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 | 32 | |
85f103d8 LE |
33 | |
34 | enum al_transaction_types { | |
35 | AL_TR_UPDATE = 0, | |
36 | AL_TR_INITIALIZED = 0xffff | |
37 | }; | |
7ad651b5 LE |
38 | /* all fields on disc in big endian */ |
39 | struct __packed al_transaction_on_disk { | |
40 | /* don't we all like magic */ | |
41 | __be32 magic; | |
42 | ||
43 | /* to identify the most recent transaction block | |
44 | * in the on disk ring buffer */ | |
45 | __be32 tr_number; | |
46 | ||
47 | /* checksum on the full 4k block, with this field set to 0. */ | |
48 | __be32 crc32c; | |
49 | ||
50 | /* type of transaction, special transaction types like: | |
85f103d8 LE |
51 | * purge-all, set-all-idle, set-all-active, ... to-be-defined |
52 | * see also enum al_transaction_types */ | |
7ad651b5 LE |
53 | __be16 transaction_type; |
54 | ||
55 | /* we currently allow only a few thousand extents, | |
56 | * so 16bit will be enough for the slot number. */ | |
57 | ||
58 | /* how many updates in this transaction */ | |
59 | __be16 n_updates; | |
60 | ||
61 | /* maximum slot number, "al-extents" in drbd.conf speak. | |
62 | * Having this in each transaction should make reconfiguration | |
63 | * of that parameter easier. */ | |
64 | __be16 context_size; | |
65 | ||
66 | /* slot number the context starts with */ | |
67 | __be16 context_start_slot_nr; | |
68 | ||
69 | /* Some reserved bytes. Expected usage is a 64bit counter of | |
70 | * sectors-written since device creation, and other data generation tag | |
71 | * supporting usage */ | |
72 | __be32 __reserved[4]; | |
73 | ||
74 | /* --- 36 byte used --- */ | |
75 | ||
76 | /* Reserve space for up to AL_UPDATES_PER_TRANSACTION changes | |
77 | * in one transaction, then use the remaining byte in the 4k block for | |
78 | * context information. "Flexible" number of updates per transaction | |
79 | * does not help, as we have to account for the case when all update | |
80 | * slots are used anyways, so it would only complicate code without | |
81 | * additional benefit. | |
82 | */ | |
83 | __be16 update_slot_nr[AL_UPDATES_PER_TRANSACTION]; | |
84 | ||
85 | /* but the extent number is 32bit, which at an extent size of 4 MiB | |
86 | * allows to cover device sizes of up to 2**54 Byte (16 PiB) */ | |
87 | __be32 update_extent_nr[AL_UPDATES_PER_TRANSACTION]; | |
88 | ||
89 | /* --- 420 bytes used (36 + 64*6) --- */ | |
90 | ||
91 | /* 4096 - 420 = 3676 = 919 * 4 */ | |
92 | __be32 context[AL_CONTEXT_PER_TRANSACTION]; | |
b411b363 PR |
93 | }; |
94 | ||
e37d2438 | 95 | void *drbd_md_get_buffer(struct drbd_device *device, const char *intent) |
cdfda633 PR |
96 | { |
97 | int r; | |
98 | ||
b30ab791 | 99 | wait_event(device->misc_wait, |
e37d2438 | 100 | (r = atomic_cmpxchg(&device->md_io.in_use, 0, 1)) == 0 || |
b30ab791 | 101 | device->state.disk <= D_FAILED); |
cdfda633 | 102 | |
e37d2438 LE |
103 | if (r) |
104 | return NULL; | |
105 | ||
106 | device->md_io.current_use = intent; | |
107 | device->md_io.start_jif = jiffies; | |
108 | device->md_io.submit_jif = device->md_io.start_jif - 1; | |
109 | return page_address(device->md_io.page); | |
cdfda633 PR |
110 | } |
111 | ||
b30ab791 | 112 | void drbd_md_put_buffer(struct drbd_device *device) |
cdfda633 | 113 | { |
e37d2438 | 114 | if (atomic_dec_and_test(&device->md_io.in_use)) |
b30ab791 | 115 | wake_up(&device->misc_wait); |
cdfda633 PR |
116 | } |
117 | ||
b30ab791 | 118 | void wait_until_done_or_force_detached(struct drbd_device *device, struct drbd_backing_dev *bdev, |
32db80f6 | 119 | unsigned int *done) |
cdfda633 | 120 | { |
32db80f6 PR |
121 | long dt; |
122 | ||
123 | rcu_read_lock(); | |
124 | dt = rcu_dereference(bdev->disk_conf)->disk_timeout; | |
125 | rcu_read_unlock(); | |
126 | dt = dt * HZ / 10; | |
127 | if (dt == 0) | |
128 | dt = MAX_SCHEDULE_TIMEOUT; | |
129 | ||
b30ab791 AG |
130 | dt = wait_event_timeout(device->misc_wait, |
131 | *done || test_bit(FORCE_DETACH, &device->flags), dt); | |
e34b677d | 132 | if (dt == 0) { |
d0180171 | 133 | drbd_err(device, "meta-data IO operation timed out\n"); |
b30ab791 | 134 | drbd_chk_io_error(device, 1, DRBD_FORCE_DETACH); |
e34b677d | 135 | } |
cdfda633 PR |
136 | } |
137 | ||
b30ab791 | 138 | static int _drbd_md_sync_page_io(struct drbd_device *device, |
b411b363 | 139 | struct drbd_backing_dev *bdev, |
193cb00c | 140 | sector_t sector, int rw) |
b411b363 PR |
141 | { |
142 | struct bio *bio; | |
193cb00c LE |
143 | /* we do all our meta data IO in aligned 4k blocks. */ |
144 | const int size = 4096; | |
ac29f403 | 145 | int err; |
b411b363 | 146 | |
b30ab791 AG |
147 | device->md_io.done = 0; |
148 | device->md_io.error = -ENODEV; | |
b411b363 | 149 | |
b30ab791 | 150 | if ((rw & WRITE) && !test_bit(MD_NO_FUA, &device->flags)) |
86e1e98e | 151 | rw |= REQ_FUA | REQ_FLUSH; |
2ed912e9 | 152 | rw |= REQ_SYNC | REQ_NOIDLE; |
b411b363 | 153 | |
da4a75d2 | 154 | bio = bio_alloc_drbd(GFP_NOIO); |
b411b363 | 155 | bio->bi_bdev = bdev->md_bdev; |
4f024f37 | 156 | bio->bi_iter.bi_sector = sector; |
ac29f403 | 157 | err = -EIO; |
193cb00c | 158 | if (bio_add_page(bio, device->md_io.page, size, 0) != size) |
b411b363 | 159 | goto out; |
e37d2438 | 160 | bio->bi_private = device; |
ed15b795 | 161 | bio->bi_end_io = drbd_md_endio; |
b411b363 PR |
162 | bio->bi_rw = rw; |
163 | ||
b30ab791 | 164 | if (!(rw & WRITE) && device->state.disk == D_DISKLESS && device->ldev == NULL) |
c04ccaa6 LE |
165 | /* special case, drbd_md_read() during drbd_adm_attach(): no get_ldev */ |
166 | ; | |
b30ab791 | 167 | else if (!get_ldev_if_state(device, D_ATTACHING)) { |
ed15b795 | 168 | /* Corresponding put_ldev in drbd_md_endio() */ |
d0180171 | 169 | drbd_err(device, "ASSERT FAILED: get_ldev_if_state() == 1 in _drbd_md_sync_page_io()\n"); |
cdfda633 PR |
170 | err = -ENODEV; |
171 | goto out; | |
172 | } | |
173 | ||
174 | bio_get(bio); /* one bio_put() is in the completion handler */ | |
e37d2438 LE |
175 | atomic_inc(&device->md_io.in_use); /* drbd_md_put_buffer() is in the completion handler */ |
176 | device->md_io.submit_jif = jiffies; | |
b30ab791 | 177 | if (drbd_insert_fault(device, (rw & WRITE) ? DRBD_FAULT_MD_WR : DRBD_FAULT_MD_RD)) |
b411b363 PR |
178 | bio_endio(bio, -EIO); |
179 | else | |
180 | submit_bio(rw, bio); | |
b30ab791 | 181 | wait_until_done_or_force_detached(device, bdev, &device->md_io.done); |
ac29f403 | 182 | if (bio_flagged(bio, BIO_UPTODATE)) |
b30ab791 | 183 | err = device->md_io.error; |
b411b363 | 184 | |
b411b363 PR |
185 | out: |
186 | bio_put(bio); | |
ac29f403 | 187 | return err; |
b411b363 PR |
188 | } |
189 | ||
b30ab791 | 190 | int drbd_md_sync_page_io(struct drbd_device *device, struct drbd_backing_dev *bdev, |
b411b363 PR |
191 | sector_t sector, int rw) |
192 | { | |
3fbf4d21 | 193 | int err; |
e37d2438 | 194 | D_ASSERT(device, atomic_read(&device->md_io.in_use) == 1); |
b411b363 PR |
195 | |
196 | BUG_ON(!bdev->md_bdev); | |
197 | ||
e4d7d6f4 | 198 | dynamic_drbd_dbg(device, "meta_data io: %s [%d]:%s(,%llus,%s) %pS\n", |
7ad651b5 | 199 | current->comm, current->pid, __func__, |
c04ccaa6 LE |
200 | (unsigned long long)sector, (rw & WRITE) ? "WRITE" : "READ", |
201 | (void*)_RET_IP_ ); | |
b411b363 PR |
202 | |
203 | if (sector < drbd_md_first_sector(bdev) || | |
7ad651b5 | 204 | sector + 7 > drbd_md_last_sector(bdev)) |
d0180171 | 205 | drbd_alert(device, "%s [%d]:%s(,%llus,%s) out of range md access!\n", |
b411b363 PR |
206 | current->comm, current->pid, __func__, |
207 | (unsigned long long)sector, (rw & WRITE) ? "WRITE" : "READ"); | |
208 | ||
193cb00c | 209 | err = _drbd_md_sync_page_io(device, bdev, sector, rw); |
3fbf4d21 | 210 | if (err) { |
d0180171 | 211 | drbd_err(device, "drbd_md_sync_page_io(,%llus,%s) failed with error %d\n", |
935be260 | 212 | (unsigned long long)sector, (rw & WRITE) ? "WRITE" : "READ", err); |
b411b363 | 213 | } |
3fbf4d21 | 214 | return err; |
b411b363 PR |
215 | } |
216 | ||
b30ab791 | 217 | static struct bm_extent *find_active_resync_extent(struct drbd_device *device, unsigned int enr) |
b411b363 | 218 | { |
b411b363 | 219 | struct lc_element *tmp; |
b30ab791 | 220 | tmp = lc_find(device->resync, enr/AL_EXT_PER_BM_SECT); |
b411b363 PR |
221 | if (unlikely(tmp != NULL)) { |
222 | struct bm_extent *bm_ext = lc_entry(tmp, struct bm_extent, lce); | |
6c3c4355 LE |
223 | if (test_bit(BME_NO_WRITES, &bm_ext->flags)) |
224 | return bm_ext; | |
b411b363 | 225 | } |
6c3c4355 LE |
226 | return NULL; |
227 | } | |
228 | ||
b30ab791 | 229 | static struct lc_element *_al_get(struct drbd_device *device, unsigned int enr, bool nonblock) |
6c3c4355 LE |
230 | { |
231 | struct lc_element *al_ext; | |
232 | struct bm_extent *bm_ext; | |
233 | int wake; | |
234 | ||
b30ab791 AG |
235 | spin_lock_irq(&device->al_lock); |
236 | bm_ext = find_active_resync_extent(device, enr); | |
6c3c4355 LE |
237 | if (bm_ext) { |
238 | wake = !test_and_set_bit(BME_PRIORITY, &bm_ext->flags); | |
b30ab791 | 239 | spin_unlock_irq(&device->al_lock); |
6c3c4355 | 240 | if (wake) |
b30ab791 | 241 | wake_up(&device->al_wait); |
6c3c4355 LE |
242 | return NULL; |
243 | } | |
244 | if (nonblock) | |
b30ab791 | 245 | al_ext = lc_try_get(device->act_log, enr); |
6c3c4355 | 246 | else |
b30ab791 AG |
247 | al_ext = lc_get(device->act_log, enr); |
248 | spin_unlock_irq(&device->al_lock); | |
b411b363 PR |
249 | return al_ext; |
250 | } | |
251 | ||
b30ab791 | 252 | bool drbd_al_begin_io_fastpath(struct drbd_device *device, struct drbd_interval *i) |
b411b363 | 253 | { |
7726547e LE |
254 | /* for bios crossing activity log extent boundaries, |
255 | * we may need to activate two extents in one go */ | |
e15766e9 | 256 | unsigned first = i->sector >> (AL_EXTENT_SHIFT-9); |
81a3537a | 257 | unsigned last = i->size == 0 ? first : (i->sector + (i->size >> 9) - 1) >> (AL_EXTENT_SHIFT-9); |
7dc1d67f | 258 | |
0b0ba1ef AG |
259 | D_ASSERT(device, (unsigned)(last - first) <= 1); |
260 | D_ASSERT(device, atomic_read(&device->local_cnt) > 0); | |
b5bc8e08 LE |
261 | |
262 | /* FIXME figure out a fast path for bios crossing AL extent boundaries */ | |
263 | if (first != last) | |
264 | return false; | |
265 | ||
b30ab791 | 266 | return _al_get(device, first, true); |
b5bc8e08 | 267 | } |
56392d2f | 268 | |
b30ab791 | 269 | bool drbd_al_begin_io_prepare(struct drbd_device *device, struct drbd_interval *i) |
b5bc8e08 LE |
270 | { |
271 | /* for bios crossing activity log extent boundaries, | |
272 | * we may need to activate two extents in one go */ | |
273 | unsigned first = i->sector >> (AL_EXTENT_SHIFT-9); | |
274 | unsigned last = i->size == 0 ? first : (i->sector + (i->size >> 9) - 1) >> (AL_EXTENT_SHIFT-9); | |
275 | unsigned enr; | |
276 | bool need_transaction = false; | |
b411b363 | 277 | |
0b0ba1ef AG |
278 | D_ASSERT(device, first <= last); |
279 | D_ASSERT(device, atomic_read(&device->local_cnt) > 0); | |
b411b363 | 280 | |
ebfd5d8f LE |
281 | for (enr = first; enr <= last; enr++) { |
282 | struct lc_element *al_ext; | |
b30ab791 AG |
283 | wait_event(device->al_wait, |
284 | (al_ext = _al_get(device, enr, false)) != NULL); | |
ebfd5d8f LE |
285 | if (al_ext->lc_number != enr) |
286 | need_transaction = true; | |
287 | } | |
b5bc8e08 LE |
288 | return need_transaction; |
289 | } | |
290 | ||
4dd726f0 | 291 | static int al_write_transaction(struct drbd_device *device); |
b5bc8e08 | 292 | |
4dd726f0 | 293 | void drbd_al_begin_io_commit(struct drbd_device *device) |
b5bc8e08 LE |
294 | { |
295 | bool locked = false; | |
ebfd5d8f | 296 | |
7dc1d67f LE |
297 | /* Serialize multiple transactions. |
298 | * This uses test_and_set_bit, memory barrier is implicit. | |
299 | */ | |
b30ab791 AG |
300 | wait_event(device->al_wait, |
301 | device->act_log->pending_changes == 0 || | |
302 | (locked = lc_try_lock_for_transaction(device->act_log))); | |
7dc1d67f LE |
303 | |
304 | if (locked) { | |
7ad651b5 LE |
305 | /* Double check: it may have been committed by someone else, |
306 | * while we have been waiting for the lock. */ | |
b30ab791 | 307 | if (device->act_log->pending_changes) { |
9a51ab1c PR |
308 | bool write_al_updates; |
309 | ||
310 | rcu_read_lock(); | |
b30ab791 | 311 | write_al_updates = rcu_dereference(device->ldev->disk_conf)->al_updates; |
9a51ab1c PR |
312 | rcu_read_unlock(); |
313 | ||
b5bc8e08 | 314 | if (write_al_updates) |
4dd726f0 | 315 | al_write_transaction(device); |
b30ab791 | 316 | spin_lock_irq(&device->al_lock); |
7ad651b5 | 317 | /* FIXME |
1b7ab15b | 318 | if (err) |
7ad651b5 LE |
319 | we need an "lc_cancel" here; |
320 | */ | |
b30ab791 AG |
321 | lc_committed(device->act_log); |
322 | spin_unlock_irq(&device->al_lock); | |
7ad651b5 | 323 | } |
b30ab791 AG |
324 | lc_unlock(device->act_log); |
325 | wake_up(&device->al_wait); | |
b411b363 PR |
326 | } |
327 | } | |
328 | ||
b5bc8e08 LE |
329 | /* |
330 | * @delegate: delegate activity log I/O to the worker thread | |
331 | */ | |
4dd726f0 | 332 | void drbd_al_begin_io(struct drbd_device *device, struct drbd_interval *i) |
b5bc8e08 | 333 | { |
b30ab791 | 334 | if (drbd_al_begin_io_prepare(device, i)) |
4dd726f0 | 335 | drbd_al_begin_io_commit(device); |
b5bc8e08 LE |
336 | } |
337 | ||
b30ab791 | 338 | int drbd_al_begin_io_nonblock(struct drbd_device *device, struct drbd_interval *i) |
08a1ddab | 339 | { |
b30ab791 | 340 | struct lru_cache *al = device->act_log; |
08a1ddab LE |
341 | /* for bios crossing activity log extent boundaries, |
342 | * we may need to activate two extents in one go */ | |
343 | unsigned first = i->sector >> (AL_EXTENT_SHIFT-9); | |
344 | unsigned last = i->size == 0 ? first : (i->sector + (i->size >> 9) - 1) >> (AL_EXTENT_SHIFT-9); | |
345 | unsigned nr_al_extents; | |
346 | unsigned available_update_slots; | |
347 | unsigned enr; | |
348 | ||
0b0ba1ef | 349 | D_ASSERT(device, first <= last); |
08a1ddab LE |
350 | |
351 | nr_al_extents = 1 + last - first; /* worst case: all touched extends are cold. */ | |
352 | available_update_slots = min(al->nr_elements - al->used, | |
353 | al->max_pending_changes - al->pending_changes); | |
354 | ||
355 | /* We want all necessary updates for a given request within the same transaction | |
356 | * We could first check how many updates are *actually* needed, | |
357 | * and use that instead of the worst-case nr_al_extents */ | |
f5b90b6b LE |
358 | if (available_update_slots < nr_al_extents) { |
359 | /* Too many activity log extents are currently "hot". | |
360 | * | |
361 | * If we have accumulated pending changes already, | |
362 | * we made progress. | |
363 | * | |
364 | * If we cannot get even a single pending change through, | |
365 | * stop the fast path until we made some progress, | |
366 | * or requests to "cold" extents could be starved. */ | |
367 | if (!al->pending_changes) | |
368 | __set_bit(__LC_STARVING, &device->act_log->flags); | |
369 | return -ENOBUFS; | |
370 | } | |
08a1ddab LE |
371 | |
372 | /* Is resync active in this area? */ | |
373 | for (enr = first; enr <= last; enr++) { | |
374 | struct lc_element *tmp; | |
b30ab791 | 375 | tmp = lc_find(device->resync, enr/AL_EXT_PER_BM_SECT); |
08a1ddab LE |
376 | if (unlikely(tmp != NULL)) { |
377 | struct bm_extent *bm_ext = lc_entry(tmp, struct bm_extent, lce); | |
378 | if (test_bit(BME_NO_WRITES, &bm_ext->flags)) { | |
0b6ef416 | 379 | if (!test_and_set_bit(BME_PRIORITY, &bm_ext->flags)) |
08a1ddab LE |
380 | return -EBUSY; |
381 | return -EWOULDBLOCK; | |
382 | } | |
383 | } | |
384 | } | |
385 | ||
386 | /* Checkout the refcounts. | |
387 | * Given that we checked for available elements and update slots above, | |
388 | * this has to be successful. */ | |
389 | for (enr = first; enr <= last; enr++) { | |
390 | struct lc_element *al_ext; | |
b30ab791 | 391 | al_ext = lc_get_cumulative(device->act_log, enr); |
08a1ddab | 392 | if (!al_ext) |
d0180171 | 393 | drbd_info(device, "LOGIC BUG for enr=%u\n", enr); |
08a1ddab LE |
394 | } |
395 | return 0; | |
396 | } | |
397 | ||
b30ab791 | 398 | void drbd_al_complete_io(struct drbd_device *device, struct drbd_interval *i) |
b411b363 | 399 | { |
e15766e9 LE |
400 | /* for bios crossing activity log extent boundaries, |
401 | * we may need to activate two extents in one go */ | |
402 | unsigned first = i->sector >> (AL_EXTENT_SHIFT-9); | |
81a3537a | 403 | unsigned last = i->size == 0 ? first : (i->sector + (i->size >> 9) - 1) >> (AL_EXTENT_SHIFT-9); |
e15766e9 | 404 | unsigned enr; |
b411b363 PR |
405 | struct lc_element *extent; |
406 | unsigned long flags; | |
407 | ||
0b0ba1ef | 408 | D_ASSERT(device, first <= last); |
b30ab791 | 409 | spin_lock_irqsave(&device->al_lock, flags); |
b411b363 | 410 | |
e15766e9 | 411 | for (enr = first; enr <= last; enr++) { |
b30ab791 | 412 | extent = lc_find(device->act_log, enr); |
e15766e9 | 413 | if (!extent) { |
d0180171 | 414 | drbd_err(device, "al_complete_io() called on inactive extent %u\n", enr); |
e15766e9 LE |
415 | continue; |
416 | } | |
b30ab791 | 417 | lc_put(device->act_log, extent); |
b411b363 | 418 | } |
b30ab791 AG |
419 | spin_unlock_irqrestore(&device->al_lock, flags); |
420 | wake_up(&device->al_wait); | |
b411b363 PR |
421 | } |
422 | ||
19f843aa LE |
423 | #if (PAGE_SHIFT + 3) < (AL_EXTENT_SHIFT - BM_BLOCK_SHIFT) |
424 | /* Currently BM_BLOCK_SHIFT, BM_EXT_SHIFT and AL_EXTENT_SHIFT | |
425 | * are still coupled, or assume too much about their relation. | |
426 | * Code below will not work if this is violated. | |
427 | * Will be cleaned up with some followup patch. | |
428 | */ | |
429 | # error FIXME | |
430 | #endif | |
431 | ||
432 | static unsigned int al_extent_to_bm_page(unsigned int al_enr) | |
433 | { | |
434 | return al_enr >> | |
435 | /* bit to page */ | |
436 | ((PAGE_SHIFT + 3) - | |
437 | /* al extent number to bit */ | |
438 | (AL_EXTENT_SHIFT - BM_BLOCK_SHIFT)); | |
439 | } | |
440 | ||
b30ab791 | 441 | static sector_t al_tr_number_to_on_disk_sector(struct drbd_device *device) |
ae8bf312 | 442 | { |
b30ab791 AG |
443 | const unsigned int stripes = device->ldev->md.al_stripes; |
444 | const unsigned int stripe_size_4kB = device->ldev->md.al_stripe_size_4k; | |
ae8bf312 LE |
445 | |
446 | /* transaction number, modulo on-disk ring buffer wrap around */ | |
b30ab791 | 447 | unsigned int t = device->al_tr_number % (device->ldev->md.al_size_4k); |
ae8bf312 LE |
448 | |
449 | /* ... to aligned 4k on disk block */ | |
450 | t = ((t % stripes) * stripe_size_4kB) + t/stripes; | |
451 | ||
452 | /* ... to 512 byte sector in activity log */ | |
453 | t *= 8; | |
454 | ||
455 | /* ... plus offset to the on disk position */ | |
b30ab791 | 456 | return device->ldev->md.md_offset + device->ldev->md.al_offset + t; |
ae8bf312 LE |
457 | } |
458 | ||
4dd726f0 | 459 | int al_write_transaction(struct drbd_device *device) |
b411b363 | 460 | { |
7ad651b5 LE |
461 | struct al_transaction_on_disk *buffer; |
462 | struct lc_element *e; | |
b411b363 | 463 | sector_t sector; |
7ad651b5 LE |
464 | int i, mx; |
465 | unsigned extent_nr; | |
466 | unsigned crc = 0; | |
1b7ab15b | 467 | int err = 0; |
b411b363 | 468 | |
b30ab791 | 469 | if (!get_ldev(device)) { |
d0180171 | 470 | drbd_err(device, "disk is %s, cannot start al transaction\n", |
b30ab791 | 471 | drbd_disk_str(device->state.disk)); |
1b7ab15b | 472 | return -EIO; |
b411b363 | 473 | } |
b411b363 | 474 | |
6719fb03 | 475 | /* The bitmap write may have failed, causing a state change. */ |
b30ab791 | 476 | if (device->state.disk < D_INCONSISTENT) { |
d0180171 | 477 | drbd_err(device, |
7ad651b5 | 478 | "disk is %s, cannot write al transaction\n", |
b30ab791 AG |
479 | drbd_disk_str(device->state.disk)); |
480 | put_ldev(device); | |
1b7ab15b | 481 | return -EIO; |
6719fb03 LE |
482 | } |
483 | ||
e37d2438 LE |
484 | /* protects md_io_buffer, al_tr_cycle, ... */ |
485 | buffer = drbd_md_get_buffer(device, __func__); | |
cdfda633 | 486 | if (!buffer) { |
d0180171 | 487 | drbd_err(device, "disk failed while waiting for md_io buffer\n"); |
b30ab791 | 488 | put_ldev(device); |
1b7ab15b | 489 | return -ENODEV; |
cdfda633 | 490 | } |
b411b363 | 491 | |
7ad651b5 LE |
492 | memset(buffer, 0, sizeof(*buffer)); |
493 | buffer->magic = cpu_to_be32(DRBD_AL_MAGIC); | |
b30ab791 | 494 | buffer->tr_number = cpu_to_be32(device->al_tr_number); |
b411b363 | 495 | |
7ad651b5 LE |
496 | i = 0; |
497 | ||
498 | /* Even though no one can start to change this list | |
499 | * once we set the LC_LOCKED -- from drbd_al_begin_io(), | |
500 | * lc_try_lock_for_transaction() --, someone may still | |
501 | * be in the process of changing it. */ | |
b30ab791 AG |
502 | spin_lock_irq(&device->al_lock); |
503 | list_for_each_entry(e, &device->act_log->to_be_changed, list) { | |
7ad651b5 LE |
504 | if (i == AL_UPDATES_PER_TRANSACTION) { |
505 | i++; | |
506 | break; | |
507 | } | |
508 | buffer->update_slot_nr[i] = cpu_to_be16(e->lc_index); | |
509 | buffer->update_extent_nr[i] = cpu_to_be32(e->lc_new_number); | |
510 | if (e->lc_number != LC_FREE) | |
b30ab791 | 511 | drbd_bm_mark_for_writeout(device, |
7ad651b5 LE |
512 | al_extent_to_bm_page(e->lc_number)); |
513 | i++; | |
514 | } | |
b30ab791 | 515 | spin_unlock_irq(&device->al_lock); |
7ad651b5 | 516 | BUG_ON(i > AL_UPDATES_PER_TRANSACTION); |
b411b363 | 517 | |
7ad651b5 LE |
518 | buffer->n_updates = cpu_to_be16(i); |
519 | for ( ; i < AL_UPDATES_PER_TRANSACTION; i++) { | |
520 | buffer->update_slot_nr[i] = cpu_to_be16(-1); | |
521 | buffer->update_extent_nr[i] = cpu_to_be32(LC_FREE); | |
522 | } | |
b411b363 | 523 | |
b30ab791 AG |
524 | buffer->context_size = cpu_to_be16(device->act_log->nr_elements); |
525 | buffer->context_start_slot_nr = cpu_to_be16(device->al_tr_cycle); | |
b411b363 | 526 | |
7ad651b5 | 527 | mx = min_t(int, AL_CONTEXT_PER_TRANSACTION, |
b30ab791 | 528 | device->act_log->nr_elements - device->al_tr_cycle); |
b411b363 | 529 | for (i = 0; i < mx; i++) { |
b30ab791 AG |
530 | unsigned idx = device->al_tr_cycle + i; |
531 | extent_nr = lc_element_by_index(device->act_log, idx)->lc_number; | |
7ad651b5 | 532 | buffer->context[i] = cpu_to_be32(extent_nr); |
b411b363 | 533 | } |
7ad651b5 LE |
534 | for (; i < AL_CONTEXT_PER_TRANSACTION; i++) |
535 | buffer->context[i] = cpu_to_be32(LC_FREE); | |
536 | ||
b30ab791 AG |
537 | device->al_tr_cycle += AL_CONTEXT_PER_TRANSACTION; |
538 | if (device->al_tr_cycle >= device->act_log->nr_elements) | |
539 | device->al_tr_cycle = 0; | |
b411b363 | 540 | |
b30ab791 | 541 | sector = al_tr_number_to_on_disk_sector(device); |
b411b363 | 542 | |
7ad651b5 LE |
543 | crc = crc32c(0, buffer, 4096); |
544 | buffer->crc32c = cpu_to_be32(crc); | |
b411b363 | 545 | |
b30ab791 | 546 | if (drbd_bm_write_hinted(device)) |
1b7ab15b | 547 | err = -EIO; |
b5bc8e08 LE |
548 | else { |
549 | bool write_al_updates; | |
550 | rcu_read_lock(); | |
b30ab791 | 551 | write_al_updates = rcu_dereference(device->ldev->disk_conf)->al_updates; |
b5bc8e08 LE |
552 | rcu_read_unlock(); |
553 | if (write_al_updates) { | |
b30ab791 | 554 | if (drbd_md_sync_page_io(device, device->ldev, sector, WRITE)) { |
b5bc8e08 | 555 | err = -EIO; |
b30ab791 | 556 | drbd_chk_io_error(device, 1, DRBD_META_IO_ERROR); |
b5bc8e08 | 557 | } else { |
b30ab791 AG |
558 | device->al_tr_number++; |
559 | device->al_writ_cnt++; | |
b5bc8e08 LE |
560 | } |
561 | } | |
7ad651b5 | 562 | } |
b411b363 | 563 | |
b30ab791 AG |
564 | drbd_md_put_buffer(device); |
565 | put_ldev(device); | |
b411b363 | 566 | |
1b7ab15b PR |
567 | return err; |
568 | } | |
569 | ||
b30ab791 | 570 | static int _try_lc_del(struct drbd_device *device, struct lc_element *al_ext) |
b411b363 PR |
571 | { |
572 | int rv; | |
573 | ||
b30ab791 | 574 | spin_lock_irq(&device->al_lock); |
b411b363 PR |
575 | rv = (al_ext->refcnt == 0); |
576 | if (likely(rv)) | |
b30ab791 AG |
577 | lc_del(device->act_log, al_ext); |
578 | spin_unlock_irq(&device->al_lock); | |
b411b363 PR |
579 | |
580 | return rv; | |
581 | } | |
582 | ||
583 | /** | |
584 | * drbd_al_shrink() - Removes all active extents form the activity log | |
b30ab791 | 585 | * @device: DRBD device. |
b411b363 PR |
586 | * |
587 | * Removes all active extents form the activity log, waiting until | |
588 | * the reference count of each entry dropped to 0 first, of course. | |
589 | * | |
b30ab791 | 590 | * You need to lock device->act_log with lc_try_lock() / lc_unlock() |
b411b363 | 591 | */ |
b30ab791 | 592 | void drbd_al_shrink(struct drbd_device *device) |
b411b363 PR |
593 | { |
594 | struct lc_element *al_ext; | |
595 | int i; | |
596 | ||
0b0ba1ef | 597 | D_ASSERT(device, test_bit(__LC_LOCKED, &device->act_log->flags)); |
b411b363 | 598 | |
b30ab791 AG |
599 | for (i = 0; i < device->act_log->nr_elements; i++) { |
600 | al_ext = lc_element_by_index(device->act_log, i); | |
b411b363 PR |
601 | if (al_ext->lc_number == LC_FREE) |
602 | continue; | |
b30ab791 | 603 | wait_event(device->al_wait, _try_lc_del(device, al_ext)); |
b411b363 PR |
604 | } |
605 | ||
b30ab791 | 606 | wake_up(&device->al_wait); |
b411b363 PR |
607 | } |
608 | ||
b30ab791 | 609 | int drbd_initialize_al(struct drbd_device *device, void *buffer) |
d752b269 PR |
610 | { |
611 | struct al_transaction_on_disk *al = buffer; | |
b30ab791 | 612 | struct drbd_md *md = &device->ldev->md; |
d752b269 PR |
613 | sector_t al_base = md->md_offset + md->al_offset; |
614 | int al_size_4k = md->al_stripes * md->al_stripe_size_4k; | |
615 | int i; | |
616 | ||
617 | memset(al, 0, 4096); | |
618 | al->magic = cpu_to_be32(DRBD_AL_MAGIC); | |
619 | al->transaction_type = cpu_to_be16(AL_TR_INITIALIZED); | |
620 | al->crc32c = cpu_to_be32(crc32c(0, al, 4096)); | |
621 | ||
622 | for (i = 0; i < al_size_4k; i++) { | |
b30ab791 | 623 | int err = drbd_md_sync_page_io(device, device->ldev, al_base + i * 8, WRITE); |
d752b269 PR |
624 | if (err) |
625 | return err; | |
626 | } | |
627 | return 0; | |
628 | } | |
629 | ||
5ab7d2c0 LE |
630 | static const char *drbd_change_sync_fname[] = { |
631 | [RECORD_RS_FAILED] = "drbd_rs_failed_io", | |
632 | [SET_IN_SYNC] = "drbd_set_in_sync", | |
633 | [SET_OUT_OF_SYNC] = "drbd_set_out_of_sync" | |
634 | }; | |
635 | ||
b411b363 PR |
636 | /* ATTENTION. The AL's extents are 4MB each, while the extents in the |
637 | * resync LRU-cache are 16MB each. | |
638 | * The caller of this function has to hold an get_ldev() reference. | |
639 | * | |
5ab7d2c0 LE |
640 | * Adjusts the caching members ->rs_left (success) or ->rs_failed (!success), |
641 | * potentially pulling in (and recounting the corresponding bits) | |
642 | * this resync extent into the resync extent lru cache. | |
643 | * | |
644 | * Returns whether all bits have been cleared for this resync extent, | |
645 | * precisely: (rs_left <= rs_failed) | |
646 | * | |
b411b363 PR |
647 | * TODO will be obsoleted once we have a caching lru of the on disk bitmap |
648 | */ | |
5ab7d2c0 LE |
649 | static bool update_rs_extent(struct drbd_device *device, |
650 | unsigned int enr, int count, | |
651 | enum update_sync_bits_mode mode) | |
b411b363 PR |
652 | { |
653 | struct lc_element *e; | |
b411b363 | 654 | |
0b0ba1ef | 655 | D_ASSERT(device, atomic_read(&device->local_cnt)); |
b411b363 | 656 | |
5ab7d2c0 LE |
657 | /* When setting out-of-sync bits, |
658 | * we don't need it cached (lc_find). | |
659 | * But if it is present in the cache, | |
660 | * we should update the cached bit count. | |
661 | * Otherwise, that extent should be in the resync extent lru cache | |
662 | * already -- or we want to pull it in if necessary -- (lc_get), | |
663 | * then update and check rs_left and rs_failed. */ | |
664 | if (mode == SET_OUT_OF_SYNC) | |
665 | e = lc_find(device->resync, enr); | |
666 | else | |
667 | e = lc_get(device->resync, enr); | |
b411b363 PR |
668 | if (e) { |
669 | struct bm_extent *ext = lc_entry(e, struct bm_extent, lce); | |
670 | if (ext->lce.lc_number == enr) { | |
5ab7d2c0 | 671 | if (mode == SET_IN_SYNC) |
b411b363 | 672 | ext->rs_left -= count; |
5ab7d2c0 LE |
673 | else if (mode == SET_OUT_OF_SYNC) |
674 | ext->rs_left += count; | |
b411b363 PR |
675 | else |
676 | ext->rs_failed += count; | |
677 | if (ext->rs_left < ext->rs_failed) { | |
5ab7d2c0 | 678 | drbd_warn(device, "BAD! enr=%u rs_left=%d " |
975b2979 | 679 | "rs_failed=%d count=%d cstate=%s\n", |
b411b363 | 680 | ext->lce.lc_number, ext->rs_left, |
975b2979 | 681 | ext->rs_failed, count, |
b30ab791 | 682 | drbd_conn_str(device->state.conn)); |
975b2979 PR |
683 | |
684 | /* We don't expect to be able to clear more bits | |
685 | * than have been set when we originally counted | |
686 | * the set bits to cache that value in ext->rs_left. | |
687 | * Whatever the reason (disconnect during resync, | |
688 | * delayed local completion of an application write), | |
689 | * try to fix it up by recounting here. */ | |
b30ab791 | 690 | ext->rs_left = drbd_bm_e_weight(device, enr); |
b411b363 PR |
691 | } |
692 | } else { | |
693 | /* Normally this element should be in the cache, | |
694 | * since drbd_rs_begin_io() pulled it already in. | |
695 | * | |
696 | * But maybe an application write finished, and we set | |
697 | * something outside the resync lru_cache in sync. | |
698 | */ | |
b30ab791 | 699 | int rs_left = drbd_bm_e_weight(device, enr); |
b411b363 | 700 | if (ext->flags != 0) { |
d0180171 | 701 | drbd_warn(device, "changing resync lce: %d[%u;%02lx]" |
b411b363 PR |
702 | " -> %d[%u;00]\n", |
703 | ext->lce.lc_number, ext->rs_left, | |
704 | ext->flags, enr, rs_left); | |
705 | ext->flags = 0; | |
706 | } | |
707 | if (ext->rs_failed) { | |
d0180171 | 708 | drbd_warn(device, "Kicking resync_lru element enr=%u " |
b411b363 PR |
709 | "out with rs_failed=%d\n", |
710 | ext->lce.lc_number, ext->rs_failed); | |
b411b363 PR |
711 | } |
712 | ext->rs_left = rs_left; | |
5ab7d2c0 | 713 | ext->rs_failed = (mode == RECORD_RS_FAILED) ? count : 0; |
46a15bc3 LE |
714 | /* we don't keep a persistent log of the resync lru, |
715 | * we can commit any change right away. */ | |
b30ab791 | 716 | lc_committed(device->resync); |
b411b363 | 717 | } |
5ab7d2c0 LE |
718 | if (mode != SET_OUT_OF_SYNC) |
719 | lc_put(device->resync, &ext->lce); | |
b411b363 PR |
720 | /* no race, we are within the al_lock! */ |
721 | ||
5ab7d2c0 | 722 | if (ext->rs_left <= ext->rs_failed) { |
b411b363 | 723 | ext->rs_failed = 0; |
5ab7d2c0 | 724 | return true; |
b411b363 | 725 | } |
5ab7d2c0 LE |
726 | } else if (mode != SET_OUT_OF_SYNC) { |
727 | /* be quiet if lc_find() did not find it. */ | |
d0180171 | 728 | drbd_err(device, "lc_get() failed! locked=%d/%d flags=%lu\n", |
b30ab791 AG |
729 | device->resync_locked, |
730 | device->resync->nr_elements, | |
731 | device->resync->flags); | |
b411b363 | 732 | } |
5ab7d2c0 | 733 | return false; |
b411b363 PR |
734 | } |
735 | ||
b30ab791 | 736 | void drbd_advance_rs_marks(struct drbd_device *device, unsigned long still_to_go) |
c6ea14df LE |
737 | { |
738 | unsigned long now = jiffies; | |
b30ab791 AG |
739 | unsigned long last = device->rs_mark_time[device->rs_last_mark]; |
740 | int next = (device->rs_last_mark + 1) % DRBD_SYNC_MARKS; | |
c6ea14df | 741 | if (time_after_eq(now, last + DRBD_SYNC_MARK_STEP)) { |
b30ab791 AG |
742 | if (device->rs_mark_left[device->rs_last_mark] != still_to_go && |
743 | device->state.conn != C_PAUSED_SYNC_T && | |
744 | device->state.conn != C_PAUSED_SYNC_S) { | |
745 | device->rs_mark_time[next] = now; | |
746 | device->rs_mark_left[next] = still_to_go; | |
747 | device->rs_last_mark = next; | |
c6ea14df LE |
748 | } |
749 | } | |
750 | } | |
751 | ||
5ab7d2c0 LE |
752 | /* It is called lazy update, so don't do write-out too often. */ |
753 | static bool lazy_bitmap_update_due(struct drbd_device *device) | |
b411b363 | 754 | { |
5ab7d2c0 LE |
755 | return time_after(jiffies, device->rs_last_bcast + 2*HZ); |
756 | } | |
b411b363 | 757 | |
5ab7d2c0 LE |
758 | static void maybe_schedule_on_disk_bitmap_update(struct drbd_device *device, bool rs_done) |
759 | { | |
5ab7d2c0 LE |
760 | if (rs_done) |
761 | set_bit(RS_DONE, &device->flags); | |
762 | /* and also set RS_PROGRESS below */ | |
763 | else if (!lazy_bitmap_update_due(device)) | |
b411b363 | 764 | return; |
518a4d53 | 765 | |
e334f550 | 766 | drbd_device_post_work(device, RS_PROGRESS); |
5ab7d2c0 | 767 | } |
b411b363 | 768 | |
5ab7d2c0 LE |
769 | static int update_sync_bits(struct drbd_device *device, |
770 | unsigned long sbnr, unsigned long ebnr, | |
771 | enum update_sync_bits_mode mode) | |
772 | { | |
b411b363 | 773 | /* |
5ab7d2c0 LE |
774 | * We keep a count of set bits per resync-extent in the ->rs_left |
775 | * caching member, so we need to loop and work within the resync extent | |
776 | * alignment. Typically this loop will execute exactly once. | |
b411b363 | 777 | */ |
5ab7d2c0 LE |
778 | unsigned long flags; |
779 | unsigned long count = 0; | |
780 | unsigned int cleared = 0; | |
781 | while (sbnr <= ebnr) { | |
782 | /* set temporary boundary bit number to last bit number within | |
783 | * the resync extent of the current start bit number, | |
784 | * but cap at provided end bit number */ | |
785 | unsigned long tbnr = min(ebnr, sbnr | BM_BLOCKS_PER_BM_EXT_MASK); | |
786 | unsigned long c; | |
787 | ||
788 | if (mode == RECORD_RS_FAILED) | |
789 | /* Only called from drbd_rs_failed_io(), bits | |
790 | * supposedly still set. Recount, maybe some | |
791 | * of the bits have been successfully cleared | |
792 | * by application IO meanwhile. | |
793 | */ | |
794 | c = drbd_bm_count_bits(device, sbnr, tbnr); | |
795 | else if (mode == SET_IN_SYNC) | |
796 | c = drbd_bm_clear_bits(device, sbnr, tbnr); | |
797 | else /* if (mode == SET_OUT_OF_SYNC) */ | |
798 | c = drbd_bm_set_bits(device, sbnr, tbnr); | |
799 | ||
800 | if (c) { | |
801 | spin_lock_irqsave(&device->al_lock, flags); | |
802 | cleared += update_rs_extent(device, BM_BIT_TO_EXT(sbnr), c, mode); | |
803 | spin_unlock_irqrestore(&device->al_lock, flags); | |
804 | count += c; | |
805 | } | |
806 | sbnr = tbnr + 1; | |
b411b363 | 807 | } |
5ab7d2c0 LE |
808 | if (count) { |
809 | if (mode == SET_IN_SYNC) { | |
810 | unsigned long still_to_go = drbd_bm_total_weight(device); | |
811 | bool rs_is_done = (still_to_go <= device->rs_failed); | |
812 | drbd_advance_rs_marks(device, still_to_go); | |
813 | if (cleared || rs_is_done) | |
814 | maybe_schedule_on_disk_bitmap_update(device, rs_is_done); | |
815 | } else if (mode == RECORD_RS_FAILED) | |
816 | device->rs_failed += count; | |
b30ab791 | 817 | wake_up(&device->al_wait); |
5ab7d2c0 LE |
818 | } |
819 | return count; | |
b411b363 PR |
820 | } |
821 | ||
5ab7d2c0 LE |
822 | /* clear the bit corresponding to the piece of storage in question: |
823 | * size byte of data starting from sector. Only clear a bits of the affected | |
824 | * one ore more _aligned_ BM_BLOCK_SIZE blocks. | |
825 | * | |
826 | * called by worker on C_SYNC_TARGET and receiver on SyncSource. | |
b411b363 | 827 | * |
b411b363 | 828 | */ |
5ab7d2c0 LE |
829 | int __drbd_change_sync(struct drbd_device *device, sector_t sector, int size, |
830 | enum update_sync_bits_mode mode, | |
831 | const char *file, const unsigned int line) | |
b411b363 | 832 | { |
5ab7d2c0 LE |
833 | /* Is called from worker and receiver context _only_ */ |
834 | unsigned long sbnr, ebnr, lbnr; | |
835 | unsigned long count = 0; | |
b411b363 | 836 | sector_t esector, nr_sectors; |
b411b363 | 837 | |
5ab7d2c0 LE |
838 | /* This would be an empty REQ_FLUSH, be silent. */ |
839 | if ((mode == SET_OUT_OF_SYNC) && size == 0) | |
81a3537a LE |
840 | return 0; |
841 | ||
5ab7d2c0 LE |
842 | if (size <= 0 || !IS_ALIGNED(size, 512) || size > DRBD_MAX_DISCARD_SIZE) { |
843 | drbd_err(device, "%s: sector=%llus size=%d nonsense!\n", | |
844 | drbd_change_sync_fname[mode], | |
845 | (unsigned long long)sector, size); | |
73a01a18 | 846 | return 0; |
b411b363 PR |
847 | } |
848 | ||
b30ab791 | 849 | if (!get_ldev(device)) |
5ab7d2c0 | 850 | return 0; /* no disk, no metadata, no bitmap to manipulate bits in */ |
b411b363 | 851 | |
b30ab791 | 852 | nr_sectors = drbd_get_capacity(device->this_bdev); |
b411b363 PR |
853 | esector = sector + (size >> 9) - 1; |
854 | ||
841ce241 | 855 | if (!expect(sector < nr_sectors)) |
b411b363 | 856 | goto out; |
841ce241 AG |
857 | if (!expect(esector < nr_sectors)) |
858 | esector = nr_sectors - 1; | |
b411b363 | 859 | |
5ab7d2c0 | 860 | lbnr = BM_SECT_TO_BIT(nr_sectors-1); |
b411b363 | 861 | |
5ab7d2c0 LE |
862 | if (mode == SET_IN_SYNC) { |
863 | /* Round up start sector, round down end sector. We make sure | |
864 | * we only clear full, aligned, BM_BLOCK_SIZE blocks. */ | |
865 | if (unlikely(esector < BM_SECT_PER_BIT-1)) | |
866 | goto out; | |
867 | if (unlikely(esector == (nr_sectors-1))) | |
868 | ebnr = lbnr; | |
869 | else | |
870 | ebnr = BM_SECT_TO_BIT(esector - (BM_SECT_PER_BIT-1)); | |
871 | sbnr = BM_SECT_TO_BIT(sector + BM_SECT_PER_BIT-1); | |
872 | } else { | |
873 | /* We set it out of sync, or record resync failure. | |
874 | * Should not round anything here. */ | |
875 | sbnr = BM_SECT_TO_BIT(sector); | |
876 | ebnr = BM_SECT_TO_BIT(esector); | |
877 | } | |
b411b363 | 878 | |
5ab7d2c0 | 879 | count = update_sync_bits(device, sbnr, ebnr, mode); |
b411b363 | 880 | out: |
b30ab791 | 881 | put_ldev(device); |
73a01a18 | 882 | return count; |
b411b363 PR |
883 | } |
884 | ||
885 | static | |
b30ab791 | 886 | struct bm_extent *_bme_get(struct drbd_device *device, unsigned int enr) |
b411b363 PR |
887 | { |
888 | struct lc_element *e; | |
889 | struct bm_extent *bm_ext; | |
890 | int wakeup = 0; | |
891 | unsigned long rs_flags; | |
892 | ||
b30ab791 AG |
893 | spin_lock_irq(&device->al_lock); |
894 | if (device->resync_locked > device->resync->nr_elements/2) { | |
895 | spin_unlock_irq(&device->al_lock); | |
b411b363 PR |
896 | return NULL; |
897 | } | |
b30ab791 | 898 | e = lc_get(device->resync, enr); |
b411b363 PR |
899 | bm_ext = e ? lc_entry(e, struct bm_extent, lce) : NULL; |
900 | if (bm_ext) { | |
901 | if (bm_ext->lce.lc_number != enr) { | |
b30ab791 | 902 | bm_ext->rs_left = drbd_bm_e_weight(device, enr); |
b411b363 | 903 | bm_ext->rs_failed = 0; |
b30ab791 | 904 | lc_committed(device->resync); |
b411b363 PR |
905 | wakeup = 1; |
906 | } | |
907 | if (bm_ext->lce.refcnt == 1) | |
b30ab791 | 908 | device->resync_locked++; |
b411b363 PR |
909 | set_bit(BME_NO_WRITES, &bm_ext->flags); |
910 | } | |
b30ab791 AG |
911 | rs_flags = device->resync->flags; |
912 | spin_unlock_irq(&device->al_lock); | |
b411b363 | 913 | if (wakeup) |
b30ab791 | 914 | wake_up(&device->al_wait); |
b411b363 PR |
915 | |
916 | if (!bm_ext) { | |
917 | if (rs_flags & LC_STARVING) | |
d0180171 | 918 | drbd_warn(device, "Have to wait for element" |
b411b363 | 919 | " (resync LRU too small?)\n"); |
46a15bc3 | 920 | BUG_ON(rs_flags & LC_LOCKED); |
b411b363 PR |
921 | } |
922 | ||
923 | return bm_ext; | |
924 | } | |
925 | ||
b30ab791 | 926 | static int _is_in_al(struct drbd_device *device, unsigned int enr) |
b411b363 | 927 | { |
46a15bc3 | 928 | int rv; |
b411b363 | 929 | |
b30ab791 AG |
930 | spin_lock_irq(&device->al_lock); |
931 | rv = lc_is_used(device->act_log, enr); | |
932 | spin_unlock_irq(&device->al_lock); | |
b411b363 | 933 | |
b411b363 PR |
934 | return rv; |
935 | } | |
936 | ||
937 | /** | |
938 | * drbd_rs_begin_io() - Gets an extent in the resync LRU cache and sets it to BME_LOCKED | |
b30ab791 | 939 | * @device: DRBD device. |
b411b363 PR |
940 | * @sector: The sector number. |
941 | * | |
80a40e43 | 942 | * This functions sleeps on al_wait. Returns 0 on success, -EINTR if interrupted. |
b411b363 | 943 | */ |
b30ab791 | 944 | int drbd_rs_begin_io(struct drbd_device *device, sector_t sector) |
b411b363 PR |
945 | { |
946 | unsigned int enr = BM_SECT_TO_EXT(sector); | |
947 | struct bm_extent *bm_ext; | |
948 | int i, sig; | |
e8299874 | 949 | bool sa; |
b411b363 | 950 | |
f91ab628 | 951 | retry: |
b30ab791 AG |
952 | sig = wait_event_interruptible(device->al_wait, |
953 | (bm_ext = _bme_get(device, enr))); | |
b411b363 | 954 | if (sig) |
80a40e43 | 955 | return -EINTR; |
b411b363 PR |
956 | |
957 | if (test_bit(BME_LOCKED, &bm_ext->flags)) | |
80a40e43 | 958 | return 0; |
b411b363 | 959 | |
e8299874 LE |
960 | /* step aside only while we are above c-min-rate; unless disabled. */ |
961 | sa = drbd_rs_c_min_rate_throttle(device); | |
962 | ||
b411b363 | 963 | for (i = 0; i < AL_EXT_PER_BM_SECT; i++) { |
b30ab791 AG |
964 | sig = wait_event_interruptible(device->al_wait, |
965 | !_is_in_al(device, enr * AL_EXT_PER_BM_SECT + i) || | |
e8299874 | 966 | (sa && test_bit(BME_PRIORITY, &bm_ext->flags))); |
f91ab628 | 967 | |
e8299874 | 968 | if (sig || (sa && test_bit(BME_PRIORITY, &bm_ext->flags))) { |
b30ab791 AG |
969 | spin_lock_irq(&device->al_lock); |
970 | if (lc_put(device->resync, &bm_ext->lce) == 0) { | |
f91ab628 | 971 | bm_ext->flags = 0; /* clears BME_NO_WRITES and eventually BME_PRIORITY */ |
b30ab791 AG |
972 | device->resync_locked--; |
973 | wake_up(&device->al_wait); | |
b411b363 | 974 | } |
b30ab791 | 975 | spin_unlock_irq(&device->al_lock); |
f91ab628 PR |
976 | if (sig) |
977 | return -EINTR; | |
978 | if (schedule_timeout_interruptible(HZ/10)) | |
979 | return -EINTR; | |
f91ab628 | 980 | goto retry; |
b411b363 PR |
981 | } |
982 | } | |
b411b363 | 983 | set_bit(BME_LOCKED, &bm_ext->flags); |
80a40e43 | 984 | return 0; |
b411b363 PR |
985 | } |
986 | ||
987 | /** | |
988 | * drbd_try_rs_begin_io() - Gets an extent in the resync LRU cache, does not sleep | |
b30ab791 | 989 | * @device: DRBD device. |
b411b363 PR |
990 | * @sector: The sector number. |
991 | * | |
992 | * Gets an extent in the resync LRU cache, sets it to BME_NO_WRITES, then | |
993 | * tries to set it to BME_LOCKED. Returns 0 upon success, and -EAGAIN | |
994 | * if there is still application IO going on in this area. | |
995 | */ | |
b30ab791 | 996 | int drbd_try_rs_begin_io(struct drbd_device *device, sector_t sector) |
b411b363 PR |
997 | { |
998 | unsigned int enr = BM_SECT_TO_EXT(sector); | |
999 | const unsigned int al_enr = enr*AL_EXT_PER_BM_SECT; | |
1000 | struct lc_element *e; | |
1001 | struct bm_extent *bm_ext; | |
1002 | int i; | |
ad3fee79 LE |
1003 | bool throttle = drbd_rs_should_slow_down(device, sector, true); |
1004 | ||
1005 | /* If we need to throttle, a half-locked (only marked BME_NO_WRITES, | |
1006 | * not yet BME_LOCKED) extent needs to be kicked out explicitly if we | |
1007 | * need to throttle. There is at most one such half-locked extent, | |
1008 | * which is remembered in resync_wenr. */ | |
1009 | ||
1010 | if (throttle && device->resync_wenr != enr) | |
1011 | return -EAGAIN; | |
b411b363 | 1012 | |
b30ab791 AG |
1013 | spin_lock_irq(&device->al_lock); |
1014 | if (device->resync_wenr != LC_FREE && device->resync_wenr != enr) { | |
b411b363 PR |
1015 | /* in case you have very heavy scattered io, it may |
1016 | * stall the syncer undefined if we give up the ref count | |
1017 | * when we try again and requeue. | |
1018 | * | |
1019 | * if we don't give up the refcount, but the next time | |
1020 | * we are scheduled this extent has been "synced" by new | |
1021 | * application writes, we'd miss the lc_put on the | |
1022 | * extent we keep the refcount on. | |
1023 | * so we remembered which extent we had to try again, and | |
1024 | * if the next requested one is something else, we do | |
1025 | * the lc_put here... | |
1026 | * we also have to wake_up | |
1027 | */ | |
b30ab791 | 1028 | e = lc_find(device->resync, device->resync_wenr); |
b411b363 PR |
1029 | bm_ext = e ? lc_entry(e, struct bm_extent, lce) : NULL; |
1030 | if (bm_ext) { | |
0b0ba1ef AG |
1031 | D_ASSERT(device, !test_bit(BME_LOCKED, &bm_ext->flags)); |
1032 | D_ASSERT(device, test_bit(BME_NO_WRITES, &bm_ext->flags)); | |
b411b363 | 1033 | clear_bit(BME_NO_WRITES, &bm_ext->flags); |
b30ab791 | 1034 | device->resync_wenr = LC_FREE; |
ad3fee79 LE |
1035 | if (lc_put(device->resync, &bm_ext->lce) == 0) { |
1036 | bm_ext->flags = 0; | |
b30ab791 | 1037 | device->resync_locked--; |
ad3fee79 | 1038 | } |
b30ab791 | 1039 | wake_up(&device->al_wait); |
b411b363 | 1040 | } else { |
d0180171 | 1041 | drbd_alert(device, "LOGIC BUG\n"); |
b411b363 PR |
1042 | } |
1043 | } | |
1044 | /* TRY. */ | |
b30ab791 | 1045 | e = lc_try_get(device->resync, enr); |
b411b363 PR |
1046 | bm_ext = e ? lc_entry(e, struct bm_extent, lce) : NULL; |
1047 | if (bm_ext) { | |
1048 | if (test_bit(BME_LOCKED, &bm_ext->flags)) | |
1049 | goto proceed; | |
1050 | if (!test_and_set_bit(BME_NO_WRITES, &bm_ext->flags)) { | |
b30ab791 | 1051 | device->resync_locked++; |
b411b363 PR |
1052 | } else { |
1053 | /* we did set the BME_NO_WRITES, | |
1054 | * but then could not set BME_LOCKED, | |
1055 | * so we tried again. | |
1056 | * drop the extra reference. */ | |
b411b363 | 1057 | bm_ext->lce.refcnt--; |
0b0ba1ef | 1058 | D_ASSERT(device, bm_ext->lce.refcnt > 0); |
b411b363 PR |
1059 | } |
1060 | goto check_al; | |
1061 | } else { | |
1062 | /* do we rather want to try later? */ | |
b30ab791 | 1063 | if (device->resync_locked > device->resync->nr_elements-3) |
b411b363 | 1064 | goto try_again; |
b411b363 | 1065 | /* Do or do not. There is no try. -- Yoda */ |
b30ab791 | 1066 | e = lc_get(device->resync, enr); |
b411b363 PR |
1067 | bm_ext = e ? lc_entry(e, struct bm_extent, lce) : NULL; |
1068 | if (!bm_ext) { | |
b30ab791 | 1069 | const unsigned long rs_flags = device->resync->flags; |
b411b363 | 1070 | if (rs_flags & LC_STARVING) |
d0180171 | 1071 | drbd_warn(device, "Have to wait for element" |
b411b363 | 1072 | " (resync LRU too small?)\n"); |
46a15bc3 | 1073 | BUG_ON(rs_flags & LC_LOCKED); |
b411b363 PR |
1074 | goto try_again; |
1075 | } | |
1076 | if (bm_ext->lce.lc_number != enr) { | |
b30ab791 | 1077 | bm_ext->rs_left = drbd_bm_e_weight(device, enr); |
b411b363 | 1078 | bm_ext->rs_failed = 0; |
b30ab791 AG |
1079 | lc_committed(device->resync); |
1080 | wake_up(&device->al_wait); | |
0b0ba1ef | 1081 | D_ASSERT(device, test_bit(BME_LOCKED, &bm_ext->flags) == 0); |
b411b363 PR |
1082 | } |
1083 | set_bit(BME_NO_WRITES, &bm_ext->flags); | |
0b0ba1ef | 1084 | D_ASSERT(device, bm_ext->lce.refcnt == 1); |
b30ab791 | 1085 | device->resync_locked++; |
b411b363 PR |
1086 | goto check_al; |
1087 | } | |
1088 | check_al: | |
b411b363 | 1089 | for (i = 0; i < AL_EXT_PER_BM_SECT; i++) { |
b30ab791 | 1090 | if (lc_is_used(device->act_log, al_enr+i)) |
b411b363 PR |
1091 | goto try_again; |
1092 | } | |
1093 | set_bit(BME_LOCKED, &bm_ext->flags); | |
1094 | proceed: | |
b30ab791 AG |
1095 | device->resync_wenr = LC_FREE; |
1096 | spin_unlock_irq(&device->al_lock); | |
b411b363 PR |
1097 | return 0; |
1098 | ||
1099 | try_again: | |
ad3fee79 LE |
1100 | if (bm_ext) { |
1101 | if (throttle) { | |
1102 | D_ASSERT(device, !test_bit(BME_LOCKED, &bm_ext->flags)); | |
1103 | D_ASSERT(device, test_bit(BME_NO_WRITES, &bm_ext->flags)); | |
1104 | clear_bit(BME_NO_WRITES, &bm_ext->flags); | |
1105 | device->resync_wenr = LC_FREE; | |
1106 | if (lc_put(device->resync, &bm_ext->lce) == 0) { | |
1107 | bm_ext->flags = 0; | |
1108 | device->resync_locked--; | |
1109 | } | |
1110 | wake_up(&device->al_wait); | |
1111 | } else | |
1112 | device->resync_wenr = enr; | |
1113 | } | |
b30ab791 | 1114 | spin_unlock_irq(&device->al_lock); |
b411b363 PR |
1115 | return -EAGAIN; |
1116 | } | |
1117 | ||
b30ab791 | 1118 | void drbd_rs_complete_io(struct drbd_device *device, sector_t sector) |
b411b363 PR |
1119 | { |
1120 | unsigned int enr = BM_SECT_TO_EXT(sector); | |
1121 | struct lc_element *e; | |
1122 | struct bm_extent *bm_ext; | |
1123 | unsigned long flags; | |
1124 | ||
b30ab791 AG |
1125 | spin_lock_irqsave(&device->al_lock, flags); |
1126 | e = lc_find(device->resync, enr); | |
b411b363 PR |
1127 | bm_ext = e ? lc_entry(e, struct bm_extent, lce) : NULL; |
1128 | if (!bm_ext) { | |
b30ab791 | 1129 | spin_unlock_irqrestore(&device->al_lock, flags); |
b411b363 | 1130 | if (__ratelimit(&drbd_ratelimit_state)) |
d0180171 | 1131 | drbd_err(device, "drbd_rs_complete_io() called, but extent not found\n"); |
b411b363 PR |
1132 | return; |
1133 | } | |
1134 | ||
1135 | if (bm_ext->lce.refcnt == 0) { | |
b30ab791 | 1136 | spin_unlock_irqrestore(&device->al_lock, flags); |
d0180171 | 1137 | drbd_err(device, "drbd_rs_complete_io(,%llu [=%u]) called, " |
b411b363 PR |
1138 | "but refcnt is 0!?\n", |
1139 | (unsigned long long)sector, enr); | |
1140 | return; | |
1141 | } | |
1142 | ||
b30ab791 | 1143 | if (lc_put(device->resync, &bm_ext->lce) == 0) { |
e3555d85 | 1144 | bm_ext->flags = 0; /* clear BME_LOCKED, BME_NO_WRITES and BME_PRIORITY */ |
b30ab791 AG |
1145 | device->resync_locked--; |
1146 | wake_up(&device->al_wait); | |
b411b363 PR |
1147 | } |
1148 | ||
b30ab791 | 1149 | spin_unlock_irqrestore(&device->al_lock, flags); |
b411b363 PR |
1150 | } |
1151 | ||
1152 | /** | |
1153 | * drbd_rs_cancel_all() - Removes all extents from the resync LRU (even BME_LOCKED) | |
b30ab791 | 1154 | * @device: DRBD device. |
b411b363 | 1155 | */ |
b30ab791 | 1156 | void drbd_rs_cancel_all(struct drbd_device *device) |
b411b363 | 1157 | { |
b30ab791 | 1158 | spin_lock_irq(&device->al_lock); |
b411b363 | 1159 | |
b30ab791 AG |
1160 | if (get_ldev_if_state(device, D_FAILED)) { /* Makes sure ->resync is there. */ |
1161 | lc_reset(device->resync); | |
1162 | put_ldev(device); | |
b411b363 | 1163 | } |
b30ab791 AG |
1164 | device->resync_locked = 0; |
1165 | device->resync_wenr = LC_FREE; | |
1166 | spin_unlock_irq(&device->al_lock); | |
1167 | wake_up(&device->al_wait); | |
b411b363 PR |
1168 | } |
1169 | ||
1170 | /** | |
1171 | * drbd_rs_del_all() - Gracefully remove all extents from the resync LRU | |
b30ab791 | 1172 | * @device: DRBD device. |
b411b363 PR |
1173 | * |
1174 | * Returns 0 upon success, -EAGAIN if at least one reference count was | |
1175 | * not zero. | |
1176 | */ | |
b30ab791 | 1177 | int drbd_rs_del_all(struct drbd_device *device) |
b411b363 PR |
1178 | { |
1179 | struct lc_element *e; | |
1180 | struct bm_extent *bm_ext; | |
1181 | int i; | |
1182 | ||
b30ab791 | 1183 | spin_lock_irq(&device->al_lock); |
b411b363 | 1184 | |
b30ab791 | 1185 | if (get_ldev_if_state(device, D_FAILED)) { |
b411b363 | 1186 | /* ok, ->resync is there. */ |
b30ab791 AG |
1187 | for (i = 0; i < device->resync->nr_elements; i++) { |
1188 | e = lc_element_by_index(device->resync, i); | |
b2b163dd | 1189 | bm_ext = lc_entry(e, struct bm_extent, lce); |
b411b363 PR |
1190 | if (bm_ext->lce.lc_number == LC_FREE) |
1191 | continue; | |
b30ab791 | 1192 | if (bm_ext->lce.lc_number == device->resync_wenr) { |
d0180171 | 1193 | drbd_info(device, "dropping %u in drbd_rs_del_all, apparently" |
b411b363 | 1194 | " got 'synced' by application io\n", |
b30ab791 | 1195 | device->resync_wenr); |
0b0ba1ef AG |
1196 | D_ASSERT(device, !test_bit(BME_LOCKED, &bm_ext->flags)); |
1197 | D_ASSERT(device, test_bit(BME_NO_WRITES, &bm_ext->flags)); | |
b411b363 | 1198 | clear_bit(BME_NO_WRITES, &bm_ext->flags); |
b30ab791 AG |
1199 | device->resync_wenr = LC_FREE; |
1200 | lc_put(device->resync, &bm_ext->lce); | |
b411b363 PR |
1201 | } |
1202 | if (bm_ext->lce.refcnt != 0) { | |
d0180171 | 1203 | drbd_info(device, "Retrying drbd_rs_del_all() later. " |
b411b363 | 1204 | "refcnt=%d\n", bm_ext->lce.refcnt); |
b30ab791 AG |
1205 | put_ldev(device); |
1206 | spin_unlock_irq(&device->al_lock); | |
b411b363 PR |
1207 | return -EAGAIN; |
1208 | } | |
0b0ba1ef AG |
1209 | D_ASSERT(device, !test_bit(BME_LOCKED, &bm_ext->flags)); |
1210 | D_ASSERT(device, !test_bit(BME_NO_WRITES, &bm_ext->flags)); | |
b30ab791 | 1211 | lc_del(device->resync, &bm_ext->lce); |
b411b363 | 1212 | } |
0b0ba1ef | 1213 | D_ASSERT(device, device->resync->used == 0); |
b30ab791 | 1214 | put_ldev(device); |
b411b363 | 1215 | } |
b30ab791 AG |
1216 | spin_unlock_irq(&device->al_lock); |
1217 | wake_up(&device->al_wait); | |
b411b363 PR |
1218 | |
1219 | return 0; | |
1220 | } |