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/pipe_fs_i.h>
35 #include <linux/mount.h>
37 #define MLOG_MASK_PREFIX ML_INODE
38 #include <cluster/masklog.h>
46 #include "extent_map.h"
56 #include "buffer_head_io.h"
58 static int ocfs2_sync_inode(struct inode *inode)
60 filemap_fdatawrite(inode->i_mapping);
61 return sync_mapping_buffers(inode->i_mapping);
64 static int ocfs2_file_open(struct inode *inode, struct file *file)
67 int mode = file->f_flags;
68 struct ocfs2_inode_info *oi = OCFS2_I(inode);
70 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
71 file->f_dentry->d_name.len, file->f_dentry->d_name.name);
73 spin_lock(&oi->ip_lock);
75 /* Check that the inode hasn't been wiped from disk by another
76 * node. If it hasn't then we're safe as long as we hold the
77 * spin lock until our increment of open count. */
78 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
79 spin_unlock(&oi->ip_lock);
86 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
89 spin_unlock(&oi->ip_lock);
96 static int ocfs2_file_release(struct inode *inode, struct file *file)
98 struct ocfs2_inode_info *oi = OCFS2_I(inode);
100 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
101 file->f_dentry->d_name.len,
102 file->f_dentry->d_name.name);
104 spin_lock(&oi->ip_lock);
105 if (!--oi->ip_open_count)
106 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
107 spin_unlock(&oi->ip_lock);
114 static int ocfs2_sync_file(struct file *file,
115 struct dentry *dentry,
120 struct inode *inode = dentry->d_inode;
121 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
123 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync,
124 dentry->d_name.len, dentry->d_name.name);
126 err = ocfs2_sync_inode(dentry->d_inode);
130 journal = osb->journal->j_journal;
131 err = journal_force_commit(journal);
136 return (err < 0) ? -EIO : 0;
139 int ocfs2_should_update_atime(struct inode *inode,
140 struct vfsmount *vfsmnt)
143 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
145 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
148 if ((inode->i_flags & S_NOATIME) ||
149 ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
152 if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
153 ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
157 if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
163 int ocfs2_update_inode_atime(struct inode *inode,
164 struct buffer_head *bh)
167 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
172 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
173 if (handle == NULL) {
179 inode->i_atime = CURRENT_TIME;
180 ret = ocfs2_mark_inode_dirty(handle, inode, bh);
184 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
190 int ocfs2_set_inode_size(handle_t *handle,
192 struct buffer_head *fe_bh,
198 i_size_write(inode, new_i_size);
199 inode->i_blocks = ocfs2_align_bytes_to_sectors(new_i_size);
200 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
202 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
213 static int ocfs2_simple_size_update(struct inode *inode,
214 struct buffer_head *di_bh,
218 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
219 handle_t *handle = NULL;
221 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
222 if (handle == NULL) {
228 ret = ocfs2_set_inode_size(handle, inode, di_bh,
233 ocfs2_commit_trans(osb, handle);
238 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
240 struct buffer_head *fe_bh,
248 /* TODO: This needs to actually orphan the inode in this
251 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
252 if (IS_ERR(handle)) {
253 status = PTR_ERR(handle);
258 status = ocfs2_set_inode_size(handle, inode, fe_bh, new_i_size);
262 ocfs2_commit_trans(osb, handle);
268 static int ocfs2_truncate_file(struct inode *inode,
269 struct buffer_head *di_bh,
273 struct ocfs2_dinode *fe = NULL;
274 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
275 struct ocfs2_truncate_context *tc = NULL;
277 mlog_entry("(inode = %llu, new_i_size = %llu\n",
278 (unsigned long long)OCFS2_I(inode)->ip_blkno,
279 (unsigned long long)new_i_size);
281 truncate_inode_pages(inode->i_mapping, new_i_size);
283 fe = (struct ocfs2_dinode *) di_bh->b_data;
284 if (!OCFS2_IS_VALID_DINODE(fe)) {
285 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
290 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
291 "Inode %llu, inode i_size = %lld != di "
292 "i_size = %llu, i_flags = 0x%x\n",
293 (unsigned long long)OCFS2_I(inode)->ip_blkno,
295 (unsigned long long)le64_to_cpu(fe->i_size),
296 le32_to_cpu(fe->i_flags));
298 if (new_i_size > le64_to_cpu(fe->i_size)) {
299 mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
300 (unsigned long long)le64_to_cpu(fe->i_size),
301 (unsigned long long)new_i_size);
307 mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
308 (unsigned long long)le64_to_cpu(fe->i_blkno),
309 (unsigned long long)le64_to_cpu(fe->i_size),
310 (unsigned long long)new_i_size);
312 /* lets handle the simple truncate cases before doing any more
313 * cluster locking. */
314 if (new_i_size == le64_to_cpu(fe->i_size))
317 /* This forces other nodes to sync and drop their pages. Do
318 * this even if we have a truncate without allocation change -
319 * ocfs2 cluster sizes can be much greater than page size, so
320 * we have to truncate them anyway. */
321 status = ocfs2_data_lock(inode, 1);
326 ocfs2_data_unlock(inode, 1);
328 if (le32_to_cpu(fe->i_clusters) ==
329 ocfs2_clusters_for_bytes(osb->sb, new_i_size)) {
330 mlog(0, "fe->i_clusters = %u, so we do a simple truncate\n",
332 /* No allocation change is required, so lets fast path
334 status = ocfs2_simple_size_update(inode, di_bh, new_i_size);
340 /* alright, we're going to need to do a full blown alloc size
341 * change. Orphan the inode so that recovery can complete the
342 * truncate if necessary. This does the task of marking
344 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
350 status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc);
356 status = ocfs2_commit_truncate(osb, inode, di_bh, tc);
362 /* TODO: orphan dir cleanup here. */
370 * extend allocation only here.
371 * we'll update all the disk stuff, and oip->alloc_size
373 * expect stuff to be locked, a transaction started and enough data /
374 * metadata reservations in the contexts.
376 * Will return -EAGAIN, and a reason if a restart is needed.
377 * If passed in, *reason will always be set, even in error.
379 int ocfs2_do_extend_allocation(struct ocfs2_super *osb,
382 struct buffer_head *fe_bh,
384 struct ocfs2_alloc_context *data_ac,
385 struct ocfs2_alloc_context *meta_ac,
386 enum ocfs2_alloc_restarted *reason_ret)
390 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
391 enum ocfs2_alloc_restarted reason = RESTART_NONE;
392 u32 bit_off, num_bits;
395 BUG_ON(!clusters_to_add);
397 free_extents = ocfs2_num_free_extents(osb, inode, fe);
398 if (free_extents < 0) {
399 status = free_extents;
404 /* there are two cases which could cause us to EAGAIN in the
405 * we-need-more-metadata case:
406 * 1) we haven't reserved *any*
407 * 2) we are so fragmented, we've needed to add metadata too
409 if (!free_extents && !meta_ac) {
410 mlog(0, "we haven't reserved any metadata!\n");
412 reason = RESTART_META;
414 } else if ((!free_extents)
415 && (ocfs2_alloc_context_bits_left(meta_ac)
416 < ocfs2_extend_meta_needed(fe))) {
417 mlog(0, "filesystem is really fragmented...\n");
419 reason = RESTART_META;
423 status = ocfs2_claim_clusters(osb, handle, data_ac, 1,
424 &bit_off, &num_bits);
426 if (status != -ENOSPC)
431 BUG_ON(num_bits > clusters_to_add);
433 /* reserve our write early -- insert_extent may update the inode */
434 status = ocfs2_journal_access(handle, inode, fe_bh,
435 OCFS2_JOURNAL_ACCESS_WRITE);
441 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
442 mlog(0, "Allocating %u clusters at block %u for inode %llu\n",
443 num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
444 status = ocfs2_insert_extent(osb, handle, inode, fe_bh, block,
451 le32_add_cpu(&fe->i_clusters, num_bits);
452 spin_lock(&OCFS2_I(inode)->ip_lock);
453 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
454 spin_unlock(&OCFS2_I(inode)->ip_lock);
456 status = ocfs2_journal_dirty(handle, fe_bh);
462 clusters_to_add -= num_bits;
464 if (clusters_to_add) {
465 mlog(0, "need to alloc once more, clusters = %u, wanted = "
466 "%u\n", fe->i_clusters, clusters_to_add);
468 reason = RESTART_TRANS;
474 *reason_ret = reason;
478 static int ocfs2_extend_allocation(struct inode *inode,
482 int restart_func = 0;
483 int drop_alloc_sem = 0;
484 int credits, num_free_extents;
486 struct buffer_head *bh = NULL;
487 struct ocfs2_dinode *fe = NULL;
488 handle_t *handle = NULL;
489 struct ocfs2_alloc_context *data_ac = NULL;
490 struct ocfs2_alloc_context *meta_ac = NULL;
491 enum ocfs2_alloc_restarted why;
492 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
494 mlog_entry("(clusters_to_add = %u)\n", clusters_to_add);
496 status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &bh,
497 OCFS2_BH_CACHED, inode);
503 fe = (struct ocfs2_dinode *) bh->b_data;
504 if (!OCFS2_IS_VALID_DINODE(fe)) {
505 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
511 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
513 mlog(0, "extend inode %llu, i_size = %lld, fe->i_clusters = %u, "
514 "clusters_to_add = %u\n",
515 (unsigned long long)OCFS2_I(inode)->ip_blkno, i_size_read(inode),
516 fe->i_clusters, clusters_to_add);
518 num_free_extents = ocfs2_num_free_extents(osb,
521 if (num_free_extents < 0) {
522 status = num_free_extents;
527 if (!num_free_extents) {
528 status = ocfs2_reserve_new_metadata(osb, fe, &meta_ac);
530 if (status != -ENOSPC)
536 status = ocfs2_reserve_clusters(osb, clusters_to_add, &data_ac);
538 if (status != -ENOSPC)
543 /* blocks peope in read/write from reading our allocation
544 * until we're done changing it. We depend on i_mutex to block
545 * other extend/truncate calls while we're here. Ordering wrt
546 * start_trans is important here -- always do it before! */
547 down_write(&OCFS2_I(inode)->ip_alloc_sem);
550 credits = ocfs2_calc_extend_credits(osb->sb, fe, clusters_to_add);
551 handle = ocfs2_start_trans(osb, credits);
552 if (IS_ERR(handle)) {
553 status = PTR_ERR(handle);
559 restarted_transaction:
560 /* reserve a write to the file entry early on - that we if we
561 * run out of credits in the allocation path, we can still
563 status = ocfs2_journal_access(handle, inode, bh,
564 OCFS2_JOURNAL_ACCESS_WRITE);
570 prev_clusters = OCFS2_I(inode)->ip_clusters;
572 status = ocfs2_do_extend_allocation(osb,
580 if ((status < 0) && (status != -EAGAIN)) {
581 if (status != -ENOSPC)
586 status = ocfs2_journal_dirty(handle, bh);
592 spin_lock(&OCFS2_I(inode)->ip_lock);
593 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
594 spin_unlock(&OCFS2_I(inode)->ip_lock);
596 if (why != RESTART_NONE && clusters_to_add) {
597 if (why == RESTART_META) {
598 mlog(0, "restarting function.\n");
601 BUG_ON(why != RESTART_TRANS);
603 mlog(0, "restarting transaction.\n");
604 /* TODO: This can be more intelligent. */
605 credits = ocfs2_calc_extend_credits(osb->sb,
608 status = ocfs2_extend_trans(handle, credits);
610 /* handle still has to be committed at
616 goto restarted_transaction;
620 mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
621 fe->i_clusters, (unsigned long long)fe->i_size);
622 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
623 OCFS2_I(inode)->ip_clusters, i_size_read(inode));
626 if (drop_alloc_sem) {
627 up_write(&OCFS2_I(inode)->ip_alloc_sem);
631 ocfs2_commit_trans(osb, handle);
635 ocfs2_free_alloc_context(data_ac);
639 ocfs2_free_alloc_context(meta_ac);
642 if ((!status) && restart_func) {
655 /* Some parts of this taken from generic_cont_expand, which turned out
656 * to be too fragile to do exactly what we need without us having to
657 * worry about recursive locking in ->prepare_write() and
658 * ->commit_write(). */
659 static int ocfs2_write_zero_page(struct inode *inode,
662 struct address_space *mapping = inode->i_mapping;
666 handle_t *handle = NULL;
669 offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */
670 /* ugh. in prepare/commit_write, if from==to==start of block, we
671 ** skip the prepare. make sure we never send an offset for the start
674 if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) {
677 index = size >> PAGE_CACHE_SHIFT;
679 page = grab_cache_page(mapping, index);
686 ret = ocfs2_prepare_write_nolock(inode, page, offset, offset);
692 if (ocfs2_should_order_data(inode)) {
693 handle = ocfs2_start_walk_page_trans(inode, page, offset,
695 if (IS_ERR(handle)) {
696 ret = PTR_ERR(handle);
702 /* must not update i_size! */
703 ret = block_commit_write(page, offset, offset);
710 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
713 page_cache_release(page);
718 static int ocfs2_zero_extend(struct inode *inode,
723 struct super_block *sb = inode->i_sb;
725 start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
726 while (start_off < zero_to_size) {
727 ret = ocfs2_write_zero_page(inode, start_off);
733 start_off += sb->s_blocksize;
736 * Very large extends have the potential to lock up
737 * the cpu for extended periods of time.
747 * A tail_to_skip value > 0 indicates that we're being called from
748 * ocfs2_file_aio_write(). This has the following implications:
750 * - we don't want to update i_size
751 * - di_bh will be NULL, which is fine because it's only used in the
752 * case where we want to update i_size.
753 * - ocfs2_zero_extend() will then only be filling the hole created
754 * between i_size and the start of the write.
756 static int ocfs2_extend_file(struct inode *inode,
757 struct buffer_head *di_bh,
764 BUG_ON(!tail_to_skip && !di_bh);
766 /* setattr sometimes calls us like this. */
770 if (i_size_read(inode) == new_i_size)
772 BUG_ON(new_i_size < i_size_read(inode));
774 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size) -
775 OCFS2_I(inode)->ip_clusters;
778 * protect the pages that ocfs2_zero_extend is going to be
779 * pulling into the page cache.. we do this before the
780 * metadata extend so that we don't get into the situation
781 * where we've extended the metadata but can't get the data
784 ret = ocfs2_data_lock(inode, 1);
790 if (clusters_to_add) {
791 ret = ocfs2_extend_allocation(inode, clusters_to_add);
799 * Call this even if we don't add any clusters to the tree. We
800 * still need to zero the area between the old i_size and the
803 ret = ocfs2_zero_extend(inode, (u64)new_i_size - tail_to_skip);
810 /* We're being called from ocfs2_setattr() which wants
811 * us to update i_size */
812 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
818 ocfs2_data_unlock(inode, 1);
824 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
826 int status = 0, size_change;
827 struct inode *inode = dentry->d_inode;
828 struct super_block *sb = inode->i_sb;
829 struct ocfs2_super *osb = OCFS2_SB(sb);
830 struct buffer_head *bh = NULL;
831 handle_t *handle = NULL;
833 mlog_entry("(0x%p, '%.*s')\n", dentry,
834 dentry->d_name.len, dentry->d_name.name);
836 if (attr->ia_valid & ATTR_MODE)
837 mlog(0, "mode change: %d\n", attr->ia_mode);
838 if (attr->ia_valid & ATTR_UID)
839 mlog(0, "uid change: %d\n", attr->ia_uid);
840 if (attr->ia_valid & ATTR_GID)
841 mlog(0, "gid change: %d\n", attr->ia_gid);
842 if (attr->ia_valid & ATTR_SIZE)
843 mlog(0, "size change...\n");
844 if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME))
845 mlog(0, "time change...\n");
847 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
848 | ATTR_GID | ATTR_UID | ATTR_MODE)
849 if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) {
850 mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid);
854 status = inode_change_ok(inode, attr);
858 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
860 status = ocfs2_rw_lock(inode, 1);
867 status = ocfs2_meta_lock(inode, &bh, 1);
869 if (status != -ENOENT)
874 if (size_change && attr->ia_size != i_size_read(inode)) {
875 if (i_size_read(inode) > attr->ia_size)
876 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
878 status = ocfs2_extend_file(inode, bh, attr->ia_size, 0);
880 if (status != -ENOSPC)
887 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
888 if (IS_ERR(handle)) {
889 status = PTR_ERR(handle);
894 status = inode_setattr(inode, attr);
900 status = ocfs2_mark_inode_dirty(handle, inode, bh);
905 ocfs2_commit_trans(osb, handle);
907 ocfs2_meta_unlock(inode, 1);
910 ocfs2_rw_unlock(inode, 1);
919 int ocfs2_getattr(struct vfsmount *mnt,
920 struct dentry *dentry,
923 struct inode *inode = dentry->d_inode;
924 struct super_block *sb = dentry->d_inode->i_sb;
925 struct ocfs2_super *osb = sb->s_fs_info;
930 err = ocfs2_inode_revalidate(dentry);
937 generic_fillattr(inode, stat);
939 /* We set the blksize from the cluster size for performance */
940 stat->blksize = osb->s_clustersize;
948 int ocfs2_permission(struct inode *inode, int mask, struct nameidata *nd)
954 ret = ocfs2_meta_lock(inode, NULL, 0);
960 ret = generic_permission(inode, mask, NULL);
964 ocfs2_meta_unlock(inode, 0);
970 static int ocfs2_write_remove_suid(struct inode *inode)
973 struct buffer_head *bh = NULL;
974 struct ocfs2_inode_info *oi = OCFS2_I(inode);
976 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
977 struct ocfs2_dinode *di;
979 mlog_entry("(Inode %llu, mode 0%o)\n",
980 (unsigned long long)oi->ip_blkno, inode->i_mode);
982 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
983 if (handle == NULL) {
989 ret = ocfs2_read_block(osb, oi->ip_blkno, &bh, OCFS2_BH_CACHED, inode);
995 ret = ocfs2_journal_access(handle, inode, bh,
996 OCFS2_JOURNAL_ACCESS_WRITE);
1002 inode->i_mode &= ~S_ISUID;
1003 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1004 inode->i_mode &= ~S_ISGID;
1006 di = (struct ocfs2_dinode *) bh->b_data;
1007 di->i_mode = cpu_to_le16(inode->i_mode);
1009 ret = ocfs2_journal_dirty(handle, bh);
1015 ocfs2_commit_trans(osb, handle);
1021 static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
1026 int ret = 0, meta_level = appending;
1027 struct inode *inode = dentry->d_inode;
1029 loff_t newsize, saved_pos;
1032 * We sample i_size under a read level meta lock to see if our write
1033 * is extending the file, if it is we back off and get a write level
1037 ret = ocfs2_meta_lock(inode, NULL, meta_level);
1044 /* Clear suid / sgid if necessary. We do this here
1045 * instead of later in the write path because
1046 * remove_suid() calls ->setattr without any hint that
1047 * we may have already done our cluster locking. Since
1048 * ocfs2_setattr() *must* take cluster locks to
1049 * proceeed, this will lead us to recursively lock the
1050 * inode. There's also the dinode i_size state which
1051 * can be lost via setattr during extending writes (we
1052 * set inode->i_size at the end of a write. */
1053 if (should_remove_suid(dentry)) {
1054 if (meta_level == 0) {
1055 ocfs2_meta_unlock(inode, meta_level);
1060 ret = ocfs2_write_remove_suid(inode);
1067 /* work on a copy of ppos until we're sure that we won't have
1068 * to recalculate it due to relocking. */
1070 saved_pos = i_size_read(inode);
1071 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos);
1075 newsize = count + saved_pos;
1077 mlog(0, "pos=%lld newsize=%lld cursize=%lld\n",
1078 (long long) saved_pos, (long long) newsize,
1079 (long long) i_size_read(inode));
1081 /* No need for a higher level metadata lock if we're
1082 * never going past i_size. */
1083 if (newsize <= i_size_read(inode))
1086 if (meta_level == 0) {
1087 ocfs2_meta_unlock(inode, meta_level);
1092 spin_lock(&OCFS2_I(inode)->ip_lock);
1093 clusters = ocfs2_clusters_for_bytes(inode->i_sb, newsize) -
1094 OCFS2_I(inode)->ip_clusters;
1095 spin_unlock(&OCFS2_I(inode)->ip_lock);
1097 mlog(0, "Writing at EOF, may need more allocation: "
1098 "i_size = %lld, newsize = %lld, need %u clusters\n",
1099 (long long) i_size_read(inode), (long long) newsize,
1102 /* We only want to continue the rest of this loop if
1103 * our extend will actually require more
1108 ret = ocfs2_extend_file(inode, NULL, newsize, count);
1121 ocfs2_meta_unlock(inode, meta_level);
1127 static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
1128 const struct iovec *iov,
1129 unsigned long nr_segs,
1132 int ret, rw_level, have_alloc_sem = 0;
1133 struct file *filp = iocb->ki_filp;
1134 struct inode *inode = filp->f_dentry->d_inode;
1135 int appending = filp->f_flags & O_APPEND ? 1 : 0;
1137 mlog_entry("(0x%p, %u, '%.*s')\n", filp,
1138 (unsigned int)nr_segs,
1139 filp->f_dentry->d_name.len,
1140 filp->f_dentry->d_name.name);
1142 /* happy write of zero bytes */
1143 if (iocb->ki_left == 0)
1146 mutex_lock(&inode->i_mutex);
1147 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
1148 if (filp->f_flags & O_DIRECT) {
1150 down_read(&inode->i_alloc_sem);
1153 /* concurrent O_DIRECT writes are allowed */
1154 rw_level = (filp->f_flags & O_DIRECT) ? 0 : 1;
1155 ret = ocfs2_rw_lock(inode, rw_level);
1162 ret = ocfs2_prepare_inode_for_write(filp->f_dentry, &iocb->ki_pos,
1163 iocb->ki_left, appending);
1169 /* communicate with ocfs2_dio_end_io */
1170 ocfs2_iocb_set_rw_locked(iocb);
1172 ret = generic_file_aio_write_nolock(iocb, iov, nr_segs, iocb->ki_pos);
1174 /* buffered aio wouldn't have proper lock coverage today */
1175 BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
1178 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
1179 * function pointer which is called when o_direct io completes so that
1180 * it can unlock our rw lock. (it's the clustered equivalent of
1181 * i_alloc_sem; protects truncate from racing with pending ios).
1182 * Unfortunately there are error cases which call end_io and others
1183 * that don't. so we don't have to unlock the rw_lock if either an
1184 * async dio is going to do it in the future or an end_io after an
1185 * error has already done it.
1187 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
1194 up_read(&inode->i_alloc_sem);
1196 ocfs2_rw_unlock(inode, rw_level);
1197 mutex_unlock(&inode->i_mutex);
1203 static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
1210 struct inode *inode = out->f_dentry->d_inode;
1212 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe,
1214 out->f_dentry->d_name.len,
1215 out->f_dentry->d_name.name);
1217 inode_double_lock(inode, pipe->inode);
1219 ret = ocfs2_rw_lock(inode, 1);
1225 ret = ocfs2_prepare_inode_for_write(out->f_dentry, ppos, len, 0);
1231 /* ok, we're done with i_size and alloc work */
1232 ret = generic_file_splice_write_nolock(pipe, out, ppos, len, flags);
1235 ocfs2_rw_unlock(inode, 1);
1237 inode_double_unlock(inode, pipe->inode);
1243 static ssize_t ocfs2_file_splice_read(struct file *in,
1245 struct pipe_inode_info *pipe,
1250 struct inode *inode = in->f_dentry->d_inode;
1252 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe,
1254 in->f_dentry->d_name.len,
1255 in->f_dentry->d_name.name);
1258 * See the comment in ocfs2_file_aio_read()
1260 ret = ocfs2_meta_lock(inode, NULL, 0);
1265 ocfs2_meta_unlock(inode, 0);
1267 ret = generic_file_splice_read(in, ppos, pipe, len, flags);
1274 static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
1275 const struct iovec *iov,
1276 unsigned long nr_segs,
1279 int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
1280 struct file *filp = iocb->ki_filp;
1281 struct inode *inode = filp->f_dentry->d_inode;
1283 mlog_entry("(0x%p, %u, '%.*s')\n", filp,
1284 (unsigned int)nr_segs,
1285 filp->f_dentry->d_name.len,
1286 filp->f_dentry->d_name.name);
1295 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
1296 * need locks to protect pending reads from racing with truncate.
1298 if (filp->f_flags & O_DIRECT) {
1299 down_read(&inode->i_alloc_sem);
1302 ret = ocfs2_rw_lock(inode, 0);
1308 /* communicate with ocfs2_dio_end_io */
1309 ocfs2_iocb_set_rw_locked(iocb);
1313 * We're fine letting folks race truncates and extending
1314 * writes with read across the cluster, just like they can
1315 * locally. Hence no rw_lock during read.
1317 * Take and drop the meta data lock to update inode fields
1318 * like i_size. This allows the checks down below
1319 * generic_file_aio_read() a chance of actually working.
1321 ret = ocfs2_meta_lock_atime(inode, filp->f_vfsmnt, &lock_level);
1326 ocfs2_meta_unlock(inode, lock_level);
1328 ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
1330 mlog(ML_ERROR, "generic_file_aio_read returned -EINVAL\n");
1332 /* buffered aio wouldn't have proper lock coverage today */
1333 BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
1335 /* see ocfs2_file_aio_write */
1336 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
1343 up_read(&inode->i_alloc_sem);
1345 ocfs2_rw_unlock(inode, rw_level);
1351 struct inode_operations ocfs2_file_iops = {
1352 .setattr = ocfs2_setattr,
1353 .getattr = ocfs2_getattr,
1354 .permission = ocfs2_permission,
1357 struct inode_operations ocfs2_special_file_iops = {
1358 .setattr = ocfs2_setattr,
1359 .getattr = ocfs2_getattr,
1360 .permission = ocfs2_permission,
1363 const struct file_operations ocfs2_fops = {
1364 .read = do_sync_read,
1365 .write = do_sync_write,
1366 .sendfile = generic_file_sendfile,
1368 .fsync = ocfs2_sync_file,
1369 .release = ocfs2_file_release,
1370 .open = ocfs2_file_open,
1371 .aio_read = ocfs2_file_aio_read,
1372 .aio_write = ocfs2_file_aio_write,
1373 .ioctl = ocfs2_ioctl,
1374 .splice_read = ocfs2_file_splice_read,
1375 .splice_write = ocfs2_file_splice_write,
1378 const struct file_operations ocfs2_dops = {
1379 .read = generic_read_dir,
1380 .readdir = ocfs2_readdir,
1381 .fsync = ocfs2_sync_file,
1382 .ioctl = ocfs2_ioctl,