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
33enum al_transaction_types {
34 AL_TR_UPDATE = 0,
35 AL_TR_INITIALIZED = 0xffff
36};
37/* all fields on disc in big endian */
38struct __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
94void *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
111void 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
117void 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
137static 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
189int 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
218static 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
230static 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
253bool 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
270bool 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
301static 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
310static 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
328static 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
414static 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
451void 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 */
490void 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
496int 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
556void 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
581static 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 */
603void 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
620int 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
644static 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 */
663static 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
750void 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. */
767static bool lazy_bitmap_update_due(struct drbd_device *device)
768{
769 return time_after(jiffies, device->rs_last_bcast + 2*HZ);
770}
771
772static 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
791static 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
844static 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 */
858int __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);
908out:
909 put_ldev(device);
910 return count;
911}
912
913static
914struct 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
954static 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 */
972int 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
979retry:
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 */
1024int 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 }
1116check_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);
1122proceed:
1123 device->resync_wenr = LC_FREE;
1124 spin_unlock_irq(&device->al_lock);
1125 return 0;
1126
1127try_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
1146void 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 */
1184void 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 */
1205int 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