blob: 100754de16b8ba110a535474583d39132cc224d0 [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 * file.c
5 *
6 * File open, close, extend, truncate
7 *
8 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation; either
13 * version 2 of the License, or (at your option) any later version.
14 *
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
19 *
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
24 */
25
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080026#include <linux/capability.h>
Mark Fashehccd979b2005-12-15 14:31:24 -080027#include <linux/fs.h>
28#include <linux/types.h>
29#include <linux/slab.h>
30#include <linux/highmem.h>
31#include <linux/pagemap.h>
32#include <linux/uio.h>
Mark Fashehe2057c52006-10-03 17:53:05 -070033#include <linux/sched.h>
Mark Fashehccd979b2005-12-15 14:31:24 -080034
35#define MLOG_MASK_PREFIX ML_INODE
36#include <cluster/masklog.h>
37
38#include "ocfs2.h"
39
40#include "alloc.h"
41#include "aops.h"
42#include "dir.h"
43#include "dlmglue.h"
44#include "extent_map.h"
45#include "file.h"
46#include "sysfile.h"
47#include "inode.h"
Herbert Poetzlca4d1472006-07-03 17:27:12 -070048#include "ioctl.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080049#include "journal.h"
50#include "mmap.h"
51#include "suballoc.h"
52#include "super.h"
53
54#include "buffer_head_io.h"
55
56static int ocfs2_sync_inode(struct inode *inode)
57{
58 filemap_fdatawrite(inode->i_mapping);
59 return sync_mapping_buffers(inode->i_mapping);
60}
61
62static int ocfs2_file_open(struct inode *inode, struct file *file)
63{
64 int status;
65 int mode = file->f_flags;
66 struct ocfs2_inode_info *oi = OCFS2_I(inode);
67
68 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
69 file->f_dentry->d_name.len, file->f_dentry->d_name.name);
70
71 spin_lock(&oi->ip_lock);
72
73 /* Check that the inode hasn't been wiped from disk by another
74 * node. If it hasn't then we're safe as long as we hold the
75 * spin lock until our increment of open count. */
76 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
77 spin_unlock(&oi->ip_lock);
78
79 status = -ENOENT;
80 goto leave;
81 }
82
83 if (mode & O_DIRECT)
84 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
85
86 oi->ip_open_count++;
87 spin_unlock(&oi->ip_lock);
88 status = 0;
89leave:
90 mlog_exit(status);
91 return status;
92}
93
94static int ocfs2_file_release(struct inode *inode, struct file *file)
95{
96 struct ocfs2_inode_info *oi = OCFS2_I(inode);
97
98 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
99 file->f_dentry->d_name.len,
100 file->f_dentry->d_name.name);
101
102 spin_lock(&oi->ip_lock);
103 if (!--oi->ip_open_count)
104 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
105 spin_unlock(&oi->ip_lock);
106
107 mlog_exit(0);
108
109 return 0;
110}
111
112static int ocfs2_sync_file(struct file *file,
113 struct dentry *dentry,
114 int datasync)
115{
116 int err = 0;
117 journal_t *journal;
118 struct inode *inode = dentry->d_inode;
119 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
120
121 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync,
122 dentry->d_name.len, dentry->d_name.name);
123
124 err = ocfs2_sync_inode(dentry->d_inode);
125 if (err)
126 goto bail;
127
128 journal = osb->journal->j_journal;
129 err = journal_force_commit(journal);
130
131bail:
132 mlog_exit(err);
133
134 return (err < 0) ? -EIO : 0;
135}
136
137int ocfs2_set_inode_size(struct ocfs2_journal_handle *handle,
138 struct inode *inode,
139 struct buffer_head *fe_bh,
140 u64 new_i_size)
141{
142 int status;
143
144 mlog_entry_void();
145 i_size_write(inode, new_i_size);
146 inode->i_blocks = ocfs2_align_bytes_to_sectors(new_i_size);
147 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
148
149 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
150 if (status < 0) {
151 mlog_errno(status);
152 goto bail;
153 }
154
155bail:
156 mlog_exit(status);
157 return status;
158}
159
160static int ocfs2_simple_size_update(struct inode *inode,
161 struct buffer_head *di_bh,
162 u64 new_i_size)
163{
164 int ret;
165 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
166 struct ocfs2_journal_handle *handle = NULL;
167
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700168 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800169 if (handle == NULL) {
170 ret = -ENOMEM;
171 mlog_errno(ret);
172 goto out;
173 }
174
175 ret = ocfs2_set_inode_size(handle, inode, di_bh,
176 new_i_size);
177 if (ret < 0)
178 mlog_errno(ret);
179
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700180 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800181out:
182 return ret;
183}
184
185static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
186 struct inode *inode,
187 struct buffer_head *fe_bh,
188 u64 new_i_size)
189{
190 int status;
191 struct ocfs2_journal_handle *handle;
192
193 mlog_entry_void();
194
195 /* TODO: This needs to actually orphan the inode in this
196 * transaction. */
197
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700198 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800199 if (IS_ERR(handle)) {
200 status = PTR_ERR(handle);
201 mlog_errno(status);
202 goto out;
203 }
204
205 status = ocfs2_set_inode_size(handle, inode, fe_bh, new_i_size);
206 if (status < 0)
207 mlog_errno(status);
208
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700209 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800210out:
211 mlog_exit(status);
212 return status;
213}
214
215static int ocfs2_truncate_file(struct inode *inode,
216 struct buffer_head *di_bh,
217 u64 new_i_size)
218{
219 int status = 0;
220 struct ocfs2_dinode *fe = NULL;
221 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
222 struct ocfs2_truncate_context *tc = NULL;
223
Mark Fashehb06970532006-03-03 10:24:33 -0800224 mlog_entry("(inode = %llu, new_i_size = %llu\n",
225 (unsigned long long)OCFS2_I(inode)->ip_blkno,
226 (unsigned long long)new_i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800227
228 truncate_inode_pages(inode->i_mapping, new_i_size);
229
230 fe = (struct ocfs2_dinode *) di_bh->b_data;
231 if (!OCFS2_IS_VALID_DINODE(fe)) {
232 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
233 status = -EIO;
234 goto bail;
235 }
236
237 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
Mark Fashehb06970532006-03-03 10:24:33 -0800238 "Inode %llu, inode i_size = %lld != di "
239 "i_size = %llu, i_flags = 0x%x\n",
240 (unsigned long long)OCFS2_I(inode)->ip_blkno,
Mark Fashehccd979b2005-12-15 14:31:24 -0800241 i_size_read(inode),
Mark Fashehb06970532006-03-03 10:24:33 -0800242 (unsigned long long)le64_to_cpu(fe->i_size),
243 le32_to_cpu(fe->i_flags));
Mark Fashehccd979b2005-12-15 14:31:24 -0800244
245 if (new_i_size > le64_to_cpu(fe->i_size)) {
Mark Fashehb06970532006-03-03 10:24:33 -0800246 mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
247 (unsigned long long)le64_to_cpu(fe->i_size),
248 (unsigned long long)new_i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800249 status = -EINVAL;
250 mlog_errno(status);
251 goto bail;
252 }
253
Mark Fashehb06970532006-03-03 10:24:33 -0800254 mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
255 (unsigned long long)le64_to_cpu(fe->i_blkno),
256 (unsigned long long)le64_to_cpu(fe->i_size),
257 (unsigned long long)new_i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800258
259 /* lets handle the simple truncate cases before doing any more
260 * cluster locking. */
261 if (new_i_size == le64_to_cpu(fe->i_size))
262 goto bail;
263
Mark Fashehab0920c2006-03-16 15:06:37 -0800264 /* This forces other nodes to sync and drop their pages. Do
265 * this even if we have a truncate without allocation change -
266 * ocfs2 cluster sizes can be much greater than page size, so
267 * we have to truncate them anyway. */
268 status = ocfs2_data_lock(inode, 1);
269 if (status < 0) {
270 mlog_errno(status);
271 goto bail;
272 }
273 ocfs2_data_unlock(inode, 1);
274
Mark Fashehccd979b2005-12-15 14:31:24 -0800275 if (le32_to_cpu(fe->i_clusters) ==
276 ocfs2_clusters_for_bytes(osb->sb, new_i_size)) {
277 mlog(0, "fe->i_clusters = %u, so we do a simple truncate\n",
278 fe->i_clusters);
279 /* No allocation change is required, so lets fast path
280 * this truncate. */
281 status = ocfs2_simple_size_update(inode, di_bh, new_i_size);
282 if (status < 0)
283 mlog_errno(status);
284 goto bail;
285 }
286
Mark Fashehccd979b2005-12-15 14:31:24 -0800287 /* alright, we're going to need to do a full blown alloc size
288 * change. Orphan the inode so that recovery can complete the
289 * truncate if necessary. This does the task of marking
290 * i_size. */
291 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
292 if (status < 0) {
293 mlog_errno(status);
294 goto bail;
295 }
296
297 status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc);
298 if (status < 0) {
299 mlog_errno(status);
300 goto bail;
301 }
302
303 status = ocfs2_commit_truncate(osb, inode, di_bh, tc);
304 if (status < 0) {
305 mlog_errno(status);
306 goto bail;
307 }
308
309 /* TODO: orphan dir cleanup here. */
310bail:
311
312 mlog_exit(status);
313 return status;
314}
315
316/*
317 * extend allocation only here.
318 * we'll update all the disk stuff, and oip->alloc_size
319 *
320 * expect stuff to be locked, a transaction started and enough data /
321 * metadata reservations in the contexts.
322 *
323 * Will return -EAGAIN, and a reason if a restart is needed.
324 * If passed in, *reason will always be set, even in error.
325 */
326int ocfs2_do_extend_allocation(struct ocfs2_super *osb,
327 struct inode *inode,
328 u32 clusters_to_add,
329 struct buffer_head *fe_bh,
330 struct ocfs2_journal_handle *handle,
331 struct ocfs2_alloc_context *data_ac,
332 struct ocfs2_alloc_context *meta_ac,
333 enum ocfs2_alloc_restarted *reason_ret)
334{
335 int status = 0;
336 int free_extents;
337 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
338 enum ocfs2_alloc_restarted reason = RESTART_NONE;
339 u32 bit_off, num_bits;
340 u64 block;
341
342 BUG_ON(!clusters_to_add);
343
344 free_extents = ocfs2_num_free_extents(osb, inode, fe);
345 if (free_extents < 0) {
346 status = free_extents;
347 mlog_errno(status);
348 goto leave;
349 }
350
351 /* there are two cases which could cause us to EAGAIN in the
352 * we-need-more-metadata case:
353 * 1) we haven't reserved *any*
354 * 2) we are so fragmented, we've needed to add metadata too
355 * many times. */
356 if (!free_extents && !meta_ac) {
357 mlog(0, "we haven't reserved any metadata!\n");
358 status = -EAGAIN;
359 reason = RESTART_META;
360 goto leave;
361 } else if ((!free_extents)
362 && (ocfs2_alloc_context_bits_left(meta_ac)
363 < ocfs2_extend_meta_needed(fe))) {
364 mlog(0, "filesystem is really fragmented...\n");
365 status = -EAGAIN;
366 reason = RESTART_META;
367 goto leave;
368 }
369
370 status = ocfs2_claim_clusters(osb, handle, data_ac, 1,
371 &bit_off, &num_bits);
372 if (status < 0) {
373 if (status != -ENOSPC)
374 mlog_errno(status);
375 goto leave;
376 }
377
378 BUG_ON(num_bits > clusters_to_add);
379
380 /* reserve our write early -- insert_extent may update the inode */
381 status = ocfs2_journal_access(handle, inode, fe_bh,
382 OCFS2_JOURNAL_ACCESS_WRITE);
383 if (status < 0) {
384 mlog_errno(status);
385 goto leave;
386 }
387
388 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
Mark Fashehb06970532006-03-03 10:24:33 -0800389 mlog(0, "Allocating %u clusters at block %u for inode %llu\n",
390 num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800391 status = ocfs2_insert_extent(osb, handle, inode, fe_bh, block,
392 num_bits, meta_ac);
393 if (status < 0) {
394 mlog_errno(status);
395 goto leave;
396 }
397
398 le32_add_cpu(&fe->i_clusters, num_bits);
399 spin_lock(&OCFS2_I(inode)->ip_lock);
400 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
401 spin_unlock(&OCFS2_I(inode)->ip_lock);
402
403 status = ocfs2_journal_dirty(handle, fe_bh);
404 if (status < 0) {
405 mlog_errno(status);
406 goto leave;
407 }
408
409 clusters_to_add -= num_bits;
410
411 if (clusters_to_add) {
412 mlog(0, "need to alloc once more, clusters = %u, wanted = "
413 "%u\n", fe->i_clusters, clusters_to_add);
414 status = -EAGAIN;
415 reason = RESTART_TRANS;
416 }
417
418leave:
419 mlog_exit(status);
420 if (reason_ret)
421 *reason_ret = reason;
422 return status;
423}
424
425static int ocfs2_extend_allocation(struct inode *inode,
426 u32 clusters_to_add)
427{
428 int status = 0;
429 int restart_func = 0;
430 int drop_alloc_sem = 0;
431 int credits, num_free_extents;
432 u32 prev_clusters;
433 struct buffer_head *bh = NULL;
434 struct ocfs2_dinode *fe = NULL;
435 struct ocfs2_journal_handle *handle = NULL;
436 struct ocfs2_alloc_context *data_ac = NULL;
437 struct ocfs2_alloc_context *meta_ac = NULL;
438 enum ocfs2_alloc_restarted why;
439 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
440
441 mlog_entry("(clusters_to_add = %u)\n", clusters_to_add);
442
443 status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &bh,
444 OCFS2_BH_CACHED, inode);
445 if (status < 0) {
446 mlog_errno(status);
447 goto leave;
448 }
449
450 fe = (struct ocfs2_dinode *) bh->b_data;
451 if (!OCFS2_IS_VALID_DINODE(fe)) {
452 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
453 status = -EIO;
454 goto leave;
455 }
456
457restart_all:
458 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
459
Mark Fashehb06970532006-03-03 10:24:33 -0800460 mlog(0, "extend inode %llu, i_size = %lld, fe->i_clusters = %u, "
Mark Fashehccd979b2005-12-15 14:31:24 -0800461 "clusters_to_add = %u\n",
Mark Fashehb06970532006-03-03 10:24:33 -0800462 (unsigned long long)OCFS2_I(inode)->ip_blkno, i_size_read(inode),
Mark Fashehccd979b2005-12-15 14:31:24 -0800463 fe->i_clusters, clusters_to_add);
464
Mark Fashehccd979b2005-12-15 14:31:24 -0800465 num_free_extents = ocfs2_num_free_extents(osb,
466 inode,
467 fe);
468 if (num_free_extents < 0) {
469 status = num_free_extents;
470 mlog_errno(status);
471 goto leave;
472 }
473
474 if (!num_free_extents) {
Mark Fashehda5cbf22006-10-06 18:34:35 -0700475 status = ocfs2_reserve_new_metadata(osb, fe, &meta_ac);
Mark Fashehccd979b2005-12-15 14:31:24 -0800476 if (status < 0) {
477 if (status != -ENOSPC)
478 mlog_errno(status);
479 goto leave;
480 }
481 }
482
Mark Fashehda5cbf22006-10-06 18:34:35 -0700483 status = ocfs2_reserve_clusters(osb, clusters_to_add, &data_ac);
Mark Fashehccd979b2005-12-15 14:31:24 -0800484 if (status < 0) {
485 if (status != -ENOSPC)
486 mlog_errno(status);
487 goto leave;
488 }
489
490 /* blocks peope in read/write from reading our allocation
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800491 * until we're done changing it. We depend on i_mutex to block
Mark Fashehccd979b2005-12-15 14:31:24 -0800492 * other extend/truncate calls while we're here. Ordering wrt
493 * start_trans is important here -- always do it before! */
494 down_write(&OCFS2_I(inode)->ip_alloc_sem);
495 drop_alloc_sem = 1;
496
497 credits = ocfs2_calc_extend_credits(osb->sb, fe, clusters_to_add);
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700498 handle = ocfs2_start_trans(osb, credits);
Mark Fashehccd979b2005-12-15 14:31:24 -0800499 if (IS_ERR(handle)) {
500 status = PTR_ERR(handle);
501 handle = NULL;
502 mlog_errno(status);
503 goto leave;
504 }
505
506restarted_transaction:
507 /* reserve a write to the file entry early on - that we if we
508 * run out of credits in the allocation path, we can still
509 * update i_size. */
510 status = ocfs2_journal_access(handle, inode, bh,
511 OCFS2_JOURNAL_ACCESS_WRITE);
512 if (status < 0) {
513 mlog_errno(status);
514 goto leave;
515 }
516
517 prev_clusters = OCFS2_I(inode)->ip_clusters;
518
519 status = ocfs2_do_extend_allocation(osb,
520 inode,
521 clusters_to_add,
522 bh,
523 handle,
524 data_ac,
525 meta_ac,
526 &why);
527 if ((status < 0) && (status != -EAGAIN)) {
528 if (status != -ENOSPC)
529 mlog_errno(status);
530 goto leave;
531 }
532
533 status = ocfs2_journal_dirty(handle, bh);
534 if (status < 0) {
535 mlog_errno(status);
536 goto leave;
537 }
538
539 spin_lock(&OCFS2_I(inode)->ip_lock);
540 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
541 spin_unlock(&OCFS2_I(inode)->ip_lock);
542
543 if (why != RESTART_NONE && clusters_to_add) {
544 if (why == RESTART_META) {
545 mlog(0, "restarting function.\n");
546 restart_func = 1;
547 } else {
548 BUG_ON(why != RESTART_TRANS);
549
550 mlog(0, "restarting transaction.\n");
551 /* TODO: This can be more intelligent. */
552 credits = ocfs2_calc_extend_credits(osb->sb,
553 fe,
554 clusters_to_add);
Mark Fasheh1fc58142006-10-05 14:15:36 -0700555 status = ocfs2_extend_trans(handle->k_handle, credits);
Mark Fashehccd979b2005-12-15 14:31:24 -0800556 if (status < 0) {
557 /* handle still has to be committed at
558 * this point. */
559 status = -ENOMEM;
560 mlog_errno(status);
561 goto leave;
562 }
563 goto restarted_transaction;
564 }
565 }
566
Mark Fashehb06970532006-03-03 10:24:33 -0800567 mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
568 fe->i_clusters, (unsigned long long)fe->i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800569 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
570 OCFS2_I(inode)->ip_clusters, i_size_read(inode));
571
572leave:
573 if (drop_alloc_sem) {
574 up_write(&OCFS2_I(inode)->ip_alloc_sem);
575 drop_alloc_sem = 0;
576 }
577 if (handle) {
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700578 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800579 handle = NULL;
580 }
581 if (data_ac) {
582 ocfs2_free_alloc_context(data_ac);
583 data_ac = NULL;
584 }
585 if (meta_ac) {
586 ocfs2_free_alloc_context(meta_ac);
587 meta_ac = NULL;
588 }
589 if ((!status) && restart_func) {
590 restart_func = 0;
591 goto restart_all;
592 }
593 if (bh) {
594 brelse(bh);
595 bh = NULL;
596 }
597
598 mlog_exit(status);
599 return status;
600}
601
602/* Some parts of this taken from generic_cont_expand, which turned out
603 * to be too fragile to do exactly what we need without us having to
Mark Fasheh53013cb2006-05-05 19:04:03 -0700604 * worry about recursive locking in ->prepare_write() and
605 * ->commit_write(). */
Mark Fashehccd979b2005-12-15 14:31:24 -0800606static int ocfs2_write_zero_page(struct inode *inode,
607 u64 size)
608{
609 struct address_space *mapping = inode->i_mapping;
610 struct page *page;
611 unsigned long index;
612 unsigned int offset;
613 struct ocfs2_journal_handle *handle = NULL;
614 int ret;
615
616 offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */
617 /* ugh. in prepare/commit_write, if from==to==start of block, we
618 ** skip the prepare. make sure we never send an offset for the start
619 ** of a block
620 */
621 if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) {
622 offset++;
623 }
624 index = size >> PAGE_CACHE_SHIFT;
625
626 page = grab_cache_page(mapping, index);
627 if (!page) {
628 ret = -ENOMEM;
629 mlog_errno(ret);
630 goto out;
631 }
632
Mark Fasheh53013cb2006-05-05 19:04:03 -0700633 ret = ocfs2_prepare_write_nolock(inode, page, offset, offset);
Mark Fashehccd979b2005-12-15 14:31:24 -0800634 if (ret < 0) {
635 mlog_errno(ret);
636 goto out_unlock;
637 }
638
639 if (ocfs2_should_order_data(inode)) {
640 handle = ocfs2_start_walk_page_trans(inode, page, offset,
641 offset);
642 if (IS_ERR(handle)) {
643 ret = PTR_ERR(handle);
644 handle = NULL;
645 goto out_unlock;
646 }
647 }
648
649 /* must not update i_size! */
650 ret = block_commit_write(page, offset, offset);
651 if (ret < 0)
652 mlog_errno(ret);
653 else
654 ret = 0;
655
656 if (handle)
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700657 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800658out_unlock:
659 unlock_page(page);
660 page_cache_release(page);
661out:
662 return ret;
663}
664
665static int ocfs2_zero_extend(struct inode *inode,
666 u64 zero_to_size)
667{
668 int ret = 0;
669 u64 start_off;
670 struct super_block *sb = inode->i_sb;
671
672 start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
673 while (start_off < zero_to_size) {
674 ret = ocfs2_write_zero_page(inode, start_off);
675 if (ret < 0) {
676 mlog_errno(ret);
677 goto out;
678 }
679
680 start_off += sb->s_blocksize;
Mark Fashehe2057c52006-10-03 17:53:05 -0700681
682 /*
683 * Very large extends have the potential to lock up
684 * the cpu for extended periods of time.
685 */
686 cond_resched();
Mark Fashehccd979b2005-12-15 14:31:24 -0800687 }
688
689out:
690 return ret;
691}
692
Mark Fasheh53013cb2006-05-05 19:04:03 -0700693/*
694 * A tail_to_skip value > 0 indicates that we're being called from
695 * ocfs2_file_aio_write(). This has the following implications:
696 *
697 * - we don't want to update i_size
698 * - di_bh will be NULL, which is fine because it's only used in the
699 * case where we want to update i_size.
700 * - ocfs2_zero_extend() will then only be filling the hole created
701 * between i_size and the start of the write.
702 */
Mark Fashehccd979b2005-12-15 14:31:24 -0800703static int ocfs2_extend_file(struct inode *inode,
704 struct buffer_head *di_bh,
Mark Fasheh53013cb2006-05-05 19:04:03 -0700705 u64 new_i_size,
706 size_t tail_to_skip)
Mark Fashehccd979b2005-12-15 14:31:24 -0800707{
708 int ret = 0;
709 u32 clusters_to_add;
710
Mark Fasheh53013cb2006-05-05 19:04:03 -0700711 BUG_ON(!tail_to_skip && !di_bh);
712
Mark Fashehccd979b2005-12-15 14:31:24 -0800713 /* setattr sometimes calls us like this. */
714 if (new_i_size == 0)
715 goto out;
716
717 if (i_size_read(inode) == new_i_size)
718 goto out;
719 BUG_ON(new_i_size < i_size_read(inode));
720
721 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size) -
722 OCFS2_I(inode)->ip_clusters;
723
Mark Fasheh0effef72006-10-03 17:44:42 -0700724 /*
725 * protect the pages that ocfs2_zero_extend is going to be
726 * pulling into the page cache.. we do this before the
727 * metadata extend so that we don't get into the situation
728 * where we've extended the metadata but can't get the data
729 * lock to zero.
730 */
731 ret = ocfs2_data_lock(inode, 1);
732 if (ret < 0) {
733 mlog_errno(ret);
734 goto out;
735 }
Mark Fasheh53013cb2006-05-05 19:04:03 -0700736
Mark Fasheh0effef72006-10-03 17:44:42 -0700737 if (clusters_to_add) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800738 ret = ocfs2_extend_allocation(inode, clusters_to_add);
739 if (ret < 0) {
740 mlog_errno(ret);
Mark Fasheh53013cb2006-05-05 19:04:03 -0700741 goto out_unlock;
Mark Fashehccd979b2005-12-15 14:31:24 -0800742 }
Mark Fasheh0effef72006-10-03 17:44:42 -0700743 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800744
Mark Fasheh0effef72006-10-03 17:44:42 -0700745 /*
746 * Call this even if we don't add any clusters to the tree. We
747 * still need to zero the area between the old i_size and the
748 * new i_size.
749 */
750 ret = ocfs2_zero_extend(inode, (u64)new_i_size - tail_to_skip);
751 if (ret < 0) {
752 mlog_errno(ret);
753 goto out_unlock;
Mark Fashehccd979b2005-12-15 14:31:24 -0800754 }
755
Mark Fasheh53013cb2006-05-05 19:04:03 -0700756 if (!tail_to_skip) {
757 /* We're being called from ocfs2_setattr() which wants
758 * us to update i_size */
759 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
760 if (ret < 0)
761 mlog_errno(ret);
762 }
763
764out_unlock:
Mark Fasheh0effef72006-10-03 17:44:42 -0700765 ocfs2_data_unlock(inode, 1);
Mark Fasheh53013cb2006-05-05 19:04:03 -0700766
Mark Fashehccd979b2005-12-15 14:31:24 -0800767out:
768 return ret;
769}
770
771int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
772{
773 int status = 0, size_change;
774 struct inode *inode = dentry->d_inode;
775 struct super_block *sb = inode->i_sb;
776 struct ocfs2_super *osb = OCFS2_SB(sb);
777 struct buffer_head *bh = NULL;
778 struct ocfs2_journal_handle *handle = NULL;
779
780 mlog_entry("(0x%p, '%.*s')\n", dentry,
781 dentry->d_name.len, dentry->d_name.name);
782
783 if (attr->ia_valid & ATTR_MODE)
784 mlog(0, "mode change: %d\n", attr->ia_mode);
785 if (attr->ia_valid & ATTR_UID)
786 mlog(0, "uid change: %d\n", attr->ia_uid);
787 if (attr->ia_valid & ATTR_GID)
788 mlog(0, "gid change: %d\n", attr->ia_gid);
789 if (attr->ia_valid & ATTR_SIZE)
790 mlog(0, "size change...\n");
791 if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME))
792 mlog(0, "time change...\n");
793
794#define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
795 | ATTR_GID | ATTR_UID | ATTR_MODE)
796 if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) {
797 mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid);
798 return 0;
799 }
800
801 status = inode_change_ok(inode, attr);
802 if (status)
803 return status;
804
805 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
806 if (size_change) {
807 status = ocfs2_rw_lock(inode, 1);
808 if (status < 0) {
809 mlog_errno(status);
810 goto bail;
811 }
812 }
813
Mark Fasheh4bcec182006-10-09 16:02:40 -0700814 status = ocfs2_meta_lock(inode, &bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800815 if (status < 0) {
816 if (status != -ENOENT)
817 mlog_errno(status);
818 goto bail_unlock_rw;
819 }
820
821 if (size_change && attr->ia_size != i_size_read(inode)) {
822 if (i_size_read(inode) > attr->ia_size)
823 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
824 else
Mark Fasheh53013cb2006-05-05 19:04:03 -0700825 status = ocfs2_extend_file(inode, bh, attr->ia_size, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -0800826 if (status < 0) {
827 if (status != -ENOSPC)
828 mlog_errno(status);
829 status = -ENOSPC;
830 goto bail_unlock;
831 }
832 }
833
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700834 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800835 if (IS_ERR(handle)) {
836 status = PTR_ERR(handle);
837 mlog_errno(status);
838 goto bail_unlock;
839 }
840
841 status = inode_setattr(inode, attr);
842 if (status < 0) {
843 mlog_errno(status);
844 goto bail_commit;
845 }
846
847 status = ocfs2_mark_inode_dirty(handle, inode, bh);
848 if (status < 0)
849 mlog_errno(status);
850
851bail_commit:
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700852 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800853bail_unlock:
854 ocfs2_meta_unlock(inode, 1);
855bail_unlock_rw:
856 if (size_change)
857 ocfs2_rw_unlock(inode, 1);
858bail:
859 if (bh)
860 brelse(bh);
861
862 mlog_exit(status);
863 return status;
864}
865
866int ocfs2_getattr(struct vfsmount *mnt,
867 struct dentry *dentry,
868 struct kstat *stat)
869{
870 struct inode *inode = dentry->d_inode;
871 struct super_block *sb = dentry->d_inode->i_sb;
872 struct ocfs2_super *osb = sb->s_fs_info;
873 int err;
874
875 mlog_entry_void();
876
877 err = ocfs2_inode_revalidate(dentry);
878 if (err) {
879 if (err != -ENOENT)
880 mlog_errno(err);
881 goto bail;
882 }
883
884 generic_fillattr(inode, stat);
885
886 /* We set the blksize from the cluster size for performance */
887 stat->blksize = osb->s_clustersize;
888
889bail:
890 mlog_exit(err);
891
892 return err;
893}
894
895static int ocfs2_write_remove_suid(struct inode *inode)
896{
897 int ret;
898 struct buffer_head *bh = NULL;
899 struct ocfs2_inode_info *oi = OCFS2_I(inode);
900 struct ocfs2_journal_handle *handle;
901 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
902 struct ocfs2_dinode *di;
903
Mark Fashehb06970532006-03-03 10:24:33 -0800904 mlog_entry("(Inode %llu, mode 0%o)\n",
905 (unsigned long long)oi->ip_blkno, inode->i_mode);
Mark Fashehccd979b2005-12-15 14:31:24 -0800906
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700907 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800908 if (handle == NULL) {
909 ret = -ENOMEM;
910 mlog_errno(ret);
911 goto out;
912 }
913
914 ret = ocfs2_read_block(osb, oi->ip_blkno, &bh, OCFS2_BH_CACHED, inode);
915 if (ret < 0) {
916 mlog_errno(ret);
917 goto out_trans;
918 }
919
920 ret = ocfs2_journal_access(handle, inode, bh,
921 OCFS2_JOURNAL_ACCESS_WRITE);
922 if (ret < 0) {
923 mlog_errno(ret);
924 goto out_bh;
925 }
926
927 inode->i_mode &= ~S_ISUID;
928 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
929 inode->i_mode &= ~S_ISGID;
930
931 di = (struct ocfs2_dinode *) bh->b_data;
932 di->i_mode = cpu_to_le16(inode->i_mode);
933
934 ret = ocfs2_journal_dirty(handle, bh);
935 if (ret < 0)
936 mlog_errno(ret);
937out_bh:
938 brelse(bh);
939out_trans:
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700940 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800941out:
942 mlog_exit(ret);
943 return ret;
944}
945
946static inline int ocfs2_write_should_remove_suid(struct inode *inode)
947{
948 mode_t mode = inode->i_mode;
949
950 if (!capable(CAP_FSETID)) {
951 if (unlikely(mode & S_ISUID))
952 return 1;
953
954 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
955 return 1;
956 }
957 return 0;
958}
959
960static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
Badari Pulavarty027445c2006-09-30 23:28:46 -0700961 const struct iovec *iov,
962 unsigned long nr_segs,
Mark Fashehccd979b2005-12-15 14:31:24 -0800963 loff_t pos)
964{
Mark Fashehccd979b2005-12-15 14:31:24 -0800965 int ret, rw_level = -1, meta_level = -1, have_alloc_sem = 0;
966 u32 clusters;
967 struct file *filp = iocb->ki_filp;
968 struct inode *inode = filp->f_dentry->d_inode;
969 loff_t newsize, saved_pos;
Mark Fashehccd979b2005-12-15 14:31:24 -0800970
Badari Pulavarty027445c2006-09-30 23:28:46 -0700971 mlog_entry("(0x%p, %u, '%.*s')\n", filp,
972 (unsigned int)nr_segs,
Mark Fashehccd979b2005-12-15 14:31:24 -0800973 filp->f_dentry->d_name.len,
974 filp->f_dentry->d_name.name);
975
976 /* happy write of zero bytes */
Badari Pulavarty027445c2006-09-30 23:28:46 -0700977 if (iocb->ki_left == 0)
Mark Fashehccd979b2005-12-15 14:31:24 -0800978 return 0;
979
980 if (!inode) {
981 mlog(0, "bad inode\n");
982 return -EIO;
983 }
984
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800985 mutex_lock(&inode->i_mutex);
986 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
Mark Fashehccd979b2005-12-15 14:31:24 -0800987 if (filp->f_flags & O_DIRECT) {
988 have_alloc_sem = 1;
989 down_read(&inode->i_alloc_sem);
990 }
991
992 /* concurrent O_DIRECT writes are allowed */
993 rw_level = (filp->f_flags & O_DIRECT) ? 0 : 1;
994 ret = ocfs2_rw_lock(inode, rw_level);
995 if (ret < 0) {
996 rw_level = -1;
997 mlog_errno(ret);
998 goto out;
999 }
1000
1001 /*
1002 * We sample i_size under a read level meta lock to see if our write
1003 * is extending the file, if it is we back off and get a write level
1004 * meta lock.
1005 */
1006 meta_level = (filp->f_flags & O_APPEND) ? 1 : 0;
1007 for(;;) {
Mark Fasheh4bcec182006-10-09 16:02:40 -07001008 ret = ocfs2_meta_lock(inode, NULL, meta_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08001009 if (ret < 0) {
1010 meta_level = -1;
1011 mlog_errno(ret);
1012 goto out;
1013 }
1014
1015 /* Clear suid / sgid if necessary. We do this here
1016 * instead of later in the write path because
1017 * remove_suid() calls ->setattr without any hint that
1018 * we may have already done our cluster locking. Since
1019 * ocfs2_setattr() *must* take cluster locks to
1020 * proceeed, this will lead us to recursively lock the
1021 * inode. There's also the dinode i_size state which
1022 * can be lost via setattr during extending writes (we
1023 * set inode->i_size at the end of a write. */
1024 if (ocfs2_write_should_remove_suid(inode)) {
1025 if (meta_level == 0) {
1026 ocfs2_meta_unlock(inode, meta_level);
1027 meta_level = 1;
1028 continue;
1029 }
1030
1031 ret = ocfs2_write_remove_suid(inode);
1032 if (ret < 0) {
1033 mlog_errno(ret);
1034 goto out;
1035 }
1036 }
1037
1038 /* work on a copy of ppos until we're sure that we won't have
1039 * to recalculate it due to relocking. */
1040 if (filp->f_flags & O_APPEND) {
1041 saved_pos = i_size_read(inode);
1042 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos);
1043 } else {
1044 saved_pos = iocb->ki_pos;
1045 }
Badari Pulavarty027445c2006-09-30 23:28:46 -07001046 newsize = iocb->ki_left + saved_pos;
Mark Fashehccd979b2005-12-15 14:31:24 -08001047
Mark Fasheh215c7f92006-02-01 16:42:10 -08001048 mlog(0, "pos=%lld newsize=%lld cursize=%lld\n",
1049 (long long) saved_pos, (long long) newsize,
1050 (long long) i_size_read(inode));
Mark Fashehccd979b2005-12-15 14:31:24 -08001051
1052 /* No need for a higher level metadata lock if we're
1053 * never going past i_size. */
1054 if (newsize <= i_size_read(inode))
1055 break;
1056
1057 if (meta_level == 0) {
1058 ocfs2_meta_unlock(inode, meta_level);
1059 meta_level = 1;
1060 continue;
1061 }
1062
1063 spin_lock(&OCFS2_I(inode)->ip_lock);
1064 clusters = ocfs2_clusters_for_bytes(inode->i_sb, newsize) -
1065 OCFS2_I(inode)->ip_clusters;
1066 spin_unlock(&OCFS2_I(inode)->ip_lock);
1067
1068 mlog(0, "Writing at EOF, may need more allocation: "
Mark Fasheh215c7f92006-02-01 16:42:10 -08001069 "i_size = %lld, newsize = %lld, need %u clusters\n",
1070 (long long) i_size_read(inode), (long long) newsize,
1071 clusters);
Mark Fashehccd979b2005-12-15 14:31:24 -08001072
1073 /* We only want to continue the rest of this loop if
1074 * our extend will actually require more
1075 * allocation. */
1076 if (!clusters)
1077 break;
1078
Badari Pulavarty027445c2006-09-30 23:28:46 -07001079 ret = ocfs2_extend_file(inode, NULL, newsize, iocb->ki_left);
Mark Fashehccd979b2005-12-15 14:31:24 -08001080 if (ret < 0) {
1081 if (ret != -ENOSPC)
1082 mlog_errno(ret);
1083 goto out;
1084 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001085 break;
1086 }
1087
1088 /* ok, we're done with i_size and alloc work */
1089 iocb->ki_pos = saved_pos;
1090 ocfs2_meta_unlock(inode, meta_level);
1091 meta_level = -1;
1092
1093 /* communicate with ocfs2_dio_end_io */
1094 ocfs2_iocb_set_rw_locked(iocb);
1095
Badari Pulavarty027445c2006-09-30 23:28:46 -07001096 ret = generic_file_aio_write_nolock(iocb, iov, nr_segs, iocb->ki_pos);
Mark Fashehccd979b2005-12-15 14:31:24 -08001097
1098 /* buffered aio wouldn't have proper lock coverage today */
1099 BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
1100
1101 /*
1102 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
1103 * function pointer which is called when o_direct io completes so that
1104 * it can unlock our rw lock. (it's the clustered equivalent of
1105 * i_alloc_sem; protects truncate from racing with pending ios).
1106 * Unfortunately there are error cases which call end_io and others
1107 * that don't. so we don't have to unlock the rw_lock if either an
1108 * async dio is going to do it in the future or an end_io after an
1109 * error has already done it.
1110 */
1111 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
1112 rw_level = -1;
1113 have_alloc_sem = 0;
1114 }
1115
1116out:
1117 if (meta_level != -1)
1118 ocfs2_meta_unlock(inode, meta_level);
1119 if (have_alloc_sem)
1120 up_read(&inode->i_alloc_sem);
1121 if (rw_level != -1)
1122 ocfs2_rw_unlock(inode, rw_level);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001123 mutex_unlock(&inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08001124
1125 mlog_exit(ret);
1126 return ret;
1127}
1128
1129static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
Badari Pulavarty027445c2006-09-30 23:28:46 -07001130 const struct iovec *iov,
1131 unsigned long nr_segs,
Mark Fashehccd979b2005-12-15 14:31:24 -08001132 loff_t pos)
1133{
1134 int ret = 0, rw_level = -1, have_alloc_sem = 0;
1135 struct file *filp = iocb->ki_filp;
1136 struct inode *inode = filp->f_dentry->d_inode;
Mark Fashehccd979b2005-12-15 14:31:24 -08001137
Badari Pulavarty027445c2006-09-30 23:28:46 -07001138 mlog_entry("(0x%p, %u, '%.*s')\n", filp,
1139 (unsigned int)nr_segs,
Mark Fashehccd979b2005-12-15 14:31:24 -08001140 filp->f_dentry->d_name.len,
1141 filp->f_dentry->d_name.name);
1142
1143 if (!inode) {
1144 ret = -EINVAL;
1145 mlog_errno(ret);
1146 goto bail;
1147 }
1148
Mark Fashehccd979b2005-12-15 14:31:24 -08001149 /*
1150 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
1151 * need locks to protect pending reads from racing with truncate.
1152 */
1153 if (filp->f_flags & O_DIRECT) {
1154 down_read(&inode->i_alloc_sem);
1155 have_alloc_sem = 1;
1156
1157 ret = ocfs2_rw_lock(inode, 0);
1158 if (ret < 0) {
1159 mlog_errno(ret);
1160 goto bail;
1161 }
1162 rw_level = 0;
1163 /* communicate with ocfs2_dio_end_io */
1164 ocfs2_iocb_set_rw_locked(iocb);
1165 }
1166
Mark Fashehc4374f82006-05-05 19:04:35 -07001167 /*
1168 * We're fine letting folks race truncates and extending
1169 * writes with read across the cluster, just like they can
1170 * locally. Hence no rw_lock during read.
1171 *
1172 * Take and drop the meta data lock to update inode fields
1173 * like i_size. This allows the checks down below
1174 * generic_file_aio_read() a chance of actually working.
1175 */
Mark Fasheh4bcec182006-10-09 16:02:40 -07001176 ret = ocfs2_meta_lock(inode, NULL, 0);
Mark Fashehc4374f82006-05-05 19:04:35 -07001177 if (ret < 0) {
1178 mlog_errno(ret);
1179 goto bail;
1180 }
1181 ocfs2_meta_unlock(inode, 0);
1182
Badari Pulavarty027445c2006-09-30 23:28:46 -07001183 ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
Mark Fashehccd979b2005-12-15 14:31:24 -08001184 if (ret == -EINVAL)
1185 mlog(ML_ERROR, "generic_file_aio_read returned -EINVAL\n");
1186
1187 /* buffered aio wouldn't have proper lock coverage today */
1188 BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
1189
1190 /* see ocfs2_file_aio_write */
1191 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
1192 rw_level = -1;
1193 have_alloc_sem = 0;
1194 }
1195
1196bail:
1197 if (have_alloc_sem)
1198 up_read(&inode->i_alloc_sem);
1199 if (rw_level != -1)
1200 ocfs2_rw_unlock(inode, rw_level);
1201 mlog_exit(ret);
1202
1203 return ret;
1204}
1205
1206struct inode_operations ocfs2_file_iops = {
1207 .setattr = ocfs2_setattr,
1208 .getattr = ocfs2_getattr,
1209};
1210
1211struct inode_operations ocfs2_special_file_iops = {
1212 .setattr = ocfs2_setattr,
1213 .getattr = ocfs2_getattr,
1214};
1215
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001216const struct file_operations ocfs2_fops = {
Mark Fashehccd979b2005-12-15 14:31:24 -08001217 .read = do_sync_read,
1218 .write = do_sync_write,
1219 .sendfile = generic_file_sendfile,
1220 .mmap = ocfs2_mmap,
1221 .fsync = ocfs2_sync_file,
1222 .release = ocfs2_file_release,
1223 .open = ocfs2_file_open,
1224 .aio_read = ocfs2_file_aio_read,
1225 .aio_write = ocfs2_file_aio_write,
Herbert Poetzlca4d1472006-07-03 17:27:12 -07001226 .ioctl = ocfs2_ioctl,
Mark Fashehccd979b2005-12-15 14:31:24 -08001227};
1228
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001229const struct file_operations ocfs2_dops = {
Mark Fashehccd979b2005-12-15 14:31:24 -08001230 .read = generic_read_dir,
1231 .readdir = ocfs2_readdir,
1232 .fsync = ocfs2_sync_file,
Herbert Poetzlca4d1472006-07-03 17:27:12 -07001233 .ioctl = ocfs2_ioctl,
Mark Fashehccd979b2005-12-15 14:31:24 -08001234};