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(0, "(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(0, "(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(0, "(0x%p, %d, 0x%p, '%.*s')\n", file, datasync,
181 file->f_path.dentry, file->f_path.dentry->d_name.len,
182 file->f_path.dentry->d_name.name);
184 if (datasync && !(inode->i_state & I_DIRTY_DATASYNC)) {
186 * We still have to flush drive's caches to get data to the
189 if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
190 blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
194 journal = osb->journal->j_journal;
195 err = jbd2_journal_force_commit(journal);
200 return (err < 0) ? -EIO : 0;
203 int ocfs2_should_update_atime(struct inode *inode,
204 struct vfsmount *vfsmnt)
207 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
209 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
212 if ((inode->i_flags & S_NOATIME) ||
213 ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
217 * We can be called with no vfsmnt structure - NFSD will
220 * Note that our action here is different than touch_atime() -
221 * if we can't tell whether this is a noatime mount, then we
222 * don't know whether to trust the value of s_atime_quantum.
227 if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
228 ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
231 if (vfsmnt->mnt_flags & MNT_RELATIME) {
232 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
233 (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
240 if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
246 int ocfs2_update_inode_atime(struct inode *inode,
247 struct buffer_head *bh)
250 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
252 struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
254 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
255 if (IS_ERR(handle)) {
256 ret = PTR_ERR(handle);
261 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
262 OCFS2_JOURNAL_ACCESS_WRITE);
269 * Don't use ocfs2_mark_inode_dirty() here as we don't always
270 * have i_mutex to guard against concurrent changes to other
273 inode->i_atime = CURRENT_TIME;
274 di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
275 di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
276 ocfs2_journal_dirty(handle, bh);
279 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
285 static int ocfs2_set_inode_size(handle_t *handle,
287 struct buffer_head *fe_bh,
292 i_size_write(inode, new_i_size);
293 inode->i_blocks = ocfs2_inode_sector_count(inode);
294 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
296 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
307 int ocfs2_simple_size_update(struct inode *inode,
308 struct buffer_head *di_bh,
312 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
313 handle_t *handle = NULL;
315 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
316 if (IS_ERR(handle)) {
317 ret = PTR_ERR(handle);
322 ret = ocfs2_set_inode_size(handle, inode, di_bh,
327 ocfs2_commit_trans(osb, handle);
332 static int ocfs2_cow_file_pos(struct inode *inode,
333 struct buffer_head *fe_bh,
337 u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
338 unsigned int num_clusters = 0;
339 unsigned int ext_flags = 0;
342 * If the new offset is aligned to the range of the cluster, there is
343 * no space for ocfs2_zero_range_for_truncate to fill, so no need to
346 if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
349 status = ocfs2_get_clusters(inode, cpos, &phys,
350 &num_clusters, &ext_flags);
356 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
359 return ocfs2_refcount_cow(inode, NULL, fe_bh, cpos, 1, cpos+1);
365 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
367 struct buffer_head *fe_bh,
372 struct ocfs2_dinode *di;
376 * We need to CoW the cluster contains the offset if it is reflinked
377 * since we will call ocfs2_zero_range_for_truncate later which will
378 * write "0" from offset to the end of the cluster.
380 status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
386 /* TODO: This needs to actually orphan the inode in this
389 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
390 if (IS_ERR(handle)) {
391 status = PTR_ERR(handle);
396 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
397 OCFS2_JOURNAL_ACCESS_WRITE);
404 * Do this before setting i_size.
406 cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
407 status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
414 i_size_write(inode, new_i_size);
415 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
417 di = (struct ocfs2_dinode *) fe_bh->b_data;
418 di->i_size = cpu_to_le64(new_i_size);
419 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
420 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
422 ocfs2_journal_dirty(handle, fe_bh);
425 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 mlog(0, "(inode = %llu, new_i_size = %llu\n",
441 (unsigned long long)OCFS2_I(inode)->ip_blkno,
442 (unsigned long long)new_i_size);
444 /* We trust di_bh because it comes from ocfs2_inode_lock(), which
445 * already validated it */
446 fe = (struct ocfs2_dinode *) di_bh->b_data;
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 mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
458 (unsigned long long)le64_to_cpu(fe->i_size),
459 (unsigned long long)new_i_size);
465 mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
466 (unsigned long long)le64_to_cpu(fe->i_blkno),
467 (unsigned long long)le64_to_cpu(fe->i_size),
468 (unsigned long long)new_i_size);
470 /* lets handle the simple truncate cases before doing any more
471 * cluster locking. */
472 if (new_i_size == le64_to_cpu(fe->i_size))
475 down_write(&OCFS2_I(inode)->ip_alloc_sem);
477 ocfs2_resv_discard(&osb->osb_la_resmap,
478 &OCFS2_I(inode)->ip_la_data_resv);
481 * The inode lock forced other nodes to sync and drop their
482 * pages, which (correctly) happens even if we have a truncate
483 * without allocation change - ocfs2 cluster sizes can be much
484 * greater than page size, so we have to truncate them
487 unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
488 truncate_inode_pages(inode->i_mapping, new_i_size);
490 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
491 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
492 i_size_read(inode), 1);
496 goto bail_unlock_sem;
499 /* alright, we're going to need to do a full blown alloc size
500 * change. Orphan the inode so that recovery can complete the
501 * truncate if necessary. This does the task of marking
503 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
506 goto bail_unlock_sem;
509 status = ocfs2_commit_truncate(osb, inode, di_bh);
512 goto bail_unlock_sem;
515 /* TODO: orphan dir cleanup here. */
517 up_write(&OCFS2_I(inode)->ip_alloc_sem);
520 if (!status && OCFS2_I(inode)->ip_clusters == 0)
521 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
528 * extend file allocation only here.
529 * we'll update all the disk stuff, and oip->alloc_size
531 * expect stuff to be locked, a transaction started and enough data /
532 * metadata reservations in the contexts.
534 * Will return -EAGAIN, and a reason if a restart is needed.
535 * If passed in, *reason will always be set, even in error.
537 int ocfs2_add_inode_data(struct ocfs2_super *osb,
542 struct buffer_head *fe_bh,
544 struct ocfs2_alloc_context *data_ac,
545 struct ocfs2_alloc_context *meta_ac,
546 enum ocfs2_alloc_restarted *reason_ret)
549 struct ocfs2_extent_tree et;
551 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
552 ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
553 clusters_to_add, mark_unwritten,
554 data_ac, meta_ac, reason_ret);
559 static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
560 u32 clusters_to_add, int mark_unwritten)
563 int restart_func = 0;
566 struct buffer_head *bh = NULL;
567 struct ocfs2_dinode *fe = NULL;
568 handle_t *handle = NULL;
569 struct ocfs2_alloc_context *data_ac = NULL;
570 struct ocfs2_alloc_context *meta_ac = NULL;
571 enum ocfs2_alloc_restarted why;
572 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
573 struct ocfs2_extent_tree et;
576 mlog(0, "(clusters_to_add = %u)\n", clusters_to_add);
579 * This function only exists for file systems which don't
582 BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
584 status = ocfs2_read_inode_block(inode, &bh);
589 fe = (struct ocfs2_dinode *) bh->b_data;
592 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
594 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, "
595 "clusters_to_add = %u\n",
596 (unsigned long long)OCFS2_I(inode)->ip_blkno,
597 (long long)i_size_read(inode), le32_to_cpu(fe->i_clusters),
599 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
600 status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
607 credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list,
609 handle = ocfs2_start_trans(osb, credits);
610 if (IS_ERR(handle)) {
611 status = PTR_ERR(handle);
617 restarted_transaction:
618 status = dquot_alloc_space_nodirty(inode,
619 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
624 /* reserve a write to the file entry early on - that we if we
625 * run out of credits in the allocation path, we can still
627 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
628 OCFS2_JOURNAL_ACCESS_WRITE);
634 prev_clusters = OCFS2_I(inode)->ip_clusters;
636 status = ocfs2_add_inode_data(osb,
646 if ((status < 0) && (status != -EAGAIN)) {
647 if (status != -ENOSPC)
652 ocfs2_journal_dirty(handle, bh);
654 spin_lock(&OCFS2_I(inode)->ip_lock);
655 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
656 spin_unlock(&OCFS2_I(inode)->ip_lock);
657 /* Release unused quota reservation */
658 dquot_free_space(inode,
659 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
662 if (why != RESTART_NONE && clusters_to_add) {
663 if (why == RESTART_META) {
664 mlog(0, "restarting function.\n");
668 BUG_ON(why != RESTART_TRANS);
670 mlog(0, "restarting transaction.\n");
671 /* TODO: This can be more intelligent. */
672 credits = ocfs2_calc_extend_credits(osb->sb,
675 status = ocfs2_extend_trans(handle, credits);
677 /* handle still has to be committed at
683 goto restarted_transaction;
687 mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
688 le32_to_cpu(fe->i_clusters),
689 (unsigned long long)le64_to_cpu(fe->i_size));
690 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
691 OCFS2_I(inode)->ip_clusters, (long long)i_size_read(inode));
694 if (status < 0 && did_quota)
695 dquot_free_space(inode,
696 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
698 ocfs2_commit_trans(osb, handle);
702 ocfs2_free_alloc_context(data_ac);
706 ocfs2_free_alloc_context(meta_ac);
709 if ((!status) && restart_func) {
721 * While a write will already be ordering the data, a truncate will not.
722 * Thus, we need to explicitly order the zeroed pages.
724 static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode)
726 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
727 handle_t *handle = NULL;
730 if (!ocfs2_should_order_data(inode))
733 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
734 if (IS_ERR(handle)) {
740 ret = ocfs2_jbd2_file_inode(handle, inode);
747 ocfs2_commit_trans(osb, handle);
748 handle = ERR_PTR(ret);
753 /* Some parts of this taken from generic_cont_expand, which turned out
754 * to be too fragile to do exactly what we need without us having to
755 * worry about recursive locking in ->write_begin() and ->write_end(). */
756 static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
759 struct address_space *mapping = inode->i_mapping;
761 unsigned long index = abs_from >> PAGE_CACHE_SHIFT;
762 handle_t *handle = NULL;
764 unsigned zero_from, zero_to, block_start, block_end;
766 BUG_ON(abs_from >= abs_to);
767 BUG_ON(abs_to > (((u64)index + 1) << PAGE_CACHE_SHIFT));
768 BUG_ON(abs_from & (inode->i_blkbits - 1));
770 page = find_or_create_page(mapping, index, GFP_NOFS);
777 /* Get the offsets within the page that we want to zero */
778 zero_from = abs_from & (PAGE_CACHE_SIZE - 1);
779 zero_to = abs_to & (PAGE_CACHE_SIZE - 1);
781 zero_to = PAGE_CACHE_SIZE;
784 "abs_from = %llu, abs_to = %llu, index = %lu, zero_from = %u, zero_to = %u\n",
785 (unsigned long long)abs_from, (unsigned long long)abs_to,
786 index, zero_from, zero_to);
788 /* We know that zero_from is block aligned */
789 for (block_start = zero_from; block_start < zero_to;
790 block_start = block_end) {
791 block_end = block_start + (1 << inode->i_blkbits);
794 * block_start is block-aligned. Bump it by one to force
795 * __block_write_begin and block_commit_write to zero the
798 ret = __block_write_begin(page, block_start + 1, 0,
806 handle = ocfs2_zero_start_ordered_transaction(inode);
807 if (IS_ERR(handle)) {
808 ret = PTR_ERR(handle);
814 /* must not update i_size! */
815 ret = block_commit_write(page, block_start + 1,
824 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
828 page_cache_release(page);
834 * Find the next range to zero. We do this in terms of bytes because
835 * that's what ocfs2_zero_extend() wants, and it is dealing with the
836 * pagecache. We may return multiple extents.
838 * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
839 * needs to be zeroed. range_start and range_end return the next zeroing
840 * range. A subsequent call should pass the previous range_end as its
841 * zero_start. If range_end is 0, there's nothing to do.
843 * Unwritten extents are skipped over. Refcounted extents are CoWd.
845 static int ocfs2_zero_extend_get_range(struct inode *inode,
846 struct buffer_head *di_bh,
847 u64 zero_start, u64 zero_end,
848 u64 *range_start, u64 *range_end)
850 int rc = 0, needs_cow = 0;
851 u32 p_cpos, zero_clusters = 0;
853 zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
854 u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
855 unsigned int num_clusters = 0;
856 unsigned int ext_flags = 0;
858 while (zero_cpos < last_cpos) {
859 rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
860 &num_clusters, &ext_flags);
866 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
867 zero_clusters = num_clusters;
868 if (ext_flags & OCFS2_EXT_REFCOUNTED)
873 zero_cpos += num_clusters;
875 if (!zero_clusters) {
880 while ((zero_cpos + zero_clusters) < last_cpos) {
881 rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
882 &p_cpos, &num_clusters,
889 if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
891 if (ext_flags & OCFS2_EXT_REFCOUNTED)
893 zero_clusters += num_clusters;
895 if ((zero_cpos + zero_clusters) > last_cpos)
896 zero_clusters = last_cpos - zero_cpos;
899 rc = ocfs2_refcount_cow(inode, NULL, di_bh, zero_cpos,
900 zero_clusters, UINT_MAX);
907 *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
908 *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
909 zero_cpos + zero_clusters);
916 * Zero one range returned from ocfs2_zero_extend_get_range(). The caller
917 * has made sure that the entire range needs zeroing.
919 static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
924 u64 zero_pos = range_start;
926 mlog(0, "range_start = %llu, range_end = %llu\n",
927 (unsigned long long)range_start,
928 (unsigned long long)range_end);
929 BUG_ON(range_start >= range_end);
931 while (zero_pos < range_end) {
932 next_pos = (zero_pos & PAGE_CACHE_MASK) + PAGE_CACHE_SIZE;
933 if (next_pos > range_end)
934 next_pos = range_end;
935 rc = ocfs2_write_zero_page(inode, zero_pos, next_pos);
943 * Very large extends have the potential to lock up
944 * the cpu for extended periods of time.
952 int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
956 u64 zero_start, range_start = 0, range_end = 0;
957 struct super_block *sb = inode->i_sb;
959 zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
960 mlog(0, "zero_start %llu for i_size %llu\n",
961 (unsigned long long)zero_start,
962 (unsigned long long)i_size_read(inode));
963 while (zero_start < zero_to_size) {
964 ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
975 if (range_start < zero_start)
976 range_start = zero_start;
977 if (range_end > zero_to_size)
978 range_end = zero_to_size;
980 ret = ocfs2_zero_extend_range(inode, range_start,
986 zero_start = range_end;
992 int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
993 u64 new_i_size, u64 zero_to)
997 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1000 * Only quota files call this without a bh, and they can't be
1003 BUG_ON(!di_bh && (oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
1004 BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1006 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1007 if (clusters_to_add < oi->ip_clusters)
1008 clusters_to_add = 0;
1010 clusters_to_add -= oi->ip_clusters;
1012 if (clusters_to_add) {
1013 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
1014 clusters_to_add, 0);
1022 * Call this even if we don't add any clusters to the tree. We
1023 * still need to zero the area between the old i_size and the
1026 ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1034 static int ocfs2_extend_file(struct inode *inode,
1035 struct buffer_head *di_bh,
1039 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1043 /* setattr sometimes calls us like this. */
1044 if (new_i_size == 0)
1047 if (i_size_read(inode) == new_i_size)
1049 BUG_ON(new_i_size < i_size_read(inode));
1052 * The alloc sem blocks people in read/write from reading our
1053 * allocation until we're done changing it. We depend on
1054 * i_mutex to block other extend/truncate calls while we're
1055 * here. We even have to hold it for sparse files because there
1056 * might be some tail zeroing.
1058 down_write(&oi->ip_alloc_sem);
1060 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1062 * We can optimize small extends by keeping the inodes
1065 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1066 up_write(&oi->ip_alloc_sem);
1067 goto out_update_size;
1070 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1072 up_write(&oi->ip_alloc_sem);
1078 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1079 ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1081 ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1084 up_write(&oi->ip_alloc_sem);
1092 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1100 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1102 int status = 0, size_change;
1103 struct inode *inode = dentry->d_inode;
1104 struct super_block *sb = inode->i_sb;
1105 struct ocfs2_super *osb = OCFS2_SB(sb);
1106 struct buffer_head *bh = NULL;
1107 handle_t *handle = NULL;
1108 struct dquot *transfer_to[MAXQUOTAS] = { };
1111 mlog(0, "(0x%p, '%.*s')\n", dentry,
1112 dentry->d_name.len, dentry->d_name.name);
1114 /* ensuring we don't even attempt to truncate a symlink */
1115 if (S_ISLNK(inode->i_mode))
1116 attr->ia_valid &= ~ATTR_SIZE;
1118 if (attr->ia_valid & ATTR_MODE)
1119 mlog(0, "mode change: %d\n", attr->ia_mode);
1120 if (attr->ia_valid & ATTR_UID)
1121 mlog(0, "uid change: %d\n", attr->ia_uid);
1122 if (attr->ia_valid & ATTR_GID)
1123 mlog(0, "gid change: %d\n", attr->ia_gid);
1124 if (attr->ia_valid & ATTR_SIZE)
1125 mlog(0, "size change...\n");
1126 if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME))
1127 mlog(0, "time change...\n");
1129 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1130 | ATTR_GID | ATTR_UID | ATTR_MODE)
1131 if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) {
1132 mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid);
1136 status = inode_change_ok(inode, attr);
1140 if (is_quota_modification(inode, attr))
1141 dquot_initialize(inode);
1142 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1144 status = ocfs2_rw_lock(inode, 1);
1151 status = ocfs2_inode_lock(inode, &bh, 1);
1153 if (status != -ENOENT)
1155 goto bail_unlock_rw;
1158 if (size_change && attr->ia_size != i_size_read(inode)) {
1159 status = inode_newsize_ok(inode, attr->ia_size);
1163 if (i_size_read(inode) > attr->ia_size) {
1164 if (ocfs2_should_order_data(inode)) {
1165 status = ocfs2_begin_ordered_truncate(inode,
1170 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1172 status = ocfs2_extend_file(inode, bh, attr->ia_size);
1174 if (status != -ENOSPC)
1181 if ((attr->ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
1182 (attr->ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
1184 * Gather pointers to quota structures so that allocation /
1185 * freeing of quota structures happens here and not inside
1186 * dquot_transfer() where we have problems with lock ordering
1188 if (attr->ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid
1189 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1190 OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1191 transfer_to[USRQUOTA] = dqget(sb, attr->ia_uid,
1193 if (!transfer_to[USRQUOTA]) {
1198 if (attr->ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid
1199 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1200 OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1201 transfer_to[GRPQUOTA] = dqget(sb, attr->ia_gid,
1203 if (!transfer_to[GRPQUOTA]) {
1208 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1209 2 * ocfs2_quota_trans_credits(sb));
1210 if (IS_ERR(handle)) {
1211 status = PTR_ERR(handle);
1215 status = __dquot_transfer(inode, transfer_to);
1219 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1220 if (IS_ERR(handle)) {
1221 status = PTR_ERR(handle);
1228 * This will intentionally not wind up calling truncate_setsize(),
1229 * since all the work for a size change has been done above.
1230 * Otherwise, we could get into problems with truncate as
1231 * ip_alloc_sem is used there to protect against i_size
1234 * XXX: this means the conditional below can probably be removed.
1236 if ((attr->ia_valid & ATTR_SIZE) &&
1237 attr->ia_size != i_size_read(inode)) {
1238 status = vmtruncate(inode, attr->ia_size);
1245 setattr_copy(inode, attr);
1246 mark_inode_dirty(inode);
1248 status = ocfs2_mark_inode_dirty(handle, inode, bh);
1253 ocfs2_commit_trans(osb, handle);
1255 ocfs2_inode_unlock(inode, 1);
1258 ocfs2_rw_unlock(inode, 1);
1262 /* Release quota pointers in case we acquired them */
1263 for (qtype = 0; qtype < MAXQUOTAS; qtype++)
1264 dqput(transfer_to[qtype]);
1266 if (!status && attr->ia_valid & ATTR_MODE) {
1267 status = ocfs2_acl_chmod(inode);
1276 int ocfs2_getattr(struct vfsmount *mnt,
1277 struct dentry *dentry,
1280 struct inode *inode = dentry->d_inode;
1281 struct super_block *sb = dentry->d_inode->i_sb;
1282 struct ocfs2_super *osb = sb->s_fs_info;
1285 err = ocfs2_inode_revalidate(dentry);
1292 generic_fillattr(inode, stat);
1294 /* We set the blksize from the cluster size for performance */
1295 stat->blksize = osb->s_clustersize;
1303 int ocfs2_permission(struct inode *inode, int mask, unsigned int flags)
1307 if (flags & IPERM_FLAG_RCU)
1310 ret = ocfs2_inode_lock(inode, NULL, 0);
1317 ret = generic_permission(inode, mask, flags, ocfs2_check_acl);
1319 ocfs2_inode_unlock(inode, 0);
1325 static int __ocfs2_write_remove_suid(struct inode *inode,
1326 struct buffer_head *bh)
1330 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1331 struct ocfs2_dinode *di;
1333 mlog(0, "(Inode %llu, mode 0%o)\n",
1334 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_mode);
1336 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1337 if (IS_ERR(handle)) {
1338 ret = PTR_ERR(handle);
1343 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1344 OCFS2_JOURNAL_ACCESS_WRITE);
1350 inode->i_mode &= ~S_ISUID;
1351 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1352 inode->i_mode &= ~S_ISGID;
1354 di = (struct ocfs2_dinode *) bh->b_data;
1355 di->i_mode = cpu_to_le16(inode->i_mode);
1357 ocfs2_journal_dirty(handle, bh);
1360 ocfs2_commit_trans(osb, handle);
1367 * Will look for holes and unwritten extents in the range starting at
1368 * pos for count bytes (inclusive).
1370 static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1374 unsigned int extent_flags;
1375 u32 cpos, clusters, extent_len, phys_cpos;
1376 struct super_block *sb = inode->i_sb;
1378 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1379 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1382 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1389 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
1394 if (extent_len > clusters)
1395 extent_len = clusters;
1397 clusters -= extent_len;
1404 static int ocfs2_write_remove_suid(struct inode *inode)
1407 struct buffer_head *bh = NULL;
1409 ret = ocfs2_read_inode_block(inode, &bh);
1415 ret = __ocfs2_write_remove_suid(inode, bh);
1422 * Allocate enough extents to cover the region starting at byte offset
1423 * start for len bytes. Existing extents are skipped, any extents
1424 * added are marked as "unwritten".
1426 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1430 u32 cpos, phys_cpos, clusters, alloc_size;
1431 u64 end = start + len;
1432 struct buffer_head *di_bh = NULL;
1434 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1435 ret = ocfs2_read_inode_block(inode, &di_bh);
1442 * Nothing to do if the requested reservation range
1443 * fits within the inode.
1445 if (ocfs2_size_fits_inline_data(di_bh, end))
1448 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1456 * We consider both start and len to be inclusive.
1458 cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1459 clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1463 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1471 * Hole or existing extent len can be arbitrary, so
1472 * cap it to our own allocation request.
1474 if (alloc_size > clusters)
1475 alloc_size = clusters;
1479 * We already have an allocation at this
1480 * region so we can safely skip it.
1485 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1494 clusters -= alloc_size;
1505 * Truncate a byte range, avoiding pages within partial clusters. This
1506 * preserves those pages for the zeroing code to write to.
1508 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1511 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1513 struct address_space *mapping = inode->i_mapping;
1515 start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1516 end = byte_start + byte_len;
1517 end = end & ~(osb->s_clustersize - 1);
1520 unmap_mapping_range(mapping, start, end - start, 0);
1521 truncate_inode_pages_range(mapping, start, end - 1);
1525 static int ocfs2_zero_partial_clusters(struct inode *inode,
1529 u64 tmpend, end = start + len;
1530 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1531 unsigned int csize = osb->s_clustersize;
1535 * The "start" and "end" values are NOT necessarily part of
1536 * the range whose allocation is being deleted. Rather, this
1537 * is what the user passed in with the request. We must zero
1538 * partial clusters here. There's no need to worry about
1539 * physical allocation - the zeroing code knows to skip holes.
1541 mlog(0, "byte start: %llu, end: %llu\n",
1542 (unsigned long long)start, (unsigned long long)end);
1545 * If both edges are on a cluster boundary then there's no
1546 * zeroing required as the region is part of the allocation to
1549 if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1552 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1553 if (IS_ERR(handle)) {
1554 ret = PTR_ERR(handle);
1560 * We want to get the byte offset of the end of the 1st cluster.
1562 tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1566 mlog(0, "1st range: start: %llu, tmpend: %llu\n",
1567 (unsigned long long)start, (unsigned long long)tmpend);
1569 ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1575 * This may make start and end equal, but the zeroing
1576 * code will skip any work in that case so there's no
1577 * need to catch it up here.
1579 start = end & ~(osb->s_clustersize - 1);
1581 mlog(0, "2nd range: start: %llu, end: %llu\n",
1582 (unsigned long long)start, (unsigned long long)end);
1584 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1589 ocfs2_commit_trans(osb, handle);
1594 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1597 struct ocfs2_extent_rec *rec = NULL;
1599 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1601 rec = &el->l_recs[i];
1603 if (le32_to_cpu(rec->e_cpos) < pos)
1611 * Helper to calculate the punching pos and length in one run, we handle the
1612 * following three cases in order:
1614 * - remove the entire record
1615 * - remove a partial record
1616 * - no record needs to be removed (hole-punching completed)
1618 static void ocfs2_calc_trunc_pos(struct inode *inode,
1619 struct ocfs2_extent_list *el,
1620 struct ocfs2_extent_rec *rec,
1621 u32 trunc_start, u32 *trunc_cpos,
1622 u32 *trunc_len, u32 *trunc_end,
1623 u64 *blkno, int *done)
1628 range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1630 if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1631 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1633 * Skip holes if any.
1635 if (range < *trunc_end)
1637 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1638 *blkno = le64_to_cpu(rec->e_blkno);
1639 *trunc_end = le32_to_cpu(rec->e_cpos);
1640 } else if (range > trunc_start) {
1641 *trunc_cpos = trunc_start;
1642 *trunc_len = *trunc_end - trunc_start;
1643 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1644 *blkno = le64_to_cpu(rec->e_blkno) +
1645 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1646 *trunc_end = trunc_start;
1649 * It may have two following possibilities:
1651 * - last record has been removed
1652 * - trunc_start was within a hole
1654 * both two cases mean the completion of hole punching.
1662 static int ocfs2_remove_inode_range(struct inode *inode,
1663 struct buffer_head *di_bh, u64 byte_start,
1666 int ret = 0, flags = 0, done = 0, i;
1667 u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1669 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1670 struct ocfs2_cached_dealloc_ctxt dealloc;
1671 struct address_space *mapping = inode->i_mapping;
1672 struct ocfs2_extent_tree et;
1673 struct ocfs2_path *path = NULL;
1674 struct ocfs2_extent_list *el = NULL;
1675 struct ocfs2_extent_rec *rec = NULL;
1676 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1677 u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1679 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1680 ocfs2_init_dealloc_ctxt(&dealloc);
1685 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1686 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1687 byte_start + byte_len, 0);
1693 * There's no need to get fancy with the page cache
1694 * truncate of an inline-data inode. We're talking
1695 * about less than a page here, which will be cached
1696 * in the dinode buffer anyway.
1698 unmap_mapping_range(mapping, 0, 0, 0);
1699 truncate_inode_pages(mapping, 0);
1704 * For reflinks, we may need to CoW 2 clusters which might be
1705 * partially zero'd later, if hole's start and end offset were
1706 * within one cluster(means is not exactly aligned to clustersize).
1709 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
1711 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1717 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1724 trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1725 trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1726 cluster_in_el = trunc_end;
1728 mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, cend: %u\n",
1729 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1730 (unsigned long long)byte_start,
1731 (unsigned long long)byte_len, trunc_start, trunc_end);
1733 ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1739 path = ocfs2_new_path_from_et(&et);
1746 while (trunc_end > trunc_start) {
1748 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1755 el = path_leaf_el(path);
1757 i = ocfs2_find_rec(el, trunc_end);
1759 * Need to go to previous extent block.
1762 if (path->p_tree_depth == 0)
1765 ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1774 * We've reached the leftmost extent block,
1775 * it's safe to leave.
1777 if (cluster_in_el == 0)
1781 * The 'pos' searched for previous extent block is
1782 * always one cluster less than actual trunc_end.
1784 trunc_end = cluster_in_el + 1;
1786 ocfs2_reinit_path(path, 1);
1791 rec = &el->l_recs[i];
1793 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1794 &trunc_len, &trunc_end, &blkno, &done);
1798 flags = rec->e_flags;
1799 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1801 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1802 phys_cpos, trunc_len, flags,
1803 &dealloc, refcount_loc);
1809 cluster_in_el = trunc_end;
1811 ocfs2_reinit_path(path, 1);
1814 ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1817 ocfs2_schedule_truncate_log_flush(osb, 1);
1818 ocfs2_run_deallocs(osb, &dealloc);
1824 * Parts of this function taken from xfs_change_file_space()
1826 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1827 loff_t f_pos, unsigned int cmd,
1828 struct ocfs2_space_resv *sr,
1834 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1835 struct buffer_head *di_bh = NULL;
1837 unsigned long long max_off = inode->i_sb->s_maxbytes;
1839 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1842 mutex_lock(&inode->i_mutex);
1845 * This prevents concurrent writes on other nodes
1847 ret = ocfs2_rw_lock(inode, 1);
1853 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1859 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1861 goto out_inode_unlock;
1864 switch (sr->l_whence) {
1865 case 0: /*SEEK_SET*/
1867 case 1: /*SEEK_CUR*/
1868 sr->l_start += f_pos;
1870 case 2: /*SEEK_END*/
1871 sr->l_start += i_size_read(inode);
1875 goto out_inode_unlock;
1879 llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1882 || sr->l_start > max_off
1883 || (sr->l_start + llen) < 0
1884 || (sr->l_start + llen) > max_off) {
1886 goto out_inode_unlock;
1888 size = sr->l_start + sr->l_len;
1890 if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) {
1891 if (sr->l_len <= 0) {
1893 goto out_inode_unlock;
1897 if (file && should_remove_suid(file->f_path.dentry)) {
1898 ret = __ocfs2_write_remove_suid(inode, di_bh);
1901 goto out_inode_unlock;
1905 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1907 case OCFS2_IOC_RESVSP:
1908 case OCFS2_IOC_RESVSP64:
1910 * This takes unsigned offsets, but the signed ones we
1911 * pass have been checked against overflow above.
1913 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1916 case OCFS2_IOC_UNRESVSP:
1917 case OCFS2_IOC_UNRESVSP64:
1918 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1924 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1927 goto out_inode_unlock;
1931 * We update c/mtime for these changes
1933 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1934 if (IS_ERR(handle)) {
1935 ret = PTR_ERR(handle);
1937 goto out_inode_unlock;
1940 if (change_size && i_size_read(inode) < size)
1941 i_size_write(inode, size);
1943 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1944 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1948 ocfs2_commit_trans(osb, handle);
1952 ocfs2_inode_unlock(inode, 1);
1954 ocfs2_rw_unlock(inode, 1);
1957 mutex_unlock(&inode->i_mutex);
1961 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1962 struct ocfs2_space_resv *sr)
1964 struct inode *inode = file->f_path.dentry->d_inode;
1965 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1967 if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1968 !ocfs2_writes_unwritten_extents(osb))
1970 else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1971 !ocfs2_sparse_alloc(osb))
1974 if (!S_ISREG(inode->i_mode))
1977 if (!(file->f_mode & FMODE_WRITE))
1980 return __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1983 static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
1986 struct inode *inode = file->f_path.dentry->d_inode;
1987 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1988 struct ocfs2_space_resv sr;
1989 int change_size = 1;
1990 int cmd = OCFS2_IOC_RESVSP64;
1992 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
1994 if (!ocfs2_writes_unwritten_extents(osb))
1997 if (mode & FALLOC_FL_KEEP_SIZE)
2000 if (mode & FALLOC_FL_PUNCH_HOLE)
2001 cmd = OCFS2_IOC_UNRESVSP64;
2004 sr.l_start = (s64)offset;
2005 sr.l_len = (s64)len;
2007 return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2011 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2015 unsigned int extent_flags;
2016 u32 cpos, clusters, extent_len, phys_cpos;
2017 struct super_block *sb = inode->i_sb;
2019 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2020 !(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) ||
2021 OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2024 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2025 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2028 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2035 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2040 if (extent_len > clusters)
2041 extent_len = clusters;
2043 clusters -= extent_len;
2050 static int ocfs2_prepare_inode_for_refcount(struct inode *inode,
2052 loff_t pos, size_t count,
2056 struct buffer_head *di_bh = NULL;
2057 u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2059 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2061 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2069 ret = ocfs2_refcount_cow(inode, file, di_bh, cpos, clusters, UINT_MAX);
2077 static int ocfs2_prepare_inode_for_write(struct file *file,
2084 int ret = 0, meta_level = 0;
2085 struct dentry *dentry = file->f_path.dentry;
2086 struct inode *inode = dentry->d_inode;
2087 loff_t saved_pos, end;
2090 * We start with a read level meta lock and only jump to an ex
2091 * if we need to make modifications here.
2094 ret = ocfs2_inode_lock(inode, NULL, meta_level);
2101 /* Clear suid / sgid if necessary. We do this here
2102 * instead of later in the write path because
2103 * remove_suid() calls ->setattr without any hint that
2104 * we may have already done our cluster locking. Since
2105 * ocfs2_setattr() *must* take cluster locks to
2106 * proceeed, this will lead us to recursively lock the
2107 * inode. There's also the dinode i_size state which
2108 * can be lost via setattr during extending writes (we
2109 * set inode->i_size at the end of a write. */
2110 if (should_remove_suid(dentry)) {
2111 if (meta_level == 0) {
2112 ocfs2_inode_unlock(inode, meta_level);
2117 ret = ocfs2_write_remove_suid(inode);
2124 /* work on a copy of ppos until we're sure that we won't have
2125 * to recalculate it due to relocking. */
2127 saved_pos = i_size_read(inode);
2128 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos);
2133 end = saved_pos + count;
2135 ret = ocfs2_check_range_for_refcount(inode, saved_pos, count);
2137 ocfs2_inode_unlock(inode, meta_level);
2140 ret = ocfs2_prepare_inode_for_refcount(inode,
2157 * Skip the O_DIRECT checks if we don't need
2160 if (!direct_io || !(*direct_io))
2164 * There's no sane way to do direct writes to an inode
2167 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2173 * Allowing concurrent direct writes means
2174 * i_size changes wouldn't be synchronized, so
2175 * one node could wind up truncating another
2178 if (end > i_size_read(inode)) {
2184 * We don't fill holes during direct io, so
2185 * check for them here. If any are found, the
2186 * caller will have to retake some cluster
2187 * locks and initiate the io as buffered.
2189 ret = ocfs2_check_range_for_holes(inode, saved_pos, count);
2202 if (meta_level >= 0)
2203 ocfs2_inode_unlock(inode, meta_level);
2209 static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
2210 const struct iovec *iov,
2211 unsigned long nr_segs,
2214 int ret, direct_io, appending, rw_level, have_alloc_sem = 0;
2215 int can_do_direct, has_refcount = 0;
2216 ssize_t written = 0;
2217 size_t ocount; /* original count */
2218 size_t count; /* after file limit checks */
2219 loff_t old_size, *ppos = &iocb->ki_pos;
2221 struct file *file = iocb->ki_filp;
2222 struct inode *inode = file->f_path.dentry->d_inode;
2223 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2224 int full_coherency = !(osb->s_mount_opt &
2225 OCFS2_MOUNT_COHERENCY_BUFFERED);
2227 mlog(0, "(0x%p, %u, '%.*s')\n", file,
2228 (unsigned int)nr_segs,
2229 file->f_path.dentry->d_name.len,
2230 file->f_path.dentry->d_name.name);
2232 if (iocb->ki_left == 0)
2235 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2237 appending = file->f_flags & O_APPEND ? 1 : 0;
2238 direct_io = file->f_flags & O_DIRECT ? 1 : 0;
2240 mutex_lock(&inode->i_mutex);
2242 ocfs2_iocb_clear_sem_locked(iocb);
2245 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
2247 down_read(&inode->i_alloc_sem);
2249 /* communicate with ocfs2_dio_end_io */
2250 ocfs2_iocb_set_sem_locked(iocb);
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);
2387 if (have_alloc_sem) {
2388 up_read(&inode->i_alloc_sem);
2389 ocfs2_iocb_clear_sem_locked(iocb);
2392 mutex_unlock(&inode->i_mutex);
2400 static int ocfs2_splice_to_file(struct pipe_inode_info *pipe,
2402 struct splice_desc *sd)
2406 ret = ocfs2_prepare_inode_for_write(out, &sd->pos,
2407 sd->total_len, 0, NULL, NULL);
2413 return splice_from_pipe_feed(pipe, sd, pipe_to_file);
2416 static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
2423 struct address_space *mapping = out->f_mapping;
2424 struct inode *inode = mapping->host;
2425 struct splice_desc sd = {
2432 mlog(0, "(0x%p, 0x%p, %u, '%.*s')\n", out, pipe,
2434 out->f_path.dentry->d_name.len,
2435 out->f_path.dentry->d_name.name);
2438 mutex_lock_nested(&pipe->inode->i_mutex, I_MUTEX_PARENT);
2440 splice_from_pipe_begin(&sd);
2442 ret = splice_from_pipe_next(pipe, &sd);
2446 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2447 ret = ocfs2_rw_lock(inode, 1);
2451 ret = ocfs2_splice_to_file(pipe, out, &sd);
2452 ocfs2_rw_unlock(inode, 1);
2454 mutex_unlock(&inode->i_mutex);
2456 splice_from_pipe_end(pipe, &sd);
2459 mutex_unlock(&pipe->inode->i_mutex);
2462 ret = sd.num_spliced;
2465 unsigned long nr_pages;
2468 nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
2470 err = generic_write_sync(out, *ppos, ret);
2476 balance_dirty_pages_ratelimited_nr(mapping, nr_pages);
2483 static ssize_t ocfs2_file_splice_read(struct file *in,
2485 struct pipe_inode_info *pipe,
2489 int ret = 0, lock_level = 0;
2490 struct inode *inode = in->f_path.dentry->d_inode;
2492 mlog(0, "(0x%p, 0x%p, %u, '%.*s')\n", in, pipe,
2494 in->f_path.dentry->d_name.len,
2495 in->f_path.dentry->d_name.name);
2498 * See the comment in ocfs2_file_aio_read()
2500 ret = ocfs2_inode_lock_atime(inode, in->f_vfsmnt, &lock_level);
2505 ocfs2_inode_unlock(inode, lock_level);
2507 ret = generic_file_splice_read(in, ppos, pipe, len, flags);
2514 static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
2515 const struct iovec *iov,
2516 unsigned long nr_segs,
2519 int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
2520 struct file *filp = iocb->ki_filp;
2521 struct inode *inode = filp->f_path.dentry->d_inode;
2523 mlog(0, "(0x%p, %u, '%.*s')\n", filp,
2524 (unsigned int)nr_segs,
2525 filp->f_path.dentry->d_name.len,
2526 filp->f_path.dentry->d_name.name);
2534 ocfs2_iocb_clear_sem_locked(iocb);
2537 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
2538 * need locks to protect pending reads from racing with truncate.
2540 if (filp->f_flags & O_DIRECT) {
2541 down_read(&inode->i_alloc_sem);
2543 ocfs2_iocb_set_sem_locked(iocb);
2545 ret = ocfs2_rw_lock(inode, 0);
2551 /* communicate with ocfs2_dio_end_io */
2552 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2556 * We're fine letting folks race truncates and extending
2557 * writes with read across the cluster, just like they can
2558 * locally. Hence no rw_lock during read.
2560 * Take and drop the meta data lock to update inode fields
2561 * like i_size. This allows the checks down below
2562 * generic_file_aio_read() a chance of actually working.
2564 ret = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level);
2569 ocfs2_inode_unlock(inode, lock_level);
2571 ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
2573 mlog(0, "generic_file_aio_read returned -EINVAL\n");
2575 /* buffered aio wouldn't have proper lock coverage today */
2576 BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
2578 /* see ocfs2_file_aio_write */
2579 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2585 if (have_alloc_sem) {
2586 up_read(&inode->i_alloc_sem);
2587 ocfs2_iocb_clear_sem_locked(iocb);
2590 ocfs2_rw_unlock(inode, rw_level);
2596 const struct inode_operations ocfs2_file_iops = {
2597 .setattr = ocfs2_setattr,
2598 .getattr = ocfs2_getattr,
2599 .permission = ocfs2_permission,
2600 .setxattr = generic_setxattr,
2601 .getxattr = generic_getxattr,
2602 .listxattr = ocfs2_listxattr,
2603 .removexattr = generic_removexattr,
2604 .fiemap = ocfs2_fiemap,
2607 const struct inode_operations ocfs2_special_file_iops = {
2608 .setattr = ocfs2_setattr,
2609 .getattr = ocfs2_getattr,
2610 .permission = ocfs2_permission,
2614 * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2615 * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2617 const struct file_operations ocfs2_fops = {
2618 .llseek = generic_file_llseek,
2619 .read = do_sync_read,
2620 .write = do_sync_write,
2622 .fsync = ocfs2_sync_file,
2623 .release = ocfs2_file_release,
2624 .open = ocfs2_file_open,
2625 .aio_read = ocfs2_file_aio_read,
2626 .aio_write = ocfs2_file_aio_write,
2627 .unlocked_ioctl = ocfs2_ioctl,
2628 #ifdef CONFIG_COMPAT
2629 .compat_ioctl = ocfs2_compat_ioctl,
2632 .flock = ocfs2_flock,
2633 .splice_read = ocfs2_file_splice_read,
2634 .splice_write = ocfs2_file_splice_write,
2635 .fallocate = ocfs2_fallocate,
2638 const struct file_operations ocfs2_dops = {
2639 .llseek = generic_file_llseek,
2640 .read = generic_read_dir,
2641 .readdir = ocfs2_readdir,
2642 .fsync = ocfs2_sync_file,
2643 .release = ocfs2_dir_release,
2644 .open = ocfs2_dir_open,
2645 .unlocked_ioctl = ocfs2_ioctl,
2646 #ifdef CONFIG_COMPAT
2647 .compat_ioctl = ocfs2_compat_ioctl,
2650 .flock = ocfs2_flock,
2654 * POSIX-lockless variants of our file_operations.
2656 * These will be used if the underlying cluster stack does not support
2657 * posix file locking, if the user passes the "localflocks" mount
2658 * option, or if we have a local-only fs.
2660 * ocfs2_flock is in here because all stacks handle UNIX file locks,
2661 * so we still want it in the case of no stack support for
2662 * plocks. Internally, it will do the right thing when asked to ignore
2665 const struct file_operations ocfs2_fops_no_plocks = {
2666 .llseek = generic_file_llseek,
2667 .read = do_sync_read,
2668 .write = do_sync_write,
2670 .fsync = ocfs2_sync_file,
2671 .release = ocfs2_file_release,
2672 .open = ocfs2_file_open,
2673 .aio_read = ocfs2_file_aio_read,
2674 .aio_write = ocfs2_file_aio_write,
2675 .unlocked_ioctl = ocfs2_ioctl,
2676 #ifdef CONFIG_COMPAT
2677 .compat_ioctl = ocfs2_compat_ioctl,
2679 .flock = ocfs2_flock,
2680 .splice_read = ocfs2_file_splice_read,
2681 .splice_write = ocfs2_file_splice_write,
2684 const struct file_operations ocfs2_dops_no_plocks = {
2685 .llseek = generic_file_llseek,
2686 .read = generic_read_dir,
2687 .readdir = ocfs2_readdir,
2688 .fsync = ocfs2_sync_file,
2689 .release = ocfs2_dir_release,
2690 .open = ocfs2_dir_open,
2691 .unlocked_ioctl = ocfs2_ioctl,
2692 #ifdef CONFIG_COMPAT
2693 .compat_ioctl = ocfs2_compat_ioctl,
2695 .flock = ocfs2_flock,