1 // SPDX-License-Identifier: GPL-2.0-or-later 1
2 /*
3 * file.c
4 *
5 * File open, close, extend, truncate
6 *
7 * Copyright (C) 2002, 2004 Oracle. All right
8 */
9
10 #include <linux/capability.h>
11 #include <linux/fs.h>
12 #include <linux/types.h>
13 #include <linux/slab.h>
14 #include <linux/highmem.h>
15 #include <linux/pagemap.h>
16 #include <linux/uio.h>
17 #include <linux/sched.h>
18 #include <linux/splice.h>
19 #include <linux/mount.h>
20 #include <linux/writeback.h>
21 #include <linux/falloc.h>
22 #include <linux/quotaops.h>
23 #include <linux/blkdev.h>
24 #include <linux/backing-dev.h>
25
26 #include <cluster/masklog.h>
27
28 #include "ocfs2.h"
29
30 #include "alloc.h"
31 #include "aops.h"
32 #include "dir.h"
33 #include "dlmglue.h"
34 #include "extent_map.h"
35 #include "file.h"
36 #include "sysfile.h"
37 #include "inode.h"
38 #include "ioctl.h"
39 #include "journal.h"
40 #include "locks.h"
41 #include "mmap.h"
42 #include "suballoc.h"
43 #include "super.h"
44 #include "xattr.h"
45 #include "acl.h"
46 #include "quota.h"
47 #include "refcounttree.h"
48 #include "ocfs2_trace.h"
49
50 #include "buffer_head_io.h"
51
52 static int ocfs2_init_file_private(struct inod
53 {
54 struct ocfs2_file_private *fp;
55
56 fp = kzalloc(sizeof(struct ocfs2_file_
57 if (!fp)
58 return -ENOMEM;
59
60 fp->fp_file = file;
61 mutex_init(&fp->fp_mutex);
62 ocfs2_file_lock_res_init(&fp->fp_flock
63 file->private_data = fp;
64
65 return 0;
66 }
67
68 static void ocfs2_free_file_private(struct ino
69 {
70 struct ocfs2_file_private *fp = file->
71 struct ocfs2_super *osb = OCFS2_SB(ino
72
73 if (fp) {
74 ocfs2_simple_drop_lockres(osb,
75 ocfs2_lock_res_free(&fp->fp_fl
76 kfree(fp);
77 file->private_data = NULL;
78 }
79 }
80
81 static int ocfs2_file_open(struct inode *inode
82 {
83 int status;
84 int mode = file->f_flags;
85 struct ocfs2_inode_info *oi = OCFS2_I(
86
87 trace_ocfs2_file_open(inode, file, fil
88 (unsigned long l
89 file->f_path.den
90 file->f_path.den
91
92 if (file->f_mode & FMODE_WRITE) {
93 status = dquot_initialize(inod
94 if (status)
95 goto leave;
96 }
97
98 spin_lock(&oi->ip_lock);
99
100 /* Check that the inode hasn't been wi
101 * node. If it hasn't then we're safe
102 * spin lock until our increment of op
103 if (oi->ip_flags & OCFS2_INODE_DELETED
104 spin_unlock(&oi->ip_lock);
105
106 status = -ENOENT;
107 goto leave;
108 }
109
110 if (mode & O_DIRECT)
111 oi->ip_flags |= OCFS2_INODE_OP
112
113 oi->ip_open_count++;
114 spin_unlock(&oi->ip_lock);
115
116 status = ocfs2_init_file_private(inode
117 if (status) {
118 /*
119 * We want to set open count b
120 * open.
121 */
122 spin_lock(&oi->ip_lock);
123 oi->ip_open_count--;
124 spin_unlock(&oi->ip_lock);
125 }
126
127 file->f_mode |= FMODE_NOWAIT;
128
129 leave:
130 return status;
131 }
132
133 static int ocfs2_file_release(struct inode *in
134 {
135 struct ocfs2_inode_info *oi = OCFS2_I(
136
137 spin_lock(&oi->ip_lock);
138 if (!--oi->ip_open_count)
139 oi->ip_flags &= ~OCFS2_INODE_O
140
141 trace_ocfs2_file_release(inode, file,
142 oi->ip_blkno,
143 file->f_path.
144 file->f_path.
145 oi->ip_open_c
146 spin_unlock(&oi->ip_lock);
147
148 ocfs2_free_file_private(inode, file);
149
150 return 0;
151 }
152
153 static int ocfs2_dir_open(struct inode *inode,
154 {
155 return ocfs2_init_file_private(inode,
156 }
157
158 static int ocfs2_dir_release(struct inode *ino
159 {
160 ocfs2_free_file_private(inode, file);
161 return 0;
162 }
163
164 static int ocfs2_sync_file(struct file *file,
165 int datasync)
166 {
167 int err = 0;
168 struct inode *inode = file->f_mapping-
169 struct ocfs2_super *osb = OCFS2_SB(ino
170 struct ocfs2_inode_info *oi = OCFS2_I(
171 journal_t *journal = osb->journal->j_j
172 int ret;
173 tid_t commit_tid;
174 bool needs_barrier = false;
175
176 trace_ocfs2_sync_file(inode, file, fil
177 oi->ip_blkno,
178 file->f_path.den
179 file->f_path.den
180 (unsigned long l
181
182 if (ocfs2_is_hard_readonly(osb) || ocf
183 return -EROFS;
184
185 err = file_write_and_wait_range(file,
186 if (err)
187 return err;
188
189 commit_tid = datasync ? oi->i_datasync
190 if (journal->j_flags & JBD2_BARRIER &&
191 !jbd2_trans_will_send_data_barrier
192 needs_barrier = true;
193 err = jbd2_complete_transaction(journa
194 if (needs_barrier) {
195 ret = blkdev_issue_flush(inode
196 if (!err)
197 err = ret;
198 }
199
200 if (err)
201 mlog_errno(err);
202
203 return (err < 0) ? -EIO : 0;
204 }
205
206 int ocfs2_should_update_atime(struct inode *in
207 struct vfsmount
208 {
209 struct timespec64 now;
210 struct ocfs2_super *osb = OCFS2_SB(ino
211
212 if (ocfs2_is_hard_readonly(osb) || ocf
213 return 0;
214
215 if ((inode->i_flags & S_NOATIME) ||
216 ((inode->i_sb->s_flags & SB_NODIRA
217 return 0;
218
219 /*
220 * We can be called with no vfsmnt str
221 * sometimes do this.
222 *
223 * Note that our action here is differ
224 * if we can't tell whether this is a
225 * don't know whether to trust the val
226 */
227 if (vfsmnt == NULL)
228 return 0;
229
230 if ((vfsmnt->mnt_flags & MNT_NOATIME)
231 ((vfsmnt->mnt_flags & MNT_NODIRATI
232 return 0;
233
234 if (vfsmnt->mnt_flags & MNT_RELATIME)
235 struct timespec64 ctime = inod
236 struct timespec64 atime = inod
237 struct timespec64 mtime = inod
238
239 if ((timespec64_compare(&atime
240 (timespec64_compare(&atime
241 return 1;
242
243 return 0;
244 }
245
246 now = current_time(inode);
247 if ((now.tv_sec - inode_get_atime_sec(
248 return 0;
249 else
250 return 1;
251 }
252
253 int ocfs2_update_inode_atime(struct inode *ino
254 struct buffer_hea
255 {
256 int ret;
257 struct ocfs2_super *osb = OCFS2_SB(ino
258 handle_t *handle;
259 struct ocfs2_dinode *di = (struct ocfs
260
261 handle = ocfs2_start_trans(osb, OCFS2_
262 if (IS_ERR(handle)) {
263 ret = PTR_ERR(handle);
264 mlog_errno(ret);
265 goto out;
266 }
267
268 ret = ocfs2_journal_access_di(handle,
269 OCFS2_JO
270 if (ret) {
271 mlog_errno(ret);
272 goto out_commit;
273 }
274
275 /*
276 * Don't use ocfs2_mark_inode_dirty()
277 * have i_rwsem to guard against concu
278 * inode fields.
279 */
280 inode_set_atime_to_ts(inode, current_t
281 di->i_atime = cpu_to_le64(inode_get_at
282 di->i_atime_nsec = cpu_to_le32(inode_g
283 ocfs2_update_inode_fsync_trans(handle,
284 ocfs2_journal_dirty(handle, bh);
285
286 out_commit:
287 ocfs2_commit_trans(osb, handle);
288 out:
289 return ret;
290 }
291
292 int ocfs2_set_inode_size(handle_t *handle,
293 struct inode *
294 struct buffer_
295 u64 new_i_size
296 {
297 int status;
298
299 i_size_write(inode, new_i_size);
300 inode->i_blocks = ocfs2_inode_sector_c
301 inode_set_mtime_to_ts(inode, inode_set
302
303 status = ocfs2_mark_inode_dirty(handle
304 if (status < 0) {
305 mlog_errno(status);
306 goto bail;
307 }
308
309 bail:
310 return status;
311 }
312
313 int ocfs2_simple_size_update(struct inode *ino
314 struct buffer_hea
315 u64 new_i_size)
316 {
317 int ret;
318 struct ocfs2_super *osb = OCFS2_SB(ino
319 handle_t *handle = NULL;
320
321 handle = ocfs2_start_trans(osb, OCFS2_
322 if (IS_ERR(handle)) {
323 ret = PTR_ERR(handle);
324 mlog_errno(ret);
325 goto out;
326 }
327
328 ret = ocfs2_set_inode_size(handle, ino
329 new_i_size)
330 if (ret < 0)
331 mlog_errno(ret);
332
333 ocfs2_update_inode_fsync_trans(handle,
334 ocfs2_commit_trans(osb, handle);
335 out:
336 return ret;
337 }
338
339 static int ocfs2_cow_file_pos(struct inode *in
340 struct buffer_he
341 u64 offset)
342 {
343 int status;
344 u32 phys, cpos = offset >> OCFS2_SB(in
345 unsigned int num_clusters = 0;
346 unsigned int ext_flags = 0;
347
348 /*
349 * If the new offset is aligned to the
350 * no space for ocfs2_zero_range_for_t
351 * CoW either.
352 */
353 if ((offset & (OCFS2_SB(inode->i_sb)->
354 return 0;
355
356 status = ocfs2_get_clusters(inode, cpo
357 &num_clust
358 if (status) {
359 mlog_errno(status);
360 goto out;
361 }
362
363 if (!(ext_flags & OCFS2_EXT_REFCOUNTED
364 goto out;
365
366 return ocfs2_refcount_cow(inode, fe_bh
367
368 out:
369 return status;
370 }
371
372 static int ocfs2_orphan_for_truncate(struct oc
373 struct in
374 struct bu
375 u64 new_i
376 {
377 int status;
378 handle_t *handle;
379 struct ocfs2_dinode *di;
380 u64 cluster_bytes;
381
382 /*
383 * We need to CoW the cluster contains
384 * since we will call ocfs2_zero_range
385 * write "" from offset to the end of
386 */
387 status = ocfs2_cow_file_pos(inode, fe_
388 if (status) {
389 mlog_errno(status);
390 return status;
391 }
392
393 /* TODO: This needs to actually orphan
394 * transaction. */
395
396 handle = ocfs2_start_trans(osb, OCFS2_
397 if (IS_ERR(handle)) {
398 status = PTR_ERR(handle);
399 mlog_errno(status);
400 goto out;
401 }
402
403 status = ocfs2_journal_access_di(handl
404 OCFS2
405 if (status < 0) {
406 mlog_errno(status);
407 goto out_commit;
408 }
409
410 /*
411 * Do this before setting i_size.
412 */
413 cluster_bytes = ocfs2_align_bytes_to_c
414 status = ocfs2_zero_range_for_truncate
415
416 if (status) {
417 mlog_errno(status);
418 goto out_commit;
419 }
420
421 i_size_write(inode, new_i_size);
422 inode_set_mtime_to_ts(inode, inode_set
423
424 di = (struct ocfs2_dinode *) fe_bh->b_
425 di->i_size = cpu_to_le64(new_i_size);
426 di->i_ctime = di->i_mtime = cpu_to_le6
427 di->i_ctime_nsec = di->i_mtime_nsec =
428 ocfs2_update_inode_fsync_trans(handle,
429
430 ocfs2_journal_dirty(handle, fe_bh);
431
432 out_commit:
433 ocfs2_commit_trans(osb, handle);
434 out:
435 return status;
436 }
437
438 int ocfs2_truncate_file(struct inode *inode,
439 struct buffer_h
440 u64 new_i_size)
441 {
442 int status = 0;
443 struct ocfs2_dinode *fe = NULL;
444 struct ocfs2_super *osb = OCFS2_SB(ino
445
446 /* We trust di_bh because it comes fro
447 * already validated it */
448 fe = (struct ocfs2_dinode *) di_bh->b_
449
450 trace_ocfs2_truncate_file((unsigned lo
451 (unsigned lo
452 (unsigned lo
453
454 mlog_bug_on_msg(le64_to_cpu(fe->i_size
455 "Inode %llu, inode i_s
456 "i_size = %llu, i_flag
457 (unsigned long long)OC
458 i_size_read(inode),
459 (unsigned long long)le
460 le32_to_cpu(fe->i_flag
461
462 if (new_i_size > le64_to_cpu(fe->i_siz
463 trace_ocfs2_truncate_file_erro
464 (unsigned long long)le
465 (unsigned long long)ne
466 status = -EINVAL;
467 mlog_errno(status);
468 goto bail;
469 }
470
471 down_write(&OCFS2_I(inode)->ip_alloc_s
472
473 ocfs2_resv_discard(&osb->osb_la_resmap
474 &OCFS2_I(inode)->ip
475
476 /*
477 * The inode lock forced other nodes t
478 * pages, which (correctly) happens ev
479 * without allocation change - ocfs2 c
480 * greater than page size, so we have
481 * anyway.
482 */
483
484 if (OCFS2_I(inode)->ip_dyn_features &
485 unmap_mapping_range(inode->i_m
486 new_i_size
487 truncate_inode_pages(inode->i_
488 status = ocfs2_truncate_inline
489
490 if (status)
491 mlog_errno(status);
492
493 goto bail_unlock_sem;
494 }
495
496 /* alright, we're going to need to do
497 * change. Orphan the inode so that re
498 * truncate if necessary. This does th
499 * i_size. */
500 status = ocfs2_orphan_for_truncate(osb
501 if (status < 0) {
502 mlog_errno(status);
503 goto bail_unlock_sem;
504 }
505
506 unmap_mapping_range(inode->i_mapping,
507 truncate_inode_pages(inode->i_mapping,
508
509 status = ocfs2_commit_truncate(osb, in
510 if (status < 0) {
511 mlog_errno(status);
512 goto bail_unlock_sem;
513 }
514
515 /* TODO: orphan dir cleanup here. */
516 bail_unlock_sem:
517 up_write(&OCFS2_I(inode)->ip_alloc_sem
518
519 bail:
520 if (!status && OCFS2_I(inode)->ip_clus
521 status = ocfs2_try_remove_refc
522
523 return status;
524 }
525
526 /*
527 * extend file allocation only here.
528 * we'll update all the disk stuff, and oip->a
529 *
530 * expect stuff to be locked, a transaction st
531 * metadata reservations in the contexts.
532 *
533 * Will return -EAGAIN, and a reason if a rest
534 * If passed in, *reason will always be set, e
535 */
536 int ocfs2_add_inode_data(struct ocfs2_super *o
537 struct inode *inode,
538 u32 *logical_offset,
539 u32 clusters_to_add,
540 int mark_unwritten,
541 struct buffer_head *f
542 handle_t *handle,
543 struct ocfs2_alloc_co
544 struct ocfs2_alloc_co
545 enum ocfs2_alloc_rest
546 {
547 struct ocfs2_extent_tree et;
548
549 ocfs2_init_dinode_extent_tree(&et, INO
550 return ocfs2_add_clusters_in_btree(han
551 clu
552 dat
553 }
554
555 static int ocfs2_extend_allocation(struct inod
556 u32 cluster
557 {
558 int status = 0;
559 int restart_func = 0;
560 int credits;
561 u32 prev_clusters;
562 struct buffer_head *bh = NULL;
563 struct ocfs2_dinode *fe = NULL;
564 handle_t *handle = NULL;
565 struct ocfs2_alloc_context *data_ac =
566 struct ocfs2_alloc_context *meta_ac =
567 enum ocfs2_alloc_restarted why = RESTA
568 struct ocfs2_super *osb = OCFS2_SB(ino
569 struct ocfs2_extent_tree et;
570 int did_quota = 0;
571
572 /*
573 * Unwritten extent only exists for fi
574 * support holes.
575 */
576 BUG_ON(mark_unwritten && !ocfs2_sparse
577
578 status = ocfs2_read_inode_block(inode,
579 if (status < 0) {
580 mlog_errno(status);
581 goto leave;
582 }
583 fe = (struct ocfs2_dinode *) bh->b_dat
584
585 restart_all:
586 BUG_ON(le32_to_cpu(fe->i_clusters) !=
587
588 ocfs2_init_dinode_extent_tree(&et, INO
589 status = ocfs2_lock_allocators(inode,
590 &data_a
591 if (status) {
592 mlog_errno(status);
593 goto leave;
594 }
595
596 credits = ocfs2_calc_extend_credits(os
597 handle = ocfs2_start_trans(osb, credit
598 if (IS_ERR(handle)) {
599 status = PTR_ERR(handle);
600 handle = NULL;
601 mlog_errno(status);
602 goto leave;
603 }
604
605 restarted_transaction:
606 trace_ocfs2_extend_allocation(
607 (unsigned long long)OCFS2_I(in
608 (unsigned long long)i_size_rea
609 le32_to_cpu(fe->i_clusters), c
610 why, restart_func);
611
612 status = dquot_alloc_space_nodirty(ino
613 ocfs2_clusters_to_byte
614 if (status)
615 goto leave;
616 did_quota = 1;
617
618 /* reserve a write to the file entry e
619 * run out of credits in the allocatio
620 * update i_size. */
621 status = ocfs2_journal_access_di(handl
622 OCFS2
623 if (status < 0) {
624 mlog_errno(status);
625 goto leave;
626 }
627
628 prev_clusters = OCFS2_I(inode)->ip_clu
629
630 status = ocfs2_add_inode_data(osb,
631 inode,
632 &logical
633 clusters
634 mark_unw
635 bh,
636 handle,
637 data_ac,
638 meta_ac,
639 &why);
640 if ((status < 0) && (status != -EAGAIN
641 if (status != -ENOSPC)
642 mlog_errno(status);
643 goto leave;
644 }
645 ocfs2_update_inode_fsync_trans(handle,
646 ocfs2_journal_dirty(handle, bh);
647
648 spin_lock(&OCFS2_I(inode)->ip_lock);
649 clusters_to_add -= (OCFS2_I(inode)->ip
650 spin_unlock(&OCFS2_I(inode)->ip_lock);
651 /* Release unused quota reservation */
652 dquot_free_space(inode,
653 ocfs2_clusters_to_byte
654 did_quota = 0;
655
656 if (why != RESTART_NONE && clusters_to
657 if (why == RESTART_META) {
658 restart_func = 1;
659 status = 0;
660 } else {
661 BUG_ON(why != RESTART_
662
663 status = ocfs2_allocat
664 if (status < 0) {
665 /* handle stil
666 * this point.
667 status = -ENOM
668 mlog_errno(sta
669 goto leave;
670 }
671 goto restarted_transac
672 }
673 }
674
675 trace_ocfs2_extend_allocation_end(OCFS
676 le32_to_cpu(fe->i_clusters),
677 (unsigned long long)le64_to_cpu(f
678 OCFS2_I(inode)->ip_clusters,
679 (unsigned long long)i_size_read(i
680
681 leave:
682 if (status < 0 && did_quota)
683 dquot_free_space(inode,
684 ocfs2_clusters_to_byte
685 if (handle) {
686 ocfs2_commit_trans(osb, handle
687 handle = NULL;
688 }
689 if (data_ac) {
690 ocfs2_free_alloc_context(data_
691 data_ac = NULL;
692 }
693 if (meta_ac) {
694 ocfs2_free_alloc_context(meta_
695 meta_ac = NULL;
696 }
697 if ((!status) && restart_func) {
698 restart_func = 0;
699 goto restart_all;
700 }
701 brelse(bh);
702 bh = NULL;
703
704 return status;
705 }
706
707 /*
708 * While a write will already be ordering the
709 * Thus, we need to explicitly order the zeroe
710 */
711 static handle_t *ocfs2_zero_start_ordered_tran
712
713
714
715 {
716 struct ocfs2_super *osb = OCFS2_SB(ino
717 handle_t *handle = NULL;
718 int ret = 0;
719
720 if (!ocfs2_should_order_data(inode))
721 goto out;
722
723 handle = ocfs2_start_trans(osb, OCFS2_
724 if (IS_ERR(handle)) {
725 ret = -ENOMEM;
726 mlog_errno(ret);
727 goto out;
728 }
729
730 ret = ocfs2_jbd2_inode_add_write(handl
731 if (ret < 0) {
732 mlog_errno(ret);
733 goto out;
734 }
735
736 ret = ocfs2_journal_access_di(handle,
737 OCFS2_JO
738 if (ret)
739 mlog_errno(ret);
740 ocfs2_update_inode_fsync_trans(handle,
741
742 out:
743 if (ret) {
744 if (!IS_ERR(handle))
745 ocfs2_commit_trans(osb
746 handle = ERR_PTR(ret);
747 }
748 return handle;
749 }
750
751 /* Some parts of this taken from generic_cont_
752 * to be too fragile to do exactly what we nee
753 * worry about recursive locking in ->write_be
754 static int ocfs2_write_zero_page(struct inode
755 u64 abs_to, s
756 {
757 struct address_space *mapping = inode-
758 struct folio *folio;
759 unsigned long index = abs_from >> PAGE
760 handle_t *handle;
761 int ret = 0;
762 unsigned zero_from, zero_to, block_sta
763 struct ocfs2_dinode *di = (struct ocfs
764
765 BUG_ON(abs_from >= abs_to);
766 BUG_ON(abs_to > (((u64)index + 1) << P
767 BUG_ON(abs_from & (inode->i_blkbits -
768
769 handle = ocfs2_zero_start_ordered_tran
770
771
772 if (IS_ERR(handle)) {
773 ret = PTR_ERR(handle);
774 goto out;
775 }
776
777 folio = __filemap_get_folio(mapping, i
778 FGP_LOCK | FGP_ACCESSE
779 if (IS_ERR(folio)) {
780 ret = PTR_ERR(folio);
781 mlog_errno(ret);
782 goto out_commit_trans;
783 }
784
785 /* Get the offsets within the page tha
786 zero_from = abs_from & (PAGE_SIZE - 1)
787 zero_to = abs_to & (PAGE_SIZE - 1);
788 if (!zero_to)
789 zero_to = PAGE_SIZE;
790
791 trace_ocfs2_write_zero_page(
792 (unsigned long long)OC
793 (unsigned long long)ab
794 (unsigned long long)ab
795 index, zero_from, zero
796
797 /* We know that zero_from is block ali
798 for (block_start = zero_from; block_st
799 block_start = block_end) {
800 block_end = block_start + i_bl
801
802 /*
803 * block_start is block-aligne
804 * __block_write_begin and blo
805 * whole block.
806 */
807 ret = __block_write_begin(foli
808 ocfs
809 if (ret < 0) {
810 mlog_errno(ret);
811 goto out_unlock;
812 }
813
814
815 /* must not update i_size! */
816 block_commit_write(&folio->pag
817 }
818
819 /*
820 * fs-writeback will release the dirty
821 * whose offset are over inode size, t
822 * block_write_full_folio().
823 */
824 i_size_write(inode, abs_to);
825 inode->i_blocks = ocfs2_inode_sector_c
826 di->i_size = cpu_to_le64((u64)i_size_r
827 inode_set_mtime_to_ts(inode, inode_set
828 di->i_mtime = di->i_ctime = cpu_to_le6
829 di->i_ctime_nsec = cpu_to_le32(inode_g
830 di->i_mtime_nsec = di->i_ctime_nsec;
831 if (handle) {
832 ocfs2_journal_dirty(handle, di
833 ocfs2_update_inode_fsync_trans
834 }
835
836 out_unlock:
837 folio_unlock(folio);
838 folio_put(folio);
839 out_commit_trans:
840 if (handle)
841 ocfs2_commit_trans(OCFS2_SB(in
842 out:
843 return ret;
844 }
845
846 /*
847 * Find the next range to zero. We do this in
848 * that's what ocfs2_zero_extend() wants, and
849 * pagecache. We may return multiple extents.
850 *
851 * zero_start and zero_end are ocfs2_zero_exte
852 * needs to be zeroed. range_start and range_
853 * range. A subsequent call should pass the p
854 * zero_start. If range_end is 0, there's not
855 *
856 * Unwritten extents are skipped over. Refcou
857 */
858 static int ocfs2_zero_extend_get_range(struct
859 struct
860 u64 zer
861 u64 *ra
862 {
863 int rc = 0, needs_cow = 0;
864 u32 p_cpos, zero_clusters = 0;
865 u32 zero_cpos =
866 zero_start >> OCFS2_SB(inode->
867 u32 last_cpos = ocfs2_clusters_for_byt
868 unsigned int num_clusters = 0;
869 unsigned int ext_flags = 0;
870
871 while (zero_cpos < last_cpos) {
872 rc = ocfs2_get_clusters(inode,
873 &num_c
874 if (rc) {
875 mlog_errno(rc);
876 goto out;
877 }
878
879 if (p_cpos && !(ext_flags & OC
880 zero_clusters = num_cl
881 if (ext_flags & OCFS2_
882 needs_cow = 1;
883 break;
884 }
885
886 zero_cpos += num_clusters;
887 }
888 if (!zero_clusters) {
889 *range_end = 0;
890 goto out;
891 }
892
893 while ((zero_cpos + zero_clusters) < l
894 rc = ocfs2_get_clusters(inode,
895 &p_cpo
896 &ext_f
897 if (rc) {
898 mlog_errno(rc);
899 goto out;
900 }
901
902 if (!p_cpos || (ext_flags & OC
903 break;
904 if (ext_flags & OCFS2_EXT_REFC
905 needs_cow = 1;
906 zero_clusters += num_clusters;
907 }
908 if ((zero_cpos + zero_clusters) > last
909 zero_clusters = last_cpos - ze
910
911 if (needs_cow) {
912 rc = ocfs2_refcount_cow(inode,
913 zero_c
914 if (rc) {
915 mlog_errno(rc);
916 goto out;
917 }
918 }
919
920 *range_start = ocfs2_clusters_to_bytes
921 *range_end = ocfs2_clusters_to_bytes(i
922 z
923
924 out:
925 return rc;
926 }
927
928 /*
929 * Zero one range returned from ocfs2_zero_ext
930 * has made sure that the entire range needs z
931 */
932 static int ocfs2_zero_extend_range(struct inod
933 u64 range_e
934 {
935 int rc = 0;
936 u64 next_pos;
937 u64 zero_pos = range_start;
938
939 trace_ocfs2_zero_extend_range(
940 (unsigned long long)OC
941 (unsigned long long)ra
942 (unsigned long long)ra
943 BUG_ON(range_start >= range_end);
944
945 while (zero_pos < range_end) {
946 next_pos = (zero_pos & PAGE_MA
947 if (next_pos > range_end)
948 next_pos = range_end;
949 rc = ocfs2_write_zero_page(ino
950 if (rc < 0) {
951 mlog_errno(rc);
952 break;
953 }
954 zero_pos = next_pos;
955
956 /*
957 * Very large extends have the
958 * the cpu for extended period
959 */
960 cond_resched();
961 }
962
963 return rc;
964 }
965
966 int ocfs2_zero_extend(struct inode *inode, str
967 loff_t zero_to_size)
968 {
969 int ret = 0;
970 u64 zero_start, range_start = 0, range
971 struct super_block *sb = inode->i_sb;
972
973 zero_start = ocfs2_align_bytes_to_bloc
974 trace_ocfs2_zero_extend((unsigned long
975 (unsigned long
976 (unsigned long
977 while (zero_start < zero_to_size) {
978 ret = ocfs2_zero_extend_get_ra
979
980
981
982 if (ret) {
983 mlog_errno(ret);
984 break;
985 }
986 if (!range_end)
987 break;
988 /* Trim the ends */
989 if (range_start < zero_start)
990 range_start = zero_sta
991 if (range_end > zero_to_size)
992 range_end = zero_to_si
993
994 ret = ocfs2_zero_extend_range(
995
996 if (ret) {
997 mlog_errno(ret);
998 break;
999 }
1000 zero_start = range_end;
1001 }
1002
1003 return ret;
1004 }
1005
1006 int ocfs2_extend_no_holes(struct inode *inode
1007 u64 new_i_size, u64
1008 {
1009 int ret;
1010 u32 clusters_to_add;
1011 struct ocfs2_inode_info *oi = OCFS2_I
1012
1013 /*
1014 * Only quota files call this without
1015 * refcounted.
1016 */
1017 BUG_ON(!di_bh && ocfs2_is_refcount_in
1018 BUG_ON(!di_bh && !(oi->ip_flags & OCF
1019
1020 clusters_to_add = ocfs2_clusters_for_
1021 if (clusters_to_add < oi->ip_clusters
1022 clusters_to_add = 0;
1023 else
1024 clusters_to_add -= oi->ip_clu
1025
1026 if (clusters_to_add) {
1027 ret = ocfs2_extend_allocation
1028
1029 if (ret) {
1030 mlog_errno(ret);
1031 goto out;
1032 }
1033 }
1034
1035 /*
1036 * Call this even if we don't add any
1037 * still need to zero the area betwee
1038 * new i_size.
1039 */
1040 ret = ocfs2_zero_extend(inode, di_bh,
1041 if (ret < 0)
1042 mlog_errno(ret);
1043
1044 out:
1045 return ret;
1046 }
1047
1048 static int ocfs2_extend_file(struct inode *in
1049 struct buffer_he
1050 u64 new_i_size)
1051 {
1052 int ret = 0;
1053 struct ocfs2_inode_info *oi = OCFS2_I
1054
1055 BUG_ON(!di_bh);
1056
1057 /* setattr sometimes calls us like th
1058 if (new_i_size == 0)
1059 goto out;
1060
1061 if (i_size_read(inode) == new_i_size)
1062 goto out;
1063 BUG_ON(new_i_size < i_size_read(inode
1064
1065 /*
1066 * The alloc sem blocks people in rea
1067 * allocation until we're done changi
1068 * i_rwsem to block other extend/trun
1069 * here. We even have to hold it for
1070 * might be some tail zeroing.
1071 */
1072 down_write(&oi->ip_alloc_sem);
1073
1074 if (oi->ip_dyn_features & OCFS2_INLIN
1075 /*
1076 * We can optimize small exte
1077 * inline data.
1078 */
1079 if (ocfs2_size_fits_inline_da
1080 up_write(&oi->ip_allo
1081 goto out_update_size;
1082 }
1083
1084 ret = ocfs2_convert_inline_da
1085 if (ret) {
1086 up_write(&oi->ip_allo
1087 mlog_errno(ret);
1088 goto out;
1089 }
1090 }
1091
1092 if (ocfs2_sparse_alloc(OCFS2_SB(inode
1093 ret = ocfs2_zero_extend(inode
1094 else
1095 ret = ocfs2_extend_no_holes(i
1096 n
1097
1098 up_write(&oi->ip_alloc_sem);
1099
1100 if (ret < 0) {
1101 mlog_errno(ret);
1102 goto out;
1103 }
1104
1105 out_update_size:
1106 ret = ocfs2_simple_size_update(inode,
1107 if (ret < 0)
1108 mlog_errno(ret);
1109
1110 out:
1111 return ret;
1112 }
1113
1114 int ocfs2_setattr(struct mnt_idmap *idmap, st
1115 struct iattr *attr)
1116 {
1117 int status = 0, size_change;
1118 int inode_locked = 0;
1119 struct inode *inode = d_inode(dentry)
1120 struct super_block *sb = inode->i_sb;
1121 struct ocfs2_super *osb = OCFS2_SB(sb
1122 struct buffer_head *bh = NULL;
1123 handle_t *handle = NULL;
1124 struct dquot *transfer_to[MAXQUOTAS]
1125 int qtype;
1126 int had_lock;
1127 struct ocfs2_lock_holder oh;
1128
1129 trace_ocfs2_setattr(inode, dentry,
1130 (unsigned long lo
1131 dentry->d_name.le
1132 attr->ia_valid,
1133 attr->ia_vali
1134 attr->ia_vali
1135 from_
1136 attr->ia_vali
1137 from_
1138
1139 /* ensuring we don't even attempt to
1140 if (S_ISLNK(inode->i_mode))
1141 attr->ia_valid &= ~ATTR_SIZE;
1142
1143 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_
1144 | ATTR_GID | ATTR_
1145 if (!(attr->ia_valid & OCFS2_VALID_AT
1146 return 0;
1147
1148 status = setattr_prepare(&nop_mnt_idm
1149 if (status)
1150 return status;
1151
1152 if (is_quota_modification(&nop_mnt_id
1153 status = dquot_initialize(ino
1154 if (status)
1155 return status;
1156 }
1157 size_change = S_ISREG(inode->i_mode)
1158 if (size_change) {
1159 /*
1160 * Here we should wait dio to
1161 * to avoid a deadlock betwee
1162 * ocfs2_dio_end_io_write()
1163 */
1164 inode_dio_wait(inode);
1165
1166 status = ocfs2_rw_lock(inode,
1167 if (status < 0) {
1168 mlog_errno(status);
1169 goto bail;
1170 }
1171 }
1172
1173 had_lock = ocfs2_inode_lock_tracker(i
1174 if (had_lock < 0) {
1175 status = had_lock;
1176 goto bail_unlock_rw;
1177 } else if (had_lock) {
1178 /*
1179 * As far as we know, ocfs2_s
1180 * VFS entry point in the cal
1181 * locking issue.
1182 *
1183 * For instance:
1184 * chmod_common()
1185 * notify_change()
1186 * ocfs2_setattr()
1187 * posix_acl_chmod()
1188 * ocfs2_iop_get_acl()
1189 *
1190 * But, we're not 100% sure i
1191 * ordering of the VFS entry
1192 * of our control. So, we'd b
1193 * catch the other cases of r
1194 */
1195 mlog(ML_ERROR, "Another case
1196 dump_stack();
1197 }
1198 inode_locked = 1;
1199
1200 if (size_change) {
1201 status = inode_newsize_ok(ino
1202 if (status)
1203 goto bail_unlock;
1204
1205 if (i_size_read(inode) >= att
1206 if (ocfs2_should_orde
1207 status = ocfs
1208
1209 if (status)
1210 goto
1211 }
1212 status = ocfs2_trunca
1213 } else
1214 status = ocfs2_extend
1215 if (status < 0) {
1216 if (status != -ENOSPC
1217 mlog_errno(st
1218 status = -ENOSPC;
1219 goto bail_unlock;
1220 }
1221 }
1222
1223 if ((attr->ia_valid & ATTR_UID && !ui
1224 (attr->ia_valid & ATTR_GID && !gi
1225 /*
1226 * Gather pointers to quota s
1227 * freeing of quota structure
1228 * dquot_transfer() where we
1229 */
1230 if (attr->ia_valid & ATTR_UID
1231 && OCFS2_HAS_RO_COMPAT_FE
1232 OCFS2_FEATURE_RO_COMPAT_U
1233 transfer_to[USRQUOTA]
1234 if (IS_ERR(transfer_t
1235 status = PTR_
1236 transfer_to[U
1237 goto bail_unl
1238 }
1239 }
1240 if (attr->ia_valid & ATTR_GID
1241 && OCFS2_HAS_RO_COMPAT_FE
1242 OCFS2_FEATURE_RO_COMPAT_G
1243 transfer_to[GRPQUOTA]
1244 if (IS_ERR(transfer_t
1245 status = PTR_
1246 transfer_to[G
1247 goto bail_unl
1248 }
1249 }
1250 down_write(&OCFS2_I(inode)->i
1251 handle = ocfs2_start_trans(os
1252 2
1253 if (IS_ERR(handle)) {
1254 status = PTR_ERR(hand
1255 mlog_errno(status);
1256 goto bail_unlock_allo
1257 }
1258 status = __dquot_transfer(ino
1259 if (status < 0)
1260 goto bail_commit;
1261 } else {
1262 down_write(&OCFS2_I(inode)->i
1263 handle = ocfs2_start_trans(os
1264 if (IS_ERR(handle)) {
1265 status = PTR_ERR(hand
1266 mlog_errno(status);
1267 goto bail_unlock_allo
1268 }
1269 }
1270
1271 setattr_copy(&nop_mnt_idmap, inode, a
1272 mark_inode_dirty(inode);
1273
1274 status = ocfs2_mark_inode_dirty(handl
1275 if (status < 0)
1276 mlog_errno(status);
1277
1278 bail_commit:
1279 ocfs2_commit_trans(osb, handle);
1280 bail_unlock_alloc:
1281 up_write(&OCFS2_I(inode)->ip_alloc_se
1282 bail_unlock:
1283 if (status && inode_locked) {
1284 ocfs2_inode_unlock_tracker(in
1285 inode_locked = 0;
1286 }
1287 bail_unlock_rw:
1288 if (size_change)
1289 ocfs2_rw_unlock(inode, 1);
1290 bail:
1291
1292 /* Release quota pointers in case we
1293 for (qtype = 0; qtype < OCFS2_MAXQUOT
1294 dqput(transfer_to[qtype]);
1295
1296 if (!status && attr->ia_valid & ATTR_
1297 status = ocfs2_acl_chmod(inod
1298 if (status < 0)
1299 mlog_errno(status);
1300 }
1301 if (inode_locked)
1302 ocfs2_inode_unlock_tracker(in
1303
1304 brelse(bh);
1305 return status;
1306 }
1307
1308 int ocfs2_getattr(struct mnt_idmap *idmap, co
1309 struct kstat *stat, u32 req
1310 {
1311 struct inode *inode = d_inode(path->d
1312 struct super_block *sb = path->dentry
1313 struct ocfs2_super *osb = sb->s_fs_in
1314 int err;
1315
1316 err = ocfs2_inode_revalidate(path->de
1317 if (err) {
1318 if (err != -ENOENT)
1319 mlog_errno(err);
1320 goto bail;
1321 }
1322
1323 generic_fillattr(&nop_mnt_idmap, requ
1324 /*
1325 * If there is inline data in the ino
1326 * have data blocks allocated (it may
1327 * Report at least one sector for suc
1328 * others don't incorrectly think the
1329 */
1330 if (unlikely(OCFS2_I(inode)->ip_dyn_f
1331 stat->blocks += (stat->size +
1332
1333 /* We set the blksize from the cluste
1334 stat->blksize = osb->s_clustersize;
1335
1336 bail:
1337 return err;
1338 }
1339
1340 int ocfs2_permission(struct mnt_idmap *idmap,
1341 int mask)
1342 {
1343 int ret, had_lock;
1344 struct ocfs2_lock_holder oh;
1345
1346 if (mask & MAY_NOT_BLOCK)
1347 return -ECHILD;
1348
1349 had_lock = ocfs2_inode_lock_tracker(i
1350 if (had_lock < 0) {
1351 ret = had_lock;
1352 goto out;
1353 } else if (had_lock) {
1354 /* See comments in ocfs2_seta
1355 * The call chain of this cas
1356 * do_sys_open()
1357 * may_open()
1358 * inode_permission()
1359 * ocfs2_permission()
1360 * ocfs2_iop_get_acl()
1361 */
1362 mlog(ML_ERROR, "Another case
1363 dump_stack();
1364 }
1365
1366 ret = generic_permission(&nop_mnt_idm
1367
1368 ocfs2_inode_unlock_tracker(inode, 0,
1369 out:
1370 return ret;
1371 }
1372
1373 static int __ocfs2_write_remove_suid(struct i
1374 struct b
1375 {
1376 int ret;
1377 handle_t *handle;
1378 struct ocfs2_super *osb = OCFS2_SB(in
1379 struct ocfs2_dinode *di;
1380
1381 trace_ocfs2_write_remove_suid(
1382 (unsigned long long)O
1383 inode->i_mode);
1384
1385 handle = ocfs2_start_trans(osb, OCFS2
1386 if (IS_ERR(handle)) {
1387 ret = PTR_ERR(handle);
1388 mlog_errno(ret);
1389 goto out;
1390 }
1391
1392 ret = ocfs2_journal_access_di(handle,
1393 OCFS2_J
1394 if (ret < 0) {
1395 mlog_errno(ret);
1396 goto out_trans;
1397 }
1398
1399 inode->i_mode &= ~S_ISUID;
1400 if ((inode->i_mode & S_ISGID) && (ino
1401 inode->i_mode &= ~S_ISGID;
1402
1403 di = (struct ocfs2_dinode *) bh->b_da
1404 di->i_mode = cpu_to_le16(inode->i_mod
1405 ocfs2_update_inode_fsync_trans(handle
1406
1407 ocfs2_journal_dirty(handle, bh);
1408
1409 out_trans:
1410 ocfs2_commit_trans(osb, handle);
1411 out:
1412 return ret;
1413 }
1414
1415 static int ocfs2_write_remove_suid(struct ino
1416 {
1417 int ret;
1418 struct buffer_head *bh = NULL;
1419
1420 ret = ocfs2_read_inode_block(inode, &
1421 if (ret < 0) {
1422 mlog_errno(ret);
1423 goto out;
1424 }
1425
1426 ret = __ocfs2_write_remove_suid(inod
1427 out:
1428 brelse(bh);
1429 return ret;
1430 }
1431
1432 /*
1433 * Allocate enough extents to cover the regio
1434 * start for len bytes. Existing extents are
1435 * added are marked as "unwritten".
1436 */
1437 static int ocfs2_allocate_unwritten_extents(s
1438 u
1439 {
1440 int ret;
1441 u32 cpos, phys_cpos, clusters, alloc_
1442 u64 end = start + len;
1443 struct buffer_head *di_bh = NULL;
1444
1445 if (OCFS2_I(inode)->ip_dyn_features &
1446 ret = ocfs2_read_inode_block(
1447 if (ret) {
1448 mlog_errno(ret);
1449 goto out;
1450 }
1451
1452 /*
1453 * Nothing to do if the reque
1454 * fits within the inode.
1455 */
1456 if (ocfs2_size_fits_inline_da
1457 goto out;
1458
1459 ret = ocfs2_convert_inline_da
1460 if (ret) {
1461 mlog_errno(ret);
1462 goto out;
1463 }
1464 }
1465
1466 /*
1467 * We consider both start and len to
1468 */
1469 cpos = start >> OCFS2_SB(inode->i_sb)
1470 clusters = ocfs2_clusters_for_bytes(i
1471 clusters -= cpos;
1472
1473 while (clusters) {
1474 ret = ocfs2_get_clusters(inod
1475 &all
1476 if (ret) {
1477 mlog_errno(ret);
1478 goto out;
1479 }
1480
1481 /*
1482 * Hole or existing extent le
1483 * cap it to our own allocati
1484 */
1485 if (alloc_size > clusters)
1486 alloc_size = clusters
1487
1488 if (phys_cpos) {
1489 /*
1490 * We already have an
1491 * region so we can s
1492 */
1493 goto next;
1494 }
1495
1496 ret = ocfs2_extend_allocation
1497 if (ret) {
1498 if (ret != -ENOSPC)
1499 mlog_errno(re
1500 goto out;
1501 }
1502
1503 next:
1504 cpos += alloc_size;
1505 clusters -= alloc_size;
1506 }
1507
1508 ret = 0;
1509 out:
1510
1511 brelse(di_bh);
1512 return ret;
1513 }
1514
1515 /*
1516 * Truncate a byte range, avoiding pages with
1517 * preserves those pages for the zeroing code
1518 */
1519 static void ocfs2_truncate_cluster_pages(stru
1520 u64
1521 {
1522 struct ocfs2_super *osb = OCFS2_SB(in
1523 loff_t start, end;
1524 struct address_space *mapping = inode
1525
1526 start = (loff_t)ocfs2_align_bytes_to_
1527 end = byte_start + byte_len;
1528 end = end & ~(osb->s_clustersize - 1)
1529
1530 if (start < end) {
1531 unmap_mapping_range(mapping,
1532 truncate_inode_pages_range(ma
1533 }
1534 }
1535
1536 /*
1537 * zero out partial blocks of one cluster.
1538 *
1539 * start: file offset where zero starts, will
1540 * len: it will be trimmed to the end of curr
1541 * is bigger than it.
1542 */
1543 static int ocfs2_zeroout_partial_cluster(stru
1544 u64 s
1545 {
1546 int ret;
1547 u64 start_block, end_block, nr_blocks
1548 u64 p_block, offset;
1549 u32 cluster, p_cluster, nr_clusters;
1550 struct super_block *sb = inode->i_sb;
1551 u64 end = ocfs2_align_bytes_to_cluste
1552
1553 if (start + len < end)
1554 end = start + len;
1555
1556 start_block = ocfs2_blocks_for_bytes(
1557 end_block = ocfs2_blocks_for_bytes(sb
1558 nr_blocks = end_block - start_block;
1559 if (!nr_blocks)
1560 return 0;
1561
1562 cluster = ocfs2_bytes_to_clusters(sb,
1563 ret = ocfs2_get_clusters(inode, clust
1564 &nr_clusters,
1565 if (ret)
1566 return ret;
1567 if (!p_cluster)
1568 return 0;
1569
1570 offset = start_block - ocfs2_clusters
1571 p_block = ocfs2_clusters_to_blocks(sb
1572 return sb_issue_zeroout(sb, p_block,
1573 }
1574
1575 static int ocfs2_zero_partial_clusters(struct
1576 u64 st
1577 {
1578 int ret = 0;
1579 u64 tmpend = 0;
1580 u64 end = start + len;
1581 struct ocfs2_super *osb = OCFS2_SB(in
1582 unsigned int csize = osb->s_clustersi
1583 handle_t *handle;
1584 loff_t isize = i_size_read(inode);
1585
1586 /*
1587 * The "start" and "end" values are N
1588 * the range whose allocation is bein
1589 * is what the user passed in with th
1590 * partial clusters here. There's no
1591 * physical allocation - the zeroing
1592 */
1593 trace_ocfs2_zero_partial_clusters(
1594 (unsigned long long)OCFS2_I(i
1595 (unsigned long long)start, (u
1596
1597 /*
1598 * If both edges are on a cluster bou
1599 * zeroing required as the region is
1600 * be truncated.
1601 */
1602 if ((start & (csize - 1)) == 0 && (en
1603 goto out;
1604
1605 /* No page cache for EOF blocks, issu
1606 if (end > isize) {
1607 /*
1608 * zeroout eof blocks in last
1609 * "isize" even "start" > "is
1610 * complicated to zeroout jus
1611 * may be not aligned with bl
1612 * would be required to do th
1613 * write is not supported.
1614 */
1615 ret = ocfs2_zeroout_partial_c
1616 end -
1617 if (ret) {
1618 mlog_errno(ret);
1619 goto out;
1620 }
1621 if (start >= isize)
1622 goto out;
1623 end = isize;
1624 }
1625 handle = ocfs2_start_trans(osb, OCFS2
1626 if (IS_ERR(handle)) {
1627 ret = PTR_ERR(handle);
1628 mlog_errno(ret);
1629 goto out;
1630 }
1631
1632 /*
1633 * If start is on a cluster boundary
1634 * cluster, we have not COWed the clu
1635 * end is also within the same cluste
1636 * first call to ocfs2_zero_range_for
1637 * to the next one.
1638 */
1639 if ((start & (csize - 1)) != 0) {
1640 /*
1641 * We want to get the byte of
1642 * cluster.
1643 */
1644 tmpend = (u64)osb->s_clusters
1645 (start & ~(osb->s_clu
1646 if (tmpend > end)
1647 tmpend = end;
1648
1649 trace_ocfs2_zero_partial_clus
1650 (unsigned long long)s
1651 (unsigned long long)t
1652
1653 ret = ocfs2_zero_range_for_tr
1654
1655 if (ret)
1656 mlog_errno(ret);
1657 }
1658
1659 if (tmpend < end) {
1660 /*
1661 * This may make start and en
1662 * code will skip any work in
1663 * need to catch it up here.
1664 */
1665 start = end & ~(osb->s_cluste
1666
1667 trace_ocfs2_zero_partial_clus
1668 (unsigned long long)s
1669
1670 ret = ocfs2_zero_range_for_tr
1671 if (ret)
1672 mlog_errno(ret);
1673 }
1674 ocfs2_update_inode_fsync_trans(handle
1675
1676 ocfs2_commit_trans(osb, handle);
1677 out:
1678 return ret;
1679 }
1680
1681 static int ocfs2_find_rec(struct ocfs2_extent
1682 {
1683 int i;
1684 struct ocfs2_extent_rec *rec = NULL;
1685
1686 for (i = le16_to_cpu(el->l_next_free_
1687
1688 rec = &el->l_recs[i];
1689
1690 if (le32_to_cpu(rec->e_cpos)
1691 break;
1692 }
1693
1694 return i;
1695 }
1696
1697 /*
1698 * Helper to calculate the punching pos and l
1699 * following three cases in order:
1700 *
1701 * - remove the entire record
1702 * - remove a partial record
1703 * - no record needs to be removed (hole-punc
1704 */
1705 static void ocfs2_calc_trunc_pos(struct inode
1706 struct ocfs2
1707 struct ocfs2
1708 u32 trunc_st
1709 u32 *trunc_l
1710 u64 *blkno,
1711 {
1712 int ret = 0;
1713 u32 coff, range;
1714
1715 range = le32_to_cpu(rec->e_cpos) + oc
1716
1717 if (le32_to_cpu(rec->e_cpos) >= trunc
1718 /*
1719 * remove an entire extent re
1720 */
1721 *trunc_cpos = le32_to_cpu(rec
1722 /*
1723 * Skip holes if any.
1724 */
1725 if (range < *trunc_end)
1726 *trunc_end = range;
1727 *trunc_len = *trunc_end - le3
1728 *blkno = le64_to_cpu(rec->e_b
1729 *trunc_end = le32_to_cpu(rec-
1730 } else if (range > trunc_start) {
1731 /*
1732 * remove a partial extent re
1733 * removing the last extent r
1734 */
1735 *trunc_cpos = trunc_start;
1736 /*
1737 * skip hole if any.
1738 */
1739 if (range < *trunc_end)
1740 *trunc_end = range;
1741 *trunc_len = *trunc_end - tru
1742 coff = trunc_start - le32_to_
1743 *blkno = le64_to_cpu(rec->e_b
1744 ocfs2_cluster
1745 *trunc_end = trunc_start;
1746 } else {
1747 /*
1748 * It may have two following
1749 *
1750 * - last record has been rem
1751 * - trunc_start was within a
1752 *
1753 * both two cases mean the co
1754 */
1755 ret = 1;
1756 }
1757
1758 *done = ret;
1759 }
1760
1761 int ocfs2_remove_inode_range(struct inode *in
1762 struct buffer_he
1763 u64 byte_len)
1764 {
1765 int ret = 0, flags = 0, done = 0, i;
1766 u32 trunc_start, trunc_len, trunc_end
1767 u32 cluster_in_el;
1768 struct ocfs2_super *osb = OCFS2_SB(in
1769 struct ocfs2_cached_dealloc_ctxt deal
1770 struct address_space *mapping = inode
1771 struct ocfs2_extent_tree et;
1772 struct ocfs2_path *path = NULL;
1773 struct ocfs2_extent_list *el = NULL;
1774 struct ocfs2_extent_rec *rec = NULL;
1775 struct ocfs2_dinode *di = (struct ocf
1776 u64 blkno, refcount_loc = le64_to_cpu
1777
1778 ocfs2_init_dinode_extent_tree(&et, IN
1779 ocfs2_init_dealloc_ctxt(&dealloc);
1780
1781 trace_ocfs2_remove_inode_range(
1782 (unsigned long long)O
1783 (unsigned long long)b
1784 (unsigned long long)b
1785
1786 if (byte_len == 0)
1787 return 0;
1788
1789 if (OCFS2_I(inode)->ip_dyn_features &
1790 int id_count = ocfs2_max_inli
1791
1792 if (byte_start > id_count ||
1793 ret = -EINVAL;
1794 mlog_errno(ret);
1795 goto out;
1796 }
1797
1798 ret = ocfs2_truncate_inline(i
1799 b
1800 if (ret) {
1801 mlog_errno(ret);
1802 goto out;
1803 }
1804 /*
1805 * There's no need to get fan
1806 * truncate of an inline-data
1807 * about less than a page her
1808 * in the dinode buffer anywa
1809 */
1810 unmap_mapping_range(mapping,
1811 truncate_inode_pages(mapping,
1812 goto out;
1813 }
1814
1815 /*
1816 * For reflinks, we may need to CoW 2
1817 * partially zero'd later, if hole's
1818 * within one cluster(means is not ex
1819 */
1820
1821 if (ocfs2_is_refcount_inode(inode)) {
1822 ret = ocfs2_cow_file_pos(inod
1823 if (ret) {
1824 mlog_errno(ret);
1825 goto out;
1826 }
1827
1828 ret = ocfs2_cow_file_pos(inod
1829 if (ret) {
1830 mlog_errno(ret);
1831 goto out;
1832 }
1833 }
1834
1835 trunc_start = ocfs2_clusters_for_byte
1836 trunc_end = (byte_start + byte_len) >
1837 cluster_in_el = trunc_end;
1838
1839 ret = ocfs2_zero_partial_clusters(ino
1840 if (ret) {
1841 mlog_errno(ret);
1842 goto out;
1843 }
1844
1845 path = ocfs2_new_path_from_et(&et);
1846 if (!path) {
1847 ret = -ENOMEM;
1848 mlog_errno(ret);
1849 goto out;
1850 }
1851
1852 while (trunc_end > trunc_start) {
1853
1854 ret = ocfs2_find_path(INODE_C
1855 cluster
1856 if (ret) {
1857 mlog_errno(ret);
1858 goto out;
1859 }
1860
1861 el = path_leaf_el(path);
1862
1863 i = ocfs2_find_rec(el, trunc_
1864 /*
1865 * Need to go to previous ext
1866 */
1867 if (i < 0) {
1868 if (path->p_tree_dept
1869 break;
1870
1871 ret = ocfs2_find_cpos
1872
1873
1874 if (ret) {
1875 mlog_errno(re
1876 goto out;
1877 }
1878
1879 /*
1880 * We've reached the
1881 * it's safe to leave
1882 */
1883 if (cluster_in_el ==
1884 break;
1885
1886 /*
1887 * The 'pos' searched
1888 * always one cluster
1889 */
1890 trunc_end = cluster_i
1891
1892 ocfs2_reinit_path(pat
1893
1894 continue;
1895
1896 } else
1897 rec = &el->l_recs[i];
1898
1899 ocfs2_calc_trunc_pos(inode, e
1900 &trunc_l
1901 if (done)
1902 break;
1903
1904 flags = rec->e_flags;
1905 phys_cpos = ocfs2_blocks_to_c
1906
1907 ret = ocfs2_remove_btree_rang
1908
1909
1910 if (ret < 0) {
1911 mlog_errno(ret);
1912 goto out;
1913 }
1914
1915 cluster_in_el = trunc_end;
1916
1917 ocfs2_reinit_path(path, 1);
1918 }
1919
1920 ocfs2_truncate_cluster_pages(inode, b
1921
1922 out:
1923 ocfs2_free_path(path);
1924 ocfs2_schedule_truncate_log_flush(osb
1925 ocfs2_run_deallocs(osb, &dealloc);
1926
1927 return ret;
1928 }
1929
1930 /*
1931 * Parts of this function taken from xfs_chan
1932 */
1933 static int __ocfs2_change_file_space(struct f
1934 loff_t f
1935 struct o
1936 int chan
1937 {
1938 int ret;
1939 s64 llen;
1940 loff_t size, orig_isize;
1941 struct ocfs2_super *osb = OCFS2_SB(in
1942 struct buffer_head *di_bh = NULL;
1943 handle_t *handle;
1944 unsigned long long max_off = inode->i
1945
1946 if (ocfs2_is_hard_readonly(osb) || oc
1947 return -EROFS;
1948
1949 inode_lock(inode);
1950
1951 /* Wait all existing dio workers, new
1952 inode_dio_wait(inode);
1953 /*
1954 * This prevents concurrent writes on
1955 */
1956 ret = ocfs2_rw_lock(inode, 1);
1957 if (ret) {
1958 mlog_errno(ret);
1959 goto out;
1960 }
1961
1962 ret = ocfs2_inode_lock(inode, &di_bh,
1963 if (ret) {
1964 mlog_errno(ret);
1965 goto out_rw_unlock;
1966 }
1967
1968 if (inode->i_flags & (S_IMMUTABLE|S_A
1969 ret = -EPERM;
1970 goto out_inode_unlock;
1971 }
1972
1973 switch (sr->l_whence) {
1974 case 0: /*SEEK_SET*/
1975 break;
1976 case 1: /*SEEK_CUR*/
1977 sr->l_start += f_pos;
1978 break;
1979 case 2: /*SEEK_END*/
1980 sr->l_start += i_size_read(in
1981 break;
1982 default:
1983 ret = -EINVAL;
1984 goto out_inode_unlock;
1985 }
1986 sr->l_whence = 0;
1987
1988 llen = sr->l_len > 0 ? sr->l_len - 1
1989
1990 if (sr->l_start < 0
1991 || sr->l_start > max_off
1992 || (sr->l_start + llen) < 0
1993 || (sr->l_start + llen) > max_off
1994 ret = -EINVAL;
1995 goto out_inode_unlock;
1996 }
1997 size = sr->l_start + sr->l_len;
1998
1999 if (cmd == OCFS2_IOC_RESVSP || cmd ==
2000 cmd == OCFS2_IOC_UNRESVSP || cmd
2001 if (sr->l_len <= 0) {
2002 ret = -EINVAL;
2003 goto out_inode_unlock
2004 }
2005 }
2006
2007 if (file && setattr_should_drop_suidg
2008 ret = __ocfs2_write_remove_su
2009 if (ret) {
2010 mlog_errno(ret);
2011 goto out_inode_unlock
2012 }
2013 }
2014
2015 down_write(&OCFS2_I(inode)->ip_alloc_
2016 switch (cmd) {
2017 case OCFS2_IOC_RESVSP:
2018 case OCFS2_IOC_RESVSP64:
2019 /*
2020 * This takes unsigned offset
2021 * pass have been checked aga
2022 */
2023 ret = ocfs2_allocate_unwritte
2024
2025 break;
2026 case OCFS2_IOC_UNRESVSP:
2027 case OCFS2_IOC_UNRESVSP64:
2028 ret = ocfs2_remove_inode_rang
2029
2030 break;
2031 default:
2032 ret = -EINVAL;
2033 }
2034
2035 orig_isize = i_size_read(inode);
2036 /* zeroout eof blocks in the cluster.
2037 if (!ret && change_size && orig_isize
2038 ret = ocfs2_zeroout_partial_c
2039 size
2040 if (!ret)
2041 i_size_write(inode, s
2042 }
2043 up_write(&OCFS2_I(inode)->ip_alloc_se
2044 if (ret) {
2045 mlog_errno(ret);
2046 goto out_inode_unlock;
2047 }
2048
2049 /*
2050 * We update c/mtime for these change
2051 */
2052 handle = ocfs2_start_trans(osb, OCFS2
2053 if (IS_ERR(handle)) {
2054 ret = PTR_ERR(handle);
2055 mlog_errno(ret);
2056 goto out_inode_unlock;
2057 }
2058
2059 inode_set_mtime_to_ts(inode, inode_se
2060 ret = ocfs2_mark_inode_dirty(handle,
2061 if (ret < 0)
2062 mlog_errno(ret);
2063
2064 if (file && (file->f_flags & O_SYNC))
2065 handle->h_sync = 1;
2066
2067 ocfs2_commit_trans(osb, handle);
2068
2069 out_inode_unlock:
2070 brelse(di_bh);
2071 ocfs2_inode_unlock(inode, 1);
2072 out_rw_unlock:
2073 ocfs2_rw_unlock(inode, 1);
2074
2075 out:
2076 inode_unlock(inode);
2077 return ret;
2078 }
2079
2080 int ocfs2_change_file_space(struct file *file
2081 struct ocfs2_spac
2082 {
2083 struct inode *inode = file_inode(file
2084 struct ocfs2_super *osb = OCFS2_SB(in
2085 int ret;
2086
2087 if ((cmd == OCFS2_IOC_RESVSP || cmd =
2088 !ocfs2_writes_unwritten_extents(o
2089 return -ENOTTY;
2090 else if ((cmd == OCFS2_IOC_UNRESVSP |
2091 !ocfs2_sparse_alloc(osb))
2092 return -ENOTTY;
2093
2094 if (!S_ISREG(inode->i_mode))
2095 return -EINVAL;
2096
2097 if (!(file->f_mode & FMODE_WRITE))
2098 return -EBADF;
2099
2100 ret = mnt_want_write_file(file);
2101 if (ret)
2102 return ret;
2103 ret = __ocfs2_change_file_space(file,
2104 mnt_drop_write_file(file);
2105 return ret;
2106 }
2107
2108 static long ocfs2_fallocate(struct file *file
2109 loff_t len)
2110 {
2111 struct inode *inode = file_inode(file
2112 struct ocfs2_super *osb = OCFS2_SB(in
2113 struct ocfs2_space_resv sr;
2114 int change_size = 1;
2115 int cmd = OCFS2_IOC_RESVSP64;
2116 int ret = 0;
2117
2118 if (mode & ~(FALLOC_FL_KEEP_SIZE | FA
2119 return -EOPNOTSUPP;
2120 if (!ocfs2_writes_unwritten_extents(o
2121 return -EOPNOTSUPP;
2122
2123 if (mode & FALLOC_FL_KEEP_SIZE) {
2124 change_size = 0;
2125 } else {
2126 ret = inode_newsize_ok(inode,
2127 if (ret)
2128 return ret;
2129 }
2130
2131 if (mode & FALLOC_FL_PUNCH_HOLE)
2132 cmd = OCFS2_IOC_UNRESVSP64;
2133
2134 sr.l_whence = 0;
2135 sr.l_start = (s64)offset;
2136 sr.l_len = (s64)len;
2137
2138 return __ocfs2_change_file_space(NULL
2139 chan
2140 }
2141
2142 int ocfs2_check_range_for_refcount(struct ino
2143 size_t cou
2144 {
2145 int ret = 0;
2146 unsigned int extent_flags;
2147 u32 cpos, clusters, extent_len, phys_
2148 struct super_block *sb = inode->i_sb;
2149
2150 if (!ocfs2_refcount_tree(OCFS2_SB(ino
2151 !ocfs2_is_refcount_inode(inode) |
2152 OCFS2_I(inode)->ip_dyn_features &
2153 return 0;
2154
2155 cpos = pos >> OCFS2_SB(sb)->s_cluster
2156 clusters = ocfs2_clusters_for_bytes(s
2157
2158 while (clusters) {
2159 ret = ocfs2_get_clusters(inod
2160 &ext
2161 if (ret < 0) {
2162 mlog_errno(ret);
2163 goto out;
2164 }
2165
2166 if (phys_cpos && (extent_flag
2167 ret = 1;
2168 break;
2169 }
2170
2171 if (extent_len > clusters)
2172 extent_len = clusters
2173
2174 clusters -= extent_len;
2175 cpos += extent_len;
2176 }
2177 out:
2178 return ret;
2179 }
2180
2181 static int ocfs2_is_io_unaligned(struct inode
2182 {
2183 int blockmask = inode->i_sb->s_blocks
2184 loff_t final_size = pos + count;
2185
2186 if ((pos & blockmask) || (final_size
2187 return 1;
2188 return 0;
2189 }
2190
2191 static int ocfs2_inode_lock_for_extent_tree(s
2192 s
2193 i
2194 i
2195 i
2196 {
2197 int ret = 0;
2198
2199 if (wait)
2200 ret = ocfs2_inode_lock(inode,
2201 else
2202 ret = ocfs2_try_inode_lock(in
2203 if (ret < 0)
2204 goto out;
2205
2206 if (wait) {
2207 if (write_sem)
2208 down_write(&OCFS2_I(i
2209 else
2210 down_read(&OCFS2_I(in
2211 } else {
2212 if (write_sem)
2213 ret = down_write_tryl
2214 else
2215 ret = down_read_trylo
2216
2217 if (!ret) {
2218 ret = -EAGAIN;
2219 goto out_unlock;
2220 }
2221 }
2222
2223 return ret;
2224
2225 out_unlock:
2226 brelse(*di_bh);
2227 *di_bh = NULL;
2228 ocfs2_inode_unlock(inode, meta_level)
2229 out:
2230 return ret;
2231 }
2232
2233 static void ocfs2_inode_unlock_for_extent_tre
2234
2235
2236
2237 {
2238 if (write_sem)
2239 up_write(&OCFS2_I(inode)->ip_
2240 else
2241 up_read(&OCFS2_I(inode)->ip_a
2242
2243 brelse(*di_bh);
2244 *di_bh = NULL;
2245
2246 if (meta_level >= 0)
2247 ocfs2_inode_unlock(inode, met
2248 }
2249
2250 static int ocfs2_prepare_inode_for_write(stru
2251 loff
2252 {
2253 int ret = 0, meta_level = 0, overwrit
2254 int write_sem = 0;
2255 struct dentry *dentry = file->f_path.
2256 struct inode *inode = d_inode(dentry)
2257 struct buffer_head *di_bh = NULL;
2258 u32 cpos;
2259 u32 clusters;
2260
2261 /*
2262 * We start with a read level meta lo
2263 * if we need to make modifications h
2264 */
2265 for(;;) {
2266 ret = ocfs2_inode_lock_for_ex
2267
2268
2269
2270
2271 if (ret < 0) {
2272 if (ret != -EAGAIN)
2273 mlog_errno(re
2274 goto out;
2275 }
2276
2277 /*
2278 * Check if IO will overwrite
2279 * IOCB_NOWAIT flag is set.
2280 */
2281 if (!wait && !overwrite_io) {
2282 overwrite_io = 1;
2283
2284 ret = ocfs2_overwrite
2285 if (ret < 0) {
2286 if (ret != -E
2287 mlog_
2288 goto out_unlo
2289 }
2290 }
2291
2292 /* Clear suid / sgid if neces
2293 * instead of later in the wr
2294 * remove_suid() calls ->seta
2295 * we may have already done o
2296 * ocfs2_setattr() *must* tak
2297 * proceed, this will lead us
2298 * inode. There's also the di
2299 * can be lost via setattr du
2300 * set inode->i_size at the e
2301 if (setattr_should_drop_suidg
2302 if (meta_level == 0)
2303 ocfs2_inode_u
2304
2305
2306
2307 meta_level =
2308 continue;
2309 }
2310
2311 ret = ocfs2_write_rem
2312 if (ret < 0) {
2313 mlog_errno(re
2314 goto out_unlo
2315 }
2316 }
2317
2318 ret = ocfs2_check_range_for_r
2319 if (ret == 1) {
2320 ocfs2_inode_unlock_fo
2321
2322
2323
2324 meta_level = 1;
2325 write_sem = 1;
2326 ret = ocfs2_inode_loc
2327
2328
2329
2330
2331 if (ret < 0) {
2332 if (ret != -E
2333 mlog_
2334 goto out;
2335 }
2336
2337 cpos = pos >> OCFS2_S
2338 clusters =
2339 ocfs2_cluster
2340 ret = ocfs2_refcount_
2341 }
2342
2343 if (ret < 0) {
2344 if (ret != -EAGAIN)
2345 mlog_errno(re
2346 goto out_unlock;
2347 }
2348
2349 break;
2350 }
2351
2352 out_unlock:
2353 trace_ocfs2_prepare_inode_for_write(O
2354 p
2355
2356 ocfs2_inode_unlock_for_extent_tree(in
2357 &d
2358 me
2359 wr
2360
2361 out:
2362 return ret;
2363 }
2364
2365 static ssize_t ocfs2_file_write_iter(struct k
2366 struct io
2367 {
2368 int rw_level;
2369 ssize_t written = 0;
2370 ssize_t ret;
2371 size_t count = iov_iter_count(from);
2372 struct file *file = iocb->ki_filp;
2373 struct inode *inode = file_inode(file
2374 struct ocfs2_super *osb = OCFS2_SB(in
2375 int full_coherency = !(osb->s_mount_o
2376 OCFS2_MOUNT_CO
2377 void *saved_ki_complete = NULL;
2378 int append_write = ((iocb->ki_pos + c
2379 i_size_read(inode) ?
2380 int direct_io = iocb->ki_flags & IOCB
2381 int nowait = iocb->ki_flags & IOCB_NO
2382
2383 trace_ocfs2_file_write_iter(inode, fi
2384 (unsigned long long)OCFS2_I(i
2385 file->f_path.dentry->d_name.l
2386 file->f_path.dentry->d_name.n
2387 (unsigned int)from->nr_segs);
2388
2389 if (!direct_io && nowait)
2390 return -EOPNOTSUPP;
2391
2392 if (count == 0)
2393 return 0;
2394
2395 if (nowait) {
2396 if (!inode_trylock(inode))
2397 return -EAGAIN;
2398 } else
2399 inode_lock(inode);
2400
2401 /*
2402 * Concurrent O_DIRECT writes are all
2403 * mount_option "coherency=buffered".
2404 * For append write, we must take rw
2405 */
2406 rw_level = (!direct_io || full_cohere
2407
2408 if (nowait)
2409 ret = ocfs2_try_rw_lock(inode
2410 else
2411 ret = ocfs2_rw_lock(inode, rw
2412 if (ret < 0) {
2413 if (ret != -EAGAIN)
2414 mlog_errno(ret);
2415 goto out_mutex;
2416 }
2417
2418 /*
2419 * O_DIRECT writes with "coherency=fu
2420 * inode_lock to guarantee coherency.
2421 */
2422 if (direct_io && full_coherency) {
2423 /*
2424 * We need to take and drop t
2425 * other nodes to drop their
2426 * already does this in write
2427 */
2428 if (nowait)
2429 ret = ocfs2_try_inode
2430 else
2431 ret = ocfs2_inode_loc
2432 if (ret < 0) {
2433 if (ret != -EAGAIN)
2434 mlog_errno(re
2435 goto out;
2436 }
2437
2438 ocfs2_inode_unlock(inode, 1);
2439 }
2440
2441 ret = generic_write_checks(iocb, from
2442 if (ret <= 0) {
2443 if (ret)
2444 mlog_errno(ret);
2445 goto out;
2446 }
2447 count = ret;
2448
2449 ret = ocfs2_prepare_inode_for_write(f
2450 if (ret < 0) {
2451 if (ret != -EAGAIN)
2452 mlog_errno(ret);
2453 goto out;
2454 }
2455
2456 if (direct_io && !is_sync_kiocb(iocb)
2457 ocfs2_is_io_unaligned(inode, coun
2458 /*
2459 * Make it a sync io if it's
2460 */
2461 saved_ki_complete = xchg(&ioc
2462 }
2463
2464 /* communicate with ocfs2_dio_end_io
2465 ocfs2_iocb_set_rw_locked(iocb, rw_lev
2466
2467 written = __generic_file_write_iter(i
2468 /* buffered aio wouldn't have proper
2469 BUG_ON(written == -EIOCBQUEUED && !di
2470
2471 /*
2472 * deep in g_f_a_w_n()->ocfs2_direct_
2473 * function pointer which is called w
2474 * it can unlock our rw lock.
2475 * Unfortunately there are error case
2476 * that don't. so we don't have to u
2477 * async dio is going to do it in the
2478 * error has already done it.
2479 */
2480 if ((written == -EIOCBQUEUED) || (!oc
2481 rw_level = -1;
2482 }
2483
2484 if (unlikely(written <= 0))
2485 goto out;
2486
2487 if (((file->f_flags & O_DSYNC) && !di
2488 IS_SYNC(inode)) {
2489 ret = filemap_fdatawrite_rang
2490
2491
2492 if (ret < 0)
2493 written = ret;
2494
2495 if (!ret) {
2496 ret = jbd2_journal_fo
2497 if (ret < 0)
2498 written = ret
2499 }
2500
2501 if (!ret)
2502 ret = filemap_fdatawa
2503
2504
2505 }
2506
2507 out:
2508 if (saved_ki_complete)
2509 xchg(&iocb->ki_complete, save
2510
2511 if (rw_level != -1)
2512 ocfs2_rw_unlock(inode, rw_lev
2513
2514 out_mutex:
2515 inode_unlock(inode);
2516
2517 if (written)
2518 ret = written;
2519 return ret;
2520 }
2521
2522 static ssize_t ocfs2_file_read_iter(struct ki
2523 struct iov
2524 {
2525 int ret = 0, rw_level = -1, lock_leve
2526 struct file *filp = iocb->ki_filp;
2527 struct inode *inode = file_inode(filp
2528 int direct_io = iocb->ki_flags & IOCB
2529 int nowait = iocb->ki_flags & IOCB_NO
2530
2531 trace_ocfs2_file_read_iter(inode, fil
2532 (unsigned long long)O
2533 filp->f_path.dentry->
2534 filp->f_path.dentry->
2535 to->nr_segs); /* GR
2536
2537
2538 if (!inode) {
2539 ret = -EINVAL;
2540 mlog_errno(ret);
2541 goto bail;
2542 }
2543
2544 if (!direct_io && nowait)
2545 return -EOPNOTSUPP;
2546
2547 /*
2548 * buffered reads protect themselves
2549 * need locks to protect pending read
2550 */
2551 if (direct_io) {
2552 if (nowait)
2553 ret = ocfs2_try_rw_lo
2554 else
2555 ret = ocfs2_rw_lock(i
2556
2557 if (ret < 0) {
2558 if (ret != -EAGAIN)
2559 mlog_errno(re
2560 goto bail;
2561 }
2562 rw_level = 0;
2563 /* communicate with ocfs2_dio
2564 ocfs2_iocb_set_rw_locked(iocb
2565 }
2566
2567 /*
2568 * We're fine letting folks race trun
2569 * writes with read across the cluste
2570 * locally. Hence no rw_lock during r
2571 *
2572 * Take and drop the meta data lock t
2573 * like i_size. This allows the check
2574 * copy_splice_read() a chance of act
2575 */
2576 ret = ocfs2_inode_lock_atime(inode, f
2577 !nowait)
2578 if (ret < 0) {
2579 if (ret != -EAGAIN)
2580 mlog_errno(ret);
2581 goto bail;
2582 }
2583 ocfs2_inode_unlock(inode, lock_level)
2584
2585 ret = generic_file_read_iter(iocb, to
2586 trace_generic_file_read_iter_ret(ret)
2587
2588 /* buffered aio wouldn't have proper
2589 BUG_ON(ret == -EIOCBQUEUED && !direct
2590
2591 /* see ocfs2_file_write_iter */
2592 if (ret == -EIOCBQUEUED || !ocfs2_ioc
2593 rw_level = -1;
2594 }
2595
2596 bail:
2597 if (rw_level != -1)
2598 ocfs2_rw_unlock(inode, rw_lev
2599
2600 return ret;
2601 }
2602
2603 static ssize_t ocfs2_file_splice_read(struct
2604 struct
2605 size_t
2606 {
2607 struct inode *inode = file_inode(in);
2608 ssize_t ret = 0;
2609 int lock_level = 0;
2610
2611 trace_ocfs2_file_splice_read(inode, i
2612 (unsigne
2613 in->f_pa
2614 in->f_pa
2615 flags);
2616
2617 /*
2618 * We're fine letting folks race trun
2619 * read across the cluster, just like
2620 * rw_lock during read.
2621 *
2622 * Take and drop the meta data lock t
2623 * This allows the checks down below
2624 * actually working.
2625 */
2626 ret = ocfs2_inode_lock_atime(inode, i
2627 if (ret < 0) {
2628 if (ret != -EAGAIN)
2629 mlog_errno(ret);
2630 goto bail;
2631 }
2632 ocfs2_inode_unlock(inode, lock_level)
2633
2634 ret = filemap_splice_read(in, ppos, p
2635 trace_filemap_splice_read_ret(ret);
2636 bail:
2637 return ret;
2638 }
2639
2640 /* Refer generic_file_llseek_unlocked() */
2641 static loff_t ocfs2_file_llseek(struct file *
2642 {
2643 struct inode *inode = file->f_mapping
2644 int ret = 0;
2645
2646 inode_lock(inode);
2647
2648 switch (whence) {
2649 case SEEK_SET:
2650 break;
2651 case SEEK_END:
2652 /* SEEK_END requires the OCFS
2653 * because it references the
2654 */
2655 ret = ocfs2_inode_lock(inode,
2656 if (ret < 0) {
2657 mlog_errno(ret);
2658 goto out;
2659 }
2660 offset += i_size_read(inode);
2661 ocfs2_inode_unlock(inode, 0);
2662 break;
2663 case SEEK_CUR:
2664 if (offset == 0) {
2665 offset = file->f_pos;
2666 goto out;
2667 }
2668 offset += file->f_pos;
2669 break;
2670 case SEEK_DATA:
2671 case SEEK_HOLE:
2672 ret = ocfs2_seek_data_hole_of
2673 if (ret)
2674 goto out;
2675 break;
2676 default:
2677 ret = -EINVAL;
2678 goto out;
2679 }
2680
2681 offset = vfs_setpos(file, offset, ino
2682
2683 out:
2684 inode_unlock(inode);
2685 if (ret)
2686 return ret;
2687 return offset;
2688 }
2689
2690 static loff_t ocfs2_remap_file_range(struct f
2691 struct f
2692 loff_t l
2693 {
2694 struct inode *inode_in = file_inode(f
2695 struct inode *inode_out = file_inode(
2696 struct ocfs2_super *osb = OCFS2_SB(in
2697 struct buffer_head *in_bh = NULL, *ou
2698 bool same_inode = (inode_in == inode_
2699 loff_t remapped = 0;
2700 ssize_t ret;
2701
2702 if (remap_flags & ~(REMAP_FILE_DEDUP
2703 return -EINVAL;
2704 if (!ocfs2_refcount_tree(osb))
2705 return -EOPNOTSUPP;
2706 if (ocfs2_is_hard_readonly(osb) || oc
2707 return -EROFS;
2708
2709 /* Lock both files against IO */
2710 ret = ocfs2_reflink_inodes_lock(inode
2711 if (ret)
2712 return ret;
2713
2714 /* Check file eligibility and prepare
2715 ret = -EINVAL;
2716 if ((OCFS2_I(inode_in)->ip_flags & OC
2717 (OCFS2_I(inode_out)->ip_flags & O
2718 goto out_unlock;
2719
2720 ret = generic_remap_file_range_prep(f
2721 &len, remap_flags);
2722 if (ret < 0 || len == 0)
2723 goto out_unlock;
2724
2725 /* Lock out changes to the allocation
2726 down_write(&OCFS2_I(inode_in)->ip_all
2727 if (!same_inode)
2728 down_write_nested(&OCFS2_I(in
2729 SINGLE_DEPT
2730
2731 /* Zap any page cache for the destina
2732 truncate_inode_pages_range(&inode_out
2733 round_down
2734 round_up(p
2735
2736 remapped = ocfs2_reflink_remap_blocks
2737 inode_out, out_bh, po
2738 up_write(&OCFS2_I(inode_in)->ip_alloc
2739 if (!same_inode)
2740 up_write(&OCFS2_I(inode_out)-
2741 if (remapped < 0) {
2742 ret = remapped;
2743 mlog_errno(ret);
2744 goto out_unlock;
2745 }
2746
2747 /*
2748 * Empty the extent map so that we ma
2749 * record from the disk.
2750 */
2751 ocfs2_extent_map_trunc(inode_in, 0);
2752 ocfs2_extent_map_trunc(inode_out, 0);
2753
2754 ret = ocfs2_reflink_update_dest(inode
2755 if (ret) {
2756 mlog_errno(ret);
2757 goto out_unlock;
2758 }
2759
2760 out_unlock:
2761 ocfs2_reflink_inodes_unlock(inode_in,
2762 return remapped > 0 ? remapped : ret;
2763 }
2764
2765 static loff_t ocfs2_dir_llseek(struct file *f
2766 {
2767 struct ocfs2_file_private *fp = file-
2768
2769 return generic_llseek_cookie(file, of
2770 }
2771
2772 const struct inode_operations ocfs2_file_iops
2773 .setattr = ocfs2_setattr,
2774 .getattr = ocfs2_getattr,
2775 .permission = ocfs2_permission,
2776 .listxattr = ocfs2_listxattr,
2777 .fiemap = ocfs2_fiemap,
2778 .get_inode_acl = ocfs2_iop_get_acl,
2779 .set_acl = ocfs2_iop_set_acl,
2780 .fileattr_get = ocfs2_fileattr_get,
2781 .fileattr_set = ocfs2_fileattr_set,
2782 };
2783
2784 const struct inode_operations ocfs2_special_f
2785 .setattr = ocfs2_setattr,
2786 .getattr = ocfs2_getattr,
2787 .listxattr = ocfs2_listxattr,
2788 .permission = ocfs2_permission,
2789 .get_inode_acl = ocfs2_iop_get_acl,
2790 .set_acl = ocfs2_iop_set_acl,
2791 };
2792
2793 /*
2794 * Other than ->lock, keep ocfs2_fops and ocf
2795 * ocfs2_fops_no_plocks and ocfs2_dops_no_plo
2796 */
2797 const struct file_operations ocfs2_fops = {
2798 .llseek = ocfs2_file_llseek,
2799 .mmap = ocfs2_mmap,
2800 .fsync = ocfs2_sync_file,
2801 .release = ocfs2_file_release,
2802 .open = ocfs2_file_open,
2803 .read_iter = ocfs2_file_read_ite
2804 .write_iter = ocfs2_file_write_it
2805 .unlocked_ioctl = ocfs2_ioctl,
2806 #ifdef CONFIG_COMPAT
2807 .compat_ioctl = ocfs2_compat_ioctl,
2808 #endif
2809 .lock = ocfs2_lock,
2810 .flock = ocfs2_flock,
2811 .splice_read = ocfs2_file_splice_r
2812 .splice_write = iter_file_splice_wr
2813 .fallocate = ocfs2_fallocate,
2814 .remap_file_range = ocfs2_remap_file_
2815 };
2816
2817 WRAP_DIR_ITER(ocfs2_readdir) // FIXME!
2818 const struct file_operations ocfs2_dops = {
2819 .llseek = ocfs2_dir_llseek,
2820 .read = generic_read_dir,
2821 .iterate_shared = shared_ocfs2_readdi
2822 .fsync = ocfs2_sync_file,
2823 .release = ocfs2_dir_release,
2824 .open = ocfs2_dir_open,
2825 .unlocked_ioctl = ocfs2_ioctl,
2826 #ifdef CONFIG_COMPAT
2827 .compat_ioctl = ocfs2_compat_ioctl,
2828 #endif
2829 .lock = ocfs2_lock,
2830 .flock = ocfs2_flock,
2831 };
2832
2833 /*
2834 * POSIX-lockless variants of our file_operat
2835 *
2836 * These will be used if the underlying clust
2837 * posix file locking, if the user passes the
2838 * option, or if we have a local-only fs.
2839 *
2840 * ocfs2_flock is in here because all stacks
2841 * so we still want it in the case of no stac
2842 * plocks. Internally, it will do the right t
2843 * the cluster.
2844 */
2845 const struct file_operations ocfs2_fops_no_pl
2846 .llseek = ocfs2_file_llseek,
2847 .mmap = ocfs2_mmap,
2848 .fsync = ocfs2_sync_file,
2849 .release = ocfs2_file_release,
2850 .open = ocfs2_file_open,
2851 .read_iter = ocfs2_file_read_ite
2852 .write_iter = ocfs2_file_write_it
2853 .unlocked_ioctl = ocfs2_ioctl,
2854 #ifdef CONFIG_COMPAT
2855 .compat_ioctl = ocfs2_compat_ioctl,
2856 #endif
2857 .flock = ocfs2_flock,
2858 .splice_read = filemap_splice_read
2859 .splice_write = iter_file_splice_wr
2860 .fallocate = ocfs2_fallocate,
2861 .remap_file_range = ocfs2_remap_file_
2862 };
2863
2864 const struct file_operations ocfs2_dops_no_pl
2865 .llseek = ocfs2_dir_llseek,
2866 .read = generic_read_dir,
2867 .iterate_shared = shared_ocfs2_readdi
2868 .fsync = ocfs2_sync_file,
2869 .release = ocfs2_dir_release,
2870 .open = ocfs2_dir_open,
2871 .unlocked_ioctl = ocfs2_ioctl,
2872 #ifdef CONFIG_COMPAT
2873 .compat_ioctl = ocfs2_compat_ioctl,
2874 #endif
2875 .flock = ocfs2_flock,
2876 };
2877
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.