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 PR |
161 | bio->bi_bdev = bdev->md_bdev; |
162 | bio->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 | ||
181286ad | 362 | void drbd_al_complete_io(struct drbd_conf *mdev, struct drbd_interval *i) |
b411b363 | 363 | { |
e15766e9 LE |
364 | /* for bios crossing activity log extent boundaries, |
365 | * we may need to activate two extents in one go */ | |
366 | unsigned first = i->sector >> (AL_EXTENT_SHIFT-9); | |
81a3537a | 367 | unsigned last = i->size == 0 ? first : (i->sector + (i->size >> 9) - 1) >> (AL_EXTENT_SHIFT-9); |
e15766e9 | 368 | unsigned enr; |
b411b363 PR |
369 | struct lc_element *extent; |
370 | unsigned long flags; | |
371 | ||
81a3537a | 372 | D_ASSERT(first <= last); |
b411b363 PR |
373 | spin_lock_irqsave(&mdev->al_lock, flags); |
374 | ||
e15766e9 LE |
375 | for (enr = first; enr <= last; enr++) { |
376 | extent = lc_find(mdev->act_log, enr); | |
377 | if (!extent) { | |
378 | dev_err(DEV, "al_complete_io() called on inactive extent %u\n", enr); | |
379 | continue; | |
380 | } | |
376694a0 | 381 | lc_put(mdev->act_log, extent); |
b411b363 | 382 | } |
b411b363 | 383 | spin_unlock_irqrestore(&mdev->al_lock, flags); |
e15766e9 | 384 | wake_up(&mdev->al_wait); |
b411b363 PR |
385 | } |
386 | ||
19f843aa LE |
387 | #if (PAGE_SHIFT + 3) < (AL_EXTENT_SHIFT - BM_BLOCK_SHIFT) |
388 | /* Currently BM_BLOCK_SHIFT, BM_EXT_SHIFT and AL_EXTENT_SHIFT | |
389 | * are still coupled, or assume too much about their relation. | |
390 | * Code below will not work if this is violated. | |
391 | * Will be cleaned up with some followup patch. | |
392 | */ | |
393 | # error FIXME | |
394 | #endif | |
395 | ||
396 | static unsigned int al_extent_to_bm_page(unsigned int al_enr) | |
397 | { | |
398 | return al_enr >> | |
399 | /* bit to page */ | |
400 | ((PAGE_SHIFT + 3) - | |
401 | /* al extent number to bit */ | |
402 | (AL_EXTENT_SHIFT - BM_BLOCK_SHIFT)); | |
403 | } | |
404 | ||
405 | static unsigned int rs_extent_to_bm_page(unsigned int rs_enr) | |
406 | { | |
407 | return rs_enr >> | |
408 | /* bit to page */ | |
409 | ((PAGE_SHIFT + 3) - | |
acb104c3 | 410 | /* resync extent number to bit */ |
19f843aa LE |
411 | (BM_EXT_SHIFT - BM_BLOCK_SHIFT)); |
412 | } | |
413 | ||
ae8bf312 LE |
414 | static sector_t al_tr_number_to_on_disk_sector(struct drbd_conf *mdev) |
415 | { | |
3a4d4eb3 LE |
416 | const unsigned int stripes = mdev->ldev->md.al_stripes; |
417 | const unsigned int stripe_size_4kB = mdev->ldev->md.al_stripe_size_4k; | |
ae8bf312 LE |
418 | |
419 | /* transaction number, modulo on-disk ring buffer wrap around */ | |
3a4d4eb3 | 420 | unsigned int t = mdev->al_tr_number % (mdev->ldev->md.al_size_4k); |
ae8bf312 LE |
421 | |
422 | /* ... to aligned 4k on disk block */ | |
423 | t = ((t % stripes) * stripe_size_4kB) + t/stripes; | |
424 | ||
425 | /* ... to 512 byte sector in activity log */ | |
426 | t *= 8; | |
427 | ||
428 | /* ... plus offset to the on disk position */ | |
429 | return mdev->ldev->md.md_offset + mdev->ldev->md.al_offset + t; | |
430 | } | |
431 | ||
99920dc5 | 432 | static int |
1b7ab15b | 433 | _al_write_transaction(struct drbd_conf *mdev) |
b411b363 | 434 | { |
7ad651b5 LE |
435 | struct al_transaction_on_disk *buffer; |
436 | struct lc_element *e; | |
b411b363 | 437 | sector_t sector; |
7ad651b5 LE |
438 | int i, mx; |
439 | unsigned extent_nr; | |
440 | unsigned crc = 0; | |
1b7ab15b | 441 | int err = 0; |
b411b363 PR |
442 | |
443 | if (!get_ldev(mdev)) { | |
7ad651b5 LE |
444 | dev_err(DEV, "disk is %s, cannot start al transaction\n", |
445 | drbd_disk_str(mdev->state.disk)); | |
1b7ab15b | 446 | return -EIO; |
b411b363 | 447 | } |
b411b363 | 448 | |
6719fb03 LE |
449 | /* The bitmap write may have failed, causing a state change. */ |
450 | if (mdev->state.disk < D_INCONSISTENT) { | |
451 | dev_err(DEV, | |
7ad651b5 LE |
452 | "disk is %s, cannot write al transaction\n", |
453 | drbd_disk_str(mdev->state.disk)); | |
6719fb03 | 454 | put_ldev(mdev); |
1b7ab15b | 455 | return -EIO; |
6719fb03 LE |
456 | } |
457 | ||
cdfda633 PR |
458 | buffer = drbd_md_get_buffer(mdev); /* protects md_io_buffer, al_tr_cycle, ... */ |
459 | if (!buffer) { | |
460 | dev_err(DEV, "disk failed while waiting for md_io buffer\n"); | |
cdfda633 | 461 | put_ldev(mdev); |
1b7ab15b | 462 | return -ENODEV; |
cdfda633 | 463 | } |
b411b363 | 464 | |
7ad651b5 LE |
465 | memset(buffer, 0, sizeof(*buffer)); |
466 | buffer->magic = cpu_to_be32(DRBD_AL_MAGIC); | |
b411b363 PR |
467 | buffer->tr_number = cpu_to_be32(mdev->al_tr_number); |
468 | ||
7ad651b5 LE |
469 | i = 0; |
470 | ||
471 | /* Even though no one can start to change this list | |
472 | * once we set the LC_LOCKED -- from drbd_al_begin_io(), | |
473 | * lc_try_lock_for_transaction() --, someone may still | |
474 | * be in the process of changing it. */ | |
475 | spin_lock_irq(&mdev->al_lock); | |
476 | list_for_each_entry(e, &mdev->act_log->to_be_changed, list) { | |
477 | if (i == AL_UPDATES_PER_TRANSACTION) { | |
478 | i++; | |
479 | break; | |
480 | } | |
481 | buffer->update_slot_nr[i] = cpu_to_be16(e->lc_index); | |
482 | buffer->update_extent_nr[i] = cpu_to_be32(e->lc_new_number); | |
483 | if (e->lc_number != LC_FREE) | |
484 | drbd_bm_mark_for_writeout(mdev, | |
485 | al_extent_to_bm_page(e->lc_number)); | |
486 | i++; | |
487 | } | |
488 | spin_unlock_irq(&mdev->al_lock); | |
489 | BUG_ON(i > AL_UPDATES_PER_TRANSACTION); | |
b411b363 | 490 | |
7ad651b5 LE |
491 | buffer->n_updates = cpu_to_be16(i); |
492 | for ( ; i < AL_UPDATES_PER_TRANSACTION; i++) { | |
493 | buffer->update_slot_nr[i] = cpu_to_be16(-1); | |
494 | buffer->update_extent_nr[i] = cpu_to_be32(LC_FREE); | |
495 | } | |
b411b363 | 496 | |
7ad651b5 LE |
497 | buffer->context_size = cpu_to_be16(mdev->act_log->nr_elements); |
498 | buffer->context_start_slot_nr = cpu_to_be16(mdev->al_tr_cycle); | |
b411b363 | 499 | |
7ad651b5 | 500 | mx = min_t(int, AL_CONTEXT_PER_TRANSACTION, |
b411b363 PR |
501 | mdev->act_log->nr_elements - mdev->al_tr_cycle); |
502 | for (i = 0; i < mx; i++) { | |
503 | unsigned idx = mdev->al_tr_cycle + i; | |
504 | extent_nr = lc_element_by_index(mdev->act_log, idx)->lc_number; | |
7ad651b5 | 505 | buffer->context[i] = cpu_to_be32(extent_nr); |
b411b363 | 506 | } |
7ad651b5 LE |
507 | for (; i < AL_CONTEXT_PER_TRANSACTION; i++) |
508 | buffer->context[i] = cpu_to_be32(LC_FREE); | |
509 | ||
510 | mdev->al_tr_cycle += AL_CONTEXT_PER_TRANSACTION; | |
b411b363 PR |
511 | if (mdev->al_tr_cycle >= mdev->act_log->nr_elements) |
512 | mdev->al_tr_cycle = 0; | |
513 | ||
ae8bf312 | 514 | sector = al_tr_number_to_on_disk_sector(mdev); |
b411b363 | 515 | |
7ad651b5 LE |
516 | crc = crc32c(0, buffer, 4096); |
517 | buffer->crc32c = cpu_to_be32(crc); | |
b411b363 | 518 | |
7ad651b5 | 519 | if (drbd_bm_write_hinted(mdev)) |
1b7ab15b | 520 | err = -EIO; |
b5bc8e08 LE |
521 | else { |
522 | bool write_al_updates; | |
523 | rcu_read_lock(); | |
524 | write_al_updates = rcu_dereference(mdev->ldev->disk_conf)->al_updates; | |
525 | rcu_read_unlock(); | |
526 | if (write_al_updates) { | |
527 | if (drbd_md_sync_page_io(mdev, mdev->ldev, sector, WRITE)) { | |
528 | err = -EIO; | |
529 | drbd_chk_io_error(mdev, 1, DRBD_META_IO_ERROR); | |
530 | } else { | |
531 | mdev->al_tr_number++; | |
532 | mdev->al_writ_cnt++; | |
533 | } | |
534 | } | |
7ad651b5 | 535 | } |
b411b363 | 536 | |
cdfda633 | 537 | drbd_md_put_buffer(mdev); |
b411b363 PR |
538 | put_ldev(mdev); |
539 | ||
1b7ab15b PR |
540 | return err; |
541 | } | |
542 | ||
543 | ||
544 | static int w_al_write_transaction(struct drbd_work *w, int unused) | |
545 | { | |
546 | struct update_al_work *aw = container_of(w, struct update_al_work, w); | |
547 | struct drbd_conf *mdev = w->mdev; | |
548 | int err; | |
549 | ||
550 | err = _al_write_transaction(mdev); | |
551 | aw->err = err; | |
552 | complete(&aw->event); | |
553 | ||
554 | return err != -EIO ? err : 0; | |
555 | } | |
556 | ||
557 | /* Calls from worker context (see w_restart_disk_io()) need to write the | |
558 | transaction directly. Others came through generic_make_request(), | |
559 | those need to delegate it to the worker. */ | |
56392d2f | 560 | static int al_write_transaction(struct drbd_conf *mdev, bool delegate) |
1b7ab15b | 561 | { |
56392d2f LE |
562 | if (delegate) { |
563 | struct update_al_work al_work; | |
564 | init_completion(&al_work.event); | |
565 | al_work.w.cb = w_al_write_transaction; | |
566 | al_work.w.mdev = mdev; | |
567 | drbd_queue_work_front(&mdev->tconn->sender_work, &al_work.w); | |
568 | wait_for_completion(&al_work.event); | |
569 | return al_work.err; | |
570 | } else | |
1b7ab15b | 571 | return _al_write_transaction(mdev); |
b411b363 PR |
572 | } |
573 | ||
b411b363 PR |
574 | static int _try_lc_del(struct drbd_conf *mdev, struct lc_element *al_ext) |
575 | { | |
576 | int rv; | |
577 | ||
578 | spin_lock_irq(&mdev->al_lock); | |
579 | rv = (al_ext->refcnt == 0); | |
580 | if (likely(rv)) | |
581 | lc_del(mdev->act_log, al_ext); | |
582 | spin_unlock_irq(&mdev->al_lock); | |
583 | ||
584 | return rv; | |
585 | } | |
586 | ||
587 | /** | |
588 | * drbd_al_shrink() - Removes all active extents form the activity log | |
589 | * @mdev: DRBD device. | |
590 | * | |
591 | * Removes all active extents form the activity log, waiting until | |
592 | * the reference count of each entry dropped to 0 first, of course. | |
593 | * | |
594 | * You need to lock mdev->act_log with lc_try_lock() / lc_unlock() | |
595 | */ | |
596 | void drbd_al_shrink(struct drbd_conf *mdev) | |
597 | { | |
598 | struct lc_element *al_ext; | |
599 | int i; | |
600 | ||
46a15bc3 | 601 | D_ASSERT(test_bit(__LC_LOCKED, &mdev->act_log->flags)); |
b411b363 PR |
602 | |
603 | for (i = 0; i < mdev->act_log->nr_elements; i++) { | |
604 | al_ext = lc_element_by_index(mdev->act_log, i); | |
605 | if (al_ext->lc_number == LC_FREE) | |
606 | continue; | |
607 | wait_event(mdev->al_wait, _try_lc_del(mdev, al_ext)); | |
608 | } | |
609 | ||
610 | wake_up(&mdev->al_wait); | |
611 | } | |
612 | ||
99920dc5 | 613 | static int w_update_odbm(struct drbd_work *w, int unused) |
b411b363 PR |
614 | { |
615 | struct update_odbm_work *udw = container_of(w, struct update_odbm_work, w); | |
00d56944 | 616 | struct drbd_conf *mdev = w->mdev; |
3b98c0c2 | 617 | struct sib_info sib = { .sib_reason = SIB_SYNC_PROGRESS, }; |
b411b363 PR |
618 | |
619 | if (!get_ldev(mdev)) { | |
620 | if (__ratelimit(&drbd_ratelimit_state)) | |
621 | dev_warn(DEV, "Can not update on disk bitmap, local IO disabled.\n"); | |
622 | kfree(udw); | |
99920dc5 | 623 | return 0; |
b411b363 PR |
624 | } |
625 | ||
19f843aa | 626 | drbd_bm_write_page(mdev, rs_extent_to_bm_page(udw->enr)); |
b411b363 PR |
627 | put_ldev(mdev); |
628 | ||
629 | kfree(udw); | |
630 | ||
631 | if (drbd_bm_total_weight(mdev) <= mdev->rs_failed) { | |
632 | switch (mdev->state.conn) { | |
633 | case C_SYNC_SOURCE: case C_SYNC_TARGET: | |
634 | case C_PAUSED_SYNC_S: case C_PAUSED_SYNC_T: | |
635 | drbd_resync_finished(mdev); | |
636 | default: | |
637 | /* nothing to do */ | |
638 | break; | |
639 | } | |
640 | } | |
3b98c0c2 | 641 | drbd_bcast_event(mdev, &sib); |
b411b363 | 642 | |
99920dc5 | 643 | return 0; |
b411b363 PR |
644 | } |
645 | ||
646 | ||
647 | /* ATTENTION. The AL's extents are 4MB each, while the extents in the | |
648 | * resync LRU-cache are 16MB each. | |
649 | * The caller of this function has to hold an get_ldev() reference. | |
650 | * | |
651 | * TODO will be obsoleted once we have a caching lru of the on disk bitmap | |
652 | */ | |
653 | static void drbd_try_clear_on_disk_bm(struct drbd_conf *mdev, sector_t sector, | |
654 | int count, int success) | |
655 | { | |
656 | struct lc_element *e; | |
657 | struct update_odbm_work *udw; | |
658 | ||
659 | unsigned int enr; | |
660 | ||
661 | D_ASSERT(atomic_read(&mdev->local_cnt)); | |
662 | ||
663 | /* I simply assume that a sector/size pair never crosses | |
664 | * a 16 MB extent border. (Currently this is true...) */ | |
665 | enr = BM_SECT_TO_EXT(sector); | |
666 | ||
667 | e = lc_get(mdev->resync, enr); | |
668 | if (e) { | |
669 | struct bm_extent *ext = lc_entry(e, struct bm_extent, lce); | |
670 | if (ext->lce.lc_number == enr) { | |
671 | if (success) | |
672 | ext->rs_left -= count; | |
673 | else | |
674 | ext->rs_failed += count; | |
675 | if (ext->rs_left < ext->rs_failed) { | |
975b2979 PR |
676 | dev_warn(DEV, "BAD! sector=%llus enr=%u rs_left=%d " |
677 | "rs_failed=%d count=%d cstate=%s\n", | |
b411b363 PR |
678 | (unsigned long long)sector, |
679 | ext->lce.lc_number, ext->rs_left, | |
975b2979 PR |
680 | ext->rs_failed, count, |
681 | drbd_conn_str(mdev->state.conn)); | |
682 | ||
683 | /* We don't expect to be able to clear more bits | |
684 | * than have been set when we originally counted | |
685 | * the set bits to cache that value in ext->rs_left. | |
686 | * Whatever the reason (disconnect during resync, | |
687 | * delayed local completion of an application write), | |
688 | * try to fix it up by recounting here. */ | |
689 | ext->rs_left = drbd_bm_e_weight(mdev, enr); | |
b411b363 PR |
690 | } |
691 | } else { | |
692 | /* Normally this element should be in the cache, | |
693 | * since drbd_rs_begin_io() pulled it already in. | |
694 | * | |
695 | * But maybe an application write finished, and we set | |
696 | * something outside the resync lru_cache in sync. | |
697 | */ | |
698 | int rs_left = drbd_bm_e_weight(mdev, enr); | |
699 | if (ext->flags != 0) { | |
700 | dev_warn(DEV, "changing resync lce: %d[%u;%02lx]" | |
701 | " -> %d[%u;00]\n", | |
702 | ext->lce.lc_number, ext->rs_left, | |
703 | ext->flags, enr, rs_left); | |
704 | ext->flags = 0; | |
705 | } | |
706 | if (ext->rs_failed) { | |
707 | dev_warn(DEV, "Kicking resync_lru element enr=%u " | |
708 | "out with rs_failed=%d\n", | |
709 | ext->lce.lc_number, ext->rs_failed); | |
b411b363 PR |
710 | } |
711 | ext->rs_left = rs_left; | |
712 | ext->rs_failed = success ? 0 : count; | |
46a15bc3 LE |
713 | /* we don't keep a persistent log of the resync lru, |
714 | * we can commit any change right away. */ | |
715 | lc_committed(mdev->resync); | |
b411b363 PR |
716 | } |
717 | lc_put(mdev->resync, &ext->lce); | |
718 | /* no race, we are within the al_lock! */ | |
719 | ||
720 | if (ext->rs_left == ext->rs_failed) { | |
721 | ext->rs_failed = 0; | |
722 | ||
723 | udw = kmalloc(sizeof(*udw), GFP_ATOMIC); | |
724 | if (udw) { | |
725 | udw->enr = ext->lce.lc_number; | |
726 | udw->w.cb = w_update_odbm; | |
a21e9298 | 727 | udw->w.mdev = mdev; |
d5b27b01 | 728 | drbd_queue_work_front(&mdev->tconn->sender_work, &udw->w); |
b411b363 PR |
729 | } else { |
730 | dev_warn(DEV, "Could not kmalloc an udw\n"); | |
b411b363 PR |
731 | } |
732 | } | |
733 | } else { | |
734 | dev_err(DEV, "lc_get() failed! locked=%d/%d flags=%lu\n", | |
735 | mdev->resync_locked, | |
736 | mdev->resync->nr_elements, | |
737 | mdev->resync->flags); | |
738 | } | |
739 | } | |
740 | ||
c6ea14df LE |
741 | void drbd_advance_rs_marks(struct drbd_conf *mdev, unsigned long still_to_go) |
742 | { | |
743 | unsigned long now = jiffies; | |
744 | unsigned long last = mdev->rs_mark_time[mdev->rs_last_mark]; | |
745 | int next = (mdev->rs_last_mark + 1) % DRBD_SYNC_MARKS; | |
746 | if (time_after_eq(now, last + DRBD_SYNC_MARK_STEP)) { | |
747 | if (mdev->rs_mark_left[mdev->rs_last_mark] != still_to_go && | |
748 | mdev->state.conn != C_PAUSED_SYNC_T && | |
749 | mdev->state.conn != C_PAUSED_SYNC_S) { | |
750 | mdev->rs_mark_time[next] = now; | |
751 | mdev->rs_mark_left[next] = still_to_go; | |
752 | mdev->rs_last_mark = next; | |
753 | } | |
754 | } | |
755 | } | |
756 | ||
b411b363 PR |
757 | /* clear the bit corresponding to the piece of storage in question: |
758 | * size byte of data starting from sector. Only clear a bits of the affected | |
759 | * one ore more _aligned_ BM_BLOCK_SIZE blocks. | |
760 | * | |
761 | * called by worker on C_SYNC_TARGET and receiver on SyncSource. | |
762 | * | |
763 | */ | |
764 | void __drbd_set_in_sync(struct drbd_conf *mdev, sector_t sector, int size, | |
765 | const char *file, const unsigned int line) | |
766 | { | |
767 | /* Is called from worker and receiver context _only_ */ | |
768 | unsigned long sbnr, ebnr, lbnr; | |
769 | unsigned long count = 0; | |
770 | sector_t esector, nr_sectors; | |
771 | int wake_up = 0; | |
772 | unsigned long flags; | |
773 | ||
c670a398 | 774 | if (size <= 0 || !IS_ALIGNED(size, 512) || size > DRBD_MAX_BIO_SIZE) { |
b411b363 PR |
775 | dev_err(DEV, "drbd_set_in_sync: sector=%llus size=%d nonsense!\n", |
776 | (unsigned long long)sector, size); | |
777 | return; | |
778 | } | |
518a4d53 PR |
779 | |
780 | if (!get_ldev(mdev)) | |
781 | return; /* no disk, no metadata, no bitmap to clear bits in */ | |
782 | ||
b411b363 PR |
783 | nr_sectors = drbd_get_capacity(mdev->this_bdev); |
784 | esector = sector + (size >> 9) - 1; | |
785 | ||
841ce241 | 786 | if (!expect(sector < nr_sectors)) |
518a4d53 | 787 | goto out; |
841ce241 AG |
788 | if (!expect(esector < nr_sectors)) |
789 | esector = nr_sectors - 1; | |
b411b363 PR |
790 | |
791 | lbnr = BM_SECT_TO_BIT(nr_sectors-1); | |
792 | ||
793 | /* we clear it (in sync). | |
794 | * round up start sector, round down end sector. we make sure we only | |
795 | * clear full, aligned, BM_BLOCK_SIZE (4K) blocks */ | |
796 | if (unlikely(esector < BM_SECT_PER_BIT-1)) | |
518a4d53 | 797 | goto out; |
b411b363 PR |
798 | if (unlikely(esector == (nr_sectors-1))) |
799 | ebnr = lbnr; | |
800 | else | |
801 | ebnr = BM_SECT_TO_BIT(esector - (BM_SECT_PER_BIT-1)); | |
802 | sbnr = BM_SECT_TO_BIT(sector + BM_SECT_PER_BIT-1); | |
803 | ||
b411b363 | 804 | if (sbnr > ebnr) |
518a4d53 | 805 | goto out; |
b411b363 PR |
806 | |
807 | /* | |
808 | * ok, (capacity & 7) != 0 sometimes, but who cares... | |
809 | * we count rs_{total,left} in bits, not sectors. | |
810 | */ | |
b411b363 | 811 | count = drbd_bm_clear_bits(mdev, sbnr, ebnr); |
518a4d53 | 812 | if (count) { |
c6ea14df | 813 | drbd_advance_rs_marks(mdev, drbd_bm_total_weight(mdev)); |
1d7734a0 | 814 | spin_lock_irqsave(&mdev->al_lock, flags); |
81e84650 | 815 | drbd_try_clear_on_disk_bm(mdev, sector, count, true); |
1d7734a0 LE |
816 | spin_unlock_irqrestore(&mdev->al_lock, flags); |
817 | ||
b411b363 PR |
818 | /* just wake_up unconditional now, various lc_chaged(), |
819 | * lc_put() in drbd_try_clear_on_disk_bm(). */ | |
820 | wake_up = 1; | |
821 | } | |
518a4d53 PR |
822 | out: |
823 | put_ldev(mdev); | |
b411b363 PR |
824 | if (wake_up) |
825 | wake_up(&mdev->al_wait); | |
826 | } | |
827 | ||
828 | /* | |
829 | * this is intended to set one request worth of data out of sync. | |
830 | * affects at least 1 bit, | |
1816a2b4 | 831 | * and at most 1+DRBD_MAX_BIO_SIZE/BM_BLOCK_SIZE bits. |
b411b363 PR |
832 | * |
833 | * called by tl_clear and drbd_send_dblock (==drbd_make_request). | |
834 | * so this can be _any_ process. | |
835 | */ | |
73a01a18 | 836 | int __drbd_set_out_of_sync(struct drbd_conf *mdev, sector_t sector, int size, |
b411b363 PR |
837 | const char *file, const unsigned int line) |
838 | { | |
376694a0 | 839 | unsigned long sbnr, ebnr, flags; |
b411b363 | 840 | sector_t esector, nr_sectors; |
73a01a18 | 841 | unsigned int enr, count = 0; |
b411b363 PR |
842 | struct lc_element *e; |
843 | ||
81a3537a LE |
844 | /* this should be an empty REQ_FLUSH */ |
845 | if (size == 0) | |
846 | return 0; | |
847 | ||
848 | if (size < 0 || !IS_ALIGNED(size, 512) || size > DRBD_MAX_BIO_SIZE) { | |
b411b363 PR |
849 | dev_err(DEV, "sector: %llus, size: %d\n", |
850 | (unsigned long long)sector, size); | |
73a01a18 | 851 | return 0; |
b411b363 PR |
852 | } |
853 | ||
854 | if (!get_ldev(mdev)) | |
73a01a18 | 855 | return 0; /* no disk, no metadata, no bitmap to set bits in */ |
b411b363 PR |
856 | |
857 | nr_sectors = drbd_get_capacity(mdev->this_bdev); | |
858 | esector = sector + (size >> 9) - 1; | |
859 | ||
841ce241 | 860 | if (!expect(sector < nr_sectors)) |
b411b363 | 861 | goto out; |
841ce241 AG |
862 | if (!expect(esector < nr_sectors)) |
863 | esector = nr_sectors - 1; | |
b411b363 | 864 | |
b411b363 PR |
865 | /* we set it out of sync, |
866 | * we do not need to round anything here */ | |
867 | sbnr = BM_SECT_TO_BIT(sector); | |
868 | ebnr = BM_SECT_TO_BIT(esector); | |
869 | ||
b411b363 PR |
870 | /* ok, (capacity & 7) != 0 sometimes, but who cares... |
871 | * we count rs_{total,left} in bits, not sectors. */ | |
872 | spin_lock_irqsave(&mdev->al_lock, flags); | |
873 | count = drbd_bm_set_bits(mdev, sbnr, ebnr); | |
874 | ||
875 | enr = BM_SECT_TO_EXT(sector); | |
876 | e = lc_find(mdev->resync, enr); | |
877 | if (e) | |
878 | lc_entry(e, struct bm_extent, lce)->rs_left += count; | |
879 | spin_unlock_irqrestore(&mdev->al_lock, flags); | |
880 | ||
881 | out: | |
882 | put_ldev(mdev); | |
73a01a18 PR |
883 | |
884 | return count; | |
b411b363 PR |
885 | } |
886 | ||
887 | static | |
888 | struct bm_extent *_bme_get(struct drbd_conf *mdev, unsigned int enr) | |
889 | { | |
890 | struct lc_element *e; | |
891 | struct bm_extent *bm_ext; | |
892 | int wakeup = 0; | |
893 | unsigned long rs_flags; | |
894 | ||
895 | spin_lock_irq(&mdev->al_lock); | |
896 | if (mdev->resync_locked > mdev->resync->nr_elements/2) { | |
897 | spin_unlock_irq(&mdev->al_lock); | |
898 | return NULL; | |
899 | } | |
900 | e = lc_get(mdev->resync, enr); | |
901 | bm_ext = e ? lc_entry(e, struct bm_extent, lce) : NULL; | |
902 | if (bm_ext) { | |
903 | if (bm_ext->lce.lc_number != enr) { | |
904 | bm_ext->rs_left = drbd_bm_e_weight(mdev, enr); | |
905 | bm_ext->rs_failed = 0; | |
46a15bc3 | 906 | lc_committed(mdev->resync); |
b411b363 PR |
907 | wakeup = 1; |
908 | } | |
909 | if (bm_ext->lce.refcnt == 1) | |
910 | mdev->resync_locked++; | |
911 | set_bit(BME_NO_WRITES, &bm_ext->flags); | |
912 | } | |
913 | rs_flags = mdev->resync->flags; | |
914 | spin_unlock_irq(&mdev->al_lock); | |
915 | if (wakeup) | |
916 | wake_up(&mdev->al_wait); | |
917 | ||
918 | if (!bm_ext) { | |
919 | if (rs_flags & LC_STARVING) | |
920 | dev_warn(DEV, "Have to wait for element" | |
921 | " (resync LRU too small?)\n"); | |
46a15bc3 | 922 | BUG_ON(rs_flags & LC_LOCKED); |
b411b363 PR |
923 | } |
924 | ||
925 | return bm_ext; | |
926 | } | |
927 | ||
928 | static int _is_in_al(struct drbd_conf *mdev, unsigned int enr) | |
929 | { | |
46a15bc3 | 930 | int rv; |
b411b363 PR |
931 | |
932 | spin_lock_irq(&mdev->al_lock); | |
46a15bc3 | 933 | rv = lc_is_used(mdev->act_log, enr); |
b411b363 PR |
934 | spin_unlock_irq(&mdev->al_lock); |
935 | ||
b411b363 PR |
936 | return rv; |
937 | } | |
938 | ||
939 | /** | |
940 | * drbd_rs_begin_io() - Gets an extent in the resync LRU cache and sets it to BME_LOCKED | |
941 | * @mdev: DRBD device. | |
942 | * @sector: The sector number. | |
943 | * | |
80a40e43 | 944 | * This functions sleeps on al_wait. Returns 0 on success, -EINTR if interrupted. |
b411b363 PR |
945 | */ |
946 | int drbd_rs_begin_io(struct drbd_conf *mdev, sector_t sector) | |
947 | { | |
948 | unsigned int enr = BM_SECT_TO_EXT(sector); | |
949 | struct bm_extent *bm_ext; | |
950 | int i, sig; | |
f91ab628 PR |
951 | int sa = 200; /* Step aside 200 times, then grab the extent and let app-IO wait. |
952 | 200 times -> 20 seconds. */ | |
b411b363 | 953 | |
f91ab628 | 954 | retry: |
b411b363 PR |
955 | sig = wait_event_interruptible(mdev->al_wait, |
956 | (bm_ext = _bme_get(mdev, enr))); | |
957 | if (sig) | |
80a40e43 | 958 | return -EINTR; |
b411b363 PR |
959 | |
960 | if (test_bit(BME_LOCKED, &bm_ext->flags)) | |
80a40e43 | 961 | return 0; |
b411b363 PR |
962 | |
963 | for (i = 0; i < AL_EXT_PER_BM_SECT; i++) { | |
964 | sig = wait_event_interruptible(mdev->al_wait, | |
f91ab628 | 965 | !_is_in_al(mdev, enr * AL_EXT_PER_BM_SECT + i) || |
c507f46f | 966 | test_bit(BME_PRIORITY, &bm_ext->flags)); |
f91ab628 PR |
967 | |
968 | if (sig || (test_bit(BME_PRIORITY, &bm_ext->flags) && sa)) { | |
b411b363 PR |
969 | spin_lock_irq(&mdev->al_lock); |
970 | if (lc_put(mdev->resync, &bm_ext->lce) == 0) { | |
f91ab628 | 971 | bm_ext->flags = 0; /* clears BME_NO_WRITES and eventually BME_PRIORITY */ |
b411b363 PR |
972 | mdev->resync_locked--; |
973 | wake_up(&mdev->al_wait); | |
974 | } | |
975 | spin_unlock_irq(&mdev->al_lock); | |
f91ab628 PR |
976 | if (sig) |
977 | return -EINTR; | |
978 | if (schedule_timeout_interruptible(HZ/10)) | |
979 | return -EINTR; | |
c507f46f PR |
980 | if (sa && --sa == 0) |
981 | dev_warn(DEV,"drbd_rs_begin_io() stepped aside for 20sec." | |
982 | "Resync stalled?\n"); | |
f91ab628 | 983 | goto retry; |
b411b363 PR |
984 | } |
985 | } | |
b411b363 | 986 | set_bit(BME_LOCKED, &bm_ext->flags); |
80a40e43 | 987 | return 0; |
b411b363 PR |
988 | } |
989 | ||
990 | /** | |
991 | * drbd_try_rs_begin_io() - Gets an extent in the resync LRU cache, does not sleep | |
992 | * @mdev: DRBD device. | |
993 | * @sector: The sector number. | |
994 | * | |
995 | * Gets an extent in the resync LRU cache, sets it to BME_NO_WRITES, then | |
996 | * tries to set it to BME_LOCKED. Returns 0 upon success, and -EAGAIN | |
997 | * if there is still application IO going on in this area. | |
998 | */ | |
999 | int drbd_try_rs_begin_io(struct drbd_conf *mdev, sector_t sector) | |
1000 | { | |
1001 | unsigned int enr = BM_SECT_TO_EXT(sector); | |
1002 | const unsigned int al_enr = enr*AL_EXT_PER_BM_SECT; | |
1003 | struct lc_element *e; | |
1004 | struct bm_extent *bm_ext; | |
1005 | int i; | |
1006 | ||
b411b363 PR |
1007 | spin_lock_irq(&mdev->al_lock); |
1008 | if (mdev->resync_wenr != LC_FREE && mdev->resync_wenr != enr) { | |
1009 | /* in case you have very heavy scattered io, it may | |
1010 | * stall the syncer undefined if we give up the ref count | |
1011 | * when we try again and requeue. | |
1012 | * | |
1013 | * if we don't give up the refcount, but the next time | |
1014 | * we are scheduled this extent has been "synced" by new | |
1015 | * application writes, we'd miss the lc_put on the | |
1016 | * extent we keep the refcount on. | |
1017 | * so we remembered which extent we had to try again, and | |
1018 | * if the next requested one is something else, we do | |
1019 | * the lc_put here... | |
1020 | * we also have to wake_up | |
1021 | */ | |
b411b363 PR |
1022 | e = lc_find(mdev->resync, mdev->resync_wenr); |
1023 | bm_ext = e ? lc_entry(e, struct bm_extent, lce) : NULL; | |
1024 | if (bm_ext) { | |
1025 | D_ASSERT(!test_bit(BME_LOCKED, &bm_ext->flags)); | |
1026 | D_ASSERT(test_bit(BME_NO_WRITES, &bm_ext->flags)); | |
1027 | clear_bit(BME_NO_WRITES, &bm_ext->flags); | |
1028 | mdev->resync_wenr = LC_FREE; | |
1029 | if (lc_put(mdev->resync, &bm_ext->lce) == 0) | |
1030 | mdev->resync_locked--; | |
1031 | wake_up(&mdev->al_wait); | |
1032 | } else { | |
1033 | dev_alert(DEV, "LOGIC BUG\n"); | |
1034 | } | |
1035 | } | |
1036 | /* TRY. */ | |
1037 | e = lc_try_get(mdev->resync, enr); | |
1038 | bm_ext = e ? lc_entry(e, struct bm_extent, lce) : NULL; | |
1039 | if (bm_ext) { | |
1040 | if (test_bit(BME_LOCKED, &bm_ext->flags)) | |
1041 | goto proceed; | |
1042 | if (!test_and_set_bit(BME_NO_WRITES, &bm_ext->flags)) { | |
1043 | mdev->resync_locked++; | |
1044 | } else { | |
1045 | /* we did set the BME_NO_WRITES, | |
1046 | * but then could not set BME_LOCKED, | |
1047 | * so we tried again. | |
1048 | * drop the extra reference. */ | |
b411b363 PR |
1049 | bm_ext->lce.refcnt--; |
1050 | D_ASSERT(bm_ext->lce.refcnt > 0); | |
1051 | } | |
1052 | goto check_al; | |
1053 | } else { | |
1054 | /* do we rather want to try later? */ | |
6a0afdf5 | 1055 | if (mdev->resync_locked > mdev->resync->nr_elements-3) |
b411b363 | 1056 | goto try_again; |
b411b363 PR |
1057 | /* Do or do not. There is no try. -- Yoda */ |
1058 | e = lc_get(mdev->resync, enr); | |
1059 | bm_ext = e ? lc_entry(e, struct bm_extent, lce) : NULL; | |
1060 | if (!bm_ext) { | |
1061 | const unsigned long rs_flags = mdev->resync->flags; | |
1062 | if (rs_flags & LC_STARVING) | |
1063 | dev_warn(DEV, "Have to wait for element" | |
1064 | " (resync LRU too small?)\n"); | |
46a15bc3 | 1065 | BUG_ON(rs_flags & LC_LOCKED); |
b411b363 PR |
1066 | goto try_again; |
1067 | } | |
1068 | if (bm_ext->lce.lc_number != enr) { | |
1069 | bm_ext->rs_left = drbd_bm_e_weight(mdev, enr); | |
1070 | bm_ext->rs_failed = 0; | |
46a15bc3 | 1071 | lc_committed(mdev->resync); |
b411b363 PR |
1072 | wake_up(&mdev->al_wait); |
1073 | D_ASSERT(test_bit(BME_LOCKED, &bm_ext->flags) == 0); | |
1074 | } | |
1075 | set_bit(BME_NO_WRITES, &bm_ext->flags); | |
1076 | D_ASSERT(bm_ext->lce.refcnt == 1); | |
1077 | mdev->resync_locked++; | |
1078 | goto check_al; | |
1079 | } | |
1080 | check_al: | |
b411b363 | 1081 | for (i = 0; i < AL_EXT_PER_BM_SECT; i++) { |
b411b363 PR |
1082 | if (lc_is_used(mdev->act_log, al_enr+i)) |
1083 | goto try_again; | |
1084 | } | |
1085 | set_bit(BME_LOCKED, &bm_ext->flags); | |
1086 | proceed: | |
1087 | mdev->resync_wenr = LC_FREE; | |
1088 | spin_unlock_irq(&mdev->al_lock); | |
1089 | return 0; | |
1090 | ||
1091 | try_again: | |
b411b363 PR |
1092 | if (bm_ext) |
1093 | mdev->resync_wenr = enr; | |
1094 | spin_unlock_irq(&mdev->al_lock); | |
1095 | return -EAGAIN; | |
1096 | } | |
1097 | ||
1098 | void drbd_rs_complete_io(struct drbd_conf *mdev, sector_t sector) | |
1099 | { | |
1100 | unsigned int enr = BM_SECT_TO_EXT(sector); | |
1101 | struct lc_element *e; | |
1102 | struct bm_extent *bm_ext; | |
1103 | unsigned long flags; | |
1104 | ||
b411b363 PR |
1105 | spin_lock_irqsave(&mdev->al_lock, flags); |
1106 | e = lc_find(mdev->resync, enr); | |
1107 | bm_ext = e ? lc_entry(e, struct bm_extent, lce) : NULL; | |
1108 | if (!bm_ext) { | |
1109 | spin_unlock_irqrestore(&mdev->al_lock, flags); | |
1110 | if (__ratelimit(&drbd_ratelimit_state)) | |
1111 | dev_err(DEV, "drbd_rs_complete_io() called, but extent not found\n"); | |
1112 | return; | |
1113 | } | |
1114 | ||
1115 | if (bm_ext->lce.refcnt == 0) { | |
1116 | spin_unlock_irqrestore(&mdev->al_lock, flags); | |
1117 | dev_err(DEV, "drbd_rs_complete_io(,%llu [=%u]) called, " | |
1118 | "but refcnt is 0!?\n", | |
1119 | (unsigned long long)sector, enr); | |
1120 | return; | |
1121 | } | |
1122 | ||
1123 | if (lc_put(mdev->resync, &bm_ext->lce) == 0) { | |
e3555d85 | 1124 | bm_ext->flags = 0; /* clear BME_LOCKED, BME_NO_WRITES and BME_PRIORITY */ |
b411b363 PR |
1125 | mdev->resync_locked--; |
1126 | wake_up(&mdev->al_wait); | |
1127 | } | |
1128 | ||
1129 | spin_unlock_irqrestore(&mdev->al_lock, flags); | |
1130 | } | |
1131 | ||
1132 | /** | |
1133 | * drbd_rs_cancel_all() - Removes all extents from the resync LRU (even BME_LOCKED) | |
1134 | * @mdev: DRBD device. | |
1135 | */ | |
1136 | void drbd_rs_cancel_all(struct drbd_conf *mdev) | |
1137 | { | |
b411b363 PR |
1138 | spin_lock_irq(&mdev->al_lock); |
1139 | ||
1140 | if (get_ldev_if_state(mdev, D_FAILED)) { /* Makes sure ->resync is there. */ | |
1141 | lc_reset(mdev->resync); | |
1142 | put_ldev(mdev); | |
1143 | } | |
1144 | mdev->resync_locked = 0; | |
1145 | mdev->resync_wenr = LC_FREE; | |
1146 | spin_unlock_irq(&mdev->al_lock); | |
1147 | wake_up(&mdev->al_wait); | |
1148 | } | |
1149 | ||
1150 | /** | |
1151 | * drbd_rs_del_all() - Gracefully remove all extents from the resync LRU | |
1152 | * @mdev: DRBD device. | |
1153 | * | |
1154 | * Returns 0 upon success, -EAGAIN if at least one reference count was | |
1155 | * not zero. | |
1156 | */ | |
1157 | int drbd_rs_del_all(struct drbd_conf *mdev) | |
1158 | { | |
1159 | struct lc_element *e; | |
1160 | struct bm_extent *bm_ext; | |
1161 | int i; | |
1162 | ||
b411b363 PR |
1163 | spin_lock_irq(&mdev->al_lock); |
1164 | ||
1165 | if (get_ldev_if_state(mdev, D_FAILED)) { | |
1166 | /* ok, ->resync is there. */ | |
1167 | for (i = 0; i < mdev->resync->nr_elements; i++) { | |
1168 | e = lc_element_by_index(mdev->resync, i); | |
b2b163dd | 1169 | bm_ext = lc_entry(e, struct bm_extent, lce); |
b411b363 PR |
1170 | if (bm_ext->lce.lc_number == LC_FREE) |
1171 | continue; | |
1172 | if (bm_ext->lce.lc_number == mdev->resync_wenr) { | |
1173 | dev_info(DEV, "dropping %u in drbd_rs_del_all, apparently" | |
1174 | " got 'synced' by application io\n", | |
1175 | mdev->resync_wenr); | |
1176 | D_ASSERT(!test_bit(BME_LOCKED, &bm_ext->flags)); | |
1177 | D_ASSERT(test_bit(BME_NO_WRITES, &bm_ext->flags)); | |
1178 | clear_bit(BME_NO_WRITES, &bm_ext->flags); | |
1179 | mdev->resync_wenr = LC_FREE; | |
1180 | lc_put(mdev->resync, &bm_ext->lce); | |
1181 | } | |
1182 | if (bm_ext->lce.refcnt != 0) { | |
1183 | dev_info(DEV, "Retrying drbd_rs_del_all() later. " | |
1184 | "refcnt=%d\n", bm_ext->lce.refcnt); | |
1185 | put_ldev(mdev); | |
1186 | spin_unlock_irq(&mdev->al_lock); | |
1187 | return -EAGAIN; | |
1188 | } | |
1189 | D_ASSERT(!test_bit(BME_LOCKED, &bm_ext->flags)); | |
1190 | D_ASSERT(!test_bit(BME_NO_WRITES, &bm_ext->flags)); | |
1191 | lc_del(mdev->resync, &bm_ext->lce); | |
1192 | } | |
1193 | D_ASSERT(mdev->resync->used == 0); | |
1194 | put_ldev(mdev); | |
1195 | } | |
1196 | spin_unlock_irq(&mdev->al_lock); | |
a6a7d4f0 | 1197 | wake_up(&mdev->al_wait); |
b411b363 PR |
1198 | |
1199 | return 0; | |
1200 | } | |
1201 | ||
1202 | /** | |
1203 | * drbd_rs_failed_io() - Record information on a failure to resync the specified blocks | |
1204 | * @mdev: DRBD device. | |
1205 | * @sector: The sector number. | |
1206 | * @size: Size of failed IO operation, in byte. | |
1207 | */ | |
1208 | void drbd_rs_failed_io(struct drbd_conf *mdev, sector_t sector, int size) | |
1209 | { | |
1210 | /* Is called from worker and receiver context _only_ */ | |
1211 | unsigned long sbnr, ebnr, lbnr; | |
1212 | unsigned long count; | |
1213 | sector_t esector, nr_sectors; | |
1214 | int wake_up = 0; | |
1215 | ||
c670a398 | 1216 | if (size <= 0 || !IS_ALIGNED(size, 512) || size > DRBD_MAX_BIO_SIZE) { |
b411b363 PR |
1217 | dev_err(DEV, "drbd_rs_failed_io: sector=%llus size=%d nonsense!\n", |
1218 | (unsigned long long)sector, size); | |
1219 | return; | |
1220 | } | |
1221 | nr_sectors = drbd_get_capacity(mdev->this_bdev); | |
1222 | esector = sector + (size >> 9) - 1; | |
1223 | ||
841ce241 AG |
1224 | if (!expect(sector < nr_sectors)) |
1225 | return; | |
1226 | if (!expect(esector < nr_sectors)) | |
1227 | esector = nr_sectors - 1; | |
b411b363 PR |
1228 | |
1229 | lbnr = BM_SECT_TO_BIT(nr_sectors-1); | |
1230 | ||
1231 | /* | |
1232 | * round up start sector, round down end sector. we make sure we only | |
1233 | * handle full, aligned, BM_BLOCK_SIZE (4K) blocks */ | |
1234 | if (unlikely(esector < BM_SECT_PER_BIT-1)) | |
1235 | return; | |
1236 | if (unlikely(esector == (nr_sectors-1))) | |
1237 | ebnr = lbnr; | |
1238 | else | |
1239 | ebnr = BM_SECT_TO_BIT(esector - (BM_SECT_PER_BIT-1)); | |
1240 | sbnr = BM_SECT_TO_BIT(sector + BM_SECT_PER_BIT-1); | |
1241 | ||
1242 | if (sbnr > ebnr) | |
1243 | return; | |
1244 | ||
1245 | /* | |
1246 | * ok, (capacity & 7) != 0 sometimes, but who cares... | |
1247 | * we count rs_{total,left} in bits, not sectors. | |
1248 | */ | |
1249 | spin_lock_irq(&mdev->al_lock); | |
1250 | count = drbd_bm_count_bits(mdev, sbnr, ebnr); | |
1251 | if (count) { | |
1252 | mdev->rs_failed += count; | |
1253 | ||
1254 | if (get_ldev(mdev)) { | |
81e84650 | 1255 | drbd_try_clear_on_disk_bm(mdev, sector, count, false); |
b411b363 PR |
1256 | put_ldev(mdev); |
1257 | } | |
1258 | ||
1259 | /* just wake_up unconditional now, various lc_chaged(), | |
1260 | * lc_put() in drbd_try_clear_on_disk_bm(). */ | |
1261 | wake_up = 1; | |
1262 | } | |
1263 | spin_unlock_irq(&mdev->al_lock); | |
1264 | if (wake_up) | |
1265 | wake_up(&mdev->al_wait); | |
1266 | } |