1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 * Copyright (c) 2016-2018 Christoph Hellwig.
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_mount.h"
13 #include "xfs_inode.h"
14 #include "xfs_trans.h"
15 #include "xfs_iomap.h"
16 #include "xfs_trace.h"
18 #include "xfs_bmap_util.h"
19 #include "xfs_reflink.h"
21 struct xfs_writepage_ctx {
22 struct iomap_writepage_ctx ctx;
23 unsigned int data_seq;
27 static inline struct xfs_writepage_ctx *
28 XFS_WPC(struct iomap_writepage_ctx *ctx)
30 return container_of(ctx, struct xfs_writepage_ctx, ctx);
34 * Fast and loose check if this write could update the on-disk inode size.
36 static inline bool xfs_ioend_is_append(struct iomap_ioend *ioend)
38 return ioend->io_offset + ioend->io_size >
39 XFS_I(ioend->io_inode)->i_disk_size;
43 * Update on-disk file size now that data has been written to disk.
51 struct xfs_mount *mp = ip->i_mount;
56 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
60 xfs_ilock(ip, XFS_ILOCK_EXCL);
61 isize = xfs_new_eof(ip, offset + size);
63 xfs_iunlock(ip, XFS_ILOCK_EXCL);
68 trace_xfs_setfilesize(ip, offset, size);
70 ip->i_disk_size = isize;
71 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
72 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
74 return xfs_trans_commit(tp);
78 * IO write completion.
82 struct iomap_ioend *ioend)
84 struct xfs_inode *ip = XFS_I(ioend->io_inode);
85 struct xfs_mount *mp = ip->i_mount;
86 xfs_off_t offset = ioend->io_offset;
87 size_t size = ioend->io_size;
88 unsigned int nofs_flag;
92 * We can allocate memory here while doing writeback on behalf of
93 * memory reclaim. To avoid memory allocation deadlocks set the
94 * task-wide nofs context for the following operations.
96 nofs_flag = memalloc_nofs_save();
99 * Just clean up the in-memory structures if the fs has been shut down.
101 if (xfs_is_shutdown(mp)) {
107 * Clean up all COW blocks and underlying data fork delalloc blocks on
108 * I/O error. The delalloc punch is required because this ioend was
109 * mapped to blocks in the COW fork and the associated pages are no
110 * longer dirty. If we don't remove delalloc blocks here, they become
111 * stale and can corrupt free space accounting on unmount.
113 error = blk_status_to_errno(ioend->io_bio->bi_status);
114 if (unlikely(error)) {
115 if (ioend->io_flags & IOMAP_F_SHARED) {
116 xfs_reflink_cancel_cow_range(ip, offset, size, true);
117 xfs_bmap_punch_delalloc_range(ip,
118 XFS_B_TO_FSBT(mp, offset),
119 XFS_B_TO_FSB(mp, size));
125 * Success: commit the COW or unwritten blocks if needed.
127 if (ioend->io_flags & IOMAP_F_SHARED)
128 error = xfs_reflink_end_cow(ip, offset, size);
129 else if (ioend->io_type == IOMAP_UNWRITTEN)
130 error = xfs_iomap_write_unwritten(ip, offset, size, false);
132 if (!error && xfs_ioend_is_append(ioend))
133 error = xfs_setfilesize(ip, ioend->io_offset, ioend->io_size);
135 iomap_finish_ioends(ioend, error);
136 memalloc_nofs_restore(nofs_flag);
140 * Finish all pending IO completions that require transactional modifications.
142 * We try to merge physical and logically contiguous ioends before completion to
143 * minimise the number of transactions we need to perform during IO completion.
144 * Both unwritten extent conversion and COW remapping need to iterate and modify
145 * one physical extent at a time, so we gain nothing by merging physically
146 * discontiguous extents here.
148 * The ioend chain length that we can be processing here is largely unbound in
149 * length and we may have to perform significant amounts of work on each ioend
150 * to complete it. Hence we have to be careful about holding the CPU for too
155 struct work_struct *work)
157 struct xfs_inode *ip =
158 container_of(work, struct xfs_inode, i_ioend_work);
159 struct iomap_ioend *ioend;
160 struct list_head tmp;
163 spin_lock_irqsave(&ip->i_ioend_lock, flags);
164 list_replace_init(&ip->i_ioend_list, &tmp);
165 spin_unlock_irqrestore(&ip->i_ioend_lock, flags);
167 iomap_sort_ioends(&tmp);
168 while ((ioend = list_first_entry_or_null(&tmp, struct iomap_ioend,
170 list_del_init(&ioend->io_list);
171 iomap_ioend_try_merge(ioend, &tmp);
172 xfs_end_ioend(ioend);
181 struct iomap_ioend *ioend = bio->bi_private;
182 struct xfs_inode *ip = XFS_I(ioend->io_inode);
185 spin_lock_irqsave(&ip->i_ioend_lock, flags);
186 if (list_empty(&ip->i_ioend_list))
187 WARN_ON_ONCE(!queue_work(ip->i_mount->m_unwritten_workqueue,
189 list_add_tail(&ioend->io_list, &ip->i_ioend_list);
190 spin_unlock_irqrestore(&ip->i_ioend_lock, flags);
194 * Fast revalidation of the cached writeback mapping. Return true if the current
195 * mapping is valid, false otherwise.
199 struct iomap_writepage_ctx *wpc,
200 struct xfs_inode *ip,
203 if (offset < wpc->iomap.offset ||
204 offset >= wpc->iomap.offset + wpc->iomap.length)
207 * If this is a COW mapping, it is sufficient to check that the mapping
208 * covers the offset. Be careful to check this first because the caller
209 * can revalidate a COW mapping without updating the data seqno.
211 if (wpc->iomap.flags & IOMAP_F_SHARED)
215 * This is not a COW mapping. Check the sequence number of the data fork
216 * because concurrent changes could have invalidated the extent. Check
217 * the COW fork because concurrent changes since the last time we
218 * checked (and found nothing at this offset) could have added
219 * overlapping blocks.
221 if (XFS_WPC(wpc)->data_seq != READ_ONCE(ip->i_df.if_seq))
223 if (xfs_inode_has_cow_data(ip) &&
224 XFS_WPC(wpc)->cow_seq != READ_ONCE(ip->i_cowfp->if_seq))
230 * Pass in a dellalloc extent and convert it to real extents, return the real
231 * extent that maps offset_fsb in wpc->iomap.
233 * The current page is held locked so nothing could have removed the block
234 * backing offset_fsb, although it could have moved from the COW to the data
235 * fork by another thread.
239 struct iomap_writepage_ctx *wpc,
240 struct xfs_inode *ip,
247 if (whichfork == XFS_COW_FORK)
248 seq = &XFS_WPC(wpc)->cow_seq;
250 seq = &XFS_WPC(wpc)->data_seq;
253 * Attempt to allocate whatever delalloc extent currently backs offset
254 * and put the result into wpc->iomap. Allocate in a loop because it
255 * may take several attempts to allocate real blocks for a contiguous
256 * delalloc extent if free space is sufficiently fragmented.
259 error = xfs_bmapi_convert_delalloc(ip, whichfork, offset,
263 } while (wpc->iomap.offset + wpc->iomap.length <= offset);
270 struct iomap_writepage_ctx *wpc,
274 struct xfs_inode *ip = XFS_I(inode);
275 struct xfs_mount *mp = ip->i_mount;
276 ssize_t count = i_blocksize(inode);
277 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
278 xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, offset + count);
279 xfs_fileoff_t cow_fsb;
281 struct xfs_bmbt_irec imap;
282 struct xfs_iext_cursor icur;
286 if (xfs_is_shutdown(mp))
290 * COW fork blocks can overlap data fork blocks even if the blocks
291 * aren't shared. COW I/O always takes precedent, so we must always
292 * check for overlap on reflink inodes unless the mapping is already a
293 * COW one, or the COW fork hasn't changed from the last time we looked
296 * It's safe to check the COW fork if_seq here without the ILOCK because
297 * we've indirectly protected against concurrent updates: writeback has
298 * the page locked, which prevents concurrent invalidations by reflink
299 * and directio and prevents concurrent buffered writes to the same
300 * page. Changes to if_seq always happen under i_lock, which protects
301 * against concurrent updates and provides a memory barrier on the way
302 * out that ensures that we always see the current value.
304 if (xfs_imap_valid(wpc, ip, offset))
308 * If we don't have a valid map, now it's time to get a new one for this
309 * offset. This will convert delayed allocations (including COW ones)
310 * into real extents. If we return without a valid map, it means we
311 * landed in a hole and we skip the block.
314 cow_fsb = NULLFILEOFF;
315 whichfork = XFS_DATA_FORK;
316 xfs_ilock(ip, XFS_ILOCK_SHARED);
317 ASSERT(!xfs_need_iread_extents(&ip->i_df));
320 * Check if this is offset is covered by a COW extents, and if yes use
321 * it directly instead of looking up anything in the data fork.
323 if (xfs_inode_has_cow_data(ip) &&
324 xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb, &icur, &imap))
325 cow_fsb = imap.br_startoff;
326 if (cow_fsb != NULLFILEOFF && cow_fsb <= offset_fsb) {
327 XFS_WPC(wpc)->cow_seq = READ_ONCE(ip->i_cowfp->if_seq);
328 xfs_iunlock(ip, XFS_ILOCK_SHARED);
330 whichfork = XFS_COW_FORK;
331 goto allocate_blocks;
335 * No COW extent overlap. Revalidate now that we may have updated
336 * ->cow_seq. If the data mapping is still valid, we're done.
338 if (xfs_imap_valid(wpc, ip, offset)) {
339 xfs_iunlock(ip, XFS_ILOCK_SHARED);
344 * If we don't have a valid map, now it's time to get a new one for this
345 * offset. This will convert delayed allocations (including COW ones)
348 if (!xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap))
349 imap.br_startoff = end_fsb; /* fake a hole past EOF */
350 XFS_WPC(wpc)->data_seq = READ_ONCE(ip->i_df.if_seq);
351 xfs_iunlock(ip, XFS_ILOCK_SHARED);
353 /* landed in a hole or beyond EOF? */
354 if (imap.br_startoff > offset_fsb) {
355 imap.br_blockcount = imap.br_startoff - offset_fsb;
356 imap.br_startoff = offset_fsb;
357 imap.br_startblock = HOLESTARTBLOCK;
358 imap.br_state = XFS_EXT_NORM;
362 * Truncate to the next COW extent if there is one. This is the only
363 * opportunity to do this because we can skip COW fork lookups for the
364 * subsequent blocks in the mapping; however, the requirement to treat
365 * the COW range separately remains.
367 if (cow_fsb != NULLFILEOFF &&
368 cow_fsb < imap.br_startoff + imap.br_blockcount)
369 imap.br_blockcount = cow_fsb - imap.br_startoff;
371 /* got a delalloc extent? */
372 if (imap.br_startblock != HOLESTARTBLOCK &&
373 isnullstartblock(imap.br_startblock))
374 goto allocate_blocks;
376 xfs_bmbt_to_iomap(ip, &wpc->iomap, &imap, 0, 0);
377 trace_xfs_map_blocks_found(ip, offset, count, whichfork, &imap);
380 error = xfs_convert_blocks(wpc, ip, whichfork, offset);
383 * If we failed to find the extent in the COW fork we might have
384 * raced with a COW to data fork conversion or truncate.
385 * Restart the lookup to catch the extent in the data fork for
386 * the former case, but prevent additional retries to avoid
387 * looping forever for the latter case.
389 if (error == -EAGAIN && whichfork == XFS_COW_FORK && !retries++)
391 ASSERT(error != -EAGAIN);
396 * Due to merging the return real extent might be larger than the
397 * original delalloc one. Trim the return extent to the next COW
398 * boundary again to force a re-lookup.
400 if (whichfork != XFS_COW_FORK && cow_fsb != NULLFILEOFF) {
401 loff_t cow_offset = XFS_FSB_TO_B(mp, cow_fsb);
403 if (cow_offset < wpc->iomap.offset + wpc->iomap.length)
404 wpc->iomap.length = cow_offset - wpc->iomap.offset;
407 ASSERT(wpc->iomap.offset <= offset);
408 ASSERT(wpc->iomap.offset + wpc->iomap.length > offset);
409 trace_xfs_map_blocks_alloc(ip, offset, count, whichfork, &imap);
415 struct iomap_ioend *ioend,
418 unsigned int nofs_flag;
421 * We can allocate memory here while doing writeback on behalf of
422 * memory reclaim. To avoid memory allocation deadlocks set the
423 * task-wide nofs context for the following operations.
425 nofs_flag = memalloc_nofs_save();
427 /* Convert CoW extents to regular */
428 if (!status && (ioend->io_flags & IOMAP_F_SHARED)) {
429 status = xfs_reflink_convert_cow(XFS_I(ioend->io_inode),
430 ioend->io_offset, ioend->io_size);
433 memalloc_nofs_restore(nofs_flag);
435 /* send ioends that might require a transaction to the completion wq */
436 if (xfs_ioend_is_append(ioend) || ioend->io_type == IOMAP_UNWRITTEN ||
437 (ioend->io_flags & IOMAP_F_SHARED))
438 ioend->io_bio->bi_end_io = xfs_end_bio;
443 * If the page has delalloc blocks on it, we need to punch them out before we
444 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
445 * inode that can trip up a later direct I/O read operation on the same region.
447 * We prevent this by truncating away the delalloc regions on the page. Because
448 * they are delalloc, we can do this without needing a transaction. Indeed - if
449 * we get ENOSPC errors, we have to be able to do this truncation without a
450 * transaction as there is no space left for block reservation (typically why we
451 * see a ENOSPC in writeback).
458 struct inode *inode = folio->mapping->host;
459 struct xfs_inode *ip = XFS_I(inode);
460 struct xfs_mount *mp = ip->i_mount;
461 size_t offset = offset_in_folio(folio, pos);
462 xfs_fileoff_t start_fsb = XFS_B_TO_FSBT(mp, pos);
463 xfs_fileoff_t pageoff_fsb = XFS_B_TO_FSBT(mp, offset);
466 if (xfs_is_shutdown(mp))
469 xfs_alert_ratelimited(mp,
470 "page discard on page "PTR_FMT", inode 0x%llx, pos %llu.",
471 folio, ip->i_ino, pos);
473 error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
474 i_blocks_per_folio(inode, folio) - pageoff_fsb);
475 if (error && !xfs_is_shutdown(mp))
476 xfs_alert(mp, "page discard unable to remove delalloc mapping.");
479 static const struct iomap_writeback_ops xfs_writeback_ops = {
480 .map_blocks = xfs_map_blocks,
481 .prepare_ioend = xfs_prepare_ioend,
482 .discard_folio = xfs_discard_folio,
487 struct address_space *mapping,
488 struct writeback_control *wbc)
490 struct xfs_writepage_ctx wpc = { };
493 * Writing back data in a transaction context can result in recursive
494 * transactions. This is bad, so issue a warning and get out of here.
496 if (WARN_ON_ONCE(current->journal_info))
499 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
500 return iomap_writepages(mapping, wbc, &wpc.ctx, &xfs_writeback_ops);
505 struct address_space *mapping,
506 struct writeback_control *wbc)
508 struct xfs_inode *ip = XFS_I(mapping->host);
510 xfs_iflags_clear(ip, XFS_ITRUNCATED);
511 return dax_writeback_mapping_range(mapping,
512 xfs_inode_buftarg(ip)->bt_daxdev, wbc);
517 struct address_space *mapping,
520 struct xfs_inode *ip = XFS_I(mapping->host);
522 trace_xfs_vm_bmap(ip);
525 * The swap code (ab-)uses ->bmap to get a block mapping and then
526 * bypasses the file system for actual I/O. We really can't allow
527 * that on reflinks inodes, so we have to skip out here. And yes,
528 * 0 is the magic code for a bmap error.
530 * Since we don't pass back blockdev info, we can't return bmap
531 * information for rt files either.
533 if (xfs_is_cow_inode(ip) || XFS_IS_REALTIME_INODE(ip))
535 return iomap_bmap(mapping, block, &xfs_read_iomap_ops);
543 return iomap_read_folio(folio, &xfs_read_iomap_ops);
548 struct readahead_control *rac)
550 iomap_readahead(rac, &xfs_read_iomap_ops);
554 xfs_iomap_swapfile_activate(
555 struct swap_info_struct *sis,
556 struct file *swap_file,
559 sis->bdev = xfs_inode_buftarg(XFS_I(file_inode(swap_file)))->bt_bdev;
560 return iomap_swapfile_activate(sis, swap_file, span,
561 &xfs_read_iomap_ops);
564 const struct address_space_operations xfs_address_space_operations = {
565 .read_folio = xfs_vm_read_folio,
566 .readahead = xfs_vm_readahead,
567 .writepages = xfs_vm_writepages,
568 .dirty_folio = filemap_dirty_folio,
569 .release_folio = iomap_release_folio,
570 .invalidate_folio = iomap_invalidate_folio,
572 .direct_IO = noop_direct_IO,
573 .migrate_folio = filemap_migrate_folio,
574 .is_partially_uptodate = iomap_is_partially_uptodate,
575 .error_remove_page = generic_error_remove_page,
576 .swap_activate = xfs_iomap_swapfile_activate,
579 const struct address_space_operations xfs_dax_aops = {
580 .writepages = xfs_dax_writepages,
581 .direct_IO = noop_direct_IO,
582 .dirty_folio = noop_dirty_folio,
583 .swap_activate = xfs_iomap_swapfile_activate,