Loading...
1/* -*- 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
26#include <linux/capability.h>
27#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#include <linux/sched.h>
34#include <linux/splice.h>
35#include <linux/mount.h>
36#include <linux/writeback.h>
37#include <linux/falloc.h>
38#include <linux/quotaops.h>
39#include <linux/blkdev.h>
40
41#include <cluster/masklog.h>
42
43#include "ocfs2.h"
44
45#include "alloc.h"
46#include "aops.h"
47#include "dir.h"
48#include "dlmglue.h"
49#include "extent_map.h"
50#include "file.h"
51#include "sysfile.h"
52#include "inode.h"
53#include "ioctl.h"
54#include "journal.h"
55#include "locks.h"
56#include "mmap.h"
57#include "suballoc.h"
58#include "super.h"
59#include "xattr.h"
60#include "acl.h"
61#include "quota.h"
62#include "refcounttree.h"
63#include "ocfs2_trace.h"
64
65#include "buffer_head_io.h"
66
67static int ocfs2_init_file_private(struct inode *inode, struct file *file)
68{
69 struct ocfs2_file_private *fp;
70
71 fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
72 if (!fp)
73 return -ENOMEM;
74
75 fp->fp_file = file;
76 mutex_init(&fp->fp_mutex);
77 ocfs2_file_lock_res_init(&fp->fp_flock, fp);
78 file->private_data = fp;
79
80 return 0;
81}
82
83static void ocfs2_free_file_private(struct inode *inode, struct file *file)
84{
85 struct ocfs2_file_private *fp = file->private_data;
86 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
87
88 if (fp) {
89 ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
90 ocfs2_lock_res_free(&fp->fp_flock);
91 kfree(fp);
92 file->private_data = NULL;
93 }
94}
95
96static int ocfs2_file_open(struct inode *inode, struct file *file)
97{
98 int status;
99 int mode = file->f_flags;
100 struct ocfs2_inode_info *oi = OCFS2_I(inode);
101
102 trace_ocfs2_file_open(inode, file, file->f_path.dentry,
103 (unsigned long long)OCFS2_I(inode)->ip_blkno,
104 file->f_path.dentry->d_name.len,
105 file->f_path.dentry->d_name.name, mode);
106
107 if (file->f_mode & FMODE_WRITE)
108 dquot_initialize(inode);
109
110 spin_lock(&oi->ip_lock);
111
112 /* Check that the inode hasn't been wiped from disk by another
113 * node. If it hasn't then we're safe as long as we hold the
114 * spin lock until our increment of open count. */
115 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
116 spin_unlock(&oi->ip_lock);
117
118 status = -ENOENT;
119 goto leave;
120 }
121
122 if (mode & O_DIRECT)
123 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
124
125 oi->ip_open_count++;
126 spin_unlock(&oi->ip_lock);
127
128 status = ocfs2_init_file_private(inode, file);
129 if (status) {
130 /*
131 * We want to set open count back if we're failing the
132 * open.
133 */
134 spin_lock(&oi->ip_lock);
135 oi->ip_open_count--;
136 spin_unlock(&oi->ip_lock);
137 }
138
139leave:
140 return status;
141}
142
143static int ocfs2_file_release(struct inode *inode, struct file *file)
144{
145 struct ocfs2_inode_info *oi = OCFS2_I(inode);
146
147 spin_lock(&oi->ip_lock);
148 if (!--oi->ip_open_count)
149 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
150
151 trace_ocfs2_file_release(inode, file, file->f_path.dentry,
152 oi->ip_blkno,
153 file->f_path.dentry->d_name.len,
154 file->f_path.dentry->d_name.name,
155 oi->ip_open_count);
156 spin_unlock(&oi->ip_lock);
157
158 ocfs2_free_file_private(inode, file);
159
160 return 0;
161}
162
163static int ocfs2_dir_open(struct inode *inode, struct file *file)
164{
165 return ocfs2_init_file_private(inode, file);
166}
167
168static int ocfs2_dir_release(struct inode *inode, struct file *file)
169{
170 ocfs2_free_file_private(inode, file);
171 return 0;
172}
173
174static int ocfs2_sync_file(struct file *file, loff_t start, loff_t end,
175 int datasync)
176{
177 int err = 0;
178 journal_t *journal;
179 struct inode *inode = file->f_mapping->host;
180 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
181
182 trace_ocfs2_sync_file(inode, file, file->f_path.dentry,
183 OCFS2_I(inode)->ip_blkno,
184 file->f_path.dentry->d_name.len,
185 file->f_path.dentry->d_name.name,
186 (unsigned long long)datasync);
187
188 err = filemap_write_and_wait_range(inode->i_mapping, start, end);
189 if (err)
190 return err;
191
192 /*
193 * Probably don't need the i_mutex at all in here, just putting it here
194 * to be consistent with how fsync used to be called, someone more
195 * familiar with the fs could possibly remove it.
196 */
197 mutex_lock(&inode->i_mutex);
198 if (datasync && !(inode->i_state & I_DIRTY_DATASYNC)) {
199 /*
200 * We still have to flush drive's caches to get data to the
201 * platter
202 */
203 if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
204 blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
205 goto bail;
206 }
207
208 journal = osb->journal->j_journal;
209 err = jbd2_journal_force_commit(journal);
210
211bail:
212 if (err)
213 mlog_errno(err);
214 mutex_unlock(&inode->i_mutex);
215
216 return (err < 0) ? -EIO : 0;
217}
218
219int ocfs2_should_update_atime(struct inode *inode,
220 struct vfsmount *vfsmnt)
221{
222 struct timespec now;
223 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
224
225 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
226 return 0;
227
228 if ((inode->i_flags & S_NOATIME) ||
229 ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
230 return 0;
231
232 /*
233 * We can be called with no vfsmnt structure - NFSD will
234 * sometimes do this.
235 *
236 * Note that our action here is different than touch_atime() -
237 * if we can't tell whether this is a noatime mount, then we
238 * don't know whether to trust the value of s_atime_quantum.
239 */
240 if (vfsmnt == NULL)
241 return 0;
242
243 if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
244 ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
245 return 0;
246
247 if (vfsmnt->mnt_flags & MNT_RELATIME) {
248 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
249 (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
250 return 1;
251
252 return 0;
253 }
254
255 now = CURRENT_TIME;
256 if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
257 return 0;
258 else
259 return 1;
260}
261
262int ocfs2_update_inode_atime(struct inode *inode,
263 struct buffer_head *bh)
264{
265 int ret;
266 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
267 handle_t *handle;
268 struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
269
270 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
271 if (IS_ERR(handle)) {
272 ret = PTR_ERR(handle);
273 mlog_errno(ret);
274 goto out;
275 }
276
277 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
278 OCFS2_JOURNAL_ACCESS_WRITE);
279 if (ret) {
280 mlog_errno(ret);
281 goto out_commit;
282 }
283
284 /*
285 * Don't use ocfs2_mark_inode_dirty() here as we don't always
286 * have i_mutex to guard against concurrent changes to other
287 * inode fields.
288 */
289 inode->i_atime = CURRENT_TIME;
290 di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
291 di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
292 ocfs2_journal_dirty(handle, bh);
293
294out_commit:
295 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
296out:
297 return ret;
298}
299
300static int ocfs2_set_inode_size(handle_t *handle,
301 struct inode *inode,
302 struct buffer_head *fe_bh,
303 u64 new_i_size)
304{
305 int status;
306
307 i_size_write(inode, new_i_size);
308 inode->i_blocks = ocfs2_inode_sector_count(inode);
309 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
310
311 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
312 if (status < 0) {
313 mlog_errno(status);
314 goto bail;
315 }
316
317bail:
318 return status;
319}
320
321int ocfs2_simple_size_update(struct inode *inode,
322 struct buffer_head *di_bh,
323 u64 new_i_size)
324{
325 int ret;
326 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
327 handle_t *handle = NULL;
328
329 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
330 if (IS_ERR(handle)) {
331 ret = PTR_ERR(handle);
332 mlog_errno(ret);
333 goto out;
334 }
335
336 ret = ocfs2_set_inode_size(handle, inode, di_bh,
337 new_i_size);
338 if (ret < 0)
339 mlog_errno(ret);
340
341 ocfs2_commit_trans(osb, handle);
342out:
343 return ret;
344}
345
346static int ocfs2_cow_file_pos(struct inode *inode,
347 struct buffer_head *fe_bh,
348 u64 offset)
349{
350 int status;
351 u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
352 unsigned int num_clusters = 0;
353 unsigned int ext_flags = 0;
354
355 /*
356 * If the new offset is aligned to the range of the cluster, there is
357 * no space for ocfs2_zero_range_for_truncate to fill, so no need to
358 * CoW either.
359 */
360 if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
361 return 0;
362
363 status = ocfs2_get_clusters(inode, cpos, &phys,
364 &num_clusters, &ext_flags);
365 if (status) {
366 mlog_errno(status);
367 goto out;
368 }
369
370 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
371 goto out;
372
373 return ocfs2_refcount_cow(inode, NULL, fe_bh, cpos, 1, cpos+1);
374
375out:
376 return status;
377}
378
379static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
380 struct inode *inode,
381 struct buffer_head *fe_bh,
382 u64 new_i_size)
383{
384 int status;
385 handle_t *handle;
386 struct ocfs2_dinode *di;
387 u64 cluster_bytes;
388
389 /*
390 * We need to CoW the cluster contains the offset if it is reflinked
391 * since we will call ocfs2_zero_range_for_truncate later which will
392 * write "0" from offset to the end of the cluster.
393 */
394 status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
395 if (status) {
396 mlog_errno(status);
397 return status;
398 }
399
400 /* TODO: This needs to actually orphan the inode in this
401 * transaction. */
402
403 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
404 if (IS_ERR(handle)) {
405 status = PTR_ERR(handle);
406 mlog_errno(status);
407 goto out;
408 }
409
410 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
411 OCFS2_JOURNAL_ACCESS_WRITE);
412 if (status < 0) {
413 mlog_errno(status);
414 goto out_commit;
415 }
416
417 /*
418 * Do this before setting i_size.
419 */
420 cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
421 status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
422 cluster_bytes);
423 if (status) {
424 mlog_errno(status);
425 goto out_commit;
426 }
427
428 i_size_write(inode, new_i_size);
429 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
430
431 di = (struct ocfs2_dinode *) fe_bh->b_data;
432 di->i_size = cpu_to_le64(new_i_size);
433 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
434 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
435
436 ocfs2_journal_dirty(handle, fe_bh);
437
438out_commit:
439 ocfs2_commit_trans(osb, handle);
440out:
441 return status;
442}
443
444static int ocfs2_truncate_file(struct inode *inode,
445 struct buffer_head *di_bh,
446 u64 new_i_size)
447{
448 int status = 0;
449 struct ocfs2_dinode *fe = NULL;
450 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
451
452 /* We trust di_bh because it comes from ocfs2_inode_lock(), which
453 * already validated it */
454 fe = (struct ocfs2_dinode *) di_bh->b_data;
455
456 trace_ocfs2_truncate_file((unsigned long long)OCFS2_I(inode)->ip_blkno,
457 (unsigned long long)le64_to_cpu(fe->i_size),
458 (unsigned long long)new_i_size);
459
460 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
461 "Inode %llu, inode i_size = %lld != di "
462 "i_size = %llu, i_flags = 0x%x\n",
463 (unsigned long long)OCFS2_I(inode)->ip_blkno,
464 i_size_read(inode),
465 (unsigned long long)le64_to_cpu(fe->i_size),
466 le32_to_cpu(fe->i_flags));
467
468 if (new_i_size > le64_to_cpu(fe->i_size)) {
469 trace_ocfs2_truncate_file_error(
470 (unsigned long long)le64_to_cpu(fe->i_size),
471 (unsigned long long)new_i_size);
472 status = -EINVAL;
473 mlog_errno(status);
474 goto bail;
475 }
476
477 /* lets handle the simple truncate cases before doing any more
478 * cluster locking. */
479 if (new_i_size == le64_to_cpu(fe->i_size))
480 goto bail;
481
482 down_write(&OCFS2_I(inode)->ip_alloc_sem);
483
484 ocfs2_resv_discard(&osb->osb_la_resmap,
485 &OCFS2_I(inode)->ip_la_data_resv);
486
487 /*
488 * The inode lock forced other nodes to sync and drop their
489 * pages, which (correctly) happens even if we have a truncate
490 * without allocation change - ocfs2 cluster sizes can be much
491 * greater than page size, so we have to truncate them
492 * anyway.
493 */
494 unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
495 truncate_inode_pages(inode->i_mapping, new_i_size);
496
497 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
498 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
499 i_size_read(inode), 1);
500 if (status)
501 mlog_errno(status);
502
503 goto bail_unlock_sem;
504 }
505
506 /* alright, we're going to need to do a full blown alloc size
507 * change. Orphan the inode so that recovery can complete the
508 * truncate if necessary. This does the task of marking
509 * i_size. */
510 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
511 if (status < 0) {
512 mlog_errno(status);
513 goto bail_unlock_sem;
514 }
515
516 status = ocfs2_commit_truncate(osb, inode, di_bh);
517 if (status < 0) {
518 mlog_errno(status);
519 goto bail_unlock_sem;
520 }
521
522 /* TODO: orphan dir cleanup here. */
523bail_unlock_sem:
524 up_write(&OCFS2_I(inode)->ip_alloc_sem);
525
526bail:
527 if (!status && OCFS2_I(inode)->ip_clusters == 0)
528 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
529
530 return status;
531}
532
533/*
534 * extend file allocation only here.
535 * we'll update all the disk stuff, and oip->alloc_size
536 *
537 * expect stuff to be locked, a transaction started and enough data /
538 * metadata reservations in the contexts.
539 *
540 * Will return -EAGAIN, and a reason if a restart is needed.
541 * If passed in, *reason will always be set, even in error.
542 */
543int ocfs2_add_inode_data(struct ocfs2_super *osb,
544 struct inode *inode,
545 u32 *logical_offset,
546 u32 clusters_to_add,
547 int mark_unwritten,
548 struct buffer_head *fe_bh,
549 handle_t *handle,
550 struct ocfs2_alloc_context *data_ac,
551 struct ocfs2_alloc_context *meta_ac,
552 enum ocfs2_alloc_restarted *reason_ret)
553{
554 int ret;
555 struct ocfs2_extent_tree et;
556
557 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
558 ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
559 clusters_to_add, mark_unwritten,
560 data_ac, meta_ac, reason_ret);
561
562 return ret;
563}
564
565static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
566 u32 clusters_to_add, int mark_unwritten)
567{
568 int status = 0;
569 int restart_func = 0;
570 int credits;
571 u32 prev_clusters;
572 struct buffer_head *bh = NULL;
573 struct ocfs2_dinode *fe = NULL;
574 handle_t *handle = NULL;
575 struct ocfs2_alloc_context *data_ac = NULL;
576 struct ocfs2_alloc_context *meta_ac = NULL;
577 enum ocfs2_alloc_restarted why;
578 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
579 struct ocfs2_extent_tree et;
580 int did_quota = 0;
581
582 /*
583 * This function only exists for file systems which don't
584 * support holes.
585 */
586 BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
587
588 status = ocfs2_read_inode_block(inode, &bh);
589 if (status < 0) {
590 mlog_errno(status);
591 goto leave;
592 }
593 fe = (struct ocfs2_dinode *) bh->b_data;
594
595restart_all:
596 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
597
598 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
599 status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
600 &data_ac, &meta_ac);
601 if (status) {
602 mlog_errno(status);
603 goto leave;
604 }
605
606 credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list,
607 clusters_to_add);
608 handle = ocfs2_start_trans(osb, credits);
609 if (IS_ERR(handle)) {
610 status = PTR_ERR(handle);
611 handle = NULL;
612 mlog_errno(status);
613 goto leave;
614 }
615
616restarted_transaction:
617 trace_ocfs2_extend_allocation(
618 (unsigned long long)OCFS2_I(inode)->ip_blkno,
619 (unsigned long long)i_size_read(inode),
620 le32_to_cpu(fe->i_clusters), clusters_to_add,
621 why, restart_func);
622
623 status = dquot_alloc_space_nodirty(inode,
624 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
625 if (status)
626 goto leave;
627 did_quota = 1;
628
629 /* reserve a write to the file entry early on - that we if we
630 * run out of credits in the allocation path, we can still
631 * update i_size. */
632 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
633 OCFS2_JOURNAL_ACCESS_WRITE);
634 if (status < 0) {
635 mlog_errno(status);
636 goto leave;
637 }
638
639 prev_clusters = OCFS2_I(inode)->ip_clusters;
640
641 status = ocfs2_add_inode_data(osb,
642 inode,
643 &logical_start,
644 clusters_to_add,
645 mark_unwritten,
646 bh,
647 handle,
648 data_ac,
649 meta_ac,
650 &why);
651 if ((status < 0) && (status != -EAGAIN)) {
652 if (status != -ENOSPC)
653 mlog_errno(status);
654 goto leave;
655 }
656
657 ocfs2_journal_dirty(handle, bh);
658
659 spin_lock(&OCFS2_I(inode)->ip_lock);
660 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
661 spin_unlock(&OCFS2_I(inode)->ip_lock);
662 /* Release unused quota reservation */
663 dquot_free_space(inode,
664 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
665 did_quota = 0;
666
667 if (why != RESTART_NONE && clusters_to_add) {
668 if (why == RESTART_META) {
669 restart_func = 1;
670 status = 0;
671 } else {
672 BUG_ON(why != RESTART_TRANS);
673
674 /* TODO: This can be more intelligent. */
675 credits = ocfs2_calc_extend_credits(osb->sb,
676 &fe->id2.i_list,
677 clusters_to_add);
678 status = ocfs2_extend_trans(handle, credits);
679 if (status < 0) {
680 /* handle still has to be committed at
681 * this point. */
682 status = -ENOMEM;
683 mlog_errno(status);
684 goto leave;
685 }
686 goto restarted_transaction;
687 }
688 }
689
690 trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno,
691 le32_to_cpu(fe->i_clusters),
692 (unsigned long long)le64_to_cpu(fe->i_size),
693 OCFS2_I(inode)->ip_clusters,
694 (unsigned long long)i_size_read(inode));
695
696leave:
697 if (status < 0 && did_quota)
698 dquot_free_space(inode,
699 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
700 if (handle) {
701 ocfs2_commit_trans(osb, handle);
702 handle = NULL;
703 }
704 if (data_ac) {
705 ocfs2_free_alloc_context(data_ac);
706 data_ac = NULL;
707 }
708 if (meta_ac) {
709 ocfs2_free_alloc_context(meta_ac);
710 meta_ac = NULL;
711 }
712 if ((!status) && restart_func) {
713 restart_func = 0;
714 goto restart_all;
715 }
716 brelse(bh);
717 bh = NULL;
718
719 return status;
720}
721
722/*
723 * While a write will already be ordering the data, a truncate will not.
724 * Thus, we need to explicitly order the zeroed pages.
725 */
726static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode)
727{
728 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
729 handle_t *handle = NULL;
730 int ret = 0;
731
732 if (!ocfs2_should_order_data(inode))
733 goto out;
734
735 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
736 if (IS_ERR(handle)) {
737 ret = -ENOMEM;
738 mlog_errno(ret);
739 goto out;
740 }
741
742 ret = ocfs2_jbd2_file_inode(handle, inode);
743 if (ret < 0)
744 mlog_errno(ret);
745
746out:
747 if (ret) {
748 if (!IS_ERR(handle))
749 ocfs2_commit_trans(osb, handle);
750 handle = ERR_PTR(ret);
751 }
752 return handle;
753}
754
755/* Some parts of this taken from generic_cont_expand, which turned out
756 * to be too fragile to do exactly what we need without us having to
757 * worry about recursive locking in ->write_begin() and ->write_end(). */
758static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
759 u64 abs_to)
760{
761 struct address_space *mapping = inode->i_mapping;
762 struct page *page;
763 unsigned long index = abs_from >> PAGE_CACHE_SHIFT;
764 handle_t *handle = NULL;
765 int ret = 0;
766 unsigned zero_from, zero_to, block_start, block_end;
767
768 BUG_ON(abs_from >= abs_to);
769 BUG_ON(abs_to > (((u64)index + 1) << PAGE_CACHE_SHIFT));
770 BUG_ON(abs_from & (inode->i_blkbits - 1));
771
772 page = find_or_create_page(mapping, index, GFP_NOFS);
773 if (!page) {
774 ret = -ENOMEM;
775 mlog_errno(ret);
776 goto out;
777 }
778
779 /* Get the offsets within the page that we want to zero */
780 zero_from = abs_from & (PAGE_CACHE_SIZE - 1);
781 zero_to = abs_to & (PAGE_CACHE_SIZE - 1);
782 if (!zero_to)
783 zero_to = PAGE_CACHE_SIZE;
784
785 trace_ocfs2_write_zero_page(
786 (unsigned long long)OCFS2_I(inode)->ip_blkno,
787 (unsigned long long)abs_from,
788 (unsigned long long)abs_to,
789 index, zero_from, zero_to);
790
791 /* We know that zero_from is block aligned */
792 for (block_start = zero_from; block_start < zero_to;
793 block_start = block_end) {
794 block_end = block_start + (1 << inode->i_blkbits);
795
796 /*
797 * block_start is block-aligned. Bump it by one to force
798 * __block_write_begin and block_commit_write to zero the
799 * whole block.
800 */
801 ret = __block_write_begin(page, block_start + 1, 0,
802 ocfs2_get_block);
803 if (ret < 0) {
804 mlog_errno(ret);
805 goto out_unlock;
806 }
807
808 if (!handle) {
809 handle = ocfs2_zero_start_ordered_transaction(inode);
810 if (IS_ERR(handle)) {
811 ret = PTR_ERR(handle);
812 handle = NULL;
813 break;
814 }
815 }
816
817 /* must not update i_size! */
818 ret = block_commit_write(page, block_start + 1,
819 block_start + 1);
820 if (ret < 0)
821 mlog_errno(ret);
822 else
823 ret = 0;
824 }
825
826 if (handle)
827 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
828
829out_unlock:
830 unlock_page(page);
831 page_cache_release(page);
832out:
833 return ret;
834}
835
836/*
837 * Find the next range to zero. We do this in terms of bytes because
838 * that's what ocfs2_zero_extend() wants, and it is dealing with the
839 * pagecache. We may return multiple extents.
840 *
841 * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
842 * needs to be zeroed. range_start and range_end return the next zeroing
843 * range. A subsequent call should pass the previous range_end as its
844 * zero_start. If range_end is 0, there's nothing to do.
845 *
846 * Unwritten extents are skipped over. Refcounted extents are CoWd.
847 */
848static int ocfs2_zero_extend_get_range(struct inode *inode,
849 struct buffer_head *di_bh,
850 u64 zero_start, u64 zero_end,
851 u64 *range_start, u64 *range_end)
852{
853 int rc = 0, needs_cow = 0;
854 u32 p_cpos, zero_clusters = 0;
855 u32 zero_cpos =
856 zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
857 u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
858 unsigned int num_clusters = 0;
859 unsigned int ext_flags = 0;
860
861 while (zero_cpos < last_cpos) {
862 rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
863 &num_clusters, &ext_flags);
864 if (rc) {
865 mlog_errno(rc);
866 goto out;
867 }
868
869 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
870 zero_clusters = num_clusters;
871 if (ext_flags & OCFS2_EXT_REFCOUNTED)
872 needs_cow = 1;
873 break;
874 }
875
876 zero_cpos += num_clusters;
877 }
878 if (!zero_clusters) {
879 *range_end = 0;
880 goto out;
881 }
882
883 while ((zero_cpos + zero_clusters) < last_cpos) {
884 rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
885 &p_cpos, &num_clusters,
886 &ext_flags);
887 if (rc) {
888 mlog_errno(rc);
889 goto out;
890 }
891
892 if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
893 break;
894 if (ext_flags & OCFS2_EXT_REFCOUNTED)
895 needs_cow = 1;
896 zero_clusters += num_clusters;
897 }
898 if ((zero_cpos + zero_clusters) > last_cpos)
899 zero_clusters = last_cpos - zero_cpos;
900
901 if (needs_cow) {
902 rc = ocfs2_refcount_cow(inode, NULL, di_bh, zero_cpos,
903 zero_clusters, UINT_MAX);
904 if (rc) {
905 mlog_errno(rc);
906 goto out;
907 }
908 }
909
910 *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
911 *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
912 zero_cpos + zero_clusters);
913
914out:
915 return rc;
916}
917
918/*
919 * Zero one range returned from ocfs2_zero_extend_get_range(). The caller
920 * has made sure that the entire range needs zeroing.
921 */
922static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
923 u64 range_end)
924{
925 int rc = 0;
926 u64 next_pos;
927 u64 zero_pos = range_start;
928
929 trace_ocfs2_zero_extend_range(
930 (unsigned long long)OCFS2_I(inode)->ip_blkno,
931 (unsigned long long)range_start,
932 (unsigned long long)range_end);
933 BUG_ON(range_start >= range_end);
934
935 while (zero_pos < range_end) {
936 next_pos = (zero_pos & PAGE_CACHE_MASK) + PAGE_CACHE_SIZE;
937 if (next_pos > range_end)
938 next_pos = range_end;
939 rc = ocfs2_write_zero_page(inode, zero_pos, next_pos);
940 if (rc < 0) {
941 mlog_errno(rc);
942 break;
943 }
944 zero_pos = next_pos;
945
946 /*
947 * Very large extends have the potential to lock up
948 * the cpu for extended periods of time.
949 */
950 cond_resched();
951 }
952
953 return rc;
954}
955
956int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
957 loff_t zero_to_size)
958{
959 int ret = 0;
960 u64 zero_start, range_start = 0, range_end = 0;
961 struct super_block *sb = inode->i_sb;
962
963 zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
964 trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
965 (unsigned long long)zero_start,
966 (unsigned long long)i_size_read(inode));
967 while (zero_start < zero_to_size) {
968 ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
969 zero_to_size,
970 &range_start,
971 &range_end);
972 if (ret) {
973 mlog_errno(ret);
974 break;
975 }
976 if (!range_end)
977 break;
978 /* Trim the ends */
979 if (range_start < zero_start)
980 range_start = zero_start;
981 if (range_end > zero_to_size)
982 range_end = zero_to_size;
983
984 ret = ocfs2_zero_extend_range(inode, range_start,
985 range_end);
986 if (ret) {
987 mlog_errno(ret);
988 break;
989 }
990 zero_start = range_end;
991 }
992
993 return ret;
994}
995
996int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
997 u64 new_i_size, u64 zero_to)
998{
999 int ret;
1000 u32 clusters_to_add;
1001 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1002
1003 /*
1004 * Only quota files call this without a bh, and they can't be
1005 * refcounted.
1006 */
1007 BUG_ON(!di_bh && (oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
1008 BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1009
1010 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1011 if (clusters_to_add < oi->ip_clusters)
1012 clusters_to_add = 0;
1013 else
1014 clusters_to_add -= oi->ip_clusters;
1015
1016 if (clusters_to_add) {
1017 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
1018 clusters_to_add, 0);
1019 if (ret) {
1020 mlog_errno(ret);
1021 goto out;
1022 }
1023 }
1024
1025 /*
1026 * Call this even if we don't add any clusters to the tree. We
1027 * still need to zero the area between the old i_size and the
1028 * new i_size.
1029 */
1030 ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1031 if (ret < 0)
1032 mlog_errno(ret);
1033
1034out:
1035 return ret;
1036}
1037
1038static int ocfs2_extend_file(struct inode *inode,
1039 struct buffer_head *di_bh,
1040 u64 new_i_size)
1041{
1042 int ret = 0;
1043 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1044
1045 BUG_ON(!di_bh);
1046
1047 /* setattr sometimes calls us like this. */
1048 if (new_i_size == 0)
1049 goto out;
1050
1051 if (i_size_read(inode) == new_i_size)
1052 goto out;
1053 BUG_ON(new_i_size < i_size_read(inode));
1054
1055 /*
1056 * The alloc sem blocks people in read/write from reading our
1057 * allocation until we're done changing it. We depend on
1058 * i_mutex to block other extend/truncate calls while we're
1059 * here. We even have to hold it for sparse files because there
1060 * might be some tail zeroing.
1061 */
1062 down_write(&oi->ip_alloc_sem);
1063
1064 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1065 /*
1066 * We can optimize small extends by keeping the inodes
1067 * inline data.
1068 */
1069 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1070 up_write(&oi->ip_alloc_sem);
1071 goto out_update_size;
1072 }
1073
1074 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1075 if (ret) {
1076 up_write(&oi->ip_alloc_sem);
1077 mlog_errno(ret);
1078 goto out;
1079 }
1080 }
1081
1082 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1083 ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1084 else
1085 ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1086 new_i_size);
1087
1088 up_write(&oi->ip_alloc_sem);
1089
1090 if (ret < 0) {
1091 mlog_errno(ret);
1092 goto out;
1093 }
1094
1095out_update_size:
1096 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1097 if (ret < 0)
1098 mlog_errno(ret);
1099
1100out:
1101 return ret;
1102}
1103
1104int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1105{
1106 int status = 0, size_change;
1107 struct inode *inode = dentry->d_inode;
1108 struct super_block *sb = inode->i_sb;
1109 struct ocfs2_super *osb = OCFS2_SB(sb);
1110 struct buffer_head *bh = NULL;
1111 handle_t *handle = NULL;
1112 struct dquot *transfer_to[MAXQUOTAS] = { };
1113 int qtype;
1114
1115 trace_ocfs2_setattr(inode, dentry,
1116 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1117 dentry->d_name.len, dentry->d_name.name,
1118 attr->ia_valid, attr->ia_mode,
1119 attr->ia_uid, attr->ia_gid);
1120
1121 /* ensuring we don't even attempt to truncate a symlink */
1122 if (S_ISLNK(inode->i_mode))
1123 attr->ia_valid &= ~ATTR_SIZE;
1124
1125#define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1126 | ATTR_GID | ATTR_UID | ATTR_MODE)
1127 if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
1128 return 0;
1129
1130 status = inode_change_ok(inode, attr);
1131 if (status)
1132 return status;
1133
1134 if (is_quota_modification(inode, attr))
1135 dquot_initialize(inode);
1136 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1137 if (size_change) {
1138 status = ocfs2_rw_lock(inode, 1);
1139 if (status < 0) {
1140 mlog_errno(status);
1141 goto bail;
1142 }
1143 }
1144
1145 status = ocfs2_inode_lock(inode, &bh, 1);
1146 if (status < 0) {
1147 if (status != -ENOENT)
1148 mlog_errno(status);
1149 goto bail_unlock_rw;
1150 }
1151
1152 if (size_change && attr->ia_size != i_size_read(inode)) {
1153 status = inode_newsize_ok(inode, attr->ia_size);
1154 if (status)
1155 goto bail_unlock;
1156
1157 inode_dio_wait(inode);
1158
1159 if (i_size_read(inode) > attr->ia_size) {
1160 if (ocfs2_should_order_data(inode)) {
1161 status = ocfs2_begin_ordered_truncate(inode,
1162 attr->ia_size);
1163 if (status)
1164 goto bail_unlock;
1165 }
1166 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1167 } else
1168 status = ocfs2_extend_file(inode, bh, attr->ia_size);
1169 if (status < 0) {
1170 if (status != -ENOSPC)
1171 mlog_errno(status);
1172 status = -ENOSPC;
1173 goto bail_unlock;
1174 }
1175 }
1176
1177 if ((attr->ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
1178 (attr->ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
1179 /*
1180 * Gather pointers to quota structures so that allocation /
1181 * freeing of quota structures happens here and not inside
1182 * dquot_transfer() where we have problems with lock ordering
1183 */
1184 if (attr->ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid
1185 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1186 OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1187 transfer_to[USRQUOTA] = dqget(sb, attr->ia_uid,
1188 USRQUOTA);
1189 if (!transfer_to[USRQUOTA]) {
1190 status = -ESRCH;
1191 goto bail_unlock;
1192 }
1193 }
1194 if (attr->ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid
1195 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1196 OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1197 transfer_to[GRPQUOTA] = dqget(sb, attr->ia_gid,
1198 GRPQUOTA);
1199 if (!transfer_to[GRPQUOTA]) {
1200 status = -ESRCH;
1201 goto bail_unlock;
1202 }
1203 }
1204 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1205 2 * ocfs2_quota_trans_credits(sb));
1206 if (IS_ERR(handle)) {
1207 status = PTR_ERR(handle);
1208 mlog_errno(status);
1209 goto bail_unlock;
1210 }
1211 status = __dquot_transfer(inode, transfer_to);
1212 if (status < 0)
1213 goto bail_commit;
1214 } else {
1215 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1216 if (IS_ERR(handle)) {
1217 status = PTR_ERR(handle);
1218 mlog_errno(status);
1219 goto bail_unlock;
1220 }
1221 }
1222
1223 /*
1224 * This will intentionally not wind up calling truncate_setsize(),
1225 * since all the work for a size change has been done above.
1226 * Otherwise, we could get into problems with truncate as
1227 * ip_alloc_sem is used there to protect against i_size
1228 * changes.
1229 *
1230 * XXX: this means the conditional below can probably be removed.
1231 */
1232 if ((attr->ia_valid & ATTR_SIZE) &&
1233 attr->ia_size != i_size_read(inode)) {
1234 status = vmtruncate(inode, attr->ia_size);
1235 if (status) {
1236 mlog_errno(status);
1237 goto bail_commit;
1238 }
1239 }
1240
1241 setattr_copy(inode, attr);
1242 mark_inode_dirty(inode);
1243
1244 status = ocfs2_mark_inode_dirty(handle, inode, bh);
1245 if (status < 0)
1246 mlog_errno(status);
1247
1248bail_commit:
1249 ocfs2_commit_trans(osb, handle);
1250bail_unlock:
1251 ocfs2_inode_unlock(inode, 1);
1252bail_unlock_rw:
1253 if (size_change)
1254 ocfs2_rw_unlock(inode, 1);
1255bail:
1256 brelse(bh);
1257
1258 /* Release quota pointers in case we acquired them */
1259 for (qtype = 0; qtype < MAXQUOTAS; qtype++)
1260 dqput(transfer_to[qtype]);
1261
1262 if (!status && attr->ia_valid & ATTR_MODE) {
1263 status = ocfs2_acl_chmod(inode);
1264 if (status < 0)
1265 mlog_errno(status);
1266 }
1267
1268 return status;
1269}
1270
1271int ocfs2_getattr(struct vfsmount *mnt,
1272 struct dentry *dentry,
1273 struct kstat *stat)
1274{
1275 struct inode *inode = dentry->d_inode;
1276 struct super_block *sb = dentry->d_inode->i_sb;
1277 struct ocfs2_super *osb = sb->s_fs_info;
1278 int err;
1279
1280 err = ocfs2_inode_revalidate(dentry);
1281 if (err) {
1282 if (err != -ENOENT)
1283 mlog_errno(err);
1284 goto bail;
1285 }
1286
1287 generic_fillattr(inode, stat);
1288
1289 /* We set the blksize from the cluster size for performance */
1290 stat->blksize = osb->s_clustersize;
1291
1292bail:
1293 return err;
1294}
1295
1296int ocfs2_permission(struct inode *inode, int mask)
1297{
1298 int ret;
1299
1300 if (mask & MAY_NOT_BLOCK)
1301 return -ECHILD;
1302
1303 ret = ocfs2_inode_lock(inode, NULL, 0);
1304 if (ret) {
1305 if (ret != -ENOENT)
1306 mlog_errno(ret);
1307 goto out;
1308 }
1309
1310 ret = generic_permission(inode, mask);
1311
1312 ocfs2_inode_unlock(inode, 0);
1313out:
1314 return ret;
1315}
1316
1317static int __ocfs2_write_remove_suid(struct inode *inode,
1318 struct buffer_head *bh)
1319{
1320 int ret;
1321 handle_t *handle;
1322 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1323 struct ocfs2_dinode *di;
1324
1325 trace_ocfs2_write_remove_suid(
1326 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1327 inode->i_mode);
1328
1329 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1330 if (IS_ERR(handle)) {
1331 ret = PTR_ERR(handle);
1332 mlog_errno(ret);
1333 goto out;
1334 }
1335
1336 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1337 OCFS2_JOURNAL_ACCESS_WRITE);
1338 if (ret < 0) {
1339 mlog_errno(ret);
1340 goto out_trans;
1341 }
1342
1343 inode->i_mode &= ~S_ISUID;
1344 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1345 inode->i_mode &= ~S_ISGID;
1346
1347 di = (struct ocfs2_dinode *) bh->b_data;
1348 di->i_mode = cpu_to_le16(inode->i_mode);
1349
1350 ocfs2_journal_dirty(handle, bh);
1351
1352out_trans:
1353 ocfs2_commit_trans(osb, handle);
1354out:
1355 return ret;
1356}
1357
1358/*
1359 * Will look for holes and unwritten extents in the range starting at
1360 * pos for count bytes (inclusive).
1361 */
1362static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1363 size_t count)
1364{
1365 int ret = 0;
1366 unsigned int extent_flags;
1367 u32 cpos, clusters, extent_len, phys_cpos;
1368 struct super_block *sb = inode->i_sb;
1369
1370 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1371 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1372
1373 while (clusters) {
1374 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1375 &extent_flags);
1376 if (ret < 0) {
1377 mlog_errno(ret);
1378 goto out;
1379 }
1380
1381 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
1382 ret = 1;
1383 break;
1384 }
1385
1386 if (extent_len > clusters)
1387 extent_len = clusters;
1388
1389 clusters -= extent_len;
1390 cpos += extent_len;
1391 }
1392out:
1393 return ret;
1394}
1395
1396static int ocfs2_write_remove_suid(struct inode *inode)
1397{
1398 int ret;
1399 struct buffer_head *bh = NULL;
1400
1401 ret = ocfs2_read_inode_block(inode, &bh);
1402 if (ret < 0) {
1403 mlog_errno(ret);
1404 goto out;
1405 }
1406
1407 ret = __ocfs2_write_remove_suid(inode, bh);
1408out:
1409 brelse(bh);
1410 return ret;
1411}
1412
1413/*
1414 * Allocate enough extents to cover the region starting at byte offset
1415 * start for len bytes. Existing extents are skipped, any extents
1416 * added are marked as "unwritten".
1417 */
1418static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1419 u64 start, u64 len)
1420{
1421 int ret;
1422 u32 cpos, phys_cpos, clusters, alloc_size;
1423 u64 end = start + len;
1424 struct buffer_head *di_bh = NULL;
1425
1426 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1427 ret = ocfs2_read_inode_block(inode, &di_bh);
1428 if (ret) {
1429 mlog_errno(ret);
1430 goto out;
1431 }
1432
1433 /*
1434 * Nothing to do if the requested reservation range
1435 * fits within the inode.
1436 */
1437 if (ocfs2_size_fits_inline_data(di_bh, end))
1438 goto out;
1439
1440 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1441 if (ret) {
1442 mlog_errno(ret);
1443 goto out;
1444 }
1445 }
1446
1447 /*
1448 * We consider both start and len to be inclusive.
1449 */
1450 cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1451 clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1452 clusters -= cpos;
1453
1454 while (clusters) {
1455 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1456 &alloc_size, NULL);
1457 if (ret) {
1458 mlog_errno(ret);
1459 goto out;
1460 }
1461
1462 /*
1463 * Hole or existing extent len can be arbitrary, so
1464 * cap it to our own allocation request.
1465 */
1466 if (alloc_size > clusters)
1467 alloc_size = clusters;
1468
1469 if (phys_cpos) {
1470 /*
1471 * We already have an allocation at this
1472 * region so we can safely skip it.
1473 */
1474 goto next;
1475 }
1476
1477 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1478 if (ret) {
1479 if (ret != -ENOSPC)
1480 mlog_errno(ret);
1481 goto out;
1482 }
1483
1484next:
1485 cpos += alloc_size;
1486 clusters -= alloc_size;
1487 }
1488
1489 ret = 0;
1490out:
1491
1492 brelse(di_bh);
1493 return ret;
1494}
1495
1496/*
1497 * Truncate a byte range, avoiding pages within partial clusters. This
1498 * preserves those pages for the zeroing code to write to.
1499 */
1500static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1501 u64 byte_len)
1502{
1503 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1504 loff_t start, end;
1505 struct address_space *mapping = inode->i_mapping;
1506
1507 start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1508 end = byte_start + byte_len;
1509 end = end & ~(osb->s_clustersize - 1);
1510
1511 if (start < end) {
1512 unmap_mapping_range(mapping, start, end - start, 0);
1513 truncate_inode_pages_range(mapping, start, end - 1);
1514 }
1515}
1516
1517static int ocfs2_zero_partial_clusters(struct inode *inode,
1518 u64 start, u64 len)
1519{
1520 int ret = 0;
1521 u64 tmpend, end = start + len;
1522 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1523 unsigned int csize = osb->s_clustersize;
1524 handle_t *handle;
1525
1526 /*
1527 * The "start" and "end" values are NOT necessarily part of
1528 * the range whose allocation is being deleted. Rather, this
1529 * is what the user passed in with the request. We must zero
1530 * partial clusters here. There's no need to worry about
1531 * physical allocation - the zeroing code knows to skip holes.
1532 */
1533 trace_ocfs2_zero_partial_clusters(
1534 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1535 (unsigned long long)start, (unsigned long long)end);
1536
1537 /*
1538 * If both edges are on a cluster boundary then there's no
1539 * zeroing required as the region is part of the allocation to
1540 * be truncated.
1541 */
1542 if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1543 goto out;
1544
1545 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1546 if (IS_ERR(handle)) {
1547 ret = PTR_ERR(handle);
1548 mlog_errno(ret);
1549 goto out;
1550 }
1551
1552 /*
1553 * We want to get the byte offset of the end of the 1st cluster.
1554 */
1555 tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1556 if (tmpend > end)
1557 tmpend = end;
1558
1559 trace_ocfs2_zero_partial_clusters_range1((unsigned long long)start,
1560 (unsigned long long)tmpend);
1561
1562 ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1563 if (ret)
1564 mlog_errno(ret);
1565
1566 if (tmpend < end) {
1567 /*
1568 * This may make start and end equal, but the zeroing
1569 * code will skip any work in that case so there's no
1570 * need to catch it up here.
1571 */
1572 start = end & ~(osb->s_clustersize - 1);
1573
1574 trace_ocfs2_zero_partial_clusters_range2(
1575 (unsigned long long)start, (unsigned long long)end);
1576
1577 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1578 if (ret)
1579 mlog_errno(ret);
1580 }
1581
1582 ocfs2_commit_trans(osb, handle);
1583out:
1584 return ret;
1585}
1586
1587static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1588{
1589 int i;
1590 struct ocfs2_extent_rec *rec = NULL;
1591
1592 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1593
1594 rec = &el->l_recs[i];
1595
1596 if (le32_to_cpu(rec->e_cpos) < pos)
1597 break;
1598 }
1599
1600 return i;
1601}
1602
1603/*
1604 * Helper to calculate the punching pos and length in one run, we handle the
1605 * following three cases in order:
1606 *
1607 * - remove the entire record
1608 * - remove a partial record
1609 * - no record needs to be removed (hole-punching completed)
1610*/
1611static void ocfs2_calc_trunc_pos(struct inode *inode,
1612 struct ocfs2_extent_list *el,
1613 struct ocfs2_extent_rec *rec,
1614 u32 trunc_start, u32 *trunc_cpos,
1615 u32 *trunc_len, u32 *trunc_end,
1616 u64 *blkno, int *done)
1617{
1618 int ret = 0;
1619 u32 coff, range;
1620
1621 range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1622
1623 if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1624 /*
1625 * remove an entire extent record.
1626 */
1627 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1628 /*
1629 * Skip holes if any.
1630 */
1631 if (range < *trunc_end)
1632 *trunc_end = range;
1633 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1634 *blkno = le64_to_cpu(rec->e_blkno);
1635 *trunc_end = le32_to_cpu(rec->e_cpos);
1636 } else if (range > trunc_start) {
1637 /*
1638 * remove a partial extent record, which means we're
1639 * removing the last extent record.
1640 */
1641 *trunc_cpos = trunc_start;
1642 /*
1643 * skip hole if any.
1644 */
1645 if (range < *trunc_end)
1646 *trunc_end = range;
1647 *trunc_len = *trunc_end - trunc_start;
1648 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1649 *blkno = le64_to_cpu(rec->e_blkno) +
1650 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1651 *trunc_end = trunc_start;
1652 } else {
1653 /*
1654 * It may have two following possibilities:
1655 *
1656 * - last record has been removed
1657 * - trunc_start was within a hole
1658 *
1659 * both two cases mean the completion of hole punching.
1660 */
1661 ret = 1;
1662 }
1663
1664 *done = ret;
1665}
1666
1667static int ocfs2_remove_inode_range(struct inode *inode,
1668 struct buffer_head *di_bh, u64 byte_start,
1669 u64 byte_len)
1670{
1671 int ret = 0, flags = 0, done = 0, i;
1672 u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1673 u32 cluster_in_el;
1674 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1675 struct ocfs2_cached_dealloc_ctxt dealloc;
1676 struct address_space *mapping = inode->i_mapping;
1677 struct ocfs2_extent_tree et;
1678 struct ocfs2_path *path = NULL;
1679 struct ocfs2_extent_list *el = NULL;
1680 struct ocfs2_extent_rec *rec = NULL;
1681 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1682 u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1683
1684 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1685 ocfs2_init_dealloc_ctxt(&dealloc);
1686
1687 trace_ocfs2_remove_inode_range(
1688 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1689 (unsigned long long)byte_start,
1690 (unsigned long long)byte_len);
1691
1692 if (byte_len == 0)
1693 return 0;
1694
1695 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1696 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1697 byte_start + byte_len, 0);
1698 if (ret) {
1699 mlog_errno(ret);
1700 goto out;
1701 }
1702 /*
1703 * There's no need to get fancy with the page cache
1704 * truncate of an inline-data inode. We're talking
1705 * about less than a page here, which will be cached
1706 * in the dinode buffer anyway.
1707 */
1708 unmap_mapping_range(mapping, 0, 0, 0);
1709 truncate_inode_pages(mapping, 0);
1710 goto out;
1711 }
1712
1713 /*
1714 * For reflinks, we may need to CoW 2 clusters which might be
1715 * partially zero'd later, if hole's start and end offset were
1716 * within one cluster(means is not exactly aligned to clustersize).
1717 */
1718
1719 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
1720
1721 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1722 if (ret) {
1723 mlog_errno(ret);
1724 goto out;
1725 }
1726
1727 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1728 if (ret) {
1729 mlog_errno(ret);
1730 goto out;
1731 }
1732 }
1733
1734 trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1735 trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1736 cluster_in_el = trunc_end;
1737
1738 ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1739 if (ret) {
1740 mlog_errno(ret);
1741 goto out;
1742 }
1743
1744 path = ocfs2_new_path_from_et(&et);
1745 if (!path) {
1746 ret = -ENOMEM;
1747 mlog_errno(ret);
1748 goto out;
1749 }
1750
1751 while (trunc_end > trunc_start) {
1752
1753 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1754 cluster_in_el);
1755 if (ret) {
1756 mlog_errno(ret);
1757 goto out;
1758 }
1759
1760 el = path_leaf_el(path);
1761
1762 i = ocfs2_find_rec(el, trunc_end);
1763 /*
1764 * Need to go to previous extent block.
1765 */
1766 if (i < 0) {
1767 if (path->p_tree_depth == 0)
1768 break;
1769
1770 ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1771 path,
1772 &cluster_in_el);
1773 if (ret) {
1774 mlog_errno(ret);
1775 goto out;
1776 }
1777
1778 /*
1779 * We've reached the leftmost extent block,
1780 * it's safe to leave.
1781 */
1782 if (cluster_in_el == 0)
1783 break;
1784
1785 /*
1786 * The 'pos' searched for previous extent block is
1787 * always one cluster less than actual trunc_end.
1788 */
1789 trunc_end = cluster_in_el + 1;
1790
1791 ocfs2_reinit_path(path, 1);
1792
1793 continue;
1794
1795 } else
1796 rec = &el->l_recs[i];
1797
1798 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1799 &trunc_len, &trunc_end, &blkno, &done);
1800 if (done)
1801 break;
1802
1803 flags = rec->e_flags;
1804 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1805
1806 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1807 phys_cpos, trunc_len, flags,
1808 &dealloc, refcount_loc);
1809 if (ret < 0) {
1810 mlog_errno(ret);
1811 goto out;
1812 }
1813
1814 cluster_in_el = trunc_end;
1815
1816 ocfs2_reinit_path(path, 1);
1817 }
1818
1819 ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1820
1821out:
1822 ocfs2_schedule_truncate_log_flush(osb, 1);
1823 ocfs2_run_deallocs(osb, &dealloc);
1824
1825 return ret;
1826}
1827
1828/*
1829 * Parts of this function taken from xfs_change_file_space()
1830 */
1831static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1832 loff_t f_pos, unsigned int cmd,
1833 struct ocfs2_space_resv *sr,
1834 int change_size)
1835{
1836 int ret;
1837 s64 llen;
1838 loff_t size;
1839 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1840 struct buffer_head *di_bh = NULL;
1841 handle_t *handle;
1842 unsigned long long max_off = inode->i_sb->s_maxbytes;
1843
1844 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1845 return -EROFS;
1846
1847 mutex_lock(&inode->i_mutex);
1848
1849 /*
1850 * This prevents concurrent writes on other nodes
1851 */
1852 ret = ocfs2_rw_lock(inode, 1);
1853 if (ret) {
1854 mlog_errno(ret);
1855 goto out;
1856 }
1857
1858 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1859 if (ret) {
1860 mlog_errno(ret);
1861 goto out_rw_unlock;
1862 }
1863
1864 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1865 ret = -EPERM;
1866 goto out_inode_unlock;
1867 }
1868
1869 switch (sr->l_whence) {
1870 case 0: /*SEEK_SET*/
1871 break;
1872 case 1: /*SEEK_CUR*/
1873 sr->l_start += f_pos;
1874 break;
1875 case 2: /*SEEK_END*/
1876 sr->l_start += i_size_read(inode);
1877 break;
1878 default:
1879 ret = -EINVAL;
1880 goto out_inode_unlock;
1881 }
1882 sr->l_whence = 0;
1883
1884 llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1885
1886 if (sr->l_start < 0
1887 || sr->l_start > max_off
1888 || (sr->l_start + llen) < 0
1889 || (sr->l_start + llen) > max_off) {
1890 ret = -EINVAL;
1891 goto out_inode_unlock;
1892 }
1893 size = sr->l_start + sr->l_len;
1894
1895 if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) {
1896 if (sr->l_len <= 0) {
1897 ret = -EINVAL;
1898 goto out_inode_unlock;
1899 }
1900 }
1901
1902 if (file && should_remove_suid(file->f_path.dentry)) {
1903 ret = __ocfs2_write_remove_suid(inode, di_bh);
1904 if (ret) {
1905 mlog_errno(ret);
1906 goto out_inode_unlock;
1907 }
1908 }
1909
1910 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1911 switch (cmd) {
1912 case OCFS2_IOC_RESVSP:
1913 case OCFS2_IOC_RESVSP64:
1914 /*
1915 * This takes unsigned offsets, but the signed ones we
1916 * pass have been checked against overflow above.
1917 */
1918 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1919 sr->l_len);
1920 break;
1921 case OCFS2_IOC_UNRESVSP:
1922 case OCFS2_IOC_UNRESVSP64:
1923 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1924 sr->l_len);
1925 break;
1926 default:
1927 ret = -EINVAL;
1928 }
1929 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1930 if (ret) {
1931 mlog_errno(ret);
1932 goto out_inode_unlock;
1933 }
1934
1935 /*
1936 * We update c/mtime for these changes
1937 */
1938 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1939 if (IS_ERR(handle)) {
1940 ret = PTR_ERR(handle);
1941 mlog_errno(ret);
1942 goto out_inode_unlock;
1943 }
1944
1945 if (change_size && i_size_read(inode) < size)
1946 i_size_write(inode, size);
1947
1948 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1949 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1950 if (ret < 0)
1951 mlog_errno(ret);
1952
1953 if (file && (file->f_flags & O_SYNC))
1954 handle->h_sync = 1;
1955
1956 ocfs2_commit_trans(osb, handle);
1957
1958out_inode_unlock:
1959 brelse(di_bh);
1960 ocfs2_inode_unlock(inode, 1);
1961out_rw_unlock:
1962 ocfs2_rw_unlock(inode, 1);
1963
1964out:
1965 mutex_unlock(&inode->i_mutex);
1966 return ret;
1967}
1968
1969int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1970 struct ocfs2_space_resv *sr)
1971{
1972 struct inode *inode = file->f_path.dentry->d_inode;
1973 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1974
1975 if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1976 !ocfs2_writes_unwritten_extents(osb))
1977 return -ENOTTY;
1978 else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1979 !ocfs2_sparse_alloc(osb))
1980 return -ENOTTY;
1981
1982 if (!S_ISREG(inode->i_mode))
1983 return -EINVAL;
1984
1985 if (!(file->f_mode & FMODE_WRITE))
1986 return -EBADF;
1987
1988 return __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1989}
1990
1991static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
1992 loff_t len)
1993{
1994 struct inode *inode = file->f_path.dentry->d_inode;
1995 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1996 struct ocfs2_space_resv sr;
1997 int change_size = 1;
1998 int cmd = OCFS2_IOC_RESVSP64;
1999
2000 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2001 return -EOPNOTSUPP;
2002 if (!ocfs2_writes_unwritten_extents(osb))
2003 return -EOPNOTSUPP;
2004
2005 if (mode & FALLOC_FL_KEEP_SIZE)
2006 change_size = 0;
2007
2008 if (mode & FALLOC_FL_PUNCH_HOLE)
2009 cmd = OCFS2_IOC_UNRESVSP64;
2010
2011 sr.l_whence = 0;
2012 sr.l_start = (s64)offset;
2013 sr.l_len = (s64)len;
2014
2015 return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2016 change_size);
2017}
2018
2019int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2020 size_t count)
2021{
2022 int ret = 0;
2023 unsigned int extent_flags;
2024 u32 cpos, clusters, extent_len, phys_cpos;
2025 struct super_block *sb = inode->i_sb;
2026
2027 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2028 !(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) ||
2029 OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2030 return 0;
2031
2032 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2033 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2034
2035 while (clusters) {
2036 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2037 &extent_flags);
2038 if (ret < 0) {
2039 mlog_errno(ret);
2040 goto out;
2041 }
2042
2043 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2044 ret = 1;
2045 break;
2046 }
2047
2048 if (extent_len > clusters)
2049 extent_len = clusters;
2050
2051 clusters -= extent_len;
2052 cpos += extent_len;
2053 }
2054out:
2055 return ret;
2056}
2057
2058static void ocfs2_aiodio_wait(struct inode *inode)
2059{
2060 wait_queue_head_t *wq = ocfs2_ioend_wq(inode);
2061
2062 wait_event(*wq, (atomic_read(&OCFS2_I(inode)->ip_unaligned_aio) == 0));
2063}
2064
2065static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
2066{
2067 int blockmask = inode->i_sb->s_blocksize - 1;
2068 loff_t final_size = pos + count;
2069
2070 if ((pos & blockmask) || (final_size & blockmask))
2071 return 1;
2072 return 0;
2073}
2074
2075static int ocfs2_prepare_inode_for_refcount(struct inode *inode,
2076 struct file *file,
2077 loff_t pos, size_t count,
2078 int *meta_level)
2079{
2080 int ret;
2081 struct buffer_head *di_bh = NULL;
2082 u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2083 u32 clusters =
2084 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2085
2086 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2087 if (ret) {
2088 mlog_errno(ret);
2089 goto out;
2090 }
2091
2092 *meta_level = 1;
2093
2094 ret = ocfs2_refcount_cow(inode, file, di_bh, cpos, clusters, UINT_MAX);
2095 if (ret)
2096 mlog_errno(ret);
2097out:
2098 brelse(di_bh);
2099 return ret;
2100}
2101
2102static int ocfs2_prepare_inode_for_write(struct file *file,
2103 loff_t *ppos,
2104 size_t count,
2105 int appending,
2106 int *direct_io,
2107 int *has_refcount)
2108{
2109 int ret = 0, meta_level = 0;
2110 struct dentry *dentry = file->f_path.dentry;
2111 struct inode *inode = dentry->d_inode;
2112 loff_t saved_pos = 0, end;
2113
2114 /*
2115 * We start with a read level meta lock and only jump to an ex
2116 * if we need to make modifications here.
2117 */
2118 for(;;) {
2119 ret = ocfs2_inode_lock(inode, NULL, meta_level);
2120 if (ret < 0) {
2121 meta_level = -1;
2122 mlog_errno(ret);
2123 goto out;
2124 }
2125
2126 /* Clear suid / sgid if necessary. We do this here
2127 * instead of later in the write path because
2128 * remove_suid() calls ->setattr without any hint that
2129 * we may have already done our cluster locking. Since
2130 * ocfs2_setattr() *must* take cluster locks to
2131 * proceed, this will lead us to recursively lock the
2132 * inode. There's also the dinode i_size state which
2133 * can be lost via setattr during extending writes (we
2134 * set inode->i_size at the end of a write. */
2135 if (should_remove_suid(dentry)) {
2136 if (meta_level == 0) {
2137 ocfs2_inode_unlock(inode, meta_level);
2138 meta_level = 1;
2139 continue;
2140 }
2141
2142 ret = ocfs2_write_remove_suid(inode);
2143 if (ret < 0) {
2144 mlog_errno(ret);
2145 goto out_unlock;
2146 }
2147 }
2148
2149 /* work on a copy of ppos until we're sure that we won't have
2150 * to recalculate it due to relocking. */
2151 if (appending)
2152 saved_pos = i_size_read(inode);
2153 else
2154 saved_pos = *ppos;
2155
2156 end = saved_pos + count;
2157
2158 ret = ocfs2_check_range_for_refcount(inode, saved_pos, count);
2159 if (ret == 1) {
2160 ocfs2_inode_unlock(inode, meta_level);
2161 meta_level = -1;
2162
2163 ret = ocfs2_prepare_inode_for_refcount(inode,
2164 file,
2165 saved_pos,
2166 count,
2167 &meta_level);
2168 if (has_refcount)
2169 *has_refcount = 1;
2170 if (direct_io)
2171 *direct_io = 0;
2172 }
2173
2174 if (ret < 0) {
2175 mlog_errno(ret);
2176 goto out_unlock;
2177 }
2178
2179 /*
2180 * Skip the O_DIRECT checks if we don't need
2181 * them.
2182 */
2183 if (!direct_io || !(*direct_io))
2184 break;
2185
2186 /*
2187 * There's no sane way to do direct writes to an inode
2188 * with inline data.
2189 */
2190 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2191 *direct_io = 0;
2192 break;
2193 }
2194
2195 /*
2196 * Allowing concurrent direct writes means
2197 * i_size changes wouldn't be synchronized, so
2198 * one node could wind up truncating another
2199 * nodes writes.
2200 */
2201 if (end > i_size_read(inode)) {
2202 *direct_io = 0;
2203 break;
2204 }
2205
2206 /*
2207 * We don't fill holes during direct io, so
2208 * check for them here. If any are found, the
2209 * caller will have to retake some cluster
2210 * locks and initiate the io as buffered.
2211 */
2212 ret = ocfs2_check_range_for_holes(inode, saved_pos, count);
2213 if (ret == 1) {
2214 *direct_io = 0;
2215 ret = 0;
2216 } else if (ret < 0)
2217 mlog_errno(ret);
2218 break;
2219 }
2220
2221 if (appending)
2222 *ppos = saved_pos;
2223
2224out_unlock:
2225 trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
2226 saved_pos, appending, count,
2227 direct_io, has_refcount);
2228
2229 if (meta_level >= 0)
2230 ocfs2_inode_unlock(inode, meta_level);
2231
2232out:
2233 return ret;
2234}
2235
2236static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
2237 const struct iovec *iov,
2238 unsigned long nr_segs,
2239 loff_t pos)
2240{
2241 int ret, direct_io, appending, rw_level, have_alloc_sem = 0;
2242 int can_do_direct, has_refcount = 0;
2243 ssize_t written = 0;
2244 size_t ocount; /* original count */
2245 size_t count; /* after file limit checks */
2246 loff_t old_size, *ppos = &iocb->ki_pos;
2247 u32 old_clusters;
2248 struct file *file = iocb->ki_filp;
2249 struct inode *inode = file->f_path.dentry->d_inode;
2250 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2251 int full_coherency = !(osb->s_mount_opt &
2252 OCFS2_MOUNT_COHERENCY_BUFFERED);
2253 int unaligned_dio = 0;
2254
2255 trace_ocfs2_file_aio_write(inode, file, file->f_path.dentry,
2256 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2257 file->f_path.dentry->d_name.len,
2258 file->f_path.dentry->d_name.name,
2259 (unsigned int)nr_segs);
2260
2261 if (iocb->ki_left == 0)
2262 return 0;
2263
2264 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2265
2266 appending = file->f_flags & O_APPEND ? 1 : 0;
2267 direct_io = file->f_flags & O_DIRECT ? 1 : 0;
2268
2269 mutex_lock(&inode->i_mutex);
2270
2271 ocfs2_iocb_clear_sem_locked(iocb);
2272
2273relock:
2274 /* to match setattr's i_mutex -> rw_lock ordering */
2275 if (direct_io) {
2276 have_alloc_sem = 1;
2277 /* communicate with ocfs2_dio_end_io */
2278 ocfs2_iocb_set_sem_locked(iocb);
2279 }
2280
2281 /*
2282 * Concurrent O_DIRECT writes are allowed with
2283 * mount_option "coherency=buffered".
2284 */
2285 rw_level = (!direct_io || full_coherency);
2286
2287 ret = ocfs2_rw_lock(inode, rw_level);
2288 if (ret < 0) {
2289 mlog_errno(ret);
2290 goto out_sems;
2291 }
2292
2293 /*
2294 * O_DIRECT writes with "coherency=full" need to take EX cluster
2295 * inode_lock to guarantee coherency.
2296 */
2297 if (direct_io && full_coherency) {
2298 /*
2299 * We need to take and drop the inode lock to force
2300 * other nodes to drop their caches. Buffered I/O
2301 * already does this in write_begin().
2302 */
2303 ret = ocfs2_inode_lock(inode, NULL, 1);
2304 if (ret < 0) {
2305 mlog_errno(ret);
2306 goto out_sems;
2307 }
2308
2309 ocfs2_inode_unlock(inode, 1);
2310 }
2311
2312 can_do_direct = direct_io;
2313 ret = ocfs2_prepare_inode_for_write(file, ppos,
2314 iocb->ki_left, appending,
2315 &can_do_direct, &has_refcount);
2316 if (ret < 0) {
2317 mlog_errno(ret);
2318 goto out;
2319 }
2320
2321 if (direct_io && !is_sync_kiocb(iocb))
2322 unaligned_dio = ocfs2_is_io_unaligned(inode, iocb->ki_left,
2323 *ppos);
2324
2325 /*
2326 * We can't complete the direct I/O as requested, fall back to
2327 * buffered I/O.
2328 */
2329 if (direct_io && !can_do_direct) {
2330 ocfs2_rw_unlock(inode, rw_level);
2331
2332 have_alloc_sem = 0;
2333 rw_level = -1;
2334
2335 direct_io = 0;
2336 goto relock;
2337 }
2338
2339 if (unaligned_dio) {
2340 /*
2341 * Wait on previous unaligned aio to complete before
2342 * proceeding.
2343 */
2344 ocfs2_aiodio_wait(inode);
2345
2346 /* Mark the iocb as needing a decrement in ocfs2_dio_end_io */
2347 atomic_inc(&OCFS2_I(inode)->ip_unaligned_aio);
2348 ocfs2_iocb_set_unaligned_aio(iocb);
2349 }
2350
2351 /*
2352 * To later detect whether a journal commit for sync writes is
2353 * necessary, we sample i_size, and cluster count here.
2354 */
2355 old_size = i_size_read(inode);
2356 old_clusters = OCFS2_I(inode)->ip_clusters;
2357
2358 /* communicate with ocfs2_dio_end_io */
2359 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2360
2361 ret = generic_segment_checks(iov, &nr_segs, &ocount,
2362 VERIFY_READ);
2363 if (ret)
2364 goto out_dio;
2365
2366 count = ocount;
2367 ret = generic_write_checks(file, ppos, &count,
2368 S_ISBLK(inode->i_mode));
2369 if (ret)
2370 goto out_dio;
2371
2372 if (direct_io) {
2373 written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
2374 ppos, count, ocount);
2375 if (written < 0) {
2376 ret = written;
2377 goto out_dio;
2378 }
2379 } else {
2380 current->backing_dev_info = file->f_mapping->backing_dev_info;
2381 written = generic_file_buffered_write(iocb, iov, nr_segs, *ppos,
2382 ppos, count, 0);
2383 current->backing_dev_info = NULL;
2384 }
2385
2386out_dio:
2387 /* buffered aio wouldn't have proper lock coverage today */
2388 BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
2389
2390 if (((file->f_flags & O_DSYNC) && !direct_io) || IS_SYNC(inode) ||
2391 ((file->f_flags & O_DIRECT) && !direct_io)) {
2392 ret = filemap_fdatawrite_range(file->f_mapping, pos,
2393 pos + count - 1);
2394 if (ret < 0)
2395 written = ret;
2396
2397 if (!ret && ((old_size != i_size_read(inode)) ||
2398 (old_clusters != OCFS2_I(inode)->ip_clusters) ||
2399 has_refcount)) {
2400 ret = jbd2_journal_force_commit(osb->journal->j_journal);
2401 if (ret < 0)
2402 written = ret;
2403 }
2404
2405 if (!ret)
2406 ret = filemap_fdatawait_range(file->f_mapping, pos,
2407 pos + count - 1);
2408 }
2409
2410 /*
2411 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2412 * function pointer which is called when o_direct io completes so that
2413 * it can unlock our rw lock.
2414 * Unfortunately there are error cases which call end_io and others
2415 * that don't. so we don't have to unlock the rw_lock if either an
2416 * async dio is going to do it in the future or an end_io after an
2417 * error has already done it.
2418 */
2419 if ((ret == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2420 rw_level = -1;
2421 have_alloc_sem = 0;
2422 unaligned_dio = 0;
2423 }
2424
2425 if (unaligned_dio) {
2426 ocfs2_iocb_clear_unaligned_aio(iocb);
2427 atomic_dec(&OCFS2_I(inode)->ip_unaligned_aio);
2428 }
2429
2430out:
2431 if (rw_level != -1)
2432 ocfs2_rw_unlock(inode, rw_level);
2433
2434out_sems:
2435 if (have_alloc_sem)
2436 ocfs2_iocb_clear_sem_locked(iocb);
2437
2438 mutex_unlock(&inode->i_mutex);
2439
2440 if (written)
2441 ret = written;
2442 return ret;
2443}
2444
2445static int ocfs2_splice_to_file(struct pipe_inode_info *pipe,
2446 struct file *out,
2447 struct splice_desc *sd)
2448{
2449 int ret;
2450
2451 ret = ocfs2_prepare_inode_for_write(out, &sd->pos,
2452 sd->total_len, 0, NULL, NULL);
2453 if (ret < 0) {
2454 mlog_errno(ret);
2455 return ret;
2456 }
2457
2458 return splice_from_pipe_feed(pipe, sd, pipe_to_file);
2459}
2460
2461static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
2462 struct file *out,
2463 loff_t *ppos,
2464 size_t len,
2465 unsigned int flags)
2466{
2467 int ret;
2468 struct address_space *mapping = out->f_mapping;
2469 struct inode *inode = mapping->host;
2470 struct splice_desc sd = {
2471 .total_len = len,
2472 .flags = flags,
2473 .pos = *ppos,
2474 .u.file = out,
2475 };
2476
2477
2478 trace_ocfs2_file_splice_write(inode, out, out->f_path.dentry,
2479 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2480 out->f_path.dentry->d_name.len,
2481 out->f_path.dentry->d_name.name, len);
2482
2483 if (pipe->inode)
2484 mutex_lock_nested(&pipe->inode->i_mutex, I_MUTEX_PARENT);
2485
2486 splice_from_pipe_begin(&sd);
2487 do {
2488 ret = splice_from_pipe_next(pipe, &sd);
2489 if (ret <= 0)
2490 break;
2491
2492 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2493 ret = ocfs2_rw_lock(inode, 1);
2494 if (ret < 0)
2495 mlog_errno(ret);
2496 else {
2497 ret = ocfs2_splice_to_file(pipe, out, &sd);
2498 ocfs2_rw_unlock(inode, 1);
2499 }
2500 mutex_unlock(&inode->i_mutex);
2501 } while (ret > 0);
2502 splice_from_pipe_end(pipe, &sd);
2503
2504 if (pipe->inode)
2505 mutex_unlock(&pipe->inode->i_mutex);
2506
2507 if (sd.num_spliced)
2508 ret = sd.num_spliced;
2509
2510 if (ret > 0) {
2511 unsigned long nr_pages;
2512 int err;
2513
2514 nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
2515
2516 err = generic_write_sync(out, *ppos, ret);
2517 if (err)
2518 ret = err;
2519 else
2520 *ppos += ret;
2521
2522 balance_dirty_pages_ratelimited_nr(mapping, nr_pages);
2523 }
2524
2525 return ret;
2526}
2527
2528static ssize_t ocfs2_file_splice_read(struct file *in,
2529 loff_t *ppos,
2530 struct pipe_inode_info *pipe,
2531 size_t len,
2532 unsigned int flags)
2533{
2534 int ret = 0, lock_level = 0;
2535 struct inode *inode = in->f_path.dentry->d_inode;
2536
2537 trace_ocfs2_file_splice_read(inode, in, in->f_path.dentry,
2538 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2539 in->f_path.dentry->d_name.len,
2540 in->f_path.dentry->d_name.name, len);
2541
2542 /*
2543 * See the comment in ocfs2_file_aio_read()
2544 */
2545 ret = ocfs2_inode_lock_atime(inode, in->f_vfsmnt, &lock_level);
2546 if (ret < 0) {
2547 mlog_errno(ret);
2548 goto bail;
2549 }
2550 ocfs2_inode_unlock(inode, lock_level);
2551
2552 ret = generic_file_splice_read(in, ppos, pipe, len, flags);
2553
2554bail:
2555 return ret;
2556}
2557
2558static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
2559 const struct iovec *iov,
2560 unsigned long nr_segs,
2561 loff_t pos)
2562{
2563 int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
2564 struct file *filp = iocb->ki_filp;
2565 struct inode *inode = filp->f_path.dentry->d_inode;
2566
2567 trace_ocfs2_file_aio_read(inode, filp, filp->f_path.dentry,
2568 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2569 filp->f_path.dentry->d_name.len,
2570 filp->f_path.dentry->d_name.name, nr_segs);
2571
2572
2573 if (!inode) {
2574 ret = -EINVAL;
2575 mlog_errno(ret);
2576 goto bail;
2577 }
2578
2579 ocfs2_iocb_clear_sem_locked(iocb);
2580
2581 /*
2582 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
2583 * need locks to protect pending reads from racing with truncate.
2584 */
2585 if (filp->f_flags & O_DIRECT) {
2586 have_alloc_sem = 1;
2587 ocfs2_iocb_set_sem_locked(iocb);
2588
2589 ret = ocfs2_rw_lock(inode, 0);
2590 if (ret < 0) {
2591 mlog_errno(ret);
2592 goto bail;
2593 }
2594 rw_level = 0;
2595 /* communicate with ocfs2_dio_end_io */
2596 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2597 }
2598
2599 /*
2600 * We're fine letting folks race truncates and extending
2601 * writes with read across the cluster, just like they can
2602 * locally. Hence no rw_lock during read.
2603 *
2604 * Take and drop the meta data lock to update inode fields
2605 * like i_size. This allows the checks down below
2606 * generic_file_aio_read() a chance of actually working.
2607 */
2608 ret = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level);
2609 if (ret < 0) {
2610 mlog_errno(ret);
2611 goto bail;
2612 }
2613 ocfs2_inode_unlock(inode, lock_level);
2614
2615 ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
2616 trace_generic_file_aio_read_ret(ret);
2617
2618 /* buffered aio wouldn't have proper lock coverage today */
2619 BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
2620
2621 /* see ocfs2_file_aio_write */
2622 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2623 rw_level = -1;
2624 have_alloc_sem = 0;
2625 }
2626
2627bail:
2628 if (have_alloc_sem)
2629 ocfs2_iocb_clear_sem_locked(iocb);
2630
2631 if (rw_level != -1)
2632 ocfs2_rw_unlock(inode, rw_level);
2633
2634 return ret;
2635}
2636
2637/* Refer generic_file_llseek_unlocked() */
2638static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int origin)
2639{
2640 struct inode *inode = file->f_mapping->host;
2641 int ret = 0;
2642
2643 mutex_lock(&inode->i_mutex);
2644
2645 switch (origin) {
2646 case SEEK_SET:
2647 break;
2648 case SEEK_END:
2649 offset += inode->i_size;
2650 break;
2651 case SEEK_CUR:
2652 if (offset == 0) {
2653 offset = file->f_pos;
2654 goto out;
2655 }
2656 offset += file->f_pos;
2657 break;
2658 case SEEK_DATA:
2659 case SEEK_HOLE:
2660 ret = ocfs2_seek_data_hole_offset(file, &offset, origin);
2661 if (ret)
2662 goto out;
2663 break;
2664 default:
2665 ret = -EINVAL;
2666 goto out;
2667 }
2668
2669 if (offset < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET))
2670 ret = -EINVAL;
2671 if (!ret && offset > inode->i_sb->s_maxbytes)
2672 ret = -EINVAL;
2673 if (ret)
2674 goto out;
2675
2676 if (offset != file->f_pos) {
2677 file->f_pos = offset;
2678 file->f_version = 0;
2679 }
2680
2681out:
2682 mutex_unlock(&inode->i_mutex);
2683 if (ret)
2684 return ret;
2685 return offset;
2686}
2687
2688const struct inode_operations ocfs2_file_iops = {
2689 .setattr = ocfs2_setattr,
2690 .getattr = ocfs2_getattr,
2691 .permission = ocfs2_permission,
2692 .setxattr = generic_setxattr,
2693 .getxattr = generic_getxattr,
2694 .listxattr = ocfs2_listxattr,
2695 .removexattr = generic_removexattr,
2696 .fiemap = ocfs2_fiemap,
2697 .get_acl = ocfs2_iop_get_acl,
2698};
2699
2700const struct inode_operations ocfs2_special_file_iops = {
2701 .setattr = ocfs2_setattr,
2702 .getattr = ocfs2_getattr,
2703 .permission = ocfs2_permission,
2704 .get_acl = ocfs2_iop_get_acl,
2705};
2706
2707/*
2708 * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2709 * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2710 */
2711const struct file_operations ocfs2_fops = {
2712 .llseek = ocfs2_file_llseek,
2713 .read = do_sync_read,
2714 .write = do_sync_write,
2715 .mmap = ocfs2_mmap,
2716 .fsync = ocfs2_sync_file,
2717 .release = ocfs2_file_release,
2718 .open = ocfs2_file_open,
2719 .aio_read = ocfs2_file_aio_read,
2720 .aio_write = ocfs2_file_aio_write,
2721 .unlocked_ioctl = ocfs2_ioctl,
2722#ifdef CONFIG_COMPAT
2723 .compat_ioctl = ocfs2_compat_ioctl,
2724#endif
2725 .lock = ocfs2_lock,
2726 .flock = ocfs2_flock,
2727 .splice_read = ocfs2_file_splice_read,
2728 .splice_write = ocfs2_file_splice_write,
2729 .fallocate = ocfs2_fallocate,
2730};
2731
2732const struct file_operations ocfs2_dops = {
2733 .llseek = generic_file_llseek,
2734 .read = generic_read_dir,
2735 .readdir = ocfs2_readdir,
2736 .fsync = ocfs2_sync_file,
2737 .release = ocfs2_dir_release,
2738 .open = ocfs2_dir_open,
2739 .unlocked_ioctl = ocfs2_ioctl,
2740#ifdef CONFIG_COMPAT
2741 .compat_ioctl = ocfs2_compat_ioctl,
2742#endif
2743 .lock = ocfs2_lock,
2744 .flock = ocfs2_flock,
2745};
2746
2747/*
2748 * POSIX-lockless variants of our file_operations.
2749 *
2750 * These will be used if the underlying cluster stack does not support
2751 * posix file locking, if the user passes the "localflocks" mount
2752 * option, or if we have a local-only fs.
2753 *
2754 * ocfs2_flock is in here because all stacks handle UNIX file locks,
2755 * so we still want it in the case of no stack support for
2756 * plocks. Internally, it will do the right thing when asked to ignore
2757 * the cluster.
2758 */
2759const struct file_operations ocfs2_fops_no_plocks = {
2760 .llseek = ocfs2_file_llseek,
2761 .read = do_sync_read,
2762 .write = do_sync_write,
2763 .mmap = ocfs2_mmap,
2764 .fsync = ocfs2_sync_file,
2765 .release = ocfs2_file_release,
2766 .open = ocfs2_file_open,
2767 .aio_read = ocfs2_file_aio_read,
2768 .aio_write = ocfs2_file_aio_write,
2769 .unlocked_ioctl = ocfs2_ioctl,
2770#ifdef CONFIG_COMPAT
2771 .compat_ioctl = ocfs2_compat_ioctl,
2772#endif
2773 .flock = ocfs2_flock,
2774 .splice_read = ocfs2_file_splice_read,
2775 .splice_write = ocfs2_file_splice_write,
2776 .fallocate = ocfs2_fallocate,
2777};
2778
2779const struct file_operations ocfs2_dops_no_plocks = {
2780 .llseek = generic_file_llseek,
2781 .read = generic_read_dir,
2782 .readdir = ocfs2_readdir,
2783 .fsync = ocfs2_sync_file,
2784 .release = ocfs2_dir_release,
2785 .open = ocfs2_dir_open,
2786 .unlocked_ioctl = ocfs2_ioctl,
2787#ifdef CONFIG_COMPAT
2788 .compat_ioctl = ocfs2_compat_ioctl,
2789#endif
2790 .flock = ocfs2_flock,
2791};