projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
The DRBD driver
[linux-block.git] / drivers / block / drbd / drbd_main.c
CommitLineData
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
29#include <linux/autoconf.h>
30#include <linux/module.h>
31#include <linux/version.h>
32#include <linux/drbd.h>
33#include <asm/uaccess.h>
34#include <asm/types.h>
35#include <net/sock.h>
36#include <linux/ctype.h>
37#include <linux/smp_lock.h>
38#include <linux/fs.h>
39#include <linux/file.h>
40#include <linux/proc_fs.h>
41#include <linux/init.h>
42#include <linux/mm.h>
43#include <linux/memcontrol.h>
44#include <linux/mm_inline.h>
45#include <linux/slab.h>
46#include <linux/random.h>
47#include <linux/reboot.h>
48#include <linux/notifier.h>
49#include <linux/kthread.h>
50
51#define __KERNEL_SYSCALLS__
52#include <linux/unistd.h>
53#include <linux/vmalloc.h>
54
55#include <linux/drbd_limits.h>
56#include "drbd_int.h"
57#include "drbd_tracing.h"
58#include "drbd_req.h" /* only for _req_mod in tl_release and tl_clear */
59
60#include "drbd_vli.h"
61
62struct after_state_chg_work {
63 struct drbd_work w;
64 union drbd_state os;
65 union drbd_state ns;
66 enum chg_state_flags flags;
67 struct completion *done;
68};
69
70int drbdd_init(struct drbd_thread *);
71int drbd_worker(struct drbd_thread *);
72int drbd_asender(struct drbd_thread *);
73
74int drbd_init(void);
75static int drbd_open(struct block_device *bdev, fmode_t mode);
76static int drbd_release(struct gendisk *gd, fmode_t mode);
77static int w_after_state_ch(struct drbd_conf *mdev, struct drbd_work *w, int unused);
78static void after_state_ch(struct drbd_conf *mdev, union drbd_state os,
79 union drbd_state ns, enum chg_state_flags flags);
80static int w_md_sync(struct drbd_conf *mdev, struct drbd_work *w, int unused);
81static void md_sync_timer_fn(unsigned long data);
82static int w_bitmap_io(struct drbd_conf *mdev, struct drbd_work *w, int unused);
83
84DEFINE_TRACE(drbd_unplug);
85DEFINE_TRACE(drbd_uuid);
86DEFINE_TRACE(drbd_ee);
87DEFINE_TRACE(drbd_packet);
88DEFINE_TRACE(drbd_md_io);
89DEFINE_TRACE(drbd_epoch);
90DEFINE_TRACE(drbd_netlink);
91DEFINE_TRACE(drbd_actlog);
92DEFINE_TRACE(drbd_bio);
93DEFINE_TRACE(_drbd_resync);
94DEFINE_TRACE(drbd_req);
95
96MODULE_AUTHOR("Philipp Reisner <phil@linbit.com>, "
97 "Lars Ellenberg <lars@linbit.com>");
98MODULE_DESCRIPTION("drbd - Distributed Replicated Block Device v" REL_VERSION);
99MODULE_VERSION(REL_VERSION);
100MODULE_LICENSE("GPL");
101MODULE_PARM_DESC(minor_count, "Maximum number of drbd devices (1-255)");
102MODULE_ALIAS_BLOCKDEV_MAJOR(DRBD_MAJOR);
103
104#include <linux/moduleparam.h>
105/* allow_open_on_secondary */
106MODULE_PARM_DESC(allow_oos, "DONT USE!");
107/* thanks to these macros, if compiled into the kernel (not-module),
108 * this becomes the boot parameter drbd.minor_count */
109module_param(minor_count, uint, 0444);
110module_param(disable_sendpage, bool, 0644);
111module_param(allow_oos, bool, 0);
112module_param(cn_idx, uint, 0444);
113module_param(proc_details, int, 0644);
114
115#ifdef CONFIG_DRBD_FAULT_INJECTION
116int enable_faults;
117int fault_rate;
118static int fault_count;
119int fault_devs;
120/* bitmap of enabled faults */
121module_param(enable_faults, int, 0664);
122/* fault rate % value - applies to all enabled faults */
123module_param(fault_rate, int, 0664);
124/* count of faults inserted */
125module_param(fault_count, int, 0664);
126/* bitmap of devices to insert faults on */
127module_param(fault_devs, int, 0644);
128#endif
129
130/* module parameter, defined */
131unsigned int minor_count = 32;
132int disable_sendpage;
133int allow_oos;
134unsigned int cn_idx = CN_IDX_DRBD;
135int proc_details; /* Detail level in proc drbd*/
136
137/* Module parameter for setting the user mode helper program
138 * to run. Default is /sbin/drbdadm */
139char usermode_helper[80] = "/sbin/drbdadm";
140
141module_param_string(usermode_helper, usermode_helper, sizeof(usermode_helper), 0644);
142
143/* in 2.6.x, our device mapping and config info contains our virtual gendisks
144 * as member "struct gendisk *vdisk;"
145 */
146struct drbd_conf **minor_table;
147
148struct kmem_cache *drbd_request_cache;
149struct kmem_cache *drbd_ee_cache; /* epoch entries */
150struct kmem_cache *drbd_bm_ext_cache; /* bitmap extents */
151struct kmem_cache *drbd_al_ext_cache; /* activity log extents */
152mempool_t *drbd_request_mempool;
153mempool_t *drbd_ee_mempool;
154
155/* I do not use a standard mempool, because:
156 1) I want to hand out the pre-allocated objects first.
157 2) I want to be able to interrupt sleeping allocation with a signal.
158 Note: This is a single linked list, the next pointer is the private
159 member of struct page.
160 */
161struct page *drbd_pp_pool;
162spinlock_t drbd_pp_lock;
163int drbd_pp_vacant;
164wait_queue_head_t drbd_pp_wait;
165
166DEFINE_RATELIMIT_STATE(drbd_ratelimit_state, 5 * HZ, 5);
167
168static struct block_device_operations drbd_ops = {
169 .owner = THIS_MODULE,
170 .open = drbd_open,
171 .release = drbd_release,
172};
173
174#define ARRY_SIZE(A) (sizeof(A)/sizeof(A[0]))
175
176#ifdef __CHECKER__
177/* When checking with sparse, and this is an inline function, sparse will
178 give tons of false positives. When this is a real functions sparse works.
179 */
180int _get_ldev_if_state(struct drbd_conf *mdev, enum drbd_disk_state mins)
181{
182 int io_allowed;
183
184 atomic_inc(&mdev->local_cnt);
185 io_allowed = (mdev->state.disk >= mins);
186 if (!io_allowed) {
187 if (atomic_dec_and_test(&mdev->local_cnt))
188 wake_up(&mdev->misc_wait);
189 }
190 return io_allowed;
191}
192
193#endif
194
195/**
196 * DOC: The transfer log
197 *
198 * The transfer log is a single linked list of &struct drbd_tl_epoch objects.
199 * mdev->newest_tle points to the head, mdev->oldest_tle points to the tail
200 * of the list. There is always at least one &struct drbd_tl_epoch object.
201 *
202 * Each &struct drbd_tl_epoch has a circular double linked list of requests
203 * attached.
204 */
205static int tl_init(struct drbd_conf *mdev)
206{
207 struct drbd_tl_epoch *b;
208
209 /* during device minor initialization, we may well use GFP_KERNEL */
210 b = kmalloc(sizeof(struct drbd_tl_epoch), GFP_KERNEL);
211 if (!b)
212 return 0;
213 INIT_LIST_HEAD(&b->requests);
214 INIT_LIST_HEAD(&b->w.list);
215 b->next = NULL;
216 b->br_number = 4711;
217 b->n_req = 0;
218 b->w.cb = NULL; /* if this is != NULL, we need to dec_ap_pending in tl_clear */
219
220 mdev->oldest_tle = b;
221 mdev->newest_tle = b;
222 INIT_LIST_HEAD(&mdev->out_of_sequence_requests);
223
224 mdev->tl_hash = NULL;
225 mdev->tl_hash_s = 0;
226
227 return 1;
228}
229
230static void tl_cleanup(struct drbd_conf *mdev)
231{
232 D_ASSERT(mdev->oldest_tle == mdev->newest_tle);
233 D_ASSERT(list_empty(&mdev->out_of_sequence_requests));
234 kfree(mdev->oldest_tle);
235 mdev->oldest_tle = NULL;
236 kfree(mdev->unused_spare_tle);
237 mdev->unused_spare_tle = NULL;
238 kfree(mdev->tl_hash);
239 mdev->tl_hash = NULL;
240 mdev->tl_hash_s = 0;
241}
242
243/**
244 * _tl_add_barrier() - Adds a barrier to the transfer log
245 * @mdev: DRBD device.
246 * @new: Barrier to be added before the current head of the TL.
247 *
248 * The caller must hold the req_lock.
249 */
250void _tl_add_barrier(struct drbd_conf *mdev, struct drbd_tl_epoch *new)
251{
252 struct drbd_tl_epoch *newest_before;
253
254 INIT_LIST_HEAD(&new->requests);
255 INIT_LIST_HEAD(&new->w.list);
256 new->w.cb = NULL; /* if this is != NULL, we need to dec_ap_pending in tl_clear */
257 new->next = NULL;
258 new->n_req = 0;
259
260 newest_before = mdev->newest_tle;
261 /* never send a barrier number == 0, because that is special-cased
262 * when using TCQ for our write ordering code */
263 new->br_number = (newest_before->br_number+1) ?: 1;
264 if (mdev->newest_tle != new) {
265 mdev->newest_tle->next = new;
266 mdev->newest_tle = new;
267 }
268}
269
270/**
271 * tl_release() - Free or recycle the oldest &struct drbd_tl_epoch object of the TL
272 * @mdev: DRBD device.
273 * @barrier_nr: Expected identifier of the DRBD write barrier packet.
274 * @set_size: Expected number of requests before that barrier.
275 *
276 * In case the passed barrier_nr or set_size does not match the oldest
277 * &struct drbd_tl_epoch objects this function will cause a termination
278 * of the connection.
279 */
280void tl_release(struct drbd_conf *mdev, unsigned int barrier_nr,
281 unsigned int set_size)
282{
283 struct drbd_tl_epoch *b, *nob; /* next old barrier */
284 struct list_head *le, *tle;
285 struct drbd_request *r;
286
287 spin_lock_irq(&mdev->req_lock);
288
289 b = mdev->oldest_tle;
290
291 /* first some paranoia code */
292 if (b == NULL) {
293 dev_err(DEV, "BAD! BarrierAck #%u received, but no epoch in tl!?\n",
294 barrier_nr);
295 goto bail;
296 }
297 if (b->br_number != barrier_nr) {
298 dev_err(DEV, "BAD! BarrierAck #%u received, expected #%u!\n",
299 barrier_nr, b->br_number);
300 goto bail;
301 }
302 if (b->n_req != set_size) {
303 dev_err(DEV, "BAD! BarrierAck #%u received with n_req=%u, expected n_req=%u!\n",
304 barrier_nr, set_size, b->n_req);
305 goto bail;
306 }
307
308 /* Clean up list of requests processed during current epoch */
309 list_for_each_safe(le, tle, &b->requests) {
310 r = list_entry(le, struct drbd_request, tl_requests);
311 _req_mod(r, barrier_acked);
312 }
313 /* There could be requests on the list waiting for completion
314 of the write to the local disk. To avoid corruptions of
315 slab's data structures we have to remove the lists head.
316
317 Also there could have been a barrier ack out of sequence, overtaking
318 the write acks - which would be a bug and violating write ordering.
319 To not deadlock in case we lose connection while such requests are
320 still pending, we need some way to find them for the
321 _req_mode(connection_lost_while_pending).
322
323 These have been list_move'd to the out_of_sequence_requests list in
324 _req_mod(, barrier_acked) above.
325 */
326 list_del_init(&b->requests);
327
328 nob = b->next;
329 if (test_and_clear_bit(CREATE_BARRIER, &mdev->flags)) {
330 _tl_add_barrier(mdev, b);
331 if (nob)
332 mdev->oldest_tle = nob;
333 /* if nob == NULL b was the only barrier, and becomes the new
334 barrier. Therefore mdev->oldest_tle points already to b */
335 } else {
336 D_ASSERT(nob != NULL);
337 mdev->oldest_tle = nob;
338 kfree(b);
339 }
340
341 spin_unlock_irq(&mdev->req_lock);
342 dec_ap_pending(mdev);
343
344 return;
345
346bail:
347 spin_unlock_irq(&mdev->req_lock);
348 drbd_force_state(mdev, NS(conn, C_PROTOCOL_ERROR));
349}
350
351
352/**
353 * tl_clear() - Clears all requests and &struct drbd_tl_epoch objects out of the TL
354 * @mdev: DRBD device.
355 *
356 * This is called after the connection to the peer was lost. The storage covered
357 * by the requests on the transfer gets marked as our of sync. Called from the
358 * receiver thread and the worker thread.
359 */
360void tl_clear(struct drbd_conf *mdev)
361{
362 struct drbd_tl_epoch *b, *tmp;
363 struct list_head *le, *tle;
364 struct drbd_request *r;
365 int new_initial_bnr = net_random();
366
367 spin_lock_irq(&mdev->req_lock);
368
369 b = mdev->oldest_tle;
370 while (b) {
371 list_for_each_safe(le, tle, &b->requests) {
372 r = list_entry(le, struct drbd_request, tl_requests);
373 /* It would be nice to complete outside of spinlock.
374 * But this is easier for now. */
375 _req_mod(r, connection_lost_while_pending);
376 }
377 tmp = b->next;
378
379 /* there could still be requests on that ring list,
380 * in case local io is still pending */
381 list_del(&b->requests);
382
383 /* dec_ap_pending corresponding to queue_barrier.
384 * the newest barrier may not have been queued yet,
385 * in which case w.cb is still NULL. */
386 if (b->w.cb != NULL)
387 dec_ap_pending(mdev);
388
389 if (b == mdev->newest_tle) {
390 /* recycle, but reinit! */
391 D_ASSERT(tmp == NULL);
392 INIT_LIST_HEAD(&b->requests);
393 INIT_LIST_HEAD(&b->w.list);
394 b->w.cb = NULL;
395 b->br_number = new_initial_bnr;
396 b->n_req = 0;
397
398 mdev->oldest_tle = b;
399 break;
400 }
401 kfree(b);
402 b = tmp;
403 }
404
405 /* we expect this list to be empty. */
406 D_ASSERT(list_empty(&mdev->out_of_sequence_requests));
407
408 /* but just in case, clean it up anyways! */
409 list_for_each_safe(le, tle, &mdev->out_of_sequence_requests) {
410 r = list_entry(le, struct drbd_request, tl_requests);
411 /* It would be nice to complete outside of spinlock.
412 * But this is easier for now. */
413 _req_mod(r, connection_lost_while_pending);
414 }
415
416 /* ensure bit indicating barrier is required is clear */
417 clear_bit(CREATE_BARRIER, &mdev->flags);
418
419 spin_unlock_irq(&mdev->req_lock);
420}
421
422/**
423 * cl_wide_st_chg() - TRUE if the state change is a cluster wide one
424 * @mdev: DRBD device.
425 * @os: old (current) state.
426 * @ns: new (wanted) state.
427 */
428static int cl_wide_st_chg(struct drbd_conf *mdev,
429 union drbd_state os, union drbd_state ns)
430{
431 return (os.conn >= C_CONNECTED && ns.conn >= C_CONNECTED &&
432 ((os.role != R_PRIMARY && ns.role == R_PRIMARY) ||
433 (os.conn != C_STARTING_SYNC_T && ns.conn == C_STARTING_SYNC_T) ||
434 (os.conn != C_STARTING_SYNC_S && ns.conn == C_STARTING_SYNC_S) ||
435 (os.disk != D_DISKLESS && ns.disk == D_DISKLESS))) ||
436 (os.conn >= C_CONNECTED && ns.conn == C_DISCONNECTING) ||
437 (os.conn == C_CONNECTED && ns.conn == C_VERIFY_S);
438}
439
440int drbd_change_state(struct drbd_conf *mdev, enum chg_state_flags f,
441 union drbd_state mask, union drbd_state val)
442{
443 unsigned long flags;
444 union drbd_state os, ns;
445 int rv;
446
447 spin_lock_irqsave(&mdev->req_lock, flags);
448 os = mdev->state;
449 ns.i = (os.i & ~mask.i) | val.i;
450 rv = _drbd_set_state(mdev, ns, f, NULL);
451 ns = mdev->state;
452 spin_unlock_irqrestore(&mdev->req_lock, flags);
453
454 return rv;
455}
456
457/**
458 * drbd_force_state() - Impose a change which happens outside our control on our state
459 * @mdev: DRBD device.
460 * @mask: mask of state bits to change.
461 * @val: value of new state bits.
462 */
463void drbd_force_state(struct drbd_conf *mdev,
464 union drbd_state mask, union drbd_state val)
465{
466 drbd_change_state(mdev, CS_HARD, mask, val);
467}
468
469static int is_valid_state(struct drbd_conf *mdev, union drbd_state ns);
470static int is_valid_state_transition(struct drbd_conf *,
471 union drbd_state, union drbd_state);
472static union drbd_state sanitize_state(struct drbd_conf *mdev, union drbd_state os,
473 union drbd_state ns, int *warn_sync_abort);
474int drbd_send_state_req(struct drbd_conf *,
475 union drbd_state, union drbd_state);
476
477static enum drbd_state_ret_codes _req_st_cond(struct drbd_conf *mdev,
478 union drbd_state mask, union drbd_state val)
479{
480 union drbd_state os, ns;
481 unsigned long flags;
482 int rv;
483
484 if (test_and_clear_bit(CL_ST_CHG_SUCCESS, &mdev->flags))
485 return SS_CW_SUCCESS;
486
487 if (test_and_clear_bit(CL_ST_CHG_FAIL, &mdev->flags))
488 return SS_CW_FAILED_BY_PEER;
489
490 rv = 0;
491 spin_lock_irqsave(&mdev->req_lock, flags);
492 os = mdev->state;
493 ns.i = (os.i & ~mask.i) | val.i;
494 ns = sanitize_state(mdev, os, ns, NULL);
495
496 if (!cl_wide_st_chg(mdev, os, ns))
497 rv = SS_CW_NO_NEED;
498 if (!rv) {
499 rv = is_valid_state(mdev, ns);
500 if (rv == SS_SUCCESS) {
501 rv = is_valid_state_transition(mdev, ns, os);
502 if (rv == SS_SUCCESS)
503 rv = 0; /* cont waiting, otherwise fail. */
504 }
505 }
506 spin_unlock_irqrestore(&mdev->req_lock, flags);
507
508 return rv;
509}
510
511/**
512 * drbd_req_state() - Perform an eventually cluster wide state change
513 * @mdev: DRBD device.
514 * @mask: mask of state bits to change.
515 * @val: value of new state bits.
516 * @f: flags
517 *
518 * Should not be called directly, use drbd_request_state() or
519 * _drbd_request_state().
520 */
521static int drbd_req_state(struct drbd_conf *mdev,
522 union drbd_state mask, union drbd_state val,
523 enum chg_state_flags f)
524{
525 struct completion done;
526 unsigned long flags;
527 union drbd_state os, ns;
528 int rv;
529
530 init_completion(&done);
531
532 if (f & CS_SERIALIZE)
533 mutex_lock(&mdev->state_mutex);
534
535 spin_lock_irqsave(&mdev->req_lock, flags);
536 os = mdev->state;
537 ns.i = (os.i & ~mask.i) | val.i;
538 ns = sanitize_state(mdev, os, ns, NULL);
539
540 if (cl_wide_st_chg(mdev, os, ns)) {
541 rv = is_valid_state(mdev, ns);
542 if (rv == SS_SUCCESS)
543 rv = is_valid_state_transition(mdev, ns, os);
544 spin_unlock_irqrestore(&mdev->req_lock, flags);
545
546 if (rv < SS_SUCCESS) {
547 if (f & CS_VERBOSE)
548 print_st_err(mdev, os, ns, rv);
549 goto abort;
550 }
551
552 drbd_state_lock(mdev);
553 if (!drbd_send_state_req(mdev, mask, val)) {
554 drbd_state_unlock(mdev);
555 rv = SS_CW_FAILED_BY_PEER;
556 if (f & CS_VERBOSE)
557 print_st_err(mdev, os, ns, rv);
558 goto abort;
559 }
560
561 wait_event(mdev->state_wait,
562 (rv = _req_st_cond(mdev, mask, val)));
563
564 if (rv < SS_SUCCESS) {
565 drbd_state_unlock(mdev);
566 if (f & CS_VERBOSE)
567 print_st_err(mdev, os, ns, rv);
568 goto abort;
569 }
570 spin_lock_irqsave(&mdev->req_lock, flags);
571 os = mdev->state;
572 ns.i = (os.i & ~mask.i) | val.i;
573 rv = _drbd_set_state(mdev, ns, f, &done);
574 drbd_state_unlock(mdev);
575 } else {
576 rv = _drbd_set_state(mdev, ns, f, &done);
577 }
578
579 spin_unlock_irqrestore(&mdev->req_lock, flags);
580
581 if (f & CS_WAIT_COMPLETE && rv == SS_SUCCESS) {
582 D_ASSERT(current != mdev->worker.task);
583 wait_for_completion(&done);
584 }
585
586abort:
587 if (f & CS_SERIALIZE)
588 mutex_unlock(&mdev->state_mutex);
589
590 return rv;
591}
592
593/**
594 * _drbd_request_state() - Request a state change (with flags)
595 * @mdev: DRBD device.
596 * @mask: mask of state bits to change.
597 * @val: value of new state bits.
598 * @f: flags
599 *
600 * Cousin of drbd_request_state(), useful with the CS_WAIT_COMPLETE
601 * flag, or when logging of failed state change requests is not desired.
602 */
603int _drbd_request_state(struct drbd_conf *mdev, union drbd_state mask,
604 union drbd_state val, enum chg_state_flags f)
605{
606 int rv;
607
608 wait_event(mdev->state_wait,
609 (rv = drbd_req_state(mdev, mask, val, f)) != SS_IN_TRANSIENT_STATE);
610
611 return rv;
612}
613
614static void print_st(struct drbd_conf *mdev, char *name, union drbd_state ns)
615{
616 dev_err(DEV, " %s = { cs:%s ro:%s/%s ds:%s/%s %c%c%c%c }\n",
617 name,
618 drbd_conn_str(ns.conn),
619 drbd_role_str(ns.role),
620 drbd_role_str(ns.peer),
621 drbd_disk_str(ns.disk),
622 drbd_disk_str(ns.pdsk),
623 ns.susp ? 's' : 'r',
624 ns.aftr_isp ? 'a' : '-',
625 ns.peer_isp ? 'p' : '-',
626 ns.user_isp ? 'u' : '-'
627 );
628}
629
630void print_st_err(struct drbd_conf *mdev,
631 union drbd_state os, union drbd_state ns, int err)
632{
633 if (err == SS_IN_TRANSIENT_STATE)
634 return;
635 dev_err(DEV, "State change failed: %s\n", drbd_set_st_err_str(err));
636 print_st(mdev, " state", os);
637 print_st(mdev, "wanted", ns);
638}
639
640
641#define drbd_peer_str drbd_role_str
642#define drbd_pdsk_str drbd_disk_str
643
644#define drbd_susp_str(A) ((A) ? "1" : "0")
645#define drbd_aftr_isp_str(A) ((A) ? "1" : "0")
646#define drbd_peer_isp_str(A) ((A) ? "1" : "0")
647#define drbd_user_isp_str(A) ((A) ? "1" : "0")
648
649#define PSC(A) \
650 ({ if (ns.A != os.A) { \
651 pbp += sprintf(pbp, #A "( %s -> %s ) ", \
652 drbd_##A##_str(os.A), \
653 drbd_##A##_str(ns.A)); \
654 } })
655
656/**
657 * is_valid_state() - Returns an SS_ error code if ns is not valid
658 * @mdev: DRBD device.
659 * @ns: State to consider.
660 */
661static int is_valid_state(struct drbd_conf *mdev, union drbd_state ns)
662{
663 /* See drbd_state_sw_errors in drbd_strings.c */
664
665 enum drbd_fencing_p fp;
666 int rv = SS_SUCCESS;
667
668 fp = FP_DONT_CARE;
669 if (get_ldev(mdev)) {
670 fp = mdev->ldev->dc.fencing;
671 put_ldev(mdev);
672 }
673
674 if (get_net_conf(mdev)) {
675 if (!mdev->net_conf->two_primaries &&
676 ns.role == R_PRIMARY && ns.peer == R_PRIMARY)
677 rv = SS_TWO_PRIMARIES;
678 put_net_conf(mdev);
679 }
680
681 if (rv <= 0)
682 /* already found a reason to abort */;
683 else if (ns.role == R_SECONDARY && mdev->open_cnt)
684 rv = SS_DEVICE_IN_USE;
685
686 else if (ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.disk < D_UP_TO_DATE)
687 rv = SS_NO_UP_TO_DATE_DISK;
688
689 else if (fp >= FP_RESOURCE &&
690 ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.pdsk >= D_UNKNOWN)
691 rv = SS_PRIMARY_NOP;
692
693 else if (ns.role == R_PRIMARY && ns.disk <= D_INCONSISTENT && ns.pdsk <= D_INCONSISTENT)
694 rv = SS_NO_UP_TO_DATE_DISK;
695
696 else if (ns.conn > C_CONNECTED && ns.disk < D_INCONSISTENT)
697 rv = SS_NO_LOCAL_DISK;
698
699 else if (ns.conn > C_CONNECTED && ns.pdsk < D_INCONSISTENT)
700 rv = SS_NO_REMOTE_DISK;
701
702 else if ((ns.conn == C_CONNECTED ||
703 ns.conn == C_WF_BITMAP_S ||
704 ns.conn == C_SYNC_SOURCE ||
705 ns.conn == C_PAUSED_SYNC_S) &&
706 ns.disk == D_OUTDATED)
707 rv = SS_CONNECTED_OUTDATES;
708
709 else if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
710 (mdev->sync_conf.verify_alg[0] == 0))
711 rv = SS_NO_VERIFY_ALG;
712
713 else if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
714 mdev->agreed_pro_version < 88)
715 rv = SS_NOT_SUPPORTED;
716
717 return rv;
718}
719
720/**
721 * is_valid_state_transition() - Returns an SS_ error code if the state transition is not possible
722 * @mdev: DRBD device.
723 * @ns: new state.
724 * @os: old state.
725 */
726static int is_valid_state_transition(struct drbd_conf *mdev,
727 union drbd_state ns, union drbd_state os)
728{
729 int rv = SS_SUCCESS;
730
731 if ((ns.conn == C_STARTING_SYNC_T || ns.conn == C_STARTING_SYNC_S) &&
732 os.conn > C_CONNECTED)
733 rv = SS_RESYNC_RUNNING;
734
735 if (ns.conn == C_DISCONNECTING && os.conn == C_STANDALONE)
736 rv = SS_ALREADY_STANDALONE;
737
738 if (ns.disk > D_ATTACHING && os.disk == D_DISKLESS)
739 rv = SS_IS_DISKLESS;
740
741 if (ns.conn == C_WF_CONNECTION && os.conn < C_UNCONNECTED)
742 rv = SS_NO_NET_CONFIG;
743
744 if (ns.disk == D_OUTDATED && os.disk < D_OUTDATED && os.disk != D_ATTACHING)
745 rv = SS_LOWER_THAN_OUTDATED;
746
747 if (ns.conn == C_DISCONNECTING && os.conn == C_UNCONNECTED)
748 rv = SS_IN_TRANSIENT_STATE;
749
750 if (ns.conn == os.conn && ns.conn == C_WF_REPORT_PARAMS)
751 rv = SS_IN_TRANSIENT_STATE;
752
753 if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) && os.conn < C_CONNECTED)
754 rv = SS_NEED_CONNECTION;
755
756 if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
757 ns.conn != os.conn && os.conn > C_CONNECTED)
758 rv = SS_RESYNC_RUNNING;
759
760 if ((ns.conn == C_STARTING_SYNC_S || ns.conn == C_STARTING_SYNC_T) &&
761 os.conn < C_CONNECTED)
762 rv = SS_NEED_CONNECTION;
763
764 return rv;
765}
766
767/**
768 * sanitize_state() - Resolves implicitly necessary additional changes to a state transition
769 * @mdev: DRBD device.
770 * @os: old state.
771 * @ns: new state.
772 * @warn_sync_abort:
773 *
774 * When we loose connection, we have to set the state of the peers disk (pdsk)
775 * to D_UNKNOWN. This rule and many more along those lines are in this function.
776 */
777static union drbd_state sanitize_state(struct drbd_conf *mdev, union drbd_state os,
778 union drbd_state ns, int *warn_sync_abort)
779{
780 enum drbd_fencing_p fp;
781
782 fp = FP_DONT_CARE;
783 if (get_ldev(mdev)) {
784 fp = mdev->ldev->dc.fencing;
785 put_ldev(mdev);
786 }
787
788 /* Disallow Network errors to configure a device's network part */
789 if ((ns.conn >= C_TIMEOUT && ns.conn <= C_TEAR_DOWN) &&
790 os.conn <= C_DISCONNECTING)
791 ns.conn = os.conn;
792
793 /* After a network error (+C_TEAR_DOWN) only C_UNCONNECTED or C_DISCONNECTING can follow */
794 if (os.conn >= C_TIMEOUT && os.conn <= C_TEAR_DOWN &&
795 ns.conn != C_UNCONNECTED && ns.conn != C_DISCONNECTING)
796 ns.conn = os.conn;
797
798 /* After C_DISCONNECTING only C_STANDALONE may follow */
799 if (os.conn == C_DISCONNECTING && ns.conn != C_STANDALONE)
800 ns.conn = os.conn;
801
802 if (ns.conn < C_CONNECTED) {
803 ns.peer_isp = 0;
804 ns.peer = R_UNKNOWN;
805 if (ns.pdsk > D_UNKNOWN || ns.pdsk < D_INCONSISTENT)
806 ns.pdsk = D_UNKNOWN;
807 }
808
809 /* Clear the aftr_isp when becoming unconfigured */
810 if (ns.conn == C_STANDALONE && ns.disk == D_DISKLESS && ns.role == R_SECONDARY)
811 ns.aftr_isp = 0;
812
813 if (ns.conn <= C_DISCONNECTING && ns.disk == D_DISKLESS)
814 ns.pdsk = D_UNKNOWN;
815
816 /* Abort resync if a disk fails/detaches */
817 if (os.conn > C_CONNECTED && ns.conn > C_CONNECTED &&
818 (ns.disk <= D_FAILED || ns.pdsk <= D_FAILED)) {
819 if (warn_sync_abort)
820 *warn_sync_abort = 1;
821 ns.conn = C_CONNECTED;
822 }
823
824 if (ns.conn >= C_CONNECTED &&
825 ((ns.disk == D_CONSISTENT || ns.disk == D_OUTDATED) ||
826 (ns.disk == D_NEGOTIATING && ns.conn == C_WF_BITMAP_T))) {
827 switch (ns.conn) {
828 case C_WF_BITMAP_T:
829 case C_PAUSED_SYNC_T:
830 ns.disk = D_OUTDATED;
831 break;
832 case C_CONNECTED:
833 case C_WF_BITMAP_S:
834 case C_SYNC_SOURCE:
835 case C_PAUSED_SYNC_S:
836 ns.disk = D_UP_TO_DATE;
837 break;
838 case C_SYNC_TARGET:
839 ns.disk = D_INCONSISTENT;
840 dev_warn(DEV, "Implicitly set disk state Inconsistent!\n");
841 break;
842 }
843 if (os.disk == D_OUTDATED && ns.disk == D_UP_TO_DATE)
844 dev_warn(DEV, "Implicitly set disk from Outdated to UpToDate\n");
845 }
846
847 if (ns.conn >= C_CONNECTED &&
848 (ns.pdsk == D_CONSISTENT || ns.pdsk == D_OUTDATED)) {
849 switch (ns.conn) {
850 case C_CONNECTED:
851 case C_WF_BITMAP_T:
852 case C_PAUSED_SYNC_T:
853 case C_SYNC_TARGET:
854 ns.pdsk = D_UP_TO_DATE;
855 break;
856 case C_WF_BITMAP_S:
857 case C_PAUSED_SYNC_S:
858 ns.pdsk = D_OUTDATED;
859 break;
860 case C_SYNC_SOURCE:
861 ns.pdsk = D_INCONSISTENT;
862 dev_warn(DEV, "Implicitly set pdsk Inconsistent!\n");
863 break;
864 }
865 if (os.pdsk == D_OUTDATED && ns.pdsk == D_UP_TO_DATE)
866 dev_warn(DEV, "Implicitly set pdsk from Outdated to UpToDate\n");
867 }
868
869 /* Connection breaks down before we finished "Negotiating" */
870 if (ns.conn < C_CONNECTED && ns.disk == D_NEGOTIATING &&
871 get_ldev_if_state(mdev, D_NEGOTIATING)) {
872 if (mdev->ed_uuid == mdev->ldev->md.uuid[UI_CURRENT]) {
873 ns.disk = mdev->new_state_tmp.disk;
874 ns.pdsk = mdev->new_state_tmp.pdsk;
875 } else {
876 dev_alert(DEV, "Connection lost while negotiating, no data!\n");
877 ns.disk = D_DISKLESS;
878 ns.pdsk = D_UNKNOWN;
879 }
880 put_ldev(mdev);
881 }
882
883 if (fp == FP_STONITH &&
884 (ns.role == R_PRIMARY &&
885 ns.conn < C_CONNECTED &&
886 ns.pdsk > D_OUTDATED))
887 ns.susp = 1;
888
889 if (ns.aftr_isp || ns.peer_isp || ns.user_isp) {
890 if (ns.conn == C_SYNC_SOURCE)
891 ns.conn = C_PAUSED_SYNC_S;
892 if (ns.conn == C_SYNC_TARGET)
893 ns.conn = C_PAUSED_SYNC_T;
894 } else {
895 if (ns.conn == C_PAUSED_SYNC_S)
896 ns.conn = C_SYNC_SOURCE;
897 if (ns.conn == C_PAUSED_SYNC_T)
898 ns.conn = C_SYNC_TARGET;
899 }
900
901 return ns;
902}
903
904/* helper for __drbd_set_state */
905static void set_ov_position(struct drbd_conf *mdev, enum drbd_conns cs)
906{
907 if (cs == C_VERIFY_T) {
908 /* starting online verify from an arbitrary position
909 * does not fit well into the existing protocol.
910 * on C_VERIFY_T, we initialize ov_left and friends
911 * implicitly in receive_DataRequest once the
912 * first P_OV_REQUEST is received */
913 mdev->ov_start_sector = ~(sector_t)0;
914 } else {
915 unsigned long bit = BM_SECT_TO_BIT(mdev->ov_start_sector);
916 if (bit >= mdev->rs_total)
917 mdev->ov_start_sector =
918 BM_BIT_TO_SECT(mdev->rs_total - 1);
919 mdev->ov_position = mdev->ov_start_sector;
920 }
921}
922
923/**
924 * __drbd_set_state() - Set a new DRBD state
925 * @mdev: DRBD device.
926 * @ns: new state.
927 * @flags: Flags
928 * @done: Optional completion, that will get completed after the after_state_ch() finished
929 *
930 * Caller needs to hold req_lock, and global_state_lock. Do not call directly.
931 */
932int __drbd_set_state(struct drbd_conf *mdev,
933 union drbd_state ns, enum chg_state_flags flags,
934 struct completion *done)
935{
936 union drbd_state os;
937 int rv = SS_SUCCESS;
938 int warn_sync_abort = 0;
939 struct after_state_chg_work *ascw;
940
941 os = mdev->state;
942
943 ns = sanitize_state(mdev, os, ns, &warn_sync_abort);
944
945 if (ns.i == os.i)
946 return SS_NOTHING_TO_DO;
947
948 if (!(flags & CS_HARD)) {
949 /* pre-state-change checks ; only look at ns */
950 /* See drbd_state_sw_errors in drbd_strings.c */
951
952 rv = is_valid_state(mdev, ns);
953 if (rv < SS_SUCCESS) {
954 /* If the old state was illegal as well, then let
955 this happen...*/
956
957 if (is_valid_state(mdev, os) == rv) {
958 dev_err(DEV, "Considering state change from bad state. "
959 "Error would be: '%s'\n",
960 drbd_set_st_err_str(rv));
961 print_st(mdev, "old", os);
962 print_st(mdev, "new", ns);
963 rv = is_valid_state_transition(mdev, ns, os);
964 }
965 } else
966 rv = is_valid_state_transition(mdev, ns, os);
967 }
968
969 if (rv < SS_SUCCESS) {
970 if (flags & CS_VERBOSE)
971 print_st_err(mdev, os, ns, rv);
972 return rv;
973 }
974
975 if (warn_sync_abort)
976 dev_warn(DEV, "Resync aborted.\n");
977
978 {
979 char *pbp, pb[300];
980 pbp = pb;
981 *pbp = 0;
982 PSC(role);
983 PSC(peer);
984 PSC(conn);
985 PSC(disk);
986 PSC(pdsk);
987 PSC(susp);
988 PSC(aftr_isp);
989 PSC(peer_isp);
990 PSC(user_isp);
991 dev_info(DEV, "%s\n", pb);
992 }
993
994 /* solve the race between becoming unconfigured,
995 * worker doing the cleanup, and
996 * admin reconfiguring us:
997 * on (re)configure, first set CONFIG_PENDING,
998 * then wait for a potentially exiting worker,
999 * start the worker, and schedule one no_op.
1000 * then proceed with configuration.
1001 */
1002 if (ns.disk == D_DISKLESS &&
1003 ns.conn == C_STANDALONE &&
1004 ns.role == R_SECONDARY &&
1005 !test_and_set_bit(CONFIG_PENDING, &mdev->flags))
1006 set_bit(DEVICE_DYING, &mdev->flags);
1007
1008 mdev->state.i = ns.i;
1009 wake_up(&mdev->misc_wait);
1010 wake_up(&mdev->state_wait);
1011
1012 /* post-state-change actions */
1013 if (os.conn >= C_SYNC_SOURCE && ns.conn <= C_CONNECTED) {
1014 set_bit(STOP_SYNC_TIMER, &mdev->flags);
1015 mod_timer(&mdev->resync_timer, jiffies);
1016 }
1017
1018 /* aborted verify run. log the last position */
1019 if ((os.conn == C_VERIFY_S || os.conn == C_VERIFY_T) &&
1020 ns.conn < C_CONNECTED) {
1021 mdev->ov_start_sector =
1022 BM_BIT_TO_SECT(mdev->rs_total - mdev->ov_left);
1023 dev_info(DEV, "Online Verify reached sector %llu\n",
1024 (unsigned long long)mdev->ov_start_sector);
1025 }
1026
1027 if ((os.conn == C_PAUSED_SYNC_T || os.conn == C_PAUSED_SYNC_S) &&
1028 (ns.conn == C_SYNC_TARGET || ns.conn == C_SYNC_SOURCE)) {
1029 dev_info(DEV, "Syncer continues.\n");
1030 mdev->rs_paused += (long)jiffies-(long)mdev->rs_mark_time;
1031 if (ns.conn == C_SYNC_TARGET) {
1032 if (!test_and_clear_bit(STOP_SYNC_TIMER, &mdev->flags))
1033 mod_timer(&mdev->resync_timer, jiffies);
1034 /* This if (!test_bit) is only needed for the case
1035 that a device that has ceased to used its timer,
1036 i.e. it is already in drbd_resync_finished() gets
1037 paused and resumed. */
1038 }
1039 }
1040
1041 if ((os.conn == C_SYNC_TARGET || os.conn == C_SYNC_SOURCE) &&
1042 (ns.conn == C_PAUSED_SYNC_T || ns.conn == C_PAUSED_SYNC_S)) {
1043 dev_info(DEV, "Resync suspended\n");
1044 mdev->rs_mark_time = jiffies;
1045 if (ns.conn == C_PAUSED_SYNC_T)
1046 set_bit(STOP_SYNC_TIMER, &mdev->flags);
1047 }
1048
1049 if (os.conn == C_CONNECTED &&
1050 (ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T)) {
1051 mdev->ov_position = 0;
1052 mdev->rs_total =
1053 mdev->rs_mark_left = drbd_bm_bits(mdev);
1054 if (mdev->agreed_pro_version >= 90)
1055 set_ov_position(mdev, ns.conn);
1056 else
1057 mdev->ov_start_sector = 0;
1058 mdev->ov_left = mdev->rs_total
1059 - BM_SECT_TO_BIT(mdev->ov_position);
1060 mdev->rs_start =
1061 mdev->rs_mark_time = jiffies;
1062 mdev->ov_last_oos_size = 0;
1063 mdev->ov_last_oos_start = 0;
1064
1065 if (ns.conn == C_VERIFY_S) {
1066 dev_info(DEV, "Starting Online Verify from sector %llu\n",
1067 (unsigned long long)mdev->ov_position);
1068 mod_timer(&mdev->resync_timer, jiffies);
1069 }
1070 }
1071
1072 if (get_ldev(mdev)) {
1073 u32 mdf = mdev->ldev->md.flags & ~(MDF_CONSISTENT|MDF_PRIMARY_IND|
1074 MDF_CONNECTED_IND|MDF_WAS_UP_TO_DATE|
1075 MDF_PEER_OUT_DATED|MDF_CRASHED_PRIMARY);
1076
1077 if (test_bit(CRASHED_PRIMARY, &mdev->flags))
1078 mdf |= MDF_CRASHED_PRIMARY;
1079 if (mdev->state.role == R_PRIMARY ||
1080 (mdev->state.pdsk < D_INCONSISTENT && mdev->state.peer == R_PRIMARY))
1081 mdf |= MDF_PRIMARY_IND;
1082 if (mdev->state.conn > C_WF_REPORT_PARAMS)
1083 mdf |= MDF_CONNECTED_IND;
1084 if (mdev->state.disk > D_INCONSISTENT)
1085 mdf |= MDF_CONSISTENT;
1086 if (mdev->state.disk > D_OUTDATED)
1087 mdf |= MDF_WAS_UP_TO_DATE;
1088 if (mdev->state.pdsk <= D_OUTDATED && mdev->state.pdsk >= D_INCONSISTENT)
1089 mdf |= MDF_PEER_OUT_DATED;
1090 if (mdf != mdev->ldev->md.flags) {
1091 mdev->ldev->md.flags = mdf;
1092 drbd_md_mark_dirty(mdev);
1093 }
1094 if (os.disk < D_CONSISTENT && ns.disk >= D_CONSISTENT)
1095 drbd_set_ed_uuid(mdev, mdev->ldev->md.uuid[UI_CURRENT]);
1096 put_ldev(mdev);
1097 }
1098
1099 /* Peer was forced D_UP_TO_DATE & R_PRIMARY, consider to resync */
1100 if (os.disk == D_INCONSISTENT && os.pdsk == D_INCONSISTENT &&
1101 os.peer == R_SECONDARY && ns.peer == R_PRIMARY)
1102 set_bit(CONSIDER_RESYNC, &mdev->flags);
1103
1104 /* Receiver should clean up itself */
1105 if (os.conn != C_DISCONNECTING && ns.conn == C_DISCONNECTING)
1106 drbd_thread_stop_nowait(&mdev->receiver);
1107
1108 /* Now the receiver finished cleaning up itself, it should die */
1109 if (os.conn != C_STANDALONE && ns.conn == C_STANDALONE)
1110 drbd_thread_stop_nowait(&mdev->receiver);
1111
1112 /* Upon network failure, we need to restart the receiver. */
1113 if (os.conn > C_TEAR_DOWN &&
1114 ns.conn <= C_TEAR_DOWN && ns.conn >= C_TIMEOUT)
1115 drbd_thread_restart_nowait(&mdev->receiver);
1116
1117 ascw = kmalloc(sizeof(*ascw), GFP_ATOMIC);
1118 if (ascw) {
1119 ascw->os = os;
1120 ascw->ns = ns;
1121 ascw->flags = flags;
1122 ascw->w.cb = w_after_state_ch;
1123 ascw->done = done;
1124 drbd_queue_work(&mdev->data.work, &ascw->w);
1125 } else {
1126 dev_warn(DEV, "Could not kmalloc an ascw\n");
1127 }
1128
1129 return rv;
1130}
1131
1132static int w_after_state_ch(struct drbd_conf *mdev, struct drbd_work *w, int unused)
1133{
1134 struct after_state_chg_work *ascw =
1135 container_of(w, struct after_state_chg_work, w);
1136 after_state_ch(mdev, ascw->os, ascw->ns, ascw->flags);
1137 if (ascw->flags & CS_WAIT_COMPLETE) {
1138 D_ASSERT(ascw->done != NULL);
1139 complete(ascw->done);
1140 }
1141 kfree(ascw);
1142
1143 return 1;
1144}
1145
1146static void abw_start_sync(struct drbd_conf *mdev, int rv)
1147{
1148 if (rv) {
1149 dev_err(DEV, "Writing the bitmap failed not starting resync.\n");
1150 _drbd_request_state(mdev, NS(conn, C_CONNECTED), CS_VERBOSE);
1151 return;
1152 }
1153
1154 switch (mdev->state.conn) {
1155 case C_STARTING_SYNC_T:
1156 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE);
1157 break;
1158 case C_STARTING_SYNC_S:
1159 drbd_start_resync(mdev, C_SYNC_SOURCE);
1160 break;
1161 }
1162}
1163
1164/**
1165 * after_state_ch() - Perform after state change actions that may sleep
1166 * @mdev: DRBD device.
1167 * @os: old state.
1168 * @ns: new state.
1169 * @flags: Flags
1170 */
1171static void after_state_ch(struct drbd_conf *mdev, union drbd_state os,
1172 union drbd_state ns, enum chg_state_flags flags)
1173{
1174 enum drbd_fencing_p fp;
1175
1176 if (os.conn != C_CONNECTED && ns.conn == C_CONNECTED) {
1177 clear_bit(CRASHED_PRIMARY, &mdev->flags);
1178 if (mdev->p_uuid)
1179 mdev->p_uuid[UI_FLAGS] &= ~((u64)2);
1180 }
1181
1182 fp = FP_DONT_CARE;
1183 if (get_ldev(mdev)) {
1184 fp = mdev->ldev->dc.fencing;
1185 put_ldev(mdev);
1186 }
1187
1188 /* Inform userspace about the change... */
1189 drbd_bcast_state(mdev, ns);
1190
1191 if (!(os.role == R_PRIMARY && os.disk < D_UP_TO_DATE && os.pdsk < D_UP_TO_DATE) &&
1192 (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE))
1193 drbd_khelper(mdev, "pri-on-incon-degr");
1194
1195 /* Here we have the actions that are performed after a
1196 state change. This function might sleep */
1197
1198 if (fp == FP_STONITH && ns.susp) {
1199 /* case1: The outdate peer handler is successful:
1200 * case2: The connection was established again: */
1201 if ((os.pdsk > D_OUTDATED && ns.pdsk <= D_OUTDATED) ||
1202 (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED)) {
1203 tl_clear(mdev);
1204 spin_lock_irq(&mdev->req_lock);
1205 _drbd_set_state(_NS(mdev, susp, 0), CS_VERBOSE, NULL);
1206 spin_unlock_irq(&mdev->req_lock);
1207 }
1208 }
1209 /* Do not change the order of the if above and the two below... */
1210 if (os.pdsk == D_DISKLESS && ns.pdsk > D_DISKLESS) { /* attach on the peer */
1211 drbd_send_uuids(mdev);
1212 drbd_send_state(mdev);
1213 }
1214 if (os.conn != C_WF_BITMAP_S && ns.conn == C_WF_BITMAP_S)
1215 drbd_queue_bitmap_io(mdev, &drbd_send_bitmap, NULL, "send_bitmap (WFBitMapS)");
1216
1217 /* Lost contact to peer's copy of the data */
1218 if ((os.pdsk >= D_INCONSISTENT &&
1219 os.pdsk != D_UNKNOWN &&
1220 os.pdsk != D_OUTDATED)
1221 && (ns.pdsk < D_INCONSISTENT ||
1222 ns.pdsk == D_UNKNOWN ||
1223 ns.pdsk == D_OUTDATED)) {
1224 kfree(mdev->p_uuid);
1225 mdev->p_uuid = NULL;
1226 if (get_ldev(mdev)) {
1227 if ((ns.role == R_PRIMARY || ns.peer == R_PRIMARY) &&
1228 mdev->ldev->md.uuid[UI_BITMAP] == 0 && ns.disk >= D_UP_TO_DATE) {
1229 drbd_uuid_new_current(mdev);
1230 drbd_send_uuids(mdev);
1231 }
1232 put_ldev(mdev);
1233 }
1234 }
1235
1236 if (ns.pdsk < D_INCONSISTENT && get_ldev(mdev)) {
1237 if (ns.peer == R_PRIMARY && mdev->ldev->md.uuid[UI_BITMAP] == 0)
1238 drbd_uuid_new_current(mdev);
1239
1240 /* D_DISKLESS Peer becomes secondary */
1241 if (os.peer == R_PRIMARY && ns.peer == R_SECONDARY)
1242 drbd_al_to_on_disk_bm(mdev);
1243 put_ldev(mdev);
1244 }
1245
1246 /* Last part of the attaching process ... */
1247 if (ns.conn >= C_CONNECTED &&
1248 os.disk == D_ATTACHING && ns.disk == D_NEGOTIATING) {
1249 kfree(mdev->p_uuid); /* We expect to receive up-to-date UUIDs soon. */
1250 mdev->p_uuid = NULL; /* ...to not use the old ones in the mean time */
1251 drbd_send_sizes(mdev, 0); /* to start sync... */
1252 drbd_send_uuids(mdev);
1253 drbd_send_state(mdev);
1254 }
1255
1256 /* We want to pause/continue resync, tell peer. */
1257 if (ns.conn >= C_CONNECTED &&
1258 ((os.aftr_isp != ns.aftr_isp) ||
1259 (os.user_isp != ns.user_isp)))
1260 drbd_send_state(mdev);
1261
1262 /* In case one of the isp bits got set, suspend other devices. */
1263 if ((!os.aftr_isp && !os.peer_isp && !os.user_isp) &&
1264 (ns.aftr_isp || ns.peer_isp || ns.user_isp))
1265 suspend_other_sg(mdev);
1266
1267 /* Make sure the peer gets informed about eventual state
1268 changes (ISP bits) while we were in WFReportParams. */
1269 if (os.conn == C_WF_REPORT_PARAMS && ns.conn >= C_CONNECTED)
1270 drbd_send_state(mdev);
1271
1272 /* We are in the progress to start a full sync... */
1273 if ((os.conn != C_STARTING_SYNC_T && ns.conn == C_STARTING_SYNC_T) ||
1274 (os.conn != C_STARTING_SYNC_S && ns.conn == C_STARTING_SYNC_S))
1275 drbd_queue_bitmap_io(mdev, &drbd_bmio_set_n_write, &abw_start_sync, "set_n_write from StartingSync");
1276
1277 /* We are invalidating our self... */
1278 if (os.conn < C_CONNECTED && ns.conn < C_CONNECTED &&
1279 os.disk > D_INCONSISTENT && ns.disk == D_INCONSISTENT)
1280 drbd_queue_bitmap_io(mdev, &drbd_bmio_set_n_write, NULL, "set_n_write from invalidate");
1281
1282 if (os.disk > D_FAILED && ns.disk == D_FAILED) {
1283 enum drbd_io_error_p eh;
1284
1285 eh = EP_PASS_ON;
1286 if (get_ldev_if_state(mdev, D_FAILED)) {
1287 eh = mdev->ldev->dc.on_io_error;
1288 put_ldev(mdev);
1289 }
1290
1291 drbd_rs_cancel_all(mdev);
1292 /* since get_ldev() only works as long as disk>=D_INCONSISTENT,
1293 and it is D_DISKLESS here, local_cnt can only go down, it can
1294 not increase... It will reach zero */
1295 wait_event(mdev->misc_wait, !atomic_read(&mdev->local_cnt));
1296 mdev->rs_total = 0;
1297 mdev->rs_failed = 0;
1298 atomic_set(&mdev->rs_pending_cnt, 0);
1299
1300 spin_lock_irq(&mdev->req_lock);
1301 _drbd_set_state(_NS(mdev, disk, D_DISKLESS), CS_HARD, NULL);
1302 spin_unlock_irq(&mdev->req_lock);
1303
1304 if (eh == EP_CALL_HELPER)
1305 drbd_khelper(mdev, "local-io-error");
1306 }
1307
1308 if (os.disk > D_DISKLESS && ns.disk == D_DISKLESS) {
1309
1310 if (os.disk == D_FAILED) /* && ns.disk == D_DISKLESS*/ {
1311 if (drbd_send_state(mdev))
1312 dev_warn(DEV, "Notified peer that my disk is broken.\n");
1313 else
1314 dev_err(DEV, "Sending state in drbd_io_error() failed\n");
1315 }
1316
1317 lc_destroy(mdev->resync);
1318 mdev->resync = NULL;
1319 lc_destroy(mdev->act_log);
1320 mdev->act_log = NULL;
1321 __no_warn(local,
1322 drbd_free_bc(mdev->ldev);
1323 mdev->ldev = NULL;);
1324
1325 if (mdev->md_io_tmpp)
1326 __free_page(mdev->md_io_tmpp);
1327 }
1328
1329 /* Disks got bigger while they were detached */
1330 if (ns.disk > D_NEGOTIATING && ns.pdsk > D_NEGOTIATING &&
1331 test_and_clear_bit(RESYNC_AFTER_NEG, &mdev->flags)) {
1332 if (ns.conn == C_CONNECTED)
1333 resync_after_online_grow(mdev);
1334 }
1335
1336 /* A resync finished or aborted, wake paused devices... */
1337 if ((os.conn > C_CONNECTED && ns.conn <= C_CONNECTED) ||
1338 (os.peer_isp && !ns.peer_isp) ||
1339 (os.user_isp && !ns.user_isp))
1340 resume_next_sg(mdev);
1341
1342 /* Upon network connection, we need to start the receiver */
1343 if (os.conn == C_STANDALONE && ns.conn == C_UNCONNECTED)
1344 drbd_thread_start(&mdev->receiver);
1345
1346 /* Terminate worker thread if we are unconfigured - it will be
1347 restarted as needed... */
1348 if (ns.disk == D_DISKLESS &&
1349 ns.conn == C_STANDALONE &&
1350 ns.role == R_SECONDARY) {
1351 if (os.aftr_isp != ns.aftr_isp)
1352 resume_next_sg(mdev);
1353 /* set in __drbd_set_state, unless CONFIG_PENDING was set */
1354 if (test_bit(DEVICE_DYING, &mdev->flags))
1355 drbd_thread_stop_nowait(&mdev->worker);
1356 }
1357
1358 drbd_md_sync(mdev);
1359}
1360
1361
1362static int drbd_thread_setup(void *arg)
1363{
1364 struct drbd_thread *thi = (struct drbd_thread *) arg;
1365 struct drbd_conf *mdev = thi->mdev;
1366 unsigned long flags;
1367 int retval;
1368
1369restart:
1370 retval = thi->function(thi);
1371
1372 spin_lock_irqsave(&thi->t_lock, flags);
1373
1374 /* if the receiver has been "Exiting", the last thing it did
1375 * was set the conn state to "StandAlone",
1376 * if now a re-connect request comes in, conn state goes C_UNCONNECTED,
1377 * and receiver thread will be "started".
1378 * drbd_thread_start needs to set "Restarting" in that case.
1379 * t_state check and assignment needs to be within the same spinlock,
1380 * so either thread_start sees Exiting, and can remap to Restarting,
1381 * or thread_start see None, and can proceed as normal.
1382 */
1383
1384 if (thi->t_state == Restarting) {
1385 dev_info(DEV, "Restarting %s\n", current->comm);
1386 thi->t_state = Running;
1387 spin_unlock_irqrestore(&thi->t_lock, flags);
1388 goto restart;
1389 }
1390
1391 thi->task = NULL;
1392 thi->t_state = None;
1393 smp_mb();
1394 complete(&thi->stop);
1395 spin_unlock_irqrestore(&thi->t_lock, flags);
1396
1397 dev_info(DEV, "Terminating %s\n", current->comm);
1398
1399 /* Release mod reference taken when thread was started */
1400 module_put(THIS_MODULE);
1401 return retval;
1402}
1403
1404static void drbd_thread_init(struct drbd_conf *mdev, struct drbd_thread *thi,
1405 int (*func) (struct drbd_thread *))
1406{
1407 spin_lock_init(&thi->t_lock);
1408 thi->task = NULL;
1409 thi->t_state = None;
1410 thi->function = func;
1411 thi->mdev = mdev;
1412}
1413
1414int drbd_thread_start(struct drbd_thread *thi)
1415{
1416 struct drbd_conf *mdev = thi->mdev;
1417 struct task_struct *nt;
1418 unsigned long flags;
1419
1420 const char *me =
1421 thi == &mdev->receiver ? "receiver" :
1422 thi == &mdev->asender ? "asender" :
1423 thi == &mdev->worker ? "worker" : "NONSENSE";
1424
1425 /* is used from state engine doing drbd_thread_stop_nowait,
1426 * while holding the req lock irqsave */
1427 spin_lock_irqsave(&thi->t_lock, flags);
1428
1429 switch (thi->t_state) {
1430 case None:
1431 dev_info(DEV, "Starting %s thread (from %s [%d])\n",
1432 me, current->comm, current->pid);
1433
1434 /* Get ref on module for thread - this is released when thread exits */
1435 if (!try_module_get(THIS_MODULE)) {
1436 dev_err(DEV, "Failed to get module reference in drbd_thread_start\n");
1437 spin_unlock_irqrestore(&thi->t_lock, flags);
1438 return FALSE;
1439 }
1440
1441 init_completion(&thi->stop);
1442 D_ASSERT(thi->task == NULL);
1443 thi->reset_cpu_mask = 1;
1444 thi->t_state = Running;
1445 spin_unlock_irqrestore(&thi->t_lock, flags);
1446 flush_signals(current); /* otherw. may get -ERESTARTNOINTR */
1447
1448 nt = kthread_create(drbd_thread_setup, (void *) thi,
1449 "drbd%d_%s", mdev_to_minor(mdev), me);
1450
1451 if (IS_ERR(nt)) {
1452 dev_err(DEV, "Couldn't start thread\n");
1453
1454 module_put(THIS_MODULE);
1455 return FALSE;
1456 }
1457 spin_lock_irqsave(&thi->t_lock, flags);
1458 thi->task = nt;
1459 thi->t_state = Running;
1460 spin_unlock_irqrestore(&thi->t_lock, flags);
1461 wake_up_process(nt);
1462 break;
1463 case Exiting:
1464 thi->t_state = Restarting;
1465 dev_info(DEV, "Restarting %s thread (from %s [%d])\n",
1466 me, current->comm, current->pid);
1467 /* fall through */
1468 case Running:
1469 case Restarting:
1470 default:
1471 spin_unlock_irqrestore(&thi->t_lock, flags);
1472 break;
1473 }
1474
1475 return TRUE;
1476}
1477
1478
1479void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait)
1480{
1481 unsigned long flags;
1482
1483 enum drbd_thread_state ns = restart ? Restarting : Exiting;
1484
1485 /* may be called from state engine, holding the req lock irqsave */
1486 spin_lock_irqsave(&thi->t_lock, flags);
1487
1488 if (thi->t_state == None) {
1489 spin_unlock_irqrestore(&thi->t_lock, flags);
1490 if (restart)
1491 drbd_thread_start(thi);
1492 return;
1493 }
1494
1495 if (thi->t_state != ns) {
1496 if (thi->task == NULL) {
1497 spin_unlock_irqrestore(&thi->t_lock, flags);
1498 return;
1499 }
1500
1501 thi->t_state = ns;
1502 smp_mb();
1503 init_completion(&thi->stop);
1504 if (thi->task != current)
1505 force_sig(DRBD_SIGKILL, thi->task);
1506
1507 }
1508
1509 spin_unlock_irqrestore(&thi->t_lock, flags);
1510
1511 if (wait)
1512 wait_for_completion(&thi->stop);
1513}
1514
1515#ifdef CONFIG_SMP
1516/**
1517 * drbd_calc_cpu_mask() - Generate CPU masks, spread over all CPUs
1518 * @mdev: DRBD device.
1519 *
1520 * Forces all threads of a device onto the same CPU. This is beneficial for
1521 * DRBD's performance. May be overwritten by user's configuration.
1522 */
1523void drbd_calc_cpu_mask(struct drbd_conf *mdev)
1524{
1525 int ord, cpu;
1526
1527 /* user override. */
1528 if (cpumask_weight(mdev->cpu_mask))
1529 return;
1530
1531 ord = mdev_to_minor(mdev) % cpumask_weight(cpu_online_mask);
1532 for_each_online_cpu(cpu) {
1533 if (ord-- == 0) {
1534 cpumask_set_cpu(cpu, mdev->cpu_mask);
1535 return;
1536 }
1537 }
1538 /* should not be reached */
1539 cpumask_setall(mdev->cpu_mask);
1540}
1541
1542/**
1543 * drbd_thread_current_set_cpu() - modifies the cpu mask of the _current_ thread
1544 * @mdev: DRBD device.
1545 *
1546 * call in the "main loop" of _all_ threads, no need for any mutex, current won't die
1547 * prematurely.
1548 */
1549void drbd_thread_current_set_cpu(struct drbd_conf *mdev)
1550{
1551 struct task_struct *p = current;
1552 struct drbd_thread *thi =
1553 p == mdev->asender.task ? &mdev->asender :
1554 p == mdev->receiver.task ? &mdev->receiver :
1555 p == mdev->worker.task ? &mdev->worker :
1556 NULL;
1557 ERR_IF(thi == NULL)
1558 return;
1559 if (!thi->reset_cpu_mask)
1560 return;
1561 thi->reset_cpu_mask = 0;
1562 set_cpus_allowed_ptr(p, mdev->cpu_mask);
1563}
1564#endif
1565
1566/* the appropriate socket mutex must be held already */
1567int _drbd_send_cmd(struct drbd_conf *mdev, struct socket *sock,
1568 enum drbd_packets cmd, struct p_header *h,
1569 size_t size, unsigned msg_flags)
1570{
1571 int sent, ok;
1572
1573 ERR_IF(!h) return FALSE;
1574 ERR_IF(!size) return FALSE;
1575
1576 h->magic = BE_DRBD_MAGIC;
1577 h->command = cpu_to_be16(cmd);
1578 h->length = cpu_to_be16(size-sizeof(struct p_header));
1579
1580 trace_drbd_packet(mdev, sock, 0, (void *)h, __FILE__, __LINE__);
1581 sent = drbd_send(mdev, sock, h, size, msg_flags);
1582
1583 ok = (sent == size);
1584 if (!ok)
1585 dev_err(DEV, "short sent %s size=%d sent=%d\n",
1586 cmdname(cmd), (int)size, sent);
1587 return ok;
1588}
1589
1590/* don't pass the socket. we may only look at it
1591 * when we hold the appropriate socket mutex.
1592 */
1593int drbd_send_cmd(struct drbd_conf *mdev, int use_data_socket,
1594 enum drbd_packets cmd, struct p_header *h, size_t size)
1595{
1596 int ok = 0;
1597 struct socket *sock;
1598
1599 if (use_data_socket) {
1600 mutex_lock(&mdev->data.mutex);
1601 sock = mdev->data.socket;
1602 } else {
1603 mutex_lock(&mdev->meta.mutex);
1604 sock = mdev->meta.socket;
1605 }
1606
1607 /* drbd_disconnect() could have called drbd_free_sock()
1608 * while we were waiting in down()... */
1609 if (likely(sock != NULL))
1610 ok = _drbd_send_cmd(mdev, sock, cmd, h, size, 0);
1611
1612 if (use_data_socket)
1613 mutex_unlock(&mdev->data.mutex);
1614 else
1615 mutex_unlock(&mdev->meta.mutex);
1616 return ok;
1617}
1618
1619int drbd_send_cmd2(struct drbd_conf *mdev, enum drbd_packets cmd, char *data,
1620 size_t size)
1621{
1622 struct p_header h;
1623 int ok;
1624
1625 h.magic = BE_DRBD_MAGIC;
1626 h.command = cpu_to_be16(cmd);
1627 h.length = cpu_to_be16(size);
1628
1629 if (!drbd_get_data_sock(mdev))
1630 return 0;
1631
1632 trace_drbd_packet(mdev, mdev->data.socket, 0, (void *)&h, __FILE__, __LINE__);
1633
1634 ok = (sizeof(h) ==
1635 drbd_send(mdev, mdev->data.socket, &h, sizeof(h), 0));
1636 ok = ok && (size ==
1637 drbd_send(mdev, mdev->data.socket, data, size, 0));
1638
1639 drbd_put_data_sock(mdev);
1640
1641 return ok;
1642}
1643
1644int drbd_send_sync_param(struct drbd_conf *mdev, struct syncer_conf *sc)
1645{
1646 struct p_rs_param_89 *p;
1647 struct socket *sock;
1648 int size, rv;
1649 const int apv = mdev->agreed_pro_version;
1650
1651 size = apv <= 87 ? sizeof(struct p_rs_param)
1652 : apv == 88 ? sizeof(struct p_rs_param)
1653 + strlen(mdev->sync_conf.verify_alg) + 1
1654 : /* 89 */ sizeof(struct p_rs_param_89);
1655
1656 /* used from admin command context and receiver/worker context.
1657 * to avoid kmalloc, grab the socket right here,
1658 * then use the pre-allocated sbuf there */
1659 mutex_lock(&mdev->data.mutex);
1660 sock = mdev->data.socket;
1661
1662 if (likely(sock != NULL)) {
1663 enum drbd_packets cmd = apv >= 89 ? P_SYNC_PARAM89 : P_SYNC_PARAM;
1664
1665 p = &mdev->data.sbuf.rs_param_89;
1666
1667 /* initialize verify_alg and csums_alg */
1668 memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX);
1669
1670 p->rate = cpu_to_be32(sc->rate);
1671
1672 if (apv >= 88)
1673 strcpy(p->verify_alg, mdev->sync_conf.verify_alg);
1674 if (apv >= 89)
1675 strcpy(p->csums_alg, mdev->sync_conf.csums_alg);
1676
1677 rv = _drbd_send_cmd(mdev, sock, cmd, &p->head, size, 0);
1678 } else
1679 rv = 0; /* not ok */
1680
1681 mutex_unlock(&mdev->data.mutex);
1682
1683 return rv;
1684}
1685
1686int drbd_send_protocol(struct drbd_conf *mdev)
1687{
1688 struct p_protocol *p;
1689 int size, rv;
1690
1691 size = sizeof(struct p_protocol);
1692
1693 if (mdev->agreed_pro_version >= 87)
1694 size += strlen(mdev->net_conf->integrity_alg) + 1;
1695
1696 /* we must not recurse into our own queue,
1697 * as that is blocked during handshake */
1698 p = kmalloc(size, GFP_NOIO);
1699 if (p == NULL)
1700 return 0;
1701
1702 p->protocol = cpu_to_be32(mdev->net_conf->wire_protocol);
1703 p->after_sb_0p = cpu_to_be32(mdev->net_conf->after_sb_0p);
1704 p->after_sb_1p = cpu_to_be32(mdev->net_conf->after_sb_1p);
1705 p->after_sb_2p = cpu_to_be32(mdev->net_conf->after_sb_2p);
1706 p->want_lose = cpu_to_be32(mdev->net_conf->want_lose);
1707 p->two_primaries = cpu_to_be32(mdev->net_conf->two_primaries);
1708
1709 if (mdev->agreed_pro_version >= 87)
1710 strcpy(p->integrity_alg, mdev->net_conf->integrity_alg);
1711
1712 rv = drbd_send_cmd(mdev, USE_DATA_SOCKET, P_PROTOCOL,
1713 (struct p_header *)p, size);
1714 kfree(p);
1715 return rv;
1716}
1717
1718int _drbd_send_uuids(struct drbd_conf *mdev, u64 uuid_flags)
1719{
1720 struct p_uuids p;
1721 int i;
1722
1723 if (!get_ldev_if_state(mdev, D_NEGOTIATING))
1724 return 1;
1725
1726 for (i = UI_CURRENT; i < UI_SIZE; i++)
1727 p.uuid[i] = mdev->ldev ? cpu_to_be64(mdev->ldev->md.uuid[i]) : 0;
1728
1729 mdev->comm_bm_set = drbd_bm_total_weight(mdev);
1730 p.uuid[UI_SIZE] = cpu_to_be64(mdev->comm_bm_set);
1731 uuid_flags |= mdev->net_conf->want_lose ? 1 : 0;
1732 uuid_flags |= test_bit(CRASHED_PRIMARY, &mdev->flags) ? 2 : 0;
1733 uuid_flags |= mdev->new_state_tmp.disk == D_INCONSISTENT ? 4 : 0;
1734 p.uuid[UI_FLAGS] = cpu_to_be64(uuid_flags);
1735
1736 put_ldev(mdev);
1737
1738 return drbd_send_cmd(mdev, USE_DATA_SOCKET, P_UUIDS,
1739 (struct p_header *)&p, sizeof(p));
1740}
1741
1742int drbd_send_uuids(struct drbd_conf *mdev)
1743{
1744 return _drbd_send_uuids(mdev, 0);
1745}
1746
1747int drbd_send_uuids_skip_initial_sync(struct drbd_conf *mdev)
1748{
1749 return _drbd_send_uuids(mdev, 8);
1750}
1751
1752
1753int drbd_send_sync_uuid(struct drbd_conf *mdev, u64 val)
1754{
1755 struct p_rs_uuid p;
1756
1757 p.uuid = cpu_to_be64(val);
1758
1759 return drbd_send_cmd(mdev, USE_DATA_SOCKET, P_SYNC_UUID,
1760 (struct p_header *)&p, sizeof(p));
1761}
1762
1763int drbd_send_sizes(struct drbd_conf *mdev, int trigger_reply)
1764{
1765 struct p_sizes p;
1766 sector_t d_size, u_size;
1767 int q_order_type;
1768 int ok;
1769
1770 if (get_ldev_if_state(mdev, D_NEGOTIATING)) {
1771 D_ASSERT(mdev->ldev->backing_bdev);
1772 d_size = drbd_get_max_capacity(mdev->ldev);
1773 u_size = mdev->ldev->dc.disk_size;
1774 q_order_type = drbd_queue_order_type(mdev);
1775 p.queue_order_type = cpu_to_be32(drbd_queue_order_type(mdev));
1776 put_ldev(mdev);
1777 } else {
1778 d_size = 0;
1779 u_size = 0;
1780 q_order_type = QUEUE_ORDERED_NONE;
1781 }
1782
1783 p.d_size = cpu_to_be64(d_size);
1784 p.u_size = cpu_to_be64(u_size);
1785 p.c_size = cpu_to_be64(trigger_reply ? 0 : drbd_get_capacity(mdev->this_bdev));
1786 p.max_segment_size = cpu_to_be32(queue_max_segment_size(mdev->rq_queue));
1787 p.queue_order_type = cpu_to_be32(q_order_type);
1788
1789 ok = drbd_send_cmd(mdev, USE_DATA_SOCKET, P_SIZES,
1790 (struct p_header *)&p, sizeof(p));
1791 return ok;
1792}
1793
1794/**
1795 * drbd_send_state() - Sends the drbd state to the peer
1796 * @mdev: DRBD device.
1797 */
1798int drbd_send_state(struct drbd_conf *mdev)
1799{
1800 struct socket *sock;
1801 struct p_state p;
1802 int ok = 0;
1803
1804 /* Grab state lock so we wont send state if we're in the middle
1805 * of a cluster wide state change on another thread */
1806 drbd_state_lock(mdev);
1807
1808 mutex_lock(&mdev->data.mutex);
1809
1810 p.state = cpu_to_be32(mdev->state.i); /* Within the send mutex */
1811 sock = mdev->data.socket;
1812
1813 if (likely(sock != NULL)) {
1814 ok = _drbd_send_cmd(mdev, sock, P_STATE,
1815 (struct p_header *)&p, sizeof(p), 0);
1816 }
1817
1818 mutex_unlock(&mdev->data.mutex);
1819
1820 drbd_state_unlock(mdev);
1821 return ok;
1822}
1823
1824int drbd_send_state_req(struct drbd_conf *mdev,
1825 union drbd_state mask, union drbd_state val)
1826{
1827 struct p_req_state p;
1828
1829 p.mask = cpu_to_be32(mask.i);
1830 p.val = cpu_to_be32(val.i);
1831
1832 return drbd_send_cmd(mdev, USE_DATA_SOCKET, P_STATE_CHG_REQ,
1833 (struct p_header *)&p, sizeof(p));
1834}
1835
1836int drbd_send_sr_reply(struct drbd_conf *mdev, int retcode)
1837{
1838 struct p_req_state_reply p;
1839
1840 p.retcode = cpu_to_be32(retcode);
1841
1842 return drbd_send_cmd(mdev, USE_META_SOCKET, P_STATE_CHG_REPLY,
1843 (struct p_header *)&p, sizeof(p));
1844}
1845
1846int fill_bitmap_rle_bits(struct drbd_conf *mdev,
1847 struct p_compressed_bm *p,
1848 struct bm_xfer_ctx *c)
1849{
1850 struct bitstream bs;
1851 unsigned long plain_bits;
1852 unsigned long tmp;
1853 unsigned long rl;
1854 unsigned len;
1855 unsigned toggle;
1856 int bits;
1857
1858 /* may we use this feature? */
1859 if ((mdev->sync_conf.use_rle == 0) ||
1860 (mdev->agreed_pro_version < 90))
1861 return 0;
1862
1863 if (c->bit_offset >= c->bm_bits)
1864 return 0; /* nothing to do. */
1865
1866 /* use at most thus many bytes */
1867 bitstream_init(&bs, p->code, BM_PACKET_VLI_BYTES_MAX, 0);
1868 memset(p->code, 0, BM_PACKET_VLI_BYTES_MAX);
1869 /* plain bits covered in this code string */
1870 plain_bits = 0;
1871
1872 /* p->encoding & 0x80 stores whether the first run length is set.
1873 * bit offset is implicit.
1874 * start with toggle == 2 to be able to tell the first iteration */
1875 toggle = 2;
1876
1877 /* see how much plain bits we can stuff into one packet
1878 * using RLE and VLI. */
1879 do {
1880 tmp = (toggle == 0) ? _drbd_bm_find_next_zero(mdev, c->bit_offset)
1881 : _drbd_bm_find_next(mdev, c->bit_offset);
1882 if (tmp == -1UL)
1883 tmp = c->bm_bits;
1884 rl = tmp - c->bit_offset;
1885
1886 if (toggle == 2) { /* first iteration */
1887 if (rl == 0) {
1888 /* the first checked bit was set,
1889 * store start value, */
1890 DCBP_set_start(p, 1);
1891 /* but skip encoding of zero run length */
1892 toggle = !toggle;
1893 continue;
1894 }
1895 DCBP_set_start(p, 0);
1896 }
1897
1898 /* paranoia: catch zero runlength.
1899 * can only happen if bitmap is modified while we scan it. */
1900 if (rl == 0) {
1901 dev_err(DEV, "unexpected zero runlength while encoding bitmap "
1902 "t:%u bo:%lu\n", toggle, c->bit_offset);
1903 return -1;
1904 }
1905
1906 bits = vli_encode_bits(&bs, rl);
1907 if (bits == -ENOBUFS) /* buffer full */
1908 break;
1909 if (bits <= 0) {
1910 dev_err(DEV, "error while encoding bitmap: %d\n", bits);
1911 return 0;
1912 }
1913
1914 toggle = !toggle;
1915 plain_bits += rl;
1916 c->bit_offset = tmp;
1917 } while (c->bit_offset < c->bm_bits);
1918
1919 len = bs.cur.b - p->code + !!bs.cur.bit;
1920
1921 if (plain_bits < (len << 3)) {
1922 /* incompressible with this method.
1923 * we need to rewind both word and bit position. */
1924 c->bit_offset -= plain_bits;
1925 bm_xfer_ctx_bit_to_word_offset(c);
1926 c->bit_offset = c->word_offset * BITS_PER_LONG;
1927 return 0;
1928 }
1929
1930 /* RLE + VLI was able to compress it just fine.
1931 * update c->word_offset. */
1932 bm_xfer_ctx_bit_to_word_offset(c);
1933
1934 /* store pad_bits */
1935 DCBP_set_pad_bits(p, (8 - bs.cur.bit) & 0x7);
1936
1937 return len;
1938}
1939
1940enum { OK, FAILED, DONE }
1941send_bitmap_rle_or_plain(struct drbd_conf *mdev,
1942 struct p_header *h, struct bm_xfer_ctx *c)
1943{
1944 struct p_compressed_bm *p = (void*)h;
1945 unsigned long num_words;
1946 int len;
1947 int ok;
1948
1949 len = fill_bitmap_rle_bits(mdev, p, c);
1950
1951 if (len < 0)
1952 return FAILED;
1953
1954 if (len) {
1955 DCBP_set_code(p, RLE_VLI_Bits);
1956 ok = _drbd_send_cmd(mdev, mdev->data.socket, P_COMPRESSED_BITMAP, h,
1957 sizeof(*p) + len, 0);
1958
1959 c->packets[0]++;
1960 c->bytes[0] += sizeof(*p) + len;
1961
1962 if (c->bit_offset >= c->bm_bits)
1963 len = 0; /* DONE */
1964 } else {
1965 /* was not compressible.
1966 * send a buffer full of plain text bits instead. */
1967 num_words = min_t(size_t, BM_PACKET_WORDS, c->bm_words - c->word_offset);
1968 len = num_words * sizeof(long);
1969 if (len)
1970 drbd_bm_get_lel(mdev, c->word_offset, num_words, (unsigned long*)h->payload);
1971 ok = _drbd_send_cmd(mdev, mdev->data.socket, P_BITMAP,
1972 h, sizeof(struct p_header) + len, 0);
1973 c->word_offset += num_words;
1974 c->bit_offset = c->word_offset * BITS_PER_LONG;
1975
1976 c->packets[1]++;
1977 c->bytes[1] += sizeof(struct p_header) + len;
1978
1979 if (c->bit_offset > c->bm_bits)
1980 c->bit_offset = c->bm_bits;
1981 }
1982 ok = ok ? ((len == 0) ? DONE : OK) : FAILED;
1983
1984 if (ok == DONE)
1985 INFO_bm_xfer_stats(mdev, "send", c);
1986 return ok;
1987}
1988
1989/* See the comment at receive_bitmap() */
1990int _drbd_send_bitmap(struct drbd_conf *mdev)
1991{
1992 struct bm_xfer_ctx c;
1993 struct p_header *p;
1994 int ret;
1995
1996 ERR_IF(!mdev->bitmap) return FALSE;
1997
1998 /* maybe we should use some per thread scratch page,
1999 * and allocate that during initial device creation? */
2000 p = (struct p_header *) __get_free_page(GFP_NOIO);
2001 if (!p) {
2002 dev_err(DEV, "failed to allocate one page buffer in %s\n", __func__);
2003 return FALSE;
2004 }
2005
2006 if (get_ldev(mdev)) {
2007 if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
2008 dev_info(DEV, "Writing the whole bitmap, MDF_FullSync was set.\n");
2009 drbd_bm_set_all(mdev);
2010 if (drbd_bm_write(mdev)) {
2011 /* write_bm did fail! Leave full sync flag set in Meta P_DATA
2012 * but otherwise process as per normal - need to tell other
2013 * side that a full resync is required! */
2014 dev_err(DEV, "Failed to write bitmap to disk!\n");
2015 } else {
2016 drbd_md_clear_flag(mdev, MDF_FULL_SYNC);
2017 drbd_md_sync(mdev);
2018 }
2019 }
2020 put_ldev(mdev);
2021 }
2022
2023 c = (struct bm_xfer_ctx) {
2024 .bm_bits = drbd_bm_bits(mdev),
2025 .bm_words = drbd_bm_words(mdev),
2026 };
2027
2028 do {
2029 ret = send_bitmap_rle_or_plain(mdev, p, &c);
2030 } while (ret == OK);
2031
2032 free_page((unsigned long) p);
2033 return (ret == DONE);
2034}
2035
2036int drbd_send_bitmap(struct drbd_conf *mdev)
2037{
2038 int err;
2039
2040 if (!drbd_get_data_sock(mdev))
2041 return -1;
2042 err = !_drbd_send_bitmap(mdev);
2043 drbd_put_data_sock(mdev);
2044 return err;
2045}
2046
2047int drbd_send_b_ack(struct drbd_conf *mdev, u32 barrier_nr, u32 set_size)
2048{
2049 int ok;
2050 struct p_barrier_ack p;
2051
2052 p.barrier = barrier_nr;
2053 p.set_size = cpu_to_be32(set_size);
2054
2055 if (mdev->state.conn < C_CONNECTED)
2056 return FALSE;
2057 ok = drbd_send_cmd(mdev, USE_META_SOCKET, P_BARRIER_ACK,
2058 (struct p_header *)&p, sizeof(p));
2059 return ok;
2060}
2061
2062/**
2063 * _drbd_send_ack() - Sends an ack packet
2064 * @mdev: DRBD device.
2065 * @cmd: Packet command code.
2066 * @sector: sector, needs to be in big endian byte order
2067 * @blksize: size in byte, needs to be in big endian byte order
2068 * @block_id: Id, big endian byte order
2069 */
2070static int _drbd_send_ack(struct drbd_conf *mdev, enum drbd_packets cmd,
2071 u64 sector,
2072 u32 blksize,
2073 u64 block_id)
2074{
2075 int ok;
2076 struct p_block_ack p;
2077
2078 p.sector = sector;
2079 p.block_id = block_id;
2080 p.blksize = blksize;
2081 p.seq_num = cpu_to_be32(atomic_add_return(1, &mdev->packet_seq));
2082
2083 if (!mdev->meta.socket || mdev->state.conn < C_CONNECTED)
2084 return FALSE;
2085 ok = drbd_send_cmd(mdev, USE_META_SOCKET, cmd,
2086 (struct p_header *)&p, sizeof(p));
2087 return ok;
2088}
2089
2090int drbd_send_ack_dp(struct drbd_conf *mdev, enum drbd_packets cmd,
2091 struct p_data *dp)
2092{
2093 const int header_size = sizeof(struct p_data)
2094 - sizeof(struct p_header);
2095 int data_size = ((struct p_header *)dp)->length - header_size;
2096
2097 return _drbd_send_ack(mdev, cmd, dp->sector, cpu_to_be32(data_size),
2098 dp->block_id);
2099}
2100
2101int drbd_send_ack_rp(struct drbd_conf *mdev, enum drbd_packets cmd,
2102 struct p_block_req *rp)
2103{
2104 return _drbd_send_ack(mdev, cmd, rp->sector, rp->blksize, rp->block_id);
2105}
2106
2107/**
2108 * drbd_send_ack() - Sends an ack packet
2109 * @mdev: DRBD device.
2110 * @cmd: Packet command code.
2111 * @e: Epoch entry.
2112 */
2113int drbd_send_ack(struct drbd_conf *mdev,
2114 enum drbd_packets cmd, struct drbd_epoch_entry *e)
2115{
2116 return _drbd_send_ack(mdev, cmd,
2117 cpu_to_be64(e->sector),
2118 cpu_to_be32(e->size),
2119 e->block_id);
2120}
2121
2122/* This function misuses the block_id field to signal if the blocks
2123 * are is sync or not. */
2124int drbd_send_ack_ex(struct drbd_conf *mdev, enum drbd_packets cmd,
2125 sector_t sector, int blksize, u64 block_id)
2126{
2127 return _drbd_send_ack(mdev, cmd,
2128 cpu_to_be64(sector),
2129 cpu_to_be32(blksize),
2130 cpu_to_be64(block_id));
2131}
2132
2133int drbd_send_drequest(struct drbd_conf *mdev, int cmd,
2134 sector_t sector, int size, u64 block_id)
2135{
2136 int ok;
2137 struct p_block_req p;
2138
2139 p.sector = cpu_to_be64(sector);
2140 p.block_id = block_id;
2141 p.blksize = cpu_to_be32(size);
2142
2143 ok = drbd_send_cmd(mdev, USE_DATA_SOCKET, cmd,
2144 (struct p_header *)&p, sizeof(p));
2145 return ok;
2146}
2147
2148int drbd_send_drequest_csum(struct drbd_conf *mdev,
2149 sector_t sector, int size,
2150 void *digest, int digest_size,
2151 enum drbd_packets cmd)
2152{
2153 int ok;
2154 struct p_block_req p;
2155
2156 p.sector = cpu_to_be64(sector);
2157 p.block_id = BE_DRBD_MAGIC + 0xbeef;
2158 p.blksize = cpu_to_be32(size);
2159
2160 p.head.magic = BE_DRBD_MAGIC;
2161 p.head.command = cpu_to_be16(cmd);
2162 p.head.length = cpu_to_be16(sizeof(p) - sizeof(struct p_header) + digest_size);
2163
2164 mutex_lock(&mdev->data.mutex);
2165
2166 ok = (sizeof(p) == drbd_send(mdev, mdev->data.socket, &p, sizeof(p), 0));
2167 ok = ok && (digest_size == drbd_send(mdev, mdev->data.socket, digest, digest_size, 0));
2168
2169 mutex_unlock(&mdev->data.mutex);
2170
2171 return ok;
2172}
2173
2174int drbd_send_ov_request(struct drbd_conf *mdev, sector_t sector, int size)
2175{
2176 int ok;
2177 struct p_block_req p;
2178
2179 p.sector = cpu_to_be64(sector);
2180 p.block_id = BE_DRBD_MAGIC + 0xbabe;
2181 p.blksize = cpu_to_be32(size);
2182
2183 ok = drbd_send_cmd(mdev, USE_DATA_SOCKET, P_OV_REQUEST,
2184 (struct p_header *)&p, sizeof(p));
2185 return ok;
2186}
2187
2188/* called on sndtimeo
2189 * returns FALSE if we should retry,
2190 * TRUE if we think connection is dead
2191 */
2192static int we_should_drop_the_connection(struct drbd_conf *mdev, struct socket *sock)
2193{
2194 int drop_it;
2195 /* long elapsed = (long)(jiffies - mdev->last_received); */
2196
2197 drop_it = mdev->meta.socket == sock
2198 || !mdev->asender.task
2199 || get_t_state(&mdev->asender) != Running
2200 || mdev->state.conn < C_CONNECTED;
2201
2202 if (drop_it)
2203 return TRUE;
2204
2205 drop_it = !--mdev->ko_count;
2206 if (!drop_it) {
2207 dev_err(DEV, "[%s/%d] sock_sendmsg time expired, ko = %u\n",
2208 current->comm, current->pid, mdev->ko_count);
2209 request_ping(mdev);
2210 }
2211
2212 return drop_it; /* && (mdev->state == R_PRIMARY) */;
2213}
2214
2215/* The idea of sendpage seems to be to put some kind of reference
2216 * to the page into the skb, and to hand it over to the NIC. In
2217 * this process get_page() gets called.
2218 *
2219 * As soon as the page was really sent over the network put_page()
2220 * gets called by some part of the network layer. [ NIC driver? ]
2221 *
2222 * [ get_page() / put_page() increment/decrement the count. If count
2223 * reaches 0 the page will be freed. ]
2224 *
2225 * This works nicely with pages from FSs.
2226 * But this means that in protocol A we might signal IO completion too early!
2227 *
2228 * In order not to corrupt data during a resync we must make sure
2229 * that we do not reuse our own buffer pages (EEs) to early, therefore
2230 * we have the net_ee list.
2231 *
2232 * XFS seems to have problems, still, it submits pages with page_count == 0!
2233 * As a workaround, we disable sendpage on pages
2234 * with page_count == 0 or PageSlab.
2235 */
2236static int _drbd_no_send_page(struct drbd_conf *mdev, struct page *page,
2237 int offset, size_t size)
2238{
2239 int sent = drbd_send(mdev, mdev->data.socket, kmap(page) + offset, size, 0);
2240 kunmap(page);
2241 if (sent == size)
2242 mdev->send_cnt += size>>9;
2243 return sent == size;
2244}
2245
2246static int _drbd_send_page(struct drbd_conf *mdev, struct page *page,
2247 int offset, size_t size)
2248{
2249 mm_segment_t oldfs = get_fs();
2250 int sent, ok;
2251 int len = size;
2252
2253 /* e.g. XFS meta- & log-data is in slab pages, which have a
2254 * page_count of 0 and/or have PageSlab() set.
2255 * we cannot use send_page for those, as that does get_page();
2256 * put_page(); and would cause either a VM_BUG directly, or
2257 * __page_cache_release a page that would actually still be referenced
2258 * by someone, leading to some obscure delayed Oops somewhere else. */
2259 if (disable_sendpage || (page_count(page) < 1) || PageSlab(page))
2260 return _drbd_no_send_page(mdev, page, offset, size);
2261
2262 drbd_update_congested(mdev);
2263 set_fs(KERNEL_DS);
2264 do {
2265 sent = mdev->data.socket->ops->sendpage(mdev->data.socket, page,
2266 offset, len,
2267 MSG_NOSIGNAL);
2268 if (sent == -EAGAIN) {
2269 if (we_should_drop_the_connection(mdev,
2270 mdev->data.socket))
2271 break;
2272 else
2273 continue;
2274 }
2275 if (sent <= 0) {
2276 dev_warn(DEV, "%s: size=%d len=%d sent=%d\n",
2277 __func__, (int)size, len, sent);
2278 break;
2279 }
2280 len -= sent;
2281 offset += sent;
2282 } while (len > 0 /* THINK && mdev->cstate >= C_CONNECTED*/);
2283 set_fs(oldfs);
2284 clear_bit(NET_CONGESTED, &mdev->flags);
2285
2286 ok = (len == 0);
2287 if (likely(ok))
2288 mdev->send_cnt += size>>9;
2289 return ok;
2290}
2291
2292static int _drbd_send_bio(struct drbd_conf *mdev, struct bio *bio)
2293{
2294 struct bio_vec *bvec;
2295 int i;
2296 __bio_for_each_segment(bvec, bio, i, 0) {
2297 if (!_drbd_no_send_page(mdev, bvec->bv_page,
2298 bvec->bv_offset, bvec->bv_len))
2299 return 0;
2300 }
2301 return 1;
2302}
2303
2304static int _drbd_send_zc_bio(struct drbd_conf *mdev, struct bio *bio)
2305{
2306 struct bio_vec *bvec;
2307 int i;
2308 __bio_for_each_segment(bvec, bio, i, 0) {
2309 if (!_drbd_send_page(mdev, bvec->bv_page,
2310 bvec->bv_offset, bvec->bv_len))
2311 return 0;
2312 }
2313
2314 return 1;
2315}
2316
2317/* Used to send write requests
2318 * R_PRIMARY -> Peer (P_DATA)
2319 */
2320int drbd_send_dblock(struct drbd_conf *mdev, struct drbd_request *req)
2321{
2322 int ok = 1;
2323 struct p_data p;
2324 unsigned int dp_flags = 0;
2325 void *dgb;
2326 int dgs;
2327
2328 if (!drbd_get_data_sock(mdev))
2329 return 0;
2330
2331 dgs = (mdev->agreed_pro_version >= 87 && mdev->integrity_w_tfm) ?
2332 crypto_hash_digestsize(mdev->integrity_w_tfm) : 0;
2333
2334 p.head.magic = BE_DRBD_MAGIC;
2335 p.head.command = cpu_to_be16(P_DATA);
2336 p.head.length =
2337 cpu_to_be16(sizeof(p) - sizeof(struct p_header) + dgs + req->size);
2338
2339 p.sector = cpu_to_be64(req->sector);
2340 p.block_id = (unsigned long)req;
2341 p.seq_num = cpu_to_be32(req->seq_num =
2342 atomic_add_return(1, &mdev->packet_seq));
2343 dp_flags = 0;
2344
2345 /* NOTE: no need to check if barriers supported here as we would
2346 * not pass the test in make_request_common in that case
2347 */
2348 if (bio_rw_flagged(req->master_bio, BIO_RW_BARRIER)) {
2349 dev_err(DEV, "ASSERT FAILED would have set DP_HARDBARRIER\n");
2350 /* dp_flags |= DP_HARDBARRIER; */
2351 }
2352 if (bio_rw_flagged(req->master_bio, BIO_RW_SYNCIO))
2353 dp_flags |= DP_RW_SYNC;
2354 /* for now handle SYNCIO and UNPLUG
2355 * as if they still were one and the same flag */
2356 if (bio_rw_flagged(req->master_bio, BIO_RW_UNPLUG))
2357 dp_flags |= DP_RW_SYNC;
2358 if (mdev->state.conn >= C_SYNC_SOURCE &&
2359 mdev->state.conn <= C_PAUSED_SYNC_T)
2360 dp_flags |= DP_MAY_SET_IN_SYNC;
2361
2362 p.dp_flags = cpu_to_be32(dp_flags);
2363 trace_drbd_packet(mdev, mdev->data.socket, 0, (void *)&p, __FILE__, __LINE__);
2364 set_bit(UNPLUG_REMOTE, &mdev->flags);
2365 ok = (sizeof(p) ==
2366 drbd_send(mdev, mdev->data.socket, &p, sizeof(p), MSG_MORE));
2367 if (ok && dgs) {
2368 dgb = mdev->int_dig_out;
2369 drbd_csum(mdev, mdev->integrity_w_tfm, req->master_bio, dgb);
2370 ok = drbd_send(mdev, mdev->data.socket, dgb, dgs, MSG_MORE);
2371 }
2372 if (ok) {
2373 if (mdev->net_conf->wire_protocol == DRBD_PROT_A)
2374 ok = _drbd_send_bio(mdev, req->master_bio);
2375 else
2376 ok = _drbd_send_zc_bio(mdev, req->master_bio);
2377 }
2378
2379 drbd_put_data_sock(mdev);
2380 return ok;
2381}
2382
2383/* answer packet, used to send data back for read requests:
2384 * Peer -> (diskless) R_PRIMARY (P_DATA_REPLY)
2385 * C_SYNC_SOURCE -> C_SYNC_TARGET (P_RS_DATA_REPLY)
2386 */
2387int drbd_send_block(struct drbd_conf *mdev, enum drbd_packets cmd,
2388 struct drbd_epoch_entry *e)
2389{
2390 int ok;
2391 struct p_data p;
2392 void *dgb;
2393 int dgs;
2394
2395 dgs = (mdev->agreed_pro_version >= 87 && mdev->integrity_w_tfm) ?
2396 crypto_hash_digestsize(mdev->integrity_w_tfm) : 0;
2397
2398 p.head.magic = BE_DRBD_MAGIC;
2399 p.head.command = cpu_to_be16(cmd);
2400 p.head.length =
2401 cpu_to_be16(sizeof(p) - sizeof(struct p_header) + dgs + e->size);
2402
2403 p.sector = cpu_to_be64(e->sector);
2404 p.block_id = e->block_id;
2405 /* p.seq_num = 0; No sequence numbers here.. */
2406
2407 /* Only called by our kernel thread.
2408 * This one may be interrupted by DRBD_SIG and/or DRBD_SIGKILL
2409 * in response to admin command or module unload.
2410 */
2411 if (!drbd_get_data_sock(mdev))
2412 return 0;
2413
2414 trace_drbd_packet(mdev, mdev->data.socket, 0, (void *)&p, __FILE__, __LINE__);
2415 ok = sizeof(p) == drbd_send(mdev, mdev->data.socket, &p,
2416 sizeof(p), MSG_MORE);
2417 if (ok && dgs) {
2418 dgb = mdev->int_dig_out;
2419 drbd_csum(mdev, mdev->integrity_w_tfm, e->private_bio, dgb);
2420 ok = drbd_send(mdev, mdev->data.socket, dgb, dgs, MSG_MORE);
2421 }
2422 if (ok)
2423 ok = _drbd_send_zc_bio(mdev, e->private_bio);
2424
2425 drbd_put_data_sock(mdev);
2426 return ok;
2427}
2428
2429/*
2430 drbd_send distinguishes two cases:
2431
2432 Packets sent via the data socket "sock"
2433 and packets sent via the meta data socket "msock"
2434
2435 sock msock
2436 -----------------+-------------------------+------------------------------
2437 timeout conf.timeout / 2 conf.timeout / 2
2438 timeout action send a ping via msock Abort communication
2439 and close all sockets
2440*/
2441
2442/*
2443 * you must have down()ed the appropriate [m]sock_mutex elsewhere!
2444 */
2445int drbd_send(struct drbd_conf *mdev, struct socket *sock,
2446 void *buf, size_t size, unsigned msg_flags)
2447{
2448 struct kvec iov;
2449 struct msghdr msg;
2450 int rv, sent = 0;
2451
2452 if (!sock)
2453 return -1000;
2454
2455 /* THINK if (signal_pending) return ... ? */
2456
2457 iov.iov_base = buf;
2458 iov.iov_len = size;
2459
2460 msg.msg_name = NULL;
2461 msg.msg_namelen = 0;
2462 msg.msg_control = NULL;
2463 msg.msg_controllen = 0;
2464 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
2465
2466 if (sock == mdev->data.socket) {
2467 mdev->ko_count = mdev->net_conf->ko_count;
2468 drbd_update_congested(mdev);
2469 }
2470 do {
2471 /* STRANGE
2472 * tcp_sendmsg does _not_ use its size parameter at all ?
2473 *
2474 * -EAGAIN on timeout, -EINTR on signal.
2475 */
2476/* THINK
2477 * do we need to block DRBD_SIG if sock == &meta.socket ??
2478 * otherwise wake_asender() might interrupt some send_*Ack !
2479 */
2480 rv = kernel_sendmsg(sock, &msg, &iov, 1, size);
2481 if (rv == -EAGAIN) {
2482 if (we_should_drop_the_connection(mdev, sock))
2483 break;
2484 else
2485 continue;
2486 }
2487 D_ASSERT(rv != 0);
2488 if (rv == -EINTR) {
2489 flush_signals(current);
2490 rv = 0;
2491 }
2492 if (rv < 0)
2493 break;
2494 sent += rv;
2495 iov.iov_base += rv;
2496 iov.iov_len -= rv;
2497 } while (sent < size);
2498
2499 if (sock == mdev->data.socket)
2500 clear_bit(NET_CONGESTED, &mdev->flags);
2501
2502 if (rv <= 0) {
2503 if (rv != -EAGAIN) {
2504 dev_err(DEV, "%s_sendmsg returned %d\n",
2505 sock == mdev->meta.socket ? "msock" : "sock",
2506 rv);
2507 drbd_force_state(mdev, NS(conn, C_BROKEN_PIPE));
2508 } else
2509 drbd_force_state(mdev, NS(conn, C_TIMEOUT));
2510 }
2511
2512 return sent;
2513}
2514
2515static int drbd_open(struct block_device *bdev, fmode_t mode)
2516{
2517 struct drbd_conf *mdev = bdev->bd_disk->private_data;
2518 unsigned long flags;
2519 int rv = 0;
2520
2521 spin_lock_irqsave(&mdev->req_lock, flags);
2522 /* to have a stable mdev->state.role
2523 * and no race with updating open_cnt */
2524
2525 if (mdev->state.role != R_PRIMARY) {
2526 if (mode & FMODE_WRITE)
2527 rv = -EROFS;
2528 else if (!allow_oos)
2529 rv = -EMEDIUMTYPE;
2530 }
2531
2532 if (!rv)
2533 mdev->open_cnt++;
2534 spin_unlock_irqrestore(&mdev->req_lock, flags);
2535
2536 return rv;
2537}
2538
2539static int drbd_release(struct gendisk *gd, fmode_t mode)
2540{
2541 struct drbd_conf *mdev = gd->private_data;
2542 mdev->open_cnt--;
2543 return 0;
2544}
2545
2546static void drbd_unplug_fn(struct request_queue *q)
2547{
2548 struct drbd_conf *mdev = q->queuedata;
2549
2550 trace_drbd_unplug(mdev, "got unplugged");
2551
2552 /* unplug FIRST */
2553 spin_lock_irq(q->queue_lock);
2554 blk_remove_plug(q);
2555 spin_unlock_irq(q->queue_lock);
2556
2557 /* only if connected */
2558 spin_lock_irq(&mdev->req_lock);
2559 if (mdev->state.pdsk >= D_INCONSISTENT && mdev->state.conn >= C_CONNECTED) {
2560 D_ASSERT(mdev->state.role == R_PRIMARY);
2561 if (test_and_clear_bit(UNPLUG_REMOTE, &mdev->flags)) {
2562 /* add to the data.work queue,
2563 * unless already queued.
2564 * XXX this might be a good addition to drbd_queue_work
2565 * anyways, to detect "double queuing" ... */
2566 if (list_empty(&mdev->unplug_work.list))
2567 drbd_queue_work(&mdev->data.work,
2568 &mdev->unplug_work);
2569 }
2570 }
2571 spin_unlock_irq(&mdev->req_lock);
2572
2573 if (mdev->state.disk >= D_INCONSISTENT)
2574 drbd_kick_lo(mdev);
2575}
2576
2577static void drbd_set_defaults(struct drbd_conf *mdev)
2578{
2579 mdev->sync_conf.after = DRBD_AFTER_DEF;
2580 mdev->sync_conf.rate = DRBD_RATE_DEF;
2581 mdev->sync_conf.al_extents = DRBD_AL_EXTENTS_DEF;
2582 mdev->state = (union drbd_state) {
2583 { .role = R_SECONDARY,
2584 .peer = R_UNKNOWN,
2585 .conn = C_STANDALONE,
2586 .disk = D_DISKLESS,
2587 .pdsk = D_UNKNOWN,
2588 .susp = 0
2589 } };
2590}
2591
2592void drbd_init_set_defaults(struct drbd_conf *mdev)
2593{
2594 /* the memset(,0,) did most of this.
2595 * note: only assignments, no allocation in here */
2596
2597 drbd_set_defaults(mdev);
2598
2599 /* for now, we do NOT yet support it,
2600 * even though we start some framework
2601 * to eventually support barriers */
2602 set_bit(NO_BARRIER_SUPP, &mdev->flags);
2603
2604 atomic_set(&mdev->ap_bio_cnt, 0);
2605 atomic_set(&mdev->ap_pending_cnt, 0);
2606 atomic_set(&mdev->rs_pending_cnt, 0);
2607 atomic_set(&mdev->unacked_cnt, 0);
2608 atomic_set(&mdev->local_cnt, 0);
2609 atomic_set(&mdev->net_cnt, 0);
2610 atomic_set(&mdev->packet_seq, 0);
2611 atomic_set(&mdev->pp_in_use, 0);
2612
2613 mutex_init(&mdev->md_io_mutex);
2614 mutex_init(&mdev->data.mutex);
2615 mutex_init(&mdev->meta.mutex);
2616 sema_init(&mdev->data.work.s, 0);
2617 sema_init(&mdev->meta.work.s, 0);
2618 mutex_init(&mdev->state_mutex);
2619
2620 spin_lock_init(&mdev->data.work.q_lock);
2621 spin_lock_init(&mdev->meta.work.q_lock);
2622
2623 spin_lock_init(&mdev->al_lock);
2624 spin_lock_init(&mdev->req_lock);
2625 spin_lock_init(&mdev->peer_seq_lock);
2626 spin_lock_init(&mdev->epoch_lock);
2627
2628 INIT_LIST_HEAD(&mdev->active_ee);
2629 INIT_LIST_HEAD(&mdev->sync_ee);
2630 INIT_LIST_HEAD(&mdev->done_ee);
2631 INIT_LIST_HEAD(&mdev->read_ee);
2632 INIT_LIST_HEAD(&mdev->net_ee);
2633 INIT_LIST_HEAD(&mdev->resync_reads);
2634 INIT_LIST_HEAD(&mdev->data.work.q);
2635 INIT_LIST_HEAD(&mdev->meta.work.q);
2636 INIT_LIST_HEAD(&mdev->resync_work.list);
2637 INIT_LIST_HEAD(&mdev->unplug_work.list);
2638 INIT_LIST_HEAD(&mdev->md_sync_work.list);
2639 INIT_LIST_HEAD(&mdev->bm_io_work.w.list);
2640 mdev->resync_work.cb = w_resync_inactive;
2641 mdev->unplug_work.cb = w_send_write_hint;
2642 mdev->md_sync_work.cb = w_md_sync;
2643 mdev->bm_io_work.w.cb = w_bitmap_io;
2644 init_timer(&mdev->resync_timer);
2645 init_timer(&mdev->md_sync_timer);
2646 mdev->resync_timer.function = resync_timer_fn;
2647 mdev->resync_timer.data = (unsigned long) mdev;
2648 mdev->md_sync_timer.function = md_sync_timer_fn;
2649 mdev->md_sync_timer.data = (unsigned long) mdev;
2650
2651 init_waitqueue_head(&mdev->misc_wait);
2652 init_waitqueue_head(&mdev->state_wait);
2653 init_waitqueue_head(&mdev->ee_wait);
2654 init_waitqueue_head(&mdev->al_wait);
2655 init_waitqueue_head(&mdev->seq_wait);
2656
2657 drbd_thread_init(mdev, &mdev->receiver, drbdd_init);
2658 drbd_thread_init(mdev, &mdev->worker, drbd_worker);
2659 drbd_thread_init(mdev, &mdev->asender, drbd_asender);
2660
2661 mdev->agreed_pro_version = PRO_VERSION_MAX;
2662 mdev->write_ordering = WO_bio_barrier;
2663 mdev->resync_wenr = LC_FREE;
2664}
2665
2666void drbd_mdev_cleanup(struct drbd_conf *mdev)
2667{
2668 if (mdev->receiver.t_state != None)
2669 dev_err(DEV, "ASSERT FAILED: receiver t_state == %d expected 0.\n",
2670 mdev->receiver.t_state);
2671
2672 /* no need to lock it, I'm the only thread alive */
2673 if (atomic_read(&mdev->current_epoch->epoch_size) != 0)
2674 dev_err(DEV, "epoch_size:%d\n", atomic_read(&mdev->current_epoch->epoch_size));
2675 mdev->al_writ_cnt =
2676 mdev->bm_writ_cnt =
2677 mdev->read_cnt =
2678 mdev->recv_cnt =
2679 mdev->send_cnt =
2680 mdev->writ_cnt =
2681 mdev->p_size =
2682 mdev->rs_start =
2683 mdev->rs_total =
2684 mdev->rs_failed =
2685 mdev->rs_mark_left =
2686 mdev->rs_mark_time = 0;
2687 D_ASSERT(mdev->net_conf == NULL);
2688
2689 drbd_set_my_capacity(mdev, 0);
2690 if (mdev->bitmap) {
2691 /* maybe never allocated. */
2692 drbd_bm_resize(mdev, 0);
2693 drbd_bm_cleanup(mdev);
2694 }
2695
2696 drbd_free_resources(mdev);
2697
2698 /*
2699 * currently we drbd_init_ee only on module load, so
2700 * we may do drbd_release_ee only on module unload!
2701 */
2702 D_ASSERT(list_empty(&mdev->active_ee));
2703 D_ASSERT(list_empty(&mdev->sync_ee));
2704 D_ASSERT(list_empty(&mdev->done_ee));
2705 D_ASSERT(list_empty(&mdev->read_ee));
2706 D_ASSERT(list_empty(&mdev->net_ee));
2707 D_ASSERT(list_empty(&mdev->resync_reads));
2708 D_ASSERT(list_empty(&mdev->data.work.q));
2709 D_ASSERT(list_empty(&mdev->meta.work.q));
2710 D_ASSERT(list_empty(&mdev->resync_work.list));
2711 D_ASSERT(list_empty(&mdev->unplug_work.list));
2712
2713}
2714
2715
2716static void drbd_destroy_mempools(void)
2717{
2718 struct page *page;
2719
2720 while (drbd_pp_pool) {
2721 page = drbd_pp_pool;
2722 drbd_pp_pool = (struct page *)page_private(page);
2723 __free_page(page);
2724 drbd_pp_vacant--;
2725 }
2726
2727 /* D_ASSERT(atomic_read(&drbd_pp_vacant)==0); */
2728
2729 if (drbd_ee_mempool)
2730 mempool_destroy(drbd_ee_mempool);
2731 if (drbd_request_mempool)
2732 mempool_destroy(drbd_request_mempool);
2733 if (drbd_ee_cache)
2734 kmem_cache_destroy(drbd_ee_cache);
2735 if (drbd_request_cache)
2736 kmem_cache_destroy(drbd_request_cache);
2737 if (drbd_bm_ext_cache)
2738 kmem_cache_destroy(drbd_bm_ext_cache);
2739 if (drbd_al_ext_cache)
2740 kmem_cache_destroy(drbd_al_ext_cache);
2741
2742 drbd_ee_mempool = NULL;
2743 drbd_request_mempool = NULL;
2744 drbd_ee_cache = NULL;
2745 drbd_request_cache = NULL;
2746 drbd_bm_ext_cache = NULL;
2747 drbd_al_ext_cache = NULL;
2748
2749 return;
2750}
2751
2752static int drbd_create_mempools(void)
2753{
2754 struct page *page;
2755 const int number = (DRBD_MAX_SEGMENT_SIZE/PAGE_SIZE) * minor_count;
2756 int i;
2757
2758 /* prepare our caches and mempools */
2759 drbd_request_mempool = NULL;
2760 drbd_ee_cache = NULL;
2761 drbd_request_cache = NULL;
2762 drbd_bm_ext_cache = NULL;
2763 drbd_al_ext_cache = NULL;
2764 drbd_pp_pool = NULL;
2765
2766 /* caches */
2767 drbd_request_cache = kmem_cache_create(
2768 "drbd_req", sizeof(struct drbd_request), 0, 0, NULL);
2769 if (drbd_request_cache == NULL)
2770 goto Enomem;
2771
2772 drbd_ee_cache = kmem_cache_create(
2773 "drbd_ee", sizeof(struct drbd_epoch_entry), 0, 0, NULL);
2774 if (drbd_ee_cache == NULL)
2775 goto Enomem;
2776
2777 drbd_bm_ext_cache = kmem_cache_create(
2778 "drbd_bm", sizeof(struct bm_extent), 0, 0, NULL);
2779 if (drbd_bm_ext_cache == NULL)
2780 goto Enomem;
2781
2782 drbd_al_ext_cache = kmem_cache_create(
2783 "drbd_al", sizeof(struct lc_element), 0, 0, NULL);
2784 if (drbd_al_ext_cache == NULL)
2785 goto Enomem;
2786
2787 /* mempools */
2788 drbd_request_mempool = mempool_create(number,
2789 mempool_alloc_slab, mempool_free_slab, drbd_request_cache);
2790 if (drbd_request_mempool == NULL)
2791 goto Enomem;
2792
2793 drbd_ee_mempool = mempool_create(number,
2794 mempool_alloc_slab, mempool_free_slab, drbd_ee_cache);
2795 if (drbd_request_mempool == NULL)
2796 goto Enomem;
2797
2798 /* drbd's page pool */
2799 spin_lock_init(&drbd_pp_lock);
2800
2801 for (i = 0; i < number; i++) {
2802 page = alloc_page(GFP_HIGHUSER);
2803 if (!page)
2804 goto Enomem;
2805 set_page_private(page, (unsigned long)drbd_pp_pool);
2806 drbd_pp_pool = page;
2807 }
2808 drbd_pp_vacant = number;
2809
2810 return 0;
2811
2812Enomem:
2813 drbd_destroy_mempools(); /* in case we allocated some */
2814 return -ENOMEM;
2815}
2816
2817static int drbd_notify_sys(struct notifier_block *this, unsigned long code,
2818 void *unused)
2819{
2820 /* just so we have it. you never know what interesting things we
2821 * might want to do here some day...
2822 */
2823
2824 return NOTIFY_DONE;
2825}
2826
2827static struct notifier_block drbd_notifier = {
2828 .notifier_call = drbd_notify_sys,
2829};
2830
2831static void drbd_release_ee_lists(struct drbd_conf *mdev)
2832{
2833 int rr;
2834
2835 rr = drbd_release_ee(mdev, &mdev->active_ee);
2836 if (rr)
2837 dev_err(DEV, "%d EEs in active list found!\n", rr);
2838
2839 rr = drbd_release_ee(mdev, &mdev->sync_ee);
2840 if (rr)
2841 dev_err(DEV, "%d EEs in sync list found!\n", rr);
2842
2843 rr = drbd_release_ee(mdev, &mdev->read_ee);
2844 if (rr)
2845 dev_err(DEV, "%d EEs in read list found!\n", rr);
2846
2847 rr = drbd_release_ee(mdev, &mdev->done_ee);
2848 if (rr)
2849 dev_err(DEV, "%d EEs in done list found!\n", rr);
2850
2851 rr = drbd_release_ee(mdev, &mdev->net_ee);
2852 if (rr)
2853 dev_err(DEV, "%d EEs in net list found!\n", rr);
2854}
2855
2856/* caution. no locking.
2857 * currently only used from module cleanup code. */
2858static void drbd_delete_device(unsigned int minor)
2859{
2860 struct drbd_conf *mdev = minor_to_mdev(minor);
2861
2862 if (!mdev)
2863 return;
2864
2865 /* paranoia asserts */
2866 if (mdev->open_cnt != 0)
2867 dev_err(DEV, "open_cnt = %d in %s:%u", mdev->open_cnt,
2868 __FILE__ , __LINE__);
2869
2870 ERR_IF (!list_empty(&mdev->data.work.q)) {
2871 struct list_head *lp;
2872 list_for_each(lp, &mdev->data.work.q) {
2873 dev_err(DEV, "lp = %p\n", lp);
2874 }
2875 };
2876 /* end paranoia asserts */
2877
2878 del_gendisk(mdev->vdisk);
2879
2880 /* cleanup stuff that may have been allocated during
2881 * device (re-)configuration or state changes */
2882
2883 if (mdev->this_bdev)
2884 bdput(mdev->this_bdev);
2885
2886 drbd_free_resources(mdev);
2887
2888 drbd_release_ee_lists(mdev);
2889
2890 /* should be free'd on disconnect? */
2891 kfree(mdev->ee_hash);
2892 /*
2893 mdev->ee_hash_s = 0;
2894 mdev->ee_hash = NULL;
2895 */
2896
2897 lc_destroy(mdev->act_log);
2898 lc_destroy(mdev->resync);
2899
2900 kfree(mdev->p_uuid);
2901 /* mdev->p_uuid = NULL; */
2902
2903 kfree(mdev->int_dig_out);
2904 kfree(mdev->int_dig_in);
2905 kfree(mdev->int_dig_vv);
2906
2907 /* cleanup the rest that has been
2908 * allocated from drbd_new_device
2909 * and actually free the mdev itself */
2910 drbd_free_mdev(mdev);
2911}
2912
2913static void drbd_cleanup(void)
2914{
2915 unsigned int i;
2916
2917 unregister_reboot_notifier(&drbd_notifier);
2918
2919 drbd_nl_cleanup();
2920
2921 if (minor_table) {
2922 if (drbd_proc)
2923 remove_proc_entry("drbd", NULL);
2924 i = minor_count;
2925 while (i--)
2926 drbd_delete_device(i);
2927 drbd_destroy_mempools();
2928 }
2929
2930 kfree(minor_table);
2931
2932 unregister_blkdev(DRBD_MAJOR, "drbd");
2933
2934 printk(KERN_INFO "drbd: module cleanup done.\n");
2935}
2936
2937/**
2938 * drbd_congested() - Callback for pdflush
2939 * @congested_data: User data
2940 * @bdi_bits: Bits pdflush is currently interested in
2941 *
2942 * Returns 1<<BDI_async_congested and/or 1<<BDI_sync_congested if we are congested.
2943 */
2944static int drbd_congested(void *congested_data, int bdi_bits)
2945{
2946 struct drbd_conf *mdev = congested_data;
2947 struct request_queue *q;
2948 char reason = '-';
2949 int r = 0;
2950
2951 if (!__inc_ap_bio_cond(mdev)) {
2952 /* DRBD has frozen IO */
2953 r = bdi_bits;
2954 reason = 'd';
2955 goto out;
2956 }
2957
2958 if (get_ldev(mdev)) {
2959 q = bdev_get_queue(mdev->ldev->backing_bdev);
2960 r = bdi_congested(&q->backing_dev_info, bdi_bits);
2961 put_ldev(mdev);
2962 if (r)
2963 reason = 'b';
2964 }
2965
2966 if (bdi_bits & (1 << BDI_async_congested) && test_bit(NET_CONGESTED, &mdev->flags)) {
2967 r |= (1 << BDI_async_congested);
2968 reason = reason == 'b' ? 'a' : 'n';
2969 }
2970
2971out:
2972 mdev->congestion_reason = reason;
2973 return r;
2974}
2975
2976struct drbd_conf *drbd_new_device(unsigned int minor)
2977{
2978 struct drbd_conf *mdev;
2979 struct gendisk *disk;
2980 struct request_queue *q;
2981
2982 /* GFP_KERNEL, we are outside of all write-out paths */
2983 mdev = kzalloc(sizeof(struct drbd_conf), GFP_KERNEL);
2984 if (!mdev)
2985 return NULL;
2986 if (!zalloc_cpumask_var(&mdev->cpu_mask, GFP_KERNEL))
2987 goto out_no_cpumask;
2988
2989 mdev->minor = minor;
2990
2991 drbd_init_set_defaults(mdev);
2992
2993 q = blk_alloc_queue(GFP_KERNEL);
2994 if (!q)
2995 goto out_no_q;
2996 mdev->rq_queue = q;
2997 q->queuedata = mdev;
2998 blk_queue_max_segment_size(q, DRBD_MAX_SEGMENT_SIZE);
2999
3000 disk = alloc_disk(1);
3001 if (!disk)
3002 goto out_no_disk;
3003 mdev->vdisk = disk;
3004
3005 set_disk_ro(disk, TRUE);
3006
3007 disk->queue = q;
3008 disk->major = DRBD_MAJOR;
3009 disk->first_minor = minor;
3010 disk->fops = &drbd_ops;
3011 sprintf(disk->disk_name, "drbd%d", minor);
3012 disk->private_data = mdev;
3013
3014 mdev->this_bdev = bdget(MKDEV(DRBD_MAJOR, minor));
3015 /* we have no partitions. we contain only ourselves. */
3016 mdev->this_bdev->bd_contains = mdev->this_bdev;
3017
3018 q->backing_dev_info.congested_fn = drbd_congested;
3019 q->backing_dev_info.congested_data = mdev;
3020
3021 blk_queue_make_request(q, drbd_make_request_26);
3022 blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
3023 blk_queue_merge_bvec(q, drbd_merge_bvec);
3024 q->queue_lock = &mdev->req_lock; /* needed since we use */
3025 /* plugging on a queue, that actually has no requests! */
3026 q->unplug_fn = drbd_unplug_fn;
3027
3028 mdev->md_io_page = alloc_page(GFP_KERNEL);
3029 if (!mdev->md_io_page)
3030 goto out_no_io_page;
3031
3032 if (drbd_bm_init(mdev))
3033 goto out_no_bitmap;
3034 /* no need to lock access, we are still initializing this minor device. */
3035 if (!tl_init(mdev))
3036 goto out_no_tl;
3037
3038 mdev->app_reads_hash = kzalloc(APP_R_HSIZE*sizeof(void *), GFP_KERNEL);
3039 if (!mdev->app_reads_hash)
3040 goto out_no_app_reads;
3041
3042 mdev->current_epoch = kzalloc(sizeof(struct drbd_epoch), GFP_KERNEL);
3043 if (!mdev->current_epoch)
3044 goto out_no_epoch;
3045
3046 INIT_LIST_HEAD(&mdev->current_epoch->list);
3047 mdev->epochs = 1;
3048
3049 return mdev;
3050
3051/* out_whatever_else:
3052 kfree(mdev->current_epoch); */
3053out_no_epoch:
3054 kfree(mdev->app_reads_hash);
3055out_no_app_reads:
3056 tl_cleanup(mdev);
3057out_no_tl:
3058 drbd_bm_cleanup(mdev);
3059out_no_bitmap:
3060 __free_page(mdev->md_io_page);
3061out_no_io_page:
3062 put_disk(disk);
3063out_no_disk:
3064 blk_cleanup_queue(q);
3065out_no_q:
3066 free_cpumask_var(mdev->cpu_mask);
3067out_no_cpumask:
3068 kfree(mdev);
3069 return NULL;
3070}
3071
3072/* counterpart of drbd_new_device.
3073 * last part of drbd_delete_device. */
3074void drbd_free_mdev(struct drbd_conf *mdev)
3075{
3076 kfree(mdev->current_epoch);
3077 kfree(mdev->app_reads_hash);
3078 tl_cleanup(mdev);
3079 if (mdev->bitmap) /* should no longer be there. */
3080 drbd_bm_cleanup(mdev);
3081 __free_page(mdev->md_io_page);
3082 put_disk(mdev->vdisk);
3083 blk_cleanup_queue(mdev->rq_queue);
3084 free_cpumask_var(mdev->cpu_mask);
3085 kfree(mdev);
3086}
3087
3088
3089int __init drbd_init(void)
3090{
3091 int err;
3092
3093 if (sizeof(struct p_handshake) != 80) {
3094 printk(KERN_ERR
3095 "drbd: never change the size or layout "
3096 "of the HandShake packet.\n");
3097 return -EINVAL;
3098 }
3099
3100 if (1 > minor_count || minor_count > 255) {
3101 printk(KERN_ERR
3102 "drbd: invalid minor_count (%d)\n", minor_count);
3103#ifdef MODULE
3104 return -EINVAL;
3105#else
3106 minor_count = 8;
3107#endif
3108 }
3109
3110 err = drbd_nl_init();
3111 if (err)
3112 return err;
3113
3114 err = register_blkdev(DRBD_MAJOR, "drbd");
3115 if (err) {
3116 printk(KERN_ERR
3117 "drbd: unable to register block device major %d\n",
3118 DRBD_MAJOR);
3119 return err;
3120 }
3121
3122 register_reboot_notifier(&drbd_notifier);
3123
3124 /*
3125 * allocate all necessary structs
3126 */
3127 err = -ENOMEM;
3128
3129 init_waitqueue_head(&drbd_pp_wait);
3130
3131 drbd_proc = NULL; /* play safe for drbd_cleanup */
3132 minor_table = kzalloc(sizeof(struct drbd_conf *)*minor_count,
3133 GFP_KERNEL);
3134 if (!minor_table)
3135 goto Enomem;
3136
3137 err = drbd_create_mempools();
3138 if (err)
3139 goto Enomem;
3140
3141 drbd_proc = proc_create("drbd", S_IFREG | S_IRUGO , NULL, &drbd_proc_fops);
3142 if (!drbd_proc) {
3143 printk(KERN_ERR "drbd: unable to register proc file\n");
3144 goto Enomem;
3145 }
3146
3147 rwlock_init(&global_state_lock);
3148
3149 printk(KERN_INFO "drbd: initialized. "
3150 "Version: " REL_VERSION " (api:%d/proto:%d-%d)\n",
3151 API_VERSION, PRO_VERSION_MIN, PRO_VERSION_MAX);
3152 printk(KERN_INFO "drbd: %s\n", drbd_buildtag());
3153 printk(KERN_INFO "drbd: registered as block device major %d\n",
3154 DRBD_MAJOR);
3155 printk(KERN_INFO "drbd: minor_table @ 0x%p\n", minor_table);
3156
3157 return 0; /* Success! */
3158
3159Enomem:
3160 drbd_cleanup();
3161 if (err == -ENOMEM)
3162 /* currently always the case */
3163 printk(KERN_ERR "drbd: ran out of memory\n");
3164 else
3165 printk(KERN_ERR "drbd: initialization failure\n");
3166 return err;
3167}
3168
3169void drbd_free_bc(struct drbd_backing_dev *ldev)
3170{
3171 if (ldev == NULL)
3172 return;
3173
3174 bd_release(ldev->backing_bdev);
3175 bd_release(ldev->md_bdev);
3176
3177 fput(ldev->lo_file);
3178 fput(ldev->md_file);
3179
3180 kfree(ldev);
3181}
3182
3183void drbd_free_sock(struct drbd_conf *mdev)
3184{
3185 if (mdev->data.socket) {
3186 kernel_sock_shutdown(mdev->data.socket, SHUT_RDWR);
3187 sock_release(mdev->data.socket);
3188 mdev->data.socket = NULL;
3189 }
3190 if (mdev->meta.socket) {
3191 kernel_sock_shutdown(mdev->meta.socket, SHUT_RDWR);
3192 sock_release(mdev->meta.socket);
3193 mdev->meta.socket = NULL;
3194 }
3195}
3196
3197
3198void drbd_free_resources(struct drbd_conf *mdev)
3199{
3200 crypto_free_hash(mdev->csums_tfm);
3201 mdev->csums_tfm = NULL;
3202 crypto_free_hash(mdev->verify_tfm);
3203 mdev->verify_tfm = NULL;
3204 crypto_free_hash(mdev->cram_hmac_tfm);
3205 mdev->cram_hmac_tfm = NULL;
3206 crypto_free_hash(mdev->integrity_w_tfm);
3207 mdev->integrity_w_tfm = NULL;
3208 crypto_free_hash(mdev->integrity_r_tfm);
3209 mdev->integrity_r_tfm = NULL;
3210
3211 drbd_free_sock(mdev);
3212
3213 __no_warn(local,
3214 drbd_free_bc(mdev->ldev);
3215 mdev->ldev = NULL;);
3216}
3217
3218/* meta data management */
3219
3220struct meta_data_on_disk {
3221 u64 la_size; /* last agreed size. */
3222 u64 uuid[UI_SIZE]; /* UUIDs. */
3223 u64 device_uuid;
3224 u64 reserved_u64_1;
3225 u32 flags; /* MDF */
3226 u32 magic;
3227 u32 md_size_sect;
3228 u32 al_offset; /* offset to this block */
3229 u32 al_nr_extents; /* important for restoring the AL */
3230 /* `-- act_log->nr_elements <-- sync_conf.al_extents */
3231 u32 bm_offset; /* offset to the bitmap, from here */
3232 u32 bm_bytes_per_bit; /* BM_BLOCK_SIZE */
3233 u32 reserved_u32[4];
3234
3235} __packed;
3236
3237/**
3238 * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set
3239 * @mdev: DRBD device.
3240 */
3241void drbd_md_sync(struct drbd_conf *mdev)
3242{
3243 struct meta_data_on_disk *buffer;
3244 sector_t sector;
3245 int i;
3246
3247 if (!test_and_clear_bit(MD_DIRTY, &mdev->flags))
3248 return;
3249 del_timer(&mdev->md_sync_timer);
3250
3251 /* We use here D_FAILED and not D_ATTACHING because we try to write
3252 * metadata even if we detach due to a disk failure! */
3253 if (!get_ldev_if_state(mdev, D_FAILED))
3254 return;
3255
3256 trace_drbd_md_io(mdev, WRITE, mdev->ldev);
3257
3258 mutex_lock(&mdev->md_io_mutex);
3259 buffer = (struct meta_data_on_disk *)page_address(mdev->md_io_page);
3260 memset(buffer, 0, 512);
3261
3262 buffer->la_size = cpu_to_be64(drbd_get_capacity(mdev->this_bdev));
3263 for (i = UI_CURRENT; i < UI_SIZE; i++)
3264 buffer->uuid[i] = cpu_to_be64(mdev->ldev->md.uuid[i]);
3265 buffer->flags = cpu_to_be32(mdev->ldev->md.flags);
3266 buffer->magic = cpu_to_be32(DRBD_MD_MAGIC);
3267
3268 buffer->md_size_sect = cpu_to_be32(mdev->ldev->md.md_size_sect);
3269 buffer->al_offset = cpu_to_be32(mdev->ldev->md.al_offset);
3270 buffer->al_nr_extents = cpu_to_be32(mdev->act_log->nr_elements);
3271 buffer->bm_bytes_per_bit = cpu_to_be32(BM_BLOCK_SIZE);
3272 buffer->device_uuid = cpu_to_be64(mdev->ldev->md.device_uuid);
3273
3274 buffer->bm_offset = cpu_to_be32(mdev->ldev->md.bm_offset);
3275
3276 D_ASSERT(drbd_md_ss__(mdev, mdev->ldev) == mdev->ldev->md.md_offset);
3277 sector = mdev->ldev->md.md_offset;
3278
3279 if (drbd_md_sync_page_io(mdev, mdev->ldev, sector, WRITE)) {
3280 clear_bit(MD_DIRTY, &mdev->flags);
3281 } else {
3282 /* this was a try anyways ... */
3283 dev_err(DEV, "meta data update failed!\n");
3284
3285 drbd_chk_io_error(mdev, 1, TRUE);
3286 }
3287
3288 /* Update mdev->ldev->md.la_size_sect,
3289 * since we updated it on metadata. */
3290 mdev->ldev->md.la_size_sect = drbd_get_capacity(mdev->this_bdev);
3291
3292 mutex_unlock(&mdev->md_io_mutex);
3293 put_ldev(mdev);
3294}
3295
3296/**
3297 * drbd_md_read() - Reads in the meta data super block
3298 * @mdev: DRBD device.
3299 * @bdev: Device from which the meta data should be read in.
3300 *
3301 * Return 0 (NO_ERROR) on success, and an enum drbd_ret_codes in case
3302 * something goes wrong. Currently only: ERR_IO_MD_DISK, ERR_MD_INVALID.
3303 */
3304int drbd_md_read(struct drbd_conf *mdev, struct drbd_backing_dev *bdev)
3305{
3306 struct meta_data_on_disk *buffer;
3307 int i, rv = NO_ERROR;
3308
3309 if (!get_ldev_if_state(mdev, D_ATTACHING))
3310 return ERR_IO_MD_DISK;
3311
3312 trace_drbd_md_io(mdev, READ, bdev);
3313
3314 mutex_lock(&mdev->md_io_mutex);
3315 buffer = (struct meta_data_on_disk *)page_address(mdev->md_io_page);
3316
3317 if (!drbd_md_sync_page_io(mdev, bdev, bdev->md.md_offset, READ)) {
3318 /* NOTE: cant do normal error processing here as this is
3319 called BEFORE disk is attached */
3320 dev_err(DEV, "Error while reading metadata.\n");
3321 rv = ERR_IO_MD_DISK;
3322 goto err;
3323 }
3324
3325 if (be32_to_cpu(buffer->magic) != DRBD_MD_MAGIC) {
3326 dev_err(DEV, "Error while reading metadata, magic not found.\n");
3327 rv = ERR_MD_INVALID;
3328 goto err;
3329 }
3330 if (be32_to_cpu(buffer->al_offset) != bdev->md.al_offset) {
3331 dev_err(DEV, "unexpected al_offset: %d (expected %d)\n",
3332 be32_to_cpu(buffer->al_offset), bdev->md.al_offset);
3333 rv = ERR_MD_INVALID;
3334 goto err;
3335 }
3336 if (be32_to_cpu(buffer->bm_offset) != bdev->md.bm_offset) {
3337 dev_err(DEV, "unexpected bm_offset: %d (expected %d)\n",
3338 be32_to_cpu(buffer->bm_offset), bdev->md.bm_offset);
3339 rv = ERR_MD_INVALID;
3340 goto err;
3341 }
3342 if (be32_to_cpu(buffer->md_size_sect) != bdev->md.md_size_sect) {
3343 dev_err(DEV, "unexpected md_size: %u (expected %u)\n",
3344 be32_to_cpu(buffer->md_size_sect), bdev->md.md_size_sect);
3345 rv = ERR_MD_INVALID;
3346 goto err;
3347 }
3348
3349 if (be32_to_cpu(buffer->bm_bytes_per_bit) != BM_BLOCK_SIZE) {
3350 dev_err(DEV, "unexpected bm_bytes_per_bit: %u (expected %u)\n",
3351 be32_to_cpu(buffer->bm_bytes_per_bit), BM_BLOCK_SIZE);
3352 rv = ERR_MD_INVALID;
3353 goto err;
3354 }
3355
3356 bdev->md.la_size_sect = be64_to_cpu(buffer->la_size);
3357 for (i = UI_CURRENT; i < UI_SIZE; i++)
3358 bdev->md.uuid[i] = be64_to_cpu(buffer->uuid[i]);
3359 bdev->md.flags = be32_to_cpu(buffer->flags);
3360 mdev->sync_conf.al_extents = be32_to_cpu(buffer->al_nr_extents);
3361 bdev->md.device_uuid = be64_to_cpu(buffer->device_uuid);
3362
3363 if (mdev->sync_conf.al_extents < 7)
3364 mdev->sync_conf.al_extents = 127;
3365
3366 err:
3367 mutex_unlock(&mdev->md_io_mutex);
3368 put_ldev(mdev);
3369
3370 return rv;
3371}
3372
3373/**
3374 * drbd_md_mark_dirty() - Mark meta data super block as dirty
3375 * @mdev: DRBD device.
3376 *
3377 * Call this function if you change anything that should be written to
3378 * the meta-data super block. This function sets MD_DIRTY, and starts a
3379 * timer that ensures that within five seconds you have to call drbd_md_sync().
3380 */
3381void drbd_md_mark_dirty(struct drbd_conf *mdev)
3382{
3383 set_bit(MD_DIRTY, &mdev->flags);
3384 mod_timer(&mdev->md_sync_timer, jiffies + 5*HZ);
3385}
3386
3387
3388static void drbd_uuid_move_history(struct drbd_conf *mdev) __must_hold(local)
3389{
3390 int i;
3391
3392 for (i = UI_HISTORY_START; i < UI_HISTORY_END; i++) {
3393 mdev->ldev->md.uuid[i+1] = mdev->ldev->md.uuid[i];
3394
3395 trace_drbd_uuid(mdev, i+1);
3396 }
3397}
3398
3399void _drbd_uuid_set(struct drbd_conf *mdev, int idx, u64 val) __must_hold(local)
3400{
3401 if (idx == UI_CURRENT) {
3402 if (mdev->state.role == R_PRIMARY)
3403 val |= 1;
3404 else
3405 val &= ~((u64)1);
3406
3407 drbd_set_ed_uuid(mdev, val);
3408 }
3409
3410 mdev->ldev->md.uuid[idx] = val;
3411 trace_drbd_uuid(mdev, idx);
3412 drbd_md_mark_dirty(mdev);
3413}
3414
3415
3416void drbd_uuid_set(struct drbd_conf *mdev, int idx, u64 val) __must_hold(local)
3417{
3418 if (mdev->ldev->md.uuid[idx]) {
3419 drbd_uuid_move_history(mdev);
3420 mdev->ldev->md.uuid[UI_HISTORY_START] = mdev->ldev->md.uuid[idx];
3421 trace_drbd_uuid(mdev, UI_HISTORY_START);
3422 }
3423 _drbd_uuid_set(mdev, idx, val);
3424}
3425
3426/**
3427 * drbd_uuid_new_current() - Creates a new current UUID
3428 * @mdev: DRBD device.
3429 *
3430 * Creates a new current UUID, and rotates the old current UUID into
3431 * the bitmap slot. Causes an incremental resync upon next connect.
3432 */
3433void drbd_uuid_new_current(struct drbd_conf *mdev) __must_hold(local)
3434{
3435 u64 val;
3436
3437 dev_info(DEV, "Creating new current UUID\n");
3438 D_ASSERT(mdev->ldev->md.uuid[UI_BITMAP] == 0);
3439 mdev->ldev->md.uuid[UI_BITMAP] = mdev->ldev->md.uuid[UI_CURRENT];
3440 trace_drbd_uuid(mdev, UI_BITMAP);
3441
3442 get_random_bytes(&val, sizeof(u64));
3443 _drbd_uuid_set(mdev, UI_CURRENT, val);
3444}
3445
3446void drbd_uuid_set_bm(struct drbd_conf *mdev, u64 val) __must_hold(local)
3447{
3448 if (mdev->ldev->md.uuid[UI_BITMAP] == 0 && val == 0)
3449 return;
3450
3451 if (val == 0) {
3452 drbd_uuid_move_history(mdev);
3453 mdev->ldev->md.uuid[UI_HISTORY_START] = mdev->ldev->md.uuid[UI_BITMAP];
3454 mdev->ldev->md.uuid[UI_BITMAP] = 0;
3455 trace_drbd_uuid(mdev, UI_HISTORY_START);
3456 trace_drbd_uuid(mdev, UI_BITMAP);
3457 } else {
3458 if (mdev->ldev->md.uuid[UI_BITMAP])
3459 dev_warn(DEV, "bm UUID already set");
3460
3461 mdev->ldev->md.uuid[UI_BITMAP] = val;
3462 mdev->ldev->md.uuid[UI_BITMAP] &= ~((u64)1);
3463
3464 trace_drbd_uuid(mdev, UI_BITMAP);
3465 }
3466 drbd_md_mark_dirty(mdev);
3467}
3468
3469/**
3470 * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3471 * @mdev: DRBD device.
3472 *
3473 * Sets all bits in the bitmap and writes the whole bitmap to stable storage.
3474 */
3475int drbd_bmio_set_n_write(struct drbd_conf *mdev)
3476{
3477 int rv = -EIO;
3478
3479 if (get_ldev_if_state(mdev, D_ATTACHING)) {
3480 drbd_md_set_flag(mdev, MDF_FULL_SYNC);
3481 drbd_md_sync(mdev);
3482 drbd_bm_set_all(mdev);
3483
3484 rv = drbd_bm_write(mdev);
3485
3486 if (!rv) {
3487 drbd_md_clear_flag(mdev, MDF_FULL_SYNC);
3488 drbd_md_sync(mdev);
3489 }
3490
3491 put_ldev(mdev);
3492 }
3493
3494 return rv;
3495}
3496
3497/**
3498 * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3499 * @mdev: DRBD device.
3500 *
3501 * Clears all bits in the bitmap and writes the whole bitmap to stable storage.
3502 */
3503int drbd_bmio_clear_n_write(struct drbd_conf *mdev)
3504{
3505 int rv = -EIO;
3506
3507 if (get_ldev_if_state(mdev, D_ATTACHING)) {
3508 drbd_bm_clear_all(mdev);
3509 rv = drbd_bm_write(mdev);
3510 put_ldev(mdev);
3511 }
3512
3513 return rv;
3514}
3515
3516static int w_bitmap_io(struct drbd_conf *mdev, struct drbd_work *w, int unused)
3517{
3518 struct bm_io_work *work = container_of(w, struct bm_io_work, w);
3519 int rv;
3520
3521 D_ASSERT(atomic_read(&mdev->ap_bio_cnt) == 0);
3522
3523 drbd_bm_lock(mdev, work->why);
3524 rv = work->io_fn(mdev);
3525 drbd_bm_unlock(mdev);
3526
3527 clear_bit(BITMAP_IO, &mdev->flags);
3528 wake_up(&mdev->misc_wait);
3529
3530 if (work->done)
3531 work->done(mdev, rv);
3532
3533 clear_bit(BITMAP_IO_QUEUED, &mdev->flags);
3534 work->why = NULL;
3535
3536 return 1;
3537}
3538
3539/**
3540 * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap
3541 * @mdev: DRBD device.
3542 * @io_fn: IO callback to be called when bitmap IO is possible
3543 * @done: callback to be called after the bitmap IO was performed
3544 * @why: Descriptive text of the reason for doing the IO
3545 *
3546 * While IO on the bitmap happens we freeze application IO thus we ensure
3547 * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be
3548 * called from worker context. It MUST NOT be used while a previous such
3549 * work is still pending!
3550 */
3551void drbd_queue_bitmap_io(struct drbd_conf *mdev,
3552 int (*io_fn)(struct drbd_conf *),
3553 void (*done)(struct drbd_conf *, int),
3554 char *why)
3555{
3556 D_ASSERT(current == mdev->worker.task);
3557
3558 D_ASSERT(!test_bit(BITMAP_IO_QUEUED, &mdev->flags));
3559 D_ASSERT(!test_bit(BITMAP_IO, &mdev->flags));
3560 D_ASSERT(list_empty(&mdev->bm_io_work.w.list));
3561 if (mdev->bm_io_work.why)
3562 dev_err(DEV, "FIXME going to queue '%s' but '%s' still pending?\n",
3563 why, mdev->bm_io_work.why);
3564
3565 mdev->bm_io_work.io_fn = io_fn;
3566 mdev->bm_io_work.done = done;
3567 mdev->bm_io_work.why = why;
3568
3569 set_bit(BITMAP_IO, &mdev->flags);
3570 if (atomic_read(&mdev->ap_bio_cnt) == 0) {
3571 if (list_empty(&mdev->bm_io_work.w.list)) {
3572 set_bit(BITMAP_IO_QUEUED, &mdev->flags);
3573 drbd_queue_work(&mdev->data.work, &mdev->bm_io_work.w);
3574 } else
3575 dev_err(DEV, "FIXME avoided double queuing bm_io_work\n");
3576 }
3577}
3578
3579/**
3580 * drbd_bitmap_io() - Does an IO operation on the whole bitmap
3581 * @mdev: DRBD device.
3582 * @io_fn: IO callback to be called when bitmap IO is possible
3583 * @why: Descriptive text of the reason for doing the IO
3584 *
3585 * freezes application IO while that the actual IO operations runs. This
3586 * functions MAY NOT be called from worker context.
3587 */
3588int drbd_bitmap_io(struct drbd_conf *mdev, int (*io_fn)(struct drbd_conf *), char *why)
3589{
3590 int rv;
3591
3592 D_ASSERT(current != mdev->worker.task);
3593
3594 drbd_suspend_io(mdev);
3595
3596 drbd_bm_lock(mdev, why);
3597 rv = io_fn(mdev);
3598 drbd_bm_unlock(mdev);
3599
3600 drbd_resume_io(mdev);
3601
3602 return rv;
3603}
3604
3605void drbd_md_set_flag(struct drbd_conf *mdev, int flag) __must_hold(local)
3606{
3607 if ((mdev->ldev->md.flags & flag) != flag) {
3608 drbd_md_mark_dirty(mdev);
3609 mdev->ldev->md.flags |= flag;
3610 }
3611}
3612
3613void drbd_md_clear_flag(struct drbd_conf *mdev, int flag) __must_hold(local)
3614{
3615 if ((mdev->ldev->md.flags & flag) != 0) {
3616 drbd_md_mark_dirty(mdev);
3617 mdev->ldev->md.flags &= ~flag;
3618 }
3619}
3620int drbd_md_test_flag(struct drbd_backing_dev *bdev, int flag)
3621{
3622 return (bdev->md.flags & flag) != 0;
3623}
3624
3625static void md_sync_timer_fn(unsigned long data)
3626{
3627 struct drbd_conf *mdev = (struct drbd_conf *) data;
3628
3629 drbd_queue_work_front(&mdev->data.work, &mdev->md_sync_work);
3630}
3631
3632static int w_md_sync(struct drbd_conf *mdev, struct drbd_work *w, int unused)
3633{
3634 dev_warn(DEV, "md_sync_timer expired! Worker calls drbd_md_sync().\n");
3635 drbd_md_sync(mdev);
3636
3637 return 1;
3638}
3639
3640#ifdef CONFIG_DRBD_FAULT_INJECTION
3641/* Fault insertion support including random number generator shamelessly
3642 * stolen from kernel/rcutorture.c */
3643struct fault_random_state {
3644 unsigned long state;
3645 unsigned long count;
3646};
3647
3648#define FAULT_RANDOM_MULT 39916801 /* prime */
3649#define FAULT_RANDOM_ADD 479001701 /* prime */
3650#define FAULT_RANDOM_REFRESH 10000
3651
3652/*
3653 * Crude but fast random-number generator. Uses a linear congruential
3654 * generator, with occasional help from get_random_bytes().
3655 */
3656static unsigned long
3657_drbd_fault_random(struct fault_random_state *rsp)
3658{
3659 long refresh;
3660
3661 if (--rsp->count < 0) {
3662 get_random_bytes(&refresh, sizeof(refresh));
3663 rsp->state += refresh;
3664 rsp->count = FAULT_RANDOM_REFRESH;
3665 }
3666 rsp->state = rsp->state * FAULT_RANDOM_MULT + FAULT_RANDOM_ADD;
3667 return swahw32(rsp->state);
3668}
3669
3670static char *
3671_drbd_fault_str(unsigned int type) {
3672 static char *_faults[] = {
3673 [DRBD_FAULT_MD_WR] = "Meta-data write",
3674 [DRBD_FAULT_MD_RD] = "Meta-data read",
3675 [DRBD_FAULT_RS_WR] = "Resync write",
3676 [DRBD_FAULT_RS_RD] = "Resync read",
3677 [DRBD_FAULT_DT_WR] = "Data write",
3678 [DRBD_FAULT_DT_RD] = "Data read",
3679 [DRBD_FAULT_DT_RA] = "Data read ahead",
3680 [DRBD_FAULT_BM_ALLOC] = "BM allocation",
3681 [DRBD_FAULT_AL_EE] = "EE allocation"
3682 };
3683
3684 return (type < DRBD_FAULT_MAX) ? _faults[type] : "**Unknown**";
3685}
3686
3687unsigned int
3688_drbd_insert_fault(struct drbd_conf *mdev, unsigned int type)
3689{
3690 static struct fault_random_state rrs = {0, 0};
3691
3692 unsigned int ret = (
3693 (fault_devs == 0 ||
3694 ((1 << mdev_to_minor(mdev)) & fault_devs) != 0) &&
3695 (((_drbd_fault_random(&rrs) % 100) + 1) <= fault_rate));
3696
3697 if (ret) {
3698 fault_count++;
3699
3700 if (printk_ratelimit())
3701 dev_warn(DEV, "***Simulating %s failure\n",
3702 _drbd_fault_str(type));
3703 }
3704
3705 return ret;
3706}
3707#endif
3708
3709const char *drbd_buildtag(void)
3710{
3711 /* DRBD built from external sources has here a reference to the
3712 git hash of the source code. */
3713
3714 static char buildtag[38] = "\0uilt-in";
3715
3716 if (buildtag[0] == 0) {
3717#ifdef CONFIG_MODULES
3718 if (THIS_MODULE != NULL)
3719 sprintf(buildtag, "srcversion: %-24s", THIS_MODULE->srcversion);
3720 else
3721#endif
3722 buildtag[0] = 'b';
3723 }
3724
3725 return buildtag;
3726}
3727
3728module_init(drbd_init)
3729module_exit(drbd_cleanup)
3730
3731/* For drbd_tracing: */
3732EXPORT_SYMBOL(drbd_conn_str);
3733EXPORT_SYMBOL(drbd_role_str);
3734EXPORT_SYMBOL(drbd_disk_str);
3735EXPORT_SYMBOL(drbd_set_st_err_str);
Linux 6.x block layer and io_uring tree(s)