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>
41 #include <cluster/masklog.h>
49 #include "extent_map.h"
62 #include "refcounttree.h"
63 #include "ocfs2_trace.h"
65 #include "buffer_head_io.h"
67 static int ocfs2_init_file_private(struct inode *inode, struct file *file)
69 struct ocfs2_file_private *fp;
71 fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
76 mutex_init(&fp->fp_mutex);
77 ocfs2_file_lock_res_init(&fp->fp_flock, fp);
78 file->private_data = fp;
83 static void ocfs2_free_file_private(struct inode *inode, struct file *file)
85 struct ocfs2_file_private *fp = file->private_data;
86 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
89 ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
90 ocfs2_lock_res_free(&fp->fp_flock);
92 file->private_data = NULL;
96 static int ocfs2_file_open(struct inode *inode, struct file *file)
99 int mode = file->f_flags;
100 struct ocfs2_inode_info *oi = OCFS2_I(inode);
102 trace_ocfs2_file_open(inode, file, file->f_path.dentry,
103 (unsigned long long)OCFS2_I(inode)->ip_blkno,
104 file->f_path.dentry->d_name.len,
105 file->f_path.dentry->d_name.name, mode);
107 if (file->f_mode & FMODE_WRITE)
108 dquot_initialize(inode);
110 spin_lock(&oi->ip_lock);
112 /* Check that the inode hasn't been wiped from disk by another
113 * node. If it hasn't then we're safe as long as we hold the
114 * spin lock until our increment of open count. */
115 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
116 spin_unlock(&oi->ip_lock);
123 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
126 spin_unlock(&oi->ip_lock);
128 status = ocfs2_init_file_private(inode, file);
131 * We want to set open count back if we're failing the
134 spin_lock(&oi->ip_lock);
136 spin_unlock(&oi->ip_lock);
143 static int ocfs2_file_release(struct inode *inode, struct file *file)
145 struct ocfs2_inode_info *oi = OCFS2_I(inode);
147 spin_lock(&oi->ip_lock);
148 if (!--oi->ip_open_count)
149 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
151 trace_ocfs2_file_release(inode, file, file->f_path.dentry,
153 file->f_path.dentry->d_name.len,
154 file->f_path.dentry->d_name.name,
156 spin_unlock(&oi->ip_lock);
158 ocfs2_free_file_private(inode, file);
163 static int ocfs2_dir_open(struct inode *inode, struct file *file)
165 return ocfs2_init_file_private(inode, file);
168 static int ocfs2_dir_release(struct inode *inode, struct file *file)
170 ocfs2_free_file_private(inode, file);
174 static int ocfs2_sync_file(struct file *file, loff_t start, loff_t end,
179 struct inode *inode = file->f_mapping->host;
180 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
182 trace_ocfs2_sync_file(inode, file, file->f_path.dentry,
183 OCFS2_I(inode)->ip_blkno,
184 file->f_path.dentry->d_name.len,
185 file->f_path.dentry->d_name.name,
186 (unsigned long long)datasync);
188 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
191 err = filemap_write_and_wait_range(inode->i_mapping, start, end);
196 * Probably don't need the i_mutex at all in here, just putting it here
197 * to be consistent with how fsync used to be called, someone more
198 * familiar with the fs could possibly remove it.
200 mutex_lock(&inode->i_mutex);
201 if (datasync && !(inode->i_state & I_DIRTY_DATASYNC)) {
203 * We still have to flush drive's caches to get data to the
206 if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
207 blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
211 journal = osb->journal->j_journal;
212 err = jbd2_journal_force_commit(journal);
217 mutex_unlock(&inode->i_mutex);
219 return (err < 0) ? -EIO : 0;
222 int ocfs2_should_update_atime(struct inode *inode,
223 struct vfsmount *vfsmnt)
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 & MS_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 ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
252 (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
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;
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_journal_dirty(handle, bh);
298 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
303 static int ocfs2_set_inode_size(handle_t *handle,
305 struct buffer_head *fe_bh,
310 i_size_write(inode, new_i_size);
311 inode->i_blocks = ocfs2_inode_sector_count(inode);
312 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
314 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
324 int ocfs2_simple_size_update(struct inode *inode,
325 struct buffer_head *di_bh,
329 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
330 handle_t *handle = NULL;
332 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
333 if (IS_ERR(handle)) {
334 ret = PTR_ERR(handle);
339 ret = ocfs2_set_inode_size(handle, inode, di_bh,
344 ocfs2_commit_trans(osb, handle);
349 static int ocfs2_cow_file_pos(struct inode *inode,
350 struct buffer_head *fe_bh,
354 u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
355 unsigned int num_clusters = 0;
356 unsigned int ext_flags = 0;
359 * If the new offset is aligned to the range of the cluster, there is
360 * no space for ocfs2_zero_range_for_truncate to fill, so no need to
363 if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
366 status = ocfs2_get_clusters(inode, cpos, &phys,
367 &num_clusters, &ext_flags);
373 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
376 return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1);
382 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
384 struct buffer_head *fe_bh,
389 struct ocfs2_dinode *di;
393 * We need to CoW the cluster contains the offset if it is reflinked
394 * since we will call ocfs2_zero_range_for_truncate later which will
395 * write "0" from offset to the end of the cluster.
397 status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
403 /* TODO: This needs to actually orphan the inode in this
406 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
407 if (IS_ERR(handle)) {
408 status = PTR_ERR(handle);
413 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
414 OCFS2_JOURNAL_ACCESS_WRITE);
421 * Do this before setting i_size.
423 cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
424 status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
431 i_size_write(inode, new_i_size);
432 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
434 di = (struct ocfs2_dinode *) fe_bh->b_data;
435 di->i_size = cpu_to_le64(new_i_size);
436 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
437 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
439 ocfs2_journal_dirty(handle, fe_bh);
442 ocfs2_commit_trans(osb, handle);
447 static int ocfs2_truncate_file(struct inode *inode,
448 struct buffer_head *di_bh,
452 struct ocfs2_dinode *fe = NULL;
453 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
455 /* We trust di_bh because it comes from ocfs2_inode_lock(), which
456 * already validated it */
457 fe = (struct ocfs2_dinode *) di_bh->b_data;
459 trace_ocfs2_truncate_file((unsigned long long)OCFS2_I(inode)->ip_blkno,
460 (unsigned long long)le64_to_cpu(fe->i_size),
461 (unsigned long long)new_i_size);
463 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
464 "Inode %llu, inode i_size = %lld != di "
465 "i_size = %llu, i_flags = 0x%x\n",
466 (unsigned long long)OCFS2_I(inode)->ip_blkno,
468 (unsigned long long)le64_to_cpu(fe->i_size),
469 le32_to_cpu(fe->i_flags));
471 if (new_i_size > le64_to_cpu(fe->i_size)) {
472 trace_ocfs2_truncate_file_error(
473 (unsigned long long)le64_to_cpu(fe->i_size),
474 (unsigned long long)new_i_size);
480 /* lets handle the simple truncate cases before doing any more
481 * cluster locking. */
482 if (new_i_size == le64_to_cpu(fe->i_size))
485 down_write(&OCFS2_I(inode)->ip_alloc_sem);
487 ocfs2_resv_discard(&osb->osb_la_resmap,
488 &OCFS2_I(inode)->ip_la_data_resv);
491 * The inode lock forced other nodes to sync and drop their
492 * pages, which (correctly) happens even if we have a truncate
493 * without allocation change - ocfs2 cluster sizes can be much
494 * greater than page size, so we have to truncate them
497 unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
498 truncate_inode_pages(inode->i_mapping, new_i_size);
500 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
501 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
502 i_size_read(inode), 1);
506 goto bail_unlock_sem;
509 /* alright, we're going to need to do a full blown alloc size
510 * change. Orphan the inode so that recovery can complete the
511 * truncate if necessary. This does the task of marking
513 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
516 goto bail_unlock_sem;
519 status = ocfs2_commit_truncate(osb, inode, di_bh);
522 goto bail_unlock_sem;
525 /* TODO: orphan dir cleanup here. */
527 up_write(&OCFS2_I(inode)->ip_alloc_sem);
530 if (!status && OCFS2_I(inode)->ip_clusters == 0)
531 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
537 * extend file allocation only here.
538 * we'll update all the disk stuff, and oip->alloc_size
540 * expect stuff to be locked, a transaction started and enough data /
541 * metadata reservations in the contexts.
543 * Will return -EAGAIN, and a reason if a restart is needed.
544 * If passed in, *reason will always be set, even in error.
546 int ocfs2_add_inode_data(struct ocfs2_super *osb,
551 struct buffer_head *fe_bh,
553 struct ocfs2_alloc_context *data_ac,
554 struct ocfs2_alloc_context *meta_ac,
555 enum ocfs2_alloc_restarted *reason_ret)
558 struct ocfs2_extent_tree et;
560 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
561 ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
562 clusters_to_add, mark_unwritten,
563 data_ac, meta_ac, reason_ret);
568 static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
569 u32 clusters_to_add, int mark_unwritten)
572 int restart_func = 0;
575 struct buffer_head *bh = NULL;
576 struct ocfs2_dinode *fe = NULL;
577 handle_t *handle = NULL;
578 struct ocfs2_alloc_context *data_ac = NULL;
579 struct ocfs2_alloc_context *meta_ac = NULL;
580 enum ocfs2_alloc_restarted why;
581 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
582 struct ocfs2_extent_tree et;
586 * Unwritten extent only exists for file systems which
589 BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
591 status = ocfs2_read_inode_block(inode, &bh);
596 fe = (struct ocfs2_dinode *) bh->b_data;
599 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
601 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
602 status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
609 credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list);
610 handle = ocfs2_start_trans(osb, credits);
611 if (IS_ERR(handle)) {
612 status = PTR_ERR(handle);
618 restarted_transaction:
619 trace_ocfs2_extend_allocation(
620 (unsigned long long)OCFS2_I(inode)->ip_blkno,
621 (unsigned long long)i_size_read(inode),
622 le32_to_cpu(fe->i_clusters), clusters_to_add,
625 status = dquot_alloc_space_nodirty(inode,
626 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
631 /* reserve a write to the file entry early on - that we if we
632 * run out of credits in the allocation path, we can still
634 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
635 OCFS2_JOURNAL_ACCESS_WRITE);
641 prev_clusters = OCFS2_I(inode)->ip_clusters;
643 status = ocfs2_add_inode_data(osb,
653 if ((status < 0) && (status != -EAGAIN)) {
654 if (status != -ENOSPC)
659 ocfs2_journal_dirty(handle, bh);
661 spin_lock(&OCFS2_I(inode)->ip_lock);
662 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
663 spin_unlock(&OCFS2_I(inode)->ip_lock);
664 /* Release unused quota reservation */
665 dquot_free_space(inode,
666 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
669 if (why != RESTART_NONE && clusters_to_add) {
670 if (why == RESTART_META) {
674 BUG_ON(why != RESTART_TRANS);
676 status = ocfs2_allocate_extend_trans(handle, 1);
678 /* handle still has to be committed at
684 goto restarted_transaction;
688 trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno,
689 le32_to_cpu(fe->i_clusters),
690 (unsigned long long)le64_to_cpu(fe->i_size),
691 OCFS2_I(inode)->ip_clusters,
692 (unsigned long long)i_size_read(inode));
695 if (status < 0 && did_quota)
696 dquot_free_space(inode,
697 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
699 ocfs2_commit_trans(osb, handle);
703 ocfs2_free_alloc_context(data_ac);
707 ocfs2_free_alloc_context(meta_ac);
710 if ((!status) && restart_func) {
721 * While a write will already be ordering the data, a truncate will not.
722 * Thus, we need to explicitly order the zeroed pages.
724 static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode)
726 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
727 handle_t *handle = NULL;
730 if (!ocfs2_should_order_data(inode))
733 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
734 if (IS_ERR(handle)) {
740 ret = ocfs2_jbd2_file_inode(handle, inode);
747 ocfs2_commit_trans(osb, handle);
748 handle = ERR_PTR(ret);
753 /* Some parts of this taken from generic_cont_expand, which turned out
754 * to be too fragile to do exactly what we need without us having to
755 * worry about recursive locking in ->write_begin() and ->write_end(). */
756 static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
759 struct address_space *mapping = inode->i_mapping;
761 unsigned long index = abs_from >> PAGE_CACHE_SHIFT;
762 handle_t *handle = NULL;
764 unsigned zero_from, zero_to, block_start, block_end;
766 BUG_ON(abs_from >= abs_to);
767 BUG_ON(abs_to > (((u64)index + 1) << PAGE_CACHE_SHIFT));
768 BUG_ON(abs_from & (inode->i_blkbits - 1));
770 page = find_or_create_page(mapping, index, GFP_NOFS);
777 /* Get the offsets within the page that we want to zero */
778 zero_from = abs_from & (PAGE_CACHE_SIZE - 1);
779 zero_to = abs_to & (PAGE_CACHE_SIZE - 1);
781 zero_to = PAGE_CACHE_SIZE;
783 trace_ocfs2_write_zero_page(
784 (unsigned long long)OCFS2_I(inode)->ip_blkno,
785 (unsigned long long)abs_from,
786 (unsigned long long)abs_to,
787 index, zero_from, zero_to);
789 /* We know that zero_from is block aligned */
790 for (block_start = zero_from; block_start < zero_to;
791 block_start = block_end) {
792 block_end = block_start + (1 << inode->i_blkbits);
795 * block_start is block-aligned. Bump it by one to force
796 * __block_write_begin and block_commit_write to zero the
799 ret = __block_write_begin(page, block_start + 1, 0,
807 handle = ocfs2_zero_start_ordered_transaction(inode);
808 if (IS_ERR(handle)) {
809 ret = PTR_ERR(handle);
815 /* must not update i_size! */
816 ret = block_commit_write(page, block_start + 1,
825 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
829 page_cache_release(page);
835 * Find the next range to zero. We do this in terms of bytes because
836 * that's what ocfs2_zero_extend() wants, and it is dealing with the
837 * pagecache. We may return multiple extents.
839 * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
840 * needs to be zeroed. range_start and range_end return the next zeroing
841 * range. A subsequent call should pass the previous range_end as its
842 * zero_start. If range_end is 0, there's nothing to do.
844 * Unwritten extents are skipped over. Refcounted extents are CoWd.
846 static int ocfs2_zero_extend_get_range(struct inode *inode,
847 struct buffer_head *di_bh,
848 u64 zero_start, u64 zero_end,
849 u64 *range_start, u64 *range_end)
851 int rc = 0, needs_cow = 0;
852 u32 p_cpos, zero_clusters = 0;
854 zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
855 u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
856 unsigned int num_clusters = 0;
857 unsigned int ext_flags = 0;
859 while (zero_cpos < last_cpos) {
860 rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
861 &num_clusters, &ext_flags);
867 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
868 zero_clusters = num_clusters;
869 if (ext_flags & OCFS2_EXT_REFCOUNTED)
874 zero_cpos += num_clusters;
876 if (!zero_clusters) {
881 while ((zero_cpos + zero_clusters) < last_cpos) {
882 rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
883 &p_cpos, &num_clusters,
890 if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
892 if (ext_flags & OCFS2_EXT_REFCOUNTED)
894 zero_clusters += num_clusters;
896 if ((zero_cpos + zero_clusters) > last_cpos)
897 zero_clusters = last_cpos - zero_cpos;
900 rc = ocfs2_refcount_cow(inode, di_bh, zero_cpos,
901 zero_clusters, UINT_MAX);
908 *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
909 *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
910 zero_cpos + zero_clusters);
917 * Zero one range returned from ocfs2_zero_extend_get_range(). The caller
918 * has made sure that the entire range needs zeroing.
920 static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
925 u64 zero_pos = range_start;
927 trace_ocfs2_zero_extend_range(
928 (unsigned long long)OCFS2_I(inode)->ip_blkno,
929 (unsigned long long)range_start,
930 (unsigned long long)range_end);
931 BUG_ON(range_start >= range_end);
933 while (zero_pos < range_end) {
934 next_pos = (zero_pos & PAGE_CACHE_MASK) + PAGE_CACHE_SIZE;
935 if (next_pos > range_end)
936 next_pos = range_end;
937 rc = ocfs2_write_zero_page(inode, zero_pos, next_pos);
945 * Very large extends have the potential to lock up
946 * the cpu for extended periods of time.
954 int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
958 u64 zero_start, range_start = 0, range_end = 0;
959 struct super_block *sb = inode->i_sb;
961 zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
962 trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
963 (unsigned long long)zero_start,
964 (unsigned long long)i_size_read(inode));
965 while (zero_start < zero_to_size) {
966 ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
977 if (range_start < zero_start)
978 range_start = zero_start;
979 if (range_end > zero_to_size)
980 range_end = zero_to_size;
982 ret = ocfs2_zero_extend_range(inode, range_start,
988 zero_start = range_end;
994 int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
995 u64 new_i_size, u64 zero_to)
999 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1002 * Only quota files call this without a bh, and they can't be
1005 BUG_ON(!di_bh && (oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
1006 BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1008 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1009 if (clusters_to_add < oi->ip_clusters)
1010 clusters_to_add = 0;
1012 clusters_to_add -= oi->ip_clusters;
1014 if (clusters_to_add) {
1015 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
1016 clusters_to_add, 0);
1024 * Call this even if we don't add any clusters to the tree. We
1025 * still need to zero the area between the old i_size and the
1028 ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1036 static int ocfs2_extend_file(struct inode *inode,
1037 struct buffer_head *di_bh,
1041 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1045 /* setattr sometimes calls us like this. */
1046 if (new_i_size == 0)
1049 if (i_size_read(inode) == new_i_size)
1051 BUG_ON(new_i_size < i_size_read(inode));
1054 * The alloc sem blocks people in read/write from reading our
1055 * allocation until we're done changing it. We depend on
1056 * i_mutex to block other extend/truncate calls while we're
1057 * here. We even have to hold it for sparse files because there
1058 * might be some tail zeroing.
1060 down_write(&oi->ip_alloc_sem);
1062 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1064 * We can optimize small extends by keeping the inodes
1067 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1068 up_write(&oi->ip_alloc_sem);
1069 goto out_update_size;
1072 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1074 up_write(&oi->ip_alloc_sem);
1080 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1081 ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1083 ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1086 up_write(&oi->ip_alloc_sem);
1094 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1102 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1104 int status = 0, size_change;
1105 struct inode *inode = dentry->d_inode;
1106 struct super_block *sb = inode->i_sb;
1107 struct ocfs2_super *osb = OCFS2_SB(sb);
1108 struct buffer_head *bh = NULL;
1109 handle_t *handle = NULL;
1110 struct dquot *transfer_to[MAXQUOTAS] = { };
1113 trace_ocfs2_setattr(inode, dentry,
1114 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1115 dentry->d_name.len, dentry->d_name.name,
1116 attr->ia_valid, attr->ia_mode,
1117 from_kuid(&init_user_ns, attr->ia_uid),
1118 from_kgid(&init_user_ns, attr->ia_gid));
1120 /* ensuring we don't even attempt to truncate a symlink */
1121 if (S_ISLNK(inode->i_mode))
1122 attr->ia_valid &= ~ATTR_SIZE;
1124 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1125 | ATTR_GID | ATTR_UID | ATTR_MODE)
1126 if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
1129 status = inode_change_ok(inode, attr);
1133 if (is_quota_modification(inode, attr))
1134 dquot_initialize(inode);
1135 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1137 status = ocfs2_rw_lock(inode, 1);
1144 status = ocfs2_inode_lock(inode, &bh, 1);
1146 if (status != -ENOENT)
1148 goto bail_unlock_rw;
1151 if (size_change && attr->ia_size != i_size_read(inode)) {
1152 status = inode_newsize_ok(inode, attr->ia_size);
1156 inode_dio_wait(inode);
1158 if (i_size_read(inode) > attr->ia_size) {
1159 if (ocfs2_should_order_data(inode)) {
1160 status = ocfs2_begin_ordered_truncate(inode,
1165 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1167 status = ocfs2_extend_file(inode, bh, attr->ia_size);
1169 if (status != -ENOSPC)
1176 if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
1177 (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
1179 * Gather pointers to quota structures so that allocation /
1180 * freeing of quota structures happens here and not inside
1181 * dquot_transfer() where we have problems with lock ordering
1183 if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)
1184 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1185 OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1186 transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid));
1187 if (!transfer_to[USRQUOTA]) {
1192 if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid)
1193 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1194 OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1195 transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid));
1196 if (!transfer_to[GRPQUOTA]) {
1201 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1202 2 * ocfs2_quota_trans_credits(sb));
1203 if (IS_ERR(handle)) {
1204 status = PTR_ERR(handle);
1208 status = __dquot_transfer(inode, transfer_to);
1212 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1213 if (IS_ERR(handle)) {
1214 status = PTR_ERR(handle);
1220 setattr_copy(inode, attr);
1221 mark_inode_dirty(inode);
1223 status = ocfs2_mark_inode_dirty(handle, inode, bh);
1228 ocfs2_commit_trans(osb, handle);
1230 ocfs2_inode_unlock(inode, 1);
1233 ocfs2_rw_unlock(inode, 1);
1237 /* Release quota pointers in case we acquired them */
1238 for (qtype = 0; qtype < MAXQUOTAS; qtype++)
1239 dqput(transfer_to[qtype]);
1241 if (!status && attr->ia_valid & ATTR_MODE) {
1242 status = posix_acl_chmod(inode, inode->i_mode);
1250 int ocfs2_getattr(struct vfsmount *mnt,
1251 struct dentry *dentry,
1254 struct inode *inode = dentry->d_inode;
1255 struct super_block *sb = dentry->d_inode->i_sb;
1256 struct ocfs2_super *osb = sb->s_fs_info;
1259 err = ocfs2_inode_revalidate(dentry);
1266 generic_fillattr(inode, stat);
1268 /* We set the blksize from the cluster size for performance */
1269 stat->blksize = osb->s_clustersize;
1275 int ocfs2_permission(struct inode *inode, int mask)
1279 if (mask & MAY_NOT_BLOCK)
1282 ret = ocfs2_inode_lock(inode, NULL, 0);
1289 ret = generic_permission(inode, mask);
1291 ocfs2_inode_unlock(inode, 0);
1296 static int __ocfs2_write_remove_suid(struct inode *inode,
1297 struct buffer_head *bh)
1301 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1302 struct ocfs2_dinode *di;
1304 trace_ocfs2_write_remove_suid(
1305 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1308 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1309 if (IS_ERR(handle)) {
1310 ret = PTR_ERR(handle);
1315 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1316 OCFS2_JOURNAL_ACCESS_WRITE);
1322 inode->i_mode &= ~S_ISUID;
1323 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1324 inode->i_mode &= ~S_ISGID;
1326 di = (struct ocfs2_dinode *) bh->b_data;
1327 di->i_mode = cpu_to_le16(inode->i_mode);
1329 ocfs2_journal_dirty(handle, bh);
1332 ocfs2_commit_trans(osb, handle);
1338 * Will look for holes and unwritten extents in the range starting at
1339 * pos for count bytes (inclusive).
1341 static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1345 unsigned int extent_flags;
1346 u32 cpos, clusters, extent_len, phys_cpos;
1347 struct super_block *sb = inode->i_sb;
1349 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1350 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1353 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1360 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
1365 if (extent_len > clusters)
1366 extent_len = clusters;
1368 clusters -= extent_len;
1375 static int ocfs2_write_remove_suid(struct inode *inode)
1378 struct buffer_head *bh = NULL;
1380 ret = ocfs2_read_inode_block(inode, &bh);
1386 ret = __ocfs2_write_remove_suid(inode, bh);
1393 * Allocate enough extents to cover the region starting at byte offset
1394 * start for len bytes. Existing extents are skipped, any extents
1395 * added are marked as "unwritten".
1397 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1401 u32 cpos, phys_cpos, clusters, alloc_size;
1402 u64 end = start + len;
1403 struct buffer_head *di_bh = NULL;
1405 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1406 ret = ocfs2_read_inode_block(inode, &di_bh);
1413 * Nothing to do if the requested reservation range
1414 * fits within the inode.
1416 if (ocfs2_size_fits_inline_data(di_bh, end))
1419 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1427 * We consider both start and len to be inclusive.
1429 cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1430 clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1434 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1442 * Hole or existing extent len can be arbitrary, so
1443 * cap it to our own allocation request.
1445 if (alloc_size > clusters)
1446 alloc_size = clusters;
1450 * We already have an allocation at this
1451 * region so we can safely skip it.
1456 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1465 clusters -= alloc_size;
1476 * Truncate a byte range, avoiding pages within partial clusters. This
1477 * preserves those pages for the zeroing code to write to.
1479 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1482 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1484 struct address_space *mapping = inode->i_mapping;
1486 start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1487 end = byte_start + byte_len;
1488 end = end & ~(osb->s_clustersize - 1);
1491 unmap_mapping_range(mapping, start, end - start, 0);
1492 truncate_inode_pages_range(mapping, start, end - 1);
1496 static int ocfs2_zero_partial_clusters(struct inode *inode,
1500 u64 tmpend, end = start + len;
1501 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1502 unsigned int csize = osb->s_clustersize;
1506 * The "start" and "end" values are NOT necessarily part of
1507 * the range whose allocation is being deleted. Rather, this
1508 * is what the user passed in with the request. We must zero
1509 * partial clusters here. There's no need to worry about
1510 * physical allocation - the zeroing code knows to skip holes.
1512 trace_ocfs2_zero_partial_clusters(
1513 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1514 (unsigned long long)start, (unsigned long long)end);
1517 * If both edges are on a cluster boundary then there's no
1518 * zeroing required as the region is part of the allocation to
1521 if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1524 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1525 if (IS_ERR(handle)) {
1526 ret = PTR_ERR(handle);
1532 * We want to get the byte offset of the end of the 1st cluster.
1534 tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1538 trace_ocfs2_zero_partial_clusters_range1((unsigned long long)start,
1539 (unsigned long long)tmpend);
1541 ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1547 * This may make start and end equal, but the zeroing
1548 * code will skip any work in that case so there's no
1549 * need to catch it up here.
1551 start = end & ~(osb->s_clustersize - 1);
1553 trace_ocfs2_zero_partial_clusters_range2(
1554 (unsigned long long)start, (unsigned long long)end);
1556 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1561 ocfs2_commit_trans(osb, handle);
1566 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1569 struct ocfs2_extent_rec *rec = NULL;
1571 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1573 rec = &el->l_recs[i];
1575 if (le32_to_cpu(rec->e_cpos) < pos)
1583 * Helper to calculate the punching pos and length in one run, we handle the
1584 * following three cases in order:
1586 * - remove the entire record
1587 * - remove a partial record
1588 * - no record needs to be removed (hole-punching completed)
1590 static void ocfs2_calc_trunc_pos(struct inode *inode,
1591 struct ocfs2_extent_list *el,
1592 struct ocfs2_extent_rec *rec,
1593 u32 trunc_start, u32 *trunc_cpos,
1594 u32 *trunc_len, u32 *trunc_end,
1595 u64 *blkno, int *done)
1600 range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1602 if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1604 * remove an entire extent record.
1606 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1608 * Skip holes if any.
1610 if (range < *trunc_end)
1612 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1613 *blkno = le64_to_cpu(rec->e_blkno);
1614 *trunc_end = le32_to_cpu(rec->e_cpos);
1615 } else if (range > trunc_start) {
1617 * remove a partial extent record, which means we're
1618 * removing the last extent record.
1620 *trunc_cpos = trunc_start;
1624 if (range < *trunc_end)
1626 *trunc_len = *trunc_end - trunc_start;
1627 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1628 *blkno = le64_to_cpu(rec->e_blkno) +
1629 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1630 *trunc_end = trunc_start;
1633 * It may have two following possibilities:
1635 * - last record has been removed
1636 * - trunc_start was within a hole
1638 * both two cases mean the completion of hole punching.
1646 static int ocfs2_remove_inode_range(struct inode *inode,
1647 struct buffer_head *di_bh, u64 byte_start,
1650 int ret = 0, flags = 0, done = 0, i;
1651 u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1653 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1654 struct ocfs2_cached_dealloc_ctxt dealloc;
1655 struct address_space *mapping = inode->i_mapping;
1656 struct ocfs2_extent_tree et;
1657 struct ocfs2_path *path = NULL;
1658 struct ocfs2_extent_list *el = NULL;
1659 struct ocfs2_extent_rec *rec = NULL;
1660 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1661 u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1663 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1664 ocfs2_init_dealloc_ctxt(&dealloc);
1666 trace_ocfs2_remove_inode_range(
1667 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1668 (unsigned long long)byte_start,
1669 (unsigned long long)byte_len);
1674 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1675 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1676 byte_start + byte_len, 0);
1682 * There's no need to get fancy with the page cache
1683 * truncate of an inline-data inode. We're talking
1684 * about less than a page here, which will be cached
1685 * in the dinode buffer anyway.
1687 unmap_mapping_range(mapping, 0, 0, 0);
1688 truncate_inode_pages(mapping, 0);
1693 * For reflinks, we may need to CoW 2 clusters which might be
1694 * partially zero'd later, if hole's start and end offset were
1695 * within one cluster(means is not exactly aligned to clustersize).
1698 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
1700 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1706 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1713 trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1714 trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1715 cluster_in_el = trunc_end;
1717 ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1723 path = ocfs2_new_path_from_et(&et);
1730 while (trunc_end > trunc_start) {
1732 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1739 el = path_leaf_el(path);
1741 i = ocfs2_find_rec(el, trunc_end);
1743 * Need to go to previous extent block.
1746 if (path->p_tree_depth == 0)
1749 ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1758 * We've reached the leftmost extent block,
1759 * it's safe to leave.
1761 if (cluster_in_el == 0)
1765 * The 'pos' searched for previous extent block is
1766 * always one cluster less than actual trunc_end.
1768 trunc_end = cluster_in_el + 1;
1770 ocfs2_reinit_path(path, 1);
1775 rec = &el->l_recs[i];
1777 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1778 &trunc_len, &trunc_end, &blkno, &done);
1782 flags = rec->e_flags;
1783 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1785 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1786 phys_cpos, trunc_len, flags,
1787 &dealloc, refcount_loc);
1793 cluster_in_el = trunc_end;
1795 ocfs2_reinit_path(path, 1);
1798 ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1801 ocfs2_free_path(path);
1802 ocfs2_schedule_truncate_log_flush(osb, 1);
1803 ocfs2_run_deallocs(osb, &dealloc);
1809 * Parts of this function taken from xfs_change_file_space()
1811 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1812 loff_t f_pos, unsigned int cmd,
1813 struct ocfs2_space_resv *sr,
1819 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1820 struct buffer_head *di_bh = NULL;
1822 unsigned long long max_off = inode->i_sb->s_maxbytes;
1824 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1827 mutex_lock(&inode->i_mutex);
1830 * This prevents concurrent writes on other nodes
1832 ret = ocfs2_rw_lock(inode, 1);
1838 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1844 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1846 goto out_inode_unlock;
1849 switch (sr->l_whence) {
1850 case 0: /*SEEK_SET*/
1852 case 1: /*SEEK_CUR*/
1853 sr->l_start += f_pos;
1855 case 2: /*SEEK_END*/
1856 sr->l_start += i_size_read(inode);
1860 goto out_inode_unlock;
1864 llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1867 || sr->l_start > max_off
1868 || (sr->l_start + llen) < 0
1869 || (sr->l_start + llen) > max_off) {
1871 goto out_inode_unlock;
1873 size = sr->l_start + sr->l_len;
1875 if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64 ||
1876 cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) {
1877 if (sr->l_len <= 0) {
1879 goto out_inode_unlock;
1883 if (file && should_remove_suid(file->f_path.dentry)) {
1884 ret = __ocfs2_write_remove_suid(inode, di_bh);
1887 goto out_inode_unlock;
1891 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1893 case OCFS2_IOC_RESVSP:
1894 case OCFS2_IOC_RESVSP64:
1896 * This takes unsigned offsets, but the signed ones we
1897 * pass have been checked against overflow above.
1899 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1902 case OCFS2_IOC_UNRESVSP:
1903 case OCFS2_IOC_UNRESVSP64:
1904 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1910 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1913 goto out_inode_unlock;
1917 * We update c/mtime for these changes
1919 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1920 if (IS_ERR(handle)) {
1921 ret = PTR_ERR(handle);
1923 goto out_inode_unlock;
1926 if (change_size && i_size_read(inode) < size)
1927 i_size_write(inode, size);
1929 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1930 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1934 if (file && (file->f_flags & O_SYNC))
1937 ocfs2_commit_trans(osb, handle);
1941 ocfs2_inode_unlock(inode, 1);
1943 ocfs2_rw_unlock(inode, 1);
1946 mutex_unlock(&inode->i_mutex);
1950 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1951 struct ocfs2_space_resv *sr)
1953 struct inode *inode = file_inode(file);
1954 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1957 if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1958 !ocfs2_writes_unwritten_extents(osb))
1960 else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1961 !ocfs2_sparse_alloc(osb))
1964 if (!S_ISREG(inode->i_mode))
1967 if (!(file->f_mode & FMODE_WRITE))
1970 ret = mnt_want_write_file(file);
1973 ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1974 mnt_drop_write_file(file);
1978 static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
1981 struct inode *inode = file_inode(file);
1982 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1983 struct ocfs2_space_resv sr;
1984 int change_size = 1;
1985 int cmd = OCFS2_IOC_RESVSP64;
1987 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
1989 if (!ocfs2_writes_unwritten_extents(osb))
1992 if (mode & FALLOC_FL_KEEP_SIZE)
1995 if (mode & FALLOC_FL_PUNCH_HOLE)
1996 cmd = OCFS2_IOC_UNRESVSP64;
1999 sr.l_start = (s64)offset;
2000 sr.l_len = (s64)len;
2002 return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2006 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2010 unsigned int extent_flags;
2011 u32 cpos, clusters, extent_len, phys_cpos;
2012 struct super_block *sb = inode->i_sb;
2014 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2015 !(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) ||
2016 OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2019 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2020 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2023 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2030 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2035 if (extent_len > clusters)
2036 extent_len = clusters;
2038 clusters -= extent_len;
2045 static void ocfs2_aiodio_wait(struct inode *inode)
2047 wait_queue_head_t *wq = ocfs2_ioend_wq(inode);
2049 wait_event(*wq, (atomic_read(&OCFS2_I(inode)->ip_unaligned_aio) == 0));
2052 static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
2054 int blockmask = inode->i_sb->s_blocksize - 1;
2055 loff_t final_size = pos + count;
2057 if ((pos & blockmask) || (final_size & blockmask))
2062 static int ocfs2_prepare_inode_for_refcount(struct inode *inode,
2064 loff_t pos, size_t count,
2068 struct buffer_head *di_bh = NULL;
2069 u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2071 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2073 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2081 ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
2089 static int ocfs2_prepare_inode_for_write(struct file *file,
2096 int ret = 0, meta_level = 0;
2097 struct dentry *dentry = file->f_path.dentry;
2098 struct inode *inode = dentry->d_inode;
2099 loff_t saved_pos = 0, end;
2102 * We start with a read level meta lock and only jump to an ex
2103 * if we need to make modifications here.
2106 ret = ocfs2_inode_lock(inode, NULL, meta_level);
2113 /* Clear suid / sgid if necessary. We do this here
2114 * instead of later in the write path because
2115 * remove_suid() calls ->setattr without any hint that
2116 * we may have already done our cluster locking. Since
2117 * ocfs2_setattr() *must* take cluster locks to
2118 * proceed, this will lead us to recursively lock the
2119 * inode. There's also the dinode i_size state which
2120 * can be lost via setattr during extending writes (we
2121 * set inode->i_size at the end of a write. */
2122 if (should_remove_suid(dentry)) {
2123 if (meta_level == 0) {
2124 ocfs2_inode_unlock(inode, meta_level);
2129 ret = ocfs2_write_remove_suid(inode);
2136 /* work on a copy of ppos until we're sure that we won't have
2137 * to recalculate it due to relocking. */
2139 saved_pos = i_size_read(inode);
2143 end = saved_pos + count;
2145 ret = ocfs2_check_range_for_refcount(inode, saved_pos, count);
2147 ocfs2_inode_unlock(inode, meta_level);
2150 ret = ocfs2_prepare_inode_for_refcount(inode,
2167 * Skip the O_DIRECT checks if we don't need
2170 if (!direct_io || !(*direct_io))
2174 * There's no sane way to do direct writes to an inode
2177 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2183 * Allowing concurrent direct writes means
2184 * i_size changes wouldn't be synchronized, so
2185 * one node could wind up truncating another
2188 if (end > i_size_read(inode)) {
2194 * We don't fill holes during direct io, so
2195 * check for them here. If any are found, the
2196 * caller will have to retake some cluster
2197 * locks and initiate the io as buffered.
2199 ret = ocfs2_check_range_for_holes(inode, saved_pos, count);
2212 trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
2213 saved_pos, appending, count,
2214 direct_io, has_refcount);
2216 if (meta_level >= 0)
2217 ocfs2_inode_unlock(inode, meta_level);
2223 static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
2224 const struct iovec *iov,
2225 unsigned long nr_segs,
2228 int ret, direct_io, appending, rw_level, have_alloc_sem = 0;
2229 int can_do_direct, has_refcount = 0;
2230 ssize_t written = 0;
2231 size_t ocount; /* original count */
2232 size_t count; /* after file limit checks */
2233 loff_t old_size, *ppos = &iocb->ki_pos;
2235 struct file *file = iocb->ki_filp;
2236 struct inode *inode = file_inode(file);
2237 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2238 int full_coherency = !(osb->s_mount_opt &
2239 OCFS2_MOUNT_COHERENCY_BUFFERED);
2240 int unaligned_dio = 0;
2242 trace_ocfs2_file_aio_write(inode, file, file->f_path.dentry,
2243 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2244 file->f_path.dentry->d_name.len,
2245 file->f_path.dentry->d_name.name,
2246 (unsigned int)nr_segs);
2248 if (iocb->ki_nbytes == 0)
2251 appending = file->f_flags & O_APPEND ? 1 : 0;
2252 direct_io = file->f_flags & O_DIRECT ? 1 : 0;
2254 mutex_lock(&inode->i_mutex);
2256 ocfs2_iocb_clear_sem_locked(iocb);
2259 /* to match setattr's i_mutex -> rw_lock ordering */
2262 /* communicate with ocfs2_dio_end_io */
2263 ocfs2_iocb_set_sem_locked(iocb);
2267 * Concurrent O_DIRECT writes are allowed with
2268 * mount_option "coherency=buffered".
2270 rw_level = (!direct_io || full_coherency);
2272 ret = ocfs2_rw_lock(inode, rw_level);
2279 * O_DIRECT writes with "coherency=full" need to take EX cluster
2280 * inode_lock to guarantee coherency.
2282 if (direct_io && full_coherency) {
2284 * We need to take and drop the inode lock to force
2285 * other nodes to drop their caches. Buffered I/O
2286 * already does this in write_begin().
2288 ret = ocfs2_inode_lock(inode, NULL, 1);
2294 ocfs2_inode_unlock(inode, 1);
2297 can_do_direct = direct_io;
2298 ret = ocfs2_prepare_inode_for_write(file, ppos,
2299 iocb->ki_nbytes, appending,
2300 &can_do_direct, &has_refcount);
2306 if (direct_io && !is_sync_kiocb(iocb))
2307 unaligned_dio = ocfs2_is_io_unaligned(inode, iocb->ki_nbytes,
2311 * We can't complete the direct I/O as requested, fall back to
2314 if (direct_io && !can_do_direct) {
2315 ocfs2_rw_unlock(inode, rw_level);
2324 if (unaligned_dio) {
2326 * Wait on previous unaligned aio to complete before
2329 ocfs2_aiodio_wait(inode);
2331 /* Mark the iocb as needing a decrement in ocfs2_dio_end_io */
2332 atomic_inc(&OCFS2_I(inode)->ip_unaligned_aio);
2333 ocfs2_iocb_set_unaligned_aio(iocb);
2337 * To later detect whether a journal commit for sync writes is
2338 * necessary, we sample i_size, and cluster count here.
2340 old_size = i_size_read(inode);
2341 old_clusters = OCFS2_I(inode)->ip_clusters;
2343 /* communicate with ocfs2_dio_end_io */
2344 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2346 ret = generic_segment_checks(iov, &nr_segs, &ocount,
2352 ret = generic_write_checks(file, ppos, &count,
2353 S_ISBLK(inode->i_mode));
2358 written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
2359 ppos, count, ocount);
2365 current->backing_dev_info = file->f_mapping->backing_dev_info;
2366 written = generic_file_buffered_write(iocb, iov, nr_segs, *ppos,
2368 current->backing_dev_info = NULL;
2372 /* buffered aio wouldn't have proper lock coverage today */
2373 BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
2375 if (((file->f_flags & O_DSYNC) && !direct_io) || IS_SYNC(inode) ||
2376 ((file->f_flags & O_DIRECT) && !direct_io)) {
2377 ret = filemap_fdatawrite_range(file->f_mapping, pos,
2382 if (!ret && ((old_size != i_size_read(inode)) ||
2383 (old_clusters != OCFS2_I(inode)->ip_clusters) ||
2385 ret = jbd2_journal_force_commit(osb->journal->j_journal);
2391 ret = filemap_fdatawait_range(file->f_mapping, pos,
2396 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2397 * function pointer which is called when o_direct io completes so that
2398 * it can unlock our rw lock.
2399 * Unfortunately there are error cases which call end_io and others
2400 * that don't. so we don't have to unlock the rw_lock if either an
2401 * async dio is going to do it in the future or an end_io after an
2402 * error has already done it.
2404 if ((ret == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2410 if (unaligned_dio) {
2411 ocfs2_iocb_clear_unaligned_aio(iocb);
2412 atomic_dec(&OCFS2_I(inode)->ip_unaligned_aio);
2417 ocfs2_rw_unlock(inode, rw_level);
2421 ocfs2_iocb_clear_sem_locked(iocb);
2423 mutex_unlock(&inode->i_mutex);
2430 static int ocfs2_splice_to_file(struct pipe_inode_info *pipe,
2432 struct splice_desc *sd)
2436 ret = ocfs2_prepare_inode_for_write(out, &sd->pos,
2437 sd->total_len, 0, NULL, NULL);
2443 return splice_from_pipe_feed(pipe, sd, pipe_to_file);
2446 static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
2453 struct address_space *mapping = out->f_mapping;
2454 struct inode *inode = mapping->host;
2455 struct splice_desc sd = {
2463 trace_ocfs2_file_splice_write(inode, out, out->f_path.dentry,
2464 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2465 out->f_path.dentry->d_name.len,
2466 out->f_path.dentry->d_name.name, len);
2470 splice_from_pipe_begin(&sd);
2472 ret = splice_from_pipe_next(pipe, &sd);
2476 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2477 ret = ocfs2_rw_lock(inode, 1);
2481 ret = ocfs2_splice_to_file(pipe, out, &sd);
2482 ocfs2_rw_unlock(inode, 1);
2484 mutex_unlock(&inode->i_mutex);
2486 splice_from_pipe_end(pipe, &sd);
2491 ret = sd.num_spliced;
2496 err = generic_write_sync(out, *ppos, ret);
2502 balance_dirty_pages_ratelimited(mapping);
2508 static ssize_t ocfs2_file_splice_read(struct file *in,
2510 struct pipe_inode_info *pipe,
2514 int ret = 0, lock_level = 0;
2515 struct inode *inode = file_inode(in);
2517 trace_ocfs2_file_splice_read(inode, in, in->f_path.dentry,
2518 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2519 in->f_path.dentry->d_name.len,
2520 in->f_path.dentry->d_name.name, len);
2523 * See the comment in ocfs2_file_aio_read()
2525 ret = ocfs2_inode_lock_atime(inode, in->f_path.mnt, &lock_level);
2530 ocfs2_inode_unlock(inode, lock_level);
2532 ret = generic_file_splice_read(in, ppos, pipe, len, flags);
2538 static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
2539 const struct iovec *iov,
2540 unsigned long nr_segs,
2543 int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
2544 struct file *filp = iocb->ki_filp;
2545 struct inode *inode = file_inode(filp);
2547 trace_ocfs2_file_aio_read(inode, filp, filp->f_path.dentry,
2548 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2549 filp->f_path.dentry->d_name.len,
2550 filp->f_path.dentry->d_name.name, nr_segs);
2559 ocfs2_iocb_clear_sem_locked(iocb);
2562 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
2563 * need locks to protect pending reads from racing with truncate.
2565 if (filp->f_flags & O_DIRECT) {
2567 ocfs2_iocb_set_sem_locked(iocb);
2569 ret = ocfs2_rw_lock(inode, 0);
2575 /* communicate with ocfs2_dio_end_io */
2576 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2580 * We're fine letting folks race truncates and extending
2581 * writes with read across the cluster, just like they can
2582 * locally. Hence no rw_lock during read.
2584 * Take and drop the meta data lock to update inode fields
2585 * like i_size. This allows the checks down below
2586 * generic_file_aio_read() a chance of actually working.
2588 ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level);
2593 ocfs2_inode_unlock(inode, lock_level);
2595 ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
2596 trace_generic_file_aio_read_ret(ret);
2598 /* buffered aio wouldn't have proper lock coverage today */
2599 BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
2601 /* see ocfs2_file_aio_write */
2602 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2609 ocfs2_iocb_clear_sem_locked(iocb);
2612 ocfs2_rw_unlock(inode, rw_level);
2617 /* Refer generic_file_llseek_unlocked() */
2618 static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence)
2620 struct inode *inode = file->f_mapping->host;
2623 mutex_lock(&inode->i_mutex);
2629 offset += inode->i_size;
2633 offset = file->f_pos;
2636 offset += file->f_pos;
2640 ret = ocfs2_seek_data_hole_offset(file, &offset, whence);
2649 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2652 mutex_unlock(&inode->i_mutex);
2658 const struct inode_operations ocfs2_file_iops = {
2659 .setattr = ocfs2_setattr,
2660 .getattr = ocfs2_getattr,
2661 .permission = ocfs2_permission,
2662 .setxattr = generic_setxattr,
2663 .getxattr = generic_getxattr,
2664 .listxattr = ocfs2_listxattr,
2665 .removexattr = generic_removexattr,
2666 .fiemap = ocfs2_fiemap,
2667 .get_acl = ocfs2_iop_get_acl,
2668 .set_acl = ocfs2_iop_set_acl,
2671 const struct inode_operations ocfs2_special_file_iops = {
2672 .setattr = ocfs2_setattr,
2673 .getattr = ocfs2_getattr,
2674 .permission = ocfs2_permission,
2675 .get_acl = ocfs2_iop_get_acl,
2676 .set_acl = ocfs2_iop_set_acl,
2680 * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2681 * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2683 const struct file_operations ocfs2_fops = {
2684 .llseek = ocfs2_file_llseek,
2685 .read = do_sync_read,
2686 .write = do_sync_write,
2688 .fsync = ocfs2_sync_file,
2689 .release = ocfs2_file_release,
2690 .open = ocfs2_file_open,
2691 .aio_read = ocfs2_file_aio_read,
2692 .aio_write = ocfs2_file_aio_write,
2693 .unlocked_ioctl = ocfs2_ioctl,
2694 #ifdef CONFIG_COMPAT
2695 .compat_ioctl = ocfs2_compat_ioctl,
2698 .flock = ocfs2_flock,
2699 .splice_read = ocfs2_file_splice_read,
2700 .splice_write = ocfs2_file_splice_write,
2701 .fallocate = ocfs2_fallocate,
2704 const struct file_operations ocfs2_dops = {
2705 .llseek = generic_file_llseek,
2706 .read = generic_read_dir,
2707 .iterate = ocfs2_readdir,
2708 .fsync = ocfs2_sync_file,
2709 .release = ocfs2_dir_release,
2710 .open = ocfs2_dir_open,
2711 .unlocked_ioctl = ocfs2_ioctl,
2712 #ifdef CONFIG_COMPAT
2713 .compat_ioctl = ocfs2_compat_ioctl,
2716 .flock = ocfs2_flock,
2720 * POSIX-lockless variants of our file_operations.
2722 * These will be used if the underlying cluster stack does not support
2723 * posix file locking, if the user passes the "localflocks" mount
2724 * option, or if we have a local-only fs.
2726 * ocfs2_flock is in here because all stacks handle UNIX file locks,
2727 * so we still want it in the case of no stack support for
2728 * plocks. Internally, it will do the right thing when asked to ignore
2731 const struct file_operations ocfs2_fops_no_plocks = {
2732 .llseek = ocfs2_file_llseek,
2733 .read = do_sync_read,
2734 .write = do_sync_write,
2736 .fsync = ocfs2_sync_file,
2737 .release = ocfs2_file_release,
2738 .open = ocfs2_file_open,
2739 .aio_read = ocfs2_file_aio_read,
2740 .aio_write = ocfs2_file_aio_write,
2741 .unlocked_ioctl = ocfs2_ioctl,
2742 #ifdef CONFIG_COMPAT
2743 .compat_ioctl = ocfs2_compat_ioctl,
2745 .flock = ocfs2_flock,
2746 .splice_read = ocfs2_file_splice_read,
2747 .splice_write = ocfs2_file_splice_write,
2748 .fallocate = ocfs2_fallocate,
2751 const struct file_operations ocfs2_dops_no_plocks = {
2752 .llseek = generic_file_llseek,
2753 .read = generic_read_dir,
2754 .iterate = ocfs2_readdir,
2755 .fsync = ocfs2_sync_file,
2756 .release = ocfs2_dir_release,
2757 .open = ocfs2_dir_open,
2758 .unlocked_ioctl = ocfs2_ioctl,
2759 #ifdef CONFIG_COMPAT
2760 .compat_ioctl = ocfs2_compat_ioctl,
2762 .flock = ocfs2_flock,