1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * File open, close, extend, truncate
8 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
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.
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.
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.
26 #include <linux/capability.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 #include <linux/backing-dev.h>
42 #include <cluster/masklog.h>
50 #include "extent_map.h"
63 #include "refcounttree.h"
64 #include "ocfs2_trace.h"
66 #include "buffer_head_io.h"
68 static int ocfs2_init_file_private(struct inode *inode, struct file *file)
70 struct ocfs2_file_private *fp;
72 fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
77 mutex_init(&fp->fp_mutex);
78 ocfs2_file_lock_res_init(&fp->fp_flock, fp);
79 file->private_data = fp;
84 static void ocfs2_free_file_private(struct inode *inode, struct file *file)
86 struct ocfs2_file_private *fp = file->private_data;
87 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
90 ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
91 ocfs2_lock_res_free(&fp->fp_flock);
93 file->private_data = NULL;
97 static int ocfs2_file_open(struct inode *inode, struct file *file)
100 int mode = file->f_flags;
101 struct ocfs2_inode_info *oi = OCFS2_I(inode);
103 trace_ocfs2_file_open(inode, file, file->f_path.dentry,
104 (unsigned long long)OCFS2_I(inode)->ip_blkno,
105 file->f_path.dentry->d_name.len,
106 file->f_path.dentry->d_name.name, mode);
108 if (file->f_mode & FMODE_WRITE) {
109 status = dquot_initialize(inode);
114 spin_lock(&oi->ip_lock);
116 /* Check that the inode hasn't been wiped from disk by another
117 * node. If it hasn't then we're safe as long as we hold the
118 * spin lock until our increment of open count. */
119 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
120 spin_unlock(&oi->ip_lock);
127 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
130 spin_unlock(&oi->ip_lock);
132 status = ocfs2_init_file_private(inode, file);
135 * We want to set open count back if we're failing the
138 spin_lock(&oi->ip_lock);
140 spin_unlock(&oi->ip_lock);
147 static int ocfs2_file_release(struct inode *inode, struct file *file)
149 struct ocfs2_inode_info *oi = OCFS2_I(inode);
151 spin_lock(&oi->ip_lock);
152 if (!--oi->ip_open_count)
153 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
155 trace_ocfs2_file_release(inode, file, file->f_path.dentry,
157 file->f_path.dentry->d_name.len,
158 file->f_path.dentry->d_name.name,
160 spin_unlock(&oi->ip_lock);
162 ocfs2_free_file_private(inode, file);
167 static int ocfs2_dir_open(struct inode *inode, struct file *file)
169 return ocfs2_init_file_private(inode, file);
172 static int ocfs2_dir_release(struct inode *inode, struct file *file)
174 ocfs2_free_file_private(inode, file);
178 static int ocfs2_sync_file(struct file *file, loff_t start, loff_t end,
182 struct inode *inode = file->f_mapping->host;
183 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
184 struct ocfs2_inode_info *oi = OCFS2_I(inode);
185 journal_t *journal = osb->journal->j_journal;
188 bool needs_barrier = false;
190 trace_ocfs2_sync_file(inode, file, file->f_path.dentry,
191 OCFS2_I(inode)->ip_blkno,
192 file->f_path.dentry->d_name.len,
193 file->f_path.dentry->d_name.name,
194 (unsigned long long)datasync);
196 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
199 err = filemap_write_and_wait_range(inode->i_mapping, start, end);
203 commit_tid = datasync ? oi->i_datasync_tid : oi->i_sync_tid;
204 if (journal->j_flags & JBD2_BARRIER &&
205 !jbd2_trans_will_send_data_barrier(journal, commit_tid))
206 needs_barrier = true;
207 err = jbd2_complete_transaction(journal, commit_tid);
209 ret = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
217 return (err < 0) ? -EIO : 0;
220 int ocfs2_should_update_atime(struct inode *inode,
221 struct vfsmount *vfsmnt)
224 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
226 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
229 if ((inode->i_flags & S_NOATIME) ||
230 ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
234 * We can be called with no vfsmnt structure - NFSD will
237 * Note that our action here is different than touch_atime() -
238 * if we can't tell whether this is a noatime mount, then we
239 * don't know whether to trust the value of s_atime_quantum.
244 if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
245 ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
248 if (vfsmnt->mnt_flags & MNT_RELATIME) {
249 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
250 (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
257 if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
263 int ocfs2_update_inode_atime(struct inode *inode,
264 struct buffer_head *bh)
267 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
269 struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
271 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
272 if (IS_ERR(handle)) {
273 ret = PTR_ERR(handle);
278 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
279 OCFS2_JOURNAL_ACCESS_WRITE);
286 * Don't use ocfs2_mark_inode_dirty() here as we don't always
287 * have i_mutex to guard against concurrent changes to other
290 inode->i_atime = CURRENT_TIME;
291 di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
292 di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
293 ocfs2_update_inode_fsync_trans(handle, inode, 0);
294 ocfs2_journal_dirty(handle, bh);
297 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
302 int ocfs2_set_inode_size(handle_t *handle,
304 struct buffer_head *fe_bh,
309 i_size_write(inode, new_i_size);
310 inode->i_blocks = ocfs2_inode_sector_count(inode);
311 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
313 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
323 int ocfs2_simple_size_update(struct inode *inode,
324 struct buffer_head *di_bh,
328 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
329 handle_t *handle = NULL;
331 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
332 if (IS_ERR(handle)) {
333 ret = PTR_ERR(handle);
338 ret = ocfs2_set_inode_size(handle, inode, di_bh,
343 ocfs2_update_inode_fsync_trans(handle, inode, 0);
344 ocfs2_commit_trans(osb, handle);
349 static int ocfs2_cow_file_pos(struct inode *inode,
350 struct buffer_head *fe_bh,
354 u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
355 unsigned int num_clusters = 0;
356 unsigned int ext_flags = 0;
359 * If the new offset is aligned to the range of the cluster, there is
360 * no space for ocfs2_zero_range_for_truncate to fill, so no need to
363 if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
366 status = ocfs2_get_clusters(inode, cpos, &phys,
367 &num_clusters, &ext_flags);
373 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
376 return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1);
382 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
384 struct buffer_head *fe_bh,
389 struct ocfs2_dinode *di;
393 * We need to CoW the cluster contains the offset if it is reflinked
394 * since we will call ocfs2_zero_range_for_truncate later which will
395 * write "0" from offset to the end of the cluster.
397 status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
403 /* TODO: This needs to actually orphan the inode in this
406 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
407 if (IS_ERR(handle)) {
408 status = PTR_ERR(handle);
413 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
414 OCFS2_JOURNAL_ACCESS_WRITE);
421 * Do this before setting i_size.
423 cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
424 status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
431 i_size_write(inode, new_i_size);
432 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
434 di = (struct ocfs2_dinode *) fe_bh->b_data;
435 di->i_size = cpu_to_le64(new_i_size);
436 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
437 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
438 ocfs2_update_inode_fsync_trans(handle, inode, 0);
440 ocfs2_journal_dirty(handle, fe_bh);
443 ocfs2_commit_trans(osb, handle);
448 int ocfs2_truncate_file(struct inode *inode,
449 struct buffer_head *di_bh,
453 struct ocfs2_dinode *fe = NULL;
454 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
456 /* We trust di_bh because it comes from ocfs2_inode_lock(), which
457 * already validated it */
458 fe = (struct ocfs2_dinode *) di_bh->b_data;
460 trace_ocfs2_truncate_file((unsigned long long)OCFS2_I(inode)->ip_blkno,
461 (unsigned long long)le64_to_cpu(fe->i_size),
462 (unsigned long long)new_i_size);
464 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
465 "Inode %llu, inode i_size = %lld != di "
466 "i_size = %llu, i_flags = 0x%x\n",
467 (unsigned long long)OCFS2_I(inode)->ip_blkno,
469 (unsigned long long)le64_to_cpu(fe->i_size),
470 le32_to_cpu(fe->i_flags));
472 if (new_i_size > le64_to_cpu(fe->i_size)) {
473 trace_ocfs2_truncate_file_error(
474 (unsigned long long)le64_to_cpu(fe->i_size),
475 (unsigned long long)new_i_size);
481 down_write(&OCFS2_I(inode)->ip_alloc_sem);
483 ocfs2_resv_discard(&osb->osb_la_resmap,
484 &OCFS2_I(inode)->ip_la_data_resv);
487 * The inode lock forced other nodes to sync and drop their
488 * pages, which (correctly) happens even if we have a truncate
489 * without allocation change - ocfs2 cluster sizes can be much
490 * greater than page size, so we have to truncate them
493 unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
494 truncate_inode_pages(inode->i_mapping, new_i_size);
496 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
497 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
498 i_size_read(inode), 1);
502 goto bail_unlock_sem;
505 /* alright, we're going to need to do a full blown alloc size
506 * change. Orphan the inode so that recovery can complete the
507 * truncate if necessary. This does the task of marking
509 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
512 goto bail_unlock_sem;
515 status = ocfs2_commit_truncate(osb, inode, di_bh);
518 goto bail_unlock_sem;
521 /* TODO: orphan dir cleanup here. */
523 up_write(&OCFS2_I(inode)->ip_alloc_sem);
526 if (!status && OCFS2_I(inode)->ip_clusters == 0)
527 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
533 * extend file allocation only here.
534 * we'll update all the disk stuff, and oip->alloc_size
536 * expect stuff to be locked, a transaction started and enough data /
537 * metadata reservations in the contexts.
539 * Will return -EAGAIN, and a reason if a restart is needed.
540 * If passed in, *reason will always be set, even in error.
542 int ocfs2_add_inode_data(struct ocfs2_super *osb,
547 struct buffer_head *fe_bh,
549 struct ocfs2_alloc_context *data_ac,
550 struct ocfs2_alloc_context *meta_ac,
551 enum ocfs2_alloc_restarted *reason_ret)
554 struct ocfs2_extent_tree et;
556 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
557 ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
558 clusters_to_add, mark_unwritten,
559 data_ac, meta_ac, reason_ret);
564 static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
565 u32 clusters_to_add, int mark_unwritten)
568 int restart_func = 0;
571 struct buffer_head *bh = NULL;
572 struct ocfs2_dinode *fe = NULL;
573 handle_t *handle = NULL;
574 struct ocfs2_alloc_context *data_ac = NULL;
575 struct ocfs2_alloc_context *meta_ac = NULL;
576 enum ocfs2_alloc_restarted why = RESTART_NONE;
577 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
578 struct ocfs2_extent_tree et;
582 * Unwritten extent only exists for file systems which
585 BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
587 status = ocfs2_read_inode_block(inode, &bh);
592 fe = (struct ocfs2_dinode *) bh->b_data;
595 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
597 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
598 status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
605 credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list);
606 handle = ocfs2_start_trans(osb, credits);
607 if (IS_ERR(handle)) {
608 status = PTR_ERR(handle);
614 restarted_transaction:
615 trace_ocfs2_extend_allocation(
616 (unsigned long long)OCFS2_I(inode)->ip_blkno,
617 (unsigned long long)i_size_read(inode),
618 le32_to_cpu(fe->i_clusters), clusters_to_add,
621 status = dquot_alloc_space_nodirty(inode,
622 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
627 /* reserve a write to the file entry early on - that we if we
628 * run out of credits in the allocation path, we can still
630 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
631 OCFS2_JOURNAL_ACCESS_WRITE);
637 prev_clusters = OCFS2_I(inode)->ip_clusters;
639 status = ocfs2_add_inode_data(osb,
649 if ((status < 0) && (status != -EAGAIN)) {
650 if (status != -ENOSPC)
654 ocfs2_update_inode_fsync_trans(handle, inode, 1);
655 ocfs2_journal_dirty(handle, bh);
657 spin_lock(&OCFS2_I(inode)->ip_lock);
658 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
659 spin_unlock(&OCFS2_I(inode)->ip_lock);
660 /* Release unused quota reservation */
661 dquot_free_space(inode,
662 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
665 if (why != RESTART_NONE && clusters_to_add) {
666 if (why == RESTART_META) {
670 BUG_ON(why != RESTART_TRANS);
672 status = ocfs2_allocate_extend_trans(handle, 1);
674 /* handle still has to be committed at
680 goto restarted_transaction;
684 trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno,
685 le32_to_cpu(fe->i_clusters),
686 (unsigned long long)le64_to_cpu(fe->i_size),
687 OCFS2_I(inode)->ip_clusters,
688 (unsigned long long)i_size_read(inode));
691 if (status < 0 && did_quota)
692 dquot_free_space(inode,
693 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
695 ocfs2_commit_trans(osb, handle);
699 ocfs2_free_alloc_context(data_ac);
703 ocfs2_free_alloc_context(meta_ac);
706 if ((!status) && restart_func) {
716 int ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
717 u32 clusters_to_add, int mark_unwritten)
719 return __ocfs2_extend_allocation(inode, logical_start,
720 clusters_to_add, mark_unwritten);
724 * While a write will already be ordering the data, a truncate will not.
725 * Thus, we need to explicitly order the zeroed pages.
727 static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode,
728 struct buffer_head *di_bh)
730 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
731 handle_t *handle = NULL;
734 if (!ocfs2_should_order_data(inode))
737 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
738 if (IS_ERR(handle)) {
744 ret = ocfs2_jbd2_file_inode(handle, inode);
750 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
751 OCFS2_JOURNAL_ACCESS_WRITE);
754 ocfs2_update_inode_fsync_trans(handle, inode, 1);
759 ocfs2_commit_trans(osb, handle);
760 handle = ERR_PTR(ret);
765 /* Some parts of this taken from generic_cont_expand, which turned out
766 * to be too fragile to do exactly what we need without us having to
767 * worry about recursive locking in ->write_begin() and ->write_end(). */
768 static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
769 u64 abs_to, struct buffer_head *di_bh)
771 struct address_space *mapping = inode->i_mapping;
773 unsigned long index = abs_from >> PAGE_CACHE_SHIFT;
776 unsigned zero_from, zero_to, block_start, block_end;
777 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
779 BUG_ON(abs_from >= abs_to);
780 BUG_ON(abs_to > (((u64)index + 1) << PAGE_CACHE_SHIFT));
781 BUG_ON(abs_from & (inode->i_blkbits - 1));
783 handle = ocfs2_zero_start_ordered_transaction(inode, di_bh);
784 if (IS_ERR(handle)) {
785 ret = PTR_ERR(handle);
789 page = find_or_create_page(mapping, index, GFP_NOFS);
793 goto out_commit_trans;
796 /* Get the offsets within the page that we want to zero */
797 zero_from = abs_from & (PAGE_CACHE_SIZE - 1);
798 zero_to = abs_to & (PAGE_CACHE_SIZE - 1);
800 zero_to = PAGE_CACHE_SIZE;
802 trace_ocfs2_write_zero_page(
803 (unsigned long long)OCFS2_I(inode)->ip_blkno,
804 (unsigned long long)abs_from,
805 (unsigned long long)abs_to,
806 index, zero_from, zero_to);
808 /* We know that zero_from is block aligned */
809 for (block_start = zero_from; block_start < zero_to;
810 block_start = block_end) {
811 block_end = block_start + (1 << inode->i_blkbits);
814 * block_start is block-aligned. Bump it by one to force
815 * __block_write_begin and block_commit_write to zero the
818 ret = __block_write_begin(page, block_start + 1, 0,
826 /* must not update i_size! */
827 ret = block_commit_write(page, block_start + 1,
836 * fs-writeback will release the dirty pages without page lock
837 * whose offset are over inode size, the release happens at
838 * block_write_full_page().
840 i_size_write(inode, abs_to);
841 inode->i_blocks = ocfs2_inode_sector_count(inode);
842 di->i_size = cpu_to_le64((u64)i_size_read(inode));
843 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
844 di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
845 di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
846 di->i_mtime_nsec = di->i_ctime_nsec;
848 ocfs2_journal_dirty(handle, di_bh);
849 ocfs2_update_inode_fsync_trans(handle, inode, 1);
854 page_cache_release(page);
857 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
863 * Find the next range to zero. We do this in terms of bytes because
864 * that's what ocfs2_zero_extend() wants, and it is dealing with the
865 * pagecache. We may return multiple extents.
867 * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
868 * needs to be zeroed. range_start and range_end return the next zeroing
869 * range. A subsequent call should pass the previous range_end as its
870 * zero_start. If range_end is 0, there's nothing to do.
872 * Unwritten extents are skipped over. Refcounted extents are CoWd.
874 static int ocfs2_zero_extend_get_range(struct inode *inode,
875 struct buffer_head *di_bh,
876 u64 zero_start, u64 zero_end,
877 u64 *range_start, u64 *range_end)
879 int rc = 0, needs_cow = 0;
880 u32 p_cpos, zero_clusters = 0;
882 zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
883 u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
884 unsigned int num_clusters = 0;
885 unsigned int ext_flags = 0;
887 while (zero_cpos < last_cpos) {
888 rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
889 &num_clusters, &ext_flags);
895 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
896 zero_clusters = num_clusters;
897 if (ext_flags & OCFS2_EXT_REFCOUNTED)
902 zero_cpos += num_clusters;
904 if (!zero_clusters) {
909 while ((zero_cpos + zero_clusters) < last_cpos) {
910 rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
911 &p_cpos, &num_clusters,
918 if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
920 if (ext_flags & OCFS2_EXT_REFCOUNTED)
922 zero_clusters += num_clusters;
924 if ((zero_cpos + zero_clusters) > last_cpos)
925 zero_clusters = last_cpos - zero_cpos;
928 rc = ocfs2_refcount_cow(inode, di_bh, zero_cpos,
929 zero_clusters, UINT_MAX);
936 *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
937 *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
938 zero_cpos + zero_clusters);
945 * Zero one range returned from ocfs2_zero_extend_get_range(). The caller
946 * has made sure that the entire range needs zeroing.
948 static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
949 u64 range_end, struct buffer_head *di_bh)
953 u64 zero_pos = range_start;
955 trace_ocfs2_zero_extend_range(
956 (unsigned long long)OCFS2_I(inode)->ip_blkno,
957 (unsigned long long)range_start,
958 (unsigned long long)range_end);
959 BUG_ON(range_start >= range_end);
961 while (zero_pos < range_end) {
962 next_pos = (zero_pos & PAGE_CACHE_MASK) + PAGE_CACHE_SIZE;
963 if (next_pos > range_end)
964 next_pos = range_end;
965 rc = ocfs2_write_zero_page(inode, zero_pos, next_pos, di_bh);
973 * Very large extends have the potential to lock up
974 * the cpu for extended periods of time.
982 int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
986 u64 zero_start, range_start = 0, range_end = 0;
987 struct super_block *sb = inode->i_sb;
989 zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
990 trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
991 (unsigned long long)zero_start,
992 (unsigned long long)i_size_read(inode));
993 while (zero_start < zero_to_size) {
994 ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
1005 if (range_start < zero_start)
1006 range_start = zero_start;
1007 if (range_end > zero_to_size)
1008 range_end = zero_to_size;
1010 ret = ocfs2_zero_extend_range(inode, range_start,
1016 zero_start = range_end;
1022 int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
1023 u64 new_i_size, u64 zero_to)
1026 u32 clusters_to_add;
1027 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1030 * Only quota files call this without a bh, and they can't be
1033 BUG_ON(!di_bh && (oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
1034 BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1036 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1037 if (clusters_to_add < oi->ip_clusters)
1038 clusters_to_add = 0;
1040 clusters_to_add -= oi->ip_clusters;
1042 if (clusters_to_add) {
1043 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
1044 clusters_to_add, 0);
1052 * Call this even if we don't add any clusters to the tree. We
1053 * still need to zero the area between the old i_size and the
1056 ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1064 static int ocfs2_extend_file(struct inode *inode,
1065 struct buffer_head *di_bh,
1069 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1073 /* setattr sometimes calls us like this. */
1074 if (new_i_size == 0)
1077 if (i_size_read(inode) == new_i_size)
1079 BUG_ON(new_i_size < i_size_read(inode));
1082 * The alloc sem blocks people in read/write from reading our
1083 * allocation until we're done changing it. We depend on
1084 * i_mutex to block other extend/truncate calls while we're
1085 * here. We even have to hold it for sparse files because there
1086 * might be some tail zeroing.
1088 down_write(&oi->ip_alloc_sem);
1090 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1092 * We can optimize small extends by keeping the inodes
1095 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1096 up_write(&oi->ip_alloc_sem);
1097 goto out_update_size;
1100 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1102 up_write(&oi->ip_alloc_sem);
1108 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1109 ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1111 ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1114 up_write(&oi->ip_alloc_sem);
1122 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1130 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1132 int status = 0, size_change;
1133 struct inode *inode = d_inode(dentry);
1134 struct super_block *sb = inode->i_sb;
1135 struct ocfs2_super *osb = OCFS2_SB(sb);
1136 struct buffer_head *bh = NULL;
1137 handle_t *handle = NULL;
1138 struct dquot *transfer_to[MAXQUOTAS] = { };
1141 trace_ocfs2_setattr(inode, dentry,
1142 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1143 dentry->d_name.len, dentry->d_name.name,
1144 attr->ia_valid, attr->ia_mode,
1145 from_kuid(&init_user_ns, attr->ia_uid),
1146 from_kgid(&init_user_ns, attr->ia_gid));
1148 /* ensuring we don't even attempt to truncate a symlink */
1149 if (S_ISLNK(inode->i_mode))
1150 attr->ia_valid &= ~ATTR_SIZE;
1152 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1153 | ATTR_GID | ATTR_UID | ATTR_MODE)
1154 if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
1157 status = inode_change_ok(inode, attr);
1161 if (is_quota_modification(inode, attr)) {
1162 status = dquot_initialize(inode);
1166 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1168 status = ocfs2_rw_lock(inode, 1);
1175 status = ocfs2_inode_lock(inode, &bh, 1);
1177 if (status != -ENOENT)
1179 goto bail_unlock_rw;
1183 status = inode_newsize_ok(inode, attr->ia_size);
1187 inode_dio_wait(inode);
1189 if (i_size_read(inode) >= attr->ia_size) {
1190 if (ocfs2_should_order_data(inode)) {
1191 status = ocfs2_begin_ordered_truncate(inode,
1196 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1198 status = ocfs2_extend_file(inode, bh, attr->ia_size);
1200 if (status != -ENOSPC)
1207 if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
1208 (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
1210 * Gather pointers to quota structures so that allocation /
1211 * freeing of quota structures happens here and not inside
1212 * dquot_transfer() where we have problems with lock ordering
1214 if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)
1215 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1216 OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1217 transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid));
1218 if (IS_ERR(transfer_to[USRQUOTA])) {
1219 status = PTR_ERR(transfer_to[USRQUOTA]);
1223 if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid)
1224 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1225 OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1226 transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid));
1227 if (IS_ERR(transfer_to[GRPQUOTA])) {
1228 status = PTR_ERR(transfer_to[GRPQUOTA]);
1232 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1233 2 * ocfs2_quota_trans_credits(sb));
1234 if (IS_ERR(handle)) {
1235 status = PTR_ERR(handle);
1239 status = __dquot_transfer(inode, transfer_to);
1243 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1244 if (IS_ERR(handle)) {
1245 status = PTR_ERR(handle);
1251 setattr_copy(inode, attr);
1252 mark_inode_dirty(inode);
1254 status = ocfs2_mark_inode_dirty(handle, inode, bh);
1259 ocfs2_commit_trans(osb, handle);
1261 ocfs2_inode_unlock(inode, 1);
1264 ocfs2_rw_unlock(inode, 1);
1268 /* Release quota pointers in case we acquired them */
1269 for (qtype = 0; qtype < OCFS2_MAXQUOTAS; qtype++)
1270 dqput(transfer_to[qtype]);
1272 if (!status && attr->ia_valid & ATTR_MODE) {
1273 status = posix_acl_chmod(inode, inode->i_mode);
1281 int ocfs2_getattr(struct vfsmount *mnt,
1282 struct dentry *dentry,
1285 struct inode *inode = d_inode(dentry);
1286 struct super_block *sb = d_inode(dentry)->i_sb;
1287 struct ocfs2_super *osb = sb->s_fs_info;
1290 err = ocfs2_inode_revalidate(dentry);
1297 generic_fillattr(inode, stat);
1299 /* We set the blksize from the cluster size for performance */
1300 stat->blksize = osb->s_clustersize;
1306 int ocfs2_permission(struct inode *inode, int mask)
1310 if (mask & MAY_NOT_BLOCK)
1313 ret = ocfs2_inode_lock(inode, NULL, 0);
1320 ret = generic_permission(inode, mask);
1322 ocfs2_inode_unlock(inode, 0);
1327 static int __ocfs2_write_remove_suid(struct inode *inode,
1328 struct buffer_head *bh)
1332 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1333 struct ocfs2_dinode *di;
1335 trace_ocfs2_write_remove_suid(
1336 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1339 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1340 if (IS_ERR(handle)) {
1341 ret = PTR_ERR(handle);
1346 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1347 OCFS2_JOURNAL_ACCESS_WRITE);
1353 inode->i_mode &= ~S_ISUID;
1354 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1355 inode->i_mode &= ~S_ISGID;
1357 di = (struct ocfs2_dinode *) bh->b_data;
1358 di->i_mode = cpu_to_le16(inode->i_mode);
1359 ocfs2_update_inode_fsync_trans(handle, inode, 0);
1361 ocfs2_journal_dirty(handle, bh);
1364 ocfs2_commit_trans(osb, handle);
1370 * Will look for holes and unwritten extents in the range starting at
1371 * pos for count bytes (inclusive).
1373 static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1377 unsigned int extent_flags;
1378 u32 cpos, clusters, extent_len, phys_cpos;
1379 struct super_block *sb = inode->i_sb;
1381 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1382 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1385 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1392 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
1397 if (extent_len > clusters)
1398 extent_len = clusters;
1400 clusters -= extent_len;
1407 static int ocfs2_write_remove_suid(struct inode *inode)
1410 struct buffer_head *bh = NULL;
1412 ret = ocfs2_read_inode_block(inode, &bh);
1418 ret = __ocfs2_write_remove_suid(inode, bh);
1425 * Allocate enough extents to cover the region starting at byte offset
1426 * start for len bytes. Existing extents are skipped, any extents
1427 * added are marked as "unwritten".
1429 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1433 u32 cpos, phys_cpos, clusters, alloc_size;
1434 u64 end = start + len;
1435 struct buffer_head *di_bh = NULL;
1437 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1438 ret = ocfs2_read_inode_block(inode, &di_bh);
1445 * Nothing to do if the requested reservation range
1446 * fits within the inode.
1448 if (ocfs2_size_fits_inline_data(di_bh, end))
1451 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1459 * We consider both start and len to be inclusive.
1461 cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1462 clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1466 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1474 * Hole or existing extent len can be arbitrary, so
1475 * cap it to our own allocation request.
1477 if (alloc_size > clusters)
1478 alloc_size = clusters;
1482 * We already have an allocation at this
1483 * region so we can safely skip it.
1488 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1497 clusters -= alloc_size;
1508 * Truncate a byte range, avoiding pages within partial clusters. This
1509 * preserves those pages for the zeroing code to write to.
1511 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1514 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1516 struct address_space *mapping = inode->i_mapping;
1518 start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1519 end = byte_start + byte_len;
1520 end = end & ~(osb->s_clustersize - 1);
1523 unmap_mapping_range(mapping, start, end - start, 0);
1524 truncate_inode_pages_range(mapping, start, end - 1);
1528 static int ocfs2_zero_partial_clusters(struct inode *inode,
1532 u64 tmpend, end = start + len;
1533 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1534 unsigned int csize = osb->s_clustersize;
1538 * The "start" and "end" values are NOT necessarily part of
1539 * the range whose allocation is being deleted. Rather, this
1540 * is what the user passed in with the request. We must zero
1541 * partial clusters here. There's no need to worry about
1542 * physical allocation - the zeroing code knows to skip holes.
1544 trace_ocfs2_zero_partial_clusters(
1545 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1546 (unsigned long long)start, (unsigned long long)end);
1549 * If both edges are on a cluster boundary then there's no
1550 * zeroing required as the region is part of the allocation to
1553 if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1556 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1557 if (IS_ERR(handle)) {
1558 ret = PTR_ERR(handle);
1564 * We want to get the byte offset of the end of the 1st cluster.
1566 tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1570 trace_ocfs2_zero_partial_clusters_range1((unsigned long long)start,
1571 (unsigned long long)tmpend);
1573 ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1579 * This may make start and end equal, but the zeroing
1580 * code will skip any work in that case so there's no
1581 * need to catch it up here.
1583 start = end & ~(osb->s_clustersize - 1);
1585 trace_ocfs2_zero_partial_clusters_range2(
1586 (unsigned long long)start, (unsigned long long)end);
1588 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1592 ocfs2_update_inode_fsync_trans(handle, inode, 1);
1594 ocfs2_commit_trans(osb, handle);
1599 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1602 struct ocfs2_extent_rec *rec = NULL;
1604 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1606 rec = &el->l_recs[i];
1608 if (le32_to_cpu(rec->e_cpos) < pos)
1616 * Helper to calculate the punching pos and length in one run, we handle the
1617 * following three cases in order:
1619 * - remove the entire record
1620 * - remove a partial record
1621 * - no record needs to be removed (hole-punching completed)
1623 static void ocfs2_calc_trunc_pos(struct inode *inode,
1624 struct ocfs2_extent_list *el,
1625 struct ocfs2_extent_rec *rec,
1626 u32 trunc_start, u32 *trunc_cpos,
1627 u32 *trunc_len, u32 *trunc_end,
1628 u64 *blkno, int *done)
1633 range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1635 if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1637 * remove an entire extent record.
1639 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1641 * Skip holes if any.
1643 if (range < *trunc_end)
1645 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1646 *blkno = le64_to_cpu(rec->e_blkno);
1647 *trunc_end = le32_to_cpu(rec->e_cpos);
1648 } else if (range > trunc_start) {
1650 * remove a partial extent record, which means we're
1651 * removing the last extent record.
1653 *trunc_cpos = trunc_start;
1657 if (range < *trunc_end)
1659 *trunc_len = *trunc_end - trunc_start;
1660 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1661 *blkno = le64_to_cpu(rec->e_blkno) +
1662 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1663 *trunc_end = trunc_start;
1666 * It may have two following possibilities:
1668 * - last record has been removed
1669 * - trunc_start was within a hole
1671 * both two cases mean the completion of hole punching.
1679 static int ocfs2_remove_inode_range(struct inode *inode,
1680 struct buffer_head *di_bh, u64 byte_start,
1683 int ret = 0, flags = 0, done = 0, i;
1684 u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1686 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1687 struct ocfs2_cached_dealloc_ctxt dealloc;
1688 struct address_space *mapping = inode->i_mapping;
1689 struct ocfs2_extent_tree et;
1690 struct ocfs2_path *path = NULL;
1691 struct ocfs2_extent_list *el = NULL;
1692 struct ocfs2_extent_rec *rec = NULL;
1693 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1694 u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1696 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1697 ocfs2_init_dealloc_ctxt(&dealloc);
1699 trace_ocfs2_remove_inode_range(
1700 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1701 (unsigned long long)byte_start,
1702 (unsigned long long)byte_len);
1707 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1708 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1709 byte_start + byte_len, 0);
1715 * There's no need to get fancy with the page cache
1716 * truncate of an inline-data inode. We're talking
1717 * about less than a page here, which will be cached
1718 * in the dinode buffer anyway.
1720 unmap_mapping_range(mapping, 0, 0, 0);
1721 truncate_inode_pages(mapping, 0);
1726 * For reflinks, we may need to CoW 2 clusters which might be
1727 * partially zero'd later, if hole's start and end offset were
1728 * within one cluster(means is not exactly aligned to clustersize).
1731 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
1733 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1739 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1746 trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1747 trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1748 cluster_in_el = trunc_end;
1750 ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1756 path = ocfs2_new_path_from_et(&et);
1763 while (trunc_end > trunc_start) {
1765 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1772 el = path_leaf_el(path);
1774 i = ocfs2_find_rec(el, trunc_end);
1776 * Need to go to previous extent block.
1779 if (path->p_tree_depth == 0)
1782 ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1791 * We've reached the leftmost extent block,
1792 * it's safe to leave.
1794 if (cluster_in_el == 0)
1798 * The 'pos' searched for previous extent block is
1799 * always one cluster less than actual trunc_end.
1801 trunc_end = cluster_in_el + 1;
1803 ocfs2_reinit_path(path, 1);
1808 rec = &el->l_recs[i];
1810 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1811 &trunc_len, &trunc_end, &blkno, &done);
1815 flags = rec->e_flags;
1816 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1818 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1819 phys_cpos, trunc_len, flags,
1820 &dealloc, refcount_loc, false);
1826 cluster_in_el = trunc_end;
1828 ocfs2_reinit_path(path, 1);
1831 ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1834 ocfs2_free_path(path);
1835 ocfs2_schedule_truncate_log_flush(osb, 1);
1836 ocfs2_run_deallocs(osb, &dealloc);
1842 * Parts of this function taken from xfs_change_file_space()
1844 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1845 loff_t f_pos, unsigned int cmd,
1846 struct ocfs2_space_resv *sr,
1852 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1853 struct buffer_head *di_bh = NULL;
1855 unsigned long long max_off = inode->i_sb->s_maxbytes;
1857 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1860 mutex_lock(&inode->i_mutex);
1863 * This prevents concurrent writes on other nodes
1865 ret = ocfs2_rw_lock(inode, 1);
1871 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1877 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1879 goto out_inode_unlock;
1882 switch (sr->l_whence) {
1883 case 0: /*SEEK_SET*/
1885 case 1: /*SEEK_CUR*/
1886 sr->l_start += f_pos;
1888 case 2: /*SEEK_END*/
1889 sr->l_start += i_size_read(inode);
1893 goto out_inode_unlock;
1897 llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1900 || sr->l_start > max_off
1901 || (sr->l_start + llen) < 0
1902 || (sr->l_start + llen) > max_off) {
1904 goto out_inode_unlock;
1906 size = sr->l_start + sr->l_len;
1908 if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64 ||
1909 cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) {
1910 if (sr->l_len <= 0) {
1912 goto out_inode_unlock;
1916 if (file && should_remove_suid(file->f_path.dentry)) {
1917 ret = __ocfs2_write_remove_suid(inode, di_bh);
1920 goto out_inode_unlock;
1924 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1926 case OCFS2_IOC_RESVSP:
1927 case OCFS2_IOC_RESVSP64:
1929 * This takes unsigned offsets, but the signed ones we
1930 * pass have been checked against overflow above.
1932 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1935 case OCFS2_IOC_UNRESVSP:
1936 case OCFS2_IOC_UNRESVSP64:
1937 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1943 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1946 goto out_inode_unlock;
1950 * We update c/mtime for these changes
1952 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1953 if (IS_ERR(handle)) {
1954 ret = PTR_ERR(handle);
1956 goto out_inode_unlock;
1959 if (change_size && i_size_read(inode) < size)
1960 i_size_write(inode, size);
1962 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1963 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1967 if (file && (file->f_flags & O_SYNC))
1970 ocfs2_commit_trans(osb, handle);
1974 ocfs2_inode_unlock(inode, 1);
1976 ocfs2_rw_unlock(inode, 1);
1979 mutex_unlock(&inode->i_mutex);
1983 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1984 struct ocfs2_space_resv *sr)
1986 struct inode *inode = file_inode(file);
1987 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1990 if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1991 !ocfs2_writes_unwritten_extents(osb))
1993 else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1994 !ocfs2_sparse_alloc(osb))
1997 if (!S_ISREG(inode->i_mode))
2000 if (!(file->f_mode & FMODE_WRITE))
2003 ret = mnt_want_write_file(file);
2006 ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
2007 mnt_drop_write_file(file);
2011 static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
2014 struct inode *inode = file_inode(file);
2015 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2016 struct ocfs2_space_resv sr;
2017 int change_size = 1;
2018 int cmd = OCFS2_IOC_RESVSP64;
2020 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2022 if (!ocfs2_writes_unwritten_extents(osb))
2025 if (mode & FALLOC_FL_KEEP_SIZE)
2028 if (mode & FALLOC_FL_PUNCH_HOLE)
2029 cmd = OCFS2_IOC_UNRESVSP64;
2032 sr.l_start = (s64)offset;
2033 sr.l_len = (s64)len;
2035 return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2039 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2043 unsigned int extent_flags;
2044 u32 cpos, clusters, extent_len, phys_cpos;
2045 struct super_block *sb = inode->i_sb;
2047 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2048 !(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) ||
2049 OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2052 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2053 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2056 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2063 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2068 if (extent_len > clusters)
2069 extent_len = clusters;
2071 clusters -= extent_len;
2078 static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
2080 int blockmask = inode->i_sb->s_blocksize - 1;
2081 loff_t final_size = pos + count;
2083 if ((pos & blockmask) || (final_size & blockmask))
2088 static int ocfs2_prepare_inode_for_refcount(struct inode *inode,
2090 loff_t pos, size_t count,
2094 struct buffer_head *di_bh = NULL;
2095 u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2097 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2099 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2107 ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
2115 static int ocfs2_prepare_inode_for_write(struct file *file,
2122 int ret = 0, meta_level = 0;
2123 struct dentry *dentry = file->f_path.dentry;
2124 struct inode *inode = d_inode(dentry);
2126 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2127 int full_coherency = !(osb->s_mount_opt &
2128 OCFS2_MOUNT_COHERENCY_BUFFERED);
2131 * We start with a read level meta lock and only jump to an ex
2132 * if we need to make modifications here.
2135 ret = ocfs2_inode_lock(inode, NULL, meta_level);
2142 /* Clear suid / sgid if necessary. We do this here
2143 * instead of later in the write path because
2144 * remove_suid() calls ->setattr without any hint that
2145 * we may have already done our cluster locking. Since
2146 * ocfs2_setattr() *must* take cluster locks to
2147 * proceed, this will lead us to recursively lock the
2148 * inode. There's also the dinode i_size state which
2149 * can be lost via setattr during extending writes (we
2150 * set inode->i_size at the end of a write. */
2151 if (should_remove_suid(dentry)) {
2152 if (meta_level == 0) {
2153 ocfs2_inode_unlock(inode, meta_level);
2158 ret = ocfs2_write_remove_suid(inode);
2167 ret = ocfs2_check_range_for_refcount(inode, pos, count);
2169 ocfs2_inode_unlock(inode, meta_level);
2172 ret = ocfs2_prepare_inode_for_refcount(inode,
2189 * Skip the O_DIRECT checks if we don't need
2192 if (!direct_io || !(*direct_io))
2196 * There's no sane way to do direct writes to an inode
2199 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2205 * Allowing concurrent direct writes means
2206 * i_size changes wouldn't be synchronized, so
2207 * one node could wind up truncating another
2210 if (end > i_size_read(inode) && !full_coherency) {
2216 * Fallback to old way if the feature bit is not set.
2218 if (end > i_size_read(inode) &&
2219 !ocfs2_supports_append_dio(osb)) {
2225 * We don't fill holes during direct io, so
2226 * check for them here. If any are found, the
2227 * caller will have to retake some cluster
2228 * locks and initiate the io as buffered.
2230 ret = ocfs2_check_range_for_holes(inode, pos, count);
2233 * Fallback to old way if the feature bit is not set.
2234 * Otherwise try dio first and then complete the rest
2235 * request through buffer io.
2237 if (!ocfs2_supports_append_dio(osb))
2246 trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
2247 pos, appending, count,
2248 direct_io, has_refcount);
2250 if (meta_level >= 0)
2251 ocfs2_inode_unlock(inode, meta_level);
2257 static ssize_t ocfs2_file_write_iter(struct kiocb *iocb,
2258 struct iov_iter *from)
2260 int direct_io, appending, rw_level;
2261 int can_do_direct, has_refcount = 0;
2262 ssize_t written = 0;
2264 size_t count = iov_iter_count(from), orig_count;
2267 struct file *file = iocb->ki_filp;
2268 struct inode *inode = file_inode(file);
2269 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2270 int full_coherency = !(osb->s_mount_opt &
2271 OCFS2_MOUNT_COHERENCY_BUFFERED);
2272 int unaligned_dio = 0;
2273 int dropped_dio = 0;
2275 trace_ocfs2_file_aio_write(inode, file, file->f_path.dentry,
2276 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2277 file->f_path.dentry->d_name.len,
2278 file->f_path.dentry->d_name.name,
2279 (unsigned int)from->nr_segs); /* GRRRRR */
2284 appending = iocb->ki_flags & IOCB_APPEND ? 1 : 0;
2285 direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2287 mutex_lock(&inode->i_mutex);
2291 * Concurrent O_DIRECT writes are allowed with
2292 * mount_option "coherency=buffered".
2294 rw_level = (!direct_io || full_coherency);
2296 ret = ocfs2_rw_lock(inode, rw_level);
2303 * O_DIRECT writes with "coherency=full" need to take EX cluster
2304 * inode_lock to guarantee coherency.
2306 if (direct_io && full_coherency) {
2308 * We need to take and drop the inode lock to force
2309 * other nodes to drop their caches. Buffered I/O
2310 * already does this in write_begin().
2312 ret = ocfs2_inode_lock(inode, NULL, 1);
2318 ocfs2_inode_unlock(inode, 1);
2321 orig_count = iov_iter_count(from);
2322 ret = generic_write_checks(iocb, from);
2330 can_do_direct = direct_io;
2331 ret = ocfs2_prepare_inode_for_write(file, iocb->ki_pos, count, appending,
2332 &can_do_direct, &has_refcount);
2338 if (direct_io && !is_sync_kiocb(iocb))
2339 unaligned_dio = ocfs2_is_io_unaligned(inode, count, iocb->ki_pos);
2342 * We can't complete the direct I/O as requested, fall back to
2345 if (direct_io && !can_do_direct) {
2346 ocfs2_rw_unlock(inode, rw_level);
2351 iocb->ki_flags &= ~IOCB_DIRECT;
2352 iov_iter_reexpand(from, orig_count);
2357 if (unaligned_dio) {
2359 * Wait on previous unaligned aio to complete before
2362 mutex_lock(&OCFS2_I(inode)->ip_unaligned_aio);
2363 /* Mark the iocb as needing an unlock in ocfs2_dio_end_io */
2364 ocfs2_iocb_set_unaligned_aio(iocb);
2368 * To later detect whether a journal commit for sync writes is
2369 * necessary, we sample i_size, and cluster count here.
2371 old_size = i_size_read(inode);
2372 old_clusters = OCFS2_I(inode)->ip_clusters;
2374 /* communicate with ocfs2_dio_end_io */
2375 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2377 written = __generic_file_write_iter(iocb, from);
2378 /* buffered aio wouldn't have proper lock coverage today */
2379 BUG_ON(written == -EIOCBQUEUED && !(iocb->ki_flags & IOCB_DIRECT));
2381 if (unlikely(written <= 0))
2384 if (((file->f_flags & O_DSYNC) && !direct_io) ||
2385 IS_SYNC(inode) || dropped_dio) {
2386 ret = filemap_fdatawrite_range(file->f_mapping,
2387 iocb->ki_pos - written,
2393 ret = jbd2_journal_force_commit(osb->journal->j_journal);
2399 ret = filemap_fdatawait_range(file->f_mapping,
2400 iocb->ki_pos - written,
2406 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2407 * function pointer which is called when o_direct io completes so that
2408 * it can unlock our rw lock.
2409 * Unfortunately there are error cases which call end_io and others
2410 * that don't. so we don't have to unlock the rw_lock if either an
2411 * async dio is going to do it in the future or an end_io after an
2412 * error has already done it.
2414 if ((ret == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2419 if (unaligned_dio) {
2420 ocfs2_iocb_clear_unaligned_aio(iocb);
2421 mutex_unlock(&OCFS2_I(inode)->ip_unaligned_aio);
2426 ocfs2_rw_unlock(inode, rw_level);
2429 mutex_unlock(&inode->i_mutex);
2436 static ssize_t ocfs2_file_splice_read(struct file *in,
2438 struct pipe_inode_info *pipe,
2442 int ret = 0, lock_level = 0;
2443 struct inode *inode = file_inode(in);
2445 trace_ocfs2_file_splice_read(inode, in, in->f_path.dentry,
2446 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2447 in->f_path.dentry->d_name.len,
2448 in->f_path.dentry->d_name.name, len);
2451 * See the comment in ocfs2_file_read_iter()
2453 ret = ocfs2_inode_lock_atime(inode, in->f_path.mnt, &lock_level);
2458 ocfs2_inode_unlock(inode, lock_level);
2460 ret = generic_file_splice_read(in, ppos, pipe, len, flags);
2466 static ssize_t ocfs2_file_read_iter(struct kiocb *iocb,
2467 struct iov_iter *to)
2469 int ret = 0, rw_level = -1, lock_level = 0;
2470 struct file *filp = iocb->ki_filp;
2471 struct inode *inode = file_inode(filp);
2473 trace_ocfs2_file_aio_read(inode, filp, filp->f_path.dentry,
2474 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2475 filp->f_path.dentry->d_name.len,
2476 filp->f_path.dentry->d_name.name,
2477 to->nr_segs); /* GRRRRR */
2487 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
2488 * need locks to protect pending reads from racing with truncate.
2490 if (iocb->ki_flags & IOCB_DIRECT) {
2491 ret = ocfs2_rw_lock(inode, 0);
2497 /* communicate with ocfs2_dio_end_io */
2498 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2502 * We're fine letting folks race truncates and extending
2503 * writes with read across the cluster, just like they can
2504 * locally. Hence no rw_lock during read.
2506 * Take and drop the meta data lock to update inode fields
2507 * like i_size. This allows the checks down below
2508 * generic_file_aio_read() a chance of actually working.
2510 ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level);
2515 ocfs2_inode_unlock(inode, lock_level);
2517 ret = generic_file_read_iter(iocb, to);
2518 trace_generic_file_aio_read_ret(ret);
2520 /* buffered aio wouldn't have proper lock coverage today */
2521 BUG_ON(ret == -EIOCBQUEUED && !(iocb->ki_flags & IOCB_DIRECT));
2523 /* see ocfs2_file_write_iter */
2524 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2530 ocfs2_rw_unlock(inode, rw_level);
2535 /* Refer generic_file_llseek_unlocked() */
2536 static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence)
2538 struct inode *inode = file->f_mapping->host;
2541 mutex_lock(&inode->i_mutex);
2547 /* SEEK_END requires the OCFS2 inode lock for the file
2548 * because it references the file's size.
2550 ret = ocfs2_inode_lock(inode, NULL, 0);
2555 offset += i_size_read(inode);
2556 ocfs2_inode_unlock(inode, 0);
2560 offset = file->f_pos;
2563 offset += file->f_pos;
2567 ret = ocfs2_seek_data_hole_offset(file, &offset, whence);
2576 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2579 mutex_unlock(&inode->i_mutex);
2585 const struct inode_operations ocfs2_file_iops = {
2586 .setattr = ocfs2_setattr,
2587 .getattr = ocfs2_getattr,
2588 .permission = ocfs2_permission,
2589 .setxattr = generic_setxattr,
2590 .getxattr = generic_getxattr,
2591 .listxattr = ocfs2_listxattr,
2592 .removexattr = generic_removexattr,
2593 .fiemap = ocfs2_fiemap,
2594 .get_acl = ocfs2_iop_get_acl,
2595 .set_acl = ocfs2_iop_set_acl,
2598 const struct inode_operations ocfs2_special_file_iops = {
2599 .setattr = ocfs2_setattr,
2600 .getattr = ocfs2_getattr,
2601 .permission = ocfs2_permission,
2602 .get_acl = ocfs2_iop_get_acl,
2603 .set_acl = ocfs2_iop_set_acl,
2607 * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2608 * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2610 const struct file_operations ocfs2_fops = {
2611 .llseek = ocfs2_file_llseek,
2613 .fsync = ocfs2_sync_file,
2614 .release = ocfs2_file_release,
2615 .open = ocfs2_file_open,
2616 .read_iter = ocfs2_file_read_iter,
2617 .write_iter = ocfs2_file_write_iter,
2618 .unlocked_ioctl = ocfs2_ioctl,
2619 #ifdef CONFIG_COMPAT
2620 .compat_ioctl = ocfs2_compat_ioctl,
2623 .flock = ocfs2_flock,
2624 .splice_read = ocfs2_file_splice_read,
2625 .splice_write = iter_file_splice_write,
2626 .fallocate = ocfs2_fallocate,
2629 const struct file_operations ocfs2_dops = {
2630 .llseek = generic_file_llseek,
2631 .read = generic_read_dir,
2632 .iterate = ocfs2_readdir,
2633 .fsync = ocfs2_sync_file,
2634 .release = ocfs2_dir_release,
2635 .open = ocfs2_dir_open,
2636 .unlocked_ioctl = ocfs2_ioctl,
2637 #ifdef CONFIG_COMPAT
2638 .compat_ioctl = ocfs2_compat_ioctl,
2641 .flock = ocfs2_flock,
2645 * POSIX-lockless variants of our file_operations.
2647 * These will be used if the underlying cluster stack does not support
2648 * posix file locking, if the user passes the "localflocks" mount
2649 * option, or if we have a local-only fs.
2651 * ocfs2_flock is in here because all stacks handle UNIX file locks,
2652 * so we still want it in the case of no stack support for
2653 * plocks. Internally, it will do the right thing when asked to ignore
2656 const struct file_operations ocfs2_fops_no_plocks = {
2657 .llseek = ocfs2_file_llseek,
2659 .fsync = ocfs2_sync_file,
2660 .release = ocfs2_file_release,
2661 .open = ocfs2_file_open,
2662 .read_iter = ocfs2_file_read_iter,
2663 .write_iter = ocfs2_file_write_iter,
2664 .unlocked_ioctl = ocfs2_ioctl,
2665 #ifdef CONFIG_COMPAT
2666 .compat_ioctl = ocfs2_compat_ioctl,
2668 .flock = ocfs2_flock,
2669 .splice_read = ocfs2_file_splice_read,
2670 .splice_write = iter_file_splice_write,
2671 .fallocate = ocfs2_fallocate,
2674 const struct file_operations ocfs2_dops_no_plocks = {
2675 .llseek = generic_file_llseek,
2676 .read = generic_read_dir,
2677 .iterate = ocfs2_readdir,
2678 .fsync = ocfs2_sync_file,
2679 .release = ocfs2_dir_release,
2680 .open = ocfs2_dir_open,
2681 .unlocked_ioctl = ocfs2_ioctl,
2682 #ifdef CONFIG_COMPAT
2683 .compat_ioctl = ocfs2_compat_ioctl,
2685 .flock = ocfs2_flock,