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 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
401 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
402 i_size_read(inode), 0);
406 goto bail_unlock_data;
409 /* alright, we're going to need to do a full blown alloc size
410 * change. Orphan the inode so that recovery can complete the
411 * truncate if necessary. This does the task of marking
413 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
416 goto bail_unlock_data;
419 status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc);
422 goto bail_unlock_data;
425 status = ocfs2_commit_truncate(osb, inode, di_bh, tc);
428 goto bail_unlock_data;
431 /* TODO: orphan dir cleanup here. */
433 ocfs2_data_unlock(inode, 1);
435 up_write(&OCFS2_I(inode)->ip_alloc_sem);
444 * extend allocation only here.
445 * we'll update all the disk stuff, and oip->alloc_size
447 * expect stuff to be locked, a transaction started and enough data /
448 * metadata reservations in the contexts.
450 * Will return -EAGAIN, and a reason if a restart is needed.
451 * If passed in, *reason will always be set, even in error.
453 int ocfs2_do_extend_allocation(struct ocfs2_super *osb,
458 struct buffer_head *fe_bh,
460 struct ocfs2_alloc_context *data_ac,
461 struct ocfs2_alloc_context *meta_ac,
462 enum ocfs2_alloc_restarted *reason_ret)
466 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
467 enum ocfs2_alloc_restarted reason = RESTART_NONE;
468 u32 bit_off, num_bits;
472 BUG_ON(!clusters_to_add);
475 flags = OCFS2_EXT_UNWRITTEN;
477 free_extents = ocfs2_num_free_extents(osb, inode, fe);
478 if (free_extents < 0) {
479 status = free_extents;
484 /* there are two cases which could cause us to EAGAIN in the
485 * we-need-more-metadata case:
486 * 1) we haven't reserved *any*
487 * 2) we are so fragmented, we've needed to add metadata too
489 if (!free_extents && !meta_ac) {
490 mlog(0, "we haven't reserved any metadata!\n");
492 reason = RESTART_META;
494 } else if ((!free_extents)
495 && (ocfs2_alloc_context_bits_left(meta_ac)
496 < ocfs2_extend_meta_needed(fe))) {
497 mlog(0, "filesystem is really fragmented...\n");
499 reason = RESTART_META;
503 status = __ocfs2_claim_clusters(osb, handle, data_ac, 1,
504 clusters_to_add, &bit_off, &num_bits);
506 if (status != -ENOSPC)
511 BUG_ON(num_bits > clusters_to_add);
513 /* reserve our write early -- insert_extent may update the inode */
514 status = ocfs2_journal_access(handle, inode, fe_bh,
515 OCFS2_JOURNAL_ACCESS_WRITE);
521 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
522 mlog(0, "Allocating %u clusters at block %u for inode %llu\n",
523 num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
524 status = ocfs2_insert_extent(osb, handle, inode, fe_bh,
525 *logical_offset, block, num_bits,
532 status = ocfs2_journal_dirty(handle, fe_bh);
538 clusters_to_add -= num_bits;
539 *logical_offset += num_bits;
541 if (clusters_to_add) {
542 mlog(0, "need to alloc once more, clusters = %u, wanted = "
543 "%u\n", fe->i_clusters, clusters_to_add);
545 reason = RESTART_TRANS;
551 *reason_ret = reason;
556 * For a given allocation, determine which allocators will need to be
557 * accessed, and lock them, reserving the appropriate number of bits.
559 * Sparse file systems call this from ocfs2_write_begin_nolock()
560 * and ocfs2_allocate_unwritten_extents().
562 * File systems which don't support holes call this from
563 * ocfs2_extend_allocation().
565 int ocfs2_lock_allocators(struct inode *inode, struct ocfs2_dinode *di,
566 u32 clusters_to_add, u32 extents_to_split,
567 struct ocfs2_alloc_context **data_ac,
568 struct ocfs2_alloc_context **meta_ac)
570 int ret = 0, num_free_extents;
571 unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
572 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
578 BUG_ON(clusters_to_add != 0 && data_ac == NULL);
580 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, "
581 "clusters_to_add = %u, extents_to_split = %u\n",
582 (unsigned long long)OCFS2_I(inode)->ip_blkno, i_size_read(inode),
583 le32_to_cpu(di->i_clusters), clusters_to_add, extents_to_split);
585 num_free_extents = ocfs2_num_free_extents(osb, inode, di);
586 if (num_free_extents < 0) {
587 ret = num_free_extents;
593 * Sparse allocation file systems need to be more conservative
594 * with reserving room for expansion - the actual allocation
595 * happens while we've got a journal handle open so re-taking
596 * a cluster lock (because we ran out of room for another
597 * extent) will violate ordering rules.
599 * Most of the time we'll only be seeing this 1 cluster at a time
602 * Always lock for any unwritten extents - we might want to
603 * add blocks during a split.
605 if (!num_free_extents ||
606 (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
607 ret = ocfs2_reserve_new_metadata(osb, di, meta_ac);
615 if (clusters_to_add == 0)
618 ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
628 ocfs2_free_alloc_context(*meta_ac);
633 * We cannot have an error and a non null *data_ac.
640 static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
641 u32 clusters_to_add, int mark_unwritten)
644 int restart_func = 0;
647 struct buffer_head *bh = NULL;
648 struct ocfs2_dinode *fe = NULL;
649 handle_t *handle = NULL;
650 struct ocfs2_alloc_context *data_ac = NULL;
651 struct ocfs2_alloc_context *meta_ac = NULL;
652 enum ocfs2_alloc_restarted why;
653 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
655 mlog_entry("(clusters_to_add = %u)\n", clusters_to_add);
658 * This function only exists for file systems which don't
661 BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
663 status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &bh,
664 OCFS2_BH_CACHED, inode);
670 fe = (struct ocfs2_dinode *) bh->b_data;
671 if (!OCFS2_IS_VALID_DINODE(fe)) {
672 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
678 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
680 status = ocfs2_lock_allocators(inode, fe, clusters_to_add, 0, &data_ac,
687 credits = ocfs2_calc_extend_credits(osb->sb, fe, clusters_to_add);
688 handle = ocfs2_start_trans(osb, credits);
689 if (IS_ERR(handle)) {
690 status = PTR_ERR(handle);
696 restarted_transaction:
697 /* reserve a write to the file entry early on - that we if we
698 * run out of credits in the allocation path, we can still
700 status = ocfs2_journal_access(handle, inode, bh,
701 OCFS2_JOURNAL_ACCESS_WRITE);
707 prev_clusters = OCFS2_I(inode)->ip_clusters;
709 status = ocfs2_do_extend_allocation(osb,
719 if ((status < 0) && (status != -EAGAIN)) {
720 if (status != -ENOSPC)
725 status = ocfs2_journal_dirty(handle, bh);
731 spin_lock(&OCFS2_I(inode)->ip_lock);
732 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
733 spin_unlock(&OCFS2_I(inode)->ip_lock);
735 if (why != RESTART_NONE && clusters_to_add) {
736 if (why == RESTART_META) {
737 mlog(0, "restarting function.\n");
740 BUG_ON(why != RESTART_TRANS);
742 mlog(0, "restarting transaction.\n");
743 /* TODO: This can be more intelligent. */
744 credits = ocfs2_calc_extend_credits(osb->sb,
747 status = ocfs2_extend_trans(handle, credits);
749 /* handle still has to be committed at
755 goto restarted_transaction;
759 mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
760 le32_to_cpu(fe->i_clusters),
761 (unsigned long long)le64_to_cpu(fe->i_size));
762 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
763 OCFS2_I(inode)->ip_clusters, i_size_read(inode));
767 ocfs2_commit_trans(osb, handle);
771 ocfs2_free_alloc_context(data_ac);
775 ocfs2_free_alloc_context(meta_ac);
778 if ((!status) && restart_func) {
791 /* Some parts of this taken from generic_cont_expand, which turned out
792 * to be too fragile to do exactly what we need without us having to
793 * worry about recursive locking in ->prepare_write() and
794 * ->commit_write(). */
795 static int ocfs2_write_zero_page(struct inode *inode,
798 struct address_space *mapping = inode->i_mapping;
802 handle_t *handle = NULL;
805 offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */
806 /* ugh. in prepare/commit_write, if from==to==start of block, we
807 ** skip the prepare. make sure we never send an offset for the start
810 if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) {
813 index = size >> PAGE_CACHE_SHIFT;
815 page = grab_cache_page(mapping, index);
822 ret = ocfs2_prepare_write_nolock(inode, page, offset, offset);
828 if (ocfs2_should_order_data(inode)) {
829 handle = ocfs2_start_walk_page_trans(inode, page, offset,
831 if (IS_ERR(handle)) {
832 ret = PTR_ERR(handle);
838 /* must not update i_size! */
839 ret = block_commit_write(page, offset, offset);
846 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
849 page_cache_release(page);
854 static int ocfs2_zero_extend(struct inode *inode,
859 struct super_block *sb = inode->i_sb;
861 start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
862 while (start_off < zero_to_size) {
863 ret = ocfs2_write_zero_page(inode, start_off);
869 start_off += sb->s_blocksize;
872 * Very large extends have the potential to lock up
873 * the cpu for extended periods of time.
882 int ocfs2_extend_no_holes(struct inode *inode, u64 new_i_size, u64 zero_to)
886 struct ocfs2_inode_info *oi = OCFS2_I(inode);
888 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
889 if (clusters_to_add < oi->ip_clusters)
892 clusters_to_add -= oi->ip_clusters;
894 if (clusters_to_add) {
895 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
904 * Call this even if we don't add any clusters to the tree. We
905 * still need to zero the area between the old i_size and the
908 ret = ocfs2_zero_extend(inode, zero_to);
916 static int ocfs2_extend_file(struct inode *inode,
917 struct buffer_head *di_bh,
920 int ret = 0, data_locked = 0;
921 struct ocfs2_inode_info *oi = OCFS2_I(inode);
925 /* setattr sometimes calls us like this. */
929 if (i_size_read(inode) == new_i_size)
931 BUG_ON(new_i_size < i_size_read(inode));
934 * Fall through for converting inline data, even if the fs
935 * supports sparse files.
937 * The check for inline data here is legal - nobody can add
938 * the feature since we have i_mutex. We must check it again
939 * after acquiring ip_alloc_sem though, as paths like mmap
940 * might have raced us to converting the inode to extents.
942 if (!(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
943 && ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
944 goto out_update_size;
947 * protect the pages that ocfs2_zero_extend is going to be
948 * pulling into the page cache.. we do this before the
949 * metadata extend so that we don't get into the situation
950 * where we've extended the metadata but can't get the data
953 ret = ocfs2_data_lock(inode, 1);
961 * The alloc sem blocks people in read/write from reading our
962 * allocation until we're done changing it. We depend on
963 * i_mutex to block other extend/truncate calls while we're
966 down_write(&oi->ip_alloc_sem);
968 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
970 * We can optimize small extends by keeping the inodes
973 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
974 up_write(&oi->ip_alloc_sem);
975 goto out_update_size;
978 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
980 up_write(&oi->ip_alloc_sem);
987 if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
988 ret = ocfs2_extend_no_holes(inode, new_i_size, new_i_size);
990 up_write(&oi->ip_alloc_sem);
998 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1004 ocfs2_data_unlock(inode, 1);
1010 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1012 int status = 0, size_change;
1013 struct inode *inode = dentry->d_inode;
1014 struct super_block *sb = inode->i_sb;
1015 struct ocfs2_super *osb = OCFS2_SB(sb);
1016 struct buffer_head *bh = NULL;
1017 handle_t *handle = NULL;
1019 mlog_entry("(0x%p, '%.*s')\n", dentry,
1020 dentry->d_name.len, dentry->d_name.name);
1022 if (attr->ia_valid & ATTR_MODE)
1023 mlog(0, "mode change: %d\n", attr->ia_mode);
1024 if (attr->ia_valid & ATTR_UID)
1025 mlog(0, "uid change: %d\n", attr->ia_uid);
1026 if (attr->ia_valid & ATTR_GID)
1027 mlog(0, "gid change: %d\n", attr->ia_gid);
1028 if (attr->ia_valid & ATTR_SIZE)
1029 mlog(0, "size change...\n");
1030 if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME))
1031 mlog(0, "time change...\n");
1033 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1034 | ATTR_GID | ATTR_UID | ATTR_MODE)
1035 if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) {
1036 mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid);
1040 status = inode_change_ok(inode, attr);
1044 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1046 status = ocfs2_rw_lock(inode, 1);
1053 status = ocfs2_meta_lock(inode, &bh, 1);
1055 if (status != -ENOENT)
1057 goto bail_unlock_rw;
1060 if (size_change && attr->ia_size != i_size_read(inode)) {
1061 if (attr->ia_size > sb->s_maxbytes) {
1066 if (i_size_read(inode) > attr->ia_size)
1067 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1069 status = ocfs2_extend_file(inode, bh, attr->ia_size);
1071 if (status != -ENOSPC)
1078 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1079 if (IS_ERR(handle)) {
1080 status = PTR_ERR(handle);
1086 * This will intentionally not wind up calling vmtruncate(),
1087 * since all the work for a size change has been done above.
1088 * Otherwise, we could get into problems with truncate as
1089 * ip_alloc_sem is used there to protect against i_size
1092 status = inode_setattr(inode, attr);
1098 status = ocfs2_mark_inode_dirty(handle, inode, bh);
1103 ocfs2_commit_trans(osb, handle);
1105 ocfs2_meta_unlock(inode, 1);
1108 ocfs2_rw_unlock(inode, 1);
1117 int ocfs2_getattr(struct vfsmount *mnt,
1118 struct dentry *dentry,
1121 struct inode *inode = dentry->d_inode;
1122 struct super_block *sb = dentry->d_inode->i_sb;
1123 struct ocfs2_super *osb = sb->s_fs_info;
1128 err = ocfs2_inode_revalidate(dentry);
1135 generic_fillattr(inode, stat);
1137 /* We set the blksize from the cluster size for performance */
1138 stat->blksize = osb->s_clustersize;
1146 int ocfs2_permission(struct inode *inode, int mask, struct nameidata *nd)
1152 ret = ocfs2_meta_lock(inode, NULL, 0);
1159 ret = generic_permission(inode, mask, NULL);
1161 ocfs2_meta_unlock(inode, 0);
1167 static int __ocfs2_write_remove_suid(struct inode *inode,
1168 struct buffer_head *bh)
1172 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1173 struct ocfs2_dinode *di;
1175 mlog_entry("(Inode %llu, mode 0%o)\n",
1176 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_mode);
1178 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1179 if (handle == NULL) {
1185 ret = ocfs2_journal_access(handle, inode, bh,
1186 OCFS2_JOURNAL_ACCESS_WRITE);
1192 inode->i_mode &= ~S_ISUID;
1193 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1194 inode->i_mode &= ~S_ISGID;
1196 di = (struct ocfs2_dinode *) bh->b_data;
1197 di->i_mode = cpu_to_le16(inode->i_mode);
1199 ret = ocfs2_journal_dirty(handle, bh);
1204 ocfs2_commit_trans(osb, handle);
1211 * Will look for holes and unwritten extents in the range starting at
1212 * pos for count bytes (inclusive).
1214 static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1218 unsigned int extent_flags;
1219 u32 cpos, clusters, extent_len, phys_cpos;
1220 struct super_block *sb = inode->i_sb;
1222 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1223 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1226 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1233 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
1238 if (extent_len > clusters)
1239 extent_len = clusters;
1241 clusters -= extent_len;
1248 static int ocfs2_write_remove_suid(struct inode *inode)
1251 struct buffer_head *bh = NULL;
1252 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1254 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
1255 oi->ip_blkno, &bh, OCFS2_BH_CACHED, inode);
1261 ret = __ocfs2_write_remove_suid(inode, bh);
1268 * Allocate enough extents to cover the region starting at byte offset
1269 * start for len bytes. Existing extents are skipped, any extents
1270 * added are marked as "unwritten".
1272 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1276 u32 cpos, phys_cpos, clusters, alloc_size;
1277 u64 end = start + len;
1278 struct buffer_head *di_bh = NULL;
1280 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1281 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
1282 OCFS2_I(inode)->ip_blkno, &di_bh,
1283 OCFS2_BH_CACHED, inode);
1290 * Nothing to do if the requested reservation range
1291 * fits within the inode.
1293 if (ocfs2_size_fits_inline_data(di_bh, end))
1296 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1304 * We consider both start and len to be inclusive.
1306 cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1307 clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1311 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1319 * Hole or existing extent len can be arbitrary, so
1320 * cap it to our own allocation request.
1322 if (alloc_size > clusters)
1323 alloc_size = clusters;
1327 * We already have an allocation at this
1328 * region so we can safely skip it.
1333 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1342 clusters -= alloc_size;
1352 static int __ocfs2_remove_inode_range(struct inode *inode,
1353 struct buffer_head *di_bh,
1354 u32 cpos, u32 phys_cpos, u32 len,
1355 struct ocfs2_cached_dealloc_ctxt *dealloc)
1358 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
1359 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1360 struct inode *tl_inode = osb->osb_tl_inode;
1362 struct ocfs2_alloc_context *meta_ac = NULL;
1363 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1365 ret = ocfs2_lock_allocators(inode, di, 0, 1, NULL, &meta_ac);
1371 mutex_lock(&tl_inode->i_mutex);
1373 if (ocfs2_truncate_log_needs_flush(osb)) {
1374 ret = __ocfs2_flush_truncate_log(osb);
1381 handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
1382 if (handle == NULL) {
1388 ret = ocfs2_journal_access(handle, inode, di_bh,
1389 OCFS2_JOURNAL_ACCESS_WRITE);
1395 ret = ocfs2_remove_extent(inode, di_bh, cpos, len, handle, meta_ac,
1402 OCFS2_I(inode)->ip_clusters -= len;
1403 di->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1405 ret = ocfs2_journal_dirty(handle, di_bh);
1411 ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len);
1416 ocfs2_commit_trans(osb, handle);
1418 mutex_unlock(&tl_inode->i_mutex);
1421 ocfs2_free_alloc_context(meta_ac);
1427 * Truncate a byte range, avoiding pages within partial clusters. This
1428 * preserves those pages for the zeroing code to write to.
1430 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1433 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1435 struct address_space *mapping = inode->i_mapping;
1437 start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1438 end = byte_start + byte_len;
1439 end = end & ~(osb->s_clustersize - 1);
1442 unmap_mapping_range(mapping, start, end - start, 0);
1443 truncate_inode_pages_range(mapping, start, end - 1);
1447 static int ocfs2_zero_partial_clusters(struct inode *inode,
1451 u64 tmpend, end = start + len;
1452 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1453 unsigned int csize = osb->s_clustersize;
1457 * The "start" and "end" values are NOT necessarily part of
1458 * the range whose allocation is being deleted. Rather, this
1459 * is what the user passed in with the request. We must zero
1460 * partial clusters here. There's no need to worry about
1461 * physical allocation - the zeroing code knows to skip holes.
1463 mlog(0, "byte start: %llu, end: %llu\n",
1464 (unsigned long long)start, (unsigned long long)end);
1467 * If both edges are on a cluster boundary then there's no
1468 * zeroing required as the region is part of the allocation to
1471 if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1474 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1475 if (handle == NULL) {
1482 * We want to get the byte offset of the end of the 1st cluster.
1484 tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1488 mlog(0, "1st range: start: %llu, tmpend: %llu\n",
1489 (unsigned long long)start, (unsigned long long)tmpend);
1491 ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1497 * This may make start and end equal, but the zeroing
1498 * code will skip any work in that case so there's no
1499 * need to catch it up here.
1501 start = end & ~(osb->s_clustersize - 1);
1503 mlog(0, "2nd range: start: %llu, end: %llu\n",
1504 (unsigned long long)start, (unsigned long long)end);
1506 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1511 ocfs2_commit_trans(osb, handle);
1516 static int ocfs2_remove_inode_range(struct inode *inode,
1517 struct buffer_head *di_bh, u64 byte_start,
1521 u32 trunc_start, trunc_len, cpos, phys_cpos, alloc_size;
1522 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1523 struct ocfs2_cached_dealloc_ctxt dealloc;
1525 ocfs2_init_dealloc_ctxt(&dealloc);
1530 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1531 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1532 byte_start + byte_len, 1);
1538 trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1539 trunc_len = (byte_start + byte_len) >> osb->s_clustersize_bits;
1540 if (trunc_len >= trunc_start)
1541 trunc_len -= trunc_start;
1545 mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u\n",
1546 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1547 (unsigned long long)byte_start,
1548 (unsigned long long)byte_len, trunc_start, trunc_len);
1550 ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1558 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1565 if (alloc_size > trunc_len)
1566 alloc_size = trunc_len;
1568 /* Only do work for non-holes */
1569 if (phys_cpos != 0) {
1570 ret = __ocfs2_remove_inode_range(inode, di_bh, cpos,
1571 phys_cpos, alloc_size,
1580 trunc_len -= alloc_size;
1583 ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1586 ocfs2_schedule_truncate_log_flush(osb, 1);
1587 ocfs2_run_deallocs(osb, &dealloc);
1593 * Parts of this function taken from xfs_change_file_space()
1595 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1596 loff_t f_pos, unsigned int cmd,
1597 struct ocfs2_space_resv *sr,
1603 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1604 struct buffer_head *di_bh = NULL;
1606 unsigned long long max_off = inode->i_sb->s_maxbytes;
1608 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1611 mutex_lock(&inode->i_mutex);
1614 * This prevents concurrent writes on other nodes
1616 ret = ocfs2_rw_lock(inode, 1);
1622 ret = ocfs2_meta_lock(inode, &di_bh, 1);
1628 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1630 goto out_meta_unlock;
1633 switch (sr->l_whence) {
1634 case 0: /*SEEK_SET*/
1636 case 1: /*SEEK_CUR*/
1637 sr->l_start += f_pos;
1639 case 2: /*SEEK_END*/
1640 sr->l_start += i_size_read(inode);
1644 goto out_meta_unlock;
1648 llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1651 || sr->l_start > max_off
1652 || (sr->l_start + llen) < 0
1653 || (sr->l_start + llen) > max_off) {
1655 goto out_meta_unlock;
1657 size = sr->l_start + sr->l_len;
1659 if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) {
1660 if (sr->l_len <= 0) {
1662 goto out_meta_unlock;
1666 if (file && should_remove_suid(file->f_path.dentry)) {
1667 ret = __ocfs2_write_remove_suid(inode, di_bh);
1670 goto out_meta_unlock;
1674 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1676 case OCFS2_IOC_RESVSP:
1677 case OCFS2_IOC_RESVSP64:
1679 * This takes unsigned offsets, but the signed ones we
1680 * pass have been checked against overflow above.
1682 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1685 case OCFS2_IOC_UNRESVSP:
1686 case OCFS2_IOC_UNRESVSP64:
1687 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1693 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1696 goto out_meta_unlock;
1700 * We update c/mtime for these changes
1702 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1703 if (IS_ERR(handle)) {
1704 ret = PTR_ERR(handle);
1706 goto out_meta_unlock;
1709 if (change_size && i_size_read(inode) < size)
1710 i_size_write(inode, size);
1712 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1713 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1717 ocfs2_commit_trans(osb, handle);
1721 ocfs2_meta_unlock(inode, 1);
1723 ocfs2_rw_unlock(inode, 1);
1725 mutex_unlock(&inode->i_mutex);
1730 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1731 struct ocfs2_space_resv *sr)
1733 struct inode *inode = file->f_path.dentry->d_inode;
1734 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);;
1736 if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1737 !ocfs2_writes_unwritten_extents(osb))
1739 else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1740 !ocfs2_sparse_alloc(osb))
1743 if (!S_ISREG(inode->i_mode))
1746 if (!(file->f_mode & FMODE_WRITE))
1749 return __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1752 static long ocfs2_fallocate(struct inode *inode, int mode, loff_t offset,
1755 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1756 struct ocfs2_space_resv sr;
1757 int change_size = 1;
1759 if (!ocfs2_writes_unwritten_extents(osb))
1762 if (S_ISDIR(inode->i_mode))
1765 if (mode & FALLOC_FL_KEEP_SIZE)
1769 sr.l_start = (s64)offset;
1770 sr.l_len = (s64)len;
1772 return __ocfs2_change_file_space(NULL, inode, offset,
1773 OCFS2_IOC_RESVSP64, &sr, change_size);
1776 static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
1782 int ret = 0, meta_level = 0;
1783 struct inode *inode = dentry->d_inode;
1784 loff_t saved_pos, end;
1787 * We start with a read level meta lock and only jump to an ex
1788 * if we need to make modifications here.
1791 ret = ocfs2_meta_lock(inode, NULL, meta_level);
1798 /* Clear suid / sgid if necessary. We do this here
1799 * instead of later in the write path because
1800 * remove_suid() calls ->setattr without any hint that
1801 * we may have already done our cluster locking. Since
1802 * ocfs2_setattr() *must* take cluster locks to
1803 * proceeed, this will lead us to recursively lock the
1804 * inode. There's also the dinode i_size state which
1805 * can be lost via setattr during extending writes (we
1806 * set inode->i_size at the end of a write. */
1807 if (should_remove_suid(dentry)) {
1808 if (meta_level == 0) {
1809 ocfs2_meta_unlock(inode, meta_level);
1814 ret = ocfs2_write_remove_suid(inode);
1821 /* work on a copy of ppos until we're sure that we won't have
1822 * to recalculate it due to relocking. */
1824 saved_pos = i_size_read(inode);
1825 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos);
1830 end = saved_pos + count;
1833 * Skip the O_DIRECT checks if we don't need
1836 if (!direct_io || !(*direct_io))
1840 * There's no sane way to do direct writes to an inode
1843 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1849 * Allowing concurrent direct writes means
1850 * i_size changes wouldn't be synchronized, so
1851 * one node could wind up truncating another
1854 if (end > i_size_read(inode)) {
1860 * We don't fill holes during direct io, so
1861 * check for them here. If any are found, the
1862 * caller will have to retake some cluster
1863 * locks and initiate the io as buffered.
1865 ret = ocfs2_check_range_for_holes(inode, saved_pos, count);
1878 ocfs2_meta_unlock(inode, meta_level);
1884 static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
1885 const struct iovec *iov,
1886 unsigned long nr_segs,
1889 int ret, direct_io, appending, rw_level, have_alloc_sem = 0;
1891 ssize_t written = 0;
1892 size_t ocount; /* original count */
1893 size_t count; /* after file limit checks */
1894 loff_t *ppos = &iocb->ki_pos;
1895 struct file *file = iocb->ki_filp;
1896 struct inode *inode = file->f_path.dentry->d_inode;
1898 mlog_entry("(0x%p, %u, '%.*s')\n", file,
1899 (unsigned int)nr_segs,
1900 file->f_path.dentry->d_name.len,
1901 file->f_path.dentry->d_name.name);
1903 if (iocb->ki_left == 0)
1906 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1908 appending = file->f_flags & O_APPEND ? 1 : 0;
1909 direct_io = file->f_flags & O_DIRECT ? 1 : 0;
1911 mutex_lock(&inode->i_mutex);
1914 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
1916 down_read(&inode->i_alloc_sem);
1920 /* concurrent O_DIRECT writes are allowed */
1921 rw_level = !direct_io;
1922 ret = ocfs2_rw_lock(inode, rw_level);
1928 can_do_direct = direct_io;
1929 ret = ocfs2_prepare_inode_for_write(file->f_path.dentry, ppos,
1930 iocb->ki_left, appending,
1938 * We can't complete the direct I/O as requested, fall back to
1941 if (direct_io && !can_do_direct) {
1942 ocfs2_rw_unlock(inode, rw_level);
1943 up_read(&inode->i_alloc_sem);
1952 /* communicate with ocfs2_dio_end_io */
1953 ocfs2_iocb_set_rw_locked(iocb, rw_level);
1956 ret = generic_segment_checks(iov, &nr_segs, &ocount,
1961 ret = generic_write_checks(file, ppos, &count,
1962 S_ISBLK(inode->i_mode));
1966 written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
1967 ppos, count, ocount);
1973 written = generic_file_aio_write_nolock(iocb, iov, nr_segs,
1978 /* buffered aio wouldn't have proper lock coverage today */
1979 BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
1982 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
1983 * function pointer which is called when o_direct io completes so that
1984 * it can unlock our rw lock. (it's the clustered equivalent of
1985 * i_alloc_sem; protects truncate from racing with pending ios).
1986 * Unfortunately there are error cases which call end_io and others
1987 * that don't. so we don't have to unlock the rw_lock if either an
1988 * async dio is going to do it in the future or an end_io after an
1989 * error has already done it.
1991 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
1998 ocfs2_rw_unlock(inode, rw_level);
2002 up_read(&inode->i_alloc_sem);
2004 mutex_unlock(&inode->i_mutex);
2007 return written ? written : ret;
2010 static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
2017 struct inode *inode = out->f_path.dentry->d_inode;
2019 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe,
2021 out->f_path.dentry->d_name.len,
2022 out->f_path.dentry->d_name.name);
2024 inode_double_lock(inode, pipe->inode);
2026 ret = ocfs2_rw_lock(inode, 1);
2032 ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, ppos, len, 0,
2039 ret = generic_file_splice_write_nolock(pipe, out, ppos, len, flags);
2042 ocfs2_rw_unlock(inode, 1);
2044 inode_double_unlock(inode, pipe->inode);
2050 static ssize_t ocfs2_file_splice_read(struct file *in,
2052 struct pipe_inode_info *pipe,
2057 struct inode *inode = in->f_path.dentry->d_inode;
2059 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe,
2061 in->f_path.dentry->d_name.len,
2062 in->f_path.dentry->d_name.name);
2065 * See the comment in ocfs2_file_aio_read()
2067 ret = ocfs2_meta_lock(inode, NULL, 0);
2072 ocfs2_meta_unlock(inode, 0);
2074 ret = generic_file_splice_read(in, ppos, pipe, len, flags);
2081 static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
2082 const struct iovec *iov,
2083 unsigned long nr_segs,
2086 int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
2087 struct file *filp = iocb->ki_filp;
2088 struct inode *inode = filp->f_path.dentry->d_inode;
2090 mlog_entry("(0x%p, %u, '%.*s')\n", filp,
2091 (unsigned int)nr_segs,
2092 filp->f_path.dentry->d_name.len,
2093 filp->f_path.dentry->d_name.name);
2102 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
2103 * need locks to protect pending reads from racing with truncate.
2105 if (filp->f_flags & O_DIRECT) {
2106 down_read(&inode->i_alloc_sem);
2109 ret = ocfs2_rw_lock(inode, 0);
2115 /* communicate with ocfs2_dio_end_io */
2116 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2120 * We're fine letting folks race truncates and extending
2121 * writes with read across the cluster, just like they can
2122 * locally. Hence no rw_lock during read.
2124 * Take and drop the meta data lock to update inode fields
2125 * like i_size. This allows the checks down below
2126 * generic_file_aio_read() a chance of actually working.
2128 ret = ocfs2_meta_lock_atime(inode, filp->f_vfsmnt, &lock_level);
2133 ocfs2_meta_unlock(inode, lock_level);
2135 ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
2137 mlog(ML_ERROR, "generic_file_aio_read returned -EINVAL\n");
2139 /* buffered aio wouldn't have proper lock coverage today */
2140 BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
2142 /* see ocfs2_file_aio_write */
2143 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2150 up_read(&inode->i_alloc_sem);
2152 ocfs2_rw_unlock(inode, rw_level);
2158 const struct inode_operations ocfs2_file_iops = {
2159 .setattr = ocfs2_setattr,
2160 .getattr = ocfs2_getattr,
2161 .permission = ocfs2_permission,
2162 .fallocate = ocfs2_fallocate,
2165 const struct inode_operations ocfs2_special_file_iops = {
2166 .setattr = ocfs2_setattr,
2167 .getattr = ocfs2_getattr,
2168 .permission = ocfs2_permission,
2171 const struct file_operations ocfs2_fops = {
2172 .read = do_sync_read,
2173 .write = do_sync_write,
2175 .fsync = ocfs2_sync_file,
2176 .release = ocfs2_file_release,
2177 .open = ocfs2_file_open,
2178 .aio_read = ocfs2_file_aio_read,
2179 .aio_write = ocfs2_file_aio_write,
2180 .ioctl = ocfs2_ioctl,
2181 #ifdef CONFIG_COMPAT
2182 .compat_ioctl = ocfs2_compat_ioctl,
2184 .splice_read = ocfs2_file_splice_read,
2185 .splice_write = ocfs2_file_splice_write,
2188 const struct file_operations ocfs2_dops = {
2189 .read = generic_read_dir,
2190 .readdir = ocfs2_readdir,
2191 .fsync = ocfs2_sync_file,
2192 .ioctl = ocfs2_ioctl,
2193 #ifdef CONFIG_COMPAT
2194 .compat_ioctl = ocfs2_compat_ioctl,