]>
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 | ||
8dacc7da SW |
85 | /* |
86 | * connection states | |
87 | */ | |
88 | #define CON_STATE_CLOSED 1 /* -> PREOPEN */ | |
89 | #define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */ | |
90 | #define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */ | |
91 | #define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */ | |
92 | #define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */ | |
93 | #define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */ | |
94 | ||
4a861692 SW |
95 | /* |
96 | * ceph_connection flag bits | |
97 | */ | |
98 | #define CON_FLAG_LOSSYTX 0 /* we can close channel or drop | |
99 | * messages on errors */ | |
100 | #define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */ | |
101 | #define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */ | |
102 | #define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */ | |
103 | #define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */ | |
8dacc7da | 104 | |
c9ffc77a AE |
105 | static bool con_flag_valid(unsigned long con_flag) |
106 | { | |
107 | switch (con_flag) { | |
108 | case CON_FLAG_LOSSYTX: | |
109 | case CON_FLAG_KEEPALIVE_PENDING: | |
110 | case CON_FLAG_WRITE_PENDING: | |
111 | case CON_FLAG_SOCK_CLOSED: | |
112 | case CON_FLAG_BACKOFF: | |
113 | return true; | |
114 | default: | |
115 | return false; | |
116 | } | |
117 | } | |
118 | ||
119 | static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag) | |
120 | { | |
121 | BUG_ON(!con_flag_valid(con_flag)); | |
122 | ||
123 | clear_bit(con_flag, &con->flags); | |
124 | } | |
125 | ||
126 | static void con_flag_set(struct ceph_connection *con, unsigned long con_flag) | |
127 | { | |
128 | BUG_ON(!con_flag_valid(con_flag)); | |
129 | ||
130 | set_bit(con_flag, &con->flags); | |
131 | } | |
132 | ||
133 | static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag) | |
134 | { | |
135 | BUG_ON(!con_flag_valid(con_flag)); | |
136 | ||
137 | return test_bit(con_flag, &con->flags); | |
138 | } | |
139 | ||
140 | static bool con_flag_test_and_clear(struct ceph_connection *con, | |
141 | unsigned long con_flag) | |
142 | { | |
143 | BUG_ON(!con_flag_valid(con_flag)); | |
144 | ||
145 | return test_and_clear_bit(con_flag, &con->flags); | |
146 | } | |
147 | ||
148 | static bool con_flag_test_and_set(struct ceph_connection *con, | |
149 | unsigned long con_flag) | |
150 | { | |
151 | BUG_ON(!con_flag_valid(con_flag)); | |
152 | ||
153 | return test_and_set_bit(con_flag, &con->flags); | |
154 | } | |
155 | ||
e3d5d638 AE |
156 | /* Slab caches for frequently-allocated structures */ |
157 | ||
158 | static struct kmem_cache *ceph_msg_cache; | |
159 | ||
31b8006e SW |
160 | /* static tag bytes (protocol control messages) */ |
161 | static char tag_msg = CEPH_MSGR_TAG_MSG; | |
162 | static char tag_ack = CEPH_MSGR_TAG_ACK; | |
163 | static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE; | |
8b9558aa | 164 | static char tag_keepalive2 = CEPH_MSGR_TAG_KEEPALIVE2; |
31b8006e | 165 | |
a6a5349d SW |
166 | #ifdef CONFIG_LOCKDEP |
167 | static struct lock_class_key socket_class; | |
168 | #endif | |
169 | ||
31b8006e | 170 | static void queue_con(struct ceph_connection *con); |
37ab77ac | 171 | static void cancel_con(struct ceph_connection *con); |
68931622 | 172 | static void ceph_con_workfn(struct work_struct *); |
93209264 | 173 | static void con_fault(struct ceph_connection *con); |
31b8006e | 174 | |
31b8006e | 175 | /* |
f64a9317 AE |
176 | * Nicely render a sockaddr as a string. An array of formatted |
177 | * strings is used, to approximate reentrancy. | |
31b8006e | 178 | */ |
f64a9317 AE |
179 | #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */ |
180 | #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG) | |
181 | #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1) | |
182 | #define MAX_ADDR_STR_LEN 64 /* 54 is enough */ | |
183 | ||
184 | static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN]; | |
185 | static atomic_t addr_str_seq = ATOMIC_INIT(0); | |
31b8006e | 186 | |
57666519 | 187 | static struct page *zero_page; /* used in certain error cases */ |
57666519 | 188 | |
b726ec97 | 189 | const char *ceph_pr_addr(const struct ceph_entity_addr *addr) |
31b8006e SW |
190 | { |
191 | int i; | |
192 | char *s; | |
b726ec97 JL |
193 | struct sockaddr_storage ss = addr->in_addr; /* align */ |
194 | struct sockaddr_in *in4 = (struct sockaddr_in *)&ss; | |
195 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *)&ss; | |
31b8006e | 196 | |
f64a9317 | 197 | i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK; |
31b8006e SW |
198 | s = addr_str[i]; |
199 | ||
b726ec97 | 200 | switch (ss.ss_family) { |
31b8006e | 201 | case AF_INET: |
d3c3c0a8 JL |
202 | snprintf(s, MAX_ADDR_STR_LEN, "(%d)%pI4:%hu", |
203 | le32_to_cpu(addr->type), &in4->sin_addr, | |
bd406145 | 204 | ntohs(in4->sin_port)); |
31b8006e SW |
205 | break; |
206 | ||
207 | case AF_INET6: | |
d3c3c0a8 JL |
208 | snprintf(s, MAX_ADDR_STR_LEN, "(%d)[%pI6c]:%hu", |
209 | le32_to_cpu(addr->type), &in6->sin6_addr, | |
bd406145 | 210 | ntohs(in6->sin6_port)); |
31b8006e SW |
211 | break; |
212 | ||
213 | default: | |
d3002b97 | 214 | snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)", |
b726ec97 | 215 | ss.ss_family); |
31b8006e SW |
216 | } |
217 | ||
218 | return s; | |
219 | } | |
3d14c5d2 | 220 | EXPORT_SYMBOL(ceph_pr_addr); |
31b8006e | 221 | |
63f2d211 SW |
222 | static void encode_my_addr(struct ceph_messenger *msgr) |
223 | { | |
224 | memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr)); | |
2c66de56 | 225 | ceph_encode_banner_addr(&msgr->my_enc_addr); |
63f2d211 SW |
226 | } |
227 | ||
31b8006e SW |
228 | /* |
229 | * work queue for all reading and writing to/from the socket. | |
230 | */ | |
e0f43c94 | 231 | static struct workqueue_struct *ceph_msgr_wq; |
31b8006e | 232 | |
e3d5d638 AE |
233 | static int ceph_msgr_slab_init(void) |
234 | { | |
235 | BUG_ON(ceph_msg_cache); | |
5ee61e95 | 236 | ceph_msg_cache = KMEM_CACHE(ceph_msg, 0); |
81b36be4 AE |
237 | if (!ceph_msg_cache) |
238 | return -ENOMEM; | |
239 | ||
0d9c1ab3 | 240 | return 0; |
e3d5d638 AE |
241 | } |
242 | ||
243 | static void ceph_msgr_slab_exit(void) | |
244 | { | |
245 | BUG_ON(!ceph_msg_cache); | |
246 | kmem_cache_destroy(ceph_msg_cache); | |
247 | ceph_msg_cache = NULL; | |
248 | } | |
249 | ||
15417167 | 250 | static void _ceph_msgr_exit(void) |
6173d1f0 | 251 | { |
d3002b97 | 252 | if (ceph_msgr_wq) { |
6173d1f0 | 253 | destroy_workqueue(ceph_msgr_wq); |
d3002b97 AE |
254 | ceph_msgr_wq = NULL; |
255 | } | |
6173d1f0 | 256 | |
6173d1f0 | 257 | BUG_ON(zero_page == NULL); |
09cbfeaf | 258 | put_page(zero_page); |
6173d1f0 | 259 | zero_page = NULL; |
d920ff6f BC |
260 | |
261 | ceph_msgr_slab_exit(); | |
6173d1f0 AE |
262 | } |
263 | ||
57a35dfb | 264 | int __init ceph_msgr_init(void) |
31b8006e | 265 | { |
d920ff6f BC |
266 | if (ceph_msgr_slab_init()) |
267 | return -ENOMEM; | |
268 | ||
57666519 AE |
269 | BUG_ON(zero_page != NULL); |
270 | zero_page = ZERO_PAGE(0); | |
09cbfeaf | 271 | get_page(zero_page); |
57666519 | 272 | |
f9865f06 ID |
273 | /* |
274 | * The number of active work items is limited by the number of | |
275 | * connections, so leave @max_active at default. | |
276 | */ | |
277 | ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0); | |
6173d1f0 AE |
278 | if (ceph_msgr_wq) |
279 | return 0; | |
57666519 | 280 | |
6173d1f0 AE |
281 | pr_err("msgr_init failed to create workqueue\n"); |
282 | _ceph_msgr_exit(); | |
57666519 | 283 | |
6173d1f0 | 284 | return -ENOMEM; |
31b8006e SW |
285 | } |
286 | ||
287 | void ceph_msgr_exit(void) | |
288 | { | |
57666519 | 289 | BUG_ON(ceph_msgr_wq == NULL); |
57666519 | 290 | |
6173d1f0 | 291 | _ceph_msgr_exit(); |
31b8006e SW |
292 | } |
293 | ||
cd84db6e | 294 | void ceph_msgr_flush(void) |
a922d38f SW |
295 | { |
296 | flush_workqueue(ceph_msgr_wq); | |
297 | } | |
3d14c5d2 | 298 | EXPORT_SYMBOL(ceph_msgr_flush); |
a922d38f | 299 | |
ce2c8903 AE |
300 | /* Connection socket state transition functions */ |
301 | ||
302 | static void con_sock_state_init(struct ceph_connection *con) | |
303 | { | |
304 | int old_state; | |
305 | ||
306 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED); | |
307 | if (WARN_ON(old_state != CON_SOCK_STATE_NEW)) | |
308 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
309 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
310 | CON_SOCK_STATE_CLOSED); | |
ce2c8903 AE |
311 | } |
312 | ||
313 | static void con_sock_state_connecting(struct ceph_connection *con) | |
314 | { | |
315 | int old_state; | |
316 | ||
317 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING); | |
318 | if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED)) | |
319 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
320 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
321 | CON_SOCK_STATE_CONNECTING); | |
ce2c8903 AE |
322 | } |
323 | ||
324 | static void con_sock_state_connected(struct ceph_connection *con) | |
325 | { | |
326 | int old_state; | |
327 | ||
328 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED); | |
329 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING)) | |
330 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
331 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
332 | CON_SOCK_STATE_CONNECTED); | |
ce2c8903 AE |
333 | } |
334 | ||
335 | static void con_sock_state_closing(struct ceph_connection *con) | |
336 | { | |
337 | int old_state; | |
338 | ||
339 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING); | |
340 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING && | |
341 | old_state != CON_SOCK_STATE_CONNECTED && | |
342 | old_state != CON_SOCK_STATE_CLOSING)) | |
343 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
344 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
345 | CON_SOCK_STATE_CLOSING); | |
ce2c8903 AE |
346 | } |
347 | ||
348 | static void con_sock_state_closed(struct ceph_connection *con) | |
349 | { | |
350 | int old_state; | |
351 | ||
352 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED); | |
353 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED && | |
fbb85a47 | 354 | old_state != CON_SOCK_STATE_CLOSING && |
8007b8d6 SW |
355 | old_state != CON_SOCK_STATE_CONNECTING && |
356 | old_state != CON_SOCK_STATE_CLOSED)) | |
ce2c8903 | 357 | printk("%s: unexpected old state %d\n", __func__, old_state); |
8007b8d6 SW |
358 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
359 | CON_SOCK_STATE_CLOSED); | |
ce2c8903 | 360 | } |
a922d38f | 361 | |
31b8006e SW |
362 | /* |
363 | * socket callback functions | |
364 | */ | |
365 | ||
366 | /* data available on socket, or listen socket received a connect */ | |
676d2369 | 367 | static void ceph_sock_data_ready(struct sock *sk) |
31b8006e | 368 | { |
bd406145 | 369 | struct ceph_connection *con = sk->sk_user_data; |
a2a32584 GH |
370 | if (atomic_read(&con->msgr->stopping)) { |
371 | return; | |
372 | } | |
bd406145 | 373 | |
31b8006e | 374 | if (sk->sk_state != TCP_CLOSE_WAIT) { |
327800bd | 375 | dout("%s on %p state = %lu, queueing work\n", __func__, |
31b8006e SW |
376 | con, con->state); |
377 | queue_con(con); | |
378 | } | |
379 | } | |
380 | ||
381 | /* socket has buffer space for writing */ | |
327800bd | 382 | static void ceph_sock_write_space(struct sock *sk) |
31b8006e | 383 | { |
d3002b97 | 384 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e | 385 | |
182fac26 JS |
386 | /* only queue to workqueue if there is data we want to write, |
387 | * and there is sufficient space in the socket buffer to accept | |
327800bd | 388 | * more data. clear SOCK_NOSPACE so that ceph_sock_write_space() |
182fac26 JS |
389 | * doesn't get called again until try_write() fills the socket |
390 | * buffer. See net/ipv4/tcp_input.c:tcp_check_space() | |
391 | * and net/core/stream.c:sk_stream_write_space(). | |
392 | */ | |
c9ffc77a | 393 | if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) { |
64dc6130 | 394 | if (sk_stream_is_writeable(sk)) { |
327800bd | 395 | dout("%s %p queueing write work\n", __func__, con); |
182fac26 JS |
396 | clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags); |
397 | queue_con(con); | |
398 | } | |
31b8006e | 399 | } else { |
327800bd | 400 | dout("%s %p nothing to write\n", __func__, con); |
31b8006e | 401 | } |
31b8006e SW |
402 | } |
403 | ||
404 | /* socket's state has changed */ | |
327800bd | 405 | static void ceph_sock_state_change(struct sock *sk) |
31b8006e | 406 | { |
bd406145 | 407 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e | 408 | |
327800bd | 409 | dout("%s %p state = %lu sk_state = %u\n", __func__, |
31b8006e SW |
410 | con, con->state, sk->sk_state); |
411 | ||
31b8006e SW |
412 | switch (sk->sk_state) { |
413 | case TCP_CLOSE: | |
327800bd | 414 | dout("%s TCP_CLOSE\n", __func__); |
df561f66 | 415 | fallthrough; |
31b8006e | 416 | case TCP_CLOSE_WAIT: |
327800bd | 417 | dout("%s TCP_CLOSE_WAIT\n", __func__); |
ce2c8903 | 418 | con_sock_state_closing(con); |
c9ffc77a | 419 | con_flag_set(con, CON_FLAG_SOCK_CLOSED); |
d65c9e0b | 420 | queue_con(con); |
31b8006e SW |
421 | break; |
422 | case TCP_ESTABLISHED: | |
327800bd | 423 | dout("%s TCP_ESTABLISHED\n", __func__); |
ce2c8903 | 424 | con_sock_state_connected(con); |
31b8006e SW |
425 | queue_con(con); |
426 | break; | |
d3002b97 AE |
427 | default: /* Everything else is uninteresting */ |
428 | break; | |
31b8006e SW |
429 | } |
430 | } | |
431 | ||
432 | /* | |
433 | * set up socket callbacks | |
434 | */ | |
435 | static void set_sock_callbacks(struct socket *sock, | |
436 | struct ceph_connection *con) | |
437 | { | |
438 | struct sock *sk = sock->sk; | |
bd406145 | 439 | sk->sk_user_data = con; |
327800bd AE |
440 | sk->sk_data_ready = ceph_sock_data_ready; |
441 | sk->sk_write_space = ceph_sock_write_space; | |
442 | sk->sk_state_change = ceph_sock_state_change; | |
31b8006e SW |
443 | } |
444 | ||
445 | ||
446 | /* | |
447 | * socket helpers | |
448 | */ | |
449 | ||
450 | /* | |
451 | * initiate connection to a remote socket. | |
452 | */ | |
41617d0c | 453 | static int ceph_tcp_connect(struct ceph_connection *con) |
31b8006e | 454 | { |
cede185b | 455 | struct sockaddr_storage ss = con->peer_addr.in_addr; /* align */ |
31b8006e | 456 | struct socket *sock; |
633ee407 | 457 | unsigned int noio_flag; |
31b8006e SW |
458 | int ret; |
459 | ||
460 | BUG_ON(con->sock); | |
633ee407 ID |
461 | |
462 | /* sock_create_kern() allocates with GFP_KERNEL */ | |
463 | noio_flag = memalloc_noio_save(); | |
cede185b | 464 | ret = sock_create_kern(read_pnet(&con->msgr->net), ss.ss_family, |
eeb1bd5c | 465 | SOCK_STREAM, IPPROTO_TCP, &sock); |
633ee407 | 466 | memalloc_noio_restore(noio_flag); |
31b8006e | 467 | if (ret) |
41617d0c | 468 | return ret; |
6d7fdb0a | 469 | sock->sk->sk_allocation = GFP_NOFS; |
31b8006e | 470 | |
a6a5349d SW |
471 | #ifdef CONFIG_LOCKDEP |
472 | lockdep_set_class(&sock->sk->sk_lock, &socket_class); | |
473 | #endif | |
474 | ||
31b8006e SW |
475 | set_sock_callbacks(sock, con); |
476 | ||
b726ec97 | 477 | dout("connect %s\n", ceph_pr_addr(&con->peer_addr)); |
31b8006e | 478 | |
89a86be0 | 479 | con_sock_state_connecting(con); |
cede185b | 480 | ret = sock->ops->connect(sock, (struct sockaddr *)&ss, sizeof(ss), |
f91d3471 | 481 | O_NONBLOCK); |
31b8006e SW |
482 | if (ret == -EINPROGRESS) { |
483 | dout("connect %s EINPROGRESS sk_state = %u\n", | |
b726ec97 | 484 | ceph_pr_addr(&con->peer_addr), |
31b8006e | 485 | sock->sk->sk_state); |
a5bc3129 | 486 | } else if (ret < 0) { |
31b8006e | 487 | pr_err("connect %s error %d\n", |
b726ec97 | 488 | ceph_pr_addr(&con->peer_addr), ret); |
31b8006e | 489 | sock_release(sock); |
41617d0c | 490 | return ret; |
a5bc3129 | 491 | } |
89baaa57 | 492 | |
12abc5ee CH |
493 | if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY)) |
494 | tcp_sock_set_nodelay(sock->sk); | |
ba988f87 | 495 | |
a5bc3129 | 496 | con->sock = sock; |
41617d0c | 497 | return 0; |
31b8006e SW |
498 | } |
499 | ||
e5c93883 ID |
500 | /* |
501 | * If @buf is NULL, discard up to @len bytes. | |
502 | */ | |
31b8006e SW |
503 | static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len) |
504 | { | |
505 | struct kvec iov = {buf, len}; | |
506 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
98bdb0aa | 507 | int r; |
31b8006e | 508 | |
e5c93883 ID |
509 | if (!buf) |
510 | msg.msg_flags |= MSG_TRUNC; | |
511 | ||
aa563d7b | 512 | iov_iter_kvec(&msg.msg_iter, READ, &iov, 1, len); |
100803a8 | 513 | r = sock_recvmsg(sock, &msg, msg.msg_flags); |
98bdb0aa SW |
514 | if (r == -EAGAIN) |
515 | r = 0; | |
516 | return r; | |
31b8006e SW |
517 | } |
518 | ||
afb3d90e AE |
519 | static int ceph_tcp_recvpage(struct socket *sock, struct page *page, |
520 | int page_offset, size_t length) | |
521 | { | |
100803a8 AV |
522 | struct bio_vec bvec = { |
523 | .bv_page = page, | |
524 | .bv_offset = page_offset, | |
525 | .bv_len = length | |
526 | }; | |
527 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
528 | int r; | |
afb3d90e AE |
529 | |
530 | BUG_ON(page_offset + length > PAGE_SIZE); | |
aa563d7b | 531 | iov_iter_bvec(&msg.msg_iter, READ, &bvec, 1, length); |
100803a8 AV |
532 | r = sock_recvmsg(sock, &msg, msg.msg_flags); |
533 | if (r == -EAGAIN) | |
534 | r = 0; | |
535 | return r; | |
afb3d90e AE |
536 | } |
537 | ||
31b8006e SW |
538 | /* |
539 | * write something. @more is true if caller will be sending more data | |
540 | * shortly. | |
541 | */ | |
542 | static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov, | |
87349cda | 543 | size_t kvlen, size_t len, bool more) |
31b8006e SW |
544 | { |
545 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
42961d23 | 546 | int r; |
31b8006e SW |
547 | |
548 | if (more) | |
549 | msg.msg_flags |= MSG_MORE; | |
550 | else | |
551 | msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */ | |
552 | ||
42961d23 SW |
553 | r = kernel_sendmsg(sock, &msg, iov, kvlen, len); |
554 | if (r == -EAGAIN) | |
555 | r = 0; | |
556 | return r; | |
31b8006e SW |
557 | } |
558 | ||
433b0a12 ID |
559 | /* |
560 | * @more: either or both of MSG_MORE and MSG_SENDPAGE_NOTLAST | |
561 | */ | |
178eda29 | 562 | static int ceph_tcp_sendpage(struct socket *sock, struct page *page, |
433b0a12 | 563 | int offset, size_t size, int more) |
178eda29 | 564 | { |
3239eb52 ID |
565 | ssize_t (*sendpage)(struct socket *sock, struct page *page, |
566 | int offset, size_t size, int flags); | |
433b0a12 | 567 | int flags = MSG_DONTWAIT | MSG_NOSIGNAL | more; |
178eda29 | 568 | int ret; |
178eda29 | 569 | |
7e241f64 ID |
570 | /* |
571 | * sendpage cannot properly handle pages with page_count == 0, | |
572 | * we need to fall back to sendmsg if that's the case. | |
573 | * | |
574 | * Same goes for slab pages: skb_can_coalesce() allows | |
575 | * coalescing neighboring slab objects into a single frag which | |
576 | * triggers one of hardened usercopy checks. | |
577 | */ | |
40efc4dc | 578 | if (sendpage_ok(page)) |
3239eb52 | 579 | sendpage = sock->ops->sendpage; |
61ff6e9b | 580 | else |
3239eb52 | 581 | sendpage = sock_no_sendpage; |
61ff6e9b | 582 | |
3239eb52 | 583 | ret = sendpage(sock, page, offset, size, flags); |
61ff6e9b AV |
584 | if (ret == -EAGAIN) |
585 | ret = 0; | |
178eda29 CC |
586 | |
587 | return ret; | |
588 | } | |
31b8006e SW |
589 | |
590 | /* | |
591 | * Shutdown/close the socket for the given connection. | |
592 | */ | |
593 | static int con_close_socket(struct ceph_connection *con) | |
594 | { | |
8007b8d6 | 595 | int rc = 0; |
31b8006e SW |
596 | |
597 | dout("con_close_socket on %p sock %p\n", con, con->sock); | |
8007b8d6 SW |
598 | if (con->sock) { |
599 | rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR); | |
600 | sock_release(con->sock); | |
601 | con->sock = NULL; | |
602 | } | |
456ea468 AE |
603 | |
604 | /* | |
4a861692 | 605 | * Forcibly clear the SOCK_CLOSED flag. It gets set |
456ea468 AE |
606 | * independent of the connection mutex, and we could have |
607 | * received a socket close event before we had the chance to | |
608 | * shut the socket down. | |
609 | */ | |
c9ffc77a | 610 | con_flag_clear(con, CON_FLAG_SOCK_CLOSED); |
8007b8d6 | 611 | |
ce2c8903 | 612 | con_sock_state_closed(con); |
31b8006e SW |
613 | return rc; |
614 | } | |
615 | ||
616 | /* | |
617 | * Reset a connection. Discard all incoming and outgoing messages | |
618 | * and clear *_seq state. | |
619 | */ | |
620 | static void ceph_msg_remove(struct ceph_msg *msg) | |
621 | { | |
622 | list_del_init(&msg->list_head); | |
38941f80 | 623 | |
31b8006e SW |
624 | ceph_msg_put(msg); |
625 | } | |
626 | static void ceph_msg_remove_list(struct list_head *head) | |
627 | { | |
628 | while (!list_empty(head)) { | |
629 | struct ceph_msg *msg = list_first_entry(head, struct ceph_msg, | |
630 | list_head); | |
631 | ceph_msg_remove(msg); | |
632 | } | |
633 | } | |
634 | ||
635 | static void reset_connection(struct ceph_connection *con) | |
636 | { | |
637 | /* reset connection, out_queue, msg_ and connect_seq */ | |
638 | /* discard existing out_queue and msg_seq */ | |
0fa6ebc6 | 639 | dout("reset_connection %p\n", con); |
31b8006e SW |
640 | ceph_msg_remove_list(&con->out_queue); |
641 | ceph_msg_remove_list(&con->out_sent); | |
642 | ||
cf3e5c40 | 643 | if (con->in_msg) { |
38941f80 | 644 | BUG_ON(con->in_msg->con != con); |
cf3e5c40 SW |
645 | ceph_msg_put(con->in_msg); |
646 | con->in_msg = NULL; | |
647 | } | |
648 | ||
31b8006e SW |
649 | con->connect_seq = 0; |
650 | con->out_seq = 0; | |
c86a2930 | 651 | if (con->out_msg) { |
583d0fef | 652 | BUG_ON(con->out_msg->con != con); |
c86a2930 SW |
653 | ceph_msg_put(con->out_msg); |
654 | con->out_msg = NULL; | |
655 | } | |
31b8006e | 656 | con->in_seq = 0; |
0e0d5e0c | 657 | con->in_seq_acked = 0; |
67645d76 ID |
658 | |
659 | con->out_skip = 0; | |
31b8006e SW |
660 | } |
661 | ||
662 | /* | |
663 | * mark a peer down. drop any open connections. | |
664 | */ | |
665 | void ceph_con_close(struct ceph_connection *con) | |
666 | { | |
8c50c817 | 667 | mutex_lock(&con->mutex); |
b726ec97 | 668 | dout("con_close %p peer %s\n", con, ceph_pr_addr(&con->peer_addr)); |
8dacc7da | 669 | con->state = CON_STATE_CLOSED; |
a5988c49 | 670 | |
c9ffc77a AE |
671 | con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */ |
672 | con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING); | |
673 | con_flag_clear(con, CON_FLAG_WRITE_PENDING); | |
674 | con_flag_clear(con, CON_FLAG_BACKOFF); | |
a5988c49 | 675 | |
31b8006e | 676 | reset_connection(con); |
6f2bc3ff | 677 | con->peer_global_seq = 0; |
37ab77ac | 678 | cancel_con(con); |
ee76e073 | 679 | con_close_socket(con); |
ec302645 | 680 | mutex_unlock(&con->mutex); |
31b8006e | 681 | } |
3d14c5d2 | 682 | EXPORT_SYMBOL(ceph_con_close); |
31b8006e | 683 | |
31b8006e SW |
684 | /* |
685 | * Reopen a closed connection, with a new peer address. | |
686 | */ | |
b7a9e5dd SW |
687 | void ceph_con_open(struct ceph_connection *con, |
688 | __u8 entity_type, __u64 entity_num, | |
689 | struct ceph_entity_addr *addr) | |
31b8006e | 690 | { |
5469155f | 691 | mutex_lock(&con->mutex); |
b726ec97 | 692 | dout("con_open %p %s\n", con, ceph_pr_addr(addr)); |
8dacc7da | 693 | |
122070a2 | 694 | WARN_ON(con->state != CON_STATE_CLOSED); |
8dacc7da | 695 | con->state = CON_STATE_PREOPEN; |
a5988c49 | 696 | |
b7a9e5dd SW |
697 | con->peer_name.type = (__u8) entity_type; |
698 | con->peer_name.num = cpu_to_le64(entity_num); | |
699 | ||
31b8006e | 700 | memcpy(&con->peer_addr, addr, sizeof(*addr)); |
03c677e1 | 701 | con->delay = 0; /* reset backoff memory */ |
5469155f | 702 | mutex_unlock(&con->mutex); |
31b8006e SW |
703 | queue_con(con); |
704 | } | |
3d14c5d2 | 705 | EXPORT_SYMBOL(ceph_con_open); |
31b8006e | 706 | |
87b315a5 SW |
707 | /* |
708 | * return true if this connection ever successfully opened | |
709 | */ | |
710 | bool ceph_con_opened(struct ceph_connection *con) | |
711 | { | |
712 | return con->connect_seq > 0; | |
713 | } | |
714 | ||
31b8006e SW |
715 | /* |
716 | * initialize a new connection. | |
717 | */ | |
1bfd89f4 AE |
718 | void ceph_con_init(struct ceph_connection *con, void *private, |
719 | const struct ceph_connection_operations *ops, | |
b7a9e5dd | 720 | struct ceph_messenger *msgr) |
31b8006e SW |
721 | { |
722 | dout("con_init %p\n", con); | |
723 | memset(con, 0, sizeof(*con)); | |
1bfd89f4 AE |
724 | con->private = private; |
725 | con->ops = ops; | |
31b8006e | 726 | con->msgr = msgr; |
ce2c8903 AE |
727 | |
728 | con_sock_state_init(con); | |
729 | ||
ec302645 | 730 | mutex_init(&con->mutex); |
31b8006e SW |
731 | INIT_LIST_HEAD(&con->out_queue); |
732 | INIT_LIST_HEAD(&con->out_sent); | |
68931622 | 733 | INIT_DELAYED_WORK(&con->work, ceph_con_workfn); |
a5988c49 | 734 | |
8dacc7da | 735 | con->state = CON_STATE_CLOSED; |
31b8006e | 736 | } |
3d14c5d2 | 737 | EXPORT_SYMBOL(ceph_con_init); |
31b8006e SW |
738 | |
739 | ||
740 | /* | |
741 | * We maintain a global counter to order connection attempts. Get | |
742 | * a unique seq greater than @gt. | |
743 | */ | |
744 | static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt) | |
745 | { | |
746 | u32 ret; | |
747 | ||
748 | spin_lock(&msgr->global_seq_lock); | |
749 | if (msgr->global_seq < gt) | |
750 | msgr->global_seq = gt; | |
751 | ret = ++msgr->global_seq; | |
752 | spin_unlock(&msgr->global_seq_lock); | |
753 | return ret; | |
754 | } | |
755 | ||
e2200423 | 756 | static void con_out_kvec_reset(struct ceph_connection *con) |
859eb799 | 757 | { |
67645d76 ID |
758 | BUG_ON(con->out_skip); |
759 | ||
859eb799 AE |
760 | con->out_kvec_left = 0; |
761 | con->out_kvec_bytes = 0; | |
762 | con->out_kvec_cur = &con->out_kvec[0]; | |
763 | } | |
764 | ||
e2200423 | 765 | static void con_out_kvec_add(struct ceph_connection *con, |
859eb799 AE |
766 | size_t size, void *data) |
767 | { | |
67645d76 | 768 | int index = con->out_kvec_left; |
859eb799 | 769 | |
67645d76 | 770 | BUG_ON(con->out_skip); |
859eb799 AE |
771 | BUG_ON(index >= ARRAY_SIZE(con->out_kvec)); |
772 | ||
773 | con->out_kvec[index].iov_len = size; | |
774 | con->out_kvec[index].iov_base = data; | |
775 | con->out_kvec_left++; | |
776 | con->out_kvec_bytes += size; | |
777 | } | |
31b8006e | 778 | |
67645d76 ID |
779 | /* |
780 | * Chop off a kvec from the end. Return residual number of bytes for | |
781 | * that kvec, i.e. how many bytes would have been written if the kvec | |
782 | * hadn't been nuked. | |
783 | */ | |
784 | static int con_out_kvec_skip(struct ceph_connection *con) | |
785 | { | |
786 | int off = con->out_kvec_cur - con->out_kvec; | |
787 | int skip = 0; | |
788 | ||
789 | if (con->out_kvec_bytes > 0) { | |
790 | skip = con->out_kvec[off + con->out_kvec_left - 1].iov_len; | |
791 | BUG_ON(con->out_kvec_bytes < skip); | |
792 | BUG_ON(!con->out_kvec_left); | |
793 | con->out_kvec_bytes -= skip; | |
794 | con->out_kvec_left--; | |
795 | } | |
796 | ||
797 | return skip; | |
798 | } | |
799 | ||
df6ad1f9 | 800 | #ifdef CONFIG_BLOCK |
6aaa4511 AE |
801 | |
802 | /* | |
803 | * For a bio data item, a piece is whatever remains of the next | |
804 | * entry in the current bio iovec, or the first entry in the next | |
805 | * bio in the list. | |
806 | */ | |
8ae4f4f5 | 807 | static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor, |
25aff7c5 | 808 | size_t length) |
6aaa4511 | 809 | { |
8ae4f4f5 | 810 | struct ceph_msg_data *data = cursor->data; |
5359a17d | 811 | struct ceph_bio_iter *it = &cursor->bio_iter; |
6aaa4511 | 812 | |
5359a17d ID |
813 | cursor->resid = min_t(size_t, length, data->bio_length); |
814 | *it = data->bio_pos; | |
815 | if (cursor->resid < it->iter.bi_size) | |
816 | it->iter.bi_size = cursor->resid; | |
6aaa4511 | 817 | |
5359a17d ID |
818 | BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter)); |
819 | cursor->last_piece = cursor->resid == bio_iter_len(it->bio, it->iter); | |
6aaa4511 AE |
820 | } |
821 | ||
8ae4f4f5 | 822 | static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor, |
6aaa4511 AE |
823 | size_t *page_offset, |
824 | size_t *length) | |
825 | { | |
5359a17d ID |
826 | struct bio_vec bv = bio_iter_iovec(cursor->bio_iter.bio, |
827 | cursor->bio_iter.iter); | |
6aaa4511 | 828 | |
5359a17d ID |
829 | *page_offset = bv.bv_offset; |
830 | *length = bv.bv_len; | |
831 | return bv.bv_page; | |
6aaa4511 AE |
832 | } |
833 | ||
8ae4f4f5 AE |
834 | static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor, |
835 | size_t bytes) | |
6aaa4511 | 836 | { |
5359a17d | 837 | struct ceph_bio_iter *it = &cursor->bio_iter; |
187df763 | 838 | struct page *page = bio_iter_page(it->bio, it->iter); |
6aaa4511 | 839 | |
5359a17d ID |
840 | BUG_ON(bytes > cursor->resid); |
841 | BUG_ON(bytes > bio_iter_len(it->bio, it->iter)); | |
25aff7c5 | 842 | cursor->resid -= bytes; |
5359a17d | 843 | bio_advance_iter(it->bio, &it->iter, bytes); |
f38a5181 | 844 | |
5359a17d ID |
845 | if (!cursor->resid) { |
846 | BUG_ON(!cursor->last_piece); | |
847 | return false; /* no more data */ | |
848 | } | |
f38a5181 | 849 | |
187df763 ID |
850 | if (!bytes || (it->iter.bi_size && it->iter.bi_bvec_done && |
851 | page == bio_iter_page(it->bio, it->iter))) | |
6aaa4511 AE |
852 | return false; /* more bytes to process in this segment */ |
853 | ||
5359a17d ID |
854 | if (!it->iter.bi_size) { |
855 | it->bio = it->bio->bi_next; | |
856 | it->iter = it->bio->bi_iter; | |
857 | if (cursor->resid < it->iter.bi_size) | |
858 | it->iter.bi_size = cursor->resid; | |
25aff7c5 | 859 | } |
6aaa4511 | 860 | |
5359a17d ID |
861 | BUG_ON(cursor->last_piece); |
862 | BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter)); | |
863 | cursor->last_piece = cursor->resid == bio_iter_len(it->bio, it->iter); | |
6aaa4511 AE |
864 | return true; |
865 | } | |
ea96571f | 866 | #endif /* CONFIG_BLOCK */ |
df6ad1f9 | 867 | |
b9e281c2 ID |
868 | static void ceph_msg_data_bvecs_cursor_init(struct ceph_msg_data_cursor *cursor, |
869 | size_t length) | |
870 | { | |
871 | struct ceph_msg_data *data = cursor->data; | |
872 | struct bio_vec *bvecs = data->bvec_pos.bvecs; | |
873 | ||
874 | cursor->resid = min_t(size_t, length, data->bvec_pos.iter.bi_size); | |
875 | cursor->bvec_iter = data->bvec_pos.iter; | |
876 | cursor->bvec_iter.bi_size = cursor->resid; | |
877 | ||
878 | BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter)); | |
879 | cursor->last_piece = | |
880 | cursor->resid == bvec_iter_len(bvecs, cursor->bvec_iter); | |
881 | } | |
882 | ||
883 | static struct page *ceph_msg_data_bvecs_next(struct ceph_msg_data_cursor *cursor, | |
884 | size_t *page_offset, | |
885 | size_t *length) | |
886 | { | |
887 | struct bio_vec bv = bvec_iter_bvec(cursor->data->bvec_pos.bvecs, | |
888 | cursor->bvec_iter); | |
889 | ||
890 | *page_offset = bv.bv_offset; | |
891 | *length = bv.bv_len; | |
892 | return bv.bv_page; | |
893 | } | |
894 | ||
895 | static bool ceph_msg_data_bvecs_advance(struct ceph_msg_data_cursor *cursor, | |
896 | size_t bytes) | |
897 | { | |
898 | struct bio_vec *bvecs = cursor->data->bvec_pos.bvecs; | |
187df763 | 899 | struct page *page = bvec_iter_page(bvecs, cursor->bvec_iter); |
b9e281c2 ID |
900 | |
901 | BUG_ON(bytes > cursor->resid); | |
902 | BUG_ON(bytes > bvec_iter_len(bvecs, cursor->bvec_iter)); | |
903 | cursor->resid -= bytes; | |
904 | bvec_iter_advance(bvecs, &cursor->bvec_iter, bytes); | |
905 | ||
906 | if (!cursor->resid) { | |
907 | BUG_ON(!cursor->last_piece); | |
908 | return false; /* no more data */ | |
909 | } | |
910 | ||
187df763 ID |
911 | if (!bytes || (cursor->bvec_iter.bi_bvec_done && |
912 | page == bvec_iter_page(bvecs, cursor->bvec_iter))) | |
b9e281c2 ID |
913 | return false; /* more bytes to process in this segment */ |
914 | ||
915 | BUG_ON(cursor->last_piece); | |
916 | BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter)); | |
917 | cursor->last_piece = | |
918 | cursor->resid == bvec_iter_len(bvecs, cursor->bvec_iter); | |
919 | return true; | |
920 | } | |
921 | ||
e766d7b5 AE |
922 | /* |
923 | * For a page array, a piece comes from the first page in the array | |
924 | * that has not already been fully consumed. | |
925 | */ | |
8ae4f4f5 | 926 | static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor, |
25aff7c5 | 927 | size_t length) |
e766d7b5 | 928 | { |
8ae4f4f5 | 929 | struct ceph_msg_data *data = cursor->data; |
e766d7b5 AE |
930 | int page_count; |
931 | ||
932 | BUG_ON(data->type != CEPH_MSG_DATA_PAGES); | |
933 | ||
934 | BUG_ON(!data->pages); | |
935 | BUG_ON(!data->length); | |
936 | ||
ca8b3a69 | 937 | cursor->resid = min(length, data->length); |
e766d7b5 | 938 | page_count = calc_pages_for(data->alignment, (u64)data->length); |
e766d7b5 AE |
939 | cursor->page_offset = data->alignment & ~PAGE_MASK; |
940 | cursor->page_index = 0; | |
56fc5659 AE |
941 | BUG_ON(page_count > (int)USHRT_MAX); |
942 | cursor->page_count = (unsigned short)page_count; | |
943 | BUG_ON(length > SIZE_MAX - cursor->page_offset); | |
5f740d7e | 944 | cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE; |
e766d7b5 AE |
945 | } |
946 | ||
8ae4f4f5 AE |
947 | static struct page * |
948 | ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor, | |
949 | size_t *page_offset, size_t *length) | |
e766d7b5 | 950 | { |
8ae4f4f5 | 951 | struct ceph_msg_data *data = cursor->data; |
e766d7b5 AE |
952 | |
953 | BUG_ON(data->type != CEPH_MSG_DATA_PAGES); | |
954 | ||
955 | BUG_ON(cursor->page_index >= cursor->page_count); | |
956 | BUG_ON(cursor->page_offset >= PAGE_SIZE); | |
e766d7b5 AE |
957 | |
958 | *page_offset = cursor->page_offset; | |
25aff7c5 | 959 | if (cursor->last_piece) |
e766d7b5 | 960 | *length = cursor->resid; |
25aff7c5 | 961 | else |
e766d7b5 | 962 | *length = PAGE_SIZE - *page_offset; |
e766d7b5 AE |
963 | |
964 | return data->pages[cursor->page_index]; | |
965 | } | |
966 | ||
8ae4f4f5 | 967 | static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor, |
e766d7b5 AE |
968 | size_t bytes) |
969 | { | |
8ae4f4f5 | 970 | BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES); |
e766d7b5 AE |
971 | |
972 | BUG_ON(cursor->page_offset + bytes > PAGE_SIZE); | |
e766d7b5 AE |
973 | |
974 | /* Advance the cursor page offset */ | |
975 | ||
976 | cursor->resid -= bytes; | |
5df521b1 AE |
977 | cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK; |
978 | if (!bytes || cursor->page_offset) | |
e766d7b5 AE |
979 | return false; /* more bytes to process in the current page */ |
980 | ||
d90deda6 YZ |
981 | if (!cursor->resid) |
982 | return false; /* no more data */ | |
983 | ||
5df521b1 | 984 | /* Move on to the next page; offset is already at 0 */ |
e766d7b5 AE |
985 | |
986 | BUG_ON(cursor->page_index >= cursor->page_count); | |
e766d7b5 | 987 | cursor->page_index++; |
25aff7c5 | 988 | cursor->last_piece = cursor->resid <= PAGE_SIZE; |
e766d7b5 AE |
989 | |
990 | return true; | |
991 | } | |
992 | ||
fe38a2b6 | 993 | /* |
dd236fcb AE |
994 | * For a pagelist, a piece is whatever remains to be consumed in the |
995 | * first page in the list, or the front of the next page. | |
fe38a2b6 | 996 | */ |
8ae4f4f5 AE |
997 | static void |
998 | ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor, | |
25aff7c5 | 999 | size_t length) |
fe38a2b6 | 1000 | { |
8ae4f4f5 | 1001 | struct ceph_msg_data *data = cursor->data; |
fe38a2b6 AE |
1002 | struct ceph_pagelist *pagelist; |
1003 | struct page *page; | |
1004 | ||
dd236fcb | 1005 | BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST); |
fe38a2b6 AE |
1006 | |
1007 | pagelist = data->pagelist; | |
1008 | BUG_ON(!pagelist); | |
25aff7c5 AE |
1009 | |
1010 | if (!length) | |
fe38a2b6 AE |
1011 | return; /* pagelist can be assigned but empty */ |
1012 | ||
1013 | BUG_ON(list_empty(&pagelist->head)); | |
1014 | page = list_first_entry(&pagelist->head, struct page, lru); | |
1015 | ||
ca8b3a69 | 1016 | cursor->resid = min(length, pagelist->length); |
fe38a2b6 AE |
1017 | cursor->page = page; |
1018 | cursor->offset = 0; | |
a51b272e | 1019 | cursor->last_piece = cursor->resid <= PAGE_SIZE; |
fe38a2b6 AE |
1020 | } |
1021 | ||
8ae4f4f5 AE |
1022 | static struct page * |
1023 | ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor, | |
1024 | size_t *page_offset, size_t *length) | |
fe38a2b6 | 1025 | { |
8ae4f4f5 | 1026 | struct ceph_msg_data *data = cursor->data; |
fe38a2b6 | 1027 | struct ceph_pagelist *pagelist; |
fe38a2b6 AE |
1028 | |
1029 | BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST); | |
1030 | ||
1031 | pagelist = data->pagelist; | |
1032 | BUG_ON(!pagelist); | |
1033 | ||
1034 | BUG_ON(!cursor->page); | |
25aff7c5 | 1035 | BUG_ON(cursor->offset + cursor->resid != pagelist->length); |
fe38a2b6 | 1036 | |
5df521b1 | 1037 | /* offset of first page in pagelist is always 0 */ |
fe38a2b6 | 1038 | *page_offset = cursor->offset & ~PAGE_MASK; |
5df521b1 | 1039 | if (cursor->last_piece) |
25aff7c5 AE |
1040 | *length = cursor->resid; |
1041 | else | |
1042 | *length = PAGE_SIZE - *page_offset; | |
fe38a2b6 | 1043 | |
8ae4f4f5 | 1044 | return cursor->page; |
fe38a2b6 AE |
1045 | } |
1046 | ||
8ae4f4f5 | 1047 | static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor, |
dd236fcb | 1048 | size_t bytes) |
fe38a2b6 | 1049 | { |
8ae4f4f5 | 1050 | struct ceph_msg_data *data = cursor->data; |
fe38a2b6 AE |
1051 | struct ceph_pagelist *pagelist; |
1052 | ||
1053 | BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST); | |
1054 | ||
1055 | pagelist = data->pagelist; | |
1056 | BUG_ON(!pagelist); | |
25aff7c5 AE |
1057 | |
1058 | BUG_ON(cursor->offset + cursor->resid != pagelist->length); | |
fe38a2b6 AE |
1059 | BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE); |
1060 | ||
1061 | /* Advance the cursor offset */ | |
1062 | ||
25aff7c5 | 1063 | cursor->resid -= bytes; |
fe38a2b6 | 1064 | cursor->offset += bytes; |
5df521b1 | 1065 | /* offset of first page in pagelist is always 0 */ |
fe38a2b6 AE |
1066 | if (!bytes || cursor->offset & ~PAGE_MASK) |
1067 | return false; /* more bytes to process in the current page */ | |
1068 | ||
d90deda6 YZ |
1069 | if (!cursor->resid) |
1070 | return false; /* no more data */ | |
1071 | ||
fe38a2b6 AE |
1072 | /* Move on to the next page */ |
1073 | ||
1074 | BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head)); | |
17ddc49b | 1075 | cursor->page = list_next_entry(cursor->page, lru); |
25aff7c5 | 1076 | cursor->last_piece = cursor->resid <= PAGE_SIZE; |
fe38a2b6 AE |
1077 | |
1078 | return true; | |
1079 | } | |
1080 | ||
dd236fcb AE |
1081 | /* |
1082 | * Message data is handled (sent or received) in pieces, where each | |
1083 | * piece resides on a single page. The network layer might not | |
1084 | * consume an entire piece at once. A data item's cursor keeps | |
1085 | * track of which piece is next to process and how much remains to | |
1086 | * be processed in that piece. It also tracks whether the current | |
1087 | * piece is the last one in the data item. | |
1088 | */ | |
ca8b3a69 | 1089 | static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor) |
dd236fcb | 1090 | { |
ca8b3a69 | 1091 | size_t length = cursor->total_resid; |
8ae4f4f5 | 1092 | |
8ae4f4f5 | 1093 | switch (cursor->data->type) { |
dd236fcb | 1094 | case CEPH_MSG_DATA_PAGELIST: |
8ae4f4f5 | 1095 | ceph_msg_data_pagelist_cursor_init(cursor, length); |
dd236fcb | 1096 | break; |
e766d7b5 | 1097 | case CEPH_MSG_DATA_PAGES: |
8ae4f4f5 | 1098 | ceph_msg_data_pages_cursor_init(cursor, length); |
e766d7b5 | 1099 | break; |
dd236fcb AE |
1100 | #ifdef CONFIG_BLOCK |
1101 | case CEPH_MSG_DATA_BIO: | |
8ae4f4f5 | 1102 | ceph_msg_data_bio_cursor_init(cursor, length); |
6aaa4511 | 1103 | break; |
dd236fcb | 1104 | #endif /* CONFIG_BLOCK */ |
b9e281c2 ID |
1105 | case CEPH_MSG_DATA_BVECS: |
1106 | ceph_msg_data_bvecs_cursor_init(cursor, length); | |
1107 | break; | |
6aaa4511 | 1108 | case CEPH_MSG_DATA_NONE: |
dd236fcb AE |
1109 | default: |
1110 | /* BUG(); */ | |
1111 | break; | |
1112 | } | |
8ae4f4f5 | 1113 | cursor->need_crc = true; |
dd236fcb AE |
1114 | } |
1115 | ||
ca8b3a69 AE |
1116 | static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length) |
1117 | { | |
1118 | struct ceph_msg_data_cursor *cursor = &msg->cursor; | |
ca8b3a69 AE |
1119 | |
1120 | BUG_ON(!length); | |
1121 | BUG_ON(length > msg->data_length); | |
0d9c1ab3 | 1122 | BUG_ON(!msg->num_data_items); |
ca8b3a69 | 1123 | |
ca8b3a69 | 1124 | cursor->total_resid = length; |
0d9c1ab3 | 1125 | cursor->data = msg->data; |
ca8b3a69 AE |
1126 | |
1127 | __ceph_msg_data_cursor_init(cursor); | |
1128 | } | |
1129 | ||
dd236fcb AE |
1130 | /* |
1131 | * Return the page containing the next piece to process for a given | |
1132 | * data item, and supply the page offset and length of that piece. | |
1133 | * Indicate whether this is the last piece in this data item. | |
1134 | */ | |
8ae4f4f5 AE |
1135 | static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor, |
1136 | size_t *page_offset, size_t *length, | |
dd236fcb AE |
1137 | bool *last_piece) |
1138 | { | |
1139 | struct page *page; | |
1140 | ||
8ae4f4f5 | 1141 | switch (cursor->data->type) { |
dd236fcb | 1142 | case CEPH_MSG_DATA_PAGELIST: |
8ae4f4f5 | 1143 | page = ceph_msg_data_pagelist_next(cursor, page_offset, length); |
dd236fcb | 1144 | break; |
e766d7b5 | 1145 | case CEPH_MSG_DATA_PAGES: |
8ae4f4f5 | 1146 | page = ceph_msg_data_pages_next(cursor, page_offset, length); |
e766d7b5 | 1147 | break; |
dd236fcb AE |
1148 | #ifdef CONFIG_BLOCK |
1149 | case CEPH_MSG_DATA_BIO: | |
8ae4f4f5 | 1150 | page = ceph_msg_data_bio_next(cursor, page_offset, length); |
6aaa4511 | 1151 | break; |
dd236fcb | 1152 | #endif /* CONFIG_BLOCK */ |
b9e281c2 ID |
1153 | case CEPH_MSG_DATA_BVECS: |
1154 | page = ceph_msg_data_bvecs_next(cursor, page_offset, length); | |
1155 | break; | |
6aaa4511 | 1156 | case CEPH_MSG_DATA_NONE: |
dd236fcb AE |
1157 | default: |
1158 | page = NULL; | |
1159 | break; | |
1160 | } | |
5359a17d | 1161 | |
dd236fcb AE |
1162 | BUG_ON(!page); |
1163 | BUG_ON(*page_offset + *length > PAGE_SIZE); | |
1164 | BUG_ON(!*length); | |
5359a17d | 1165 | BUG_ON(*length > cursor->resid); |
dd236fcb | 1166 | if (last_piece) |
8ae4f4f5 | 1167 | *last_piece = cursor->last_piece; |
dd236fcb AE |
1168 | |
1169 | return page; | |
1170 | } | |
1171 | ||
1172 | /* | |
1173 | * Returns true if the result moves the cursor on to the next piece | |
1174 | * of the data item. | |
1175 | */ | |
1759f7b0 ID |
1176 | static void ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor, |
1177 | size_t bytes) | |
dd236fcb AE |
1178 | { |
1179 | bool new_piece; | |
1180 | ||
25aff7c5 | 1181 | BUG_ON(bytes > cursor->resid); |
8ae4f4f5 | 1182 | switch (cursor->data->type) { |
dd236fcb | 1183 | case CEPH_MSG_DATA_PAGELIST: |
8ae4f4f5 | 1184 | new_piece = ceph_msg_data_pagelist_advance(cursor, bytes); |
dd236fcb | 1185 | break; |
e766d7b5 | 1186 | case CEPH_MSG_DATA_PAGES: |
8ae4f4f5 | 1187 | new_piece = ceph_msg_data_pages_advance(cursor, bytes); |
e766d7b5 | 1188 | break; |
dd236fcb AE |
1189 | #ifdef CONFIG_BLOCK |
1190 | case CEPH_MSG_DATA_BIO: | |
8ae4f4f5 | 1191 | new_piece = ceph_msg_data_bio_advance(cursor, bytes); |
6aaa4511 | 1192 | break; |
dd236fcb | 1193 | #endif /* CONFIG_BLOCK */ |
b9e281c2 ID |
1194 | case CEPH_MSG_DATA_BVECS: |
1195 | new_piece = ceph_msg_data_bvecs_advance(cursor, bytes); | |
1196 | break; | |
6aaa4511 | 1197 | case CEPH_MSG_DATA_NONE: |
dd236fcb AE |
1198 | default: |
1199 | BUG(); | |
1200 | break; | |
1201 | } | |
ca8b3a69 | 1202 | cursor->total_resid -= bytes; |
dd236fcb | 1203 | |
ca8b3a69 AE |
1204 | if (!cursor->resid && cursor->total_resid) { |
1205 | WARN_ON(!cursor->last_piece); | |
0d9c1ab3 | 1206 | cursor->data++; |
ca8b3a69 | 1207 | __ceph_msg_data_cursor_init(cursor); |
a51b272e | 1208 | new_piece = true; |
ca8b3a69 | 1209 | } |
a51b272e | 1210 | cursor->need_crc = new_piece; |
dd236fcb AE |
1211 | } |
1212 | ||
dbc0d3ca ID |
1213 | static size_t sizeof_footer(struct ceph_connection *con) |
1214 | { | |
1215 | return (con->peer_features & CEPH_FEATURE_MSG_AUTH) ? | |
1216 | sizeof(struct ceph_msg_footer) : | |
1217 | sizeof(struct ceph_msg_footer_old); | |
1218 | } | |
1219 | ||
98fa5dd8 | 1220 | static void prepare_message_data(struct ceph_msg *msg, u32 data_len) |
739c905b | 1221 | { |
4c59b4a2 | 1222 | /* Initialize data cursor */ |
fe38a2b6 | 1223 | |
8ae4f4f5 | 1224 | ceph_msg_data_cursor_init(msg, (size_t)data_len); |
739c905b AE |
1225 | } |
1226 | ||
31b8006e SW |
1227 | /* |
1228 | * Prepare footer for currently outgoing message, and finish things | |
1229 | * off. Assumes out_kvec* are already valid.. we just add on to the end. | |
1230 | */ | |
859eb799 | 1231 | static void prepare_write_message_footer(struct ceph_connection *con) |
31b8006e SW |
1232 | { |
1233 | struct ceph_msg *m = con->out_msg; | |
1234 | ||
fd154f3c AE |
1235 | m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE; |
1236 | ||
31b8006e | 1237 | dout("prepare_write_message_footer %p\n", con); |
89f08173 | 1238 | con_out_kvec_add(con, sizeof_footer(con), &m->footer); |
33d07337 YZ |
1239 | if (con->peer_features & CEPH_FEATURE_MSG_AUTH) { |
1240 | if (con->ops->sign_message) | |
79dbd1ba | 1241 | con->ops->sign_message(m); |
33d07337 YZ |
1242 | else |
1243 | m->footer.sig = 0; | |
33d07337 YZ |
1244 | } else { |
1245 | m->old_footer.flags = m->footer.flags; | |
33d07337 | 1246 | } |
31b8006e | 1247 | con->out_more = m->more_to_follow; |
c86a2930 | 1248 | con->out_msg_done = true; |
31b8006e SW |
1249 | } |
1250 | ||
1251 | /* | |
1252 | * Prepare headers for the next outgoing message. | |
1253 | */ | |
1254 | static void prepare_write_message(struct ceph_connection *con) | |
1255 | { | |
1256 | struct ceph_msg *m; | |
a9a0c51a | 1257 | u32 crc; |
31b8006e | 1258 | |
e2200423 | 1259 | con_out_kvec_reset(con); |
c86a2930 | 1260 | con->out_msg_done = false; |
31b8006e SW |
1261 | |
1262 | /* Sneak an ack in there first? If we can get it into the same | |
1263 | * TCP packet that's a good thing. */ | |
1264 | if (con->in_seq > con->in_seq_acked) { | |
1265 | con->in_seq_acked = con->in_seq; | |
e2200423 | 1266 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
31b8006e | 1267 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 1268 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 | 1269 | &con->out_temp_ack); |
31b8006e SW |
1270 | } |
1271 | ||
38941f80 | 1272 | BUG_ON(list_empty(&con->out_queue)); |
859eb799 | 1273 | m = list_first_entry(&con->out_queue, struct ceph_msg, list_head); |
c86a2930 | 1274 | con->out_msg = m; |
38941f80 | 1275 | BUG_ON(m->con != con); |
4cf9d544 SW |
1276 | |
1277 | /* put message on sent list */ | |
1278 | ceph_msg_get(m); | |
1279 | list_move_tail(&m->list_head, &con->out_sent); | |
31b8006e | 1280 | |
e84346b7 SW |
1281 | /* |
1282 | * only assign outgoing seq # if we haven't sent this message | |
1283 | * yet. if it is requeued, resend with it's original seq. | |
1284 | */ | |
1285 | if (m->needs_out_seq) { | |
1286 | m->hdr.seq = cpu_to_le64(++con->out_seq); | |
1287 | m->needs_out_seq = false; | |
98ad5ebd | 1288 | |
4690faf0 ID |
1289 | if (con->ops->reencode_message) |
1290 | con->ops->reencode_message(m); | |
1291 | } | |
31b8006e | 1292 | |
98fa5dd8 | 1293 | dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n", |
31b8006e SW |
1294 | m, con->out_seq, le16_to_cpu(m->hdr.type), |
1295 | le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len), | |
98fa5dd8 | 1296 | m->data_length); |
98ad5ebd ID |
1297 | WARN_ON(m->front.iov_len != le32_to_cpu(m->hdr.front_len)); |
1298 | WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len)); | |
31b8006e SW |
1299 | |
1300 | /* tag + hdr + front + middle */ | |
e2200423 | 1301 | con_out_kvec_add(con, sizeof (tag_msg), &tag_msg); |
67645d76 | 1302 | con_out_kvec_add(con, sizeof(con->out_hdr), &con->out_hdr); |
e2200423 | 1303 | con_out_kvec_add(con, m->front.iov_len, m->front.iov_base); |
859eb799 | 1304 | |
31b8006e | 1305 | if (m->middle) |
e2200423 | 1306 | con_out_kvec_add(con, m->middle->vec.iov_len, |
859eb799 | 1307 | m->middle->vec.iov_base); |
31b8006e | 1308 | |
67645d76 | 1309 | /* fill in hdr crc and finalize hdr */ |
a9a0c51a AE |
1310 | crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc)); |
1311 | con->out_msg->hdr.crc = cpu_to_le32(crc); | |
67645d76 | 1312 | memcpy(&con->out_hdr, &con->out_msg->hdr, sizeof(con->out_hdr)); |
a9a0c51a | 1313 | |
67645d76 | 1314 | /* fill in front and middle crc, footer */ |
a9a0c51a AE |
1315 | crc = crc32c(0, m->front.iov_base, m->front.iov_len); |
1316 | con->out_msg->footer.front_crc = cpu_to_le32(crc); | |
1317 | if (m->middle) { | |
1318 | crc = crc32c(0, m->middle->vec.iov_base, | |
1319 | m->middle->vec.iov_len); | |
1320 | con->out_msg->footer.middle_crc = cpu_to_le32(crc); | |
1321 | } else | |
31b8006e | 1322 | con->out_msg->footer.middle_crc = 0; |
739c905b | 1323 | dout("%s front_crc %u middle_crc %u\n", __func__, |
31b8006e SW |
1324 | le32_to_cpu(con->out_msg->footer.front_crc), |
1325 | le32_to_cpu(con->out_msg->footer.middle_crc)); | |
67645d76 | 1326 | con->out_msg->footer.flags = 0; |
31b8006e SW |
1327 | |
1328 | /* is there a data payload? */ | |
739c905b | 1329 | con->out_msg->footer.data_crc = 0; |
98fa5dd8 AE |
1330 | if (m->data_length) { |
1331 | prepare_message_data(con->out_msg, m->data_length); | |
78625051 AE |
1332 | con->out_more = 1; /* data + footer will follow */ |
1333 | } else { | |
31b8006e | 1334 | /* no, queue up footer too and be done */ |
859eb799 | 1335 | prepare_write_message_footer(con); |
78625051 | 1336 | } |
31b8006e | 1337 | |
c9ffc77a | 1338 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
31b8006e SW |
1339 | } |
1340 | ||
1341 | /* | |
1342 | * Prepare an ack. | |
1343 | */ | |
1344 | static void prepare_write_ack(struct ceph_connection *con) | |
1345 | { | |
1346 | dout("prepare_write_ack %p %llu -> %llu\n", con, | |
1347 | con->in_seq_acked, con->in_seq); | |
1348 | con->in_seq_acked = con->in_seq; | |
1349 | ||
e2200423 | 1350 | con_out_kvec_reset(con); |
859eb799 | 1351 | |
e2200423 | 1352 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
859eb799 | 1353 | |
31b8006e | 1354 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 1355 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 AE |
1356 | &con->out_temp_ack); |
1357 | ||
31b8006e | 1358 | con->out_more = 1; /* more will follow.. eventually.. */ |
c9ffc77a | 1359 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
31b8006e SW |
1360 | } |
1361 | ||
3a23083b SW |
1362 | /* |
1363 | * Prepare to share the seq during handshake | |
1364 | */ | |
1365 | static void prepare_write_seq(struct ceph_connection *con) | |
1366 | { | |
1367 | dout("prepare_write_seq %p %llu -> %llu\n", con, | |
1368 | con->in_seq_acked, con->in_seq); | |
1369 | con->in_seq_acked = con->in_seq; | |
1370 | ||
1371 | con_out_kvec_reset(con); | |
1372 | ||
1373 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); | |
1374 | con_out_kvec_add(con, sizeof (con->out_temp_ack), | |
1375 | &con->out_temp_ack); | |
1376 | ||
1377 | con_flag_set(con, CON_FLAG_WRITE_PENDING); | |
1378 | } | |
1379 | ||
31b8006e SW |
1380 | /* |
1381 | * Prepare to write keepalive byte. | |
1382 | */ | |
1383 | static void prepare_write_keepalive(struct ceph_connection *con) | |
1384 | { | |
1385 | dout("prepare_write_keepalive %p\n", con); | |
e2200423 | 1386 | con_out_kvec_reset(con); |
8b9558aa | 1387 | if (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2) { |
473bd2d7 | 1388 | struct timespec64 now; |
7f61f545 | 1389 | |
473bd2d7 | 1390 | ktime_get_real_ts64(&now); |
8b9558aa | 1391 | con_out_kvec_add(con, sizeof(tag_keepalive2), &tag_keepalive2); |
473bd2d7 | 1392 | ceph_encode_timespec64(&con->out_temp_keepalive2, &now); |
7f61f545 ID |
1393 | con_out_kvec_add(con, sizeof(con->out_temp_keepalive2), |
1394 | &con->out_temp_keepalive2); | |
8b9558aa YZ |
1395 | } else { |
1396 | con_out_kvec_add(con, sizeof(tag_keepalive), &tag_keepalive); | |
1397 | } | |
c9ffc77a | 1398 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
31b8006e SW |
1399 | } |
1400 | ||
1401 | /* | |
1402 | * Connection negotiation. | |
1403 | */ | |
1404 | ||
262614c4 | 1405 | static int get_connect_authorizer(struct ceph_connection *con) |
4e7a5dcd | 1406 | { |
a3530df3 | 1407 | struct ceph_auth_handshake *auth; |
262614c4 | 1408 | int auth_proto; |
b1c6b980 AE |
1409 | |
1410 | if (!con->ops->get_authorizer) { | |
262614c4 | 1411 | con->auth = NULL; |
b1c6b980 AE |
1412 | con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN; |
1413 | con->out_connect.authorizer_len = 0; | |
262614c4 | 1414 | return 0; |
b1c6b980 AE |
1415 | } |
1416 | ||
262614c4 | 1417 | auth = con->ops->get_authorizer(con, &auth_proto, con->auth_retry); |
a3530df3 | 1418 | if (IS_ERR(auth)) |
262614c4 | 1419 | return PTR_ERR(auth); |
0da5d703 | 1420 | |
262614c4 ID |
1421 | con->auth = auth; |
1422 | con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto); | |
1423 | con->out_connect.authorizer_len = cpu_to_le32(auth->authorizer_buf_len); | |
1424 | return 0; | |
4e7a5dcd SW |
1425 | } |
1426 | ||
31b8006e SW |
1427 | /* |
1428 | * We connected to a peer and are saying hello. | |
1429 | */ | |
e825a66d | 1430 | static void prepare_write_banner(struct ceph_connection *con) |
31b8006e | 1431 | { |
e2200423 AE |
1432 | con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER); |
1433 | con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr), | |
e825a66d | 1434 | &con->msgr->my_enc_addr); |
eed0ef2c | 1435 | |
eed0ef2c | 1436 | con->out_more = 0; |
c9ffc77a | 1437 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
eed0ef2c SW |
1438 | } |
1439 | ||
c0f56b48 ID |
1440 | static void __prepare_write_connect(struct ceph_connection *con) |
1441 | { | |
1442 | con_out_kvec_add(con, sizeof(con->out_connect), &con->out_connect); | |
1443 | if (con->auth) | |
1444 | con_out_kvec_add(con, con->auth->authorizer_buf_len, | |
1445 | con->auth->authorizer_buf); | |
1446 | ||
1447 | con->out_more = 0; | |
1448 | con_flag_set(con, CON_FLAG_WRITE_PENDING); | |
1449 | } | |
1450 | ||
e825a66d | 1451 | static int prepare_write_connect(struct ceph_connection *con) |
eed0ef2c | 1452 | { |
95c96174 | 1453 | unsigned int global_seq = get_global_seq(con->msgr, 0); |
31b8006e | 1454 | int proto; |
262614c4 | 1455 | int ret; |
31b8006e SW |
1456 | |
1457 | switch (con->peer_name.type) { | |
1458 | case CEPH_ENTITY_TYPE_MON: | |
1459 | proto = CEPH_MONC_PROTOCOL; | |
1460 | break; | |
1461 | case CEPH_ENTITY_TYPE_OSD: | |
1462 | proto = CEPH_OSDC_PROTOCOL; | |
1463 | break; | |
1464 | case CEPH_ENTITY_TYPE_MDS: | |
1465 | proto = CEPH_MDSC_PROTOCOL; | |
1466 | break; | |
1467 | default: | |
1468 | BUG(); | |
1469 | } | |
1470 | ||
1471 | dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con, | |
1472 | con->connect_seq, global_seq, proto); | |
4e7a5dcd | 1473 | |
859bff51 ID |
1474 | con->out_connect.features = |
1475 | cpu_to_le64(from_msgr(con->msgr)->supported_features); | |
31b8006e SW |
1476 | con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT); |
1477 | con->out_connect.connect_seq = cpu_to_le32(con->connect_seq); | |
1478 | con->out_connect.global_seq = cpu_to_le32(global_seq); | |
1479 | con->out_connect.protocol_version = cpu_to_le32(proto); | |
1480 | con->out_connect.flags = 0; | |
31b8006e | 1481 | |
262614c4 ID |
1482 | ret = get_connect_authorizer(con); |
1483 | if (ret) | |
1484 | return ret; | |
3da54776 | 1485 | |
c0f56b48 | 1486 | __prepare_write_connect(con); |
e10c758e | 1487 | return 0; |
31b8006e SW |
1488 | } |
1489 | ||
31b8006e SW |
1490 | /* |
1491 | * write as much of pending kvecs to the socket as we can. | |
1492 | * 1 -> done | |
1493 | * 0 -> socket full, but more to do | |
1494 | * <0 -> error | |
1495 | */ | |
1496 | static int write_partial_kvec(struct ceph_connection *con) | |
1497 | { | |
1498 | int ret; | |
1499 | ||
1500 | dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes); | |
1501 | while (con->out_kvec_bytes > 0) { | |
1502 | ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur, | |
1503 | con->out_kvec_left, con->out_kvec_bytes, | |
1504 | con->out_more); | |
1505 | if (ret <= 0) | |
1506 | goto out; | |
1507 | con->out_kvec_bytes -= ret; | |
1508 | if (con->out_kvec_bytes == 0) | |
1509 | break; /* done */ | |
f42299e6 AE |
1510 | |
1511 | /* account for full iov entries consumed */ | |
1512 | while (ret >= con->out_kvec_cur->iov_len) { | |
1513 | BUG_ON(!con->out_kvec_left); | |
1514 | ret -= con->out_kvec_cur->iov_len; | |
1515 | con->out_kvec_cur++; | |
1516 | con->out_kvec_left--; | |
1517 | } | |
1518 | /* and for a partially-consumed entry */ | |
1519 | if (ret) { | |
1520 | con->out_kvec_cur->iov_len -= ret; | |
1521 | con->out_kvec_cur->iov_base += ret; | |
31b8006e SW |
1522 | } |
1523 | } | |
1524 | con->out_kvec_left = 0; | |
31b8006e SW |
1525 | ret = 1; |
1526 | out: | |
1527 | dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con, | |
1528 | con->out_kvec_bytes, con->out_kvec_left, ret); | |
1529 | return ret; /* done! */ | |
1530 | } | |
1531 | ||
35b62808 AE |
1532 | static u32 ceph_crc32c_page(u32 crc, struct page *page, |
1533 | unsigned int page_offset, | |
1534 | unsigned int length) | |
1535 | { | |
1536 | char *kaddr; | |
1537 | ||
1538 | kaddr = kmap(page); | |
1539 | BUG_ON(kaddr == NULL); | |
1540 | crc = crc32c(crc, kaddr + page_offset, length); | |
1541 | kunmap(page); | |
1542 | ||
1543 | return crc; | |
1544 | } | |
31b8006e SW |
1545 | /* |
1546 | * Write as much message data payload as we can. If we finish, queue | |
1547 | * up the footer. | |
1548 | * 1 -> done, footer is now queued in out_kvec[]. | |
1549 | * 0 -> socket full, but more to do | |
1550 | * <0 -> error | |
1551 | */ | |
34d2d200 | 1552 | static int write_partial_message_data(struct ceph_connection *con) |
31b8006e SW |
1553 | { |
1554 | struct ceph_msg *msg = con->out_msg; | |
8ae4f4f5 | 1555 | struct ceph_msg_data_cursor *cursor = &msg->cursor; |
859bff51 | 1556 | bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC); |
433b0a12 | 1557 | int more = MSG_MORE | MSG_SENDPAGE_NOTLAST; |
f5db90bc | 1558 | u32 crc; |
31b8006e | 1559 | |
859a35d5 | 1560 | dout("%s %p msg %p\n", __func__, con, msg); |
31b8006e | 1561 | |
0d9c1ab3 | 1562 | if (!msg->num_data_items) |
4c59b4a2 AE |
1563 | return -EINVAL; |
1564 | ||
5821bd8c AE |
1565 | /* |
1566 | * Iterate through each page that contains data to be | |
1567 | * written, and send as much as possible for each. | |
1568 | * | |
1569 | * If we are calculating the data crc (the default), we will | |
1570 | * need to map the page. If we have no pages, they have | |
1571 | * been revoked, so use the zero page. | |
1572 | */ | |
f5db90bc | 1573 | crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0; |
45a267db | 1574 | while (cursor->total_resid) { |
8a166d05 | 1575 | struct page *page; |
e387d525 AE |
1576 | size_t page_offset; |
1577 | size_t length; | |
f5db90bc | 1578 | int ret; |
68b4476b | 1579 | |
45a267db ID |
1580 | if (!cursor->resid) { |
1581 | ceph_msg_data_advance(cursor, 0); | |
1582 | continue; | |
1583 | } | |
1584 | ||
1f6b821a | 1585 | page = ceph_msg_data_next(cursor, &page_offset, &length, NULL); |
433b0a12 ID |
1586 | if (length == cursor->total_resid) |
1587 | more = MSG_MORE; | |
1f6b821a | 1588 | ret = ceph_tcp_sendpage(con->sock, page, page_offset, length, |
433b0a12 | 1589 | more); |
f5db90bc AE |
1590 | if (ret <= 0) { |
1591 | if (do_datacrc) | |
1592 | msg->footer.data_crc = cpu_to_le32(crc); | |
31b8006e | 1593 | |
f5db90bc AE |
1594 | return ret; |
1595 | } | |
143334ff AE |
1596 | if (do_datacrc && cursor->need_crc) |
1597 | crc = ceph_crc32c_page(crc, page, page_offset, length); | |
1759f7b0 | 1598 | ceph_msg_data_advance(cursor, (size_t)ret); |
31b8006e SW |
1599 | } |
1600 | ||
34d2d200 | 1601 | dout("%s %p msg %p done\n", __func__, con, msg); |
31b8006e SW |
1602 | |
1603 | /* prepare and queue up footer, too */ | |
f5db90bc AE |
1604 | if (do_datacrc) |
1605 | msg->footer.data_crc = cpu_to_le32(crc); | |
1606 | else | |
84ca8fc8 | 1607 | msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC; |
e2200423 | 1608 | con_out_kvec_reset(con); |
859eb799 | 1609 | prepare_write_message_footer(con); |
f5db90bc AE |
1610 | |
1611 | return 1; /* must return > 0 to indicate success */ | |
31b8006e SW |
1612 | } |
1613 | ||
1614 | /* | |
1615 | * write some zeros | |
1616 | */ | |
1617 | static int write_partial_skip(struct ceph_connection *con) | |
1618 | { | |
433b0a12 | 1619 | int more = MSG_MORE | MSG_SENDPAGE_NOTLAST; |
31b8006e SW |
1620 | int ret; |
1621 | ||
67645d76 | 1622 | dout("%s %p %d left\n", __func__, con, con->out_skip); |
31b8006e | 1623 | while (con->out_skip > 0) { |
09cbfeaf | 1624 | size_t size = min(con->out_skip, (int) PAGE_SIZE); |
31b8006e | 1625 | |
433b0a12 ID |
1626 | if (size == con->out_skip) |
1627 | more = MSG_MORE; | |
1628 | ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, more); | |
31b8006e SW |
1629 | if (ret <= 0) |
1630 | goto out; | |
1631 | con->out_skip -= ret; | |
1632 | } | |
1633 | ret = 1; | |
1634 | out: | |
1635 | return ret; | |
1636 | } | |
1637 | ||
1638 | /* | |
1639 | * Prepare to read connection handshake, or an ack. | |
1640 | */ | |
eed0ef2c SW |
1641 | static void prepare_read_banner(struct ceph_connection *con) |
1642 | { | |
1643 | dout("prepare_read_banner %p\n", con); | |
1644 | con->in_base_pos = 0; | |
1645 | } | |
1646 | ||
31b8006e SW |
1647 | static void prepare_read_connect(struct ceph_connection *con) |
1648 | { | |
1649 | dout("prepare_read_connect %p\n", con); | |
1650 | con->in_base_pos = 0; | |
1651 | } | |
1652 | ||
1653 | static void prepare_read_ack(struct ceph_connection *con) | |
1654 | { | |
1655 | dout("prepare_read_ack %p\n", con); | |
1656 | con->in_base_pos = 0; | |
1657 | } | |
1658 | ||
3a23083b SW |
1659 | static void prepare_read_seq(struct ceph_connection *con) |
1660 | { | |
1661 | dout("prepare_read_seq %p\n", con); | |
1662 | con->in_base_pos = 0; | |
1663 | con->in_tag = CEPH_MSGR_TAG_SEQ; | |
1664 | } | |
1665 | ||
31b8006e SW |
1666 | static void prepare_read_tag(struct ceph_connection *con) |
1667 | { | |
1668 | dout("prepare_read_tag %p\n", con); | |
1669 | con->in_base_pos = 0; | |
1670 | con->in_tag = CEPH_MSGR_TAG_READY; | |
1671 | } | |
1672 | ||
8b9558aa YZ |
1673 | static void prepare_read_keepalive_ack(struct ceph_connection *con) |
1674 | { | |
1675 | dout("prepare_read_keepalive_ack %p\n", con); | |
1676 | con->in_base_pos = 0; | |
1677 | } | |
1678 | ||
31b8006e SW |
1679 | /* |
1680 | * Prepare to read a message. | |
1681 | */ | |
1682 | static int prepare_read_message(struct ceph_connection *con) | |
1683 | { | |
1684 | dout("prepare_read_message %p\n", con); | |
1685 | BUG_ON(con->in_msg != NULL); | |
1686 | con->in_base_pos = 0; | |
1687 | con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0; | |
1688 | return 0; | |
1689 | } | |
1690 | ||
1691 | ||
1692 | static int read_partial(struct ceph_connection *con, | |
fd51653f | 1693 | int end, int size, void *object) |
31b8006e | 1694 | { |
e6cee71f AE |
1695 | while (con->in_base_pos < end) { |
1696 | int left = end - con->in_base_pos; | |
31b8006e SW |
1697 | int have = size - left; |
1698 | int ret = ceph_tcp_recvmsg(con->sock, object + have, left); | |
1699 | if (ret <= 0) | |
1700 | return ret; | |
1701 | con->in_base_pos += ret; | |
1702 | } | |
1703 | return 1; | |
1704 | } | |
1705 | ||
1706 | ||
1707 | /* | |
1708 | * Read all or part of the connect-side handshake on a new connection | |
1709 | */ | |
eed0ef2c | 1710 | static int read_partial_banner(struct ceph_connection *con) |
31b8006e | 1711 | { |
fd51653f AE |
1712 | int size; |
1713 | int end; | |
1714 | int ret; | |
31b8006e | 1715 | |
eed0ef2c | 1716 | dout("read_partial_banner %p at %d\n", con, con->in_base_pos); |
31b8006e SW |
1717 | |
1718 | /* peer's banner */ | |
fd51653f AE |
1719 | size = strlen(CEPH_BANNER); |
1720 | end = size; | |
1721 | ret = read_partial(con, end, size, con->in_banner); | |
31b8006e SW |
1722 | if (ret <= 0) |
1723 | goto out; | |
fd51653f AE |
1724 | |
1725 | size = sizeof (con->actual_peer_addr); | |
1726 | end += size; | |
1727 | ret = read_partial(con, end, size, &con->actual_peer_addr); | |
31b8006e SW |
1728 | if (ret <= 0) |
1729 | goto out; | |
2c66de56 | 1730 | ceph_decode_banner_addr(&con->actual_peer_addr); |
fd51653f AE |
1731 | |
1732 | size = sizeof (con->peer_addr_for_me); | |
1733 | end += size; | |
1734 | ret = read_partial(con, end, size, &con->peer_addr_for_me); | |
31b8006e SW |
1735 | if (ret <= 0) |
1736 | goto out; | |
2c66de56 | 1737 | ceph_decode_banner_addr(&con->peer_addr_for_me); |
fd51653f | 1738 | |
eed0ef2c SW |
1739 | out: |
1740 | return ret; | |
1741 | } | |
1742 | ||
1743 | static int read_partial_connect(struct ceph_connection *con) | |
1744 | { | |
fd51653f AE |
1745 | int size; |
1746 | int end; | |
1747 | int ret; | |
eed0ef2c SW |
1748 | |
1749 | dout("read_partial_connect %p at %d\n", con, con->in_base_pos); | |
1750 | ||
fd51653f AE |
1751 | size = sizeof (con->in_reply); |
1752 | end = size; | |
1753 | ret = read_partial(con, end, size, &con->in_reply); | |
31b8006e SW |
1754 | if (ret <= 0) |
1755 | goto out; | |
fd51653f | 1756 | |
262614c4 ID |
1757 | if (con->auth) { |
1758 | size = le32_to_cpu(con->in_reply.authorizer_len); | |
130f52f2 ID |
1759 | if (size > con->auth->authorizer_reply_buf_len) { |
1760 | pr_err("authorizer reply too big: %d > %zu\n", size, | |
1761 | con->auth->authorizer_reply_buf_len); | |
1762 | ret = -EINVAL; | |
1763 | goto out; | |
1764 | } | |
1765 | ||
262614c4 ID |
1766 | end += size; |
1767 | ret = read_partial(con, end, size, | |
1768 | con->auth->authorizer_reply_buf); | |
1769 | if (ret <= 0) | |
1770 | goto out; | |
1771 | } | |
31b8006e | 1772 | |
4e7a5dcd SW |
1773 | dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n", |
1774 | con, (int)con->in_reply.tag, | |
1775 | le32_to_cpu(con->in_reply.connect_seq), | |
31b8006e SW |
1776 | le32_to_cpu(con->in_reply.global_seq)); |
1777 | out: | |
1778 | return ret; | |
1779 | } | |
1780 | ||
1781 | /* | |
1782 | * Verify the hello banner looks okay. | |
1783 | */ | |
1784 | static int verify_hello(struct ceph_connection *con) | |
1785 | { | |
1786 | if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) { | |
13e38c8a | 1787 | pr_err("connect to %s got bad banner\n", |
b726ec97 | 1788 | ceph_pr_addr(&con->peer_addr)); |
31b8006e SW |
1789 | con->error_msg = "protocol error, bad banner"; |
1790 | return -1; | |
1791 | } | |
1792 | return 0; | |
1793 | } | |
1794 | ||
cede185b | 1795 | static bool addr_is_blank(struct ceph_entity_addr *addr) |
31b8006e | 1796 | { |
cede185b JL |
1797 | struct sockaddr_storage ss = addr->in_addr; /* align */ |
1798 | struct in_addr *addr4 = &((struct sockaddr_in *)&ss)->sin_addr; | |
1799 | struct in6_addr *addr6 = &((struct sockaddr_in6 *)&ss)->sin6_addr; | |
c44bd69c | 1800 | |
cede185b | 1801 | switch (ss.ss_family) { |
31b8006e | 1802 | case AF_INET: |
cede185b | 1803 | return addr4->s_addr == htonl(INADDR_ANY); |
31b8006e | 1804 | case AF_INET6: |
c44bd69c ID |
1805 | return ipv6_addr_any(addr6); |
1806 | default: | |
1807 | return true; | |
31b8006e | 1808 | } |
31b8006e SW |
1809 | } |
1810 | ||
cede185b | 1811 | static int addr_port(struct ceph_entity_addr *addr) |
31b8006e | 1812 | { |
cede185b | 1813 | switch (get_unaligned(&addr->in_addr.ss_family)) { |
31b8006e | 1814 | case AF_INET: |
cede185b | 1815 | return ntohs(get_unaligned(&((struct sockaddr_in *)&addr->in_addr)->sin_port)); |
31b8006e | 1816 | case AF_INET6: |
cede185b | 1817 | return ntohs(get_unaligned(&((struct sockaddr_in6 *)&addr->in_addr)->sin6_port)); |
31b8006e SW |
1818 | } |
1819 | return 0; | |
1820 | } | |
1821 | ||
cede185b | 1822 | static void addr_set_port(struct ceph_entity_addr *addr, int p) |
31b8006e | 1823 | { |
cede185b | 1824 | switch (get_unaligned(&addr->in_addr.ss_family)) { |
31b8006e | 1825 | case AF_INET: |
cede185b | 1826 | put_unaligned(htons(p), &((struct sockaddr_in *)&addr->in_addr)->sin_port); |
a2a79609 | 1827 | break; |
31b8006e | 1828 | case AF_INET6: |
cede185b | 1829 | put_unaligned(htons(p), &((struct sockaddr_in6 *)&addr->in_addr)->sin6_port); |
a2a79609 | 1830 | break; |
31b8006e SW |
1831 | } |
1832 | } | |
1833 | ||
ee3b56f2 NW |
1834 | /* |
1835 | * Unlike other *_pton function semantics, zero indicates success. | |
1836 | */ | |
cede185b | 1837 | static int ceph_pton(const char *str, size_t len, struct ceph_entity_addr *addr, |
ee3b56f2 NW |
1838 | char delim, const char **ipend) |
1839 | { | |
cede185b | 1840 | memset(&addr->in_addr, 0, sizeof(addr->in_addr)); |
ee3b56f2 | 1841 | |
cede185b JL |
1842 | if (in4_pton(str, len, (u8 *)&((struct sockaddr_in *)&addr->in_addr)->sin_addr.s_addr, delim, ipend)) { |
1843 | put_unaligned(AF_INET, &addr->in_addr.ss_family); | |
ee3b56f2 NW |
1844 | return 0; |
1845 | } | |
1846 | ||
cede185b JL |
1847 | if (in6_pton(str, len, (u8 *)&((struct sockaddr_in6 *)&addr->in_addr)->sin6_addr.s6_addr, delim, ipend)) { |
1848 | put_unaligned(AF_INET6, &addr->in_addr.ss_family); | |
ee3b56f2 NW |
1849 | return 0; |
1850 | } | |
1851 | ||
1852 | return -EINVAL; | |
1853 | } | |
1854 | ||
1855 | /* | |
1856 | * Extract hostname string and resolve using kernel DNS facility. | |
1857 | */ | |
1858 | #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER | |
1859 | static int ceph_dns_resolve_name(const char *name, size_t namelen, | |
cede185b | 1860 | struct ceph_entity_addr *addr, char delim, const char **ipend) |
ee3b56f2 NW |
1861 | { |
1862 | const char *end, *delim_p; | |
1863 | char *colon_p, *ip_addr = NULL; | |
1864 | int ip_len, ret; | |
1865 | ||
1866 | /* | |
1867 | * The end of the hostname occurs immediately preceding the delimiter or | |
1868 | * the port marker (':') where the delimiter takes precedence. | |
1869 | */ | |
1870 | delim_p = memchr(name, delim, namelen); | |
1871 | colon_p = memchr(name, ':', namelen); | |
1872 | ||
1873 | if (delim_p && colon_p) | |
1874 | end = delim_p < colon_p ? delim_p : colon_p; | |
1875 | else if (!delim_p && colon_p) | |
1876 | end = colon_p; | |
1877 | else { | |
1878 | end = delim_p; | |
1879 | if (!end) /* case: hostname:/ */ | |
1880 | end = name + namelen; | |
1881 | } | |
1882 | ||
1883 | if (end <= name) | |
1884 | return -EINVAL; | |
1885 | ||
1886 | /* do dns_resolve upcall */ | |
a58946c1 DH |
1887 | ip_len = dns_query(current->nsproxy->net_ns, |
1888 | NULL, name, end - name, NULL, &ip_addr, NULL, false); | |
ee3b56f2 | 1889 | if (ip_len > 0) |
cede185b | 1890 | ret = ceph_pton(ip_addr, ip_len, addr, -1, NULL); |
ee3b56f2 NW |
1891 | else |
1892 | ret = -ESRCH; | |
1893 | ||
1894 | kfree(ip_addr); | |
1895 | ||
1896 | *ipend = end; | |
1897 | ||
1898 | pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name, | |
b726ec97 | 1899 | ret, ret ? "failed" : ceph_pr_addr(addr)); |
ee3b56f2 NW |
1900 | |
1901 | return ret; | |
1902 | } | |
1903 | #else | |
1904 | static inline int ceph_dns_resolve_name(const char *name, size_t namelen, | |
cede185b | 1905 | struct ceph_entity_addr *addr, char delim, const char **ipend) |
ee3b56f2 NW |
1906 | { |
1907 | return -EINVAL; | |
1908 | } | |
1909 | #endif | |
1910 | ||
1911 | /* | |
1912 | * Parse a server name (IP or hostname). If a valid IP address is not found | |
1913 | * then try to extract a hostname to resolve using userspace DNS upcall. | |
1914 | */ | |
1915 | static int ceph_parse_server_name(const char *name, size_t namelen, | |
cede185b | 1916 | struct ceph_entity_addr *addr, char delim, const char **ipend) |
ee3b56f2 NW |
1917 | { |
1918 | int ret; | |
1919 | ||
cede185b | 1920 | ret = ceph_pton(name, namelen, addr, delim, ipend); |
ee3b56f2 | 1921 | if (ret) |
cede185b | 1922 | ret = ceph_dns_resolve_name(name, namelen, addr, delim, ipend); |
ee3b56f2 NW |
1923 | |
1924 | return ret; | |
1925 | } | |
1926 | ||
31b8006e SW |
1927 | /* |
1928 | * Parse an ip[:port] list into an addr array. Use the default | |
1929 | * monitor port if a port isn't specified. | |
1930 | */ | |
1931 | int ceph_parse_ips(const char *c, const char *end, | |
1932 | struct ceph_entity_addr *addr, | |
1933 | int max_count, int *count) | |
1934 | { | |
ee3b56f2 | 1935 | int i, ret = -EINVAL; |
31b8006e SW |
1936 | const char *p = c; |
1937 | ||
1938 | dout("parse_ips on '%.*s'\n", (int)(end-c), c); | |
1939 | for (i = 0; i < max_count; i++) { | |
1940 | const char *ipend; | |
31b8006e | 1941 | int port; |
39139f64 SW |
1942 | char delim = ','; |
1943 | ||
1944 | if (*p == '[') { | |
1945 | delim = ']'; | |
1946 | p++; | |
1947 | } | |
31b8006e | 1948 | |
cede185b | 1949 | ret = ceph_parse_server_name(p, end - p, &addr[i], delim, &ipend); |
ee3b56f2 | 1950 | if (ret) |
31b8006e | 1951 | goto bad; |
ee3b56f2 NW |
1952 | ret = -EINVAL; |
1953 | ||
31b8006e SW |
1954 | p = ipend; |
1955 | ||
39139f64 SW |
1956 | if (delim == ']') { |
1957 | if (*p != ']') { | |
1958 | dout("missing matching ']'\n"); | |
1959 | goto bad; | |
1960 | } | |
1961 | p++; | |
1962 | } | |
1963 | ||
31b8006e SW |
1964 | /* port? */ |
1965 | if (p < end && *p == ':') { | |
1966 | port = 0; | |
1967 | p++; | |
1968 | while (p < end && *p >= '0' && *p <= '9') { | |
1969 | port = (port * 10) + (*p - '0'); | |
1970 | p++; | |
1971 | } | |
f48db1e9 ID |
1972 | if (port == 0) |
1973 | port = CEPH_MON_PORT; | |
1974 | else if (port > 65535) | |
31b8006e SW |
1975 | goto bad; |
1976 | } else { | |
1977 | port = CEPH_MON_PORT; | |
1978 | } | |
1979 | ||
cede185b | 1980 | addr_set_port(&addr[i], port); |
d3c3c0a8 | 1981 | addr[i].type = CEPH_ENTITY_ADDR_TYPE_LEGACY; |
31b8006e | 1982 | |
b726ec97 | 1983 | dout("parse_ips got %s\n", ceph_pr_addr(&addr[i])); |
31b8006e SW |
1984 | |
1985 | if (p == end) | |
1986 | break; | |
1987 | if (*p != ',') | |
1988 | goto bad; | |
1989 | p++; | |
1990 | } | |
1991 | ||
1992 | if (p != end) | |
1993 | goto bad; | |
1994 | ||
1995 | if (count) | |
1996 | *count = i + 1; | |
1997 | return 0; | |
1998 | ||
1999 | bad: | |
ee3b56f2 | 2000 | return ret; |
31b8006e SW |
2001 | } |
2002 | ||
eed0ef2c | 2003 | static int process_banner(struct ceph_connection *con) |
31b8006e | 2004 | { |
eed0ef2c | 2005 | dout("process_banner on %p\n", con); |
31b8006e SW |
2006 | |
2007 | if (verify_hello(con) < 0) | |
2008 | return -1; | |
2009 | ||
2010 | /* | |
2011 | * Make sure the other end is who we wanted. note that the other | |
2012 | * end may not yet know their ip address, so if it's 0.0.0.0, give | |
2013 | * them the benefit of the doubt. | |
2014 | */ | |
103e2d3a SW |
2015 | if (memcmp(&con->peer_addr, &con->actual_peer_addr, |
2016 | sizeof(con->peer_addr)) != 0 && | |
cede185b | 2017 | !(addr_is_blank(&con->actual_peer_addr) && |
31b8006e | 2018 | con->actual_peer_addr.nonce == con->peer_addr.nonce)) { |
a9dfe31e | 2019 | pr_warn("wrong peer, want %s/%u, got %s/%u\n", |
b726ec97 | 2020 | ceph_pr_addr(&con->peer_addr), |
a9dfe31e | 2021 | le32_to_cpu(con->peer_addr.nonce), |
b726ec97 | 2022 | ceph_pr_addr(&con->actual_peer_addr), |
a9dfe31e | 2023 | le32_to_cpu(con->actual_peer_addr.nonce)); |
58bb3b37 | 2024 | con->error_msg = "wrong peer at address"; |
31b8006e SW |
2025 | return -1; |
2026 | } | |
2027 | ||
2028 | /* | |
2029 | * did we learn our address? | |
2030 | */ | |
cede185b JL |
2031 | if (addr_is_blank(&con->msgr->inst.addr)) { |
2032 | int port = addr_port(&con->msgr->inst.addr); | |
31b8006e SW |
2033 | |
2034 | memcpy(&con->msgr->inst.addr.in_addr, | |
2035 | &con->peer_addr_for_me.in_addr, | |
2036 | sizeof(con->peer_addr_for_me.in_addr)); | |
cede185b | 2037 | addr_set_port(&con->msgr->inst.addr, port); |
63f2d211 | 2038 | encode_my_addr(con->msgr); |
eed0ef2c | 2039 | dout("process_banner learned my addr is %s\n", |
b726ec97 | 2040 | ceph_pr_addr(&con->msgr->inst.addr)); |
31b8006e SW |
2041 | } |
2042 | ||
eed0ef2c SW |
2043 | return 0; |
2044 | } | |
2045 | ||
2046 | static int process_connect(struct ceph_connection *con) | |
2047 | { | |
859bff51 ID |
2048 | u64 sup_feat = from_msgr(con->msgr)->supported_features; |
2049 | u64 req_feat = from_msgr(con->msgr)->required_features; | |
dcbbd97c | 2050 | u64 server_feat = le64_to_cpu(con->in_reply.features); |
0da5d703 | 2051 | int ret; |
04a419f9 | 2052 | |
eed0ef2c SW |
2053 | dout("process_connect on %p tag %d\n", con, (int)con->in_tag); |
2054 | ||
262614c4 | 2055 | if (con->auth) { |
0fd3fd0a ID |
2056 | int len = le32_to_cpu(con->in_reply.authorizer_len); |
2057 | ||
5c056fdc ID |
2058 | /* |
2059 | * Any connection that defines ->get_authorizer() | |
6daca13d ID |
2060 | * should also define ->add_authorizer_challenge() and |
2061 | * ->verify_authorizer_reply(). | |
2062 | * | |
5c056fdc ID |
2063 | * See get_connect_authorizer(). |
2064 | */ | |
6daca13d ID |
2065 | if (con->in_reply.tag == CEPH_MSGR_TAG_CHALLENGE_AUTHORIZER) { |
2066 | ret = con->ops->add_authorizer_challenge( | |
0fd3fd0a | 2067 | con, con->auth->authorizer_reply_buf, len); |
6daca13d ID |
2068 | if (ret < 0) |
2069 | return ret; | |
2070 | ||
2071 | con_out_kvec_reset(con); | |
2072 | __prepare_write_connect(con); | |
2073 | prepare_read_connect(con); | |
2074 | return 0; | |
2075 | } | |
2076 | ||
0fd3fd0a ID |
2077 | if (len) { |
2078 | ret = con->ops->verify_authorizer_reply(con); | |
2079 | if (ret < 0) { | |
2080 | con->error_msg = "bad authorize reply"; | |
2081 | return ret; | |
2082 | } | |
5c056fdc ID |
2083 | } |
2084 | } | |
2085 | ||
31b8006e | 2086 | switch (con->in_reply.tag) { |
04a419f9 SW |
2087 | case CEPH_MSGR_TAG_FEATURES: |
2088 | pr_err("%s%lld %s feature set mismatch," | |
2089 | " my %llx < server's %llx, missing %llx\n", | |
2090 | ENTITY_NAME(con->peer_name), | |
b726ec97 | 2091 | ceph_pr_addr(&con->peer_addr), |
04a419f9 SW |
2092 | sup_feat, server_feat, server_feat & ~sup_feat); |
2093 | con->error_msg = "missing required protocol features"; | |
04a419f9 SW |
2094 | return -1; |
2095 | ||
31b8006e | 2096 | case CEPH_MSGR_TAG_BADPROTOVER: |
31b8006e SW |
2097 | pr_err("%s%lld %s protocol version mismatch," |
2098 | " my %d != server's %d\n", | |
2099 | ENTITY_NAME(con->peer_name), | |
b726ec97 | 2100 | ceph_pr_addr(&con->peer_addr), |
31b8006e SW |
2101 | le32_to_cpu(con->out_connect.protocol_version), |
2102 | le32_to_cpu(con->in_reply.protocol_version)); | |
2103 | con->error_msg = "protocol version mismatch"; | |
31b8006e SW |
2104 | return -1; |
2105 | ||
4e7a5dcd SW |
2106 | case CEPH_MSGR_TAG_BADAUTHORIZER: |
2107 | con->auth_retry++; | |
2108 | dout("process_connect %p got BADAUTHORIZER attempt %d\n", con, | |
2109 | con->auth_retry); | |
2110 | if (con->auth_retry == 2) { | |
2111 | con->error_msg = "connect authorization failure"; | |
4e7a5dcd SW |
2112 | return -1; |
2113 | } | |
6d4221b5 | 2114 | con_out_kvec_reset(con); |
e825a66d | 2115 | ret = prepare_write_connect(con); |
0da5d703 SW |
2116 | if (ret < 0) |
2117 | return ret; | |
63733a0f | 2118 | prepare_read_connect(con); |
4e7a5dcd | 2119 | break; |
31b8006e SW |
2120 | |
2121 | case CEPH_MSGR_TAG_RESETSESSION: | |
2122 | /* | |
2123 | * If we connected with a large connect_seq but the peer | |
2124 | * has no record of a session with us (no connection, or | |
2125 | * connect_seq == 0), they will send RESETSESION to indicate | |
2126 | * that they must have reset their session, and may have | |
2127 | * dropped messages. | |
2128 | */ | |
2129 | dout("process_connect got RESET peer seq %u\n", | |
5bdca4e0 | 2130 | le32_to_cpu(con->in_reply.connect_seq)); |
31b8006e SW |
2131 | pr_err("%s%lld %s connection reset\n", |
2132 | ENTITY_NAME(con->peer_name), | |
b726ec97 | 2133 | ceph_pr_addr(&con->peer_addr)); |
31b8006e | 2134 | reset_connection(con); |
6d4221b5 | 2135 | con_out_kvec_reset(con); |
5a0f8fdd AE |
2136 | ret = prepare_write_connect(con); |
2137 | if (ret < 0) | |
2138 | return ret; | |
31b8006e SW |
2139 | prepare_read_connect(con); |
2140 | ||
2141 | /* Tell ceph about it. */ | |
ec302645 | 2142 | mutex_unlock(&con->mutex); |
31b8006e SW |
2143 | pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name)); |
2144 | if (con->ops->peer_reset) | |
2145 | con->ops->peer_reset(con); | |
ec302645 | 2146 | mutex_lock(&con->mutex); |
8dacc7da | 2147 | if (con->state != CON_STATE_NEGOTIATING) |
0da5d703 | 2148 | return -EAGAIN; |
31b8006e SW |
2149 | break; |
2150 | ||
2151 | case CEPH_MSGR_TAG_RETRY_SESSION: | |
2152 | /* | |
2153 | * If we sent a smaller connect_seq than the peer has, try | |
2154 | * again with a larger value. | |
2155 | */ | |
5bdca4e0 | 2156 | dout("process_connect got RETRY_SESSION my seq %u, peer %u\n", |
31b8006e | 2157 | le32_to_cpu(con->out_connect.connect_seq), |
5bdca4e0 SW |
2158 | le32_to_cpu(con->in_reply.connect_seq)); |
2159 | con->connect_seq = le32_to_cpu(con->in_reply.connect_seq); | |
6d4221b5 | 2160 | con_out_kvec_reset(con); |
5a0f8fdd AE |
2161 | ret = prepare_write_connect(con); |
2162 | if (ret < 0) | |
2163 | return ret; | |
31b8006e SW |
2164 | prepare_read_connect(con); |
2165 | break; | |
2166 | ||
2167 | case CEPH_MSGR_TAG_RETRY_GLOBAL: | |
2168 | /* | |
2169 | * If we sent a smaller global_seq than the peer has, try | |
2170 | * again with a larger value. | |
2171 | */ | |
eed0ef2c | 2172 | dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n", |
31b8006e | 2173 | con->peer_global_seq, |
5bdca4e0 | 2174 | le32_to_cpu(con->in_reply.global_seq)); |
31b8006e | 2175 | get_global_seq(con->msgr, |
5bdca4e0 | 2176 | le32_to_cpu(con->in_reply.global_seq)); |
6d4221b5 | 2177 | con_out_kvec_reset(con); |
5a0f8fdd AE |
2178 | ret = prepare_write_connect(con); |
2179 | if (ret < 0) | |
2180 | return ret; | |
31b8006e SW |
2181 | prepare_read_connect(con); |
2182 | break; | |
2183 | ||
3a23083b | 2184 | case CEPH_MSGR_TAG_SEQ: |
31b8006e | 2185 | case CEPH_MSGR_TAG_READY: |
04a419f9 SW |
2186 | if (req_feat & ~server_feat) { |
2187 | pr_err("%s%lld %s protocol feature mismatch," | |
2188 | " my required %llx > server's %llx, need %llx\n", | |
2189 | ENTITY_NAME(con->peer_name), | |
b726ec97 | 2190 | ceph_pr_addr(&con->peer_addr), |
04a419f9 SW |
2191 | req_feat, server_feat, req_feat & ~server_feat); |
2192 | con->error_msg = "missing required protocol features"; | |
04a419f9 SW |
2193 | return -1; |
2194 | } | |
8dacc7da | 2195 | |
122070a2 | 2196 | WARN_ON(con->state != CON_STATE_NEGOTIATING); |
8dacc7da | 2197 | con->state = CON_STATE_OPEN; |
20e55c4c | 2198 | con->auth_retry = 0; /* we authenticated; clear flag */ |
31b8006e SW |
2199 | con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq); |
2200 | con->connect_seq++; | |
aba558e2 | 2201 | con->peer_features = server_feat; |
31b8006e SW |
2202 | dout("process_connect got READY gseq %d cseq %d (%d)\n", |
2203 | con->peer_global_seq, | |
2204 | le32_to_cpu(con->in_reply.connect_seq), | |
2205 | con->connect_seq); | |
2206 | WARN_ON(con->connect_seq != | |
2207 | le32_to_cpu(con->in_reply.connect_seq)); | |
92ac41d0 SW |
2208 | |
2209 | if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY) | |
c9ffc77a | 2210 | con_flag_set(con, CON_FLAG_LOSSYTX); |
92ac41d0 | 2211 | |
85effe18 | 2212 | con->delay = 0; /* reset backoff memory */ |
92ac41d0 | 2213 | |
3a23083b SW |
2214 | if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) { |
2215 | prepare_write_seq(con); | |
2216 | prepare_read_seq(con); | |
2217 | } else { | |
2218 | prepare_read_tag(con); | |
2219 | } | |
31b8006e SW |
2220 | break; |
2221 | ||
2222 | case CEPH_MSGR_TAG_WAIT: | |
2223 | /* | |
2224 | * If there is a connection race (we are opening | |
2225 | * connections to each other), one of us may just have | |
2226 | * to WAIT. This shouldn't happen if we are the | |
2227 | * client. | |
2228 | */ | |
04177882 SW |
2229 | con->error_msg = "protocol error, got WAIT as client"; |
2230 | return -1; | |
31b8006e SW |
2231 | |
2232 | default: | |
31b8006e SW |
2233 | con->error_msg = "protocol error, garbage tag during connect"; |
2234 | return -1; | |
2235 | } | |
2236 | return 0; | |
2237 | } | |
2238 | ||
2239 | ||
2240 | /* | |
2241 | * read (part of) an ack | |
2242 | */ | |
2243 | static int read_partial_ack(struct ceph_connection *con) | |
2244 | { | |
fd51653f AE |
2245 | int size = sizeof (con->in_temp_ack); |
2246 | int end = size; | |
31b8006e | 2247 | |
fd51653f | 2248 | return read_partial(con, end, size, &con->in_temp_ack); |
31b8006e SW |
2249 | } |
2250 | ||
31b8006e SW |
2251 | /* |
2252 | * We can finally discard anything that's been acked. | |
2253 | */ | |
2254 | static void process_ack(struct ceph_connection *con) | |
2255 | { | |
2256 | struct ceph_msg *m; | |
2257 | u64 ack = le64_to_cpu(con->in_temp_ack); | |
2258 | u64 seq; | |
0a2ad541 YZ |
2259 | bool reconnect = (con->in_tag == CEPH_MSGR_TAG_SEQ); |
2260 | struct list_head *list = reconnect ? &con->out_queue : &con->out_sent; | |
31b8006e | 2261 | |
0a2ad541 YZ |
2262 | /* |
2263 | * In the reconnect case, con_fault() has requeued messages | |
2264 | * in out_sent. We should cleanup old messages according to | |
2265 | * the reconnect seq. | |
2266 | */ | |
2267 | while (!list_empty(list)) { | |
2268 | m = list_first_entry(list, struct ceph_msg, list_head); | |
2269 | if (reconnect && m->needs_out_seq) | |
2270 | break; | |
31b8006e SW |
2271 | seq = le64_to_cpu(m->hdr.seq); |
2272 | if (seq > ack) | |
2273 | break; | |
2274 | dout("got ack for seq %llu type %d at %p\n", seq, | |
2275 | le16_to_cpu(m->hdr.type), m); | |
4cf9d544 | 2276 | m->ack_stamp = jiffies; |
31b8006e SW |
2277 | ceph_msg_remove(m); |
2278 | } | |
0a2ad541 | 2279 | |
31b8006e SW |
2280 | prepare_read_tag(con); |
2281 | } | |
2282 | ||
2283 | ||
2450418c | 2284 | static int read_partial_message_section(struct ceph_connection *con, |
213c99ee SW |
2285 | struct kvec *section, |
2286 | unsigned int sec_len, u32 *crc) | |
2450418c | 2287 | { |
68b4476b | 2288 | int ret, left; |
2450418c YS |
2289 | |
2290 | BUG_ON(!section); | |
2291 | ||
2292 | while (section->iov_len < sec_len) { | |
2293 | BUG_ON(section->iov_base == NULL); | |
2294 | left = sec_len - section->iov_len; | |
2295 | ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base + | |
2296 | section->iov_len, left); | |
2297 | if (ret <= 0) | |
2298 | return ret; | |
2299 | section->iov_len += ret; | |
2450418c | 2300 | } |
fe3ad593 AE |
2301 | if (section->iov_len == sec_len) |
2302 | *crc = crc32c(0, section->iov_base, section->iov_len); | |
31b8006e | 2303 | |
2450418c YS |
2304 | return 1; |
2305 | } | |
31b8006e | 2306 | |
34d2d200 AE |
2307 | static int read_partial_msg_data(struct ceph_connection *con) |
2308 | { | |
2309 | struct ceph_msg *msg = con->in_msg; | |
8ae4f4f5 | 2310 | struct ceph_msg_data_cursor *cursor = &msg->cursor; |
859bff51 | 2311 | bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC); |
686be208 AE |
2312 | struct page *page; |
2313 | size_t page_offset; | |
2314 | size_t length; | |
f5db90bc | 2315 | u32 crc = 0; |
34d2d200 AE |
2316 | int ret; |
2317 | ||
0d9c1ab3 | 2318 | if (!msg->num_data_items) |
4c59b4a2 | 2319 | return -EIO; |
34d2d200 | 2320 | |
f5db90bc AE |
2321 | if (do_datacrc) |
2322 | crc = con->in_data_crc; | |
45a267db ID |
2323 | while (cursor->total_resid) { |
2324 | if (!cursor->resid) { | |
2325 | ceph_msg_data_advance(cursor, 0); | |
2326 | continue; | |
2327 | } | |
2328 | ||
343128ce | 2329 | page = ceph_msg_data_next(cursor, &page_offset, &length, NULL); |
686be208 | 2330 | ret = ceph_tcp_recvpage(con->sock, page, page_offset, length); |
f5db90bc AE |
2331 | if (ret <= 0) { |
2332 | if (do_datacrc) | |
2333 | con->in_data_crc = crc; | |
2334 | ||
686be208 | 2335 | return ret; |
f5db90bc | 2336 | } |
686be208 AE |
2337 | |
2338 | if (do_datacrc) | |
f5db90bc | 2339 | crc = ceph_crc32c_page(crc, page, page_offset, ret); |
1759f7b0 | 2340 | ceph_msg_data_advance(cursor, (size_t)ret); |
34d2d200 | 2341 | } |
f5db90bc AE |
2342 | if (do_datacrc) |
2343 | con->in_data_crc = crc; | |
34d2d200 AE |
2344 | |
2345 | return 1; /* must return > 0 to indicate success */ | |
2346 | } | |
2347 | ||
31b8006e SW |
2348 | /* |
2349 | * read (part of) a message. | |
2350 | */ | |
686be208 AE |
2351 | static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip); |
2352 | ||
31b8006e SW |
2353 | static int read_partial_message(struct ceph_connection *con) |
2354 | { | |
2355 | struct ceph_msg *m = con->in_msg; | |
fd51653f AE |
2356 | int size; |
2357 | int end; | |
31b8006e | 2358 | int ret; |
95c96174 | 2359 | unsigned int front_len, middle_len, data_len; |
859bff51 | 2360 | bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC); |
33d07337 | 2361 | bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH); |
ae18756b | 2362 | u64 seq; |
fe3ad593 | 2363 | u32 crc; |
31b8006e SW |
2364 | |
2365 | dout("read_partial_message con %p msg %p\n", con, m); | |
2366 | ||
2367 | /* header */ | |
fd51653f AE |
2368 | size = sizeof (con->in_hdr); |
2369 | end = size; | |
2370 | ret = read_partial(con, end, size, &con->in_hdr); | |
57dac9d1 AE |
2371 | if (ret <= 0) |
2372 | return ret; | |
fe3ad593 AE |
2373 | |
2374 | crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc)); | |
2375 | if (cpu_to_le32(crc) != con->in_hdr.crc) { | |
67c64eb7 | 2376 | pr_err("read_partial_message bad hdr crc %u != expected %u\n", |
fe3ad593 AE |
2377 | crc, con->in_hdr.crc); |
2378 | return -EBADMSG; | |
2379 | } | |
2380 | ||
31b8006e SW |
2381 | front_len = le32_to_cpu(con->in_hdr.front_len); |
2382 | if (front_len > CEPH_MSG_MAX_FRONT_LEN) | |
2383 | return -EIO; | |
2384 | middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
7b11ba37 | 2385 | if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN) |
31b8006e SW |
2386 | return -EIO; |
2387 | data_len = le32_to_cpu(con->in_hdr.data_len); | |
2388 | if (data_len > CEPH_MSG_MAX_DATA_LEN) | |
2389 | return -EIO; | |
2390 | ||
ae18756b SW |
2391 | /* verify seq# */ |
2392 | seq = le64_to_cpu(con->in_hdr.seq); | |
2393 | if ((s64)seq - (s64)con->in_seq < 1) { | |
df9f86fa | 2394 | pr_info("skipping %s%lld %s seq %lld expected %lld\n", |
ae18756b | 2395 | ENTITY_NAME(con->peer_name), |
b726ec97 | 2396 | ceph_pr_addr(&con->peer_addr), |
ae18756b SW |
2397 | seq, con->in_seq + 1); |
2398 | con->in_base_pos = -front_len - middle_len - data_len - | |
dbc0d3ca | 2399 | sizeof_footer(con); |
ae18756b | 2400 | con->in_tag = CEPH_MSGR_TAG_READY; |
e7a88e82 | 2401 | return 1; |
ae18756b SW |
2402 | } else if ((s64)seq - (s64)con->in_seq > 1) { |
2403 | pr_err("read_partial_message bad seq %lld expected %lld\n", | |
2404 | seq, con->in_seq + 1); | |
2405 | con->error_msg = "bad message sequence # for incoming message"; | |
67c64eb7 | 2406 | return -EBADE; |
ae18756b SW |
2407 | } |
2408 | ||
31b8006e SW |
2409 | /* allocate message? */ |
2410 | if (!con->in_msg) { | |
4740a623 SW |
2411 | int skip = 0; |
2412 | ||
31b8006e | 2413 | dout("got hdr type %d front %d data %d\n", con->in_hdr.type, |
6ebc8b32 | 2414 | front_len, data_len); |
4740a623 SW |
2415 | ret = ceph_con_in_msg_alloc(con, &skip); |
2416 | if (ret < 0) | |
2417 | return ret; | |
f759ebb9 AE |
2418 | |
2419 | BUG_ON(!con->in_msg ^ skip); | |
2450418c | 2420 | if (skip) { |
31b8006e | 2421 | /* skip this message */ |
a79832f2 | 2422 | dout("alloc_msg said skip message\n"); |
31b8006e | 2423 | con->in_base_pos = -front_len - middle_len - data_len - |
dbc0d3ca | 2424 | sizeof_footer(con); |
31b8006e | 2425 | con->in_tag = CEPH_MSGR_TAG_READY; |
684be25c | 2426 | con->in_seq++; |
e7a88e82 | 2427 | return 1; |
31b8006e | 2428 | } |
38941f80 | 2429 | |
4740a623 | 2430 | BUG_ON(!con->in_msg); |
38941f80 | 2431 | BUG_ON(con->in_msg->con != con); |
31b8006e SW |
2432 | m = con->in_msg; |
2433 | m->front.iov_len = 0; /* haven't read it yet */ | |
2450418c YS |
2434 | if (m->middle) |
2435 | m->middle->vec.iov_len = 0; | |
9d7f0f13 | 2436 | |
78625051 | 2437 | /* prepare for data payload, if any */ |
a4107026 | 2438 | |
78625051 | 2439 | if (data_len) |
98fa5dd8 | 2440 | prepare_message_data(con->in_msg, data_len); |
31b8006e SW |
2441 | } |
2442 | ||
2443 | /* front */ | |
2450418c YS |
2444 | ret = read_partial_message_section(con, &m->front, front_len, |
2445 | &con->in_front_crc); | |
2446 | if (ret <= 0) | |
2447 | return ret; | |
31b8006e SW |
2448 | |
2449 | /* middle */ | |
2450418c | 2450 | if (m->middle) { |
213c99ee SW |
2451 | ret = read_partial_message_section(con, &m->middle->vec, |
2452 | middle_len, | |
2450418c | 2453 | &con->in_middle_crc); |
31b8006e SW |
2454 | if (ret <= 0) |
2455 | return ret; | |
31b8006e SW |
2456 | } |
2457 | ||
2458 | /* (page) data */ | |
34d2d200 AE |
2459 | if (data_len) { |
2460 | ret = read_partial_msg_data(con); | |
2461 | if (ret <= 0) | |
2462 | return ret; | |
31b8006e SW |
2463 | } |
2464 | ||
31b8006e | 2465 | /* footer */ |
89f08173 | 2466 | size = sizeof_footer(con); |
fd51653f AE |
2467 | end += size; |
2468 | ret = read_partial(con, end, size, &m->footer); | |
57dac9d1 AE |
2469 | if (ret <= 0) |
2470 | return ret; | |
2471 | ||
33d07337 YZ |
2472 | if (!need_sign) { |
2473 | m->footer.flags = m->old_footer.flags; | |
2474 | m->footer.sig = 0; | |
2475 | } | |
2476 | ||
31b8006e SW |
2477 | dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n", |
2478 | m, front_len, m->footer.front_crc, middle_len, | |
2479 | m->footer.middle_crc, data_len, m->footer.data_crc); | |
2480 | ||
2481 | /* crc ok? */ | |
2482 | if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) { | |
2483 | pr_err("read_partial_message %p front crc %u != exp. %u\n", | |
2484 | m, con->in_front_crc, m->footer.front_crc); | |
2485 | return -EBADMSG; | |
2486 | } | |
2487 | if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) { | |
2488 | pr_err("read_partial_message %p middle crc %u != exp %u\n", | |
2489 | m, con->in_middle_crc, m->footer.middle_crc); | |
2490 | return -EBADMSG; | |
2491 | } | |
bca064d2 | 2492 | if (do_datacrc && |
31b8006e SW |
2493 | (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 && |
2494 | con->in_data_crc != le32_to_cpu(m->footer.data_crc)) { | |
2495 | pr_err("read_partial_message %p data crc %u != exp. %u\n", m, | |
2496 | con->in_data_crc, le32_to_cpu(m->footer.data_crc)); | |
2497 | return -EBADMSG; | |
2498 | } | |
2499 | ||
33d07337 | 2500 | if (need_sign && con->ops->check_message_signature && |
79dbd1ba | 2501 | con->ops->check_message_signature(m)) { |
33d07337 YZ |
2502 | pr_err("read_partial_message %p signature check failed\n", m); |
2503 | return -EBADMSG; | |
2504 | } | |
2505 | ||
31b8006e SW |
2506 | return 1; /* done! */ |
2507 | } | |
2508 | ||
2509 | /* | |
2510 | * Process message. This happens in the worker thread. The callback should | |
2511 | * be careful not to do anything that waits on other incoming messages or it | |
2512 | * may deadlock. | |
2513 | */ | |
2514 | static void process_message(struct ceph_connection *con) | |
2515 | { | |
583d0fef | 2516 | struct ceph_msg *msg = con->in_msg; |
31b8006e | 2517 | |
38941f80 | 2518 | BUG_ON(con->in_msg->con != con); |
31b8006e SW |
2519 | con->in_msg = NULL; |
2520 | ||
2521 | /* if first message, set peer_name */ | |
2522 | if (con->peer_name.type == 0) | |
dbad185d | 2523 | con->peer_name = msg->hdr.src; |
31b8006e | 2524 | |
31b8006e | 2525 | con->in_seq++; |
ec302645 | 2526 | mutex_unlock(&con->mutex); |
31b8006e | 2527 | |
b77f8f0e | 2528 | dout("===== %p %llu from %s%lld %d=%s len %d+%d+%d (%u %u %u) =====\n", |
31b8006e | 2529 | msg, le64_to_cpu(msg->hdr.seq), |
dbad185d | 2530 | ENTITY_NAME(msg->hdr.src), |
31b8006e SW |
2531 | le16_to_cpu(msg->hdr.type), |
2532 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
2533 | le32_to_cpu(msg->hdr.front_len), | |
b77f8f0e | 2534 | le32_to_cpu(msg->hdr.middle_len), |
31b8006e SW |
2535 | le32_to_cpu(msg->hdr.data_len), |
2536 | con->in_front_crc, con->in_middle_crc, con->in_data_crc); | |
2537 | con->ops->dispatch(con, msg); | |
ec302645 SW |
2538 | |
2539 | mutex_lock(&con->mutex); | |
31b8006e SW |
2540 | } |
2541 | ||
8b9558aa YZ |
2542 | static int read_keepalive_ack(struct ceph_connection *con) |
2543 | { | |
2544 | struct ceph_timespec ceph_ts; | |
2545 | size_t size = sizeof(ceph_ts); | |
2546 | int ret = read_partial(con, size, size, &ceph_ts); | |
2547 | if (ret <= 0) | |
2548 | return ret; | |
473bd2d7 | 2549 | ceph_decode_timespec64(&con->last_keepalive_ack, &ceph_ts); |
8b9558aa YZ |
2550 | prepare_read_tag(con); |
2551 | return 1; | |
2552 | } | |
31b8006e SW |
2553 | |
2554 | /* | |
2555 | * Write something to the socket. Called in a worker thread when the | |
2556 | * socket appears to be writeable and we have something ready to send. | |
2557 | */ | |
2558 | static int try_write(struct ceph_connection *con) | |
2559 | { | |
31b8006e SW |
2560 | int ret = 1; |
2561 | ||
d59315ca | 2562 | dout("try_write start %p state %lu\n", con, con->state); |
9c55ad1c ID |
2563 | if (con->state != CON_STATE_PREOPEN && |
2564 | con->state != CON_STATE_CONNECTING && | |
2565 | con->state != CON_STATE_NEGOTIATING && | |
2566 | con->state != CON_STATE_OPEN) | |
2567 | return 0; | |
31b8006e | 2568 | |
31b8006e | 2569 | /* open the socket first? */ |
8dacc7da SW |
2570 | if (con->state == CON_STATE_PREOPEN) { |
2571 | BUG_ON(con->sock); | |
2572 | con->state = CON_STATE_CONNECTING; | |
a5988c49 | 2573 | |
e2200423 | 2574 | con_out_kvec_reset(con); |
e825a66d | 2575 | prepare_write_banner(con); |
eed0ef2c | 2576 | prepare_read_banner(con); |
31b8006e | 2577 | |
cf3e5c40 | 2578 | BUG_ON(con->in_msg); |
31b8006e SW |
2579 | con->in_tag = CEPH_MSGR_TAG_READY; |
2580 | dout("try_write initiating connect on %p new state %lu\n", | |
2581 | con, con->state); | |
41617d0c AE |
2582 | ret = ceph_tcp_connect(con); |
2583 | if (ret < 0) { | |
31b8006e | 2584 | con->error_msg = "connect error"; |
31b8006e SW |
2585 | goto out; |
2586 | } | |
2587 | } | |
2588 | ||
d2935d6f ID |
2589 | more: |
2590 | dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes); | |
9c55ad1c ID |
2591 | BUG_ON(!con->sock); |
2592 | ||
31b8006e | 2593 | /* kvec data queued? */ |
67645d76 ID |
2594 | if (con->out_kvec_left) { |
2595 | ret = write_partial_kvec(con); | |
31b8006e | 2596 | if (ret <= 0) |
42961d23 | 2597 | goto out; |
31b8006e | 2598 | } |
67645d76 ID |
2599 | if (con->out_skip) { |
2600 | ret = write_partial_skip(con); | |
31b8006e | 2601 | if (ret <= 0) |
42961d23 | 2602 | goto out; |
31b8006e SW |
2603 | } |
2604 | ||
2605 | /* msg pages? */ | |
2606 | if (con->out_msg) { | |
c86a2930 SW |
2607 | if (con->out_msg_done) { |
2608 | ceph_msg_put(con->out_msg); | |
2609 | con->out_msg = NULL; /* we're done with this one */ | |
2610 | goto do_next; | |
2611 | } | |
2612 | ||
34d2d200 | 2613 | ret = write_partial_message_data(con); |
31b8006e | 2614 | if (ret == 1) |
d2935d6f | 2615 | goto more; /* we need to send the footer, too! */ |
31b8006e | 2616 | if (ret == 0) |
42961d23 | 2617 | goto out; |
31b8006e | 2618 | if (ret < 0) { |
34d2d200 | 2619 | dout("try_write write_partial_message_data err %d\n", |
31b8006e | 2620 | ret); |
42961d23 | 2621 | goto out; |
31b8006e SW |
2622 | } |
2623 | } | |
2624 | ||
c86a2930 | 2625 | do_next: |
8dacc7da | 2626 | if (con->state == CON_STATE_OPEN) { |
8b9558aa YZ |
2627 | if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) { |
2628 | prepare_write_keepalive(con); | |
2629 | goto more; | |
2630 | } | |
31b8006e SW |
2631 | /* is anything else pending? */ |
2632 | if (!list_empty(&con->out_queue)) { | |
2633 | prepare_write_message(con); | |
2634 | goto more; | |
2635 | } | |
2636 | if (con->in_seq > con->in_seq_acked) { | |
2637 | prepare_write_ack(con); | |
2638 | goto more; | |
2639 | } | |
31b8006e SW |
2640 | } |
2641 | ||
2642 | /* Nothing to do! */ | |
c9ffc77a | 2643 | con_flag_clear(con, CON_FLAG_WRITE_PENDING); |
31b8006e | 2644 | dout("try_write nothing else to write.\n"); |
31b8006e SW |
2645 | ret = 0; |
2646 | out: | |
42961d23 | 2647 | dout("try_write done on %p ret %d\n", con, ret); |
31b8006e SW |
2648 | return ret; |
2649 | } | |
2650 | ||
31b8006e SW |
2651 | /* |
2652 | * Read what we can from the socket. | |
2653 | */ | |
2654 | static int try_read(struct ceph_connection *con) | |
2655 | { | |
31b8006e SW |
2656 | int ret = -1; |
2657 | ||
8dacc7da SW |
2658 | more: |
2659 | dout("try_read start on %p state %lu\n", con, con->state); | |
2660 | if (con->state != CON_STATE_CONNECTING && | |
2661 | con->state != CON_STATE_NEGOTIATING && | |
2662 | con->state != CON_STATE_OPEN) | |
31b8006e SW |
2663 | return 0; |
2664 | ||
8dacc7da | 2665 | BUG_ON(!con->sock); |
ec302645 | 2666 | |
31b8006e SW |
2667 | dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag, |
2668 | con->in_base_pos); | |
0da5d703 | 2669 | |
8dacc7da | 2670 | if (con->state == CON_STATE_CONNECTING) { |
7593af92 AE |
2671 | dout("try_read connecting\n"); |
2672 | ret = read_partial_banner(con); | |
2673 | if (ret <= 0) | |
ab166d5a | 2674 | goto out; |
7593af92 AE |
2675 | ret = process_banner(con); |
2676 | if (ret < 0) | |
2677 | goto out; | |
2678 | ||
8dacc7da | 2679 | con->state = CON_STATE_NEGOTIATING; |
7593af92 | 2680 | |
6d4221b5 JS |
2681 | /* |
2682 | * Received banner is good, exchange connection info. | |
2683 | * Do not reset out_kvec, as sending our banner raced | |
2684 | * with receiving peer banner after connect completed. | |
2685 | */ | |
7593af92 AE |
2686 | ret = prepare_write_connect(con); |
2687 | if (ret < 0) | |
2688 | goto out; | |
2689 | prepare_read_connect(con); | |
2690 | ||
2691 | /* Send connection info before awaiting response */ | |
0da5d703 SW |
2692 | goto out; |
2693 | } | |
2694 | ||
8dacc7da | 2695 | if (con->state == CON_STATE_NEGOTIATING) { |
7593af92 | 2696 | dout("try_read negotiating\n"); |
31b8006e SW |
2697 | ret = read_partial_connect(con); |
2698 | if (ret <= 0) | |
31b8006e | 2699 | goto out; |
98bdb0aa SW |
2700 | ret = process_connect(con); |
2701 | if (ret < 0) | |
2702 | goto out; | |
31b8006e SW |
2703 | goto more; |
2704 | } | |
2705 | ||
122070a2 | 2706 | WARN_ON(con->state != CON_STATE_OPEN); |
8dacc7da | 2707 | |
31b8006e SW |
2708 | if (con->in_base_pos < 0) { |
2709 | /* | |
2710 | * skipping + discarding content. | |
31b8006e | 2711 | */ |
e5c93883 | 2712 | ret = ceph_tcp_recvmsg(con->sock, NULL, -con->in_base_pos); |
31b8006e | 2713 | if (ret <= 0) |
98bdb0aa | 2714 | goto out; |
e5c93883 | 2715 | dout("skipped %d / %d bytes\n", ret, -con->in_base_pos); |
31b8006e SW |
2716 | con->in_base_pos += ret; |
2717 | if (con->in_base_pos) | |
2718 | goto more; | |
2719 | } | |
2720 | if (con->in_tag == CEPH_MSGR_TAG_READY) { | |
2721 | /* | |
2722 | * what's next? | |
2723 | */ | |
2724 | ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1); | |
2725 | if (ret <= 0) | |
98bdb0aa | 2726 | goto out; |
31b8006e SW |
2727 | dout("try_read got tag %d\n", (int)con->in_tag); |
2728 | switch (con->in_tag) { | |
2729 | case CEPH_MSGR_TAG_MSG: | |
2730 | prepare_read_message(con); | |
2731 | break; | |
2732 | case CEPH_MSGR_TAG_ACK: | |
2733 | prepare_read_ack(con); | |
2734 | break; | |
8b9558aa YZ |
2735 | case CEPH_MSGR_TAG_KEEPALIVE2_ACK: |
2736 | prepare_read_keepalive_ack(con); | |
2737 | break; | |
31b8006e | 2738 | case CEPH_MSGR_TAG_CLOSE: |
8dacc7da SW |
2739 | con_close_socket(con); |
2740 | con->state = CON_STATE_CLOSED; | |
98bdb0aa | 2741 | goto out; |
31b8006e SW |
2742 | default: |
2743 | goto bad_tag; | |
2744 | } | |
2745 | } | |
2746 | if (con->in_tag == CEPH_MSGR_TAG_MSG) { | |
2747 | ret = read_partial_message(con); | |
2748 | if (ret <= 0) { | |
2749 | switch (ret) { | |
2750 | case -EBADMSG: | |
a51983e4 | 2751 | con->error_msg = "bad crc/signature"; |
df561f66 | 2752 | fallthrough; |
67c64eb7 | 2753 | case -EBADE: |
31b8006e | 2754 | ret = -EIO; |
98bdb0aa | 2755 | break; |
31b8006e SW |
2756 | case -EIO: |
2757 | con->error_msg = "io error"; | |
98bdb0aa | 2758 | break; |
31b8006e | 2759 | } |
98bdb0aa | 2760 | goto out; |
31b8006e SW |
2761 | } |
2762 | if (con->in_tag == CEPH_MSGR_TAG_READY) | |
2763 | goto more; | |
2764 | process_message(con); | |
7b862e07 SW |
2765 | if (con->state == CON_STATE_OPEN) |
2766 | prepare_read_tag(con); | |
31b8006e SW |
2767 | goto more; |
2768 | } | |
3a23083b SW |
2769 | if (con->in_tag == CEPH_MSGR_TAG_ACK || |
2770 | con->in_tag == CEPH_MSGR_TAG_SEQ) { | |
2771 | /* | |
2772 | * the final handshake seq exchange is semantically | |
2773 | * equivalent to an ACK | |
2774 | */ | |
31b8006e SW |
2775 | ret = read_partial_ack(con); |
2776 | if (ret <= 0) | |
98bdb0aa | 2777 | goto out; |
31b8006e SW |
2778 | process_ack(con); |
2779 | goto more; | |
2780 | } | |
8b9558aa YZ |
2781 | if (con->in_tag == CEPH_MSGR_TAG_KEEPALIVE2_ACK) { |
2782 | ret = read_keepalive_ack(con); | |
2783 | if (ret <= 0) | |
2784 | goto out; | |
2785 | goto more; | |
2786 | } | |
31b8006e | 2787 | |
31b8006e | 2788 | out: |
98bdb0aa | 2789 | dout("try_read done on %p ret %d\n", con, ret); |
31b8006e SW |
2790 | return ret; |
2791 | ||
2792 | bad_tag: | |
2793 | pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag); | |
2794 | con->error_msg = "protocol error, garbage tag"; | |
2795 | ret = -1; | |
2796 | goto out; | |
2797 | } | |
2798 | ||
2799 | ||
2800 | /* | |
802c6d96 AE |
2801 | * Atomically queue work on a connection after the specified delay. |
2802 | * Bump @con reference to avoid races with connection teardown. | |
2803 | * Returns 0 if work was queued, or an error code otherwise. | |
31b8006e | 2804 | */ |
802c6d96 | 2805 | static int queue_con_delay(struct ceph_connection *con, unsigned long delay) |
31b8006e | 2806 | { |
31b8006e | 2807 | if (!con->ops->get(con)) { |
802c6d96 | 2808 | dout("%s %p ref count 0\n", __func__, con); |
802c6d96 | 2809 | return -ENOENT; |
31b8006e SW |
2810 | } |
2811 | ||
418af5b3 ID |
2812 | if (delay >= HZ) |
2813 | delay = round_jiffies_relative(delay); | |
2814 | ||
5a5036c8 | 2815 | dout("%s %p %lu\n", __func__, con, delay); |
802c6d96 AE |
2816 | if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) { |
2817 | dout("%s %p - already queued\n", __func__, con); | |
31b8006e | 2818 | con->ops->put(con); |
802c6d96 | 2819 | return -EBUSY; |
31b8006e | 2820 | } |
802c6d96 | 2821 | |
802c6d96 AE |
2822 | return 0; |
2823 | } | |
2824 | ||
2825 | static void queue_con(struct ceph_connection *con) | |
2826 | { | |
2827 | (void) queue_con_delay(con, 0); | |
31b8006e SW |
2828 | } |
2829 | ||
37ab77ac ID |
2830 | static void cancel_con(struct ceph_connection *con) |
2831 | { | |
2832 | if (cancel_delayed_work(&con->work)) { | |
2833 | dout("%s %p\n", __func__, con); | |
2834 | con->ops->put(con); | |
2835 | } | |
2836 | } | |
2837 | ||
7bb21d68 AE |
2838 | static bool con_sock_closed(struct ceph_connection *con) |
2839 | { | |
c9ffc77a | 2840 | if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED)) |
7bb21d68 AE |
2841 | return false; |
2842 | ||
2843 | #define CASE(x) \ | |
2844 | case CON_STATE_ ## x: \ | |
2845 | con->error_msg = "socket closed (con state " #x ")"; \ | |
2846 | break; | |
2847 | ||
2848 | switch (con->state) { | |
2849 | CASE(CLOSED); | |
2850 | CASE(PREOPEN); | |
2851 | CASE(CONNECTING); | |
2852 | CASE(NEGOTIATING); | |
2853 | CASE(OPEN); | |
2854 | CASE(STANDBY); | |
2855 | default: | |
b9a67899 | 2856 | pr_warn("%s con %p unrecognized state %lu\n", |
7bb21d68 AE |
2857 | __func__, con, con->state); |
2858 | con->error_msg = "unrecognized con state"; | |
2859 | BUG(); | |
2860 | break; | |
2861 | } | |
2862 | #undef CASE | |
2863 | ||
2864 | return true; | |
2865 | } | |
2866 | ||
f20a39fd AE |
2867 | static bool con_backoff(struct ceph_connection *con) |
2868 | { | |
2869 | int ret; | |
2870 | ||
2871 | if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF)) | |
2872 | return false; | |
2873 | ||
418af5b3 | 2874 | ret = queue_con_delay(con, con->delay); |
f20a39fd AE |
2875 | if (ret) { |
2876 | dout("%s: con %p FAILED to back off %lu\n", __func__, | |
2877 | con, con->delay); | |
2878 | BUG_ON(ret == -ENOENT); | |
2879 | con_flag_set(con, CON_FLAG_BACKOFF); | |
2880 | } | |
2881 | ||
2882 | return true; | |
2883 | } | |
2884 | ||
93209264 AE |
2885 | /* Finish fault handling; con->mutex must *not* be held here */ |
2886 | ||
2887 | static void con_fault_finish(struct ceph_connection *con) | |
2888 | { | |
f6330cc1 ID |
2889 | dout("%s %p\n", __func__, con); |
2890 | ||
93209264 AE |
2891 | /* |
2892 | * in case we faulted due to authentication, invalidate our | |
2893 | * current tickets so that we can get new ones. | |
2894 | */ | |
f6330cc1 ID |
2895 | if (con->auth_retry) { |
2896 | dout("auth_retry %d, invalidating\n", con->auth_retry); | |
2897 | if (con->ops->invalidate_authorizer) | |
2898 | con->ops->invalidate_authorizer(con); | |
2899 | con->auth_retry = 0; | |
93209264 AE |
2900 | } |
2901 | ||
2902 | if (con->ops->fault) | |
2903 | con->ops->fault(con); | |
2904 | } | |
2905 | ||
31b8006e SW |
2906 | /* |
2907 | * Do some work on a connection. Drop a connection ref when we're done. | |
2908 | */ | |
68931622 | 2909 | static void ceph_con_workfn(struct work_struct *work) |
31b8006e SW |
2910 | { |
2911 | struct ceph_connection *con = container_of(work, struct ceph_connection, | |
2912 | work.work); | |
49659416 | 2913 | bool fault; |
31b8006e | 2914 | |
9dd4658d | 2915 | mutex_lock(&con->mutex); |
49659416 AE |
2916 | while (true) { |
2917 | int ret; | |
31b8006e | 2918 | |
49659416 AE |
2919 | if ((fault = con_sock_closed(con))) { |
2920 | dout("%s: con %p SOCK_CLOSED\n", __func__, con); | |
2921 | break; | |
2922 | } | |
2923 | if (con_backoff(con)) { | |
2924 | dout("%s: con %p BACKOFF\n", __func__, con); | |
2925 | break; | |
2926 | } | |
2927 | if (con->state == CON_STATE_STANDBY) { | |
2928 | dout("%s: con %p STANDBY\n", __func__, con); | |
2929 | break; | |
2930 | } | |
2931 | if (con->state == CON_STATE_CLOSED) { | |
2932 | dout("%s: con %p CLOSED\n", __func__, con); | |
2933 | BUG_ON(con->sock); | |
2934 | break; | |
2935 | } | |
2936 | if (con->state == CON_STATE_PREOPEN) { | |
2937 | dout("%s: con %p PREOPEN\n", __func__, con); | |
2938 | BUG_ON(con->sock); | |
2939 | } | |
0da5d703 | 2940 | |
49659416 AE |
2941 | ret = try_read(con); |
2942 | if (ret < 0) { | |
2943 | if (ret == -EAGAIN) | |
2944 | continue; | |
67c64eb7 ID |
2945 | if (!con->error_msg) |
2946 | con->error_msg = "socket error on read"; | |
49659416 AE |
2947 | fault = true; |
2948 | break; | |
2949 | } | |
2950 | ||
2951 | ret = try_write(con); | |
2952 | if (ret < 0) { | |
2953 | if (ret == -EAGAIN) | |
2954 | continue; | |
67c64eb7 ID |
2955 | if (!con->error_msg) |
2956 | con->error_msg = "socket error on write"; | |
49659416 AE |
2957 | fault = true; |
2958 | } | |
2959 | ||
2960 | break; /* If we make it to here, we're done */ | |
3a140a0d | 2961 | } |
b6e7b6a1 AE |
2962 | if (fault) |
2963 | con_fault(con); | |
9dd4658d | 2964 | mutex_unlock(&con->mutex); |
0da5d703 | 2965 | |
b6e7b6a1 AE |
2966 | if (fault) |
2967 | con_fault_finish(con); | |
2968 | ||
2969 | con->ops->put(con); | |
31b8006e SW |
2970 | } |
2971 | ||
31b8006e SW |
2972 | /* |
2973 | * Generic error/fault handler. A retry mechanism is used with | |
2974 | * exponential backoff | |
2975 | */ | |
93209264 | 2976 | static void con_fault(struct ceph_connection *con) |
31b8006e | 2977 | { |
31b8006e | 2978 | dout("fault %p state %lu to peer %s\n", |
b726ec97 | 2979 | con, con->state, ceph_pr_addr(&con->peer_addr)); |
31b8006e | 2980 | |
67c64eb7 | 2981 | pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name), |
b726ec97 | 2982 | ceph_pr_addr(&con->peer_addr), con->error_msg); |
67c64eb7 ID |
2983 | con->error_msg = NULL; |
2984 | ||
122070a2 | 2985 | WARN_ON(con->state != CON_STATE_CONNECTING && |
8dacc7da SW |
2986 | con->state != CON_STATE_NEGOTIATING && |
2987 | con->state != CON_STATE_OPEN); | |
ec302645 | 2988 | |
31b8006e | 2989 | con_close_socket(con); |
5e095e8b | 2990 | |
c9ffc77a | 2991 | if (con_flag_test(con, CON_FLAG_LOSSYTX)) { |
8dacc7da SW |
2992 | dout("fault on LOSSYTX channel, marking CLOSED\n"); |
2993 | con->state = CON_STATE_CLOSED; | |
93209264 | 2994 | return; |
3b5ede07 SW |
2995 | } |
2996 | ||
5e095e8b | 2997 | if (con->in_msg) { |
38941f80 | 2998 | BUG_ON(con->in_msg->con != con); |
5e095e8b SW |
2999 | ceph_msg_put(con->in_msg); |
3000 | con->in_msg = NULL; | |
3001 | } | |
28e1581c ID |
3002 | if (con->out_msg) { |
3003 | BUG_ON(con->out_msg->con != con); | |
3004 | ceph_msg_put(con->out_msg); | |
3005 | con->out_msg = NULL; | |
3006 | } | |
31b8006e | 3007 | |
e80a52d1 SW |
3008 | /* Requeue anything that hasn't been acked */ |
3009 | list_splice_init(&con->out_sent, &con->out_queue); | |
9bd2e6f8 | 3010 | |
e76661d0 SW |
3011 | /* If there are no messages queued or keepalive pending, place |
3012 | * the connection in a STANDBY state */ | |
3013 | if (list_empty(&con->out_queue) && | |
c9ffc77a | 3014 | !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) { |
e00de341 | 3015 | dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con); |
c9ffc77a | 3016 | con_flag_clear(con, CON_FLAG_WRITE_PENDING); |
8dacc7da | 3017 | con->state = CON_STATE_STANDBY; |
e80a52d1 SW |
3018 | } else { |
3019 | /* retry after a delay. */ | |
8dacc7da | 3020 | con->state = CON_STATE_PREOPEN; |
418af5b3 | 3021 | if (!con->delay) { |
e80a52d1 | 3022 | con->delay = BASE_DELAY_INTERVAL; |
418af5b3 | 3023 | } else if (con->delay < MAX_DELAY_INTERVAL) { |
e80a52d1 | 3024 | con->delay *= 2; |
418af5b3 ID |
3025 | if (con->delay > MAX_DELAY_INTERVAL) |
3026 | con->delay = MAX_DELAY_INTERVAL; | |
3027 | } | |
c9ffc77a | 3028 | con_flag_set(con, CON_FLAG_BACKOFF); |
8618e30b | 3029 | queue_con(con); |
31b8006e | 3030 | } |
31b8006e SW |
3031 | } |
3032 | ||
3033 | ||
120a75ea YZ |
3034 | void ceph_messenger_reset_nonce(struct ceph_messenger *msgr) |
3035 | { | |
3036 | u32 nonce = le32_to_cpu(msgr->inst.addr.nonce) + 1000000; | |
3037 | msgr->inst.addr.nonce = cpu_to_le32(nonce); | |
3038 | encode_my_addr(msgr); | |
3039 | } | |
31b8006e SW |
3040 | |
3041 | /* | |
15d9882c | 3042 | * initialize a new messenger instance |
31b8006e | 3043 | */ |
15d9882c | 3044 | void ceph_messenger_init(struct ceph_messenger *msgr, |
859bff51 | 3045 | struct ceph_entity_addr *myaddr) |
31b8006e | 3046 | { |
31b8006e SW |
3047 | spin_lock_init(&msgr->global_seq_lock); |
3048 | ||
31b8006e SW |
3049 | if (myaddr) |
3050 | msgr->inst.addr = *myaddr; | |
3051 | ||
3052 | /* select a random nonce */ | |
ac8839d7 | 3053 | msgr->inst.addr.type = 0; |
103e2d3a | 3054 | get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce)); |
63f2d211 | 3055 | encode_my_addr(msgr); |
31b8006e | 3056 | |
a2a32584 | 3057 | atomic_set(&msgr->stopping, 0); |
757856d2 | 3058 | write_pnet(&msgr->net, get_net(current->nsproxy->net_ns)); |
31b8006e | 3059 | |
15d9882c | 3060 | dout("%s %p\n", __func__, msgr); |
31b8006e | 3061 | } |
15d9882c | 3062 | EXPORT_SYMBOL(ceph_messenger_init); |
31b8006e | 3063 | |
757856d2 ID |
3064 | void ceph_messenger_fini(struct ceph_messenger *msgr) |
3065 | { | |
3066 | put_net(read_pnet(&msgr->net)); | |
3067 | } | |
3068 | EXPORT_SYMBOL(ceph_messenger_fini); | |
3069 | ||
583d0fef ID |
3070 | static void msg_con_set(struct ceph_msg *msg, struct ceph_connection *con) |
3071 | { | |
3072 | if (msg->con) | |
3073 | msg->con->ops->put(msg->con); | |
3074 | ||
3075 | msg->con = con ? con->ops->get(con) : NULL; | |
3076 | BUG_ON(msg->con != con); | |
3077 | } | |
3078 | ||
e00de341 SW |
3079 | static void clear_standby(struct ceph_connection *con) |
3080 | { | |
3081 | /* come back from STANDBY? */ | |
8dacc7da | 3082 | if (con->state == CON_STATE_STANDBY) { |
e00de341 | 3083 | dout("clear_standby %p and ++connect_seq\n", con); |
8dacc7da | 3084 | con->state = CON_STATE_PREOPEN; |
e00de341 | 3085 | con->connect_seq++; |
c9ffc77a AE |
3086 | WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING)); |
3087 | WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)); | |
e00de341 SW |
3088 | } |
3089 | } | |
3090 | ||
31b8006e SW |
3091 | /* |
3092 | * Queue up an outgoing message on the given connection. | |
3093 | */ | |
3094 | void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg) | |
3095 | { | |
31b8006e | 3096 | /* set src+dst */ |
dbad185d | 3097 | msg->hdr.src = con->msgr->inst.name; |
3ca02ef9 | 3098 | BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len)); |
e84346b7 SW |
3099 | msg->needs_out_seq = true; |
3100 | ||
ec302645 | 3101 | mutex_lock(&con->mutex); |
92ce034b | 3102 | |
8dacc7da | 3103 | if (con->state == CON_STATE_CLOSED) { |
a59b55a6 SW |
3104 | dout("con_send %p closed, dropping %p\n", con, msg); |
3105 | ceph_msg_put(msg); | |
3106 | mutex_unlock(&con->mutex); | |
3107 | return; | |
3108 | } | |
3109 | ||
583d0fef | 3110 | msg_con_set(msg, con); |
92ce034b | 3111 | |
31b8006e SW |
3112 | BUG_ON(!list_empty(&msg->list_head)); |
3113 | list_add_tail(&msg->list_head, &con->out_queue); | |
3114 | dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg, | |
3115 | ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type), | |
3116 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
3117 | le32_to_cpu(msg->hdr.front_len), | |
3118 | le32_to_cpu(msg->hdr.middle_len), | |
3119 | le32_to_cpu(msg->hdr.data_len)); | |
00650931 SW |
3120 | |
3121 | clear_standby(con); | |
ec302645 | 3122 | mutex_unlock(&con->mutex); |
31b8006e SW |
3123 | |
3124 | /* if there wasn't anything waiting to send before, queue | |
3125 | * new work */ | |
c9ffc77a | 3126 | if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0) |
31b8006e SW |
3127 | queue_con(con); |
3128 | } | |
3d14c5d2 | 3129 | EXPORT_SYMBOL(ceph_con_send); |
31b8006e SW |
3130 | |
3131 | /* | |
3132 | * Revoke a message that was previously queued for send | |
3133 | */ | |
6740a845 | 3134 | void ceph_msg_revoke(struct ceph_msg *msg) |
31b8006e | 3135 | { |
6740a845 AE |
3136 | struct ceph_connection *con = msg->con; |
3137 | ||
583d0fef ID |
3138 | if (!con) { |
3139 | dout("%s msg %p null con\n", __func__, msg); | |
6740a845 | 3140 | return; /* Message not in our possession */ |
583d0fef | 3141 | } |
6740a845 | 3142 | |
ec302645 | 3143 | mutex_lock(&con->mutex); |
31b8006e | 3144 | if (!list_empty(&msg->list_head)) { |
38941f80 | 3145 | dout("%s %p msg %p - was on queue\n", __func__, con, msg); |
31b8006e | 3146 | list_del_init(&msg->list_head); |
31b8006e | 3147 | msg->hdr.seq = 0; |
38941f80 | 3148 | |
31b8006e | 3149 | ceph_msg_put(msg); |
ed98adad SW |
3150 | } |
3151 | if (con->out_msg == msg) { | |
67645d76 ID |
3152 | BUG_ON(con->out_skip); |
3153 | /* footer */ | |
3154 | if (con->out_msg_done) { | |
3155 | con->out_skip += con_out_kvec_skip(con); | |
3156 | } else { | |
3157 | BUG_ON(!msg->data_length); | |
89f08173 | 3158 | con->out_skip += sizeof_footer(con); |
31b8006e | 3159 | } |
67645d76 ID |
3160 | /* data, middle, front */ |
3161 | if (msg->data_length) | |
3162 | con->out_skip += msg->cursor.total_resid; | |
3163 | if (msg->middle) | |
3164 | con->out_skip += con_out_kvec_skip(con); | |
3165 | con->out_skip += con_out_kvec_skip(con); | |
3166 | ||
3167 | dout("%s %p msg %p - was sending, will write %d skip %d\n", | |
3168 | __func__, con, msg, con->out_kvec_bytes, con->out_skip); | |
ed98adad | 3169 | msg->hdr.seq = 0; |
67645d76 | 3170 | con->out_msg = NULL; |
92ce034b | 3171 | ceph_msg_put(msg); |
31b8006e | 3172 | } |
67645d76 | 3173 | |
ec302645 | 3174 | mutex_unlock(&con->mutex); |
31b8006e SW |
3175 | } |
3176 | ||
350b1c32 | 3177 | /* |
0d59ab81 | 3178 | * Revoke a message that we may be reading data into |
350b1c32 | 3179 | */ |
8921d114 | 3180 | void ceph_msg_revoke_incoming(struct ceph_msg *msg) |
350b1c32 | 3181 | { |
583d0fef | 3182 | struct ceph_connection *con = msg->con; |
8921d114 | 3183 | |
583d0fef | 3184 | if (!con) { |
8921d114 | 3185 | dout("%s msg %p null con\n", __func__, msg); |
8921d114 AE |
3186 | return; /* Message not in our possession */ |
3187 | } | |
3188 | ||
350b1c32 | 3189 | mutex_lock(&con->mutex); |
8921d114 | 3190 | if (con->in_msg == msg) { |
95c96174 ED |
3191 | unsigned int front_len = le32_to_cpu(con->in_hdr.front_len); |
3192 | unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
3193 | unsigned int data_len = le32_to_cpu(con->in_hdr.data_len); | |
350b1c32 SW |
3194 | |
3195 | /* skip rest of message */ | |
8921d114 AE |
3196 | dout("%s %p msg %p revoked\n", __func__, con, msg); |
3197 | con->in_base_pos = con->in_base_pos - | |
350b1c32 | 3198 | sizeof(struct ceph_msg_header) - |
0d59ab81 YS |
3199 | front_len - |
3200 | middle_len - | |
3201 | data_len - | |
350b1c32 | 3202 | sizeof(struct ceph_msg_footer); |
350b1c32 SW |
3203 | ceph_msg_put(con->in_msg); |
3204 | con->in_msg = NULL; | |
3205 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 3206 | con->in_seq++; |
350b1c32 | 3207 | } else { |
8921d114 AE |
3208 | dout("%s %p in_msg %p msg %p no-op\n", |
3209 | __func__, con, con->in_msg, msg); | |
350b1c32 SW |
3210 | } |
3211 | mutex_unlock(&con->mutex); | |
3212 | } | |
3213 | ||
31b8006e SW |
3214 | /* |
3215 | * Queue a keepalive byte to ensure the tcp connection is alive. | |
3216 | */ | |
3217 | void ceph_con_keepalive(struct ceph_connection *con) | |
3218 | { | |
e00de341 | 3219 | dout("con_keepalive %p\n", con); |
00650931 | 3220 | mutex_lock(&con->mutex); |
e00de341 | 3221 | clear_standby(con); |
4aac9228 | 3222 | con_flag_set(con, CON_FLAG_KEEPALIVE_PENDING); |
00650931 | 3223 | mutex_unlock(&con->mutex); |
4aac9228 ID |
3224 | |
3225 | if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0) | |
31b8006e SW |
3226 | queue_con(con); |
3227 | } | |
3d14c5d2 | 3228 | EXPORT_SYMBOL(ceph_con_keepalive); |
31b8006e | 3229 | |
8b9558aa YZ |
3230 | bool ceph_con_keepalive_expired(struct ceph_connection *con, |
3231 | unsigned long interval) | |
3232 | { | |
3233 | if (interval > 0 && | |
3234 | (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) { | |
473bd2d7 AB |
3235 | struct timespec64 now; |
3236 | struct timespec64 ts; | |
3237 | ktime_get_real_ts64(&now); | |
3238 | jiffies_to_timespec64(interval, &ts); | |
3239 | ts = timespec64_add(con->last_keepalive_ack, ts); | |
3240 | return timespec64_compare(&now, &ts) >= 0; | |
8b9558aa YZ |
3241 | } |
3242 | return false; | |
3243 | } | |
3244 | ||
0d9c1ab3 | 3245 | static struct ceph_msg_data *ceph_msg_data_add(struct ceph_msg *msg) |
43794509 | 3246 | { |
0d9c1ab3 ID |
3247 | BUG_ON(msg->num_data_items >= msg->max_data_items); |
3248 | return &msg->data[msg->num_data_items++]; | |
6644ed7b AE |
3249 | } |
3250 | ||
3251 | static void ceph_msg_data_destroy(struct ceph_msg_data *data) | |
3252 | { | |
e8862740 ID |
3253 | if (data->type == CEPH_MSG_DATA_PAGES && data->own_pages) { |
3254 | int num_pages = calc_pages_for(data->alignment, data->length); | |
3255 | ceph_release_page_vector(data->pages, num_pages); | |
3256 | } else if (data->type == CEPH_MSG_DATA_PAGELIST) { | |
6644ed7b | 3257 | ceph_pagelist_release(data->pagelist); |
e8862740 | 3258 | } |
43794509 AE |
3259 | } |
3260 | ||
90af3602 | 3261 | void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages, |
e8862740 | 3262 | size_t length, size_t alignment, bool own_pages) |
02afca6c | 3263 | { |
6644ed7b AE |
3264 | struct ceph_msg_data *data; |
3265 | ||
07aa1558 AE |
3266 | BUG_ON(!pages); |
3267 | BUG_ON(!length); | |
6644ed7b | 3268 | |
0d9c1ab3 ID |
3269 | data = ceph_msg_data_add(msg); |
3270 | data->type = CEPH_MSG_DATA_PAGES; | |
6644ed7b AE |
3271 | data->pages = pages; |
3272 | data->length = length; | |
3273 | data->alignment = alignment & ~PAGE_MASK; | |
e8862740 | 3274 | data->own_pages = own_pages; |
02afca6c | 3275 | |
5240d9f9 | 3276 | msg->data_length += length; |
02afca6c | 3277 | } |
90af3602 | 3278 | EXPORT_SYMBOL(ceph_msg_data_add_pages); |
31b8006e | 3279 | |
90af3602 | 3280 | void ceph_msg_data_add_pagelist(struct ceph_msg *msg, |
27fa8385 AE |
3281 | struct ceph_pagelist *pagelist) |
3282 | { | |
6644ed7b AE |
3283 | struct ceph_msg_data *data; |
3284 | ||
07aa1558 AE |
3285 | BUG_ON(!pagelist); |
3286 | BUG_ON(!pagelist->length); | |
27fa8385 | 3287 | |
0d9c1ab3 ID |
3288 | data = ceph_msg_data_add(msg); |
3289 | data->type = CEPH_MSG_DATA_PAGELIST; | |
89486833 | 3290 | refcount_inc(&pagelist->refcnt); |
6644ed7b AE |
3291 | data->pagelist = pagelist; |
3292 | ||
5240d9f9 | 3293 | msg->data_length += pagelist->length; |
27fa8385 | 3294 | } |
90af3602 | 3295 | EXPORT_SYMBOL(ceph_msg_data_add_pagelist); |
27fa8385 | 3296 | |
ea96571f | 3297 | #ifdef CONFIG_BLOCK |
5359a17d ID |
3298 | void ceph_msg_data_add_bio(struct ceph_msg *msg, struct ceph_bio_iter *bio_pos, |
3299 | u32 length) | |
27fa8385 | 3300 | { |
6644ed7b AE |
3301 | struct ceph_msg_data *data; |
3302 | ||
0d9c1ab3 ID |
3303 | data = ceph_msg_data_add(msg); |
3304 | data->type = CEPH_MSG_DATA_BIO; | |
5359a17d | 3305 | data->bio_pos = *bio_pos; |
c851c495 | 3306 | data->bio_length = length; |
6644ed7b | 3307 | |
5240d9f9 | 3308 | msg->data_length += length; |
27fa8385 | 3309 | } |
90af3602 | 3310 | EXPORT_SYMBOL(ceph_msg_data_add_bio); |
ea96571f | 3311 | #endif /* CONFIG_BLOCK */ |
27fa8385 | 3312 | |
b9e281c2 ID |
3313 | void ceph_msg_data_add_bvecs(struct ceph_msg *msg, |
3314 | struct ceph_bvec_iter *bvec_pos) | |
3315 | { | |
3316 | struct ceph_msg_data *data; | |
3317 | ||
0d9c1ab3 ID |
3318 | data = ceph_msg_data_add(msg); |
3319 | data->type = CEPH_MSG_DATA_BVECS; | |
b9e281c2 ID |
3320 | data->bvec_pos = *bvec_pos; |
3321 | ||
b9e281c2 ID |
3322 | msg->data_length += bvec_pos->iter.bi_size; |
3323 | } | |
3324 | EXPORT_SYMBOL(ceph_msg_data_add_bvecs); | |
3325 | ||
31b8006e SW |
3326 | /* |
3327 | * construct a new message with given type, size | |
3328 | * the new msg has a ref count of 1. | |
3329 | */ | |
0d9c1ab3 ID |
3330 | struct ceph_msg *ceph_msg_new2(int type, int front_len, int max_data_items, |
3331 | gfp_t flags, bool can_fail) | |
31b8006e SW |
3332 | { |
3333 | struct ceph_msg *m; | |
3334 | ||
e3d5d638 | 3335 | m = kmem_cache_zalloc(ceph_msg_cache, flags); |
31b8006e SW |
3336 | if (m == NULL) |
3337 | goto out; | |
31b8006e SW |
3338 | |
3339 | m->hdr.type = cpu_to_le16(type); | |
45c6ceb5 | 3340 | m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT); |
31b8006e | 3341 | m->hdr.front_len = cpu_to_le32(front_len); |
ca20892d | 3342 | |
9516e45b AE |
3343 | INIT_LIST_HEAD(&m->list_head); |
3344 | kref_init(&m->kref); | |
ca20892d | 3345 | |
31b8006e SW |
3346 | /* front */ |
3347 | if (front_len) { | |
eeb0bed5 | 3348 | m->front.iov_base = ceph_kvmalloc(front_len, flags); |
31b8006e | 3349 | if (m->front.iov_base == NULL) { |
b61c2763 | 3350 | dout("ceph_msg_new can't allocate %d bytes\n", |
31b8006e SW |
3351 | front_len); |
3352 | goto out2; | |
3353 | } | |
3354 | } else { | |
3355 | m->front.iov_base = NULL; | |
3356 | } | |
f2be82b0 | 3357 | m->front_alloc_len = m->front.iov_len = front_len; |
31b8006e | 3358 | |
0d9c1ab3 ID |
3359 | if (max_data_items) { |
3360 | m->data = kmalloc_array(max_data_items, sizeof(*m->data), | |
3361 | flags); | |
3362 | if (!m->data) | |
3363 | goto out2; | |
3364 | ||
3365 | m->max_data_items = max_data_items; | |
3366 | } | |
3367 | ||
bb257664 | 3368 | dout("ceph_msg_new %p front %d\n", m, front_len); |
31b8006e SW |
3369 | return m; |
3370 | ||
3371 | out2: | |
3372 | ceph_msg_put(m); | |
3373 | out: | |
b61c2763 SW |
3374 | if (!can_fail) { |
3375 | pr_err("msg_new can't create type %d front %d\n", type, | |
3376 | front_len); | |
f0ed1b7c | 3377 | WARN_ON(1); |
b61c2763 SW |
3378 | } else { |
3379 | dout("msg_new can't create type %d front %d\n", type, | |
3380 | front_len); | |
3381 | } | |
a79832f2 | 3382 | return NULL; |
31b8006e | 3383 | } |
0d9c1ab3 ID |
3384 | EXPORT_SYMBOL(ceph_msg_new2); |
3385 | ||
3386 | struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags, | |
3387 | bool can_fail) | |
3388 | { | |
3389 | return ceph_msg_new2(type, front_len, 0, flags, can_fail); | |
3390 | } | |
3d14c5d2 | 3391 | EXPORT_SYMBOL(ceph_msg_new); |
31b8006e | 3392 | |
31b8006e SW |
3393 | /* |
3394 | * Allocate "middle" portion of a message, if it is needed and wasn't | |
3395 | * allocated by alloc_msg. This allows us to read a small fixed-size | |
3396 | * per-type header in the front and then gracefully fail (i.e., | |
3397 | * propagate the error to the caller based on info in the front) when | |
3398 | * the middle is too large. | |
3399 | */ | |
2450418c | 3400 | static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg) |
31b8006e SW |
3401 | { |
3402 | int type = le16_to_cpu(msg->hdr.type); | |
3403 | int middle_len = le32_to_cpu(msg->hdr.middle_len); | |
3404 | ||
3405 | dout("alloc_middle %p type %d %s middle_len %d\n", msg, type, | |
3406 | ceph_msg_type_name(type), middle_len); | |
3407 | BUG_ON(!middle_len); | |
3408 | BUG_ON(msg->middle); | |
3409 | ||
b6c1d5b8 | 3410 | msg->middle = ceph_buffer_new(middle_len, GFP_NOFS); |
31b8006e SW |
3411 | if (!msg->middle) |
3412 | return -ENOMEM; | |
3413 | return 0; | |
3414 | } | |
3415 | ||
2450418c | 3416 | /* |
1c20f2d2 AE |
3417 | * Allocate a message for receiving an incoming message on a |
3418 | * connection, and save the result in con->in_msg. Uses the | |
3419 | * connection's private alloc_msg op if available. | |
3420 | * | |
4740a623 SW |
3421 | * Returns 0 on success, or a negative error code. |
3422 | * | |
3423 | * On success, if we set *skip = 1: | |
3424 | * - the next message should be skipped and ignored. | |
3425 | * - con->in_msg == NULL | |
3426 | * or if we set *skip = 0: | |
3427 | * - con->in_msg is non-null. | |
3428 | * On error (ENOMEM, EAGAIN, ...), | |
3429 | * - con->in_msg == NULL | |
2450418c | 3430 | */ |
4740a623 | 3431 | static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip) |
2450418c | 3432 | { |
4740a623 | 3433 | struct ceph_msg_header *hdr = &con->in_hdr; |
2450418c | 3434 | int middle_len = le32_to_cpu(hdr->middle_len); |
1d866d1c | 3435 | struct ceph_msg *msg; |
4740a623 | 3436 | int ret = 0; |
2450418c | 3437 | |
1c20f2d2 | 3438 | BUG_ON(con->in_msg != NULL); |
53ded495 | 3439 | BUG_ON(!con->ops->alloc_msg); |
2450418c | 3440 | |
53ded495 AE |
3441 | mutex_unlock(&con->mutex); |
3442 | msg = con->ops->alloc_msg(con, hdr, skip); | |
3443 | mutex_lock(&con->mutex); | |
3444 | if (con->state != CON_STATE_OPEN) { | |
3445 | if (msg) | |
1d866d1c | 3446 | ceph_msg_put(msg); |
53ded495 AE |
3447 | return -EAGAIN; |
3448 | } | |
4137577a AE |
3449 | if (msg) { |
3450 | BUG_ON(*skip); | |
583d0fef | 3451 | msg_con_set(msg, con); |
4137577a | 3452 | con->in_msg = msg; |
4137577a AE |
3453 | } else { |
3454 | /* | |
3455 | * Null message pointer means either we should skip | |
3456 | * this message or we couldn't allocate memory. The | |
3457 | * former is not an error. | |
3458 | */ | |
3459 | if (*skip) | |
3460 | return 0; | |
4137577a | 3461 | |
67c64eb7 | 3462 | con->error_msg = "error allocating memory for incoming message"; |
53ded495 | 3463 | return -ENOMEM; |
2450418c | 3464 | } |
1c20f2d2 | 3465 | memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr)); |
2450418c | 3466 | |
1c20f2d2 AE |
3467 | if (middle_len && !con->in_msg->middle) { |
3468 | ret = ceph_alloc_middle(con, con->in_msg); | |
2450418c | 3469 | if (ret < 0) { |
1c20f2d2 AE |
3470 | ceph_msg_put(con->in_msg); |
3471 | con->in_msg = NULL; | |
2450418c YS |
3472 | } |
3473 | } | |
9d7f0f13 | 3474 | |
4740a623 | 3475 | return ret; |
2450418c YS |
3476 | } |
3477 | ||
31b8006e SW |
3478 | |
3479 | /* | |
3480 | * Free a generically kmalloc'd message. | |
3481 | */ | |
0215e44b | 3482 | static void ceph_msg_free(struct ceph_msg *m) |
31b8006e | 3483 | { |
0215e44b | 3484 | dout("%s %p\n", __func__, m); |
4965fc38 | 3485 | kvfree(m->front.iov_base); |
0d9c1ab3 | 3486 | kfree(m->data); |
e3d5d638 | 3487 | kmem_cache_free(ceph_msg_cache, m); |
31b8006e SW |
3488 | } |
3489 | ||
0215e44b | 3490 | static void ceph_msg_release(struct kref *kref) |
c2e552e7 SW |
3491 | { |
3492 | struct ceph_msg *m = container_of(kref, struct ceph_msg, kref); | |
0d9c1ab3 | 3493 | int i; |
31b8006e | 3494 | |
0215e44b | 3495 | dout("%s %p\n", __func__, m); |
c2e552e7 SW |
3496 | WARN_ON(!list_empty(&m->list_head)); |
3497 | ||
583d0fef ID |
3498 | msg_con_set(m, NULL); |
3499 | ||
c2e552e7 SW |
3500 | /* drop middle, data, if any */ |
3501 | if (m->middle) { | |
3502 | ceph_buffer_put(m->middle); | |
3503 | m->middle = NULL; | |
31b8006e | 3504 | } |
5240d9f9 | 3505 | |
0d9c1ab3 ID |
3506 | for (i = 0; i < m->num_data_items; i++) |
3507 | ceph_msg_data_destroy(&m->data[i]); | |
58bb3b37 | 3508 | |
c2e552e7 SW |
3509 | if (m->pool) |
3510 | ceph_msgpool_put(m->pool, m); | |
3511 | else | |
0215e44b ID |
3512 | ceph_msg_free(m); |
3513 | } | |
3514 | ||
3515 | struct ceph_msg *ceph_msg_get(struct ceph_msg *msg) | |
3516 | { | |
3517 | dout("%s %p (was %d)\n", __func__, msg, | |
2c935bc5 | 3518 | kref_read(&msg->kref)); |
0215e44b ID |
3519 | kref_get(&msg->kref); |
3520 | return msg; | |
3521 | } | |
3522 | EXPORT_SYMBOL(ceph_msg_get); | |
3523 | ||
3524 | void ceph_msg_put(struct ceph_msg *msg) | |
3525 | { | |
3526 | dout("%s %p (was %d)\n", __func__, msg, | |
2c935bc5 | 3527 | kref_read(&msg->kref)); |
0215e44b | 3528 | kref_put(&msg->kref, ceph_msg_release); |
31b8006e | 3529 | } |
0215e44b | 3530 | EXPORT_SYMBOL(ceph_msg_put); |
9ec7cab1 SW |
3531 | |
3532 | void ceph_msg_dump(struct ceph_msg *msg) | |
3533 | { | |
3cea4c30 ID |
3534 | pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg, |
3535 | msg->front_alloc_len, msg->data_length); | |
9ec7cab1 SW |
3536 | print_hex_dump(KERN_DEBUG, "header: ", |
3537 | DUMP_PREFIX_OFFSET, 16, 1, | |
3538 | &msg->hdr, sizeof(msg->hdr), true); | |
3539 | print_hex_dump(KERN_DEBUG, " front: ", | |
3540 | DUMP_PREFIX_OFFSET, 16, 1, | |
3541 | msg->front.iov_base, msg->front.iov_len, true); | |
3542 | if (msg->middle) | |
3543 | print_hex_dump(KERN_DEBUG, "middle: ", | |
3544 | DUMP_PREFIX_OFFSET, 16, 1, | |
3545 | msg->middle->vec.iov_base, | |
3546 | msg->middle->vec.iov_len, true); | |
3547 | print_hex_dump(KERN_DEBUG, "footer: ", | |
3548 | DUMP_PREFIX_OFFSET, 16, 1, | |
3549 | &msg->footer, sizeof(msg->footer), true); | |
3550 | } | |
3d14c5d2 | 3551 | EXPORT_SYMBOL(ceph_msg_dump); |