blob: 8582e3f4f120647d05df81556639e627882b5811 [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
Mark Fashehccd979b2005-12-15 14:31:24 -080046#include <cluster/masklog.h>
47
48#include "ocfs2.h"
49
50#include "alloc.h"
Joel Beckerc175a512008-12-10 17:58:22 -080051#include "blockcheck.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080052#include "dir.h"
53#include "dlmglue.h"
54#include "extent_map.h"
55#include "file.h"
56#include "inode.h"
57#include "journal.h"
58#include "namei.h"
59#include "suballoc.h"
Mark Fasheh316f4b92007-09-07 18:21:26 -070060#include "super.h"
Mark Fasheh9b7895e2008-11-12 16:27:44 -080061#include "sysfile.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080062#include "uptodate.h"
Tao Maf1088d42011-02-23 22:30:23 +080063#include "ocfs2_trace.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080064
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
Joel Becker0cf2f762009-02-12 16:41:25 -0800179 ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
Mark Fashehe7c17e42009-01-29 18:17:46 -0800180 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
Tao Ma1dd9ffc2011-01-24 23:23:30 +0800325 if (unlikely(rlen < OCFS2_DIR_REC_LEN(1)))
Mark Fasheh316f4b92007-09-07 18:21:26 -0700326 error_msg = "rec_len is smaller than minimal";
Tao Ma1dd9ffc2011-01-24 23:23:30 +0800327 else if (unlikely(rlen % 4 != 0))
Mark Fasheh316f4b92007-09-07 18:21:26 -0700328 error_msg = "rec_len % 4 != 0";
Tao Ma1dd9ffc2011-01-24 23:23:30 +0800329 else if (unlikely(rlen < OCFS2_DIR_REC_LEN(de->name_len)))
Mark Fasheh316f4b92007-09-07 18:21:26 -0700330 error_msg = "rec_len is too small for name_len";
Tao Ma1dd9ffc2011-01-24 23:23:30 +0800331 else if (unlikely(
332 ((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize))
Mark Fasheh316f4b92007-09-07 18:21:26 -0700333 error_msg = "directory entry across blocks";
334
Tao Ma1dd9ffc2011-01-24 23:23:30 +0800335 if (unlikely(error_msg != NULL))
Mark Fasheh316f4b92007-09-07 18:21:26 -0700336 mlog(ML_ERROR, "bad entry in directory #%llu: %s - "
337 "offset=%lu, inode=%llu, rec_len=%d, name_len=%d\n",
338 (unsigned long long)OCFS2_I(dir)->ip_blkno, error_msg,
339 offset, (unsigned long long)le64_to_cpu(de->inode), rlen,
340 de->name_len);
Tao Ma1dd9ffc2011-01-24 23:23:30 +0800341
Mark Fasheh316f4b92007-09-07 18:21:26 -0700342 return error_msg == NULL ? 1 : 0;
343}
344
345static inline int ocfs2_match(int len,
346 const char * const name,
347 struct ocfs2_dir_entry *de)
348{
349 if (len != de->name_len)
350 return 0;
351 if (!de->inode)
352 return 0;
353 return !memcmp(name, de->name, len);
354}
355
356/*
357 * Returns 0 if not found, -1 on failure, and 1 on success
358 */
Jesper Juhl42b16b32011-01-17 00:09:38 +0100359static inline int ocfs2_search_dirblock(struct buffer_head *bh,
Mark Fasheh316f4b92007-09-07 18:21:26 -0700360 struct inode *dir,
361 const char *name, int namelen,
362 unsigned long offset,
Mark Fasheh23193e52007-09-12 13:01:18 -0700363 char *first_de,
364 unsigned int bytes,
Mark Fasheh316f4b92007-09-07 18:21:26 -0700365 struct ocfs2_dir_entry **res_dir)
366{
367 struct ocfs2_dir_entry *de;
368 char *dlimit, *de_buf;
369 int de_len;
370 int ret = 0;
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:
Tao Maf1088d42011-02-23 22:30:23 +0800405 trace_ocfs2_search_dirblock(ret);
Mark Fasheh316f4b92007-09-07 18:21:26 -0700406 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 */
Tao Maf1088d42011-02-23 22:30:23 +0800450 trace_ocfs2_validate_dir_block((unsigned long long)bh->b_blocknr);
Joel Beckera22305c2008-11-13 14:49:17 -0800451
Joel Beckerc175a512008-12-10 17:58:22 -0800452 BUG_ON(!buffer_uptodate(bh));
453
454 /*
455 * If the ecc fails, we return the error but otherwise
456 * leave the filesystem running. We know any error is
457 * local to this block.
458 *
459 * Note that we are safe to call this even if the directory
460 * doesn't have a trailer. Filesystems without metaecc will do
461 * nothing, and filesystems with it will have one.
462 */
463 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &trailer->db_check);
464 if (rc)
465 mlog(ML_ERROR, "Checksum failed for dinode %llu\n",
466 (unsigned long long)bh->b_blocknr);
467
468 return rc;
Joel Beckera22305c2008-11-13 14:49:17 -0800469}
470
471/*
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800472 * Validate a directory trailer.
473 *
474 * We check the trailer here rather than in ocfs2_validate_dir_block()
475 * because that function doesn't have the inode to test.
476 */
477static int ocfs2_check_dir_trailer(struct inode *dir, struct buffer_head *bh)
478{
479 int rc = 0;
480 struct ocfs2_dir_block_trailer *trailer;
481
482 trailer = ocfs2_trailer_from_bh(bh, dir->i_sb);
483 if (!OCFS2_IS_VALID_DIR_TRAILER(trailer)) {
484 rc = -EINVAL;
485 ocfs2_error(dir->i_sb,
486 "Invalid dirblock #%llu: "
487 "signature = %.*s\n",
488 (unsigned long long)bh->b_blocknr, 7,
489 trailer->db_signature);
490 goto out;
491 }
492 if (le64_to_cpu(trailer->db_blkno) != bh->b_blocknr) {
493 rc = -EINVAL;
494 ocfs2_error(dir->i_sb,
495 "Directory block #%llu has an invalid "
496 "db_blkno of %llu",
497 (unsigned long long)bh->b_blocknr,
498 (unsigned long long)le64_to_cpu(trailer->db_blkno));
499 goto out;
500 }
501 if (le64_to_cpu(trailer->db_parent_dinode) !=
502 OCFS2_I(dir)->ip_blkno) {
503 rc = -EINVAL;
504 ocfs2_error(dir->i_sb,
505 "Directory block #%llu on dinode "
506 "#%llu has an invalid parent_dinode "
507 "of %llu",
508 (unsigned long long)bh->b_blocknr,
509 (unsigned long long)OCFS2_I(dir)->ip_blkno,
510 (unsigned long long)le64_to_cpu(trailer->db_blkno));
511 goto out;
512 }
513out:
514 return rc;
515}
516
517/*
Joel Beckera22305c2008-11-13 14:49:17 -0800518 * This function forces all errors to -EIO for consistency with its
519 * predecessor, ocfs2_bread(). We haven't audited what returning the
520 * real error codes would do to callers. We log the real codes with
521 * mlog_errno() before we squash them.
522 */
523static int ocfs2_read_dir_block(struct inode *inode, u64 v_block,
524 struct buffer_head **bh, int flags)
525{
526 int rc = 0;
527 struct buffer_head *tmp = *bh;
Joel Beckera22305c2008-11-13 14:49:17 -0800528
Joel Becker511308d2008-11-13 14:49:21 -0800529 rc = ocfs2_read_virt_blocks(inode, v_block, 1, &tmp, flags,
530 ocfs2_validate_dir_block);
Mark Fasheh87d35a72008-12-10 17:36:25 -0800531 if (rc) {
Joel Beckera22305c2008-11-13 14:49:17 -0800532 mlog_errno(rc);
Mark Fasheh87d35a72008-12-10 17:36:25 -0800533 goto out;
534 }
535
Mark Fasheh87d35a72008-12-10 17:36:25 -0800536 if (!(flags & OCFS2_BH_READAHEAD) &&
Mark Fashehe7c17e42009-01-29 18:17:46 -0800537 ocfs2_supports_dir_trailer(inode)) {
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800538 rc = ocfs2_check_dir_trailer(inode, tmp);
539 if (rc) {
540 if (!*bh)
541 brelse(tmp);
542 mlog_errno(rc);
Mark Fasheh87d35a72008-12-10 17:36:25 -0800543 goto out;
544 }
545 }
Joel Beckera22305c2008-11-13 14:49:17 -0800546
Joel Becker511308d2008-11-13 14:49:21 -0800547 /* If ocfs2_read_virt_blocks() got us a new bh, pass it up. */
Mark Fasheh87d35a72008-12-10 17:36:25 -0800548 if (!*bh)
Joel Beckera22305c2008-11-13 14:49:17 -0800549 *bh = tmp;
550
Mark Fasheh87d35a72008-12-10 17:36:25 -0800551out:
Joel Beckera22305c2008-11-13 14:49:17 -0800552 return rc ? -EIO : 0;
553}
554
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800555/*
556 * Read the block at 'phys' which belongs to this directory
557 * inode. This function does no virtual->physical block translation -
558 * what's passed in is assumed to be a valid directory block.
559 */
560static int ocfs2_read_dir_block_direct(struct inode *dir, u64 phys,
561 struct buffer_head **bh)
562{
563 int ret;
564 struct buffer_head *tmp = *bh;
565
Joel Becker8cb471e2009-02-10 20:00:41 -0800566 ret = ocfs2_read_block(INODE_CACHE(dir), phys, &tmp,
567 ocfs2_validate_dir_block);
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800568 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
Joel Becker8cb471e2009-02-10 20:00:41 -0800625 ret = ocfs2_read_block(INODE_CACHE(dir), blkno, &tmp,
626 ocfs2_validate_dx_root);
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800627
628 /* If ocfs2_read_block() got us a new bh, pass it up. */
629 if (!ret && !*dx_root_bh)
630 *dx_root_bh = tmp;
631
632 return ret;
633}
634
635static int ocfs2_validate_dx_leaf(struct super_block *sb,
636 struct buffer_head *bh)
637{
638 int ret;
639 struct ocfs2_dx_leaf *dx_leaf = (struct ocfs2_dx_leaf *)bh->b_data;
640
641 BUG_ON(!buffer_uptodate(bh));
642
643 ret = ocfs2_validate_meta_ecc(sb, bh->b_data, &dx_leaf->dl_check);
644 if (ret) {
645 mlog(ML_ERROR,
646 "Checksum failed for dir index leaf block %llu\n",
647 (unsigned long long)bh->b_blocknr);
648 return ret;
649 }
650
651 if (!OCFS2_IS_VALID_DX_LEAF(dx_leaf)) {
652 ocfs2_error(sb, "Dir Index Leaf has bad signature %.*s",
653 7, dx_leaf->dl_signature);
654 return -EROFS;
655 }
656
657 return 0;
658}
659
660static int ocfs2_read_dx_leaf(struct inode *dir, u64 blkno,
661 struct buffer_head **dx_leaf_bh)
662{
663 int ret;
664 struct buffer_head *tmp = *dx_leaf_bh;
665
Joel Becker8cb471e2009-02-10 20:00:41 -0800666 ret = ocfs2_read_block(INODE_CACHE(dir), blkno, &tmp,
667 ocfs2_validate_dx_leaf);
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800668
669 /* If ocfs2_read_block() got us a new bh, pass it up. */
670 if (!ret && !*dx_leaf_bh)
671 *dx_leaf_bh = tmp;
672
673 return ret;
674}
675
676/*
677 * Read a series of dx_leaf blocks. This expects all buffer_head
678 * pointers to be NULL on function entry.
679 */
680static int ocfs2_read_dx_leaves(struct inode *dir, u64 start, int num,
681 struct buffer_head **dx_leaf_bhs)
682{
683 int ret;
684
Joel Becker8cb471e2009-02-10 20:00:41 -0800685 ret = ocfs2_read_blocks(INODE_CACHE(dir), start, num, dx_leaf_bhs, 0,
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800686 ocfs2_validate_dx_leaf);
687 if (ret)
688 mlog_errno(ret);
689
690 return ret;
691}
692
Adrian Bunk0af4bd32007-10-24 18:23:27 +0200693static struct buffer_head *ocfs2_find_entry_el(const char *name, int namelen,
694 struct inode *dir,
695 struct ocfs2_dir_entry **res_dir)
Mark Fasheh316f4b92007-09-07 18:21:26 -0700696{
697 struct super_block *sb;
698 struct buffer_head *bh_use[NAMEI_RA_SIZE];
699 struct buffer_head *bh, *ret = NULL;
700 unsigned long start, block, b;
701 int ra_max = 0; /* Number of bh's in the readahead
702 buffer, bh_use[] */
703 int ra_ptr = 0; /* Current index into readahead
704 buffer */
705 int num = 0;
706 int nblocks, i, err;
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
Tao Maf1088d42011-02-23 22:30:23 +0800788 trace_ocfs2_find_entry_el(ret);
Mark Fasheh316f4b92007-09-07 18:21:26 -0700789 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) {
Joel Beckerfacdb772009-02-12 18:08:48 -0800805 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, major_hash,
806 &eb_bh);
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800807 if (ret) {
808 mlog_errno(ret);
809 goto out;
810 }
811
812 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
813 el = &eb->h_list;
814
815 if (el->l_tree_depth) {
816 ocfs2_error(inode->i_sb,
817 "Inode %lu has non zero tree depth in "
818 "btree tree block %llu\n", inode->i_ino,
819 (unsigned long long)eb_bh->b_blocknr);
820 ret = -EROFS;
821 goto out;
822 }
823 }
824
825 found = 0;
826 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
827 rec = &el->l_recs[i];
828
829 if (le32_to_cpu(rec->e_cpos) <= major_hash) {
830 found = 1;
831 break;
832 }
833 }
834
835 if (!found) {
836 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
837 "record (%u, %u, 0) in btree", inode->i_ino,
838 le32_to_cpu(rec->e_cpos),
839 ocfs2_rec_clusters(el, rec));
840 ret = -EROFS;
841 goto out;
842 }
843
844 if (ret_phys_blkno)
845 *ret_phys_blkno = le64_to_cpu(rec->e_blkno);
846 if (ret_cpos)
847 *ret_cpos = le32_to_cpu(rec->e_cpos);
848 if (ret_clen)
849 *ret_clen = le16_to_cpu(rec->e_leaf_clusters);
850
851out:
852 brelse(eb_bh);
853 return ret;
854}
855
856/*
857 * Returns the block index, from the start of the cluster which this
858 * hash belongs too.
859 */
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -0800860static inline unsigned int __ocfs2_dx_dir_hash_idx(struct ocfs2_super *osb,
861 u32 minor_hash)
862{
863 return minor_hash & osb->osb_dx_mask;
864}
865
866static inline unsigned int ocfs2_dx_dir_hash_idx(struct ocfs2_super *osb,
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800867 struct ocfs2_dx_hinfo *hinfo)
868{
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -0800869 return __ocfs2_dx_dir_hash_idx(osb, hinfo->minor_hash);
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800870}
871
872static int ocfs2_dx_dir_lookup(struct inode *inode,
873 struct ocfs2_extent_list *el,
874 struct ocfs2_dx_hinfo *hinfo,
875 u32 *ret_cpos,
876 u64 *ret_phys_blkno)
877{
878 int ret = 0;
879 unsigned int cend, uninitialized_var(clen);
880 u32 uninitialized_var(cpos);
881 u64 uninitialized_var(blkno);
882 u32 name_hash = hinfo->major_hash;
883
884 ret = ocfs2_dx_dir_lookup_rec(inode, el, name_hash, &cpos, &blkno,
885 &clen);
886 if (ret) {
887 mlog_errno(ret);
888 goto out;
889 }
890
891 cend = cpos + clen;
892 if (name_hash >= cend) {
893 /* We want the last cluster */
894 blkno += ocfs2_clusters_to_blocks(inode->i_sb, clen - 1);
895 cpos += clen - 1;
896 } else {
897 blkno += ocfs2_clusters_to_blocks(inode->i_sb,
898 name_hash - cpos);
899 cpos = name_hash;
900 }
901
902 /*
903 * We now have the cluster which should hold our entry. To
904 * find the exact block from the start of the cluster to
905 * search, we take the lower bits of the hash.
906 */
907 blkno += ocfs2_dx_dir_hash_idx(OCFS2_SB(inode->i_sb), hinfo);
908
909 if (ret_phys_blkno)
910 *ret_phys_blkno = blkno;
911 if (ret_cpos)
912 *ret_cpos = cpos;
913
914out:
915
916 return ret;
917}
918
919static int ocfs2_dx_dir_search(const char *name, int namelen,
920 struct inode *dir,
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -0800921 struct ocfs2_dx_root_block *dx_root,
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800922 struct ocfs2_dir_lookup_result *res)
923{
924 int ret, i, found;
925 u64 uninitialized_var(phys);
926 struct buffer_head *dx_leaf_bh = NULL;
927 struct ocfs2_dx_leaf *dx_leaf;
928 struct ocfs2_dx_entry *dx_entry = NULL;
929 struct buffer_head *dir_ent_bh = NULL;
930 struct ocfs2_dir_entry *dir_ent = NULL;
931 struct ocfs2_dx_hinfo *hinfo = &res->dl_hinfo;
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -0800932 struct ocfs2_extent_list *dr_el;
933 struct ocfs2_dx_entry_list *entry_list;
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800934
935 ocfs2_dx_dir_name_hash(dir, name, namelen, &res->dl_hinfo);
936
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -0800937 if (ocfs2_dx_root_inline(dx_root)) {
938 entry_list = &dx_root->dr_entries;
939 goto search;
940 }
941
942 dr_el = &dx_root->dr_list;
943
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800944 ret = ocfs2_dx_dir_lookup(dir, dr_el, hinfo, NULL, &phys);
945 if (ret) {
946 mlog_errno(ret);
947 goto out;
948 }
949
Tao Maf1088d42011-02-23 22:30:23 +0800950 trace_ocfs2_dx_dir_search((unsigned long long)OCFS2_I(dir)->ip_blkno,
951 namelen, name, hinfo->major_hash,
952 hinfo->minor_hash, (unsigned long long)phys);
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800953
954 ret = ocfs2_read_dx_leaf(dir, phys, &dx_leaf_bh);
955 if (ret) {
956 mlog_errno(ret);
957 goto out;
958 }
959
960 dx_leaf = (struct ocfs2_dx_leaf *) dx_leaf_bh->b_data;
961
Tao Maf1088d42011-02-23 22:30:23 +0800962 trace_ocfs2_dx_dir_search_leaf_info(
963 le16_to_cpu(dx_leaf->dl_list.de_num_used),
964 le16_to_cpu(dx_leaf->dl_list.de_count));
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800965
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -0800966 entry_list = &dx_leaf->dl_list;
967
968search:
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800969 /*
970 * Empty leaf is legal, so no need to check for that.
971 */
972 found = 0;
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -0800973 for (i = 0; i < le16_to_cpu(entry_list->de_num_used); i++) {
974 dx_entry = &entry_list->de_entries[i];
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800975
976 if (hinfo->major_hash != le32_to_cpu(dx_entry->dx_major_hash)
977 || hinfo->minor_hash != le32_to_cpu(dx_entry->dx_minor_hash))
978 continue;
979
980 /*
981 * Search unindexed leaf block now. We're not
982 * guaranteed to find anything.
983 */
984 ret = ocfs2_read_dir_block_direct(dir,
985 le64_to_cpu(dx_entry->dx_dirent_blk),
986 &dir_ent_bh);
987 if (ret) {
988 mlog_errno(ret);
989 goto out;
990 }
991
992 /*
993 * XXX: We should check the unindexed block here,
994 * before using it.
995 */
996
997 found = ocfs2_search_dirblock(dir_ent_bh, dir, name, namelen,
998 0, dir_ent_bh->b_data,
999 dir->i_sb->s_blocksize, &dir_ent);
1000 if (found == 1)
1001 break;
1002
1003 if (found == -1) {
1004 /* This means we found a bad directory entry. */
1005 ret = -EIO;
1006 mlog_errno(ret);
1007 goto out;
1008 }
1009
1010 brelse(dir_ent_bh);
1011 dir_ent_bh = NULL;
1012 }
1013
1014 if (found <= 0) {
1015 ret = -ENOENT;
1016 goto out;
1017 }
1018
1019 res->dl_leaf_bh = dir_ent_bh;
1020 res->dl_entry = dir_ent;
1021 res->dl_dx_leaf_bh = dx_leaf_bh;
1022 res->dl_dx_entry = dx_entry;
1023
1024 ret = 0;
1025out:
1026 if (ret) {
1027 brelse(dx_leaf_bh);
1028 brelse(dir_ent_bh);
1029 }
1030 return ret;
1031}
1032
1033static int ocfs2_find_entry_dx(const char *name, int namelen,
1034 struct inode *dir,
1035 struct ocfs2_dir_lookup_result *lookup)
1036{
1037 int ret;
1038 struct buffer_head *di_bh = NULL;
1039 struct ocfs2_dinode *di;
1040 struct buffer_head *dx_root_bh = NULL;
1041 struct ocfs2_dx_root_block *dx_root;
1042
1043 ret = ocfs2_read_inode_block(dir, &di_bh);
1044 if (ret) {
1045 mlog_errno(ret);
1046 goto out;
1047 }
1048
1049 di = (struct ocfs2_dinode *)di_bh->b_data;
1050
1051 ret = ocfs2_read_dx_root(dir, di, &dx_root_bh);
1052 if (ret) {
1053 mlog_errno(ret);
1054 goto out;
1055 }
1056 dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data;
1057
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001058 ret = ocfs2_dx_dir_search(name, namelen, dir, dx_root, lookup);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001059 if (ret) {
1060 if (ret != -ENOENT)
1061 mlog_errno(ret);
1062 goto out;
1063 }
1064
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001065 lookup->dl_dx_root_bh = dx_root_bh;
1066 dx_root_bh = NULL;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001067out:
1068 brelse(di_bh);
1069 brelse(dx_root_bh);
1070 return ret;
1071}
1072
Mark Fasheh23193e52007-09-12 13:01:18 -07001073/*
1074 * Try to find an entry of the provided name within 'dir'.
1075 *
Mark Fasheh4a12ca32008-11-12 15:43:34 -08001076 * If nothing was found, -ENOENT is returned. Otherwise, zero is
1077 * returned and the struct 'res' will contain information useful to
1078 * other directory manipulation functions.
Mark Fasheh23193e52007-09-12 13:01:18 -07001079 *
1080 * Caller can NOT assume anything about the contents of the
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001081 * buffer_heads - they are passed back only so that it can be passed
1082 * into any one of the manipulation functions (add entry, delete
1083 * entry, etc). As an example, bh in the extent directory case is a
1084 * data block, in the inline-data case it actually points to an inode,
1085 * in the indexed directory case, multiple buffers are involved.
Mark Fasheh23193e52007-09-12 13:01:18 -07001086 */
Mark Fasheh4a12ca32008-11-12 15:43:34 -08001087int ocfs2_find_entry(const char *name, int namelen,
1088 struct inode *dir, struct ocfs2_dir_lookup_result *lookup)
Mark Fasheh23193e52007-09-12 13:01:18 -07001089{
Mark Fasheh4a12ca32008-11-12 15:43:34 -08001090 struct buffer_head *bh;
1091 struct ocfs2_dir_entry *res_dir = NULL;
Mark Fasheh23193e52007-09-12 13:01:18 -07001092
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001093 if (ocfs2_dir_indexed(dir))
1094 return ocfs2_find_entry_dx(name, namelen, dir, lookup);
1095
1096 /*
1097 * The unindexed dir code only uses part of the lookup
1098 * structure, so there's no reason to push it down further
1099 * than this.
1100 */
Mark Fasheh23193e52007-09-12 13:01:18 -07001101 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
Mark Fasheh4a12ca32008-11-12 15:43:34 -08001102 bh = ocfs2_find_entry_id(name, namelen, dir, &res_dir);
1103 else
1104 bh = ocfs2_find_entry_el(name, namelen, dir, &res_dir);
Mark Fasheh23193e52007-09-12 13:01:18 -07001105
Mark Fasheh4a12ca32008-11-12 15:43:34 -08001106 if (bh == NULL)
1107 return -ENOENT;
1108
1109 lookup->dl_leaf_bh = bh;
1110 lookup->dl_entry = res_dir;
1111 return 0;
Mark Fasheh23193e52007-09-12 13:01:18 -07001112}
1113
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001114/*
1115 * Update inode number and type of a previously found directory entry.
1116 */
Mark Fasheh38760e22007-09-11 17:21:56 -07001117int ocfs2_update_entry(struct inode *dir, handle_t *handle,
Mark Fasheh4a12ca32008-11-12 15:43:34 -08001118 struct ocfs2_dir_lookup_result *res,
Mark Fasheh38760e22007-09-11 17:21:56 -07001119 struct inode *new_entry_inode)
1120{
1121 int ret;
Joel Becker13723d02008-10-17 19:25:01 -07001122 ocfs2_journal_access_func access = ocfs2_journal_access_db;
Mark Fasheh4a12ca32008-11-12 15:43:34 -08001123 struct ocfs2_dir_entry *de = res->dl_entry;
1124 struct buffer_head *de_bh = res->dl_leaf_bh;
Mark Fasheh38760e22007-09-11 17:21:56 -07001125
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001126 /*
1127 * The same code works fine for both inline-data and extent
Joel Becker13723d02008-10-17 19:25:01 -07001128 * based directories, so no need to split this up. The only
1129 * difference is the journal_access function.
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001130 */
1131
Joel Becker13723d02008-10-17 19:25:01 -07001132 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
1133 access = ocfs2_journal_access_di;
1134
Joel Becker0cf2f762009-02-12 16:41:25 -08001135 ret = access(handle, INODE_CACHE(dir), de_bh,
1136 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
Joel Becker13723d02008-10-17 19:25:01 -07001164 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
1165 access = ocfs2_journal_access_di;
1166
Mark Fasheh316f4b92007-09-07 18:21:26 -07001167 i = 0;
1168 pde = NULL;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001169 de = (struct ocfs2_dir_entry *) first_de;
1170 while (i < bytes) {
Mark Fasheh316f4b92007-09-07 18:21:26 -07001171 if (!ocfs2_check_dir_entry(dir, de, bh, i)) {
1172 status = -EIO;
1173 mlog_errno(status);
1174 goto bail;
1175 }
1176 if (de == de_del) {
Joel Becker0cf2f762009-02-12 16:41:25 -08001177 status = access(handle, INODE_CACHE(dir), bh,
Joel Becker13723d02008-10-17 19:25:01 -07001178 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh316f4b92007-09-07 18:21:26 -07001179 if (status < 0) {
1180 status = -EIO;
1181 mlog_errno(status);
1182 goto bail;
1183 }
1184 if (pde)
Marcin Slusarz0dd32562008-02-13 00:06:18 +01001185 le16_add_cpu(&pde->rec_len,
1186 le16_to_cpu(de->rec_len));
Mark Fasheh316f4b92007-09-07 18:21:26 -07001187 else
1188 de->inode = 0;
1189 dir->i_version++;
Joel Beckerec20cec2010-03-19 14:13:52 -07001190 ocfs2_journal_dirty(handle, bh);
Mark Fasheh316f4b92007-09-07 18:21:26 -07001191 goto bail;
1192 }
1193 i += le16_to_cpu(de->rec_len);
1194 pde = de;
1195 de = (struct ocfs2_dir_entry *)((char *)de + le16_to_cpu(de->rec_len));
1196 }
1197bail:
Mark Fasheh316f4b92007-09-07 18:21:26 -07001198 return status;
1199}
1200
Mark Fashehe7c17e42009-01-29 18:17:46 -08001201static unsigned int ocfs2_figure_dirent_hole(struct ocfs2_dir_entry *de)
1202{
1203 unsigned int hole;
1204
1205 if (le64_to_cpu(de->inode) == 0)
1206 hole = le16_to_cpu(de->rec_len);
1207 else
1208 hole = le16_to_cpu(de->rec_len) -
1209 OCFS2_DIR_REC_LEN(de->name_len);
1210
1211 return hole;
1212}
1213
1214static int ocfs2_find_max_rec_len(struct super_block *sb,
1215 struct buffer_head *dirblock_bh)
1216{
1217 int size, this_hole, largest_hole = 0;
1218 char *trailer, *de_buf, *limit, *start = dirblock_bh->b_data;
1219 struct ocfs2_dir_entry *de;
1220
1221 trailer = (char *)ocfs2_trailer_from_bh(dirblock_bh, sb);
1222 size = ocfs2_dir_trailer_blk_off(sb);
1223 limit = start + size;
1224 de_buf = start;
1225 de = (struct ocfs2_dir_entry *)de_buf;
1226 do {
1227 if (de_buf != trailer) {
1228 this_hole = ocfs2_figure_dirent_hole(de);
1229 if (this_hole > largest_hole)
1230 largest_hole = this_hole;
1231 }
1232
1233 de_buf += le16_to_cpu(de->rec_len);
1234 de = (struct ocfs2_dir_entry *)de_buf;
1235 } while (de_buf < limit);
1236
1237 if (largest_hole >= OCFS2_DIR_MIN_REC_LEN)
1238 return largest_hole;
1239 return 0;
1240}
1241
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001242static void ocfs2_dx_list_remove_entry(struct ocfs2_dx_entry_list *entry_list,
1243 int index)
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001244{
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001245 int num_used = le16_to_cpu(entry_list->de_num_used);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001246
1247 if (num_used == 1 || index == (num_used - 1))
1248 goto clear;
1249
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001250 memmove(&entry_list->de_entries[index],
1251 &entry_list->de_entries[index + 1],
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001252 (num_used - index - 1)*sizeof(struct ocfs2_dx_entry));
1253clear:
1254 num_used--;
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001255 memset(&entry_list->de_entries[num_used], 0,
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001256 sizeof(struct ocfs2_dx_entry));
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001257 entry_list->de_num_used = cpu_to_le16(num_used);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001258}
1259
1260static int ocfs2_delete_entry_dx(handle_t *handle, struct inode *dir,
1261 struct ocfs2_dir_lookup_result *lookup)
1262{
Mark Fashehe7c17e42009-01-29 18:17:46 -08001263 int ret, index, max_rec_len, add_to_free_list = 0;
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001264 struct buffer_head *dx_root_bh = lookup->dl_dx_root_bh;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001265 struct buffer_head *leaf_bh = lookup->dl_leaf_bh;
1266 struct ocfs2_dx_leaf *dx_leaf;
1267 struct ocfs2_dx_entry *dx_entry = lookup->dl_dx_entry;
Mark Fashehe7c17e42009-01-29 18:17:46 -08001268 struct ocfs2_dir_block_trailer *trailer;
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001269 struct ocfs2_dx_root_block *dx_root;
1270 struct ocfs2_dx_entry_list *entry_list;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001271
Mark Fashehe7c17e42009-01-29 18:17:46 -08001272 /*
1273 * This function gets a bit messy because we might have to
1274 * modify the root block, regardless of whether the indexed
1275 * entries are stored inline.
1276 */
1277
1278 /*
1279 * *Only* set 'entry_list' here, based on where we're looking
1280 * for the indexed entries. Later, we might still want to
1281 * journal both blocks, based on free list state.
1282 */
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001283 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
1284 if (ocfs2_dx_root_inline(dx_root)) {
1285 entry_list = &dx_root->dr_entries;
1286 } else {
1287 dx_leaf = (struct ocfs2_dx_leaf *) lookup->dl_dx_leaf_bh->b_data;
1288 entry_list = &dx_leaf->dl_list;
1289 }
1290
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001291 /* Neither of these are a disk corruption - that should have
1292 * been caught by lookup, before we got here. */
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001293 BUG_ON(le16_to_cpu(entry_list->de_count) <= 0);
1294 BUG_ON(le16_to_cpu(entry_list->de_num_used) <= 0);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001295
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001296 index = (char *)dx_entry - (char *)entry_list->de_entries;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001297 index /= sizeof(*dx_entry);
1298
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001299 if (index >= le16_to_cpu(entry_list->de_num_used)) {
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001300 mlog(ML_ERROR, "Dir %llu: Bad dx_entry ptr idx %d, (%p, %p)\n",
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001301 (unsigned long long)OCFS2_I(dir)->ip_blkno, index,
1302 entry_list, dx_entry);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001303 return -EIO;
1304 }
1305
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001306 /*
Mark Fashehe7c17e42009-01-29 18:17:46 -08001307 * We know that removal of this dirent will leave enough room
1308 * for a new one, so add this block to the free list if it
1309 * isn't already there.
1310 */
1311 trailer = ocfs2_trailer_from_bh(leaf_bh, dir->i_sb);
1312 if (trailer->db_free_rec_len == 0)
1313 add_to_free_list = 1;
1314
1315 /*
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001316 * Add the block holding our index into the journal before
1317 * removing the unindexed entry. If we get an error return
1318 * from __ocfs2_delete_entry(), then it hasn't removed the
1319 * entry yet. Likewise, successful return means we *must*
1320 * remove the indexed entry.
1321 *
Mark Fashehe3a93c22009-02-17 15:29:35 -08001322 * We're also careful to journal the root tree block here as
1323 * the entry count needs to be updated. Also, we might be
1324 * adding to the start of the free list.
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001325 */
Joel Becker0cf2f762009-02-12 16:41:25 -08001326 ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
Mark Fashehe3a93c22009-02-17 15:29:35 -08001327 OCFS2_JOURNAL_ACCESS_WRITE);
1328 if (ret) {
1329 mlog_errno(ret);
1330 goto out;
Mark Fashehe7c17e42009-01-29 18:17:46 -08001331 }
1332
1333 if (!ocfs2_dx_root_inline(dx_root)) {
Joel Becker0cf2f762009-02-12 16:41:25 -08001334 ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir),
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001335 lookup->dl_dx_leaf_bh,
1336 OCFS2_JOURNAL_ACCESS_WRITE);
1337 if (ret) {
1338 mlog_errno(ret);
1339 goto out;
1340 }
1341 }
1342
Tao Maf1088d42011-02-23 22:30:23 +08001343 trace_ocfs2_delete_entry_dx((unsigned long long)OCFS2_I(dir)->ip_blkno,
1344 index);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001345
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001346 ret = __ocfs2_delete_entry(handle, dir, lookup->dl_entry,
1347 leaf_bh, leaf_bh->b_data, leaf_bh->b_size);
1348 if (ret) {
1349 mlog_errno(ret);
1350 goto out;
1351 }
1352
Mark Fashehe7c17e42009-01-29 18:17:46 -08001353 max_rec_len = ocfs2_find_max_rec_len(dir->i_sb, leaf_bh);
1354 trailer->db_free_rec_len = cpu_to_le16(max_rec_len);
1355 if (add_to_free_list) {
1356 trailer->db_free_next = dx_root->dr_free_blk;
1357 dx_root->dr_free_blk = cpu_to_le64(leaf_bh->b_blocknr);
1358 ocfs2_journal_dirty(handle, dx_root_bh);
1359 }
1360
1361 /* leaf_bh was journal_accessed for us in __ocfs2_delete_entry */
1362 ocfs2_journal_dirty(handle, leaf_bh);
1363
Mark Fashehe3a93c22009-02-17 15:29:35 -08001364 le32_add_cpu(&dx_root->dr_num_entries, -1);
1365 ocfs2_journal_dirty(handle, dx_root_bh);
1366
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001367 ocfs2_dx_list_remove_entry(entry_list, index);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001368
Mark Fashehe3a93c22009-02-17 15:29:35 -08001369 if (!ocfs2_dx_root_inline(dx_root))
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001370 ocfs2_journal_dirty(handle, lookup->dl_dx_leaf_bh);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001371
1372out:
1373 return ret;
1374}
1375
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001376static inline int ocfs2_delete_entry_id(handle_t *handle,
1377 struct inode *dir,
1378 struct ocfs2_dir_entry *de_del,
1379 struct buffer_head *bh)
1380{
1381 int ret;
1382 struct buffer_head *di_bh = NULL;
1383 struct ocfs2_dinode *di;
1384 struct ocfs2_inline_data *data;
1385
Joel Beckerb657c952008-11-13 14:49:11 -08001386 ret = ocfs2_read_inode_block(dir, &di_bh);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001387 if (ret) {
1388 mlog_errno(ret);
1389 goto out;
1390 }
1391
1392 di = (struct ocfs2_dinode *)di_bh->b_data;
1393 data = &di->id2.i_data;
1394
1395 ret = __ocfs2_delete_entry(handle, dir, de_del, bh, data->id_data,
1396 i_size_read(dir));
1397
1398 brelse(di_bh);
1399out:
1400 return ret;
1401}
1402
1403static inline int ocfs2_delete_entry_el(handle_t *handle,
1404 struct inode *dir,
1405 struct ocfs2_dir_entry *de_del,
1406 struct buffer_head *bh)
1407{
1408 return __ocfs2_delete_entry(handle, dir, de_del, bh, bh->b_data,
1409 bh->b_size);
1410}
1411
1412/*
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001413 * Delete a directory entry. Hide the details of directory
1414 * implementation from the caller.
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001415 */
1416int ocfs2_delete_entry(handle_t *handle,
1417 struct inode *dir,
Mark Fasheh4a12ca32008-11-12 15:43:34 -08001418 struct ocfs2_dir_lookup_result *res)
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001419{
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001420 if (ocfs2_dir_indexed(dir))
1421 return ocfs2_delete_entry_dx(handle, dir, res);
1422
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001423 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
Mark Fasheh4a12ca32008-11-12 15:43:34 -08001424 return ocfs2_delete_entry_id(handle, dir, res->dl_entry,
1425 res->dl_leaf_bh);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001426
Mark Fasheh4a12ca32008-11-12 15:43:34 -08001427 return ocfs2_delete_entry_el(handle, dir, res->dl_entry,
1428 res->dl_leaf_bh);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001429}
1430
Mark Fasheh8553cf42007-09-13 16:29:01 -07001431/*
1432 * Check whether 'de' has enough room to hold an entry of
1433 * 'new_rec_len' bytes.
1434 */
1435static inline int ocfs2_dirent_would_fit(struct ocfs2_dir_entry *de,
1436 unsigned int new_rec_len)
1437{
1438 unsigned int de_really_used;
1439
1440 /* Check whether this is an empty record with enough space */
1441 if (le64_to_cpu(de->inode) == 0 &&
1442 le16_to_cpu(de->rec_len) >= new_rec_len)
1443 return 1;
1444
1445 /*
1446 * Record might have free space at the end which we can
1447 * use.
1448 */
1449 de_really_used = OCFS2_DIR_REC_LEN(de->name_len);
1450 if (le16_to_cpu(de->rec_len) >= (de_really_used + new_rec_len))
1451 return 1;
1452
1453 return 0;
1454}
1455
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001456static void ocfs2_dx_dir_leaf_insert_tail(struct ocfs2_dx_leaf *dx_leaf,
1457 struct ocfs2_dx_entry *dx_new_entry)
1458{
1459 int i;
1460
1461 i = le16_to_cpu(dx_leaf->dl_list.de_num_used);
1462 dx_leaf->dl_list.de_entries[i] = *dx_new_entry;
1463
1464 le16_add_cpu(&dx_leaf->dl_list.de_num_used, 1);
1465}
1466
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001467static void ocfs2_dx_entry_list_insert(struct ocfs2_dx_entry_list *entry_list,
1468 struct ocfs2_dx_hinfo *hinfo,
1469 u64 dirent_blk)
1470{
1471 int i;
1472 struct ocfs2_dx_entry *dx_entry;
1473
1474 i = le16_to_cpu(entry_list->de_num_used);
1475 dx_entry = &entry_list->de_entries[i];
1476
1477 memset(dx_entry, 0, sizeof(*dx_entry));
1478 dx_entry->dx_major_hash = cpu_to_le32(hinfo->major_hash);
1479 dx_entry->dx_minor_hash = cpu_to_le32(hinfo->minor_hash);
1480 dx_entry->dx_dirent_blk = cpu_to_le64(dirent_blk);
1481
1482 le16_add_cpu(&entry_list->de_num_used, 1);
1483}
1484
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001485static int __ocfs2_dx_dir_leaf_insert(struct inode *dir, handle_t *handle,
1486 struct ocfs2_dx_hinfo *hinfo,
1487 u64 dirent_blk,
1488 struct buffer_head *dx_leaf_bh)
1489{
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001490 int ret;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001491 struct ocfs2_dx_leaf *dx_leaf;
1492
Joel Becker0cf2f762009-02-12 16:41:25 -08001493 ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir), dx_leaf_bh,
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001494 OCFS2_JOURNAL_ACCESS_WRITE);
1495 if (ret) {
1496 mlog_errno(ret);
1497 goto out;
1498 }
1499
1500 dx_leaf = (struct ocfs2_dx_leaf *)dx_leaf_bh->b_data;
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001501 ocfs2_dx_entry_list_insert(&dx_leaf->dl_list, hinfo, dirent_blk);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001502 ocfs2_journal_dirty(handle, dx_leaf_bh);
1503
1504out:
1505 return ret;
1506}
1507
Mark Fashehe3a93c22009-02-17 15:29:35 -08001508static void ocfs2_dx_inline_root_insert(struct inode *dir, handle_t *handle,
1509 struct ocfs2_dx_hinfo *hinfo,
1510 u64 dirent_blk,
1511 struct ocfs2_dx_root_block *dx_root)
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001512{
Mark Fashehe3a93c22009-02-17 15:29:35 -08001513 ocfs2_dx_entry_list_insert(&dx_root->dr_entries, hinfo, dirent_blk);
1514}
1515
1516static int ocfs2_dx_dir_insert(struct inode *dir, handle_t *handle,
1517 struct ocfs2_dir_lookup_result *lookup)
1518{
1519 int ret = 0;
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001520 struct ocfs2_dx_root_block *dx_root;
Mark Fashehe3a93c22009-02-17 15:29:35 -08001521 struct buffer_head *dx_root_bh = lookup->dl_dx_root_bh;
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001522
Joel Becker0cf2f762009-02-12 16:41:25 -08001523 ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001524 OCFS2_JOURNAL_ACCESS_WRITE);
1525 if (ret) {
1526 mlog_errno(ret);
1527 goto out;
1528 }
1529
Mark Fashehe3a93c22009-02-17 15:29:35 -08001530 dx_root = (struct ocfs2_dx_root_block *)lookup->dl_dx_root_bh->b_data;
1531 if (ocfs2_dx_root_inline(dx_root)) {
1532 ocfs2_dx_inline_root_insert(dir, handle,
1533 &lookup->dl_hinfo,
1534 lookup->dl_leaf_bh->b_blocknr,
1535 dx_root);
1536 } else {
1537 ret = __ocfs2_dx_dir_leaf_insert(dir, handle, &lookup->dl_hinfo,
1538 lookup->dl_leaf_bh->b_blocknr,
1539 lookup->dl_dx_leaf_bh);
1540 if (ret)
1541 goto out;
1542 }
1543
1544 le32_add_cpu(&dx_root->dr_num_entries, 1);
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001545 ocfs2_journal_dirty(handle, dx_root_bh);
1546
1547out:
1548 return ret;
1549}
1550
Mark Fashehe7c17e42009-01-29 18:17:46 -08001551static void ocfs2_remove_block_from_free_list(struct inode *dir,
1552 handle_t *handle,
1553 struct ocfs2_dir_lookup_result *lookup)
1554{
1555 struct ocfs2_dir_block_trailer *trailer, *prev;
1556 struct ocfs2_dx_root_block *dx_root;
1557 struct buffer_head *bh;
1558
1559 trailer = ocfs2_trailer_from_bh(lookup->dl_leaf_bh, dir->i_sb);
1560
1561 if (ocfs2_free_list_at_root(lookup)) {
1562 bh = lookup->dl_dx_root_bh;
1563 dx_root = (struct ocfs2_dx_root_block *)bh->b_data;
1564 dx_root->dr_free_blk = trailer->db_free_next;
1565 } else {
1566 bh = lookup->dl_prev_leaf_bh;
1567 prev = ocfs2_trailer_from_bh(bh, dir->i_sb);
1568 prev->db_free_next = trailer->db_free_next;
1569 }
1570
1571 trailer->db_free_rec_len = cpu_to_le16(0);
1572 trailer->db_free_next = cpu_to_le64(0);
1573
1574 ocfs2_journal_dirty(handle, bh);
1575 ocfs2_journal_dirty(handle, lookup->dl_leaf_bh);
1576}
1577
1578/*
1579 * This expects that a journal write has been reserved on
1580 * lookup->dl_prev_leaf_bh or lookup->dl_dx_root_bh
1581 */
1582static void ocfs2_recalc_free_list(struct inode *dir, handle_t *handle,
1583 struct ocfs2_dir_lookup_result *lookup)
1584{
1585 int max_rec_len;
1586 struct ocfs2_dir_block_trailer *trailer;
1587
1588 /* Walk dl_leaf_bh to figure out what the new free rec_len is. */
1589 max_rec_len = ocfs2_find_max_rec_len(dir->i_sb, lookup->dl_leaf_bh);
1590 if (max_rec_len) {
1591 /*
1592 * There's still room in this block, so no need to remove it
1593 * from the free list. In this case, we just want to update
1594 * the rec len accounting.
1595 */
1596 trailer = ocfs2_trailer_from_bh(lookup->dl_leaf_bh, dir->i_sb);
1597 trailer->db_free_rec_len = cpu_to_le16(max_rec_len);
1598 ocfs2_journal_dirty(handle, lookup->dl_leaf_bh);
1599 } else {
1600 ocfs2_remove_block_from_free_list(dir, handle, lookup);
1601 }
1602}
1603
Mark Fasheh316f4b92007-09-07 18:21:26 -07001604/* we don't always have a dentry for what we want to add, so people
1605 * like orphan dir can call this instead.
1606 *
Mark Fasheh4a12ca32008-11-12 15:43:34 -08001607 * The lookup context must have been filled from
1608 * ocfs2_prepare_dir_for_insert.
Mark Fasheh316f4b92007-09-07 18:21:26 -07001609 */
1610int __ocfs2_add_entry(handle_t *handle,
1611 struct inode *dir,
1612 const char *name, int namelen,
1613 struct inode *inode, u64 blkno,
1614 struct buffer_head *parent_fe_bh,
Mark Fasheh4a12ca32008-11-12 15:43:34 -08001615 struct ocfs2_dir_lookup_result *lookup)
Mark Fasheh316f4b92007-09-07 18:21:26 -07001616{
1617 unsigned long offset;
1618 unsigned short rec_len;
1619 struct ocfs2_dir_entry *de, *de1;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001620 struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_fe_bh->b_data;
1621 struct super_block *sb = dir->i_sb;
Mark Fasheh316f4b92007-09-07 18:21:26 -07001622 int retval, status;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001623 unsigned int size = sb->s_blocksize;
Mark Fasheh4a12ca32008-11-12 15:43:34 -08001624 struct buffer_head *insert_bh = lookup->dl_leaf_bh;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001625 char *data_start = insert_bh->b_data;
Mark Fasheh316f4b92007-09-07 18:21:26 -07001626
Mark Fasheh316f4b92007-09-07 18:21:26 -07001627 if (!namelen)
1628 return -EINVAL;
1629
Mark Fashehe7c17e42009-01-29 18:17:46 -08001630 if (ocfs2_dir_indexed(dir)) {
1631 struct buffer_head *bh;
1632
1633 /*
1634 * An indexed dir may require that we update the free space
1635 * list. Reserve a write to the previous node in the list so
1636 * that we don't fail later.
1637 *
1638 * XXX: This can be either a dx_root_block, or an unindexed
1639 * directory tree leaf block.
1640 */
1641 if (ocfs2_free_list_at_root(lookup)) {
1642 bh = lookup->dl_dx_root_bh;
Joel Becker0cf2f762009-02-12 16:41:25 -08001643 retval = ocfs2_journal_access_dr(handle,
1644 INODE_CACHE(dir), bh,
Mark Fashehe7c17e42009-01-29 18:17:46 -08001645 OCFS2_JOURNAL_ACCESS_WRITE);
1646 } else {
1647 bh = lookup->dl_prev_leaf_bh;
Joel Becker0cf2f762009-02-12 16:41:25 -08001648 retval = ocfs2_journal_access_db(handle,
1649 INODE_CACHE(dir), bh,
Mark Fashehe7c17e42009-01-29 18:17:46 -08001650 OCFS2_JOURNAL_ACCESS_WRITE);
1651 }
1652 if (retval) {
1653 mlog_errno(retval);
1654 return retval;
1655 }
1656 } else if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001657 data_start = di->id2.i_data.id_data;
1658 size = i_size_read(dir);
1659
1660 BUG_ON(insert_bh != parent_fe_bh);
1661 }
1662
Mark Fasheh316f4b92007-09-07 18:21:26 -07001663 rec_len = OCFS2_DIR_REC_LEN(namelen);
1664 offset = 0;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001665 de = (struct ocfs2_dir_entry *) data_start;
Mark Fasheh316f4b92007-09-07 18:21:26 -07001666 while (1) {
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07001667 BUG_ON((char *)de >= (size + data_start));
1668
Mark Fasheh316f4b92007-09-07 18:21:26 -07001669 /* These checks should've already been passed by the
1670 * prepare function, but I guess we can leave them
1671 * here anyway. */
1672 if (!ocfs2_check_dir_entry(dir, de, insert_bh, offset)) {
1673 retval = -ENOENT;
1674 goto bail;
1675 }
1676 if (ocfs2_match(namelen, name, de)) {
1677 retval = -EEXIST;
1678 goto bail;
1679 }
Mark Fasheh8553cf42007-09-13 16:29:01 -07001680
Mark Fasheh87d35a72008-12-10 17:36:25 -08001681 /* We're guaranteed that we should have space, so we
1682 * can't possibly have hit the trailer...right? */
1683 mlog_bug_on_msg(ocfs2_skip_dir_trailer(dir, de, offset, size),
1684 "Hit dir trailer trying to insert %.*s "
1685 "(namelen %d) into directory %llu. "
1686 "offset is %lu, trailer offset is %d\n",
1687 namelen, name, namelen,
1688 (unsigned long long)parent_fe_bh->b_blocknr,
1689 offset, ocfs2_dir_trailer_blk_off(dir->i_sb));
1690
Mark Fasheh8553cf42007-09-13 16:29:01 -07001691 if (ocfs2_dirent_would_fit(de, rec_len)) {
Mark Fasheh316f4b92007-09-07 18:21:26 -07001692 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1693 retval = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
1694 if (retval < 0) {
1695 mlog_errno(retval);
1696 goto bail;
1697 }
1698
Joel Becker13723d02008-10-17 19:25:01 -07001699 if (insert_bh == parent_fe_bh)
Joel Becker0cf2f762009-02-12 16:41:25 -08001700 status = ocfs2_journal_access_di(handle,
1701 INODE_CACHE(dir),
Joel Becker13723d02008-10-17 19:25:01 -07001702 insert_bh,
1703 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001704 else {
Joel Becker0cf2f762009-02-12 16:41:25 -08001705 status = ocfs2_journal_access_db(handle,
1706 INODE_CACHE(dir),
Joel Becker13723d02008-10-17 19:25:01 -07001707 insert_bh,
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001708 OCFS2_JOURNAL_ACCESS_WRITE);
1709
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001710 if (ocfs2_dir_indexed(dir)) {
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08001711 status = ocfs2_dx_dir_insert(dir,
1712 handle,
1713 lookup);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08001714 if (status) {
1715 mlog_errno(status);
1716 goto bail;
1717 }
1718 }
1719 }
1720
Mark Fasheh316f4b92007-09-07 18:21:26 -07001721 /* By now the buffer is marked for journaling */
1722 offset += le16_to_cpu(de->rec_len);
1723 if (le64_to_cpu(de->inode)) {
1724 de1 = (struct ocfs2_dir_entry *)((char *) de +
1725 OCFS2_DIR_REC_LEN(de->name_len));
1726 de1->rec_len =
1727 cpu_to_le16(le16_to_cpu(de->rec_len) -
1728 OCFS2_DIR_REC_LEN(de->name_len));
1729 de->rec_len = cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
1730 de = de1;
1731 }
1732 de->file_type = OCFS2_FT_UNKNOWN;
1733 if (blkno) {
1734 de->inode = cpu_to_le64(blkno);
1735 ocfs2_set_de_type(de, inode->i_mode);
1736 } else
1737 de->inode = 0;
1738 de->name_len = namelen;
1739 memcpy(de->name, name, namelen);
1740
Mark Fashehe7c17e42009-01-29 18:17:46 -08001741 if (ocfs2_dir_indexed(dir))
1742 ocfs2_recalc_free_list(dir, handle, lookup);
1743
Mark Fasheh316f4b92007-09-07 18:21:26 -07001744 dir->i_version++;
Joel Beckerec20cec2010-03-19 14:13:52 -07001745 ocfs2_journal_dirty(handle, insert_bh);
Mark Fasheh316f4b92007-09-07 18:21:26 -07001746 retval = 0;
1747 goto bail;
1748 }
Mark Fasheh87d35a72008-12-10 17:36:25 -08001749
Mark Fasheh316f4b92007-09-07 18:21:26 -07001750 offset += le16_to_cpu(de->rec_len);
1751 de = (struct ocfs2_dir_entry *) ((char *) de + le16_to_cpu(de->rec_len));
1752 }
1753
1754 /* when you think about it, the assert above should prevent us
1755 * from ever getting here. */
1756 retval = -ENOSPC;
1757bail:
Tao Mac1e8d352011-03-07 16:43:21 +08001758 if (retval)
1759 mlog_errno(retval);
Mark Fasheh316f4b92007-09-07 18:21:26 -07001760
Mark Fasheh316f4b92007-09-07 18:21:26 -07001761 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
Tao Maf1088d42011-02-23 22:30:23 +08002022 trace_ocfs2_readdir((unsigned long long)OCFS2_I(inode)->ip_blkno);
Mark Fashehb8bc5f42007-09-10 17:17:52 -07002023
Mark Fashehe63aecb62007-10-18 15:30:42 -07002024 error = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level);
Mark Fashehb8bc5f42007-09-10 17:17:52 -07002025 if (lock_level && error >= 0) {
2026 /* We release EX lock which used to update atime
2027 * and get PR lock again to reduce contention
2028 * on commonly accessed directories. */
Mark Fashehe63aecb62007-10-18 15:30:42 -07002029 ocfs2_inode_unlock(inode, 1);
Mark Fashehb8bc5f42007-09-10 17:17:52 -07002030 lock_level = 0;
Mark Fashehe63aecb62007-10-18 15:30:42 -07002031 error = ocfs2_inode_lock(inode, NULL, 0);
Mark Fashehb8bc5f42007-09-10 17:17:52 -07002032 }
2033 if (error < 0) {
2034 if (error != -ENOENT)
2035 mlog_errno(error);
2036 /* we haven't got any yet, so propagate the error. */
2037 goto bail_nolock;
2038 }
2039
2040 error = ocfs2_dir_foreach_blk(inode, &filp->f_version, &filp->f_pos,
Mark Fashehe7b34012007-09-24 14:25:27 -07002041 dirent, filldir, NULL);
Mark Fashehb8bc5f42007-09-10 17:17:52 -07002042
Mark Fashehe63aecb62007-10-18 15:30:42 -07002043 ocfs2_inode_unlock(inode, lock_level);
Tao Mac1e8d352011-03-07 16:43:21 +08002044 if (error)
2045 mlog_errno(error);
Mark Fashehccd979b2005-12-15 14:31:24 -08002046
Mark Fashehaa958872006-04-21 13:49:02 -07002047bail_nolock:
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
Tao Maf1088d42011-02-23 22:30:23 +08002063 trace_ocfs2_find_files_on_disk(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
Tao Maf1088d42011-02-23 22:30:23 +08002108 trace_ocfs2_check_dir_for_entry(
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
Tao Mac1e8d352011-03-07 16:43:21 +08002119 if (ret)
2120 mlog_errno(ret);
Mark Fashehccd979b2005-12-15 14:31:24 -08002121 return ret;
2122}
2123
Mark Fasheh0bfbbf62007-09-12 11:19:00 -07002124struct ocfs2_empty_dir_priv {
2125 unsigned seen_dot;
2126 unsigned seen_dot_dot;
2127 unsigned seen_other;
Mark Fashehe3a93c22009-02-17 15:29:35 -08002128 unsigned dx_dir;
Mark Fasheh0bfbbf62007-09-12 11:19:00 -07002129};
2130static int ocfs2_empty_dir_filldir(void *priv, const char *name, int name_len,
2131 loff_t pos, u64 ino, unsigned type)
2132{
2133 struct ocfs2_empty_dir_priv *p = priv;
2134
2135 /*
2136 * Check the positions of "." and ".." records to be sure
2137 * they're in the correct place.
Mark Fashehe3a93c22009-02-17 15:29:35 -08002138 *
2139 * Indexed directories don't need to proceed past the first
2140 * two entries, so we end the scan after seeing '..'. Despite
2141 * that, we allow the scan to proceed In the event that we
2142 * have a corrupted indexed directory (no dot or dot dot
2143 * entries). This allows us to double check for existing
2144 * entries which might not have been found in the index.
Mark Fasheh0bfbbf62007-09-12 11:19:00 -07002145 */
2146 if (name_len == 1 && !strncmp(".", name, 1) && pos == 0) {
2147 p->seen_dot = 1;
2148 return 0;
2149 }
2150
2151 if (name_len == 2 && !strncmp("..", name, 2) &&
2152 pos == OCFS2_DIR_REC_LEN(1)) {
2153 p->seen_dot_dot = 1;
Mark Fashehe3a93c22009-02-17 15:29:35 -08002154
2155 if (p->dx_dir && p->seen_dot)
2156 return 1;
2157
Mark Fasheh0bfbbf62007-09-12 11:19:00 -07002158 return 0;
2159 }
2160
2161 p->seen_other = 1;
2162 return 1;
2163}
Mark Fashehe3a93c22009-02-17 15:29:35 -08002164
2165static int ocfs2_empty_dir_dx(struct inode *inode,
2166 struct ocfs2_empty_dir_priv *priv)
2167{
2168 int ret;
2169 struct buffer_head *di_bh = NULL;
2170 struct buffer_head *dx_root_bh = NULL;
2171 struct ocfs2_dinode *di;
2172 struct ocfs2_dx_root_block *dx_root;
2173
2174 priv->dx_dir = 1;
2175
2176 ret = ocfs2_read_inode_block(inode, &di_bh);
2177 if (ret) {
2178 mlog_errno(ret);
2179 goto out;
2180 }
2181 di = (struct ocfs2_dinode *)di_bh->b_data;
2182
2183 ret = ocfs2_read_dx_root(inode, di, &dx_root_bh);
2184 if (ret) {
2185 mlog_errno(ret);
2186 goto out;
2187 }
2188 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
2189
2190 if (le32_to_cpu(dx_root->dr_num_entries) != 2)
2191 priv->seen_other = 1;
2192
2193out:
2194 brelse(di_bh);
2195 brelse(dx_root_bh);
2196 return ret;
2197}
2198
Mark Fashehccd979b2005-12-15 14:31:24 -08002199/*
2200 * routine to check that the specified directory is empty (for rmdir)
Mark Fasheh0bfbbf62007-09-12 11:19:00 -07002201 *
2202 * Returns 1 if dir is empty, zero otherwise.
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002203 *
Mark Fashehe3a93c22009-02-17 15:29:35 -08002204 * XXX: This is a performance problem for unindexed directories.
Mark Fashehccd979b2005-12-15 14:31:24 -08002205 */
2206int ocfs2_empty_dir(struct inode *inode)
2207{
Mark Fasheh0bfbbf62007-09-12 11:19:00 -07002208 int ret;
2209 loff_t start = 0;
2210 struct ocfs2_empty_dir_priv priv;
Mark Fashehccd979b2005-12-15 14:31:24 -08002211
Mark Fasheh0bfbbf62007-09-12 11:19:00 -07002212 memset(&priv, 0, sizeof(priv));
2213
Mark Fashehe3a93c22009-02-17 15:29:35 -08002214 if (ocfs2_dir_indexed(inode)) {
2215 ret = ocfs2_empty_dir_dx(inode, &priv);
2216 if (ret)
2217 mlog_errno(ret);
2218 /*
2219 * We still run ocfs2_dir_foreach to get the checks
2220 * for "." and "..".
2221 */
2222 }
2223
Mark Fasheh0bfbbf62007-09-12 11:19:00 -07002224 ret = ocfs2_dir_foreach(inode, &start, &priv, ocfs2_empty_dir_filldir);
2225 if (ret)
2226 mlog_errno(ret);
2227
2228 if (!priv.seen_dot || !priv.seen_dot_dot) {
2229 mlog(ML_ERROR, "bad directory (dir #%llu) - no `.' or `..'\n",
Mark Fashehb0697052006-03-03 10:24:33 -08002230 (unsigned long long)OCFS2_I(inode)->ip_blkno);
Mark Fasheh0bfbbf62007-09-12 11:19:00 -07002231 /*
2232 * XXX: Is it really safe to allow an unlink to continue?
2233 */
Mark Fashehccd979b2005-12-15 14:31:24 -08002234 return 1;
2235 }
2236
Mark Fasheh0bfbbf62007-09-12 11:19:00 -07002237 return !priv.seen_other;
Mark Fashehccd979b2005-12-15 14:31:24 -08002238}
2239
Mark Fasheh87d35a72008-12-10 17:36:25 -08002240/*
2241 * Fills "." and ".." dirents in a new directory block. Returns dirent for
2242 * "..", which might be used during creation of a directory with a trailing
2243 * header. It is otherwise safe to ignore the return code.
2244 */
2245static struct ocfs2_dir_entry *ocfs2_fill_initial_dirents(struct inode *inode,
2246 struct inode *parent,
2247 char *start,
2248 unsigned int size)
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002249{
2250 struct ocfs2_dir_entry *de = (struct ocfs2_dir_entry *)start;
2251
2252 de->inode = cpu_to_le64(OCFS2_I(inode)->ip_blkno);
2253 de->name_len = 1;
2254 de->rec_len =
2255 cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
2256 strcpy(de->name, ".");
2257 ocfs2_set_de_type(de, S_IFDIR);
2258
2259 de = (struct ocfs2_dir_entry *) ((char *)de + le16_to_cpu(de->rec_len));
2260 de->inode = cpu_to_le64(OCFS2_I(parent)->ip_blkno);
2261 de->rec_len = cpu_to_le16(size - OCFS2_DIR_REC_LEN(1));
2262 de->name_len = 2;
2263 strcpy(de->name, "..");
2264 ocfs2_set_de_type(de, S_IFDIR);
Mark Fasheh87d35a72008-12-10 17:36:25 -08002265
2266 return de;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002267}
2268
2269/*
2270 * This works together with code in ocfs2_mknod_locked() which sets
2271 * the inline-data flag and initializes the inline-data section.
2272 */
2273static int ocfs2_fill_new_dir_id(struct ocfs2_super *osb,
2274 handle_t *handle,
2275 struct inode *parent,
2276 struct inode *inode,
2277 struct buffer_head *di_bh)
2278{
2279 int ret;
2280 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2281 struct ocfs2_inline_data *data = &di->id2.i_data;
2282 unsigned int size = le16_to_cpu(data->id_count);
2283
Joel Becker0cf2f762009-02-12 16:41:25 -08002284 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
Joel Becker13723d02008-10-17 19:25:01 -07002285 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002286 if (ret) {
2287 mlog_errno(ret);
2288 goto out;
2289 }
2290
2291 ocfs2_fill_initial_dirents(inode, parent, data->id_data, size);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002292 ocfs2_journal_dirty(handle, di_bh);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002293
2294 i_size_write(inode, size);
2295 inode->i_nlink = 2;
2296 inode->i_blocks = ocfs2_inode_sector_count(inode);
2297
2298 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
2299 if (ret < 0)
2300 mlog_errno(ret);
2301
2302out:
2303 return ret;
2304}
2305
2306static int ocfs2_fill_new_dir_el(struct ocfs2_super *osb,
2307 handle_t *handle,
2308 struct inode *parent,
2309 struct inode *inode,
2310 struct buffer_head *fe_bh,
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002311 struct ocfs2_alloc_context *data_ac,
2312 struct buffer_head **ret_new_bh)
Mark Fasheh316f4b92007-09-07 18:21:26 -07002313{
2314 int status;
Mark Fasheh87d35a72008-12-10 17:36:25 -08002315 unsigned int size = osb->sb->s_blocksize;
Mark Fasheh316f4b92007-09-07 18:21:26 -07002316 struct buffer_head *new_bh = NULL;
Mark Fasheh87d35a72008-12-10 17:36:25 -08002317 struct ocfs2_dir_entry *de;
Mark Fasheh316f4b92007-09-07 18:21:26 -07002318
Mark Fashehe7c17e42009-01-29 18:17:46 -08002319 if (ocfs2_new_dir_wants_trailer(inode))
Mark Fasheh87d35a72008-12-10 17:36:25 -08002320 size = ocfs2_dir_trailer_blk_off(parent->i_sb);
2321
Mark Fasheh316f4b92007-09-07 18:21:26 -07002322 status = ocfs2_do_extend_dir(osb->sb, handle, inode, fe_bh,
2323 data_ac, NULL, &new_bh);
2324 if (status < 0) {
2325 mlog_errno(status);
2326 goto bail;
2327 }
2328
Joel Becker8cb471e2009-02-10 20:00:41 -08002329 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
Mark Fasheh316f4b92007-09-07 18:21:26 -07002330
Joel Becker0cf2f762009-02-12 16:41:25 -08002331 status = ocfs2_journal_access_db(handle, INODE_CACHE(inode), new_bh,
Joel Becker13723d02008-10-17 19:25:01 -07002332 OCFS2_JOURNAL_ACCESS_CREATE);
Mark Fasheh316f4b92007-09-07 18:21:26 -07002333 if (status < 0) {
2334 mlog_errno(status);
2335 goto bail;
2336 }
2337 memset(new_bh->b_data, 0, osb->sb->s_blocksize);
2338
Mark Fasheh87d35a72008-12-10 17:36:25 -08002339 de = ocfs2_fill_initial_dirents(inode, parent, new_bh->b_data, size);
Mark Fashehe7c17e42009-01-29 18:17:46 -08002340 if (ocfs2_new_dir_wants_trailer(inode)) {
2341 int size = le16_to_cpu(de->rec_len);
2342
2343 /*
2344 * Figure out the size of the hole left over after
2345 * insertion of '.' and '..'. The trailer wants this
2346 * information.
2347 */
2348 size -= OCFS2_DIR_REC_LEN(2);
2349 size -= sizeof(struct ocfs2_dir_block_trailer);
2350
2351 ocfs2_init_dir_trailer(inode, new_bh, size);
2352 }
Mark Fasheh316f4b92007-09-07 18:21:26 -07002353
Joel Beckerec20cec2010-03-19 14:13:52 -07002354 ocfs2_journal_dirty(handle, new_bh);
Mark Fasheh316f4b92007-09-07 18:21:26 -07002355
2356 i_size_write(inode, inode->i_sb->s_blocksize);
2357 inode->i_nlink = 2;
2358 inode->i_blocks = ocfs2_inode_sector_count(inode);
2359 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
2360 if (status < 0) {
2361 mlog_errno(status);
2362 goto bail;
2363 }
2364
2365 status = 0;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002366 if (ret_new_bh) {
2367 *ret_new_bh = new_bh;
2368 new_bh = NULL;
2369 }
Mark Fasheh316f4b92007-09-07 18:21:26 -07002370bail:
Mark Fasheha81cb882008-10-07 14:25:16 -07002371 brelse(new_bh);
Mark Fasheh316f4b92007-09-07 18:21:26 -07002372
Mark Fasheh316f4b92007-09-07 18:21:26 -07002373 return status;
2374}
2375
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002376static int ocfs2_dx_dir_attach_index(struct ocfs2_super *osb,
2377 handle_t *handle, struct inode *dir,
2378 struct buffer_head *di_bh,
Mark Fashehe7c17e42009-01-29 18:17:46 -08002379 struct buffer_head *dirdata_bh,
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002380 struct ocfs2_alloc_context *meta_ac,
Mark Fashehe3a93c22009-02-17 15:29:35 -08002381 int dx_inline, u32 num_entries,
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002382 struct buffer_head **ret_dx_root_bh)
2383{
2384 int ret;
2385 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
2386 u16 dr_suballoc_bit;
Joel Becker2b6cb572010-03-26 10:09:15 +08002387 u64 suballoc_loc, dr_blkno;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002388 unsigned int num_bits;
2389 struct buffer_head *dx_root_bh = NULL;
2390 struct ocfs2_dx_root_block *dx_root;
Mark Fashehe7c17e42009-01-29 18:17:46 -08002391 struct ocfs2_dir_block_trailer *trailer =
2392 ocfs2_trailer_from_bh(dirdata_bh, dir->i_sb);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002393
Joel Becker2b6cb572010-03-26 10:09:15 +08002394 ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
2395 &dr_suballoc_bit, &num_bits, &dr_blkno);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002396 if (ret) {
2397 mlog_errno(ret);
2398 goto out;
2399 }
2400
Tao Maf1088d42011-02-23 22:30:23 +08002401 trace_ocfs2_dx_dir_attach_index(
2402 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2403 (unsigned long long)dr_blkno);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002404
2405 dx_root_bh = sb_getblk(osb->sb, dr_blkno);
2406 if (dx_root_bh == NULL) {
2407 ret = -EIO;
2408 goto out;
2409 }
Joel Becker8cb471e2009-02-10 20:00:41 -08002410 ocfs2_set_new_buffer_uptodate(INODE_CACHE(dir), dx_root_bh);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002411
Joel Becker0cf2f762009-02-12 16:41:25 -08002412 ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002413 OCFS2_JOURNAL_ACCESS_CREATE);
2414 if (ret < 0) {
2415 mlog_errno(ret);
2416 goto out;
2417 }
2418
2419 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
2420 memset(dx_root, 0, osb->sb->s_blocksize);
2421 strcpy(dx_root->dr_signature, OCFS2_DX_ROOT_SIGNATURE);
Tiger Yangb89c5422010-01-25 14:11:06 +08002422 dx_root->dr_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
Joel Becker2b6cb572010-03-26 10:09:15 +08002423 dx_root->dr_suballoc_loc = cpu_to_le64(suballoc_loc);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002424 dx_root->dr_suballoc_bit = cpu_to_le16(dr_suballoc_bit);
2425 dx_root->dr_fs_generation = cpu_to_le32(osb->fs_generation);
2426 dx_root->dr_blkno = cpu_to_le64(dr_blkno);
2427 dx_root->dr_dir_blkno = cpu_to_le64(OCFS2_I(dir)->ip_blkno);
Mark Fashehe3a93c22009-02-17 15:29:35 -08002428 dx_root->dr_num_entries = cpu_to_le32(num_entries);
Mark Fashehe7c17e42009-01-29 18:17:46 -08002429 if (le16_to_cpu(trailer->db_free_rec_len))
2430 dx_root->dr_free_blk = cpu_to_le64(dirdata_bh->b_blocknr);
2431 else
2432 dx_root->dr_free_blk = cpu_to_le64(0);
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08002433
2434 if (dx_inline) {
2435 dx_root->dr_flags |= OCFS2_DX_FLAG_INLINE;
2436 dx_root->dr_entries.de_count =
2437 cpu_to_le16(ocfs2_dx_entries_per_root(osb->sb));
2438 } else {
2439 dx_root->dr_list.l_count =
2440 cpu_to_le16(ocfs2_extent_recs_per_dx_root(osb->sb));
2441 }
Joel Beckerec20cec2010-03-19 14:13:52 -07002442 ocfs2_journal_dirty(handle, dx_root_bh);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002443
Joel Becker0cf2f762009-02-12 16:41:25 -08002444 ret = ocfs2_journal_access_di(handle, INODE_CACHE(dir), di_bh,
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002445 OCFS2_JOURNAL_ACCESS_CREATE);
2446 if (ret) {
2447 mlog_errno(ret);
2448 goto out;
2449 }
2450
2451 di->i_dx_root = cpu_to_le64(dr_blkno);
2452
Tao Ma8ac33dc2010-12-15 16:30:00 +08002453 spin_lock(&OCFS2_I(dir)->ip_lock);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002454 OCFS2_I(dir)->ip_dyn_features |= OCFS2_INDEXED_DIR_FL;
2455 di->i_dyn_features = cpu_to_le16(OCFS2_I(dir)->ip_dyn_features);
Tao Ma8ac33dc2010-12-15 16:30:00 +08002456 spin_unlock(&OCFS2_I(dir)->ip_lock);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002457
Joel Beckerec20cec2010-03-19 14:13:52 -07002458 ocfs2_journal_dirty(handle, di_bh);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002459
2460 *ret_dx_root_bh = dx_root_bh;
2461 dx_root_bh = NULL;
2462
2463out:
2464 brelse(dx_root_bh);
2465 return ret;
2466}
2467
2468static int ocfs2_dx_dir_format_cluster(struct ocfs2_super *osb,
2469 handle_t *handle, struct inode *dir,
2470 struct buffer_head **dx_leaves,
2471 int num_dx_leaves, u64 start_blk)
2472{
2473 int ret, i;
2474 struct ocfs2_dx_leaf *dx_leaf;
2475 struct buffer_head *bh;
2476
2477 for (i = 0; i < num_dx_leaves; i++) {
2478 bh = sb_getblk(osb->sb, start_blk + i);
2479 if (bh == NULL) {
2480 ret = -EIO;
2481 goto out;
2482 }
2483 dx_leaves[i] = bh;
2484
Joel Becker8cb471e2009-02-10 20:00:41 -08002485 ocfs2_set_new_buffer_uptodate(INODE_CACHE(dir), bh);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002486
Joel Becker0cf2f762009-02-12 16:41:25 -08002487 ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir), bh,
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002488 OCFS2_JOURNAL_ACCESS_CREATE);
2489 if (ret < 0) {
2490 mlog_errno(ret);
2491 goto out;
2492 }
2493
2494 dx_leaf = (struct ocfs2_dx_leaf *) bh->b_data;
2495
2496 memset(dx_leaf, 0, osb->sb->s_blocksize);
2497 strcpy(dx_leaf->dl_signature, OCFS2_DX_LEAF_SIGNATURE);
2498 dx_leaf->dl_fs_generation = cpu_to_le32(osb->fs_generation);
2499 dx_leaf->dl_blkno = cpu_to_le64(bh->b_blocknr);
2500 dx_leaf->dl_list.de_count =
2501 cpu_to_le16(ocfs2_dx_entries_per_leaf(osb->sb));
2502
Tao Maf1088d42011-02-23 22:30:23 +08002503 trace_ocfs2_dx_dir_format_cluster(
2504 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2505 (unsigned long long)bh->b_blocknr,
2506 le16_to_cpu(dx_leaf->dl_list.de_count));
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002507
2508 ocfs2_journal_dirty(handle, bh);
2509 }
2510
2511 ret = 0;
2512out:
2513 return ret;
2514}
2515
2516/*
2517 * Allocates and formats a new cluster for use in an indexed dir
2518 * leaf. This version will not do the extent insert, so that it can be
2519 * used by operations which need careful ordering.
2520 */
2521static int __ocfs2_dx_dir_new_cluster(struct inode *dir,
2522 u32 cpos, handle_t *handle,
2523 struct ocfs2_alloc_context *data_ac,
2524 struct buffer_head **dx_leaves,
2525 int num_dx_leaves, u64 *ret_phys_blkno)
2526{
2527 int ret;
2528 u32 phys, num;
2529 u64 phys_blkno;
2530 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2531
2532 /*
2533 * XXX: For create, this should claim cluster for the index
2534 * *before* the unindexed insert so that we have a better
2535 * chance of contiguousness as the directory grows in number
2536 * of entries.
2537 */
Joel Becker1ed9b772010-05-06 13:59:06 +08002538 ret = __ocfs2_claim_clusters(handle, data_ac, 1, 1, &phys, &num);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002539 if (ret) {
2540 mlog_errno(ret);
2541 goto out;
2542 }
2543
2544 /*
2545 * Format the new cluster first. That way, we're inserting
2546 * valid data.
2547 */
2548 phys_blkno = ocfs2_clusters_to_blocks(osb->sb, phys);
2549 ret = ocfs2_dx_dir_format_cluster(osb, handle, dir, dx_leaves,
2550 num_dx_leaves, phys_blkno);
2551 if (ret) {
2552 mlog_errno(ret);
2553 goto out;
2554 }
2555
2556 *ret_phys_blkno = phys_blkno;
2557out:
2558 return ret;
2559}
2560
2561static int ocfs2_dx_dir_new_cluster(struct inode *dir,
2562 struct ocfs2_extent_tree *et,
2563 u32 cpos, handle_t *handle,
2564 struct ocfs2_alloc_context *data_ac,
2565 struct ocfs2_alloc_context *meta_ac,
2566 struct buffer_head **dx_leaves,
2567 int num_dx_leaves)
2568{
2569 int ret;
2570 u64 phys_blkno;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002571
2572 ret = __ocfs2_dx_dir_new_cluster(dir, cpos, handle, data_ac, dx_leaves,
2573 num_dx_leaves, &phys_blkno);
2574 if (ret) {
2575 mlog_errno(ret);
2576 goto out;
2577 }
2578
Joel Beckercc79d8c2009-02-13 03:24:43 -08002579 ret = ocfs2_insert_extent(handle, et, cpos, phys_blkno, 1, 0,
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002580 meta_ac);
2581 if (ret)
2582 mlog_errno(ret);
2583out:
2584 return ret;
2585}
2586
2587static struct buffer_head **ocfs2_dx_dir_kmalloc_leaves(struct super_block *sb,
2588 int *ret_num_leaves)
2589{
2590 int num_dx_leaves = ocfs2_clusters_to_blocks(sb, 1);
2591 struct buffer_head **dx_leaves;
2592
2593 dx_leaves = kcalloc(num_dx_leaves, sizeof(struct buffer_head *),
2594 GFP_NOFS);
2595 if (dx_leaves && ret_num_leaves)
2596 *ret_num_leaves = num_dx_leaves;
2597
2598 return dx_leaves;
2599}
2600
2601static int ocfs2_fill_new_dir_dx(struct ocfs2_super *osb,
2602 handle_t *handle,
2603 struct inode *parent,
2604 struct inode *inode,
2605 struct buffer_head *di_bh,
2606 struct ocfs2_alloc_context *data_ac,
2607 struct ocfs2_alloc_context *meta_ac)
2608{
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08002609 int ret;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002610 struct buffer_head *leaf_bh = NULL;
2611 struct buffer_head *dx_root_bh = NULL;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002612 struct ocfs2_dx_hinfo hinfo;
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08002613 struct ocfs2_dx_root_block *dx_root;
2614 struct ocfs2_dx_entry_list *entry_list;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002615
2616 /*
2617 * Our strategy is to create the directory as though it were
2618 * unindexed, then add the index block. This works with very
2619 * little complication since the state of a new directory is a
2620 * very well known quantity.
2621 *
2622 * Essentially, we have two dirents ("." and ".."), in the 1st
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08002623 * block which need indexing. These are easily inserted into
2624 * the index block.
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002625 */
2626
2627 ret = ocfs2_fill_new_dir_el(osb, handle, parent, inode, di_bh,
2628 data_ac, &leaf_bh);
2629 if (ret) {
2630 mlog_errno(ret);
2631 goto out;
2632 }
2633
Mark Fashehe7c17e42009-01-29 18:17:46 -08002634 ret = ocfs2_dx_dir_attach_index(osb, handle, inode, di_bh, leaf_bh,
Mark Fashehe3a93c22009-02-17 15:29:35 -08002635 meta_ac, 1, 2, &dx_root_bh);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002636 if (ret) {
2637 mlog_errno(ret);
2638 goto out;
2639 }
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08002640 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
2641 entry_list = &dx_root->dr_entries;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002642
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08002643 /* Buffer has been journaled for us by ocfs2_dx_dir_attach_index */
Mark Fashehe7c17e42009-01-29 18:17:46 -08002644 ocfs2_dx_dir_name_hash(inode, ".", 1, &hinfo);
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08002645 ocfs2_dx_entry_list_insert(entry_list, &hinfo, leaf_bh->b_blocknr);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002646
2647 ocfs2_dx_dir_name_hash(inode, "..", 2, &hinfo);
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08002648 ocfs2_dx_entry_list_insert(entry_list, &hinfo, leaf_bh->b_blocknr);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002649
2650out:
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002651 brelse(dx_root_bh);
2652 brelse(leaf_bh);
2653 return ret;
2654}
2655
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002656int ocfs2_fill_new_dir(struct ocfs2_super *osb,
2657 handle_t *handle,
2658 struct inode *parent,
2659 struct inode *inode,
2660 struct buffer_head *fe_bh,
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002661 struct ocfs2_alloc_context *data_ac,
2662 struct ocfs2_alloc_context *meta_ac)
2663
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002664{
2665 BUG_ON(!ocfs2_supports_inline_data(osb) && data_ac == NULL);
2666
2667 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2668 return ocfs2_fill_new_dir_id(osb, handle, parent, inode, fe_bh);
2669
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002670 if (ocfs2_supports_indexed_dirs(osb))
2671 return ocfs2_fill_new_dir_dx(osb, handle, parent, inode, fe_bh,
2672 data_ac, meta_ac);
2673
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002674 return ocfs2_fill_new_dir_el(osb, handle, parent, inode, fe_bh,
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002675 data_ac, NULL);
2676}
2677
2678static int ocfs2_dx_dir_index_block(struct inode *dir,
2679 handle_t *handle,
2680 struct buffer_head **dx_leaves,
2681 int num_dx_leaves,
Mark Fashehe3a93c22009-02-17 15:29:35 -08002682 u32 *num_dx_entries,
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002683 struct buffer_head *dirent_bh)
2684{
Tao Ma0fba8132009-03-19 05:08:43 +08002685 int ret = 0, namelen, i;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002686 char *de_buf, *limit;
2687 struct ocfs2_dir_entry *de;
2688 struct buffer_head *dx_leaf_bh;
2689 struct ocfs2_dx_hinfo hinfo;
2690 u64 dirent_blk = dirent_bh->b_blocknr;
2691
2692 de_buf = dirent_bh->b_data;
2693 limit = de_buf + dir->i_sb->s_blocksize;
2694
2695 while (de_buf < limit) {
2696 de = (struct ocfs2_dir_entry *)de_buf;
2697
2698 namelen = de->name_len;
2699 if (!namelen || !de->inode)
2700 goto inc;
2701
2702 ocfs2_dx_dir_name_hash(dir, de->name, namelen, &hinfo);
2703
2704 i = ocfs2_dx_dir_hash_idx(OCFS2_SB(dir->i_sb), &hinfo);
2705 dx_leaf_bh = dx_leaves[i];
2706
2707 ret = __ocfs2_dx_dir_leaf_insert(dir, handle, &hinfo,
2708 dirent_blk, dx_leaf_bh);
2709 if (ret) {
2710 mlog_errno(ret);
2711 goto out;
2712 }
2713
Mark Fashehe3a93c22009-02-17 15:29:35 -08002714 *num_dx_entries = *num_dx_entries + 1;
2715
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002716inc:
2717 de_buf += le16_to_cpu(de->rec_len);
2718 }
2719
2720out:
2721 return ret;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002722}
Mark Fashehe7c17e42009-01-29 18:17:46 -08002723
2724/*
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08002725 * XXX: This expects dx_root_bh to already be part of the transaction.
2726 */
2727static void ocfs2_dx_dir_index_root_block(struct inode *dir,
2728 struct buffer_head *dx_root_bh,
2729 struct buffer_head *dirent_bh)
2730{
2731 char *de_buf, *limit;
2732 struct ocfs2_dx_root_block *dx_root;
2733 struct ocfs2_dir_entry *de;
2734 struct ocfs2_dx_hinfo hinfo;
2735 u64 dirent_blk = dirent_bh->b_blocknr;
2736
2737 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
2738
2739 de_buf = dirent_bh->b_data;
2740 limit = de_buf + dir->i_sb->s_blocksize;
2741
2742 while (de_buf < limit) {
2743 de = (struct ocfs2_dir_entry *)de_buf;
2744
2745 if (!de->name_len || !de->inode)
2746 goto inc;
2747
2748 ocfs2_dx_dir_name_hash(dir, de->name, de->name_len, &hinfo);
2749
Tao Maf1088d42011-02-23 22:30:23 +08002750 trace_ocfs2_dx_dir_index_root_block(
2751 (unsigned long long)dir->i_ino,
2752 hinfo.major_hash, hinfo.minor_hash,
2753 de->name_len, de->name,
2754 le16_to_cpu(dx_root->dr_entries.de_num_used));
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08002755
2756 ocfs2_dx_entry_list_insert(&dx_root->dr_entries, &hinfo,
2757 dirent_blk);
Mark Fashehe3a93c22009-02-17 15:29:35 -08002758
2759 le32_add_cpu(&dx_root->dr_num_entries, 1);
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08002760inc:
2761 de_buf += le16_to_cpu(de->rec_len);
2762 }
2763}
2764
2765/*
2766 * Count the number of inline directory entries in di_bh and compare
2767 * them against the number of entries we can hold in an inline dx root
2768 * block.
2769 */
2770static int ocfs2_new_dx_should_be_inline(struct inode *dir,
2771 struct buffer_head *di_bh)
2772{
2773 int dirent_count = 0;
2774 char *de_buf, *limit;
2775 struct ocfs2_dir_entry *de;
2776 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2777
2778 de_buf = di->id2.i_data.id_data;
2779 limit = de_buf + i_size_read(dir);
2780
2781 while (de_buf < limit) {
2782 de = (struct ocfs2_dir_entry *)de_buf;
2783
2784 if (de->name_len && de->inode)
2785 dirent_count++;
2786
2787 de_buf += le16_to_cpu(de->rec_len);
2788 }
2789
2790 /* We are careful to leave room for one extra record. */
2791 return dirent_count < ocfs2_dx_entries_per_root(dir->i_sb);
2792}
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002793
Mark Fasheh87d35a72008-12-10 17:36:25 -08002794/*
2795 * Expand rec_len of the rightmost dirent in a directory block so that it
2796 * contains the end of our valid space for dirents. We do this during
2797 * expansion from an inline directory to one with extents. The first dir block
2798 * in that case is taken from the inline data portion of the inode block.
2799 *
Mark Fashehe7c17e42009-01-29 18:17:46 -08002800 * This will also return the largest amount of contiguous space for a dirent
2801 * in the block. That value is *not* necessarily the last dirent, even after
2802 * expansion. The directory indexing code wants this value for free space
2803 * accounting. We do this here since we're already walking the entire dir
2804 * block.
2805 *
Mark Fasheh87d35a72008-12-10 17:36:25 -08002806 * We add the dir trailer if this filesystem wants it.
2807 */
Mark Fashehe7c17e42009-01-29 18:17:46 -08002808static unsigned int ocfs2_expand_last_dirent(char *start, unsigned int old_size,
2809 struct inode *dir)
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002810{
Mark Fashehe7c17e42009-01-29 18:17:46 -08002811 struct super_block *sb = dir->i_sb;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002812 struct ocfs2_dir_entry *de;
2813 struct ocfs2_dir_entry *prev_de;
2814 char *de_buf, *limit;
Mark Fasheh87d35a72008-12-10 17:36:25 -08002815 unsigned int new_size = sb->s_blocksize;
Mark Fashehe7c17e42009-01-29 18:17:46 -08002816 unsigned int bytes, this_hole;
2817 unsigned int largest_hole = 0;
Mark Fasheh87d35a72008-12-10 17:36:25 -08002818
Mark Fashehe7c17e42009-01-29 18:17:46 -08002819 if (ocfs2_new_dir_wants_trailer(dir))
Mark Fasheh87d35a72008-12-10 17:36:25 -08002820 new_size = ocfs2_dir_trailer_blk_off(sb);
2821
2822 bytes = new_size - old_size;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002823
2824 limit = start + old_size;
2825 de_buf = start;
2826 de = (struct ocfs2_dir_entry *)de_buf;
2827 do {
Mark Fashehe7c17e42009-01-29 18:17:46 -08002828 this_hole = ocfs2_figure_dirent_hole(de);
2829 if (this_hole > largest_hole)
2830 largest_hole = this_hole;
2831
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002832 prev_de = de;
2833 de_buf += le16_to_cpu(de->rec_len);
2834 de = (struct ocfs2_dir_entry *)de_buf;
2835 } while (de_buf < limit);
2836
2837 le16_add_cpu(&prev_de->rec_len, bytes);
Mark Fashehe7c17e42009-01-29 18:17:46 -08002838
2839 /* We need to double check this after modification of the final
2840 * dirent. */
2841 this_hole = ocfs2_figure_dirent_hole(prev_de);
2842 if (this_hole > largest_hole)
2843 largest_hole = this_hole;
2844
2845 if (largest_hole >= OCFS2_DIR_MIN_REC_LEN)
2846 return largest_hole;
2847 return 0;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002848}
2849
2850/*
2851 * We allocate enough clusters to fulfill "blocks_wanted", but set
2852 * i_size to exactly one block. Ocfs2_extend_dir() will handle the
2853 * rest automatically for us.
2854 *
2855 * *first_block_bh is a pointer to the 1st data block allocated to the
2856 * directory.
2857 */
2858static int ocfs2_expand_inline_dir(struct inode *dir, struct buffer_head *di_bh,
2859 unsigned int blocks_wanted,
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002860 struct ocfs2_dir_lookup_result *lookup,
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002861 struct buffer_head **first_block_bh)
2862{
Mark Fashehe3a93c22009-02-17 15:29:35 -08002863 u32 alloc, dx_alloc, bit_off, len, num_dx_entries = 0;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002864 struct super_block *sb = dir->i_sb;
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08002865 int ret, i, num_dx_leaves = 0, dx_inline = 0,
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002866 credits = ocfs2_inline_to_extents_credits(sb);
2867 u64 dx_insert_blkno, blkno,
2868 bytes = blocks_wanted << sb->s_blocksize_bits;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002869 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2870 struct ocfs2_inode_info *oi = OCFS2_I(dir);
Marcus Meissner5d446702011-05-05 10:44:11 -07002871 struct ocfs2_alloc_context *data_ac = NULL;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002872 struct ocfs2_alloc_context *meta_ac = NULL;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002873 struct buffer_head *dirdata_bh = NULL;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002874 struct buffer_head *dx_root_bh = NULL;
2875 struct buffer_head **dx_leaves = NULL;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002876 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2877 handle_t *handle;
Joel Beckerf99b9b72008-08-20 19:36:33 -07002878 struct ocfs2_extent_tree et;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002879 struct ocfs2_extent_tree dx_et;
2880 int did_quota = 0, bytes_allocated = 0;
Joel Beckerf99b9b72008-08-20 19:36:33 -07002881
Joel Becker5e404e92009-02-13 03:54:22 -08002882 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(dir), di_bh);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002883
2884 alloc = ocfs2_clusters_for_bytes(sb, bytes);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002885 dx_alloc = 0;
2886
Jan Karaedd45c02009-06-02 14:24:03 +02002887 down_write(&oi->ip_alloc_sem);
2888
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002889 if (ocfs2_supports_indexed_dirs(osb)) {
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002890 credits += ocfs2_add_dir_index_credits(sb);
2891
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08002892 dx_inline = ocfs2_new_dx_should_be_inline(dir, di_bh);
2893 if (!dx_inline) {
2894 /* Add one more cluster for an index leaf */
2895 dx_alloc++;
2896 dx_leaves = ocfs2_dx_dir_kmalloc_leaves(sb,
2897 &num_dx_leaves);
2898 if (!dx_leaves) {
2899 ret = -ENOMEM;
2900 mlog_errno(ret);
2901 goto out;
2902 }
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002903 }
2904
2905 /* This gets us the dx_root */
2906 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
2907 if (ret) {
2908 mlog_errno(ret);
2909 goto out;
2910 }
2911 }
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002912
2913 /*
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002914 * We should never need more than 2 clusters for the unindexed
2915 * tree - maximum dirent size is far less than one block. In
2916 * fact, the only time we'd need more than one cluster is if
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002917 * blocksize == clustersize and the dirent won't fit in the
2918 * extra space that the expansion to a single block gives. As
2919 * of today, that only happens on 4k/4k file systems.
2920 */
2921 BUG_ON(alloc > 2);
2922
Tao Ma035a5712009-04-07 07:40:57 +08002923 ret = ocfs2_reserve_clusters(osb, alloc + dx_alloc, &data_ac);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002924 if (ret) {
2925 mlog_errno(ret);
2926 goto out;
2927 }
2928
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002929 /*
Joe Perchesc78bad12008-02-03 17:33:42 +02002930 * Prepare for worst case allocation scenario of two separate
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002931 * extents in the unindexed tree.
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002932 */
2933 if (alloc == 2)
2934 credits += OCFS2_SUBALLOC_ALLOC;
2935
2936 handle = ocfs2_start_trans(osb, credits);
2937 if (IS_ERR(handle)) {
2938 ret = PTR_ERR(handle);
2939 mlog_errno(ret);
Jan Karaedd45c02009-06-02 14:24:03 +02002940 goto out;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002941 }
2942
Christoph Hellwig5dd40562010-03-03 09:05:00 -05002943 ret = dquot_alloc_space_nodirty(dir,
2944 ocfs2_clusters_to_bytes(osb->sb, alloc + dx_alloc));
2945 if (ret)
Jan Karaa90714c2008-10-09 19:38:40 +02002946 goto out_commit;
Jan Karaa90714c2008-10-09 19:38:40 +02002947 did_quota = 1;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002948
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08002949 if (ocfs2_supports_indexed_dirs(osb) && !dx_inline) {
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002950 /*
2951 * Allocate our index cluster first, to maximize the
2952 * possibility that unindexed leaves grow
2953 * contiguously.
2954 */
2955 ret = __ocfs2_dx_dir_new_cluster(dir, 0, handle, data_ac,
2956 dx_leaves, num_dx_leaves,
2957 &dx_insert_blkno);
2958 if (ret) {
2959 mlog_errno(ret);
2960 goto out_commit;
2961 }
2962 bytes_allocated += ocfs2_clusters_to_bytes(dir->i_sb, 1);
2963 }
2964
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002965 /*
2966 * Try to claim as many clusters as the bitmap can give though
2967 * if we only get one now, that's enough to continue. The rest
2968 * will be claimed after the conversion to extents.
2969 */
Mark Fasheh83f92312010-04-05 18:17:16 -07002970 if (ocfs2_dir_resv_allowed(osb))
2971 data_ac->ac_resv = &oi->ip_la_data_resv;
Joel Becker1ed9b772010-05-06 13:59:06 +08002972 ret = ocfs2_claim_clusters(handle, data_ac, 1, &bit_off, &len);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002973 if (ret) {
2974 mlog_errno(ret);
2975 goto out_commit;
2976 }
Mark Fasheh9b7895e2008-11-12 16:27:44 -08002977 bytes_allocated += ocfs2_clusters_to_bytes(dir->i_sb, 1);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002978
2979 /*
2980 * Operations are carefully ordered so that we set up the new
2981 * data block first. The conversion from inline data to
2982 * extents follows.
2983 */
2984 blkno = ocfs2_clusters_to_blocks(dir->i_sb, bit_off);
2985 dirdata_bh = sb_getblk(sb, blkno);
2986 if (!dirdata_bh) {
2987 ret = -EIO;
2988 mlog_errno(ret);
2989 goto out_commit;
2990 }
2991
Joel Becker8cb471e2009-02-10 20:00:41 -08002992 ocfs2_set_new_buffer_uptodate(INODE_CACHE(dir), dirdata_bh);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002993
Joel Becker0cf2f762009-02-12 16:41:25 -08002994 ret = ocfs2_journal_access_db(handle, INODE_CACHE(dir), dirdata_bh,
Joel Becker13723d02008-10-17 19:25:01 -07002995 OCFS2_JOURNAL_ACCESS_CREATE);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07002996 if (ret) {
2997 mlog_errno(ret);
2998 goto out_commit;
2999 }
3000
3001 memcpy(dirdata_bh->b_data, di->id2.i_data.id_data, i_size_read(dir));
3002 memset(dirdata_bh->b_data + i_size_read(dir), 0,
3003 sb->s_blocksize - i_size_read(dir));
Mark Fashehe7c17e42009-01-29 18:17:46 -08003004 i = ocfs2_expand_last_dirent(dirdata_bh->b_data, i_size_read(dir), dir);
3005 if (ocfs2_new_dir_wants_trailer(dir)) {
3006 /*
3007 * Prepare the dir trailer up front. It will otherwise look
3008 * like a valid dirent. Even if inserting the index fails
3009 * (unlikely), then all we'll have done is given first dir
3010 * block a small amount of fragmentation.
3011 */
3012 ocfs2_init_dir_trailer(dir, dirdata_bh, i);
3013 }
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003014
Joel Beckerec20cec2010-03-19 14:13:52 -07003015 ocfs2_journal_dirty(handle, dirdata_bh);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003016
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08003017 if (ocfs2_supports_indexed_dirs(osb) && !dx_inline) {
3018 /*
3019 * Dx dirs with an external cluster need to do this up
3020 * front. Inline dx root's get handled later, after
Mark Fashehe3a93c22009-02-17 15:29:35 -08003021 * we've allocated our root block. We get passed back
3022 * a total number of items so that dr_num_entries can
3023 * be correctly set once the dx_root has been
3024 * allocated.
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08003025 */
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003026 ret = ocfs2_dx_dir_index_block(dir, handle, dx_leaves,
Mark Fashehe3a93c22009-02-17 15:29:35 -08003027 num_dx_leaves, &num_dx_entries,
3028 dirdata_bh);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003029 if (ret) {
3030 mlog_errno(ret);
3031 goto out_commit;
3032 }
3033 }
3034
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003035 /*
3036 * Set extent, i_size, etc on the directory. After this, the
3037 * inode should contain the same exact dirents as before and
3038 * be fully accessible from system calls.
3039 *
3040 * We let the later dirent insert modify c/mtime - to the user
3041 * the data hasn't changed.
3042 */
Joel Becker0cf2f762009-02-12 16:41:25 -08003043 ret = ocfs2_journal_access_di(handle, INODE_CACHE(dir), di_bh,
Joel Becker13723d02008-10-17 19:25:01 -07003044 OCFS2_JOURNAL_ACCESS_CREATE);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003045 if (ret) {
3046 mlog_errno(ret);
3047 goto out_commit;
3048 }
3049
3050 spin_lock(&oi->ip_lock);
3051 oi->ip_dyn_features &= ~OCFS2_INLINE_DATA_FL;
3052 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
3053 spin_unlock(&oi->ip_lock);
3054
3055 ocfs2_dinode_new_extent_list(dir, di);
3056
3057 i_size_write(dir, sb->s_blocksize);
3058 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3059
3060 di->i_size = cpu_to_le64(sb->s_blocksize);
3061 di->i_ctime = di->i_mtime = cpu_to_le64(dir->i_ctime.tv_sec);
3062 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(dir->i_ctime.tv_nsec);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003063
3064 /*
3065 * This should never fail as our extent list is empty and all
3066 * related blocks have been journaled already.
3067 */
Joel Beckercc79d8c2009-02-13 03:24:43 -08003068 ret = ocfs2_insert_extent(handle, &et, 0, blkno, len,
Joel Beckerf99b9b72008-08-20 19:36:33 -07003069 0, NULL);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003070 if (ret) {
3071 mlog_errno(ret);
Tao Ma83cab532008-08-21 14:14:27 +08003072 goto out_commit;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003073 }
3074
Mark Fasheh9780eb62008-08-05 11:32:46 -07003075 /*
3076 * Set i_blocks after the extent insert for the most up to
3077 * date ip_clusters value.
3078 */
3079 dir->i_blocks = ocfs2_inode_sector_count(dir);
3080
Joel Beckerec20cec2010-03-19 14:13:52 -07003081 ocfs2_journal_dirty(handle, di_bh);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003082
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003083 if (ocfs2_supports_indexed_dirs(osb)) {
3084 ret = ocfs2_dx_dir_attach_index(osb, handle, dir, di_bh,
Mark Fashehe7c17e42009-01-29 18:17:46 -08003085 dirdata_bh, meta_ac, dx_inline,
Mark Fashehe3a93c22009-02-17 15:29:35 -08003086 num_dx_entries, &dx_root_bh);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003087 if (ret) {
3088 mlog_errno(ret);
3089 goto out_commit;
3090 }
3091
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08003092 if (dx_inline) {
3093 ocfs2_dx_dir_index_root_block(dir, dx_root_bh,
3094 dirdata_bh);
3095 } else {
Joel Becker5e404e92009-02-13 03:54:22 -08003096 ocfs2_init_dx_root_extent_tree(&dx_et,
3097 INODE_CACHE(dir),
3098 dx_root_bh);
Joel Beckercc79d8c2009-02-13 03:24:43 -08003099 ret = ocfs2_insert_extent(handle, &dx_et, 0,
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08003100 dx_insert_blkno, 1, 0, NULL);
3101 if (ret)
3102 mlog_errno(ret);
3103 }
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003104 }
3105
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003106 /*
3107 * We asked for two clusters, but only got one in the 1st
3108 * pass. Claim the 2nd cluster as a separate extent.
3109 */
3110 if (alloc > len) {
Joel Becker1ed9b772010-05-06 13:59:06 +08003111 ret = ocfs2_claim_clusters(handle, data_ac, 1, &bit_off,
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003112 &len);
3113 if (ret) {
3114 mlog_errno(ret);
3115 goto out_commit;
3116 }
3117 blkno = ocfs2_clusters_to_blocks(dir->i_sb, bit_off);
3118
Joel Beckercc79d8c2009-02-13 03:24:43 -08003119 ret = ocfs2_insert_extent(handle, &et, 1,
Joel Beckerf99b9b72008-08-20 19:36:33 -07003120 blkno, len, 0, NULL);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003121 if (ret) {
3122 mlog_errno(ret);
Tao Ma83cab532008-08-21 14:14:27 +08003123 goto out_commit;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003124 }
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003125 bytes_allocated += ocfs2_clusters_to_bytes(dir->i_sb, 1);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003126 }
3127
3128 *first_block_bh = dirdata_bh;
3129 dirdata_bh = NULL;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003130 if (ocfs2_supports_indexed_dirs(osb)) {
3131 unsigned int off;
3132
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08003133 if (!dx_inline) {
3134 /*
3135 * We need to return the correct block within the
3136 * cluster which should hold our entry.
3137 */
3138 off = ocfs2_dx_dir_hash_idx(OCFS2_SB(dir->i_sb),
3139 &lookup->dl_hinfo);
3140 get_bh(dx_leaves[off]);
3141 lookup->dl_dx_leaf_bh = dx_leaves[off];
3142 }
3143 lookup->dl_dx_root_bh = dx_root_bh;
3144 dx_root_bh = NULL;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003145 }
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003146
3147out_commit:
Jan Karaa90714c2008-10-09 19:38:40 +02003148 if (ret < 0 && did_quota)
Christoph Hellwig5dd40562010-03-03 09:05:00 -05003149 dquot_free_space_nodirty(dir, bytes_allocated);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003150
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003151 ocfs2_commit_trans(osb, handle);
3152
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003153out:
Jan Karaedd45c02009-06-02 14:24:03 +02003154 up_write(&oi->ip_alloc_sem);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003155 if (data_ac)
3156 ocfs2_free_alloc_context(data_ac);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003157 if (meta_ac)
3158 ocfs2_free_alloc_context(meta_ac);
3159
3160 if (dx_leaves) {
3161 for (i = 0; i < num_dx_leaves; i++)
3162 brelse(dx_leaves[i]);
3163 kfree(dx_leaves);
3164 }
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003165
3166 brelse(dirdata_bh);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003167 brelse(dx_root_bh);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003168
3169 return ret;
3170}
3171
Mark Fashehccd979b2005-12-15 14:31:24 -08003172/* returns a bh of the 1st new block in the allocation. */
Mark Fasheh316f4b92007-09-07 18:21:26 -07003173static int ocfs2_do_extend_dir(struct super_block *sb,
3174 handle_t *handle,
3175 struct inode *dir,
3176 struct buffer_head *parent_fe_bh,
3177 struct ocfs2_alloc_context *data_ac,
3178 struct ocfs2_alloc_context *meta_ac,
3179 struct buffer_head **new_bh)
Mark Fashehccd979b2005-12-15 14:31:24 -08003180{
3181 int status;
Jan Karaa90714c2008-10-09 19:38:40 +02003182 int extend, did_quota = 0;
Mark Fasheh8110b072007-03-22 16:53:23 -07003183 u64 p_blkno, v_blkno;
Mark Fashehccd979b2005-12-15 14:31:24 -08003184
3185 spin_lock(&OCFS2_I(dir)->ip_lock);
3186 extend = (i_size_read(dir) == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters));
3187 spin_unlock(&OCFS2_I(dir)->ip_lock);
3188
3189 if (extend) {
Mark Fashehdcd05382007-01-16 11:32:23 -08003190 u32 offset = OCFS2_I(dir)->ip_clusters;
3191
Christoph Hellwig5dd40562010-03-03 09:05:00 -05003192 status = dquot_alloc_space_nodirty(dir,
3193 ocfs2_clusters_to_bytes(sb, 1));
3194 if (status)
Jan Karaa90714c2008-10-09 19:38:40 +02003195 goto bail;
Jan Karaa90714c2008-10-09 19:38:40 +02003196 did_quota = 1;
3197
Tao Ma0eb8d472008-08-18 17:38:45 +08003198 status = ocfs2_add_inode_data(OCFS2_SB(sb), dir, &offset,
3199 1, 0, parent_fe_bh, handle,
3200 data_ac, meta_ac, NULL);
Mark Fashehccd979b2005-12-15 14:31:24 -08003201 BUG_ON(status == -EAGAIN);
3202 if (status < 0) {
3203 mlog_errno(status);
3204 goto bail;
3205 }
3206 }
3207
Mark Fasheh8110b072007-03-22 16:53:23 -07003208 v_blkno = ocfs2_blocks_for_bytes(sb, i_size_read(dir));
3209 status = ocfs2_extent_map_get_blocks(dir, v_blkno, &p_blkno, NULL, NULL);
Mark Fashehccd979b2005-12-15 14:31:24 -08003210 if (status < 0) {
3211 mlog_errno(status);
3212 goto bail;
3213 }
3214
3215 *new_bh = sb_getblk(sb, p_blkno);
3216 if (!*new_bh) {
3217 status = -EIO;
3218 mlog_errno(status);
3219 goto bail;
3220 }
3221 status = 0;
3222bail:
Jan Karaa90714c2008-10-09 19:38:40 +02003223 if (did_quota && status < 0)
Christoph Hellwig5dd40562010-03-03 09:05:00 -05003224 dquot_free_space_nodirty(dir, ocfs2_clusters_to_bytes(sb, 1));
Mark Fashehccd979b2005-12-15 14:31:24 -08003225 return status;
3226}
3227
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003228/*
3229 * Assumes you already have a cluster lock on the directory.
3230 *
3231 * 'blocks_wanted' is only used if we have an inline directory which
3232 * is to be turned into an extent based one. The size of the dirent to
3233 * insert might be larger than the space gained by growing to just one
3234 * block, so we may have to grow the inode by two blocks in that case.
Mark Fashehe7c17e42009-01-29 18:17:46 -08003235 *
3236 * If the directory is already indexed, dx_root_bh must be provided.
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003237 */
Mark Fashehccd979b2005-12-15 14:31:24 -08003238static int ocfs2_extend_dir(struct ocfs2_super *osb,
3239 struct inode *dir,
3240 struct buffer_head *parent_fe_bh,
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003241 unsigned int blocks_wanted,
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003242 struct ocfs2_dir_lookup_result *lookup,
Mark Fashehccd979b2005-12-15 14:31:24 -08003243 struct buffer_head **new_de_bh)
3244{
3245 int status = 0;
Joel Beckeree19a772007-03-28 18:27:07 -07003246 int credits, num_free_extents, drop_alloc_sem = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08003247 loff_t dir_i_size;
3248 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
Tao Ma811f9332008-08-18 17:38:43 +08003249 struct ocfs2_extent_list *el = &fe->id2.i_list;
Mark Fashehccd979b2005-12-15 14:31:24 -08003250 struct ocfs2_alloc_context *data_ac = NULL;
3251 struct ocfs2_alloc_context *meta_ac = NULL;
Mark Fasheh1fabe142006-10-09 18:11:45 -07003252 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08003253 struct buffer_head *new_bh = NULL;
3254 struct ocfs2_dir_entry * de;
3255 struct super_block *sb = osb->sb;
Joel Beckerf99b9b72008-08-20 19:36:33 -07003256 struct ocfs2_extent_tree et;
Mark Fashehe7c17e42009-01-29 18:17:46 -08003257 struct buffer_head *dx_root_bh = lookup->dl_dx_root_bh;
Mark Fashehccd979b2005-12-15 14:31:24 -08003258
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003259 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
Mark Fashehe7c17e42009-01-29 18:17:46 -08003260 /*
3261 * This would be a code error as an inline directory should
3262 * never have an index root.
3263 */
3264 BUG_ON(dx_root_bh);
3265
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003266 status = ocfs2_expand_inline_dir(dir, parent_fe_bh,
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003267 blocks_wanted, lookup,
3268 &new_bh);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003269 if (status) {
3270 mlog_errno(status);
3271 goto bail;
3272 }
3273
Mark Fashehe7c17e42009-01-29 18:17:46 -08003274 /* Expansion from inline to an indexed directory will
3275 * have given us this. */
3276 dx_root_bh = lookup->dl_dx_root_bh;
3277
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003278 if (blocks_wanted == 1) {
3279 /*
3280 * If the new dirent will fit inside the space
3281 * created by pushing out to one block, then
3282 * we can complete the operation
3283 * here. Otherwise we have to expand i_size
3284 * and format the 2nd block below.
3285 */
3286 BUG_ON(new_bh == NULL);
3287 goto bail_bh;
3288 }
3289
3290 /*
3291 * Get rid of 'new_bh' - we want to format the 2nd
3292 * data block and return that instead.
3293 */
3294 brelse(new_bh);
3295 new_bh = NULL;
3296
Jan Karaedd45c02009-06-02 14:24:03 +02003297 down_write(&OCFS2_I(dir)->ip_alloc_sem);
3298 drop_alloc_sem = 1;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003299 dir_i_size = i_size_read(dir);
3300 credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS;
3301 goto do_extend;
3302 }
3303
Jan Karaedd45c02009-06-02 14:24:03 +02003304 down_write(&OCFS2_I(dir)->ip_alloc_sem);
3305 drop_alloc_sem = 1;
Mark Fashehccd979b2005-12-15 14:31:24 -08003306 dir_i_size = i_size_read(dir);
Tao Maf1088d42011-02-23 22:30:23 +08003307 trace_ocfs2_extend_dir((unsigned long long)OCFS2_I(dir)->ip_blkno,
3308 dir_i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -08003309
Mark Fashehccd979b2005-12-15 14:31:24 -08003310 /* dir->i_size is always block aligned. */
3311 spin_lock(&OCFS2_I(dir)->ip_lock);
3312 if (dir_i_size == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters)) {
3313 spin_unlock(&OCFS2_I(dir)->ip_lock);
Joel Becker5e404e92009-02-13 03:54:22 -08003314 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(dir),
3315 parent_fe_bh);
Joel Becker3d03a302009-02-12 17:49:26 -08003316 num_free_extents = ocfs2_num_free_extents(osb, &et);
Mark Fashehccd979b2005-12-15 14:31:24 -08003317 if (num_free_extents < 0) {
3318 status = num_free_extents;
3319 mlog_errno(status);
3320 goto bail;
3321 }
3322
3323 if (!num_free_extents) {
Tao Ma811f9332008-08-18 17:38:43 +08003324 status = ocfs2_reserve_new_metadata(osb, el, &meta_ac);
Mark Fashehccd979b2005-12-15 14:31:24 -08003325 if (status < 0) {
3326 if (status != -ENOSPC)
3327 mlog_errno(status);
3328 goto bail;
3329 }
3330 }
3331
Mark Fashehda5cbf22006-10-06 18:34:35 -07003332 status = ocfs2_reserve_clusters(osb, 1, &data_ac);
Mark Fashehccd979b2005-12-15 14:31:24 -08003333 if (status < 0) {
3334 if (status != -ENOSPC)
3335 mlog_errno(status);
3336 goto bail;
3337 }
3338
Mark Fasheh83f92312010-04-05 18:17:16 -07003339 if (ocfs2_dir_resv_allowed(osb))
3340 data_ac->ac_resv = &OCFS2_I(dir)->ip_la_data_resv;
Mark Fashehe3b4a972009-12-07 13:16:07 -08003341
Tao Ma811f9332008-08-18 17:38:43 +08003342 credits = ocfs2_calc_extend_credits(sb, el, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08003343 } else {
3344 spin_unlock(&OCFS2_I(dir)->ip_lock);
3345 credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS;
3346 }
3347
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003348do_extend:
Mark Fashehe7c17e42009-01-29 18:17:46 -08003349 if (ocfs2_dir_indexed(dir))
3350 credits++; /* For attaching the new dirent block to the
3351 * dx_root */
3352
Mark Fasheh65eff9c2006-10-09 17:26:22 -07003353 handle = ocfs2_start_trans(osb, credits);
Mark Fashehccd979b2005-12-15 14:31:24 -08003354 if (IS_ERR(handle)) {
3355 status = PTR_ERR(handle);
3356 handle = NULL;
3357 mlog_errno(status);
3358 goto bail;
3359 }
3360
3361 status = ocfs2_do_extend_dir(osb->sb, handle, dir, parent_fe_bh,
3362 data_ac, meta_ac, &new_bh);
3363 if (status < 0) {
3364 mlog_errno(status);
3365 goto bail;
3366 }
3367
Joel Becker8cb471e2009-02-10 20:00:41 -08003368 ocfs2_set_new_buffer_uptodate(INODE_CACHE(dir), new_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08003369
Joel Becker0cf2f762009-02-12 16:41:25 -08003370 status = ocfs2_journal_access_db(handle, INODE_CACHE(dir), new_bh,
Joel Becker13723d02008-10-17 19:25:01 -07003371 OCFS2_JOURNAL_ACCESS_CREATE);
Mark Fashehccd979b2005-12-15 14:31:24 -08003372 if (status < 0) {
3373 mlog_errno(status);
3374 goto bail;
3375 }
3376 memset(new_bh->b_data, 0, sb->s_blocksize);
Mark Fasheh87d35a72008-12-10 17:36:25 -08003377
Mark Fashehccd979b2005-12-15 14:31:24 -08003378 de = (struct ocfs2_dir_entry *) new_bh->b_data;
3379 de->inode = 0;
Mark Fashehe7c17e42009-01-29 18:17:46 -08003380 if (ocfs2_supports_dir_trailer(dir)) {
Mark Fasheh87d35a72008-12-10 17:36:25 -08003381 de->rec_len = cpu_to_le16(ocfs2_dir_trailer_blk_off(sb));
Mark Fashehe7c17e42009-01-29 18:17:46 -08003382
3383 ocfs2_init_dir_trailer(dir, new_bh, le16_to_cpu(de->rec_len));
3384
3385 if (ocfs2_dir_indexed(dir)) {
3386 status = ocfs2_dx_dir_link_trailer(dir, handle,
3387 dx_root_bh, new_bh);
3388 if (status) {
3389 mlog_errno(status);
3390 goto bail;
3391 }
3392 }
Mark Fasheh87d35a72008-12-10 17:36:25 -08003393 } else {
3394 de->rec_len = cpu_to_le16(sb->s_blocksize);
3395 }
Joel Beckerec20cec2010-03-19 14:13:52 -07003396 ocfs2_journal_dirty(handle, new_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08003397
3398 dir_i_size += dir->i_sb->s_blocksize;
3399 i_size_write(dir, dir_i_size);
Mark Fasheh8110b072007-03-22 16:53:23 -07003400 dir->i_blocks = ocfs2_inode_sector_count(dir);
Mark Fashehccd979b2005-12-15 14:31:24 -08003401 status = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
3402 if (status < 0) {
3403 mlog_errno(status);
3404 goto bail;
3405 }
3406
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003407bail_bh:
Mark Fashehccd979b2005-12-15 14:31:24 -08003408 *new_de_bh = new_bh;
3409 get_bh(*new_de_bh);
3410bail:
3411 if (handle)
Mark Fasheh02dc1af2006-10-09 16:48:10 -07003412 ocfs2_commit_trans(osb, handle);
Jan Karaedd45c02009-06-02 14:24:03 +02003413 if (drop_alloc_sem)
3414 up_write(&OCFS2_I(dir)->ip_alloc_sem);
Mark Fashehccd979b2005-12-15 14:31:24 -08003415
3416 if (data_ac)
3417 ocfs2_free_alloc_context(data_ac);
3418 if (meta_ac)
3419 ocfs2_free_alloc_context(meta_ac);
3420
Mark Fasheha81cb882008-10-07 14:25:16 -07003421 brelse(new_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08003422
Mark Fashehccd979b2005-12-15 14:31:24 -08003423 return status;
3424}
3425
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003426static int ocfs2_find_dir_space_id(struct inode *dir, struct buffer_head *di_bh,
3427 const char *name, int namelen,
3428 struct buffer_head **ret_de_bh,
3429 unsigned int *blocks_wanted)
3430{
3431 int ret;
Mark Fasheh87d35a72008-12-10 17:36:25 -08003432 struct super_block *sb = dir->i_sb;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003433 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3434 struct ocfs2_dir_entry *de, *last_de = NULL;
3435 char *de_buf, *limit;
3436 unsigned long offset = 0;
Mark Fasheh87d35a72008-12-10 17:36:25 -08003437 unsigned int rec_len, new_rec_len, free_space = dir->i_sb->s_blocksize;
3438
3439 /*
3440 * This calculates how many free bytes we'd have in block zero, should
3441 * this function force expansion to an extent tree.
3442 */
Mark Fashehe7c17e42009-01-29 18:17:46 -08003443 if (ocfs2_new_dir_wants_trailer(dir))
Mark Fasheh87d35a72008-12-10 17:36:25 -08003444 free_space = ocfs2_dir_trailer_blk_off(sb) - i_size_read(dir);
3445 else
3446 free_space = dir->i_sb->s_blocksize - i_size_read(dir);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003447
3448 de_buf = di->id2.i_data.id_data;
3449 limit = de_buf + i_size_read(dir);
3450 rec_len = OCFS2_DIR_REC_LEN(namelen);
3451
3452 while (de_buf < limit) {
3453 de = (struct ocfs2_dir_entry *)de_buf;
3454
3455 if (!ocfs2_check_dir_entry(dir, de, di_bh, offset)) {
3456 ret = -ENOENT;
3457 goto out;
3458 }
3459 if (ocfs2_match(namelen, name, de)) {
3460 ret = -EEXIST;
3461 goto out;
3462 }
Mark Fasheh87d35a72008-12-10 17:36:25 -08003463 /*
3464 * No need to check for a trailing dirent record here as
3465 * they're not used for inline dirs.
3466 */
3467
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003468 if (ocfs2_dirent_would_fit(de, rec_len)) {
3469 /* Ok, we found a spot. Return this bh and let
3470 * the caller actually fill it in. */
3471 *ret_de_bh = di_bh;
3472 get_bh(*ret_de_bh);
3473 ret = 0;
3474 goto out;
3475 }
3476
3477 last_de = de;
3478 de_buf += le16_to_cpu(de->rec_len);
3479 offset += le16_to_cpu(de->rec_len);
3480 }
3481
3482 /*
3483 * We're going to require expansion of the directory - figure
3484 * out how many blocks we'll need so that a place for the
3485 * dirent can be found.
3486 */
3487 *blocks_wanted = 1;
Mark Fasheh87d35a72008-12-10 17:36:25 -08003488 new_rec_len = le16_to_cpu(last_de->rec_len) + free_space;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003489 if (new_rec_len < (rec_len + OCFS2_DIR_REC_LEN(last_de->name_len)))
3490 *blocks_wanted = 2;
3491
3492 ret = -ENOSPC;
3493out:
3494 return ret;
3495}
3496
3497static int ocfs2_find_dir_space_el(struct inode *dir, const char *name,
3498 int namelen, struct buffer_head **ret_de_bh)
Mark Fashehccd979b2005-12-15 14:31:24 -08003499{
3500 unsigned long offset;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003501 struct buffer_head *bh = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08003502 unsigned short rec_len;
Mark Fashehccd979b2005-12-15 14:31:24 -08003503 struct ocfs2_dir_entry *de;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003504 struct super_block *sb = dir->i_sb;
Mark Fashehccd979b2005-12-15 14:31:24 -08003505 int status;
Mark Fasheh87d35a72008-12-10 17:36:25 -08003506 int blocksize = dir->i_sb->s_blocksize;
Mark Fashehccd979b2005-12-15 14:31:24 -08003507
Joel Beckera22305c2008-11-13 14:49:17 -08003508 status = ocfs2_read_dir_block(dir, 0, &bh, 0);
3509 if (status) {
Mark Fashehccd979b2005-12-15 14:31:24 -08003510 mlog_errno(status);
3511 goto bail;
3512 }
3513
3514 rec_len = OCFS2_DIR_REC_LEN(namelen);
3515 offset = 0;
3516 de = (struct ocfs2_dir_entry *) bh->b_data;
3517 while (1) {
3518 if ((char *)de >= sb->s_blocksize + bh->b_data) {
3519 brelse(bh);
3520 bh = NULL;
3521
3522 if (i_size_read(dir) <= offset) {
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003523 /*
3524 * Caller will have to expand this
3525 * directory.
3526 */
3527 status = -ENOSPC;
Mark Fashehccd979b2005-12-15 14:31:24 -08003528 goto bail;
3529 }
Joel Beckera22305c2008-11-13 14:49:17 -08003530 status = ocfs2_read_dir_block(dir,
3531 offset >> sb->s_blocksize_bits,
3532 &bh, 0);
3533 if (status) {
Mark Fashehccd979b2005-12-15 14:31:24 -08003534 mlog_errno(status);
3535 goto bail;
3536 }
3537 /* move to next block */
3538 de = (struct ocfs2_dir_entry *) bh->b_data;
3539 }
3540 if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
3541 status = -ENOENT;
3542 goto bail;
3543 }
3544 if (ocfs2_match(namelen, name, de)) {
3545 status = -EEXIST;
3546 goto bail;
3547 }
Mark Fasheh87d35a72008-12-10 17:36:25 -08003548
3549 if (ocfs2_skip_dir_trailer(dir, de, offset % blocksize,
3550 blocksize))
3551 goto next;
3552
Mark Fasheh8553cf42007-09-13 16:29:01 -07003553 if (ocfs2_dirent_would_fit(de, rec_len)) {
Mark Fashehccd979b2005-12-15 14:31:24 -08003554 /* Ok, we found a spot. Return this bh and let
3555 * the caller actually fill it in. */
3556 *ret_de_bh = bh;
3557 get_bh(*ret_de_bh);
3558 status = 0;
3559 goto bail;
3560 }
Mark Fasheh87d35a72008-12-10 17:36:25 -08003561next:
Mark Fashehccd979b2005-12-15 14:31:24 -08003562 offset += le16_to_cpu(de->rec_len);
3563 de = (struct ocfs2_dir_entry *)((char *) de + le16_to_cpu(de->rec_len));
3564 }
3565
3566 status = 0;
3567bail:
Mark Fasheha81cb882008-10-07 14:25:16 -07003568 brelse(bh);
Tao Mac1e8d352011-03-07 16:43:21 +08003569 if (status)
3570 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08003571
Mark Fashehccd979b2005-12-15 14:31:24 -08003572 return status;
3573}
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07003574
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003575static int dx_leaf_sort_cmp(const void *a, const void *b)
3576{
3577 const struct ocfs2_dx_entry *entry1 = a;
3578 const struct ocfs2_dx_entry *entry2 = b;
3579 u32 major_hash1 = le32_to_cpu(entry1->dx_major_hash);
3580 u32 major_hash2 = le32_to_cpu(entry2->dx_major_hash);
3581 u32 minor_hash1 = le32_to_cpu(entry1->dx_minor_hash);
3582 u32 minor_hash2 = le32_to_cpu(entry2->dx_minor_hash);
3583
3584 if (major_hash1 > major_hash2)
3585 return 1;
3586 if (major_hash1 < major_hash2)
3587 return -1;
3588
3589 /*
3590 * It is not strictly necessary to sort by minor
3591 */
3592 if (minor_hash1 > minor_hash2)
3593 return 1;
3594 if (minor_hash1 < minor_hash2)
3595 return -1;
3596 return 0;
3597}
3598
3599static void dx_leaf_sort_swap(void *a, void *b, int size)
3600{
3601 struct ocfs2_dx_entry *entry1 = a;
3602 struct ocfs2_dx_entry *entry2 = b;
3603 struct ocfs2_dx_entry tmp;
3604
3605 BUG_ON(size != sizeof(*entry1));
3606
3607 tmp = *entry1;
3608 *entry1 = *entry2;
3609 *entry2 = tmp;
3610}
3611
3612static int ocfs2_dx_leaf_same_major(struct ocfs2_dx_leaf *dx_leaf)
3613{
3614 struct ocfs2_dx_entry_list *dl_list = &dx_leaf->dl_list;
3615 int i, num = le16_to_cpu(dl_list->de_num_used);
3616
3617 for (i = 0; i < (num - 1); i++) {
3618 if (le32_to_cpu(dl_list->de_entries[i].dx_major_hash) !=
3619 le32_to_cpu(dl_list->de_entries[i + 1].dx_major_hash))
3620 return 0;
3621 }
3622
3623 return 1;
3624}
3625
3626/*
3627 * Find the optimal value to split this leaf on. This expects the leaf
3628 * entries to be in sorted order.
3629 *
3630 * leaf_cpos is the cpos of the leaf we're splitting. insert_hash is
3631 * the hash we want to insert.
3632 *
3633 * This function is only concerned with the major hash - that which
3634 * determines which cluster an item belongs to.
3635 */
3636static int ocfs2_dx_dir_find_leaf_split(struct ocfs2_dx_leaf *dx_leaf,
3637 u32 leaf_cpos, u32 insert_hash,
3638 u32 *split_hash)
3639{
3640 struct ocfs2_dx_entry_list *dl_list = &dx_leaf->dl_list;
3641 int i, num_used = le16_to_cpu(dl_list->de_num_used);
3642 int allsame;
3643
3644 /*
3645 * There's a couple rare, but nasty corner cases we have to
3646 * check for here. All of them involve a leaf where all value
3647 * have the same hash, which is what we look for first.
3648 *
3649 * Most of the time, all of the above is false, and we simply
3650 * pick the median value for a split.
3651 */
3652 allsame = ocfs2_dx_leaf_same_major(dx_leaf);
3653 if (allsame) {
3654 u32 val = le32_to_cpu(dl_list->de_entries[0].dx_major_hash);
3655
3656 if (val == insert_hash) {
3657 /*
3658 * No matter where we would choose to split,
3659 * the new entry would want to occupy the same
3660 * block as these. Since there's no space left
3661 * in their existing block, we know there
3662 * won't be space after the split.
3663 */
3664 return -ENOSPC;
3665 }
3666
3667 if (val == leaf_cpos) {
3668 /*
3669 * Because val is the same as leaf_cpos (which
3670 * is the smallest value this leaf can have),
3671 * yet is not equal to insert_hash, then we
3672 * know that insert_hash *must* be larger than
3673 * val (and leaf_cpos). At least cpos+1 in value.
3674 *
3675 * We also know then, that there cannot be an
3676 * adjacent extent (otherwise we'd be looking
3677 * at it). Choosing this value gives us a
3678 * chance to get some contiguousness.
3679 */
3680 *split_hash = leaf_cpos + 1;
3681 return 0;
3682 }
3683
3684 if (val > insert_hash) {
3685 /*
3686 * val can not be the same as insert hash, and
3687 * also must be larger than leaf_cpos. Also,
3688 * we know that there can't be a leaf between
3689 * cpos and val, otherwise the entries with
3690 * hash 'val' would be there.
3691 */
3692 *split_hash = val;
3693 return 0;
3694 }
3695
3696 *split_hash = insert_hash;
3697 return 0;
3698 }
3699
3700 /*
3701 * Since the records are sorted and the checks above
3702 * guaranteed that not all records in this block are the same,
3703 * we simple travel forward, from the median, and pick the 1st
3704 * record whose value is larger than leaf_cpos.
3705 */
3706 for (i = (num_used / 2); i < num_used; i++)
3707 if (le32_to_cpu(dl_list->de_entries[i].dx_major_hash) >
3708 leaf_cpos)
3709 break;
3710
3711 BUG_ON(i == num_used); /* Should be impossible */
3712 *split_hash = le32_to_cpu(dl_list->de_entries[i].dx_major_hash);
3713 return 0;
3714}
3715
3716/*
3717 * Transfer all entries in orig_dx_leaves whose major hash is equal to or
3718 * larger than split_hash into new_dx_leaves. We use a temporary
3719 * buffer (tmp_dx_leaf) to make the changes to the original leaf blocks.
3720 *
3721 * Since the block offset inside a leaf (cluster) is a constant mask
3722 * of minor_hash, we can optimize - an item at block offset X within
3723 * the original cluster, will be at offset X within the new cluster.
3724 */
3725static void ocfs2_dx_dir_transfer_leaf(struct inode *dir, u32 split_hash,
3726 handle_t *handle,
3727 struct ocfs2_dx_leaf *tmp_dx_leaf,
3728 struct buffer_head **orig_dx_leaves,
3729 struct buffer_head **new_dx_leaves,
3730 int num_dx_leaves)
3731{
3732 int i, j, num_used;
3733 u32 major_hash;
3734 struct ocfs2_dx_leaf *orig_dx_leaf, *new_dx_leaf;
3735 struct ocfs2_dx_entry_list *orig_list, *new_list, *tmp_list;
3736 struct ocfs2_dx_entry *dx_entry;
3737
3738 tmp_list = &tmp_dx_leaf->dl_list;
3739
3740 for (i = 0; i < num_dx_leaves; i++) {
3741 orig_dx_leaf = (struct ocfs2_dx_leaf *) orig_dx_leaves[i]->b_data;
3742 orig_list = &orig_dx_leaf->dl_list;
3743 new_dx_leaf = (struct ocfs2_dx_leaf *) new_dx_leaves[i]->b_data;
3744 new_list = &new_dx_leaf->dl_list;
3745
3746 num_used = le16_to_cpu(orig_list->de_num_used);
3747
3748 memcpy(tmp_dx_leaf, orig_dx_leaf, dir->i_sb->s_blocksize);
3749 tmp_list->de_num_used = cpu_to_le16(0);
3750 memset(&tmp_list->de_entries, 0, sizeof(*dx_entry)*num_used);
3751
3752 for (j = 0; j < num_used; j++) {
3753 dx_entry = &orig_list->de_entries[j];
3754 major_hash = le32_to_cpu(dx_entry->dx_major_hash);
3755 if (major_hash >= split_hash)
3756 ocfs2_dx_dir_leaf_insert_tail(new_dx_leaf,
3757 dx_entry);
3758 else
3759 ocfs2_dx_dir_leaf_insert_tail(tmp_dx_leaf,
3760 dx_entry);
3761 }
3762 memcpy(orig_dx_leaf, tmp_dx_leaf, dir->i_sb->s_blocksize);
3763
3764 ocfs2_journal_dirty(handle, orig_dx_leaves[i]);
3765 ocfs2_journal_dirty(handle, new_dx_leaves[i]);
3766 }
3767}
3768
3769static int ocfs2_dx_dir_rebalance_credits(struct ocfs2_super *osb,
3770 struct ocfs2_dx_root_block *dx_root)
3771{
3772 int credits = ocfs2_clusters_to_blocks(osb->sb, 2);
3773
3774 credits += ocfs2_calc_extend_credits(osb->sb, &dx_root->dr_list, 1);
3775 credits += ocfs2_quota_trans_credits(osb->sb);
3776 return credits;
3777}
3778
3779/*
3780 * Find the median value in dx_leaf_bh and allocate a new leaf to move
3781 * half our entries into.
3782 */
3783static int ocfs2_dx_dir_rebalance(struct ocfs2_super *osb, struct inode *dir,
3784 struct buffer_head *dx_root_bh,
3785 struct buffer_head *dx_leaf_bh,
3786 struct ocfs2_dx_hinfo *hinfo, u32 leaf_cpos,
3787 u64 leaf_blkno)
3788{
3789 struct ocfs2_dx_leaf *dx_leaf = (struct ocfs2_dx_leaf *)dx_leaf_bh->b_data;
3790 int credits, ret, i, num_used, did_quota = 0;
3791 u32 cpos, split_hash, insert_hash = hinfo->major_hash;
3792 u64 orig_leaves_start;
3793 int num_dx_leaves;
3794 struct buffer_head **orig_dx_leaves = NULL;
3795 struct buffer_head **new_dx_leaves = NULL;
3796 struct ocfs2_alloc_context *data_ac = NULL, *meta_ac = NULL;
3797 struct ocfs2_extent_tree et;
3798 handle_t *handle = NULL;
3799 struct ocfs2_dx_root_block *dx_root;
3800 struct ocfs2_dx_leaf *tmp_dx_leaf = NULL;
3801
Tao Maf1088d42011-02-23 22:30:23 +08003802 trace_ocfs2_dx_dir_rebalance((unsigned long long)OCFS2_I(dir)->ip_blkno,
3803 (unsigned long long)leaf_blkno,
3804 insert_hash);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003805
Joel Becker5e404e92009-02-13 03:54:22 -08003806 ocfs2_init_dx_root_extent_tree(&et, INODE_CACHE(dir), dx_root_bh);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003807
3808 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
3809 /*
3810 * XXX: This is a rather large limit. We should use a more
3811 * realistic value.
3812 */
3813 if (le32_to_cpu(dx_root->dr_clusters) == UINT_MAX)
3814 return -ENOSPC;
3815
3816 num_used = le16_to_cpu(dx_leaf->dl_list.de_num_used);
3817 if (num_used < le16_to_cpu(dx_leaf->dl_list.de_count)) {
3818 mlog(ML_ERROR, "DX Dir: %llu, Asked to rebalance empty leaf: "
3819 "%llu, %d\n", (unsigned long long)OCFS2_I(dir)->ip_blkno,
3820 (unsigned long long)leaf_blkno, num_used);
3821 ret = -EIO;
3822 goto out;
3823 }
3824
3825 orig_dx_leaves = ocfs2_dx_dir_kmalloc_leaves(osb->sb, &num_dx_leaves);
3826 if (!orig_dx_leaves) {
3827 ret = -ENOMEM;
3828 mlog_errno(ret);
3829 goto out;
3830 }
3831
3832 new_dx_leaves = ocfs2_dx_dir_kmalloc_leaves(osb->sb, NULL);
3833 if (!new_dx_leaves) {
3834 ret = -ENOMEM;
3835 mlog_errno(ret);
3836 goto out;
3837 }
3838
3839 ret = ocfs2_lock_allocators(dir, &et, 1, 0, &data_ac, &meta_ac);
3840 if (ret) {
3841 if (ret != -ENOSPC)
3842 mlog_errno(ret);
3843 goto out;
3844 }
3845
3846 credits = ocfs2_dx_dir_rebalance_credits(osb, dx_root);
3847 handle = ocfs2_start_trans(osb, credits);
3848 if (IS_ERR(handle)) {
3849 ret = PTR_ERR(handle);
3850 handle = NULL;
3851 mlog_errno(ret);
3852 goto out;
3853 }
3854
Christoph Hellwig5dd40562010-03-03 09:05:00 -05003855 ret = dquot_alloc_space_nodirty(dir,
3856 ocfs2_clusters_to_bytes(dir->i_sb, 1));
3857 if (ret)
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003858 goto out_commit;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003859 did_quota = 1;
3860
Joel Becker0cf2f762009-02-12 16:41:25 -08003861 ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir), dx_leaf_bh,
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003862 OCFS2_JOURNAL_ACCESS_WRITE);
3863 if (ret) {
3864 mlog_errno(ret);
3865 goto out_commit;
3866 }
3867
3868 /*
3869 * This block is changing anyway, so we can sort it in place.
3870 */
3871 sort(dx_leaf->dl_list.de_entries, num_used,
3872 sizeof(struct ocfs2_dx_entry), dx_leaf_sort_cmp,
3873 dx_leaf_sort_swap);
3874
Joel Beckerec20cec2010-03-19 14:13:52 -07003875 ocfs2_journal_dirty(handle, dx_leaf_bh);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003876
3877 ret = ocfs2_dx_dir_find_leaf_split(dx_leaf, leaf_cpos, insert_hash,
3878 &split_hash);
3879 if (ret) {
3880 mlog_errno(ret);
3881 goto out_commit;
3882 }
3883
Tao Maf1088d42011-02-23 22:30:23 +08003884 trace_ocfs2_dx_dir_rebalance_split(leaf_cpos, split_hash, insert_hash);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003885
3886 /*
3887 * We have to carefully order operations here. There are items
3888 * which want to be in the new cluster before insert, but in
3889 * order to put those items in the new cluster, we alter the
3890 * old cluster. A failure to insert gets nasty.
3891 *
3892 * So, start by reserving writes to the old
3893 * cluster. ocfs2_dx_dir_new_cluster will reserve writes on
3894 * the new cluster for us, before inserting it. The insert
3895 * won't happen if there's an error before that. Once the
3896 * insert is done then, we can transfer from one leaf into the
3897 * other without fear of hitting any error.
3898 */
3899
3900 /*
3901 * The leaf transfer wants some scratch space so that we don't
3902 * wind up doing a bunch of expensive memmove().
3903 */
3904 tmp_dx_leaf = kmalloc(osb->sb->s_blocksize, GFP_NOFS);
3905 if (!tmp_dx_leaf) {
3906 ret = -ENOMEM;
3907 mlog_errno(ret);
3908 goto out_commit;
3909 }
3910
Mark Fasheh1d46dc02009-02-19 13:17:05 -08003911 orig_leaves_start = ocfs2_block_to_cluster_start(dir->i_sb, leaf_blkno);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003912 ret = ocfs2_read_dx_leaves(dir, orig_leaves_start, num_dx_leaves,
3913 orig_dx_leaves);
3914 if (ret) {
3915 mlog_errno(ret);
3916 goto out_commit;
3917 }
3918
Tristan Ye0f4da212010-09-08 17:12:38 +08003919 cpos = split_hash;
3920 ret = ocfs2_dx_dir_new_cluster(dir, &et, cpos, handle,
3921 data_ac, meta_ac, new_dx_leaves,
3922 num_dx_leaves);
3923 if (ret) {
3924 mlog_errno(ret);
3925 goto out_commit;
3926 }
3927
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003928 for (i = 0; i < num_dx_leaves; i++) {
Joel Becker0cf2f762009-02-12 16:41:25 -08003929 ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir),
3930 orig_dx_leaves[i],
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003931 OCFS2_JOURNAL_ACCESS_WRITE);
3932 if (ret) {
3933 mlog_errno(ret);
3934 goto out_commit;
3935 }
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003936
Tristan Ye0f4da212010-09-08 17:12:38 +08003937 ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir),
3938 new_dx_leaves[i],
3939 OCFS2_JOURNAL_ACCESS_WRITE);
3940 if (ret) {
3941 mlog_errno(ret);
3942 goto out_commit;
3943 }
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003944 }
3945
3946 ocfs2_dx_dir_transfer_leaf(dir, split_hash, handle, tmp_dx_leaf,
3947 orig_dx_leaves, new_dx_leaves, num_dx_leaves);
3948
3949out_commit:
3950 if (ret < 0 && did_quota)
Christoph Hellwig5dd40562010-03-03 09:05:00 -05003951 dquot_free_space_nodirty(dir,
Mark Fasheh9b7895e2008-11-12 16:27:44 -08003952 ocfs2_clusters_to_bytes(dir->i_sb, 1));
3953
3954 ocfs2_commit_trans(osb, handle);
3955
3956out:
3957 if (orig_dx_leaves || new_dx_leaves) {
3958 for (i = 0; i < num_dx_leaves; i++) {
3959 if (orig_dx_leaves)
3960 brelse(orig_dx_leaves[i]);
3961 if (new_dx_leaves)
3962 brelse(new_dx_leaves[i]);
3963 }
3964 kfree(orig_dx_leaves);
3965 kfree(new_dx_leaves);
3966 }
3967
3968 if (meta_ac)
3969 ocfs2_free_alloc_context(meta_ac);
3970 if (data_ac)
3971 ocfs2_free_alloc_context(data_ac);
3972
3973 kfree(tmp_dx_leaf);
3974 return ret;
3975}
3976
Mark Fashehe7c17e42009-01-29 18:17:46 -08003977static int ocfs2_find_dir_space_dx(struct ocfs2_super *osb, struct inode *dir,
3978 struct buffer_head *di_bh,
3979 struct buffer_head *dx_root_bh,
3980 const char *name, int namelen,
3981 struct ocfs2_dir_lookup_result *lookup)
3982{
3983 int ret, rebalanced = 0;
3984 struct ocfs2_dx_root_block *dx_root;
3985 struct buffer_head *dx_leaf_bh = NULL;
3986 struct ocfs2_dx_leaf *dx_leaf;
3987 u64 blkno;
3988 u32 leaf_cpos;
3989
3990 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
3991
3992restart_search:
3993 ret = ocfs2_dx_dir_lookup(dir, &dx_root->dr_list, &lookup->dl_hinfo,
3994 &leaf_cpos, &blkno);
3995 if (ret) {
3996 mlog_errno(ret);
3997 goto out;
3998 }
3999
4000 ret = ocfs2_read_dx_leaf(dir, blkno, &dx_leaf_bh);
4001 if (ret) {
4002 mlog_errno(ret);
4003 goto out;
4004 }
4005
4006 dx_leaf = (struct ocfs2_dx_leaf *)dx_leaf_bh->b_data;
4007
4008 if (le16_to_cpu(dx_leaf->dl_list.de_num_used) >=
4009 le16_to_cpu(dx_leaf->dl_list.de_count)) {
4010 if (rebalanced) {
4011 /*
4012 * Rebalancing should have provided us with
4013 * space in an appropriate leaf.
4014 *
4015 * XXX: Is this an abnormal condition then?
4016 * Should we print a message here?
4017 */
4018 ret = -ENOSPC;
4019 goto out;
4020 }
4021
4022 ret = ocfs2_dx_dir_rebalance(osb, dir, dx_root_bh, dx_leaf_bh,
4023 &lookup->dl_hinfo, leaf_cpos,
4024 blkno);
4025 if (ret) {
4026 if (ret != -ENOSPC)
4027 mlog_errno(ret);
4028 goto out;
4029 }
4030
4031 /*
4032 * Restart the lookup. The rebalance might have
4033 * changed which block our item fits into. Mark our
4034 * progress, so we only execute this once.
4035 */
4036 brelse(dx_leaf_bh);
4037 dx_leaf_bh = NULL;
4038 rebalanced = 1;
4039 goto restart_search;
4040 }
4041
4042 lookup->dl_dx_leaf_bh = dx_leaf_bh;
4043 dx_leaf_bh = NULL;
4044
4045out:
4046 brelse(dx_leaf_bh);
4047 return ret;
4048}
4049
4050static int ocfs2_search_dx_free_list(struct inode *dir,
4051 struct buffer_head *dx_root_bh,
4052 int namelen,
4053 struct ocfs2_dir_lookup_result *lookup)
4054{
4055 int ret = -ENOSPC;
4056 struct buffer_head *leaf_bh = NULL, *prev_leaf_bh = NULL;
4057 struct ocfs2_dir_block_trailer *db;
4058 u64 next_block;
4059 int rec_len = OCFS2_DIR_REC_LEN(namelen);
4060 struct ocfs2_dx_root_block *dx_root;
4061
4062 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
4063 next_block = le64_to_cpu(dx_root->dr_free_blk);
4064
4065 while (next_block) {
4066 brelse(prev_leaf_bh);
4067 prev_leaf_bh = leaf_bh;
4068 leaf_bh = NULL;
4069
4070 ret = ocfs2_read_dir_block_direct(dir, next_block, &leaf_bh);
4071 if (ret) {
4072 mlog_errno(ret);
4073 goto out;
4074 }
4075
4076 db = ocfs2_trailer_from_bh(leaf_bh, dir->i_sb);
4077 if (rec_len <= le16_to_cpu(db->db_free_rec_len)) {
4078 lookup->dl_leaf_bh = leaf_bh;
4079 lookup->dl_prev_leaf_bh = prev_leaf_bh;
4080 leaf_bh = NULL;
4081 prev_leaf_bh = NULL;
4082 break;
4083 }
4084
4085 next_block = le64_to_cpu(db->db_free_next);
4086 }
4087
4088 if (!next_block)
4089 ret = -ENOSPC;
4090
4091out:
4092
4093 brelse(leaf_bh);
4094 brelse(prev_leaf_bh);
4095 return ret;
4096}
4097
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08004098static int ocfs2_expand_inline_dx_root(struct inode *dir,
4099 struct buffer_head *dx_root_bh)
4100{
4101 int ret, num_dx_leaves, i, j, did_quota = 0;
4102 struct buffer_head **dx_leaves = NULL;
4103 struct ocfs2_extent_tree et;
4104 u64 insert_blkno;
4105 struct ocfs2_alloc_context *data_ac = NULL;
4106 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
4107 handle_t *handle = NULL;
4108 struct ocfs2_dx_root_block *dx_root;
4109 struct ocfs2_dx_entry_list *entry_list;
4110 struct ocfs2_dx_entry *dx_entry;
4111 struct ocfs2_dx_leaf *target_leaf;
4112
4113 ret = ocfs2_reserve_clusters(osb, 1, &data_ac);
4114 if (ret) {
4115 mlog_errno(ret);
4116 goto out;
4117 }
4118
4119 dx_leaves = ocfs2_dx_dir_kmalloc_leaves(osb->sb, &num_dx_leaves);
4120 if (!dx_leaves) {
4121 ret = -ENOMEM;
4122 mlog_errno(ret);
4123 goto out;
4124 }
4125
4126 handle = ocfs2_start_trans(osb, ocfs2_calc_dxi_expand_credits(osb->sb));
4127 if (IS_ERR(handle)) {
4128 ret = PTR_ERR(handle);
4129 mlog_errno(ret);
4130 goto out;
4131 }
4132
Christoph Hellwig5dd40562010-03-03 09:05:00 -05004133 ret = dquot_alloc_space_nodirty(dir,
4134 ocfs2_clusters_to_bytes(osb->sb, 1));
4135 if (ret)
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08004136 goto out_commit;
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08004137 did_quota = 1;
4138
4139 /*
4140 * We do this up front, before the allocation, so that a
4141 * failure to add the dx_root_bh to the journal won't result
4142 * us losing clusters.
4143 */
Joel Becker0cf2f762009-02-12 16:41:25 -08004144 ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08004145 OCFS2_JOURNAL_ACCESS_WRITE);
4146 if (ret) {
4147 mlog_errno(ret);
4148 goto out_commit;
4149 }
4150
4151 ret = __ocfs2_dx_dir_new_cluster(dir, 0, handle, data_ac, dx_leaves,
4152 num_dx_leaves, &insert_blkno);
4153 if (ret) {
4154 mlog_errno(ret);
4155 goto out_commit;
4156 }
4157
4158 /*
4159 * Transfer the entries from our dx_root into the appropriate
4160 * block
4161 */
4162 dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data;
4163 entry_list = &dx_root->dr_entries;
4164
4165 for (i = 0; i < le16_to_cpu(entry_list->de_num_used); i++) {
4166 dx_entry = &entry_list->de_entries[i];
4167
4168 j = __ocfs2_dx_dir_hash_idx(osb,
4169 le32_to_cpu(dx_entry->dx_minor_hash));
4170 target_leaf = (struct ocfs2_dx_leaf *)dx_leaves[j]->b_data;
4171
4172 ocfs2_dx_dir_leaf_insert_tail(target_leaf, dx_entry);
4173
4174 /* Each leaf has been passed to the journal already
4175 * via __ocfs2_dx_dir_new_cluster() */
4176 }
4177
4178 dx_root->dr_flags &= ~OCFS2_DX_FLAG_INLINE;
4179 memset(&dx_root->dr_list, 0, osb->sb->s_blocksize -
4180 offsetof(struct ocfs2_dx_root_block, dr_list));
4181 dx_root->dr_list.l_count =
4182 cpu_to_le16(ocfs2_extent_recs_per_dx_root(osb->sb));
4183
4184 /* This should never fail considering we start with an empty
4185 * dx_root. */
Joel Becker5e404e92009-02-13 03:54:22 -08004186 ocfs2_init_dx_root_extent_tree(&et, INODE_CACHE(dir), dx_root_bh);
Joel Beckercc79d8c2009-02-13 03:24:43 -08004187 ret = ocfs2_insert_extent(handle, &et, 0, insert_blkno, 1, 0, NULL);
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08004188 if (ret)
4189 mlog_errno(ret);
4190 did_quota = 0;
4191
4192 ocfs2_journal_dirty(handle, dx_root_bh);
4193
4194out_commit:
4195 if (ret < 0 && did_quota)
Christoph Hellwig5dd40562010-03-03 09:05:00 -05004196 dquot_free_space_nodirty(dir,
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08004197 ocfs2_clusters_to_bytes(dir->i_sb, 1));
4198
4199 ocfs2_commit_trans(osb, handle);
4200
4201out:
4202 if (data_ac)
4203 ocfs2_free_alloc_context(data_ac);
4204
4205 if (dx_leaves) {
4206 for (i = 0; i < num_dx_leaves; i++)
4207 brelse(dx_leaves[i]);
4208 kfree(dx_leaves);
4209 }
4210 return ret;
4211}
4212
4213static int ocfs2_inline_dx_has_space(struct buffer_head *dx_root_bh)
4214{
4215 struct ocfs2_dx_root_block *dx_root;
4216 struct ocfs2_dx_entry_list *entry_list;
4217
4218 dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data;
4219 entry_list = &dx_root->dr_entries;
4220
4221 if (le16_to_cpu(entry_list->de_num_used) >=
4222 le16_to_cpu(entry_list->de_count))
4223 return -ENOSPC;
4224
4225 return 0;
4226}
4227
Mark Fashehe7c17e42009-01-29 18:17:46 -08004228static int ocfs2_prepare_dx_dir_for_insert(struct inode *dir,
4229 struct buffer_head *di_bh,
4230 const char *name,
4231 int namelen,
4232 struct ocfs2_dir_lookup_result *lookup)
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004233{
Mark Fashehe7c17e42009-01-29 18:17:46 -08004234 int ret, free_dx_root = 1;
4235 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004236 struct buffer_head *dx_root_bh = NULL;
Mark Fashehe7c17e42009-01-29 18:17:46 -08004237 struct buffer_head *leaf_bh = NULL;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004238 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
Mark Fashehe7c17e42009-01-29 18:17:46 -08004239 struct ocfs2_dx_root_block *dx_root;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004240
4241 ret = ocfs2_read_dx_root(dir, di, &dx_root_bh);
4242 if (ret) {
4243 mlog_errno(ret);
4244 goto out;
4245 }
4246
4247 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
Mark Fashehe3a93c22009-02-17 15:29:35 -08004248 if (le32_to_cpu(dx_root->dr_num_entries) == OCFS2_DX_ENTRIES_MAX) {
4249 ret = -ENOSPC;
4250 mlog_errno(ret);
4251 goto out;
4252 }
4253
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08004254 if (ocfs2_dx_root_inline(dx_root)) {
4255 ret = ocfs2_inline_dx_has_space(dx_root_bh);
4256
4257 if (ret == 0)
4258 goto search_el;
4259
4260 /*
4261 * We ran out of room in the root block. Expand it to
4262 * an extent, then allow ocfs2_find_dir_space_dx to do
4263 * the rest.
4264 */
4265 ret = ocfs2_expand_inline_dx_root(dir, dx_root_bh);
4266 if (ret) {
4267 mlog_errno(ret);
4268 goto out;
4269 }
4270 }
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004271
Mark Fashehe7c17e42009-01-29 18:17:46 -08004272 /*
4273 * Insert preparation for an indexed directory is split into two
4274 * steps. The call to find_dir_space_dx reserves room in the index for
4275 * an additional item. If we run out of space there, it's a real error
4276 * we can't continue on.
4277 */
4278 ret = ocfs2_find_dir_space_dx(osb, dir, di_bh, dx_root_bh, name,
4279 namelen, lookup);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004280 if (ret) {
4281 mlog_errno(ret);
4282 goto out;
4283 }
4284
Mark Fashehe7c17e42009-01-29 18:17:46 -08004285search_el:
4286 /*
4287 * Next, we need to find space in the unindexed tree. This call
4288 * searches using the free space linked list. If the unindexed tree
4289 * lacks sufficient space, we'll expand it below. The expansion code
4290 * is smart enough to add any new blocks to the free space list.
4291 */
4292 ret = ocfs2_search_dx_free_list(dir, dx_root_bh, namelen, lookup);
4293 if (ret && ret != -ENOSPC) {
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004294 mlog_errno(ret);
4295 goto out;
4296 }
4297
Mark Fashehe7c17e42009-01-29 18:17:46 -08004298 /* Do this up here - ocfs2_extend_dir might need the dx_root */
4299 lookup->dl_dx_root_bh = dx_root_bh;
4300 free_dx_root = 0;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004301
Mark Fashehe7c17e42009-01-29 18:17:46 -08004302 if (ret == -ENOSPC) {
4303 ret = ocfs2_extend_dir(osb, dir, di_bh, 1, lookup, &leaf_bh);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004304
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004305 if (ret) {
Mark Fashehe7c17e42009-01-29 18:17:46 -08004306 mlog_errno(ret);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004307 goto out;
4308 }
4309
4310 /*
Mark Fashehe7c17e42009-01-29 18:17:46 -08004311 * We make the assumption here that new leaf blocks are added
4312 * to the front of our free list.
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004313 */
Mark Fashehe7c17e42009-01-29 18:17:46 -08004314 lookup->dl_prev_leaf_bh = NULL;
4315 lookup->dl_leaf_bh = leaf_bh;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004316 }
4317
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004318out:
Mark Fashehe7c17e42009-01-29 18:17:46 -08004319 if (free_dx_root)
4320 brelse(dx_root_bh);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004321 return ret;
4322}
4323
Mark Fasheh4a12ca32008-11-12 15:43:34 -08004324/*
4325 * Get a directory ready for insert. Any directory allocation required
4326 * happens here. Success returns zero, and enough context in the dir
4327 * lookup result that ocfs2_add_entry() will be able complete the task
4328 * with minimal performance impact.
4329 */
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07004330int ocfs2_prepare_dir_for_insert(struct ocfs2_super *osb,
4331 struct inode *dir,
4332 struct buffer_head *parent_fe_bh,
4333 const char *name,
4334 int namelen,
Mark Fasheh4a12ca32008-11-12 15:43:34 -08004335 struct ocfs2_dir_lookup_result *lookup)
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07004336{
4337 int ret;
4338 unsigned int blocks_wanted = 1;
4339 struct buffer_head *bh = NULL;
4340
Tao Maf1088d42011-02-23 22:30:23 +08004341 trace_ocfs2_prepare_dir_for_insert(
4342 (unsigned long long)OCFS2_I(dir)->ip_blkno, namelen);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07004343
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07004344 if (!namelen) {
4345 ret = -EINVAL;
4346 mlog_errno(ret);
4347 goto out;
4348 }
4349
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004350 /*
4351 * Do this up front to reduce confusion.
4352 *
4353 * The directory might start inline, then be turned into an
4354 * indexed one, in which case we'd need to hash deep inside
4355 * ocfs2_find_dir_space_id(). Since
4356 * ocfs2_prepare_dx_dir_for_insert() also needs this hash
4357 * done, there seems no point in spreading out the calls. We
4358 * can optimize away the case where the file system doesn't
4359 * support indexing.
4360 */
4361 if (ocfs2_supports_indexed_dirs(osb))
4362 ocfs2_dx_dir_name_hash(dir, name, namelen, &lookup->dl_hinfo);
4363
4364 if (ocfs2_dir_indexed(dir)) {
Mark Fashehe7c17e42009-01-29 18:17:46 -08004365 ret = ocfs2_prepare_dx_dir_for_insert(dir, parent_fe_bh,
4366 name, namelen, lookup);
4367 if (ret)
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004368 mlog_errno(ret);
Mark Fashehe7c17e42009-01-29 18:17:46 -08004369 goto out;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004370 }
4371
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07004372 if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
4373 ret = ocfs2_find_dir_space_id(dir, parent_fe_bh, name,
4374 namelen, &bh, &blocks_wanted);
4375 } else
4376 ret = ocfs2_find_dir_space_el(dir, name, namelen, &bh);
4377
4378 if (ret && ret != -ENOSPC) {
4379 mlog_errno(ret);
4380 goto out;
4381 }
4382
4383 if (ret == -ENOSPC) {
4384 /*
4385 * We have to expand the directory to add this name.
4386 */
4387 BUG_ON(bh);
4388
4389 ret = ocfs2_extend_dir(osb, dir, parent_fe_bh, blocks_wanted,
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004390 lookup, &bh);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07004391 if (ret) {
4392 if (ret != -ENOSPC)
4393 mlog_errno(ret);
4394 goto out;
4395 }
4396
4397 BUG_ON(!bh);
4398 }
4399
Mark Fasheh4a12ca32008-11-12 15:43:34 -08004400 lookup->dl_leaf_bh = bh;
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07004401 bh = NULL;
4402out:
Mark Fasheha81cb882008-10-07 14:25:16 -07004403 brelse(bh);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07004404 return ret;
4405}
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004406
4407static int ocfs2_dx_dir_remove_index(struct inode *dir,
4408 struct buffer_head *di_bh,
4409 struct buffer_head *dx_root_bh)
4410{
4411 int ret;
4412 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
4413 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
4414 struct ocfs2_dx_root_block *dx_root;
4415 struct inode *dx_alloc_inode = NULL;
4416 struct buffer_head *dx_alloc_bh = NULL;
4417 handle_t *handle;
4418 u64 blk;
4419 u16 bit;
4420 u64 bg_blkno;
4421
4422 dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data;
4423
4424 dx_alloc_inode = ocfs2_get_system_file_inode(osb,
4425 EXTENT_ALLOC_SYSTEM_INODE,
4426 le16_to_cpu(dx_root->dr_suballoc_slot));
4427 if (!dx_alloc_inode) {
4428 ret = -ENOMEM;
4429 mlog_errno(ret);
4430 goto out;
4431 }
4432 mutex_lock(&dx_alloc_inode->i_mutex);
4433
4434 ret = ocfs2_inode_lock(dx_alloc_inode, &dx_alloc_bh, 1);
4435 if (ret) {
4436 mlog_errno(ret);
4437 goto out_mutex;
4438 }
4439
4440 handle = ocfs2_start_trans(osb, OCFS2_DX_ROOT_REMOVE_CREDITS);
4441 if (IS_ERR(handle)) {
4442 ret = PTR_ERR(handle);
4443 mlog_errno(ret);
4444 goto out_unlock;
4445 }
4446
Joel Becker0cf2f762009-02-12 16:41:25 -08004447 ret = ocfs2_journal_access_di(handle, INODE_CACHE(dir), di_bh,
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004448 OCFS2_JOURNAL_ACCESS_WRITE);
4449 if (ret) {
4450 mlog_errno(ret);
4451 goto out_commit;
4452 }
4453
Tao Ma8ac33dc2010-12-15 16:30:00 +08004454 spin_lock(&OCFS2_I(dir)->ip_lock);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004455 OCFS2_I(dir)->ip_dyn_features &= ~OCFS2_INDEXED_DIR_FL;
4456 di->i_dyn_features = cpu_to_le16(OCFS2_I(dir)->ip_dyn_features);
Tao Ma8ac33dc2010-12-15 16:30:00 +08004457 spin_unlock(&OCFS2_I(dir)->ip_lock);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004458 di->i_dx_root = cpu_to_le64(0ULL);
4459
4460 ocfs2_journal_dirty(handle, di_bh);
4461
4462 blk = le64_to_cpu(dx_root->dr_blkno);
4463 bit = le16_to_cpu(dx_root->dr_suballoc_bit);
Tao Ma74380c42010-03-22 14:20:18 +08004464 if (dx_root->dr_suballoc_loc)
4465 bg_blkno = le64_to_cpu(dx_root->dr_suballoc_loc);
4466 else
4467 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004468 ret = ocfs2_free_suballoc_bits(handle, dx_alloc_inode, dx_alloc_bh,
4469 bit, bg_blkno, 1);
4470 if (ret)
4471 mlog_errno(ret);
4472
4473out_commit:
4474 ocfs2_commit_trans(osb, handle);
4475
4476out_unlock:
4477 ocfs2_inode_unlock(dx_alloc_inode, 1);
4478
4479out_mutex:
4480 mutex_unlock(&dx_alloc_inode->i_mutex);
4481 brelse(dx_alloc_bh);
4482out:
4483 iput(dx_alloc_inode);
4484 return ret;
4485}
4486
4487int ocfs2_dx_dir_truncate(struct inode *dir, struct buffer_head *di_bh)
4488{
4489 int ret;
4490 unsigned int uninitialized_var(clen);
4491 u32 major_hash = UINT_MAX, p_cpos, uninitialized_var(cpos);
4492 u64 uninitialized_var(blkno);
4493 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
4494 struct buffer_head *dx_root_bh = NULL;
4495 struct ocfs2_dx_root_block *dx_root;
4496 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
4497 struct ocfs2_cached_dealloc_ctxt dealloc;
4498 struct ocfs2_extent_tree et;
4499
4500 ocfs2_init_dealloc_ctxt(&dealloc);
4501
4502 if (!ocfs2_dir_indexed(dir))
4503 return 0;
4504
4505 ret = ocfs2_read_dx_root(dir, di, &dx_root_bh);
4506 if (ret) {
4507 mlog_errno(ret);
4508 goto out;
4509 }
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08004510 dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
4511
4512 if (ocfs2_dx_root_inline(dx_root))
4513 goto remove_index;
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004514
Joel Becker5e404e92009-02-13 03:54:22 -08004515 ocfs2_init_dx_root_extent_tree(&et, INODE_CACHE(dir), dx_root_bh);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004516
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004517 /* XXX: What if dr_clusters is too large? */
4518 while (le32_to_cpu(dx_root->dr_clusters)) {
4519 ret = ocfs2_dx_dir_lookup_rec(dir, &dx_root->dr_list,
4520 major_hash, &cpos, &blkno, &clen);
4521 if (ret) {
4522 mlog_errno(ret);
4523 goto out;
4524 }
4525
4526 p_cpos = ocfs2_blocks_to_clusters(dir->i_sb, blkno);
4527
Tristan Ye78f94672010-05-11 17:54:42 +08004528 ret = ocfs2_remove_btree_range(dir, &et, cpos, p_cpos, clen, 0,
4529 &dealloc, 0);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004530 if (ret) {
4531 mlog_errno(ret);
4532 goto out;
4533 }
4534
4535 if (cpos == 0)
4536 break;
4537
4538 major_hash = cpos - 1;
4539 }
4540
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -08004541remove_index:
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004542 ret = ocfs2_dx_dir_remove_index(dir, di_bh, dx_root_bh);
4543 if (ret) {
4544 mlog_errno(ret);
4545 goto out;
4546 }
4547
Joel Becker8cb471e2009-02-10 20:00:41 -08004548 ocfs2_remove_from_cache(INODE_CACHE(dir), dx_root_bh);
Mark Fasheh9b7895e2008-11-12 16:27:44 -08004549out:
4550 ocfs2_schedule_truncate_log_flush(osb, 1);
4551 ocfs2_run_deallocs(osb, &dealloc);
4552
4553 brelse(dx_root_bh);
4554 return ret;
4555}