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, int datasync)
178 struct inode *inode = file->f_mapping->host;
179 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
181 trace_ocfs2_sync_file(inode, file, file->f_path.dentry,
182 OCFS2_I(inode)->ip_blkno,
183 file->f_path.dentry->d_name.len,
184 file->f_path.dentry->d_name.name,
185 (unsigned long long)datasync);
187 if (datasync && !(inode->i_state & I_DIRTY_DATASYNC)) {
189 * We still have to flush drive's caches to get data to the
192 if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
193 blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
197 journal = osb->journal->j_journal;
198 err = jbd2_journal_force_commit(journal);
204 return (err < 0) ? -EIO : 0;
207 int ocfs2_should_update_atime(struct inode *inode,
208 struct vfsmount *vfsmnt)
211 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
213 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
216 if ((inode->i_flags & S_NOATIME) ||
217 ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
221 * We can be called with no vfsmnt structure - NFSD will
224 * Note that our action here is different than touch_atime() -
225 * if we can't tell whether this is a noatime mount, then we
226 * don't know whether to trust the value of s_atime_quantum.
231 if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
232 ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
235 if (vfsmnt->mnt_flags & MNT_RELATIME) {
236 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
237 (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
244 if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
250 int ocfs2_update_inode_atime(struct inode *inode,
251 struct buffer_head *bh)
254 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
256 struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
258 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
259 if (IS_ERR(handle)) {
260 ret = PTR_ERR(handle);
265 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
266 OCFS2_JOURNAL_ACCESS_WRITE);
273 * Don't use ocfs2_mark_inode_dirty() here as we don't always
274 * have i_mutex to guard against concurrent changes to other
277 inode->i_atime = CURRENT_TIME;
278 di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
279 di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
280 ocfs2_journal_dirty(handle, bh);
283 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
288 static int ocfs2_set_inode_size(handle_t *handle,
290 struct buffer_head *fe_bh,
295 i_size_write(inode, new_i_size);
296 inode->i_blocks = ocfs2_inode_sector_count(inode);
297 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
299 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
309 int ocfs2_simple_size_update(struct inode *inode,
310 struct buffer_head *di_bh,
314 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
315 handle_t *handle = NULL;
317 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
318 if (IS_ERR(handle)) {
319 ret = PTR_ERR(handle);
324 ret = ocfs2_set_inode_size(handle, inode, di_bh,
329 ocfs2_commit_trans(osb, handle);
334 static int ocfs2_cow_file_pos(struct inode *inode,
335 struct buffer_head *fe_bh,
339 u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
340 unsigned int num_clusters = 0;
341 unsigned int ext_flags = 0;
344 * If the new offset is aligned to the range of the cluster, there is
345 * no space for ocfs2_zero_range_for_truncate to fill, so no need to
348 if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
351 status = ocfs2_get_clusters(inode, cpos, &phys,
352 &num_clusters, &ext_flags);
358 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
361 return ocfs2_refcount_cow(inode, NULL, fe_bh, cpos, 1, cpos+1);
367 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
369 struct buffer_head *fe_bh,
374 struct ocfs2_dinode *di;
378 * We need to CoW the cluster contains the offset if it is reflinked
379 * since we will call ocfs2_zero_range_for_truncate later which will
380 * write "0" from offset to the end of the cluster.
382 status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
388 /* TODO: This needs to actually orphan the inode in this
391 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
392 if (IS_ERR(handle)) {
393 status = PTR_ERR(handle);
398 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
399 OCFS2_JOURNAL_ACCESS_WRITE);
406 * Do this before setting i_size.
408 cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
409 status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
416 i_size_write(inode, new_i_size);
417 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
419 di = (struct ocfs2_dinode *) fe_bh->b_data;
420 di->i_size = cpu_to_le64(new_i_size);
421 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
422 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
424 ocfs2_journal_dirty(handle, fe_bh);
427 ocfs2_commit_trans(osb, handle);
432 static int ocfs2_truncate_file(struct inode *inode,
433 struct buffer_head *di_bh,
437 struct ocfs2_dinode *fe = NULL;
438 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
440 /* We trust di_bh because it comes from ocfs2_inode_lock(), which
441 * already validated it */
442 fe = (struct ocfs2_dinode *) di_bh->b_data;
444 trace_ocfs2_truncate_file((unsigned long long)OCFS2_I(inode)->ip_blkno,
445 (unsigned long long)le64_to_cpu(fe->i_size),
446 (unsigned long long)new_i_size);
448 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
449 "Inode %llu, inode i_size = %lld != di "
450 "i_size = %llu, i_flags = 0x%x\n",
451 (unsigned long long)OCFS2_I(inode)->ip_blkno,
453 (unsigned long long)le64_to_cpu(fe->i_size),
454 le32_to_cpu(fe->i_flags));
456 if (new_i_size > le64_to_cpu(fe->i_size)) {
457 trace_ocfs2_truncate_file_error(
458 (unsigned long long)le64_to_cpu(fe->i_size),
459 (unsigned long long)new_i_size);
465 /* lets handle the simple truncate cases before doing any more
466 * cluster locking. */
467 if (new_i_size == le64_to_cpu(fe->i_size))
470 down_write(&OCFS2_I(inode)->ip_alloc_sem);
472 ocfs2_resv_discard(&osb->osb_la_resmap,
473 &OCFS2_I(inode)->ip_la_data_resv);
476 * The inode lock forced other nodes to sync and drop their
477 * pages, which (correctly) happens even if we have a truncate
478 * without allocation change - ocfs2 cluster sizes can be much
479 * greater than page size, so we have to truncate them
482 unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
483 truncate_inode_pages(inode->i_mapping, new_i_size);
485 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
486 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
487 i_size_read(inode), 1);
491 goto bail_unlock_sem;
494 /* alright, we're going to need to do a full blown alloc size
495 * change. Orphan the inode so that recovery can complete the
496 * truncate if necessary. This does the task of marking
498 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
501 goto bail_unlock_sem;
504 status = ocfs2_commit_truncate(osb, inode, di_bh);
507 goto bail_unlock_sem;
510 /* TODO: orphan dir cleanup here. */
512 up_write(&OCFS2_I(inode)->ip_alloc_sem);
515 if (!status && OCFS2_I(inode)->ip_clusters == 0)
516 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
522 * extend file allocation only here.
523 * we'll update all the disk stuff, and oip->alloc_size
525 * expect stuff to be locked, a transaction started and enough data /
526 * metadata reservations in the contexts.
528 * Will return -EAGAIN, and a reason if a restart is needed.
529 * If passed in, *reason will always be set, even in error.
531 int ocfs2_add_inode_data(struct ocfs2_super *osb,
536 struct buffer_head *fe_bh,
538 struct ocfs2_alloc_context *data_ac,
539 struct ocfs2_alloc_context *meta_ac,
540 enum ocfs2_alloc_restarted *reason_ret)
543 struct ocfs2_extent_tree et;
545 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
546 ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
547 clusters_to_add, mark_unwritten,
548 data_ac, meta_ac, reason_ret);
553 static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
554 u32 clusters_to_add, int mark_unwritten)
557 int restart_func = 0;
560 struct buffer_head *bh = NULL;
561 struct ocfs2_dinode *fe = NULL;
562 handle_t *handle = NULL;
563 struct ocfs2_alloc_context *data_ac = NULL;
564 struct ocfs2_alloc_context *meta_ac = NULL;
565 enum ocfs2_alloc_restarted why;
566 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
567 struct ocfs2_extent_tree et;
571 * This function only exists for file systems which don't
574 BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
576 status = ocfs2_read_inode_block(inode, &bh);
581 fe = (struct ocfs2_dinode *) bh->b_data;
584 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
586 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
587 status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
594 credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list,
596 handle = ocfs2_start_trans(osb, credits);
597 if (IS_ERR(handle)) {
598 status = PTR_ERR(handle);
604 restarted_transaction:
605 trace_ocfs2_extend_allocation(
606 (unsigned long long)OCFS2_I(inode)->ip_blkno,
607 (unsigned long long)i_size_read(inode),
608 le32_to_cpu(fe->i_clusters), clusters_to_add,
611 status = dquot_alloc_space_nodirty(inode,
612 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
617 /* reserve a write to the file entry early on - that we if we
618 * run out of credits in the allocation path, we can still
620 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
621 OCFS2_JOURNAL_ACCESS_WRITE);
627 prev_clusters = OCFS2_I(inode)->ip_clusters;
629 status = ocfs2_add_inode_data(osb,
639 if ((status < 0) && (status != -EAGAIN)) {
640 if (status != -ENOSPC)
645 ocfs2_journal_dirty(handle, bh);
647 spin_lock(&OCFS2_I(inode)->ip_lock);
648 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
649 spin_unlock(&OCFS2_I(inode)->ip_lock);
650 /* Release unused quota reservation */
651 dquot_free_space(inode,
652 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
655 if (why != RESTART_NONE && clusters_to_add) {
656 if (why == RESTART_META) {
660 BUG_ON(why != RESTART_TRANS);
662 /* TODO: This can be more intelligent. */
663 credits = ocfs2_calc_extend_credits(osb->sb,
666 status = ocfs2_extend_trans(handle, credits);
668 /* handle still has to be committed at
674 goto restarted_transaction;
678 trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno,
679 le32_to_cpu(fe->i_clusters),
680 (unsigned long long)le64_to_cpu(fe->i_size),
681 OCFS2_I(inode)->ip_clusters,
682 (unsigned long long)i_size_read(inode));
685 if (status < 0 && did_quota)
686 dquot_free_space(inode,
687 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
689 ocfs2_commit_trans(osb, handle);
693 ocfs2_free_alloc_context(data_ac);
697 ocfs2_free_alloc_context(meta_ac);
700 if ((!status) && restart_func) {
711 * While a write will already be ordering the data, a truncate will not.
712 * Thus, we need to explicitly order the zeroed pages.
714 static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode)
716 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
717 handle_t *handle = NULL;
720 if (!ocfs2_should_order_data(inode))
723 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
724 if (IS_ERR(handle)) {
730 ret = ocfs2_jbd2_file_inode(handle, inode);
737 ocfs2_commit_trans(osb, handle);
738 handle = ERR_PTR(ret);
743 /* Some parts of this taken from generic_cont_expand, which turned out
744 * to be too fragile to do exactly what we need without us having to
745 * worry about recursive locking in ->write_begin() and ->write_end(). */
746 static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
749 struct address_space *mapping = inode->i_mapping;
751 unsigned long index = abs_from >> PAGE_CACHE_SHIFT;
752 handle_t *handle = NULL;
754 unsigned zero_from, zero_to, block_start, block_end;
756 BUG_ON(abs_from >= abs_to);
757 BUG_ON(abs_to > (((u64)index + 1) << PAGE_CACHE_SHIFT));
758 BUG_ON(abs_from & (inode->i_blkbits - 1));
760 page = find_or_create_page(mapping, index, GFP_NOFS);
767 /* Get the offsets within the page that we want to zero */
768 zero_from = abs_from & (PAGE_CACHE_SIZE - 1);
769 zero_to = abs_to & (PAGE_CACHE_SIZE - 1);
771 zero_to = PAGE_CACHE_SIZE;
773 trace_ocfs2_write_zero_page(
774 (unsigned long long)OCFS2_I(inode)->ip_blkno,
775 (unsigned long long)abs_from,
776 (unsigned long long)abs_to,
777 index, zero_from, zero_to);
779 /* We know that zero_from is block aligned */
780 for (block_start = zero_from; block_start < zero_to;
781 block_start = block_end) {
782 block_end = block_start + (1 << inode->i_blkbits);
785 * block_start is block-aligned. Bump it by one to force
786 * __block_write_begin and block_commit_write to zero the
789 ret = __block_write_begin(page, block_start + 1, 0,
797 handle = ocfs2_zero_start_ordered_transaction(inode);
798 if (IS_ERR(handle)) {
799 ret = PTR_ERR(handle);
805 /* must not update i_size! */
806 ret = block_commit_write(page, block_start + 1,
815 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
819 page_cache_release(page);
825 * Find the next range to zero. We do this in terms of bytes because
826 * that's what ocfs2_zero_extend() wants, and it is dealing with the
827 * pagecache. We may return multiple extents.
829 * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
830 * needs to be zeroed. range_start and range_end return the next zeroing
831 * range. A subsequent call should pass the previous range_end as its
832 * zero_start. If range_end is 0, there's nothing to do.
834 * Unwritten extents are skipped over. Refcounted extents are CoWd.
836 static int ocfs2_zero_extend_get_range(struct inode *inode,
837 struct buffer_head *di_bh,
838 u64 zero_start, u64 zero_end,
839 u64 *range_start, u64 *range_end)
841 int rc = 0, needs_cow = 0;
842 u32 p_cpos, zero_clusters = 0;
844 zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
845 u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
846 unsigned int num_clusters = 0;
847 unsigned int ext_flags = 0;
849 while (zero_cpos < last_cpos) {
850 rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
851 &num_clusters, &ext_flags);
857 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
858 zero_clusters = num_clusters;
859 if (ext_flags & OCFS2_EXT_REFCOUNTED)
864 zero_cpos += num_clusters;
866 if (!zero_clusters) {
871 while ((zero_cpos + zero_clusters) < last_cpos) {
872 rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
873 &p_cpos, &num_clusters,
880 if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
882 if (ext_flags & OCFS2_EXT_REFCOUNTED)
884 zero_clusters += num_clusters;
886 if ((zero_cpos + zero_clusters) > last_cpos)
887 zero_clusters = last_cpos - zero_cpos;
890 rc = ocfs2_refcount_cow(inode, NULL, di_bh, zero_cpos,
891 zero_clusters, UINT_MAX);
898 *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
899 *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
900 zero_cpos + zero_clusters);
907 * Zero one range returned from ocfs2_zero_extend_get_range(). The caller
908 * has made sure that the entire range needs zeroing.
910 static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
915 u64 zero_pos = range_start;
917 trace_ocfs2_zero_extend_range(
918 (unsigned long long)OCFS2_I(inode)->ip_blkno,
919 (unsigned long long)range_start,
920 (unsigned long long)range_end);
921 BUG_ON(range_start >= range_end);
923 while (zero_pos < range_end) {
924 next_pos = (zero_pos & PAGE_CACHE_MASK) + PAGE_CACHE_SIZE;
925 if (next_pos > range_end)
926 next_pos = range_end;
927 rc = ocfs2_write_zero_page(inode, zero_pos, next_pos);
935 * Very large extends have the potential to lock up
936 * the cpu for extended periods of time.
944 int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
948 u64 zero_start, range_start = 0, range_end = 0;
949 struct super_block *sb = inode->i_sb;
951 zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
952 trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
953 (unsigned long long)zero_start,
954 (unsigned long long)i_size_read(inode));
955 while (zero_start < zero_to_size) {
956 ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
967 if (range_start < zero_start)
968 range_start = zero_start;
969 if (range_end > zero_to_size)
970 range_end = zero_to_size;
972 ret = ocfs2_zero_extend_range(inode, range_start,
978 zero_start = range_end;
984 int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
985 u64 new_i_size, u64 zero_to)
989 struct ocfs2_inode_info *oi = OCFS2_I(inode);
992 * Only quota files call this without a bh, and they can't be
995 BUG_ON(!di_bh && (oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
996 BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
998 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
999 if (clusters_to_add < oi->ip_clusters)
1000 clusters_to_add = 0;
1002 clusters_to_add -= oi->ip_clusters;
1004 if (clusters_to_add) {
1005 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
1006 clusters_to_add, 0);
1014 * Call this even if we don't add any clusters to the tree. We
1015 * still need to zero the area between the old i_size and the
1018 ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1026 static int ocfs2_extend_file(struct inode *inode,
1027 struct buffer_head *di_bh,
1031 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1035 /* setattr sometimes calls us like this. */
1036 if (new_i_size == 0)
1039 if (i_size_read(inode) == new_i_size)
1041 BUG_ON(new_i_size < i_size_read(inode));
1044 * The alloc sem blocks people in read/write from reading our
1045 * allocation until we're done changing it. We depend on
1046 * i_mutex to block other extend/truncate calls while we're
1047 * here. We even have to hold it for sparse files because there
1048 * might be some tail zeroing.
1050 down_write(&oi->ip_alloc_sem);
1052 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1054 * We can optimize small extends by keeping the inodes
1057 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1058 up_write(&oi->ip_alloc_sem);
1059 goto out_update_size;
1062 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1064 up_write(&oi->ip_alloc_sem);
1070 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1071 ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1073 ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1076 up_write(&oi->ip_alloc_sem);
1084 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1092 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1094 int status = 0, size_change;
1095 struct inode *inode = dentry->d_inode;
1096 struct super_block *sb = inode->i_sb;
1097 struct ocfs2_super *osb = OCFS2_SB(sb);
1098 struct buffer_head *bh = NULL;
1099 handle_t *handle = NULL;
1100 struct dquot *transfer_to[MAXQUOTAS] = { };
1103 trace_ocfs2_setattr(inode, dentry,
1104 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1105 dentry->d_name.len, dentry->d_name.name,
1106 attr->ia_valid, attr->ia_mode,
1107 attr->ia_uid, attr->ia_gid);
1109 /* ensuring we don't even attempt to truncate a symlink */
1110 if (S_ISLNK(inode->i_mode))
1111 attr->ia_valid &= ~ATTR_SIZE;
1113 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1114 | ATTR_GID | ATTR_UID | ATTR_MODE)
1115 if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
1118 status = inode_change_ok(inode, attr);
1122 if (is_quota_modification(inode, attr))
1123 dquot_initialize(inode);
1124 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1126 status = ocfs2_rw_lock(inode, 1);
1133 status = ocfs2_inode_lock(inode, &bh, 1);
1135 if (status != -ENOENT)
1137 goto bail_unlock_rw;
1140 if (size_change && attr->ia_size != i_size_read(inode)) {
1141 status = inode_newsize_ok(inode, attr->ia_size);
1145 inode_dio_wait(inode);
1147 if (i_size_read(inode) > attr->ia_size) {
1148 if (ocfs2_should_order_data(inode)) {
1149 status = ocfs2_begin_ordered_truncate(inode,
1154 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1156 status = ocfs2_extend_file(inode, bh, attr->ia_size);
1158 if (status != -ENOSPC)
1165 if ((attr->ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
1166 (attr->ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
1168 * Gather pointers to quota structures so that allocation /
1169 * freeing of quota structures happens here and not inside
1170 * dquot_transfer() where we have problems with lock ordering
1172 if (attr->ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid
1173 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1174 OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1175 transfer_to[USRQUOTA] = dqget(sb, attr->ia_uid,
1177 if (!transfer_to[USRQUOTA]) {
1182 if (attr->ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid
1183 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1184 OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1185 transfer_to[GRPQUOTA] = dqget(sb, attr->ia_gid,
1187 if (!transfer_to[GRPQUOTA]) {
1192 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1193 2 * ocfs2_quota_trans_credits(sb));
1194 if (IS_ERR(handle)) {
1195 status = PTR_ERR(handle);
1199 status = __dquot_transfer(inode, transfer_to);
1203 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1204 if (IS_ERR(handle)) {
1205 status = PTR_ERR(handle);
1212 * This will intentionally not wind up calling truncate_setsize(),
1213 * since all the work for a size change has been done above.
1214 * Otherwise, we could get into problems with truncate as
1215 * ip_alloc_sem is used there to protect against i_size
1218 * XXX: this means the conditional below can probably be removed.
1220 if ((attr->ia_valid & ATTR_SIZE) &&
1221 attr->ia_size != i_size_read(inode)) {
1222 status = vmtruncate(inode, attr->ia_size);
1229 setattr_copy(inode, attr);
1230 mark_inode_dirty(inode);
1232 status = ocfs2_mark_inode_dirty(handle, inode, bh);
1237 ocfs2_commit_trans(osb, handle);
1239 ocfs2_inode_unlock(inode, 1);
1242 ocfs2_rw_unlock(inode, 1);
1246 /* Release quota pointers in case we acquired them */
1247 for (qtype = 0; qtype < MAXQUOTAS; qtype++)
1248 dqput(transfer_to[qtype]);
1250 if (!status && attr->ia_valid & ATTR_MODE) {
1251 status = ocfs2_acl_chmod(inode);
1259 int ocfs2_getattr(struct vfsmount *mnt,
1260 struct dentry *dentry,
1263 struct inode *inode = dentry->d_inode;
1264 struct super_block *sb = dentry->d_inode->i_sb;
1265 struct ocfs2_super *osb = sb->s_fs_info;
1268 err = ocfs2_inode_revalidate(dentry);
1275 generic_fillattr(inode, stat);
1277 /* We set the blksize from the cluster size for performance */
1278 stat->blksize = osb->s_clustersize;
1284 int ocfs2_permission(struct inode *inode, int mask)
1288 if (mask & MAY_NOT_BLOCK)
1291 ret = ocfs2_inode_lock(inode, NULL, 0);
1298 ret = generic_permission(inode, mask);
1300 ocfs2_inode_unlock(inode, 0);
1305 static int __ocfs2_write_remove_suid(struct inode *inode,
1306 struct buffer_head *bh)
1310 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1311 struct ocfs2_dinode *di;
1313 trace_ocfs2_write_remove_suid(
1314 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1317 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1318 if (IS_ERR(handle)) {
1319 ret = PTR_ERR(handle);
1324 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1325 OCFS2_JOURNAL_ACCESS_WRITE);
1331 inode->i_mode &= ~S_ISUID;
1332 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1333 inode->i_mode &= ~S_ISGID;
1335 di = (struct ocfs2_dinode *) bh->b_data;
1336 di->i_mode = cpu_to_le16(inode->i_mode);
1338 ocfs2_journal_dirty(handle, bh);
1341 ocfs2_commit_trans(osb, handle);
1347 * Will look for holes and unwritten extents in the range starting at
1348 * pos for count bytes (inclusive).
1350 static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1354 unsigned int extent_flags;
1355 u32 cpos, clusters, extent_len, phys_cpos;
1356 struct super_block *sb = inode->i_sb;
1358 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1359 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1362 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1369 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
1374 if (extent_len > clusters)
1375 extent_len = clusters;
1377 clusters -= extent_len;
1384 static int ocfs2_write_remove_suid(struct inode *inode)
1387 struct buffer_head *bh = NULL;
1389 ret = ocfs2_read_inode_block(inode, &bh);
1395 ret = __ocfs2_write_remove_suid(inode, bh);
1402 * Allocate enough extents to cover the region starting at byte offset
1403 * start for len bytes. Existing extents are skipped, any extents
1404 * added are marked as "unwritten".
1406 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1410 u32 cpos, phys_cpos, clusters, alloc_size;
1411 u64 end = start + len;
1412 struct buffer_head *di_bh = NULL;
1414 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1415 ret = ocfs2_read_inode_block(inode, &di_bh);
1422 * Nothing to do if the requested reservation range
1423 * fits within the inode.
1425 if (ocfs2_size_fits_inline_data(di_bh, end))
1428 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1436 * We consider both start and len to be inclusive.
1438 cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1439 clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1443 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1451 * Hole or existing extent len can be arbitrary, so
1452 * cap it to our own allocation request.
1454 if (alloc_size > clusters)
1455 alloc_size = clusters;
1459 * We already have an allocation at this
1460 * region so we can safely skip it.
1465 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1474 clusters -= alloc_size;
1485 * Truncate a byte range, avoiding pages within partial clusters. This
1486 * preserves those pages for the zeroing code to write to.
1488 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1491 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1493 struct address_space *mapping = inode->i_mapping;
1495 start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1496 end = byte_start + byte_len;
1497 end = end & ~(osb->s_clustersize - 1);
1500 unmap_mapping_range(mapping, start, end - start, 0);
1501 truncate_inode_pages_range(mapping, start, end - 1);
1505 static int ocfs2_zero_partial_clusters(struct inode *inode,
1509 u64 tmpend, end = start + len;
1510 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1511 unsigned int csize = osb->s_clustersize;
1515 * The "start" and "end" values are NOT necessarily part of
1516 * the range whose allocation is being deleted. Rather, this
1517 * is what the user passed in with the request. We must zero
1518 * partial clusters here. There's no need to worry about
1519 * physical allocation - the zeroing code knows to skip holes.
1521 trace_ocfs2_zero_partial_clusters(
1522 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1523 (unsigned long long)start, (unsigned long long)end);
1526 * If both edges are on a cluster boundary then there's no
1527 * zeroing required as the region is part of the allocation to
1530 if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1533 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1534 if (IS_ERR(handle)) {
1535 ret = PTR_ERR(handle);
1541 * We want to get the byte offset of the end of the 1st cluster.
1543 tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1547 trace_ocfs2_zero_partial_clusters_range1((unsigned long long)start,
1548 (unsigned long long)tmpend);
1550 ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1556 * This may make start and end equal, but the zeroing
1557 * code will skip any work in that case so there's no
1558 * need to catch it up here.
1560 start = end & ~(osb->s_clustersize - 1);
1562 trace_ocfs2_zero_partial_clusters_range2(
1563 (unsigned long long)start, (unsigned long long)end);
1565 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1570 ocfs2_commit_trans(osb, handle);
1575 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1578 struct ocfs2_extent_rec *rec = NULL;
1580 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1582 rec = &el->l_recs[i];
1584 if (le32_to_cpu(rec->e_cpos) < pos)
1592 * Helper to calculate the punching pos and length in one run, we handle the
1593 * following three cases in order:
1595 * - remove the entire record
1596 * - remove a partial record
1597 * - no record needs to be removed (hole-punching completed)
1599 static void ocfs2_calc_trunc_pos(struct inode *inode,
1600 struct ocfs2_extent_list *el,
1601 struct ocfs2_extent_rec *rec,
1602 u32 trunc_start, u32 *trunc_cpos,
1603 u32 *trunc_len, u32 *trunc_end,
1604 u64 *blkno, int *done)
1609 range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1611 if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1613 * remove an entire extent record.
1615 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1617 * Skip holes if any.
1619 if (range < *trunc_end)
1621 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1622 *blkno = le64_to_cpu(rec->e_blkno);
1623 *trunc_end = le32_to_cpu(rec->e_cpos);
1624 } else if (range > trunc_start) {
1626 * remove a partial extent record, which means we're
1627 * removing the last extent record.
1629 *trunc_cpos = trunc_start;
1633 if (range < *trunc_end)
1635 *trunc_len = *trunc_end - trunc_start;
1636 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1637 *blkno = le64_to_cpu(rec->e_blkno) +
1638 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1639 *trunc_end = trunc_start;
1642 * It may have two following possibilities:
1644 * - last record has been removed
1645 * - trunc_start was within a hole
1647 * both two cases mean the completion of hole punching.
1655 static int ocfs2_remove_inode_range(struct inode *inode,
1656 struct buffer_head *di_bh, u64 byte_start,
1659 int ret = 0, flags = 0, done = 0, i;
1660 u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1662 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1663 struct ocfs2_cached_dealloc_ctxt dealloc;
1664 struct address_space *mapping = inode->i_mapping;
1665 struct ocfs2_extent_tree et;
1666 struct ocfs2_path *path = NULL;
1667 struct ocfs2_extent_list *el = NULL;
1668 struct ocfs2_extent_rec *rec = NULL;
1669 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1670 u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1672 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1673 ocfs2_init_dealloc_ctxt(&dealloc);
1675 trace_ocfs2_remove_inode_range(
1676 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1677 (unsigned long long)byte_start,
1678 (unsigned long long)byte_len);
1683 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1684 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1685 byte_start + byte_len, 0);
1691 * There's no need to get fancy with the page cache
1692 * truncate of an inline-data inode. We're talking
1693 * about less than a page here, which will be cached
1694 * in the dinode buffer anyway.
1696 unmap_mapping_range(mapping, 0, 0, 0);
1697 truncate_inode_pages(mapping, 0);
1702 * For reflinks, we may need to CoW 2 clusters which might be
1703 * partially zero'd later, if hole's start and end offset were
1704 * within one cluster(means is not exactly aligned to clustersize).
1707 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
1709 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1715 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1722 trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1723 trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1724 cluster_in_el = trunc_end;
1726 ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1732 path = ocfs2_new_path_from_et(&et);
1739 while (trunc_end > trunc_start) {
1741 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1748 el = path_leaf_el(path);
1750 i = ocfs2_find_rec(el, trunc_end);
1752 * Need to go to previous extent block.
1755 if (path->p_tree_depth == 0)
1758 ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1767 * We've reached the leftmost extent block,
1768 * it's safe to leave.
1770 if (cluster_in_el == 0)
1774 * The 'pos' searched for previous extent block is
1775 * always one cluster less than actual trunc_end.
1777 trunc_end = cluster_in_el + 1;
1779 ocfs2_reinit_path(path, 1);
1784 rec = &el->l_recs[i];
1786 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1787 &trunc_len, &trunc_end, &blkno, &done);
1791 flags = rec->e_flags;
1792 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1794 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1795 phys_cpos, trunc_len, flags,
1796 &dealloc, refcount_loc);
1802 cluster_in_el = trunc_end;
1804 ocfs2_reinit_path(path, 1);
1807 ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1810 ocfs2_schedule_truncate_log_flush(osb, 1);
1811 ocfs2_run_deallocs(osb, &dealloc);
1817 * Parts of this function taken from xfs_change_file_space()
1819 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1820 loff_t f_pos, unsigned int cmd,
1821 struct ocfs2_space_resv *sr,
1827 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1828 struct buffer_head *di_bh = NULL;
1830 unsigned long long max_off = inode->i_sb->s_maxbytes;
1832 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1835 mutex_lock(&inode->i_mutex);
1838 * This prevents concurrent writes on other nodes
1840 ret = ocfs2_rw_lock(inode, 1);
1846 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1852 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1854 goto out_inode_unlock;
1857 switch (sr->l_whence) {
1858 case 0: /*SEEK_SET*/
1860 case 1: /*SEEK_CUR*/
1861 sr->l_start += f_pos;
1863 case 2: /*SEEK_END*/
1864 sr->l_start += i_size_read(inode);
1868 goto out_inode_unlock;
1872 llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1875 || sr->l_start > max_off
1876 || (sr->l_start + llen) < 0
1877 || (sr->l_start + llen) > max_off) {
1879 goto out_inode_unlock;
1881 size = sr->l_start + sr->l_len;
1883 if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) {
1884 if (sr->l_len <= 0) {
1886 goto out_inode_unlock;
1890 if (file && should_remove_suid(file->f_path.dentry)) {
1891 ret = __ocfs2_write_remove_suid(inode, di_bh);
1894 goto out_inode_unlock;
1898 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1900 case OCFS2_IOC_RESVSP:
1901 case OCFS2_IOC_RESVSP64:
1903 * This takes unsigned offsets, but the signed ones we
1904 * pass have been checked against overflow above.
1906 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1909 case OCFS2_IOC_UNRESVSP:
1910 case OCFS2_IOC_UNRESVSP64:
1911 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1917 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1920 goto out_inode_unlock;
1924 * We update c/mtime for these changes
1926 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1927 if (IS_ERR(handle)) {
1928 ret = PTR_ERR(handle);
1930 goto out_inode_unlock;
1933 if (change_size && i_size_read(inode) < size)
1934 i_size_write(inode, size);
1936 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1937 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1941 ocfs2_commit_trans(osb, handle);
1945 ocfs2_inode_unlock(inode, 1);
1947 ocfs2_rw_unlock(inode, 1);
1950 mutex_unlock(&inode->i_mutex);
1954 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1955 struct ocfs2_space_resv *sr)
1957 struct inode *inode = file->f_path.dentry->d_inode;
1958 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1960 if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1961 !ocfs2_writes_unwritten_extents(osb))
1963 else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1964 !ocfs2_sparse_alloc(osb))
1967 if (!S_ISREG(inode->i_mode))
1970 if (!(file->f_mode & FMODE_WRITE))
1973 return __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1976 static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
1979 struct inode *inode = file->f_path.dentry->d_inode;
1980 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1981 struct ocfs2_space_resv sr;
1982 int change_size = 1;
1983 int cmd = OCFS2_IOC_RESVSP64;
1985 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
1987 if (!ocfs2_writes_unwritten_extents(osb))
1990 if (mode & FALLOC_FL_KEEP_SIZE)
1993 if (mode & FALLOC_FL_PUNCH_HOLE)
1994 cmd = OCFS2_IOC_UNRESVSP64;
1997 sr.l_start = (s64)offset;
1998 sr.l_len = (s64)len;
2000 return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2004 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2008 unsigned int extent_flags;
2009 u32 cpos, clusters, extent_len, phys_cpos;
2010 struct super_block *sb = inode->i_sb;
2012 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2013 !(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) ||
2014 OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2017 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2018 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2021 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2028 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2033 if (extent_len > clusters)
2034 extent_len = clusters;
2036 clusters -= extent_len;
2043 static int ocfs2_prepare_inode_for_refcount(struct inode *inode,
2045 loff_t pos, size_t count,
2049 struct buffer_head *di_bh = NULL;
2050 u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2052 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2054 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2062 ret = ocfs2_refcount_cow(inode, file, di_bh, cpos, clusters, UINT_MAX);
2070 static int ocfs2_prepare_inode_for_write(struct file *file,
2077 int ret = 0, meta_level = 0;
2078 struct dentry *dentry = file->f_path.dentry;
2079 struct inode *inode = dentry->d_inode;
2080 loff_t saved_pos = 0, end;
2083 * We start with a read level meta lock and only jump to an ex
2084 * if we need to make modifications here.
2087 ret = ocfs2_inode_lock(inode, NULL, meta_level);
2094 /* Clear suid / sgid if necessary. We do this here
2095 * instead of later in the write path because
2096 * remove_suid() calls ->setattr without any hint that
2097 * we may have already done our cluster locking. Since
2098 * ocfs2_setattr() *must* take cluster locks to
2099 * proceeed, this will lead us to recursively lock the
2100 * inode. There's also the dinode i_size state which
2101 * can be lost via setattr during extending writes (we
2102 * set inode->i_size at the end of a write. */
2103 if (should_remove_suid(dentry)) {
2104 if (meta_level == 0) {
2105 ocfs2_inode_unlock(inode, meta_level);
2110 ret = ocfs2_write_remove_suid(inode);
2117 /* work on a copy of ppos until we're sure that we won't have
2118 * to recalculate it due to relocking. */
2120 saved_pos = i_size_read(inode);
2124 end = saved_pos + count;
2126 ret = ocfs2_check_range_for_refcount(inode, saved_pos, count);
2128 ocfs2_inode_unlock(inode, meta_level);
2131 ret = ocfs2_prepare_inode_for_refcount(inode,
2148 * Skip the O_DIRECT checks if we don't need
2151 if (!direct_io || !(*direct_io))
2155 * There's no sane way to do direct writes to an inode
2158 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2164 * Allowing concurrent direct writes means
2165 * i_size changes wouldn't be synchronized, so
2166 * one node could wind up truncating another
2169 if (end > i_size_read(inode)) {
2175 * We don't fill holes during direct io, so
2176 * check for them here. If any are found, the
2177 * caller will have to retake some cluster
2178 * locks and initiate the io as buffered.
2180 ret = ocfs2_check_range_for_holes(inode, saved_pos, count);
2193 trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
2194 saved_pos, appending, count,
2195 direct_io, has_refcount);
2197 if (meta_level >= 0)
2198 ocfs2_inode_unlock(inode, meta_level);
2204 static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
2205 const struct iovec *iov,
2206 unsigned long nr_segs,
2209 int ret, direct_io, appending, rw_level, have_alloc_sem = 0;
2210 int can_do_direct, has_refcount = 0;
2211 ssize_t written = 0;
2212 size_t ocount; /* original count */
2213 size_t count; /* after file limit checks */
2214 loff_t old_size, *ppos = &iocb->ki_pos;
2216 struct file *file = iocb->ki_filp;
2217 struct inode *inode = file->f_path.dentry->d_inode;
2218 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2219 int full_coherency = !(osb->s_mount_opt &
2220 OCFS2_MOUNT_COHERENCY_BUFFERED);
2222 trace_ocfs2_file_aio_write(inode, file, file->f_path.dentry,
2223 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2224 file->f_path.dentry->d_name.len,
2225 file->f_path.dentry->d_name.name,
2226 (unsigned int)nr_segs);
2228 if (iocb->ki_left == 0)
2231 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2233 appending = file->f_flags & O_APPEND ? 1 : 0;
2234 direct_io = file->f_flags & O_DIRECT ? 1 : 0;
2236 mutex_lock(&inode->i_mutex);
2238 ocfs2_iocb_clear_sem_locked(iocb);
2241 /* to match setattr's i_mutex -> rw_lock ordering */
2244 /* communicate with ocfs2_dio_end_io */
2245 ocfs2_iocb_set_sem_locked(iocb);
2249 * Concurrent O_DIRECT writes are allowed with
2250 * mount_option "coherency=buffered".
2252 rw_level = (!direct_io || full_coherency);
2254 ret = ocfs2_rw_lock(inode, rw_level);
2261 * O_DIRECT writes with "coherency=full" need to take EX cluster
2262 * inode_lock to guarantee coherency.
2264 if (direct_io && full_coherency) {
2266 * We need to take and drop the inode lock to force
2267 * other nodes to drop their caches. Buffered I/O
2268 * already does this in write_begin().
2270 ret = ocfs2_inode_lock(inode, NULL, 1);
2276 ocfs2_inode_unlock(inode, 1);
2279 can_do_direct = direct_io;
2280 ret = ocfs2_prepare_inode_for_write(file, ppos,
2281 iocb->ki_left, appending,
2282 &can_do_direct, &has_refcount);
2289 * We can't complete the direct I/O as requested, fall back to
2292 if (direct_io && !can_do_direct) {
2293 ocfs2_rw_unlock(inode, rw_level);
2303 * To later detect whether a journal commit for sync writes is
2304 * necessary, we sample i_size, and cluster count here.
2306 old_size = i_size_read(inode);
2307 old_clusters = OCFS2_I(inode)->ip_clusters;
2309 /* communicate with ocfs2_dio_end_io */
2310 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2312 ret = generic_segment_checks(iov, &nr_segs, &ocount,
2318 ret = generic_write_checks(file, ppos, &count,
2319 S_ISBLK(inode->i_mode));
2324 written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
2325 ppos, count, ocount);
2331 current->backing_dev_info = file->f_mapping->backing_dev_info;
2332 written = generic_file_buffered_write(iocb, iov, nr_segs, *ppos,
2334 current->backing_dev_info = NULL;
2338 /* buffered aio wouldn't have proper lock coverage today */
2339 BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
2341 if (((file->f_flags & O_DSYNC) && !direct_io) || IS_SYNC(inode) ||
2342 ((file->f_flags & O_DIRECT) && !direct_io)) {
2343 ret = filemap_fdatawrite_range(file->f_mapping, pos,
2348 if (!ret && ((old_size != i_size_read(inode)) ||
2349 (old_clusters != OCFS2_I(inode)->ip_clusters) ||
2351 ret = jbd2_journal_force_commit(osb->journal->j_journal);
2357 ret = filemap_fdatawait_range(file->f_mapping, pos,
2362 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2363 * function pointer which is called when o_direct io completes so that
2364 * it can unlock our rw lock.
2365 * Unfortunately there are error cases which call end_io and others
2366 * that don't. so we don't have to unlock the rw_lock if either an
2367 * async dio is going to do it in the future or an end_io after an
2368 * error has already done it.
2370 if ((ret == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2377 ocfs2_rw_unlock(inode, rw_level);
2381 ocfs2_iocb_clear_sem_locked(iocb);
2383 mutex_unlock(&inode->i_mutex);
2390 static int ocfs2_splice_to_file(struct pipe_inode_info *pipe,
2392 struct splice_desc *sd)
2396 ret = ocfs2_prepare_inode_for_write(out, &sd->pos,
2397 sd->total_len, 0, NULL, NULL);
2403 return splice_from_pipe_feed(pipe, sd, pipe_to_file);
2406 static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
2413 struct address_space *mapping = out->f_mapping;
2414 struct inode *inode = mapping->host;
2415 struct splice_desc sd = {
2423 trace_ocfs2_file_splice_write(inode, out, out->f_path.dentry,
2424 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2425 out->f_path.dentry->d_name.len,
2426 out->f_path.dentry->d_name.name, len);
2429 mutex_lock_nested(&pipe->inode->i_mutex, I_MUTEX_PARENT);
2431 splice_from_pipe_begin(&sd);
2433 ret = splice_from_pipe_next(pipe, &sd);
2437 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2438 ret = ocfs2_rw_lock(inode, 1);
2442 ret = ocfs2_splice_to_file(pipe, out, &sd);
2443 ocfs2_rw_unlock(inode, 1);
2445 mutex_unlock(&inode->i_mutex);
2447 splice_from_pipe_end(pipe, &sd);
2450 mutex_unlock(&pipe->inode->i_mutex);
2453 ret = sd.num_spliced;
2456 unsigned long nr_pages;
2459 nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
2461 err = generic_write_sync(out, *ppos, ret);
2467 balance_dirty_pages_ratelimited_nr(mapping, nr_pages);
2473 static ssize_t ocfs2_file_splice_read(struct file *in,
2475 struct pipe_inode_info *pipe,
2479 int ret = 0, lock_level = 0;
2480 struct inode *inode = in->f_path.dentry->d_inode;
2482 trace_ocfs2_file_splice_read(inode, in, in->f_path.dentry,
2483 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2484 in->f_path.dentry->d_name.len,
2485 in->f_path.dentry->d_name.name, len);
2488 * See the comment in ocfs2_file_aio_read()
2490 ret = ocfs2_inode_lock_atime(inode, in->f_vfsmnt, &lock_level);
2495 ocfs2_inode_unlock(inode, lock_level);
2497 ret = generic_file_splice_read(in, ppos, pipe, len, flags);
2503 static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
2504 const struct iovec *iov,
2505 unsigned long nr_segs,
2508 int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
2509 struct file *filp = iocb->ki_filp;
2510 struct inode *inode = filp->f_path.dentry->d_inode;
2512 trace_ocfs2_file_aio_read(inode, filp, filp->f_path.dentry,
2513 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2514 filp->f_path.dentry->d_name.len,
2515 filp->f_path.dentry->d_name.name, nr_segs);
2524 ocfs2_iocb_clear_sem_locked(iocb);
2527 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
2528 * need locks to protect pending reads from racing with truncate.
2530 if (filp->f_flags & O_DIRECT) {
2532 ocfs2_iocb_set_sem_locked(iocb);
2534 ret = ocfs2_rw_lock(inode, 0);
2540 /* communicate with ocfs2_dio_end_io */
2541 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2545 * We're fine letting folks race truncates and extending
2546 * writes with read across the cluster, just like they can
2547 * locally. Hence no rw_lock during read.
2549 * Take and drop the meta data lock to update inode fields
2550 * like i_size. This allows the checks down below
2551 * generic_file_aio_read() a chance of actually working.
2553 ret = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level);
2558 ocfs2_inode_unlock(inode, lock_level);
2560 ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
2561 trace_generic_file_aio_read_ret(ret);
2563 /* buffered aio wouldn't have proper lock coverage today */
2564 BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
2566 /* see ocfs2_file_aio_write */
2567 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2574 ocfs2_iocb_clear_sem_locked(iocb);
2577 ocfs2_rw_unlock(inode, rw_level);
2582 const struct inode_operations ocfs2_file_iops = {
2583 .setattr = ocfs2_setattr,
2584 .getattr = ocfs2_getattr,
2585 .permission = ocfs2_permission,
2586 .setxattr = generic_setxattr,
2587 .getxattr = generic_getxattr,
2588 .listxattr = ocfs2_listxattr,
2589 .removexattr = generic_removexattr,
2590 .fiemap = ocfs2_fiemap,
2591 .check_acl = ocfs2_check_acl,
2594 const struct inode_operations ocfs2_special_file_iops = {
2595 .setattr = ocfs2_setattr,
2596 .getattr = ocfs2_getattr,
2597 .permission = ocfs2_permission,
2598 .check_acl = ocfs2_check_acl,
2602 * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2603 * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2605 const struct file_operations ocfs2_fops = {
2606 .llseek = generic_file_llseek,
2607 .read = do_sync_read,
2608 .write = do_sync_write,
2610 .fsync = ocfs2_sync_file,
2611 .release = ocfs2_file_release,
2612 .open = ocfs2_file_open,
2613 .aio_read = ocfs2_file_aio_read,
2614 .aio_write = ocfs2_file_aio_write,
2615 .unlocked_ioctl = ocfs2_ioctl,
2616 #ifdef CONFIG_COMPAT
2617 .compat_ioctl = ocfs2_compat_ioctl,
2620 .flock = ocfs2_flock,
2621 .splice_read = ocfs2_file_splice_read,
2622 .splice_write = ocfs2_file_splice_write,
2623 .fallocate = ocfs2_fallocate,
2626 const struct file_operations ocfs2_dops = {
2627 .llseek = generic_file_llseek,
2628 .read = generic_read_dir,
2629 .readdir = ocfs2_readdir,
2630 .fsync = ocfs2_sync_file,
2631 .release = ocfs2_dir_release,
2632 .open = ocfs2_dir_open,
2633 .unlocked_ioctl = ocfs2_ioctl,
2634 #ifdef CONFIG_COMPAT
2635 .compat_ioctl = ocfs2_compat_ioctl,
2638 .flock = ocfs2_flock,
2642 * POSIX-lockless variants of our file_operations.
2644 * These will be used if the underlying cluster stack does not support
2645 * posix file locking, if the user passes the "localflocks" mount
2646 * option, or if we have a local-only fs.
2648 * ocfs2_flock is in here because all stacks handle UNIX file locks,
2649 * so we still want it in the case of no stack support for
2650 * plocks. Internally, it will do the right thing when asked to ignore
2653 const struct file_operations ocfs2_fops_no_plocks = {
2654 .llseek = generic_file_llseek,
2655 .read = do_sync_read,
2656 .write = do_sync_write,
2658 .fsync = ocfs2_sync_file,
2659 .release = ocfs2_file_release,
2660 .open = ocfs2_file_open,
2661 .aio_read = ocfs2_file_aio_read,
2662 .aio_write = ocfs2_file_aio_write,
2663 .unlocked_ioctl = ocfs2_ioctl,
2664 #ifdef CONFIG_COMPAT
2665 .compat_ioctl = ocfs2_compat_ioctl,
2667 .flock = ocfs2_flock,
2668 .splice_read = ocfs2_file_splice_read,
2669 .splice_write = ocfs2_file_splice_write,
2670 .fallocate = ocfs2_fallocate,
2673 const struct file_operations ocfs2_dops_no_plocks = {
2674 .llseek = generic_file_llseek,
2675 .read = generic_read_dir,
2676 .readdir = ocfs2_readdir,
2677 .fsync = ocfs2_sync_file,
2678 .release = ocfs2_dir_release,
2679 .open = ocfs2_dir_open,
2680 .unlocked_ioctl = ocfs2_ioctl,
2681 #ifdef CONFIG_COMPAT
2682 .compat_ioctl = ocfs2_compat_ioctl,
2684 .flock = ocfs2_flock,