xref: /openbmc/linux/drivers/block/drbd/drbd_actlog.c (revision 82e6fdd6)
1 /*
2    drbd_actlog.c
3 
4    This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
5 
6    Copyright (C) 2003-2008, LINBIT Information Technologies GmbH.
7    Copyright (C) 2003-2008, Philipp Reisner <philipp.reisner@linbit.com>.
8    Copyright (C) 2003-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.
9 
10    drbd is free software; you can redistribute it and/or modify
11    it under the terms of the GNU General Public License as published by
12    the Free Software Foundation; either version 2, or (at your option)
13    any later version.
14 
15    drbd is distributed in the hope that it will be useful,
16    but WITHOUT ANY WARRANTY; without even the implied warranty of
17    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
18    GNU General Public License for more details.
19 
20    You should have received a copy of the GNU General Public License
21    along with drbd; see the file COPYING.  If not, write to
22    the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
23 
24  */
25 
26 #include <linux/slab.h>
27 #include <linux/crc32c.h>
28 #include <linux/drbd.h>
29 #include <linux/drbd_limits.h>
30 #include "drbd_int.h"
31 
32 
33 enum al_transaction_types {
34 	AL_TR_UPDATE = 0,
35 	AL_TR_INITIALIZED = 0xffff
36 };
37 /* all fields on disc in big endian */
38 struct __packed al_transaction_on_disk {
39 	/* don't we all like magic */
40 	__be32	magic;
41 
42 	/* to identify the most recent transaction block
43 	 * in the on disk ring buffer */
44 	__be32	tr_number;
45 
46 	/* checksum on the full 4k block, with this field set to 0. */
47 	__be32	crc32c;
48 
49 	/* type of transaction, special transaction types like:
50 	 * purge-all, set-all-idle, set-all-active, ... to-be-defined
51 	 * see also enum al_transaction_types */
52 	__be16	transaction_type;
53 
54 	/* we currently allow only a few thousand extents,
55 	 * so 16bit will be enough for the slot number. */
56 
57 	/* how many updates in this transaction */
58 	__be16	n_updates;
59 
60 	/* maximum slot number, "al-extents" in drbd.conf speak.
61 	 * Having this in each transaction should make reconfiguration
62 	 * of that parameter easier. */
63 	__be16	context_size;
64 
65 	/* slot number the context starts with */
66 	__be16	context_start_slot_nr;
67 
68 	/* Some reserved bytes.  Expected usage is a 64bit counter of
69 	 * sectors-written since device creation, and other data generation tag
70 	 * supporting usage */
71 	__be32	__reserved[4];
72 
73 	/* --- 36 byte used --- */
74 
75 	/* Reserve space for up to AL_UPDATES_PER_TRANSACTION changes
76 	 * in one transaction, then use the remaining byte in the 4k block for
77 	 * context information.  "Flexible" number of updates per transaction
78 	 * does not help, as we have to account for the case when all update
79 	 * slots are used anyways, so it would only complicate code without
80 	 * additional benefit.
81 	 */
82 	__be16	update_slot_nr[AL_UPDATES_PER_TRANSACTION];
83 
84 	/* but the extent number is 32bit, which at an extent size of 4 MiB
85 	 * allows to cover device sizes of up to 2**54 Byte (16 PiB) */
86 	__be32	update_extent_nr[AL_UPDATES_PER_TRANSACTION];
87 
88 	/* --- 420 bytes used (36 + 64*6) --- */
89 
90 	/* 4096 - 420 = 3676 = 919 * 4 */
91 	__be32	context[AL_CONTEXT_PER_TRANSACTION];
92 };
93 
94 void *drbd_md_get_buffer(struct drbd_device *device, const char *intent)
95 {
96 	int r;
97 
98 	wait_event(device->misc_wait,
99 		   (r = atomic_cmpxchg(&device->md_io.in_use, 0, 1)) == 0 ||
100 		   device->state.disk <= D_FAILED);
101 
102 	if (r)
103 		return NULL;
104 
105 	device->md_io.current_use = intent;
106 	device->md_io.start_jif = jiffies;
107 	device->md_io.submit_jif = device->md_io.start_jif - 1;
108 	return page_address(device->md_io.page);
109 }
110 
111 void drbd_md_put_buffer(struct drbd_device *device)
112 {
113 	if (atomic_dec_and_test(&device->md_io.in_use))
114 		wake_up(&device->misc_wait);
115 }
116 
117 void wait_until_done_or_force_detached(struct drbd_device *device, struct drbd_backing_dev *bdev,
118 				     unsigned int *done)
119 {
120 	long dt;
121 
122 	rcu_read_lock();
123 	dt = rcu_dereference(bdev->disk_conf)->disk_timeout;
124 	rcu_read_unlock();
125 	dt = dt * HZ / 10;
126 	if (dt == 0)
127 		dt = MAX_SCHEDULE_TIMEOUT;
128 
129 	dt = wait_event_timeout(device->misc_wait,
130 			*done || test_bit(FORCE_DETACH, &device->flags), dt);
131 	if (dt == 0) {
132 		drbd_err(device, "meta-data IO operation timed out\n");
133 		drbd_chk_io_error(device, 1, DRBD_FORCE_DETACH);
134 	}
135 }
136 
137 static int _drbd_md_sync_page_io(struct drbd_device *device,
138 				 struct drbd_backing_dev *bdev,
139 				 sector_t sector, int op)
140 {
141 	struct bio *bio;
142 	/* we do all our meta data IO in aligned 4k blocks. */
143 	const int size = 4096;
144 	int err, op_flags = 0;
145 
146 	device->md_io.done = 0;
147 	device->md_io.error = -ENODEV;
148 
149 	if ((op == REQ_OP_WRITE) && !test_bit(MD_NO_FUA, &device->flags))
150 		op_flags |= REQ_FUA | REQ_PREFLUSH;
151 	op_flags |= REQ_SYNC;
152 
153 	bio = bio_alloc_drbd(GFP_NOIO);
154 	bio_set_dev(bio, bdev->md_bdev);
155 	bio->bi_iter.bi_sector = sector;
156 	err = -EIO;
157 	if (bio_add_page(bio, device->md_io.page, size, 0) != size)
158 		goto out;
159 	bio->bi_private = device;
160 	bio->bi_end_io = drbd_md_endio;
161 	bio_set_op_attrs(bio, op, op_flags);
162 
163 	if (op != REQ_OP_WRITE && device->state.disk == D_DISKLESS && device->ldev == NULL)
164 		/* special case, drbd_md_read() during drbd_adm_attach(): no get_ldev */
165 		;
166 	else if (!get_ldev_if_state(device, D_ATTACHING)) {
167 		/* Corresponding put_ldev in drbd_md_endio() */
168 		drbd_err(device, "ASSERT FAILED: get_ldev_if_state() == 1 in _drbd_md_sync_page_io()\n");
169 		err = -ENODEV;
170 		goto out;
171 	}
172 
173 	bio_get(bio); /* one bio_put() is in the completion handler */
174 	atomic_inc(&device->md_io.in_use); /* drbd_md_put_buffer() is in the completion handler */
175 	device->md_io.submit_jif = jiffies;
176 	if (drbd_insert_fault(device, (op == REQ_OP_WRITE) ? DRBD_FAULT_MD_WR : DRBD_FAULT_MD_RD))
177 		bio_io_error(bio);
178 	else
179 		submit_bio(bio);
180 	wait_until_done_or_force_detached(device, bdev, &device->md_io.done);
181 	if (!bio->bi_status)
182 		err = device->md_io.error;
183 
184  out:
185 	bio_put(bio);
186 	return err;
187 }
188 
189 int drbd_md_sync_page_io(struct drbd_device *device, struct drbd_backing_dev *bdev,
190 			 sector_t sector, int op)
191 {
192 	int err;
193 	D_ASSERT(device, atomic_read(&device->md_io.in_use) == 1);
194 
195 	BUG_ON(!bdev->md_bdev);
196 
197 	dynamic_drbd_dbg(device, "meta_data io: %s [%d]:%s(,%llus,%s) %pS\n",
198 	     current->comm, current->pid, __func__,
199 	     (unsigned long long)sector, (op == REQ_OP_WRITE) ? "WRITE" : "READ",
200 	     (void*)_RET_IP_ );
201 
202 	if (sector < drbd_md_first_sector(bdev) ||
203 	    sector + 7 > drbd_md_last_sector(bdev))
204 		drbd_alert(device, "%s [%d]:%s(,%llus,%s) out of range md access!\n",
205 		     current->comm, current->pid, __func__,
206 		     (unsigned long long)sector,
207 		     (op == REQ_OP_WRITE) ? "WRITE" : "READ");
208 
209 	err = _drbd_md_sync_page_io(device, bdev, sector, op);
210 	if (err) {
211 		drbd_err(device, "drbd_md_sync_page_io(,%llus,%s) failed with error %d\n",
212 		    (unsigned long long)sector,
213 		    (op == REQ_OP_WRITE) ? "WRITE" : "READ", err);
214 	}
215 	return err;
216 }
217 
218 static struct bm_extent *find_active_resync_extent(struct drbd_device *device, unsigned int enr)
219 {
220 	struct lc_element *tmp;
221 	tmp = lc_find(device->resync, enr/AL_EXT_PER_BM_SECT);
222 	if (unlikely(tmp != NULL)) {
223 		struct bm_extent  *bm_ext = lc_entry(tmp, struct bm_extent, lce);
224 		if (test_bit(BME_NO_WRITES, &bm_ext->flags))
225 			return bm_ext;
226 	}
227 	return NULL;
228 }
229 
230 static struct lc_element *_al_get(struct drbd_device *device, unsigned int enr, bool nonblock)
231 {
232 	struct lc_element *al_ext;
233 	struct bm_extent *bm_ext;
234 	int wake;
235 
236 	spin_lock_irq(&device->al_lock);
237 	bm_ext = find_active_resync_extent(device, enr);
238 	if (bm_ext) {
239 		wake = !test_and_set_bit(BME_PRIORITY, &bm_ext->flags);
240 		spin_unlock_irq(&device->al_lock);
241 		if (wake)
242 			wake_up(&device->al_wait);
243 		return NULL;
244 	}
245 	if (nonblock)
246 		al_ext = lc_try_get(device->act_log, enr);
247 	else
248 		al_ext = lc_get(device->act_log, enr);
249 	spin_unlock_irq(&device->al_lock);
250 	return al_ext;
251 }
252 
253 bool drbd_al_begin_io_fastpath(struct drbd_device *device, struct drbd_interval *i)
254 {
255 	/* for bios crossing activity log extent boundaries,
256 	 * we may need to activate two extents in one go */
257 	unsigned first = i->sector >> (AL_EXTENT_SHIFT-9);
258 	unsigned last = i->size == 0 ? first : (i->sector + (i->size >> 9) - 1) >> (AL_EXTENT_SHIFT-9);
259 
260 	D_ASSERT(device, first <= last);
261 	D_ASSERT(device, atomic_read(&device->local_cnt) > 0);
262 
263 	/* FIXME figure out a fast path for bios crossing AL extent boundaries */
264 	if (first != last)
265 		return false;
266 
267 	return _al_get(device, first, true);
268 }
269 
270 bool drbd_al_begin_io_prepare(struct drbd_device *device, struct drbd_interval *i)
271 {
272 	/* for bios crossing activity log extent boundaries,
273 	 * we may need to activate two extents in one go */
274 	unsigned first = i->sector >> (AL_EXTENT_SHIFT-9);
275 	unsigned last = i->size == 0 ? first : (i->sector + (i->size >> 9) - 1) >> (AL_EXTENT_SHIFT-9);
276 	unsigned enr;
277 	bool need_transaction = false;
278 
279 	D_ASSERT(device, first <= last);
280 	D_ASSERT(device, atomic_read(&device->local_cnt) > 0);
281 
282 	for (enr = first; enr <= last; enr++) {
283 		struct lc_element *al_ext;
284 		wait_event(device->al_wait,
285 				(al_ext = _al_get(device, enr, false)) != NULL);
286 		if (al_ext->lc_number != enr)
287 			need_transaction = true;
288 	}
289 	return need_transaction;
290 }
291 
292 #if (PAGE_SHIFT + 3) < (AL_EXTENT_SHIFT - BM_BLOCK_SHIFT)
293 /* Currently BM_BLOCK_SHIFT, BM_EXT_SHIFT and AL_EXTENT_SHIFT
294  * are still coupled, or assume too much about their relation.
295  * Code below will not work if this is violated.
296  * Will be cleaned up with some followup patch.
297  */
298 # error FIXME
299 #endif
300 
301 static unsigned int al_extent_to_bm_page(unsigned int al_enr)
302 {
303 	return al_enr >>
304 		/* bit to page */
305 		((PAGE_SHIFT + 3) -
306 		/* al extent number to bit */
307 		 (AL_EXTENT_SHIFT - BM_BLOCK_SHIFT));
308 }
309 
310 static sector_t al_tr_number_to_on_disk_sector(struct drbd_device *device)
311 {
312 	const unsigned int stripes = device->ldev->md.al_stripes;
313 	const unsigned int stripe_size_4kB = device->ldev->md.al_stripe_size_4k;
314 
315 	/* transaction number, modulo on-disk ring buffer wrap around */
316 	unsigned int t = device->al_tr_number % (device->ldev->md.al_size_4k);
317 
318 	/* ... to aligned 4k on disk block */
319 	t = ((t % stripes) * stripe_size_4kB) + t/stripes;
320 
321 	/* ... to 512 byte sector in activity log */
322 	t *= 8;
323 
324 	/* ... plus offset to the on disk position */
325 	return device->ldev->md.md_offset + device->ldev->md.al_offset + t;
326 }
327 
328 static int __al_write_transaction(struct drbd_device *device, struct al_transaction_on_disk *buffer)
329 {
330 	struct lc_element *e;
331 	sector_t sector;
332 	int i, mx;
333 	unsigned extent_nr;
334 	unsigned crc = 0;
335 	int err = 0;
336 
337 	memset(buffer, 0, sizeof(*buffer));
338 	buffer->magic = cpu_to_be32(DRBD_AL_MAGIC);
339 	buffer->tr_number = cpu_to_be32(device->al_tr_number);
340 
341 	i = 0;
342 
343 	drbd_bm_reset_al_hints(device);
344 
345 	/* Even though no one can start to change this list
346 	 * once we set the LC_LOCKED -- from drbd_al_begin_io(),
347 	 * lc_try_lock_for_transaction() --, someone may still
348 	 * be in the process of changing it. */
349 	spin_lock_irq(&device->al_lock);
350 	list_for_each_entry(e, &device->act_log->to_be_changed, list) {
351 		if (i == AL_UPDATES_PER_TRANSACTION) {
352 			i++;
353 			break;
354 		}
355 		buffer->update_slot_nr[i] = cpu_to_be16(e->lc_index);
356 		buffer->update_extent_nr[i] = cpu_to_be32(e->lc_new_number);
357 		if (e->lc_number != LC_FREE)
358 			drbd_bm_mark_for_writeout(device,
359 					al_extent_to_bm_page(e->lc_number));
360 		i++;
361 	}
362 	spin_unlock_irq(&device->al_lock);
363 	BUG_ON(i > AL_UPDATES_PER_TRANSACTION);
364 
365 	buffer->n_updates = cpu_to_be16(i);
366 	for ( ; i < AL_UPDATES_PER_TRANSACTION; i++) {
367 		buffer->update_slot_nr[i] = cpu_to_be16(-1);
368 		buffer->update_extent_nr[i] = cpu_to_be32(LC_FREE);
369 	}
370 
371 	buffer->context_size = cpu_to_be16(device->act_log->nr_elements);
372 	buffer->context_start_slot_nr = cpu_to_be16(device->al_tr_cycle);
373 
374 	mx = min_t(int, AL_CONTEXT_PER_TRANSACTION,
375 		   device->act_log->nr_elements - device->al_tr_cycle);
376 	for (i = 0; i < mx; i++) {
377 		unsigned idx = device->al_tr_cycle + i;
378 		extent_nr = lc_element_by_index(device->act_log, idx)->lc_number;
379 		buffer->context[i] = cpu_to_be32(extent_nr);
380 	}
381 	for (; i < AL_CONTEXT_PER_TRANSACTION; i++)
382 		buffer->context[i] = cpu_to_be32(LC_FREE);
383 
384 	device->al_tr_cycle += AL_CONTEXT_PER_TRANSACTION;
385 	if (device->al_tr_cycle >= device->act_log->nr_elements)
386 		device->al_tr_cycle = 0;
387 
388 	sector = al_tr_number_to_on_disk_sector(device);
389 
390 	crc = crc32c(0, buffer, 4096);
391 	buffer->crc32c = cpu_to_be32(crc);
392 
393 	if (drbd_bm_write_hinted(device))
394 		err = -EIO;
395 	else {
396 		bool write_al_updates;
397 		rcu_read_lock();
398 		write_al_updates = rcu_dereference(device->ldev->disk_conf)->al_updates;
399 		rcu_read_unlock();
400 		if (write_al_updates) {
401 			if (drbd_md_sync_page_io(device, device->ldev, sector, WRITE)) {
402 				err = -EIO;
403 				drbd_chk_io_error(device, 1, DRBD_META_IO_ERROR);
404 			} else {
405 				device->al_tr_number++;
406 				device->al_writ_cnt++;
407 			}
408 		}
409 	}
410 
411 	return err;
412 }
413 
414 static int al_write_transaction(struct drbd_device *device)
415 {
416 	struct al_transaction_on_disk *buffer;
417 	int err;
418 
419 	if (!get_ldev(device)) {
420 		drbd_err(device, "disk is %s, cannot start al transaction\n",
421 			drbd_disk_str(device->state.disk));
422 		return -EIO;
423 	}
424 
425 	/* The bitmap write may have failed, causing a state change. */
426 	if (device->state.disk < D_INCONSISTENT) {
427 		drbd_err(device,
428 			"disk is %s, cannot write al transaction\n",
429 			drbd_disk_str(device->state.disk));
430 		put_ldev(device);
431 		return -EIO;
432 	}
433 
434 	/* protects md_io_buffer, al_tr_cycle, ... */
435 	buffer = drbd_md_get_buffer(device, __func__);
436 	if (!buffer) {
437 		drbd_err(device, "disk failed while waiting for md_io buffer\n");
438 		put_ldev(device);
439 		return -ENODEV;
440 	}
441 
442 	err = __al_write_transaction(device, buffer);
443 
444 	drbd_md_put_buffer(device);
445 	put_ldev(device);
446 
447 	return err;
448 }
449 
450 
451 void drbd_al_begin_io_commit(struct drbd_device *device)
452 {
453 	bool locked = false;
454 
455 	/* Serialize multiple transactions.
456 	 * This uses test_and_set_bit, memory barrier is implicit.
457 	 */
458 	wait_event(device->al_wait,
459 			device->act_log->pending_changes == 0 ||
460 			(locked = lc_try_lock_for_transaction(device->act_log)));
461 
462 	if (locked) {
463 		/* Double check: it may have been committed by someone else,
464 		 * while we have been waiting for the lock. */
465 		if (device->act_log->pending_changes) {
466 			bool write_al_updates;
467 
468 			rcu_read_lock();
469 			write_al_updates = rcu_dereference(device->ldev->disk_conf)->al_updates;
470 			rcu_read_unlock();
471 
472 			if (write_al_updates)
473 				al_write_transaction(device);
474 			spin_lock_irq(&device->al_lock);
475 			/* FIXME
476 			if (err)
477 				we need an "lc_cancel" here;
478 			*/
479 			lc_committed(device->act_log);
480 			spin_unlock_irq(&device->al_lock);
481 		}
482 		lc_unlock(device->act_log);
483 		wake_up(&device->al_wait);
484 	}
485 }
486 
487 /*
488  * @delegate:   delegate activity log I/O to the worker thread
489  */
490 void drbd_al_begin_io(struct drbd_device *device, struct drbd_interval *i)
491 {
492 	if (drbd_al_begin_io_prepare(device, i))
493 		drbd_al_begin_io_commit(device);
494 }
495 
496 int drbd_al_begin_io_nonblock(struct drbd_device *device, struct drbd_interval *i)
497 {
498 	struct lru_cache *al = device->act_log;
499 	/* for bios crossing activity log extent boundaries,
500 	 * we may need to activate two extents in one go */
501 	unsigned first = i->sector >> (AL_EXTENT_SHIFT-9);
502 	unsigned last = i->size == 0 ? first : (i->sector + (i->size >> 9) - 1) >> (AL_EXTENT_SHIFT-9);
503 	unsigned nr_al_extents;
504 	unsigned available_update_slots;
505 	unsigned enr;
506 
507 	D_ASSERT(device, first <= last);
508 
509 	nr_al_extents = 1 + last - first; /* worst case: all touched extends are cold. */
510 	available_update_slots = min(al->nr_elements - al->used,
511 				al->max_pending_changes - al->pending_changes);
512 
513 	/* We want all necessary updates for a given request within the same transaction
514 	 * We could first check how many updates are *actually* needed,
515 	 * and use that instead of the worst-case nr_al_extents */
516 	if (available_update_slots < nr_al_extents) {
517 		/* Too many activity log extents are currently "hot".
518 		 *
519 		 * If we have accumulated pending changes already,
520 		 * we made progress.
521 		 *
522 		 * If we cannot get even a single pending change through,
523 		 * stop the fast path until we made some progress,
524 		 * or requests to "cold" extents could be starved. */
525 		if (!al->pending_changes)
526 			__set_bit(__LC_STARVING, &device->act_log->flags);
527 		return -ENOBUFS;
528 	}
529 
530 	/* Is resync active in this area? */
531 	for (enr = first; enr <= last; enr++) {
532 		struct lc_element *tmp;
533 		tmp = lc_find(device->resync, enr/AL_EXT_PER_BM_SECT);
534 		if (unlikely(tmp != NULL)) {
535 			struct bm_extent  *bm_ext = lc_entry(tmp, struct bm_extent, lce);
536 			if (test_bit(BME_NO_WRITES, &bm_ext->flags)) {
537 				if (!test_and_set_bit(BME_PRIORITY, &bm_ext->flags))
538 					return -EBUSY;
539 				return -EWOULDBLOCK;
540 			}
541 		}
542 	}
543 
544 	/* Checkout the refcounts.
545 	 * Given that we checked for available elements and update slots above,
546 	 * this has to be successful. */
547 	for (enr = first; enr <= last; enr++) {
548 		struct lc_element *al_ext;
549 		al_ext = lc_get_cumulative(device->act_log, enr);
550 		if (!al_ext)
551 			drbd_info(device, "LOGIC BUG for enr=%u\n", enr);
552 	}
553 	return 0;
554 }
555 
556 void drbd_al_complete_io(struct drbd_device *device, struct drbd_interval *i)
557 {
558 	/* for bios crossing activity log extent boundaries,
559 	 * we may need to activate two extents in one go */
560 	unsigned first = i->sector >> (AL_EXTENT_SHIFT-9);
561 	unsigned last = i->size == 0 ? first : (i->sector + (i->size >> 9) - 1) >> (AL_EXTENT_SHIFT-9);
562 	unsigned enr;
563 	struct lc_element *extent;
564 	unsigned long flags;
565 
566 	D_ASSERT(device, first <= last);
567 	spin_lock_irqsave(&device->al_lock, flags);
568 
569 	for (enr = first; enr <= last; enr++) {
570 		extent = lc_find(device->act_log, enr);
571 		if (!extent) {
572 			drbd_err(device, "al_complete_io() called on inactive extent %u\n", enr);
573 			continue;
574 		}
575 		lc_put(device->act_log, extent);
576 	}
577 	spin_unlock_irqrestore(&device->al_lock, flags);
578 	wake_up(&device->al_wait);
579 }
580 
581 static int _try_lc_del(struct drbd_device *device, struct lc_element *al_ext)
582 {
583 	int rv;
584 
585 	spin_lock_irq(&device->al_lock);
586 	rv = (al_ext->refcnt == 0);
587 	if (likely(rv))
588 		lc_del(device->act_log, al_ext);
589 	spin_unlock_irq(&device->al_lock);
590 
591 	return rv;
592 }
593 
594 /**
595  * drbd_al_shrink() - Removes all active extents form the activity log
596  * @device:	DRBD device.
597  *
598  * Removes all active extents form the activity log, waiting until
599  * the reference count of each entry dropped to 0 first, of course.
600  *
601  * You need to lock device->act_log with lc_try_lock() / lc_unlock()
602  */
603 void drbd_al_shrink(struct drbd_device *device)
604 {
605 	struct lc_element *al_ext;
606 	int i;
607 
608 	D_ASSERT(device, test_bit(__LC_LOCKED, &device->act_log->flags));
609 
610 	for (i = 0; i < device->act_log->nr_elements; i++) {
611 		al_ext = lc_element_by_index(device->act_log, i);
612 		if (al_ext->lc_number == LC_FREE)
613 			continue;
614 		wait_event(device->al_wait, _try_lc_del(device, al_ext));
615 	}
616 
617 	wake_up(&device->al_wait);
618 }
619 
620 int drbd_al_initialize(struct drbd_device *device, void *buffer)
621 {
622 	struct al_transaction_on_disk *al = buffer;
623 	struct drbd_md *md = &device->ldev->md;
624 	int al_size_4k = md->al_stripes * md->al_stripe_size_4k;
625 	int i;
626 
627 	__al_write_transaction(device, al);
628 	/* There may or may not have been a pending transaction. */
629 	spin_lock_irq(&device->al_lock);
630 	lc_committed(device->act_log);
631 	spin_unlock_irq(&device->al_lock);
632 
633 	/* The rest of the transactions will have an empty "updates" list, and
634 	 * are written out only to provide the context, and to initialize the
635 	 * on-disk ring buffer. */
636 	for (i = 1; i < al_size_4k; i++) {
637 		int err = __al_write_transaction(device, al);
638 		if (err)
639 			return err;
640 	}
641 	return 0;
642 }
643 
644 static const char *drbd_change_sync_fname[] = {
645 	[RECORD_RS_FAILED] = "drbd_rs_failed_io",
646 	[SET_IN_SYNC] = "drbd_set_in_sync",
647 	[SET_OUT_OF_SYNC] = "drbd_set_out_of_sync"
648 };
649 
650 /* ATTENTION. The AL's extents are 4MB each, while the extents in the
651  * resync LRU-cache are 16MB each.
652  * The caller of this function has to hold an get_ldev() reference.
653  *
654  * Adjusts the caching members ->rs_left (success) or ->rs_failed (!success),
655  * potentially pulling in (and recounting the corresponding bits)
656  * this resync extent into the resync extent lru cache.
657  *
658  * Returns whether all bits have been cleared for this resync extent,
659  * precisely: (rs_left <= rs_failed)
660  *
661  * TODO will be obsoleted once we have a caching lru of the on disk bitmap
662  */
663 static bool update_rs_extent(struct drbd_device *device,
664 		unsigned int enr, int count,
665 		enum update_sync_bits_mode mode)
666 {
667 	struct lc_element *e;
668 
669 	D_ASSERT(device, atomic_read(&device->local_cnt));
670 
671 	/* When setting out-of-sync bits,
672 	 * we don't need it cached (lc_find).
673 	 * But if it is present in the cache,
674 	 * we should update the cached bit count.
675 	 * Otherwise, that extent should be in the resync extent lru cache
676 	 * already -- or we want to pull it in if necessary -- (lc_get),
677 	 * then update and check rs_left and rs_failed. */
678 	if (mode == SET_OUT_OF_SYNC)
679 		e = lc_find(device->resync, enr);
680 	else
681 		e = lc_get(device->resync, enr);
682 	if (e) {
683 		struct bm_extent *ext = lc_entry(e, struct bm_extent, lce);
684 		if (ext->lce.lc_number == enr) {
685 			if (mode == SET_IN_SYNC)
686 				ext->rs_left -= count;
687 			else if (mode == SET_OUT_OF_SYNC)
688 				ext->rs_left += count;
689 			else
690 				ext->rs_failed += count;
691 			if (ext->rs_left < ext->rs_failed) {
692 				drbd_warn(device, "BAD! enr=%u rs_left=%d "
693 				    "rs_failed=%d count=%d cstate=%s\n",
694 				     ext->lce.lc_number, ext->rs_left,
695 				     ext->rs_failed, count,
696 				     drbd_conn_str(device->state.conn));
697 
698 				/* We don't expect to be able to clear more bits
699 				 * than have been set when we originally counted
700 				 * the set bits to cache that value in ext->rs_left.
701 				 * Whatever the reason (disconnect during resync,
702 				 * delayed local completion of an application write),
703 				 * try to fix it up by recounting here. */
704 				ext->rs_left = drbd_bm_e_weight(device, enr);
705 			}
706 		} else {
707 			/* Normally this element should be in the cache,
708 			 * since drbd_rs_begin_io() pulled it already in.
709 			 *
710 			 * But maybe an application write finished, and we set
711 			 * something outside the resync lru_cache in sync.
712 			 */
713 			int rs_left = drbd_bm_e_weight(device, enr);
714 			if (ext->flags != 0) {
715 				drbd_warn(device, "changing resync lce: %d[%u;%02lx]"
716 				     " -> %d[%u;00]\n",
717 				     ext->lce.lc_number, ext->rs_left,
718 				     ext->flags, enr, rs_left);
719 				ext->flags = 0;
720 			}
721 			if (ext->rs_failed) {
722 				drbd_warn(device, "Kicking resync_lru element enr=%u "
723 				     "out with rs_failed=%d\n",
724 				     ext->lce.lc_number, ext->rs_failed);
725 			}
726 			ext->rs_left = rs_left;
727 			ext->rs_failed = (mode == RECORD_RS_FAILED) ? count : 0;
728 			/* we don't keep a persistent log of the resync lru,
729 			 * we can commit any change right away. */
730 			lc_committed(device->resync);
731 		}
732 		if (mode != SET_OUT_OF_SYNC)
733 			lc_put(device->resync, &ext->lce);
734 		/* no race, we are within the al_lock! */
735 
736 		if (ext->rs_left <= ext->rs_failed) {
737 			ext->rs_failed = 0;
738 			return true;
739 		}
740 	} else if (mode != SET_OUT_OF_SYNC) {
741 		/* be quiet if lc_find() did not find it. */
742 		drbd_err(device, "lc_get() failed! locked=%d/%d flags=%lu\n",
743 		    device->resync_locked,
744 		    device->resync->nr_elements,
745 		    device->resync->flags);
746 	}
747 	return false;
748 }
749 
750 void drbd_advance_rs_marks(struct drbd_device *device, unsigned long still_to_go)
751 {
752 	unsigned long now = jiffies;
753 	unsigned long last = device->rs_mark_time[device->rs_last_mark];
754 	int next = (device->rs_last_mark + 1) % DRBD_SYNC_MARKS;
755 	if (time_after_eq(now, last + DRBD_SYNC_MARK_STEP)) {
756 		if (device->rs_mark_left[device->rs_last_mark] != still_to_go &&
757 		    device->state.conn != C_PAUSED_SYNC_T &&
758 		    device->state.conn != C_PAUSED_SYNC_S) {
759 			device->rs_mark_time[next] = now;
760 			device->rs_mark_left[next] = still_to_go;
761 			device->rs_last_mark = next;
762 		}
763 	}
764 }
765 
766 /* It is called lazy update, so don't do write-out too often. */
767 static bool lazy_bitmap_update_due(struct drbd_device *device)
768 {
769 	return time_after(jiffies, device->rs_last_bcast + 2*HZ);
770 }
771 
772 static void maybe_schedule_on_disk_bitmap_update(struct drbd_device *device, bool rs_done)
773 {
774 	if (rs_done) {
775 		struct drbd_connection *connection = first_peer_device(device)->connection;
776 		if (connection->agreed_pro_version <= 95 ||
777 		    is_sync_target_state(device->state.conn))
778 			set_bit(RS_DONE, &device->flags);
779 			/* and also set RS_PROGRESS below */
780 
781 		/* Else: rather wait for explicit notification via receive_state,
782 		 * to avoid uuids-rotated-too-fast causing full resync
783 		 * in next handshake, in case the replication link breaks
784 		 * at the most unfortunate time... */
785 	} else if (!lazy_bitmap_update_due(device))
786 		return;
787 
788 	drbd_device_post_work(device, RS_PROGRESS);
789 }
790 
791 static int update_sync_bits(struct drbd_device *device,
792 		unsigned long sbnr, unsigned long ebnr,
793 		enum update_sync_bits_mode mode)
794 {
795 	/*
796 	 * We keep a count of set bits per resync-extent in the ->rs_left
797 	 * caching member, so we need to loop and work within the resync extent
798 	 * alignment. Typically this loop will execute exactly once.
799 	 */
800 	unsigned long flags;
801 	unsigned long count = 0;
802 	unsigned int cleared = 0;
803 	while (sbnr <= ebnr) {
804 		/* set temporary boundary bit number to last bit number within
805 		 * the resync extent of the current start bit number,
806 		 * but cap at provided end bit number */
807 		unsigned long tbnr = min(ebnr, sbnr | BM_BLOCKS_PER_BM_EXT_MASK);
808 		unsigned long c;
809 
810 		if (mode == RECORD_RS_FAILED)
811 			/* Only called from drbd_rs_failed_io(), bits
812 			 * supposedly still set.  Recount, maybe some
813 			 * of the bits have been successfully cleared
814 			 * by application IO meanwhile.
815 			 */
816 			c = drbd_bm_count_bits(device, sbnr, tbnr);
817 		else if (mode == SET_IN_SYNC)
818 			c = drbd_bm_clear_bits(device, sbnr, tbnr);
819 		else /* if (mode == SET_OUT_OF_SYNC) */
820 			c = drbd_bm_set_bits(device, sbnr, tbnr);
821 
822 		if (c) {
823 			spin_lock_irqsave(&device->al_lock, flags);
824 			cleared += update_rs_extent(device, BM_BIT_TO_EXT(sbnr), c, mode);
825 			spin_unlock_irqrestore(&device->al_lock, flags);
826 			count += c;
827 		}
828 		sbnr = tbnr + 1;
829 	}
830 	if (count) {
831 		if (mode == SET_IN_SYNC) {
832 			unsigned long still_to_go = drbd_bm_total_weight(device);
833 			bool rs_is_done = (still_to_go <= device->rs_failed);
834 			drbd_advance_rs_marks(device, still_to_go);
835 			if (cleared || rs_is_done)
836 				maybe_schedule_on_disk_bitmap_update(device, rs_is_done);
837 		} else if (mode == RECORD_RS_FAILED)
838 			device->rs_failed += count;
839 		wake_up(&device->al_wait);
840 	}
841 	return count;
842 }
843 
844 static bool plausible_request_size(int size)
845 {
846 	return size > 0
847 		&& size <= DRBD_MAX_BATCH_BIO_SIZE
848 		&& IS_ALIGNED(size, 512);
849 }
850 
851 /* clear the bit corresponding to the piece of storage in question:
852  * size byte of data starting from sector.  Only clear a bits of the affected
853  * one ore more _aligned_ BM_BLOCK_SIZE blocks.
854  *
855  * called by worker on C_SYNC_TARGET and receiver on SyncSource.
856  *
857  */
858 int __drbd_change_sync(struct drbd_device *device, sector_t sector, int size,
859 		enum update_sync_bits_mode mode)
860 {
861 	/* Is called from worker and receiver context _only_ */
862 	unsigned long sbnr, ebnr, lbnr;
863 	unsigned long count = 0;
864 	sector_t esector, nr_sectors;
865 
866 	/* This would be an empty REQ_PREFLUSH, be silent. */
867 	if ((mode == SET_OUT_OF_SYNC) && size == 0)
868 		return 0;
869 
870 	if (!plausible_request_size(size)) {
871 		drbd_err(device, "%s: sector=%llus size=%d nonsense!\n",
872 				drbd_change_sync_fname[mode],
873 				(unsigned long long)sector, size);
874 		return 0;
875 	}
876 
877 	if (!get_ldev(device))
878 		return 0; /* no disk, no metadata, no bitmap to manipulate bits in */
879 
880 	nr_sectors = drbd_get_capacity(device->this_bdev);
881 	esector = sector + (size >> 9) - 1;
882 
883 	if (!expect(sector < nr_sectors))
884 		goto out;
885 	if (!expect(esector < nr_sectors))
886 		esector = nr_sectors - 1;
887 
888 	lbnr = BM_SECT_TO_BIT(nr_sectors-1);
889 
890 	if (mode == SET_IN_SYNC) {
891 		/* Round up start sector, round down end sector.  We make sure
892 		 * we only clear full, aligned, BM_BLOCK_SIZE blocks. */
893 		if (unlikely(esector < BM_SECT_PER_BIT-1))
894 			goto out;
895 		if (unlikely(esector == (nr_sectors-1)))
896 			ebnr = lbnr;
897 		else
898 			ebnr = BM_SECT_TO_BIT(esector - (BM_SECT_PER_BIT-1));
899 		sbnr = BM_SECT_TO_BIT(sector + BM_SECT_PER_BIT-1);
900 	} else {
901 		/* We set it out of sync, or record resync failure.
902 		 * Should not round anything here. */
903 		sbnr = BM_SECT_TO_BIT(sector);
904 		ebnr = BM_SECT_TO_BIT(esector);
905 	}
906 
907 	count = update_sync_bits(device, sbnr, ebnr, mode);
908 out:
909 	put_ldev(device);
910 	return count;
911 }
912 
913 static
914 struct bm_extent *_bme_get(struct drbd_device *device, unsigned int enr)
915 {
916 	struct lc_element *e;
917 	struct bm_extent *bm_ext;
918 	int wakeup = 0;
919 	unsigned long rs_flags;
920 
921 	spin_lock_irq(&device->al_lock);
922 	if (device->resync_locked > device->resync->nr_elements/2) {
923 		spin_unlock_irq(&device->al_lock);
924 		return NULL;
925 	}
926 	e = lc_get(device->resync, enr);
927 	bm_ext = e ? lc_entry(e, struct bm_extent, lce) : NULL;
928 	if (bm_ext) {
929 		if (bm_ext->lce.lc_number != enr) {
930 			bm_ext->rs_left = drbd_bm_e_weight(device, enr);
931 			bm_ext->rs_failed = 0;
932 			lc_committed(device->resync);
933 			wakeup = 1;
934 		}
935 		if (bm_ext->lce.refcnt == 1)
936 			device->resync_locked++;
937 		set_bit(BME_NO_WRITES, &bm_ext->flags);
938 	}
939 	rs_flags = device->resync->flags;
940 	spin_unlock_irq(&device->al_lock);
941 	if (wakeup)
942 		wake_up(&device->al_wait);
943 
944 	if (!bm_ext) {
945 		if (rs_flags & LC_STARVING)
946 			drbd_warn(device, "Have to wait for element"
947 			     " (resync LRU too small?)\n");
948 		BUG_ON(rs_flags & LC_LOCKED);
949 	}
950 
951 	return bm_ext;
952 }
953 
954 static int _is_in_al(struct drbd_device *device, unsigned int enr)
955 {
956 	int rv;
957 
958 	spin_lock_irq(&device->al_lock);
959 	rv = lc_is_used(device->act_log, enr);
960 	spin_unlock_irq(&device->al_lock);
961 
962 	return rv;
963 }
964 
965 /**
966  * drbd_rs_begin_io() - Gets an extent in the resync LRU cache and sets it to BME_LOCKED
967  * @device:	DRBD device.
968  * @sector:	The sector number.
969  *
970  * This functions sleeps on al_wait. Returns 0 on success, -EINTR if interrupted.
971  */
972 int drbd_rs_begin_io(struct drbd_device *device, sector_t sector)
973 {
974 	unsigned int enr = BM_SECT_TO_EXT(sector);
975 	struct bm_extent *bm_ext;
976 	int i, sig;
977 	bool sa;
978 
979 retry:
980 	sig = wait_event_interruptible(device->al_wait,
981 			(bm_ext = _bme_get(device, enr)));
982 	if (sig)
983 		return -EINTR;
984 
985 	if (test_bit(BME_LOCKED, &bm_ext->flags))
986 		return 0;
987 
988 	/* step aside only while we are above c-min-rate; unless disabled. */
989 	sa = drbd_rs_c_min_rate_throttle(device);
990 
991 	for (i = 0; i < AL_EXT_PER_BM_SECT; i++) {
992 		sig = wait_event_interruptible(device->al_wait,
993 					       !_is_in_al(device, enr * AL_EXT_PER_BM_SECT + i) ||
994 					       (sa && test_bit(BME_PRIORITY, &bm_ext->flags)));
995 
996 		if (sig || (sa && test_bit(BME_PRIORITY, &bm_ext->flags))) {
997 			spin_lock_irq(&device->al_lock);
998 			if (lc_put(device->resync, &bm_ext->lce) == 0) {
999 				bm_ext->flags = 0; /* clears BME_NO_WRITES and eventually BME_PRIORITY */
1000 				device->resync_locked--;
1001 				wake_up(&device->al_wait);
1002 			}
1003 			spin_unlock_irq(&device->al_lock);
1004 			if (sig)
1005 				return -EINTR;
1006 			if (schedule_timeout_interruptible(HZ/10))
1007 				return -EINTR;
1008 			goto retry;
1009 		}
1010 	}
1011 	set_bit(BME_LOCKED, &bm_ext->flags);
1012 	return 0;
1013 }
1014 
1015 /**
1016  * drbd_try_rs_begin_io() - Gets an extent in the resync LRU cache, does not sleep
1017  * @device:	DRBD device.
1018  * @sector:	The sector number.
1019  *
1020  * Gets an extent in the resync LRU cache, sets it to BME_NO_WRITES, then
1021  * tries to set it to BME_LOCKED. Returns 0 upon success, and -EAGAIN
1022  * if there is still application IO going on in this area.
1023  */
1024 int drbd_try_rs_begin_io(struct drbd_device *device, sector_t sector)
1025 {
1026 	unsigned int enr = BM_SECT_TO_EXT(sector);
1027 	const unsigned int al_enr = enr*AL_EXT_PER_BM_SECT;
1028 	struct lc_element *e;
1029 	struct bm_extent *bm_ext;
1030 	int i;
1031 	bool throttle = drbd_rs_should_slow_down(device, sector, true);
1032 
1033 	/* If we need to throttle, a half-locked (only marked BME_NO_WRITES,
1034 	 * not yet BME_LOCKED) extent needs to be kicked out explicitly if we
1035 	 * need to throttle. There is at most one such half-locked extent,
1036 	 * which is remembered in resync_wenr. */
1037 
1038 	if (throttle && device->resync_wenr != enr)
1039 		return -EAGAIN;
1040 
1041 	spin_lock_irq(&device->al_lock);
1042 	if (device->resync_wenr != LC_FREE && device->resync_wenr != enr) {
1043 		/* in case you have very heavy scattered io, it may
1044 		 * stall the syncer undefined if we give up the ref count
1045 		 * when we try again and requeue.
1046 		 *
1047 		 * if we don't give up the refcount, but the next time
1048 		 * we are scheduled this extent has been "synced" by new
1049 		 * application writes, we'd miss the lc_put on the
1050 		 * extent we keep the refcount on.
1051 		 * so we remembered which extent we had to try again, and
1052 		 * if the next requested one is something else, we do
1053 		 * the lc_put here...
1054 		 * we also have to wake_up
1055 		 */
1056 		e = lc_find(device->resync, device->resync_wenr);
1057 		bm_ext = e ? lc_entry(e, struct bm_extent, lce) : NULL;
1058 		if (bm_ext) {
1059 			D_ASSERT(device, !test_bit(BME_LOCKED, &bm_ext->flags));
1060 			D_ASSERT(device, test_bit(BME_NO_WRITES, &bm_ext->flags));
1061 			clear_bit(BME_NO_WRITES, &bm_ext->flags);
1062 			device->resync_wenr = LC_FREE;
1063 			if (lc_put(device->resync, &bm_ext->lce) == 0) {
1064 				bm_ext->flags = 0;
1065 				device->resync_locked--;
1066 			}
1067 			wake_up(&device->al_wait);
1068 		} else {
1069 			drbd_alert(device, "LOGIC BUG\n");
1070 		}
1071 	}
1072 	/* TRY. */
1073 	e = lc_try_get(device->resync, enr);
1074 	bm_ext = e ? lc_entry(e, struct bm_extent, lce) : NULL;
1075 	if (bm_ext) {
1076 		if (test_bit(BME_LOCKED, &bm_ext->flags))
1077 			goto proceed;
1078 		if (!test_and_set_bit(BME_NO_WRITES, &bm_ext->flags)) {
1079 			device->resync_locked++;
1080 		} else {
1081 			/* we did set the BME_NO_WRITES,
1082 			 * but then could not set BME_LOCKED,
1083 			 * so we tried again.
1084 			 * drop the extra reference. */
1085 			bm_ext->lce.refcnt--;
1086 			D_ASSERT(device, bm_ext->lce.refcnt > 0);
1087 		}
1088 		goto check_al;
1089 	} else {
1090 		/* do we rather want to try later? */
1091 		if (device->resync_locked > device->resync->nr_elements-3)
1092 			goto try_again;
1093 		/* Do or do not. There is no try. -- Yoda */
1094 		e = lc_get(device->resync, enr);
1095 		bm_ext = e ? lc_entry(e, struct bm_extent, lce) : NULL;
1096 		if (!bm_ext) {
1097 			const unsigned long rs_flags = device->resync->flags;
1098 			if (rs_flags & LC_STARVING)
1099 				drbd_warn(device, "Have to wait for element"
1100 				     " (resync LRU too small?)\n");
1101 			BUG_ON(rs_flags & LC_LOCKED);
1102 			goto try_again;
1103 		}
1104 		if (bm_ext->lce.lc_number != enr) {
1105 			bm_ext->rs_left = drbd_bm_e_weight(device, enr);
1106 			bm_ext->rs_failed = 0;
1107 			lc_committed(device->resync);
1108 			wake_up(&device->al_wait);
1109 			D_ASSERT(device, test_bit(BME_LOCKED, &bm_ext->flags) == 0);
1110 		}
1111 		set_bit(BME_NO_WRITES, &bm_ext->flags);
1112 		D_ASSERT(device, bm_ext->lce.refcnt == 1);
1113 		device->resync_locked++;
1114 		goto check_al;
1115 	}
1116 check_al:
1117 	for (i = 0; i < AL_EXT_PER_BM_SECT; i++) {
1118 		if (lc_is_used(device->act_log, al_enr+i))
1119 			goto try_again;
1120 	}
1121 	set_bit(BME_LOCKED, &bm_ext->flags);
1122 proceed:
1123 	device->resync_wenr = LC_FREE;
1124 	spin_unlock_irq(&device->al_lock);
1125 	return 0;
1126 
1127 try_again:
1128 	if (bm_ext) {
1129 		if (throttle) {
1130 			D_ASSERT(device, !test_bit(BME_LOCKED, &bm_ext->flags));
1131 			D_ASSERT(device, test_bit(BME_NO_WRITES, &bm_ext->flags));
1132 			clear_bit(BME_NO_WRITES, &bm_ext->flags);
1133 			device->resync_wenr = LC_FREE;
1134 			if (lc_put(device->resync, &bm_ext->lce) == 0) {
1135 				bm_ext->flags = 0;
1136 				device->resync_locked--;
1137 			}
1138 			wake_up(&device->al_wait);
1139 		} else
1140 			device->resync_wenr = enr;
1141 	}
1142 	spin_unlock_irq(&device->al_lock);
1143 	return -EAGAIN;
1144 }
1145 
1146 void drbd_rs_complete_io(struct drbd_device *device, sector_t sector)
1147 {
1148 	unsigned int enr = BM_SECT_TO_EXT(sector);
1149 	struct lc_element *e;
1150 	struct bm_extent *bm_ext;
1151 	unsigned long flags;
1152 
1153 	spin_lock_irqsave(&device->al_lock, flags);
1154 	e = lc_find(device->resync, enr);
1155 	bm_ext = e ? lc_entry(e, struct bm_extent, lce) : NULL;
1156 	if (!bm_ext) {
1157 		spin_unlock_irqrestore(&device->al_lock, flags);
1158 		if (__ratelimit(&drbd_ratelimit_state))
1159 			drbd_err(device, "drbd_rs_complete_io() called, but extent not found\n");
1160 		return;
1161 	}
1162 
1163 	if (bm_ext->lce.refcnt == 0) {
1164 		spin_unlock_irqrestore(&device->al_lock, flags);
1165 		drbd_err(device, "drbd_rs_complete_io(,%llu [=%u]) called, "
1166 		    "but refcnt is 0!?\n",
1167 		    (unsigned long long)sector, enr);
1168 		return;
1169 	}
1170 
1171 	if (lc_put(device->resync, &bm_ext->lce) == 0) {
1172 		bm_ext->flags = 0; /* clear BME_LOCKED, BME_NO_WRITES and BME_PRIORITY */
1173 		device->resync_locked--;
1174 		wake_up(&device->al_wait);
1175 	}
1176 
1177 	spin_unlock_irqrestore(&device->al_lock, flags);
1178 }
1179 
1180 /**
1181  * drbd_rs_cancel_all() - Removes all extents from the resync LRU (even BME_LOCKED)
1182  * @device:	DRBD device.
1183  */
1184 void drbd_rs_cancel_all(struct drbd_device *device)
1185 {
1186 	spin_lock_irq(&device->al_lock);
1187 
1188 	if (get_ldev_if_state(device, D_FAILED)) { /* Makes sure ->resync is there. */
1189 		lc_reset(device->resync);
1190 		put_ldev(device);
1191 	}
1192 	device->resync_locked = 0;
1193 	device->resync_wenr = LC_FREE;
1194 	spin_unlock_irq(&device->al_lock);
1195 	wake_up(&device->al_wait);
1196 }
1197 
1198 /**
1199  * drbd_rs_del_all() - Gracefully remove all extents from the resync LRU
1200  * @device:	DRBD device.
1201  *
1202  * Returns 0 upon success, -EAGAIN if at least one reference count was
1203  * not zero.
1204  */
1205 int drbd_rs_del_all(struct drbd_device *device)
1206 {
1207 	struct lc_element *e;
1208 	struct bm_extent *bm_ext;
1209 	int i;
1210 
1211 	spin_lock_irq(&device->al_lock);
1212 
1213 	if (get_ldev_if_state(device, D_FAILED)) {
1214 		/* ok, ->resync is there. */
1215 		for (i = 0; i < device->resync->nr_elements; i++) {
1216 			e = lc_element_by_index(device->resync, i);
1217 			bm_ext = lc_entry(e, struct bm_extent, lce);
1218 			if (bm_ext->lce.lc_number == LC_FREE)
1219 				continue;
1220 			if (bm_ext->lce.lc_number == device->resync_wenr) {
1221 				drbd_info(device, "dropping %u in drbd_rs_del_all, apparently"
1222 				     " got 'synced' by application io\n",
1223 				     device->resync_wenr);
1224 				D_ASSERT(device, !test_bit(BME_LOCKED, &bm_ext->flags));
1225 				D_ASSERT(device, test_bit(BME_NO_WRITES, &bm_ext->flags));
1226 				clear_bit(BME_NO_WRITES, &bm_ext->flags);
1227 				device->resync_wenr = LC_FREE;
1228 				lc_put(device->resync, &bm_ext->lce);
1229 			}
1230 			if (bm_ext->lce.refcnt != 0) {
1231 				drbd_info(device, "Retrying drbd_rs_del_all() later. "
1232 				     "refcnt=%d\n", bm_ext->lce.refcnt);
1233 				put_ldev(device);
1234 				spin_unlock_irq(&device->al_lock);
1235 				return -EAGAIN;
1236 			}
1237 			D_ASSERT(device, !test_bit(BME_LOCKED, &bm_ext->flags));
1238 			D_ASSERT(device, !test_bit(BME_NO_WRITES, &bm_ext->flags));
1239 			lc_del(device->resync, &bm_ext->lce);
1240 		}
1241 		D_ASSERT(device, device->resync->used == 0);
1242 		put_ldev(device);
1243 	}
1244 	spin_unlock_irq(&device->al_lock);
1245 	wake_up(&device->al_wait);
1246 
1247 	return 0;
1248 }
1249