1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* -*- mode: c; c-basic-offset: 8; -*-
3 * vim: noexpandtab sw=8 ts=8 sts=0:
7 * File open, close, extend, truncate
9 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
12 #include <linux/capability.h>
14 #include <linux/types.h>
15 #include <linux/slab.h>
16 #include <linux/highmem.h>
17 #include <linux/pagemap.h>
18 #include <linux/uio.h>
19 #include <linux/sched.h>
20 #include <linux/splice.h>
21 #include <linux/mount.h>
22 #include <linux/writeback.h>
23 #include <linux/falloc.h>
24 #include <linux/quotaops.h>
25 #include <linux/blkdev.h>
26 #include <linux/backing-dev.h>
28 #include <cluster/masklog.h>
36 #include "extent_map.h"
49 #include "refcounttree.h"
50 #include "ocfs2_trace.h"
52 #include "buffer_head_io.h"
54 static int ocfs2_init_file_private(struct inode *inode, struct file *file)
56 struct ocfs2_file_private *fp;
58 fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
63 mutex_init(&fp->fp_mutex);
64 ocfs2_file_lock_res_init(&fp->fp_flock, fp);
65 file->private_data = fp;
70 static void ocfs2_free_file_private(struct inode *inode, struct file *file)
72 struct ocfs2_file_private *fp = file->private_data;
73 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
76 ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
77 ocfs2_lock_res_free(&fp->fp_flock);
79 file->private_data = NULL;
83 static int ocfs2_file_open(struct inode *inode, struct file *file)
86 int mode = file->f_flags;
87 struct ocfs2_inode_info *oi = OCFS2_I(inode);
89 trace_ocfs2_file_open(inode, file, file->f_path.dentry,
90 (unsigned long long)oi->ip_blkno,
91 file->f_path.dentry->d_name.len,
92 file->f_path.dentry->d_name.name, mode);
94 if (file->f_mode & FMODE_WRITE) {
95 status = dquot_initialize(inode);
100 spin_lock(&oi->ip_lock);
102 /* Check that the inode hasn't been wiped from disk by another
103 * node. If it hasn't then we're safe as long as we hold the
104 * spin lock until our increment of open count. */
105 if (oi->ip_flags & OCFS2_INODE_DELETED) {
106 spin_unlock(&oi->ip_lock);
113 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
116 spin_unlock(&oi->ip_lock);
118 status = ocfs2_init_file_private(inode, file);
121 * We want to set open count back if we're failing the
124 spin_lock(&oi->ip_lock);
126 spin_unlock(&oi->ip_lock);
129 file->f_mode |= FMODE_NOWAIT;
135 static int ocfs2_file_release(struct inode *inode, struct file *file)
137 struct ocfs2_inode_info *oi = OCFS2_I(inode);
139 spin_lock(&oi->ip_lock);
140 if (!--oi->ip_open_count)
141 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
143 trace_ocfs2_file_release(inode, file, file->f_path.dentry,
145 file->f_path.dentry->d_name.len,
146 file->f_path.dentry->d_name.name,
148 spin_unlock(&oi->ip_lock);
150 ocfs2_free_file_private(inode, file);
155 static int ocfs2_dir_open(struct inode *inode, struct file *file)
157 return ocfs2_init_file_private(inode, file);
160 static int ocfs2_dir_release(struct inode *inode, struct file *file)
162 ocfs2_free_file_private(inode, file);
166 static int ocfs2_sync_file(struct file *file, loff_t start, loff_t end,
170 struct inode *inode = file->f_mapping->host;
171 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
172 struct ocfs2_inode_info *oi = OCFS2_I(inode);
173 journal_t *journal = osb->journal->j_journal;
176 bool needs_barrier = false;
178 trace_ocfs2_sync_file(inode, file, file->f_path.dentry,
180 file->f_path.dentry->d_name.len,
181 file->f_path.dentry->d_name.name,
182 (unsigned long long)datasync);
184 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
187 err = file_write_and_wait_range(file, start, end);
191 commit_tid = datasync ? oi->i_datasync_tid : oi->i_sync_tid;
192 if (journal->j_flags & JBD2_BARRIER &&
193 !jbd2_trans_will_send_data_barrier(journal, commit_tid))
194 needs_barrier = true;
195 err = jbd2_complete_transaction(journal, commit_tid);
197 ret = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
205 return (err < 0) ? -EIO : 0;
208 int ocfs2_should_update_atime(struct inode *inode,
209 struct vfsmount *vfsmnt)
211 struct timespec64 now;
212 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
214 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
217 if ((inode->i_flags & S_NOATIME) ||
218 ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode)))
222 * We can be called with no vfsmnt structure - NFSD will
225 * Note that our action here is different than touch_atime() -
226 * if we can't tell whether this is a noatime mount, then we
227 * don't know whether to trust the value of s_atime_quantum.
232 if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
233 ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
236 if (vfsmnt->mnt_flags & MNT_RELATIME) {
237 if ((timespec64_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
238 (timespec64_compare(&inode->i_atime, &inode->i_ctime) <= 0))
244 now = current_time(inode);
245 if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
251 int ocfs2_update_inode_atime(struct inode *inode,
252 struct buffer_head *bh)
255 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
257 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(inode);
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_update_inode_fsync_trans(handle, inode, 0);
282 ocfs2_journal_dirty(handle, bh);
285 ocfs2_commit_trans(osb, handle);
290 int ocfs2_set_inode_size(handle_t *handle,
292 struct buffer_head *fe_bh,
297 i_size_write(inode, new_i_size);
298 inode->i_blocks = ocfs2_inode_sector_count(inode);
299 inode->i_ctime = inode->i_mtime = current_time(inode);
301 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
311 int ocfs2_simple_size_update(struct inode *inode,
312 struct buffer_head *di_bh,
316 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
317 handle_t *handle = NULL;
319 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
320 if (IS_ERR(handle)) {
321 ret = PTR_ERR(handle);
326 ret = ocfs2_set_inode_size(handle, inode, di_bh,
331 ocfs2_update_inode_fsync_trans(handle, inode, 0);
332 ocfs2_commit_trans(osb, handle);
337 static int ocfs2_cow_file_pos(struct inode *inode,
338 struct buffer_head *fe_bh,
342 u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
343 unsigned int num_clusters = 0;
344 unsigned int ext_flags = 0;
347 * If the new offset is aligned to the range of the cluster, there is
348 * no space for ocfs2_zero_range_for_truncate to fill, so no need to
351 if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
354 status = ocfs2_get_clusters(inode, cpos, &phys,
355 &num_clusters, &ext_flags);
361 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
364 return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1);
370 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
372 struct buffer_head *fe_bh,
377 struct ocfs2_dinode *di;
381 * We need to CoW the cluster contains the offset if it is reflinked
382 * since we will call ocfs2_zero_range_for_truncate later which will
383 * write "0" from offset to the end of the cluster.
385 status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
391 /* TODO: This needs to actually orphan the inode in this
394 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
395 if (IS_ERR(handle)) {
396 status = PTR_ERR(handle);
401 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
402 OCFS2_JOURNAL_ACCESS_WRITE);
409 * Do this before setting i_size.
411 cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
412 status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
419 i_size_write(inode, new_i_size);
420 inode->i_ctime = inode->i_mtime = current_time(inode);
422 di = (struct ocfs2_dinode *) fe_bh->b_data;
423 di->i_size = cpu_to_le64(new_i_size);
424 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
425 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
426 ocfs2_update_inode_fsync_trans(handle, inode, 0);
428 ocfs2_journal_dirty(handle, fe_bh);
431 ocfs2_commit_trans(osb, handle);
436 int ocfs2_truncate_file(struct inode *inode,
437 struct buffer_head *di_bh,
441 struct ocfs2_dinode *fe = NULL;
442 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
444 /* We trust di_bh because it comes from ocfs2_inode_lock(), which
445 * already validated it */
446 fe = (struct ocfs2_dinode *) di_bh->b_data;
448 trace_ocfs2_truncate_file((unsigned long long)OCFS2_I(inode)->ip_blkno,
449 (unsigned long long)le64_to_cpu(fe->i_size),
450 (unsigned long long)new_i_size);
452 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
453 "Inode %llu, inode i_size = %lld != di "
454 "i_size = %llu, i_flags = 0x%x\n",
455 (unsigned long long)OCFS2_I(inode)->ip_blkno,
457 (unsigned long long)le64_to_cpu(fe->i_size),
458 le32_to_cpu(fe->i_flags));
460 if (new_i_size > le64_to_cpu(fe->i_size)) {
461 trace_ocfs2_truncate_file_error(
462 (unsigned long long)le64_to_cpu(fe->i_size),
463 (unsigned long long)new_i_size);
469 down_write(&OCFS2_I(inode)->ip_alloc_sem);
471 ocfs2_resv_discard(&osb->osb_la_resmap,
472 &OCFS2_I(inode)->ip_la_data_resv);
475 * The inode lock forced other nodes to sync and drop their
476 * pages, which (correctly) happens even if we have a truncate
477 * without allocation change - ocfs2 cluster sizes can be much
478 * greater than page size, so we have to truncate them
481 unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
482 truncate_inode_pages(inode->i_mapping, new_i_size);
484 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
485 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
486 i_size_read(inode), 1);
490 goto bail_unlock_sem;
493 /* alright, we're going to need to do a full blown alloc size
494 * change. Orphan the inode so that recovery can complete the
495 * truncate if necessary. This does the task of marking
497 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
500 goto bail_unlock_sem;
503 status = ocfs2_commit_truncate(osb, inode, di_bh);
506 goto bail_unlock_sem;
509 /* TODO: orphan dir cleanup here. */
511 up_write(&OCFS2_I(inode)->ip_alloc_sem);
514 if (!status && OCFS2_I(inode)->ip_clusters == 0)
515 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
521 * extend file allocation only here.
522 * we'll update all the disk stuff, and oip->alloc_size
524 * expect stuff to be locked, a transaction started and enough data /
525 * metadata reservations in the contexts.
527 * Will return -EAGAIN, and a reason if a restart is needed.
528 * If passed in, *reason will always be set, even in error.
530 int ocfs2_add_inode_data(struct ocfs2_super *osb,
535 struct buffer_head *fe_bh,
537 struct ocfs2_alloc_context *data_ac,
538 struct ocfs2_alloc_context *meta_ac,
539 enum ocfs2_alloc_restarted *reason_ret)
542 struct ocfs2_extent_tree et;
544 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
545 ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
546 clusters_to_add, mark_unwritten,
547 data_ac, meta_ac, reason_ret);
552 static int ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
553 u32 clusters_to_add, int mark_unwritten)
556 int restart_func = 0;
559 struct buffer_head *bh = NULL;
560 struct ocfs2_dinode *fe = NULL;
561 handle_t *handle = NULL;
562 struct ocfs2_alloc_context *data_ac = NULL;
563 struct ocfs2_alloc_context *meta_ac = NULL;
564 enum ocfs2_alloc_restarted why = RESTART_NONE;
565 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
566 struct ocfs2_extent_tree et;
570 * Unwritten extent only exists for file systems which
573 BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
575 status = ocfs2_read_inode_block(inode, &bh);
580 fe = (struct ocfs2_dinode *) bh->b_data;
583 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
585 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
586 status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
593 credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list);
594 handle = ocfs2_start_trans(osb, credits);
595 if (IS_ERR(handle)) {
596 status = PTR_ERR(handle);
602 restarted_transaction:
603 trace_ocfs2_extend_allocation(
604 (unsigned long long)OCFS2_I(inode)->ip_blkno,
605 (unsigned long long)i_size_read(inode),
606 le32_to_cpu(fe->i_clusters), clusters_to_add,
609 status = dquot_alloc_space_nodirty(inode,
610 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
615 /* reserve a write to the file entry early on - that we if we
616 * run out of credits in the allocation path, we can still
618 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
619 OCFS2_JOURNAL_ACCESS_WRITE);
625 prev_clusters = OCFS2_I(inode)->ip_clusters;
627 status = ocfs2_add_inode_data(osb,
637 if ((status < 0) && (status != -EAGAIN)) {
638 if (status != -ENOSPC)
642 ocfs2_update_inode_fsync_trans(handle, inode, 1);
643 ocfs2_journal_dirty(handle, bh);
645 spin_lock(&OCFS2_I(inode)->ip_lock);
646 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
647 spin_unlock(&OCFS2_I(inode)->ip_lock);
648 /* Release unused quota reservation */
649 dquot_free_space(inode,
650 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
653 if (why != RESTART_NONE && clusters_to_add) {
654 if (why == RESTART_META) {
658 BUG_ON(why != RESTART_TRANS);
660 status = ocfs2_allocate_extend_trans(handle, 1);
662 /* handle still has to be committed at
668 goto restarted_transaction;
672 trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno,
673 le32_to_cpu(fe->i_clusters),
674 (unsigned long long)le64_to_cpu(fe->i_size),
675 OCFS2_I(inode)->ip_clusters,
676 (unsigned long long)i_size_read(inode));
679 if (status < 0 && did_quota)
680 dquot_free_space(inode,
681 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
683 ocfs2_commit_trans(osb, handle);
687 ocfs2_free_alloc_context(data_ac);
691 ocfs2_free_alloc_context(meta_ac);
694 if ((!status) && restart_func) {
705 * While a write will already be ordering the data, a truncate will not.
706 * Thus, we need to explicitly order the zeroed pages.
708 static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode,
709 struct buffer_head *di_bh,
713 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
714 handle_t *handle = NULL;
717 if (!ocfs2_should_order_data(inode))
720 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
721 if (IS_ERR(handle)) {
727 ret = ocfs2_jbd2_inode_add_write(handle, inode, start_byte, length);
733 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
734 OCFS2_JOURNAL_ACCESS_WRITE);
737 ocfs2_update_inode_fsync_trans(handle, inode, 1);
742 ocfs2_commit_trans(osb, handle);
743 handle = ERR_PTR(ret);
748 /* Some parts of this taken from generic_cont_expand, which turned out
749 * to be too fragile to do exactly what we need without us having to
750 * worry about recursive locking in ->write_begin() and ->write_end(). */
751 static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
752 u64 abs_to, struct buffer_head *di_bh)
754 struct address_space *mapping = inode->i_mapping;
756 unsigned long index = abs_from >> PAGE_SHIFT;
759 unsigned zero_from, zero_to, block_start, block_end;
760 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
762 BUG_ON(abs_from >= abs_to);
763 BUG_ON(abs_to > (((u64)index + 1) << PAGE_SHIFT));
764 BUG_ON(abs_from & (inode->i_blkbits - 1));
766 handle = ocfs2_zero_start_ordered_transaction(inode, di_bh,
769 if (IS_ERR(handle)) {
770 ret = PTR_ERR(handle);
774 page = find_or_create_page(mapping, index, GFP_NOFS);
778 goto out_commit_trans;
781 /* Get the offsets within the page that we want to zero */
782 zero_from = abs_from & (PAGE_SIZE - 1);
783 zero_to = abs_to & (PAGE_SIZE - 1);
787 trace_ocfs2_write_zero_page(
788 (unsigned long long)OCFS2_I(inode)->ip_blkno,
789 (unsigned long long)abs_from,
790 (unsigned long long)abs_to,
791 index, zero_from, zero_to);
793 /* We know that zero_from is block aligned */
794 for (block_start = zero_from; block_start < zero_to;
795 block_start = block_end) {
796 block_end = block_start + i_blocksize(inode);
799 * block_start is block-aligned. Bump it by one to force
800 * __block_write_begin and block_commit_write to zero the
803 ret = __block_write_begin(page, block_start + 1, 0,
811 /* must not update i_size! */
812 ret = block_commit_write(page, block_start + 1,
821 * fs-writeback will release the dirty pages without page lock
822 * whose offset are over inode size, the release happens at
823 * block_write_full_page().
825 i_size_write(inode, abs_to);
826 inode->i_blocks = ocfs2_inode_sector_count(inode);
827 di->i_size = cpu_to_le64((u64)i_size_read(inode));
828 inode->i_mtime = inode->i_ctime = current_time(inode);
829 di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
830 di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
831 di->i_mtime_nsec = di->i_ctime_nsec;
833 ocfs2_journal_dirty(handle, di_bh);
834 ocfs2_update_inode_fsync_trans(handle, inode, 1);
842 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
848 * Find the next range to zero. We do this in terms of bytes because
849 * that's what ocfs2_zero_extend() wants, and it is dealing with the
850 * pagecache. We may return multiple extents.
852 * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
853 * needs to be zeroed. range_start and range_end return the next zeroing
854 * range. A subsequent call should pass the previous range_end as its
855 * zero_start. If range_end is 0, there's nothing to do.
857 * Unwritten extents are skipped over. Refcounted extents are CoWd.
859 static int ocfs2_zero_extend_get_range(struct inode *inode,
860 struct buffer_head *di_bh,
861 u64 zero_start, u64 zero_end,
862 u64 *range_start, u64 *range_end)
864 int rc = 0, needs_cow = 0;
865 u32 p_cpos, zero_clusters = 0;
867 zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
868 u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
869 unsigned int num_clusters = 0;
870 unsigned int ext_flags = 0;
872 while (zero_cpos < last_cpos) {
873 rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
874 &num_clusters, &ext_flags);
880 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
881 zero_clusters = num_clusters;
882 if (ext_flags & OCFS2_EXT_REFCOUNTED)
887 zero_cpos += num_clusters;
889 if (!zero_clusters) {
894 while ((zero_cpos + zero_clusters) < last_cpos) {
895 rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
896 &p_cpos, &num_clusters,
903 if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
905 if (ext_flags & OCFS2_EXT_REFCOUNTED)
907 zero_clusters += num_clusters;
909 if ((zero_cpos + zero_clusters) > last_cpos)
910 zero_clusters = last_cpos - zero_cpos;
913 rc = ocfs2_refcount_cow(inode, di_bh, zero_cpos,
914 zero_clusters, UINT_MAX);
921 *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
922 *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
923 zero_cpos + zero_clusters);
930 * Zero one range returned from ocfs2_zero_extend_get_range(). The caller
931 * has made sure that the entire range needs zeroing.
933 static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
934 u64 range_end, struct buffer_head *di_bh)
938 u64 zero_pos = range_start;
940 trace_ocfs2_zero_extend_range(
941 (unsigned long long)OCFS2_I(inode)->ip_blkno,
942 (unsigned long long)range_start,
943 (unsigned long long)range_end);
944 BUG_ON(range_start >= range_end);
946 while (zero_pos < range_end) {
947 next_pos = (zero_pos & PAGE_MASK) + PAGE_SIZE;
948 if (next_pos > range_end)
949 next_pos = range_end;
950 rc = ocfs2_write_zero_page(inode, zero_pos, next_pos, di_bh);
958 * Very large extends have the potential to lock up
959 * the cpu for extended periods of time.
967 int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
971 u64 zero_start, range_start = 0, range_end = 0;
972 struct super_block *sb = inode->i_sb;
974 zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
975 trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
976 (unsigned long long)zero_start,
977 (unsigned long long)i_size_read(inode));
978 while (zero_start < zero_to_size) {
979 ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
990 if (range_start < zero_start)
991 range_start = zero_start;
992 if (range_end > zero_to_size)
993 range_end = zero_to_size;
995 ret = ocfs2_zero_extend_range(inode, range_start,
1001 zero_start = range_end;
1007 int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
1008 u64 new_i_size, u64 zero_to)
1011 u32 clusters_to_add;
1012 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1015 * Only quota files call this without a bh, and they can't be
1018 BUG_ON(!di_bh && ocfs2_is_refcount_inode(inode));
1019 BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1021 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1022 if (clusters_to_add < oi->ip_clusters)
1023 clusters_to_add = 0;
1025 clusters_to_add -= oi->ip_clusters;
1027 if (clusters_to_add) {
1028 ret = ocfs2_extend_allocation(inode, oi->ip_clusters,
1029 clusters_to_add, 0);
1037 * Call this even if we don't add any clusters to the tree. We
1038 * still need to zero the area between the old i_size and the
1041 ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1049 static int ocfs2_extend_file(struct inode *inode,
1050 struct buffer_head *di_bh,
1054 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1058 /* setattr sometimes calls us like this. */
1059 if (new_i_size == 0)
1062 if (i_size_read(inode) == new_i_size)
1064 BUG_ON(new_i_size < i_size_read(inode));
1067 * The alloc sem blocks people in read/write from reading our
1068 * allocation until we're done changing it. We depend on
1069 * i_mutex to block other extend/truncate calls while we're
1070 * here. We even have to hold it for sparse files because there
1071 * might be some tail zeroing.
1073 down_write(&oi->ip_alloc_sem);
1075 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1077 * We can optimize small extends by keeping the inodes
1080 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1081 up_write(&oi->ip_alloc_sem);
1082 goto out_update_size;
1085 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1087 up_write(&oi->ip_alloc_sem);
1093 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1094 ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1096 ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1099 up_write(&oi->ip_alloc_sem);
1107 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1115 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1117 int status = 0, size_change;
1118 int inode_locked = 0;
1119 struct inode *inode = d_inode(dentry);
1120 struct super_block *sb = inode->i_sb;
1121 struct ocfs2_super *osb = OCFS2_SB(sb);
1122 struct buffer_head *bh = NULL;
1123 handle_t *handle = NULL;
1124 struct dquot *transfer_to[MAXQUOTAS] = { };
1127 struct ocfs2_lock_holder oh;
1129 trace_ocfs2_setattr(inode, dentry,
1130 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1131 dentry->d_name.len, dentry->d_name.name,
1132 attr->ia_valid, attr->ia_mode,
1133 from_kuid(&init_user_ns, attr->ia_uid),
1134 from_kgid(&init_user_ns, attr->ia_gid));
1136 /* ensuring we don't even attempt to truncate a symlink */
1137 if (S_ISLNK(inode->i_mode))
1138 attr->ia_valid &= ~ATTR_SIZE;
1140 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1141 | ATTR_GID | ATTR_UID | ATTR_MODE)
1142 if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
1145 status = setattr_prepare(dentry, attr);
1149 if (is_quota_modification(inode, attr)) {
1150 status = dquot_initialize(inode);
1154 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1157 * Here we should wait dio to finish before inode lock
1158 * to avoid a deadlock between ocfs2_setattr() and
1159 * ocfs2_dio_end_io_write()
1161 inode_dio_wait(inode);
1163 status = ocfs2_rw_lock(inode, 1);
1170 had_lock = ocfs2_inode_lock_tracker(inode, &bh, 1, &oh);
1173 goto bail_unlock_rw;
1174 } else if (had_lock) {
1176 * As far as we know, ocfs2_setattr() could only be the first
1177 * VFS entry point in the call chain of recursive cluster
1185 * ocfs2_iop_get_acl()
1187 * But, we're not 100% sure if it's always true, because the
1188 * ordering of the VFS entry points in the call chain is out
1189 * of our control. So, we'd better dump the stack here to
1190 * catch the other cases of recursive locking.
1192 mlog(ML_ERROR, "Another case of recursive locking:\n");
1198 status = inode_newsize_ok(inode, attr->ia_size);
1202 if (i_size_read(inode) >= attr->ia_size) {
1203 if (ocfs2_should_order_data(inode)) {
1204 status = ocfs2_begin_ordered_truncate(inode,
1209 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1211 status = ocfs2_extend_file(inode, bh, attr->ia_size);
1213 if (status != -ENOSPC)
1220 if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
1221 (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
1223 * Gather pointers to quota structures so that allocation /
1224 * freeing of quota structures happens here and not inside
1225 * dquot_transfer() where we have problems with lock ordering
1227 if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)
1228 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1229 OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1230 transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid));
1231 if (IS_ERR(transfer_to[USRQUOTA])) {
1232 status = PTR_ERR(transfer_to[USRQUOTA]);
1236 if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid)
1237 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1238 OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1239 transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid));
1240 if (IS_ERR(transfer_to[GRPQUOTA])) {
1241 status = PTR_ERR(transfer_to[GRPQUOTA]);
1245 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1246 2 * ocfs2_quota_trans_credits(sb));
1247 if (IS_ERR(handle)) {
1248 status = PTR_ERR(handle);
1252 status = __dquot_transfer(inode, transfer_to);
1256 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1257 if (IS_ERR(handle)) {
1258 status = PTR_ERR(handle);
1264 setattr_copy(inode, attr);
1265 mark_inode_dirty(inode);
1267 status = ocfs2_mark_inode_dirty(handle, inode, bh);
1272 ocfs2_commit_trans(osb, handle);
1274 if (status && inode_locked) {
1275 ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
1280 ocfs2_rw_unlock(inode, 1);
1283 /* Release quota pointers in case we acquired them */
1284 for (qtype = 0; qtype < OCFS2_MAXQUOTAS; qtype++)
1285 dqput(transfer_to[qtype]);
1287 if (!status && attr->ia_valid & ATTR_MODE) {
1288 status = ocfs2_acl_chmod(inode, bh);
1293 ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
1299 int ocfs2_getattr(const struct path *path, struct kstat *stat,
1300 u32 request_mask, unsigned int flags)
1302 struct inode *inode = d_inode(path->dentry);
1303 struct super_block *sb = path->dentry->d_sb;
1304 struct ocfs2_super *osb = sb->s_fs_info;
1307 err = ocfs2_inode_revalidate(path->dentry);
1314 generic_fillattr(inode, stat);
1316 * If there is inline data in the inode, the inode will normally not
1317 * have data blocks allocated (it may have an external xattr block).
1318 * Report at least one sector for such files, so tools like tar, rsync,
1319 * others don't incorrectly think the file is completely sparse.
1321 if (unlikely(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1322 stat->blocks += (stat->size + 511)>>9;
1324 /* We set the blksize from the cluster size for performance */
1325 stat->blksize = osb->s_clustersize;
1331 int ocfs2_permission(struct inode *inode, int mask)
1334 struct ocfs2_lock_holder oh;
1336 if (mask & MAY_NOT_BLOCK)
1339 had_lock = ocfs2_inode_lock_tracker(inode, NULL, 0, &oh);
1343 } else if (had_lock) {
1344 /* See comments in ocfs2_setattr() for details.
1345 * The call chain of this case could be:
1348 * inode_permission()
1349 * ocfs2_permission()
1350 * ocfs2_iop_get_acl()
1352 mlog(ML_ERROR, "Another case of recursive locking:\n");
1356 ret = generic_permission(inode, mask);
1358 ocfs2_inode_unlock_tracker(inode, 0, &oh, had_lock);
1363 static int __ocfs2_write_remove_suid(struct inode *inode,
1364 struct buffer_head *bh)
1368 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1369 struct ocfs2_dinode *di;
1371 trace_ocfs2_write_remove_suid(
1372 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1375 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1376 if (IS_ERR(handle)) {
1377 ret = PTR_ERR(handle);
1382 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1383 OCFS2_JOURNAL_ACCESS_WRITE);
1389 inode->i_mode &= ~S_ISUID;
1390 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1391 inode->i_mode &= ~S_ISGID;
1393 di = (struct ocfs2_dinode *) bh->b_data;
1394 di->i_mode = cpu_to_le16(inode->i_mode);
1395 ocfs2_update_inode_fsync_trans(handle, inode, 0);
1397 ocfs2_journal_dirty(handle, bh);
1400 ocfs2_commit_trans(osb, handle);
1405 static int ocfs2_write_remove_suid(struct inode *inode)
1408 struct buffer_head *bh = NULL;
1410 ret = ocfs2_read_inode_block(inode, &bh);
1416 ret = __ocfs2_write_remove_suid(inode, bh);
1423 * Allocate enough extents to cover the region starting at byte offset
1424 * start for len bytes. Existing extents are skipped, any extents
1425 * added are marked as "unwritten".
1427 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1431 u32 cpos, phys_cpos, clusters, alloc_size;
1432 u64 end = start + len;
1433 struct buffer_head *di_bh = NULL;
1435 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1436 ret = ocfs2_read_inode_block(inode, &di_bh);
1443 * Nothing to do if the requested reservation range
1444 * fits within the inode.
1446 if (ocfs2_size_fits_inline_data(di_bh, end))
1449 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1457 * We consider both start and len to be inclusive.
1459 cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1460 clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1464 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1472 * Hole or existing extent len can be arbitrary, so
1473 * cap it to our own allocation request.
1475 if (alloc_size > clusters)
1476 alloc_size = clusters;
1480 * We already have an allocation at this
1481 * region so we can safely skip it.
1486 ret = ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1495 clusters -= alloc_size;
1506 * Truncate a byte range, avoiding pages within partial clusters. This
1507 * preserves those pages for the zeroing code to write to.
1509 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1512 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1514 struct address_space *mapping = inode->i_mapping;
1516 start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1517 end = byte_start + byte_len;
1518 end = end & ~(osb->s_clustersize - 1);
1521 unmap_mapping_range(mapping, start, end - start, 0);
1522 truncate_inode_pages_range(mapping, start, end - 1);
1526 static int ocfs2_zero_partial_clusters(struct inode *inode,
1531 u64 end = start + len;
1532 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1533 unsigned int csize = osb->s_clustersize;
1537 * The "start" and "end" values are NOT necessarily part of
1538 * the range whose allocation is being deleted. Rather, this
1539 * is what the user passed in with the request. We must zero
1540 * partial clusters here. There's no need to worry about
1541 * physical allocation - the zeroing code knows to skip holes.
1543 trace_ocfs2_zero_partial_clusters(
1544 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1545 (unsigned long long)start, (unsigned long long)end);
1548 * If both edges are on a cluster boundary then there's no
1549 * zeroing required as the region is part of the allocation to
1552 if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1555 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1556 if (IS_ERR(handle)) {
1557 ret = PTR_ERR(handle);
1563 * If start is on a cluster boundary and end is somewhere in another
1564 * cluster, we have not COWed the cluster starting at start, unless
1565 * end is also within the same cluster. So, in this case, we skip this
1566 * first call to ocfs2_zero_range_for_truncate() truncate and move on
1569 if ((start & (csize - 1)) != 0) {
1571 * We want to get the byte offset of the end of the 1st
1574 tmpend = (u64)osb->s_clustersize +
1575 (start & ~(osb->s_clustersize - 1));
1579 trace_ocfs2_zero_partial_clusters_range1(
1580 (unsigned long long)start,
1581 (unsigned long long)tmpend);
1583 ret = ocfs2_zero_range_for_truncate(inode, handle, start,
1591 * This may make start and end equal, but the zeroing
1592 * code will skip any work in that case so there's no
1593 * need to catch it up here.
1595 start = end & ~(osb->s_clustersize - 1);
1597 trace_ocfs2_zero_partial_clusters_range2(
1598 (unsigned long long)start, (unsigned long long)end);
1600 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1604 ocfs2_update_inode_fsync_trans(handle, inode, 1);
1606 ocfs2_commit_trans(osb, handle);
1611 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1614 struct ocfs2_extent_rec *rec = NULL;
1616 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1618 rec = &el->l_recs[i];
1620 if (le32_to_cpu(rec->e_cpos) < pos)
1628 * Helper to calculate the punching pos and length in one run, we handle the
1629 * following three cases in order:
1631 * - remove the entire record
1632 * - remove a partial record
1633 * - no record needs to be removed (hole-punching completed)
1635 static void ocfs2_calc_trunc_pos(struct inode *inode,
1636 struct ocfs2_extent_list *el,
1637 struct ocfs2_extent_rec *rec,
1638 u32 trunc_start, u32 *trunc_cpos,
1639 u32 *trunc_len, u32 *trunc_end,
1640 u64 *blkno, int *done)
1645 range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1647 if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1649 * remove an entire extent record.
1651 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1653 * Skip holes if any.
1655 if (range < *trunc_end)
1657 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1658 *blkno = le64_to_cpu(rec->e_blkno);
1659 *trunc_end = le32_to_cpu(rec->e_cpos);
1660 } else if (range > trunc_start) {
1662 * remove a partial extent record, which means we're
1663 * removing the last extent record.
1665 *trunc_cpos = trunc_start;
1669 if (range < *trunc_end)
1671 *trunc_len = *trunc_end - trunc_start;
1672 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1673 *blkno = le64_to_cpu(rec->e_blkno) +
1674 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1675 *trunc_end = trunc_start;
1678 * It may have two following possibilities:
1680 * - last record has been removed
1681 * - trunc_start was within a hole
1683 * both two cases mean the completion of hole punching.
1691 int ocfs2_remove_inode_range(struct inode *inode,
1692 struct buffer_head *di_bh, u64 byte_start,
1695 int ret = 0, flags = 0, done = 0, i;
1696 u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1698 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1699 struct ocfs2_cached_dealloc_ctxt dealloc;
1700 struct address_space *mapping = inode->i_mapping;
1701 struct ocfs2_extent_tree et;
1702 struct ocfs2_path *path = NULL;
1703 struct ocfs2_extent_list *el = NULL;
1704 struct ocfs2_extent_rec *rec = NULL;
1705 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1706 u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1708 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1709 ocfs2_init_dealloc_ctxt(&dealloc);
1711 trace_ocfs2_remove_inode_range(
1712 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1713 (unsigned long long)byte_start,
1714 (unsigned long long)byte_len);
1719 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1720 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1721 byte_start + byte_len, 0);
1727 * There's no need to get fancy with the page cache
1728 * truncate of an inline-data inode. We're talking
1729 * about less than a page here, which will be cached
1730 * in the dinode buffer anyway.
1732 unmap_mapping_range(mapping, 0, 0, 0);
1733 truncate_inode_pages(mapping, 0);
1738 * For reflinks, we may need to CoW 2 clusters which might be
1739 * partially zero'd later, if hole's start and end offset were
1740 * within one cluster(means is not exactly aligned to clustersize).
1743 if (ocfs2_is_refcount_inode(inode)) {
1744 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1750 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1757 trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1758 trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1759 cluster_in_el = trunc_end;
1761 ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1767 path = ocfs2_new_path_from_et(&et);
1774 while (trunc_end > trunc_start) {
1776 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1783 el = path_leaf_el(path);
1785 i = ocfs2_find_rec(el, trunc_end);
1787 * Need to go to previous extent block.
1790 if (path->p_tree_depth == 0)
1793 ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1802 * We've reached the leftmost extent block,
1803 * it's safe to leave.
1805 if (cluster_in_el == 0)
1809 * The 'pos' searched for previous extent block is
1810 * always one cluster less than actual trunc_end.
1812 trunc_end = cluster_in_el + 1;
1814 ocfs2_reinit_path(path, 1);
1819 rec = &el->l_recs[i];
1821 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1822 &trunc_len, &trunc_end, &blkno, &done);
1826 flags = rec->e_flags;
1827 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1829 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1830 phys_cpos, trunc_len, flags,
1831 &dealloc, refcount_loc, false);
1837 cluster_in_el = trunc_end;
1839 ocfs2_reinit_path(path, 1);
1842 ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1845 ocfs2_free_path(path);
1846 ocfs2_schedule_truncate_log_flush(osb, 1);
1847 ocfs2_run_deallocs(osb, &dealloc);
1853 * Parts of this function taken from xfs_change_file_space()
1855 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1856 loff_t f_pos, unsigned int cmd,
1857 struct ocfs2_space_resv *sr,
1863 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1864 struct buffer_head *di_bh = NULL;
1866 unsigned long long max_off = inode->i_sb->s_maxbytes;
1868 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1874 * This prevents concurrent writes on other nodes
1876 ret = ocfs2_rw_lock(inode, 1);
1882 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1888 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1890 goto out_inode_unlock;
1893 switch (sr->l_whence) {
1894 case 0: /*SEEK_SET*/
1896 case 1: /*SEEK_CUR*/
1897 sr->l_start += f_pos;
1899 case 2: /*SEEK_END*/
1900 sr->l_start += i_size_read(inode);
1904 goto out_inode_unlock;
1908 llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1911 || sr->l_start > max_off
1912 || (sr->l_start + llen) < 0
1913 || (sr->l_start + llen) > max_off) {
1915 goto out_inode_unlock;
1917 size = sr->l_start + sr->l_len;
1919 if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64 ||
1920 cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) {
1921 if (sr->l_len <= 0) {
1923 goto out_inode_unlock;
1927 if (file && should_remove_suid(file->f_path.dentry)) {
1928 ret = __ocfs2_write_remove_suid(inode, di_bh);
1931 goto out_inode_unlock;
1935 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1937 case OCFS2_IOC_RESVSP:
1938 case OCFS2_IOC_RESVSP64:
1940 * This takes unsigned offsets, but the signed ones we
1941 * pass have been checked against overflow above.
1943 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1946 case OCFS2_IOC_UNRESVSP:
1947 case OCFS2_IOC_UNRESVSP64:
1948 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1954 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1957 goto out_inode_unlock;
1961 * We update c/mtime for these changes
1963 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1964 if (IS_ERR(handle)) {
1965 ret = PTR_ERR(handle);
1967 goto out_inode_unlock;
1970 if (change_size && i_size_read(inode) < size)
1971 i_size_write(inode, size);
1973 inode->i_ctime = inode->i_mtime = current_time(inode);
1974 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1978 if (file && (file->f_flags & O_SYNC))
1981 ocfs2_commit_trans(osb, handle);
1985 ocfs2_inode_unlock(inode, 1);
1987 ocfs2_rw_unlock(inode, 1);
1990 inode_unlock(inode);
1994 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1995 struct ocfs2_space_resv *sr)
1997 struct inode *inode = file_inode(file);
1998 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2001 if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
2002 !ocfs2_writes_unwritten_extents(osb))
2004 else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
2005 !ocfs2_sparse_alloc(osb))
2008 if (!S_ISREG(inode->i_mode))
2011 if (!(file->f_mode & FMODE_WRITE))
2014 ret = mnt_want_write_file(file);
2017 ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
2018 mnt_drop_write_file(file);
2022 static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
2025 struct inode *inode = file_inode(file);
2026 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2027 struct ocfs2_space_resv sr;
2028 int change_size = 1;
2029 int cmd = OCFS2_IOC_RESVSP64;
2031 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2033 if (!ocfs2_writes_unwritten_extents(osb))
2036 if (mode & FALLOC_FL_KEEP_SIZE)
2039 if (mode & FALLOC_FL_PUNCH_HOLE)
2040 cmd = OCFS2_IOC_UNRESVSP64;
2043 sr.l_start = (s64)offset;
2044 sr.l_len = (s64)len;
2046 return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2050 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2054 unsigned int extent_flags;
2055 u32 cpos, clusters, extent_len, phys_cpos;
2056 struct super_block *sb = inode->i_sb;
2058 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2059 !ocfs2_is_refcount_inode(inode) ||
2060 OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2063 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2064 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2067 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2074 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2079 if (extent_len > clusters)
2080 extent_len = clusters;
2082 clusters -= extent_len;
2089 static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
2091 int blockmask = inode->i_sb->s_blocksize - 1;
2092 loff_t final_size = pos + count;
2094 if ((pos & blockmask) || (final_size & blockmask))
2099 static int ocfs2_prepare_inode_for_refcount(struct inode *inode,
2101 loff_t pos, size_t count,
2105 struct buffer_head *di_bh = NULL;
2106 u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2108 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2110 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2118 ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
2126 static int ocfs2_prepare_inode_for_write(struct file *file,
2127 loff_t pos, size_t count, int wait)
2129 int ret = 0, meta_level = 0, overwrite_io = 0;
2130 struct dentry *dentry = file->f_path.dentry;
2131 struct inode *inode = d_inode(dentry);
2132 struct buffer_head *di_bh = NULL;
2135 * We start with a read level meta lock and only jump to an ex
2136 * if we need to make modifications here.
2140 ret = ocfs2_inode_lock(inode, NULL, meta_level);
2142 ret = ocfs2_try_inode_lock(inode,
2143 overwrite_io ? NULL : &di_bh, meta_level);
2152 * Check if IO will overwrite allocated blocks in case
2153 * IOCB_NOWAIT flag is set.
2155 if (!wait && !overwrite_io) {
2157 if (!down_read_trylock(&OCFS2_I(inode)->ip_alloc_sem)) {
2162 ret = ocfs2_overwrite_io(inode, di_bh, pos, count);
2165 up_read(&OCFS2_I(inode)->ip_alloc_sem);
2173 /* Clear suid / sgid if necessary. We do this here
2174 * instead of later in the write path because
2175 * remove_suid() calls ->setattr without any hint that
2176 * we may have already done our cluster locking. Since
2177 * ocfs2_setattr() *must* take cluster locks to
2178 * proceed, this will lead us to recursively lock the
2179 * inode. There's also the dinode i_size state which
2180 * can be lost via setattr during extending writes (we
2181 * set inode->i_size at the end of a write. */
2182 if (should_remove_suid(dentry)) {
2183 if (meta_level == 0) {
2184 ocfs2_inode_unlock(inode, meta_level);
2189 ret = ocfs2_write_remove_suid(inode);
2196 ret = ocfs2_check_range_for_refcount(inode, pos, count);
2198 ocfs2_inode_unlock(inode, meta_level);
2201 ret = ocfs2_prepare_inode_for_refcount(inode,
2217 trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
2222 if (meta_level >= 0)
2223 ocfs2_inode_unlock(inode, meta_level);
2229 static ssize_t ocfs2_file_write_iter(struct kiocb *iocb,
2230 struct iov_iter *from)
2233 ssize_t written = 0;
2235 size_t count = iov_iter_count(from);
2236 struct file *file = iocb->ki_filp;
2237 struct inode *inode = file_inode(file);
2238 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2239 int full_coherency = !(osb->s_mount_opt &
2240 OCFS2_MOUNT_COHERENCY_BUFFERED);
2241 void *saved_ki_complete = NULL;
2242 int append_write = ((iocb->ki_pos + count) >=
2243 i_size_read(inode) ? 1 : 0);
2244 int direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2245 int nowait = iocb->ki_flags & IOCB_NOWAIT ? 1 : 0;
2247 trace_ocfs2_file_write_iter(inode, file, file->f_path.dentry,
2248 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2249 file->f_path.dentry->d_name.len,
2250 file->f_path.dentry->d_name.name,
2251 (unsigned int)from->nr_segs); /* GRRRRR */
2253 if (!direct_io && nowait)
2260 if (!inode_trylock(inode))
2266 * Concurrent O_DIRECT writes are allowed with
2267 * mount_option "coherency=buffered".
2268 * For append write, we must take rw EX.
2270 rw_level = (!direct_io || full_coherency || append_write);
2273 ret = ocfs2_try_rw_lock(inode, rw_level);
2275 ret = ocfs2_rw_lock(inode, rw_level);
2283 * O_DIRECT writes with "coherency=full" need to take EX cluster
2284 * inode_lock to guarantee coherency.
2286 if (direct_io && full_coherency) {
2288 * We need to take and drop the inode lock to force
2289 * other nodes to drop their caches. Buffered I/O
2290 * already does this in write_begin().
2293 ret = ocfs2_try_inode_lock(inode, NULL, 1);
2295 ret = ocfs2_inode_lock(inode, NULL, 1);
2302 ocfs2_inode_unlock(inode, 1);
2305 ret = generic_write_checks(iocb, from);
2313 ret = ocfs2_prepare_inode_for_write(file, iocb->ki_pos, count, !nowait);
2320 if (direct_io && !is_sync_kiocb(iocb) &&
2321 ocfs2_is_io_unaligned(inode, count, iocb->ki_pos)) {
2323 * Make it a sync io if it's an unaligned aio.
2325 saved_ki_complete = xchg(&iocb->ki_complete, NULL);
2328 /* communicate with ocfs2_dio_end_io */
2329 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2331 written = __generic_file_write_iter(iocb, from);
2332 /* buffered aio wouldn't have proper lock coverage today */
2333 BUG_ON(written == -EIOCBQUEUED && !direct_io);
2336 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2337 * function pointer which is called when o_direct io completes so that
2338 * it can unlock our rw lock.
2339 * Unfortunately there are error cases which call end_io and others
2340 * that don't. so we don't have to unlock the rw_lock if either an
2341 * async dio is going to do it in the future or an end_io after an
2342 * error has already done it.
2344 if ((written == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2348 if (unlikely(written <= 0))
2351 if (((file->f_flags & O_DSYNC) && !direct_io) ||
2353 ret = filemap_fdatawrite_range(file->f_mapping,
2354 iocb->ki_pos - written,
2360 ret = jbd2_journal_force_commit(osb->journal->j_journal);
2366 ret = filemap_fdatawait_range(file->f_mapping,
2367 iocb->ki_pos - written,
2372 if (saved_ki_complete)
2373 xchg(&iocb->ki_complete, saved_ki_complete);
2376 ocfs2_rw_unlock(inode, rw_level);
2379 inode_unlock(inode);
2386 static ssize_t ocfs2_file_read_iter(struct kiocb *iocb,
2387 struct iov_iter *to)
2389 int ret = 0, rw_level = -1, lock_level = 0;
2390 struct file *filp = iocb->ki_filp;
2391 struct inode *inode = file_inode(filp);
2392 int direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2393 int nowait = iocb->ki_flags & IOCB_NOWAIT ? 1 : 0;
2395 trace_ocfs2_file_read_iter(inode, filp, filp->f_path.dentry,
2396 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2397 filp->f_path.dentry->d_name.len,
2398 filp->f_path.dentry->d_name.name,
2399 to->nr_segs); /* GRRRRR */
2408 if (!direct_io && nowait)
2412 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
2413 * need locks to protect pending reads from racing with truncate.
2417 ret = ocfs2_try_rw_lock(inode, 0);
2419 ret = ocfs2_rw_lock(inode, 0);
2427 /* communicate with ocfs2_dio_end_io */
2428 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2432 * We're fine letting folks race truncates and extending
2433 * writes with read across the cluster, just like they can
2434 * locally. Hence no rw_lock during read.
2436 * Take and drop the meta data lock to update inode fields
2437 * like i_size. This allows the checks down below
2438 * generic_file_read_iter() a chance of actually working.
2440 ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level,
2447 ocfs2_inode_unlock(inode, lock_level);
2449 ret = generic_file_read_iter(iocb, to);
2450 trace_generic_file_read_iter_ret(ret);
2452 /* buffered aio wouldn't have proper lock coverage today */
2453 BUG_ON(ret == -EIOCBQUEUED && !direct_io);
2455 /* see ocfs2_file_write_iter */
2456 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2462 ocfs2_rw_unlock(inode, rw_level);
2467 /* Refer generic_file_llseek_unlocked() */
2468 static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence)
2470 struct inode *inode = file->f_mapping->host;
2479 /* SEEK_END requires the OCFS2 inode lock for the file
2480 * because it references the file's size.
2482 ret = ocfs2_inode_lock(inode, NULL, 0);
2487 offset += i_size_read(inode);
2488 ocfs2_inode_unlock(inode, 0);
2492 offset = file->f_pos;
2495 offset += file->f_pos;
2499 ret = ocfs2_seek_data_hole_offset(file, &offset, whence);
2508 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2511 inode_unlock(inode);
2517 static loff_t ocfs2_remap_file_range(struct file *file_in, loff_t pos_in,
2518 struct file *file_out, loff_t pos_out,
2519 loff_t len, unsigned int remap_flags)
2521 struct inode *inode_in = file_inode(file_in);
2522 struct inode *inode_out = file_inode(file_out);
2523 struct ocfs2_super *osb = OCFS2_SB(inode_in->i_sb);
2524 struct buffer_head *in_bh = NULL, *out_bh = NULL;
2525 bool same_inode = (inode_in == inode_out);
2526 loff_t remapped = 0;
2529 if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
2531 if (!ocfs2_refcount_tree(osb))
2533 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
2536 /* Lock both files against IO */
2537 ret = ocfs2_reflink_inodes_lock(inode_in, &in_bh, inode_out, &out_bh);
2541 /* Check file eligibility and prepare for block sharing. */
2543 if ((OCFS2_I(inode_in)->ip_flags & OCFS2_INODE_SYSTEM_FILE) ||
2544 (OCFS2_I(inode_out)->ip_flags & OCFS2_INODE_SYSTEM_FILE))
2547 ret = generic_remap_file_range_prep(file_in, pos_in, file_out, pos_out,
2549 if (ret < 0 || len == 0)
2552 /* Lock out changes to the allocation maps and remap. */
2553 down_write(&OCFS2_I(inode_in)->ip_alloc_sem);
2555 down_write_nested(&OCFS2_I(inode_out)->ip_alloc_sem,
2556 SINGLE_DEPTH_NESTING);
2558 /* Zap any page cache for the destination file's range. */
2559 truncate_inode_pages_range(&inode_out->i_data,
2560 round_down(pos_out, PAGE_SIZE),
2561 round_up(pos_out + len, PAGE_SIZE) - 1);
2563 remapped = ocfs2_reflink_remap_blocks(inode_in, in_bh, pos_in,
2564 inode_out, out_bh, pos_out, len);
2565 up_write(&OCFS2_I(inode_in)->ip_alloc_sem);
2567 up_write(&OCFS2_I(inode_out)->ip_alloc_sem);
2575 * Empty the extent map so that we may get the right extent
2576 * record from the disk.
2578 ocfs2_extent_map_trunc(inode_in, 0);
2579 ocfs2_extent_map_trunc(inode_out, 0);
2581 ret = ocfs2_reflink_update_dest(inode_out, out_bh, pos_out + len);
2588 ocfs2_reflink_inodes_unlock(inode_in, in_bh, inode_out, out_bh);
2589 return remapped > 0 ? remapped : ret;
2592 const struct inode_operations ocfs2_file_iops = {
2593 .setattr = ocfs2_setattr,
2594 .getattr = ocfs2_getattr,
2595 .permission = ocfs2_permission,
2596 .listxattr = ocfs2_listxattr,
2597 .fiemap = ocfs2_fiemap,
2598 .get_acl = ocfs2_iop_get_acl,
2599 .set_acl = ocfs2_iop_set_acl,
2602 const struct inode_operations ocfs2_special_file_iops = {
2603 .setattr = ocfs2_setattr,
2604 .getattr = ocfs2_getattr,
2605 .permission = ocfs2_permission,
2606 .get_acl = ocfs2_iop_get_acl,
2607 .set_acl = ocfs2_iop_set_acl,
2611 * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2612 * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2614 const struct file_operations ocfs2_fops = {
2615 .llseek = ocfs2_file_llseek,
2617 .fsync = ocfs2_sync_file,
2618 .release = ocfs2_file_release,
2619 .open = ocfs2_file_open,
2620 .read_iter = ocfs2_file_read_iter,
2621 .write_iter = ocfs2_file_write_iter,
2622 .unlocked_ioctl = ocfs2_ioctl,
2623 #ifdef CONFIG_COMPAT
2624 .compat_ioctl = ocfs2_compat_ioctl,
2627 .flock = ocfs2_flock,
2628 .splice_read = generic_file_splice_read,
2629 .splice_write = iter_file_splice_write,
2630 .fallocate = ocfs2_fallocate,
2631 .remap_file_range = ocfs2_remap_file_range,
2634 const struct file_operations ocfs2_dops = {
2635 .llseek = generic_file_llseek,
2636 .read = generic_read_dir,
2637 .iterate = ocfs2_readdir,
2638 .fsync = ocfs2_sync_file,
2639 .release = ocfs2_dir_release,
2640 .open = ocfs2_dir_open,
2641 .unlocked_ioctl = ocfs2_ioctl,
2642 #ifdef CONFIG_COMPAT
2643 .compat_ioctl = ocfs2_compat_ioctl,
2646 .flock = ocfs2_flock,
2650 * POSIX-lockless variants of our file_operations.
2652 * These will be used if the underlying cluster stack does not support
2653 * posix file locking, if the user passes the "localflocks" mount
2654 * option, or if we have a local-only fs.
2656 * ocfs2_flock is in here because all stacks handle UNIX file locks,
2657 * so we still want it in the case of no stack support for
2658 * plocks. Internally, it will do the right thing when asked to ignore
2661 const struct file_operations ocfs2_fops_no_plocks = {
2662 .llseek = ocfs2_file_llseek,
2664 .fsync = ocfs2_sync_file,
2665 .release = ocfs2_file_release,
2666 .open = ocfs2_file_open,
2667 .read_iter = ocfs2_file_read_iter,
2668 .write_iter = ocfs2_file_write_iter,
2669 .unlocked_ioctl = ocfs2_ioctl,
2670 #ifdef CONFIG_COMPAT
2671 .compat_ioctl = ocfs2_compat_ioctl,
2673 .flock = ocfs2_flock,
2674 .splice_read = generic_file_splice_read,
2675 .splice_write = iter_file_splice_write,
2676 .fallocate = ocfs2_fallocate,
2677 .remap_file_range = ocfs2_remap_file_range,
2680 const struct file_operations ocfs2_dops_no_plocks = {
2681 .llseek = generic_file_llseek,
2682 .read = generic_read_dir,
2683 .iterate = ocfs2_readdir,
2684 .fsync = ocfs2_sync_file,
2685 .release = ocfs2_dir_release,
2686 .open = ocfs2_dir_open,
2687 .unlocked_ioctl = ocfs2_ioctl,
2688 #ifdef CONFIG_COMPAT
2689 .compat_ioctl = ocfs2_compat_ioctl,
2691 .flock = ocfs2_flock,