1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * File open, close, extend, truncate
8 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation; either
13 * version 2 of the License, or (at your option) any later version.
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
26 #include <linux/capability.h>
28 #include <linux/types.h>
29 #include <linux/slab.h>
30 #include <linux/highmem.h>
31 #include <linux/pagemap.h>
32 #include <linux/uio.h>
33 #include <linux/sched.h>
34 #include <linux/splice.h>
35 #include <linux/mount.h>
36 #include <linux/writeback.h>
37 #include <linux/falloc.h>
38 #include <linux/quotaops.h>
39 #include <linux/blkdev.h>
40 #include <linux/backing-dev.h>
42 #include <cluster/masklog.h>
50 #include "extent_map.h"
63 #include "refcounttree.h"
64 #include "ocfs2_trace.h"
66 #include "buffer_head_io.h"
68 static int ocfs2_init_file_private(struct inode *inode, struct file *file)
70 struct ocfs2_file_private *fp;
72 fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
77 mutex_init(&fp->fp_mutex);
78 ocfs2_file_lock_res_init(&fp->fp_flock, fp);
79 file->private_data = fp;
84 static void ocfs2_free_file_private(struct inode *inode, struct file *file)
86 struct ocfs2_file_private *fp = file->private_data;
87 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
90 ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
91 ocfs2_lock_res_free(&fp->fp_flock);
93 file->private_data = NULL;
97 static int ocfs2_file_open(struct inode *inode, struct file *file)
100 int mode = file->f_flags;
101 struct ocfs2_inode_info *oi = OCFS2_I(inode);
103 trace_ocfs2_file_open(inode, file, file->f_path.dentry,
104 (unsigned long long)oi->ip_blkno,
105 file->f_path.dentry->d_name.len,
106 file->f_path.dentry->d_name.name, mode);
108 if (file->f_mode & FMODE_WRITE) {
109 status = dquot_initialize(inode);
114 spin_lock(&oi->ip_lock);
116 /* Check that the inode hasn't been wiped from disk by another
117 * node. If it hasn't then we're safe as long as we hold the
118 * spin lock until our increment of open count. */
119 if (oi->ip_flags & OCFS2_INODE_DELETED) {
120 spin_unlock(&oi->ip_lock);
127 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
130 spin_unlock(&oi->ip_lock);
132 status = ocfs2_init_file_private(inode, file);
135 * We want to set open count back if we're failing the
138 spin_lock(&oi->ip_lock);
140 spin_unlock(&oi->ip_lock);
143 file->f_mode |= FMODE_NOWAIT;
149 static int ocfs2_file_release(struct inode *inode, struct file *file)
151 struct ocfs2_inode_info *oi = OCFS2_I(inode);
153 spin_lock(&oi->ip_lock);
154 if (!--oi->ip_open_count)
155 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
157 trace_ocfs2_file_release(inode, file, file->f_path.dentry,
159 file->f_path.dentry->d_name.len,
160 file->f_path.dentry->d_name.name,
162 spin_unlock(&oi->ip_lock);
164 ocfs2_free_file_private(inode, file);
169 static int ocfs2_dir_open(struct inode *inode, struct file *file)
171 return ocfs2_init_file_private(inode, file);
174 static int ocfs2_dir_release(struct inode *inode, struct file *file)
176 ocfs2_free_file_private(inode, file);
180 static int ocfs2_sync_file(struct file *file, loff_t start, loff_t end,
184 struct inode *inode = file->f_mapping->host;
185 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
186 struct ocfs2_inode_info *oi = OCFS2_I(inode);
187 journal_t *journal = osb->journal->j_journal;
190 bool needs_barrier = false;
192 trace_ocfs2_sync_file(inode, file, file->f_path.dentry,
194 file->f_path.dentry->d_name.len,
195 file->f_path.dentry->d_name.name,
196 (unsigned long long)datasync);
198 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
201 err = file_write_and_wait_range(file, start, end);
205 commit_tid = datasync ? oi->i_datasync_tid : oi->i_sync_tid;
206 if (journal->j_flags & JBD2_BARRIER &&
207 !jbd2_trans_will_send_data_barrier(journal, commit_tid))
208 needs_barrier = true;
209 err = jbd2_complete_transaction(journal, commit_tid);
211 ret = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
219 return (err < 0) ? -EIO : 0;
222 int ocfs2_should_update_atime(struct inode *inode,
223 struct vfsmount *vfsmnt)
225 struct timespec64 now;
226 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
228 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
231 if ((inode->i_flags & S_NOATIME) ||
232 ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode)))
236 * We can be called with no vfsmnt structure - NFSD will
239 * Note that our action here is different than touch_atime() -
240 * if we can't tell whether this is a noatime mount, then we
241 * don't know whether to trust the value of s_atime_quantum.
246 if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
247 ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
250 if (vfsmnt->mnt_flags & MNT_RELATIME) {
251 if ((timespec64_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
252 (timespec64_compare(&inode->i_atime, &inode->i_ctime) <= 0))
258 now = current_time(inode);
259 if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
265 int ocfs2_update_inode_atime(struct inode *inode,
266 struct buffer_head *bh)
269 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
271 struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
273 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
274 if (IS_ERR(handle)) {
275 ret = PTR_ERR(handle);
280 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
281 OCFS2_JOURNAL_ACCESS_WRITE);
288 * Don't use ocfs2_mark_inode_dirty() here as we don't always
289 * have i_mutex to guard against concurrent changes to other
292 inode->i_atime = current_time(inode);
293 di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
294 di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
295 ocfs2_update_inode_fsync_trans(handle, inode, 0);
296 ocfs2_journal_dirty(handle, bh);
299 ocfs2_commit_trans(osb, handle);
304 int ocfs2_set_inode_size(handle_t *handle,
306 struct buffer_head *fe_bh,
311 i_size_write(inode, new_i_size);
312 inode->i_blocks = ocfs2_inode_sector_count(inode);
313 inode->i_ctime = inode->i_mtime = current_time(inode);
315 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
325 int ocfs2_simple_size_update(struct inode *inode,
326 struct buffer_head *di_bh,
330 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
331 handle_t *handle = NULL;
333 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
334 if (IS_ERR(handle)) {
335 ret = PTR_ERR(handle);
340 ret = ocfs2_set_inode_size(handle, inode, di_bh,
345 ocfs2_update_inode_fsync_trans(handle, inode, 0);
346 ocfs2_commit_trans(osb, handle);
351 static int ocfs2_cow_file_pos(struct inode *inode,
352 struct buffer_head *fe_bh,
356 u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
357 unsigned int num_clusters = 0;
358 unsigned int ext_flags = 0;
361 * If the new offset is aligned to the range of the cluster, there is
362 * no space for ocfs2_zero_range_for_truncate to fill, so no need to
365 if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
368 status = ocfs2_get_clusters(inode, cpos, &phys,
369 &num_clusters, &ext_flags);
375 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
378 return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1);
384 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
386 struct buffer_head *fe_bh,
391 struct ocfs2_dinode *di;
395 * We need to CoW the cluster contains the offset if it is reflinked
396 * since we will call ocfs2_zero_range_for_truncate later which will
397 * write "0" from offset to the end of the cluster.
399 status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
405 /* TODO: This needs to actually orphan the inode in this
408 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
409 if (IS_ERR(handle)) {
410 status = PTR_ERR(handle);
415 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
416 OCFS2_JOURNAL_ACCESS_WRITE);
423 * Do this before setting i_size.
425 cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
426 status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
433 i_size_write(inode, new_i_size);
434 inode->i_ctime = inode->i_mtime = current_time(inode);
436 di = (struct ocfs2_dinode *) fe_bh->b_data;
437 di->i_size = cpu_to_le64(new_i_size);
438 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
439 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
440 ocfs2_update_inode_fsync_trans(handle, inode, 0);
442 ocfs2_journal_dirty(handle, fe_bh);
445 ocfs2_commit_trans(osb, handle);
450 int ocfs2_truncate_file(struct inode *inode,
451 struct buffer_head *di_bh,
455 struct ocfs2_dinode *fe = NULL;
456 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
458 /* We trust di_bh because it comes from ocfs2_inode_lock(), which
459 * already validated it */
460 fe = (struct ocfs2_dinode *) di_bh->b_data;
462 trace_ocfs2_truncate_file((unsigned long long)OCFS2_I(inode)->ip_blkno,
463 (unsigned long long)le64_to_cpu(fe->i_size),
464 (unsigned long long)new_i_size);
466 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
467 "Inode %llu, inode i_size = %lld != di "
468 "i_size = %llu, i_flags = 0x%x\n",
469 (unsigned long long)OCFS2_I(inode)->ip_blkno,
471 (unsigned long long)le64_to_cpu(fe->i_size),
472 le32_to_cpu(fe->i_flags));
474 if (new_i_size > le64_to_cpu(fe->i_size)) {
475 trace_ocfs2_truncate_file_error(
476 (unsigned long long)le64_to_cpu(fe->i_size),
477 (unsigned long long)new_i_size);
483 down_write(&OCFS2_I(inode)->ip_alloc_sem);
485 ocfs2_resv_discard(&osb->osb_la_resmap,
486 &OCFS2_I(inode)->ip_la_data_resv);
489 * The inode lock forced other nodes to sync and drop their
490 * pages, which (correctly) happens even if we have a truncate
491 * without allocation change - ocfs2 cluster sizes can be much
492 * greater than page size, so we have to truncate them
496 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
497 unmap_mapping_range(inode->i_mapping,
498 new_i_size + PAGE_SIZE - 1, 0, 1);
499 truncate_inode_pages(inode->i_mapping, new_i_size);
500 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
501 i_size_read(inode), 1);
505 goto bail_unlock_sem;
508 /* alright, we're going to need to do a full blown alloc size
509 * change. Orphan the inode so that recovery can complete the
510 * truncate if necessary. This does the task of marking
512 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
515 goto bail_unlock_sem;
518 unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
519 truncate_inode_pages(inode->i_mapping, new_i_size);
521 status = ocfs2_commit_truncate(osb, inode, di_bh);
524 goto bail_unlock_sem;
527 /* TODO: orphan dir cleanup here. */
529 up_write(&OCFS2_I(inode)->ip_alloc_sem);
532 if (!status && OCFS2_I(inode)->ip_clusters == 0)
533 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
539 * extend file allocation only here.
540 * we'll update all the disk stuff, and oip->alloc_size
542 * expect stuff to be locked, a transaction started and enough data /
543 * metadata reservations in the contexts.
545 * Will return -EAGAIN, and a reason if a restart is needed.
546 * If passed in, *reason will always be set, even in error.
548 int ocfs2_add_inode_data(struct ocfs2_super *osb,
553 struct buffer_head *fe_bh,
555 struct ocfs2_alloc_context *data_ac,
556 struct ocfs2_alloc_context *meta_ac,
557 enum ocfs2_alloc_restarted *reason_ret)
560 struct ocfs2_extent_tree et;
562 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
563 ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
564 clusters_to_add, mark_unwritten,
565 data_ac, meta_ac, reason_ret);
570 static int ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
571 u32 clusters_to_add, int mark_unwritten)
574 int restart_func = 0;
577 struct buffer_head *bh = NULL;
578 struct ocfs2_dinode *fe = NULL;
579 handle_t *handle = NULL;
580 struct ocfs2_alloc_context *data_ac = NULL;
581 struct ocfs2_alloc_context *meta_ac = NULL;
582 enum ocfs2_alloc_restarted why = RESTART_NONE;
583 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
584 struct ocfs2_extent_tree et;
588 * Unwritten extent only exists for file systems which
591 BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
593 status = ocfs2_read_inode_block(inode, &bh);
598 fe = (struct ocfs2_dinode *) bh->b_data;
601 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
603 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
604 status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
611 credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list);
612 handle = ocfs2_start_trans(osb, credits);
613 if (IS_ERR(handle)) {
614 status = PTR_ERR(handle);
620 restarted_transaction:
621 trace_ocfs2_extend_allocation(
622 (unsigned long long)OCFS2_I(inode)->ip_blkno,
623 (unsigned long long)i_size_read(inode),
624 le32_to_cpu(fe->i_clusters), clusters_to_add,
627 status = dquot_alloc_space_nodirty(inode,
628 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
633 /* reserve a write to the file entry early on - that we if we
634 * run out of credits in the allocation path, we can still
636 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
637 OCFS2_JOURNAL_ACCESS_WRITE);
643 prev_clusters = OCFS2_I(inode)->ip_clusters;
645 status = ocfs2_add_inode_data(osb,
655 if ((status < 0) && (status != -EAGAIN)) {
656 if (status != -ENOSPC)
660 ocfs2_update_inode_fsync_trans(handle, inode, 1);
661 ocfs2_journal_dirty(handle, bh);
663 spin_lock(&OCFS2_I(inode)->ip_lock);
664 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
665 spin_unlock(&OCFS2_I(inode)->ip_lock);
666 /* Release unused quota reservation */
667 dquot_free_space(inode,
668 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
671 if (why != RESTART_NONE && clusters_to_add) {
672 if (why == RESTART_META) {
676 BUG_ON(why != RESTART_TRANS);
678 status = ocfs2_allocate_extend_trans(handle, 1);
680 /* handle still has to be committed at
686 goto restarted_transaction;
690 trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno,
691 le32_to_cpu(fe->i_clusters),
692 (unsigned long long)le64_to_cpu(fe->i_size),
693 OCFS2_I(inode)->ip_clusters,
694 (unsigned long long)i_size_read(inode));
697 if (status < 0 && did_quota)
698 dquot_free_space(inode,
699 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
701 ocfs2_commit_trans(osb, handle);
705 ocfs2_free_alloc_context(data_ac);
709 ocfs2_free_alloc_context(meta_ac);
712 if ((!status) && restart_func) {
723 * While a write will already be ordering the data, a truncate will not.
724 * Thus, we need to explicitly order the zeroed pages.
726 static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode,
727 struct buffer_head *di_bh)
729 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
730 handle_t *handle = NULL;
733 if (!ocfs2_should_order_data(inode))
736 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
737 if (IS_ERR(handle)) {
743 ret = ocfs2_jbd2_file_inode(handle, inode);
749 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
750 OCFS2_JOURNAL_ACCESS_WRITE);
753 ocfs2_update_inode_fsync_trans(handle, inode, 1);
758 ocfs2_commit_trans(osb, handle);
759 handle = ERR_PTR(ret);
764 /* Some parts of this taken from generic_cont_expand, which turned out
765 * to be too fragile to do exactly what we need without us having to
766 * worry about recursive locking in ->write_begin() and ->write_end(). */
767 static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
768 u64 abs_to, struct buffer_head *di_bh)
770 struct address_space *mapping = inode->i_mapping;
772 unsigned long index = abs_from >> PAGE_SHIFT;
775 unsigned zero_from, zero_to, block_start, block_end;
776 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
778 BUG_ON(abs_from >= abs_to);
779 BUG_ON(abs_to > (((u64)index + 1) << PAGE_SHIFT));
780 BUG_ON(abs_from & (inode->i_blkbits - 1));
782 handle = ocfs2_zero_start_ordered_transaction(inode, di_bh);
783 if (IS_ERR(handle)) {
784 ret = PTR_ERR(handle);
788 page = find_or_create_page(mapping, index, GFP_NOFS);
792 goto out_commit_trans;
795 /* Get the offsets within the page that we want to zero */
796 zero_from = abs_from & (PAGE_SIZE - 1);
797 zero_to = abs_to & (PAGE_SIZE - 1);
801 trace_ocfs2_write_zero_page(
802 (unsigned long long)OCFS2_I(inode)->ip_blkno,
803 (unsigned long long)abs_from,
804 (unsigned long long)abs_to,
805 index, zero_from, zero_to);
807 /* We know that zero_from is block aligned */
808 for (block_start = zero_from; block_start < zero_to;
809 block_start = block_end) {
810 block_end = block_start + i_blocksize(inode);
813 * block_start is block-aligned. Bump it by one to force
814 * __block_write_begin and block_commit_write to zero the
817 ret = __block_write_begin(page, block_start + 1, 0,
825 /* must not update i_size! */
826 ret = block_commit_write(page, block_start + 1,
835 * fs-writeback will release the dirty pages without page lock
836 * whose offset are over inode size, the release happens at
837 * block_write_full_page().
839 i_size_write(inode, abs_to);
840 inode->i_blocks = ocfs2_inode_sector_count(inode);
841 di->i_size = cpu_to_le64((u64)i_size_read(inode));
842 inode->i_mtime = inode->i_ctime = current_time(inode);
843 di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
844 di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
845 di->i_mtime_nsec = di->i_ctime_nsec;
847 ocfs2_journal_dirty(handle, di_bh);
848 ocfs2_update_inode_fsync_trans(handle, inode, 1);
856 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
862 * Find the next range to zero. We do this in terms of bytes because
863 * that's what ocfs2_zero_extend() wants, and it is dealing with the
864 * pagecache. We may return multiple extents.
866 * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
867 * needs to be zeroed. range_start and range_end return the next zeroing
868 * range. A subsequent call should pass the previous range_end as its
869 * zero_start. If range_end is 0, there's nothing to do.
871 * Unwritten extents are skipped over. Refcounted extents are CoWd.
873 static int ocfs2_zero_extend_get_range(struct inode *inode,
874 struct buffer_head *di_bh,
875 u64 zero_start, u64 zero_end,
876 u64 *range_start, u64 *range_end)
878 int rc = 0, needs_cow = 0;
879 u32 p_cpos, zero_clusters = 0;
881 zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
882 u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
883 unsigned int num_clusters = 0;
884 unsigned int ext_flags = 0;
886 while (zero_cpos < last_cpos) {
887 rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
888 &num_clusters, &ext_flags);
894 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
895 zero_clusters = num_clusters;
896 if (ext_flags & OCFS2_EXT_REFCOUNTED)
901 zero_cpos += num_clusters;
903 if (!zero_clusters) {
908 while ((zero_cpos + zero_clusters) < last_cpos) {
909 rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
910 &p_cpos, &num_clusters,
917 if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
919 if (ext_flags & OCFS2_EXT_REFCOUNTED)
921 zero_clusters += num_clusters;
923 if ((zero_cpos + zero_clusters) > last_cpos)
924 zero_clusters = last_cpos - zero_cpos;
927 rc = ocfs2_refcount_cow(inode, di_bh, zero_cpos,
928 zero_clusters, UINT_MAX);
935 *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
936 *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
937 zero_cpos + zero_clusters);
944 * Zero one range returned from ocfs2_zero_extend_get_range(). The caller
945 * has made sure that the entire range needs zeroing.
947 static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
948 u64 range_end, struct buffer_head *di_bh)
952 u64 zero_pos = range_start;
954 trace_ocfs2_zero_extend_range(
955 (unsigned long long)OCFS2_I(inode)->ip_blkno,
956 (unsigned long long)range_start,
957 (unsigned long long)range_end);
958 BUG_ON(range_start >= range_end);
960 while (zero_pos < range_end) {
961 next_pos = (zero_pos & PAGE_MASK) + PAGE_SIZE;
962 if (next_pos > range_end)
963 next_pos = range_end;
964 rc = ocfs2_write_zero_page(inode, zero_pos, next_pos, di_bh);
972 * Very large extends have the potential to lock up
973 * the cpu for extended periods of time.
981 int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
985 u64 zero_start, range_start = 0, range_end = 0;
986 struct super_block *sb = inode->i_sb;
988 zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
989 trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
990 (unsigned long long)zero_start,
991 (unsigned long long)i_size_read(inode));
992 while (zero_start < zero_to_size) {
993 ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
1004 if (range_start < zero_start)
1005 range_start = zero_start;
1006 if (range_end > zero_to_size)
1007 range_end = zero_to_size;
1009 ret = ocfs2_zero_extend_range(inode, range_start,
1015 zero_start = range_end;
1021 int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
1022 u64 new_i_size, u64 zero_to)
1025 u32 clusters_to_add;
1026 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1029 * Only quota files call this without a bh, and they can't be
1032 BUG_ON(!di_bh && ocfs2_is_refcount_inode(inode));
1033 BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1035 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1036 if (clusters_to_add < oi->ip_clusters)
1037 clusters_to_add = 0;
1039 clusters_to_add -= oi->ip_clusters;
1041 if (clusters_to_add) {
1042 ret = ocfs2_extend_allocation(inode, oi->ip_clusters,
1043 clusters_to_add, 0);
1051 * Call this even if we don't add any clusters to the tree. We
1052 * still need to zero the area between the old i_size and the
1055 ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1063 static int ocfs2_extend_file(struct inode *inode,
1064 struct buffer_head *di_bh,
1068 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1072 /* setattr sometimes calls us like this. */
1073 if (new_i_size == 0)
1076 if (i_size_read(inode) == new_i_size)
1078 BUG_ON(new_i_size < i_size_read(inode));
1081 * The alloc sem blocks people in read/write from reading our
1082 * allocation until we're done changing it. We depend on
1083 * i_mutex to block other extend/truncate calls while we're
1084 * here. We even have to hold it for sparse files because there
1085 * might be some tail zeroing.
1087 down_write(&oi->ip_alloc_sem);
1089 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1091 * We can optimize small extends by keeping the inodes
1094 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1095 up_write(&oi->ip_alloc_sem);
1096 goto out_update_size;
1099 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1101 up_write(&oi->ip_alloc_sem);
1107 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1108 ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1110 ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1113 up_write(&oi->ip_alloc_sem);
1121 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1129 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1131 int status = 0, size_change;
1132 int inode_locked = 0;
1133 struct inode *inode = d_inode(dentry);
1134 struct super_block *sb = inode->i_sb;
1135 struct ocfs2_super *osb = OCFS2_SB(sb);
1136 struct buffer_head *bh = NULL;
1137 handle_t *handle = NULL;
1138 struct dquot *transfer_to[MAXQUOTAS] = { };
1141 struct ocfs2_lock_holder oh;
1143 trace_ocfs2_setattr(inode, dentry,
1144 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1145 dentry->d_name.len, dentry->d_name.name,
1146 attr->ia_valid, attr->ia_mode,
1147 from_kuid(&init_user_ns, attr->ia_uid),
1148 from_kgid(&init_user_ns, attr->ia_gid));
1150 /* ensuring we don't even attempt to truncate a symlink */
1151 if (S_ISLNK(inode->i_mode))
1152 attr->ia_valid &= ~ATTR_SIZE;
1154 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1155 | ATTR_GID | ATTR_UID | ATTR_MODE)
1156 if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
1159 status = setattr_prepare(dentry, attr);
1163 if (is_quota_modification(inode, attr)) {
1164 status = dquot_initialize(inode);
1168 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1171 * Here we should wait dio to finish before inode lock
1172 * to avoid a deadlock between ocfs2_setattr() and
1173 * ocfs2_dio_end_io_write()
1175 inode_dio_wait(inode);
1177 status = ocfs2_rw_lock(inode, 1);
1184 had_lock = ocfs2_inode_lock_tracker(inode, &bh, 1, &oh);
1187 goto bail_unlock_rw;
1188 } else if (had_lock) {
1190 * As far as we know, ocfs2_setattr() could only be the first
1191 * VFS entry point in the call chain of recursive cluster
1199 * ocfs2_iop_get_acl()
1201 * But, we're not 100% sure if it's always true, because the
1202 * ordering of the VFS entry points in the call chain is out
1203 * of our control. So, we'd better dump the stack here to
1204 * catch the other cases of recursive locking.
1206 mlog(ML_ERROR, "Another case of recursive locking:\n");
1212 status = inode_newsize_ok(inode, attr->ia_size);
1216 if (i_size_read(inode) >= attr->ia_size) {
1217 if (ocfs2_should_order_data(inode)) {
1218 status = ocfs2_begin_ordered_truncate(inode,
1223 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1225 status = ocfs2_extend_file(inode, bh, attr->ia_size);
1227 if (status != -ENOSPC)
1234 if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
1235 (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
1237 * Gather pointers to quota structures so that allocation /
1238 * freeing of quota structures happens here and not inside
1239 * dquot_transfer() where we have problems with lock ordering
1241 if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)
1242 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1243 OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1244 transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid));
1245 if (IS_ERR(transfer_to[USRQUOTA])) {
1246 status = PTR_ERR(transfer_to[USRQUOTA]);
1250 if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid)
1251 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1252 OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1253 transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid));
1254 if (IS_ERR(transfer_to[GRPQUOTA])) {
1255 status = PTR_ERR(transfer_to[GRPQUOTA]);
1259 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1260 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1261 2 * ocfs2_quota_trans_credits(sb));
1262 if (IS_ERR(handle)) {
1263 status = PTR_ERR(handle);
1265 goto bail_unlock_alloc;
1267 status = __dquot_transfer(inode, transfer_to);
1271 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1272 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1273 if (IS_ERR(handle)) {
1274 status = PTR_ERR(handle);
1276 goto bail_unlock_alloc;
1280 setattr_copy(inode, attr);
1281 mark_inode_dirty(inode);
1283 status = ocfs2_mark_inode_dirty(handle, inode, bh);
1288 ocfs2_commit_trans(osb, handle);
1290 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1292 if (status && inode_locked) {
1293 ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
1298 ocfs2_rw_unlock(inode, 1);
1301 /* Release quota pointers in case we acquired them */
1302 for (qtype = 0; qtype < OCFS2_MAXQUOTAS; qtype++)
1303 dqput(transfer_to[qtype]);
1305 if (!status && attr->ia_valid & ATTR_MODE) {
1306 status = ocfs2_acl_chmod(inode, bh);
1311 ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
1317 int ocfs2_getattr(const struct path *path, struct kstat *stat,
1318 u32 request_mask, unsigned int flags)
1320 struct inode *inode = d_inode(path->dentry);
1321 struct super_block *sb = path->dentry->d_sb;
1322 struct ocfs2_super *osb = sb->s_fs_info;
1325 err = ocfs2_inode_revalidate(path->dentry);
1332 generic_fillattr(inode, stat);
1334 * If there is inline data in the inode, the inode will normally not
1335 * have data blocks allocated (it may have an external xattr block).
1336 * Report at least one sector for such files, so tools like tar, rsync,
1337 * others don't incorrectly think the file is completely sparse.
1339 if (unlikely(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1340 stat->blocks += (stat->size + 511)>>9;
1342 /* We set the blksize from the cluster size for performance */
1343 stat->blksize = osb->s_clustersize;
1349 int ocfs2_permission(struct inode *inode, int mask)
1352 struct ocfs2_lock_holder oh;
1354 if (mask & MAY_NOT_BLOCK)
1357 had_lock = ocfs2_inode_lock_tracker(inode, NULL, 0, &oh);
1361 } else if (had_lock) {
1362 /* See comments in ocfs2_setattr() for details.
1363 * The call chain of this case could be:
1366 * inode_permission()
1367 * ocfs2_permission()
1368 * ocfs2_iop_get_acl()
1370 mlog(ML_ERROR, "Another case of recursive locking:\n");
1374 ret = generic_permission(inode, mask);
1376 ocfs2_inode_unlock_tracker(inode, 0, &oh, had_lock);
1381 static int __ocfs2_write_remove_suid(struct inode *inode,
1382 struct buffer_head *bh)
1386 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1387 struct ocfs2_dinode *di;
1389 trace_ocfs2_write_remove_suid(
1390 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1393 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1394 if (IS_ERR(handle)) {
1395 ret = PTR_ERR(handle);
1400 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1401 OCFS2_JOURNAL_ACCESS_WRITE);
1407 inode->i_mode &= ~S_ISUID;
1408 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1409 inode->i_mode &= ~S_ISGID;
1411 di = (struct ocfs2_dinode *) bh->b_data;
1412 di->i_mode = cpu_to_le16(inode->i_mode);
1413 ocfs2_update_inode_fsync_trans(handle, inode, 0);
1415 ocfs2_journal_dirty(handle, bh);
1418 ocfs2_commit_trans(osb, handle);
1423 static int ocfs2_write_remove_suid(struct inode *inode)
1426 struct buffer_head *bh = NULL;
1428 ret = ocfs2_read_inode_block(inode, &bh);
1434 ret = __ocfs2_write_remove_suid(inode, bh);
1441 * Allocate enough extents to cover the region starting at byte offset
1442 * start for len bytes. Existing extents are skipped, any extents
1443 * added are marked as "unwritten".
1445 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1449 u32 cpos, phys_cpos, clusters, alloc_size;
1450 u64 end = start + len;
1451 struct buffer_head *di_bh = NULL;
1453 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1454 ret = ocfs2_read_inode_block(inode, &di_bh);
1461 * Nothing to do if the requested reservation range
1462 * fits within the inode.
1464 if (ocfs2_size_fits_inline_data(di_bh, end))
1467 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1475 * We consider both start and len to be inclusive.
1477 cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1478 clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1482 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1490 * Hole or existing extent len can be arbitrary, so
1491 * cap it to our own allocation request.
1493 if (alloc_size > clusters)
1494 alloc_size = clusters;
1498 * We already have an allocation at this
1499 * region so we can safely skip it.
1504 ret = ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1513 clusters -= alloc_size;
1524 * Truncate a byte range, avoiding pages within partial clusters. This
1525 * preserves those pages for the zeroing code to write to.
1527 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1530 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1532 struct address_space *mapping = inode->i_mapping;
1534 start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1535 end = byte_start + byte_len;
1536 end = end & ~(osb->s_clustersize - 1);
1539 unmap_mapping_range(mapping, start, end - start, 0);
1540 truncate_inode_pages_range(mapping, start, end - 1);
1545 * zero out partial blocks of one cluster.
1547 * start: file offset where zero starts, will be made upper block aligned.
1548 * len: it will be trimmed to the end of current cluster if "start + len"
1549 * is bigger than it.
1551 static int ocfs2_zeroout_partial_cluster(struct inode *inode,
1555 u64 start_block, end_block, nr_blocks;
1556 u64 p_block, offset;
1557 u32 cluster, p_cluster, nr_clusters;
1558 struct super_block *sb = inode->i_sb;
1559 u64 end = ocfs2_align_bytes_to_clusters(sb, start);
1561 if (start + len < end)
1564 start_block = ocfs2_blocks_for_bytes(sb, start);
1565 end_block = ocfs2_blocks_for_bytes(sb, end);
1566 nr_blocks = end_block - start_block;
1570 cluster = ocfs2_bytes_to_clusters(sb, start);
1571 ret = ocfs2_get_clusters(inode, cluster, &p_cluster,
1572 &nr_clusters, NULL);
1578 offset = start_block - ocfs2_clusters_to_blocks(sb, cluster);
1579 p_block = ocfs2_clusters_to_blocks(sb, p_cluster) + offset;
1580 return sb_issue_zeroout(sb, p_block, nr_blocks, GFP_NOFS);
1583 static int ocfs2_zero_partial_clusters(struct inode *inode,
1588 u64 end = start + len;
1589 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1590 unsigned int csize = osb->s_clustersize;
1592 loff_t isize = i_size_read(inode);
1595 * The "start" and "end" values are NOT necessarily part of
1596 * the range whose allocation is being deleted. Rather, this
1597 * is what the user passed in with the request. We must zero
1598 * partial clusters here. There's no need to worry about
1599 * physical allocation - the zeroing code knows to skip holes.
1601 trace_ocfs2_zero_partial_clusters(
1602 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1603 (unsigned long long)start, (unsigned long long)end);
1606 * If both edges are on a cluster boundary then there's no
1607 * zeroing required as the region is part of the allocation to
1610 if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1613 /* No page cache for EOF blocks, issue zero out to disk. */
1616 * zeroout eof blocks in last cluster starting from
1617 * "isize" even "start" > "isize" because it is
1618 * complicated to zeroout just at "start" as "start"
1619 * may be not aligned with block size, buffer write
1620 * would be required to do that, but out of eof buffer
1621 * write is not supported.
1623 ret = ocfs2_zeroout_partial_cluster(inode, isize,
1633 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1634 if (IS_ERR(handle)) {
1635 ret = PTR_ERR(handle);
1641 * If start is on a cluster boundary and end is somewhere in another
1642 * cluster, we have not COWed the cluster starting at start, unless
1643 * end is also within the same cluster. So, in this case, we skip this
1644 * first call to ocfs2_zero_range_for_truncate() truncate and move on
1647 if ((start & (csize - 1)) != 0) {
1649 * We want to get the byte offset of the end of the 1st
1652 tmpend = (u64)osb->s_clustersize +
1653 (start & ~(osb->s_clustersize - 1));
1657 trace_ocfs2_zero_partial_clusters_range1(
1658 (unsigned long long)start,
1659 (unsigned long long)tmpend);
1661 ret = ocfs2_zero_range_for_truncate(inode, handle, start,
1669 * This may make start and end equal, but the zeroing
1670 * code will skip any work in that case so there's no
1671 * need to catch it up here.
1673 start = end & ~(osb->s_clustersize - 1);
1675 trace_ocfs2_zero_partial_clusters_range2(
1676 (unsigned long long)start, (unsigned long long)end);
1678 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1682 ocfs2_update_inode_fsync_trans(handle, inode, 1);
1684 ocfs2_commit_trans(osb, handle);
1689 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1692 struct ocfs2_extent_rec *rec = NULL;
1694 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1696 rec = &el->l_recs[i];
1698 if (le32_to_cpu(rec->e_cpos) < pos)
1706 * Helper to calculate the punching pos and length in one run, we handle the
1707 * following three cases in order:
1709 * - remove the entire record
1710 * - remove a partial record
1711 * - no record needs to be removed (hole-punching completed)
1713 static void ocfs2_calc_trunc_pos(struct inode *inode,
1714 struct ocfs2_extent_list *el,
1715 struct ocfs2_extent_rec *rec,
1716 u32 trunc_start, u32 *trunc_cpos,
1717 u32 *trunc_len, u32 *trunc_end,
1718 u64 *blkno, int *done)
1723 range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1725 if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1727 * remove an entire extent record.
1729 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1731 * Skip holes if any.
1733 if (range < *trunc_end)
1735 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1736 *blkno = le64_to_cpu(rec->e_blkno);
1737 *trunc_end = le32_to_cpu(rec->e_cpos);
1738 } else if (range > trunc_start) {
1740 * remove a partial extent record, which means we're
1741 * removing the last extent record.
1743 *trunc_cpos = trunc_start;
1747 if (range < *trunc_end)
1749 *trunc_len = *trunc_end - trunc_start;
1750 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1751 *blkno = le64_to_cpu(rec->e_blkno) +
1752 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1753 *trunc_end = trunc_start;
1756 * It may have two following possibilities:
1758 * - last record has been removed
1759 * - trunc_start was within a hole
1761 * both two cases mean the completion of hole punching.
1769 int ocfs2_remove_inode_range(struct inode *inode,
1770 struct buffer_head *di_bh, u64 byte_start,
1773 int ret = 0, flags = 0, done = 0, i;
1774 u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1776 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1777 struct ocfs2_cached_dealloc_ctxt dealloc;
1778 struct address_space *mapping = inode->i_mapping;
1779 struct ocfs2_extent_tree et;
1780 struct ocfs2_path *path = NULL;
1781 struct ocfs2_extent_list *el = NULL;
1782 struct ocfs2_extent_rec *rec = NULL;
1783 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1784 u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1786 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1787 ocfs2_init_dealloc_ctxt(&dealloc);
1789 trace_ocfs2_remove_inode_range(
1790 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1791 (unsigned long long)byte_start,
1792 (unsigned long long)byte_len);
1797 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1798 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1799 byte_start + byte_len, 0);
1805 * There's no need to get fancy with the page cache
1806 * truncate of an inline-data inode. We're talking
1807 * about less than a page here, which will be cached
1808 * in the dinode buffer anyway.
1810 unmap_mapping_range(mapping, 0, 0, 0);
1811 truncate_inode_pages(mapping, 0);
1816 * For reflinks, we may need to CoW 2 clusters which might be
1817 * partially zero'd later, if hole's start and end offset were
1818 * within one cluster(means is not exactly aligned to clustersize).
1821 if (ocfs2_is_refcount_inode(inode)) {
1822 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1828 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1835 trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1836 trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1837 cluster_in_el = trunc_end;
1839 ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1845 path = ocfs2_new_path_from_et(&et);
1852 while (trunc_end > trunc_start) {
1854 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1861 el = path_leaf_el(path);
1863 i = ocfs2_find_rec(el, trunc_end);
1865 * Need to go to previous extent block.
1868 if (path->p_tree_depth == 0)
1871 ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1880 * We've reached the leftmost extent block,
1881 * it's safe to leave.
1883 if (cluster_in_el == 0)
1887 * The 'pos' searched for previous extent block is
1888 * always one cluster less than actual trunc_end.
1890 trunc_end = cluster_in_el + 1;
1892 ocfs2_reinit_path(path, 1);
1897 rec = &el->l_recs[i];
1899 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1900 &trunc_len, &trunc_end, &blkno, &done);
1904 flags = rec->e_flags;
1905 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1907 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1908 phys_cpos, trunc_len, flags,
1909 &dealloc, refcount_loc, false);
1915 cluster_in_el = trunc_end;
1917 ocfs2_reinit_path(path, 1);
1920 ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1923 ocfs2_free_path(path);
1924 ocfs2_schedule_truncate_log_flush(osb, 1);
1925 ocfs2_run_deallocs(osb, &dealloc);
1931 * Parts of this function taken from xfs_change_file_space()
1933 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1934 loff_t f_pos, unsigned int cmd,
1935 struct ocfs2_space_resv *sr,
1940 loff_t size, orig_isize;
1941 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1942 struct buffer_head *di_bh = NULL;
1944 unsigned long long max_off = inode->i_sb->s_maxbytes;
1946 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1952 * This prevents concurrent writes on other nodes
1954 ret = ocfs2_rw_lock(inode, 1);
1960 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1966 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1968 goto out_inode_unlock;
1971 switch (sr->l_whence) {
1972 case 0: /*SEEK_SET*/
1974 case 1: /*SEEK_CUR*/
1975 sr->l_start += f_pos;
1977 case 2: /*SEEK_END*/
1978 sr->l_start += i_size_read(inode);
1982 goto out_inode_unlock;
1986 llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1989 || sr->l_start > max_off
1990 || (sr->l_start + llen) < 0
1991 || (sr->l_start + llen) > max_off) {
1993 goto out_inode_unlock;
1995 size = sr->l_start + sr->l_len;
1997 if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64 ||
1998 cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) {
1999 if (sr->l_len <= 0) {
2001 goto out_inode_unlock;
2005 if (file && should_remove_suid(file->f_path.dentry)) {
2006 ret = __ocfs2_write_remove_suid(inode, di_bh);
2009 goto out_inode_unlock;
2013 down_write(&OCFS2_I(inode)->ip_alloc_sem);
2015 case OCFS2_IOC_RESVSP:
2016 case OCFS2_IOC_RESVSP64:
2018 * This takes unsigned offsets, but the signed ones we
2019 * pass have been checked against overflow above.
2021 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
2024 case OCFS2_IOC_UNRESVSP:
2025 case OCFS2_IOC_UNRESVSP64:
2026 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
2033 orig_isize = i_size_read(inode);
2034 /* zeroout eof blocks in the cluster. */
2035 if (!ret && change_size && orig_isize < size) {
2036 ret = ocfs2_zeroout_partial_cluster(inode, orig_isize,
2039 i_size_write(inode, size);
2041 up_write(&OCFS2_I(inode)->ip_alloc_sem);
2044 goto out_inode_unlock;
2048 * We update c/mtime for these changes
2050 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
2051 if (IS_ERR(handle)) {
2052 ret = PTR_ERR(handle);
2054 goto out_inode_unlock;
2057 inode->i_ctime = inode->i_mtime = current_time(inode);
2058 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
2062 if (file && (file->f_flags & O_SYNC))
2065 ocfs2_commit_trans(osb, handle);
2069 ocfs2_inode_unlock(inode, 1);
2071 ocfs2_rw_unlock(inode, 1);
2074 inode_unlock(inode);
2078 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
2079 struct ocfs2_space_resv *sr)
2081 struct inode *inode = file_inode(file);
2082 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2085 if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
2086 !ocfs2_writes_unwritten_extents(osb))
2088 else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
2089 !ocfs2_sparse_alloc(osb))
2092 if (!S_ISREG(inode->i_mode))
2095 if (!(file->f_mode & FMODE_WRITE))
2098 ret = mnt_want_write_file(file);
2101 ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
2102 mnt_drop_write_file(file);
2106 static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
2109 struct inode *inode = file_inode(file);
2110 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2111 struct ocfs2_space_resv sr;
2112 int change_size = 1;
2113 int cmd = OCFS2_IOC_RESVSP64;
2115 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2117 if (!ocfs2_writes_unwritten_extents(osb))
2120 if (mode & FALLOC_FL_KEEP_SIZE)
2123 if (mode & FALLOC_FL_PUNCH_HOLE)
2124 cmd = OCFS2_IOC_UNRESVSP64;
2127 sr.l_start = (s64)offset;
2128 sr.l_len = (s64)len;
2130 return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2134 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2138 unsigned int extent_flags;
2139 u32 cpos, clusters, extent_len, phys_cpos;
2140 struct super_block *sb = inode->i_sb;
2142 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2143 !ocfs2_is_refcount_inode(inode) ||
2144 OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2147 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2148 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2151 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2158 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2163 if (extent_len > clusters)
2164 extent_len = clusters;
2166 clusters -= extent_len;
2173 static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
2175 int blockmask = inode->i_sb->s_blocksize - 1;
2176 loff_t final_size = pos + count;
2178 if ((pos & blockmask) || (final_size & blockmask))
2183 static int ocfs2_inode_lock_for_extent_tree(struct inode *inode,
2184 struct buffer_head **di_bh,
2192 ret = ocfs2_inode_lock(inode, di_bh, meta_level);
2194 ret = ocfs2_try_inode_lock(inode, di_bh, meta_level);
2200 down_write(&OCFS2_I(inode)->ip_alloc_sem);
2202 down_read(&OCFS2_I(inode)->ip_alloc_sem);
2205 ret = down_write_trylock(&OCFS2_I(inode)->ip_alloc_sem);
2207 ret = down_read_trylock(&OCFS2_I(inode)->ip_alloc_sem);
2220 ocfs2_inode_unlock(inode, meta_level);
2225 static void ocfs2_inode_unlock_for_extent_tree(struct inode *inode,
2226 struct buffer_head **di_bh,
2231 up_write(&OCFS2_I(inode)->ip_alloc_sem);
2233 up_read(&OCFS2_I(inode)->ip_alloc_sem);
2238 if (meta_level >= 0)
2239 ocfs2_inode_unlock(inode, meta_level);
2242 static int ocfs2_prepare_inode_for_write(struct file *file,
2243 loff_t pos, size_t count, int wait)
2245 int ret = 0, meta_level = 0, overwrite_io = 0;
2247 struct dentry *dentry = file->f_path.dentry;
2248 struct inode *inode = d_inode(dentry);
2249 struct buffer_head *di_bh = NULL;
2255 * We start with a read level meta lock and only jump to an ex
2256 * if we need to make modifications here.
2259 ret = ocfs2_inode_lock_for_extent_tree(inode,
2271 * Check if IO will overwrite allocated blocks in case
2272 * IOCB_NOWAIT flag is set.
2274 if (!wait && !overwrite_io) {
2277 ret = ocfs2_overwrite_io(inode, di_bh, pos, count);
2285 /* Clear suid / sgid if necessary. We do this here
2286 * instead of later in the write path because
2287 * remove_suid() calls ->setattr without any hint that
2288 * we may have already done our cluster locking. Since
2289 * ocfs2_setattr() *must* take cluster locks to
2290 * proceed, this will lead us to recursively lock the
2291 * inode. There's also the dinode i_size state which
2292 * can be lost via setattr during extending writes (we
2293 * set inode->i_size at the end of a write. */
2294 if (should_remove_suid(dentry)) {
2295 if (meta_level == 0) {
2296 ocfs2_inode_unlock_for_extent_tree(inode,
2304 ret = ocfs2_write_remove_suid(inode);
2313 ret = ocfs2_check_range_for_refcount(inode, pos, count);
2315 ocfs2_inode_unlock_for_extent_tree(inode,
2321 ret = ocfs2_inode_lock_for_extent_tree(inode,
2332 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2334 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2335 ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
2348 trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
2351 ocfs2_inode_unlock_for_extent_tree(inode,
2360 static ssize_t ocfs2_file_write_iter(struct kiocb *iocb,
2361 struct iov_iter *from)
2364 ssize_t written = 0;
2366 size_t count = iov_iter_count(from);
2367 struct file *file = iocb->ki_filp;
2368 struct inode *inode = file_inode(file);
2369 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2370 int full_coherency = !(osb->s_mount_opt &
2371 OCFS2_MOUNT_COHERENCY_BUFFERED);
2372 void *saved_ki_complete = NULL;
2373 int append_write = ((iocb->ki_pos + count) >=
2374 i_size_read(inode) ? 1 : 0);
2375 int direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2376 int nowait = iocb->ki_flags & IOCB_NOWAIT ? 1 : 0;
2378 trace_ocfs2_file_write_iter(inode, file, file->f_path.dentry,
2379 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2380 file->f_path.dentry->d_name.len,
2381 file->f_path.dentry->d_name.name,
2382 (unsigned int)from->nr_segs); /* GRRRRR */
2384 if (!direct_io && nowait)
2391 if (!inode_trylock(inode))
2397 * Concurrent O_DIRECT writes are allowed with
2398 * mount_option "coherency=buffered".
2399 * For append write, we must take rw EX.
2401 rw_level = (!direct_io || full_coherency || append_write);
2404 ret = ocfs2_try_rw_lock(inode, rw_level);
2406 ret = ocfs2_rw_lock(inode, rw_level);
2414 * O_DIRECT writes with "coherency=full" need to take EX cluster
2415 * inode_lock to guarantee coherency.
2417 if (direct_io && full_coherency) {
2419 * We need to take and drop the inode lock to force
2420 * other nodes to drop their caches. Buffered I/O
2421 * already does this in write_begin().
2424 ret = ocfs2_try_inode_lock(inode, NULL, 1);
2426 ret = ocfs2_inode_lock(inode, NULL, 1);
2433 ocfs2_inode_unlock(inode, 1);
2436 ret = generic_write_checks(iocb, from);
2444 ret = ocfs2_prepare_inode_for_write(file, iocb->ki_pos, count, !nowait);
2451 if (direct_io && !is_sync_kiocb(iocb) &&
2452 ocfs2_is_io_unaligned(inode, count, iocb->ki_pos)) {
2454 * Make it a sync io if it's an unaligned aio.
2456 saved_ki_complete = xchg(&iocb->ki_complete, NULL);
2459 /* communicate with ocfs2_dio_end_io */
2460 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2462 written = __generic_file_write_iter(iocb, from);
2463 /* buffered aio wouldn't have proper lock coverage today */
2464 BUG_ON(written == -EIOCBQUEUED && !direct_io);
2467 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2468 * function pointer which is called when o_direct io completes so that
2469 * it can unlock our rw lock.
2470 * Unfortunately there are error cases which call end_io and others
2471 * that don't. so we don't have to unlock the rw_lock if either an
2472 * async dio is going to do it in the future or an end_io after an
2473 * error has already done it.
2475 if ((written == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2479 if (unlikely(written <= 0))
2482 if (((file->f_flags & O_DSYNC) && !direct_io) ||
2484 ret = filemap_fdatawrite_range(file->f_mapping,
2485 iocb->ki_pos - written,
2491 ret = jbd2_journal_force_commit(osb->journal->j_journal);
2497 ret = filemap_fdatawait_range(file->f_mapping,
2498 iocb->ki_pos - written,
2503 if (saved_ki_complete)
2504 xchg(&iocb->ki_complete, saved_ki_complete);
2507 ocfs2_rw_unlock(inode, rw_level);
2510 inode_unlock(inode);
2517 static ssize_t ocfs2_file_read_iter(struct kiocb *iocb,
2518 struct iov_iter *to)
2520 int ret = 0, rw_level = -1, lock_level = 0;
2521 struct file *filp = iocb->ki_filp;
2522 struct inode *inode = file_inode(filp);
2523 int direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2524 int nowait = iocb->ki_flags & IOCB_NOWAIT ? 1 : 0;
2526 trace_ocfs2_file_read_iter(inode, filp, filp->f_path.dentry,
2527 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2528 filp->f_path.dentry->d_name.len,
2529 filp->f_path.dentry->d_name.name,
2530 to->nr_segs); /* GRRRRR */
2539 if (!direct_io && nowait)
2543 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
2544 * need locks to protect pending reads from racing with truncate.
2548 ret = ocfs2_try_rw_lock(inode, 0);
2550 ret = ocfs2_rw_lock(inode, 0);
2558 /* communicate with ocfs2_dio_end_io */
2559 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2563 * We're fine letting folks race truncates and extending
2564 * writes with read across the cluster, just like they can
2565 * locally. Hence no rw_lock during read.
2567 * Take and drop the meta data lock to update inode fields
2568 * like i_size. This allows the checks down below
2569 * generic_file_read_iter() a chance of actually working.
2571 ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level,
2578 ocfs2_inode_unlock(inode, lock_level);
2580 ret = generic_file_read_iter(iocb, to);
2581 trace_generic_file_read_iter_ret(ret);
2583 /* buffered aio wouldn't have proper lock coverage today */
2584 BUG_ON(ret == -EIOCBQUEUED && !direct_io);
2586 /* see ocfs2_file_write_iter */
2587 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2593 ocfs2_rw_unlock(inode, rw_level);
2598 /* Refer generic_file_llseek_unlocked() */
2599 static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence)
2601 struct inode *inode = file->f_mapping->host;
2610 /* SEEK_END requires the OCFS2 inode lock for the file
2611 * because it references the file's size.
2613 ret = ocfs2_inode_lock(inode, NULL, 0);
2618 offset += i_size_read(inode);
2619 ocfs2_inode_unlock(inode, 0);
2623 offset = file->f_pos;
2626 offset += file->f_pos;
2630 ret = ocfs2_seek_data_hole_offset(file, &offset, whence);
2639 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2642 inode_unlock(inode);
2648 static int ocfs2_file_clone_range(struct file *file_in,
2650 struct file *file_out,
2654 return ocfs2_reflink_remap_range(file_in, pos_in, file_out, pos_out,
2658 static int ocfs2_file_dedupe_range(struct file *file_in,
2660 struct file *file_out,
2664 return ocfs2_reflink_remap_range(file_in, pos_in, file_out, pos_out,
2668 const struct inode_operations ocfs2_file_iops = {
2669 .setattr = ocfs2_setattr,
2670 .getattr = ocfs2_getattr,
2671 .permission = ocfs2_permission,
2672 .listxattr = ocfs2_listxattr,
2673 .fiemap = ocfs2_fiemap,
2674 .get_acl = ocfs2_iop_get_acl,
2675 .set_acl = ocfs2_iop_set_acl,
2678 const struct inode_operations ocfs2_special_file_iops = {
2679 .setattr = ocfs2_setattr,
2680 .getattr = ocfs2_getattr,
2681 .permission = ocfs2_permission,
2682 .get_acl = ocfs2_iop_get_acl,
2683 .set_acl = ocfs2_iop_set_acl,
2687 * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2688 * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2690 const struct file_operations ocfs2_fops = {
2691 .llseek = ocfs2_file_llseek,
2693 .fsync = ocfs2_sync_file,
2694 .release = ocfs2_file_release,
2695 .open = ocfs2_file_open,
2696 .read_iter = ocfs2_file_read_iter,
2697 .write_iter = ocfs2_file_write_iter,
2698 .unlocked_ioctl = ocfs2_ioctl,
2699 #ifdef CONFIG_COMPAT
2700 .compat_ioctl = ocfs2_compat_ioctl,
2703 .flock = ocfs2_flock,
2704 .splice_read = generic_file_splice_read,
2705 .splice_write = iter_file_splice_write,
2706 .fallocate = ocfs2_fallocate,
2707 .clone_file_range = ocfs2_file_clone_range,
2708 .dedupe_file_range = ocfs2_file_dedupe_range,
2711 const struct file_operations ocfs2_dops = {
2712 .llseek = generic_file_llseek,
2713 .read = generic_read_dir,
2714 .iterate = ocfs2_readdir,
2715 .fsync = ocfs2_sync_file,
2716 .release = ocfs2_dir_release,
2717 .open = ocfs2_dir_open,
2718 .unlocked_ioctl = ocfs2_ioctl,
2719 #ifdef CONFIG_COMPAT
2720 .compat_ioctl = ocfs2_compat_ioctl,
2723 .flock = ocfs2_flock,
2727 * POSIX-lockless variants of our file_operations.
2729 * These will be used if the underlying cluster stack does not support
2730 * posix file locking, if the user passes the "localflocks" mount
2731 * option, or if we have a local-only fs.
2733 * ocfs2_flock is in here because all stacks handle UNIX file locks,
2734 * so we still want it in the case of no stack support for
2735 * plocks. Internally, it will do the right thing when asked to ignore
2738 const struct file_operations ocfs2_fops_no_plocks = {
2739 .llseek = ocfs2_file_llseek,
2741 .fsync = ocfs2_sync_file,
2742 .release = ocfs2_file_release,
2743 .open = ocfs2_file_open,
2744 .read_iter = ocfs2_file_read_iter,
2745 .write_iter = ocfs2_file_write_iter,
2746 .unlocked_ioctl = ocfs2_ioctl,
2747 #ifdef CONFIG_COMPAT
2748 .compat_ioctl = ocfs2_compat_ioctl,
2750 .flock = ocfs2_flock,
2751 .splice_read = generic_file_splice_read,
2752 .splice_write = iter_file_splice_write,
2753 .fallocate = ocfs2_fallocate,
2754 .clone_file_range = ocfs2_file_clone_range,
2755 .dedupe_file_range = ocfs2_file_dedupe_range,
2758 const struct file_operations ocfs2_dops_no_plocks = {
2759 .llseek = generic_file_llseek,
2760 .read = generic_read_dir,
2761 .iterate = ocfs2_readdir,
2762 .fsync = ocfs2_sync_file,
2763 .release = ocfs2_dir_release,
2764 .open = ocfs2_dir_open,
2765 .unlocked_ioctl = ocfs2_ioctl,
2766 #ifdef CONFIG_COMPAT
2767 .compat_ioctl = ocfs2_compat_ioctl,
2769 .flock = ocfs2_flock,