Commit | Line | Data |
---|---|---|
b411b363 PR |
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> | |
7ad651b5 | 27 | #include <linux/crc32c.h> |
b411b363 | 28 | #include <linux/drbd.h> |
7ad651b5 LE |
29 | #include <linux/drbd_limits.h> |
30 | #include <linux/dynamic_debug.h> | |
b411b363 | 31 | #include "drbd_int.h" |
b411b363 PR |
32 | #include "drbd_wrappers.h" |
33 | ||
85f103d8 LE |
34 | |
35 | enum al_transaction_types { | |
36 | AL_TR_UPDATE = 0, | |
37 | AL_TR_INITIALIZED = 0xffff | |
38 | }; | |
7ad651b5 LE |
39 | /* all fields on disc in big endian */ |
40 | struct __packed al_transaction_on_disk { | |
41 | /* don't we all like magic */ | |
42 | __be32 magic; | |
43 | ||
44 | /* to identify the most recent transaction block | |
45 | * in the on disk ring buffer */ | |
46 | __be32 tr_number; | |
47 | ||
48 | /* checksum on the full 4k block, with this field set to 0. */ | |
49 | __be32 crc32c; | |
50 | ||
51 | /* type of transaction, special transaction types like: | |
85f103d8 LE |
52 | * purge-all, set-all-idle, set-all-active, ... to-be-defined |
53 | * see also enum al_transaction_types */ | |
7ad651b5 LE |
54 | __be16 transaction_type; |
55 | ||
56 | /* we currently allow only a few thousand extents, | |
57 | * so 16bit will be enough for the slot number. */ | |
58 | ||
59 | /* how many updates in this transaction */ | |
60 | __be16 n_updates; | |
61 | ||
62 | /* maximum slot number, "al-extents" in drbd.conf speak. | |
63 | * Having this in each transaction should make reconfiguration | |
64 | * of that parameter easier. */ | |
65 | __be16 context_size; | |
66 | ||
67 | /* slot number the context starts with */ | |
68 | __be16 context_start_slot_nr; | |
69 | ||
70 | /* Some reserved bytes. Expected usage is a 64bit counter of | |
71 | * sectors-written since device creation, and other data generation tag | |
72 | * supporting usage */ | |
73 | __be32 __reserved[4]; | |
74 | ||
75 | /* --- 36 byte used --- */ | |
76 | ||
77 | /* Reserve space for up to AL_UPDATES_PER_TRANSACTION changes | |
78 | * in one transaction, then use the remaining byte in the 4k block for | |
79 | * context information. "Flexible" number of updates per transaction | |
80 | * does not help, as we have to account for the case when all update | |
81 | * slots are used anyways, so it would only complicate code without | |
82 | * additional benefit. | |
83 | */ | |
84 | __be16 update_slot_nr[AL_UPDATES_PER_TRANSACTION]; | |
85 | ||
86 | /* but the extent number is 32bit, which at an extent size of 4 MiB | |
87 | * allows to cover device sizes of up to 2**54 Byte (16 PiB) */ | |
88 | __be32 update_extent_nr[AL_UPDATES_PER_TRANSACTION]; | |
89 | ||
90 | /* --- 420 bytes used (36 + 64*6) --- */ | |
91 | ||
92 | /* 4096 - 420 = 3676 = 919 * 4 */ | |
93 | __be32 context[AL_CONTEXT_PER_TRANSACTION]; | |
b411b363 PR |
94 | }; |
95 | ||
96 | struct update_odbm_work { | |
97 | struct drbd_work w; | |
98 | unsigned int enr; | |
99 | }; | |
100 | ||
101 | struct update_al_work { | |
102 | struct drbd_work w; | |
b411b363 | 103 | struct completion event; |
7ad651b5 | 104 | int err; |
b411b363 PR |
105 | }; |
106 | ||
b411b363 | 107 | |
cdfda633 PR |
108 | void *drbd_md_get_buffer(struct drbd_conf *mdev) |
109 | { | |
110 | int r; | |
111 | ||
112 | wait_event(mdev->misc_wait, | |
113 | (r = atomic_cmpxchg(&mdev->md_io_in_use, 0, 1)) == 0 || | |
114 | mdev->state.disk <= D_FAILED); | |
115 | ||
116 | return r ? NULL : page_address(mdev->md_io_page); | |
117 | } | |
118 | ||
119 | void drbd_md_put_buffer(struct drbd_conf *mdev) | |
120 | { | |
121 | if (atomic_dec_and_test(&mdev->md_io_in_use)) | |
122 | wake_up(&mdev->misc_wait); | |
123 | } | |
124 | ||
e34b677d | 125 | void wait_until_done_or_force_detached(struct drbd_conf *mdev, struct drbd_backing_dev *bdev, |
32db80f6 | 126 | unsigned int *done) |
cdfda633 | 127 | { |
32db80f6 PR |
128 | long dt; |
129 | ||
130 | rcu_read_lock(); | |
131 | dt = rcu_dereference(bdev->disk_conf)->disk_timeout; | |
132 | rcu_read_unlock(); | |
133 | dt = dt * HZ / 10; | |
134 | if (dt == 0) | |
135 | dt = MAX_SCHEDULE_TIMEOUT; | |
136 | ||
e34b677d LE |
137 | dt = wait_event_timeout(mdev->misc_wait, |
138 | *done || test_bit(FORCE_DETACH, &mdev->flags), dt); | |
139 | if (dt == 0) { | |
32db80f6 | 140 | dev_err(DEV, "meta-data IO operation timed out\n"); |
e34b677d LE |
141 | drbd_chk_io_error(mdev, 1, DRBD_FORCE_DETACH); |
142 | } | |
cdfda633 PR |
143 | } |
144 | ||
b411b363 PR |
145 | static int _drbd_md_sync_page_io(struct drbd_conf *mdev, |
146 | struct drbd_backing_dev *bdev, | |
147 | struct page *page, sector_t sector, | |
148 | int rw, int size) | |
149 | { | |
150 | struct bio *bio; | |
ac29f403 | 151 | int err; |
b411b363 | 152 | |
cdfda633 PR |
153 | mdev->md_io.done = 0; |
154 | mdev->md_io.error = -ENODEV; | |
b411b363 | 155 | |
a8a4e51e | 156 | if ((rw & WRITE) && !test_bit(MD_NO_FUA, &mdev->flags)) |
86e1e98e | 157 | rw |= REQ_FUA | REQ_FLUSH; |
721a9602 | 158 | rw |= REQ_SYNC; |
b411b363 | 159 | |
da4a75d2 | 160 | bio = bio_alloc_drbd(GFP_NOIO); |
b411b363 | 161 | bio->bi_bdev = bdev->md_bdev; |
4f024f37 | 162 | bio->bi_iter.bi_sector = sector; |
ac29f403 AG |
163 | err = -EIO; |
164 | if (bio_add_page(bio, page, size, 0) != size) | |
b411b363 | 165 | goto out; |
cdfda633 | 166 | bio->bi_private = &mdev->md_io; |
b411b363 PR |
167 | bio->bi_end_io = drbd_md_io_complete; |
168 | bio->bi_rw = rw; | |
169 | ||
c04ccaa6 LE |
170 | if (!(rw & WRITE) && mdev->state.disk == D_DISKLESS && mdev->ldev == NULL) |
171 | /* special case, drbd_md_read() during drbd_adm_attach(): no get_ldev */ | |
172 | ; | |
173 | else if (!get_ldev_if_state(mdev, D_ATTACHING)) { | |
174 | /* Corresponding put_ldev in drbd_md_io_complete() */ | |
cdfda633 PR |
175 | dev_err(DEV, "ASSERT FAILED: get_ldev_if_state() == 1 in _drbd_md_sync_page_io()\n"); |
176 | err = -ENODEV; | |
177 | goto out; | |
178 | } | |
179 | ||
180 | bio_get(bio); /* one bio_put() is in the completion handler */ | |
181 | atomic_inc(&mdev->md_io_in_use); /* drbd_md_put_buffer() is in the completion handler */ | |
0cf9d27e | 182 | if (drbd_insert_fault(mdev, (rw & WRITE) ? DRBD_FAULT_MD_WR : DRBD_FAULT_MD_RD)) |
b411b363 PR |
183 | bio_endio(bio, -EIO); |
184 | else | |
185 | submit_bio(rw, bio); | |
e34b677d | 186 | wait_until_done_or_force_detached(mdev, bdev, &mdev->md_io.done); |
ac29f403 | 187 | if (bio_flagged(bio, BIO_UPTODATE)) |
cdfda633 | 188 | err = mdev->md_io.error; |
b411b363 | 189 | |
b411b363 PR |
190 | out: |
191 | bio_put(bio); | |
ac29f403 | 192 | return err; |
b411b363 PR |
193 | } |
194 | ||
195 | int drbd_md_sync_page_io(struct drbd_conf *mdev, struct drbd_backing_dev *bdev, | |
196 | sector_t sector, int rw) | |
197 | { | |
3fbf4d21 | 198 | int err; |
b411b363 PR |
199 | struct page *iop = mdev->md_io_page; |
200 | ||
cdfda633 | 201 | D_ASSERT(atomic_read(&mdev->md_io_in_use) == 1); |
b411b363 PR |
202 | |
203 | BUG_ON(!bdev->md_bdev); | |
204 | ||
c04ccaa6 | 205 | dev_dbg(DEV, "meta_data io: %s [%d]:%s(,%llus,%s) %pS\n", |
7ad651b5 | 206 | current->comm, current->pid, __func__, |
c04ccaa6 LE |
207 | (unsigned long long)sector, (rw & WRITE) ? "WRITE" : "READ", |
208 | (void*)_RET_IP_ ); | |
b411b363 PR |
209 | |
210 | if (sector < drbd_md_first_sector(bdev) || | |
7ad651b5 | 211 | sector + 7 > drbd_md_last_sector(bdev)) |
b411b363 PR |
212 | dev_alert(DEV, "%s [%d]:%s(,%llus,%s) out of range md access!\n", |
213 | current->comm, current->pid, __func__, | |
214 | (unsigned long long)sector, (rw & WRITE) ? "WRITE" : "READ"); | |
215 | ||
ae8bf312 LE |
216 | /* we do all our meta data IO in aligned 4k blocks. */ |
217 | err = _drbd_md_sync_page_io(mdev, bdev, iop, sector, rw, 4096); | |
3fbf4d21 | 218 | if (err) { |
935be260 AG |
219 | dev_err(DEV, "drbd_md_sync_page_io(,%llus,%s) failed with error %d\n", |
220 | (unsigned long long)sector, (rw & WRITE) ? "WRITE" : "READ", err); | |
b411b363 | 221 | } |
3fbf4d21 | 222 | return err; |
b411b363 PR |
223 | } |
224 | ||
6c3c4355 | 225 | static struct bm_extent *find_active_resync_extent(struct drbd_conf *mdev, unsigned int enr) |
b411b363 | 226 | { |
b411b363 | 227 | struct lc_element *tmp; |
b411b363 PR |
228 | tmp = lc_find(mdev->resync, enr/AL_EXT_PER_BM_SECT); |
229 | if (unlikely(tmp != NULL)) { | |
230 | struct bm_extent *bm_ext = lc_entry(tmp, struct bm_extent, lce); | |
6c3c4355 LE |
231 | if (test_bit(BME_NO_WRITES, &bm_ext->flags)) |
232 | return bm_ext; | |
b411b363 | 233 | } |
6c3c4355 LE |
234 | return NULL; |
235 | } | |
236 | ||
237 | static struct lc_element *_al_get(struct drbd_conf *mdev, unsigned int enr, bool nonblock) | |
238 | { | |
239 | struct lc_element *al_ext; | |
240 | struct bm_extent *bm_ext; | |
241 | int wake; | |
242 | ||
243 | spin_lock_irq(&mdev->al_lock); | |
244 | bm_ext = find_active_resync_extent(mdev, enr); | |
245 | if (bm_ext) { | |
246 | wake = !test_and_set_bit(BME_PRIORITY, &bm_ext->flags); | |
247 | spin_unlock_irq(&mdev->al_lock); | |
248 | if (wake) | |
249 | wake_up(&mdev->al_wait); | |
250 | return NULL; | |
251 | } | |
252 | if (nonblock) | |
253 | al_ext = lc_try_get(mdev->act_log, enr); | |
254 | else | |
255 | al_ext = lc_get(mdev->act_log, enr); | |
b411b363 | 256 | spin_unlock_irq(&mdev->al_lock); |
b411b363 PR |
257 | return al_ext; |
258 | } | |
259 | ||
b5bc8e08 | 260 | bool drbd_al_begin_io_fastpath(struct drbd_conf *mdev, struct drbd_interval *i) |
b411b363 | 261 | { |
7726547e LE |
262 | /* for bios crossing activity log extent boundaries, |
263 | * we may need to activate two extents in one go */ | |
e15766e9 | 264 | unsigned first = i->sector >> (AL_EXTENT_SHIFT-9); |
81a3537a | 265 | unsigned last = i->size == 0 ? first : (i->sector + (i->size >> 9) - 1) >> (AL_EXTENT_SHIFT-9); |
7dc1d67f | 266 | |
b5bc8e08 LE |
267 | D_ASSERT((unsigned)(last - first) <= 1); |
268 | D_ASSERT(atomic_read(&mdev->local_cnt) > 0); | |
269 | ||
270 | /* FIXME figure out a fast path for bios crossing AL extent boundaries */ | |
271 | if (first != last) | |
272 | return false; | |
273 | ||
6c3c4355 | 274 | return _al_get(mdev, first, true); |
b5bc8e08 | 275 | } |
56392d2f | 276 | |
b5bc8e08 LE |
277 | bool drbd_al_begin_io_prepare(struct drbd_conf *mdev, struct drbd_interval *i) |
278 | { | |
279 | /* for bios crossing activity log extent boundaries, | |
280 | * we may need to activate two extents in one go */ | |
281 | unsigned first = i->sector >> (AL_EXTENT_SHIFT-9); | |
282 | unsigned last = i->size == 0 ? first : (i->sector + (i->size >> 9) - 1) >> (AL_EXTENT_SHIFT-9); | |
283 | unsigned enr; | |
284 | bool need_transaction = false; | |
b411b363 | 285 | |
81a3537a | 286 | D_ASSERT(first <= last); |
b411b363 PR |
287 | D_ASSERT(atomic_read(&mdev->local_cnt) > 0); |
288 | ||
ebfd5d8f LE |
289 | for (enr = first; enr <= last; enr++) { |
290 | struct lc_element *al_ext; | |
6c3c4355 LE |
291 | wait_event(mdev->al_wait, |
292 | (al_ext = _al_get(mdev, enr, false)) != NULL); | |
ebfd5d8f LE |
293 | if (al_ext->lc_number != enr) |
294 | need_transaction = true; | |
295 | } | |
b5bc8e08 LE |
296 | return need_transaction; |
297 | } | |
298 | ||
299 | static int al_write_transaction(struct drbd_conf *mdev, bool delegate); | |
300 | ||
301 | /* When called through generic_make_request(), we must delegate | |
302 | * activity log I/O to the worker thread: a further request | |
303 | * submitted via generic_make_request() within the same task | |
304 | * would be queued on current->bio_list, and would only start | |
305 | * after this function returns (see generic_make_request()). | |
306 | * | |
307 | * However, if we *are* the worker, we must not delegate to ourselves. | |
308 | */ | |
309 | ||
310 | /* | |
311 | * @delegate: delegate activity log I/O to the worker thread | |
312 | */ | |
313 | void drbd_al_begin_io_commit(struct drbd_conf *mdev, bool delegate) | |
314 | { | |
315 | bool locked = false; | |
ebfd5d8f | 316 | |
b5bc8e08 | 317 | BUG_ON(delegate && current == mdev->tconn->worker.task); |
b411b363 | 318 | |
7dc1d67f LE |
319 | /* Serialize multiple transactions. |
320 | * This uses test_and_set_bit, memory barrier is implicit. | |
321 | */ | |
322 | wait_event(mdev->al_wait, | |
323 | mdev->act_log->pending_changes == 0 || | |
324 | (locked = lc_try_lock_for_transaction(mdev->act_log))); | |
325 | ||
326 | if (locked) { | |
7ad651b5 LE |
327 | /* Double check: it may have been committed by someone else, |
328 | * while we have been waiting for the lock. */ | |
e15766e9 | 329 | if (mdev->act_log->pending_changes) { |
9a51ab1c PR |
330 | bool write_al_updates; |
331 | ||
332 | rcu_read_lock(); | |
333 | write_al_updates = rcu_dereference(mdev->ldev->disk_conf)->al_updates; | |
334 | rcu_read_unlock(); | |
335 | ||
b5bc8e08 | 336 | if (write_al_updates) |
56392d2f | 337 | al_write_transaction(mdev, delegate); |
7ad651b5 LE |
338 | spin_lock_irq(&mdev->al_lock); |
339 | /* FIXME | |
1b7ab15b | 340 | if (err) |
7ad651b5 LE |
341 | we need an "lc_cancel" here; |
342 | */ | |
343 | lc_committed(mdev->act_log); | |
344 | spin_unlock_irq(&mdev->al_lock); | |
345 | } | |
346 | lc_unlock(mdev->act_log); | |
b411b363 PR |
347 | wake_up(&mdev->al_wait); |
348 | } | |
349 | } | |
350 | ||
b5bc8e08 LE |
351 | /* |
352 | * @delegate: delegate activity log I/O to the worker thread | |
353 | */ | |
354 | void drbd_al_begin_io(struct drbd_conf *mdev, struct drbd_interval *i, bool delegate) | |
355 | { | |
356 | BUG_ON(delegate && current == mdev->tconn->worker.task); | |
357 | ||
358 | if (drbd_al_begin_io_prepare(mdev, i)) | |
359 | drbd_al_begin_io_commit(mdev, delegate); | |
360 | } | |
361 | ||
08a1ddab LE |
362 | int drbd_al_begin_io_nonblock(struct drbd_conf *mdev, struct drbd_interval *i) |
363 | { | |
364 | struct lru_cache *al = mdev->act_log; | |
365 | /* for bios crossing activity log extent boundaries, | |
366 | * we may need to activate two extents in one go */ | |
367 | unsigned first = i->sector >> (AL_EXTENT_SHIFT-9); | |
368 | unsigned last = i->size == 0 ? first : (i->sector + (i->size >> 9) - 1) >> (AL_EXTENT_SHIFT-9); | |
369 | unsigned nr_al_extents; | |
370 | unsigned available_update_slots; | |
371 | unsigned enr; | |
372 | ||
373 | D_ASSERT(first <= last); | |
374 | ||
375 | nr_al_extents = 1 + last - first; /* worst case: all touched extends are cold. */ | |
376 | available_update_slots = min(al->nr_elements - al->used, | |
377 | al->max_pending_changes - al->pending_changes); | |
378 | ||
379 | /* We want all necessary updates for a given request within the same transaction | |
380 | * We could first check how many updates are *actually* needed, | |
381 | * and use that instead of the worst-case nr_al_extents */ | |
382 | if (available_update_slots < nr_al_extents) | |
383 | return -EWOULDBLOCK; | |
384 | ||
385 | /* Is resync active in this area? */ | |
386 | for (enr = first; enr <= last; enr++) { | |
387 | struct lc_element *tmp; | |
388 | tmp = lc_find(mdev->resync, enr/AL_EXT_PER_BM_SECT); | |
389 | if (unlikely(tmp != NULL)) { | |
390 | struct bm_extent *bm_ext = lc_entry(tmp, struct bm_extent, lce); | |
391 | if (test_bit(BME_NO_WRITES, &bm_ext->flags)) { | |
0b6ef416 | 392 | if (!test_and_set_bit(BME_PRIORITY, &bm_ext->flags)) |
08a1ddab LE |
393 | return -EBUSY; |
394 | return -EWOULDBLOCK; | |
395 | } | |
396 | } | |
397 | } | |
398 | ||
399 | /* Checkout the refcounts. | |
400 | * Given that we checked for available elements and update slots above, | |
401 | * this has to be successful. */ | |
402 | for (enr = first; enr <= last; enr++) { | |
403 | struct lc_element *al_ext; | |
404 | al_ext = lc_get_cumulative(mdev->act_log, enr); | |
405 | if (!al_ext) | |
406 | dev_info(DEV, "LOGIC BUG for enr=%u\n", enr); | |
407 | } | |
408 | return 0; | |
409 | } | |
410 | ||
181286ad | 411 | void drbd_al_complete_io(struct drbd_conf *mdev, struct drbd_interval *i) |
b411b363 | 412 | { |
e15766e9 LE |
413 | /* for bios crossing activity log extent boundaries, |
414 | * we may need to activate two extents in one go */ | |
415 | unsigned first = i->sector >> (AL_EXTENT_SHIFT-9); | |
81a3537a | 416 | unsigned last = i->size == 0 ? first : (i->sector + (i->size >> 9) - 1) >> (AL_EXTENT_SHIFT-9); |
e15766e9 | 417 | unsigned enr; |
b411b363 PR |
418 | struct lc_element *extent; |
419 | unsigned long flags; | |
420 | ||
81a3537a | 421 | D_ASSERT(first <= last); |
b411b363 PR |
422 | spin_lock_irqsave(&mdev->al_lock, flags); |
423 | ||
e15766e9 LE |
424 | for (enr = first; enr <= last; enr++) { |
425 | extent = lc_find(mdev->act_log, enr); | |
426 | if (!extent) { | |
427 | dev_err(DEV, "al_complete_io() called on inactive extent %u\n", enr); | |
428 | continue; | |
429 | } | |
376694a0 | 430 | lc_put(mdev->act_log, extent); |
b411b363 | 431 | } |
b411b363 | 432 | spin_unlock_irqrestore(&mdev->al_lock, flags); |
e15766e9 | 433 | wake_up(&mdev->al_wait); |
b411b363 PR |
434 | } |
435 | ||
19f843aa LE |
436 | #if (PAGE_SHIFT + 3) < (AL_EXTENT_SHIFT - BM_BLOCK_SHIFT) |
437 | /* Currently BM_BLOCK_SHIFT, BM_EXT_SHIFT and AL_EXTENT_SHIFT | |
438 | * are still coupled, or assume too much about their relation. | |
439 | * Code below will not work if this is violated. | |
440 | * Will be cleaned up with some followup patch. | |
441 | */ | |
442 | # error FIXME | |
443 | #endif | |
444 | ||
445 | static unsigned int al_extent_to_bm_page(unsigned int al_enr) | |
446 | { | |
447 | return al_enr >> | |
448 | /* bit to page */ | |
449 | ((PAGE_SHIFT + 3) - | |
450 | /* al extent number to bit */ | |
451 | (AL_EXTENT_SHIFT - BM_BLOCK_SHIFT)); | |
452 | } | |
453 | ||
454 | static unsigned int rs_extent_to_bm_page(unsigned int rs_enr) | |
455 | { | |
456 | return rs_enr >> | |
457 | /* bit to page */ | |
458 | ((PAGE_SHIFT + 3) - | |
acb104c3 | 459 | /* resync extent number to bit */ |
19f843aa LE |
460 | (BM_EXT_SHIFT - BM_BLOCK_SHIFT)); |
461 | } | |
462 | ||
ae8bf312 LE |
463 | static sector_t al_tr_number_to_on_disk_sector(struct drbd_conf *mdev) |
464 | { | |
3a4d4eb3 LE |
465 | const unsigned int stripes = mdev->ldev->md.al_stripes; |
466 | const unsigned int stripe_size_4kB = mdev->ldev->md.al_stripe_size_4k; | |
ae8bf312 LE |
467 | |
468 | /* transaction number, modulo on-disk ring buffer wrap around */ | |
3a4d4eb3 | 469 | unsigned int t = mdev->al_tr_number % (mdev->ldev->md.al_size_4k); |
ae8bf312 LE |
470 | |
471 | /* ... to aligned 4k on disk block */ | |
472 | t = ((t % stripes) * stripe_size_4kB) + t/stripes; | |
473 | ||
474 | /* ... to 512 byte sector in activity log */ | |
475 | t *= 8; | |
476 | ||
477 | /* ... plus offset to the on disk position */ | |
478 | return mdev->ldev->md.md_offset + mdev->ldev->md.al_offset + t; | |
479 | } | |
480 | ||
99920dc5 | 481 | static int |
1b7ab15b | 482 | _al_write_transaction(struct drbd_conf *mdev) |
b411b363 | 483 | { |
7ad651b5 LE |
484 | struct al_transaction_on_disk *buffer; |
485 | struct lc_element *e; | |
b411b363 | 486 | sector_t sector; |
7ad651b5 LE |
487 | int i, mx; |
488 | unsigned extent_nr; | |
489 | unsigned crc = 0; | |
1b7ab15b | 490 | int err = 0; |
b411b363 PR |
491 | |
492 | if (!get_ldev(mdev)) { | |
7ad651b5 LE |
493 | dev_err(DEV, "disk is %s, cannot start al transaction\n", |
494 | drbd_disk_str(mdev->state.disk)); | |
1b7ab15b | 495 | return -EIO; |
b411b363 | 496 | } |
b411b363 | 497 | |
6719fb03 LE |
498 | /* The bitmap write may have failed, causing a state change. */ |
499 | if (mdev->state.disk < D_INCONSISTENT) { | |
500 | dev_err(DEV, | |
7ad651b5 LE |
501 | "disk is %s, cannot write al transaction\n", |
502 | drbd_disk_str(mdev->state.disk)); | |
6719fb03 | 503 | put_ldev(mdev); |
1b7ab15b | 504 | return -EIO; |
6719fb03 LE |
505 | } |
506 | ||
cdfda633 PR |
507 | buffer = drbd_md_get_buffer(mdev); /* protects md_io_buffer, al_tr_cycle, ... */ |
508 | if (!buffer) { | |
509 | dev_err(DEV, "disk failed while waiting for md_io buffer\n"); | |
cdfda633 | 510 | put_ldev(mdev); |
1b7ab15b | 511 | return -ENODEV; |
cdfda633 | 512 | } |
b411b363 | 513 | |
7ad651b5 LE |
514 | memset(buffer, 0, sizeof(*buffer)); |
515 | buffer->magic = cpu_to_be32(DRBD_AL_MAGIC); | |
b411b363 PR |
516 | buffer->tr_number = cpu_to_be32(mdev->al_tr_number); |
517 | ||
7ad651b5 LE |
518 | i = 0; |
519 | ||
520 | /* Even though no one can start to change this list | |
521 | * once we set the LC_LOCKED -- from drbd_al_begin_io(), | |
522 | * lc_try_lock_for_transaction() --, someone may still | |
523 | * be in the process of changing it. */ | |
524 | spin_lock_irq(&mdev->al_lock); | |
525 | list_for_each_entry(e, &mdev->act_log->to_be_changed, list) { | |
526 | if (i == AL_UPDATES_PER_TRANSACTION) { | |
527 | i++; | |
528 | break; | |
529 | } | |
530 | buffer->update_slot_nr[i] = cpu_to_be16(e->lc_index); | |
531 | buffer->update_extent_nr[i] = cpu_to_be32(e->lc_new_number); | |
532 | if (e->lc_number != LC_FREE) | |
533 | drbd_bm_mark_for_writeout(mdev, | |
534 | al_extent_to_bm_page(e->lc_number)); | |
535 | i++; | |
536 | } | |
537 | spin_unlock_irq(&mdev->al_lock); | |
538 | BUG_ON(i > AL_UPDATES_PER_TRANSACTION); | |
b411b363 | 539 | |
7ad651b5 LE |
540 | buffer->n_updates = cpu_to_be16(i); |
541 | for ( ; i < AL_UPDATES_PER_TRANSACTION; i++) { | |
542 | buffer->update_slot_nr[i] = cpu_to_be16(-1); | |
543 | buffer->update_extent_nr[i] = cpu_to_be32(LC_FREE); | |
544 | } | |
b411b363 | 545 | |
7ad651b5 LE |
546 | buffer->context_size = cpu_to_be16(mdev->act_log->nr_elements); |
547 | buffer->context_start_slot_nr = cpu_to_be16(mdev->al_tr_cycle); | |
b411b363 | 548 | |
7ad651b5 | 549 | mx = min_t(int, AL_CONTEXT_PER_TRANSACTION, |
b411b363 PR |
550 | mdev->act_log->nr_elements - mdev->al_tr_cycle); |
551 | for (i = 0; i < mx; i++) { | |
552 | unsigned idx = mdev->al_tr_cycle + i; | |
553 | extent_nr = lc_element_by_index(mdev->act_log, idx)->lc_number; | |
7ad651b5 | 554 | buffer->context[i] = cpu_to_be32(extent_nr); |
b411b363 | 555 | } |
7ad651b5 LE |
556 | for (; i < AL_CONTEXT_PER_TRANSACTION; i++) |
557 | buffer->context[i] = cpu_to_be32(LC_FREE); | |
558 | ||
559 | mdev->al_tr_cycle += AL_CONTEXT_PER_TRANSACTION; | |
b411b363 PR |
560 | if (mdev->al_tr_cycle >= mdev->act_log->nr_elements) |
561 | mdev->al_tr_cycle = 0; | |
562 | ||
ae8bf312 | 563 | sector = al_tr_number_to_on_disk_sector(mdev); |
b411b363 | 564 | |
7ad651b5 LE |
565 | crc = crc32c(0, buffer, 4096); |
566 | buffer->crc32c = cpu_to_be32(crc); | |
b411b363 | 567 | |
7ad651b5 | 568 | if (drbd_bm_write_hinted(mdev)) |
1b7ab15b | 569 | err = -EIO; |
b5bc8e08 LE |
570 | else { |
571 | bool write_al_updates; | |
572 | rcu_read_lock(); | |
573 | write_al_updates = rcu_dereference(mdev->ldev->disk_conf)->al_updates; | |
574 | rcu_read_unlock(); | |
575 | if (write_al_updates) { | |
576 | if (drbd_md_sync_page_io(mdev, mdev->ldev, sector, WRITE)) { | |
577 | err = -EIO; | |
578 | drbd_chk_io_error(mdev, 1, DRBD_META_IO_ERROR); | |
579 | } else { | |
580 | mdev->al_tr_number++; | |
581 | mdev->al_writ_cnt++; | |
582 | } | |
583 | } | |
7ad651b5 | 584 | } |
b411b363 | 585 | |
cdfda633 | 586 | drbd_md_put_buffer(mdev); |
b411b363 PR |
587 | put_ldev(mdev); |
588 | ||
1b7ab15b PR |
589 | return err; |
590 | } | |
591 | ||
592 | ||
593 | static int w_al_write_transaction(struct drbd_work *w, int unused) | |
594 | { | |
595 | struct update_al_work *aw = container_of(w, struct update_al_work, w); | |
596 | struct drbd_conf *mdev = w->mdev; | |
597 | int err; | |
598 | ||
599 | err = _al_write_transaction(mdev); | |
600 | aw->err = err; | |
601 | complete(&aw->event); | |
602 | ||
603 | return err != -EIO ? err : 0; | |
604 | } | |
605 | ||
606 | /* Calls from worker context (see w_restart_disk_io()) need to write the | |
607 | transaction directly. Others came through generic_make_request(), | |
608 | those need to delegate it to the worker. */ | |
56392d2f | 609 | static int al_write_transaction(struct drbd_conf *mdev, bool delegate) |
1b7ab15b | 610 | { |
56392d2f LE |
611 | if (delegate) { |
612 | struct update_al_work al_work; | |
613 | init_completion(&al_work.event); | |
614 | al_work.w.cb = w_al_write_transaction; | |
615 | al_work.w.mdev = mdev; | |
616 | drbd_queue_work_front(&mdev->tconn->sender_work, &al_work.w); | |
617 | wait_for_completion(&al_work.event); | |
618 | return al_work.err; | |
619 | } else | |
1b7ab15b | 620 | return _al_write_transaction(mdev); |
b411b363 PR |
621 | } |
622 | ||
b411b363 PR |
623 | static int _try_lc_del(struct drbd_conf *mdev, struct lc_element *al_ext) |
624 | { | |
625 | int rv; | |
626 | ||
627 | spin_lock_irq(&mdev->al_lock); | |
628 | rv = (al_ext->refcnt == 0); | |
629 | if (likely(rv)) | |
630 | lc_del(mdev->act_log, al_ext); | |
631 | spin_unlock_irq(&mdev->al_lock); | |
632 | ||
633 | return rv; | |
634 | } | |
635 | ||
636 | /** | |
637 | * drbd_al_shrink() - Removes all active extents form the activity log | |
638 | * @mdev: DRBD device. | |
639 | * | |
640 | * Removes all active extents form the activity log, waiting until | |
641 | * the reference count of each entry dropped to 0 first, of course. | |
642 | * | |
643 | * You need to lock mdev->act_log with lc_try_lock() / lc_unlock() | |
644 | */ | |
645 | void drbd_al_shrink(struct drbd_conf *mdev) | |
646 | { | |
647 | struct lc_element *al_ext; | |
648 | int i; | |
649 | ||
46a15bc3 | 650 | D_ASSERT(test_bit(__LC_LOCKED, &mdev->act_log->flags)); |
b411b363 PR |
651 | |
652 | for (i = 0; i < mdev->act_log->nr_elements; i++) { | |
653 | al_ext = lc_element_by_index(mdev->act_log, i); | |
654 | if (al_ext->lc_number == LC_FREE) | |
655 | continue; | |
656 | wait_event(mdev->al_wait, _try_lc_del(mdev, al_ext)); | |
657 | } | |
658 | ||
659 | wake_up(&mdev->al_wait); | |
660 | } | |
661 | ||
d752b269 PR |
662 | int drbd_initialize_al(struct drbd_conf *mdev, void *buffer) |
663 | { | |
664 | struct al_transaction_on_disk *al = buffer; | |
665 | struct drbd_md *md = &mdev->ldev->md; | |
666 | sector_t al_base = md->md_offset + md->al_offset; | |
667 | int al_size_4k = md->al_stripes * md->al_stripe_size_4k; | |
668 | int i; | |
669 | ||
670 | memset(al, 0, 4096); | |
671 | al->magic = cpu_to_be32(DRBD_AL_MAGIC); | |
672 | al->transaction_type = cpu_to_be16(AL_TR_INITIALIZED); | |
673 | al->crc32c = cpu_to_be32(crc32c(0, al, 4096)); | |
674 | ||
675 | for (i = 0; i < al_size_4k; i++) { | |
676 | int err = drbd_md_sync_page_io(mdev, mdev->ldev, al_base + i * 8, WRITE); | |
677 | if (err) | |
678 | return err; | |
679 | } | |
680 | return 0; | |
681 | } | |
682 | ||
99920dc5 | 683 | static int w_update_odbm(struct drbd_work *w, int unused) |
b411b363 PR |
684 | { |
685 | struct update_odbm_work *udw = container_of(w, struct update_odbm_work, w); | |
00d56944 | 686 | struct drbd_conf *mdev = w->mdev; |
3b98c0c2 | 687 | struct sib_info sib = { .sib_reason = SIB_SYNC_PROGRESS, }; |
b411b363 PR |
688 | |
689 | if (!get_ldev(mdev)) { | |
690 | if (__ratelimit(&drbd_ratelimit_state)) | |
691 | dev_warn(DEV, "Can not update on disk bitmap, local IO disabled.\n"); | |
692 | kfree(udw); | |
99920dc5 | 693 | return 0; |
b411b363 PR |
694 | } |
695 | ||
19f843aa | 696 | drbd_bm_write_page(mdev, rs_extent_to_bm_page(udw->enr)); |
b411b363 PR |
697 | put_ldev(mdev); |
698 | ||
699 | kfree(udw); | |
700 | ||
701 | if (drbd_bm_total_weight(mdev) <= mdev->rs_failed) { | |
702 | switch (mdev->state.conn) { | |
703 | case C_SYNC_SOURCE: case C_SYNC_TARGET: | |
704 | case C_PAUSED_SYNC_S: case C_PAUSED_SYNC_T: | |
705 | drbd_resync_finished(mdev); | |
706 | default: | |
707 | /* nothing to do */ | |
708 | break; | |
709 | } | |
710 | } | |
3b98c0c2 | 711 | drbd_bcast_event(mdev, &sib); |
b411b363 | 712 | |
99920dc5 | 713 | return 0; |
b411b363 PR |
714 | } |
715 | ||
716 | ||
717 | /* ATTENTION. The AL's extents are 4MB each, while the extents in the | |
718 | * resync LRU-cache are 16MB each. | |
719 | * The caller of this function has to hold an get_ldev() reference. | |
720 | * | |
721 | * TODO will be obsoleted once we have a caching lru of the on disk bitmap | |
722 | */ | |
723 | static void drbd_try_clear_on_disk_bm(struct drbd_conf *mdev, sector_t sector, | |
724 | int count, int success) | |
725 | { | |
726 | struct lc_element *e; | |
727 | struct update_odbm_work *udw; | |
728 | ||
729 | unsigned int enr; | |
730 | ||
731 | D_ASSERT(atomic_read(&mdev->local_cnt)); | |
732 | ||
733 | /* I simply assume that a sector/size pair never crosses | |
734 | * a 16 MB extent border. (Currently this is true...) */ | |
735 | enr = BM_SECT_TO_EXT(sector); | |
736 | ||
737 | e = lc_get(mdev->resync, enr); | |
738 | if (e) { | |
739 | struct bm_extent *ext = lc_entry(e, struct bm_extent, lce); | |
740 | if (ext->lce.lc_number == enr) { | |
741 | if (success) | |
742 | ext->rs_left -= count; | |
743 | else | |
744 | ext->rs_failed += count; | |
745 | if (ext->rs_left < ext->rs_failed) { | |
975b2979 PR |
746 | dev_warn(DEV, "BAD! sector=%llus enr=%u rs_left=%d " |
747 | "rs_failed=%d count=%d cstate=%s\n", | |
b411b363 PR |
748 | (unsigned long long)sector, |
749 | ext->lce.lc_number, ext->rs_left, | |
975b2979 PR |
750 | ext->rs_failed, count, |
751 | drbd_conn_str(mdev->state.conn)); | |
752 | ||
753 | /* We don't expect to be able to clear more bits | |
754 | * than have been set when we originally counted | |
755 | * the set bits to cache that value in ext->rs_left. | |
756 | * Whatever the reason (disconnect during resync, | |
757 | * delayed local completion of an application write), | |
758 | * try to fix it up by recounting here. */ | |
759 | ext->rs_left = drbd_bm_e_weight(mdev, enr); | |
b411b363 PR |
760 | } |
761 | } else { | |
762 | /* Normally this element should be in the cache, | |
763 | * since drbd_rs_begin_io() pulled it already in. | |
764 | * | |
765 | * But maybe an application write finished, and we set | |
766 | * something outside the resync lru_cache in sync. | |
767 | */ | |
768 | int rs_left = drbd_bm_e_weight(mdev, enr); | |
769 | if (ext->flags != 0) { | |
770 | dev_warn(DEV, "changing resync lce: %d[%u;%02lx]" | |
771 | " -> %d[%u;00]\n", | |
772 | ext->lce.lc_number, ext->rs_left, | |
773 | ext->flags, enr, rs_left); | |
774 | ext->flags = 0; | |
775 | } | |
776 | if (ext->rs_failed) { | |
777 | dev_warn(DEV, "Kicking resync_lru element enr=%u " | |
778 | "out with rs_failed=%d\n", | |
779 | ext->lce.lc_number, ext->rs_failed); | |
b411b363 PR |
780 | } |
781 | ext->rs_left = rs_left; | |
782 | ext->rs_failed = success ? 0 : count; | |
46a15bc3 LE |
783 | /* we don't keep a persistent log of the resync lru, |
784 | * we can commit any change right away. */ | |
785 | lc_committed(mdev->resync); | |
b411b363 PR |
786 | } |
787 | lc_put(mdev->resync, &ext->lce); | |
788 | /* no race, we are within the al_lock! */ | |
789 | ||
790 | if (ext->rs_left == ext->rs_failed) { | |
791 | ext->rs_failed = 0; | |
792 | ||
793 | udw = kmalloc(sizeof(*udw), GFP_ATOMIC); | |
794 | if (udw) { | |
795 | udw->enr = ext->lce.lc_number; | |
796 | udw->w.cb = w_update_odbm; | |
a21e9298 | 797 | udw->w.mdev = mdev; |
d5b27b01 | 798 | drbd_queue_work_front(&mdev->tconn->sender_work, &udw->w); |
b411b363 PR |
799 | } else { |
800 | dev_warn(DEV, "Could not kmalloc an udw\n"); | |
b411b363 PR |
801 | } |
802 | } | |
803 | } else { | |
804 | dev_err(DEV, "lc_get() failed! locked=%d/%d flags=%lu\n", | |
805 | mdev->resync_locked, | |
806 | mdev->resync->nr_elements, | |
807 | mdev->resync->flags); | |
808 | } | |
809 | } | |
810 | ||
c6ea14df LE |
811 | void drbd_advance_rs_marks(struct drbd_conf *mdev, unsigned long still_to_go) |
812 | { | |
813 | unsigned long now = jiffies; | |
814 | unsigned long last = mdev->rs_mark_time[mdev->rs_last_mark]; | |
815 | int next = (mdev->rs_last_mark + 1) % DRBD_SYNC_MARKS; | |
816 | if (time_after_eq(now, last + DRBD_SYNC_MARK_STEP)) { | |
817 | if (mdev->rs_mark_left[mdev->rs_last_mark] != still_to_go && | |
818 | mdev->state.conn != C_PAUSED_SYNC_T && | |
819 | mdev->state.conn != C_PAUSED_SYNC_S) { | |
820 | mdev->rs_mark_time[next] = now; | |
821 | mdev->rs_mark_left[next] = still_to_go; | |
822 | mdev->rs_last_mark = next; | |
823 | } | |
824 | } | |
825 | } | |
826 | ||
b411b363 PR |
827 | /* clear the bit corresponding to the piece of storage in question: |
828 | * size byte of data starting from sector. Only clear a bits of the affected | |
829 | * one ore more _aligned_ BM_BLOCK_SIZE blocks. | |
830 | * | |
831 | * called by worker on C_SYNC_TARGET and receiver on SyncSource. | |
832 | * | |
833 | */ | |
834 | void __drbd_set_in_sync(struct drbd_conf *mdev, sector_t sector, int size, | |
835 | const char *file, const unsigned int line) | |
836 | { | |
837 | /* Is called from worker and receiver context _only_ */ | |
838 | unsigned long sbnr, ebnr, lbnr; | |
839 | unsigned long count = 0; | |
840 | sector_t esector, nr_sectors; | |
841 | int wake_up = 0; | |
842 | unsigned long flags; | |
843 | ||
c670a398 | 844 | if (size <= 0 || !IS_ALIGNED(size, 512) || size > DRBD_MAX_BIO_SIZE) { |
b411b363 PR |
845 | dev_err(DEV, "drbd_set_in_sync: sector=%llus size=%d nonsense!\n", |
846 | (unsigned long long)sector, size); | |
847 | return; | |
848 | } | |
518a4d53 PR |
849 | |
850 | if (!get_ldev(mdev)) | |
851 | return; /* no disk, no metadata, no bitmap to clear bits in */ | |
852 | ||
b411b363 PR |
853 | nr_sectors = drbd_get_capacity(mdev->this_bdev); |
854 | esector = sector + (size >> 9) - 1; | |
855 | ||
841ce241 | 856 | if (!expect(sector < nr_sectors)) |
518a4d53 | 857 | goto out; |
841ce241 AG |
858 | if (!expect(esector < nr_sectors)) |
859 | esector = nr_sectors - 1; | |
b411b363 PR |
860 | |
861 | lbnr = BM_SECT_TO_BIT(nr_sectors-1); | |
862 | ||
863 | /* we clear it (in sync). | |
864 | * round up start sector, round down end sector. we make sure we only | |
865 | * clear full, aligned, BM_BLOCK_SIZE (4K) blocks */ | |
866 | if (unlikely(esector < BM_SECT_PER_BIT-1)) | |
518a4d53 | 867 | goto out; |
b411b363 PR |
868 | if (unlikely(esector == (nr_sectors-1))) |
869 | ebnr = lbnr; | |
870 | else | |
871 | ebnr = BM_SECT_TO_BIT(esector - (BM_SECT_PER_BIT-1)); | |
872 | sbnr = BM_SECT_TO_BIT(sector + BM_SECT_PER_BIT-1); | |
873 | ||
b411b363 | 874 | if (sbnr > ebnr) |
518a4d53 | 875 | goto out; |
b411b363 PR |
876 | |
877 | /* | |
878 | * ok, (capacity & 7) != 0 sometimes, but who cares... | |
879 | * we count rs_{total,left} in bits, not sectors. | |
880 | */ | |
b411b363 | 881 | count = drbd_bm_clear_bits(mdev, sbnr, ebnr); |
518a4d53 | 882 | if (count) { |
c6ea14df | 883 | drbd_advance_rs_marks(mdev, drbd_bm_total_weight(mdev)); |
1d7734a0 | 884 | spin_lock_irqsave(&mdev->al_lock, flags); |
81e84650 | 885 | drbd_try_clear_on_disk_bm(mdev, sector, count, true); |
1d7734a0 LE |
886 | spin_unlock_irqrestore(&mdev->al_lock, flags); |
887 | ||
b411b363 PR |
888 | /* just wake_up unconditional now, various lc_chaged(), |
889 | * lc_put() in drbd_try_clear_on_disk_bm(). */ | |
890 | wake_up = 1; | |
891 | } | |
518a4d53 PR |
892 | out: |
893 | put_ldev(mdev); | |
b411b363 PR |
894 | if (wake_up) |
895 | wake_up(&mdev->al_wait); | |
896 | } | |
897 | ||
898 | /* | |
899 | * this is intended to set one request worth of data out of sync. | |
900 | * affects at least 1 bit, | |
1816a2b4 | 901 | * and at most 1+DRBD_MAX_BIO_SIZE/BM_BLOCK_SIZE bits. |
b411b363 PR |
902 | * |
903 | * called by tl_clear and drbd_send_dblock (==drbd_make_request). | |
904 | * so this can be _any_ process. | |
905 | */ | |
73a01a18 | 906 | int __drbd_set_out_of_sync(struct drbd_conf *mdev, sector_t sector, int size, |
b411b363 PR |
907 | const char *file, const unsigned int line) |
908 | { | |
376694a0 | 909 | unsigned long sbnr, ebnr, flags; |
b411b363 | 910 | sector_t esector, nr_sectors; |
73a01a18 | 911 | unsigned int enr, count = 0; |
b411b363 PR |
912 | struct lc_element *e; |
913 | ||
81a3537a LE |
914 | /* this should be an empty REQ_FLUSH */ |
915 | if (size == 0) | |
916 | return 0; | |
917 | ||
918 | if (size < 0 || !IS_ALIGNED(size, 512) || size > DRBD_MAX_BIO_SIZE) { | |
b411b363 PR |
919 | dev_err(DEV, "sector: %llus, size: %d\n", |
920 | (unsigned long long)sector, size); | |
73a01a18 | 921 | return 0; |
b411b363 PR |
922 | } |
923 | ||
924 | if (!get_ldev(mdev)) | |
73a01a18 | 925 | return 0; /* no disk, no metadata, no bitmap to set bits in */ |
b411b363 PR |
926 | |
927 | nr_sectors = drbd_get_capacity(mdev->this_bdev); | |
928 | esector = sector + (size >> 9) - 1; | |
929 | ||
841ce241 | 930 | if (!expect(sector < nr_sectors)) |
b411b363 | 931 | goto out; |
841ce241 AG |
932 | if (!expect(esector < nr_sectors)) |
933 | esector = nr_sectors - 1; | |
b411b363 | 934 | |
b411b363 PR |
935 | /* we set it out of sync, |
936 | * we do not need to round anything here */ | |
937 | sbnr = BM_SECT_TO_BIT(sector); | |
938 | ebnr = BM_SECT_TO_BIT(esector); | |
939 | ||
b411b363 PR |
940 | /* ok, (capacity & 7) != 0 sometimes, but who cares... |
941 | * we count rs_{total,left} in bits, not sectors. */ | |
942 | spin_lock_irqsave(&mdev->al_lock, flags); | |
943 | count = drbd_bm_set_bits(mdev, sbnr, ebnr); | |
944 | ||
945 | enr = BM_SECT_TO_EXT(sector); | |
946 | e = lc_find(mdev->resync, enr); | |
947 | if (e) | |
948 | lc_entry(e, struct bm_extent, lce)->rs_left += count; | |
949 | spin_unlock_irqrestore(&mdev->al_lock, flags); | |
950 | ||
951 | out: | |
952 | put_ldev(mdev); | |
73a01a18 PR |
953 | |
954 | return count; | |
b411b363 PR |
955 | } |
956 | ||
957 | static | |
958 | struct bm_extent *_bme_get(struct drbd_conf *mdev, unsigned int enr) | |
959 | { | |
960 | struct lc_element *e; | |
961 | struct bm_extent *bm_ext; | |
962 | int wakeup = 0; | |
963 | unsigned long rs_flags; | |
964 | ||
965 | spin_lock_irq(&mdev->al_lock); | |
966 | if (mdev->resync_locked > mdev->resync->nr_elements/2) { | |
967 | spin_unlock_irq(&mdev->al_lock); | |
968 | return NULL; | |
969 | } | |
970 | e = lc_get(mdev->resync, enr); | |
971 | bm_ext = e ? lc_entry(e, struct bm_extent, lce) : NULL; | |
972 | if (bm_ext) { | |
973 | if (bm_ext->lce.lc_number != enr) { | |
974 | bm_ext->rs_left = drbd_bm_e_weight(mdev, enr); | |
975 | bm_ext->rs_failed = 0; | |
46a15bc3 | 976 | lc_committed(mdev->resync); |
b411b363 PR |
977 | wakeup = 1; |
978 | } | |
979 | if (bm_ext->lce.refcnt == 1) | |
980 | mdev->resync_locked++; | |
981 | set_bit(BME_NO_WRITES, &bm_ext->flags); | |
982 | } | |
983 | rs_flags = mdev->resync->flags; | |
984 | spin_unlock_irq(&mdev->al_lock); | |
985 | if (wakeup) | |
986 | wake_up(&mdev->al_wait); | |
987 | ||
988 | if (!bm_ext) { | |
989 | if (rs_flags & LC_STARVING) | |
990 | dev_warn(DEV, "Have to wait for element" | |
991 | " (resync LRU too small?)\n"); | |
46a15bc3 | 992 | BUG_ON(rs_flags & LC_LOCKED); |
b411b363 PR |
993 | } |
994 | ||
995 | return bm_ext; | |
996 | } | |
997 | ||
998 | static int _is_in_al(struct drbd_conf *mdev, unsigned int enr) | |
999 | { | |
46a15bc3 | 1000 | int rv; |
b411b363 PR |
1001 | |
1002 | spin_lock_irq(&mdev->al_lock); | |
46a15bc3 | 1003 | rv = lc_is_used(mdev->act_log, enr); |
b411b363 PR |
1004 | spin_unlock_irq(&mdev->al_lock); |
1005 | ||
b411b363 PR |
1006 | return rv; |
1007 | } | |
1008 | ||
1009 | /** | |
1010 | * drbd_rs_begin_io() - Gets an extent in the resync LRU cache and sets it to BME_LOCKED | |
1011 | * @mdev: DRBD device. | |
1012 | * @sector: The sector number. | |
1013 | * | |
80a40e43 | 1014 | * This functions sleeps on al_wait. Returns 0 on success, -EINTR if interrupted. |
b411b363 PR |
1015 | */ |
1016 | int drbd_rs_begin_io(struct drbd_conf *mdev, sector_t sector) | |
1017 | { | |
1018 | unsigned int enr = BM_SECT_TO_EXT(sector); | |
1019 | struct bm_extent *bm_ext; | |
1020 | int i, sig; | |
f91ab628 PR |
1021 | int sa = 200; /* Step aside 200 times, then grab the extent and let app-IO wait. |
1022 | 200 times -> 20 seconds. */ | |
b411b363 | 1023 | |
f91ab628 | 1024 | retry: |
b411b363 PR |
1025 | sig = wait_event_interruptible(mdev->al_wait, |
1026 | (bm_ext = _bme_get(mdev, enr))); | |
1027 | if (sig) | |
80a40e43 | 1028 | return -EINTR; |
b411b363 PR |
1029 | |
1030 | if (test_bit(BME_LOCKED, &bm_ext->flags)) | |
80a40e43 | 1031 | return 0; |
b411b363 PR |
1032 | |
1033 | for (i = 0; i < AL_EXT_PER_BM_SECT; i++) { | |
1034 | sig = wait_event_interruptible(mdev->al_wait, | |
f91ab628 | 1035 | !_is_in_al(mdev, enr * AL_EXT_PER_BM_SECT + i) || |
c507f46f | 1036 | test_bit(BME_PRIORITY, &bm_ext->flags)); |
f91ab628 PR |
1037 | |
1038 | if (sig || (test_bit(BME_PRIORITY, &bm_ext->flags) && sa)) { | |
b411b363 PR |
1039 | spin_lock_irq(&mdev->al_lock); |
1040 | if (lc_put(mdev->resync, &bm_ext->lce) == 0) { | |
f91ab628 | 1041 | bm_ext->flags = 0; /* clears BME_NO_WRITES and eventually BME_PRIORITY */ |
b411b363 PR |
1042 | mdev->resync_locked--; |
1043 | wake_up(&mdev->al_wait); | |
1044 | } | |
1045 | spin_unlock_irq(&mdev->al_lock); | |
f91ab628 PR |
1046 | if (sig) |
1047 | return -EINTR; | |
1048 | if (schedule_timeout_interruptible(HZ/10)) | |
1049 | return -EINTR; | |
c507f46f PR |
1050 | if (sa && --sa == 0) |
1051 | dev_warn(DEV,"drbd_rs_begin_io() stepped aside for 20sec." | |
1052 | "Resync stalled?\n"); | |
f91ab628 | 1053 | goto retry; |
b411b363 PR |
1054 | } |
1055 | } | |
b411b363 | 1056 | set_bit(BME_LOCKED, &bm_ext->flags); |
80a40e43 | 1057 | return 0; |
b411b363 PR |
1058 | } |
1059 | ||
1060 | /** | |
1061 | * drbd_try_rs_begin_io() - Gets an extent in the resync LRU cache, does not sleep | |
1062 | * @mdev: DRBD device. | |
1063 | * @sector: The sector number. | |
1064 | * | |
1065 | * Gets an extent in the resync LRU cache, sets it to BME_NO_WRITES, then | |
1066 | * tries to set it to BME_LOCKED. Returns 0 upon success, and -EAGAIN | |
1067 | * if there is still application IO going on in this area. | |
1068 | */ | |
1069 | int drbd_try_rs_begin_io(struct drbd_conf *mdev, sector_t sector) | |
1070 | { | |
1071 | unsigned int enr = BM_SECT_TO_EXT(sector); | |
1072 | const unsigned int al_enr = enr*AL_EXT_PER_BM_SECT; | |
1073 | struct lc_element *e; | |
1074 | struct bm_extent *bm_ext; | |
1075 | int i; | |
1076 | ||
b411b363 PR |
1077 | spin_lock_irq(&mdev->al_lock); |
1078 | if (mdev->resync_wenr != LC_FREE && mdev->resync_wenr != enr) { | |
1079 | /* in case you have very heavy scattered io, it may | |
1080 | * stall the syncer undefined if we give up the ref count | |
1081 | * when we try again and requeue. | |
1082 | * | |
1083 | * if we don't give up the refcount, but the next time | |
1084 | * we are scheduled this extent has been "synced" by new | |
1085 | * application writes, we'd miss the lc_put on the | |
1086 | * extent we keep the refcount on. | |
1087 | * so we remembered which extent we had to try again, and | |
1088 | * if the next requested one is something else, we do | |
1089 | * the lc_put here... | |
1090 | * we also have to wake_up | |
1091 | */ | |
b411b363 PR |
1092 | e = lc_find(mdev->resync, mdev->resync_wenr); |
1093 | bm_ext = e ? lc_entry(e, struct bm_extent, lce) : NULL; | |
1094 | if (bm_ext) { | |
1095 | D_ASSERT(!test_bit(BME_LOCKED, &bm_ext->flags)); | |
1096 | D_ASSERT(test_bit(BME_NO_WRITES, &bm_ext->flags)); | |
1097 | clear_bit(BME_NO_WRITES, &bm_ext->flags); | |
1098 | mdev->resync_wenr = LC_FREE; | |
1099 | if (lc_put(mdev->resync, &bm_ext->lce) == 0) | |
1100 | mdev->resync_locked--; | |
1101 | wake_up(&mdev->al_wait); | |
1102 | } else { | |
1103 | dev_alert(DEV, "LOGIC BUG\n"); | |
1104 | } | |
1105 | } | |
1106 | /* TRY. */ | |
1107 | e = lc_try_get(mdev->resync, enr); | |
1108 | bm_ext = e ? lc_entry(e, struct bm_extent, lce) : NULL; | |
1109 | if (bm_ext) { | |
1110 | if (test_bit(BME_LOCKED, &bm_ext->flags)) | |
1111 | goto proceed; | |
1112 | if (!test_and_set_bit(BME_NO_WRITES, &bm_ext->flags)) { | |
1113 | mdev->resync_locked++; | |
1114 | } else { | |
1115 | /* we did set the BME_NO_WRITES, | |
1116 | * but then could not set BME_LOCKED, | |
1117 | * so we tried again. | |
1118 | * drop the extra reference. */ | |
b411b363 PR |
1119 | bm_ext->lce.refcnt--; |
1120 | D_ASSERT(bm_ext->lce.refcnt > 0); | |
1121 | } | |
1122 | goto check_al; | |
1123 | } else { | |
1124 | /* do we rather want to try later? */ | |
6a0afdf5 | 1125 | if (mdev->resync_locked > mdev->resync->nr_elements-3) |
b411b363 | 1126 | goto try_again; |
b411b363 PR |
1127 | /* Do or do not. There is no try. -- Yoda */ |
1128 | e = lc_get(mdev->resync, enr); | |
1129 | bm_ext = e ? lc_entry(e, struct bm_extent, lce) : NULL; | |
1130 | if (!bm_ext) { | |
1131 | const unsigned long rs_flags = mdev->resync->flags; | |
1132 | if (rs_flags & LC_STARVING) | |
1133 | dev_warn(DEV, "Have to wait for element" | |
1134 | " (resync LRU too small?)\n"); | |
46a15bc3 | 1135 | BUG_ON(rs_flags & LC_LOCKED); |
b411b363 PR |
1136 | goto try_again; |
1137 | } | |
1138 | if (bm_ext->lce.lc_number != enr) { | |
1139 | bm_ext->rs_left = drbd_bm_e_weight(mdev, enr); | |
1140 | bm_ext->rs_failed = 0; | |
46a15bc3 | 1141 | lc_committed(mdev->resync); |
b411b363 PR |
1142 | wake_up(&mdev->al_wait); |
1143 | D_ASSERT(test_bit(BME_LOCKED, &bm_ext->flags) == 0); | |
1144 | } | |
1145 | set_bit(BME_NO_WRITES, &bm_ext->flags); | |
1146 | D_ASSERT(bm_ext->lce.refcnt == 1); | |
1147 | mdev->resync_locked++; | |
1148 | goto check_al; | |
1149 | } | |
1150 | check_al: | |
b411b363 | 1151 | for (i = 0; i < AL_EXT_PER_BM_SECT; i++) { |
b411b363 PR |
1152 | if (lc_is_used(mdev->act_log, al_enr+i)) |
1153 | goto try_again; | |
1154 | } | |
1155 | set_bit(BME_LOCKED, &bm_ext->flags); | |
1156 | proceed: | |
1157 | mdev->resync_wenr = LC_FREE; | |
1158 | spin_unlock_irq(&mdev->al_lock); | |
1159 | return 0; | |
1160 | ||
1161 | try_again: | |
b411b363 PR |
1162 | if (bm_ext) |
1163 | mdev->resync_wenr = enr; | |
1164 | spin_unlock_irq(&mdev->al_lock); | |
1165 | return -EAGAIN; | |
1166 | } | |
1167 | ||
1168 | void drbd_rs_complete_io(struct drbd_conf *mdev, sector_t sector) | |
1169 | { | |
1170 | unsigned int enr = BM_SECT_TO_EXT(sector); | |
1171 | struct lc_element *e; | |
1172 | struct bm_extent *bm_ext; | |
1173 | unsigned long flags; | |
1174 | ||
b411b363 PR |
1175 | spin_lock_irqsave(&mdev->al_lock, flags); |
1176 | e = lc_find(mdev->resync, enr); | |
1177 | bm_ext = e ? lc_entry(e, struct bm_extent, lce) : NULL; | |
1178 | if (!bm_ext) { | |
1179 | spin_unlock_irqrestore(&mdev->al_lock, flags); | |
1180 | if (__ratelimit(&drbd_ratelimit_state)) | |
1181 | dev_err(DEV, "drbd_rs_complete_io() called, but extent not found\n"); | |
1182 | return; | |
1183 | } | |
1184 | ||
1185 | if (bm_ext->lce.refcnt == 0) { | |
1186 | spin_unlock_irqrestore(&mdev->al_lock, flags); | |
1187 | dev_err(DEV, "drbd_rs_complete_io(,%llu [=%u]) called, " | |
1188 | "but refcnt is 0!?\n", | |
1189 | (unsigned long long)sector, enr); | |
1190 | return; | |
1191 | } | |
1192 | ||
1193 | if (lc_put(mdev->resync, &bm_ext->lce) == 0) { | |
e3555d85 | 1194 | bm_ext->flags = 0; /* clear BME_LOCKED, BME_NO_WRITES and BME_PRIORITY */ |
b411b363 PR |
1195 | mdev->resync_locked--; |
1196 | wake_up(&mdev->al_wait); | |
1197 | } | |
1198 | ||
1199 | spin_unlock_irqrestore(&mdev->al_lock, flags); | |
1200 | } | |
1201 | ||
1202 | /** | |
1203 | * drbd_rs_cancel_all() - Removes all extents from the resync LRU (even BME_LOCKED) | |
1204 | * @mdev: DRBD device. | |
1205 | */ | |
1206 | void drbd_rs_cancel_all(struct drbd_conf *mdev) | |
1207 | { | |
b411b363 PR |
1208 | spin_lock_irq(&mdev->al_lock); |
1209 | ||
1210 | if (get_ldev_if_state(mdev, D_FAILED)) { /* Makes sure ->resync is there. */ | |
1211 | lc_reset(mdev->resync); | |
1212 | put_ldev(mdev); | |
1213 | } | |
1214 | mdev->resync_locked = 0; | |
1215 | mdev->resync_wenr = LC_FREE; | |
1216 | spin_unlock_irq(&mdev->al_lock); | |
1217 | wake_up(&mdev->al_wait); | |
1218 | } | |
1219 | ||
1220 | /** | |
1221 | * drbd_rs_del_all() - Gracefully remove all extents from the resync LRU | |
1222 | * @mdev: DRBD device. | |
1223 | * | |
1224 | * Returns 0 upon success, -EAGAIN if at least one reference count was | |
1225 | * not zero. | |
1226 | */ | |
1227 | int drbd_rs_del_all(struct drbd_conf *mdev) | |
1228 | { | |
1229 | struct lc_element *e; | |
1230 | struct bm_extent *bm_ext; | |
1231 | int i; | |
1232 | ||
b411b363 PR |
1233 | spin_lock_irq(&mdev->al_lock); |
1234 | ||
1235 | if (get_ldev_if_state(mdev, D_FAILED)) { | |
1236 | /* ok, ->resync is there. */ | |
1237 | for (i = 0; i < mdev->resync->nr_elements; i++) { | |
1238 | e = lc_element_by_index(mdev->resync, i); | |
b2b163dd | 1239 | bm_ext = lc_entry(e, struct bm_extent, lce); |
b411b363 PR |
1240 | if (bm_ext->lce.lc_number == LC_FREE) |
1241 | continue; | |
1242 | if (bm_ext->lce.lc_number == mdev->resync_wenr) { | |
1243 | dev_info(DEV, "dropping %u in drbd_rs_del_all, apparently" | |
1244 | " got 'synced' by application io\n", | |
1245 | mdev->resync_wenr); | |
1246 | D_ASSERT(!test_bit(BME_LOCKED, &bm_ext->flags)); | |
1247 | D_ASSERT(test_bit(BME_NO_WRITES, &bm_ext->flags)); | |
1248 | clear_bit(BME_NO_WRITES, &bm_ext->flags); | |
1249 | mdev->resync_wenr = LC_FREE; | |
1250 | lc_put(mdev->resync, &bm_ext->lce); | |
1251 | } | |
1252 | if (bm_ext->lce.refcnt != 0) { | |
1253 | dev_info(DEV, "Retrying drbd_rs_del_all() later. " | |
1254 | "refcnt=%d\n", bm_ext->lce.refcnt); | |
1255 | put_ldev(mdev); | |
1256 | spin_unlock_irq(&mdev->al_lock); | |
1257 | return -EAGAIN; | |
1258 | } | |
1259 | D_ASSERT(!test_bit(BME_LOCKED, &bm_ext->flags)); | |
1260 | D_ASSERT(!test_bit(BME_NO_WRITES, &bm_ext->flags)); | |
1261 | lc_del(mdev->resync, &bm_ext->lce); | |
1262 | } | |
1263 | D_ASSERT(mdev->resync->used == 0); | |
1264 | put_ldev(mdev); | |
1265 | } | |
1266 | spin_unlock_irq(&mdev->al_lock); | |
a6a7d4f0 | 1267 | wake_up(&mdev->al_wait); |
b411b363 PR |
1268 | |
1269 | return 0; | |
1270 | } | |
1271 | ||
1272 | /** | |
1273 | * drbd_rs_failed_io() - Record information on a failure to resync the specified blocks | |
1274 | * @mdev: DRBD device. | |
1275 | * @sector: The sector number. | |
1276 | * @size: Size of failed IO operation, in byte. | |
1277 | */ | |
1278 | void drbd_rs_failed_io(struct drbd_conf *mdev, sector_t sector, int size) | |
1279 | { | |
1280 | /* Is called from worker and receiver context _only_ */ | |
1281 | unsigned long sbnr, ebnr, lbnr; | |
1282 | unsigned long count; | |
1283 | sector_t esector, nr_sectors; | |
1284 | int wake_up = 0; | |
1285 | ||
c670a398 | 1286 | if (size <= 0 || !IS_ALIGNED(size, 512) || size > DRBD_MAX_BIO_SIZE) { |
b411b363 PR |
1287 | dev_err(DEV, "drbd_rs_failed_io: sector=%llus size=%d nonsense!\n", |
1288 | (unsigned long long)sector, size); | |
1289 | return; | |
1290 | } | |
1291 | nr_sectors = drbd_get_capacity(mdev->this_bdev); | |
1292 | esector = sector + (size >> 9) - 1; | |
1293 | ||
841ce241 AG |
1294 | if (!expect(sector < nr_sectors)) |
1295 | return; | |
1296 | if (!expect(esector < nr_sectors)) | |
1297 | esector = nr_sectors - 1; | |
b411b363 PR |
1298 | |
1299 | lbnr = BM_SECT_TO_BIT(nr_sectors-1); | |
1300 | ||
1301 | /* | |
1302 | * round up start sector, round down end sector. we make sure we only | |
1303 | * handle full, aligned, BM_BLOCK_SIZE (4K) blocks */ | |
1304 | if (unlikely(esector < BM_SECT_PER_BIT-1)) | |
1305 | return; | |
1306 | if (unlikely(esector == (nr_sectors-1))) | |
1307 | ebnr = lbnr; | |
1308 | else | |
1309 | ebnr = BM_SECT_TO_BIT(esector - (BM_SECT_PER_BIT-1)); | |
1310 | sbnr = BM_SECT_TO_BIT(sector + BM_SECT_PER_BIT-1); | |
1311 | ||
1312 | if (sbnr > ebnr) | |
1313 | return; | |
1314 | ||
1315 | /* | |
1316 | * ok, (capacity & 7) != 0 sometimes, but who cares... | |
1317 | * we count rs_{total,left} in bits, not sectors. | |
1318 | */ | |
1319 | spin_lock_irq(&mdev->al_lock); | |
1320 | count = drbd_bm_count_bits(mdev, sbnr, ebnr); | |
1321 | if (count) { | |
1322 | mdev->rs_failed += count; | |
1323 | ||
1324 | if (get_ldev(mdev)) { | |
81e84650 | 1325 | drbd_try_clear_on_disk_bm(mdev, sector, count, false); |
b411b363 PR |
1326 | put_ldev(mdev); |
1327 | } | |
1328 | ||
1329 | /* just wake_up unconditional now, various lc_chaged(), | |
1330 | * lc_put() in drbd_try_clear_on_disk_bm(). */ | |
1331 | wake_up = 1; | |
1332 | } | |
1333 | spin_unlock_irq(&mdev->al_lock); | |
1334 | if (wake_up) | |
1335 | wake_up(&mdev->al_wait); | |
1336 | } |