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);
193 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
194 if (handle == NULL) {
200 inode->i_atime = CURRENT_TIME;
201 ret = ocfs2_mark_inode_dirty(handle, inode, bh);
205 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
211 static int ocfs2_set_inode_size(handle_t *handle,
213 struct buffer_head *fe_bh,
219 i_size_write(inode, new_i_size);
220 inode->i_blocks = ocfs2_inode_sector_count(inode);
221 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
223 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
234 static int ocfs2_simple_size_update(struct inode *inode,
235 struct buffer_head *di_bh,
239 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
240 handle_t *handle = NULL;
242 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
243 if (handle == NULL) {
249 ret = ocfs2_set_inode_size(handle, inode, di_bh,
254 ocfs2_commit_trans(osb, handle);
259 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
261 struct buffer_head *fe_bh,
266 struct ocfs2_dinode *di;
271 /* TODO: This needs to actually orphan the inode in this
274 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
275 if (IS_ERR(handle)) {
276 status = PTR_ERR(handle);
281 status = ocfs2_journal_access(handle, inode, fe_bh,
282 OCFS2_JOURNAL_ACCESS_WRITE);
289 * Do this before setting i_size.
291 cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
292 status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
299 i_size_write(inode, new_i_size);
300 inode->i_blocks = ocfs2_align_bytes_to_sectors(new_i_size);
301 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
303 di = (struct ocfs2_dinode *) fe_bh->b_data;
304 di->i_size = cpu_to_le64(new_i_size);
305 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
306 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
308 status = ocfs2_journal_dirty(handle, fe_bh);
313 ocfs2_commit_trans(osb, handle);
320 static int ocfs2_truncate_file(struct inode *inode,
321 struct buffer_head *di_bh,
325 struct ocfs2_dinode *fe = NULL;
326 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
327 struct ocfs2_truncate_context *tc = NULL;
329 mlog_entry("(inode = %llu, new_i_size = %llu\n",
330 (unsigned long long)OCFS2_I(inode)->ip_blkno,
331 (unsigned long long)new_i_size);
333 fe = (struct ocfs2_dinode *) di_bh->b_data;
334 if (!OCFS2_IS_VALID_DINODE(fe)) {
335 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
340 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
341 "Inode %llu, inode i_size = %lld != di "
342 "i_size = %llu, i_flags = 0x%x\n",
343 (unsigned long long)OCFS2_I(inode)->ip_blkno,
345 (unsigned long long)le64_to_cpu(fe->i_size),
346 le32_to_cpu(fe->i_flags));
348 if (new_i_size > le64_to_cpu(fe->i_size)) {
349 mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
350 (unsigned long long)le64_to_cpu(fe->i_size),
351 (unsigned long long)new_i_size);
357 mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
358 (unsigned long long)le64_to_cpu(fe->i_blkno),
359 (unsigned long long)le64_to_cpu(fe->i_size),
360 (unsigned long long)new_i_size);
362 /* lets handle the simple truncate cases before doing any more
363 * cluster locking. */
364 if (new_i_size == le64_to_cpu(fe->i_size))
367 down_write(&OCFS2_I(inode)->ip_alloc_sem);
369 /* This forces other nodes to sync and drop their pages. Do
370 * this even if we have a truncate without allocation change -
371 * ocfs2 cluster sizes can be much greater than page size, so
372 * we have to truncate them anyway. */
373 status = ocfs2_data_lock(inode, 1);
375 up_write(&OCFS2_I(inode)->ip_alloc_sem);
381 unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
382 truncate_inode_pages(inode->i_mapping, new_i_size);
384 /* alright, we're going to need to do a full blown alloc size
385 * change. Orphan the inode so that recovery can complete the
386 * truncate if necessary. This does the task of marking
388 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
391 goto bail_unlock_data;
394 status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc);
397 goto bail_unlock_data;
400 status = ocfs2_commit_truncate(osb, inode, di_bh, tc);
403 goto bail_unlock_data;
406 /* TODO: orphan dir cleanup here. */
408 ocfs2_data_unlock(inode, 1);
410 up_write(&OCFS2_I(inode)->ip_alloc_sem);
419 * extend allocation only here.
420 * we'll update all the disk stuff, and oip->alloc_size
422 * expect stuff to be locked, a transaction started and enough data /
423 * metadata reservations in the contexts.
425 * Will return -EAGAIN, and a reason if a restart is needed.
426 * If passed in, *reason will always be set, even in error.
428 int ocfs2_do_extend_allocation(struct ocfs2_super *osb,
433 struct buffer_head *fe_bh,
435 struct ocfs2_alloc_context *data_ac,
436 struct ocfs2_alloc_context *meta_ac,
437 enum ocfs2_alloc_restarted *reason_ret)
441 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
442 enum ocfs2_alloc_restarted reason = RESTART_NONE;
443 u32 bit_off, num_bits;
447 BUG_ON(!clusters_to_add);
450 flags = OCFS2_EXT_UNWRITTEN;
452 free_extents = ocfs2_num_free_extents(osb, inode, fe);
453 if (free_extents < 0) {
454 status = free_extents;
459 /* there are two cases which could cause us to EAGAIN in the
460 * we-need-more-metadata case:
461 * 1) we haven't reserved *any*
462 * 2) we are so fragmented, we've needed to add metadata too
464 if (!free_extents && !meta_ac) {
465 mlog(0, "we haven't reserved any metadata!\n");
467 reason = RESTART_META;
469 } else if ((!free_extents)
470 && (ocfs2_alloc_context_bits_left(meta_ac)
471 < ocfs2_extend_meta_needed(fe))) {
472 mlog(0, "filesystem is really fragmented...\n");
474 reason = RESTART_META;
478 status = ocfs2_claim_clusters(osb, handle, data_ac, 1,
479 &bit_off, &num_bits);
481 if (status != -ENOSPC)
486 BUG_ON(num_bits > clusters_to_add);
488 /* reserve our write early -- insert_extent may update the inode */
489 status = ocfs2_journal_access(handle, inode, fe_bh,
490 OCFS2_JOURNAL_ACCESS_WRITE);
496 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
497 mlog(0, "Allocating %u clusters at block %u for inode %llu\n",
498 num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
499 status = ocfs2_insert_extent(osb, handle, inode, fe_bh,
500 *logical_offset, block, num_bits,
507 status = ocfs2_journal_dirty(handle, fe_bh);
513 clusters_to_add -= num_bits;
514 *logical_offset += num_bits;
516 if (clusters_to_add) {
517 mlog(0, "need to alloc once more, clusters = %u, wanted = "
518 "%u\n", fe->i_clusters, clusters_to_add);
520 reason = RESTART_TRANS;
526 *reason_ret = reason;
531 * For a given allocation, determine which allocators will need to be
532 * accessed, and lock them, reserving the appropriate number of bits.
534 * Sparse file systems call this from ocfs2_write_begin_nolock()
535 * and ocfs2_allocate_unwritten_extents().
537 * File systems which don't support holes call this from
538 * ocfs2_extend_allocation().
540 int ocfs2_lock_allocators(struct inode *inode, struct ocfs2_dinode *di,
541 u32 clusters_to_add, u32 extents_to_split,
542 struct ocfs2_alloc_context **data_ac,
543 struct ocfs2_alloc_context **meta_ac)
545 int ret = 0, num_free_extents;
546 unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
547 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
553 BUG_ON(clusters_to_add != 0 && data_ac == NULL);
555 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, "
556 "clusters_to_add = %u, extents_to_split = %u\n",
557 (unsigned long long)OCFS2_I(inode)->ip_blkno, i_size_read(inode),
558 le32_to_cpu(di->i_clusters), clusters_to_add, extents_to_split);
560 num_free_extents = ocfs2_num_free_extents(osb, inode, di);
561 if (num_free_extents < 0) {
562 ret = num_free_extents;
568 * Sparse allocation file systems need to be more conservative
569 * with reserving room for expansion - the actual allocation
570 * happens while we've got a journal handle open so re-taking
571 * a cluster lock (because we ran out of room for another
572 * extent) will violate ordering rules.
574 * Most of the time we'll only be seeing this 1 cluster at a time
577 * Always lock for any unwritten extents - we might want to
578 * add blocks during a split.
580 if (!num_free_extents ||
581 (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
582 ret = ocfs2_reserve_new_metadata(osb, di, meta_ac);
590 if (clusters_to_add == 0)
593 ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
603 ocfs2_free_alloc_context(*meta_ac);
608 * We cannot have an error and a non null *data_ac.
615 static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
616 u32 clusters_to_add, int mark_unwritten)
619 int restart_func = 0;
622 struct buffer_head *bh = NULL;
623 struct ocfs2_dinode *fe = NULL;
624 handle_t *handle = NULL;
625 struct ocfs2_alloc_context *data_ac = NULL;
626 struct ocfs2_alloc_context *meta_ac = NULL;
627 enum ocfs2_alloc_restarted why;
628 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
630 mlog_entry("(clusters_to_add = %u)\n", clusters_to_add);
633 * This function only exists for file systems which don't
636 BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
638 status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &bh,
639 OCFS2_BH_CACHED, inode);
645 fe = (struct ocfs2_dinode *) bh->b_data;
646 if (!OCFS2_IS_VALID_DINODE(fe)) {
647 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
653 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
655 status = ocfs2_lock_allocators(inode, fe, clusters_to_add, 0, &data_ac,
662 credits = ocfs2_calc_extend_credits(osb->sb, fe, clusters_to_add);
663 handle = ocfs2_start_trans(osb, credits);
664 if (IS_ERR(handle)) {
665 status = PTR_ERR(handle);
671 restarted_transaction:
672 /* reserve a write to the file entry early on - that we if we
673 * run out of credits in the allocation path, we can still
675 status = ocfs2_journal_access(handle, inode, bh,
676 OCFS2_JOURNAL_ACCESS_WRITE);
682 prev_clusters = OCFS2_I(inode)->ip_clusters;
684 status = ocfs2_do_extend_allocation(osb,
694 if ((status < 0) && (status != -EAGAIN)) {
695 if (status != -ENOSPC)
700 status = ocfs2_journal_dirty(handle, bh);
706 spin_lock(&OCFS2_I(inode)->ip_lock);
707 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
708 spin_unlock(&OCFS2_I(inode)->ip_lock);
710 if (why != RESTART_NONE && clusters_to_add) {
711 if (why == RESTART_META) {
712 mlog(0, "restarting function.\n");
715 BUG_ON(why != RESTART_TRANS);
717 mlog(0, "restarting transaction.\n");
718 /* TODO: This can be more intelligent. */
719 credits = ocfs2_calc_extend_credits(osb->sb,
722 status = ocfs2_extend_trans(handle, credits);
724 /* handle still has to be committed at
730 goto restarted_transaction;
734 mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
735 le32_to_cpu(fe->i_clusters),
736 (unsigned long long)le64_to_cpu(fe->i_size));
737 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
738 OCFS2_I(inode)->ip_clusters, i_size_read(inode));
742 ocfs2_commit_trans(osb, handle);
746 ocfs2_free_alloc_context(data_ac);
750 ocfs2_free_alloc_context(meta_ac);
753 if ((!status) && restart_func) {
766 static int ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
767 u32 clusters_to_add, int mark_unwritten)
772 * The alloc sem blocks peope in read/write from reading our
773 * allocation until we're done changing it. We depend on
774 * i_mutex to block other extend/truncate calls while we're
777 down_write(&OCFS2_I(inode)->ip_alloc_sem);
778 ret = __ocfs2_extend_allocation(inode, logical_start, clusters_to_add,
780 up_write(&OCFS2_I(inode)->ip_alloc_sem);
785 /* Some parts of this taken from generic_cont_expand, which turned out
786 * to be too fragile to do exactly what we need without us having to
787 * worry about recursive locking in ->prepare_write() and
788 * ->commit_write(). */
789 static int ocfs2_write_zero_page(struct inode *inode,
792 struct address_space *mapping = inode->i_mapping;
796 handle_t *handle = NULL;
799 offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */
800 /* ugh. in prepare/commit_write, if from==to==start of block, we
801 ** skip the prepare. make sure we never send an offset for the start
804 if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) {
807 index = size >> PAGE_CACHE_SHIFT;
809 page = grab_cache_page(mapping, index);
816 ret = ocfs2_prepare_write_nolock(inode, page, offset, offset);
822 if (ocfs2_should_order_data(inode)) {
823 handle = ocfs2_start_walk_page_trans(inode, page, offset,
825 if (IS_ERR(handle)) {
826 ret = PTR_ERR(handle);
832 /* must not update i_size! */
833 ret = block_commit_write(page, offset, offset);
840 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
843 page_cache_release(page);
848 static int ocfs2_zero_extend(struct inode *inode,
853 struct super_block *sb = inode->i_sb;
855 start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
856 while (start_off < zero_to_size) {
857 ret = ocfs2_write_zero_page(inode, start_off);
863 start_off += sb->s_blocksize;
866 * Very large extends have the potential to lock up
867 * the cpu for extended periods of time.
877 * A tail_to_skip value > 0 indicates that we're being called from
878 * ocfs2_file_aio_write(). This has the following implications:
880 * - we don't want to update i_size
881 * - di_bh will be NULL, which is fine because it's only used in the
882 * case where we want to update i_size.
883 * - ocfs2_zero_extend() will then only be filling the hole created
884 * between i_size and the start of the write.
886 static int ocfs2_extend_file(struct inode *inode,
887 struct buffer_head *di_bh,
892 u32 clusters_to_add = 0;
894 BUG_ON(!tail_to_skip && !di_bh);
896 /* setattr sometimes calls us like this. */
900 if (i_size_read(inode) == new_i_size)
902 BUG_ON(new_i_size < i_size_read(inode));
904 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) {
905 BUG_ON(tail_to_skip != 0);
906 goto out_update_size;
909 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size) -
910 OCFS2_I(inode)->ip_clusters;
913 * protect the pages that ocfs2_zero_extend is going to be
914 * pulling into the page cache.. we do this before the
915 * metadata extend so that we don't get into the situation
916 * where we've extended the metadata but can't get the data
919 ret = ocfs2_data_lock(inode, 1);
925 if (clusters_to_add) {
926 ret = ocfs2_extend_allocation(inode,
927 OCFS2_I(inode)->ip_clusters,
936 * Call this even if we don't add any clusters to the tree. We
937 * still need to zero the area between the old i_size and the
940 ret = ocfs2_zero_extend(inode, (u64)new_i_size - tail_to_skip);
948 /* We're being called from ocfs2_setattr() which wants
949 * us to update i_size */
950 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
956 if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
957 ocfs2_data_unlock(inode, 1);
963 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
965 int status = 0, size_change;
966 struct inode *inode = dentry->d_inode;
967 struct super_block *sb = inode->i_sb;
968 struct ocfs2_super *osb = OCFS2_SB(sb);
969 struct buffer_head *bh = NULL;
970 handle_t *handle = NULL;
972 mlog_entry("(0x%p, '%.*s')\n", dentry,
973 dentry->d_name.len, dentry->d_name.name);
975 if (attr->ia_valid & ATTR_MODE)
976 mlog(0, "mode change: %d\n", attr->ia_mode);
977 if (attr->ia_valid & ATTR_UID)
978 mlog(0, "uid change: %d\n", attr->ia_uid);
979 if (attr->ia_valid & ATTR_GID)
980 mlog(0, "gid change: %d\n", attr->ia_gid);
981 if (attr->ia_valid & ATTR_SIZE)
982 mlog(0, "size change...\n");
983 if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME))
984 mlog(0, "time change...\n");
986 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
987 | ATTR_GID | ATTR_UID | ATTR_MODE)
988 if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) {
989 mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid);
993 status = inode_change_ok(inode, attr);
997 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
999 status = ocfs2_rw_lock(inode, 1);
1006 status = ocfs2_meta_lock(inode, &bh, 1);
1008 if (status != -ENOENT)
1010 goto bail_unlock_rw;
1013 if (size_change && attr->ia_size != i_size_read(inode)) {
1014 if (i_size_read(inode) > attr->ia_size)
1015 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1017 status = ocfs2_extend_file(inode, bh, attr->ia_size, 0);
1019 if (status != -ENOSPC)
1026 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1027 if (IS_ERR(handle)) {
1028 status = PTR_ERR(handle);
1034 * This will intentionally not wind up calling vmtruncate(),
1035 * since all the work for a size change has been done above.
1036 * Otherwise, we could get into problems with truncate as
1037 * ip_alloc_sem is used there to protect against i_size
1040 status = inode_setattr(inode, attr);
1046 status = ocfs2_mark_inode_dirty(handle, inode, bh);
1051 ocfs2_commit_trans(osb, handle);
1053 ocfs2_meta_unlock(inode, 1);
1056 ocfs2_rw_unlock(inode, 1);
1065 int ocfs2_getattr(struct vfsmount *mnt,
1066 struct dentry *dentry,
1069 struct inode *inode = dentry->d_inode;
1070 struct super_block *sb = dentry->d_inode->i_sb;
1071 struct ocfs2_super *osb = sb->s_fs_info;
1076 err = ocfs2_inode_revalidate(dentry);
1083 generic_fillattr(inode, stat);
1085 /* We set the blksize from the cluster size for performance */
1086 stat->blksize = osb->s_clustersize;
1094 int ocfs2_permission(struct inode *inode, int mask, struct nameidata *nd)
1100 ret = ocfs2_meta_lock(inode, NULL, 0);
1107 ret = generic_permission(inode, mask, NULL);
1109 ocfs2_meta_unlock(inode, 0);
1115 static int __ocfs2_write_remove_suid(struct inode *inode,
1116 struct buffer_head *bh)
1120 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1121 struct ocfs2_dinode *di;
1123 mlog_entry("(Inode %llu, mode 0%o)\n",
1124 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_mode);
1126 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1127 if (handle == NULL) {
1133 ret = ocfs2_journal_access(handle, inode, bh,
1134 OCFS2_JOURNAL_ACCESS_WRITE);
1140 inode->i_mode &= ~S_ISUID;
1141 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1142 inode->i_mode &= ~S_ISGID;
1144 di = (struct ocfs2_dinode *) bh->b_data;
1145 di->i_mode = cpu_to_le16(inode->i_mode);
1147 ret = ocfs2_journal_dirty(handle, bh);
1152 ocfs2_commit_trans(osb, handle);
1159 * Will look for holes and unwritten extents in the range starting at
1160 * pos for count bytes (inclusive).
1162 static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1166 unsigned int extent_flags;
1167 u32 cpos, clusters, extent_len, phys_cpos;
1168 struct super_block *sb = inode->i_sb;
1170 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1171 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1174 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1181 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
1186 if (extent_len > clusters)
1187 extent_len = clusters;
1189 clusters -= extent_len;
1196 static int ocfs2_write_remove_suid(struct inode *inode)
1199 struct buffer_head *bh = NULL;
1200 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1202 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
1203 oi->ip_blkno, &bh, OCFS2_BH_CACHED, inode);
1209 ret = __ocfs2_write_remove_suid(inode, bh);
1216 * Allocate enough extents to cover the region starting at byte offset
1217 * start for len bytes. Existing extents are skipped, any extents
1218 * added are marked as "unwritten".
1220 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1224 u32 cpos, phys_cpos, clusters, alloc_size;
1227 * We consider both start and len to be inclusive.
1229 cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1230 clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1234 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1242 * Hole or existing extent len can be arbitrary, so
1243 * cap it to our own allocation request.
1245 if (alloc_size > clusters)
1246 alloc_size = clusters;
1250 * We already have an allocation at this
1251 * region so we can safely skip it.
1256 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1265 clusters -= alloc_size;
1273 static int __ocfs2_remove_inode_range(struct inode *inode,
1274 struct buffer_head *di_bh,
1275 u32 cpos, u32 phys_cpos, u32 len,
1276 struct ocfs2_cached_dealloc_ctxt *dealloc)
1279 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
1280 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1281 struct inode *tl_inode = osb->osb_tl_inode;
1283 struct ocfs2_alloc_context *meta_ac = NULL;
1284 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1286 ret = ocfs2_lock_allocators(inode, di, 0, 1, NULL, &meta_ac);
1292 mutex_lock(&tl_inode->i_mutex);
1294 if (ocfs2_truncate_log_needs_flush(osb)) {
1295 ret = __ocfs2_flush_truncate_log(osb);
1302 handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
1303 if (handle == NULL) {
1309 ret = ocfs2_journal_access(handle, inode, di_bh,
1310 OCFS2_JOURNAL_ACCESS_WRITE);
1316 ret = ocfs2_remove_extent(inode, di_bh, cpos, len, handle, meta_ac,
1323 OCFS2_I(inode)->ip_clusters -= len;
1324 di->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1326 ret = ocfs2_journal_dirty(handle, di_bh);
1332 ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len);
1337 ocfs2_commit_trans(osb, handle);
1339 mutex_unlock(&tl_inode->i_mutex);
1342 ocfs2_free_alloc_context(meta_ac);
1348 * Truncate a byte range, avoiding pages within partial clusters. This
1349 * preserves those pages for the zeroing code to write to.
1351 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1354 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1356 struct address_space *mapping = inode->i_mapping;
1358 start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1359 end = byte_start + byte_len;
1360 end = end & ~(osb->s_clustersize - 1);
1363 unmap_mapping_range(mapping, start, end - start, 0);
1364 truncate_inode_pages_range(mapping, start, end - 1);
1368 static int ocfs2_zero_partial_clusters(struct inode *inode,
1372 u64 tmpend, end = start + len;
1373 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1374 unsigned int csize = osb->s_clustersize;
1378 * The "start" and "end" values are NOT necessarily part of
1379 * the range whose allocation is being deleted. Rather, this
1380 * is what the user passed in with the request. We must zero
1381 * partial clusters here. There's no need to worry about
1382 * physical allocation - the zeroing code knows to skip holes.
1384 mlog(0, "byte start: %llu, end: %llu\n",
1385 (unsigned long long)start, (unsigned long long)end);
1388 * If both edges are on a cluster boundary then there's no
1389 * zeroing required as the region is part of the allocation to
1392 if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1395 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1396 if (handle == NULL) {
1403 * We want to get the byte offset of the end of the 1st cluster.
1405 tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1409 mlog(0, "1st range: start: %llu, tmpend: %llu\n",
1410 (unsigned long long)start, (unsigned long long)tmpend);
1412 ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1418 * This may make start and end equal, but the zeroing
1419 * code will skip any work in that case so there's no
1420 * need to catch it up here.
1422 start = end & ~(osb->s_clustersize - 1);
1424 mlog(0, "2nd range: start: %llu, end: %llu\n",
1425 (unsigned long long)start, (unsigned long long)end);
1427 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1432 ocfs2_commit_trans(osb, handle);
1437 static int ocfs2_remove_inode_range(struct inode *inode,
1438 struct buffer_head *di_bh, u64 byte_start,
1442 u32 trunc_start, trunc_len, cpos, phys_cpos, alloc_size;
1443 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1444 struct ocfs2_cached_dealloc_ctxt dealloc;
1446 ocfs2_init_dealloc_ctxt(&dealloc);
1451 trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1452 trunc_len = (byte_start + byte_len) >> osb->s_clustersize_bits;
1453 if (trunc_len >= trunc_start)
1454 trunc_len -= trunc_start;
1458 mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u\n",
1459 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1460 (unsigned long long)byte_start,
1461 (unsigned long long)byte_len, trunc_start, trunc_len);
1463 ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1471 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1478 if (alloc_size > trunc_len)
1479 alloc_size = trunc_len;
1481 /* Only do work for non-holes */
1482 if (phys_cpos != 0) {
1483 ret = __ocfs2_remove_inode_range(inode, di_bh, cpos,
1484 phys_cpos, alloc_size,
1493 trunc_len -= alloc_size;
1496 ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1499 ocfs2_schedule_truncate_log_flush(osb, 1);
1500 ocfs2_run_deallocs(osb, &dealloc);
1506 * Parts of this function taken from xfs_change_file_space()
1508 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1509 loff_t f_pos, unsigned int cmd,
1510 struct ocfs2_space_resv *sr,
1516 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1517 struct buffer_head *di_bh = NULL;
1519 unsigned long long max_off = ocfs2_max_file_offset(inode->i_sb->s_blocksize_bits);
1521 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1524 mutex_lock(&inode->i_mutex);
1527 * This prevents concurrent writes on other nodes
1529 ret = ocfs2_rw_lock(inode, 1);
1535 ret = ocfs2_meta_lock(inode, &di_bh, 1);
1541 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1543 goto out_meta_unlock;
1546 switch (sr->l_whence) {
1547 case 0: /*SEEK_SET*/
1549 case 1: /*SEEK_CUR*/
1550 sr->l_start += f_pos;
1552 case 2: /*SEEK_END*/
1553 sr->l_start += i_size_read(inode);
1557 goto out_meta_unlock;
1561 llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1564 || sr->l_start > max_off
1565 || (sr->l_start + llen) < 0
1566 || (sr->l_start + llen) > max_off) {
1568 goto out_meta_unlock;
1570 size = sr->l_start + sr->l_len;
1572 if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) {
1573 if (sr->l_len <= 0) {
1575 goto out_meta_unlock;
1579 if (file && should_remove_suid(file->f_path.dentry)) {
1580 ret = __ocfs2_write_remove_suid(inode, di_bh);
1583 goto out_meta_unlock;
1587 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1589 case OCFS2_IOC_RESVSP:
1590 case OCFS2_IOC_RESVSP64:
1592 * This takes unsigned offsets, but the signed ones we
1593 * pass have been checked against overflow above.
1595 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1598 case OCFS2_IOC_UNRESVSP:
1599 case OCFS2_IOC_UNRESVSP64:
1600 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1606 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1609 goto out_meta_unlock;
1613 * We update c/mtime for these changes
1615 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1616 if (IS_ERR(handle)) {
1617 ret = PTR_ERR(handle);
1619 goto out_meta_unlock;
1622 if (change_size && i_size_read(inode) < size)
1623 i_size_write(inode, size);
1625 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1626 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1630 ocfs2_commit_trans(osb, handle);
1634 ocfs2_meta_unlock(inode, 1);
1636 ocfs2_rw_unlock(inode, 1);
1638 mutex_unlock(&inode->i_mutex);
1643 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1644 struct ocfs2_space_resv *sr)
1646 struct inode *inode = file->f_path.dentry->d_inode;
1647 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);;
1649 if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1650 !ocfs2_writes_unwritten_extents(osb))
1652 else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1653 !ocfs2_sparse_alloc(osb))
1656 if (!S_ISREG(inode->i_mode))
1659 if (!(file->f_mode & FMODE_WRITE))
1662 return __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1665 static long ocfs2_fallocate(struct inode *inode, int mode, loff_t offset,
1668 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1669 struct ocfs2_space_resv sr;
1670 int change_size = 1;
1672 if (!ocfs2_writes_unwritten_extents(osb))
1675 if (S_ISDIR(inode->i_mode))
1678 if (mode & FALLOC_FL_KEEP_SIZE)
1682 sr.l_start = (s64)offset;
1683 sr.l_len = (s64)len;
1685 return __ocfs2_change_file_space(NULL, inode, offset,
1686 OCFS2_IOC_RESVSP64, &sr, change_size);
1689 static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
1695 int ret = 0, meta_level = appending;
1696 struct inode *inode = dentry->d_inode;
1698 loff_t newsize, saved_pos;
1701 * We sample i_size under a read level meta lock to see if our write
1702 * is extending the file, if it is we back off and get a write level
1706 ret = ocfs2_meta_lock(inode, NULL, meta_level);
1713 /* Clear suid / sgid if necessary. We do this here
1714 * instead of later in the write path because
1715 * remove_suid() calls ->setattr without any hint that
1716 * we may have already done our cluster locking. Since
1717 * ocfs2_setattr() *must* take cluster locks to
1718 * proceeed, this will lead us to recursively lock the
1719 * inode. There's also the dinode i_size state which
1720 * can be lost via setattr during extending writes (we
1721 * set inode->i_size at the end of a write. */
1722 if (should_remove_suid(dentry)) {
1723 if (meta_level == 0) {
1724 ocfs2_meta_unlock(inode, meta_level);
1729 ret = ocfs2_write_remove_suid(inode);
1736 /* work on a copy of ppos until we're sure that we won't have
1737 * to recalculate it due to relocking. */
1739 saved_pos = i_size_read(inode);
1740 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos);
1745 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) {
1746 loff_t end = saved_pos + count;
1749 * Skip the O_DIRECT checks if we don't need
1752 if (!direct_io || !(*direct_io))
1756 * Allowing concurrent direct writes means
1757 * i_size changes wouldn't be synchronized, so
1758 * one node could wind up truncating another
1761 if (end > i_size_read(inode)) {
1767 * We don't fill holes during direct io, so
1768 * check for them here. If any are found, the
1769 * caller will have to retake some cluster
1770 * locks and initiate the io as buffered.
1772 ret = ocfs2_check_range_for_holes(inode, saved_pos,
1783 * The rest of this loop is concerned with legacy file
1784 * systems which don't support sparse files.
1787 newsize = count + saved_pos;
1789 mlog(0, "pos=%lld newsize=%lld cursize=%lld\n",
1790 (long long) saved_pos, (long long) newsize,
1791 (long long) i_size_read(inode));
1793 /* No need for a higher level metadata lock if we're
1794 * never going past i_size. */
1795 if (newsize <= i_size_read(inode))
1798 if (meta_level == 0) {
1799 ocfs2_meta_unlock(inode, meta_level);
1804 spin_lock(&OCFS2_I(inode)->ip_lock);
1805 clusters = ocfs2_clusters_for_bytes(inode->i_sb, newsize) -
1806 OCFS2_I(inode)->ip_clusters;
1807 spin_unlock(&OCFS2_I(inode)->ip_lock);
1809 mlog(0, "Writing at EOF, may need more allocation: "
1810 "i_size = %lld, newsize = %lld, need %u clusters\n",
1811 (long long) i_size_read(inode), (long long) newsize,
1814 /* We only want to continue the rest of this loop if
1815 * our extend will actually require more
1820 ret = ocfs2_extend_file(inode, NULL, newsize, count);
1833 ocfs2_meta_unlock(inode, meta_level);
1840 ocfs2_set_next_iovec(const struct iovec **iovp, size_t *basep, size_t bytes)
1842 const struct iovec *iov = *iovp;
1843 size_t base = *basep;
1846 int copy = min(bytes, iov->iov_len - base);
1850 if (iov->iov_len == base) {
1859 static struct page * ocfs2_get_write_source(char **ret_src_buf,
1860 const struct iovec *cur_iov,
1864 char *buf = cur_iov->iov_base + iov_offset;
1865 struct page *src_page = NULL;
1868 off = (unsigned long)(buf) & ~PAGE_CACHE_MASK;
1870 if (!segment_eq(get_fs(), KERNEL_DS)) {
1872 * Pull in the user page. We want to do this outside
1873 * of the meta data locks in order to preserve locking
1874 * order in case of page fault.
1876 ret = get_user_pages(current, current->mm,
1877 (unsigned long)buf & PAGE_CACHE_MASK, 1,
1878 0, 0, &src_page, NULL);
1880 *ret_src_buf = kmap(src_page) + off;
1882 src_page = ERR_PTR(-EFAULT);
1890 static void ocfs2_put_write_source(struct page *page)
1894 page_cache_release(page);
1898 static ssize_t ocfs2_file_buffered_write(struct file *file, loff_t *ppos,
1899 const struct iovec *iov,
1900 unsigned long nr_segs,
1902 ssize_t o_direct_written)
1905 ssize_t copied, total = 0;
1906 size_t iov_offset = 0, bytes;
1908 const struct iovec *cur_iov = iov;
1909 struct page *user_page, *page;
1910 char * uninitialized_var(buf);
1915 * handle partial DIO write. Adjust cur_iov if needed.
1917 ocfs2_set_next_iovec(&cur_iov, &iov_offset, o_direct_written);
1922 user_page = ocfs2_get_write_source(&buf, cur_iov, iov_offset);
1923 if (IS_ERR(user_page)) {
1924 ret = PTR_ERR(user_page);
1928 /* Stay within our page boundaries */
1929 bytes = min((PAGE_CACHE_SIZE - ((unsigned long)pos & ~PAGE_CACHE_MASK)),
1930 (PAGE_CACHE_SIZE - ((unsigned long)buf & ~PAGE_CACHE_MASK)));
1931 /* Stay within the vector boundary */
1932 bytes = min_t(size_t, bytes, cur_iov->iov_len - iov_offset);
1933 /* Stay within count */
1934 bytes = min(bytes, count);
1937 ret = ocfs2_write_begin(file, file->f_mapping, pos, bytes, 0,
1944 dst = kmap_atomic(page, KM_USER0);
1945 memcpy(dst + (pos & (PAGE_CACHE_SIZE - 1)), buf, bytes);
1946 kunmap_atomic(dst, KM_USER0);
1947 flush_dcache_page(page);
1948 ocfs2_put_write_source(user_page);
1950 copied = ocfs2_write_end(file, file->f_mapping, pos, bytes,
1951 bytes, page, fsdata);
1959 *ppos = pos + copied;
1962 ocfs2_set_next_iovec(&cur_iov, &iov_offset, copied);
1966 return total ? total : ret;
1969 static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
1970 const struct iovec *iov,
1971 unsigned long nr_segs,
1974 int ret, direct_io, appending, rw_level, have_alloc_sem = 0;
1975 int can_do_direct, sync = 0;
1976 ssize_t written = 0;
1977 size_t ocount; /* original count */
1978 size_t count; /* after file limit checks */
1979 loff_t *ppos = &iocb->ki_pos;
1980 struct file *file = iocb->ki_filp;
1981 struct inode *inode = file->f_path.dentry->d_inode;
1983 mlog_entry("(0x%p, %u, '%.*s')\n", file,
1984 (unsigned int)nr_segs,
1985 file->f_path.dentry->d_name.len,
1986 file->f_path.dentry->d_name.name);
1988 if (iocb->ki_left == 0)
1991 ret = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
1997 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1999 appending = file->f_flags & O_APPEND ? 1 : 0;
2000 direct_io = file->f_flags & O_DIRECT ? 1 : 0;
2002 mutex_lock(&inode->i_mutex);
2005 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
2007 down_read(&inode->i_alloc_sem);
2011 /* concurrent O_DIRECT writes are allowed */
2012 rw_level = !direct_io;
2013 ret = ocfs2_rw_lock(inode, rw_level);
2019 can_do_direct = direct_io;
2020 ret = ocfs2_prepare_inode_for_write(file->f_path.dentry, ppos,
2021 iocb->ki_left, appending,
2029 * We can't complete the direct I/O as requested, fall back to
2032 if (direct_io && !can_do_direct) {
2033 ocfs2_rw_unlock(inode, rw_level);
2034 up_read(&inode->i_alloc_sem);
2044 if (!sync && ((file->f_flags & O_SYNC) || IS_SYNC(inode)))
2048 * XXX: Is it ok to execute these checks a second time?
2050 ret = generic_write_checks(file, ppos, &count, S_ISBLK(inode->i_mode));
2055 * Set pos so that sync_page_range_nolock() below understands
2056 * where to start from. We might've moved it around via the
2057 * calls above. The range we want to actually sync starts from
2063 /* communicate with ocfs2_dio_end_io */
2064 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2067 written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
2068 ppos, count, ocount);
2074 written = ocfs2_file_buffered_write(file, ppos, iov, nr_segs,
2078 if (ret != -EFAULT || ret != -ENOSPC)
2085 /* buffered aio wouldn't have proper lock coverage today */
2086 BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
2089 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2090 * function pointer which is called when o_direct io completes so that
2091 * it can unlock our rw lock. (it's the clustered equivalent of
2092 * i_alloc_sem; protects truncate from racing with pending ios).
2093 * Unfortunately there are error cases which call end_io and others
2094 * that don't. so we don't have to unlock the rw_lock if either an
2095 * async dio is going to do it in the future or an end_io after an
2096 * error has already done it.
2098 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2105 ocfs2_rw_unlock(inode, rw_level);
2109 up_read(&inode->i_alloc_sem);
2111 if (written > 0 && sync) {
2114 err = sync_page_range_nolock(inode, file->f_mapping, pos, count);
2119 mutex_unlock(&inode->i_mutex);
2122 return written ? written : ret;
2125 static int ocfs2_splice_write_actor(struct pipe_inode_info *pipe,
2126 struct pipe_buffer *buf,
2127 struct splice_desc *sd)
2131 struct file *file = sd->u.file;
2132 unsigned int offset;
2133 struct page *page = NULL;
2137 ret = buf->ops->confirm(pipe, buf);
2141 offset = sd->pos & ~PAGE_CACHE_MASK;
2143 if (count + offset > PAGE_CACHE_SIZE)
2144 count = PAGE_CACHE_SIZE - offset;
2146 ret = ocfs2_write_begin(file, file->f_mapping, sd->pos, count, 0,
2153 src = buf->ops->map(pipe, buf, 1);
2154 dst = kmap_atomic(page, KM_USER1);
2155 memcpy(dst + offset, src + buf->offset, count);
2156 kunmap_atomic(dst, KM_USER1);
2157 buf->ops->unmap(pipe, buf, src);
2159 copied = ocfs2_write_end(file, file->f_mapping, sd->pos, count, count,
2168 return copied ? copied : ret;
2171 static ssize_t __ocfs2_file_splice_write(struct pipe_inode_info *pipe,
2178 struct address_space *mapping = out->f_mapping;
2179 struct inode *inode = mapping->host;
2180 struct splice_desc sd = {
2187 ret = __splice_from_pipe(pipe, &sd, ocfs2_splice_write_actor);
2191 if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) {
2192 err = generic_osync_inode(inode, mapping,
2193 OSYNC_METADATA|OSYNC_DATA);
2202 static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
2209 struct inode *inode = out->f_path.dentry->d_inode;
2211 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe,
2213 out->f_path.dentry->d_name.len,
2214 out->f_path.dentry->d_name.name);
2216 inode_double_lock(inode, pipe->inode);
2218 ret = ocfs2_rw_lock(inode, 1);
2224 ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, ppos, len, 0,
2231 /* ok, we're done with i_size and alloc work */
2232 ret = __ocfs2_file_splice_write(pipe, out, ppos, len, flags);
2235 ocfs2_rw_unlock(inode, 1);
2237 inode_double_unlock(inode, pipe->inode);
2243 static ssize_t ocfs2_file_splice_read(struct file *in,
2245 struct pipe_inode_info *pipe,
2250 struct inode *inode = in->f_path.dentry->d_inode;
2252 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe,
2254 in->f_path.dentry->d_name.len,
2255 in->f_path.dentry->d_name.name);
2258 * See the comment in ocfs2_file_aio_read()
2260 ret = ocfs2_meta_lock(inode, NULL, 0);
2265 ocfs2_meta_unlock(inode, 0);
2267 ret = generic_file_splice_read(in, ppos, pipe, len, flags);
2274 static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
2275 const struct iovec *iov,
2276 unsigned long nr_segs,
2279 int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
2280 struct file *filp = iocb->ki_filp;
2281 struct inode *inode = filp->f_path.dentry->d_inode;
2283 mlog_entry("(0x%p, %u, '%.*s')\n", filp,
2284 (unsigned int)nr_segs,
2285 filp->f_path.dentry->d_name.len,
2286 filp->f_path.dentry->d_name.name);
2295 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
2296 * need locks to protect pending reads from racing with truncate.
2298 if (filp->f_flags & O_DIRECT) {
2299 down_read(&inode->i_alloc_sem);
2302 ret = ocfs2_rw_lock(inode, 0);
2308 /* communicate with ocfs2_dio_end_io */
2309 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2313 * We're fine letting folks race truncates and extending
2314 * writes with read across the cluster, just like they can
2315 * locally. Hence no rw_lock during read.
2317 * Take and drop the meta data lock to update inode fields
2318 * like i_size. This allows the checks down below
2319 * generic_file_aio_read() a chance of actually working.
2321 ret = ocfs2_meta_lock_atime(inode, filp->f_vfsmnt, &lock_level);
2326 ocfs2_meta_unlock(inode, lock_level);
2328 ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
2330 mlog(ML_ERROR, "generic_file_aio_read returned -EINVAL\n");
2332 /* buffered aio wouldn't have proper lock coverage today */
2333 BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
2335 /* see ocfs2_file_aio_write */
2336 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2343 up_read(&inode->i_alloc_sem);
2345 ocfs2_rw_unlock(inode, rw_level);
2351 const struct inode_operations ocfs2_file_iops = {
2352 .setattr = ocfs2_setattr,
2353 .getattr = ocfs2_getattr,
2354 .permission = ocfs2_permission,
2355 .fallocate = ocfs2_fallocate,
2358 const struct inode_operations ocfs2_special_file_iops = {
2359 .setattr = ocfs2_setattr,
2360 .getattr = ocfs2_getattr,
2361 .permission = ocfs2_permission,
2364 const struct file_operations ocfs2_fops = {
2365 .read = do_sync_read,
2366 .write = do_sync_write,
2368 .fsync = ocfs2_sync_file,
2369 .release = ocfs2_file_release,
2370 .open = ocfs2_file_open,
2371 .aio_read = ocfs2_file_aio_read,
2372 .aio_write = ocfs2_file_aio_write,
2373 .ioctl = ocfs2_ioctl,
2374 #ifdef CONFIG_COMPAT
2375 .compat_ioctl = ocfs2_compat_ioctl,
2377 .splice_read = ocfs2_file_splice_read,
2378 .splice_write = ocfs2_file_splice_write,
2381 const struct file_operations ocfs2_dops = {
2382 .read = generic_read_dir,
2383 .readdir = ocfs2_readdir,
2384 .fsync = ocfs2_sync_file,
2385 .ioctl = ocfs2_ioctl,
2386 #ifdef CONFIG_COMPAT
2387 .compat_ioctl = ocfs2_compat_ioctl,