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 #define MLOG_MASK_PREFIX ML_INODE
42 #include <cluster/masklog.h>
50 #include "extent_map.h"
63 #include "refcounttree.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 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
103 file->f_path.dentry->d_name.len, file->f_path.dentry->d_name.name);
105 if (file->f_mode & FMODE_WRITE)
106 dquot_initialize(inode);
108 spin_lock(&oi->ip_lock);
110 /* Check that the inode hasn't been wiped from disk by another
111 * node. If it hasn't then we're safe as long as we hold the
112 * spin lock until our increment of open count. */
113 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
114 spin_unlock(&oi->ip_lock);
121 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
124 spin_unlock(&oi->ip_lock);
126 status = ocfs2_init_file_private(inode, file);
129 * We want to set open count back if we're failing the
132 spin_lock(&oi->ip_lock);
134 spin_unlock(&oi->ip_lock);
142 static int ocfs2_file_release(struct inode *inode, struct file *file)
144 struct ocfs2_inode_info *oi = OCFS2_I(inode);
146 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
147 file->f_path.dentry->d_name.len,
148 file->f_path.dentry->d_name.name);
150 spin_lock(&oi->ip_lock);
151 if (!--oi->ip_open_count)
152 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
153 spin_unlock(&oi->ip_lock);
155 ocfs2_free_file_private(inode, file);
162 static int ocfs2_dir_open(struct inode *inode, struct file *file)
164 return ocfs2_init_file_private(inode, file);
167 static int ocfs2_dir_release(struct inode *inode, struct file *file)
169 ocfs2_free_file_private(inode, file);
173 static int ocfs2_sync_file(struct file *file, int datasync)
177 struct inode *inode = file->f_mapping->host;
178 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
180 mlog_entry("(0x%p, %d, 0x%p, '%.*s')\n", file, datasync,
181 file->f_path.dentry, file->f_path.dentry->d_name.len,
182 file->f_path.dentry->d_name.name);
184 if (datasync && !(inode->i_state & I_DIRTY_DATASYNC)) {
186 * We still have to flush drive's caches to get data to the
189 if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
190 blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
194 journal = osb->journal->j_journal;
195 err = jbd2_journal_force_commit(journal);
200 return (err < 0) ? -EIO : 0;
203 int ocfs2_should_update_atime(struct inode *inode,
204 struct vfsmount *vfsmnt)
207 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
209 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
212 if ((inode->i_flags & S_NOATIME) ||
213 ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
217 * We can be called with no vfsmnt structure - NFSD will
220 * Note that our action here is different than touch_atime() -
221 * if we can't tell whether this is a noatime mount, then we
222 * don't know whether to trust the value of s_atime_quantum.
227 if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
228 ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
231 if (vfsmnt->mnt_flags & MNT_RELATIME) {
232 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
233 (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
240 if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
246 int ocfs2_update_inode_atime(struct inode *inode,
247 struct buffer_head *bh)
250 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
252 struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
256 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
257 if (IS_ERR(handle)) {
258 ret = PTR_ERR(handle);
263 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
264 OCFS2_JOURNAL_ACCESS_WRITE);
271 * Don't use ocfs2_mark_inode_dirty() here as we don't always
272 * have i_mutex to guard against concurrent changes to other
275 inode->i_atime = CURRENT_TIME;
276 di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
277 di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
278 ocfs2_journal_dirty(handle, bh);
281 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
287 static int ocfs2_set_inode_size(handle_t *handle,
289 struct buffer_head *fe_bh,
295 i_size_write(inode, new_i_size);
296 inode->i_blocks = ocfs2_inode_sector_count(inode);
297 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
299 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
310 int ocfs2_simple_size_update(struct inode *inode,
311 struct buffer_head *di_bh,
315 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
316 handle_t *handle = NULL;
318 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
319 if (IS_ERR(handle)) {
320 ret = PTR_ERR(handle);
325 ret = ocfs2_set_inode_size(handle, inode, di_bh,
330 ocfs2_commit_trans(osb, handle);
335 static int ocfs2_cow_file_pos(struct inode *inode,
336 struct buffer_head *fe_bh,
340 u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
341 unsigned int num_clusters = 0;
342 unsigned int ext_flags = 0;
345 * If the new offset is aligned to the range of the cluster, there is
346 * no space for ocfs2_zero_range_for_truncate to fill, so no need to
349 if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
352 status = ocfs2_get_clusters(inode, cpos, &phys,
353 &num_clusters, &ext_flags);
359 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
362 return ocfs2_refcount_cow(inode, NULL, fe_bh, cpos, 1, cpos+1);
368 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
370 struct buffer_head *fe_bh,
375 struct ocfs2_dinode *di;
381 * We need to CoW the cluster contains the offset if it is reflinked
382 * since we will call ocfs2_zero_range_for_truncate later which will
383 * write "0" from offset to the end of the cluster.
385 status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
391 /* TODO: This needs to actually orphan the inode in this
394 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
395 if (IS_ERR(handle)) {
396 status = PTR_ERR(handle);
401 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
402 OCFS2_JOURNAL_ACCESS_WRITE);
409 * Do this before setting i_size.
411 cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
412 status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
419 i_size_write(inode, new_i_size);
420 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
422 di = (struct ocfs2_dinode *) fe_bh->b_data;
423 di->i_size = cpu_to_le64(new_i_size);
424 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
425 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
427 ocfs2_journal_dirty(handle, fe_bh);
430 ocfs2_commit_trans(osb, handle);
437 static int ocfs2_truncate_file(struct inode *inode,
438 struct buffer_head *di_bh,
442 struct ocfs2_dinode *fe = NULL;
443 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
445 mlog_entry("(inode = %llu, new_i_size = %llu\n",
446 (unsigned long long)OCFS2_I(inode)->ip_blkno,
447 (unsigned long long)new_i_size);
449 /* We trust di_bh because it comes from ocfs2_inode_lock(), which
450 * already validated it */
451 fe = (struct ocfs2_dinode *) di_bh->b_data;
453 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
454 "Inode %llu, inode i_size = %lld != di "
455 "i_size = %llu, i_flags = 0x%x\n",
456 (unsigned long long)OCFS2_I(inode)->ip_blkno,
458 (unsigned long long)le64_to_cpu(fe->i_size),
459 le32_to_cpu(fe->i_flags));
461 if (new_i_size > le64_to_cpu(fe->i_size)) {
462 mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
463 (unsigned long long)le64_to_cpu(fe->i_size),
464 (unsigned long long)new_i_size);
470 mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
471 (unsigned long long)le64_to_cpu(fe->i_blkno),
472 (unsigned long long)le64_to_cpu(fe->i_size),
473 (unsigned long long)new_i_size);
475 /* lets handle the simple truncate cases before doing any more
476 * cluster locking. */
477 if (new_i_size == le64_to_cpu(fe->i_size))
480 down_write(&OCFS2_I(inode)->ip_alloc_sem);
482 ocfs2_resv_discard(&osb->osb_la_resmap,
483 &OCFS2_I(inode)->ip_la_data_resv);
486 * The inode lock forced other nodes to sync and drop their
487 * pages, which (correctly) happens even if we have a truncate
488 * without allocation change - ocfs2 cluster sizes can be much
489 * greater than page size, so we have to truncate them
492 unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
493 truncate_inode_pages(inode->i_mapping, new_i_size);
495 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
496 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
497 i_size_read(inode), 1);
501 goto bail_unlock_sem;
504 /* alright, we're going to need to do a full blown alloc size
505 * change. Orphan the inode so that recovery can complete the
506 * truncate if necessary. This does the task of marking
508 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
511 goto bail_unlock_sem;
514 status = ocfs2_commit_truncate(osb, inode, di_bh);
517 goto bail_unlock_sem;
520 /* TODO: orphan dir cleanup here. */
522 up_write(&OCFS2_I(inode)->ip_alloc_sem);
525 if (!status && OCFS2_I(inode)->ip_clusters == 0)
526 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
533 * extend file allocation only here.
534 * we'll update all the disk stuff, and oip->alloc_size
536 * expect stuff to be locked, a transaction started and enough data /
537 * metadata reservations in the contexts.
539 * Will return -EAGAIN, and a reason if a restart is needed.
540 * If passed in, *reason will always be set, even in error.
542 int ocfs2_add_inode_data(struct ocfs2_super *osb,
547 struct buffer_head *fe_bh,
549 struct ocfs2_alloc_context *data_ac,
550 struct ocfs2_alloc_context *meta_ac,
551 enum ocfs2_alloc_restarted *reason_ret)
554 struct ocfs2_extent_tree et;
556 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
557 ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
558 clusters_to_add, mark_unwritten,
559 data_ac, meta_ac, reason_ret);
564 static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
565 u32 clusters_to_add, int mark_unwritten)
568 int restart_func = 0;
571 struct buffer_head *bh = NULL;
572 struct ocfs2_dinode *fe = NULL;
573 handle_t *handle = NULL;
574 struct ocfs2_alloc_context *data_ac = NULL;
575 struct ocfs2_alloc_context *meta_ac = NULL;
576 enum ocfs2_alloc_restarted why;
577 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
578 struct ocfs2_extent_tree et;
581 mlog_entry("(clusters_to_add = %u)\n", clusters_to_add);
584 * This function only exists for file systems which don't
587 BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
589 status = ocfs2_read_inode_block(inode, &bh);
594 fe = (struct ocfs2_dinode *) bh->b_data;
597 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
599 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, "
600 "clusters_to_add = %u\n",
601 (unsigned long long)OCFS2_I(inode)->ip_blkno,
602 (long long)i_size_read(inode), le32_to_cpu(fe->i_clusters),
604 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
605 status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
612 credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list,
614 handle = ocfs2_start_trans(osb, credits);
615 if (IS_ERR(handle)) {
616 status = PTR_ERR(handle);
622 restarted_transaction:
623 status = dquot_alloc_space_nodirty(inode,
624 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
629 /* reserve a write to the file entry early on - that we if we
630 * run out of credits in the allocation path, we can still
632 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
633 OCFS2_JOURNAL_ACCESS_WRITE);
639 prev_clusters = OCFS2_I(inode)->ip_clusters;
641 status = ocfs2_add_inode_data(osb,
651 if ((status < 0) && (status != -EAGAIN)) {
652 if (status != -ENOSPC)
657 ocfs2_journal_dirty(handle, bh);
659 spin_lock(&OCFS2_I(inode)->ip_lock);
660 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
661 spin_unlock(&OCFS2_I(inode)->ip_lock);
662 /* Release unused quota reservation */
663 dquot_free_space(inode,
664 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
667 if (why != RESTART_NONE && clusters_to_add) {
668 if (why == RESTART_META) {
669 mlog(0, "restarting function.\n");
673 BUG_ON(why != RESTART_TRANS);
675 mlog(0, "restarting transaction.\n");
676 /* TODO: This can be more intelligent. */
677 credits = ocfs2_calc_extend_credits(osb->sb,
680 status = ocfs2_extend_trans(handle, credits);
682 /* handle still has to be committed at
688 goto restarted_transaction;
692 mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
693 le32_to_cpu(fe->i_clusters),
694 (unsigned long long)le64_to_cpu(fe->i_size));
695 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
696 OCFS2_I(inode)->ip_clusters, (long long)i_size_read(inode));
699 if (status < 0 && did_quota)
700 dquot_free_space(inode,
701 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
703 ocfs2_commit_trans(osb, handle);
707 ocfs2_free_alloc_context(data_ac);
711 ocfs2_free_alloc_context(meta_ac);
714 if ((!status) && restart_func) {
726 * While a write will already be ordering the data, a truncate will not.
727 * Thus, we need to explicitly order the zeroed pages.
729 static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode)
731 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
732 handle_t *handle = NULL;
735 if (!ocfs2_should_order_data(inode))
738 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
739 if (IS_ERR(handle)) {
745 ret = ocfs2_jbd2_file_inode(handle, inode);
752 ocfs2_commit_trans(osb, handle);
753 handle = ERR_PTR(ret);
758 /* Some parts of this taken from generic_cont_expand, which turned out
759 * to be too fragile to do exactly what we need without us having to
760 * worry about recursive locking in ->write_begin() and ->write_end(). */
761 static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
764 struct address_space *mapping = inode->i_mapping;
766 unsigned long index = abs_from >> PAGE_CACHE_SHIFT;
767 handle_t *handle = NULL;
769 unsigned zero_from, zero_to, block_start, block_end;
771 BUG_ON(abs_from >= abs_to);
772 BUG_ON(abs_to > (((u64)index + 1) << PAGE_CACHE_SHIFT));
773 BUG_ON(abs_from & (inode->i_blkbits - 1));
775 page = find_or_create_page(mapping, index, GFP_NOFS);
782 /* Get the offsets within the page that we want to zero */
783 zero_from = abs_from & (PAGE_CACHE_SIZE - 1);
784 zero_to = abs_to & (PAGE_CACHE_SIZE - 1);
786 zero_to = PAGE_CACHE_SIZE;
789 "abs_from = %llu, abs_to = %llu, index = %lu, zero_from = %u, zero_to = %u\n",
790 (unsigned long long)abs_from, (unsigned long long)abs_to,
791 index, zero_from, zero_to);
793 /* We know that zero_from is block aligned */
794 for (block_start = zero_from; block_start < zero_to;
795 block_start = block_end) {
796 block_end = block_start + (1 << inode->i_blkbits);
799 * block_start is block-aligned. Bump it by one to force
800 * __block_write_begin and block_commit_write to zero the
803 ret = __block_write_begin(page, block_start + 1, 0,
811 handle = ocfs2_zero_start_ordered_transaction(inode);
812 if (IS_ERR(handle)) {
813 ret = PTR_ERR(handle);
819 /* must not update i_size! */
820 ret = block_commit_write(page, block_start + 1,
829 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
833 page_cache_release(page);
839 * Find the next range to zero. We do this in terms of bytes because
840 * that's what ocfs2_zero_extend() wants, and it is dealing with the
841 * pagecache. We may return multiple extents.
843 * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
844 * needs to be zeroed. range_start and range_end return the next zeroing
845 * range. A subsequent call should pass the previous range_end as its
846 * zero_start. If range_end is 0, there's nothing to do.
848 * Unwritten extents are skipped over. Refcounted extents are CoWd.
850 static int ocfs2_zero_extend_get_range(struct inode *inode,
851 struct buffer_head *di_bh,
852 u64 zero_start, u64 zero_end,
853 u64 *range_start, u64 *range_end)
855 int rc = 0, needs_cow = 0;
856 u32 p_cpos, zero_clusters = 0;
858 zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
859 u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
860 unsigned int num_clusters = 0;
861 unsigned int ext_flags = 0;
863 while (zero_cpos < last_cpos) {
864 rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
865 &num_clusters, &ext_flags);
871 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
872 zero_clusters = num_clusters;
873 if (ext_flags & OCFS2_EXT_REFCOUNTED)
878 zero_cpos += num_clusters;
880 if (!zero_clusters) {
885 while ((zero_cpos + zero_clusters) < last_cpos) {
886 rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
887 &p_cpos, &num_clusters,
894 if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
896 if (ext_flags & OCFS2_EXT_REFCOUNTED)
898 zero_clusters += num_clusters;
900 if ((zero_cpos + zero_clusters) > last_cpos)
901 zero_clusters = last_cpos - zero_cpos;
904 rc = ocfs2_refcount_cow(inode, NULL, di_bh, zero_cpos,
905 zero_clusters, UINT_MAX);
912 *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
913 *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
914 zero_cpos + zero_clusters);
921 * Zero one range returned from ocfs2_zero_extend_get_range(). The caller
922 * has made sure that the entire range needs zeroing.
924 static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
929 u64 zero_pos = range_start;
931 mlog(0, "range_start = %llu, range_end = %llu\n",
932 (unsigned long long)range_start,
933 (unsigned long long)range_end);
934 BUG_ON(range_start >= range_end);
936 while (zero_pos < range_end) {
937 next_pos = (zero_pos & PAGE_CACHE_MASK) + PAGE_CACHE_SIZE;
938 if (next_pos > range_end)
939 next_pos = range_end;
940 rc = ocfs2_write_zero_page(inode, zero_pos, next_pos);
948 * Very large extends have the potential to lock up
949 * the cpu for extended periods of time.
957 int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
961 u64 zero_start, range_start = 0, range_end = 0;
962 struct super_block *sb = inode->i_sb;
964 zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
965 mlog(0, "zero_start %llu for i_size %llu\n",
966 (unsigned long long)zero_start,
967 (unsigned long long)i_size_read(inode));
968 while (zero_start < zero_to_size) {
969 ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
980 if (range_start < zero_start)
981 range_start = zero_start;
982 if (range_end > zero_to_size)
983 range_end = zero_to_size;
985 ret = ocfs2_zero_extend_range(inode, range_start,
991 zero_start = range_end;
997 int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
998 u64 new_i_size, u64 zero_to)
1001 u32 clusters_to_add;
1002 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1005 * Only quota files call this without a bh, and they can't be
1008 BUG_ON(!di_bh && (oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
1009 BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1011 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1012 if (clusters_to_add < oi->ip_clusters)
1013 clusters_to_add = 0;
1015 clusters_to_add -= oi->ip_clusters;
1017 if (clusters_to_add) {
1018 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
1019 clusters_to_add, 0);
1027 * Call this even if we don't add any clusters to the tree. We
1028 * still need to zero the area between the old i_size and the
1031 ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1039 static int ocfs2_extend_file(struct inode *inode,
1040 struct buffer_head *di_bh,
1044 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1048 /* setattr sometimes calls us like this. */
1049 if (new_i_size == 0)
1052 if (i_size_read(inode) == new_i_size)
1054 BUG_ON(new_i_size < i_size_read(inode));
1057 * The alloc sem blocks people in read/write from reading our
1058 * allocation until we're done changing it. We depend on
1059 * i_mutex to block other extend/truncate calls while we're
1060 * here. We even have to hold it for sparse files because there
1061 * might be some tail zeroing.
1063 down_write(&oi->ip_alloc_sem);
1065 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1067 * We can optimize small extends by keeping the inodes
1070 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1071 up_write(&oi->ip_alloc_sem);
1072 goto out_update_size;
1075 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1077 up_write(&oi->ip_alloc_sem);
1083 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1084 ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1086 ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1089 up_write(&oi->ip_alloc_sem);
1097 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1105 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1107 int status = 0, size_change;
1108 struct inode *inode = dentry->d_inode;
1109 struct super_block *sb = inode->i_sb;
1110 struct ocfs2_super *osb = OCFS2_SB(sb);
1111 struct buffer_head *bh = NULL;
1112 handle_t *handle = NULL;
1113 struct dquot *transfer_to[MAXQUOTAS] = { };
1116 mlog_entry("(0x%p, '%.*s')\n", dentry,
1117 dentry->d_name.len, dentry->d_name.name);
1119 /* ensuring we don't even attempt to truncate a symlink */
1120 if (S_ISLNK(inode->i_mode))
1121 attr->ia_valid &= ~ATTR_SIZE;
1123 if (attr->ia_valid & ATTR_MODE)
1124 mlog(0, "mode change: %d\n", attr->ia_mode);
1125 if (attr->ia_valid & ATTR_UID)
1126 mlog(0, "uid change: %d\n", attr->ia_uid);
1127 if (attr->ia_valid & ATTR_GID)
1128 mlog(0, "gid change: %d\n", attr->ia_gid);
1129 if (attr->ia_valid & ATTR_SIZE)
1130 mlog(0, "size change...\n");
1131 if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME))
1132 mlog(0, "time change...\n");
1134 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1135 | ATTR_GID | ATTR_UID | ATTR_MODE)
1136 if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) {
1137 mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid);
1141 status = inode_change_ok(inode, attr);
1145 if (is_quota_modification(inode, attr))
1146 dquot_initialize(inode);
1147 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1149 status = ocfs2_rw_lock(inode, 1);
1156 status = ocfs2_inode_lock(inode, &bh, 1);
1158 if (status != -ENOENT)
1160 goto bail_unlock_rw;
1163 if (size_change && attr->ia_size != i_size_read(inode)) {
1164 status = inode_newsize_ok(inode, attr->ia_size);
1168 if (i_size_read(inode) > attr->ia_size) {
1169 if (ocfs2_should_order_data(inode)) {
1170 status = ocfs2_begin_ordered_truncate(inode,
1175 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1177 status = ocfs2_extend_file(inode, bh, attr->ia_size);
1179 if (status != -ENOSPC)
1186 if ((attr->ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
1187 (attr->ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
1189 * Gather pointers to quota structures so that allocation /
1190 * freeing of quota structures happens here and not inside
1191 * dquot_transfer() where we have problems with lock ordering
1193 if (attr->ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid
1194 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1195 OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1196 transfer_to[USRQUOTA] = dqget(sb, attr->ia_uid,
1198 if (!transfer_to[USRQUOTA]) {
1203 if (attr->ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid
1204 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1205 OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1206 transfer_to[GRPQUOTA] = dqget(sb, attr->ia_gid,
1208 if (!transfer_to[GRPQUOTA]) {
1213 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1214 2 * ocfs2_quota_trans_credits(sb));
1215 if (IS_ERR(handle)) {
1216 status = PTR_ERR(handle);
1220 status = __dquot_transfer(inode, transfer_to);
1224 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1225 if (IS_ERR(handle)) {
1226 status = PTR_ERR(handle);
1233 * This will intentionally not wind up calling truncate_setsize(),
1234 * since all the work for a size change has been done above.
1235 * Otherwise, we could get into problems with truncate as
1236 * ip_alloc_sem is used there to protect against i_size
1239 * XXX: this means the conditional below can probably be removed.
1241 if ((attr->ia_valid & ATTR_SIZE) &&
1242 attr->ia_size != i_size_read(inode)) {
1243 status = vmtruncate(inode, attr->ia_size);
1250 setattr_copy(inode, attr);
1251 mark_inode_dirty(inode);
1253 status = ocfs2_mark_inode_dirty(handle, inode, bh);
1258 ocfs2_commit_trans(osb, handle);
1260 ocfs2_inode_unlock(inode, 1);
1263 ocfs2_rw_unlock(inode, 1);
1267 /* Release quota pointers in case we acquired them */
1268 for (qtype = 0; qtype < MAXQUOTAS; qtype++)
1269 dqput(transfer_to[qtype]);
1271 if (!status && attr->ia_valid & ATTR_MODE) {
1272 status = ocfs2_acl_chmod(inode);
1281 int ocfs2_getattr(struct vfsmount *mnt,
1282 struct dentry *dentry,
1285 struct inode *inode = dentry->d_inode;
1286 struct super_block *sb = dentry->d_inode->i_sb;
1287 struct ocfs2_super *osb = sb->s_fs_info;
1292 err = ocfs2_inode_revalidate(dentry);
1299 generic_fillattr(inode, stat);
1301 /* We set the blksize from the cluster size for performance */
1302 stat->blksize = osb->s_clustersize;
1310 int ocfs2_permission(struct inode *inode, int mask, unsigned int flags)
1314 if (flags & IPERM_FLAG_RCU)
1319 ret = ocfs2_inode_lock(inode, NULL, 0);
1326 ret = generic_permission(inode, mask, flags, ocfs2_check_acl);
1328 ocfs2_inode_unlock(inode, 0);
1334 static int __ocfs2_write_remove_suid(struct inode *inode,
1335 struct buffer_head *bh)
1339 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1340 struct ocfs2_dinode *di;
1342 mlog_entry("(Inode %llu, mode 0%o)\n",
1343 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_mode);
1345 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1346 if (IS_ERR(handle)) {
1347 ret = PTR_ERR(handle);
1352 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1353 OCFS2_JOURNAL_ACCESS_WRITE);
1359 inode->i_mode &= ~S_ISUID;
1360 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1361 inode->i_mode &= ~S_ISGID;
1363 di = (struct ocfs2_dinode *) bh->b_data;
1364 di->i_mode = cpu_to_le16(inode->i_mode);
1366 ocfs2_journal_dirty(handle, bh);
1369 ocfs2_commit_trans(osb, handle);
1376 * Will look for holes and unwritten extents in the range starting at
1377 * pos for count bytes (inclusive).
1379 static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1383 unsigned int extent_flags;
1384 u32 cpos, clusters, extent_len, phys_cpos;
1385 struct super_block *sb = inode->i_sb;
1387 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1388 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1391 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1398 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
1403 if (extent_len > clusters)
1404 extent_len = clusters;
1406 clusters -= extent_len;
1413 static int ocfs2_write_remove_suid(struct inode *inode)
1416 struct buffer_head *bh = NULL;
1418 ret = ocfs2_read_inode_block(inode, &bh);
1424 ret = __ocfs2_write_remove_suid(inode, bh);
1431 * Allocate enough extents to cover the region starting at byte offset
1432 * start for len bytes. Existing extents are skipped, any extents
1433 * added are marked as "unwritten".
1435 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1439 u32 cpos, phys_cpos, clusters, alloc_size;
1440 u64 end = start + len;
1441 struct buffer_head *di_bh = NULL;
1443 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1444 ret = ocfs2_read_inode_block(inode, &di_bh);
1451 * Nothing to do if the requested reservation range
1452 * fits within the inode.
1454 if (ocfs2_size_fits_inline_data(di_bh, end))
1457 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1465 * We consider both start and len to be inclusive.
1467 cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1468 clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1472 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1480 * Hole or existing extent len can be arbitrary, so
1481 * cap it to our own allocation request.
1483 if (alloc_size > clusters)
1484 alloc_size = clusters;
1488 * We already have an allocation at this
1489 * region so we can safely skip it.
1494 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1503 clusters -= alloc_size;
1514 * Truncate a byte range, avoiding pages within partial clusters. This
1515 * preserves those pages for the zeroing code to write to.
1517 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1520 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1522 struct address_space *mapping = inode->i_mapping;
1524 start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1525 end = byte_start + byte_len;
1526 end = end & ~(osb->s_clustersize - 1);
1529 unmap_mapping_range(mapping, start, end - start, 0);
1530 truncate_inode_pages_range(mapping, start, end - 1);
1534 static int ocfs2_zero_partial_clusters(struct inode *inode,
1538 u64 tmpend, end = start + len;
1539 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1540 unsigned int csize = osb->s_clustersize;
1544 * The "start" and "end" values are NOT necessarily part of
1545 * the range whose allocation is being deleted. Rather, this
1546 * is what the user passed in with the request. We must zero
1547 * partial clusters here. There's no need to worry about
1548 * physical allocation - the zeroing code knows to skip holes.
1550 mlog(0, "byte start: %llu, end: %llu\n",
1551 (unsigned long long)start, (unsigned long long)end);
1554 * If both edges are on a cluster boundary then there's no
1555 * zeroing required as the region is part of the allocation to
1558 if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1561 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1562 if (IS_ERR(handle)) {
1563 ret = PTR_ERR(handle);
1569 * We want to get the byte offset of the end of the 1st cluster.
1571 tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1575 mlog(0, "1st range: start: %llu, tmpend: %llu\n",
1576 (unsigned long long)start, (unsigned long long)tmpend);
1578 ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1584 * This may make start and end equal, but the zeroing
1585 * code will skip any work in that case so there's no
1586 * need to catch it up here.
1588 start = end & ~(osb->s_clustersize - 1);
1590 mlog(0, "2nd range: start: %llu, end: %llu\n",
1591 (unsigned long long)start, (unsigned long long)end);
1593 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1598 ocfs2_commit_trans(osb, handle);
1603 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1606 struct ocfs2_extent_rec *rec = NULL;
1608 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1610 rec = &el->l_recs[i];
1612 if (le32_to_cpu(rec->e_cpos) < pos)
1620 * Helper to calculate the punching pos and length in one run, we handle the
1621 * following three cases in order:
1623 * - remove the entire record
1624 * - remove a partial record
1625 * - no record needs to be removed (hole-punching completed)
1627 static void ocfs2_calc_trunc_pos(struct inode *inode,
1628 struct ocfs2_extent_list *el,
1629 struct ocfs2_extent_rec *rec,
1630 u32 trunc_start, u32 *trunc_cpos,
1631 u32 *trunc_len, u32 *trunc_end,
1632 u64 *blkno, int *done)
1637 range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1639 if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1640 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1642 * Skip holes if any.
1644 if (range < *trunc_end)
1646 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1647 *blkno = le64_to_cpu(rec->e_blkno);
1648 *trunc_end = le32_to_cpu(rec->e_cpos);
1649 } else if (range > trunc_start) {
1650 *trunc_cpos = trunc_start;
1651 *trunc_len = *trunc_end - trunc_start;
1652 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1653 *blkno = le64_to_cpu(rec->e_blkno) +
1654 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1655 *trunc_end = trunc_start;
1658 * It may have two following possibilities:
1660 * - last record has been removed
1661 * - trunc_start was within a hole
1663 * both two cases mean the completion of hole punching.
1671 static int ocfs2_remove_inode_range(struct inode *inode,
1672 struct buffer_head *di_bh, u64 byte_start,
1675 int ret = 0, flags = 0, done = 0, i;
1676 u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1678 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1679 struct ocfs2_cached_dealloc_ctxt dealloc;
1680 struct address_space *mapping = inode->i_mapping;
1681 struct ocfs2_extent_tree et;
1682 struct ocfs2_path *path = NULL;
1683 struct ocfs2_extent_list *el = NULL;
1684 struct ocfs2_extent_rec *rec = NULL;
1685 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1686 u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1688 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1689 ocfs2_init_dealloc_ctxt(&dealloc);
1694 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1695 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1696 byte_start + byte_len, 0);
1702 * There's no need to get fancy with the page cache
1703 * truncate of an inline-data inode. We're talking
1704 * about less than a page here, which will be cached
1705 * in the dinode buffer anyway.
1707 unmap_mapping_range(mapping, 0, 0, 0);
1708 truncate_inode_pages(mapping, 0);
1713 * For reflinks, we may need to CoW 2 clusters which might be
1714 * partially zero'd later, if hole's start and end offset were
1715 * within one cluster(means is not exactly aligned to clustersize).
1718 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
1720 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1726 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1733 trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1734 trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1735 cluster_in_el = trunc_end;
1737 mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, cend: %u\n",
1738 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1739 (unsigned long long)byte_start,
1740 (unsigned long long)byte_len, trunc_start, trunc_end);
1742 ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1748 path = ocfs2_new_path_from_et(&et);
1755 while (trunc_end > trunc_start) {
1757 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1764 el = path_leaf_el(path);
1766 i = ocfs2_find_rec(el, trunc_end);
1768 * Need to go to previous extent block.
1771 if (path->p_tree_depth == 0)
1774 ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1783 * We've reached the leftmost extent block,
1784 * it's safe to leave.
1786 if (cluster_in_el == 0)
1790 * The 'pos' searched for previous extent block is
1791 * always one cluster less than actual trunc_end.
1793 trunc_end = cluster_in_el + 1;
1795 ocfs2_reinit_path(path, 1);
1800 rec = &el->l_recs[i];
1802 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1803 &trunc_len, &trunc_end, &blkno, &done);
1807 flags = rec->e_flags;
1808 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1810 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1811 phys_cpos, trunc_len, flags,
1812 &dealloc, refcount_loc);
1818 cluster_in_el = trunc_end;
1820 ocfs2_reinit_path(path, 1);
1823 ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1826 ocfs2_schedule_truncate_log_flush(osb, 1);
1827 ocfs2_run_deallocs(osb, &dealloc);
1833 * Parts of this function taken from xfs_change_file_space()
1835 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1836 loff_t f_pos, unsigned int cmd,
1837 struct ocfs2_space_resv *sr,
1843 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1844 struct buffer_head *di_bh = NULL;
1846 unsigned long long max_off = inode->i_sb->s_maxbytes;
1848 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1851 mutex_lock(&inode->i_mutex);
1854 * This prevents concurrent writes on other nodes
1856 ret = ocfs2_rw_lock(inode, 1);
1862 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1868 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1870 goto out_inode_unlock;
1873 switch (sr->l_whence) {
1874 case 0: /*SEEK_SET*/
1876 case 1: /*SEEK_CUR*/
1877 sr->l_start += f_pos;
1879 case 2: /*SEEK_END*/
1880 sr->l_start += i_size_read(inode);
1884 goto out_inode_unlock;
1888 llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1891 || sr->l_start > max_off
1892 || (sr->l_start + llen) < 0
1893 || (sr->l_start + llen) > max_off) {
1895 goto out_inode_unlock;
1897 size = sr->l_start + sr->l_len;
1899 if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) {
1900 if (sr->l_len <= 0) {
1902 goto out_inode_unlock;
1906 if (file && should_remove_suid(file->f_path.dentry)) {
1907 ret = __ocfs2_write_remove_suid(inode, di_bh);
1910 goto out_inode_unlock;
1914 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1916 case OCFS2_IOC_RESVSP:
1917 case OCFS2_IOC_RESVSP64:
1919 * This takes unsigned offsets, but the signed ones we
1920 * pass have been checked against overflow above.
1922 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1925 case OCFS2_IOC_UNRESVSP:
1926 case OCFS2_IOC_UNRESVSP64:
1927 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1933 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1936 goto out_inode_unlock;
1940 * We update c/mtime for these changes
1942 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1943 if (IS_ERR(handle)) {
1944 ret = PTR_ERR(handle);
1946 goto out_inode_unlock;
1949 if (change_size && i_size_read(inode) < size)
1950 i_size_write(inode, size);
1952 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1953 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1957 ocfs2_commit_trans(osb, handle);
1961 ocfs2_inode_unlock(inode, 1);
1963 ocfs2_rw_unlock(inode, 1);
1966 mutex_unlock(&inode->i_mutex);
1970 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1971 struct ocfs2_space_resv *sr)
1973 struct inode *inode = file->f_path.dentry->d_inode;
1974 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1976 if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1977 !ocfs2_writes_unwritten_extents(osb))
1979 else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1980 !ocfs2_sparse_alloc(osb))
1983 if (!S_ISREG(inode->i_mode))
1986 if (!(file->f_mode & FMODE_WRITE))
1989 return __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1992 static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
1995 struct inode *inode = file->f_path.dentry->d_inode;
1996 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1997 struct ocfs2_space_resv sr;
1998 int change_size = 1;
1999 int cmd = OCFS2_IOC_RESVSP64;
2001 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2003 if (!ocfs2_writes_unwritten_extents(osb))
2006 if (mode & FALLOC_FL_KEEP_SIZE)
2009 if (mode & FALLOC_FL_PUNCH_HOLE)
2010 cmd = OCFS2_IOC_UNRESVSP64;
2013 sr.l_start = (s64)offset;
2014 sr.l_len = (s64)len;
2016 return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2020 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2024 unsigned int extent_flags;
2025 u32 cpos, clusters, extent_len, phys_cpos;
2026 struct super_block *sb = inode->i_sb;
2028 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2029 !(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) ||
2030 OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2033 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2034 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2037 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2044 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2049 if (extent_len > clusters)
2050 extent_len = clusters;
2052 clusters -= extent_len;
2059 static int ocfs2_prepare_inode_for_refcount(struct inode *inode,
2061 loff_t pos, size_t count,
2065 struct buffer_head *di_bh = NULL;
2066 u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2068 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2070 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2078 ret = ocfs2_refcount_cow(inode, file, di_bh, cpos, clusters, UINT_MAX);
2086 static int ocfs2_prepare_inode_for_write(struct file *file,
2093 int ret = 0, meta_level = 0;
2094 struct dentry *dentry = file->f_path.dentry;
2095 struct inode *inode = dentry->d_inode;
2096 loff_t saved_pos, end;
2099 * We start with a read level meta lock and only jump to an ex
2100 * if we need to make modifications here.
2103 ret = ocfs2_inode_lock(inode, NULL, meta_level);
2110 /* Clear suid / sgid if necessary. We do this here
2111 * instead of later in the write path because
2112 * remove_suid() calls ->setattr without any hint that
2113 * we may have already done our cluster locking. Since
2114 * ocfs2_setattr() *must* take cluster locks to
2115 * proceeed, this will lead us to recursively lock the
2116 * inode. There's also the dinode i_size state which
2117 * can be lost via setattr during extending writes (we
2118 * set inode->i_size at the end of a write. */
2119 if (should_remove_suid(dentry)) {
2120 if (meta_level == 0) {
2121 ocfs2_inode_unlock(inode, meta_level);
2126 ret = ocfs2_write_remove_suid(inode);
2133 /* work on a copy of ppos until we're sure that we won't have
2134 * to recalculate it due to relocking. */
2136 saved_pos = i_size_read(inode);
2137 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos);
2142 end = saved_pos + count;
2144 ret = ocfs2_check_range_for_refcount(inode, saved_pos, count);
2146 ocfs2_inode_unlock(inode, meta_level);
2149 ret = ocfs2_prepare_inode_for_refcount(inode,
2166 * Skip the O_DIRECT checks if we don't need
2169 if (!direct_io || !(*direct_io))
2173 * There's no sane way to do direct writes to an inode
2176 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2182 * Allowing concurrent direct writes means
2183 * i_size changes wouldn't be synchronized, so
2184 * one node could wind up truncating another
2187 if (end > i_size_read(inode)) {
2193 * We don't fill holes during direct io, so
2194 * check for them here. If any are found, the
2195 * caller will have to retake some cluster
2196 * locks and initiate the io as buffered.
2198 ret = ocfs2_check_range_for_holes(inode, saved_pos, count);
2211 if (meta_level >= 0)
2212 ocfs2_inode_unlock(inode, meta_level);
2218 static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
2219 const struct iovec *iov,
2220 unsigned long nr_segs,
2223 int ret, direct_io, appending, rw_level, have_alloc_sem = 0;
2224 int can_do_direct, has_refcount = 0;
2225 ssize_t written = 0;
2226 size_t ocount; /* original count */
2227 size_t count; /* after file limit checks */
2228 loff_t old_size, *ppos = &iocb->ki_pos;
2230 struct file *file = iocb->ki_filp;
2231 struct inode *inode = file->f_path.dentry->d_inode;
2232 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2233 int full_coherency = !(osb->s_mount_opt &
2234 OCFS2_MOUNT_COHERENCY_BUFFERED);
2236 mlog_entry("(0x%p, %u, '%.*s')\n", file,
2237 (unsigned int)nr_segs,
2238 file->f_path.dentry->d_name.len,
2239 file->f_path.dentry->d_name.name);
2241 if (iocb->ki_left == 0)
2244 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2246 appending = file->f_flags & O_APPEND ? 1 : 0;
2247 direct_io = file->f_flags & O_DIRECT ? 1 : 0;
2249 mutex_lock(&inode->i_mutex);
2251 ocfs2_iocb_clear_sem_locked(iocb);
2254 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
2256 down_read(&inode->i_alloc_sem);
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_left, appending,
2296 &can_do_direct, &has_refcount);
2303 * We can't complete the direct I/O as requested, fall back to
2306 if (direct_io && !can_do_direct) {
2307 ocfs2_rw_unlock(inode, rw_level);
2308 up_read(&inode->i_alloc_sem);
2318 * To later detect whether a journal commit for sync writes is
2319 * necessary, we sample i_size, and cluster count here.
2321 old_size = i_size_read(inode);
2322 old_clusters = OCFS2_I(inode)->ip_clusters;
2324 /* communicate with ocfs2_dio_end_io */
2325 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2327 ret = generic_segment_checks(iov, &nr_segs, &ocount,
2333 ret = generic_write_checks(file, ppos, &count,
2334 S_ISBLK(inode->i_mode));
2339 written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
2340 ppos, count, ocount);
2346 current->backing_dev_info = file->f_mapping->backing_dev_info;
2347 written = generic_file_buffered_write(iocb, iov, nr_segs, *ppos,
2349 current->backing_dev_info = NULL;
2353 /* buffered aio wouldn't have proper lock coverage today */
2354 BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
2356 if (((file->f_flags & O_DSYNC) && !direct_io) || IS_SYNC(inode) ||
2357 ((file->f_flags & O_DIRECT) && !direct_io)) {
2358 ret = filemap_fdatawrite_range(file->f_mapping, pos,
2363 if (!ret && ((old_size != i_size_read(inode)) ||
2364 (old_clusters != OCFS2_I(inode)->ip_clusters) ||
2366 ret = jbd2_journal_force_commit(osb->journal->j_journal);
2372 ret = filemap_fdatawait_range(file->f_mapping, pos,
2377 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2378 * function pointer which is called when o_direct io completes so that
2379 * it can unlock our rw lock. (it's the clustered equivalent of
2380 * i_alloc_sem; protects truncate from racing with pending ios).
2381 * Unfortunately there are error cases which call end_io and others
2382 * that don't. so we don't have to unlock the rw_lock if either an
2383 * async dio is going to do it in the future or an end_io after an
2384 * error has already done it.
2386 if ((ret == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2393 ocfs2_rw_unlock(inode, rw_level);
2396 if (have_alloc_sem) {
2397 up_read(&inode->i_alloc_sem);
2398 ocfs2_iocb_clear_sem_locked(iocb);
2401 mutex_unlock(&inode->i_mutex);
2409 static int ocfs2_splice_to_file(struct pipe_inode_info *pipe,
2411 struct splice_desc *sd)
2415 ret = ocfs2_prepare_inode_for_write(out, &sd->pos,
2416 sd->total_len, 0, NULL, NULL);
2422 return splice_from_pipe_feed(pipe, sd, pipe_to_file);
2425 static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
2432 struct address_space *mapping = out->f_mapping;
2433 struct inode *inode = mapping->host;
2434 struct splice_desc sd = {
2441 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe,
2443 out->f_path.dentry->d_name.len,
2444 out->f_path.dentry->d_name.name);
2447 mutex_lock_nested(&pipe->inode->i_mutex, I_MUTEX_PARENT);
2449 splice_from_pipe_begin(&sd);
2451 ret = splice_from_pipe_next(pipe, &sd);
2455 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2456 ret = ocfs2_rw_lock(inode, 1);
2460 ret = ocfs2_splice_to_file(pipe, out, &sd);
2461 ocfs2_rw_unlock(inode, 1);
2463 mutex_unlock(&inode->i_mutex);
2465 splice_from_pipe_end(pipe, &sd);
2468 mutex_unlock(&pipe->inode->i_mutex);
2471 ret = sd.num_spliced;
2474 unsigned long nr_pages;
2477 nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
2479 err = generic_write_sync(out, *ppos, ret);
2485 balance_dirty_pages_ratelimited_nr(mapping, nr_pages);
2492 static ssize_t ocfs2_file_splice_read(struct file *in,
2494 struct pipe_inode_info *pipe,
2498 int ret = 0, lock_level = 0;
2499 struct inode *inode = in->f_path.dentry->d_inode;
2501 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe,
2503 in->f_path.dentry->d_name.len,
2504 in->f_path.dentry->d_name.name);
2507 * See the comment in ocfs2_file_aio_read()
2509 ret = ocfs2_inode_lock_atime(inode, in->f_vfsmnt, &lock_level);
2514 ocfs2_inode_unlock(inode, lock_level);
2516 ret = generic_file_splice_read(in, ppos, pipe, len, flags);
2523 static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
2524 const struct iovec *iov,
2525 unsigned long nr_segs,
2528 int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
2529 struct file *filp = iocb->ki_filp;
2530 struct inode *inode = filp->f_path.dentry->d_inode;
2532 mlog_entry("(0x%p, %u, '%.*s')\n", filp,
2533 (unsigned int)nr_segs,
2534 filp->f_path.dentry->d_name.len,
2535 filp->f_path.dentry->d_name.name);
2543 ocfs2_iocb_clear_sem_locked(iocb);
2546 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
2547 * need locks to protect pending reads from racing with truncate.
2549 if (filp->f_flags & O_DIRECT) {
2550 down_read(&inode->i_alloc_sem);
2552 ocfs2_iocb_set_sem_locked(iocb);
2554 ret = ocfs2_rw_lock(inode, 0);
2560 /* communicate with ocfs2_dio_end_io */
2561 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2565 * We're fine letting folks race truncates and extending
2566 * writes with read across the cluster, just like they can
2567 * locally. Hence no rw_lock during read.
2569 * Take and drop the meta data lock to update inode fields
2570 * like i_size. This allows the checks down below
2571 * generic_file_aio_read() a chance of actually working.
2573 ret = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level);
2578 ocfs2_inode_unlock(inode, lock_level);
2580 ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
2582 mlog(0, "generic_file_aio_read returned -EINVAL\n");
2584 /* buffered aio wouldn't have proper lock coverage today */
2585 BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
2587 /* see ocfs2_file_aio_write */
2588 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2594 if (have_alloc_sem) {
2595 up_read(&inode->i_alloc_sem);
2596 ocfs2_iocb_clear_sem_locked(iocb);
2599 ocfs2_rw_unlock(inode, rw_level);
2605 const struct inode_operations ocfs2_file_iops = {
2606 .setattr = ocfs2_setattr,
2607 .getattr = ocfs2_getattr,
2608 .permission = ocfs2_permission,
2609 .setxattr = generic_setxattr,
2610 .getxattr = generic_getxattr,
2611 .listxattr = ocfs2_listxattr,
2612 .removexattr = generic_removexattr,
2613 .fiemap = ocfs2_fiemap,
2616 const struct inode_operations ocfs2_special_file_iops = {
2617 .setattr = ocfs2_setattr,
2618 .getattr = ocfs2_getattr,
2619 .permission = ocfs2_permission,
2623 * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2624 * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2626 const struct file_operations ocfs2_fops = {
2627 .llseek = generic_file_llseek,
2628 .read = do_sync_read,
2629 .write = do_sync_write,
2631 .fsync = ocfs2_sync_file,
2632 .release = ocfs2_file_release,
2633 .open = ocfs2_file_open,
2634 .aio_read = ocfs2_file_aio_read,
2635 .aio_write = ocfs2_file_aio_write,
2636 .unlocked_ioctl = ocfs2_ioctl,
2637 #ifdef CONFIG_COMPAT
2638 .compat_ioctl = ocfs2_compat_ioctl,
2641 .flock = ocfs2_flock,
2642 .splice_read = ocfs2_file_splice_read,
2643 .splice_write = ocfs2_file_splice_write,
2644 .fallocate = ocfs2_fallocate,
2647 const struct file_operations ocfs2_dops = {
2648 .llseek = generic_file_llseek,
2649 .read = generic_read_dir,
2650 .readdir = ocfs2_readdir,
2651 .fsync = ocfs2_sync_file,
2652 .release = ocfs2_dir_release,
2653 .open = ocfs2_dir_open,
2654 .unlocked_ioctl = ocfs2_ioctl,
2655 #ifdef CONFIG_COMPAT
2656 .compat_ioctl = ocfs2_compat_ioctl,
2659 .flock = ocfs2_flock,
2663 * POSIX-lockless variants of our file_operations.
2665 * These will be used if the underlying cluster stack does not support
2666 * posix file locking, if the user passes the "localflocks" mount
2667 * option, or if we have a local-only fs.
2669 * ocfs2_flock is in here because all stacks handle UNIX file locks,
2670 * so we still want it in the case of no stack support for
2671 * plocks. Internally, it will do the right thing when asked to ignore
2674 const struct file_operations ocfs2_fops_no_plocks = {
2675 .llseek = generic_file_llseek,
2676 .read = do_sync_read,
2677 .write = do_sync_write,
2679 .fsync = ocfs2_sync_file,
2680 .release = ocfs2_file_release,
2681 .open = ocfs2_file_open,
2682 .aio_read = ocfs2_file_aio_read,
2683 .aio_write = ocfs2_file_aio_write,
2684 .unlocked_ioctl = ocfs2_ioctl,
2685 #ifdef CONFIG_COMPAT
2686 .compat_ioctl = ocfs2_compat_ioctl,
2688 .flock = ocfs2_flock,
2689 .splice_read = ocfs2_file_splice_read,
2690 .splice_write = ocfs2_file_splice_write,
2693 const struct file_operations ocfs2_dops_no_plocks = {
2694 .llseek = generic_file_llseek,
2695 .read = generic_read_dir,
2696 .readdir = ocfs2_readdir,
2697 .fsync = ocfs2_sync_file,
2698 .release = ocfs2_dir_release,
2699 .open = ocfs2_dir_open,
2700 .unlocked_ioctl = ocfs2_ioctl,
2701 #ifdef CONFIG_COMPAT
2702 .compat_ioctl = ocfs2_compat_ioctl,
2704 .flock = ocfs2_flock,