xref: /openbmc/linux/fs/xfs/xfs_extent_busy.c (revision feac8c8b)
1 /*
2  * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
3  * Copyright (c) 2010 David Chinner.
4  * Copyright (c) 2011 Christoph Hellwig.
5  * All Rights Reserved.
6  *
7  * This program is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU General Public License as
9  * published by the Free Software Foundation.
10  *
11  * This program is distributed in the hope that it would be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14  * GNU General Public License for more details.
15  *
16  * You should have received a copy of the GNU General Public License
17  * along with this program; if not, write the Free Software Foundation,
18  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
19  */
20 #include "xfs.h"
21 #include "xfs_fs.h"
22 #include "xfs_format.h"
23 #include "xfs_log_format.h"
24 #include "xfs_shared.h"
25 #include "xfs_trans_resv.h"
26 #include "xfs_sb.h"
27 #include "xfs_mount.h"
28 #include "xfs_alloc.h"
29 #include "xfs_extent_busy.h"
30 #include "xfs_trace.h"
31 #include "xfs_trans.h"
32 #include "xfs_log.h"
33 
34 void
35 xfs_extent_busy_insert(
36 	struct xfs_trans	*tp,
37 	xfs_agnumber_t		agno,
38 	xfs_agblock_t		bno,
39 	xfs_extlen_t		len,
40 	unsigned int		flags)
41 {
42 	struct xfs_extent_busy	*new;
43 	struct xfs_extent_busy	*busyp;
44 	struct xfs_perag	*pag;
45 	struct rb_node		**rbp;
46 	struct rb_node		*parent = NULL;
47 
48 	new = kmem_zalloc(sizeof(struct xfs_extent_busy), KM_SLEEP);
49 	new->agno = agno;
50 	new->bno = bno;
51 	new->length = len;
52 	INIT_LIST_HEAD(&new->list);
53 	new->flags = flags;
54 
55 	/* trace before insert to be able to see failed inserts */
56 	trace_xfs_extent_busy(tp->t_mountp, agno, bno, len);
57 
58 	pag = xfs_perag_get(tp->t_mountp, new->agno);
59 	spin_lock(&pag->pagb_lock);
60 	rbp = &pag->pagb_tree.rb_node;
61 	while (*rbp) {
62 		parent = *rbp;
63 		busyp = rb_entry(parent, struct xfs_extent_busy, rb_node);
64 
65 		if (new->bno < busyp->bno) {
66 			rbp = &(*rbp)->rb_left;
67 			ASSERT(new->bno + new->length <= busyp->bno);
68 		} else if (new->bno > busyp->bno) {
69 			rbp = &(*rbp)->rb_right;
70 			ASSERT(bno >= busyp->bno + busyp->length);
71 		} else {
72 			ASSERT(0);
73 		}
74 	}
75 
76 	rb_link_node(&new->rb_node, parent, rbp);
77 	rb_insert_color(&new->rb_node, &pag->pagb_tree);
78 
79 	list_add(&new->list, &tp->t_busy);
80 	spin_unlock(&pag->pagb_lock);
81 	xfs_perag_put(pag);
82 }
83 
84 /*
85  * Search for a busy extent within the range of the extent we are about to
86  * allocate.  You need to be holding the busy extent tree lock when calling
87  * xfs_extent_busy_search(). This function returns 0 for no overlapping busy
88  * extent, -1 for an overlapping but not exact busy extent, and 1 for an exact
89  * match. This is done so that a non-zero return indicates an overlap that
90  * will require a synchronous transaction, but it can still be
91  * used to distinguish between a partial or exact match.
92  */
93 int
94 xfs_extent_busy_search(
95 	struct xfs_mount	*mp,
96 	xfs_agnumber_t		agno,
97 	xfs_agblock_t		bno,
98 	xfs_extlen_t		len)
99 {
100 	struct xfs_perag	*pag;
101 	struct rb_node		*rbp;
102 	struct xfs_extent_busy	*busyp;
103 	int			match = 0;
104 
105 	pag = xfs_perag_get(mp, agno);
106 	spin_lock(&pag->pagb_lock);
107 
108 	rbp = pag->pagb_tree.rb_node;
109 
110 	/* find closest start bno overlap */
111 	while (rbp) {
112 		busyp = rb_entry(rbp, struct xfs_extent_busy, rb_node);
113 		if (bno < busyp->bno) {
114 			/* may overlap, but exact start block is lower */
115 			if (bno + len > busyp->bno)
116 				match = -1;
117 			rbp = rbp->rb_left;
118 		} else if (bno > busyp->bno) {
119 			/* may overlap, but exact start block is higher */
120 			if (bno < busyp->bno + busyp->length)
121 				match = -1;
122 			rbp = rbp->rb_right;
123 		} else {
124 			/* bno matches busyp, length determines exact match */
125 			match = (busyp->length == len) ? 1 : -1;
126 			break;
127 		}
128 	}
129 	spin_unlock(&pag->pagb_lock);
130 	xfs_perag_put(pag);
131 	return match;
132 }
133 
134 /*
135  * The found free extent [fbno, fend] overlaps part or all of the given busy
136  * extent.  If the overlap covers the beginning, the end, or all of the busy
137  * extent, the overlapping portion can be made unbusy and used for the
138  * allocation.  We can't split a busy extent because we can't modify a
139  * transaction/CIL context busy list, but we can update an entry's block
140  * number or length.
141  *
142  * Returns true if the extent can safely be reused, or false if the search
143  * needs to be restarted.
144  */
145 STATIC bool
146 xfs_extent_busy_update_extent(
147 	struct xfs_mount	*mp,
148 	struct xfs_perag	*pag,
149 	struct xfs_extent_busy	*busyp,
150 	xfs_agblock_t		fbno,
151 	xfs_extlen_t		flen,
152 	bool			userdata) __releases(&pag->pagb_lock)
153 					  __acquires(&pag->pagb_lock)
154 {
155 	xfs_agblock_t		fend = fbno + flen;
156 	xfs_agblock_t		bbno = busyp->bno;
157 	xfs_agblock_t		bend = bbno + busyp->length;
158 
159 	/*
160 	 * This extent is currently being discarded.  Give the thread
161 	 * performing the discard a chance to mark the extent unbusy
162 	 * and retry.
163 	 */
164 	if (busyp->flags & XFS_EXTENT_BUSY_DISCARDED) {
165 		spin_unlock(&pag->pagb_lock);
166 		delay(1);
167 		spin_lock(&pag->pagb_lock);
168 		return false;
169 	}
170 
171 	/*
172 	 * If there is a busy extent overlapping a user allocation, we have
173 	 * no choice but to force the log and retry the search.
174 	 *
175 	 * Fortunately this does not happen during normal operation, but
176 	 * only if the filesystem is very low on space and has to dip into
177 	 * the AGFL for normal allocations.
178 	 */
179 	if (userdata)
180 		goto out_force_log;
181 
182 	if (bbno < fbno && bend > fend) {
183 		/*
184 		 * Case 1:
185 		 *    bbno           bend
186 		 *    +BBBBBBBBBBBBBBBBB+
187 		 *        +---------+
188 		 *        fbno   fend
189 		 */
190 
191 		/*
192 		 * We would have to split the busy extent to be able to track
193 		 * it correct, which we cannot do because we would have to
194 		 * modify the list of busy extents attached to the transaction
195 		 * or CIL context, which is immutable.
196 		 *
197 		 * Force out the log to clear the busy extent and retry the
198 		 * search.
199 		 */
200 		goto out_force_log;
201 	} else if (bbno >= fbno && bend <= fend) {
202 		/*
203 		 * Case 2:
204 		 *    bbno           bend
205 		 *    +BBBBBBBBBBBBBBBBB+
206 		 *    +-----------------+
207 		 *    fbno           fend
208 		 *
209 		 * Case 3:
210 		 *    bbno           bend
211 		 *    +BBBBBBBBBBBBBBBBB+
212 		 *    +--------------------------+
213 		 *    fbno                    fend
214 		 *
215 		 * Case 4:
216 		 *             bbno           bend
217 		 *             +BBBBBBBBBBBBBBBBB+
218 		 *    +--------------------------+
219 		 *    fbno                    fend
220 		 *
221 		 * Case 5:
222 		 *             bbno           bend
223 		 *             +BBBBBBBBBBBBBBBBB+
224 		 *    +-----------------------------------+
225 		 *    fbno                             fend
226 		 *
227 		 */
228 
229 		/*
230 		 * The busy extent is fully covered by the extent we are
231 		 * allocating, and can simply be removed from the rbtree.
232 		 * However we cannot remove it from the immutable list
233 		 * tracking busy extents in the transaction or CIL context,
234 		 * so set the length to zero to mark it invalid.
235 		 *
236 		 * We also need to restart the busy extent search from the
237 		 * tree root, because erasing the node can rearrange the
238 		 * tree topology.
239 		 */
240 		rb_erase(&busyp->rb_node, &pag->pagb_tree);
241 		busyp->length = 0;
242 		return false;
243 	} else if (fend < bend) {
244 		/*
245 		 * Case 6:
246 		 *              bbno           bend
247 		 *             +BBBBBBBBBBBBBBBBB+
248 		 *             +---------+
249 		 *             fbno   fend
250 		 *
251 		 * Case 7:
252 		 *             bbno           bend
253 		 *             +BBBBBBBBBBBBBBBBB+
254 		 *    +------------------+
255 		 *    fbno            fend
256 		 *
257 		 */
258 		busyp->bno = fend;
259 	} else if (bbno < fbno) {
260 		/*
261 		 * Case 8:
262 		 *    bbno           bend
263 		 *    +BBBBBBBBBBBBBBBBB+
264 		 *        +-------------+
265 		 *        fbno       fend
266 		 *
267 		 * Case 9:
268 		 *    bbno           bend
269 		 *    +BBBBBBBBBBBBBBBBB+
270 		 *        +----------------------+
271 		 *        fbno                fend
272 		 */
273 		busyp->length = fbno - busyp->bno;
274 	} else {
275 		ASSERT(0);
276 	}
277 
278 	trace_xfs_extent_busy_reuse(mp, pag->pag_agno, fbno, flen);
279 	return true;
280 
281 out_force_log:
282 	spin_unlock(&pag->pagb_lock);
283 	xfs_log_force(mp, XFS_LOG_SYNC);
284 	trace_xfs_extent_busy_force(mp, pag->pag_agno, fbno, flen);
285 	spin_lock(&pag->pagb_lock);
286 	return false;
287 }
288 
289 
290 /*
291  * For a given extent [fbno, flen], make sure we can reuse it safely.
292  */
293 void
294 xfs_extent_busy_reuse(
295 	struct xfs_mount	*mp,
296 	xfs_agnumber_t		agno,
297 	xfs_agblock_t		fbno,
298 	xfs_extlen_t		flen,
299 	bool			userdata)
300 {
301 	struct xfs_perag	*pag;
302 	struct rb_node		*rbp;
303 
304 	ASSERT(flen > 0);
305 
306 	pag = xfs_perag_get(mp, agno);
307 	spin_lock(&pag->pagb_lock);
308 restart:
309 	rbp = pag->pagb_tree.rb_node;
310 	while (rbp) {
311 		struct xfs_extent_busy *busyp =
312 			rb_entry(rbp, struct xfs_extent_busy, rb_node);
313 		xfs_agblock_t	bbno = busyp->bno;
314 		xfs_agblock_t	bend = bbno + busyp->length;
315 
316 		if (fbno + flen <= bbno) {
317 			rbp = rbp->rb_left;
318 			continue;
319 		} else if (fbno >= bend) {
320 			rbp = rbp->rb_right;
321 			continue;
322 		}
323 
324 		if (!xfs_extent_busy_update_extent(mp, pag, busyp, fbno, flen,
325 						  userdata))
326 			goto restart;
327 	}
328 	spin_unlock(&pag->pagb_lock);
329 	xfs_perag_put(pag);
330 }
331 
332 /*
333  * For a given extent [fbno, flen], search the busy extent list to find a
334  * subset of the extent that is not busy.  If *rlen is smaller than
335  * args->minlen no suitable extent could be found, and the higher level
336  * code needs to force out the log and retry the allocation.
337  *
338  * Return the current busy generation for the AG if the extent is busy. This
339  * value can be used to wait for at least one of the currently busy extents
340  * to be cleared. Note that the busy list is not guaranteed to be empty after
341  * the gen is woken. The state of a specific extent must always be confirmed
342  * with another call to xfs_extent_busy_trim() before it can be used.
343  */
344 bool
345 xfs_extent_busy_trim(
346 	struct xfs_alloc_arg	*args,
347 	xfs_agblock_t		*bno,
348 	xfs_extlen_t		*len,
349 	unsigned		*busy_gen)
350 {
351 	xfs_agblock_t		fbno;
352 	xfs_extlen_t		flen;
353 	struct rb_node		*rbp;
354 	bool			ret = false;
355 
356 	ASSERT(*len > 0);
357 
358 	spin_lock(&args->pag->pagb_lock);
359 restart:
360 	fbno = *bno;
361 	flen = *len;
362 	rbp = args->pag->pagb_tree.rb_node;
363 	while (rbp && flen >= args->minlen) {
364 		struct xfs_extent_busy *busyp =
365 			rb_entry(rbp, struct xfs_extent_busy, rb_node);
366 		xfs_agblock_t	fend = fbno + flen;
367 		xfs_agblock_t	bbno = busyp->bno;
368 		xfs_agblock_t	bend = bbno + busyp->length;
369 
370 		if (fend <= bbno) {
371 			rbp = rbp->rb_left;
372 			continue;
373 		} else if (fbno >= bend) {
374 			rbp = rbp->rb_right;
375 			continue;
376 		}
377 
378 		/*
379 		 * If this is a metadata allocation, try to reuse the busy
380 		 * extent instead of trimming the allocation.
381 		 */
382 		if (!xfs_alloc_is_userdata(args->datatype) &&
383 		    !(busyp->flags & XFS_EXTENT_BUSY_DISCARDED)) {
384 			if (!xfs_extent_busy_update_extent(args->mp, args->pag,
385 							  busyp, fbno, flen,
386 							  false))
387 				goto restart;
388 			continue;
389 		}
390 
391 		if (bbno <= fbno) {
392 			/* start overlap */
393 
394 			/*
395 			 * Case 1:
396 			 *    bbno           bend
397 			 *    +BBBBBBBBBBBBBBBBB+
398 			 *        +---------+
399 			 *        fbno   fend
400 			 *
401 			 * Case 2:
402 			 *    bbno           bend
403 			 *    +BBBBBBBBBBBBBBBBB+
404 			 *    +-------------+
405 			 *    fbno       fend
406 			 *
407 			 * Case 3:
408 			 *    bbno           bend
409 			 *    +BBBBBBBBBBBBBBBBB+
410 			 *        +-------------+
411 			 *        fbno       fend
412 			 *
413 			 * Case 4:
414 			 *    bbno           bend
415 			 *    +BBBBBBBBBBBBBBBBB+
416 			 *    +-----------------+
417 			 *    fbno           fend
418 			 *
419 			 * No unbusy region in extent, return failure.
420 			 */
421 			if (fend <= bend)
422 				goto fail;
423 
424 			/*
425 			 * Case 5:
426 			 *    bbno           bend
427 			 *    +BBBBBBBBBBBBBBBBB+
428 			 *        +----------------------+
429 			 *        fbno                fend
430 			 *
431 			 * Case 6:
432 			 *    bbno           bend
433 			 *    +BBBBBBBBBBBBBBBBB+
434 			 *    +--------------------------+
435 			 *    fbno                    fend
436 			 *
437 			 * Needs to be trimmed to:
438 			 *                       +-------+
439 			 *                       fbno fend
440 			 */
441 			fbno = bend;
442 		} else if (bend >= fend) {
443 			/* end overlap */
444 
445 			/*
446 			 * Case 7:
447 			 *             bbno           bend
448 			 *             +BBBBBBBBBBBBBBBBB+
449 			 *    +------------------+
450 			 *    fbno            fend
451 			 *
452 			 * Case 8:
453 			 *             bbno           bend
454 			 *             +BBBBBBBBBBBBBBBBB+
455 			 *    +--------------------------+
456 			 *    fbno                    fend
457 			 *
458 			 * Needs to be trimmed to:
459 			 *    +-------+
460 			 *    fbno fend
461 			 */
462 			fend = bbno;
463 		} else {
464 			/* middle overlap */
465 
466 			/*
467 			 * Case 9:
468 			 *             bbno           bend
469 			 *             +BBBBBBBBBBBBBBBBB+
470 			 *    +-----------------------------------+
471 			 *    fbno                             fend
472 			 *
473 			 * Can be trimmed to:
474 			 *    +-------+        OR         +-------+
475 			 *    fbno fend                   fbno fend
476 			 *
477 			 * Backward allocation leads to significant
478 			 * fragmentation of directories, which degrades
479 			 * directory performance, therefore we always want to
480 			 * choose the option that produces forward allocation
481 			 * patterns.
482 			 * Preferring the lower bno extent will make the next
483 			 * request use "fend" as the start of the next
484 			 * allocation;  if the segment is no longer busy at
485 			 * that point, we'll get a contiguous allocation, but
486 			 * even if it is still busy, we will get a forward
487 			 * allocation.
488 			 * We try to avoid choosing the segment at "bend",
489 			 * because that can lead to the next allocation
490 			 * taking the segment at "fbno", which would be a
491 			 * backward allocation.  We only use the segment at
492 			 * "fbno" if it is much larger than the current
493 			 * requested size, because in that case there's a
494 			 * good chance subsequent allocations will be
495 			 * contiguous.
496 			 */
497 			if (bbno - fbno >= args->maxlen) {
498 				/* left candidate fits perfect */
499 				fend = bbno;
500 			} else if (fend - bend >= args->maxlen * 4) {
501 				/* right candidate has enough free space */
502 				fbno = bend;
503 			} else if (bbno - fbno >= args->minlen) {
504 				/* left candidate fits minimum requirement */
505 				fend = bbno;
506 			} else {
507 				goto fail;
508 			}
509 		}
510 
511 		flen = fend - fbno;
512 	}
513 out:
514 
515 	if (fbno != *bno || flen != *len) {
516 		trace_xfs_extent_busy_trim(args->mp, args->agno, *bno, *len,
517 					  fbno, flen);
518 		*bno = fbno;
519 		*len = flen;
520 		*busy_gen = args->pag->pagb_gen;
521 		ret = true;
522 	}
523 	spin_unlock(&args->pag->pagb_lock);
524 	return ret;
525 fail:
526 	/*
527 	 * Return a zero extent length as failure indications.  All callers
528 	 * re-check if the trimmed extent satisfies the minlen requirement.
529 	 */
530 	flen = 0;
531 	goto out;
532 }
533 
534 STATIC void
535 xfs_extent_busy_clear_one(
536 	struct xfs_mount	*mp,
537 	struct xfs_perag	*pag,
538 	struct xfs_extent_busy	*busyp)
539 {
540 	if (busyp->length) {
541 		trace_xfs_extent_busy_clear(mp, busyp->agno, busyp->bno,
542 						busyp->length);
543 		rb_erase(&busyp->rb_node, &pag->pagb_tree);
544 	}
545 
546 	list_del_init(&busyp->list);
547 	kmem_free(busyp);
548 }
549 
550 static void
551 xfs_extent_busy_put_pag(
552 	struct xfs_perag	*pag,
553 	bool			wakeup)
554 		__releases(pag->pagb_lock)
555 {
556 	if (wakeup) {
557 		pag->pagb_gen++;
558 		wake_up_all(&pag->pagb_wait);
559 	}
560 
561 	spin_unlock(&pag->pagb_lock);
562 	xfs_perag_put(pag);
563 }
564 
565 /*
566  * Remove all extents on the passed in list from the busy extents tree.
567  * If do_discard is set skip extents that need to be discarded, and mark
568  * these as undergoing a discard operation instead.
569  */
570 void
571 xfs_extent_busy_clear(
572 	struct xfs_mount	*mp,
573 	struct list_head	*list,
574 	bool			do_discard)
575 {
576 	struct xfs_extent_busy	*busyp, *n;
577 	struct xfs_perag	*pag = NULL;
578 	xfs_agnumber_t		agno = NULLAGNUMBER;
579 	bool			wakeup = false;
580 
581 	list_for_each_entry_safe(busyp, n, list, list) {
582 		if (busyp->agno != agno) {
583 			if (pag)
584 				xfs_extent_busy_put_pag(pag, wakeup);
585 			agno = busyp->agno;
586 			pag = xfs_perag_get(mp, agno);
587 			spin_lock(&pag->pagb_lock);
588 			wakeup = false;
589 		}
590 
591 		if (do_discard && busyp->length &&
592 		    !(busyp->flags & XFS_EXTENT_BUSY_SKIP_DISCARD)) {
593 			busyp->flags = XFS_EXTENT_BUSY_DISCARDED;
594 		} else {
595 			xfs_extent_busy_clear_one(mp, pag, busyp);
596 			wakeup = true;
597 		}
598 	}
599 
600 	if (pag)
601 		xfs_extent_busy_put_pag(pag, wakeup);
602 }
603 
604 /*
605  * Flush out all busy extents for this AG.
606  */
607 void
608 xfs_extent_busy_flush(
609 	struct xfs_mount	*mp,
610 	struct xfs_perag	*pag,
611 	unsigned		busy_gen)
612 {
613 	DEFINE_WAIT		(wait);
614 	int			log_flushed = 0, error;
615 
616 	trace_xfs_log_force(mp, 0, _THIS_IP_);
617 	error = _xfs_log_force(mp, XFS_LOG_SYNC, &log_flushed);
618 	if (error)
619 		return;
620 
621 	do {
622 		prepare_to_wait(&pag->pagb_wait, &wait, TASK_KILLABLE);
623 		if  (busy_gen != READ_ONCE(pag->pagb_gen))
624 			break;
625 		schedule();
626 	} while (1);
627 
628 	finish_wait(&pag->pagb_wait, &wait);
629 }
630 
631 void
632 xfs_extent_busy_wait_all(
633 	struct xfs_mount	*mp)
634 {
635 	DEFINE_WAIT		(wait);
636 	xfs_agnumber_t		agno;
637 
638 	for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
639 		struct xfs_perag *pag = xfs_perag_get(mp, agno);
640 
641 		do {
642 			prepare_to_wait(&pag->pagb_wait, &wait, TASK_KILLABLE);
643 			if  (RB_EMPTY_ROOT(&pag->pagb_tree))
644 				break;
645 			schedule();
646 		} while (1);
647 		finish_wait(&pag->pagb_wait, &wait);
648 
649 		xfs_perag_put(pag);
650 	}
651 }
652 
653 /*
654  * Callback for list_sort to sort busy extents by the AG they reside in.
655  */
656 int
657 xfs_extent_busy_ag_cmp(
658 	void			*priv,
659 	struct list_head	*l1,
660 	struct list_head	*l2)
661 {
662 	struct xfs_extent_busy	*b1 =
663 		container_of(l1, struct xfs_extent_busy, list);
664 	struct xfs_extent_busy	*b2 =
665 		container_of(l2, struct xfs_extent_busy, list);
666 	s32 diff;
667 
668 	diff = b1->agno - b2->agno;
669 	if (!diff)
670 		diff = b1->bno - b2->bno;
671 	return diff;
672 }
673