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 <linux/iomap.h>
8 #include "xfs.h"
9 #include "xfs_fs.h"
10 #include "xfs_shared.h"
11 #include "xfs_format.h"
12 #include "xfs_log_format.h"
13 #include "xfs_trans_resv.h"
14 #include "xfs_mount.h"
15 #include "xfs_defer.h"
16 #include "xfs_inode.h"
17 #include "xfs_btree.h"
18 #include "xfs_bmap_btree.h"
19 #include "xfs_bmap.h"
20 #include "xfs_bmap_util.h"
21 #include "xfs_errortag.h"
22 #include "xfs_error.h"
23 #include "xfs_trans.h"
24 #include "xfs_trans_space.h"
25 #include "xfs_inode_item.h"
26 #include "xfs_iomap.h"
27 #include "xfs_trace.h"
28 #include "xfs_icache.h"
29 #include "xfs_quota.h"
30 #include "xfs_dquot_item.h"
31 #include "xfs_dquot.h"
32 #include "xfs_reflink.h"
33 
34 
35 #define XFS_WRITEIO_ALIGN(mp,off)	(((off) >> mp->m_writeio_log) \
36 						<< mp->m_writeio_log)
37 
38 void
xfs_bmbt_to_iomap(struct xfs_inode * ip,struct iomap * iomap,struct xfs_bmbt_irec * imap)39 xfs_bmbt_to_iomap(
40 	struct xfs_inode	*ip,
41 	struct iomap		*iomap,
42 	struct xfs_bmbt_irec	*imap)
43 {
44 	struct xfs_mount	*mp = ip->i_mount;
45 
46 	if (imap->br_startblock == HOLESTARTBLOCK) {
47 		iomap->addr = IOMAP_NULL_ADDR;
48 		iomap->type = IOMAP_HOLE;
49 	} else if (imap->br_startblock == DELAYSTARTBLOCK) {
50 		iomap->addr = IOMAP_NULL_ADDR;
51 		iomap->type = IOMAP_DELALLOC;
52 	} else {
53 		iomap->addr = BBTOB(xfs_fsb_to_db(ip, imap->br_startblock));
54 		if (imap->br_state == XFS_EXT_UNWRITTEN)
55 			iomap->type = IOMAP_UNWRITTEN;
56 		else
57 			iomap->type = IOMAP_MAPPED;
58 	}
59 	iomap->offset = XFS_FSB_TO_B(mp, imap->br_startoff);
60 	iomap->length = XFS_FSB_TO_B(mp, imap->br_blockcount);
61 	iomap->bdev = xfs_find_bdev_for_inode(VFS_I(ip));
62 	iomap->dax_dev = xfs_find_daxdev_for_inode(VFS_I(ip));
63 }
64 
65 xfs_extlen_t
xfs_eof_alignment(struct xfs_inode * ip,xfs_extlen_t extsize)66 xfs_eof_alignment(
67 	struct xfs_inode	*ip,
68 	xfs_extlen_t		extsize)
69 {
70 	struct xfs_mount	*mp = ip->i_mount;
71 	xfs_extlen_t		align = 0;
72 
73 	if (!XFS_IS_REALTIME_INODE(ip)) {
74 		/*
75 		 * Round up the allocation request to a stripe unit
76 		 * (m_dalign) boundary if the file size is >= stripe unit
77 		 * size, and we are allocating past the allocation eof.
78 		 *
79 		 * If mounted with the "-o swalloc" option the alignment is
80 		 * increased from the strip unit size to the stripe width.
81 		 */
82 		if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
83 			align = mp->m_swidth;
84 		else if (mp->m_dalign)
85 			align = mp->m_dalign;
86 
87 		if (align && XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, align))
88 			align = 0;
89 	}
90 
91 	/*
92 	 * Always round up the allocation request to an extent boundary
93 	 * (when file on a real-time subvolume or has di_extsize hint).
94 	 */
95 	if (extsize) {
96 		if (align)
97 			align = roundup_64(align, extsize);
98 		else
99 			align = extsize;
100 	}
101 
102 	return align;
103 }
104 
105 STATIC int
xfs_iomap_eof_align_last_fsb(struct xfs_inode * ip,xfs_extlen_t extsize,xfs_fileoff_t * last_fsb)106 xfs_iomap_eof_align_last_fsb(
107 	struct xfs_inode	*ip,
108 	xfs_extlen_t		extsize,
109 	xfs_fileoff_t		*last_fsb)
110 {
111 	xfs_extlen_t		align = xfs_eof_alignment(ip, extsize);
112 
113 	if (align) {
114 		xfs_fileoff_t	new_last_fsb = roundup_64(*last_fsb, align);
115 		int		eof, error;
116 
117 		error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
118 		if (error)
119 			return error;
120 		if (eof)
121 			*last_fsb = new_last_fsb;
122 	}
123 	return 0;
124 }
125 
126 STATIC int
xfs_alert_fsblock_zero(xfs_inode_t * ip,xfs_bmbt_irec_t * imap)127 xfs_alert_fsblock_zero(
128 	xfs_inode_t	*ip,
129 	xfs_bmbt_irec_t	*imap)
130 {
131 	xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
132 			"Access to block zero in inode %llu "
133 			"start_block: %llx start_off: %llx "
134 			"blkcnt: %llx extent-state: %x",
135 		(unsigned long long)ip->i_ino,
136 		(unsigned long long)imap->br_startblock,
137 		(unsigned long long)imap->br_startoff,
138 		(unsigned long long)imap->br_blockcount,
139 		imap->br_state);
140 	return -EFSCORRUPTED;
141 }
142 
143 int
xfs_iomap_write_direct(xfs_inode_t * ip,xfs_off_t offset,size_t count,xfs_bmbt_irec_t * imap,int nmaps)144 xfs_iomap_write_direct(
145 	xfs_inode_t	*ip,
146 	xfs_off_t	offset,
147 	size_t		count,
148 	xfs_bmbt_irec_t *imap,
149 	int		nmaps)
150 {
151 	xfs_mount_t	*mp = ip->i_mount;
152 	xfs_fileoff_t	offset_fsb;
153 	xfs_fileoff_t	last_fsb;
154 	xfs_filblks_t	count_fsb, resaligned;
155 	xfs_extlen_t	extsz;
156 	int		nimaps;
157 	int		quota_flag;
158 	int		rt;
159 	xfs_trans_t	*tp;
160 	uint		qblocks, resblks, resrtextents;
161 	int		error;
162 	int		lockmode;
163 	int		bmapi_flags = XFS_BMAPI_PREALLOC;
164 	uint		tflags = 0;
165 
166 	rt = XFS_IS_REALTIME_INODE(ip);
167 	extsz = xfs_get_extsz_hint(ip);
168 	lockmode = XFS_ILOCK_SHARED;	/* locked by caller */
169 
170 	ASSERT(xfs_isilocked(ip, lockmode));
171 
172 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
173 	last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
174 	if ((offset + count) > XFS_ISIZE(ip)) {
175 		/*
176 		 * Assert that the in-core extent list is present since this can
177 		 * call xfs_iread_extents() and we only have the ilock shared.
178 		 * This should be safe because the lock was held around a bmapi
179 		 * call in the caller and we only need it to access the in-core
180 		 * list.
181 		 */
182 		ASSERT(XFS_IFORK_PTR(ip, XFS_DATA_FORK)->if_flags &
183 								XFS_IFEXTENTS);
184 		error = xfs_iomap_eof_align_last_fsb(ip, extsz, &last_fsb);
185 		if (error)
186 			goto out_unlock;
187 	} else {
188 		if (nmaps && (imap->br_startblock == HOLESTARTBLOCK))
189 			last_fsb = min(last_fsb, (xfs_fileoff_t)
190 					imap->br_blockcount +
191 					imap->br_startoff);
192 	}
193 	count_fsb = last_fsb - offset_fsb;
194 	ASSERT(count_fsb > 0);
195 	resaligned = xfs_aligned_fsb_count(offset_fsb, count_fsb, extsz);
196 
197 	if (unlikely(rt)) {
198 		resrtextents = qblocks = resaligned;
199 		resrtextents /= mp->m_sb.sb_rextsize;
200 		resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
201 		quota_flag = XFS_QMOPT_RES_RTBLKS;
202 	} else {
203 		resrtextents = 0;
204 		resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
205 		quota_flag = XFS_QMOPT_RES_REGBLKS;
206 	}
207 
208 	/*
209 	 * Drop the shared lock acquired by the caller, attach the dquot if
210 	 * necessary and move on to transaction setup.
211 	 */
212 	xfs_iunlock(ip, lockmode);
213 	error = xfs_qm_dqattach(ip);
214 	if (error)
215 		return error;
216 
217 	/*
218 	 * For DAX, we do not allocate unwritten extents, but instead we zero
219 	 * the block before we commit the transaction.  Ideally we'd like to do
220 	 * this outside the transaction context, but if we commit and then crash
221 	 * we may not have zeroed the blocks and this will be exposed on
222 	 * recovery of the allocation. Hence we must zero before commit.
223 	 *
224 	 * Further, if we are mapping unwritten extents here, we need to zero
225 	 * and convert them to written so that we don't need an unwritten extent
226 	 * callback for DAX. This also means that we need to be able to dip into
227 	 * the reserve block pool for bmbt block allocation if there is no space
228 	 * left but we need to do unwritten extent conversion.
229 	 */
230 	if (IS_DAX(VFS_I(ip))) {
231 		bmapi_flags = XFS_BMAPI_CONVERT | XFS_BMAPI_ZERO;
232 		if (imap->br_state == XFS_EXT_UNWRITTEN) {
233 			tflags |= XFS_TRANS_RESERVE;
234 			resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
235 		}
236 	}
237 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, resrtextents,
238 			tflags, &tp);
239 	if (error)
240 		return error;
241 
242 	lockmode = XFS_ILOCK_EXCL;
243 	xfs_ilock(ip, lockmode);
244 
245 	error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
246 	if (error)
247 		goto out_trans_cancel;
248 
249 	xfs_trans_ijoin(tp, ip, 0);
250 
251 	/*
252 	 * From this point onwards we overwrite the imap pointer that the
253 	 * caller gave to us.
254 	 */
255 	nimaps = 1;
256 	error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
257 				bmapi_flags, resblks, imap, &nimaps);
258 	if (error)
259 		goto out_res_cancel;
260 
261 	/*
262 	 * Complete the transaction
263 	 */
264 	error = xfs_trans_commit(tp);
265 	if (error)
266 		goto out_unlock;
267 
268 	/*
269 	 * Copy any maps to caller's array and return any error.
270 	 */
271 	if (nimaps == 0) {
272 		error = -ENOSPC;
273 		goto out_unlock;
274 	}
275 
276 	if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
277 		error = xfs_alert_fsblock_zero(ip, imap);
278 
279 out_unlock:
280 	xfs_iunlock(ip, lockmode);
281 	return error;
282 
283 out_res_cancel:
284 	xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
285 out_trans_cancel:
286 	xfs_trans_cancel(tp);
287 	goto out_unlock;
288 }
289 
290 STATIC bool
xfs_quota_need_throttle(struct xfs_inode * ip,int type,xfs_fsblock_t alloc_blocks)291 xfs_quota_need_throttle(
292 	struct xfs_inode *ip,
293 	int type,
294 	xfs_fsblock_t alloc_blocks)
295 {
296 	struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
297 
298 	if (!dq || !xfs_this_quota_on(ip->i_mount, type))
299 		return false;
300 
301 	/* no hi watermark, no throttle */
302 	if (!dq->q_prealloc_hi_wmark)
303 		return false;
304 
305 	/* under the lo watermark, no throttle */
306 	if (dq->q_res_bcount + alloc_blocks < dq->q_prealloc_lo_wmark)
307 		return false;
308 
309 	return true;
310 }
311 
312 STATIC void
xfs_quota_calc_throttle(struct xfs_inode * ip,int type,xfs_fsblock_t * qblocks,int * qshift,int64_t * qfreesp)313 xfs_quota_calc_throttle(
314 	struct xfs_inode *ip,
315 	int type,
316 	xfs_fsblock_t *qblocks,
317 	int *qshift,
318 	int64_t	*qfreesp)
319 {
320 	int64_t freesp;
321 	int shift = 0;
322 	struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
323 
324 	/* no dq, or over hi wmark, squash the prealloc completely */
325 	if (!dq || dq->q_res_bcount >= dq->q_prealloc_hi_wmark) {
326 		*qblocks = 0;
327 		*qfreesp = 0;
328 		return;
329 	}
330 
331 	freesp = dq->q_prealloc_hi_wmark - dq->q_res_bcount;
332 	if (freesp < dq->q_low_space[XFS_QLOWSP_5_PCNT]) {
333 		shift = 2;
334 		if (freesp < dq->q_low_space[XFS_QLOWSP_3_PCNT])
335 			shift += 2;
336 		if (freesp < dq->q_low_space[XFS_QLOWSP_1_PCNT])
337 			shift += 2;
338 	}
339 
340 	if (freesp < *qfreesp)
341 		*qfreesp = freesp;
342 
343 	/* only overwrite the throttle values if we are more aggressive */
344 	if ((freesp >> shift) < (*qblocks >> *qshift)) {
345 		*qblocks = freesp;
346 		*qshift = shift;
347 	}
348 }
349 
350 /*
351  * If we are doing a write at the end of the file and there are no allocations
352  * past this one, then extend the allocation out to the file system's write
353  * iosize.
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 full, the smaller the maximum prealocation.
359  *
360  * As an exception we don't do any preallocation at all if the file is smaller
361  * than the minimum preallocation and we are using the default dynamic
362  * preallocation scheme, as it is likely this is the only write to the file that
363  * is going to be done.
364  *
365  * We clean up any extra space left over when the file is closed in
366  * xfs_inactive().
367  */
368 STATIC xfs_fsblock_t
xfs_iomap_prealloc_size(struct xfs_inode * ip,loff_t offset,loff_t count,struct xfs_iext_cursor * icur)369 xfs_iomap_prealloc_size(
370 	struct xfs_inode	*ip,
371 	loff_t			offset,
372 	loff_t			count,
373 	struct xfs_iext_cursor	*icur)
374 {
375 	struct xfs_mount	*mp = ip->i_mount;
376 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
377 	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
378 	struct xfs_bmbt_irec	prev;
379 	int			shift = 0;
380 	int64_t			freesp;
381 	xfs_fsblock_t		qblocks;
382 	int			qshift = 0;
383 	xfs_fsblock_t		alloc_blocks = 0;
384 
385 	if (offset + count <= XFS_ISIZE(ip))
386 		return 0;
387 
388 	if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) &&
389 	    (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_writeio_blocks)))
390 		return 0;
391 
392 	/*
393 	 * If an explicit allocsize is set, the file is small, or we
394 	 * are writing behind a hole, then use the minimum prealloc:
395 	 */
396 	if ((mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) ||
397 	    XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign) ||
398 	    !xfs_iext_peek_prev_extent(ifp, icur, &prev) ||
399 	    prev.br_startoff + prev.br_blockcount < offset_fsb)
400 		return mp->m_writeio_blocks;
401 
402 	/*
403 	 * Determine the initial size of the preallocation. We are beyond the
404 	 * current EOF here, but we need to take into account whether this is
405 	 * a sparse write or an extending write when determining the
406 	 * preallocation size.  Hence we need to look up the extent that ends
407 	 * at the current write offset and use the result to determine the
408 	 * preallocation size.
409 	 *
410 	 * If the extent is a hole, then preallocation is essentially disabled.
411 	 * Otherwise we take the size of the preceding data extent as the basis
412 	 * for the preallocation size. If the size of the extent is greater than
413 	 * half the maximum extent length, then use the current offset as the
414 	 * basis. This ensures that for large files the preallocation size
415 	 * always extends to MAXEXTLEN rather than falling short due to things
416 	 * like stripe unit/width alignment of real extents.
417 	 */
418 	if (prev.br_blockcount <= (MAXEXTLEN >> 1))
419 		alloc_blocks = prev.br_blockcount << 1;
420 	else
421 		alloc_blocks = XFS_B_TO_FSB(mp, offset);
422 	if (!alloc_blocks)
423 		goto check_writeio;
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 prealloc
430 	 * size, we round up first, apply appropriate throttling, round down and
431 	 * 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_DQ_USER, alloc_blocks))
454 		xfs_quota_calc_throttle(ip, XFS_DQ_USER, &qblocks, &qshift,
455 					&freesp);
456 	if (xfs_quota_need_throttle(ip, XFS_DQ_GROUP, alloc_blocks))
457 		xfs_quota_calc_throttle(ip, XFS_DQ_GROUP, &qblocks, &qshift,
458 					&freesp);
459 	if (xfs_quota_need_throttle(ip, XFS_DQ_PROJ, alloc_blocks))
460 		xfs_quota_calc_throttle(ip, XFS_DQ_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 check_writeio:
493 	if (alloc_blocks < mp->m_writeio_blocks)
494 		alloc_blocks = mp->m_writeio_blocks;
495 	trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift,
496 				      mp->m_writeio_blocks);
497 	return alloc_blocks;
498 }
499 
500 static int
xfs_file_iomap_begin_delay(struct inode * inode,loff_t offset,loff_t count,struct iomap * iomap)501 xfs_file_iomap_begin_delay(
502 	struct inode		*inode,
503 	loff_t			offset,
504 	loff_t			count,
505 	struct iomap		*iomap)
506 {
507 	struct xfs_inode	*ip = XFS_I(inode);
508 	struct xfs_mount	*mp = ip->i_mount;
509 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
510 	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
511 	xfs_fileoff_t		maxbytes_fsb =
512 		XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
513 	xfs_fileoff_t		end_fsb;
514 	int			error = 0, eof = 0;
515 	struct xfs_bmbt_irec	got;
516 	struct xfs_iext_cursor	icur;
517 	xfs_fsblock_t		prealloc_blocks = 0;
518 
519 	ASSERT(!XFS_IS_REALTIME_INODE(ip));
520 	ASSERT(!xfs_get_extsz_hint(ip));
521 
522 	xfs_ilock(ip, XFS_ILOCK_EXCL);
523 
524 	if (unlikely(XFS_TEST_ERROR(
525 	    (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
526 	     XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
527 	     mp, XFS_ERRTAG_BMAPIFORMAT))) {
528 		XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
529 		error = -EFSCORRUPTED;
530 		goto out_unlock;
531 	}
532 
533 	XFS_STATS_INC(mp, xs_blk_mapw);
534 
535 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
536 		error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
537 		if (error)
538 			goto out_unlock;
539 	}
540 
541 	eof = !xfs_iext_lookup_extent(ip, ifp, offset_fsb, &icur, &got);
542 	if (!eof && got.br_startoff <= offset_fsb) {
543 		if (xfs_is_reflink_inode(ip)) {
544 			bool		shared;
545 
546 			end_fsb = min(XFS_B_TO_FSB(mp, offset + count),
547 					maxbytes_fsb);
548 			xfs_trim_extent(&got, offset_fsb, end_fsb - offset_fsb);
549 			error = xfs_reflink_reserve_cow(ip, &got, &shared);
550 			if (error)
551 				goto out_unlock;
552 		}
553 
554 		trace_xfs_iomap_found(ip, offset, count, 0, &got);
555 		goto done;
556 	}
557 
558 	error = xfs_qm_dqattach_locked(ip, false);
559 	if (error)
560 		goto out_unlock;
561 
562 	/*
563 	 * We cap the maximum length we map here to MAX_WRITEBACK_PAGES pages
564 	 * to keep the chunks of work done where somewhat symmetric with the
565 	 * work writeback does. This is a completely arbitrary number pulled
566 	 * out of thin air as a best guess for initial testing.
567 	 *
568 	 * Note that the values needs to be less than 32-bits wide until
569 	 * the lower level functions are updated.
570 	 */
571 	count = min_t(loff_t, count, 1024 * PAGE_SIZE);
572 	end_fsb = min(XFS_B_TO_FSB(mp, offset + count), maxbytes_fsb);
573 
574 	if (eof) {
575 		prealloc_blocks = xfs_iomap_prealloc_size(ip, offset, count,
576 				&icur);
577 		if (prealloc_blocks) {
578 			xfs_extlen_t	align;
579 			xfs_off_t	end_offset;
580 			xfs_fileoff_t	p_end_fsb;
581 
582 			end_offset = XFS_WRITEIO_ALIGN(mp, offset + count - 1);
583 			p_end_fsb = XFS_B_TO_FSBT(mp, end_offset) +
584 					prealloc_blocks;
585 
586 			align = xfs_eof_alignment(ip, 0);
587 			if (align)
588 				p_end_fsb = roundup_64(p_end_fsb, align);
589 
590 			p_end_fsb = min(p_end_fsb, maxbytes_fsb);
591 			ASSERT(p_end_fsb > offset_fsb);
592 			prealloc_blocks = p_end_fsb - end_fsb;
593 		}
594 	}
595 
596 retry:
597 	error = xfs_bmapi_reserve_delalloc(ip, XFS_DATA_FORK, offset_fsb,
598 			end_fsb - offset_fsb, prealloc_blocks, &got, &icur,
599 			eof);
600 	switch (error) {
601 	case 0:
602 		break;
603 	case -ENOSPC:
604 	case -EDQUOT:
605 		/* retry without any preallocation */
606 		trace_xfs_delalloc_enospc(ip, offset, count);
607 		if (prealloc_blocks) {
608 			prealloc_blocks = 0;
609 			goto retry;
610 		}
611 		/*FALLTHRU*/
612 	default:
613 		goto out_unlock;
614 	}
615 
616 	/*
617 	 * Flag newly allocated delalloc blocks with IOMAP_F_NEW so we punch
618 	 * them out if the write happens to fail.
619 	 */
620 	iomap->flags |= IOMAP_F_NEW;
621 	trace_xfs_iomap_alloc(ip, offset, count, 0, &got);
622 done:
623 	if (isnullstartblock(got.br_startblock))
624 		got.br_startblock = DELAYSTARTBLOCK;
625 
626 	if (!got.br_startblock) {
627 		error = xfs_alert_fsblock_zero(ip, &got);
628 		if (error)
629 			goto out_unlock;
630 	}
631 
632 	xfs_bmbt_to_iomap(ip, iomap, &got);
633 
634 out_unlock:
635 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
636 	return error;
637 }
638 
639 /*
640  * Pass in a delayed allocate extent, convert it to real extents;
641  * return to the caller the extent we create which maps on top of
642  * the originating callers request.
643  *
644  * Called without a lock on the inode.
645  *
646  * We no longer bother to look at the incoming map - all we have to
647  * guarantee is that whatever we allocate fills the required range.
648  */
649 int
xfs_iomap_write_allocate(xfs_inode_t * ip,int whichfork,xfs_off_t offset,xfs_bmbt_irec_t * imap,unsigned int * cow_seq)650 xfs_iomap_write_allocate(
651 	xfs_inode_t	*ip,
652 	int		whichfork,
653 	xfs_off_t	offset,
654 	xfs_bmbt_irec_t *imap,
655 	unsigned int	*cow_seq)
656 {
657 	xfs_mount_t	*mp = ip->i_mount;
658 	struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
659 	xfs_fileoff_t	offset_fsb, last_block;
660 	xfs_fileoff_t	end_fsb, map_start_fsb;
661 	xfs_filblks_t	count_fsb;
662 	xfs_trans_t	*tp;
663 	int		nimaps;
664 	int		error = 0;
665 	int		flags = XFS_BMAPI_DELALLOC;
666 	int		nres;
667 
668 	if (whichfork == XFS_COW_FORK)
669 		flags |= XFS_BMAPI_COWFORK | XFS_BMAPI_PREALLOC;
670 
671 	/*
672 	 * Make sure that the dquots are there.
673 	 */
674 	error = xfs_qm_dqattach(ip);
675 	if (error)
676 		return error;
677 
678 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
679 	count_fsb = imap->br_blockcount;
680 	map_start_fsb = imap->br_startoff;
681 
682 	XFS_STATS_ADD(mp, xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
683 
684 	while (count_fsb != 0) {
685 		/*
686 		 * Set up a transaction with which to allocate the
687 		 * backing store for the file.  Do allocations in a
688 		 * loop until we get some space in the range we are
689 		 * interested in.  The other space that might be allocated
690 		 * is in the delayed allocation extent on which we sit
691 		 * but before our buffer starts.
692 		 */
693 		nimaps = 0;
694 		while (nimaps == 0) {
695 			nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
696 			/*
697 			 * We have already reserved space for the extent and any
698 			 * indirect blocks when creating the delalloc extent,
699 			 * there is no need to reserve space in this transaction
700 			 * again.
701 			 */
702 			error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0,
703 					0, XFS_TRANS_RESERVE, &tp);
704 			if (error)
705 				return error;
706 
707 			xfs_ilock(ip, XFS_ILOCK_EXCL);
708 			xfs_trans_ijoin(tp, ip, 0);
709 
710 			/*
711 			 * it is possible that the extents have changed since
712 			 * we did the read call as we dropped the ilock for a
713 			 * while. We have to be careful about truncates or hole
714 			 * punchs here - we are not allowed to allocate
715 			 * non-delalloc blocks here.
716 			 *
717 			 * The only protection against truncation is the pages
718 			 * for the range we are being asked to convert are
719 			 * locked and hence a truncate will block on them
720 			 * first.
721 			 *
722 			 * As a result, if we go beyond the range we really
723 			 * need and hit an delalloc extent boundary followed by
724 			 * a hole while we have excess blocks in the map, we
725 			 * will fill the hole incorrectly and overrun the
726 			 * transaction reservation.
727 			 *
728 			 * Using a single map prevents this as we are forced to
729 			 * check each map we look for overlap with the desired
730 			 * range and abort as soon as we find it. Also, given
731 			 * that we only return a single map, having one beyond
732 			 * what we can return is probably a bit silly.
733 			 *
734 			 * We also need to check that we don't go beyond EOF;
735 			 * this is a truncate optimisation as a truncate sets
736 			 * the new file size before block on the pages we
737 			 * currently have locked under writeback. Because they
738 			 * are about to be tossed, we don't need to write them
739 			 * back....
740 			 */
741 			nimaps = 1;
742 			end_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
743 			error = xfs_bmap_last_offset(ip, &last_block,
744 							XFS_DATA_FORK);
745 			if (error)
746 				goto trans_cancel;
747 
748 			last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
749 			if ((map_start_fsb + count_fsb) > last_block) {
750 				count_fsb = last_block - map_start_fsb;
751 				if (count_fsb == 0) {
752 					error = -EAGAIN;
753 					goto trans_cancel;
754 				}
755 			}
756 
757 			/*
758 			 * From this point onwards we overwrite the imap
759 			 * pointer that the caller gave to us.
760 			 */
761 			error = xfs_bmapi_write(tp, ip, map_start_fsb,
762 						count_fsb, flags, nres, imap,
763 						&nimaps);
764 			if (error)
765 				goto trans_cancel;
766 
767 			error = xfs_trans_commit(tp);
768 			if (error)
769 				goto error0;
770 
771 			if (whichfork == XFS_COW_FORK)
772 				*cow_seq = READ_ONCE(ifp->if_seq);
773 			xfs_iunlock(ip, XFS_ILOCK_EXCL);
774 		}
775 
776 		/*
777 		 * See if we were able to allocate an extent that
778 		 * covers at least part of the callers request
779 		 */
780 		if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
781 			return xfs_alert_fsblock_zero(ip, imap);
782 
783 		if ((offset_fsb >= imap->br_startoff) &&
784 		    (offset_fsb < (imap->br_startoff +
785 				   imap->br_blockcount))) {
786 			XFS_STATS_INC(mp, xs_xstrat_quick);
787 			return 0;
788 		}
789 
790 		/*
791 		 * So far we have not mapped the requested part of the
792 		 * file, just surrounding data, try again.
793 		 */
794 		count_fsb -= imap->br_blockcount;
795 		map_start_fsb = imap->br_startoff + imap->br_blockcount;
796 	}
797 
798 trans_cancel:
799 	xfs_trans_cancel(tp);
800 error0:
801 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
802 	return error;
803 }
804 
805 int
xfs_iomap_write_unwritten(xfs_inode_t * ip,xfs_off_t offset,xfs_off_t count,bool update_isize)806 xfs_iomap_write_unwritten(
807 	xfs_inode_t	*ip,
808 	xfs_off_t	offset,
809 	xfs_off_t	count,
810 	bool		update_isize)
811 {
812 	xfs_mount_t	*mp = ip->i_mount;
813 	xfs_fileoff_t	offset_fsb;
814 	xfs_filblks_t	count_fsb;
815 	xfs_filblks_t	numblks_fsb;
816 	int		nimaps;
817 	xfs_trans_t	*tp;
818 	xfs_bmbt_irec_t imap;
819 	struct inode	*inode = VFS_I(ip);
820 	xfs_fsize_t	i_size;
821 	uint		resblks;
822 	int		error;
823 
824 	trace_xfs_unwritten_convert(ip, offset, count);
825 
826 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
827 	count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
828 	count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
829 
830 	/*
831 	 * Reserve enough blocks in this transaction for two complete extent
832 	 * btree splits.  We may be converting the middle part of an unwritten
833 	 * extent and in this case we will insert two new extents in the btree
834 	 * each of which could cause a full split.
835 	 *
836 	 * This reservation amount will be used in the first call to
837 	 * xfs_bmbt_split() to select an AG with enough space to satisfy the
838 	 * rest of the operation.
839 	 */
840 	resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
841 
842 	do {
843 		/*
844 		 * Set up a transaction to convert the range of extents
845 		 * from unwritten to real. Do allocations in a loop until
846 		 * we have covered the range passed in.
847 		 *
848 		 * Note that we can't risk to recursing back into the filesystem
849 		 * here as we might be asked to write out the same inode that we
850 		 * complete here and might deadlock on the iolock.
851 		 */
852 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0,
853 				XFS_TRANS_RESERVE | XFS_TRANS_NOFS, &tp);
854 		if (error)
855 			return error;
856 
857 		xfs_ilock(ip, XFS_ILOCK_EXCL);
858 		xfs_trans_ijoin(tp, ip, 0);
859 
860 		/*
861 		 * Modify the unwritten extent state of the buffer.
862 		 */
863 		nimaps = 1;
864 		error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
865 					XFS_BMAPI_CONVERT, resblks, &imap,
866 					&nimaps);
867 		if (error)
868 			goto error_on_bmapi_transaction;
869 
870 		/*
871 		 * Log the updated inode size as we go.  We have to be careful
872 		 * to only log it up to the actual write offset if it is
873 		 * halfway into a block.
874 		 */
875 		i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb);
876 		if (i_size > offset + count)
877 			i_size = offset + count;
878 		if (update_isize && i_size > i_size_read(inode))
879 			i_size_write(inode, i_size);
880 		i_size = xfs_new_eof(ip, i_size);
881 		if (i_size) {
882 			ip->i_d.di_size = i_size;
883 			xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
884 		}
885 
886 		error = xfs_trans_commit(tp);
887 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
888 		if (error)
889 			return error;
890 
891 		if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
892 			return xfs_alert_fsblock_zero(ip, &imap);
893 
894 		if ((numblks_fsb = imap.br_blockcount) == 0) {
895 			/*
896 			 * The numblks_fsb value should always get
897 			 * smaller, otherwise the loop is stuck.
898 			 */
899 			ASSERT(imap.br_blockcount);
900 			break;
901 		}
902 		offset_fsb += numblks_fsb;
903 		count_fsb -= numblks_fsb;
904 	} while (count_fsb > 0);
905 
906 	return 0;
907 
908 error_on_bmapi_transaction:
909 	xfs_trans_cancel(tp);
910 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
911 	return error;
912 }
913 
914 static inline bool
imap_needs_alloc(struct inode * inode,struct xfs_bmbt_irec * imap,int nimaps)915 imap_needs_alloc(
916 	struct inode		*inode,
917 	struct xfs_bmbt_irec	*imap,
918 	int			nimaps)
919 {
920 	return !nimaps ||
921 		imap->br_startblock == HOLESTARTBLOCK ||
922 		imap->br_startblock == DELAYSTARTBLOCK ||
923 		(IS_DAX(inode) && imap->br_state == XFS_EXT_UNWRITTEN);
924 }
925 
926 static inline bool
needs_cow_for_zeroing(struct xfs_bmbt_irec * imap,int nimaps)927 needs_cow_for_zeroing(
928 	struct xfs_bmbt_irec	*imap,
929 	int			nimaps)
930 {
931 	return nimaps &&
932 		imap->br_startblock != HOLESTARTBLOCK &&
933 		imap->br_state != XFS_EXT_UNWRITTEN;
934 }
935 
936 static int
xfs_ilock_for_iomap(struct xfs_inode * ip,unsigned flags,unsigned * lockmode)937 xfs_ilock_for_iomap(
938 	struct xfs_inode	*ip,
939 	unsigned		flags,
940 	unsigned		*lockmode)
941 {
942 	unsigned		mode = XFS_ILOCK_SHARED;
943 	bool			is_write = flags & (IOMAP_WRITE | IOMAP_ZERO);
944 
945 	/*
946 	 * COW writes may allocate delalloc space or convert unwritten COW
947 	 * extents, so we need to make sure to take the lock exclusively here.
948 	 */
949 	if (xfs_is_reflink_inode(ip) && is_write) {
950 		/*
951 		 * FIXME: It could still overwrite on unshared extents and not
952 		 * need allocation.
953 		 */
954 		if (flags & IOMAP_NOWAIT)
955 			return -EAGAIN;
956 		mode = XFS_ILOCK_EXCL;
957 	}
958 
959 	/*
960 	 * Extents not yet cached requires exclusive access, don't block.  This
961 	 * is an opencoded xfs_ilock_data_map_shared() call but with
962 	 * non-blocking behaviour.
963 	 */
964 	if (!(ip->i_df.if_flags & XFS_IFEXTENTS)) {
965 		if (flags & IOMAP_NOWAIT)
966 			return -EAGAIN;
967 		mode = XFS_ILOCK_EXCL;
968 	}
969 
970 relock:
971 	if (flags & IOMAP_NOWAIT) {
972 		if (!xfs_ilock_nowait(ip, mode))
973 			return -EAGAIN;
974 	} else {
975 		xfs_ilock(ip, mode);
976 	}
977 
978 	/*
979 	 * The reflink iflag could have changed since the earlier unlocked
980 	 * check, so if we got ILOCK_SHARED for a write and but we're now a
981 	 * reflink inode we have to switch to ILOCK_EXCL and relock.
982 	 */
983 	if (mode == XFS_ILOCK_SHARED && is_write && xfs_is_reflink_inode(ip)) {
984 		xfs_iunlock(ip, mode);
985 		mode = XFS_ILOCK_EXCL;
986 		goto relock;
987 	}
988 
989 	*lockmode = mode;
990 	return 0;
991 }
992 
993 static int
xfs_file_iomap_begin(struct inode * inode,loff_t offset,loff_t length,unsigned flags,struct iomap * iomap)994 xfs_file_iomap_begin(
995 	struct inode		*inode,
996 	loff_t			offset,
997 	loff_t			length,
998 	unsigned		flags,
999 	struct iomap		*iomap)
1000 {
1001 	struct xfs_inode	*ip = XFS_I(inode);
1002 	struct xfs_mount	*mp = ip->i_mount;
1003 	struct xfs_bmbt_irec	imap;
1004 	xfs_fileoff_t		offset_fsb, end_fsb;
1005 	int			nimaps = 1, error = 0;
1006 	bool			shared = false, trimmed = false;
1007 	unsigned		lockmode;
1008 
1009 	if (XFS_FORCED_SHUTDOWN(mp))
1010 		return -EIO;
1011 
1012 	if (((flags & (IOMAP_WRITE | IOMAP_DIRECT)) == IOMAP_WRITE) &&
1013 			!IS_DAX(inode) && !xfs_get_extsz_hint(ip)) {
1014 		/* Reserve delalloc blocks for regular writeback. */
1015 		return xfs_file_iomap_begin_delay(inode, offset, length, iomap);
1016 	}
1017 
1018 	/*
1019 	 * Lock the inode in the manner required for the specified operation and
1020 	 * check for as many conditions that would result in blocking as
1021 	 * possible. This removes most of the non-blocking checks from the
1022 	 * mapping code below.
1023 	 */
1024 	error = xfs_ilock_for_iomap(ip, flags, &lockmode);
1025 	if (error)
1026 		return error;
1027 
1028 	ASSERT(offset <= mp->m_super->s_maxbytes);
1029 	if (offset > mp->m_super->s_maxbytes - length)
1030 		length = mp->m_super->s_maxbytes - offset;
1031 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
1032 	end_fsb = XFS_B_TO_FSB(mp, offset + length);
1033 
1034 	error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1035 			       &nimaps, 0);
1036 	if (error)
1037 		goto out_unlock;
1038 
1039 	if (flags & IOMAP_REPORT) {
1040 		/* Trim the mapping to the nearest shared extent boundary. */
1041 		error = xfs_reflink_trim_around_shared(ip, &imap, &shared,
1042 				&trimmed);
1043 		if (error)
1044 			goto out_unlock;
1045 	}
1046 
1047 	/* Non-modifying mapping requested, so we are done */
1048 	if (!(flags & (IOMAP_WRITE | IOMAP_ZERO)))
1049 		goto out_found;
1050 
1051 	/*
1052 	 * Break shared extents if necessary. Checks for non-blocking IO have
1053 	 * been done up front, so we don't need to do them here.
1054 	 */
1055 	if (xfs_is_reflink_inode(ip)) {
1056 		/* if zeroing doesn't need COW allocation, then we are done. */
1057 		if ((flags & IOMAP_ZERO) &&
1058 		    !needs_cow_for_zeroing(&imap, nimaps))
1059 			goto out_found;
1060 
1061 		if (flags & IOMAP_DIRECT) {
1062 			/* may drop and re-acquire the ilock */
1063 			error = xfs_reflink_allocate_cow(ip, &imap, &shared,
1064 					&lockmode);
1065 			if (error)
1066 				goto out_unlock;
1067 		} else {
1068 			error = xfs_reflink_reserve_cow(ip, &imap, &shared);
1069 			if (error)
1070 				goto out_unlock;
1071 		}
1072 
1073 		end_fsb = imap.br_startoff + imap.br_blockcount;
1074 		length = XFS_FSB_TO_B(mp, end_fsb) - offset;
1075 	}
1076 
1077 	/* Don't need to allocate over holes when doing zeroing operations. */
1078 	if (flags & IOMAP_ZERO)
1079 		goto out_found;
1080 
1081 	if (!imap_needs_alloc(inode, &imap, nimaps))
1082 		goto out_found;
1083 
1084 	/* If nowait is set bail since we are going to make allocations. */
1085 	if (flags & IOMAP_NOWAIT) {
1086 		error = -EAGAIN;
1087 		goto out_unlock;
1088 	}
1089 
1090 	/*
1091 	 * We cap the maximum length we map to a sane size  to keep the chunks
1092 	 * of work done where somewhat symmetric with the work writeback does.
1093 	 * This is a completely arbitrary number pulled out of thin air as a
1094 	 * best guess for initial testing.
1095 	 *
1096 	 * Note that the values needs to be less than 32-bits wide until the
1097 	 * lower level functions are updated.
1098 	 */
1099 	length = min_t(loff_t, length, 1024 * PAGE_SIZE);
1100 
1101 	/*
1102 	 * xfs_iomap_write_direct() expects the shared lock. It is unlocked on
1103 	 * return.
1104 	 */
1105 	if (lockmode == XFS_ILOCK_EXCL)
1106 		xfs_ilock_demote(ip, lockmode);
1107 	error = xfs_iomap_write_direct(ip, offset, length, &imap,
1108 			nimaps);
1109 	if (error)
1110 		return error;
1111 
1112 	iomap->flags |= IOMAP_F_NEW;
1113 	trace_xfs_iomap_alloc(ip, offset, length, 0, &imap);
1114 
1115 out_finish:
1116 	if (xfs_ipincount(ip) && (ip->i_itemp->ili_fsync_fields
1117 				& ~XFS_ILOG_TIMESTAMP))
1118 		iomap->flags |= IOMAP_F_DIRTY;
1119 
1120 	xfs_bmbt_to_iomap(ip, iomap, &imap);
1121 
1122 	if (shared)
1123 		iomap->flags |= IOMAP_F_SHARED;
1124 	return 0;
1125 
1126 out_found:
1127 	ASSERT(nimaps);
1128 	xfs_iunlock(ip, lockmode);
1129 	trace_xfs_iomap_found(ip, offset, length, 0, &imap);
1130 	goto out_finish;
1131 
1132 out_unlock:
1133 	xfs_iunlock(ip, lockmode);
1134 	return error;
1135 }
1136 
1137 static int
xfs_file_iomap_end_delalloc(struct xfs_inode * ip,loff_t offset,loff_t length,ssize_t written,struct iomap * iomap)1138 xfs_file_iomap_end_delalloc(
1139 	struct xfs_inode	*ip,
1140 	loff_t			offset,
1141 	loff_t			length,
1142 	ssize_t			written,
1143 	struct iomap		*iomap)
1144 {
1145 	struct xfs_mount	*mp = ip->i_mount;
1146 	xfs_fileoff_t		start_fsb;
1147 	xfs_fileoff_t		end_fsb;
1148 	int			error = 0;
1149 
1150 	/*
1151 	 * Behave as if the write failed if drop writes is enabled. Set the NEW
1152 	 * flag to force delalloc cleanup.
1153 	 */
1154 	if (XFS_TEST_ERROR(false, mp, XFS_ERRTAG_DROP_WRITES)) {
1155 		iomap->flags |= IOMAP_F_NEW;
1156 		written = 0;
1157 	}
1158 
1159 	/*
1160 	 * start_fsb refers to the first unused block after a short write. If
1161 	 * nothing was written, round offset down to point at the first block in
1162 	 * the range.
1163 	 */
1164 	if (unlikely(!written))
1165 		start_fsb = XFS_B_TO_FSBT(mp, offset);
1166 	else
1167 		start_fsb = XFS_B_TO_FSB(mp, offset + written);
1168 	end_fsb = XFS_B_TO_FSB(mp, offset + length);
1169 
1170 	/*
1171 	 * Trim delalloc blocks if they were allocated by this write and we
1172 	 * didn't manage to write the whole range.
1173 	 *
1174 	 * We don't need to care about racing delalloc as we hold i_mutex
1175 	 * across the reserve/allocate/unreserve calls. If there are delalloc
1176 	 * blocks in the range, they are ours.
1177 	 */
1178 	if ((iomap->flags & IOMAP_F_NEW) && start_fsb < end_fsb) {
1179 		truncate_pagecache_range(VFS_I(ip), XFS_FSB_TO_B(mp, start_fsb),
1180 					 XFS_FSB_TO_B(mp, end_fsb) - 1);
1181 
1182 		error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
1183 					       end_fsb - start_fsb);
1184 		if (error && !XFS_FORCED_SHUTDOWN(mp)) {
1185 			xfs_alert(mp, "%s: unable to clean up ino %lld",
1186 				__func__, ip->i_ino);
1187 			return error;
1188 		}
1189 	}
1190 
1191 	return 0;
1192 }
1193 
1194 static int
xfs_file_iomap_end(struct inode * inode,loff_t offset,loff_t length,ssize_t written,unsigned flags,struct iomap * iomap)1195 xfs_file_iomap_end(
1196 	struct inode		*inode,
1197 	loff_t			offset,
1198 	loff_t			length,
1199 	ssize_t			written,
1200 	unsigned		flags,
1201 	struct iomap		*iomap)
1202 {
1203 	if ((flags & IOMAP_WRITE) && iomap->type == IOMAP_DELALLOC)
1204 		return xfs_file_iomap_end_delalloc(XFS_I(inode), offset,
1205 				length, written, iomap);
1206 	return 0;
1207 }
1208 
1209 const struct iomap_ops xfs_iomap_ops = {
1210 	.iomap_begin		= xfs_file_iomap_begin,
1211 	.iomap_end		= xfs_file_iomap_end,
1212 };
1213 
1214 static int
xfs_xattr_iomap_begin(struct inode * inode,loff_t offset,loff_t length,unsigned flags,struct iomap * iomap)1215 xfs_xattr_iomap_begin(
1216 	struct inode		*inode,
1217 	loff_t			offset,
1218 	loff_t			length,
1219 	unsigned		flags,
1220 	struct iomap		*iomap)
1221 {
1222 	struct xfs_inode	*ip = XFS_I(inode);
1223 	struct xfs_mount	*mp = ip->i_mount;
1224 	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
1225 	xfs_fileoff_t		end_fsb = XFS_B_TO_FSB(mp, offset + length);
1226 	struct xfs_bmbt_irec	imap;
1227 	int			nimaps = 1, error = 0;
1228 	unsigned		lockmode;
1229 
1230 	if (XFS_FORCED_SHUTDOWN(mp))
1231 		return -EIO;
1232 
1233 	lockmode = xfs_ilock_attr_map_shared(ip);
1234 
1235 	/* if there are no attribute fork or extents, return ENOENT */
1236 	if (!XFS_IFORK_Q(ip) || !ip->i_d.di_anextents) {
1237 		error = -ENOENT;
1238 		goto out_unlock;
1239 	}
1240 
1241 	ASSERT(ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL);
1242 	error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1243 			       &nimaps, XFS_BMAPI_ATTRFORK);
1244 out_unlock:
1245 	xfs_iunlock(ip, lockmode);
1246 
1247 	if (!error) {
1248 		ASSERT(nimaps);
1249 		xfs_bmbt_to_iomap(ip, iomap, &imap);
1250 	}
1251 
1252 	return error;
1253 }
1254 
1255 const struct iomap_ops xfs_xattr_iomap_ops = {
1256 	.iomap_begin		= xfs_xattr_iomap_begin,
1257 };
1258