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>
38 #include <linux/quotaops.h>
40 #define MLOG_MASK_PREFIX ML_INODE
41 #include <cluster/masklog.h>
49 #include "extent_map.h"
63 #include "buffer_head_io.h"
65 static int ocfs2_sync_inode(struct inode *inode)
67 filemap_fdatawrite(inode->i_mapping);
68 return sync_mapping_buffers(inode->i_mapping);
71 static int ocfs2_init_file_private(struct inode *inode, struct file *file)
73 struct ocfs2_file_private *fp;
75 fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
80 mutex_init(&fp->fp_mutex);
81 ocfs2_file_lock_res_init(&fp->fp_flock, fp);
82 file->private_data = fp;
87 static void ocfs2_free_file_private(struct inode *inode, struct file *file)
89 struct ocfs2_file_private *fp = file->private_data;
90 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
93 ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
94 ocfs2_lock_res_free(&fp->fp_flock);
96 file->private_data = NULL;
100 static int ocfs2_file_open(struct inode *inode, struct file *file)
103 int mode = file->f_flags;
104 struct ocfs2_inode_info *oi = OCFS2_I(inode);
106 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
107 file->f_path.dentry->d_name.len, file->f_path.dentry->d_name.name);
109 spin_lock(&oi->ip_lock);
111 /* Check that the inode hasn't been wiped from disk by another
112 * node. If it hasn't then we're safe as long as we hold the
113 * spin lock until our increment of open count. */
114 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
115 spin_unlock(&oi->ip_lock);
122 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
125 spin_unlock(&oi->ip_lock);
127 status = ocfs2_init_file_private(inode, file);
130 * We want to set open count back if we're failing the
133 spin_lock(&oi->ip_lock);
135 spin_unlock(&oi->ip_lock);
143 static int ocfs2_file_release(struct inode *inode, struct file *file)
145 struct ocfs2_inode_info *oi = OCFS2_I(inode);
147 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
148 file->f_path.dentry->d_name.len,
149 file->f_path.dentry->d_name.name);
151 spin_lock(&oi->ip_lock);
152 if (!--oi->ip_open_count)
153 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
154 spin_unlock(&oi->ip_lock);
156 ocfs2_free_file_private(inode, file);
163 static int ocfs2_dir_open(struct inode *inode, struct file *file)
165 return ocfs2_init_file_private(inode, file);
168 static int ocfs2_dir_release(struct inode *inode, struct file *file)
170 ocfs2_free_file_private(inode, file);
174 static int ocfs2_sync_file(struct file *file,
175 struct dentry *dentry,
180 struct inode *inode = dentry->d_inode;
181 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
183 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync,
184 dentry->d_name.len, dentry->d_name.name);
186 err = ocfs2_sync_inode(dentry->d_inode);
190 journal = osb->journal->j_journal;
191 err = jbd2_journal_force_commit(journal);
196 return (err < 0) ? -EIO : 0;
199 int ocfs2_should_update_atime(struct inode *inode,
200 struct vfsmount *vfsmnt)
203 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
205 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
208 if ((inode->i_flags & S_NOATIME) ||
209 ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
213 * We can be called with no vfsmnt structure - NFSD will
216 * Note that our action here is different than touch_atime() -
217 * if we can't tell whether this is a noatime mount, then we
218 * don't know whether to trust the value of s_atime_quantum.
223 if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
224 ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
227 if (vfsmnt->mnt_flags & MNT_RELATIME) {
228 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
229 (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
236 if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
242 int ocfs2_update_inode_atime(struct inode *inode,
243 struct buffer_head *bh)
246 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
248 struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
252 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
253 if (IS_ERR(handle)) {
254 ret = PTR_ERR(handle);
259 ret = ocfs2_journal_access_di(handle, inode, bh,
260 OCFS2_JOURNAL_ACCESS_WRITE);
267 * Don't use ocfs2_mark_inode_dirty() here as we don't always
268 * have i_mutex to guard against concurrent changes to other
271 inode->i_atime = CURRENT_TIME;
272 di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
273 di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
275 ret = ocfs2_journal_dirty(handle, bh);
280 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
286 static int ocfs2_set_inode_size(handle_t *handle,
288 struct buffer_head *fe_bh,
294 i_size_write(inode, new_i_size);
295 inode->i_blocks = ocfs2_inode_sector_count(inode);
296 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
298 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
309 int ocfs2_simple_size_update(struct inode *inode,
310 struct buffer_head *di_bh,
314 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
315 handle_t *handle = NULL;
317 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
318 if (IS_ERR(handle)) {
319 ret = PTR_ERR(handle);
324 ret = ocfs2_set_inode_size(handle, inode, di_bh,
329 ocfs2_commit_trans(osb, handle);
334 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
336 struct buffer_head *fe_bh,
341 struct ocfs2_dinode *di;
346 /* TODO: This needs to actually orphan the inode in this
349 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
350 if (IS_ERR(handle)) {
351 status = PTR_ERR(handle);
356 status = ocfs2_journal_access_di(handle, inode, fe_bh,
357 OCFS2_JOURNAL_ACCESS_WRITE);
364 * Do this before setting i_size.
366 cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
367 status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
374 i_size_write(inode, new_i_size);
375 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
377 di = (struct ocfs2_dinode *) fe_bh->b_data;
378 di->i_size = cpu_to_le64(new_i_size);
379 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
380 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
382 status = ocfs2_journal_dirty(handle, fe_bh);
387 ocfs2_commit_trans(osb, handle);
394 static int ocfs2_truncate_file(struct inode *inode,
395 struct buffer_head *di_bh,
399 struct ocfs2_dinode *fe = NULL;
400 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
401 struct ocfs2_truncate_context *tc = NULL;
403 mlog_entry("(inode = %llu, new_i_size = %llu\n",
404 (unsigned long long)OCFS2_I(inode)->ip_blkno,
405 (unsigned long long)new_i_size);
407 /* We trust di_bh because it comes from ocfs2_inode_lock(), which
408 * already validated it */
409 fe = (struct ocfs2_dinode *) di_bh->b_data;
411 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
412 "Inode %llu, inode i_size = %lld != di "
413 "i_size = %llu, i_flags = 0x%x\n",
414 (unsigned long long)OCFS2_I(inode)->ip_blkno,
416 (unsigned long long)le64_to_cpu(fe->i_size),
417 le32_to_cpu(fe->i_flags));
419 if (new_i_size > le64_to_cpu(fe->i_size)) {
420 mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
421 (unsigned long long)le64_to_cpu(fe->i_size),
422 (unsigned long long)new_i_size);
428 mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
429 (unsigned long long)le64_to_cpu(fe->i_blkno),
430 (unsigned long long)le64_to_cpu(fe->i_size),
431 (unsigned long long)new_i_size);
433 /* lets handle the simple truncate cases before doing any more
434 * cluster locking. */
435 if (new_i_size == le64_to_cpu(fe->i_size))
438 down_write(&OCFS2_I(inode)->ip_alloc_sem);
441 * The inode lock forced other nodes to sync and drop their
442 * pages, which (correctly) happens even if we have a truncate
443 * without allocation change - ocfs2 cluster sizes can be much
444 * greater than page size, so we have to truncate them
447 unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
448 truncate_inode_pages(inode->i_mapping, new_i_size);
450 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
451 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
452 i_size_read(inode), 1);
456 goto bail_unlock_sem;
459 /* alright, we're going to need to do a full blown alloc size
460 * change. Orphan the inode so that recovery can complete the
461 * truncate if necessary. This does the task of marking
463 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
466 goto bail_unlock_sem;
469 status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc);
472 goto bail_unlock_sem;
475 status = ocfs2_commit_truncate(osb, inode, di_bh, tc);
478 goto bail_unlock_sem;
481 /* TODO: orphan dir cleanup here. */
483 up_write(&OCFS2_I(inode)->ip_alloc_sem);
492 * extend file allocation only here.
493 * we'll update all the disk stuff, and oip->alloc_size
495 * expect stuff to be locked, a transaction started and enough data /
496 * metadata reservations in the contexts.
498 * Will return -EAGAIN, and a reason if a restart is needed.
499 * If passed in, *reason will always be set, even in error.
501 int ocfs2_add_inode_data(struct ocfs2_super *osb,
506 struct buffer_head *fe_bh,
508 struct ocfs2_alloc_context *data_ac,
509 struct ocfs2_alloc_context *meta_ac,
510 enum ocfs2_alloc_restarted *reason_ret)
513 struct ocfs2_extent_tree et;
515 ocfs2_init_dinode_extent_tree(&et, inode, fe_bh);
516 ret = ocfs2_add_clusters_in_btree(osb, inode, logical_offset,
517 clusters_to_add, mark_unwritten,
519 data_ac, meta_ac, reason_ret);
524 static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
525 u32 clusters_to_add, int mark_unwritten)
528 int restart_func = 0;
531 struct buffer_head *bh = NULL;
532 struct ocfs2_dinode *fe = NULL;
533 handle_t *handle = NULL;
534 struct ocfs2_alloc_context *data_ac = NULL;
535 struct ocfs2_alloc_context *meta_ac = NULL;
536 enum ocfs2_alloc_restarted why;
537 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
538 struct ocfs2_extent_tree et;
541 mlog_entry("(clusters_to_add = %u)\n", clusters_to_add);
544 * This function only exists for file systems which don't
547 BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
549 status = ocfs2_read_inode_block(inode, &bh);
554 fe = (struct ocfs2_dinode *) bh->b_data;
557 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
559 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, "
560 "clusters_to_add = %u\n",
561 (unsigned long long)OCFS2_I(inode)->ip_blkno,
562 (long long)i_size_read(inode), le32_to_cpu(fe->i_clusters),
564 ocfs2_init_dinode_extent_tree(&et, inode, bh);
565 status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
572 credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list,
574 handle = ocfs2_start_trans(osb, credits);
575 if (IS_ERR(handle)) {
576 status = PTR_ERR(handle);
582 restarted_transaction:
583 if (vfs_dq_alloc_space_nodirty(inode, ocfs2_clusters_to_bytes(osb->sb,
590 /* reserve a write to the file entry early on - that we if we
591 * run out of credits in the allocation path, we can still
593 status = ocfs2_journal_access_di(handle, inode, bh,
594 OCFS2_JOURNAL_ACCESS_WRITE);
600 prev_clusters = OCFS2_I(inode)->ip_clusters;
602 status = ocfs2_add_inode_data(osb,
612 if ((status < 0) && (status != -EAGAIN)) {
613 if (status != -ENOSPC)
618 status = ocfs2_journal_dirty(handle, bh);
624 spin_lock(&OCFS2_I(inode)->ip_lock);
625 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
626 spin_unlock(&OCFS2_I(inode)->ip_lock);
627 /* Release unused quota reservation */
628 vfs_dq_free_space(inode,
629 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
632 if (why != RESTART_NONE && clusters_to_add) {
633 if (why == RESTART_META) {
634 mlog(0, "restarting function.\n");
637 BUG_ON(why != RESTART_TRANS);
639 mlog(0, "restarting transaction.\n");
640 /* TODO: This can be more intelligent. */
641 credits = ocfs2_calc_extend_credits(osb->sb,
644 status = ocfs2_extend_trans(handle, credits);
646 /* handle still has to be committed at
652 goto restarted_transaction;
656 mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
657 le32_to_cpu(fe->i_clusters),
658 (unsigned long long)le64_to_cpu(fe->i_size));
659 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
660 OCFS2_I(inode)->ip_clusters, (long long)i_size_read(inode));
663 if (status < 0 && did_quota)
664 vfs_dq_free_space(inode,
665 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
667 ocfs2_commit_trans(osb, handle);
671 ocfs2_free_alloc_context(data_ac);
675 ocfs2_free_alloc_context(meta_ac);
678 if ((!status) && restart_func) {
689 /* Some parts of this taken from generic_cont_expand, which turned out
690 * to be too fragile to do exactly what we need without us having to
691 * worry about recursive locking in ->write_begin() and ->write_end(). */
692 static int ocfs2_write_zero_page(struct inode *inode,
695 struct address_space *mapping = inode->i_mapping;
699 handle_t *handle = NULL;
702 offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */
703 /* ugh. in prepare/commit_write, if from==to==start of block, we
704 ** skip the prepare. make sure we never send an offset for the start
707 if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) {
710 index = size >> PAGE_CACHE_SHIFT;
712 page = grab_cache_page(mapping, index);
719 ret = ocfs2_prepare_write_nolock(inode, page, offset, offset);
725 if (ocfs2_should_order_data(inode)) {
726 handle = ocfs2_start_walk_page_trans(inode, page, offset,
728 if (IS_ERR(handle)) {
729 ret = PTR_ERR(handle);
735 /* must not update i_size! */
736 ret = block_commit_write(page, offset, offset);
743 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
746 page_cache_release(page);
751 static int ocfs2_zero_extend(struct inode *inode,
756 struct super_block *sb = inode->i_sb;
758 start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
759 while (start_off < zero_to_size) {
760 ret = ocfs2_write_zero_page(inode, start_off);
766 start_off += sb->s_blocksize;
769 * Very large extends have the potential to lock up
770 * the cpu for extended periods of time.
779 int ocfs2_extend_no_holes(struct inode *inode, u64 new_i_size, u64 zero_to)
783 struct ocfs2_inode_info *oi = OCFS2_I(inode);
785 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
786 if (clusters_to_add < oi->ip_clusters)
789 clusters_to_add -= oi->ip_clusters;
791 if (clusters_to_add) {
792 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
801 * Call this even if we don't add any clusters to the tree. We
802 * still need to zero the area between the old i_size and the
805 ret = ocfs2_zero_extend(inode, zero_to);
813 static int ocfs2_extend_file(struct inode *inode,
814 struct buffer_head *di_bh,
818 struct ocfs2_inode_info *oi = OCFS2_I(inode);
822 /* setattr sometimes calls us like this. */
826 if (i_size_read(inode) == new_i_size)
828 BUG_ON(new_i_size < i_size_read(inode));
831 * Fall through for converting inline data, even if the fs
832 * supports sparse files.
834 * The check for inline data here is legal - nobody can add
835 * the feature since we have i_mutex. We must check it again
836 * after acquiring ip_alloc_sem though, as paths like mmap
837 * might have raced us to converting the inode to extents.
839 if (!(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
840 && ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
841 goto out_update_size;
844 * The alloc sem blocks people in read/write from reading our
845 * allocation until we're done changing it. We depend on
846 * i_mutex to block other extend/truncate calls while we're
849 down_write(&oi->ip_alloc_sem);
851 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
853 * We can optimize small extends by keeping the inodes
856 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
857 up_write(&oi->ip_alloc_sem);
858 goto out_update_size;
861 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
863 up_write(&oi->ip_alloc_sem);
870 if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
871 ret = ocfs2_extend_no_holes(inode, new_i_size, new_i_size);
873 up_write(&oi->ip_alloc_sem);
881 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
889 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
891 int status = 0, size_change;
892 struct inode *inode = dentry->d_inode;
893 struct super_block *sb = inode->i_sb;
894 struct ocfs2_super *osb = OCFS2_SB(sb);
895 struct buffer_head *bh = NULL;
896 handle_t *handle = NULL;
897 int locked[MAXQUOTAS] = {0, 0};
899 struct ocfs2_mem_dqinfo *oinfo;
901 mlog_entry("(0x%p, '%.*s')\n", dentry,
902 dentry->d_name.len, dentry->d_name.name);
904 /* ensuring we don't even attempt to truncate a symlink */
905 if (S_ISLNK(inode->i_mode))
906 attr->ia_valid &= ~ATTR_SIZE;
908 if (attr->ia_valid & ATTR_MODE)
909 mlog(0, "mode change: %d\n", attr->ia_mode);
910 if (attr->ia_valid & ATTR_UID)
911 mlog(0, "uid change: %d\n", attr->ia_uid);
912 if (attr->ia_valid & ATTR_GID)
913 mlog(0, "gid change: %d\n", attr->ia_gid);
914 if (attr->ia_valid & ATTR_SIZE)
915 mlog(0, "size change...\n");
916 if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME))
917 mlog(0, "time change...\n");
919 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
920 | ATTR_GID | ATTR_UID | ATTR_MODE)
921 if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) {
922 mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid);
926 status = inode_change_ok(inode, attr);
930 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
932 status = ocfs2_rw_lock(inode, 1);
939 status = ocfs2_inode_lock(inode, &bh, 1);
941 if (status != -ENOENT)
946 if (size_change && attr->ia_size != i_size_read(inode)) {
947 if (attr->ia_size > sb->s_maxbytes) {
952 if (i_size_read(inode) > attr->ia_size) {
953 if (ocfs2_should_order_data(inode)) {
954 status = ocfs2_begin_ordered_truncate(inode,
959 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
961 status = ocfs2_extend_file(inode, bh, attr->ia_size);
963 if (status != -ENOSPC)
970 if ((attr->ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
971 (attr->ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
972 credits = OCFS2_INODE_UPDATE_CREDITS;
973 if (attr->ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid
974 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
975 OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
976 oinfo = sb_dqinfo(sb, USRQUOTA)->dqi_priv;
977 status = ocfs2_lock_global_qf(oinfo, 1);
980 credits += ocfs2_calc_qinit_credits(sb, USRQUOTA) +
981 ocfs2_calc_qdel_credits(sb, USRQUOTA);
982 locked[USRQUOTA] = 1;
984 if (attr->ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid
985 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
986 OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
987 oinfo = sb_dqinfo(sb, GRPQUOTA)->dqi_priv;
988 status = ocfs2_lock_global_qf(oinfo, 1);
991 credits += ocfs2_calc_qinit_credits(sb, GRPQUOTA) +
992 ocfs2_calc_qdel_credits(sb, GRPQUOTA);
993 locked[GRPQUOTA] = 1;
995 handle = ocfs2_start_trans(osb, credits);
996 if (IS_ERR(handle)) {
997 status = PTR_ERR(handle);
1001 status = vfs_dq_transfer(inode, attr) ? -EDQUOT : 0;
1005 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1006 if (IS_ERR(handle)) {
1007 status = PTR_ERR(handle);
1014 * This will intentionally not wind up calling vmtruncate(),
1015 * since all the work for a size change has been done above.
1016 * Otherwise, we could get into problems with truncate as
1017 * ip_alloc_sem is used there to protect against i_size
1020 status = inode_setattr(inode, attr);
1026 status = ocfs2_mark_inode_dirty(handle, inode, bh);
1031 ocfs2_commit_trans(osb, handle);
1033 for (qtype = 0; qtype < MAXQUOTAS; qtype++) {
1036 oinfo = sb_dqinfo(sb, qtype)->dqi_priv;
1037 ocfs2_unlock_global_qf(oinfo, 1);
1039 ocfs2_inode_unlock(inode, 1);
1042 ocfs2_rw_unlock(inode, 1);
1046 if (!status && attr->ia_valid & ATTR_MODE) {
1047 status = ocfs2_acl_chmod(inode);
1056 int ocfs2_getattr(struct vfsmount *mnt,
1057 struct dentry *dentry,
1060 struct inode *inode = dentry->d_inode;
1061 struct super_block *sb = dentry->d_inode->i_sb;
1062 struct ocfs2_super *osb = sb->s_fs_info;
1067 err = ocfs2_inode_revalidate(dentry);
1074 generic_fillattr(inode, stat);
1076 /* We set the blksize from the cluster size for performance */
1077 stat->blksize = osb->s_clustersize;
1085 int ocfs2_permission(struct inode *inode, int mask)
1091 ret = ocfs2_inode_lock(inode, NULL, 0);
1098 ret = generic_permission(inode, mask, ocfs2_check_acl);
1100 ocfs2_inode_unlock(inode, 0);
1106 static int __ocfs2_write_remove_suid(struct inode *inode,
1107 struct buffer_head *bh)
1111 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1112 struct ocfs2_dinode *di;
1114 mlog_entry("(Inode %llu, mode 0%o)\n",
1115 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_mode);
1117 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1118 if (IS_ERR(handle)) {
1119 ret = PTR_ERR(handle);
1124 ret = ocfs2_journal_access_di(handle, inode, bh,
1125 OCFS2_JOURNAL_ACCESS_WRITE);
1131 inode->i_mode &= ~S_ISUID;
1132 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1133 inode->i_mode &= ~S_ISGID;
1135 di = (struct ocfs2_dinode *) bh->b_data;
1136 di->i_mode = cpu_to_le16(inode->i_mode);
1138 ret = ocfs2_journal_dirty(handle, bh);
1143 ocfs2_commit_trans(osb, handle);
1150 * Will look for holes and unwritten extents in the range starting at
1151 * pos for count bytes (inclusive).
1153 static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1157 unsigned int extent_flags;
1158 u32 cpos, clusters, extent_len, phys_cpos;
1159 struct super_block *sb = inode->i_sb;
1161 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1162 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1165 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1172 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
1177 if (extent_len > clusters)
1178 extent_len = clusters;
1180 clusters -= extent_len;
1187 static int ocfs2_write_remove_suid(struct inode *inode)
1190 struct buffer_head *bh = NULL;
1192 ret = ocfs2_read_inode_block(inode, &bh);
1198 ret = __ocfs2_write_remove_suid(inode, bh);
1205 * Allocate enough extents to cover the region starting at byte offset
1206 * start for len bytes. Existing extents are skipped, any extents
1207 * added are marked as "unwritten".
1209 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1213 u32 cpos, phys_cpos, clusters, alloc_size;
1214 u64 end = start + len;
1215 struct buffer_head *di_bh = NULL;
1217 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1218 ret = ocfs2_read_inode_block(inode, &di_bh);
1225 * Nothing to do if the requested reservation range
1226 * fits within the inode.
1228 if (ocfs2_size_fits_inline_data(di_bh, end))
1231 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1239 * We consider both start and len to be inclusive.
1241 cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1242 clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1246 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1254 * Hole or existing extent len can be arbitrary, so
1255 * cap it to our own allocation request.
1257 if (alloc_size > clusters)
1258 alloc_size = clusters;
1262 * We already have an allocation at this
1263 * region so we can safely skip it.
1268 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1277 clusters -= alloc_size;
1288 * Truncate a byte range, avoiding pages within partial clusters. This
1289 * preserves those pages for the zeroing code to write to.
1291 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1294 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1296 struct address_space *mapping = inode->i_mapping;
1298 start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1299 end = byte_start + byte_len;
1300 end = end & ~(osb->s_clustersize - 1);
1303 unmap_mapping_range(mapping, start, end - start, 0);
1304 truncate_inode_pages_range(mapping, start, end - 1);
1308 static int ocfs2_zero_partial_clusters(struct inode *inode,
1312 u64 tmpend, end = start + len;
1313 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1314 unsigned int csize = osb->s_clustersize;
1318 * The "start" and "end" values are NOT necessarily part of
1319 * the range whose allocation is being deleted. Rather, this
1320 * is what the user passed in with the request. We must zero
1321 * partial clusters here. There's no need to worry about
1322 * physical allocation - the zeroing code knows to skip holes.
1324 mlog(0, "byte start: %llu, end: %llu\n",
1325 (unsigned long long)start, (unsigned long long)end);
1328 * If both edges are on a cluster boundary then there's no
1329 * zeroing required as the region is part of the allocation to
1332 if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1335 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1336 if (IS_ERR(handle)) {
1337 ret = PTR_ERR(handle);
1343 * We want to get the byte offset of the end of the 1st cluster.
1345 tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1349 mlog(0, "1st range: start: %llu, tmpend: %llu\n",
1350 (unsigned long long)start, (unsigned long long)tmpend);
1352 ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1358 * This may make start and end equal, but the zeroing
1359 * code will skip any work in that case so there's no
1360 * need to catch it up here.
1362 start = end & ~(osb->s_clustersize - 1);
1364 mlog(0, "2nd range: start: %llu, end: %llu\n",
1365 (unsigned long long)start, (unsigned long long)end);
1367 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1372 ocfs2_commit_trans(osb, handle);
1377 static int ocfs2_remove_inode_range(struct inode *inode,
1378 struct buffer_head *di_bh, u64 byte_start,
1382 u32 trunc_start, trunc_len, cpos, phys_cpos, alloc_size;
1383 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1384 struct ocfs2_cached_dealloc_ctxt dealloc;
1385 struct address_space *mapping = inode->i_mapping;
1386 struct ocfs2_extent_tree et;
1388 ocfs2_init_dinode_extent_tree(&et, inode, di_bh);
1389 ocfs2_init_dealloc_ctxt(&dealloc);
1394 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1395 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1396 byte_start + byte_len, 0);
1402 * There's no need to get fancy with the page cache
1403 * truncate of an inline-data inode. We're talking
1404 * about less than a page here, which will be cached
1405 * in the dinode buffer anyway.
1407 unmap_mapping_range(mapping, 0, 0, 0);
1408 truncate_inode_pages(mapping, 0);
1412 trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1413 trunc_len = (byte_start + byte_len) >> osb->s_clustersize_bits;
1414 if (trunc_len >= trunc_start)
1415 trunc_len -= trunc_start;
1419 mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u\n",
1420 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1421 (unsigned long long)byte_start,
1422 (unsigned long long)byte_len, trunc_start, trunc_len);
1424 ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1432 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1439 if (alloc_size > trunc_len)
1440 alloc_size = trunc_len;
1442 /* Only do work for non-holes */
1443 if (phys_cpos != 0) {
1444 ret = ocfs2_remove_btree_range(inode, &et, cpos,
1445 phys_cpos, alloc_size,
1454 trunc_len -= alloc_size;
1457 ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1460 ocfs2_schedule_truncate_log_flush(osb, 1);
1461 ocfs2_run_deallocs(osb, &dealloc);
1467 * Parts of this function taken from xfs_change_file_space()
1469 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1470 loff_t f_pos, unsigned int cmd,
1471 struct ocfs2_space_resv *sr,
1477 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1478 struct buffer_head *di_bh = NULL;
1480 unsigned long long max_off = inode->i_sb->s_maxbytes;
1482 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1485 mutex_lock(&inode->i_mutex);
1488 * This prevents concurrent writes on other nodes
1490 ret = ocfs2_rw_lock(inode, 1);
1496 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1502 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1504 goto out_inode_unlock;
1507 switch (sr->l_whence) {
1508 case 0: /*SEEK_SET*/
1510 case 1: /*SEEK_CUR*/
1511 sr->l_start += f_pos;
1513 case 2: /*SEEK_END*/
1514 sr->l_start += i_size_read(inode);
1518 goto out_inode_unlock;
1522 llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1525 || sr->l_start > max_off
1526 || (sr->l_start + llen) < 0
1527 || (sr->l_start + llen) > max_off) {
1529 goto out_inode_unlock;
1531 size = sr->l_start + sr->l_len;
1533 if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) {
1534 if (sr->l_len <= 0) {
1536 goto out_inode_unlock;
1540 if (file && should_remove_suid(file->f_path.dentry)) {
1541 ret = __ocfs2_write_remove_suid(inode, di_bh);
1544 goto out_inode_unlock;
1548 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1550 case OCFS2_IOC_RESVSP:
1551 case OCFS2_IOC_RESVSP64:
1553 * This takes unsigned offsets, but the signed ones we
1554 * pass have been checked against overflow above.
1556 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1559 case OCFS2_IOC_UNRESVSP:
1560 case OCFS2_IOC_UNRESVSP64:
1561 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1567 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1570 goto out_inode_unlock;
1574 * We update c/mtime for these changes
1576 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1577 if (IS_ERR(handle)) {
1578 ret = PTR_ERR(handle);
1580 goto out_inode_unlock;
1583 if (change_size && i_size_read(inode) < size)
1584 i_size_write(inode, size);
1586 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1587 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1591 ocfs2_commit_trans(osb, handle);
1595 ocfs2_inode_unlock(inode, 1);
1597 ocfs2_rw_unlock(inode, 1);
1600 mutex_unlock(&inode->i_mutex);
1604 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1605 struct ocfs2_space_resv *sr)
1607 struct inode *inode = file->f_path.dentry->d_inode;
1608 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1610 if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1611 !ocfs2_writes_unwritten_extents(osb))
1613 else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1614 !ocfs2_sparse_alloc(osb))
1617 if (!S_ISREG(inode->i_mode))
1620 if (!(file->f_mode & FMODE_WRITE))
1623 return __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1626 static long ocfs2_fallocate(struct inode *inode, int mode, loff_t offset,
1629 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1630 struct ocfs2_space_resv sr;
1631 int change_size = 1;
1633 if (!ocfs2_writes_unwritten_extents(osb))
1636 if (S_ISDIR(inode->i_mode))
1639 if (mode & FALLOC_FL_KEEP_SIZE)
1643 sr.l_start = (s64)offset;
1644 sr.l_len = (s64)len;
1646 return __ocfs2_change_file_space(NULL, inode, offset,
1647 OCFS2_IOC_RESVSP64, &sr, change_size);
1650 static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
1656 int ret = 0, meta_level = 0;
1657 struct inode *inode = dentry->d_inode;
1658 loff_t saved_pos, end;
1661 * We start with a read level meta lock and only jump to an ex
1662 * if we need to make modifications here.
1665 ret = ocfs2_inode_lock(inode, NULL, meta_level);
1672 /* Clear suid / sgid if necessary. We do this here
1673 * instead of later in the write path because
1674 * remove_suid() calls ->setattr without any hint that
1675 * we may have already done our cluster locking. Since
1676 * ocfs2_setattr() *must* take cluster locks to
1677 * proceeed, this will lead us to recursively lock the
1678 * inode. There's also the dinode i_size state which
1679 * can be lost via setattr during extending writes (we
1680 * set inode->i_size at the end of a write. */
1681 if (should_remove_suid(dentry)) {
1682 if (meta_level == 0) {
1683 ocfs2_inode_unlock(inode, meta_level);
1688 ret = ocfs2_write_remove_suid(inode);
1695 /* work on a copy of ppos until we're sure that we won't have
1696 * to recalculate it due to relocking. */
1698 saved_pos = i_size_read(inode);
1699 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos);
1704 end = saved_pos + count;
1707 * Skip the O_DIRECT checks if we don't need
1710 if (!direct_io || !(*direct_io))
1714 * There's no sane way to do direct writes to an inode
1717 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1723 * Allowing concurrent direct writes means
1724 * i_size changes wouldn't be synchronized, so
1725 * one node could wind up truncating another
1728 if (end > i_size_read(inode)) {
1734 * We don't fill holes during direct io, so
1735 * check for them here. If any are found, the
1736 * caller will have to retake some cluster
1737 * locks and initiate the io as buffered.
1739 ret = ocfs2_check_range_for_holes(inode, saved_pos, count);
1752 ocfs2_inode_unlock(inode, meta_level);
1758 static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
1759 const struct iovec *iov,
1760 unsigned long nr_segs,
1763 int ret, direct_io, appending, rw_level, have_alloc_sem = 0;
1765 ssize_t written = 0;
1766 size_t ocount; /* original count */
1767 size_t count; /* after file limit checks */
1768 loff_t old_size, *ppos = &iocb->ki_pos;
1770 struct file *file = iocb->ki_filp;
1771 struct inode *inode = file->f_path.dentry->d_inode;
1772 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1774 mlog_entry("(0x%p, %u, '%.*s')\n", file,
1775 (unsigned int)nr_segs,
1776 file->f_path.dentry->d_name.len,
1777 file->f_path.dentry->d_name.name);
1779 if (iocb->ki_left == 0)
1782 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1784 appending = file->f_flags & O_APPEND ? 1 : 0;
1785 direct_io = file->f_flags & O_DIRECT ? 1 : 0;
1787 mutex_lock(&inode->i_mutex);
1790 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
1792 down_read(&inode->i_alloc_sem);
1796 /* concurrent O_DIRECT writes are allowed */
1797 rw_level = !direct_io;
1798 ret = ocfs2_rw_lock(inode, rw_level);
1804 can_do_direct = direct_io;
1805 ret = ocfs2_prepare_inode_for_write(file->f_path.dentry, ppos,
1806 iocb->ki_left, appending,
1814 * We can't complete the direct I/O as requested, fall back to
1817 if (direct_io && !can_do_direct) {
1818 ocfs2_rw_unlock(inode, rw_level);
1819 up_read(&inode->i_alloc_sem);
1829 * To later detect whether a journal commit for sync writes is
1830 * necessary, we sample i_size, and cluster count here.
1832 old_size = i_size_read(inode);
1833 old_clusters = OCFS2_I(inode)->ip_clusters;
1835 /* communicate with ocfs2_dio_end_io */
1836 ocfs2_iocb_set_rw_locked(iocb, rw_level);
1839 ret = generic_segment_checks(iov, &nr_segs, &ocount,
1844 ret = generic_write_checks(file, ppos, &count,
1845 S_ISBLK(inode->i_mode));
1849 written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
1850 ppos, count, ocount);
1853 * direct write may have instantiated a few
1854 * blocks outside i_size. Trim these off again.
1855 * Don't need i_size_read because we hold i_mutex.
1857 if (*ppos + count > inode->i_size)
1858 vmtruncate(inode, inode->i_size);
1863 written = generic_file_aio_write_nolock(iocb, iov, nr_segs,
1868 /* buffered aio wouldn't have proper lock coverage today */
1869 BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
1871 if ((file->f_flags & O_SYNC && !direct_io) || IS_SYNC(inode)) {
1873 * The generic write paths have handled getting data
1874 * to disk, but since we don't make use of the dirty
1875 * inode list, a manual journal commit is necessary
1878 if (old_size != i_size_read(inode) ||
1879 old_clusters != OCFS2_I(inode)->ip_clusters) {
1880 ret = jbd2_journal_force_commit(osb->journal->j_journal);
1887 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
1888 * function pointer which is called when o_direct io completes so that
1889 * it can unlock our rw lock. (it's the clustered equivalent of
1890 * i_alloc_sem; protects truncate from racing with pending ios).
1891 * Unfortunately there are error cases which call end_io and others
1892 * that don't. so we don't have to unlock the rw_lock if either an
1893 * async dio is going to do it in the future or an end_io after an
1894 * error has already done it.
1896 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
1903 ocfs2_rw_unlock(inode, rw_level);
1907 up_read(&inode->i_alloc_sem);
1909 mutex_unlock(&inode->i_mutex);
1912 return written ? written : ret;
1915 static int ocfs2_splice_to_file(struct pipe_inode_info *pipe,
1917 struct splice_desc *sd)
1921 ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, &sd->pos,
1922 sd->total_len, 0, NULL);
1928 return splice_from_pipe_feed(pipe, sd, pipe_to_file);
1931 static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
1938 struct address_space *mapping = out->f_mapping;
1939 struct inode *inode = mapping->host;
1940 struct splice_desc sd = {
1947 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe,
1949 out->f_path.dentry->d_name.len,
1950 out->f_path.dentry->d_name.name);
1953 mutex_lock_nested(&pipe->inode->i_mutex, I_MUTEX_PARENT);
1955 splice_from_pipe_begin(&sd);
1957 ret = splice_from_pipe_next(pipe, &sd);
1961 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
1962 ret = ocfs2_rw_lock(inode, 1);
1966 ret = ocfs2_splice_to_file(pipe, out, &sd);
1967 ocfs2_rw_unlock(inode, 1);
1969 mutex_unlock(&inode->i_mutex);
1971 splice_from_pipe_end(pipe, &sd);
1974 mutex_unlock(&pipe->inode->i_mutex);
1977 ret = sd.num_spliced;
1980 unsigned long nr_pages;
1983 nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1986 * If file or inode is SYNC and we actually wrote some data,
1989 if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) {
1992 mutex_lock(&inode->i_mutex);
1993 err = ocfs2_rw_lock(inode, 1);
1997 err = generic_osync_inode(inode, mapping,
1998 OSYNC_METADATA|OSYNC_DATA);
1999 ocfs2_rw_unlock(inode, 1);
2001 mutex_unlock(&inode->i_mutex);
2006 balance_dirty_pages_ratelimited_nr(mapping, nr_pages);
2013 static ssize_t ocfs2_file_splice_read(struct file *in,
2015 struct pipe_inode_info *pipe,
2020 struct inode *inode = in->f_path.dentry->d_inode;
2022 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe,
2024 in->f_path.dentry->d_name.len,
2025 in->f_path.dentry->d_name.name);
2028 * See the comment in ocfs2_file_aio_read()
2030 ret = ocfs2_inode_lock(inode, NULL, 0);
2035 ocfs2_inode_unlock(inode, 0);
2037 ret = generic_file_splice_read(in, ppos, pipe, len, flags);
2044 static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
2045 const struct iovec *iov,
2046 unsigned long nr_segs,
2049 int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
2050 struct file *filp = iocb->ki_filp;
2051 struct inode *inode = filp->f_path.dentry->d_inode;
2053 mlog_entry("(0x%p, %u, '%.*s')\n", filp,
2054 (unsigned int)nr_segs,
2055 filp->f_path.dentry->d_name.len,
2056 filp->f_path.dentry->d_name.name);
2065 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
2066 * need locks to protect pending reads from racing with truncate.
2068 if (filp->f_flags & O_DIRECT) {
2069 down_read(&inode->i_alloc_sem);
2072 ret = ocfs2_rw_lock(inode, 0);
2078 /* communicate with ocfs2_dio_end_io */
2079 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2083 * We're fine letting folks race truncates and extending
2084 * writes with read across the cluster, just like they can
2085 * locally. Hence no rw_lock during read.
2087 * Take and drop the meta data lock to update inode fields
2088 * like i_size. This allows the checks down below
2089 * generic_file_aio_read() a chance of actually working.
2091 ret = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level);
2096 ocfs2_inode_unlock(inode, lock_level);
2098 ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
2100 mlog(0, "generic_file_aio_read returned -EINVAL\n");
2102 /* buffered aio wouldn't have proper lock coverage today */
2103 BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
2105 /* see ocfs2_file_aio_write */
2106 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2113 up_read(&inode->i_alloc_sem);
2115 ocfs2_rw_unlock(inode, rw_level);
2121 const struct inode_operations ocfs2_file_iops = {
2122 .setattr = ocfs2_setattr,
2123 .getattr = ocfs2_getattr,
2124 .permission = ocfs2_permission,
2125 .setxattr = generic_setxattr,
2126 .getxattr = generic_getxattr,
2127 .listxattr = ocfs2_listxattr,
2128 .removexattr = generic_removexattr,
2129 .fallocate = ocfs2_fallocate,
2130 .fiemap = ocfs2_fiemap,
2133 const struct inode_operations ocfs2_special_file_iops = {
2134 .setattr = ocfs2_setattr,
2135 .getattr = ocfs2_getattr,
2136 .permission = ocfs2_permission,
2140 * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2141 * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2143 const struct file_operations ocfs2_fops = {
2144 .llseek = generic_file_llseek,
2145 .read = do_sync_read,
2146 .write = do_sync_write,
2148 .fsync = ocfs2_sync_file,
2149 .release = ocfs2_file_release,
2150 .open = ocfs2_file_open,
2151 .aio_read = ocfs2_file_aio_read,
2152 .aio_write = ocfs2_file_aio_write,
2153 .unlocked_ioctl = ocfs2_ioctl,
2154 #ifdef CONFIG_COMPAT
2155 .compat_ioctl = ocfs2_compat_ioctl,
2158 .flock = ocfs2_flock,
2159 .splice_read = ocfs2_file_splice_read,
2160 .splice_write = ocfs2_file_splice_write,
2163 const struct file_operations ocfs2_dops = {
2164 .llseek = generic_file_llseek,
2165 .read = generic_read_dir,
2166 .readdir = ocfs2_readdir,
2167 .fsync = ocfs2_sync_file,
2168 .release = ocfs2_dir_release,
2169 .open = ocfs2_dir_open,
2170 .unlocked_ioctl = ocfs2_ioctl,
2171 #ifdef CONFIG_COMPAT
2172 .compat_ioctl = ocfs2_compat_ioctl,
2175 .flock = ocfs2_flock,
2179 * POSIX-lockless variants of our file_operations.
2181 * These will be used if the underlying cluster stack does not support
2182 * posix file locking, if the user passes the "localflocks" mount
2183 * option, or if we have a local-only fs.
2185 * ocfs2_flock is in here because all stacks handle UNIX file locks,
2186 * so we still want it in the case of no stack support for
2187 * plocks. Internally, it will do the right thing when asked to ignore
2190 const struct file_operations ocfs2_fops_no_plocks = {
2191 .llseek = generic_file_llseek,
2192 .read = do_sync_read,
2193 .write = do_sync_write,
2195 .fsync = ocfs2_sync_file,
2196 .release = ocfs2_file_release,
2197 .open = ocfs2_file_open,
2198 .aio_read = ocfs2_file_aio_read,
2199 .aio_write = ocfs2_file_aio_write,
2200 .unlocked_ioctl = ocfs2_ioctl,
2201 #ifdef CONFIG_COMPAT
2202 .compat_ioctl = ocfs2_compat_ioctl,
2204 .flock = ocfs2_flock,
2205 .splice_read = ocfs2_file_splice_read,
2206 .splice_write = ocfs2_file_splice_write,
2209 const struct file_operations ocfs2_dops_no_plocks = {
2210 .llseek = generic_file_llseek,
2211 .read = generic_read_dir,
2212 .readdir = ocfs2_readdir,
2213 .fsync = ocfs2_sync_file,
2214 .release = ocfs2_dir_release,
2215 .open = ocfs2_dir_open,
2216 .unlocked_ioctl = ocfs2_ioctl,
2217 #ifdef CONFIG_COMPAT
2218 .compat_ioctl = ocfs2_compat_ioctl,
2220 .flock = ocfs2_flock,