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"
61 #include "buffer_head_io.h"
63 static int ocfs2_sync_inode(struct inode *inode)
65 filemap_fdatawrite(inode->i_mapping);
66 return sync_mapping_buffers(inode->i_mapping);
69 static int ocfs2_init_file_private(struct inode *inode, struct file *file)
71 struct ocfs2_file_private *fp;
73 fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
78 mutex_init(&fp->fp_mutex);
79 ocfs2_file_lock_res_init(&fp->fp_flock, fp);
80 file->private_data = fp;
85 static void ocfs2_free_file_private(struct inode *inode, struct file *file)
87 struct ocfs2_file_private *fp = file->private_data;
88 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
91 ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
92 ocfs2_lock_res_free(&fp->fp_flock);
94 file->private_data = NULL;
98 static int ocfs2_file_open(struct inode *inode, struct file *file)
101 int mode = file->f_flags;
102 struct ocfs2_inode_info *oi = OCFS2_I(inode);
104 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
105 file->f_path.dentry->d_name.len, file->f_path.dentry->d_name.name);
107 spin_lock(&oi->ip_lock);
109 /* Check that the inode hasn't been wiped from disk by another
110 * node. If it hasn't then we're safe as long as we hold the
111 * spin lock until our increment of open count. */
112 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
113 spin_unlock(&oi->ip_lock);
120 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
123 spin_unlock(&oi->ip_lock);
125 status = ocfs2_init_file_private(inode, file);
128 * We want to set open count back if we're failing the
131 spin_lock(&oi->ip_lock);
133 spin_unlock(&oi->ip_lock);
141 static int ocfs2_file_release(struct inode *inode, struct file *file)
143 struct ocfs2_inode_info *oi = OCFS2_I(inode);
145 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
146 file->f_path.dentry->d_name.len,
147 file->f_path.dentry->d_name.name);
149 spin_lock(&oi->ip_lock);
150 if (!--oi->ip_open_count)
151 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
152 spin_unlock(&oi->ip_lock);
154 ocfs2_free_file_private(inode, file);
161 static int ocfs2_dir_open(struct inode *inode, struct file *file)
163 return ocfs2_init_file_private(inode, file);
166 static int ocfs2_dir_release(struct inode *inode, struct file *file)
168 ocfs2_free_file_private(inode, file);
172 static int ocfs2_sync_file(struct file *file,
173 struct dentry *dentry,
178 struct inode *inode = dentry->d_inode;
179 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
181 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync,
182 dentry->d_name.len, dentry->d_name.name);
184 err = ocfs2_sync_inode(dentry->d_inode);
188 journal = osb->journal->j_journal;
189 err = jbd2_journal_force_commit(journal);
194 return (err < 0) ? -EIO : 0;
197 int ocfs2_should_update_atime(struct inode *inode,
198 struct vfsmount *vfsmnt)
201 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
203 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
206 if ((inode->i_flags & S_NOATIME) ||
207 ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
211 * We can be called with no vfsmnt structure - NFSD will
214 * Note that our action here is different than touch_atime() -
215 * if we can't tell whether this is a noatime mount, then we
216 * don't know whether to trust the value of s_atime_quantum.
221 if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
222 ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
225 if (vfsmnt->mnt_flags & MNT_RELATIME) {
226 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
227 (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
234 if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
240 int ocfs2_update_inode_atime(struct inode *inode,
241 struct buffer_head *bh)
244 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
246 struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
250 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
251 if (IS_ERR(handle)) {
252 ret = PTR_ERR(handle);
257 ret = ocfs2_journal_access(handle, inode, bh,
258 OCFS2_JOURNAL_ACCESS_WRITE);
265 * Don't use ocfs2_mark_inode_dirty() here as we don't always
266 * have i_mutex to guard against concurrent changes to other
269 inode->i_atime = CURRENT_TIME;
270 di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
271 di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
273 ret = ocfs2_journal_dirty(handle, bh);
278 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
284 static int ocfs2_set_inode_size(handle_t *handle,
286 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 static 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_orphan_for_truncate(struct ocfs2_super *osb,
334 struct buffer_head *fe_bh,
339 struct ocfs2_dinode *di;
344 /* TODO: This needs to actually orphan the inode in this
347 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
348 if (IS_ERR(handle)) {
349 status = PTR_ERR(handle);
354 status = ocfs2_journal_access(handle, inode, fe_bh,
355 OCFS2_JOURNAL_ACCESS_WRITE);
362 * Do this before setting i_size.
364 cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
365 status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
372 i_size_write(inode, new_i_size);
373 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
375 di = (struct ocfs2_dinode *) fe_bh->b_data;
376 di->i_size = cpu_to_le64(new_i_size);
377 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
378 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
380 status = ocfs2_journal_dirty(handle, fe_bh);
385 ocfs2_commit_trans(osb, handle);
392 static int ocfs2_truncate_file(struct inode *inode,
393 struct buffer_head *di_bh,
397 struct ocfs2_dinode *fe = NULL;
398 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
399 struct ocfs2_truncate_context *tc = NULL;
401 mlog_entry("(inode = %llu, new_i_size = %llu\n",
402 (unsigned long long)OCFS2_I(inode)->ip_blkno,
403 (unsigned long long)new_i_size);
405 fe = (struct ocfs2_dinode *) di_bh->b_data;
406 if (!OCFS2_IS_VALID_DINODE(fe)) {
407 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
412 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
413 "Inode %llu, inode i_size = %lld != di "
414 "i_size = %llu, i_flags = 0x%x\n",
415 (unsigned long long)OCFS2_I(inode)->ip_blkno,
417 (unsigned long long)le64_to_cpu(fe->i_size),
418 le32_to_cpu(fe->i_flags));
420 if (new_i_size > le64_to_cpu(fe->i_size)) {
421 mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
422 (unsigned long long)le64_to_cpu(fe->i_size),
423 (unsigned long long)new_i_size);
429 mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
430 (unsigned long long)le64_to_cpu(fe->i_blkno),
431 (unsigned long long)le64_to_cpu(fe->i_size),
432 (unsigned long long)new_i_size);
434 /* lets handle the simple truncate cases before doing any more
435 * cluster locking. */
436 if (new_i_size == le64_to_cpu(fe->i_size))
439 down_write(&OCFS2_I(inode)->ip_alloc_sem);
442 * The inode lock forced other nodes to sync and drop their
443 * pages, which (correctly) happens even if we have a truncate
444 * without allocation change - ocfs2 cluster sizes can be much
445 * greater than page size, so we have to truncate them
448 unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
449 truncate_inode_pages(inode->i_mapping, new_i_size);
451 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
452 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
453 i_size_read(inode), 1);
457 goto bail_unlock_sem;
460 /* alright, we're going to need to do a full blown alloc size
461 * change. Orphan the inode so that recovery can complete the
462 * truncate if necessary. This does the task of marking
464 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
467 goto bail_unlock_sem;
470 status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc);
473 goto bail_unlock_sem;
476 status = ocfs2_commit_truncate(osb, inode, di_bh, tc);
479 goto bail_unlock_sem;
482 /* TODO: orphan dir cleanup here. */
484 up_write(&OCFS2_I(inode)->ip_alloc_sem);
493 * extend file allocation only here.
494 * we'll update all the disk stuff, and oip->alloc_size
496 * expect stuff to be locked, a transaction started and enough data /
497 * metadata reservations in the contexts.
499 * Will return -EAGAIN, and a reason if a restart is needed.
500 * If passed in, *reason will always be set, even in error.
502 int ocfs2_add_inode_data(struct ocfs2_super *osb,
507 struct buffer_head *fe_bh,
509 struct ocfs2_alloc_context *data_ac,
510 struct ocfs2_alloc_context *meta_ac,
511 enum ocfs2_alloc_restarted *reason_ret)
514 struct ocfs2_extent_tree et;
516 ocfs2_init_dinode_extent_tree(&et, inode, fe_bh);
517 ret = ocfs2_add_clusters_in_btree(osb, inode, logical_offset,
518 clusters_to_add, mark_unwritten,
520 data_ac, meta_ac, reason_ret);
525 static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
526 u32 clusters_to_add, int mark_unwritten)
529 int restart_func = 0;
532 struct buffer_head *bh = NULL;
533 struct ocfs2_dinode *fe = NULL;
534 handle_t *handle = NULL;
535 struct ocfs2_alloc_context *data_ac = NULL;
536 struct ocfs2_alloc_context *meta_ac = NULL;
537 enum ocfs2_alloc_restarted why;
538 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
539 struct ocfs2_extent_tree et;
541 mlog_entry("(clusters_to_add = %u)\n", clusters_to_add);
544 * This function only exists for file systems which don't
547 BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
549 status = ocfs2_read_block(inode, OCFS2_I(inode)->ip_blkno, &bh);
555 fe = (struct ocfs2_dinode *) bh->b_data;
556 if (!OCFS2_IS_VALID_DINODE(fe)) {
557 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
563 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
565 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, "
566 "clusters_to_add = %u\n",
567 (unsigned long long)OCFS2_I(inode)->ip_blkno,
568 (long long)i_size_read(inode), le32_to_cpu(fe->i_clusters),
570 ocfs2_init_dinode_extent_tree(&et, inode, bh);
571 status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
578 credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list,
580 handle = ocfs2_start_trans(osb, credits);
581 if (IS_ERR(handle)) {
582 status = PTR_ERR(handle);
588 restarted_transaction:
589 /* reserve a write to the file entry early on - that we if we
590 * run out of credits in the allocation path, we can still
592 status = ocfs2_journal_access(handle, inode, bh,
593 OCFS2_JOURNAL_ACCESS_WRITE);
599 prev_clusters = OCFS2_I(inode)->ip_clusters;
601 status = ocfs2_add_inode_data(osb,
611 if ((status < 0) && (status != -EAGAIN)) {
612 if (status != -ENOSPC)
617 status = ocfs2_journal_dirty(handle, bh);
623 spin_lock(&OCFS2_I(inode)->ip_lock);
624 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
625 spin_unlock(&OCFS2_I(inode)->ip_lock);
627 if (why != RESTART_NONE && clusters_to_add) {
628 if (why == RESTART_META) {
629 mlog(0, "restarting function.\n");
632 BUG_ON(why != RESTART_TRANS);
634 mlog(0, "restarting transaction.\n");
635 /* TODO: This can be more intelligent. */
636 credits = ocfs2_calc_extend_credits(osb->sb,
639 status = ocfs2_extend_trans(handle, credits);
641 /* handle still has to be committed at
647 goto restarted_transaction;
651 mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
652 le32_to_cpu(fe->i_clusters),
653 (unsigned long long)le64_to_cpu(fe->i_size));
654 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
655 OCFS2_I(inode)->ip_clusters, (long long)i_size_read(inode));
659 ocfs2_commit_trans(osb, handle);
663 ocfs2_free_alloc_context(data_ac);
667 ocfs2_free_alloc_context(meta_ac);
670 if ((!status) && restart_func) {
681 /* Some parts of this taken from generic_cont_expand, which turned out
682 * to be too fragile to do exactly what we need without us having to
683 * worry about recursive locking in ->write_begin() and ->write_end(). */
684 static int ocfs2_write_zero_page(struct inode *inode,
687 struct address_space *mapping = inode->i_mapping;
691 handle_t *handle = NULL;
694 offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */
695 /* ugh. in prepare/commit_write, if from==to==start of block, we
696 ** skip the prepare. make sure we never send an offset for the start
699 if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) {
702 index = size >> PAGE_CACHE_SHIFT;
704 page = grab_cache_page(mapping, index);
711 ret = ocfs2_prepare_write_nolock(inode, page, offset, offset);
717 if (ocfs2_should_order_data(inode)) {
718 handle = ocfs2_start_walk_page_trans(inode, page, offset,
720 if (IS_ERR(handle)) {
721 ret = PTR_ERR(handle);
727 /* must not update i_size! */
728 ret = block_commit_write(page, offset, offset);
735 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
738 page_cache_release(page);
743 static int ocfs2_zero_extend(struct inode *inode,
748 struct super_block *sb = inode->i_sb;
750 start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
751 while (start_off < zero_to_size) {
752 ret = ocfs2_write_zero_page(inode, start_off);
758 start_off += sb->s_blocksize;
761 * Very large extends have the potential to lock up
762 * the cpu for extended periods of time.
771 int ocfs2_extend_no_holes(struct inode *inode, u64 new_i_size, u64 zero_to)
775 struct ocfs2_inode_info *oi = OCFS2_I(inode);
777 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
778 if (clusters_to_add < oi->ip_clusters)
781 clusters_to_add -= oi->ip_clusters;
783 if (clusters_to_add) {
784 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
793 * Call this even if we don't add any clusters to the tree. We
794 * still need to zero the area between the old i_size and the
797 ret = ocfs2_zero_extend(inode, zero_to);
805 static int ocfs2_extend_file(struct inode *inode,
806 struct buffer_head *di_bh,
810 struct ocfs2_inode_info *oi = OCFS2_I(inode);
814 /* setattr sometimes calls us like this. */
818 if (i_size_read(inode) == new_i_size)
820 BUG_ON(new_i_size < i_size_read(inode));
823 * Fall through for converting inline data, even if the fs
824 * supports sparse files.
826 * The check for inline data here is legal - nobody can add
827 * the feature since we have i_mutex. We must check it again
828 * after acquiring ip_alloc_sem though, as paths like mmap
829 * might have raced us to converting the inode to extents.
831 if (!(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
832 && ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
833 goto out_update_size;
836 * The alloc sem blocks people in read/write from reading our
837 * allocation until we're done changing it. We depend on
838 * i_mutex to block other extend/truncate calls while we're
841 down_write(&oi->ip_alloc_sem);
843 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
845 * We can optimize small extends by keeping the inodes
848 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
849 up_write(&oi->ip_alloc_sem);
850 goto out_update_size;
853 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
855 up_write(&oi->ip_alloc_sem);
862 if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
863 ret = ocfs2_extend_no_holes(inode, new_i_size, new_i_size);
865 up_write(&oi->ip_alloc_sem);
873 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
881 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
883 int status = 0, size_change;
884 struct inode *inode = dentry->d_inode;
885 struct super_block *sb = inode->i_sb;
886 struct ocfs2_super *osb = OCFS2_SB(sb);
887 struct buffer_head *bh = NULL;
888 handle_t *handle = NULL;
890 mlog_entry("(0x%p, '%.*s')\n", dentry,
891 dentry->d_name.len, dentry->d_name.name);
893 /* ensuring we don't even attempt to truncate a symlink */
894 if (S_ISLNK(inode->i_mode))
895 attr->ia_valid &= ~ATTR_SIZE;
897 if (attr->ia_valid & ATTR_MODE)
898 mlog(0, "mode change: %d\n", attr->ia_mode);
899 if (attr->ia_valid & ATTR_UID)
900 mlog(0, "uid change: %d\n", attr->ia_uid);
901 if (attr->ia_valid & ATTR_GID)
902 mlog(0, "gid change: %d\n", attr->ia_gid);
903 if (attr->ia_valid & ATTR_SIZE)
904 mlog(0, "size change...\n");
905 if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME))
906 mlog(0, "time change...\n");
908 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
909 | ATTR_GID | ATTR_UID | ATTR_MODE)
910 if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) {
911 mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid);
915 status = inode_change_ok(inode, attr);
919 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
921 status = ocfs2_rw_lock(inode, 1);
928 status = ocfs2_inode_lock(inode, &bh, 1);
930 if (status != -ENOENT)
935 if (size_change && attr->ia_size != i_size_read(inode)) {
936 if (attr->ia_size > sb->s_maxbytes) {
941 if (i_size_read(inode) > attr->ia_size) {
942 if (ocfs2_should_order_data(inode)) {
943 status = ocfs2_begin_ordered_truncate(inode,
948 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
950 status = ocfs2_extend_file(inode, bh, attr->ia_size);
952 if (status != -ENOSPC)
959 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
960 if (IS_ERR(handle)) {
961 status = PTR_ERR(handle);
967 * This will intentionally not wind up calling vmtruncate(),
968 * since all the work for a size change has been done above.
969 * Otherwise, we could get into problems with truncate as
970 * ip_alloc_sem is used there to protect against i_size
973 status = inode_setattr(inode, attr);
979 status = ocfs2_mark_inode_dirty(handle, inode, bh);
984 ocfs2_commit_trans(osb, handle);
986 ocfs2_inode_unlock(inode, 1);
989 ocfs2_rw_unlock(inode, 1);
993 if (!status && attr->ia_valid & ATTR_MODE) {
994 status = ocfs2_acl_chmod(inode);
1003 int ocfs2_getattr(struct vfsmount *mnt,
1004 struct dentry *dentry,
1007 struct inode *inode = dentry->d_inode;
1008 struct super_block *sb = dentry->d_inode->i_sb;
1009 struct ocfs2_super *osb = sb->s_fs_info;
1014 err = ocfs2_inode_revalidate(dentry);
1021 generic_fillattr(inode, stat);
1023 /* We set the blksize from the cluster size for performance */
1024 stat->blksize = osb->s_clustersize;
1032 int ocfs2_permission(struct inode *inode, int mask)
1038 ret = ocfs2_inode_lock(inode, NULL, 0);
1045 ret = generic_permission(inode, mask, ocfs2_check_acl);
1047 ocfs2_inode_unlock(inode, 0);
1053 static int __ocfs2_write_remove_suid(struct inode *inode,
1054 struct buffer_head *bh)
1058 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1059 struct ocfs2_dinode *di;
1061 mlog_entry("(Inode %llu, mode 0%o)\n",
1062 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_mode);
1064 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1065 if (IS_ERR(handle)) {
1066 ret = PTR_ERR(handle);
1071 ret = ocfs2_journal_access(handle, inode, bh,
1072 OCFS2_JOURNAL_ACCESS_WRITE);
1078 inode->i_mode &= ~S_ISUID;
1079 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1080 inode->i_mode &= ~S_ISGID;
1082 di = (struct ocfs2_dinode *) bh->b_data;
1083 di->i_mode = cpu_to_le16(inode->i_mode);
1085 ret = ocfs2_journal_dirty(handle, bh);
1090 ocfs2_commit_trans(osb, handle);
1097 * Will look for holes and unwritten extents in the range starting at
1098 * pos for count bytes (inclusive).
1100 static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1104 unsigned int extent_flags;
1105 u32 cpos, clusters, extent_len, phys_cpos;
1106 struct super_block *sb = inode->i_sb;
1108 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1109 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1112 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1119 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
1124 if (extent_len > clusters)
1125 extent_len = clusters;
1127 clusters -= extent_len;
1134 static int ocfs2_write_remove_suid(struct inode *inode)
1137 struct buffer_head *bh = NULL;
1138 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1140 ret = ocfs2_read_block(inode, oi->ip_blkno, &bh);
1146 ret = __ocfs2_write_remove_suid(inode, bh);
1153 * Allocate enough extents to cover the region starting at byte offset
1154 * start for len bytes. Existing extents are skipped, any extents
1155 * added are marked as "unwritten".
1157 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1161 u32 cpos, phys_cpos, clusters, alloc_size;
1162 u64 end = start + len;
1163 struct buffer_head *di_bh = NULL;
1165 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1166 ret = ocfs2_read_block(inode, OCFS2_I(inode)->ip_blkno,
1174 * Nothing to do if the requested reservation range
1175 * fits within the inode.
1177 if (ocfs2_size_fits_inline_data(di_bh, end))
1180 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1188 * We consider both start and len to be inclusive.
1190 cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1191 clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1195 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1203 * Hole or existing extent len can be arbitrary, so
1204 * cap it to our own allocation request.
1206 if (alloc_size > clusters)
1207 alloc_size = clusters;
1211 * We already have an allocation at this
1212 * region so we can safely skip it.
1217 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1226 clusters -= alloc_size;
1237 * Truncate a byte range, avoiding pages within partial clusters. This
1238 * preserves those pages for the zeroing code to write to.
1240 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1243 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1245 struct address_space *mapping = inode->i_mapping;
1247 start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1248 end = byte_start + byte_len;
1249 end = end & ~(osb->s_clustersize - 1);
1252 unmap_mapping_range(mapping, start, end - start, 0);
1253 truncate_inode_pages_range(mapping, start, end - 1);
1257 static int ocfs2_zero_partial_clusters(struct inode *inode,
1261 u64 tmpend, end = start + len;
1262 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1263 unsigned int csize = osb->s_clustersize;
1267 * The "start" and "end" values are NOT necessarily part of
1268 * the range whose allocation is being deleted. Rather, this
1269 * is what the user passed in with the request. We must zero
1270 * partial clusters here. There's no need to worry about
1271 * physical allocation - the zeroing code knows to skip holes.
1273 mlog(0, "byte start: %llu, end: %llu\n",
1274 (unsigned long long)start, (unsigned long long)end);
1277 * If both edges are on a cluster boundary then there's no
1278 * zeroing required as the region is part of the allocation to
1281 if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1284 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1285 if (IS_ERR(handle)) {
1286 ret = PTR_ERR(handle);
1292 * We want to get the byte offset of the end of the 1st cluster.
1294 tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1298 mlog(0, "1st range: start: %llu, tmpend: %llu\n",
1299 (unsigned long long)start, (unsigned long long)tmpend);
1301 ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1307 * This may make start and end equal, but the zeroing
1308 * code will skip any work in that case so there's no
1309 * need to catch it up here.
1311 start = end & ~(osb->s_clustersize - 1);
1313 mlog(0, "2nd range: start: %llu, end: %llu\n",
1314 (unsigned long long)start, (unsigned long long)end);
1316 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1321 ocfs2_commit_trans(osb, handle);
1326 static int ocfs2_remove_inode_range(struct inode *inode,
1327 struct buffer_head *di_bh, u64 byte_start,
1331 u32 trunc_start, trunc_len, cpos, phys_cpos, alloc_size;
1332 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1333 struct ocfs2_cached_dealloc_ctxt dealloc;
1334 struct address_space *mapping = inode->i_mapping;
1335 struct ocfs2_extent_tree et;
1337 ocfs2_init_dinode_extent_tree(&et, inode, di_bh);
1338 ocfs2_init_dealloc_ctxt(&dealloc);
1343 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1344 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1345 byte_start + byte_len, 0);
1351 * There's no need to get fancy with the page cache
1352 * truncate of an inline-data inode. We're talking
1353 * about less than a page here, which will be cached
1354 * in the dinode buffer anyway.
1356 unmap_mapping_range(mapping, 0, 0, 0);
1357 truncate_inode_pages(mapping, 0);
1361 trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1362 trunc_len = (byte_start + byte_len) >> osb->s_clustersize_bits;
1363 if (trunc_len >= trunc_start)
1364 trunc_len -= trunc_start;
1368 mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u\n",
1369 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1370 (unsigned long long)byte_start,
1371 (unsigned long long)byte_len, trunc_start, trunc_len);
1373 ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1381 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1388 if (alloc_size > trunc_len)
1389 alloc_size = trunc_len;
1391 /* Only do work for non-holes */
1392 if (phys_cpos != 0) {
1393 ret = ocfs2_remove_btree_range(inode, &et, cpos,
1394 phys_cpos, alloc_size,
1403 trunc_len -= alloc_size;
1406 ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1409 ocfs2_schedule_truncate_log_flush(osb, 1);
1410 ocfs2_run_deallocs(osb, &dealloc);
1416 * Parts of this function taken from xfs_change_file_space()
1418 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1419 loff_t f_pos, unsigned int cmd,
1420 struct ocfs2_space_resv *sr,
1426 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1427 struct buffer_head *di_bh = NULL;
1429 unsigned long long max_off = inode->i_sb->s_maxbytes;
1431 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1434 mutex_lock(&inode->i_mutex);
1437 * This prevents concurrent writes on other nodes
1439 ret = ocfs2_rw_lock(inode, 1);
1445 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1451 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1453 goto out_inode_unlock;
1456 switch (sr->l_whence) {
1457 case 0: /*SEEK_SET*/
1459 case 1: /*SEEK_CUR*/
1460 sr->l_start += f_pos;
1462 case 2: /*SEEK_END*/
1463 sr->l_start += i_size_read(inode);
1467 goto out_inode_unlock;
1471 llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1474 || sr->l_start > max_off
1475 || (sr->l_start + llen) < 0
1476 || (sr->l_start + llen) > max_off) {
1478 goto out_inode_unlock;
1480 size = sr->l_start + sr->l_len;
1482 if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) {
1483 if (sr->l_len <= 0) {
1485 goto out_inode_unlock;
1489 if (file && should_remove_suid(file->f_path.dentry)) {
1490 ret = __ocfs2_write_remove_suid(inode, di_bh);
1493 goto out_inode_unlock;
1497 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1499 case OCFS2_IOC_RESVSP:
1500 case OCFS2_IOC_RESVSP64:
1502 * This takes unsigned offsets, but the signed ones we
1503 * pass have been checked against overflow above.
1505 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1508 case OCFS2_IOC_UNRESVSP:
1509 case OCFS2_IOC_UNRESVSP64:
1510 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1516 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1519 goto out_inode_unlock;
1523 * We update c/mtime for these changes
1525 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1526 if (IS_ERR(handle)) {
1527 ret = PTR_ERR(handle);
1529 goto out_inode_unlock;
1532 if (change_size && i_size_read(inode) < size)
1533 i_size_write(inode, size);
1535 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1536 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1540 ocfs2_commit_trans(osb, handle);
1544 ocfs2_inode_unlock(inode, 1);
1546 ocfs2_rw_unlock(inode, 1);
1549 mutex_unlock(&inode->i_mutex);
1553 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1554 struct ocfs2_space_resv *sr)
1556 struct inode *inode = file->f_path.dentry->d_inode;
1557 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);;
1559 if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1560 !ocfs2_writes_unwritten_extents(osb))
1562 else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1563 !ocfs2_sparse_alloc(osb))
1566 if (!S_ISREG(inode->i_mode))
1569 if (!(file->f_mode & FMODE_WRITE))
1572 return __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1575 static long ocfs2_fallocate(struct inode *inode, int mode, loff_t offset,
1578 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1579 struct ocfs2_space_resv sr;
1580 int change_size = 1;
1582 if (!ocfs2_writes_unwritten_extents(osb))
1585 if (S_ISDIR(inode->i_mode))
1588 if (mode & FALLOC_FL_KEEP_SIZE)
1592 sr.l_start = (s64)offset;
1593 sr.l_len = (s64)len;
1595 return __ocfs2_change_file_space(NULL, inode, offset,
1596 OCFS2_IOC_RESVSP64, &sr, change_size);
1599 static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
1605 int ret = 0, meta_level = 0;
1606 struct inode *inode = dentry->d_inode;
1607 loff_t saved_pos, end;
1610 * We start with a read level meta lock and only jump to an ex
1611 * if we need to make modifications here.
1614 ret = ocfs2_inode_lock(inode, NULL, meta_level);
1621 /* Clear suid / sgid if necessary. We do this here
1622 * instead of later in the write path because
1623 * remove_suid() calls ->setattr without any hint that
1624 * we may have already done our cluster locking. Since
1625 * ocfs2_setattr() *must* take cluster locks to
1626 * proceeed, this will lead us to recursively lock the
1627 * inode. There's also the dinode i_size state which
1628 * can be lost via setattr during extending writes (we
1629 * set inode->i_size at the end of a write. */
1630 if (should_remove_suid(dentry)) {
1631 if (meta_level == 0) {
1632 ocfs2_inode_unlock(inode, meta_level);
1637 ret = ocfs2_write_remove_suid(inode);
1644 /* work on a copy of ppos until we're sure that we won't have
1645 * to recalculate it due to relocking. */
1647 saved_pos = i_size_read(inode);
1648 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos);
1653 end = saved_pos + count;
1656 * Skip the O_DIRECT checks if we don't need
1659 if (!direct_io || !(*direct_io))
1663 * There's no sane way to do direct writes to an inode
1666 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1672 * Allowing concurrent direct writes means
1673 * i_size changes wouldn't be synchronized, so
1674 * one node could wind up truncating another
1677 if (end > i_size_read(inode)) {
1683 * We don't fill holes during direct io, so
1684 * check for them here. If any are found, the
1685 * caller will have to retake some cluster
1686 * locks and initiate the io as buffered.
1688 ret = ocfs2_check_range_for_holes(inode, saved_pos, count);
1701 ocfs2_inode_unlock(inode, meta_level);
1707 static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
1708 const struct iovec *iov,
1709 unsigned long nr_segs,
1712 int ret, direct_io, appending, rw_level, have_alloc_sem = 0;
1714 ssize_t written = 0;
1715 size_t ocount; /* original count */
1716 size_t count; /* after file limit checks */
1717 loff_t old_size, *ppos = &iocb->ki_pos;
1719 struct file *file = iocb->ki_filp;
1720 struct inode *inode = file->f_path.dentry->d_inode;
1721 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1723 mlog_entry("(0x%p, %u, '%.*s')\n", file,
1724 (unsigned int)nr_segs,
1725 file->f_path.dentry->d_name.len,
1726 file->f_path.dentry->d_name.name);
1728 if (iocb->ki_left == 0)
1731 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1733 appending = file->f_flags & O_APPEND ? 1 : 0;
1734 direct_io = file->f_flags & O_DIRECT ? 1 : 0;
1736 mutex_lock(&inode->i_mutex);
1739 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
1741 down_read(&inode->i_alloc_sem);
1745 /* concurrent O_DIRECT writes are allowed */
1746 rw_level = !direct_io;
1747 ret = ocfs2_rw_lock(inode, rw_level);
1753 can_do_direct = direct_io;
1754 ret = ocfs2_prepare_inode_for_write(file->f_path.dentry, ppos,
1755 iocb->ki_left, appending,
1763 * We can't complete the direct I/O as requested, fall back to
1766 if (direct_io && !can_do_direct) {
1767 ocfs2_rw_unlock(inode, rw_level);
1768 up_read(&inode->i_alloc_sem);
1778 * To later detect whether a journal commit for sync writes is
1779 * necessary, we sample i_size, and cluster count here.
1781 old_size = i_size_read(inode);
1782 old_clusters = OCFS2_I(inode)->ip_clusters;
1784 /* communicate with ocfs2_dio_end_io */
1785 ocfs2_iocb_set_rw_locked(iocb, rw_level);
1788 ret = generic_segment_checks(iov, &nr_segs, &ocount,
1793 ret = generic_write_checks(file, ppos, &count,
1794 S_ISBLK(inode->i_mode));
1798 written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
1799 ppos, count, ocount);
1802 * direct write may have instantiated a few
1803 * blocks outside i_size. Trim these off again.
1804 * Don't need i_size_read because we hold i_mutex.
1806 if (*ppos + count > inode->i_size)
1807 vmtruncate(inode, inode->i_size);
1812 written = generic_file_aio_write_nolock(iocb, iov, nr_segs,
1817 /* buffered aio wouldn't have proper lock coverage today */
1818 BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
1820 if ((file->f_flags & O_SYNC && !direct_io) || IS_SYNC(inode)) {
1822 * The generic write paths have handled getting data
1823 * to disk, but since we don't make use of the dirty
1824 * inode list, a manual journal commit is necessary
1827 if (old_size != i_size_read(inode) ||
1828 old_clusters != OCFS2_I(inode)->ip_clusters) {
1829 ret = jbd2_journal_force_commit(osb->journal->j_journal);
1836 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
1837 * function pointer which is called when o_direct io completes so that
1838 * it can unlock our rw lock. (it's the clustered equivalent of
1839 * i_alloc_sem; protects truncate from racing with pending ios).
1840 * Unfortunately there are error cases which call end_io and others
1841 * that don't. so we don't have to unlock the rw_lock if either an
1842 * async dio is going to do it in the future or an end_io after an
1843 * error has already done it.
1845 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
1852 ocfs2_rw_unlock(inode, rw_level);
1856 up_read(&inode->i_alloc_sem);
1858 mutex_unlock(&inode->i_mutex);
1861 return written ? written : ret;
1864 static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
1871 struct inode *inode = out->f_path.dentry->d_inode;
1873 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe,
1875 out->f_path.dentry->d_name.len,
1876 out->f_path.dentry->d_name.name);
1878 inode_double_lock(inode, pipe->inode);
1880 ret = ocfs2_rw_lock(inode, 1);
1886 ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, ppos, len, 0,
1893 ret = generic_file_splice_write_nolock(pipe, out, ppos, len, flags);
1896 ocfs2_rw_unlock(inode, 1);
1898 inode_double_unlock(inode, pipe->inode);
1904 static ssize_t ocfs2_file_splice_read(struct file *in,
1906 struct pipe_inode_info *pipe,
1911 struct inode *inode = in->f_path.dentry->d_inode;
1913 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe,
1915 in->f_path.dentry->d_name.len,
1916 in->f_path.dentry->d_name.name);
1919 * See the comment in ocfs2_file_aio_read()
1921 ret = ocfs2_inode_lock(inode, NULL, 0);
1926 ocfs2_inode_unlock(inode, 0);
1928 ret = generic_file_splice_read(in, ppos, pipe, len, flags);
1935 static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
1936 const struct iovec *iov,
1937 unsigned long nr_segs,
1940 int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
1941 struct file *filp = iocb->ki_filp;
1942 struct inode *inode = filp->f_path.dentry->d_inode;
1944 mlog_entry("(0x%p, %u, '%.*s')\n", filp,
1945 (unsigned int)nr_segs,
1946 filp->f_path.dentry->d_name.len,
1947 filp->f_path.dentry->d_name.name);
1956 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
1957 * need locks to protect pending reads from racing with truncate.
1959 if (filp->f_flags & O_DIRECT) {
1960 down_read(&inode->i_alloc_sem);
1963 ret = ocfs2_rw_lock(inode, 0);
1969 /* communicate with ocfs2_dio_end_io */
1970 ocfs2_iocb_set_rw_locked(iocb, rw_level);
1974 * We're fine letting folks race truncates and extending
1975 * writes with read across the cluster, just like they can
1976 * locally. Hence no rw_lock during read.
1978 * Take and drop the meta data lock to update inode fields
1979 * like i_size. This allows the checks down below
1980 * generic_file_aio_read() a chance of actually working.
1982 ret = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level);
1987 ocfs2_inode_unlock(inode, lock_level);
1989 ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
1991 mlog(0, "generic_file_aio_read returned -EINVAL\n");
1993 /* buffered aio wouldn't have proper lock coverage today */
1994 BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
1996 /* see ocfs2_file_aio_write */
1997 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2004 up_read(&inode->i_alloc_sem);
2006 ocfs2_rw_unlock(inode, rw_level);
2012 const struct inode_operations ocfs2_file_iops = {
2013 .setattr = ocfs2_setattr,
2014 .getattr = ocfs2_getattr,
2015 .permission = ocfs2_permission,
2016 .setxattr = generic_setxattr,
2017 .getxattr = generic_getxattr,
2018 .listxattr = ocfs2_listxattr,
2019 .removexattr = generic_removexattr,
2020 .fallocate = ocfs2_fallocate,
2021 .fiemap = ocfs2_fiemap,
2024 const struct inode_operations ocfs2_special_file_iops = {
2025 .setattr = ocfs2_setattr,
2026 .getattr = ocfs2_getattr,
2027 .permission = ocfs2_permission,
2031 * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2032 * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2034 const struct file_operations ocfs2_fops = {
2035 .llseek = generic_file_llseek,
2036 .read = do_sync_read,
2037 .write = do_sync_write,
2039 .fsync = ocfs2_sync_file,
2040 .release = ocfs2_file_release,
2041 .open = ocfs2_file_open,
2042 .aio_read = ocfs2_file_aio_read,
2043 .aio_write = ocfs2_file_aio_write,
2044 .unlocked_ioctl = ocfs2_ioctl,
2045 #ifdef CONFIG_COMPAT
2046 .compat_ioctl = ocfs2_compat_ioctl,
2049 .flock = ocfs2_flock,
2050 .splice_read = ocfs2_file_splice_read,
2051 .splice_write = ocfs2_file_splice_write,
2054 const struct file_operations ocfs2_dops = {
2055 .llseek = generic_file_llseek,
2056 .read = generic_read_dir,
2057 .readdir = ocfs2_readdir,
2058 .fsync = ocfs2_sync_file,
2059 .release = ocfs2_dir_release,
2060 .open = ocfs2_dir_open,
2061 .unlocked_ioctl = ocfs2_ioctl,
2062 #ifdef CONFIG_COMPAT
2063 .compat_ioctl = ocfs2_compat_ioctl,
2066 .flock = ocfs2_flock,
2070 * POSIX-lockless variants of our file_operations.
2072 * These will be used if the underlying cluster stack does not support
2073 * posix file locking, if the user passes the "localflocks" mount
2074 * option, or if we have a local-only fs.
2076 * ocfs2_flock is in here because all stacks handle UNIX file locks,
2077 * so we still want it in the case of no stack support for
2078 * plocks. Internally, it will do the right thing when asked to ignore
2081 const struct file_operations ocfs2_fops_no_plocks = {
2082 .llseek = generic_file_llseek,
2083 .read = do_sync_read,
2084 .write = do_sync_write,
2086 .fsync = ocfs2_sync_file,
2087 .release = ocfs2_file_release,
2088 .open = ocfs2_file_open,
2089 .aio_read = ocfs2_file_aio_read,
2090 .aio_write = ocfs2_file_aio_write,
2091 .unlocked_ioctl = ocfs2_ioctl,
2092 #ifdef CONFIG_COMPAT
2093 .compat_ioctl = ocfs2_compat_ioctl,
2095 .flock = ocfs2_flock,
2096 .splice_read = ocfs2_file_splice_read,
2097 .splice_write = ocfs2_file_splice_write,
2100 const struct file_operations ocfs2_dops_no_plocks = {
2101 .llseek = generic_file_llseek,
2102 .read = generic_read_dir,
2103 .readdir = ocfs2_readdir,
2104 .fsync = ocfs2_sync_file,
2105 .release = ocfs2_dir_release,
2106 .open = ocfs2_dir_open,
2107 .unlocked_ioctl = ocfs2_ioctl,
2108 #ifdef CONFIG_COMPAT
2109 .compat_ioctl = ocfs2_compat_ioctl,
2111 .flock = ocfs2_flock,