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 err = filemap_write_and_wait_range(inode->i_mapping, start, end);
193 * Probably don't need the i_mutex at all in here, just putting it here
194 * to be consistent with how fsync used to be called, someone more
195 * familiar with the fs could possibly remove it.
197 mutex_lock(&inode->i_mutex);
198 if (datasync && !(inode->i_state & I_DIRTY_DATASYNC)) {
200 * We still have to flush drive's caches to get data to the
203 if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
204 blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
208 journal = osb->journal->j_journal;
209 err = jbd2_journal_force_commit(journal);
214 mutex_unlock(&inode->i_mutex);
216 return (err < 0) ? -EIO : 0;
219 int ocfs2_should_update_atime(struct inode *inode,
220 struct vfsmount *vfsmnt)
223 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
225 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
228 if ((inode->i_flags & S_NOATIME) ||
229 ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
233 * We can be called with no vfsmnt structure - NFSD will
236 * Note that our action here is different than touch_atime() -
237 * if we can't tell whether this is a noatime mount, then we
238 * don't know whether to trust the value of s_atime_quantum.
243 if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
244 ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
247 if (vfsmnt->mnt_flags & MNT_RELATIME) {
248 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
249 (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
256 if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
262 int ocfs2_update_inode_atime(struct inode *inode,
263 struct buffer_head *bh)
266 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
268 struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
270 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
271 if (IS_ERR(handle)) {
272 ret = PTR_ERR(handle);
277 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
278 OCFS2_JOURNAL_ACCESS_WRITE);
285 * Don't use ocfs2_mark_inode_dirty() here as we don't always
286 * have i_mutex to guard against concurrent changes to other
289 inode->i_atime = CURRENT_TIME;
290 di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
291 di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
292 ocfs2_journal_dirty(handle, bh);
295 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
300 static int ocfs2_set_inode_size(handle_t *handle,
302 struct buffer_head *fe_bh,
307 i_size_write(inode, new_i_size);
308 inode->i_blocks = ocfs2_inode_sector_count(inode);
309 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
311 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
321 int ocfs2_simple_size_update(struct inode *inode,
322 struct buffer_head *di_bh,
326 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
327 handle_t *handle = NULL;
329 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
330 if (IS_ERR(handle)) {
331 ret = PTR_ERR(handle);
336 ret = ocfs2_set_inode_size(handle, inode, di_bh,
341 ocfs2_commit_trans(osb, handle);
346 static int ocfs2_cow_file_pos(struct inode *inode,
347 struct buffer_head *fe_bh,
351 u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
352 unsigned int num_clusters = 0;
353 unsigned int ext_flags = 0;
356 * If the new offset is aligned to the range of the cluster, there is
357 * no space for ocfs2_zero_range_for_truncate to fill, so no need to
360 if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
363 status = ocfs2_get_clusters(inode, cpos, &phys,
364 &num_clusters, &ext_flags);
370 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
373 return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1);
379 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
381 struct buffer_head *fe_bh,
386 struct ocfs2_dinode *di;
390 * We need to CoW the cluster contains the offset if it is reflinked
391 * since we will call ocfs2_zero_range_for_truncate later which will
392 * write "0" from offset to the end of the cluster.
394 status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
400 /* TODO: This needs to actually orphan the inode in this
403 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
404 if (IS_ERR(handle)) {
405 status = PTR_ERR(handle);
410 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
411 OCFS2_JOURNAL_ACCESS_WRITE);
418 * Do this before setting i_size.
420 cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
421 status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
428 i_size_write(inode, new_i_size);
429 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
431 di = (struct ocfs2_dinode *) fe_bh->b_data;
432 di->i_size = cpu_to_le64(new_i_size);
433 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
434 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
436 ocfs2_journal_dirty(handle, fe_bh);
439 ocfs2_commit_trans(osb, handle);
444 static int ocfs2_truncate_file(struct inode *inode,
445 struct buffer_head *di_bh,
449 struct ocfs2_dinode *fe = NULL;
450 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
452 /* We trust di_bh because it comes from ocfs2_inode_lock(), which
453 * already validated it */
454 fe = (struct ocfs2_dinode *) di_bh->b_data;
456 trace_ocfs2_truncate_file((unsigned long long)OCFS2_I(inode)->ip_blkno,
457 (unsigned long long)le64_to_cpu(fe->i_size),
458 (unsigned long long)new_i_size);
460 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
461 "Inode %llu, inode i_size = %lld != di "
462 "i_size = %llu, i_flags = 0x%x\n",
463 (unsigned long long)OCFS2_I(inode)->ip_blkno,
465 (unsigned long long)le64_to_cpu(fe->i_size),
466 le32_to_cpu(fe->i_flags));
468 if (new_i_size > le64_to_cpu(fe->i_size)) {
469 trace_ocfs2_truncate_file_error(
470 (unsigned long long)le64_to_cpu(fe->i_size),
471 (unsigned long long)new_i_size);
477 /* lets handle the simple truncate cases before doing any more
478 * cluster locking. */
479 if (new_i_size == le64_to_cpu(fe->i_size))
482 down_write(&OCFS2_I(inode)->ip_alloc_sem);
484 ocfs2_resv_discard(&osb->osb_la_resmap,
485 &OCFS2_I(inode)->ip_la_data_resv);
488 * The inode lock forced other nodes to sync and drop their
489 * pages, which (correctly) happens even if we have a truncate
490 * without allocation change - ocfs2 cluster sizes can be much
491 * greater than page size, so we have to truncate them
494 unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
495 truncate_inode_pages(inode->i_mapping, new_i_size);
497 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
498 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
499 i_size_read(inode), 1);
503 goto bail_unlock_sem;
506 /* alright, we're going to need to do a full blown alloc size
507 * change. Orphan the inode so that recovery can complete the
508 * truncate if necessary. This does the task of marking
510 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
513 goto bail_unlock_sem;
516 status = ocfs2_commit_truncate(osb, inode, di_bh);
519 goto bail_unlock_sem;
522 /* TODO: orphan dir cleanup here. */
524 up_write(&OCFS2_I(inode)->ip_alloc_sem);
527 if (!status && OCFS2_I(inode)->ip_clusters == 0)
528 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
534 * extend file allocation only here.
535 * we'll update all the disk stuff, and oip->alloc_size
537 * expect stuff to be locked, a transaction started and enough data /
538 * metadata reservations in the contexts.
540 * Will return -EAGAIN, and a reason if a restart is needed.
541 * If passed in, *reason will always be set, even in error.
543 int ocfs2_add_inode_data(struct ocfs2_super *osb,
548 struct buffer_head *fe_bh,
550 struct ocfs2_alloc_context *data_ac,
551 struct ocfs2_alloc_context *meta_ac,
552 enum ocfs2_alloc_restarted *reason_ret)
555 struct ocfs2_extent_tree et;
557 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
558 ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
559 clusters_to_add, mark_unwritten,
560 data_ac, meta_ac, reason_ret);
565 static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
566 u32 clusters_to_add, int mark_unwritten)
569 int restart_func = 0;
572 struct buffer_head *bh = NULL;
573 struct ocfs2_dinode *fe = NULL;
574 handle_t *handle = NULL;
575 struct ocfs2_alloc_context *data_ac = NULL;
576 struct ocfs2_alloc_context *meta_ac = NULL;
577 enum ocfs2_alloc_restarted why;
578 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
579 struct ocfs2_extent_tree et;
583 * Unwritten extent only exists for file systems which
586 BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
588 status = ocfs2_read_inode_block(inode, &bh);
593 fe = (struct ocfs2_dinode *) bh->b_data;
596 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
598 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
599 status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
606 credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list);
607 handle = ocfs2_start_trans(osb, credits);
608 if (IS_ERR(handle)) {
609 status = PTR_ERR(handle);
615 restarted_transaction:
616 trace_ocfs2_extend_allocation(
617 (unsigned long long)OCFS2_I(inode)->ip_blkno,
618 (unsigned long long)i_size_read(inode),
619 le32_to_cpu(fe->i_clusters), clusters_to_add,
622 status = dquot_alloc_space_nodirty(inode,
623 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
628 /* reserve a write to the file entry early on - that we if we
629 * run out of credits in the allocation path, we can still
631 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
632 OCFS2_JOURNAL_ACCESS_WRITE);
638 prev_clusters = OCFS2_I(inode)->ip_clusters;
640 status = ocfs2_add_inode_data(osb,
650 if ((status < 0) && (status != -EAGAIN)) {
651 if (status != -ENOSPC)
656 ocfs2_journal_dirty(handle, bh);
658 spin_lock(&OCFS2_I(inode)->ip_lock);
659 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
660 spin_unlock(&OCFS2_I(inode)->ip_lock);
661 /* Release unused quota reservation */
662 dquot_free_space(inode,
663 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
666 if (why != RESTART_NONE && clusters_to_add) {
667 if (why == RESTART_META) {
671 BUG_ON(why != RESTART_TRANS);
673 status = ocfs2_allocate_extend_trans(handle, 1);
675 /* handle still has to be committed at
681 goto restarted_transaction;
685 trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno,
686 le32_to_cpu(fe->i_clusters),
687 (unsigned long long)le64_to_cpu(fe->i_size),
688 OCFS2_I(inode)->ip_clusters,
689 (unsigned long long)i_size_read(inode));
692 if (status < 0 && did_quota)
693 dquot_free_space(inode,
694 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
696 ocfs2_commit_trans(osb, handle);
700 ocfs2_free_alloc_context(data_ac);
704 ocfs2_free_alloc_context(meta_ac);
707 if ((!status) && restart_func) {
718 * While a write will already be ordering the data, a truncate will not.
719 * Thus, we need to explicitly order the zeroed pages.
721 static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode)
723 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
724 handle_t *handle = NULL;
727 if (!ocfs2_should_order_data(inode))
730 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
731 if (IS_ERR(handle)) {
737 ret = ocfs2_jbd2_file_inode(handle, inode);
744 ocfs2_commit_trans(osb, handle);
745 handle = ERR_PTR(ret);
750 /* Some parts of this taken from generic_cont_expand, which turned out
751 * to be too fragile to do exactly what we need without us having to
752 * worry about recursive locking in ->write_begin() and ->write_end(). */
753 static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
756 struct address_space *mapping = inode->i_mapping;
758 unsigned long index = abs_from >> PAGE_CACHE_SHIFT;
759 handle_t *handle = NULL;
761 unsigned zero_from, zero_to, block_start, block_end;
763 BUG_ON(abs_from >= abs_to);
764 BUG_ON(abs_to > (((u64)index + 1) << PAGE_CACHE_SHIFT));
765 BUG_ON(abs_from & (inode->i_blkbits - 1));
767 page = find_or_create_page(mapping, index, GFP_NOFS);
774 /* Get the offsets within the page that we want to zero */
775 zero_from = abs_from & (PAGE_CACHE_SIZE - 1);
776 zero_to = abs_to & (PAGE_CACHE_SIZE - 1);
778 zero_to = PAGE_CACHE_SIZE;
780 trace_ocfs2_write_zero_page(
781 (unsigned long long)OCFS2_I(inode)->ip_blkno,
782 (unsigned long long)abs_from,
783 (unsigned long long)abs_to,
784 index, zero_from, zero_to);
786 /* We know that zero_from is block aligned */
787 for (block_start = zero_from; block_start < zero_to;
788 block_start = block_end) {
789 block_end = block_start + (1 << inode->i_blkbits);
792 * block_start is block-aligned. Bump it by one to force
793 * __block_write_begin and block_commit_write to zero the
796 ret = __block_write_begin(page, block_start + 1, 0,
804 handle = ocfs2_zero_start_ordered_transaction(inode);
805 if (IS_ERR(handle)) {
806 ret = PTR_ERR(handle);
812 /* must not update i_size! */
813 ret = block_commit_write(page, block_start + 1,
822 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
826 page_cache_release(page);
832 * Find the next range to zero. We do this in terms of bytes because
833 * that's what ocfs2_zero_extend() wants, and it is dealing with the
834 * pagecache. We may return multiple extents.
836 * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
837 * needs to be zeroed. range_start and range_end return the next zeroing
838 * range. A subsequent call should pass the previous range_end as its
839 * zero_start. If range_end is 0, there's nothing to do.
841 * Unwritten extents are skipped over. Refcounted extents are CoWd.
843 static int ocfs2_zero_extend_get_range(struct inode *inode,
844 struct buffer_head *di_bh,
845 u64 zero_start, u64 zero_end,
846 u64 *range_start, u64 *range_end)
848 int rc = 0, needs_cow = 0;
849 u32 p_cpos, zero_clusters = 0;
851 zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
852 u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
853 unsigned int num_clusters = 0;
854 unsigned int ext_flags = 0;
856 while (zero_cpos < last_cpos) {
857 rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
858 &num_clusters, &ext_flags);
864 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
865 zero_clusters = num_clusters;
866 if (ext_flags & OCFS2_EXT_REFCOUNTED)
871 zero_cpos += num_clusters;
873 if (!zero_clusters) {
878 while ((zero_cpos + zero_clusters) < last_cpos) {
879 rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
880 &p_cpos, &num_clusters,
887 if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
889 if (ext_flags & OCFS2_EXT_REFCOUNTED)
891 zero_clusters += num_clusters;
893 if ((zero_cpos + zero_clusters) > last_cpos)
894 zero_clusters = last_cpos - zero_cpos;
897 rc = ocfs2_refcount_cow(inode, di_bh, zero_cpos,
898 zero_clusters, UINT_MAX);
905 *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
906 *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
907 zero_cpos + zero_clusters);
914 * Zero one range returned from ocfs2_zero_extend_get_range(). The caller
915 * has made sure that the entire range needs zeroing.
917 static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
922 u64 zero_pos = range_start;
924 trace_ocfs2_zero_extend_range(
925 (unsigned long long)OCFS2_I(inode)->ip_blkno,
926 (unsigned long long)range_start,
927 (unsigned long long)range_end);
928 BUG_ON(range_start >= range_end);
930 while (zero_pos < range_end) {
931 next_pos = (zero_pos & PAGE_CACHE_MASK) + PAGE_CACHE_SIZE;
932 if (next_pos > range_end)
933 next_pos = range_end;
934 rc = ocfs2_write_zero_page(inode, zero_pos, next_pos);
942 * Very large extends have the potential to lock up
943 * the cpu for extended periods of time.
951 int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
955 u64 zero_start, range_start = 0, range_end = 0;
956 struct super_block *sb = inode->i_sb;
958 zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
959 trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
960 (unsigned long long)zero_start,
961 (unsigned long long)i_size_read(inode));
962 while (zero_start < zero_to_size) {
963 ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
974 if (range_start < zero_start)
975 range_start = zero_start;
976 if (range_end > zero_to_size)
977 range_end = zero_to_size;
979 ret = ocfs2_zero_extend_range(inode, range_start,
985 zero_start = range_end;
991 int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
992 u64 new_i_size, u64 zero_to)
996 struct ocfs2_inode_info *oi = OCFS2_I(inode);
999 * Only quota files call this without a bh, and they can't be
1002 BUG_ON(!di_bh && (oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
1003 BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1005 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1006 if (clusters_to_add < oi->ip_clusters)
1007 clusters_to_add = 0;
1009 clusters_to_add -= oi->ip_clusters;
1011 if (clusters_to_add) {
1012 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
1013 clusters_to_add, 0);
1021 * Call this even if we don't add any clusters to the tree. We
1022 * still need to zero the area between the old i_size and the
1025 ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1033 static int ocfs2_extend_file(struct inode *inode,
1034 struct buffer_head *di_bh,
1038 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1042 /* setattr sometimes calls us like this. */
1043 if (new_i_size == 0)
1046 if (i_size_read(inode) == new_i_size)
1048 BUG_ON(new_i_size < i_size_read(inode));
1051 * The alloc sem blocks people in read/write from reading our
1052 * allocation until we're done changing it. We depend on
1053 * i_mutex to block other extend/truncate calls while we're
1054 * here. We even have to hold it for sparse files because there
1055 * might be some tail zeroing.
1057 down_write(&oi->ip_alloc_sem);
1059 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1061 * We can optimize small extends by keeping the inodes
1064 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1065 up_write(&oi->ip_alloc_sem);
1066 goto out_update_size;
1069 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1071 up_write(&oi->ip_alloc_sem);
1077 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1078 ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1080 ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1083 up_write(&oi->ip_alloc_sem);
1091 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1099 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1101 int status = 0, size_change;
1102 struct inode *inode = dentry->d_inode;
1103 struct super_block *sb = inode->i_sb;
1104 struct ocfs2_super *osb = OCFS2_SB(sb);
1105 struct buffer_head *bh = NULL;
1106 handle_t *handle = NULL;
1107 struct dquot *transfer_to[MAXQUOTAS] = { };
1110 trace_ocfs2_setattr(inode, dentry,
1111 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1112 dentry->d_name.len, dentry->d_name.name,
1113 attr->ia_valid, attr->ia_mode,
1114 from_kuid(&init_user_ns, attr->ia_uid),
1115 from_kgid(&init_user_ns, attr->ia_gid));
1117 /* ensuring we don't even attempt to truncate a symlink */
1118 if (S_ISLNK(inode->i_mode))
1119 attr->ia_valid &= ~ATTR_SIZE;
1121 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1122 | ATTR_GID | ATTR_UID | ATTR_MODE)
1123 if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
1126 status = inode_change_ok(inode, attr);
1130 if (is_quota_modification(inode, attr))
1131 dquot_initialize(inode);
1132 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1134 status = ocfs2_rw_lock(inode, 1);
1141 status = ocfs2_inode_lock(inode, &bh, 1);
1143 if (status != -ENOENT)
1145 goto bail_unlock_rw;
1148 if (size_change && attr->ia_size != i_size_read(inode)) {
1149 status = inode_newsize_ok(inode, attr->ia_size);
1153 inode_dio_wait(inode);
1155 if (i_size_read(inode) > attr->ia_size) {
1156 if (ocfs2_should_order_data(inode)) {
1157 status = ocfs2_begin_ordered_truncate(inode,
1162 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1164 status = ocfs2_extend_file(inode, bh, attr->ia_size);
1166 if (status != -ENOSPC)
1173 if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
1174 (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
1176 * Gather pointers to quota structures so that allocation /
1177 * freeing of quota structures happens here and not inside
1178 * dquot_transfer() where we have problems with lock ordering
1180 if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)
1181 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1182 OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1183 transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid));
1184 if (!transfer_to[USRQUOTA]) {
1189 if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid)
1190 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1191 OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1192 transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid));
1193 if (!transfer_to[GRPQUOTA]) {
1198 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1199 2 * ocfs2_quota_trans_credits(sb));
1200 if (IS_ERR(handle)) {
1201 status = PTR_ERR(handle);
1205 status = __dquot_transfer(inode, transfer_to);
1209 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1210 if (IS_ERR(handle)) {
1211 status = PTR_ERR(handle);
1217 setattr_copy(inode, attr);
1218 mark_inode_dirty(inode);
1220 status = ocfs2_mark_inode_dirty(handle, inode, bh);
1225 ocfs2_commit_trans(osb, handle);
1227 ocfs2_inode_unlock(inode, 1);
1230 ocfs2_rw_unlock(inode, 1);
1234 /* Release quota pointers in case we acquired them */
1235 for (qtype = 0; qtype < MAXQUOTAS; qtype++)
1236 dqput(transfer_to[qtype]);
1238 if (!status && attr->ia_valid & ATTR_MODE) {
1239 status = ocfs2_acl_chmod(inode);
1247 int ocfs2_getattr(struct vfsmount *mnt,
1248 struct dentry *dentry,
1251 struct inode *inode = dentry->d_inode;
1252 struct super_block *sb = dentry->d_inode->i_sb;
1253 struct ocfs2_super *osb = sb->s_fs_info;
1256 err = ocfs2_inode_revalidate(dentry);
1263 generic_fillattr(inode, stat);
1265 /* We set the blksize from the cluster size for performance */
1266 stat->blksize = osb->s_clustersize;
1272 int ocfs2_permission(struct inode *inode, int mask)
1276 if (mask & MAY_NOT_BLOCK)
1279 ret = ocfs2_inode_lock(inode, NULL, 0);
1286 ret = generic_permission(inode, mask);
1288 ocfs2_inode_unlock(inode, 0);
1293 static int __ocfs2_write_remove_suid(struct inode *inode,
1294 struct buffer_head *bh)
1298 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1299 struct ocfs2_dinode *di;
1301 trace_ocfs2_write_remove_suid(
1302 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1305 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1306 if (IS_ERR(handle)) {
1307 ret = PTR_ERR(handle);
1312 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1313 OCFS2_JOURNAL_ACCESS_WRITE);
1319 inode->i_mode &= ~S_ISUID;
1320 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1321 inode->i_mode &= ~S_ISGID;
1323 di = (struct ocfs2_dinode *) bh->b_data;
1324 di->i_mode = cpu_to_le16(inode->i_mode);
1326 ocfs2_journal_dirty(handle, bh);
1329 ocfs2_commit_trans(osb, handle);
1335 * Will look for holes and unwritten extents in the range starting at
1336 * pos for count bytes (inclusive).
1338 static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1342 unsigned int extent_flags;
1343 u32 cpos, clusters, extent_len, phys_cpos;
1344 struct super_block *sb = inode->i_sb;
1346 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1347 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1350 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1357 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
1362 if (extent_len > clusters)
1363 extent_len = clusters;
1365 clusters -= extent_len;
1372 static int ocfs2_write_remove_suid(struct inode *inode)
1375 struct buffer_head *bh = NULL;
1377 ret = ocfs2_read_inode_block(inode, &bh);
1383 ret = __ocfs2_write_remove_suid(inode, bh);
1390 * Allocate enough extents to cover the region starting at byte offset
1391 * start for len bytes. Existing extents are skipped, any extents
1392 * added are marked as "unwritten".
1394 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1398 u32 cpos, phys_cpos, clusters, alloc_size;
1399 u64 end = start + len;
1400 struct buffer_head *di_bh = NULL;
1402 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1403 ret = ocfs2_read_inode_block(inode, &di_bh);
1410 * Nothing to do if the requested reservation range
1411 * fits within the inode.
1413 if (ocfs2_size_fits_inline_data(di_bh, end))
1416 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1424 * We consider both start and len to be inclusive.
1426 cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1427 clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1431 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1439 * Hole or existing extent len can be arbitrary, so
1440 * cap it to our own allocation request.
1442 if (alloc_size > clusters)
1443 alloc_size = clusters;
1447 * We already have an allocation at this
1448 * region so we can safely skip it.
1453 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1462 clusters -= alloc_size;
1473 * Truncate a byte range, avoiding pages within partial clusters. This
1474 * preserves those pages for the zeroing code to write to.
1476 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1479 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1481 struct address_space *mapping = inode->i_mapping;
1483 start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1484 end = byte_start + byte_len;
1485 end = end & ~(osb->s_clustersize - 1);
1488 unmap_mapping_range(mapping, start, end - start, 0);
1489 truncate_inode_pages_range(mapping, start, end - 1);
1493 static int ocfs2_zero_partial_clusters(struct inode *inode,
1497 u64 tmpend, end = start + len;
1498 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1499 unsigned int csize = osb->s_clustersize;
1503 * The "start" and "end" values are NOT necessarily part of
1504 * the range whose allocation is being deleted. Rather, this
1505 * is what the user passed in with the request. We must zero
1506 * partial clusters here. There's no need to worry about
1507 * physical allocation - the zeroing code knows to skip holes.
1509 trace_ocfs2_zero_partial_clusters(
1510 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1511 (unsigned long long)start, (unsigned long long)end);
1514 * If both edges are on a cluster boundary then there's no
1515 * zeroing required as the region is part of the allocation to
1518 if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1521 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1522 if (IS_ERR(handle)) {
1523 ret = PTR_ERR(handle);
1529 * We want to get the byte offset of the end of the 1st cluster.
1531 tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1535 trace_ocfs2_zero_partial_clusters_range1((unsigned long long)start,
1536 (unsigned long long)tmpend);
1538 ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1544 * This may make start and end equal, but the zeroing
1545 * code will skip any work in that case so there's no
1546 * need to catch it up here.
1548 start = end & ~(osb->s_clustersize - 1);
1550 trace_ocfs2_zero_partial_clusters_range2(
1551 (unsigned long long)start, (unsigned long long)end);
1553 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1558 ocfs2_commit_trans(osb, handle);
1563 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1566 struct ocfs2_extent_rec *rec = NULL;
1568 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1570 rec = &el->l_recs[i];
1572 if (le32_to_cpu(rec->e_cpos) < pos)
1580 * Helper to calculate the punching pos and length in one run, we handle the
1581 * following three cases in order:
1583 * - remove the entire record
1584 * - remove a partial record
1585 * - no record needs to be removed (hole-punching completed)
1587 static void ocfs2_calc_trunc_pos(struct inode *inode,
1588 struct ocfs2_extent_list *el,
1589 struct ocfs2_extent_rec *rec,
1590 u32 trunc_start, u32 *trunc_cpos,
1591 u32 *trunc_len, u32 *trunc_end,
1592 u64 *blkno, int *done)
1597 range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1599 if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1601 * remove an entire extent record.
1603 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1605 * Skip holes if any.
1607 if (range < *trunc_end)
1609 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1610 *blkno = le64_to_cpu(rec->e_blkno);
1611 *trunc_end = le32_to_cpu(rec->e_cpos);
1612 } else if (range > trunc_start) {
1614 * remove a partial extent record, which means we're
1615 * removing the last extent record.
1617 *trunc_cpos = trunc_start;
1621 if (range < *trunc_end)
1623 *trunc_len = *trunc_end - trunc_start;
1624 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1625 *blkno = le64_to_cpu(rec->e_blkno) +
1626 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1627 *trunc_end = trunc_start;
1630 * It may have two following possibilities:
1632 * - last record has been removed
1633 * - trunc_start was within a hole
1635 * both two cases mean the completion of hole punching.
1643 static int ocfs2_remove_inode_range(struct inode *inode,
1644 struct buffer_head *di_bh, u64 byte_start,
1647 int ret = 0, flags = 0, done = 0, i;
1648 u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1650 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1651 struct ocfs2_cached_dealloc_ctxt dealloc;
1652 struct address_space *mapping = inode->i_mapping;
1653 struct ocfs2_extent_tree et;
1654 struct ocfs2_path *path = NULL;
1655 struct ocfs2_extent_list *el = NULL;
1656 struct ocfs2_extent_rec *rec = NULL;
1657 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1658 u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1660 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1661 ocfs2_init_dealloc_ctxt(&dealloc);
1663 trace_ocfs2_remove_inode_range(
1664 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1665 (unsigned long long)byte_start,
1666 (unsigned long long)byte_len);
1671 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1672 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1673 byte_start + byte_len, 0);
1679 * There's no need to get fancy with the page cache
1680 * truncate of an inline-data inode. We're talking
1681 * about less than a page here, which will be cached
1682 * in the dinode buffer anyway.
1684 unmap_mapping_range(mapping, 0, 0, 0);
1685 truncate_inode_pages(mapping, 0);
1690 * For reflinks, we may need to CoW 2 clusters which might be
1691 * partially zero'd later, if hole's start and end offset were
1692 * within one cluster(means is not exactly aligned to clustersize).
1695 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
1697 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1703 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1710 trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1711 trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1712 cluster_in_el = trunc_end;
1714 ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1720 path = ocfs2_new_path_from_et(&et);
1727 while (trunc_end > trunc_start) {
1729 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1736 el = path_leaf_el(path);
1738 i = ocfs2_find_rec(el, trunc_end);
1740 * Need to go to previous extent block.
1743 if (path->p_tree_depth == 0)
1746 ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1755 * We've reached the leftmost extent block,
1756 * it's safe to leave.
1758 if (cluster_in_el == 0)
1762 * The 'pos' searched for previous extent block is
1763 * always one cluster less than actual trunc_end.
1765 trunc_end = cluster_in_el + 1;
1767 ocfs2_reinit_path(path, 1);
1772 rec = &el->l_recs[i];
1774 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1775 &trunc_len, &trunc_end, &blkno, &done);
1779 flags = rec->e_flags;
1780 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1782 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1783 phys_cpos, trunc_len, flags,
1784 &dealloc, refcount_loc);
1790 cluster_in_el = trunc_end;
1792 ocfs2_reinit_path(path, 1);
1795 ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1798 ocfs2_free_path(path);
1799 ocfs2_schedule_truncate_log_flush(osb, 1);
1800 ocfs2_run_deallocs(osb, &dealloc);
1806 * Parts of this function taken from xfs_change_file_space()
1808 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1809 loff_t f_pos, unsigned int cmd,
1810 struct ocfs2_space_resv *sr,
1816 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1817 struct buffer_head *di_bh = NULL;
1819 unsigned long long max_off = inode->i_sb->s_maxbytes;
1821 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1824 mutex_lock(&inode->i_mutex);
1827 * This prevents concurrent writes on other nodes
1829 ret = ocfs2_rw_lock(inode, 1);
1835 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1841 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1843 goto out_inode_unlock;
1846 switch (sr->l_whence) {
1847 case 0: /*SEEK_SET*/
1849 case 1: /*SEEK_CUR*/
1850 sr->l_start += f_pos;
1852 case 2: /*SEEK_END*/
1853 sr->l_start += i_size_read(inode);
1857 goto out_inode_unlock;
1861 llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1864 || sr->l_start > max_off
1865 || (sr->l_start + llen) < 0
1866 || (sr->l_start + llen) > max_off) {
1868 goto out_inode_unlock;
1870 size = sr->l_start + sr->l_len;
1872 if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) {
1873 if (sr->l_len <= 0) {
1875 goto out_inode_unlock;
1879 if (file && should_remove_suid(file->f_path.dentry)) {
1880 ret = __ocfs2_write_remove_suid(inode, di_bh);
1883 goto out_inode_unlock;
1887 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1889 case OCFS2_IOC_RESVSP:
1890 case OCFS2_IOC_RESVSP64:
1892 * This takes unsigned offsets, but the signed ones we
1893 * pass have been checked against overflow above.
1895 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1898 case OCFS2_IOC_UNRESVSP:
1899 case OCFS2_IOC_UNRESVSP64:
1900 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1906 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1909 goto out_inode_unlock;
1913 * We update c/mtime for these changes
1915 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1916 if (IS_ERR(handle)) {
1917 ret = PTR_ERR(handle);
1919 goto out_inode_unlock;
1922 if (change_size && i_size_read(inode) < size)
1923 i_size_write(inode, size);
1925 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1926 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1930 if (file && (file->f_flags & O_SYNC))
1933 ocfs2_commit_trans(osb, handle);
1937 ocfs2_inode_unlock(inode, 1);
1939 ocfs2_rw_unlock(inode, 1);
1942 mutex_unlock(&inode->i_mutex);
1946 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1947 struct ocfs2_space_resv *sr)
1949 struct inode *inode = file_inode(file);
1950 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1953 if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1954 !ocfs2_writes_unwritten_extents(osb))
1956 else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1957 !ocfs2_sparse_alloc(osb))
1960 if (!S_ISREG(inode->i_mode))
1963 if (!(file->f_mode & FMODE_WRITE))
1966 ret = mnt_want_write_file(file);
1969 ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1970 mnt_drop_write_file(file);
1974 static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
1977 struct inode *inode = file_inode(file);
1978 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1979 struct ocfs2_space_resv sr;
1980 int change_size = 1;
1981 int cmd = OCFS2_IOC_RESVSP64;
1983 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
1985 if (!ocfs2_writes_unwritten_extents(osb))
1988 if (mode & FALLOC_FL_KEEP_SIZE)
1991 if (mode & FALLOC_FL_PUNCH_HOLE)
1992 cmd = OCFS2_IOC_UNRESVSP64;
1995 sr.l_start = (s64)offset;
1996 sr.l_len = (s64)len;
1998 return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2002 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2006 unsigned int extent_flags;
2007 u32 cpos, clusters, extent_len, phys_cpos;
2008 struct super_block *sb = inode->i_sb;
2010 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2011 !(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) ||
2012 OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2015 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2016 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2019 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2026 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2031 if (extent_len > clusters)
2032 extent_len = clusters;
2034 clusters -= extent_len;
2041 static void ocfs2_aiodio_wait(struct inode *inode)
2043 wait_queue_head_t *wq = ocfs2_ioend_wq(inode);
2045 wait_event(*wq, (atomic_read(&OCFS2_I(inode)->ip_unaligned_aio) == 0));
2048 static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
2050 int blockmask = inode->i_sb->s_blocksize - 1;
2051 loff_t final_size = pos + count;
2053 if ((pos & blockmask) || (final_size & blockmask))
2058 static int ocfs2_prepare_inode_for_refcount(struct inode *inode,
2060 loff_t pos, size_t count,
2064 struct buffer_head *di_bh = NULL;
2065 u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2067 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2069 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2077 ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
2085 static int ocfs2_prepare_inode_for_write(struct file *file,
2092 int ret = 0, meta_level = 0;
2093 struct dentry *dentry = file->f_path.dentry;
2094 struct inode *inode = dentry->d_inode;
2095 loff_t saved_pos = 0, end;
2098 * We start with a read level meta lock and only jump to an ex
2099 * if we need to make modifications here.
2102 ret = ocfs2_inode_lock(inode, NULL, meta_level);
2109 /* Clear suid / sgid if necessary. We do this here
2110 * instead of later in the write path because
2111 * remove_suid() calls ->setattr without any hint that
2112 * we may have already done our cluster locking. Since
2113 * ocfs2_setattr() *must* take cluster locks to
2114 * proceed, this will lead us to recursively lock the
2115 * inode. There's also the dinode i_size state which
2116 * can be lost via setattr during extending writes (we
2117 * set inode->i_size at the end of a write. */
2118 if (should_remove_suid(dentry)) {
2119 if (meta_level == 0) {
2120 ocfs2_inode_unlock(inode, meta_level);
2125 ret = ocfs2_write_remove_suid(inode);
2132 /* work on a copy of ppos until we're sure that we won't have
2133 * to recalculate it due to relocking. */
2135 saved_pos = i_size_read(inode);
2139 end = saved_pos + count;
2141 ret = ocfs2_check_range_for_refcount(inode, saved_pos, count);
2143 ocfs2_inode_unlock(inode, meta_level);
2146 ret = ocfs2_prepare_inode_for_refcount(inode,
2163 * Skip the O_DIRECT checks if we don't need
2166 if (!direct_io || !(*direct_io))
2170 * There's no sane way to do direct writes to an inode
2173 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2179 * Allowing concurrent direct writes means
2180 * i_size changes wouldn't be synchronized, so
2181 * one node could wind up truncating another
2184 if (end > i_size_read(inode)) {
2190 * We don't fill holes during direct io, so
2191 * check for them here. If any are found, the
2192 * caller will have to retake some cluster
2193 * locks and initiate the io as buffered.
2195 ret = ocfs2_check_range_for_holes(inode, saved_pos, count);
2208 trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
2209 saved_pos, appending, count,
2210 direct_io, has_refcount);
2212 if (meta_level >= 0)
2213 ocfs2_inode_unlock(inode, meta_level);
2219 static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
2220 const struct iovec *iov,
2221 unsigned long nr_segs,
2224 int ret, direct_io, appending, rw_level, have_alloc_sem = 0;
2225 int can_do_direct, has_refcount = 0;
2226 ssize_t written = 0;
2227 size_t ocount; /* original count */
2228 size_t count; /* after file limit checks */
2229 loff_t old_size, *ppos = &iocb->ki_pos;
2231 struct file *file = iocb->ki_filp;
2232 struct inode *inode = file_inode(file);
2233 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2234 int full_coherency = !(osb->s_mount_opt &
2235 OCFS2_MOUNT_COHERENCY_BUFFERED);
2236 int unaligned_dio = 0;
2238 trace_ocfs2_file_aio_write(inode, file, file->f_path.dentry,
2239 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2240 file->f_path.dentry->d_name.len,
2241 file->f_path.dentry->d_name.name,
2242 (unsigned int)nr_segs);
2244 if (iocb->ki_nbytes == 0)
2247 appending = file->f_flags & O_APPEND ? 1 : 0;
2248 direct_io = file->f_flags & O_DIRECT ? 1 : 0;
2250 mutex_lock(&inode->i_mutex);
2252 ocfs2_iocb_clear_sem_locked(iocb);
2255 /* to match setattr's i_mutex -> rw_lock ordering */
2258 /* communicate with ocfs2_dio_end_io */
2259 ocfs2_iocb_set_sem_locked(iocb);
2263 * Concurrent O_DIRECT writes are allowed with
2264 * mount_option "coherency=buffered".
2266 rw_level = (!direct_io || full_coherency);
2268 ret = ocfs2_rw_lock(inode, rw_level);
2275 * O_DIRECT writes with "coherency=full" need to take EX cluster
2276 * inode_lock to guarantee coherency.
2278 if (direct_io && full_coherency) {
2280 * We need to take and drop the inode lock to force
2281 * other nodes to drop their caches. Buffered I/O
2282 * already does this in write_begin().
2284 ret = ocfs2_inode_lock(inode, NULL, 1);
2290 ocfs2_inode_unlock(inode, 1);
2293 can_do_direct = direct_io;
2294 ret = ocfs2_prepare_inode_for_write(file, ppos,
2295 iocb->ki_nbytes, appending,
2296 &can_do_direct, &has_refcount);
2302 if (direct_io && !is_sync_kiocb(iocb))
2303 unaligned_dio = ocfs2_is_io_unaligned(inode, iocb->ki_nbytes,
2307 * We can't complete the direct I/O as requested, fall back to
2310 if (direct_io && !can_do_direct) {
2311 ocfs2_rw_unlock(inode, rw_level);
2320 if (unaligned_dio) {
2322 * Wait on previous unaligned aio to complete before
2325 ocfs2_aiodio_wait(inode);
2327 /* Mark the iocb as needing a decrement in ocfs2_dio_end_io */
2328 atomic_inc(&OCFS2_I(inode)->ip_unaligned_aio);
2329 ocfs2_iocb_set_unaligned_aio(iocb);
2333 * To later detect whether a journal commit for sync writes is
2334 * necessary, we sample i_size, and cluster count here.
2336 old_size = i_size_read(inode);
2337 old_clusters = OCFS2_I(inode)->ip_clusters;
2339 /* communicate with ocfs2_dio_end_io */
2340 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2342 ret = generic_segment_checks(iov, &nr_segs, &ocount,
2348 ret = generic_write_checks(file, ppos, &count,
2349 S_ISBLK(inode->i_mode));
2354 written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
2355 ppos, count, ocount);
2361 current->backing_dev_info = file->f_mapping->backing_dev_info;
2362 written = generic_file_buffered_write(iocb, iov, nr_segs, *ppos,
2364 current->backing_dev_info = NULL;
2368 /* buffered aio wouldn't have proper lock coverage today */
2369 BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
2371 if (((file->f_flags & O_DSYNC) && !direct_io) || IS_SYNC(inode) ||
2372 ((file->f_flags & O_DIRECT) && !direct_io)) {
2373 ret = filemap_fdatawrite_range(file->f_mapping, pos,
2378 if (!ret && ((old_size != i_size_read(inode)) ||
2379 (old_clusters != OCFS2_I(inode)->ip_clusters) ||
2381 ret = jbd2_journal_force_commit(osb->journal->j_journal);
2387 ret = filemap_fdatawait_range(file->f_mapping, pos,
2392 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2393 * function pointer which is called when o_direct io completes so that
2394 * it can unlock our rw lock.
2395 * Unfortunately there are error cases which call end_io and others
2396 * that don't. so we don't have to unlock the rw_lock if either an
2397 * async dio is going to do it in the future or an end_io after an
2398 * error has already done it.
2400 if ((ret == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2406 if (unaligned_dio) {
2407 ocfs2_iocb_clear_unaligned_aio(iocb);
2408 atomic_dec(&OCFS2_I(inode)->ip_unaligned_aio);
2413 ocfs2_rw_unlock(inode, rw_level);
2417 ocfs2_iocb_clear_sem_locked(iocb);
2419 mutex_unlock(&inode->i_mutex);
2426 static int ocfs2_splice_to_file(struct pipe_inode_info *pipe,
2428 struct splice_desc *sd)
2432 ret = ocfs2_prepare_inode_for_write(out, &sd->pos,
2433 sd->total_len, 0, NULL, NULL);
2439 return splice_from_pipe_feed(pipe, sd, pipe_to_file);
2442 static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
2449 struct address_space *mapping = out->f_mapping;
2450 struct inode *inode = mapping->host;
2451 struct splice_desc sd = {
2459 trace_ocfs2_file_splice_write(inode, out, out->f_path.dentry,
2460 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2461 out->f_path.dentry->d_name.len,
2462 out->f_path.dentry->d_name.name, len);
2466 splice_from_pipe_begin(&sd);
2468 ret = splice_from_pipe_next(pipe, &sd);
2472 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2473 ret = ocfs2_rw_lock(inode, 1);
2477 ret = ocfs2_splice_to_file(pipe, out, &sd);
2478 ocfs2_rw_unlock(inode, 1);
2480 mutex_unlock(&inode->i_mutex);
2482 splice_from_pipe_end(pipe, &sd);
2487 ret = sd.num_spliced;
2492 err = generic_write_sync(out, *ppos, ret);
2498 balance_dirty_pages_ratelimited(mapping);
2504 static ssize_t ocfs2_file_splice_read(struct file *in,
2506 struct pipe_inode_info *pipe,
2510 int ret = 0, lock_level = 0;
2511 struct inode *inode = file_inode(in);
2513 trace_ocfs2_file_splice_read(inode, in, in->f_path.dentry,
2514 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2515 in->f_path.dentry->d_name.len,
2516 in->f_path.dentry->d_name.name, len);
2519 * See the comment in ocfs2_file_aio_read()
2521 ret = ocfs2_inode_lock_atime(inode, in->f_path.mnt, &lock_level);
2526 ocfs2_inode_unlock(inode, lock_level);
2528 ret = generic_file_splice_read(in, ppos, pipe, len, flags);
2534 static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
2535 const struct iovec *iov,
2536 unsigned long nr_segs,
2539 int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
2540 struct file *filp = iocb->ki_filp;
2541 struct inode *inode = file_inode(filp);
2543 trace_ocfs2_file_aio_read(inode, filp, filp->f_path.dentry,
2544 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2545 filp->f_path.dentry->d_name.len,
2546 filp->f_path.dentry->d_name.name, nr_segs);
2555 ocfs2_iocb_clear_sem_locked(iocb);
2558 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
2559 * need locks to protect pending reads from racing with truncate.
2561 if (filp->f_flags & O_DIRECT) {
2563 ocfs2_iocb_set_sem_locked(iocb);
2565 ret = ocfs2_rw_lock(inode, 0);
2571 /* communicate with ocfs2_dio_end_io */
2572 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2576 * We're fine letting folks race truncates and extending
2577 * writes with read across the cluster, just like they can
2578 * locally. Hence no rw_lock during read.
2580 * Take and drop the meta data lock to update inode fields
2581 * like i_size. This allows the checks down below
2582 * generic_file_aio_read() a chance of actually working.
2584 ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level);
2589 ocfs2_inode_unlock(inode, lock_level);
2591 ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
2592 trace_generic_file_aio_read_ret(ret);
2594 /* buffered aio wouldn't have proper lock coverage today */
2595 BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
2597 /* see ocfs2_file_aio_write */
2598 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2605 ocfs2_iocb_clear_sem_locked(iocb);
2608 ocfs2_rw_unlock(inode, rw_level);
2613 /* Refer generic_file_llseek_unlocked() */
2614 static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence)
2616 struct inode *inode = file->f_mapping->host;
2619 mutex_lock(&inode->i_mutex);
2625 offset += inode->i_size;
2629 offset = file->f_pos;
2632 offset += file->f_pos;
2636 ret = ocfs2_seek_data_hole_offset(file, &offset, whence);
2645 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2648 mutex_unlock(&inode->i_mutex);
2654 const struct inode_operations ocfs2_file_iops = {
2655 .setattr = ocfs2_setattr,
2656 .getattr = ocfs2_getattr,
2657 .permission = ocfs2_permission,
2658 .setxattr = generic_setxattr,
2659 .getxattr = generic_getxattr,
2660 .listxattr = ocfs2_listxattr,
2661 .removexattr = generic_removexattr,
2662 .fiemap = ocfs2_fiemap,
2663 .get_acl = ocfs2_iop_get_acl,
2666 const struct inode_operations ocfs2_special_file_iops = {
2667 .setattr = ocfs2_setattr,
2668 .getattr = ocfs2_getattr,
2669 .permission = ocfs2_permission,
2670 .get_acl = ocfs2_iop_get_acl,
2674 * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2675 * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2677 const struct file_operations ocfs2_fops = {
2678 .llseek = ocfs2_file_llseek,
2679 .read = do_sync_read,
2680 .write = do_sync_write,
2682 .fsync = ocfs2_sync_file,
2683 .release = ocfs2_file_release,
2684 .open = ocfs2_file_open,
2685 .aio_read = ocfs2_file_aio_read,
2686 .aio_write = ocfs2_file_aio_write,
2687 .unlocked_ioctl = ocfs2_ioctl,
2688 #ifdef CONFIG_COMPAT
2689 .compat_ioctl = ocfs2_compat_ioctl,
2692 .flock = ocfs2_flock,
2693 .splice_read = ocfs2_file_splice_read,
2694 .splice_write = ocfs2_file_splice_write,
2695 .fallocate = ocfs2_fallocate,
2698 const struct file_operations ocfs2_dops = {
2699 .llseek = generic_file_llseek,
2700 .read = generic_read_dir,
2701 .iterate = ocfs2_readdir,
2702 .fsync = ocfs2_sync_file,
2703 .release = ocfs2_dir_release,
2704 .open = ocfs2_dir_open,
2705 .unlocked_ioctl = ocfs2_ioctl,
2706 #ifdef CONFIG_COMPAT
2707 .compat_ioctl = ocfs2_compat_ioctl,
2710 .flock = ocfs2_flock,
2714 * POSIX-lockless variants of our file_operations.
2716 * These will be used if the underlying cluster stack does not support
2717 * posix file locking, if the user passes the "localflocks" mount
2718 * option, or if we have a local-only fs.
2720 * ocfs2_flock is in here because all stacks handle UNIX file locks,
2721 * so we still want it in the case of no stack support for
2722 * plocks. Internally, it will do the right thing when asked to ignore
2725 const struct file_operations ocfs2_fops_no_plocks = {
2726 .llseek = ocfs2_file_llseek,
2727 .read = do_sync_read,
2728 .write = do_sync_write,
2730 .fsync = ocfs2_sync_file,
2731 .release = ocfs2_file_release,
2732 .open = ocfs2_file_open,
2733 .aio_read = ocfs2_file_aio_read,
2734 .aio_write = ocfs2_file_aio_write,
2735 .unlocked_ioctl = ocfs2_ioctl,
2736 #ifdef CONFIG_COMPAT
2737 .compat_ioctl = ocfs2_compat_ioctl,
2739 .flock = ocfs2_flock,
2740 .splice_read = ocfs2_file_splice_read,
2741 .splice_write = ocfs2_file_splice_write,
2742 .fallocate = ocfs2_fallocate,
2745 const struct file_operations ocfs2_dops_no_plocks = {
2746 .llseek = generic_file_llseek,
2747 .read = generic_read_dir,
2748 .iterate = ocfs2_readdir,
2749 .fsync = ocfs2_sync_file,
2750 .release = ocfs2_dir_release,
2751 .open = ocfs2_dir_open,
2752 .unlocked_ioctl = ocfs2_ioctl,
2753 #ifdef CONFIG_COMPAT
2754 .compat_ioctl = ocfs2_compat_ioctl,
2756 .flock = ocfs2_flock,