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; | |
128 | spinlock_t drbd_pp_lock; | |
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 | |
da4a75d2 LE |
141 | struct bio *bio_alloc_drbd(gfp_t gfp_mask) |
142 | { | |
143 | struct bio *bio; | |
19f843aa | 144 | |
0892fac8 | 145 | if (!bioset_initialized(&drbd_md_io_bio_set)) |
da4a75d2 | 146 | return bio_alloc(gfp_mask, 1); |
19f843aa | 147 | |
0892fac8 | 148 | bio = bio_alloc_bioset(gfp_mask, 1, &drbd_md_io_bio_set); |
da4a75d2 LE |
149 | if (!bio) |
150 | return NULL; | |
da4a75d2 | 151 | return bio; |
19f843aa LE |
152 | } |
153 | ||
b411b363 PR |
154 | #ifdef __CHECKER__ |
155 | /* When checking with sparse, and this is an inline function, sparse will | |
156 | give tons of false positives. When this is a real functions sparse works. | |
b411b363 | 157 | */ |
b30ab791 | 158 | int _get_ldev_if_state(struct drbd_device *device, enum drbd_disk_state mins) |
b411b363 | 159 | { |
b411b363 | 160 | int io_allowed; |
b411b363 | 161 | |
b30ab791 AG |
162 | atomic_inc(&device->local_cnt); |
163 | io_allowed = (device->state.disk >= mins); | |
b411b363 | 164 | if (!io_allowed) { |
b30ab791 AG |
165 | if (atomic_dec_and_test(&device->local_cnt)) |
166 | wake_up(&device->misc_wait); | |
b411b363 | 167 | } |
b411b363 PR |
168 | return io_allowed; |
169 | } | |
b411b363 | 170 | |
b411b363 | 171 | #endif |
265be2d0 | 172 | |
b411b363 | 173 | /** |
b6dd1a89 | 174 | * tl_release() - mark as BARRIER_ACKED all requests in the corresponding transfer log epoch |
bde89a9e | 175 | * @connection: DRBD connection. |
b411b363 PR |
176 | * @barrier_nr: Expected identifier of the DRBD write barrier packet. |
177 | * @set_size: Expected number of requests before that barrier. | |
178 | * | |
179 | * In case the passed barrier_nr or set_size does not match the oldest | |
b6dd1a89 LE |
180 | * epoch of not yet barrier-acked requests, this function will cause a |
181 | * termination of the connection. | |
b411b363 | 182 | */ |
bde89a9e | 183 | void tl_release(struct drbd_connection *connection, unsigned int barrier_nr, |
2f5cdd0b | 184 | unsigned int set_size) |
b411b363 | 185 | { |
b411b363 | 186 | struct drbd_request *r; |
b6dd1a89 LE |
187 | struct drbd_request *req = NULL; |
188 | int expect_epoch = 0; | |
189 | int expect_size = 0; | |
b411b363 | 190 | |
0500813f | 191 | spin_lock_irq(&connection->resource->req_lock); |
b411b363 | 192 | |
98683650 | 193 | /* find oldest not yet barrier-acked write request, |
b6dd1a89 | 194 | * count writes in its epoch. */ |
bde89a9e | 195 | list_for_each_entry(r, &connection->transfer_log, tl_requests) { |
a0d856df | 196 | const unsigned s = r->rq_state; |
b6dd1a89 LE |
197 | if (!req) { |
198 | if (!(s & RQ_WRITE)) | |
199 | continue; | |
200 | if (!(s & RQ_NET_MASK)) | |
201 | continue; | |
202 | if (s & RQ_NET_DONE) | |
203 | continue; | |
204 | req = r; | |
205 | expect_epoch = req->epoch; | |
206 | expect_size ++; | |
207 | } else { | |
208 | if (r->epoch != expect_epoch) | |
209 | break; | |
210 | if (!(s & RQ_WRITE)) | |
211 | continue; | |
212 | /* if (s & RQ_DONE): not expected */ | |
213 | /* if (!(s & RQ_NET_MASK)): not expected */ | |
214 | expect_size++; | |
43a5182c | 215 | } |
b411b363 | 216 | } |
67098930 | 217 | |
b411b363 | 218 | /* first some paranoia code */ |
b6dd1a89 | 219 | if (req == NULL) { |
1ec861eb | 220 | drbd_err(connection, "BAD! BarrierAck #%u received, but no epoch in tl!?\n", |
2f5cdd0b | 221 | barrier_nr); |
b411b363 | 222 | goto bail; |
b411b363 | 223 | } |
b6dd1a89 | 224 | if (expect_epoch != barrier_nr) { |
1ec861eb | 225 | drbd_err(connection, "BAD! BarrierAck #%u received, expected #%u!\n", |
b6dd1a89 | 226 | barrier_nr, expect_epoch); |
b411b363 | 227 | goto bail; |
5a22db89 LE |
228 | } |
229 | ||
b6dd1a89 | 230 | if (expect_size != set_size) { |
1ec861eb | 231 | drbd_err(connection, "BAD! BarrierAck #%u received with n_writes=%u, expected n_writes=%u!\n", |
b6dd1a89 | 232 | barrier_nr, set_size, expect_size); |
b411b363 | 233 | goto bail; |
b411b363 PR |
234 | } |
235 | ||
98683650 PR |
236 | /* Clean up list of requests processed during current epoch. */ |
237 | /* this extra list walk restart is paranoia, | |
238 | * to catch requests being barrier-acked "unexpectedly". | |
239 | * It usually should find the same req again, or some READ preceding it. */ | |
bde89a9e | 240 | list_for_each_entry(req, &connection->transfer_log, tl_requests) |
98683650 PR |
241 | if (req->epoch == expect_epoch) |
242 | break; | |
bde89a9e | 243 | list_for_each_entry_safe_from(req, r, &connection->transfer_log, tl_requests) { |
b6dd1a89 LE |
244 | if (req->epoch != expect_epoch) |
245 | break; | |
246 | _req_mod(req, BARRIER_ACKED); | |
19f843aa | 247 | } |
0500813f | 248 | spin_unlock_irq(&connection->resource->req_lock); |
19f843aa | 249 | |
b411b363 | 250 | return; |
b411b363 | 251 | |
b411b363 | 252 | bail: |
0500813f | 253 | spin_unlock_irq(&connection->resource->req_lock); |
bde89a9e | 254 | conn_request_state(connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD); |
b411b363 | 255 | } |
b411b363 | 256 | |
b411b363 | 257 | |
b411b363 | 258 | /** |
11b58e73 | 259 | * _tl_restart() - Walks the transfer log, and applies an action to all requests |
e5f891b2 | 260 | * @connection: DRBD connection to operate on. |
11b58e73 | 261 | * @what: The action/event to perform with all request objects |
b411b363 | 262 | * |
8554df1c AG |
263 | * @what might be one of CONNECTION_LOST_WHILE_PENDING, RESEND, FAIL_FROZEN_DISK_IO, |
264 | * RESTART_FROZEN_DISK_IO. | |
b411b363 | 265 | */ |
b6dd1a89 | 266 | /* must hold resource->req_lock */ |
bde89a9e | 267 | void _tl_restart(struct drbd_connection *connection, enum drbd_req_event what) |
b411b363 | 268 | { |
b6dd1a89 | 269 | struct drbd_request *req, *r; |
b411b363 | 270 | |
bde89a9e | 271 | list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests) |
b6dd1a89 LE |
272 | _req_mod(req, what); |
273 | } | |
738a84b2 | 274 | |
bde89a9e | 275 | void tl_restart(struct drbd_connection *connection, enum drbd_req_event what) |
b6dd1a89 | 276 | { |
0500813f | 277 | spin_lock_irq(&connection->resource->req_lock); |
bde89a9e | 278 | _tl_restart(connection, what); |
0500813f | 279 | spin_unlock_irq(&connection->resource->req_lock); |
cdfda633 | 280 | } |
b411b363 | 281 | |
b411b363 PR |
282 | /** |
283 | * tl_clear() - Clears all requests and &struct drbd_tl_epoch objects out of the TL | |
b30ab791 | 284 | * @device: DRBD device. |
b411b363 PR |
285 | * |
286 | * This is called after the connection to the peer was lost. The storage covered | |
287 | * by the requests on the transfer gets marked as our of sync. Called from the | |
288 | * receiver thread and the worker thread. | |
289 | */ | |
bde89a9e | 290 | void tl_clear(struct drbd_connection *connection) |
b411b363 | 291 | { |
bde89a9e | 292 | tl_restart(connection, CONNECTION_LOST_WHILE_PENDING); |
b411b363 | 293 | } |
197296ff | 294 | |
cdfda633 | 295 | /** |
b30ab791 AG |
296 | * tl_abort_disk_io() - Abort disk I/O for all requests for a certain device in the TL |
297 | * @device: DRBD device. | |
cdfda633 | 298 | */ |
b30ab791 | 299 | void tl_abort_disk_io(struct drbd_device *device) |
cdfda633 | 300 | { |
a6b32bc3 | 301 | struct drbd_connection *connection = first_peer_device(device)->connection; |
b6dd1a89 | 302 | struct drbd_request *req, *r; |
02851e9f | 303 | |
0500813f | 304 | spin_lock_irq(&connection->resource->req_lock); |
bde89a9e | 305 | list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests) { |
97ddb687 LE |
306 | if (!(req->rq_state & RQ_LOCAL_PENDING)) |
307 | continue; | |
84b8c06b | 308 | if (req->device != device) |
b6dd1a89 LE |
309 | continue; |
310 | _req_mod(req, ABORT_DISK_IO); | |
b411b363 | 311 | } |
0500813f | 312 | spin_unlock_irq(&connection->resource->req_lock); |
b411b363 PR |
313 | } |
314 | ||
b411b363 PR |
315 | static int drbd_thread_setup(void *arg) |
316 | { | |
317 | struct drbd_thread *thi = (struct drbd_thread *) arg; | |
2457b6d5 | 318 | struct drbd_resource *resource = thi->resource; |
b411b363 PR |
319 | unsigned long flags; |
320 | int retval; | |
321 | ||
f1b3a6ec | 322 | snprintf(current->comm, sizeof(current->comm), "drbd_%c_%s", |
77c556f6 | 323 | thi->name[0], |
2457b6d5 | 324 | resource->name); |
f1b3a6ec | 325 | |
33da8e7c EB |
326 | allow_kernel_signal(DRBD_SIGKILL); |
327 | allow_kernel_signal(SIGXCPU); | |
b411b363 PR |
328 | restart: |
329 | retval = thi->function(thi); | |
330 | ||
331 | spin_lock_irqsave(&thi->t_lock, flags); | |
332 | ||
e77a0a5c | 333 | /* if the receiver has been "EXITING", the last thing it did |
b411b363 PR |
334 | * was set the conn state to "StandAlone", |
335 | * if now a re-connect request comes in, conn state goes C_UNCONNECTED, | |
336 | * and receiver thread will be "started". | |
e77a0a5c | 337 | * drbd_thread_start needs to set "RESTARTING" in that case. |
b411b363 | 338 | * t_state check and assignment needs to be within the same spinlock, |
e77a0a5c AG |
339 | * so either thread_start sees EXITING, and can remap to RESTARTING, |
340 | * or thread_start see NONE, and can proceed as normal. | |
b411b363 PR |
341 | */ |
342 | ||
e77a0a5c | 343 | if (thi->t_state == RESTARTING) { |
2457b6d5 | 344 | drbd_info(resource, "Restarting %s thread\n", thi->name); |
e77a0a5c | 345 | thi->t_state = RUNNING; |
b411b363 PR |
346 | spin_unlock_irqrestore(&thi->t_lock, flags); |
347 | goto restart; | |
348 | } | |
349 | ||
350 | thi->task = NULL; | |
e77a0a5c | 351 | thi->t_state = NONE; |
b411b363 | 352 | smp_mb(); |
992d6e91 | 353 | complete_all(&thi->stop); |
b411b363 PR |
354 | spin_unlock_irqrestore(&thi->t_lock, flags); |
355 | ||
2457b6d5 | 356 | drbd_info(resource, "Terminating %s\n", current->comm); |
b411b363 PR |
357 | |
358 | /* Release mod reference taken when thread was started */ | |
9dc9fbb3 | 359 | |
2457b6d5 AG |
360 | if (thi->connection) |
361 | kref_put(&thi->connection->kref, drbd_destroy_connection); | |
362 | kref_put(&resource->kref, drbd_destroy_resource); | |
b411b363 PR |
363 | module_put(THIS_MODULE); |
364 | return retval; | |
365 | } | |
366 | ||
2457b6d5 | 367 | static void drbd_thread_init(struct drbd_resource *resource, struct drbd_thread *thi, |
c60b0251 | 368 | int (*func) (struct drbd_thread *), const char *name) |
b411b363 PR |
369 | { |
370 | spin_lock_init(&thi->t_lock); | |
371 | thi->task = NULL; | |
e77a0a5c | 372 | thi->t_state = NONE; |
b411b363 | 373 | thi->function = func; |
2457b6d5 AG |
374 | thi->resource = resource; |
375 | thi->connection = NULL; | |
c60b0251 | 376 | thi->name = name; |
b411b363 PR |
377 | } |
378 | ||
379 | int drbd_thread_start(struct drbd_thread *thi) | |
380 | { | |
2457b6d5 | 381 | struct drbd_resource *resource = thi->resource; |
b411b363 PR |
382 | struct task_struct *nt; |
383 | unsigned long flags; | |
384 | ||
b411b363 PR |
385 | /* is used from state engine doing drbd_thread_stop_nowait, |
386 | * while holding the req lock irqsave */ | |
387 | spin_lock_irqsave(&thi->t_lock, flags); | |
388 | ||
389 | switch (thi->t_state) { | |
e77a0a5c | 390 | case NONE: |
2457b6d5 | 391 | drbd_info(resource, "Starting %s thread (from %s [%d])\n", |
bed879ae | 392 | thi->name, current->comm, current->pid); |
b411b363 PR |
393 | |
394 | /* Get ref on module for thread - this is released when thread exits */ | |
395 | if (!try_module_get(THIS_MODULE)) { | |
2457b6d5 | 396 | drbd_err(resource, "Failed to get module reference in drbd_thread_start\n"); |
b411b363 | 397 | spin_unlock_irqrestore(&thi->t_lock, flags); |
81e84650 | 398 | return false; |
b411b363 PR |
399 | } |
400 | ||
2457b6d5 AG |
401 | kref_get(&resource->kref); |
402 | if (thi->connection) | |
403 | kref_get(&thi->connection->kref); | |
9dc9fbb3 | 404 | |
b411b363 | 405 | init_completion(&thi->stop); |
b411b363 | 406 | thi->reset_cpu_mask = 1; |
e77a0a5c | 407 | thi->t_state = RUNNING; |
b411b363 PR |
408 | spin_unlock_irqrestore(&thi->t_lock, flags); |
409 | flush_signals(current); /* otherw. may get -ERESTARTNOINTR */ | |
410 | ||
411 | nt = kthread_create(drbd_thread_setup, (void *) thi, | |
2457b6d5 | 412 | "drbd_%c_%s", thi->name[0], thi->resource->name); |
b411b363 PR |
413 | |
414 | if (IS_ERR(nt)) { | |
2457b6d5 | 415 | drbd_err(resource, "Couldn't start thread\n"); |
b411b363 | 416 | |
2457b6d5 AG |
417 | if (thi->connection) |
418 | kref_put(&thi->connection->kref, drbd_destroy_connection); | |
419 | kref_put(&resource->kref, drbd_destroy_resource); | |
b411b363 | 420 | module_put(THIS_MODULE); |
81e84650 | 421 | return false; |
b411b363 PR |
422 | } |
423 | spin_lock_irqsave(&thi->t_lock, flags); | |
424 | thi->task = nt; | |
e77a0a5c | 425 | thi->t_state = RUNNING; |
b411b363 PR |
426 | spin_unlock_irqrestore(&thi->t_lock, flags); |
427 | wake_up_process(nt); | |
428 | break; | |
e77a0a5c AG |
429 | case EXITING: |
430 | thi->t_state = RESTARTING; | |
2457b6d5 | 431 | drbd_info(resource, "Restarting %s thread (from %s [%d])\n", |
bed879ae | 432 | thi->name, current->comm, current->pid); |
df561f66 | 433 | fallthrough; |
e77a0a5c AG |
434 | case RUNNING: |
435 | case RESTARTING: | |
b411b363 PR |
436 | default: |
437 | spin_unlock_irqrestore(&thi->t_lock, flags); | |
438 | break; | |
439 | } | |
440 | ||
81e84650 | 441 | return true; |
b411b363 PR |
442 | } |
443 | ||
444 | ||
445 | void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait) | |
446 | { | |
447 | unsigned long flags; | |
448 | ||
e77a0a5c | 449 | enum drbd_thread_state ns = restart ? RESTARTING : EXITING; |
b411b363 PR |
450 | |
451 | /* may be called from state engine, holding the req lock irqsave */ | |
452 | spin_lock_irqsave(&thi->t_lock, flags); | |
453 | ||
e77a0a5c | 454 | if (thi->t_state == NONE) { |
b411b363 PR |
455 | spin_unlock_irqrestore(&thi->t_lock, flags); |
456 | if (restart) | |
457 | drbd_thread_start(thi); | |
458 | return; | |
459 | } | |
460 | ||
461 | if (thi->t_state != ns) { | |
462 | if (thi->task == NULL) { | |
463 | spin_unlock_irqrestore(&thi->t_lock, flags); | |
464 | return; | |
465 | } | |
466 | ||
467 | thi->t_state = ns; | |
468 | smp_mb(); | |
469 | init_completion(&thi->stop); | |
470 | if (thi->task != current) | |
fee10990 | 471 | send_sig(DRBD_SIGKILL, thi->task, 1); |
b411b363 PR |
472 | } |
473 | ||
474 | spin_unlock_irqrestore(&thi->t_lock, flags); | |
475 | ||
476 | if (wait) | |
477 | wait_for_completion(&thi->stop); | |
478 | } | |
479 | ||
bde89a9e | 480 | int conn_lowest_minor(struct drbd_connection *connection) |
80822284 | 481 | { |
c06ece6b AG |
482 | struct drbd_peer_device *peer_device; |
483 | int vnr = 0, minor = -1; | |
774b3055 | 484 | |
695d08fa | 485 | rcu_read_lock(); |
c06ece6b AG |
486 | peer_device = idr_get_next(&connection->peer_devices, &vnr); |
487 | if (peer_device) | |
488 | minor = device_to_minor(peer_device->device); | |
695d08fa PR |
489 | rcu_read_unlock(); |
490 | ||
c06ece6b | 491 | return minor; |
80822284 | 492 | } |
774b3055 | 493 | |
b411b363 PR |
494 | #ifdef CONFIG_SMP |
495 | /** | |
496 | * drbd_calc_cpu_mask() - Generate CPU masks, spread over all CPUs | |
b411b363 | 497 | * |
625a6ba2 | 498 | * Forces all threads of a resource onto the same CPU. This is beneficial for |
b411b363 PR |
499 | * DRBD's performance. May be overwritten by user's configuration. |
500 | */ | |
625a6ba2 | 501 | static void drbd_calc_cpu_mask(cpumask_var_t *cpu_mask) |
b411b363 | 502 | { |
625a6ba2 | 503 | unsigned int *resources_per_cpu, min_index = ~0; |
b411b363 | 504 | |
6396bb22 KC |
505 | resources_per_cpu = kcalloc(nr_cpu_ids, sizeof(*resources_per_cpu), |
506 | GFP_KERNEL); | |
625a6ba2 AG |
507 | if (resources_per_cpu) { |
508 | struct drbd_resource *resource; | |
509 | unsigned int cpu, min = ~0; | |
b411b363 | 510 | |
625a6ba2 AG |
511 | rcu_read_lock(); |
512 | for_each_resource_rcu(resource, &drbd_resources) { | |
513 | for_each_cpu(cpu, resource->cpu_mask) | |
514 | resources_per_cpu[cpu]++; | |
b411b363 | 515 | } |
625a6ba2 AG |
516 | rcu_read_unlock(); |
517 | for_each_online_cpu(cpu) { | |
518 | if (resources_per_cpu[cpu] < min) { | |
519 | min = resources_per_cpu[cpu]; | |
520 | min_index = cpu; | |
521 | } | |
522 | } | |
523 | kfree(resources_per_cpu); | |
524 | } | |
525 | if (min_index == ~0) { | |
526 | cpumask_setall(*cpu_mask); | |
527 | return; | |
b411b363 | 528 | } |
625a6ba2 | 529 | cpumask_set_cpu(min_index, *cpu_mask); |
b411b363 PR |
530 | } |
531 | ||
532 | /** | |
533 | * drbd_thread_current_set_cpu() - modifies the cpu mask of the _current_ thread | |
b30ab791 | 534 | * @device: DRBD device. |
bc31fe33 | 535 | * @thi: drbd_thread object |
b411b363 PR |
536 | * |
537 | * call in the "main loop" of _all_ threads, no need for any mutex, current won't die | |
538 | * prematurely. | |
539 | */ | |
80822284 | 540 | void drbd_thread_current_set_cpu(struct drbd_thread *thi) |
b411b363 | 541 | { |
2457b6d5 | 542 | struct drbd_resource *resource = thi->resource; |
b411b363 | 543 | struct task_struct *p = current; |
bed879ae | 544 | |
b411b363 PR |
545 | if (!thi->reset_cpu_mask) |
546 | return; | |
547 | thi->reset_cpu_mask = 0; | |
2457b6d5 | 548 | set_cpus_allowed_ptr(p, resource->cpu_mask); |
b411b363 | 549 | } |
625a6ba2 AG |
550 | #else |
551 | #define drbd_calc_cpu_mask(A) ({}) | |
b411b363 PR |
552 | #endif |
553 | ||
52b061a4 AG |
554 | /** |
555 | * drbd_header_size - size of a packet header | |
556 | * | |
557 | * The header size is a multiple of 8, so any payload following the header is | |
558 | * word aligned on 64-bit architectures. (The bitmap send and receive code | |
559 | * relies on this.) | |
560 | */ | |
bde89a9e | 561 | unsigned int drbd_header_size(struct drbd_connection *connection) |
b411b363 | 562 | { |
bde89a9e | 563 | if (connection->agreed_pro_version >= 100) { |
0c8e36d9 AG |
564 | BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header100), 8)); |
565 | return sizeof(struct p_header100); | |
566 | } else { | |
567 | BUILD_BUG_ON(sizeof(struct p_header80) != | |
568 | sizeof(struct p_header95)); | |
569 | BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header80), 8)); | |
570 | return sizeof(struct p_header80); | |
571 | } | |
52b061a4 | 572 | } |
b411b363 | 573 | |
e658983a | 574 | static unsigned int prepare_header80(struct p_header80 *h, enum drbd_packet cmd, int size) |
fd340c12 PR |
575 | { |
576 | h->magic = cpu_to_be32(DRBD_MAGIC); | |
577 | h->command = cpu_to_be16(cmd); | |
578 | h->length = cpu_to_be16(size); | |
e658983a | 579 | return sizeof(struct p_header80); |
fd340c12 | 580 | } |
b411b363 | 581 | |
e658983a | 582 | static unsigned int prepare_header95(struct p_header95 *h, enum drbd_packet cmd, int size) |
fd340c12 PR |
583 | { |
584 | h->magic = cpu_to_be16(DRBD_MAGIC_BIG); | |
b411b363 | 585 | h->command = cpu_to_be16(cmd); |
b55d84ba | 586 | h->length = cpu_to_be32(size); |
e658983a | 587 | return sizeof(struct p_header95); |
fd340c12 | 588 | } |
b411b363 | 589 | |
0c8e36d9 AG |
590 | static unsigned int prepare_header100(struct p_header100 *h, enum drbd_packet cmd, |
591 | int size, int vnr) | |
592 | { | |
593 | h->magic = cpu_to_be32(DRBD_MAGIC_100); | |
594 | h->volume = cpu_to_be16(vnr); | |
595 | h->command = cpu_to_be16(cmd); | |
596 | h->length = cpu_to_be32(size); | |
597 | h->pad = 0; | |
598 | return sizeof(struct p_header100); | |
599 | } | |
b411b363 | 600 | |
bde89a9e | 601 | static unsigned int prepare_header(struct drbd_connection *connection, int vnr, |
0c8e36d9 | 602 | void *buffer, enum drbd_packet cmd, int size) |
d38e787e | 603 | { |
bde89a9e | 604 | if (connection->agreed_pro_version >= 100) |
0c8e36d9 | 605 | return prepare_header100(buffer, cmd, size, vnr); |
bde89a9e | 606 | else if (connection->agreed_pro_version >= 95 && |
0c8e36d9 | 607 | size > DRBD_MAX_SIZE_H80_PACKET) |
e658983a | 608 | return prepare_header95(buffer, cmd, size); |
d38e787e | 609 | else |
e658983a | 610 | return prepare_header80(buffer, cmd, size); |
b411b363 PR |
611 | } |
612 | ||
bde89a9e | 613 | static void *__conn_prepare_command(struct drbd_connection *connection, |
a7eb7bdf | 614 | struct drbd_socket *sock) |
b411b363 | 615 | { |
a7eb7bdf AG |
616 | if (!sock->socket) |
617 | return NULL; | |
bde89a9e | 618 | return sock->sbuf + drbd_header_size(connection); |
a7eb7bdf | 619 | } |
b411b363 | 620 | |
bde89a9e | 621 | void *conn_prepare_command(struct drbd_connection *connection, struct drbd_socket *sock) |
dba58587 | 622 | { |
a7eb7bdf | 623 | void *p; |
b411b363 | 624 | |
dba58587 | 625 | mutex_lock(&sock->mutex); |
bde89a9e | 626 | p = __conn_prepare_command(connection, sock); |
a7eb7bdf | 627 | if (!p) |
dba58587 | 628 | mutex_unlock(&sock->mutex); |
b411b363 | 629 | |
a7eb7bdf | 630 | return p; |
b411b363 PR |
631 | } |
632 | ||
69a22773 | 633 | void *drbd_prepare_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock) |
b411b363 | 634 | { |
69a22773 | 635 | return conn_prepare_command(peer_device->connection, sock); |
dba58587 | 636 | } |
b411b363 | 637 | |
bde89a9e | 638 | static int __send_command(struct drbd_connection *connection, int vnr, |
dba58587 AG |
639 | struct drbd_socket *sock, enum drbd_packet cmd, |
640 | unsigned int header_size, void *data, | |
641 | unsigned int size) | |
642 | { | |
643 | int msg_flags; | |
644 | int err; | |
b411b363 | 645 | |
dba58587 AG |
646 | /* |
647 | * Called with @data == NULL and the size of the data blocks in @size | |
648 | * for commands that send data blocks. For those commands, omit the | |
649 | * MSG_MORE flag: this will increase the likelihood that data blocks | |
650 | * which are page aligned on the sender will end up page aligned on the | |
651 | * receiver. | |
652 | */ | |
653 | msg_flags = data ? MSG_MORE : 0; | |
654 | ||
bde89a9e | 655 | header_size += prepare_header(connection, vnr, sock->sbuf, cmd, |
e658983a | 656 | header_size + size); |
bde89a9e | 657 | err = drbd_send_all(connection, sock->socket, sock->sbuf, header_size, |
dba58587 AG |
658 | msg_flags); |
659 | if (data && !err) | |
bde89a9e | 660 | err = drbd_send_all(connection, sock->socket, data, size, 0); |
123ff122 LE |
661 | /* DRBD protocol "pings" are latency critical. |
662 | * This is supposed to trigger tcp_push_pending_frames() */ | |
663 | if (!err && (cmd == P_PING || cmd == P_PING_ACK)) | |
12abc5ee | 664 | tcp_sock_set_nodelay(sock->socket->sk); |
123ff122 | 665 | |
dba58587 AG |
666 | return err; |
667 | } | |
668 | ||
bde89a9e | 669 | static int __conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock, |
a7eb7bdf AG |
670 | enum drbd_packet cmd, unsigned int header_size, |
671 | void *data, unsigned int size) | |
672 | { | |
bde89a9e | 673 | return __send_command(connection, 0, sock, cmd, header_size, data, size); |
a7eb7bdf AG |
674 | } |
675 | ||
bde89a9e | 676 | int conn_send_command(struct drbd_connection *connection, 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; | |
b411b363 | 681 | |
bde89a9e | 682 | err = __conn_send_command(connection, sock, cmd, header_size, data, size); |
dba58587 AG |
683 | mutex_unlock(&sock->mutex); |
684 | return err; | |
685 | } | |
686 | ||
69a22773 | 687 | int drbd_send_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock, |
dba58587 AG |
688 | enum drbd_packet cmd, unsigned int header_size, |
689 | void *data, unsigned int size) | |
690 | { | |
691 | int err; | |
692 | ||
69a22773 AG |
693 | err = __send_command(peer_device->connection, peer_device->device->vnr, |
694 | sock, cmd, header_size, data, size); | |
dba58587 AG |
695 | mutex_unlock(&sock->mutex); |
696 | return err; | |
697 | } | |
b411b363 | 698 | |
bde89a9e | 699 | int drbd_send_ping(struct drbd_connection *connection) |
e307f352 | 700 | { |
9f5bdc33 AG |
701 | struct drbd_socket *sock; |
702 | ||
bde89a9e AG |
703 | sock = &connection->meta; |
704 | if (!conn_prepare_command(connection, sock)) | |
9f5bdc33 | 705 | return -EIO; |
bde89a9e | 706 | return conn_send_command(connection, sock, P_PING, 0, NULL, 0); |
e307f352 | 707 | } |
b411b363 | 708 | |
bde89a9e | 709 | int drbd_send_ping_ack(struct drbd_connection *connection) |
e307f352 | 710 | { |
9f5bdc33 AG |
711 | struct drbd_socket *sock; |
712 | ||
bde89a9e AG |
713 | sock = &connection->meta; |
714 | if (!conn_prepare_command(connection, sock)) | |
9f5bdc33 | 715 | return -EIO; |
bde89a9e | 716 | return conn_send_command(connection, sock, P_PING_ACK, 0, NULL, 0); |
b411b363 PR |
717 | } |
718 | ||
69a22773 | 719 | int drbd_send_sync_param(struct drbd_peer_device *peer_device) |
b411b363 | 720 | { |
7c96715a | 721 | struct drbd_socket *sock; |
8e26f9cc | 722 | struct p_rs_param_95 *p; |
9f5bdc33 | 723 | int size; |
69a22773 | 724 | const int apv = peer_device->connection->agreed_pro_version; |
9f5bdc33 | 725 | enum drbd_packet cmd; |
44ed167d | 726 | struct net_conf *nc; |
daeda1cc | 727 | struct disk_conf *dc; |
9f5bdc33 | 728 | |
69a22773 AG |
729 | sock = &peer_device->connection->data; |
730 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 AG |
731 | if (!p) |
732 | return -EIO; | |
b411b363 | 733 | |
44ed167d | 734 | rcu_read_lock(); |
69a22773 | 735 | nc = rcu_dereference(peer_device->connection->net_conf); |
b411b363 PR |
736 | |
737 | size = apv <= 87 ? sizeof(struct p_rs_param) | |
738 | : apv == 88 ? sizeof(struct p_rs_param) | |
44ed167d | 739 | + strlen(nc->verify_alg) + 1 |
8e26f9cc PR |
740 | : apv <= 94 ? sizeof(struct p_rs_param_89) |
741 | : /* apv >= 95 */ sizeof(struct p_rs_param_95); | |
b411b363 | 742 | |
9f5bdc33 | 743 | cmd = apv >= 89 ? P_SYNC_PARAM89 : P_SYNC_PARAM; |
b411b363 | 744 | |
9f5bdc33 AG |
745 | /* initialize verify_alg and csums_alg */ |
746 | memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX); | |
b411b363 | 747 | |
69a22773 AG |
748 | if (get_ldev(peer_device->device)) { |
749 | dc = rcu_dereference(peer_device->device->ldev->disk_conf); | |
6394b935 | 750 | p->resync_rate = cpu_to_be32(dc->resync_rate); |
daeda1cc PR |
751 | p->c_plan_ahead = cpu_to_be32(dc->c_plan_ahead); |
752 | p->c_delay_target = cpu_to_be32(dc->c_delay_target); | |
753 | p->c_fill_target = cpu_to_be32(dc->c_fill_target); | |
754 | p->c_max_rate = cpu_to_be32(dc->c_max_rate); | |
69a22773 | 755 | put_ldev(peer_device->device); |
9f5bdc33 | 756 | } else { |
6394b935 | 757 | p->resync_rate = cpu_to_be32(DRBD_RESYNC_RATE_DEF); |
9f5bdc33 AG |
758 | p->c_plan_ahead = cpu_to_be32(DRBD_C_PLAN_AHEAD_DEF); |
759 | p->c_delay_target = cpu_to_be32(DRBD_C_DELAY_TARGET_DEF); | |
760 | p->c_fill_target = cpu_to_be32(DRBD_C_FILL_TARGET_DEF); | |
761 | p->c_max_rate = cpu_to_be32(DRBD_C_MAX_RATE_DEF); | |
762 | } | |
b411b363 | 763 | |
9f5bdc33 | 764 | if (apv >= 88) |
44ed167d | 765 | strcpy(p->verify_alg, nc->verify_alg); |
9f5bdc33 | 766 | if (apv >= 89) |
44ed167d PR |
767 | strcpy(p->csums_alg, nc->csums_alg); |
768 | rcu_read_unlock(); | |
b411b363 | 769 | |
69a22773 | 770 | return drbd_send_command(peer_device, sock, cmd, size, NULL, 0); |
b411b363 PR |
771 | } |
772 | ||
bde89a9e | 773 | int __drbd_send_protocol(struct drbd_connection *connection, enum drbd_packet cmd) |
b411b363 | 774 | { |
9f5bdc33 | 775 | struct drbd_socket *sock; |
b411b363 | 776 | struct p_protocol *p; |
44ed167d | 777 | struct net_conf *nc; |
9f5bdc33 | 778 | int size, cf; |
b411b363 | 779 | |
bde89a9e AG |
780 | sock = &connection->data; |
781 | p = __conn_prepare_command(connection, sock); | |
9f5bdc33 AG |
782 | if (!p) |
783 | return -EIO; | |
b411b363 | 784 | |
44ed167d | 785 | rcu_read_lock(); |
bde89a9e | 786 | nc = rcu_dereference(connection->net_conf); |
b411b363 | 787 | |
bde89a9e | 788 | if (nc->tentative && connection->agreed_pro_version < 92) { |
44ed167d | 789 | rcu_read_unlock(); |
1ec861eb | 790 | drbd_err(connection, "--dry-run is not supported by peer"); |
44ed167d PR |
791 | return -EOPNOTSUPP; |
792 | } | |
b411b363 | 793 | |
9f5bdc33 | 794 | size = sizeof(*p); |
bde89a9e | 795 | if (connection->agreed_pro_version >= 87) |
44ed167d | 796 | size += strlen(nc->integrity_alg) + 1; |
b411b363 | 797 | |
44ed167d PR |
798 | p->protocol = cpu_to_be32(nc->wire_protocol); |
799 | p->after_sb_0p = cpu_to_be32(nc->after_sb_0p); | |
800 | p->after_sb_1p = cpu_to_be32(nc->after_sb_1p); | |
801 | p->after_sb_2p = cpu_to_be32(nc->after_sb_2p); | |
802 | p->two_primaries = cpu_to_be32(nc->two_primaries); | |
cf14c2e9 | 803 | cf = 0; |
6139f60d AG |
804 | if (nc->discard_my_data) |
805 | cf |= CF_DISCARD_MY_DATA; | |
6dff2902 | 806 | if (nc->tentative) |
9f5bdc33 | 807 | cf |= CF_DRY_RUN; |
cf14c2e9 PR |
808 | p->conn_flags = cpu_to_be32(cf); |
809 | ||
bde89a9e | 810 | if (connection->agreed_pro_version >= 87) |
44ed167d PR |
811 | strcpy(p->integrity_alg, nc->integrity_alg); |
812 | rcu_read_unlock(); | |
b411b363 | 813 | |
bde89a9e | 814 | return __conn_send_command(connection, sock, cmd, size, NULL, 0); |
a7eb7bdf AG |
815 | } |
816 | ||
bde89a9e | 817 | int drbd_send_protocol(struct drbd_connection *connection) |
a7eb7bdf AG |
818 | { |
819 | int err; | |
820 | ||
bde89a9e AG |
821 | mutex_lock(&connection->data.mutex); |
822 | err = __drbd_send_protocol(connection, P_PROTOCOL); | |
823 | mutex_unlock(&connection->data.mutex); | |
a7eb7bdf AG |
824 | |
825 | return err; | |
b411b363 PR |
826 | } |
827 | ||
69a22773 | 828 | static int _drbd_send_uuids(struct drbd_peer_device *peer_device, u64 uuid_flags) |
b411b363 | 829 | { |
69a22773 | 830 | struct drbd_device *device = peer_device->device; |
9f5bdc33 AG |
831 | struct drbd_socket *sock; |
832 | struct p_uuids *p; | |
b411b363 PR |
833 | int i; |
834 | ||
b30ab791 | 835 | if (!get_ldev_if_state(device, D_NEGOTIATING)) |
2ae5f95b | 836 | return 0; |
b411b363 | 837 | |
69a22773 AG |
838 | sock = &peer_device->connection->data; |
839 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 | 840 | if (!p) { |
b30ab791 | 841 | put_ldev(device); |
9f5bdc33 AG |
842 | return -EIO; |
843 | } | |
b30ab791 | 844 | spin_lock_irq(&device->ldev->md.uuid_lock); |
b411b363 | 845 | for (i = UI_CURRENT; i < UI_SIZE; i++) |
b30ab791 AG |
846 | p->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]); |
847 | spin_unlock_irq(&device->ldev->md.uuid_lock); | |
b411b363 | 848 | |
b30ab791 AG |
849 | device->comm_bm_set = drbd_bm_total_weight(device); |
850 | p->uuid[UI_SIZE] = cpu_to_be64(device->comm_bm_set); | |
44ed167d | 851 | rcu_read_lock(); |
69a22773 | 852 | uuid_flags |= rcu_dereference(peer_device->connection->net_conf)->discard_my_data ? 1 : 0; |
44ed167d | 853 | rcu_read_unlock(); |
b30ab791 AG |
854 | uuid_flags |= test_bit(CRASHED_PRIMARY, &device->flags) ? 2 : 0; |
855 | uuid_flags |= device->new_state_tmp.disk == D_INCONSISTENT ? 4 : 0; | |
9f5bdc33 | 856 | p->uuid[UI_FLAGS] = cpu_to_be64(uuid_flags); |
b411b363 | 857 | |
b30ab791 | 858 | put_ldev(device); |
69a22773 | 859 | return drbd_send_command(peer_device, sock, P_UUIDS, sizeof(*p), NULL, 0); |
b411b363 PR |
860 | } |
861 | ||
69a22773 | 862 | int drbd_send_uuids(struct drbd_peer_device *peer_device) |
b411b363 | 863 | { |
69a22773 | 864 | return _drbd_send_uuids(peer_device, 0); |
b411b363 PR |
865 | } |
866 | ||
69a22773 | 867 | int drbd_send_uuids_skip_initial_sync(struct drbd_peer_device *peer_device) |
b411b363 | 868 | { |
69a22773 | 869 | return _drbd_send_uuids(peer_device, 8); |
b411b363 PR |
870 | } |
871 | ||
b30ab791 | 872 | void drbd_print_uuids(struct drbd_device *device, const char *text) |
62b0da3a | 873 | { |
b30ab791 AG |
874 | if (get_ldev_if_state(device, D_NEGOTIATING)) { |
875 | u64 *uuid = device->ldev->md.uuid; | |
d0180171 | 876 | drbd_info(device, "%s %016llX:%016llX:%016llX:%016llX\n", |
62b0da3a LE |
877 | text, |
878 | (unsigned long long)uuid[UI_CURRENT], | |
879 | (unsigned long long)uuid[UI_BITMAP], | |
880 | (unsigned long long)uuid[UI_HISTORY_START], | |
881 | (unsigned long long)uuid[UI_HISTORY_END]); | |
b30ab791 | 882 | put_ldev(device); |
62b0da3a | 883 | } else { |
d0180171 | 884 | drbd_info(device, "%s effective data uuid: %016llX\n", |
62b0da3a | 885 | text, |
b30ab791 | 886 | (unsigned long long)device->ed_uuid); |
62b0da3a LE |
887 | } |
888 | } | |
889 | ||
69a22773 | 890 | void drbd_gen_and_send_sync_uuid(struct drbd_peer_device *peer_device) |
b411b363 | 891 | { |
69a22773 | 892 | struct drbd_device *device = peer_device->device; |
9f5bdc33 AG |
893 | struct drbd_socket *sock; |
894 | struct p_rs_uuid *p; | |
5a22db89 LE |
895 | u64 uuid; |
896 | ||
0b0ba1ef | 897 | D_ASSERT(device, device->state.disk == D_UP_TO_DATE); |
b411b363 | 898 | |
b30ab791 | 899 | uuid = device->ldev->md.uuid[UI_BITMAP]; |
5ba3dac5 PR |
900 | if (uuid && uuid != UUID_JUST_CREATED) |
901 | uuid = uuid + UUID_NEW_BM_OFFSET; | |
902 | else | |
903 | get_random_bytes(&uuid, sizeof(u64)); | |
b30ab791 AG |
904 | drbd_uuid_set(device, UI_BITMAP, uuid); |
905 | drbd_print_uuids(device, "updated sync UUID"); | |
906 | drbd_md_sync(device); | |
b411b363 | 907 | |
69a22773 AG |
908 | sock = &peer_device->connection->data; |
909 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 AG |
910 | if (p) { |
911 | p->uuid = cpu_to_be64(uuid); | |
69a22773 | 912 | drbd_send_command(peer_device, sock, P_SYNC_UUID, sizeof(*p), NULL, 0); |
9f5bdc33 | 913 | } |
b411b363 PR |
914 | } |
915 | ||
9104d31a | 916 | /* communicated if (agreed_features & DRBD_FF_WSAME) */ |
1ffa7bfa BX |
917 | static void |
918 | assign_p_sizes_qlim(struct drbd_device *device, struct p_sizes *p, | |
919 | struct request_queue *q) | |
9104d31a LE |
920 | { |
921 | if (q) { | |
922 | p->qlim->physical_block_size = cpu_to_be32(queue_physical_block_size(q)); | |
923 | p->qlim->logical_block_size = cpu_to_be32(queue_logical_block_size(q)); | |
924 | p->qlim->alignment_offset = cpu_to_be32(queue_alignment_offset(q)); | |
925 | p->qlim->io_min = cpu_to_be32(queue_io_min(q)); | |
926 | p->qlim->io_opt = cpu_to_be32(queue_io_opt(q)); | |
927 | p->qlim->discard_enabled = blk_queue_discard(q); | |
9104d31a LE |
928 | p->qlim->write_same_capable = !!q->limits.max_write_same_sectors; |
929 | } else { | |
930 | q = device->rq_queue; | |
931 | p->qlim->physical_block_size = cpu_to_be32(queue_physical_block_size(q)); | |
932 | p->qlim->logical_block_size = cpu_to_be32(queue_logical_block_size(q)); | |
933 | p->qlim->alignment_offset = 0; | |
934 | p->qlim->io_min = cpu_to_be32(queue_io_min(q)); | |
935 | p->qlim->io_opt = cpu_to_be32(queue_io_opt(q)); | |
936 | p->qlim->discard_enabled = 0; | |
9104d31a LE |
937 | p->qlim->write_same_capable = 0; |
938 | } | |
939 | } | |
940 | ||
69a22773 | 941 | int drbd_send_sizes(struct drbd_peer_device *peer_device, int trigger_reply, enum dds_flags flags) |
b411b363 | 942 | { |
69a22773 | 943 | struct drbd_device *device = peer_device->device; |
9f5bdc33 AG |
944 | struct drbd_socket *sock; |
945 | struct p_sizes *p; | |
b411b363 | 946 | sector_t d_size, u_size; |
db141b2f LE |
947 | int q_order_type; |
948 | unsigned int max_bio_size; | |
9104d31a LE |
949 | unsigned int packet_size; |
950 | ||
951 | sock = &peer_device->connection->data; | |
952 | p = drbd_prepare_command(peer_device, sock); | |
953 | if (!p) | |
954 | return -EIO; | |
b411b363 | 955 | |
9104d31a LE |
956 | packet_size = sizeof(*p); |
957 | if (peer_device->connection->agreed_features & DRBD_FF_WSAME) | |
958 | packet_size += sizeof(p->qlim[0]); | |
959 | ||
960 | memset(p, 0, packet_size); | |
b30ab791 | 961 | if (get_ldev_if_state(device, D_NEGOTIATING)) { |
9104d31a | 962 | struct request_queue *q = bdev_get_queue(device->ldev->backing_bdev); |
b30ab791 | 963 | d_size = drbd_get_max_capacity(device->ldev); |
daeda1cc | 964 | rcu_read_lock(); |
b30ab791 | 965 | u_size = rcu_dereference(device->ldev->disk_conf)->disk_size; |
daeda1cc | 966 | rcu_read_unlock(); |
b30ab791 | 967 | q_order_type = drbd_queue_order_type(device); |
9104d31a | 968 | max_bio_size = queue_max_hw_sectors(q) << 9; |
db141b2f | 969 | max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE); |
9104d31a | 970 | assign_p_sizes_qlim(device, p, q); |
b30ab791 | 971 | put_ldev(device); |
b411b363 PR |
972 | } else { |
973 | d_size = 0; | |
974 | u_size = 0; | |
975 | q_order_type = QUEUE_ORDERED_NONE; | |
99432fcc | 976 | max_bio_size = DRBD_MAX_BIO_SIZE; /* ... multiple BIOs per peer_request */ |
9104d31a | 977 | assign_p_sizes_qlim(device, p, NULL); |
b411b363 PR |
978 | } |
979 | ||
69a22773 | 980 | if (peer_device->connection->agreed_pro_version <= 94) |
98683650 | 981 | max_bio_size = min(max_bio_size, DRBD_MAX_SIZE_H80_PACKET); |
69a22773 | 982 | else if (peer_device->connection->agreed_pro_version < 100) |
98683650 | 983 | max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE_P95); |
b411b363 | 984 | |
9f5bdc33 AG |
985 | p->d_size = cpu_to_be64(d_size); |
986 | p->u_size = cpu_to_be64(u_size); | |
b30ab791 | 987 | p->c_size = cpu_to_be64(trigger_reply ? 0 : drbd_get_capacity(device->this_bdev)); |
9f5bdc33 AG |
988 | p->max_bio_size = cpu_to_be32(max_bio_size); |
989 | p->queue_order_type = cpu_to_be16(q_order_type); | |
990 | p->dds_flags = cpu_to_be16(flags); | |
9104d31a LE |
991 | |
992 | return drbd_send_command(peer_device, sock, P_SIZES, packet_size, NULL, 0); | |
b411b363 PR |
993 | } |
994 | ||
995 | /** | |
f479ea06 | 996 | * drbd_send_current_state() - Sends the drbd state to the peer |
69a22773 | 997 | * @peer_device: DRBD peer device. |
b411b363 | 998 | */ |
69a22773 | 999 | int drbd_send_current_state(struct drbd_peer_device *peer_device) |
b411b363 | 1000 | { |
7c96715a | 1001 | struct drbd_socket *sock; |
9f5bdc33 | 1002 | struct p_state *p; |
b411b363 | 1003 | |
69a22773 AG |
1004 | sock = &peer_device->connection->data; |
1005 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 AG |
1006 | if (!p) |
1007 | return -EIO; | |
69a22773 AG |
1008 | p->state = cpu_to_be32(peer_device->device->state.i); /* Within the send mutex */ |
1009 | return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0); | |
b411b363 PR |
1010 | } |
1011 | ||
f479ea06 LE |
1012 | /** |
1013 | * drbd_send_state() - After a state change, sends the new state to the peer | |
69a22773 | 1014 | * @peer_device: DRBD peer device. |
43de7c85 | 1015 | * @state: the state to send, not necessarily the current state. |
f479ea06 LE |
1016 | * |
1017 | * Each state change queues an "after_state_ch" work, which will eventually | |
1018 | * send the resulting new state to the peer. If more state changes happen | |
1019 | * between queuing and processing of the after_state_ch work, we still | |
1020 | * want to send each intermediary state in the order it occurred. | |
1021 | */ | |
69a22773 | 1022 | int drbd_send_state(struct drbd_peer_device *peer_device, union drbd_state state) |
f479ea06 | 1023 | { |
43de7c85 PR |
1024 | struct drbd_socket *sock; |
1025 | struct p_state *p; | |
f479ea06 | 1026 | |
69a22773 AG |
1027 | sock = &peer_device->connection->data; |
1028 | p = drbd_prepare_command(peer_device, sock); | |
43de7c85 PR |
1029 | if (!p) |
1030 | return -EIO; | |
1031 | p->state = cpu_to_be32(state.i); /* Within the send mutex */ | |
69a22773 | 1032 | return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0); |
43de7c85 | 1033 | } |
f479ea06 | 1034 | |
69a22773 | 1035 | int drbd_send_state_req(struct drbd_peer_device *peer_device, union drbd_state mask, union drbd_state val) |
9f5bdc33 AG |
1036 | { |
1037 | struct drbd_socket *sock; | |
1038 | struct p_req_state *p; | |
f479ea06 | 1039 | |
69a22773 AG |
1040 | sock = &peer_device->connection->data; |
1041 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 AG |
1042 | if (!p) |
1043 | return -EIO; | |
1044 | p->mask = cpu_to_be32(mask.i); | |
1045 | p->val = cpu_to_be32(val.i); | |
69a22773 | 1046 | return drbd_send_command(peer_device, sock, P_STATE_CHG_REQ, sizeof(*p), NULL, 0); |
b411b363 | 1047 | } |
f479ea06 | 1048 | |
bde89a9e | 1049 | int conn_send_state_req(struct drbd_connection *connection, union drbd_state mask, union drbd_state val) |
b411b363 | 1050 | { |
9f5bdc33 AG |
1051 | enum drbd_packet cmd; |
1052 | struct drbd_socket *sock; | |
1053 | struct p_req_state *p; | |
f479ea06 | 1054 | |
bde89a9e AG |
1055 | cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REQ : P_CONN_ST_CHG_REQ; |
1056 | sock = &connection->data; | |
1057 | p = conn_prepare_command(connection, sock); | |
9f5bdc33 AG |
1058 | if (!p) |
1059 | return -EIO; | |
1060 | p->mask = cpu_to_be32(mask.i); | |
1061 | p->val = cpu_to_be32(val.i); | |
bde89a9e | 1062 | return conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0); |
f479ea06 LE |
1063 | } |
1064 | ||
69a22773 | 1065 | void drbd_send_sr_reply(struct drbd_peer_device *peer_device, enum drbd_state_rv retcode) |
b411b363 | 1066 | { |
9f5bdc33 AG |
1067 | struct drbd_socket *sock; |
1068 | struct p_req_state_reply *p; | |
b411b363 | 1069 | |
69a22773 AG |
1070 | sock = &peer_device->connection->meta; |
1071 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 AG |
1072 | if (p) { |
1073 | p->retcode = cpu_to_be32(retcode); | |
69a22773 | 1074 | drbd_send_command(peer_device, sock, P_STATE_CHG_REPLY, sizeof(*p), NULL, 0); |
9f5bdc33 | 1075 | } |
b411b363 | 1076 | } |
b411b363 | 1077 | |
bde89a9e | 1078 | void conn_send_sr_reply(struct drbd_connection *connection, enum drbd_state_rv retcode) |
047cd4a6 | 1079 | { |
9f5bdc33 AG |
1080 | struct drbd_socket *sock; |
1081 | struct p_req_state_reply *p; | |
bde89a9e | 1082 | enum drbd_packet cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REPLY : P_CONN_ST_CHG_REPLY; |
b411b363 | 1083 | |
bde89a9e AG |
1084 | sock = &connection->meta; |
1085 | p = conn_prepare_command(connection, sock); | |
9f5bdc33 AG |
1086 | if (p) { |
1087 | p->retcode = cpu_to_be32(retcode); | |
bde89a9e | 1088 | conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0); |
9f5bdc33 | 1089 | } |
b411b363 PR |
1090 | } |
1091 | ||
a02d1240 | 1092 | static void dcbp_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code) |
b411b363 | 1093 | { |
a02d1240 AG |
1094 | BUG_ON(code & ~0xf); |
1095 | p->encoding = (p->encoding & ~0xf) | code; | |
1096 | } | |
b411b363 | 1097 | |
a02d1240 AG |
1098 | static void dcbp_set_start(struct p_compressed_bm *p, int set) |
1099 | { | |
1100 | p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0); | |
1101 | } | |
b411b363 | 1102 | |
a02d1240 AG |
1103 | static void dcbp_set_pad_bits(struct p_compressed_bm *p, int n) |
1104 | { | |
1105 | BUG_ON(n & ~0x7); | |
1106 | p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4); | |
b411b363 PR |
1107 | } |
1108 | ||
b30ab791 | 1109 | static int fill_bitmap_rle_bits(struct drbd_device *device, |
50d0b1ad AG |
1110 | struct p_compressed_bm *p, |
1111 | unsigned int size, | |
1112 | struct bm_xfer_ctx *c) | |
b411b363 PR |
1113 | { |
1114 | struct bitstream bs; | |
1115 | unsigned long plain_bits; | |
1116 | unsigned long tmp; | |
1117 | unsigned long rl; | |
1118 | unsigned len; | |
1119 | unsigned toggle; | |
44ed167d | 1120 | int bits, use_rle; |
b411b363 PR |
1121 | |
1122 | /* may we use this feature? */ | |
44ed167d | 1123 | rcu_read_lock(); |
a6b32bc3 | 1124 | use_rle = rcu_dereference(first_peer_device(device)->connection->net_conf)->use_rle; |
44ed167d | 1125 | rcu_read_unlock(); |
a6b32bc3 | 1126 | if (!use_rle || first_peer_device(device)->connection->agreed_pro_version < 90) |
44ed167d | 1127 | return 0; |
b411b363 PR |
1128 | |
1129 | if (c->bit_offset >= c->bm_bits) | |
1130 | return 0; /* nothing to do. */ | |
1131 | ||
1132 | /* use at most thus many bytes */ | |
50d0b1ad AG |
1133 | bitstream_init(&bs, p->code, size, 0); |
1134 | memset(p->code, 0, size); | |
b411b363 PR |
1135 | /* plain bits covered in this code string */ |
1136 | plain_bits = 0; | |
1137 | ||
1138 | /* p->encoding & 0x80 stores whether the first run length is set. | |
1139 | * bit offset is implicit. | |
1140 | * start with toggle == 2 to be able to tell the first iteration */ | |
1141 | toggle = 2; | |
1142 | ||
1143 | /* see how much plain bits we can stuff into one packet | |
1144 | * using RLE and VLI. */ | |
1145 | do { | |
b30ab791 AG |
1146 | tmp = (toggle == 0) ? _drbd_bm_find_next_zero(device, c->bit_offset) |
1147 | : _drbd_bm_find_next(device, c->bit_offset); | |
b411b363 PR |
1148 | if (tmp == -1UL) |
1149 | tmp = c->bm_bits; | |
1150 | rl = tmp - c->bit_offset; | |
1151 | ||
1152 | if (toggle == 2) { /* first iteration */ | |
1153 | if (rl == 0) { | |
1154 | /* the first checked bit was set, | |
1155 | * store start value, */ | |
a02d1240 | 1156 | dcbp_set_start(p, 1); |
b411b363 PR |
1157 | /* but skip encoding of zero run length */ |
1158 | toggle = !toggle; | |
1159 | continue; | |
1160 | } | |
a02d1240 | 1161 | dcbp_set_start(p, 0); |
b411b363 PR |
1162 | } |
1163 | ||
1164 | /* paranoia: catch zero runlength. | |
1165 | * can only happen if bitmap is modified while we scan it. */ | |
1166 | if (rl == 0) { | |
d0180171 | 1167 | drbd_err(device, "unexpected zero runlength while encoding bitmap " |
b411b363 PR |
1168 | "t:%u bo:%lu\n", toggle, c->bit_offset); |
1169 | return -1; | |
1170 | } | |
1171 | ||
1172 | bits = vli_encode_bits(&bs, rl); | |
1173 | if (bits == -ENOBUFS) /* buffer full */ | |
1174 | break; | |
1175 | if (bits <= 0) { | |
d0180171 | 1176 | drbd_err(device, "error while encoding bitmap: %d\n", bits); |
b411b363 PR |
1177 | return 0; |
1178 | } | |
1179 | ||
1180 | toggle = !toggle; | |
1181 | plain_bits += rl; | |
1182 | c->bit_offset = tmp; | |
1183 | } while (c->bit_offset < c->bm_bits); | |
1184 | ||
1185 | len = bs.cur.b - p->code + !!bs.cur.bit; | |
1186 | ||
1187 | if (plain_bits < (len << 3)) { | |
1188 | /* incompressible with this method. | |
1189 | * we need to rewind both word and bit position. */ | |
1190 | c->bit_offset -= plain_bits; | |
1191 | bm_xfer_ctx_bit_to_word_offset(c); | |
1192 | c->bit_offset = c->word_offset * BITS_PER_LONG; | |
1193 | return 0; | |
1194 | } | |
1195 | ||
1196 | /* RLE + VLI was able to compress it just fine. | |
1197 | * update c->word_offset. */ | |
1198 | bm_xfer_ctx_bit_to_word_offset(c); | |
1199 | ||
1200 | /* store pad_bits */ | |
a02d1240 | 1201 | dcbp_set_pad_bits(p, (8 - bs.cur.bit) & 0x7); |
b411b363 PR |
1202 | |
1203 | return len; | |
1204 | } | |
1205 | ||
f70af118 AG |
1206 | /** |
1207 | * send_bitmap_rle_or_plain | |
1208 | * | |
1209 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
1210 | * code upon failure. | |
1211 | */ | |
1212 | static int | |
b30ab791 | 1213 | send_bitmap_rle_or_plain(struct drbd_device *device, struct bm_xfer_ctx *c) |
b411b363 | 1214 | { |
a6b32bc3 AG |
1215 | struct drbd_socket *sock = &first_peer_device(device)->connection->data; |
1216 | unsigned int header_size = drbd_header_size(first_peer_device(device)->connection); | |
e658983a | 1217 | struct p_compressed_bm *p = sock->sbuf + header_size; |
a982dd57 | 1218 | int len, err; |
b411b363 | 1219 | |
b30ab791 | 1220 | len = fill_bitmap_rle_bits(device, p, |
e658983a | 1221 | DRBD_SOCKET_BUFFER_SIZE - header_size - sizeof(*p), c); |
b411b363 | 1222 | if (len < 0) |
f70af118 | 1223 | return -EIO; |
b411b363 PR |
1224 | |
1225 | if (len) { | |
a02d1240 | 1226 | dcbp_set_code(p, RLE_VLI_Bits); |
a6b32bc3 | 1227 | err = __send_command(first_peer_device(device)->connection, device->vnr, sock, |
9f5bdc33 AG |
1228 | P_COMPRESSED_BITMAP, sizeof(*p) + len, |
1229 | NULL, 0); | |
b411b363 | 1230 | c->packets[0]++; |
e658983a | 1231 | c->bytes[0] += header_size + sizeof(*p) + len; |
b411b363 PR |
1232 | |
1233 | if (c->bit_offset >= c->bm_bits) | |
1234 | len = 0; /* DONE */ | |
1235 | } else { | |
1236 | /* was not compressible. | |
1237 | * send a buffer full of plain text bits instead. */ | |
50d0b1ad AG |
1238 | unsigned int data_size; |
1239 | unsigned long num_words; | |
e658983a | 1240 | unsigned long *p = sock->sbuf + header_size; |
50d0b1ad AG |
1241 | |
1242 | data_size = DRBD_SOCKET_BUFFER_SIZE - header_size; | |
e658983a | 1243 | num_words = min_t(size_t, data_size / sizeof(*p), |
50d0b1ad | 1244 | c->bm_words - c->word_offset); |
e658983a | 1245 | len = num_words * sizeof(*p); |
b411b363 | 1246 | if (len) |
b30ab791 | 1247 | drbd_bm_get_lel(device, c->word_offset, num_words, p); |
a6b32bc3 | 1248 | err = __send_command(first_peer_device(device)->connection, device->vnr, sock, P_BITMAP, len, NULL, 0); |
b411b363 PR |
1249 | c->word_offset += num_words; |
1250 | c->bit_offset = c->word_offset * BITS_PER_LONG; | |
1251 | ||
1252 | c->packets[1]++; | |
50d0b1ad | 1253 | c->bytes[1] += header_size + len; |
b411b363 PR |
1254 | |
1255 | if (c->bit_offset > c->bm_bits) | |
1256 | c->bit_offset = c->bm_bits; | |
1257 | } | |
a982dd57 | 1258 | if (!err) { |
f70af118 | 1259 | if (len == 0) { |
b30ab791 | 1260 | INFO_bm_xfer_stats(device, "send", c); |
f70af118 AG |
1261 | return 0; |
1262 | } else | |
1263 | return 1; | |
1264 | } | |
1265 | return -EIO; | |
b411b363 PR |
1266 | } |
1267 | ||
1268 | /* See the comment at receive_bitmap() */ | |
b30ab791 | 1269 | static int _drbd_send_bitmap(struct drbd_device *device) |
b411b363 PR |
1270 | { |
1271 | struct bm_xfer_ctx c; | |
f70af118 | 1272 | int err; |
b411b363 | 1273 | |
b30ab791 | 1274 | if (!expect(device->bitmap)) |
81e84650 | 1275 | return false; |
b411b363 | 1276 | |
b30ab791 AG |
1277 | if (get_ldev(device)) { |
1278 | if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC)) { | |
d0180171 | 1279 | drbd_info(device, "Writing the whole bitmap, MDF_FullSync was set.\n"); |
b30ab791 AG |
1280 | drbd_bm_set_all(device); |
1281 | if (drbd_bm_write(device)) { | |
b411b363 PR |
1282 | /* write_bm did fail! Leave full sync flag set in Meta P_DATA |
1283 | * but otherwise process as per normal - need to tell other | |
1284 | * side that a full resync is required! */ | |
d0180171 | 1285 | drbd_err(device, "Failed to write bitmap to disk!\n"); |
b411b363 | 1286 | } else { |
b30ab791 AG |
1287 | drbd_md_clear_flag(device, MDF_FULL_SYNC); |
1288 | drbd_md_sync(device); | |
b411b363 PR |
1289 | } |
1290 | } | |
b30ab791 | 1291 | put_ldev(device); |
b411b363 PR |
1292 | } |
1293 | ||
1294 | c = (struct bm_xfer_ctx) { | |
b30ab791 AG |
1295 | .bm_bits = drbd_bm_bits(device), |
1296 | .bm_words = drbd_bm_words(device), | |
b411b363 PR |
1297 | }; |
1298 | ||
1299 | do { | |
b30ab791 | 1300 | err = send_bitmap_rle_or_plain(device, &c); |
f70af118 | 1301 | } while (err > 0); |
b411b363 | 1302 | |
f70af118 | 1303 | return err == 0; |
b411b363 PR |
1304 | } |
1305 | ||
b30ab791 | 1306 | int drbd_send_bitmap(struct drbd_device *device) |
b411b363 | 1307 | { |
a6b32bc3 | 1308 | struct drbd_socket *sock = &first_peer_device(device)->connection->data; |
9f5bdc33 | 1309 | int err = -1; |
b411b363 | 1310 | |
9f5bdc33 AG |
1311 | mutex_lock(&sock->mutex); |
1312 | if (sock->socket) | |
b30ab791 | 1313 | err = !_drbd_send_bitmap(device); |
9f5bdc33 | 1314 | mutex_unlock(&sock->mutex); |
b411b363 PR |
1315 | return err; |
1316 | } | |
1317 | ||
bde89a9e | 1318 | void drbd_send_b_ack(struct drbd_connection *connection, u32 barrier_nr, u32 set_size) |
b411b363 | 1319 | { |
9f5bdc33 AG |
1320 | struct drbd_socket *sock; |
1321 | struct p_barrier_ack *p; | |
b411b363 | 1322 | |
bde89a9e | 1323 | if (connection->cstate < C_WF_REPORT_PARAMS) |
9f5bdc33 | 1324 | return; |
b411b363 | 1325 | |
bde89a9e AG |
1326 | sock = &connection->meta; |
1327 | p = conn_prepare_command(connection, sock); | |
9f5bdc33 AG |
1328 | if (!p) |
1329 | return; | |
1330 | p->barrier = barrier_nr; | |
1331 | p->set_size = cpu_to_be32(set_size); | |
bde89a9e | 1332 | conn_send_command(connection, sock, P_BARRIER_ACK, sizeof(*p), NULL, 0); |
b411b363 PR |
1333 | } |
1334 | ||
1335 | /** | |
1336 | * _drbd_send_ack() - Sends an ack packet | |
b30ab791 | 1337 | * @device: DRBD device. |
b411b363 PR |
1338 | * @cmd: Packet command code. |
1339 | * @sector: sector, needs to be in big endian byte order | |
1340 | * @blksize: size in byte, needs to be in big endian byte order | |
1341 | * @block_id: Id, big endian byte order | |
1342 | */ | |
69a22773 | 1343 | static int _drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd, |
d8763023 | 1344 | u64 sector, u32 blksize, u64 block_id) |
b411b363 | 1345 | { |
9f5bdc33 AG |
1346 | struct drbd_socket *sock; |
1347 | struct p_block_ack *p; | |
b411b363 | 1348 | |
69a22773 | 1349 | if (peer_device->device->state.conn < C_CONNECTED) |
9f5bdc33 | 1350 | return -EIO; |
b411b363 | 1351 | |
69a22773 AG |
1352 | sock = &peer_device->connection->meta; |
1353 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 | 1354 | if (!p) |
a8c32aa8 | 1355 | return -EIO; |
9f5bdc33 AG |
1356 | p->sector = sector; |
1357 | p->block_id = block_id; | |
1358 | p->blksize = blksize; | |
69a22773 AG |
1359 | p->seq_num = cpu_to_be32(atomic_inc_return(&peer_device->device->packet_seq)); |
1360 | return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0); | |
b411b363 PR |
1361 | } |
1362 | ||
2b2bf214 LE |
1363 | /* dp->sector and dp->block_id already/still in network byte order, |
1364 | * data_size is payload size according to dp->head, | |
1365 | * and may need to be corrected for digest size. */ | |
69a22773 | 1366 | void drbd_send_ack_dp(struct drbd_peer_device *peer_device, enum drbd_packet cmd, |
a9a9994d | 1367 | struct p_data *dp, int data_size) |
b411b363 | 1368 | { |
69a22773 | 1369 | if (peer_device->connection->peer_integrity_tfm) |
3d0e6375 | 1370 | data_size -= crypto_shash_digestsize(peer_device->connection->peer_integrity_tfm); |
69a22773 | 1371 | _drbd_send_ack(peer_device, cmd, dp->sector, cpu_to_be32(data_size), |
a9a9994d | 1372 | dp->block_id); |
b411b363 PR |
1373 | } |
1374 | ||
69a22773 | 1375 | void drbd_send_ack_rp(struct drbd_peer_device *peer_device, enum drbd_packet cmd, |
a9a9994d | 1376 | struct p_block_req *rp) |
b411b363 | 1377 | { |
69a22773 | 1378 | _drbd_send_ack(peer_device, cmd, rp->sector, rp->blksize, rp->block_id); |
b411b363 PR |
1379 | } |
1380 | ||
1381 | /** | |
1382 | * drbd_send_ack() - Sends an ack packet | |
b30ab791 | 1383 | * @device: DRBD device |
db830c46 AG |
1384 | * @cmd: packet command code |
1385 | * @peer_req: peer request | |
b411b363 | 1386 | */ |
69a22773 | 1387 | int drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd, |
db830c46 | 1388 | struct drbd_peer_request *peer_req) |
b411b363 | 1389 | { |
69a22773 | 1390 | return _drbd_send_ack(peer_device, cmd, |
dd516121 AG |
1391 | cpu_to_be64(peer_req->i.sector), |
1392 | cpu_to_be32(peer_req->i.size), | |
1393 | peer_req->block_id); | |
b411b363 PR |
1394 | } |
1395 | ||
1396 | /* This function misuses the block_id field to signal if the blocks | |
1397 | * are is sync or not. */ | |
69a22773 | 1398 | int drbd_send_ack_ex(struct drbd_peer_device *peer_device, enum drbd_packet cmd, |
b411b363 PR |
1399 | sector_t sector, int blksize, u64 block_id) |
1400 | { | |
69a22773 | 1401 | return _drbd_send_ack(peer_device, cmd, |
b411b363 PR |
1402 | cpu_to_be64(sector), |
1403 | cpu_to_be32(blksize), | |
1404 | cpu_to_be64(block_id)); | |
1405 | } | |
1406 | ||
700ca8c0 PR |
1407 | int drbd_send_rs_deallocated(struct drbd_peer_device *peer_device, |
1408 | struct drbd_peer_request *peer_req) | |
1409 | { | |
1410 | struct drbd_socket *sock; | |
1411 | struct p_block_desc *p; | |
1412 | ||
1413 | sock = &peer_device->connection->data; | |
1414 | p = drbd_prepare_command(peer_device, sock); | |
1415 | if (!p) | |
1416 | return -EIO; | |
1417 | p->sector = cpu_to_be64(peer_req->i.sector); | |
1418 | p->blksize = cpu_to_be32(peer_req->i.size); | |
1419 | p->pad = 0; | |
1420 | return drbd_send_command(peer_device, sock, P_RS_DEALLOCATED, sizeof(*p), NULL, 0); | |
1421 | } | |
1422 | ||
69a22773 | 1423 | int drbd_send_drequest(struct drbd_peer_device *peer_device, int cmd, |
b411b363 PR |
1424 | sector_t sector, int size, u64 block_id) |
1425 | { | |
9f5bdc33 AG |
1426 | struct drbd_socket *sock; |
1427 | struct p_block_req *p; | |
b411b363 | 1428 | |
69a22773 AG |
1429 | sock = &peer_device->connection->data; |
1430 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 AG |
1431 | if (!p) |
1432 | return -EIO; | |
1433 | p->sector = cpu_to_be64(sector); | |
1434 | p->block_id = block_id; | |
1435 | p->blksize = cpu_to_be32(size); | |
69a22773 | 1436 | return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0); |
b411b363 PR |
1437 | } |
1438 | ||
69a22773 | 1439 | int drbd_send_drequest_csum(struct drbd_peer_device *peer_device, sector_t sector, int size, |
d8763023 | 1440 | void *digest, int digest_size, enum drbd_packet cmd) |
b411b363 | 1441 | { |
9f5bdc33 AG |
1442 | struct drbd_socket *sock; |
1443 | struct p_block_req *p; | |
b411b363 | 1444 | |
9f5bdc33 | 1445 | /* FIXME: Put the digest into the preallocated socket buffer. */ |
b411b363 | 1446 | |
69a22773 AG |
1447 | sock = &peer_device->connection->data; |
1448 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 AG |
1449 | if (!p) |
1450 | return -EIO; | |
1451 | p->sector = cpu_to_be64(sector); | |
1452 | p->block_id = ID_SYNCER /* unused */; | |
1453 | p->blksize = cpu_to_be32(size); | |
69a22773 | 1454 | return drbd_send_command(peer_device, sock, cmd, sizeof(*p), digest, digest_size); |
b411b363 PR |
1455 | } |
1456 | ||
69a22773 | 1457 | int drbd_send_ov_request(struct drbd_peer_device *peer_device, sector_t sector, int size) |
b411b363 | 1458 | { |
9f5bdc33 AG |
1459 | struct drbd_socket *sock; |
1460 | struct p_block_req *p; | |
b411b363 | 1461 | |
69a22773 AG |
1462 | sock = &peer_device->connection->data; |
1463 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 AG |
1464 | if (!p) |
1465 | return -EIO; | |
1466 | p->sector = cpu_to_be64(sector); | |
1467 | p->block_id = ID_SYNCER /* unused */; | |
1468 | p->blksize = cpu_to_be32(size); | |
69a22773 | 1469 | return drbd_send_command(peer_device, sock, P_OV_REQUEST, sizeof(*p), NULL, 0); |
b411b363 PR |
1470 | } |
1471 | ||
1472 | /* called on sndtimeo | |
81e84650 AG |
1473 | * returns false if we should retry, |
1474 | * true if we think connection is dead | |
b411b363 | 1475 | */ |
bde89a9e | 1476 | static int we_should_drop_the_connection(struct drbd_connection *connection, struct socket *sock) |
b411b363 PR |
1477 | { |
1478 | int drop_it; | |
b30ab791 | 1479 | /* long elapsed = (long)(jiffies - device->last_received); */ |
b411b363 | 1480 | |
bde89a9e | 1481 | drop_it = connection->meta.socket == sock |
1c03e520 PR |
1482 | || !connection->ack_receiver.task |
1483 | || get_t_state(&connection->ack_receiver) != RUNNING | |
bde89a9e | 1484 | || connection->cstate < C_WF_REPORT_PARAMS; |
b411b363 PR |
1485 | |
1486 | if (drop_it) | |
81e84650 | 1487 | return true; |
b411b363 | 1488 | |
bde89a9e | 1489 | drop_it = !--connection->ko_count; |
b411b363 | 1490 | if (!drop_it) { |
1ec861eb | 1491 | drbd_err(connection, "[%s/%d] sock_sendmsg time expired, ko = %u\n", |
bde89a9e AG |
1492 | current->comm, current->pid, connection->ko_count); |
1493 | request_ping(connection); | |
b411b363 PR |
1494 | } |
1495 | ||
b30ab791 | 1496 | return drop_it; /* && (device->state == R_PRIMARY) */; |
b411b363 PR |
1497 | } |
1498 | ||
bde89a9e | 1499 | static void drbd_update_congested(struct drbd_connection *connection) |
9e204cdd | 1500 | { |
bde89a9e | 1501 | struct sock *sk = connection->data.socket->sk; |
9e204cdd | 1502 | if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5) |
bde89a9e | 1503 | set_bit(NET_CONGESTED, &connection->flags); |
9e204cdd AG |
1504 | } |
1505 | ||
b411b363 PR |
1506 | /* The idea of sendpage seems to be to put some kind of reference |
1507 | * to the page into the skb, and to hand it over to the NIC. In | |
1508 | * this process get_page() gets called. | |
1509 | * | |
1510 | * As soon as the page was really sent over the network put_page() | |
1511 | * gets called by some part of the network layer. [ NIC driver? ] | |
1512 | * | |
1513 | * [ get_page() / put_page() increment/decrement the count. If count | |
1514 | * reaches 0 the page will be freed. ] | |
1515 | * | |
1516 | * This works nicely with pages from FSs. | |
1517 | * But this means that in protocol A we might signal IO completion too early! | |
1518 | * | |
1519 | * In order not to corrupt data during a resync we must make sure | |
1520 | * that we do not reuse our own buffer pages (EEs) to early, therefore | |
1521 | * we have the net_ee list. | |
1522 | * | |
1523 | * XFS seems to have problems, still, it submits pages with page_count == 0! | |
1524 | * As a workaround, we disable sendpage on pages | |
1525 | * with page_count == 0 or PageSlab. | |
1526 | */ | |
69a22773 | 1527 | static int _drbd_no_send_page(struct drbd_peer_device *peer_device, struct page *page, |
b987427b | 1528 | int offset, size_t size, unsigned msg_flags) |
b411b363 | 1529 | { |
b987427b AG |
1530 | struct socket *socket; |
1531 | void *addr; | |
1532 | int err; | |
1533 | ||
69a22773 | 1534 | socket = peer_device->connection->data.socket; |
b987427b | 1535 | addr = kmap(page) + offset; |
69a22773 | 1536 | err = drbd_send_all(peer_device->connection, socket, addr, size, msg_flags); |
b411b363 | 1537 | kunmap(page); |
b987427b | 1538 | if (!err) |
69a22773 | 1539 | peer_device->device->send_cnt += size >> 9; |
b987427b | 1540 | return err; |
b411b363 PR |
1541 | } |
1542 | ||
69a22773 | 1543 | static int _drbd_send_page(struct drbd_peer_device *peer_device, struct page *page, |
ba11ad9a | 1544 | int offset, size_t size, unsigned msg_flags) |
b411b363 | 1545 | { |
69a22773 | 1546 | struct socket *socket = peer_device->connection->data.socket; |
b411b363 | 1547 | int len = size; |
88b390ff | 1548 | int err = -EIO; |
b411b363 PR |
1549 | |
1550 | /* e.g. XFS meta- & log-data is in slab pages, which have a | |
1551 | * page_count of 0 and/or have PageSlab() set. | |
1552 | * we cannot use send_page for those, as that does get_page(); | |
1553 | * put_page(); and would cause either a VM_BUG directly, or | |
1554 | * __page_cache_release a page that would actually still be referenced | |
1555 | * by someone, leading to some obscure delayed Oops somewhere else. */ | |
183ece30 | 1556 | if (drbd_disable_sendpage || (page_count(page) < 1) || PageSlab(page)) |
69a22773 | 1557 | return _drbd_no_send_page(peer_device, page, offset, size, msg_flags); |
b411b363 | 1558 | |
ba11ad9a | 1559 | msg_flags |= MSG_NOSIGNAL; |
69a22773 | 1560 | drbd_update_congested(peer_device->connection); |
b411b363 | 1561 | do { |
88b390ff AG |
1562 | int sent; |
1563 | ||
1564 | sent = socket->ops->sendpage(socket, page, offset, len, msg_flags); | |
b411b363 | 1565 | if (sent <= 0) { |
88b390ff | 1566 | if (sent == -EAGAIN) { |
69a22773 | 1567 | if (we_should_drop_the_connection(peer_device->connection, socket)) |
88b390ff AG |
1568 | break; |
1569 | continue; | |
1570 | } | |
69a22773 | 1571 | drbd_warn(peer_device->device, "%s: size=%d len=%d sent=%d\n", |
b411b363 | 1572 | __func__, (int)size, len, sent); |
88b390ff AG |
1573 | if (sent < 0) |
1574 | err = sent; | |
b411b363 PR |
1575 | break; |
1576 | } | |
1577 | len -= sent; | |
1578 | offset += sent; | |
b30ab791 | 1579 | } while (len > 0 /* THINK && device->cstate >= C_CONNECTED*/); |
69a22773 | 1580 | clear_bit(NET_CONGESTED, &peer_device->connection->flags); |
b411b363 | 1581 | |
88b390ff AG |
1582 | if (len == 0) { |
1583 | err = 0; | |
69a22773 | 1584 | peer_device->device->send_cnt += size >> 9; |
88b390ff AG |
1585 | } |
1586 | return err; | |
b411b363 PR |
1587 | } |
1588 | ||
69a22773 | 1589 | static int _drbd_send_bio(struct drbd_peer_device *peer_device, struct bio *bio) |
b411b363 | 1590 | { |
7988613b KO |
1591 | struct bio_vec bvec; |
1592 | struct bvec_iter iter; | |
1593 | ||
ba11ad9a | 1594 | /* hint all but last page with MSG_MORE */ |
7988613b | 1595 | bio_for_each_segment(bvec, bio, iter) { |
7fae55da AG |
1596 | int err; |
1597 | ||
69a22773 | 1598 | err = _drbd_no_send_page(peer_device, bvec.bv_page, |
7988613b | 1599 | bvec.bv_offset, bvec.bv_len, |
4550dd6c | 1600 | bio_iter_last(bvec, iter) |
7988613b | 1601 | ? 0 : MSG_MORE); |
7fae55da AG |
1602 | if (err) |
1603 | return err; | |
9104d31a LE |
1604 | /* REQ_OP_WRITE_SAME has only one segment */ |
1605 | if (bio_op(bio) == REQ_OP_WRITE_SAME) | |
1606 | break; | |
b411b363 | 1607 | } |
7fae55da | 1608 | return 0; |
b411b363 PR |
1609 | } |
1610 | ||
69a22773 | 1611 | static int _drbd_send_zc_bio(struct drbd_peer_device *peer_device, struct bio *bio) |
b411b363 | 1612 | { |
7988613b KO |
1613 | struct bio_vec bvec; |
1614 | struct bvec_iter iter; | |
1615 | ||
ba11ad9a | 1616 | /* hint all but last page with MSG_MORE */ |
7988613b | 1617 | bio_for_each_segment(bvec, bio, iter) { |
7fae55da AG |
1618 | int err; |
1619 | ||
69a22773 | 1620 | err = _drbd_send_page(peer_device, bvec.bv_page, |
7988613b | 1621 | bvec.bv_offset, bvec.bv_len, |
4550dd6c | 1622 | bio_iter_last(bvec, iter) ? 0 : MSG_MORE); |
7fae55da AG |
1623 | if (err) |
1624 | return err; | |
9104d31a LE |
1625 | /* REQ_OP_WRITE_SAME has only one segment */ |
1626 | if (bio_op(bio) == REQ_OP_WRITE_SAME) | |
1627 | break; | |
b411b363 | 1628 | } |
7fae55da | 1629 | return 0; |
b411b363 PR |
1630 | } |
1631 | ||
69a22773 | 1632 | static int _drbd_send_zc_ee(struct drbd_peer_device *peer_device, |
db830c46 | 1633 | struct drbd_peer_request *peer_req) |
45bb912b | 1634 | { |
db830c46 AG |
1635 | struct page *page = peer_req->pages; |
1636 | unsigned len = peer_req->i.size; | |
9f69230c | 1637 | int err; |
db830c46 | 1638 | |
ba11ad9a | 1639 | /* hint all but last page with MSG_MORE */ |
45bb912b LE |
1640 | page_chain_for_each(page) { |
1641 | unsigned l = min_t(unsigned, len, PAGE_SIZE); | |
9f69230c | 1642 | |
69a22773 | 1643 | err = _drbd_send_page(peer_device, page, 0, l, |
9f69230c AG |
1644 | page_chain_next(page) ? MSG_MORE : 0); |
1645 | if (err) | |
1646 | return err; | |
45bb912b LE |
1647 | len -= l; |
1648 | } | |
9f69230c | 1649 | return 0; |
45bb912b LE |
1650 | } |
1651 | ||
bb3cc85e MC |
1652 | static u32 bio_flags_to_wire(struct drbd_connection *connection, |
1653 | struct bio *bio) | |
76d2e7ec | 1654 | { |
69a22773 | 1655 | if (connection->agreed_pro_version >= 95) |
1eff9d32 JA |
1656 | return (bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0) | |
1657 | (bio->bi_opf & REQ_FUA ? DP_FUA : 0) | | |
1658 | (bio->bi_opf & REQ_PREFLUSH ? DP_FLUSH : 0) | | |
9104d31a | 1659 | (bio_op(bio) == REQ_OP_WRITE_SAME ? DP_WSAME : 0) | |
45c21793 | 1660 | (bio_op(bio) == REQ_OP_DISCARD ? DP_DISCARD : 0) | |
f31e583a LE |
1661 | (bio_op(bio) == REQ_OP_WRITE_ZEROES ? |
1662 | ((connection->agreed_features & DRBD_FF_WZEROES) ? | |
1663 | (DP_ZEROES |(!(bio->bi_opf & REQ_NOUNMAP) ? DP_DISCARD : 0)) | |
1664 | : DP_DISCARD) | |
1665 | : 0); | |
76d2e7ec | 1666 | else |
1eff9d32 | 1667 | return bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0; |
76d2e7ec PR |
1668 | } |
1669 | ||
9305455a | 1670 | /* Used to send write or TRIM aka REQ_OP_DISCARD requests |
2f632aeb | 1671 | * R_PRIMARY -> Peer (P_DATA, P_TRIM) |
b411b363 | 1672 | */ |
69a22773 | 1673 | int drbd_send_dblock(struct drbd_peer_device *peer_device, struct drbd_request *req) |
b411b363 | 1674 | { |
69a22773 | 1675 | struct drbd_device *device = peer_device->device; |
9f5bdc33 AG |
1676 | struct drbd_socket *sock; |
1677 | struct p_data *p; | |
9104d31a LE |
1678 | struct p_wsame *wsame = NULL; |
1679 | void *digest_out; | |
b411b363 | 1680 | unsigned int dp_flags = 0; |
11f8b2b6 | 1681 | int digest_size; |
9f5bdc33 | 1682 | int err; |
b411b363 | 1683 | |
69a22773 AG |
1684 | sock = &peer_device->connection->data; |
1685 | p = drbd_prepare_command(peer_device, sock); | |
11f8b2b6 | 1686 | digest_size = peer_device->connection->integrity_tfm ? |
3d0e6375 | 1687 | crypto_shash_digestsize(peer_device->connection->integrity_tfm) : 0; |
b411b363 | 1688 | |
9f5bdc33 AG |
1689 | if (!p) |
1690 | return -EIO; | |
1691 | p->sector = cpu_to_be64(req->i.sector); | |
1692 | p->block_id = (unsigned long)req; | |
b30ab791 | 1693 | p->seq_num = cpu_to_be32(atomic_inc_return(&device->packet_seq)); |
bb3cc85e | 1694 | dp_flags = bio_flags_to_wire(peer_device->connection, req->master_bio); |
b30ab791 AG |
1695 | if (device->state.conn >= C_SYNC_SOURCE && |
1696 | device->state.conn <= C_PAUSED_SYNC_T) | |
b411b363 | 1697 | dp_flags |= DP_MAY_SET_IN_SYNC; |
69a22773 | 1698 | if (peer_device->connection->agreed_pro_version >= 100) { |
303d1448 PR |
1699 | if (req->rq_state & RQ_EXP_RECEIVE_ACK) |
1700 | dp_flags |= DP_SEND_RECEIVE_ACK; | |
08d0dabf LE |
1701 | /* During resync, request an explicit write ack, |
1702 | * even in protocol != C */ | |
1703 | if (req->rq_state & RQ_EXP_WRITE_ACK | |
1704 | || (dp_flags & DP_MAY_SET_IN_SYNC)) | |
303d1448 PR |
1705 | dp_flags |= DP_SEND_WRITE_ACK; |
1706 | } | |
9f5bdc33 | 1707 | p->dp_flags = cpu_to_be32(dp_flags); |
2f632aeb | 1708 | |
f31e583a LE |
1709 | if (dp_flags & (DP_DISCARD|DP_ZEROES)) { |
1710 | enum drbd_packet cmd = (dp_flags & DP_ZEROES) ? P_ZEROES : P_TRIM; | |
2f632aeb LE |
1711 | struct p_trim *t = (struct p_trim*)p; |
1712 | t->size = cpu_to_be32(req->i.size); | |
f31e583a | 1713 | err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*t), NULL, 0); |
2f632aeb LE |
1714 | goto out; |
1715 | } | |
9104d31a LE |
1716 | if (dp_flags & DP_WSAME) { |
1717 | /* this will only work if DRBD_FF_WSAME is set AND the | |
1718 | * handshake agreed that all nodes and backend devices are | |
1719 | * WRITE_SAME capable and agree on logical_block_size */ | |
1720 | wsame = (struct p_wsame*)p; | |
1721 | digest_out = wsame + 1; | |
1722 | wsame->size = cpu_to_be32(req->i.size); | |
1723 | } else | |
1724 | digest_out = p + 1; | |
2f632aeb LE |
1725 | |
1726 | /* our digest is still only over the payload. | |
1727 | * TRIM does not carry any payload. */ | |
11f8b2b6 | 1728 | if (digest_size) |
9104d31a LE |
1729 | drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest_out); |
1730 | if (wsame) { | |
1731 | err = | |
1732 | __send_command(peer_device->connection, device->vnr, sock, P_WSAME, | |
1733 | sizeof(*wsame) + digest_size, NULL, | |
1734 | bio_iovec(req->master_bio).bv_len); | |
1735 | } else | |
1736 | err = | |
1737 | __send_command(peer_device->connection, device->vnr, sock, P_DATA, | |
1738 | sizeof(*p) + digest_size, NULL, req->i.size); | |
6bdb9b0e | 1739 | if (!err) { |
470be44a LE |
1740 | /* For protocol A, we have to memcpy the payload into |
1741 | * socket buffers, as we may complete right away | |
1742 | * as soon as we handed it over to tcp, at which point the data | |
1743 | * pages may become invalid. | |
1744 | * | |
1745 | * For data-integrity enabled, we copy it as well, so we can be | |
1746 | * sure that even if the bio pages may still be modified, it | |
1747 | * won't change the data on the wire, thus if the digest checks | |
1748 | * out ok after sending on this side, but does not fit on the | |
1749 | * receiving side, we sure have detected corruption elsewhere. | |
1750 | */ | |
11f8b2b6 | 1751 | if (!(req->rq_state & (RQ_EXP_RECEIVE_ACK | RQ_EXP_WRITE_ACK)) || digest_size) |
69a22773 | 1752 | err = _drbd_send_bio(peer_device, req->master_bio); |
b411b363 | 1753 | else |
69a22773 | 1754 | err = _drbd_send_zc_bio(peer_device, req->master_bio); |
470be44a LE |
1755 | |
1756 | /* double check digest, sometimes buffers have been modified in flight. */ | |
11f8b2b6 | 1757 | if (digest_size > 0 && digest_size <= 64) { |
24c4830c | 1758 | /* 64 byte, 512 bit, is the largest digest size |
470be44a LE |
1759 | * currently supported in kernel crypto. */ |
1760 | unsigned char digest[64]; | |
69a22773 | 1761 | drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest); |
11f8b2b6 | 1762 | if (memcmp(p + 1, digest, digest_size)) { |
d0180171 | 1763 | drbd_warn(device, |
470be44a | 1764 | "Digest mismatch, buffer modified by upper layers during write: %llus +%u\n", |
ace652ac | 1765 | (unsigned long long)req->i.sector, req->i.size); |
470be44a | 1766 | } |
11f8b2b6 | 1767 | } /* else if (digest_size > 64) { |
470be44a LE |
1768 | ... Be noisy about digest too large ... |
1769 | } */ | |
b411b363 | 1770 | } |
2f632aeb | 1771 | out: |
9f5bdc33 | 1772 | mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */ |
b411b363 | 1773 | |
6bdb9b0e | 1774 | return err; |
b411b363 PR |
1775 | } |
1776 | ||
1777 | /* answer packet, used to send data back for read requests: | |
1778 | * Peer -> (diskless) R_PRIMARY (P_DATA_REPLY) | |
1779 | * C_SYNC_SOURCE -> C_SYNC_TARGET (P_RS_DATA_REPLY) | |
1780 | */ | |
69a22773 | 1781 | int drbd_send_block(struct drbd_peer_device *peer_device, enum drbd_packet cmd, |
db830c46 | 1782 | struct drbd_peer_request *peer_req) |
b411b363 | 1783 | { |
69a22773 | 1784 | struct drbd_device *device = peer_device->device; |
9f5bdc33 AG |
1785 | struct drbd_socket *sock; |
1786 | struct p_data *p; | |
7b57b89d | 1787 | int err; |
11f8b2b6 | 1788 | int digest_size; |
b411b363 | 1789 | |
69a22773 AG |
1790 | sock = &peer_device->connection->data; |
1791 | p = drbd_prepare_command(peer_device, sock); | |
b411b363 | 1792 | |
11f8b2b6 | 1793 | digest_size = peer_device->connection->integrity_tfm ? |
3d0e6375 | 1794 | crypto_shash_digestsize(peer_device->connection->integrity_tfm) : 0; |
b411b363 | 1795 | |
9f5bdc33 AG |
1796 | if (!p) |
1797 | return -EIO; | |
1798 | p->sector = cpu_to_be64(peer_req->i.sector); | |
1799 | p->block_id = peer_req->block_id; | |
1800 | p->seq_num = 0; /* unused */ | |
b17f33cb | 1801 | p->dp_flags = 0; |
11f8b2b6 | 1802 | if (digest_size) |
69a22773 | 1803 | drbd_csum_ee(peer_device->connection->integrity_tfm, peer_req, p + 1); |
11f8b2b6 | 1804 | err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*p) + digest_size, NULL, peer_req->i.size); |
7b57b89d | 1805 | if (!err) |
69a22773 | 1806 | err = _drbd_send_zc_ee(peer_device, peer_req); |
9f5bdc33 | 1807 | mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */ |
bd26bfc5 | 1808 | |
7b57b89d | 1809 | return err; |
b411b363 PR |
1810 | } |
1811 | ||
69a22773 | 1812 | int drbd_send_out_of_sync(struct drbd_peer_device *peer_device, struct drbd_request *req) |
73a01a18 | 1813 | { |
9f5bdc33 AG |
1814 | struct drbd_socket *sock; |
1815 | struct p_block_desc *p; | |
73a01a18 | 1816 | |
69a22773 AG |
1817 | sock = &peer_device->connection->data; |
1818 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 AG |
1819 | if (!p) |
1820 | return -EIO; | |
1821 | p->sector = cpu_to_be64(req->i.sector); | |
1822 | p->blksize = cpu_to_be32(req->i.size); | |
69a22773 | 1823 | return drbd_send_command(peer_device, sock, P_OUT_OF_SYNC, sizeof(*p), NULL, 0); |
73a01a18 PR |
1824 | } |
1825 | ||
b411b363 PR |
1826 | /* |
1827 | drbd_send distinguishes two cases: | |
1828 | ||
1829 | Packets sent via the data socket "sock" | |
1830 | and packets sent via the meta data socket "msock" | |
1831 | ||
1832 | sock msock | |
1833 | -----------------+-------------------------+------------------------------ | |
1834 | timeout conf.timeout / 2 conf.timeout / 2 | |
1835 | timeout action send a ping via msock Abort communication | |
1836 | and close all sockets | |
1837 | */ | |
1838 | ||
1839 | /* | |
1840 | * you must have down()ed the appropriate [m]sock_mutex elsewhere! | |
1841 | */ | |
bde89a9e | 1842 | int drbd_send(struct drbd_connection *connection, struct socket *sock, |
b411b363 PR |
1843 | void *buf, size_t size, unsigned msg_flags) |
1844 | { | |
79ab80be | 1845 | struct kvec iov = {.iov_base = buf, .iov_len = size}; |
f7765c36 | 1846 | struct msghdr msg = {.msg_flags = msg_flags | MSG_NOSIGNAL}; |
b411b363 PR |
1847 | int rv, sent = 0; |
1848 | ||
1849 | if (!sock) | |
c0d42c8e | 1850 | return -EBADR; |
b411b363 PR |
1851 | |
1852 | /* THINK if (signal_pending) return ... ? */ | |
1853 | ||
aa563d7b | 1854 | iov_iter_kvec(&msg.msg_iter, WRITE, &iov, 1, size); |
79ab80be | 1855 | |
bde89a9e | 1856 | if (sock == connection->data.socket) { |
44ed167d | 1857 | rcu_read_lock(); |
bde89a9e | 1858 | connection->ko_count = rcu_dereference(connection->net_conf)->ko_count; |
44ed167d | 1859 | rcu_read_unlock(); |
bde89a9e | 1860 | drbd_update_congested(connection); |
b411b363 PR |
1861 | } |
1862 | do { | |
79ab80be | 1863 | rv = sock_sendmsg(sock, &msg); |
b411b363 | 1864 | if (rv == -EAGAIN) { |
bde89a9e | 1865 | if (we_should_drop_the_connection(connection, sock)) |
b411b363 PR |
1866 | break; |
1867 | else | |
1868 | continue; | |
1869 | } | |
b411b363 PR |
1870 | if (rv == -EINTR) { |
1871 | flush_signals(current); | |
1872 | rv = 0; | |
1873 | } | |
1874 | if (rv < 0) | |
1875 | break; | |
1876 | sent += rv; | |
b411b363 PR |
1877 | } while (sent < size); |
1878 | ||
bde89a9e AG |
1879 | if (sock == connection->data.socket) |
1880 | clear_bit(NET_CONGESTED, &connection->flags); | |
b411b363 PR |
1881 | |
1882 | if (rv <= 0) { | |
1883 | if (rv != -EAGAIN) { | |
1ec861eb | 1884 | drbd_err(connection, "%s_sendmsg returned %d\n", |
bde89a9e | 1885 | sock == connection->meta.socket ? "msock" : "sock", |
bedbd2a5 | 1886 | rv); |
bde89a9e | 1887 | conn_request_state(connection, NS(conn, C_BROKEN_PIPE), CS_HARD); |
b411b363 | 1888 | } else |
bde89a9e | 1889 | conn_request_state(connection, NS(conn, C_TIMEOUT), CS_HARD); |
b411b363 PR |
1890 | } |
1891 | ||
1892 | return sent; | |
1893 | } | |
1894 | ||
fb708e40 AG |
1895 | /** |
1896 | * drbd_send_all - Send an entire buffer | |
1897 | * | |
1898 | * Returns 0 upon success and a negative error value otherwise. | |
1899 | */ | |
bde89a9e | 1900 | int drbd_send_all(struct drbd_connection *connection, struct socket *sock, void *buffer, |
fb708e40 AG |
1901 | size_t size, unsigned msg_flags) |
1902 | { | |
1903 | int err; | |
1904 | ||
bde89a9e | 1905 | err = drbd_send(connection, sock, buffer, size, msg_flags); |
fb708e40 AG |
1906 | if (err < 0) |
1907 | return err; | |
1908 | if (err != size) | |
1909 | return -EIO; | |
1910 | return 0; | |
1911 | } | |
1912 | ||
b411b363 PR |
1913 | static int drbd_open(struct block_device *bdev, fmode_t mode) |
1914 | { | |
b30ab791 | 1915 | struct drbd_device *device = bdev->bd_disk->private_data; |
b411b363 PR |
1916 | unsigned long flags; |
1917 | int rv = 0; | |
1918 | ||
2a48fc0a | 1919 | mutex_lock(&drbd_main_mutex); |
0500813f | 1920 | spin_lock_irqsave(&device->resource->req_lock, flags); |
b30ab791 | 1921 | /* to have a stable device->state.role |
b411b363 PR |
1922 | * and no race with updating open_cnt */ |
1923 | ||
b30ab791 | 1924 | if (device->state.role != R_PRIMARY) { |
b411b363 PR |
1925 | if (mode & FMODE_WRITE) |
1926 | rv = -EROFS; | |
183ece30 | 1927 | else if (!drbd_allow_oos) |
b411b363 PR |
1928 | rv = -EMEDIUMTYPE; |
1929 | } | |
1930 | ||
1931 | if (!rv) | |
b30ab791 | 1932 | device->open_cnt++; |
0500813f | 1933 | spin_unlock_irqrestore(&device->resource->req_lock, flags); |
2a48fc0a | 1934 | mutex_unlock(&drbd_main_mutex); |
b411b363 PR |
1935 | |
1936 | return rv; | |
1937 | } | |
1938 | ||
db2a144b | 1939 | static void drbd_release(struct gendisk *gd, fmode_t mode) |
b411b363 | 1940 | { |
b30ab791 | 1941 | struct drbd_device *device = gd->private_data; |
2a48fc0a | 1942 | mutex_lock(&drbd_main_mutex); |
b30ab791 | 1943 | device->open_cnt--; |
2a48fc0a | 1944 | mutex_unlock(&drbd_main_mutex); |
b411b363 PR |
1945 | } |
1946 | ||
c51a0ef3 LE |
1947 | /* need to hold resource->req_lock */ |
1948 | void drbd_queue_unplug(struct drbd_device *device) | |
1949 | { | |
1950 | if (device->state.pdsk >= D_INCONSISTENT && device->state.conn >= C_CONNECTED) { | |
1951 | D_ASSERT(device, device->state.role == R_PRIMARY); | |
1952 | if (test_and_clear_bit(UNPLUG_REMOTE, &device->flags)) { | |
1953 | drbd_queue_work_if_unqueued( | |
1954 | &first_peer_device(device)->connection->sender_work, | |
1955 | &device->unplug_work); | |
1956 | } | |
1957 | } | |
1958 | } | |
1959 | ||
b30ab791 | 1960 | static void drbd_set_defaults(struct drbd_device *device) |
b411b363 | 1961 | { |
f399002e LE |
1962 | /* Beware! The actual layout differs |
1963 | * between big endian and little endian */ | |
b30ab791 | 1964 | device->state = (union drbd_dev_state) { |
b411b363 PR |
1965 | { .role = R_SECONDARY, |
1966 | .peer = R_UNKNOWN, | |
1967 | .conn = C_STANDALONE, | |
1968 | .disk = D_DISKLESS, | |
1969 | .pdsk = D_UNKNOWN, | |
b411b363 PR |
1970 | } }; |
1971 | } | |
1972 | ||
b30ab791 | 1973 | void drbd_init_set_defaults(struct drbd_device *device) |
b411b363 PR |
1974 | { |
1975 | /* the memset(,0,) did most of this. | |
1976 | * note: only assignments, no allocation in here */ | |
1977 | ||
b30ab791 AG |
1978 | drbd_set_defaults(device); |
1979 | ||
1980 | atomic_set(&device->ap_bio_cnt, 0); | |
ad3fee79 | 1981 | atomic_set(&device->ap_actlog_cnt, 0); |
b30ab791 AG |
1982 | atomic_set(&device->ap_pending_cnt, 0); |
1983 | atomic_set(&device->rs_pending_cnt, 0); | |
1984 | atomic_set(&device->unacked_cnt, 0); | |
1985 | atomic_set(&device->local_cnt, 0); | |
1986 | atomic_set(&device->pp_in_use_by_net, 0); | |
1987 | atomic_set(&device->rs_sect_in, 0); | |
1988 | atomic_set(&device->rs_sect_ev, 0); | |
1989 | atomic_set(&device->ap_in_flight, 0); | |
e37d2438 | 1990 | atomic_set(&device->md_io.in_use, 0); |
b30ab791 AG |
1991 | |
1992 | mutex_init(&device->own_state_mutex); | |
1993 | device->state_mutex = &device->own_state_mutex; | |
1994 | ||
1995 | spin_lock_init(&device->al_lock); | |
1996 | spin_lock_init(&device->peer_seq_lock); | |
1997 | ||
1998 | INIT_LIST_HEAD(&device->active_ee); | |
1999 | INIT_LIST_HEAD(&device->sync_ee); | |
2000 | INIT_LIST_HEAD(&device->done_ee); | |
2001 | INIT_LIST_HEAD(&device->read_ee); | |
2002 | INIT_LIST_HEAD(&device->net_ee); | |
2003 | INIT_LIST_HEAD(&device->resync_reads); | |
2004 | INIT_LIST_HEAD(&device->resync_work.list); | |
2005 | INIT_LIST_HEAD(&device->unplug_work.list); | |
b30ab791 | 2006 | INIT_LIST_HEAD(&device->bm_io_work.w.list); |
844a6ae7 LE |
2007 | INIT_LIST_HEAD(&device->pending_master_completion[0]); |
2008 | INIT_LIST_HEAD(&device->pending_master_completion[1]); | |
2009 | INIT_LIST_HEAD(&device->pending_completion[0]); | |
2010 | INIT_LIST_HEAD(&device->pending_completion[1]); | |
b30ab791 AG |
2011 | |
2012 | device->resync_work.cb = w_resync_timer; | |
2013 | device->unplug_work.cb = w_send_write_hint; | |
b30ab791 | 2014 | device->bm_io_work.w.cb = w_bitmap_io; |
b30ab791 | 2015 | |
2bccef39 KC |
2016 | timer_setup(&device->resync_timer, resync_timer_fn, 0); |
2017 | timer_setup(&device->md_sync_timer, md_sync_timer_fn, 0); | |
2018 | timer_setup(&device->start_resync_timer, start_resync_timer_fn, 0); | |
2019 | timer_setup(&device->request_timer, request_timer_fn, 0); | |
b30ab791 AG |
2020 | |
2021 | init_waitqueue_head(&device->misc_wait); | |
2022 | init_waitqueue_head(&device->state_wait); | |
2023 | init_waitqueue_head(&device->ee_wait); | |
2024 | init_waitqueue_head(&device->al_wait); | |
2025 | init_waitqueue_head(&device->seq_wait); | |
2026 | ||
2027 | device->resync_wenr = LC_FREE; | |
2028 | device->peer_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE; | |
2029 | device->local_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE; | |
2030 | } | |
2031 | ||
d5412e8d LE |
2032 | static void _drbd_set_my_capacity(struct drbd_device *device, sector_t size) |
2033 | { | |
2034 | /* set_capacity(device->this_bdev->bd_disk, size); */ | |
2035 | set_capacity(device->vdisk, size); | |
2036 | device->this_bdev->bd_inode->i_size = (loff_t)size << 9; | |
2037 | } | |
2038 | ||
2039 | void drbd_set_my_capacity(struct drbd_device *device, sector_t size) | |
2040 | { | |
2041 | char ppb[10]; | |
2042 | _drbd_set_my_capacity(device, size); | |
2043 | drbd_info(device, "size = %s (%llu KB)\n", | |
2044 | ppsize(ppb, size>>1), (unsigned long long)size>>1); | |
2045 | } | |
2046 | ||
b30ab791 | 2047 | void drbd_device_cleanup(struct drbd_device *device) |
b411b363 | 2048 | { |
1d7734a0 | 2049 | int i; |
a6b32bc3 | 2050 | if (first_peer_device(device)->connection->receiver.t_state != NONE) |
d0180171 | 2051 | drbd_err(device, "ASSERT FAILED: receiver t_state == %d expected 0.\n", |
a6b32bc3 | 2052 | first_peer_device(device)->connection->receiver.t_state); |
b30ab791 AG |
2053 | |
2054 | device->al_writ_cnt = | |
2055 | device->bm_writ_cnt = | |
2056 | device->read_cnt = | |
2057 | device->recv_cnt = | |
2058 | device->send_cnt = | |
2059 | device->writ_cnt = | |
2060 | device->p_size = | |
2061 | device->rs_start = | |
2062 | device->rs_total = | |
2063 | device->rs_failed = 0; | |
2064 | device->rs_last_events = 0; | |
2065 | device->rs_last_sect_ev = 0; | |
1d7734a0 | 2066 | for (i = 0; i < DRBD_SYNC_MARKS; i++) { |
b30ab791 AG |
2067 | device->rs_mark_left[i] = 0; |
2068 | device->rs_mark_time[i] = 0; | |
1d7734a0 | 2069 | } |
0b0ba1ef | 2070 | D_ASSERT(device, first_peer_device(device)->connection->net_conf == NULL); |
b411b363 | 2071 | |
d5412e8d | 2072 | _drbd_set_my_capacity(device, 0); |
b30ab791 | 2073 | if (device->bitmap) { |
b411b363 | 2074 | /* maybe never allocated. */ |
b30ab791 AG |
2075 | drbd_bm_resize(device, 0, 1); |
2076 | drbd_bm_cleanup(device); | |
b411b363 PR |
2077 | } |
2078 | ||
63a7c8ad | 2079 | drbd_backing_dev_free(device, device->ldev); |
b30ab791 | 2080 | device->ldev = NULL; |
1d041225 | 2081 | |
b30ab791 | 2082 | clear_bit(AL_SUSPENDED, &device->flags); |
b411b363 | 2083 | |
0b0ba1ef AG |
2084 | D_ASSERT(device, list_empty(&device->active_ee)); |
2085 | D_ASSERT(device, list_empty(&device->sync_ee)); | |
2086 | D_ASSERT(device, list_empty(&device->done_ee)); | |
2087 | D_ASSERT(device, list_empty(&device->read_ee)); | |
2088 | D_ASSERT(device, list_empty(&device->net_ee)); | |
2089 | D_ASSERT(device, list_empty(&device->resync_reads)); | |
2090 | D_ASSERT(device, list_empty(&first_peer_device(device)->connection->sender_work.q)); | |
2091 | D_ASSERT(device, list_empty(&device->resync_work.list)); | |
2092 | D_ASSERT(device, list_empty(&device->unplug_work.list)); | |
2265b473 | 2093 | |
b30ab791 | 2094 | drbd_set_defaults(device); |
b411b363 PR |
2095 | } |
2096 | ||
2097 | ||
2098 | static void drbd_destroy_mempools(void) | |
2099 | { | |
2100 | struct page *page; | |
2101 | ||
2102 | while (drbd_pp_pool) { | |
2103 | page = drbd_pp_pool; | |
2104 | drbd_pp_pool = (struct page *)page_private(page); | |
2105 | __free_page(page); | |
2106 | drbd_pp_vacant--; | |
2107 | } | |
2108 | ||
0b0ba1ef | 2109 | /* D_ASSERT(device, atomic_read(&drbd_pp_vacant)==0); */ |
b411b363 | 2110 | |
0892fac8 KO |
2111 | bioset_exit(&drbd_io_bio_set); |
2112 | bioset_exit(&drbd_md_io_bio_set); | |
2113 | mempool_exit(&drbd_md_io_page_pool); | |
2114 | mempool_exit(&drbd_ee_mempool); | |
2115 | mempool_exit(&drbd_request_mempool); | |
a12fc00b | 2116 | kmem_cache_destroy(drbd_ee_cache); |
2117 | kmem_cache_destroy(drbd_request_cache); | |
2118 | kmem_cache_destroy(drbd_bm_ext_cache); | |
2119 | kmem_cache_destroy(drbd_al_ext_cache); | |
b411b363 | 2120 | |
b411b363 PR |
2121 | drbd_ee_cache = NULL; |
2122 | drbd_request_cache = NULL; | |
2123 | drbd_bm_ext_cache = NULL; | |
2124 | drbd_al_ext_cache = NULL; | |
2125 | ||
2126 | return; | |
2127 | } | |
2128 | ||
2129 | static int drbd_create_mempools(void) | |
2130 | { | |
2131 | struct page *page; | |
183ece30 | 2132 | const int number = (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * drbd_minor_count; |
0892fac8 | 2133 | int i, ret; |
b411b363 PR |
2134 | |
2135 | /* caches */ | |
2136 | drbd_request_cache = kmem_cache_create( | |
2137 | "drbd_req", sizeof(struct drbd_request), 0, 0, NULL); | |
2138 | if (drbd_request_cache == NULL) | |
2139 | goto Enomem; | |
2140 | ||
2141 | drbd_ee_cache = kmem_cache_create( | |
f6ffca9f | 2142 | "drbd_ee", sizeof(struct drbd_peer_request), 0, 0, NULL); |
b411b363 PR |
2143 | if (drbd_ee_cache == NULL) |
2144 | goto Enomem; | |
2145 | ||
2146 | drbd_bm_ext_cache = kmem_cache_create( | |
2147 | "drbd_bm", sizeof(struct bm_extent), 0, 0, NULL); | |
2148 | if (drbd_bm_ext_cache == NULL) | |
2149 | goto Enomem; | |
2150 | ||
2151 | drbd_al_ext_cache = kmem_cache_create( | |
2152 | "drbd_al", sizeof(struct lc_element), 0, 0, NULL); | |
2153 | if (drbd_al_ext_cache == NULL) | |
2154 | goto Enomem; | |
2155 | ||
2156 | /* mempools */ | |
0892fac8 KO |
2157 | ret = bioset_init(&drbd_io_bio_set, BIO_POOL_SIZE, 0, 0); |
2158 | if (ret) | |
8cb0defb N |
2159 | goto Enomem; |
2160 | ||
0892fac8 KO |
2161 | ret = bioset_init(&drbd_md_io_bio_set, DRBD_MIN_POOL_PAGES, 0, |
2162 | BIOSET_NEED_BVECS); | |
2163 | if (ret) | |
9476f39d | 2164 | goto Enomem; |
9476f39d | 2165 | |
0892fac8 KO |
2166 | ret = mempool_init_page_pool(&drbd_md_io_page_pool, DRBD_MIN_POOL_PAGES, 0); |
2167 | if (ret) | |
4281808f LE |
2168 | goto Enomem; |
2169 | ||
0892fac8 KO |
2170 | ret = mempool_init_slab_pool(&drbd_request_mempool, number, |
2171 | drbd_request_cache); | |
2172 | if (ret) | |
b411b363 PR |
2173 | goto Enomem; |
2174 | ||
0892fac8 KO |
2175 | ret = mempool_init_slab_pool(&drbd_ee_mempool, number, drbd_ee_cache); |
2176 | if (ret) | |
b411b363 PR |
2177 | goto Enomem; |
2178 | ||
2179 | /* drbd's page pool */ | |
2180 | spin_lock_init(&drbd_pp_lock); | |
2181 | ||
2182 | for (i = 0; i < number; i++) { | |
2183 | page = alloc_page(GFP_HIGHUSER); | |
2184 | if (!page) | |
2185 | goto Enomem; | |
2186 | set_page_private(page, (unsigned long)drbd_pp_pool); | |
2187 | drbd_pp_pool = page; | |
2188 | } | |
2189 | drbd_pp_vacant = number; | |
2190 | ||
2191 | return 0; | |
2192 | ||
2193 | Enomem: | |
2194 | drbd_destroy_mempools(); /* in case we allocated some */ | |
2195 | return -ENOMEM; | |
2196 | } | |
2197 | ||
b30ab791 | 2198 | static void drbd_release_all_peer_reqs(struct drbd_device *device) |
b411b363 PR |
2199 | { |
2200 | int rr; | |
2201 | ||
b30ab791 | 2202 | rr = drbd_free_peer_reqs(device, &device->active_ee); |
b411b363 | 2203 | if (rr) |
d0180171 | 2204 | drbd_err(device, "%d EEs in active list found!\n", rr); |
b411b363 | 2205 | |
b30ab791 | 2206 | rr = drbd_free_peer_reqs(device, &device->sync_ee); |
b411b363 | 2207 | if (rr) |
d0180171 | 2208 | drbd_err(device, "%d EEs in sync list found!\n", rr); |
b411b363 | 2209 | |
b30ab791 | 2210 | rr = drbd_free_peer_reqs(device, &device->read_ee); |
b411b363 | 2211 | if (rr) |
d0180171 | 2212 | drbd_err(device, "%d EEs in read list found!\n", rr); |
b411b363 | 2213 | |
b30ab791 | 2214 | rr = drbd_free_peer_reqs(device, &device->done_ee); |
b411b363 | 2215 | if (rr) |
d0180171 | 2216 | drbd_err(device, "%d EEs in done list found!\n", rr); |
b411b363 | 2217 | |
b30ab791 | 2218 | rr = drbd_free_peer_reqs(device, &device->net_ee); |
b411b363 | 2219 | if (rr) |
d0180171 | 2220 | drbd_err(device, "%d EEs in net list found!\n", rr); |
b411b363 PR |
2221 | } |
2222 | ||
774b3055 | 2223 | /* caution. no locking. */ |
05a10ec7 | 2224 | void drbd_destroy_device(struct kref *kref) |
b411b363 | 2225 | { |
b30ab791 | 2226 | struct drbd_device *device = container_of(kref, struct drbd_device, kref); |
803ea134 | 2227 | struct drbd_resource *resource = device->resource; |
a8ba0d60 | 2228 | struct drbd_peer_device *peer_device, *tmp_peer_device; |
b411b363 | 2229 | |
b30ab791 | 2230 | del_timer_sync(&device->request_timer); |
dfa8bedb | 2231 | |
b411b363 | 2232 | /* paranoia asserts */ |
0b0ba1ef | 2233 | D_ASSERT(device, device->open_cnt == 0); |
b411b363 PR |
2234 | /* end paranoia asserts */ |
2235 | ||
b411b363 PR |
2236 | /* cleanup stuff that may have been allocated during |
2237 | * device (re-)configuration or state changes */ | |
2238 | ||
b30ab791 AG |
2239 | if (device->this_bdev) |
2240 | bdput(device->this_bdev); | |
b411b363 | 2241 | |
63a7c8ad | 2242 | drbd_backing_dev_free(device, device->ldev); |
b30ab791 | 2243 | device->ldev = NULL; |
b411b363 | 2244 | |
b30ab791 | 2245 | drbd_release_all_peer_reqs(device); |
b411b363 | 2246 | |
b30ab791 AG |
2247 | lc_destroy(device->act_log); |
2248 | lc_destroy(device->resync); | |
b411b363 | 2249 | |
b30ab791 AG |
2250 | kfree(device->p_uuid); |
2251 | /* device->p_uuid = NULL; */ | |
b411b363 | 2252 | |
b30ab791 AG |
2253 | if (device->bitmap) /* should no longer be there. */ |
2254 | drbd_bm_cleanup(device); | |
e37d2438 | 2255 | __free_page(device->md_io.page); |
b30ab791 AG |
2256 | put_disk(device->vdisk); |
2257 | blk_cleanup_queue(device->rq_queue); | |
2258 | kfree(device->rs_plan_s); | |
a8ba0d60 LE |
2259 | |
2260 | /* not for_each_connection(connection, resource): | |
2261 | * those may have been cleaned up and disassociated already. | |
2262 | */ | |
2263 | for_each_peer_device_safe(peer_device, tmp_peer_device, device) { | |
2264 | kref_put(&peer_device->connection->kref, drbd_destroy_connection); | |
2265 | kfree(peer_device); | |
2266 | } | |
c2258ffc | 2267 | memset(device, 0xfd, sizeof(*device)); |
b30ab791 | 2268 | kfree(device); |
803ea134 | 2269 | kref_put(&resource->kref, drbd_destroy_resource); |
b411b363 PR |
2270 | } |
2271 | ||
2312f0b3 LE |
2272 | /* One global retry thread, if we need to push back some bio and have it |
2273 | * reinserted through our make request function. | |
2274 | */ | |
2275 | static struct retry_worker { | |
2276 | struct workqueue_struct *wq; | |
2277 | struct work_struct worker; | |
2278 | ||
2279 | spinlock_t lock; | |
2280 | struct list_head writes; | |
2281 | } retry; | |
2282 | ||
2283 | static void do_retry(struct work_struct *ws) | |
2284 | { | |
2285 | struct retry_worker *retry = container_of(ws, struct retry_worker, worker); | |
2286 | LIST_HEAD(writes); | |
2287 | struct drbd_request *req, *tmp; | |
2288 | ||
2289 | spin_lock_irq(&retry->lock); | |
2290 | list_splice_init(&retry->writes, &writes); | |
2291 | spin_unlock_irq(&retry->lock); | |
2292 | ||
2293 | list_for_each_entry_safe(req, tmp, &writes, tl_requests) { | |
84b8c06b | 2294 | struct drbd_device *device = req->device; |
2312f0b3 | 2295 | struct bio *bio = req->master_bio; |
e5f891b2 | 2296 | unsigned long start_jif = req->start_jif; |
9a278a79 LE |
2297 | bool expected; |
2298 | ||
84b8c06b | 2299 | expected = |
9a278a79 LE |
2300 | expect(atomic_read(&req->completion_ref) == 0) && |
2301 | expect(req->rq_state & RQ_POSTPONED) && | |
2302 | expect((req->rq_state & RQ_LOCAL_PENDING) == 0 || | |
2303 | (req->rq_state & RQ_LOCAL_ABORTED) != 0); | |
2304 | ||
2305 | if (!expected) | |
d0180171 | 2306 | drbd_err(device, "req=%p completion_ref=%d rq_state=%x\n", |
9a278a79 LE |
2307 | req, atomic_read(&req->completion_ref), |
2308 | req->rq_state); | |
2309 | ||
2310 | /* We still need to put one kref associated with the | |
2311 | * "completion_ref" going zero in the code path that queued it | |
2312 | * here. The request object may still be referenced by a | |
2313 | * frozen local req->private_bio, in case we force-detached. | |
2312f0b3 | 2314 | */ |
9a278a79 | 2315 | kref_put(&req->kref, drbd_req_destroy); |
2312f0b3 LE |
2316 | |
2317 | /* A single suspended or otherwise blocking device may stall | |
2318 | * all others as well. Fortunately, this code path is to | |
2319 | * recover from a situation that "should not happen": | |
2320 | * concurrent writes in multi-primary setup. | |
2321 | * In a "normal" lifecycle, this workqueue is supposed to be | |
2322 | * destroyed without ever doing anything. | |
2323 | * If it turns out to be an issue anyways, we can do per | |
2324 | * resource (replication group) or per device (minor) retry | |
2325 | * workqueues instead. | |
2326 | */ | |
2327 | ||
ed00aabd | 2328 | /* We are not just doing submit_bio_noacct(), |
2312f0b3 | 2329 | * as we want to keep the start_time information. */ |
b30ab791 | 2330 | inc_ap_bio(device); |
e5f891b2 | 2331 | __drbd_make_request(device, bio, start_jif); |
2312f0b3 LE |
2332 | } |
2333 | } | |
2334 | ||
844a6ae7 LE |
2335 | /* called via drbd_req_put_completion_ref(), |
2336 | * holds resource->req_lock */ | |
9d05e7c4 | 2337 | void drbd_restart_request(struct drbd_request *req) |
2312f0b3 LE |
2338 | { |
2339 | unsigned long flags; | |
2340 | spin_lock_irqsave(&retry.lock, flags); | |
2341 | list_move_tail(&req->tl_requests, &retry.writes); | |
2342 | spin_unlock_irqrestore(&retry.lock, flags); | |
2343 | ||
2344 | /* Drop the extra reference that would otherwise | |
2345 | * have been dropped by complete_master_bio. | |
2346 | * do_retry() needs to grab a new one. */ | |
84b8c06b | 2347 | dec_ap_bio(req->device); |
b411b363 | 2348 | |
2312f0b3 | 2349 | queue_work(retry.wq, &retry.worker); |
b411b363 PR |
2350 | } |
2351 | ||
77c556f6 AG |
2352 | void drbd_destroy_resource(struct kref *kref) |
2353 | { | |
2354 | struct drbd_resource *resource = | |
2355 | container_of(kref, struct drbd_resource, kref); | |
2356 | ||
803ea134 | 2357 | idr_destroy(&resource->devices); |
625a6ba2 | 2358 | free_cpumask_var(resource->cpu_mask); |
77c556f6 | 2359 | kfree(resource->name); |
c2258ffc | 2360 | memset(resource, 0xf2, sizeof(*resource)); |
77c556f6 AG |
2361 | kfree(resource); |
2362 | } | |
2363 | ||
2364 | void drbd_free_resource(struct drbd_resource *resource) | |
2365 | { | |
2366 | struct drbd_connection *connection, *tmp; | |
2367 | ||
2368 | for_each_connection_safe(connection, tmp, resource) { | |
2369 | list_del(&connection->connections); | |
4d3d5aa8 | 2370 | drbd_debugfs_connection_cleanup(connection); |
77c556f6 AG |
2371 | kref_put(&connection->kref, drbd_destroy_connection); |
2372 | } | |
4d3d5aa8 | 2373 | drbd_debugfs_resource_cleanup(resource); |
77c556f6 AG |
2374 | kref_put(&resource->kref, drbd_destroy_resource); |
2375 | } | |
2312f0b3 | 2376 | |
b411b363 PR |
2377 | static void drbd_cleanup(void) |
2378 | { | |
2379 | unsigned int i; | |
b30ab791 | 2380 | struct drbd_device *device; |
77c556f6 | 2381 | struct drbd_resource *resource, *tmp; |
b411b363 | 2382 | |
17a93f30 LE |
2383 | /* first remove proc, |
2384 | * drbdsetup uses it's presence to detect | |
2385 | * whether DRBD is loaded. | |
2386 | * If we would get stuck in proc removal, | |
2387 | * but have netlink already deregistered, | |
2388 | * some drbdsetup commands may wait forever | |
2389 | * for an answer. | |
2390 | */ | |
2391 | if (drbd_proc) | |
2392 | remove_proc_entry("drbd", NULL); | |
2393 | ||
2312f0b3 LE |
2394 | if (retry.wq) |
2395 | destroy_workqueue(retry.wq); | |
b411b363 | 2396 | |
3b98c0c2 | 2397 | drbd_genl_unregister(); |
b411b363 | 2398 | |
803ea134 | 2399 | idr_for_each_entry(&drbd_devices, device, i) |
f82795d6 | 2400 | drbd_delete_device(device); |
b411b363 | 2401 | |
c141ebda | 2402 | /* not _rcu since, no other updater anymore. Genl already unregistered */ |
77c556f6 AG |
2403 | for_each_resource_safe(resource, tmp, &drbd_resources) { |
2404 | list_del(&resource->resources); | |
2405 | drbd_free_resource(resource); | |
81fa2e67 | 2406 | } |
b411b363 | 2407 | |
3f1a1b7c LE |
2408 | drbd_debugfs_cleanup(); |
2409 | ||
81a5d60e | 2410 | drbd_destroy_mempools(); |
b411b363 PR |
2411 | unregister_blkdev(DRBD_MAJOR, "drbd"); |
2412 | ||
05a10ec7 | 2413 | idr_destroy(&drbd_devices); |
81a5d60e | 2414 | |
f88c5d90 | 2415 | pr_info("module cleanup done.\n"); |
b411b363 PR |
2416 | } |
2417 | ||
6699b655 PR |
2418 | static void drbd_init_workqueue(struct drbd_work_queue* wq) |
2419 | { | |
6699b655 PR |
2420 | spin_lock_init(&wq->q_lock); |
2421 | INIT_LIST_HEAD(&wq->q); | |
8c0785a5 | 2422 | init_waitqueue_head(&wq->q_wait); |
6699b655 PR |
2423 | } |
2424 | ||
b5043c5e AG |
2425 | struct completion_work { |
2426 | struct drbd_work w; | |
2427 | struct completion done; | |
2428 | }; | |
2429 | ||
2430 | static int w_complete(struct drbd_work *w, int cancel) | |
2431 | { | |
2432 | struct completion_work *completion_work = | |
2433 | container_of(w, struct completion_work, w); | |
2434 | ||
2435 | complete(&completion_work->done); | |
2436 | return 0; | |
2437 | } | |
2438 | ||
2439 | void drbd_flush_workqueue(struct drbd_work_queue *work_queue) | |
2440 | { | |
2441 | struct completion_work completion_work; | |
2442 | ||
2443 | completion_work.w.cb = w_complete; | |
2444 | init_completion(&completion_work.done); | |
2445 | drbd_queue_work(work_queue, &completion_work.w); | |
2446 | wait_for_completion(&completion_work.done); | |
2447 | } | |
2448 | ||
4bc76048 | 2449 | struct drbd_resource *drbd_find_resource(const char *name) |
1aba4d7f | 2450 | { |
77c556f6 | 2451 | struct drbd_resource *resource; |
1aba4d7f | 2452 | |
3b98c0c2 LE |
2453 | if (!name || !name[0]) |
2454 | return NULL; | |
2455 | ||
c141ebda | 2456 | rcu_read_lock(); |
77c556f6 AG |
2457 | for_each_resource_rcu(resource, &drbd_resources) { |
2458 | if (!strcmp(resource->name, name)) { | |
4bc76048 | 2459 | kref_get(&resource->kref); |
1aba4d7f | 2460 | goto found; |
0ace9dfa | 2461 | } |
1aba4d7f | 2462 | } |
4bc76048 | 2463 | resource = NULL; |
1aba4d7f | 2464 | found: |
c141ebda | 2465 | rcu_read_unlock(); |
4bc76048 | 2466 | return resource; |
1aba4d7f PR |
2467 | } |
2468 | ||
bde89a9e | 2469 | struct drbd_connection *conn_get_by_addrs(void *my_addr, int my_addr_len, |
089c075d AG |
2470 | void *peer_addr, int peer_addr_len) |
2471 | { | |
77c556f6 | 2472 | struct drbd_resource *resource; |
bde89a9e | 2473 | struct drbd_connection *connection; |
089c075d AG |
2474 | |
2475 | rcu_read_lock(); | |
77c556f6 AG |
2476 | for_each_resource_rcu(resource, &drbd_resources) { |
2477 | for_each_connection_rcu(connection, resource) { | |
2478 | if (connection->my_addr_len == my_addr_len && | |
2479 | connection->peer_addr_len == peer_addr_len && | |
2480 | !memcmp(&connection->my_addr, my_addr, my_addr_len) && | |
2481 | !memcmp(&connection->peer_addr, peer_addr, peer_addr_len)) { | |
2482 | kref_get(&connection->kref); | |
2483 | goto found; | |
2484 | } | |
089c075d AG |
2485 | } |
2486 | } | |
bde89a9e | 2487 | connection = NULL; |
089c075d AG |
2488 | found: |
2489 | rcu_read_unlock(); | |
bde89a9e | 2490 | return connection; |
089c075d AG |
2491 | } |
2492 | ||
e6ef8a5c AG |
2493 | static int drbd_alloc_socket(struct drbd_socket *socket) |
2494 | { | |
2495 | socket->rbuf = (void *) __get_free_page(GFP_KERNEL); | |
2496 | if (!socket->rbuf) | |
2497 | return -ENOMEM; | |
5a87d920 AG |
2498 | socket->sbuf = (void *) __get_free_page(GFP_KERNEL); |
2499 | if (!socket->sbuf) | |
2500 | return -ENOMEM; | |
e6ef8a5c AG |
2501 | return 0; |
2502 | } | |
2503 | ||
2504 | static void drbd_free_socket(struct drbd_socket *socket) | |
2505 | { | |
5a87d920 | 2506 | free_page((unsigned long) socket->sbuf); |
e6ef8a5c AG |
2507 | free_page((unsigned long) socket->rbuf); |
2508 | } | |
2509 | ||
bde89a9e | 2510 | void conn_free_crypto(struct drbd_connection *connection) |
91fd4dad | 2511 | { |
bde89a9e | 2512 | drbd_free_sock(connection); |
1d041225 | 2513 | |
3d0e6375 KC |
2514 | crypto_free_shash(connection->csums_tfm); |
2515 | crypto_free_shash(connection->verify_tfm); | |
9534d671 | 2516 | crypto_free_shash(connection->cram_hmac_tfm); |
3d0e6375 KC |
2517 | crypto_free_shash(connection->integrity_tfm); |
2518 | crypto_free_shash(connection->peer_integrity_tfm); | |
bde89a9e AG |
2519 | kfree(connection->int_dig_in); |
2520 | kfree(connection->int_dig_vv); | |
1d041225 | 2521 | |
bde89a9e AG |
2522 | connection->csums_tfm = NULL; |
2523 | connection->verify_tfm = NULL; | |
2524 | connection->cram_hmac_tfm = NULL; | |
2525 | connection->integrity_tfm = NULL; | |
2526 | connection->peer_integrity_tfm = NULL; | |
2527 | connection->int_dig_in = NULL; | |
2528 | connection->int_dig_vv = NULL; | |
91fd4dad PR |
2529 | } |
2530 | ||
eb6bea67 | 2531 | int set_resource_options(struct drbd_resource *resource, struct res_opts *res_opts) |
afbbfa88 | 2532 | { |
eb6bea67 | 2533 | struct drbd_connection *connection; |
afbbfa88 AG |
2534 | cpumask_var_t new_cpu_mask; |
2535 | int err; | |
2536 | ||
2537 | if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) | |
2538 | return -ENOMEM; | |
afbbfa88 AG |
2539 | |
2540 | /* silently ignore cpu mask on UP kernel */ | |
2541 | if (nr_cpu_ids > 1 && res_opts->cpu_mask[0] != 0) { | |
f44d0436 | 2542 | err = bitmap_parse(res_opts->cpu_mask, DRBD_CPU_MASK_SIZE, |
c5b005ab | 2543 | cpumask_bits(new_cpu_mask), nr_cpu_ids); |
1e39152f LE |
2544 | if (err == -EOVERFLOW) { |
2545 | /* So what. mask it out. */ | |
2546 | cpumask_var_t tmp_cpu_mask; | |
2547 | if (zalloc_cpumask_var(&tmp_cpu_mask, GFP_KERNEL)) { | |
2548 | cpumask_setall(tmp_cpu_mask); | |
2549 | cpumask_and(new_cpu_mask, new_cpu_mask, tmp_cpu_mask); | |
2550 | drbd_warn(resource, "Overflow in bitmap_parse(%.12s%s), truncating to %u bits\n", | |
2551 | res_opts->cpu_mask, | |
2552 | strlen(res_opts->cpu_mask) > 12 ? "..." : "", | |
2553 | nr_cpu_ids); | |
2554 | free_cpumask_var(tmp_cpu_mask); | |
2555 | err = 0; | |
2556 | } | |
2557 | } | |
afbbfa88 | 2558 | if (err) { |
1ec861eb | 2559 | drbd_warn(resource, "bitmap_parse() failed with %d\n", err); |
afbbfa88 AG |
2560 | /* retcode = ERR_CPU_MASK_PARSE; */ |
2561 | goto fail; | |
2562 | } | |
2563 | } | |
eb6bea67 | 2564 | resource->res_opts = *res_opts; |
625a6ba2 AG |
2565 | if (cpumask_empty(new_cpu_mask)) |
2566 | drbd_calc_cpu_mask(&new_cpu_mask); | |
2567 | if (!cpumask_equal(resource->cpu_mask, new_cpu_mask)) { | |
2568 | cpumask_copy(resource->cpu_mask, new_cpu_mask); | |
2569 | for_each_connection_rcu(connection, resource) { | |
eb6bea67 | 2570 | connection->receiver.reset_cpu_mask = 1; |
1c03e520 | 2571 | connection->ack_receiver.reset_cpu_mask = 1; |
eb6bea67 AG |
2572 | connection->worker.reset_cpu_mask = 1; |
2573 | } | |
afbbfa88 AG |
2574 | } |
2575 | err = 0; | |
2576 | ||
2577 | fail: | |
2578 | free_cpumask_var(new_cpu_mask); | |
2579 | return err; | |
2580 | ||
2581 | } | |
2582 | ||
77c556f6 AG |
2583 | struct drbd_resource *drbd_create_resource(const char *name) |
2584 | { | |
2585 | struct drbd_resource *resource; | |
2586 | ||
6bbf53ca | 2587 | resource = kzalloc(sizeof(struct drbd_resource), GFP_KERNEL); |
77c556f6 | 2588 | if (!resource) |
625a6ba2 | 2589 | goto fail; |
77c556f6 | 2590 | resource->name = kstrdup(name, GFP_KERNEL); |
625a6ba2 AG |
2591 | if (!resource->name) |
2592 | goto fail_free_resource; | |
2593 | if (!zalloc_cpumask_var(&resource->cpu_mask, GFP_KERNEL)) | |
2594 | goto fail_free_name; | |
77c556f6 | 2595 | kref_init(&resource->kref); |
803ea134 | 2596 | idr_init(&resource->devices); |
77c556f6 | 2597 | INIT_LIST_HEAD(&resource->connections); |
f6ba8636 | 2598 | resource->write_ordering = WO_BDEV_FLUSH; |
77c556f6 | 2599 | list_add_tail_rcu(&resource->resources, &drbd_resources); |
0500813f | 2600 | mutex_init(&resource->conf_update); |
9e276872 | 2601 | mutex_init(&resource->adm_mutex); |
0500813f | 2602 | spin_lock_init(&resource->req_lock); |
4d3d5aa8 | 2603 | drbd_debugfs_resource_add(resource); |
77c556f6 | 2604 | return resource; |
625a6ba2 AG |
2605 | |
2606 | fail_free_name: | |
2607 | kfree(resource->name); | |
2608 | fail_free_resource: | |
2609 | kfree(resource); | |
2610 | fail: | |
2611 | return NULL; | |
77c556f6 AG |
2612 | } |
2613 | ||
4d3d5aa8 | 2614 | /* caller must be under adm_mutex */ |
bde89a9e | 2615 | struct drbd_connection *conn_create(const char *name, struct res_opts *res_opts) |
2111438b | 2616 | { |
77c556f6 | 2617 | struct drbd_resource *resource; |
bde89a9e | 2618 | struct drbd_connection *connection; |
2111438b | 2619 | |
bde89a9e AG |
2620 | connection = kzalloc(sizeof(struct drbd_connection), GFP_KERNEL); |
2621 | if (!connection) | |
2111438b PR |
2622 | return NULL; |
2623 | ||
bde89a9e | 2624 | if (drbd_alloc_socket(&connection->data)) |
e6ef8a5c | 2625 | goto fail; |
bde89a9e | 2626 | if (drbd_alloc_socket(&connection->meta)) |
e6ef8a5c AG |
2627 | goto fail; |
2628 | ||
bde89a9e AG |
2629 | connection->current_epoch = kzalloc(sizeof(struct drbd_epoch), GFP_KERNEL); |
2630 | if (!connection->current_epoch) | |
12038a3a | 2631 | goto fail; |
b6dd1a89 | 2632 | |
bde89a9e | 2633 | INIT_LIST_HEAD(&connection->transfer_log); |
b6dd1a89 | 2634 | |
bde89a9e AG |
2635 | INIT_LIST_HEAD(&connection->current_epoch->list); |
2636 | connection->epochs = 1; | |
2637 | spin_lock_init(&connection->epoch_lock); | |
4b0007c0 | 2638 | |
bde89a9e AG |
2639 | connection->send.seen_any_write_yet = false; |
2640 | connection->send.current_epoch_nr = 0; | |
2641 | connection->send.current_epoch_writes = 0; | |
b6dd1a89 | 2642 | |
77c556f6 AG |
2643 | resource = drbd_create_resource(name); |
2644 | if (!resource) | |
2645 | goto fail; | |
2646 | ||
bde89a9e AG |
2647 | connection->cstate = C_STANDALONE; |
2648 | mutex_init(&connection->cstate_mutex); | |
bde89a9e | 2649 | init_waitqueue_head(&connection->ping_wait); |
c06ece6b | 2650 | idr_init(&connection->peer_devices); |
b2fb6dbe | 2651 | |
bde89a9e AG |
2652 | drbd_init_workqueue(&connection->sender_work); |
2653 | mutex_init(&connection->data.mutex); | |
2654 | mutex_init(&connection->meta.mutex); | |
6699b655 | 2655 | |
2457b6d5 AG |
2656 | drbd_thread_init(resource, &connection->receiver, drbd_receiver, "receiver"); |
2657 | connection->receiver.connection = connection; | |
2658 | drbd_thread_init(resource, &connection->worker, drbd_worker, "worker"); | |
2659 | connection->worker.connection = connection; | |
1c03e520 PR |
2660 | drbd_thread_init(resource, &connection->ack_receiver, drbd_ack_receiver, "ack_recv"); |
2661 | connection->ack_receiver.connection = connection; | |
392c8801 | 2662 | |
bde89a9e | 2663 | kref_init(&connection->kref); |
77c556f6 | 2664 | |
77c556f6 | 2665 | connection->resource = resource; |
2111438b | 2666 | |
eb6bea67 AG |
2667 | if (set_resource_options(resource, res_opts)) |
2668 | goto fail_resource; | |
2669 | ||
2670 | kref_get(&resource->kref); | |
2671 | list_add_tail_rcu(&connection->connections, &resource->connections); | |
4d3d5aa8 | 2672 | drbd_debugfs_connection_add(connection); |
bde89a9e | 2673 | return connection; |
2111438b | 2674 | |
eb6bea67 AG |
2675 | fail_resource: |
2676 | list_del(&resource->resources); | |
2677 | drbd_free_resource(resource); | |
2111438b | 2678 | fail: |
bde89a9e | 2679 | kfree(connection->current_epoch); |
bde89a9e AG |
2680 | drbd_free_socket(&connection->meta); |
2681 | drbd_free_socket(&connection->data); | |
bde89a9e | 2682 | kfree(connection); |
2111438b PR |
2683 | return NULL; |
2684 | } | |
2685 | ||
05a10ec7 | 2686 | void drbd_destroy_connection(struct kref *kref) |
2111438b | 2687 | { |
bde89a9e | 2688 | struct drbd_connection *connection = container_of(kref, struct drbd_connection, kref); |
77c556f6 | 2689 | struct drbd_resource *resource = connection->resource; |
9dc9fbb3 | 2690 | |
bde89a9e | 2691 | if (atomic_read(&connection->current_epoch->epoch_size) != 0) |
1ec861eb | 2692 | drbd_err(connection, "epoch_size:%d\n", atomic_read(&connection->current_epoch->epoch_size)); |
bde89a9e | 2693 | kfree(connection->current_epoch); |
12038a3a | 2694 | |
c06ece6b | 2695 | idr_destroy(&connection->peer_devices); |
2111438b | 2696 | |
bde89a9e AG |
2697 | drbd_free_socket(&connection->meta); |
2698 | drbd_free_socket(&connection->data); | |
bde89a9e AG |
2699 | kfree(connection->int_dig_in); |
2700 | kfree(connection->int_dig_vv); | |
c2258ffc | 2701 | memset(connection, 0xfc, sizeof(*connection)); |
bde89a9e | 2702 | kfree(connection); |
77c556f6 | 2703 | kref_put(&resource->kref, drbd_destroy_resource); |
2111438b PR |
2704 | } |
2705 | ||
b30ab791 | 2706 | static int init_submitter(struct drbd_device *device) |
113fef9e LE |
2707 | { |
2708 | /* opencoded create_singlethread_workqueue(), | |
2709 | * to be able to say "drbd%d", ..., minor */ | |
39e91a60 LE |
2710 | device->submit.wq = |
2711 | alloc_ordered_workqueue("drbd%u_submit", WQ_MEM_RECLAIM, device->minor); | |
b30ab791 | 2712 | if (!device->submit.wq) |
113fef9e LE |
2713 | return -ENOMEM; |
2714 | ||
b30ab791 | 2715 | INIT_WORK(&device->submit.worker, do_submit); |
b30ab791 | 2716 | INIT_LIST_HEAD(&device->submit.writes); |
113fef9e LE |
2717 | return 0; |
2718 | } | |
2719 | ||
a910b123 | 2720 | enum drbd_ret_code drbd_create_device(struct drbd_config_context *adm_ctx, unsigned int minor) |
b411b363 | 2721 | { |
a910b123 | 2722 | struct drbd_resource *resource = adm_ctx->resource; |
b6f85ef9 | 2723 | struct drbd_connection *connection; |
b30ab791 | 2724 | struct drbd_device *device; |
b6f85ef9 | 2725 | struct drbd_peer_device *peer_device, *tmp_peer_device; |
b411b363 PR |
2726 | struct gendisk *disk; |
2727 | struct request_queue *q; | |
93e4bf7a | 2728 | int id; |
a910b123 | 2729 | int vnr = adm_ctx->volume; |
8432b314 | 2730 | enum drbd_ret_code err = ERR_NOMEM; |
774b3055 | 2731 | |
b30ab791 AG |
2732 | device = minor_to_device(minor); |
2733 | if (device) | |
179e20b8 | 2734 | return ERR_MINOR_OR_VOLUME_EXISTS; |
b411b363 PR |
2735 | |
2736 | /* GFP_KERNEL, we are outside of all write-out paths */ | |
b30ab791 AG |
2737 | device = kzalloc(sizeof(struct drbd_device), GFP_KERNEL); |
2738 | if (!device) | |
774b3055 | 2739 | return ERR_NOMEM; |
803ea134 AG |
2740 | kref_init(&device->kref); |
2741 | ||
803ea134 AG |
2742 | kref_get(&resource->kref); |
2743 | device->resource = resource; | |
b30ab791 AG |
2744 | device->minor = minor; |
2745 | device->vnr = vnr; | |
b411b363 | 2746 | |
b30ab791 | 2747 | drbd_init_set_defaults(device); |
b411b363 | 2748 | |
c62b37d9 | 2749 | q = blk_alloc_queue(NUMA_NO_NODE); |
b411b363 PR |
2750 | if (!q) |
2751 | goto out_no_q; | |
b30ab791 | 2752 | device->rq_queue = q; |
b411b363 PR |
2753 | |
2754 | disk = alloc_disk(1); | |
2755 | if (!disk) | |
2756 | goto out_no_disk; | |
b30ab791 | 2757 | device->vdisk = disk; |
b411b363 | 2758 | |
81e84650 | 2759 | set_disk_ro(disk, true); |
b411b363 PR |
2760 | |
2761 | disk->queue = q; | |
2762 | disk->major = DRBD_MAJOR; | |
2763 | disk->first_minor = minor; | |
2764 | disk->fops = &drbd_ops; | |
2765 | sprintf(disk->disk_name, "drbd%d", minor); | |
b30ab791 | 2766 | disk->private_data = device; |
b411b363 | 2767 | |
b30ab791 | 2768 | device->this_bdev = bdget(MKDEV(DRBD_MAJOR, minor)); |
b411b363 | 2769 | /* we have no partitions. we contain only ourselves. */ |
b30ab791 | 2770 | device->this_bdev->bd_contains = device->this_bdev; |
b411b363 | 2771 | |
fe8fb75e | 2772 | blk_queue_write_cache(q, true, true); |
99432fcc PR |
2773 | /* Setting the max_hw_sectors to an odd value of 8kibyte here |
2774 | This triggers a max_bio_size message upon first attach or connect */ | |
2775 | blk_queue_max_hw_sectors(q, DRBD_MAX_BIO_SIZE_SAFE >> 8); | |
b411b363 | 2776 | |
e37d2438 LE |
2777 | device->md_io.page = alloc_page(GFP_KERNEL); |
2778 | if (!device->md_io.page) | |
b411b363 PR |
2779 | goto out_no_io_page; |
2780 | ||
b30ab791 | 2781 | if (drbd_bm_init(device)) |
b411b363 | 2782 | goto out_no_bitmap; |
b30ab791 AG |
2783 | device->read_requests = RB_ROOT; |
2784 | device->write_requests = RB_ROOT; | |
b411b363 | 2785 | |
93e4bf7a AG |
2786 | id = idr_alloc(&drbd_devices, device, minor, minor + 1, GFP_KERNEL); |
2787 | if (id < 0) { | |
f221f4bc | 2788 | if (id == -ENOSPC) |
179e20b8 | 2789 | err = ERR_MINOR_OR_VOLUME_EXISTS; |
8432b314 | 2790 | goto out_no_minor_idr; |
81a5d60e | 2791 | } |
803ea134 AG |
2792 | kref_get(&device->kref); |
2793 | ||
2794 | id = idr_alloc(&resource->devices, device, vnr, vnr + 1, GFP_KERNEL); | |
2795 | if (id < 0) { | |
f221f4bc | 2796 | if (id == -ENOSPC) |
179e20b8 | 2797 | err = ERR_MINOR_OR_VOLUME_EXISTS; |
803ea134 AG |
2798 | goto out_idr_remove_minor; |
2799 | } | |
2800 | kref_get(&device->kref); | |
8432b314 | 2801 | |
b6f85ef9 | 2802 | INIT_LIST_HEAD(&device->peer_devices); |
4ce49266 | 2803 | INIT_LIST_HEAD(&device->pending_bitmap_io); |
b6f85ef9 AG |
2804 | for_each_connection(connection, resource) { |
2805 | peer_device = kzalloc(sizeof(struct drbd_peer_device), GFP_KERNEL); | |
2806 | if (!peer_device) | |
2807 | goto out_idr_remove_from_resource; | |
2808 | peer_device->connection = connection; | |
2809 | peer_device->device = device; | |
2810 | ||
2811 | list_add(&peer_device->peer_devices, &device->peer_devices); | |
2812 | kref_get(&device->kref); | |
2813 | ||
2814 | id = idr_alloc(&connection->peer_devices, peer_device, vnr, vnr + 1, GFP_KERNEL); | |
2815 | if (id < 0) { | |
f221f4bc | 2816 | if (id == -ENOSPC) |
b6f85ef9 | 2817 | err = ERR_INVALID_REQUEST; |
b6f85ef9 | 2818 | goto out_idr_remove_from_resource; |
56de2102 | 2819 | } |
b6f85ef9 | 2820 | kref_get(&connection->kref); |
668700b4 | 2821 | INIT_WORK(&peer_device->send_acks_work, drbd_send_acks_wf); |
8432b314 | 2822 | } |
56de2102 | 2823 | |
b30ab791 | 2824 | if (init_submitter(device)) { |
113fef9e | 2825 | err = ERR_NOMEM; |
113fef9e LE |
2826 | goto out_idr_remove_vol; |
2827 | } | |
2828 | ||
774b3055 PR |
2829 | add_disk(disk); |
2830 | ||
2325eb66 | 2831 | /* inherit the connection state */ |
b6f85ef9 | 2832 | device->state.conn = first_connection(resource)->cstate; |
69a22773 AG |
2833 | if (device->state.conn == C_WF_REPORT_PARAMS) { |
2834 | for_each_peer_device(peer_device, device) | |
2835 | drbd_connected(peer_device); | |
2836 | } | |
4d3d5aa8 LE |
2837 | /* move to create_peer_device() */ |
2838 | for_each_peer_device(peer_device, device) | |
2839 | drbd_debugfs_peer_device_add(peer_device); | |
2840 | drbd_debugfs_device_add(device); | |
774b3055 | 2841 | return NO_ERROR; |
b411b363 | 2842 | |
113fef9e | 2843 | out_idr_remove_vol: |
c06ece6b | 2844 | idr_remove(&connection->peer_devices, vnr); |
803ea134 | 2845 | out_idr_remove_from_resource: |
b6f85ef9 | 2846 | for_each_connection(connection, resource) { |
d3e709e6 MW |
2847 | peer_device = idr_remove(&connection->peer_devices, vnr); |
2848 | if (peer_device) | |
b6f85ef9 | 2849 | kref_put(&connection->kref, drbd_destroy_connection); |
b6f85ef9 AG |
2850 | } |
2851 | for_each_peer_device_safe(peer_device, tmp_peer_device, device) { | |
2852 | list_del(&peer_device->peer_devices); | |
2853 | kfree(peer_device); | |
2854 | } | |
803ea134 | 2855 | idr_remove(&resource->devices, vnr); |
8432b314 | 2856 | out_idr_remove_minor: |
93e4bf7a | 2857 | idr_remove(&drbd_devices, minor); |
569083c0 | 2858 | synchronize_rcu(); |
8432b314 | 2859 | out_no_minor_idr: |
b30ab791 | 2860 | drbd_bm_cleanup(device); |
b411b363 | 2861 | out_no_bitmap: |
e37d2438 | 2862 | __free_page(device->md_io.page); |
b411b363 PR |
2863 | out_no_io_page: |
2864 | put_disk(disk); | |
2865 | out_no_disk: | |
2866 | blk_cleanup_queue(q); | |
2867 | out_no_q: | |
803ea134 | 2868 | kref_put(&resource->kref, drbd_destroy_resource); |
a6b32bc3 | 2869 | kfree(device); |
8432b314 | 2870 | return err; |
b411b363 PR |
2871 | } |
2872 | ||
f82795d6 | 2873 | void drbd_delete_device(struct drbd_device *device) |
803ea134 AG |
2874 | { |
2875 | struct drbd_resource *resource = device->resource; | |
2876 | struct drbd_connection *connection; | |
4d3d5aa8 | 2877 | struct drbd_peer_device *peer_device; |
803ea134 | 2878 | |
4d3d5aa8 LE |
2879 | /* move to free_peer_device() */ |
2880 | for_each_peer_device(peer_device, device) | |
2881 | drbd_debugfs_peer_device_cleanup(peer_device); | |
2882 | drbd_debugfs_device_cleanup(device); | |
803ea134 | 2883 | for_each_connection(connection, resource) { |
c06ece6b | 2884 | idr_remove(&connection->peer_devices, device->vnr); |
bdfafc4f | 2885 | kref_put(&device->kref, drbd_destroy_device); |
803ea134 AG |
2886 | } |
2887 | idr_remove(&resource->devices, device->vnr); | |
bdfafc4f | 2888 | kref_put(&device->kref, drbd_destroy_device); |
803ea134 | 2889 | idr_remove(&drbd_devices, device_to_minor(device)); |
bdfafc4f | 2890 | kref_put(&device->kref, drbd_destroy_device); |
803ea134 AG |
2891 | del_gendisk(device->vdisk); |
2892 | synchronize_rcu(); | |
bdfafc4f | 2893 | kref_put(&device->kref, drbd_destroy_device); |
803ea134 AG |
2894 | } |
2895 | ||
8ce953aa | 2896 | static int __init drbd_init(void) |
b411b363 PR |
2897 | { |
2898 | int err; | |
2899 | ||
183ece30 RK |
2900 | if (drbd_minor_count < DRBD_MINOR_COUNT_MIN || drbd_minor_count > DRBD_MINOR_COUNT_MAX) { |
2901 | pr_err("invalid minor_count (%d)\n", drbd_minor_count); | |
b411b363 PR |
2902 | #ifdef MODULE |
2903 | return -EINVAL; | |
2904 | #else | |
5fc1efd5 | 2905 | drbd_minor_count = DRBD_MINOR_COUNT_DEF; |
b411b363 PR |
2906 | #endif |
2907 | } | |
2908 | ||
b411b363 PR |
2909 | err = register_blkdev(DRBD_MAJOR, "drbd"); |
2910 | if (err) { | |
f88c5d90 | 2911 | pr_err("unable to register block device major %d\n", |
b411b363 PR |
2912 | DRBD_MAJOR); |
2913 | return err; | |
2914 | } | |
2915 | ||
b411b363 PR |
2916 | /* |
2917 | * allocate all necessary structs | |
2918 | */ | |
b411b363 PR |
2919 | init_waitqueue_head(&drbd_pp_wait); |
2920 | ||
2921 | drbd_proc = NULL; /* play safe for drbd_cleanup */ | |
05a10ec7 | 2922 | idr_init(&drbd_devices); |
b411b363 | 2923 | |
28bc3b8c | 2924 | mutex_init(&resources_mutex); |
77c556f6 | 2925 | INIT_LIST_HEAD(&drbd_resources); |
69babf05 LE |
2926 | |
2927 | err = drbd_genl_register(); | |
2928 | if (err) { | |
f88c5d90 | 2929 | pr_err("unable to register generic netlink family\n"); |
69babf05 LE |
2930 | goto fail; |
2931 | } | |
2932 | ||
b411b363 PR |
2933 | err = drbd_create_mempools(); |
2934 | if (err) | |
3b98c0c2 | 2935 | goto fail; |
b411b363 | 2936 | |
6110d70b | 2937 | err = -ENOMEM; |
cf626b0d | 2938 | drbd_proc = proc_create_single("drbd", S_IFREG | 0444 , NULL, drbd_seq_show); |
b411b363 | 2939 | if (!drbd_proc) { |
f88c5d90 | 2940 | pr_err("unable to register proc file\n"); |
3b98c0c2 | 2941 | goto fail; |
b411b363 PR |
2942 | } |
2943 | ||
2312f0b3 LE |
2944 | retry.wq = create_singlethread_workqueue("drbd-reissue"); |
2945 | if (!retry.wq) { | |
f88c5d90 | 2946 | pr_err("unable to create retry workqueue\n"); |
2312f0b3 LE |
2947 | goto fail; |
2948 | } | |
2949 | INIT_WORK(&retry.worker, do_retry); | |
2950 | spin_lock_init(&retry.lock); | |
2951 | INIT_LIST_HEAD(&retry.writes); | |
b411b363 | 2952 | |
d27e84a3 | 2953 | drbd_debugfs_init(); |
4d3d5aa8 | 2954 | |
f88c5d90 | 2955 | pr_info("initialized. " |
b411b363 PR |
2956 | "Version: " REL_VERSION " (api:%d/proto:%d-%d)\n", |
2957 | API_VERSION, PRO_VERSION_MIN, PRO_VERSION_MAX); | |
f88c5d90 LE |
2958 | pr_info("%s\n", drbd_buildtag()); |
2959 | pr_info("registered as block device major %d\n", DRBD_MAJOR); | |
b411b363 PR |
2960 | return 0; /* Success! */ |
2961 | ||
3b98c0c2 | 2962 | fail: |
b411b363 PR |
2963 | drbd_cleanup(); |
2964 | if (err == -ENOMEM) | |
f88c5d90 | 2965 | pr_err("ran out of memory\n"); |
b411b363 | 2966 | else |
f88c5d90 | 2967 | pr_err("initialization failure\n"); |
b411b363 PR |
2968 | return err; |
2969 | } | |
2970 | ||
f418815f | 2971 | static void drbd_free_one_sock(struct drbd_socket *ds) |
b411b363 | 2972 | { |
f418815f LE |
2973 | struct socket *s; |
2974 | mutex_lock(&ds->mutex); | |
2975 | s = ds->socket; | |
2976 | ds->socket = NULL; | |
2977 | mutex_unlock(&ds->mutex); | |
2978 | if (s) { | |
2979 | /* so debugfs does not need to mutex_lock() */ | |
2980 | synchronize_rcu(); | |
2981 | kernel_sock_shutdown(s, SHUT_RDWR); | |
2982 | sock_release(s); | |
b411b363 PR |
2983 | } |
2984 | } | |
2985 | ||
f418815f LE |
2986 | void drbd_free_sock(struct drbd_connection *connection) |
2987 | { | |
2988 | if (connection->data.socket) | |
2989 | drbd_free_one_sock(&connection->data); | |
2990 | if (connection->meta.socket) | |
2991 | drbd_free_one_sock(&connection->meta); | |
2992 | } | |
2993 | ||
b411b363 | 2994 | /* meta data management */ |
b411b363 | 2995 | |
bde89a9e | 2996 | void conn_md_sync(struct drbd_connection *connection) |
b411b363 | 2997 | { |
c06ece6b | 2998 | struct drbd_peer_device *peer_device; |
19fffd7b | 2999 | int vnr; |
b411b363 | 3000 | |
19fffd7b | 3001 | rcu_read_lock(); |
c06ece6b AG |
3002 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
3003 | struct drbd_device *device = peer_device->device; | |
3004 | ||
b30ab791 | 3005 | kref_get(&device->kref); |
19fffd7b | 3006 | rcu_read_unlock(); |
b30ab791 | 3007 | drbd_md_sync(device); |
05a10ec7 | 3008 | kref_put(&device->kref, drbd_destroy_device); |
19fffd7b PR |
3009 | rcu_read_lock(); |
3010 | } | |
3011 | rcu_read_unlock(); | |
b411b363 PR |
3012 | } |
3013 | ||
ae8bf312 | 3014 | /* aligned 4kByte */ |
b411b363 | 3015 | struct meta_data_on_disk { |
cccac985 | 3016 | u64 la_size_sect; /* last agreed size. */ |
b411b363 PR |
3017 | u64 uuid[UI_SIZE]; /* UUIDs. */ |
3018 | u64 device_uuid; | |
3019 | u64 reserved_u64_1; | |
3020 | u32 flags; /* MDF */ | |
3021 | u32 magic; | |
3022 | u32 md_size_sect; | |
3023 | u32 al_offset; /* offset to this block */ | |
ae8bf312 | 3024 | u32 al_nr_extents; /* important for restoring the AL (userspace) */ |
f399002e | 3025 | /* `-- act_log->nr_elements <-- ldev->dc.al_extents */ |
b411b363 PR |
3026 | u32 bm_offset; /* offset to the bitmap, from here */ |
3027 | u32 bm_bytes_per_bit; /* BM_BLOCK_SIZE */ | |
99432fcc | 3028 | u32 la_peer_max_bio_size; /* last peer max_bio_size */ |
b411b363 | 3029 | |
3a4d4eb3 LE |
3030 | /* see al_tr_number_to_on_disk_sector() */ |
3031 | u32 al_stripes; | |
3032 | u32 al_stripe_size_4k; | |
3033 | ||
3034 | u8 reserved_u8[4096 - (7*8 + 10*4)]; | |
b411b363 PR |
3035 | } __packed; |
3036 | ||
d752b269 PR |
3037 | |
3038 | ||
b30ab791 | 3039 | void drbd_md_write(struct drbd_device *device, void *b) |
b411b363 | 3040 | { |
d752b269 | 3041 | struct meta_data_on_disk *buffer = b; |
b411b363 PR |
3042 | sector_t sector; |
3043 | int i; | |
3044 | ||
ae8bf312 | 3045 | memset(buffer, 0, sizeof(*buffer)); |
b411b363 | 3046 | |
b30ab791 | 3047 | buffer->la_size_sect = cpu_to_be64(drbd_get_capacity(device->this_bdev)); |
b411b363 | 3048 | for (i = UI_CURRENT; i < UI_SIZE; i++) |
b30ab791 AG |
3049 | buffer->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]); |
3050 | buffer->flags = cpu_to_be32(device->ldev->md.flags); | |
d5d7ebd4 | 3051 | buffer->magic = cpu_to_be32(DRBD_MD_MAGIC_84_UNCLEAN); |
b411b363 | 3052 | |
b30ab791 AG |
3053 | buffer->md_size_sect = cpu_to_be32(device->ldev->md.md_size_sect); |
3054 | buffer->al_offset = cpu_to_be32(device->ldev->md.al_offset); | |
3055 | buffer->al_nr_extents = cpu_to_be32(device->act_log->nr_elements); | |
b411b363 | 3056 | buffer->bm_bytes_per_bit = cpu_to_be32(BM_BLOCK_SIZE); |
b30ab791 | 3057 | buffer->device_uuid = cpu_to_be64(device->ldev->md.device_uuid); |
b411b363 | 3058 | |
b30ab791 AG |
3059 | buffer->bm_offset = cpu_to_be32(device->ldev->md.bm_offset); |
3060 | buffer->la_peer_max_bio_size = cpu_to_be32(device->peer_max_bio_size); | |
b411b363 | 3061 | |
b30ab791 AG |
3062 | buffer->al_stripes = cpu_to_be32(device->ldev->md.al_stripes); |
3063 | buffer->al_stripe_size_4k = cpu_to_be32(device->ldev->md.al_stripe_size_4k); | |
3a4d4eb3 | 3064 | |
0b0ba1ef | 3065 | D_ASSERT(device, drbd_md_ss(device->ldev) == device->ldev->md.md_offset); |
b30ab791 | 3066 | sector = device->ldev->md.md_offset; |
b411b363 | 3067 | |
bb3cc85e | 3068 | if (drbd_md_sync_page_io(device, device->ldev, sector, REQ_OP_WRITE)) { |
b411b363 | 3069 | /* this was a try anyways ... */ |
d0180171 | 3070 | drbd_err(device, "meta data update failed!\n"); |
b30ab791 | 3071 | drbd_chk_io_error(device, 1, DRBD_META_IO_ERROR); |
b411b363 | 3072 | } |
d752b269 PR |
3073 | } |
3074 | ||
3075 | /** | |
3076 | * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set | |
b30ab791 | 3077 | * @device: DRBD device. |
d752b269 | 3078 | */ |
b30ab791 | 3079 | void drbd_md_sync(struct drbd_device *device) |
d752b269 PR |
3080 | { |
3081 | struct meta_data_on_disk *buffer; | |
3082 | ||
3083 | /* Don't accidentally change the DRBD meta data layout. */ | |
3084 | BUILD_BUG_ON(UI_SIZE != 4); | |
3085 | BUILD_BUG_ON(sizeof(struct meta_data_on_disk) != 4096); | |
3086 | ||
b30ab791 | 3087 | del_timer(&device->md_sync_timer); |
d752b269 | 3088 | /* timer may be rearmed by drbd_md_mark_dirty() now. */ |
b30ab791 | 3089 | if (!test_and_clear_bit(MD_DIRTY, &device->flags)) |
d752b269 PR |
3090 | return; |
3091 | ||
3092 | /* We use here D_FAILED and not D_ATTACHING because we try to write | |
3093 | * metadata even if we detach due to a disk failure! */ | |
b30ab791 | 3094 | if (!get_ldev_if_state(device, D_FAILED)) |
d752b269 PR |
3095 | return; |
3096 | ||
e37d2438 | 3097 | buffer = drbd_md_get_buffer(device, __func__); |
d752b269 PR |
3098 | if (!buffer) |
3099 | goto out; | |
3100 | ||
b30ab791 | 3101 | drbd_md_write(device, buffer); |
b411b363 | 3102 | |
b30ab791 | 3103 | /* Update device->ldev->md.la_size_sect, |
b411b363 | 3104 | * since we updated it on metadata. */ |
b30ab791 | 3105 | device->ldev->md.la_size_sect = drbd_get_capacity(device->this_bdev); |
b411b363 | 3106 | |
b30ab791 | 3107 | drbd_md_put_buffer(device); |
e1711731 | 3108 | out: |
b30ab791 | 3109 | put_ldev(device); |
b411b363 PR |
3110 | } |
3111 | ||
b30ab791 | 3112 | static int check_activity_log_stripe_size(struct drbd_device *device, |
3a4d4eb3 LE |
3113 | struct meta_data_on_disk *on_disk, |
3114 | struct drbd_md *in_core) | |
3115 | { | |
3116 | u32 al_stripes = be32_to_cpu(on_disk->al_stripes); | |
3117 | u32 al_stripe_size_4k = be32_to_cpu(on_disk->al_stripe_size_4k); | |
3118 | u64 al_size_4k; | |
3119 | ||
3120 | /* both not set: default to old fixed size activity log */ | |
3121 | if (al_stripes == 0 && al_stripe_size_4k == 0) { | |
3122 | al_stripes = 1; | |
3123 | al_stripe_size_4k = MD_32kB_SECT/8; | |
3124 | } | |
3125 | ||
3126 | /* some paranoia plausibility checks */ | |
3127 | ||
3128 | /* we need both values to be set */ | |
3129 | if (al_stripes == 0 || al_stripe_size_4k == 0) | |
3130 | goto err; | |
3131 | ||
3132 | al_size_4k = (u64)al_stripes * al_stripe_size_4k; | |
3133 | ||
3134 | /* Upper limit of activity log area, to avoid potential overflow | |
3135 | * problems in al_tr_number_to_on_disk_sector(). As right now, more | |
3136 | * than 72 * 4k blocks total only increases the amount of history, | |
3137 | * limiting this arbitrarily to 16 GB is not a real limitation ;-) */ | |
3138 | if (al_size_4k > (16 * 1024 * 1024/4)) | |
3139 | goto err; | |
3140 | ||
3141 | /* Lower limit: we need at least 8 transaction slots (32kB) | |
3142 | * to not break existing setups */ | |
3143 | if (al_size_4k < MD_32kB_SECT/8) | |
3144 | goto err; | |
3145 | ||
3146 | in_core->al_stripe_size_4k = al_stripe_size_4k; | |
3147 | in_core->al_stripes = al_stripes; | |
3148 | in_core->al_size_4k = al_size_4k; | |
3149 | ||
3150 | return 0; | |
3151 | err: | |
d0180171 | 3152 | drbd_err(device, "invalid activity log striping: al_stripes=%u, al_stripe_size_4k=%u\n", |
3a4d4eb3 LE |
3153 | al_stripes, al_stripe_size_4k); |
3154 | return -EINVAL; | |
3155 | } | |
3156 | ||
b30ab791 | 3157 | static int check_offsets_and_sizes(struct drbd_device *device, struct drbd_backing_dev *bdev) |
c04ccaa6 LE |
3158 | { |
3159 | sector_t capacity = drbd_get_capacity(bdev->md_bdev); | |
3160 | struct drbd_md *in_core = &bdev->md; | |
3161 | s32 on_disk_al_sect; | |
3162 | s32 on_disk_bm_sect; | |
3163 | ||
3164 | /* The on-disk size of the activity log, calculated from offsets, and | |
3165 | * the size of the activity log calculated from the stripe settings, | |
3166 | * should match. | |
3167 | * Though we could relax this a bit: it is ok, if the striped activity log | |
3168 | * fits in the available on-disk activity log size. | |
3169 | * Right now, that would break how resize is implemented. | |
3170 | * TODO: make drbd_determine_dev_size() (and the drbdmeta tool) aware | |
3171 | * of possible unused padding space in the on disk layout. */ | |
3172 | if (in_core->al_offset < 0) { | |
3173 | if (in_core->bm_offset > in_core->al_offset) | |
3174 | goto err; | |
3175 | on_disk_al_sect = -in_core->al_offset; | |
3176 | on_disk_bm_sect = in_core->al_offset - in_core->bm_offset; | |
3177 | } else { | |
3178 | if (in_core->al_offset != MD_4kB_SECT) | |
3179 | goto err; | |
3180 | if (in_core->bm_offset < in_core->al_offset + in_core->al_size_4k * MD_4kB_SECT) | |
3181 | goto err; | |
3182 | ||
3183 | on_disk_al_sect = in_core->bm_offset - MD_4kB_SECT; | |
3184 | on_disk_bm_sect = in_core->md_size_sect - in_core->bm_offset; | |
3185 | } | |
3186 | ||
3187 | /* old fixed size meta data is exactly that: fixed. */ | |
3188 | if (in_core->meta_dev_idx >= 0) { | |
3189 | if (in_core->md_size_sect != MD_128MB_SECT | |
3190 | || in_core->al_offset != MD_4kB_SECT | |
3191 | || in_core->bm_offset != MD_4kB_SECT + MD_32kB_SECT | |
3192 | || in_core->al_stripes != 1 | |
3193 | || in_core->al_stripe_size_4k != MD_32kB_SECT/8) | |
3194 | goto err; | |
3195 | } | |
3196 | ||
3197 | if (capacity < in_core->md_size_sect) | |
3198 | goto err; | |
3199 | if (capacity - in_core->md_size_sect < drbd_md_first_sector(bdev)) | |
3200 | goto err; | |
3201 | ||
3202 | /* should be aligned, and at least 32k */ | |
3203 | if ((on_disk_al_sect & 7) || (on_disk_al_sect < MD_32kB_SECT)) | |
3204 | goto err; | |
3205 | ||
3206 | /* should fit (for now: exactly) into the available on-disk space; | |
3207 | * overflow prevention is in check_activity_log_stripe_size() above. */ | |
3208 | if (on_disk_al_sect != in_core->al_size_4k * MD_4kB_SECT) | |
3209 | goto err; | |
3210 | ||
3211 | /* again, should be aligned */ | |
3212 | if (in_core->bm_offset & 7) | |
3213 | goto err; | |
3214 | ||
3215 | /* FIXME check for device grow with flex external meta data? */ | |
3216 | ||
3217 | /* can the available bitmap space cover the last agreed device size? */ | |
3218 | if (on_disk_bm_sect < (in_core->la_size_sect+7)/MD_4kB_SECT/8/512) | |
3219 | goto err; | |
3220 | ||
3221 | return 0; | |
3222 | ||
3223 | err: | |
d0180171 | 3224 | drbd_err(device, "meta data offsets don't make sense: idx=%d " |
c04ccaa6 LE |
3225 | "al_s=%u, al_sz4k=%u, al_offset=%d, bm_offset=%d, " |
3226 | "md_size_sect=%u, la_size=%llu, md_capacity=%llu\n", | |
3227 | in_core->meta_dev_idx, | |
3228 | in_core->al_stripes, in_core->al_stripe_size_4k, | |
3229 | in_core->al_offset, in_core->bm_offset, in_core->md_size_sect, | |
3230 | (unsigned long long)in_core->la_size_sect, | |
3231 | (unsigned long long)capacity); | |
3232 | ||
3233 | return -EINVAL; | |
3234 | } | |
3235 | ||
3236 | ||
b411b363 PR |
3237 | /** |
3238 | * drbd_md_read() - Reads in the meta data super block | |
b30ab791 | 3239 | * @device: DRBD device. |
b411b363 PR |
3240 | * @bdev: Device from which the meta data should be read in. |
3241 | * | |
3a4d4eb3 | 3242 | * Return NO_ERROR on success, and an enum drbd_ret_code in case |
d5d7ebd4 | 3243 | * something goes wrong. |
3a4d4eb3 | 3244 | * |
c04ccaa6 | 3245 | * Called exactly once during drbd_adm_attach(), while still being D_DISKLESS, |
b30ab791 | 3246 | * even before @bdev is assigned to @device->ldev. |
b411b363 | 3247 | */ |
b30ab791 | 3248 | int drbd_md_read(struct drbd_device *device, struct drbd_backing_dev *bdev) |
b411b363 PR |
3249 | { |
3250 | struct meta_data_on_disk *buffer; | |
d5d7ebd4 | 3251 | u32 magic, flags; |
b411b363 PR |
3252 | int i, rv = NO_ERROR; |
3253 | ||
b30ab791 | 3254 | if (device->state.disk != D_DISKLESS) |
c04ccaa6 | 3255 | return ERR_DISK_CONFIGURED; |
b411b363 | 3256 | |
e37d2438 | 3257 | buffer = drbd_md_get_buffer(device, __func__); |
e1711731 | 3258 | if (!buffer) |
c04ccaa6 | 3259 | return ERR_NOMEM; |
b411b363 | 3260 | |
c04ccaa6 LE |
3261 | /* First, figure out where our meta data superblock is located, |
3262 | * and read it. */ | |
3a4d4eb3 LE |
3263 | bdev->md.meta_dev_idx = bdev->disk_conf->meta_dev_idx; |
3264 | bdev->md.md_offset = drbd_md_ss(bdev); | |
edb5e5f6 LE |
3265 | /* Even for (flexible or indexed) external meta data, |
3266 | * initially restrict us to the 4k superblock for now. | |
3267 | * Affects the paranoia out-of-range access check in drbd_md_sync_page_io(). */ | |
3268 | bdev->md.md_size_sect = 8; | |
b411b363 | 3269 | |
bb3cc85e MC |
3270 | if (drbd_md_sync_page_io(device, bdev, bdev->md.md_offset, |
3271 | REQ_OP_READ)) { | |
25985edc | 3272 | /* NOTE: can't do normal error processing here as this is |
b411b363 | 3273 | called BEFORE disk is attached */ |
d0180171 | 3274 | drbd_err(device, "Error while reading metadata.\n"); |
b411b363 PR |
3275 | rv = ERR_IO_MD_DISK; |
3276 | goto err; | |
3277 | } | |
3278 | ||
d5d7ebd4 LE |
3279 | magic = be32_to_cpu(buffer->magic); |
3280 | flags = be32_to_cpu(buffer->flags); | |
3281 | if (magic == DRBD_MD_MAGIC_84_UNCLEAN || | |
3282 | (magic == DRBD_MD_MAGIC_08 && !(flags & MDF_AL_CLEAN))) { | |
3283 | /* btw: that's Activity Log clean, not "all" clean. */ | |
d0180171 | 3284 | drbd_err(device, "Found unclean meta data. Did you \"drbdadm apply-al\"?\n"); |
d5d7ebd4 LE |
3285 | rv = ERR_MD_UNCLEAN; |
3286 | goto err; | |
3287 | } | |
3a4d4eb3 LE |
3288 | |
3289 | rv = ERR_MD_INVALID; | |
d5d7ebd4 | 3290 | if (magic != DRBD_MD_MAGIC_08) { |
43de7c85 | 3291 | if (magic == DRBD_MD_MAGIC_07) |
d0180171 | 3292 | drbd_err(device, "Found old (0.7) meta data magic. Did you \"drbdadm create-md\"?\n"); |
d5d7ebd4 | 3293 | else |
d0180171 | 3294 | drbd_err(device, "Meta data magic not found. Did you \"drbdadm create-md\"?\n"); |
b411b363 PR |
3295 | goto err; |
3296 | } | |
3a4d4eb3 | 3297 | |
c04ccaa6 | 3298 | if (be32_to_cpu(buffer->bm_bytes_per_bit) != BM_BLOCK_SIZE) { |
d0180171 | 3299 | drbd_err(device, "unexpected bm_bytes_per_bit: %u (expected %u)\n", |
c04ccaa6 | 3300 | be32_to_cpu(buffer->bm_bytes_per_bit), BM_BLOCK_SIZE); |
b411b363 PR |
3301 | goto err; |
3302 | } | |
3a4d4eb3 | 3303 | |
c04ccaa6 LE |
3304 | |
3305 | /* convert to in_core endian */ | |
3306 | bdev->md.la_size_sect = be64_to_cpu(buffer->la_size_sect); | |
3307 | for (i = UI_CURRENT; i < UI_SIZE; i++) | |
3308 | bdev->md.uuid[i] = be64_to_cpu(buffer->uuid[i]); | |
3309 | bdev->md.flags = be32_to_cpu(buffer->flags); | |
3310 | bdev->md.device_uuid = be64_to_cpu(buffer->device_uuid); | |
3311 | ||
3312 | bdev->md.md_size_sect = be32_to_cpu(buffer->md_size_sect); | |
3313 | bdev->md.al_offset = be32_to_cpu(buffer->al_offset); | |
3314 | bdev->md.bm_offset = be32_to_cpu(buffer->bm_offset); | |
3315 | ||
b30ab791 | 3316 | if (check_activity_log_stripe_size(device, buffer, &bdev->md)) |
b411b363 | 3317 | goto err; |
b30ab791 | 3318 | if (check_offsets_and_sizes(device, bdev)) |
c04ccaa6 LE |
3319 | goto err; |
3320 | ||
b411b363 | 3321 | if (be32_to_cpu(buffer->bm_offset) != bdev->md.bm_offset) { |
d0180171 | 3322 | drbd_err(device, "unexpected bm_offset: %d (expected %d)\n", |
b411b363 | 3323 | be32_to_cpu(buffer->bm_offset), bdev->md.bm_offset); |
b411b363 PR |
3324 | goto err; |
3325 | } | |
3326 | if (be32_to_cpu(buffer->md_size_sect) != bdev->md.md_size_sect) { | |
d0180171 | 3327 | drbd_err(device, "unexpected md_size: %u (expected %u)\n", |
b411b363 | 3328 | be32_to_cpu(buffer->md_size_sect), bdev->md.md_size_sect); |
b411b363 PR |
3329 | goto err; |
3330 | } | |
3331 | ||
3a4d4eb3 | 3332 | rv = NO_ERROR; |
b411b363 | 3333 | |
0500813f | 3334 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 3335 | if (device->state.conn < C_CONNECTED) { |
db141b2f | 3336 | unsigned int peer; |
99432fcc | 3337 | peer = be32_to_cpu(buffer->la_peer_max_bio_size); |
db141b2f | 3338 | peer = max(peer, DRBD_MAX_BIO_SIZE_SAFE); |
b30ab791 | 3339 | device->peer_max_bio_size = peer; |
99432fcc | 3340 | } |
0500813f | 3341 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 PR |
3342 | |
3343 | err: | |
b30ab791 | 3344 | drbd_md_put_buffer(device); |
b411b363 PR |
3345 | |
3346 | return rv; | |
3347 | } | |
3348 | ||
3349 | /** | |
3350 | * drbd_md_mark_dirty() - Mark meta data super block as dirty | |
b30ab791 | 3351 | * @device: DRBD device. |
b411b363 PR |
3352 | * |
3353 | * Call this function if you change anything that should be written to | |
3354 | * the meta-data super block. This function sets MD_DIRTY, and starts a | |
3355 | * timer that ensures that within five seconds you have to call drbd_md_sync(). | |
3356 | */ | |
b30ab791 | 3357 | void drbd_md_mark_dirty(struct drbd_device *device) |
b411b363 | 3358 | { |
b30ab791 AG |
3359 | if (!test_and_set_bit(MD_DIRTY, &device->flags)) |
3360 | mod_timer(&device->md_sync_timer, jiffies + 5*HZ); | |
b411b363 | 3361 | } |
b411b363 | 3362 | |
b30ab791 | 3363 | void drbd_uuid_move_history(struct drbd_device *device) __must_hold(local) |
b411b363 PR |
3364 | { |
3365 | int i; | |
3366 | ||
62b0da3a | 3367 | for (i = UI_HISTORY_START; i < UI_HISTORY_END; i++) |
b30ab791 | 3368 | device->ldev->md.uuid[i+1] = device->ldev->md.uuid[i]; |
b411b363 PR |
3369 | } |
3370 | ||
b30ab791 | 3371 | void __drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local) |
b411b363 PR |
3372 | { |
3373 | if (idx == UI_CURRENT) { | |
b30ab791 | 3374 | if (device->state.role == R_PRIMARY) |
b411b363 PR |
3375 | val |= 1; |
3376 | else | |
3377 | val &= ~((u64)1); | |
3378 | ||
b30ab791 | 3379 | drbd_set_ed_uuid(device, val); |
b411b363 PR |
3380 | } |
3381 | ||
b30ab791 AG |
3382 | device->ldev->md.uuid[idx] = val; |
3383 | drbd_md_mark_dirty(device); | |
b411b363 PR |
3384 | } |
3385 | ||
b30ab791 | 3386 | void _drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local) |
9f2247bb PR |
3387 | { |
3388 | unsigned long flags; | |
b30ab791 AG |
3389 | spin_lock_irqsave(&device->ldev->md.uuid_lock, flags); |
3390 | __drbd_uuid_set(device, idx, val); | |
3391 | spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags); | |
9f2247bb | 3392 | } |
b411b363 | 3393 | |
b30ab791 | 3394 | void drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local) |
b411b363 | 3395 | { |
9f2247bb | 3396 | unsigned long flags; |
b30ab791 AG |
3397 | spin_lock_irqsave(&device->ldev->md.uuid_lock, flags); |
3398 | if (device->ldev->md.uuid[idx]) { | |
3399 | drbd_uuid_move_history(device); | |
3400 | device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[idx]; | |
b411b363 | 3401 | } |
b30ab791 AG |
3402 | __drbd_uuid_set(device, idx, val); |
3403 | spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags); | |
b411b363 PR |
3404 | } |
3405 | ||
3406 | /** | |
3407 | * drbd_uuid_new_current() - Creates a new current UUID | |
b30ab791 | 3408 | * @device: DRBD device. |
b411b363 PR |
3409 | * |
3410 | * Creates a new current UUID, and rotates the old current UUID into | |
3411 | * the bitmap slot. Causes an incremental resync upon next connect. | |
3412 | */ | |
b30ab791 | 3413 | void drbd_uuid_new_current(struct drbd_device *device) __must_hold(local) |
b411b363 PR |
3414 | { |
3415 | u64 val; | |
9f2247bb PR |
3416 | unsigned long long bm_uuid; |
3417 | ||
3418 | get_random_bytes(&val, sizeof(u64)); | |
3419 | ||
b30ab791 AG |
3420 | spin_lock_irq(&device->ldev->md.uuid_lock); |
3421 | bm_uuid = device->ldev->md.uuid[UI_BITMAP]; | |
62b0da3a LE |
3422 | |
3423 | if (bm_uuid) | |
d0180171 | 3424 | drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid); |
b411b363 | 3425 | |
b30ab791 AG |
3426 | device->ldev->md.uuid[UI_BITMAP] = device->ldev->md.uuid[UI_CURRENT]; |
3427 | __drbd_uuid_set(device, UI_CURRENT, val); | |
3428 | spin_unlock_irq(&device->ldev->md.uuid_lock); | |
b411b363 | 3429 | |
b30ab791 | 3430 | drbd_print_uuids(device, "new current UUID"); |
aaa8e2b3 | 3431 | /* get it to stable storage _now_ */ |
b30ab791 | 3432 | drbd_md_sync(device); |
b411b363 PR |
3433 | } |
3434 | ||
b30ab791 | 3435 | void drbd_uuid_set_bm(struct drbd_device *device, u64 val) __must_hold(local) |
b411b363 | 3436 | { |
9f2247bb | 3437 | unsigned long flags; |
b30ab791 | 3438 | if (device->ldev->md.uuid[UI_BITMAP] == 0 && val == 0) |
b411b363 PR |
3439 | return; |
3440 | ||
b30ab791 | 3441 | spin_lock_irqsave(&device->ldev->md.uuid_lock, flags); |
b411b363 | 3442 | if (val == 0) { |
b30ab791 AG |
3443 | drbd_uuid_move_history(device); |
3444 | device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[UI_BITMAP]; | |
3445 | device->ldev->md.uuid[UI_BITMAP] = 0; | |
b411b363 | 3446 | } else { |
b30ab791 | 3447 | unsigned long long bm_uuid = device->ldev->md.uuid[UI_BITMAP]; |
62b0da3a | 3448 | if (bm_uuid) |
d0180171 | 3449 | drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid); |
b411b363 | 3450 | |
b30ab791 | 3451 | device->ldev->md.uuid[UI_BITMAP] = val & ~((u64)1); |
b411b363 | 3452 | } |
b30ab791 | 3453 | spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags); |
9f2247bb | 3454 | |
b30ab791 | 3455 | drbd_md_mark_dirty(device); |
b411b363 PR |
3456 | } |
3457 | ||
3458 | /** | |
3459 | * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io() | |
b30ab791 | 3460 | * @device: DRBD device. |
b411b363 PR |
3461 | * |
3462 | * Sets all bits in the bitmap and writes the whole bitmap to stable storage. | |
3463 | */ | |
8fe39aac | 3464 | int drbd_bmio_set_n_write(struct drbd_device *device) __must_hold(local) |
b411b363 PR |
3465 | { |
3466 | int rv = -EIO; | |
3467 | ||
8fe39aac PR |
3468 | drbd_md_set_flag(device, MDF_FULL_SYNC); |
3469 | drbd_md_sync(device); | |
3470 | drbd_bm_set_all(device); | |
b411b363 | 3471 | |
8fe39aac | 3472 | rv = drbd_bm_write(device); |
b411b363 | 3473 | |
8fe39aac PR |
3474 | if (!rv) { |
3475 | drbd_md_clear_flag(device, MDF_FULL_SYNC); | |
3476 | drbd_md_sync(device); | |
b411b363 PR |
3477 | } |
3478 | ||
3479 | return rv; | |
3480 | } | |
3481 | ||
3482 | /** | |
3483 | * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io() | |
b30ab791 | 3484 | * @device: DRBD device. |
b411b363 PR |
3485 | * |
3486 | * Clears all bits in the bitmap and writes the whole bitmap to stable storage. | |
3487 | */ | |
8fe39aac | 3488 | int drbd_bmio_clear_n_write(struct drbd_device *device) __must_hold(local) |
b411b363 | 3489 | { |
b30ab791 | 3490 | drbd_resume_al(device); |
8fe39aac PR |
3491 | drbd_bm_clear_all(device); |
3492 | return drbd_bm_write(device); | |
b411b363 PR |
3493 | } |
3494 | ||
99920dc5 | 3495 | static int w_bitmap_io(struct drbd_work *w, int unused) |
b411b363 | 3496 | { |
84b8c06b AG |
3497 | struct drbd_device *device = |
3498 | container_of(w, struct drbd_device, bm_io_work.w); | |
3499 | struct bm_io_work *work = &device->bm_io_work; | |
02851e9f | 3500 | int rv = -EIO; |
b411b363 | 3501 | |
bca1cbae LE |
3502 | if (work->flags != BM_LOCKED_CHANGE_ALLOWED) { |
3503 | int cnt = atomic_read(&device->ap_bio_cnt); | |
3504 | if (cnt) | |
3505 | drbd_err(device, "FIXME: ap_bio_cnt %d, expected 0; queued for '%s'\n", | |
3506 | cnt, work->why); | |
3507 | } | |
b411b363 | 3508 | |
b30ab791 AG |
3509 | if (get_ldev(device)) { |
3510 | drbd_bm_lock(device, work->why, work->flags); | |
3511 | rv = work->io_fn(device); | |
3512 | drbd_bm_unlock(device); | |
3513 | put_ldev(device); | |
02851e9f | 3514 | } |
b411b363 | 3515 | |
b30ab791 AG |
3516 | clear_bit_unlock(BITMAP_IO, &device->flags); |
3517 | wake_up(&device->misc_wait); | |
b411b363 PR |
3518 | |
3519 | if (work->done) | |
b30ab791 | 3520 | work->done(device, rv); |
b411b363 | 3521 | |
b30ab791 | 3522 | clear_bit(BITMAP_IO_QUEUED, &device->flags); |
b411b363 | 3523 | work->why = NULL; |
20ceb2b2 | 3524 | work->flags = 0; |
b411b363 | 3525 | |
99920dc5 | 3526 | return 0; |
b411b363 PR |
3527 | } |
3528 | ||
3529 | /** | |
3530 | * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap | |
b30ab791 | 3531 | * @device: DRBD device. |
b411b363 PR |
3532 | * @io_fn: IO callback to be called when bitmap IO is possible |
3533 | * @done: callback to be called after the bitmap IO was performed | |
3534 | * @why: Descriptive text of the reason for doing the IO | |
3535 | * | |
3536 | * While IO on the bitmap happens we freeze application IO thus we ensure | |
3537 | * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be | |
3538 | * called from worker context. It MUST NOT be used while a previous such | |
3539 | * work is still pending! | |
8fe39aac PR |
3540 | * |
3541 | * Its worker function encloses the call of io_fn() by get_ldev() and | |
3542 | * put_ldev(). | |
b411b363 | 3543 | */ |
b30ab791 | 3544 | void drbd_queue_bitmap_io(struct drbd_device *device, |
54761697 AG |
3545 | int (*io_fn)(struct drbd_device *), |
3546 | void (*done)(struct drbd_device *, int), | |
20ceb2b2 | 3547 | char *why, enum bm_flag flags) |
b411b363 | 3548 | { |
0b0ba1ef | 3549 | D_ASSERT(device, current == first_peer_device(device)->connection->worker.task); |
b411b363 | 3550 | |
0b0ba1ef AG |
3551 | D_ASSERT(device, !test_bit(BITMAP_IO_QUEUED, &device->flags)); |
3552 | D_ASSERT(device, !test_bit(BITMAP_IO, &device->flags)); | |
3553 | D_ASSERT(device, list_empty(&device->bm_io_work.w.list)); | |
b30ab791 | 3554 | if (device->bm_io_work.why) |
d0180171 | 3555 | drbd_err(device, "FIXME going to queue '%s' but '%s' still pending?\n", |
b30ab791 | 3556 | why, device->bm_io_work.why); |
b411b363 | 3557 | |
b30ab791 AG |
3558 | device->bm_io_work.io_fn = io_fn; |
3559 | device->bm_io_work.done = done; | |
3560 | device->bm_io_work.why = why; | |
3561 | device->bm_io_work.flags = flags; | |
b411b363 | 3562 | |
0500813f | 3563 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 3564 | set_bit(BITMAP_IO, &device->flags); |
5bded4ef LE |
3565 | /* don't wait for pending application IO if the caller indicates that |
3566 | * application IO does not conflict anyways. */ | |
3567 | if (flags == BM_LOCKED_CHANGE_ALLOWED || atomic_read(&device->ap_bio_cnt) == 0) { | |
b30ab791 | 3568 | if (!test_and_set_bit(BITMAP_IO_QUEUED, &device->flags)) |
84b8c06b AG |
3569 | drbd_queue_work(&first_peer_device(device)->connection->sender_work, |
3570 | &device->bm_io_work.w); | |
b411b363 | 3571 | } |
0500813f | 3572 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 PR |
3573 | } |
3574 | ||
3575 | /** | |
3576 | * drbd_bitmap_io() - Does an IO operation on the whole bitmap | |
b30ab791 | 3577 | * @device: DRBD device. |
b411b363 PR |
3578 | * @io_fn: IO callback to be called when bitmap IO is possible |
3579 | * @why: Descriptive text of the reason for doing the IO | |
3580 | * | |
3581 | * freezes application IO while that the actual IO operations runs. This | |
3582 | * functions MAY NOT be called from worker context. | |
3583 | */ | |
b30ab791 | 3584 | int drbd_bitmap_io(struct drbd_device *device, int (*io_fn)(struct drbd_device *), |
20ceb2b2 | 3585 | char *why, enum bm_flag flags) |
b411b363 | 3586 | { |
c0065f98 LE |
3587 | /* Only suspend io, if some operation is supposed to be locked out */ |
3588 | const bool do_suspend_io = flags & (BM_DONT_CLEAR|BM_DONT_SET|BM_DONT_TEST); | |
b411b363 PR |
3589 | int rv; |
3590 | ||
0b0ba1ef | 3591 | D_ASSERT(device, current != first_peer_device(device)->connection->worker.task); |
b411b363 | 3592 | |
c0065f98 | 3593 | if (do_suspend_io) |
b30ab791 | 3594 | drbd_suspend_io(device); |
b411b363 | 3595 | |
b30ab791 AG |
3596 | drbd_bm_lock(device, why, flags); |
3597 | rv = io_fn(device); | |
3598 | drbd_bm_unlock(device); | |
b411b363 | 3599 | |
c0065f98 | 3600 | if (do_suspend_io) |
b30ab791 | 3601 | drbd_resume_io(device); |
b411b363 PR |
3602 | |
3603 | return rv; | |
3604 | } | |
3605 | ||
b30ab791 | 3606 | void drbd_md_set_flag(struct drbd_device *device, int flag) __must_hold(local) |
b411b363 | 3607 | { |
b30ab791 AG |
3608 | if ((device->ldev->md.flags & flag) != flag) { |
3609 | drbd_md_mark_dirty(device); | |
3610 | device->ldev->md.flags |= flag; | |
b411b363 PR |
3611 | } |
3612 | } | |
3613 | ||
b30ab791 | 3614 | void drbd_md_clear_flag(struct drbd_device *device, int flag) __must_hold(local) |
b411b363 | 3615 | { |
b30ab791 AG |
3616 | if ((device->ldev->md.flags & flag) != 0) { |
3617 | drbd_md_mark_dirty(device); | |
3618 | device->ldev->md.flags &= ~flag; | |
b411b363 PR |
3619 | } |
3620 | } | |
3621 | int drbd_md_test_flag(struct drbd_backing_dev *bdev, int flag) | |
3622 | { | |
3623 | return (bdev->md.flags & flag) != 0; | |
3624 | } | |
3625 | ||
2bccef39 | 3626 | static void md_sync_timer_fn(struct timer_list *t) |
b411b363 | 3627 | { |
2bccef39 | 3628 | struct drbd_device *device = from_timer(device, t, md_sync_timer); |
ac0acb9e | 3629 | drbd_device_post_work(device, MD_SYNC); |
b411b363 PR |
3630 | } |
3631 | ||
d8763023 | 3632 | const char *cmdname(enum drbd_packet cmd) |
f2ad9063 AG |
3633 | { |
3634 | /* THINK may need to become several global tables | |
3635 | * when we want to support more than | |
3636 | * one PRO_VERSION */ | |
3637 | static const char *cmdnames[] = { | |
3638 | [P_DATA] = "Data", | |
9104d31a LE |
3639 | [P_WSAME] = "WriteSame", |
3640 | [P_TRIM] = "Trim", | |
f2ad9063 AG |
3641 | [P_DATA_REPLY] = "DataReply", |
3642 | [P_RS_DATA_REPLY] = "RSDataReply", | |
3643 | [P_BARRIER] = "Barrier", | |
3644 | [P_BITMAP] = "ReportBitMap", | |
3645 | [P_BECOME_SYNC_TARGET] = "BecomeSyncTarget", | |
3646 | [P_BECOME_SYNC_SOURCE] = "BecomeSyncSource", | |
3647 | [P_UNPLUG_REMOTE] = "UnplugRemote", | |
3648 | [P_DATA_REQUEST] = "DataRequest", | |
3649 | [P_RS_DATA_REQUEST] = "RSDataRequest", | |
3650 | [P_SYNC_PARAM] = "SyncParam", | |
3651 | [P_SYNC_PARAM89] = "SyncParam89", | |
3652 | [P_PROTOCOL] = "ReportProtocol", | |
3653 | [P_UUIDS] = "ReportUUIDs", | |
3654 | [P_SIZES] = "ReportSizes", | |
3655 | [P_STATE] = "ReportState", | |
3656 | [P_SYNC_UUID] = "ReportSyncUUID", | |
3657 | [P_AUTH_CHALLENGE] = "AuthChallenge", | |
3658 | [P_AUTH_RESPONSE] = "AuthResponse", | |
3659 | [P_PING] = "Ping", | |
3660 | [P_PING_ACK] = "PingAck", | |
3661 | [P_RECV_ACK] = "RecvAck", | |
3662 | [P_WRITE_ACK] = "WriteAck", | |
3663 | [P_RS_WRITE_ACK] = "RSWriteAck", | |
d4dabbe2 | 3664 | [P_SUPERSEDED] = "Superseded", |
f2ad9063 AG |
3665 | [P_NEG_ACK] = "NegAck", |
3666 | [P_NEG_DREPLY] = "NegDReply", | |
3667 | [P_NEG_RS_DREPLY] = "NegRSDReply", | |
3668 | [P_BARRIER_ACK] = "BarrierAck", | |
3669 | [P_STATE_CHG_REQ] = "StateChgRequest", | |
3670 | [P_STATE_CHG_REPLY] = "StateChgReply", | |
3671 | [P_OV_REQUEST] = "OVRequest", | |
3672 | [P_OV_REPLY] = "OVReply", | |
3673 | [P_OV_RESULT] = "OVResult", | |
3674 | [P_CSUM_RS_REQUEST] = "CsumRSRequest", | |
3675 | [P_RS_IS_IN_SYNC] = "CsumRSIsInSync", | |
3676 | [P_COMPRESSED_BITMAP] = "CBitmap", | |
3677 | [P_DELAY_PROBE] = "DelayProbe", | |
3678 | [P_OUT_OF_SYNC] = "OutOfSync", | |
7be8da07 | 3679 | [P_RETRY_WRITE] = "RetryWrite", |
ae25b336 LE |
3680 | [P_RS_CANCEL] = "RSCancel", |
3681 | [P_CONN_ST_CHG_REQ] = "conn_st_chg_req", | |
3682 | [P_CONN_ST_CHG_REPLY] = "conn_st_chg_reply", | |
036b17ea PR |
3683 | [P_RETRY_WRITE] = "retry_write", |
3684 | [P_PROTOCOL_UPDATE] = "protocol_update", | |
700ca8c0 PR |
3685 | [P_RS_THIN_REQ] = "rs_thin_req", |
3686 | [P_RS_DEALLOCATED] = "rs_deallocated", | |
ae25b336 LE |
3687 | |
3688 | /* enum drbd_packet, but not commands - obsoleted flags: | |
3689 | * P_MAY_IGNORE | |
3690 | * P_MAX_OPT_CMD | |
3691 | */ | |
f2ad9063 AG |
3692 | }; |
3693 | ||
ae25b336 | 3694 | /* too big for the array: 0xfffX */ |
e5d6f33a AG |
3695 | if (cmd == P_INITIAL_META) |
3696 | return "InitialMeta"; | |
3697 | if (cmd == P_INITIAL_DATA) | |
3698 | return "InitialData"; | |
6038178e AG |
3699 | if (cmd == P_CONNECTION_FEATURES) |
3700 | return "ConnectionFeatures"; | |
6e849ce8 | 3701 | if (cmd >= ARRAY_SIZE(cmdnames)) |
f2ad9063 AG |
3702 | return "Unknown"; |
3703 | return cmdnames[cmd]; | |
3704 | } | |
3705 | ||
7be8da07 AG |
3706 | /** |
3707 | * drbd_wait_misc - wait for a request to make progress | |
b30ab791 | 3708 | * @device: device associated with the request |
7be8da07 AG |
3709 | * @i: the struct drbd_interval embedded in struct drbd_request or |
3710 | * struct drbd_peer_request | |
3711 | */ | |
b30ab791 | 3712 | int drbd_wait_misc(struct drbd_device *device, struct drbd_interval *i) |
7be8da07 | 3713 | { |
44ed167d | 3714 | struct net_conf *nc; |
7be8da07 AG |
3715 | DEFINE_WAIT(wait); |
3716 | long timeout; | |
3717 | ||
44ed167d | 3718 | rcu_read_lock(); |
a6b32bc3 | 3719 | nc = rcu_dereference(first_peer_device(device)->connection->net_conf); |
44ed167d PR |
3720 | if (!nc) { |
3721 | rcu_read_unlock(); | |
7be8da07 | 3722 | return -ETIMEDOUT; |
44ed167d PR |
3723 | } |
3724 | timeout = nc->ko_count ? nc->timeout * HZ / 10 * nc->ko_count : MAX_SCHEDULE_TIMEOUT; | |
3725 | rcu_read_unlock(); | |
7be8da07 | 3726 | |
b30ab791 | 3727 | /* Indicate to wake up device->misc_wait on progress. */ |
7be8da07 | 3728 | i->waiting = true; |
b30ab791 | 3729 | prepare_to_wait(&device->misc_wait, &wait, TASK_INTERRUPTIBLE); |
0500813f | 3730 | spin_unlock_irq(&device->resource->req_lock); |
7be8da07 | 3731 | timeout = schedule_timeout(timeout); |
b30ab791 | 3732 | finish_wait(&device->misc_wait, &wait); |
0500813f | 3733 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 3734 | if (!timeout || device->state.conn < C_CONNECTED) |
7be8da07 AG |
3735 | return -ETIMEDOUT; |
3736 | if (signal_pending(current)) | |
3737 | return -ERESTARTSYS; | |
3738 | return 0; | |
b411b363 PR |
3739 | } |
3740 | ||
28bc3b8c AG |
3741 | void lock_all_resources(void) |
3742 | { | |
3743 | struct drbd_resource *resource; | |
3744 | int __maybe_unused i = 0; | |
3745 | ||
3746 | mutex_lock(&resources_mutex); | |
3747 | local_irq_disable(); | |
3748 | for_each_resource(resource, &drbd_resources) | |
3749 | spin_lock_nested(&resource->req_lock, i++); | |
3750 | } | |
3751 | ||
3752 | void unlock_all_resources(void) | |
3753 | { | |
3754 | struct drbd_resource *resource; | |
3755 | ||
3756 | for_each_resource(resource, &drbd_resources) | |
3757 | spin_unlock(&resource->req_lock); | |
3758 | local_irq_enable(); | |
3759 | mutex_unlock(&resources_mutex); | |
3760 | } | |
3761 | ||
b411b363 PR |
3762 | #ifdef CONFIG_DRBD_FAULT_INJECTION |
3763 | /* Fault insertion support including random number generator shamelessly | |
3764 | * stolen from kernel/rcutorture.c */ | |
3765 | struct fault_random_state { | |
3766 | unsigned long state; | |
3767 | unsigned long count; | |
3768 | }; | |
3769 | ||
3770 | #define FAULT_RANDOM_MULT 39916801 /* prime */ | |
3771 | #define FAULT_RANDOM_ADD 479001701 /* prime */ | |
3772 | #define FAULT_RANDOM_REFRESH 10000 | |
3773 | ||
3774 | /* | |
3775 | * Crude but fast random-number generator. Uses a linear congruential | |
3776 | * generator, with occasional help from get_random_bytes(). | |
3777 | */ | |
3778 | static unsigned long | |
3779 | _drbd_fault_random(struct fault_random_state *rsp) | |
3780 | { | |
3781 | long refresh; | |
3782 | ||
49829ea7 | 3783 | if (!rsp->count--) { |
b411b363 PR |
3784 | get_random_bytes(&refresh, sizeof(refresh)); |
3785 | rsp->state += refresh; | |
3786 | rsp->count = FAULT_RANDOM_REFRESH; | |
3787 | } | |
3788 | rsp->state = rsp->state * FAULT_RANDOM_MULT + FAULT_RANDOM_ADD; | |
3789 | return swahw32(rsp->state); | |
3790 | } | |
3791 | ||
3792 | static char * | |
3793 | _drbd_fault_str(unsigned int type) { | |
3794 | static char *_faults[] = { | |
3795 | [DRBD_FAULT_MD_WR] = "Meta-data write", | |
3796 | [DRBD_FAULT_MD_RD] = "Meta-data read", | |
3797 | [DRBD_FAULT_RS_WR] = "Resync write", | |
3798 | [DRBD_FAULT_RS_RD] = "Resync read", | |
3799 | [DRBD_FAULT_DT_WR] = "Data write", | |
3800 | [DRBD_FAULT_DT_RD] = "Data read", | |
3801 | [DRBD_FAULT_DT_RA] = "Data read ahead", | |
3802 | [DRBD_FAULT_BM_ALLOC] = "BM allocation", | |
6b4388ac PR |
3803 | [DRBD_FAULT_AL_EE] = "EE allocation", |
3804 | [DRBD_FAULT_RECEIVE] = "receive data corruption", | |
b411b363 PR |
3805 | }; |
3806 | ||
3807 | return (type < DRBD_FAULT_MAX) ? _faults[type] : "**Unknown**"; | |
3808 | } | |
3809 | ||
3810 | unsigned int | |
b30ab791 | 3811 | _drbd_insert_fault(struct drbd_device *device, unsigned int type) |
b411b363 PR |
3812 | { |
3813 | static struct fault_random_state rrs = {0, 0}; | |
3814 | ||
3815 | unsigned int ret = ( | |
183ece30 RK |
3816 | (drbd_fault_devs == 0 || |
3817 | ((1 << device_to_minor(device)) & drbd_fault_devs) != 0) && | |
3818 | (((_drbd_fault_random(&rrs) % 100) + 1) <= drbd_fault_rate)); | |
b411b363 PR |
3819 | |
3820 | if (ret) { | |
183ece30 | 3821 | drbd_fault_count++; |
b411b363 | 3822 | |
7383506c | 3823 | if (__ratelimit(&drbd_ratelimit_state)) |
d0180171 | 3824 | drbd_warn(device, "***Simulating %s failure\n", |
b411b363 PR |
3825 | _drbd_fault_str(type)); |
3826 | } | |
3827 | ||
3828 | return ret; | |
3829 | } | |
3830 | #endif | |
3831 | ||
3832 | const char *drbd_buildtag(void) | |
3833 | { | |
3834 | /* DRBD built from external sources has here a reference to the | |
3835 | git hash of the source code. */ | |
3836 | ||
3837 | static char buildtag[38] = "\0uilt-in"; | |
3838 | ||
3839 | if (buildtag[0] == 0) { | |
bc4854bc CW |
3840 | #ifdef MODULE |
3841 | sprintf(buildtag, "srcversion: %-24s", THIS_MODULE->srcversion); | |
3842 | #else | |
3843 | buildtag[0] = 'b'; | |
b411b363 | 3844 | #endif |
b411b363 PR |
3845 | } |
3846 | ||
3847 | return buildtag; | |
3848 | } | |
3849 | ||
3850 | module_init(drbd_init) | |
3851 | module_exit(drbd_cleanup) | |
3852 | ||
b411b363 PR |
3853 | EXPORT_SYMBOL(drbd_conn_str); |
3854 | EXPORT_SYMBOL(drbd_role_str); | |
3855 | EXPORT_SYMBOL(drbd_disk_str); | |
3856 | EXPORT_SYMBOL(drbd_set_st_err_str); |