xref: /linux/fs/xfs/xfs_aops.c (revision c148bc7535650fbfa95a1f571b9ffa2ab478ea33)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4  * Copyright (c) 2016-2025 Christoph Hellwig.
5  * All Rights Reserved.
6  */
7 #include "xfs.h"
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"
17 #include "xfs_bmap.h"
18 #include "xfs_bmap_util.h"
19 #include "xfs_reflink.h"
20 #include "xfs_errortag.h"
21 #include "xfs_error.h"
22 #include "xfs_icache.h"
23 #include "xfs_zone_alloc.h"
24 #include "xfs_rtgroup.h"
25 
26 struct xfs_writepage_ctx {
27 	struct iomap_writepage_ctx ctx;
28 	unsigned int		data_seq;
29 	unsigned int		cow_seq;
30 };
31 
32 static inline struct xfs_writepage_ctx *
XFS_WPC(struct iomap_writepage_ctx * ctx)33 XFS_WPC(struct iomap_writepage_ctx *ctx)
34 {
35 	return container_of(ctx, struct xfs_writepage_ctx, ctx);
36 }
37 
38 /*
39  * Fast and loose check if this write could update the on-disk inode size.
40  */
xfs_ioend_is_append(struct iomap_ioend * ioend)41 static inline bool xfs_ioend_is_append(struct iomap_ioend *ioend)
42 {
43 	return ioend->io_offset + ioend->io_size >
44 		XFS_I(ioend->io_inode)->i_disk_size;
45 }
46 
47 /*
48  * Update on-disk file size now that data has been written to disk.
49  */
50 int
xfs_setfilesize(struct xfs_inode * ip,xfs_off_t offset,size_t size)51 xfs_setfilesize(
52 	struct xfs_inode	*ip,
53 	xfs_off_t		offset,
54 	size_t			size)
55 {
56 	struct xfs_mount	*mp = ip->i_mount;
57 	struct xfs_trans	*tp;
58 	xfs_fsize_t		isize;
59 	int			error;
60 
61 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
62 	if (error)
63 		return error;
64 
65 	xfs_ilock(ip, XFS_ILOCK_EXCL);
66 	isize = xfs_new_eof(ip, offset + size);
67 	if (!isize) {
68 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
69 		xfs_trans_cancel(tp);
70 		return 0;
71 	}
72 
73 	trace_xfs_setfilesize(ip, offset, size);
74 
75 	ip->i_disk_size = isize;
76 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
77 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
78 
79 	return xfs_trans_commit(tp);
80 }
81 
82 static void
xfs_ioend_put_open_zones(struct iomap_ioend * ioend)83 xfs_ioend_put_open_zones(
84 	struct iomap_ioend	*ioend)
85 {
86 	struct iomap_ioend *tmp;
87 
88 	/*
89 	 * Put the open zone for all ioends merged into this one (if any).
90 	 */
91 	list_for_each_entry(tmp, &ioend->io_list, io_list)
92 		xfs_open_zone_put(tmp->io_private);
93 
94 	/*
95 	 * The main ioend might not have an open zone if the submission failed
96 	 * before xfs_zone_alloc_and_submit got called.
97 	 */
98 	if (ioend->io_private)
99 		xfs_open_zone_put(ioend->io_private);
100 }
101 
102 /*
103  * IO write completion.
104  */
105 STATIC void
xfs_end_ioend(struct iomap_ioend * ioend)106 xfs_end_ioend(
107 	struct iomap_ioend	*ioend)
108 {
109 	struct xfs_inode	*ip = XFS_I(ioend->io_inode);
110 	struct xfs_mount	*mp = ip->i_mount;
111 	bool			is_zoned = xfs_is_zoned_inode(ip);
112 	xfs_off_t		offset = ioend->io_offset;
113 	size_t			size = ioend->io_size;
114 	unsigned int		nofs_flag;
115 	int			error;
116 
117 	/*
118 	 * We can allocate memory here while doing writeback on behalf of
119 	 * memory reclaim.  To avoid memory allocation deadlocks set the
120 	 * task-wide nofs context for the following operations.
121 	 */
122 	nofs_flag = memalloc_nofs_save();
123 
124 	/*
125 	 * Just clean up the in-memory structures if the fs has been shut down.
126 	 */
127 	if (xfs_is_shutdown(mp)) {
128 		error = -EIO;
129 		goto done;
130 	}
131 
132 	/*
133 	 * Clean up all COW blocks and underlying data fork delalloc blocks on
134 	 * I/O error. The delalloc punch is required because this ioend was
135 	 * mapped to blocks in the COW fork and the associated pages are no
136 	 * longer dirty. If we don't remove delalloc blocks here, they become
137 	 * stale and can corrupt free space accounting on unmount.
138 	 */
139 	error = blk_status_to_errno(ioend->io_bio.bi_status);
140 	if (unlikely(error)) {
141 		if (ioend->io_flags & IOMAP_IOEND_SHARED) {
142 			ASSERT(!is_zoned);
143 			xfs_reflink_cancel_cow_range(ip, offset, size, true);
144 			xfs_bmap_punch_delalloc_range(ip, XFS_DATA_FORK, offset,
145 					offset + size, NULL);
146 		}
147 		goto done;
148 	}
149 
150 	/*
151 	 * Success: commit the COW or unwritten blocks if needed.
152 	 */
153 	if (is_zoned)
154 		error = xfs_zoned_end_io(ip, offset, size, ioend->io_sector,
155 				ioend->io_private, NULLFSBLOCK);
156 	else if (ioend->io_flags & IOMAP_IOEND_SHARED)
157 		error = xfs_reflink_end_cow(ip, offset, size);
158 	else if (ioend->io_flags & IOMAP_IOEND_UNWRITTEN)
159 		error = xfs_iomap_write_unwritten(ip, offset, size, false);
160 
161 	if (!error &&
162 	    !(ioend->io_flags & IOMAP_IOEND_DIRECT) &&
163 	    xfs_ioend_is_append(ioend))
164 		error = xfs_setfilesize(ip, offset, size);
165 done:
166 	if (is_zoned)
167 		xfs_ioend_put_open_zones(ioend);
168 	iomap_finish_ioends(ioend, error);
169 	memalloc_nofs_restore(nofs_flag);
170 }
171 
172 /*
173  * Finish all pending IO completions that require transactional modifications.
174  *
175  * We try to merge physical and logically contiguous ioends before completion to
176  * minimise the number of transactions we need to perform during IO completion.
177  * Both unwritten extent conversion and COW remapping need to iterate and modify
178  * one physical extent at a time, so we gain nothing by merging physically
179  * discontiguous extents here.
180  *
181  * The ioend chain length that we can be processing here is largely unbound in
182  * length and we may have to perform significant amounts of work on each ioend
183  * to complete it. Hence we have to be careful about holding the CPU for too
184  * long in this loop.
185  */
186 void
xfs_end_io(struct work_struct * work)187 xfs_end_io(
188 	struct work_struct	*work)
189 {
190 	struct xfs_inode	*ip =
191 		container_of(work, struct xfs_inode, i_ioend_work);
192 	struct iomap_ioend	*ioend;
193 	struct list_head	tmp;
194 	unsigned long		flags;
195 
196 	spin_lock_irqsave(&ip->i_ioend_lock, flags);
197 	list_replace_init(&ip->i_ioend_list, &tmp);
198 	spin_unlock_irqrestore(&ip->i_ioend_lock, flags);
199 
200 	iomap_sort_ioends(&tmp);
201 	while ((ioend = list_first_entry_or_null(&tmp, struct iomap_ioend,
202 			io_list))) {
203 		list_del_init(&ioend->io_list);
204 		iomap_ioend_try_merge(ioend, &tmp);
205 		xfs_end_ioend(ioend);
206 		cond_resched();
207 	}
208 }
209 
210 void
xfs_end_bio(struct bio * bio)211 xfs_end_bio(
212 	struct bio		*bio)
213 {
214 	struct iomap_ioend	*ioend = iomap_ioend_from_bio(bio);
215 	struct xfs_inode	*ip = XFS_I(ioend->io_inode);
216 	struct xfs_mount	*mp = ip->i_mount;
217 	unsigned long		flags;
218 
219 	/*
220 	 * For Appends record the actually written block number and set the
221 	 * boundary flag if needed.
222 	 */
223 	if (IS_ENABLED(CONFIG_XFS_RT) && bio_is_zone_append(bio)) {
224 		ioend->io_sector = bio->bi_iter.bi_sector;
225 		xfs_mark_rtg_boundary(ioend);
226 	}
227 
228 	spin_lock_irqsave(&ip->i_ioend_lock, flags);
229 	if (list_empty(&ip->i_ioend_list))
230 		WARN_ON_ONCE(!queue_work(mp->m_unwritten_workqueue,
231 					 &ip->i_ioend_work));
232 	list_add_tail(&ioend->io_list, &ip->i_ioend_list);
233 	spin_unlock_irqrestore(&ip->i_ioend_lock, flags);
234 }
235 
236 /*
237  * Fast revalidation of the cached writeback mapping. Return true if the current
238  * mapping is valid, false otherwise.
239  */
240 static bool
xfs_imap_valid(struct iomap_writepage_ctx * wpc,struct xfs_inode * ip,loff_t offset)241 xfs_imap_valid(
242 	struct iomap_writepage_ctx	*wpc,
243 	struct xfs_inode		*ip,
244 	loff_t				offset)
245 {
246 	if (offset < wpc->iomap.offset ||
247 	    offset >= wpc->iomap.offset + wpc->iomap.length)
248 		return false;
249 	/*
250 	 * If this is a COW mapping, it is sufficient to check that the mapping
251 	 * covers the offset. Be careful to check this first because the caller
252 	 * can revalidate a COW mapping without updating the data seqno.
253 	 */
254 	if (wpc->iomap.flags & IOMAP_F_SHARED)
255 		return true;
256 
257 	/*
258 	 * This is not a COW mapping. Check the sequence number of the data fork
259 	 * because concurrent changes could have invalidated the extent. Check
260 	 * the COW fork because concurrent changes since the last time we
261 	 * checked (and found nothing at this offset) could have added
262 	 * overlapping blocks.
263 	 */
264 	if (XFS_WPC(wpc)->data_seq != READ_ONCE(ip->i_df.if_seq)) {
265 		trace_xfs_wb_data_iomap_invalid(ip, &wpc->iomap,
266 				XFS_WPC(wpc)->data_seq, XFS_DATA_FORK);
267 		return false;
268 	}
269 	if (xfs_inode_has_cow_data(ip) &&
270 	    XFS_WPC(wpc)->cow_seq != READ_ONCE(ip->i_cowfp->if_seq)) {
271 		trace_xfs_wb_cow_iomap_invalid(ip, &wpc->iomap,
272 				XFS_WPC(wpc)->cow_seq, XFS_COW_FORK);
273 		return false;
274 	}
275 	return true;
276 }
277 
278 static int
xfs_map_blocks(struct iomap_writepage_ctx * wpc,struct inode * inode,loff_t offset,unsigned int len)279 xfs_map_blocks(
280 	struct iomap_writepage_ctx *wpc,
281 	struct inode		*inode,
282 	loff_t			offset,
283 	unsigned int		len)
284 {
285 	struct xfs_inode	*ip = XFS_I(inode);
286 	struct xfs_mount	*mp = ip->i_mount;
287 	ssize_t			count = i_blocksize(inode);
288 	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
289 	xfs_fileoff_t		end_fsb = XFS_B_TO_FSB(mp, offset + count);
290 	xfs_fileoff_t		cow_fsb;
291 	int			whichfork;
292 	struct xfs_bmbt_irec	imap;
293 	struct xfs_iext_cursor	icur;
294 	int			retries = 0;
295 	int			error = 0;
296 	unsigned int		*seq;
297 
298 	if (xfs_is_shutdown(mp))
299 		return -EIO;
300 
301 	XFS_ERRORTAG_DELAY(mp, XFS_ERRTAG_WB_DELAY_MS);
302 
303 	/*
304 	 * COW fork blocks can overlap data fork blocks even if the blocks
305 	 * aren't shared.  COW I/O always takes precedent, so we must always
306 	 * check for overlap on reflink inodes unless the mapping is already a
307 	 * COW one, or the COW fork hasn't changed from the last time we looked
308 	 * at it.
309 	 *
310 	 * It's safe to check the COW fork if_seq here without the ILOCK because
311 	 * we've indirectly protected against concurrent updates: writeback has
312 	 * the page locked, which prevents concurrent invalidations by reflink
313 	 * and directio and prevents concurrent buffered writes to the same
314 	 * page.  Changes to if_seq always happen under i_lock, which protects
315 	 * against concurrent updates and provides a memory barrier on the way
316 	 * out that ensures that we always see the current value.
317 	 */
318 	if (xfs_imap_valid(wpc, ip, offset))
319 		return 0;
320 
321 	/*
322 	 * If we don't have a valid map, now it's time to get a new one for this
323 	 * offset.  This will convert delayed allocations (including COW ones)
324 	 * into real extents.  If we return without a valid map, it means we
325 	 * landed in a hole and we skip the block.
326 	 */
327 retry:
328 	cow_fsb = NULLFILEOFF;
329 	whichfork = XFS_DATA_FORK;
330 	xfs_ilock(ip, XFS_ILOCK_SHARED);
331 	ASSERT(!xfs_need_iread_extents(&ip->i_df));
332 
333 	/*
334 	 * Check if this is offset is covered by a COW extents, and if yes use
335 	 * it directly instead of looking up anything in the data fork.
336 	 */
337 	if (xfs_inode_has_cow_data(ip) &&
338 	    xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb, &icur, &imap))
339 		cow_fsb = imap.br_startoff;
340 	if (cow_fsb != NULLFILEOFF && cow_fsb <= offset_fsb) {
341 		XFS_WPC(wpc)->cow_seq = READ_ONCE(ip->i_cowfp->if_seq);
342 		xfs_iunlock(ip, XFS_ILOCK_SHARED);
343 
344 		whichfork = XFS_COW_FORK;
345 		goto allocate_blocks;
346 	}
347 
348 	/*
349 	 * No COW extent overlap. Revalidate now that we may have updated
350 	 * ->cow_seq. If the data mapping is still valid, we're done.
351 	 */
352 	if (xfs_imap_valid(wpc, ip, offset)) {
353 		xfs_iunlock(ip, XFS_ILOCK_SHARED);
354 		return 0;
355 	}
356 
357 	/*
358 	 * If we don't have a valid map, now it's time to get a new one for this
359 	 * offset.  This will convert delayed allocations (including COW ones)
360 	 * into real extents.
361 	 */
362 	if (!xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap))
363 		imap.br_startoff = end_fsb;	/* fake a hole past EOF */
364 	XFS_WPC(wpc)->data_seq = READ_ONCE(ip->i_df.if_seq);
365 	xfs_iunlock(ip, XFS_ILOCK_SHARED);
366 
367 	/* landed in a hole or beyond EOF? */
368 	if (imap.br_startoff > offset_fsb) {
369 		imap.br_blockcount = imap.br_startoff - offset_fsb;
370 		imap.br_startoff = offset_fsb;
371 		imap.br_startblock = HOLESTARTBLOCK;
372 		imap.br_state = XFS_EXT_NORM;
373 	}
374 
375 	/*
376 	 * Truncate to the next COW extent if there is one.  This is the only
377 	 * opportunity to do this because we can skip COW fork lookups for the
378 	 * subsequent blocks in the mapping; however, the requirement to treat
379 	 * the COW range separately remains.
380 	 */
381 	if (cow_fsb != NULLFILEOFF &&
382 	    cow_fsb < imap.br_startoff + imap.br_blockcount)
383 		imap.br_blockcount = cow_fsb - imap.br_startoff;
384 
385 	/* got a delalloc extent? */
386 	if (imap.br_startblock != HOLESTARTBLOCK &&
387 	    isnullstartblock(imap.br_startblock))
388 		goto allocate_blocks;
389 
390 	xfs_bmbt_to_iomap(ip, &wpc->iomap, &imap, 0, 0, XFS_WPC(wpc)->data_seq);
391 	trace_xfs_map_blocks_found(ip, offset, count, whichfork, &imap);
392 	return 0;
393 allocate_blocks:
394 	/*
395 	 * Convert a dellalloc extent to a real one. The current page is held
396 	 * locked so nothing could have removed the block backing offset_fsb,
397 	 * although it could have moved from the COW to the data fork by another
398 	 * thread.
399 	 */
400 	if (whichfork == XFS_COW_FORK)
401 		seq = &XFS_WPC(wpc)->cow_seq;
402 	else
403 		seq = &XFS_WPC(wpc)->data_seq;
404 
405 	error = xfs_bmapi_convert_delalloc(ip, whichfork, offset,
406 				&wpc->iomap, seq);
407 	if (error) {
408 		/*
409 		 * If we failed to find the extent in the COW fork we might have
410 		 * raced with a COW to data fork conversion or truncate.
411 		 * Restart the lookup to catch the extent in the data fork for
412 		 * the former case, but prevent additional retries to avoid
413 		 * looping forever for the latter case.
414 		 */
415 		if (error == -EAGAIN && whichfork == XFS_COW_FORK && !retries++)
416 			goto retry;
417 		ASSERT(error != -EAGAIN);
418 		return error;
419 	}
420 
421 	/*
422 	 * Due to merging the return real extent might be larger than the
423 	 * original delalloc one.  Trim the return extent to the next COW
424 	 * boundary again to force a re-lookup.
425 	 */
426 	if (whichfork != XFS_COW_FORK && cow_fsb != NULLFILEOFF) {
427 		loff_t		cow_offset = XFS_FSB_TO_B(mp, cow_fsb);
428 
429 		if (cow_offset < wpc->iomap.offset + wpc->iomap.length)
430 			wpc->iomap.length = cow_offset - wpc->iomap.offset;
431 	}
432 
433 	ASSERT(wpc->iomap.offset <= offset);
434 	ASSERT(wpc->iomap.offset + wpc->iomap.length > offset);
435 	trace_xfs_map_blocks_alloc(ip, offset, count, whichfork, &imap);
436 	return 0;
437 }
438 
439 static int
xfs_submit_ioend(struct iomap_writepage_ctx * wpc,int status)440 xfs_submit_ioend(
441 	struct iomap_writepage_ctx *wpc,
442 	int			status)
443 {
444 	struct iomap_ioend	*ioend = wpc->ioend;
445 	unsigned int		nofs_flag;
446 
447 	/*
448 	 * We can allocate memory here while doing writeback on behalf of
449 	 * memory reclaim.  To avoid memory allocation deadlocks set the
450 	 * task-wide nofs context for the following operations.
451 	 */
452 	nofs_flag = memalloc_nofs_save();
453 
454 	/* Convert CoW extents to regular */
455 	if (!status && (ioend->io_flags & IOMAP_IOEND_SHARED)) {
456 		status = xfs_reflink_convert_cow(XFS_I(ioend->io_inode),
457 				ioend->io_offset, ioend->io_size);
458 	}
459 
460 	memalloc_nofs_restore(nofs_flag);
461 
462 	/* send ioends that might require a transaction to the completion wq */
463 	if (xfs_ioend_is_append(ioend) ||
464 	    (ioend->io_flags & (IOMAP_IOEND_UNWRITTEN | IOMAP_IOEND_SHARED)))
465 		ioend->io_bio.bi_end_io = xfs_end_bio;
466 
467 	if (status)
468 		return status;
469 	submit_bio(&ioend->io_bio);
470 	return 0;
471 }
472 
473 /*
474  * If the folio has delalloc blocks on it, the caller is asking us to punch them
475  * out. If we don't, we can leave a stale delalloc mapping covered by a clean
476  * page that needs to be dirtied again before the delalloc mapping can be
477  * converted. This stale delalloc mapping can trip up a later direct I/O read
478  * operation on the same region.
479  *
480  * We prevent this by truncating away the delalloc regions on the folio. Because
481  * they are delalloc, we can do this without needing a transaction. Indeed - if
482  * we get ENOSPC errors, we have to be able to do this truncation without a
483  * transaction as there is no space left for block reservation (typically why
484  * we see a ENOSPC in writeback).
485  */
486 static void
xfs_discard_folio(struct folio * folio,loff_t pos)487 xfs_discard_folio(
488 	struct folio		*folio,
489 	loff_t			pos)
490 {
491 	struct xfs_inode	*ip = XFS_I(folio->mapping->host);
492 	struct xfs_mount	*mp = ip->i_mount;
493 
494 	if (xfs_is_shutdown(mp))
495 		return;
496 
497 	xfs_alert_ratelimited(mp,
498 		"page discard on page "PTR_FMT", inode 0x%llx, pos %llu.",
499 			folio, ip->i_ino, pos);
500 
501 	/*
502 	 * The end of the punch range is always the offset of the first
503 	 * byte of the next folio. Hence the end offset is only dependent on the
504 	 * folio itself and not the start offset that is passed in.
505 	 */
506 	xfs_bmap_punch_delalloc_range(ip, XFS_DATA_FORK, pos,
507 				folio_pos(folio) + folio_size(folio), NULL);
508 }
509 
510 static const struct iomap_writeback_ops xfs_writeback_ops = {
511 	.map_blocks		= xfs_map_blocks,
512 	.submit_ioend		= xfs_submit_ioend,
513 	.discard_folio		= xfs_discard_folio,
514 };
515 
516 struct xfs_zoned_writepage_ctx {
517 	struct iomap_writepage_ctx	ctx;
518 	struct xfs_open_zone		*open_zone;
519 };
520 
521 static inline struct xfs_zoned_writepage_ctx *
XFS_ZWPC(struct iomap_writepage_ctx * ctx)522 XFS_ZWPC(struct iomap_writepage_ctx *ctx)
523 {
524 	return container_of(ctx, struct xfs_zoned_writepage_ctx, ctx);
525 }
526 
527 static int
xfs_zoned_map_blocks(struct iomap_writepage_ctx * wpc,struct inode * inode,loff_t offset,unsigned int len)528 xfs_zoned_map_blocks(
529 	struct iomap_writepage_ctx *wpc,
530 	struct inode		*inode,
531 	loff_t			offset,
532 	unsigned int		len)
533 {
534 	struct xfs_inode	*ip = XFS_I(inode);
535 	struct xfs_mount	*mp = ip->i_mount;
536 	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
537 	xfs_fileoff_t		end_fsb = XFS_B_TO_FSB(mp, offset + len);
538 	xfs_filblks_t		count_fsb;
539 	struct xfs_bmbt_irec	imap, del;
540 	struct xfs_iext_cursor	icur;
541 
542 	if (xfs_is_shutdown(mp))
543 		return -EIO;
544 
545 	XFS_ERRORTAG_DELAY(mp, XFS_ERRTAG_WB_DELAY_MS);
546 
547 	/*
548 	 * All dirty data must be covered by delalloc extents.  But truncate can
549 	 * remove delalloc extents underneath us or reduce their size.
550 	 * Returning a hole tells iomap to not write back any data from this
551 	 * range, which is the right thing to do in that case.
552 	 *
553 	 * Otherwise just tell iomap to treat ranges previously covered by a
554 	 * delalloc extent as mapped.  The actual block allocation will be done
555 	 * just before submitting the bio.
556 	 *
557 	 * This implies we never map outside folios that are locked or marked
558 	 * as under writeback, and thus there is no need check the fork sequence
559 	 * count here.
560 	 */
561 	xfs_ilock(ip, XFS_ILOCK_EXCL);
562 	if (!xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb, &icur, &imap))
563 		imap.br_startoff = end_fsb;	/* fake a hole past EOF */
564 	if (imap.br_startoff > offset_fsb) {
565 		imap.br_blockcount = imap.br_startoff - offset_fsb;
566 		imap.br_startoff = offset_fsb;
567 		imap.br_startblock = HOLESTARTBLOCK;
568 		imap.br_state = XFS_EXT_NORM;
569 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
570 		xfs_bmbt_to_iomap(ip, &wpc->iomap, &imap, 0, 0, 0);
571 		return 0;
572 	}
573 	end_fsb = min(end_fsb, imap.br_startoff + imap.br_blockcount);
574 	count_fsb = end_fsb - offset_fsb;
575 
576 	del = imap;
577 	xfs_trim_extent(&del, offset_fsb, count_fsb);
578 	xfs_bmap_del_extent_delay(ip, XFS_COW_FORK, &icur, &imap, &del,
579 			XFS_BMAPI_REMAP);
580 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
581 
582 	wpc->iomap.type = IOMAP_MAPPED;
583 	wpc->iomap.flags = IOMAP_F_DIRTY;
584 	wpc->iomap.bdev = mp->m_rtdev_targp->bt_bdev;
585 	wpc->iomap.offset = offset;
586 	wpc->iomap.length = XFS_FSB_TO_B(mp, count_fsb);
587 	wpc->iomap.flags = IOMAP_F_ANON_WRITE;
588 
589 	trace_xfs_zoned_map_blocks(ip, offset, wpc->iomap.length);
590 	return 0;
591 }
592 
593 static int
xfs_zoned_submit_ioend(struct iomap_writepage_ctx * wpc,int status)594 xfs_zoned_submit_ioend(
595 	struct iomap_writepage_ctx *wpc,
596 	int			status)
597 {
598 	wpc->ioend->io_bio.bi_end_io = xfs_end_bio;
599 	if (status)
600 		return status;
601 	xfs_zone_alloc_and_submit(wpc->ioend, &XFS_ZWPC(wpc)->open_zone);
602 	return 0;
603 }
604 
605 static const struct iomap_writeback_ops xfs_zoned_writeback_ops = {
606 	.map_blocks		= xfs_zoned_map_blocks,
607 	.submit_ioend		= xfs_zoned_submit_ioend,
608 	.discard_folio		= xfs_discard_folio,
609 };
610 
611 STATIC int
xfs_vm_writepages(struct address_space * mapping,struct writeback_control * wbc)612 xfs_vm_writepages(
613 	struct address_space	*mapping,
614 	struct writeback_control *wbc)
615 {
616 	struct xfs_inode	*ip = XFS_I(mapping->host);
617 
618 	xfs_iflags_clear(ip, XFS_ITRUNCATED);
619 
620 	if (xfs_is_zoned_inode(ip)) {
621 		struct xfs_zoned_writepage_ctx	xc = { };
622 		int				error;
623 
624 		error = iomap_writepages(mapping, wbc, &xc.ctx,
625 					 &xfs_zoned_writeback_ops);
626 		if (xc.open_zone)
627 			xfs_open_zone_put(xc.open_zone);
628 		return error;
629 	} else {
630 		struct xfs_writepage_ctx	wpc = { };
631 
632 		return iomap_writepages(mapping, wbc, &wpc.ctx,
633 				&xfs_writeback_ops);
634 	}
635 }
636 
637 STATIC int
xfs_dax_writepages(struct address_space * mapping,struct writeback_control * wbc)638 xfs_dax_writepages(
639 	struct address_space	*mapping,
640 	struct writeback_control *wbc)
641 {
642 	struct xfs_inode	*ip = XFS_I(mapping->host);
643 
644 	xfs_iflags_clear(ip, XFS_ITRUNCATED);
645 	return dax_writeback_mapping_range(mapping,
646 			xfs_inode_buftarg(ip)->bt_daxdev, wbc);
647 }
648 
649 STATIC sector_t
xfs_vm_bmap(struct address_space * mapping,sector_t block)650 xfs_vm_bmap(
651 	struct address_space	*mapping,
652 	sector_t		block)
653 {
654 	struct xfs_inode	*ip = XFS_I(mapping->host);
655 
656 	trace_xfs_vm_bmap(ip);
657 
658 	/*
659 	 * The swap code (ab-)uses ->bmap to get a block mapping and then
660 	 * bypasses the file system for actual I/O.  We really can't allow
661 	 * that on reflinks inodes, so we have to skip out here.  And yes,
662 	 * 0 is the magic code for a bmap error.
663 	 *
664 	 * Since we don't pass back blockdev info, we can't return bmap
665 	 * information for rt files either.
666 	 */
667 	if (xfs_is_cow_inode(ip) || XFS_IS_REALTIME_INODE(ip))
668 		return 0;
669 	return iomap_bmap(mapping, block, &xfs_read_iomap_ops);
670 }
671 
672 STATIC int
xfs_vm_read_folio(struct file * unused,struct folio * folio)673 xfs_vm_read_folio(
674 	struct file		*unused,
675 	struct folio		*folio)
676 {
677 	return iomap_read_folio(folio, &xfs_read_iomap_ops);
678 }
679 
680 STATIC void
xfs_vm_readahead(struct readahead_control * rac)681 xfs_vm_readahead(
682 	struct readahead_control	*rac)
683 {
684 	iomap_readahead(rac, &xfs_read_iomap_ops);
685 }
686 
687 static int
xfs_vm_swap_activate(struct swap_info_struct * sis,struct file * swap_file,sector_t * span)688 xfs_vm_swap_activate(
689 	struct swap_info_struct		*sis,
690 	struct file			*swap_file,
691 	sector_t			*span)
692 {
693 	struct xfs_inode		*ip = XFS_I(file_inode(swap_file));
694 
695 	/*
696 	 * Swap file activation can race against concurrent shared extent
697 	 * removal in files that have been cloned.  If this happens,
698 	 * iomap_swapfile_iter() can fail because it encountered a shared
699 	 * extent even though an operation is in progress to remove those
700 	 * shared extents.
701 	 *
702 	 * This race becomes problematic when we defer extent removal
703 	 * operations beyond the end of a syscall (i.e. use async background
704 	 * processing algorithms).  Users think the extents are no longer
705 	 * shared, but iomap_swapfile_iter() still sees them as shared
706 	 * because the refcountbt entries for the extents being removed have
707 	 * not yet been updated.  Hence the swapon call fails unexpectedly.
708 	 *
709 	 * The race condition is currently most obvious from the unlink()
710 	 * operation as extent removal is deferred until after the last
711 	 * reference to the inode goes away.  We then process the extent
712 	 * removal asynchronously, hence triggers the "syscall completed but
713 	 * work not done" condition mentioned above.  To close this race
714 	 * window, we need to flush any pending inodegc operations to ensure
715 	 * they have updated the refcountbt records before we try to map the
716 	 * swapfile.
717 	 */
718 	xfs_inodegc_flush(ip->i_mount);
719 
720 	/*
721 	 * Direct the swap code to the correct block device when this file
722 	 * sits on the RT device.
723 	 */
724 	sis->bdev = xfs_inode_buftarg(ip)->bt_bdev;
725 
726 	return iomap_swapfile_activate(sis, swap_file, span,
727 			&xfs_read_iomap_ops);
728 }
729 
730 const struct address_space_operations xfs_address_space_operations = {
731 	.read_folio		= xfs_vm_read_folio,
732 	.readahead		= xfs_vm_readahead,
733 	.writepages		= xfs_vm_writepages,
734 	.dirty_folio		= iomap_dirty_folio,
735 	.release_folio		= iomap_release_folio,
736 	.invalidate_folio	= iomap_invalidate_folio,
737 	.bmap			= xfs_vm_bmap,
738 	.migrate_folio		= filemap_migrate_folio,
739 	.is_partially_uptodate  = iomap_is_partially_uptodate,
740 	.error_remove_folio	= generic_error_remove_folio,
741 	.swap_activate		= xfs_vm_swap_activate,
742 };
743 
744 const struct address_space_operations xfs_dax_aops = {
745 	.writepages		= xfs_dax_writepages,
746 	.dirty_folio		= noop_dirty_folio,
747 	.swap_activate		= xfs_vm_swap_activate,
748 };
749