]>
Commit | Line | Data |
---|---|---|
b411b363 PR |
1 | /* |
2 | drbd.c | |
3 | ||
4 | This file is part of DRBD by Philipp Reisner and Lars Ellenberg. | |
5 | ||
6 | Copyright (C) 2001-2008, LINBIT Information Technologies GmbH. | |
7 | Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>. | |
8 | Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>. | |
9 | ||
10 | Thanks to Carter Burden, Bart Grantham and Gennadiy Nerubayev | |
11 | from Logicworks, Inc. for making SDP replication support possible. | |
12 | ||
13 | drbd is free software; you can redistribute it and/or modify | |
14 | it under the terms of the GNU General Public License as published by | |
15 | the Free Software Foundation; either version 2, or (at your option) | |
16 | any later version. | |
17 | ||
18 | drbd is distributed in the hope that it will be useful, | |
19 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
20 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
21 | GNU General Public License for more details. | |
22 | ||
23 | You should have received a copy of the GNU General Public License | |
24 | along with drbd; see the file COPYING. If not, write to | |
25 | the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. | |
26 | ||
27 | */ | |
28 | ||
b411b363 | 29 | #include <linux/module.h> |
b411b363 PR |
30 | #include <linux/drbd.h> |
31 | #include <asm/uaccess.h> | |
32 | #include <asm/types.h> | |
33 | #include <net/sock.h> | |
34 | #include <linux/ctype.h> | |
2a48fc0a | 35 | #include <linux/mutex.h> |
b411b363 PR |
36 | #include <linux/fs.h> |
37 | #include <linux/file.h> | |
38 | #include <linux/proc_fs.h> | |
39 | #include <linux/init.h> | |
40 | #include <linux/mm.h> | |
41 | #include <linux/memcontrol.h> | |
42 | #include <linux/mm_inline.h> | |
43 | #include <linux/slab.h> | |
44 | #include <linux/random.h> | |
45 | #include <linux/reboot.h> | |
46 | #include <linux/notifier.h> | |
47 | #include <linux/kthread.h> | |
48 | ||
49 | #define __KERNEL_SYSCALLS__ | |
50 | #include <linux/unistd.h> | |
51 | #include <linux/vmalloc.h> | |
52 | ||
53 | #include <linux/drbd_limits.h> | |
54 | #include "drbd_int.h" | |
b411b363 PR |
55 | #include "drbd_req.h" /* only for _req_mod in tl_release and tl_clear */ |
56 | ||
57 | #include "drbd_vli.h" | |
58 | ||
2a48fc0a | 59 | static DEFINE_MUTEX(drbd_main_mutex); |
b411b363 PR |
60 | int drbdd_init(struct drbd_thread *); |
61 | int drbd_worker(struct drbd_thread *); | |
62 | int drbd_asender(struct drbd_thread *); | |
63 | ||
64 | int drbd_init(void); | |
65 | static int drbd_open(struct block_device *bdev, fmode_t mode); | |
66 | static int drbd_release(struct gendisk *gd, fmode_t mode); | |
99920dc5 | 67 | static int w_md_sync(struct drbd_work *w, int unused); |
b411b363 | 68 | static void md_sync_timer_fn(unsigned long data); |
99920dc5 AG |
69 | static int w_bitmap_io(struct drbd_work *w, int unused); |
70 | static int w_go_diskless(struct drbd_work *w, int unused); | |
b411b363 | 71 | |
b411b363 PR |
72 | MODULE_AUTHOR("Philipp Reisner <phil@linbit.com>, " |
73 | "Lars Ellenberg <lars@linbit.com>"); | |
74 | MODULE_DESCRIPTION("drbd - Distributed Replicated Block Device v" REL_VERSION); | |
75 | MODULE_VERSION(REL_VERSION); | |
76 | MODULE_LICENSE("GPL"); | |
81a5d60e | 77 | MODULE_PARM_DESC(minor_count, "Approximate number of drbd devices (" |
2b8a90b5 | 78 | __stringify(DRBD_MINOR_COUNT_MIN) "-" __stringify(DRBD_MINOR_COUNT_MAX) ")"); |
b411b363 PR |
79 | MODULE_ALIAS_BLOCKDEV_MAJOR(DRBD_MAJOR); |
80 | ||
81 | #include <linux/moduleparam.h> | |
82 | /* allow_open_on_secondary */ | |
83 | MODULE_PARM_DESC(allow_oos, "DONT USE!"); | |
84 | /* thanks to these macros, if compiled into the kernel (not-module), | |
85 | * this becomes the boot parameter drbd.minor_count */ | |
86 | module_param(minor_count, uint, 0444); | |
87 | module_param(disable_sendpage, bool, 0644); | |
88 | module_param(allow_oos, bool, 0); | |
b411b363 PR |
89 | module_param(proc_details, int, 0644); |
90 | ||
91 | #ifdef CONFIG_DRBD_FAULT_INJECTION | |
92 | int enable_faults; | |
93 | int fault_rate; | |
94 | static int fault_count; | |
95 | int fault_devs; | |
96 | /* bitmap of enabled faults */ | |
97 | module_param(enable_faults, int, 0664); | |
98 | /* fault rate % value - applies to all enabled faults */ | |
99 | module_param(fault_rate, int, 0664); | |
100 | /* count of faults inserted */ | |
101 | module_param(fault_count, int, 0664); | |
102 | /* bitmap of devices to insert faults on */ | |
103 | module_param(fault_devs, int, 0644); | |
104 | #endif | |
105 | ||
106 | /* module parameter, defined */ | |
2b8a90b5 | 107 | unsigned int minor_count = DRBD_MINOR_COUNT_DEF; |
b411b363 PR |
108 | int disable_sendpage; |
109 | int allow_oos; | |
b411b363 PR |
110 | int proc_details; /* Detail level in proc drbd*/ |
111 | ||
112 | /* Module parameter for setting the user mode helper program | |
113 | * to run. Default is /sbin/drbdadm */ | |
114 | char usermode_helper[80] = "/sbin/drbdadm"; | |
115 | ||
116 | module_param_string(usermode_helper, usermode_helper, sizeof(usermode_helper), 0644); | |
117 | ||
118 | /* in 2.6.x, our device mapping and config info contains our virtual gendisks | |
119 | * as member "struct gendisk *vdisk;" | |
120 | */ | |
81a5d60e | 121 | struct idr minors; |
2111438b | 122 | struct list_head drbd_tconns; /* list of struct drbd_tconn */ |
b411b363 PR |
123 | |
124 | struct kmem_cache *drbd_request_cache; | |
6c852bec | 125 | struct kmem_cache *drbd_ee_cache; /* peer requests */ |
b411b363 PR |
126 | struct kmem_cache *drbd_bm_ext_cache; /* bitmap extents */ |
127 | struct kmem_cache *drbd_al_ext_cache; /* activity log extents */ | |
128 | mempool_t *drbd_request_mempool; | |
129 | mempool_t *drbd_ee_mempool; | |
35abf594 | 130 | mempool_t *drbd_md_io_page_pool; |
da4a75d2 | 131 | struct bio_set *drbd_md_io_bio_set; |
b411b363 PR |
132 | |
133 | /* I do not use a standard mempool, because: | |
134 | 1) I want to hand out the pre-allocated objects first. | |
135 | 2) I want to be able to interrupt sleeping allocation with a signal. | |
136 | Note: This is a single linked list, the next pointer is the private | |
137 | member of struct page. | |
138 | */ | |
139 | struct page *drbd_pp_pool; | |
140 | spinlock_t drbd_pp_lock; | |
141 | int drbd_pp_vacant; | |
142 | wait_queue_head_t drbd_pp_wait; | |
143 | ||
144 | DEFINE_RATELIMIT_STATE(drbd_ratelimit_state, 5 * HZ, 5); | |
145 | ||
7d4e9d09 | 146 | static const struct block_device_operations drbd_ops = { |
b411b363 PR |
147 | .owner = THIS_MODULE, |
148 | .open = drbd_open, | |
149 | .release = drbd_release, | |
150 | }; | |
151 | ||
da4a75d2 LE |
152 | static void bio_destructor_drbd(struct bio *bio) |
153 | { | |
154 | bio_free(bio, drbd_md_io_bio_set); | |
155 | } | |
156 | ||
157 | struct bio *bio_alloc_drbd(gfp_t gfp_mask) | |
158 | { | |
159 | struct bio *bio; | |
160 | ||
161 | if (!drbd_md_io_bio_set) | |
162 | return bio_alloc(gfp_mask, 1); | |
163 | ||
164 | bio = bio_alloc_bioset(gfp_mask, 1, drbd_md_io_bio_set); | |
165 | if (!bio) | |
166 | return NULL; | |
167 | bio->bi_destructor = bio_destructor_drbd; | |
168 | return bio; | |
169 | } | |
170 | ||
b411b363 PR |
171 | #ifdef __CHECKER__ |
172 | /* When checking with sparse, and this is an inline function, sparse will | |
173 | give tons of false positives. When this is a real functions sparse works. | |
174 | */ | |
175 | int _get_ldev_if_state(struct drbd_conf *mdev, enum drbd_disk_state mins) | |
176 | { | |
177 | int io_allowed; | |
178 | ||
179 | atomic_inc(&mdev->local_cnt); | |
180 | io_allowed = (mdev->state.disk >= mins); | |
181 | if (!io_allowed) { | |
182 | if (atomic_dec_and_test(&mdev->local_cnt)) | |
183 | wake_up(&mdev->misc_wait); | |
184 | } | |
185 | return io_allowed; | |
186 | } | |
187 | ||
188 | #endif | |
189 | ||
190 | /** | |
191 | * DOC: The transfer log | |
192 | * | |
193 | * The transfer log is a single linked list of &struct drbd_tl_epoch objects. | |
87eeee41 | 194 | * mdev->tconn->newest_tle points to the head, mdev->tconn->oldest_tle points to the tail |
b411b363 PR |
195 | * of the list. There is always at least one &struct drbd_tl_epoch object. |
196 | * | |
197 | * Each &struct drbd_tl_epoch has a circular double linked list of requests | |
198 | * attached. | |
199 | */ | |
2f5cdd0b | 200 | static int tl_init(struct drbd_tconn *tconn) |
b411b363 PR |
201 | { |
202 | struct drbd_tl_epoch *b; | |
203 | ||
204 | /* during device minor initialization, we may well use GFP_KERNEL */ | |
205 | b = kmalloc(sizeof(struct drbd_tl_epoch), GFP_KERNEL); | |
206 | if (!b) | |
207 | return 0; | |
208 | INIT_LIST_HEAD(&b->requests); | |
209 | INIT_LIST_HEAD(&b->w.list); | |
210 | b->next = NULL; | |
211 | b->br_number = 4711; | |
7e602c0a | 212 | b->n_writes = 0; |
b411b363 PR |
213 | b->w.cb = NULL; /* if this is != NULL, we need to dec_ap_pending in tl_clear */ |
214 | ||
2f5cdd0b PR |
215 | tconn->oldest_tle = b; |
216 | tconn->newest_tle = b; | |
217 | INIT_LIST_HEAD(&tconn->out_of_sequence_requests); | |
cdfda633 | 218 | INIT_LIST_HEAD(&tconn->barrier_acked_requests); |
b411b363 | 219 | |
b411b363 PR |
220 | return 1; |
221 | } | |
222 | ||
2f5cdd0b | 223 | static void tl_cleanup(struct drbd_tconn *tconn) |
b411b363 | 224 | { |
2f5cdd0b PR |
225 | if (tconn->oldest_tle != tconn->newest_tle) |
226 | conn_err(tconn, "ASSERT FAILED: oldest_tle == newest_tle\n"); | |
227 | if (!list_empty(&tconn->out_of_sequence_requests)) | |
228 | conn_err(tconn, "ASSERT FAILED: list_empty(out_of_sequence_requests)\n"); | |
229 | kfree(tconn->oldest_tle); | |
230 | tconn->oldest_tle = NULL; | |
231 | kfree(tconn->unused_spare_tle); | |
232 | tconn->unused_spare_tle = NULL; | |
d628769b AG |
233 | } |
234 | ||
b411b363 PR |
235 | /** |
236 | * _tl_add_barrier() - Adds a barrier to the transfer log | |
237 | * @mdev: DRBD device. | |
238 | * @new: Barrier to be added before the current head of the TL. | |
239 | * | |
240 | * The caller must hold the req_lock. | |
241 | */ | |
2f5cdd0b | 242 | void _tl_add_barrier(struct drbd_tconn *tconn, struct drbd_tl_epoch *new) |
b411b363 PR |
243 | { |
244 | struct drbd_tl_epoch *newest_before; | |
245 | ||
246 | INIT_LIST_HEAD(&new->requests); | |
247 | INIT_LIST_HEAD(&new->w.list); | |
248 | new->w.cb = NULL; /* if this is != NULL, we need to dec_ap_pending in tl_clear */ | |
249 | new->next = NULL; | |
7e602c0a | 250 | new->n_writes = 0; |
b411b363 | 251 | |
2f5cdd0b | 252 | newest_before = tconn->newest_tle; |
b411b363 PR |
253 | /* never send a barrier number == 0, because that is special-cased |
254 | * when using TCQ for our write ordering code */ | |
255 | new->br_number = (newest_before->br_number+1) ?: 1; | |
2f5cdd0b PR |
256 | if (tconn->newest_tle != new) { |
257 | tconn->newest_tle->next = new; | |
258 | tconn->newest_tle = new; | |
b411b363 PR |
259 | } |
260 | } | |
261 | ||
262 | /** | |
263 | * tl_release() - Free or recycle the oldest &struct drbd_tl_epoch object of the TL | |
264 | * @mdev: DRBD device. | |
265 | * @barrier_nr: Expected identifier of the DRBD write barrier packet. | |
266 | * @set_size: Expected number of requests before that barrier. | |
267 | * | |
268 | * In case the passed barrier_nr or set_size does not match the oldest | |
269 | * &struct drbd_tl_epoch objects this function will cause a termination | |
270 | * of the connection. | |
271 | */ | |
2f5cdd0b PR |
272 | void tl_release(struct drbd_tconn *tconn, unsigned int barrier_nr, |
273 | unsigned int set_size) | |
b411b363 | 274 | { |
2f5cdd0b | 275 | struct drbd_conf *mdev; |
b411b363 PR |
276 | struct drbd_tl_epoch *b, *nob; /* next old barrier */ |
277 | struct list_head *le, *tle; | |
278 | struct drbd_request *r; | |
279 | ||
2f5cdd0b | 280 | spin_lock_irq(&tconn->req_lock); |
b411b363 | 281 | |
2f5cdd0b | 282 | b = tconn->oldest_tle; |
b411b363 PR |
283 | |
284 | /* first some paranoia code */ | |
285 | if (b == NULL) { | |
2f5cdd0b PR |
286 | conn_err(tconn, "BAD! BarrierAck #%u received, but no epoch in tl!?\n", |
287 | barrier_nr); | |
b411b363 PR |
288 | goto bail; |
289 | } | |
290 | if (b->br_number != barrier_nr) { | |
2f5cdd0b PR |
291 | conn_err(tconn, "BAD! BarrierAck #%u received, expected #%u!\n", |
292 | barrier_nr, b->br_number); | |
b411b363 PR |
293 | goto bail; |
294 | } | |
7e602c0a | 295 | if (b->n_writes != set_size) { |
2f5cdd0b PR |
296 | conn_err(tconn, "BAD! BarrierAck #%u received with n_writes=%u, expected n_writes=%u!\n", |
297 | barrier_nr, set_size, b->n_writes); | |
b411b363 PR |
298 | goto bail; |
299 | } | |
300 | ||
301 | /* Clean up list of requests processed during current epoch */ | |
302 | list_for_each_safe(le, tle, &b->requests) { | |
303 | r = list_entry(le, struct drbd_request, tl_requests); | |
8554df1c | 304 | _req_mod(r, BARRIER_ACKED); |
b411b363 PR |
305 | } |
306 | /* There could be requests on the list waiting for completion | |
307 | of the write to the local disk. To avoid corruptions of | |
308 | slab's data structures we have to remove the lists head. | |
309 | ||
310 | Also there could have been a barrier ack out of sequence, overtaking | |
311 | the write acks - which would be a bug and violating write ordering. | |
312 | To not deadlock in case we lose connection while such requests are | |
313 | still pending, we need some way to find them for the | |
8554df1c | 314 | _req_mode(CONNECTION_LOST_WHILE_PENDING). |
b411b363 PR |
315 | |
316 | These have been list_move'd to the out_of_sequence_requests list in | |
8554df1c | 317 | _req_mod(, BARRIER_ACKED) above. |
b411b363 | 318 | */ |
cdfda633 | 319 | list_splice_init(&b->requests, &tconn->barrier_acked_requests); |
2f5cdd0b | 320 | mdev = b->w.mdev; |
b411b363 PR |
321 | |
322 | nob = b->next; | |
6936fcb4 | 323 | if (test_and_clear_bit(CREATE_BARRIER, &tconn->flags)) { |
2f5cdd0b | 324 | _tl_add_barrier(tconn, b); |
b411b363 | 325 | if (nob) |
2f5cdd0b | 326 | tconn->oldest_tle = nob; |
b411b363 | 327 | /* if nob == NULL b was the only barrier, and becomes the new |
2f5cdd0b | 328 | barrier. Therefore tconn->oldest_tle points already to b */ |
b411b363 PR |
329 | } else { |
330 | D_ASSERT(nob != NULL); | |
2f5cdd0b | 331 | tconn->oldest_tle = nob; |
b411b363 PR |
332 | kfree(b); |
333 | } | |
334 | ||
2f5cdd0b | 335 | spin_unlock_irq(&tconn->req_lock); |
b411b363 PR |
336 | dec_ap_pending(mdev); |
337 | ||
338 | return; | |
339 | ||
340 | bail: | |
2f5cdd0b PR |
341 | spin_unlock_irq(&tconn->req_lock); |
342 | conn_request_state(tconn, NS(conn, C_PROTOCOL_ERROR), CS_HARD); | |
b411b363 PR |
343 | } |
344 | ||
617049aa | 345 | |
b411b363 | 346 | /** |
11b58e73 | 347 | * _tl_restart() - Walks the transfer log, and applies an action to all requests |
b411b363 | 348 | * @mdev: DRBD device. |
11b58e73 | 349 | * @what: The action/event to perform with all request objects |
b411b363 | 350 | * |
8554df1c AG |
351 | * @what might be one of CONNECTION_LOST_WHILE_PENDING, RESEND, FAIL_FROZEN_DISK_IO, |
352 | * RESTART_FROZEN_DISK_IO. | |
b411b363 | 353 | */ |
2f5cdd0b | 354 | void _tl_restart(struct drbd_tconn *tconn, enum drbd_req_event what) |
b411b363 | 355 | { |
11b58e73 | 356 | struct drbd_tl_epoch *b, *tmp, **pn; |
b9b98716 | 357 | struct list_head *le, *tle, carry_reads; |
11b58e73 PR |
358 | struct drbd_request *req; |
359 | int rv, n_writes, n_reads; | |
b411b363 | 360 | |
2f5cdd0b PR |
361 | b = tconn->oldest_tle; |
362 | pn = &tconn->oldest_tle; | |
b411b363 | 363 | while (b) { |
11b58e73 PR |
364 | n_writes = 0; |
365 | n_reads = 0; | |
b9b98716 | 366 | INIT_LIST_HEAD(&carry_reads); |
b411b363 | 367 | list_for_each_safe(le, tle, &b->requests) { |
11b58e73 PR |
368 | req = list_entry(le, struct drbd_request, tl_requests); |
369 | rv = _req_mod(req, what); | |
370 | ||
f497609e AG |
371 | if (rv & MR_WRITE) |
372 | n_writes++; | |
373 | if (rv & MR_READ) | |
374 | n_reads++; | |
b411b363 PR |
375 | } |
376 | tmp = b->next; | |
377 | ||
b9b98716 | 378 | if (n_writes) { |
8554df1c | 379 | if (what == RESEND) { |
11b58e73 PR |
380 | b->n_writes = n_writes; |
381 | if (b->w.cb == NULL) { | |
382 | b->w.cb = w_send_barrier; | |
2f5cdd0b | 383 | inc_ap_pending(b->w.mdev); |
6936fcb4 | 384 | set_bit(CREATE_BARRIER, &tconn->flags); |
11b58e73 PR |
385 | } |
386 | ||
2f5cdd0b | 387 | drbd_queue_work(&tconn->data.work, &b->w); |
11b58e73 PR |
388 | } |
389 | pn = &b->next; | |
390 | } else { | |
b9b98716 PR |
391 | if (n_reads) |
392 | list_add(&carry_reads, &b->requests); | |
11b58e73 PR |
393 | /* there could still be requests on that ring list, |
394 | * in case local io is still pending */ | |
395 | list_del(&b->requests); | |
396 | ||
397 | /* dec_ap_pending corresponding to queue_barrier. | |
398 | * the newest barrier may not have been queued yet, | |
399 | * in which case w.cb is still NULL. */ | |
400 | if (b->w.cb != NULL) | |
2f5cdd0b | 401 | dec_ap_pending(b->w.mdev); |
11b58e73 | 402 | |
2f5cdd0b | 403 | if (b == tconn->newest_tle) { |
11b58e73 | 404 | /* recycle, but reinit! */ |
2f5cdd0b PR |
405 | if (tmp != NULL) |
406 | conn_err(tconn, "ASSERT FAILED tmp == NULL"); | |
11b58e73 | 407 | INIT_LIST_HEAD(&b->requests); |
b9b98716 | 408 | list_splice(&carry_reads, &b->requests); |
11b58e73 PR |
409 | INIT_LIST_HEAD(&b->w.list); |
410 | b->w.cb = NULL; | |
411 | b->br_number = net_random(); | |
412 | b->n_writes = 0; | |
413 | ||
414 | *pn = b; | |
415 | break; | |
416 | } | |
417 | *pn = tmp; | |
418 | kfree(b); | |
b411b363 | 419 | } |
b411b363 | 420 | b = tmp; |
b9b98716 | 421 | list_splice(&carry_reads, &b->requests); |
b411b363 | 422 | } |
11b58e73 | 423 | |
cdfda633 PR |
424 | /* Actions operating on the disk state, also want to work on |
425 | requests that got barrier acked. */ | |
426 | switch (what) { | |
427 | case FAIL_FROZEN_DISK_IO: | |
428 | case RESTART_FROZEN_DISK_IO: | |
429 | list_for_each_safe(le, tle, &tconn->barrier_acked_requests) { | |
430 | req = list_entry(le, struct drbd_request, tl_requests); | |
431 | _req_mod(req, what); | |
432 | } | |
433 | case CONNECTION_LOST_WHILE_PENDING: | |
434 | case RESEND: | |
435 | break; | |
436 | default: | |
437 | conn_err(tconn, "what = %d in _tl_restart()\n", what); | |
438 | } | |
439 | } | |
b411b363 PR |
440 | |
441 | /** | |
442 | * tl_clear() - Clears all requests and &struct drbd_tl_epoch objects out of the TL | |
443 | * @mdev: DRBD device. | |
444 | * | |
445 | * This is called after the connection to the peer was lost. The storage covered | |
446 | * by the requests on the transfer gets marked as our of sync. Called from the | |
447 | * receiver thread and the worker thread. | |
448 | */ | |
2f5cdd0b | 449 | void tl_clear(struct drbd_tconn *tconn) |
b411b363 | 450 | { |
b411b363 PR |
451 | struct list_head *le, *tle; |
452 | struct drbd_request *r; | |
b411b363 | 453 | |
2f5cdd0b | 454 | spin_lock_irq(&tconn->req_lock); |
b411b363 | 455 | |
2f5cdd0b | 456 | _tl_restart(tconn, CONNECTION_LOST_WHILE_PENDING); |
b411b363 PR |
457 | |
458 | /* we expect this list to be empty. */ | |
2f5cdd0b PR |
459 | if (!list_empty(&tconn->out_of_sequence_requests)) |
460 | conn_err(tconn, "ASSERT FAILED list_empty(&out_of_sequence_requests)\n"); | |
b411b363 PR |
461 | |
462 | /* but just in case, clean it up anyways! */ | |
2f5cdd0b | 463 | list_for_each_safe(le, tle, &tconn->out_of_sequence_requests) { |
b411b363 PR |
464 | r = list_entry(le, struct drbd_request, tl_requests); |
465 | /* It would be nice to complete outside of spinlock. | |
466 | * But this is easier for now. */ | |
8554df1c | 467 | _req_mod(r, CONNECTION_LOST_WHILE_PENDING); |
b411b363 PR |
468 | } |
469 | ||
470 | /* ensure bit indicating barrier is required is clear */ | |
6936fcb4 | 471 | clear_bit(CREATE_BARRIER, &tconn->flags); |
b411b363 | 472 | |
2f5cdd0b | 473 | spin_unlock_irq(&tconn->req_lock); |
b411b363 PR |
474 | } |
475 | ||
2f5cdd0b | 476 | void tl_restart(struct drbd_tconn *tconn, enum drbd_req_event what) |
11b58e73 | 477 | { |
2f5cdd0b PR |
478 | spin_lock_irq(&tconn->req_lock); |
479 | _tl_restart(tconn, what); | |
480 | spin_unlock_irq(&tconn->req_lock); | |
b411b363 PR |
481 | } |
482 | ||
cdfda633 | 483 | /** |
71fc7eed | 484 | * tl_abort_disk_io() - Abort disk I/O for all requests for a certain mdev in the TL |
cdfda633 | 485 | * @mdev: DRBD device. |
cdfda633 | 486 | */ |
71fc7eed | 487 | void tl_abort_disk_io(struct drbd_conf *mdev) |
cdfda633 PR |
488 | { |
489 | struct drbd_tconn *tconn = mdev->tconn; | |
490 | struct drbd_tl_epoch *b; | |
491 | struct list_head *le, *tle; | |
492 | struct drbd_request *req; | |
493 | ||
cdfda633 PR |
494 | spin_lock_irq(&tconn->req_lock); |
495 | b = tconn->oldest_tle; | |
496 | while (b) { | |
497 | list_for_each_safe(le, tle, &b->requests) { | |
498 | req = list_entry(le, struct drbd_request, tl_requests); | |
97ddb687 LE |
499 | if (!(req->rq_state & RQ_LOCAL_PENDING)) |
500 | continue; | |
cdfda633 | 501 | if (req->w.mdev == mdev) |
71fc7eed | 502 | _req_mod(req, ABORT_DISK_IO); |
cdfda633 PR |
503 | } |
504 | b = b->next; | |
505 | } | |
506 | ||
507 | list_for_each_safe(le, tle, &tconn->barrier_acked_requests) { | |
508 | req = list_entry(le, struct drbd_request, tl_requests); | |
97ddb687 LE |
509 | if (!(req->rq_state & RQ_LOCAL_PENDING)) |
510 | continue; | |
cdfda633 | 511 | if (req->w.mdev == mdev) |
71fc7eed | 512 | _req_mod(req, ABORT_DISK_IO); |
cdfda633 PR |
513 | } |
514 | ||
515 | spin_unlock_irq(&tconn->req_lock); | |
516 | } | |
517 | ||
b411b363 PR |
518 | static int drbd_thread_setup(void *arg) |
519 | { | |
520 | struct drbd_thread *thi = (struct drbd_thread *) arg; | |
392c8801 | 521 | struct drbd_tconn *tconn = thi->tconn; |
b411b363 PR |
522 | unsigned long flags; |
523 | int retval; | |
524 | ||
f1b3a6ec | 525 | snprintf(current->comm, sizeof(current->comm), "drbd_%c_%s", |
392c8801 | 526 | thi->name[0], thi->tconn->name); |
f1b3a6ec | 527 | |
b411b363 PR |
528 | restart: |
529 | retval = thi->function(thi); | |
530 | ||
531 | spin_lock_irqsave(&thi->t_lock, flags); | |
532 | ||
e77a0a5c | 533 | /* if the receiver has been "EXITING", the last thing it did |
b411b363 PR |
534 | * was set the conn state to "StandAlone", |
535 | * if now a re-connect request comes in, conn state goes C_UNCONNECTED, | |
536 | * and receiver thread will be "started". | |
e77a0a5c | 537 | * drbd_thread_start needs to set "RESTARTING" in that case. |
b411b363 | 538 | * t_state check and assignment needs to be within the same spinlock, |
e77a0a5c AG |
539 | * so either thread_start sees EXITING, and can remap to RESTARTING, |
540 | * or thread_start see NONE, and can proceed as normal. | |
b411b363 PR |
541 | */ |
542 | ||
e77a0a5c | 543 | if (thi->t_state == RESTARTING) { |
392c8801 | 544 | conn_info(tconn, "Restarting %s thread\n", thi->name); |
e77a0a5c | 545 | thi->t_state = RUNNING; |
b411b363 PR |
546 | spin_unlock_irqrestore(&thi->t_lock, flags); |
547 | goto restart; | |
548 | } | |
549 | ||
550 | thi->task = NULL; | |
e77a0a5c | 551 | thi->t_state = NONE; |
b411b363 | 552 | smp_mb(); |
992d6e91 | 553 | complete_all(&thi->stop); |
b411b363 PR |
554 | spin_unlock_irqrestore(&thi->t_lock, flags); |
555 | ||
392c8801 | 556 | conn_info(tconn, "Terminating %s\n", current->comm); |
b411b363 PR |
557 | |
558 | /* Release mod reference taken when thread was started */ | |
9dc9fbb3 PR |
559 | |
560 | kref_put(&tconn->kref, &conn_destroy); | |
b411b363 PR |
561 | module_put(THIS_MODULE); |
562 | return retval; | |
563 | } | |
564 | ||
392c8801 | 565 | static void drbd_thread_init(struct drbd_tconn *tconn, struct drbd_thread *thi, |
bed879ae | 566 | int (*func) (struct drbd_thread *), char *name) |
b411b363 PR |
567 | { |
568 | spin_lock_init(&thi->t_lock); | |
569 | thi->task = NULL; | |
e77a0a5c | 570 | thi->t_state = NONE; |
b411b363 | 571 | thi->function = func; |
392c8801 | 572 | thi->tconn = tconn; |
bed879ae | 573 | strncpy(thi->name, name, ARRAY_SIZE(thi->name)); |
b411b363 PR |
574 | } |
575 | ||
576 | int drbd_thread_start(struct drbd_thread *thi) | |
577 | { | |
392c8801 | 578 | struct drbd_tconn *tconn = thi->tconn; |
b411b363 PR |
579 | struct task_struct *nt; |
580 | unsigned long flags; | |
581 | ||
b411b363 PR |
582 | /* is used from state engine doing drbd_thread_stop_nowait, |
583 | * while holding the req lock irqsave */ | |
584 | spin_lock_irqsave(&thi->t_lock, flags); | |
585 | ||
586 | switch (thi->t_state) { | |
e77a0a5c | 587 | case NONE: |
392c8801 | 588 | conn_info(tconn, "Starting %s thread (from %s [%d])\n", |
bed879ae | 589 | thi->name, current->comm, current->pid); |
b411b363 PR |
590 | |
591 | /* Get ref on module for thread - this is released when thread exits */ | |
592 | if (!try_module_get(THIS_MODULE)) { | |
392c8801 | 593 | conn_err(tconn, "Failed to get module reference in drbd_thread_start\n"); |
b411b363 | 594 | spin_unlock_irqrestore(&thi->t_lock, flags); |
81e84650 | 595 | return false; |
b411b363 PR |
596 | } |
597 | ||
9dc9fbb3 PR |
598 | kref_get(&thi->tconn->kref); |
599 | ||
b411b363 | 600 | init_completion(&thi->stop); |
b411b363 | 601 | thi->reset_cpu_mask = 1; |
e77a0a5c | 602 | thi->t_state = RUNNING; |
b411b363 PR |
603 | spin_unlock_irqrestore(&thi->t_lock, flags); |
604 | flush_signals(current); /* otherw. may get -ERESTARTNOINTR */ | |
605 | ||
606 | nt = kthread_create(drbd_thread_setup, (void *) thi, | |
392c8801 | 607 | "drbd_%c_%s", thi->name[0], thi->tconn->name); |
b411b363 PR |
608 | |
609 | if (IS_ERR(nt)) { | |
392c8801 | 610 | conn_err(tconn, "Couldn't start thread\n"); |
b411b363 | 611 | |
9dc9fbb3 | 612 | kref_put(&tconn->kref, &conn_destroy); |
b411b363 | 613 | module_put(THIS_MODULE); |
81e84650 | 614 | return false; |
b411b363 PR |
615 | } |
616 | spin_lock_irqsave(&thi->t_lock, flags); | |
617 | thi->task = nt; | |
e77a0a5c | 618 | thi->t_state = RUNNING; |
b411b363 PR |
619 | spin_unlock_irqrestore(&thi->t_lock, flags); |
620 | wake_up_process(nt); | |
621 | break; | |
e77a0a5c AG |
622 | case EXITING: |
623 | thi->t_state = RESTARTING; | |
392c8801 | 624 | conn_info(tconn, "Restarting %s thread (from %s [%d])\n", |
bed879ae | 625 | thi->name, current->comm, current->pid); |
b411b363 | 626 | /* fall through */ |
e77a0a5c AG |
627 | case RUNNING: |
628 | case RESTARTING: | |
b411b363 PR |
629 | default: |
630 | spin_unlock_irqrestore(&thi->t_lock, flags); | |
631 | break; | |
632 | } | |
633 | ||
81e84650 | 634 | return true; |
b411b363 PR |
635 | } |
636 | ||
637 | ||
638 | void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait) | |
639 | { | |
640 | unsigned long flags; | |
641 | ||
e77a0a5c | 642 | enum drbd_thread_state ns = restart ? RESTARTING : EXITING; |
b411b363 PR |
643 | |
644 | /* may be called from state engine, holding the req lock irqsave */ | |
645 | spin_lock_irqsave(&thi->t_lock, flags); | |
646 | ||
e77a0a5c | 647 | if (thi->t_state == NONE) { |
b411b363 PR |
648 | spin_unlock_irqrestore(&thi->t_lock, flags); |
649 | if (restart) | |
650 | drbd_thread_start(thi); | |
651 | return; | |
652 | } | |
653 | ||
654 | if (thi->t_state != ns) { | |
655 | if (thi->task == NULL) { | |
656 | spin_unlock_irqrestore(&thi->t_lock, flags); | |
657 | return; | |
658 | } | |
659 | ||
660 | thi->t_state = ns; | |
661 | smp_mb(); | |
662 | init_completion(&thi->stop); | |
663 | if (thi->task != current) | |
664 | force_sig(DRBD_SIGKILL, thi->task); | |
b411b363 PR |
665 | } |
666 | ||
667 | spin_unlock_irqrestore(&thi->t_lock, flags); | |
668 | ||
669 | if (wait) | |
670 | wait_for_completion(&thi->stop); | |
671 | } | |
672 | ||
392c8801 | 673 | static struct drbd_thread *drbd_task_to_thread(struct drbd_tconn *tconn, struct task_struct *task) |
bed879ae | 674 | { |
bed879ae PR |
675 | struct drbd_thread *thi = |
676 | task == tconn->receiver.task ? &tconn->receiver : | |
677 | task == tconn->asender.task ? &tconn->asender : | |
678 | task == tconn->worker.task ? &tconn->worker : NULL; | |
679 | ||
680 | return thi; | |
681 | } | |
682 | ||
392c8801 | 683 | char *drbd_task_to_thread_name(struct drbd_tconn *tconn, struct task_struct *task) |
bed879ae | 684 | { |
392c8801 | 685 | struct drbd_thread *thi = drbd_task_to_thread(tconn, task); |
bed879ae PR |
686 | return thi ? thi->name : task->comm; |
687 | } | |
688 | ||
80883197 | 689 | int conn_lowest_minor(struct drbd_tconn *tconn) |
80822284 | 690 | { |
e90285e0 | 691 | struct drbd_conf *mdev; |
695d08fa | 692 | int vnr = 0, m; |
774b3055 | 693 | |
695d08fa | 694 | rcu_read_lock(); |
e90285e0 | 695 | mdev = idr_get_next(&tconn->volumes, &vnr); |
695d08fa PR |
696 | m = mdev ? mdev_to_minor(mdev) : -1; |
697 | rcu_read_unlock(); | |
698 | ||
699 | return m; | |
80822284 | 700 | } |
774b3055 PR |
701 | |
702 | #ifdef CONFIG_SMP | |
b411b363 PR |
703 | /** |
704 | * drbd_calc_cpu_mask() - Generate CPU masks, spread over all CPUs | |
705 | * @mdev: DRBD device. | |
706 | * | |
707 | * Forces all threads of a device onto the same CPU. This is beneficial for | |
708 | * DRBD's performance. May be overwritten by user's configuration. | |
709 | */ | |
80822284 | 710 | void drbd_calc_cpu_mask(struct drbd_tconn *tconn) |
b411b363 PR |
711 | { |
712 | int ord, cpu; | |
713 | ||
714 | /* user override. */ | |
80822284 | 715 | if (cpumask_weight(tconn->cpu_mask)) |
b411b363 PR |
716 | return; |
717 | ||
80822284 | 718 | ord = conn_lowest_minor(tconn) % cpumask_weight(cpu_online_mask); |
b411b363 PR |
719 | for_each_online_cpu(cpu) { |
720 | if (ord-- == 0) { | |
80822284 | 721 | cpumask_set_cpu(cpu, tconn->cpu_mask); |
b411b363 PR |
722 | return; |
723 | } | |
724 | } | |
725 | /* should not be reached */ | |
80822284 | 726 | cpumask_setall(tconn->cpu_mask); |
b411b363 PR |
727 | } |
728 | ||
729 | /** | |
730 | * drbd_thread_current_set_cpu() - modifies the cpu mask of the _current_ thread | |
731 | * @mdev: DRBD device. | |
bc31fe33 | 732 | * @thi: drbd_thread object |
b411b363 PR |
733 | * |
734 | * call in the "main loop" of _all_ threads, no need for any mutex, current won't die | |
735 | * prematurely. | |
736 | */ | |
80822284 | 737 | void drbd_thread_current_set_cpu(struct drbd_thread *thi) |
b411b363 PR |
738 | { |
739 | struct task_struct *p = current; | |
bed879ae | 740 | |
b411b363 PR |
741 | if (!thi->reset_cpu_mask) |
742 | return; | |
743 | thi->reset_cpu_mask = 0; | |
392c8801 | 744 | set_cpus_allowed_ptr(p, thi->tconn->cpu_mask); |
b411b363 PR |
745 | } |
746 | #endif | |
747 | ||
52b061a4 AG |
748 | /** |
749 | * drbd_header_size - size of a packet header | |
750 | * | |
751 | * The header size is a multiple of 8, so any payload following the header is | |
752 | * word aligned on 64-bit architectures. (The bitmap send and receive code | |
753 | * relies on this.) | |
754 | */ | |
755 | unsigned int drbd_header_size(struct drbd_tconn *tconn) | |
756 | { | |
0c8e36d9 AG |
757 | if (tconn->agreed_pro_version >= 100) { |
758 | BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header100), 8)); | |
759 | return sizeof(struct p_header100); | |
760 | } else { | |
761 | BUILD_BUG_ON(sizeof(struct p_header80) != | |
762 | sizeof(struct p_header95)); | |
763 | BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header80), 8)); | |
764 | return sizeof(struct p_header80); | |
765 | } | |
52b061a4 AG |
766 | } |
767 | ||
e658983a | 768 | static unsigned int prepare_header80(struct p_header80 *h, enum drbd_packet cmd, int size) |
fd340c12 PR |
769 | { |
770 | h->magic = cpu_to_be32(DRBD_MAGIC); | |
771 | h->command = cpu_to_be16(cmd); | |
772 | h->length = cpu_to_be16(size); | |
e658983a | 773 | return sizeof(struct p_header80); |
fd340c12 PR |
774 | } |
775 | ||
e658983a | 776 | static unsigned int prepare_header95(struct p_header95 *h, enum drbd_packet cmd, int size) |
fd340c12 PR |
777 | { |
778 | h->magic = cpu_to_be16(DRBD_MAGIC_BIG); | |
779 | h->command = cpu_to_be16(cmd); | |
b55d84ba | 780 | h->length = cpu_to_be32(size); |
e658983a | 781 | return sizeof(struct p_header95); |
fd340c12 PR |
782 | } |
783 | ||
0c8e36d9 AG |
784 | static unsigned int prepare_header100(struct p_header100 *h, enum drbd_packet cmd, |
785 | int size, int vnr) | |
786 | { | |
787 | h->magic = cpu_to_be32(DRBD_MAGIC_100); | |
788 | h->volume = cpu_to_be16(vnr); | |
789 | h->command = cpu_to_be16(cmd); | |
790 | h->length = cpu_to_be32(size); | |
791 | h->pad = 0; | |
792 | return sizeof(struct p_header100); | |
793 | } | |
794 | ||
795 | static unsigned int prepare_header(struct drbd_tconn *tconn, int vnr, | |
796 | void *buffer, enum drbd_packet cmd, int size) | |
d38e787e | 797 | { |
0c8e36d9 AG |
798 | if (tconn->agreed_pro_version >= 100) |
799 | return prepare_header100(buffer, cmd, size, vnr); | |
800 | else if (tconn->agreed_pro_version >= 95 && | |
801 | size > DRBD_MAX_SIZE_H80_PACKET) | |
e658983a | 802 | return prepare_header95(buffer, cmd, size); |
d38e787e | 803 | else |
e658983a | 804 | return prepare_header80(buffer, cmd, size); |
d38e787e PR |
805 | } |
806 | ||
a7eb7bdf AG |
807 | static void *__conn_prepare_command(struct drbd_tconn *tconn, |
808 | struct drbd_socket *sock) | |
809 | { | |
810 | if (!sock->socket) | |
811 | return NULL; | |
812 | return sock->sbuf + drbd_header_size(tconn); | |
813 | } | |
814 | ||
dba58587 AG |
815 | void *conn_prepare_command(struct drbd_tconn *tconn, struct drbd_socket *sock) |
816 | { | |
a7eb7bdf AG |
817 | void *p; |
818 | ||
dba58587 | 819 | mutex_lock(&sock->mutex); |
a7eb7bdf AG |
820 | p = __conn_prepare_command(tconn, sock); |
821 | if (!p) | |
dba58587 | 822 | mutex_unlock(&sock->mutex); |
a7eb7bdf AG |
823 | |
824 | return p; | |
dba58587 AG |
825 | } |
826 | ||
827 | void *drbd_prepare_command(struct drbd_conf *mdev, struct drbd_socket *sock) | |
828 | { | |
829 | return conn_prepare_command(mdev->tconn, sock); | |
830 | } | |
831 | ||
832 | static int __send_command(struct drbd_tconn *tconn, int vnr, | |
833 | struct drbd_socket *sock, enum drbd_packet cmd, | |
834 | unsigned int header_size, void *data, | |
835 | unsigned int size) | |
836 | { | |
837 | int msg_flags; | |
838 | int err; | |
839 | ||
840 | /* | |
841 | * Called with @data == NULL and the size of the data blocks in @size | |
842 | * for commands that send data blocks. For those commands, omit the | |
843 | * MSG_MORE flag: this will increase the likelihood that data blocks | |
844 | * which are page aligned on the sender will end up page aligned on the | |
845 | * receiver. | |
846 | */ | |
847 | msg_flags = data ? MSG_MORE : 0; | |
848 | ||
e658983a AG |
849 | header_size += prepare_header(tconn, vnr, sock->sbuf, cmd, |
850 | header_size + size); | |
dba58587 AG |
851 | err = drbd_send_all(tconn, sock->socket, sock->sbuf, header_size, |
852 | msg_flags); | |
853 | if (data && !err) | |
854 | err = drbd_send_all(tconn, sock->socket, data, size, 0); | |
855 | return err; | |
856 | } | |
857 | ||
a7eb7bdf AG |
858 | static int __conn_send_command(struct drbd_tconn *tconn, struct drbd_socket *sock, |
859 | enum drbd_packet cmd, unsigned int header_size, | |
860 | void *data, unsigned int size) | |
861 | { | |
862 | return __send_command(tconn, 0, sock, cmd, header_size, data, size); | |
863 | } | |
864 | ||
dba58587 AG |
865 | int conn_send_command(struct drbd_tconn *tconn, struct drbd_socket *sock, |
866 | enum drbd_packet cmd, unsigned int header_size, | |
867 | void *data, unsigned int size) | |
868 | { | |
869 | int err; | |
870 | ||
a7eb7bdf | 871 | err = __conn_send_command(tconn, sock, cmd, header_size, data, size); |
dba58587 AG |
872 | mutex_unlock(&sock->mutex); |
873 | return err; | |
874 | } | |
875 | ||
876 | int drbd_send_command(struct drbd_conf *mdev, struct drbd_socket *sock, | |
877 | enum drbd_packet cmd, unsigned int header_size, | |
878 | void *data, unsigned int size) | |
879 | { | |
880 | int err; | |
881 | ||
882 | err = __send_command(mdev->tconn, mdev->vnr, sock, cmd, header_size, | |
883 | data, size); | |
884 | mutex_unlock(&sock->mutex); | |
885 | return err; | |
886 | } | |
887 | ||
e307f352 AG |
888 | int drbd_send_ping(struct drbd_tconn *tconn) |
889 | { | |
9f5bdc33 AG |
890 | struct drbd_socket *sock; |
891 | ||
892 | sock = &tconn->meta; | |
893 | if (!conn_prepare_command(tconn, sock)) | |
894 | return -EIO; | |
e658983a | 895 | return conn_send_command(tconn, sock, P_PING, 0, NULL, 0); |
e307f352 AG |
896 | } |
897 | ||
898 | int drbd_send_ping_ack(struct drbd_tconn *tconn) | |
899 | { | |
9f5bdc33 AG |
900 | struct drbd_socket *sock; |
901 | ||
902 | sock = &tconn->meta; | |
903 | if (!conn_prepare_command(tconn, sock)) | |
904 | return -EIO; | |
e658983a | 905 | return conn_send_command(tconn, sock, P_PING_ACK, 0, NULL, 0); |
e307f352 AG |
906 | } |
907 | ||
f399002e | 908 | int drbd_send_sync_param(struct drbd_conf *mdev) |
b411b363 | 909 | { |
7c96715a | 910 | struct drbd_socket *sock; |
9f5bdc33 AG |
911 | struct p_rs_param_95 *p; |
912 | int size; | |
31890f4a | 913 | const int apv = mdev->tconn->agreed_pro_version; |
9f5bdc33 | 914 | enum drbd_packet cmd; |
44ed167d | 915 | struct net_conf *nc; |
daeda1cc | 916 | struct disk_conf *dc; |
9f5bdc33 AG |
917 | |
918 | sock = &mdev->tconn->data; | |
919 | p = drbd_prepare_command(mdev, sock); | |
920 | if (!p) | |
921 | return -EIO; | |
b411b363 | 922 | |
44ed167d PR |
923 | rcu_read_lock(); |
924 | nc = rcu_dereference(mdev->tconn->net_conf); | |
925 | ||
b411b363 PR |
926 | size = apv <= 87 ? sizeof(struct p_rs_param) |
927 | : apv == 88 ? sizeof(struct p_rs_param) | |
44ed167d | 928 | + strlen(nc->verify_alg) + 1 |
8e26f9cc PR |
929 | : apv <= 94 ? sizeof(struct p_rs_param_89) |
930 | : /* apv >= 95 */ sizeof(struct p_rs_param_95); | |
b411b363 | 931 | |
9f5bdc33 | 932 | cmd = apv >= 89 ? P_SYNC_PARAM89 : P_SYNC_PARAM; |
b411b363 | 933 | |
9f5bdc33 AG |
934 | /* initialize verify_alg and csums_alg */ |
935 | memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX); | |
b411b363 | 936 | |
9f5bdc33 | 937 | if (get_ldev(mdev)) { |
daeda1cc | 938 | dc = rcu_dereference(mdev->ldev->disk_conf); |
6394b935 | 939 | p->resync_rate = cpu_to_be32(dc->resync_rate); |
daeda1cc PR |
940 | p->c_plan_ahead = cpu_to_be32(dc->c_plan_ahead); |
941 | p->c_delay_target = cpu_to_be32(dc->c_delay_target); | |
942 | p->c_fill_target = cpu_to_be32(dc->c_fill_target); | |
943 | p->c_max_rate = cpu_to_be32(dc->c_max_rate); | |
9f5bdc33 AG |
944 | put_ldev(mdev); |
945 | } else { | |
6394b935 | 946 | p->resync_rate = cpu_to_be32(DRBD_RESYNC_RATE_DEF); |
9f5bdc33 AG |
947 | p->c_plan_ahead = cpu_to_be32(DRBD_C_PLAN_AHEAD_DEF); |
948 | p->c_delay_target = cpu_to_be32(DRBD_C_DELAY_TARGET_DEF); | |
949 | p->c_fill_target = cpu_to_be32(DRBD_C_FILL_TARGET_DEF); | |
950 | p->c_max_rate = cpu_to_be32(DRBD_C_MAX_RATE_DEF); | |
951 | } | |
b411b363 | 952 | |
9f5bdc33 | 953 | if (apv >= 88) |
44ed167d | 954 | strcpy(p->verify_alg, nc->verify_alg); |
9f5bdc33 | 955 | if (apv >= 89) |
44ed167d PR |
956 | strcpy(p->csums_alg, nc->csums_alg); |
957 | rcu_read_unlock(); | |
b411b363 | 958 | |
9f5bdc33 | 959 | return drbd_send_command(mdev, sock, cmd, size, NULL, 0); |
b411b363 PR |
960 | } |
961 | ||
d659f2aa | 962 | int __drbd_send_protocol(struct drbd_tconn *tconn, enum drbd_packet cmd) |
b411b363 | 963 | { |
9f5bdc33 | 964 | struct drbd_socket *sock; |
b411b363 | 965 | struct p_protocol *p; |
44ed167d | 966 | struct net_conf *nc; |
9f5bdc33 | 967 | int size, cf; |
b411b363 | 968 | |
9f5bdc33 | 969 | sock = &tconn->data; |
a7eb7bdf | 970 | p = __conn_prepare_command(tconn, sock); |
9f5bdc33 AG |
971 | if (!p) |
972 | return -EIO; | |
973 | ||
44ed167d PR |
974 | rcu_read_lock(); |
975 | nc = rcu_dereference(tconn->net_conf); | |
976 | ||
6dff2902 | 977 | if (nc->tentative && tconn->agreed_pro_version < 92) { |
44ed167d PR |
978 | rcu_read_unlock(); |
979 | mutex_unlock(&sock->mutex); | |
980 | conn_err(tconn, "--dry-run is not supported by peer"); | |
981 | return -EOPNOTSUPP; | |
982 | } | |
983 | ||
9f5bdc33 | 984 | size = sizeof(*p); |
dc8228d1 | 985 | if (tconn->agreed_pro_version >= 87) |
44ed167d | 986 | size += strlen(nc->integrity_alg) + 1; |
b411b363 | 987 | |
44ed167d PR |
988 | p->protocol = cpu_to_be32(nc->wire_protocol); |
989 | p->after_sb_0p = cpu_to_be32(nc->after_sb_0p); | |
990 | p->after_sb_1p = cpu_to_be32(nc->after_sb_1p); | |
991 | p->after_sb_2p = cpu_to_be32(nc->after_sb_2p); | |
992 | p->two_primaries = cpu_to_be32(nc->two_primaries); | |
cf14c2e9 | 993 | cf = 0; |
6139f60d AG |
994 | if (nc->discard_my_data) |
995 | cf |= CF_DISCARD_MY_DATA; | |
6dff2902 | 996 | if (nc->tentative) |
9f5bdc33 | 997 | cf |= CF_DRY_RUN; |
cf14c2e9 PR |
998 | p->conn_flags = cpu_to_be32(cf); |
999 | ||
dc8228d1 | 1000 | if (tconn->agreed_pro_version >= 87) |
44ed167d PR |
1001 | strcpy(p->integrity_alg, nc->integrity_alg); |
1002 | rcu_read_unlock(); | |
1003 | ||
d659f2aa | 1004 | return __conn_send_command(tconn, sock, cmd, size, NULL, 0); |
a7eb7bdf AG |
1005 | } |
1006 | ||
1007 | int drbd_send_protocol(struct drbd_tconn *tconn) | |
1008 | { | |
1009 | int err; | |
1010 | ||
1011 | mutex_lock(&tconn->data.mutex); | |
d659f2aa | 1012 | err = __drbd_send_protocol(tconn, P_PROTOCOL); |
a7eb7bdf AG |
1013 | mutex_unlock(&tconn->data.mutex); |
1014 | ||
1015 | return err; | |
b411b363 PR |
1016 | } |
1017 | ||
1018 | int _drbd_send_uuids(struct drbd_conf *mdev, u64 uuid_flags) | |
1019 | { | |
9f5bdc33 AG |
1020 | struct drbd_socket *sock; |
1021 | struct p_uuids *p; | |
b411b363 PR |
1022 | int i; |
1023 | ||
1024 | if (!get_ldev_if_state(mdev, D_NEGOTIATING)) | |
2ae5f95b | 1025 | return 0; |
b411b363 | 1026 | |
9f5bdc33 AG |
1027 | sock = &mdev->tconn->data; |
1028 | p = drbd_prepare_command(mdev, sock); | |
1029 | if (!p) { | |
1030 | put_ldev(mdev); | |
1031 | return -EIO; | |
1032 | } | |
b411b363 | 1033 | for (i = UI_CURRENT; i < UI_SIZE; i++) |
9f5bdc33 | 1034 | p->uuid[i] = mdev->ldev ? cpu_to_be64(mdev->ldev->md.uuid[i]) : 0; |
b411b363 PR |
1035 | |
1036 | mdev->comm_bm_set = drbd_bm_total_weight(mdev); | |
9f5bdc33 | 1037 | p->uuid[UI_SIZE] = cpu_to_be64(mdev->comm_bm_set); |
44ed167d | 1038 | rcu_read_lock(); |
6139f60d | 1039 | uuid_flags |= rcu_dereference(mdev->tconn->net_conf)->discard_my_data ? 1 : 0; |
44ed167d | 1040 | rcu_read_unlock(); |
b411b363 PR |
1041 | uuid_flags |= test_bit(CRASHED_PRIMARY, &mdev->flags) ? 2 : 0; |
1042 | uuid_flags |= mdev->new_state_tmp.disk == D_INCONSISTENT ? 4 : 0; | |
9f5bdc33 | 1043 | p->uuid[UI_FLAGS] = cpu_to_be64(uuid_flags); |
b411b363 PR |
1044 | |
1045 | put_ldev(mdev); | |
9f5bdc33 | 1046 | return drbd_send_command(mdev, sock, P_UUIDS, sizeof(*p), NULL, 0); |
b411b363 PR |
1047 | } |
1048 | ||
1049 | int drbd_send_uuids(struct drbd_conf *mdev) | |
1050 | { | |
1051 | return _drbd_send_uuids(mdev, 0); | |
1052 | } | |
1053 | ||
1054 | int drbd_send_uuids_skip_initial_sync(struct drbd_conf *mdev) | |
1055 | { | |
1056 | return _drbd_send_uuids(mdev, 8); | |
1057 | } | |
1058 | ||
62b0da3a LE |
1059 | void drbd_print_uuids(struct drbd_conf *mdev, const char *text) |
1060 | { | |
1061 | if (get_ldev_if_state(mdev, D_NEGOTIATING)) { | |
1062 | u64 *uuid = mdev->ldev->md.uuid; | |
1063 | dev_info(DEV, "%s %016llX:%016llX:%016llX:%016llX\n", | |
1064 | text, | |
1065 | (unsigned long long)uuid[UI_CURRENT], | |
1066 | (unsigned long long)uuid[UI_BITMAP], | |
1067 | (unsigned long long)uuid[UI_HISTORY_START], | |
1068 | (unsigned long long)uuid[UI_HISTORY_END]); | |
1069 | put_ldev(mdev); | |
1070 | } else { | |
1071 | dev_info(DEV, "%s effective data uuid: %016llX\n", | |
1072 | text, | |
1073 | (unsigned long long)mdev->ed_uuid); | |
1074 | } | |
1075 | } | |
1076 | ||
9c1b7f72 | 1077 | void drbd_gen_and_send_sync_uuid(struct drbd_conf *mdev) |
b411b363 | 1078 | { |
9f5bdc33 AG |
1079 | struct drbd_socket *sock; |
1080 | struct p_rs_uuid *p; | |
5a22db89 LE |
1081 | u64 uuid; |
1082 | ||
1083 | D_ASSERT(mdev->state.disk == D_UP_TO_DATE); | |
b411b363 | 1084 | |
0cfac5dd PR |
1085 | uuid = mdev->ldev->md.uuid[UI_BITMAP]; |
1086 | if (uuid && uuid != UUID_JUST_CREATED) | |
1087 | uuid = uuid + UUID_NEW_BM_OFFSET; | |
1088 | else | |
1089 | get_random_bytes(&uuid, sizeof(u64)); | |
5a22db89 | 1090 | drbd_uuid_set(mdev, UI_BITMAP, uuid); |
62b0da3a | 1091 | drbd_print_uuids(mdev, "updated sync UUID"); |
5a22db89 | 1092 | drbd_md_sync(mdev); |
b411b363 | 1093 | |
9f5bdc33 AG |
1094 | sock = &mdev->tconn->data; |
1095 | p = drbd_prepare_command(mdev, sock); | |
1096 | if (p) { | |
1097 | p->uuid = cpu_to_be64(uuid); | |
1098 | drbd_send_command(mdev, sock, P_SYNC_UUID, sizeof(*p), NULL, 0); | |
1099 | } | |
b411b363 PR |
1100 | } |
1101 | ||
e89b591c | 1102 | int drbd_send_sizes(struct drbd_conf *mdev, int trigger_reply, enum dds_flags flags) |
b411b363 | 1103 | { |
9f5bdc33 AG |
1104 | struct drbd_socket *sock; |
1105 | struct p_sizes *p; | |
b411b363 | 1106 | sector_t d_size, u_size; |
99432fcc | 1107 | int q_order_type, max_bio_size; |
b411b363 PR |
1108 | |
1109 | if (get_ldev_if_state(mdev, D_NEGOTIATING)) { | |
1110 | D_ASSERT(mdev->ldev->backing_bdev); | |
1111 | d_size = drbd_get_max_capacity(mdev->ldev); | |
daeda1cc PR |
1112 | rcu_read_lock(); |
1113 | u_size = rcu_dereference(mdev->ldev->disk_conf)->disk_size; | |
1114 | rcu_read_unlock(); | |
b411b363 | 1115 | q_order_type = drbd_queue_order_type(mdev); |
99432fcc PR |
1116 | max_bio_size = queue_max_hw_sectors(mdev->ldev->backing_bdev->bd_disk->queue) << 9; |
1117 | max_bio_size = min_t(int, max_bio_size, DRBD_MAX_BIO_SIZE); | |
b411b363 PR |
1118 | put_ldev(mdev); |
1119 | } else { | |
1120 | d_size = 0; | |
1121 | u_size = 0; | |
1122 | q_order_type = QUEUE_ORDERED_NONE; | |
99432fcc | 1123 | max_bio_size = DRBD_MAX_BIO_SIZE; /* ... multiple BIOs per peer_request */ |
b411b363 PR |
1124 | } |
1125 | ||
9f5bdc33 AG |
1126 | sock = &mdev->tconn->data; |
1127 | p = drbd_prepare_command(mdev, sock); | |
1128 | if (!p) | |
1129 | return -EIO; | |
2ffca4f3 PR |
1130 | |
1131 | if (mdev->tconn->agreed_pro_version <= 94) | |
1132 | max_bio_size = min_t(int, max_bio_size, DRBD_MAX_SIZE_H80_PACKET); | |
1133 | else if (mdev->tconn->agreed_pro_version < 100) | |
1134 | max_bio_size = min_t(int, max_bio_size, DRBD_MAX_BIO_SIZE_P95); | |
1135 | ||
9f5bdc33 AG |
1136 | p->d_size = cpu_to_be64(d_size); |
1137 | p->u_size = cpu_to_be64(u_size); | |
1138 | p->c_size = cpu_to_be64(trigger_reply ? 0 : drbd_get_capacity(mdev->this_bdev)); | |
1139 | p->max_bio_size = cpu_to_be32(max_bio_size); | |
1140 | p->queue_order_type = cpu_to_be16(q_order_type); | |
1141 | p->dds_flags = cpu_to_be16(flags); | |
1142 | return drbd_send_command(mdev, sock, P_SIZES, sizeof(*p), NULL, 0); | |
b411b363 PR |
1143 | } |
1144 | ||
1145 | /** | |
43de7c85 | 1146 | * drbd_send_current_state() - Sends the drbd state to the peer |
b411b363 PR |
1147 | * @mdev: DRBD device. |
1148 | */ | |
43de7c85 | 1149 | int drbd_send_current_state(struct drbd_conf *mdev) |
b411b363 | 1150 | { |
7c96715a | 1151 | struct drbd_socket *sock; |
9f5bdc33 | 1152 | struct p_state *p; |
b411b363 | 1153 | |
7c96715a | 1154 | sock = &mdev->tconn->data; |
9f5bdc33 AG |
1155 | p = drbd_prepare_command(mdev, sock); |
1156 | if (!p) | |
1157 | return -EIO; | |
1158 | p->state = cpu_to_be32(mdev->state.i); /* Within the send mutex */ | |
1159 | return drbd_send_command(mdev, sock, P_STATE, sizeof(*p), NULL, 0); | |
1160 | } | |
b411b363 | 1161 | |
43de7c85 PR |
1162 | /** |
1163 | * drbd_send_state() - After a state change, sends the new state to the peer | |
1164 | * @mdev: DRBD device. | |
1165 | * @state: the state to send, not necessarily the current state. | |
1166 | * | |
1167 | * Each state change queues an "after_state_ch" work, which will eventually | |
1168 | * send the resulting new state to the peer. If more state changes happen | |
1169 | * between queuing and processing of the after_state_ch work, we still | |
1170 | * want to send each intermediary state in the order it occurred. | |
1171 | */ | |
1172 | int drbd_send_state(struct drbd_conf *mdev, union drbd_state state) | |
1173 | { | |
1174 | struct drbd_socket *sock; | |
1175 | struct p_state *p; | |
1176 | ||
1177 | sock = &mdev->tconn->data; | |
1178 | p = drbd_prepare_command(mdev, sock); | |
1179 | if (!p) | |
1180 | return -EIO; | |
1181 | p->state = cpu_to_be32(state.i); /* Within the send mutex */ | |
1182 | return drbd_send_command(mdev, sock, P_STATE, sizeof(*p), NULL, 0); | |
1183 | } | |
1184 | ||
9f5bdc33 AG |
1185 | int drbd_send_state_req(struct drbd_conf *mdev, union drbd_state mask, union drbd_state val) |
1186 | { | |
1187 | struct drbd_socket *sock; | |
1188 | struct p_req_state *p; | |
b411b363 | 1189 | |
9f5bdc33 AG |
1190 | sock = &mdev->tconn->data; |
1191 | p = drbd_prepare_command(mdev, sock); | |
1192 | if (!p) | |
1193 | return -EIO; | |
1194 | p->mask = cpu_to_be32(mask.i); | |
1195 | p->val = cpu_to_be32(val.i); | |
1196 | return drbd_send_command(mdev, sock, P_STATE_CHG_REQ, sizeof(*p), NULL, 0); | |
b411b363 PR |
1197 | } |
1198 | ||
9f5bdc33 | 1199 | int conn_send_state_req(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val) |
b411b363 | 1200 | { |
9f5bdc33 AG |
1201 | enum drbd_packet cmd; |
1202 | struct drbd_socket *sock; | |
1203 | struct p_req_state *p; | |
b411b363 | 1204 | |
9f5bdc33 AG |
1205 | cmd = tconn->agreed_pro_version < 100 ? P_STATE_CHG_REQ : P_CONN_ST_CHG_REQ; |
1206 | sock = &tconn->data; | |
1207 | p = conn_prepare_command(tconn, sock); | |
1208 | if (!p) | |
1209 | return -EIO; | |
1210 | p->mask = cpu_to_be32(mask.i); | |
1211 | p->val = cpu_to_be32(val.i); | |
1212 | return conn_send_command(tconn, sock, cmd, sizeof(*p), NULL, 0); | |
b411b363 PR |
1213 | } |
1214 | ||
2f4e7abe | 1215 | void drbd_send_sr_reply(struct drbd_conf *mdev, enum drbd_state_rv retcode) |
b411b363 | 1216 | { |
9f5bdc33 AG |
1217 | struct drbd_socket *sock; |
1218 | struct p_req_state_reply *p; | |
b411b363 | 1219 | |
9f5bdc33 AG |
1220 | sock = &mdev->tconn->meta; |
1221 | p = drbd_prepare_command(mdev, sock); | |
1222 | if (p) { | |
1223 | p->retcode = cpu_to_be32(retcode); | |
1224 | drbd_send_command(mdev, sock, P_STATE_CHG_REPLY, sizeof(*p), NULL, 0); | |
1225 | } | |
b411b363 PR |
1226 | } |
1227 | ||
9f5bdc33 | 1228 | void conn_send_sr_reply(struct drbd_tconn *tconn, enum drbd_state_rv retcode) |
047cd4a6 | 1229 | { |
9f5bdc33 AG |
1230 | struct drbd_socket *sock; |
1231 | struct p_req_state_reply *p; | |
047cd4a6 PR |
1232 | enum drbd_packet cmd = tconn->agreed_pro_version < 100 ? P_STATE_CHG_REPLY : P_CONN_ST_CHG_REPLY; |
1233 | ||
9f5bdc33 AG |
1234 | sock = &tconn->meta; |
1235 | p = conn_prepare_command(tconn, sock); | |
1236 | if (p) { | |
1237 | p->retcode = cpu_to_be32(retcode); | |
1238 | conn_send_command(tconn, sock, cmd, sizeof(*p), NULL, 0); | |
1239 | } | |
047cd4a6 PR |
1240 | } |
1241 | ||
a02d1240 AG |
1242 | static void dcbp_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code) |
1243 | { | |
1244 | BUG_ON(code & ~0xf); | |
1245 | p->encoding = (p->encoding & ~0xf) | code; | |
1246 | } | |
1247 | ||
1248 | static void dcbp_set_start(struct p_compressed_bm *p, int set) | |
1249 | { | |
1250 | p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0); | |
1251 | } | |
1252 | ||
1253 | static void dcbp_set_pad_bits(struct p_compressed_bm *p, int n) | |
1254 | { | |
1255 | BUG_ON(n & ~0x7); | |
1256 | p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4); | |
1257 | } | |
1258 | ||
b411b363 | 1259 | int fill_bitmap_rle_bits(struct drbd_conf *mdev, |
50d0b1ad AG |
1260 | struct p_compressed_bm *p, |
1261 | unsigned int size, | |
1262 | struct bm_xfer_ctx *c) | |
b411b363 PR |
1263 | { |
1264 | struct bitstream bs; | |
1265 | unsigned long plain_bits; | |
1266 | unsigned long tmp; | |
1267 | unsigned long rl; | |
1268 | unsigned len; | |
1269 | unsigned toggle; | |
44ed167d | 1270 | int bits, use_rle; |
b411b363 PR |
1271 | |
1272 | /* may we use this feature? */ | |
44ed167d PR |
1273 | rcu_read_lock(); |
1274 | use_rle = rcu_dereference(mdev->tconn->net_conf)->use_rle; | |
1275 | rcu_read_unlock(); | |
1276 | if (!use_rle || mdev->tconn->agreed_pro_version < 90) | |
1277 | return 0; | |
b411b363 PR |
1278 | |
1279 | if (c->bit_offset >= c->bm_bits) | |
1280 | return 0; /* nothing to do. */ | |
1281 | ||
1282 | /* use at most thus many bytes */ | |
50d0b1ad AG |
1283 | bitstream_init(&bs, p->code, size, 0); |
1284 | memset(p->code, 0, size); | |
b411b363 PR |
1285 | /* plain bits covered in this code string */ |
1286 | plain_bits = 0; | |
1287 | ||
1288 | /* p->encoding & 0x80 stores whether the first run length is set. | |
1289 | * bit offset is implicit. | |
1290 | * start with toggle == 2 to be able to tell the first iteration */ | |
1291 | toggle = 2; | |
1292 | ||
1293 | /* see how much plain bits we can stuff into one packet | |
1294 | * using RLE and VLI. */ | |
1295 | do { | |
1296 | tmp = (toggle == 0) ? _drbd_bm_find_next_zero(mdev, c->bit_offset) | |
1297 | : _drbd_bm_find_next(mdev, c->bit_offset); | |
1298 | if (tmp == -1UL) | |
1299 | tmp = c->bm_bits; | |
1300 | rl = tmp - c->bit_offset; | |
1301 | ||
1302 | if (toggle == 2) { /* first iteration */ | |
1303 | if (rl == 0) { | |
1304 | /* the first checked bit was set, | |
1305 | * store start value, */ | |
a02d1240 | 1306 | dcbp_set_start(p, 1); |
b411b363 PR |
1307 | /* but skip encoding of zero run length */ |
1308 | toggle = !toggle; | |
1309 | continue; | |
1310 | } | |
a02d1240 | 1311 | dcbp_set_start(p, 0); |
b411b363 PR |
1312 | } |
1313 | ||
1314 | /* paranoia: catch zero runlength. | |
1315 | * can only happen if bitmap is modified while we scan it. */ | |
1316 | if (rl == 0) { | |
1317 | dev_err(DEV, "unexpected zero runlength while encoding bitmap " | |
1318 | "t:%u bo:%lu\n", toggle, c->bit_offset); | |
1319 | return -1; | |
1320 | } | |
1321 | ||
1322 | bits = vli_encode_bits(&bs, rl); | |
1323 | if (bits == -ENOBUFS) /* buffer full */ | |
1324 | break; | |
1325 | if (bits <= 0) { | |
1326 | dev_err(DEV, "error while encoding bitmap: %d\n", bits); | |
1327 | return 0; | |
1328 | } | |
1329 | ||
1330 | toggle = !toggle; | |
1331 | plain_bits += rl; | |
1332 | c->bit_offset = tmp; | |
1333 | } while (c->bit_offset < c->bm_bits); | |
1334 | ||
1335 | len = bs.cur.b - p->code + !!bs.cur.bit; | |
1336 | ||
1337 | if (plain_bits < (len << 3)) { | |
1338 | /* incompressible with this method. | |
1339 | * we need to rewind both word and bit position. */ | |
1340 | c->bit_offset -= plain_bits; | |
1341 | bm_xfer_ctx_bit_to_word_offset(c); | |
1342 | c->bit_offset = c->word_offset * BITS_PER_LONG; | |
1343 | return 0; | |
1344 | } | |
1345 | ||
1346 | /* RLE + VLI was able to compress it just fine. | |
1347 | * update c->word_offset. */ | |
1348 | bm_xfer_ctx_bit_to_word_offset(c); | |
1349 | ||
1350 | /* store pad_bits */ | |
a02d1240 | 1351 | dcbp_set_pad_bits(p, (8 - bs.cur.bit) & 0x7); |
b411b363 PR |
1352 | |
1353 | return len; | |
1354 | } | |
1355 | ||
f70af118 AG |
1356 | /** |
1357 | * send_bitmap_rle_or_plain | |
1358 | * | |
1359 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
1360 | * code upon failure. | |
1361 | */ | |
1362 | static int | |
79ed9bd0 | 1363 | send_bitmap_rle_or_plain(struct drbd_conf *mdev, struct bm_xfer_ctx *c) |
b411b363 | 1364 | { |
9f5bdc33 | 1365 | struct drbd_socket *sock = &mdev->tconn->data; |
50d0b1ad | 1366 | unsigned int header_size = drbd_header_size(mdev->tconn); |
e658983a | 1367 | struct p_compressed_bm *p = sock->sbuf + header_size; |
a982dd57 | 1368 | int len, err; |
b411b363 | 1369 | |
e658983a AG |
1370 | len = fill_bitmap_rle_bits(mdev, p, |
1371 | DRBD_SOCKET_BUFFER_SIZE - header_size - sizeof(*p), c); | |
b411b363 | 1372 | if (len < 0) |
f70af118 | 1373 | return -EIO; |
b411b363 PR |
1374 | |
1375 | if (len) { | |
a02d1240 | 1376 | dcbp_set_code(p, RLE_VLI_Bits); |
9f5bdc33 AG |
1377 | err = __send_command(mdev->tconn, mdev->vnr, sock, |
1378 | P_COMPRESSED_BITMAP, sizeof(*p) + len, | |
1379 | NULL, 0); | |
b411b363 | 1380 | c->packets[0]++; |
e658983a | 1381 | c->bytes[0] += header_size + sizeof(*p) + len; |
b411b363 PR |
1382 | |
1383 | if (c->bit_offset >= c->bm_bits) | |
1384 | len = 0; /* DONE */ | |
1385 | } else { | |
1386 | /* was not compressible. | |
1387 | * send a buffer full of plain text bits instead. */ | |
50d0b1ad AG |
1388 | unsigned int data_size; |
1389 | unsigned long num_words; | |
e658983a | 1390 | unsigned long *p = sock->sbuf + header_size; |
50d0b1ad AG |
1391 | |
1392 | data_size = DRBD_SOCKET_BUFFER_SIZE - header_size; | |
e658983a | 1393 | num_words = min_t(size_t, data_size / sizeof(*p), |
50d0b1ad | 1394 | c->bm_words - c->word_offset); |
e658983a | 1395 | len = num_words * sizeof(*p); |
b411b363 | 1396 | if (len) |
e658983a AG |
1397 | drbd_bm_get_lel(mdev, c->word_offset, num_words, p); |
1398 | err = __send_command(mdev->tconn, mdev->vnr, sock, P_BITMAP, len, NULL, 0); | |
b411b363 PR |
1399 | c->word_offset += num_words; |
1400 | c->bit_offset = c->word_offset * BITS_PER_LONG; | |
1401 | ||
1402 | c->packets[1]++; | |
50d0b1ad | 1403 | c->bytes[1] += header_size + len; |
b411b363 PR |
1404 | |
1405 | if (c->bit_offset > c->bm_bits) | |
1406 | c->bit_offset = c->bm_bits; | |
1407 | } | |
a982dd57 | 1408 | if (!err) { |
f70af118 AG |
1409 | if (len == 0) { |
1410 | INFO_bm_xfer_stats(mdev, "send", c); | |
1411 | return 0; | |
1412 | } else | |
1413 | return 1; | |
1414 | } | |
1415 | return -EIO; | |
b411b363 PR |
1416 | } |
1417 | ||
1418 | /* See the comment at receive_bitmap() */ | |
058820cd | 1419 | static int _drbd_send_bitmap(struct drbd_conf *mdev) |
b411b363 PR |
1420 | { |
1421 | struct bm_xfer_ctx c; | |
f70af118 | 1422 | int err; |
b411b363 | 1423 | |
841ce241 AG |
1424 | if (!expect(mdev->bitmap)) |
1425 | return false; | |
b411b363 | 1426 | |
b411b363 PR |
1427 | if (get_ldev(mdev)) { |
1428 | if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) { | |
1429 | dev_info(DEV, "Writing the whole bitmap, MDF_FullSync was set.\n"); | |
1430 | drbd_bm_set_all(mdev); | |
1431 | if (drbd_bm_write(mdev)) { | |
1432 | /* write_bm did fail! Leave full sync flag set in Meta P_DATA | |
1433 | * but otherwise process as per normal - need to tell other | |
1434 | * side that a full resync is required! */ | |
1435 | dev_err(DEV, "Failed to write bitmap to disk!\n"); | |
1436 | } else { | |
1437 | drbd_md_clear_flag(mdev, MDF_FULL_SYNC); | |
1438 | drbd_md_sync(mdev); | |
1439 | } | |
1440 | } | |
1441 | put_ldev(mdev); | |
1442 | } | |
1443 | ||
1444 | c = (struct bm_xfer_ctx) { | |
1445 | .bm_bits = drbd_bm_bits(mdev), | |
1446 | .bm_words = drbd_bm_words(mdev), | |
1447 | }; | |
1448 | ||
1449 | do { | |
79ed9bd0 | 1450 | err = send_bitmap_rle_or_plain(mdev, &c); |
f70af118 | 1451 | } while (err > 0); |
b411b363 | 1452 | |
f70af118 | 1453 | return err == 0; |
b411b363 PR |
1454 | } |
1455 | ||
1456 | int drbd_send_bitmap(struct drbd_conf *mdev) | |
1457 | { | |
9f5bdc33 AG |
1458 | struct drbd_socket *sock = &mdev->tconn->data; |
1459 | int err = -1; | |
b411b363 | 1460 | |
9f5bdc33 AG |
1461 | mutex_lock(&sock->mutex); |
1462 | if (sock->socket) | |
1463 | err = !_drbd_send_bitmap(mdev); | |
1464 | mutex_unlock(&sock->mutex); | |
b411b363 PR |
1465 | return err; |
1466 | } | |
9f5bdc33 | 1467 | |
d4e67d7c | 1468 | void drbd_send_b_ack(struct drbd_conf *mdev, u32 barrier_nr, u32 set_size) |
b411b363 | 1469 | { |
9f5bdc33 AG |
1470 | struct drbd_socket *sock; |
1471 | struct p_barrier_ack *p; | |
b411b363 | 1472 | |
9f5bdc33 AG |
1473 | if (mdev->state.conn < C_CONNECTED) |
1474 | return; | |
b411b363 | 1475 | |
9f5bdc33 AG |
1476 | sock = &mdev->tconn->meta; |
1477 | p = drbd_prepare_command(mdev, sock); | |
1478 | if (!p) | |
1479 | return; | |
1480 | p->barrier = barrier_nr; | |
1481 | p->set_size = cpu_to_be32(set_size); | |
1482 | drbd_send_command(mdev, sock, P_BARRIER_ACK, sizeof(*p), NULL, 0); | |
b411b363 PR |
1483 | } |
1484 | ||
1485 | /** | |
1486 | * _drbd_send_ack() - Sends an ack packet | |
1487 | * @mdev: DRBD device. | |
1488 | * @cmd: Packet command code. | |
1489 | * @sector: sector, needs to be in big endian byte order | |
1490 | * @blksize: size in byte, needs to be in big endian byte order | |
1491 | * @block_id: Id, big endian byte order | |
1492 | */ | |
d8763023 AG |
1493 | static int _drbd_send_ack(struct drbd_conf *mdev, enum drbd_packet cmd, |
1494 | u64 sector, u32 blksize, u64 block_id) | |
b411b363 | 1495 | { |
9f5bdc33 AG |
1496 | struct drbd_socket *sock; |
1497 | struct p_block_ack *p; | |
b411b363 | 1498 | |
9f5bdc33 AG |
1499 | if (mdev->state.conn < C_CONNECTED) |
1500 | return -EIO; | |
b411b363 | 1501 | |
9f5bdc33 AG |
1502 | sock = &mdev->tconn->meta; |
1503 | p = drbd_prepare_command(mdev, sock); | |
1504 | if (!p) | |
a8c32aa8 | 1505 | return -EIO; |
9f5bdc33 AG |
1506 | p->sector = sector; |
1507 | p->block_id = block_id; | |
1508 | p->blksize = blksize; | |
1509 | p->seq_num = cpu_to_be32(atomic_inc_return(&mdev->packet_seq)); | |
1510 | return drbd_send_command(mdev, sock, cmd, sizeof(*p), NULL, 0); | |
b411b363 PR |
1511 | } |
1512 | ||
2b2bf214 LE |
1513 | /* dp->sector and dp->block_id already/still in network byte order, |
1514 | * data_size is payload size according to dp->head, | |
1515 | * and may need to be corrected for digest size. */ | |
a9a9994d AG |
1516 | void drbd_send_ack_dp(struct drbd_conf *mdev, enum drbd_packet cmd, |
1517 | struct p_data *dp, int data_size) | |
b411b363 | 1518 | { |
88104ca4 AG |
1519 | if (mdev->tconn->peer_integrity_tfm) |
1520 | data_size -= crypto_hash_digestsize(mdev->tconn->peer_integrity_tfm); | |
a9a9994d AG |
1521 | _drbd_send_ack(mdev, cmd, dp->sector, cpu_to_be32(data_size), |
1522 | dp->block_id); | |
b411b363 PR |
1523 | } |
1524 | ||
a9a9994d AG |
1525 | void drbd_send_ack_rp(struct drbd_conf *mdev, enum drbd_packet cmd, |
1526 | struct p_block_req *rp) | |
b411b363 | 1527 | { |
a9a9994d | 1528 | _drbd_send_ack(mdev, cmd, rp->sector, rp->blksize, rp->block_id); |
b411b363 PR |
1529 | } |
1530 | ||
1531 | /** | |
1532 | * drbd_send_ack() - Sends an ack packet | |
db830c46 AG |
1533 | * @mdev: DRBD device |
1534 | * @cmd: packet command code | |
1535 | * @peer_req: peer request | |
b411b363 | 1536 | */ |
d8763023 | 1537 | int drbd_send_ack(struct drbd_conf *mdev, enum drbd_packet cmd, |
db830c46 | 1538 | struct drbd_peer_request *peer_req) |
b411b363 | 1539 | { |
dd516121 AG |
1540 | return _drbd_send_ack(mdev, cmd, |
1541 | cpu_to_be64(peer_req->i.sector), | |
1542 | cpu_to_be32(peer_req->i.size), | |
1543 | peer_req->block_id); | |
b411b363 PR |
1544 | } |
1545 | ||
1546 | /* This function misuses the block_id field to signal if the blocks | |
1547 | * are is sync or not. */ | |
d8763023 | 1548 | int drbd_send_ack_ex(struct drbd_conf *mdev, enum drbd_packet cmd, |
b411b363 PR |
1549 | sector_t sector, int blksize, u64 block_id) |
1550 | { | |
fa79abd8 AG |
1551 | return _drbd_send_ack(mdev, cmd, |
1552 | cpu_to_be64(sector), | |
1553 | cpu_to_be32(blksize), | |
1554 | cpu_to_be64(block_id)); | |
b411b363 PR |
1555 | } |
1556 | ||
1557 | int drbd_send_drequest(struct drbd_conf *mdev, int cmd, | |
1558 | sector_t sector, int size, u64 block_id) | |
1559 | { | |
9f5bdc33 AG |
1560 | struct drbd_socket *sock; |
1561 | struct p_block_req *p; | |
b411b363 | 1562 | |
9f5bdc33 AG |
1563 | sock = &mdev->tconn->data; |
1564 | p = drbd_prepare_command(mdev, sock); | |
1565 | if (!p) | |
1566 | return -EIO; | |
1567 | p->sector = cpu_to_be64(sector); | |
1568 | p->block_id = block_id; | |
1569 | p->blksize = cpu_to_be32(size); | |
1570 | return drbd_send_command(mdev, sock, cmd, sizeof(*p), NULL, 0); | |
b411b363 PR |
1571 | } |
1572 | ||
d8763023 AG |
1573 | int drbd_send_drequest_csum(struct drbd_conf *mdev, sector_t sector, int size, |
1574 | void *digest, int digest_size, enum drbd_packet cmd) | |
b411b363 | 1575 | { |
9f5bdc33 AG |
1576 | struct drbd_socket *sock; |
1577 | struct p_block_req *p; | |
b411b363 | 1578 | |
9f5bdc33 | 1579 | /* FIXME: Put the digest into the preallocated socket buffer. */ |
b411b363 | 1580 | |
9f5bdc33 AG |
1581 | sock = &mdev->tconn->data; |
1582 | p = drbd_prepare_command(mdev, sock); | |
1583 | if (!p) | |
1584 | return -EIO; | |
1585 | p->sector = cpu_to_be64(sector); | |
1586 | p->block_id = ID_SYNCER /* unused */; | |
1587 | p->blksize = cpu_to_be32(size); | |
1588 | return drbd_send_command(mdev, sock, cmd, sizeof(*p), | |
1589 | digest, digest_size); | |
b411b363 PR |
1590 | } |
1591 | ||
1592 | int drbd_send_ov_request(struct drbd_conf *mdev, sector_t sector, int size) | |
1593 | { | |
9f5bdc33 AG |
1594 | struct drbd_socket *sock; |
1595 | struct p_block_req *p; | |
b411b363 | 1596 | |
9f5bdc33 AG |
1597 | sock = &mdev->tconn->data; |
1598 | p = drbd_prepare_command(mdev, sock); | |
1599 | if (!p) | |
1600 | return -EIO; | |
1601 | p->sector = cpu_to_be64(sector); | |
1602 | p->block_id = ID_SYNCER /* unused */; | |
1603 | p->blksize = cpu_to_be32(size); | |
1604 | return drbd_send_command(mdev, sock, P_OV_REQUEST, sizeof(*p), NULL, 0); | |
b411b363 PR |
1605 | } |
1606 | ||
1607 | /* called on sndtimeo | |
81e84650 AG |
1608 | * returns false if we should retry, |
1609 | * true if we think connection is dead | |
b411b363 | 1610 | */ |
1a7ba646 | 1611 | static int we_should_drop_the_connection(struct drbd_tconn *tconn, struct socket *sock) |
b411b363 PR |
1612 | { |
1613 | int drop_it; | |
1614 | /* long elapsed = (long)(jiffies - mdev->last_received); */ | |
1615 | ||
1a7ba646 PR |
1616 | drop_it = tconn->meta.socket == sock |
1617 | || !tconn->asender.task | |
1618 | || get_t_state(&tconn->asender) != RUNNING | |
bbeb641c | 1619 | || tconn->cstate < C_WF_REPORT_PARAMS; |
b411b363 PR |
1620 | |
1621 | if (drop_it) | |
81e84650 | 1622 | return true; |
b411b363 | 1623 | |
1a7ba646 | 1624 | drop_it = !--tconn->ko_count; |
b411b363 | 1625 | if (!drop_it) { |
1a7ba646 PR |
1626 | conn_err(tconn, "[%s/%d] sock_sendmsg time expired, ko = %u\n", |
1627 | current->comm, current->pid, tconn->ko_count); | |
1628 | request_ping(tconn); | |
b411b363 PR |
1629 | } |
1630 | ||
1631 | return drop_it; /* && (mdev->state == R_PRIMARY) */; | |
1632 | } | |
1633 | ||
1a7ba646 | 1634 | static void drbd_update_congested(struct drbd_tconn *tconn) |
9e204cdd | 1635 | { |
1a7ba646 | 1636 | struct sock *sk = tconn->data.socket->sk; |
9e204cdd | 1637 | if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5) |
1a7ba646 | 1638 | set_bit(NET_CONGESTED, &tconn->flags); |
9e204cdd AG |
1639 | } |
1640 | ||
b411b363 PR |
1641 | /* The idea of sendpage seems to be to put some kind of reference |
1642 | * to the page into the skb, and to hand it over to the NIC. In | |
1643 | * this process get_page() gets called. | |
1644 | * | |
1645 | * As soon as the page was really sent over the network put_page() | |
1646 | * gets called by some part of the network layer. [ NIC driver? ] | |
1647 | * | |
1648 | * [ get_page() / put_page() increment/decrement the count. If count | |
1649 | * reaches 0 the page will be freed. ] | |
1650 | * | |
1651 | * This works nicely with pages from FSs. | |
1652 | * But this means that in protocol A we might signal IO completion too early! | |
1653 | * | |
1654 | * In order not to corrupt data during a resync we must make sure | |
1655 | * that we do not reuse our own buffer pages (EEs) to early, therefore | |
1656 | * we have the net_ee list. | |
1657 | * | |
1658 | * XFS seems to have problems, still, it submits pages with page_count == 0! | |
1659 | * As a workaround, we disable sendpage on pages | |
1660 | * with page_count == 0 or PageSlab. | |
1661 | */ | |
1662 | static int _drbd_no_send_page(struct drbd_conf *mdev, struct page *page, | |
b987427b | 1663 | int offset, size_t size, unsigned msg_flags) |
b411b363 | 1664 | { |
b987427b AG |
1665 | struct socket *socket; |
1666 | void *addr; | |
1667 | int err; | |
1668 | ||
1669 | socket = mdev->tconn->data.socket; | |
1670 | addr = kmap(page) + offset; | |
1671 | err = drbd_send_all(mdev->tconn, socket, addr, size, msg_flags); | |
b411b363 | 1672 | kunmap(page); |
b987427b AG |
1673 | if (!err) |
1674 | mdev->send_cnt += size >> 9; | |
1675 | return err; | |
b411b363 PR |
1676 | } |
1677 | ||
1678 | static int _drbd_send_page(struct drbd_conf *mdev, struct page *page, | |
ba11ad9a | 1679 | int offset, size_t size, unsigned msg_flags) |
b411b363 | 1680 | { |
88b390ff | 1681 | struct socket *socket = mdev->tconn->data.socket; |
b411b363 | 1682 | mm_segment_t oldfs = get_fs(); |
b411b363 | 1683 | int len = size; |
88b390ff | 1684 | int err = -EIO; |
b411b363 PR |
1685 | |
1686 | /* e.g. XFS meta- & log-data is in slab pages, which have a | |
1687 | * page_count of 0 and/or have PageSlab() set. | |
1688 | * we cannot use send_page for those, as that does get_page(); | |
1689 | * put_page(); and would cause either a VM_BUG directly, or | |
1690 | * __page_cache_release a page that would actually still be referenced | |
1691 | * by someone, leading to some obscure delayed Oops somewhere else. */ | |
1692 | if (disable_sendpage || (page_count(page) < 1) || PageSlab(page)) | |
88b390ff | 1693 | return _drbd_no_send_page(mdev, page, offset, size, msg_flags); |
b411b363 | 1694 | |
ba11ad9a | 1695 | msg_flags |= MSG_NOSIGNAL; |
1a7ba646 | 1696 | drbd_update_congested(mdev->tconn); |
b411b363 PR |
1697 | set_fs(KERNEL_DS); |
1698 | do { | |
88b390ff AG |
1699 | int sent; |
1700 | ||
1701 | sent = socket->ops->sendpage(socket, page, offset, len, msg_flags); | |
b411b363 | 1702 | if (sent <= 0) { |
88b390ff AG |
1703 | if (sent == -EAGAIN) { |
1704 | if (we_should_drop_the_connection(mdev->tconn, socket)) | |
1705 | break; | |
1706 | continue; | |
1707 | } | |
b411b363 PR |
1708 | dev_warn(DEV, "%s: size=%d len=%d sent=%d\n", |
1709 | __func__, (int)size, len, sent); | |
88b390ff AG |
1710 | if (sent < 0) |
1711 | err = sent; | |
b411b363 PR |
1712 | break; |
1713 | } | |
1714 | len -= sent; | |
1715 | offset += sent; | |
1716 | } while (len > 0 /* THINK && mdev->cstate >= C_CONNECTED*/); | |
1717 | set_fs(oldfs); | |
01a311a5 | 1718 | clear_bit(NET_CONGESTED, &mdev->tconn->flags); |
b411b363 | 1719 | |
88b390ff AG |
1720 | if (len == 0) { |
1721 | err = 0; | |
1722 | mdev->send_cnt += size >> 9; | |
1723 | } | |
1724 | return err; | |
b411b363 PR |
1725 | } |
1726 | ||
1727 | static int _drbd_send_bio(struct drbd_conf *mdev, struct bio *bio) | |
1728 | { | |
1729 | struct bio_vec *bvec; | |
1730 | int i; | |
ba11ad9a | 1731 | /* hint all but last page with MSG_MORE */ |
b411b363 | 1732 | __bio_for_each_segment(bvec, bio, i, 0) { |
7fae55da AG |
1733 | int err; |
1734 | ||
1735 | err = _drbd_no_send_page(mdev, bvec->bv_page, | |
1736 | bvec->bv_offset, bvec->bv_len, | |
1737 | i == bio->bi_vcnt - 1 ? 0 : MSG_MORE); | |
1738 | if (err) | |
1739 | return err; | |
b411b363 | 1740 | } |
7fae55da | 1741 | return 0; |
b411b363 PR |
1742 | } |
1743 | ||
1744 | static int _drbd_send_zc_bio(struct drbd_conf *mdev, struct bio *bio) | |
1745 | { | |
1746 | struct bio_vec *bvec; | |
1747 | int i; | |
ba11ad9a | 1748 | /* hint all but last page with MSG_MORE */ |
b411b363 | 1749 | __bio_for_each_segment(bvec, bio, i, 0) { |
7fae55da AG |
1750 | int err; |
1751 | ||
1752 | err = _drbd_send_page(mdev, bvec->bv_page, | |
1753 | bvec->bv_offset, bvec->bv_len, | |
1754 | i == bio->bi_vcnt - 1 ? 0 : MSG_MORE); | |
1755 | if (err) | |
1756 | return err; | |
b411b363 | 1757 | } |
7fae55da | 1758 | return 0; |
b411b363 PR |
1759 | } |
1760 | ||
db830c46 AG |
1761 | static int _drbd_send_zc_ee(struct drbd_conf *mdev, |
1762 | struct drbd_peer_request *peer_req) | |
45bb912b | 1763 | { |
db830c46 AG |
1764 | struct page *page = peer_req->pages; |
1765 | unsigned len = peer_req->i.size; | |
9f69230c | 1766 | int err; |
db830c46 | 1767 | |
ba11ad9a | 1768 | /* hint all but last page with MSG_MORE */ |
45bb912b LE |
1769 | page_chain_for_each(page) { |
1770 | unsigned l = min_t(unsigned, len, PAGE_SIZE); | |
9f69230c AG |
1771 | |
1772 | err = _drbd_send_page(mdev, page, 0, l, | |
1773 | page_chain_next(page) ? MSG_MORE : 0); | |
1774 | if (err) | |
1775 | return err; | |
45bb912b LE |
1776 | len -= l; |
1777 | } | |
9f69230c | 1778 | return 0; |
45bb912b LE |
1779 | } |
1780 | ||
76d2e7ec PR |
1781 | static u32 bio_flags_to_wire(struct drbd_conf *mdev, unsigned long bi_rw) |
1782 | { | |
31890f4a | 1783 | if (mdev->tconn->agreed_pro_version >= 95) |
76d2e7ec | 1784 | return (bi_rw & REQ_SYNC ? DP_RW_SYNC : 0) | |
76d2e7ec PR |
1785 | (bi_rw & REQ_FUA ? DP_FUA : 0) | |
1786 | (bi_rw & REQ_FLUSH ? DP_FLUSH : 0) | | |
1787 | (bi_rw & REQ_DISCARD ? DP_DISCARD : 0); | |
1788 | else | |
721a9602 | 1789 | return bi_rw & REQ_SYNC ? DP_RW_SYNC : 0; |
76d2e7ec PR |
1790 | } |
1791 | ||
b411b363 PR |
1792 | /* Used to send write requests |
1793 | * R_PRIMARY -> Peer (P_DATA) | |
1794 | */ | |
1795 | int drbd_send_dblock(struct drbd_conf *mdev, struct drbd_request *req) | |
1796 | { | |
9f5bdc33 AG |
1797 | struct drbd_socket *sock; |
1798 | struct p_data *p; | |
b411b363 | 1799 | unsigned int dp_flags = 0; |
b411b363 | 1800 | int dgs; |
9f5bdc33 | 1801 | int err; |
b411b363 | 1802 | |
46e1ce41 PR |
1803 | sock = &mdev->tconn->data; |
1804 | p = drbd_prepare_command(mdev, sock); | |
7d4c782c | 1805 | dgs = mdev->tconn->integrity_tfm ? crypto_hash_digestsize(mdev->tconn->integrity_tfm) : 0; |
b411b363 | 1806 | |
9f5bdc33 AG |
1807 | if (!p) |
1808 | return -EIO; | |
1809 | p->sector = cpu_to_be64(req->i.sector); | |
1810 | p->block_id = (unsigned long)req; | |
1811 | p->seq_num = cpu_to_be32(req->seq_num = atomic_inc_return(&mdev->packet_seq)); | |
76d2e7ec | 1812 | dp_flags = bio_flags_to_wire(mdev, req->master_bio->bi_rw); |
b411b363 PR |
1813 | if (mdev->state.conn >= C_SYNC_SOURCE && |
1814 | mdev->state.conn <= C_PAUSED_SYNC_T) | |
1815 | dp_flags |= DP_MAY_SET_IN_SYNC; | |
303d1448 PR |
1816 | if (mdev->tconn->agreed_pro_version >= 100) { |
1817 | if (req->rq_state & RQ_EXP_RECEIVE_ACK) | |
1818 | dp_flags |= DP_SEND_RECEIVE_ACK; | |
1819 | if (req->rq_state & RQ_EXP_WRITE_ACK) | |
1820 | dp_flags |= DP_SEND_WRITE_ACK; | |
1821 | } | |
9f5bdc33 AG |
1822 | p->dp_flags = cpu_to_be32(dp_flags); |
1823 | if (dgs) | |
8d412fc6 | 1824 | drbd_csum_bio(mdev, mdev->tconn->integrity_tfm, req->master_bio, p + 1); |
9f5bdc33 | 1825 | err = __send_command(mdev->tconn, mdev->vnr, sock, P_DATA, sizeof(*p) + dgs, NULL, req->i.size); |
6bdb9b0e | 1826 | if (!err) { |
470be44a LE |
1827 | /* For protocol A, we have to memcpy the payload into |
1828 | * socket buffers, as we may complete right away | |
1829 | * as soon as we handed it over to tcp, at which point the data | |
1830 | * pages may become invalid. | |
1831 | * | |
1832 | * For data-integrity enabled, we copy it as well, so we can be | |
1833 | * sure that even if the bio pages may still be modified, it | |
1834 | * won't change the data on the wire, thus if the digest checks | |
1835 | * out ok after sending on this side, but does not fit on the | |
1836 | * receiving side, we sure have detected corruption elsewhere. | |
1837 | */ | |
303d1448 | 1838 | if (!(req->rq_state & (RQ_EXP_RECEIVE_ACK | RQ_EXP_WRITE_ACK)) || dgs) |
6bdb9b0e | 1839 | err = _drbd_send_bio(mdev, req->master_bio); |
b411b363 | 1840 | else |
6bdb9b0e | 1841 | err = _drbd_send_zc_bio(mdev, req->master_bio); |
470be44a LE |
1842 | |
1843 | /* double check digest, sometimes buffers have been modified in flight. */ | |
1844 | if (dgs > 0 && dgs <= 64) { | |
24c4830c | 1845 | /* 64 byte, 512 bit, is the largest digest size |
470be44a LE |
1846 | * currently supported in kernel crypto. */ |
1847 | unsigned char digest[64]; | |
8d412fc6 | 1848 | drbd_csum_bio(mdev, mdev->tconn->integrity_tfm, req->master_bio, digest); |
9f5bdc33 | 1849 | if (memcmp(p + 1, digest, dgs)) { |
470be44a LE |
1850 | dev_warn(DEV, |
1851 | "Digest mismatch, buffer modified by upper layers during write: %llus +%u\n", | |
ace652ac | 1852 | (unsigned long long)req->i.sector, req->i.size); |
470be44a LE |
1853 | } |
1854 | } /* else if (dgs > 64) { | |
1855 | ... Be noisy about digest too large ... | |
1856 | } */ | |
b411b363 | 1857 | } |
9f5bdc33 | 1858 | mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */ |
bd26bfc5 | 1859 | |
6bdb9b0e | 1860 | return err; |
b411b363 PR |
1861 | } |
1862 | ||
1863 | /* answer packet, used to send data back for read requests: | |
1864 | * Peer -> (diskless) R_PRIMARY (P_DATA_REPLY) | |
1865 | * C_SYNC_SOURCE -> C_SYNC_TARGET (P_RS_DATA_REPLY) | |
1866 | */ | |
d8763023 | 1867 | int drbd_send_block(struct drbd_conf *mdev, enum drbd_packet cmd, |
db830c46 | 1868 | struct drbd_peer_request *peer_req) |
b411b363 | 1869 | { |
9f5bdc33 AG |
1870 | struct drbd_socket *sock; |
1871 | struct p_data *p; | |
7b57b89d | 1872 | int err; |
b411b363 PR |
1873 | int dgs; |
1874 | ||
46e1ce41 PR |
1875 | sock = &mdev->tconn->data; |
1876 | p = drbd_prepare_command(mdev, sock); | |
1877 | ||
7d4c782c | 1878 | dgs = mdev->tconn->integrity_tfm ? crypto_hash_digestsize(mdev->tconn->integrity_tfm) : 0; |
b411b363 | 1879 | |
9f5bdc33 AG |
1880 | if (!p) |
1881 | return -EIO; | |
1882 | p->sector = cpu_to_be64(peer_req->i.sector); | |
1883 | p->block_id = peer_req->block_id; | |
1884 | p->seq_num = 0; /* unused */ | |
1885 | if (dgs) | |
8d412fc6 | 1886 | drbd_csum_ee(mdev, mdev->tconn->integrity_tfm, peer_req, p + 1); |
9f5bdc33 | 1887 | err = __send_command(mdev->tconn, mdev->vnr, sock, cmd, sizeof(*p) + dgs, NULL, peer_req->i.size); |
7b57b89d AG |
1888 | if (!err) |
1889 | err = _drbd_send_zc_ee(mdev, peer_req); | |
9f5bdc33 | 1890 | mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */ |
bd26bfc5 | 1891 | |
7b57b89d | 1892 | return err; |
b411b363 PR |
1893 | } |
1894 | ||
8f7bed77 | 1895 | int drbd_send_out_of_sync(struct drbd_conf *mdev, struct drbd_request *req) |
73a01a18 | 1896 | { |
9f5bdc33 AG |
1897 | struct drbd_socket *sock; |
1898 | struct p_block_desc *p; | |
73a01a18 | 1899 | |
9f5bdc33 AG |
1900 | sock = &mdev->tconn->data; |
1901 | p = drbd_prepare_command(mdev, sock); | |
1902 | if (!p) | |
1903 | return -EIO; | |
1904 | p->sector = cpu_to_be64(req->i.sector); | |
1905 | p->blksize = cpu_to_be32(req->i.size); | |
1906 | return drbd_send_command(mdev, sock, P_OUT_OF_SYNC, sizeof(*p), NULL, 0); | |
73a01a18 PR |
1907 | } |
1908 | ||
b411b363 PR |
1909 | /* |
1910 | drbd_send distinguishes two cases: | |
1911 | ||
1912 | Packets sent via the data socket "sock" | |
1913 | and packets sent via the meta data socket "msock" | |
1914 | ||
1915 | sock msock | |
1916 | -----------------+-------------------------+------------------------------ | |
1917 | timeout conf.timeout / 2 conf.timeout / 2 | |
1918 | timeout action send a ping via msock Abort communication | |
1919 | and close all sockets | |
1920 | */ | |
1921 | ||
1922 | /* | |
1923 | * you must have down()ed the appropriate [m]sock_mutex elsewhere! | |
1924 | */ | |
bedbd2a5 | 1925 | int drbd_send(struct drbd_tconn *tconn, struct socket *sock, |
b411b363 PR |
1926 | void *buf, size_t size, unsigned msg_flags) |
1927 | { | |
1928 | struct kvec iov; | |
1929 | struct msghdr msg; | |
1930 | int rv, sent = 0; | |
1931 | ||
1932 | if (!sock) | |
c0d42c8e | 1933 | return -EBADR; |
b411b363 PR |
1934 | |
1935 | /* THINK if (signal_pending) return ... ? */ | |
1936 | ||
1937 | iov.iov_base = buf; | |
1938 | iov.iov_len = size; | |
1939 | ||
1940 | msg.msg_name = NULL; | |
1941 | msg.msg_namelen = 0; | |
1942 | msg.msg_control = NULL; | |
1943 | msg.msg_controllen = 0; | |
1944 | msg.msg_flags = msg_flags | MSG_NOSIGNAL; | |
1945 | ||
bedbd2a5 | 1946 | if (sock == tconn->data.socket) { |
44ed167d PR |
1947 | rcu_read_lock(); |
1948 | tconn->ko_count = rcu_dereference(tconn->net_conf)->ko_count; | |
1949 | rcu_read_unlock(); | |
bedbd2a5 | 1950 | drbd_update_congested(tconn); |
b411b363 PR |
1951 | } |
1952 | do { | |
1953 | /* STRANGE | |
1954 | * tcp_sendmsg does _not_ use its size parameter at all ? | |
1955 | * | |
1956 | * -EAGAIN on timeout, -EINTR on signal. | |
1957 | */ | |
1958 | /* THINK | |
1959 | * do we need to block DRBD_SIG if sock == &meta.socket ?? | |
1960 | * otherwise wake_asender() might interrupt some send_*Ack ! | |
1961 | */ | |
1962 | rv = kernel_sendmsg(sock, &msg, &iov, 1, size); | |
1963 | if (rv == -EAGAIN) { | |
bedbd2a5 | 1964 | if (we_should_drop_the_connection(tconn, sock)) |
b411b363 PR |
1965 | break; |
1966 | else | |
1967 | continue; | |
1968 | } | |
b411b363 PR |
1969 | if (rv == -EINTR) { |
1970 | flush_signals(current); | |
1971 | rv = 0; | |
1972 | } | |
1973 | if (rv < 0) | |
1974 | break; | |
1975 | sent += rv; | |
1976 | iov.iov_base += rv; | |
1977 | iov.iov_len -= rv; | |
1978 | } while (sent < size); | |
1979 | ||
bedbd2a5 PR |
1980 | if (sock == tconn->data.socket) |
1981 | clear_bit(NET_CONGESTED, &tconn->flags); | |
b411b363 PR |
1982 | |
1983 | if (rv <= 0) { | |
1984 | if (rv != -EAGAIN) { | |
bedbd2a5 PR |
1985 | conn_err(tconn, "%s_sendmsg returned %d\n", |
1986 | sock == tconn->meta.socket ? "msock" : "sock", | |
1987 | rv); | |
bbeb641c | 1988 | conn_request_state(tconn, NS(conn, C_BROKEN_PIPE), CS_HARD); |
b411b363 | 1989 | } else |
bbeb641c | 1990 | conn_request_state(tconn, NS(conn, C_TIMEOUT), CS_HARD); |
b411b363 PR |
1991 | } |
1992 | ||
1993 | return sent; | |
1994 | } | |
1995 | ||
fb708e40 AG |
1996 | /** |
1997 | * drbd_send_all - Send an entire buffer | |
1998 | * | |
1999 | * Returns 0 upon success and a negative error value otherwise. | |
2000 | */ | |
2001 | int drbd_send_all(struct drbd_tconn *tconn, struct socket *sock, void *buffer, | |
2002 | size_t size, unsigned msg_flags) | |
2003 | { | |
2004 | int err; | |
2005 | ||
2006 | err = drbd_send(tconn, sock, buffer, size, msg_flags); | |
2007 | if (err < 0) | |
2008 | return err; | |
2009 | if (err != size) | |
2010 | return -EIO; | |
2011 | return 0; | |
2012 | } | |
2013 | ||
b411b363 PR |
2014 | static int drbd_open(struct block_device *bdev, fmode_t mode) |
2015 | { | |
2016 | struct drbd_conf *mdev = bdev->bd_disk->private_data; | |
2017 | unsigned long flags; | |
2018 | int rv = 0; | |
2019 | ||
2a48fc0a | 2020 | mutex_lock(&drbd_main_mutex); |
87eeee41 | 2021 | spin_lock_irqsave(&mdev->tconn->req_lock, flags); |
b411b363 PR |
2022 | /* to have a stable mdev->state.role |
2023 | * and no race with updating open_cnt */ | |
2024 | ||
2025 | if (mdev->state.role != R_PRIMARY) { | |
2026 | if (mode & FMODE_WRITE) | |
2027 | rv = -EROFS; | |
2028 | else if (!allow_oos) | |
2029 | rv = -EMEDIUMTYPE; | |
2030 | } | |
2031 | ||
2032 | if (!rv) | |
2033 | mdev->open_cnt++; | |
87eeee41 | 2034 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); |
2a48fc0a | 2035 | mutex_unlock(&drbd_main_mutex); |
b411b363 PR |
2036 | |
2037 | return rv; | |
2038 | } | |
2039 | ||
2040 | static int drbd_release(struct gendisk *gd, fmode_t mode) | |
2041 | { | |
2042 | struct drbd_conf *mdev = gd->private_data; | |
2a48fc0a | 2043 | mutex_lock(&drbd_main_mutex); |
b411b363 | 2044 | mdev->open_cnt--; |
2a48fc0a | 2045 | mutex_unlock(&drbd_main_mutex); |
b411b363 PR |
2046 | return 0; |
2047 | } | |
2048 | ||
b411b363 PR |
2049 | static void drbd_set_defaults(struct drbd_conf *mdev) |
2050 | { | |
f399002e LE |
2051 | /* Beware! The actual layout differs |
2052 | * between big endian and little endian */ | |
da9fbc27 | 2053 | mdev->state = (union drbd_dev_state) { |
b411b363 PR |
2054 | { .role = R_SECONDARY, |
2055 | .peer = R_UNKNOWN, | |
2056 | .conn = C_STANDALONE, | |
2057 | .disk = D_DISKLESS, | |
2058 | .pdsk = D_UNKNOWN, | |
b411b363 PR |
2059 | } }; |
2060 | } | |
2061 | ||
2062 | void drbd_init_set_defaults(struct drbd_conf *mdev) | |
2063 | { | |
2064 | /* the memset(,0,) did most of this. | |
2065 | * note: only assignments, no allocation in here */ | |
2066 | ||
2067 | drbd_set_defaults(mdev); | |
2068 | ||
b411b363 PR |
2069 | atomic_set(&mdev->ap_bio_cnt, 0); |
2070 | atomic_set(&mdev->ap_pending_cnt, 0); | |
2071 | atomic_set(&mdev->rs_pending_cnt, 0); | |
2072 | atomic_set(&mdev->unacked_cnt, 0); | |
2073 | atomic_set(&mdev->local_cnt, 0); | |
435f0740 | 2074 | atomic_set(&mdev->pp_in_use_by_net, 0); |
778f271d | 2075 | atomic_set(&mdev->rs_sect_in, 0); |
0f0601f4 | 2076 | atomic_set(&mdev->rs_sect_ev, 0); |
759fbdfb | 2077 | atomic_set(&mdev->ap_in_flight, 0); |
cdfda633 | 2078 | atomic_set(&mdev->md_io_in_use, 0); |
b411b363 | 2079 | |
8410da8f PR |
2080 | mutex_init(&mdev->own_state_mutex); |
2081 | mdev->state_mutex = &mdev->own_state_mutex; | |
b411b363 | 2082 | |
b411b363 | 2083 | spin_lock_init(&mdev->al_lock); |
b411b363 PR |
2084 | spin_lock_init(&mdev->peer_seq_lock); |
2085 | spin_lock_init(&mdev->epoch_lock); | |
2086 | ||
2087 | INIT_LIST_HEAD(&mdev->active_ee); | |
2088 | INIT_LIST_HEAD(&mdev->sync_ee); | |
2089 | INIT_LIST_HEAD(&mdev->done_ee); | |
2090 | INIT_LIST_HEAD(&mdev->read_ee); | |
2091 | INIT_LIST_HEAD(&mdev->net_ee); | |
2092 | INIT_LIST_HEAD(&mdev->resync_reads); | |
b411b363 PR |
2093 | INIT_LIST_HEAD(&mdev->resync_work.list); |
2094 | INIT_LIST_HEAD(&mdev->unplug_work.list); | |
e9e6f3ec | 2095 | INIT_LIST_HEAD(&mdev->go_diskless.list); |
b411b363 | 2096 | INIT_LIST_HEAD(&mdev->md_sync_work.list); |
c4752ef1 | 2097 | INIT_LIST_HEAD(&mdev->start_resync_work.list); |
b411b363 | 2098 | INIT_LIST_HEAD(&mdev->bm_io_work.w.list); |
0ced55a3 | 2099 | |
794abb75 | 2100 | mdev->resync_work.cb = w_resync_timer; |
b411b363 | 2101 | mdev->unplug_work.cb = w_send_write_hint; |
e9e6f3ec | 2102 | mdev->go_diskless.cb = w_go_diskless; |
b411b363 PR |
2103 | mdev->md_sync_work.cb = w_md_sync; |
2104 | mdev->bm_io_work.w.cb = w_bitmap_io; | |
370a43e7 | 2105 | mdev->start_resync_work.cb = w_start_resync; |
a21e9298 PR |
2106 | |
2107 | mdev->resync_work.mdev = mdev; | |
2108 | mdev->unplug_work.mdev = mdev; | |
2109 | mdev->go_diskless.mdev = mdev; | |
2110 | mdev->md_sync_work.mdev = mdev; | |
2111 | mdev->bm_io_work.w.mdev = mdev; | |
2112 | mdev->start_resync_work.mdev = mdev; | |
2113 | ||
b411b363 PR |
2114 | init_timer(&mdev->resync_timer); |
2115 | init_timer(&mdev->md_sync_timer); | |
370a43e7 | 2116 | init_timer(&mdev->start_resync_timer); |
7fde2be9 | 2117 | init_timer(&mdev->request_timer); |
b411b363 PR |
2118 | mdev->resync_timer.function = resync_timer_fn; |
2119 | mdev->resync_timer.data = (unsigned long) mdev; | |
2120 | mdev->md_sync_timer.function = md_sync_timer_fn; | |
2121 | mdev->md_sync_timer.data = (unsigned long) mdev; | |
370a43e7 PR |
2122 | mdev->start_resync_timer.function = start_resync_timer_fn; |
2123 | mdev->start_resync_timer.data = (unsigned long) mdev; | |
7fde2be9 PR |
2124 | mdev->request_timer.function = request_timer_fn; |
2125 | mdev->request_timer.data = (unsigned long) mdev; | |
b411b363 PR |
2126 | |
2127 | init_waitqueue_head(&mdev->misc_wait); | |
2128 | init_waitqueue_head(&mdev->state_wait); | |
2129 | init_waitqueue_head(&mdev->ee_wait); | |
2130 | init_waitqueue_head(&mdev->al_wait); | |
2131 | init_waitqueue_head(&mdev->seq_wait); | |
2132 | ||
b411b363 | 2133 | mdev->resync_wenr = LC_FREE; |
99432fcc PR |
2134 | mdev->peer_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE; |
2135 | mdev->local_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE; | |
b411b363 PR |
2136 | } |
2137 | ||
2138 | void drbd_mdev_cleanup(struct drbd_conf *mdev) | |
2139 | { | |
1d7734a0 | 2140 | int i; |
e6b3ea83 | 2141 | if (mdev->tconn->receiver.t_state != NONE) |
b411b363 | 2142 | dev_err(DEV, "ASSERT FAILED: receiver t_state == %d expected 0.\n", |
e6b3ea83 | 2143 | mdev->tconn->receiver.t_state); |
b411b363 PR |
2144 | |
2145 | /* no need to lock it, I'm the only thread alive */ | |
2146 | if (atomic_read(&mdev->current_epoch->epoch_size) != 0) | |
2147 | dev_err(DEV, "epoch_size:%d\n", atomic_read(&mdev->current_epoch->epoch_size)); | |
2148 | mdev->al_writ_cnt = | |
2149 | mdev->bm_writ_cnt = | |
2150 | mdev->read_cnt = | |
2151 | mdev->recv_cnt = | |
2152 | mdev->send_cnt = | |
2153 | mdev->writ_cnt = | |
2154 | mdev->p_size = | |
2155 | mdev->rs_start = | |
2156 | mdev->rs_total = | |
1d7734a0 LE |
2157 | mdev->rs_failed = 0; |
2158 | mdev->rs_last_events = 0; | |
0f0601f4 | 2159 | mdev->rs_last_sect_ev = 0; |
1d7734a0 LE |
2160 | for (i = 0; i < DRBD_SYNC_MARKS; i++) { |
2161 | mdev->rs_mark_left[i] = 0; | |
2162 | mdev->rs_mark_time[i] = 0; | |
2163 | } | |
89e58e75 | 2164 | D_ASSERT(mdev->tconn->net_conf == NULL); |
b411b363 PR |
2165 | |
2166 | drbd_set_my_capacity(mdev, 0); | |
2167 | if (mdev->bitmap) { | |
2168 | /* maybe never allocated. */ | |
02d9a94b | 2169 | drbd_bm_resize(mdev, 0, 1); |
b411b363 PR |
2170 | drbd_bm_cleanup(mdev); |
2171 | } | |
2172 | ||
1d041225 PR |
2173 | drbd_free_bc(mdev->ldev); |
2174 | mdev->ldev = NULL; | |
2175 | ||
0778286a | 2176 | clear_bit(AL_SUSPENDED, &mdev->flags); |
b411b363 | 2177 | |
b411b363 PR |
2178 | D_ASSERT(list_empty(&mdev->active_ee)); |
2179 | D_ASSERT(list_empty(&mdev->sync_ee)); | |
2180 | D_ASSERT(list_empty(&mdev->done_ee)); | |
2181 | D_ASSERT(list_empty(&mdev->read_ee)); | |
2182 | D_ASSERT(list_empty(&mdev->net_ee)); | |
2183 | D_ASSERT(list_empty(&mdev->resync_reads)); | |
e42325a5 PR |
2184 | D_ASSERT(list_empty(&mdev->tconn->data.work.q)); |
2185 | D_ASSERT(list_empty(&mdev->tconn->meta.work.q)); | |
b411b363 PR |
2186 | D_ASSERT(list_empty(&mdev->resync_work.list)); |
2187 | D_ASSERT(list_empty(&mdev->unplug_work.list)); | |
e9e6f3ec | 2188 | D_ASSERT(list_empty(&mdev->go_diskless.list)); |
2265b473 LE |
2189 | |
2190 | drbd_set_defaults(mdev); | |
b411b363 PR |
2191 | } |
2192 | ||
2193 | ||
2194 | static void drbd_destroy_mempools(void) | |
2195 | { | |
2196 | struct page *page; | |
2197 | ||
2198 | while (drbd_pp_pool) { | |
2199 | page = drbd_pp_pool; | |
2200 | drbd_pp_pool = (struct page *)page_private(page); | |
2201 | __free_page(page); | |
2202 | drbd_pp_vacant--; | |
2203 | } | |
2204 | ||
2205 | /* D_ASSERT(atomic_read(&drbd_pp_vacant)==0); */ | |
2206 | ||
da4a75d2 LE |
2207 | if (drbd_md_io_bio_set) |
2208 | bioset_free(drbd_md_io_bio_set); | |
35abf594 LE |
2209 | if (drbd_md_io_page_pool) |
2210 | mempool_destroy(drbd_md_io_page_pool); | |
b411b363 PR |
2211 | if (drbd_ee_mempool) |
2212 | mempool_destroy(drbd_ee_mempool); | |
2213 | if (drbd_request_mempool) | |
2214 | mempool_destroy(drbd_request_mempool); | |
2215 | if (drbd_ee_cache) | |
2216 | kmem_cache_destroy(drbd_ee_cache); | |
2217 | if (drbd_request_cache) | |
2218 | kmem_cache_destroy(drbd_request_cache); | |
2219 | if (drbd_bm_ext_cache) | |
2220 | kmem_cache_destroy(drbd_bm_ext_cache); | |
2221 | if (drbd_al_ext_cache) | |
2222 | kmem_cache_destroy(drbd_al_ext_cache); | |
2223 | ||
da4a75d2 | 2224 | drbd_md_io_bio_set = NULL; |
35abf594 | 2225 | drbd_md_io_page_pool = NULL; |
b411b363 PR |
2226 | drbd_ee_mempool = NULL; |
2227 | drbd_request_mempool = NULL; | |
2228 | drbd_ee_cache = NULL; | |
2229 | drbd_request_cache = NULL; | |
2230 | drbd_bm_ext_cache = NULL; | |
2231 | drbd_al_ext_cache = NULL; | |
2232 | ||
2233 | return; | |
2234 | } | |
2235 | ||
2236 | static int drbd_create_mempools(void) | |
2237 | { | |
2238 | struct page *page; | |
1816a2b4 | 2239 | const int number = (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * minor_count; |
b411b363 PR |
2240 | int i; |
2241 | ||
2242 | /* prepare our caches and mempools */ | |
2243 | drbd_request_mempool = NULL; | |
2244 | drbd_ee_cache = NULL; | |
2245 | drbd_request_cache = NULL; | |
2246 | drbd_bm_ext_cache = NULL; | |
2247 | drbd_al_ext_cache = NULL; | |
2248 | drbd_pp_pool = NULL; | |
35abf594 | 2249 | drbd_md_io_page_pool = NULL; |
da4a75d2 | 2250 | drbd_md_io_bio_set = NULL; |
b411b363 PR |
2251 | |
2252 | /* caches */ | |
2253 | drbd_request_cache = kmem_cache_create( | |
2254 | "drbd_req", sizeof(struct drbd_request), 0, 0, NULL); | |
2255 | if (drbd_request_cache == NULL) | |
2256 | goto Enomem; | |
2257 | ||
2258 | drbd_ee_cache = kmem_cache_create( | |
f6ffca9f | 2259 | "drbd_ee", sizeof(struct drbd_peer_request), 0, 0, NULL); |
b411b363 PR |
2260 | if (drbd_ee_cache == NULL) |
2261 | goto Enomem; | |
2262 | ||
2263 | drbd_bm_ext_cache = kmem_cache_create( | |
2264 | "drbd_bm", sizeof(struct bm_extent), 0, 0, NULL); | |
2265 | if (drbd_bm_ext_cache == NULL) | |
2266 | goto Enomem; | |
2267 | ||
2268 | drbd_al_ext_cache = kmem_cache_create( | |
2269 | "drbd_al", sizeof(struct lc_element), 0, 0, NULL); | |
2270 | if (drbd_al_ext_cache == NULL) | |
2271 | goto Enomem; | |
2272 | ||
2273 | /* mempools */ | |
da4a75d2 LE |
2274 | drbd_md_io_bio_set = bioset_create(DRBD_MIN_POOL_PAGES, 0); |
2275 | if (drbd_md_io_bio_set == NULL) | |
2276 | goto Enomem; | |
2277 | ||
35abf594 LE |
2278 | drbd_md_io_page_pool = mempool_create_page_pool(DRBD_MIN_POOL_PAGES, 0); |
2279 | if (drbd_md_io_page_pool == NULL) | |
2280 | goto Enomem; | |
2281 | ||
b411b363 PR |
2282 | drbd_request_mempool = mempool_create(number, |
2283 | mempool_alloc_slab, mempool_free_slab, drbd_request_cache); | |
2284 | if (drbd_request_mempool == NULL) | |
2285 | goto Enomem; | |
2286 | ||
2287 | drbd_ee_mempool = mempool_create(number, | |
2288 | mempool_alloc_slab, mempool_free_slab, drbd_ee_cache); | |
2027ae1f | 2289 | if (drbd_ee_mempool == NULL) |
b411b363 PR |
2290 | goto Enomem; |
2291 | ||
2292 | /* drbd's page pool */ | |
2293 | spin_lock_init(&drbd_pp_lock); | |
2294 | ||
2295 | for (i = 0; i < number; i++) { | |
2296 | page = alloc_page(GFP_HIGHUSER); | |
2297 | if (!page) | |
2298 | goto Enomem; | |
2299 | set_page_private(page, (unsigned long)drbd_pp_pool); | |
2300 | drbd_pp_pool = page; | |
2301 | } | |
2302 | drbd_pp_vacant = number; | |
2303 | ||
2304 | return 0; | |
2305 | ||
2306 | Enomem: | |
2307 | drbd_destroy_mempools(); /* in case we allocated some */ | |
2308 | return -ENOMEM; | |
2309 | } | |
2310 | ||
2311 | static int drbd_notify_sys(struct notifier_block *this, unsigned long code, | |
2312 | void *unused) | |
2313 | { | |
2314 | /* just so we have it. you never know what interesting things we | |
2315 | * might want to do here some day... | |
2316 | */ | |
2317 | ||
2318 | return NOTIFY_DONE; | |
2319 | } | |
2320 | ||
2321 | static struct notifier_block drbd_notifier = { | |
2322 | .notifier_call = drbd_notify_sys, | |
2323 | }; | |
2324 | ||
7721f567 | 2325 | static void drbd_release_all_peer_reqs(struct drbd_conf *mdev) |
b411b363 PR |
2326 | { |
2327 | int rr; | |
2328 | ||
7721f567 | 2329 | rr = drbd_free_peer_reqs(mdev, &mdev->active_ee); |
b411b363 PR |
2330 | if (rr) |
2331 | dev_err(DEV, "%d EEs in active list found!\n", rr); | |
2332 | ||
7721f567 | 2333 | rr = drbd_free_peer_reqs(mdev, &mdev->sync_ee); |
b411b363 PR |
2334 | if (rr) |
2335 | dev_err(DEV, "%d EEs in sync list found!\n", rr); | |
2336 | ||
7721f567 | 2337 | rr = drbd_free_peer_reqs(mdev, &mdev->read_ee); |
b411b363 PR |
2338 | if (rr) |
2339 | dev_err(DEV, "%d EEs in read list found!\n", rr); | |
2340 | ||
7721f567 | 2341 | rr = drbd_free_peer_reqs(mdev, &mdev->done_ee); |
b411b363 PR |
2342 | if (rr) |
2343 | dev_err(DEV, "%d EEs in done list found!\n", rr); | |
2344 | ||
7721f567 | 2345 | rr = drbd_free_peer_reqs(mdev, &mdev->net_ee); |
b411b363 PR |
2346 | if (rr) |
2347 | dev_err(DEV, "%d EEs in net list found!\n", rr); | |
2348 | } | |
2349 | ||
774b3055 | 2350 | /* caution. no locking. */ |
81fa2e67 | 2351 | void drbd_minor_destroy(struct kref *kref) |
b411b363 | 2352 | { |
81fa2e67 | 2353 | struct drbd_conf *mdev = container_of(kref, struct drbd_conf, kref); |
9dc9fbb3 PR |
2354 | struct drbd_tconn *tconn = mdev->tconn; |
2355 | ||
cdfda633 PR |
2356 | del_timer_sync(&mdev->request_timer); |
2357 | ||
b411b363 | 2358 | /* paranoia asserts */ |
70dc65e1 | 2359 | D_ASSERT(mdev->open_cnt == 0); |
e42325a5 | 2360 | D_ASSERT(list_empty(&mdev->tconn->data.work.q)); |
b411b363 PR |
2361 | /* end paranoia asserts */ |
2362 | ||
b411b363 PR |
2363 | /* cleanup stuff that may have been allocated during |
2364 | * device (re-)configuration or state changes */ | |
2365 | ||
2366 | if (mdev->this_bdev) | |
2367 | bdput(mdev->this_bdev); | |
2368 | ||
1d041225 PR |
2369 | drbd_free_bc(mdev->ldev); |
2370 | mdev->ldev = NULL; | |
b411b363 | 2371 | |
7721f567 | 2372 | drbd_release_all_peer_reqs(mdev); |
b411b363 | 2373 | |
b411b363 PR |
2374 | lc_destroy(mdev->act_log); |
2375 | lc_destroy(mdev->resync); | |
2376 | ||
2377 | kfree(mdev->p_uuid); | |
2378 | /* mdev->p_uuid = NULL; */ | |
2379 | ||
cd1d9950 PR |
2380 | kfree(mdev->current_epoch); |
2381 | if (mdev->bitmap) /* should no longer be there. */ | |
2382 | drbd_bm_cleanup(mdev); | |
2383 | __free_page(mdev->md_io_page); | |
2384 | put_disk(mdev->vdisk); | |
2385 | blk_cleanup_queue(mdev->rq_queue); | |
9958c857 | 2386 | kfree(mdev->rs_plan_s); |
cd1d9950 | 2387 | kfree(mdev); |
9dc9fbb3 PR |
2388 | |
2389 | kref_put(&tconn->kref, &conn_destroy); | |
b411b363 PR |
2390 | } |
2391 | ||
2392 | static void drbd_cleanup(void) | |
2393 | { | |
2394 | unsigned int i; | |
81a5d60e | 2395 | struct drbd_conf *mdev; |
81fa2e67 | 2396 | struct drbd_tconn *tconn, *tmp; |
b411b363 PR |
2397 | |
2398 | unregister_reboot_notifier(&drbd_notifier); | |
2399 | ||
17a93f30 LE |
2400 | /* first remove proc, |
2401 | * drbdsetup uses it's presence to detect | |
2402 | * whether DRBD is loaded. | |
2403 | * If we would get stuck in proc removal, | |
2404 | * but have netlink already deregistered, | |
2405 | * some drbdsetup commands may wait forever | |
2406 | * for an answer. | |
2407 | */ | |
2408 | if (drbd_proc) | |
2409 | remove_proc_entry("drbd", NULL); | |
2410 | ||
3b98c0c2 | 2411 | drbd_genl_unregister(); |
b411b363 | 2412 | |
81fa2e67 PR |
2413 | idr_for_each_entry(&minors, mdev, i) { |
2414 | idr_remove(&minors, mdev_to_minor(mdev)); | |
2415 | idr_remove(&mdev->tconn->volumes, mdev->vnr); | |
2416 | del_gendisk(mdev->vdisk); | |
c141ebda | 2417 | /* synchronize_rcu(); No other threads running at this point */ |
81fa2e67 PR |
2418 | kref_put(&mdev->kref, &drbd_minor_destroy); |
2419 | } | |
2420 | ||
c141ebda | 2421 | /* not _rcu since, no other updater anymore. Genl already unregistered */ |
81fa2e67 | 2422 | list_for_each_entry_safe(tconn, tmp, &drbd_tconns, all_tconn) { |
c141ebda PR |
2423 | list_del(&tconn->all_tconn); /* not _rcu no proc, not other threads */ |
2424 | /* synchronize_rcu(); */ | |
81fa2e67 PR |
2425 | kref_put(&tconn->kref, &conn_destroy); |
2426 | } | |
ff370e5a | 2427 | |
81a5d60e | 2428 | drbd_destroy_mempools(); |
b411b363 PR |
2429 | unregister_blkdev(DRBD_MAJOR, "drbd"); |
2430 | ||
81a5d60e PR |
2431 | idr_destroy(&minors); |
2432 | ||
b411b363 PR |
2433 | printk(KERN_INFO "drbd: module cleanup done.\n"); |
2434 | } | |
2435 | ||
2436 | /** | |
2437 | * drbd_congested() - Callback for pdflush | |
2438 | * @congested_data: User data | |
2439 | * @bdi_bits: Bits pdflush is currently interested in | |
2440 | * | |
2441 | * Returns 1<<BDI_async_congested and/or 1<<BDI_sync_congested if we are congested. | |
2442 | */ | |
2443 | static int drbd_congested(void *congested_data, int bdi_bits) | |
2444 | { | |
2445 | struct drbd_conf *mdev = congested_data; | |
2446 | struct request_queue *q; | |
2447 | char reason = '-'; | |
2448 | int r = 0; | |
2449 | ||
1b881ef7 | 2450 | if (!may_inc_ap_bio(mdev)) { |
b411b363 PR |
2451 | /* DRBD has frozen IO */ |
2452 | r = bdi_bits; | |
2453 | reason = 'd'; | |
2454 | goto out; | |
2455 | } | |
2456 | ||
2457 | if (get_ldev(mdev)) { | |
2458 | q = bdev_get_queue(mdev->ldev->backing_bdev); | |
2459 | r = bdi_congested(&q->backing_dev_info, bdi_bits); | |
2460 | put_ldev(mdev); | |
2461 | if (r) | |
2462 | reason = 'b'; | |
2463 | } | |
2464 | ||
01a311a5 | 2465 | if (bdi_bits & (1 << BDI_async_congested) && test_bit(NET_CONGESTED, &mdev->tconn->flags)) { |
b411b363 PR |
2466 | r |= (1 << BDI_async_congested); |
2467 | reason = reason == 'b' ? 'a' : 'n'; | |
2468 | } | |
2469 | ||
2470 | out: | |
2471 | mdev->congestion_reason = reason; | |
2472 | return r; | |
2473 | } | |
2474 | ||
6699b655 PR |
2475 | static void drbd_init_workqueue(struct drbd_work_queue* wq) |
2476 | { | |
2477 | sema_init(&wq->s, 0); | |
2478 | spin_lock_init(&wq->q_lock); | |
2479 | INIT_LIST_HEAD(&wq->q); | |
2480 | } | |
2481 | ||
0ace9dfa | 2482 | struct drbd_tconn *conn_get_by_name(const char *name) |
1aba4d7f PR |
2483 | { |
2484 | struct drbd_tconn *tconn; | |
2485 | ||
3b98c0c2 LE |
2486 | if (!name || !name[0]) |
2487 | return NULL; | |
2488 | ||
c141ebda | 2489 | rcu_read_lock(); |
ec0bddbc | 2490 | list_for_each_entry_rcu(tconn, &drbd_tconns, all_tconn) { |
0ace9dfa PR |
2491 | if (!strcmp(tconn->name, name)) { |
2492 | kref_get(&tconn->kref); | |
1aba4d7f | 2493 | goto found; |
0ace9dfa | 2494 | } |
1aba4d7f PR |
2495 | } |
2496 | tconn = NULL; | |
2497 | found: | |
c141ebda | 2498 | rcu_read_unlock(); |
1aba4d7f PR |
2499 | return tconn; |
2500 | } | |
2501 | ||
089c075d AG |
2502 | struct drbd_tconn *conn_get_by_addrs(void *my_addr, int my_addr_len, |
2503 | void *peer_addr, int peer_addr_len) | |
2504 | { | |
2505 | struct drbd_tconn *tconn; | |
2506 | ||
2507 | rcu_read_lock(); | |
2508 | list_for_each_entry_rcu(tconn, &drbd_tconns, all_tconn) { | |
2509 | if (tconn->my_addr_len == my_addr_len && | |
2510 | tconn->peer_addr_len == peer_addr_len && | |
2511 | !memcmp(&tconn->my_addr, my_addr, my_addr_len) && | |
2512 | !memcmp(&tconn->peer_addr, peer_addr, peer_addr_len)) { | |
2513 | kref_get(&tconn->kref); | |
2514 | goto found; | |
2515 | } | |
2516 | } | |
2517 | tconn = NULL; | |
2518 | found: | |
2519 | rcu_read_unlock(); | |
2520 | return tconn; | |
2521 | } | |
2522 | ||
e6ef8a5c AG |
2523 | static int drbd_alloc_socket(struct drbd_socket *socket) |
2524 | { | |
2525 | socket->rbuf = (void *) __get_free_page(GFP_KERNEL); | |
2526 | if (!socket->rbuf) | |
2527 | return -ENOMEM; | |
5a87d920 AG |
2528 | socket->sbuf = (void *) __get_free_page(GFP_KERNEL); |
2529 | if (!socket->sbuf) | |
2530 | return -ENOMEM; | |
e6ef8a5c AG |
2531 | return 0; |
2532 | } | |
2533 | ||
2534 | static void drbd_free_socket(struct drbd_socket *socket) | |
2535 | { | |
5a87d920 | 2536 | free_page((unsigned long) socket->sbuf); |
e6ef8a5c AG |
2537 | free_page((unsigned long) socket->rbuf); |
2538 | } | |
2539 | ||
91fd4dad PR |
2540 | void conn_free_crypto(struct drbd_tconn *tconn) |
2541 | { | |
1d041225 PR |
2542 | drbd_free_sock(tconn); |
2543 | ||
2544 | crypto_free_hash(tconn->csums_tfm); | |
2545 | crypto_free_hash(tconn->verify_tfm); | |
91fd4dad | 2546 | crypto_free_hash(tconn->cram_hmac_tfm); |
8d412fc6 | 2547 | crypto_free_hash(tconn->integrity_tfm); |
5b614abe | 2548 | crypto_free_hash(tconn->peer_integrity_tfm); |
91fd4dad PR |
2549 | kfree(tconn->int_dig_in); |
2550 | kfree(tconn->int_dig_vv); | |
1d041225 PR |
2551 | |
2552 | tconn->csums_tfm = NULL; | |
2553 | tconn->verify_tfm = NULL; | |
91fd4dad | 2554 | tconn->cram_hmac_tfm = NULL; |
8d412fc6 | 2555 | tconn->integrity_tfm = NULL; |
5b614abe | 2556 | tconn->peer_integrity_tfm = NULL; |
91fd4dad PR |
2557 | tconn->int_dig_in = NULL; |
2558 | tconn->int_dig_vv = NULL; | |
2559 | } | |
2560 | ||
afbbfa88 AG |
2561 | int set_resource_options(struct drbd_tconn *tconn, struct res_opts *res_opts) |
2562 | { | |
2563 | cpumask_var_t new_cpu_mask; | |
2564 | int err; | |
2565 | ||
2566 | if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) | |
2567 | return -ENOMEM; | |
2568 | /* | |
2569 | retcode = ERR_NOMEM; | |
2570 | drbd_msg_put_info("unable to allocate cpumask"); | |
2571 | */ | |
2572 | ||
2573 | /* silently ignore cpu mask on UP kernel */ | |
2574 | if (nr_cpu_ids > 1 && res_opts->cpu_mask[0] != 0) { | |
2575 | /* FIXME: Get rid of constant 32 here */ | |
2576 | err = __bitmap_parse(res_opts->cpu_mask, 32, 0, | |
2577 | cpumask_bits(new_cpu_mask), nr_cpu_ids); | |
2578 | if (err) { | |
2579 | conn_warn(tconn, "__bitmap_parse() failed with %d\n", err); | |
2580 | /* retcode = ERR_CPU_MASK_PARSE; */ | |
2581 | goto fail; | |
2582 | } | |
2583 | } | |
2584 | tconn->res_opts = *res_opts; | |
2585 | if (!cpumask_equal(tconn->cpu_mask, new_cpu_mask)) { | |
2586 | cpumask_copy(tconn->cpu_mask, new_cpu_mask); | |
2587 | drbd_calc_cpu_mask(tconn); | |
2588 | tconn->receiver.reset_cpu_mask = 1; | |
2589 | tconn->asender.reset_cpu_mask = 1; | |
2590 | tconn->worker.reset_cpu_mask = 1; | |
2591 | } | |
2592 | err = 0; | |
2593 | ||
2594 | fail: | |
2595 | free_cpumask_var(new_cpu_mask); | |
2596 | return err; | |
2597 | ||
2598 | } | |
2599 | ||
ec0bddbc | 2600 | /* caller must be under genl_lock() */ |
afbbfa88 | 2601 | struct drbd_tconn *conn_create(const char *name, struct res_opts *res_opts) |
2111438b PR |
2602 | { |
2603 | struct drbd_tconn *tconn; | |
2604 | ||
2605 | tconn = kzalloc(sizeof(struct drbd_tconn), GFP_KERNEL); | |
2606 | if (!tconn) | |
2607 | return NULL; | |
2608 | ||
2609 | tconn->name = kstrdup(name, GFP_KERNEL); | |
2610 | if (!tconn->name) | |
2611 | goto fail; | |
2612 | ||
e6ef8a5c AG |
2613 | if (drbd_alloc_socket(&tconn->data)) |
2614 | goto fail; | |
2615 | if (drbd_alloc_socket(&tconn->meta)) | |
2616 | goto fail; | |
2617 | ||
774b3055 PR |
2618 | if (!zalloc_cpumask_var(&tconn->cpu_mask, GFP_KERNEL)) |
2619 | goto fail; | |
2620 | ||
afbbfa88 AG |
2621 | if (set_resource_options(tconn, res_opts)) |
2622 | goto fail; | |
2623 | ||
2f5cdd0b PR |
2624 | if (!tl_init(tconn)) |
2625 | goto fail; | |
2626 | ||
4b0007c0 PR |
2627 | tconn->write_ordering = WO_bdev_flush; |
2628 | ||
bbeb641c | 2629 | tconn->cstate = C_STANDALONE; |
8410da8f | 2630 | mutex_init(&tconn->cstate_mutex); |
6699b655 | 2631 | spin_lock_init(&tconn->req_lock); |
a0095508 | 2632 | mutex_init(&tconn->conf_update); |
2a67d8b9 | 2633 | init_waitqueue_head(&tconn->ping_wait); |
062e879c | 2634 | idr_init(&tconn->volumes); |
b2fb6dbe | 2635 | |
6699b655 PR |
2636 | drbd_init_workqueue(&tconn->data.work); |
2637 | mutex_init(&tconn->data.mutex); | |
2638 | ||
2639 | drbd_init_workqueue(&tconn->meta.work); | |
2640 | mutex_init(&tconn->meta.mutex); | |
2641 | ||
392c8801 PR |
2642 | drbd_thread_init(tconn, &tconn->receiver, drbdd_init, "receiver"); |
2643 | drbd_thread_init(tconn, &tconn->worker, drbd_worker, "worker"); | |
2644 | drbd_thread_init(tconn, &tconn->asender, drbd_asender, "asender"); | |
2645 | ||
9dc9fbb3 | 2646 | kref_init(&tconn->kref); |
ec0bddbc | 2647 | list_add_tail_rcu(&tconn->all_tconn, &drbd_tconns); |
2111438b PR |
2648 | |
2649 | return tconn; | |
2650 | ||
2651 | fail: | |
2f5cdd0b | 2652 | tl_cleanup(tconn); |
774b3055 | 2653 | free_cpumask_var(tconn->cpu_mask); |
e6ef8a5c AG |
2654 | drbd_free_socket(&tconn->meta); |
2655 | drbd_free_socket(&tconn->data); | |
2111438b PR |
2656 | kfree(tconn->name); |
2657 | kfree(tconn); | |
2658 | ||
2659 | return NULL; | |
2660 | } | |
2661 | ||
9dc9fbb3 | 2662 | void conn_destroy(struct kref *kref) |
2111438b | 2663 | { |
9dc9fbb3 PR |
2664 | struct drbd_tconn *tconn = container_of(kref, struct drbd_tconn, kref); |
2665 | ||
062e879c | 2666 | idr_destroy(&tconn->volumes); |
2111438b | 2667 | |
774b3055 | 2668 | free_cpumask_var(tconn->cpu_mask); |
e6ef8a5c AG |
2669 | drbd_free_socket(&tconn->meta); |
2670 | drbd_free_socket(&tconn->data); | |
2111438b | 2671 | kfree(tconn->name); |
b42a70ad PR |
2672 | kfree(tconn->int_dig_in); |
2673 | kfree(tconn->int_dig_vv); | |
2111438b PR |
2674 | kfree(tconn); |
2675 | } | |
2676 | ||
774b3055 | 2677 | enum drbd_ret_code conn_new_minor(struct drbd_tconn *tconn, unsigned int minor, int vnr) |
b411b363 PR |
2678 | { |
2679 | struct drbd_conf *mdev; | |
2680 | struct gendisk *disk; | |
2681 | struct request_queue *q; | |
774b3055 | 2682 | int vnr_got = vnr; |
81a5d60e | 2683 | int minor_got = minor; |
8432b314 | 2684 | enum drbd_ret_code err = ERR_NOMEM; |
774b3055 PR |
2685 | |
2686 | mdev = minor_to_mdev(minor); | |
2687 | if (mdev) | |
2688 | return ERR_MINOR_EXISTS; | |
b411b363 PR |
2689 | |
2690 | /* GFP_KERNEL, we are outside of all write-out paths */ | |
2691 | mdev = kzalloc(sizeof(struct drbd_conf), GFP_KERNEL); | |
2692 | if (!mdev) | |
774b3055 PR |
2693 | return ERR_NOMEM; |
2694 | ||
9dc9fbb3 | 2695 | kref_get(&tconn->kref); |
774b3055 | 2696 | mdev->tconn = tconn; |
9dc9fbb3 | 2697 | |
b411b363 | 2698 | mdev->minor = minor; |
3b98c0c2 | 2699 | mdev->vnr = vnr; |
b411b363 PR |
2700 | |
2701 | drbd_init_set_defaults(mdev); | |
2702 | ||
2703 | q = blk_alloc_queue(GFP_KERNEL); | |
2704 | if (!q) | |
2705 | goto out_no_q; | |
2706 | mdev->rq_queue = q; | |
2707 | q->queuedata = mdev; | |
b411b363 PR |
2708 | |
2709 | disk = alloc_disk(1); | |
2710 | if (!disk) | |
2711 | goto out_no_disk; | |
2712 | mdev->vdisk = disk; | |
2713 | ||
81e84650 | 2714 | set_disk_ro(disk, true); |
b411b363 PR |
2715 | |
2716 | disk->queue = q; | |
2717 | disk->major = DRBD_MAJOR; | |
2718 | disk->first_minor = minor; | |
2719 | disk->fops = &drbd_ops; | |
2720 | sprintf(disk->disk_name, "drbd%d", minor); | |
2721 | disk->private_data = mdev; | |
2722 | ||
2723 | mdev->this_bdev = bdget(MKDEV(DRBD_MAJOR, minor)); | |
2724 | /* we have no partitions. we contain only ourselves. */ | |
2725 | mdev->this_bdev->bd_contains = mdev->this_bdev; | |
2726 | ||
2727 | q->backing_dev_info.congested_fn = drbd_congested; | |
2728 | q->backing_dev_info.congested_data = mdev; | |
2729 | ||
2f58dcfc | 2730 | blk_queue_make_request(q, drbd_make_request); |
99432fcc PR |
2731 | /* Setting the max_hw_sectors to an odd value of 8kibyte here |
2732 | This triggers a max_bio_size message upon first attach or connect */ | |
2733 | blk_queue_max_hw_sectors(q, DRBD_MAX_BIO_SIZE_SAFE >> 8); | |
b411b363 PR |
2734 | blk_queue_bounce_limit(q, BLK_BOUNCE_ANY); |
2735 | blk_queue_merge_bvec(q, drbd_merge_bvec); | |
87eeee41 | 2736 | q->queue_lock = &mdev->tconn->req_lock; /* needed since we use */ |
b411b363 PR |
2737 | |
2738 | mdev->md_io_page = alloc_page(GFP_KERNEL); | |
2739 | if (!mdev->md_io_page) | |
2740 | goto out_no_io_page; | |
2741 | ||
2742 | if (drbd_bm_init(mdev)) | |
2743 | goto out_no_bitmap; | |
dac1389c | 2744 | mdev->read_requests = RB_ROOT; |
de696716 | 2745 | mdev->write_requests = RB_ROOT; |
b411b363 | 2746 | |
b411b363 PR |
2747 | mdev->current_epoch = kzalloc(sizeof(struct drbd_epoch), GFP_KERNEL); |
2748 | if (!mdev->current_epoch) | |
2749 | goto out_no_epoch; | |
2750 | ||
2751 | INIT_LIST_HEAD(&mdev->current_epoch->list); | |
2752 | mdev->epochs = 1; | |
2753 | ||
81a5d60e | 2754 | if (!idr_pre_get(&minors, GFP_KERNEL)) |
8432b314 LE |
2755 | goto out_no_minor_idr; |
2756 | if (idr_get_new_above(&minors, mdev, minor, &minor_got)) | |
2757 | goto out_no_minor_idr; | |
81a5d60e | 2758 | if (minor_got != minor) { |
8432b314 LE |
2759 | err = ERR_MINOR_EXISTS; |
2760 | drbd_msg_put_info("requested minor exists already"); | |
569083c0 | 2761 | goto out_idr_remove_minor; |
81a5d60e | 2762 | } |
8432b314 LE |
2763 | |
2764 | if (!idr_pre_get(&tconn->volumes, GFP_KERNEL)) | |
2765 | goto out_idr_remove_minor; | |
2766 | if (idr_get_new_above(&tconn->volumes, mdev, vnr, &vnr_got)) | |
2767 | goto out_idr_remove_minor; | |
2768 | if (vnr_got != vnr) { | |
2769 | err = ERR_INVALID_REQUEST; | |
2770 | drbd_msg_put_info("requested volume exists already"); | |
2771 | goto out_idr_remove_vol; | |
2772 | } | |
774b3055 | 2773 | add_disk(disk); |
81fa2e67 | 2774 | kref_init(&mdev->kref); /* one ref for both idrs and the the add_disk */ |
774b3055 | 2775 | |
2325eb66 PR |
2776 | /* inherit the connection state */ |
2777 | mdev->state.conn = tconn->cstate; | |
2778 | if (mdev->state.conn == C_WF_REPORT_PARAMS) | |
c141ebda | 2779 | drbd_connected(mdev); |
2325eb66 | 2780 | |
774b3055 | 2781 | return NO_ERROR; |
b411b363 | 2782 | |
569083c0 LE |
2783 | out_idr_remove_vol: |
2784 | idr_remove(&tconn->volumes, vnr_got); | |
8432b314 LE |
2785 | out_idr_remove_minor: |
2786 | idr_remove(&minors, minor_got); | |
569083c0 | 2787 | synchronize_rcu(); |
8432b314 | 2788 | out_no_minor_idr: |
81a5d60e | 2789 | kfree(mdev->current_epoch); |
b411b363 | 2790 | out_no_epoch: |
b411b363 PR |
2791 | drbd_bm_cleanup(mdev); |
2792 | out_no_bitmap: | |
2793 | __free_page(mdev->md_io_page); | |
2794 | out_no_io_page: | |
2795 | put_disk(disk); | |
2796 | out_no_disk: | |
2797 | blk_cleanup_queue(q); | |
2798 | out_no_q: | |
b411b363 | 2799 | kfree(mdev); |
9dc9fbb3 | 2800 | kref_put(&tconn->kref, &conn_destroy); |
8432b314 | 2801 | return err; |
b411b363 PR |
2802 | } |
2803 | ||
b411b363 PR |
2804 | int __init drbd_init(void) |
2805 | { | |
2806 | int err; | |
2807 | ||
2b8a90b5 | 2808 | if (minor_count < DRBD_MINOR_COUNT_MIN || minor_count > DRBD_MINOR_COUNT_MAX) { |
b411b363 | 2809 | printk(KERN_ERR |
81a5d60e | 2810 | "drbd: invalid minor_count (%d)\n", minor_count); |
b411b363 PR |
2811 | #ifdef MODULE |
2812 | return -EINVAL; | |
2813 | #else | |
46530e85 | 2814 | minor_count = DRBD_MINOR_COUNT_DEF; |
b411b363 PR |
2815 | #endif |
2816 | } | |
2817 | ||
b411b363 PR |
2818 | err = register_blkdev(DRBD_MAJOR, "drbd"); |
2819 | if (err) { | |
2820 | printk(KERN_ERR | |
2821 | "drbd: unable to register block device major %d\n", | |
2822 | DRBD_MAJOR); | |
2823 | return err; | |
2824 | } | |
2825 | ||
3b98c0c2 LE |
2826 | err = drbd_genl_register(); |
2827 | if (err) { | |
2828 | printk(KERN_ERR "drbd: unable to register generic netlink family\n"); | |
2829 | goto fail; | |
2830 | } | |
2831 | ||
2832 | ||
b411b363 PR |
2833 | register_reboot_notifier(&drbd_notifier); |
2834 | ||
2835 | /* | |
2836 | * allocate all necessary structs | |
2837 | */ | |
2838 | err = -ENOMEM; | |
2839 | ||
2840 | init_waitqueue_head(&drbd_pp_wait); | |
2841 | ||
2842 | drbd_proc = NULL; /* play safe for drbd_cleanup */ | |
81a5d60e | 2843 | idr_init(&minors); |
b411b363 PR |
2844 | |
2845 | err = drbd_create_mempools(); | |
2846 | if (err) | |
3b98c0c2 | 2847 | goto fail; |
b411b363 | 2848 | |
8c484ee4 | 2849 | drbd_proc = proc_create_data("drbd", S_IFREG | S_IRUGO , NULL, &drbd_proc_fops, NULL); |
b411b363 PR |
2850 | if (!drbd_proc) { |
2851 | printk(KERN_ERR "drbd: unable to register proc file\n"); | |
3b98c0c2 | 2852 | goto fail; |
b411b363 PR |
2853 | } |
2854 | ||
2855 | rwlock_init(&global_state_lock); | |
2111438b | 2856 | INIT_LIST_HEAD(&drbd_tconns); |
b411b363 PR |
2857 | |
2858 | printk(KERN_INFO "drbd: initialized. " | |
2859 | "Version: " REL_VERSION " (api:%d/proto:%d-%d)\n", | |
2860 | API_VERSION, PRO_VERSION_MIN, PRO_VERSION_MAX); | |
2861 | printk(KERN_INFO "drbd: %s\n", drbd_buildtag()); | |
2862 | printk(KERN_INFO "drbd: registered as block device major %d\n", | |
2863 | DRBD_MAJOR); | |
b411b363 PR |
2864 | |
2865 | return 0; /* Success! */ | |
2866 | ||
3b98c0c2 | 2867 | fail: |
b411b363 PR |
2868 | drbd_cleanup(); |
2869 | if (err == -ENOMEM) | |
2870 | /* currently always the case */ | |
2871 | printk(KERN_ERR "drbd: ran out of memory\n"); | |
2872 | else | |
2873 | printk(KERN_ERR "drbd: initialization failure\n"); | |
2874 | return err; | |
2875 | } | |
2876 | ||
2877 | void drbd_free_bc(struct drbd_backing_dev *ldev) | |
2878 | { | |
2879 | if (ldev == NULL) | |
2880 | return; | |
2881 | ||
e525fd89 TH |
2882 | blkdev_put(ldev->backing_bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
2883 | blkdev_put(ldev->md_bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); | |
b411b363 PR |
2884 | |
2885 | kfree(ldev); | |
2886 | } | |
2887 | ||
360cc740 PR |
2888 | void drbd_free_sock(struct drbd_tconn *tconn) |
2889 | { | |
2890 | if (tconn->data.socket) { | |
2891 | mutex_lock(&tconn->data.mutex); | |
2892 | kernel_sock_shutdown(tconn->data.socket, SHUT_RDWR); | |
2893 | sock_release(tconn->data.socket); | |
2894 | tconn->data.socket = NULL; | |
2895 | mutex_unlock(&tconn->data.mutex); | |
b411b363 | 2896 | } |
360cc740 PR |
2897 | if (tconn->meta.socket) { |
2898 | mutex_lock(&tconn->meta.mutex); | |
2899 | kernel_sock_shutdown(tconn->meta.socket, SHUT_RDWR); | |
2900 | sock_release(tconn->meta.socket); | |
2901 | tconn->meta.socket = NULL; | |
2902 | mutex_unlock(&tconn->meta.mutex); | |
b411b363 PR |
2903 | } |
2904 | } | |
2905 | ||
b411b363 PR |
2906 | /* meta data management */ |
2907 | ||
2908 | struct meta_data_on_disk { | |
2909 | u64 la_size; /* last agreed size. */ | |
2910 | u64 uuid[UI_SIZE]; /* UUIDs. */ | |
2911 | u64 device_uuid; | |
2912 | u64 reserved_u64_1; | |
2913 | u32 flags; /* MDF */ | |
2914 | u32 magic; | |
2915 | u32 md_size_sect; | |
2916 | u32 al_offset; /* offset to this block */ | |
2917 | u32 al_nr_extents; /* important for restoring the AL */ | |
f399002e | 2918 | /* `-- act_log->nr_elements <-- ldev->dc.al_extents */ |
b411b363 PR |
2919 | u32 bm_offset; /* offset to the bitmap, from here */ |
2920 | u32 bm_bytes_per_bit; /* BM_BLOCK_SIZE */ | |
99432fcc PR |
2921 | u32 la_peer_max_bio_size; /* last peer max_bio_size */ |
2922 | u32 reserved_u32[3]; | |
b411b363 PR |
2923 | |
2924 | } __packed; | |
2925 | ||
2926 | /** | |
2927 | * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set | |
2928 | * @mdev: DRBD device. | |
2929 | */ | |
2930 | void drbd_md_sync(struct drbd_conf *mdev) | |
2931 | { | |
2932 | struct meta_data_on_disk *buffer; | |
2933 | sector_t sector; | |
2934 | int i; | |
2935 | ||
ee15b038 LE |
2936 | del_timer(&mdev->md_sync_timer); |
2937 | /* timer may be rearmed by drbd_md_mark_dirty() now. */ | |
b411b363 PR |
2938 | if (!test_and_clear_bit(MD_DIRTY, &mdev->flags)) |
2939 | return; | |
b411b363 PR |
2940 | |
2941 | /* We use here D_FAILED and not D_ATTACHING because we try to write | |
2942 | * metadata even if we detach due to a disk failure! */ | |
2943 | if (!get_ldev_if_state(mdev, D_FAILED)) | |
2944 | return; | |
2945 | ||
cdfda633 PR |
2946 | buffer = drbd_md_get_buffer(mdev); |
2947 | if (!buffer) | |
2948 | goto out; | |
2949 | ||
b411b363 PR |
2950 | memset(buffer, 0, 512); |
2951 | ||
2952 | buffer->la_size = cpu_to_be64(drbd_get_capacity(mdev->this_bdev)); | |
2953 | for (i = UI_CURRENT; i < UI_SIZE; i++) | |
2954 | buffer->uuid[i] = cpu_to_be64(mdev->ldev->md.uuid[i]); | |
2955 | buffer->flags = cpu_to_be32(mdev->ldev->md.flags); | |
d5d7ebd4 | 2956 | buffer->magic = cpu_to_be32(DRBD_MD_MAGIC_84_UNCLEAN); |
b411b363 PR |
2957 | |
2958 | buffer->md_size_sect = cpu_to_be32(mdev->ldev->md.md_size_sect); | |
2959 | buffer->al_offset = cpu_to_be32(mdev->ldev->md.al_offset); | |
2960 | buffer->al_nr_extents = cpu_to_be32(mdev->act_log->nr_elements); | |
2961 | buffer->bm_bytes_per_bit = cpu_to_be32(BM_BLOCK_SIZE); | |
2962 | buffer->device_uuid = cpu_to_be64(mdev->ldev->md.device_uuid); | |
2963 | ||
2964 | buffer->bm_offset = cpu_to_be32(mdev->ldev->md.bm_offset); | |
99432fcc | 2965 | buffer->la_peer_max_bio_size = cpu_to_be32(mdev->peer_max_bio_size); |
b411b363 PR |
2966 | |
2967 | D_ASSERT(drbd_md_ss__(mdev, mdev->ldev) == mdev->ldev->md.md_offset); | |
2968 | sector = mdev->ldev->md.md_offset; | |
2969 | ||
3fbf4d21 | 2970 | if (drbd_md_sync_page_io(mdev, mdev->ldev, sector, WRITE)) { |
b411b363 PR |
2971 | /* this was a try anyways ... */ |
2972 | dev_err(DEV, "meta data update failed!\n"); | |
81e84650 | 2973 | drbd_chk_io_error(mdev, 1, true); |
b411b363 PR |
2974 | } |
2975 | ||
2976 | /* Update mdev->ldev->md.la_size_sect, | |
2977 | * since we updated it on metadata. */ | |
2978 | mdev->ldev->md.la_size_sect = drbd_get_capacity(mdev->this_bdev); | |
2979 | ||
cdfda633 PR |
2980 | drbd_md_put_buffer(mdev); |
2981 | out: | |
b411b363 PR |
2982 | put_ldev(mdev); |
2983 | } | |
2984 | ||
2985 | /** | |
2986 | * drbd_md_read() - Reads in the meta data super block | |
2987 | * @mdev: DRBD device. | |
2988 | * @bdev: Device from which the meta data should be read in. | |
2989 | * | |
116676ca | 2990 | * Return 0 (NO_ERROR) on success, and an enum drbd_ret_code in case |
d5d7ebd4 | 2991 | * something goes wrong. |
b411b363 PR |
2992 | */ |
2993 | int drbd_md_read(struct drbd_conf *mdev, struct drbd_backing_dev *bdev) | |
2994 | { | |
2995 | struct meta_data_on_disk *buffer; | |
d5d7ebd4 | 2996 | u32 magic, flags; |
b411b363 PR |
2997 | int i, rv = NO_ERROR; |
2998 | ||
2999 | if (!get_ldev_if_state(mdev, D_ATTACHING)) | |
3000 | return ERR_IO_MD_DISK; | |
3001 | ||
cdfda633 PR |
3002 | buffer = drbd_md_get_buffer(mdev); |
3003 | if (!buffer) | |
3004 | goto out; | |
b411b363 | 3005 | |
3fbf4d21 | 3006 | if (drbd_md_sync_page_io(mdev, bdev, bdev->md.md_offset, READ)) { |
25985edc | 3007 | /* NOTE: can't do normal error processing here as this is |
b411b363 PR |
3008 | called BEFORE disk is attached */ |
3009 | dev_err(DEV, "Error while reading metadata.\n"); | |
3010 | rv = ERR_IO_MD_DISK; | |
3011 | goto err; | |
3012 | } | |
3013 | ||
d5d7ebd4 LE |
3014 | magic = be32_to_cpu(buffer->magic); |
3015 | flags = be32_to_cpu(buffer->flags); | |
3016 | if (magic == DRBD_MD_MAGIC_84_UNCLEAN || | |
3017 | (magic == DRBD_MD_MAGIC_08 && !(flags & MDF_AL_CLEAN))) { | |
3018 | /* btw: that's Activity Log clean, not "all" clean. */ | |
3019 | dev_err(DEV, "Found unclean meta data. Did you \"drbdadm apply-al\"?\n"); | |
3020 | rv = ERR_MD_UNCLEAN; | |
3021 | goto err; | |
3022 | } | |
3023 | if (magic != DRBD_MD_MAGIC_08) { | |
43de7c85 | 3024 | if (magic == DRBD_MD_MAGIC_07) |
d5d7ebd4 LE |
3025 | dev_err(DEV, "Found old (0.7) meta data magic. Did you \"drbdadm create-md\"?\n"); |
3026 | else | |
3027 | dev_err(DEV, "Meta data magic not found. Did you \"drbdadm create-md\"?\n"); | |
b411b363 PR |
3028 | rv = ERR_MD_INVALID; |
3029 | goto err; | |
3030 | } | |
3031 | if (be32_to_cpu(buffer->al_offset) != bdev->md.al_offset) { | |
3032 | dev_err(DEV, "unexpected al_offset: %d (expected %d)\n", | |
3033 | be32_to_cpu(buffer->al_offset), bdev->md.al_offset); | |
3034 | rv = ERR_MD_INVALID; | |
3035 | goto err; | |
3036 | } | |
3037 | if (be32_to_cpu(buffer->bm_offset) != bdev->md.bm_offset) { | |
3038 | dev_err(DEV, "unexpected bm_offset: %d (expected %d)\n", | |
3039 | be32_to_cpu(buffer->bm_offset), bdev->md.bm_offset); | |
3040 | rv = ERR_MD_INVALID; | |
3041 | goto err; | |
3042 | } | |
3043 | if (be32_to_cpu(buffer->md_size_sect) != bdev->md.md_size_sect) { | |
3044 | dev_err(DEV, "unexpected md_size: %u (expected %u)\n", | |
3045 | be32_to_cpu(buffer->md_size_sect), bdev->md.md_size_sect); | |
3046 | rv = ERR_MD_INVALID; | |
3047 | goto err; | |
3048 | } | |
3049 | ||
3050 | if (be32_to_cpu(buffer->bm_bytes_per_bit) != BM_BLOCK_SIZE) { | |
3051 | dev_err(DEV, "unexpected bm_bytes_per_bit: %u (expected %u)\n", | |
3052 | be32_to_cpu(buffer->bm_bytes_per_bit), BM_BLOCK_SIZE); | |
3053 | rv = ERR_MD_INVALID; | |
3054 | goto err; | |
3055 | } | |
3056 | ||
3057 | bdev->md.la_size_sect = be64_to_cpu(buffer->la_size); | |
3058 | for (i = UI_CURRENT; i < UI_SIZE; i++) | |
3059 | bdev->md.uuid[i] = be64_to_cpu(buffer->uuid[i]); | |
3060 | bdev->md.flags = be32_to_cpu(buffer->flags); | |
b411b363 PR |
3061 | bdev->md.device_uuid = be64_to_cpu(buffer->device_uuid); |
3062 | ||
87eeee41 | 3063 | spin_lock_irq(&mdev->tconn->req_lock); |
99432fcc PR |
3064 | if (mdev->state.conn < C_CONNECTED) { |
3065 | int peer; | |
3066 | peer = be32_to_cpu(buffer->la_peer_max_bio_size); | |
3067 | peer = max_t(int, peer, DRBD_MAX_BIO_SIZE_SAFE); | |
3068 | mdev->peer_max_bio_size = peer; | |
3069 | } | |
87eeee41 | 3070 | spin_unlock_irq(&mdev->tconn->req_lock); |
99432fcc | 3071 | |
b411b363 | 3072 | err: |
cdfda633 PR |
3073 | drbd_md_put_buffer(mdev); |
3074 | out: | |
b411b363 PR |
3075 | put_ldev(mdev); |
3076 | ||
3077 | return rv; | |
3078 | } | |
3079 | ||
3080 | /** | |
3081 | * drbd_md_mark_dirty() - Mark meta data super block as dirty | |
3082 | * @mdev: DRBD device. | |
3083 | * | |
3084 | * Call this function if you change anything that should be written to | |
3085 | * the meta-data super block. This function sets MD_DIRTY, and starts a | |
3086 | * timer that ensures that within five seconds you have to call drbd_md_sync(). | |
3087 | */ | |
ca0e6098 | 3088 | #ifdef DEBUG |
ee15b038 LE |
3089 | void drbd_md_mark_dirty_(struct drbd_conf *mdev, unsigned int line, const char *func) |
3090 | { | |
3091 | if (!test_and_set_bit(MD_DIRTY, &mdev->flags)) { | |
3092 | mod_timer(&mdev->md_sync_timer, jiffies + HZ); | |
3093 | mdev->last_md_mark_dirty.line = line; | |
3094 | mdev->last_md_mark_dirty.func = func; | |
3095 | } | |
3096 | } | |
3097 | #else | |
b411b363 PR |
3098 | void drbd_md_mark_dirty(struct drbd_conf *mdev) |
3099 | { | |
ee15b038 | 3100 | if (!test_and_set_bit(MD_DIRTY, &mdev->flags)) |
ca0e6098 | 3101 | mod_timer(&mdev->md_sync_timer, jiffies + 5*HZ); |
b411b363 | 3102 | } |
ee15b038 | 3103 | #endif |
b411b363 PR |
3104 | |
3105 | static void drbd_uuid_move_history(struct drbd_conf *mdev) __must_hold(local) | |
3106 | { | |
3107 | int i; | |
3108 | ||
62b0da3a | 3109 | for (i = UI_HISTORY_START; i < UI_HISTORY_END; i++) |
b411b363 | 3110 | mdev->ldev->md.uuid[i+1] = mdev->ldev->md.uuid[i]; |
b411b363 PR |
3111 | } |
3112 | ||
3113 | void _drbd_uuid_set(struct drbd_conf *mdev, int idx, u64 val) __must_hold(local) | |
3114 | { | |
3115 | if (idx == UI_CURRENT) { | |
3116 | if (mdev->state.role == R_PRIMARY) | |
3117 | val |= 1; | |
3118 | else | |
3119 | val &= ~((u64)1); | |
3120 | ||
3121 | drbd_set_ed_uuid(mdev, val); | |
3122 | } | |
3123 | ||
3124 | mdev->ldev->md.uuid[idx] = val; | |
b411b363 PR |
3125 | drbd_md_mark_dirty(mdev); |
3126 | } | |
3127 | ||
3128 | ||
3129 | void drbd_uuid_set(struct drbd_conf *mdev, int idx, u64 val) __must_hold(local) | |
3130 | { | |
3131 | if (mdev->ldev->md.uuid[idx]) { | |
3132 | drbd_uuid_move_history(mdev); | |
3133 | mdev->ldev->md.uuid[UI_HISTORY_START] = mdev->ldev->md.uuid[idx]; | |
b411b363 PR |
3134 | } |
3135 | _drbd_uuid_set(mdev, idx, val); | |
3136 | } | |
3137 | ||
3138 | /** | |
3139 | * drbd_uuid_new_current() - Creates a new current UUID | |
3140 | * @mdev: DRBD device. | |
3141 | * | |
3142 | * Creates a new current UUID, and rotates the old current UUID into | |
3143 | * the bitmap slot. Causes an incremental resync upon next connect. | |
3144 | */ | |
3145 | void drbd_uuid_new_current(struct drbd_conf *mdev) __must_hold(local) | |
3146 | { | |
3147 | u64 val; | |
62b0da3a LE |
3148 | unsigned long long bm_uuid = mdev->ldev->md.uuid[UI_BITMAP]; |
3149 | ||
3150 | if (bm_uuid) | |
3151 | dev_warn(DEV, "bm UUID was already set: %llX\n", bm_uuid); | |
b411b363 | 3152 | |
b411b363 | 3153 | mdev->ldev->md.uuid[UI_BITMAP] = mdev->ldev->md.uuid[UI_CURRENT]; |
b411b363 PR |
3154 | |
3155 | get_random_bytes(&val, sizeof(u64)); | |
3156 | _drbd_uuid_set(mdev, UI_CURRENT, val); | |
62b0da3a | 3157 | drbd_print_uuids(mdev, "new current UUID"); |
aaa8e2b3 LE |
3158 | /* get it to stable storage _now_ */ |
3159 | drbd_md_sync(mdev); | |
b411b363 PR |
3160 | } |
3161 | ||
3162 | void drbd_uuid_set_bm(struct drbd_conf *mdev, u64 val) __must_hold(local) | |
3163 | { | |
3164 | if (mdev->ldev->md.uuid[UI_BITMAP] == 0 && val == 0) | |
3165 | return; | |
3166 | ||
3167 | if (val == 0) { | |
3168 | drbd_uuid_move_history(mdev); | |
3169 | mdev->ldev->md.uuid[UI_HISTORY_START] = mdev->ldev->md.uuid[UI_BITMAP]; | |
3170 | mdev->ldev->md.uuid[UI_BITMAP] = 0; | |
b411b363 | 3171 | } else { |
62b0da3a LE |
3172 | unsigned long long bm_uuid = mdev->ldev->md.uuid[UI_BITMAP]; |
3173 | if (bm_uuid) | |
3174 | dev_warn(DEV, "bm UUID was already set: %llX\n", bm_uuid); | |
b411b363 | 3175 | |
62b0da3a | 3176 | mdev->ldev->md.uuid[UI_BITMAP] = val & ~((u64)1); |
b411b363 PR |
3177 | } |
3178 | drbd_md_mark_dirty(mdev); | |
3179 | } | |
3180 | ||
3181 | /** | |
3182 | * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io() | |
3183 | * @mdev: DRBD device. | |
3184 | * | |
3185 | * Sets all bits in the bitmap and writes the whole bitmap to stable storage. | |
3186 | */ | |
3187 | int drbd_bmio_set_n_write(struct drbd_conf *mdev) | |
3188 | { | |
3189 | int rv = -EIO; | |
3190 | ||
3191 | if (get_ldev_if_state(mdev, D_ATTACHING)) { | |
3192 | drbd_md_set_flag(mdev, MDF_FULL_SYNC); | |
3193 | drbd_md_sync(mdev); | |
3194 | drbd_bm_set_all(mdev); | |
3195 | ||
3196 | rv = drbd_bm_write(mdev); | |
3197 | ||
3198 | if (!rv) { | |
3199 | drbd_md_clear_flag(mdev, MDF_FULL_SYNC); | |
3200 | drbd_md_sync(mdev); | |
3201 | } | |
3202 | ||
3203 | put_ldev(mdev); | |
3204 | } | |
3205 | ||
3206 | return rv; | |
3207 | } | |
3208 | ||
3209 | /** | |
3210 | * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io() | |
3211 | * @mdev: DRBD device. | |
3212 | * | |
3213 | * Clears all bits in the bitmap and writes the whole bitmap to stable storage. | |
3214 | */ | |
3215 | int drbd_bmio_clear_n_write(struct drbd_conf *mdev) | |
3216 | { | |
3217 | int rv = -EIO; | |
3218 | ||
0778286a | 3219 | drbd_resume_al(mdev); |
b411b363 PR |
3220 | if (get_ldev_if_state(mdev, D_ATTACHING)) { |
3221 | drbd_bm_clear_all(mdev); | |
3222 | rv = drbd_bm_write(mdev); | |
3223 | put_ldev(mdev); | |
3224 | } | |
3225 | ||
3226 | return rv; | |
3227 | } | |
3228 | ||
99920dc5 | 3229 | static int w_bitmap_io(struct drbd_work *w, int unused) |
b411b363 PR |
3230 | { |
3231 | struct bm_io_work *work = container_of(w, struct bm_io_work, w); | |
00d56944 | 3232 | struct drbd_conf *mdev = w->mdev; |
02851e9f | 3233 | int rv = -EIO; |
b411b363 PR |
3234 | |
3235 | D_ASSERT(atomic_read(&mdev->ap_bio_cnt) == 0); | |
3236 | ||
02851e9f | 3237 | if (get_ldev(mdev)) { |
20ceb2b2 | 3238 | drbd_bm_lock(mdev, work->why, work->flags); |
02851e9f LE |
3239 | rv = work->io_fn(mdev); |
3240 | drbd_bm_unlock(mdev); | |
3241 | put_ldev(mdev); | |
3242 | } | |
b411b363 | 3243 | |
4738fa16 | 3244 | clear_bit_unlock(BITMAP_IO, &mdev->flags); |
b411b363 PR |
3245 | wake_up(&mdev->misc_wait); |
3246 | ||
3247 | if (work->done) | |
3248 | work->done(mdev, rv); | |
3249 | ||
3250 | clear_bit(BITMAP_IO_QUEUED, &mdev->flags); | |
3251 | work->why = NULL; | |
20ceb2b2 | 3252 | work->flags = 0; |
b411b363 | 3253 | |
99920dc5 | 3254 | return 0; |
b411b363 PR |
3255 | } |
3256 | ||
82f59cc6 LE |
3257 | void drbd_ldev_destroy(struct drbd_conf *mdev) |
3258 | { | |
3259 | lc_destroy(mdev->resync); | |
3260 | mdev->resync = NULL; | |
3261 | lc_destroy(mdev->act_log); | |
3262 | mdev->act_log = NULL; | |
3263 | __no_warn(local, | |
3264 | drbd_free_bc(mdev->ldev); | |
3265 | mdev->ldev = NULL;); | |
3266 | ||
82f59cc6 LE |
3267 | clear_bit(GO_DISKLESS, &mdev->flags); |
3268 | } | |
3269 | ||
99920dc5 | 3270 | static int w_go_diskless(struct drbd_work *w, int unused) |
e9e6f3ec | 3271 | { |
00d56944 PR |
3272 | struct drbd_conf *mdev = w->mdev; |
3273 | ||
e9e6f3ec | 3274 | D_ASSERT(mdev->state.disk == D_FAILED); |
9d282875 LE |
3275 | /* we cannot assert local_cnt == 0 here, as get_ldev_if_state will |
3276 | * inc/dec it frequently. Once we are D_DISKLESS, no one will touch | |
82f59cc6 LE |
3277 | * the protected members anymore, though, so once put_ldev reaches zero |
3278 | * again, it will be safe to free them. */ | |
e9e6f3ec | 3279 | drbd_force_state(mdev, NS(disk, D_DISKLESS)); |
99920dc5 | 3280 | return 0; |
e9e6f3ec LE |
3281 | } |
3282 | ||
3283 | void drbd_go_diskless(struct drbd_conf *mdev) | |
3284 | { | |
3285 | D_ASSERT(mdev->state.disk == D_FAILED); | |
3286 | if (!test_and_set_bit(GO_DISKLESS, &mdev->flags)) | |
e42325a5 | 3287 | drbd_queue_work(&mdev->tconn->data.work, &mdev->go_diskless); |
e9e6f3ec LE |
3288 | } |
3289 | ||
b411b363 PR |
3290 | /** |
3291 | * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap | |
3292 | * @mdev: DRBD device. | |
3293 | * @io_fn: IO callback to be called when bitmap IO is possible | |
3294 | * @done: callback to be called after the bitmap IO was performed | |
3295 | * @why: Descriptive text of the reason for doing the IO | |
3296 | * | |
3297 | * While IO on the bitmap happens we freeze application IO thus we ensure | |
3298 | * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be | |
3299 | * called from worker context. It MUST NOT be used while a previous such | |
3300 | * work is still pending! | |
3301 | */ | |
3302 | void drbd_queue_bitmap_io(struct drbd_conf *mdev, | |
3303 | int (*io_fn)(struct drbd_conf *), | |
3304 | void (*done)(struct drbd_conf *, int), | |
20ceb2b2 | 3305 | char *why, enum bm_flag flags) |
b411b363 | 3306 | { |
e6b3ea83 | 3307 | D_ASSERT(current == mdev->tconn->worker.task); |
b411b363 PR |
3308 | |
3309 | D_ASSERT(!test_bit(BITMAP_IO_QUEUED, &mdev->flags)); | |
3310 | D_ASSERT(!test_bit(BITMAP_IO, &mdev->flags)); | |
3311 | D_ASSERT(list_empty(&mdev->bm_io_work.w.list)); | |
3312 | if (mdev->bm_io_work.why) | |
3313 | dev_err(DEV, "FIXME going to queue '%s' but '%s' still pending?\n", | |
3314 | why, mdev->bm_io_work.why); | |
3315 | ||
3316 | mdev->bm_io_work.io_fn = io_fn; | |
3317 | mdev->bm_io_work.done = done; | |
3318 | mdev->bm_io_work.why = why; | |
20ceb2b2 | 3319 | mdev->bm_io_work.flags = flags; |
b411b363 | 3320 | |
87eeee41 | 3321 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
3322 | set_bit(BITMAP_IO, &mdev->flags); |
3323 | if (atomic_read(&mdev->ap_bio_cnt) == 0) { | |
127b3178 | 3324 | if (!test_and_set_bit(BITMAP_IO_QUEUED, &mdev->flags)) |
e42325a5 | 3325 | drbd_queue_work(&mdev->tconn->data.work, &mdev->bm_io_work.w); |
b411b363 | 3326 | } |
87eeee41 | 3327 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
3328 | } |
3329 | ||
3330 | /** | |
3331 | * drbd_bitmap_io() - Does an IO operation on the whole bitmap | |
3332 | * @mdev: DRBD device. | |
3333 | * @io_fn: IO callback to be called when bitmap IO is possible | |
3334 | * @why: Descriptive text of the reason for doing the IO | |
3335 | * | |
3336 | * freezes application IO while that the actual IO operations runs. This | |
3337 | * functions MAY NOT be called from worker context. | |
3338 | */ | |
20ceb2b2 LE |
3339 | int drbd_bitmap_io(struct drbd_conf *mdev, int (*io_fn)(struct drbd_conf *), |
3340 | char *why, enum bm_flag flags) | |
b411b363 PR |
3341 | { |
3342 | int rv; | |
3343 | ||
e6b3ea83 | 3344 | D_ASSERT(current != mdev->tconn->worker.task); |
b411b363 | 3345 | |
20ceb2b2 LE |
3346 | if ((flags & BM_LOCKED_SET_ALLOWED) == 0) |
3347 | drbd_suspend_io(mdev); | |
b411b363 | 3348 | |
20ceb2b2 | 3349 | drbd_bm_lock(mdev, why, flags); |
b411b363 PR |
3350 | rv = io_fn(mdev); |
3351 | drbd_bm_unlock(mdev); | |
3352 | ||
20ceb2b2 LE |
3353 | if ((flags & BM_LOCKED_SET_ALLOWED) == 0) |
3354 | drbd_resume_io(mdev); | |
b411b363 PR |
3355 | |
3356 | return rv; | |
3357 | } | |
3358 | ||
3359 | void drbd_md_set_flag(struct drbd_conf *mdev, int flag) __must_hold(local) | |
3360 | { | |
3361 | if ((mdev->ldev->md.flags & flag) != flag) { | |
3362 | drbd_md_mark_dirty(mdev); | |
3363 | mdev->ldev->md.flags |= flag; | |
3364 | } | |
3365 | } | |
3366 | ||
3367 | void drbd_md_clear_flag(struct drbd_conf *mdev, int flag) __must_hold(local) | |
3368 | { | |
3369 | if ((mdev->ldev->md.flags & flag) != 0) { | |
3370 | drbd_md_mark_dirty(mdev); | |
3371 | mdev->ldev->md.flags &= ~flag; | |
3372 | } | |
3373 | } | |
3374 | int drbd_md_test_flag(struct drbd_backing_dev *bdev, int flag) | |
3375 | { | |
3376 | return (bdev->md.flags & flag) != 0; | |
3377 | } | |
3378 | ||
3379 | static void md_sync_timer_fn(unsigned long data) | |
3380 | { | |
3381 | struct drbd_conf *mdev = (struct drbd_conf *) data; | |
3382 | ||
e42325a5 | 3383 | drbd_queue_work_front(&mdev->tconn->data.work, &mdev->md_sync_work); |
b411b363 PR |
3384 | } |
3385 | ||
99920dc5 | 3386 | static int w_md_sync(struct drbd_work *w, int unused) |
b411b363 | 3387 | { |
00d56944 PR |
3388 | struct drbd_conf *mdev = w->mdev; |
3389 | ||
b411b363 | 3390 | dev_warn(DEV, "md_sync_timer expired! Worker calls drbd_md_sync().\n"); |
ee15b038 LE |
3391 | #ifdef DEBUG |
3392 | dev_warn(DEV, "last md_mark_dirty: %s:%u\n", | |
3393 | mdev->last_md_mark_dirty.func, mdev->last_md_mark_dirty.line); | |
3394 | #endif | |
b411b363 | 3395 | drbd_md_sync(mdev); |
99920dc5 | 3396 | return 0; |
b411b363 PR |
3397 | } |
3398 | ||
d8763023 | 3399 | const char *cmdname(enum drbd_packet cmd) |
f2ad9063 AG |
3400 | { |
3401 | /* THINK may need to become several global tables | |
3402 | * when we want to support more than | |
3403 | * one PRO_VERSION */ | |
3404 | static const char *cmdnames[] = { | |
3405 | [P_DATA] = "Data", | |
3406 | [P_DATA_REPLY] = "DataReply", | |
3407 | [P_RS_DATA_REPLY] = "RSDataReply", | |
3408 | [P_BARRIER] = "Barrier", | |
3409 | [P_BITMAP] = "ReportBitMap", | |
3410 | [P_BECOME_SYNC_TARGET] = "BecomeSyncTarget", | |
3411 | [P_BECOME_SYNC_SOURCE] = "BecomeSyncSource", | |
3412 | [P_UNPLUG_REMOTE] = "UnplugRemote", | |
3413 | [P_DATA_REQUEST] = "DataRequest", | |
3414 | [P_RS_DATA_REQUEST] = "RSDataRequest", | |
3415 | [P_SYNC_PARAM] = "SyncParam", | |
3416 | [P_SYNC_PARAM89] = "SyncParam89", | |
3417 | [P_PROTOCOL] = "ReportProtocol", | |
3418 | [P_UUIDS] = "ReportUUIDs", | |
3419 | [P_SIZES] = "ReportSizes", | |
3420 | [P_STATE] = "ReportState", | |
3421 | [P_SYNC_UUID] = "ReportSyncUUID", | |
3422 | [P_AUTH_CHALLENGE] = "AuthChallenge", | |
3423 | [P_AUTH_RESPONSE] = "AuthResponse", | |
3424 | [P_PING] = "Ping", | |
3425 | [P_PING_ACK] = "PingAck", | |
3426 | [P_RECV_ACK] = "RecvAck", | |
3427 | [P_WRITE_ACK] = "WriteAck", | |
3428 | [P_RS_WRITE_ACK] = "RSWriteAck", | |
7be8da07 | 3429 | [P_DISCARD_WRITE] = "DiscardWrite", |
f2ad9063 AG |
3430 | [P_NEG_ACK] = "NegAck", |
3431 | [P_NEG_DREPLY] = "NegDReply", | |
3432 | [P_NEG_RS_DREPLY] = "NegRSDReply", | |
3433 | [P_BARRIER_ACK] = "BarrierAck", | |
3434 | [P_STATE_CHG_REQ] = "StateChgRequest", | |
3435 | [P_STATE_CHG_REPLY] = "StateChgReply", | |
3436 | [P_OV_REQUEST] = "OVRequest", | |
3437 | [P_OV_REPLY] = "OVReply", | |
3438 | [P_OV_RESULT] = "OVResult", | |
3439 | [P_CSUM_RS_REQUEST] = "CsumRSRequest", | |
3440 | [P_RS_IS_IN_SYNC] = "CsumRSIsInSync", | |
3441 | [P_COMPRESSED_BITMAP] = "CBitmap", | |
3442 | [P_DELAY_PROBE] = "DelayProbe", | |
3443 | [P_OUT_OF_SYNC] = "OutOfSync", | |
7be8da07 | 3444 | [P_RETRY_WRITE] = "RetryWrite", |
ae25b336 LE |
3445 | [P_RS_CANCEL] = "RSCancel", |
3446 | [P_CONN_ST_CHG_REQ] = "conn_st_chg_req", | |
3447 | [P_CONN_ST_CHG_REPLY] = "conn_st_chg_reply", | |
036b17ea PR |
3448 | [P_RETRY_WRITE] = "retry_write", |
3449 | [P_PROTOCOL_UPDATE] = "protocol_update", | |
ae25b336 LE |
3450 | |
3451 | /* enum drbd_packet, but not commands - obsoleted flags: | |
3452 | * P_MAY_IGNORE | |
3453 | * P_MAX_OPT_CMD | |
3454 | */ | |
f2ad9063 AG |
3455 | }; |
3456 | ||
ae25b336 | 3457 | /* too big for the array: 0xfffX */ |
e5d6f33a AG |
3458 | if (cmd == P_INITIAL_META) |
3459 | return "InitialMeta"; | |
3460 | if (cmd == P_INITIAL_DATA) | |
3461 | return "InitialData"; | |
6038178e AG |
3462 | if (cmd == P_CONNECTION_FEATURES) |
3463 | return "ConnectionFeatures"; | |
6e849ce8 | 3464 | if (cmd >= ARRAY_SIZE(cmdnames)) |
f2ad9063 AG |
3465 | return "Unknown"; |
3466 | return cmdnames[cmd]; | |
3467 | } | |
3468 | ||
7be8da07 AG |
3469 | /** |
3470 | * drbd_wait_misc - wait for a request to make progress | |
3471 | * @mdev: device associated with the request | |
3472 | * @i: the struct drbd_interval embedded in struct drbd_request or | |
3473 | * struct drbd_peer_request | |
3474 | */ | |
3475 | int drbd_wait_misc(struct drbd_conf *mdev, struct drbd_interval *i) | |
3476 | { | |
44ed167d | 3477 | struct net_conf *nc; |
7be8da07 AG |
3478 | DEFINE_WAIT(wait); |
3479 | long timeout; | |
3480 | ||
44ed167d PR |
3481 | rcu_read_lock(); |
3482 | nc = rcu_dereference(mdev->tconn->net_conf); | |
3483 | if (!nc) { | |
3484 | rcu_read_unlock(); | |
7be8da07 | 3485 | return -ETIMEDOUT; |
44ed167d PR |
3486 | } |
3487 | timeout = nc->ko_count ? nc->timeout * HZ / 10 * nc->ko_count : MAX_SCHEDULE_TIMEOUT; | |
3488 | rcu_read_unlock(); | |
7be8da07 AG |
3489 | |
3490 | /* Indicate to wake up mdev->misc_wait on progress. */ | |
3491 | i->waiting = true; | |
3492 | prepare_to_wait(&mdev->misc_wait, &wait, TASK_INTERRUPTIBLE); | |
3493 | spin_unlock_irq(&mdev->tconn->req_lock); | |
3494 | timeout = schedule_timeout(timeout); | |
3495 | finish_wait(&mdev->misc_wait, &wait); | |
3496 | spin_lock_irq(&mdev->tconn->req_lock); | |
3497 | if (!timeout || mdev->state.conn < C_CONNECTED) | |
3498 | return -ETIMEDOUT; | |
3499 | if (signal_pending(current)) | |
3500 | return -ERESTARTSYS; | |
3501 | return 0; | |
3502 | } | |
3503 | ||
b411b363 PR |
3504 | #ifdef CONFIG_DRBD_FAULT_INJECTION |
3505 | /* Fault insertion support including random number generator shamelessly | |
3506 | * stolen from kernel/rcutorture.c */ | |
3507 | struct fault_random_state { | |
3508 | unsigned long state; | |
3509 | unsigned long count; | |
3510 | }; | |
3511 | ||
3512 | #define FAULT_RANDOM_MULT 39916801 /* prime */ | |
3513 | #define FAULT_RANDOM_ADD 479001701 /* prime */ | |
3514 | #define FAULT_RANDOM_REFRESH 10000 | |
3515 | ||
3516 | /* | |
3517 | * Crude but fast random-number generator. Uses a linear congruential | |
3518 | * generator, with occasional help from get_random_bytes(). | |
3519 | */ | |
3520 | static unsigned long | |
3521 | _drbd_fault_random(struct fault_random_state *rsp) | |
3522 | { | |
3523 | long refresh; | |
3524 | ||
49829ea7 | 3525 | if (!rsp->count--) { |
b411b363 PR |
3526 | get_random_bytes(&refresh, sizeof(refresh)); |
3527 | rsp->state += refresh; | |
3528 | rsp->count = FAULT_RANDOM_REFRESH; | |
3529 | } | |
3530 | rsp->state = rsp->state * FAULT_RANDOM_MULT + FAULT_RANDOM_ADD; | |
3531 | return swahw32(rsp->state); | |
3532 | } | |
3533 | ||
3534 | static char * | |
3535 | _drbd_fault_str(unsigned int type) { | |
3536 | static char *_faults[] = { | |
3537 | [DRBD_FAULT_MD_WR] = "Meta-data write", | |
3538 | [DRBD_FAULT_MD_RD] = "Meta-data read", | |
3539 | [DRBD_FAULT_RS_WR] = "Resync write", | |
3540 | [DRBD_FAULT_RS_RD] = "Resync read", | |
3541 | [DRBD_FAULT_DT_WR] = "Data write", | |
3542 | [DRBD_FAULT_DT_RD] = "Data read", | |
3543 | [DRBD_FAULT_DT_RA] = "Data read ahead", | |
3544 | [DRBD_FAULT_BM_ALLOC] = "BM allocation", | |
6b4388ac PR |
3545 | [DRBD_FAULT_AL_EE] = "EE allocation", |
3546 | [DRBD_FAULT_RECEIVE] = "receive data corruption", | |
b411b363 PR |
3547 | }; |
3548 | ||
3549 | return (type < DRBD_FAULT_MAX) ? _faults[type] : "**Unknown**"; | |
3550 | } | |
3551 | ||
3552 | unsigned int | |
3553 | _drbd_insert_fault(struct drbd_conf *mdev, unsigned int type) | |
3554 | { | |
3555 | static struct fault_random_state rrs = {0, 0}; | |
3556 | ||
3557 | unsigned int ret = ( | |
3558 | (fault_devs == 0 || | |
3559 | ((1 << mdev_to_minor(mdev)) & fault_devs) != 0) && | |
3560 | (((_drbd_fault_random(&rrs) % 100) + 1) <= fault_rate)); | |
3561 | ||
3562 | if (ret) { | |
3563 | fault_count++; | |
3564 | ||
7383506c | 3565 | if (__ratelimit(&drbd_ratelimit_state)) |
b411b363 PR |
3566 | dev_warn(DEV, "***Simulating %s failure\n", |
3567 | _drbd_fault_str(type)); | |
3568 | } | |
3569 | ||
3570 | return ret; | |
3571 | } | |
3572 | #endif | |
3573 | ||
3574 | const char *drbd_buildtag(void) | |
3575 | { | |
3576 | /* DRBD built from external sources has here a reference to the | |
3577 | git hash of the source code. */ | |
3578 | ||
3579 | static char buildtag[38] = "\0uilt-in"; | |
3580 | ||
3581 | if (buildtag[0] == 0) { | |
3582 | #ifdef CONFIG_MODULES | |
3583 | if (THIS_MODULE != NULL) | |
3584 | sprintf(buildtag, "srcversion: %-24s", THIS_MODULE->srcversion); | |
3585 | else | |
3586 | #endif | |
3587 | buildtag[0] = 'b'; | |
3588 | } | |
3589 | ||
3590 | return buildtag; | |
3591 | } | |
3592 | ||
3593 | module_init(drbd_init) | |
3594 | module_exit(drbd_cleanup) | |
3595 | ||
b411b363 PR |
3596 | EXPORT_SYMBOL(drbd_conn_str); |
3597 | EXPORT_SYMBOL(drbd_role_str); | |
3598 | EXPORT_SYMBOL(drbd_disk_str); | |
3599 | EXPORT_SYMBOL(drbd_set_st_err_str); |