Commit | Line | Data |
---|---|---|
c6ae4c04 | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
b411b363 PR |
2 | /* |
3 | drbd.c | |
4 | ||
5 | This file is part of DRBD by Philipp Reisner and Lars Ellenberg. | |
6 | ||
7 | Copyright (C) 2001-2008, LINBIT Information Technologies GmbH. | |
8 | Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>. | |
9 | Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>. | |
10 | ||
11 | Thanks to Carter Burden, Bart Grantham and Gennadiy Nerubayev | |
12 | from Logicworks, Inc. for making SDP replication support possible. | |
13 | ||
b411b363 PR |
14 | |
15 | */ | |
16 | ||
f88c5d90 LE |
17 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
18 | ||
b411b363 | 19 | #include <linux/module.h> |
e5f891b2 | 20 | #include <linux/jiffies.h> |
b411b363 | 21 | #include <linux/drbd.h> |
7e5fec31 | 22 | #include <linux/uaccess.h> |
b411b363 PR |
23 | #include <asm/types.h> |
24 | #include <net/sock.h> | |
25 | #include <linux/ctype.h> | |
2a48fc0a | 26 | #include <linux/mutex.h> |
b411b363 PR |
27 | #include <linux/fs.h> |
28 | #include <linux/file.h> | |
29 | #include <linux/proc_fs.h> | |
30 | #include <linux/init.h> | |
31 | #include <linux/mm.h> | |
32 | #include <linux/memcontrol.h> | |
33 | #include <linux/mm_inline.h> | |
34 | #include <linux/slab.h> | |
35 | #include <linux/random.h> | |
36 | #include <linux/reboot.h> | |
37 | #include <linux/notifier.h> | |
38 | #include <linux/kthread.h> | |
113fef9e | 39 | #include <linux/workqueue.h> |
b411b363 PR |
40 | #define __KERNEL_SYSCALLS__ |
41 | #include <linux/unistd.h> | |
42 | #include <linux/vmalloc.h> | |
174cd4b1 | 43 | #include <linux/sched/signal.h> |
b411b363 PR |
44 | |
45 | #include <linux/drbd_limits.h> | |
46 | #include "drbd_int.h" | |
a3603a6e | 47 | #include "drbd_protocol.h" |
b411b363 | 48 | #include "drbd_req.h" /* only for _req_mod in tl_release and tl_clear */ |
b411b363 | 49 | #include "drbd_vli.h" |
4d3d5aa8 | 50 | #include "drbd_debugfs.h" |
b411b363 | 51 | |
2a48fc0a | 52 | static DEFINE_MUTEX(drbd_main_mutex); |
b411b363 | 53 | static int drbd_open(struct block_device *bdev, fmode_t mode); |
db2a144b | 54 | static void drbd_release(struct gendisk *gd, fmode_t mode); |
2bccef39 | 55 | static void md_sync_timer_fn(struct timer_list *t); |
99920dc5 | 56 | static int w_bitmap_io(struct drbd_work *w, int unused); |
b411b363 | 57 | |
b411b363 PR |
58 | MODULE_AUTHOR("Philipp Reisner <phil@linbit.com>, " |
59 | "Lars Ellenberg <lars@linbit.com>"); | |
60 | MODULE_DESCRIPTION("drbd - Distributed Replicated Block Device v" REL_VERSION); | |
61 | MODULE_VERSION(REL_VERSION); | |
62 | MODULE_LICENSE("GPL"); | |
81a5d60e | 63 | MODULE_PARM_DESC(minor_count, "Approximate number of drbd devices (" |
2b8a90b5 | 64 | __stringify(DRBD_MINOR_COUNT_MIN) "-" __stringify(DRBD_MINOR_COUNT_MAX) ")"); |
b411b363 PR |
65 | MODULE_ALIAS_BLOCKDEV_MAJOR(DRBD_MAJOR); |
66 | ||
67 | #include <linux/moduleparam.h> | |
b411b363 | 68 | /* thanks to these macros, if compiled into the kernel (not-module), |
183ece30 | 69 | * these become boot parameters (e.g., drbd.minor_count) */ |
b411b363 PR |
70 | |
71 | #ifdef CONFIG_DRBD_FAULT_INJECTION | |
183ece30 RK |
72 | int drbd_enable_faults; |
73 | int drbd_fault_rate; | |
74 | static int drbd_fault_count; | |
75 | static int drbd_fault_devs; | |
b411b363 | 76 | /* bitmap of enabled faults */ |
183ece30 | 77 | module_param_named(enable_faults, drbd_enable_faults, int, 0664); |
b411b363 | 78 | /* fault rate % value - applies to all enabled faults */ |
183ece30 | 79 | module_param_named(fault_rate, drbd_fault_rate, int, 0664); |
b411b363 | 80 | /* count of faults inserted */ |
183ece30 | 81 | module_param_named(fault_count, drbd_fault_count, int, 0664); |
b411b363 | 82 | /* bitmap of devices to insert faults on */ |
183ece30 | 83 | module_param_named(fault_devs, drbd_fault_devs, int, 0644); |
b411b363 PR |
84 | #endif |
85 | ||
183ece30 RK |
86 | /* module parameters we can keep static */ |
87 | static bool drbd_allow_oos; /* allow_open_on_secondary */ | |
88 | static bool drbd_disable_sendpage; | |
89 | MODULE_PARM_DESC(allow_oos, "DONT USE!"); | |
90 | module_param_named(allow_oos, drbd_allow_oos, bool, 0); | |
91 | module_param_named(disable_sendpage, drbd_disable_sendpage, bool, 0644); | |
92 | ||
93 | /* module parameters we share */ | |
94 | int drbd_proc_details; /* Detail level in proc drbd*/ | |
95 | module_param_named(proc_details, drbd_proc_details, int, 0644); | |
96 | /* module parameters shared with defaults */ | |
97 | unsigned int drbd_minor_count = DRBD_MINOR_COUNT_DEF; | |
b411b363 PR |
98 | /* Module parameter for setting the user mode helper program |
99 | * to run. Default is /sbin/drbdadm */ | |
8ab761e1 | 100 | char drbd_usermode_helper[80] = "/sbin/drbdadm"; |
183ece30 | 101 | module_param_named(minor_count, drbd_minor_count, uint, 0444); |
8ab761e1 | 102 | module_param_string(usermode_helper, drbd_usermode_helper, sizeof(drbd_usermode_helper), 0644); |
b411b363 PR |
103 | |
104 | /* in 2.6.x, our device mapping and config info contains our virtual gendisks | |
105 | * as member "struct gendisk *vdisk;" | |
106 | */ | |
05a10ec7 | 107 | struct idr drbd_devices; |
77c556f6 | 108 | struct list_head drbd_resources; |
28bc3b8c | 109 | struct mutex resources_mutex; |
b411b363 PR |
110 | |
111 | struct kmem_cache *drbd_request_cache; | |
6c852bec | 112 | struct kmem_cache *drbd_ee_cache; /* peer requests */ |
b411b363 PR |
113 | struct kmem_cache *drbd_bm_ext_cache; /* bitmap extents */ |
114 | struct kmem_cache *drbd_al_ext_cache; /* activity log extents */ | |
0892fac8 KO |
115 | mempool_t drbd_request_mempool; |
116 | mempool_t drbd_ee_mempool; | |
117 | mempool_t drbd_md_io_page_pool; | |
118 | struct bio_set drbd_md_io_bio_set; | |
119 | struct bio_set drbd_io_bio_set; | |
b411b363 PR |
120 | |
121 | /* I do not use a standard mempool, because: | |
122 | 1) I want to hand out the pre-allocated objects first. | |
123 | 2) I want to be able to interrupt sleeping allocation with a signal. | |
124 | Note: This is a single linked list, the next pointer is the private | |
125 | member of struct page. | |
126 | */ | |
127 | struct page *drbd_pp_pool; | |
9c282c29 | 128 | DEFINE_SPINLOCK(drbd_pp_lock); |
b411b363 PR |
129 | int drbd_pp_vacant; |
130 | wait_queue_head_t drbd_pp_wait; | |
131 | ||
132 | DEFINE_RATELIMIT_STATE(drbd_ratelimit_state, 5 * HZ, 5); | |
133 | ||
7d4e9d09 | 134 | static const struct block_device_operations drbd_ops = { |
c62b37d9 CH |
135 | .owner = THIS_MODULE, |
136 | .submit_bio = drbd_submit_bio, | |
137 | .open = drbd_open, | |
138 | .release = drbd_release, | |
b411b363 | 139 | }; |
19f843aa | 140 | |
b411b363 PR |
141 | #ifdef __CHECKER__ |
142 | /* When checking with sparse, and this is an inline function, sparse will | |
143 | give tons of false positives. When this is a real functions sparse works. | |
b411b363 | 144 | */ |
b30ab791 | 145 | int _get_ldev_if_state(struct drbd_device *device, enum drbd_disk_state mins) |
b411b363 | 146 | { |
b411b363 | 147 | int io_allowed; |
b411b363 | 148 | |
b30ab791 AG |
149 | atomic_inc(&device->local_cnt); |
150 | io_allowed = (device->state.disk >= mins); | |
b411b363 | 151 | if (!io_allowed) { |
b30ab791 AG |
152 | if (atomic_dec_and_test(&device->local_cnt)) |
153 | wake_up(&device->misc_wait); | |
b411b363 | 154 | } |
b411b363 PR |
155 | return io_allowed; |
156 | } | |
b411b363 | 157 | |
b411b363 | 158 | #endif |
265be2d0 | 159 | |
b411b363 | 160 | /** |
b6dd1a89 | 161 | * tl_release() - mark as BARRIER_ACKED all requests in the corresponding transfer log epoch |
bde89a9e | 162 | * @connection: DRBD connection. |
b411b363 PR |
163 | * @barrier_nr: Expected identifier of the DRBD write barrier packet. |
164 | * @set_size: Expected number of requests before that barrier. | |
165 | * | |
166 | * In case the passed barrier_nr or set_size does not match the oldest | |
b6dd1a89 LE |
167 | * epoch of not yet barrier-acked requests, this function will cause a |
168 | * termination of the connection. | |
b411b363 | 169 | */ |
bde89a9e | 170 | void tl_release(struct drbd_connection *connection, unsigned int barrier_nr, |
2f5cdd0b | 171 | unsigned int set_size) |
b411b363 | 172 | { |
b411b363 | 173 | struct drbd_request *r; |
901aeda6 | 174 | struct drbd_request *req = NULL, *tmp = NULL; |
b6dd1a89 LE |
175 | int expect_epoch = 0; |
176 | int expect_size = 0; | |
b411b363 | 177 | |
0500813f | 178 | spin_lock_irq(&connection->resource->req_lock); |
b411b363 | 179 | |
98683650 | 180 | /* find oldest not yet barrier-acked write request, |
b6dd1a89 | 181 | * count writes in its epoch. */ |
bde89a9e | 182 | list_for_each_entry(r, &connection->transfer_log, tl_requests) { |
a0d856df | 183 | const unsigned s = r->rq_state; |
b6dd1a89 LE |
184 | if (!req) { |
185 | if (!(s & RQ_WRITE)) | |
186 | continue; | |
187 | if (!(s & RQ_NET_MASK)) | |
188 | continue; | |
189 | if (s & RQ_NET_DONE) | |
190 | continue; | |
191 | req = r; | |
192 | expect_epoch = req->epoch; | |
193 | expect_size ++; | |
194 | } else { | |
195 | if (r->epoch != expect_epoch) | |
196 | break; | |
197 | if (!(s & RQ_WRITE)) | |
198 | continue; | |
199 | /* if (s & RQ_DONE): not expected */ | |
200 | /* if (!(s & RQ_NET_MASK)): not expected */ | |
201 | expect_size++; | |
43a5182c | 202 | } |
b411b363 | 203 | } |
67098930 | 204 | |
b411b363 | 205 | /* first some paranoia code */ |
b6dd1a89 | 206 | if (req == NULL) { |
1ec861eb | 207 | drbd_err(connection, "BAD! BarrierAck #%u received, but no epoch in tl!?\n", |
2f5cdd0b | 208 | barrier_nr); |
b411b363 | 209 | goto bail; |
b411b363 | 210 | } |
b6dd1a89 | 211 | if (expect_epoch != barrier_nr) { |
1ec861eb | 212 | drbd_err(connection, "BAD! BarrierAck #%u received, expected #%u!\n", |
b6dd1a89 | 213 | barrier_nr, expect_epoch); |
b411b363 | 214 | goto bail; |
5a22db89 LE |
215 | } |
216 | ||
b6dd1a89 | 217 | if (expect_size != set_size) { |
1ec861eb | 218 | drbd_err(connection, "BAD! BarrierAck #%u received with n_writes=%u, expected n_writes=%u!\n", |
b6dd1a89 | 219 | barrier_nr, set_size, expect_size); |
b411b363 | 220 | goto bail; |
b411b363 PR |
221 | } |
222 | ||
98683650 PR |
223 | /* Clean up list of requests processed during current epoch. */ |
224 | /* this extra list walk restart is paranoia, | |
225 | * to catch requests being barrier-acked "unexpectedly". | |
226 | * It usually should find the same req again, or some READ preceding it. */ | |
bde89a9e | 227 | list_for_each_entry(req, &connection->transfer_log, tl_requests) |
901aeda6 JK |
228 | if (req->epoch == expect_epoch) { |
229 | tmp = req; | |
98683650 | 230 | break; |
901aeda6 JK |
231 | } |
232 | req = list_prepare_entry(tmp, &connection->transfer_log, tl_requests); | |
bde89a9e | 233 | list_for_each_entry_safe_from(req, r, &connection->transfer_log, tl_requests) { |
b6dd1a89 LE |
234 | if (req->epoch != expect_epoch) |
235 | break; | |
236 | _req_mod(req, BARRIER_ACKED); | |
19f843aa | 237 | } |
0500813f | 238 | spin_unlock_irq(&connection->resource->req_lock); |
19f843aa | 239 | |
b411b363 | 240 | return; |
b411b363 | 241 | |
b411b363 | 242 | bail: |
0500813f | 243 | spin_unlock_irq(&connection->resource->req_lock); |
bde89a9e | 244 | conn_request_state(connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD); |
b411b363 | 245 | } |
b411b363 | 246 | |
b411b363 | 247 | |
b411b363 | 248 | /** |
11b58e73 | 249 | * _tl_restart() - Walks the transfer log, and applies an action to all requests |
e5f891b2 | 250 | * @connection: DRBD connection to operate on. |
11b58e73 | 251 | * @what: The action/event to perform with all request objects |
b411b363 | 252 | * |
8554df1c AG |
253 | * @what might be one of CONNECTION_LOST_WHILE_PENDING, RESEND, FAIL_FROZEN_DISK_IO, |
254 | * RESTART_FROZEN_DISK_IO. | |
b411b363 | 255 | */ |
b6dd1a89 | 256 | /* must hold resource->req_lock */ |
bde89a9e | 257 | void _tl_restart(struct drbd_connection *connection, enum drbd_req_event what) |
b411b363 | 258 | { |
b6dd1a89 | 259 | struct drbd_request *req, *r; |
b411b363 | 260 | |
bde89a9e | 261 | list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests) |
b6dd1a89 LE |
262 | _req_mod(req, what); |
263 | } | |
738a84b2 | 264 | |
bde89a9e | 265 | void tl_restart(struct drbd_connection *connection, enum drbd_req_event what) |
b6dd1a89 | 266 | { |
0500813f | 267 | spin_lock_irq(&connection->resource->req_lock); |
bde89a9e | 268 | _tl_restart(connection, what); |
0500813f | 269 | spin_unlock_irq(&connection->resource->req_lock); |
cdfda633 | 270 | } |
b411b363 | 271 | |
b411b363 PR |
272 | /** |
273 | * tl_clear() - Clears all requests and &struct drbd_tl_epoch objects out of the TL | |
584164c8 | 274 | * @connection: DRBD connection. |
b411b363 PR |
275 | * |
276 | * This is called after the connection to the peer was lost. The storage covered | |
277 | * by the requests on the transfer gets marked as our of sync. Called from the | |
278 | * receiver thread and the worker thread. | |
279 | */ | |
bde89a9e | 280 | void tl_clear(struct drbd_connection *connection) |
b411b363 | 281 | { |
bde89a9e | 282 | tl_restart(connection, CONNECTION_LOST_WHILE_PENDING); |
b411b363 | 283 | } |
197296ff | 284 | |
cdfda633 | 285 | /** |
b30ab791 AG |
286 | * tl_abort_disk_io() - Abort disk I/O for all requests for a certain device in the TL |
287 | * @device: DRBD device. | |
cdfda633 | 288 | */ |
b30ab791 | 289 | void tl_abort_disk_io(struct drbd_device *device) |
cdfda633 | 290 | { |
a6b32bc3 | 291 | struct drbd_connection *connection = first_peer_device(device)->connection; |
b6dd1a89 | 292 | struct drbd_request *req, *r; |
02851e9f | 293 | |
0500813f | 294 | spin_lock_irq(&connection->resource->req_lock); |
bde89a9e | 295 | list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests) { |
97ddb687 LE |
296 | if (!(req->rq_state & RQ_LOCAL_PENDING)) |
297 | continue; | |
84b8c06b | 298 | if (req->device != device) |
b6dd1a89 LE |
299 | continue; |
300 | _req_mod(req, ABORT_DISK_IO); | |
b411b363 | 301 | } |
0500813f | 302 | spin_unlock_irq(&connection->resource->req_lock); |
b411b363 PR |
303 | } |
304 | ||
b411b363 PR |
305 | static int drbd_thread_setup(void *arg) |
306 | { | |
307 | struct drbd_thread *thi = (struct drbd_thread *) arg; | |
2457b6d5 | 308 | struct drbd_resource *resource = thi->resource; |
b411b363 PR |
309 | unsigned long flags; |
310 | int retval; | |
311 | ||
f1b3a6ec | 312 | snprintf(current->comm, sizeof(current->comm), "drbd_%c_%s", |
77c556f6 | 313 | thi->name[0], |
2457b6d5 | 314 | resource->name); |
f1b3a6ec | 315 | |
33da8e7c EB |
316 | allow_kernel_signal(DRBD_SIGKILL); |
317 | allow_kernel_signal(SIGXCPU); | |
b411b363 PR |
318 | restart: |
319 | retval = thi->function(thi); | |
320 | ||
321 | spin_lock_irqsave(&thi->t_lock, flags); | |
322 | ||
e77a0a5c | 323 | /* if the receiver has been "EXITING", the last thing it did |
b411b363 PR |
324 | * was set the conn state to "StandAlone", |
325 | * if now a re-connect request comes in, conn state goes C_UNCONNECTED, | |
326 | * and receiver thread will be "started". | |
e77a0a5c | 327 | * drbd_thread_start needs to set "RESTARTING" in that case. |
b411b363 | 328 | * t_state check and assignment needs to be within the same spinlock, |
e77a0a5c AG |
329 | * so either thread_start sees EXITING, and can remap to RESTARTING, |
330 | * or thread_start see NONE, and can proceed as normal. | |
b411b363 PR |
331 | */ |
332 | ||
e77a0a5c | 333 | if (thi->t_state == RESTARTING) { |
2457b6d5 | 334 | drbd_info(resource, "Restarting %s thread\n", thi->name); |
e77a0a5c | 335 | thi->t_state = RUNNING; |
b411b363 PR |
336 | spin_unlock_irqrestore(&thi->t_lock, flags); |
337 | goto restart; | |
338 | } | |
339 | ||
340 | thi->task = NULL; | |
e77a0a5c | 341 | thi->t_state = NONE; |
b411b363 | 342 | smp_mb(); |
992d6e91 | 343 | complete_all(&thi->stop); |
b411b363 PR |
344 | spin_unlock_irqrestore(&thi->t_lock, flags); |
345 | ||
2457b6d5 | 346 | drbd_info(resource, "Terminating %s\n", current->comm); |
b411b363 PR |
347 | |
348 | /* Release mod reference taken when thread was started */ | |
9dc9fbb3 | 349 | |
2457b6d5 AG |
350 | if (thi->connection) |
351 | kref_put(&thi->connection->kref, drbd_destroy_connection); | |
352 | kref_put(&resource->kref, drbd_destroy_resource); | |
b411b363 PR |
353 | module_put(THIS_MODULE); |
354 | return retval; | |
355 | } | |
356 | ||
2457b6d5 | 357 | static void drbd_thread_init(struct drbd_resource *resource, struct drbd_thread *thi, |
c60b0251 | 358 | int (*func) (struct drbd_thread *), const char *name) |
b411b363 PR |
359 | { |
360 | spin_lock_init(&thi->t_lock); | |
361 | thi->task = NULL; | |
e77a0a5c | 362 | thi->t_state = NONE; |
b411b363 | 363 | thi->function = func; |
2457b6d5 AG |
364 | thi->resource = resource; |
365 | thi->connection = NULL; | |
c60b0251 | 366 | thi->name = name; |
b411b363 PR |
367 | } |
368 | ||
369 | int drbd_thread_start(struct drbd_thread *thi) | |
370 | { | |
2457b6d5 | 371 | struct drbd_resource *resource = thi->resource; |
b411b363 PR |
372 | struct task_struct *nt; |
373 | unsigned long flags; | |
374 | ||
b411b363 PR |
375 | /* is used from state engine doing drbd_thread_stop_nowait, |
376 | * while holding the req lock irqsave */ | |
377 | spin_lock_irqsave(&thi->t_lock, flags); | |
378 | ||
379 | switch (thi->t_state) { | |
e77a0a5c | 380 | case NONE: |
2457b6d5 | 381 | drbd_info(resource, "Starting %s thread (from %s [%d])\n", |
bed879ae | 382 | thi->name, current->comm, current->pid); |
b411b363 PR |
383 | |
384 | /* Get ref on module for thread - this is released when thread exits */ | |
385 | if (!try_module_get(THIS_MODULE)) { | |
2457b6d5 | 386 | drbd_err(resource, "Failed to get module reference in drbd_thread_start\n"); |
b411b363 | 387 | spin_unlock_irqrestore(&thi->t_lock, flags); |
81e84650 | 388 | return false; |
b411b363 PR |
389 | } |
390 | ||
2457b6d5 AG |
391 | kref_get(&resource->kref); |
392 | if (thi->connection) | |
393 | kref_get(&thi->connection->kref); | |
9dc9fbb3 | 394 | |
b411b363 | 395 | init_completion(&thi->stop); |
b411b363 | 396 | thi->reset_cpu_mask = 1; |
e77a0a5c | 397 | thi->t_state = RUNNING; |
b411b363 PR |
398 | spin_unlock_irqrestore(&thi->t_lock, flags); |
399 | flush_signals(current); /* otherw. may get -ERESTARTNOINTR */ | |
400 | ||
401 | nt = kthread_create(drbd_thread_setup, (void *) thi, | |
2457b6d5 | 402 | "drbd_%c_%s", thi->name[0], thi->resource->name); |
b411b363 PR |
403 | |
404 | if (IS_ERR(nt)) { | |
2457b6d5 | 405 | drbd_err(resource, "Couldn't start thread\n"); |
b411b363 | 406 | |
2457b6d5 AG |
407 | if (thi->connection) |
408 | kref_put(&thi->connection->kref, drbd_destroy_connection); | |
409 | kref_put(&resource->kref, drbd_destroy_resource); | |
b411b363 | 410 | module_put(THIS_MODULE); |
81e84650 | 411 | return false; |
b411b363 PR |
412 | } |
413 | spin_lock_irqsave(&thi->t_lock, flags); | |
414 | thi->task = nt; | |
e77a0a5c | 415 | thi->t_state = RUNNING; |
b411b363 PR |
416 | spin_unlock_irqrestore(&thi->t_lock, flags); |
417 | wake_up_process(nt); | |
418 | break; | |
e77a0a5c AG |
419 | case EXITING: |
420 | thi->t_state = RESTARTING; | |
2457b6d5 | 421 | drbd_info(resource, "Restarting %s thread (from %s [%d])\n", |
bed879ae | 422 | thi->name, current->comm, current->pid); |
df561f66 | 423 | fallthrough; |
e77a0a5c AG |
424 | case RUNNING: |
425 | case RESTARTING: | |
b411b363 PR |
426 | default: |
427 | spin_unlock_irqrestore(&thi->t_lock, flags); | |
428 | break; | |
429 | } | |
430 | ||
81e84650 | 431 | return true; |
b411b363 PR |
432 | } |
433 | ||
434 | ||
435 | void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait) | |
436 | { | |
437 | unsigned long flags; | |
438 | ||
e77a0a5c | 439 | enum drbd_thread_state ns = restart ? RESTARTING : EXITING; |
b411b363 PR |
440 | |
441 | /* may be called from state engine, holding the req lock irqsave */ | |
442 | spin_lock_irqsave(&thi->t_lock, flags); | |
443 | ||
e77a0a5c | 444 | if (thi->t_state == NONE) { |
b411b363 PR |
445 | spin_unlock_irqrestore(&thi->t_lock, flags); |
446 | if (restart) | |
447 | drbd_thread_start(thi); | |
448 | return; | |
449 | } | |
450 | ||
451 | if (thi->t_state != ns) { | |
452 | if (thi->task == NULL) { | |
453 | spin_unlock_irqrestore(&thi->t_lock, flags); | |
454 | return; | |
455 | } | |
456 | ||
457 | thi->t_state = ns; | |
458 | smp_mb(); | |
459 | init_completion(&thi->stop); | |
460 | if (thi->task != current) | |
fee10990 | 461 | send_sig(DRBD_SIGKILL, thi->task, 1); |
b411b363 PR |
462 | } |
463 | ||
464 | spin_unlock_irqrestore(&thi->t_lock, flags); | |
465 | ||
466 | if (wait) | |
467 | wait_for_completion(&thi->stop); | |
468 | } | |
469 | ||
bde89a9e | 470 | int conn_lowest_minor(struct drbd_connection *connection) |
80822284 | 471 | { |
c06ece6b AG |
472 | struct drbd_peer_device *peer_device; |
473 | int vnr = 0, minor = -1; | |
774b3055 | 474 | |
695d08fa | 475 | rcu_read_lock(); |
c06ece6b AG |
476 | peer_device = idr_get_next(&connection->peer_devices, &vnr); |
477 | if (peer_device) | |
478 | minor = device_to_minor(peer_device->device); | |
695d08fa PR |
479 | rcu_read_unlock(); |
480 | ||
c06ece6b | 481 | return minor; |
80822284 | 482 | } |
774b3055 | 483 | |
b411b363 | 484 | #ifdef CONFIG_SMP |
584164c8 | 485 | /* |
b411b363 | 486 | * drbd_calc_cpu_mask() - Generate CPU masks, spread over all CPUs |
b411b363 | 487 | * |
625a6ba2 | 488 | * Forces all threads of a resource onto the same CPU. This is beneficial for |
b411b363 PR |
489 | * DRBD's performance. May be overwritten by user's configuration. |
490 | */ | |
625a6ba2 | 491 | static void drbd_calc_cpu_mask(cpumask_var_t *cpu_mask) |
b411b363 | 492 | { |
625a6ba2 | 493 | unsigned int *resources_per_cpu, min_index = ~0; |
b411b363 | 494 | |
6396bb22 KC |
495 | resources_per_cpu = kcalloc(nr_cpu_ids, sizeof(*resources_per_cpu), |
496 | GFP_KERNEL); | |
625a6ba2 AG |
497 | if (resources_per_cpu) { |
498 | struct drbd_resource *resource; | |
499 | unsigned int cpu, min = ~0; | |
b411b363 | 500 | |
625a6ba2 AG |
501 | rcu_read_lock(); |
502 | for_each_resource_rcu(resource, &drbd_resources) { | |
503 | for_each_cpu(cpu, resource->cpu_mask) | |
504 | resources_per_cpu[cpu]++; | |
b411b363 | 505 | } |
625a6ba2 AG |
506 | rcu_read_unlock(); |
507 | for_each_online_cpu(cpu) { | |
508 | if (resources_per_cpu[cpu] < min) { | |
509 | min = resources_per_cpu[cpu]; | |
510 | min_index = cpu; | |
511 | } | |
512 | } | |
513 | kfree(resources_per_cpu); | |
514 | } | |
515 | if (min_index == ~0) { | |
516 | cpumask_setall(*cpu_mask); | |
517 | return; | |
b411b363 | 518 | } |
625a6ba2 | 519 | cpumask_set_cpu(min_index, *cpu_mask); |
b411b363 PR |
520 | } |
521 | ||
522 | /** | |
523 | * drbd_thread_current_set_cpu() - modifies the cpu mask of the _current_ thread | |
bc31fe33 | 524 | * @thi: drbd_thread object |
b411b363 PR |
525 | * |
526 | * call in the "main loop" of _all_ threads, no need for any mutex, current won't die | |
527 | * prematurely. | |
528 | */ | |
80822284 | 529 | void drbd_thread_current_set_cpu(struct drbd_thread *thi) |
b411b363 | 530 | { |
2457b6d5 | 531 | struct drbd_resource *resource = thi->resource; |
b411b363 | 532 | struct task_struct *p = current; |
bed879ae | 533 | |
b411b363 PR |
534 | if (!thi->reset_cpu_mask) |
535 | return; | |
536 | thi->reset_cpu_mask = 0; | |
2457b6d5 | 537 | set_cpus_allowed_ptr(p, resource->cpu_mask); |
b411b363 | 538 | } |
625a6ba2 AG |
539 | #else |
540 | #define drbd_calc_cpu_mask(A) ({}) | |
b411b363 PR |
541 | #endif |
542 | ||
584164c8 | 543 | /* |
52b061a4 AG |
544 | * drbd_header_size - size of a packet header |
545 | * | |
546 | * The header size is a multiple of 8, so any payload following the header is | |
547 | * word aligned on 64-bit architectures. (The bitmap send and receive code | |
548 | * relies on this.) | |
549 | */ | |
bde89a9e | 550 | unsigned int drbd_header_size(struct drbd_connection *connection) |
b411b363 | 551 | { |
bde89a9e | 552 | if (connection->agreed_pro_version >= 100) { |
0c8e36d9 AG |
553 | BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header100), 8)); |
554 | return sizeof(struct p_header100); | |
555 | } else { | |
556 | BUILD_BUG_ON(sizeof(struct p_header80) != | |
557 | sizeof(struct p_header95)); | |
558 | BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header80), 8)); | |
559 | return sizeof(struct p_header80); | |
560 | } | |
52b061a4 | 561 | } |
b411b363 | 562 | |
e658983a | 563 | static unsigned int prepare_header80(struct p_header80 *h, enum drbd_packet cmd, int size) |
fd340c12 PR |
564 | { |
565 | h->magic = cpu_to_be32(DRBD_MAGIC); | |
566 | h->command = cpu_to_be16(cmd); | |
567 | h->length = cpu_to_be16(size); | |
e658983a | 568 | return sizeof(struct p_header80); |
fd340c12 | 569 | } |
b411b363 | 570 | |
e658983a | 571 | static unsigned int prepare_header95(struct p_header95 *h, enum drbd_packet cmd, int size) |
fd340c12 PR |
572 | { |
573 | h->magic = cpu_to_be16(DRBD_MAGIC_BIG); | |
b411b363 | 574 | h->command = cpu_to_be16(cmd); |
b55d84ba | 575 | h->length = cpu_to_be32(size); |
e658983a | 576 | return sizeof(struct p_header95); |
fd340c12 | 577 | } |
b411b363 | 578 | |
0c8e36d9 AG |
579 | static unsigned int prepare_header100(struct p_header100 *h, enum drbd_packet cmd, |
580 | int size, int vnr) | |
581 | { | |
582 | h->magic = cpu_to_be32(DRBD_MAGIC_100); | |
583 | h->volume = cpu_to_be16(vnr); | |
584 | h->command = cpu_to_be16(cmd); | |
585 | h->length = cpu_to_be32(size); | |
586 | h->pad = 0; | |
587 | return sizeof(struct p_header100); | |
588 | } | |
b411b363 | 589 | |
bde89a9e | 590 | static unsigned int prepare_header(struct drbd_connection *connection, int vnr, |
0c8e36d9 | 591 | void *buffer, enum drbd_packet cmd, int size) |
d38e787e | 592 | { |
bde89a9e | 593 | if (connection->agreed_pro_version >= 100) |
0c8e36d9 | 594 | return prepare_header100(buffer, cmd, size, vnr); |
bde89a9e | 595 | else if (connection->agreed_pro_version >= 95 && |
0c8e36d9 | 596 | size > DRBD_MAX_SIZE_H80_PACKET) |
e658983a | 597 | return prepare_header95(buffer, cmd, size); |
d38e787e | 598 | else |
e658983a | 599 | return prepare_header80(buffer, cmd, size); |
b411b363 PR |
600 | } |
601 | ||
bde89a9e | 602 | static void *__conn_prepare_command(struct drbd_connection *connection, |
a7eb7bdf | 603 | struct drbd_socket *sock) |
b411b363 | 604 | { |
a7eb7bdf AG |
605 | if (!sock->socket) |
606 | return NULL; | |
bde89a9e | 607 | return sock->sbuf + drbd_header_size(connection); |
a7eb7bdf | 608 | } |
b411b363 | 609 | |
bde89a9e | 610 | void *conn_prepare_command(struct drbd_connection *connection, struct drbd_socket *sock) |
dba58587 | 611 | { |
a7eb7bdf | 612 | void *p; |
b411b363 | 613 | |
dba58587 | 614 | mutex_lock(&sock->mutex); |
bde89a9e | 615 | p = __conn_prepare_command(connection, sock); |
a7eb7bdf | 616 | if (!p) |
dba58587 | 617 | mutex_unlock(&sock->mutex); |
b411b363 | 618 | |
a7eb7bdf | 619 | return p; |
b411b363 PR |
620 | } |
621 | ||
69a22773 | 622 | void *drbd_prepare_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock) |
b411b363 | 623 | { |
69a22773 | 624 | return conn_prepare_command(peer_device->connection, sock); |
dba58587 | 625 | } |
b411b363 | 626 | |
bde89a9e | 627 | static int __send_command(struct drbd_connection *connection, int vnr, |
dba58587 AG |
628 | struct drbd_socket *sock, enum drbd_packet cmd, |
629 | unsigned int header_size, void *data, | |
630 | unsigned int size) | |
631 | { | |
632 | int msg_flags; | |
633 | int err; | |
b411b363 | 634 | |
dba58587 AG |
635 | /* |
636 | * Called with @data == NULL and the size of the data blocks in @size | |
637 | * for commands that send data blocks. For those commands, omit the | |
638 | * MSG_MORE flag: this will increase the likelihood that data blocks | |
639 | * which are page aligned on the sender will end up page aligned on the | |
640 | * receiver. | |
641 | */ | |
642 | msg_flags = data ? MSG_MORE : 0; | |
643 | ||
bde89a9e | 644 | header_size += prepare_header(connection, vnr, sock->sbuf, cmd, |
e658983a | 645 | header_size + size); |
bde89a9e | 646 | err = drbd_send_all(connection, sock->socket, sock->sbuf, header_size, |
dba58587 AG |
647 | msg_flags); |
648 | if (data && !err) | |
bde89a9e | 649 | err = drbd_send_all(connection, sock->socket, data, size, 0); |
123ff122 LE |
650 | /* DRBD protocol "pings" are latency critical. |
651 | * This is supposed to trigger tcp_push_pending_frames() */ | |
652 | if (!err && (cmd == P_PING || cmd == P_PING_ACK)) | |
12abc5ee | 653 | tcp_sock_set_nodelay(sock->socket->sk); |
123ff122 | 654 | |
dba58587 AG |
655 | return err; |
656 | } | |
657 | ||
bde89a9e | 658 | static int __conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock, |
a7eb7bdf AG |
659 | enum drbd_packet cmd, unsigned int header_size, |
660 | void *data, unsigned int size) | |
661 | { | |
bde89a9e | 662 | return __send_command(connection, 0, sock, cmd, header_size, data, size); |
a7eb7bdf AG |
663 | } |
664 | ||
bde89a9e | 665 | int conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock, |
dba58587 AG |
666 | enum drbd_packet cmd, unsigned int header_size, |
667 | void *data, unsigned int size) | |
668 | { | |
669 | int err; | |
b411b363 | 670 | |
bde89a9e | 671 | err = __conn_send_command(connection, sock, cmd, header_size, data, size); |
dba58587 AG |
672 | mutex_unlock(&sock->mutex); |
673 | return err; | |
674 | } | |
675 | ||
69a22773 | 676 | int drbd_send_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock, |
dba58587 AG |
677 | enum drbd_packet cmd, unsigned int header_size, |
678 | void *data, unsigned int size) | |
679 | { | |
680 | int err; | |
681 | ||
69a22773 AG |
682 | err = __send_command(peer_device->connection, peer_device->device->vnr, |
683 | sock, cmd, header_size, data, size); | |
dba58587 AG |
684 | mutex_unlock(&sock->mutex); |
685 | return err; | |
686 | } | |
b411b363 | 687 | |
bde89a9e | 688 | int drbd_send_ping(struct drbd_connection *connection) |
e307f352 | 689 | { |
9f5bdc33 AG |
690 | struct drbd_socket *sock; |
691 | ||
bde89a9e AG |
692 | sock = &connection->meta; |
693 | if (!conn_prepare_command(connection, sock)) | |
9f5bdc33 | 694 | return -EIO; |
bde89a9e | 695 | return conn_send_command(connection, sock, P_PING, 0, NULL, 0); |
e307f352 | 696 | } |
b411b363 | 697 | |
bde89a9e | 698 | int drbd_send_ping_ack(struct drbd_connection *connection) |
e307f352 | 699 | { |
9f5bdc33 AG |
700 | struct drbd_socket *sock; |
701 | ||
bde89a9e AG |
702 | sock = &connection->meta; |
703 | if (!conn_prepare_command(connection, sock)) | |
9f5bdc33 | 704 | return -EIO; |
bde89a9e | 705 | return conn_send_command(connection, sock, P_PING_ACK, 0, NULL, 0); |
b411b363 PR |
706 | } |
707 | ||
69a22773 | 708 | int drbd_send_sync_param(struct drbd_peer_device *peer_device) |
b411b363 | 709 | { |
7c96715a | 710 | struct drbd_socket *sock; |
8e26f9cc | 711 | struct p_rs_param_95 *p; |
9f5bdc33 | 712 | int size; |
69a22773 | 713 | const int apv = peer_device->connection->agreed_pro_version; |
9f5bdc33 | 714 | enum drbd_packet cmd; |
44ed167d | 715 | struct net_conf *nc; |
daeda1cc | 716 | struct disk_conf *dc; |
9f5bdc33 | 717 | |
69a22773 AG |
718 | sock = &peer_device->connection->data; |
719 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 AG |
720 | if (!p) |
721 | return -EIO; | |
b411b363 | 722 | |
44ed167d | 723 | rcu_read_lock(); |
69a22773 | 724 | nc = rcu_dereference(peer_device->connection->net_conf); |
b411b363 PR |
725 | |
726 | size = apv <= 87 ? sizeof(struct p_rs_param) | |
727 | : apv == 88 ? sizeof(struct p_rs_param) | |
44ed167d | 728 | + strlen(nc->verify_alg) + 1 |
8e26f9cc PR |
729 | : apv <= 94 ? sizeof(struct p_rs_param_89) |
730 | : /* apv >= 95 */ sizeof(struct p_rs_param_95); | |
b411b363 | 731 | |
9f5bdc33 | 732 | cmd = apv >= 89 ? P_SYNC_PARAM89 : P_SYNC_PARAM; |
b411b363 | 733 | |
9f5bdc33 | 734 | /* initialize verify_alg and csums_alg */ |
52a0cab3 KC |
735 | BUILD_BUG_ON(sizeof(p->algs) != 2 * SHARED_SECRET_MAX); |
736 | memset(&p->algs, 0, sizeof(p->algs)); | |
b411b363 | 737 | |
69a22773 AG |
738 | if (get_ldev(peer_device->device)) { |
739 | dc = rcu_dereference(peer_device->device->ldev->disk_conf); | |
6394b935 | 740 | p->resync_rate = cpu_to_be32(dc->resync_rate); |
daeda1cc PR |
741 | p->c_plan_ahead = cpu_to_be32(dc->c_plan_ahead); |
742 | p->c_delay_target = cpu_to_be32(dc->c_delay_target); | |
743 | p->c_fill_target = cpu_to_be32(dc->c_fill_target); | |
744 | p->c_max_rate = cpu_to_be32(dc->c_max_rate); | |
69a22773 | 745 | put_ldev(peer_device->device); |
9f5bdc33 | 746 | } else { |
6394b935 | 747 | p->resync_rate = cpu_to_be32(DRBD_RESYNC_RATE_DEF); |
9f5bdc33 AG |
748 | p->c_plan_ahead = cpu_to_be32(DRBD_C_PLAN_AHEAD_DEF); |
749 | p->c_delay_target = cpu_to_be32(DRBD_C_DELAY_TARGET_DEF); | |
750 | p->c_fill_target = cpu_to_be32(DRBD_C_FILL_TARGET_DEF); | |
751 | p->c_max_rate = cpu_to_be32(DRBD_C_MAX_RATE_DEF); | |
752 | } | |
b411b363 | 753 | |
9f5bdc33 | 754 | if (apv >= 88) |
44ed167d | 755 | strcpy(p->verify_alg, nc->verify_alg); |
9f5bdc33 | 756 | if (apv >= 89) |
44ed167d PR |
757 | strcpy(p->csums_alg, nc->csums_alg); |
758 | rcu_read_unlock(); | |
b411b363 | 759 | |
69a22773 | 760 | return drbd_send_command(peer_device, sock, cmd, size, NULL, 0); |
b411b363 PR |
761 | } |
762 | ||
bde89a9e | 763 | int __drbd_send_protocol(struct drbd_connection *connection, enum drbd_packet cmd) |
b411b363 | 764 | { |
9f5bdc33 | 765 | struct drbd_socket *sock; |
b411b363 | 766 | struct p_protocol *p; |
44ed167d | 767 | struct net_conf *nc; |
9f5bdc33 | 768 | int size, cf; |
b411b363 | 769 | |
bde89a9e AG |
770 | sock = &connection->data; |
771 | p = __conn_prepare_command(connection, sock); | |
9f5bdc33 AG |
772 | if (!p) |
773 | return -EIO; | |
b411b363 | 774 | |
44ed167d | 775 | rcu_read_lock(); |
bde89a9e | 776 | nc = rcu_dereference(connection->net_conf); |
b411b363 | 777 | |
bde89a9e | 778 | if (nc->tentative && connection->agreed_pro_version < 92) { |
44ed167d | 779 | rcu_read_unlock(); |
1ec861eb | 780 | drbd_err(connection, "--dry-run is not supported by peer"); |
44ed167d PR |
781 | return -EOPNOTSUPP; |
782 | } | |
b411b363 | 783 | |
9f5bdc33 | 784 | size = sizeof(*p); |
bde89a9e | 785 | if (connection->agreed_pro_version >= 87) |
44ed167d | 786 | size += strlen(nc->integrity_alg) + 1; |
b411b363 | 787 | |
44ed167d PR |
788 | p->protocol = cpu_to_be32(nc->wire_protocol); |
789 | p->after_sb_0p = cpu_to_be32(nc->after_sb_0p); | |
790 | p->after_sb_1p = cpu_to_be32(nc->after_sb_1p); | |
791 | p->after_sb_2p = cpu_to_be32(nc->after_sb_2p); | |
792 | p->two_primaries = cpu_to_be32(nc->two_primaries); | |
cf14c2e9 | 793 | cf = 0; |
6139f60d AG |
794 | if (nc->discard_my_data) |
795 | cf |= CF_DISCARD_MY_DATA; | |
6dff2902 | 796 | if (nc->tentative) |
9f5bdc33 | 797 | cf |= CF_DRY_RUN; |
cf14c2e9 PR |
798 | p->conn_flags = cpu_to_be32(cf); |
799 | ||
bde89a9e | 800 | if (connection->agreed_pro_version >= 87) |
44ed167d PR |
801 | strcpy(p->integrity_alg, nc->integrity_alg); |
802 | rcu_read_unlock(); | |
b411b363 | 803 | |
bde89a9e | 804 | return __conn_send_command(connection, sock, cmd, size, NULL, 0); |
a7eb7bdf AG |
805 | } |
806 | ||
bde89a9e | 807 | int drbd_send_protocol(struct drbd_connection *connection) |
a7eb7bdf AG |
808 | { |
809 | int err; | |
810 | ||
bde89a9e AG |
811 | mutex_lock(&connection->data.mutex); |
812 | err = __drbd_send_protocol(connection, P_PROTOCOL); | |
813 | mutex_unlock(&connection->data.mutex); | |
a7eb7bdf AG |
814 | |
815 | return err; | |
b411b363 PR |
816 | } |
817 | ||
69a22773 | 818 | static int _drbd_send_uuids(struct drbd_peer_device *peer_device, u64 uuid_flags) |
b411b363 | 819 | { |
69a22773 | 820 | struct drbd_device *device = peer_device->device; |
9f5bdc33 AG |
821 | struct drbd_socket *sock; |
822 | struct p_uuids *p; | |
b411b363 PR |
823 | int i; |
824 | ||
b30ab791 | 825 | if (!get_ldev_if_state(device, D_NEGOTIATING)) |
2ae5f95b | 826 | return 0; |
b411b363 | 827 | |
69a22773 AG |
828 | sock = &peer_device->connection->data; |
829 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 | 830 | if (!p) { |
b30ab791 | 831 | put_ldev(device); |
9f5bdc33 AG |
832 | return -EIO; |
833 | } | |
b30ab791 | 834 | spin_lock_irq(&device->ldev->md.uuid_lock); |
b411b363 | 835 | for (i = UI_CURRENT; i < UI_SIZE; i++) |
b30ab791 AG |
836 | p->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]); |
837 | spin_unlock_irq(&device->ldev->md.uuid_lock); | |
b411b363 | 838 | |
b30ab791 AG |
839 | device->comm_bm_set = drbd_bm_total_weight(device); |
840 | p->uuid[UI_SIZE] = cpu_to_be64(device->comm_bm_set); | |
44ed167d | 841 | rcu_read_lock(); |
69a22773 | 842 | uuid_flags |= rcu_dereference(peer_device->connection->net_conf)->discard_my_data ? 1 : 0; |
44ed167d | 843 | rcu_read_unlock(); |
b30ab791 AG |
844 | uuid_flags |= test_bit(CRASHED_PRIMARY, &device->flags) ? 2 : 0; |
845 | uuid_flags |= device->new_state_tmp.disk == D_INCONSISTENT ? 4 : 0; | |
9f5bdc33 | 846 | p->uuid[UI_FLAGS] = cpu_to_be64(uuid_flags); |
b411b363 | 847 | |
b30ab791 | 848 | put_ldev(device); |
69a22773 | 849 | return drbd_send_command(peer_device, sock, P_UUIDS, sizeof(*p), NULL, 0); |
b411b363 PR |
850 | } |
851 | ||
69a22773 | 852 | int drbd_send_uuids(struct drbd_peer_device *peer_device) |
b411b363 | 853 | { |
69a22773 | 854 | return _drbd_send_uuids(peer_device, 0); |
b411b363 PR |
855 | } |
856 | ||
69a22773 | 857 | int drbd_send_uuids_skip_initial_sync(struct drbd_peer_device *peer_device) |
b411b363 | 858 | { |
69a22773 | 859 | return _drbd_send_uuids(peer_device, 8); |
b411b363 PR |
860 | } |
861 | ||
b30ab791 | 862 | void drbd_print_uuids(struct drbd_device *device, const char *text) |
62b0da3a | 863 | { |
b30ab791 AG |
864 | if (get_ldev_if_state(device, D_NEGOTIATING)) { |
865 | u64 *uuid = device->ldev->md.uuid; | |
d0180171 | 866 | drbd_info(device, "%s %016llX:%016llX:%016llX:%016llX\n", |
62b0da3a LE |
867 | text, |
868 | (unsigned long long)uuid[UI_CURRENT], | |
869 | (unsigned long long)uuid[UI_BITMAP], | |
870 | (unsigned long long)uuid[UI_HISTORY_START], | |
871 | (unsigned long long)uuid[UI_HISTORY_END]); | |
b30ab791 | 872 | put_ldev(device); |
62b0da3a | 873 | } else { |
d0180171 | 874 | drbd_info(device, "%s effective data uuid: %016llX\n", |
62b0da3a | 875 | text, |
b30ab791 | 876 | (unsigned long long)device->ed_uuid); |
62b0da3a LE |
877 | } |
878 | } | |
879 | ||
69a22773 | 880 | void drbd_gen_and_send_sync_uuid(struct drbd_peer_device *peer_device) |
b411b363 | 881 | { |
69a22773 | 882 | struct drbd_device *device = peer_device->device; |
9f5bdc33 AG |
883 | struct drbd_socket *sock; |
884 | struct p_rs_uuid *p; | |
5a22db89 LE |
885 | u64 uuid; |
886 | ||
0b0ba1ef | 887 | D_ASSERT(device, device->state.disk == D_UP_TO_DATE); |
b411b363 | 888 | |
b30ab791 | 889 | uuid = device->ldev->md.uuid[UI_BITMAP]; |
5ba3dac5 PR |
890 | if (uuid && uuid != UUID_JUST_CREATED) |
891 | uuid = uuid + UUID_NEW_BM_OFFSET; | |
892 | else | |
893 | get_random_bytes(&uuid, sizeof(u64)); | |
b30ab791 AG |
894 | drbd_uuid_set(device, UI_BITMAP, uuid); |
895 | drbd_print_uuids(device, "updated sync UUID"); | |
896 | drbd_md_sync(device); | |
b411b363 | 897 | |
69a22773 AG |
898 | sock = &peer_device->connection->data; |
899 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 AG |
900 | if (p) { |
901 | p->uuid = cpu_to_be64(uuid); | |
69a22773 | 902 | drbd_send_command(peer_device, sock, P_SYNC_UUID, sizeof(*p), NULL, 0); |
9f5bdc33 | 903 | } |
b411b363 PR |
904 | } |
905 | ||
69a22773 | 906 | int drbd_send_sizes(struct drbd_peer_device *peer_device, int trigger_reply, enum dds_flags flags) |
b411b363 | 907 | { |
69a22773 | 908 | struct drbd_device *device = peer_device->device; |
9f5bdc33 AG |
909 | struct drbd_socket *sock; |
910 | struct p_sizes *p; | |
b411b363 | 911 | sector_t d_size, u_size; |
db141b2f LE |
912 | int q_order_type; |
913 | unsigned int max_bio_size; | |
9104d31a LE |
914 | unsigned int packet_size; |
915 | ||
916 | sock = &peer_device->connection->data; | |
917 | p = drbd_prepare_command(peer_device, sock); | |
918 | if (!p) | |
919 | return -EIO; | |
b411b363 | 920 | |
9104d31a LE |
921 | packet_size = sizeof(*p); |
922 | if (peer_device->connection->agreed_features & DRBD_FF_WSAME) | |
923 | packet_size += sizeof(p->qlim[0]); | |
924 | ||
925 | memset(p, 0, packet_size); | |
b30ab791 | 926 | if (get_ldev_if_state(device, D_NEGOTIATING)) { |
7a38acce CH |
927 | struct block_device *bdev = device->ldev->backing_bdev; |
928 | struct request_queue *q = bdev_get_queue(bdev); | |
929 | ||
b30ab791 | 930 | d_size = drbd_get_max_capacity(device->ldev); |
daeda1cc | 931 | rcu_read_lock(); |
b30ab791 | 932 | u_size = rcu_dereference(device->ldev->disk_conf)->disk_size; |
daeda1cc | 933 | rcu_read_unlock(); |
b30ab791 | 934 | q_order_type = drbd_queue_order_type(device); |
9104d31a | 935 | max_bio_size = queue_max_hw_sectors(q) << 9; |
db141b2f | 936 | max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE); |
40349d0e | 937 | p->qlim->physical_block_size = |
7a38acce | 938 | cpu_to_be32(bdev_physical_block_size(bdev)); |
40349d0e | 939 | p->qlim->logical_block_size = |
7a38acce | 940 | cpu_to_be32(bdev_logical_block_size(bdev)); |
40349d0e | 941 | p->qlim->alignment_offset = |
c6f23b1a | 942 | cpu_to_be32(bdev_alignment_offset(bdev)); |
7a38acce CH |
943 | p->qlim->io_min = cpu_to_be32(bdev_io_min(bdev)); |
944 | p->qlim->io_opt = cpu_to_be32(bdev_io_opt(bdev)); | |
70200574 | 945 | p->qlim->discard_enabled = !!bdev_max_discard_sectors(bdev); |
b30ab791 | 946 | put_ldev(device); |
b411b363 | 947 | } else { |
40349d0e CH |
948 | struct request_queue *q = device->rq_queue; |
949 | ||
950 | p->qlim->physical_block_size = | |
951 | cpu_to_be32(queue_physical_block_size(q)); | |
952 | p->qlim->logical_block_size = | |
953 | cpu_to_be32(queue_logical_block_size(q)); | |
954 | p->qlim->alignment_offset = 0; | |
955 | p->qlim->io_min = cpu_to_be32(queue_io_min(q)); | |
956 | p->qlim->io_opt = cpu_to_be32(queue_io_opt(q)); | |
957 | p->qlim->discard_enabled = 0; | |
958 | ||
b411b363 PR |
959 | d_size = 0; |
960 | u_size = 0; | |
961 | q_order_type = QUEUE_ORDERED_NONE; | |
99432fcc | 962 | max_bio_size = DRBD_MAX_BIO_SIZE; /* ... multiple BIOs per peer_request */ |
b411b363 PR |
963 | } |
964 | ||
69a22773 | 965 | if (peer_device->connection->agreed_pro_version <= 94) |
98683650 | 966 | max_bio_size = min(max_bio_size, DRBD_MAX_SIZE_H80_PACKET); |
69a22773 | 967 | else if (peer_device->connection->agreed_pro_version < 100) |
98683650 | 968 | max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE_P95); |
b411b363 | 969 | |
9f5bdc33 AG |
970 | p->d_size = cpu_to_be64(d_size); |
971 | p->u_size = cpu_to_be64(u_size); | |
155bd9d1 CH |
972 | if (trigger_reply) |
973 | p->c_size = 0; | |
974 | else | |
975 | p->c_size = cpu_to_be64(get_capacity(device->vdisk)); | |
9f5bdc33 AG |
976 | p->max_bio_size = cpu_to_be32(max_bio_size); |
977 | p->queue_order_type = cpu_to_be16(q_order_type); | |
978 | p->dds_flags = cpu_to_be16(flags); | |
9104d31a LE |
979 | |
980 | return drbd_send_command(peer_device, sock, P_SIZES, packet_size, NULL, 0); | |
b411b363 PR |
981 | } |
982 | ||
983 | /** | |
f479ea06 | 984 | * drbd_send_current_state() - Sends the drbd state to the peer |
69a22773 | 985 | * @peer_device: DRBD peer device. |
b411b363 | 986 | */ |
69a22773 | 987 | int drbd_send_current_state(struct drbd_peer_device *peer_device) |
b411b363 | 988 | { |
7c96715a | 989 | struct drbd_socket *sock; |
9f5bdc33 | 990 | struct p_state *p; |
b411b363 | 991 | |
69a22773 AG |
992 | sock = &peer_device->connection->data; |
993 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 AG |
994 | if (!p) |
995 | return -EIO; | |
69a22773 AG |
996 | p->state = cpu_to_be32(peer_device->device->state.i); /* Within the send mutex */ |
997 | return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0); | |
b411b363 PR |
998 | } |
999 | ||
f479ea06 LE |
1000 | /** |
1001 | * drbd_send_state() - After a state change, sends the new state to the peer | |
69a22773 | 1002 | * @peer_device: DRBD peer device. |
43de7c85 | 1003 | * @state: the state to send, not necessarily the current state. |
f479ea06 LE |
1004 | * |
1005 | * Each state change queues an "after_state_ch" work, which will eventually | |
1006 | * send the resulting new state to the peer. If more state changes happen | |
1007 | * between queuing and processing of the after_state_ch work, we still | |
1008 | * want to send each intermediary state in the order it occurred. | |
1009 | */ | |
69a22773 | 1010 | int drbd_send_state(struct drbd_peer_device *peer_device, union drbd_state state) |
f479ea06 | 1011 | { |
43de7c85 PR |
1012 | struct drbd_socket *sock; |
1013 | struct p_state *p; | |
f479ea06 | 1014 | |
69a22773 AG |
1015 | sock = &peer_device->connection->data; |
1016 | p = drbd_prepare_command(peer_device, sock); | |
43de7c85 PR |
1017 | if (!p) |
1018 | return -EIO; | |
1019 | p->state = cpu_to_be32(state.i); /* Within the send mutex */ | |
69a22773 | 1020 | return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0); |
43de7c85 | 1021 | } |
f479ea06 | 1022 | |
69a22773 | 1023 | int drbd_send_state_req(struct drbd_peer_device *peer_device, union drbd_state mask, union drbd_state val) |
9f5bdc33 AG |
1024 | { |
1025 | struct drbd_socket *sock; | |
1026 | struct p_req_state *p; | |
f479ea06 | 1027 | |
69a22773 AG |
1028 | sock = &peer_device->connection->data; |
1029 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 AG |
1030 | if (!p) |
1031 | return -EIO; | |
1032 | p->mask = cpu_to_be32(mask.i); | |
1033 | p->val = cpu_to_be32(val.i); | |
69a22773 | 1034 | return drbd_send_command(peer_device, sock, P_STATE_CHG_REQ, sizeof(*p), NULL, 0); |
b411b363 | 1035 | } |
f479ea06 | 1036 | |
bde89a9e | 1037 | int conn_send_state_req(struct drbd_connection *connection, union drbd_state mask, union drbd_state val) |
b411b363 | 1038 | { |
9f5bdc33 AG |
1039 | enum drbd_packet cmd; |
1040 | struct drbd_socket *sock; | |
1041 | struct p_req_state *p; | |
f479ea06 | 1042 | |
bde89a9e AG |
1043 | cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REQ : P_CONN_ST_CHG_REQ; |
1044 | sock = &connection->data; | |
1045 | p = conn_prepare_command(connection, sock); | |
9f5bdc33 AG |
1046 | if (!p) |
1047 | return -EIO; | |
1048 | p->mask = cpu_to_be32(mask.i); | |
1049 | p->val = cpu_to_be32(val.i); | |
bde89a9e | 1050 | return conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0); |
f479ea06 LE |
1051 | } |
1052 | ||
69a22773 | 1053 | void drbd_send_sr_reply(struct drbd_peer_device *peer_device, enum drbd_state_rv retcode) |
b411b363 | 1054 | { |
9f5bdc33 AG |
1055 | struct drbd_socket *sock; |
1056 | struct p_req_state_reply *p; | |
b411b363 | 1057 | |
69a22773 AG |
1058 | sock = &peer_device->connection->meta; |
1059 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 AG |
1060 | if (p) { |
1061 | p->retcode = cpu_to_be32(retcode); | |
69a22773 | 1062 | drbd_send_command(peer_device, sock, P_STATE_CHG_REPLY, sizeof(*p), NULL, 0); |
9f5bdc33 | 1063 | } |
b411b363 | 1064 | } |
b411b363 | 1065 | |
bde89a9e | 1066 | void conn_send_sr_reply(struct drbd_connection *connection, enum drbd_state_rv retcode) |
047cd4a6 | 1067 | { |
9f5bdc33 AG |
1068 | struct drbd_socket *sock; |
1069 | struct p_req_state_reply *p; | |
bde89a9e | 1070 | enum drbd_packet cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REPLY : P_CONN_ST_CHG_REPLY; |
b411b363 | 1071 | |
bde89a9e AG |
1072 | sock = &connection->meta; |
1073 | p = conn_prepare_command(connection, sock); | |
9f5bdc33 AG |
1074 | if (p) { |
1075 | p->retcode = cpu_to_be32(retcode); | |
bde89a9e | 1076 | conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0); |
9f5bdc33 | 1077 | } |
b411b363 PR |
1078 | } |
1079 | ||
a02d1240 | 1080 | static void dcbp_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code) |
b411b363 | 1081 | { |
a02d1240 AG |
1082 | BUG_ON(code & ~0xf); |
1083 | p->encoding = (p->encoding & ~0xf) | code; | |
1084 | } | |
b411b363 | 1085 | |
a02d1240 AG |
1086 | static void dcbp_set_start(struct p_compressed_bm *p, int set) |
1087 | { | |
1088 | p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0); | |
1089 | } | |
b411b363 | 1090 | |
a02d1240 AG |
1091 | static void dcbp_set_pad_bits(struct p_compressed_bm *p, int n) |
1092 | { | |
1093 | BUG_ON(n & ~0x7); | |
1094 | p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4); | |
b411b363 PR |
1095 | } |
1096 | ||
b30ab791 | 1097 | static int fill_bitmap_rle_bits(struct drbd_device *device, |
50d0b1ad AG |
1098 | struct p_compressed_bm *p, |
1099 | unsigned int size, | |
1100 | struct bm_xfer_ctx *c) | |
b411b363 PR |
1101 | { |
1102 | struct bitstream bs; | |
1103 | unsigned long plain_bits; | |
1104 | unsigned long tmp; | |
1105 | unsigned long rl; | |
1106 | unsigned len; | |
1107 | unsigned toggle; | |
44ed167d | 1108 | int bits, use_rle; |
b411b363 PR |
1109 | |
1110 | /* may we use this feature? */ | |
44ed167d | 1111 | rcu_read_lock(); |
a6b32bc3 | 1112 | use_rle = rcu_dereference(first_peer_device(device)->connection->net_conf)->use_rle; |
44ed167d | 1113 | rcu_read_unlock(); |
a6b32bc3 | 1114 | if (!use_rle || first_peer_device(device)->connection->agreed_pro_version < 90) |
44ed167d | 1115 | return 0; |
b411b363 PR |
1116 | |
1117 | if (c->bit_offset >= c->bm_bits) | |
1118 | return 0; /* nothing to do. */ | |
1119 | ||
1120 | /* use at most thus many bytes */ | |
50d0b1ad AG |
1121 | bitstream_init(&bs, p->code, size, 0); |
1122 | memset(p->code, 0, size); | |
b411b363 PR |
1123 | /* plain bits covered in this code string */ |
1124 | plain_bits = 0; | |
1125 | ||
1126 | /* p->encoding & 0x80 stores whether the first run length is set. | |
1127 | * bit offset is implicit. | |
1128 | * start with toggle == 2 to be able to tell the first iteration */ | |
1129 | toggle = 2; | |
1130 | ||
1131 | /* see how much plain bits we can stuff into one packet | |
1132 | * using RLE and VLI. */ | |
1133 | do { | |
b30ab791 AG |
1134 | tmp = (toggle == 0) ? _drbd_bm_find_next_zero(device, c->bit_offset) |
1135 | : _drbd_bm_find_next(device, c->bit_offset); | |
b411b363 PR |
1136 | if (tmp == -1UL) |
1137 | tmp = c->bm_bits; | |
1138 | rl = tmp - c->bit_offset; | |
1139 | ||
1140 | if (toggle == 2) { /* first iteration */ | |
1141 | if (rl == 0) { | |
1142 | /* the first checked bit was set, | |
1143 | * store start value, */ | |
a02d1240 | 1144 | dcbp_set_start(p, 1); |
b411b363 PR |
1145 | /* but skip encoding of zero run length */ |
1146 | toggle = !toggle; | |
1147 | continue; | |
1148 | } | |
a02d1240 | 1149 | dcbp_set_start(p, 0); |
b411b363 PR |
1150 | } |
1151 | ||
1152 | /* paranoia: catch zero runlength. | |
1153 | * can only happen if bitmap is modified while we scan it. */ | |
1154 | if (rl == 0) { | |
d0180171 | 1155 | drbd_err(device, "unexpected zero runlength while encoding bitmap " |
b411b363 PR |
1156 | "t:%u bo:%lu\n", toggle, c->bit_offset); |
1157 | return -1; | |
1158 | } | |
1159 | ||
1160 | bits = vli_encode_bits(&bs, rl); | |
1161 | if (bits == -ENOBUFS) /* buffer full */ | |
1162 | break; | |
1163 | if (bits <= 0) { | |
d0180171 | 1164 | drbd_err(device, "error while encoding bitmap: %d\n", bits); |
b411b363 PR |
1165 | return 0; |
1166 | } | |
1167 | ||
1168 | toggle = !toggle; | |
1169 | plain_bits += rl; | |
1170 | c->bit_offset = tmp; | |
1171 | } while (c->bit_offset < c->bm_bits); | |
1172 | ||
1173 | len = bs.cur.b - p->code + !!bs.cur.bit; | |
1174 | ||
1175 | if (plain_bits < (len << 3)) { | |
1176 | /* incompressible with this method. | |
1177 | * we need to rewind both word and bit position. */ | |
1178 | c->bit_offset -= plain_bits; | |
1179 | bm_xfer_ctx_bit_to_word_offset(c); | |
1180 | c->bit_offset = c->word_offset * BITS_PER_LONG; | |
1181 | return 0; | |
1182 | } | |
1183 | ||
1184 | /* RLE + VLI was able to compress it just fine. | |
1185 | * update c->word_offset. */ | |
1186 | bm_xfer_ctx_bit_to_word_offset(c); | |
1187 | ||
1188 | /* store pad_bits */ | |
a02d1240 | 1189 | dcbp_set_pad_bits(p, (8 - bs.cur.bit) & 0x7); |
b411b363 PR |
1190 | |
1191 | return len; | |
1192 | } | |
1193 | ||
584164c8 | 1194 | /* |
f70af118 AG |
1195 | * send_bitmap_rle_or_plain |
1196 | * | |
1197 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
1198 | * code upon failure. | |
1199 | */ | |
1200 | static int | |
b30ab791 | 1201 | send_bitmap_rle_or_plain(struct drbd_device *device, struct bm_xfer_ctx *c) |
b411b363 | 1202 | { |
a6b32bc3 AG |
1203 | struct drbd_socket *sock = &first_peer_device(device)->connection->data; |
1204 | unsigned int header_size = drbd_header_size(first_peer_device(device)->connection); | |
e658983a | 1205 | struct p_compressed_bm *p = sock->sbuf + header_size; |
a982dd57 | 1206 | int len, err; |
b411b363 | 1207 | |
b30ab791 | 1208 | len = fill_bitmap_rle_bits(device, p, |
e658983a | 1209 | DRBD_SOCKET_BUFFER_SIZE - header_size - sizeof(*p), c); |
b411b363 | 1210 | if (len < 0) |
f70af118 | 1211 | return -EIO; |
b411b363 PR |
1212 | |
1213 | if (len) { | |
a02d1240 | 1214 | dcbp_set_code(p, RLE_VLI_Bits); |
a6b32bc3 | 1215 | err = __send_command(first_peer_device(device)->connection, device->vnr, sock, |
9f5bdc33 AG |
1216 | P_COMPRESSED_BITMAP, sizeof(*p) + len, |
1217 | NULL, 0); | |
b411b363 | 1218 | c->packets[0]++; |
e658983a | 1219 | c->bytes[0] += header_size + sizeof(*p) + len; |
b411b363 PR |
1220 | |
1221 | if (c->bit_offset >= c->bm_bits) | |
1222 | len = 0; /* DONE */ | |
1223 | } else { | |
1224 | /* was not compressible. | |
1225 | * send a buffer full of plain text bits instead. */ | |
50d0b1ad AG |
1226 | unsigned int data_size; |
1227 | unsigned long num_words; | |
e658983a | 1228 | unsigned long *p = sock->sbuf + header_size; |
50d0b1ad AG |
1229 | |
1230 | data_size = DRBD_SOCKET_BUFFER_SIZE - header_size; | |
e658983a | 1231 | num_words = min_t(size_t, data_size / sizeof(*p), |
50d0b1ad | 1232 | c->bm_words - c->word_offset); |
e658983a | 1233 | len = num_words * sizeof(*p); |
b411b363 | 1234 | if (len) |
b30ab791 | 1235 | drbd_bm_get_lel(device, c->word_offset, num_words, p); |
a6b32bc3 | 1236 | err = __send_command(first_peer_device(device)->connection, device->vnr, sock, P_BITMAP, len, NULL, 0); |
b411b363 PR |
1237 | c->word_offset += num_words; |
1238 | c->bit_offset = c->word_offset * BITS_PER_LONG; | |
1239 | ||
1240 | c->packets[1]++; | |
50d0b1ad | 1241 | c->bytes[1] += header_size + len; |
b411b363 PR |
1242 | |
1243 | if (c->bit_offset > c->bm_bits) | |
1244 | c->bit_offset = c->bm_bits; | |
1245 | } | |
a982dd57 | 1246 | if (!err) { |
f70af118 | 1247 | if (len == 0) { |
b30ab791 | 1248 | INFO_bm_xfer_stats(device, "send", c); |
f70af118 AG |
1249 | return 0; |
1250 | } else | |
1251 | return 1; | |
1252 | } | |
1253 | return -EIO; | |
b411b363 PR |
1254 | } |
1255 | ||
1256 | /* See the comment at receive_bitmap() */ | |
b30ab791 | 1257 | static int _drbd_send_bitmap(struct drbd_device *device) |
b411b363 PR |
1258 | { |
1259 | struct bm_xfer_ctx c; | |
f70af118 | 1260 | int err; |
b411b363 | 1261 | |
b30ab791 | 1262 | if (!expect(device->bitmap)) |
81e84650 | 1263 | return false; |
b411b363 | 1264 | |
b30ab791 AG |
1265 | if (get_ldev(device)) { |
1266 | if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC)) { | |
d0180171 | 1267 | drbd_info(device, "Writing the whole bitmap, MDF_FullSync was set.\n"); |
b30ab791 AG |
1268 | drbd_bm_set_all(device); |
1269 | if (drbd_bm_write(device)) { | |
b411b363 PR |
1270 | /* write_bm did fail! Leave full sync flag set in Meta P_DATA |
1271 | * but otherwise process as per normal - need to tell other | |
1272 | * side that a full resync is required! */ | |
d0180171 | 1273 | drbd_err(device, "Failed to write bitmap to disk!\n"); |
b411b363 | 1274 | } else { |
b30ab791 AG |
1275 | drbd_md_clear_flag(device, MDF_FULL_SYNC); |
1276 | drbd_md_sync(device); | |
b411b363 PR |
1277 | } |
1278 | } | |
b30ab791 | 1279 | put_ldev(device); |
b411b363 PR |
1280 | } |
1281 | ||
1282 | c = (struct bm_xfer_ctx) { | |
b30ab791 AG |
1283 | .bm_bits = drbd_bm_bits(device), |
1284 | .bm_words = drbd_bm_words(device), | |
b411b363 PR |
1285 | }; |
1286 | ||
1287 | do { | |
b30ab791 | 1288 | err = send_bitmap_rle_or_plain(device, &c); |
f70af118 | 1289 | } while (err > 0); |
b411b363 | 1290 | |
f70af118 | 1291 | return err == 0; |
b411b363 PR |
1292 | } |
1293 | ||
b30ab791 | 1294 | int drbd_send_bitmap(struct drbd_device *device) |
b411b363 | 1295 | { |
a6b32bc3 | 1296 | struct drbd_socket *sock = &first_peer_device(device)->connection->data; |
9f5bdc33 | 1297 | int err = -1; |
b411b363 | 1298 | |
9f5bdc33 AG |
1299 | mutex_lock(&sock->mutex); |
1300 | if (sock->socket) | |
b30ab791 | 1301 | err = !_drbd_send_bitmap(device); |
9f5bdc33 | 1302 | mutex_unlock(&sock->mutex); |
b411b363 PR |
1303 | return err; |
1304 | } | |
1305 | ||
bde89a9e | 1306 | void drbd_send_b_ack(struct drbd_connection *connection, u32 barrier_nr, u32 set_size) |
b411b363 | 1307 | { |
9f5bdc33 AG |
1308 | struct drbd_socket *sock; |
1309 | struct p_barrier_ack *p; | |
b411b363 | 1310 | |
bde89a9e | 1311 | if (connection->cstate < C_WF_REPORT_PARAMS) |
9f5bdc33 | 1312 | return; |
b411b363 | 1313 | |
bde89a9e AG |
1314 | sock = &connection->meta; |
1315 | p = conn_prepare_command(connection, sock); | |
9f5bdc33 AG |
1316 | if (!p) |
1317 | return; | |
1318 | p->barrier = barrier_nr; | |
1319 | p->set_size = cpu_to_be32(set_size); | |
bde89a9e | 1320 | conn_send_command(connection, sock, P_BARRIER_ACK, sizeof(*p), NULL, 0); |
b411b363 PR |
1321 | } |
1322 | ||
1323 | /** | |
1324 | * _drbd_send_ack() - Sends an ack packet | |
584164c8 LJ |
1325 | * @peer_device: DRBD peer device. |
1326 | * @cmd: Packet command code. | |
1327 | * @sector: sector, needs to be in big endian byte order | |
1328 | * @blksize: size in byte, needs to be in big endian byte order | |
1329 | * @block_id: Id, big endian byte order | |
b411b363 | 1330 | */ |
69a22773 | 1331 | static int _drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd, |
d8763023 | 1332 | u64 sector, u32 blksize, u64 block_id) |
b411b363 | 1333 | { |
9f5bdc33 AG |
1334 | struct drbd_socket *sock; |
1335 | struct p_block_ack *p; | |
b411b363 | 1336 | |
69a22773 | 1337 | if (peer_device->device->state.conn < C_CONNECTED) |
9f5bdc33 | 1338 | return -EIO; |
b411b363 | 1339 | |
69a22773 AG |
1340 | sock = &peer_device->connection->meta; |
1341 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 | 1342 | if (!p) |
a8c32aa8 | 1343 | return -EIO; |
9f5bdc33 AG |
1344 | p->sector = sector; |
1345 | p->block_id = block_id; | |
1346 | p->blksize = blksize; | |
69a22773 AG |
1347 | p->seq_num = cpu_to_be32(atomic_inc_return(&peer_device->device->packet_seq)); |
1348 | return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0); | |
b411b363 PR |
1349 | } |
1350 | ||
2b2bf214 LE |
1351 | /* dp->sector and dp->block_id already/still in network byte order, |
1352 | * data_size is payload size according to dp->head, | |
1353 | * and may need to be corrected for digest size. */ | |
69a22773 | 1354 | void drbd_send_ack_dp(struct drbd_peer_device *peer_device, enum drbd_packet cmd, |
a9a9994d | 1355 | struct p_data *dp, int data_size) |
b411b363 | 1356 | { |
69a22773 | 1357 | if (peer_device->connection->peer_integrity_tfm) |
3d0e6375 | 1358 | data_size -= crypto_shash_digestsize(peer_device->connection->peer_integrity_tfm); |
69a22773 | 1359 | _drbd_send_ack(peer_device, cmd, dp->sector, cpu_to_be32(data_size), |
a9a9994d | 1360 | dp->block_id); |
b411b363 PR |
1361 | } |
1362 | ||
69a22773 | 1363 | void drbd_send_ack_rp(struct drbd_peer_device *peer_device, enum drbd_packet cmd, |
a9a9994d | 1364 | struct p_block_req *rp) |
b411b363 | 1365 | { |
69a22773 | 1366 | _drbd_send_ack(peer_device, cmd, rp->sector, rp->blksize, rp->block_id); |
b411b363 PR |
1367 | } |
1368 | ||
1369 | /** | |
1370 | * drbd_send_ack() - Sends an ack packet | |
584164c8 LJ |
1371 | * @peer_device: DRBD peer device |
1372 | * @cmd: packet command code | |
1373 | * @peer_req: peer request | |
b411b363 | 1374 | */ |
69a22773 | 1375 | int drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd, |
db830c46 | 1376 | struct drbd_peer_request *peer_req) |
b411b363 | 1377 | { |
69a22773 | 1378 | return _drbd_send_ack(peer_device, cmd, |
dd516121 AG |
1379 | cpu_to_be64(peer_req->i.sector), |
1380 | cpu_to_be32(peer_req->i.size), | |
1381 | peer_req->block_id); | |
b411b363 PR |
1382 | } |
1383 | ||
1384 | /* This function misuses the block_id field to signal if the blocks | |
1385 | * are is sync or not. */ | |
69a22773 | 1386 | int drbd_send_ack_ex(struct drbd_peer_device *peer_device, enum drbd_packet cmd, |
b411b363 PR |
1387 | sector_t sector, int blksize, u64 block_id) |
1388 | { | |
69a22773 | 1389 | return _drbd_send_ack(peer_device, cmd, |
b411b363 PR |
1390 | cpu_to_be64(sector), |
1391 | cpu_to_be32(blksize), | |
1392 | cpu_to_be64(block_id)); | |
1393 | } | |
1394 | ||
700ca8c0 PR |
1395 | int drbd_send_rs_deallocated(struct drbd_peer_device *peer_device, |
1396 | struct drbd_peer_request *peer_req) | |
1397 | { | |
1398 | struct drbd_socket *sock; | |
1399 | struct p_block_desc *p; | |
1400 | ||
1401 | sock = &peer_device->connection->data; | |
1402 | p = drbd_prepare_command(peer_device, sock); | |
1403 | if (!p) | |
1404 | return -EIO; | |
1405 | p->sector = cpu_to_be64(peer_req->i.sector); | |
1406 | p->blksize = cpu_to_be32(peer_req->i.size); | |
1407 | p->pad = 0; | |
1408 | return drbd_send_command(peer_device, sock, P_RS_DEALLOCATED, sizeof(*p), NULL, 0); | |
1409 | } | |
1410 | ||
69a22773 | 1411 | int drbd_send_drequest(struct drbd_peer_device *peer_device, int cmd, |
b411b363 PR |
1412 | sector_t sector, int size, u64 block_id) |
1413 | { | |
9f5bdc33 AG |
1414 | struct drbd_socket *sock; |
1415 | struct p_block_req *p; | |
b411b363 | 1416 | |
69a22773 AG |
1417 | sock = &peer_device->connection->data; |
1418 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 AG |
1419 | if (!p) |
1420 | return -EIO; | |
1421 | p->sector = cpu_to_be64(sector); | |
1422 | p->block_id = block_id; | |
1423 | p->blksize = cpu_to_be32(size); | |
69a22773 | 1424 | return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0); |
b411b363 PR |
1425 | } |
1426 | ||
69a22773 | 1427 | int drbd_send_drequest_csum(struct drbd_peer_device *peer_device, sector_t sector, int size, |
d8763023 | 1428 | void *digest, int digest_size, enum drbd_packet cmd) |
b411b363 | 1429 | { |
9f5bdc33 AG |
1430 | struct drbd_socket *sock; |
1431 | struct p_block_req *p; | |
b411b363 | 1432 | |
9f5bdc33 | 1433 | /* FIXME: Put the digest into the preallocated socket buffer. */ |
b411b363 | 1434 | |
69a22773 AG |
1435 | sock = &peer_device->connection->data; |
1436 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 AG |
1437 | if (!p) |
1438 | return -EIO; | |
1439 | p->sector = cpu_to_be64(sector); | |
1440 | p->block_id = ID_SYNCER /* unused */; | |
1441 | p->blksize = cpu_to_be32(size); | |
69a22773 | 1442 | return drbd_send_command(peer_device, sock, cmd, sizeof(*p), digest, digest_size); |
b411b363 PR |
1443 | } |
1444 | ||
69a22773 | 1445 | int drbd_send_ov_request(struct drbd_peer_device *peer_device, sector_t sector, int size) |
b411b363 | 1446 | { |
9f5bdc33 AG |
1447 | struct drbd_socket *sock; |
1448 | struct p_block_req *p; | |
b411b363 | 1449 | |
69a22773 AG |
1450 | sock = &peer_device->connection->data; |
1451 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 AG |
1452 | if (!p) |
1453 | return -EIO; | |
1454 | p->sector = cpu_to_be64(sector); | |
1455 | p->block_id = ID_SYNCER /* unused */; | |
1456 | p->blksize = cpu_to_be32(size); | |
69a22773 | 1457 | return drbd_send_command(peer_device, sock, P_OV_REQUEST, sizeof(*p), NULL, 0); |
b411b363 PR |
1458 | } |
1459 | ||
1460 | /* called on sndtimeo | |
81e84650 AG |
1461 | * returns false if we should retry, |
1462 | * true if we think connection is dead | |
b411b363 | 1463 | */ |
bde89a9e | 1464 | static int we_should_drop_the_connection(struct drbd_connection *connection, struct socket *sock) |
b411b363 PR |
1465 | { |
1466 | int drop_it; | |
b30ab791 | 1467 | /* long elapsed = (long)(jiffies - device->last_received); */ |
b411b363 | 1468 | |
bde89a9e | 1469 | drop_it = connection->meta.socket == sock |
1c03e520 PR |
1470 | || !connection->ack_receiver.task |
1471 | || get_t_state(&connection->ack_receiver) != RUNNING | |
bde89a9e | 1472 | || connection->cstate < C_WF_REPORT_PARAMS; |
b411b363 PR |
1473 | |
1474 | if (drop_it) | |
81e84650 | 1475 | return true; |
b411b363 | 1476 | |
bde89a9e | 1477 | drop_it = !--connection->ko_count; |
b411b363 | 1478 | if (!drop_it) { |
1ec861eb | 1479 | drbd_err(connection, "[%s/%d] sock_sendmsg time expired, ko = %u\n", |
bde89a9e AG |
1480 | current->comm, current->pid, connection->ko_count); |
1481 | request_ping(connection); | |
b411b363 PR |
1482 | } |
1483 | ||
b30ab791 | 1484 | return drop_it; /* && (device->state == R_PRIMARY) */; |
b411b363 PR |
1485 | } |
1486 | ||
bde89a9e | 1487 | static void drbd_update_congested(struct drbd_connection *connection) |
9e204cdd | 1488 | { |
bde89a9e | 1489 | struct sock *sk = connection->data.socket->sk; |
9e204cdd | 1490 | if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5) |
bde89a9e | 1491 | set_bit(NET_CONGESTED, &connection->flags); |
9e204cdd AG |
1492 | } |
1493 | ||
b411b363 PR |
1494 | /* The idea of sendpage seems to be to put some kind of reference |
1495 | * to the page into the skb, and to hand it over to the NIC. In | |
1496 | * this process get_page() gets called. | |
1497 | * | |
1498 | * As soon as the page was really sent over the network put_page() | |
1499 | * gets called by some part of the network layer. [ NIC driver? ] | |
1500 | * | |
1501 | * [ get_page() / put_page() increment/decrement the count. If count | |
1502 | * reaches 0 the page will be freed. ] | |
1503 | * | |
1504 | * This works nicely with pages from FSs. | |
1505 | * But this means that in protocol A we might signal IO completion too early! | |
1506 | * | |
1507 | * In order not to corrupt data during a resync we must make sure | |
1508 | * that we do not reuse our own buffer pages (EEs) to early, therefore | |
1509 | * we have the net_ee list. | |
1510 | * | |
1511 | * XFS seems to have problems, still, it submits pages with page_count == 0! | |
1512 | * As a workaround, we disable sendpage on pages | |
1513 | * with page_count == 0 or PageSlab. | |
1514 | */ | |
69a22773 | 1515 | static int _drbd_no_send_page(struct drbd_peer_device *peer_device, struct page *page, |
b987427b | 1516 | int offset, size_t size, unsigned msg_flags) |
b411b363 | 1517 | { |
b987427b AG |
1518 | struct socket *socket; |
1519 | void *addr; | |
1520 | int err; | |
1521 | ||
69a22773 | 1522 | socket = peer_device->connection->data.socket; |
b987427b | 1523 | addr = kmap(page) + offset; |
69a22773 | 1524 | err = drbd_send_all(peer_device->connection, socket, addr, size, msg_flags); |
b411b363 | 1525 | kunmap(page); |
b987427b | 1526 | if (!err) |
69a22773 | 1527 | peer_device->device->send_cnt += size >> 9; |
b987427b | 1528 | return err; |
b411b363 PR |
1529 | } |
1530 | ||
69a22773 | 1531 | static int _drbd_send_page(struct drbd_peer_device *peer_device, struct page *page, |
ba11ad9a | 1532 | int offset, size_t size, unsigned msg_flags) |
b411b363 | 1533 | { |
69a22773 | 1534 | struct socket *socket = peer_device->connection->data.socket; |
b411b363 | 1535 | int len = size; |
88b390ff | 1536 | int err = -EIO; |
b411b363 PR |
1537 | |
1538 | /* e.g. XFS meta- & log-data is in slab pages, which have a | |
1539 | * page_count of 0 and/or have PageSlab() set. | |
1540 | * we cannot use send_page for those, as that does get_page(); | |
1541 | * put_page(); and would cause either a VM_BUG directly, or | |
1542 | * __page_cache_release a page that would actually still be referenced | |
1543 | * by someone, leading to some obscure delayed Oops somewhere else. */ | |
fb25ebe1 | 1544 | if (drbd_disable_sendpage || !sendpage_ok(page)) |
69a22773 | 1545 | return _drbd_no_send_page(peer_device, page, offset, size, msg_flags); |
b411b363 | 1546 | |
ba11ad9a | 1547 | msg_flags |= MSG_NOSIGNAL; |
69a22773 | 1548 | drbd_update_congested(peer_device->connection); |
b411b363 | 1549 | do { |
88b390ff AG |
1550 | int sent; |
1551 | ||
1552 | sent = socket->ops->sendpage(socket, page, offset, len, msg_flags); | |
b411b363 | 1553 | if (sent <= 0) { |
88b390ff | 1554 | if (sent == -EAGAIN) { |
69a22773 | 1555 | if (we_should_drop_the_connection(peer_device->connection, socket)) |
88b390ff AG |
1556 | break; |
1557 | continue; | |
1558 | } | |
69a22773 | 1559 | drbd_warn(peer_device->device, "%s: size=%d len=%d sent=%d\n", |
b411b363 | 1560 | __func__, (int)size, len, sent); |
88b390ff AG |
1561 | if (sent < 0) |
1562 | err = sent; | |
b411b363 PR |
1563 | break; |
1564 | } | |
1565 | len -= sent; | |
1566 | offset += sent; | |
b30ab791 | 1567 | } while (len > 0 /* THINK && device->cstate >= C_CONNECTED*/); |
69a22773 | 1568 | clear_bit(NET_CONGESTED, &peer_device->connection->flags); |
b411b363 | 1569 | |
88b390ff AG |
1570 | if (len == 0) { |
1571 | err = 0; | |
69a22773 | 1572 | peer_device->device->send_cnt += size >> 9; |
88b390ff AG |
1573 | } |
1574 | return err; | |
b411b363 PR |
1575 | } |
1576 | ||
69a22773 | 1577 | static int _drbd_send_bio(struct drbd_peer_device *peer_device, struct bio *bio) |
b411b363 | 1578 | { |
7988613b KO |
1579 | struct bio_vec bvec; |
1580 | struct bvec_iter iter; | |
1581 | ||
ba11ad9a | 1582 | /* hint all but last page with MSG_MORE */ |
7988613b | 1583 | bio_for_each_segment(bvec, bio, iter) { |
7fae55da AG |
1584 | int err; |
1585 | ||
69a22773 | 1586 | err = _drbd_no_send_page(peer_device, bvec.bv_page, |
7988613b | 1587 | bvec.bv_offset, bvec.bv_len, |
4550dd6c | 1588 | bio_iter_last(bvec, iter) |
7988613b | 1589 | ? 0 : MSG_MORE); |
7fae55da AG |
1590 | if (err) |
1591 | return err; | |
b411b363 | 1592 | } |
7fae55da | 1593 | return 0; |
b411b363 PR |
1594 | } |
1595 | ||
69a22773 | 1596 | static int _drbd_send_zc_bio(struct drbd_peer_device *peer_device, struct bio *bio) |
b411b363 | 1597 | { |
7988613b KO |
1598 | struct bio_vec bvec; |
1599 | struct bvec_iter iter; | |
1600 | ||
ba11ad9a | 1601 | /* hint all but last page with MSG_MORE */ |
7988613b | 1602 | bio_for_each_segment(bvec, bio, iter) { |
7fae55da AG |
1603 | int err; |
1604 | ||
69a22773 | 1605 | err = _drbd_send_page(peer_device, bvec.bv_page, |
7988613b | 1606 | bvec.bv_offset, bvec.bv_len, |
4550dd6c | 1607 | bio_iter_last(bvec, iter) ? 0 : MSG_MORE); |
7fae55da AG |
1608 | if (err) |
1609 | return err; | |
b411b363 | 1610 | } |
7fae55da | 1611 | return 0; |
b411b363 PR |
1612 | } |
1613 | ||
69a22773 | 1614 | static int _drbd_send_zc_ee(struct drbd_peer_device *peer_device, |
db830c46 | 1615 | struct drbd_peer_request *peer_req) |
45bb912b | 1616 | { |
db830c46 AG |
1617 | struct page *page = peer_req->pages; |
1618 | unsigned len = peer_req->i.size; | |
9f69230c | 1619 | int err; |
db830c46 | 1620 | |
ba11ad9a | 1621 | /* hint all but last page with MSG_MORE */ |
45bb912b LE |
1622 | page_chain_for_each(page) { |
1623 | unsigned l = min_t(unsigned, len, PAGE_SIZE); | |
9f69230c | 1624 | |
69a22773 | 1625 | err = _drbd_send_page(peer_device, page, 0, l, |
9f69230c AG |
1626 | page_chain_next(page) ? MSG_MORE : 0); |
1627 | if (err) | |
1628 | return err; | |
45bb912b LE |
1629 | len -= l; |
1630 | } | |
9f69230c | 1631 | return 0; |
45bb912b LE |
1632 | } |
1633 | ||
bb3cc85e MC |
1634 | static u32 bio_flags_to_wire(struct drbd_connection *connection, |
1635 | struct bio *bio) | |
76d2e7ec | 1636 | { |
69a22773 | 1637 | if (connection->agreed_pro_version >= 95) |
1eff9d32 JA |
1638 | return (bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0) | |
1639 | (bio->bi_opf & REQ_FUA ? DP_FUA : 0) | | |
1640 | (bio->bi_opf & REQ_PREFLUSH ? DP_FLUSH : 0) | | |
45c21793 | 1641 | (bio_op(bio) == REQ_OP_DISCARD ? DP_DISCARD : 0) | |
f31e583a LE |
1642 | (bio_op(bio) == REQ_OP_WRITE_ZEROES ? |
1643 | ((connection->agreed_features & DRBD_FF_WZEROES) ? | |
1644 | (DP_ZEROES |(!(bio->bi_opf & REQ_NOUNMAP) ? DP_DISCARD : 0)) | |
1645 | : DP_DISCARD) | |
1646 | : 0); | |
76d2e7ec | 1647 | else |
1eff9d32 | 1648 | return bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0; |
76d2e7ec PR |
1649 | } |
1650 | ||
9305455a | 1651 | /* Used to send write or TRIM aka REQ_OP_DISCARD requests |
2f632aeb | 1652 | * R_PRIMARY -> Peer (P_DATA, P_TRIM) |
b411b363 | 1653 | */ |
69a22773 | 1654 | int drbd_send_dblock(struct drbd_peer_device *peer_device, struct drbd_request *req) |
b411b363 | 1655 | { |
69a22773 | 1656 | struct drbd_device *device = peer_device->device; |
9f5bdc33 AG |
1657 | struct drbd_socket *sock; |
1658 | struct p_data *p; | |
9104d31a | 1659 | void *digest_out; |
b411b363 | 1660 | unsigned int dp_flags = 0; |
11f8b2b6 | 1661 | int digest_size; |
9f5bdc33 | 1662 | int err; |
b411b363 | 1663 | |
69a22773 AG |
1664 | sock = &peer_device->connection->data; |
1665 | p = drbd_prepare_command(peer_device, sock); | |
11f8b2b6 | 1666 | digest_size = peer_device->connection->integrity_tfm ? |
3d0e6375 | 1667 | crypto_shash_digestsize(peer_device->connection->integrity_tfm) : 0; |
b411b363 | 1668 | |
9f5bdc33 AG |
1669 | if (!p) |
1670 | return -EIO; | |
1671 | p->sector = cpu_to_be64(req->i.sector); | |
1672 | p->block_id = (unsigned long)req; | |
b30ab791 | 1673 | p->seq_num = cpu_to_be32(atomic_inc_return(&device->packet_seq)); |
bb3cc85e | 1674 | dp_flags = bio_flags_to_wire(peer_device->connection, req->master_bio); |
b30ab791 AG |
1675 | if (device->state.conn >= C_SYNC_SOURCE && |
1676 | device->state.conn <= C_PAUSED_SYNC_T) | |
b411b363 | 1677 | dp_flags |= DP_MAY_SET_IN_SYNC; |
69a22773 | 1678 | if (peer_device->connection->agreed_pro_version >= 100) { |
303d1448 PR |
1679 | if (req->rq_state & RQ_EXP_RECEIVE_ACK) |
1680 | dp_flags |= DP_SEND_RECEIVE_ACK; | |
08d0dabf LE |
1681 | /* During resync, request an explicit write ack, |
1682 | * even in protocol != C */ | |
1683 | if (req->rq_state & RQ_EXP_WRITE_ACK | |
1684 | || (dp_flags & DP_MAY_SET_IN_SYNC)) | |
303d1448 PR |
1685 | dp_flags |= DP_SEND_WRITE_ACK; |
1686 | } | |
9f5bdc33 | 1687 | p->dp_flags = cpu_to_be32(dp_flags); |
2f632aeb | 1688 | |
f31e583a LE |
1689 | if (dp_flags & (DP_DISCARD|DP_ZEROES)) { |
1690 | enum drbd_packet cmd = (dp_flags & DP_ZEROES) ? P_ZEROES : P_TRIM; | |
2f632aeb LE |
1691 | struct p_trim *t = (struct p_trim*)p; |
1692 | t->size = cpu_to_be32(req->i.size); | |
f31e583a | 1693 | err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*t), NULL, 0); |
2f632aeb LE |
1694 | goto out; |
1695 | } | |
a34592ff | 1696 | digest_out = p + 1; |
2f632aeb LE |
1697 | |
1698 | /* our digest is still only over the payload. | |
1699 | * TRIM does not carry any payload. */ | |
11f8b2b6 | 1700 | if (digest_size) |
9104d31a | 1701 | drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest_out); |
a34592ff CH |
1702 | err = __send_command(peer_device->connection, device->vnr, sock, P_DATA, |
1703 | sizeof(*p) + digest_size, NULL, req->i.size); | |
6bdb9b0e | 1704 | if (!err) { |
470be44a LE |
1705 | /* For protocol A, we have to memcpy the payload into |
1706 | * socket buffers, as we may complete right away | |
1707 | * as soon as we handed it over to tcp, at which point the data | |
1708 | * pages may become invalid. | |
1709 | * | |
1710 | * For data-integrity enabled, we copy it as well, so we can be | |
1711 | * sure that even if the bio pages may still be modified, it | |
1712 | * won't change the data on the wire, thus if the digest checks | |
1713 | * out ok after sending on this side, but does not fit on the | |
1714 | * receiving side, we sure have detected corruption elsewhere. | |
1715 | */ | |
11f8b2b6 | 1716 | if (!(req->rq_state & (RQ_EXP_RECEIVE_ACK | RQ_EXP_WRITE_ACK)) || digest_size) |
69a22773 | 1717 | err = _drbd_send_bio(peer_device, req->master_bio); |
b411b363 | 1718 | else |
69a22773 | 1719 | err = _drbd_send_zc_bio(peer_device, req->master_bio); |
470be44a LE |
1720 | |
1721 | /* double check digest, sometimes buffers have been modified in flight. */ | |
11f8b2b6 | 1722 | if (digest_size > 0 && digest_size <= 64) { |
24c4830c | 1723 | /* 64 byte, 512 bit, is the largest digest size |
470be44a LE |
1724 | * currently supported in kernel crypto. */ |
1725 | unsigned char digest[64]; | |
69a22773 | 1726 | drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest); |
11f8b2b6 | 1727 | if (memcmp(p + 1, digest, digest_size)) { |
d0180171 | 1728 | drbd_warn(device, |
470be44a | 1729 | "Digest mismatch, buffer modified by upper layers during write: %llus +%u\n", |
ace652ac | 1730 | (unsigned long long)req->i.sector, req->i.size); |
470be44a | 1731 | } |
11f8b2b6 | 1732 | } /* else if (digest_size > 64) { |
470be44a LE |
1733 | ... Be noisy about digest too large ... |
1734 | } */ | |
b411b363 | 1735 | } |
2f632aeb | 1736 | out: |
9f5bdc33 | 1737 | mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */ |
b411b363 | 1738 | |
6bdb9b0e | 1739 | return err; |
b411b363 PR |
1740 | } |
1741 | ||
1742 | /* answer packet, used to send data back for read requests: | |
1743 | * Peer -> (diskless) R_PRIMARY (P_DATA_REPLY) | |
1744 | * C_SYNC_SOURCE -> C_SYNC_TARGET (P_RS_DATA_REPLY) | |
1745 | */ | |
69a22773 | 1746 | int drbd_send_block(struct drbd_peer_device *peer_device, enum drbd_packet cmd, |
db830c46 | 1747 | struct drbd_peer_request *peer_req) |
b411b363 | 1748 | { |
69a22773 | 1749 | struct drbd_device *device = peer_device->device; |
9f5bdc33 AG |
1750 | struct drbd_socket *sock; |
1751 | struct p_data *p; | |
7b57b89d | 1752 | int err; |
11f8b2b6 | 1753 | int digest_size; |
b411b363 | 1754 | |
69a22773 AG |
1755 | sock = &peer_device->connection->data; |
1756 | p = drbd_prepare_command(peer_device, sock); | |
b411b363 | 1757 | |
11f8b2b6 | 1758 | digest_size = peer_device->connection->integrity_tfm ? |
3d0e6375 | 1759 | crypto_shash_digestsize(peer_device->connection->integrity_tfm) : 0; |
b411b363 | 1760 | |
9f5bdc33 AG |
1761 | if (!p) |
1762 | return -EIO; | |
1763 | p->sector = cpu_to_be64(peer_req->i.sector); | |
1764 | p->block_id = peer_req->block_id; | |
1765 | p->seq_num = 0; /* unused */ | |
b17f33cb | 1766 | p->dp_flags = 0; |
11f8b2b6 | 1767 | if (digest_size) |
69a22773 | 1768 | drbd_csum_ee(peer_device->connection->integrity_tfm, peer_req, p + 1); |
11f8b2b6 | 1769 | err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*p) + digest_size, NULL, peer_req->i.size); |
7b57b89d | 1770 | if (!err) |
69a22773 | 1771 | err = _drbd_send_zc_ee(peer_device, peer_req); |
9f5bdc33 | 1772 | mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */ |
bd26bfc5 | 1773 | |
7b57b89d | 1774 | return err; |
b411b363 PR |
1775 | } |
1776 | ||
69a22773 | 1777 | int drbd_send_out_of_sync(struct drbd_peer_device *peer_device, struct drbd_request *req) |
73a01a18 | 1778 | { |
9f5bdc33 AG |
1779 | struct drbd_socket *sock; |
1780 | struct p_block_desc *p; | |
73a01a18 | 1781 | |
69a22773 AG |
1782 | sock = &peer_device->connection->data; |
1783 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 AG |
1784 | if (!p) |
1785 | return -EIO; | |
1786 | p->sector = cpu_to_be64(req->i.sector); | |
1787 | p->blksize = cpu_to_be32(req->i.size); | |
69a22773 | 1788 | return drbd_send_command(peer_device, sock, P_OUT_OF_SYNC, sizeof(*p), NULL, 0); |
73a01a18 PR |
1789 | } |
1790 | ||
b411b363 PR |
1791 | /* |
1792 | drbd_send distinguishes two cases: | |
1793 | ||
1794 | Packets sent via the data socket "sock" | |
1795 | and packets sent via the meta data socket "msock" | |
1796 | ||
1797 | sock msock | |
1798 | -----------------+-------------------------+------------------------------ | |
1799 | timeout conf.timeout / 2 conf.timeout / 2 | |
1800 | timeout action send a ping via msock Abort communication | |
1801 | and close all sockets | |
1802 | */ | |
1803 | ||
1804 | /* | |
1805 | * you must have down()ed the appropriate [m]sock_mutex elsewhere! | |
1806 | */ | |
bde89a9e | 1807 | int drbd_send(struct drbd_connection *connection, struct socket *sock, |
b411b363 PR |
1808 | void *buf, size_t size, unsigned msg_flags) |
1809 | { | |
79ab80be | 1810 | struct kvec iov = {.iov_base = buf, .iov_len = size}; |
f7765c36 | 1811 | struct msghdr msg = {.msg_flags = msg_flags | MSG_NOSIGNAL}; |
b411b363 PR |
1812 | int rv, sent = 0; |
1813 | ||
1814 | if (!sock) | |
c0d42c8e | 1815 | return -EBADR; |
b411b363 PR |
1816 | |
1817 | /* THINK if (signal_pending) return ... ? */ | |
1818 | ||
de4eda9d | 1819 | iov_iter_kvec(&msg.msg_iter, ITER_SOURCE, &iov, 1, size); |
79ab80be | 1820 | |
bde89a9e | 1821 | if (sock == connection->data.socket) { |
44ed167d | 1822 | rcu_read_lock(); |
bde89a9e | 1823 | connection->ko_count = rcu_dereference(connection->net_conf)->ko_count; |
44ed167d | 1824 | rcu_read_unlock(); |
bde89a9e | 1825 | drbd_update_congested(connection); |
b411b363 PR |
1826 | } |
1827 | do { | |
79ab80be | 1828 | rv = sock_sendmsg(sock, &msg); |
b411b363 | 1829 | if (rv == -EAGAIN) { |
bde89a9e | 1830 | if (we_should_drop_the_connection(connection, sock)) |
b411b363 PR |
1831 | break; |
1832 | else | |
1833 | continue; | |
1834 | } | |
b411b363 PR |
1835 | if (rv == -EINTR) { |
1836 | flush_signals(current); | |
1837 | rv = 0; | |
1838 | } | |
1839 | if (rv < 0) | |
1840 | break; | |
1841 | sent += rv; | |
b411b363 PR |
1842 | } while (sent < size); |
1843 | ||
bde89a9e AG |
1844 | if (sock == connection->data.socket) |
1845 | clear_bit(NET_CONGESTED, &connection->flags); | |
b411b363 PR |
1846 | |
1847 | if (rv <= 0) { | |
1848 | if (rv != -EAGAIN) { | |
1ec861eb | 1849 | drbd_err(connection, "%s_sendmsg returned %d\n", |
bde89a9e | 1850 | sock == connection->meta.socket ? "msock" : "sock", |
bedbd2a5 | 1851 | rv); |
bde89a9e | 1852 | conn_request_state(connection, NS(conn, C_BROKEN_PIPE), CS_HARD); |
b411b363 | 1853 | } else |
bde89a9e | 1854 | conn_request_state(connection, NS(conn, C_TIMEOUT), CS_HARD); |
b411b363 PR |
1855 | } |
1856 | ||
1857 | return sent; | |
1858 | } | |
1859 | ||
584164c8 | 1860 | /* |
fb708e40 AG |
1861 | * drbd_send_all - Send an entire buffer |
1862 | * | |
1863 | * Returns 0 upon success and a negative error value otherwise. | |
1864 | */ | |
bde89a9e | 1865 | int drbd_send_all(struct drbd_connection *connection, struct socket *sock, void *buffer, |
fb708e40 AG |
1866 | size_t size, unsigned msg_flags) |
1867 | { | |
1868 | int err; | |
1869 | ||
bde89a9e | 1870 | err = drbd_send(connection, sock, buffer, size, msg_flags); |
fb708e40 AG |
1871 | if (err < 0) |
1872 | return err; | |
1873 | if (err != size) | |
1874 | return -EIO; | |
1875 | return 0; | |
1876 | } | |
1877 | ||
b411b363 PR |
1878 | static int drbd_open(struct block_device *bdev, fmode_t mode) |
1879 | { | |
b30ab791 | 1880 | struct drbd_device *device = bdev->bd_disk->private_data; |
b411b363 PR |
1881 | unsigned long flags; |
1882 | int rv = 0; | |
1883 | ||
2a48fc0a | 1884 | mutex_lock(&drbd_main_mutex); |
0500813f | 1885 | spin_lock_irqsave(&device->resource->req_lock, flags); |
b30ab791 | 1886 | /* to have a stable device->state.role |
b411b363 PR |
1887 | * and no race with updating open_cnt */ |
1888 | ||
b30ab791 | 1889 | if (device->state.role != R_PRIMARY) { |
b411b363 PR |
1890 | if (mode & FMODE_WRITE) |
1891 | rv = -EROFS; | |
183ece30 | 1892 | else if (!drbd_allow_oos) |
b411b363 PR |
1893 | rv = -EMEDIUMTYPE; |
1894 | } | |
1895 | ||
1896 | if (!rv) | |
b30ab791 | 1897 | device->open_cnt++; |
0500813f | 1898 | spin_unlock_irqrestore(&device->resource->req_lock, flags); |
2a48fc0a | 1899 | mutex_unlock(&drbd_main_mutex); |
b411b363 PR |
1900 | |
1901 | return rv; | |
1902 | } | |
1903 | ||
db2a144b | 1904 | static void drbd_release(struct gendisk *gd, fmode_t mode) |
b411b363 | 1905 | { |
b30ab791 | 1906 | struct drbd_device *device = gd->private_data; |
2a48fc0a | 1907 | mutex_lock(&drbd_main_mutex); |
b30ab791 | 1908 | device->open_cnt--; |
2a48fc0a | 1909 | mutex_unlock(&drbd_main_mutex); |
b411b363 PR |
1910 | } |
1911 | ||
c51a0ef3 LE |
1912 | /* need to hold resource->req_lock */ |
1913 | void drbd_queue_unplug(struct drbd_device *device) | |
1914 | { | |
1915 | if (device->state.pdsk >= D_INCONSISTENT && device->state.conn >= C_CONNECTED) { | |
1916 | D_ASSERT(device, device->state.role == R_PRIMARY); | |
1917 | if (test_and_clear_bit(UNPLUG_REMOTE, &device->flags)) { | |
1918 | drbd_queue_work_if_unqueued( | |
1919 | &first_peer_device(device)->connection->sender_work, | |
1920 | &device->unplug_work); | |
1921 | } | |
1922 | } | |
1923 | } | |
1924 | ||
b30ab791 | 1925 | static void drbd_set_defaults(struct drbd_device *device) |
b411b363 | 1926 | { |
f399002e LE |
1927 | /* Beware! The actual layout differs |
1928 | * between big endian and little endian */ | |
b30ab791 | 1929 | device->state = (union drbd_dev_state) { |
b411b363 PR |
1930 | { .role = R_SECONDARY, |
1931 | .peer = R_UNKNOWN, | |
1932 | .conn = C_STANDALONE, | |
1933 | .disk = D_DISKLESS, | |
1934 | .pdsk = D_UNKNOWN, | |
b411b363 PR |
1935 | } }; |
1936 | } | |
1937 | ||
b30ab791 | 1938 | void drbd_init_set_defaults(struct drbd_device *device) |
b411b363 PR |
1939 | { |
1940 | /* the memset(,0,) did most of this. | |
1941 | * note: only assignments, no allocation in here */ | |
1942 | ||
b30ab791 AG |
1943 | drbd_set_defaults(device); |
1944 | ||
1945 | atomic_set(&device->ap_bio_cnt, 0); | |
ad3fee79 | 1946 | atomic_set(&device->ap_actlog_cnt, 0); |
b30ab791 AG |
1947 | atomic_set(&device->ap_pending_cnt, 0); |
1948 | atomic_set(&device->rs_pending_cnt, 0); | |
1949 | atomic_set(&device->unacked_cnt, 0); | |
1950 | atomic_set(&device->local_cnt, 0); | |
1951 | atomic_set(&device->pp_in_use_by_net, 0); | |
1952 | atomic_set(&device->rs_sect_in, 0); | |
1953 | atomic_set(&device->rs_sect_ev, 0); | |
1954 | atomic_set(&device->ap_in_flight, 0); | |
e37d2438 | 1955 | atomic_set(&device->md_io.in_use, 0); |
b30ab791 AG |
1956 | |
1957 | mutex_init(&device->own_state_mutex); | |
1958 | device->state_mutex = &device->own_state_mutex; | |
1959 | ||
1960 | spin_lock_init(&device->al_lock); | |
1961 | spin_lock_init(&device->peer_seq_lock); | |
1962 | ||
1963 | INIT_LIST_HEAD(&device->active_ee); | |
1964 | INIT_LIST_HEAD(&device->sync_ee); | |
1965 | INIT_LIST_HEAD(&device->done_ee); | |
1966 | INIT_LIST_HEAD(&device->read_ee); | |
1967 | INIT_LIST_HEAD(&device->net_ee); | |
1968 | INIT_LIST_HEAD(&device->resync_reads); | |
1969 | INIT_LIST_HEAD(&device->resync_work.list); | |
1970 | INIT_LIST_HEAD(&device->unplug_work.list); | |
b30ab791 | 1971 | INIT_LIST_HEAD(&device->bm_io_work.w.list); |
844a6ae7 LE |
1972 | INIT_LIST_HEAD(&device->pending_master_completion[0]); |
1973 | INIT_LIST_HEAD(&device->pending_master_completion[1]); | |
1974 | INIT_LIST_HEAD(&device->pending_completion[0]); | |
1975 | INIT_LIST_HEAD(&device->pending_completion[1]); | |
b30ab791 AG |
1976 | |
1977 | device->resync_work.cb = w_resync_timer; | |
1978 | device->unplug_work.cb = w_send_write_hint; | |
b30ab791 | 1979 | device->bm_io_work.w.cb = w_bitmap_io; |
b30ab791 | 1980 | |
2bccef39 KC |
1981 | timer_setup(&device->resync_timer, resync_timer_fn, 0); |
1982 | timer_setup(&device->md_sync_timer, md_sync_timer_fn, 0); | |
1983 | timer_setup(&device->start_resync_timer, start_resync_timer_fn, 0); | |
1984 | timer_setup(&device->request_timer, request_timer_fn, 0); | |
b30ab791 AG |
1985 | |
1986 | init_waitqueue_head(&device->misc_wait); | |
1987 | init_waitqueue_head(&device->state_wait); | |
1988 | init_waitqueue_head(&device->ee_wait); | |
1989 | init_waitqueue_head(&device->al_wait); | |
1990 | init_waitqueue_head(&device->seq_wait); | |
1991 | ||
1992 | device->resync_wenr = LC_FREE; | |
1993 | device->peer_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE; | |
1994 | device->local_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE; | |
1995 | } | |
1996 | ||
d5412e8d LE |
1997 | void drbd_set_my_capacity(struct drbd_device *device, sector_t size) |
1998 | { | |
1999 | char ppb[10]; | |
155bd9d1 | 2000 | |
bc254eb4 | 2001 | set_capacity_and_notify(device->vdisk, size); |
155bd9d1 | 2002 | |
d5412e8d LE |
2003 | drbd_info(device, "size = %s (%llu KB)\n", |
2004 | ppsize(ppb, size>>1), (unsigned long long)size>>1); | |
2005 | } | |
2006 | ||
b30ab791 | 2007 | void drbd_device_cleanup(struct drbd_device *device) |
b411b363 | 2008 | { |
1d7734a0 | 2009 | int i; |
a6b32bc3 | 2010 | if (first_peer_device(device)->connection->receiver.t_state != NONE) |
d0180171 | 2011 | drbd_err(device, "ASSERT FAILED: receiver t_state == %d expected 0.\n", |
a6b32bc3 | 2012 | first_peer_device(device)->connection->receiver.t_state); |
b30ab791 AG |
2013 | |
2014 | device->al_writ_cnt = | |
2015 | device->bm_writ_cnt = | |
2016 | device->read_cnt = | |
2017 | device->recv_cnt = | |
2018 | device->send_cnt = | |
2019 | device->writ_cnt = | |
2020 | device->p_size = | |
2021 | device->rs_start = | |
2022 | device->rs_total = | |
2023 | device->rs_failed = 0; | |
2024 | device->rs_last_events = 0; | |
2025 | device->rs_last_sect_ev = 0; | |
1d7734a0 | 2026 | for (i = 0; i < DRBD_SYNC_MARKS; i++) { |
b30ab791 AG |
2027 | device->rs_mark_left[i] = 0; |
2028 | device->rs_mark_time[i] = 0; | |
1d7734a0 | 2029 | } |
0b0ba1ef | 2030 | D_ASSERT(device, first_peer_device(device)->connection->net_conf == NULL); |
b411b363 | 2031 | |
bc254eb4 | 2032 | set_capacity_and_notify(device->vdisk, 0); |
b30ab791 | 2033 | if (device->bitmap) { |
b411b363 | 2034 | /* maybe never allocated. */ |
b30ab791 AG |
2035 | drbd_bm_resize(device, 0, 1); |
2036 | drbd_bm_cleanup(device); | |
b411b363 PR |
2037 | } |
2038 | ||
63a7c8ad | 2039 | drbd_backing_dev_free(device, device->ldev); |
b30ab791 | 2040 | device->ldev = NULL; |
1d041225 | 2041 | |
b30ab791 | 2042 | clear_bit(AL_SUSPENDED, &device->flags); |
b411b363 | 2043 | |
0b0ba1ef AG |
2044 | D_ASSERT(device, list_empty(&device->active_ee)); |
2045 | D_ASSERT(device, list_empty(&device->sync_ee)); | |
2046 | D_ASSERT(device, list_empty(&device->done_ee)); | |
2047 | D_ASSERT(device, list_empty(&device->read_ee)); | |
2048 | D_ASSERT(device, list_empty(&device->net_ee)); | |
2049 | D_ASSERT(device, list_empty(&device->resync_reads)); | |
2050 | D_ASSERT(device, list_empty(&first_peer_device(device)->connection->sender_work.q)); | |
2051 | D_ASSERT(device, list_empty(&device->resync_work.list)); | |
2052 | D_ASSERT(device, list_empty(&device->unplug_work.list)); | |
2265b473 | 2053 | |
b30ab791 | 2054 | drbd_set_defaults(device); |
b411b363 PR |
2055 | } |
2056 | ||
2057 | ||
2058 | static void drbd_destroy_mempools(void) | |
2059 | { | |
2060 | struct page *page; | |
2061 | ||
2062 | while (drbd_pp_pool) { | |
2063 | page = drbd_pp_pool; | |
2064 | drbd_pp_pool = (struct page *)page_private(page); | |
2065 | __free_page(page); | |
2066 | drbd_pp_vacant--; | |
2067 | } | |
2068 | ||
0b0ba1ef | 2069 | /* D_ASSERT(device, atomic_read(&drbd_pp_vacant)==0); */ |
b411b363 | 2070 | |
0892fac8 KO |
2071 | bioset_exit(&drbd_io_bio_set); |
2072 | bioset_exit(&drbd_md_io_bio_set); | |
2073 | mempool_exit(&drbd_md_io_page_pool); | |
2074 | mempool_exit(&drbd_ee_mempool); | |
2075 | mempool_exit(&drbd_request_mempool); | |
a12fc00b | 2076 | kmem_cache_destroy(drbd_ee_cache); |
2077 | kmem_cache_destroy(drbd_request_cache); | |
2078 | kmem_cache_destroy(drbd_bm_ext_cache); | |
2079 | kmem_cache_destroy(drbd_al_ext_cache); | |
b411b363 | 2080 | |
b411b363 PR |
2081 | drbd_ee_cache = NULL; |
2082 | drbd_request_cache = NULL; | |
2083 | drbd_bm_ext_cache = NULL; | |
2084 | drbd_al_ext_cache = NULL; | |
2085 | ||
2086 | return; | |
2087 | } | |
2088 | ||
2089 | static int drbd_create_mempools(void) | |
2090 | { | |
2091 | struct page *page; | |
183ece30 | 2092 | const int number = (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * drbd_minor_count; |
0892fac8 | 2093 | int i, ret; |
b411b363 PR |
2094 | |
2095 | /* caches */ | |
2096 | drbd_request_cache = kmem_cache_create( | |
2097 | "drbd_req", sizeof(struct drbd_request), 0, 0, NULL); | |
2098 | if (drbd_request_cache == NULL) | |
2099 | goto Enomem; | |
2100 | ||
2101 | drbd_ee_cache = kmem_cache_create( | |
f6ffca9f | 2102 | "drbd_ee", sizeof(struct drbd_peer_request), 0, 0, NULL); |
b411b363 PR |
2103 | if (drbd_ee_cache == NULL) |
2104 | goto Enomem; | |
2105 | ||
2106 | drbd_bm_ext_cache = kmem_cache_create( | |
2107 | "drbd_bm", sizeof(struct bm_extent), 0, 0, NULL); | |
2108 | if (drbd_bm_ext_cache == NULL) | |
2109 | goto Enomem; | |
2110 | ||
2111 | drbd_al_ext_cache = kmem_cache_create( | |
2112 | "drbd_al", sizeof(struct lc_element), 0, 0, NULL); | |
2113 | if (drbd_al_ext_cache == NULL) | |
2114 | goto Enomem; | |
2115 | ||
2116 | /* mempools */ | |
0892fac8 KO |
2117 | ret = bioset_init(&drbd_io_bio_set, BIO_POOL_SIZE, 0, 0); |
2118 | if (ret) | |
8cb0defb N |
2119 | goto Enomem; |
2120 | ||
0892fac8 KO |
2121 | ret = bioset_init(&drbd_md_io_bio_set, DRBD_MIN_POOL_PAGES, 0, |
2122 | BIOSET_NEED_BVECS); | |
2123 | if (ret) | |
9476f39d | 2124 | goto Enomem; |
9476f39d | 2125 | |
0892fac8 KO |
2126 | ret = mempool_init_page_pool(&drbd_md_io_page_pool, DRBD_MIN_POOL_PAGES, 0); |
2127 | if (ret) | |
4281808f LE |
2128 | goto Enomem; |
2129 | ||
0892fac8 KO |
2130 | ret = mempool_init_slab_pool(&drbd_request_mempool, number, |
2131 | drbd_request_cache); | |
2132 | if (ret) | |
b411b363 PR |
2133 | goto Enomem; |
2134 | ||
0892fac8 KO |
2135 | ret = mempool_init_slab_pool(&drbd_ee_mempool, number, drbd_ee_cache); |
2136 | if (ret) | |
b411b363 PR |
2137 | goto Enomem; |
2138 | ||
b411b363 PR |
2139 | for (i = 0; i < number; i++) { |
2140 | page = alloc_page(GFP_HIGHUSER); | |
2141 | if (!page) | |
2142 | goto Enomem; | |
2143 | set_page_private(page, (unsigned long)drbd_pp_pool); | |
2144 | drbd_pp_pool = page; | |
2145 | } | |
2146 | drbd_pp_vacant = number; | |
2147 | ||
2148 | return 0; | |
2149 | ||
2150 | Enomem: | |
2151 | drbd_destroy_mempools(); /* in case we allocated some */ | |
2152 | return -ENOMEM; | |
2153 | } | |
2154 | ||
b30ab791 | 2155 | static void drbd_release_all_peer_reqs(struct drbd_device *device) |
b411b363 PR |
2156 | { |
2157 | int rr; | |
2158 | ||
b30ab791 | 2159 | rr = drbd_free_peer_reqs(device, &device->active_ee); |
b411b363 | 2160 | if (rr) |
d0180171 | 2161 | drbd_err(device, "%d EEs in active list found!\n", rr); |
b411b363 | 2162 | |
b30ab791 | 2163 | rr = drbd_free_peer_reqs(device, &device->sync_ee); |
b411b363 | 2164 | if (rr) |
d0180171 | 2165 | drbd_err(device, "%d EEs in sync list found!\n", rr); |
b411b363 | 2166 | |
b30ab791 | 2167 | rr = drbd_free_peer_reqs(device, &device->read_ee); |
b411b363 | 2168 | if (rr) |
d0180171 | 2169 | drbd_err(device, "%d EEs in read list found!\n", rr); |
b411b363 | 2170 | |
b30ab791 | 2171 | rr = drbd_free_peer_reqs(device, &device->done_ee); |
b411b363 | 2172 | if (rr) |
d0180171 | 2173 | drbd_err(device, "%d EEs in done list found!\n", rr); |
b411b363 | 2174 | |
b30ab791 | 2175 | rr = drbd_free_peer_reqs(device, &device->net_ee); |
b411b363 | 2176 | if (rr) |
d0180171 | 2177 | drbd_err(device, "%d EEs in net list found!\n", rr); |
b411b363 PR |
2178 | } |
2179 | ||
774b3055 | 2180 | /* caution. no locking. */ |
05a10ec7 | 2181 | void drbd_destroy_device(struct kref *kref) |
b411b363 | 2182 | { |
b30ab791 | 2183 | struct drbd_device *device = container_of(kref, struct drbd_device, kref); |
803ea134 | 2184 | struct drbd_resource *resource = device->resource; |
a8ba0d60 | 2185 | struct drbd_peer_device *peer_device, *tmp_peer_device; |
b411b363 | 2186 | |
b30ab791 | 2187 | del_timer_sync(&device->request_timer); |
dfa8bedb | 2188 | |
b411b363 | 2189 | /* paranoia asserts */ |
0b0ba1ef | 2190 | D_ASSERT(device, device->open_cnt == 0); |
b411b363 PR |
2191 | /* end paranoia asserts */ |
2192 | ||
b411b363 PR |
2193 | /* cleanup stuff that may have been allocated during |
2194 | * device (re-)configuration or state changes */ | |
2195 | ||
63a7c8ad | 2196 | drbd_backing_dev_free(device, device->ldev); |
b30ab791 | 2197 | device->ldev = NULL; |
b411b363 | 2198 | |
b30ab791 | 2199 | drbd_release_all_peer_reqs(device); |
b411b363 | 2200 | |
b30ab791 AG |
2201 | lc_destroy(device->act_log); |
2202 | lc_destroy(device->resync); | |
b411b363 | 2203 | |
b30ab791 AG |
2204 | kfree(device->p_uuid); |
2205 | /* device->p_uuid = NULL; */ | |
b411b363 | 2206 | |
b30ab791 AG |
2207 | if (device->bitmap) /* should no longer be there. */ |
2208 | drbd_bm_cleanup(device); | |
e37d2438 | 2209 | __free_page(device->md_io.page); |
8b9ab626 | 2210 | put_disk(device->vdisk); |
b30ab791 | 2211 | kfree(device->rs_plan_s); |
a8ba0d60 LE |
2212 | |
2213 | /* not for_each_connection(connection, resource): | |
2214 | * those may have been cleaned up and disassociated already. | |
2215 | */ | |
2216 | for_each_peer_device_safe(peer_device, tmp_peer_device, device) { | |
2217 | kref_put(&peer_device->connection->kref, drbd_destroy_connection); | |
2218 | kfree(peer_device); | |
2219 | } | |
c2258ffc | 2220 | memset(device, 0xfd, sizeof(*device)); |
b30ab791 | 2221 | kfree(device); |
803ea134 | 2222 | kref_put(&resource->kref, drbd_destroy_resource); |
b411b363 PR |
2223 | } |
2224 | ||
2312f0b3 LE |
2225 | /* One global retry thread, if we need to push back some bio and have it |
2226 | * reinserted through our make request function. | |
2227 | */ | |
2228 | static struct retry_worker { | |
2229 | struct workqueue_struct *wq; | |
2230 | struct work_struct worker; | |
2231 | ||
2232 | spinlock_t lock; | |
2233 | struct list_head writes; | |
2234 | } retry; | |
2235 | ||
2236 | static void do_retry(struct work_struct *ws) | |
2237 | { | |
2238 | struct retry_worker *retry = container_of(ws, struct retry_worker, worker); | |
2239 | LIST_HEAD(writes); | |
2240 | struct drbd_request *req, *tmp; | |
2241 | ||
2242 | spin_lock_irq(&retry->lock); | |
2243 | list_splice_init(&retry->writes, &writes); | |
2244 | spin_unlock_irq(&retry->lock); | |
2245 | ||
2246 | list_for_each_entry_safe(req, tmp, &writes, tl_requests) { | |
84b8c06b | 2247 | struct drbd_device *device = req->device; |
2312f0b3 | 2248 | struct bio *bio = req->master_bio; |
9a278a79 LE |
2249 | bool expected; |
2250 | ||
84b8c06b | 2251 | expected = |
9a278a79 LE |
2252 | expect(atomic_read(&req->completion_ref) == 0) && |
2253 | expect(req->rq_state & RQ_POSTPONED) && | |
2254 | expect((req->rq_state & RQ_LOCAL_PENDING) == 0 || | |
2255 | (req->rq_state & RQ_LOCAL_ABORTED) != 0); | |
2256 | ||
2257 | if (!expected) | |
d0180171 | 2258 | drbd_err(device, "req=%p completion_ref=%d rq_state=%x\n", |
9a278a79 LE |
2259 | req, atomic_read(&req->completion_ref), |
2260 | req->rq_state); | |
2261 | ||
2262 | /* We still need to put one kref associated with the | |
2263 | * "completion_ref" going zero in the code path that queued it | |
2264 | * here. The request object may still be referenced by a | |
2265 | * frozen local req->private_bio, in case we force-detached. | |
2312f0b3 | 2266 | */ |
9a278a79 | 2267 | kref_put(&req->kref, drbd_req_destroy); |
2312f0b3 LE |
2268 | |
2269 | /* A single suspended or otherwise blocking device may stall | |
2270 | * all others as well. Fortunately, this code path is to | |
2271 | * recover from a situation that "should not happen": | |
2272 | * concurrent writes in multi-primary setup. | |
2273 | * In a "normal" lifecycle, this workqueue is supposed to be | |
2274 | * destroyed without ever doing anything. | |
2275 | * If it turns out to be an issue anyways, we can do per | |
2276 | * resource (replication group) or per device (minor) retry | |
2277 | * workqueues instead. | |
2278 | */ | |
2279 | ||
ed00aabd | 2280 | /* We are not just doing submit_bio_noacct(), |
2312f0b3 | 2281 | * as we want to keep the start_time information. */ |
b30ab791 | 2282 | inc_ap_bio(device); |
370276ba | 2283 | __drbd_make_request(device, bio); |
2312f0b3 LE |
2284 | } |
2285 | } | |
2286 | ||
844a6ae7 LE |
2287 | /* called via drbd_req_put_completion_ref(), |
2288 | * holds resource->req_lock */ | |
9d05e7c4 | 2289 | void drbd_restart_request(struct drbd_request *req) |
2312f0b3 LE |
2290 | { |
2291 | unsigned long flags; | |
2292 | spin_lock_irqsave(&retry.lock, flags); | |
2293 | list_move_tail(&req->tl_requests, &retry.writes); | |
2294 | spin_unlock_irqrestore(&retry.lock, flags); | |
2295 | ||
2296 | /* Drop the extra reference that would otherwise | |
2297 | * have been dropped by complete_master_bio. | |
2298 | * do_retry() needs to grab a new one. */ | |
84b8c06b | 2299 | dec_ap_bio(req->device); |
b411b363 | 2300 | |
2312f0b3 | 2301 | queue_work(retry.wq, &retry.worker); |
b411b363 PR |
2302 | } |
2303 | ||
77c556f6 AG |
2304 | void drbd_destroy_resource(struct kref *kref) |
2305 | { | |
2306 | struct drbd_resource *resource = | |
2307 | container_of(kref, struct drbd_resource, kref); | |
2308 | ||
803ea134 | 2309 | idr_destroy(&resource->devices); |
625a6ba2 | 2310 | free_cpumask_var(resource->cpu_mask); |
77c556f6 | 2311 | kfree(resource->name); |
c2258ffc | 2312 | memset(resource, 0xf2, sizeof(*resource)); |
77c556f6 AG |
2313 | kfree(resource); |
2314 | } | |
2315 | ||
2316 | void drbd_free_resource(struct drbd_resource *resource) | |
2317 | { | |
2318 | struct drbd_connection *connection, *tmp; | |
2319 | ||
2320 | for_each_connection_safe(connection, tmp, resource) { | |
2321 | list_del(&connection->connections); | |
4d3d5aa8 | 2322 | drbd_debugfs_connection_cleanup(connection); |
77c556f6 AG |
2323 | kref_put(&connection->kref, drbd_destroy_connection); |
2324 | } | |
4d3d5aa8 | 2325 | drbd_debugfs_resource_cleanup(resource); |
77c556f6 AG |
2326 | kref_put(&resource->kref, drbd_destroy_resource); |
2327 | } | |
2312f0b3 | 2328 | |
b411b363 PR |
2329 | static void drbd_cleanup(void) |
2330 | { | |
2331 | unsigned int i; | |
b30ab791 | 2332 | struct drbd_device *device; |
77c556f6 | 2333 | struct drbd_resource *resource, *tmp; |
b411b363 | 2334 | |
17a93f30 LE |
2335 | /* first remove proc, |
2336 | * drbdsetup uses it's presence to detect | |
2337 | * whether DRBD is loaded. | |
2338 | * If we would get stuck in proc removal, | |
2339 | * but have netlink already deregistered, | |
2340 | * some drbdsetup commands may wait forever | |
2341 | * for an answer. | |
2342 | */ | |
2343 | if (drbd_proc) | |
2344 | remove_proc_entry("drbd", NULL); | |
2345 | ||
2312f0b3 LE |
2346 | if (retry.wq) |
2347 | destroy_workqueue(retry.wq); | |
b411b363 | 2348 | |
3b98c0c2 | 2349 | drbd_genl_unregister(); |
b411b363 | 2350 | |
803ea134 | 2351 | idr_for_each_entry(&drbd_devices, device, i) |
f82795d6 | 2352 | drbd_delete_device(device); |
b411b363 | 2353 | |
c141ebda | 2354 | /* not _rcu since, no other updater anymore. Genl already unregistered */ |
77c556f6 AG |
2355 | for_each_resource_safe(resource, tmp, &drbd_resources) { |
2356 | list_del(&resource->resources); | |
2357 | drbd_free_resource(resource); | |
81fa2e67 | 2358 | } |
b411b363 | 2359 | |
3f1a1b7c LE |
2360 | drbd_debugfs_cleanup(); |
2361 | ||
81a5d60e | 2362 | drbd_destroy_mempools(); |
b411b363 PR |
2363 | unregister_blkdev(DRBD_MAJOR, "drbd"); |
2364 | ||
05a10ec7 | 2365 | idr_destroy(&drbd_devices); |
81a5d60e | 2366 | |
f88c5d90 | 2367 | pr_info("module cleanup done.\n"); |
b411b363 PR |
2368 | } |
2369 | ||
6699b655 PR |
2370 | static void drbd_init_workqueue(struct drbd_work_queue* wq) |
2371 | { | |
6699b655 PR |
2372 | spin_lock_init(&wq->q_lock); |
2373 | INIT_LIST_HEAD(&wq->q); | |
8c0785a5 | 2374 | init_waitqueue_head(&wq->q_wait); |
6699b655 PR |
2375 | } |
2376 | ||
b5043c5e AG |
2377 | struct completion_work { |
2378 | struct drbd_work w; | |
2379 | struct completion done; | |
2380 | }; | |
2381 | ||
2382 | static int w_complete(struct drbd_work *w, int cancel) | |
2383 | { | |
2384 | struct completion_work *completion_work = | |
2385 | container_of(w, struct completion_work, w); | |
2386 | ||
2387 | complete(&completion_work->done); | |
2388 | return 0; | |
2389 | } | |
2390 | ||
2391 | void drbd_flush_workqueue(struct drbd_work_queue *work_queue) | |
2392 | { | |
2393 | struct completion_work completion_work; | |
2394 | ||
2395 | completion_work.w.cb = w_complete; | |
2396 | init_completion(&completion_work.done); | |
2397 | drbd_queue_work(work_queue, &completion_work.w); | |
2398 | wait_for_completion(&completion_work.done); | |
2399 | } | |
2400 | ||
4bc76048 | 2401 | struct drbd_resource *drbd_find_resource(const char *name) |
1aba4d7f | 2402 | { |
77c556f6 | 2403 | struct drbd_resource *resource; |
1aba4d7f | 2404 | |
3b98c0c2 LE |
2405 | if (!name || !name[0]) |
2406 | return NULL; | |
2407 | ||
c141ebda | 2408 | rcu_read_lock(); |
77c556f6 AG |
2409 | for_each_resource_rcu(resource, &drbd_resources) { |
2410 | if (!strcmp(resource->name, name)) { | |
4bc76048 | 2411 | kref_get(&resource->kref); |
1aba4d7f | 2412 | goto found; |
0ace9dfa | 2413 | } |
1aba4d7f | 2414 | } |
4bc76048 | 2415 | resource = NULL; |
1aba4d7f | 2416 | found: |
c141ebda | 2417 | rcu_read_unlock(); |
4bc76048 | 2418 | return resource; |
1aba4d7f PR |
2419 | } |
2420 | ||
bde89a9e | 2421 | struct drbd_connection *conn_get_by_addrs(void *my_addr, int my_addr_len, |
089c075d AG |
2422 | void *peer_addr, int peer_addr_len) |
2423 | { | |
77c556f6 | 2424 | struct drbd_resource *resource; |
bde89a9e | 2425 | struct drbd_connection *connection; |
089c075d AG |
2426 | |
2427 | rcu_read_lock(); | |
77c556f6 AG |
2428 | for_each_resource_rcu(resource, &drbd_resources) { |
2429 | for_each_connection_rcu(connection, resource) { | |
2430 | if (connection->my_addr_len == my_addr_len && | |
2431 | connection->peer_addr_len == peer_addr_len && | |
2432 | !memcmp(&connection->my_addr, my_addr, my_addr_len) && | |
2433 | !memcmp(&connection->peer_addr, peer_addr, peer_addr_len)) { | |
2434 | kref_get(&connection->kref); | |
2435 | goto found; | |
2436 | } | |
089c075d AG |
2437 | } |
2438 | } | |
bde89a9e | 2439 | connection = NULL; |
089c075d AG |
2440 | found: |
2441 | rcu_read_unlock(); | |
bde89a9e | 2442 | return connection; |
089c075d AG |
2443 | } |
2444 | ||
e6ef8a5c AG |
2445 | static int drbd_alloc_socket(struct drbd_socket *socket) |
2446 | { | |
2447 | socket->rbuf = (void *) __get_free_page(GFP_KERNEL); | |
2448 | if (!socket->rbuf) | |
2449 | return -ENOMEM; | |
5a87d920 AG |
2450 | socket->sbuf = (void *) __get_free_page(GFP_KERNEL); |
2451 | if (!socket->sbuf) | |
2452 | return -ENOMEM; | |
e6ef8a5c AG |
2453 | return 0; |
2454 | } | |
2455 | ||
2456 | static void drbd_free_socket(struct drbd_socket *socket) | |
2457 | { | |
5a87d920 | 2458 | free_page((unsigned long) socket->sbuf); |
e6ef8a5c AG |
2459 | free_page((unsigned long) socket->rbuf); |
2460 | } | |
2461 | ||
bde89a9e | 2462 | void conn_free_crypto(struct drbd_connection *connection) |
91fd4dad | 2463 | { |
bde89a9e | 2464 | drbd_free_sock(connection); |
1d041225 | 2465 | |
3d0e6375 KC |
2466 | crypto_free_shash(connection->csums_tfm); |
2467 | crypto_free_shash(connection->verify_tfm); | |
9534d671 | 2468 | crypto_free_shash(connection->cram_hmac_tfm); |
3d0e6375 KC |
2469 | crypto_free_shash(connection->integrity_tfm); |
2470 | crypto_free_shash(connection->peer_integrity_tfm); | |
bde89a9e AG |
2471 | kfree(connection->int_dig_in); |
2472 | kfree(connection->int_dig_vv); | |
1d041225 | 2473 | |
bde89a9e AG |
2474 | connection->csums_tfm = NULL; |
2475 | connection->verify_tfm = NULL; | |
2476 | connection->cram_hmac_tfm = NULL; | |
2477 | connection->integrity_tfm = NULL; | |
2478 | connection->peer_integrity_tfm = NULL; | |
2479 | connection->int_dig_in = NULL; | |
2480 | connection->int_dig_vv = NULL; | |
91fd4dad PR |
2481 | } |
2482 | ||
eb6bea67 | 2483 | int set_resource_options(struct drbd_resource *resource, struct res_opts *res_opts) |
afbbfa88 | 2484 | { |
eb6bea67 | 2485 | struct drbd_connection *connection; |
afbbfa88 AG |
2486 | cpumask_var_t new_cpu_mask; |
2487 | int err; | |
2488 | ||
2489 | if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) | |
2490 | return -ENOMEM; | |
afbbfa88 AG |
2491 | |
2492 | /* silently ignore cpu mask on UP kernel */ | |
2493 | if (nr_cpu_ids > 1 && res_opts->cpu_mask[0] != 0) { | |
f44d0436 | 2494 | err = bitmap_parse(res_opts->cpu_mask, DRBD_CPU_MASK_SIZE, |
c5b005ab | 2495 | cpumask_bits(new_cpu_mask), nr_cpu_ids); |
1e39152f LE |
2496 | if (err == -EOVERFLOW) { |
2497 | /* So what. mask it out. */ | |
2498 | cpumask_var_t tmp_cpu_mask; | |
2499 | if (zalloc_cpumask_var(&tmp_cpu_mask, GFP_KERNEL)) { | |
2500 | cpumask_setall(tmp_cpu_mask); | |
2501 | cpumask_and(new_cpu_mask, new_cpu_mask, tmp_cpu_mask); | |
2502 | drbd_warn(resource, "Overflow in bitmap_parse(%.12s%s), truncating to %u bits\n", | |
2503 | res_opts->cpu_mask, | |
2504 | strlen(res_opts->cpu_mask) > 12 ? "..." : "", | |
2505 | nr_cpu_ids); | |
2506 | free_cpumask_var(tmp_cpu_mask); | |
2507 | err = 0; | |
2508 | } | |
2509 | } | |
afbbfa88 | 2510 | if (err) { |
1ec861eb | 2511 | drbd_warn(resource, "bitmap_parse() failed with %d\n", err); |
afbbfa88 AG |
2512 | /* retcode = ERR_CPU_MASK_PARSE; */ |
2513 | goto fail; | |
2514 | } | |
2515 | } | |
eb6bea67 | 2516 | resource->res_opts = *res_opts; |
625a6ba2 AG |
2517 | if (cpumask_empty(new_cpu_mask)) |
2518 | drbd_calc_cpu_mask(&new_cpu_mask); | |
2519 | if (!cpumask_equal(resource->cpu_mask, new_cpu_mask)) { | |
2520 | cpumask_copy(resource->cpu_mask, new_cpu_mask); | |
2521 | for_each_connection_rcu(connection, resource) { | |
eb6bea67 | 2522 | connection->receiver.reset_cpu_mask = 1; |
1c03e520 | 2523 | connection->ack_receiver.reset_cpu_mask = 1; |
eb6bea67 AG |
2524 | connection->worker.reset_cpu_mask = 1; |
2525 | } | |
afbbfa88 AG |
2526 | } |
2527 | err = 0; | |
2528 | ||
2529 | fail: | |
2530 | free_cpumask_var(new_cpu_mask); | |
2531 | return err; | |
2532 | ||
2533 | } | |
2534 | ||
77c556f6 AG |
2535 | struct drbd_resource *drbd_create_resource(const char *name) |
2536 | { | |
2537 | struct drbd_resource *resource; | |
2538 | ||
6bbf53ca | 2539 | resource = kzalloc(sizeof(struct drbd_resource), GFP_KERNEL); |
77c556f6 | 2540 | if (!resource) |
625a6ba2 | 2541 | goto fail; |
77c556f6 | 2542 | resource->name = kstrdup(name, GFP_KERNEL); |
625a6ba2 AG |
2543 | if (!resource->name) |
2544 | goto fail_free_resource; | |
2545 | if (!zalloc_cpumask_var(&resource->cpu_mask, GFP_KERNEL)) | |
2546 | goto fail_free_name; | |
77c556f6 | 2547 | kref_init(&resource->kref); |
803ea134 | 2548 | idr_init(&resource->devices); |
77c556f6 | 2549 | INIT_LIST_HEAD(&resource->connections); |
f6ba8636 | 2550 | resource->write_ordering = WO_BDEV_FLUSH; |
77c556f6 | 2551 | list_add_tail_rcu(&resource->resources, &drbd_resources); |
0500813f | 2552 | mutex_init(&resource->conf_update); |
9e276872 | 2553 | mutex_init(&resource->adm_mutex); |
0500813f | 2554 | spin_lock_init(&resource->req_lock); |
4d3d5aa8 | 2555 | drbd_debugfs_resource_add(resource); |
77c556f6 | 2556 | return resource; |
625a6ba2 AG |
2557 | |
2558 | fail_free_name: | |
2559 | kfree(resource->name); | |
2560 | fail_free_resource: | |
2561 | kfree(resource); | |
2562 | fail: | |
2563 | return NULL; | |
77c556f6 AG |
2564 | } |
2565 | ||
4d3d5aa8 | 2566 | /* caller must be under adm_mutex */ |
bde89a9e | 2567 | struct drbd_connection *conn_create(const char *name, struct res_opts *res_opts) |
2111438b | 2568 | { |
77c556f6 | 2569 | struct drbd_resource *resource; |
bde89a9e | 2570 | struct drbd_connection *connection; |
2111438b | 2571 | |
bde89a9e AG |
2572 | connection = kzalloc(sizeof(struct drbd_connection), GFP_KERNEL); |
2573 | if (!connection) | |
2111438b PR |
2574 | return NULL; |
2575 | ||
bde89a9e | 2576 | if (drbd_alloc_socket(&connection->data)) |
e6ef8a5c | 2577 | goto fail; |
bde89a9e | 2578 | if (drbd_alloc_socket(&connection->meta)) |
e6ef8a5c AG |
2579 | goto fail; |
2580 | ||
bde89a9e AG |
2581 | connection->current_epoch = kzalloc(sizeof(struct drbd_epoch), GFP_KERNEL); |
2582 | if (!connection->current_epoch) | |
12038a3a | 2583 | goto fail; |
b6dd1a89 | 2584 | |
bde89a9e | 2585 | INIT_LIST_HEAD(&connection->transfer_log); |
b6dd1a89 | 2586 | |
bde89a9e AG |
2587 | INIT_LIST_HEAD(&connection->current_epoch->list); |
2588 | connection->epochs = 1; | |
2589 | spin_lock_init(&connection->epoch_lock); | |
4b0007c0 | 2590 | |
bde89a9e AG |
2591 | connection->send.seen_any_write_yet = false; |
2592 | connection->send.current_epoch_nr = 0; | |
2593 | connection->send.current_epoch_writes = 0; | |
b6dd1a89 | 2594 | |
77c556f6 AG |
2595 | resource = drbd_create_resource(name); |
2596 | if (!resource) | |
2597 | goto fail; | |
2598 | ||
bde89a9e AG |
2599 | connection->cstate = C_STANDALONE; |
2600 | mutex_init(&connection->cstate_mutex); | |
bde89a9e | 2601 | init_waitqueue_head(&connection->ping_wait); |
c06ece6b | 2602 | idr_init(&connection->peer_devices); |
b2fb6dbe | 2603 | |
bde89a9e AG |
2604 | drbd_init_workqueue(&connection->sender_work); |
2605 | mutex_init(&connection->data.mutex); | |
2606 | mutex_init(&connection->meta.mutex); | |
6699b655 | 2607 | |
2457b6d5 AG |
2608 | drbd_thread_init(resource, &connection->receiver, drbd_receiver, "receiver"); |
2609 | connection->receiver.connection = connection; | |
2610 | drbd_thread_init(resource, &connection->worker, drbd_worker, "worker"); | |
2611 | connection->worker.connection = connection; | |
1c03e520 PR |
2612 | drbd_thread_init(resource, &connection->ack_receiver, drbd_ack_receiver, "ack_recv"); |
2613 | connection->ack_receiver.connection = connection; | |
392c8801 | 2614 | |
bde89a9e | 2615 | kref_init(&connection->kref); |
77c556f6 | 2616 | |
77c556f6 | 2617 | connection->resource = resource; |
2111438b | 2618 | |
eb6bea67 AG |
2619 | if (set_resource_options(resource, res_opts)) |
2620 | goto fail_resource; | |
2621 | ||
2622 | kref_get(&resource->kref); | |
2623 | list_add_tail_rcu(&connection->connections, &resource->connections); | |
4d3d5aa8 | 2624 | drbd_debugfs_connection_add(connection); |
bde89a9e | 2625 | return connection; |
2111438b | 2626 | |
eb6bea67 AG |
2627 | fail_resource: |
2628 | list_del(&resource->resources); | |
2629 | drbd_free_resource(resource); | |
2111438b | 2630 | fail: |
bde89a9e | 2631 | kfree(connection->current_epoch); |
bde89a9e AG |
2632 | drbd_free_socket(&connection->meta); |
2633 | drbd_free_socket(&connection->data); | |
bde89a9e | 2634 | kfree(connection); |
2111438b PR |
2635 | return NULL; |
2636 | } | |
2637 | ||
05a10ec7 | 2638 | void drbd_destroy_connection(struct kref *kref) |
2111438b | 2639 | { |
bde89a9e | 2640 | struct drbd_connection *connection = container_of(kref, struct drbd_connection, kref); |
77c556f6 | 2641 | struct drbd_resource *resource = connection->resource; |
9dc9fbb3 | 2642 | |
bde89a9e | 2643 | if (atomic_read(&connection->current_epoch->epoch_size) != 0) |
1ec861eb | 2644 | drbd_err(connection, "epoch_size:%d\n", atomic_read(&connection->current_epoch->epoch_size)); |
bde89a9e | 2645 | kfree(connection->current_epoch); |
12038a3a | 2646 | |
c06ece6b | 2647 | idr_destroy(&connection->peer_devices); |
2111438b | 2648 | |
bde89a9e AG |
2649 | drbd_free_socket(&connection->meta); |
2650 | drbd_free_socket(&connection->data); | |
bde89a9e AG |
2651 | kfree(connection->int_dig_in); |
2652 | kfree(connection->int_dig_vv); | |
c2258ffc | 2653 | memset(connection, 0xfc, sizeof(*connection)); |
bde89a9e | 2654 | kfree(connection); |
77c556f6 | 2655 | kref_put(&resource->kref, drbd_destroy_resource); |
2111438b PR |
2656 | } |
2657 | ||
b30ab791 | 2658 | static int init_submitter(struct drbd_device *device) |
113fef9e LE |
2659 | { |
2660 | /* opencoded create_singlethread_workqueue(), | |
2661 | * to be able to say "drbd%d", ..., minor */ | |
39e91a60 LE |
2662 | device->submit.wq = |
2663 | alloc_ordered_workqueue("drbd%u_submit", WQ_MEM_RECLAIM, device->minor); | |
b30ab791 | 2664 | if (!device->submit.wq) |
113fef9e LE |
2665 | return -ENOMEM; |
2666 | ||
b30ab791 | 2667 | INIT_WORK(&device->submit.worker, do_submit); |
b30ab791 | 2668 | INIT_LIST_HEAD(&device->submit.writes); |
113fef9e LE |
2669 | return 0; |
2670 | } | |
2671 | ||
a910b123 | 2672 | enum drbd_ret_code drbd_create_device(struct drbd_config_context *adm_ctx, unsigned int minor) |
b411b363 | 2673 | { |
a910b123 | 2674 | struct drbd_resource *resource = adm_ctx->resource; |
a7a15981 | 2675 | struct drbd_connection *connection, *n; |
b30ab791 | 2676 | struct drbd_device *device; |
b6f85ef9 | 2677 | struct drbd_peer_device *peer_device, *tmp_peer_device; |
b411b363 | 2678 | struct gendisk *disk; |
93e4bf7a | 2679 | int id; |
a910b123 | 2680 | int vnr = adm_ctx->volume; |
8432b314 | 2681 | enum drbd_ret_code err = ERR_NOMEM; |
774b3055 | 2682 | |
b30ab791 AG |
2683 | device = minor_to_device(minor); |
2684 | if (device) | |
179e20b8 | 2685 | return ERR_MINOR_OR_VOLUME_EXISTS; |
b411b363 PR |
2686 | |
2687 | /* GFP_KERNEL, we are outside of all write-out paths */ | |
b30ab791 AG |
2688 | device = kzalloc(sizeof(struct drbd_device), GFP_KERNEL); |
2689 | if (!device) | |
774b3055 | 2690 | return ERR_NOMEM; |
803ea134 AG |
2691 | kref_init(&device->kref); |
2692 | ||
803ea134 AG |
2693 | kref_get(&resource->kref); |
2694 | device->resource = resource; | |
b30ab791 AG |
2695 | device->minor = minor; |
2696 | device->vnr = vnr; | |
b411b363 | 2697 | |
b30ab791 | 2698 | drbd_init_set_defaults(device); |
b411b363 | 2699 | |
b647ad02 | 2700 | disk = blk_alloc_disk(NUMA_NO_NODE); |
b411b363 PR |
2701 | if (!disk) |
2702 | goto out_no_disk; | |
b647ad02 | 2703 | |
b30ab791 | 2704 | device->vdisk = disk; |
b647ad02 | 2705 | device->rq_queue = disk->queue; |
b411b363 | 2706 | |
81e84650 | 2707 | set_disk_ro(disk, true); |
b411b363 | 2708 | |
b411b363 PR |
2709 | disk->major = DRBD_MAJOR; |
2710 | disk->first_minor = minor; | |
b647ad02 | 2711 | disk->minors = 1; |
b411b363 | 2712 | disk->fops = &drbd_ops; |
1ebe2e5f | 2713 | disk->flags |= GENHD_FL_NO_PART; |
b411b363 | 2714 | sprintf(disk->disk_name, "drbd%d", minor); |
b30ab791 | 2715 | disk->private_data = device; |
b411b363 | 2716 | |
28690194 | 2717 | blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, disk->queue); |
b647ad02 | 2718 | blk_queue_write_cache(disk->queue, true, true); |
99432fcc PR |
2719 | /* Setting the max_hw_sectors to an odd value of 8kibyte here |
2720 | This triggers a max_bio_size message upon first attach or connect */ | |
b647ad02 | 2721 | blk_queue_max_hw_sectors(disk->queue, DRBD_MAX_BIO_SIZE_SAFE >> 8); |
b411b363 | 2722 | |
e37d2438 LE |
2723 | device->md_io.page = alloc_page(GFP_KERNEL); |
2724 | if (!device->md_io.page) | |
b411b363 PR |
2725 | goto out_no_io_page; |
2726 | ||
b30ab791 | 2727 | if (drbd_bm_init(device)) |
b411b363 | 2728 | goto out_no_bitmap; |
b30ab791 AG |
2729 | device->read_requests = RB_ROOT; |
2730 | device->write_requests = RB_ROOT; | |
b411b363 | 2731 | |
93e4bf7a AG |
2732 | id = idr_alloc(&drbd_devices, device, minor, minor + 1, GFP_KERNEL); |
2733 | if (id < 0) { | |
f221f4bc | 2734 | if (id == -ENOSPC) |
179e20b8 | 2735 | err = ERR_MINOR_OR_VOLUME_EXISTS; |
8432b314 | 2736 | goto out_no_minor_idr; |
81a5d60e | 2737 | } |
803ea134 AG |
2738 | kref_get(&device->kref); |
2739 | ||
2740 | id = idr_alloc(&resource->devices, device, vnr, vnr + 1, GFP_KERNEL); | |
2741 | if (id < 0) { | |
f221f4bc | 2742 | if (id == -ENOSPC) |
179e20b8 | 2743 | err = ERR_MINOR_OR_VOLUME_EXISTS; |
803ea134 AG |
2744 | goto out_idr_remove_minor; |
2745 | } | |
2746 | kref_get(&device->kref); | |
8432b314 | 2747 | |
b6f85ef9 | 2748 | INIT_LIST_HEAD(&device->peer_devices); |
4ce49266 | 2749 | INIT_LIST_HEAD(&device->pending_bitmap_io); |
b6f85ef9 AG |
2750 | for_each_connection(connection, resource) { |
2751 | peer_device = kzalloc(sizeof(struct drbd_peer_device), GFP_KERNEL); | |
2752 | if (!peer_device) | |
2753 | goto out_idr_remove_from_resource; | |
2754 | peer_device->connection = connection; | |
2755 | peer_device->device = device; | |
2756 | ||
2757 | list_add(&peer_device->peer_devices, &device->peer_devices); | |
2758 | kref_get(&device->kref); | |
2759 | ||
2760 | id = idr_alloc(&connection->peer_devices, peer_device, vnr, vnr + 1, GFP_KERNEL); | |
2761 | if (id < 0) { | |
f221f4bc | 2762 | if (id == -ENOSPC) |
b6f85ef9 | 2763 | err = ERR_INVALID_REQUEST; |
b6f85ef9 | 2764 | goto out_idr_remove_from_resource; |
56de2102 | 2765 | } |
b6f85ef9 | 2766 | kref_get(&connection->kref); |
668700b4 | 2767 | INIT_WORK(&peer_device->send_acks_work, drbd_send_acks_wf); |
8432b314 | 2768 | } |
56de2102 | 2769 | |
b30ab791 | 2770 | if (init_submitter(device)) { |
113fef9e | 2771 | err = ERR_NOMEM; |
ae4d37b5 | 2772 | goto out_idr_remove_from_resource; |
113fef9e LE |
2773 | } |
2774 | ||
e92ab4ed LC |
2775 | err = add_disk(disk); |
2776 | if (err) | |
ae4d37b5 | 2777 | goto out_idr_remove_from_resource; |
774b3055 | 2778 | |
2325eb66 | 2779 | /* inherit the connection state */ |
b6f85ef9 | 2780 | device->state.conn = first_connection(resource)->cstate; |
69a22773 AG |
2781 | if (device->state.conn == C_WF_REPORT_PARAMS) { |
2782 | for_each_peer_device(peer_device, device) | |
2783 | drbd_connected(peer_device); | |
2784 | } | |
4d3d5aa8 LE |
2785 | /* move to create_peer_device() */ |
2786 | for_each_peer_device(peer_device, device) | |
2787 | drbd_debugfs_peer_device_add(peer_device); | |
2788 | drbd_debugfs_device_add(device); | |
774b3055 | 2789 | return NO_ERROR; |
b411b363 | 2790 | |
803ea134 | 2791 | out_idr_remove_from_resource: |
a7a15981 | 2792 | for_each_connection_safe(connection, n, resource) { |
d3e709e6 MW |
2793 | peer_device = idr_remove(&connection->peer_devices, vnr); |
2794 | if (peer_device) | |
b6f85ef9 | 2795 | kref_put(&connection->kref, drbd_destroy_connection); |
b6f85ef9 AG |
2796 | } |
2797 | for_each_peer_device_safe(peer_device, tmp_peer_device, device) { | |
2798 | list_del(&peer_device->peer_devices); | |
2799 | kfree(peer_device); | |
2800 | } | |
803ea134 | 2801 | idr_remove(&resource->devices, vnr); |
8432b314 | 2802 | out_idr_remove_minor: |
93e4bf7a | 2803 | idr_remove(&drbd_devices, minor); |
569083c0 | 2804 | synchronize_rcu(); |
8432b314 | 2805 | out_no_minor_idr: |
b30ab791 | 2806 | drbd_bm_cleanup(device); |
b411b363 | 2807 | out_no_bitmap: |
e37d2438 | 2808 | __free_page(device->md_io.page); |
b411b363 | 2809 | out_no_io_page: |
8b9ab626 | 2810 | put_disk(disk); |
b411b363 | 2811 | out_no_disk: |
803ea134 | 2812 | kref_put(&resource->kref, drbd_destroy_resource); |
a6b32bc3 | 2813 | kfree(device); |
8432b314 | 2814 | return err; |
b411b363 PR |
2815 | } |
2816 | ||
f82795d6 | 2817 | void drbd_delete_device(struct drbd_device *device) |
803ea134 AG |
2818 | { |
2819 | struct drbd_resource *resource = device->resource; | |
2820 | struct drbd_connection *connection; | |
4d3d5aa8 | 2821 | struct drbd_peer_device *peer_device; |
803ea134 | 2822 | |
4d3d5aa8 LE |
2823 | /* move to free_peer_device() */ |
2824 | for_each_peer_device(peer_device, device) | |
2825 | drbd_debugfs_peer_device_cleanup(peer_device); | |
2826 | drbd_debugfs_device_cleanup(device); | |
803ea134 | 2827 | for_each_connection(connection, resource) { |
c06ece6b | 2828 | idr_remove(&connection->peer_devices, device->vnr); |
bdfafc4f | 2829 | kref_put(&device->kref, drbd_destroy_device); |
803ea134 AG |
2830 | } |
2831 | idr_remove(&resource->devices, device->vnr); | |
bdfafc4f | 2832 | kref_put(&device->kref, drbd_destroy_device); |
803ea134 | 2833 | idr_remove(&drbd_devices, device_to_minor(device)); |
bdfafc4f | 2834 | kref_put(&device->kref, drbd_destroy_device); |
803ea134 AG |
2835 | del_gendisk(device->vdisk); |
2836 | synchronize_rcu(); | |
bdfafc4f | 2837 | kref_put(&device->kref, drbd_destroy_device); |
803ea134 AG |
2838 | } |
2839 | ||
8ce953aa | 2840 | static int __init drbd_init(void) |
b411b363 PR |
2841 | { |
2842 | int err; | |
2843 | ||
183ece30 RK |
2844 | if (drbd_minor_count < DRBD_MINOR_COUNT_MIN || drbd_minor_count > DRBD_MINOR_COUNT_MAX) { |
2845 | pr_err("invalid minor_count (%d)\n", drbd_minor_count); | |
b411b363 PR |
2846 | #ifdef MODULE |
2847 | return -EINVAL; | |
2848 | #else | |
5fc1efd5 | 2849 | drbd_minor_count = DRBD_MINOR_COUNT_DEF; |
b411b363 PR |
2850 | #endif |
2851 | } | |
2852 | ||
b411b363 PR |
2853 | err = register_blkdev(DRBD_MAJOR, "drbd"); |
2854 | if (err) { | |
f88c5d90 | 2855 | pr_err("unable to register block device major %d\n", |
b411b363 PR |
2856 | DRBD_MAJOR); |
2857 | return err; | |
2858 | } | |
2859 | ||
b411b363 PR |
2860 | /* |
2861 | * allocate all necessary structs | |
2862 | */ | |
b411b363 PR |
2863 | init_waitqueue_head(&drbd_pp_wait); |
2864 | ||
2865 | drbd_proc = NULL; /* play safe for drbd_cleanup */ | |
05a10ec7 | 2866 | idr_init(&drbd_devices); |
b411b363 | 2867 | |
28bc3b8c | 2868 | mutex_init(&resources_mutex); |
77c556f6 | 2869 | INIT_LIST_HEAD(&drbd_resources); |
69babf05 LE |
2870 | |
2871 | err = drbd_genl_register(); | |
2872 | if (err) { | |
f88c5d90 | 2873 | pr_err("unable to register generic netlink family\n"); |
69babf05 LE |
2874 | goto fail; |
2875 | } | |
2876 | ||
b411b363 PR |
2877 | err = drbd_create_mempools(); |
2878 | if (err) | |
3b98c0c2 | 2879 | goto fail; |
b411b363 | 2880 | |
6110d70b | 2881 | err = -ENOMEM; |
cf626b0d | 2882 | drbd_proc = proc_create_single("drbd", S_IFREG | 0444 , NULL, drbd_seq_show); |
b411b363 | 2883 | if (!drbd_proc) { |
f88c5d90 | 2884 | pr_err("unable to register proc file\n"); |
3b98c0c2 | 2885 | goto fail; |
b411b363 PR |
2886 | } |
2887 | ||
2312f0b3 LE |
2888 | retry.wq = create_singlethread_workqueue("drbd-reissue"); |
2889 | if (!retry.wq) { | |
f88c5d90 | 2890 | pr_err("unable to create retry workqueue\n"); |
2312f0b3 LE |
2891 | goto fail; |
2892 | } | |
2893 | INIT_WORK(&retry.worker, do_retry); | |
2894 | spin_lock_init(&retry.lock); | |
2895 | INIT_LIST_HEAD(&retry.writes); | |
b411b363 | 2896 | |
d27e84a3 | 2897 | drbd_debugfs_init(); |
4d3d5aa8 | 2898 | |
f88c5d90 | 2899 | pr_info("initialized. " |
b411b363 PR |
2900 | "Version: " REL_VERSION " (api:%d/proto:%d-%d)\n", |
2901 | API_VERSION, PRO_VERSION_MIN, PRO_VERSION_MAX); | |
f88c5d90 LE |
2902 | pr_info("%s\n", drbd_buildtag()); |
2903 | pr_info("registered as block device major %d\n", DRBD_MAJOR); | |
b411b363 PR |
2904 | return 0; /* Success! */ |
2905 | ||
3b98c0c2 | 2906 | fail: |
b411b363 PR |
2907 | drbd_cleanup(); |
2908 | if (err == -ENOMEM) | |
f88c5d90 | 2909 | pr_err("ran out of memory\n"); |
b411b363 | 2910 | else |
f88c5d90 | 2911 | pr_err("initialization failure\n"); |
b411b363 PR |
2912 | return err; |
2913 | } | |
2914 | ||
f418815f | 2915 | static void drbd_free_one_sock(struct drbd_socket *ds) |
b411b363 | 2916 | { |
f418815f LE |
2917 | struct socket *s; |
2918 | mutex_lock(&ds->mutex); | |
2919 | s = ds->socket; | |
2920 | ds->socket = NULL; | |
2921 | mutex_unlock(&ds->mutex); | |
2922 | if (s) { | |
2923 | /* so debugfs does not need to mutex_lock() */ | |
2924 | synchronize_rcu(); | |
2925 | kernel_sock_shutdown(s, SHUT_RDWR); | |
2926 | sock_release(s); | |
b411b363 PR |
2927 | } |
2928 | } | |
2929 | ||
f418815f LE |
2930 | void drbd_free_sock(struct drbd_connection *connection) |
2931 | { | |
2932 | if (connection->data.socket) | |
2933 | drbd_free_one_sock(&connection->data); | |
2934 | if (connection->meta.socket) | |
2935 | drbd_free_one_sock(&connection->meta); | |
2936 | } | |
2937 | ||
b411b363 | 2938 | /* meta data management */ |
b411b363 | 2939 | |
bde89a9e | 2940 | void conn_md_sync(struct drbd_connection *connection) |
b411b363 | 2941 | { |
c06ece6b | 2942 | struct drbd_peer_device *peer_device; |
19fffd7b | 2943 | int vnr; |
b411b363 | 2944 | |
19fffd7b | 2945 | rcu_read_lock(); |
c06ece6b AG |
2946 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
2947 | struct drbd_device *device = peer_device->device; | |
2948 | ||
b30ab791 | 2949 | kref_get(&device->kref); |
19fffd7b | 2950 | rcu_read_unlock(); |
b30ab791 | 2951 | drbd_md_sync(device); |
05a10ec7 | 2952 | kref_put(&device->kref, drbd_destroy_device); |
19fffd7b PR |
2953 | rcu_read_lock(); |
2954 | } | |
2955 | rcu_read_unlock(); | |
b411b363 PR |
2956 | } |
2957 | ||
ae8bf312 | 2958 | /* aligned 4kByte */ |
b411b363 | 2959 | struct meta_data_on_disk { |
cccac985 | 2960 | u64 la_size_sect; /* last agreed size. */ |
b411b363 PR |
2961 | u64 uuid[UI_SIZE]; /* UUIDs. */ |
2962 | u64 device_uuid; | |
2963 | u64 reserved_u64_1; | |
2964 | u32 flags; /* MDF */ | |
2965 | u32 magic; | |
2966 | u32 md_size_sect; | |
2967 | u32 al_offset; /* offset to this block */ | |
ae8bf312 | 2968 | u32 al_nr_extents; /* important for restoring the AL (userspace) */ |
f399002e | 2969 | /* `-- act_log->nr_elements <-- ldev->dc.al_extents */ |
b411b363 PR |
2970 | u32 bm_offset; /* offset to the bitmap, from here */ |
2971 | u32 bm_bytes_per_bit; /* BM_BLOCK_SIZE */ | |
99432fcc | 2972 | u32 la_peer_max_bio_size; /* last peer max_bio_size */ |
b411b363 | 2973 | |
3a4d4eb3 LE |
2974 | /* see al_tr_number_to_on_disk_sector() */ |
2975 | u32 al_stripes; | |
2976 | u32 al_stripe_size_4k; | |
2977 | ||
2978 | u8 reserved_u8[4096 - (7*8 + 10*4)]; | |
b411b363 PR |
2979 | } __packed; |
2980 | ||
d752b269 PR |
2981 | |
2982 | ||
b30ab791 | 2983 | void drbd_md_write(struct drbd_device *device, void *b) |
b411b363 | 2984 | { |
d752b269 | 2985 | struct meta_data_on_disk *buffer = b; |
b411b363 PR |
2986 | sector_t sector; |
2987 | int i; | |
2988 | ||
ae8bf312 | 2989 | memset(buffer, 0, sizeof(*buffer)); |
b411b363 | 2990 | |
155bd9d1 | 2991 | buffer->la_size_sect = cpu_to_be64(get_capacity(device->vdisk)); |
b411b363 | 2992 | for (i = UI_CURRENT; i < UI_SIZE; i++) |
b30ab791 AG |
2993 | buffer->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]); |
2994 | buffer->flags = cpu_to_be32(device->ldev->md.flags); | |
d5d7ebd4 | 2995 | buffer->magic = cpu_to_be32(DRBD_MD_MAGIC_84_UNCLEAN); |
b411b363 | 2996 | |
b30ab791 AG |
2997 | buffer->md_size_sect = cpu_to_be32(device->ldev->md.md_size_sect); |
2998 | buffer->al_offset = cpu_to_be32(device->ldev->md.al_offset); | |
2999 | buffer->al_nr_extents = cpu_to_be32(device->act_log->nr_elements); | |
b411b363 | 3000 | buffer->bm_bytes_per_bit = cpu_to_be32(BM_BLOCK_SIZE); |
b30ab791 | 3001 | buffer->device_uuid = cpu_to_be64(device->ldev->md.device_uuid); |
b411b363 | 3002 | |
b30ab791 AG |
3003 | buffer->bm_offset = cpu_to_be32(device->ldev->md.bm_offset); |
3004 | buffer->la_peer_max_bio_size = cpu_to_be32(device->peer_max_bio_size); | |
b411b363 | 3005 | |
b30ab791 AG |
3006 | buffer->al_stripes = cpu_to_be32(device->ldev->md.al_stripes); |
3007 | buffer->al_stripe_size_4k = cpu_to_be32(device->ldev->md.al_stripe_size_4k); | |
3a4d4eb3 | 3008 | |
0b0ba1ef | 3009 | D_ASSERT(device, drbd_md_ss(device->ldev) == device->ldev->md.md_offset); |
b30ab791 | 3010 | sector = device->ldev->md.md_offset; |
b411b363 | 3011 | |
bb3cc85e | 3012 | if (drbd_md_sync_page_io(device, device->ldev, sector, REQ_OP_WRITE)) { |
b411b363 | 3013 | /* this was a try anyways ... */ |
d0180171 | 3014 | drbd_err(device, "meta data update failed!\n"); |
b30ab791 | 3015 | drbd_chk_io_error(device, 1, DRBD_META_IO_ERROR); |
b411b363 | 3016 | } |
d752b269 PR |
3017 | } |
3018 | ||
3019 | /** | |
3020 | * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set | |
b30ab791 | 3021 | * @device: DRBD device. |
d752b269 | 3022 | */ |
b30ab791 | 3023 | void drbd_md_sync(struct drbd_device *device) |
d752b269 PR |
3024 | { |
3025 | struct meta_data_on_disk *buffer; | |
3026 | ||
3027 | /* Don't accidentally change the DRBD meta data layout. */ | |
3028 | BUILD_BUG_ON(UI_SIZE != 4); | |
3029 | BUILD_BUG_ON(sizeof(struct meta_data_on_disk) != 4096); | |
3030 | ||
b30ab791 | 3031 | del_timer(&device->md_sync_timer); |
d752b269 | 3032 | /* timer may be rearmed by drbd_md_mark_dirty() now. */ |
b30ab791 | 3033 | if (!test_and_clear_bit(MD_DIRTY, &device->flags)) |
d752b269 PR |
3034 | return; |
3035 | ||
3036 | /* We use here D_FAILED and not D_ATTACHING because we try to write | |
3037 | * metadata even if we detach due to a disk failure! */ | |
b30ab791 | 3038 | if (!get_ldev_if_state(device, D_FAILED)) |
d752b269 PR |
3039 | return; |
3040 | ||
e37d2438 | 3041 | buffer = drbd_md_get_buffer(device, __func__); |
d752b269 PR |
3042 | if (!buffer) |
3043 | goto out; | |
3044 | ||
b30ab791 | 3045 | drbd_md_write(device, buffer); |
b411b363 | 3046 | |
b30ab791 | 3047 | /* Update device->ldev->md.la_size_sect, |
b411b363 | 3048 | * since we updated it on metadata. */ |
155bd9d1 | 3049 | device->ldev->md.la_size_sect = get_capacity(device->vdisk); |
b411b363 | 3050 | |
b30ab791 | 3051 | drbd_md_put_buffer(device); |
e1711731 | 3052 | out: |
b30ab791 | 3053 | put_ldev(device); |
b411b363 PR |
3054 | } |
3055 | ||
b30ab791 | 3056 | static int check_activity_log_stripe_size(struct drbd_device *device, |
3a4d4eb3 LE |
3057 | struct meta_data_on_disk *on_disk, |
3058 | struct drbd_md *in_core) | |
3059 | { | |
3060 | u32 al_stripes = be32_to_cpu(on_disk->al_stripes); | |
3061 | u32 al_stripe_size_4k = be32_to_cpu(on_disk->al_stripe_size_4k); | |
3062 | u64 al_size_4k; | |
3063 | ||
3064 | /* both not set: default to old fixed size activity log */ | |
3065 | if (al_stripes == 0 && al_stripe_size_4k == 0) { | |
3066 | al_stripes = 1; | |
3067 | al_stripe_size_4k = MD_32kB_SECT/8; | |
3068 | } | |
3069 | ||
3070 | /* some paranoia plausibility checks */ | |
3071 | ||
3072 | /* we need both values to be set */ | |
3073 | if (al_stripes == 0 || al_stripe_size_4k == 0) | |
3074 | goto err; | |
3075 | ||
3076 | al_size_4k = (u64)al_stripes * al_stripe_size_4k; | |
3077 | ||
3078 | /* Upper limit of activity log area, to avoid potential overflow | |
3079 | * problems in al_tr_number_to_on_disk_sector(). As right now, more | |
3080 | * than 72 * 4k blocks total only increases the amount of history, | |
3081 | * limiting this arbitrarily to 16 GB is not a real limitation ;-) */ | |
3082 | if (al_size_4k > (16 * 1024 * 1024/4)) | |
3083 | goto err; | |
3084 | ||
3085 | /* Lower limit: we need at least 8 transaction slots (32kB) | |
3086 | * to not break existing setups */ | |
3087 | if (al_size_4k < MD_32kB_SECT/8) | |
3088 | goto err; | |
3089 | ||
3090 | in_core->al_stripe_size_4k = al_stripe_size_4k; | |
3091 | in_core->al_stripes = al_stripes; | |
3092 | in_core->al_size_4k = al_size_4k; | |
3093 | ||
3094 | return 0; | |
3095 | err: | |
d0180171 | 3096 | drbd_err(device, "invalid activity log striping: al_stripes=%u, al_stripe_size_4k=%u\n", |
3a4d4eb3 LE |
3097 | al_stripes, al_stripe_size_4k); |
3098 | return -EINVAL; | |
3099 | } | |
3100 | ||
b30ab791 | 3101 | static int check_offsets_and_sizes(struct drbd_device *device, struct drbd_backing_dev *bdev) |
c04ccaa6 LE |
3102 | { |
3103 | sector_t capacity = drbd_get_capacity(bdev->md_bdev); | |
3104 | struct drbd_md *in_core = &bdev->md; | |
3105 | s32 on_disk_al_sect; | |
3106 | s32 on_disk_bm_sect; | |
3107 | ||
3108 | /* The on-disk size of the activity log, calculated from offsets, and | |
3109 | * the size of the activity log calculated from the stripe settings, | |
3110 | * should match. | |
3111 | * Though we could relax this a bit: it is ok, if the striped activity log | |
3112 | * fits in the available on-disk activity log size. | |
3113 | * Right now, that would break how resize is implemented. | |
3114 | * TODO: make drbd_determine_dev_size() (and the drbdmeta tool) aware | |
3115 | * of possible unused padding space in the on disk layout. */ | |
3116 | if (in_core->al_offset < 0) { | |
3117 | if (in_core->bm_offset > in_core->al_offset) | |
3118 | goto err; | |
3119 | on_disk_al_sect = -in_core->al_offset; | |
3120 | on_disk_bm_sect = in_core->al_offset - in_core->bm_offset; | |
3121 | } else { | |
3122 | if (in_core->al_offset != MD_4kB_SECT) | |
3123 | goto err; | |
3124 | if (in_core->bm_offset < in_core->al_offset + in_core->al_size_4k * MD_4kB_SECT) | |
3125 | goto err; | |
3126 | ||
3127 | on_disk_al_sect = in_core->bm_offset - MD_4kB_SECT; | |
3128 | on_disk_bm_sect = in_core->md_size_sect - in_core->bm_offset; | |
3129 | } | |
3130 | ||
3131 | /* old fixed size meta data is exactly that: fixed. */ | |
3132 | if (in_core->meta_dev_idx >= 0) { | |
3133 | if (in_core->md_size_sect != MD_128MB_SECT | |
3134 | || in_core->al_offset != MD_4kB_SECT | |
3135 | || in_core->bm_offset != MD_4kB_SECT + MD_32kB_SECT | |
3136 | || in_core->al_stripes != 1 | |
3137 | || in_core->al_stripe_size_4k != MD_32kB_SECT/8) | |
3138 | goto err; | |
3139 | } | |
3140 | ||
3141 | if (capacity < in_core->md_size_sect) | |
3142 | goto err; | |
3143 | if (capacity - in_core->md_size_sect < drbd_md_first_sector(bdev)) | |
3144 | goto err; | |
3145 | ||
3146 | /* should be aligned, and at least 32k */ | |
3147 | if ((on_disk_al_sect & 7) || (on_disk_al_sect < MD_32kB_SECT)) | |
3148 | goto err; | |
3149 | ||
3150 | /* should fit (for now: exactly) into the available on-disk space; | |
3151 | * overflow prevention is in check_activity_log_stripe_size() above. */ | |
3152 | if (on_disk_al_sect != in_core->al_size_4k * MD_4kB_SECT) | |
3153 | goto err; | |
3154 | ||
3155 | /* again, should be aligned */ | |
3156 | if (in_core->bm_offset & 7) | |
3157 | goto err; | |
3158 | ||
3159 | /* FIXME check for device grow with flex external meta data? */ | |
3160 | ||
3161 | /* can the available bitmap space cover the last agreed device size? */ | |
3162 | if (on_disk_bm_sect < (in_core->la_size_sect+7)/MD_4kB_SECT/8/512) | |
3163 | goto err; | |
3164 | ||
3165 | return 0; | |
3166 | ||
3167 | err: | |
d0180171 | 3168 | drbd_err(device, "meta data offsets don't make sense: idx=%d " |
c04ccaa6 LE |
3169 | "al_s=%u, al_sz4k=%u, al_offset=%d, bm_offset=%d, " |
3170 | "md_size_sect=%u, la_size=%llu, md_capacity=%llu\n", | |
3171 | in_core->meta_dev_idx, | |
3172 | in_core->al_stripes, in_core->al_stripe_size_4k, | |
3173 | in_core->al_offset, in_core->bm_offset, in_core->md_size_sect, | |
3174 | (unsigned long long)in_core->la_size_sect, | |
3175 | (unsigned long long)capacity); | |
3176 | ||
3177 | return -EINVAL; | |
3178 | } | |
3179 | ||
3180 | ||
b411b363 PR |
3181 | /** |
3182 | * drbd_md_read() - Reads in the meta data super block | |
b30ab791 | 3183 | * @device: DRBD device. |
b411b363 PR |
3184 | * @bdev: Device from which the meta data should be read in. |
3185 | * | |
3a4d4eb3 | 3186 | * Return NO_ERROR on success, and an enum drbd_ret_code in case |
d5d7ebd4 | 3187 | * something goes wrong. |
3a4d4eb3 | 3188 | * |
c04ccaa6 | 3189 | * Called exactly once during drbd_adm_attach(), while still being D_DISKLESS, |
b30ab791 | 3190 | * even before @bdev is assigned to @device->ldev. |
b411b363 | 3191 | */ |
b30ab791 | 3192 | int drbd_md_read(struct drbd_device *device, struct drbd_backing_dev *bdev) |
b411b363 PR |
3193 | { |
3194 | struct meta_data_on_disk *buffer; | |
d5d7ebd4 | 3195 | u32 magic, flags; |
b411b363 PR |
3196 | int i, rv = NO_ERROR; |
3197 | ||
b30ab791 | 3198 | if (device->state.disk != D_DISKLESS) |
c04ccaa6 | 3199 | return ERR_DISK_CONFIGURED; |
b411b363 | 3200 | |
e37d2438 | 3201 | buffer = drbd_md_get_buffer(device, __func__); |
e1711731 | 3202 | if (!buffer) |
c04ccaa6 | 3203 | return ERR_NOMEM; |
b411b363 | 3204 | |
c04ccaa6 LE |
3205 | /* First, figure out where our meta data superblock is located, |
3206 | * and read it. */ | |
3a4d4eb3 LE |
3207 | bdev->md.meta_dev_idx = bdev->disk_conf->meta_dev_idx; |
3208 | bdev->md.md_offset = drbd_md_ss(bdev); | |
edb5e5f6 LE |
3209 | /* Even for (flexible or indexed) external meta data, |
3210 | * initially restrict us to the 4k superblock for now. | |
3211 | * Affects the paranoia out-of-range access check in drbd_md_sync_page_io(). */ | |
3212 | bdev->md.md_size_sect = 8; | |
b411b363 | 3213 | |
bb3cc85e MC |
3214 | if (drbd_md_sync_page_io(device, bdev, bdev->md.md_offset, |
3215 | REQ_OP_READ)) { | |
25985edc | 3216 | /* NOTE: can't do normal error processing here as this is |
b411b363 | 3217 | called BEFORE disk is attached */ |
d0180171 | 3218 | drbd_err(device, "Error while reading metadata.\n"); |
b411b363 PR |
3219 | rv = ERR_IO_MD_DISK; |
3220 | goto err; | |
3221 | } | |
3222 | ||
d5d7ebd4 LE |
3223 | magic = be32_to_cpu(buffer->magic); |
3224 | flags = be32_to_cpu(buffer->flags); | |
3225 | if (magic == DRBD_MD_MAGIC_84_UNCLEAN || | |
3226 | (magic == DRBD_MD_MAGIC_08 && !(flags & MDF_AL_CLEAN))) { | |
3227 | /* btw: that's Activity Log clean, not "all" clean. */ | |
d0180171 | 3228 | drbd_err(device, "Found unclean meta data. Did you \"drbdadm apply-al\"?\n"); |
d5d7ebd4 LE |
3229 | rv = ERR_MD_UNCLEAN; |
3230 | goto err; | |
3231 | } | |
3a4d4eb3 LE |
3232 | |
3233 | rv = ERR_MD_INVALID; | |
d5d7ebd4 | 3234 | if (magic != DRBD_MD_MAGIC_08) { |
43de7c85 | 3235 | if (magic == DRBD_MD_MAGIC_07) |
d0180171 | 3236 | drbd_err(device, "Found old (0.7) meta data magic. Did you \"drbdadm create-md\"?\n"); |
d5d7ebd4 | 3237 | else |
d0180171 | 3238 | drbd_err(device, "Meta data magic not found. Did you \"drbdadm create-md\"?\n"); |
b411b363 PR |
3239 | goto err; |
3240 | } | |
3a4d4eb3 | 3241 | |
c04ccaa6 | 3242 | if (be32_to_cpu(buffer->bm_bytes_per_bit) != BM_BLOCK_SIZE) { |
d0180171 | 3243 | drbd_err(device, "unexpected bm_bytes_per_bit: %u (expected %u)\n", |
c04ccaa6 | 3244 | be32_to_cpu(buffer->bm_bytes_per_bit), BM_BLOCK_SIZE); |
b411b363 PR |
3245 | goto err; |
3246 | } | |
3a4d4eb3 | 3247 | |
c04ccaa6 LE |
3248 | |
3249 | /* convert to in_core endian */ | |
3250 | bdev->md.la_size_sect = be64_to_cpu(buffer->la_size_sect); | |
3251 | for (i = UI_CURRENT; i < UI_SIZE; i++) | |
3252 | bdev->md.uuid[i] = be64_to_cpu(buffer->uuid[i]); | |
3253 | bdev->md.flags = be32_to_cpu(buffer->flags); | |
3254 | bdev->md.device_uuid = be64_to_cpu(buffer->device_uuid); | |
3255 | ||
3256 | bdev->md.md_size_sect = be32_to_cpu(buffer->md_size_sect); | |
3257 | bdev->md.al_offset = be32_to_cpu(buffer->al_offset); | |
3258 | bdev->md.bm_offset = be32_to_cpu(buffer->bm_offset); | |
3259 | ||
b30ab791 | 3260 | if (check_activity_log_stripe_size(device, buffer, &bdev->md)) |
b411b363 | 3261 | goto err; |
b30ab791 | 3262 | if (check_offsets_and_sizes(device, bdev)) |
c04ccaa6 LE |
3263 | goto err; |
3264 | ||
b411b363 | 3265 | if (be32_to_cpu(buffer->bm_offset) != bdev->md.bm_offset) { |
d0180171 | 3266 | drbd_err(device, "unexpected bm_offset: %d (expected %d)\n", |
b411b363 | 3267 | be32_to_cpu(buffer->bm_offset), bdev->md.bm_offset); |
b411b363 PR |
3268 | goto err; |
3269 | } | |
3270 | if (be32_to_cpu(buffer->md_size_sect) != bdev->md.md_size_sect) { | |
d0180171 | 3271 | drbd_err(device, "unexpected md_size: %u (expected %u)\n", |
b411b363 | 3272 | be32_to_cpu(buffer->md_size_sect), bdev->md.md_size_sect); |
b411b363 PR |
3273 | goto err; |
3274 | } | |
3275 | ||
3a4d4eb3 | 3276 | rv = NO_ERROR; |
b411b363 | 3277 | |
0500813f | 3278 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 3279 | if (device->state.conn < C_CONNECTED) { |
db141b2f | 3280 | unsigned int peer; |
99432fcc | 3281 | peer = be32_to_cpu(buffer->la_peer_max_bio_size); |
db141b2f | 3282 | peer = max(peer, DRBD_MAX_BIO_SIZE_SAFE); |
b30ab791 | 3283 | device->peer_max_bio_size = peer; |
99432fcc | 3284 | } |
0500813f | 3285 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 PR |
3286 | |
3287 | err: | |
b30ab791 | 3288 | drbd_md_put_buffer(device); |
b411b363 PR |
3289 | |
3290 | return rv; | |
3291 | } | |
3292 | ||
3293 | /** | |
3294 | * drbd_md_mark_dirty() - Mark meta data super block as dirty | |
b30ab791 | 3295 | * @device: DRBD device. |
b411b363 PR |
3296 | * |
3297 | * Call this function if you change anything that should be written to | |
3298 | * the meta-data super block. This function sets MD_DIRTY, and starts a | |
3299 | * timer that ensures that within five seconds you have to call drbd_md_sync(). | |
3300 | */ | |
b30ab791 | 3301 | void drbd_md_mark_dirty(struct drbd_device *device) |
b411b363 | 3302 | { |
b30ab791 AG |
3303 | if (!test_and_set_bit(MD_DIRTY, &device->flags)) |
3304 | mod_timer(&device->md_sync_timer, jiffies + 5*HZ); | |
b411b363 | 3305 | } |
b411b363 | 3306 | |
b30ab791 | 3307 | void drbd_uuid_move_history(struct drbd_device *device) __must_hold(local) |
b411b363 PR |
3308 | { |
3309 | int i; | |
3310 | ||
62b0da3a | 3311 | for (i = UI_HISTORY_START; i < UI_HISTORY_END; i++) |
b30ab791 | 3312 | device->ldev->md.uuid[i+1] = device->ldev->md.uuid[i]; |
b411b363 PR |
3313 | } |
3314 | ||
b30ab791 | 3315 | void __drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local) |
b411b363 PR |
3316 | { |
3317 | if (idx == UI_CURRENT) { | |
b30ab791 | 3318 | if (device->state.role == R_PRIMARY) |
b411b363 PR |
3319 | val |= 1; |
3320 | else | |
3321 | val &= ~((u64)1); | |
3322 | ||
b30ab791 | 3323 | drbd_set_ed_uuid(device, val); |
b411b363 PR |
3324 | } |
3325 | ||
b30ab791 AG |
3326 | device->ldev->md.uuid[idx] = val; |
3327 | drbd_md_mark_dirty(device); | |
b411b363 PR |
3328 | } |
3329 | ||
b30ab791 | 3330 | void _drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local) |
9f2247bb PR |
3331 | { |
3332 | unsigned long flags; | |
b30ab791 AG |
3333 | spin_lock_irqsave(&device->ldev->md.uuid_lock, flags); |
3334 | __drbd_uuid_set(device, idx, val); | |
3335 | spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags); | |
9f2247bb | 3336 | } |
b411b363 | 3337 | |
b30ab791 | 3338 | void drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local) |
b411b363 | 3339 | { |
9f2247bb | 3340 | unsigned long flags; |
b30ab791 AG |
3341 | spin_lock_irqsave(&device->ldev->md.uuid_lock, flags); |
3342 | if (device->ldev->md.uuid[idx]) { | |
3343 | drbd_uuid_move_history(device); | |
3344 | device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[idx]; | |
b411b363 | 3345 | } |
b30ab791 AG |
3346 | __drbd_uuid_set(device, idx, val); |
3347 | spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags); | |
b411b363 PR |
3348 | } |
3349 | ||
3350 | /** | |
3351 | * drbd_uuid_new_current() - Creates a new current UUID | |
b30ab791 | 3352 | * @device: DRBD device. |
b411b363 PR |
3353 | * |
3354 | * Creates a new current UUID, and rotates the old current UUID into | |
3355 | * the bitmap slot. Causes an incremental resync upon next connect. | |
3356 | */ | |
b30ab791 | 3357 | void drbd_uuid_new_current(struct drbd_device *device) __must_hold(local) |
b411b363 PR |
3358 | { |
3359 | u64 val; | |
9f2247bb PR |
3360 | unsigned long long bm_uuid; |
3361 | ||
3362 | get_random_bytes(&val, sizeof(u64)); | |
3363 | ||
b30ab791 AG |
3364 | spin_lock_irq(&device->ldev->md.uuid_lock); |
3365 | bm_uuid = device->ldev->md.uuid[UI_BITMAP]; | |
62b0da3a LE |
3366 | |
3367 | if (bm_uuid) | |
d0180171 | 3368 | drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid); |
b411b363 | 3369 | |
b30ab791 AG |
3370 | device->ldev->md.uuid[UI_BITMAP] = device->ldev->md.uuid[UI_CURRENT]; |
3371 | __drbd_uuid_set(device, UI_CURRENT, val); | |
3372 | spin_unlock_irq(&device->ldev->md.uuid_lock); | |
b411b363 | 3373 | |
b30ab791 | 3374 | drbd_print_uuids(device, "new current UUID"); |
aaa8e2b3 | 3375 | /* get it to stable storage _now_ */ |
b30ab791 | 3376 | drbd_md_sync(device); |
b411b363 PR |
3377 | } |
3378 | ||
b30ab791 | 3379 | void drbd_uuid_set_bm(struct drbd_device *device, u64 val) __must_hold(local) |
b411b363 | 3380 | { |
9f2247bb | 3381 | unsigned long flags; |
b30ab791 | 3382 | if (device->ldev->md.uuid[UI_BITMAP] == 0 && val == 0) |
b411b363 PR |
3383 | return; |
3384 | ||
b30ab791 | 3385 | spin_lock_irqsave(&device->ldev->md.uuid_lock, flags); |
b411b363 | 3386 | if (val == 0) { |
b30ab791 AG |
3387 | drbd_uuid_move_history(device); |
3388 | device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[UI_BITMAP]; | |
3389 | device->ldev->md.uuid[UI_BITMAP] = 0; | |
b411b363 | 3390 | } else { |
b30ab791 | 3391 | unsigned long long bm_uuid = device->ldev->md.uuid[UI_BITMAP]; |
62b0da3a | 3392 | if (bm_uuid) |
d0180171 | 3393 | drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid); |
b411b363 | 3394 | |
b30ab791 | 3395 | device->ldev->md.uuid[UI_BITMAP] = val & ~((u64)1); |
b411b363 | 3396 | } |
b30ab791 | 3397 | spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags); |
9f2247bb | 3398 | |
b30ab791 | 3399 | drbd_md_mark_dirty(device); |
b411b363 PR |
3400 | } |
3401 | ||
3402 | /** | |
3403 | * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io() | |
b30ab791 | 3404 | * @device: DRBD device. |
b411b363 PR |
3405 | * |
3406 | * Sets all bits in the bitmap and writes the whole bitmap to stable storage. | |
3407 | */ | |
8fe39aac | 3408 | int drbd_bmio_set_n_write(struct drbd_device *device) __must_hold(local) |
b411b363 PR |
3409 | { |
3410 | int rv = -EIO; | |
3411 | ||
8fe39aac PR |
3412 | drbd_md_set_flag(device, MDF_FULL_SYNC); |
3413 | drbd_md_sync(device); | |
3414 | drbd_bm_set_all(device); | |
b411b363 | 3415 | |
8fe39aac | 3416 | rv = drbd_bm_write(device); |
b411b363 | 3417 | |
8fe39aac PR |
3418 | if (!rv) { |
3419 | drbd_md_clear_flag(device, MDF_FULL_SYNC); | |
3420 | drbd_md_sync(device); | |
b411b363 PR |
3421 | } |
3422 | ||
3423 | return rv; | |
3424 | } | |
3425 | ||
3426 | /** | |
3427 | * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io() | |
b30ab791 | 3428 | * @device: DRBD device. |
b411b363 PR |
3429 | * |
3430 | * Clears all bits in the bitmap and writes the whole bitmap to stable storage. | |
3431 | */ | |
8fe39aac | 3432 | int drbd_bmio_clear_n_write(struct drbd_device *device) __must_hold(local) |
b411b363 | 3433 | { |
b30ab791 | 3434 | drbd_resume_al(device); |
8fe39aac PR |
3435 | drbd_bm_clear_all(device); |
3436 | return drbd_bm_write(device); | |
b411b363 PR |
3437 | } |
3438 | ||
99920dc5 | 3439 | static int w_bitmap_io(struct drbd_work *w, int unused) |
b411b363 | 3440 | { |
84b8c06b AG |
3441 | struct drbd_device *device = |
3442 | container_of(w, struct drbd_device, bm_io_work.w); | |
3443 | struct bm_io_work *work = &device->bm_io_work; | |
02851e9f | 3444 | int rv = -EIO; |
b411b363 | 3445 | |
bca1cbae LE |
3446 | if (work->flags != BM_LOCKED_CHANGE_ALLOWED) { |
3447 | int cnt = atomic_read(&device->ap_bio_cnt); | |
3448 | if (cnt) | |
3449 | drbd_err(device, "FIXME: ap_bio_cnt %d, expected 0; queued for '%s'\n", | |
3450 | cnt, work->why); | |
3451 | } | |
b411b363 | 3452 | |
b30ab791 AG |
3453 | if (get_ldev(device)) { |
3454 | drbd_bm_lock(device, work->why, work->flags); | |
3455 | rv = work->io_fn(device); | |
3456 | drbd_bm_unlock(device); | |
3457 | put_ldev(device); | |
02851e9f | 3458 | } |
b411b363 | 3459 | |
b30ab791 AG |
3460 | clear_bit_unlock(BITMAP_IO, &device->flags); |
3461 | wake_up(&device->misc_wait); | |
b411b363 PR |
3462 | |
3463 | if (work->done) | |
b30ab791 | 3464 | work->done(device, rv); |
b411b363 | 3465 | |
b30ab791 | 3466 | clear_bit(BITMAP_IO_QUEUED, &device->flags); |
b411b363 | 3467 | work->why = NULL; |
20ceb2b2 | 3468 | work->flags = 0; |
b411b363 | 3469 | |
99920dc5 | 3470 | return 0; |
b411b363 PR |
3471 | } |
3472 | ||
3473 | /** | |
3474 | * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap | |
b30ab791 | 3475 | * @device: DRBD device. |
b411b363 PR |
3476 | * @io_fn: IO callback to be called when bitmap IO is possible |
3477 | * @done: callback to be called after the bitmap IO was performed | |
3478 | * @why: Descriptive text of the reason for doing the IO | |
584164c8 | 3479 | * @flags: Bitmap flags |
b411b363 PR |
3480 | * |
3481 | * While IO on the bitmap happens we freeze application IO thus we ensure | |
3482 | * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be | |
3483 | * called from worker context. It MUST NOT be used while a previous such | |
3484 | * work is still pending! | |
8fe39aac PR |
3485 | * |
3486 | * Its worker function encloses the call of io_fn() by get_ldev() and | |
3487 | * put_ldev(). | |
b411b363 | 3488 | */ |
b30ab791 | 3489 | void drbd_queue_bitmap_io(struct drbd_device *device, |
54761697 AG |
3490 | int (*io_fn)(struct drbd_device *), |
3491 | void (*done)(struct drbd_device *, int), | |
20ceb2b2 | 3492 | char *why, enum bm_flag flags) |
b411b363 | 3493 | { |
0b0ba1ef | 3494 | D_ASSERT(device, current == first_peer_device(device)->connection->worker.task); |
b411b363 | 3495 | |
0b0ba1ef AG |
3496 | D_ASSERT(device, !test_bit(BITMAP_IO_QUEUED, &device->flags)); |
3497 | D_ASSERT(device, !test_bit(BITMAP_IO, &device->flags)); | |
3498 | D_ASSERT(device, list_empty(&device->bm_io_work.w.list)); | |
b30ab791 | 3499 | if (device->bm_io_work.why) |
d0180171 | 3500 | drbd_err(device, "FIXME going to queue '%s' but '%s' still pending?\n", |
b30ab791 | 3501 | why, device->bm_io_work.why); |
b411b363 | 3502 | |
b30ab791 AG |
3503 | device->bm_io_work.io_fn = io_fn; |
3504 | device->bm_io_work.done = done; | |
3505 | device->bm_io_work.why = why; | |
3506 | device->bm_io_work.flags = flags; | |
b411b363 | 3507 | |
0500813f | 3508 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 3509 | set_bit(BITMAP_IO, &device->flags); |
5bded4ef LE |
3510 | /* don't wait for pending application IO if the caller indicates that |
3511 | * application IO does not conflict anyways. */ | |
3512 | if (flags == BM_LOCKED_CHANGE_ALLOWED || atomic_read(&device->ap_bio_cnt) == 0) { | |
b30ab791 | 3513 | if (!test_and_set_bit(BITMAP_IO_QUEUED, &device->flags)) |
84b8c06b AG |
3514 | drbd_queue_work(&first_peer_device(device)->connection->sender_work, |
3515 | &device->bm_io_work.w); | |
b411b363 | 3516 | } |
0500813f | 3517 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 PR |
3518 | } |
3519 | ||
3520 | /** | |
3521 | * drbd_bitmap_io() - Does an IO operation on the whole bitmap | |
b30ab791 | 3522 | * @device: DRBD device. |
b411b363 PR |
3523 | * @io_fn: IO callback to be called when bitmap IO is possible |
3524 | * @why: Descriptive text of the reason for doing the IO | |
584164c8 | 3525 | * @flags: Bitmap flags |
b411b363 PR |
3526 | * |
3527 | * freezes application IO while that the actual IO operations runs. This | |
3528 | * functions MAY NOT be called from worker context. | |
3529 | */ | |
b30ab791 | 3530 | int drbd_bitmap_io(struct drbd_device *device, int (*io_fn)(struct drbd_device *), |
20ceb2b2 | 3531 | char *why, enum bm_flag flags) |
b411b363 | 3532 | { |
c0065f98 LE |
3533 | /* Only suspend io, if some operation is supposed to be locked out */ |
3534 | const bool do_suspend_io = flags & (BM_DONT_CLEAR|BM_DONT_SET|BM_DONT_TEST); | |
b411b363 PR |
3535 | int rv; |
3536 | ||
0b0ba1ef | 3537 | D_ASSERT(device, current != first_peer_device(device)->connection->worker.task); |
b411b363 | 3538 | |
c0065f98 | 3539 | if (do_suspend_io) |
b30ab791 | 3540 | drbd_suspend_io(device); |
b411b363 | 3541 | |
b30ab791 AG |
3542 | drbd_bm_lock(device, why, flags); |
3543 | rv = io_fn(device); | |
3544 | drbd_bm_unlock(device); | |
b411b363 | 3545 | |
c0065f98 | 3546 | if (do_suspend_io) |
b30ab791 | 3547 | drbd_resume_io(device); |
b411b363 PR |
3548 | |
3549 | return rv; | |
3550 | } | |
3551 | ||
b30ab791 | 3552 | void drbd_md_set_flag(struct drbd_device *device, int flag) __must_hold(local) |
b411b363 | 3553 | { |
b30ab791 AG |
3554 | if ((device->ldev->md.flags & flag) != flag) { |
3555 | drbd_md_mark_dirty(device); | |
3556 | device->ldev->md.flags |= flag; | |
b411b363 PR |
3557 | } |
3558 | } | |
3559 | ||
b30ab791 | 3560 | void drbd_md_clear_flag(struct drbd_device *device, int flag) __must_hold(local) |
b411b363 | 3561 | { |
b30ab791 AG |
3562 | if ((device->ldev->md.flags & flag) != 0) { |
3563 | drbd_md_mark_dirty(device); | |
3564 | device->ldev->md.flags &= ~flag; | |
b411b363 PR |
3565 | } |
3566 | } | |
3567 | int drbd_md_test_flag(struct drbd_backing_dev *bdev, int flag) | |
3568 | { | |
3569 | return (bdev->md.flags & flag) != 0; | |
3570 | } | |
3571 | ||
2bccef39 | 3572 | static void md_sync_timer_fn(struct timer_list *t) |
b411b363 | 3573 | { |
2bccef39 | 3574 | struct drbd_device *device = from_timer(device, t, md_sync_timer); |
ac0acb9e | 3575 | drbd_device_post_work(device, MD_SYNC); |
b411b363 PR |
3576 | } |
3577 | ||
d8763023 | 3578 | const char *cmdname(enum drbd_packet cmd) |
f2ad9063 AG |
3579 | { |
3580 | /* THINK may need to become several global tables | |
3581 | * when we want to support more than | |
3582 | * one PRO_VERSION */ | |
3583 | static const char *cmdnames[] = { | |
33cb0917 | 3584 | |
f2ad9063 AG |
3585 | [P_DATA] = "Data", |
3586 | [P_DATA_REPLY] = "DataReply", | |
3587 | [P_RS_DATA_REPLY] = "RSDataReply", | |
3588 | [P_BARRIER] = "Barrier", | |
3589 | [P_BITMAP] = "ReportBitMap", | |
3590 | [P_BECOME_SYNC_TARGET] = "BecomeSyncTarget", | |
3591 | [P_BECOME_SYNC_SOURCE] = "BecomeSyncSource", | |
3592 | [P_UNPLUG_REMOTE] = "UnplugRemote", | |
3593 | [P_DATA_REQUEST] = "DataRequest", | |
3594 | [P_RS_DATA_REQUEST] = "RSDataRequest", | |
3595 | [P_SYNC_PARAM] = "SyncParam", | |
f2ad9063 AG |
3596 | [P_PROTOCOL] = "ReportProtocol", |
3597 | [P_UUIDS] = "ReportUUIDs", | |
3598 | [P_SIZES] = "ReportSizes", | |
3599 | [P_STATE] = "ReportState", | |
3600 | [P_SYNC_UUID] = "ReportSyncUUID", | |
3601 | [P_AUTH_CHALLENGE] = "AuthChallenge", | |
3602 | [P_AUTH_RESPONSE] = "AuthResponse", | |
33cb0917 | 3603 | [P_STATE_CHG_REQ] = "StateChgRequest", |
f2ad9063 AG |
3604 | [P_PING] = "Ping", |
3605 | [P_PING_ACK] = "PingAck", | |
3606 | [P_RECV_ACK] = "RecvAck", | |
3607 | [P_WRITE_ACK] = "WriteAck", | |
3608 | [P_RS_WRITE_ACK] = "RSWriteAck", | |
d4dabbe2 | 3609 | [P_SUPERSEDED] = "Superseded", |
f2ad9063 AG |
3610 | [P_NEG_ACK] = "NegAck", |
3611 | [P_NEG_DREPLY] = "NegDReply", | |
3612 | [P_NEG_RS_DREPLY] = "NegRSDReply", | |
3613 | [P_BARRIER_ACK] = "BarrierAck", | |
f2ad9063 AG |
3614 | [P_STATE_CHG_REPLY] = "StateChgReply", |
3615 | [P_OV_REQUEST] = "OVRequest", | |
3616 | [P_OV_REPLY] = "OVReply", | |
3617 | [P_OV_RESULT] = "OVResult", | |
3618 | [P_CSUM_RS_REQUEST] = "CsumRSRequest", | |
3619 | [P_RS_IS_IN_SYNC] = "CsumRSIsInSync", | |
33cb0917 | 3620 | [P_SYNC_PARAM89] = "SyncParam89", |
f2ad9063 AG |
3621 | [P_COMPRESSED_BITMAP] = "CBitmap", |
3622 | [P_DELAY_PROBE] = "DelayProbe", | |
3623 | [P_OUT_OF_SYNC] = "OutOfSync", | |
ae25b336 LE |
3624 | [P_RS_CANCEL] = "RSCancel", |
3625 | [P_CONN_ST_CHG_REQ] = "conn_st_chg_req", | |
3626 | [P_CONN_ST_CHG_REPLY] = "conn_st_chg_reply", | |
036b17ea | 3627 | [P_PROTOCOL_UPDATE] = "protocol_update", |
33cb0917 | 3628 | [P_TRIM] = "Trim", |
700ca8c0 PR |
3629 | [P_RS_THIN_REQ] = "rs_thin_req", |
3630 | [P_RS_DEALLOCATED] = "rs_deallocated", | |
33cb0917 AB |
3631 | [P_WSAME] = "WriteSame", |
3632 | [P_ZEROES] = "Zeroes", | |
ae25b336 LE |
3633 | |
3634 | /* enum drbd_packet, but not commands - obsoleted flags: | |
3635 | * P_MAY_IGNORE | |
3636 | * P_MAX_OPT_CMD | |
3637 | */ | |
f2ad9063 AG |
3638 | }; |
3639 | ||
ae25b336 | 3640 | /* too big for the array: 0xfffX */ |
e5d6f33a AG |
3641 | if (cmd == P_INITIAL_META) |
3642 | return "InitialMeta"; | |
3643 | if (cmd == P_INITIAL_DATA) | |
3644 | return "InitialData"; | |
6038178e AG |
3645 | if (cmd == P_CONNECTION_FEATURES) |
3646 | return "ConnectionFeatures"; | |
6e849ce8 | 3647 | if (cmd >= ARRAY_SIZE(cmdnames)) |
f2ad9063 AG |
3648 | return "Unknown"; |
3649 | return cmdnames[cmd]; | |
3650 | } | |
3651 | ||
7be8da07 AG |
3652 | /** |
3653 | * drbd_wait_misc - wait for a request to make progress | |
b30ab791 | 3654 | * @device: device associated with the request |
7be8da07 AG |
3655 | * @i: the struct drbd_interval embedded in struct drbd_request or |
3656 | * struct drbd_peer_request | |
3657 | */ | |
b30ab791 | 3658 | int drbd_wait_misc(struct drbd_device *device, struct drbd_interval *i) |
7be8da07 | 3659 | { |
44ed167d | 3660 | struct net_conf *nc; |
7be8da07 AG |
3661 | DEFINE_WAIT(wait); |
3662 | long timeout; | |
3663 | ||
44ed167d | 3664 | rcu_read_lock(); |
a6b32bc3 | 3665 | nc = rcu_dereference(first_peer_device(device)->connection->net_conf); |
44ed167d PR |
3666 | if (!nc) { |
3667 | rcu_read_unlock(); | |
7be8da07 | 3668 | return -ETIMEDOUT; |
44ed167d PR |
3669 | } |
3670 | timeout = nc->ko_count ? nc->timeout * HZ / 10 * nc->ko_count : MAX_SCHEDULE_TIMEOUT; | |
3671 | rcu_read_unlock(); | |
7be8da07 | 3672 | |
b30ab791 | 3673 | /* Indicate to wake up device->misc_wait on progress. */ |
7be8da07 | 3674 | i->waiting = true; |
b30ab791 | 3675 | prepare_to_wait(&device->misc_wait, &wait, TASK_INTERRUPTIBLE); |
0500813f | 3676 | spin_unlock_irq(&device->resource->req_lock); |
7be8da07 | 3677 | timeout = schedule_timeout(timeout); |
b30ab791 | 3678 | finish_wait(&device->misc_wait, &wait); |
0500813f | 3679 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 3680 | if (!timeout || device->state.conn < C_CONNECTED) |
7be8da07 AG |
3681 | return -ETIMEDOUT; |
3682 | if (signal_pending(current)) | |
3683 | return -ERESTARTSYS; | |
3684 | return 0; | |
b411b363 PR |
3685 | } |
3686 | ||
28bc3b8c AG |
3687 | void lock_all_resources(void) |
3688 | { | |
3689 | struct drbd_resource *resource; | |
3690 | int __maybe_unused i = 0; | |
3691 | ||
3692 | mutex_lock(&resources_mutex); | |
3693 | local_irq_disable(); | |
3694 | for_each_resource(resource, &drbd_resources) | |
3695 | spin_lock_nested(&resource->req_lock, i++); | |
3696 | } | |
3697 | ||
3698 | void unlock_all_resources(void) | |
3699 | { | |
3700 | struct drbd_resource *resource; | |
3701 | ||
3702 | for_each_resource(resource, &drbd_resources) | |
3703 | spin_unlock(&resource->req_lock); | |
3704 | local_irq_enable(); | |
3705 | mutex_unlock(&resources_mutex); | |
3706 | } | |
3707 | ||
b411b363 PR |
3708 | #ifdef CONFIG_DRBD_FAULT_INJECTION |
3709 | /* Fault insertion support including random number generator shamelessly | |
3710 | * stolen from kernel/rcutorture.c */ | |
3711 | struct fault_random_state { | |
3712 | unsigned long state; | |
3713 | unsigned long count; | |
3714 | }; | |
3715 | ||
3716 | #define FAULT_RANDOM_MULT 39916801 /* prime */ | |
3717 | #define FAULT_RANDOM_ADD 479001701 /* prime */ | |
3718 | #define FAULT_RANDOM_REFRESH 10000 | |
3719 | ||
3720 | /* | |
3721 | * Crude but fast random-number generator. Uses a linear congruential | |
3722 | * generator, with occasional help from get_random_bytes(). | |
3723 | */ | |
3724 | static unsigned long | |
3725 | _drbd_fault_random(struct fault_random_state *rsp) | |
3726 | { | |
3727 | long refresh; | |
3728 | ||
49829ea7 | 3729 | if (!rsp->count--) { |
b411b363 PR |
3730 | get_random_bytes(&refresh, sizeof(refresh)); |
3731 | rsp->state += refresh; | |
3732 | rsp->count = FAULT_RANDOM_REFRESH; | |
3733 | } | |
3734 | rsp->state = rsp->state * FAULT_RANDOM_MULT + FAULT_RANDOM_ADD; | |
3735 | return swahw32(rsp->state); | |
3736 | } | |
3737 | ||
3738 | static char * | |
3739 | _drbd_fault_str(unsigned int type) { | |
3740 | static char *_faults[] = { | |
3741 | [DRBD_FAULT_MD_WR] = "Meta-data write", | |
3742 | [DRBD_FAULT_MD_RD] = "Meta-data read", | |
3743 | [DRBD_FAULT_RS_WR] = "Resync write", | |
3744 | [DRBD_FAULT_RS_RD] = "Resync read", | |
3745 | [DRBD_FAULT_DT_WR] = "Data write", | |
3746 | [DRBD_FAULT_DT_RD] = "Data read", | |
3747 | [DRBD_FAULT_DT_RA] = "Data read ahead", | |
3748 | [DRBD_FAULT_BM_ALLOC] = "BM allocation", | |
6b4388ac PR |
3749 | [DRBD_FAULT_AL_EE] = "EE allocation", |
3750 | [DRBD_FAULT_RECEIVE] = "receive data corruption", | |
b411b363 PR |
3751 | }; |
3752 | ||
3753 | return (type < DRBD_FAULT_MAX) ? _faults[type] : "**Unknown**"; | |
3754 | } | |
3755 | ||
3756 | unsigned int | |
b30ab791 | 3757 | _drbd_insert_fault(struct drbd_device *device, unsigned int type) |
b411b363 PR |
3758 | { |
3759 | static struct fault_random_state rrs = {0, 0}; | |
3760 | ||
3761 | unsigned int ret = ( | |
183ece30 RK |
3762 | (drbd_fault_devs == 0 || |
3763 | ((1 << device_to_minor(device)) & drbd_fault_devs) != 0) && | |
3764 | (((_drbd_fault_random(&rrs) % 100) + 1) <= drbd_fault_rate)); | |
b411b363 PR |
3765 | |
3766 | if (ret) { | |
183ece30 | 3767 | drbd_fault_count++; |
b411b363 | 3768 | |
7383506c | 3769 | if (__ratelimit(&drbd_ratelimit_state)) |
d0180171 | 3770 | drbd_warn(device, "***Simulating %s failure\n", |
b411b363 PR |
3771 | _drbd_fault_str(type)); |
3772 | } | |
3773 | ||
3774 | return ret; | |
3775 | } | |
3776 | #endif | |
3777 | ||
3778 | const char *drbd_buildtag(void) | |
3779 | { | |
3780 | /* DRBD built from external sources has here a reference to the | |
3781 | git hash of the source code. */ | |
3782 | ||
3783 | static char buildtag[38] = "\0uilt-in"; | |
3784 | ||
3785 | if (buildtag[0] == 0) { | |
bc4854bc CW |
3786 | #ifdef MODULE |
3787 | sprintf(buildtag, "srcversion: %-24s", THIS_MODULE->srcversion); | |
3788 | #else | |
3789 | buildtag[0] = 'b'; | |
b411b363 | 3790 | #endif |
b411b363 PR |
3791 | } |
3792 | ||
3793 | return buildtag; | |
3794 | } | |
3795 | ||
3796 | module_init(drbd_init) | |
3797 | module_exit(drbd_cleanup) | |
3798 | ||
b411b363 PR |
3799 | EXPORT_SYMBOL(drbd_conn_str); |
3800 | EXPORT_SYMBOL(drbd_role_str); | |
3801 | EXPORT_SYMBOL(drbd_disk_str); | |
3802 | EXPORT_SYMBOL(drbd_set_st_err_str); |