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>
39 #define MLOG_MASK_PREFIX ML_INODE
40 #include <cluster/masklog.h>
48 #include "extent_map.h"
60 #include "buffer_head_io.h"
62 static int ocfs2_sync_inode(struct inode *inode)
64 filemap_fdatawrite(inode->i_mapping);
65 return sync_mapping_buffers(inode->i_mapping);
68 static int ocfs2_init_file_private(struct inode *inode, struct file *file)
70 struct ocfs2_file_private *fp;
72 fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
77 mutex_init(&fp->fp_mutex);
78 ocfs2_file_lock_res_init(&fp->fp_flock, fp);
79 file->private_data = fp;
84 static void ocfs2_free_file_private(struct inode *inode, struct file *file)
86 struct ocfs2_file_private *fp = file->private_data;
87 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
90 ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
91 ocfs2_lock_res_free(&fp->fp_flock);
93 file->private_data = NULL;
97 static int ocfs2_file_open(struct inode *inode, struct file *file)
100 int mode = file->f_flags;
101 struct ocfs2_inode_info *oi = OCFS2_I(inode);
103 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
104 file->f_path.dentry->d_name.len, file->f_path.dentry->d_name.name);
106 spin_lock(&oi->ip_lock);
108 /* Check that the inode hasn't been wiped from disk by another
109 * node. If it hasn't then we're safe as long as we hold the
110 * spin lock until our increment of open count. */
111 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
112 spin_unlock(&oi->ip_lock);
119 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
122 spin_unlock(&oi->ip_lock);
124 status = ocfs2_init_file_private(inode, file);
127 * We want to set open count back if we're failing the
130 spin_lock(&oi->ip_lock);
132 spin_unlock(&oi->ip_lock);
140 static int ocfs2_file_release(struct inode *inode, struct file *file)
142 struct ocfs2_inode_info *oi = OCFS2_I(inode);
144 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
145 file->f_path.dentry->d_name.len,
146 file->f_path.dentry->d_name.name);
148 spin_lock(&oi->ip_lock);
149 if (!--oi->ip_open_count)
150 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
151 spin_unlock(&oi->ip_lock);
153 ocfs2_free_file_private(inode, file);
160 static int ocfs2_dir_open(struct inode *inode, struct file *file)
162 return ocfs2_init_file_private(inode, file);
165 static int ocfs2_dir_release(struct inode *inode, struct file *file)
167 ocfs2_free_file_private(inode, file);
171 static int ocfs2_sync_file(struct file *file,
172 struct dentry *dentry,
177 struct inode *inode = dentry->d_inode;
178 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
180 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync,
181 dentry->d_name.len, dentry->d_name.name);
183 err = ocfs2_sync_inode(dentry->d_inode);
187 journal = osb->journal->j_journal;
188 err = jbd2_journal_force_commit(journal);
193 return (err < 0) ? -EIO : 0;
196 int ocfs2_should_update_atime(struct inode *inode,
197 struct vfsmount *vfsmnt)
200 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
202 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
205 if ((inode->i_flags & S_NOATIME) ||
206 ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
210 * We can be called with no vfsmnt structure - NFSD will
213 * Note that our action here is different than touch_atime() -
214 * if we can't tell whether this is a noatime mount, then we
215 * don't know whether to trust the value of s_atime_quantum.
220 if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
221 ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
224 if (vfsmnt->mnt_flags & MNT_RELATIME) {
225 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
226 (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
233 if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
239 int ocfs2_update_inode_atime(struct inode *inode,
240 struct buffer_head *bh)
243 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
245 struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
249 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
250 if (IS_ERR(handle)) {
251 ret = PTR_ERR(handle);
256 ret = ocfs2_journal_access(handle, inode, bh,
257 OCFS2_JOURNAL_ACCESS_WRITE);
264 * Don't use ocfs2_mark_inode_dirty() here as we don't always
265 * have i_mutex to guard against concurrent changes to other
268 inode->i_atime = CURRENT_TIME;
269 di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
270 di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
272 ret = ocfs2_journal_dirty(handle, bh);
277 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
283 static int ocfs2_set_inode_size(handle_t *handle,
285 struct buffer_head *fe_bh,
291 i_size_write(inode, new_i_size);
292 inode->i_blocks = ocfs2_inode_sector_count(inode);
293 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
295 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
306 static int ocfs2_simple_size_update(struct inode *inode,
307 struct buffer_head *di_bh,
311 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
312 handle_t *handle = NULL;
314 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
315 if (IS_ERR(handle)) {
316 ret = PTR_ERR(handle);
321 ret = ocfs2_set_inode_size(handle, inode, di_bh,
326 ocfs2_commit_trans(osb, handle);
331 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
333 struct buffer_head *fe_bh,
338 struct ocfs2_dinode *di;
343 /* TODO: This needs to actually orphan the inode in this
346 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
347 if (IS_ERR(handle)) {
348 status = PTR_ERR(handle);
353 status = ocfs2_journal_access(handle, inode, fe_bh,
354 OCFS2_JOURNAL_ACCESS_WRITE);
361 * Do this before setting i_size.
363 cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
364 status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
371 i_size_write(inode, new_i_size);
372 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
374 di = (struct ocfs2_dinode *) fe_bh->b_data;
375 di->i_size = cpu_to_le64(new_i_size);
376 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
377 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
379 status = ocfs2_journal_dirty(handle, fe_bh);
384 ocfs2_commit_trans(osb, handle);
391 static int ocfs2_truncate_file(struct inode *inode,
392 struct buffer_head *di_bh,
396 struct ocfs2_dinode *fe = NULL;
397 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
398 struct ocfs2_truncate_context *tc = NULL;
400 mlog_entry("(inode = %llu, new_i_size = %llu\n",
401 (unsigned long long)OCFS2_I(inode)->ip_blkno,
402 (unsigned long long)new_i_size);
404 fe = (struct ocfs2_dinode *) di_bh->b_data;
405 if (!OCFS2_IS_VALID_DINODE(fe)) {
406 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
411 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
412 "Inode %llu, inode i_size = %lld != di "
413 "i_size = %llu, i_flags = 0x%x\n",
414 (unsigned long long)OCFS2_I(inode)->ip_blkno,
416 (unsigned long long)le64_to_cpu(fe->i_size),
417 le32_to_cpu(fe->i_flags));
419 if (new_i_size > le64_to_cpu(fe->i_size)) {
420 mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
421 (unsigned long long)le64_to_cpu(fe->i_size),
422 (unsigned long long)new_i_size);
428 mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
429 (unsigned long long)le64_to_cpu(fe->i_blkno),
430 (unsigned long long)le64_to_cpu(fe->i_size),
431 (unsigned long long)new_i_size);
433 /* lets handle the simple truncate cases before doing any more
434 * cluster locking. */
435 if (new_i_size == le64_to_cpu(fe->i_size))
438 down_write(&OCFS2_I(inode)->ip_alloc_sem);
441 * The inode lock forced other nodes to sync and drop their
442 * pages, which (correctly) happens even if we have a truncate
443 * without allocation change - ocfs2 cluster sizes can be much
444 * greater than page size, so we have to truncate them
447 unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
448 truncate_inode_pages(inode->i_mapping, new_i_size);
450 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
451 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
452 i_size_read(inode), 1);
456 goto bail_unlock_sem;
459 /* alright, we're going to need to do a full blown alloc size
460 * change. Orphan the inode so that recovery can complete the
461 * truncate if necessary. This does the task of marking
463 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
466 goto bail_unlock_sem;
469 status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc);
472 goto bail_unlock_sem;
475 status = ocfs2_commit_truncate(osb, inode, di_bh, tc);
478 goto bail_unlock_sem;
481 /* TODO: orphan dir cleanup here. */
483 up_write(&OCFS2_I(inode)->ip_alloc_sem);
492 * extend file allocation only here.
493 * we'll update all the disk stuff, and oip->alloc_size
495 * expect stuff to be locked, a transaction started and enough data /
496 * metadata reservations in the contexts.
498 * Will return -EAGAIN, and a reason if a restart is needed.
499 * If passed in, *reason will always be set, even in error.
501 int ocfs2_add_inode_data(struct ocfs2_super *osb,
506 struct buffer_head *fe_bh,
508 struct ocfs2_alloc_context *data_ac,
509 struct ocfs2_alloc_context *meta_ac,
510 enum ocfs2_alloc_restarted *reason_ret)
513 struct ocfs2_extent_tree et;
515 ocfs2_init_dinode_extent_tree(&et, inode, fe_bh);
516 ret = ocfs2_add_clusters_in_btree(osb, inode, logical_offset,
517 clusters_to_add, mark_unwritten,
519 data_ac, meta_ac, reason_ret);
524 static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
525 u32 clusters_to_add, int mark_unwritten)
528 int restart_func = 0;
531 struct buffer_head *bh = NULL;
532 struct ocfs2_dinode *fe = NULL;
533 handle_t *handle = NULL;
534 struct ocfs2_alloc_context *data_ac = NULL;
535 struct ocfs2_alloc_context *meta_ac = NULL;
536 enum ocfs2_alloc_restarted why;
537 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
538 struct ocfs2_extent_tree et;
540 mlog_entry("(clusters_to_add = %u)\n", clusters_to_add);
543 * This function only exists for file systems which don't
546 BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
548 status = ocfs2_read_block(inode, OCFS2_I(inode)->ip_blkno, &bh);
554 fe = (struct ocfs2_dinode *) bh->b_data;
555 if (!OCFS2_IS_VALID_DINODE(fe)) {
556 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
562 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
564 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, "
565 "clusters_to_add = %u\n",
566 (unsigned long long)OCFS2_I(inode)->ip_blkno,
567 (long long)i_size_read(inode), le32_to_cpu(fe->i_clusters),
569 ocfs2_init_dinode_extent_tree(&et, inode, bh);
570 status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
577 credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list,
579 handle = ocfs2_start_trans(osb, credits);
580 if (IS_ERR(handle)) {
581 status = PTR_ERR(handle);
587 restarted_transaction:
588 /* reserve a write to the file entry early on - that we if we
589 * run out of credits in the allocation path, we can still
591 status = ocfs2_journal_access(handle, inode, bh,
592 OCFS2_JOURNAL_ACCESS_WRITE);
598 prev_clusters = OCFS2_I(inode)->ip_clusters;
600 status = ocfs2_add_inode_data(osb,
610 if ((status < 0) && (status != -EAGAIN)) {
611 if (status != -ENOSPC)
616 status = ocfs2_journal_dirty(handle, bh);
622 spin_lock(&OCFS2_I(inode)->ip_lock);
623 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
624 spin_unlock(&OCFS2_I(inode)->ip_lock);
626 if (why != RESTART_NONE && clusters_to_add) {
627 if (why == RESTART_META) {
628 mlog(0, "restarting function.\n");
631 BUG_ON(why != RESTART_TRANS);
633 mlog(0, "restarting transaction.\n");
634 /* TODO: This can be more intelligent. */
635 credits = ocfs2_calc_extend_credits(osb->sb,
638 status = ocfs2_extend_trans(handle, credits);
640 /* handle still has to be committed at
646 goto restarted_transaction;
650 mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
651 le32_to_cpu(fe->i_clusters),
652 (unsigned long long)le64_to_cpu(fe->i_size));
653 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
654 OCFS2_I(inode)->ip_clusters, (long long)i_size_read(inode));
658 ocfs2_commit_trans(osb, handle);
662 ocfs2_free_alloc_context(data_ac);
666 ocfs2_free_alloc_context(meta_ac);
669 if ((!status) && restart_func) {
680 /* Some parts of this taken from generic_cont_expand, which turned out
681 * to be too fragile to do exactly what we need without us having to
682 * worry about recursive locking in ->write_begin() and ->write_end(). */
683 static int ocfs2_write_zero_page(struct inode *inode,
686 struct address_space *mapping = inode->i_mapping;
690 handle_t *handle = NULL;
693 offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */
694 /* ugh. in prepare/commit_write, if from==to==start of block, we
695 ** skip the prepare. make sure we never send an offset for the start
698 if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) {
701 index = size >> PAGE_CACHE_SHIFT;
703 page = grab_cache_page(mapping, index);
710 ret = ocfs2_prepare_write_nolock(inode, page, offset, offset);
716 if (ocfs2_should_order_data(inode)) {
717 handle = ocfs2_start_walk_page_trans(inode, page, offset,
719 if (IS_ERR(handle)) {
720 ret = PTR_ERR(handle);
726 /* must not update i_size! */
727 ret = block_commit_write(page, offset, offset);
734 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
737 page_cache_release(page);
742 static int ocfs2_zero_extend(struct inode *inode,
747 struct super_block *sb = inode->i_sb;
749 start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
750 while (start_off < zero_to_size) {
751 ret = ocfs2_write_zero_page(inode, start_off);
757 start_off += sb->s_blocksize;
760 * Very large extends have the potential to lock up
761 * the cpu for extended periods of time.
770 int ocfs2_extend_no_holes(struct inode *inode, u64 new_i_size, u64 zero_to)
774 struct ocfs2_inode_info *oi = OCFS2_I(inode);
776 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
777 if (clusters_to_add < oi->ip_clusters)
780 clusters_to_add -= oi->ip_clusters;
782 if (clusters_to_add) {
783 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
792 * Call this even if we don't add any clusters to the tree. We
793 * still need to zero the area between the old i_size and the
796 ret = ocfs2_zero_extend(inode, zero_to);
804 static int ocfs2_extend_file(struct inode *inode,
805 struct buffer_head *di_bh,
809 struct ocfs2_inode_info *oi = OCFS2_I(inode);
813 /* setattr sometimes calls us like this. */
817 if (i_size_read(inode) == new_i_size)
819 BUG_ON(new_i_size < i_size_read(inode));
822 * Fall through for converting inline data, even if the fs
823 * supports sparse files.
825 * The check for inline data here is legal - nobody can add
826 * the feature since we have i_mutex. We must check it again
827 * after acquiring ip_alloc_sem though, as paths like mmap
828 * might have raced us to converting the inode to extents.
830 if (!(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
831 && ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
832 goto out_update_size;
835 * The alloc sem blocks people in read/write from reading our
836 * allocation until we're done changing it. We depend on
837 * i_mutex to block other extend/truncate calls while we're
840 down_write(&oi->ip_alloc_sem);
842 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
844 * We can optimize small extends by keeping the inodes
847 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
848 up_write(&oi->ip_alloc_sem);
849 goto out_update_size;
852 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
854 up_write(&oi->ip_alloc_sem);
861 if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
862 ret = ocfs2_extend_no_holes(inode, new_i_size, new_i_size);
864 up_write(&oi->ip_alloc_sem);
872 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
880 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
882 int status = 0, size_change;
883 struct inode *inode = dentry->d_inode;
884 struct super_block *sb = inode->i_sb;
885 struct ocfs2_super *osb = OCFS2_SB(sb);
886 struct buffer_head *bh = NULL;
887 handle_t *handle = NULL;
889 mlog_entry("(0x%p, '%.*s')\n", dentry,
890 dentry->d_name.len, dentry->d_name.name);
892 /* ensuring we don't even attempt to truncate a symlink */
893 if (S_ISLNK(inode->i_mode))
894 attr->ia_valid &= ~ATTR_SIZE;
896 if (attr->ia_valid & ATTR_MODE)
897 mlog(0, "mode change: %d\n", attr->ia_mode);
898 if (attr->ia_valid & ATTR_UID)
899 mlog(0, "uid change: %d\n", attr->ia_uid);
900 if (attr->ia_valid & ATTR_GID)
901 mlog(0, "gid change: %d\n", attr->ia_gid);
902 if (attr->ia_valid & ATTR_SIZE)
903 mlog(0, "size change...\n");
904 if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME))
905 mlog(0, "time change...\n");
907 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
908 | ATTR_GID | ATTR_UID | ATTR_MODE)
909 if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) {
910 mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid);
914 status = inode_change_ok(inode, attr);
918 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
920 status = ocfs2_rw_lock(inode, 1);
927 status = ocfs2_inode_lock(inode, &bh, 1);
929 if (status != -ENOENT)
934 if (size_change && attr->ia_size != i_size_read(inode)) {
935 if (attr->ia_size > sb->s_maxbytes) {
940 if (i_size_read(inode) > attr->ia_size) {
941 if (ocfs2_should_order_data(inode)) {
942 status = ocfs2_begin_ordered_truncate(inode,
947 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
949 status = ocfs2_extend_file(inode, bh, attr->ia_size);
951 if (status != -ENOSPC)
958 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
959 if (IS_ERR(handle)) {
960 status = PTR_ERR(handle);
966 * This will intentionally not wind up calling vmtruncate(),
967 * since all the work for a size change has been done above.
968 * Otherwise, we could get into problems with truncate as
969 * ip_alloc_sem is used there to protect against i_size
972 status = inode_setattr(inode, attr);
978 status = ocfs2_mark_inode_dirty(handle, inode, bh);
983 ocfs2_commit_trans(osb, handle);
985 ocfs2_inode_unlock(inode, 1);
988 ocfs2_rw_unlock(inode, 1);
996 int ocfs2_getattr(struct vfsmount *mnt,
997 struct dentry *dentry,
1000 struct inode *inode = dentry->d_inode;
1001 struct super_block *sb = dentry->d_inode->i_sb;
1002 struct ocfs2_super *osb = sb->s_fs_info;
1007 err = ocfs2_inode_revalidate(dentry);
1014 generic_fillattr(inode, stat);
1016 /* We set the blksize from the cluster size for performance */
1017 stat->blksize = osb->s_clustersize;
1025 int ocfs2_permission(struct inode *inode, int mask)
1031 ret = ocfs2_inode_lock(inode, NULL, 0);
1038 ret = generic_permission(inode, mask, NULL);
1040 ocfs2_inode_unlock(inode, 0);
1046 static int __ocfs2_write_remove_suid(struct inode *inode,
1047 struct buffer_head *bh)
1051 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1052 struct ocfs2_dinode *di;
1054 mlog_entry("(Inode %llu, mode 0%o)\n",
1055 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_mode);
1057 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1058 if (IS_ERR(handle)) {
1059 ret = PTR_ERR(handle);
1064 ret = ocfs2_journal_access(handle, inode, bh,
1065 OCFS2_JOURNAL_ACCESS_WRITE);
1071 inode->i_mode &= ~S_ISUID;
1072 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1073 inode->i_mode &= ~S_ISGID;
1075 di = (struct ocfs2_dinode *) bh->b_data;
1076 di->i_mode = cpu_to_le16(inode->i_mode);
1078 ret = ocfs2_journal_dirty(handle, bh);
1083 ocfs2_commit_trans(osb, handle);
1090 * Will look for holes and unwritten extents in the range starting at
1091 * pos for count bytes (inclusive).
1093 static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1097 unsigned int extent_flags;
1098 u32 cpos, clusters, extent_len, phys_cpos;
1099 struct super_block *sb = inode->i_sb;
1101 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1102 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1105 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1112 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
1117 if (extent_len > clusters)
1118 extent_len = clusters;
1120 clusters -= extent_len;
1127 static int ocfs2_write_remove_suid(struct inode *inode)
1130 struct buffer_head *bh = NULL;
1131 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1133 ret = ocfs2_read_block(inode, oi->ip_blkno, &bh);
1139 ret = __ocfs2_write_remove_suid(inode, bh);
1146 * Allocate enough extents to cover the region starting at byte offset
1147 * start for len bytes. Existing extents are skipped, any extents
1148 * added are marked as "unwritten".
1150 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1154 u32 cpos, phys_cpos, clusters, alloc_size;
1155 u64 end = start + len;
1156 struct buffer_head *di_bh = NULL;
1158 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1159 ret = ocfs2_read_block(inode, OCFS2_I(inode)->ip_blkno,
1167 * Nothing to do if the requested reservation range
1168 * fits within the inode.
1170 if (ocfs2_size_fits_inline_data(di_bh, end))
1173 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1181 * We consider both start and len to be inclusive.
1183 cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1184 clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1188 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1196 * Hole or existing extent len can be arbitrary, so
1197 * cap it to our own allocation request.
1199 if (alloc_size > clusters)
1200 alloc_size = clusters;
1204 * We already have an allocation at this
1205 * region so we can safely skip it.
1210 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1219 clusters -= alloc_size;
1229 static int __ocfs2_remove_inode_range(struct inode *inode,
1230 struct buffer_head *di_bh,
1231 u32 cpos, u32 phys_cpos, u32 len,
1232 struct ocfs2_cached_dealloc_ctxt *dealloc)
1235 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
1236 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1237 struct inode *tl_inode = osb->osb_tl_inode;
1239 struct ocfs2_alloc_context *meta_ac = NULL;
1240 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1241 struct ocfs2_extent_tree et;
1243 ocfs2_init_dinode_extent_tree(&et, inode, di_bh);
1245 ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
1251 mutex_lock(&tl_inode->i_mutex);
1253 if (ocfs2_truncate_log_needs_flush(osb)) {
1254 ret = __ocfs2_flush_truncate_log(osb);
1261 handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
1262 if (IS_ERR(handle)) {
1263 ret = PTR_ERR(handle);
1268 ret = ocfs2_journal_access(handle, inode, di_bh,
1269 OCFS2_JOURNAL_ACCESS_WRITE);
1275 ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
1282 OCFS2_I(inode)->ip_clusters -= len;
1283 di->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1285 ret = ocfs2_journal_dirty(handle, di_bh);
1291 ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len);
1296 ocfs2_commit_trans(osb, handle);
1298 mutex_unlock(&tl_inode->i_mutex);
1301 ocfs2_free_alloc_context(meta_ac);
1307 * Truncate a byte range, avoiding pages within partial clusters. This
1308 * preserves those pages for the zeroing code to write to.
1310 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1313 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1315 struct address_space *mapping = inode->i_mapping;
1317 start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1318 end = byte_start + byte_len;
1319 end = end & ~(osb->s_clustersize - 1);
1322 unmap_mapping_range(mapping, start, end - start, 0);
1323 truncate_inode_pages_range(mapping, start, end - 1);
1327 static int ocfs2_zero_partial_clusters(struct inode *inode,
1331 u64 tmpend, end = start + len;
1332 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1333 unsigned int csize = osb->s_clustersize;
1337 * The "start" and "end" values are NOT necessarily part of
1338 * the range whose allocation is being deleted. Rather, this
1339 * is what the user passed in with the request. We must zero
1340 * partial clusters here. There's no need to worry about
1341 * physical allocation - the zeroing code knows to skip holes.
1343 mlog(0, "byte start: %llu, end: %llu\n",
1344 (unsigned long long)start, (unsigned long long)end);
1347 * If both edges are on a cluster boundary then there's no
1348 * zeroing required as the region is part of the allocation to
1351 if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1354 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1355 if (IS_ERR(handle)) {
1356 ret = PTR_ERR(handle);
1362 * We want to get the byte offset of the end of the 1st cluster.
1364 tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1368 mlog(0, "1st range: start: %llu, tmpend: %llu\n",
1369 (unsigned long long)start, (unsigned long long)tmpend);
1371 ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1377 * This may make start and end equal, but the zeroing
1378 * code will skip any work in that case so there's no
1379 * need to catch it up here.
1381 start = end & ~(osb->s_clustersize - 1);
1383 mlog(0, "2nd range: start: %llu, end: %llu\n",
1384 (unsigned long long)start, (unsigned long long)end);
1386 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1391 ocfs2_commit_trans(osb, handle);
1396 static int ocfs2_remove_inode_range(struct inode *inode,
1397 struct buffer_head *di_bh, u64 byte_start,
1401 u32 trunc_start, trunc_len, cpos, phys_cpos, alloc_size;
1402 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1403 struct ocfs2_cached_dealloc_ctxt dealloc;
1404 struct address_space *mapping = inode->i_mapping;
1406 ocfs2_init_dealloc_ctxt(&dealloc);
1411 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1412 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1413 byte_start + byte_len, 0);
1419 * There's no need to get fancy with the page cache
1420 * truncate of an inline-data inode. We're talking
1421 * about less than a page here, which will be cached
1422 * in the dinode buffer anyway.
1424 unmap_mapping_range(mapping, 0, 0, 0);
1425 truncate_inode_pages(mapping, 0);
1429 trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1430 trunc_len = (byte_start + byte_len) >> osb->s_clustersize_bits;
1431 if (trunc_len >= trunc_start)
1432 trunc_len -= trunc_start;
1436 mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u\n",
1437 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1438 (unsigned long long)byte_start,
1439 (unsigned long long)byte_len, trunc_start, trunc_len);
1441 ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1449 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1456 if (alloc_size > trunc_len)
1457 alloc_size = trunc_len;
1459 /* Only do work for non-holes */
1460 if (phys_cpos != 0) {
1461 ret = __ocfs2_remove_inode_range(inode, di_bh, cpos,
1462 phys_cpos, alloc_size,
1471 trunc_len -= alloc_size;
1474 ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1477 ocfs2_schedule_truncate_log_flush(osb, 1);
1478 ocfs2_run_deallocs(osb, &dealloc);
1484 * Parts of this function taken from xfs_change_file_space()
1486 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1487 loff_t f_pos, unsigned int cmd,
1488 struct ocfs2_space_resv *sr,
1494 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1495 struct buffer_head *di_bh = NULL;
1497 unsigned long long max_off = inode->i_sb->s_maxbytes;
1499 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1502 mutex_lock(&inode->i_mutex);
1505 * This prevents concurrent writes on other nodes
1507 ret = ocfs2_rw_lock(inode, 1);
1513 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1519 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1521 goto out_inode_unlock;
1524 switch (sr->l_whence) {
1525 case 0: /*SEEK_SET*/
1527 case 1: /*SEEK_CUR*/
1528 sr->l_start += f_pos;
1530 case 2: /*SEEK_END*/
1531 sr->l_start += i_size_read(inode);
1535 goto out_inode_unlock;
1539 llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1542 || sr->l_start > max_off
1543 || (sr->l_start + llen) < 0
1544 || (sr->l_start + llen) > max_off) {
1546 goto out_inode_unlock;
1548 size = sr->l_start + sr->l_len;
1550 if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) {
1551 if (sr->l_len <= 0) {
1553 goto out_inode_unlock;
1557 if (file && should_remove_suid(file->f_path.dentry)) {
1558 ret = __ocfs2_write_remove_suid(inode, di_bh);
1561 goto out_inode_unlock;
1565 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1567 case OCFS2_IOC_RESVSP:
1568 case OCFS2_IOC_RESVSP64:
1570 * This takes unsigned offsets, but the signed ones we
1571 * pass have been checked against overflow above.
1573 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1576 case OCFS2_IOC_UNRESVSP:
1577 case OCFS2_IOC_UNRESVSP64:
1578 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1584 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1587 goto out_inode_unlock;
1591 * We update c/mtime for these changes
1593 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1594 if (IS_ERR(handle)) {
1595 ret = PTR_ERR(handle);
1597 goto out_inode_unlock;
1600 if (change_size && i_size_read(inode) < size)
1601 i_size_write(inode, size);
1603 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1604 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1608 ocfs2_commit_trans(osb, handle);
1612 ocfs2_inode_unlock(inode, 1);
1614 ocfs2_rw_unlock(inode, 1);
1617 mutex_unlock(&inode->i_mutex);
1621 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1622 struct ocfs2_space_resv *sr)
1624 struct inode *inode = file->f_path.dentry->d_inode;
1625 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);;
1627 if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1628 !ocfs2_writes_unwritten_extents(osb))
1630 else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1631 !ocfs2_sparse_alloc(osb))
1634 if (!S_ISREG(inode->i_mode))
1637 if (!(file->f_mode & FMODE_WRITE))
1640 return __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1643 static long ocfs2_fallocate(struct inode *inode, int mode, loff_t offset,
1646 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1647 struct ocfs2_space_resv sr;
1648 int change_size = 1;
1650 if (!ocfs2_writes_unwritten_extents(osb))
1653 if (S_ISDIR(inode->i_mode))
1656 if (mode & FALLOC_FL_KEEP_SIZE)
1660 sr.l_start = (s64)offset;
1661 sr.l_len = (s64)len;
1663 return __ocfs2_change_file_space(NULL, inode, offset,
1664 OCFS2_IOC_RESVSP64, &sr, change_size);
1667 static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
1673 int ret = 0, meta_level = 0;
1674 struct inode *inode = dentry->d_inode;
1675 loff_t saved_pos, end;
1678 * We start with a read level meta lock and only jump to an ex
1679 * if we need to make modifications here.
1682 ret = ocfs2_inode_lock(inode, NULL, meta_level);
1689 /* Clear suid / sgid if necessary. We do this here
1690 * instead of later in the write path because
1691 * remove_suid() calls ->setattr without any hint that
1692 * we may have already done our cluster locking. Since
1693 * ocfs2_setattr() *must* take cluster locks to
1694 * proceeed, this will lead us to recursively lock the
1695 * inode. There's also the dinode i_size state which
1696 * can be lost via setattr during extending writes (we
1697 * set inode->i_size at the end of a write. */
1698 if (should_remove_suid(dentry)) {
1699 if (meta_level == 0) {
1700 ocfs2_inode_unlock(inode, meta_level);
1705 ret = ocfs2_write_remove_suid(inode);
1712 /* work on a copy of ppos until we're sure that we won't have
1713 * to recalculate it due to relocking. */
1715 saved_pos = i_size_read(inode);
1716 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos);
1721 end = saved_pos + count;
1724 * Skip the O_DIRECT checks if we don't need
1727 if (!direct_io || !(*direct_io))
1731 * There's no sane way to do direct writes to an inode
1734 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1740 * Allowing concurrent direct writes means
1741 * i_size changes wouldn't be synchronized, so
1742 * one node could wind up truncating another
1745 if (end > i_size_read(inode)) {
1751 * We don't fill holes during direct io, so
1752 * check for them here. If any are found, the
1753 * caller will have to retake some cluster
1754 * locks and initiate the io as buffered.
1756 ret = ocfs2_check_range_for_holes(inode, saved_pos, count);
1769 ocfs2_inode_unlock(inode, meta_level);
1775 static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
1776 const struct iovec *iov,
1777 unsigned long nr_segs,
1780 int ret, direct_io, appending, rw_level, have_alloc_sem = 0;
1782 ssize_t written = 0;
1783 size_t ocount; /* original count */
1784 size_t count; /* after file limit checks */
1785 loff_t old_size, *ppos = &iocb->ki_pos;
1787 struct file *file = iocb->ki_filp;
1788 struct inode *inode = file->f_path.dentry->d_inode;
1789 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1791 mlog_entry("(0x%p, %u, '%.*s')\n", file,
1792 (unsigned int)nr_segs,
1793 file->f_path.dentry->d_name.len,
1794 file->f_path.dentry->d_name.name);
1796 if (iocb->ki_left == 0)
1799 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1801 appending = file->f_flags & O_APPEND ? 1 : 0;
1802 direct_io = file->f_flags & O_DIRECT ? 1 : 0;
1804 mutex_lock(&inode->i_mutex);
1807 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
1809 down_read(&inode->i_alloc_sem);
1813 /* concurrent O_DIRECT writes are allowed */
1814 rw_level = !direct_io;
1815 ret = ocfs2_rw_lock(inode, rw_level);
1821 can_do_direct = direct_io;
1822 ret = ocfs2_prepare_inode_for_write(file->f_path.dentry, ppos,
1823 iocb->ki_left, appending,
1831 * We can't complete the direct I/O as requested, fall back to
1834 if (direct_io && !can_do_direct) {
1835 ocfs2_rw_unlock(inode, rw_level);
1836 up_read(&inode->i_alloc_sem);
1846 * To later detect whether a journal commit for sync writes is
1847 * necessary, we sample i_size, and cluster count here.
1849 old_size = i_size_read(inode);
1850 old_clusters = OCFS2_I(inode)->ip_clusters;
1852 /* communicate with ocfs2_dio_end_io */
1853 ocfs2_iocb_set_rw_locked(iocb, rw_level);
1856 ret = generic_segment_checks(iov, &nr_segs, &ocount,
1861 ret = generic_write_checks(file, ppos, &count,
1862 S_ISBLK(inode->i_mode));
1866 written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
1867 ppos, count, ocount);
1870 * direct write may have instantiated a few
1871 * blocks outside i_size. Trim these off again.
1872 * Don't need i_size_read because we hold i_mutex.
1874 if (*ppos + count > inode->i_size)
1875 vmtruncate(inode, inode->i_size);
1880 written = generic_file_aio_write_nolock(iocb, iov, nr_segs,
1885 /* buffered aio wouldn't have proper lock coverage today */
1886 BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
1888 if ((file->f_flags & O_SYNC && !direct_io) || IS_SYNC(inode)) {
1890 * The generic write paths have handled getting data
1891 * to disk, but since we don't make use of the dirty
1892 * inode list, a manual journal commit is necessary
1895 if (old_size != i_size_read(inode) ||
1896 old_clusters != OCFS2_I(inode)->ip_clusters) {
1897 ret = jbd2_journal_force_commit(osb->journal->j_journal);
1904 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
1905 * function pointer which is called when o_direct io completes so that
1906 * it can unlock our rw lock. (it's the clustered equivalent of
1907 * i_alloc_sem; protects truncate from racing with pending ios).
1908 * Unfortunately there are error cases which call end_io and others
1909 * that don't. so we don't have to unlock the rw_lock if either an
1910 * async dio is going to do it in the future or an end_io after an
1911 * error has already done it.
1913 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
1920 ocfs2_rw_unlock(inode, rw_level);
1924 up_read(&inode->i_alloc_sem);
1926 mutex_unlock(&inode->i_mutex);
1929 return written ? written : ret;
1932 static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
1939 struct inode *inode = out->f_path.dentry->d_inode;
1941 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe,
1943 out->f_path.dentry->d_name.len,
1944 out->f_path.dentry->d_name.name);
1946 inode_double_lock(inode, pipe->inode);
1948 ret = ocfs2_rw_lock(inode, 1);
1954 ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, ppos, len, 0,
1961 ret = generic_file_splice_write_nolock(pipe, out, ppos, len, flags);
1964 ocfs2_rw_unlock(inode, 1);
1966 inode_double_unlock(inode, pipe->inode);
1972 static ssize_t ocfs2_file_splice_read(struct file *in,
1974 struct pipe_inode_info *pipe,
1979 struct inode *inode = in->f_path.dentry->d_inode;
1981 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe,
1983 in->f_path.dentry->d_name.len,
1984 in->f_path.dentry->d_name.name);
1987 * See the comment in ocfs2_file_aio_read()
1989 ret = ocfs2_inode_lock(inode, NULL, 0);
1994 ocfs2_inode_unlock(inode, 0);
1996 ret = generic_file_splice_read(in, ppos, pipe, len, flags);
2003 static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
2004 const struct iovec *iov,
2005 unsigned long nr_segs,
2008 int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
2009 struct file *filp = iocb->ki_filp;
2010 struct inode *inode = filp->f_path.dentry->d_inode;
2012 mlog_entry("(0x%p, %u, '%.*s')\n", filp,
2013 (unsigned int)nr_segs,
2014 filp->f_path.dentry->d_name.len,
2015 filp->f_path.dentry->d_name.name);
2024 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
2025 * need locks to protect pending reads from racing with truncate.
2027 if (filp->f_flags & O_DIRECT) {
2028 down_read(&inode->i_alloc_sem);
2031 ret = ocfs2_rw_lock(inode, 0);
2037 /* communicate with ocfs2_dio_end_io */
2038 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2042 * We're fine letting folks race truncates and extending
2043 * writes with read across the cluster, just like they can
2044 * locally. Hence no rw_lock during read.
2046 * Take and drop the meta data lock to update inode fields
2047 * like i_size. This allows the checks down below
2048 * generic_file_aio_read() a chance of actually working.
2050 ret = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level);
2055 ocfs2_inode_unlock(inode, lock_level);
2057 ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
2059 mlog(0, "generic_file_aio_read returned -EINVAL\n");
2061 /* buffered aio wouldn't have proper lock coverage today */
2062 BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
2064 /* see ocfs2_file_aio_write */
2065 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2072 up_read(&inode->i_alloc_sem);
2074 ocfs2_rw_unlock(inode, rw_level);
2080 const struct inode_operations ocfs2_file_iops = {
2081 .setattr = ocfs2_setattr,
2082 .getattr = ocfs2_getattr,
2083 .permission = ocfs2_permission,
2084 .setxattr = generic_setxattr,
2085 .getxattr = generic_getxattr,
2086 .listxattr = ocfs2_listxattr,
2087 .removexattr = generic_removexattr,
2088 .fallocate = ocfs2_fallocate,
2089 .fiemap = ocfs2_fiemap,
2092 const struct inode_operations ocfs2_special_file_iops = {
2093 .setattr = ocfs2_setattr,
2094 .getattr = ocfs2_getattr,
2095 .permission = ocfs2_permission,
2099 * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2100 * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2102 const struct file_operations ocfs2_fops = {
2103 .llseek = generic_file_llseek,
2104 .read = do_sync_read,
2105 .write = do_sync_write,
2107 .fsync = ocfs2_sync_file,
2108 .release = ocfs2_file_release,
2109 .open = ocfs2_file_open,
2110 .aio_read = ocfs2_file_aio_read,
2111 .aio_write = ocfs2_file_aio_write,
2112 .unlocked_ioctl = ocfs2_ioctl,
2113 #ifdef CONFIG_COMPAT
2114 .compat_ioctl = ocfs2_compat_ioctl,
2117 .flock = ocfs2_flock,
2118 .splice_read = ocfs2_file_splice_read,
2119 .splice_write = ocfs2_file_splice_write,
2122 const struct file_operations ocfs2_dops = {
2123 .llseek = generic_file_llseek,
2124 .read = generic_read_dir,
2125 .readdir = ocfs2_readdir,
2126 .fsync = ocfs2_sync_file,
2127 .release = ocfs2_dir_release,
2128 .open = ocfs2_dir_open,
2129 .unlocked_ioctl = ocfs2_ioctl,
2130 #ifdef CONFIG_COMPAT
2131 .compat_ioctl = ocfs2_compat_ioctl,
2134 .flock = ocfs2_flock,
2138 * POSIX-lockless variants of our file_operations.
2140 * These will be used if the underlying cluster stack does not support
2141 * posix file locking, if the user passes the "localflocks" mount
2142 * option, or if we have a local-only fs.
2144 * ocfs2_flock is in here because all stacks handle UNIX file locks,
2145 * so we still want it in the case of no stack support for
2146 * plocks. Internally, it will do the right thing when asked to ignore
2149 const struct file_operations ocfs2_fops_no_plocks = {
2150 .llseek = generic_file_llseek,
2151 .read = do_sync_read,
2152 .write = do_sync_write,
2154 .fsync = ocfs2_sync_file,
2155 .release = ocfs2_file_release,
2156 .open = ocfs2_file_open,
2157 .aio_read = ocfs2_file_aio_read,
2158 .aio_write = ocfs2_file_aio_write,
2159 .unlocked_ioctl = ocfs2_ioctl,
2160 #ifdef CONFIG_COMPAT
2161 .compat_ioctl = ocfs2_compat_ioctl,
2163 .flock = ocfs2_flock,
2164 .splice_read = ocfs2_file_splice_read,
2165 .splice_write = ocfs2_file_splice_write,
2168 const struct file_operations ocfs2_dops_no_plocks = {
2169 .llseek = generic_file_llseek,
2170 .read = generic_read_dir,
2171 .readdir = ocfs2_readdir,
2172 .fsync = ocfs2_sync_file,
2173 .release = ocfs2_dir_release,
2174 .open = ocfs2_dir_open,
2175 .unlocked_ioctl = ocfs2_ioctl,
2176 #ifdef CONFIG_COMPAT
2177 .compat_ioctl = ocfs2_compat_ioctl,
2179 .flock = ocfs2_flock,