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_mutex 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
479 unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
480 truncate_inode_pages(inode->i_mapping, new_i_size);
482 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
483 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
484 i_size_read(inode), 1);
488 goto bail_unlock_sem;
491 /* alright, we're going to need to do a full blown alloc size
492 * change. Orphan the inode so that recovery can complete the
493 * truncate if necessary. This does the task of marking
495 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
498 goto bail_unlock_sem;
501 status = ocfs2_commit_truncate(osb, inode, di_bh);
504 goto bail_unlock_sem;
507 /* TODO: orphan dir cleanup here. */
509 up_write(&OCFS2_I(inode)->ip_alloc_sem);
512 if (!status && OCFS2_I(inode)->ip_clusters == 0)
513 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
519 * extend file allocation only here.
520 * we'll update all the disk stuff, and oip->alloc_size
522 * expect stuff to be locked, a transaction started and enough data /
523 * metadata reservations in the contexts.
525 * Will return -EAGAIN, and a reason if a restart is needed.
526 * If passed in, *reason will always be set, even in error.
528 int ocfs2_add_inode_data(struct ocfs2_super *osb,
533 struct buffer_head *fe_bh,
535 struct ocfs2_alloc_context *data_ac,
536 struct ocfs2_alloc_context *meta_ac,
537 enum ocfs2_alloc_restarted *reason_ret)
540 struct ocfs2_extent_tree et;
542 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
543 ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
544 clusters_to_add, mark_unwritten,
545 data_ac, meta_ac, reason_ret);
550 static int ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
551 u32 clusters_to_add, int mark_unwritten)
554 int restart_func = 0;
557 struct buffer_head *bh = NULL;
558 struct ocfs2_dinode *fe = NULL;
559 handle_t *handle = NULL;
560 struct ocfs2_alloc_context *data_ac = NULL;
561 struct ocfs2_alloc_context *meta_ac = NULL;
562 enum ocfs2_alloc_restarted why = RESTART_NONE;
563 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
564 struct ocfs2_extent_tree et;
568 * Unwritten extent only exists for file systems which
571 BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
573 status = ocfs2_read_inode_block(inode, &bh);
578 fe = (struct ocfs2_dinode *) bh->b_data;
581 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
583 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
584 status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
591 credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list);
592 handle = ocfs2_start_trans(osb, credits);
593 if (IS_ERR(handle)) {
594 status = PTR_ERR(handle);
600 restarted_transaction:
601 trace_ocfs2_extend_allocation(
602 (unsigned long long)OCFS2_I(inode)->ip_blkno,
603 (unsigned long long)i_size_read(inode),
604 le32_to_cpu(fe->i_clusters), clusters_to_add,
607 status = dquot_alloc_space_nodirty(inode,
608 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
613 /* reserve a write to the file entry early on - that we if we
614 * run out of credits in the allocation path, we can still
616 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
617 OCFS2_JOURNAL_ACCESS_WRITE);
623 prev_clusters = OCFS2_I(inode)->ip_clusters;
625 status = ocfs2_add_inode_data(osb,
635 if ((status < 0) && (status != -EAGAIN)) {
636 if (status != -ENOSPC)
640 ocfs2_update_inode_fsync_trans(handle, inode, 1);
641 ocfs2_journal_dirty(handle, bh);
643 spin_lock(&OCFS2_I(inode)->ip_lock);
644 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
645 spin_unlock(&OCFS2_I(inode)->ip_lock);
646 /* Release unused quota reservation */
647 dquot_free_space(inode,
648 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
651 if (why != RESTART_NONE && clusters_to_add) {
652 if (why == RESTART_META) {
656 BUG_ON(why != RESTART_TRANS);
658 status = ocfs2_allocate_extend_trans(handle, 1);
660 /* handle still has to be committed at
666 goto restarted_transaction;
670 trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno,
671 le32_to_cpu(fe->i_clusters),
672 (unsigned long long)le64_to_cpu(fe->i_size),
673 OCFS2_I(inode)->ip_clusters,
674 (unsigned long long)i_size_read(inode));
677 if (status < 0 && did_quota)
678 dquot_free_space(inode,
679 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
681 ocfs2_commit_trans(osb, handle);
685 ocfs2_free_alloc_context(data_ac);
689 ocfs2_free_alloc_context(meta_ac);
692 if ((!status) && restart_func) {
703 * While a write will already be ordering the data, a truncate will not.
704 * Thus, we need to explicitly order the zeroed pages.
706 static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode,
707 struct buffer_head *di_bh,
711 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
712 handle_t *handle = NULL;
715 if (!ocfs2_should_order_data(inode))
718 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
719 if (IS_ERR(handle)) {
725 ret = ocfs2_jbd2_inode_add_write(handle, inode, start_byte, length);
731 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
732 OCFS2_JOURNAL_ACCESS_WRITE);
735 ocfs2_update_inode_fsync_trans(handle, inode, 1);
740 ocfs2_commit_trans(osb, handle);
741 handle = ERR_PTR(ret);
746 /* Some parts of this taken from generic_cont_expand, which turned out
747 * to be too fragile to do exactly what we need without us having to
748 * worry about recursive locking in ->write_begin() and ->write_end(). */
749 static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
750 u64 abs_to, struct buffer_head *di_bh)
752 struct address_space *mapping = inode->i_mapping;
754 unsigned long index = abs_from >> PAGE_SHIFT;
757 unsigned zero_from, zero_to, block_start, block_end;
758 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
760 BUG_ON(abs_from >= abs_to);
761 BUG_ON(abs_to > (((u64)index + 1) << PAGE_SHIFT));
762 BUG_ON(abs_from & (inode->i_blkbits - 1));
764 handle = ocfs2_zero_start_ordered_transaction(inode, di_bh,
767 if (IS_ERR(handle)) {
768 ret = PTR_ERR(handle);
772 page = find_or_create_page(mapping, index, GFP_NOFS);
776 goto out_commit_trans;
779 /* Get the offsets within the page that we want to zero */
780 zero_from = abs_from & (PAGE_SIZE - 1);
781 zero_to = abs_to & (PAGE_SIZE - 1);
785 trace_ocfs2_write_zero_page(
786 (unsigned long long)OCFS2_I(inode)->ip_blkno,
787 (unsigned long long)abs_from,
788 (unsigned long long)abs_to,
789 index, zero_from, zero_to);
791 /* We know that zero_from is block aligned */
792 for (block_start = zero_from; block_start < zero_to;
793 block_start = block_end) {
794 block_end = block_start + i_blocksize(inode);
797 * block_start is block-aligned. Bump it by one to force
798 * __block_write_begin and block_commit_write to zero the
801 ret = __block_write_begin(page, block_start + 1, 0,
809 /* must not update i_size! */
810 ret = block_commit_write(page, block_start + 1,
819 * fs-writeback will release the dirty pages without page lock
820 * whose offset are over inode size, the release happens at
821 * block_write_full_page().
823 i_size_write(inode, abs_to);
824 inode->i_blocks = ocfs2_inode_sector_count(inode);
825 di->i_size = cpu_to_le64((u64)i_size_read(inode));
826 inode->i_mtime = inode->i_ctime = current_time(inode);
827 di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
828 di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
829 di->i_mtime_nsec = di->i_ctime_nsec;
831 ocfs2_journal_dirty(handle, di_bh);
832 ocfs2_update_inode_fsync_trans(handle, inode, 1);
840 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
846 * Find the next range to zero. We do this in terms of bytes because
847 * that's what ocfs2_zero_extend() wants, and it is dealing with the
848 * pagecache. We may return multiple extents.
850 * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
851 * needs to be zeroed. range_start and range_end return the next zeroing
852 * range. A subsequent call should pass the previous range_end as its
853 * zero_start. If range_end is 0, there's nothing to do.
855 * Unwritten extents are skipped over. Refcounted extents are CoWd.
857 static int ocfs2_zero_extend_get_range(struct inode *inode,
858 struct buffer_head *di_bh,
859 u64 zero_start, u64 zero_end,
860 u64 *range_start, u64 *range_end)
862 int rc = 0, needs_cow = 0;
863 u32 p_cpos, zero_clusters = 0;
865 zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
866 u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
867 unsigned int num_clusters = 0;
868 unsigned int ext_flags = 0;
870 while (zero_cpos < last_cpos) {
871 rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
872 &num_clusters, &ext_flags);
878 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
879 zero_clusters = num_clusters;
880 if (ext_flags & OCFS2_EXT_REFCOUNTED)
885 zero_cpos += num_clusters;
887 if (!zero_clusters) {
892 while ((zero_cpos + zero_clusters) < last_cpos) {
893 rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
894 &p_cpos, &num_clusters,
901 if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
903 if (ext_flags & OCFS2_EXT_REFCOUNTED)
905 zero_clusters += num_clusters;
907 if ((zero_cpos + zero_clusters) > last_cpos)
908 zero_clusters = last_cpos - zero_cpos;
911 rc = ocfs2_refcount_cow(inode, di_bh, zero_cpos,
912 zero_clusters, UINT_MAX);
919 *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
920 *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
921 zero_cpos + zero_clusters);
928 * Zero one range returned from ocfs2_zero_extend_get_range(). The caller
929 * has made sure that the entire range needs zeroing.
931 static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
932 u64 range_end, struct buffer_head *di_bh)
936 u64 zero_pos = range_start;
938 trace_ocfs2_zero_extend_range(
939 (unsigned long long)OCFS2_I(inode)->ip_blkno,
940 (unsigned long long)range_start,
941 (unsigned long long)range_end);
942 BUG_ON(range_start >= range_end);
944 while (zero_pos < range_end) {
945 next_pos = (zero_pos & PAGE_MASK) + PAGE_SIZE;
946 if (next_pos > range_end)
947 next_pos = range_end;
948 rc = ocfs2_write_zero_page(inode, zero_pos, next_pos, di_bh);
956 * Very large extends have the potential to lock up
957 * the cpu for extended periods of time.
965 int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
969 u64 zero_start, range_start = 0, range_end = 0;
970 struct super_block *sb = inode->i_sb;
972 zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
973 trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
974 (unsigned long long)zero_start,
975 (unsigned long long)i_size_read(inode));
976 while (zero_start < zero_to_size) {
977 ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
988 if (range_start < zero_start)
989 range_start = zero_start;
990 if (range_end > zero_to_size)
991 range_end = zero_to_size;
993 ret = ocfs2_zero_extend_range(inode, range_start,
999 zero_start = range_end;
1005 int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
1006 u64 new_i_size, u64 zero_to)
1009 u32 clusters_to_add;
1010 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1013 * Only quota files call this without a bh, and they can't be
1016 BUG_ON(!di_bh && ocfs2_is_refcount_inode(inode));
1017 BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1019 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1020 if (clusters_to_add < oi->ip_clusters)
1021 clusters_to_add = 0;
1023 clusters_to_add -= oi->ip_clusters;
1025 if (clusters_to_add) {
1026 ret = ocfs2_extend_allocation(inode, oi->ip_clusters,
1027 clusters_to_add, 0);
1035 * Call this even if we don't add any clusters to the tree. We
1036 * still need to zero the area between the old i_size and the
1039 ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1047 static int ocfs2_extend_file(struct inode *inode,
1048 struct buffer_head *di_bh,
1052 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1056 /* setattr sometimes calls us like this. */
1057 if (new_i_size == 0)
1060 if (i_size_read(inode) == new_i_size)
1062 BUG_ON(new_i_size < i_size_read(inode));
1065 * The alloc sem blocks people in read/write from reading our
1066 * allocation until we're done changing it. We depend on
1067 * i_mutex to block other extend/truncate calls while we're
1068 * here. We even have to hold it for sparse files because there
1069 * might be some tail zeroing.
1071 down_write(&oi->ip_alloc_sem);
1073 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1075 * We can optimize small extends by keeping the inodes
1078 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1079 up_write(&oi->ip_alloc_sem);
1080 goto out_update_size;
1083 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1085 up_write(&oi->ip_alloc_sem);
1091 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1092 ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1094 ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1097 up_write(&oi->ip_alloc_sem);
1105 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1113 int ocfs2_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
1116 int status = 0, size_change;
1117 int inode_locked = 0;
1118 struct inode *inode = d_inode(dentry);
1119 struct super_block *sb = inode->i_sb;
1120 struct ocfs2_super *osb = OCFS2_SB(sb);
1121 struct buffer_head *bh = NULL;
1122 handle_t *handle = NULL;
1123 struct dquot *transfer_to[MAXQUOTAS] = { };
1126 struct ocfs2_lock_holder oh;
1128 trace_ocfs2_setattr(inode, dentry,
1129 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1130 dentry->d_name.len, dentry->d_name.name,
1131 attr->ia_valid, attr->ia_mode,
1132 from_kuid(&init_user_ns, attr->ia_uid),
1133 from_kgid(&init_user_ns, attr->ia_gid));
1135 /* ensuring we don't even attempt to truncate a symlink */
1136 if (S_ISLNK(inode->i_mode))
1137 attr->ia_valid &= ~ATTR_SIZE;
1139 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1140 | ATTR_GID | ATTR_UID | ATTR_MODE)
1141 if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
1144 status = setattr_prepare(&init_user_ns, dentry, attr);
1148 if (is_quota_modification(inode, attr)) {
1149 status = dquot_initialize(inode);
1153 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1156 * Here we should wait dio to finish before inode lock
1157 * to avoid a deadlock between ocfs2_setattr() and
1158 * ocfs2_dio_end_io_write()
1160 inode_dio_wait(inode);
1162 status = ocfs2_rw_lock(inode, 1);
1169 had_lock = ocfs2_inode_lock_tracker(inode, &bh, 1, &oh);
1172 goto bail_unlock_rw;
1173 } else if (had_lock) {
1175 * As far as we know, ocfs2_setattr() could only be the first
1176 * VFS entry point in the call chain of recursive cluster
1184 * ocfs2_iop_get_acl()
1186 * But, we're not 100% sure if it's always true, because the
1187 * ordering of the VFS entry points in the call chain is out
1188 * of our control. So, we'd better dump the stack here to
1189 * catch the other cases of recursive locking.
1191 mlog(ML_ERROR, "Another case of recursive locking:\n");
1197 status = inode_newsize_ok(inode, attr->ia_size);
1201 if (i_size_read(inode) >= attr->ia_size) {
1202 if (ocfs2_should_order_data(inode)) {
1203 status = ocfs2_begin_ordered_truncate(inode,
1208 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1210 status = ocfs2_extend_file(inode, bh, attr->ia_size);
1212 if (status != -ENOSPC)
1219 if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
1220 (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
1222 * Gather pointers to quota structures so that allocation /
1223 * freeing of quota structures happens here and not inside
1224 * dquot_transfer() where we have problems with lock ordering
1226 if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)
1227 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1228 OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1229 transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid));
1230 if (IS_ERR(transfer_to[USRQUOTA])) {
1231 status = PTR_ERR(transfer_to[USRQUOTA]);
1232 transfer_to[USRQUOTA] = NULL;
1236 if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid)
1237 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1238 OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1239 transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid));
1240 if (IS_ERR(transfer_to[GRPQUOTA])) {
1241 status = PTR_ERR(transfer_to[GRPQUOTA]);
1242 transfer_to[GRPQUOTA] = NULL;
1246 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1247 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1248 2 * ocfs2_quota_trans_credits(sb));
1249 if (IS_ERR(handle)) {
1250 status = PTR_ERR(handle);
1252 goto bail_unlock_alloc;
1254 status = __dquot_transfer(inode, transfer_to);
1258 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1259 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1260 if (IS_ERR(handle)) {
1261 status = PTR_ERR(handle);
1263 goto bail_unlock_alloc;
1267 setattr_copy(&init_user_ns, inode, attr);
1268 mark_inode_dirty(inode);
1270 status = ocfs2_mark_inode_dirty(handle, inode, bh);
1275 ocfs2_commit_trans(osb, handle);
1277 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1279 if (status && inode_locked) {
1280 ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
1285 ocfs2_rw_unlock(inode, 1);
1288 /* Release quota pointers in case we acquired them */
1289 for (qtype = 0; qtype < OCFS2_MAXQUOTAS; qtype++)
1290 dqput(transfer_to[qtype]);
1292 if (!status && attr->ia_valid & ATTR_MODE) {
1293 status = ocfs2_acl_chmod(inode, bh);
1298 ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
1304 int ocfs2_getattr(struct user_namespace *mnt_userns, const struct path *path,
1305 struct kstat *stat, u32 request_mask, unsigned int flags)
1307 struct inode *inode = d_inode(path->dentry);
1308 struct super_block *sb = path->dentry->d_sb;
1309 struct ocfs2_super *osb = sb->s_fs_info;
1312 err = ocfs2_inode_revalidate(path->dentry);
1319 generic_fillattr(&init_user_ns, inode, stat);
1321 * If there is inline data in the inode, the inode will normally not
1322 * have data blocks allocated (it may have an external xattr block).
1323 * Report at least one sector for such files, so tools like tar, rsync,
1324 * others don't incorrectly think the file is completely sparse.
1326 if (unlikely(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1327 stat->blocks += (stat->size + 511)>>9;
1329 /* We set the blksize from the cluster size for performance */
1330 stat->blksize = osb->s_clustersize;
1336 int ocfs2_permission(struct user_namespace *mnt_userns, struct inode *inode,
1340 struct ocfs2_lock_holder oh;
1342 if (mask & MAY_NOT_BLOCK)
1345 had_lock = ocfs2_inode_lock_tracker(inode, NULL, 0, &oh);
1349 } else if (had_lock) {
1350 /* See comments in ocfs2_setattr() for details.
1351 * The call chain of this case could be:
1354 * inode_permission()
1355 * ocfs2_permission()
1356 * ocfs2_iop_get_acl()
1358 mlog(ML_ERROR, "Another case of recursive locking:\n");
1362 ret = generic_permission(&init_user_ns, inode, mask);
1364 ocfs2_inode_unlock_tracker(inode, 0, &oh, had_lock);
1369 static int __ocfs2_write_remove_suid(struct inode *inode,
1370 struct buffer_head *bh)
1374 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1375 struct ocfs2_dinode *di;
1377 trace_ocfs2_write_remove_suid(
1378 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1381 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1382 if (IS_ERR(handle)) {
1383 ret = PTR_ERR(handle);
1388 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1389 OCFS2_JOURNAL_ACCESS_WRITE);
1395 inode->i_mode &= ~S_ISUID;
1396 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1397 inode->i_mode &= ~S_ISGID;
1399 di = (struct ocfs2_dinode *) bh->b_data;
1400 di->i_mode = cpu_to_le16(inode->i_mode);
1401 ocfs2_update_inode_fsync_trans(handle, inode, 0);
1403 ocfs2_journal_dirty(handle, bh);
1406 ocfs2_commit_trans(osb, handle);
1411 static int ocfs2_write_remove_suid(struct inode *inode)
1414 struct buffer_head *bh = NULL;
1416 ret = ocfs2_read_inode_block(inode, &bh);
1422 ret = __ocfs2_write_remove_suid(inode, bh);
1429 * Allocate enough extents to cover the region starting at byte offset
1430 * start for len bytes. Existing extents are skipped, any extents
1431 * added are marked as "unwritten".
1433 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1437 u32 cpos, phys_cpos, clusters, alloc_size;
1438 u64 end = start + len;
1439 struct buffer_head *di_bh = NULL;
1441 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1442 ret = ocfs2_read_inode_block(inode, &di_bh);
1449 * Nothing to do if the requested reservation range
1450 * fits within the inode.
1452 if (ocfs2_size_fits_inline_data(di_bh, end))
1455 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1463 * We consider both start and len to be inclusive.
1465 cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1466 clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1470 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1478 * Hole or existing extent len can be arbitrary, so
1479 * cap it to our own allocation request.
1481 if (alloc_size > clusters)
1482 alloc_size = clusters;
1486 * We already have an allocation at this
1487 * region so we can safely skip it.
1492 ret = ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1501 clusters -= alloc_size;
1512 * Truncate a byte range, avoiding pages within partial clusters. This
1513 * preserves those pages for the zeroing code to write to.
1515 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1518 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1520 struct address_space *mapping = inode->i_mapping;
1522 start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1523 end = byte_start + byte_len;
1524 end = end & ~(osb->s_clustersize - 1);
1527 unmap_mapping_range(mapping, start, end - start, 0);
1528 truncate_inode_pages_range(mapping, start, end - 1);
1532 static int ocfs2_zero_partial_clusters(struct inode *inode,
1537 u64 end = start + len;
1538 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1539 unsigned int csize = osb->s_clustersize;
1543 * The "start" and "end" values are NOT necessarily part of
1544 * the range whose allocation is being deleted. Rather, this
1545 * is what the user passed in with the request. We must zero
1546 * partial clusters here. There's no need to worry about
1547 * physical allocation - the zeroing code knows to skip holes.
1549 trace_ocfs2_zero_partial_clusters(
1550 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1551 (unsigned long long)start, (unsigned long long)end);
1554 * If both edges are on a cluster boundary then there's no
1555 * zeroing required as the region is part of the allocation to
1558 if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1561 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1562 if (IS_ERR(handle)) {
1563 ret = PTR_ERR(handle);
1569 * If start is on a cluster boundary and end is somewhere in another
1570 * cluster, we have not COWed the cluster starting at start, unless
1571 * end is also within the same cluster. So, in this case, we skip this
1572 * first call to ocfs2_zero_range_for_truncate() truncate and move on
1575 if ((start & (csize - 1)) != 0) {
1577 * We want to get the byte offset of the end of the 1st
1580 tmpend = (u64)osb->s_clustersize +
1581 (start & ~(osb->s_clustersize - 1));
1585 trace_ocfs2_zero_partial_clusters_range1(
1586 (unsigned long long)start,
1587 (unsigned long long)tmpend);
1589 ret = ocfs2_zero_range_for_truncate(inode, handle, start,
1597 * This may make start and end equal, but the zeroing
1598 * code will skip any work in that case so there's no
1599 * need to catch it up here.
1601 start = end & ~(osb->s_clustersize - 1);
1603 trace_ocfs2_zero_partial_clusters_range2(
1604 (unsigned long long)start, (unsigned long long)end);
1606 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1610 ocfs2_update_inode_fsync_trans(handle, inode, 1);
1612 ocfs2_commit_trans(osb, handle);
1617 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1620 struct ocfs2_extent_rec *rec = NULL;
1622 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1624 rec = &el->l_recs[i];
1626 if (le32_to_cpu(rec->e_cpos) < pos)
1634 * Helper to calculate the punching pos and length in one run, we handle the
1635 * following three cases in order:
1637 * - remove the entire record
1638 * - remove a partial record
1639 * - no record needs to be removed (hole-punching completed)
1641 static void ocfs2_calc_trunc_pos(struct inode *inode,
1642 struct ocfs2_extent_list *el,
1643 struct ocfs2_extent_rec *rec,
1644 u32 trunc_start, u32 *trunc_cpos,
1645 u32 *trunc_len, u32 *trunc_end,
1646 u64 *blkno, int *done)
1651 range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1653 if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1655 * remove an entire extent record.
1657 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1659 * Skip holes if any.
1661 if (range < *trunc_end)
1663 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1664 *blkno = le64_to_cpu(rec->e_blkno);
1665 *trunc_end = le32_to_cpu(rec->e_cpos);
1666 } else if (range > trunc_start) {
1668 * remove a partial extent record, which means we're
1669 * removing the last extent record.
1671 *trunc_cpos = trunc_start;
1675 if (range < *trunc_end)
1677 *trunc_len = *trunc_end - trunc_start;
1678 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1679 *blkno = le64_to_cpu(rec->e_blkno) +
1680 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1681 *trunc_end = trunc_start;
1684 * It may have two following possibilities:
1686 * - last record has been removed
1687 * - trunc_start was within a hole
1689 * both two cases mean the completion of hole punching.
1697 int ocfs2_remove_inode_range(struct inode *inode,
1698 struct buffer_head *di_bh, u64 byte_start,
1701 int ret = 0, flags = 0, done = 0, i;
1702 u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1704 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1705 struct ocfs2_cached_dealloc_ctxt dealloc;
1706 struct address_space *mapping = inode->i_mapping;
1707 struct ocfs2_extent_tree et;
1708 struct ocfs2_path *path = NULL;
1709 struct ocfs2_extent_list *el = NULL;
1710 struct ocfs2_extent_rec *rec = NULL;
1711 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1712 u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1714 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1715 ocfs2_init_dealloc_ctxt(&dealloc);
1717 trace_ocfs2_remove_inode_range(
1718 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1719 (unsigned long long)byte_start,
1720 (unsigned long long)byte_len);
1725 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1726 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1727 byte_start + byte_len, 0);
1733 * There's no need to get fancy with the page cache
1734 * truncate of an inline-data inode. We're talking
1735 * about less than a page here, which will be cached
1736 * in the dinode buffer anyway.
1738 unmap_mapping_range(mapping, 0, 0, 0);
1739 truncate_inode_pages(mapping, 0);
1744 * For reflinks, we may need to CoW 2 clusters which might be
1745 * partially zero'd later, if hole's start and end offset were
1746 * within one cluster(means is not exactly aligned to clustersize).
1749 if (ocfs2_is_refcount_inode(inode)) {
1750 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1756 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1763 trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1764 trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1765 cluster_in_el = trunc_end;
1767 ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1773 path = ocfs2_new_path_from_et(&et);
1780 while (trunc_end > trunc_start) {
1782 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1789 el = path_leaf_el(path);
1791 i = ocfs2_find_rec(el, trunc_end);
1793 * Need to go to previous extent block.
1796 if (path->p_tree_depth == 0)
1799 ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1808 * We've reached the leftmost extent block,
1809 * it's safe to leave.
1811 if (cluster_in_el == 0)
1815 * The 'pos' searched for previous extent block is
1816 * always one cluster less than actual trunc_end.
1818 trunc_end = cluster_in_el + 1;
1820 ocfs2_reinit_path(path, 1);
1825 rec = &el->l_recs[i];
1827 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1828 &trunc_len, &trunc_end, &blkno, &done);
1832 flags = rec->e_flags;
1833 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1835 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1836 phys_cpos, trunc_len, flags,
1837 &dealloc, refcount_loc, false);
1843 cluster_in_el = trunc_end;
1845 ocfs2_reinit_path(path, 1);
1848 ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1851 ocfs2_free_path(path);
1852 ocfs2_schedule_truncate_log_flush(osb, 1);
1853 ocfs2_run_deallocs(osb, &dealloc);
1859 * Parts of this function taken from xfs_change_file_space()
1861 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1862 loff_t f_pos, unsigned int cmd,
1863 struct ocfs2_space_resv *sr,
1869 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1870 struct buffer_head *di_bh = NULL;
1872 unsigned long long max_off = inode->i_sb->s_maxbytes;
1874 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1880 * This prevents concurrent writes on other nodes
1882 ret = ocfs2_rw_lock(inode, 1);
1888 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1894 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1896 goto out_inode_unlock;
1899 switch (sr->l_whence) {
1900 case 0: /*SEEK_SET*/
1902 case 1: /*SEEK_CUR*/
1903 sr->l_start += f_pos;
1905 case 2: /*SEEK_END*/
1906 sr->l_start += i_size_read(inode);
1910 goto out_inode_unlock;
1914 llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1917 || sr->l_start > max_off
1918 || (sr->l_start + llen) < 0
1919 || (sr->l_start + llen) > max_off) {
1921 goto out_inode_unlock;
1923 size = sr->l_start + sr->l_len;
1925 if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64 ||
1926 cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) {
1927 if (sr->l_len <= 0) {
1929 goto out_inode_unlock;
1933 if (file && should_remove_suid(file->f_path.dentry)) {
1934 ret = __ocfs2_write_remove_suid(inode, di_bh);
1937 goto out_inode_unlock;
1941 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1943 case OCFS2_IOC_RESVSP:
1944 case OCFS2_IOC_RESVSP64:
1946 * This takes unsigned offsets, but the signed ones we
1947 * pass have been checked against overflow above.
1949 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1952 case OCFS2_IOC_UNRESVSP:
1953 case OCFS2_IOC_UNRESVSP64:
1954 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1960 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1963 goto out_inode_unlock;
1967 * We update c/mtime for these changes
1969 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1970 if (IS_ERR(handle)) {
1971 ret = PTR_ERR(handle);
1973 goto out_inode_unlock;
1976 if (change_size && i_size_read(inode) < size)
1977 i_size_write(inode, size);
1979 inode->i_ctime = inode->i_mtime = current_time(inode);
1980 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1984 if (file && (file->f_flags & O_SYNC))
1987 ocfs2_commit_trans(osb, handle);
1991 ocfs2_inode_unlock(inode, 1);
1993 ocfs2_rw_unlock(inode, 1);
1996 inode_unlock(inode);
2000 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
2001 struct ocfs2_space_resv *sr)
2003 struct inode *inode = file_inode(file);
2004 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2007 if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
2008 !ocfs2_writes_unwritten_extents(osb))
2010 else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
2011 !ocfs2_sparse_alloc(osb))
2014 if (!S_ISREG(inode->i_mode))
2017 if (!(file->f_mode & FMODE_WRITE))
2020 ret = mnt_want_write_file(file);
2023 ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
2024 mnt_drop_write_file(file);
2028 static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
2031 struct inode *inode = file_inode(file);
2032 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2033 struct ocfs2_space_resv sr;
2034 int change_size = 1;
2035 int cmd = OCFS2_IOC_RESVSP64;
2037 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2039 if (!ocfs2_writes_unwritten_extents(osb))
2042 if (mode & FALLOC_FL_KEEP_SIZE)
2045 if (mode & FALLOC_FL_PUNCH_HOLE)
2046 cmd = OCFS2_IOC_UNRESVSP64;
2049 sr.l_start = (s64)offset;
2050 sr.l_len = (s64)len;
2052 return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2056 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2060 unsigned int extent_flags;
2061 u32 cpos, clusters, extent_len, phys_cpos;
2062 struct super_block *sb = inode->i_sb;
2064 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2065 !ocfs2_is_refcount_inode(inode) ||
2066 OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2069 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2070 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2073 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2080 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2085 if (extent_len > clusters)
2086 extent_len = clusters;
2088 clusters -= extent_len;
2095 static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
2097 int blockmask = inode->i_sb->s_blocksize - 1;
2098 loff_t final_size = pos + count;
2100 if ((pos & blockmask) || (final_size & blockmask))
2105 static int ocfs2_inode_lock_for_extent_tree(struct inode *inode,
2106 struct buffer_head **di_bh,
2114 ret = ocfs2_inode_lock(inode, di_bh, meta_level);
2116 ret = ocfs2_try_inode_lock(inode, di_bh, meta_level);
2122 down_write(&OCFS2_I(inode)->ip_alloc_sem);
2124 down_read(&OCFS2_I(inode)->ip_alloc_sem);
2127 ret = down_write_trylock(&OCFS2_I(inode)->ip_alloc_sem);
2129 ret = down_read_trylock(&OCFS2_I(inode)->ip_alloc_sem);
2142 ocfs2_inode_unlock(inode, meta_level);
2147 static void ocfs2_inode_unlock_for_extent_tree(struct inode *inode,
2148 struct buffer_head **di_bh,
2153 up_write(&OCFS2_I(inode)->ip_alloc_sem);
2155 up_read(&OCFS2_I(inode)->ip_alloc_sem);
2160 if (meta_level >= 0)
2161 ocfs2_inode_unlock(inode, meta_level);
2164 static int ocfs2_prepare_inode_for_write(struct file *file,
2165 loff_t pos, size_t count, int wait)
2167 int ret = 0, meta_level = 0, overwrite_io = 0;
2169 struct dentry *dentry = file->f_path.dentry;
2170 struct inode *inode = d_inode(dentry);
2171 struct buffer_head *di_bh = NULL;
2176 * We start with a read level meta lock and only jump to an ex
2177 * if we need to make modifications here.
2180 ret = ocfs2_inode_lock_for_extent_tree(inode,
2192 * Check if IO will overwrite allocated blocks in case
2193 * IOCB_NOWAIT flag is set.
2195 if (!wait && !overwrite_io) {
2198 ret = ocfs2_overwrite_io(inode, di_bh, pos, count);
2206 /* Clear suid / sgid if necessary. We do this here
2207 * instead of later in the write path because
2208 * remove_suid() calls ->setattr without any hint that
2209 * we may have already done our cluster locking. Since
2210 * ocfs2_setattr() *must* take cluster locks to
2211 * proceed, this will lead us to recursively lock the
2212 * inode. There's also the dinode i_size state which
2213 * can be lost via setattr during extending writes (we
2214 * set inode->i_size at the end of a write. */
2215 if (should_remove_suid(dentry)) {
2216 if (meta_level == 0) {
2217 ocfs2_inode_unlock_for_extent_tree(inode,
2225 ret = ocfs2_write_remove_suid(inode);
2232 ret = ocfs2_check_range_for_refcount(inode, pos, count);
2234 ocfs2_inode_unlock_for_extent_tree(inode,
2240 ret = ocfs2_inode_lock_for_extent_tree(inode,
2251 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2253 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2254 ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
2267 trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
2270 ocfs2_inode_unlock_for_extent_tree(inode,
2279 static ssize_t ocfs2_file_write_iter(struct kiocb *iocb,
2280 struct iov_iter *from)
2283 ssize_t written = 0;
2285 size_t count = iov_iter_count(from);
2286 struct file *file = iocb->ki_filp;
2287 struct inode *inode = file_inode(file);
2288 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2289 int full_coherency = !(osb->s_mount_opt &
2290 OCFS2_MOUNT_COHERENCY_BUFFERED);
2291 void *saved_ki_complete = NULL;
2292 int append_write = ((iocb->ki_pos + count) >=
2293 i_size_read(inode) ? 1 : 0);
2294 int direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2295 int nowait = iocb->ki_flags & IOCB_NOWAIT ? 1 : 0;
2297 trace_ocfs2_file_write_iter(inode, file, file->f_path.dentry,
2298 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2299 file->f_path.dentry->d_name.len,
2300 file->f_path.dentry->d_name.name,
2301 (unsigned int)from->nr_segs); /* GRRRRR */
2303 if (!direct_io && nowait)
2310 if (!inode_trylock(inode))
2316 * Concurrent O_DIRECT writes are allowed with
2317 * mount_option "coherency=buffered".
2318 * For append write, we must take rw EX.
2320 rw_level = (!direct_io || full_coherency || append_write);
2323 ret = ocfs2_try_rw_lock(inode, rw_level);
2325 ret = ocfs2_rw_lock(inode, rw_level);
2333 * O_DIRECT writes with "coherency=full" need to take EX cluster
2334 * inode_lock to guarantee coherency.
2336 if (direct_io && full_coherency) {
2338 * We need to take and drop the inode lock to force
2339 * other nodes to drop their caches. Buffered I/O
2340 * already does this in write_begin().
2343 ret = ocfs2_try_inode_lock(inode, NULL, 1);
2345 ret = ocfs2_inode_lock(inode, NULL, 1);
2352 ocfs2_inode_unlock(inode, 1);
2355 ret = generic_write_checks(iocb, from);
2363 ret = ocfs2_prepare_inode_for_write(file, iocb->ki_pos, count, !nowait);
2370 if (direct_io && !is_sync_kiocb(iocb) &&
2371 ocfs2_is_io_unaligned(inode, count, iocb->ki_pos)) {
2373 * Make it a sync io if it's an unaligned aio.
2375 saved_ki_complete = xchg(&iocb->ki_complete, NULL);
2378 /* communicate with ocfs2_dio_end_io */
2379 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2381 written = __generic_file_write_iter(iocb, from);
2382 /* buffered aio wouldn't have proper lock coverage today */
2383 BUG_ON(written == -EIOCBQUEUED && !direct_io);
2386 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2387 * function pointer which is called when o_direct io completes so that
2388 * it can unlock our rw lock.
2389 * Unfortunately there are error cases which call end_io and others
2390 * that don't. so we don't have to unlock the rw_lock if either an
2391 * async dio is going to do it in the future or an end_io after an
2392 * error has already done it.
2394 if ((written == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2398 if (unlikely(written <= 0))
2401 if (((file->f_flags & O_DSYNC) && !direct_io) ||
2403 ret = filemap_fdatawrite_range(file->f_mapping,
2404 iocb->ki_pos - written,
2410 ret = jbd2_journal_force_commit(osb->journal->j_journal);
2416 ret = filemap_fdatawait_range(file->f_mapping,
2417 iocb->ki_pos - written,
2422 if (saved_ki_complete)
2423 xchg(&iocb->ki_complete, saved_ki_complete);
2426 ocfs2_rw_unlock(inode, rw_level);
2429 inode_unlock(inode);
2436 static ssize_t ocfs2_file_read_iter(struct kiocb *iocb,
2437 struct iov_iter *to)
2439 int ret = 0, rw_level = -1, lock_level = 0;
2440 struct file *filp = iocb->ki_filp;
2441 struct inode *inode = file_inode(filp);
2442 int direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2443 int nowait = iocb->ki_flags & IOCB_NOWAIT ? 1 : 0;
2445 trace_ocfs2_file_read_iter(inode, filp, filp->f_path.dentry,
2446 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2447 filp->f_path.dentry->d_name.len,
2448 filp->f_path.dentry->d_name.name,
2449 to->nr_segs); /* GRRRRR */
2458 if (!direct_io && nowait)
2462 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
2463 * need locks to protect pending reads from racing with truncate.
2467 ret = ocfs2_try_rw_lock(inode, 0);
2469 ret = ocfs2_rw_lock(inode, 0);
2477 /* communicate with ocfs2_dio_end_io */
2478 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2482 * We're fine letting folks race truncates and extending
2483 * writes with read across the cluster, just like they can
2484 * locally. Hence no rw_lock during read.
2486 * Take and drop the meta data lock to update inode fields
2487 * like i_size. This allows the checks down below
2488 * generic_file_read_iter() a chance of actually working.
2490 ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level,
2497 ocfs2_inode_unlock(inode, lock_level);
2499 ret = generic_file_read_iter(iocb, to);
2500 trace_generic_file_read_iter_ret(ret);
2502 /* buffered aio wouldn't have proper lock coverage today */
2503 BUG_ON(ret == -EIOCBQUEUED && !direct_io);
2505 /* see ocfs2_file_write_iter */
2506 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2512 ocfs2_rw_unlock(inode, rw_level);
2517 /* Refer generic_file_llseek_unlocked() */
2518 static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence)
2520 struct inode *inode = file->f_mapping->host;
2529 /* SEEK_END requires the OCFS2 inode lock for the file
2530 * because it references the file's size.
2532 ret = ocfs2_inode_lock(inode, NULL, 0);
2537 offset += i_size_read(inode);
2538 ocfs2_inode_unlock(inode, 0);
2542 offset = file->f_pos;
2545 offset += file->f_pos;
2549 ret = ocfs2_seek_data_hole_offset(file, &offset, whence);
2558 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2561 inode_unlock(inode);
2567 static loff_t ocfs2_remap_file_range(struct file *file_in, loff_t pos_in,
2568 struct file *file_out, loff_t pos_out,
2569 loff_t len, unsigned int remap_flags)
2571 struct inode *inode_in = file_inode(file_in);
2572 struct inode *inode_out = file_inode(file_out);
2573 struct ocfs2_super *osb = OCFS2_SB(inode_in->i_sb);
2574 struct buffer_head *in_bh = NULL, *out_bh = NULL;
2575 bool same_inode = (inode_in == inode_out);
2576 loff_t remapped = 0;
2579 if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
2581 if (!ocfs2_refcount_tree(osb))
2583 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
2586 /* Lock both files against IO */
2587 ret = ocfs2_reflink_inodes_lock(inode_in, &in_bh, inode_out, &out_bh);
2591 /* Check file eligibility and prepare for block sharing. */
2593 if ((OCFS2_I(inode_in)->ip_flags & OCFS2_INODE_SYSTEM_FILE) ||
2594 (OCFS2_I(inode_out)->ip_flags & OCFS2_INODE_SYSTEM_FILE))
2597 ret = generic_remap_file_range_prep(file_in, pos_in, file_out, pos_out,
2599 if (ret < 0 || len == 0)
2602 /* Lock out changes to the allocation maps and remap. */
2603 down_write(&OCFS2_I(inode_in)->ip_alloc_sem);
2605 down_write_nested(&OCFS2_I(inode_out)->ip_alloc_sem,
2606 SINGLE_DEPTH_NESTING);
2608 /* Zap any page cache for the destination file's range. */
2609 truncate_inode_pages_range(&inode_out->i_data,
2610 round_down(pos_out, PAGE_SIZE),
2611 round_up(pos_out + len, PAGE_SIZE) - 1);
2613 remapped = ocfs2_reflink_remap_blocks(inode_in, in_bh, pos_in,
2614 inode_out, out_bh, pos_out, len);
2615 up_write(&OCFS2_I(inode_in)->ip_alloc_sem);
2617 up_write(&OCFS2_I(inode_out)->ip_alloc_sem);
2625 * Empty the extent map so that we may get the right extent
2626 * record from the disk.
2628 ocfs2_extent_map_trunc(inode_in, 0);
2629 ocfs2_extent_map_trunc(inode_out, 0);
2631 ret = ocfs2_reflink_update_dest(inode_out, out_bh, pos_out + len);
2638 ocfs2_reflink_inodes_unlock(inode_in, in_bh, inode_out, out_bh);
2639 return remapped > 0 ? remapped : ret;
2642 const struct inode_operations ocfs2_file_iops = {
2643 .setattr = ocfs2_setattr,
2644 .getattr = ocfs2_getattr,
2645 .permission = ocfs2_permission,
2646 .listxattr = ocfs2_listxattr,
2647 .fiemap = ocfs2_fiemap,
2648 .get_acl = ocfs2_iop_get_acl,
2649 .set_acl = ocfs2_iop_set_acl,
2650 .fileattr_get = ocfs2_fileattr_get,
2651 .fileattr_set = ocfs2_fileattr_set,
2654 const struct inode_operations ocfs2_special_file_iops = {
2655 .setattr = ocfs2_setattr,
2656 .getattr = ocfs2_getattr,
2657 .permission = ocfs2_permission,
2658 .get_acl = ocfs2_iop_get_acl,
2659 .set_acl = ocfs2_iop_set_acl,
2663 * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2664 * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2666 const struct file_operations ocfs2_fops = {
2667 .llseek = ocfs2_file_llseek,
2669 .fsync = ocfs2_sync_file,
2670 .release = ocfs2_file_release,
2671 .open = ocfs2_file_open,
2672 .read_iter = ocfs2_file_read_iter,
2673 .write_iter = ocfs2_file_write_iter,
2674 .unlocked_ioctl = ocfs2_ioctl,
2675 #ifdef CONFIG_COMPAT
2676 .compat_ioctl = ocfs2_compat_ioctl,
2679 .flock = ocfs2_flock,
2680 .splice_read = generic_file_splice_read,
2681 .splice_write = iter_file_splice_write,
2682 .fallocate = ocfs2_fallocate,
2683 .remap_file_range = ocfs2_remap_file_range,
2686 const struct file_operations ocfs2_dops = {
2687 .llseek = generic_file_llseek,
2688 .read = generic_read_dir,
2689 .iterate = ocfs2_readdir,
2690 .fsync = ocfs2_sync_file,
2691 .release = ocfs2_dir_release,
2692 .open = ocfs2_dir_open,
2693 .unlocked_ioctl = ocfs2_ioctl,
2694 #ifdef CONFIG_COMPAT
2695 .compat_ioctl = ocfs2_compat_ioctl,
2698 .flock = ocfs2_flock,
2702 * POSIX-lockless variants of our file_operations.
2704 * These will be used if the underlying cluster stack does not support
2705 * posix file locking, if the user passes the "localflocks" mount
2706 * option, or if we have a local-only fs.
2708 * ocfs2_flock is in here because all stacks handle UNIX file locks,
2709 * so we still want it in the case of no stack support for
2710 * plocks. Internally, it will do the right thing when asked to ignore
2713 const struct file_operations ocfs2_fops_no_plocks = {
2714 .llseek = ocfs2_file_llseek,
2716 .fsync = ocfs2_sync_file,
2717 .release = ocfs2_file_release,
2718 .open = ocfs2_file_open,
2719 .read_iter = ocfs2_file_read_iter,
2720 .write_iter = ocfs2_file_write_iter,
2721 .unlocked_ioctl = ocfs2_ioctl,
2722 #ifdef CONFIG_COMPAT
2723 .compat_ioctl = ocfs2_compat_ioctl,
2725 .flock = ocfs2_flock,
2726 .splice_read = generic_file_splice_read,
2727 .splice_write = iter_file_splice_write,
2728 .fallocate = ocfs2_fallocate,
2729 .remap_file_range = ocfs2_remap_file_range,
2732 const struct file_operations ocfs2_dops_no_plocks = {
2733 .llseek = generic_file_llseek,
2734 .read = generic_read_dir,
2735 .iterate = ocfs2_readdir,
2736 .fsync = ocfs2_sync_file,
2737 .release = ocfs2_dir_release,
2738 .open = ocfs2_dir_open,
2739 .unlocked_ioctl = ocfs2_ioctl,
2740 #ifdef CONFIG_COMPAT
2741 .compat_ioctl = ocfs2_compat_ioctl,
2743 .flock = ocfs2_flock,