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__); |
18370b36 | 415 | /* fall through */ |
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 | |
859bff51 | 493 | if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY)) { |
ba988f87 CH |
494 | int optval = 1; |
495 | ||
496 | ret = kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY, | |
497 | (char *)&optval, sizeof(optval)); | |
498 | if (ret) | |
499 | pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d", | |
500 | ret); | |
501 | } | |
502 | ||
a5bc3129 | 503 | con->sock = sock; |
41617d0c | 504 | return 0; |
31b8006e SW |
505 | } |
506 | ||
e5c93883 ID |
507 | /* |
508 | * If @buf is NULL, discard up to @len bytes. | |
509 | */ | |
31b8006e SW |
510 | static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len) |
511 | { | |
512 | struct kvec iov = {buf, len}; | |
513 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
98bdb0aa | 514 | int r; |
31b8006e | 515 | |
e5c93883 ID |
516 | if (!buf) |
517 | msg.msg_flags |= MSG_TRUNC; | |
518 | ||
aa563d7b | 519 | iov_iter_kvec(&msg.msg_iter, READ, &iov, 1, len); |
100803a8 | 520 | r = sock_recvmsg(sock, &msg, msg.msg_flags); |
98bdb0aa SW |
521 | if (r == -EAGAIN) |
522 | r = 0; | |
523 | return r; | |
31b8006e SW |
524 | } |
525 | ||
afb3d90e AE |
526 | static int ceph_tcp_recvpage(struct socket *sock, struct page *page, |
527 | int page_offset, size_t length) | |
528 | { | |
100803a8 AV |
529 | struct bio_vec bvec = { |
530 | .bv_page = page, | |
531 | .bv_offset = page_offset, | |
532 | .bv_len = length | |
533 | }; | |
534 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
535 | int r; | |
afb3d90e AE |
536 | |
537 | BUG_ON(page_offset + length > PAGE_SIZE); | |
aa563d7b | 538 | iov_iter_bvec(&msg.msg_iter, READ, &bvec, 1, length); |
100803a8 AV |
539 | r = sock_recvmsg(sock, &msg, msg.msg_flags); |
540 | if (r == -EAGAIN) | |
541 | r = 0; | |
542 | return r; | |
afb3d90e AE |
543 | } |
544 | ||
31b8006e SW |
545 | /* |
546 | * write something. @more is true if caller will be sending more data | |
547 | * shortly. | |
548 | */ | |
549 | static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov, | |
87349cda | 550 | size_t kvlen, size_t len, bool more) |
31b8006e SW |
551 | { |
552 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
42961d23 | 553 | int r; |
31b8006e SW |
554 | |
555 | if (more) | |
556 | msg.msg_flags |= MSG_MORE; | |
557 | else | |
558 | msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */ | |
559 | ||
42961d23 SW |
560 | r = kernel_sendmsg(sock, &msg, iov, kvlen, len); |
561 | if (r == -EAGAIN) | |
562 | r = 0; | |
563 | return r; | |
31b8006e SW |
564 | } |
565 | ||
433b0a12 ID |
566 | /* |
567 | * @more: either or both of MSG_MORE and MSG_SENDPAGE_NOTLAST | |
568 | */ | |
178eda29 | 569 | static int ceph_tcp_sendpage(struct socket *sock, struct page *page, |
433b0a12 | 570 | int offset, size_t size, int more) |
178eda29 | 571 | { |
3239eb52 ID |
572 | ssize_t (*sendpage)(struct socket *sock, struct page *page, |
573 | int offset, size_t size, int flags); | |
433b0a12 | 574 | int flags = MSG_DONTWAIT | MSG_NOSIGNAL | more; |
178eda29 | 575 | int ret; |
178eda29 | 576 | |
7e241f64 ID |
577 | /* |
578 | * sendpage cannot properly handle pages with page_count == 0, | |
579 | * we need to fall back to sendmsg if that's the case. | |
580 | * | |
581 | * Same goes for slab pages: skb_can_coalesce() allows | |
582 | * coalescing neighboring slab objects into a single frag which | |
583 | * triggers one of hardened usercopy checks. | |
584 | */ | |
585 | if (page_count(page) >= 1 && !PageSlab(page)) | |
3239eb52 | 586 | sendpage = sock->ops->sendpage; |
61ff6e9b | 587 | else |
3239eb52 | 588 | sendpage = sock_no_sendpage; |
61ff6e9b | 589 | |
3239eb52 | 590 | ret = sendpage(sock, page, offset, size, flags); |
61ff6e9b AV |
591 | if (ret == -EAGAIN) |
592 | ret = 0; | |
178eda29 CC |
593 | |
594 | return ret; | |
595 | } | |
31b8006e SW |
596 | |
597 | /* | |
598 | * Shutdown/close the socket for the given connection. | |
599 | */ | |
600 | static int con_close_socket(struct ceph_connection *con) | |
601 | { | |
8007b8d6 | 602 | int rc = 0; |
31b8006e SW |
603 | |
604 | dout("con_close_socket on %p sock %p\n", con, con->sock); | |
8007b8d6 SW |
605 | if (con->sock) { |
606 | rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR); | |
607 | sock_release(con->sock); | |
608 | con->sock = NULL; | |
609 | } | |
456ea468 AE |
610 | |
611 | /* | |
4a861692 | 612 | * Forcibly clear the SOCK_CLOSED flag. It gets set |
456ea468 AE |
613 | * independent of the connection mutex, and we could have |
614 | * received a socket close event before we had the chance to | |
615 | * shut the socket down. | |
616 | */ | |
c9ffc77a | 617 | con_flag_clear(con, CON_FLAG_SOCK_CLOSED); |
8007b8d6 | 618 | |
ce2c8903 | 619 | con_sock_state_closed(con); |
31b8006e SW |
620 | return rc; |
621 | } | |
622 | ||
623 | /* | |
624 | * Reset a connection. Discard all incoming and outgoing messages | |
625 | * and clear *_seq state. | |
626 | */ | |
627 | static void ceph_msg_remove(struct ceph_msg *msg) | |
628 | { | |
629 | list_del_init(&msg->list_head); | |
38941f80 | 630 | |
31b8006e SW |
631 | ceph_msg_put(msg); |
632 | } | |
633 | static void ceph_msg_remove_list(struct list_head *head) | |
634 | { | |
635 | while (!list_empty(head)) { | |
636 | struct ceph_msg *msg = list_first_entry(head, struct ceph_msg, | |
637 | list_head); | |
638 | ceph_msg_remove(msg); | |
639 | } | |
640 | } | |
641 | ||
642 | static void reset_connection(struct ceph_connection *con) | |
643 | { | |
644 | /* reset connection, out_queue, msg_ and connect_seq */ | |
645 | /* discard existing out_queue and msg_seq */ | |
0fa6ebc6 | 646 | dout("reset_connection %p\n", con); |
31b8006e SW |
647 | ceph_msg_remove_list(&con->out_queue); |
648 | ceph_msg_remove_list(&con->out_sent); | |
649 | ||
cf3e5c40 | 650 | if (con->in_msg) { |
38941f80 | 651 | BUG_ON(con->in_msg->con != con); |
cf3e5c40 SW |
652 | ceph_msg_put(con->in_msg); |
653 | con->in_msg = NULL; | |
654 | } | |
655 | ||
31b8006e SW |
656 | con->connect_seq = 0; |
657 | con->out_seq = 0; | |
c86a2930 | 658 | if (con->out_msg) { |
583d0fef | 659 | BUG_ON(con->out_msg->con != con); |
c86a2930 SW |
660 | ceph_msg_put(con->out_msg); |
661 | con->out_msg = NULL; | |
662 | } | |
31b8006e | 663 | con->in_seq = 0; |
0e0d5e0c | 664 | con->in_seq_acked = 0; |
67645d76 ID |
665 | |
666 | con->out_skip = 0; | |
31b8006e SW |
667 | } |
668 | ||
669 | /* | |
670 | * mark a peer down. drop any open connections. | |
671 | */ | |
672 | void ceph_con_close(struct ceph_connection *con) | |
673 | { | |
8c50c817 | 674 | mutex_lock(&con->mutex); |
b726ec97 | 675 | dout("con_close %p peer %s\n", con, ceph_pr_addr(&con->peer_addr)); |
8dacc7da | 676 | con->state = CON_STATE_CLOSED; |
a5988c49 | 677 | |
c9ffc77a AE |
678 | con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */ |
679 | con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING); | |
680 | con_flag_clear(con, CON_FLAG_WRITE_PENDING); | |
681 | con_flag_clear(con, CON_FLAG_BACKOFF); | |
a5988c49 | 682 | |
31b8006e | 683 | reset_connection(con); |
6f2bc3ff | 684 | con->peer_global_seq = 0; |
37ab77ac | 685 | cancel_con(con); |
ee76e073 | 686 | con_close_socket(con); |
ec302645 | 687 | mutex_unlock(&con->mutex); |
31b8006e | 688 | } |
3d14c5d2 | 689 | EXPORT_SYMBOL(ceph_con_close); |
31b8006e | 690 | |
31b8006e SW |
691 | /* |
692 | * Reopen a closed connection, with a new peer address. | |
693 | */ | |
b7a9e5dd SW |
694 | void ceph_con_open(struct ceph_connection *con, |
695 | __u8 entity_type, __u64 entity_num, | |
696 | struct ceph_entity_addr *addr) | |
31b8006e | 697 | { |
5469155f | 698 | mutex_lock(&con->mutex); |
b726ec97 | 699 | dout("con_open %p %s\n", con, ceph_pr_addr(addr)); |
8dacc7da | 700 | |
122070a2 | 701 | WARN_ON(con->state != CON_STATE_CLOSED); |
8dacc7da | 702 | con->state = CON_STATE_PREOPEN; |
a5988c49 | 703 | |
b7a9e5dd SW |
704 | con->peer_name.type = (__u8) entity_type; |
705 | con->peer_name.num = cpu_to_le64(entity_num); | |
706 | ||
31b8006e | 707 | memcpy(&con->peer_addr, addr, sizeof(*addr)); |
03c677e1 | 708 | con->delay = 0; /* reset backoff memory */ |
5469155f | 709 | mutex_unlock(&con->mutex); |
31b8006e SW |
710 | queue_con(con); |
711 | } | |
3d14c5d2 | 712 | EXPORT_SYMBOL(ceph_con_open); |
31b8006e | 713 | |
87b315a5 SW |
714 | /* |
715 | * return true if this connection ever successfully opened | |
716 | */ | |
717 | bool ceph_con_opened(struct ceph_connection *con) | |
718 | { | |
719 | return con->connect_seq > 0; | |
720 | } | |
721 | ||
31b8006e SW |
722 | /* |
723 | * initialize a new connection. | |
724 | */ | |
1bfd89f4 AE |
725 | void ceph_con_init(struct ceph_connection *con, void *private, |
726 | const struct ceph_connection_operations *ops, | |
b7a9e5dd | 727 | struct ceph_messenger *msgr) |
31b8006e SW |
728 | { |
729 | dout("con_init %p\n", con); | |
730 | memset(con, 0, sizeof(*con)); | |
1bfd89f4 AE |
731 | con->private = private; |
732 | con->ops = ops; | |
31b8006e | 733 | con->msgr = msgr; |
ce2c8903 AE |
734 | |
735 | con_sock_state_init(con); | |
736 | ||
ec302645 | 737 | mutex_init(&con->mutex); |
31b8006e SW |
738 | INIT_LIST_HEAD(&con->out_queue); |
739 | INIT_LIST_HEAD(&con->out_sent); | |
68931622 | 740 | INIT_DELAYED_WORK(&con->work, ceph_con_workfn); |
a5988c49 | 741 | |
8dacc7da | 742 | con->state = CON_STATE_CLOSED; |
31b8006e | 743 | } |
3d14c5d2 | 744 | EXPORT_SYMBOL(ceph_con_init); |
31b8006e SW |
745 | |
746 | ||
747 | /* | |
748 | * We maintain a global counter to order connection attempts. Get | |
749 | * a unique seq greater than @gt. | |
750 | */ | |
751 | static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt) | |
752 | { | |
753 | u32 ret; | |
754 | ||
755 | spin_lock(&msgr->global_seq_lock); | |
756 | if (msgr->global_seq < gt) | |
757 | msgr->global_seq = gt; | |
758 | ret = ++msgr->global_seq; | |
759 | spin_unlock(&msgr->global_seq_lock); | |
760 | return ret; | |
761 | } | |
762 | ||
e2200423 | 763 | static void con_out_kvec_reset(struct ceph_connection *con) |
859eb799 | 764 | { |
67645d76 ID |
765 | BUG_ON(con->out_skip); |
766 | ||
859eb799 AE |
767 | con->out_kvec_left = 0; |
768 | con->out_kvec_bytes = 0; | |
769 | con->out_kvec_cur = &con->out_kvec[0]; | |
770 | } | |
771 | ||
e2200423 | 772 | static void con_out_kvec_add(struct ceph_connection *con, |
859eb799 AE |
773 | size_t size, void *data) |
774 | { | |
67645d76 | 775 | int index = con->out_kvec_left; |
859eb799 | 776 | |
67645d76 | 777 | BUG_ON(con->out_skip); |
859eb799 AE |
778 | BUG_ON(index >= ARRAY_SIZE(con->out_kvec)); |
779 | ||
780 | con->out_kvec[index].iov_len = size; | |
781 | con->out_kvec[index].iov_base = data; | |
782 | con->out_kvec_left++; | |
783 | con->out_kvec_bytes += size; | |
784 | } | |
31b8006e | 785 | |
67645d76 ID |
786 | /* |
787 | * Chop off a kvec from the end. Return residual number of bytes for | |
788 | * that kvec, i.e. how many bytes would have been written if the kvec | |
789 | * hadn't been nuked. | |
790 | */ | |
791 | static int con_out_kvec_skip(struct ceph_connection *con) | |
792 | { | |
793 | int off = con->out_kvec_cur - con->out_kvec; | |
794 | int skip = 0; | |
795 | ||
796 | if (con->out_kvec_bytes > 0) { | |
797 | skip = con->out_kvec[off + con->out_kvec_left - 1].iov_len; | |
798 | BUG_ON(con->out_kvec_bytes < skip); | |
799 | BUG_ON(!con->out_kvec_left); | |
800 | con->out_kvec_bytes -= skip; | |
801 | con->out_kvec_left--; | |
802 | } | |
803 | ||
804 | return skip; | |
805 | } | |
806 | ||
df6ad1f9 | 807 | #ifdef CONFIG_BLOCK |
6aaa4511 AE |
808 | |
809 | /* | |
810 | * For a bio data item, a piece is whatever remains of the next | |
811 | * entry in the current bio iovec, or the first entry in the next | |
812 | * bio in the list. | |
813 | */ | |
8ae4f4f5 | 814 | static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor, |
25aff7c5 | 815 | size_t length) |
6aaa4511 | 816 | { |
8ae4f4f5 | 817 | struct ceph_msg_data *data = cursor->data; |
5359a17d | 818 | struct ceph_bio_iter *it = &cursor->bio_iter; |
6aaa4511 | 819 | |
5359a17d ID |
820 | cursor->resid = min_t(size_t, length, data->bio_length); |
821 | *it = data->bio_pos; | |
822 | if (cursor->resid < it->iter.bi_size) | |
823 | it->iter.bi_size = cursor->resid; | |
6aaa4511 | 824 | |
5359a17d ID |
825 | BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter)); |
826 | cursor->last_piece = cursor->resid == bio_iter_len(it->bio, it->iter); | |
6aaa4511 AE |
827 | } |
828 | ||
8ae4f4f5 | 829 | static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor, |
6aaa4511 AE |
830 | size_t *page_offset, |
831 | size_t *length) | |
832 | { | |
5359a17d ID |
833 | struct bio_vec bv = bio_iter_iovec(cursor->bio_iter.bio, |
834 | cursor->bio_iter.iter); | |
6aaa4511 | 835 | |
5359a17d ID |
836 | *page_offset = bv.bv_offset; |
837 | *length = bv.bv_len; | |
838 | return bv.bv_page; | |
6aaa4511 AE |
839 | } |
840 | ||
8ae4f4f5 AE |
841 | static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor, |
842 | size_t bytes) | |
6aaa4511 | 843 | { |
5359a17d | 844 | struct ceph_bio_iter *it = &cursor->bio_iter; |
187df763 | 845 | struct page *page = bio_iter_page(it->bio, it->iter); |
6aaa4511 | 846 | |
5359a17d ID |
847 | BUG_ON(bytes > cursor->resid); |
848 | BUG_ON(bytes > bio_iter_len(it->bio, it->iter)); | |
25aff7c5 | 849 | cursor->resid -= bytes; |
5359a17d | 850 | bio_advance_iter(it->bio, &it->iter, bytes); |
f38a5181 | 851 | |
5359a17d ID |
852 | if (!cursor->resid) { |
853 | BUG_ON(!cursor->last_piece); | |
854 | return false; /* no more data */ | |
855 | } | |
f38a5181 | 856 | |
187df763 ID |
857 | if (!bytes || (it->iter.bi_size && it->iter.bi_bvec_done && |
858 | page == bio_iter_page(it->bio, it->iter))) | |
6aaa4511 AE |
859 | return false; /* more bytes to process in this segment */ |
860 | ||
5359a17d ID |
861 | if (!it->iter.bi_size) { |
862 | it->bio = it->bio->bi_next; | |
863 | it->iter = it->bio->bi_iter; | |
864 | if (cursor->resid < it->iter.bi_size) | |
865 | it->iter.bi_size = cursor->resid; | |
25aff7c5 | 866 | } |
6aaa4511 | 867 | |
5359a17d ID |
868 | BUG_ON(cursor->last_piece); |
869 | BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter)); | |
870 | cursor->last_piece = cursor->resid == bio_iter_len(it->bio, it->iter); | |
6aaa4511 AE |
871 | return true; |
872 | } | |
ea96571f | 873 | #endif /* CONFIG_BLOCK */ |
df6ad1f9 | 874 | |
b9e281c2 ID |
875 | static void ceph_msg_data_bvecs_cursor_init(struct ceph_msg_data_cursor *cursor, |
876 | size_t length) | |
877 | { | |
878 | struct ceph_msg_data *data = cursor->data; | |
879 | struct bio_vec *bvecs = data->bvec_pos.bvecs; | |
880 | ||
881 | cursor->resid = min_t(size_t, length, data->bvec_pos.iter.bi_size); | |
882 | cursor->bvec_iter = data->bvec_pos.iter; | |
883 | cursor->bvec_iter.bi_size = cursor->resid; | |
884 | ||
885 | BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter)); | |
886 | cursor->last_piece = | |
887 | cursor->resid == bvec_iter_len(bvecs, cursor->bvec_iter); | |
888 | } | |
889 | ||
890 | static struct page *ceph_msg_data_bvecs_next(struct ceph_msg_data_cursor *cursor, | |
891 | size_t *page_offset, | |
892 | size_t *length) | |
893 | { | |
894 | struct bio_vec bv = bvec_iter_bvec(cursor->data->bvec_pos.bvecs, | |
895 | cursor->bvec_iter); | |
896 | ||
897 | *page_offset = bv.bv_offset; | |
898 | *length = bv.bv_len; | |
899 | return bv.bv_page; | |
900 | } | |
901 | ||
902 | static bool ceph_msg_data_bvecs_advance(struct ceph_msg_data_cursor *cursor, | |
903 | size_t bytes) | |
904 | { | |
905 | struct bio_vec *bvecs = cursor->data->bvec_pos.bvecs; | |
187df763 | 906 | struct page *page = bvec_iter_page(bvecs, cursor->bvec_iter); |
b9e281c2 ID |
907 | |
908 | BUG_ON(bytes > cursor->resid); | |
909 | BUG_ON(bytes > bvec_iter_len(bvecs, cursor->bvec_iter)); | |
910 | cursor->resid -= bytes; | |
911 | bvec_iter_advance(bvecs, &cursor->bvec_iter, bytes); | |
912 | ||
913 | if (!cursor->resid) { | |
914 | BUG_ON(!cursor->last_piece); | |
915 | return false; /* no more data */ | |
916 | } | |
917 | ||
187df763 ID |
918 | if (!bytes || (cursor->bvec_iter.bi_bvec_done && |
919 | page == bvec_iter_page(bvecs, cursor->bvec_iter))) | |
b9e281c2 ID |
920 | return false; /* more bytes to process in this segment */ |
921 | ||
922 | BUG_ON(cursor->last_piece); | |
923 | BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter)); | |
924 | cursor->last_piece = | |
925 | cursor->resid == bvec_iter_len(bvecs, cursor->bvec_iter); | |
926 | return true; | |
927 | } | |
928 | ||
e766d7b5 AE |
929 | /* |
930 | * For a page array, a piece comes from the first page in the array | |
931 | * that has not already been fully consumed. | |
932 | */ | |
8ae4f4f5 | 933 | static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor, |
25aff7c5 | 934 | size_t length) |
e766d7b5 | 935 | { |
8ae4f4f5 | 936 | struct ceph_msg_data *data = cursor->data; |
e766d7b5 AE |
937 | int page_count; |
938 | ||
939 | BUG_ON(data->type != CEPH_MSG_DATA_PAGES); | |
940 | ||
941 | BUG_ON(!data->pages); | |
942 | BUG_ON(!data->length); | |
943 | ||
ca8b3a69 | 944 | cursor->resid = min(length, data->length); |
e766d7b5 | 945 | page_count = calc_pages_for(data->alignment, (u64)data->length); |
e766d7b5 AE |
946 | cursor->page_offset = data->alignment & ~PAGE_MASK; |
947 | cursor->page_index = 0; | |
56fc5659 AE |
948 | BUG_ON(page_count > (int)USHRT_MAX); |
949 | cursor->page_count = (unsigned short)page_count; | |
950 | BUG_ON(length > SIZE_MAX - cursor->page_offset); | |
5f740d7e | 951 | cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE; |
e766d7b5 AE |
952 | } |
953 | ||
8ae4f4f5 AE |
954 | static struct page * |
955 | ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor, | |
956 | size_t *page_offset, size_t *length) | |
e766d7b5 | 957 | { |
8ae4f4f5 | 958 | struct ceph_msg_data *data = cursor->data; |
e766d7b5 AE |
959 | |
960 | BUG_ON(data->type != CEPH_MSG_DATA_PAGES); | |
961 | ||
962 | BUG_ON(cursor->page_index >= cursor->page_count); | |
963 | BUG_ON(cursor->page_offset >= PAGE_SIZE); | |
e766d7b5 AE |
964 | |
965 | *page_offset = cursor->page_offset; | |
25aff7c5 | 966 | if (cursor->last_piece) |
e766d7b5 | 967 | *length = cursor->resid; |
25aff7c5 | 968 | else |
e766d7b5 | 969 | *length = PAGE_SIZE - *page_offset; |
e766d7b5 AE |
970 | |
971 | return data->pages[cursor->page_index]; | |
972 | } | |
973 | ||
8ae4f4f5 | 974 | static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor, |
e766d7b5 AE |
975 | size_t bytes) |
976 | { | |
8ae4f4f5 | 977 | BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES); |
e766d7b5 AE |
978 | |
979 | BUG_ON(cursor->page_offset + bytes > PAGE_SIZE); | |
e766d7b5 AE |
980 | |
981 | /* Advance the cursor page offset */ | |
982 | ||
983 | cursor->resid -= bytes; | |
5df521b1 AE |
984 | cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK; |
985 | if (!bytes || cursor->page_offset) | |
e766d7b5 AE |
986 | return false; /* more bytes to process in the current page */ |
987 | ||
d90deda6 YZ |
988 | if (!cursor->resid) |
989 | return false; /* no more data */ | |
990 | ||
5df521b1 | 991 | /* Move on to the next page; offset is already at 0 */ |
e766d7b5 AE |
992 | |
993 | BUG_ON(cursor->page_index >= cursor->page_count); | |
e766d7b5 | 994 | cursor->page_index++; |
25aff7c5 | 995 | cursor->last_piece = cursor->resid <= PAGE_SIZE; |
e766d7b5 AE |
996 | |
997 | return true; | |
998 | } | |
999 | ||
fe38a2b6 | 1000 | /* |
dd236fcb AE |
1001 | * For a pagelist, a piece is whatever remains to be consumed in the |
1002 | * first page in the list, or the front of the next page. | |
fe38a2b6 | 1003 | */ |
8ae4f4f5 AE |
1004 | static void |
1005 | ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor, | |
25aff7c5 | 1006 | size_t length) |
fe38a2b6 | 1007 | { |
8ae4f4f5 | 1008 | struct ceph_msg_data *data = cursor->data; |
fe38a2b6 AE |
1009 | struct ceph_pagelist *pagelist; |
1010 | struct page *page; | |
1011 | ||
dd236fcb | 1012 | BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST); |
fe38a2b6 AE |
1013 | |
1014 | pagelist = data->pagelist; | |
1015 | BUG_ON(!pagelist); | |
25aff7c5 AE |
1016 | |
1017 | if (!length) | |
fe38a2b6 AE |
1018 | return; /* pagelist can be assigned but empty */ |
1019 | ||
1020 | BUG_ON(list_empty(&pagelist->head)); | |
1021 | page = list_first_entry(&pagelist->head, struct page, lru); | |
1022 | ||
ca8b3a69 | 1023 | cursor->resid = min(length, pagelist->length); |
fe38a2b6 AE |
1024 | cursor->page = page; |
1025 | cursor->offset = 0; | |
a51b272e | 1026 | cursor->last_piece = cursor->resid <= PAGE_SIZE; |
fe38a2b6 AE |
1027 | } |
1028 | ||
8ae4f4f5 AE |
1029 | static struct page * |
1030 | ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor, | |
1031 | size_t *page_offset, size_t *length) | |
fe38a2b6 | 1032 | { |
8ae4f4f5 | 1033 | struct ceph_msg_data *data = cursor->data; |
fe38a2b6 | 1034 | struct ceph_pagelist *pagelist; |
fe38a2b6 AE |
1035 | |
1036 | BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST); | |
1037 | ||
1038 | pagelist = data->pagelist; | |
1039 | BUG_ON(!pagelist); | |
1040 | ||
1041 | BUG_ON(!cursor->page); | |
25aff7c5 | 1042 | BUG_ON(cursor->offset + cursor->resid != pagelist->length); |
fe38a2b6 | 1043 | |
5df521b1 | 1044 | /* offset of first page in pagelist is always 0 */ |
fe38a2b6 | 1045 | *page_offset = cursor->offset & ~PAGE_MASK; |
5df521b1 | 1046 | if (cursor->last_piece) |
25aff7c5 AE |
1047 | *length = cursor->resid; |
1048 | else | |
1049 | *length = PAGE_SIZE - *page_offset; | |
fe38a2b6 | 1050 | |
8ae4f4f5 | 1051 | return cursor->page; |
fe38a2b6 AE |
1052 | } |
1053 | ||
8ae4f4f5 | 1054 | static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor, |
dd236fcb | 1055 | size_t bytes) |
fe38a2b6 | 1056 | { |
8ae4f4f5 | 1057 | struct ceph_msg_data *data = cursor->data; |
fe38a2b6 AE |
1058 | struct ceph_pagelist *pagelist; |
1059 | ||
1060 | BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST); | |
1061 | ||
1062 | pagelist = data->pagelist; | |
1063 | BUG_ON(!pagelist); | |
25aff7c5 AE |
1064 | |
1065 | BUG_ON(cursor->offset + cursor->resid != pagelist->length); | |
fe38a2b6 AE |
1066 | BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE); |
1067 | ||
1068 | /* Advance the cursor offset */ | |
1069 | ||
25aff7c5 | 1070 | cursor->resid -= bytes; |
fe38a2b6 | 1071 | cursor->offset += bytes; |
5df521b1 | 1072 | /* offset of first page in pagelist is always 0 */ |
fe38a2b6 AE |
1073 | if (!bytes || cursor->offset & ~PAGE_MASK) |
1074 | return false; /* more bytes to process in the current page */ | |
1075 | ||
d90deda6 YZ |
1076 | if (!cursor->resid) |
1077 | return false; /* no more data */ | |
1078 | ||
fe38a2b6 AE |
1079 | /* Move on to the next page */ |
1080 | ||
1081 | BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head)); | |
17ddc49b | 1082 | cursor->page = list_next_entry(cursor->page, lru); |
25aff7c5 | 1083 | cursor->last_piece = cursor->resid <= PAGE_SIZE; |
fe38a2b6 AE |
1084 | |
1085 | return true; | |
1086 | } | |
1087 | ||
dd236fcb AE |
1088 | /* |
1089 | * Message data is handled (sent or received) in pieces, where each | |
1090 | * piece resides on a single page. The network layer might not | |
1091 | * consume an entire piece at once. A data item's cursor keeps | |
1092 | * track of which piece is next to process and how much remains to | |
1093 | * be processed in that piece. It also tracks whether the current | |
1094 | * piece is the last one in the data item. | |
1095 | */ | |
ca8b3a69 | 1096 | static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor) |
dd236fcb | 1097 | { |
ca8b3a69 | 1098 | size_t length = cursor->total_resid; |
8ae4f4f5 | 1099 | |
8ae4f4f5 | 1100 | switch (cursor->data->type) { |
dd236fcb | 1101 | case CEPH_MSG_DATA_PAGELIST: |
8ae4f4f5 | 1102 | ceph_msg_data_pagelist_cursor_init(cursor, length); |
dd236fcb | 1103 | break; |
e766d7b5 | 1104 | case CEPH_MSG_DATA_PAGES: |
8ae4f4f5 | 1105 | ceph_msg_data_pages_cursor_init(cursor, length); |
e766d7b5 | 1106 | break; |
dd236fcb AE |
1107 | #ifdef CONFIG_BLOCK |
1108 | case CEPH_MSG_DATA_BIO: | |
8ae4f4f5 | 1109 | ceph_msg_data_bio_cursor_init(cursor, length); |
6aaa4511 | 1110 | break; |
dd236fcb | 1111 | #endif /* CONFIG_BLOCK */ |
b9e281c2 ID |
1112 | case CEPH_MSG_DATA_BVECS: |
1113 | ceph_msg_data_bvecs_cursor_init(cursor, length); | |
1114 | break; | |
6aaa4511 | 1115 | case CEPH_MSG_DATA_NONE: |
dd236fcb AE |
1116 | default: |
1117 | /* BUG(); */ | |
1118 | break; | |
1119 | } | |
8ae4f4f5 | 1120 | cursor->need_crc = true; |
dd236fcb AE |
1121 | } |
1122 | ||
ca8b3a69 AE |
1123 | static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length) |
1124 | { | |
1125 | struct ceph_msg_data_cursor *cursor = &msg->cursor; | |
ca8b3a69 AE |
1126 | |
1127 | BUG_ON(!length); | |
1128 | BUG_ON(length > msg->data_length); | |
0d9c1ab3 | 1129 | BUG_ON(!msg->num_data_items); |
ca8b3a69 | 1130 | |
ca8b3a69 | 1131 | cursor->total_resid = length; |
0d9c1ab3 | 1132 | cursor->data = msg->data; |
ca8b3a69 AE |
1133 | |
1134 | __ceph_msg_data_cursor_init(cursor); | |
1135 | } | |
1136 | ||
dd236fcb AE |
1137 | /* |
1138 | * Return the page containing the next piece to process for a given | |
1139 | * data item, and supply the page offset and length of that piece. | |
1140 | * Indicate whether this is the last piece in this data item. | |
1141 | */ | |
8ae4f4f5 AE |
1142 | static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor, |
1143 | size_t *page_offset, size_t *length, | |
dd236fcb AE |
1144 | bool *last_piece) |
1145 | { | |
1146 | struct page *page; | |
1147 | ||
8ae4f4f5 | 1148 | switch (cursor->data->type) { |
dd236fcb | 1149 | case CEPH_MSG_DATA_PAGELIST: |
8ae4f4f5 | 1150 | page = ceph_msg_data_pagelist_next(cursor, page_offset, length); |
dd236fcb | 1151 | break; |
e766d7b5 | 1152 | case CEPH_MSG_DATA_PAGES: |
8ae4f4f5 | 1153 | page = ceph_msg_data_pages_next(cursor, page_offset, length); |
e766d7b5 | 1154 | break; |
dd236fcb AE |
1155 | #ifdef CONFIG_BLOCK |
1156 | case CEPH_MSG_DATA_BIO: | |
8ae4f4f5 | 1157 | page = ceph_msg_data_bio_next(cursor, page_offset, length); |
6aaa4511 | 1158 | break; |
dd236fcb | 1159 | #endif /* CONFIG_BLOCK */ |
b9e281c2 ID |
1160 | case CEPH_MSG_DATA_BVECS: |
1161 | page = ceph_msg_data_bvecs_next(cursor, page_offset, length); | |
1162 | break; | |
6aaa4511 | 1163 | case CEPH_MSG_DATA_NONE: |
dd236fcb AE |
1164 | default: |
1165 | page = NULL; | |
1166 | break; | |
1167 | } | |
5359a17d | 1168 | |
dd236fcb AE |
1169 | BUG_ON(!page); |
1170 | BUG_ON(*page_offset + *length > PAGE_SIZE); | |
1171 | BUG_ON(!*length); | |
5359a17d | 1172 | BUG_ON(*length > cursor->resid); |
dd236fcb | 1173 | if (last_piece) |
8ae4f4f5 | 1174 | *last_piece = cursor->last_piece; |
dd236fcb AE |
1175 | |
1176 | return page; | |
1177 | } | |
1178 | ||
1179 | /* | |
1180 | * Returns true if the result moves the cursor on to the next piece | |
1181 | * of the data item. | |
1182 | */ | |
1759f7b0 ID |
1183 | static void ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor, |
1184 | size_t bytes) | |
dd236fcb AE |
1185 | { |
1186 | bool new_piece; | |
1187 | ||
25aff7c5 | 1188 | BUG_ON(bytes > cursor->resid); |
8ae4f4f5 | 1189 | switch (cursor->data->type) { |
dd236fcb | 1190 | case CEPH_MSG_DATA_PAGELIST: |
8ae4f4f5 | 1191 | new_piece = ceph_msg_data_pagelist_advance(cursor, bytes); |
dd236fcb | 1192 | break; |
e766d7b5 | 1193 | case CEPH_MSG_DATA_PAGES: |
8ae4f4f5 | 1194 | new_piece = ceph_msg_data_pages_advance(cursor, bytes); |
e766d7b5 | 1195 | break; |
dd236fcb AE |
1196 | #ifdef CONFIG_BLOCK |
1197 | case CEPH_MSG_DATA_BIO: | |
8ae4f4f5 | 1198 | new_piece = ceph_msg_data_bio_advance(cursor, bytes); |
6aaa4511 | 1199 | break; |
dd236fcb | 1200 | #endif /* CONFIG_BLOCK */ |
b9e281c2 ID |
1201 | case CEPH_MSG_DATA_BVECS: |
1202 | new_piece = ceph_msg_data_bvecs_advance(cursor, bytes); | |
1203 | break; | |
6aaa4511 | 1204 | case CEPH_MSG_DATA_NONE: |
dd236fcb AE |
1205 | default: |
1206 | BUG(); | |
1207 | break; | |
1208 | } | |
ca8b3a69 | 1209 | cursor->total_resid -= bytes; |
dd236fcb | 1210 | |
ca8b3a69 AE |
1211 | if (!cursor->resid && cursor->total_resid) { |
1212 | WARN_ON(!cursor->last_piece); | |
0d9c1ab3 | 1213 | cursor->data++; |
ca8b3a69 | 1214 | __ceph_msg_data_cursor_init(cursor); |
a51b272e | 1215 | new_piece = true; |
ca8b3a69 | 1216 | } |
a51b272e | 1217 | cursor->need_crc = new_piece; |
dd236fcb AE |
1218 | } |
1219 | ||
dbc0d3ca ID |
1220 | static size_t sizeof_footer(struct ceph_connection *con) |
1221 | { | |
1222 | return (con->peer_features & CEPH_FEATURE_MSG_AUTH) ? | |
1223 | sizeof(struct ceph_msg_footer) : | |
1224 | sizeof(struct ceph_msg_footer_old); | |
1225 | } | |
1226 | ||
98fa5dd8 | 1227 | static void prepare_message_data(struct ceph_msg *msg, u32 data_len) |
739c905b | 1228 | { |
4c59b4a2 | 1229 | /* Initialize data cursor */ |
fe38a2b6 | 1230 | |
8ae4f4f5 | 1231 | ceph_msg_data_cursor_init(msg, (size_t)data_len); |
739c905b AE |
1232 | } |
1233 | ||
31b8006e SW |
1234 | /* |
1235 | * Prepare footer for currently outgoing message, and finish things | |
1236 | * off. Assumes out_kvec* are already valid.. we just add on to the end. | |
1237 | */ | |
859eb799 | 1238 | static void prepare_write_message_footer(struct ceph_connection *con) |
31b8006e SW |
1239 | { |
1240 | struct ceph_msg *m = con->out_msg; | |
1241 | ||
fd154f3c AE |
1242 | m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE; |
1243 | ||
31b8006e | 1244 | dout("prepare_write_message_footer %p\n", con); |
89f08173 | 1245 | con_out_kvec_add(con, sizeof_footer(con), &m->footer); |
33d07337 YZ |
1246 | if (con->peer_features & CEPH_FEATURE_MSG_AUTH) { |
1247 | if (con->ops->sign_message) | |
79dbd1ba | 1248 | con->ops->sign_message(m); |
33d07337 YZ |
1249 | else |
1250 | m->footer.sig = 0; | |
33d07337 YZ |
1251 | } else { |
1252 | m->old_footer.flags = m->footer.flags; | |
33d07337 | 1253 | } |
31b8006e | 1254 | con->out_more = m->more_to_follow; |
c86a2930 | 1255 | con->out_msg_done = true; |
31b8006e SW |
1256 | } |
1257 | ||
1258 | /* | |
1259 | * Prepare headers for the next outgoing message. | |
1260 | */ | |
1261 | static void prepare_write_message(struct ceph_connection *con) | |
1262 | { | |
1263 | struct ceph_msg *m; | |
a9a0c51a | 1264 | u32 crc; |
31b8006e | 1265 | |
e2200423 | 1266 | con_out_kvec_reset(con); |
c86a2930 | 1267 | con->out_msg_done = false; |
31b8006e SW |
1268 | |
1269 | /* Sneak an ack in there first? If we can get it into the same | |
1270 | * TCP packet that's a good thing. */ | |
1271 | if (con->in_seq > con->in_seq_acked) { | |
1272 | con->in_seq_acked = con->in_seq; | |
e2200423 | 1273 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
31b8006e | 1274 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 1275 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 | 1276 | &con->out_temp_ack); |
31b8006e SW |
1277 | } |
1278 | ||
38941f80 | 1279 | BUG_ON(list_empty(&con->out_queue)); |
859eb799 | 1280 | m = list_first_entry(&con->out_queue, struct ceph_msg, list_head); |
c86a2930 | 1281 | con->out_msg = m; |
38941f80 | 1282 | BUG_ON(m->con != con); |
4cf9d544 SW |
1283 | |
1284 | /* put message on sent list */ | |
1285 | ceph_msg_get(m); | |
1286 | list_move_tail(&m->list_head, &con->out_sent); | |
31b8006e | 1287 | |
e84346b7 SW |
1288 | /* |
1289 | * only assign outgoing seq # if we haven't sent this message | |
1290 | * yet. if it is requeued, resend with it's original seq. | |
1291 | */ | |
1292 | if (m->needs_out_seq) { | |
1293 | m->hdr.seq = cpu_to_le64(++con->out_seq); | |
1294 | m->needs_out_seq = false; | |
98ad5ebd | 1295 | |
4690faf0 ID |
1296 | if (con->ops->reencode_message) |
1297 | con->ops->reencode_message(m); | |
1298 | } | |
31b8006e | 1299 | |
98fa5dd8 | 1300 | dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n", |
31b8006e SW |
1301 | m, con->out_seq, le16_to_cpu(m->hdr.type), |
1302 | le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len), | |
98fa5dd8 | 1303 | m->data_length); |
98ad5ebd ID |
1304 | WARN_ON(m->front.iov_len != le32_to_cpu(m->hdr.front_len)); |
1305 | WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len)); | |
31b8006e SW |
1306 | |
1307 | /* tag + hdr + front + middle */ | |
e2200423 | 1308 | con_out_kvec_add(con, sizeof (tag_msg), &tag_msg); |
67645d76 | 1309 | con_out_kvec_add(con, sizeof(con->out_hdr), &con->out_hdr); |
e2200423 | 1310 | con_out_kvec_add(con, m->front.iov_len, m->front.iov_base); |
859eb799 | 1311 | |
31b8006e | 1312 | if (m->middle) |
e2200423 | 1313 | con_out_kvec_add(con, m->middle->vec.iov_len, |
859eb799 | 1314 | m->middle->vec.iov_base); |
31b8006e | 1315 | |
67645d76 | 1316 | /* fill in hdr crc and finalize hdr */ |
a9a0c51a AE |
1317 | crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc)); |
1318 | con->out_msg->hdr.crc = cpu_to_le32(crc); | |
67645d76 | 1319 | memcpy(&con->out_hdr, &con->out_msg->hdr, sizeof(con->out_hdr)); |
a9a0c51a | 1320 | |
67645d76 | 1321 | /* fill in front and middle crc, footer */ |
a9a0c51a AE |
1322 | crc = crc32c(0, m->front.iov_base, m->front.iov_len); |
1323 | con->out_msg->footer.front_crc = cpu_to_le32(crc); | |
1324 | if (m->middle) { | |
1325 | crc = crc32c(0, m->middle->vec.iov_base, | |
1326 | m->middle->vec.iov_len); | |
1327 | con->out_msg->footer.middle_crc = cpu_to_le32(crc); | |
1328 | } else | |
31b8006e | 1329 | con->out_msg->footer.middle_crc = 0; |
739c905b | 1330 | dout("%s front_crc %u middle_crc %u\n", __func__, |
31b8006e SW |
1331 | le32_to_cpu(con->out_msg->footer.front_crc), |
1332 | le32_to_cpu(con->out_msg->footer.middle_crc)); | |
67645d76 | 1333 | con->out_msg->footer.flags = 0; |
31b8006e SW |
1334 | |
1335 | /* is there a data payload? */ | |
739c905b | 1336 | con->out_msg->footer.data_crc = 0; |
98fa5dd8 AE |
1337 | if (m->data_length) { |
1338 | prepare_message_data(con->out_msg, m->data_length); | |
78625051 AE |
1339 | con->out_more = 1; /* data + footer will follow */ |
1340 | } else { | |
31b8006e | 1341 | /* no, queue up footer too and be done */ |
859eb799 | 1342 | prepare_write_message_footer(con); |
78625051 | 1343 | } |
31b8006e | 1344 | |
c9ffc77a | 1345 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
31b8006e SW |
1346 | } |
1347 | ||
1348 | /* | |
1349 | * Prepare an ack. | |
1350 | */ | |
1351 | static void prepare_write_ack(struct ceph_connection *con) | |
1352 | { | |
1353 | dout("prepare_write_ack %p %llu -> %llu\n", con, | |
1354 | con->in_seq_acked, con->in_seq); | |
1355 | con->in_seq_acked = con->in_seq; | |
1356 | ||
e2200423 | 1357 | con_out_kvec_reset(con); |
859eb799 | 1358 | |
e2200423 | 1359 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
859eb799 | 1360 | |
31b8006e | 1361 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 1362 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 AE |
1363 | &con->out_temp_ack); |
1364 | ||
31b8006e | 1365 | con->out_more = 1; /* more will follow.. eventually.. */ |
c9ffc77a | 1366 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
31b8006e SW |
1367 | } |
1368 | ||
3a23083b SW |
1369 | /* |
1370 | * Prepare to share the seq during handshake | |
1371 | */ | |
1372 | static void prepare_write_seq(struct ceph_connection *con) | |
1373 | { | |
1374 | dout("prepare_write_seq %p %llu -> %llu\n", con, | |
1375 | con->in_seq_acked, con->in_seq); | |
1376 | con->in_seq_acked = con->in_seq; | |
1377 | ||
1378 | con_out_kvec_reset(con); | |
1379 | ||
1380 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); | |
1381 | con_out_kvec_add(con, sizeof (con->out_temp_ack), | |
1382 | &con->out_temp_ack); | |
1383 | ||
1384 | con_flag_set(con, CON_FLAG_WRITE_PENDING); | |
1385 | } | |
1386 | ||
31b8006e SW |
1387 | /* |
1388 | * Prepare to write keepalive byte. | |
1389 | */ | |
1390 | static void prepare_write_keepalive(struct ceph_connection *con) | |
1391 | { | |
1392 | dout("prepare_write_keepalive %p\n", con); | |
e2200423 | 1393 | con_out_kvec_reset(con); |
8b9558aa | 1394 | if (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2) { |
473bd2d7 | 1395 | struct timespec64 now; |
7f61f545 | 1396 | |
473bd2d7 | 1397 | ktime_get_real_ts64(&now); |
8b9558aa | 1398 | con_out_kvec_add(con, sizeof(tag_keepalive2), &tag_keepalive2); |
473bd2d7 | 1399 | ceph_encode_timespec64(&con->out_temp_keepalive2, &now); |
7f61f545 ID |
1400 | con_out_kvec_add(con, sizeof(con->out_temp_keepalive2), |
1401 | &con->out_temp_keepalive2); | |
8b9558aa YZ |
1402 | } else { |
1403 | con_out_kvec_add(con, sizeof(tag_keepalive), &tag_keepalive); | |
1404 | } | |
c9ffc77a | 1405 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
31b8006e SW |
1406 | } |
1407 | ||
1408 | /* | |
1409 | * Connection negotiation. | |
1410 | */ | |
1411 | ||
262614c4 | 1412 | static int get_connect_authorizer(struct ceph_connection *con) |
4e7a5dcd | 1413 | { |
a3530df3 | 1414 | struct ceph_auth_handshake *auth; |
262614c4 | 1415 | int auth_proto; |
b1c6b980 AE |
1416 | |
1417 | if (!con->ops->get_authorizer) { | |
262614c4 | 1418 | con->auth = NULL; |
b1c6b980 AE |
1419 | con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN; |
1420 | con->out_connect.authorizer_len = 0; | |
262614c4 | 1421 | return 0; |
b1c6b980 AE |
1422 | } |
1423 | ||
262614c4 | 1424 | auth = con->ops->get_authorizer(con, &auth_proto, con->auth_retry); |
a3530df3 | 1425 | if (IS_ERR(auth)) |
262614c4 | 1426 | return PTR_ERR(auth); |
0da5d703 | 1427 | |
262614c4 ID |
1428 | con->auth = auth; |
1429 | con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto); | |
1430 | con->out_connect.authorizer_len = cpu_to_le32(auth->authorizer_buf_len); | |
1431 | return 0; | |
4e7a5dcd SW |
1432 | } |
1433 | ||
31b8006e SW |
1434 | /* |
1435 | * We connected to a peer and are saying hello. | |
1436 | */ | |
e825a66d | 1437 | static void prepare_write_banner(struct ceph_connection *con) |
31b8006e | 1438 | { |
e2200423 AE |
1439 | con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER); |
1440 | con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr), | |
e825a66d | 1441 | &con->msgr->my_enc_addr); |
eed0ef2c | 1442 | |
eed0ef2c | 1443 | con->out_more = 0; |
c9ffc77a | 1444 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
eed0ef2c SW |
1445 | } |
1446 | ||
c0f56b48 ID |
1447 | static void __prepare_write_connect(struct ceph_connection *con) |
1448 | { | |
1449 | con_out_kvec_add(con, sizeof(con->out_connect), &con->out_connect); | |
1450 | if (con->auth) | |
1451 | con_out_kvec_add(con, con->auth->authorizer_buf_len, | |
1452 | con->auth->authorizer_buf); | |
1453 | ||
1454 | con->out_more = 0; | |
1455 | con_flag_set(con, CON_FLAG_WRITE_PENDING); | |
1456 | } | |
1457 | ||
e825a66d | 1458 | static int prepare_write_connect(struct ceph_connection *con) |
eed0ef2c | 1459 | { |
95c96174 | 1460 | unsigned int global_seq = get_global_seq(con->msgr, 0); |
31b8006e | 1461 | int proto; |
262614c4 | 1462 | int ret; |
31b8006e SW |
1463 | |
1464 | switch (con->peer_name.type) { | |
1465 | case CEPH_ENTITY_TYPE_MON: | |
1466 | proto = CEPH_MONC_PROTOCOL; | |
1467 | break; | |
1468 | case CEPH_ENTITY_TYPE_OSD: | |
1469 | proto = CEPH_OSDC_PROTOCOL; | |
1470 | break; | |
1471 | case CEPH_ENTITY_TYPE_MDS: | |
1472 | proto = CEPH_MDSC_PROTOCOL; | |
1473 | break; | |
1474 | default: | |
1475 | BUG(); | |
1476 | } | |
1477 | ||
1478 | dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con, | |
1479 | con->connect_seq, global_seq, proto); | |
4e7a5dcd | 1480 | |
859bff51 ID |
1481 | con->out_connect.features = |
1482 | cpu_to_le64(from_msgr(con->msgr)->supported_features); | |
31b8006e SW |
1483 | con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT); |
1484 | con->out_connect.connect_seq = cpu_to_le32(con->connect_seq); | |
1485 | con->out_connect.global_seq = cpu_to_le32(global_seq); | |
1486 | con->out_connect.protocol_version = cpu_to_le32(proto); | |
1487 | con->out_connect.flags = 0; | |
31b8006e | 1488 | |
262614c4 ID |
1489 | ret = get_connect_authorizer(con); |
1490 | if (ret) | |
1491 | return ret; | |
3da54776 | 1492 | |
c0f56b48 | 1493 | __prepare_write_connect(con); |
e10c758e | 1494 | return 0; |
31b8006e SW |
1495 | } |
1496 | ||
31b8006e SW |
1497 | /* |
1498 | * write as much of pending kvecs to the socket as we can. | |
1499 | * 1 -> done | |
1500 | * 0 -> socket full, but more to do | |
1501 | * <0 -> error | |
1502 | */ | |
1503 | static int write_partial_kvec(struct ceph_connection *con) | |
1504 | { | |
1505 | int ret; | |
1506 | ||
1507 | dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes); | |
1508 | while (con->out_kvec_bytes > 0) { | |
1509 | ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur, | |
1510 | con->out_kvec_left, con->out_kvec_bytes, | |
1511 | con->out_more); | |
1512 | if (ret <= 0) | |
1513 | goto out; | |
1514 | con->out_kvec_bytes -= ret; | |
1515 | if (con->out_kvec_bytes == 0) | |
1516 | break; /* done */ | |
f42299e6 AE |
1517 | |
1518 | /* account for full iov entries consumed */ | |
1519 | while (ret >= con->out_kvec_cur->iov_len) { | |
1520 | BUG_ON(!con->out_kvec_left); | |
1521 | ret -= con->out_kvec_cur->iov_len; | |
1522 | con->out_kvec_cur++; | |
1523 | con->out_kvec_left--; | |
1524 | } | |
1525 | /* and for a partially-consumed entry */ | |
1526 | if (ret) { | |
1527 | con->out_kvec_cur->iov_len -= ret; | |
1528 | con->out_kvec_cur->iov_base += ret; | |
31b8006e SW |
1529 | } |
1530 | } | |
1531 | con->out_kvec_left = 0; | |
31b8006e SW |
1532 | ret = 1; |
1533 | out: | |
1534 | dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con, | |
1535 | con->out_kvec_bytes, con->out_kvec_left, ret); | |
1536 | return ret; /* done! */ | |
1537 | } | |
1538 | ||
35b62808 AE |
1539 | static u32 ceph_crc32c_page(u32 crc, struct page *page, |
1540 | unsigned int page_offset, | |
1541 | unsigned int length) | |
1542 | { | |
1543 | char *kaddr; | |
1544 | ||
1545 | kaddr = kmap(page); | |
1546 | BUG_ON(kaddr == NULL); | |
1547 | crc = crc32c(crc, kaddr + page_offset, length); | |
1548 | kunmap(page); | |
1549 | ||
1550 | return crc; | |
1551 | } | |
31b8006e SW |
1552 | /* |
1553 | * Write as much message data payload as we can. If we finish, queue | |
1554 | * up the footer. | |
1555 | * 1 -> done, footer is now queued in out_kvec[]. | |
1556 | * 0 -> socket full, but more to do | |
1557 | * <0 -> error | |
1558 | */ | |
34d2d200 | 1559 | static int write_partial_message_data(struct ceph_connection *con) |
31b8006e SW |
1560 | { |
1561 | struct ceph_msg *msg = con->out_msg; | |
8ae4f4f5 | 1562 | struct ceph_msg_data_cursor *cursor = &msg->cursor; |
859bff51 | 1563 | bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC); |
433b0a12 | 1564 | int more = MSG_MORE | MSG_SENDPAGE_NOTLAST; |
f5db90bc | 1565 | u32 crc; |
31b8006e | 1566 | |
859a35d5 | 1567 | dout("%s %p msg %p\n", __func__, con, msg); |
31b8006e | 1568 | |
0d9c1ab3 | 1569 | if (!msg->num_data_items) |
4c59b4a2 AE |
1570 | return -EINVAL; |
1571 | ||
5821bd8c AE |
1572 | /* |
1573 | * Iterate through each page that contains data to be | |
1574 | * written, and send as much as possible for each. | |
1575 | * | |
1576 | * If we are calculating the data crc (the default), we will | |
1577 | * need to map the page. If we have no pages, they have | |
1578 | * been revoked, so use the zero page. | |
1579 | */ | |
f5db90bc | 1580 | crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0; |
45a267db | 1581 | while (cursor->total_resid) { |
8a166d05 | 1582 | struct page *page; |
e387d525 AE |
1583 | size_t page_offset; |
1584 | size_t length; | |
f5db90bc | 1585 | int ret; |
68b4476b | 1586 | |
45a267db ID |
1587 | if (!cursor->resid) { |
1588 | ceph_msg_data_advance(cursor, 0); | |
1589 | continue; | |
1590 | } | |
1591 | ||
1f6b821a | 1592 | page = ceph_msg_data_next(cursor, &page_offset, &length, NULL); |
433b0a12 ID |
1593 | if (length == cursor->total_resid) |
1594 | more = MSG_MORE; | |
1f6b821a | 1595 | ret = ceph_tcp_sendpage(con->sock, page, page_offset, length, |
433b0a12 | 1596 | more); |
f5db90bc AE |
1597 | if (ret <= 0) { |
1598 | if (do_datacrc) | |
1599 | msg->footer.data_crc = cpu_to_le32(crc); | |
31b8006e | 1600 | |
f5db90bc AE |
1601 | return ret; |
1602 | } | |
143334ff AE |
1603 | if (do_datacrc && cursor->need_crc) |
1604 | crc = ceph_crc32c_page(crc, page, page_offset, length); | |
1759f7b0 | 1605 | ceph_msg_data_advance(cursor, (size_t)ret); |
31b8006e SW |
1606 | } |
1607 | ||
34d2d200 | 1608 | dout("%s %p msg %p done\n", __func__, con, msg); |
31b8006e SW |
1609 | |
1610 | /* prepare and queue up footer, too */ | |
f5db90bc AE |
1611 | if (do_datacrc) |
1612 | msg->footer.data_crc = cpu_to_le32(crc); | |
1613 | else | |
84ca8fc8 | 1614 | msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC; |
e2200423 | 1615 | con_out_kvec_reset(con); |
859eb799 | 1616 | prepare_write_message_footer(con); |
f5db90bc AE |
1617 | |
1618 | return 1; /* must return > 0 to indicate success */ | |
31b8006e SW |
1619 | } |
1620 | ||
1621 | /* | |
1622 | * write some zeros | |
1623 | */ | |
1624 | static int write_partial_skip(struct ceph_connection *con) | |
1625 | { | |
433b0a12 | 1626 | int more = MSG_MORE | MSG_SENDPAGE_NOTLAST; |
31b8006e SW |
1627 | int ret; |
1628 | ||
67645d76 | 1629 | dout("%s %p %d left\n", __func__, con, con->out_skip); |
31b8006e | 1630 | while (con->out_skip > 0) { |
09cbfeaf | 1631 | size_t size = min(con->out_skip, (int) PAGE_SIZE); |
31b8006e | 1632 | |
433b0a12 ID |
1633 | if (size == con->out_skip) |
1634 | more = MSG_MORE; | |
1635 | ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, more); | |
31b8006e SW |
1636 | if (ret <= 0) |
1637 | goto out; | |
1638 | con->out_skip -= ret; | |
1639 | } | |
1640 | ret = 1; | |
1641 | out: | |
1642 | return ret; | |
1643 | } | |
1644 | ||
1645 | /* | |
1646 | * Prepare to read connection handshake, or an ack. | |
1647 | */ | |
eed0ef2c SW |
1648 | static void prepare_read_banner(struct ceph_connection *con) |
1649 | { | |
1650 | dout("prepare_read_banner %p\n", con); | |
1651 | con->in_base_pos = 0; | |
1652 | } | |
1653 | ||
31b8006e SW |
1654 | static void prepare_read_connect(struct ceph_connection *con) |
1655 | { | |
1656 | dout("prepare_read_connect %p\n", con); | |
1657 | con->in_base_pos = 0; | |
1658 | } | |
1659 | ||
1660 | static void prepare_read_ack(struct ceph_connection *con) | |
1661 | { | |
1662 | dout("prepare_read_ack %p\n", con); | |
1663 | con->in_base_pos = 0; | |
1664 | } | |
1665 | ||
3a23083b SW |
1666 | static void prepare_read_seq(struct ceph_connection *con) |
1667 | { | |
1668 | dout("prepare_read_seq %p\n", con); | |
1669 | con->in_base_pos = 0; | |
1670 | con->in_tag = CEPH_MSGR_TAG_SEQ; | |
1671 | } | |
1672 | ||
31b8006e SW |
1673 | static void prepare_read_tag(struct ceph_connection *con) |
1674 | { | |
1675 | dout("prepare_read_tag %p\n", con); | |
1676 | con->in_base_pos = 0; | |
1677 | con->in_tag = CEPH_MSGR_TAG_READY; | |
1678 | } | |
1679 | ||
8b9558aa YZ |
1680 | static void prepare_read_keepalive_ack(struct ceph_connection *con) |
1681 | { | |
1682 | dout("prepare_read_keepalive_ack %p\n", con); | |
1683 | con->in_base_pos = 0; | |
1684 | } | |
1685 | ||
31b8006e SW |
1686 | /* |
1687 | * Prepare to read a message. | |
1688 | */ | |
1689 | static int prepare_read_message(struct ceph_connection *con) | |
1690 | { | |
1691 | dout("prepare_read_message %p\n", con); | |
1692 | BUG_ON(con->in_msg != NULL); | |
1693 | con->in_base_pos = 0; | |
1694 | con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0; | |
1695 | return 0; | |
1696 | } | |
1697 | ||
1698 | ||
1699 | static int read_partial(struct ceph_connection *con, | |
fd51653f | 1700 | int end, int size, void *object) |
31b8006e | 1701 | { |
e6cee71f AE |
1702 | while (con->in_base_pos < end) { |
1703 | int left = end - con->in_base_pos; | |
31b8006e SW |
1704 | int have = size - left; |
1705 | int ret = ceph_tcp_recvmsg(con->sock, object + have, left); | |
1706 | if (ret <= 0) | |
1707 | return ret; | |
1708 | con->in_base_pos += ret; | |
1709 | } | |
1710 | return 1; | |
1711 | } | |
1712 | ||
1713 | ||
1714 | /* | |
1715 | * Read all or part of the connect-side handshake on a new connection | |
1716 | */ | |
eed0ef2c | 1717 | static int read_partial_banner(struct ceph_connection *con) |
31b8006e | 1718 | { |
fd51653f AE |
1719 | int size; |
1720 | int end; | |
1721 | int ret; | |
31b8006e | 1722 | |
eed0ef2c | 1723 | dout("read_partial_banner %p at %d\n", con, con->in_base_pos); |
31b8006e SW |
1724 | |
1725 | /* peer's banner */ | |
fd51653f AE |
1726 | size = strlen(CEPH_BANNER); |
1727 | end = size; | |
1728 | ret = read_partial(con, end, size, con->in_banner); | |
31b8006e SW |
1729 | if (ret <= 0) |
1730 | goto out; | |
fd51653f AE |
1731 | |
1732 | size = sizeof (con->actual_peer_addr); | |
1733 | end += size; | |
1734 | ret = read_partial(con, end, size, &con->actual_peer_addr); | |
31b8006e SW |
1735 | if (ret <= 0) |
1736 | goto out; | |
2c66de56 | 1737 | ceph_decode_banner_addr(&con->actual_peer_addr); |
fd51653f AE |
1738 | |
1739 | size = sizeof (con->peer_addr_for_me); | |
1740 | end += size; | |
1741 | ret = read_partial(con, end, size, &con->peer_addr_for_me); | |
31b8006e SW |
1742 | if (ret <= 0) |
1743 | goto out; | |
2c66de56 | 1744 | ceph_decode_banner_addr(&con->peer_addr_for_me); |
fd51653f | 1745 | |
eed0ef2c SW |
1746 | out: |
1747 | return ret; | |
1748 | } | |
1749 | ||
1750 | static int read_partial_connect(struct ceph_connection *con) | |
1751 | { | |
fd51653f AE |
1752 | int size; |
1753 | int end; | |
1754 | int ret; | |
eed0ef2c SW |
1755 | |
1756 | dout("read_partial_connect %p at %d\n", con, con->in_base_pos); | |
1757 | ||
fd51653f AE |
1758 | size = sizeof (con->in_reply); |
1759 | end = size; | |
1760 | ret = read_partial(con, end, size, &con->in_reply); | |
31b8006e SW |
1761 | if (ret <= 0) |
1762 | goto out; | |
fd51653f | 1763 | |
262614c4 ID |
1764 | if (con->auth) { |
1765 | size = le32_to_cpu(con->in_reply.authorizer_len); | |
130f52f2 ID |
1766 | if (size > con->auth->authorizer_reply_buf_len) { |
1767 | pr_err("authorizer reply too big: %d > %zu\n", size, | |
1768 | con->auth->authorizer_reply_buf_len); | |
1769 | ret = -EINVAL; | |
1770 | goto out; | |
1771 | } | |
1772 | ||
262614c4 ID |
1773 | end += size; |
1774 | ret = read_partial(con, end, size, | |
1775 | con->auth->authorizer_reply_buf); | |
1776 | if (ret <= 0) | |
1777 | goto out; | |
1778 | } | |
31b8006e | 1779 | |
4e7a5dcd SW |
1780 | dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n", |
1781 | con, (int)con->in_reply.tag, | |
1782 | le32_to_cpu(con->in_reply.connect_seq), | |
31b8006e SW |
1783 | le32_to_cpu(con->in_reply.global_seq)); |
1784 | out: | |
1785 | return ret; | |
1786 | } | |
1787 | ||
1788 | /* | |
1789 | * Verify the hello banner looks okay. | |
1790 | */ | |
1791 | static int verify_hello(struct ceph_connection *con) | |
1792 | { | |
1793 | if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) { | |
13e38c8a | 1794 | pr_err("connect to %s got bad banner\n", |
b726ec97 | 1795 | ceph_pr_addr(&con->peer_addr)); |
31b8006e SW |
1796 | con->error_msg = "protocol error, bad banner"; |
1797 | return -1; | |
1798 | } | |
1799 | return 0; | |
1800 | } | |
1801 | ||
cede185b | 1802 | static bool addr_is_blank(struct ceph_entity_addr *addr) |
31b8006e | 1803 | { |
cede185b JL |
1804 | struct sockaddr_storage ss = addr->in_addr; /* align */ |
1805 | struct in_addr *addr4 = &((struct sockaddr_in *)&ss)->sin_addr; | |
1806 | struct in6_addr *addr6 = &((struct sockaddr_in6 *)&ss)->sin6_addr; | |
c44bd69c | 1807 | |
cede185b | 1808 | switch (ss.ss_family) { |
31b8006e | 1809 | case AF_INET: |
cede185b | 1810 | return addr4->s_addr == htonl(INADDR_ANY); |
31b8006e | 1811 | case AF_INET6: |
c44bd69c ID |
1812 | return ipv6_addr_any(addr6); |
1813 | default: | |
1814 | return true; | |
31b8006e | 1815 | } |
31b8006e SW |
1816 | } |
1817 | ||
cede185b | 1818 | static int addr_port(struct ceph_entity_addr *addr) |
31b8006e | 1819 | { |
cede185b | 1820 | switch (get_unaligned(&addr->in_addr.ss_family)) { |
31b8006e | 1821 | case AF_INET: |
cede185b | 1822 | return ntohs(get_unaligned(&((struct sockaddr_in *)&addr->in_addr)->sin_port)); |
31b8006e | 1823 | case AF_INET6: |
cede185b | 1824 | return ntohs(get_unaligned(&((struct sockaddr_in6 *)&addr->in_addr)->sin6_port)); |
31b8006e SW |
1825 | } |
1826 | return 0; | |
1827 | } | |
1828 | ||
cede185b | 1829 | static void addr_set_port(struct ceph_entity_addr *addr, int p) |
31b8006e | 1830 | { |
cede185b | 1831 | switch (get_unaligned(&addr->in_addr.ss_family)) { |
31b8006e | 1832 | case AF_INET: |
cede185b | 1833 | put_unaligned(htons(p), &((struct sockaddr_in *)&addr->in_addr)->sin_port); |
a2a79609 | 1834 | break; |
31b8006e | 1835 | case AF_INET6: |
cede185b | 1836 | put_unaligned(htons(p), &((struct sockaddr_in6 *)&addr->in_addr)->sin6_port); |
a2a79609 | 1837 | break; |
31b8006e SW |
1838 | } |
1839 | } | |
1840 | ||
ee3b56f2 NW |
1841 | /* |
1842 | * Unlike other *_pton function semantics, zero indicates success. | |
1843 | */ | |
cede185b | 1844 | static int ceph_pton(const char *str, size_t len, struct ceph_entity_addr *addr, |
ee3b56f2 NW |
1845 | char delim, const char **ipend) |
1846 | { | |
cede185b | 1847 | memset(&addr->in_addr, 0, sizeof(addr->in_addr)); |
ee3b56f2 | 1848 | |
cede185b JL |
1849 | if (in4_pton(str, len, (u8 *)&((struct sockaddr_in *)&addr->in_addr)->sin_addr.s_addr, delim, ipend)) { |
1850 | put_unaligned(AF_INET, &addr->in_addr.ss_family); | |
ee3b56f2 NW |
1851 | return 0; |
1852 | } | |
1853 | ||
cede185b JL |
1854 | if (in6_pton(str, len, (u8 *)&((struct sockaddr_in6 *)&addr->in_addr)->sin6_addr.s6_addr, delim, ipend)) { |
1855 | put_unaligned(AF_INET6, &addr->in_addr.ss_family); | |
ee3b56f2 NW |
1856 | return 0; |
1857 | } | |
1858 | ||
1859 | return -EINVAL; | |
1860 | } | |
1861 | ||
1862 | /* | |
1863 | * Extract hostname string and resolve using kernel DNS facility. | |
1864 | */ | |
1865 | #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER | |
1866 | static int ceph_dns_resolve_name(const char *name, size_t namelen, | |
cede185b | 1867 | struct ceph_entity_addr *addr, char delim, const char **ipend) |
ee3b56f2 NW |
1868 | { |
1869 | const char *end, *delim_p; | |
1870 | char *colon_p, *ip_addr = NULL; | |
1871 | int ip_len, ret; | |
1872 | ||
1873 | /* | |
1874 | * The end of the hostname occurs immediately preceding the delimiter or | |
1875 | * the port marker (':') where the delimiter takes precedence. | |
1876 | */ | |
1877 | delim_p = memchr(name, delim, namelen); | |
1878 | colon_p = memchr(name, ':', namelen); | |
1879 | ||
1880 | if (delim_p && colon_p) | |
1881 | end = delim_p < colon_p ? delim_p : colon_p; | |
1882 | else if (!delim_p && colon_p) | |
1883 | end = colon_p; | |
1884 | else { | |
1885 | end = delim_p; | |
1886 | if (!end) /* case: hostname:/ */ | |
1887 | end = name + namelen; | |
1888 | } | |
1889 | ||
1890 | if (end <= name) | |
1891 | return -EINVAL; | |
1892 | ||
1893 | /* do dns_resolve upcall */ | |
a58946c1 DH |
1894 | ip_len = dns_query(current->nsproxy->net_ns, |
1895 | NULL, name, end - name, NULL, &ip_addr, NULL, false); | |
ee3b56f2 | 1896 | if (ip_len > 0) |
cede185b | 1897 | ret = ceph_pton(ip_addr, ip_len, addr, -1, NULL); |
ee3b56f2 NW |
1898 | else |
1899 | ret = -ESRCH; | |
1900 | ||
1901 | kfree(ip_addr); | |
1902 | ||
1903 | *ipend = end; | |
1904 | ||
1905 | pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name, | |
b726ec97 | 1906 | ret, ret ? "failed" : ceph_pr_addr(addr)); |
ee3b56f2 NW |
1907 | |
1908 | return ret; | |
1909 | } | |
1910 | #else | |
1911 | static inline int ceph_dns_resolve_name(const char *name, size_t namelen, | |
cede185b | 1912 | struct ceph_entity_addr *addr, char delim, const char **ipend) |
ee3b56f2 NW |
1913 | { |
1914 | return -EINVAL; | |
1915 | } | |
1916 | #endif | |
1917 | ||
1918 | /* | |
1919 | * Parse a server name (IP or hostname). If a valid IP address is not found | |
1920 | * then try to extract a hostname to resolve using userspace DNS upcall. | |
1921 | */ | |
1922 | static int ceph_parse_server_name(const char *name, size_t namelen, | |
cede185b | 1923 | struct ceph_entity_addr *addr, char delim, const char **ipend) |
ee3b56f2 NW |
1924 | { |
1925 | int ret; | |
1926 | ||
cede185b | 1927 | ret = ceph_pton(name, namelen, addr, delim, ipend); |
ee3b56f2 | 1928 | if (ret) |
cede185b | 1929 | ret = ceph_dns_resolve_name(name, namelen, addr, delim, ipend); |
ee3b56f2 NW |
1930 | |
1931 | return ret; | |
1932 | } | |
1933 | ||
31b8006e SW |
1934 | /* |
1935 | * Parse an ip[:port] list into an addr array. Use the default | |
1936 | * monitor port if a port isn't specified. | |
1937 | */ | |
1938 | int ceph_parse_ips(const char *c, const char *end, | |
1939 | struct ceph_entity_addr *addr, | |
1940 | int max_count, int *count) | |
1941 | { | |
ee3b56f2 | 1942 | int i, ret = -EINVAL; |
31b8006e SW |
1943 | const char *p = c; |
1944 | ||
1945 | dout("parse_ips on '%.*s'\n", (int)(end-c), c); | |
1946 | for (i = 0; i < max_count; i++) { | |
1947 | const char *ipend; | |
31b8006e | 1948 | int port; |
39139f64 SW |
1949 | char delim = ','; |
1950 | ||
1951 | if (*p == '[') { | |
1952 | delim = ']'; | |
1953 | p++; | |
1954 | } | |
31b8006e | 1955 | |
cede185b | 1956 | ret = ceph_parse_server_name(p, end - p, &addr[i], delim, &ipend); |
ee3b56f2 | 1957 | if (ret) |
31b8006e | 1958 | goto bad; |
ee3b56f2 NW |
1959 | ret = -EINVAL; |
1960 | ||
31b8006e SW |
1961 | p = ipend; |
1962 | ||
39139f64 SW |
1963 | if (delim == ']') { |
1964 | if (*p != ']') { | |
1965 | dout("missing matching ']'\n"); | |
1966 | goto bad; | |
1967 | } | |
1968 | p++; | |
1969 | } | |
1970 | ||
31b8006e SW |
1971 | /* port? */ |
1972 | if (p < end && *p == ':') { | |
1973 | port = 0; | |
1974 | p++; | |
1975 | while (p < end && *p >= '0' && *p <= '9') { | |
1976 | port = (port * 10) + (*p - '0'); | |
1977 | p++; | |
1978 | } | |
f48db1e9 ID |
1979 | if (port == 0) |
1980 | port = CEPH_MON_PORT; | |
1981 | else if (port > 65535) | |
31b8006e SW |
1982 | goto bad; |
1983 | } else { | |
1984 | port = CEPH_MON_PORT; | |
1985 | } | |
1986 | ||
cede185b | 1987 | addr_set_port(&addr[i], port); |
d3c3c0a8 | 1988 | addr[i].type = CEPH_ENTITY_ADDR_TYPE_LEGACY; |
31b8006e | 1989 | |
b726ec97 | 1990 | dout("parse_ips got %s\n", ceph_pr_addr(&addr[i])); |
31b8006e SW |
1991 | |
1992 | if (p == end) | |
1993 | break; | |
1994 | if (*p != ',') | |
1995 | goto bad; | |
1996 | p++; | |
1997 | } | |
1998 | ||
1999 | if (p != end) | |
2000 | goto bad; | |
2001 | ||
2002 | if (count) | |
2003 | *count = i + 1; | |
2004 | return 0; | |
2005 | ||
2006 | bad: | |
39139f64 | 2007 | pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c); |
ee3b56f2 | 2008 | return ret; |
31b8006e | 2009 | } |
3d14c5d2 | 2010 | EXPORT_SYMBOL(ceph_parse_ips); |
31b8006e | 2011 | |
eed0ef2c | 2012 | static int process_banner(struct ceph_connection *con) |
31b8006e | 2013 | { |
eed0ef2c | 2014 | dout("process_banner on %p\n", con); |
31b8006e SW |
2015 | |
2016 | if (verify_hello(con) < 0) | |
2017 | return -1; | |
2018 | ||
2019 | /* | |
2020 | * Make sure the other end is who we wanted. note that the other | |
2021 | * end may not yet know their ip address, so if it's 0.0.0.0, give | |
2022 | * them the benefit of the doubt. | |
2023 | */ | |
103e2d3a SW |
2024 | if (memcmp(&con->peer_addr, &con->actual_peer_addr, |
2025 | sizeof(con->peer_addr)) != 0 && | |
cede185b | 2026 | !(addr_is_blank(&con->actual_peer_addr) && |
31b8006e | 2027 | con->actual_peer_addr.nonce == con->peer_addr.nonce)) { |
b9a67899 | 2028 | pr_warn("wrong peer, want %s/%d, got %s/%d\n", |
b726ec97 | 2029 | ceph_pr_addr(&con->peer_addr), |
b9a67899 | 2030 | (int)le32_to_cpu(con->peer_addr.nonce), |
b726ec97 | 2031 | ceph_pr_addr(&con->actual_peer_addr), |
b9a67899 | 2032 | (int)le32_to_cpu(con->actual_peer_addr.nonce)); |
58bb3b37 | 2033 | con->error_msg = "wrong peer at address"; |
31b8006e SW |
2034 | return -1; |
2035 | } | |
2036 | ||
2037 | /* | |
2038 | * did we learn our address? | |
2039 | */ | |
cede185b JL |
2040 | if (addr_is_blank(&con->msgr->inst.addr)) { |
2041 | int port = addr_port(&con->msgr->inst.addr); | |
31b8006e SW |
2042 | |
2043 | memcpy(&con->msgr->inst.addr.in_addr, | |
2044 | &con->peer_addr_for_me.in_addr, | |
2045 | sizeof(con->peer_addr_for_me.in_addr)); | |
cede185b | 2046 | addr_set_port(&con->msgr->inst.addr, port); |
63f2d211 | 2047 | encode_my_addr(con->msgr); |
eed0ef2c | 2048 | dout("process_banner learned my addr is %s\n", |
b726ec97 | 2049 | ceph_pr_addr(&con->msgr->inst.addr)); |
31b8006e SW |
2050 | } |
2051 | ||
eed0ef2c SW |
2052 | return 0; |
2053 | } | |
2054 | ||
2055 | static int process_connect(struct ceph_connection *con) | |
2056 | { | |
859bff51 ID |
2057 | u64 sup_feat = from_msgr(con->msgr)->supported_features; |
2058 | u64 req_feat = from_msgr(con->msgr)->required_features; | |
dcbbd97c | 2059 | u64 server_feat = le64_to_cpu(con->in_reply.features); |
0da5d703 | 2060 | int ret; |
04a419f9 | 2061 | |
eed0ef2c SW |
2062 | dout("process_connect on %p tag %d\n", con, (int)con->in_tag); |
2063 | ||
262614c4 | 2064 | if (con->auth) { |
0fd3fd0a ID |
2065 | int len = le32_to_cpu(con->in_reply.authorizer_len); |
2066 | ||
5c056fdc ID |
2067 | /* |
2068 | * Any connection that defines ->get_authorizer() | |
6daca13d ID |
2069 | * should also define ->add_authorizer_challenge() and |
2070 | * ->verify_authorizer_reply(). | |
2071 | * | |
5c056fdc ID |
2072 | * See get_connect_authorizer(). |
2073 | */ | |
6daca13d ID |
2074 | if (con->in_reply.tag == CEPH_MSGR_TAG_CHALLENGE_AUTHORIZER) { |
2075 | ret = con->ops->add_authorizer_challenge( | |
0fd3fd0a | 2076 | con, con->auth->authorizer_reply_buf, len); |
6daca13d ID |
2077 | if (ret < 0) |
2078 | return ret; | |
2079 | ||
2080 | con_out_kvec_reset(con); | |
2081 | __prepare_write_connect(con); | |
2082 | prepare_read_connect(con); | |
2083 | return 0; | |
2084 | } | |
2085 | ||
0fd3fd0a ID |
2086 | if (len) { |
2087 | ret = con->ops->verify_authorizer_reply(con); | |
2088 | if (ret < 0) { | |
2089 | con->error_msg = "bad authorize reply"; | |
2090 | return ret; | |
2091 | } | |
5c056fdc ID |
2092 | } |
2093 | } | |
2094 | ||
31b8006e | 2095 | switch (con->in_reply.tag) { |
04a419f9 SW |
2096 | case CEPH_MSGR_TAG_FEATURES: |
2097 | pr_err("%s%lld %s feature set mismatch," | |
2098 | " my %llx < server's %llx, missing %llx\n", | |
2099 | ENTITY_NAME(con->peer_name), | |
b726ec97 | 2100 | ceph_pr_addr(&con->peer_addr), |
04a419f9 SW |
2101 | sup_feat, server_feat, server_feat & ~sup_feat); |
2102 | con->error_msg = "missing required protocol features"; | |
0fa6ebc6 | 2103 | reset_connection(con); |
04a419f9 SW |
2104 | return -1; |
2105 | ||
31b8006e | 2106 | case CEPH_MSGR_TAG_BADPROTOVER: |
31b8006e SW |
2107 | pr_err("%s%lld %s protocol version mismatch," |
2108 | " my %d != server's %d\n", | |
2109 | ENTITY_NAME(con->peer_name), | |
b726ec97 | 2110 | ceph_pr_addr(&con->peer_addr), |
31b8006e SW |
2111 | le32_to_cpu(con->out_connect.protocol_version), |
2112 | le32_to_cpu(con->in_reply.protocol_version)); | |
2113 | con->error_msg = "protocol version mismatch"; | |
0fa6ebc6 | 2114 | reset_connection(con); |
31b8006e SW |
2115 | return -1; |
2116 | ||
4e7a5dcd SW |
2117 | case CEPH_MSGR_TAG_BADAUTHORIZER: |
2118 | con->auth_retry++; | |
2119 | dout("process_connect %p got BADAUTHORIZER attempt %d\n", con, | |
2120 | con->auth_retry); | |
2121 | if (con->auth_retry == 2) { | |
2122 | con->error_msg = "connect authorization failure"; | |
4e7a5dcd SW |
2123 | return -1; |
2124 | } | |
6d4221b5 | 2125 | con_out_kvec_reset(con); |
e825a66d | 2126 | ret = prepare_write_connect(con); |
0da5d703 SW |
2127 | if (ret < 0) |
2128 | return ret; | |
63733a0f | 2129 | prepare_read_connect(con); |
4e7a5dcd | 2130 | break; |
31b8006e SW |
2131 | |
2132 | case CEPH_MSGR_TAG_RESETSESSION: | |
2133 | /* | |
2134 | * If we connected with a large connect_seq but the peer | |
2135 | * has no record of a session with us (no connection, or | |
2136 | * connect_seq == 0), they will send RESETSESION to indicate | |
2137 | * that they must have reset their session, and may have | |
2138 | * dropped messages. | |
2139 | */ | |
2140 | dout("process_connect got RESET peer seq %u\n", | |
5bdca4e0 | 2141 | le32_to_cpu(con->in_reply.connect_seq)); |
31b8006e SW |
2142 | pr_err("%s%lld %s connection reset\n", |
2143 | ENTITY_NAME(con->peer_name), | |
b726ec97 | 2144 | ceph_pr_addr(&con->peer_addr)); |
31b8006e | 2145 | reset_connection(con); |
6d4221b5 | 2146 | con_out_kvec_reset(con); |
5a0f8fdd AE |
2147 | ret = prepare_write_connect(con); |
2148 | if (ret < 0) | |
2149 | return ret; | |
31b8006e SW |
2150 | prepare_read_connect(con); |
2151 | ||
2152 | /* Tell ceph about it. */ | |
ec302645 | 2153 | mutex_unlock(&con->mutex); |
31b8006e SW |
2154 | pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name)); |
2155 | if (con->ops->peer_reset) | |
2156 | con->ops->peer_reset(con); | |
ec302645 | 2157 | mutex_lock(&con->mutex); |
8dacc7da | 2158 | if (con->state != CON_STATE_NEGOTIATING) |
0da5d703 | 2159 | return -EAGAIN; |
31b8006e SW |
2160 | break; |
2161 | ||
2162 | case CEPH_MSGR_TAG_RETRY_SESSION: | |
2163 | /* | |
2164 | * If we sent a smaller connect_seq than the peer has, try | |
2165 | * again with a larger value. | |
2166 | */ | |
5bdca4e0 | 2167 | dout("process_connect got RETRY_SESSION my seq %u, peer %u\n", |
31b8006e | 2168 | le32_to_cpu(con->out_connect.connect_seq), |
5bdca4e0 SW |
2169 | le32_to_cpu(con->in_reply.connect_seq)); |
2170 | con->connect_seq = le32_to_cpu(con->in_reply.connect_seq); | |
6d4221b5 | 2171 | con_out_kvec_reset(con); |
5a0f8fdd AE |
2172 | ret = prepare_write_connect(con); |
2173 | if (ret < 0) | |
2174 | return ret; | |
31b8006e SW |
2175 | prepare_read_connect(con); |
2176 | break; | |
2177 | ||
2178 | case CEPH_MSGR_TAG_RETRY_GLOBAL: | |
2179 | /* | |
2180 | * If we sent a smaller global_seq than the peer has, try | |
2181 | * again with a larger value. | |
2182 | */ | |
eed0ef2c | 2183 | dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n", |
31b8006e | 2184 | con->peer_global_seq, |
5bdca4e0 | 2185 | le32_to_cpu(con->in_reply.global_seq)); |
31b8006e | 2186 | get_global_seq(con->msgr, |
5bdca4e0 | 2187 | le32_to_cpu(con->in_reply.global_seq)); |
6d4221b5 | 2188 | con_out_kvec_reset(con); |
5a0f8fdd AE |
2189 | ret = prepare_write_connect(con); |
2190 | if (ret < 0) | |
2191 | return ret; | |
31b8006e SW |
2192 | prepare_read_connect(con); |
2193 | break; | |
2194 | ||
3a23083b | 2195 | case CEPH_MSGR_TAG_SEQ: |
31b8006e | 2196 | case CEPH_MSGR_TAG_READY: |
04a419f9 SW |
2197 | if (req_feat & ~server_feat) { |
2198 | pr_err("%s%lld %s protocol feature mismatch," | |
2199 | " my required %llx > server's %llx, need %llx\n", | |
2200 | ENTITY_NAME(con->peer_name), | |
b726ec97 | 2201 | ceph_pr_addr(&con->peer_addr), |
04a419f9 SW |
2202 | req_feat, server_feat, req_feat & ~server_feat); |
2203 | con->error_msg = "missing required protocol features"; | |
0fa6ebc6 | 2204 | reset_connection(con); |
04a419f9 SW |
2205 | return -1; |
2206 | } | |
8dacc7da | 2207 | |
122070a2 | 2208 | WARN_ON(con->state != CON_STATE_NEGOTIATING); |
8dacc7da | 2209 | con->state = CON_STATE_OPEN; |
20e55c4c | 2210 | con->auth_retry = 0; /* we authenticated; clear flag */ |
31b8006e SW |
2211 | con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq); |
2212 | con->connect_seq++; | |
aba558e2 | 2213 | con->peer_features = server_feat; |
31b8006e SW |
2214 | dout("process_connect got READY gseq %d cseq %d (%d)\n", |
2215 | con->peer_global_seq, | |
2216 | le32_to_cpu(con->in_reply.connect_seq), | |
2217 | con->connect_seq); | |
2218 | WARN_ON(con->connect_seq != | |
2219 | le32_to_cpu(con->in_reply.connect_seq)); | |
92ac41d0 SW |
2220 | |
2221 | if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY) | |
c9ffc77a | 2222 | con_flag_set(con, CON_FLAG_LOSSYTX); |
92ac41d0 | 2223 | |
85effe18 | 2224 | con->delay = 0; /* reset backoff memory */ |
92ac41d0 | 2225 | |
3a23083b SW |
2226 | if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) { |
2227 | prepare_write_seq(con); | |
2228 | prepare_read_seq(con); | |
2229 | } else { | |
2230 | prepare_read_tag(con); | |
2231 | } | |
31b8006e SW |
2232 | break; |
2233 | ||
2234 | case CEPH_MSGR_TAG_WAIT: | |
2235 | /* | |
2236 | * If there is a connection race (we are opening | |
2237 | * connections to each other), one of us may just have | |
2238 | * to WAIT. This shouldn't happen if we are the | |
2239 | * client. | |
2240 | */ | |
04177882 SW |
2241 | con->error_msg = "protocol error, got WAIT as client"; |
2242 | return -1; | |
31b8006e SW |
2243 | |
2244 | default: | |
31b8006e SW |
2245 | con->error_msg = "protocol error, garbage tag during connect"; |
2246 | return -1; | |
2247 | } | |
2248 | return 0; | |
2249 | } | |
2250 | ||
2251 | ||
2252 | /* | |
2253 | * read (part of) an ack | |
2254 | */ | |
2255 | static int read_partial_ack(struct ceph_connection *con) | |
2256 | { | |
fd51653f AE |
2257 | int size = sizeof (con->in_temp_ack); |
2258 | int end = size; | |
31b8006e | 2259 | |
fd51653f | 2260 | return read_partial(con, end, size, &con->in_temp_ack); |
31b8006e SW |
2261 | } |
2262 | ||
31b8006e SW |
2263 | /* |
2264 | * We can finally discard anything that's been acked. | |
2265 | */ | |
2266 | static void process_ack(struct ceph_connection *con) | |
2267 | { | |
2268 | struct ceph_msg *m; | |
2269 | u64 ack = le64_to_cpu(con->in_temp_ack); | |
2270 | u64 seq; | |
0a2ad541 YZ |
2271 | bool reconnect = (con->in_tag == CEPH_MSGR_TAG_SEQ); |
2272 | struct list_head *list = reconnect ? &con->out_queue : &con->out_sent; | |
31b8006e | 2273 | |
0a2ad541 YZ |
2274 | /* |
2275 | * In the reconnect case, con_fault() has requeued messages | |
2276 | * in out_sent. We should cleanup old messages according to | |
2277 | * the reconnect seq. | |
2278 | */ | |
2279 | while (!list_empty(list)) { | |
2280 | m = list_first_entry(list, struct ceph_msg, list_head); | |
2281 | if (reconnect && m->needs_out_seq) | |
2282 | break; | |
31b8006e SW |
2283 | seq = le64_to_cpu(m->hdr.seq); |
2284 | if (seq > ack) | |
2285 | break; | |
2286 | dout("got ack for seq %llu type %d at %p\n", seq, | |
2287 | le16_to_cpu(m->hdr.type), m); | |
4cf9d544 | 2288 | m->ack_stamp = jiffies; |
31b8006e SW |
2289 | ceph_msg_remove(m); |
2290 | } | |
0a2ad541 | 2291 | |
31b8006e SW |
2292 | prepare_read_tag(con); |
2293 | } | |
2294 | ||
2295 | ||
2450418c | 2296 | static int read_partial_message_section(struct ceph_connection *con, |
213c99ee SW |
2297 | struct kvec *section, |
2298 | unsigned int sec_len, u32 *crc) | |
2450418c | 2299 | { |
68b4476b | 2300 | int ret, left; |
2450418c YS |
2301 | |
2302 | BUG_ON(!section); | |
2303 | ||
2304 | while (section->iov_len < sec_len) { | |
2305 | BUG_ON(section->iov_base == NULL); | |
2306 | left = sec_len - section->iov_len; | |
2307 | ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base + | |
2308 | section->iov_len, left); | |
2309 | if (ret <= 0) | |
2310 | return ret; | |
2311 | section->iov_len += ret; | |
2450418c | 2312 | } |
fe3ad593 AE |
2313 | if (section->iov_len == sec_len) |
2314 | *crc = crc32c(0, section->iov_base, section->iov_len); | |
31b8006e | 2315 | |
2450418c YS |
2316 | return 1; |
2317 | } | |
31b8006e | 2318 | |
34d2d200 AE |
2319 | static int read_partial_msg_data(struct ceph_connection *con) |
2320 | { | |
2321 | struct ceph_msg *msg = con->in_msg; | |
8ae4f4f5 | 2322 | struct ceph_msg_data_cursor *cursor = &msg->cursor; |
859bff51 | 2323 | bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC); |
686be208 AE |
2324 | struct page *page; |
2325 | size_t page_offset; | |
2326 | size_t length; | |
f5db90bc | 2327 | u32 crc = 0; |
34d2d200 AE |
2328 | int ret; |
2329 | ||
0d9c1ab3 | 2330 | if (!msg->num_data_items) |
4c59b4a2 | 2331 | return -EIO; |
34d2d200 | 2332 | |
f5db90bc AE |
2333 | if (do_datacrc) |
2334 | crc = con->in_data_crc; | |
45a267db ID |
2335 | while (cursor->total_resid) { |
2336 | if (!cursor->resid) { | |
2337 | ceph_msg_data_advance(cursor, 0); | |
2338 | continue; | |
2339 | } | |
2340 | ||
343128ce | 2341 | page = ceph_msg_data_next(cursor, &page_offset, &length, NULL); |
686be208 | 2342 | ret = ceph_tcp_recvpage(con->sock, page, page_offset, length); |
f5db90bc AE |
2343 | if (ret <= 0) { |
2344 | if (do_datacrc) | |
2345 | con->in_data_crc = crc; | |
2346 | ||
686be208 | 2347 | return ret; |
f5db90bc | 2348 | } |
686be208 AE |
2349 | |
2350 | if (do_datacrc) | |
f5db90bc | 2351 | crc = ceph_crc32c_page(crc, page, page_offset, ret); |
1759f7b0 | 2352 | ceph_msg_data_advance(cursor, (size_t)ret); |
34d2d200 | 2353 | } |
f5db90bc AE |
2354 | if (do_datacrc) |
2355 | con->in_data_crc = crc; | |
34d2d200 AE |
2356 | |
2357 | return 1; /* must return > 0 to indicate success */ | |
2358 | } | |
2359 | ||
31b8006e SW |
2360 | /* |
2361 | * read (part of) a message. | |
2362 | */ | |
686be208 AE |
2363 | static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip); |
2364 | ||
31b8006e SW |
2365 | static int read_partial_message(struct ceph_connection *con) |
2366 | { | |
2367 | struct ceph_msg *m = con->in_msg; | |
fd51653f AE |
2368 | int size; |
2369 | int end; | |
31b8006e | 2370 | int ret; |
95c96174 | 2371 | unsigned int front_len, middle_len, data_len; |
859bff51 | 2372 | bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC); |
33d07337 | 2373 | bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH); |
ae18756b | 2374 | u64 seq; |
fe3ad593 | 2375 | u32 crc; |
31b8006e SW |
2376 | |
2377 | dout("read_partial_message con %p msg %p\n", con, m); | |
2378 | ||
2379 | /* header */ | |
fd51653f AE |
2380 | size = sizeof (con->in_hdr); |
2381 | end = size; | |
2382 | ret = read_partial(con, end, size, &con->in_hdr); | |
57dac9d1 AE |
2383 | if (ret <= 0) |
2384 | return ret; | |
fe3ad593 AE |
2385 | |
2386 | crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc)); | |
2387 | if (cpu_to_le32(crc) != con->in_hdr.crc) { | |
67c64eb7 | 2388 | pr_err("read_partial_message bad hdr crc %u != expected %u\n", |
fe3ad593 AE |
2389 | crc, con->in_hdr.crc); |
2390 | return -EBADMSG; | |
2391 | } | |
2392 | ||
31b8006e SW |
2393 | front_len = le32_to_cpu(con->in_hdr.front_len); |
2394 | if (front_len > CEPH_MSG_MAX_FRONT_LEN) | |
2395 | return -EIO; | |
2396 | middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
7b11ba37 | 2397 | if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN) |
31b8006e SW |
2398 | return -EIO; |
2399 | data_len = le32_to_cpu(con->in_hdr.data_len); | |
2400 | if (data_len > CEPH_MSG_MAX_DATA_LEN) | |
2401 | return -EIO; | |
2402 | ||
ae18756b SW |
2403 | /* verify seq# */ |
2404 | seq = le64_to_cpu(con->in_hdr.seq); | |
2405 | if ((s64)seq - (s64)con->in_seq < 1) { | |
df9f86fa | 2406 | pr_info("skipping %s%lld %s seq %lld expected %lld\n", |
ae18756b | 2407 | ENTITY_NAME(con->peer_name), |
b726ec97 | 2408 | ceph_pr_addr(&con->peer_addr), |
ae18756b SW |
2409 | seq, con->in_seq + 1); |
2410 | con->in_base_pos = -front_len - middle_len - data_len - | |
dbc0d3ca | 2411 | sizeof_footer(con); |
ae18756b | 2412 | con->in_tag = CEPH_MSGR_TAG_READY; |
e7a88e82 | 2413 | return 1; |
ae18756b SW |
2414 | } else if ((s64)seq - (s64)con->in_seq > 1) { |
2415 | pr_err("read_partial_message bad seq %lld expected %lld\n", | |
2416 | seq, con->in_seq + 1); | |
2417 | con->error_msg = "bad message sequence # for incoming message"; | |
67c64eb7 | 2418 | return -EBADE; |
ae18756b SW |
2419 | } |
2420 | ||
31b8006e SW |
2421 | /* allocate message? */ |
2422 | if (!con->in_msg) { | |
4740a623 SW |
2423 | int skip = 0; |
2424 | ||
31b8006e | 2425 | dout("got hdr type %d front %d data %d\n", con->in_hdr.type, |
6ebc8b32 | 2426 | front_len, data_len); |
4740a623 SW |
2427 | ret = ceph_con_in_msg_alloc(con, &skip); |
2428 | if (ret < 0) | |
2429 | return ret; | |
f759ebb9 AE |
2430 | |
2431 | BUG_ON(!con->in_msg ^ skip); | |
2450418c | 2432 | if (skip) { |
31b8006e | 2433 | /* skip this message */ |
a79832f2 | 2434 | dout("alloc_msg said skip message\n"); |
31b8006e | 2435 | con->in_base_pos = -front_len - middle_len - data_len - |
dbc0d3ca | 2436 | sizeof_footer(con); |
31b8006e | 2437 | con->in_tag = CEPH_MSGR_TAG_READY; |
684be25c | 2438 | con->in_seq++; |
e7a88e82 | 2439 | return 1; |
31b8006e | 2440 | } |
38941f80 | 2441 | |
4740a623 | 2442 | BUG_ON(!con->in_msg); |
38941f80 | 2443 | BUG_ON(con->in_msg->con != con); |
31b8006e SW |
2444 | m = con->in_msg; |
2445 | m->front.iov_len = 0; /* haven't read it yet */ | |
2450418c YS |
2446 | if (m->middle) |
2447 | m->middle->vec.iov_len = 0; | |
9d7f0f13 | 2448 | |
78625051 | 2449 | /* prepare for data payload, if any */ |
a4107026 | 2450 | |
78625051 | 2451 | if (data_len) |
98fa5dd8 | 2452 | prepare_message_data(con->in_msg, data_len); |
31b8006e SW |
2453 | } |
2454 | ||
2455 | /* front */ | |
2450418c YS |
2456 | ret = read_partial_message_section(con, &m->front, front_len, |
2457 | &con->in_front_crc); | |
2458 | if (ret <= 0) | |
2459 | return ret; | |
31b8006e SW |
2460 | |
2461 | /* middle */ | |
2450418c | 2462 | if (m->middle) { |
213c99ee SW |
2463 | ret = read_partial_message_section(con, &m->middle->vec, |
2464 | middle_len, | |
2450418c | 2465 | &con->in_middle_crc); |
31b8006e SW |
2466 | if (ret <= 0) |
2467 | return ret; | |
31b8006e SW |
2468 | } |
2469 | ||
2470 | /* (page) data */ | |
34d2d200 AE |
2471 | if (data_len) { |
2472 | ret = read_partial_msg_data(con); | |
2473 | if (ret <= 0) | |
2474 | return ret; | |
31b8006e SW |
2475 | } |
2476 | ||
31b8006e | 2477 | /* footer */ |
89f08173 | 2478 | size = sizeof_footer(con); |
fd51653f AE |
2479 | end += size; |
2480 | ret = read_partial(con, end, size, &m->footer); | |
57dac9d1 AE |
2481 | if (ret <= 0) |
2482 | return ret; | |
2483 | ||
33d07337 YZ |
2484 | if (!need_sign) { |
2485 | m->footer.flags = m->old_footer.flags; | |
2486 | m->footer.sig = 0; | |
2487 | } | |
2488 | ||
31b8006e SW |
2489 | dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n", |
2490 | m, front_len, m->footer.front_crc, middle_len, | |
2491 | m->footer.middle_crc, data_len, m->footer.data_crc); | |
2492 | ||
2493 | /* crc ok? */ | |
2494 | if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) { | |
2495 | pr_err("read_partial_message %p front crc %u != exp. %u\n", | |
2496 | m, con->in_front_crc, m->footer.front_crc); | |
2497 | return -EBADMSG; | |
2498 | } | |
2499 | if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) { | |
2500 | pr_err("read_partial_message %p middle crc %u != exp %u\n", | |
2501 | m, con->in_middle_crc, m->footer.middle_crc); | |
2502 | return -EBADMSG; | |
2503 | } | |
bca064d2 | 2504 | if (do_datacrc && |
31b8006e SW |
2505 | (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 && |
2506 | con->in_data_crc != le32_to_cpu(m->footer.data_crc)) { | |
2507 | pr_err("read_partial_message %p data crc %u != exp. %u\n", m, | |
2508 | con->in_data_crc, le32_to_cpu(m->footer.data_crc)); | |
2509 | return -EBADMSG; | |
2510 | } | |
2511 | ||
33d07337 | 2512 | if (need_sign && con->ops->check_message_signature && |
79dbd1ba | 2513 | con->ops->check_message_signature(m)) { |
33d07337 YZ |
2514 | pr_err("read_partial_message %p signature check failed\n", m); |
2515 | return -EBADMSG; | |
2516 | } | |
2517 | ||
31b8006e SW |
2518 | return 1; /* done! */ |
2519 | } | |
2520 | ||
2521 | /* | |
2522 | * Process message. This happens in the worker thread. The callback should | |
2523 | * be careful not to do anything that waits on other incoming messages or it | |
2524 | * may deadlock. | |
2525 | */ | |
2526 | static void process_message(struct ceph_connection *con) | |
2527 | { | |
583d0fef | 2528 | struct ceph_msg *msg = con->in_msg; |
31b8006e | 2529 | |
38941f80 | 2530 | BUG_ON(con->in_msg->con != con); |
31b8006e SW |
2531 | con->in_msg = NULL; |
2532 | ||
2533 | /* if first message, set peer_name */ | |
2534 | if (con->peer_name.type == 0) | |
dbad185d | 2535 | con->peer_name = msg->hdr.src; |
31b8006e | 2536 | |
31b8006e | 2537 | con->in_seq++; |
ec302645 | 2538 | mutex_unlock(&con->mutex); |
31b8006e SW |
2539 | |
2540 | dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n", | |
2541 | msg, le64_to_cpu(msg->hdr.seq), | |
dbad185d | 2542 | ENTITY_NAME(msg->hdr.src), |
31b8006e SW |
2543 | le16_to_cpu(msg->hdr.type), |
2544 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
2545 | le32_to_cpu(msg->hdr.front_len), | |
2546 | le32_to_cpu(msg->hdr.data_len), | |
2547 | con->in_front_crc, con->in_middle_crc, con->in_data_crc); | |
2548 | con->ops->dispatch(con, msg); | |
ec302645 SW |
2549 | |
2550 | mutex_lock(&con->mutex); | |
31b8006e SW |
2551 | } |
2552 | ||
8b9558aa YZ |
2553 | static int read_keepalive_ack(struct ceph_connection *con) |
2554 | { | |
2555 | struct ceph_timespec ceph_ts; | |
2556 | size_t size = sizeof(ceph_ts); | |
2557 | int ret = read_partial(con, size, size, &ceph_ts); | |
2558 | if (ret <= 0) | |
2559 | return ret; | |
473bd2d7 | 2560 | ceph_decode_timespec64(&con->last_keepalive_ack, &ceph_ts); |
8b9558aa YZ |
2561 | prepare_read_tag(con); |
2562 | return 1; | |
2563 | } | |
31b8006e SW |
2564 | |
2565 | /* | |
2566 | * Write something to the socket. Called in a worker thread when the | |
2567 | * socket appears to be writeable and we have something ready to send. | |
2568 | */ | |
2569 | static int try_write(struct ceph_connection *con) | |
2570 | { | |
31b8006e SW |
2571 | int ret = 1; |
2572 | ||
d59315ca | 2573 | dout("try_write start %p state %lu\n", con, con->state); |
9c55ad1c ID |
2574 | if (con->state != CON_STATE_PREOPEN && |
2575 | con->state != CON_STATE_CONNECTING && | |
2576 | con->state != CON_STATE_NEGOTIATING && | |
2577 | con->state != CON_STATE_OPEN) | |
2578 | return 0; | |
31b8006e | 2579 | |
31b8006e | 2580 | /* open the socket first? */ |
8dacc7da SW |
2581 | if (con->state == CON_STATE_PREOPEN) { |
2582 | BUG_ON(con->sock); | |
2583 | con->state = CON_STATE_CONNECTING; | |
a5988c49 | 2584 | |
e2200423 | 2585 | con_out_kvec_reset(con); |
e825a66d | 2586 | prepare_write_banner(con); |
eed0ef2c | 2587 | prepare_read_banner(con); |
31b8006e | 2588 | |
cf3e5c40 | 2589 | BUG_ON(con->in_msg); |
31b8006e SW |
2590 | con->in_tag = CEPH_MSGR_TAG_READY; |
2591 | dout("try_write initiating connect on %p new state %lu\n", | |
2592 | con, con->state); | |
41617d0c AE |
2593 | ret = ceph_tcp_connect(con); |
2594 | if (ret < 0) { | |
31b8006e | 2595 | con->error_msg = "connect error"; |
31b8006e SW |
2596 | goto out; |
2597 | } | |
2598 | } | |
2599 | ||
d2935d6f ID |
2600 | more: |
2601 | dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes); | |
9c55ad1c ID |
2602 | BUG_ON(!con->sock); |
2603 | ||
31b8006e | 2604 | /* kvec data queued? */ |
67645d76 ID |
2605 | if (con->out_kvec_left) { |
2606 | ret = write_partial_kvec(con); | |
31b8006e | 2607 | if (ret <= 0) |
42961d23 | 2608 | goto out; |
31b8006e | 2609 | } |
67645d76 ID |
2610 | if (con->out_skip) { |
2611 | ret = write_partial_skip(con); | |
31b8006e | 2612 | if (ret <= 0) |
42961d23 | 2613 | goto out; |
31b8006e SW |
2614 | } |
2615 | ||
2616 | /* msg pages? */ | |
2617 | if (con->out_msg) { | |
c86a2930 SW |
2618 | if (con->out_msg_done) { |
2619 | ceph_msg_put(con->out_msg); | |
2620 | con->out_msg = NULL; /* we're done with this one */ | |
2621 | goto do_next; | |
2622 | } | |
2623 | ||
34d2d200 | 2624 | ret = write_partial_message_data(con); |
31b8006e | 2625 | if (ret == 1) |
d2935d6f | 2626 | goto more; /* we need to send the footer, too! */ |
31b8006e | 2627 | if (ret == 0) |
42961d23 | 2628 | goto out; |
31b8006e | 2629 | if (ret < 0) { |
34d2d200 | 2630 | dout("try_write write_partial_message_data err %d\n", |
31b8006e | 2631 | ret); |
42961d23 | 2632 | goto out; |
31b8006e SW |
2633 | } |
2634 | } | |
2635 | ||
c86a2930 | 2636 | do_next: |
8dacc7da | 2637 | if (con->state == CON_STATE_OPEN) { |
8b9558aa YZ |
2638 | if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) { |
2639 | prepare_write_keepalive(con); | |
2640 | goto more; | |
2641 | } | |
31b8006e SW |
2642 | /* is anything else pending? */ |
2643 | if (!list_empty(&con->out_queue)) { | |
2644 | prepare_write_message(con); | |
2645 | goto more; | |
2646 | } | |
2647 | if (con->in_seq > con->in_seq_acked) { | |
2648 | prepare_write_ack(con); | |
2649 | goto more; | |
2650 | } | |
31b8006e SW |
2651 | } |
2652 | ||
2653 | /* Nothing to do! */ | |
c9ffc77a | 2654 | con_flag_clear(con, CON_FLAG_WRITE_PENDING); |
31b8006e | 2655 | dout("try_write nothing else to write.\n"); |
31b8006e SW |
2656 | ret = 0; |
2657 | out: | |
42961d23 | 2658 | dout("try_write done on %p ret %d\n", con, ret); |
31b8006e SW |
2659 | return ret; |
2660 | } | |
2661 | ||
31b8006e SW |
2662 | /* |
2663 | * Read what we can from the socket. | |
2664 | */ | |
2665 | static int try_read(struct ceph_connection *con) | |
2666 | { | |
31b8006e SW |
2667 | int ret = -1; |
2668 | ||
8dacc7da SW |
2669 | more: |
2670 | dout("try_read start on %p state %lu\n", con, con->state); | |
2671 | if (con->state != CON_STATE_CONNECTING && | |
2672 | con->state != CON_STATE_NEGOTIATING && | |
2673 | con->state != CON_STATE_OPEN) | |
31b8006e SW |
2674 | return 0; |
2675 | ||
8dacc7da | 2676 | BUG_ON(!con->sock); |
ec302645 | 2677 | |
31b8006e SW |
2678 | dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag, |
2679 | con->in_base_pos); | |
0da5d703 | 2680 | |
8dacc7da | 2681 | if (con->state == CON_STATE_CONNECTING) { |
7593af92 AE |
2682 | dout("try_read connecting\n"); |
2683 | ret = read_partial_banner(con); | |
2684 | if (ret <= 0) | |
ab166d5a | 2685 | goto out; |
7593af92 AE |
2686 | ret = process_banner(con); |
2687 | if (ret < 0) | |
2688 | goto out; | |
2689 | ||
8dacc7da | 2690 | con->state = CON_STATE_NEGOTIATING; |
7593af92 | 2691 | |
6d4221b5 JS |
2692 | /* |
2693 | * Received banner is good, exchange connection info. | |
2694 | * Do not reset out_kvec, as sending our banner raced | |
2695 | * with receiving peer banner after connect completed. | |
2696 | */ | |
7593af92 AE |
2697 | ret = prepare_write_connect(con); |
2698 | if (ret < 0) | |
2699 | goto out; | |
2700 | prepare_read_connect(con); | |
2701 | ||
2702 | /* Send connection info before awaiting response */ | |
0da5d703 SW |
2703 | goto out; |
2704 | } | |
2705 | ||
8dacc7da | 2706 | if (con->state == CON_STATE_NEGOTIATING) { |
7593af92 | 2707 | dout("try_read negotiating\n"); |
31b8006e SW |
2708 | ret = read_partial_connect(con); |
2709 | if (ret <= 0) | |
31b8006e | 2710 | goto out; |
98bdb0aa SW |
2711 | ret = process_connect(con); |
2712 | if (ret < 0) | |
2713 | goto out; | |
31b8006e SW |
2714 | goto more; |
2715 | } | |
2716 | ||
122070a2 | 2717 | WARN_ON(con->state != CON_STATE_OPEN); |
8dacc7da | 2718 | |
31b8006e SW |
2719 | if (con->in_base_pos < 0) { |
2720 | /* | |
2721 | * skipping + discarding content. | |
31b8006e | 2722 | */ |
e5c93883 | 2723 | ret = ceph_tcp_recvmsg(con->sock, NULL, -con->in_base_pos); |
31b8006e | 2724 | if (ret <= 0) |
98bdb0aa | 2725 | goto out; |
e5c93883 | 2726 | dout("skipped %d / %d bytes\n", ret, -con->in_base_pos); |
31b8006e SW |
2727 | con->in_base_pos += ret; |
2728 | if (con->in_base_pos) | |
2729 | goto more; | |
2730 | } | |
2731 | if (con->in_tag == CEPH_MSGR_TAG_READY) { | |
2732 | /* | |
2733 | * what's next? | |
2734 | */ | |
2735 | ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1); | |
2736 | if (ret <= 0) | |
98bdb0aa | 2737 | goto out; |
31b8006e SW |
2738 | dout("try_read got tag %d\n", (int)con->in_tag); |
2739 | switch (con->in_tag) { | |
2740 | case CEPH_MSGR_TAG_MSG: | |
2741 | prepare_read_message(con); | |
2742 | break; | |
2743 | case CEPH_MSGR_TAG_ACK: | |
2744 | prepare_read_ack(con); | |
2745 | break; | |
8b9558aa YZ |
2746 | case CEPH_MSGR_TAG_KEEPALIVE2_ACK: |
2747 | prepare_read_keepalive_ack(con); | |
2748 | break; | |
31b8006e | 2749 | case CEPH_MSGR_TAG_CLOSE: |
8dacc7da SW |
2750 | con_close_socket(con); |
2751 | con->state = CON_STATE_CLOSED; | |
98bdb0aa | 2752 | goto out; |
31b8006e SW |
2753 | default: |
2754 | goto bad_tag; | |
2755 | } | |
2756 | } | |
2757 | if (con->in_tag == CEPH_MSGR_TAG_MSG) { | |
2758 | ret = read_partial_message(con); | |
2759 | if (ret <= 0) { | |
2760 | switch (ret) { | |
2761 | case -EBADMSG: | |
a51983e4 | 2762 | con->error_msg = "bad crc/signature"; |
67c64eb7 ID |
2763 | /* fall through */ |
2764 | case -EBADE: | |
31b8006e | 2765 | ret = -EIO; |
98bdb0aa | 2766 | break; |
31b8006e SW |
2767 | case -EIO: |
2768 | con->error_msg = "io error"; | |
98bdb0aa | 2769 | break; |
31b8006e | 2770 | } |
98bdb0aa | 2771 | goto out; |
31b8006e SW |
2772 | } |
2773 | if (con->in_tag == CEPH_MSGR_TAG_READY) | |
2774 | goto more; | |
2775 | process_message(con); | |
7b862e07 SW |
2776 | if (con->state == CON_STATE_OPEN) |
2777 | prepare_read_tag(con); | |
31b8006e SW |
2778 | goto more; |
2779 | } | |
3a23083b SW |
2780 | if (con->in_tag == CEPH_MSGR_TAG_ACK || |
2781 | con->in_tag == CEPH_MSGR_TAG_SEQ) { | |
2782 | /* | |
2783 | * the final handshake seq exchange is semantically | |
2784 | * equivalent to an ACK | |
2785 | */ | |
31b8006e SW |
2786 | ret = read_partial_ack(con); |
2787 | if (ret <= 0) | |
98bdb0aa | 2788 | goto out; |
31b8006e SW |
2789 | process_ack(con); |
2790 | goto more; | |
2791 | } | |
8b9558aa YZ |
2792 | if (con->in_tag == CEPH_MSGR_TAG_KEEPALIVE2_ACK) { |
2793 | ret = read_keepalive_ack(con); | |
2794 | if (ret <= 0) | |
2795 | goto out; | |
2796 | goto more; | |
2797 | } | |
31b8006e | 2798 | |
31b8006e | 2799 | out: |
98bdb0aa | 2800 | dout("try_read done on %p ret %d\n", con, ret); |
31b8006e SW |
2801 | return ret; |
2802 | ||
2803 | bad_tag: | |
2804 | pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag); | |
2805 | con->error_msg = "protocol error, garbage tag"; | |
2806 | ret = -1; | |
2807 | goto out; | |
2808 | } | |
2809 | ||
2810 | ||
2811 | /* | |
802c6d96 AE |
2812 | * Atomically queue work on a connection after the specified delay. |
2813 | * Bump @con reference to avoid races with connection teardown. | |
2814 | * Returns 0 if work was queued, or an error code otherwise. | |
31b8006e | 2815 | */ |
802c6d96 | 2816 | static int queue_con_delay(struct ceph_connection *con, unsigned long delay) |
31b8006e | 2817 | { |
31b8006e | 2818 | if (!con->ops->get(con)) { |
802c6d96 | 2819 | dout("%s %p ref count 0\n", __func__, con); |
802c6d96 | 2820 | return -ENOENT; |
31b8006e SW |
2821 | } |
2822 | ||
802c6d96 AE |
2823 | if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) { |
2824 | dout("%s %p - already queued\n", __func__, con); | |
31b8006e | 2825 | con->ops->put(con); |
802c6d96 | 2826 | return -EBUSY; |
31b8006e | 2827 | } |
802c6d96 AE |
2828 | |
2829 | dout("%s %p %lu\n", __func__, con, delay); | |
802c6d96 AE |
2830 | return 0; |
2831 | } | |
2832 | ||
2833 | static void queue_con(struct ceph_connection *con) | |
2834 | { | |
2835 | (void) queue_con_delay(con, 0); | |
31b8006e SW |
2836 | } |
2837 | ||
37ab77ac ID |
2838 | static void cancel_con(struct ceph_connection *con) |
2839 | { | |
2840 | if (cancel_delayed_work(&con->work)) { | |
2841 | dout("%s %p\n", __func__, con); | |
2842 | con->ops->put(con); | |
2843 | } | |
2844 | } | |
2845 | ||
7bb21d68 AE |
2846 | static bool con_sock_closed(struct ceph_connection *con) |
2847 | { | |
c9ffc77a | 2848 | if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED)) |
7bb21d68 AE |
2849 | return false; |
2850 | ||
2851 | #define CASE(x) \ | |
2852 | case CON_STATE_ ## x: \ | |
2853 | con->error_msg = "socket closed (con state " #x ")"; \ | |
2854 | break; | |
2855 | ||
2856 | switch (con->state) { | |
2857 | CASE(CLOSED); | |
2858 | CASE(PREOPEN); | |
2859 | CASE(CONNECTING); | |
2860 | CASE(NEGOTIATING); | |
2861 | CASE(OPEN); | |
2862 | CASE(STANDBY); | |
2863 | default: | |
b9a67899 | 2864 | pr_warn("%s con %p unrecognized state %lu\n", |
7bb21d68 AE |
2865 | __func__, con, con->state); |
2866 | con->error_msg = "unrecognized con state"; | |
2867 | BUG(); | |
2868 | break; | |
2869 | } | |
2870 | #undef CASE | |
2871 | ||
2872 | return true; | |
2873 | } | |
2874 | ||
f20a39fd AE |
2875 | static bool con_backoff(struct ceph_connection *con) |
2876 | { | |
2877 | int ret; | |
2878 | ||
2879 | if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF)) | |
2880 | return false; | |
2881 | ||
2882 | ret = queue_con_delay(con, round_jiffies_relative(con->delay)); | |
2883 | if (ret) { | |
2884 | dout("%s: con %p FAILED to back off %lu\n", __func__, | |
2885 | con, con->delay); | |
2886 | BUG_ON(ret == -ENOENT); | |
2887 | con_flag_set(con, CON_FLAG_BACKOFF); | |
2888 | } | |
2889 | ||
2890 | return true; | |
2891 | } | |
2892 | ||
93209264 AE |
2893 | /* Finish fault handling; con->mutex must *not* be held here */ |
2894 | ||
2895 | static void con_fault_finish(struct ceph_connection *con) | |
2896 | { | |
f6330cc1 ID |
2897 | dout("%s %p\n", __func__, con); |
2898 | ||
93209264 AE |
2899 | /* |
2900 | * in case we faulted due to authentication, invalidate our | |
2901 | * current tickets so that we can get new ones. | |
2902 | */ | |
f6330cc1 ID |
2903 | if (con->auth_retry) { |
2904 | dout("auth_retry %d, invalidating\n", con->auth_retry); | |
2905 | if (con->ops->invalidate_authorizer) | |
2906 | con->ops->invalidate_authorizer(con); | |
2907 | con->auth_retry = 0; | |
93209264 AE |
2908 | } |
2909 | ||
2910 | if (con->ops->fault) | |
2911 | con->ops->fault(con); | |
2912 | } | |
2913 | ||
31b8006e SW |
2914 | /* |
2915 | * Do some work on a connection. Drop a connection ref when we're done. | |
2916 | */ | |
68931622 | 2917 | static void ceph_con_workfn(struct work_struct *work) |
31b8006e SW |
2918 | { |
2919 | struct ceph_connection *con = container_of(work, struct ceph_connection, | |
2920 | work.work); | |
49659416 | 2921 | bool fault; |
31b8006e | 2922 | |
9dd4658d | 2923 | mutex_lock(&con->mutex); |
49659416 AE |
2924 | while (true) { |
2925 | int ret; | |
31b8006e | 2926 | |
49659416 AE |
2927 | if ((fault = con_sock_closed(con))) { |
2928 | dout("%s: con %p SOCK_CLOSED\n", __func__, con); | |
2929 | break; | |
2930 | } | |
2931 | if (con_backoff(con)) { | |
2932 | dout("%s: con %p BACKOFF\n", __func__, con); | |
2933 | break; | |
2934 | } | |
2935 | if (con->state == CON_STATE_STANDBY) { | |
2936 | dout("%s: con %p STANDBY\n", __func__, con); | |
2937 | break; | |
2938 | } | |
2939 | if (con->state == CON_STATE_CLOSED) { | |
2940 | dout("%s: con %p CLOSED\n", __func__, con); | |
2941 | BUG_ON(con->sock); | |
2942 | break; | |
2943 | } | |
2944 | if (con->state == CON_STATE_PREOPEN) { | |
2945 | dout("%s: con %p PREOPEN\n", __func__, con); | |
2946 | BUG_ON(con->sock); | |
2947 | } | |
0da5d703 | 2948 | |
49659416 AE |
2949 | ret = try_read(con); |
2950 | if (ret < 0) { | |
2951 | if (ret == -EAGAIN) | |
2952 | continue; | |
67c64eb7 ID |
2953 | if (!con->error_msg) |
2954 | con->error_msg = "socket error on read"; | |
49659416 AE |
2955 | fault = true; |
2956 | break; | |
2957 | } | |
2958 | ||
2959 | ret = try_write(con); | |
2960 | if (ret < 0) { | |
2961 | if (ret == -EAGAIN) | |
2962 | continue; | |
67c64eb7 ID |
2963 | if (!con->error_msg) |
2964 | con->error_msg = "socket error on write"; | |
49659416 AE |
2965 | fault = true; |
2966 | } | |
2967 | ||
2968 | break; /* If we make it to here, we're done */ | |
3a140a0d | 2969 | } |
b6e7b6a1 AE |
2970 | if (fault) |
2971 | con_fault(con); | |
9dd4658d | 2972 | mutex_unlock(&con->mutex); |
0da5d703 | 2973 | |
b6e7b6a1 AE |
2974 | if (fault) |
2975 | con_fault_finish(con); | |
2976 | ||
2977 | con->ops->put(con); | |
31b8006e SW |
2978 | } |
2979 | ||
31b8006e SW |
2980 | /* |
2981 | * Generic error/fault handler. A retry mechanism is used with | |
2982 | * exponential backoff | |
2983 | */ | |
93209264 | 2984 | static void con_fault(struct ceph_connection *con) |
31b8006e | 2985 | { |
31b8006e | 2986 | dout("fault %p state %lu to peer %s\n", |
b726ec97 | 2987 | con, con->state, ceph_pr_addr(&con->peer_addr)); |
31b8006e | 2988 | |
67c64eb7 | 2989 | pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name), |
b726ec97 | 2990 | ceph_pr_addr(&con->peer_addr), con->error_msg); |
67c64eb7 ID |
2991 | con->error_msg = NULL; |
2992 | ||
122070a2 | 2993 | WARN_ON(con->state != CON_STATE_CONNECTING && |
8dacc7da SW |
2994 | con->state != CON_STATE_NEGOTIATING && |
2995 | con->state != CON_STATE_OPEN); | |
ec302645 | 2996 | |
31b8006e | 2997 | con_close_socket(con); |
5e095e8b | 2998 | |
c9ffc77a | 2999 | if (con_flag_test(con, CON_FLAG_LOSSYTX)) { |
8dacc7da SW |
3000 | dout("fault on LOSSYTX channel, marking CLOSED\n"); |
3001 | con->state = CON_STATE_CLOSED; | |
93209264 | 3002 | return; |
3b5ede07 SW |
3003 | } |
3004 | ||
5e095e8b | 3005 | if (con->in_msg) { |
38941f80 | 3006 | BUG_ON(con->in_msg->con != con); |
5e095e8b SW |
3007 | ceph_msg_put(con->in_msg); |
3008 | con->in_msg = NULL; | |
3009 | } | |
31b8006e | 3010 | |
e80a52d1 SW |
3011 | /* Requeue anything that hasn't been acked */ |
3012 | list_splice_init(&con->out_sent, &con->out_queue); | |
9bd2e6f8 | 3013 | |
e76661d0 SW |
3014 | /* If there are no messages queued or keepalive pending, place |
3015 | * the connection in a STANDBY state */ | |
3016 | if (list_empty(&con->out_queue) && | |
c9ffc77a | 3017 | !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) { |
e00de341 | 3018 | dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con); |
c9ffc77a | 3019 | con_flag_clear(con, CON_FLAG_WRITE_PENDING); |
8dacc7da | 3020 | con->state = CON_STATE_STANDBY; |
e80a52d1 SW |
3021 | } else { |
3022 | /* retry after a delay. */ | |
8dacc7da | 3023 | con->state = CON_STATE_PREOPEN; |
e80a52d1 SW |
3024 | if (con->delay == 0) |
3025 | con->delay = BASE_DELAY_INTERVAL; | |
3026 | else if (con->delay < MAX_DELAY_INTERVAL) | |
3027 | con->delay *= 2; | |
c9ffc77a | 3028 | con_flag_set(con, CON_FLAG_BACKOFF); |
8618e30b | 3029 | queue_con(con); |
31b8006e | 3030 | } |
31b8006e SW |
3031 | } |
3032 | ||
3033 | ||
3034 | ||
3035 | /* | |
15d9882c | 3036 | * initialize a new messenger instance |
31b8006e | 3037 | */ |
15d9882c | 3038 | void ceph_messenger_init(struct ceph_messenger *msgr, |
859bff51 | 3039 | struct ceph_entity_addr *myaddr) |
31b8006e | 3040 | { |
31b8006e SW |
3041 | spin_lock_init(&msgr->global_seq_lock); |
3042 | ||
31b8006e SW |
3043 | if (myaddr) |
3044 | msgr->inst.addr = *myaddr; | |
3045 | ||
3046 | /* select a random nonce */ | |
ac8839d7 | 3047 | msgr->inst.addr.type = 0; |
103e2d3a | 3048 | get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce)); |
63f2d211 | 3049 | encode_my_addr(msgr); |
31b8006e | 3050 | |
a2a32584 | 3051 | atomic_set(&msgr->stopping, 0); |
757856d2 | 3052 | write_pnet(&msgr->net, get_net(current->nsproxy->net_ns)); |
31b8006e | 3053 | |
15d9882c | 3054 | dout("%s %p\n", __func__, msgr); |
31b8006e | 3055 | } |
15d9882c | 3056 | EXPORT_SYMBOL(ceph_messenger_init); |
31b8006e | 3057 | |
757856d2 ID |
3058 | void ceph_messenger_fini(struct ceph_messenger *msgr) |
3059 | { | |
3060 | put_net(read_pnet(&msgr->net)); | |
3061 | } | |
3062 | EXPORT_SYMBOL(ceph_messenger_fini); | |
3063 | ||
583d0fef ID |
3064 | static void msg_con_set(struct ceph_msg *msg, struct ceph_connection *con) |
3065 | { | |
3066 | if (msg->con) | |
3067 | msg->con->ops->put(msg->con); | |
3068 | ||
3069 | msg->con = con ? con->ops->get(con) : NULL; | |
3070 | BUG_ON(msg->con != con); | |
3071 | } | |
3072 | ||
e00de341 SW |
3073 | static void clear_standby(struct ceph_connection *con) |
3074 | { | |
3075 | /* come back from STANDBY? */ | |
8dacc7da | 3076 | if (con->state == CON_STATE_STANDBY) { |
e00de341 | 3077 | dout("clear_standby %p and ++connect_seq\n", con); |
8dacc7da | 3078 | con->state = CON_STATE_PREOPEN; |
e00de341 | 3079 | con->connect_seq++; |
c9ffc77a AE |
3080 | WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING)); |
3081 | WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)); | |
e00de341 SW |
3082 | } |
3083 | } | |
3084 | ||
31b8006e SW |
3085 | /* |
3086 | * Queue up an outgoing message on the given connection. | |
3087 | */ | |
3088 | void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg) | |
3089 | { | |
31b8006e | 3090 | /* set src+dst */ |
dbad185d | 3091 | msg->hdr.src = con->msgr->inst.name; |
3ca02ef9 | 3092 | BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len)); |
e84346b7 SW |
3093 | msg->needs_out_seq = true; |
3094 | ||
ec302645 | 3095 | mutex_lock(&con->mutex); |
92ce034b | 3096 | |
8dacc7da | 3097 | if (con->state == CON_STATE_CLOSED) { |
a59b55a6 SW |
3098 | dout("con_send %p closed, dropping %p\n", con, msg); |
3099 | ceph_msg_put(msg); | |
3100 | mutex_unlock(&con->mutex); | |
3101 | return; | |
3102 | } | |
3103 | ||
583d0fef | 3104 | msg_con_set(msg, con); |
92ce034b | 3105 | |
31b8006e SW |
3106 | BUG_ON(!list_empty(&msg->list_head)); |
3107 | list_add_tail(&msg->list_head, &con->out_queue); | |
3108 | dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg, | |
3109 | ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type), | |
3110 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
3111 | le32_to_cpu(msg->hdr.front_len), | |
3112 | le32_to_cpu(msg->hdr.middle_len), | |
3113 | le32_to_cpu(msg->hdr.data_len)); | |
00650931 SW |
3114 | |
3115 | clear_standby(con); | |
ec302645 | 3116 | mutex_unlock(&con->mutex); |
31b8006e SW |
3117 | |
3118 | /* if there wasn't anything waiting to send before, queue | |
3119 | * new work */ | |
c9ffc77a | 3120 | if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0) |
31b8006e SW |
3121 | queue_con(con); |
3122 | } | |
3d14c5d2 | 3123 | EXPORT_SYMBOL(ceph_con_send); |
31b8006e SW |
3124 | |
3125 | /* | |
3126 | * Revoke a message that was previously queued for send | |
3127 | */ | |
6740a845 | 3128 | void ceph_msg_revoke(struct ceph_msg *msg) |
31b8006e | 3129 | { |
6740a845 AE |
3130 | struct ceph_connection *con = msg->con; |
3131 | ||
583d0fef ID |
3132 | if (!con) { |
3133 | dout("%s msg %p null con\n", __func__, msg); | |
6740a845 | 3134 | return; /* Message not in our possession */ |
583d0fef | 3135 | } |
6740a845 | 3136 | |
ec302645 | 3137 | mutex_lock(&con->mutex); |
31b8006e | 3138 | if (!list_empty(&msg->list_head)) { |
38941f80 | 3139 | dout("%s %p msg %p - was on queue\n", __func__, con, msg); |
31b8006e | 3140 | list_del_init(&msg->list_head); |
31b8006e | 3141 | msg->hdr.seq = 0; |
38941f80 | 3142 | |
31b8006e | 3143 | ceph_msg_put(msg); |
ed98adad SW |
3144 | } |
3145 | if (con->out_msg == msg) { | |
67645d76 ID |
3146 | BUG_ON(con->out_skip); |
3147 | /* footer */ | |
3148 | if (con->out_msg_done) { | |
3149 | con->out_skip += con_out_kvec_skip(con); | |
3150 | } else { | |
3151 | BUG_ON(!msg->data_length); | |
89f08173 | 3152 | con->out_skip += sizeof_footer(con); |
31b8006e | 3153 | } |
67645d76 ID |
3154 | /* data, middle, front */ |
3155 | if (msg->data_length) | |
3156 | con->out_skip += msg->cursor.total_resid; | |
3157 | if (msg->middle) | |
3158 | con->out_skip += con_out_kvec_skip(con); | |
3159 | con->out_skip += con_out_kvec_skip(con); | |
3160 | ||
3161 | dout("%s %p msg %p - was sending, will write %d skip %d\n", | |
3162 | __func__, con, msg, con->out_kvec_bytes, con->out_skip); | |
ed98adad | 3163 | msg->hdr.seq = 0; |
67645d76 | 3164 | con->out_msg = NULL; |
92ce034b | 3165 | ceph_msg_put(msg); |
31b8006e | 3166 | } |
67645d76 | 3167 | |
ec302645 | 3168 | mutex_unlock(&con->mutex); |
31b8006e SW |
3169 | } |
3170 | ||
350b1c32 | 3171 | /* |
0d59ab81 | 3172 | * Revoke a message that we may be reading data into |
350b1c32 | 3173 | */ |
8921d114 | 3174 | void ceph_msg_revoke_incoming(struct ceph_msg *msg) |
350b1c32 | 3175 | { |
583d0fef | 3176 | struct ceph_connection *con = msg->con; |
8921d114 | 3177 | |
583d0fef | 3178 | if (!con) { |
8921d114 | 3179 | dout("%s msg %p null con\n", __func__, msg); |
8921d114 AE |
3180 | return; /* Message not in our possession */ |
3181 | } | |
3182 | ||
350b1c32 | 3183 | mutex_lock(&con->mutex); |
8921d114 | 3184 | if (con->in_msg == msg) { |
95c96174 ED |
3185 | unsigned int front_len = le32_to_cpu(con->in_hdr.front_len); |
3186 | unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
3187 | unsigned int data_len = le32_to_cpu(con->in_hdr.data_len); | |
350b1c32 SW |
3188 | |
3189 | /* skip rest of message */ | |
8921d114 AE |
3190 | dout("%s %p msg %p revoked\n", __func__, con, msg); |
3191 | con->in_base_pos = con->in_base_pos - | |
350b1c32 | 3192 | sizeof(struct ceph_msg_header) - |
0d59ab81 YS |
3193 | front_len - |
3194 | middle_len - | |
3195 | data_len - | |
350b1c32 | 3196 | sizeof(struct ceph_msg_footer); |
350b1c32 SW |
3197 | ceph_msg_put(con->in_msg); |
3198 | con->in_msg = NULL; | |
3199 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 3200 | con->in_seq++; |
350b1c32 | 3201 | } else { |
8921d114 AE |
3202 | dout("%s %p in_msg %p msg %p no-op\n", |
3203 | __func__, con, con->in_msg, msg); | |
350b1c32 SW |
3204 | } |
3205 | mutex_unlock(&con->mutex); | |
3206 | } | |
3207 | ||
31b8006e SW |
3208 | /* |
3209 | * Queue a keepalive byte to ensure the tcp connection is alive. | |
3210 | */ | |
3211 | void ceph_con_keepalive(struct ceph_connection *con) | |
3212 | { | |
e00de341 | 3213 | dout("con_keepalive %p\n", con); |
00650931 | 3214 | mutex_lock(&con->mutex); |
e00de341 | 3215 | clear_standby(con); |
4aac9228 | 3216 | con_flag_set(con, CON_FLAG_KEEPALIVE_PENDING); |
00650931 | 3217 | mutex_unlock(&con->mutex); |
4aac9228 ID |
3218 | |
3219 | if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0) | |
31b8006e SW |
3220 | queue_con(con); |
3221 | } | |
3d14c5d2 | 3222 | EXPORT_SYMBOL(ceph_con_keepalive); |
31b8006e | 3223 | |
8b9558aa YZ |
3224 | bool ceph_con_keepalive_expired(struct ceph_connection *con, |
3225 | unsigned long interval) | |
3226 | { | |
3227 | if (interval > 0 && | |
3228 | (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) { | |
473bd2d7 AB |
3229 | struct timespec64 now; |
3230 | struct timespec64 ts; | |
3231 | ktime_get_real_ts64(&now); | |
3232 | jiffies_to_timespec64(interval, &ts); | |
3233 | ts = timespec64_add(con->last_keepalive_ack, ts); | |
3234 | return timespec64_compare(&now, &ts) >= 0; | |
8b9558aa YZ |
3235 | } |
3236 | return false; | |
3237 | } | |
3238 | ||
0d9c1ab3 | 3239 | static struct ceph_msg_data *ceph_msg_data_add(struct ceph_msg *msg) |
43794509 | 3240 | { |
0d9c1ab3 ID |
3241 | BUG_ON(msg->num_data_items >= msg->max_data_items); |
3242 | return &msg->data[msg->num_data_items++]; | |
6644ed7b AE |
3243 | } |
3244 | ||
3245 | static void ceph_msg_data_destroy(struct ceph_msg_data *data) | |
3246 | { | |
e4339d28 | 3247 | if (data->type == CEPH_MSG_DATA_PAGELIST) |
6644ed7b | 3248 | ceph_pagelist_release(data->pagelist); |
43794509 AE |
3249 | } |
3250 | ||
90af3602 | 3251 | void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages, |
f1baeb2b | 3252 | size_t length, size_t alignment) |
02afca6c | 3253 | { |
6644ed7b AE |
3254 | struct ceph_msg_data *data; |
3255 | ||
07aa1558 AE |
3256 | BUG_ON(!pages); |
3257 | BUG_ON(!length); | |
6644ed7b | 3258 | |
0d9c1ab3 ID |
3259 | data = ceph_msg_data_add(msg); |
3260 | data->type = CEPH_MSG_DATA_PAGES; | |
6644ed7b AE |
3261 | data->pages = pages; |
3262 | data->length = length; | |
3263 | data->alignment = alignment & ~PAGE_MASK; | |
02afca6c | 3264 | |
5240d9f9 | 3265 | msg->data_length += length; |
02afca6c | 3266 | } |
90af3602 | 3267 | EXPORT_SYMBOL(ceph_msg_data_add_pages); |
31b8006e | 3268 | |
90af3602 | 3269 | void ceph_msg_data_add_pagelist(struct ceph_msg *msg, |
27fa8385 AE |
3270 | struct ceph_pagelist *pagelist) |
3271 | { | |
6644ed7b AE |
3272 | struct ceph_msg_data *data; |
3273 | ||
07aa1558 AE |
3274 | BUG_ON(!pagelist); |
3275 | BUG_ON(!pagelist->length); | |
27fa8385 | 3276 | |
0d9c1ab3 ID |
3277 | data = ceph_msg_data_add(msg); |
3278 | data->type = CEPH_MSG_DATA_PAGELIST; | |
89486833 | 3279 | refcount_inc(&pagelist->refcnt); |
6644ed7b AE |
3280 | data->pagelist = pagelist; |
3281 | ||
5240d9f9 | 3282 | msg->data_length += pagelist->length; |
27fa8385 | 3283 | } |
90af3602 | 3284 | EXPORT_SYMBOL(ceph_msg_data_add_pagelist); |
27fa8385 | 3285 | |
ea96571f | 3286 | #ifdef CONFIG_BLOCK |
5359a17d ID |
3287 | void ceph_msg_data_add_bio(struct ceph_msg *msg, struct ceph_bio_iter *bio_pos, |
3288 | u32 length) | |
27fa8385 | 3289 | { |
6644ed7b AE |
3290 | struct ceph_msg_data *data; |
3291 | ||
0d9c1ab3 ID |
3292 | data = ceph_msg_data_add(msg); |
3293 | data->type = CEPH_MSG_DATA_BIO; | |
5359a17d | 3294 | data->bio_pos = *bio_pos; |
c851c495 | 3295 | data->bio_length = length; |
6644ed7b | 3296 | |
5240d9f9 | 3297 | msg->data_length += length; |
27fa8385 | 3298 | } |
90af3602 | 3299 | EXPORT_SYMBOL(ceph_msg_data_add_bio); |
ea96571f | 3300 | #endif /* CONFIG_BLOCK */ |
27fa8385 | 3301 | |
b9e281c2 ID |
3302 | void ceph_msg_data_add_bvecs(struct ceph_msg *msg, |
3303 | struct ceph_bvec_iter *bvec_pos) | |
3304 | { | |
3305 | struct ceph_msg_data *data; | |
3306 | ||
0d9c1ab3 ID |
3307 | data = ceph_msg_data_add(msg); |
3308 | data->type = CEPH_MSG_DATA_BVECS; | |
b9e281c2 ID |
3309 | data->bvec_pos = *bvec_pos; |
3310 | ||
b9e281c2 ID |
3311 | msg->data_length += bvec_pos->iter.bi_size; |
3312 | } | |
3313 | EXPORT_SYMBOL(ceph_msg_data_add_bvecs); | |
3314 | ||
31b8006e SW |
3315 | /* |
3316 | * construct a new message with given type, size | |
3317 | * the new msg has a ref count of 1. | |
3318 | */ | |
0d9c1ab3 ID |
3319 | struct ceph_msg *ceph_msg_new2(int type, int front_len, int max_data_items, |
3320 | gfp_t flags, bool can_fail) | |
31b8006e SW |
3321 | { |
3322 | struct ceph_msg *m; | |
3323 | ||
e3d5d638 | 3324 | m = kmem_cache_zalloc(ceph_msg_cache, flags); |
31b8006e SW |
3325 | if (m == NULL) |
3326 | goto out; | |
31b8006e SW |
3327 | |
3328 | m->hdr.type = cpu_to_le16(type); | |
45c6ceb5 | 3329 | m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT); |
31b8006e | 3330 | m->hdr.front_len = cpu_to_le32(front_len); |
ca20892d | 3331 | |
9516e45b AE |
3332 | INIT_LIST_HEAD(&m->list_head); |
3333 | kref_init(&m->kref); | |
ca20892d | 3334 | |
31b8006e SW |
3335 | /* front */ |
3336 | if (front_len) { | |
eeb0bed5 | 3337 | m->front.iov_base = ceph_kvmalloc(front_len, flags); |
31b8006e | 3338 | if (m->front.iov_base == NULL) { |
b61c2763 | 3339 | dout("ceph_msg_new can't allocate %d bytes\n", |
31b8006e SW |
3340 | front_len); |
3341 | goto out2; | |
3342 | } | |
3343 | } else { | |
3344 | m->front.iov_base = NULL; | |
3345 | } | |
f2be82b0 | 3346 | m->front_alloc_len = m->front.iov_len = front_len; |
31b8006e | 3347 | |
0d9c1ab3 ID |
3348 | if (max_data_items) { |
3349 | m->data = kmalloc_array(max_data_items, sizeof(*m->data), | |
3350 | flags); | |
3351 | if (!m->data) | |
3352 | goto out2; | |
3353 | ||
3354 | m->max_data_items = max_data_items; | |
3355 | } | |
3356 | ||
bb257664 | 3357 | dout("ceph_msg_new %p front %d\n", m, front_len); |
31b8006e SW |
3358 | return m; |
3359 | ||
3360 | out2: | |
3361 | ceph_msg_put(m); | |
3362 | out: | |
b61c2763 SW |
3363 | if (!can_fail) { |
3364 | pr_err("msg_new can't create type %d front %d\n", type, | |
3365 | front_len); | |
f0ed1b7c | 3366 | WARN_ON(1); |
b61c2763 SW |
3367 | } else { |
3368 | dout("msg_new can't create type %d front %d\n", type, | |
3369 | front_len); | |
3370 | } | |
a79832f2 | 3371 | return NULL; |
31b8006e | 3372 | } |
0d9c1ab3 ID |
3373 | EXPORT_SYMBOL(ceph_msg_new2); |
3374 | ||
3375 | struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags, | |
3376 | bool can_fail) | |
3377 | { | |
3378 | return ceph_msg_new2(type, front_len, 0, flags, can_fail); | |
3379 | } | |
3d14c5d2 | 3380 | EXPORT_SYMBOL(ceph_msg_new); |
31b8006e | 3381 | |
31b8006e SW |
3382 | /* |
3383 | * Allocate "middle" portion of a message, if it is needed and wasn't | |
3384 | * allocated by alloc_msg. This allows us to read a small fixed-size | |
3385 | * per-type header in the front and then gracefully fail (i.e., | |
3386 | * propagate the error to the caller based on info in the front) when | |
3387 | * the middle is too large. | |
3388 | */ | |
2450418c | 3389 | static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg) |
31b8006e SW |
3390 | { |
3391 | int type = le16_to_cpu(msg->hdr.type); | |
3392 | int middle_len = le32_to_cpu(msg->hdr.middle_len); | |
3393 | ||
3394 | dout("alloc_middle %p type %d %s middle_len %d\n", msg, type, | |
3395 | ceph_msg_type_name(type), middle_len); | |
3396 | BUG_ON(!middle_len); | |
3397 | BUG_ON(msg->middle); | |
3398 | ||
b6c1d5b8 | 3399 | msg->middle = ceph_buffer_new(middle_len, GFP_NOFS); |
31b8006e SW |
3400 | if (!msg->middle) |
3401 | return -ENOMEM; | |
3402 | return 0; | |
3403 | } | |
3404 | ||
2450418c | 3405 | /* |
1c20f2d2 AE |
3406 | * Allocate a message for receiving an incoming message on a |
3407 | * connection, and save the result in con->in_msg. Uses the | |
3408 | * connection's private alloc_msg op if available. | |
3409 | * | |
4740a623 SW |
3410 | * Returns 0 on success, or a negative error code. |
3411 | * | |
3412 | * On success, if we set *skip = 1: | |
3413 | * - the next message should be skipped and ignored. | |
3414 | * - con->in_msg == NULL | |
3415 | * or if we set *skip = 0: | |
3416 | * - con->in_msg is non-null. | |
3417 | * On error (ENOMEM, EAGAIN, ...), | |
3418 | * - con->in_msg == NULL | |
2450418c | 3419 | */ |
4740a623 | 3420 | static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip) |
2450418c | 3421 | { |
4740a623 | 3422 | struct ceph_msg_header *hdr = &con->in_hdr; |
2450418c | 3423 | int middle_len = le32_to_cpu(hdr->middle_len); |
1d866d1c | 3424 | struct ceph_msg *msg; |
4740a623 | 3425 | int ret = 0; |
2450418c | 3426 | |
1c20f2d2 | 3427 | BUG_ON(con->in_msg != NULL); |
53ded495 | 3428 | BUG_ON(!con->ops->alloc_msg); |
2450418c | 3429 | |
53ded495 AE |
3430 | mutex_unlock(&con->mutex); |
3431 | msg = con->ops->alloc_msg(con, hdr, skip); | |
3432 | mutex_lock(&con->mutex); | |
3433 | if (con->state != CON_STATE_OPEN) { | |
3434 | if (msg) | |
1d866d1c | 3435 | ceph_msg_put(msg); |
53ded495 AE |
3436 | return -EAGAIN; |
3437 | } | |
4137577a AE |
3438 | if (msg) { |
3439 | BUG_ON(*skip); | |
583d0fef | 3440 | msg_con_set(msg, con); |
4137577a | 3441 | con->in_msg = msg; |
4137577a AE |
3442 | } else { |
3443 | /* | |
3444 | * Null message pointer means either we should skip | |
3445 | * this message or we couldn't allocate memory. The | |
3446 | * former is not an error. | |
3447 | */ | |
3448 | if (*skip) | |
3449 | return 0; | |
4137577a | 3450 | |
67c64eb7 | 3451 | con->error_msg = "error allocating memory for incoming message"; |
53ded495 | 3452 | return -ENOMEM; |
2450418c | 3453 | } |
1c20f2d2 | 3454 | memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr)); |
2450418c | 3455 | |
1c20f2d2 AE |
3456 | if (middle_len && !con->in_msg->middle) { |
3457 | ret = ceph_alloc_middle(con, con->in_msg); | |
2450418c | 3458 | if (ret < 0) { |
1c20f2d2 AE |
3459 | ceph_msg_put(con->in_msg); |
3460 | con->in_msg = NULL; | |
2450418c YS |
3461 | } |
3462 | } | |
9d7f0f13 | 3463 | |
4740a623 | 3464 | return ret; |
2450418c YS |
3465 | } |
3466 | ||
31b8006e SW |
3467 | |
3468 | /* | |
3469 | * Free a generically kmalloc'd message. | |
3470 | */ | |
0215e44b | 3471 | static void ceph_msg_free(struct ceph_msg *m) |
31b8006e | 3472 | { |
0215e44b | 3473 | dout("%s %p\n", __func__, m); |
4965fc38 | 3474 | kvfree(m->front.iov_base); |
0d9c1ab3 | 3475 | kfree(m->data); |
e3d5d638 | 3476 | kmem_cache_free(ceph_msg_cache, m); |
31b8006e SW |
3477 | } |
3478 | ||
0215e44b | 3479 | static void ceph_msg_release(struct kref *kref) |
c2e552e7 SW |
3480 | { |
3481 | struct ceph_msg *m = container_of(kref, struct ceph_msg, kref); | |
0d9c1ab3 | 3482 | int i; |
31b8006e | 3483 | |
0215e44b | 3484 | dout("%s %p\n", __func__, m); |
c2e552e7 SW |
3485 | WARN_ON(!list_empty(&m->list_head)); |
3486 | ||
583d0fef ID |
3487 | msg_con_set(m, NULL); |
3488 | ||
c2e552e7 SW |
3489 | /* drop middle, data, if any */ |
3490 | if (m->middle) { | |
3491 | ceph_buffer_put(m->middle); | |
3492 | m->middle = NULL; | |
31b8006e | 3493 | } |
5240d9f9 | 3494 | |
0d9c1ab3 ID |
3495 | for (i = 0; i < m->num_data_items; i++) |
3496 | ceph_msg_data_destroy(&m->data[i]); | |
58bb3b37 | 3497 | |
c2e552e7 SW |
3498 | if (m->pool) |
3499 | ceph_msgpool_put(m->pool, m); | |
3500 | else | |
0215e44b ID |
3501 | ceph_msg_free(m); |
3502 | } | |
3503 | ||
3504 | struct ceph_msg *ceph_msg_get(struct ceph_msg *msg) | |
3505 | { | |
3506 | dout("%s %p (was %d)\n", __func__, msg, | |
2c935bc5 | 3507 | kref_read(&msg->kref)); |
0215e44b ID |
3508 | kref_get(&msg->kref); |
3509 | return msg; | |
3510 | } | |
3511 | EXPORT_SYMBOL(ceph_msg_get); | |
3512 | ||
3513 | void ceph_msg_put(struct ceph_msg *msg) | |
3514 | { | |
3515 | dout("%s %p (was %d)\n", __func__, msg, | |
2c935bc5 | 3516 | kref_read(&msg->kref)); |
0215e44b | 3517 | kref_put(&msg->kref, ceph_msg_release); |
31b8006e | 3518 | } |
0215e44b | 3519 | EXPORT_SYMBOL(ceph_msg_put); |
9ec7cab1 SW |
3520 | |
3521 | void ceph_msg_dump(struct ceph_msg *msg) | |
3522 | { | |
3cea4c30 ID |
3523 | pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg, |
3524 | msg->front_alloc_len, msg->data_length); | |
9ec7cab1 SW |
3525 | print_hex_dump(KERN_DEBUG, "header: ", |
3526 | DUMP_PREFIX_OFFSET, 16, 1, | |
3527 | &msg->hdr, sizeof(msg->hdr), true); | |
3528 | print_hex_dump(KERN_DEBUG, " front: ", | |
3529 | DUMP_PREFIX_OFFSET, 16, 1, | |
3530 | msg->front.iov_base, msg->front.iov_len, true); | |
3531 | if (msg->middle) | |
3532 | print_hex_dump(KERN_DEBUG, "middle: ", | |
3533 | DUMP_PREFIX_OFFSET, 16, 1, | |
3534 | msg->middle->vec.iov_base, | |
3535 | msg->middle->vec.iov_len, true); | |
3536 | print_hex_dump(KERN_DEBUG, "footer: ", | |
3537 | DUMP_PREFIX_OFFSET, 16, 1, | |
3538 | &msg->footer, sizeof(msg->footer), true); | |
3539 | } | |
3d14c5d2 | 3540 | EXPORT_SYMBOL(ceph_msg_dump); |