
Mark Fashehe7c17e42009-01-29 18:17:46 -08002816 * This will also return the largest amount of contiguous space for a dirent
2817 * in the block. That value is *not* necessarily the last dirent, even after
2818 * expansion. The directory indexing code wants this value for free space
2819 * accounting. We do this here since we're already walking the entire dir
2820 * block.
2821 *
Mark Fasheh87d35a72008-12-10 17:36:25 -08002822 * We add the dir trailer if this filesystem wants it.
2823 */
Mark Fashehe7c17e42009-01-29 18:17:46 -08002824static unsigned int ocfs2_expand_last_dirent(char *start, unsigned int old_size,
2825 struct inode *dir)
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002826{
Mark Fashehe7c17e42009-01-29 18:17:46 -08002827 struct super_block *sb = dir->i_sb;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002828 struct ocfs2_dir_entry *de;
2829 struct ocfs2_dir_entry *prev_de;
2830 char *de_buf, *limit;
Mark Fasheh87d35a72008-12-10 17:36:25 -08002831 unsigned int new_size = sb->s_blocksize;
Mark Fashehe7c17e42009-01-29 18:17:46 -08002832 unsigned int bytes, this_hole;
2833 unsigned int largest_hole = 0;
Mark Fasheh87d35a72008-12-10 17:36:25 -08002834
Mark Fashehe7c17e42009-01-29 18:17:46 -08002835 if (ocfs2_new_dir_wants_trailer(dir))
Mark Fasheh87d35a72008-12-10 17:36:25 -08002836 new_size = ocfs2_dir_trailer_blk_off(sb);
2837
2838 bytes = new_size - old_size;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002839
2840 limit = start + old_size;
2841 de_buf = start;
2842 de = (struct ocfs2_dir_entry *)de_buf;
2843 do {
Mark Fashehe7c17e42009-01-29 18:17:46 -08002844 this_hole = ocfs2_figure_dirent_hole(de);
2845 if (this_hole > largest_hole)
2846 largest_hole = this_hole;
2847
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002848 prev_de = de;
2849 de_buf += le16_to_cpu(de->rec_len);
2850 de = (struct ocfs2_dir_entry *)de_buf;
2851 } while (de_buf < limit);
2852
2853 le16_add_cpu(&prev_de->rec_len, bytes);
Mark Fashehe7c17e42009-01-29 18:17:46 -08002854
2855 /* We need to double check this after modification of the final
2856 * dirent. */
2857 this_hole = ocfs2_figure_dirent_hole(prev_de);
2858 if (this_hole > largest_hole)
2859 largest_hole = this_hole;
2860
2861 if (largest_hole >= OCFS2_DIR_MIN_REC_LEN)
2862 return largest_hole;
2863 return 0;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002864}
2865
2866/*
2867 * We allocate enough clusters to fulfill "blocks_wanted", but set
2868 * i_size to exactly one block. Ocfs2_extend_dir() will handle the
2869 * rest automatically for us.
2870 *
2871 * *first_block_bh is a pointer to the 1st data block allocated to the
2872 * directory.
2873 */
2874static int ocfs2_expand_inline_dir(struct inode *dir, struct buffer_head *di_bh,
2875 unsigned int blocks_wanted,
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002876 struct ocfs2_dir_lookup_result *lookup,
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002877 struct buffer_head **first_block_bh)
2878{
Mark Fashehe3a93c22009-02-17 15:29:35 -08002879 u32 alloc, dx_alloc, bit_off, len, num_dx_entries = 0;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002880 struct super_block *sb = dir->i_sb;
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08002881 int ret, i, num_dx_leaves = 0, dx_inline = 0,
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002882 credits = ocfs2_inline_to_extents_credits(sb);
2883 u64 dx_insert_blkno, blkno,
2884 bytes = blocks_wanted << sb->s_blocksize_bits;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002885 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2886 struct ocfs2_inode_info *oi = OCFS2_I(dir);
2887 struct ocfs2_alloc_context *data_ac;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002888 struct ocfs2_alloc_context *meta_ac = NULL;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002889 struct buffer_head *dirdata_bh = NULL;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002890 struct buffer_head *dx_root_bh = NULL;
2891 struct buffer_head **dx_leaves = NULL;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002892 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2893 handle_t *handle;
Joel Beckerf99b9b72008-08-20 19:36:33 -07002894 struct ocfs2_extent_tree et;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002895 struct ocfs2_extent_tree dx_et;
2896 int did_quota = 0, bytes_allocated = 0;
Joel Beckerf99b9b72008-08-20 19:36:33 -07002897
Joel Becker8d6220d2008-08-22 12:46:09 -07002898 ocfs2_init_dinode_extent_tree(&et, dir, di_bh);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002899
2900 alloc = ocfs2_clusters_for_bytes(sb, bytes);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002901 dx_alloc = 0;
2902
2903 if (ocfs2_supports_indexed_dirs(osb)) {
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002904 credits += ocfs2_add_dir_index_credits(sb);
2905
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08002906 dx_inline = ocfs2_new_dx_should_be_inline(dir, di_bh);
2907 if (!dx_inline) {
2908 /* Add one more cluster for an index leaf */
2909 dx_alloc++;
2910 dx_leaves = ocfs2_dx_dir_kmalloc_leaves(sb,
2911 &num_dx_leaves);
2912 if (!dx_leaves) {
2913 ret = -ENOMEM;
2914 mlog_errno(ret);
2915 goto out;
2916 }
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002917 }
2918
2919 /* This gets us the dx_root */
2920 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
2921 if (ret) {
2922 mlog_errno(ret);
2923 goto out;
2924 }
2925 }
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002926
2927 /*
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002928 * We should never need more than 2 clusters for the unindexed
2929 * tree - maximum dirent size is far less than one block. In
2930 * fact, the only time we'd need more than one cluster is if
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002931 * blocksize == clustersize and the dirent won't fit in the
2932 * extra space that the expansion to a single block gives. As
2933 * of today, that only happens on 4k/4k file systems.
2934 */
2935 BUG_ON(alloc > 2);
2936
2937 ret = ocfs2_reserve_clusters(osb, alloc, &data_ac);
2938 if (ret) {
2939 mlog_errno(ret);
2940 goto out;
2941 }
2942
2943 down_write(&oi->ip_alloc_sem);
2944
2945 /*
Joe Perchesc78bad12008-02-03 17:33:42 +02002946 * Prepare for worst case allocation scenario of two separate
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002947 * extents in the unindexed tree.
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002948 */
2949 if (alloc == 2)
2950 credits += OCFS2_SUBALLOC_ALLOC;
2951
2952 handle = ocfs2_start_trans(osb, credits);
2953 if (IS_ERR(handle)) {
2954 ret = PTR_ERR(handle);
2955 mlog_errno(ret);
2956 goto out_sem;
2957 }
2958
Jan Karaa90714c2008-10-09 19:38:40 +02002959 if (vfs_dq_alloc_space_nodirty(dir,
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002960 ocfs2_clusters_to_bytes(osb->sb,
2961 alloc + dx_alloc))) {
Jan Karaa90714c2008-10-09 19:38:40 +02002962 ret = -EDQUOT;
2963 goto out_commit;
2964 }
2965 did_quota = 1;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002966
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08002967 if (ocfs2_supports_indexed_dirs(osb) && !dx_inline) {
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002968 /*
2969 * Allocate our index cluster first, to maximize the
2970 * possibility that unindexed leaves grow
2971 * contiguously.
2972 */
2973 ret = __ocfs2_dx_dir_new_cluster(dir, 0, handle, data_ac,
2974 dx_leaves, num_dx_leaves,
2975 &dx_insert_blkno);
2976 if (ret) {
2977 mlog_errno(ret);
2978 goto out_commit;
2979 }
2980 bytes_allocated += ocfs2_clusters_to_bytes(dir->i_sb, 1);
2981 }
2982
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002983 /*
2984 * Try to claim as many clusters as the bitmap can give though
2985 * if we only get one now, that's enough to continue. The rest
2986 * will be claimed after the conversion to extents.
2987 */
2988 ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off, &len);
2989 if (ret) {
2990 mlog_errno(ret);
2991 goto out_commit;
2992 }
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002993 bytes_allocated += ocfs2_clusters_to_bytes(dir->i_sb, 1);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002994
2995 /*
2996 * Operations are carefully ordered so that we set up the new
2997 * data block first. The conversion from inline data to
2998 * extents follows.
2999 */
3000 blkno = ocfs2_clusters_to_blocks(dir->i_sb, bit_off);
3001 dirdata_bh = sb_getblk(sb, blkno);
3002 if (!dirdata_bh) {
3003 ret = -EIO;
3004 mlog_errno(ret);
3005 goto out_commit;
3006 }
3007
3008 ocfs2_set_new_buffer_uptodate(dir, dirdata_bh);
3009
Joel Becker13723d02008-10-17 19:25:01 -07003010 ret = ocfs2_journal_access_db(handle, dir, dirdata_bh,
3011 OCFS2_JOURNAL_ACCESS_CREATE);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003012 if (ret) {
3013 mlog_errno(ret);
3014 goto out_commit;
3015 }
3016
3017 memcpy(dirdata_bh->b_data, di->id2.i_data.id_data, i_size_read(dir));
3018 memset(dirdata_bh->b_data + i_size_read(dir), 0,
3019 sb->s_blocksize - i_size_read(dir));
Mark Fashehe7c17e42009-01-29 18:17:46 -08003020 i = ocfs2_expand_last_dirent(dirdata_bh->b_data, i_size_read(dir), dir);
3021 if (ocfs2_new_dir_wants_trailer(dir)) {
3022 /*
3023 * Prepare the dir trailer up front. It will otherwise look
3024 * like a valid dirent. Even if inserting the index fails
3025 * (unlikely), then all we'll have done is given first dir
3026 * block a small amount of fragmentation.
3027 */
3028 ocfs2_init_dir_trailer(dir, dirdata_bh, i);
3029 }
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003030
3031 ret = ocfs2_journal_dirty(handle, dirdata_bh);
3032 if (ret) {
3033 mlog_errno(ret);
3034 goto out_commit;
3035 }
3036
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08003037 if (ocfs2_supports_indexed_dirs(osb) && !dx_inline) {
3038 /*
3039 * Dx dirs with an external cluster need to do this up
3040 * front. Inline dx root's get handled later, after
Mark Fashehe3a93c22009-02-17 15:29:35 -08003041 * we've allocated our root block. We get passed back
3042 * a total number of items so that dr_num_entries can
3043 * be correctly set once the dx_root has been
3044 * allocated.
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08003045 */
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003046 ret = ocfs2_dx_dir_index_block(dir, handle, dx_leaves,
Mark Fashehe3a93c22009-02-17 15:29:35 -08003047 num_dx_leaves, &num_dx_entries,
3048 dirdata_bh);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003049 if (ret) {
3050 mlog_errno(ret);
3051 goto out_commit;
3052 }
3053 }
3054
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003055 /*
3056 * Set extent, i_size, etc on the directory. After this, the
3057 * inode should contain the same exact dirents as before and
3058 * be fully accessible from system calls.
3059 *
3060 * We let the later dirent insert modify c/mtime - to the user
3061 * the data hasn't changed.
3062 */
Joel Becker13723d02008-10-17 19:25:01 -07003063 ret = ocfs2_journal_access_di(handle, dir, di_bh,
3064 OCFS2_JOURNAL_ACCESS_CREATE);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003065 if (ret) {
3066 mlog_errno(ret);
3067 goto out_commit;
3068 }
3069
3070 spin_lock(&oi->ip_lock);
3071 oi->ip_dyn_features &= ~OCFS2_INLINE_DATA_FL;
3072 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
3073 spin_unlock(&oi->ip_lock);
3074
3075 ocfs2_dinode_new_extent_list(dir, di);
3076
3077 i_size_write(dir, sb->s_blocksize);
3078 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3079
3080 di->i_size = cpu_to_le64(sb->s_blocksize);
3081 di->i_ctime = di->i_mtime = cpu_to_le64(dir->i_ctime.tv_sec);
3082 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(dir->i_ctime.tv_nsec);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003083
3084 /*
3085 * This should never fail as our extent list is empty and all
3086 * related blocks have been journaled already.
3087 */
Joel Beckerf99b9b72008-08-20 19:36:33 -07003088 ret = ocfs2_insert_extent(osb, handle, dir, &et, 0, blkno, len,
3089 0, NULL);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003090 if (ret) {
3091 mlog_errno(ret);
Tao Ma83cab532008-08-21 14:14:27 +08003092 goto out_commit;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003093 }
3094
Mark Fasheh9780eb62008-08-05 11:32:46 -07003095 /*
3096 * Set i_blocks after the extent insert for the most up to
3097 * date ip_clusters value.
3098 */
3099 dir->i_blocks = ocfs2_inode_sector_count(dir);
3100
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003101 ret = ocfs2_journal_dirty(handle, di_bh);
3102 if (ret) {
3103 mlog_errno(ret);
3104 goto out_commit;
3105 }
3106
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003107 if (ocfs2_supports_indexed_dirs(osb)) {
3108 ret = ocfs2_dx_dir_attach_index(osb, handle, dir, di_bh,
Mark Fashehe7c17e42009-01-29 18:17:46 -08003109 dirdata_bh, meta_ac, dx_inline,
Mark Fashehe3a93c22009-02-17 15:29:35 -08003110 num_dx_entries, &dx_root_bh);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003111 if (ret) {
3112 mlog_errno(ret);
3113 goto out_commit;
3114 }
3115
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08003116 if (dx_inline) {
3117 ocfs2_dx_dir_index_root_block(dir, dx_root_bh,
3118 dirdata_bh);
3119 } else {
3120 ocfs2_init_dx_root_extent_tree(&dx_et, dir, dx_root_bh);
3121 ret = ocfs2_insert_extent(osb, handle, dir, &dx_et, 0,
3122 dx_insert_blkno, 1, 0, NULL);
3123 if (ret)
3124 mlog_errno(ret);
3125 }
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003126 }
3127
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003128 /*
3129 * We asked for two clusters, but only got one in the 1st
3130 * pass. Claim the 2nd cluster as a separate extent.
3131 */
3132 if (alloc > len) {
3133 ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off,
3134 &len);
3135 if (ret) {
3136 mlog_errno(ret);
3137 goto out_commit;
3138 }
3139 blkno = ocfs2_clusters_to_blocks(dir->i_sb, bit_off);
3140
Joel Beckerf99b9b72008-08-20 19:36:33 -07003141 ret = ocfs2_insert_extent(osb, handle, dir, &et, 1,
3142 blkno, len, 0, NULL);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003143 if (ret) {
3144 mlog_errno(ret);
Tao Ma83cab532008-08-21 14:14:27 +08003145 goto out_commit;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003146 }
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003147 bytes_allocated += ocfs2_clusters_to_bytes(dir->i_sb, 1);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003148 }
3149
3150 *first_block_bh = dirdata_bh;
3151 dirdata_bh = NULL;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003152 if (ocfs2_supports_indexed_dirs(osb)) {
3153 unsigned int off;
3154
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08003155 if (!dx_inline) {
3156 /*
3157 * We need to return the correct block within the
3158 * cluster which should hold our entry.
3159 */
3160 off = ocfs2_dx_dir_hash_idx(OCFS2_SB(dir->i_sb),
3161 &lookup->dl_hinfo);
3162 get_bh(dx_leaves[off]);
3163 lookup->dl_dx_leaf_bh = dx_leaves[off];
3164 }
3165 lookup->dl_dx_root_bh = dx_root_bh;
3166 dx_root_bh = NULL;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003167 }
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003168
3169out_commit:
Jan Karaa90714c2008-10-09 19:38:40 +02003170 if (ret < 0 && did_quota)
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003171 vfs_dq_free_space_nodirty(dir, bytes_allocated);
3172
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003173 ocfs2_commit_trans(osb, handle);
3174
3175out_sem:
3176 up_write(&oi->ip_alloc_sem);
3177
3178out:
3179 if (data_ac)
3180 ocfs2_free_alloc_context(data_ac);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003181 if (meta_ac)
3182 ocfs2_free_alloc_context(meta_ac);
3183
3184 if (dx_leaves) {
3185 for (i = 0; i < num_dx_leaves; i++)
3186 brelse(dx_leaves[i]);
3187 kfree(dx_leaves);
3188 }
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003189
3190 brelse(dirdata_bh);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003191 brelse(dx_root_bh);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003192
3193 return ret;
3194}
3195
Mark Fashehccd979b2005-12-15 14:31:24 -08003196/* returns a bh of the 1st new block in the allocation. */
Mark Fasheh316f4b92007-09-07 18:21:26 -07003197static int ocfs2_do_extend_dir(struct super_block *sb,
3198 handle_t *handle,
3199 struct inode *dir,
3200 struct buffer_head *parent_fe_bh,
3201 struct ocfs2_alloc_context *data_ac,
3202 struct ocfs2_alloc_context *meta_ac,
3203 struct buffer_head **new_bh)
Mark Fashehccd979b2005-12-15 14:31:24 -08003204{
3205 int status;
Jan Karaa90714c2008-10-09 19:38:40 +02003206 int extend, did_quota = 0;
Mark Fasheh8110b072007-03-22 16:53:23 -07003207 u64 p_blkno, v_blkno;
Mark Fashehccd979b2005-12-15 14:31:24 -08003208
3209 spin_lock(&OCFS2_I(dir)->ip_lock);
3210 extend = (i_size_read(dir) == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters));
3211 spin_unlock(&OCFS2_I(dir)->ip_lock);
3212
3213 if (extend) {
Mark Fashehdcd05382007-01-16 11:32:23 -08003214 u32 offset = OCFS2_I(dir)->ip_clusters;
3215
Jan Karaa90714c2008-10-09 19:38:40 +02003216 if (vfs_dq_alloc_space_nodirty(dir,
3217 ocfs2_clusters_to_bytes(sb, 1))) {
3218 status = -EDQUOT;
3219 goto bail;
3220 }
3221 did_quota = 1;
3222
Tao Ma0eb8d472008-08-18 17:38:45 +08003223 status = ocfs2_add_inode_data(OCFS2_SB(sb), dir, &offset,
3224 1, 0, parent_fe_bh, handle,
3225 data_ac, meta_ac, NULL);
Mark Fashehccd979b2005-12-15 14:31:24 -08003226 BUG_ON(status == -EAGAIN);
3227 if (status < 0) {
3228 mlog_errno(status);
3229 goto bail;
3230 }
3231 }
3232
Mark Fasheh8110b072007-03-22 16:53:23 -07003233 v_blkno = ocfs2_blocks_for_bytes(sb, i_size_read(dir));
3234 status = ocfs2_extent_map_get_blocks(dir, v_blkno, &p_blkno, NULL, NULL);
Mark Fashehccd979b2005-12-15 14:31:24 -08003235 if (status < 0) {
3236 mlog_errno(status);
3237 goto bail;
3238 }
3239
3240 *new_bh = sb_getblk(sb, p_blkno);
3241 if (!*new_bh) {
3242 status = -EIO;
3243 mlog_errno(status);
3244 goto bail;
3245 }
3246 status = 0;
3247bail:
Jan Karaa90714c2008-10-09 19:38:40 +02003248 if (did_quota && status < 0)
3249 vfs_dq_free_space_nodirty(dir, ocfs2_clusters_to_bytes(sb, 1));
Mark Fashehccd979b2005-12-15 14:31:24 -08003250 mlog_exit(status);
3251 return status;
3252}
3253
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003254/*
3255 * Assumes you already have a cluster lock on the directory.
3256 *
3257 * 'blocks_wanted' is only used if we have an inline directory which
3258 * is to be turned into an extent based one. The size of the dirent to
3259 * insert might be larger than the space gained by growing to just one
3260 * block, so we may have to grow the inode by two blocks in that case.
Mark Fashehe7c17e42009-01-29 18:17:46 -08003261 *
3262 * If the directory is already indexed, dx_root_bh must be provided.
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003263 */
Mark Fashehccd979b2005-12-15 14:31:24 -08003264static int ocfs2_extend_dir(struct ocfs2_super *osb,
3265 struct inode *dir,
3266 struct buffer_head *parent_fe_bh,
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003267 unsigned int blocks_wanted,
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003268 struct ocfs2_dir_lookup_result *lookup,
Mark Fashehccd979b2005-12-15 14:31:24 -08003269 struct buffer_head **new_de_bh)
3270{
3271 int status = 0;
Joel Beckeree19a772007-03-28 18:27:07 -07003272 int credits, num_free_extents, drop_alloc_sem = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08003273 loff_t dir_i_size;
3274 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
Tao Ma811f9332008-08-18 17:38:43 +08003275 struct ocfs2_extent_list *el = &fe->id2.i_list;
Mark Fashehccd979b2005-12-15 14:31:24 -08003276 struct ocfs2_alloc_context *data_ac = NULL;
3277 struct ocfs2_alloc_context *meta_ac = NULL;
Mark Fasheh1fabe142006-10-09 18:11:45 -07003278 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08003279 struct buffer_head *new_bh = NULL;
3280 struct ocfs2_dir_entry * de;
3281 struct super_block *sb = osb->sb;
Joel Beckerf99b9b72008-08-20 19:36:33 -07003282 struct ocfs2_extent_tree et;
Mark Fashehe7c17e42009-01-29 18:17:46 -08003283 struct buffer_head *dx_root_bh = lookup->dl_dx_root_bh;
Mark Fashehccd979b2005-12-15 14:31:24 -08003284
3285 mlog_entry_void();
3286
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003287 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
Mark Fashehe7c17e42009-01-29 18:17:46 -08003288 /*
3289 * This would be a code error as an inline directory should
3290 * never have an index root.
3291 */
3292 BUG_ON(dx_root_bh);
3293
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003294 status = ocfs2_expand_inline_dir(dir, parent_fe_bh,
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003295 blocks_wanted, lookup,
3296 &new_bh);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003297 if (status) {
3298 mlog_errno(status);
3299 goto bail;
3300 }
3301
Mark Fashehe7c17e42009-01-29 18:17:46 -08003302 /* Expansion from inline to an indexed directory will
3303 * have given us this. */
3304 dx_root_bh = lookup->dl_dx_root_bh;
3305
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003306 if (blocks_wanted == 1) {
3307 /*
3308 * If the new dirent will fit inside the space
3309 * created by pushing out to one block, then
3310 * we can complete the operation
3311 * here. Otherwise we have to expand i_size
3312 * and format the 2nd block below.
3313 */
3314 BUG_ON(new_bh == NULL);
3315 goto bail_bh;
3316 }
3317
3318 /*
3319 * Get rid of 'new_bh' - we want to format the 2nd
3320 * data block and return that instead.
3321 */
3322 brelse(new_bh);
3323 new_bh = NULL;
3324
3325 dir_i_size = i_size_read(dir);
3326 credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS;
3327 goto do_extend;
3328 }
3329
Mark Fashehccd979b2005-12-15 14:31:24 -08003330 dir_i_size = i_size_read(dir);
Mark Fashehb06970532006-03-03 10:24:33 -08003331 mlog(0, "extending dir %llu (i_size = %lld)\n",
3332 (unsigned long long)OCFS2_I(dir)->ip_blkno, dir_i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -08003333
Mark Fashehccd979b2005-12-15 14:31:24 -08003334 /* dir->i_size is always block aligned. */
3335 spin_lock(&OCFS2_I(dir)->ip_lock);
3336 if (dir_i_size == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters)) {
3337 spin_unlock(&OCFS2_I(dir)->ip_lock);
Joel Becker8d6220d2008-08-22 12:46:09 -07003338 ocfs2_init_dinode_extent_tree(&et, dir, parent_fe_bh);
Joel Beckerf99b9b72008-08-20 19:36:33 -07003339 num_free_extents = ocfs2_num_free_extents(osb, dir, &et);
Mark Fashehccd979b2005-12-15 14:31:24 -08003340 if (num_free_extents < 0) {
3341 status = num_free_extents;
3342 mlog_errno(status);
3343 goto bail;
3344 }
3345
3346 if (!num_free_extents) {
Tao Ma811f9332008-08-18 17:38:43 +08003347 status = ocfs2_reserve_new_metadata(osb, el, &meta_ac);
Mark Fashehccd979b2005-12-15 14:31:24 -08003348 if (status < 0) {
3349 if (status != -ENOSPC)
3350 mlog_errno(status);
3351 goto bail;
3352 }
3353 }
3354
Mark Fashehda5cbf22006-10-06 18:34:35 -07003355 status = ocfs2_reserve_clusters(osb, 1, &data_ac);
Mark Fashehccd979b2005-12-15 14:31:24 -08003356 if (status < 0) {
3357 if (status != -ENOSPC)
3358 mlog_errno(status);
3359 goto bail;
3360 }
3361
Tao Ma811f9332008-08-18 17:38:43 +08003362 credits = ocfs2_calc_extend_credits(sb, el, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08003363 } else {
3364 spin_unlock(&OCFS2_I(dir)->ip_lock);
3365 credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS;
3366 }
3367
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003368do_extend:
Mark Fashehe7c17e42009-01-29 18:17:46 -08003369 if (ocfs2_dir_indexed(dir))
3370 credits++; /* For attaching the new dirent block to the
3371 * dx_root */
3372
Joel Beckeree19a772007-03-28 18:27:07 -07003373 down_write(&OCFS2_I(dir)->ip_alloc_sem);
3374 drop_alloc_sem = 1;
3375
Mark Fasheh65eff9c2006-10-09 17:26:22 -07003376 handle = ocfs2_start_trans(osb, credits);
Mark Fashehccd979b2005-12-15 14:31:24 -08003377 if (IS_ERR(handle)) {
3378 status = PTR_ERR(handle);
3379 handle = NULL;
3380 mlog_errno(status);
3381 goto bail;
3382 }
3383
3384 status = ocfs2_do_extend_dir(osb->sb, handle, dir, parent_fe_bh,
3385 data_ac, meta_ac, &new_bh);
3386 if (status < 0) {
3387 mlog_errno(status);
3388 goto bail;
3389 }
3390
3391 ocfs2_set_new_buffer_uptodate(dir, new_bh);
3392
Joel Becker13723d02008-10-17 19:25:01 -07003393 status = ocfs2_journal_access_db(handle, dir, new_bh,
3394 OCFS2_JOURNAL_ACCESS_CREATE);
Mark Fashehccd979b2005-12-15 14:31:24 -08003395 if (status < 0) {
3396 mlog_errno(status);
3397 goto bail;
3398 }
3399 memset(new_bh->b_data, 0, sb->s_blocksize);
Mark Fasheh87d35a72008-12-10 17:36:25 -08003400
Mark Fashehccd979b2005-12-15 14:31:24 -08003401 de = (struct ocfs2_dir_entry *) new_bh->b_data;
3402 de->inode = 0;
Mark Fashehe7c17e42009-01-29 18:17:46 -08003403 if (ocfs2_supports_dir_trailer(dir)) {
Mark Fasheh87d35a72008-12-10 17:36:25 -08003404 de->rec_len = cpu_to_le16(ocfs2_dir_trailer_blk_off(sb));
Mark Fashehe7c17e42009-01-29 18:17:46 -08003405
3406 ocfs2_init_dir_trailer(dir, new_bh, le16_to_cpu(de->rec_len));
3407
3408 if (ocfs2_dir_indexed(dir)) {
3409 status = ocfs2_dx_dir_link_trailer(dir, handle,
3410 dx_root_bh, new_bh);
3411 if (status) {
3412 mlog_errno(status);
3413 goto bail;
3414 }
3415 }
Mark Fasheh87d35a72008-12-10 17:36:25 -08003416 } else {
3417 de->rec_len = cpu_to_le16(sb->s_blocksize);
3418 }
Mark Fashehccd979b2005-12-15 14:31:24 -08003419 status = ocfs2_journal_dirty(handle, new_bh);
3420 if (status < 0) {
3421 mlog_errno(status);
3422 goto bail;
3423 }
3424
3425 dir_i_size += dir->i_sb->s_blocksize;
3426 i_size_write(dir, dir_i_size);
Mark Fasheh8110b072007-03-22 16:53:23 -07003427 dir->i_blocks = ocfs2_inode_sector_count(dir);
Mark Fashehccd979b2005-12-15 14:31:24 -08003428 status = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
3429 if (status < 0) {
3430 mlog_errno(status);
3431 goto bail;
3432 }
3433
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003434bail_bh:
Mark Fashehccd979b2005-12-15 14:31:24 -08003435 *new_de_bh = new_bh;
3436 get_bh(*new_de_bh);
3437bail:
Joel Beckeree19a772007-03-28 18:27:07 -07003438 if (drop_alloc_sem)
3439 up_write(&OCFS2_I(dir)->ip_alloc_sem);
Mark Fashehccd979b2005-12-15 14:31:24 -08003440 if (handle)
Mark Fasheh02dc1af2006-10-09 16:48:10 -07003441 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08003442
3443 if (data_ac)
3444 ocfs2_free_alloc_context(data_ac);
3445 if (meta_ac)
3446 ocfs2_free_alloc_context(meta_ac);
3447
Mark Fasheha81cb882008-10-07 14:25:16 -07003448 brelse(new_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08003449
3450 mlog_exit(status);
3451 return status;
3452}
3453
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003454static int ocfs2_find_dir_space_id(struct inode *dir, struct buffer_head *di_bh,
3455 const char *name, int namelen,
3456 struct buffer_head **ret_de_bh,
3457 unsigned int *blocks_wanted)
3458{
3459 int ret;
Mark Fasheh87d35a72008-12-10 17:36:25 -08003460 struct super_block *sb = dir->i_sb;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003461 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3462 struct ocfs2_dir_entry *de, *last_de = NULL;
3463 char *de_buf, *limit;
3464 unsigned long offset = 0;
Mark Fasheh87d35a72008-12-10 17:36:25 -08003465 unsigned int rec_len, new_rec_len, free_space = dir->i_sb->s_blocksize;
3466
3467 /*
3468 * This calculates how many free bytes we'd have in block zero, should
3469 * this function force expansion to an extent tree.
3470 */
Mark Fashehe7c17e42009-01-29 18:17:46 -08003471 if (ocfs2_new_dir_wants_trailer(dir))
Mark Fasheh87d35a72008-12-10 17:36:25 -08003472 free_space = ocfs2_dir_trailer_blk_off(sb) - i_size_read(dir);
3473 else
3474 free_space = dir->i_sb->s_blocksize - i_size_read(dir);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003475
3476 de_buf = di->id2.i_data.id_data;
3477 limit = de_buf + i_size_read(dir);
3478 rec_len = OCFS2_DIR_REC_LEN(namelen);
3479
3480 while (de_buf < limit) {
3481 de = (struct ocfs2_dir_entry *)de_buf;
3482
3483 if (!ocfs2_check_dir_entry(dir, de, di_bh, offset)) {
3484 ret = -ENOENT;
3485 goto out;
3486 }
3487 if (ocfs2_match(namelen, name, de)) {
3488 ret = -EEXIST;
3489 goto out;
3490 }
Mark Fasheh87d35a72008-12-10 17:36:25 -08003491 /*
3492 * No need to check for a trailing dirent record here as
3493 * they're not used for inline dirs.
3494 */
3495
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003496 if (ocfs2_dirent_would_fit(de, rec_len)) {
3497 /* Ok, we found a spot. Return this bh and let
3498 * the caller actually fill it in. */
3499 *ret_de_bh = di_bh;
3500 get_bh(*ret_de_bh);
3501 ret = 0;
3502 goto out;
3503 }
3504
3505 last_de = de;
3506 de_buf += le16_to_cpu(de->rec_len);
3507 offset += le16_to_cpu(de->rec_len);
3508 }
3509
3510 /*
3511 * We're going to require expansion of the directory - figure
3512 * out how many blocks we'll need so that a place for the
3513 * dirent can be found.
3514 */
3515 *blocks_wanted = 1;
Mark Fasheh87d35a72008-12-10 17:36:25 -08003516 new_rec_len = le16_to_cpu(last_de->rec_len) + free_space;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003517 if (new_rec_len < (rec_len + OCFS2_DIR_REC_LEN(last_de->name_len)))
3518 *blocks_wanted = 2;
3519
3520 ret = -ENOSPC;
3521out:
3522 return ret;
3523}
3524
3525static int ocfs2_find_dir_space_el(struct inode *dir, const char *name,
3526 int namelen, struct buffer_head **ret_de_bh)
Mark Fashehccd979b2005-12-15 14:31:24 -08003527{
3528 unsigned long offset;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003529 struct buffer_head *bh = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08003530 unsigned short rec_len;
Mark Fashehccd979b2005-12-15 14:31:24 -08003531 struct ocfs2_dir_entry *de;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003532 struct super_block *sb = dir->i_sb;
Mark Fashehccd979b2005-12-15 14:31:24 -08003533 int status;
Mark Fasheh87d35a72008-12-10 17:36:25 -08003534 int blocksize = dir->i_sb->s_blocksize;
Mark Fashehccd979b2005-12-15 14:31:24 -08003535
Joel Beckera22305c2008-11-13 14:49:17 -08003536 status = ocfs2_read_dir_block(dir, 0, &bh, 0);
3537 if (status) {
Mark Fashehccd979b2005-12-15 14:31:24 -08003538 mlog_errno(status);
3539 goto bail;
3540 }
3541
3542 rec_len = OCFS2_DIR_REC_LEN(namelen);
3543 offset = 0;
3544 de = (struct ocfs2_dir_entry *) bh->b_data;
3545 while (1) {
3546 if ((char *)de >= sb->s_blocksize + bh->b_data) {
3547 brelse(bh);
3548 bh = NULL;
3549
3550 if (i_size_read(dir) <= offset) {
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003551 /*
3552 * Caller will have to expand this
3553 * directory.
3554 */
3555 status = -ENOSPC;
Mark Fashehccd979b2005-12-15 14:31:24 -08003556 goto bail;
3557 }
Joel Beckera22305c2008-11-13 14:49:17 -08003558 status = ocfs2_read_dir_block(dir,
3559 offset >> sb->s_blocksize_bits,
3560 &bh, 0);
3561 if (status) {
Mark Fashehccd979b2005-12-15 14:31:24 -08003562 mlog_errno(status);
3563 goto bail;
3564 }
3565 /* move to next block */
3566 de = (struct ocfs2_dir_entry *) bh->b_data;
3567 }
3568 if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
3569 status = -ENOENT;
3570 goto bail;
3571 }
3572 if (ocfs2_match(namelen, name, de)) {
3573 status = -EEXIST;
3574 goto bail;
3575 }
Mark Fasheh87d35a72008-12-10 17:36:25 -08003576
3577 if (ocfs2_skip_dir_trailer(dir, de, offset % blocksize,
3578 blocksize))
3579 goto next;
3580
Mark Fasheh8553cf42007-09-13 16:29:01 -07003581 if (ocfs2_dirent_would_fit(de, rec_len)) {
Mark Fashehccd979b2005-12-15 14:31:24 -08003582 /* Ok, we found a spot. Return this bh and let
3583 * the caller actually fill it in. */
3584 *ret_de_bh = bh;
3585 get_bh(*ret_de_bh);
3586 status = 0;
3587 goto bail;
3588 }
Mark Fasheh87d35a72008-12-10 17:36:25 -08003589next:
Mark Fashehccd979b2005-12-15 14:31:24 -08003590 offset += le16_to_cpu(de->rec_len);
3591 de = (struct ocfs2_dir_entry *)((char *) de + le16_to_cpu(de->rec_len));
3592 }
3593
3594 status = 0;
3595bail:
Mark Fasheha81cb882008-10-07 14:25:16 -07003596 brelse(bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08003597
3598 mlog_exit(status);
3599 return status;
3600}
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003601
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003602static int dx_leaf_sort_cmp(const void *a, const void *b)
3603{
3604 const struct ocfs2_dx_entry *entry1 = a;
3605 const struct ocfs2_dx_entry *entry2 = b;
3606 u32 major_hash1 = le32_to_cpu(entry1->dx_major_hash);
3607 u32 major_hash2 = le32_to_cpu(entry2->dx_major_hash);
3608 u32 minor_hash1 = le32_to_cpu(entry1->dx_minor_hash);
3609 u32 minor_hash2 = le32_to_cpu(entry2->dx_minor_hash);
3610
3611 if (major_hash1 > major_hash2)
3612 return 1;
3613 if (major_hash1 < major_hash2)
3614 return -1;
3615
3616 /*
3617 * It is not strictly necessary to sort by minor
3618 */
3619 if (minor_hash1 > minor_hash2)
3620 return 1;
3621 if (minor_hash1 < minor_hash2)
3622 return -1;
3623 return 0;
3624}
3625
3626static void dx_leaf_sort_swap(void *a, void *b, int size)
3627{
3628 struct ocfs2_dx_entry *entry1 = a;
3629 struct ocfs2_dx_entry *entry2 = b;
3630 struct ocfs2_dx_entry tmp;
3631
3632 BUG_ON(size != sizeof(*entry1));
3633
3634 tmp = *entry1;
3635 *entry1 = *entry2;
3636 *entry2 = tmp;
3637}
3638
3639static int ocfs2_dx_leaf_same_major(struct ocfs2_dx_leaf *dx_leaf)
3640{
3641 struct ocfs2_dx_entry_list *dl_list = &dx_leaf->dl_list;
3642 int i, num = le16_to_cpu(dl_list->de_num_used);
3643
3644 for (i = 0; i < (num - 1); i++) {
3645 if (le32_to_cpu(dl_list->de_entries[i].dx_major_hash) !=
3646 le32_to_cpu(dl_list->de_entries[i + 1].dx_major_hash))
3647 return 0;
3648 }
3649
3650 return 1;
3651}
3652
3653/*
3654 * Find the optimal value to split this leaf on. This expects the leaf
3655 * entries to be in sorted order.
3656 *
3657 * leaf_cpos is the cpos of the leaf we're splitting. insert_hash is
3658 * the hash we want to insert.
3659 *
3660 * This function is only concerned with the major hash - that which
3661 * determines which cluster an item belongs to.
3662 */
3663static int ocfs2_dx_dir_find_leaf_split(struct ocfs2_dx_leaf *dx_leaf,
3664 u32 leaf_cpos, u32 insert_hash,
3665 u32 *split_hash)
3666{
3667 struct ocfs2_dx_entry_list *dl_list = &dx_leaf->dl_list;
3668 int i, num_used = le16_to_cpu(dl_list->de_num_used);
3669 int allsame;
3670
3671 /*
3672 * There's a couple rare, but nasty corner cases we have to
3673 * check for here. All of them involve a leaf where all value
3674 * have the same hash, which is what we look for first.
3675 *
3676 * Most of the time, all of the above is false, and we simply
3677 * pick the median value for a split.
3678 */
3679 allsame = ocfs2_dx_leaf_same_major(dx_leaf);
3680 if (allsame) {
3681 u32 val = le32_to_cpu(dl_list->de_entries[0].dx_major_hash);
3682
3683 if (val == insert_hash) {
3684 /*
3685 * No matter where we would choose to split,
3686 * the new entry would want to occupy the same
3687 * block as these. Since there's no space left
3688 * in their existing block, we know there
3689 * won't be space after the split.
3690 */
3691 return -ENOSPC;
3692 }
3693
3694 if (val == leaf_cpos) {
3695 /*
3696 * Because val is the same as leaf_cpos (which
3697 * is the smallest value this leaf can have),
3698 * yet is not equal to insert_hash, then we
3699 * know that insert_hash *must* be larger than
3700 * val (and leaf_cpos). At least cpos+1 in value.
3701 *
3702 * We also know then, that there cannot be an
3703 * adjacent extent (otherwise we'd be looking
3704 * at it). Choosing this value gives us a
3705 * chance to get some contiguousness.
3706 */
3707 *split_hash = leaf_cpos + 1;
3708 return 0;
3709 }
3710
3711 if (val > insert_hash) {
3712 /*
3713 * val can not be the same as insert hash, and
3714 * also must be larger than leaf_cpos. Also,
3715 * we know that there can't be a leaf between
3716 * cpos and val, otherwise the entries with
3717 * hash 'val' would be there.
3718 */
3719 *split_hash = val;
3720 return 0;
3721 }
3722
3723 *split_hash = insert_hash;
3724 return 0;
3725 }
3726
3727 /*
3728 * Since the records are sorted and the checks above
3729 * guaranteed that not all records in this block are the same,
3730 * we simple travel forward, from the median, and pick the 1st
3731 * record whose value is larger than leaf_cpos.
3732 */
3733 for (i = (num_used / 2); i < num_used; i++)
3734 if (le32_to_cpu(dl_list->de_entries[i].dx_major_hash) >
3735 leaf_cpos)
3736 break;
3737
3738 BUG_ON(i == num_used); /* Should be impossible */
3739 *split_hash = le32_to_cpu(dl_list->de_entries[i].dx_major_hash);
3740 return 0;
3741}
3742
3743/*
3744 * Transfer all entries in orig_dx_leaves whose major hash is equal to or
3745 * larger than split_hash into new_dx_leaves. We use a temporary
3746 * buffer (tmp_dx_leaf) to make the changes to the original leaf blocks.
3747 *
3748 * Since the block offset inside a leaf (cluster) is a constant mask
3749 * of minor_hash, we can optimize - an item at block offset X within
3750 * the original cluster, will be at offset X within the new cluster.
3751 */
3752static void ocfs2_dx_dir_transfer_leaf(struct inode *dir, u32 split_hash,
3753 handle_t *handle,
3754 struct ocfs2_dx_leaf *tmp_dx_leaf,
3755 struct buffer_head **orig_dx_leaves,
3756 struct buffer_head **new_dx_leaves,
3757 int num_dx_leaves)
3758{
3759 int i, j, num_used;
3760 u32 major_hash;
3761 struct ocfs2_dx_leaf *orig_dx_leaf, *new_dx_leaf;
3762 struct ocfs2_dx_entry_list *orig_list, *new_list, *tmp_list;
3763 struct ocfs2_dx_entry *dx_entry;
3764
3765 tmp_list = &tmp_dx_leaf->dl_list;
3766
3767 for (i = 0; i < num_dx_leaves; i++) {
3768 orig_dx_leaf = (struct ocfs2_dx_leaf *) orig_dx_leaves[i]->b_data;
3769 orig_list = &orig_dx_leaf->dl_list;
3770 new_dx_leaf = (struct ocfs2_dx_leaf *) new_dx_leaves[i]->b_data;
3771 new_list = &new_dx_leaf->dl_list;
3772
3773 num_used = le16_to_cpu(orig_list->de_num_used);
3774
3775 memcpy(tmp_dx_leaf, orig_dx_leaf, dir->i_sb->s_blocksize);
3776 tmp_list->de_num_used = cpu_to_le16(0);
3777 memset(&tmp_list->de_entries, 0, sizeof(*dx_entry)*num_used);
3778
3779 for (j = 0; j < num_used; j++) {
3780 dx_entry = &orig_list->de_entries[j];
3781 major_hash = le32_to_cpu(dx_entry->dx_major_hash);
3782 if (major_hash >= split_hash)
3783 ocfs2_dx_dir_leaf_insert_tail(new_dx_leaf,
3784 dx_entry);
3785 else
3786 ocfs2_dx_dir_leaf_insert_tail(tmp_dx_leaf,
3787 dx_entry);
3788 }
3789 memcpy(orig_dx_leaf, tmp_dx_leaf, dir->i_sb->s_blocksize);
3790
3791 ocfs2_journal_dirty(handle, orig_dx_leaves[i]);
3792 ocfs2_journal_dirty(handle, new_dx_leaves[i]);
3793 }
3794}
3795
3796static int ocfs2_dx_dir_rebalance_credits(struct ocfs2_super *osb,
3797 struct ocfs2_dx_root_block *dx_root)
3798{
3799 int credits = ocfs2_clusters_to_blocks(osb->sb, 2);
3800
3801 credits += ocfs2_calc_extend_credits(osb->sb, &dx_root->dr_list, 1);
3802 credits += ocfs2_quota_trans_credits(osb->sb);
3803 return credits;
3804}
3805
3806/*
3807 * Find the median value in dx_leaf_bh and allocate a new leaf to move
3808 * half our entries into.
3809 */
3810static int ocfs2_dx_dir_rebalance(struct ocfs2_super *osb, struct inode *dir,
3811 struct buffer_head *dx_root_bh,
3812 struct buffer_head *dx_leaf_bh,
3813 struct ocfs2_dx_hinfo *hinfo, u32 leaf_cpos,
3814 u64 leaf_blkno)
3815{
3816 struct ocfs2_dx_leaf *dx_leaf = (struct ocfs2_dx_leaf *)dx_leaf_bh->b_data;
3817 int credits, ret, i, num_used, did_quota = 0;
3818 u32 cpos, split_hash, insert_hash = hinfo->major_hash;
3819 u64 orig_leaves_start;
3820 int num_dx_leaves;
3821 struct buffer_head **orig_dx_leaves = NULL;
3822 struct buffer_head **new_dx_leaves = NULL;
3823 struct ocfs2_alloc_context *data_ac = NULL, *meta_ac = NULL;
3824 struct ocfs2_extent_tree et;
3825 handle_t *handle = NULL;
3826 struct ocfs2_dx_root_block *dx_root;
3827 struct ocfs2_dx_leaf *tmp_dx_leaf = NULL;
3828
3829 mlog(0, "DX Dir: %llu, rebalance leaf leaf_blkno: %llu insert: %u\n",
3830 (unsigned long long)OCFS2_I(dir)->ip_blkno,
3831 (unsigned long long)leaf_blkno, insert_hash);
3832
3833 ocfs2_init_dx_root_extent_tree(&et, dir, dx_root_bh);
3834
3835 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
3836 /*
3837 * XXX: This is a rather large limit. We should use a more
3838 * realistic value.
3839 */
3840 if (le32_to_cpu(dx_root->dr_clusters) == UINT_MAX)
3841 return -ENOSPC;
3842
3843 num_used = le16_to_cpu(dx_leaf->dl_list.de_num_used);
3844 if (num_used < le16_to_cpu(dx_leaf->dl_list.de_count)) {
3845 mlog(ML_ERROR, "DX Dir: %llu, Asked to rebalance empty leaf: "
3846 "%llu, %d\n", (unsigned long long)OCFS2_I(dir)->ip_blkno,
3847 (unsigned long long)leaf_blkno, num_used);
3848 ret = -EIO;
3849 goto out;
3850 }
3851
3852 orig_dx_leaves = ocfs2_dx_dir_kmalloc_leaves(osb->sb, &num_dx_leaves);
3853 if (!orig_dx_leaves) {
3854 ret = -ENOMEM;
3855 mlog_errno(ret);
3856 goto out;
3857 }
3858
3859 new_dx_leaves = ocfs2_dx_dir_kmalloc_leaves(osb->sb, NULL);
3860 if (!new_dx_leaves) {
3861 ret = -ENOMEM;
3862 mlog_errno(ret);
3863 goto out;
3864 }
3865
3866 ret = ocfs2_lock_allocators(dir, &et, 1, 0, &data_ac, &meta_ac);
3867 if (ret) {
3868 if (ret != -ENOSPC)
3869 mlog_errno(ret);
3870 goto out;
3871 }
3872
3873 credits = ocfs2_dx_dir_rebalance_credits(osb, dx_root);
3874 handle = ocfs2_start_trans(osb, credits);
3875 if (IS_ERR(handle)) {
3876 ret = PTR_ERR(handle);
3877 handle = NULL;
3878 mlog_errno(ret);
3879 goto out;
3880 }
3881
3882 if (vfs_dq_alloc_space_nodirty(dir,
3883 ocfs2_clusters_to_bytes(dir->i_sb, 1))) {
3884 ret = -EDQUOT;
3885 goto out_commit;
3886 }
3887 did_quota = 1;
3888
3889 ret = ocfs2_journal_access_dl(handle, dir, dx_leaf_bh,
3890 OCFS2_JOURNAL_ACCESS_WRITE);
3891 if (ret) {
3892 mlog_errno(ret);
3893 goto out_commit;
3894 }
3895
3896 /*
3897 * This block is changing anyway, so we can sort it in place.
3898 */
3899 sort(dx_leaf->dl_list.de_entries, num_used,
3900 sizeof(struct ocfs2_dx_entry), dx_leaf_sort_cmp,
3901 dx_leaf_sort_swap);
3902
3903 ret = ocfs2_journal_dirty(handle, dx_leaf_bh);
3904 if (ret) {
3905 mlog_errno(ret);
3906 goto out_commit;
3907 }
3908
3909 ret = ocfs2_dx_dir_find_leaf_split(dx_leaf, leaf_cpos, insert_hash,
3910 &split_hash);
3911 if (ret) {
3912 mlog_errno(ret);
3913 goto out_commit;
3914 }
3915
3916 mlog(0, "Split leaf (%u) at %u, insert major hash is %u\n",
3917 leaf_cpos, split_hash, insert_hash);
3918
3919 /*
3920 * We have to carefully order operations here. There are items
3921 * which want to be in the new cluster before insert, but in
3922 * order to put those items in the new cluster, we alter the
3923 * old cluster. A failure to insert gets nasty.
3924 *
3925 * So, start by reserving writes to the old
3926 * cluster. ocfs2_dx_dir_new_cluster will reserve writes on
3927 * the new cluster for us, before inserting it. The insert
3928 * won't happen if there's an error before that. Once the
3929 * insert is done then, we can transfer from one leaf into the
3930 * other without fear of hitting any error.
3931 */
3932
3933 /*
3934 * The leaf transfer wants some scratch space so that we don't
3935 * wind up doing a bunch of expensive memmove().
3936 */
3937 tmp_dx_leaf = kmalloc(osb->sb->s_blocksize, GFP_NOFS);
3938 if (!tmp_dx_leaf) {
3939 ret = -ENOMEM;
3940 mlog_errno(ret);
3941 goto out_commit;
3942 }
3943
3944 orig_leaves_start = leaf_blkno & ~(osb->s_clustersize_bits -
3945 osb->sb->s_blocksize_bits);
3946 ret = ocfs2_read_dx_leaves(dir, orig_leaves_start, num_dx_leaves,
3947 orig_dx_leaves);
3948 if (ret) {
3949 mlog_errno(ret);
3950 goto out_commit;
3951 }
3952
3953 for (i = 0; i < num_dx_leaves; i++) {
3954 ret = ocfs2_journal_access_dl(handle, dir, orig_dx_leaves[i],
3955 OCFS2_JOURNAL_ACCESS_WRITE);
3956 if (ret) {
3957 mlog_errno(ret);
3958 goto out_commit;
3959 }
3960 }
3961
3962 cpos = split_hash;
3963 ret = ocfs2_dx_dir_new_cluster(dir, &et, cpos, handle,
3964 data_ac, meta_ac, new_dx_leaves,
3965 num_dx_leaves);
3966 if (ret) {
3967 mlog_errno(ret);
3968 goto out_commit;
3969 }
3970
3971 ocfs2_dx_dir_transfer_leaf(dir, split_hash, handle, tmp_dx_leaf,
3972 orig_dx_leaves, new_dx_leaves, num_dx_leaves);
3973
3974out_commit:
3975 if (ret < 0 && did_quota)
3976 vfs_dq_free_space_nodirty(dir,
3977 ocfs2_clusters_to_bytes(dir->i_sb, 1));
3978
3979 ocfs2_commit_trans(osb, handle);
3980
3981out:
3982 if (orig_dx_leaves || new_dx_leaves) {
3983 for (i = 0; i < num_dx_leaves; i++) {
3984 if (orig_dx_leaves)
3985 brelse(orig_dx_leaves[i]);
3986 if (new_dx_leaves)
3987 brelse(new_dx_leaves[i]);
3988 }
3989 kfree(orig_dx_leaves);
3990 kfree(new_dx_leaves);
3991 }
3992
3993 if (meta_ac)
3994 ocfs2_free_alloc_context(meta_ac);
3995 if (data_ac)
3996 ocfs2_free_alloc_context(data_ac);
3997
3998 kfree(tmp_dx_leaf);
3999 return ret;
4000}
4001
Mark Fashehe7c17e42009-01-29 18:17:46 -08004002static int ocfs2_find_dir_space_dx(struct ocfs2_super *osb, struct inode *dir,
4003 struct buffer_head *di_bh,
4004 struct buffer_head *dx_root_bh,
4005 const char *name, int namelen,
4006 struct ocfs2_dir_lookup_result *lookup)
4007{
4008 int ret, rebalanced = 0;
4009 struct ocfs2_dx_root_block *dx_root;
4010 struct buffer_head *dx_leaf_bh = NULL;
4011 struct ocfs2_dx_leaf *dx_leaf;
4012 u64 blkno;
4013 u32 leaf_cpos;
4014
4015 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
4016
4017restart_search:
4018 ret = ocfs2_dx_dir_lookup(dir, &dx_root->dr_list, &lookup->dl_hinfo,
4019 &leaf_cpos, &blkno);
4020 if (ret) {
4021 mlog_errno(ret);
4022 goto out;
4023 }
4024
4025 ret = ocfs2_read_dx_leaf(dir, blkno, &dx_leaf_bh);
4026 if (ret) {
4027 mlog_errno(ret);
4028 goto out;
4029 }
4030
4031 dx_leaf = (struct ocfs2_dx_leaf *)dx_leaf_bh->b_data;
4032
4033 if (le16_to_cpu(dx_leaf->dl_list.de_num_used) >=
4034 le16_to_cpu(dx_leaf->dl_list.de_count)) {
4035 if (rebalanced) {
4036 /*
4037 * Rebalancing should have provided us with
4038 * space in an appropriate leaf.
4039 *
4040 * XXX: Is this an abnormal condition then?
4041 * Should we print a message here?
4042 */
4043 ret = -ENOSPC;
4044 goto out;
4045 }
4046
4047 ret = ocfs2_dx_dir_rebalance(osb, dir, dx_root_bh, dx_leaf_bh,
4048 &lookup->dl_hinfo, leaf_cpos,
4049 blkno);
4050 if (ret) {
4051 if (ret != -ENOSPC)
4052 mlog_errno(ret);
4053 goto out;
4054 }
4055
4056 /*
4057 * Restart the lookup. The rebalance might have
4058 * changed which block our item fits into. Mark our
4059 * progress, so we only execute this once.
4060 */
4061 brelse(dx_leaf_bh);
4062 dx_leaf_bh = NULL;
4063 rebalanced = 1;
4064 goto restart_search;
4065 }
4066
4067 lookup->dl_dx_leaf_bh = dx_leaf_bh;
4068 dx_leaf_bh = NULL;
4069
4070out:
4071 brelse(dx_leaf_bh);
4072 return ret;
4073}
4074
4075static int ocfs2_search_dx_free_list(struct inode *dir,
4076 struct buffer_head *dx_root_bh,
4077 int namelen,
4078 struct ocfs2_dir_lookup_result *lookup)
4079{
4080 int ret = -ENOSPC;
4081 struct buffer_head *leaf_bh = NULL, *prev_leaf_bh = NULL;
4082 struct ocfs2_dir_block_trailer *db;
4083 u64 next_block;
4084 int rec_len = OCFS2_DIR_REC_LEN(namelen);
4085 struct ocfs2_dx_root_block *dx_root;
4086
4087 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
4088 next_block = le64_to_cpu(dx_root->dr_free_blk);
4089
4090 while (next_block) {
4091 brelse(prev_leaf_bh);
4092 prev_leaf_bh = leaf_bh;
4093 leaf_bh = NULL;
4094
4095 ret = ocfs2_read_dir_block_direct(dir, next_block, &leaf_bh);
4096 if (ret) {
4097 mlog_errno(ret);
4098 goto out;
4099 }
4100
4101 db = ocfs2_trailer_from_bh(leaf_bh, dir->i_sb);
4102 if (rec_len <= le16_to_cpu(db->db_free_rec_len)) {
4103 lookup->dl_leaf_bh = leaf_bh;
4104 lookup->dl_prev_leaf_bh = prev_leaf_bh;
4105 leaf_bh = NULL;
4106 prev_leaf_bh = NULL;
4107 break;
4108 }
4109
4110 next_block = le64_to_cpu(db->db_free_next);
4111 }
4112
4113 if (!next_block)
4114 ret = -ENOSPC;
4115
4116out:
4117
4118 brelse(leaf_bh);
4119 brelse(prev_leaf_bh);
4120 return ret;
4121}
4122
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08004123static int ocfs2_expand_inline_dx_root(struct inode *dir,
4124 struct buffer_head *dx_root_bh)
4125{
4126 int ret, num_dx_leaves, i, j, did_quota = 0;
4127 struct buffer_head **dx_leaves = NULL;
4128 struct ocfs2_extent_tree et;
4129 u64 insert_blkno;
4130 struct ocfs2_alloc_context *data_ac = NULL;
4131 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
4132 handle_t *handle = NULL;
4133 struct ocfs2_dx_root_block *dx_root;
4134 struct ocfs2_dx_entry_list *entry_list;
4135 struct ocfs2_dx_entry *dx_entry;
4136 struct ocfs2_dx_leaf *target_leaf;
4137
4138 ret = ocfs2_reserve_clusters(osb, 1, &data_ac);
4139 if (ret) {
4140 mlog_errno(ret);
4141 goto out;
4142 }
4143
4144 dx_leaves = ocfs2_dx_dir_kmalloc_leaves(osb->sb, &num_dx_leaves);
4145 if (!dx_leaves) {
4146 ret = -ENOMEM;
4147 mlog_errno(ret);
4148 goto out;
4149 }
4150
4151 handle = ocfs2_start_trans(osb, ocfs2_calc_dxi_expand_credits(osb->sb));
4152 if (IS_ERR(handle)) {
4153 ret = PTR_ERR(handle);
4154 mlog_errno(ret);
4155 goto out;
4156 }
4157
4158 if (vfs_dq_alloc_space_nodirty(dir,
4159 ocfs2_clusters_to_bytes(osb->sb, 1))) {
4160 ret = -EDQUOT;
4161 goto out_commit;
4162 }
4163 did_quota = 1;
4164
4165 /*
4166 * We do this up front, before the allocation, so that a
4167 * failure to add the dx_root_bh to the journal won't result
4168 * us losing clusters.
4169 */
4170 ret = ocfs2_journal_access_dr(handle, dir, dx_root_bh,
4171 OCFS2_JOURNAL_ACCESS_WRITE);
4172 if (ret) {
4173 mlog_errno(ret);
4174 goto out_commit;
4175 }
4176
4177 ret = __ocfs2_dx_dir_new_cluster(dir, 0, handle, data_ac, dx_leaves,
4178 num_dx_leaves, &insert_blkno);
4179 if (ret) {
4180 mlog_errno(ret);
4181 goto out_commit;
4182 }
4183
4184 /*
4185 * Transfer the entries from our dx_root into the appropriate
4186 * block
4187 */
4188 dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data;
4189 entry_list = &dx_root->dr_entries;
4190
4191 for (i = 0; i < le16_to_cpu(entry_list->de_num_used); i++) {
4192 dx_entry = &entry_list->de_entries[i];
4193
4194 j = __ocfs2_dx_dir_hash_idx(osb,
4195 le32_to_cpu(dx_entry->dx_minor_hash));
4196 target_leaf = (struct ocfs2_dx_leaf *)dx_leaves[j]->b_data;
4197
4198 ocfs2_dx_dir_leaf_insert_tail(target_leaf, dx_entry);
4199
4200 /* Each leaf has been passed to the journal already
4201 * via __ocfs2_dx_dir_new_cluster() */
4202 }
4203
4204 dx_root->dr_flags &= ~OCFS2_DX_FLAG_INLINE;
4205 memset(&dx_root->dr_list, 0, osb->sb->s_blocksize -
4206 offsetof(struct ocfs2_dx_root_block, dr_list));
4207 dx_root->dr_list.l_count =
4208 cpu_to_le16(ocfs2_extent_recs_per_dx_root(osb->sb));
4209
4210 /* This should never fail considering we start with an empty
4211 * dx_root. */
4212 ocfs2_init_dx_root_extent_tree(&et, dir, dx_root_bh);
4213 ret = ocfs2_insert_extent(osb, handle, dir, &et, 0,
4214 insert_blkno, 1, 0, NULL);
4215 if (ret)
4216 mlog_errno(ret);
4217 did_quota = 0;
4218
4219 ocfs2_journal_dirty(handle, dx_root_bh);
4220
4221out_commit:
4222 if (ret < 0 && did_quota)
4223 vfs_dq_free_space_nodirty(dir,
4224 ocfs2_clusters_to_bytes(dir->i_sb, 1));
4225
4226 ocfs2_commit_trans(osb, handle);
4227
4228out:
4229 if (data_ac)
4230 ocfs2_free_alloc_context(data_ac);
4231
4232 if (dx_leaves) {
4233 for (i = 0; i < num_dx_leaves; i++)
4234 brelse(dx_leaves[i]);
4235 kfree(dx_leaves);
4236 }
4237 return ret;
4238}
4239
4240static int ocfs2_inline_dx_has_space(struct buffer_head *dx_root_bh)
4241{
4242 struct ocfs2_dx_root_block *dx_root;
4243 struct ocfs2_dx_entry_list *entry_list;
4244
4245 dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data;
4246 entry_list = &dx_root->dr_entries;
4247
4248 if (le16_to_cpu(entry_list->de_num_used) >=
4249 le16_to_cpu(entry_list->de_count))
4250 return -ENOSPC;
4251
4252 return 0;
4253}
4254
Mark Fashehe7c17e42009-01-29 18:17:46 -08004255static int ocfs2_prepare_dx_dir_for_insert(struct inode *dir,
4256 struct buffer_head *di_bh,
4257 const char *name,
4258 int namelen,
4259 struct ocfs2_dir_lookup_result *lookup)
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004260{
Mark Fashehe7c17e42009-01-29 18:17:46 -08004261 int ret, free_dx_root = 1;
4262 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004263 struct buffer_head *dx_root_bh = NULL;
Mark Fashehe7c17e42009-01-29 18:17:46 -08004264 struct buffer_head *leaf_bh = NULL;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004265 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
Mark Fashehe7c17e42009-01-29 18:17:46 -08004266 struct ocfs2_dx_root_block *dx_root;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004267
4268 ret = ocfs2_read_dx_root(dir, di, &dx_root_bh);
4269 if (ret) {
4270 mlog_errno(ret);
4271 goto out;
4272 }
4273
4274 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
Mark Fashehe3a93c22009-02-17 15:29:35 -08004275 if (le32_to_cpu(dx_root->dr_num_entries) == OCFS2_DX_ENTRIES_MAX) {
4276 ret = -ENOSPC;
4277 mlog_errno(ret);
4278 goto out;
4279 }
4280
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08004281 if (ocfs2_dx_root_inline(dx_root)) {
4282 ret = ocfs2_inline_dx_has_space(dx_root_bh);
4283
4284 if (ret == 0)
4285 goto search_el;
4286
4287 /*
4288 * We ran out of room in the root block. Expand it to
4289 * an extent, then allow ocfs2_find_dir_space_dx to do
4290 * the rest.
4291 */
4292 ret = ocfs2_expand_inline_dx_root(dir, dx_root_bh);
4293 if (ret) {
4294 mlog_errno(ret);
4295 goto out;
4296 }
4297 }
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004298
Mark Fashehe7c17e42009-01-29 18:17:46 -08004299 /*
4300 * Insert preparation for an indexed directory is split into two
4301 * steps. The call to find_dir_space_dx reserves room in the index for
4302 * an additional item. If we run out of space there, it's a real error
4303 * we can't continue on.
4304 */
4305 ret = ocfs2_find_dir_space_dx(osb, dir, di_bh, dx_root_bh, name,
4306 namelen, lookup);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004307 if (ret) {
4308 mlog_errno(ret);
4309 goto out;
4310 }
4311
Mark Fashehe7c17e42009-01-29 18:17:46 -08004312search_el:
4313 /*
4314 * Next, we need to find space in the unindexed tree. This call
4315 * searches using the free space linked list. If the unindexed tree
4316 * lacks sufficient space, we'll expand it below. The expansion code
4317 * is smart enough to add any new blocks to the free space list.
4318 */
4319 ret = ocfs2_search_dx_free_list(dir, dx_root_bh, namelen, lookup);
4320 if (ret && ret != -ENOSPC) {
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004321 mlog_errno(ret);
4322 goto out;
4323 }
4324
Mark Fashehe7c17e42009-01-29 18:17:46 -08004325 /* Do this up here - ocfs2_extend_dir might need the dx_root */
4326 lookup->dl_dx_root_bh = dx_root_bh;
4327 free_dx_root = 0;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004328
Mark Fashehe7c17e42009-01-29 18:17:46 -08004329 if (ret == -ENOSPC) {
4330 ret = ocfs2_extend_dir(osb, dir, di_bh, 1, lookup, &leaf_bh);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004331
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004332 if (ret) {
Mark Fashehe7c17e42009-01-29 18:17:46 -08004333 mlog_errno(ret);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004334 goto out;
4335 }
4336
4337 /*
Mark Fashehe7c17e42009-01-29 18:17:46 -08004338 * We make the assumption here that new leaf blocks are added
4339 * to the front of our free list.
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004340 */
Mark Fashehe7c17e42009-01-29 18:17:46 -08004341 lookup->dl_prev_leaf_bh = NULL;
4342 lookup->dl_leaf_bh = leaf_bh;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004343 }
4344
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004345out:
Mark Fashehe7c17e42009-01-29 18:17:46 -08004346 if (free_dx_root)
4347 brelse(dx_root_bh);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004348 return ret;
4349}
4350
Mark Fasheh4a12ca32008-11-12 15:43:34 -08004351/*
4352 * Get a directory ready for insert. Any directory allocation required
4353 * happens here. Success returns zero, and enough context in the dir
4354 * lookup result that ocfs2_add_entry() will be able complete the task
4355 * with minimal performance impact.
4356 */
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07004357int ocfs2_prepare_dir_for_insert(struct ocfs2_super *osb,
4358 struct inode *dir,
4359 struct buffer_head *parent_fe_bh,
4360 const char *name,
4361 int namelen,
Mark Fasheh4a12ca32008-11-12 15:43:34 -08004362 struct ocfs2_dir_lookup_result *lookup)
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07004363{
4364 int ret;
4365 unsigned int blocks_wanted = 1;
4366 struct buffer_head *bh = NULL;
4367
4368 mlog(0, "getting ready to insert namelen %d into dir %llu\n",
4369 namelen, (unsigned long long)OCFS2_I(dir)->ip_blkno);
4370
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07004371 if (!namelen) {
4372 ret = -EINVAL;
4373 mlog_errno(ret);
4374 goto out;
4375 }
4376
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004377 /*
4378 * Do this up front to reduce confusion.
4379 *
4380 * The directory might start inline, then be turned into an
4381 * indexed one, in which case we'd need to hash deep inside
4382 * ocfs2_find_dir_space_id(). Since
4383 * ocfs2_prepare_dx_dir_for_insert() also needs this hash
4384 * done, there seems no point in spreading out the calls. We
4385 * can optimize away the case where the file system doesn't
4386 * support indexing.
4387 */
4388 if (ocfs2_supports_indexed_dirs(osb))
4389 ocfs2_dx_dir_name_hash(dir, name, namelen, &lookup->dl_hinfo);
4390
4391 if (ocfs2_dir_indexed(dir)) {
Mark Fashehe7c17e42009-01-29 18:17:46 -08004392 ret = ocfs2_prepare_dx_dir_for_insert(dir, parent_fe_bh,
4393 name, namelen, lookup);
4394 if (ret)
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004395 mlog_errno(ret);
Mark Fashehe7c17e42009-01-29 18:17:46 -08004396 goto out;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004397 }
4398
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07004399 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
4400 ret = ocfs2_find_dir_space_id(dir, parent_fe_bh, name,
4401 namelen, &bh, &blocks_wanted);
4402 } else
4403 ret = ocfs2_find_dir_space_el(dir, name, namelen, &bh);
4404
4405 if (ret && ret != -ENOSPC) {
4406 mlog_errno(ret);
4407 goto out;
4408 }
4409
4410 if (ret == -ENOSPC) {
4411 /*
4412 * We have to expand the directory to add this name.
4413 */
4414 BUG_ON(bh);
4415
4416 ret = ocfs2_extend_dir(osb, dir, parent_fe_bh, blocks_wanted,
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004417 lookup, &bh);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07004418 if (ret) {
4419 if (ret != -ENOSPC)
4420 mlog_errno(ret);
4421 goto out;
4422 }
4423
4424 BUG_ON(!bh);
4425 }
4426
Mark Fasheh4a12ca32008-11-12 15:43:34 -08004427 lookup->dl_leaf_bh = bh;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07004428 bh = NULL;
4429out:
Mark Fasheha81cb882008-10-07 14:25:16 -07004430 brelse(bh);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07004431 return ret;
4432}
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004433
4434static int ocfs2_dx_dir_remove_index(struct inode *dir,
4435 struct buffer_head *di_bh,
4436 struct buffer_head *dx_root_bh)
4437{
4438 int ret;
4439 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
4440 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
4441 struct ocfs2_dx_root_block *dx_root;
4442 struct inode *dx_alloc_inode = NULL;
4443 struct buffer_head *dx_alloc_bh = NULL;
4444 handle_t *handle;
4445 u64 blk;
4446 u16 bit;
4447 u64 bg_blkno;
4448
4449 dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data;
4450
4451 dx_alloc_inode = ocfs2_get_system_file_inode(osb,
4452 EXTENT_ALLOC_SYSTEM_INODE,
4453 le16_to_cpu(dx_root->dr_suballoc_slot));
4454 if (!dx_alloc_inode) {
4455 ret = -ENOMEM;
4456 mlog_errno(ret);
4457 goto out;
4458 }
4459 mutex_lock(&dx_alloc_inode->i_mutex);
4460
4461 ret = ocfs2_inode_lock(dx_alloc_inode, &dx_alloc_bh, 1);
4462 if (ret) {
4463 mlog_errno(ret);
4464 goto out_mutex;
4465 }
4466
4467 handle = ocfs2_start_trans(osb, OCFS2_DX_ROOT_REMOVE_CREDITS);
4468 if (IS_ERR(handle)) {
4469 ret = PTR_ERR(handle);
4470 mlog_errno(ret);
4471 goto out_unlock;
4472 }
4473
4474 ret = ocfs2_journal_access_di(handle, dir, di_bh,
4475 OCFS2_JOURNAL_ACCESS_WRITE);
4476 if (ret) {
4477 mlog_errno(ret);
4478 goto out_commit;
4479 }
4480
4481 OCFS2_I(dir)->ip_dyn_features &= ~OCFS2_INDEXED_DIR_FL;
4482 di->i_dyn_features = cpu_to_le16(OCFS2_I(dir)->ip_dyn_features);
4483 di->i_dx_root = cpu_to_le64(0ULL);
4484
4485 ocfs2_journal_dirty(handle, di_bh);
4486
4487 blk = le64_to_cpu(dx_root->dr_blkno);
4488 bit = le16_to_cpu(dx_root->dr_suballoc_bit);
4489 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
4490 ret = ocfs2_free_suballoc_bits(handle, dx_alloc_inode, dx_alloc_bh,
4491 bit, bg_blkno, 1);
4492 if (ret)
4493 mlog_errno(ret);
4494
4495out_commit:
4496 ocfs2_commit_trans(osb, handle);
4497
4498out_unlock:
4499 ocfs2_inode_unlock(dx_alloc_inode, 1);
4500
4501out_mutex:
4502 mutex_unlock(&dx_alloc_inode->i_mutex);
4503 brelse(dx_alloc_bh);
4504out:
4505 iput(dx_alloc_inode);
4506 return ret;
4507}
4508
4509int ocfs2_dx_dir_truncate(struct inode *dir, struct buffer_head *di_bh)
4510{
4511 int ret;
4512 unsigned int uninitialized_var(clen);
4513 u32 major_hash = UINT_MAX, p_cpos, uninitialized_var(cpos);
4514 u64 uninitialized_var(blkno);
4515 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
4516 struct buffer_head *dx_root_bh = NULL;
4517 struct ocfs2_dx_root_block *dx_root;
4518 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
4519 struct ocfs2_cached_dealloc_ctxt dealloc;
4520 struct ocfs2_extent_tree et;
4521
4522 ocfs2_init_dealloc_ctxt(&dealloc);
4523
4524 if (!ocfs2_dir_indexed(dir))
4525 return 0;
4526
4527 ret = ocfs2_read_dx_root(dir, di, &dx_root_bh);
4528 if (ret) {
4529 mlog_errno(ret);
4530 goto out;
4531 }
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08004532 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
4533
4534 if (ocfs2_dx_root_inline(dx_root))
4535 goto remove_index;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004536
4537 ocfs2_init_dx_root_extent_tree(&et, dir, dx_root_bh);
4538
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004539 /* XXX: What if dr_clusters is too large? */
4540 while (le32_to_cpu(dx_root->dr_clusters)) {
4541 ret = ocfs2_dx_dir_lookup_rec(dir, &dx_root->dr_list,
4542 major_hash, &cpos, &blkno, &clen);
4543 if (ret) {
4544 mlog_errno(ret);
4545 goto out;
4546 }
4547
4548 p_cpos = ocfs2_blocks_to_clusters(dir->i_sb, blkno);
4549
4550 ret = ocfs2_remove_btree_range(dir, &et, cpos, p_cpos, clen,
4551 &dealloc);
4552 if (ret) {
4553 mlog_errno(ret);
4554 goto out;
4555 }
4556
4557 if (cpos == 0)
4558 break;
4559
4560 major_hash = cpos - 1;
4561 }
4562
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08004563remove_index:
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004564 ret = ocfs2_dx_dir_remove_index(dir, di_bh, dx_root_bh);
4565 if (ret) {
4566 mlog_errno(ret);
4567 goto out;
4568 }
4569
4570 ocfs2_remove_from_cache(dir, dx_root_bh);
4571out:
4572 ocfs2_schedule_truncate_log_flush(osb, 1);
4573 ocfs2_run_deallocs(osb, &dealloc);
4574
4575 brelse(dx_root_bh);
4576 return ret;
4577}