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"
62 #include "refcounttree.h"
64 #include "buffer_head_io.h"
66 static int ocfs2_sync_inode(struct inode *inode)
68 filemap_fdatawrite(inode->i_mapping);
69 return sync_mapping_buffers(inode->i_mapping);
72 static int ocfs2_init_file_private(struct inode *inode, struct file *file)
74 struct ocfs2_file_private *fp;
76 fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
81 mutex_init(&fp->fp_mutex);
82 ocfs2_file_lock_res_init(&fp->fp_flock, fp);
83 file->private_data = fp;
88 static void ocfs2_free_file_private(struct inode *inode, struct file *file)
90 struct ocfs2_file_private *fp = file->private_data;
91 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
94 ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
95 ocfs2_lock_res_free(&fp->fp_flock);
97 file->private_data = NULL;
101 static int ocfs2_file_open(struct inode *inode, struct file *file)
104 int mode = file->f_flags;
105 struct ocfs2_inode_info *oi = OCFS2_I(inode);
107 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
108 file->f_path.dentry->d_name.len, file->f_path.dentry->d_name.name);
110 if (file->f_mode & FMODE_WRITE)
111 dquot_initialize(inode);
113 spin_lock(&oi->ip_lock);
115 /* Check that the inode hasn't been wiped from disk by another
116 * node. If it hasn't then we're safe as long as we hold the
117 * spin lock until our increment of open count. */
118 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
119 spin_unlock(&oi->ip_lock);
126 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
129 spin_unlock(&oi->ip_lock);
131 status = ocfs2_init_file_private(inode, file);
134 * We want to set open count back if we're failing the
137 spin_lock(&oi->ip_lock);
139 spin_unlock(&oi->ip_lock);
147 static int ocfs2_file_release(struct inode *inode, struct file *file)
149 struct ocfs2_inode_info *oi = OCFS2_I(inode);
151 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
152 file->f_path.dentry->d_name.len,
153 file->f_path.dentry->d_name.name);
155 spin_lock(&oi->ip_lock);
156 if (!--oi->ip_open_count)
157 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
158 spin_unlock(&oi->ip_lock);
160 ocfs2_free_file_private(inode, file);
167 static int ocfs2_dir_open(struct inode *inode, struct file *file)
169 return ocfs2_init_file_private(inode, file);
172 static int ocfs2_dir_release(struct inode *inode, struct file *file)
174 ocfs2_free_file_private(inode, file);
178 static int ocfs2_sync_file(struct file *file,
179 struct dentry *dentry,
184 struct inode *inode = dentry->d_inode;
185 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
187 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync,
188 dentry->d_name.len, dentry->d_name.name);
190 err = ocfs2_sync_inode(dentry->d_inode);
194 if (datasync && !(inode->i_state & I_DIRTY_DATASYNC))
197 journal = osb->journal->j_journal;
198 err = jbd2_journal_force_commit(journal);
203 return (err < 0) ? -EIO : 0;
206 int ocfs2_should_update_atime(struct inode *inode,
207 struct vfsmount *vfsmnt)
210 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
212 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
215 if ((inode->i_flags & S_NOATIME) ||
216 ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
220 * We can be called with no vfsmnt structure - NFSD will
223 * Note that our action here is different than touch_atime() -
224 * if we can't tell whether this is a noatime mount, then we
225 * don't know whether to trust the value of s_atime_quantum.
230 if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
231 ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
234 if (vfsmnt->mnt_flags & MNT_RELATIME) {
235 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
236 (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
243 if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
249 int ocfs2_update_inode_atime(struct inode *inode,
250 struct buffer_head *bh)
253 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
255 struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
259 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
260 if (IS_ERR(handle)) {
261 ret = PTR_ERR(handle);
266 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
267 OCFS2_JOURNAL_ACCESS_WRITE);
274 * Don't use ocfs2_mark_inode_dirty() here as we don't always
275 * have i_mutex to guard against concurrent changes to other
278 inode->i_atime = CURRENT_TIME;
279 di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
280 di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
281 ocfs2_journal_dirty(handle, bh);
284 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
290 static int ocfs2_set_inode_size(handle_t *handle,
292 struct buffer_head *fe_bh,
298 i_size_write(inode, new_i_size);
299 inode->i_blocks = ocfs2_inode_sector_count(inode);
300 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
302 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
313 int ocfs2_simple_size_update(struct inode *inode,
314 struct buffer_head *di_bh,
318 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
319 handle_t *handle = NULL;
321 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
322 if (IS_ERR(handle)) {
323 ret = PTR_ERR(handle);
328 ret = ocfs2_set_inode_size(handle, inode, di_bh,
333 ocfs2_commit_trans(osb, handle);
338 static int ocfs2_cow_file_pos(struct inode *inode,
339 struct buffer_head *fe_bh,
343 u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
344 unsigned int num_clusters = 0;
345 unsigned int ext_flags = 0;
348 * If the new offset is aligned to the range of the cluster, there is
349 * no space for ocfs2_zero_range_for_truncate to fill, so no need to
352 if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
355 status = ocfs2_get_clusters(inode, cpos, &phys,
356 &num_clusters, &ext_flags);
362 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
365 return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1);
371 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
373 struct buffer_head *fe_bh,
378 struct ocfs2_dinode *di;
384 * We need to CoW the cluster contains the offset if it is reflinked
385 * since we will call ocfs2_zero_range_for_truncate later which will
386 * write "0" from offset to the end of the cluster.
388 status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
394 /* TODO: This needs to actually orphan the inode in this
397 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
398 if (IS_ERR(handle)) {
399 status = PTR_ERR(handle);
404 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
405 OCFS2_JOURNAL_ACCESS_WRITE);
412 * Do this before setting i_size.
414 cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
415 status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
422 i_size_write(inode, new_i_size);
423 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
425 di = (struct ocfs2_dinode *) fe_bh->b_data;
426 di->i_size = cpu_to_le64(new_i_size);
427 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
428 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
430 ocfs2_journal_dirty(handle, fe_bh);
433 ocfs2_commit_trans(osb, handle);
440 static int ocfs2_truncate_file(struct inode *inode,
441 struct buffer_head *di_bh,
445 struct ocfs2_dinode *fe = NULL;
446 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
448 mlog_entry("(inode = %llu, new_i_size = %llu\n",
449 (unsigned long long)OCFS2_I(inode)->ip_blkno,
450 (unsigned long long)new_i_size);
452 /* We trust di_bh because it comes from ocfs2_inode_lock(), which
453 * already validated it */
454 fe = (struct ocfs2_dinode *) di_bh->b_data;
456 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
457 "Inode %llu, inode i_size = %lld != di "
458 "i_size = %llu, i_flags = 0x%x\n",
459 (unsigned long long)OCFS2_I(inode)->ip_blkno,
461 (unsigned long long)le64_to_cpu(fe->i_size),
462 le32_to_cpu(fe->i_flags));
464 if (new_i_size > le64_to_cpu(fe->i_size)) {
465 mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
466 (unsigned long long)le64_to_cpu(fe->i_size),
467 (unsigned long long)new_i_size);
473 mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
474 (unsigned long long)le64_to_cpu(fe->i_blkno),
475 (unsigned long long)le64_to_cpu(fe->i_size),
476 (unsigned long long)new_i_size);
478 /* lets handle the simple truncate cases before doing any more
479 * cluster locking. */
480 if (new_i_size == le64_to_cpu(fe->i_size))
483 down_write(&OCFS2_I(inode)->ip_alloc_sem);
485 ocfs2_resv_discard(&osb->osb_la_resmap,
486 &OCFS2_I(inode)->ip_la_data_resv);
489 * The inode lock forced other nodes to sync and drop their
490 * pages, which (correctly) happens even if we have a truncate
491 * without allocation change - ocfs2 cluster sizes can be much
492 * greater than page size, so we have to truncate them
495 unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
496 truncate_inode_pages(inode->i_mapping, new_i_size);
498 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
499 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
500 i_size_read(inode), 1);
504 goto bail_unlock_sem;
507 /* alright, we're going to need to do a full blown alloc size
508 * change. Orphan the inode so that recovery can complete the
509 * truncate if necessary. This does the task of marking
511 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
514 goto bail_unlock_sem;
517 status = ocfs2_commit_truncate(osb, inode, di_bh);
520 goto bail_unlock_sem;
523 /* TODO: orphan dir cleanup here. */
525 up_write(&OCFS2_I(inode)->ip_alloc_sem);
528 if (!status && OCFS2_I(inode)->ip_clusters == 0)
529 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
536 * extend file allocation only here.
537 * we'll update all the disk stuff, and oip->alloc_size
539 * expect stuff to be locked, a transaction started and enough data /
540 * metadata reservations in the contexts.
542 * Will return -EAGAIN, and a reason if a restart is needed.
543 * If passed in, *reason will always be set, even in error.
545 int ocfs2_add_inode_data(struct ocfs2_super *osb,
550 struct buffer_head *fe_bh,
552 struct ocfs2_alloc_context *data_ac,
553 struct ocfs2_alloc_context *meta_ac,
554 enum ocfs2_alloc_restarted *reason_ret)
557 struct ocfs2_extent_tree et;
559 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
560 ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
561 clusters_to_add, mark_unwritten,
562 data_ac, meta_ac, reason_ret);
567 static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
568 u32 clusters_to_add, int mark_unwritten)
571 int restart_func = 0;
574 struct buffer_head *bh = NULL;
575 struct ocfs2_dinode *fe = NULL;
576 handle_t *handle = NULL;
577 struct ocfs2_alloc_context *data_ac = NULL;
578 struct ocfs2_alloc_context *meta_ac = NULL;
579 enum ocfs2_alloc_restarted why;
580 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
581 struct ocfs2_extent_tree et;
584 mlog_entry("(clusters_to_add = %u)\n", clusters_to_add);
587 * This function only exists for file systems which don't
590 BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
592 status = ocfs2_read_inode_block(inode, &bh);
597 fe = (struct ocfs2_dinode *) bh->b_data;
600 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
602 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, "
603 "clusters_to_add = %u\n",
604 (unsigned long long)OCFS2_I(inode)->ip_blkno,
605 (long long)i_size_read(inode), le32_to_cpu(fe->i_clusters),
607 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
608 status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
615 credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list,
617 handle = ocfs2_start_trans(osb, credits);
618 if (IS_ERR(handle)) {
619 status = PTR_ERR(handle);
625 restarted_transaction:
626 status = dquot_alloc_space_nodirty(inode,
627 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
632 /* reserve a write to the file entry early on - that we if we
633 * run out of credits in the allocation path, we can still
635 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
636 OCFS2_JOURNAL_ACCESS_WRITE);
642 prev_clusters = OCFS2_I(inode)->ip_clusters;
644 status = ocfs2_add_inode_data(osb,
654 if ((status < 0) && (status != -EAGAIN)) {
655 if (status != -ENOSPC)
660 ocfs2_journal_dirty(handle, bh);
662 spin_lock(&OCFS2_I(inode)->ip_lock);
663 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
664 spin_unlock(&OCFS2_I(inode)->ip_lock);
665 /* Release unused quota reservation */
666 dquot_free_space(inode,
667 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
670 if (why != RESTART_NONE && clusters_to_add) {
671 if (why == RESTART_META) {
672 mlog(0, "restarting function.\n");
676 BUG_ON(why != RESTART_TRANS);
678 mlog(0, "restarting transaction.\n");
679 /* TODO: This can be more intelligent. */
680 credits = ocfs2_calc_extend_credits(osb->sb,
683 status = ocfs2_extend_trans(handle, credits);
685 /* handle still has to be committed at
691 goto restarted_transaction;
695 mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
696 le32_to_cpu(fe->i_clusters),
697 (unsigned long long)le64_to_cpu(fe->i_size));
698 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
699 OCFS2_I(inode)->ip_clusters, (long long)i_size_read(inode));
702 if (status < 0 && did_quota)
703 dquot_free_space(inode,
704 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
706 ocfs2_commit_trans(osb, handle);
710 ocfs2_free_alloc_context(data_ac);
714 ocfs2_free_alloc_context(meta_ac);
717 if ((!status) && restart_func) {
728 /* Some parts of this taken from generic_cont_expand, which turned out
729 * to be too fragile to do exactly what we need without us having to
730 * worry about recursive locking in ->write_begin() and ->write_end(). */
731 static int ocfs2_write_zero_page(struct inode *inode,
734 struct address_space *mapping = inode->i_mapping;
738 handle_t *handle = NULL;
741 offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */
742 /* ugh. in prepare/commit_write, if from==to==start of block, we
743 ** skip the prepare. make sure we never send an offset for the start
746 if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) {
749 index = size >> PAGE_CACHE_SHIFT;
751 page = grab_cache_page(mapping, index);
758 ret = ocfs2_prepare_write_nolock(inode, page, offset, offset);
764 if (ocfs2_should_order_data(inode)) {
765 handle = ocfs2_start_walk_page_trans(inode, page, offset,
767 if (IS_ERR(handle)) {
768 ret = PTR_ERR(handle);
774 /* must not update i_size! */
775 ret = block_commit_write(page, offset, offset);
782 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
785 page_cache_release(page);
790 static int ocfs2_zero_extend(struct inode *inode,
795 struct super_block *sb = inode->i_sb;
797 start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
798 while (start_off < zero_to_size) {
799 ret = ocfs2_write_zero_page(inode, start_off);
805 start_off += sb->s_blocksize;
808 * Very large extends have the potential to lock up
809 * the cpu for extended periods of time.
818 int ocfs2_extend_no_holes(struct inode *inode, u64 new_i_size, u64 zero_to)
822 struct ocfs2_inode_info *oi = OCFS2_I(inode);
824 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
825 if (clusters_to_add < oi->ip_clusters)
828 clusters_to_add -= oi->ip_clusters;
830 if (clusters_to_add) {
831 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
840 * Call this even if we don't add any clusters to the tree. We
841 * still need to zero the area between the old i_size and the
844 ret = ocfs2_zero_extend(inode, zero_to);
852 static int ocfs2_extend_file(struct inode *inode,
853 struct buffer_head *di_bh,
857 struct ocfs2_inode_info *oi = OCFS2_I(inode);
861 /* setattr sometimes calls us like this. */
865 if (i_size_read(inode) == new_i_size)
867 BUG_ON(new_i_size < i_size_read(inode));
870 * Fall through for converting inline data, even if the fs
871 * supports sparse files.
873 * The check for inline data here is legal - nobody can add
874 * the feature since we have i_mutex. We must check it again
875 * after acquiring ip_alloc_sem though, as paths like mmap
876 * might have raced us to converting the inode to extents.
878 if (!(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
879 && ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
880 goto out_update_size;
883 * The alloc sem blocks people in read/write from reading our
884 * allocation until we're done changing it. We depend on
885 * i_mutex to block other extend/truncate calls while we're
888 down_write(&oi->ip_alloc_sem);
890 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
892 * We can optimize small extends by keeping the inodes
895 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
896 up_write(&oi->ip_alloc_sem);
897 goto out_update_size;
900 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
902 up_write(&oi->ip_alloc_sem);
909 if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
910 ret = ocfs2_extend_no_holes(inode, new_i_size, new_i_size);
912 up_write(&oi->ip_alloc_sem);
920 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
928 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
930 int status = 0, size_change;
931 struct inode *inode = dentry->d_inode;
932 struct super_block *sb = inode->i_sb;
933 struct ocfs2_super *osb = OCFS2_SB(sb);
934 struct buffer_head *bh = NULL;
935 handle_t *handle = NULL;
937 struct dquot *transfer_from[MAXQUOTAS] = { };
938 struct dquot *transfer_to[MAXQUOTAS] = { };
940 mlog_entry("(0x%p, '%.*s')\n", dentry,
941 dentry->d_name.len, dentry->d_name.name);
943 /* ensuring we don't even attempt to truncate a symlink */
944 if (S_ISLNK(inode->i_mode))
945 attr->ia_valid &= ~ATTR_SIZE;
947 if (attr->ia_valid & ATTR_MODE)
948 mlog(0, "mode change: %d\n", attr->ia_mode);
949 if (attr->ia_valid & ATTR_UID)
950 mlog(0, "uid change: %d\n", attr->ia_uid);
951 if (attr->ia_valid & ATTR_GID)
952 mlog(0, "gid change: %d\n", attr->ia_gid);
953 if (attr->ia_valid & ATTR_SIZE)
954 mlog(0, "size change...\n");
955 if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME))
956 mlog(0, "time change...\n");
958 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
959 | ATTR_GID | ATTR_UID | ATTR_MODE)
960 if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) {
961 mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid);
965 status = inode_change_ok(inode, attr);
969 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
971 dquot_initialize(inode);
973 status = ocfs2_rw_lock(inode, 1);
980 status = ocfs2_inode_lock(inode, &bh, 1);
982 if (status != -ENOENT)
987 if (size_change && attr->ia_size != i_size_read(inode)) {
988 status = inode_newsize_ok(inode, attr->ia_size);
992 if (i_size_read(inode) > attr->ia_size) {
993 if (ocfs2_should_order_data(inode)) {
994 status = ocfs2_begin_ordered_truncate(inode,
999 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1001 status = ocfs2_extend_file(inode, bh, attr->ia_size);
1003 if (status != -ENOSPC)
1010 if ((attr->ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
1011 (attr->ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
1013 * Gather pointers to quota structures so that allocation /
1014 * freeing of quota structures happens here and not inside
1015 * dquot_transfer() where we have problems with lock ordering
1017 if (attr->ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid
1018 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1019 OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1020 transfer_to[USRQUOTA] = dqget(sb, attr->ia_uid,
1022 transfer_from[USRQUOTA] = dqget(sb, inode->i_uid,
1024 if (!transfer_to[USRQUOTA] || !transfer_from[USRQUOTA]) {
1029 if (attr->ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid
1030 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1031 OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1032 transfer_to[GRPQUOTA] = dqget(sb, attr->ia_gid,
1034 transfer_from[GRPQUOTA] = dqget(sb, inode->i_gid,
1036 if (!transfer_to[GRPQUOTA] || !transfer_from[GRPQUOTA]) {
1041 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1042 2 * ocfs2_quota_trans_credits(sb));
1043 if (IS_ERR(handle)) {
1044 status = PTR_ERR(handle);
1048 status = dquot_transfer(inode, attr);
1052 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1053 if (IS_ERR(handle)) {
1054 status = PTR_ERR(handle);
1061 * This will intentionally not wind up calling vmtruncate(),
1062 * since all the work for a size change has been done above.
1063 * Otherwise, we could get into problems with truncate as
1064 * ip_alloc_sem is used there to protect against i_size
1067 status = inode_setattr(inode, attr);
1073 status = ocfs2_mark_inode_dirty(handle, inode, bh);
1078 ocfs2_commit_trans(osb, handle);
1080 ocfs2_inode_unlock(inode, 1);
1083 ocfs2_rw_unlock(inode, 1);
1087 /* Release quota pointers in case we acquired them */
1088 for (qtype = 0; qtype < MAXQUOTAS; qtype++) {
1089 dqput(transfer_to[qtype]);
1090 dqput(transfer_from[qtype]);
1093 if (!status && attr->ia_valid & ATTR_MODE) {
1094 status = ocfs2_acl_chmod(inode);
1103 int ocfs2_getattr(struct vfsmount *mnt,
1104 struct dentry *dentry,
1107 struct inode *inode = dentry->d_inode;
1108 struct super_block *sb = dentry->d_inode->i_sb;
1109 struct ocfs2_super *osb = sb->s_fs_info;
1114 err = ocfs2_inode_revalidate(dentry);
1121 generic_fillattr(inode, stat);
1123 /* We set the blksize from the cluster size for performance */
1124 stat->blksize = osb->s_clustersize;
1132 int ocfs2_permission(struct inode *inode, int mask)
1138 ret = ocfs2_inode_lock(inode, NULL, 0);
1145 ret = generic_permission(inode, mask, ocfs2_check_acl);
1147 ocfs2_inode_unlock(inode, 0);
1153 static int __ocfs2_write_remove_suid(struct inode *inode,
1154 struct buffer_head *bh)
1158 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1159 struct ocfs2_dinode *di;
1161 mlog_entry("(Inode %llu, mode 0%o)\n",
1162 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_mode);
1164 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1165 if (IS_ERR(handle)) {
1166 ret = PTR_ERR(handle);
1171 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1172 OCFS2_JOURNAL_ACCESS_WRITE);
1178 inode->i_mode &= ~S_ISUID;
1179 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1180 inode->i_mode &= ~S_ISGID;
1182 di = (struct ocfs2_dinode *) bh->b_data;
1183 di->i_mode = cpu_to_le16(inode->i_mode);
1185 ocfs2_journal_dirty(handle, bh);
1188 ocfs2_commit_trans(osb, handle);
1195 * Will look for holes and unwritten extents in the range starting at
1196 * pos for count bytes (inclusive).
1198 static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1202 unsigned int extent_flags;
1203 u32 cpos, clusters, extent_len, phys_cpos;
1204 struct super_block *sb = inode->i_sb;
1206 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1207 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1210 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1217 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
1222 if (extent_len > clusters)
1223 extent_len = clusters;
1225 clusters -= extent_len;
1232 static int ocfs2_write_remove_suid(struct inode *inode)
1235 struct buffer_head *bh = NULL;
1237 ret = ocfs2_read_inode_block(inode, &bh);
1243 ret = __ocfs2_write_remove_suid(inode, bh);
1250 * Allocate enough extents to cover the region starting at byte offset
1251 * start for len bytes. Existing extents are skipped, any extents
1252 * added are marked as "unwritten".
1254 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1258 u32 cpos, phys_cpos, clusters, alloc_size;
1259 u64 end = start + len;
1260 struct buffer_head *di_bh = NULL;
1262 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1263 ret = ocfs2_read_inode_block(inode, &di_bh);
1270 * Nothing to do if the requested reservation range
1271 * fits within the inode.
1273 if (ocfs2_size_fits_inline_data(di_bh, end))
1276 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1284 * We consider both start and len to be inclusive.
1286 cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1287 clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1291 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1299 * Hole or existing extent len can be arbitrary, so
1300 * cap it to our own allocation request.
1302 if (alloc_size > clusters)
1303 alloc_size = clusters;
1307 * We already have an allocation at this
1308 * region so we can safely skip it.
1313 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1322 clusters -= alloc_size;
1333 * Truncate a byte range, avoiding pages within partial clusters. This
1334 * preserves those pages for the zeroing code to write to.
1336 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1339 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1341 struct address_space *mapping = inode->i_mapping;
1343 start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1344 end = byte_start + byte_len;
1345 end = end & ~(osb->s_clustersize - 1);
1348 unmap_mapping_range(mapping, start, end - start, 0);
1349 truncate_inode_pages_range(mapping, start, end - 1);
1353 static int ocfs2_zero_partial_clusters(struct inode *inode,
1357 u64 tmpend, end = start + len;
1358 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1359 unsigned int csize = osb->s_clustersize;
1363 * The "start" and "end" values are NOT necessarily part of
1364 * the range whose allocation is being deleted. Rather, this
1365 * is what the user passed in with the request. We must zero
1366 * partial clusters here. There's no need to worry about
1367 * physical allocation - the zeroing code knows to skip holes.
1369 mlog(0, "byte start: %llu, end: %llu\n",
1370 (unsigned long long)start, (unsigned long long)end);
1373 * If both edges are on a cluster boundary then there's no
1374 * zeroing required as the region is part of the allocation to
1377 if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1380 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1381 if (IS_ERR(handle)) {
1382 ret = PTR_ERR(handle);
1388 * We want to get the byte offset of the end of the 1st cluster.
1390 tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1394 mlog(0, "1st range: start: %llu, tmpend: %llu\n",
1395 (unsigned long long)start, (unsigned long long)tmpend);
1397 ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1403 * This may make start and end equal, but the zeroing
1404 * code will skip any work in that case so there's no
1405 * need to catch it up here.
1407 start = end & ~(osb->s_clustersize - 1);
1409 mlog(0, "2nd range: start: %llu, end: %llu\n",
1410 (unsigned long long)start, (unsigned long long)end);
1412 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1417 ocfs2_commit_trans(osb, handle);
1422 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1425 struct ocfs2_extent_rec *rec = NULL;
1427 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1429 rec = &el->l_recs[i];
1431 if (le32_to_cpu(rec->e_cpos) < pos)
1439 * Helper to calculate the punching pos and length in one run, we handle the
1440 * following three cases in order:
1442 * - remove the entire record
1443 * - remove a partial record
1444 * - no record needs to be removed (hole-punching completed)
1446 static void ocfs2_calc_trunc_pos(struct inode *inode,
1447 struct ocfs2_extent_list *el,
1448 struct ocfs2_extent_rec *rec,
1449 u32 trunc_start, u32 *trunc_cpos,
1450 u32 *trunc_len, u32 *trunc_end,
1451 u64 *blkno, int *done)
1456 range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1458 if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1459 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1461 * Skip holes if any.
1463 if (range < *trunc_end)
1465 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1466 *blkno = le64_to_cpu(rec->e_blkno);
1467 *trunc_end = le32_to_cpu(rec->e_cpos);
1468 } else if (range > trunc_start) {
1469 *trunc_cpos = trunc_start;
1470 *trunc_len = *trunc_end - trunc_start;
1471 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1472 *blkno = le64_to_cpu(rec->e_blkno) +
1473 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1474 *trunc_end = trunc_start;
1477 * It may have two following possibilities:
1479 * - last record has been removed
1480 * - trunc_start was within a hole
1482 * both two cases mean the completion of hole punching.
1490 static int ocfs2_remove_inode_range(struct inode *inode,
1491 struct buffer_head *di_bh, u64 byte_start,
1494 int ret = 0, flags = 0, done = 0, i;
1495 u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1497 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1498 struct ocfs2_cached_dealloc_ctxt dealloc;
1499 struct address_space *mapping = inode->i_mapping;
1500 struct ocfs2_extent_tree et;
1501 struct ocfs2_path *path = NULL;
1502 struct ocfs2_extent_list *el = NULL;
1503 struct ocfs2_extent_rec *rec = NULL;
1504 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1505 u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1507 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1508 ocfs2_init_dealloc_ctxt(&dealloc);
1513 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1514 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1515 byte_start + byte_len, 0);
1521 * There's no need to get fancy with the page cache
1522 * truncate of an inline-data inode. We're talking
1523 * about less than a page here, which will be cached
1524 * in the dinode buffer anyway.
1526 unmap_mapping_range(mapping, 0, 0, 0);
1527 truncate_inode_pages(mapping, 0);
1532 * For reflinks, we may need to CoW 2 clusters which might be
1533 * partially zero'd later, if hole's start and end offset were
1534 * within one cluster(means is not exactly aligned to clustersize).
1537 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
1539 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1545 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1552 trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1553 trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1554 cluster_in_el = trunc_end;
1556 mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, cend: %u\n",
1557 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1558 (unsigned long long)byte_start,
1559 (unsigned long long)byte_len, trunc_start, trunc_end);
1561 ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1567 path = ocfs2_new_path_from_et(&et);
1574 while (trunc_end > trunc_start) {
1576 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1583 el = path_leaf_el(path);
1585 i = ocfs2_find_rec(el, trunc_end);
1587 * Need to go to previous extent block.
1590 if (path->p_tree_depth == 0)
1593 ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1602 * We've reached the leftmost extent block,
1603 * it's safe to leave.
1605 if (cluster_in_el == 0)
1609 * The 'pos' searched for previous extent block is
1610 * always one cluster less than actual trunc_end.
1612 trunc_end = cluster_in_el + 1;
1614 ocfs2_reinit_path(path, 1);
1619 rec = &el->l_recs[i];
1621 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1622 &trunc_len, &trunc_end, &blkno, &done);
1626 flags = rec->e_flags;
1627 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1629 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1630 phys_cpos, trunc_len, flags,
1631 &dealloc, refcount_loc);
1637 cluster_in_el = trunc_end;
1639 ocfs2_reinit_path(path, 1);
1642 ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1645 ocfs2_schedule_truncate_log_flush(osb, 1);
1646 ocfs2_run_deallocs(osb, &dealloc);
1652 * Parts of this function taken from xfs_change_file_space()
1654 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1655 loff_t f_pos, unsigned int cmd,
1656 struct ocfs2_space_resv *sr,
1662 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1663 struct buffer_head *di_bh = NULL;
1665 unsigned long long max_off = inode->i_sb->s_maxbytes;
1667 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1670 mutex_lock(&inode->i_mutex);
1673 * This prevents concurrent writes on other nodes
1675 ret = ocfs2_rw_lock(inode, 1);
1681 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1687 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1689 goto out_inode_unlock;
1692 switch (sr->l_whence) {
1693 case 0: /*SEEK_SET*/
1695 case 1: /*SEEK_CUR*/
1696 sr->l_start += f_pos;
1698 case 2: /*SEEK_END*/
1699 sr->l_start += i_size_read(inode);
1703 goto out_inode_unlock;
1707 llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1710 || sr->l_start > max_off
1711 || (sr->l_start + llen) < 0
1712 || (sr->l_start + llen) > max_off) {
1714 goto out_inode_unlock;
1716 size = sr->l_start + sr->l_len;
1718 if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) {
1719 if (sr->l_len <= 0) {
1721 goto out_inode_unlock;
1725 if (file && should_remove_suid(file->f_path.dentry)) {
1726 ret = __ocfs2_write_remove_suid(inode, di_bh);
1729 goto out_inode_unlock;
1733 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1735 case OCFS2_IOC_RESVSP:
1736 case OCFS2_IOC_RESVSP64:
1738 * This takes unsigned offsets, but the signed ones we
1739 * pass have been checked against overflow above.
1741 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1744 case OCFS2_IOC_UNRESVSP:
1745 case OCFS2_IOC_UNRESVSP64:
1746 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1752 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1755 goto out_inode_unlock;
1759 * We update c/mtime for these changes
1761 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1762 if (IS_ERR(handle)) {
1763 ret = PTR_ERR(handle);
1765 goto out_inode_unlock;
1768 if (change_size && i_size_read(inode) < size)
1769 i_size_write(inode, size);
1771 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1772 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1776 ocfs2_commit_trans(osb, handle);
1780 ocfs2_inode_unlock(inode, 1);
1782 ocfs2_rw_unlock(inode, 1);
1785 mutex_unlock(&inode->i_mutex);
1789 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1790 struct ocfs2_space_resv *sr)
1792 struct inode *inode = file->f_path.dentry->d_inode;
1793 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1795 if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1796 !ocfs2_writes_unwritten_extents(osb))
1798 else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1799 !ocfs2_sparse_alloc(osb))
1802 if (!S_ISREG(inode->i_mode))
1805 if (!(file->f_mode & FMODE_WRITE))
1808 return __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1811 static long ocfs2_fallocate(struct inode *inode, int mode, loff_t offset,
1814 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1815 struct ocfs2_space_resv sr;
1816 int change_size = 1;
1818 if (!ocfs2_writes_unwritten_extents(osb))
1821 if (S_ISDIR(inode->i_mode))
1824 if (mode & FALLOC_FL_KEEP_SIZE)
1828 sr.l_start = (s64)offset;
1829 sr.l_len = (s64)len;
1831 return __ocfs2_change_file_space(NULL, inode, offset,
1832 OCFS2_IOC_RESVSP64, &sr, change_size);
1835 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
1839 unsigned int extent_flags;
1840 u32 cpos, clusters, extent_len, phys_cpos;
1841 struct super_block *sb = inode->i_sb;
1843 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
1844 !(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) ||
1845 OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
1848 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1849 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1852 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1859 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
1864 if (extent_len > clusters)
1865 extent_len = clusters;
1867 clusters -= extent_len;
1874 static int ocfs2_prepare_inode_for_refcount(struct inode *inode,
1875 loff_t pos, size_t count,
1879 struct buffer_head *di_bh = NULL;
1880 u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1882 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
1884 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1892 ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
1900 static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
1907 int ret = 0, meta_level = 0;
1908 struct inode *inode = dentry->d_inode;
1909 loff_t saved_pos, end;
1912 * We start with a read level meta lock and only jump to an ex
1913 * if we need to make modifications here.
1916 ret = ocfs2_inode_lock(inode, NULL, meta_level);
1923 /* Clear suid / sgid if necessary. We do this here
1924 * instead of later in the write path because
1925 * remove_suid() calls ->setattr without any hint that
1926 * we may have already done our cluster locking. Since
1927 * ocfs2_setattr() *must* take cluster locks to
1928 * proceeed, this will lead us to recursively lock the
1929 * inode. There's also the dinode i_size state which
1930 * can be lost via setattr during extending writes (we
1931 * set inode->i_size at the end of a write. */
1932 if (should_remove_suid(dentry)) {
1933 if (meta_level == 0) {
1934 ocfs2_inode_unlock(inode, meta_level);
1939 ret = ocfs2_write_remove_suid(inode);
1946 /* work on a copy of ppos until we're sure that we won't have
1947 * to recalculate it due to relocking. */
1949 saved_pos = i_size_read(inode);
1950 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos);
1955 end = saved_pos + count;
1957 ret = ocfs2_check_range_for_refcount(inode, saved_pos, count);
1959 ocfs2_inode_unlock(inode, meta_level);
1962 ret = ocfs2_prepare_inode_for_refcount(inode,
1978 * Skip the O_DIRECT checks if we don't need
1981 if (!direct_io || !(*direct_io))
1985 * There's no sane way to do direct writes to an inode
1988 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1994 * Allowing concurrent direct writes means
1995 * i_size changes wouldn't be synchronized, so
1996 * one node could wind up truncating another
1999 if (end > i_size_read(inode)) {
2005 * We don't fill holes during direct io, so
2006 * check for them here. If any are found, the
2007 * caller will have to retake some cluster
2008 * locks and initiate the io as buffered.
2010 ret = ocfs2_check_range_for_holes(inode, saved_pos, count);
2023 if (meta_level >= 0)
2024 ocfs2_inode_unlock(inode, meta_level);
2030 static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
2031 const struct iovec *iov,
2032 unsigned long nr_segs,
2035 int ret, direct_io, appending, rw_level, have_alloc_sem = 0;
2036 int can_do_direct, has_refcount = 0;
2037 ssize_t written = 0;
2038 size_t ocount; /* original count */
2039 size_t count; /* after file limit checks */
2040 loff_t old_size, *ppos = &iocb->ki_pos;
2042 struct file *file = iocb->ki_filp;
2043 struct inode *inode = file->f_path.dentry->d_inode;
2044 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2046 mlog_entry("(0x%p, %u, '%.*s')\n", file,
2047 (unsigned int)nr_segs,
2048 file->f_path.dentry->d_name.len,
2049 file->f_path.dentry->d_name.name);
2051 if (iocb->ki_left == 0)
2054 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2056 appending = file->f_flags & O_APPEND ? 1 : 0;
2057 direct_io = file->f_flags & O_DIRECT ? 1 : 0;
2059 mutex_lock(&inode->i_mutex);
2062 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
2064 down_read(&inode->i_alloc_sem);
2068 /* concurrent O_DIRECT writes are allowed */
2069 rw_level = !direct_io;
2070 ret = ocfs2_rw_lock(inode, rw_level);
2076 can_do_direct = direct_io;
2077 ret = ocfs2_prepare_inode_for_write(file->f_path.dentry, ppos,
2078 iocb->ki_left, appending,
2079 &can_do_direct, &has_refcount);
2086 * We can't complete the direct I/O as requested, fall back to
2089 if (direct_io && !can_do_direct) {
2090 ocfs2_rw_unlock(inode, rw_level);
2091 up_read(&inode->i_alloc_sem);
2101 * To later detect whether a journal commit for sync writes is
2102 * necessary, we sample i_size, and cluster count here.
2104 old_size = i_size_read(inode);
2105 old_clusters = OCFS2_I(inode)->ip_clusters;
2107 /* communicate with ocfs2_dio_end_io */
2108 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2110 ret = generic_segment_checks(iov, &nr_segs, &ocount,
2116 ret = generic_write_checks(file, ppos, &count,
2117 S_ISBLK(inode->i_mode));
2122 written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
2123 ppos, count, ocount);
2126 * direct write may have instantiated a few
2127 * blocks outside i_size. Trim these off again.
2128 * Don't need i_size_read because we hold i_mutex.
2130 if (*ppos + count > inode->i_size)
2131 vmtruncate(inode, inode->i_size);
2136 current->backing_dev_info = file->f_mapping->backing_dev_info;
2137 written = generic_file_buffered_write(iocb, iov, nr_segs, *ppos,
2139 current->backing_dev_info = NULL;
2143 /* buffered aio wouldn't have proper lock coverage today */
2144 BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
2146 if (((file->f_flags & O_DSYNC) && !direct_io) || IS_SYNC(inode) ||
2147 ((file->f_flags & O_DIRECT) && has_refcount)) {
2148 ret = filemap_fdatawrite_range(file->f_mapping, pos,
2153 if (!ret && ((old_size != i_size_read(inode)) ||
2154 (old_clusters != OCFS2_I(inode)->ip_clusters) ||
2156 ret = jbd2_journal_force_commit(osb->journal->j_journal);
2162 ret = filemap_fdatawait_range(file->f_mapping, pos,
2167 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2168 * function pointer which is called when o_direct io completes so that
2169 * it can unlock our rw lock. (it's the clustered equivalent of
2170 * i_alloc_sem; protects truncate from racing with pending ios).
2171 * Unfortunately there are error cases which call end_io and others
2172 * that don't. so we don't have to unlock the rw_lock if either an
2173 * async dio is going to do it in the future or an end_io after an
2174 * error has already done it.
2176 if ((ret == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2183 ocfs2_rw_unlock(inode, rw_level);
2187 up_read(&inode->i_alloc_sem);
2189 mutex_unlock(&inode->i_mutex);
2197 static int ocfs2_splice_to_file(struct pipe_inode_info *pipe,
2199 struct splice_desc *sd)
2203 ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, &sd->pos,
2204 sd->total_len, 0, NULL, NULL);
2210 return splice_from_pipe_feed(pipe, sd, pipe_to_file);
2213 static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
2220 struct address_space *mapping = out->f_mapping;
2221 struct inode *inode = mapping->host;
2222 struct splice_desc sd = {
2229 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe,
2231 out->f_path.dentry->d_name.len,
2232 out->f_path.dentry->d_name.name);
2235 mutex_lock_nested(&pipe->inode->i_mutex, I_MUTEX_PARENT);
2237 splice_from_pipe_begin(&sd);
2239 ret = splice_from_pipe_next(pipe, &sd);
2243 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2244 ret = ocfs2_rw_lock(inode, 1);
2248 ret = ocfs2_splice_to_file(pipe, out, &sd);
2249 ocfs2_rw_unlock(inode, 1);
2251 mutex_unlock(&inode->i_mutex);
2253 splice_from_pipe_end(pipe, &sd);
2256 mutex_unlock(&pipe->inode->i_mutex);
2259 ret = sd.num_spliced;
2262 unsigned long nr_pages;
2265 nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
2267 err = generic_write_sync(out, *ppos, ret);
2273 balance_dirty_pages_ratelimited_nr(mapping, nr_pages);
2280 static ssize_t ocfs2_file_splice_read(struct file *in,
2282 struct pipe_inode_info *pipe,
2286 int ret = 0, lock_level = 0;
2287 struct inode *inode = in->f_path.dentry->d_inode;
2289 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe,
2291 in->f_path.dentry->d_name.len,
2292 in->f_path.dentry->d_name.name);
2295 * See the comment in ocfs2_file_aio_read()
2297 ret = ocfs2_inode_lock_atime(inode, in->f_vfsmnt, &lock_level);
2302 ocfs2_inode_unlock(inode, lock_level);
2304 ret = generic_file_splice_read(in, ppos, pipe, len, flags);
2311 static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
2312 const struct iovec *iov,
2313 unsigned long nr_segs,
2316 int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
2317 struct file *filp = iocb->ki_filp;
2318 struct inode *inode = filp->f_path.dentry->d_inode;
2320 mlog_entry("(0x%p, %u, '%.*s')\n", filp,
2321 (unsigned int)nr_segs,
2322 filp->f_path.dentry->d_name.len,
2323 filp->f_path.dentry->d_name.name);
2332 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
2333 * need locks to protect pending reads from racing with truncate.
2335 if (filp->f_flags & O_DIRECT) {
2336 down_read(&inode->i_alloc_sem);
2339 ret = ocfs2_rw_lock(inode, 0);
2345 /* communicate with ocfs2_dio_end_io */
2346 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2350 * We're fine letting folks race truncates and extending
2351 * writes with read across the cluster, just like they can
2352 * locally. Hence no rw_lock during read.
2354 * Take and drop the meta data lock to update inode fields
2355 * like i_size. This allows the checks down below
2356 * generic_file_aio_read() a chance of actually working.
2358 ret = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level);
2363 ocfs2_inode_unlock(inode, lock_level);
2365 ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
2367 mlog(0, "generic_file_aio_read returned -EINVAL\n");
2369 /* buffered aio wouldn't have proper lock coverage today */
2370 BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
2372 /* see ocfs2_file_aio_write */
2373 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2380 up_read(&inode->i_alloc_sem);
2382 ocfs2_rw_unlock(inode, rw_level);
2388 const struct inode_operations ocfs2_file_iops = {
2389 .setattr = ocfs2_setattr,
2390 .getattr = ocfs2_getattr,
2391 .permission = ocfs2_permission,
2392 .setxattr = generic_setxattr,
2393 .getxattr = generic_getxattr,
2394 .listxattr = ocfs2_listxattr,
2395 .removexattr = generic_removexattr,
2396 .fallocate = ocfs2_fallocate,
2397 .fiemap = ocfs2_fiemap,
2400 const struct inode_operations ocfs2_special_file_iops = {
2401 .setattr = ocfs2_setattr,
2402 .getattr = ocfs2_getattr,
2403 .permission = ocfs2_permission,
2407 * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2408 * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2410 const struct file_operations ocfs2_fops = {
2411 .llseek = generic_file_llseek,
2412 .read = do_sync_read,
2413 .write = do_sync_write,
2415 .fsync = ocfs2_sync_file,
2416 .release = ocfs2_file_release,
2417 .open = ocfs2_file_open,
2418 .aio_read = ocfs2_file_aio_read,
2419 .aio_write = ocfs2_file_aio_write,
2420 .unlocked_ioctl = ocfs2_ioctl,
2421 #ifdef CONFIG_COMPAT
2422 .compat_ioctl = ocfs2_compat_ioctl,
2425 .flock = ocfs2_flock,
2426 .splice_read = ocfs2_file_splice_read,
2427 .splice_write = ocfs2_file_splice_write,
2430 const struct file_operations ocfs2_dops = {
2431 .llseek = generic_file_llseek,
2432 .read = generic_read_dir,
2433 .readdir = ocfs2_readdir,
2434 .fsync = ocfs2_sync_file,
2435 .release = ocfs2_dir_release,
2436 .open = ocfs2_dir_open,
2437 .unlocked_ioctl = ocfs2_ioctl,
2438 #ifdef CONFIG_COMPAT
2439 .compat_ioctl = ocfs2_compat_ioctl,
2442 .flock = ocfs2_flock,
2446 * POSIX-lockless variants of our file_operations.
2448 * These will be used if the underlying cluster stack does not support
2449 * posix file locking, if the user passes the "localflocks" mount
2450 * option, or if we have a local-only fs.
2452 * ocfs2_flock is in here because all stacks handle UNIX file locks,
2453 * so we still want it in the case of no stack support for
2454 * plocks. Internally, it will do the right thing when asked to ignore
2457 const struct file_operations ocfs2_fops_no_plocks = {
2458 .llseek = generic_file_llseek,
2459 .read = do_sync_read,
2460 .write = do_sync_write,
2462 .fsync = ocfs2_sync_file,
2463 .release = ocfs2_file_release,
2464 .open = ocfs2_file_open,
2465 .aio_read = ocfs2_file_aio_read,
2466 .aio_write = ocfs2_file_aio_write,
2467 .unlocked_ioctl = ocfs2_ioctl,
2468 #ifdef CONFIG_COMPAT
2469 .compat_ioctl = ocfs2_compat_ioctl,
2471 .flock = ocfs2_flock,
2472 .splice_read = ocfs2_file_splice_read,
2473 .splice_write = ocfs2_file_splice_write,
2476 const struct file_operations ocfs2_dops_no_plocks = {
2477 .llseek = generic_file_llseek,
2478 .read = generic_read_dir,
2479 .readdir = ocfs2_readdir,
2480 .fsync = ocfs2_sync_file,
2481 .release = ocfs2_dir_release,
2482 .open = ocfs2_dir_open,
2483 .unlocked_ioctl = ocfs2_ioctl,
2484 #ifdef CONFIG_COMPAT
2485 .compat_ioctl = ocfs2_compat_ioctl,
2487 .flock = ocfs2_flock,