blob: 8a4048b55fdc41d5dda4dc5e8b9908e4a765cc64 [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>
33
34#define MLOG_MASK_PREFIX ML_INODE
35#include <cluster/masklog.h>
36
37#include "ocfs2.h"
38
39#include "alloc.h"
40#include "aops.h"
41#include "dir.h"
42#include "dlmglue.h"
43#include "extent_map.h"
44#include "file.h"
45#include "sysfile.h"
46#include "inode.h"
47#include "journal.h"
48#include "mmap.h"
49#include "suballoc.h"
50#include "super.h"
51
52#include "buffer_head_io.h"
53
54static int ocfs2_sync_inode(struct inode *inode)
55{
56 filemap_fdatawrite(inode->i_mapping);
57 return sync_mapping_buffers(inode->i_mapping);
58}
59
60static int ocfs2_file_open(struct inode *inode, struct file *file)
61{
62 int status;
63 int mode = file->f_flags;
64 struct ocfs2_inode_info *oi = OCFS2_I(inode);
65
66 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
67 file->f_dentry->d_name.len, file->f_dentry->d_name.name);
68
69 spin_lock(&oi->ip_lock);
70
71 /* Check that the inode hasn't been wiped from disk by another
72 * node. If it hasn't then we're safe as long as we hold the
73 * spin lock until our increment of open count. */
74 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
75 spin_unlock(&oi->ip_lock);
76
77 status = -ENOENT;
78 goto leave;
79 }
80
81 if (mode & O_DIRECT)
82 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
83
84 oi->ip_open_count++;
85 spin_unlock(&oi->ip_lock);
86 status = 0;
87leave:
88 mlog_exit(status);
89 return status;
90}
91
92static int ocfs2_file_release(struct inode *inode, struct file *file)
93{
94 struct ocfs2_inode_info *oi = OCFS2_I(inode);
95
96 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
97 file->f_dentry->d_name.len,
98 file->f_dentry->d_name.name);
99
100 spin_lock(&oi->ip_lock);
101 if (!--oi->ip_open_count)
102 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
103 spin_unlock(&oi->ip_lock);
104
105 mlog_exit(0);
106
107 return 0;
108}
109
110static int ocfs2_sync_file(struct file *file,
111 struct dentry *dentry,
112 int datasync)
113{
114 int err = 0;
115 journal_t *journal;
116 struct inode *inode = dentry->d_inode;
117 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
118
119 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync,
120 dentry->d_name.len, dentry->d_name.name);
121
122 err = ocfs2_sync_inode(dentry->d_inode);
123 if (err)
124 goto bail;
125
126 journal = osb->journal->j_journal;
127 err = journal_force_commit(journal);
128
129bail:
130 mlog_exit(err);
131
132 return (err < 0) ? -EIO : 0;
133}
134
135int ocfs2_set_inode_size(struct ocfs2_journal_handle *handle,
136 struct inode *inode,
137 struct buffer_head *fe_bh,
138 u64 new_i_size)
139{
140 int status;
141
142 mlog_entry_void();
143 i_size_write(inode, new_i_size);
144 inode->i_blocks = ocfs2_align_bytes_to_sectors(new_i_size);
145 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
146
147 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
148 if (status < 0) {
149 mlog_errno(status);
150 goto bail;
151 }
152
153bail:
154 mlog_exit(status);
155 return status;
156}
157
158static int ocfs2_simple_size_update(struct inode *inode,
159 struct buffer_head *di_bh,
160 u64 new_i_size)
161{
162 int ret;
163 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
164 struct ocfs2_journal_handle *handle = NULL;
165
166 handle = ocfs2_start_trans(osb, NULL,
167 OCFS2_INODE_UPDATE_CREDITS);
168 if (handle == NULL) {
169 ret = -ENOMEM;
170 mlog_errno(ret);
171 goto out;
172 }
173
174 ret = ocfs2_set_inode_size(handle, inode, di_bh,
175 new_i_size);
176 if (ret < 0)
177 mlog_errno(ret);
178
179 ocfs2_commit_trans(handle);
180out:
181 return ret;
182}
183
184static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
185 struct inode *inode,
186 struct buffer_head *fe_bh,
187 u64 new_i_size)
188{
189 int status;
190 struct ocfs2_journal_handle *handle;
191
192 mlog_entry_void();
193
194 /* TODO: This needs to actually orphan the inode in this
195 * transaction. */
196
197 handle = ocfs2_start_trans(osb, NULL, OCFS2_INODE_UPDATE_CREDITS);
198 if (IS_ERR(handle)) {
199 status = PTR_ERR(handle);
200 mlog_errno(status);
201 goto out;
202 }
203
204 status = ocfs2_set_inode_size(handle, inode, fe_bh, new_i_size);
205 if (status < 0)
206 mlog_errno(status);
207
208 ocfs2_commit_trans(handle);
209out:
210 mlog_exit(status);
211 return status;
212}
213
214static int ocfs2_truncate_file(struct inode *inode,
215 struct buffer_head *di_bh,
216 u64 new_i_size)
217{
218 int status = 0;
219 struct ocfs2_dinode *fe = NULL;
220 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
221 struct ocfs2_truncate_context *tc = NULL;
222
223 mlog_entry("(inode = %"MLFu64", new_i_size = %"MLFu64"\n",
224 OCFS2_I(inode)->ip_blkno, new_i_size);
225
226 truncate_inode_pages(inode->i_mapping, new_i_size);
227
228 fe = (struct ocfs2_dinode *) di_bh->b_data;
229 if (!OCFS2_IS_VALID_DINODE(fe)) {
230 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
231 status = -EIO;
232 goto bail;
233 }
234
235 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
236 "Inode %"MLFu64", inode i_size = %lld != di "
237 "i_size = %"MLFu64", i_flags = 0x%x\n",
238 OCFS2_I(inode)->ip_blkno,
239 i_size_read(inode),
240 le64_to_cpu(fe->i_size), le32_to_cpu(fe->i_flags));
241
242 if (new_i_size > le64_to_cpu(fe->i_size)) {
243 mlog(0, "asked to truncate file with size (%"MLFu64") "
244 "to size (%"MLFu64")!\n",
245 le64_to_cpu(fe->i_size), new_i_size);
246 status = -EINVAL;
247 mlog_errno(status);
248 goto bail;
249 }
250
251 mlog(0, "inode %"MLFu64", i_size = %"MLFu64", new_i_size = %"MLFu64"\n",
252 le64_to_cpu(fe->i_blkno), le64_to_cpu(fe->i_size), new_i_size);
253
254 /* lets handle the simple truncate cases before doing any more
255 * cluster locking. */
256 if (new_i_size == le64_to_cpu(fe->i_size))
257 goto bail;
258
259 if (le32_to_cpu(fe->i_clusters) ==
260 ocfs2_clusters_for_bytes(osb->sb, new_i_size)) {
261 mlog(0, "fe->i_clusters = %u, so we do a simple truncate\n",
262 fe->i_clusters);
263 /* No allocation change is required, so lets fast path
264 * this truncate. */
265 status = ocfs2_simple_size_update(inode, di_bh, new_i_size);
266 if (status < 0)
267 mlog_errno(status);
268 goto bail;
269 }
270
271 /* This forces other nodes to sync and drop their pages */
272 status = ocfs2_data_lock(inode, 1);
273 if (status < 0) {
274 mlog_errno(status);
275 goto bail;
276 }
277 ocfs2_data_unlock(inode, 1);
278
279 /* alright, we're going to need to do a full blown alloc size
280 * change. Orphan the inode so that recovery can complete the
281 * truncate if necessary. This does the task of marking
282 * i_size. */
283 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
284 if (status < 0) {
285 mlog_errno(status);
286 goto bail;
287 }
288
289 status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc);
290 if (status < 0) {
291 mlog_errno(status);
292 goto bail;
293 }
294
295 status = ocfs2_commit_truncate(osb, inode, di_bh, tc);
296 if (status < 0) {
297 mlog_errno(status);
298 goto bail;
299 }
300
301 /* TODO: orphan dir cleanup here. */
302bail:
303
304 mlog_exit(status);
305 return status;
306}
307
308/*
309 * extend allocation only here.
310 * we'll update all the disk stuff, and oip->alloc_size
311 *
312 * expect stuff to be locked, a transaction started and enough data /
313 * metadata reservations in the contexts.
314 *
315 * Will return -EAGAIN, and a reason if a restart is needed.
316 * If passed in, *reason will always be set, even in error.
317 */
318int ocfs2_do_extend_allocation(struct ocfs2_super *osb,
319 struct inode *inode,
320 u32 clusters_to_add,
321 struct buffer_head *fe_bh,
322 struct ocfs2_journal_handle *handle,
323 struct ocfs2_alloc_context *data_ac,
324 struct ocfs2_alloc_context *meta_ac,
325 enum ocfs2_alloc_restarted *reason_ret)
326{
327 int status = 0;
328 int free_extents;
329 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
330 enum ocfs2_alloc_restarted reason = RESTART_NONE;
331 u32 bit_off, num_bits;
332 u64 block;
333
334 BUG_ON(!clusters_to_add);
335
336 free_extents = ocfs2_num_free_extents(osb, inode, fe);
337 if (free_extents < 0) {
338 status = free_extents;
339 mlog_errno(status);
340 goto leave;
341 }
342
343 /* there are two cases which could cause us to EAGAIN in the
344 * we-need-more-metadata case:
345 * 1) we haven't reserved *any*
346 * 2) we are so fragmented, we've needed to add metadata too
347 * many times. */
348 if (!free_extents && !meta_ac) {
349 mlog(0, "we haven't reserved any metadata!\n");
350 status = -EAGAIN;
351 reason = RESTART_META;
352 goto leave;
353 } else if ((!free_extents)
354 && (ocfs2_alloc_context_bits_left(meta_ac)
355 < ocfs2_extend_meta_needed(fe))) {
356 mlog(0, "filesystem is really fragmented...\n");
357 status = -EAGAIN;
358 reason = RESTART_META;
359 goto leave;
360 }
361
362 status = ocfs2_claim_clusters(osb, handle, data_ac, 1,
363 &bit_off, &num_bits);
364 if (status < 0) {
365 if (status != -ENOSPC)
366 mlog_errno(status);
367 goto leave;
368 }
369
370 BUG_ON(num_bits > clusters_to_add);
371
372 /* reserve our write early -- insert_extent may update the inode */
373 status = ocfs2_journal_access(handle, inode, fe_bh,
374 OCFS2_JOURNAL_ACCESS_WRITE);
375 if (status < 0) {
376 mlog_errno(status);
377 goto leave;
378 }
379
380 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
381 mlog(0, "Allocating %u clusters at block %u for inode %"MLFu64"\n",
382 num_bits, bit_off, OCFS2_I(inode)->ip_blkno);
383 status = ocfs2_insert_extent(osb, handle, inode, fe_bh, block,
384 num_bits, meta_ac);
385 if (status < 0) {
386 mlog_errno(status);
387 goto leave;
388 }
389
390 le32_add_cpu(&fe->i_clusters, num_bits);
391 spin_lock(&OCFS2_I(inode)->ip_lock);
392 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
393 spin_unlock(&OCFS2_I(inode)->ip_lock);
394
395 status = ocfs2_journal_dirty(handle, fe_bh);
396 if (status < 0) {
397 mlog_errno(status);
398 goto leave;
399 }
400
401 clusters_to_add -= num_bits;
402
403 if (clusters_to_add) {
404 mlog(0, "need to alloc once more, clusters = %u, wanted = "
405 "%u\n", fe->i_clusters, clusters_to_add);
406 status = -EAGAIN;
407 reason = RESTART_TRANS;
408 }
409
410leave:
411 mlog_exit(status);
412 if (reason_ret)
413 *reason_ret = reason;
414 return status;
415}
416
417static int ocfs2_extend_allocation(struct inode *inode,
418 u32 clusters_to_add)
419{
420 int status = 0;
421 int restart_func = 0;
422 int drop_alloc_sem = 0;
423 int credits, num_free_extents;
424 u32 prev_clusters;
425 struct buffer_head *bh = NULL;
426 struct ocfs2_dinode *fe = NULL;
427 struct ocfs2_journal_handle *handle = NULL;
428 struct ocfs2_alloc_context *data_ac = NULL;
429 struct ocfs2_alloc_context *meta_ac = NULL;
430 enum ocfs2_alloc_restarted why;
431 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
432
433 mlog_entry("(clusters_to_add = %u)\n", clusters_to_add);
434
435 status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &bh,
436 OCFS2_BH_CACHED, inode);
437 if (status < 0) {
438 mlog_errno(status);
439 goto leave;
440 }
441
442 fe = (struct ocfs2_dinode *) bh->b_data;
443 if (!OCFS2_IS_VALID_DINODE(fe)) {
444 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
445 status = -EIO;
446 goto leave;
447 }
448
449restart_all:
450 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
451
452 mlog(0, "extend inode %"MLFu64", i_size = %lld, fe->i_clusters = %u, "
453 "clusters_to_add = %u\n",
454 OCFS2_I(inode)->ip_blkno, i_size_read(inode),
455 fe->i_clusters, clusters_to_add);
456
457 handle = ocfs2_alloc_handle(osb);
458 if (handle == NULL) {
459 status = -ENOMEM;
460 mlog_errno(status);
461 goto leave;
462 }
463
464 num_free_extents = ocfs2_num_free_extents(osb,
465 inode,
466 fe);
467 if (num_free_extents < 0) {
468 status = num_free_extents;
469 mlog_errno(status);
470 goto leave;
471 }
472
473 if (!num_free_extents) {
474 status = ocfs2_reserve_new_metadata(osb,
475 handle,
476 fe,
477 &meta_ac);
478 if (status < 0) {
479 if (status != -ENOSPC)
480 mlog_errno(status);
481 goto leave;
482 }
483 }
484
485 status = ocfs2_reserve_clusters(osb,
486 handle,
487 clusters_to_add,
488 &data_ac);
489 if (status < 0) {
490 if (status != -ENOSPC)
491 mlog_errno(status);
492 goto leave;
493 }
494
495 /* blocks peope in read/write from reading our allocation
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800496 * until we're done changing it. We depend on i_mutex to block
Mark Fashehccd979b2005-12-15 14:31:24 -0800497 * other extend/truncate calls while we're here. Ordering wrt
498 * start_trans is important here -- always do it before! */
499 down_write(&OCFS2_I(inode)->ip_alloc_sem);
500 drop_alloc_sem = 1;
501
502 credits = ocfs2_calc_extend_credits(osb->sb, fe, clusters_to_add);
503 handle = ocfs2_start_trans(osb, handle, credits);
504 if (IS_ERR(handle)) {
505 status = PTR_ERR(handle);
506 handle = NULL;
507 mlog_errno(status);
508 goto leave;
509 }
510
511restarted_transaction:
512 /* reserve a write to the file entry early on - that we if we
513 * run out of credits in the allocation path, we can still
514 * update i_size. */
515 status = ocfs2_journal_access(handle, inode, bh,
516 OCFS2_JOURNAL_ACCESS_WRITE);
517 if (status < 0) {
518 mlog_errno(status);
519 goto leave;
520 }
521
522 prev_clusters = OCFS2_I(inode)->ip_clusters;
523
524 status = ocfs2_do_extend_allocation(osb,
525 inode,
526 clusters_to_add,
527 bh,
528 handle,
529 data_ac,
530 meta_ac,
531 &why);
532 if ((status < 0) && (status != -EAGAIN)) {
533 if (status != -ENOSPC)
534 mlog_errno(status);
535 goto leave;
536 }
537
538 status = ocfs2_journal_dirty(handle, bh);
539 if (status < 0) {
540 mlog_errno(status);
541 goto leave;
542 }
543
544 spin_lock(&OCFS2_I(inode)->ip_lock);
545 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
546 spin_unlock(&OCFS2_I(inode)->ip_lock);
547
548 if (why != RESTART_NONE && clusters_to_add) {
549 if (why == RESTART_META) {
550 mlog(0, "restarting function.\n");
551 restart_func = 1;
552 } else {
553 BUG_ON(why != RESTART_TRANS);
554
555 mlog(0, "restarting transaction.\n");
556 /* TODO: This can be more intelligent. */
557 credits = ocfs2_calc_extend_credits(osb->sb,
558 fe,
559 clusters_to_add);
560 status = ocfs2_extend_trans(handle, credits);
561 if (status < 0) {
562 /* handle still has to be committed at
563 * this point. */
564 status = -ENOMEM;
565 mlog_errno(status);
566 goto leave;
567 }
568 goto restarted_transaction;
569 }
570 }
571
572 mlog(0, "fe: i_clusters = %u, i_size=%"MLFu64"\n",
573 fe->i_clusters, fe->i_size);
574 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
575 OCFS2_I(inode)->ip_clusters, i_size_read(inode));
576
577leave:
578 if (drop_alloc_sem) {
579 up_write(&OCFS2_I(inode)->ip_alloc_sem);
580 drop_alloc_sem = 0;
581 }
582 if (handle) {
583 ocfs2_commit_trans(handle);
584 handle = NULL;
585 }
586 if (data_ac) {
587 ocfs2_free_alloc_context(data_ac);
588 data_ac = NULL;
589 }
590 if (meta_ac) {
591 ocfs2_free_alloc_context(meta_ac);
592 meta_ac = NULL;
593 }
594 if ((!status) && restart_func) {
595 restart_func = 0;
596 goto restart_all;
597 }
598 if (bh) {
599 brelse(bh);
600 bh = NULL;
601 }
602
603 mlog_exit(status);
604 return status;
605}
606
607/* Some parts of this taken from generic_cont_expand, which turned out
608 * to be too fragile to do exactly what we need without us having to
609 * worry about recursive locking in ->commit_write(). */
610static int ocfs2_write_zero_page(struct inode *inode,
611 u64 size)
612{
613 struct address_space *mapping = inode->i_mapping;
614 struct page *page;
615 unsigned long index;
616 unsigned int offset;
617 struct ocfs2_journal_handle *handle = NULL;
618 int ret;
619
620 offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */
621 /* ugh. in prepare/commit_write, if from==to==start of block, we
622 ** skip the prepare. make sure we never send an offset for the start
623 ** of a block
624 */
625 if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) {
626 offset++;
627 }
628 index = size >> PAGE_CACHE_SHIFT;
629
630 page = grab_cache_page(mapping, index);
631 if (!page) {
632 ret = -ENOMEM;
633 mlog_errno(ret);
634 goto out;
635 }
636
637 ret = ocfs2_prepare_write(NULL, page, offset, offset);
638 if (ret < 0) {
639 mlog_errno(ret);
640 goto out_unlock;
641 }
642
643 if (ocfs2_should_order_data(inode)) {
644 handle = ocfs2_start_walk_page_trans(inode, page, offset,
645 offset);
646 if (IS_ERR(handle)) {
647 ret = PTR_ERR(handle);
648 handle = NULL;
649 goto out_unlock;
650 }
651 }
652
653 /* must not update i_size! */
654 ret = block_commit_write(page, offset, offset);
655 if (ret < 0)
656 mlog_errno(ret);
657 else
658 ret = 0;
659
660 if (handle)
661 ocfs2_commit_trans(handle);
662out_unlock:
663 unlock_page(page);
664 page_cache_release(page);
665out:
666 return ret;
667}
668
669static int ocfs2_zero_extend(struct inode *inode,
670 u64 zero_to_size)
671{
672 int ret = 0;
673 u64 start_off;
674 struct super_block *sb = inode->i_sb;
675
676 start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
677 while (start_off < zero_to_size) {
678 ret = ocfs2_write_zero_page(inode, start_off);
679 if (ret < 0) {
680 mlog_errno(ret);
681 goto out;
682 }
683
684 start_off += sb->s_blocksize;
685 }
686
687out:
688 return ret;
689}
690
691static int ocfs2_extend_file(struct inode *inode,
692 struct buffer_head *di_bh,
693 u64 new_i_size)
694{
695 int ret = 0;
696 u32 clusters_to_add;
697
698 /* setattr sometimes calls us like this. */
699 if (new_i_size == 0)
700 goto out;
701
702 if (i_size_read(inode) == new_i_size)
703 goto out;
704 BUG_ON(new_i_size < i_size_read(inode));
705
706 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size) -
707 OCFS2_I(inode)->ip_clusters;
708
709 if (clusters_to_add) {
710 ret = ocfs2_extend_allocation(inode, clusters_to_add);
711 if (ret < 0) {
712 mlog_errno(ret);
713 goto out;
714 }
715
716 ret = ocfs2_zero_extend(inode, new_i_size);
717 if (ret < 0) {
718 mlog_errno(ret);
719 goto out;
720 }
721 }
722
723 /* No allocation required, we just use this helper to
724 * do a trivial update of i_size. */
725 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
726 if (ret < 0) {
727 mlog_errno(ret);
728 goto out;
729 }
730
731out:
732 return ret;
733}
734
735int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
736{
737 int status = 0, size_change;
738 struct inode *inode = dentry->d_inode;
739 struct super_block *sb = inode->i_sb;
740 struct ocfs2_super *osb = OCFS2_SB(sb);
741 struct buffer_head *bh = NULL;
742 struct ocfs2_journal_handle *handle = NULL;
743
744 mlog_entry("(0x%p, '%.*s')\n", dentry,
745 dentry->d_name.len, dentry->d_name.name);
746
747 if (attr->ia_valid & ATTR_MODE)
748 mlog(0, "mode change: %d\n", attr->ia_mode);
749 if (attr->ia_valid & ATTR_UID)
750 mlog(0, "uid change: %d\n", attr->ia_uid);
751 if (attr->ia_valid & ATTR_GID)
752 mlog(0, "gid change: %d\n", attr->ia_gid);
753 if (attr->ia_valid & ATTR_SIZE)
754 mlog(0, "size change...\n");
755 if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME))
756 mlog(0, "time change...\n");
757
758#define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
759 | ATTR_GID | ATTR_UID | ATTR_MODE)
760 if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) {
761 mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid);
762 return 0;
763 }
764
765 status = inode_change_ok(inode, attr);
766 if (status)
767 return status;
768
769 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
770 if (size_change) {
771 status = ocfs2_rw_lock(inode, 1);
772 if (status < 0) {
773 mlog_errno(status);
774 goto bail;
775 }
776 }
777
778 status = ocfs2_meta_lock(inode, NULL, &bh, 1);
779 if (status < 0) {
780 if (status != -ENOENT)
781 mlog_errno(status);
782 goto bail_unlock_rw;
783 }
784
785 if (size_change && attr->ia_size != i_size_read(inode)) {
786 if (i_size_read(inode) > attr->ia_size)
787 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
788 else
789 status = ocfs2_extend_file(inode, bh, attr->ia_size);
790 if (status < 0) {
791 if (status != -ENOSPC)
792 mlog_errno(status);
793 status = -ENOSPC;
794 goto bail_unlock;
795 }
796 }
797
798 handle = ocfs2_start_trans(osb, NULL, OCFS2_INODE_UPDATE_CREDITS);
799 if (IS_ERR(handle)) {
800 status = PTR_ERR(handle);
801 mlog_errno(status);
802 goto bail_unlock;
803 }
804
805 status = inode_setattr(inode, attr);
806 if (status < 0) {
807 mlog_errno(status);
808 goto bail_commit;
809 }
810
811 status = ocfs2_mark_inode_dirty(handle, inode, bh);
812 if (status < 0)
813 mlog_errno(status);
814
815bail_commit:
816 ocfs2_commit_trans(handle);
817bail_unlock:
818 ocfs2_meta_unlock(inode, 1);
819bail_unlock_rw:
820 if (size_change)
821 ocfs2_rw_unlock(inode, 1);
822bail:
823 if (bh)
824 brelse(bh);
825
826 mlog_exit(status);
827 return status;
828}
829
830int ocfs2_getattr(struct vfsmount *mnt,
831 struct dentry *dentry,
832 struct kstat *stat)
833{
834 struct inode *inode = dentry->d_inode;
835 struct super_block *sb = dentry->d_inode->i_sb;
836 struct ocfs2_super *osb = sb->s_fs_info;
837 int err;
838
839 mlog_entry_void();
840
841 err = ocfs2_inode_revalidate(dentry);
842 if (err) {
843 if (err != -ENOENT)
844 mlog_errno(err);
845 goto bail;
846 }
847
848 generic_fillattr(inode, stat);
849
850 /* We set the blksize from the cluster size for performance */
851 stat->blksize = osb->s_clustersize;
852
853bail:
854 mlog_exit(err);
855
856 return err;
857}
858
859static int ocfs2_write_remove_suid(struct inode *inode)
860{
861 int ret;
862 struct buffer_head *bh = NULL;
863 struct ocfs2_inode_info *oi = OCFS2_I(inode);
864 struct ocfs2_journal_handle *handle;
865 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
866 struct ocfs2_dinode *di;
867
868 mlog_entry("(Inode %"MLFu64", mode 0%o)\n", oi->ip_blkno,
869 inode->i_mode);
870
871 handle = ocfs2_start_trans(osb, NULL, OCFS2_INODE_UPDATE_CREDITS);
872 if (handle == NULL) {
873 ret = -ENOMEM;
874 mlog_errno(ret);
875 goto out;
876 }
877
878 ret = ocfs2_read_block(osb, oi->ip_blkno, &bh, OCFS2_BH_CACHED, inode);
879 if (ret < 0) {
880 mlog_errno(ret);
881 goto out_trans;
882 }
883
884 ret = ocfs2_journal_access(handle, inode, bh,
885 OCFS2_JOURNAL_ACCESS_WRITE);
886 if (ret < 0) {
887 mlog_errno(ret);
888 goto out_bh;
889 }
890
891 inode->i_mode &= ~S_ISUID;
892 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
893 inode->i_mode &= ~S_ISGID;
894
895 di = (struct ocfs2_dinode *) bh->b_data;
896 di->i_mode = cpu_to_le16(inode->i_mode);
897
898 ret = ocfs2_journal_dirty(handle, bh);
899 if (ret < 0)
900 mlog_errno(ret);
901out_bh:
902 brelse(bh);
903out_trans:
904 ocfs2_commit_trans(handle);
905out:
906 mlog_exit(ret);
907 return ret;
908}
909
910static inline int ocfs2_write_should_remove_suid(struct inode *inode)
911{
912 mode_t mode = inode->i_mode;
913
914 if (!capable(CAP_FSETID)) {
915 if (unlikely(mode & S_ISUID))
916 return 1;
917
918 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
919 return 1;
920 }
921 return 0;
922}
923
924static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
925 const char __user *buf,
926 size_t count,
927 loff_t pos)
928{
929 struct iovec local_iov = { .iov_base = (void __user *)buf,
930 .iov_len = count };
931 int ret, rw_level = -1, meta_level = -1, have_alloc_sem = 0;
932 u32 clusters;
933 struct file *filp = iocb->ki_filp;
934 struct inode *inode = filp->f_dentry->d_inode;
935 loff_t newsize, saved_pos;
Mark Fashehccd979b2005-12-15 14:31:24 -0800936
937 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", filp, buf,
938 (unsigned int)count,
939 filp->f_dentry->d_name.len,
940 filp->f_dentry->d_name.name);
941
942 /* happy write of zero bytes */
943 if (count == 0)
944 return 0;
945
946 if (!inode) {
947 mlog(0, "bad inode\n");
948 return -EIO;
949 }
950
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800951 mutex_lock(&inode->i_mutex);
952 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
Mark Fashehccd979b2005-12-15 14:31:24 -0800953 if (filp->f_flags & O_DIRECT) {
954 have_alloc_sem = 1;
955 down_read(&inode->i_alloc_sem);
956 }
957
958 /* concurrent O_DIRECT writes are allowed */
959 rw_level = (filp->f_flags & O_DIRECT) ? 0 : 1;
960 ret = ocfs2_rw_lock(inode, rw_level);
961 if (ret < 0) {
962 rw_level = -1;
963 mlog_errno(ret);
964 goto out;
965 }
966
967 /*
968 * We sample i_size under a read level meta lock to see if our write
969 * is extending the file, if it is we back off and get a write level
970 * meta lock.
971 */
972 meta_level = (filp->f_flags & O_APPEND) ? 1 : 0;
973 for(;;) {
974 ret = ocfs2_meta_lock(inode, NULL, NULL, meta_level);
975 if (ret < 0) {
976 meta_level = -1;
977 mlog_errno(ret);
978 goto out;
979 }
980
981 /* Clear suid / sgid if necessary. We do this here
982 * instead of later in the write path because
983 * remove_suid() calls ->setattr without any hint that
984 * we may have already done our cluster locking. Since
985 * ocfs2_setattr() *must* take cluster locks to
986 * proceeed, this will lead us to recursively lock the
987 * inode. There's also the dinode i_size state which
988 * can be lost via setattr during extending writes (we
989 * set inode->i_size at the end of a write. */
990 if (ocfs2_write_should_remove_suid(inode)) {
991 if (meta_level == 0) {
992 ocfs2_meta_unlock(inode, meta_level);
993 meta_level = 1;
994 continue;
995 }
996
997 ret = ocfs2_write_remove_suid(inode);
998 if (ret < 0) {
999 mlog_errno(ret);
1000 goto out;
1001 }
1002 }
1003
1004 /* work on a copy of ppos until we're sure that we won't have
1005 * to recalculate it due to relocking. */
1006 if (filp->f_flags & O_APPEND) {
1007 saved_pos = i_size_read(inode);
1008 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos);
1009 } else {
1010 saved_pos = iocb->ki_pos;
1011 }
1012 newsize = count + saved_pos;
1013
Mark Fasheh215c7f92006-02-01 16:42:10 -08001014 mlog(0, "pos=%lld newsize=%lld cursize=%lld\n",
1015 (long long) saved_pos, (long long) newsize,
1016 (long long) i_size_read(inode));
Mark Fashehccd979b2005-12-15 14:31:24 -08001017
1018 /* No need for a higher level metadata lock if we're
1019 * never going past i_size. */
1020 if (newsize <= i_size_read(inode))
1021 break;
1022
1023 if (meta_level == 0) {
1024 ocfs2_meta_unlock(inode, meta_level);
1025 meta_level = 1;
1026 continue;
1027 }
1028
1029 spin_lock(&OCFS2_I(inode)->ip_lock);
1030 clusters = ocfs2_clusters_for_bytes(inode->i_sb, newsize) -
1031 OCFS2_I(inode)->ip_clusters;
1032 spin_unlock(&OCFS2_I(inode)->ip_lock);
1033
1034 mlog(0, "Writing at EOF, may need more allocation: "
Mark Fasheh215c7f92006-02-01 16:42:10 -08001035 "i_size = %lld, newsize = %lld, need %u clusters\n",
1036 (long long) i_size_read(inode), (long long) newsize,
1037 clusters);
Mark Fashehccd979b2005-12-15 14:31:24 -08001038
1039 /* We only want to continue the rest of this loop if
1040 * our extend will actually require more
1041 * allocation. */
1042 if (!clusters)
1043 break;
1044
1045 ret = ocfs2_extend_allocation(inode, clusters);
1046 if (ret < 0) {
1047 if (ret != -ENOSPC)
1048 mlog_errno(ret);
1049 goto out;
1050 }
1051
1052 /* Fill any holes which would've been created by this
1053 * write. If we're O_APPEND, this will wind up
1054 * (correctly) being a noop. */
1055 ret = ocfs2_zero_extend(inode, (u64) newsize - count);
1056 if (ret < 0) {
1057 mlog_errno(ret);
1058 goto out;
1059 }
1060 break;
1061 }
1062
1063 /* ok, we're done with i_size and alloc work */
1064 iocb->ki_pos = saved_pos;
1065 ocfs2_meta_unlock(inode, meta_level);
1066 meta_level = -1;
1067
1068 /* communicate with ocfs2_dio_end_io */
1069 ocfs2_iocb_set_rw_locked(iocb);
1070
Mark Fashehd267a562006-02-23 13:23:39 -08001071 ret = generic_file_aio_write_nolock(iocb, &local_iov, 1, &iocb->ki_pos);
Mark Fashehccd979b2005-12-15 14:31:24 -08001072
1073 /* buffered aio wouldn't have proper lock coverage today */
1074 BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
1075
1076 /*
1077 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
1078 * function pointer which is called when o_direct io completes so that
1079 * it can unlock our rw lock. (it's the clustered equivalent of
1080 * i_alloc_sem; protects truncate from racing with pending ios).
1081 * Unfortunately there are error cases which call end_io and others
1082 * that don't. so we don't have to unlock the rw_lock if either an
1083 * async dio is going to do it in the future or an end_io after an
1084 * error has already done it.
1085 */
1086 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
1087 rw_level = -1;
1088 have_alloc_sem = 0;
1089 }
1090
1091out:
1092 if (meta_level != -1)
1093 ocfs2_meta_unlock(inode, meta_level);
1094 if (have_alloc_sem)
1095 up_read(&inode->i_alloc_sem);
1096 if (rw_level != -1)
1097 ocfs2_rw_unlock(inode, rw_level);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001098 mutex_unlock(&inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08001099
1100 mlog_exit(ret);
1101 return ret;
1102}
1103
1104static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
1105 char __user *buf,
1106 size_t count,
1107 loff_t pos)
1108{
1109 int ret = 0, rw_level = -1, have_alloc_sem = 0;
1110 struct file *filp = iocb->ki_filp;
1111 struct inode *inode = filp->f_dentry->d_inode;
Mark Fashehccd979b2005-12-15 14:31:24 -08001112
1113 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", filp, buf,
1114 (unsigned int)count,
1115 filp->f_dentry->d_name.len,
1116 filp->f_dentry->d_name.name);
1117
1118 if (!inode) {
1119 ret = -EINVAL;
1120 mlog_errno(ret);
1121 goto bail;
1122 }
1123
Mark Fashehccd979b2005-12-15 14:31:24 -08001124 /*
1125 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
1126 * need locks to protect pending reads from racing with truncate.
1127 */
1128 if (filp->f_flags & O_DIRECT) {
1129 down_read(&inode->i_alloc_sem);
1130 have_alloc_sem = 1;
1131
1132 ret = ocfs2_rw_lock(inode, 0);
1133 if (ret < 0) {
1134 mlog_errno(ret);
1135 goto bail;
1136 }
1137 rw_level = 0;
1138 /* communicate with ocfs2_dio_end_io */
1139 ocfs2_iocb_set_rw_locked(iocb);
1140 }
1141
1142 ret = generic_file_aio_read(iocb, buf, count, iocb->ki_pos);
1143 if (ret == -EINVAL)
1144 mlog(ML_ERROR, "generic_file_aio_read returned -EINVAL\n");
1145
1146 /* buffered aio wouldn't have proper lock coverage today */
1147 BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
1148
1149 /* see ocfs2_file_aio_write */
1150 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
1151 rw_level = -1;
1152 have_alloc_sem = 0;
1153 }
1154
1155bail:
1156 if (have_alloc_sem)
1157 up_read(&inode->i_alloc_sem);
1158 if (rw_level != -1)
1159 ocfs2_rw_unlock(inode, rw_level);
1160 mlog_exit(ret);
1161
1162 return ret;
1163}
1164
1165struct inode_operations ocfs2_file_iops = {
1166 .setattr = ocfs2_setattr,
1167 .getattr = ocfs2_getattr,
1168};
1169
1170struct inode_operations ocfs2_special_file_iops = {
1171 .setattr = ocfs2_setattr,
1172 .getattr = ocfs2_getattr,
1173};
1174
1175struct file_operations ocfs2_fops = {
1176 .read = do_sync_read,
1177 .write = do_sync_write,
1178 .sendfile = generic_file_sendfile,
1179 .mmap = ocfs2_mmap,
1180 .fsync = ocfs2_sync_file,
1181 .release = ocfs2_file_release,
1182 .open = ocfs2_file_open,
1183 .aio_read = ocfs2_file_aio_read,
1184 .aio_write = ocfs2_file_aio_write,
1185};
1186
1187struct file_operations ocfs2_dops = {
1188 .read = generic_read_dir,
1189 .readdir = ocfs2_readdir,
1190 .fsync = ocfs2_sync_file,
1191};