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"
58 #include "buffer_head_io.h"
60 static int ocfs2_sync_inode(struct inode *inode)
62 filemap_fdatawrite(inode->i_mapping);
63 return sync_mapping_buffers(inode->i_mapping);
66 static int ocfs2_file_open(struct inode *inode, struct file *file)
69 int mode = file->f_flags;
70 struct ocfs2_inode_info *oi = OCFS2_I(inode);
72 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
73 file->f_path.dentry->d_name.len, file->f_path.dentry->d_name.name);
75 spin_lock(&oi->ip_lock);
77 /* Check that the inode hasn't been wiped from disk by another
78 * node. If it hasn't then we're safe as long as we hold the
79 * spin lock until our increment of open count. */
80 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
81 spin_unlock(&oi->ip_lock);
88 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
91 spin_unlock(&oi->ip_lock);
98 static int ocfs2_file_release(struct inode *inode, struct file *file)
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,
104 file->f_path.dentry->d_name.name);
106 spin_lock(&oi->ip_lock);
107 if (!--oi->ip_open_count)
108 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
109 spin_unlock(&oi->ip_lock);
116 static int ocfs2_sync_file(struct file *file,
117 struct dentry *dentry,
122 struct inode *inode = dentry->d_inode;
123 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
125 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync,
126 dentry->d_name.len, dentry->d_name.name);
128 err = ocfs2_sync_inode(dentry->d_inode);
132 journal = osb->journal->j_journal;
133 err = journal_force_commit(journal);
138 return (err < 0) ? -EIO : 0;
141 int ocfs2_should_update_atime(struct inode *inode,
142 struct vfsmount *vfsmnt)
145 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
147 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
150 if ((inode->i_flags & S_NOATIME) ||
151 ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
155 * We can be called with no vfsmnt structure - NFSD will
158 * Note that our action here is different than touch_atime() -
159 * if we can't tell whether this is a noatime mount, then we
160 * don't know whether to trust the value of s_atime_quantum.
165 if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
166 ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
169 if (vfsmnt->mnt_flags & MNT_RELATIME) {
170 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
171 (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
178 if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
184 int ocfs2_update_inode_atime(struct inode *inode,
185 struct buffer_head *bh)
188 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
190 struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
194 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
195 if (handle == NULL) {
201 ret = ocfs2_journal_access(handle, inode, bh,
202 OCFS2_JOURNAL_ACCESS_WRITE);
209 * Don't use ocfs2_mark_inode_dirty() here as we don't always
210 * have i_mutex to guard against concurrent changes to other
213 inode->i_atime = CURRENT_TIME;
214 di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
215 di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
217 ret = ocfs2_journal_dirty(handle, bh);
222 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
228 static int ocfs2_set_inode_size(handle_t *handle,
230 struct buffer_head *fe_bh,
236 i_size_write(inode, new_i_size);
237 inode->i_blocks = ocfs2_inode_sector_count(inode);
238 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
240 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
251 static int ocfs2_simple_size_update(struct inode *inode,
252 struct buffer_head *di_bh,
256 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
257 handle_t *handle = NULL;
259 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
260 if (handle == NULL) {
266 ret = ocfs2_set_inode_size(handle, inode, di_bh,
271 ocfs2_commit_trans(osb, handle);
276 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
278 struct buffer_head *fe_bh,
283 struct ocfs2_dinode *di;
288 /* TODO: This needs to actually orphan the inode in this
291 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
292 if (IS_ERR(handle)) {
293 status = PTR_ERR(handle);
298 status = ocfs2_journal_access(handle, inode, fe_bh,
299 OCFS2_JOURNAL_ACCESS_WRITE);
306 * Do this before setting i_size.
308 cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
309 status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
316 i_size_write(inode, new_i_size);
317 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
319 di = (struct ocfs2_dinode *) fe_bh->b_data;
320 di->i_size = cpu_to_le64(new_i_size);
321 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
322 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
324 status = ocfs2_journal_dirty(handle, fe_bh);
329 ocfs2_commit_trans(osb, handle);
336 static int ocfs2_truncate_file(struct inode *inode,
337 struct buffer_head *di_bh,
341 struct ocfs2_dinode *fe = NULL;
342 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
343 struct ocfs2_truncate_context *tc = NULL;
345 mlog_entry("(inode = %llu, new_i_size = %llu\n",
346 (unsigned long long)OCFS2_I(inode)->ip_blkno,
347 (unsigned long long)new_i_size);
349 fe = (struct ocfs2_dinode *) di_bh->b_data;
350 if (!OCFS2_IS_VALID_DINODE(fe)) {
351 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
356 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
357 "Inode %llu, inode i_size = %lld != di "
358 "i_size = %llu, i_flags = 0x%x\n",
359 (unsigned long long)OCFS2_I(inode)->ip_blkno,
361 (unsigned long long)le64_to_cpu(fe->i_size),
362 le32_to_cpu(fe->i_flags));
364 if (new_i_size > le64_to_cpu(fe->i_size)) {
365 mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
366 (unsigned long long)le64_to_cpu(fe->i_size),
367 (unsigned long long)new_i_size);
373 mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
374 (unsigned long long)le64_to_cpu(fe->i_blkno),
375 (unsigned long long)le64_to_cpu(fe->i_size),
376 (unsigned long long)new_i_size);
378 /* lets handle the simple truncate cases before doing any more
379 * cluster locking. */
380 if (new_i_size == le64_to_cpu(fe->i_size))
383 down_write(&OCFS2_I(inode)->ip_alloc_sem);
385 /* This forces other nodes to sync and drop their pages. Do
386 * this even if we have a truncate without allocation change -
387 * ocfs2 cluster sizes can be much greater than page size, so
388 * we have to truncate them anyway. */
389 status = ocfs2_data_lock(inode, 1);
391 up_write(&OCFS2_I(inode)->ip_alloc_sem);
397 unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
398 truncate_inode_pages(inode->i_mapping, new_i_size);
400 /* alright, we're going to need to do a full blown alloc size
401 * change. Orphan the inode so that recovery can complete the
402 * truncate if necessary. This does the task of marking
404 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
407 goto bail_unlock_data;
410 status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc);
413 goto bail_unlock_data;
416 status = ocfs2_commit_truncate(osb, inode, di_bh, tc);
419 goto bail_unlock_data;
422 /* TODO: orphan dir cleanup here. */
424 ocfs2_data_unlock(inode, 1);
426 up_write(&OCFS2_I(inode)->ip_alloc_sem);
435 * extend allocation only here.
436 * we'll update all the disk stuff, and oip->alloc_size
438 * expect stuff to be locked, a transaction started and enough data /
439 * metadata reservations in the contexts.
441 * Will return -EAGAIN, and a reason if a restart is needed.
442 * If passed in, *reason will always be set, even in error.
444 int ocfs2_do_extend_allocation(struct ocfs2_super *osb,
449 struct buffer_head *fe_bh,
451 struct ocfs2_alloc_context *data_ac,
452 struct ocfs2_alloc_context *meta_ac,
453 enum ocfs2_alloc_restarted *reason_ret)
457 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
458 enum ocfs2_alloc_restarted reason = RESTART_NONE;
459 u32 bit_off, num_bits;
463 BUG_ON(!clusters_to_add);
466 flags = OCFS2_EXT_UNWRITTEN;
468 free_extents = ocfs2_num_free_extents(osb, inode, fe);
469 if (free_extents < 0) {
470 status = free_extents;
475 /* there are two cases which could cause us to EAGAIN in the
476 * we-need-more-metadata case:
477 * 1) we haven't reserved *any*
478 * 2) we are so fragmented, we've needed to add metadata too
480 if (!free_extents && !meta_ac) {
481 mlog(0, "we haven't reserved any metadata!\n");
483 reason = RESTART_META;
485 } else if ((!free_extents)
486 && (ocfs2_alloc_context_bits_left(meta_ac)
487 < ocfs2_extend_meta_needed(fe))) {
488 mlog(0, "filesystem is really fragmented...\n");
490 reason = RESTART_META;
494 status = __ocfs2_claim_clusters(osb, handle, data_ac, 1,
495 clusters_to_add, &bit_off, &num_bits);
497 if (status != -ENOSPC)
502 BUG_ON(num_bits > clusters_to_add);
504 /* reserve our write early -- insert_extent may update the inode */
505 status = ocfs2_journal_access(handle, inode, fe_bh,
506 OCFS2_JOURNAL_ACCESS_WRITE);
512 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
513 mlog(0, "Allocating %u clusters at block %u for inode %llu\n",
514 num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
515 status = ocfs2_insert_extent(osb, handle, inode, fe_bh,
516 *logical_offset, block, num_bits,
523 status = ocfs2_journal_dirty(handle, fe_bh);
529 clusters_to_add -= num_bits;
530 *logical_offset += num_bits;
532 if (clusters_to_add) {
533 mlog(0, "need to alloc once more, clusters = %u, wanted = "
534 "%u\n", fe->i_clusters, clusters_to_add);
536 reason = RESTART_TRANS;
542 *reason_ret = reason;
547 * For a given allocation, determine which allocators will need to be
548 * accessed, and lock them, reserving the appropriate number of bits.
550 * Sparse file systems call this from ocfs2_write_begin_nolock()
551 * and ocfs2_allocate_unwritten_extents().
553 * File systems which don't support holes call this from
554 * ocfs2_extend_allocation().
556 int ocfs2_lock_allocators(struct inode *inode, struct ocfs2_dinode *di,
557 u32 clusters_to_add, u32 extents_to_split,
558 struct ocfs2_alloc_context **data_ac,
559 struct ocfs2_alloc_context **meta_ac)
561 int ret = 0, num_free_extents;
562 unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
563 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
569 BUG_ON(clusters_to_add != 0 && data_ac == NULL);
571 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, "
572 "clusters_to_add = %u, extents_to_split = %u\n",
573 (unsigned long long)OCFS2_I(inode)->ip_blkno, i_size_read(inode),
574 le32_to_cpu(di->i_clusters), clusters_to_add, extents_to_split);
576 num_free_extents = ocfs2_num_free_extents(osb, inode, di);
577 if (num_free_extents < 0) {
578 ret = num_free_extents;
584 * Sparse allocation file systems need to be more conservative
585 * with reserving room for expansion - the actual allocation
586 * happens while we've got a journal handle open so re-taking
587 * a cluster lock (because we ran out of room for another
588 * extent) will violate ordering rules.
590 * Most of the time we'll only be seeing this 1 cluster at a time
593 * Always lock for any unwritten extents - we might want to
594 * add blocks during a split.
596 if (!num_free_extents ||
597 (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
598 ret = ocfs2_reserve_new_metadata(osb, di, meta_ac);
606 if (clusters_to_add == 0)
609 ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
619 ocfs2_free_alloc_context(*meta_ac);
624 * We cannot have an error and a non null *data_ac.
631 static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
632 u32 clusters_to_add, int mark_unwritten)
635 int restart_func = 0;
638 struct buffer_head *bh = NULL;
639 struct ocfs2_dinode *fe = NULL;
640 handle_t *handle = NULL;
641 struct ocfs2_alloc_context *data_ac = NULL;
642 struct ocfs2_alloc_context *meta_ac = NULL;
643 enum ocfs2_alloc_restarted why;
644 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
646 mlog_entry("(clusters_to_add = %u)\n", clusters_to_add);
649 * This function only exists for file systems which don't
652 BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
654 status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &bh,
655 OCFS2_BH_CACHED, inode);
661 fe = (struct ocfs2_dinode *) bh->b_data;
662 if (!OCFS2_IS_VALID_DINODE(fe)) {
663 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
669 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
671 status = ocfs2_lock_allocators(inode, fe, clusters_to_add, 0, &data_ac,
678 credits = ocfs2_calc_extend_credits(osb->sb, fe, clusters_to_add);
679 handle = ocfs2_start_trans(osb, credits);
680 if (IS_ERR(handle)) {
681 status = PTR_ERR(handle);
687 restarted_transaction:
688 /* reserve a write to the file entry early on - that we if we
689 * run out of credits in the allocation path, we can still
691 status = ocfs2_journal_access(handle, inode, bh,
692 OCFS2_JOURNAL_ACCESS_WRITE);
698 prev_clusters = OCFS2_I(inode)->ip_clusters;
700 status = ocfs2_do_extend_allocation(osb,
710 if ((status < 0) && (status != -EAGAIN)) {
711 if (status != -ENOSPC)
716 status = ocfs2_journal_dirty(handle, bh);
722 spin_lock(&OCFS2_I(inode)->ip_lock);
723 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
724 spin_unlock(&OCFS2_I(inode)->ip_lock);
726 if (why != RESTART_NONE && clusters_to_add) {
727 if (why == RESTART_META) {
728 mlog(0, "restarting function.\n");
731 BUG_ON(why != RESTART_TRANS);
733 mlog(0, "restarting transaction.\n");
734 /* TODO: This can be more intelligent. */
735 credits = ocfs2_calc_extend_credits(osb->sb,
738 status = ocfs2_extend_trans(handle, credits);
740 /* handle still has to be committed at
746 goto restarted_transaction;
750 mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
751 le32_to_cpu(fe->i_clusters),
752 (unsigned long long)le64_to_cpu(fe->i_size));
753 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
754 OCFS2_I(inode)->ip_clusters, i_size_read(inode));
758 ocfs2_commit_trans(osb, handle);
762 ocfs2_free_alloc_context(data_ac);
766 ocfs2_free_alloc_context(meta_ac);
769 if ((!status) && restart_func) {
782 /* Some parts of this taken from generic_cont_expand, which turned out
783 * to be too fragile to do exactly what we need without us having to
784 * worry about recursive locking in ->prepare_write() and
785 * ->commit_write(). */
786 static int ocfs2_write_zero_page(struct inode *inode,
789 struct address_space *mapping = inode->i_mapping;
793 handle_t *handle = NULL;
796 offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */
797 /* ugh. in prepare/commit_write, if from==to==start of block, we
798 ** skip the prepare. make sure we never send an offset for the start
801 if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) {
804 index = size >> PAGE_CACHE_SHIFT;
806 page = grab_cache_page(mapping, index);
813 ret = ocfs2_prepare_write_nolock(inode, page, offset, offset);
819 if (ocfs2_should_order_data(inode)) {
820 handle = ocfs2_start_walk_page_trans(inode, page, offset,
822 if (IS_ERR(handle)) {
823 ret = PTR_ERR(handle);
829 /* must not update i_size! */
830 ret = block_commit_write(page, offset, offset);
837 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
840 page_cache_release(page);
845 static int ocfs2_zero_extend(struct inode *inode,
850 struct super_block *sb = inode->i_sb;
852 start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
853 while (start_off < zero_to_size) {
854 ret = ocfs2_write_zero_page(inode, start_off);
860 start_off += sb->s_blocksize;
863 * Very large extends have the potential to lock up
864 * the cpu for extended periods of time.
873 int ocfs2_extend_no_holes(struct inode *inode, u64 new_i_size, u64 zero_to)
877 struct ocfs2_inode_info *oi = OCFS2_I(inode);
879 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
880 if (clusters_to_add < oi->ip_clusters)
883 clusters_to_add -= oi->ip_clusters;
885 if (clusters_to_add) {
886 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
895 * Call this even if we don't add any clusters to the tree. We
896 * still need to zero the area between the old i_size and the
899 ret = ocfs2_zero_extend(inode, zero_to);
907 static int ocfs2_extend_file(struct inode *inode,
908 struct buffer_head *di_bh,
915 /* setattr sometimes calls us like this. */
919 if (i_size_read(inode) == new_i_size)
921 BUG_ON(new_i_size < i_size_read(inode));
923 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
924 goto out_update_size;
927 * protect the pages that ocfs2_zero_extend is going to be
928 * pulling into the page cache.. we do this before the
929 * metadata extend so that we don't get into the situation
930 * where we've extended the metadata but can't get the data
933 ret = ocfs2_data_lock(inode, 1);
940 * The alloc sem blocks people in read/write from reading our
941 * allocation until we're done changing it. We depend on
942 * i_mutex to block other extend/truncate calls while we're
945 down_write(&OCFS2_I(inode)->ip_alloc_sem);
946 ret = ocfs2_extend_no_holes(inode, new_i_size, new_i_size);
947 up_write(&OCFS2_I(inode)->ip_alloc_sem);
955 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
960 if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
961 ocfs2_data_unlock(inode, 1);
967 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
969 int status = 0, size_change;
970 struct inode *inode = dentry->d_inode;
971 struct super_block *sb = inode->i_sb;
972 struct ocfs2_super *osb = OCFS2_SB(sb);
973 struct buffer_head *bh = NULL;
974 handle_t *handle = NULL;
976 mlog_entry("(0x%p, '%.*s')\n", dentry,
977 dentry->d_name.len, dentry->d_name.name);
979 if (attr->ia_valid & ATTR_MODE)
980 mlog(0, "mode change: %d\n", attr->ia_mode);
981 if (attr->ia_valid & ATTR_UID)
982 mlog(0, "uid change: %d\n", attr->ia_uid);
983 if (attr->ia_valid & ATTR_GID)
984 mlog(0, "gid change: %d\n", attr->ia_gid);
985 if (attr->ia_valid & ATTR_SIZE)
986 mlog(0, "size change...\n");
987 if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME))
988 mlog(0, "time change...\n");
990 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
991 | ATTR_GID | ATTR_UID | ATTR_MODE)
992 if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) {
993 mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid);
997 status = inode_change_ok(inode, attr);
1001 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1003 status = ocfs2_rw_lock(inode, 1);
1010 status = ocfs2_meta_lock(inode, &bh, 1);
1012 if (status != -ENOENT)
1014 goto bail_unlock_rw;
1017 if (size_change && attr->ia_size != i_size_read(inode)) {
1018 if (attr->ia_size > sb->s_maxbytes) {
1023 if (i_size_read(inode) > attr->ia_size)
1024 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1026 status = ocfs2_extend_file(inode, bh, attr->ia_size);
1028 if (status != -ENOSPC)
1035 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1036 if (IS_ERR(handle)) {
1037 status = PTR_ERR(handle);
1043 * This will intentionally not wind up calling vmtruncate(),
1044 * since all the work for a size change has been done above.
1045 * Otherwise, we could get into problems with truncate as
1046 * ip_alloc_sem is used there to protect against i_size
1049 status = inode_setattr(inode, attr);
1055 status = ocfs2_mark_inode_dirty(handle, inode, bh);
1060 ocfs2_commit_trans(osb, handle);
1062 ocfs2_meta_unlock(inode, 1);
1065 ocfs2_rw_unlock(inode, 1);
1074 int ocfs2_getattr(struct vfsmount *mnt,
1075 struct dentry *dentry,
1078 struct inode *inode = dentry->d_inode;
1079 struct super_block *sb = dentry->d_inode->i_sb;
1080 struct ocfs2_super *osb = sb->s_fs_info;
1085 err = ocfs2_inode_revalidate(dentry);
1092 generic_fillattr(inode, stat);
1094 /* We set the blksize from the cluster size for performance */
1095 stat->blksize = osb->s_clustersize;
1103 int ocfs2_permission(struct inode *inode, int mask, struct nameidata *nd)
1109 ret = ocfs2_meta_lock(inode, NULL, 0);
1116 ret = generic_permission(inode, mask, NULL);
1118 ocfs2_meta_unlock(inode, 0);
1124 static int __ocfs2_write_remove_suid(struct inode *inode,
1125 struct buffer_head *bh)
1129 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1130 struct ocfs2_dinode *di;
1132 mlog_entry("(Inode %llu, mode 0%o)\n",
1133 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_mode);
1135 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1136 if (handle == NULL) {
1142 ret = ocfs2_journal_access(handle, inode, bh,
1143 OCFS2_JOURNAL_ACCESS_WRITE);
1149 inode->i_mode &= ~S_ISUID;
1150 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1151 inode->i_mode &= ~S_ISGID;
1153 di = (struct ocfs2_dinode *) bh->b_data;
1154 di->i_mode = cpu_to_le16(inode->i_mode);
1156 ret = ocfs2_journal_dirty(handle, bh);
1161 ocfs2_commit_trans(osb, handle);
1168 * Will look for holes and unwritten extents in the range starting at
1169 * pos for count bytes (inclusive).
1171 static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1175 unsigned int extent_flags;
1176 u32 cpos, clusters, extent_len, phys_cpos;
1177 struct super_block *sb = inode->i_sb;
1179 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1180 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1183 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1190 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
1195 if (extent_len > clusters)
1196 extent_len = clusters;
1198 clusters -= extent_len;
1205 static int ocfs2_write_remove_suid(struct inode *inode)
1208 struct buffer_head *bh = NULL;
1209 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1211 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
1212 oi->ip_blkno, &bh, OCFS2_BH_CACHED, inode);
1218 ret = __ocfs2_write_remove_suid(inode, bh);
1225 * Allocate enough extents to cover the region starting at byte offset
1226 * start for len bytes. Existing extents are skipped, any extents
1227 * added are marked as "unwritten".
1229 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1233 u32 cpos, phys_cpos, clusters, alloc_size;
1236 * We consider both start and len to be inclusive.
1238 cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1239 clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1243 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1251 * Hole or existing extent len can be arbitrary, so
1252 * cap it to our own allocation request.
1254 if (alloc_size > clusters)
1255 alloc_size = clusters;
1259 * We already have an allocation at this
1260 * region so we can safely skip it.
1265 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1274 clusters -= alloc_size;
1282 static int __ocfs2_remove_inode_range(struct inode *inode,
1283 struct buffer_head *di_bh,
1284 u32 cpos, u32 phys_cpos, u32 len,
1285 struct ocfs2_cached_dealloc_ctxt *dealloc)
1288 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
1289 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1290 struct inode *tl_inode = osb->osb_tl_inode;
1292 struct ocfs2_alloc_context *meta_ac = NULL;
1293 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1295 ret = ocfs2_lock_allocators(inode, di, 0, 1, NULL, &meta_ac);
1301 mutex_lock(&tl_inode->i_mutex);
1303 if (ocfs2_truncate_log_needs_flush(osb)) {
1304 ret = __ocfs2_flush_truncate_log(osb);
1311 handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
1312 if (handle == NULL) {
1318 ret = ocfs2_journal_access(handle, inode, di_bh,
1319 OCFS2_JOURNAL_ACCESS_WRITE);
1325 ret = ocfs2_remove_extent(inode, di_bh, cpos, len, handle, meta_ac,
1332 OCFS2_I(inode)->ip_clusters -= len;
1333 di->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1335 ret = ocfs2_journal_dirty(handle, di_bh);
1341 ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len);
1346 ocfs2_commit_trans(osb, handle);
1348 mutex_unlock(&tl_inode->i_mutex);
1351 ocfs2_free_alloc_context(meta_ac);
1357 * Truncate a byte range, avoiding pages within partial clusters. This
1358 * preserves those pages for the zeroing code to write to.
1360 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1363 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1365 struct address_space *mapping = inode->i_mapping;
1367 start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1368 end = byte_start + byte_len;
1369 end = end & ~(osb->s_clustersize - 1);
1372 unmap_mapping_range(mapping, start, end - start, 0);
1373 truncate_inode_pages_range(mapping, start, end - 1);
1377 static int ocfs2_zero_partial_clusters(struct inode *inode,
1381 u64 tmpend, end = start + len;
1382 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1383 unsigned int csize = osb->s_clustersize;
1387 * The "start" and "end" values are NOT necessarily part of
1388 * the range whose allocation is being deleted. Rather, this
1389 * is what the user passed in with the request. We must zero
1390 * partial clusters here. There's no need to worry about
1391 * physical allocation - the zeroing code knows to skip holes.
1393 mlog(0, "byte start: %llu, end: %llu\n",
1394 (unsigned long long)start, (unsigned long long)end);
1397 * If both edges are on a cluster boundary then there's no
1398 * zeroing required as the region is part of the allocation to
1401 if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1404 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1405 if (handle == NULL) {
1412 * We want to get the byte offset of the end of the 1st cluster.
1414 tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1418 mlog(0, "1st range: start: %llu, tmpend: %llu\n",
1419 (unsigned long long)start, (unsigned long long)tmpend);
1421 ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1427 * This may make start and end equal, but the zeroing
1428 * code will skip any work in that case so there's no
1429 * need to catch it up here.
1431 start = end & ~(osb->s_clustersize - 1);
1433 mlog(0, "2nd range: start: %llu, end: %llu\n",
1434 (unsigned long long)start, (unsigned long long)end);
1436 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1441 ocfs2_commit_trans(osb, handle);
1446 static int ocfs2_remove_inode_range(struct inode *inode,
1447 struct buffer_head *di_bh, u64 byte_start,
1451 u32 trunc_start, trunc_len, cpos, phys_cpos, alloc_size;
1452 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1453 struct ocfs2_cached_dealloc_ctxt dealloc;
1455 ocfs2_init_dealloc_ctxt(&dealloc);
1460 trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1461 trunc_len = (byte_start + byte_len) >> osb->s_clustersize_bits;
1462 if (trunc_len >= trunc_start)
1463 trunc_len -= trunc_start;
1467 mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u\n",
1468 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1469 (unsigned long long)byte_start,
1470 (unsigned long long)byte_len, trunc_start, trunc_len);
1472 ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1480 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1487 if (alloc_size > trunc_len)
1488 alloc_size = trunc_len;
1490 /* Only do work for non-holes */
1491 if (phys_cpos != 0) {
1492 ret = __ocfs2_remove_inode_range(inode, di_bh, cpos,
1493 phys_cpos, alloc_size,
1502 trunc_len -= alloc_size;
1505 ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1508 ocfs2_schedule_truncate_log_flush(osb, 1);
1509 ocfs2_run_deallocs(osb, &dealloc);
1515 * Parts of this function taken from xfs_change_file_space()
1517 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1518 loff_t f_pos, unsigned int cmd,
1519 struct ocfs2_space_resv *sr,
1525 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1526 struct buffer_head *di_bh = NULL;
1528 unsigned long long max_off = inode->i_sb->s_maxbytes;
1530 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1533 mutex_lock(&inode->i_mutex);
1536 * This prevents concurrent writes on other nodes
1538 ret = ocfs2_rw_lock(inode, 1);
1544 ret = ocfs2_meta_lock(inode, &di_bh, 1);
1550 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1552 goto out_meta_unlock;
1555 switch (sr->l_whence) {
1556 case 0: /*SEEK_SET*/
1558 case 1: /*SEEK_CUR*/
1559 sr->l_start += f_pos;
1561 case 2: /*SEEK_END*/
1562 sr->l_start += i_size_read(inode);
1566 goto out_meta_unlock;
1570 llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1573 || sr->l_start > max_off
1574 || (sr->l_start + llen) < 0
1575 || (sr->l_start + llen) > max_off) {
1577 goto out_meta_unlock;
1579 size = sr->l_start + sr->l_len;
1581 if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) {
1582 if (sr->l_len <= 0) {
1584 goto out_meta_unlock;
1588 if (file && should_remove_suid(file->f_path.dentry)) {
1589 ret = __ocfs2_write_remove_suid(inode, di_bh);
1592 goto out_meta_unlock;
1596 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1598 case OCFS2_IOC_RESVSP:
1599 case OCFS2_IOC_RESVSP64:
1601 * This takes unsigned offsets, but the signed ones we
1602 * pass have been checked against overflow above.
1604 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1607 case OCFS2_IOC_UNRESVSP:
1608 case OCFS2_IOC_UNRESVSP64:
1609 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1615 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1618 goto out_meta_unlock;
1622 * We update c/mtime for these changes
1624 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1625 if (IS_ERR(handle)) {
1626 ret = PTR_ERR(handle);
1628 goto out_meta_unlock;
1631 if (change_size && i_size_read(inode) < size)
1632 i_size_write(inode, size);
1634 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1635 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1639 ocfs2_commit_trans(osb, handle);
1643 ocfs2_meta_unlock(inode, 1);
1645 ocfs2_rw_unlock(inode, 1);
1647 mutex_unlock(&inode->i_mutex);
1652 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1653 struct ocfs2_space_resv *sr)
1655 struct inode *inode = file->f_path.dentry->d_inode;
1656 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);;
1658 if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1659 !ocfs2_writes_unwritten_extents(osb))
1661 else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1662 !ocfs2_sparse_alloc(osb))
1665 if (!S_ISREG(inode->i_mode))
1668 if (!(file->f_mode & FMODE_WRITE))
1671 return __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1674 static long ocfs2_fallocate(struct inode *inode, int mode, loff_t offset,
1677 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1678 struct ocfs2_space_resv sr;
1679 int change_size = 1;
1681 if (!ocfs2_writes_unwritten_extents(osb))
1684 if (S_ISDIR(inode->i_mode))
1687 if (mode & FALLOC_FL_KEEP_SIZE)
1691 sr.l_start = (s64)offset;
1692 sr.l_len = (s64)len;
1694 return __ocfs2_change_file_space(NULL, inode, offset,
1695 OCFS2_IOC_RESVSP64, &sr, change_size);
1698 static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
1704 int ret = 0, meta_level = 0;
1705 struct inode *inode = dentry->d_inode;
1706 loff_t saved_pos, end;
1709 * We start with a read level meta lock and only jump to an ex
1710 * if we need to make modifications here.
1713 ret = ocfs2_meta_lock(inode, NULL, meta_level);
1720 /* Clear suid / sgid if necessary. We do this here
1721 * instead of later in the write path because
1722 * remove_suid() calls ->setattr without any hint that
1723 * we may have already done our cluster locking. Since
1724 * ocfs2_setattr() *must* take cluster locks to
1725 * proceeed, this will lead us to recursively lock the
1726 * inode. There's also the dinode i_size state which
1727 * can be lost via setattr during extending writes (we
1728 * set inode->i_size at the end of a write. */
1729 if (should_remove_suid(dentry)) {
1730 if (meta_level == 0) {
1731 ocfs2_meta_unlock(inode, meta_level);
1736 ret = ocfs2_write_remove_suid(inode);
1743 /* work on a copy of ppos until we're sure that we won't have
1744 * to recalculate it due to relocking. */
1746 saved_pos = i_size_read(inode);
1747 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos);
1752 end = saved_pos + count;
1755 * Skip the O_DIRECT checks if we don't need
1758 if (!direct_io || !(*direct_io))
1762 * Allowing concurrent direct writes means
1763 * i_size changes wouldn't be synchronized, so
1764 * one node could wind up truncating another
1767 if (end > i_size_read(inode)) {
1773 * We don't fill holes during direct io, so
1774 * check for them here. If any are found, the
1775 * caller will have to retake some cluster
1776 * locks and initiate the io as buffered.
1778 ret = ocfs2_check_range_for_holes(inode, saved_pos, count);
1791 ocfs2_meta_unlock(inode, meta_level);
1798 ocfs2_set_next_iovec(const struct iovec **iovp, size_t *basep, size_t bytes)
1800 const struct iovec *iov = *iovp;
1801 size_t base = *basep;
1804 int copy = min(bytes, iov->iov_len - base);
1808 if (iov->iov_len == base) {
1817 static struct page * ocfs2_get_write_source(char **ret_src_buf,
1818 const struct iovec *cur_iov,
1822 char *buf = cur_iov->iov_base + iov_offset;
1823 struct page *src_page = NULL;
1826 off = (unsigned long)(buf) & ~PAGE_CACHE_MASK;
1828 if (!segment_eq(get_fs(), KERNEL_DS)) {
1830 * Pull in the user page. We want to do this outside
1831 * of the meta data locks in order to preserve locking
1832 * order in case of page fault.
1834 ret = get_user_pages(current, current->mm,
1835 (unsigned long)buf & PAGE_CACHE_MASK, 1,
1836 0, 0, &src_page, NULL);
1838 *ret_src_buf = kmap(src_page) + off;
1840 src_page = ERR_PTR(-EFAULT);
1848 static void ocfs2_put_write_source(struct page *page)
1852 page_cache_release(page);
1856 static ssize_t ocfs2_file_buffered_write(struct file *file, loff_t *ppos,
1857 const struct iovec *iov,
1858 unsigned long nr_segs,
1860 ssize_t o_direct_written)
1863 ssize_t copied, total = 0;
1864 size_t iov_offset = 0, bytes;
1866 const struct iovec *cur_iov = iov;
1867 struct page *user_page, *page;
1868 char * uninitialized_var(buf);
1873 * handle partial DIO write. Adjust cur_iov if needed.
1875 ocfs2_set_next_iovec(&cur_iov, &iov_offset, o_direct_written);
1880 user_page = ocfs2_get_write_source(&buf, cur_iov, iov_offset);
1881 if (IS_ERR(user_page)) {
1882 ret = PTR_ERR(user_page);
1886 /* Stay within our page boundaries */
1887 bytes = min((PAGE_CACHE_SIZE - ((unsigned long)pos & ~PAGE_CACHE_MASK)),
1888 (PAGE_CACHE_SIZE - ((unsigned long)buf & ~PAGE_CACHE_MASK)));
1889 /* Stay within the vector boundary */
1890 bytes = min_t(size_t, bytes, cur_iov->iov_len - iov_offset);
1891 /* Stay within count */
1892 bytes = min(bytes, count);
1895 ret = ocfs2_write_begin(file, file->f_mapping, pos, bytes, 0,
1902 dst = kmap_atomic(page, KM_USER0);
1903 memcpy(dst + (pos & (loff_t)(PAGE_CACHE_SIZE - 1)), buf, bytes);
1904 kunmap_atomic(dst, KM_USER0);
1905 flush_dcache_page(page);
1906 ocfs2_put_write_source(user_page);
1908 copied = ocfs2_write_end(file, file->f_mapping, pos, bytes,
1909 bytes, page, fsdata);
1917 *ppos = pos + copied;
1920 ocfs2_set_next_iovec(&cur_iov, &iov_offset, copied);
1924 return total ? total : ret;
1927 static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
1928 const struct iovec *iov,
1929 unsigned long nr_segs,
1932 int ret, direct_io, appending, rw_level, have_alloc_sem = 0;
1933 int can_do_direct, sync = 0;
1934 ssize_t written = 0;
1935 size_t ocount; /* original count */
1936 size_t count; /* after file limit checks */
1937 loff_t *ppos = &iocb->ki_pos;
1938 struct file *file = iocb->ki_filp;
1939 struct inode *inode = file->f_path.dentry->d_inode;
1941 mlog_entry("(0x%p, %u, '%.*s')\n", file,
1942 (unsigned int)nr_segs,
1943 file->f_path.dentry->d_name.len,
1944 file->f_path.dentry->d_name.name);
1946 if (iocb->ki_left == 0)
1949 ret = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
1955 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1957 appending = file->f_flags & O_APPEND ? 1 : 0;
1958 direct_io = file->f_flags & O_DIRECT ? 1 : 0;
1960 mutex_lock(&inode->i_mutex);
1963 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
1965 down_read(&inode->i_alloc_sem);
1969 /* concurrent O_DIRECT writes are allowed */
1970 rw_level = !direct_io;
1971 ret = ocfs2_rw_lock(inode, rw_level);
1977 can_do_direct = direct_io;
1978 ret = ocfs2_prepare_inode_for_write(file->f_path.dentry, ppos,
1979 iocb->ki_left, appending,
1987 * We can't complete the direct I/O as requested, fall back to
1990 if (direct_io && !can_do_direct) {
1991 ocfs2_rw_unlock(inode, rw_level);
1992 up_read(&inode->i_alloc_sem);
2002 if (!sync && ((file->f_flags & O_SYNC) || IS_SYNC(inode)))
2006 * XXX: Is it ok to execute these checks a second time?
2008 ret = generic_write_checks(file, ppos, &count, S_ISBLK(inode->i_mode));
2013 * Set pos so that sync_page_range_nolock() below understands
2014 * where to start from. We might've moved it around via the
2015 * calls above. The range we want to actually sync starts from
2021 /* communicate with ocfs2_dio_end_io */
2022 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2025 written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
2026 ppos, count, ocount);
2032 written = ocfs2_file_buffered_write(file, ppos, iov, nr_segs,
2036 if (ret != -EFAULT || ret != -ENOSPC)
2043 /* buffered aio wouldn't have proper lock coverage today */
2044 BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
2047 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2048 * function pointer which is called when o_direct io completes so that
2049 * it can unlock our rw lock. (it's the clustered equivalent of
2050 * i_alloc_sem; protects truncate from racing with pending ios).
2051 * Unfortunately there are error cases which call end_io and others
2052 * that don't. so we don't have to unlock the rw_lock if either an
2053 * async dio is going to do it in the future or an end_io after an
2054 * error has already done it.
2056 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2063 ocfs2_rw_unlock(inode, rw_level);
2067 up_read(&inode->i_alloc_sem);
2069 if (written > 0 && sync) {
2072 err = sync_page_range_nolock(inode, file->f_mapping, pos, count);
2077 mutex_unlock(&inode->i_mutex);
2080 return written ? written : ret;
2083 static int ocfs2_splice_write_actor(struct pipe_inode_info *pipe,
2084 struct pipe_buffer *buf,
2085 struct splice_desc *sd)
2089 struct file *file = sd->u.file;
2090 unsigned int offset;
2091 struct page *page = NULL;
2095 ret = buf->ops->confirm(pipe, buf);
2099 offset = sd->pos & ~PAGE_CACHE_MASK;
2101 if (count + offset > PAGE_CACHE_SIZE)
2102 count = PAGE_CACHE_SIZE - offset;
2104 ret = ocfs2_write_begin(file, file->f_mapping, sd->pos, count, 0,
2111 src = buf->ops->map(pipe, buf, 1);
2112 dst = kmap_atomic(page, KM_USER1);
2113 memcpy(dst + offset, src + buf->offset, count);
2114 kunmap_atomic(dst, KM_USER1);
2115 buf->ops->unmap(pipe, buf, src);
2117 copied = ocfs2_write_end(file, file->f_mapping, sd->pos, count, count,
2126 return copied ? copied : ret;
2129 static ssize_t __ocfs2_file_splice_write(struct pipe_inode_info *pipe,
2136 struct address_space *mapping = out->f_mapping;
2137 struct inode *inode = mapping->host;
2138 struct splice_desc sd = {
2145 ret = __splice_from_pipe(pipe, &sd, ocfs2_splice_write_actor);
2149 if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) {
2150 err = generic_osync_inode(inode, mapping,
2151 OSYNC_METADATA|OSYNC_DATA);
2160 static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
2167 struct inode *inode = out->f_path.dentry->d_inode;
2169 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe,
2171 out->f_path.dentry->d_name.len,
2172 out->f_path.dentry->d_name.name);
2174 inode_double_lock(inode, pipe->inode);
2176 ret = ocfs2_rw_lock(inode, 1);
2182 ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, ppos, len, 0,
2189 /* ok, we're done with i_size and alloc work */
2190 ret = __ocfs2_file_splice_write(pipe, out, ppos, len, flags);
2193 ocfs2_rw_unlock(inode, 1);
2195 inode_double_unlock(inode, pipe->inode);
2201 static ssize_t ocfs2_file_splice_read(struct file *in,
2203 struct pipe_inode_info *pipe,
2208 struct inode *inode = in->f_path.dentry->d_inode;
2210 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe,
2212 in->f_path.dentry->d_name.len,
2213 in->f_path.dentry->d_name.name);
2216 * See the comment in ocfs2_file_aio_read()
2218 ret = ocfs2_meta_lock(inode, NULL, 0);
2223 ocfs2_meta_unlock(inode, 0);
2225 ret = generic_file_splice_read(in, ppos, pipe, len, flags);
2232 static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
2233 const struct iovec *iov,
2234 unsigned long nr_segs,
2237 int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
2238 struct file *filp = iocb->ki_filp;
2239 struct inode *inode = filp->f_path.dentry->d_inode;
2241 mlog_entry("(0x%p, %u, '%.*s')\n", filp,
2242 (unsigned int)nr_segs,
2243 filp->f_path.dentry->d_name.len,
2244 filp->f_path.dentry->d_name.name);
2253 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
2254 * need locks to protect pending reads from racing with truncate.
2256 if (filp->f_flags & O_DIRECT) {
2257 down_read(&inode->i_alloc_sem);
2260 ret = ocfs2_rw_lock(inode, 0);
2266 /* communicate with ocfs2_dio_end_io */
2267 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2271 * We're fine letting folks race truncates and extending
2272 * writes with read across the cluster, just like they can
2273 * locally. Hence no rw_lock during read.
2275 * Take and drop the meta data lock to update inode fields
2276 * like i_size. This allows the checks down below
2277 * generic_file_aio_read() a chance of actually working.
2279 ret = ocfs2_meta_lock_atime(inode, filp->f_vfsmnt, &lock_level);
2284 ocfs2_meta_unlock(inode, lock_level);
2286 ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
2288 mlog(ML_ERROR, "generic_file_aio_read returned -EINVAL\n");
2290 /* buffered aio wouldn't have proper lock coverage today */
2291 BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
2293 /* see ocfs2_file_aio_write */
2294 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2301 up_read(&inode->i_alloc_sem);
2303 ocfs2_rw_unlock(inode, rw_level);
2309 const struct inode_operations ocfs2_file_iops = {
2310 .setattr = ocfs2_setattr,
2311 .getattr = ocfs2_getattr,
2312 .permission = ocfs2_permission,
2313 .fallocate = ocfs2_fallocate,
2316 const struct inode_operations ocfs2_special_file_iops = {
2317 .setattr = ocfs2_setattr,
2318 .getattr = ocfs2_getattr,
2319 .permission = ocfs2_permission,
2322 const struct file_operations ocfs2_fops = {
2323 .read = do_sync_read,
2324 .write = do_sync_write,
2326 .fsync = ocfs2_sync_file,
2327 .release = ocfs2_file_release,
2328 .open = ocfs2_file_open,
2329 .aio_read = ocfs2_file_aio_read,
2330 .aio_write = ocfs2_file_aio_write,
2331 .ioctl = ocfs2_ioctl,
2332 #ifdef CONFIG_COMPAT
2333 .compat_ioctl = ocfs2_compat_ioctl,
2335 .splice_read = ocfs2_file_splice_read,
2336 .splice_write = ocfs2_file_splice_write,
2339 const struct file_operations ocfs2_dops = {
2340 .read = generic_read_dir,
2341 .readdir = ocfs2_readdir,
2342 .fsync = ocfs2_sync_file,
2343 .ioctl = ocfs2_ioctl,
2344 #ifdef CONFIG_COMPAT
2345 .compat_ioctl = ocfs2_compat_ioctl,