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,
191 NULL, BLKDEV_IFL_WAIT);
195 journal = osb->journal->j_journal;
196 err = jbd2_journal_force_commit(journal);
201 return (err < 0) ? -EIO : 0;
204 int ocfs2_should_update_atime(struct inode *inode,
205 struct vfsmount *vfsmnt)
208 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
210 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
213 if ((inode->i_flags & S_NOATIME) ||
214 ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
218 * We can be called with no vfsmnt structure - NFSD will
221 * Note that our action here is different than touch_atime() -
222 * if we can't tell whether this is a noatime mount, then we
223 * don't know whether to trust the value of s_atime_quantum.
228 if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
229 ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
232 if (vfsmnt->mnt_flags & MNT_RELATIME) {
233 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
234 (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
241 if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
247 int ocfs2_update_inode_atime(struct inode *inode,
248 struct buffer_head *bh)
251 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
253 struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
257 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
258 if (IS_ERR(handle)) {
259 ret = PTR_ERR(handle);
264 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
265 OCFS2_JOURNAL_ACCESS_WRITE);
272 * Don't use ocfs2_mark_inode_dirty() here as we don't always
273 * have i_mutex to guard against concurrent changes to other
276 inode->i_atime = CURRENT_TIME;
277 di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
278 di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
279 ocfs2_journal_dirty(handle, bh);
282 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,
296 i_size_write(inode, new_i_size);
297 inode->i_blocks = ocfs2_inode_sector_count(inode);
298 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
300 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
311 int ocfs2_simple_size_update(struct inode *inode,
312 struct buffer_head *di_bh,
316 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
317 handle_t *handle = NULL;
319 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
320 if (IS_ERR(handle)) {
321 ret = PTR_ERR(handle);
326 ret = ocfs2_set_inode_size(handle, inode, di_bh,
331 ocfs2_commit_trans(osb, handle);
336 static int ocfs2_cow_file_pos(struct inode *inode,
337 struct buffer_head *fe_bh,
341 u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
342 unsigned int num_clusters = 0;
343 unsigned int ext_flags = 0;
346 * If the new offset is aligned to the range of the cluster, there is
347 * no space for ocfs2_zero_range_for_truncate to fill, so no need to
350 if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
353 status = ocfs2_get_clusters(inode, cpos, &phys,
354 &num_clusters, &ext_flags);
360 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
363 return ocfs2_refcount_cow(inode, NULL, fe_bh, cpos, 1, cpos+1);
369 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
371 struct buffer_head *fe_bh,
376 struct ocfs2_dinode *di;
382 * We need to CoW the cluster contains the offset if it is reflinked
383 * since we will call ocfs2_zero_range_for_truncate later which will
384 * write "0" from offset to the end of the cluster.
386 status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
392 /* TODO: This needs to actually orphan the inode in this
395 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
396 if (IS_ERR(handle)) {
397 status = PTR_ERR(handle);
402 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
403 OCFS2_JOURNAL_ACCESS_WRITE);
410 * Do this before setting i_size.
412 cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
413 status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
420 i_size_write(inode, new_i_size);
421 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
423 di = (struct ocfs2_dinode *) fe_bh->b_data;
424 di->i_size = cpu_to_le64(new_i_size);
425 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
426 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
428 ocfs2_journal_dirty(handle, fe_bh);
431 ocfs2_commit_trans(osb, handle);
438 static int ocfs2_truncate_file(struct inode *inode,
439 struct buffer_head *di_bh,
443 struct ocfs2_dinode *fe = NULL;
444 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
446 mlog_entry("(inode = %llu, new_i_size = %llu\n",
447 (unsigned long long)OCFS2_I(inode)->ip_blkno,
448 (unsigned long long)new_i_size);
450 /* We trust di_bh because it comes from ocfs2_inode_lock(), which
451 * already validated it */
452 fe = (struct ocfs2_dinode *) di_bh->b_data;
454 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
455 "Inode %llu, inode i_size = %lld != di "
456 "i_size = %llu, i_flags = 0x%x\n",
457 (unsigned long long)OCFS2_I(inode)->ip_blkno,
459 (unsigned long long)le64_to_cpu(fe->i_size),
460 le32_to_cpu(fe->i_flags));
462 if (new_i_size > le64_to_cpu(fe->i_size)) {
463 mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
464 (unsigned long long)le64_to_cpu(fe->i_size),
465 (unsigned long long)new_i_size);
471 mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
472 (unsigned long long)le64_to_cpu(fe->i_blkno),
473 (unsigned long long)le64_to_cpu(fe->i_size),
474 (unsigned long long)new_i_size);
476 /* lets handle the simple truncate cases before doing any more
477 * cluster locking. */
478 if (new_i_size == le64_to_cpu(fe->i_size))
481 down_write(&OCFS2_I(inode)->ip_alloc_sem);
483 ocfs2_resv_discard(&osb->osb_la_resmap,
484 &OCFS2_I(inode)->ip_la_data_resv);
487 * The inode lock forced other nodes to sync and drop their
488 * pages, which (correctly) happens even if we have a truncate
489 * without allocation change - ocfs2 cluster sizes can be much
490 * greater than page size, so we have to truncate them
493 unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
494 truncate_inode_pages(inode->i_mapping, new_i_size);
496 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
497 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
498 i_size_read(inode), 1);
502 goto bail_unlock_sem;
505 /* alright, we're going to need to do a full blown alloc size
506 * change. Orphan the inode so that recovery can complete the
507 * truncate if necessary. This does the task of marking
509 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
512 goto bail_unlock_sem;
515 status = ocfs2_commit_truncate(osb, inode, di_bh);
518 goto bail_unlock_sem;
521 /* TODO: orphan dir cleanup here. */
523 up_write(&OCFS2_I(inode)->ip_alloc_sem);
526 if (!status && OCFS2_I(inode)->ip_clusters == 0)
527 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
534 * extend file allocation only here.
535 * we'll update all the disk stuff, and oip->alloc_size
537 * expect stuff to be locked, a transaction started and enough data /
538 * metadata reservations in the contexts.
540 * Will return -EAGAIN, and a reason if a restart is needed.
541 * If passed in, *reason will always be set, even in error.
543 int ocfs2_add_inode_data(struct ocfs2_super *osb,
548 struct buffer_head *fe_bh,
550 struct ocfs2_alloc_context *data_ac,
551 struct ocfs2_alloc_context *meta_ac,
552 enum ocfs2_alloc_restarted *reason_ret)
555 struct ocfs2_extent_tree et;
557 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
558 ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
559 clusters_to_add, mark_unwritten,
560 data_ac, meta_ac, reason_ret);
565 static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
566 u32 clusters_to_add, int mark_unwritten)
569 int restart_func = 0;
572 struct buffer_head *bh = NULL;
573 struct ocfs2_dinode *fe = NULL;
574 handle_t *handle = NULL;
575 struct ocfs2_alloc_context *data_ac = NULL;
576 struct ocfs2_alloc_context *meta_ac = NULL;
577 enum ocfs2_alloc_restarted why;
578 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
579 struct ocfs2_extent_tree et;
582 mlog_entry("(clusters_to_add = %u)\n", clusters_to_add);
585 * This function only exists for file systems which don't
588 BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
590 status = ocfs2_read_inode_block(inode, &bh);
595 fe = (struct ocfs2_dinode *) bh->b_data;
598 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
600 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, "
601 "clusters_to_add = %u\n",
602 (unsigned long long)OCFS2_I(inode)->ip_blkno,
603 (long long)i_size_read(inode), le32_to_cpu(fe->i_clusters),
605 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
606 status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
613 credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list,
615 handle = ocfs2_start_trans(osb, credits);
616 if (IS_ERR(handle)) {
617 status = PTR_ERR(handle);
623 restarted_transaction:
624 status = dquot_alloc_space_nodirty(inode,
625 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
630 /* reserve a write to the file entry early on - that we if we
631 * run out of credits in the allocation path, we can still
633 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
634 OCFS2_JOURNAL_ACCESS_WRITE);
640 prev_clusters = OCFS2_I(inode)->ip_clusters;
642 status = ocfs2_add_inode_data(osb,
652 if ((status < 0) && (status != -EAGAIN)) {
653 if (status != -ENOSPC)
658 ocfs2_journal_dirty(handle, bh);
660 spin_lock(&OCFS2_I(inode)->ip_lock);
661 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
662 spin_unlock(&OCFS2_I(inode)->ip_lock);
663 /* Release unused quota reservation */
664 dquot_free_space(inode,
665 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
668 if (why != RESTART_NONE && clusters_to_add) {
669 if (why == RESTART_META) {
670 mlog(0, "restarting function.\n");
674 BUG_ON(why != RESTART_TRANS);
676 mlog(0, "restarting transaction.\n");
677 /* TODO: This can be more intelligent. */
678 credits = ocfs2_calc_extend_credits(osb->sb,
681 status = ocfs2_extend_trans(handle, credits);
683 /* handle still has to be committed at
689 goto restarted_transaction;
693 mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
694 le32_to_cpu(fe->i_clusters),
695 (unsigned long long)le64_to_cpu(fe->i_size));
696 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
697 OCFS2_I(inode)->ip_clusters, (long long)i_size_read(inode));
700 if (status < 0 && did_quota)
701 dquot_free_space(inode,
702 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
704 ocfs2_commit_trans(osb, handle);
708 ocfs2_free_alloc_context(data_ac);
712 ocfs2_free_alloc_context(meta_ac);
715 if ((!status) && restart_func) {
727 * While a write will already be ordering the data, a truncate will not.
728 * Thus, we need to explicitly order the zeroed pages.
730 static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode)
732 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
733 handle_t *handle = NULL;
736 if (!ocfs2_should_order_data(inode))
739 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
740 if (IS_ERR(handle)) {
746 ret = ocfs2_jbd2_file_inode(handle, inode);
753 ocfs2_commit_trans(osb, handle);
754 handle = ERR_PTR(ret);
759 /* Some parts of this taken from generic_cont_expand, which turned out
760 * to be too fragile to do exactly what we need without us having to
761 * worry about recursive locking in ->write_begin() and ->write_end(). */
762 static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
765 struct address_space *mapping = inode->i_mapping;
767 unsigned long index = abs_from >> PAGE_CACHE_SHIFT;
768 handle_t *handle = NULL;
770 unsigned zero_from, zero_to, block_start, block_end;
772 BUG_ON(abs_from >= abs_to);
773 BUG_ON(abs_to > (((u64)index + 1) << PAGE_CACHE_SHIFT));
774 BUG_ON(abs_from & (inode->i_blkbits - 1));
776 page = find_or_create_page(mapping, index, GFP_NOFS);
783 /* Get the offsets within the page that we want to zero */
784 zero_from = abs_from & (PAGE_CACHE_SIZE - 1);
785 zero_to = abs_to & (PAGE_CACHE_SIZE - 1);
787 zero_to = PAGE_CACHE_SIZE;
790 "abs_from = %llu, abs_to = %llu, index = %lu, zero_from = %u, zero_to = %u\n",
791 (unsigned long long)abs_from, (unsigned long long)abs_to,
792 index, zero_from, zero_to);
794 /* We know that zero_from is block aligned */
795 for (block_start = zero_from; block_start < zero_to;
796 block_start = block_end) {
797 block_end = block_start + (1 << inode->i_blkbits);
800 * block_start is block-aligned. Bump it by one to
801 * force ocfs2_{prepare,commit}_write() to zero the
804 ret = ocfs2_prepare_write_nolock(inode, page,
813 handle = ocfs2_zero_start_ordered_transaction(inode);
814 if (IS_ERR(handle)) {
815 ret = PTR_ERR(handle);
821 /* must not update i_size! */
822 ret = block_commit_write(page, block_start + 1,
831 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
835 page_cache_release(page);
841 * Find the next range to zero. We do this in terms of bytes because
842 * that's what ocfs2_zero_extend() wants, and it is dealing with the
843 * pagecache. We may return multiple extents.
845 * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
846 * needs to be zeroed. range_start and range_end return the next zeroing
847 * range. A subsequent call should pass the previous range_end as its
848 * zero_start. If range_end is 0, there's nothing to do.
850 * Unwritten extents are skipped over. Refcounted extents are CoWd.
852 static int ocfs2_zero_extend_get_range(struct inode *inode,
853 struct buffer_head *di_bh,
854 u64 zero_start, u64 zero_end,
855 u64 *range_start, u64 *range_end)
857 int rc = 0, needs_cow = 0;
858 u32 p_cpos, zero_clusters = 0;
860 zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
861 u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
862 unsigned int num_clusters = 0;
863 unsigned int ext_flags = 0;
865 while (zero_cpos < last_cpos) {
866 rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
867 &num_clusters, &ext_flags);
873 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
874 zero_clusters = num_clusters;
875 if (ext_flags & OCFS2_EXT_REFCOUNTED)
880 zero_cpos += num_clusters;
882 if (!zero_clusters) {
887 while ((zero_cpos + zero_clusters) < last_cpos) {
888 rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
889 &p_cpos, &num_clusters,
896 if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
898 if (ext_flags & OCFS2_EXT_REFCOUNTED)
900 zero_clusters += num_clusters;
902 if ((zero_cpos + zero_clusters) > last_cpos)
903 zero_clusters = last_cpos - zero_cpos;
906 rc = ocfs2_refcount_cow(inode, NULL, di_bh, zero_cpos,
907 zero_clusters, UINT_MAX);
914 *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
915 *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
916 zero_cpos + zero_clusters);
923 * Zero one range returned from ocfs2_zero_extend_get_range(). The caller
924 * has made sure that the entire range needs zeroing.
926 static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
931 u64 zero_pos = range_start;
933 mlog(0, "range_start = %llu, range_end = %llu\n",
934 (unsigned long long)range_start,
935 (unsigned long long)range_end);
936 BUG_ON(range_start >= range_end);
938 while (zero_pos < range_end) {
939 next_pos = (zero_pos & PAGE_CACHE_MASK) + PAGE_CACHE_SIZE;
940 if (next_pos > range_end)
941 next_pos = range_end;
942 rc = ocfs2_write_zero_page(inode, zero_pos, next_pos);
950 * Very large extends have the potential to lock up
951 * the cpu for extended periods of time.
959 int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
963 u64 zero_start, range_start = 0, range_end = 0;
964 struct super_block *sb = inode->i_sb;
966 zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
967 mlog(0, "zero_start %llu for i_size %llu\n",
968 (unsigned long long)zero_start,
969 (unsigned long long)i_size_read(inode));
970 while (zero_start < zero_to_size) {
971 ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
982 if (range_start < zero_start)
983 range_start = zero_start;
984 if (range_end > zero_to_size)
985 range_end = zero_to_size;
987 ret = ocfs2_zero_extend_range(inode, range_start,
993 zero_start = range_end;
999 int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
1000 u64 new_i_size, u64 zero_to)
1003 u32 clusters_to_add;
1004 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1007 * Only quota files call this without a bh, and they can't be
1010 BUG_ON(!di_bh && (oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
1011 BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1013 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1014 if (clusters_to_add < oi->ip_clusters)
1015 clusters_to_add = 0;
1017 clusters_to_add -= oi->ip_clusters;
1019 if (clusters_to_add) {
1020 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
1021 clusters_to_add, 0);
1029 * Call this even if we don't add any clusters to the tree. We
1030 * still need to zero the area between the old i_size and the
1033 ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1041 static int ocfs2_extend_file(struct inode *inode,
1042 struct buffer_head *di_bh,
1046 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1050 /* setattr sometimes calls us like this. */
1051 if (new_i_size == 0)
1054 if (i_size_read(inode) == new_i_size)
1056 BUG_ON(new_i_size < i_size_read(inode));
1059 * The alloc sem blocks people in read/write from reading our
1060 * allocation until we're done changing it. We depend on
1061 * i_mutex to block other extend/truncate calls while we're
1062 * here. We even have to hold it for sparse files because there
1063 * might be some tail zeroing.
1065 down_write(&oi->ip_alloc_sem);
1067 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1069 * We can optimize small extends by keeping the inodes
1072 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1073 up_write(&oi->ip_alloc_sem);
1074 goto out_update_size;
1077 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1079 up_write(&oi->ip_alloc_sem);
1085 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1086 ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1088 ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1091 up_write(&oi->ip_alloc_sem);
1099 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1107 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1109 int status = 0, size_change;
1110 struct inode *inode = dentry->d_inode;
1111 struct super_block *sb = inode->i_sb;
1112 struct ocfs2_super *osb = OCFS2_SB(sb);
1113 struct buffer_head *bh = NULL;
1114 handle_t *handle = NULL;
1115 struct dquot *transfer_to[MAXQUOTAS] = { };
1118 mlog_entry("(0x%p, '%.*s')\n", dentry,
1119 dentry->d_name.len, dentry->d_name.name);
1121 /* ensuring we don't even attempt to truncate a symlink */
1122 if (S_ISLNK(inode->i_mode))
1123 attr->ia_valid &= ~ATTR_SIZE;
1125 if (attr->ia_valid & ATTR_MODE)
1126 mlog(0, "mode change: %d\n", attr->ia_mode);
1127 if (attr->ia_valid & ATTR_UID)
1128 mlog(0, "uid change: %d\n", attr->ia_uid);
1129 if (attr->ia_valid & ATTR_GID)
1130 mlog(0, "gid change: %d\n", attr->ia_gid);
1131 if (attr->ia_valid & ATTR_SIZE)
1132 mlog(0, "size change...\n");
1133 if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME))
1134 mlog(0, "time change...\n");
1136 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1137 | ATTR_GID | ATTR_UID | ATTR_MODE)
1138 if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) {
1139 mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid);
1143 status = inode_change_ok(inode, attr);
1147 if (is_quota_modification(inode, attr))
1148 dquot_initialize(inode);
1149 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1151 status = ocfs2_rw_lock(inode, 1);
1158 status = ocfs2_inode_lock(inode, &bh, 1);
1160 if (status != -ENOENT)
1162 goto bail_unlock_rw;
1165 if (size_change && attr->ia_size != i_size_read(inode)) {
1166 status = inode_newsize_ok(inode, attr->ia_size);
1170 if (i_size_read(inode) > attr->ia_size) {
1171 if (ocfs2_should_order_data(inode)) {
1172 status = ocfs2_begin_ordered_truncate(inode,
1177 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1179 status = ocfs2_extend_file(inode, bh, attr->ia_size);
1181 if (status != -ENOSPC)
1188 if ((attr->ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
1189 (attr->ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
1191 * Gather pointers to quota structures so that allocation /
1192 * freeing of quota structures happens here and not inside
1193 * dquot_transfer() where we have problems with lock ordering
1195 if (attr->ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid
1196 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1197 OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1198 transfer_to[USRQUOTA] = dqget(sb, attr->ia_uid,
1200 if (!transfer_to[USRQUOTA]) {
1205 if (attr->ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid
1206 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1207 OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1208 transfer_to[GRPQUOTA] = dqget(sb, attr->ia_gid,
1210 if (!transfer_to[GRPQUOTA]) {
1215 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1216 2 * ocfs2_quota_trans_credits(sb));
1217 if (IS_ERR(handle)) {
1218 status = PTR_ERR(handle);
1222 status = __dquot_transfer(inode, transfer_to);
1226 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1227 if (IS_ERR(handle)) {
1228 status = PTR_ERR(handle);
1235 * This will intentionally not wind up calling truncate_setsize(),
1236 * since all the work for a size change has been done above.
1237 * Otherwise, we could get into problems with truncate as
1238 * ip_alloc_sem is used there to protect against i_size
1241 * XXX: this means the conditional below can probably be removed.
1243 if ((attr->ia_valid & ATTR_SIZE) &&
1244 attr->ia_size != i_size_read(inode)) {
1245 status = vmtruncate(inode, attr->ia_size);
1252 setattr_copy(inode, attr);
1253 mark_inode_dirty(inode);
1255 status = ocfs2_mark_inode_dirty(handle, inode, bh);
1260 ocfs2_commit_trans(osb, handle);
1262 ocfs2_inode_unlock(inode, 1);
1265 ocfs2_rw_unlock(inode, 1);
1269 /* Release quota pointers in case we acquired them */
1270 for (qtype = 0; qtype < MAXQUOTAS; qtype++)
1271 dqput(transfer_to[qtype]);
1273 if (!status && attr->ia_valid & ATTR_MODE) {
1274 status = ocfs2_acl_chmod(inode);
1283 int ocfs2_getattr(struct vfsmount *mnt,
1284 struct dentry *dentry,
1287 struct inode *inode = dentry->d_inode;
1288 struct super_block *sb = dentry->d_inode->i_sb;
1289 struct ocfs2_super *osb = sb->s_fs_info;
1294 err = ocfs2_inode_revalidate(dentry);
1301 generic_fillattr(inode, stat);
1303 /* We set the blksize from the cluster size for performance */
1304 stat->blksize = osb->s_clustersize;
1312 int ocfs2_permission(struct inode *inode, int mask)
1318 ret = ocfs2_inode_lock(inode, NULL, 0);
1325 ret = generic_permission(inode, mask, ocfs2_check_acl);
1327 ocfs2_inode_unlock(inode, 0);
1333 static int __ocfs2_write_remove_suid(struct inode *inode,
1334 struct buffer_head *bh)
1338 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1339 struct ocfs2_dinode *di;
1341 mlog_entry("(Inode %llu, mode 0%o)\n",
1342 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_mode);
1344 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1345 if (IS_ERR(handle)) {
1346 ret = PTR_ERR(handle);
1351 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1352 OCFS2_JOURNAL_ACCESS_WRITE);
1358 inode->i_mode &= ~S_ISUID;
1359 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1360 inode->i_mode &= ~S_ISGID;
1362 di = (struct ocfs2_dinode *) bh->b_data;
1363 di->i_mode = cpu_to_le16(inode->i_mode);
1365 ocfs2_journal_dirty(handle, bh);
1368 ocfs2_commit_trans(osb, handle);
1375 * Will look for holes and unwritten extents in the range starting at
1376 * pos for count bytes (inclusive).
1378 static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1382 unsigned int extent_flags;
1383 u32 cpos, clusters, extent_len, phys_cpos;
1384 struct super_block *sb = inode->i_sb;
1386 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1387 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1390 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1397 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
1402 if (extent_len > clusters)
1403 extent_len = clusters;
1405 clusters -= extent_len;
1412 static int ocfs2_write_remove_suid(struct inode *inode)
1415 struct buffer_head *bh = NULL;
1417 ret = ocfs2_read_inode_block(inode, &bh);
1423 ret = __ocfs2_write_remove_suid(inode, bh);
1430 * Allocate enough extents to cover the region starting at byte offset
1431 * start for len bytes. Existing extents are skipped, any extents
1432 * added are marked as "unwritten".
1434 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1438 u32 cpos, phys_cpos, clusters, alloc_size;
1439 u64 end = start + len;
1440 struct buffer_head *di_bh = NULL;
1442 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1443 ret = ocfs2_read_inode_block(inode, &di_bh);
1450 * Nothing to do if the requested reservation range
1451 * fits within the inode.
1453 if (ocfs2_size_fits_inline_data(di_bh, end))
1456 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1464 * We consider both start and len to be inclusive.
1466 cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1467 clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1471 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1479 * Hole or existing extent len can be arbitrary, so
1480 * cap it to our own allocation request.
1482 if (alloc_size > clusters)
1483 alloc_size = clusters;
1487 * We already have an allocation at this
1488 * region so we can safely skip it.
1493 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1502 clusters -= alloc_size;
1513 * Truncate a byte range, avoiding pages within partial clusters. This
1514 * preserves those pages for the zeroing code to write to.
1516 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1519 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1521 struct address_space *mapping = inode->i_mapping;
1523 start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1524 end = byte_start + byte_len;
1525 end = end & ~(osb->s_clustersize - 1);
1528 unmap_mapping_range(mapping, start, end - start, 0);
1529 truncate_inode_pages_range(mapping, start, end - 1);
1533 static int ocfs2_zero_partial_clusters(struct inode *inode,
1537 u64 tmpend, end = start + len;
1538 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1539 unsigned int csize = osb->s_clustersize;
1543 * The "start" and "end" values are NOT necessarily part of
1544 * the range whose allocation is being deleted. Rather, this
1545 * is what the user passed in with the request. We must zero
1546 * partial clusters here. There's no need to worry about
1547 * physical allocation - the zeroing code knows to skip holes.
1549 mlog(0, "byte start: %llu, end: %llu\n",
1550 (unsigned long long)start, (unsigned long long)end);
1553 * If both edges are on a cluster boundary then there's no
1554 * zeroing required as the region is part of the allocation to
1557 if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1560 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1561 if (IS_ERR(handle)) {
1562 ret = PTR_ERR(handle);
1568 * We want to get the byte offset of the end of the 1st cluster.
1570 tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1574 mlog(0, "1st range: start: %llu, tmpend: %llu\n",
1575 (unsigned long long)start, (unsigned long long)tmpend);
1577 ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1583 * This may make start and end equal, but the zeroing
1584 * code will skip any work in that case so there's no
1585 * need to catch it up here.
1587 start = end & ~(osb->s_clustersize - 1);
1589 mlog(0, "2nd range: start: %llu, end: %llu\n",
1590 (unsigned long long)start, (unsigned long long)end);
1592 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1597 ocfs2_commit_trans(osb, handle);
1602 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1605 struct ocfs2_extent_rec *rec = NULL;
1607 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1609 rec = &el->l_recs[i];
1611 if (le32_to_cpu(rec->e_cpos) < pos)
1619 * Helper to calculate the punching pos and length in one run, we handle the
1620 * following three cases in order:
1622 * - remove the entire record
1623 * - remove a partial record
1624 * - no record needs to be removed (hole-punching completed)
1626 static void ocfs2_calc_trunc_pos(struct inode *inode,
1627 struct ocfs2_extent_list *el,
1628 struct ocfs2_extent_rec *rec,
1629 u32 trunc_start, u32 *trunc_cpos,
1630 u32 *trunc_len, u32 *trunc_end,
1631 u64 *blkno, int *done)
1636 range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1638 if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1639 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1641 * Skip holes if any.
1643 if (range < *trunc_end)
1645 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1646 *blkno = le64_to_cpu(rec->e_blkno);
1647 *trunc_end = le32_to_cpu(rec->e_cpos);
1648 } else if (range > trunc_start) {
1649 *trunc_cpos = trunc_start;
1650 *trunc_len = *trunc_end - trunc_start;
1651 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1652 *blkno = le64_to_cpu(rec->e_blkno) +
1653 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1654 *trunc_end = trunc_start;
1657 * It may have two following possibilities:
1659 * - last record has been removed
1660 * - trunc_start was within a hole
1662 * both two cases mean the completion of hole punching.
1670 static int ocfs2_remove_inode_range(struct inode *inode,
1671 struct buffer_head *di_bh, u64 byte_start,
1674 int ret = 0, flags = 0, done = 0, i;
1675 u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1677 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1678 struct ocfs2_cached_dealloc_ctxt dealloc;
1679 struct address_space *mapping = inode->i_mapping;
1680 struct ocfs2_extent_tree et;
1681 struct ocfs2_path *path = NULL;
1682 struct ocfs2_extent_list *el = NULL;
1683 struct ocfs2_extent_rec *rec = NULL;
1684 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1685 u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1687 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1688 ocfs2_init_dealloc_ctxt(&dealloc);
1693 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1694 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1695 byte_start + byte_len, 0);
1701 * There's no need to get fancy with the page cache
1702 * truncate of an inline-data inode. We're talking
1703 * about less than a page here, which will be cached
1704 * in the dinode buffer anyway.
1706 unmap_mapping_range(mapping, 0, 0, 0);
1707 truncate_inode_pages(mapping, 0);
1712 * For reflinks, we may need to CoW 2 clusters which might be
1713 * partially zero'd later, if hole's start and end offset were
1714 * within one cluster(means is not exactly aligned to clustersize).
1717 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
1719 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1725 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1732 trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1733 trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1734 cluster_in_el = trunc_end;
1736 mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, cend: %u\n",
1737 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1738 (unsigned long long)byte_start,
1739 (unsigned long long)byte_len, trunc_start, trunc_end);
1741 ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1747 path = ocfs2_new_path_from_et(&et);
1754 while (trunc_end > trunc_start) {
1756 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1763 el = path_leaf_el(path);
1765 i = ocfs2_find_rec(el, trunc_end);
1767 * Need to go to previous extent block.
1770 if (path->p_tree_depth == 0)
1773 ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1782 * We've reached the leftmost extent block,
1783 * it's safe to leave.
1785 if (cluster_in_el == 0)
1789 * The 'pos' searched for previous extent block is
1790 * always one cluster less than actual trunc_end.
1792 trunc_end = cluster_in_el + 1;
1794 ocfs2_reinit_path(path, 1);
1799 rec = &el->l_recs[i];
1801 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1802 &trunc_len, &trunc_end, &blkno, &done);
1806 flags = rec->e_flags;
1807 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1809 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1810 phys_cpos, trunc_len, flags,
1811 &dealloc, refcount_loc);
1817 cluster_in_el = trunc_end;
1819 ocfs2_reinit_path(path, 1);
1822 ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1825 ocfs2_schedule_truncate_log_flush(osb, 1);
1826 ocfs2_run_deallocs(osb, &dealloc);
1832 * Parts of this function taken from xfs_change_file_space()
1834 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1835 loff_t f_pos, unsigned int cmd,
1836 struct ocfs2_space_resv *sr,
1842 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1843 struct buffer_head *di_bh = NULL;
1845 unsigned long long max_off = inode->i_sb->s_maxbytes;
1847 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1850 mutex_lock(&inode->i_mutex);
1853 * This prevents concurrent writes on other nodes
1855 ret = ocfs2_rw_lock(inode, 1);
1861 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1867 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1869 goto out_inode_unlock;
1872 switch (sr->l_whence) {
1873 case 0: /*SEEK_SET*/
1875 case 1: /*SEEK_CUR*/
1876 sr->l_start += f_pos;
1878 case 2: /*SEEK_END*/
1879 sr->l_start += i_size_read(inode);
1883 goto out_inode_unlock;
1887 llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1890 || sr->l_start > max_off
1891 || (sr->l_start + llen) < 0
1892 || (sr->l_start + llen) > max_off) {
1894 goto out_inode_unlock;
1896 size = sr->l_start + sr->l_len;
1898 if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) {
1899 if (sr->l_len <= 0) {
1901 goto out_inode_unlock;
1905 if (file && should_remove_suid(file->f_path.dentry)) {
1906 ret = __ocfs2_write_remove_suid(inode, di_bh);
1909 goto out_inode_unlock;
1913 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1915 case OCFS2_IOC_RESVSP:
1916 case OCFS2_IOC_RESVSP64:
1918 * This takes unsigned offsets, but the signed ones we
1919 * pass have been checked against overflow above.
1921 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1924 case OCFS2_IOC_UNRESVSP:
1925 case OCFS2_IOC_UNRESVSP64:
1926 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1932 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1935 goto out_inode_unlock;
1939 * We update c/mtime for these changes
1941 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1942 if (IS_ERR(handle)) {
1943 ret = PTR_ERR(handle);
1945 goto out_inode_unlock;
1948 if (change_size && i_size_read(inode) < size)
1949 i_size_write(inode, size);
1951 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1952 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1956 ocfs2_commit_trans(osb, handle);
1960 ocfs2_inode_unlock(inode, 1);
1962 ocfs2_rw_unlock(inode, 1);
1965 mutex_unlock(&inode->i_mutex);
1969 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1970 struct ocfs2_space_resv *sr)
1972 struct inode *inode = file->f_path.dentry->d_inode;
1973 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1975 if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1976 !ocfs2_writes_unwritten_extents(osb))
1978 else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1979 !ocfs2_sparse_alloc(osb))
1982 if (!S_ISREG(inode->i_mode))
1985 if (!(file->f_mode & FMODE_WRITE))
1988 return __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1991 static long ocfs2_fallocate(struct inode *inode, int mode, loff_t offset,
1994 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1995 struct ocfs2_space_resv sr;
1996 int change_size = 1;
1998 if (!ocfs2_writes_unwritten_extents(osb))
2001 if (S_ISDIR(inode->i_mode))
2004 if (mode & FALLOC_FL_KEEP_SIZE)
2008 sr.l_start = (s64)offset;
2009 sr.l_len = (s64)len;
2011 return __ocfs2_change_file_space(NULL, inode, offset,
2012 OCFS2_IOC_RESVSP64, &sr, change_size);
2015 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2019 unsigned int extent_flags;
2020 u32 cpos, clusters, extent_len, phys_cpos;
2021 struct super_block *sb = inode->i_sb;
2023 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2024 !(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) ||
2025 OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2028 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2029 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2032 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2039 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2044 if (extent_len > clusters)
2045 extent_len = clusters;
2047 clusters -= extent_len;
2054 static int ocfs2_prepare_inode_for_refcount(struct inode *inode,
2056 loff_t pos, size_t count,
2060 struct buffer_head *di_bh = NULL;
2061 u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2063 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2065 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2073 ret = ocfs2_refcount_cow(inode, file, di_bh, cpos, clusters, UINT_MAX);
2081 static int ocfs2_prepare_inode_for_write(struct file *file,
2088 int ret = 0, meta_level = 0;
2089 struct dentry *dentry = file->f_path.dentry;
2090 struct inode *inode = dentry->d_inode;
2091 loff_t saved_pos, end;
2094 * We start with a read level meta lock and only jump to an ex
2095 * if we need to make modifications here.
2098 ret = ocfs2_inode_lock(inode, NULL, meta_level);
2105 /* Clear suid / sgid if necessary. We do this here
2106 * instead of later in the write path because
2107 * remove_suid() calls ->setattr without any hint that
2108 * we may have already done our cluster locking. Since
2109 * ocfs2_setattr() *must* take cluster locks to
2110 * proceeed, this will lead us to recursively lock the
2111 * inode. There's also the dinode i_size state which
2112 * can be lost via setattr during extending writes (we
2113 * set inode->i_size at the end of a write. */
2114 if (should_remove_suid(dentry)) {
2115 if (meta_level == 0) {
2116 ocfs2_inode_unlock(inode, meta_level);
2121 ret = ocfs2_write_remove_suid(inode);
2128 /* work on a copy of ppos until we're sure that we won't have
2129 * to recalculate it due to relocking. */
2131 saved_pos = i_size_read(inode);
2132 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos);
2137 end = saved_pos + count;
2139 ret = ocfs2_check_range_for_refcount(inode, saved_pos, count);
2141 ocfs2_inode_unlock(inode, meta_level);
2144 ret = ocfs2_prepare_inode_for_refcount(inode,
2161 * Skip the O_DIRECT checks if we don't need
2164 if (!direct_io || !(*direct_io))
2168 * There's no sane way to do direct writes to an inode
2171 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2177 * Allowing concurrent direct writes means
2178 * i_size changes wouldn't be synchronized, so
2179 * one node could wind up truncating another
2182 if (end > i_size_read(inode)) {
2188 * We don't fill holes during direct io, so
2189 * check for them here. If any are found, the
2190 * caller will have to retake some cluster
2191 * locks and initiate the io as buffered.
2193 ret = ocfs2_check_range_for_holes(inode, saved_pos, count);
2206 if (meta_level >= 0)
2207 ocfs2_inode_unlock(inode, meta_level);
2213 static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
2214 const struct iovec *iov,
2215 unsigned long nr_segs,
2218 int ret, direct_io, appending, rw_level, have_alloc_sem = 0;
2219 int can_do_direct, has_refcount = 0;
2220 ssize_t written = 0;
2221 size_t ocount; /* original count */
2222 size_t count; /* after file limit checks */
2223 loff_t old_size, *ppos = &iocb->ki_pos;
2225 struct file *file = iocb->ki_filp;
2226 struct inode *inode = file->f_path.dentry->d_inode;
2227 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2228 int full_coherency = !(osb->s_mount_opt &
2229 OCFS2_MOUNT_COHERENCY_BUFFERED);
2231 mlog_entry("(0x%p, %u, '%.*s')\n", file,
2232 (unsigned int)nr_segs,
2233 file->f_path.dentry->d_name.len,
2234 file->f_path.dentry->d_name.name);
2236 if (iocb->ki_left == 0)
2239 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2241 appending = file->f_flags & O_APPEND ? 1 : 0;
2242 direct_io = file->f_flags & O_DIRECT ? 1 : 0;
2244 mutex_lock(&inode->i_mutex);
2247 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
2249 down_read(&inode->i_alloc_sem);
2254 * Concurrent O_DIRECT writes are allowed with
2255 * mount_option "coherency=buffered".
2257 rw_level = (!direct_io || full_coherency);
2259 ret = ocfs2_rw_lock(inode, rw_level);
2266 * O_DIRECT writes with "coherency=full" need to take EX cluster
2267 * inode_lock to guarantee coherency.
2269 if (direct_io && full_coherency) {
2271 * We need to take and drop the inode lock to force
2272 * other nodes to drop their caches. Buffered I/O
2273 * already does this in write_begin().
2275 ret = ocfs2_inode_lock(inode, NULL, 1);
2281 ocfs2_inode_unlock(inode, 1);
2284 can_do_direct = direct_io;
2285 ret = ocfs2_prepare_inode_for_write(file, ppos,
2286 iocb->ki_left, appending,
2287 &can_do_direct, &has_refcount);
2294 * We can't complete the direct I/O as requested, fall back to
2297 if (direct_io && !can_do_direct) {
2298 ocfs2_rw_unlock(inode, rw_level);
2299 up_read(&inode->i_alloc_sem);
2309 * To later detect whether a journal commit for sync writes is
2310 * necessary, we sample i_size, and cluster count here.
2312 old_size = i_size_read(inode);
2313 old_clusters = OCFS2_I(inode)->ip_clusters;
2315 /* communicate with ocfs2_dio_end_io */
2316 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2318 ret = generic_segment_checks(iov, &nr_segs, &ocount,
2324 ret = generic_write_checks(file, ppos, &count,
2325 S_ISBLK(inode->i_mode));
2330 written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
2331 ppos, count, ocount);
2337 current->backing_dev_info = file->f_mapping->backing_dev_info;
2338 written = generic_file_buffered_write(iocb, iov, nr_segs, *ppos,
2340 current->backing_dev_info = NULL;
2344 /* buffered aio wouldn't have proper lock coverage today */
2345 BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
2347 if (((file->f_flags & O_DSYNC) && !direct_io) || IS_SYNC(inode) ||
2348 ((file->f_flags & O_DIRECT) && !direct_io)) {
2349 ret = filemap_fdatawrite_range(file->f_mapping, pos,
2354 if (!ret && ((old_size != i_size_read(inode)) ||
2355 (old_clusters != OCFS2_I(inode)->ip_clusters) ||
2357 ret = jbd2_journal_force_commit(osb->journal->j_journal);
2363 ret = filemap_fdatawait_range(file->f_mapping, pos,
2368 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2369 * function pointer which is called when o_direct io completes so that
2370 * it can unlock our rw lock. (it's the clustered equivalent of
2371 * i_alloc_sem; protects truncate from racing with pending ios).
2372 * Unfortunately there are error cases which call end_io and others
2373 * that don't. so we don't have to unlock the rw_lock if either an
2374 * async dio is going to do it in the future or an end_io after an
2375 * error has already done it.
2377 if ((ret == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2384 ocfs2_rw_unlock(inode, rw_level);
2388 up_read(&inode->i_alloc_sem);
2390 mutex_unlock(&inode->i_mutex);
2398 static int ocfs2_splice_to_file(struct pipe_inode_info *pipe,
2400 struct splice_desc *sd)
2404 ret = ocfs2_prepare_inode_for_write(out, &sd->pos,
2405 sd->total_len, 0, NULL, NULL);
2411 return splice_from_pipe_feed(pipe, sd, pipe_to_file);
2414 static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
2421 struct address_space *mapping = out->f_mapping;
2422 struct inode *inode = mapping->host;
2423 struct splice_desc sd = {
2430 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe,
2432 out->f_path.dentry->d_name.len,
2433 out->f_path.dentry->d_name.name);
2436 mutex_lock_nested(&pipe->inode->i_mutex, I_MUTEX_PARENT);
2438 splice_from_pipe_begin(&sd);
2440 ret = splice_from_pipe_next(pipe, &sd);
2444 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2445 ret = ocfs2_rw_lock(inode, 1);
2449 ret = ocfs2_splice_to_file(pipe, out, &sd);
2450 ocfs2_rw_unlock(inode, 1);
2452 mutex_unlock(&inode->i_mutex);
2454 splice_from_pipe_end(pipe, &sd);
2457 mutex_unlock(&pipe->inode->i_mutex);
2460 ret = sd.num_spliced;
2463 unsigned long nr_pages;
2466 nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
2468 err = generic_write_sync(out, *ppos, ret);
2474 balance_dirty_pages_ratelimited_nr(mapping, nr_pages);
2481 static ssize_t ocfs2_file_splice_read(struct file *in,
2483 struct pipe_inode_info *pipe,
2487 int ret = 0, lock_level = 0;
2488 struct inode *inode = in->f_path.dentry->d_inode;
2490 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe,
2492 in->f_path.dentry->d_name.len,
2493 in->f_path.dentry->d_name.name);
2496 * See the comment in ocfs2_file_aio_read()
2498 ret = ocfs2_inode_lock_atime(inode, in->f_vfsmnt, &lock_level);
2503 ocfs2_inode_unlock(inode, lock_level);
2505 ret = generic_file_splice_read(in, ppos, pipe, len, flags);
2512 static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
2513 const struct iovec *iov,
2514 unsigned long nr_segs,
2517 int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
2518 struct file *filp = iocb->ki_filp;
2519 struct inode *inode = filp->f_path.dentry->d_inode;
2521 mlog_entry("(0x%p, %u, '%.*s')\n", filp,
2522 (unsigned int)nr_segs,
2523 filp->f_path.dentry->d_name.len,
2524 filp->f_path.dentry->d_name.name);
2533 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
2534 * need locks to protect pending reads from racing with truncate.
2536 if (filp->f_flags & O_DIRECT) {
2537 down_read(&inode->i_alloc_sem);
2540 ret = ocfs2_rw_lock(inode, 0);
2546 /* communicate with ocfs2_dio_end_io */
2547 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2551 * We're fine letting folks race truncates and extending
2552 * writes with read across the cluster, just like they can
2553 * locally. Hence no rw_lock during read.
2555 * Take and drop the meta data lock to update inode fields
2556 * like i_size. This allows the checks down below
2557 * generic_file_aio_read() a chance of actually working.
2559 ret = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level);
2564 ocfs2_inode_unlock(inode, lock_level);
2566 ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
2568 mlog(0, "generic_file_aio_read returned -EINVAL\n");
2570 /* buffered aio wouldn't have proper lock coverage today */
2571 BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
2573 /* see ocfs2_file_aio_write */
2574 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2581 up_read(&inode->i_alloc_sem);
2583 ocfs2_rw_unlock(inode, rw_level);
2589 const struct inode_operations ocfs2_file_iops = {
2590 .setattr = ocfs2_setattr,
2591 .getattr = ocfs2_getattr,
2592 .permission = ocfs2_permission,
2593 .setxattr = generic_setxattr,
2594 .getxattr = generic_getxattr,
2595 .listxattr = ocfs2_listxattr,
2596 .removexattr = generic_removexattr,
2597 .fallocate = ocfs2_fallocate,
2598 .fiemap = ocfs2_fiemap,
2601 const struct inode_operations ocfs2_special_file_iops = {
2602 .setattr = ocfs2_setattr,
2603 .getattr = ocfs2_getattr,
2604 .permission = ocfs2_permission,
2608 * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2609 * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2611 const struct file_operations ocfs2_fops = {
2612 .llseek = generic_file_llseek,
2613 .read = do_sync_read,
2614 .write = do_sync_write,
2616 .fsync = ocfs2_sync_file,
2617 .release = ocfs2_file_release,
2618 .open = ocfs2_file_open,
2619 .aio_read = ocfs2_file_aio_read,
2620 .aio_write = ocfs2_file_aio_write,
2621 .unlocked_ioctl = ocfs2_ioctl,
2622 #ifdef CONFIG_COMPAT
2623 .compat_ioctl = ocfs2_compat_ioctl,
2626 .flock = ocfs2_flock,
2627 .splice_read = ocfs2_file_splice_read,
2628 .splice_write = ocfs2_file_splice_write,
2631 const struct file_operations ocfs2_dops = {
2632 .llseek = generic_file_llseek,
2633 .read = generic_read_dir,
2634 .readdir = ocfs2_readdir,
2635 .fsync = ocfs2_sync_file,
2636 .release = ocfs2_dir_release,
2637 .open = ocfs2_dir_open,
2638 .unlocked_ioctl = ocfs2_ioctl,
2639 #ifdef CONFIG_COMPAT
2640 .compat_ioctl = ocfs2_compat_ioctl,
2643 .flock = ocfs2_flock,
2647 * POSIX-lockless variants of our file_operations.
2649 * These will be used if the underlying cluster stack does not support
2650 * posix file locking, if the user passes the "localflocks" mount
2651 * option, or if we have a local-only fs.
2653 * ocfs2_flock is in here because all stacks handle UNIX file locks,
2654 * so we still want it in the case of no stack support for
2655 * plocks. Internally, it will do the right thing when asked to ignore
2658 const struct file_operations ocfs2_fops_no_plocks = {
2659 .llseek = generic_file_llseek,
2660 .read = do_sync_read,
2661 .write = do_sync_write,
2663 .fsync = ocfs2_sync_file,
2664 .release = ocfs2_file_release,
2665 .open = ocfs2_file_open,
2666 .aio_read = ocfs2_file_aio_read,
2667 .aio_write = ocfs2_file_aio_write,
2668 .unlocked_ioctl = ocfs2_ioctl,
2669 #ifdef CONFIG_COMPAT
2670 .compat_ioctl = ocfs2_compat_ioctl,
2672 .flock = ocfs2_flock,
2673 .splice_read = ocfs2_file_splice_read,
2674 .splice_write = ocfs2_file_splice_write,
2677 const struct file_operations ocfs2_dops_no_plocks = {
2678 .llseek = generic_file_llseek,
2679 .read = generic_read_dir,
2680 .readdir = ocfs2_readdir,
2681 .fsync = ocfs2_sync_file,
2682 .release = ocfs2_dir_release,
2683 .open = ocfs2_dir_open,
2684 .unlocked_ioctl = ocfs2_ioctl,
2685 #ifdef CONFIG_COMPAT
2686 .compat_ioctl = ocfs2_compat_ioctl,
2688 .flock = ocfs2_flock,