]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
3d14c5d2 | 2 | #include <linux/ceph/ceph_debug.h> |
31b8006e SW |
3 | |
4 | #include <linux/crc32c.h> | |
5 | #include <linux/ctype.h> | |
6 | #include <linux/highmem.h> | |
7 | #include <linux/inet.h> | |
8 | #include <linux/kthread.h> | |
9 | #include <linux/net.h> | |
757856d2 | 10 | #include <linux/nsproxy.h> |
633ee407 | 11 | #include <linux/sched/mm.h> |
5a0e3ad6 | 12 | #include <linux/slab.h> |
31b8006e SW |
13 | #include <linux/socket.h> |
14 | #include <linux/string.h> | |
3ebc21f7 | 15 | #ifdef CONFIG_BLOCK |
68b4476b | 16 | #include <linux/bio.h> |
3ebc21f7 | 17 | #endif /* CONFIG_BLOCK */ |
ee3b56f2 | 18 | #include <linux/dns_resolver.h> |
31b8006e SW |
19 | #include <net/tcp.h> |
20 | ||
2b3e0c90 | 21 | #include <linux/ceph/ceph_features.h> |
3d14c5d2 YS |
22 | #include <linux/ceph/libceph.h> |
23 | #include <linux/ceph/messenger.h> | |
24 | #include <linux/ceph/decode.h> | |
25 | #include <linux/ceph/pagelist.h> | |
bc3b2d7f | 26 | #include <linux/export.h> |
31b8006e SW |
27 | |
28 | /* | |
29 | * Ceph uses the messenger to exchange ceph_msg messages with other | |
30 | * hosts in the system. The messenger provides ordered and reliable | |
31 | * delivery. We tolerate TCP disconnects by reconnecting (with | |
32 | * exponential backoff) in the case of a fault (disconnection, bad | |
33 | * crc, protocol error). Acks allow sent messages to be discarded by | |
34 | * the sender. | |
35 | */ | |
36 | ||
bc18f4b1 AE |
37 | /* |
38 | * We track the state of the socket on a given connection using | |
39 | * values defined below. The transition to a new socket state is | |
40 | * handled by a function which verifies we aren't coming from an | |
41 | * unexpected state. | |
42 | * | |
43 | * -------- | |
44 | * | NEW* | transient initial state | |
45 | * -------- | |
46 | * | con_sock_state_init() | |
47 | * v | |
48 | * ---------- | |
49 | * | CLOSED | initialized, but no socket (and no | |
50 | * ---------- TCP connection) | |
51 | * ^ \ | |
52 | * | \ con_sock_state_connecting() | |
53 | * | ---------------------- | |
54 | * | \ | |
55 | * + con_sock_state_closed() \ | |
fbb85a47 SW |
56 | * |+--------------------------- \ |
57 | * | \ \ \ | |
58 | * | ----------- \ \ | |
59 | * | | CLOSING | socket event; \ \ | |
60 | * | ----------- await close \ \ | |
61 | * | ^ \ | | |
62 | * | | \ | | |
63 | * | + con_sock_state_closing() \ | | |
64 | * | / \ | | | |
65 | * | / --------------- | | | |
66 | * | / \ v v | |
bc18f4b1 AE |
67 | * | / -------------- |
68 | * | / -----------------| CONNECTING | socket created, TCP | |
69 | * | | / -------------- connect initiated | |
70 | * | | | con_sock_state_connected() | |
71 | * | | v | |
72 | * ------------- | |
73 | * | CONNECTED | TCP connection established | |
74 | * ------------- | |
75 | * | |
76 | * State values for ceph_connection->sock_state; NEW is assumed to be 0. | |
77 | */ | |
ce2c8903 AE |
78 | |
79 | #define CON_SOCK_STATE_NEW 0 /* -> CLOSED */ | |
80 | #define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */ | |
81 | #define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */ | |
82 | #define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */ | |
83 | #define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */ | |
84 | ||
c9ffc77a AE |
85 | static bool con_flag_valid(unsigned long con_flag) |
86 | { | |
87 | switch (con_flag) { | |
3fefd43e ID |
88 | case CEPH_CON_F_LOSSYTX: |
89 | case CEPH_CON_F_KEEPALIVE_PENDING: | |
90 | case CEPH_CON_F_WRITE_PENDING: | |
91 | case CEPH_CON_F_SOCK_CLOSED: | |
92 | case CEPH_CON_F_BACKOFF: | |
c9ffc77a AE |
93 | return true; |
94 | default: | |
95 | return false; | |
96 | } | |
97 | } | |
98 | ||
6503e0b6 | 99 | void ceph_con_flag_clear(struct ceph_connection *con, unsigned long con_flag) |
c9ffc77a AE |
100 | { |
101 | BUG_ON(!con_flag_valid(con_flag)); | |
102 | ||
103 | clear_bit(con_flag, &con->flags); | |
104 | } | |
105 | ||
6503e0b6 | 106 | void ceph_con_flag_set(struct ceph_connection *con, unsigned long con_flag) |
c9ffc77a AE |
107 | { |
108 | BUG_ON(!con_flag_valid(con_flag)); | |
109 | ||
110 | set_bit(con_flag, &con->flags); | |
111 | } | |
112 | ||
6503e0b6 | 113 | bool ceph_con_flag_test(struct ceph_connection *con, unsigned long con_flag) |
c9ffc77a AE |
114 | { |
115 | BUG_ON(!con_flag_valid(con_flag)); | |
116 | ||
117 | return test_bit(con_flag, &con->flags); | |
118 | } | |
119 | ||
6503e0b6 ID |
120 | bool ceph_con_flag_test_and_clear(struct ceph_connection *con, |
121 | unsigned long con_flag) | |
c9ffc77a AE |
122 | { |
123 | BUG_ON(!con_flag_valid(con_flag)); | |
124 | ||
125 | return test_and_clear_bit(con_flag, &con->flags); | |
126 | } | |
127 | ||
6503e0b6 ID |
128 | bool ceph_con_flag_test_and_set(struct ceph_connection *con, |
129 | unsigned long con_flag) | |
c9ffc77a AE |
130 | { |
131 | BUG_ON(!con_flag_valid(con_flag)); | |
132 | ||
133 | return test_and_set_bit(con_flag, &con->flags); | |
134 | } | |
135 | ||
e3d5d638 AE |
136 | /* Slab caches for frequently-allocated structures */ |
137 | ||
138 | static struct kmem_cache *ceph_msg_cache; | |
139 | ||
a6a5349d SW |
140 | #ifdef CONFIG_LOCKDEP |
141 | static struct lock_class_key socket_class; | |
142 | #endif | |
143 | ||
31b8006e | 144 | static void queue_con(struct ceph_connection *con); |
37ab77ac | 145 | static void cancel_con(struct ceph_connection *con); |
68931622 | 146 | static void ceph_con_workfn(struct work_struct *); |
93209264 | 147 | static void con_fault(struct ceph_connection *con); |
31b8006e | 148 | |
31b8006e | 149 | /* |
f64a9317 AE |
150 | * Nicely render a sockaddr as a string. An array of formatted |
151 | * strings is used, to approximate reentrancy. | |
31b8006e | 152 | */ |
f64a9317 AE |
153 | #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */ |
154 | #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG) | |
155 | #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1) | |
156 | #define MAX_ADDR_STR_LEN 64 /* 54 is enough */ | |
157 | ||
158 | static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN]; | |
159 | static atomic_t addr_str_seq = ATOMIC_INIT(0); | |
31b8006e | 160 | |
699921d9 | 161 | struct page *ceph_zero_page; /* used in certain error cases */ |
57666519 | 162 | |
b726ec97 | 163 | const char *ceph_pr_addr(const struct ceph_entity_addr *addr) |
31b8006e SW |
164 | { |
165 | int i; | |
166 | char *s; | |
b726ec97 JL |
167 | struct sockaddr_storage ss = addr->in_addr; /* align */ |
168 | struct sockaddr_in *in4 = (struct sockaddr_in *)&ss; | |
169 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *)&ss; | |
31b8006e | 170 | |
f64a9317 | 171 | i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK; |
31b8006e SW |
172 | s = addr_str[i]; |
173 | ||
b726ec97 | 174 | switch (ss.ss_family) { |
31b8006e | 175 | case AF_INET: |
d3c3c0a8 JL |
176 | snprintf(s, MAX_ADDR_STR_LEN, "(%d)%pI4:%hu", |
177 | le32_to_cpu(addr->type), &in4->sin_addr, | |
bd406145 | 178 | ntohs(in4->sin_port)); |
31b8006e SW |
179 | break; |
180 | ||
181 | case AF_INET6: | |
d3c3c0a8 JL |
182 | snprintf(s, MAX_ADDR_STR_LEN, "(%d)[%pI6c]:%hu", |
183 | le32_to_cpu(addr->type), &in6->sin6_addr, | |
bd406145 | 184 | ntohs(in6->sin6_port)); |
31b8006e SW |
185 | break; |
186 | ||
187 | default: | |
d3002b97 | 188 | snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)", |
b726ec97 | 189 | ss.ss_family); |
31b8006e SW |
190 | } |
191 | ||
192 | return s; | |
193 | } | |
3d14c5d2 | 194 | EXPORT_SYMBOL(ceph_pr_addr); |
31b8006e | 195 | |
6503e0b6 | 196 | void ceph_encode_my_addr(struct ceph_messenger *msgr) |
63f2d211 | 197 | { |
cd1a677c ID |
198 | if (!ceph_msgr2(from_msgr(msgr))) { |
199 | memcpy(&msgr->my_enc_addr, &msgr->inst.addr, | |
200 | sizeof(msgr->my_enc_addr)); | |
201 | ceph_encode_banner_addr(&msgr->my_enc_addr); | |
202 | } | |
63f2d211 SW |
203 | } |
204 | ||
31b8006e SW |
205 | /* |
206 | * work queue for all reading and writing to/from the socket. | |
207 | */ | |
e0f43c94 | 208 | static struct workqueue_struct *ceph_msgr_wq; |
31b8006e | 209 | |
e3d5d638 AE |
210 | static int ceph_msgr_slab_init(void) |
211 | { | |
212 | BUG_ON(ceph_msg_cache); | |
5ee61e95 | 213 | ceph_msg_cache = KMEM_CACHE(ceph_msg, 0); |
81b36be4 AE |
214 | if (!ceph_msg_cache) |
215 | return -ENOMEM; | |
216 | ||
0d9c1ab3 | 217 | return 0; |
e3d5d638 AE |
218 | } |
219 | ||
220 | static void ceph_msgr_slab_exit(void) | |
221 | { | |
222 | BUG_ON(!ceph_msg_cache); | |
223 | kmem_cache_destroy(ceph_msg_cache); | |
224 | ceph_msg_cache = NULL; | |
225 | } | |
226 | ||
15417167 | 227 | static void _ceph_msgr_exit(void) |
6173d1f0 | 228 | { |
d3002b97 | 229 | if (ceph_msgr_wq) { |
6173d1f0 | 230 | destroy_workqueue(ceph_msgr_wq); |
d3002b97 AE |
231 | ceph_msgr_wq = NULL; |
232 | } | |
6173d1f0 | 233 | |
699921d9 ID |
234 | BUG_ON(!ceph_zero_page); |
235 | put_page(ceph_zero_page); | |
236 | ceph_zero_page = NULL; | |
d920ff6f BC |
237 | |
238 | ceph_msgr_slab_exit(); | |
6173d1f0 AE |
239 | } |
240 | ||
57a35dfb | 241 | int __init ceph_msgr_init(void) |
31b8006e | 242 | { |
d920ff6f BC |
243 | if (ceph_msgr_slab_init()) |
244 | return -ENOMEM; | |
245 | ||
699921d9 ID |
246 | BUG_ON(ceph_zero_page); |
247 | ceph_zero_page = ZERO_PAGE(0); | |
248 | get_page(ceph_zero_page); | |
57666519 | 249 | |
f9865f06 ID |
250 | /* |
251 | * The number of active work items is limited by the number of | |
252 | * connections, so leave @max_active at default. | |
253 | */ | |
254 | ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0); | |
6173d1f0 AE |
255 | if (ceph_msgr_wq) |
256 | return 0; | |
57666519 | 257 | |
6173d1f0 AE |
258 | pr_err("msgr_init failed to create workqueue\n"); |
259 | _ceph_msgr_exit(); | |
57666519 | 260 | |
6173d1f0 | 261 | return -ENOMEM; |
31b8006e SW |
262 | } |
263 | ||
264 | void ceph_msgr_exit(void) | |
265 | { | |
57666519 | 266 | BUG_ON(ceph_msgr_wq == NULL); |
57666519 | 267 | |
6173d1f0 | 268 | _ceph_msgr_exit(); |
31b8006e SW |
269 | } |
270 | ||
cd84db6e | 271 | void ceph_msgr_flush(void) |
a922d38f SW |
272 | { |
273 | flush_workqueue(ceph_msgr_wq); | |
274 | } | |
3d14c5d2 | 275 | EXPORT_SYMBOL(ceph_msgr_flush); |
a922d38f | 276 | |
ce2c8903 AE |
277 | /* Connection socket state transition functions */ |
278 | ||
279 | static void con_sock_state_init(struct ceph_connection *con) | |
280 | { | |
281 | int old_state; | |
282 | ||
283 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED); | |
284 | if (WARN_ON(old_state != CON_SOCK_STATE_NEW)) | |
285 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
286 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
287 | CON_SOCK_STATE_CLOSED); | |
ce2c8903 AE |
288 | } |
289 | ||
290 | static void con_sock_state_connecting(struct ceph_connection *con) | |
291 | { | |
292 | int old_state; | |
293 | ||
294 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING); | |
295 | if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED)) | |
296 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
297 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
298 | CON_SOCK_STATE_CONNECTING); | |
ce2c8903 AE |
299 | } |
300 | ||
301 | static void con_sock_state_connected(struct ceph_connection *con) | |
302 | { | |
303 | int old_state; | |
304 | ||
305 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED); | |
306 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING)) | |
307 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
308 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
309 | CON_SOCK_STATE_CONNECTED); | |
ce2c8903 AE |
310 | } |
311 | ||
312 | static void con_sock_state_closing(struct ceph_connection *con) | |
313 | { | |
314 | int old_state; | |
315 | ||
316 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING); | |
317 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING && | |
318 | old_state != CON_SOCK_STATE_CONNECTED && | |
319 | old_state != CON_SOCK_STATE_CLOSING)) | |
320 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
321 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
322 | CON_SOCK_STATE_CLOSING); | |
ce2c8903 AE |
323 | } |
324 | ||
325 | static void con_sock_state_closed(struct ceph_connection *con) | |
326 | { | |
327 | int old_state; | |
328 | ||
329 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED); | |
330 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED && | |
fbb85a47 | 331 | old_state != CON_SOCK_STATE_CLOSING && |
8007b8d6 SW |
332 | old_state != CON_SOCK_STATE_CONNECTING && |
333 | old_state != CON_SOCK_STATE_CLOSED)) | |
ce2c8903 | 334 | printk("%s: unexpected old state %d\n", __func__, old_state); |
8007b8d6 SW |
335 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
336 | CON_SOCK_STATE_CLOSED); | |
ce2c8903 | 337 | } |
a922d38f | 338 | |
31b8006e SW |
339 | /* |
340 | * socket callback functions | |
341 | */ | |
342 | ||
343 | /* data available on socket, or listen socket received a connect */ | |
676d2369 | 344 | static void ceph_sock_data_ready(struct sock *sk) |
31b8006e | 345 | { |
bd406145 | 346 | struct ceph_connection *con = sk->sk_user_data; |
a2a32584 GH |
347 | if (atomic_read(&con->msgr->stopping)) { |
348 | return; | |
349 | } | |
bd406145 | 350 | |
31b8006e | 351 | if (sk->sk_state != TCP_CLOSE_WAIT) { |
30be780a | 352 | dout("%s %p state = %d, queueing work\n", __func__, |
31b8006e SW |
353 | con, con->state); |
354 | queue_con(con); | |
355 | } | |
356 | } | |
357 | ||
358 | /* socket has buffer space for writing */ | |
327800bd | 359 | static void ceph_sock_write_space(struct sock *sk) |
31b8006e | 360 | { |
d3002b97 | 361 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e | 362 | |
182fac26 JS |
363 | /* only queue to workqueue if there is data we want to write, |
364 | * and there is sufficient space in the socket buffer to accept | |
327800bd | 365 | * more data. clear SOCK_NOSPACE so that ceph_sock_write_space() |
182fac26 JS |
366 | * doesn't get called again until try_write() fills the socket |
367 | * buffer. See net/ipv4/tcp_input.c:tcp_check_space() | |
368 | * and net/core/stream.c:sk_stream_write_space(). | |
369 | */ | |
6503e0b6 | 370 | if (ceph_con_flag_test(con, CEPH_CON_F_WRITE_PENDING)) { |
64dc6130 | 371 | if (sk_stream_is_writeable(sk)) { |
327800bd | 372 | dout("%s %p queueing write work\n", __func__, con); |
182fac26 JS |
373 | clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags); |
374 | queue_con(con); | |
375 | } | |
31b8006e | 376 | } else { |
327800bd | 377 | dout("%s %p nothing to write\n", __func__, con); |
31b8006e | 378 | } |
31b8006e SW |
379 | } |
380 | ||
381 | /* socket's state has changed */ | |
327800bd | 382 | static void ceph_sock_state_change(struct sock *sk) |
31b8006e | 383 | { |
bd406145 | 384 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e | 385 | |
30be780a | 386 | dout("%s %p state = %d sk_state = %u\n", __func__, |
31b8006e SW |
387 | con, con->state, sk->sk_state); |
388 | ||
31b8006e SW |
389 | switch (sk->sk_state) { |
390 | case TCP_CLOSE: | |
327800bd | 391 | dout("%s TCP_CLOSE\n", __func__); |
df561f66 | 392 | fallthrough; |
31b8006e | 393 | case TCP_CLOSE_WAIT: |
327800bd | 394 | dout("%s TCP_CLOSE_WAIT\n", __func__); |
ce2c8903 | 395 | con_sock_state_closing(con); |
6503e0b6 | 396 | ceph_con_flag_set(con, CEPH_CON_F_SOCK_CLOSED); |
d65c9e0b | 397 | queue_con(con); |
31b8006e SW |
398 | break; |
399 | case TCP_ESTABLISHED: | |
327800bd | 400 | dout("%s TCP_ESTABLISHED\n", __func__); |
ce2c8903 | 401 | con_sock_state_connected(con); |
31b8006e SW |
402 | queue_con(con); |
403 | break; | |
d3002b97 AE |
404 | default: /* Everything else is uninteresting */ |
405 | break; | |
31b8006e SW |
406 | } |
407 | } | |
408 | ||
409 | /* | |
410 | * set up socket callbacks | |
411 | */ | |
412 | static void set_sock_callbacks(struct socket *sock, | |
413 | struct ceph_connection *con) | |
414 | { | |
415 | struct sock *sk = sock->sk; | |
bd406145 | 416 | sk->sk_user_data = con; |
327800bd AE |
417 | sk->sk_data_ready = ceph_sock_data_ready; |
418 | sk->sk_write_space = ceph_sock_write_space; | |
419 | sk->sk_state_change = ceph_sock_state_change; | |
31b8006e SW |
420 | } |
421 | ||
422 | ||
423 | /* | |
424 | * socket helpers | |
425 | */ | |
426 | ||
427 | /* | |
428 | * initiate connection to a remote socket. | |
429 | */ | |
6503e0b6 | 430 | int ceph_tcp_connect(struct ceph_connection *con) |
31b8006e | 431 | { |
cede185b | 432 | struct sockaddr_storage ss = con->peer_addr.in_addr; /* align */ |
31b8006e | 433 | struct socket *sock; |
633ee407 | 434 | unsigned int noio_flag; |
31b8006e SW |
435 | int ret; |
436 | ||
6503e0b6 ID |
437 | dout("%s con %p peer_addr %s\n", __func__, con, |
438 | ceph_pr_addr(&con->peer_addr)); | |
31b8006e | 439 | BUG_ON(con->sock); |
633ee407 ID |
440 | |
441 | /* sock_create_kern() allocates with GFP_KERNEL */ | |
442 | noio_flag = memalloc_noio_save(); | |
cede185b | 443 | ret = sock_create_kern(read_pnet(&con->msgr->net), ss.ss_family, |
eeb1bd5c | 444 | SOCK_STREAM, IPPROTO_TCP, &sock); |
633ee407 | 445 | memalloc_noio_restore(noio_flag); |
31b8006e | 446 | if (ret) |
41617d0c | 447 | return ret; |
6d7fdb0a | 448 | sock->sk->sk_allocation = GFP_NOFS; |
31b8006e | 449 | |
a6a5349d SW |
450 | #ifdef CONFIG_LOCKDEP |
451 | lockdep_set_class(&sock->sk->sk_lock, &socket_class); | |
452 | #endif | |
453 | ||
31b8006e SW |
454 | set_sock_callbacks(sock, con); |
455 | ||
89a86be0 | 456 | con_sock_state_connecting(con); |
cede185b | 457 | ret = sock->ops->connect(sock, (struct sockaddr *)&ss, sizeof(ss), |
f91d3471 | 458 | O_NONBLOCK); |
31b8006e SW |
459 | if (ret == -EINPROGRESS) { |
460 | dout("connect %s EINPROGRESS sk_state = %u\n", | |
b726ec97 | 461 | ceph_pr_addr(&con->peer_addr), |
31b8006e | 462 | sock->sk->sk_state); |
a5bc3129 | 463 | } else if (ret < 0) { |
31b8006e | 464 | pr_err("connect %s error %d\n", |
b726ec97 | 465 | ceph_pr_addr(&con->peer_addr), ret); |
31b8006e | 466 | sock_release(sock); |
41617d0c | 467 | return ret; |
a5bc3129 | 468 | } |
89baaa57 | 469 | |
12abc5ee CH |
470 | if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY)) |
471 | tcp_sock_set_nodelay(sock->sk); | |
ba988f87 | 472 | |
a5bc3129 | 473 | con->sock = sock; |
41617d0c | 474 | return 0; |
31b8006e SW |
475 | } |
476 | ||
31b8006e SW |
477 | /* |
478 | * Shutdown/close the socket for the given connection. | |
479 | */ | |
6503e0b6 | 480 | int ceph_con_close_socket(struct ceph_connection *con) |
31b8006e | 481 | { |
8007b8d6 | 482 | int rc = 0; |
31b8006e | 483 | |
6503e0b6 | 484 | dout("%s con %p sock %p\n", __func__, con, con->sock); |
8007b8d6 SW |
485 | if (con->sock) { |
486 | rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR); | |
487 | sock_release(con->sock); | |
488 | con->sock = NULL; | |
489 | } | |
456ea468 AE |
490 | |
491 | /* | |
4a861692 | 492 | * Forcibly clear the SOCK_CLOSED flag. It gets set |
456ea468 AE |
493 | * independent of the connection mutex, and we could have |
494 | * received a socket close event before we had the chance to | |
495 | * shut the socket down. | |
496 | */ | |
6503e0b6 | 497 | ceph_con_flag_clear(con, CEPH_CON_F_SOCK_CLOSED); |
8007b8d6 | 498 | |
ce2c8903 | 499 | con_sock_state_closed(con); |
31b8006e SW |
500 | return rc; |
501 | } | |
502 | ||
3596f4c1 ID |
503 | static void ceph_con_reset_protocol(struct ceph_connection *con) |
504 | { | |
505 | dout("%s con %p\n", __func__, con); | |
506 | ||
6503e0b6 | 507 | ceph_con_close_socket(con); |
3596f4c1 ID |
508 | if (con->in_msg) { |
509 | WARN_ON(con->in_msg->con != con); | |
510 | ceph_msg_put(con->in_msg); | |
511 | con->in_msg = NULL; | |
512 | } | |
513 | if (con->out_msg) { | |
514 | WARN_ON(con->out_msg->con != con); | |
515 | ceph_msg_put(con->out_msg); | |
516 | con->out_msg = NULL; | |
517 | } | |
518 | ||
cd1a677c ID |
519 | if (ceph_msgr2(from_msgr(con->msgr))) |
520 | ceph_con_v2_reset_protocol(con); | |
521 | else | |
522 | ceph_con_v1_reset_protocol(con); | |
3596f4c1 ID |
523 | } |
524 | ||
31b8006e SW |
525 | /* |
526 | * Reset a connection. Discard all incoming and outgoing messages | |
527 | * and clear *_seq state. | |
528 | */ | |
529 | static void ceph_msg_remove(struct ceph_msg *msg) | |
530 | { | |
531 | list_del_init(&msg->list_head); | |
38941f80 | 532 | |
31b8006e SW |
533 | ceph_msg_put(msg); |
534 | } | |
cd1a677c | 535 | |
31b8006e SW |
536 | static void ceph_msg_remove_list(struct list_head *head) |
537 | { | |
538 | while (!list_empty(head)) { | |
539 | struct ceph_msg *msg = list_first_entry(head, struct ceph_msg, | |
540 | list_head); | |
541 | ceph_msg_remove(msg); | |
542 | } | |
543 | } | |
544 | ||
6503e0b6 | 545 | void ceph_con_reset_session(struct ceph_connection *con) |
31b8006e | 546 | { |
5963c3d0 | 547 | dout("%s con %p\n", __func__, con); |
3596f4c1 ID |
548 | |
549 | WARN_ON(con->in_msg); | |
550 | WARN_ON(con->out_msg); | |
31b8006e SW |
551 | ceph_msg_remove_list(&con->out_queue); |
552 | ceph_msg_remove_list(&con->out_sent); | |
31b8006e | 553 | con->out_seq = 0; |
31b8006e | 554 | con->in_seq = 0; |
0e0d5e0c | 555 | con->in_seq_acked = 0; |
a3da057b | 556 | |
cd1a677c ID |
557 | if (ceph_msgr2(from_msgr(con->msgr))) |
558 | ceph_con_v2_reset_session(con); | |
559 | else | |
560 | ceph_con_v1_reset_session(con); | |
31b8006e SW |
561 | } |
562 | ||
563 | /* | |
564 | * mark a peer down. drop any open connections. | |
565 | */ | |
566 | void ceph_con_close(struct ceph_connection *con) | |
567 | { | |
8c50c817 | 568 | mutex_lock(&con->mutex); |
b726ec97 | 569 | dout("con_close %p peer %s\n", con, ceph_pr_addr(&con->peer_addr)); |
6d7f62bf | 570 | con->state = CEPH_CON_S_CLOSED; |
a5988c49 | 571 | |
6503e0b6 ID |
572 | ceph_con_flag_clear(con, CEPH_CON_F_LOSSYTX); /* so we retry next |
573 | connect */ | |
574 | ceph_con_flag_clear(con, CEPH_CON_F_KEEPALIVE_PENDING); | |
575 | ceph_con_flag_clear(con, CEPH_CON_F_WRITE_PENDING); | |
576 | ceph_con_flag_clear(con, CEPH_CON_F_BACKOFF); | |
a5988c49 | 577 | |
3596f4c1 | 578 | ceph_con_reset_protocol(con); |
5963c3d0 | 579 | ceph_con_reset_session(con); |
37ab77ac | 580 | cancel_con(con); |
ec302645 | 581 | mutex_unlock(&con->mutex); |
31b8006e | 582 | } |
3d14c5d2 | 583 | EXPORT_SYMBOL(ceph_con_close); |
31b8006e | 584 | |
31b8006e SW |
585 | /* |
586 | * Reopen a closed connection, with a new peer address. | |
587 | */ | |
b7a9e5dd SW |
588 | void ceph_con_open(struct ceph_connection *con, |
589 | __u8 entity_type, __u64 entity_num, | |
590 | struct ceph_entity_addr *addr) | |
31b8006e | 591 | { |
5469155f | 592 | mutex_lock(&con->mutex); |
b726ec97 | 593 | dout("con_open %p %s\n", con, ceph_pr_addr(addr)); |
8dacc7da | 594 | |
6d7f62bf ID |
595 | WARN_ON(con->state != CEPH_CON_S_CLOSED); |
596 | con->state = CEPH_CON_S_PREOPEN; | |
a5988c49 | 597 | |
b7a9e5dd SW |
598 | con->peer_name.type = (__u8) entity_type; |
599 | con->peer_name.num = cpu_to_le64(entity_num); | |
600 | ||
31b8006e | 601 | memcpy(&con->peer_addr, addr, sizeof(*addr)); |
03c677e1 | 602 | con->delay = 0; /* reset backoff memory */ |
5469155f | 603 | mutex_unlock(&con->mutex); |
31b8006e SW |
604 | queue_con(con); |
605 | } | |
3d14c5d2 | 606 | EXPORT_SYMBOL(ceph_con_open); |
31b8006e | 607 | |
87b315a5 SW |
608 | /* |
609 | * return true if this connection ever successfully opened | |
610 | */ | |
611 | bool ceph_con_opened(struct ceph_connection *con) | |
612 | { | |
cd1a677c ID |
613 | if (ceph_msgr2(from_msgr(con->msgr))) |
614 | return ceph_con_v2_opened(con); | |
615 | ||
566050e1 | 616 | return ceph_con_v1_opened(con); |
87b315a5 SW |
617 | } |
618 | ||
31b8006e SW |
619 | /* |
620 | * initialize a new connection. | |
621 | */ | |
1bfd89f4 AE |
622 | void ceph_con_init(struct ceph_connection *con, void *private, |
623 | const struct ceph_connection_operations *ops, | |
b7a9e5dd | 624 | struct ceph_messenger *msgr) |
31b8006e SW |
625 | { |
626 | dout("con_init %p\n", con); | |
627 | memset(con, 0, sizeof(*con)); | |
1bfd89f4 AE |
628 | con->private = private; |
629 | con->ops = ops; | |
31b8006e | 630 | con->msgr = msgr; |
ce2c8903 AE |
631 | |
632 | con_sock_state_init(con); | |
633 | ||
ec302645 | 634 | mutex_init(&con->mutex); |
31b8006e SW |
635 | INIT_LIST_HEAD(&con->out_queue); |
636 | INIT_LIST_HEAD(&con->out_sent); | |
68931622 | 637 | INIT_DELAYED_WORK(&con->work, ceph_con_workfn); |
a5988c49 | 638 | |
6d7f62bf | 639 | con->state = CEPH_CON_S_CLOSED; |
31b8006e | 640 | } |
3d14c5d2 | 641 | EXPORT_SYMBOL(ceph_con_init); |
31b8006e | 642 | |
31b8006e SW |
643 | /* |
644 | * We maintain a global counter to order connection attempts. Get | |
645 | * a unique seq greater than @gt. | |
646 | */ | |
6503e0b6 | 647 | u32 ceph_get_global_seq(struct ceph_messenger *msgr, u32 gt) |
31b8006e SW |
648 | { |
649 | u32 ret; | |
650 | ||
651 | spin_lock(&msgr->global_seq_lock); | |
652 | if (msgr->global_seq < gt) | |
653 | msgr->global_seq = gt; | |
654 | ret = ++msgr->global_seq; | |
655 | spin_unlock(&msgr->global_seq_lock); | |
656 | return ret; | |
657 | } | |
658 | ||
02471928 ID |
659 | /* |
660 | * Discard messages that have been acked by the server. | |
661 | */ | |
6503e0b6 | 662 | void ceph_con_discard_sent(struct ceph_connection *con, u64 ack_seq) |
02471928 ID |
663 | { |
664 | struct ceph_msg *msg; | |
665 | u64 seq; | |
666 | ||
667 | dout("%s con %p ack_seq %llu\n", __func__, con, ack_seq); | |
668 | while (!list_empty(&con->out_sent)) { | |
669 | msg = list_first_entry(&con->out_sent, struct ceph_msg, | |
670 | list_head); | |
671 | WARN_ON(msg->needs_out_seq); | |
672 | seq = le64_to_cpu(msg->hdr.seq); | |
673 | if (seq > ack_seq) | |
674 | break; | |
675 | ||
676 | dout("%s con %p discarding msg %p seq %llu\n", __func__, con, | |
677 | msg, seq); | |
678 | ceph_msg_remove(msg); | |
679 | } | |
680 | } | |
681 | ||
682 | /* | |
683 | * Discard messages that have been requeued in con_fault(), up to | |
684 | * reconnect_seq. This avoids gratuitously resending messages that | |
685 | * the server had received and handled prior to reconnect. | |
686 | */ | |
6503e0b6 | 687 | void ceph_con_discard_requeued(struct ceph_connection *con, u64 reconnect_seq) |
02471928 ID |
688 | { |
689 | struct ceph_msg *msg; | |
690 | u64 seq; | |
691 | ||
692 | dout("%s con %p reconnect_seq %llu\n", __func__, con, reconnect_seq); | |
693 | while (!list_empty(&con->out_queue)) { | |
694 | msg = list_first_entry(&con->out_queue, struct ceph_msg, | |
695 | list_head); | |
696 | if (msg->needs_out_seq) | |
697 | break; | |
698 | seq = le64_to_cpu(msg->hdr.seq); | |
699 | if (seq > reconnect_seq) | |
700 | break; | |
701 | ||
702 | dout("%s con %p discarding msg %p seq %llu\n", __func__, con, | |
703 | msg, seq); | |
704 | ceph_msg_remove(msg); | |
705 | } | |
706 | } | |
707 | ||
df6ad1f9 | 708 | #ifdef CONFIG_BLOCK |
6aaa4511 AE |
709 | |
710 | /* | |
711 | * For a bio data item, a piece is whatever remains of the next | |
712 | * entry in the current bio iovec, or the first entry in the next | |
713 | * bio in the list. | |
714 | */ | |
8ae4f4f5 | 715 | static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor, |
25aff7c5 | 716 | size_t length) |
6aaa4511 | 717 | { |
8ae4f4f5 | 718 | struct ceph_msg_data *data = cursor->data; |
5359a17d | 719 | struct ceph_bio_iter *it = &cursor->bio_iter; |
6aaa4511 | 720 | |
5359a17d ID |
721 | cursor->resid = min_t(size_t, length, data->bio_length); |
722 | *it = data->bio_pos; | |
723 | if (cursor->resid < it->iter.bi_size) | |
724 | it->iter.bi_size = cursor->resid; | |
6aaa4511 | 725 | |
5359a17d ID |
726 | BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter)); |
727 | cursor->last_piece = cursor->resid == bio_iter_len(it->bio, it->iter); | |
6aaa4511 AE |
728 | } |
729 | ||
8ae4f4f5 | 730 | static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor, |
6aaa4511 AE |
731 | size_t *page_offset, |
732 | size_t *length) | |
733 | { | |
5359a17d ID |
734 | struct bio_vec bv = bio_iter_iovec(cursor->bio_iter.bio, |
735 | cursor->bio_iter.iter); | |
6aaa4511 | 736 | |
5359a17d ID |
737 | *page_offset = bv.bv_offset; |
738 | *length = bv.bv_len; | |
739 | return bv.bv_page; | |
6aaa4511 AE |
740 | } |
741 | ||
8ae4f4f5 AE |
742 | static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor, |
743 | size_t bytes) | |
6aaa4511 | 744 | { |
5359a17d | 745 | struct ceph_bio_iter *it = &cursor->bio_iter; |
187df763 | 746 | struct page *page = bio_iter_page(it->bio, it->iter); |
6aaa4511 | 747 | |
5359a17d ID |
748 | BUG_ON(bytes > cursor->resid); |
749 | BUG_ON(bytes > bio_iter_len(it->bio, it->iter)); | |
25aff7c5 | 750 | cursor->resid -= bytes; |
5359a17d | 751 | bio_advance_iter(it->bio, &it->iter, bytes); |
f38a5181 | 752 | |
5359a17d ID |
753 | if (!cursor->resid) { |
754 | BUG_ON(!cursor->last_piece); | |
755 | return false; /* no more data */ | |
756 | } | |
f38a5181 | 757 | |
187df763 ID |
758 | if (!bytes || (it->iter.bi_size && it->iter.bi_bvec_done && |
759 | page == bio_iter_page(it->bio, it->iter))) | |
6aaa4511 AE |
760 | return false; /* more bytes to process in this segment */ |
761 | ||
5359a17d ID |
762 | if (!it->iter.bi_size) { |
763 | it->bio = it->bio->bi_next; | |
764 | it->iter = it->bio->bi_iter; | |
765 | if (cursor->resid < it->iter.bi_size) | |
766 | it->iter.bi_size = cursor->resid; | |
25aff7c5 | 767 | } |
6aaa4511 | 768 | |
5359a17d ID |
769 | BUG_ON(cursor->last_piece); |
770 | BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter)); | |
771 | cursor->last_piece = cursor->resid == bio_iter_len(it->bio, it->iter); | |
6aaa4511 AE |
772 | return true; |
773 | } | |
ea96571f | 774 | #endif /* CONFIG_BLOCK */ |
df6ad1f9 | 775 | |
b9e281c2 ID |
776 | static void ceph_msg_data_bvecs_cursor_init(struct ceph_msg_data_cursor *cursor, |
777 | size_t length) | |
778 | { | |
779 | struct ceph_msg_data *data = cursor->data; | |
780 | struct bio_vec *bvecs = data->bvec_pos.bvecs; | |
781 | ||
782 | cursor->resid = min_t(size_t, length, data->bvec_pos.iter.bi_size); | |
783 | cursor->bvec_iter = data->bvec_pos.iter; | |
784 | cursor->bvec_iter.bi_size = cursor->resid; | |
785 | ||
786 | BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter)); | |
787 | cursor->last_piece = | |
788 | cursor->resid == bvec_iter_len(bvecs, cursor->bvec_iter); | |
789 | } | |
790 | ||
791 | static struct page *ceph_msg_data_bvecs_next(struct ceph_msg_data_cursor *cursor, | |
792 | size_t *page_offset, | |
793 | size_t *length) | |
794 | { | |
795 | struct bio_vec bv = bvec_iter_bvec(cursor->data->bvec_pos.bvecs, | |
796 | cursor->bvec_iter); | |
797 | ||
798 | *page_offset = bv.bv_offset; | |
799 | *length = bv.bv_len; | |
800 | return bv.bv_page; | |
801 | } | |
802 | ||
803 | static bool ceph_msg_data_bvecs_advance(struct ceph_msg_data_cursor *cursor, | |
804 | size_t bytes) | |
805 | { | |
806 | struct bio_vec *bvecs = cursor->data->bvec_pos.bvecs; | |
187df763 | 807 | struct page *page = bvec_iter_page(bvecs, cursor->bvec_iter); |
b9e281c2 ID |
808 | |
809 | BUG_ON(bytes > cursor->resid); | |
810 | BUG_ON(bytes > bvec_iter_len(bvecs, cursor->bvec_iter)); | |
811 | cursor->resid -= bytes; | |
812 | bvec_iter_advance(bvecs, &cursor->bvec_iter, bytes); | |
813 | ||
814 | if (!cursor->resid) { | |
815 | BUG_ON(!cursor->last_piece); | |
816 | return false; /* no more data */ | |
817 | } | |
818 | ||
187df763 ID |
819 | if (!bytes || (cursor->bvec_iter.bi_bvec_done && |
820 | page == bvec_iter_page(bvecs, cursor->bvec_iter))) | |
b9e281c2 ID |
821 | return false; /* more bytes to process in this segment */ |
822 | ||
823 | BUG_ON(cursor->last_piece); | |
824 | BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter)); | |
825 | cursor->last_piece = | |
826 | cursor->resid == bvec_iter_len(bvecs, cursor->bvec_iter); | |
827 | return true; | |
828 | } | |
829 | ||
e766d7b5 AE |
830 | /* |
831 | * For a page array, a piece comes from the first page in the array | |
832 | * that has not already been fully consumed. | |
833 | */ | |
8ae4f4f5 | 834 | static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor, |
25aff7c5 | 835 | size_t length) |
e766d7b5 | 836 | { |
8ae4f4f5 | 837 | struct ceph_msg_data *data = cursor->data; |
e766d7b5 AE |
838 | int page_count; |
839 | ||
840 | BUG_ON(data->type != CEPH_MSG_DATA_PAGES); | |
841 | ||
842 | BUG_ON(!data->pages); | |
843 | BUG_ON(!data->length); | |
844 | ||
ca8b3a69 | 845 | cursor->resid = min(length, data->length); |
e766d7b5 | 846 | page_count = calc_pages_for(data->alignment, (u64)data->length); |
e766d7b5 AE |
847 | cursor->page_offset = data->alignment & ~PAGE_MASK; |
848 | cursor->page_index = 0; | |
56fc5659 AE |
849 | BUG_ON(page_count > (int)USHRT_MAX); |
850 | cursor->page_count = (unsigned short)page_count; | |
851 | BUG_ON(length > SIZE_MAX - cursor->page_offset); | |
5f740d7e | 852 | cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE; |
e766d7b5 AE |
853 | } |
854 | ||
8ae4f4f5 AE |
855 | static struct page * |
856 | ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor, | |
857 | size_t *page_offset, size_t *length) | |
e766d7b5 | 858 | { |
8ae4f4f5 | 859 | struct ceph_msg_data *data = cursor->data; |
e766d7b5 AE |
860 | |
861 | BUG_ON(data->type != CEPH_MSG_DATA_PAGES); | |
862 | ||
863 | BUG_ON(cursor->page_index >= cursor->page_count); | |
864 | BUG_ON(cursor->page_offset >= PAGE_SIZE); | |
e766d7b5 AE |
865 | |
866 | *page_offset = cursor->page_offset; | |
25aff7c5 | 867 | if (cursor->last_piece) |
e766d7b5 | 868 | *length = cursor->resid; |
25aff7c5 | 869 | else |
e766d7b5 | 870 | *length = PAGE_SIZE - *page_offset; |
e766d7b5 AE |
871 | |
872 | return data->pages[cursor->page_index]; | |
873 | } | |
874 | ||
8ae4f4f5 | 875 | static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor, |
e766d7b5 AE |
876 | size_t bytes) |
877 | { | |
8ae4f4f5 | 878 | BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES); |
e766d7b5 AE |
879 | |
880 | BUG_ON(cursor->page_offset + bytes > PAGE_SIZE); | |
e766d7b5 AE |
881 | |
882 | /* Advance the cursor page offset */ | |
883 | ||
884 | cursor->resid -= bytes; | |
5df521b1 AE |
885 | cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK; |
886 | if (!bytes || cursor->page_offset) | |
e766d7b5 AE |
887 | return false; /* more bytes to process in the current page */ |
888 | ||
d90deda6 YZ |
889 | if (!cursor->resid) |
890 | return false; /* no more data */ | |
891 | ||
5df521b1 | 892 | /* Move on to the next page; offset is already at 0 */ |
e766d7b5 AE |
893 | |
894 | BUG_ON(cursor->page_index >= cursor->page_count); | |
e766d7b5 | 895 | cursor->page_index++; |
25aff7c5 | 896 | cursor->last_piece = cursor->resid <= PAGE_SIZE; |
e766d7b5 AE |
897 | |
898 | return true; | |
899 | } | |
900 | ||
fe38a2b6 | 901 | /* |
dd236fcb AE |
902 | * For a pagelist, a piece is whatever remains to be consumed in the |
903 | * first page in the list, or the front of the next page. | |
fe38a2b6 | 904 | */ |
8ae4f4f5 AE |
905 | static void |
906 | ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor, | |
25aff7c5 | 907 | size_t length) |
fe38a2b6 | 908 | { |
8ae4f4f5 | 909 | struct ceph_msg_data *data = cursor->data; |
fe38a2b6 AE |
910 | struct ceph_pagelist *pagelist; |
911 | struct page *page; | |
912 | ||
dd236fcb | 913 | BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST); |
fe38a2b6 AE |
914 | |
915 | pagelist = data->pagelist; | |
916 | BUG_ON(!pagelist); | |
25aff7c5 AE |
917 | |
918 | if (!length) | |
fe38a2b6 AE |
919 | return; /* pagelist can be assigned but empty */ |
920 | ||
921 | BUG_ON(list_empty(&pagelist->head)); | |
922 | page = list_first_entry(&pagelist->head, struct page, lru); | |
923 | ||
ca8b3a69 | 924 | cursor->resid = min(length, pagelist->length); |
fe38a2b6 AE |
925 | cursor->page = page; |
926 | cursor->offset = 0; | |
a51b272e | 927 | cursor->last_piece = cursor->resid <= PAGE_SIZE; |
fe38a2b6 AE |
928 | } |
929 | ||
8ae4f4f5 AE |
930 | static struct page * |
931 | ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor, | |
932 | size_t *page_offset, size_t *length) | |
fe38a2b6 | 933 | { |
8ae4f4f5 | 934 | struct ceph_msg_data *data = cursor->data; |
fe38a2b6 | 935 | struct ceph_pagelist *pagelist; |
fe38a2b6 AE |
936 | |
937 | BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST); | |
938 | ||
939 | pagelist = data->pagelist; | |
940 | BUG_ON(!pagelist); | |
941 | ||
942 | BUG_ON(!cursor->page); | |
25aff7c5 | 943 | BUG_ON(cursor->offset + cursor->resid != pagelist->length); |
fe38a2b6 | 944 | |
5df521b1 | 945 | /* offset of first page in pagelist is always 0 */ |
fe38a2b6 | 946 | *page_offset = cursor->offset & ~PAGE_MASK; |
5df521b1 | 947 | if (cursor->last_piece) |
25aff7c5 AE |
948 | *length = cursor->resid; |
949 | else | |
950 | *length = PAGE_SIZE - *page_offset; | |
fe38a2b6 | 951 | |
8ae4f4f5 | 952 | return cursor->page; |
fe38a2b6 AE |
953 | } |
954 | ||
8ae4f4f5 | 955 | static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor, |
dd236fcb | 956 | size_t bytes) |
fe38a2b6 | 957 | { |
8ae4f4f5 | 958 | struct ceph_msg_data *data = cursor->data; |
fe38a2b6 AE |
959 | struct ceph_pagelist *pagelist; |
960 | ||
961 | BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST); | |
962 | ||
963 | pagelist = data->pagelist; | |
964 | BUG_ON(!pagelist); | |
25aff7c5 AE |
965 | |
966 | BUG_ON(cursor->offset + cursor->resid != pagelist->length); | |
fe38a2b6 AE |
967 | BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE); |
968 | ||
969 | /* Advance the cursor offset */ | |
970 | ||
25aff7c5 | 971 | cursor->resid -= bytes; |
fe38a2b6 | 972 | cursor->offset += bytes; |
5df521b1 | 973 | /* offset of first page in pagelist is always 0 */ |
fe38a2b6 AE |
974 | if (!bytes || cursor->offset & ~PAGE_MASK) |
975 | return false; /* more bytes to process in the current page */ | |
976 | ||
d90deda6 YZ |
977 | if (!cursor->resid) |
978 | return false; /* no more data */ | |
979 | ||
fe38a2b6 AE |
980 | /* Move on to the next page */ |
981 | ||
982 | BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head)); | |
17ddc49b | 983 | cursor->page = list_next_entry(cursor->page, lru); |
25aff7c5 | 984 | cursor->last_piece = cursor->resid <= PAGE_SIZE; |
fe38a2b6 AE |
985 | |
986 | return true; | |
987 | } | |
988 | ||
dd236fcb AE |
989 | /* |
990 | * Message data is handled (sent or received) in pieces, where each | |
991 | * piece resides on a single page. The network layer might not | |
992 | * consume an entire piece at once. A data item's cursor keeps | |
993 | * track of which piece is next to process and how much remains to | |
994 | * be processed in that piece. It also tracks whether the current | |
995 | * piece is the last one in the data item. | |
996 | */ | |
ca8b3a69 | 997 | static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor) |
dd236fcb | 998 | { |
ca8b3a69 | 999 | size_t length = cursor->total_resid; |
8ae4f4f5 | 1000 | |
8ae4f4f5 | 1001 | switch (cursor->data->type) { |
dd236fcb | 1002 | case CEPH_MSG_DATA_PAGELIST: |
8ae4f4f5 | 1003 | ceph_msg_data_pagelist_cursor_init(cursor, length); |
dd236fcb | 1004 | break; |
e766d7b5 | 1005 | case CEPH_MSG_DATA_PAGES: |
8ae4f4f5 | 1006 | ceph_msg_data_pages_cursor_init(cursor, length); |
e766d7b5 | 1007 | break; |
dd236fcb AE |
1008 | #ifdef CONFIG_BLOCK |
1009 | case CEPH_MSG_DATA_BIO: | |
8ae4f4f5 | 1010 | ceph_msg_data_bio_cursor_init(cursor, length); |
6aaa4511 | 1011 | break; |
dd236fcb | 1012 | #endif /* CONFIG_BLOCK */ |
b9e281c2 ID |
1013 | case CEPH_MSG_DATA_BVECS: |
1014 | ceph_msg_data_bvecs_cursor_init(cursor, length); | |
1015 | break; | |
6aaa4511 | 1016 | case CEPH_MSG_DATA_NONE: |
dd236fcb AE |
1017 | default: |
1018 | /* BUG(); */ | |
1019 | break; | |
1020 | } | |
8ae4f4f5 | 1021 | cursor->need_crc = true; |
dd236fcb AE |
1022 | } |
1023 | ||
6503e0b6 ID |
1024 | void ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor, |
1025 | struct ceph_msg *msg, size_t length) | |
ca8b3a69 | 1026 | { |
ca8b3a69 AE |
1027 | BUG_ON(!length); |
1028 | BUG_ON(length > msg->data_length); | |
0d9c1ab3 | 1029 | BUG_ON(!msg->num_data_items); |
ca8b3a69 | 1030 | |
ca8b3a69 | 1031 | cursor->total_resid = length; |
0d9c1ab3 | 1032 | cursor->data = msg->data; |
ca8b3a69 AE |
1033 | |
1034 | __ceph_msg_data_cursor_init(cursor); | |
1035 | } | |
1036 | ||
dd236fcb AE |
1037 | /* |
1038 | * Return the page containing the next piece to process for a given | |
1039 | * data item, and supply the page offset and length of that piece. | |
1040 | * Indicate whether this is the last piece in this data item. | |
1041 | */ | |
6503e0b6 ID |
1042 | struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor, |
1043 | size_t *page_offset, size_t *length, | |
1044 | bool *last_piece) | |
dd236fcb AE |
1045 | { |
1046 | struct page *page; | |
1047 | ||
8ae4f4f5 | 1048 | switch (cursor->data->type) { |
dd236fcb | 1049 | case CEPH_MSG_DATA_PAGELIST: |
8ae4f4f5 | 1050 | page = ceph_msg_data_pagelist_next(cursor, page_offset, length); |
dd236fcb | 1051 | break; |
e766d7b5 | 1052 | case CEPH_MSG_DATA_PAGES: |
8ae4f4f5 | 1053 | page = ceph_msg_data_pages_next(cursor, page_offset, length); |
e766d7b5 | 1054 | break; |
dd236fcb AE |
1055 | #ifdef CONFIG_BLOCK |
1056 | case CEPH_MSG_DATA_BIO: | |
8ae4f4f5 | 1057 | page = ceph_msg_data_bio_next(cursor, page_offset, length); |
6aaa4511 | 1058 | break; |
dd236fcb | 1059 | #endif /* CONFIG_BLOCK */ |
b9e281c2 ID |
1060 | case CEPH_MSG_DATA_BVECS: |
1061 | page = ceph_msg_data_bvecs_next(cursor, page_offset, length); | |
1062 | break; | |
6aaa4511 | 1063 | case CEPH_MSG_DATA_NONE: |
dd236fcb AE |
1064 | default: |
1065 | page = NULL; | |
1066 | break; | |
1067 | } | |
5359a17d | 1068 | |
dd236fcb AE |
1069 | BUG_ON(!page); |
1070 | BUG_ON(*page_offset + *length > PAGE_SIZE); | |
1071 | BUG_ON(!*length); | |
5359a17d | 1072 | BUG_ON(*length > cursor->resid); |
dd236fcb | 1073 | if (last_piece) |
8ae4f4f5 | 1074 | *last_piece = cursor->last_piece; |
dd236fcb AE |
1075 | |
1076 | return page; | |
1077 | } | |
1078 | ||
1079 | /* | |
1080 | * Returns true if the result moves the cursor on to the next piece | |
1081 | * of the data item. | |
1082 | */ | |
6503e0b6 | 1083 | void ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor, size_t bytes) |
dd236fcb AE |
1084 | { |
1085 | bool new_piece; | |
1086 | ||
25aff7c5 | 1087 | BUG_ON(bytes > cursor->resid); |
8ae4f4f5 | 1088 | switch (cursor->data->type) { |
dd236fcb | 1089 | case CEPH_MSG_DATA_PAGELIST: |
8ae4f4f5 | 1090 | new_piece = ceph_msg_data_pagelist_advance(cursor, bytes); |
dd236fcb | 1091 | break; |
e766d7b5 | 1092 | case CEPH_MSG_DATA_PAGES: |
8ae4f4f5 | 1093 | new_piece = ceph_msg_data_pages_advance(cursor, bytes); |
e766d7b5 | 1094 | break; |
dd236fcb AE |
1095 | #ifdef CONFIG_BLOCK |
1096 | case CEPH_MSG_DATA_BIO: | |
8ae4f4f5 | 1097 | new_piece = ceph_msg_data_bio_advance(cursor, bytes); |
6aaa4511 | 1098 | break; |
dd236fcb | 1099 | #endif /* CONFIG_BLOCK */ |
b9e281c2 ID |
1100 | case CEPH_MSG_DATA_BVECS: |
1101 | new_piece = ceph_msg_data_bvecs_advance(cursor, bytes); | |
1102 | break; | |
6aaa4511 | 1103 | case CEPH_MSG_DATA_NONE: |
dd236fcb AE |
1104 | default: |
1105 | BUG(); | |
1106 | break; | |
1107 | } | |
ca8b3a69 | 1108 | cursor->total_resid -= bytes; |
dd236fcb | 1109 | |
ca8b3a69 AE |
1110 | if (!cursor->resid && cursor->total_resid) { |
1111 | WARN_ON(!cursor->last_piece); | |
0d9c1ab3 | 1112 | cursor->data++; |
ca8b3a69 | 1113 | __ceph_msg_data_cursor_init(cursor); |
a51b272e | 1114 | new_piece = true; |
ca8b3a69 | 1115 | } |
a51b272e | 1116 | cursor->need_crc = new_piece; |
dd236fcb AE |
1117 | } |
1118 | ||
6503e0b6 ID |
1119 | u32 ceph_crc32c_page(u32 crc, struct page *page, unsigned int page_offset, |
1120 | unsigned int length) | |
35b62808 AE |
1121 | { |
1122 | char *kaddr; | |
1123 | ||
1124 | kaddr = kmap(page); | |
1125 | BUG_ON(kaddr == NULL); | |
1126 | crc = crc32c(crc, kaddr + page_offset, length); | |
1127 | kunmap(page); | |
1128 | ||
1129 | return crc; | |
1130 | } | |
31b8006e | 1131 | |
6503e0b6 | 1132 | bool ceph_addr_is_blank(const struct ceph_entity_addr *addr) |
31b8006e | 1133 | { |
cede185b JL |
1134 | struct sockaddr_storage ss = addr->in_addr; /* align */ |
1135 | struct in_addr *addr4 = &((struct sockaddr_in *)&ss)->sin_addr; | |
1136 | struct in6_addr *addr6 = &((struct sockaddr_in6 *)&ss)->sin6_addr; | |
c44bd69c | 1137 | |
cede185b | 1138 | switch (ss.ss_family) { |
31b8006e | 1139 | case AF_INET: |
cede185b | 1140 | return addr4->s_addr == htonl(INADDR_ANY); |
31b8006e | 1141 | case AF_INET6: |
c44bd69c ID |
1142 | return ipv6_addr_any(addr6); |
1143 | default: | |
1144 | return true; | |
31b8006e | 1145 | } |
31b8006e SW |
1146 | } |
1147 | ||
6503e0b6 | 1148 | int ceph_addr_port(const struct ceph_entity_addr *addr) |
31b8006e | 1149 | { |
cede185b | 1150 | switch (get_unaligned(&addr->in_addr.ss_family)) { |
31b8006e | 1151 | case AF_INET: |
cede185b | 1152 | return ntohs(get_unaligned(&((struct sockaddr_in *)&addr->in_addr)->sin_port)); |
31b8006e | 1153 | case AF_INET6: |
cede185b | 1154 | return ntohs(get_unaligned(&((struct sockaddr_in6 *)&addr->in_addr)->sin6_port)); |
31b8006e SW |
1155 | } |
1156 | return 0; | |
1157 | } | |
1158 | ||
6503e0b6 | 1159 | void ceph_addr_set_port(struct ceph_entity_addr *addr, int p) |
31b8006e | 1160 | { |
cede185b | 1161 | switch (get_unaligned(&addr->in_addr.ss_family)) { |
31b8006e | 1162 | case AF_INET: |
cede185b | 1163 | put_unaligned(htons(p), &((struct sockaddr_in *)&addr->in_addr)->sin_port); |
a2a79609 | 1164 | break; |
31b8006e | 1165 | case AF_INET6: |
cede185b | 1166 | put_unaligned(htons(p), &((struct sockaddr_in6 *)&addr->in_addr)->sin6_port); |
a2a79609 | 1167 | break; |
31b8006e SW |
1168 | } |
1169 | } | |
1170 | ||
ee3b56f2 NW |
1171 | /* |
1172 | * Unlike other *_pton function semantics, zero indicates success. | |
1173 | */ | |
cede185b | 1174 | static int ceph_pton(const char *str, size_t len, struct ceph_entity_addr *addr, |
ee3b56f2 NW |
1175 | char delim, const char **ipend) |
1176 | { | |
cede185b | 1177 | memset(&addr->in_addr, 0, sizeof(addr->in_addr)); |
ee3b56f2 | 1178 | |
cede185b JL |
1179 | if (in4_pton(str, len, (u8 *)&((struct sockaddr_in *)&addr->in_addr)->sin_addr.s_addr, delim, ipend)) { |
1180 | put_unaligned(AF_INET, &addr->in_addr.ss_family); | |
ee3b56f2 NW |
1181 | return 0; |
1182 | } | |
1183 | ||
cede185b JL |
1184 | if (in6_pton(str, len, (u8 *)&((struct sockaddr_in6 *)&addr->in_addr)->sin6_addr.s6_addr, delim, ipend)) { |
1185 | put_unaligned(AF_INET6, &addr->in_addr.ss_family); | |
ee3b56f2 NW |
1186 | return 0; |
1187 | } | |
1188 | ||
1189 | return -EINVAL; | |
1190 | } | |
1191 | ||
1192 | /* | |
1193 | * Extract hostname string and resolve using kernel DNS facility. | |
1194 | */ | |
1195 | #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER | |
1196 | static int ceph_dns_resolve_name(const char *name, size_t namelen, | |
cede185b | 1197 | struct ceph_entity_addr *addr, char delim, const char **ipend) |
ee3b56f2 NW |
1198 | { |
1199 | const char *end, *delim_p; | |
1200 | char *colon_p, *ip_addr = NULL; | |
1201 | int ip_len, ret; | |
1202 | ||
1203 | /* | |
1204 | * The end of the hostname occurs immediately preceding the delimiter or | |
1205 | * the port marker (':') where the delimiter takes precedence. | |
1206 | */ | |
1207 | delim_p = memchr(name, delim, namelen); | |
1208 | colon_p = memchr(name, ':', namelen); | |
1209 | ||
1210 | if (delim_p && colon_p) | |
1211 | end = delim_p < colon_p ? delim_p : colon_p; | |
1212 | else if (!delim_p && colon_p) | |
1213 | end = colon_p; | |
1214 | else { | |
1215 | end = delim_p; | |
1216 | if (!end) /* case: hostname:/ */ | |
1217 | end = name + namelen; | |
1218 | } | |
1219 | ||
1220 | if (end <= name) | |
1221 | return -EINVAL; | |
1222 | ||
1223 | /* do dns_resolve upcall */ | |
a58946c1 DH |
1224 | ip_len = dns_query(current->nsproxy->net_ns, |
1225 | NULL, name, end - name, NULL, &ip_addr, NULL, false); | |
ee3b56f2 | 1226 | if (ip_len > 0) |
cede185b | 1227 | ret = ceph_pton(ip_addr, ip_len, addr, -1, NULL); |
ee3b56f2 NW |
1228 | else |
1229 | ret = -ESRCH; | |
1230 | ||
1231 | kfree(ip_addr); | |
1232 | ||
1233 | *ipend = end; | |
1234 | ||
1235 | pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name, | |
b726ec97 | 1236 | ret, ret ? "failed" : ceph_pr_addr(addr)); |
ee3b56f2 NW |
1237 | |
1238 | return ret; | |
1239 | } | |
1240 | #else | |
1241 | static inline int ceph_dns_resolve_name(const char *name, size_t namelen, | |
cede185b | 1242 | struct ceph_entity_addr *addr, char delim, const char **ipend) |
ee3b56f2 NW |
1243 | { |
1244 | return -EINVAL; | |
1245 | } | |
1246 | #endif | |
1247 | ||
1248 | /* | |
1249 | * Parse a server name (IP or hostname). If a valid IP address is not found | |
1250 | * then try to extract a hostname to resolve using userspace DNS upcall. | |
1251 | */ | |
1252 | static int ceph_parse_server_name(const char *name, size_t namelen, | |
cede185b | 1253 | struct ceph_entity_addr *addr, char delim, const char **ipend) |
ee3b56f2 NW |
1254 | { |
1255 | int ret; | |
1256 | ||
cede185b | 1257 | ret = ceph_pton(name, namelen, addr, delim, ipend); |
ee3b56f2 | 1258 | if (ret) |
cede185b | 1259 | ret = ceph_dns_resolve_name(name, namelen, addr, delim, ipend); |
ee3b56f2 NW |
1260 | |
1261 | return ret; | |
1262 | } | |
1263 | ||
31b8006e SW |
1264 | /* |
1265 | * Parse an ip[:port] list into an addr array. Use the default | |
1266 | * monitor port if a port isn't specified. | |
1267 | */ | |
1268 | int ceph_parse_ips(const char *c, const char *end, | |
1269 | struct ceph_entity_addr *addr, | |
1270 | int max_count, int *count) | |
1271 | { | |
ee3b56f2 | 1272 | int i, ret = -EINVAL; |
31b8006e SW |
1273 | const char *p = c; |
1274 | ||
1275 | dout("parse_ips on '%.*s'\n", (int)(end-c), c); | |
1276 | for (i = 0; i < max_count; i++) { | |
1277 | const char *ipend; | |
31b8006e | 1278 | int port; |
39139f64 SW |
1279 | char delim = ','; |
1280 | ||
1281 | if (*p == '[') { | |
1282 | delim = ']'; | |
1283 | p++; | |
1284 | } | |
31b8006e | 1285 | |
cede185b | 1286 | ret = ceph_parse_server_name(p, end - p, &addr[i], delim, &ipend); |
ee3b56f2 | 1287 | if (ret) |
31b8006e | 1288 | goto bad; |
ee3b56f2 NW |
1289 | ret = -EINVAL; |
1290 | ||
31b8006e SW |
1291 | p = ipend; |
1292 | ||
39139f64 SW |
1293 | if (delim == ']') { |
1294 | if (*p != ']') { | |
1295 | dout("missing matching ']'\n"); | |
1296 | goto bad; | |
1297 | } | |
1298 | p++; | |
1299 | } | |
1300 | ||
31b8006e SW |
1301 | /* port? */ |
1302 | if (p < end && *p == ':') { | |
1303 | port = 0; | |
1304 | p++; | |
1305 | while (p < end && *p >= '0' && *p <= '9') { | |
1306 | port = (port * 10) + (*p - '0'); | |
1307 | p++; | |
1308 | } | |
f48db1e9 ID |
1309 | if (port == 0) |
1310 | port = CEPH_MON_PORT; | |
1311 | else if (port > 65535) | |
31b8006e SW |
1312 | goto bad; |
1313 | } else { | |
1314 | port = CEPH_MON_PORT; | |
1315 | } | |
1316 | ||
6503e0b6 | 1317 | ceph_addr_set_port(&addr[i], port); |
cd1a677c ID |
1318 | /* |
1319 | * We want the type to be set according to ms_mode | |
1320 | * option, but options are normally parsed after mon | |
1321 | * addresses. Rather than complicating parsing, set | |
1322 | * to LEGACY and override in build_initial_monmap() | |
1323 | * for mon addresses and ceph_messenger_init() for | |
1324 | * ip option. | |
1325 | */ | |
d3c3c0a8 | 1326 | addr[i].type = CEPH_ENTITY_ADDR_TYPE_LEGACY; |
cd1a677c | 1327 | addr[i].nonce = 0; |
31b8006e | 1328 | |
b726ec97 | 1329 | dout("parse_ips got %s\n", ceph_pr_addr(&addr[i])); |
31b8006e SW |
1330 | |
1331 | if (p == end) | |
1332 | break; | |
1333 | if (*p != ',') | |
1334 | goto bad; | |
1335 | p++; | |
1336 | } | |
1337 | ||
1338 | if (p != end) | |
1339 | goto bad; | |
1340 | ||
1341 | if (count) | |
1342 | *count = i + 1; | |
1343 | return 0; | |
1344 | ||
1345 | bad: | |
ee3b56f2 | 1346 | return ret; |
31b8006e SW |
1347 | } |
1348 | ||
31b8006e SW |
1349 | /* |
1350 | * Process message. This happens in the worker thread. The callback should | |
1351 | * be careful not to do anything that waits on other incoming messages or it | |
1352 | * may deadlock. | |
1353 | */ | |
6503e0b6 | 1354 | void ceph_con_process_message(struct ceph_connection *con) |
31b8006e | 1355 | { |
583d0fef | 1356 | struct ceph_msg *msg = con->in_msg; |
31b8006e | 1357 | |
38941f80 | 1358 | BUG_ON(con->in_msg->con != con); |
31b8006e SW |
1359 | con->in_msg = NULL; |
1360 | ||
1361 | /* if first message, set peer_name */ | |
1362 | if (con->peer_name.type == 0) | |
dbad185d | 1363 | con->peer_name = msg->hdr.src; |
31b8006e | 1364 | |
31b8006e | 1365 | con->in_seq++; |
ec302645 | 1366 | mutex_unlock(&con->mutex); |
31b8006e | 1367 | |
b77f8f0e | 1368 | dout("===== %p %llu from %s%lld %d=%s len %d+%d+%d (%u %u %u) =====\n", |
31b8006e | 1369 | msg, le64_to_cpu(msg->hdr.seq), |
dbad185d | 1370 | ENTITY_NAME(msg->hdr.src), |
31b8006e SW |
1371 | le16_to_cpu(msg->hdr.type), |
1372 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
1373 | le32_to_cpu(msg->hdr.front_len), | |
b77f8f0e | 1374 | le32_to_cpu(msg->hdr.middle_len), |
31b8006e SW |
1375 | le32_to_cpu(msg->hdr.data_len), |
1376 | con->in_front_crc, con->in_middle_crc, con->in_data_crc); | |
1377 | con->ops->dispatch(con, msg); | |
ec302645 SW |
1378 | |
1379 | mutex_lock(&con->mutex); | |
31b8006e SW |
1380 | } |
1381 | ||
31b8006e | 1382 | /* |
802c6d96 AE |
1383 | * Atomically queue work on a connection after the specified delay. |
1384 | * Bump @con reference to avoid races with connection teardown. | |
1385 | * Returns 0 if work was queued, or an error code otherwise. | |
31b8006e | 1386 | */ |
802c6d96 | 1387 | static int queue_con_delay(struct ceph_connection *con, unsigned long delay) |
31b8006e | 1388 | { |
31b8006e | 1389 | if (!con->ops->get(con)) { |
802c6d96 | 1390 | dout("%s %p ref count 0\n", __func__, con); |
802c6d96 | 1391 | return -ENOENT; |
31b8006e SW |
1392 | } |
1393 | ||
418af5b3 ID |
1394 | if (delay >= HZ) |
1395 | delay = round_jiffies_relative(delay); | |
1396 | ||
5a5036c8 | 1397 | dout("%s %p %lu\n", __func__, con, delay); |
802c6d96 AE |
1398 | if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) { |
1399 | dout("%s %p - already queued\n", __func__, con); | |
31b8006e | 1400 | con->ops->put(con); |
802c6d96 | 1401 | return -EBUSY; |
31b8006e | 1402 | } |
802c6d96 | 1403 | |
802c6d96 AE |
1404 | return 0; |
1405 | } | |
1406 | ||
1407 | static void queue_con(struct ceph_connection *con) | |
1408 | { | |
1409 | (void) queue_con_delay(con, 0); | |
31b8006e SW |
1410 | } |
1411 | ||
37ab77ac ID |
1412 | static void cancel_con(struct ceph_connection *con) |
1413 | { | |
1414 | if (cancel_delayed_work(&con->work)) { | |
1415 | dout("%s %p\n", __func__, con); | |
1416 | con->ops->put(con); | |
1417 | } | |
1418 | } | |
1419 | ||
7bb21d68 AE |
1420 | static bool con_sock_closed(struct ceph_connection *con) |
1421 | { | |
6503e0b6 | 1422 | if (!ceph_con_flag_test_and_clear(con, CEPH_CON_F_SOCK_CLOSED)) |
7bb21d68 AE |
1423 | return false; |
1424 | ||
1425 | #define CASE(x) \ | |
6d7f62bf | 1426 | case CEPH_CON_S_ ## x: \ |
7bb21d68 AE |
1427 | con->error_msg = "socket closed (con state " #x ")"; \ |
1428 | break; | |
1429 | ||
1430 | switch (con->state) { | |
1431 | CASE(CLOSED); | |
1432 | CASE(PREOPEN); | |
6d7f62bf ID |
1433 | CASE(V1_BANNER); |
1434 | CASE(V1_CONNECT_MSG); | |
cd1a677c ID |
1435 | CASE(V2_BANNER_PREFIX); |
1436 | CASE(V2_BANNER_PAYLOAD); | |
1437 | CASE(V2_HELLO); | |
1438 | CASE(V2_AUTH); | |
1439 | CASE(V2_AUTH_SIGNATURE); | |
1440 | CASE(V2_SESSION_CONNECT); | |
1441 | CASE(V2_SESSION_RECONNECT); | |
7bb21d68 AE |
1442 | CASE(OPEN); |
1443 | CASE(STANDBY); | |
1444 | default: | |
7bb21d68 | 1445 | BUG(); |
7bb21d68 AE |
1446 | } |
1447 | #undef CASE | |
1448 | ||
1449 | return true; | |
1450 | } | |
1451 | ||
f20a39fd AE |
1452 | static bool con_backoff(struct ceph_connection *con) |
1453 | { | |
1454 | int ret; | |
1455 | ||
6503e0b6 | 1456 | if (!ceph_con_flag_test_and_clear(con, CEPH_CON_F_BACKOFF)) |
f20a39fd AE |
1457 | return false; |
1458 | ||
418af5b3 | 1459 | ret = queue_con_delay(con, con->delay); |
f20a39fd AE |
1460 | if (ret) { |
1461 | dout("%s: con %p FAILED to back off %lu\n", __func__, | |
1462 | con, con->delay); | |
1463 | BUG_ON(ret == -ENOENT); | |
6503e0b6 | 1464 | ceph_con_flag_set(con, CEPH_CON_F_BACKOFF); |
f20a39fd AE |
1465 | } |
1466 | ||
1467 | return true; | |
1468 | } | |
1469 | ||
93209264 AE |
1470 | /* Finish fault handling; con->mutex must *not* be held here */ |
1471 | ||
1472 | static void con_fault_finish(struct ceph_connection *con) | |
1473 | { | |
f6330cc1 ID |
1474 | dout("%s %p\n", __func__, con); |
1475 | ||
93209264 AE |
1476 | /* |
1477 | * in case we faulted due to authentication, invalidate our | |
1478 | * current tickets so that we can get new ones. | |
1479 | */ | |
a56dd9bf ID |
1480 | if (con->v1.auth_retry) { |
1481 | dout("auth_retry %d, invalidating\n", con->v1.auth_retry); | |
f6330cc1 ID |
1482 | if (con->ops->invalidate_authorizer) |
1483 | con->ops->invalidate_authorizer(con); | |
a56dd9bf | 1484 | con->v1.auth_retry = 0; |
93209264 AE |
1485 | } |
1486 | ||
1487 | if (con->ops->fault) | |
1488 | con->ops->fault(con); | |
1489 | } | |
1490 | ||
31b8006e SW |
1491 | /* |
1492 | * Do some work on a connection. Drop a connection ref when we're done. | |
1493 | */ | |
68931622 | 1494 | static void ceph_con_workfn(struct work_struct *work) |
31b8006e SW |
1495 | { |
1496 | struct ceph_connection *con = container_of(work, struct ceph_connection, | |
1497 | work.work); | |
49659416 | 1498 | bool fault; |
31b8006e | 1499 | |
9dd4658d | 1500 | mutex_lock(&con->mutex); |
49659416 AE |
1501 | while (true) { |
1502 | int ret; | |
31b8006e | 1503 | |
49659416 AE |
1504 | if ((fault = con_sock_closed(con))) { |
1505 | dout("%s: con %p SOCK_CLOSED\n", __func__, con); | |
1506 | break; | |
1507 | } | |
1508 | if (con_backoff(con)) { | |
1509 | dout("%s: con %p BACKOFF\n", __func__, con); | |
1510 | break; | |
1511 | } | |
6d7f62bf | 1512 | if (con->state == CEPH_CON_S_STANDBY) { |
49659416 AE |
1513 | dout("%s: con %p STANDBY\n", __func__, con); |
1514 | break; | |
1515 | } | |
6d7f62bf | 1516 | if (con->state == CEPH_CON_S_CLOSED) { |
49659416 AE |
1517 | dout("%s: con %p CLOSED\n", __func__, con); |
1518 | BUG_ON(con->sock); | |
1519 | break; | |
1520 | } | |
6d7f62bf | 1521 | if (con->state == CEPH_CON_S_PREOPEN) { |
49659416 AE |
1522 | dout("%s: con %p PREOPEN\n", __func__, con); |
1523 | BUG_ON(con->sock); | |
1524 | } | |
0da5d703 | 1525 | |
cd1a677c ID |
1526 | if (ceph_msgr2(from_msgr(con->msgr))) |
1527 | ret = ceph_con_v2_try_read(con); | |
1528 | else | |
1529 | ret = ceph_con_v1_try_read(con); | |
49659416 AE |
1530 | if (ret < 0) { |
1531 | if (ret == -EAGAIN) | |
1532 | continue; | |
67c64eb7 ID |
1533 | if (!con->error_msg) |
1534 | con->error_msg = "socket error on read"; | |
49659416 AE |
1535 | fault = true; |
1536 | break; | |
1537 | } | |
1538 | ||
cd1a677c ID |
1539 | if (ceph_msgr2(from_msgr(con->msgr))) |
1540 | ret = ceph_con_v2_try_write(con); | |
1541 | else | |
1542 | ret = ceph_con_v1_try_write(con); | |
49659416 AE |
1543 | if (ret < 0) { |
1544 | if (ret == -EAGAIN) | |
1545 | continue; | |
67c64eb7 ID |
1546 | if (!con->error_msg) |
1547 | con->error_msg = "socket error on write"; | |
49659416 AE |
1548 | fault = true; |
1549 | } | |
1550 | ||
1551 | break; /* If we make it to here, we're done */ | |
3a140a0d | 1552 | } |
b6e7b6a1 AE |
1553 | if (fault) |
1554 | con_fault(con); | |
9dd4658d | 1555 | mutex_unlock(&con->mutex); |
0da5d703 | 1556 | |
b6e7b6a1 AE |
1557 | if (fault) |
1558 | con_fault_finish(con); | |
1559 | ||
1560 | con->ops->put(con); | |
31b8006e SW |
1561 | } |
1562 | ||
31b8006e SW |
1563 | /* |
1564 | * Generic error/fault handler. A retry mechanism is used with | |
1565 | * exponential backoff | |
1566 | */ | |
93209264 | 1567 | static void con_fault(struct ceph_connection *con) |
31b8006e | 1568 | { |
30be780a | 1569 | dout("fault %p state %d to peer %s\n", |
b726ec97 | 1570 | con, con->state, ceph_pr_addr(&con->peer_addr)); |
31b8006e | 1571 | |
67c64eb7 | 1572 | pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name), |
b726ec97 | 1573 | ceph_pr_addr(&con->peer_addr), con->error_msg); |
67c64eb7 ID |
1574 | con->error_msg = NULL; |
1575 | ||
cd1a677c ID |
1576 | WARN_ON(con->state == CEPH_CON_S_STANDBY || |
1577 | con->state == CEPH_CON_S_CLOSED); | |
ec302645 | 1578 | |
3596f4c1 | 1579 | ceph_con_reset_protocol(con); |
5e095e8b | 1580 | |
6503e0b6 | 1581 | if (ceph_con_flag_test(con, CEPH_CON_F_LOSSYTX)) { |
8dacc7da | 1582 | dout("fault on LOSSYTX channel, marking CLOSED\n"); |
6d7f62bf | 1583 | con->state = CEPH_CON_S_CLOSED; |
93209264 | 1584 | return; |
3b5ede07 SW |
1585 | } |
1586 | ||
e80a52d1 SW |
1587 | /* Requeue anything that hasn't been acked */ |
1588 | list_splice_init(&con->out_sent, &con->out_queue); | |
9bd2e6f8 | 1589 | |
e76661d0 SW |
1590 | /* If there are no messages queued or keepalive pending, place |
1591 | * the connection in a STANDBY state */ | |
1592 | if (list_empty(&con->out_queue) && | |
6503e0b6 | 1593 | !ceph_con_flag_test(con, CEPH_CON_F_KEEPALIVE_PENDING)) { |
e00de341 | 1594 | dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con); |
6503e0b6 | 1595 | ceph_con_flag_clear(con, CEPH_CON_F_WRITE_PENDING); |
6d7f62bf | 1596 | con->state = CEPH_CON_S_STANDBY; |
e80a52d1 SW |
1597 | } else { |
1598 | /* retry after a delay. */ | |
6d7f62bf | 1599 | con->state = CEPH_CON_S_PREOPEN; |
418af5b3 | 1600 | if (!con->delay) { |
e80a52d1 | 1601 | con->delay = BASE_DELAY_INTERVAL; |
418af5b3 | 1602 | } else if (con->delay < MAX_DELAY_INTERVAL) { |
e80a52d1 | 1603 | con->delay *= 2; |
418af5b3 ID |
1604 | if (con->delay > MAX_DELAY_INTERVAL) |
1605 | con->delay = MAX_DELAY_INTERVAL; | |
1606 | } | |
6503e0b6 | 1607 | ceph_con_flag_set(con, CEPH_CON_F_BACKOFF); |
8618e30b | 1608 | queue_con(con); |
31b8006e | 1609 | } |
31b8006e SW |
1610 | } |
1611 | ||
120a75ea YZ |
1612 | void ceph_messenger_reset_nonce(struct ceph_messenger *msgr) |
1613 | { | |
1614 | u32 nonce = le32_to_cpu(msgr->inst.addr.nonce) + 1000000; | |
1615 | msgr->inst.addr.nonce = cpu_to_le32(nonce); | |
6503e0b6 | 1616 | ceph_encode_my_addr(msgr); |
120a75ea | 1617 | } |
31b8006e SW |
1618 | |
1619 | /* | |
15d9882c | 1620 | * initialize a new messenger instance |
31b8006e | 1621 | */ |
15d9882c | 1622 | void ceph_messenger_init(struct ceph_messenger *msgr, |
859bff51 | 1623 | struct ceph_entity_addr *myaddr) |
31b8006e | 1624 | { |
31b8006e SW |
1625 | spin_lock_init(&msgr->global_seq_lock); |
1626 | ||
fd1a154c ID |
1627 | if (myaddr) { |
1628 | memcpy(&msgr->inst.addr.in_addr, &myaddr->in_addr, | |
1629 | sizeof(msgr->inst.addr.in_addr)); | |
6503e0b6 | 1630 | ceph_addr_set_port(&msgr->inst.addr, 0); |
fd1a154c | 1631 | } |
31b8006e | 1632 | |
cd1a677c ID |
1633 | /* |
1634 | * Since nautilus, clients are identified using type ANY. | |
1635 | * For msgr1, ceph_encode_banner_addr() munges it to NONE. | |
1636 | */ | |
1637 | msgr->inst.addr.type = CEPH_ENTITY_ADDR_TYPE_ANY; | |
fd1a154c ID |
1638 | |
1639 | /* generate a random non-zero nonce */ | |
1640 | do { | |
1641 | get_random_bytes(&msgr->inst.addr.nonce, | |
1642 | sizeof(msgr->inst.addr.nonce)); | |
1643 | } while (!msgr->inst.addr.nonce); | |
6503e0b6 | 1644 | ceph_encode_my_addr(msgr); |
31b8006e | 1645 | |
a2a32584 | 1646 | atomic_set(&msgr->stopping, 0); |
757856d2 | 1647 | write_pnet(&msgr->net, get_net(current->nsproxy->net_ns)); |
31b8006e | 1648 | |
15d9882c | 1649 | dout("%s %p\n", __func__, msgr); |
31b8006e SW |
1650 | } |
1651 | ||
757856d2 ID |
1652 | void ceph_messenger_fini(struct ceph_messenger *msgr) |
1653 | { | |
1654 | put_net(read_pnet(&msgr->net)); | |
1655 | } | |
757856d2 | 1656 | |
583d0fef ID |
1657 | static void msg_con_set(struct ceph_msg *msg, struct ceph_connection *con) |
1658 | { | |
1659 | if (msg->con) | |
1660 | msg->con->ops->put(msg->con); | |
1661 | ||
1662 | msg->con = con ? con->ops->get(con) : NULL; | |
1663 | BUG_ON(msg->con != con); | |
1664 | } | |
1665 | ||
e00de341 SW |
1666 | static void clear_standby(struct ceph_connection *con) |
1667 | { | |
1668 | /* come back from STANDBY? */ | |
6d7f62bf | 1669 | if (con->state == CEPH_CON_S_STANDBY) { |
e00de341 | 1670 | dout("clear_standby %p and ++connect_seq\n", con); |
6d7f62bf | 1671 | con->state = CEPH_CON_S_PREOPEN; |
a56dd9bf | 1672 | con->v1.connect_seq++; |
6503e0b6 ID |
1673 | WARN_ON(ceph_con_flag_test(con, CEPH_CON_F_WRITE_PENDING)); |
1674 | WARN_ON(ceph_con_flag_test(con, CEPH_CON_F_KEEPALIVE_PENDING)); | |
e00de341 SW |
1675 | } |
1676 | } | |
1677 | ||
31b8006e SW |
1678 | /* |
1679 | * Queue up an outgoing message on the given connection. | |
771294fe ID |
1680 | * |
1681 | * Consumes a ref on @msg. | |
31b8006e SW |
1682 | */ |
1683 | void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg) | |
1684 | { | |
31b8006e | 1685 | /* set src+dst */ |
dbad185d | 1686 | msg->hdr.src = con->msgr->inst.name; |
3ca02ef9 | 1687 | BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len)); |
e84346b7 SW |
1688 | msg->needs_out_seq = true; |
1689 | ||
ec302645 | 1690 | mutex_lock(&con->mutex); |
92ce034b | 1691 | |
6d7f62bf | 1692 | if (con->state == CEPH_CON_S_CLOSED) { |
a59b55a6 SW |
1693 | dout("con_send %p closed, dropping %p\n", con, msg); |
1694 | ceph_msg_put(msg); | |
1695 | mutex_unlock(&con->mutex); | |
1696 | return; | |
1697 | } | |
1698 | ||
583d0fef | 1699 | msg_con_set(msg, con); |
92ce034b | 1700 | |
31b8006e SW |
1701 | BUG_ON(!list_empty(&msg->list_head)); |
1702 | list_add_tail(&msg->list_head, &con->out_queue); | |
1703 | dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg, | |
1704 | ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type), | |
1705 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
1706 | le32_to_cpu(msg->hdr.front_len), | |
1707 | le32_to_cpu(msg->hdr.middle_len), | |
1708 | le32_to_cpu(msg->hdr.data_len)); | |
00650931 SW |
1709 | |
1710 | clear_standby(con); | |
ec302645 | 1711 | mutex_unlock(&con->mutex); |
31b8006e SW |
1712 | |
1713 | /* if there wasn't anything waiting to send before, queue | |
1714 | * new work */ | |
6503e0b6 | 1715 | if (!ceph_con_flag_test_and_set(con, CEPH_CON_F_WRITE_PENDING)) |
31b8006e SW |
1716 | queue_con(con); |
1717 | } | |
3d14c5d2 | 1718 | EXPORT_SYMBOL(ceph_con_send); |
31b8006e SW |
1719 | |
1720 | /* | |
1721 | * Revoke a message that was previously queued for send | |
1722 | */ | |
6740a845 | 1723 | void ceph_msg_revoke(struct ceph_msg *msg) |
31b8006e | 1724 | { |
6740a845 AE |
1725 | struct ceph_connection *con = msg->con; |
1726 | ||
583d0fef ID |
1727 | if (!con) { |
1728 | dout("%s msg %p null con\n", __func__, msg); | |
6740a845 | 1729 | return; /* Message not in our possession */ |
583d0fef | 1730 | } |
6740a845 | 1731 | |
ec302645 | 1732 | mutex_lock(&con->mutex); |
566050e1 ID |
1733 | if (list_empty(&msg->list_head)) { |
1734 | WARN_ON(con->out_msg == msg); | |
1735 | dout("%s con %p msg %p not linked\n", __func__, con, msg); | |
1736 | mutex_unlock(&con->mutex); | |
1737 | return; | |
ed98adad | 1738 | } |
67645d76 | 1739 | |
566050e1 ID |
1740 | dout("%s con %p msg %p was linked\n", __func__, con, msg); |
1741 | msg->hdr.seq = 0; | |
1742 | ceph_msg_remove(msg); | |
1743 | ||
1744 | if (con->out_msg == msg) { | |
1745 | WARN_ON(con->state != CEPH_CON_S_OPEN); | |
1746 | dout("%s con %p msg %p was sending\n", __func__, con, msg); | |
cd1a677c ID |
1747 | if (ceph_msgr2(from_msgr(con->msgr))) |
1748 | ceph_con_v2_revoke(con); | |
1749 | else | |
1750 | ceph_con_v1_revoke(con); | |
566050e1 | 1751 | ceph_msg_put(con->out_msg); |
67645d76 | 1752 | con->out_msg = NULL; |
566050e1 ID |
1753 | } else { |
1754 | dout("%s con %p msg %p not current, out_msg %p\n", __func__, | |
1755 | con, msg, con->out_msg); | |
31b8006e | 1756 | } |
ec302645 | 1757 | mutex_unlock(&con->mutex); |
31b8006e SW |
1758 | } |
1759 | ||
350b1c32 | 1760 | /* |
0d59ab81 | 1761 | * Revoke a message that we may be reading data into |
350b1c32 | 1762 | */ |
8921d114 | 1763 | void ceph_msg_revoke_incoming(struct ceph_msg *msg) |
350b1c32 | 1764 | { |
583d0fef | 1765 | struct ceph_connection *con = msg->con; |
8921d114 | 1766 | |
583d0fef | 1767 | if (!con) { |
8921d114 | 1768 | dout("%s msg %p null con\n", __func__, msg); |
8921d114 AE |
1769 | return; /* Message not in our possession */ |
1770 | } | |
1771 | ||
350b1c32 | 1772 | mutex_lock(&con->mutex); |
8921d114 | 1773 | if (con->in_msg == msg) { |
566050e1 ID |
1774 | WARN_ON(con->state != CEPH_CON_S_OPEN); |
1775 | dout("%s con %p msg %p was recving\n", __func__, con, msg); | |
cd1a677c ID |
1776 | if (ceph_msgr2(from_msgr(con->msgr))) |
1777 | ceph_con_v2_revoke_incoming(con); | |
1778 | else | |
1779 | ceph_con_v1_revoke_incoming(con); | |
350b1c32 SW |
1780 | ceph_msg_put(con->in_msg); |
1781 | con->in_msg = NULL; | |
350b1c32 | 1782 | } else { |
566050e1 ID |
1783 | dout("%s con %p msg %p not current, in_msg %p\n", __func__, |
1784 | con, msg, con->in_msg); | |
350b1c32 SW |
1785 | } |
1786 | mutex_unlock(&con->mutex); | |
1787 | } | |
1788 | ||
31b8006e SW |
1789 | /* |
1790 | * Queue a keepalive byte to ensure the tcp connection is alive. | |
1791 | */ | |
1792 | void ceph_con_keepalive(struct ceph_connection *con) | |
1793 | { | |
e00de341 | 1794 | dout("con_keepalive %p\n", con); |
00650931 | 1795 | mutex_lock(&con->mutex); |
e00de341 | 1796 | clear_standby(con); |
6503e0b6 | 1797 | ceph_con_flag_set(con, CEPH_CON_F_KEEPALIVE_PENDING); |
00650931 | 1798 | mutex_unlock(&con->mutex); |
4aac9228 | 1799 | |
6503e0b6 | 1800 | if (!ceph_con_flag_test_and_set(con, CEPH_CON_F_WRITE_PENDING)) |
31b8006e SW |
1801 | queue_con(con); |
1802 | } | |
3d14c5d2 | 1803 | EXPORT_SYMBOL(ceph_con_keepalive); |
31b8006e | 1804 | |
8b9558aa YZ |
1805 | bool ceph_con_keepalive_expired(struct ceph_connection *con, |
1806 | unsigned long interval) | |
1807 | { | |
1808 | if (interval > 0 && | |
1809 | (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) { | |
473bd2d7 AB |
1810 | struct timespec64 now; |
1811 | struct timespec64 ts; | |
1812 | ktime_get_real_ts64(&now); | |
1813 | jiffies_to_timespec64(interval, &ts); | |
1814 | ts = timespec64_add(con->last_keepalive_ack, ts); | |
1815 | return timespec64_compare(&now, &ts) >= 0; | |
8b9558aa YZ |
1816 | } |
1817 | return false; | |
1818 | } | |
1819 | ||
0d9c1ab3 | 1820 | static struct ceph_msg_data *ceph_msg_data_add(struct ceph_msg *msg) |
43794509 | 1821 | { |
0d9c1ab3 ID |
1822 | BUG_ON(msg->num_data_items >= msg->max_data_items); |
1823 | return &msg->data[msg->num_data_items++]; | |
6644ed7b AE |
1824 | } |
1825 | ||
1826 | static void ceph_msg_data_destroy(struct ceph_msg_data *data) | |
1827 | { | |
e8862740 ID |
1828 | if (data->type == CEPH_MSG_DATA_PAGES && data->own_pages) { |
1829 | int num_pages = calc_pages_for(data->alignment, data->length); | |
1830 | ceph_release_page_vector(data->pages, num_pages); | |
1831 | } else if (data->type == CEPH_MSG_DATA_PAGELIST) { | |
6644ed7b | 1832 | ceph_pagelist_release(data->pagelist); |
e8862740 | 1833 | } |
43794509 AE |
1834 | } |
1835 | ||
90af3602 | 1836 | void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages, |
e8862740 | 1837 | size_t length, size_t alignment, bool own_pages) |
02afca6c | 1838 | { |
6644ed7b AE |
1839 | struct ceph_msg_data *data; |
1840 | ||
07aa1558 AE |
1841 | BUG_ON(!pages); |
1842 | BUG_ON(!length); | |
6644ed7b | 1843 | |
0d9c1ab3 ID |
1844 | data = ceph_msg_data_add(msg); |
1845 | data->type = CEPH_MSG_DATA_PAGES; | |
6644ed7b AE |
1846 | data->pages = pages; |
1847 | data->length = length; | |
1848 | data->alignment = alignment & ~PAGE_MASK; | |
e8862740 | 1849 | data->own_pages = own_pages; |
02afca6c | 1850 | |
5240d9f9 | 1851 | msg->data_length += length; |
02afca6c | 1852 | } |
90af3602 | 1853 | EXPORT_SYMBOL(ceph_msg_data_add_pages); |
31b8006e | 1854 | |
90af3602 | 1855 | void ceph_msg_data_add_pagelist(struct ceph_msg *msg, |
27fa8385 AE |
1856 | struct ceph_pagelist *pagelist) |
1857 | { | |
6644ed7b AE |
1858 | struct ceph_msg_data *data; |
1859 | ||
07aa1558 AE |
1860 | BUG_ON(!pagelist); |
1861 | BUG_ON(!pagelist->length); | |
27fa8385 | 1862 | |
0d9c1ab3 ID |
1863 | data = ceph_msg_data_add(msg); |
1864 | data->type = CEPH_MSG_DATA_PAGELIST; | |
89486833 | 1865 | refcount_inc(&pagelist->refcnt); |
6644ed7b AE |
1866 | data->pagelist = pagelist; |
1867 | ||
5240d9f9 | 1868 | msg->data_length += pagelist->length; |
27fa8385 | 1869 | } |
90af3602 | 1870 | EXPORT_SYMBOL(ceph_msg_data_add_pagelist); |
27fa8385 | 1871 | |
ea96571f | 1872 | #ifdef CONFIG_BLOCK |
5359a17d ID |
1873 | void ceph_msg_data_add_bio(struct ceph_msg *msg, struct ceph_bio_iter *bio_pos, |
1874 | u32 length) | |
27fa8385 | 1875 | { |
6644ed7b AE |
1876 | struct ceph_msg_data *data; |
1877 | ||
0d9c1ab3 ID |
1878 | data = ceph_msg_data_add(msg); |
1879 | data->type = CEPH_MSG_DATA_BIO; | |
5359a17d | 1880 | data->bio_pos = *bio_pos; |
c851c495 | 1881 | data->bio_length = length; |
6644ed7b | 1882 | |
5240d9f9 | 1883 | msg->data_length += length; |
27fa8385 | 1884 | } |
90af3602 | 1885 | EXPORT_SYMBOL(ceph_msg_data_add_bio); |
ea96571f | 1886 | #endif /* CONFIG_BLOCK */ |
27fa8385 | 1887 | |
b9e281c2 ID |
1888 | void ceph_msg_data_add_bvecs(struct ceph_msg *msg, |
1889 | struct ceph_bvec_iter *bvec_pos) | |
1890 | { | |
1891 | struct ceph_msg_data *data; | |
1892 | ||
0d9c1ab3 ID |
1893 | data = ceph_msg_data_add(msg); |
1894 | data->type = CEPH_MSG_DATA_BVECS; | |
b9e281c2 ID |
1895 | data->bvec_pos = *bvec_pos; |
1896 | ||
b9e281c2 ID |
1897 | msg->data_length += bvec_pos->iter.bi_size; |
1898 | } | |
1899 | EXPORT_SYMBOL(ceph_msg_data_add_bvecs); | |
1900 | ||
31b8006e SW |
1901 | /* |
1902 | * construct a new message with given type, size | |
1903 | * the new msg has a ref count of 1. | |
1904 | */ | |
0d9c1ab3 ID |
1905 | struct ceph_msg *ceph_msg_new2(int type, int front_len, int max_data_items, |
1906 | gfp_t flags, bool can_fail) | |
31b8006e SW |
1907 | { |
1908 | struct ceph_msg *m; | |
1909 | ||
e3d5d638 | 1910 | m = kmem_cache_zalloc(ceph_msg_cache, flags); |
31b8006e SW |
1911 | if (m == NULL) |
1912 | goto out; | |
31b8006e SW |
1913 | |
1914 | m->hdr.type = cpu_to_le16(type); | |
45c6ceb5 | 1915 | m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT); |
31b8006e | 1916 | m->hdr.front_len = cpu_to_le32(front_len); |
ca20892d | 1917 | |
9516e45b AE |
1918 | INIT_LIST_HEAD(&m->list_head); |
1919 | kref_init(&m->kref); | |
ca20892d | 1920 | |
31b8006e SW |
1921 | /* front */ |
1922 | if (front_len) { | |
eeb0bed5 | 1923 | m->front.iov_base = ceph_kvmalloc(front_len, flags); |
31b8006e | 1924 | if (m->front.iov_base == NULL) { |
b61c2763 | 1925 | dout("ceph_msg_new can't allocate %d bytes\n", |
31b8006e SW |
1926 | front_len); |
1927 | goto out2; | |
1928 | } | |
1929 | } else { | |
1930 | m->front.iov_base = NULL; | |
1931 | } | |
f2be82b0 | 1932 | m->front_alloc_len = m->front.iov_len = front_len; |
31b8006e | 1933 | |
0d9c1ab3 ID |
1934 | if (max_data_items) { |
1935 | m->data = kmalloc_array(max_data_items, sizeof(*m->data), | |
1936 | flags); | |
1937 | if (!m->data) | |
1938 | goto out2; | |
1939 | ||
1940 | m->max_data_items = max_data_items; | |
1941 | } | |
1942 | ||
bb257664 | 1943 | dout("ceph_msg_new %p front %d\n", m, front_len); |
31b8006e SW |
1944 | return m; |
1945 | ||
1946 | out2: | |
1947 | ceph_msg_put(m); | |
1948 | out: | |
b61c2763 SW |
1949 | if (!can_fail) { |
1950 | pr_err("msg_new can't create type %d front %d\n", type, | |
1951 | front_len); | |
f0ed1b7c | 1952 | WARN_ON(1); |
b61c2763 SW |
1953 | } else { |
1954 | dout("msg_new can't create type %d front %d\n", type, | |
1955 | front_len); | |
1956 | } | |
a79832f2 | 1957 | return NULL; |
31b8006e | 1958 | } |
0d9c1ab3 ID |
1959 | EXPORT_SYMBOL(ceph_msg_new2); |
1960 | ||
1961 | struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags, | |
1962 | bool can_fail) | |
1963 | { | |
1964 | return ceph_msg_new2(type, front_len, 0, flags, can_fail); | |
1965 | } | |
3d14c5d2 | 1966 | EXPORT_SYMBOL(ceph_msg_new); |
31b8006e | 1967 | |
31b8006e SW |
1968 | /* |
1969 | * Allocate "middle" portion of a message, if it is needed and wasn't | |
1970 | * allocated by alloc_msg. This allows us to read a small fixed-size | |
1971 | * per-type header in the front and then gracefully fail (i.e., | |
1972 | * propagate the error to the caller based on info in the front) when | |
1973 | * the middle is too large. | |
1974 | */ | |
2450418c | 1975 | static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg) |
31b8006e SW |
1976 | { |
1977 | int type = le16_to_cpu(msg->hdr.type); | |
1978 | int middle_len = le32_to_cpu(msg->hdr.middle_len); | |
1979 | ||
1980 | dout("alloc_middle %p type %d %s middle_len %d\n", msg, type, | |
1981 | ceph_msg_type_name(type), middle_len); | |
1982 | BUG_ON(!middle_len); | |
1983 | BUG_ON(msg->middle); | |
1984 | ||
b6c1d5b8 | 1985 | msg->middle = ceph_buffer_new(middle_len, GFP_NOFS); |
31b8006e SW |
1986 | if (!msg->middle) |
1987 | return -ENOMEM; | |
1988 | return 0; | |
1989 | } | |
1990 | ||
2450418c | 1991 | /* |
1c20f2d2 AE |
1992 | * Allocate a message for receiving an incoming message on a |
1993 | * connection, and save the result in con->in_msg. Uses the | |
1994 | * connection's private alloc_msg op if available. | |
1995 | * | |
4740a623 SW |
1996 | * Returns 0 on success, or a negative error code. |
1997 | * | |
1998 | * On success, if we set *skip = 1: | |
1999 | * - the next message should be skipped and ignored. | |
2000 | * - con->in_msg == NULL | |
2001 | * or if we set *skip = 0: | |
2002 | * - con->in_msg is non-null. | |
2003 | * On error (ENOMEM, EAGAIN, ...), | |
2004 | * - con->in_msg == NULL | |
2450418c | 2005 | */ |
6503e0b6 ID |
2006 | int ceph_con_in_msg_alloc(struct ceph_connection *con, |
2007 | struct ceph_msg_header *hdr, int *skip) | |
2450418c | 2008 | { |
2450418c | 2009 | int middle_len = le32_to_cpu(hdr->middle_len); |
1d866d1c | 2010 | struct ceph_msg *msg; |
4740a623 | 2011 | int ret = 0; |
2450418c | 2012 | |
1c20f2d2 | 2013 | BUG_ON(con->in_msg != NULL); |
53ded495 | 2014 | BUG_ON(!con->ops->alloc_msg); |
2450418c | 2015 | |
53ded495 AE |
2016 | mutex_unlock(&con->mutex); |
2017 | msg = con->ops->alloc_msg(con, hdr, skip); | |
2018 | mutex_lock(&con->mutex); | |
6d7f62bf | 2019 | if (con->state != CEPH_CON_S_OPEN) { |
53ded495 | 2020 | if (msg) |
1d866d1c | 2021 | ceph_msg_put(msg); |
53ded495 AE |
2022 | return -EAGAIN; |
2023 | } | |
4137577a AE |
2024 | if (msg) { |
2025 | BUG_ON(*skip); | |
583d0fef | 2026 | msg_con_set(msg, con); |
4137577a | 2027 | con->in_msg = msg; |
4137577a AE |
2028 | } else { |
2029 | /* | |
2030 | * Null message pointer means either we should skip | |
2031 | * this message or we couldn't allocate memory. The | |
2032 | * former is not an error. | |
2033 | */ | |
2034 | if (*skip) | |
2035 | return 0; | |
4137577a | 2036 | |
67c64eb7 | 2037 | con->error_msg = "error allocating memory for incoming message"; |
53ded495 | 2038 | return -ENOMEM; |
2450418c | 2039 | } |
fc4c128e | 2040 | memcpy(&con->in_msg->hdr, hdr, sizeof(*hdr)); |
2450418c | 2041 | |
1c20f2d2 AE |
2042 | if (middle_len && !con->in_msg->middle) { |
2043 | ret = ceph_alloc_middle(con, con->in_msg); | |
2450418c | 2044 | if (ret < 0) { |
1c20f2d2 AE |
2045 | ceph_msg_put(con->in_msg); |
2046 | con->in_msg = NULL; | |
2450418c YS |
2047 | } |
2048 | } | |
9d7f0f13 | 2049 | |
4740a623 | 2050 | return ret; |
2450418c YS |
2051 | } |
2052 | ||
6503e0b6 | 2053 | void ceph_con_get_out_msg(struct ceph_connection *con) |
771294fe ID |
2054 | { |
2055 | struct ceph_msg *msg; | |
2056 | ||
2057 | BUG_ON(list_empty(&con->out_queue)); | |
2058 | msg = list_first_entry(&con->out_queue, struct ceph_msg, list_head); | |
2059 | WARN_ON(msg->con != con); | |
2060 | ||
2061 | /* | |
2062 | * Put the message on "sent" list using a ref from ceph_con_send(). | |
2063 | * It is put when the message is acked or revoked. | |
2064 | */ | |
2065 | list_move_tail(&msg->list_head, &con->out_sent); | |
2066 | ||
2067 | /* | |
2068 | * Only assign outgoing seq # if we haven't sent this message | |
2069 | * yet. If it is requeued, resend with it's original seq. | |
2070 | */ | |
2071 | if (msg->needs_out_seq) { | |
2072 | msg->hdr.seq = cpu_to_le64(++con->out_seq); | |
2073 | msg->needs_out_seq = false; | |
2074 | ||
2075 | if (con->ops->reencode_message) | |
2076 | con->ops->reencode_message(msg); | |
2077 | } | |
2078 | ||
2079 | /* | |
2080 | * Get a ref for out_msg. It is put when we are done sending the | |
2081 | * message or in case of a fault. | |
2082 | */ | |
2083 | WARN_ON(con->out_msg); | |
2084 | con->out_msg = ceph_msg_get(msg); | |
2085 | } | |
31b8006e SW |
2086 | |
2087 | /* | |
2088 | * Free a generically kmalloc'd message. | |
2089 | */ | |
0215e44b | 2090 | static void ceph_msg_free(struct ceph_msg *m) |
31b8006e | 2091 | { |
0215e44b | 2092 | dout("%s %p\n", __func__, m); |
4965fc38 | 2093 | kvfree(m->front.iov_base); |
0d9c1ab3 | 2094 | kfree(m->data); |
e3d5d638 | 2095 | kmem_cache_free(ceph_msg_cache, m); |
31b8006e SW |
2096 | } |
2097 | ||
0215e44b | 2098 | static void ceph_msg_release(struct kref *kref) |
c2e552e7 SW |
2099 | { |
2100 | struct ceph_msg *m = container_of(kref, struct ceph_msg, kref); | |
0d9c1ab3 | 2101 | int i; |
31b8006e | 2102 | |
0215e44b | 2103 | dout("%s %p\n", __func__, m); |
c2e552e7 SW |
2104 | WARN_ON(!list_empty(&m->list_head)); |
2105 | ||
583d0fef ID |
2106 | msg_con_set(m, NULL); |
2107 | ||
c2e552e7 SW |
2108 | /* drop middle, data, if any */ |
2109 | if (m->middle) { | |
2110 | ceph_buffer_put(m->middle); | |
2111 | m->middle = NULL; | |
31b8006e | 2112 | } |
5240d9f9 | 2113 | |
0d9c1ab3 ID |
2114 | for (i = 0; i < m->num_data_items; i++) |
2115 | ceph_msg_data_destroy(&m->data[i]); | |
58bb3b37 | 2116 | |
c2e552e7 SW |
2117 | if (m->pool) |
2118 | ceph_msgpool_put(m->pool, m); | |
2119 | else | |
0215e44b ID |
2120 | ceph_msg_free(m); |
2121 | } | |
2122 | ||
2123 | struct ceph_msg *ceph_msg_get(struct ceph_msg *msg) | |
2124 | { | |
2125 | dout("%s %p (was %d)\n", __func__, msg, | |
2c935bc5 | 2126 | kref_read(&msg->kref)); |
0215e44b ID |
2127 | kref_get(&msg->kref); |
2128 | return msg; | |
2129 | } | |
2130 | EXPORT_SYMBOL(ceph_msg_get); | |
2131 | ||
2132 | void ceph_msg_put(struct ceph_msg *msg) | |
2133 | { | |
2134 | dout("%s %p (was %d)\n", __func__, msg, | |
2c935bc5 | 2135 | kref_read(&msg->kref)); |
0215e44b | 2136 | kref_put(&msg->kref, ceph_msg_release); |
31b8006e | 2137 | } |
0215e44b | 2138 | EXPORT_SYMBOL(ceph_msg_put); |
9ec7cab1 SW |
2139 | |
2140 | void ceph_msg_dump(struct ceph_msg *msg) | |
2141 | { | |
3cea4c30 ID |
2142 | pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg, |
2143 | msg->front_alloc_len, msg->data_length); | |
9ec7cab1 SW |
2144 | print_hex_dump(KERN_DEBUG, "header: ", |
2145 | DUMP_PREFIX_OFFSET, 16, 1, | |
2146 | &msg->hdr, sizeof(msg->hdr), true); | |
2147 | print_hex_dump(KERN_DEBUG, " front: ", | |
2148 | DUMP_PREFIX_OFFSET, 16, 1, | |
2149 | msg->front.iov_base, msg->front.iov_len, true); | |
2150 | if (msg->middle) | |
2151 | print_hex_dump(KERN_DEBUG, "middle: ", | |
2152 | DUMP_PREFIX_OFFSET, 16, 1, | |
2153 | msg->middle->vec.iov_base, | |
2154 | msg->middle->vec.iov_len, true); | |
2155 | print_hex_dump(KERN_DEBUG, "footer: ", | |
2156 | DUMP_PREFIX_OFFSET, 16, 1, | |
2157 | &msg->footer, sizeof(msg->footer), true); | |
2158 | } | |
3d14c5d2 | 2159 | EXPORT_SYMBOL(ceph_msg_dump); |