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"
59 #include "buffer_head_io.h"
61 static int ocfs2_sync_inode(struct inode *inode)
63 filemap_fdatawrite(inode->i_mapping);
64 return sync_mapping_buffers(inode->i_mapping);
67 static int ocfs2_init_file_private(struct inode *inode, struct file *file)
69 struct ocfs2_file_private *fp;
71 fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
76 mutex_init(&fp->fp_mutex);
77 ocfs2_file_lock_res_init(&fp->fp_flock, fp);
78 file->private_data = fp;
83 static void ocfs2_free_file_private(struct inode *inode, struct file *file)
85 struct ocfs2_file_private *fp = file->private_data;
86 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
89 ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
90 ocfs2_lock_res_free(&fp->fp_flock);
92 file->private_data = NULL;
96 static int ocfs2_file_open(struct inode *inode, struct file *file)
99 int mode = file->f_flags;
100 struct ocfs2_inode_info *oi = OCFS2_I(inode);
102 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
103 file->f_path.dentry->d_name.len, file->f_path.dentry->d_name.name);
105 spin_lock(&oi->ip_lock);
107 /* Check that the inode hasn't been wiped from disk by another
108 * node. If it hasn't then we're safe as long as we hold the
109 * spin lock until our increment of open count. */
110 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
111 spin_unlock(&oi->ip_lock);
118 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
121 spin_unlock(&oi->ip_lock);
123 status = ocfs2_init_file_private(inode, file);
126 * We want to set open count back if we're failing the
129 spin_lock(&oi->ip_lock);
131 spin_unlock(&oi->ip_lock);
139 static int ocfs2_file_release(struct inode *inode, struct file *file)
141 struct ocfs2_inode_info *oi = OCFS2_I(inode);
143 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
144 file->f_path.dentry->d_name.len,
145 file->f_path.dentry->d_name.name);
147 spin_lock(&oi->ip_lock);
148 if (!--oi->ip_open_count)
149 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
150 spin_unlock(&oi->ip_lock);
152 ocfs2_free_file_private(inode, file);
159 static int ocfs2_dir_open(struct inode *inode, struct file *file)
161 return ocfs2_init_file_private(inode, file);
164 static int ocfs2_dir_release(struct inode *inode, struct file *file)
166 ocfs2_free_file_private(inode, file);
170 static int ocfs2_sync_file(struct file *file,
171 struct dentry *dentry,
176 struct inode *inode = dentry->d_inode;
177 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
179 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync,
180 dentry->d_name.len, dentry->d_name.name);
182 err = ocfs2_sync_inode(dentry->d_inode);
186 journal = osb->journal->j_journal;
187 err = journal_force_commit(journal);
192 return (err < 0) ? -EIO : 0;
195 int ocfs2_should_update_atime(struct inode *inode,
196 struct vfsmount *vfsmnt)
199 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
201 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
204 if ((inode->i_flags & S_NOATIME) ||
205 ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
209 * We can be called with no vfsmnt structure - NFSD will
212 * Note that our action here is different than touch_atime() -
213 * if we can't tell whether this is a noatime mount, then we
214 * don't know whether to trust the value of s_atime_quantum.
219 if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
220 ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
223 if (vfsmnt->mnt_flags & MNT_RELATIME) {
224 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
225 (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
232 if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
238 int ocfs2_update_inode_atime(struct inode *inode,
239 struct buffer_head *bh)
242 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
244 struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
248 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
249 if (handle == NULL) {
255 ret = ocfs2_journal_access(handle, inode, bh,
256 OCFS2_JOURNAL_ACCESS_WRITE);
263 * Don't use ocfs2_mark_inode_dirty() here as we don't always
264 * have i_mutex to guard against concurrent changes to other
267 inode->i_atime = CURRENT_TIME;
268 di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
269 di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
271 ret = ocfs2_journal_dirty(handle, bh);
276 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
282 static int ocfs2_set_inode_size(handle_t *handle,
284 struct buffer_head *fe_bh,
290 i_size_write(inode, new_i_size);
291 inode->i_blocks = ocfs2_inode_sector_count(inode);
292 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
294 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
305 static int ocfs2_simple_size_update(struct inode *inode,
306 struct buffer_head *di_bh,
310 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
311 handle_t *handle = NULL;
313 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
314 if (handle == NULL) {
320 ret = ocfs2_set_inode_size(handle, inode, di_bh,
325 ocfs2_commit_trans(osb, handle);
330 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
332 struct buffer_head *fe_bh,
337 struct ocfs2_dinode *di;
342 /* TODO: This needs to actually orphan the inode in this
345 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
346 if (IS_ERR(handle)) {
347 status = PTR_ERR(handle);
352 status = ocfs2_journal_access(handle, inode, fe_bh,
353 OCFS2_JOURNAL_ACCESS_WRITE);
360 * Do this before setting i_size.
362 cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
363 status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
370 i_size_write(inode, new_i_size);
371 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
373 di = (struct ocfs2_dinode *) fe_bh->b_data;
374 di->i_size = cpu_to_le64(new_i_size);
375 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
376 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
378 status = ocfs2_journal_dirty(handle, fe_bh);
383 ocfs2_commit_trans(osb, handle);
390 static int ocfs2_truncate_file(struct inode *inode,
391 struct buffer_head *di_bh,
395 struct ocfs2_dinode *fe = NULL;
396 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
397 struct ocfs2_truncate_context *tc = NULL;
399 mlog_entry("(inode = %llu, new_i_size = %llu\n",
400 (unsigned long long)OCFS2_I(inode)->ip_blkno,
401 (unsigned long long)new_i_size);
403 fe = (struct ocfs2_dinode *) di_bh->b_data;
404 if (!OCFS2_IS_VALID_DINODE(fe)) {
405 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
410 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
411 "Inode %llu, inode i_size = %lld != di "
412 "i_size = %llu, i_flags = 0x%x\n",
413 (unsigned long long)OCFS2_I(inode)->ip_blkno,
415 (unsigned long long)le64_to_cpu(fe->i_size),
416 le32_to_cpu(fe->i_flags));
418 if (new_i_size > le64_to_cpu(fe->i_size)) {
419 mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
420 (unsigned long long)le64_to_cpu(fe->i_size),
421 (unsigned long long)new_i_size);
427 mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
428 (unsigned long long)le64_to_cpu(fe->i_blkno),
429 (unsigned long long)le64_to_cpu(fe->i_size),
430 (unsigned long long)new_i_size);
432 /* lets handle the simple truncate cases before doing any more
433 * cluster locking. */
434 if (new_i_size == le64_to_cpu(fe->i_size))
437 down_write(&OCFS2_I(inode)->ip_alloc_sem);
440 * The inode lock forced other nodes to sync and drop their
441 * pages, which (correctly) happens even if we have a truncate
442 * without allocation change - ocfs2 cluster sizes can be much
443 * greater than page size, so we have to truncate them
446 unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
447 truncate_inode_pages(inode->i_mapping, new_i_size);
449 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
450 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
451 i_size_read(inode), 1);
455 goto bail_unlock_sem;
458 /* alright, we're going to need to do a full blown alloc size
459 * change. Orphan the inode so that recovery can complete the
460 * truncate if necessary. This does the task of marking
462 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
465 goto bail_unlock_sem;
468 status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc);
471 goto bail_unlock_sem;
474 status = ocfs2_commit_truncate(osb, inode, di_bh, tc);
477 goto bail_unlock_sem;
480 /* TODO: orphan dir cleanup here. */
482 up_write(&OCFS2_I(inode)->ip_alloc_sem);
491 * extend file allocation only here.
492 * we'll update all the disk stuff, and oip->alloc_size
494 * expect stuff to be locked, a transaction started and enough data /
495 * metadata reservations in the contexts.
497 * Will return -EAGAIN, and a reason if a restart is needed.
498 * If passed in, *reason will always be set, even in error.
500 int ocfs2_add_inode_data(struct ocfs2_super *osb,
505 struct buffer_head *fe_bh,
507 struct ocfs2_alloc_context *data_ac,
508 struct ocfs2_alloc_context *meta_ac,
509 enum ocfs2_alloc_restarted *reason_ret)
511 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
512 struct ocfs2_extent_list *el = &fe->id2.i_list;
514 return ocfs2_add_clusters_in_btree(osb, inode, logical_offset,
515 clusters_to_add, mark_unwritten,
517 data_ac, meta_ac, reason_ret,
518 OCFS2_DINODE_EXTENT, NULL);
521 static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
522 u32 clusters_to_add, int mark_unwritten)
525 int restart_func = 0;
528 struct buffer_head *bh = NULL;
529 struct ocfs2_dinode *fe = NULL;
530 handle_t *handle = NULL;
531 struct ocfs2_alloc_context *data_ac = NULL;
532 struct ocfs2_alloc_context *meta_ac = NULL;
533 enum ocfs2_alloc_restarted why;
534 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
536 mlog_entry("(clusters_to_add = %u)\n", clusters_to_add);
539 * This function only exists for file systems which don't
542 BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
544 status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &bh,
545 OCFS2_BH_CACHED, inode);
551 fe = (struct ocfs2_dinode *) bh->b_data;
552 if (!OCFS2_IS_VALID_DINODE(fe)) {
553 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
559 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
561 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, "
562 "clusters_to_add = %u\n",
563 (unsigned long long)OCFS2_I(inode)->ip_blkno,
564 (long long)i_size_read(inode), le32_to_cpu(fe->i_clusters),
566 status = ocfs2_lock_allocators(inode, bh, &fe->id2.i_list,
567 clusters_to_add, 0, &data_ac,
568 &meta_ac, OCFS2_DINODE_EXTENT, NULL);
574 credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list,
576 handle = ocfs2_start_trans(osb, credits);
577 if (IS_ERR(handle)) {
578 status = PTR_ERR(handle);
584 restarted_transaction:
585 /* reserve a write to the file entry early on - that we if we
586 * run out of credits in the allocation path, we can still
588 status = ocfs2_journal_access(handle, inode, bh,
589 OCFS2_JOURNAL_ACCESS_WRITE);
595 prev_clusters = OCFS2_I(inode)->ip_clusters;
597 status = ocfs2_add_inode_data(osb,
607 if ((status < 0) && (status != -EAGAIN)) {
608 if (status != -ENOSPC)
613 status = ocfs2_journal_dirty(handle, bh);
619 spin_lock(&OCFS2_I(inode)->ip_lock);
620 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
621 spin_unlock(&OCFS2_I(inode)->ip_lock);
623 if (why != RESTART_NONE && clusters_to_add) {
624 if (why == RESTART_META) {
625 mlog(0, "restarting function.\n");
628 BUG_ON(why != RESTART_TRANS);
630 mlog(0, "restarting transaction.\n");
631 /* TODO: This can be more intelligent. */
632 credits = ocfs2_calc_extend_credits(osb->sb,
635 status = ocfs2_extend_trans(handle, credits);
637 /* handle still has to be committed at
643 goto restarted_transaction;
647 mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
648 le32_to_cpu(fe->i_clusters),
649 (unsigned long long)le64_to_cpu(fe->i_size));
650 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
651 OCFS2_I(inode)->ip_clusters, (long long)i_size_read(inode));
655 ocfs2_commit_trans(osb, handle);
659 ocfs2_free_alloc_context(data_ac);
663 ocfs2_free_alloc_context(meta_ac);
666 if ((!status) && restart_func) {
679 /* Some parts of this taken from generic_cont_expand, which turned out
680 * to be too fragile to do exactly what we need without us having to
681 * worry about recursive locking in ->prepare_write() and
682 * ->commit_write(). */
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 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
943 status = ocfs2_extend_file(inode, bh, attr->ia_size);
945 if (status != -ENOSPC)
952 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
953 if (IS_ERR(handle)) {
954 status = PTR_ERR(handle);
960 * This will intentionally not wind up calling vmtruncate(),
961 * since all the work for a size change has been done above.
962 * Otherwise, we could get into problems with truncate as
963 * ip_alloc_sem is used there to protect against i_size
966 status = inode_setattr(inode, attr);
972 status = ocfs2_mark_inode_dirty(handle, inode, bh);
977 ocfs2_commit_trans(osb, handle);
979 ocfs2_inode_unlock(inode, 1);
982 ocfs2_rw_unlock(inode, 1);
991 int ocfs2_getattr(struct vfsmount *mnt,
992 struct dentry *dentry,
995 struct inode *inode = dentry->d_inode;
996 struct super_block *sb = dentry->d_inode->i_sb;
997 struct ocfs2_super *osb = sb->s_fs_info;
1002 err = ocfs2_inode_revalidate(dentry);
1009 generic_fillattr(inode, stat);
1011 /* We set the blksize from the cluster size for performance */
1012 stat->blksize = osb->s_clustersize;
1020 int ocfs2_permission(struct inode *inode, int mask)
1026 ret = ocfs2_inode_lock(inode, NULL, 0);
1033 ret = generic_permission(inode, mask, NULL);
1035 ocfs2_inode_unlock(inode, 0);
1041 static int __ocfs2_write_remove_suid(struct inode *inode,
1042 struct buffer_head *bh)
1046 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1047 struct ocfs2_dinode *di;
1049 mlog_entry("(Inode %llu, mode 0%o)\n",
1050 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_mode);
1052 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1053 if (handle == NULL) {
1059 ret = ocfs2_journal_access(handle, inode, bh,
1060 OCFS2_JOURNAL_ACCESS_WRITE);
1066 inode->i_mode &= ~S_ISUID;
1067 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1068 inode->i_mode &= ~S_ISGID;
1070 di = (struct ocfs2_dinode *) bh->b_data;
1071 di->i_mode = cpu_to_le16(inode->i_mode);
1073 ret = ocfs2_journal_dirty(handle, bh);
1078 ocfs2_commit_trans(osb, handle);
1085 * Will look for holes and unwritten extents in the range starting at
1086 * pos for count bytes (inclusive).
1088 static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1092 unsigned int extent_flags;
1093 u32 cpos, clusters, extent_len, phys_cpos;
1094 struct super_block *sb = inode->i_sb;
1096 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1097 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1100 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1107 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
1112 if (extent_len > clusters)
1113 extent_len = clusters;
1115 clusters -= extent_len;
1122 static int ocfs2_write_remove_suid(struct inode *inode)
1125 struct buffer_head *bh = NULL;
1126 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1128 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
1129 oi->ip_blkno, &bh, OCFS2_BH_CACHED, inode);
1135 ret = __ocfs2_write_remove_suid(inode, bh);
1142 * Allocate enough extents to cover the region starting at byte offset
1143 * start for len bytes. Existing extents are skipped, any extents
1144 * added are marked as "unwritten".
1146 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1150 u32 cpos, phys_cpos, clusters, alloc_size;
1151 u64 end = start + len;
1152 struct buffer_head *di_bh = NULL;
1154 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1155 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
1156 OCFS2_I(inode)->ip_blkno, &di_bh,
1157 OCFS2_BH_CACHED, inode);
1164 * Nothing to do if the requested reservation range
1165 * fits within the inode.
1167 if (ocfs2_size_fits_inline_data(di_bh, end))
1170 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1178 * We consider both start and len to be inclusive.
1180 cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1181 clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1185 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1193 * Hole or existing extent len can be arbitrary, so
1194 * cap it to our own allocation request.
1196 if (alloc_size > clusters)
1197 alloc_size = clusters;
1201 * We already have an allocation at this
1202 * region so we can safely skip it.
1207 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1216 clusters -= alloc_size;
1226 static int __ocfs2_remove_inode_range(struct inode *inode,
1227 struct buffer_head *di_bh,
1228 u32 cpos, u32 phys_cpos, u32 len,
1229 struct ocfs2_cached_dealloc_ctxt *dealloc)
1232 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
1233 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1234 struct inode *tl_inode = osb->osb_tl_inode;
1236 struct ocfs2_alloc_context *meta_ac = NULL;
1237 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1239 ret = ocfs2_lock_allocators(inode, di_bh, &di->id2.i_list,
1240 0, 1, NULL, &meta_ac,
1241 OCFS2_DINODE_EXTENT, NULL);
1247 mutex_lock(&tl_inode->i_mutex);
1249 if (ocfs2_truncate_log_needs_flush(osb)) {
1250 ret = __ocfs2_flush_truncate_log(osb);
1257 handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
1258 if (handle == NULL) {
1264 ret = ocfs2_journal_access(handle, inode, di_bh,
1265 OCFS2_JOURNAL_ACCESS_WRITE);
1271 ret = ocfs2_remove_extent(inode, di_bh, cpos, len, handle, meta_ac,
1272 dealloc, OCFS2_DINODE_EXTENT, NULL);
1278 OCFS2_I(inode)->ip_clusters -= len;
1279 di->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1281 ret = ocfs2_journal_dirty(handle, di_bh);
1287 ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len);
1292 ocfs2_commit_trans(osb, handle);
1294 mutex_unlock(&tl_inode->i_mutex);
1297 ocfs2_free_alloc_context(meta_ac);
1303 * Truncate a byte range, avoiding pages within partial clusters. This
1304 * preserves those pages for the zeroing code to write to.
1306 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1309 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1311 struct address_space *mapping = inode->i_mapping;
1313 start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1314 end = byte_start + byte_len;
1315 end = end & ~(osb->s_clustersize - 1);
1318 unmap_mapping_range(mapping, start, end - start, 0);
1319 truncate_inode_pages_range(mapping, start, end - 1);
1323 static int ocfs2_zero_partial_clusters(struct inode *inode,
1327 u64 tmpend, end = start + len;
1328 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1329 unsigned int csize = osb->s_clustersize;
1333 * The "start" and "end" values are NOT necessarily part of
1334 * the range whose allocation is being deleted. Rather, this
1335 * is what the user passed in with the request. We must zero
1336 * partial clusters here. There's no need to worry about
1337 * physical allocation - the zeroing code knows to skip holes.
1339 mlog(0, "byte start: %llu, end: %llu\n",
1340 (unsigned long long)start, (unsigned long long)end);
1343 * If both edges are on a cluster boundary then there's no
1344 * zeroing required as the region is part of the allocation to
1347 if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1350 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1351 if (handle == NULL) {
1358 * We want to get the byte offset of the end of the 1st cluster.
1360 tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1364 mlog(0, "1st range: start: %llu, tmpend: %llu\n",
1365 (unsigned long long)start, (unsigned long long)tmpend);
1367 ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1373 * This may make start and end equal, but the zeroing
1374 * code will skip any work in that case so there's no
1375 * need to catch it up here.
1377 start = end & ~(osb->s_clustersize - 1);
1379 mlog(0, "2nd range: start: %llu, end: %llu\n",
1380 (unsigned long long)start, (unsigned long long)end);
1382 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1387 ocfs2_commit_trans(osb, handle);
1392 static int ocfs2_remove_inode_range(struct inode *inode,
1393 struct buffer_head *di_bh, u64 byte_start,
1397 u32 trunc_start, trunc_len, cpos, phys_cpos, alloc_size;
1398 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1399 struct ocfs2_cached_dealloc_ctxt dealloc;
1400 struct address_space *mapping = inode->i_mapping;
1402 ocfs2_init_dealloc_ctxt(&dealloc);
1407 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1408 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1409 byte_start + byte_len, 0);
1415 * There's no need to get fancy with the page cache
1416 * truncate of an inline-data inode. We're talking
1417 * about less than a page here, which will be cached
1418 * in the dinode buffer anyway.
1420 unmap_mapping_range(mapping, 0, 0, 0);
1421 truncate_inode_pages(mapping, 0);
1425 trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1426 trunc_len = (byte_start + byte_len) >> osb->s_clustersize_bits;
1427 if (trunc_len >= trunc_start)
1428 trunc_len -= trunc_start;
1432 mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u\n",
1433 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1434 (unsigned long long)byte_start,
1435 (unsigned long long)byte_len, trunc_start, trunc_len);
1437 ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1445 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1452 if (alloc_size > trunc_len)
1453 alloc_size = trunc_len;
1455 /* Only do work for non-holes */
1456 if (phys_cpos != 0) {
1457 ret = __ocfs2_remove_inode_range(inode, di_bh, cpos,
1458 phys_cpos, alloc_size,
1467 trunc_len -= alloc_size;
1470 ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1473 ocfs2_schedule_truncate_log_flush(osb, 1);
1474 ocfs2_run_deallocs(osb, &dealloc);
1480 * Parts of this function taken from xfs_change_file_space()
1482 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1483 loff_t f_pos, unsigned int cmd,
1484 struct ocfs2_space_resv *sr,
1490 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1491 struct buffer_head *di_bh = NULL;
1493 unsigned long long max_off = inode->i_sb->s_maxbytes;
1495 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1498 mutex_lock(&inode->i_mutex);
1501 * This prevents concurrent writes on other nodes
1503 ret = ocfs2_rw_lock(inode, 1);
1509 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1515 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1517 goto out_inode_unlock;
1520 switch (sr->l_whence) {
1521 case 0: /*SEEK_SET*/
1523 case 1: /*SEEK_CUR*/
1524 sr->l_start += f_pos;
1526 case 2: /*SEEK_END*/
1527 sr->l_start += i_size_read(inode);
1531 goto out_inode_unlock;
1535 llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1538 || sr->l_start > max_off
1539 || (sr->l_start + llen) < 0
1540 || (sr->l_start + llen) > max_off) {
1542 goto out_inode_unlock;
1544 size = sr->l_start + sr->l_len;
1546 if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) {
1547 if (sr->l_len <= 0) {
1549 goto out_inode_unlock;
1553 if (file && should_remove_suid(file->f_path.dentry)) {
1554 ret = __ocfs2_write_remove_suid(inode, di_bh);
1557 goto out_inode_unlock;
1561 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1563 case OCFS2_IOC_RESVSP:
1564 case OCFS2_IOC_RESVSP64:
1566 * This takes unsigned offsets, but the signed ones we
1567 * pass have been checked against overflow above.
1569 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1572 case OCFS2_IOC_UNRESVSP:
1573 case OCFS2_IOC_UNRESVSP64:
1574 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1580 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1583 goto out_inode_unlock;
1587 * We update c/mtime for these changes
1589 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1590 if (IS_ERR(handle)) {
1591 ret = PTR_ERR(handle);
1593 goto out_inode_unlock;
1596 if (change_size && i_size_read(inode) < size)
1597 i_size_write(inode, size);
1599 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1600 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1604 ocfs2_commit_trans(osb, handle);
1608 ocfs2_inode_unlock(inode, 1);
1610 ocfs2_rw_unlock(inode, 1);
1613 mutex_unlock(&inode->i_mutex);
1617 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1618 struct ocfs2_space_resv *sr)
1620 struct inode *inode = file->f_path.dentry->d_inode;
1621 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);;
1623 if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1624 !ocfs2_writes_unwritten_extents(osb))
1626 else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1627 !ocfs2_sparse_alloc(osb))
1630 if (!S_ISREG(inode->i_mode))
1633 if (!(file->f_mode & FMODE_WRITE))
1636 return __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1639 static long ocfs2_fallocate(struct inode *inode, int mode, loff_t offset,
1642 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1643 struct ocfs2_space_resv sr;
1644 int change_size = 1;
1646 if (!ocfs2_writes_unwritten_extents(osb))
1649 if (S_ISDIR(inode->i_mode))
1652 if (mode & FALLOC_FL_KEEP_SIZE)
1656 sr.l_start = (s64)offset;
1657 sr.l_len = (s64)len;
1659 return __ocfs2_change_file_space(NULL, inode, offset,
1660 OCFS2_IOC_RESVSP64, &sr, change_size);
1663 static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
1669 int ret = 0, meta_level = 0;
1670 struct inode *inode = dentry->d_inode;
1671 loff_t saved_pos, end;
1674 * We start with a read level meta lock and only jump to an ex
1675 * if we need to make modifications here.
1678 ret = ocfs2_inode_lock(inode, NULL, meta_level);
1685 /* Clear suid / sgid if necessary. We do this here
1686 * instead of later in the write path because
1687 * remove_suid() calls ->setattr without any hint that
1688 * we may have already done our cluster locking. Since
1689 * ocfs2_setattr() *must* take cluster locks to
1690 * proceeed, this will lead us to recursively lock the
1691 * inode. There's also the dinode i_size state which
1692 * can be lost via setattr during extending writes (we
1693 * set inode->i_size at the end of a write. */
1694 if (should_remove_suid(dentry)) {
1695 if (meta_level == 0) {
1696 ocfs2_inode_unlock(inode, meta_level);
1701 ret = ocfs2_write_remove_suid(inode);
1708 /* work on a copy of ppos until we're sure that we won't have
1709 * to recalculate it due to relocking. */
1711 saved_pos = i_size_read(inode);
1712 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos);
1717 end = saved_pos + count;
1720 * Skip the O_DIRECT checks if we don't need
1723 if (!direct_io || !(*direct_io))
1727 * There's no sane way to do direct writes to an inode
1730 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1736 * Allowing concurrent direct writes means
1737 * i_size changes wouldn't be synchronized, so
1738 * one node could wind up truncating another
1741 if (end > i_size_read(inode)) {
1747 * We don't fill holes during direct io, so
1748 * check for them here. If any are found, the
1749 * caller will have to retake some cluster
1750 * locks and initiate the io as buffered.
1752 ret = ocfs2_check_range_for_holes(inode, saved_pos, count);
1765 ocfs2_inode_unlock(inode, meta_level);
1771 static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
1772 const struct iovec *iov,
1773 unsigned long nr_segs,
1776 int ret, direct_io, appending, rw_level, have_alloc_sem = 0;
1778 ssize_t written = 0;
1779 size_t ocount; /* original count */
1780 size_t count; /* after file limit checks */
1781 loff_t old_size, *ppos = &iocb->ki_pos;
1783 struct file *file = iocb->ki_filp;
1784 struct inode *inode = file->f_path.dentry->d_inode;
1785 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1787 mlog_entry("(0x%p, %u, '%.*s')\n", file,
1788 (unsigned int)nr_segs,
1789 file->f_path.dentry->d_name.len,
1790 file->f_path.dentry->d_name.name);
1792 if (iocb->ki_left == 0)
1795 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1797 appending = file->f_flags & O_APPEND ? 1 : 0;
1798 direct_io = file->f_flags & O_DIRECT ? 1 : 0;
1800 mutex_lock(&inode->i_mutex);
1803 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
1805 down_read(&inode->i_alloc_sem);
1809 /* concurrent O_DIRECT writes are allowed */
1810 rw_level = !direct_io;
1811 ret = ocfs2_rw_lock(inode, rw_level);
1817 can_do_direct = direct_io;
1818 ret = ocfs2_prepare_inode_for_write(file->f_path.dentry, ppos,
1819 iocb->ki_left, appending,
1827 * We can't complete the direct I/O as requested, fall back to
1830 if (direct_io && !can_do_direct) {
1831 ocfs2_rw_unlock(inode, rw_level);
1832 up_read(&inode->i_alloc_sem);
1842 * To later detect whether a journal commit for sync writes is
1843 * necessary, we sample i_size, and cluster count here.
1845 old_size = i_size_read(inode);
1846 old_clusters = OCFS2_I(inode)->ip_clusters;
1848 /* communicate with ocfs2_dio_end_io */
1849 ocfs2_iocb_set_rw_locked(iocb, rw_level);
1852 ret = generic_segment_checks(iov, &nr_segs, &ocount,
1857 ret = generic_write_checks(file, ppos, &count,
1858 S_ISBLK(inode->i_mode));
1862 written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
1863 ppos, count, ocount);
1869 written = generic_file_aio_write_nolock(iocb, iov, nr_segs,
1874 /* buffered aio wouldn't have proper lock coverage today */
1875 BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
1877 if ((file->f_flags & O_SYNC && !direct_io) || IS_SYNC(inode)) {
1879 * The generic write paths have handled getting data
1880 * to disk, but since we don't make use of the dirty
1881 * inode list, a manual journal commit is necessary
1884 if (old_size != i_size_read(inode) ||
1885 old_clusters != OCFS2_I(inode)->ip_clusters) {
1886 ret = journal_force_commit(osb->journal->j_journal);
1893 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
1894 * function pointer which is called when o_direct io completes so that
1895 * it can unlock our rw lock. (it's the clustered equivalent of
1896 * i_alloc_sem; protects truncate from racing with pending ios).
1897 * Unfortunately there are error cases which call end_io and others
1898 * that don't. so we don't have to unlock the rw_lock if either an
1899 * async dio is going to do it in the future or an end_io after an
1900 * error has already done it.
1902 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
1909 ocfs2_rw_unlock(inode, rw_level);
1913 up_read(&inode->i_alloc_sem);
1915 mutex_unlock(&inode->i_mutex);
1918 return written ? written : ret;
1921 static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
1928 struct inode *inode = out->f_path.dentry->d_inode;
1930 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe,
1932 out->f_path.dentry->d_name.len,
1933 out->f_path.dentry->d_name.name);
1935 inode_double_lock(inode, pipe->inode);
1937 ret = ocfs2_rw_lock(inode, 1);
1943 ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, ppos, len, 0,
1950 ret = generic_file_splice_write_nolock(pipe, out, ppos, len, flags);
1953 ocfs2_rw_unlock(inode, 1);
1955 inode_double_unlock(inode, pipe->inode);
1961 static ssize_t ocfs2_file_splice_read(struct file *in,
1963 struct pipe_inode_info *pipe,
1968 struct inode *inode = in->f_path.dentry->d_inode;
1970 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe,
1972 in->f_path.dentry->d_name.len,
1973 in->f_path.dentry->d_name.name);
1976 * See the comment in ocfs2_file_aio_read()
1978 ret = ocfs2_inode_lock(inode, NULL, 0);
1983 ocfs2_inode_unlock(inode, 0);
1985 ret = generic_file_splice_read(in, ppos, pipe, len, flags);
1992 static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
1993 const struct iovec *iov,
1994 unsigned long nr_segs,
1997 int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
1998 struct file *filp = iocb->ki_filp;
1999 struct inode *inode = filp->f_path.dentry->d_inode;
2001 mlog_entry("(0x%p, %u, '%.*s')\n", filp,
2002 (unsigned int)nr_segs,
2003 filp->f_path.dentry->d_name.len,
2004 filp->f_path.dentry->d_name.name);
2013 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
2014 * need locks to protect pending reads from racing with truncate.
2016 if (filp->f_flags & O_DIRECT) {
2017 down_read(&inode->i_alloc_sem);
2020 ret = ocfs2_rw_lock(inode, 0);
2026 /* communicate with ocfs2_dio_end_io */
2027 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2031 * We're fine letting folks race truncates and extending
2032 * writes with read across the cluster, just like they can
2033 * locally. Hence no rw_lock during read.
2035 * Take and drop the meta data lock to update inode fields
2036 * like i_size. This allows the checks down below
2037 * generic_file_aio_read() a chance of actually working.
2039 ret = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level);
2044 ocfs2_inode_unlock(inode, lock_level);
2046 ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
2048 mlog(0, "generic_file_aio_read returned -EINVAL\n");
2050 /* buffered aio wouldn't have proper lock coverage today */
2051 BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
2053 /* see ocfs2_file_aio_write */
2054 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2061 up_read(&inode->i_alloc_sem);
2063 ocfs2_rw_unlock(inode, rw_level);
2069 const struct inode_operations ocfs2_file_iops = {
2070 .setattr = ocfs2_setattr,
2071 .getattr = ocfs2_getattr,
2072 .permission = ocfs2_permission,
2073 .fallocate = ocfs2_fallocate,
2074 .fiemap = ocfs2_fiemap,
2077 const struct inode_operations ocfs2_special_file_iops = {
2078 .setattr = ocfs2_setattr,
2079 .getattr = ocfs2_getattr,
2080 .permission = ocfs2_permission,
2084 * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2085 * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2087 const struct file_operations ocfs2_fops = {
2088 .llseek = generic_file_llseek,
2089 .read = do_sync_read,
2090 .write = do_sync_write,
2092 .fsync = ocfs2_sync_file,
2093 .release = ocfs2_file_release,
2094 .open = ocfs2_file_open,
2095 .aio_read = ocfs2_file_aio_read,
2096 .aio_write = ocfs2_file_aio_write,
2097 .unlocked_ioctl = ocfs2_ioctl,
2098 #ifdef CONFIG_COMPAT
2099 .compat_ioctl = ocfs2_compat_ioctl,
2102 .flock = ocfs2_flock,
2103 .splice_read = ocfs2_file_splice_read,
2104 .splice_write = ocfs2_file_splice_write,
2107 const struct file_operations ocfs2_dops = {
2108 .llseek = generic_file_llseek,
2109 .read = generic_read_dir,
2110 .readdir = ocfs2_readdir,
2111 .fsync = ocfs2_sync_file,
2112 .release = ocfs2_dir_release,
2113 .open = ocfs2_dir_open,
2114 .unlocked_ioctl = ocfs2_ioctl,
2115 #ifdef CONFIG_COMPAT
2116 .compat_ioctl = ocfs2_compat_ioctl,
2119 .flock = ocfs2_flock,
2123 * POSIX-lockless variants of our file_operations.
2125 * These will be used if the underlying cluster stack does not support
2126 * posix file locking, if the user passes the "localflocks" mount
2127 * option, or if we have a local-only fs.
2129 * ocfs2_flock is in here because all stacks handle UNIX file locks,
2130 * so we still want it in the case of no stack support for
2131 * plocks. Internally, it will do the right thing when asked to ignore
2134 const struct file_operations ocfs2_fops_no_plocks = {
2135 .llseek = generic_file_llseek,
2136 .read = do_sync_read,
2137 .write = do_sync_write,
2139 .fsync = ocfs2_sync_file,
2140 .release = ocfs2_file_release,
2141 .open = ocfs2_file_open,
2142 .aio_read = ocfs2_file_aio_read,
2143 .aio_write = ocfs2_file_aio_write,
2144 .unlocked_ioctl = ocfs2_ioctl,
2145 #ifdef CONFIG_COMPAT
2146 .compat_ioctl = ocfs2_compat_ioctl,
2148 .flock = ocfs2_flock,
2149 .splice_read = ocfs2_file_splice_read,
2150 .splice_write = ocfs2_file_splice_write,
2153 const struct file_operations ocfs2_dops_no_plocks = {
2154 .llseek = generic_file_llseek,
2155 .read = generic_read_dir,
2156 .readdir = ocfs2_readdir,
2157 .fsync = ocfs2_sync_file,
2158 .release = ocfs2_dir_release,
2159 .open = ocfs2_dir_open,
2160 .unlocked_ioctl = ocfs2_ioctl,
2161 #ifdef CONFIG_COMPAT
2162 .compat_ioctl = ocfs2_compat_ioctl,
2164 .flock = ocfs2_flock,