• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4  * Copyright (c) 2016-2018 Christoph Hellwig.
5  * All Rights Reserved.
6  */
7 #include "xfs.h"
8 #include "xfs_fs.h"
9 #include "xfs_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h"
13 #include "xfs_mount.h"
14 #include "xfs_inode.h"
15 #include "xfs_btree.h"
16 #include "xfs_bmap_btree.h"
17 #include "xfs_bmap.h"
18 #include "xfs_bmap_util.h"
19 #include "xfs_errortag.h"
20 #include "xfs_error.h"
21 #include "xfs_trans.h"
22 #include "xfs_trans_space.h"
23 #include "xfs_inode_item.h"
24 #include "xfs_iomap.h"
25 #include "xfs_trace.h"
26 #include "xfs_quota.h"
27 #include "xfs_dquot_item.h"
28 #include "xfs_dquot.h"
29 #include "xfs_reflink.h"
30 
31 
32 #define XFS_ALLOC_ALIGN(mp, off) \
33 	(((off) >> mp->m_allocsize_log) << mp->m_allocsize_log)
34 
35 static int
xfs_alert_fsblock_zero(xfs_inode_t * ip,xfs_bmbt_irec_t * imap)36 xfs_alert_fsblock_zero(
37 	xfs_inode_t	*ip,
38 	xfs_bmbt_irec_t	*imap)
39 {
40 	xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
41 			"Access to block zero in inode %llu "
42 			"start_block: %llx start_off: %llx "
43 			"blkcnt: %llx extent-state: %x",
44 		(unsigned long long)ip->i_ino,
45 		(unsigned long long)imap->br_startblock,
46 		(unsigned long long)imap->br_startoff,
47 		(unsigned long long)imap->br_blockcount,
48 		imap->br_state);
49 	return -EFSCORRUPTED;
50 }
51 
52 int
xfs_bmbt_to_iomap(struct xfs_inode * ip,struct iomap * iomap,struct xfs_bmbt_irec * imap,u16 flags)53 xfs_bmbt_to_iomap(
54 	struct xfs_inode	*ip,
55 	struct iomap		*iomap,
56 	struct xfs_bmbt_irec	*imap,
57 	u16			flags)
58 {
59 	struct xfs_mount	*mp = ip->i_mount;
60 	struct xfs_buftarg	*target = xfs_inode_buftarg(ip);
61 
62 	if (unlikely(!xfs_valid_startblock(ip, imap->br_startblock)))
63 		return xfs_alert_fsblock_zero(ip, imap);
64 
65 	if (imap->br_startblock == HOLESTARTBLOCK) {
66 		iomap->addr = IOMAP_NULL_ADDR;
67 		iomap->type = IOMAP_HOLE;
68 	} else if (imap->br_startblock == DELAYSTARTBLOCK ||
69 		   isnullstartblock(imap->br_startblock)) {
70 		iomap->addr = IOMAP_NULL_ADDR;
71 		iomap->type = IOMAP_DELALLOC;
72 	} else {
73 		iomap->addr = BBTOB(xfs_fsb_to_db(ip, imap->br_startblock));
74 		if (imap->br_state == XFS_EXT_UNWRITTEN)
75 			iomap->type = IOMAP_UNWRITTEN;
76 		else
77 			iomap->type = IOMAP_MAPPED;
78 	}
79 	iomap->offset = XFS_FSB_TO_B(mp, imap->br_startoff);
80 	iomap->length = XFS_FSB_TO_B(mp, imap->br_blockcount);
81 	iomap->bdev = target->bt_bdev;
82 	iomap->dax_dev = target->bt_daxdev;
83 	iomap->flags = flags;
84 
85 	if (xfs_ipincount(ip) &&
86 	    (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP))
87 		iomap->flags |= IOMAP_F_DIRTY;
88 	return 0;
89 }
90 
91 static void
xfs_hole_to_iomap(struct xfs_inode * ip,struct iomap * iomap,xfs_fileoff_t offset_fsb,xfs_fileoff_t end_fsb)92 xfs_hole_to_iomap(
93 	struct xfs_inode	*ip,
94 	struct iomap		*iomap,
95 	xfs_fileoff_t		offset_fsb,
96 	xfs_fileoff_t		end_fsb)
97 {
98 	struct xfs_buftarg	*target = xfs_inode_buftarg(ip);
99 
100 	iomap->addr = IOMAP_NULL_ADDR;
101 	iomap->type = IOMAP_HOLE;
102 	iomap->offset = XFS_FSB_TO_B(ip->i_mount, offset_fsb);
103 	iomap->length = XFS_FSB_TO_B(ip->i_mount, end_fsb - offset_fsb);
104 	iomap->bdev = target->bt_bdev;
105 	iomap->dax_dev = target->bt_daxdev;
106 }
107 
108 static inline xfs_fileoff_t
xfs_iomap_end_fsb(struct xfs_mount * mp,loff_t offset,loff_t count)109 xfs_iomap_end_fsb(
110 	struct xfs_mount	*mp,
111 	loff_t			offset,
112 	loff_t			count)
113 {
114 	ASSERT(offset <= mp->m_super->s_maxbytes);
115 	return min(XFS_B_TO_FSB(mp, offset + count),
116 		   XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes));
117 }
118 
119 static xfs_extlen_t
xfs_eof_alignment(struct xfs_inode * ip)120 xfs_eof_alignment(
121 	struct xfs_inode	*ip)
122 {
123 	struct xfs_mount	*mp = ip->i_mount;
124 	xfs_extlen_t		align = 0;
125 
126 	if (!XFS_IS_REALTIME_INODE(ip)) {
127 		/*
128 		 * Round up the allocation request to a stripe unit
129 		 * (m_dalign) boundary if the file size is >= stripe unit
130 		 * size, and we are allocating past the allocation eof.
131 		 *
132 		 * If mounted with the "-o swalloc" option the alignment is
133 		 * increased from the strip unit size to the stripe width.
134 		 */
135 		if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
136 			align = mp->m_swidth;
137 		else if (mp->m_dalign)
138 			align = mp->m_dalign;
139 
140 		if (align && XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, align))
141 			align = 0;
142 	}
143 
144 	return align;
145 }
146 
147 /*
148  * Check if last_fsb is outside the last extent, and if so grow it to the next
149  * stripe unit boundary.
150  */
151 xfs_fileoff_t
xfs_iomap_eof_align_last_fsb(struct xfs_inode * ip,xfs_fileoff_t end_fsb)152 xfs_iomap_eof_align_last_fsb(
153 	struct xfs_inode	*ip,
154 	xfs_fileoff_t		end_fsb)
155 {
156 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
157 	xfs_extlen_t		extsz = xfs_get_extsz_hint(ip);
158 	xfs_extlen_t		align = xfs_eof_alignment(ip);
159 	struct xfs_bmbt_irec	irec;
160 	struct xfs_iext_cursor	icur;
161 
162 	ASSERT(ifp->if_flags & XFS_IFEXTENTS);
163 
164 	/*
165 	 * Always round up the allocation request to the extent hint boundary.
166 	 */
167 	if (extsz) {
168 		if (align)
169 			align = roundup_64(align, extsz);
170 		else
171 			align = extsz;
172 	}
173 
174 	if (align) {
175 		xfs_fileoff_t	aligned_end_fsb = roundup_64(end_fsb, align);
176 
177 		xfs_iext_last(ifp, &icur);
178 		if (!xfs_iext_get_extent(ifp, &icur, &irec) ||
179 		    aligned_end_fsb >= irec.br_startoff + irec.br_blockcount)
180 			return aligned_end_fsb;
181 	}
182 
183 	return end_fsb;
184 }
185 
186 int
xfs_iomap_write_direct(struct xfs_inode * ip,xfs_fileoff_t offset_fsb,xfs_fileoff_t count_fsb,struct xfs_bmbt_irec * imap)187 xfs_iomap_write_direct(
188 	struct xfs_inode	*ip,
189 	xfs_fileoff_t		offset_fsb,
190 	xfs_fileoff_t		count_fsb,
191 	struct xfs_bmbt_irec	*imap)
192 {
193 	struct xfs_mount	*mp = ip->i_mount;
194 	struct xfs_trans	*tp;
195 	xfs_filblks_t		resaligned;
196 	int			nimaps;
197 	int			quota_flag;
198 	uint			qblocks, resblks;
199 	unsigned int		resrtextents = 0;
200 	int			error;
201 	int			bmapi_flags = XFS_BMAPI_PREALLOC;
202 	uint			tflags = 0;
203 
204 	ASSERT(count_fsb > 0);
205 
206 	resaligned = xfs_aligned_fsb_count(offset_fsb, count_fsb,
207 					   xfs_get_extsz_hint(ip));
208 	if (unlikely(XFS_IS_REALTIME_INODE(ip))) {
209 		resrtextents = qblocks = resaligned;
210 		resrtextents /= mp->m_sb.sb_rextsize;
211 		resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
212 		quota_flag = XFS_QMOPT_RES_RTBLKS;
213 	} else {
214 		resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
215 		quota_flag = XFS_QMOPT_RES_REGBLKS;
216 	}
217 
218 	error = xfs_qm_dqattach(ip);
219 	if (error)
220 		return error;
221 
222 	/*
223 	 * For DAX, we do not allocate unwritten extents, but instead we zero
224 	 * the block before we commit the transaction.  Ideally we'd like to do
225 	 * this outside the transaction context, but if we commit and then crash
226 	 * we may not have zeroed the blocks and this will be exposed on
227 	 * recovery of the allocation. Hence we must zero before commit.
228 	 *
229 	 * Further, if we are mapping unwritten extents here, we need to zero
230 	 * and convert them to written so that we don't need an unwritten extent
231 	 * callback for DAX. This also means that we need to be able to dip into
232 	 * the reserve block pool for bmbt block allocation if there is no space
233 	 * left but we need to do unwritten extent conversion.
234 	 */
235 	if (IS_DAX(VFS_I(ip))) {
236 		bmapi_flags = XFS_BMAPI_CONVERT | XFS_BMAPI_ZERO;
237 		if (imap->br_state == XFS_EXT_UNWRITTEN) {
238 			tflags |= XFS_TRANS_RESERVE;
239 			resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
240 		}
241 	}
242 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, resrtextents,
243 			tflags, &tp);
244 	if (error)
245 		return error;
246 
247 	xfs_ilock(ip, XFS_ILOCK_EXCL);
248 
249 	error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
250 	if (error)
251 		goto out_trans_cancel;
252 
253 	xfs_trans_ijoin(tp, ip, 0);
254 
255 	/*
256 	 * From this point onwards we overwrite the imap pointer that the
257 	 * caller gave to us.
258 	 */
259 	nimaps = 1;
260 	error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, bmapi_flags, 0,
261 				imap, &nimaps);
262 	if (error)
263 		goto out_res_cancel;
264 
265 	/*
266 	 * Complete the transaction
267 	 */
268 	error = xfs_trans_commit(tp);
269 	if (error)
270 		goto out_unlock;
271 
272 	/*
273 	 * Copy any maps to caller's array and return any error.
274 	 */
275 	if (nimaps == 0) {
276 		error = -ENOSPC;
277 		goto out_unlock;
278 	}
279 
280 	if (unlikely(!xfs_valid_startblock(ip, imap->br_startblock)))
281 		error = xfs_alert_fsblock_zero(ip, imap);
282 
283 out_unlock:
284 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
285 	return error;
286 
287 out_res_cancel:
288 	xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
289 out_trans_cancel:
290 	xfs_trans_cancel(tp);
291 	goto out_unlock;
292 }
293 
294 STATIC bool
xfs_quota_need_throttle(struct xfs_inode * ip,xfs_dqtype_t type,xfs_fsblock_t alloc_blocks)295 xfs_quota_need_throttle(
296 	struct xfs_inode	*ip,
297 	xfs_dqtype_t		type,
298 	xfs_fsblock_t		alloc_blocks)
299 {
300 	struct xfs_dquot	*dq = xfs_inode_dquot(ip, type);
301 
302 	if (!dq || !xfs_this_quota_on(ip->i_mount, type))
303 		return false;
304 
305 	/* no hi watermark, no throttle */
306 	if (!dq->q_prealloc_hi_wmark)
307 		return false;
308 
309 	/* under the lo watermark, no throttle */
310 	if (dq->q_blk.reserved + alloc_blocks < dq->q_prealloc_lo_wmark)
311 		return false;
312 
313 	return true;
314 }
315 
316 STATIC void
xfs_quota_calc_throttle(struct xfs_inode * ip,xfs_dqtype_t type,xfs_fsblock_t * qblocks,int * qshift,int64_t * qfreesp)317 xfs_quota_calc_throttle(
318 	struct xfs_inode	*ip,
319 	xfs_dqtype_t		type,
320 	xfs_fsblock_t		*qblocks,
321 	int			*qshift,
322 	int64_t			*qfreesp)
323 {
324 	struct xfs_dquot	*dq = xfs_inode_dquot(ip, type);
325 	int64_t			freesp;
326 	int			shift = 0;
327 
328 	/* no dq, or over hi wmark, squash the prealloc completely */
329 	if (!dq || dq->q_blk.reserved >= dq->q_prealloc_hi_wmark) {
330 		*qblocks = 0;
331 		*qfreesp = 0;
332 		return;
333 	}
334 
335 	freesp = dq->q_prealloc_hi_wmark - dq->q_blk.reserved;
336 	if (freesp < dq->q_low_space[XFS_QLOWSP_5_PCNT]) {
337 		shift = 2;
338 		if (freesp < dq->q_low_space[XFS_QLOWSP_3_PCNT])
339 			shift += 2;
340 		if (freesp < dq->q_low_space[XFS_QLOWSP_1_PCNT])
341 			shift += 2;
342 	}
343 
344 	if (freesp < *qfreesp)
345 		*qfreesp = freesp;
346 
347 	/* only overwrite the throttle values if we are more aggressive */
348 	if ((freesp >> shift) < (*qblocks >> *qshift)) {
349 		*qblocks = freesp;
350 		*qshift = shift;
351 	}
352 }
353 
354 /*
355  * If we don't have a user specified preallocation size, dynamically increase
356  * the preallocation size as the size of the file grows.  Cap the maximum size
357  * at a single extent or less if the filesystem is near full. The closer the
358  * filesystem is to being full, the smaller the maximum preallocation.
359  */
360 STATIC xfs_fsblock_t
xfs_iomap_prealloc_size(struct xfs_inode * ip,int whichfork,loff_t offset,loff_t count,struct xfs_iext_cursor * icur)361 xfs_iomap_prealloc_size(
362 	struct xfs_inode	*ip,
363 	int			whichfork,
364 	loff_t			offset,
365 	loff_t			count,
366 	struct xfs_iext_cursor	*icur)
367 {
368 	struct xfs_iext_cursor	ncur = *icur;
369 	struct xfs_bmbt_irec	prev, got;
370 	struct xfs_mount	*mp = ip->i_mount;
371 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
372 	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
373 	int64_t			freesp;
374 	xfs_fsblock_t		qblocks;
375 	xfs_fsblock_t		alloc_blocks = 0;
376 	xfs_extlen_t		plen;
377 	int			shift = 0;
378 	int			qshift = 0;
379 
380 	/*
381 	 * As an exception we don't do any preallocation at all if the file is
382 	 * smaller than the minimum preallocation and we are using the default
383 	 * dynamic preallocation scheme, as it is likely this is the only write
384 	 * to the file that is going to be done.
385 	 */
386 	if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_allocsize_blocks))
387 		return 0;
388 
389 	/*
390 	 * Use the minimum preallocation size for small files or if we are
391 	 * writing right after a hole.
392 	 */
393 	if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign) ||
394 	    !xfs_iext_prev_extent(ifp, &ncur, &prev) ||
395 	    prev.br_startoff + prev.br_blockcount < offset_fsb)
396 		return mp->m_allocsize_blocks;
397 
398 	/*
399 	 * Take the size of the preceding data extents as the basis for the
400 	 * preallocation size. Note that we don't care if the previous extents
401 	 * are written or not.
402 	 */
403 	plen = prev.br_blockcount;
404 	while (xfs_iext_prev_extent(ifp, &ncur, &got)) {
405 		if (plen > MAXEXTLEN / 2 ||
406 		    isnullstartblock(got.br_startblock) ||
407 		    got.br_startoff + got.br_blockcount != prev.br_startoff ||
408 		    got.br_startblock + got.br_blockcount != prev.br_startblock)
409 			break;
410 		plen += got.br_blockcount;
411 		prev = got;
412 	}
413 
414 	/*
415 	 * If the size of the extents is greater than half the maximum extent
416 	 * length, then use the current offset as the basis.  This ensures that
417 	 * for large files the preallocation size always extends to MAXEXTLEN
418 	 * rather than falling short due to things like stripe unit/width
419 	 * alignment of real extents.
420 	 */
421 	alloc_blocks = plen * 2;
422 	if (alloc_blocks > MAXEXTLEN)
423 		alloc_blocks = XFS_B_TO_FSB(mp, offset);
424 	qblocks = alloc_blocks;
425 
426 	/*
427 	 * MAXEXTLEN is not a power of two value but we round the prealloc down
428 	 * to the nearest power of two value after throttling. To prevent the
429 	 * round down from unconditionally reducing the maximum supported
430 	 * prealloc size, we round up first, apply appropriate throttling,
431 	 * round down and cap the value to MAXEXTLEN.
432 	 */
433 	alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(MAXEXTLEN),
434 				       alloc_blocks);
435 
436 	freesp = percpu_counter_read_positive(&mp->m_fdblocks);
437 	if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) {
438 		shift = 2;
439 		if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT])
440 			shift++;
441 		if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT])
442 			shift++;
443 		if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT])
444 			shift++;
445 		if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT])
446 			shift++;
447 	}
448 
449 	/*
450 	 * Check each quota to cap the prealloc size, provide a shift value to
451 	 * throttle with and adjust amount of available space.
452 	 */
453 	if (xfs_quota_need_throttle(ip, XFS_DQTYPE_USER, alloc_blocks))
454 		xfs_quota_calc_throttle(ip, XFS_DQTYPE_USER, &qblocks, &qshift,
455 					&freesp);
456 	if (xfs_quota_need_throttle(ip, XFS_DQTYPE_GROUP, alloc_blocks))
457 		xfs_quota_calc_throttle(ip, XFS_DQTYPE_GROUP, &qblocks, &qshift,
458 					&freesp);
459 	if (xfs_quota_need_throttle(ip, XFS_DQTYPE_PROJ, alloc_blocks))
460 		xfs_quota_calc_throttle(ip, XFS_DQTYPE_PROJ, &qblocks, &qshift,
461 					&freesp);
462 
463 	/*
464 	 * The final prealloc size is set to the minimum of free space available
465 	 * in each of the quotas and the overall filesystem.
466 	 *
467 	 * The shift throttle value is set to the maximum value as determined by
468 	 * the global low free space values and per-quota low free space values.
469 	 */
470 	alloc_blocks = min(alloc_blocks, qblocks);
471 	shift = max(shift, qshift);
472 
473 	if (shift)
474 		alloc_blocks >>= shift;
475 	/*
476 	 * rounddown_pow_of_two() returns an undefined result if we pass in
477 	 * alloc_blocks = 0.
478 	 */
479 	if (alloc_blocks)
480 		alloc_blocks = rounddown_pow_of_two(alloc_blocks);
481 	if (alloc_blocks > MAXEXTLEN)
482 		alloc_blocks = MAXEXTLEN;
483 
484 	/*
485 	 * If we are still trying to allocate more space than is
486 	 * available, squash the prealloc hard. This can happen if we
487 	 * have a large file on a small filesystem and the above
488 	 * lowspace thresholds are smaller than MAXEXTLEN.
489 	 */
490 	while (alloc_blocks && alloc_blocks >= freesp)
491 		alloc_blocks >>= 4;
492 	if (alloc_blocks < mp->m_allocsize_blocks)
493 		alloc_blocks = mp->m_allocsize_blocks;
494 	trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift,
495 				      mp->m_allocsize_blocks);
496 	return alloc_blocks;
497 }
498 
499 int
xfs_iomap_write_unwritten(xfs_inode_t * ip,xfs_off_t offset,xfs_off_t count,bool update_isize)500 xfs_iomap_write_unwritten(
501 	xfs_inode_t	*ip,
502 	xfs_off_t	offset,
503 	xfs_off_t	count,
504 	bool		update_isize)
505 {
506 	xfs_mount_t	*mp = ip->i_mount;
507 	xfs_fileoff_t	offset_fsb;
508 	xfs_filblks_t	count_fsb;
509 	xfs_filblks_t	numblks_fsb;
510 	int		nimaps;
511 	xfs_trans_t	*tp;
512 	xfs_bmbt_irec_t imap;
513 	struct inode	*inode = VFS_I(ip);
514 	xfs_fsize_t	i_size;
515 	uint		resblks;
516 	int		error;
517 
518 	trace_xfs_unwritten_convert(ip, offset, count);
519 
520 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
521 	count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
522 	count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
523 
524 	/*
525 	 * Reserve enough blocks in this transaction for two complete extent
526 	 * btree splits.  We may be converting the middle part of an unwritten
527 	 * extent and in this case we will insert two new extents in the btree
528 	 * each of which could cause a full split.
529 	 *
530 	 * This reservation amount will be used in the first call to
531 	 * xfs_bmbt_split() to select an AG with enough space to satisfy the
532 	 * rest of the operation.
533 	 */
534 	resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
535 
536 	/* Attach dquots so that bmbt splits are accounted correctly. */
537 	error = xfs_qm_dqattach(ip);
538 	if (error)
539 		return error;
540 
541 	do {
542 		/*
543 		 * Set up a transaction to convert the range of extents
544 		 * from unwritten to real. Do allocations in a loop until
545 		 * we have covered the range passed in.
546 		 *
547 		 * Note that we can't risk to recursing back into the filesystem
548 		 * here as we might be asked to write out the same inode that we
549 		 * complete here and might deadlock on the iolock.
550 		 */
551 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0,
552 				XFS_TRANS_RESERVE, &tp);
553 		if (error)
554 			return error;
555 
556 		xfs_ilock(ip, XFS_ILOCK_EXCL);
557 		xfs_trans_ijoin(tp, ip, 0);
558 
559 		error = xfs_trans_reserve_quota_nblks(tp, ip, resblks, 0,
560 				XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES);
561 		if (error)
562 			goto error_on_bmapi_transaction;
563 
564 		/*
565 		 * Modify the unwritten extent state of the buffer.
566 		 */
567 		nimaps = 1;
568 		error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
569 					XFS_BMAPI_CONVERT, resblks, &imap,
570 					&nimaps);
571 		if (error)
572 			goto error_on_bmapi_transaction;
573 
574 		/*
575 		 * Log the updated inode size as we go.  We have to be careful
576 		 * to only log it up to the actual write offset if it is
577 		 * halfway into a block.
578 		 */
579 		i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb);
580 		if (i_size > offset + count)
581 			i_size = offset + count;
582 		if (update_isize && i_size > i_size_read(inode))
583 			i_size_write(inode, i_size);
584 		i_size = xfs_new_eof(ip, i_size);
585 		if (i_size) {
586 			ip->i_d.di_size = i_size;
587 			xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
588 		}
589 
590 		error = xfs_trans_commit(tp);
591 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
592 		if (error)
593 			return error;
594 
595 		if (unlikely(!xfs_valid_startblock(ip, imap.br_startblock)))
596 			return xfs_alert_fsblock_zero(ip, &imap);
597 
598 		if ((numblks_fsb = imap.br_blockcount) == 0) {
599 			/*
600 			 * The numblks_fsb value should always get
601 			 * smaller, otherwise the loop is stuck.
602 			 */
603 			ASSERT(imap.br_blockcount);
604 			break;
605 		}
606 		offset_fsb += numblks_fsb;
607 		count_fsb -= numblks_fsb;
608 	} while (count_fsb > 0);
609 
610 	return 0;
611 
612 error_on_bmapi_transaction:
613 	xfs_trans_cancel(tp);
614 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
615 	return error;
616 }
617 
618 static inline bool
imap_needs_alloc(struct inode * inode,unsigned flags,struct xfs_bmbt_irec * imap,int nimaps)619 imap_needs_alloc(
620 	struct inode		*inode,
621 	unsigned		flags,
622 	struct xfs_bmbt_irec	*imap,
623 	int			nimaps)
624 {
625 	/* don't allocate blocks when just zeroing */
626 	if (flags & IOMAP_ZERO)
627 		return false;
628 	if (!nimaps ||
629 	    imap->br_startblock == HOLESTARTBLOCK ||
630 	    imap->br_startblock == DELAYSTARTBLOCK)
631 		return true;
632 	/* we convert unwritten extents before copying the data for DAX */
633 	if (IS_DAX(inode) && imap->br_state == XFS_EXT_UNWRITTEN)
634 		return true;
635 	return false;
636 }
637 
638 static inline bool
imap_needs_cow(struct xfs_inode * ip,unsigned int flags,struct xfs_bmbt_irec * imap,int nimaps)639 imap_needs_cow(
640 	struct xfs_inode	*ip,
641 	unsigned int		flags,
642 	struct xfs_bmbt_irec	*imap,
643 	int			nimaps)
644 {
645 	if (!xfs_is_cow_inode(ip))
646 		return false;
647 
648 	/* when zeroing we don't have to COW holes or unwritten extents */
649 	if (flags & IOMAP_ZERO) {
650 		if (!nimaps ||
651 		    imap->br_startblock == HOLESTARTBLOCK ||
652 		    imap->br_state == XFS_EXT_UNWRITTEN)
653 			return false;
654 	}
655 
656 	return true;
657 }
658 
659 static int
xfs_ilock_for_iomap(struct xfs_inode * ip,unsigned flags,unsigned * lockmode)660 xfs_ilock_for_iomap(
661 	struct xfs_inode	*ip,
662 	unsigned		flags,
663 	unsigned		*lockmode)
664 {
665 	unsigned		mode = XFS_ILOCK_SHARED;
666 	bool			is_write = flags & (IOMAP_WRITE | IOMAP_ZERO);
667 
668 	/*
669 	 * COW writes may allocate delalloc space or convert unwritten COW
670 	 * extents, so we need to make sure to take the lock exclusively here.
671 	 */
672 	if (xfs_is_cow_inode(ip) && is_write)
673 		mode = XFS_ILOCK_EXCL;
674 
675 	/*
676 	 * Extents not yet cached requires exclusive access, don't block.  This
677 	 * is an opencoded xfs_ilock_data_map_shared() call but with
678 	 * non-blocking behaviour.
679 	 */
680 	if (!(ip->i_df.if_flags & XFS_IFEXTENTS)) {
681 		if (flags & IOMAP_NOWAIT)
682 			return -EAGAIN;
683 		mode = XFS_ILOCK_EXCL;
684 	}
685 
686 relock:
687 	if (flags & IOMAP_NOWAIT) {
688 		if (!xfs_ilock_nowait(ip, mode))
689 			return -EAGAIN;
690 	} else {
691 		xfs_ilock(ip, mode);
692 	}
693 
694 	/*
695 	 * The reflink iflag could have changed since the earlier unlocked
696 	 * check, so if we got ILOCK_SHARED for a write and but we're now a
697 	 * reflink inode we have to switch to ILOCK_EXCL and relock.
698 	 */
699 	if (mode == XFS_ILOCK_SHARED && is_write && xfs_is_cow_inode(ip)) {
700 		xfs_iunlock(ip, mode);
701 		mode = XFS_ILOCK_EXCL;
702 		goto relock;
703 	}
704 
705 	*lockmode = mode;
706 	return 0;
707 }
708 
709 /*
710  * Check that the imap we are going to return to the caller spans the entire
711  * range that the caller requested for the IO.
712  */
713 static bool
imap_spans_range(struct xfs_bmbt_irec * imap,xfs_fileoff_t offset_fsb,xfs_fileoff_t end_fsb)714 imap_spans_range(
715 	struct xfs_bmbt_irec	*imap,
716 	xfs_fileoff_t		offset_fsb,
717 	xfs_fileoff_t		end_fsb)
718 {
719 	if (imap->br_startoff > offset_fsb)
720 		return false;
721 	if (imap->br_startoff + imap->br_blockcount < end_fsb)
722 		return false;
723 	return true;
724 }
725 
726 static int
xfs_direct_write_iomap_begin(struct inode * inode,loff_t offset,loff_t length,unsigned flags,struct iomap * iomap,struct iomap * srcmap)727 xfs_direct_write_iomap_begin(
728 	struct inode		*inode,
729 	loff_t			offset,
730 	loff_t			length,
731 	unsigned		flags,
732 	struct iomap		*iomap,
733 	struct iomap		*srcmap)
734 {
735 	struct xfs_inode	*ip = XFS_I(inode);
736 	struct xfs_mount	*mp = ip->i_mount;
737 	struct xfs_bmbt_irec	imap, cmap;
738 	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
739 	xfs_fileoff_t		end_fsb = xfs_iomap_end_fsb(mp, offset, length);
740 	int			nimaps = 1, error = 0;
741 	bool			shared = false;
742 	u16			iomap_flags = 0;
743 	unsigned		lockmode;
744 
745 	ASSERT(flags & (IOMAP_WRITE | IOMAP_ZERO));
746 
747 	if (XFS_FORCED_SHUTDOWN(mp))
748 		return -EIO;
749 
750 	/*
751 	 * Writes that span EOF might trigger an IO size update on completion,
752 	 * so consider them to be dirty for the purposes of O_DSYNC even if
753 	 * there is no other metadata changes pending or have been made here.
754 	 */
755 	if (offset + length > i_size_read(inode))
756 		iomap_flags |= IOMAP_F_DIRTY;
757 
758 	error = xfs_ilock_for_iomap(ip, flags, &lockmode);
759 	if (error)
760 		return error;
761 
762 	error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
763 			       &nimaps, 0);
764 	if (error)
765 		goto out_unlock;
766 
767 	if (imap_needs_cow(ip, flags, &imap, nimaps)) {
768 		error = -EAGAIN;
769 		if (flags & IOMAP_NOWAIT)
770 			goto out_unlock;
771 
772 		/* may drop and re-acquire the ilock */
773 		error = xfs_reflink_allocate_cow(ip, &imap, &cmap, &shared,
774 				&lockmode, flags & IOMAP_DIRECT);
775 		if (error)
776 			goto out_unlock;
777 		if (shared)
778 			goto out_found_cow;
779 		end_fsb = imap.br_startoff + imap.br_blockcount;
780 		length = XFS_FSB_TO_B(mp, end_fsb) - offset;
781 	}
782 
783 	if (imap_needs_alloc(inode, flags, &imap, nimaps))
784 		goto allocate_blocks;
785 
786 	/*
787 	 * NOWAIT IO needs to span the entire requested IO with a single map so
788 	 * that we avoid partial IO failures due to the rest of the IO range not
789 	 * covered by this map triggering an EAGAIN condition when it is
790 	 * subsequently mapped and aborting the IO.
791 	 */
792 	if ((flags & IOMAP_NOWAIT) &&
793 	    !imap_spans_range(&imap, offset_fsb, end_fsb)) {
794 		error = -EAGAIN;
795 		goto out_unlock;
796 	}
797 
798 	xfs_iunlock(ip, lockmode);
799 	trace_xfs_iomap_found(ip, offset, length, XFS_DATA_FORK, &imap);
800 	return xfs_bmbt_to_iomap(ip, iomap, &imap, iomap_flags);
801 
802 allocate_blocks:
803 	error = -EAGAIN;
804 	if (flags & IOMAP_NOWAIT)
805 		goto out_unlock;
806 
807 	/*
808 	 * We cap the maximum length we map to a sane size  to keep the chunks
809 	 * of work done where somewhat symmetric with the work writeback does.
810 	 * This is a completely arbitrary number pulled out of thin air as a
811 	 * best guess for initial testing.
812 	 *
813 	 * Note that the values needs to be less than 32-bits wide until the
814 	 * lower level functions are updated.
815 	 */
816 	length = min_t(loff_t, length, 1024 * PAGE_SIZE);
817 	end_fsb = xfs_iomap_end_fsb(mp, offset, length);
818 
819 	if (offset + length > XFS_ISIZE(ip))
820 		end_fsb = xfs_iomap_eof_align_last_fsb(ip, end_fsb);
821 	else if (nimaps && imap.br_startblock == HOLESTARTBLOCK)
822 		end_fsb = min(end_fsb, imap.br_startoff + imap.br_blockcount);
823 	xfs_iunlock(ip, lockmode);
824 
825 	error = xfs_iomap_write_direct(ip, offset_fsb, end_fsb - offset_fsb,
826 			&imap);
827 	if (error)
828 		return error;
829 
830 	trace_xfs_iomap_alloc(ip, offset, length, XFS_DATA_FORK, &imap);
831 	return xfs_bmbt_to_iomap(ip, iomap, &imap, iomap_flags | IOMAP_F_NEW);
832 
833 out_found_cow:
834 	xfs_iunlock(ip, lockmode);
835 	length = XFS_FSB_TO_B(mp, cmap.br_startoff + cmap.br_blockcount);
836 	trace_xfs_iomap_found(ip, offset, length - offset, XFS_COW_FORK, &cmap);
837 	if (imap.br_startblock != HOLESTARTBLOCK) {
838 		error = xfs_bmbt_to_iomap(ip, srcmap, &imap, 0);
839 		if (error)
840 			return error;
841 	}
842 	return xfs_bmbt_to_iomap(ip, iomap, &cmap, IOMAP_F_SHARED);
843 
844 out_unlock:
845 	xfs_iunlock(ip, lockmode);
846 	return error;
847 }
848 
849 const struct iomap_ops xfs_direct_write_iomap_ops = {
850 	.iomap_begin		= xfs_direct_write_iomap_begin,
851 };
852 
853 static int
xfs_buffered_write_iomap_begin(struct inode * inode,loff_t offset,loff_t count,unsigned flags,struct iomap * iomap,struct iomap * srcmap)854 xfs_buffered_write_iomap_begin(
855 	struct inode		*inode,
856 	loff_t			offset,
857 	loff_t			count,
858 	unsigned		flags,
859 	struct iomap		*iomap,
860 	struct iomap		*srcmap)
861 {
862 	struct xfs_inode	*ip = XFS_I(inode);
863 	struct xfs_mount	*mp = ip->i_mount;
864 	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
865 	xfs_fileoff_t		end_fsb = xfs_iomap_end_fsb(mp, offset, count);
866 	struct xfs_bmbt_irec	imap, cmap;
867 	struct xfs_iext_cursor	icur, ccur;
868 	xfs_fsblock_t		prealloc_blocks = 0;
869 	bool			eof = false, cow_eof = false, shared = false;
870 	int			allocfork = XFS_DATA_FORK;
871 	int			error = 0;
872 
873 	if (XFS_FORCED_SHUTDOWN(mp))
874 		return -EIO;
875 
876 	/* we can't use delayed allocations when using extent size hints */
877 	if (xfs_get_extsz_hint(ip))
878 		return xfs_direct_write_iomap_begin(inode, offset, count,
879 				flags, iomap, srcmap);
880 
881 	ASSERT(!XFS_IS_REALTIME_INODE(ip));
882 
883 	xfs_ilock(ip, XFS_ILOCK_EXCL);
884 
885 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(&ip->i_df)) ||
886 	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
887 		error = -EFSCORRUPTED;
888 		goto out_unlock;
889 	}
890 
891 	XFS_STATS_INC(mp, xs_blk_mapw);
892 
893 	if (!(ip->i_df.if_flags & XFS_IFEXTENTS)) {
894 		error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
895 		if (error)
896 			goto out_unlock;
897 	}
898 
899 	/*
900 	 * Search the data fork first to look up our source mapping.  We
901 	 * always need the data fork map, as we have to return it to the
902 	 * iomap code so that the higher level write code can read data in to
903 	 * perform read-modify-write cycles for unaligned writes.
904 	 */
905 	eof = !xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap);
906 	if (eof)
907 		imap.br_startoff = end_fsb; /* fake hole until the end */
908 
909 	/* We never need to allocate blocks for zeroing a hole. */
910 	if ((flags & IOMAP_ZERO) && imap.br_startoff > offset_fsb) {
911 		xfs_hole_to_iomap(ip, iomap, offset_fsb, imap.br_startoff);
912 		goto out_unlock;
913 	}
914 
915 	/*
916 	 * Search the COW fork extent list even if we did not find a data fork
917 	 * extent.  This serves two purposes: first this implements the
918 	 * speculative preallocation using cowextsize, so that we also unshare
919 	 * block adjacent to shared blocks instead of just the shared blocks
920 	 * themselves.  Second the lookup in the extent list is generally faster
921 	 * than going out to the shared extent tree.
922 	 */
923 	if (xfs_is_cow_inode(ip)) {
924 		if (!ip->i_cowfp) {
925 			ASSERT(!xfs_is_reflink_inode(ip));
926 			xfs_ifork_init_cow(ip);
927 		}
928 		cow_eof = !xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb,
929 				&ccur, &cmap);
930 		if (!cow_eof && cmap.br_startoff <= offset_fsb) {
931 			trace_xfs_reflink_cow_found(ip, &cmap);
932 			goto found_cow;
933 		}
934 	}
935 
936 	if (imap.br_startoff <= offset_fsb) {
937 		/*
938 		 * For reflink files we may need a delalloc reservation when
939 		 * overwriting shared extents.   This includes zeroing of
940 		 * existing extents that contain data.
941 		 */
942 		if (!xfs_is_cow_inode(ip) ||
943 		    ((flags & IOMAP_ZERO) && imap.br_state != XFS_EXT_NORM)) {
944 			trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK,
945 					&imap);
946 			goto found_imap;
947 		}
948 
949 		xfs_trim_extent(&imap, offset_fsb, end_fsb - offset_fsb);
950 
951 		/* Trim the mapping to the nearest shared extent boundary. */
952 		error = xfs_bmap_trim_cow(ip, &imap, &shared);
953 		if (error)
954 			goto out_unlock;
955 
956 		/* Not shared?  Just report the (potentially capped) extent. */
957 		if (!shared) {
958 			trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK,
959 					&imap);
960 			goto found_imap;
961 		}
962 
963 		/*
964 		 * Fork all the shared blocks from our write offset until the
965 		 * end of the extent.
966 		 */
967 		allocfork = XFS_COW_FORK;
968 		end_fsb = imap.br_startoff + imap.br_blockcount;
969 	} else {
970 		/*
971 		 * We cap the maximum length we map here to MAX_WRITEBACK_PAGES
972 		 * pages to keep the chunks of work done where somewhat
973 		 * symmetric with the work writeback does.  This is a completely
974 		 * arbitrary number pulled out of thin air.
975 		 *
976 		 * Note that the values needs to be less than 32-bits wide until
977 		 * the lower level functions are updated.
978 		 */
979 		count = min_t(loff_t, count, 1024 * PAGE_SIZE);
980 		end_fsb = xfs_iomap_end_fsb(mp, offset, count);
981 
982 		if (xfs_is_always_cow_inode(ip))
983 			allocfork = XFS_COW_FORK;
984 	}
985 
986 	error = xfs_qm_dqattach_locked(ip, false);
987 	if (error)
988 		goto out_unlock;
989 
990 	if (eof && offset + count > XFS_ISIZE(ip)) {
991 		/*
992 		 * Determine the initial size of the preallocation.
993 		 * We clean up any extra preallocation when the file is closed.
994 		 */
995 		if (mp->m_flags & XFS_MOUNT_ALLOCSIZE)
996 			prealloc_blocks = mp->m_allocsize_blocks;
997 		else
998 			prealloc_blocks = xfs_iomap_prealloc_size(ip, allocfork,
999 						offset, count, &icur);
1000 		if (prealloc_blocks) {
1001 			xfs_extlen_t	align;
1002 			xfs_off_t	end_offset;
1003 			xfs_fileoff_t	p_end_fsb;
1004 
1005 			end_offset = XFS_ALLOC_ALIGN(mp, offset + count - 1);
1006 			p_end_fsb = XFS_B_TO_FSBT(mp, end_offset) +
1007 					prealloc_blocks;
1008 
1009 			align = xfs_eof_alignment(ip);
1010 			if (align)
1011 				p_end_fsb = roundup_64(p_end_fsb, align);
1012 
1013 			p_end_fsb = min(p_end_fsb,
1014 				XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes));
1015 			ASSERT(p_end_fsb > offset_fsb);
1016 			prealloc_blocks = p_end_fsb - end_fsb;
1017 		}
1018 	}
1019 
1020 retry:
1021 	error = xfs_bmapi_reserve_delalloc(ip, allocfork, offset_fsb,
1022 			end_fsb - offset_fsb, prealloc_blocks,
1023 			allocfork == XFS_DATA_FORK ? &imap : &cmap,
1024 			allocfork == XFS_DATA_FORK ? &icur : &ccur,
1025 			allocfork == XFS_DATA_FORK ? eof : cow_eof);
1026 	switch (error) {
1027 	case 0:
1028 		break;
1029 	case -ENOSPC:
1030 	case -EDQUOT:
1031 		/* retry without any preallocation */
1032 		trace_xfs_delalloc_enospc(ip, offset, count);
1033 		if (prealloc_blocks) {
1034 			prealloc_blocks = 0;
1035 			goto retry;
1036 		}
1037 		/*FALLTHRU*/
1038 	default:
1039 		goto out_unlock;
1040 	}
1041 
1042 	if (allocfork == XFS_COW_FORK) {
1043 		trace_xfs_iomap_alloc(ip, offset, count, allocfork, &cmap);
1044 		goto found_cow;
1045 	}
1046 
1047 	/*
1048 	 * Flag newly allocated delalloc blocks with IOMAP_F_NEW so we punch
1049 	 * them out if the write happens to fail.
1050 	 */
1051 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1052 	trace_xfs_iomap_alloc(ip, offset, count, allocfork, &imap);
1053 	return xfs_bmbt_to_iomap(ip, iomap, &imap, IOMAP_F_NEW);
1054 
1055 found_imap:
1056 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1057 	return xfs_bmbt_to_iomap(ip, iomap, &imap, 0);
1058 
1059 found_cow:
1060 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1061 	if (imap.br_startoff <= offset_fsb) {
1062 		error = xfs_bmbt_to_iomap(ip, srcmap, &imap, 0);
1063 		if (error)
1064 			return error;
1065 		return xfs_bmbt_to_iomap(ip, iomap, &cmap, IOMAP_F_SHARED);
1066 	}
1067 
1068 	xfs_trim_extent(&cmap, offset_fsb, imap.br_startoff - offset_fsb);
1069 	return xfs_bmbt_to_iomap(ip, iomap, &cmap, 0);
1070 
1071 out_unlock:
1072 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1073 	return error;
1074 }
1075 
1076 static int
xfs_buffered_write_iomap_end(struct inode * inode,loff_t offset,loff_t length,ssize_t written,unsigned flags,struct iomap * iomap)1077 xfs_buffered_write_iomap_end(
1078 	struct inode		*inode,
1079 	loff_t			offset,
1080 	loff_t			length,
1081 	ssize_t			written,
1082 	unsigned		flags,
1083 	struct iomap		*iomap)
1084 {
1085 	struct xfs_inode	*ip = XFS_I(inode);
1086 	struct xfs_mount	*mp = ip->i_mount;
1087 	xfs_fileoff_t		start_fsb;
1088 	xfs_fileoff_t		end_fsb;
1089 	int			error = 0;
1090 
1091 	if (iomap->type != IOMAP_DELALLOC)
1092 		return 0;
1093 
1094 	/*
1095 	 * Behave as if the write failed if drop writes is enabled. Set the NEW
1096 	 * flag to force delalloc cleanup.
1097 	 */
1098 	if (XFS_TEST_ERROR(false, mp, XFS_ERRTAG_DROP_WRITES)) {
1099 		iomap->flags |= IOMAP_F_NEW;
1100 		written = 0;
1101 	}
1102 
1103 	/*
1104 	 * start_fsb refers to the first unused block after a short write. If
1105 	 * nothing was written, round offset down to point at the first block in
1106 	 * the range.
1107 	 */
1108 	if (unlikely(!written))
1109 		start_fsb = XFS_B_TO_FSBT(mp, offset);
1110 	else
1111 		start_fsb = XFS_B_TO_FSB(mp, offset + written);
1112 	end_fsb = XFS_B_TO_FSB(mp, offset + length);
1113 
1114 	/*
1115 	 * Trim delalloc blocks if they were allocated by this write and we
1116 	 * didn't manage to write the whole range.
1117 	 *
1118 	 * We don't need to care about racing delalloc as we hold i_mutex
1119 	 * across the reserve/allocate/unreserve calls. If there are delalloc
1120 	 * blocks in the range, they are ours.
1121 	 */
1122 	if ((iomap->flags & IOMAP_F_NEW) && start_fsb < end_fsb) {
1123 		truncate_pagecache_range(VFS_I(ip), XFS_FSB_TO_B(mp, start_fsb),
1124 					 XFS_FSB_TO_B(mp, end_fsb) - 1);
1125 
1126 		error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
1127 					       end_fsb - start_fsb);
1128 		if (error && !XFS_FORCED_SHUTDOWN(mp)) {
1129 			xfs_alert(mp, "%s: unable to clean up ino %lld",
1130 				__func__, ip->i_ino);
1131 			return error;
1132 		}
1133 	}
1134 
1135 	return 0;
1136 }
1137 
1138 const struct iomap_ops xfs_buffered_write_iomap_ops = {
1139 	.iomap_begin		= xfs_buffered_write_iomap_begin,
1140 	.iomap_end		= xfs_buffered_write_iomap_end,
1141 };
1142 
1143 static int
xfs_read_iomap_begin(struct inode * inode,loff_t offset,loff_t length,unsigned flags,struct iomap * iomap,struct iomap * srcmap)1144 xfs_read_iomap_begin(
1145 	struct inode		*inode,
1146 	loff_t			offset,
1147 	loff_t			length,
1148 	unsigned		flags,
1149 	struct iomap		*iomap,
1150 	struct iomap		*srcmap)
1151 {
1152 	struct xfs_inode	*ip = XFS_I(inode);
1153 	struct xfs_mount	*mp = ip->i_mount;
1154 	struct xfs_bmbt_irec	imap;
1155 	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
1156 	xfs_fileoff_t		end_fsb = xfs_iomap_end_fsb(mp, offset, length);
1157 	int			nimaps = 1, error = 0;
1158 	bool			shared = false;
1159 	unsigned		lockmode;
1160 
1161 	ASSERT(!(flags & (IOMAP_WRITE | IOMAP_ZERO)));
1162 
1163 	if (XFS_FORCED_SHUTDOWN(mp))
1164 		return -EIO;
1165 
1166 	error = xfs_ilock_for_iomap(ip, flags, &lockmode);
1167 	if (error)
1168 		return error;
1169 	error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1170 			       &nimaps, 0);
1171 	if (!error && (flags & IOMAP_REPORT))
1172 		error = xfs_reflink_trim_around_shared(ip, &imap, &shared);
1173 	xfs_iunlock(ip, lockmode);
1174 
1175 	if (error)
1176 		return error;
1177 	trace_xfs_iomap_found(ip, offset, length, XFS_DATA_FORK, &imap);
1178 	return xfs_bmbt_to_iomap(ip, iomap, &imap, shared ? IOMAP_F_SHARED : 0);
1179 }
1180 
1181 const struct iomap_ops xfs_read_iomap_ops = {
1182 	.iomap_begin		= xfs_read_iomap_begin,
1183 };
1184 
1185 static int
xfs_seek_iomap_begin(struct inode * inode,loff_t offset,loff_t length,unsigned flags,struct iomap * iomap,struct iomap * srcmap)1186 xfs_seek_iomap_begin(
1187 	struct inode		*inode,
1188 	loff_t			offset,
1189 	loff_t			length,
1190 	unsigned		flags,
1191 	struct iomap		*iomap,
1192 	struct iomap		*srcmap)
1193 {
1194 	struct xfs_inode	*ip = XFS_I(inode);
1195 	struct xfs_mount	*mp = ip->i_mount;
1196 	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
1197 	xfs_fileoff_t		end_fsb = XFS_B_TO_FSB(mp, offset + length);
1198 	xfs_fileoff_t		cow_fsb = NULLFILEOFF, data_fsb = NULLFILEOFF;
1199 	struct xfs_iext_cursor	icur;
1200 	struct xfs_bmbt_irec	imap, cmap;
1201 	int			error = 0;
1202 	unsigned		lockmode;
1203 
1204 	if (XFS_FORCED_SHUTDOWN(mp))
1205 		return -EIO;
1206 
1207 	lockmode = xfs_ilock_data_map_shared(ip);
1208 	if (!(ip->i_df.if_flags & XFS_IFEXTENTS)) {
1209 		error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
1210 		if (error)
1211 			goto out_unlock;
1212 	}
1213 
1214 	if (xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap)) {
1215 		/*
1216 		 * If we found a data extent we are done.
1217 		 */
1218 		if (imap.br_startoff <= offset_fsb)
1219 			goto done;
1220 		data_fsb = imap.br_startoff;
1221 	} else {
1222 		/*
1223 		 * Fake a hole until the end of the file.
1224 		 */
1225 		data_fsb = xfs_iomap_end_fsb(mp, offset, length);
1226 	}
1227 
1228 	/*
1229 	 * If a COW fork extent covers the hole, report it - capped to the next
1230 	 * data fork extent:
1231 	 */
1232 	if (xfs_inode_has_cow_data(ip) &&
1233 	    xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb, &icur, &cmap))
1234 		cow_fsb = cmap.br_startoff;
1235 	if (cow_fsb != NULLFILEOFF && cow_fsb <= offset_fsb) {
1236 		if (data_fsb < cow_fsb + cmap.br_blockcount)
1237 			end_fsb = min(end_fsb, data_fsb);
1238 		xfs_trim_extent(&cmap, offset_fsb, end_fsb);
1239 		error = xfs_bmbt_to_iomap(ip, iomap, &cmap, IOMAP_F_SHARED);
1240 		/*
1241 		 * This is a COW extent, so we must probe the page cache
1242 		 * because there could be dirty page cache being backed
1243 		 * by this extent.
1244 		 */
1245 		iomap->type = IOMAP_UNWRITTEN;
1246 		goto out_unlock;
1247 	}
1248 
1249 	/*
1250 	 * Else report a hole, capped to the next found data or COW extent.
1251 	 */
1252 	if (cow_fsb != NULLFILEOFF && cow_fsb < data_fsb)
1253 		imap.br_blockcount = cow_fsb - offset_fsb;
1254 	else
1255 		imap.br_blockcount = data_fsb - offset_fsb;
1256 	imap.br_startoff = offset_fsb;
1257 	imap.br_startblock = HOLESTARTBLOCK;
1258 	imap.br_state = XFS_EXT_NORM;
1259 done:
1260 	xfs_trim_extent(&imap, offset_fsb, end_fsb);
1261 	error = xfs_bmbt_to_iomap(ip, iomap, &imap, 0);
1262 out_unlock:
1263 	xfs_iunlock(ip, lockmode);
1264 	return error;
1265 }
1266 
1267 const struct iomap_ops xfs_seek_iomap_ops = {
1268 	.iomap_begin		= xfs_seek_iomap_begin,
1269 };
1270 
1271 static int
xfs_xattr_iomap_begin(struct inode * inode,loff_t offset,loff_t length,unsigned flags,struct iomap * iomap,struct iomap * srcmap)1272 xfs_xattr_iomap_begin(
1273 	struct inode		*inode,
1274 	loff_t			offset,
1275 	loff_t			length,
1276 	unsigned		flags,
1277 	struct iomap		*iomap,
1278 	struct iomap		*srcmap)
1279 {
1280 	struct xfs_inode	*ip = XFS_I(inode);
1281 	struct xfs_mount	*mp = ip->i_mount;
1282 	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
1283 	xfs_fileoff_t		end_fsb = XFS_B_TO_FSB(mp, offset + length);
1284 	struct xfs_bmbt_irec	imap;
1285 	int			nimaps = 1, error = 0;
1286 	unsigned		lockmode;
1287 
1288 	if (XFS_FORCED_SHUTDOWN(mp))
1289 		return -EIO;
1290 
1291 	lockmode = xfs_ilock_attr_map_shared(ip);
1292 
1293 	/* if there are no attribute fork or extents, return ENOENT */
1294 	if (!XFS_IFORK_Q(ip) || !ip->i_afp->if_nextents) {
1295 		error = -ENOENT;
1296 		goto out_unlock;
1297 	}
1298 
1299 	ASSERT(ip->i_afp->if_format != XFS_DINODE_FMT_LOCAL);
1300 	error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1301 			       &nimaps, XFS_BMAPI_ATTRFORK);
1302 out_unlock:
1303 	xfs_iunlock(ip, lockmode);
1304 
1305 	if (error)
1306 		return error;
1307 	ASSERT(nimaps);
1308 	return xfs_bmbt_to_iomap(ip, iomap, &imap, 0);
1309 }
1310 
1311 const struct iomap_ops xfs_xattr_iomap_ops = {
1312 	.iomap_begin		= xfs_xattr_iomap_begin,
1313 };
1314