1 // SPDX-License-Identifier: GPL-2.0-or-later
5 * File open, close, extend, truncate
7 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
10 #include <linux/capability.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>
26 #include <cluster/masklog.h>
34 #include "extent_map.h"
47 #include "refcounttree.h"
48 #include "ocfs2_trace.h"
50 #include "buffer_head_io.h"
52 static int ocfs2_init_file_private(struct inode *inode, struct file *file)
54 struct ocfs2_file_private *fp;
56 fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
61 mutex_init(&fp->fp_mutex);
62 ocfs2_file_lock_res_init(&fp->fp_flock, fp);
63 file->private_data = fp;
68 static void ocfs2_free_file_private(struct inode *inode, struct file *file)
70 struct ocfs2_file_private *fp = file->private_data;
71 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
74 ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
75 ocfs2_lock_res_free(&fp->fp_flock);
77 file->private_data = NULL;
81 static int ocfs2_file_open(struct inode *inode, struct file *file)
84 int mode = file->f_flags;
85 struct ocfs2_inode_info *oi = OCFS2_I(inode);
87 trace_ocfs2_file_open(inode, file, file->f_path.dentry,
88 (unsigned long long)oi->ip_blkno,
89 file->f_path.dentry->d_name.len,
90 file->f_path.dentry->d_name.name, mode);
92 if (file->f_mode & FMODE_WRITE) {
93 status = dquot_initialize(inode);
98 spin_lock(&oi->ip_lock);
100 /* Check that the inode hasn't been wiped from disk by another
101 * node. If it hasn't then we're safe as long as we hold the
102 * spin lock until our increment of open count. */
103 if (oi->ip_flags & OCFS2_INODE_DELETED) {
104 spin_unlock(&oi->ip_lock);
111 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
114 spin_unlock(&oi->ip_lock);
116 status = ocfs2_init_file_private(inode, file);
119 * We want to set open count back if we're failing the
122 spin_lock(&oi->ip_lock);
124 spin_unlock(&oi->ip_lock);
127 file->f_mode |= FMODE_NOWAIT;
133 static int ocfs2_file_release(struct inode *inode, struct file *file)
135 struct ocfs2_inode_info *oi = OCFS2_I(inode);
137 spin_lock(&oi->ip_lock);
138 if (!--oi->ip_open_count)
139 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
141 trace_ocfs2_file_release(inode, file, file->f_path.dentry,
143 file->f_path.dentry->d_name.len,
144 file->f_path.dentry->d_name.name,
146 spin_unlock(&oi->ip_lock);
148 ocfs2_free_file_private(inode, file);
153 static int ocfs2_dir_open(struct inode *inode, struct file *file)
155 return ocfs2_init_file_private(inode, file);
158 static int ocfs2_dir_release(struct inode *inode, struct file *file)
160 ocfs2_free_file_private(inode, file);
164 static int ocfs2_sync_file(struct file *file, loff_t start, loff_t end,
168 struct inode *inode = file->f_mapping->host;
169 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
170 struct ocfs2_inode_info *oi = OCFS2_I(inode);
171 journal_t *journal = osb->journal->j_journal;
174 bool needs_barrier = false;
176 trace_ocfs2_sync_file(inode, file, file->f_path.dentry,
178 file->f_path.dentry->d_name.len,
179 file->f_path.dentry->d_name.name,
180 (unsigned long long)datasync);
182 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
185 err = file_write_and_wait_range(file, start, end);
189 commit_tid = datasync ? oi->i_datasync_tid : oi->i_sync_tid;
190 if (journal->j_flags & JBD2_BARRIER &&
191 !jbd2_trans_will_send_data_barrier(journal, commit_tid))
192 needs_barrier = true;
193 err = jbd2_complete_transaction(journal, commit_tid);
195 ret = blkdev_issue_flush(inode->i_sb->s_bdev);
203 return (err < 0) ? -EIO : 0;
206 int ocfs2_should_update_atime(struct inode *inode,
207 struct vfsmount *vfsmnt)
209 struct timespec64 now;
210 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
212 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
215 if ((inode->i_flags & S_NOATIME) ||
216 ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode)))
220 * We can be called with no vfsmnt structure - NFSD will
223 * Note that our action here is different than touch_atime() -
224 * if we can't tell whether this is a noatime mount, then we
225 * don't know whether to trust the value of s_atime_quantum.
230 if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
231 ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
234 if (vfsmnt->mnt_flags & MNT_RELATIME) {
235 if ((timespec64_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
236 (timespec64_compare(&inode->i_atime, &inode->i_ctime) <= 0))
242 now = current_time(inode);
243 if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
249 int ocfs2_update_inode_atime(struct inode *inode,
250 struct buffer_head *bh)
253 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
255 struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
257 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
258 if (IS_ERR(handle)) {
259 ret = PTR_ERR(handle);
264 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
265 OCFS2_JOURNAL_ACCESS_WRITE);
272 * Don't use ocfs2_mark_inode_dirty() here as we don't always
273 * have i_rwsem to guard against concurrent changes to other
276 inode->i_atime = current_time(inode);
277 di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
278 di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
279 ocfs2_update_inode_fsync_trans(handle, inode, 0);
280 ocfs2_journal_dirty(handle, bh);
283 ocfs2_commit_trans(osb, handle);
288 int ocfs2_set_inode_size(handle_t *handle,
290 struct buffer_head *fe_bh,
295 i_size_write(inode, new_i_size);
296 inode->i_blocks = ocfs2_inode_sector_count(inode);
297 inode->i_ctime = inode->i_mtime = current_time(inode);
299 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
309 int ocfs2_simple_size_update(struct inode *inode,
310 struct buffer_head *di_bh,
314 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
315 handle_t *handle = NULL;
317 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
318 if (IS_ERR(handle)) {
319 ret = PTR_ERR(handle);
324 ret = ocfs2_set_inode_size(handle, inode, di_bh,
329 ocfs2_update_inode_fsync_trans(handle, inode, 0);
330 ocfs2_commit_trans(osb, handle);
335 static int ocfs2_cow_file_pos(struct inode *inode,
336 struct buffer_head *fe_bh,
340 u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
341 unsigned int num_clusters = 0;
342 unsigned int ext_flags = 0;
345 * If the new offset is aligned to the range of the cluster, there is
346 * no space for ocfs2_zero_range_for_truncate to fill, so no need to
349 if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
352 status = ocfs2_get_clusters(inode, cpos, &phys,
353 &num_clusters, &ext_flags);
359 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
362 return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1);
368 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
370 struct buffer_head *fe_bh,
375 struct ocfs2_dinode *di;
379 * We need to CoW the cluster contains the offset if it is reflinked
380 * since we will call ocfs2_zero_range_for_truncate later which will
381 * write "0" from offset to the end of the cluster.
383 status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
389 /* TODO: This needs to actually orphan the inode in this
392 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
393 if (IS_ERR(handle)) {
394 status = PTR_ERR(handle);
399 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
400 OCFS2_JOURNAL_ACCESS_WRITE);
407 * Do this before setting i_size.
409 cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
410 status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
417 i_size_write(inode, new_i_size);
418 inode->i_ctime = inode->i_mtime = current_time(inode);
420 di = (struct ocfs2_dinode *) fe_bh->b_data;
421 di->i_size = cpu_to_le64(new_i_size);
422 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
423 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
424 ocfs2_update_inode_fsync_trans(handle, inode, 0);
426 ocfs2_journal_dirty(handle, fe_bh);
429 ocfs2_commit_trans(osb, handle);
434 int ocfs2_truncate_file(struct inode *inode,
435 struct buffer_head *di_bh,
439 struct ocfs2_dinode *fe = NULL;
440 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
442 /* We trust di_bh because it comes from ocfs2_inode_lock(), which
443 * already validated it */
444 fe = (struct ocfs2_dinode *) di_bh->b_data;
446 trace_ocfs2_truncate_file((unsigned long long)OCFS2_I(inode)->ip_blkno,
447 (unsigned long long)le64_to_cpu(fe->i_size),
448 (unsigned long long)new_i_size);
450 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
451 "Inode %llu, inode i_size = %lld != di "
452 "i_size = %llu, i_flags = 0x%x\n",
453 (unsigned long long)OCFS2_I(inode)->ip_blkno,
455 (unsigned long long)le64_to_cpu(fe->i_size),
456 le32_to_cpu(fe->i_flags));
458 if (new_i_size > le64_to_cpu(fe->i_size)) {
459 trace_ocfs2_truncate_file_error(
460 (unsigned long long)le64_to_cpu(fe->i_size),
461 (unsigned long long)new_i_size);
467 down_write(&OCFS2_I(inode)->ip_alloc_sem);
469 ocfs2_resv_discard(&osb->osb_la_resmap,
470 &OCFS2_I(inode)->ip_la_data_resv);
473 * The inode lock forced other nodes to sync and drop their
474 * pages, which (correctly) happens even if we have a truncate
475 * without allocation change - ocfs2 cluster sizes can be much
476 * greater than page size, so we have to truncate them
480 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
481 unmap_mapping_range(inode->i_mapping,
482 new_i_size + PAGE_SIZE - 1, 0, 1);
483 truncate_inode_pages(inode->i_mapping, new_i_size);
484 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
485 i_size_read(inode), 1);
489 goto bail_unlock_sem;
492 /* alright, we're going to need to do a full blown alloc size
493 * change. Orphan the inode so that recovery can complete the
494 * truncate if necessary. This does the task of marking
496 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
499 goto bail_unlock_sem;
502 unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
503 truncate_inode_pages(inode->i_mapping, new_i_size);
505 status = ocfs2_commit_truncate(osb, inode, di_bh);
508 goto bail_unlock_sem;
511 /* TODO: orphan dir cleanup here. */
513 up_write(&OCFS2_I(inode)->ip_alloc_sem);
516 if (!status && OCFS2_I(inode)->ip_clusters == 0)
517 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
523 * extend file allocation only here.
524 * we'll update all the disk stuff, and oip->alloc_size
526 * expect stuff to be locked, a transaction started and enough data /
527 * metadata reservations in the contexts.
529 * Will return -EAGAIN, and a reason if a restart is needed.
530 * If passed in, *reason will always be set, even in error.
532 int ocfs2_add_inode_data(struct ocfs2_super *osb,
537 struct buffer_head *fe_bh,
539 struct ocfs2_alloc_context *data_ac,
540 struct ocfs2_alloc_context *meta_ac,
541 enum ocfs2_alloc_restarted *reason_ret)
543 struct ocfs2_extent_tree et;
545 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
546 return ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
547 clusters_to_add, mark_unwritten,
548 data_ac, meta_ac, reason_ret);
551 static int ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
552 u32 clusters_to_add, int mark_unwritten)
555 int restart_func = 0;
558 struct buffer_head *bh = NULL;
559 struct ocfs2_dinode *fe = NULL;
560 handle_t *handle = NULL;
561 struct ocfs2_alloc_context *data_ac = NULL;
562 struct ocfs2_alloc_context *meta_ac = NULL;
563 enum ocfs2_alloc_restarted why = RESTART_NONE;
564 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
565 struct ocfs2_extent_tree et;
569 * Unwritten extent only exists for file systems which
572 BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
574 status = ocfs2_read_inode_block(inode, &bh);
579 fe = (struct ocfs2_dinode *) bh->b_data;
582 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
584 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
585 status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
592 credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list);
593 handle = ocfs2_start_trans(osb, credits);
594 if (IS_ERR(handle)) {
595 status = PTR_ERR(handle);
601 restarted_transaction:
602 trace_ocfs2_extend_allocation(
603 (unsigned long long)OCFS2_I(inode)->ip_blkno,
604 (unsigned long long)i_size_read(inode),
605 le32_to_cpu(fe->i_clusters), clusters_to_add,
608 status = dquot_alloc_space_nodirty(inode,
609 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
614 /* reserve a write to the file entry early on - that we if we
615 * run out of credits in the allocation path, we can still
617 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
618 OCFS2_JOURNAL_ACCESS_WRITE);
624 prev_clusters = OCFS2_I(inode)->ip_clusters;
626 status = ocfs2_add_inode_data(osb,
636 if ((status < 0) && (status != -EAGAIN)) {
637 if (status != -ENOSPC)
641 ocfs2_update_inode_fsync_trans(handle, inode, 1);
642 ocfs2_journal_dirty(handle, bh);
644 spin_lock(&OCFS2_I(inode)->ip_lock);
645 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
646 spin_unlock(&OCFS2_I(inode)->ip_lock);
647 /* Release unused quota reservation */
648 dquot_free_space(inode,
649 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
652 if (why != RESTART_NONE && clusters_to_add) {
653 if (why == RESTART_META) {
657 BUG_ON(why != RESTART_TRANS);
659 status = ocfs2_allocate_extend_trans(handle, 1);
661 /* handle still has to be committed at
667 goto restarted_transaction;
671 trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno,
672 le32_to_cpu(fe->i_clusters),
673 (unsigned long long)le64_to_cpu(fe->i_size),
674 OCFS2_I(inode)->ip_clusters,
675 (unsigned long long)i_size_read(inode));
678 if (status < 0 && did_quota)
679 dquot_free_space(inode,
680 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
682 ocfs2_commit_trans(osb, handle);
686 ocfs2_free_alloc_context(data_ac);
690 ocfs2_free_alloc_context(meta_ac);
693 if ((!status) && restart_func) {
704 * While a write will already be ordering the data, a truncate will not.
705 * Thus, we need to explicitly order the zeroed pages.
707 static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode,
708 struct buffer_head *di_bh,
712 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
713 handle_t *handle = NULL;
716 if (!ocfs2_should_order_data(inode))
719 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
720 if (IS_ERR(handle)) {
726 ret = ocfs2_jbd2_inode_add_write(handle, inode, start_byte, length);
732 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
733 OCFS2_JOURNAL_ACCESS_WRITE);
736 ocfs2_update_inode_fsync_trans(handle, inode, 1);
741 ocfs2_commit_trans(osb, handle);
742 handle = ERR_PTR(ret);
747 /* Some parts of this taken from generic_cont_expand, which turned out
748 * to be too fragile to do exactly what we need without us having to
749 * worry about recursive locking in ->write_begin() and ->write_end(). */
750 static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
751 u64 abs_to, struct buffer_head *di_bh)
753 struct address_space *mapping = inode->i_mapping;
755 unsigned long index = abs_from >> PAGE_SHIFT;
758 unsigned zero_from, zero_to, block_start, block_end;
759 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
761 BUG_ON(abs_from >= abs_to);
762 BUG_ON(abs_to > (((u64)index + 1) << PAGE_SHIFT));
763 BUG_ON(abs_from & (inode->i_blkbits - 1));
765 handle = ocfs2_zero_start_ordered_transaction(inode, di_bh,
768 if (IS_ERR(handle)) {
769 ret = PTR_ERR(handle);
773 page = find_or_create_page(mapping, index, GFP_NOFS);
777 goto out_commit_trans;
780 /* Get the offsets within the page that we want to zero */
781 zero_from = abs_from & (PAGE_SIZE - 1);
782 zero_to = abs_to & (PAGE_SIZE - 1);
786 trace_ocfs2_write_zero_page(
787 (unsigned long long)OCFS2_I(inode)->ip_blkno,
788 (unsigned long long)abs_from,
789 (unsigned long long)abs_to,
790 index, zero_from, zero_to);
792 /* We know that zero_from is block aligned */
793 for (block_start = zero_from; block_start < zero_to;
794 block_start = block_end) {
795 block_end = block_start + i_blocksize(inode);
798 * block_start is block-aligned. Bump it by one to force
799 * __block_write_begin and block_commit_write to zero the
802 ret = __block_write_begin(page, block_start + 1, 0,
810 /* must not update i_size! */
811 block_commit_write(page, block_start + 1, block_start + 1);
815 * fs-writeback will release the dirty pages without page lock
816 * whose offset are over inode size, the release happens at
817 * block_write_full_page().
819 i_size_write(inode, abs_to);
820 inode->i_blocks = ocfs2_inode_sector_count(inode);
821 di->i_size = cpu_to_le64((u64)i_size_read(inode));
822 inode->i_mtime = inode->i_ctime = current_time(inode);
823 di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
824 di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
825 di->i_mtime_nsec = di->i_ctime_nsec;
827 ocfs2_journal_dirty(handle, di_bh);
828 ocfs2_update_inode_fsync_trans(handle, inode, 1);
836 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
842 * Find the next range to zero. We do this in terms of bytes because
843 * that's what ocfs2_zero_extend() wants, and it is dealing with the
844 * pagecache. We may return multiple extents.
846 * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
847 * needs to be zeroed. range_start and range_end return the next zeroing
848 * range. A subsequent call should pass the previous range_end as its
849 * zero_start. If range_end is 0, there's nothing to do.
851 * Unwritten extents are skipped over. Refcounted extents are CoWd.
853 static int ocfs2_zero_extend_get_range(struct inode *inode,
854 struct buffer_head *di_bh,
855 u64 zero_start, u64 zero_end,
856 u64 *range_start, u64 *range_end)
858 int rc = 0, needs_cow = 0;
859 u32 p_cpos, zero_clusters = 0;
861 zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
862 u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
863 unsigned int num_clusters = 0;
864 unsigned int ext_flags = 0;
866 while (zero_cpos < last_cpos) {
867 rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
868 &num_clusters, &ext_flags);
874 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
875 zero_clusters = num_clusters;
876 if (ext_flags & OCFS2_EXT_REFCOUNTED)
881 zero_cpos += num_clusters;
883 if (!zero_clusters) {
888 while ((zero_cpos + zero_clusters) < last_cpos) {
889 rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
890 &p_cpos, &num_clusters,
897 if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
899 if (ext_flags & OCFS2_EXT_REFCOUNTED)
901 zero_clusters += num_clusters;
903 if ((zero_cpos + zero_clusters) > last_cpos)
904 zero_clusters = last_cpos - zero_cpos;
907 rc = ocfs2_refcount_cow(inode, di_bh, zero_cpos,
908 zero_clusters, UINT_MAX);
915 *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
916 *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
917 zero_cpos + zero_clusters);
924 * Zero one range returned from ocfs2_zero_extend_get_range(). The caller
925 * has made sure that the entire range needs zeroing.
927 static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
928 u64 range_end, struct buffer_head *di_bh)
932 u64 zero_pos = range_start;
934 trace_ocfs2_zero_extend_range(
935 (unsigned long long)OCFS2_I(inode)->ip_blkno,
936 (unsigned long long)range_start,
937 (unsigned long long)range_end);
938 BUG_ON(range_start >= range_end);
940 while (zero_pos < range_end) {
941 next_pos = (zero_pos & PAGE_MASK) + PAGE_SIZE;
942 if (next_pos > range_end)
943 next_pos = range_end;
944 rc = ocfs2_write_zero_page(inode, zero_pos, next_pos, di_bh);
952 * Very large extends have the potential to lock up
953 * the cpu for extended periods of time.
961 int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
965 u64 zero_start, range_start = 0, range_end = 0;
966 struct super_block *sb = inode->i_sb;
968 zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
969 trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
970 (unsigned long long)zero_start,
971 (unsigned long long)i_size_read(inode));
972 while (zero_start < zero_to_size) {
973 ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
984 if (range_start < zero_start)
985 range_start = zero_start;
986 if (range_end > zero_to_size)
987 range_end = zero_to_size;
989 ret = ocfs2_zero_extend_range(inode, range_start,
995 zero_start = range_end;
1001 int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
1002 u64 new_i_size, u64 zero_to)
1005 u32 clusters_to_add;
1006 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1009 * Only quota files call this without a bh, and they can't be
1012 BUG_ON(!di_bh && ocfs2_is_refcount_inode(inode));
1013 BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1015 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1016 if (clusters_to_add < oi->ip_clusters)
1017 clusters_to_add = 0;
1019 clusters_to_add -= oi->ip_clusters;
1021 if (clusters_to_add) {
1022 ret = ocfs2_extend_allocation(inode, oi->ip_clusters,
1023 clusters_to_add, 0);
1031 * Call this even if we don't add any clusters to the tree. We
1032 * still need to zero the area between the old i_size and the
1035 ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1043 static int ocfs2_extend_file(struct inode *inode,
1044 struct buffer_head *di_bh,
1048 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1052 /* setattr sometimes calls us like this. */
1053 if (new_i_size == 0)
1056 if (i_size_read(inode) == new_i_size)
1058 BUG_ON(new_i_size < i_size_read(inode));
1061 * The alloc sem blocks people in read/write from reading our
1062 * allocation until we're done changing it. We depend on
1063 * i_rwsem to block other extend/truncate calls while we're
1064 * here. We even have to hold it for sparse files because there
1065 * might be some tail zeroing.
1067 down_write(&oi->ip_alloc_sem);
1069 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1071 * We can optimize small extends by keeping the inodes
1074 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1075 up_write(&oi->ip_alloc_sem);
1076 goto out_update_size;
1079 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1081 up_write(&oi->ip_alloc_sem);
1087 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1088 ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1090 ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1093 up_write(&oi->ip_alloc_sem);
1101 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1109 int ocfs2_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
1112 int status = 0, size_change;
1113 int inode_locked = 0;
1114 struct inode *inode = d_inode(dentry);
1115 struct super_block *sb = inode->i_sb;
1116 struct ocfs2_super *osb = OCFS2_SB(sb);
1117 struct buffer_head *bh = NULL;
1118 handle_t *handle = NULL;
1119 struct dquot *transfer_to[MAXQUOTAS] = { };
1122 struct ocfs2_lock_holder oh;
1124 trace_ocfs2_setattr(inode, dentry,
1125 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1126 dentry->d_name.len, dentry->d_name.name,
1127 attr->ia_valid, attr->ia_mode,
1128 from_kuid(&init_user_ns, attr->ia_uid),
1129 from_kgid(&init_user_ns, attr->ia_gid));
1131 /* ensuring we don't even attempt to truncate a symlink */
1132 if (S_ISLNK(inode->i_mode))
1133 attr->ia_valid &= ~ATTR_SIZE;
1135 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1136 | ATTR_GID | ATTR_UID | ATTR_MODE)
1137 if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
1140 status = setattr_prepare(&nop_mnt_idmap, dentry, attr);
1144 if (is_quota_modification(&nop_mnt_idmap, inode, attr)) {
1145 status = dquot_initialize(inode);
1149 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1152 * Here we should wait dio to finish before inode lock
1153 * to avoid a deadlock between ocfs2_setattr() and
1154 * ocfs2_dio_end_io_write()
1156 inode_dio_wait(inode);
1158 status = ocfs2_rw_lock(inode, 1);
1165 had_lock = ocfs2_inode_lock_tracker(inode, &bh, 1, &oh);
1168 goto bail_unlock_rw;
1169 } else if (had_lock) {
1171 * As far as we know, ocfs2_setattr() could only be the first
1172 * VFS entry point in the call chain of recursive cluster
1180 * ocfs2_iop_get_acl()
1182 * But, we're not 100% sure if it's always true, because the
1183 * ordering of the VFS entry points in the call chain is out
1184 * of our control. So, we'd better dump the stack here to
1185 * catch the other cases of recursive locking.
1187 mlog(ML_ERROR, "Another case of recursive locking:\n");
1193 status = inode_newsize_ok(inode, attr->ia_size);
1197 if (i_size_read(inode) >= attr->ia_size) {
1198 if (ocfs2_should_order_data(inode)) {
1199 status = ocfs2_begin_ordered_truncate(inode,
1204 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1206 status = ocfs2_extend_file(inode, bh, attr->ia_size);
1208 if (status != -ENOSPC)
1215 if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
1216 (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
1218 * Gather pointers to quota structures so that allocation /
1219 * freeing of quota structures happens here and not inside
1220 * dquot_transfer() where we have problems with lock ordering
1222 if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)
1223 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1224 OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1225 transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid));
1226 if (IS_ERR(transfer_to[USRQUOTA])) {
1227 status = PTR_ERR(transfer_to[USRQUOTA]);
1228 transfer_to[USRQUOTA] = NULL;
1232 if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid)
1233 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1234 OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1235 transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid));
1236 if (IS_ERR(transfer_to[GRPQUOTA])) {
1237 status = PTR_ERR(transfer_to[GRPQUOTA]);
1238 transfer_to[GRPQUOTA] = NULL;
1242 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1243 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1244 2 * ocfs2_quota_trans_credits(sb));
1245 if (IS_ERR(handle)) {
1246 status = PTR_ERR(handle);
1248 goto bail_unlock_alloc;
1250 status = __dquot_transfer(inode, transfer_to);
1254 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1255 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1256 if (IS_ERR(handle)) {
1257 status = PTR_ERR(handle);
1259 goto bail_unlock_alloc;
1263 setattr_copy(&nop_mnt_idmap, inode, attr);
1264 mark_inode_dirty(inode);
1266 status = ocfs2_mark_inode_dirty(handle, inode, bh);
1271 ocfs2_commit_trans(osb, handle);
1273 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1275 if (status && inode_locked) {
1276 ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
1281 ocfs2_rw_unlock(inode, 1);
1284 /* Release quota pointers in case we acquired them */
1285 for (qtype = 0; qtype < OCFS2_MAXQUOTAS; qtype++)
1286 dqput(transfer_to[qtype]);
1288 if (!status && attr->ia_valid & ATTR_MODE) {
1289 status = ocfs2_acl_chmod(inode, bh);
1294 ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
1300 int ocfs2_getattr(struct mnt_idmap *idmap, const struct path *path,
1301 struct kstat *stat, u32 request_mask, unsigned int flags)
1303 struct inode *inode = d_inode(path->dentry);
1304 struct super_block *sb = path->dentry->d_sb;
1305 struct ocfs2_super *osb = sb->s_fs_info;
1308 err = ocfs2_inode_revalidate(path->dentry);
1315 generic_fillattr(&nop_mnt_idmap, inode, stat);
1317 * If there is inline data in the inode, the inode will normally not
1318 * have data blocks allocated (it may have an external xattr block).
1319 * Report at least one sector for such files, so tools like tar, rsync,
1320 * others don't incorrectly think the file is completely sparse.
1322 if (unlikely(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1323 stat->blocks += (stat->size + 511)>>9;
1325 /* We set the blksize from the cluster size for performance */
1326 stat->blksize = osb->s_clustersize;
1332 int ocfs2_permission(struct mnt_idmap *idmap, struct inode *inode,
1336 struct ocfs2_lock_holder oh;
1338 if (mask & MAY_NOT_BLOCK)
1341 had_lock = ocfs2_inode_lock_tracker(inode, NULL, 0, &oh);
1345 } else if (had_lock) {
1346 /* See comments in ocfs2_setattr() for details.
1347 * The call chain of this case could be:
1350 * inode_permission()
1351 * ocfs2_permission()
1352 * ocfs2_iop_get_acl()
1354 mlog(ML_ERROR, "Another case of recursive locking:\n");
1358 ret = generic_permission(&nop_mnt_idmap, inode, mask);
1360 ocfs2_inode_unlock_tracker(inode, 0, &oh, had_lock);
1365 static int __ocfs2_write_remove_suid(struct inode *inode,
1366 struct buffer_head *bh)
1370 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1371 struct ocfs2_dinode *di;
1373 trace_ocfs2_write_remove_suid(
1374 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1377 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1378 if (IS_ERR(handle)) {
1379 ret = PTR_ERR(handle);
1384 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1385 OCFS2_JOURNAL_ACCESS_WRITE);
1391 inode->i_mode &= ~S_ISUID;
1392 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1393 inode->i_mode &= ~S_ISGID;
1395 di = (struct ocfs2_dinode *) bh->b_data;
1396 di->i_mode = cpu_to_le16(inode->i_mode);
1397 ocfs2_update_inode_fsync_trans(handle, inode, 0);
1399 ocfs2_journal_dirty(handle, bh);
1402 ocfs2_commit_trans(osb, handle);
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);
1529 * zero out partial blocks of one cluster.
1531 * start: file offset where zero starts, will be made upper block aligned.
1532 * len: it will be trimmed to the end of current cluster if "start + len"
1533 * is bigger than it.
1535 static int ocfs2_zeroout_partial_cluster(struct inode *inode,
1539 u64 start_block, end_block, nr_blocks;
1540 u64 p_block, offset;
1541 u32 cluster, p_cluster, nr_clusters;
1542 struct super_block *sb = inode->i_sb;
1543 u64 end = ocfs2_align_bytes_to_clusters(sb, start);
1545 if (start + len < end)
1548 start_block = ocfs2_blocks_for_bytes(sb, start);
1549 end_block = ocfs2_blocks_for_bytes(sb, end);
1550 nr_blocks = end_block - start_block;
1554 cluster = ocfs2_bytes_to_clusters(sb, start);
1555 ret = ocfs2_get_clusters(inode, cluster, &p_cluster,
1556 &nr_clusters, NULL);
1562 offset = start_block - ocfs2_clusters_to_blocks(sb, cluster);
1563 p_block = ocfs2_clusters_to_blocks(sb, p_cluster) + offset;
1564 return sb_issue_zeroout(sb, p_block, nr_blocks, GFP_NOFS);
1567 static int ocfs2_zero_partial_clusters(struct inode *inode,
1572 u64 end = start + len;
1573 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1574 unsigned int csize = osb->s_clustersize;
1576 loff_t isize = i_size_read(inode);
1579 * The "start" and "end" values are NOT necessarily part of
1580 * the range whose allocation is being deleted. Rather, this
1581 * is what the user passed in with the request. We must zero
1582 * partial clusters here. There's no need to worry about
1583 * physical allocation - the zeroing code knows to skip holes.
1585 trace_ocfs2_zero_partial_clusters(
1586 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1587 (unsigned long long)start, (unsigned long long)end);
1590 * If both edges are on a cluster boundary then there's no
1591 * zeroing required as the region is part of the allocation to
1594 if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1597 /* No page cache for EOF blocks, issue zero out to disk. */
1600 * zeroout eof blocks in last cluster starting from
1601 * "isize" even "start" > "isize" because it is
1602 * complicated to zeroout just at "start" as "start"
1603 * may be not aligned with block size, buffer write
1604 * would be required to do that, but out of eof buffer
1605 * write is not supported.
1607 ret = ocfs2_zeroout_partial_cluster(inode, isize,
1617 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1618 if (IS_ERR(handle)) {
1619 ret = PTR_ERR(handle);
1625 * If start is on a cluster boundary and end is somewhere in another
1626 * cluster, we have not COWed the cluster starting at start, unless
1627 * end is also within the same cluster. So, in this case, we skip this
1628 * first call to ocfs2_zero_range_for_truncate() truncate and move on
1631 if ((start & (csize - 1)) != 0) {
1633 * We want to get the byte offset of the end of the 1st
1636 tmpend = (u64)osb->s_clustersize +
1637 (start & ~(osb->s_clustersize - 1));
1641 trace_ocfs2_zero_partial_clusters_range1(
1642 (unsigned long long)start,
1643 (unsigned long long)tmpend);
1645 ret = ocfs2_zero_range_for_truncate(inode, handle, start,
1653 * This may make start and end equal, but the zeroing
1654 * code will skip any work in that case so there's no
1655 * need to catch it up here.
1657 start = end & ~(osb->s_clustersize - 1);
1659 trace_ocfs2_zero_partial_clusters_range2(
1660 (unsigned long long)start, (unsigned long long)end);
1662 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1666 ocfs2_update_inode_fsync_trans(handle, inode, 1);
1668 ocfs2_commit_trans(osb, handle);
1673 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1676 struct ocfs2_extent_rec *rec = NULL;
1678 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1680 rec = &el->l_recs[i];
1682 if (le32_to_cpu(rec->e_cpos) < pos)
1690 * Helper to calculate the punching pos and length in one run, we handle the
1691 * following three cases in order:
1693 * - remove the entire record
1694 * - remove a partial record
1695 * - no record needs to be removed (hole-punching completed)
1697 static void ocfs2_calc_trunc_pos(struct inode *inode,
1698 struct ocfs2_extent_list *el,
1699 struct ocfs2_extent_rec *rec,
1700 u32 trunc_start, u32 *trunc_cpos,
1701 u32 *trunc_len, u32 *trunc_end,
1702 u64 *blkno, int *done)
1707 range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1709 if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1711 * remove an entire extent record.
1713 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1715 * Skip holes if any.
1717 if (range < *trunc_end)
1719 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1720 *blkno = le64_to_cpu(rec->e_blkno);
1721 *trunc_end = le32_to_cpu(rec->e_cpos);
1722 } else if (range > trunc_start) {
1724 * remove a partial extent record, which means we're
1725 * removing the last extent record.
1727 *trunc_cpos = trunc_start;
1731 if (range < *trunc_end)
1733 *trunc_len = *trunc_end - trunc_start;
1734 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1735 *blkno = le64_to_cpu(rec->e_blkno) +
1736 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1737 *trunc_end = trunc_start;
1740 * It may have two following possibilities:
1742 * - last record has been removed
1743 * - trunc_start was within a hole
1745 * both two cases mean the completion of hole punching.
1753 int ocfs2_remove_inode_range(struct inode *inode,
1754 struct buffer_head *di_bh, u64 byte_start,
1757 int ret = 0, flags = 0, done = 0, i;
1758 u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1760 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1761 struct ocfs2_cached_dealloc_ctxt dealloc;
1762 struct address_space *mapping = inode->i_mapping;
1763 struct ocfs2_extent_tree et;
1764 struct ocfs2_path *path = NULL;
1765 struct ocfs2_extent_list *el = NULL;
1766 struct ocfs2_extent_rec *rec = NULL;
1767 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1768 u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1770 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1771 ocfs2_init_dealloc_ctxt(&dealloc);
1773 trace_ocfs2_remove_inode_range(
1774 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1775 (unsigned long long)byte_start,
1776 (unsigned long long)byte_len);
1781 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1782 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1783 byte_start + byte_len, 0);
1789 * There's no need to get fancy with the page cache
1790 * truncate of an inline-data inode. We're talking
1791 * about less than a page here, which will be cached
1792 * in the dinode buffer anyway.
1794 unmap_mapping_range(mapping, 0, 0, 0);
1795 truncate_inode_pages(mapping, 0);
1800 * For reflinks, we may need to CoW 2 clusters which might be
1801 * partially zero'd later, if hole's start and end offset were
1802 * within one cluster(means is not exactly aligned to clustersize).
1805 if (ocfs2_is_refcount_inode(inode)) {
1806 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1812 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1819 trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1820 trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1821 cluster_in_el = trunc_end;
1823 ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1829 path = ocfs2_new_path_from_et(&et);
1836 while (trunc_end > trunc_start) {
1838 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1845 el = path_leaf_el(path);
1847 i = ocfs2_find_rec(el, trunc_end);
1849 * Need to go to previous extent block.
1852 if (path->p_tree_depth == 0)
1855 ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1864 * We've reached the leftmost extent block,
1865 * it's safe to leave.
1867 if (cluster_in_el == 0)
1871 * The 'pos' searched for previous extent block is
1872 * always one cluster less than actual trunc_end.
1874 trunc_end = cluster_in_el + 1;
1876 ocfs2_reinit_path(path, 1);
1881 rec = &el->l_recs[i];
1883 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1884 &trunc_len, &trunc_end, &blkno, &done);
1888 flags = rec->e_flags;
1889 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1891 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1892 phys_cpos, trunc_len, flags,
1893 &dealloc, refcount_loc, false);
1899 cluster_in_el = trunc_end;
1901 ocfs2_reinit_path(path, 1);
1904 ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1907 ocfs2_free_path(path);
1908 ocfs2_schedule_truncate_log_flush(osb, 1);
1909 ocfs2_run_deallocs(osb, &dealloc);
1915 * Parts of this function taken from xfs_change_file_space()
1917 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1918 loff_t f_pos, unsigned int cmd,
1919 struct ocfs2_space_resv *sr,
1924 loff_t size, orig_isize;
1925 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1926 struct buffer_head *di_bh = NULL;
1928 unsigned long long max_off = inode->i_sb->s_maxbytes;
1930 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1936 * This prevents concurrent writes on other nodes
1938 ret = ocfs2_rw_lock(inode, 1);
1944 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1950 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1952 goto out_inode_unlock;
1955 switch (sr->l_whence) {
1956 case 0: /*SEEK_SET*/
1958 case 1: /*SEEK_CUR*/
1959 sr->l_start += f_pos;
1961 case 2: /*SEEK_END*/
1962 sr->l_start += i_size_read(inode);
1966 goto out_inode_unlock;
1970 llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1973 || sr->l_start > max_off
1974 || (sr->l_start + llen) < 0
1975 || (sr->l_start + llen) > max_off) {
1977 goto out_inode_unlock;
1979 size = sr->l_start + sr->l_len;
1981 if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64 ||
1982 cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) {
1983 if (sr->l_len <= 0) {
1985 goto out_inode_unlock;
1989 if (file && setattr_should_drop_suidgid(&nop_mnt_idmap, file_inode(file))) {
1990 ret = __ocfs2_write_remove_suid(inode, di_bh);
1993 goto out_inode_unlock;
1997 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1999 case OCFS2_IOC_RESVSP:
2000 case OCFS2_IOC_RESVSP64:
2002 * This takes unsigned offsets, but the signed ones we
2003 * pass have been checked against overflow above.
2005 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
2008 case OCFS2_IOC_UNRESVSP:
2009 case OCFS2_IOC_UNRESVSP64:
2010 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
2017 orig_isize = i_size_read(inode);
2018 /* zeroout eof blocks in the cluster. */
2019 if (!ret && change_size && orig_isize < size) {
2020 ret = ocfs2_zeroout_partial_cluster(inode, orig_isize,
2023 i_size_write(inode, size);
2025 up_write(&OCFS2_I(inode)->ip_alloc_sem);
2028 goto out_inode_unlock;
2032 * We update c/mtime for these changes
2034 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
2035 if (IS_ERR(handle)) {
2036 ret = PTR_ERR(handle);
2038 goto out_inode_unlock;
2041 inode->i_ctime = inode->i_mtime = current_time(inode);
2042 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
2046 if (file && (file->f_flags & O_SYNC))
2049 ocfs2_commit_trans(osb, handle);
2053 ocfs2_inode_unlock(inode, 1);
2055 ocfs2_rw_unlock(inode, 1);
2058 inode_unlock(inode);
2062 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
2063 struct ocfs2_space_resv *sr)
2065 struct inode *inode = file_inode(file);
2066 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2069 if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
2070 !ocfs2_writes_unwritten_extents(osb))
2072 else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
2073 !ocfs2_sparse_alloc(osb))
2076 if (!S_ISREG(inode->i_mode))
2079 if (!(file->f_mode & FMODE_WRITE))
2082 ret = mnt_want_write_file(file);
2085 ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
2086 mnt_drop_write_file(file);
2090 static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
2093 struct inode *inode = file_inode(file);
2094 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2095 struct ocfs2_space_resv sr;
2096 int change_size = 1;
2097 int cmd = OCFS2_IOC_RESVSP64;
2100 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2102 if (!ocfs2_writes_unwritten_extents(osb))
2105 if (mode & FALLOC_FL_KEEP_SIZE) {
2108 ret = inode_newsize_ok(inode, offset + len);
2113 if (mode & FALLOC_FL_PUNCH_HOLE)
2114 cmd = OCFS2_IOC_UNRESVSP64;
2117 sr.l_start = (s64)offset;
2118 sr.l_len = (s64)len;
2120 return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2124 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2128 unsigned int extent_flags;
2129 u32 cpos, clusters, extent_len, phys_cpos;
2130 struct super_block *sb = inode->i_sb;
2132 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2133 !ocfs2_is_refcount_inode(inode) ||
2134 OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2137 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2138 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2141 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2148 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2153 if (extent_len > clusters)
2154 extent_len = clusters;
2156 clusters -= extent_len;
2163 static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
2165 int blockmask = inode->i_sb->s_blocksize - 1;
2166 loff_t final_size = pos + count;
2168 if ((pos & blockmask) || (final_size & blockmask))
2173 static int ocfs2_inode_lock_for_extent_tree(struct inode *inode,
2174 struct buffer_head **di_bh,
2182 ret = ocfs2_inode_lock(inode, di_bh, meta_level);
2184 ret = ocfs2_try_inode_lock(inode, di_bh, meta_level);
2190 down_write(&OCFS2_I(inode)->ip_alloc_sem);
2192 down_read(&OCFS2_I(inode)->ip_alloc_sem);
2195 ret = down_write_trylock(&OCFS2_I(inode)->ip_alloc_sem);
2197 ret = down_read_trylock(&OCFS2_I(inode)->ip_alloc_sem);
2210 ocfs2_inode_unlock(inode, meta_level);
2215 static void ocfs2_inode_unlock_for_extent_tree(struct inode *inode,
2216 struct buffer_head **di_bh,
2221 up_write(&OCFS2_I(inode)->ip_alloc_sem);
2223 up_read(&OCFS2_I(inode)->ip_alloc_sem);
2228 if (meta_level >= 0)
2229 ocfs2_inode_unlock(inode, meta_level);
2232 static int ocfs2_prepare_inode_for_write(struct file *file,
2233 loff_t pos, size_t count, int wait)
2235 int ret = 0, meta_level = 0, overwrite_io = 0;
2237 struct dentry *dentry = file->f_path.dentry;
2238 struct inode *inode = d_inode(dentry);
2239 struct buffer_head *di_bh = NULL;
2244 * We start with a read level meta lock and only jump to an ex
2245 * if we need to make modifications here.
2248 ret = ocfs2_inode_lock_for_extent_tree(inode,
2260 * Check if IO will overwrite allocated blocks in case
2261 * IOCB_NOWAIT flag is set.
2263 if (!wait && !overwrite_io) {
2266 ret = ocfs2_overwrite_io(inode, di_bh, pos, count);
2274 /* Clear suid / sgid if necessary. We do this here
2275 * instead of later in the write path because
2276 * remove_suid() calls ->setattr without any hint that
2277 * we may have already done our cluster locking. Since
2278 * ocfs2_setattr() *must* take cluster locks to
2279 * proceed, this will lead us to recursively lock the
2280 * inode. There's also the dinode i_size state which
2281 * can be lost via setattr during extending writes (we
2282 * set inode->i_size at the end of a write. */
2283 if (setattr_should_drop_suidgid(&nop_mnt_idmap, inode)) {
2284 if (meta_level == 0) {
2285 ocfs2_inode_unlock_for_extent_tree(inode,
2293 ret = ocfs2_write_remove_suid(inode);
2300 ret = ocfs2_check_range_for_refcount(inode, pos, count);
2302 ocfs2_inode_unlock_for_extent_tree(inode,
2308 ret = ocfs2_inode_lock_for_extent_tree(inode,
2319 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2321 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2322 ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
2335 trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
2338 ocfs2_inode_unlock_for_extent_tree(inode,
2347 static ssize_t ocfs2_file_write_iter(struct kiocb *iocb,
2348 struct iov_iter *from)
2351 ssize_t written = 0;
2353 size_t count = iov_iter_count(from);
2354 struct file *file = iocb->ki_filp;
2355 struct inode *inode = file_inode(file);
2356 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2357 int full_coherency = !(osb->s_mount_opt &
2358 OCFS2_MOUNT_COHERENCY_BUFFERED);
2359 void *saved_ki_complete = NULL;
2360 int append_write = ((iocb->ki_pos + count) >=
2361 i_size_read(inode) ? 1 : 0);
2362 int direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2363 int nowait = iocb->ki_flags & IOCB_NOWAIT ? 1 : 0;
2365 trace_ocfs2_file_write_iter(inode, file, file->f_path.dentry,
2366 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2367 file->f_path.dentry->d_name.len,
2368 file->f_path.dentry->d_name.name,
2369 (unsigned int)from->nr_segs); /* GRRRRR */
2371 if (!direct_io && nowait)
2378 if (!inode_trylock(inode))
2384 * Concurrent O_DIRECT writes are allowed with
2385 * mount_option "coherency=buffered".
2386 * For append write, we must take rw EX.
2388 rw_level = (!direct_io || full_coherency || append_write);
2391 ret = ocfs2_try_rw_lock(inode, rw_level);
2393 ret = ocfs2_rw_lock(inode, rw_level);
2401 * O_DIRECT writes with "coherency=full" need to take EX cluster
2402 * inode_lock to guarantee coherency.
2404 if (direct_io && full_coherency) {
2406 * We need to take and drop the inode lock to force
2407 * other nodes to drop their caches. Buffered I/O
2408 * already does this in write_begin().
2411 ret = ocfs2_try_inode_lock(inode, NULL, 1);
2413 ret = ocfs2_inode_lock(inode, NULL, 1);
2420 ocfs2_inode_unlock(inode, 1);
2423 ret = generic_write_checks(iocb, from);
2431 ret = ocfs2_prepare_inode_for_write(file, iocb->ki_pos, count, !nowait);
2438 if (direct_io && !is_sync_kiocb(iocb) &&
2439 ocfs2_is_io_unaligned(inode, count, iocb->ki_pos)) {
2441 * Make it a sync io if it's an unaligned aio.
2443 saved_ki_complete = xchg(&iocb->ki_complete, NULL);
2446 /* communicate with ocfs2_dio_end_io */
2447 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2449 written = __generic_file_write_iter(iocb, from);
2450 /* buffered aio wouldn't have proper lock coverage today */
2451 BUG_ON(written == -EIOCBQUEUED && !direct_io);
2454 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2455 * function pointer which is called when o_direct io completes so that
2456 * it can unlock our rw lock.
2457 * Unfortunately there are error cases which call end_io and others
2458 * that don't. so we don't have to unlock the rw_lock if either an
2459 * async dio is going to do it in the future or an end_io after an
2460 * error has already done it.
2462 if ((written == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2466 if (unlikely(written <= 0))
2469 if (((file->f_flags & O_DSYNC) && !direct_io) ||
2471 ret = filemap_fdatawrite_range(file->f_mapping,
2472 iocb->ki_pos - written,
2478 ret = jbd2_journal_force_commit(osb->journal->j_journal);
2484 ret = filemap_fdatawait_range(file->f_mapping,
2485 iocb->ki_pos - written,
2490 if (saved_ki_complete)
2491 xchg(&iocb->ki_complete, saved_ki_complete);
2494 ocfs2_rw_unlock(inode, rw_level);
2497 inode_unlock(inode);
2504 static ssize_t ocfs2_file_read_iter(struct kiocb *iocb,
2505 struct iov_iter *to)
2507 int ret = 0, rw_level = -1, lock_level = 0;
2508 struct file *filp = iocb->ki_filp;
2509 struct inode *inode = file_inode(filp);
2510 int direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2511 int nowait = iocb->ki_flags & IOCB_NOWAIT ? 1 : 0;
2513 trace_ocfs2_file_read_iter(inode, filp, filp->f_path.dentry,
2514 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2515 filp->f_path.dentry->d_name.len,
2516 filp->f_path.dentry->d_name.name,
2517 to->nr_segs); /* GRRRRR */
2526 if (!direct_io && nowait)
2530 * buffered reads protect themselves in ->read_folio(). O_DIRECT reads
2531 * need locks to protect pending reads from racing with truncate.
2535 ret = ocfs2_try_rw_lock(inode, 0);
2537 ret = ocfs2_rw_lock(inode, 0);
2545 /* communicate with ocfs2_dio_end_io */
2546 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2550 * We're fine letting folks race truncates and extending
2551 * writes with read across the cluster, just like they can
2552 * locally. Hence no rw_lock during read.
2554 * Take and drop the meta data lock to update inode fields
2555 * like i_size. This allows the checks down below
2556 * copy_splice_read() a chance of actually working.
2558 ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level,
2565 ocfs2_inode_unlock(inode, lock_level);
2567 ret = generic_file_read_iter(iocb, to);
2568 trace_generic_file_read_iter_ret(ret);
2570 /* buffered aio wouldn't have proper lock coverage today */
2571 BUG_ON(ret == -EIOCBQUEUED && !direct_io);
2573 /* see ocfs2_file_write_iter */
2574 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2580 ocfs2_rw_unlock(inode, rw_level);
2585 static ssize_t ocfs2_file_splice_read(struct file *in, loff_t *ppos,
2586 struct pipe_inode_info *pipe,
2587 size_t len, unsigned int flags)
2589 struct inode *inode = file_inode(in);
2593 trace_ocfs2_file_splice_read(inode, in, in->f_path.dentry,
2594 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2595 in->f_path.dentry->d_name.len,
2596 in->f_path.dentry->d_name.name,
2600 * We're fine letting folks race truncates and extending writes with
2601 * read across the cluster, just like they can locally. Hence no
2602 * rw_lock during read.
2604 * Take and drop the meta data lock to update inode fields like i_size.
2605 * This allows the checks down below filemap_splice_read() a chance of
2608 ret = ocfs2_inode_lock_atime(inode, in->f_path.mnt, &lock_level, 1);
2614 ocfs2_inode_unlock(inode, lock_level);
2616 ret = filemap_splice_read(in, ppos, pipe, len, flags);
2617 trace_filemap_splice_read_ret(ret);
2622 /* Refer generic_file_llseek_unlocked() */
2623 static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence)
2625 struct inode *inode = file->f_mapping->host;
2634 /* SEEK_END requires the OCFS2 inode lock for the file
2635 * because it references the file's size.
2637 ret = ocfs2_inode_lock(inode, NULL, 0);
2642 offset += i_size_read(inode);
2643 ocfs2_inode_unlock(inode, 0);
2647 offset = file->f_pos;
2650 offset += file->f_pos;
2654 ret = ocfs2_seek_data_hole_offset(file, &offset, whence);
2663 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2666 inode_unlock(inode);
2672 static loff_t ocfs2_remap_file_range(struct file *file_in, loff_t pos_in,
2673 struct file *file_out, loff_t pos_out,
2674 loff_t len, unsigned int remap_flags)
2676 struct inode *inode_in = file_inode(file_in);
2677 struct inode *inode_out = file_inode(file_out);
2678 struct ocfs2_super *osb = OCFS2_SB(inode_in->i_sb);
2679 struct buffer_head *in_bh = NULL, *out_bh = NULL;
2680 bool same_inode = (inode_in == inode_out);
2681 loff_t remapped = 0;
2684 if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
2686 if (!ocfs2_refcount_tree(osb))
2688 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
2691 /* Lock both files against IO */
2692 ret = ocfs2_reflink_inodes_lock(inode_in, &in_bh, inode_out, &out_bh);
2696 /* Check file eligibility and prepare for block sharing. */
2698 if ((OCFS2_I(inode_in)->ip_flags & OCFS2_INODE_SYSTEM_FILE) ||
2699 (OCFS2_I(inode_out)->ip_flags & OCFS2_INODE_SYSTEM_FILE))
2702 ret = generic_remap_file_range_prep(file_in, pos_in, file_out, pos_out,
2704 if (ret < 0 || len == 0)
2707 /* Lock out changes to the allocation maps and remap. */
2708 down_write(&OCFS2_I(inode_in)->ip_alloc_sem);
2710 down_write_nested(&OCFS2_I(inode_out)->ip_alloc_sem,
2711 SINGLE_DEPTH_NESTING);
2713 /* Zap any page cache for the destination file's range. */
2714 truncate_inode_pages_range(&inode_out->i_data,
2715 round_down(pos_out, PAGE_SIZE),
2716 round_up(pos_out + len, PAGE_SIZE) - 1);
2718 remapped = ocfs2_reflink_remap_blocks(inode_in, in_bh, pos_in,
2719 inode_out, out_bh, pos_out, len);
2720 up_write(&OCFS2_I(inode_in)->ip_alloc_sem);
2722 up_write(&OCFS2_I(inode_out)->ip_alloc_sem);
2730 * Empty the extent map so that we may get the right extent
2731 * record from the disk.
2733 ocfs2_extent_map_trunc(inode_in, 0);
2734 ocfs2_extent_map_trunc(inode_out, 0);
2736 ret = ocfs2_reflink_update_dest(inode_out, out_bh, pos_out + len);
2743 ocfs2_reflink_inodes_unlock(inode_in, in_bh, inode_out, out_bh);
2744 return remapped > 0 ? remapped : ret;
2747 const struct inode_operations ocfs2_file_iops = {
2748 .setattr = ocfs2_setattr,
2749 .getattr = ocfs2_getattr,
2750 .permission = ocfs2_permission,
2751 .listxattr = ocfs2_listxattr,
2752 .fiemap = ocfs2_fiemap,
2753 .get_inode_acl = ocfs2_iop_get_acl,
2754 .set_acl = ocfs2_iop_set_acl,
2755 .fileattr_get = ocfs2_fileattr_get,
2756 .fileattr_set = ocfs2_fileattr_set,
2759 const struct inode_operations ocfs2_special_file_iops = {
2760 .setattr = ocfs2_setattr,
2761 .getattr = ocfs2_getattr,
2762 .permission = ocfs2_permission,
2763 .get_inode_acl = ocfs2_iop_get_acl,
2764 .set_acl = ocfs2_iop_set_acl,
2768 * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2769 * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2771 const struct file_operations ocfs2_fops = {
2772 .llseek = ocfs2_file_llseek,
2774 .fsync = ocfs2_sync_file,
2775 .release = ocfs2_file_release,
2776 .open = ocfs2_file_open,
2777 .read_iter = ocfs2_file_read_iter,
2778 .write_iter = ocfs2_file_write_iter,
2779 .unlocked_ioctl = ocfs2_ioctl,
2780 #ifdef CONFIG_COMPAT
2781 .compat_ioctl = ocfs2_compat_ioctl,
2784 .flock = ocfs2_flock,
2785 .splice_read = ocfs2_file_splice_read,
2786 .splice_write = iter_file_splice_write,
2787 .fallocate = ocfs2_fallocate,
2788 .remap_file_range = ocfs2_remap_file_range,
2791 const struct file_operations ocfs2_dops = {
2792 .llseek = generic_file_llseek,
2793 .read = generic_read_dir,
2794 .iterate = ocfs2_readdir,
2795 .fsync = ocfs2_sync_file,
2796 .release = ocfs2_dir_release,
2797 .open = ocfs2_dir_open,
2798 .unlocked_ioctl = ocfs2_ioctl,
2799 #ifdef CONFIG_COMPAT
2800 .compat_ioctl = ocfs2_compat_ioctl,
2803 .flock = ocfs2_flock,
2807 * POSIX-lockless variants of our file_operations.
2809 * These will be used if the underlying cluster stack does not support
2810 * posix file locking, if the user passes the "localflocks" mount
2811 * option, or if we have a local-only fs.
2813 * ocfs2_flock is in here because all stacks handle UNIX file locks,
2814 * so we still want it in the case of no stack support for
2815 * plocks. Internally, it will do the right thing when asked to ignore
2818 const struct file_operations ocfs2_fops_no_plocks = {
2819 .llseek = ocfs2_file_llseek,
2821 .fsync = ocfs2_sync_file,
2822 .release = ocfs2_file_release,
2823 .open = ocfs2_file_open,
2824 .read_iter = ocfs2_file_read_iter,
2825 .write_iter = ocfs2_file_write_iter,
2826 .unlocked_ioctl = ocfs2_ioctl,
2827 #ifdef CONFIG_COMPAT
2828 .compat_ioctl = ocfs2_compat_ioctl,
2830 .flock = ocfs2_flock,
2831 .splice_read = filemap_splice_read,
2832 .splice_write = iter_file_splice_write,
2833 .fallocate = ocfs2_fallocate,
2834 .remap_file_range = ocfs2_remap_file_range,
2837 const struct file_operations ocfs2_dops_no_plocks = {
2838 .llseek = generic_file_llseek,
2839 .read = generic_read_dir,
2840 .iterate = ocfs2_readdir,
2841 .fsync = ocfs2_sync_file,
2842 .release = ocfs2_dir_release,
2843 .open = ocfs2_dir_open,
2844 .unlocked_ioctl = ocfs2_ioctl,
2845 #ifdef CONFIG_COMPAT
2846 .compat_ioctl = ocfs2_compat_ioctl,
2848 .flock = ocfs2_flock,