| 1 | #include <linux/ceph/ceph_debug.h> |
| 2 | |
| 3 | #include <linux/crc32c.h> |
| 4 | #include <linux/ctype.h> |
| 5 | #include <linux/highmem.h> |
| 6 | #include <linux/inet.h> |
| 7 | #include <linux/kthread.h> |
| 8 | #include <linux/net.h> |
| 9 | #include <linux/nsproxy.h> |
| 10 | #include <linux/slab.h> |
| 11 | #include <linux/socket.h> |
| 12 | #include <linux/string.h> |
| 13 | #ifdef CONFIG_BLOCK |
| 14 | #include <linux/bio.h> |
| 15 | #endif /* CONFIG_BLOCK */ |
| 16 | #include <linux/dns_resolver.h> |
| 17 | #include <net/tcp.h> |
| 18 | |
| 19 | #include <linux/ceph/ceph_features.h> |
| 20 | #include <linux/ceph/libceph.h> |
| 21 | #include <linux/ceph/messenger.h> |
| 22 | #include <linux/ceph/decode.h> |
| 23 | #include <linux/ceph/pagelist.h> |
| 24 | #include <linux/export.h> |
| 25 | |
| 26 | #define list_entry_next(pos, member) \ |
| 27 | list_entry(pos->member.next, typeof(*pos), member) |
| 28 | |
| 29 | /* |
| 30 | * Ceph uses the messenger to exchange ceph_msg messages with other |
| 31 | * hosts in the system. The messenger provides ordered and reliable |
| 32 | * delivery. We tolerate TCP disconnects by reconnecting (with |
| 33 | * exponential backoff) in the case of a fault (disconnection, bad |
| 34 | * crc, protocol error). Acks allow sent messages to be discarded by |
| 35 | * the sender. |
| 36 | */ |
| 37 | |
| 38 | /* |
| 39 | * We track the state of the socket on a given connection using |
| 40 | * values defined below. The transition to a new socket state is |
| 41 | * handled by a function which verifies we aren't coming from an |
| 42 | * unexpected state. |
| 43 | * |
| 44 | * -------- |
| 45 | * | NEW* | transient initial state |
| 46 | * -------- |
| 47 | * | con_sock_state_init() |
| 48 | * v |
| 49 | * ---------- |
| 50 | * | CLOSED | initialized, but no socket (and no |
| 51 | * ---------- TCP connection) |
| 52 | * ^ \ |
| 53 | * | \ con_sock_state_connecting() |
| 54 | * | ---------------------- |
| 55 | * | \ |
| 56 | * + con_sock_state_closed() \ |
| 57 | * |+--------------------------- \ |
| 58 | * | \ \ \ |
| 59 | * | ----------- \ \ |
| 60 | * | | CLOSING | socket event; \ \ |
| 61 | * | ----------- await close \ \ |
| 62 | * | ^ \ | |
| 63 | * | | \ | |
| 64 | * | + con_sock_state_closing() \ | |
| 65 | * | / \ | | |
| 66 | * | / --------------- | | |
| 67 | * | / \ v v |
| 68 | * | / -------------- |
| 69 | * | / -----------------| CONNECTING | socket created, TCP |
| 70 | * | | / -------------- connect initiated |
| 71 | * | | | con_sock_state_connected() |
| 72 | * | | v |
| 73 | * ------------- |
| 74 | * | CONNECTED | TCP connection established |
| 75 | * ------------- |
| 76 | * |
| 77 | * State values for ceph_connection->sock_state; NEW is assumed to be 0. |
| 78 | */ |
| 79 | |
| 80 | #define CON_SOCK_STATE_NEW 0 /* -> CLOSED */ |
| 81 | #define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */ |
| 82 | #define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */ |
| 83 | #define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */ |
| 84 | #define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */ |
| 85 | |
| 86 | /* |
| 87 | * connection states |
| 88 | */ |
| 89 | #define CON_STATE_CLOSED 1 /* -> PREOPEN */ |
| 90 | #define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */ |
| 91 | #define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */ |
| 92 | #define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */ |
| 93 | #define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */ |
| 94 | #define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */ |
| 95 | |
| 96 | /* |
| 97 | * ceph_connection flag bits |
| 98 | */ |
| 99 | #define CON_FLAG_LOSSYTX 0 /* we can close channel or drop |
| 100 | * messages on errors */ |
| 101 | #define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */ |
| 102 | #define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */ |
| 103 | #define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */ |
| 104 | #define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */ |
| 105 | |
| 106 | static bool con_flag_valid(unsigned long con_flag) |
| 107 | { |
| 108 | switch (con_flag) { |
| 109 | case CON_FLAG_LOSSYTX: |
| 110 | case CON_FLAG_KEEPALIVE_PENDING: |
| 111 | case CON_FLAG_WRITE_PENDING: |
| 112 | case CON_FLAG_SOCK_CLOSED: |
| 113 | case CON_FLAG_BACKOFF: |
| 114 | return true; |
| 115 | default: |
| 116 | return false; |
| 117 | } |
| 118 | } |
| 119 | |
| 120 | static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag) |
| 121 | { |
| 122 | BUG_ON(!con_flag_valid(con_flag)); |
| 123 | |
| 124 | clear_bit(con_flag, &con->flags); |
| 125 | } |
| 126 | |
| 127 | static void con_flag_set(struct ceph_connection *con, unsigned long con_flag) |
| 128 | { |
| 129 | BUG_ON(!con_flag_valid(con_flag)); |
| 130 | |
| 131 | set_bit(con_flag, &con->flags); |
| 132 | } |
| 133 | |
| 134 | static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag) |
| 135 | { |
| 136 | BUG_ON(!con_flag_valid(con_flag)); |
| 137 | |
| 138 | return test_bit(con_flag, &con->flags); |
| 139 | } |
| 140 | |
| 141 | static bool con_flag_test_and_clear(struct ceph_connection *con, |
| 142 | unsigned long con_flag) |
| 143 | { |
| 144 | BUG_ON(!con_flag_valid(con_flag)); |
| 145 | |
| 146 | return test_and_clear_bit(con_flag, &con->flags); |
| 147 | } |
| 148 | |
| 149 | static bool con_flag_test_and_set(struct ceph_connection *con, |
| 150 | unsigned long con_flag) |
| 151 | { |
| 152 | BUG_ON(!con_flag_valid(con_flag)); |
| 153 | |
| 154 | return test_and_set_bit(con_flag, &con->flags); |
| 155 | } |
| 156 | |
| 157 | /* Slab caches for frequently-allocated structures */ |
| 158 | |
| 159 | static struct kmem_cache *ceph_msg_cache; |
| 160 | static struct kmem_cache *ceph_msg_data_cache; |
| 161 | |
| 162 | /* static tag bytes (protocol control messages) */ |
| 163 | static char tag_msg = CEPH_MSGR_TAG_MSG; |
| 164 | static char tag_ack = CEPH_MSGR_TAG_ACK; |
| 165 | static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE; |
| 166 | static char tag_keepalive2 = CEPH_MSGR_TAG_KEEPALIVE2; |
| 167 | |
| 168 | #ifdef CONFIG_LOCKDEP |
| 169 | static struct lock_class_key socket_class; |
| 170 | #endif |
| 171 | |
| 172 | /* |
| 173 | * When skipping (ignoring) a block of input we read it into a "skip |
| 174 | * buffer," which is this many bytes in size. |
| 175 | */ |
| 176 | #define SKIP_BUF_SIZE 1024 |
| 177 | |
| 178 | static void queue_con(struct ceph_connection *con); |
| 179 | static void cancel_con(struct ceph_connection *con); |
| 180 | static void ceph_con_workfn(struct work_struct *); |
| 181 | static void con_fault(struct ceph_connection *con); |
| 182 | |
| 183 | /* |
| 184 | * Nicely render a sockaddr as a string. An array of formatted |
| 185 | * strings is used, to approximate reentrancy. |
| 186 | */ |
| 187 | #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */ |
| 188 | #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG) |
| 189 | #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1) |
| 190 | #define MAX_ADDR_STR_LEN 64 /* 54 is enough */ |
| 191 | |
| 192 | static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN]; |
| 193 | static atomic_t addr_str_seq = ATOMIC_INIT(0); |
| 194 | |
| 195 | static struct page *zero_page; /* used in certain error cases */ |
| 196 | |
| 197 | const char *ceph_pr_addr(const struct sockaddr_storage *ss) |
| 198 | { |
| 199 | int i; |
| 200 | char *s; |
| 201 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
| 202 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; |
| 203 | |
| 204 | i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK; |
| 205 | s = addr_str[i]; |
| 206 | |
| 207 | switch (ss->ss_family) { |
| 208 | case AF_INET: |
| 209 | snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr, |
| 210 | ntohs(in4->sin_port)); |
| 211 | break; |
| 212 | |
| 213 | case AF_INET6: |
| 214 | snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr, |
| 215 | ntohs(in6->sin6_port)); |
| 216 | break; |
| 217 | |
| 218 | default: |
| 219 | snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)", |
| 220 | ss->ss_family); |
| 221 | } |
| 222 | |
| 223 | return s; |
| 224 | } |
| 225 | EXPORT_SYMBOL(ceph_pr_addr); |
| 226 | |
| 227 | static void encode_my_addr(struct ceph_messenger *msgr) |
| 228 | { |
| 229 | memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr)); |
| 230 | ceph_encode_addr(&msgr->my_enc_addr); |
| 231 | } |
| 232 | |
| 233 | /* |
| 234 | * work queue for all reading and writing to/from the socket. |
| 235 | */ |
| 236 | static struct workqueue_struct *ceph_msgr_wq; |
| 237 | |
| 238 | static int ceph_msgr_slab_init(void) |
| 239 | { |
| 240 | BUG_ON(ceph_msg_cache); |
| 241 | ceph_msg_cache = kmem_cache_create("ceph_msg", |
| 242 | sizeof (struct ceph_msg), |
| 243 | __alignof__(struct ceph_msg), 0, NULL); |
| 244 | |
| 245 | if (!ceph_msg_cache) |
| 246 | return -ENOMEM; |
| 247 | |
| 248 | BUG_ON(ceph_msg_data_cache); |
| 249 | ceph_msg_data_cache = kmem_cache_create("ceph_msg_data", |
| 250 | sizeof (struct ceph_msg_data), |
| 251 | __alignof__(struct ceph_msg_data), |
| 252 | 0, NULL); |
| 253 | if (ceph_msg_data_cache) |
| 254 | return 0; |
| 255 | |
| 256 | kmem_cache_destroy(ceph_msg_cache); |
| 257 | ceph_msg_cache = NULL; |
| 258 | |
| 259 | return -ENOMEM; |
| 260 | } |
| 261 | |
| 262 | static void ceph_msgr_slab_exit(void) |
| 263 | { |
| 264 | BUG_ON(!ceph_msg_data_cache); |
| 265 | kmem_cache_destroy(ceph_msg_data_cache); |
| 266 | ceph_msg_data_cache = NULL; |
| 267 | |
| 268 | BUG_ON(!ceph_msg_cache); |
| 269 | kmem_cache_destroy(ceph_msg_cache); |
| 270 | ceph_msg_cache = NULL; |
| 271 | } |
| 272 | |
| 273 | static void _ceph_msgr_exit(void) |
| 274 | { |
| 275 | if (ceph_msgr_wq) { |
| 276 | destroy_workqueue(ceph_msgr_wq); |
| 277 | ceph_msgr_wq = NULL; |
| 278 | } |
| 279 | |
| 280 | BUG_ON(zero_page == NULL); |
| 281 | page_cache_release(zero_page); |
| 282 | zero_page = NULL; |
| 283 | |
| 284 | ceph_msgr_slab_exit(); |
| 285 | } |
| 286 | |
| 287 | int ceph_msgr_init(void) |
| 288 | { |
| 289 | if (ceph_msgr_slab_init()) |
| 290 | return -ENOMEM; |
| 291 | |
| 292 | BUG_ON(zero_page != NULL); |
| 293 | zero_page = ZERO_PAGE(0); |
| 294 | page_cache_get(zero_page); |
| 295 | |
| 296 | /* |
| 297 | * The number of active work items is limited by the number of |
| 298 | * connections, so leave @max_active at default. |
| 299 | */ |
| 300 | ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0); |
| 301 | if (ceph_msgr_wq) |
| 302 | return 0; |
| 303 | |
| 304 | pr_err("msgr_init failed to create workqueue\n"); |
| 305 | _ceph_msgr_exit(); |
| 306 | |
| 307 | return -ENOMEM; |
| 308 | } |
| 309 | EXPORT_SYMBOL(ceph_msgr_init); |
| 310 | |
| 311 | void ceph_msgr_exit(void) |
| 312 | { |
| 313 | BUG_ON(ceph_msgr_wq == NULL); |
| 314 | |
| 315 | _ceph_msgr_exit(); |
| 316 | } |
| 317 | EXPORT_SYMBOL(ceph_msgr_exit); |
| 318 | |
| 319 | void ceph_msgr_flush(void) |
| 320 | { |
| 321 | flush_workqueue(ceph_msgr_wq); |
| 322 | } |
| 323 | EXPORT_SYMBOL(ceph_msgr_flush); |
| 324 | |
| 325 | /* Connection socket state transition functions */ |
| 326 | |
| 327 | static void con_sock_state_init(struct ceph_connection *con) |
| 328 | { |
| 329 | int old_state; |
| 330 | |
| 331 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED); |
| 332 | if (WARN_ON(old_state != CON_SOCK_STATE_NEW)) |
| 333 | printk("%s: unexpected old state %d\n", __func__, old_state); |
| 334 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
| 335 | CON_SOCK_STATE_CLOSED); |
| 336 | } |
| 337 | |
| 338 | static void con_sock_state_connecting(struct ceph_connection *con) |
| 339 | { |
| 340 | int old_state; |
| 341 | |
| 342 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING); |
| 343 | if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED)) |
| 344 | printk("%s: unexpected old state %d\n", __func__, old_state); |
| 345 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
| 346 | CON_SOCK_STATE_CONNECTING); |
| 347 | } |
| 348 | |
| 349 | static void con_sock_state_connected(struct ceph_connection *con) |
| 350 | { |
| 351 | int old_state; |
| 352 | |
| 353 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED); |
| 354 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING)) |
| 355 | printk("%s: unexpected old state %d\n", __func__, old_state); |
| 356 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
| 357 | CON_SOCK_STATE_CONNECTED); |
| 358 | } |
| 359 | |
| 360 | static void con_sock_state_closing(struct ceph_connection *con) |
| 361 | { |
| 362 | int old_state; |
| 363 | |
| 364 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING); |
| 365 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING && |
| 366 | old_state != CON_SOCK_STATE_CONNECTED && |
| 367 | old_state != CON_SOCK_STATE_CLOSING)) |
| 368 | printk("%s: unexpected old state %d\n", __func__, old_state); |
| 369 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
| 370 | CON_SOCK_STATE_CLOSING); |
| 371 | } |
| 372 | |
| 373 | static void con_sock_state_closed(struct ceph_connection *con) |
| 374 | { |
| 375 | int old_state; |
| 376 | |
| 377 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED); |
| 378 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED && |
| 379 | old_state != CON_SOCK_STATE_CLOSING && |
| 380 | old_state != CON_SOCK_STATE_CONNECTING && |
| 381 | old_state != CON_SOCK_STATE_CLOSED)) |
| 382 | printk("%s: unexpected old state %d\n", __func__, old_state); |
| 383 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
| 384 | CON_SOCK_STATE_CLOSED); |
| 385 | } |
| 386 | |
| 387 | /* |
| 388 | * socket callback functions |
| 389 | */ |
| 390 | |
| 391 | /* data available on socket, or listen socket received a connect */ |
| 392 | static void ceph_sock_data_ready(struct sock *sk) |
| 393 | { |
| 394 | struct ceph_connection *con = sk->sk_user_data; |
| 395 | if (atomic_read(&con->msgr->stopping)) { |
| 396 | return; |
| 397 | } |
| 398 | |
| 399 | if (sk->sk_state != TCP_CLOSE_WAIT) { |
| 400 | dout("%s on %p state = %lu, queueing work\n", __func__, |
| 401 | con, con->state); |
| 402 | queue_con(con); |
| 403 | } |
| 404 | } |
| 405 | |
| 406 | /* socket has buffer space for writing */ |
| 407 | static void ceph_sock_write_space(struct sock *sk) |
| 408 | { |
| 409 | struct ceph_connection *con = sk->sk_user_data; |
| 410 | |
| 411 | /* only queue to workqueue if there is data we want to write, |
| 412 | * and there is sufficient space in the socket buffer to accept |
| 413 | * more data. clear SOCK_NOSPACE so that ceph_sock_write_space() |
| 414 | * doesn't get called again until try_write() fills the socket |
| 415 | * buffer. See net/ipv4/tcp_input.c:tcp_check_space() |
| 416 | * and net/core/stream.c:sk_stream_write_space(). |
| 417 | */ |
| 418 | if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) { |
| 419 | if (sk_stream_is_writeable(sk)) { |
| 420 | dout("%s %p queueing write work\n", __func__, con); |
| 421 | clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags); |
| 422 | queue_con(con); |
| 423 | } |
| 424 | } else { |
| 425 | dout("%s %p nothing to write\n", __func__, con); |
| 426 | } |
| 427 | } |
| 428 | |
| 429 | /* socket's state has changed */ |
| 430 | static void ceph_sock_state_change(struct sock *sk) |
| 431 | { |
| 432 | struct ceph_connection *con = sk->sk_user_data; |
| 433 | |
| 434 | dout("%s %p state = %lu sk_state = %u\n", __func__, |
| 435 | con, con->state, sk->sk_state); |
| 436 | |
| 437 | switch (sk->sk_state) { |
| 438 | case TCP_CLOSE: |
| 439 | dout("%s TCP_CLOSE\n", __func__); |
| 440 | case TCP_CLOSE_WAIT: |
| 441 | dout("%s TCP_CLOSE_WAIT\n", __func__); |
| 442 | con_sock_state_closing(con); |
| 443 | con_flag_set(con, CON_FLAG_SOCK_CLOSED); |
| 444 | queue_con(con); |
| 445 | break; |
| 446 | case TCP_ESTABLISHED: |
| 447 | dout("%s TCP_ESTABLISHED\n", __func__); |
| 448 | con_sock_state_connected(con); |
| 449 | queue_con(con); |
| 450 | break; |
| 451 | default: /* Everything else is uninteresting */ |
| 452 | break; |
| 453 | } |
| 454 | } |
| 455 | |
| 456 | /* |
| 457 | * set up socket callbacks |
| 458 | */ |
| 459 | static void set_sock_callbacks(struct socket *sock, |
| 460 | struct ceph_connection *con) |
| 461 | { |
| 462 | struct sock *sk = sock->sk; |
| 463 | sk->sk_user_data = con; |
| 464 | sk->sk_data_ready = ceph_sock_data_ready; |
| 465 | sk->sk_write_space = ceph_sock_write_space; |
| 466 | sk->sk_state_change = ceph_sock_state_change; |
| 467 | } |
| 468 | |
| 469 | |
| 470 | /* |
| 471 | * socket helpers |
| 472 | */ |
| 473 | |
| 474 | /* |
| 475 | * initiate connection to a remote socket. |
| 476 | */ |
| 477 | static int ceph_tcp_connect(struct ceph_connection *con) |
| 478 | { |
| 479 | struct sockaddr_storage *paddr = &con->peer_addr.in_addr; |
| 480 | struct socket *sock; |
| 481 | int ret; |
| 482 | |
| 483 | BUG_ON(con->sock); |
| 484 | ret = sock_create_kern(read_pnet(&con->msgr->net), paddr->ss_family, |
| 485 | SOCK_STREAM, IPPROTO_TCP, &sock); |
| 486 | if (ret) |
| 487 | return ret; |
| 488 | sock->sk->sk_allocation = GFP_NOFS; |
| 489 | |
| 490 | #ifdef CONFIG_LOCKDEP |
| 491 | lockdep_set_class(&sock->sk->sk_lock, &socket_class); |
| 492 | #endif |
| 493 | |
| 494 | set_sock_callbacks(sock, con); |
| 495 | |
| 496 | dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr)); |
| 497 | |
| 498 | con_sock_state_connecting(con); |
| 499 | ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr), |
| 500 | O_NONBLOCK); |
| 501 | if (ret == -EINPROGRESS) { |
| 502 | dout("connect %s EINPROGRESS sk_state = %u\n", |
| 503 | ceph_pr_addr(&con->peer_addr.in_addr), |
| 504 | sock->sk->sk_state); |
| 505 | } else if (ret < 0) { |
| 506 | pr_err("connect %s error %d\n", |
| 507 | ceph_pr_addr(&con->peer_addr.in_addr), ret); |
| 508 | sock_release(sock); |
| 509 | return ret; |
| 510 | } |
| 511 | |
| 512 | if (con->msgr->tcp_nodelay) { |
| 513 | int optval = 1; |
| 514 | |
| 515 | ret = kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY, |
| 516 | (char *)&optval, sizeof(optval)); |
| 517 | if (ret) |
| 518 | pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d", |
| 519 | ret); |
| 520 | } |
| 521 | |
| 522 | con->sock = sock; |
| 523 | return 0; |
| 524 | } |
| 525 | |
| 526 | static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len) |
| 527 | { |
| 528 | struct kvec iov = {buf, len}; |
| 529 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; |
| 530 | int r; |
| 531 | |
| 532 | r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags); |
| 533 | if (r == -EAGAIN) |
| 534 | r = 0; |
| 535 | return r; |
| 536 | } |
| 537 | |
| 538 | static int ceph_tcp_recvpage(struct socket *sock, struct page *page, |
| 539 | int page_offset, size_t length) |
| 540 | { |
| 541 | void *kaddr; |
| 542 | int ret; |
| 543 | |
| 544 | BUG_ON(page_offset + length > PAGE_SIZE); |
| 545 | |
| 546 | kaddr = kmap(page); |
| 547 | BUG_ON(!kaddr); |
| 548 | ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length); |
| 549 | kunmap(page); |
| 550 | |
| 551 | return ret; |
| 552 | } |
| 553 | |
| 554 | /* |
| 555 | * write something. @more is true if caller will be sending more data |
| 556 | * shortly. |
| 557 | */ |
| 558 | static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov, |
| 559 | size_t kvlen, size_t len, int more) |
| 560 | { |
| 561 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; |
| 562 | int r; |
| 563 | |
| 564 | if (more) |
| 565 | msg.msg_flags |= MSG_MORE; |
| 566 | else |
| 567 | msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */ |
| 568 | |
| 569 | r = kernel_sendmsg(sock, &msg, iov, kvlen, len); |
| 570 | if (r == -EAGAIN) |
| 571 | r = 0; |
| 572 | return r; |
| 573 | } |
| 574 | |
| 575 | static int __ceph_tcp_sendpage(struct socket *sock, struct page *page, |
| 576 | int offset, size_t size, bool more) |
| 577 | { |
| 578 | int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR); |
| 579 | int ret; |
| 580 | |
| 581 | ret = kernel_sendpage(sock, page, offset, size, flags); |
| 582 | if (ret == -EAGAIN) |
| 583 | ret = 0; |
| 584 | |
| 585 | return ret; |
| 586 | } |
| 587 | |
| 588 | static int ceph_tcp_sendpage(struct socket *sock, struct page *page, |
| 589 | int offset, size_t size, bool more) |
| 590 | { |
| 591 | int ret; |
| 592 | struct kvec iov; |
| 593 | |
| 594 | /* sendpage cannot properly handle pages with page_count == 0, |
| 595 | * we need to fallback to sendmsg if that's the case */ |
| 596 | if (page_count(page) >= 1) |
| 597 | return __ceph_tcp_sendpage(sock, page, offset, size, more); |
| 598 | |
| 599 | iov.iov_base = kmap(page) + offset; |
| 600 | iov.iov_len = size; |
| 601 | ret = ceph_tcp_sendmsg(sock, &iov, 1, size, more); |
| 602 | kunmap(page); |
| 603 | |
| 604 | return ret; |
| 605 | } |
| 606 | |
| 607 | /* |
| 608 | * Shutdown/close the socket for the given connection. |
| 609 | */ |
| 610 | static int con_close_socket(struct ceph_connection *con) |
| 611 | { |
| 612 | int rc = 0; |
| 613 | |
| 614 | dout("con_close_socket on %p sock %p\n", con, con->sock); |
| 615 | if (con->sock) { |
| 616 | rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR); |
| 617 | sock_release(con->sock); |
| 618 | con->sock = NULL; |
| 619 | } |
| 620 | |
| 621 | /* |
| 622 | * Forcibly clear the SOCK_CLOSED flag. It gets set |
| 623 | * independent of the connection mutex, and we could have |
| 624 | * received a socket close event before we had the chance to |
| 625 | * shut the socket down. |
| 626 | */ |
| 627 | con_flag_clear(con, CON_FLAG_SOCK_CLOSED); |
| 628 | |
| 629 | con_sock_state_closed(con); |
| 630 | return rc; |
| 631 | } |
| 632 | |
| 633 | /* |
| 634 | * Reset a connection. Discard all incoming and outgoing messages |
| 635 | * and clear *_seq state. |
| 636 | */ |
| 637 | static void ceph_msg_remove(struct ceph_msg *msg) |
| 638 | { |
| 639 | list_del_init(&msg->list_head); |
| 640 | BUG_ON(msg->con == NULL); |
| 641 | msg->con->ops->put(msg->con); |
| 642 | msg->con = NULL; |
| 643 | |
| 644 | ceph_msg_put(msg); |
| 645 | } |
| 646 | static void ceph_msg_remove_list(struct list_head *head) |
| 647 | { |
| 648 | while (!list_empty(head)) { |
| 649 | struct ceph_msg *msg = list_first_entry(head, struct ceph_msg, |
| 650 | list_head); |
| 651 | ceph_msg_remove(msg); |
| 652 | } |
| 653 | } |
| 654 | |
| 655 | static void reset_connection(struct ceph_connection *con) |
| 656 | { |
| 657 | /* reset connection, out_queue, msg_ and connect_seq */ |
| 658 | /* discard existing out_queue and msg_seq */ |
| 659 | dout("reset_connection %p\n", con); |
| 660 | ceph_msg_remove_list(&con->out_queue); |
| 661 | ceph_msg_remove_list(&con->out_sent); |
| 662 | |
| 663 | if (con->in_msg) { |
| 664 | BUG_ON(con->in_msg->con != con); |
| 665 | con->in_msg->con = NULL; |
| 666 | ceph_msg_put(con->in_msg); |
| 667 | con->in_msg = NULL; |
| 668 | con->ops->put(con); |
| 669 | } |
| 670 | |
| 671 | con->connect_seq = 0; |
| 672 | con->out_seq = 0; |
| 673 | if (con->out_msg) { |
| 674 | ceph_msg_put(con->out_msg); |
| 675 | con->out_msg = NULL; |
| 676 | } |
| 677 | con->in_seq = 0; |
| 678 | con->in_seq_acked = 0; |
| 679 | } |
| 680 | |
| 681 | /* |
| 682 | * mark a peer down. drop any open connections. |
| 683 | */ |
| 684 | void ceph_con_close(struct ceph_connection *con) |
| 685 | { |
| 686 | mutex_lock(&con->mutex); |
| 687 | dout("con_close %p peer %s\n", con, |
| 688 | ceph_pr_addr(&con->peer_addr.in_addr)); |
| 689 | con->state = CON_STATE_CLOSED; |
| 690 | |
| 691 | con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */ |
| 692 | con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING); |
| 693 | con_flag_clear(con, CON_FLAG_WRITE_PENDING); |
| 694 | con_flag_clear(con, CON_FLAG_BACKOFF); |
| 695 | |
| 696 | reset_connection(con); |
| 697 | con->peer_global_seq = 0; |
| 698 | cancel_con(con); |
| 699 | con_close_socket(con); |
| 700 | mutex_unlock(&con->mutex); |
| 701 | } |
| 702 | EXPORT_SYMBOL(ceph_con_close); |
| 703 | |
| 704 | /* |
| 705 | * Reopen a closed connection, with a new peer address. |
| 706 | */ |
| 707 | void ceph_con_open(struct ceph_connection *con, |
| 708 | __u8 entity_type, __u64 entity_num, |
| 709 | struct ceph_entity_addr *addr) |
| 710 | { |
| 711 | mutex_lock(&con->mutex); |
| 712 | dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr)); |
| 713 | |
| 714 | WARN_ON(con->state != CON_STATE_CLOSED); |
| 715 | con->state = CON_STATE_PREOPEN; |
| 716 | |
| 717 | con->peer_name.type = (__u8) entity_type; |
| 718 | con->peer_name.num = cpu_to_le64(entity_num); |
| 719 | |
| 720 | memcpy(&con->peer_addr, addr, sizeof(*addr)); |
| 721 | con->delay = 0; /* reset backoff memory */ |
| 722 | mutex_unlock(&con->mutex); |
| 723 | queue_con(con); |
| 724 | } |
| 725 | EXPORT_SYMBOL(ceph_con_open); |
| 726 | |
| 727 | /* |
| 728 | * return true if this connection ever successfully opened |
| 729 | */ |
| 730 | bool ceph_con_opened(struct ceph_connection *con) |
| 731 | { |
| 732 | return con->connect_seq > 0; |
| 733 | } |
| 734 | |
| 735 | /* |
| 736 | * initialize a new connection. |
| 737 | */ |
| 738 | void ceph_con_init(struct ceph_connection *con, void *private, |
| 739 | const struct ceph_connection_operations *ops, |
| 740 | struct ceph_messenger *msgr) |
| 741 | { |
| 742 | dout("con_init %p\n", con); |
| 743 | memset(con, 0, sizeof(*con)); |
| 744 | con->private = private; |
| 745 | con->ops = ops; |
| 746 | con->msgr = msgr; |
| 747 | |
| 748 | con_sock_state_init(con); |
| 749 | |
| 750 | mutex_init(&con->mutex); |
| 751 | INIT_LIST_HEAD(&con->out_queue); |
| 752 | INIT_LIST_HEAD(&con->out_sent); |
| 753 | INIT_DELAYED_WORK(&con->work, ceph_con_workfn); |
| 754 | |
| 755 | con->state = CON_STATE_CLOSED; |
| 756 | } |
| 757 | EXPORT_SYMBOL(ceph_con_init); |
| 758 | |
| 759 | |
| 760 | /* |
| 761 | * We maintain a global counter to order connection attempts. Get |
| 762 | * a unique seq greater than @gt. |
| 763 | */ |
| 764 | static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt) |
| 765 | { |
| 766 | u32 ret; |
| 767 | |
| 768 | spin_lock(&msgr->global_seq_lock); |
| 769 | if (msgr->global_seq < gt) |
| 770 | msgr->global_seq = gt; |
| 771 | ret = ++msgr->global_seq; |
| 772 | spin_unlock(&msgr->global_seq_lock); |
| 773 | return ret; |
| 774 | } |
| 775 | |
| 776 | static void con_out_kvec_reset(struct ceph_connection *con) |
| 777 | { |
| 778 | con->out_kvec_left = 0; |
| 779 | con->out_kvec_bytes = 0; |
| 780 | con->out_kvec_cur = &con->out_kvec[0]; |
| 781 | } |
| 782 | |
| 783 | static void con_out_kvec_add(struct ceph_connection *con, |
| 784 | size_t size, void *data) |
| 785 | { |
| 786 | int index; |
| 787 | |
| 788 | index = con->out_kvec_left; |
| 789 | BUG_ON(index >= ARRAY_SIZE(con->out_kvec)); |
| 790 | |
| 791 | con->out_kvec[index].iov_len = size; |
| 792 | con->out_kvec[index].iov_base = data; |
| 793 | con->out_kvec_left++; |
| 794 | con->out_kvec_bytes += size; |
| 795 | } |
| 796 | |
| 797 | #ifdef CONFIG_BLOCK |
| 798 | |
| 799 | /* |
| 800 | * For a bio data item, a piece is whatever remains of the next |
| 801 | * entry in the current bio iovec, or the first entry in the next |
| 802 | * bio in the list. |
| 803 | */ |
| 804 | static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor, |
| 805 | size_t length) |
| 806 | { |
| 807 | struct ceph_msg_data *data = cursor->data; |
| 808 | struct bio *bio; |
| 809 | |
| 810 | BUG_ON(data->type != CEPH_MSG_DATA_BIO); |
| 811 | |
| 812 | bio = data->bio; |
| 813 | BUG_ON(!bio); |
| 814 | |
| 815 | cursor->resid = min(length, data->bio_length); |
| 816 | cursor->bio = bio; |
| 817 | cursor->bvec_iter = bio->bi_iter; |
| 818 | cursor->last_piece = |
| 819 | cursor->resid <= bio_iter_len(bio, cursor->bvec_iter); |
| 820 | } |
| 821 | |
| 822 | static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor, |
| 823 | size_t *page_offset, |
| 824 | size_t *length) |
| 825 | { |
| 826 | struct ceph_msg_data *data = cursor->data; |
| 827 | struct bio *bio; |
| 828 | struct bio_vec bio_vec; |
| 829 | |
| 830 | BUG_ON(data->type != CEPH_MSG_DATA_BIO); |
| 831 | |
| 832 | bio = cursor->bio; |
| 833 | BUG_ON(!bio); |
| 834 | |
| 835 | bio_vec = bio_iter_iovec(bio, cursor->bvec_iter); |
| 836 | |
| 837 | *page_offset = (size_t) bio_vec.bv_offset; |
| 838 | BUG_ON(*page_offset >= PAGE_SIZE); |
| 839 | if (cursor->last_piece) /* pagelist offset is always 0 */ |
| 840 | *length = cursor->resid; |
| 841 | else |
| 842 | *length = (size_t) bio_vec.bv_len; |
| 843 | BUG_ON(*length > cursor->resid); |
| 844 | BUG_ON(*page_offset + *length > PAGE_SIZE); |
| 845 | |
| 846 | return bio_vec.bv_page; |
| 847 | } |
| 848 | |
| 849 | static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor, |
| 850 | size_t bytes) |
| 851 | { |
| 852 | struct bio *bio; |
| 853 | struct bio_vec bio_vec; |
| 854 | |
| 855 | BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO); |
| 856 | |
| 857 | bio = cursor->bio; |
| 858 | BUG_ON(!bio); |
| 859 | |
| 860 | bio_vec = bio_iter_iovec(bio, cursor->bvec_iter); |
| 861 | |
| 862 | /* Advance the cursor offset */ |
| 863 | |
| 864 | BUG_ON(cursor->resid < bytes); |
| 865 | cursor->resid -= bytes; |
| 866 | |
| 867 | bio_advance_iter(bio, &cursor->bvec_iter, bytes); |
| 868 | |
| 869 | if (bytes < bio_vec.bv_len) |
| 870 | return false; /* more bytes to process in this segment */ |
| 871 | |
| 872 | /* Move on to the next segment, and possibly the next bio */ |
| 873 | |
| 874 | if (!cursor->bvec_iter.bi_size) { |
| 875 | bio = bio->bi_next; |
| 876 | cursor->bio = bio; |
| 877 | if (bio) |
| 878 | cursor->bvec_iter = bio->bi_iter; |
| 879 | else |
| 880 | memset(&cursor->bvec_iter, 0, |
| 881 | sizeof(cursor->bvec_iter)); |
| 882 | } |
| 883 | |
| 884 | if (!cursor->last_piece) { |
| 885 | BUG_ON(!cursor->resid); |
| 886 | BUG_ON(!bio); |
| 887 | /* A short read is OK, so use <= rather than == */ |
| 888 | if (cursor->resid <= bio_iter_len(bio, cursor->bvec_iter)) |
| 889 | cursor->last_piece = true; |
| 890 | } |
| 891 | |
| 892 | return true; |
| 893 | } |
| 894 | #endif /* CONFIG_BLOCK */ |
| 895 | |
| 896 | /* |
| 897 | * For a page array, a piece comes from the first page in the array |
| 898 | * that has not already been fully consumed. |
| 899 | */ |
| 900 | static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor, |
| 901 | size_t length) |
| 902 | { |
| 903 | struct ceph_msg_data *data = cursor->data; |
| 904 | int page_count; |
| 905 | |
| 906 | BUG_ON(data->type != CEPH_MSG_DATA_PAGES); |
| 907 | |
| 908 | BUG_ON(!data->pages); |
| 909 | BUG_ON(!data->length); |
| 910 | |
| 911 | cursor->resid = min(length, data->length); |
| 912 | page_count = calc_pages_for(data->alignment, (u64)data->length); |
| 913 | cursor->page_offset = data->alignment & ~PAGE_MASK; |
| 914 | cursor->page_index = 0; |
| 915 | BUG_ON(page_count > (int)USHRT_MAX); |
| 916 | cursor->page_count = (unsigned short)page_count; |
| 917 | BUG_ON(length > SIZE_MAX - cursor->page_offset); |
| 918 | cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE; |
| 919 | } |
| 920 | |
| 921 | static struct page * |
| 922 | ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor, |
| 923 | size_t *page_offset, size_t *length) |
| 924 | { |
| 925 | struct ceph_msg_data *data = cursor->data; |
| 926 | |
| 927 | BUG_ON(data->type != CEPH_MSG_DATA_PAGES); |
| 928 | |
| 929 | BUG_ON(cursor->page_index >= cursor->page_count); |
| 930 | BUG_ON(cursor->page_offset >= PAGE_SIZE); |
| 931 | |
| 932 | *page_offset = cursor->page_offset; |
| 933 | if (cursor->last_piece) |
| 934 | *length = cursor->resid; |
| 935 | else |
| 936 | *length = PAGE_SIZE - *page_offset; |
| 937 | |
| 938 | return data->pages[cursor->page_index]; |
| 939 | } |
| 940 | |
| 941 | static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor, |
| 942 | size_t bytes) |
| 943 | { |
| 944 | BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES); |
| 945 | |
| 946 | BUG_ON(cursor->page_offset + bytes > PAGE_SIZE); |
| 947 | |
| 948 | /* Advance the cursor page offset */ |
| 949 | |
| 950 | cursor->resid -= bytes; |
| 951 | cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK; |
| 952 | if (!bytes || cursor->page_offset) |
| 953 | return false; /* more bytes to process in the current page */ |
| 954 | |
| 955 | if (!cursor->resid) |
| 956 | return false; /* no more data */ |
| 957 | |
| 958 | /* Move on to the next page; offset is already at 0 */ |
| 959 | |
| 960 | BUG_ON(cursor->page_index >= cursor->page_count); |
| 961 | cursor->page_index++; |
| 962 | cursor->last_piece = cursor->resid <= PAGE_SIZE; |
| 963 | |
| 964 | return true; |
| 965 | } |
| 966 | |
| 967 | /* |
| 968 | * For a pagelist, a piece is whatever remains to be consumed in the |
| 969 | * first page in the list, or the front of the next page. |
| 970 | */ |
| 971 | static void |
| 972 | ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor, |
| 973 | size_t length) |
| 974 | { |
| 975 | struct ceph_msg_data *data = cursor->data; |
| 976 | struct ceph_pagelist *pagelist; |
| 977 | struct page *page; |
| 978 | |
| 979 | BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST); |
| 980 | |
| 981 | pagelist = data->pagelist; |
| 982 | BUG_ON(!pagelist); |
| 983 | |
| 984 | if (!length) |
| 985 | return; /* pagelist can be assigned but empty */ |
| 986 | |
| 987 | BUG_ON(list_empty(&pagelist->head)); |
| 988 | page = list_first_entry(&pagelist->head, struct page, lru); |
| 989 | |
| 990 | cursor->resid = min(length, pagelist->length); |
| 991 | cursor->page = page; |
| 992 | cursor->offset = 0; |
| 993 | cursor->last_piece = cursor->resid <= PAGE_SIZE; |
| 994 | } |
| 995 | |
| 996 | static struct page * |
| 997 | ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor, |
| 998 | size_t *page_offset, size_t *length) |
| 999 | { |
| 1000 | struct ceph_msg_data *data = cursor->data; |
| 1001 | struct ceph_pagelist *pagelist; |
| 1002 | |
| 1003 | BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST); |
| 1004 | |
| 1005 | pagelist = data->pagelist; |
| 1006 | BUG_ON(!pagelist); |
| 1007 | |
| 1008 | BUG_ON(!cursor->page); |
| 1009 | BUG_ON(cursor->offset + cursor->resid != pagelist->length); |
| 1010 | |
| 1011 | /* offset of first page in pagelist is always 0 */ |
| 1012 | *page_offset = cursor->offset & ~PAGE_MASK; |
| 1013 | if (cursor->last_piece) |
| 1014 | *length = cursor->resid; |
| 1015 | else |
| 1016 | *length = PAGE_SIZE - *page_offset; |
| 1017 | |
| 1018 | return cursor->page; |
| 1019 | } |
| 1020 | |
| 1021 | static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor, |
| 1022 | size_t bytes) |
| 1023 | { |
| 1024 | struct ceph_msg_data *data = cursor->data; |
| 1025 | struct ceph_pagelist *pagelist; |
| 1026 | |
| 1027 | BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST); |
| 1028 | |
| 1029 | pagelist = data->pagelist; |
| 1030 | BUG_ON(!pagelist); |
| 1031 | |
| 1032 | BUG_ON(cursor->offset + cursor->resid != pagelist->length); |
| 1033 | BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE); |
| 1034 | |
| 1035 | /* Advance the cursor offset */ |
| 1036 | |
| 1037 | cursor->resid -= bytes; |
| 1038 | cursor->offset += bytes; |
| 1039 | /* offset of first page in pagelist is always 0 */ |
| 1040 | if (!bytes || cursor->offset & ~PAGE_MASK) |
| 1041 | return false; /* more bytes to process in the current page */ |
| 1042 | |
| 1043 | if (!cursor->resid) |
| 1044 | return false; /* no more data */ |
| 1045 | |
| 1046 | /* Move on to the next page */ |
| 1047 | |
| 1048 | BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head)); |
| 1049 | cursor->page = list_entry_next(cursor->page, lru); |
| 1050 | cursor->last_piece = cursor->resid <= PAGE_SIZE; |
| 1051 | |
| 1052 | return true; |
| 1053 | } |
| 1054 | |
| 1055 | /* |
| 1056 | * Message data is handled (sent or received) in pieces, where each |
| 1057 | * piece resides on a single page. The network layer might not |
| 1058 | * consume an entire piece at once. A data item's cursor keeps |
| 1059 | * track of which piece is next to process and how much remains to |
| 1060 | * be processed in that piece. It also tracks whether the current |
| 1061 | * piece is the last one in the data item. |
| 1062 | */ |
| 1063 | static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor) |
| 1064 | { |
| 1065 | size_t length = cursor->total_resid; |
| 1066 | |
| 1067 | switch (cursor->data->type) { |
| 1068 | case CEPH_MSG_DATA_PAGELIST: |
| 1069 | ceph_msg_data_pagelist_cursor_init(cursor, length); |
| 1070 | break; |
| 1071 | case CEPH_MSG_DATA_PAGES: |
| 1072 | ceph_msg_data_pages_cursor_init(cursor, length); |
| 1073 | break; |
| 1074 | #ifdef CONFIG_BLOCK |
| 1075 | case CEPH_MSG_DATA_BIO: |
| 1076 | ceph_msg_data_bio_cursor_init(cursor, length); |
| 1077 | break; |
| 1078 | #endif /* CONFIG_BLOCK */ |
| 1079 | case CEPH_MSG_DATA_NONE: |
| 1080 | default: |
| 1081 | /* BUG(); */ |
| 1082 | break; |
| 1083 | } |
| 1084 | cursor->need_crc = true; |
| 1085 | } |
| 1086 | |
| 1087 | static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length) |
| 1088 | { |
| 1089 | struct ceph_msg_data_cursor *cursor = &msg->cursor; |
| 1090 | struct ceph_msg_data *data; |
| 1091 | |
| 1092 | BUG_ON(!length); |
| 1093 | BUG_ON(length > msg->data_length); |
| 1094 | BUG_ON(list_empty(&msg->data)); |
| 1095 | |
| 1096 | cursor->data_head = &msg->data; |
| 1097 | cursor->total_resid = length; |
| 1098 | data = list_first_entry(&msg->data, struct ceph_msg_data, links); |
| 1099 | cursor->data = data; |
| 1100 | |
| 1101 | __ceph_msg_data_cursor_init(cursor); |
| 1102 | } |
| 1103 | |
| 1104 | /* |
| 1105 | * Return the page containing the next piece to process for a given |
| 1106 | * data item, and supply the page offset and length of that piece. |
| 1107 | * Indicate whether this is the last piece in this data item. |
| 1108 | */ |
| 1109 | static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor, |
| 1110 | size_t *page_offset, size_t *length, |
| 1111 | bool *last_piece) |
| 1112 | { |
| 1113 | struct page *page; |
| 1114 | |
| 1115 | switch (cursor->data->type) { |
| 1116 | case CEPH_MSG_DATA_PAGELIST: |
| 1117 | page = ceph_msg_data_pagelist_next(cursor, page_offset, length); |
| 1118 | break; |
| 1119 | case CEPH_MSG_DATA_PAGES: |
| 1120 | page = ceph_msg_data_pages_next(cursor, page_offset, length); |
| 1121 | break; |
| 1122 | #ifdef CONFIG_BLOCK |
| 1123 | case CEPH_MSG_DATA_BIO: |
| 1124 | page = ceph_msg_data_bio_next(cursor, page_offset, length); |
| 1125 | break; |
| 1126 | #endif /* CONFIG_BLOCK */ |
| 1127 | case CEPH_MSG_DATA_NONE: |
| 1128 | default: |
| 1129 | page = NULL; |
| 1130 | break; |
| 1131 | } |
| 1132 | BUG_ON(!page); |
| 1133 | BUG_ON(*page_offset + *length > PAGE_SIZE); |
| 1134 | BUG_ON(!*length); |
| 1135 | if (last_piece) |
| 1136 | *last_piece = cursor->last_piece; |
| 1137 | |
| 1138 | return page; |
| 1139 | } |
| 1140 | |
| 1141 | /* |
| 1142 | * Returns true if the result moves the cursor on to the next piece |
| 1143 | * of the data item. |
| 1144 | */ |
| 1145 | static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor, |
| 1146 | size_t bytes) |
| 1147 | { |
| 1148 | bool new_piece; |
| 1149 | |
| 1150 | BUG_ON(bytes > cursor->resid); |
| 1151 | switch (cursor->data->type) { |
| 1152 | case CEPH_MSG_DATA_PAGELIST: |
| 1153 | new_piece = ceph_msg_data_pagelist_advance(cursor, bytes); |
| 1154 | break; |
| 1155 | case CEPH_MSG_DATA_PAGES: |
| 1156 | new_piece = ceph_msg_data_pages_advance(cursor, bytes); |
| 1157 | break; |
| 1158 | #ifdef CONFIG_BLOCK |
| 1159 | case CEPH_MSG_DATA_BIO: |
| 1160 | new_piece = ceph_msg_data_bio_advance(cursor, bytes); |
| 1161 | break; |
| 1162 | #endif /* CONFIG_BLOCK */ |
| 1163 | case CEPH_MSG_DATA_NONE: |
| 1164 | default: |
| 1165 | BUG(); |
| 1166 | break; |
| 1167 | } |
| 1168 | cursor->total_resid -= bytes; |
| 1169 | |
| 1170 | if (!cursor->resid && cursor->total_resid) { |
| 1171 | WARN_ON(!cursor->last_piece); |
| 1172 | BUG_ON(list_is_last(&cursor->data->links, cursor->data_head)); |
| 1173 | cursor->data = list_entry_next(cursor->data, links); |
| 1174 | __ceph_msg_data_cursor_init(cursor); |
| 1175 | new_piece = true; |
| 1176 | } |
| 1177 | cursor->need_crc = new_piece; |
| 1178 | |
| 1179 | return new_piece; |
| 1180 | } |
| 1181 | |
| 1182 | static void prepare_message_data(struct ceph_msg *msg, u32 data_len) |
| 1183 | { |
| 1184 | BUG_ON(!msg); |
| 1185 | BUG_ON(!data_len); |
| 1186 | |
| 1187 | /* Initialize data cursor */ |
| 1188 | |
| 1189 | ceph_msg_data_cursor_init(msg, (size_t)data_len); |
| 1190 | } |
| 1191 | |
| 1192 | /* |
| 1193 | * Prepare footer for currently outgoing message, and finish things |
| 1194 | * off. Assumes out_kvec* are already valid.. we just add on to the end. |
| 1195 | */ |
| 1196 | static void prepare_write_message_footer(struct ceph_connection *con) |
| 1197 | { |
| 1198 | struct ceph_msg *m = con->out_msg; |
| 1199 | int v = con->out_kvec_left; |
| 1200 | |
| 1201 | m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE; |
| 1202 | |
| 1203 | dout("prepare_write_message_footer %p\n", con); |
| 1204 | con->out_kvec_is_msg = true; |
| 1205 | con->out_kvec[v].iov_base = &m->footer; |
| 1206 | if (con->peer_features & CEPH_FEATURE_MSG_AUTH) { |
| 1207 | if (con->ops->sign_message) |
| 1208 | con->ops->sign_message(con, m); |
| 1209 | else |
| 1210 | m->footer.sig = 0; |
| 1211 | con->out_kvec[v].iov_len = sizeof(m->footer); |
| 1212 | con->out_kvec_bytes += sizeof(m->footer); |
| 1213 | } else { |
| 1214 | m->old_footer.flags = m->footer.flags; |
| 1215 | con->out_kvec[v].iov_len = sizeof(m->old_footer); |
| 1216 | con->out_kvec_bytes += sizeof(m->old_footer); |
| 1217 | } |
| 1218 | con->out_kvec_left++; |
| 1219 | con->out_more = m->more_to_follow; |
| 1220 | con->out_msg_done = true; |
| 1221 | } |
| 1222 | |
| 1223 | /* |
| 1224 | * Prepare headers for the next outgoing message. |
| 1225 | */ |
| 1226 | static void prepare_write_message(struct ceph_connection *con) |
| 1227 | { |
| 1228 | struct ceph_msg *m; |
| 1229 | u32 crc; |
| 1230 | |
| 1231 | con_out_kvec_reset(con); |
| 1232 | con->out_kvec_is_msg = true; |
| 1233 | con->out_msg_done = false; |
| 1234 | |
| 1235 | /* Sneak an ack in there first? If we can get it into the same |
| 1236 | * TCP packet that's a good thing. */ |
| 1237 | if (con->in_seq > con->in_seq_acked) { |
| 1238 | con->in_seq_acked = con->in_seq; |
| 1239 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
| 1240 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
| 1241 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
| 1242 | &con->out_temp_ack); |
| 1243 | } |
| 1244 | |
| 1245 | BUG_ON(list_empty(&con->out_queue)); |
| 1246 | m = list_first_entry(&con->out_queue, struct ceph_msg, list_head); |
| 1247 | con->out_msg = m; |
| 1248 | BUG_ON(m->con != con); |
| 1249 | |
| 1250 | /* put message on sent list */ |
| 1251 | ceph_msg_get(m); |
| 1252 | list_move_tail(&m->list_head, &con->out_sent); |
| 1253 | |
| 1254 | /* |
| 1255 | * only assign outgoing seq # if we haven't sent this message |
| 1256 | * yet. if it is requeued, resend with it's original seq. |
| 1257 | */ |
| 1258 | if (m->needs_out_seq) { |
| 1259 | m->hdr.seq = cpu_to_le64(++con->out_seq); |
| 1260 | m->needs_out_seq = false; |
| 1261 | } |
| 1262 | WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len)); |
| 1263 | |
| 1264 | dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n", |
| 1265 | m, con->out_seq, le16_to_cpu(m->hdr.type), |
| 1266 | le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len), |
| 1267 | m->data_length); |
| 1268 | BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len); |
| 1269 | |
| 1270 | /* tag + hdr + front + middle */ |
| 1271 | con_out_kvec_add(con, sizeof (tag_msg), &tag_msg); |
| 1272 | con_out_kvec_add(con, sizeof (m->hdr), &m->hdr); |
| 1273 | con_out_kvec_add(con, m->front.iov_len, m->front.iov_base); |
| 1274 | |
| 1275 | if (m->middle) |
| 1276 | con_out_kvec_add(con, m->middle->vec.iov_len, |
| 1277 | m->middle->vec.iov_base); |
| 1278 | |
| 1279 | /* fill in crc (except data pages), footer */ |
| 1280 | crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc)); |
| 1281 | con->out_msg->hdr.crc = cpu_to_le32(crc); |
| 1282 | con->out_msg->footer.flags = 0; |
| 1283 | |
| 1284 | crc = crc32c(0, m->front.iov_base, m->front.iov_len); |
| 1285 | con->out_msg->footer.front_crc = cpu_to_le32(crc); |
| 1286 | if (m->middle) { |
| 1287 | crc = crc32c(0, m->middle->vec.iov_base, |
| 1288 | m->middle->vec.iov_len); |
| 1289 | con->out_msg->footer.middle_crc = cpu_to_le32(crc); |
| 1290 | } else |
| 1291 | con->out_msg->footer.middle_crc = 0; |
| 1292 | dout("%s front_crc %u middle_crc %u\n", __func__, |
| 1293 | le32_to_cpu(con->out_msg->footer.front_crc), |
| 1294 | le32_to_cpu(con->out_msg->footer.middle_crc)); |
| 1295 | |
| 1296 | /* is there a data payload? */ |
| 1297 | con->out_msg->footer.data_crc = 0; |
| 1298 | if (m->data_length) { |
| 1299 | prepare_message_data(con->out_msg, m->data_length); |
| 1300 | con->out_more = 1; /* data + footer will follow */ |
| 1301 | } else { |
| 1302 | /* no, queue up footer too and be done */ |
| 1303 | prepare_write_message_footer(con); |
| 1304 | } |
| 1305 | |
| 1306 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
| 1307 | } |
| 1308 | |
| 1309 | /* |
| 1310 | * Prepare an ack. |
| 1311 | */ |
| 1312 | static void prepare_write_ack(struct ceph_connection *con) |
| 1313 | { |
| 1314 | dout("prepare_write_ack %p %llu -> %llu\n", con, |
| 1315 | con->in_seq_acked, con->in_seq); |
| 1316 | con->in_seq_acked = con->in_seq; |
| 1317 | |
| 1318 | con_out_kvec_reset(con); |
| 1319 | |
| 1320 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
| 1321 | |
| 1322 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
| 1323 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
| 1324 | &con->out_temp_ack); |
| 1325 | |
| 1326 | con->out_more = 1; /* more will follow.. eventually.. */ |
| 1327 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
| 1328 | } |
| 1329 | |
| 1330 | /* |
| 1331 | * Prepare to share the seq during handshake |
| 1332 | */ |
| 1333 | static void prepare_write_seq(struct ceph_connection *con) |
| 1334 | { |
| 1335 | dout("prepare_write_seq %p %llu -> %llu\n", con, |
| 1336 | con->in_seq_acked, con->in_seq); |
| 1337 | con->in_seq_acked = con->in_seq; |
| 1338 | |
| 1339 | con_out_kvec_reset(con); |
| 1340 | |
| 1341 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
| 1342 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
| 1343 | &con->out_temp_ack); |
| 1344 | |
| 1345 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
| 1346 | } |
| 1347 | |
| 1348 | /* |
| 1349 | * Prepare to write keepalive byte. |
| 1350 | */ |
| 1351 | static void prepare_write_keepalive(struct ceph_connection *con) |
| 1352 | { |
| 1353 | dout("prepare_write_keepalive %p\n", con); |
| 1354 | con_out_kvec_reset(con); |
| 1355 | if (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2) { |
| 1356 | struct timespec now = CURRENT_TIME; |
| 1357 | |
| 1358 | con_out_kvec_add(con, sizeof(tag_keepalive2), &tag_keepalive2); |
| 1359 | ceph_encode_timespec(&con->out_temp_keepalive2, &now); |
| 1360 | con_out_kvec_add(con, sizeof(con->out_temp_keepalive2), |
| 1361 | &con->out_temp_keepalive2); |
| 1362 | } else { |
| 1363 | con_out_kvec_add(con, sizeof(tag_keepalive), &tag_keepalive); |
| 1364 | } |
| 1365 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
| 1366 | } |
| 1367 | |
| 1368 | /* |
| 1369 | * Connection negotiation. |
| 1370 | */ |
| 1371 | |
| 1372 | static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con, |
| 1373 | int *auth_proto) |
| 1374 | { |
| 1375 | struct ceph_auth_handshake *auth; |
| 1376 | |
| 1377 | if (!con->ops->get_authorizer) { |
| 1378 | con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN; |
| 1379 | con->out_connect.authorizer_len = 0; |
| 1380 | return NULL; |
| 1381 | } |
| 1382 | |
| 1383 | /* Can't hold the mutex while getting authorizer */ |
| 1384 | mutex_unlock(&con->mutex); |
| 1385 | auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry); |
| 1386 | mutex_lock(&con->mutex); |
| 1387 | |
| 1388 | if (IS_ERR(auth)) |
| 1389 | return auth; |
| 1390 | if (con->state != CON_STATE_NEGOTIATING) |
| 1391 | return ERR_PTR(-EAGAIN); |
| 1392 | |
| 1393 | con->auth_reply_buf = auth->authorizer_reply_buf; |
| 1394 | con->auth_reply_buf_len = auth->authorizer_reply_buf_len; |
| 1395 | return auth; |
| 1396 | } |
| 1397 | |
| 1398 | /* |
| 1399 | * We connected to a peer and are saying hello. |
| 1400 | */ |
| 1401 | static void prepare_write_banner(struct ceph_connection *con) |
| 1402 | { |
| 1403 | con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER); |
| 1404 | con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr), |
| 1405 | &con->msgr->my_enc_addr); |
| 1406 | |
| 1407 | con->out_more = 0; |
| 1408 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
| 1409 | } |
| 1410 | |
| 1411 | static int prepare_write_connect(struct ceph_connection *con) |
| 1412 | { |
| 1413 | unsigned int global_seq = get_global_seq(con->msgr, 0); |
| 1414 | int proto; |
| 1415 | int auth_proto; |
| 1416 | struct ceph_auth_handshake *auth; |
| 1417 | |
| 1418 | switch (con->peer_name.type) { |
| 1419 | case CEPH_ENTITY_TYPE_MON: |
| 1420 | proto = CEPH_MONC_PROTOCOL; |
| 1421 | break; |
| 1422 | case CEPH_ENTITY_TYPE_OSD: |
| 1423 | proto = CEPH_OSDC_PROTOCOL; |
| 1424 | break; |
| 1425 | case CEPH_ENTITY_TYPE_MDS: |
| 1426 | proto = CEPH_MDSC_PROTOCOL; |
| 1427 | break; |
| 1428 | default: |
| 1429 | BUG(); |
| 1430 | } |
| 1431 | |
| 1432 | dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con, |
| 1433 | con->connect_seq, global_seq, proto); |
| 1434 | |
| 1435 | con->out_connect.features = cpu_to_le64(con->msgr->supported_features); |
| 1436 | con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT); |
| 1437 | con->out_connect.connect_seq = cpu_to_le32(con->connect_seq); |
| 1438 | con->out_connect.global_seq = cpu_to_le32(global_seq); |
| 1439 | con->out_connect.protocol_version = cpu_to_le32(proto); |
| 1440 | con->out_connect.flags = 0; |
| 1441 | |
| 1442 | auth_proto = CEPH_AUTH_UNKNOWN; |
| 1443 | auth = get_connect_authorizer(con, &auth_proto); |
| 1444 | if (IS_ERR(auth)) |
| 1445 | return PTR_ERR(auth); |
| 1446 | |
| 1447 | con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto); |
| 1448 | con->out_connect.authorizer_len = auth ? |
| 1449 | cpu_to_le32(auth->authorizer_buf_len) : 0; |
| 1450 | |
| 1451 | con_out_kvec_add(con, sizeof (con->out_connect), |
| 1452 | &con->out_connect); |
| 1453 | if (auth && auth->authorizer_buf_len) |
| 1454 | con_out_kvec_add(con, auth->authorizer_buf_len, |
| 1455 | auth->authorizer_buf); |
| 1456 | |
| 1457 | con->out_more = 0; |
| 1458 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
| 1459 | |
| 1460 | return 0; |
| 1461 | } |
| 1462 | |
| 1463 | /* |
| 1464 | * write as much of pending kvecs to the socket as we can. |
| 1465 | * 1 -> done |
| 1466 | * 0 -> socket full, but more to do |
| 1467 | * <0 -> error |
| 1468 | */ |
| 1469 | static int write_partial_kvec(struct ceph_connection *con) |
| 1470 | { |
| 1471 | int ret; |
| 1472 | |
| 1473 | dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes); |
| 1474 | while (con->out_kvec_bytes > 0) { |
| 1475 | ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur, |
| 1476 | con->out_kvec_left, con->out_kvec_bytes, |
| 1477 | con->out_more); |
| 1478 | if (ret <= 0) |
| 1479 | goto out; |
| 1480 | con->out_kvec_bytes -= ret; |
| 1481 | if (con->out_kvec_bytes == 0) |
| 1482 | break; /* done */ |
| 1483 | |
| 1484 | /* account for full iov entries consumed */ |
| 1485 | while (ret >= con->out_kvec_cur->iov_len) { |
| 1486 | BUG_ON(!con->out_kvec_left); |
| 1487 | ret -= con->out_kvec_cur->iov_len; |
| 1488 | con->out_kvec_cur++; |
| 1489 | con->out_kvec_left--; |
| 1490 | } |
| 1491 | /* and for a partially-consumed entry */ |
| 1492 | if (ret) { |
| 1493 | con->out_kvec_cur->iov_len -= ret; |
| 1494 | con->out_kvec_cur->iov_base += ret; |
| 1495 | } |
| 1496 | } |
| 1497 | con->out_kvec_left = 0; |
| 1498 | con->out_kvec_is_msg = false; |
| 1499 | ret = 1; |
| 1500 | out: |
| 1501 | dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con, |
| 1502 | con->out_kvec_bytes, con->out_kvec_left, ret); |
| 1503 | return ret; /* done! */ |
| 1504 | } |
| 1505 | |
| 1506 | static u32 ceph_crc32c_page(u32 crc, struct page *page, |
| 1507 | unsigned int page_offset, |
| 1508 | unsigned int length) |
| 1509 | { |
| 1510 | char *kaddr; |
| 1511 | |
| 1512 | kaddr = kmap(page); |
| 1513 | BUG_ON(kaddr == NULL); |
| 1514 | crc = crc32c(crc, kaddr + page_offset, length); |
| 1515 | kunmap(page); |
| 1516 | |
| 1517 | return crc; |
| 1518 | } |
| 1519 | /* |
| 1520 | * Write as much message data payload as we can. If we finish, queue |
| 1521 | * up the footer. |
| 1522 | * 1 -> done, footer is now queued in out_kvec[]. |
| 1523 | * 0 -> socket full, but more to do |
| 1524 | * <0 -> error |
| 1525 | */ |
| 1526 | static int write_partial_message_data(struct ceph_connection *con) |
| 1527 | { |
| 1528 | struct ceph_msg *msg = con->out_msg; |
| 1529 | struct ceph_msg_data_cursor *cursor = &msg->cursor; |
| 1530 | bool do_datacrc = !con->msgr->nocrc; |
| 1531 | u32 crc; |
| 1532 | |
| 1533 | dout("%s %p msg %p\n", __func__, con, msg); |
| 1534 | |
| 1535 | if (list_empty(&msg->data)) |
| 1536 | return -EINVAL; |
| 1537 | |
| 1538 | /* |
| 1539 | * Iterate through each page that contains data to be |
| 1540 | * written, and send as much as possible for each. |
| 1541 | * |
| 1542 | * If we are calculating the data crc (the default), we will |
| 1543 | * need to map the page. If we have no pages, they have |
| 1544 | * been revoked, so use the zero page. |
| 1545 | */ |
| 1546 | crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0; |
| 1547 | while (cursor->resid) { |
| 1548 | struct page *page; |
| 1549 | size_t page_offset; |
| 1550 | size_t length; |
| 1551 | bool last_piece; |
| 1552 | bool need_crc; |
| 1553 | int ret; |
| 1554 | |
| 1555 | page = ceph_msg_data_next(&msg->cursor, &page_offset, &length, |
| 1556 | &last_piece); |
| 1557 | ret = ceph_tcp_sendpage(con->sock, page, page_offset, |
| 1558 | length, !last_piece); |
| 1559 | if (ret <= 0) { |
| 1560 | if (do_datacrc) |
| 1561 | msg->footer.data_crc = cpu_to_le32(crc); |
| 1562 | |
| 1563 | return ret; |
| 1564 | } |
| 1565 | if (do_datacrc && cursor->need_crc) |
| 1566 | crc = ceph_crc32c_page(crc, page, page_offset, length); |
| 1567 | need_crc = ceph_msg_data_advance(&msg->cursor, (size_t)ret); |
| 1568 | } |
| 1569 | |
| 1570 | dout("%s %p msg %p done\n", __func__, con, msg); |
| 1571 | |
| 1572 | /* prepare and queue up footer, too */ |
| 1573 | if (do_datacrc) |
| 1574 | msg->footer.data_crc = cpu_to_le32(crc); |
| 1575 | else |
| 1576 | msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC; |
| 1577 | con_out_kvec_reset(con); |
| 1578 | prepare_write_message_footer(con); |
| 1579 | |
| 1580 | return 1; /* must return > 0 to indicate success */ |
| 1581 | } |
| 1582 | |
| 1583 | /* |
| 1584 | * write some zeros |
| 1585 | */ |
| 1586 | static int write_partial_skip(struct ceph_connection *con) |
| 1587 | { |
| 1588 | int ret; |
| 1589 | |
| 1590 | while (con->out_skip > 0) { |
| 1591 | size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE); |
| 1592 | |
| 1593 | ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true); |
| 1594 | if (ret <= 0) |
| 1595 | goto out; |
| 1596 | con->out_skip -= ret; |
| 1597 | } |
| 1598 | ret = 1; |
| 1599 | out: |
| 1600 | return ret; |
| 1601 | } |
| 1602 | |
| 1603 | /* |
| 1604 | * Prepare to read connection handshake, or an ack. |
| 1605 | */ |
| 1606 | static void prepare_read_banner(struct ceph_connection *con) |
| 1607 | { |
| 1608 | dout("prepare_read_banner %p\n", con); |
| 1609 | con->in_base_pos = 0; |
| 1610 | } |
| 1611 | |
| 1612 | static void prepare_read_connect(struct ceph_connection *con) |
| 1613 | { |
| 1614 | dout("prepare_read_connect %p\n", con); |
| 1615 | con->in_base_pos = 0; |
| 1616 | } |
| 1617 | |
| 1618 | static void prepare_read_ack(struct ceph_connection *con) |
| 1619 | { |
| 1620 | dout("prepare_read_ack %p\n", con); |
| 1621 | con->in_base_pos = 0; |
| 1622 | } |
| 1623 | |
| 1624 | static void prepare_read_seq(struct ceph_connection *con) |
| 1625 | { |
| 1626 | dout("prepare_read_seq %p\n", con); |
| 1627 | con->in_base_pos = 0; |
| 1628 | con->in_tag = CEPH_MSGR_TAG_SEQ; |
| 1629 | } |
| 1630 | |
| 1631 | static void prepare_read_tag(struct ceph_connection *con) |
| 1632 | { |
| 1633 | dout("prepare_read_tag %p\n", con); |
| 1634 | con->in_base_pos = 0; |
| 1635 | con->in_tag = CEPH_MSGR_TAG_READY; |
| 1636 | } |
| 1637 | |
| 1638 | static void prepare_read_keepalive_ack(struct ceph_connection *con) |
| 1639 | { |
| 1640 | dout("prepare_read_keepalive_ack %p\n", con); |
| 1641 | con->in_base_pos = 0; |
| 1642 | } |
| 1643 | |
| 1644 | /* |
| 1645 | * Prepare to read a message. |
| 1646 | */ |
| 1647 | static int prepare_read_message(struct ceph_connection *con) |
| 1648 | { |
| 1649 | dout("prepare_read_message %p\n", con); |
| 1650 | BUG_ON(con->in_msg != NULL); |
| 1651 | con->in_base_pos = 0; |
| 1652 | con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0; |
| 1653 | return 0; |
| 1654 | } |
| 1655 | |
| 1656 | |
| 1657 | static int read_partial(struct ceph_connection *con, |
| 1658 | int end, int size, void *object) |
| 1659 | { |
| 1660 | while (con->in_base_pos < end) { |
| 1661 | int left = end - con->in_base_pos; |
| 1662 | int have = size - left; |
| 1663 | int ret = ceph_tcp_recvmsg(con->sock, object + have, left); |
| 1664 | if (ret <= 0) |
| 1665 | return ret; |
| 1666 | con->in_base_pos += ret; |
| 1667 | } |
| 1668 | return 1; |
| 1669 | } |
| 1670 | |
| 1671 | |
| 1672 | /* |
| 1673 | * Read all or part of the connect-side handshake on a new connection |
| 1674 | */ |
| 1675 | static int read_partial_banner(struct ceph_connection *con) |
| 1676 | { |
| 1677 | int size; |
| 1678 | int end; |
| 1679 | int ret; |
| 1680 | |
| 1681 | dout("read_partial_banner %p at %d\n", con, con->in_base_pos); |
| 1682 | |
| 1683 | /* peer's banner */ |
| 1684 | size = strlen(CEPH_BANNER); |
| 1685 | end = size; |
| 1686 | ret = read_partial(con, end, size, con->in_banner); |
| 1687 | if (ret <= 0) |
| 1688 | goto out; |
| 1689 | |
| 1690 | size = sizeof (con->actual_peer_addr); |
| 1691 | end += size; |
| 1692 | ret = read_partial(con, end, size, &con->actual_peer_addr); |
| 1693 | if (ret <= 0) |
| 1694 | goto out; |
| 1695 | |
| 1696 | size = sizeof (con->peer_addr_for_me); |
| 1697 | end += size; |
| 1698 | ret = read_partial(con, end, size, &con->peer_addr_for_me); |
| 1699 | if (ret <= 0) |
| 1700 | goto out; |
| 1701 | |
| 1702 | out: |
| 1703 | return ret; |
| 1704 | } |
| 1705 | |
| 1706 | static int read_partial_connect(struct ceph_connection *con) |
| 1707 | { |
| 1708 | int size; |
| 1709 | int end; |
| 1710 | int ret; |
| 1711 | |
| 1712 | dout("read_partial_connect %p at %d\n", con, con->in_base_pos); |
| 1713 | |
| 1714 | size = sizeof (con->in_reply); |
| 1715 | end = size; |
| 1716 | ret = read_partial(con, end, size, &con->in_reply); |
| 1717 | if (ret <= 0) |
| 1718 | goto out; |
| 1719 | |
| 1720 | size = le32_to_cpu(con->in_reply.authorizer_len); |
| 1721 | end += size; |
| 1722 | ret = read_partial(con, end, size, con->auth_reply_buf); |
| 1723 | if (ret <= 0) |
| 1724 | goto out; |
| 1725 | |
| 1726 | dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n", |
| 1727 | con, (int)con->in_reply.tag, |
| 1728 | le32_to_cpu(con->in_reply.connect_seq), |
| 1729 | le32_to_cpu(con->in_reply.global_seq)); |
| 1730 | out: |
| 1731 | return ret; |
| 1732 | |
| 1733 | } |
| 1734 | |
| 1735 | /* |
| 1736 | * Verify the hello banner looks okay. |
| 1737 | */ |
| 1738 | static int verify_hello(struct ceph_connection *con) |
| 1739 | { |
| 1740 | if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) { |
| 1741 | pr_err("connect to %s got bad banner\n", |
| 1742 | ceph_pr_addr(&con->peer_addr.in_addr)); |
| 1743 | con->error_msg = "protocol error, bad banner"; |
| 1744 | return -1; |
| 1745 | } |
| 1746 | return 0; |
| 1747 | } |
| 1748 | |
| 1749 | static bool addr_is_blank(struct sockaddr_storage *ss) |
| 1750 | { |
| 1751 | struct in_addr *addr = &((struct sockaddr_in *)ss)->sin_addr; |
| 1752 | struct in6_addr *addr6 = &((struct sockaddr_in6 *)ss)->sin6_addr; |
| 1753 | |
| 1754 | switch (ss->ss_family) { |
| 1755 | case AF_INET: |
| 1756 | return addr->s_addr == htonl(INADDR_ANY); |
| 1757 | case AF_INET6: |
| 1758 | return ipv6_addr_any(addr6); |
| 1759 | default: |
| 1760 | return true; |
| 1761 | } |
| 1762 | } |
| 1763 | |
| 1764 | static int addr_port(struct sockaddr_storage *ss) |
| 1765 | { |
| 1766 | switch (ss->ss_family) { |
| 1767 | case AF_INET: |
| 1768 | return ntohs(((struct sockaddr_in *)ss)->sin_port); |
| 1769 | case AF_INET6: |
| 1770 | return ntohs(((struct sockaddr_in6 *)ss)->sin6_port); |
| 1771 | } |
| 1772 | return 0; |
| 1773 | } |
| 1774 | |
| 1775 | static void addr_set_port(struct sockaddr_storage *ss, int p) |
| 1776 | { |
| 1777 | switch (ss->ss_family) { |
| 1778 | case AF_INET: |
| 1779 | ((struct sockaddr_in *)ss)->sin_port = htons(p); |
| 1780 | break; |
| 1781 | case AF_INET6: |
| 1782 | ((struct sockaddr_in6 *)ss)->sin6_port = htons(p); |
| 1783 | break; |
| 1784 | } |
| 1785 | } |
| 1786 | |
| 1787 | /* |
| 1788 | * Unlike other *_pton function semantics, zero indicates success. |
| 1789 | */ |
| 1790 | static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss, |
| 1791 | char delim, const char **ipend) |
| 1792 | { |
| 1793 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
| 1794 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; |
| 1795 | |
| 1796 | memset(ss, 0, sizeof(*ss)); |
| 1797 | |
| 1798 | if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) { |
| 1799 | ss->ss_family = AF_INET; |
| 1800 | return 0; |
| 1801 | } |
| 1802 | |
| 1803 | if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) { |
| 1804 | ss->ss_family = AF_INET6; |
| 1805 | return 0; |
| 1806 | } |
| 1807 | |
| 1808 | return -EINVAL; |
| 1809 | } |
| 1810 | |
| 1811 | /* |
| 1812 | * Extract hostname string and resolve using kernel DNS facility. |
| 1813 | */ |
| 1814 | #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER |
| 1815 | static int ceph_dns_resolve_name(const char *name, size_t namelen, |
| 1816 | struct sockaddr_storage *ss, char delim, const char **ipend) |
| 1817 | { |
| 1818 | const char *end, *delim_p; |
| 1819 | char *colon_p, *ip_addr = NULL; |
| 1820 | int ip_len, ret; |
| 1821 | |
| 1822 | /* |
| 1823 | * The end of the hostname occurs immediately preceding the delimiter or |
| 1824 | * the port marker (':') where the delimiter takes precedence. |
| 1825 | */ |
| 1826 | delim_p = memchr(name, delim, namelen); |
| 1827 | colon_p = memchr(name, ':', namelen); |
| 1828 | |
| 1829 | if (delim_p && colon_p) |
| 1830 | end = delim_p < colon_p ? delim_p : colon_p; |
| 1831 | else if (!delim_p && colon_p) |
| 1832 | end = colon_p; |
| 1833 | else { |
| 1834 | end = delim_p; |
| 1835 | if (!end) /* case: hostname:/ */ |
| 1836 | end = name + namelen; |
| 1837 | } |
| 1838 | |
| 1839 | if (end <= name) |
| 1840 | return -EINVAL; |
| 1841 | |
| 1842 | /* do dns_resolve upcall */ |
| 1843 | ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL); |
| 1844 | if (ip_len > 0) |
| 1845 | ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL); |
| 1846 | else |
| 1847 | ret = -ESRCH; |
| 1848 | |
| 1849 | kfree(ip_addr); |
| 1850 | |
| 1851 | *ipend = end; |
| 1852 | |
| 1853 | pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name, |
| 1854 | ret, ret ? "failed" : ceph_pr_addr(ss)); |
| 1855 | |
| 1856 | return ret; |
| 1857 | } |
| 1858 | #else |
| 1859 | static inline int ceph_dns_resolve_name(const char *name, size_t namelen, |
| 1860 | struct sockaddr_storage *ss, char delim, const char **ipend) |
| 1861 | { |
| 1862 | return -EINVAL; |
| 1863 | } |
| 1864 | #endif |
| 1865 | |
| 1866 | /* |
| 1867 | * Parse a server name (IP or hostname). If a valid IP address is not found |
| 1868 | * then try to extract a hostname to resolve using userspace DNS upcall. |
| 1869 | */ |
| 1870 | static int ceph_parse_server_name(const char *name, size_t namelen, |
| 1871 | struct sockaddr_storage *ss, char delim, const char **ipend) |
| 1872 | { |
| 1873 | int ret; |
| 1874 | |
| 1875 | ret = ceph_pton(name, namelen, ss, delim, ipend); |
| 1876 | if (ret) |
| 1877 | ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend); |
| 1878 | |
| 1879 | return ret; |
| 1880 | } |
| 1881 | |
| 1882 | /* |
| 1883 | * Parse an ip[:port] list into an addr array. Use the default |
| 1884 | * monitor port if a port isn't specified. |
| 1885 | */ |
| 1886 | int ceph_parse_ips(const char *c, const char *end, |
| 1887 | struct ceph_entity_addr *addr, |
| 1888 | int max_count, int *count) |
| 1889 | { |
| 1890 | int i, ret = -EINVAL; |
| 1891 | const char *p = c; |
| 1892 | |
| 1893 | dout("parse_ips on '%.*s'\n", (int)(end-c), c); |
| 1894 | for (i = 0; i < max_count; i++) { |
| 1895 | const char *ipend; |
| 1896 | struct sockaddr_storage *ss = &addr[i].in_addr; |
| 1897 | int port; |
| 1898 | char delim = ','; |
| 1899 | |
| 1900 | if (*p == '[') { |
| 1901 | delim = ']'; |
| 1902 | p++; |
| 1903 | } |
| 1904 | |
| 1905 | ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend); |
| 1906 | if (ret) |
| 1907 | goto bad; |
| 1908 | ret = -EINVAL; |
| 1909 | |
| 1910 | p = ipend; |
| 1911 | |
| 1912 | if (delim == ']') { |
| 1913 | if (*p != ']') { |
| 1914 | dout("missing matching ']'\n"); |
| 1915 | goto bad; |
| 1916 | } |
| 1917 | p++; |
| 1918 | } |
| 1919 | |
| 1920 | /* port? */ |
| 1921 | if (p < end && *p == ':') { |
| 1922 | port = 0; |
| 1923 | p++; |
| 1924 | while (p < end && *p >= '0' && *p <= '9') { |
| 1925 | port = (port * 10) + (*p - '0'); |
| 1926 | p++; |
| 1927 | } |
| 1928 | if (port == 0) |
| 1929 | port = CEPH_MON_PORT; |
| 1930 | else if (port > 65535) |
| 1931 | goto bad; |
| 1932 | } else { |
| 1933 | port = CEPH_MON_PORT; |
| 1934 | } |
| 1935 | |
| 1936 | addr_set_port(ss, port); |
| 1937 | |
| 1938 | dout("parse_ips got %s\n", ceph_pr_addr(ss)); |
| 1939 | |
| 1940 | if (p == end) |
| 1941 | break; |
| 1942 | if (*p != ',') |
| 1943 | goto bad; |
| 1944 | p++; |
| 1945 | } |
| 1946 | |
| 1947 | if (p != end) |
| 1948 | goto bad; |
| 1949 | |
| 1950 | if (count) |
| 1951 | *count = i + 1; |
| 1952 | return 0; |
| 1953 | |
| 1954 | bad: |
| 1955 | pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c); |
| 1956 | return ret; |
| 1957 | } |
| 1958 | EXPORT_SYMBOL(ceph_parse_ips); |
| 1959 | |
| 1960 | static int process_banner(struct ceph_connection *con) |
| 1961 | { |
| 1962 | dout("process_banner on %p\n", con); |
| 1963 | |
| 1964 | if (verify_hello(con) < 0) |
| 1965 | return -1; |
| 1966 | |
| 1967 | ceph_decode_addr(&con->actual_peer_addr); |
| 1968 | ceph_decode_addr(&con->peer_addr_for_me); |
| 1969 | |
| 1970 | /* |
| 1971 | * Make sure the other end is who we wanted. note that the other |
| 1972 | * end may not yet know their ip address, so if it's 0.0.0.0, give |
| 1973 | * them the benefit of the doubt. |
| 1974 | */ |
| 1975 | if (memcmp(&con->peer_addr, &con->actual_peer_addr, |
| 1976 | sizeof(con->peer_addr)) != 0 && |
| 1977 | !(addr_is_blank(&con->actual_peer_addr.in_addr) && |
| 1978 | con->actual_peer_addr.nonce == con->peer_addr.nonce)) { |
| 1979 | pr_warn("wrong peer, want %s/%d, got %s/%d\n", |
| 1980 | ceph_pr_addr(&con->peer_addr.in_addr), |
| 1981 | (int)le32_to_cpu(con->peer_addr.nonce), |
| 1982 | ceph_pr_addr(&con->actual_peer_addr.in_addr), |
| 1983 | (int)le32_to_cpu(con->actual_peer_addr.nonce)); |
| 1984 | con->error_msg = "wrong peer at address"; |
| 1985 | return -1; |
| 1986 | } |
| 1987 | |
| 1988 | /* |
| 1989 | * did we learn our address? |
| 1990 | */ |
| 1991 | if (addr_is_blank(&con->msgr->inst.addr.in_addr)) { |
| 1992 | int port = addr_port(&con->msgr->inst.addr.in_addr); |
| 1993 | |
| 1994 | memcpy(&con->msgr->inst.addr.in_addr, |
| 1995 | &con->peer_addr_for_me.in_addr, |
| 1996 | sizeof(con->peer_addr_for_me.in_addr)); |
| 1997 | addr_set_port(&con->msgr->inst.addr.in_addr, port); |
| 1998 | encode_my_addr(con->msgr); |
| 1999 | dout("process_banner learned my addr is %s\n", |
| 2000 | ceph_pr_addr(&con->msgr->inst.addr.in_addr)); |
| 2001 | } |
| 2002 | |
| 2003 | return 0; |
| 2004 | } |
| 2005 | |
| 2006 | static int process_connect(struct ceph_connection *con) |
| 2007 | { |
| 2008 | u64 sup_feat = con->msgr->supported_features; |
| 2009 | u64 req_feat = con->msgr->required_features; |
| 2010 | u64 server_feat = ceph_sanitize_features( |
| 2011 | le64_to_cpu(con->in_reply.features)); |
| 2012 | int ret; |
| 2013 | |
| 2014 | dout("process_connect on %p tag %d\n", con, (int)con->in_tag); |
| 2015 | |
| 2016 | switch (con->in_reply.tag) { |
| 2017 | case CEPH_MSGR_TAG_FEATURES: |
| 2018 | pr_err("%s%lld %s feature set mismatch," |
| 2019 | " my %llx < server's %llx, missing %llx\n", |
| 2020 | ENTITY_NAME(con->peer_name), |
| 2021 | ceph_pr_addr(&con->peer_addr.in_addr), |
| 2022 | sup_feat, server_feat, server_feat & ~sup_feat); |
| 2023 | con->error_msg = "missing required protocol features"; |
| 2024 | reset_connection(con); |
| 2025 | return -1; |
| 2026 | |
| 2027 | case CEPH_MSGR_TAG_BADPROTOVER: |
| 2028 | pr_err("%s%lld %s protocol version mismatch," |
| 2029 | " my %d != server's %d\n", |
| 2030 | ENTITY_NAME(con->peer_name), |
| 2031 | ceph_pr_addr(&con->peer_addr.in_addr), |
| 2032 | le32_to_cpu(con->out_connect.protocol_version), |
| 2033 | le32_to_cpu(con->in_reply.protocol_version)); |
| 2034 | con->error_msg = "protocol version mismatch"; |
| 2035 | reset_connection(con); |
| 2036 | return -1; |
| 2037 | |
| 2038 | case CEPH_MSGR_TAG_BADAUTHORIZER: |
| 2039 | con->auth_retry++; |
| 2040 | dout("process_connect %p got BADAUTHORIZER attempt %d\n", con, |
| 2041 | con->auth_retry); |
| 2042 | if (con->auth_retry == 2) { |
| 2043 | con->error_msg = "connect authorization failure"; |
| 2044 | return -1; |
| 2045 | } |
| 2046 | con_out_kvec_reset(con); |
| 2047 | ret = prepare_write_connect(con); |
| 2048 | if (ret < 0) |
| 2049 | return ret; |
| 2050 | prepare_read_connect(con); |
| 2051 | break; |
| 2052 | |
| 2053 | case CEPH_MSGR_TAG_RESETSESSION: |
| 2054 | /* |
| 2055 | * If we connected with a large connect_seq but the peer |
| 2056 | * has no record of a session with us (no connection, or |
| 2057 | * connect_seq == 0), they will send RESETSESION to indicate |
| 2058 | * that they must have reset their session, and may have |
| 2059 | * dropped messages. |
| 2060 | */ |
| 2061 | dout("process_connect got RESET peer seq %u\n", |
| 2062 | le32_to_cpu(con->in_reply.connect_seq)); |
| 2063 | pr_err("%s%lld %s connection reset\n", |
| 2064 | ENTITY_NAME(con->peer_name), |
| 2065 | ceph_pr_addr(&con->peer_addr.in_addr)); |
| 2066 | reset_connection(con); |
| 2067 | con_out_kvec_reset(con); |
| 2068 | ret = prepare_write_connect(con); |
| 2069 | if (ret < 0) |
| 2070 | return ret; |
| 2071 | prepare_read_connect(con); |
| 2072 | |
| 2073 | /* Tell ceph about it. */ |
| 2074 | mutex_unlock(&con->mutex); |
| 2075 | pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name)); |
| 2076 | if (con->ops->peer_reset) |
| 2077 | con->ops->peer_reset(con); |
| 2078 | mutex_lock(&con->mutex); |
| 2079 | if (con->state != CON_STATE_NEGOTIATING) |
| 2080 | return -EAGAIN; |
| 2081 | break; |
| 2082 | |
| 2083 | case CEPH_MSGR_TAG_RETRY_SESSION: |
| 2084 | /* |
| 2085 | * If we sent a smaller connect_seq than the peer has, try |
| 2086 | * again with a larger value. |
| 2087 | */ |
| 2088 | dout("process_connect got RETRY_SESSION my seq %u, peer %u\n", |
| 2089 | le32_to_cpu(con->out_connect.connect_seq), |
| 2090 | le32_to_cpu(con->in_reply.connect_seq)); |
| 2091 | con->connect_seq = le32_to_cpu(con->in_reply.connect_seq); |
| 2092 | con_out_kvec_reset(con); |
| 2093 | ret = prepare_write_connect(con); |
| 2094 | if (ret < 0) |
| 2095 | return ret; |
| 2096 | prepare_read_connect(con); |
| 2097 | break; |
| 2098 | |
| 2099 | case CEPH_MSGR_TAG_RETRY_GLOBAL: |
| 2100 | /* |
| 2101 | * If we sent a smaller global_seq than the peer has, try |
| 2102 | * again with a larger value. |
| 2103 | */ |
| 2104 | dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n", |
| 2105 | con->peer_global_seq, |
| 2106 | le32_to_cpu(con->in_reply.global_seq)); |
| 2107 | get_global_seq(con->msgr, |
| 2108 | le32_to_cpu(con->in_reply.global_seq)); |
| 2109 | con_out_kvec_reset(con); |
| 2110 | ret = prepare_write_connect(con); |
| 2111 | if (ret < 0) |
| 2112 | return ret; |
| 2113 | prepare_read_connect(con); |
| 2114 | break; |
| 2115 | |
| 2116 | case CEPH_MSGR_TAG_SEQ: |
| 2117 | case CEPH_MSGR_TAG_READY: |
| 2118 | if (req_feat & ~server_feat) { |
| 2119 | pr_err("%s%lld %s protocol feature mismatch," |
| 2120 | " my required %llx > server's %llx, need %llx\n", |
| 2121 | ENTITY_NAME(con->peer_name), |
| 2122 | ceph_pr_addr(&con->peer_addr.in_addr), |
| 2123 | req_feat, server_feat, req_feat & ~server_feat); |
| 2124 | con->error_msg = "missing required protocol features"; |
| 2125 | reset_connection(con); |
| 2126 | return -1; |
| 2127 | } |
| 2128 | |
| 2129 | WARN_ON(con->state != CON_STATE_NEGOTIATING); |
| 2130 | con->state = CON_STATE_OPEN; |
| 2131 | con->auth_retry = 0; /* we authenticated; clear flag */ |
| 2132 | con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq); |
| 2133 | con->connect_seq++; |
| 2134 | con->peer_features = server_feat; |
| 2135 | dout("process_connect got READY gseq %d cseq %d (%d)\n", |
| 2136 | con->peer_global_seq, |
| 2137 | le32_to_cpu(con->in_reply.connect_seq), |
| 2138 | con->connect_seq); |
| 2139 | WARN_ON(con->connect_seq != |
| 2140 | le32_to_cpu(con->in_reply.connect_seq)); |
| 2141 | |
| 2142 | if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY) |
| 2143 | con_flag_set(con, CON_FLAG_LOSSYTX); |
| 2144 | |
| 2145 | con->delay = 0; /* reset backoff memory */ |
| 2146 | |
| 2147 | if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) { |
| 2148 | prepare_write_seq(con); |
| 2149 | prepare_read_seq(con); |
| 2150 | } else { |
| 2151 | prepare_read_tag(con); |
| 2152 | } |
| 2153 | break; |
| 2154 | |
| 2155 | case CEPH_MSGR_TAG_WAIT: |
| 2156 | /* |
| 2157 | * If there is a connection race (we are opening |
| 2158 | * connections to each other), one of us may just have |
| 2159 | * to WAIT. This shouldn't happen if we are the |
| 2160 | * client. |
| 2161 | */ |
| 2162 | con->error_msg = "protocol error, got WAIT as client"; |
| 2163 | return -1; |
| 2164 | |
| 2165 | default: |
| 2166 | con->error_msg = "protocol error, garbage tag during connect"; |
| 2167 | return -1; |
| 2168 | } |
| 2169 | return 0; |
| 2170 | } |
| 2171 | |
| 2172 | |
| 2173 | /* |
| 2174 | * read (part of) an ack |
| 2175 | */ |
| 2176 | static int read_partial_ack(struct ceph_connection *con) |
| 2177 | { |
| 2178 | int size = sizeof (con->in_temp_ack); |
| 2179 | int end = size; |
| 2180 | |
| 2181 | return read_partial(con, end, size, &con->in_temp_ack); |
| 2182 | } |
| 2183 | |
| 2184 | /* |
| 2185 | * We can finally discard anything that's been acked. |
| 2186 | */ |
| 2187 | static void process_ack(struct ceph_connection *con) |
| 2188 | { |
| 2189 | struct ceph_msg *m; |
| 2190 | u64 ack = le64_to_cpu(con->in_temp_ack); |
| 2191 | u64 seq; |
| 2192 | |
| 2193 | while (!list_empty(&con->out_sent)) { |
| 2194 | m = list_first_entry(&con->out_sent, struct ceph_msg, |
| 2195 | list_head); |
| 2196 | seq = le64_to_cpu(m->hdr.seq); |
| 2197 | if (seq > ack) |
| 2198 | break; |
| 2199 | dout("got ack for seq %llu type %d at %p\n", seq, |
| 2200 | le16_to_cpu(m->hdr.type), m); |
| 2201 | m->ack_stamp = jiffies; |
| 2202 | ceph_msg_remove(m); |
| 2203 | } |
| 2204 | prepare_read_tag(con); |
| 2205 | } |
| 2206 | |
| 2207 | |
| 2208 | static int read_partial_message_section(struct ceph_connection *con, |
| 2209 | struct kvec *section, |
| 2210 | unsigned int sec_len, u32 *crc) |
| 2211 | { |
| 2212 | int ret, left; |
| 2213 | |
| 2214 | BUG_ON(!section); |
| 2215 | |
| 2216 | while (section->iov_len < sec_len) { |
| 2217 | BUG_ON(section->iov_base == NULL); |
| 2218 | left = sec_len - section->iov_len; |
| 2219 | ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base + |
| 2220 | section->iov_len, left); |
| 2221 | if (ret <= 0) |
| 2222 | return ret; |
| 2223 | section->iov_len += ret; |
| 2224 | } |
| 2225 | if (section->iov_len == sec_len) |
| 2226 | *crc = crc32c(0, section->iov_base, section->iov_len); |
| 2227 | |
| 2228 | return 1; |
| 2229 | } |
| 2230 | |
| 2231 | static int read_partial_msg_data(struct ceph_connection *con) |
| 2232 | { |
| 2233 | struct ceph_msg *msg = con->in_msg; |
| 2234 | struct ceph_msg_data_cursor *cursor = &msg->cursor; |
| 2235 | const bool do_datacrc = !con->msgr->nocrc; |
| 2236 | struct page *page; |
| 2237 | size_t page_offset; |
| 2238 | size_t length; |
| 2239 | u32 crc = 0; |
| 2240 | int ret; |
| 2241 | |
| 2242 | BUG_ON(!msg); |
| 2243 | if (list_empty(&msg->data)) |
| 2244 | return -EIO; |
| 2245 | |
| 2246 | if (do_datacrc) |
| 2247 | crc = con->in_data_crc; |
| 2248 | while (cursor->resid) { |
| 2249 | page = ceph_msg_data_next(&msg->cursor, &page_offset, &length, |
| 2250 | NULL); |
| 2251 | ret = ceph_tcp_recvpage(con->sock, page, page_offset, length); |
| 2252 | if (ret <= 0) { |
| 2253 | if (do_datacrc) |
| 2254 | con->in_data_crc = crc; |
| 2255 | |
| 2256 | return ret; |
| 2257 | } |
| 2258 | |
| 2259 | if (do_datacrc) |
| 2260 | crc = ceph_crc32c_page(crc, page, page_offset, ret); |
| 2261 | (void) ceph_msg_data_advance(&msg->cursor, (size_t)ret); |
| 2262 | } |
| 2263 | if (do_datacrc) |
| 2264 | con->in_data_crc = crc; |
| 2265 | |
| 2266 | return 1; /* must return > 0 to indicate success */ |
| 2267 | } |
| 2268 | |
| 2269 | /* |
| 2270 | * read (part of) a message. |
| 2271 | */ |
| 2272 | static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip); |
| 2273 | |
| 2274 | static int read_partial_message(struct ceph_connection *con) |
| 2275 | { |
| 2276 | struct ceph_msg *m = con->in_msg; |
| 2277 | int size; |
| 2278 | int end; |
| 2279 | int ret; |
| 2280 | unsigned int front_len, middle_len, data_len; |
| 2281 | bool do_datacrc = !con->msgr->nocrc; |
| 2282 | bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH); |
| 2283 | u64 seq; |
| 2284 | u32 crc; |
| 2285 | |
| 2286 | dout("read_partial_message con %p msg %p\n", con, m); |
| 2287 | |
| 2288 | /* header */ |
| 2289 | size = sizeof (con->in_hdr); |
| 2290 | end = size; |
| 2291 | ret = read_partial(con, end, size, &con->in_hdr); |
| 2292 | if (ret <= 0) |
| 2293 | return ret; |
| 2294 | |
| 2295 | crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc)); |
| 2296 | if (cpu_to_le32(crc) != con->in_hdr.crc) { |
| 2297 | pr_err("read_partial_message bad hdr crc %u != expected %u\n", |
| 2298 | crc, con->in_hdr.crc); |
| 2299 | return -EBADMSG; |
| 2300 | } |
| 2301 | |
| 2302 | front_len = le32_to_cpu(con->in_hdr.front_len); |
| 2303 | if (front_len > CEPH_MSG_MAX_FRONT_LEN) |
| 2304 | return -EIO; |
| 2305 | middle_len = le32_to_cpu(con->in_hdr.middle_len); |
| 2306 | if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN) |
| 2307 | return -EIO; |
| 2308 | data_len = le32_to_cpu(con->in_hdr.data_len); |
| 2309 | if (data_len > CEPH_MSG_MAX_DATA_LEN) |
| 2310 | return -EIO; |
| 2311 | |
| 2312 | /* verify seq# */ |
| 2313 | seq = le64_to_cpu(con->in_hdr.seq); |
| 2314 | if ((s64)seq - (s64)con->in_seq < 1) { |
| 2315 | pr_info("skipping %s%lld %s seq %lld expected %lld\n", |
| 2316 | ENTITY_NAME(con->peer_name), |
| 2317 | ceph_pr_addr(&con->peer_addr.in_addr), |
| 2318 | seq, con->in_seq + 1); |
| 2319 | con->in_base_pos = -front_len - middle_len - data_len - |
| 2320 | sizeof(m->footer); |
| 2321 | con->in_tag = CEPH_MSGR_TAG_READY; |
| 2322 | return 0; |
| 2323 | } else if ((s64)seq - (s64)con->in_seq > 1) { |
| 2324 | pr_err("read_partial_message bad seq %lld expected %lld\n", |
| 2325 | seq, con->in_seq + 1); |
| 2326 | con->error_msg = "bad message sequence # for incoming message"; |
| 2327 | return -EBADE; |
| 2328 | } |
| 2329 | |
| 2330 | /* allocate message? */ |
| 2331 | if (!con->in_msg) { |
| 2332 | int skip = 0; |
| 2333 | |
| 2334 | dout("got hdr type %d front %d data %d\n", con->in_hdr.type, |
| 2335 | front_len, data_len); |
| 2336 | ret = ceph_con_in_msg_alloc(con, &skip); |
| 2337 | if (ret < 0) |
| 2338 | return ret; |
| 2339 | |
| 2340 | BUG_ON(!con->in_msg ^ skip); |
| 2341 | if (skip) { |
| 2342 | /* skip this message */ |
| 2343 | dout("alloc_msg said skip message\n"); |
| 2344 | con->in_base_pos = -front_len - middle_len - data_len - |
| 2345 | sizeof(m->footer); |
| 2346 | con->in_tag = CEPH_MSGR_TAG_READY; |
| 2347 | con->in_seq++; |
| 2348 | return 0; |
| 2349 | } |
| 2350 | |
| 2351 | BUG_ON(!con->in_msg); |
| 2352 | BUG_ON(con->in_msg->con != con); |
| 2353 | m = con->in_msg; |
| 2354 | m->front.iov_len = 0; /* haven't read it yet */ |
| 2355 | if (m->middle) |
| 2356 | m->middle->vec.iov_len = 0; |
| 2357 | |
| 2358 | /* prepare for data payload, if any */ |
| 2359 | |
| 2360 | if (data_len) |
| 2361 | prepare_message_data(con->in_msg, data_len); |
| 2362 | } |
| 2363 | |
| 2364 | /* front */ |
| 2365 | ret = read_partial_message_section(con, &m->front, front_len, |
| 2366 | &con->in_front_crc); |
| 2367 | if (ret <= 0) |
| 2368 | return ret; |
| 2369 | |
| 2370 | /* middle */ |
| 2371 | if (m->middle) { |
| 2372 | ret = read_partial_message_section(con, &m->middle->vec, |
| 2373 | middle_len, |
| 2374 | &con->in_middle_crc); |
| 2375 | if (ret <= 0) |
| 2376 | return ret; |
| 2377 | } |
| 2378 | |
| 2379 | /* (page) data */ |
| 2380 | if (data_len) { |
| 2381 | ret = read_partial_msg_data(con); |
| 2382 | if (ret <= 0) |
| 2383 | return ret; |
| 2384 | } |
| 2385 | |
| 2386 | /* footer */ |
| 2387 | if (need_sign) |
| 2388 | size = sizeof(m->footer); |
| 2389 | else |
| 2390 | size = sizeof(m->old_footer); |
| 2391 | |
| 2392 | end += size; |
| 2393 | ret = read_partial(con, end, size, &m->footer); |
| 2394 | if (ret <= 0) |
| 2395 | return ret; |
| 2396 | |
| 2397 | if (!need_sign) { |
| 2398 | m->footer.flags = m->old_footer.flags; |
| 2399 | m->footer.sig = 0; |
| 2400 | } |
| 2401 | |
| 2402 | dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n", |
| 2403 | m, front_len, m->footer.front_crc, middle_len, |
| 2404 | m->footer.middle_crc, data_len, m->footer.data_crc); |
| 2405 | |
| 2406 | /* crc ok? */ |
| 2407 | if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) { |
| 2408 | pr_err("read_partial_message %p front crc %u != exp. %u\n", |
| 2409 | m, con->in_front_crc, m->footer.front_crc); |
| 2410 | return -EBADMSG; |
| 2411 | } |
| 2412 | if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) { |
| 2413 | pr_err("read_partial_message %p middle crc %u != exp %u\n", |
| 2414 | m, con->in_middle_crc, m->footer.middle_crc); |
| 2415 | return -EBADMSG; |
| 2416 | } |
| 2417 | if (do_datacrc && |
| 2418 | (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 && |
| 2419 | con->in_data_crc != le32_to_cpu(m->footer.data_crc)) { |
| 2420 | pr_err("read_partial_message %p data crc %u != exp. %u\n", m, |
| 2421 | con->in_data_crc, le32_to_cpu(m->footer.data_crc)); |
| 2422 | return -EBADMSG; |
| 2423 | } |
| 2424 | |
| 2425 | if (need_sign && con->ops->check_message_signature && |
| 2426 | con->ops->check_message_signature(con, m)) { |
| 2427 | pr_err("read_partial_message %p signature check failed\n", m); |
| 2428 | return -EBADMSG; |
| 2429 | } |
| 2430 | |
| 2431 | return 1; /* done! */ |
| 2432 | } |
| 2433 | |
| 2434 | /* |
| 2435 | * Process message. This happens in the worker thread. The callback should |
| 2436 | * be careful not to do anything that waits on other incoming messages or it |
| 2437 | * may deadlock. |
| 2438 | */ |
| 2439 | static void process_message(struct ceph_connection *con) |
| 2440 | { |
| 2441 | struct ceph_msg *msg; |
| 2442 | |
| 2443 | BUG_ON(con->in_msg->con != con); |
| 2444 | con->in_msg->con = NULL; |
| 2445 | msg = con->in_msg; |
| 2446 | con->in_msg = NULL; |
| 2447 | con->ops->put(con); |
| 2448 | |
| 2449 | /* if first message, set peer_name */ |
| 2450 | if (con->peer_name.type == 0) |
| 2451 | con->peer_name = msg->hdr.src; |
| 2452 | |
| 2453 | con->in_seq++; |
| 2454 | mutex_unlock(&con->mutex); |
| 2455 | |
| 2456 | dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n", |
| 2457 | msg, le64_to_cpu(msg->hdr.seq), |
| 2458 | ENTITY_NAME(msg->hdr.src), |
| 2459 | le16_to_cpu(msg->hdr.type), |
| 2460 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), |
| 2461 | le32_to_cpu(msg->hdr.front_len), |
| 2462 | le32_to_cpu(msg->hdr.data_len), |
| 2463 | con->in_front_crc, con->in_middle_crc, con->in_data_crc); |
| 2464 | con->ops->dispatch(con, msg); |
| 2465 | |
| 2466 | mutex_lock(&con->mutex); |
| 2467 | } |
| 2468 | |
| 2469 | static int read_keepalive_ack(struct ceph_connection *con) |
| 2470 | { |
| 2471 | struct ceph_timespec ceph_ts; |
| 2472 | size_t size = sizeof(ceph_ts); |
| 2473 | int ret = read_partial(con, size, size, &ceph_ts); |
| 2474 | if (ret <= 0) |
| 2475 | return ret; |
| 2476 | ceph_decode_timespec(&con->last_keepalive_ack, &ceph_ts); |
| 2477 | prepare_read_tag(con); |
| 2478 | return 1; |
| 2479 | } |
| 2480 | |
| 2481 | /* |
| 2482 | * Write something to the socket. Called in a worker thread when the |
| 2483 | * socket appears to be writeable and we have something ready to send. |
| 2484 | */ |
| 2485 | static int try_write(struct ceph_connection *con) |
| 2486 | { |
| 2487 | int ret = 1; |
| 2488 | |
| 2489 | dout("try_write start %p state %lu\n", con, con->state); |
| 2490 | |
| 2491 | more: |
| 2492 | dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes); |
| 2493 | |
| 2494 | /* open the socket first? */ |
| 2495 | if (con->state == CON_STATE_PREOPEN) { |
| 2496 | BUG_ON(con->sock); |
| 2497 | con->state = CON_STATE_CONNECTING; |
| 2498 | |
| 2499 | con_out_kvec_reset(con); |
| 2500 | prepare_write_banner(con); |
| 2501 | prepare_read_banner(con); |
| 2502 | |
| 2503 | BUG_ON(con->in_msg); |
| 2504 | con->in_tag = CEPH_MSGR_TAG_READY; |
| 2505 | dout("try_write initiating connect on %p new state %lu\n", |
| 2506 | con, con->state); |
| 2507 | ret = ceph_tcp_connect(con); |
| 2508 | if (ret < 0) { |
| 2509 | con->error_msg = "connect error"; |
| 2510 | goto out; |
| 2511 | } |
| 2512 | } |
| 2513 | |
| 2514 | more_kvec: |
| 2515 | /* kvec data queued? */ |
| 2516 | if (con->out_skip) { |
| 2517 | ret = write_partial_skip(con); |
| 2518 | if (ret <= 0) |
| 2519 | goto out; |
| 2520 | } |
| 2521 | if (con->out_kvec_left) { |
| 2522 | ret = write_partial_kvec(con); |
| 2523 | if (ret <= 0) |
| 2524 | goto out; |
| 2525 | } |
| 2526 | |
| 2527 | /* msg pages? */ |
| 2528 | if (con->out_msg) { |
| 2529 | if (con->out_msg_done) { |
| 2530 | ceph_msg_put(con->out_msg); |
| 2531 | con->out_msg = NULL; /* we're done with this one */ |
| 2532 | goto do_next; |
| 2533 | } |
| 2534 | |
| 2535 | ret = write_partial_message_data(con); |
| 2536 | if (ret == 1) |
| 2537 | goto more_kvec; /* we need to send the footer, too! */ |
| 2538 | if (ret == 0) |
| 2539 | goto out; |
| 2540 | if (ret < 0) { |
| 2541 | dout("try_write write_partial_message_data err %d\n", |
| 2542 | ret); |
| 2543 | goto out; |
| 2544 | } |
| 2545 | } |
| 2546 | |
| 2547 | do_next: |
| 2548 | if (con->state == CON_STATE_OPEN) { |
| 2549 | if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) { |
| 2550 | prepare_write_keepalive(con); |
| 2551 | goto more; |
| 2552 | } |
| 2553 | /* is anything else pending? */ |
| 2554 | if (!list_empty(&con->out_queue)) { |
| 2555 | prepare_write_message(con); |
| 2556 | goto more; |
| 2557 | } |
| 2558 | if (con->in_seq > con->in_seq_acked) { |
| 2559 | prepare_write_ack(con); |
| 2560 | goto more; |
| 2561 | } |
| 2562 | } |
| 2563 | |
| 2564 | /* Nothing to do! */ |
| 2565 | con_flag_clear(con, CON_FLAG_WRITE_PENDING); |
| 2566 | dout("try_write nothing else to write.\n"); |
| 2567 | ret = 0; |
| 2568 | out: |
| 2569 | dout("try_write done on %p ret %d\n", con, ret); |
| 2570 | return ret; |
| 2571 | } |
| 2572 | |
| 2573 | |
| 2574 | |
| 2575 | /* |
| 2576 | * Read what we can from the socket. |
| 2577 | */ |
| 2578 | static int try_read(struct ceph_connection *con) |
| 2579 | { |
| 2580 | int ret = -1; |
| 2581 | |
| 2582 | more: |
| 2583 | dout("try_read start on %p state %lu\n", con, con->state); |
| 2584 | if (con->state != CON_STATE_CONNECTING && |
| 2585 | con->state != CON_STATE_NEGOTIATING && |
| 2586 | con->state != CON_STATE_OPEN) |
| 2587 | return 0; |
| 2588 | |
| 2589 | BUG_ON(!con->sock); |
| 2590 | |
| 2591 | dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag, |
| 2592 | con->in_base_pos); |
| 2593 | |
| 2594 | if (con->state == CON_STATE_CONNECTING) { |
| 2595 | dout("try_read connecting\n"); |
| 2596 | ret = read_partial_banner(con); |
| 2597 | if (ret <= 0) |
| 2598 | goto out; |
| 2599 | ret = process_banner(con); |
| 2600 | if (ret < 0) |
| 2601 | goto out; |
| 2602 | |
| 2603 | con->state = CON_STATE_NEGOTIATING; |
| 2604 | |
| 2605 | /* |
| 2606 | * Received banner is good, exchange connection info. |
| 2607 | * Do not reset out_kvec, as sending our banner raced |
| 2608 | * with receiving peer banner after connect completed. |
| 2609 | */ |
| 2610 | ret = prepare_write_connect(con); |
| 2611 | if (ret < 0) |
| 2612 | goto out; |
| 2613 | prepare_read_connect(con); |
| 2614 | |
| 2615 | /* Send connection info before awaiting response */ |
| 2616 | goto out; |
| 2617 | } |
| 2618 | |
| 2619 | if (con->state == CON_STATE_NEGOTIATING) { |
| 2620 | dout("try_read negotiating\n"); |
| 2621 | ret = read_partial_connect(con); |
| 2622 | if (ret <= 0) |
| 2623 | goto out; |
| 2624 | ret = process_connect(con); |
| 2625 | if (ret < 0) |
| 2626 | goto out; |
| 2627 | goto more; |
| 2628 | } |
| 2629 | |
| 2630 | WARN_ON(con->state != CON_STATE_OPEN); |
| 2631 | |
| 2632 | if (con->in_base_pos < 0) { |
| 2633 | /* |
| 2634 | * skipping + discarding content. |
| 2635 | * |
| 2636 | * FIXME: there must be a better way to do this! |
| 2637 | */ |
| 2638 | static char buf[SKIP_BUF_SIZE]; |
| 2639 | int skip = min((int) sizeof (buf), -con->in_base_pos); |
| 2640 | |
| 2641 | dout("skipping %d / %d bytes\n", skip, -con->in_base_pos); |
| 2642 | ret = ceph_tcp_recvmsg(con->sock, buf, skip); |
| 2643 | if (ret <= 0) |
| 2644 | goto out; |
| 2645 | con->in_base_pos += ret; |
| 2646 | if (con->in_base_pos) |
| 2647 | goto more; |
| 2648 | } |
| 2649 | if (con->in_tag == CEPH_MSGR_TAG_READY) { |
| 2650 | /* |
| 2651 | * what's next? |
| 2652 | */ |
| 2653 | ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1); |
| 2654 | if (ret <= 0) |
| 2655 | goto out; |
| 2656 | dout("try_read got tag %d\n", (int)con->in_tag); |
| 2657 | switch (con->in_tag) { |
| 2658 | case CEPH_MSGR_TAG_MSG: |
| 2659 | prepare_read_message(con); |
| 2660 | break; |
| 2661 | case CEPH_MSGR_TAG_ACK: |
| 2662 | prepare_read_ack(con); |
| 2663 | break; |
| 2664 | case CEPH_MSGR_TAG_KEEPALIVE2_ACK: |
| 2665 | prepare_read_keepalive_ack(con); |
| 2666 | break; |
| 2667 | case CEPH_MSGR_TAG_CLOSE: |
| 2668 | con_close_socket(con); |
| 2669 | con->state = CON_STATE_CLOSED; |
| 2670 | goto out; |
| 2671 | default: |
| 2672 | goto bad_tag; |
| 2673 | } |
| 2674 | } |
| 2675 | if (con->in_tag == CEPH_MSGR_TAG_MSG) { |
| 2676 | ret = read_partial_message(con); |
| 2677 | if (ret <= 0) { |
| 2678 | switch (ret) { |
| 2679 | case -EBADMSG: |
| 2680 | con->error_msg = "bad crc"; |
| 2681 | /* fall through */ |
| 2682 | case -EBADE: |
| 2683 | ret = -EIO; |
| 2684 | break; |
| 2685 | case -EIO: |
| 2686 | con->error_msg = "io error"; |
| 2687 | break; |
| 2688 | } |
| 2689 | goto out; |
| 2690 | } |
| 2691 | if (con->in_tag == CEPH_MSGR_TAG_READY) |
| 2692 | goto more; |
| 2693 | process_message(con); |
| 2694 | if (con->state == CON_STATE_OPEN) |
| 2695 | prepare_read_tag(con); |
| 2696 | goto more; |
| 2697 | } |
| 2698 | if (con->in_tag == CEPH_MSGR_TAG_ACK || |
| 2699 | con->in_tag == CEPH_MSGR_TAG_SEQ) { |
| 2700 | /* |
| 2701 | * the final handshake seq exchange is semantically |
| 2702 | * equivalent to an ACK |
| 2703 | */ |
| 2704 | ret = read_partial_ack(con); |
| 2705 | if (ret <= 0) |
| 2706 | goto out; |
| 2707 | process_ack(con); |
| 2708 | goto more; |
| 2709 | } |
| 2710 | if (con->in_tag == CEPH_MSGR_TAG_KEEPALIVE2_ACK) { |
| 2711 | ret = read_keepalive_ack(con); |
| 2712 | if (ret <= 0) |
| 2713 | goto out; |
| 2714 | goto more; |
| 2715 | } |
| 2716 | |
| 2717 | out: |
| 2718 | dout("try_read done on %p ret %d\n", con, ret); |
| 2719 | return ret; |
| 2720 | |
| 2721 | bad_tag: |
| 2722 | pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag); |
| 2723 | con->error_msg = "protocol error, garbage tag"; |
| 2724 | ret = -1; |
| 2725 | goto out; |
| 2726 | } |
| 2727 | |
| 2728 | |
| 2729 | /* |
| 2730 | * Atomically queue work on a connection after the specified delay. |
| 2731 | * Bump @con reference to avoid races with connection teardown. |
| 2732 | * Returns 0 if work was queued, or an error code otherwise. |
| 2733 | */ |
| 2734 | static int queue_con_delay(struct ceph_connection *con, unsigned long delay) |
| 2735 | { |
| 2736 | if (!con->ops->get(con)) { |
| 2737 | dout("%s %p ref count 0\n", __func__, con); |
| 2738 | return -ENOENT; |
| 2739 | } |
| 2740 | |
| 2741 | if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) { |
| 2742 | dout("%s %p - already queued\n", __func__, con); |
| 2743 | con->ops->put(con); |
| 2744 | return -EBUSY; |
| 2745 | } |
| 2746 | |
| 2747 | dout("%s %p %lu\n", __func__, con, delay); |
| 2748 | return 0; |
| 2749 | } |
| 2750 | |
| 2751 | static void queue_con(struct ceph_connection *con) |
| 2752 | { |
| 2753 | (void) queue_con_delay(con, 0); |
| 2754 | } |
| 2755 | |
| 2756 | static void cancel_con(struct ceph_connection *con) |
| 2757 | { |
| 2758 | if (cancel_delayed_work(&con->work)) { |
| 2759 | dout("%s %p\n", __func__, con); |
| 2760 | con->ops->put(con); |
| 2761 | } |
| 2762 | } |
| 2763 | |
| 2764 | static bool con_sock_closed(struct ceph_connection *con) |
| 2765 | { |
| 2766 | if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED)) |
| 2767 | return false; |
| 2768 | |
| 2769 | #define CASE(x) \ |
| 2770 | case CON_STATE_ ## x: \ |
| 2771 | con->error_msg = "socket closed (con state " #x ")"; \ |
| 2772 | break; |
| 2773 | |
| 2774 | switch (con->state) { |
| 2775 | CASE(CLOSED); |
| 2776 | CASE(PREOPEN); |
| 2777 | CASE(CONNECTING); |
| 2778 | CASE(NEGOTIATING); |
| 2779 | CASE(OPEN); |
| 2780 | CASE(STANDBY); |
| 2781 | default: |
| 2782 | pr_warn("%s con %p unrecognized state %lu\n", |
| 2783 | __func__, con, con->state); |
| 2784 | con->error_msg = "unrecognized con state"; |
| 2785 | BUG(); |
| 2786 | break; |
| 2787 | } |
| 2788 | #undef CASE |
| 2789 | |
| 2790 | return true; |
| 2791 | } |
| 2792 | |
| 2793 | static bool con_backoff(struct ceph_connection *con) |
| 2794 | { |
| 2795 | int ret; |
| 2796 | |
| 2797 | if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF)) |
| 2798 | return false; |
| 2799 | |
| 2800 | ret = queue_con_delay(con, round_jiffies_relative(con->delay)); |
| 2801 | if (ret) { |
| 2802 | dout("%s: con %p FAILED to back off %lu\n", __func__, |
| 2803 | con, con->delay); |
| 2804 | BUG_ON(ret == -ENOENT); |
| 2805 | con_flag_set(con, CON_FLAG_BACKOFF); |
| 2806 | } |
| 2807 | |
| 2808 | return true; |
| 2809 | } |
| 2810 | |
| 2811 | /* Finish fault handling; con->mutex must *not* be held here */ |
| 2812 | |
| 2813 | static void con_fault_finish(struct ceph_connection *con) |
| 2814 | { |
| 2815 | /* |
| 2816 | * in case we faulted due to authentication, invalidate our |
| 2817 | * current tickets so that we can get new ones. |
| 2818 | */ |
| 2819 | if (con->auth_retry && con->ops->invalidate_authorizer) { |
| 2820 | dout("calling invalidate_authorizer()\n"); |
| 2821 | con->ops->invalidate_authorizer(con); |
| 2822 | } |
| 2823 | |
| 2824 | if (con->ops->fault) |
| 2825 | con->ops->fault(con); |
| 2826 | } |
| 2827 | |
| 2828 | /* |
| 2829 | * Do some work on a connection. Drop a connection ref when we're done. |
| 2830 | */ |
| 2831 | static void ceph_con_workfn(struct work_struct *work) |
| 2832 | { |
| 2833 | struct ceph_connection *con = container_of(work, struct ceph_connection, |
| 2834 | work.work); |
| 2835 | bool fault; |
| 2836 | |
| 2837 | mutex_lock(&con->mutex); |
| 2838 | while (true) { |
| 2839 | int ret; |
| 2840 | |
| 2841 | if ((fault = con_sock_closed(con))) { |
| 2842 | dout("%s: con %p SOCK_CLOSED\n", __func__, con); |
| 2843 | break; |
| 2844 | } |
| 2845 | if (con_backoff(con)) { |
| 2846 | dout("%s: con %p BACKOFF\n", __func__, con); |
| 2847 | break; |
| 2848 | } |
| 2849 | if (con->state == CON_STATE_STANDBY) { |
| 2850 | dout("%s: con %p STANDBY\n", __func__, con); |
| 2851 | break; |
| 2852 | } |
| 2853 | if (con->state == CON_STATE_CLOSED) { |
| 2854 | dout("%s: con %p CLOSED\n", __func__, con); |
| 2855 | BUG_ON(con->sock); |
| 2856 | break; |
| 2857 | } |
| 2858 | if (con->state == CON_STATE_PREOPEN) { |
| 2859 | dout("%s: con %p PREOPEN\n", __func__, con); |
| 2860 | BUG_ON(con->sock); |
| 2861 | } |
| 2862 | |
| 2863 | ret = try_read(con); |
| 2864 | if (ret < 0) { |
| 2865 | if (ret == -EAGAIN) |
| 2866 | continue; |
| 2867 | if (!con->error_msg) |
| 2868 | con->error_msg = "socket error on read"; |
| 2869 | fault = true; |
| 2870 | break; |
| 2871 | } |
| 2872 | |
| 2873 | ret = try_write(con); |
| 2874 | if (ret < 0) { |
| 2875 | if (ret == -EAGAIN) |
| 2876 | continue; |
| 2877 | if (!con->error_msg) |
| 2878 | con->error_msg = "socket error on write"; |
| 2879 | fault = true; |
| 2880 | } |
| 2881 | |
| 2882 | break; /* If we make it to here, we're done */ |
| 2883 | } |
| 2884 | if (fault) |
| 2885 | con_fault(con); |
| 2886 | mutex_unlock(&con->mutex); |
| 2887 | |
| 2888 | if (fault) |
| 2889 | con_fault_finish(con); |
| 2890 | |
| 2891 | con->ops->put(con); |
| 2892 | } |
| 2893 | |
| 2894 | /* |
| 2895 | * Generic error/fault handler. A retry mechanism is used with |
| 2896 | * exponential backoff |
| 2897 | */ |
| 2898 | static void con_fault(struct ceph_connection *con) |
| 2899 | { |
| 2900 | dout("fault %p state %lu to peer %s\n", |
| 2901 | con, con->state, ceph_pr_addr(&con->peer_addr.in_addr)); |
| 2902 | |
| 2903 | pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name), |
| 2904 | ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg); |
| 2905 | con->error_msg = NULL; |
| 2906 | |
| 2907 | WARN_ON(con->state != CON_STATE_CONNECTING && |
| 2908 | con->state != CON_STATE_NEGOTIATING && |
| 2909 | con->state != CON_STATE_OPEN); |
| 2910 | |
| 2911 | con_close_socket(con); |
| 2912 | |
| 2913 | if (con_flag_test(con, CON_FLAG_LOSSYTX)) { |
| 2914 | dout("fault on LOSSYTX channel, marking CLOSED\n"); |
| 2915 | con->state = CON_STATE_CLOSED; |
| 2916 | return; |
| 2917 | } |
| 2918 | |
| 2919 | if (con->in_msg) { |
| 2920 | BUG_ON(con->in_msg->con != con); |
| 2921 | con->in_msg->con = NULL; |
| 2922 | ceph_msg_put(con->in_msg); |
| 2923 | con->in_msg = NULL; |
| 2924 | con->ops->put(con); |
| 2925 | } |
| 2926 | |
| 2927 | /* Requeue anything that hasn't been acked */ |
| 2928 | list_splice_init(&con->out_sent, &con->out_queue); |
| 2929 | |
| 2930 | /* If there are no messages queued or keepalive pending, place |
| 2931 | * the connection in a STANDBY state */ |
| 2932 | if (list_empty(&con->out_queue) && |
| 2933 | !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) { |
| 2934 | dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con); |
| 2935 | con_flag_clear(con, CON_FLAG_WRITE_PENDING); |
| 2936 | con->state = CON_STATE_STANDBY; |
| 2937 | } else { |
| 2938 | /* retry after a delay. */ |
| 2939 | con->state = CON_STATE_PREOPEN; |
| 2940 | if (con->delay == 0) |
| 2941 | con->delay = BASE_DELAY_INTERVAL; |
| 2942 | else if (con->delay < MAX_DELAY_INTERVAL) |
| 2943 | con->delay *= 2; |
| 2944 | con_flag_set(con, CON_FLAG_BACKOFF); |
| 2945 | queue_con(con); |
| 2946 | } |
| 2947 | } |
| 2948 | |
| 2949 | |
| 2950 | |
| 2951 | /* |
| 2952 | * initialize a new messenger instance |
| 2953 | */ |
| 2954 | void ceph_messenger_init(struct ceph_messenger *msgr, |
| 2955 | struct ceph_entity_addr *myaddr, |
| 2956 | u64 supported_features, |
| 2957 | u64 required_features, |
| 2958 | bool nocrc, |
| 2959 | bool tcp_nodelay) |
| 2960 | { |
| 2961 | msgr->supported_features = supported_features; |
| 2962 | msgr->required_features = required_features; |
| 2963 | |
| 2964 | spin_lock_init(&msgr->global_seq_lock); |
| 2965 | |
| 2966 | if (myaddr) |
| 2967 | msgr->inst.addr = *myaddr; |
| 2968 | |
| 2969 | /* select a random nonce */ |
| 2970 | msgr->inst.addr.type = 0; |
| 2971 | get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce)); |
| 2972 | encode_my_addr(msgr); |
| 2973 | msgr->nocrc = nocrc; |
| 2974 | msgr->tcp_nodelay = tcp_nodelay; |
| 2975 | |
| 2976 | atomic_set(&msgr->stopping, 0); |
| 2977 | write_pnet(&msgr->net, get_net(current->nsproxy->net_ns)); |
| 2978 | |
| 2979 | dout("%s %p\n", __func__, msgr); |
| 2980 | } |
| 2981 | EXPORT_SYMBOL(ceph_messenger_init); |
| 2982 | |
| 2983 | void ceph_messenger_fini(struct ceph_messenger *msgr) |
| 2984 | { |
| 2985 | put_net(read_pnet(&msgr->net)); |
| 2986 | } |
| 2987 | EXPORT_SYMBOL(ceph_messenger_fini); |
| 2988 | |
| 2989 | static void clear_standby(struct ceph_connection *con) |
| 2990 | { |
| 2991 | /* come back from STANDBY? */ |
| 2992 | if (con->state == CON_STATE_STANDBY) { |
| 2993 | dout("clear_standby %p and ++connect_seq\n", con); |
| 2994 | con->state = CON_STATE_PREOPEN; |
| 2995 | con->connect_seq++; |
| 2996 | WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING)); |
| 2997 | WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)); |
| 2998 | } |
| 2999 | } |
| 3000 | |
| 3001 | /* |
| 3002 | * Queue up an outgoing message on the given connection. |
| 3003 | */ |
| 3004 | void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg) |
| 3005 | { |
| 3006 | /* set src+dst */ |
| 3007 | msg->hdr.src = con->msgr->inst.name; |
| 3008 | BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len)); |
| 3009 | msg->needs_out_seq = true; |
| 3010 | |
| 3011 | mutex_lock(&con->mutex); |
| 3012 | |
| 3013 | if (con->state == CON_STATE_CLOSED) { |
| 3014 | dout("con_send %p closed, dropping %p\n", con, msg); |
| 3015 | ceph_msg_put(msg); |
| 3016 | mutex_unlock(&con->mutex); |
| 3017 | return; |
| 3018 | } |
| 3019 | |
| 3020 | BUG_ON(msg->con != NULL); |
| 3021 | msg->con = con->ops->get(con); |
| 3022 | BUG_ON(msg->con == NULL); |
| 3023 | |
| 3024 | BUG_ON(!list_empty(&msg->list_head)); |
| 3025 | list_add_tail(&msg->list_head, &con->out_queue); |
| 3026 | dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg, |
| 3027 | ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type), |
| 3028 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), |
| 3029 | le32_to_cpu(msg->hdr.front_len), |
| 3030 | le32_to_cpu(msg->hdr.middle_len), |
| 3031 | le32_to_cpu(msg->hdr.data_len)); |
| 3032 | |
| 3033 | clear_standby(con); |
| 3034 | mutex_unlock(&con->mutex); |
| 3035 | |
| 3036 | /* if there wasn't anything waiting to send before, queue |
| 3037 | * new work */ |
| 3038 | if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0) |
| 3039 | queue_con(con); |
| 3040 | } |
| 3041 | EXPORT_SYMBOL(ceph_con_send); |
| 3042 | |
| 3043 | /* |
| 3044 | * Revoke a message that was previously queued for send |
| 3045 | */ |
| 3046 | void ceph_msg_revoke(struct ceph_msg *msg) |
| 3047 | { |
| 3048 | struct ceph_connection *con = msg->con; |
| 3049 | |
| 3050 | if (!con) |
| 3051 | return; /* Message not in our possession */ |
| 3052 | |
| 3053 | mutex_lock(&con->mutex); |
| 3054 | if (!list_empty(&msg->list_head)) { |
| 3055 | dout("%s %p msg %p - was on queue\n", __func__, con, msg); |
| 3056 | list_del_init(&msg->list_head); |
| 3057 | BUG_ON(msg->con == NULL); |
| 3058 | msg->con->ops->put(msg->con); |
| 3059 | msg->con = NULL; |
| 3060 | msg->hdr.seq = 0; |
| 3061 | |
| 3062 | ceph_msg_put(msg); |
| 3063 | } |
| 3064 | if (con->out_msg == msg) { |
| 3065 | dout("%s %p msg %p - was sending\n", __func__, con, msg); |
| 3066 | con->out_msg = NULL; |
| 3067 | if (con->out_kvec_is_msg) { |
| 3068 | con->out_skip = con->out_kvec_bytes; |
| 3069 | con->out_kvec_is_msg = false; |
| 3070 | } |
| 3071 | msg->hdr.seq = 0; |
| 3072 | |
| 3073 | ceph_msg_put(msg); |
| 3074 | } |
| 3075 | mutex_unlock(&con->mutex); |
| 3076 | } |
| 3077 | |
| 3078 | /* |
| 3079 | * Revoke a message that we may be reading data into |
| 3080 | */ |
| 3081 | void ceph_msg_revoke_incoming(struct ceph_msg *msg) |
| 3082 | { |
| 3083 | struct ceph_connection *con; |
| 3084 | |
| 3085 | BUG_ON(msg == NULL); |
| 3086 | if (!msg->con) { |
| 3087 | dout("%s msg %p null con\n", __func__, msg); |
| 3088 | |
| 3089 | return; /* Message not in our possession */ |
| 3090 | } |
| 3091 | |
| 3092 | con = msg->con; |
| 3093 | mutex_lock(&con->mutex); |
| 3094 | if (con->in_msg == msg) { |
| 3095 | unsigned int front_len = le32_to_cpu(con->in_hdr.front_len); |
| 3096 | unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len); |
| 3097 | unsigned int data_len = le32_to_cpu(con->in_hdr.data_len); |
| 3098 | |
| 3099 | /* skip rest of message */ |
| 3100 | dout("%s %p msg %p revoked\n", __func__, con, msg); |
| 3101 | con->in_base_pos = con->in_base_pos - |
| 3102 | sizeof(struct ceph_msg_header) - |
| 3103 | front_len - |
| 3104 | middle_len - |
| 3105 | data_len - |
| 3106 | sizeof(struct ceph_msg_footer); |
| 3107 | ceph_msg_put(con->in_msg); |
| 3108 | con->in_msg = NULL; |
| 3109 | con->in_tag = CEPH_MSGR_TAG_READY; |
| 3110 | con->in_seq++; |
| 3111 | } else { |
| 3112 | dout("%s %p in_msg %p msg %p no-op\n", |
| 3113 | __func__, con, con->in_msg, msg); |
| 3114 | } |
| 3115 | mutex_unlock(&con->mutex); |
| 3116 | } |
| 3117 | |
| 3118 | /* |
| 3119 | * Queue a keepalive byte to ensure the tcp connection is alive. |
| 3120 | */ |
| 3121 | void ceph_con_keepalive(struct ceph_connection *con) |
| 3122 | { |
| 3123 | dout("con_keepalive %p\n", con); |
| 3124 | mutex_lock(&con->mutex); |
| 3125 | clear_standby(con); |
| 3126 | mutex_unlock(&con->mutex); |
| 3127 | if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 && |
| 3128 | con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0) |
| 3129 | queue_con(con); |
| 3130 | } |
| 3131 | EXPORT_SYMBOL(ceph_con_keepalive); |
| 3132 | |
| 3133 | bool ceph_con_keepalive_expired(struct ceph_connection *con, |
| 3134 | unsigned long interval) |
| 3135 | { |
| 3136 | if (interval > 0 && |
| 3137 | (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) { |
| 3138 | struct timespec now = CURRENT_TIME; |
| 3139 | struct timespec ts; |
| 3140 | jiffies_to_timespec(interval, &ts); |
| 3141 | ts = timespec_add(con->last_keepalive_ack, ts); |
| 3142 | return timespec_compare(&now, &ts) >= 0; |
| 3143 | } |
| 3144 | return false; |
| 3145 | } |
| 3146 | |
| 3147 | static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type) |
| 3148 | { |
| 3149 | struct ceph_msg_data *data; |
| 3150 | |
| 3151 | if (WARN_ON(!ceph_msg_data_type_valid(type))) |
| 3152 | return NULL; |
| 3153 | |
| 3154 | data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS); |
| 3155 | if (data) |
| 3156 | data->type = type; |
| 3157 | INIT_LIST_HEAD(&data->links); |
| 3158 | |
| 3159 | return data; |
| 3160 | } |
| 3161 | |
| 3162 | static void ceph_msg_data_destroy(struct ceph_msg_data *data) |
| 3163 | { |
| 3164 | if (!data) |
| 3165 | return; |
| 3166 | |
| 3167 | WARN_ON(!list_empty(&data->links)); |
| 3168 | if (data->type == CEPH_MSG_DATA_PAGELIST) |
| 3169 | ceph_pagelist_release(data->pagelist); |
| 3170 | kmem_cache_free(ceph_msg_data_cache, data); |
| 3171 | } |
| 3172 | |
| 3173 | void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages, |
| 3174 | size_t length, size_t alignment) |
| 3175 | { |
| 3176 | struct ceph_msg_data *data; |
| 3177 | |
| 3178 | BUG_ON(!pages); |
| 3179 | BUG_ON(!length); |
| 3180 | |
| 3181 | data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES); |
| 3182 | BUG_ON(!data); |
| 3183 | data->pages = pages; |
| 3184 | data->length = length; |
| 3185 | data->alignment = alignment & ~PAGE_MASK; |
| 3186 | |
| 3187 | list_add_tail(&data->links, &msg->data); |
| 3188 | msg->data_length += length; |
| 3189 | } |
| 3190 | EXPORT_SYMBOL(ceph_msg_data_add_pages); |
| 3191 | |
| 3192 | void ceph_msg_data_add_pagelist(struct ceph_msg *msg, |
| 3193 | struct ceph_pagelist *pagelist) |
| 3194 | { |
| 3195 | struct ceph_msg_data *data; |
| 3196 | |
| 3197 | BUG_ON(!pagelist); |
| 3198 | BUG_ON(!pagelist->length); |
| 3199 | |
| 3200 | data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST); |
| 3201 | BUG_ON(!data); |
| 3202 | data->pagelist = pagelist; |
| 3203 | |
| 3204 | list_add_tail(&data->links, &msg->data); |
| 3205 | msg->data_length += pagelist->length; |
| 3206 | } |
| 3207 | EXPORT_SYMBOL(ceph_msg_data_add_pagelist); |
| 3208 | |
| 3209 | #ifdef CONFIG_BLOCK |
| 3210 | void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio, |
| 3211 | size_t length) |
| 3212 | { |
| 3213 | struct ceph_msg_data *data; |
| 3214 | |
| 3215 | BUG_ON(!bio); |
| 3216 | |
| 3217 | data = ceph_msg_data_create(CEPH_MSG_DATA_BIO); |
| 3218 | BUG_ON(!data); |
| 3219 | data->bio = bio; |
| 3220 | data->bio_length = length; |
| 3221 | |
| 3222 | list_add_tail(&data->links, &msg->data); |
| 3223 | msg->data_length += length; |
| 3224 | } |
| 3225 | EXPORT_SYMBOL(ceph_msg_data_add_bio); |
| 3226 | #endif /* CONFIG_BLOCK */ |
| 3227 | |
| 3228 | /* |
| 3229 | * construct a new message with given type, size |
| 3230 | * the new msg has a ref count of 1. |
| 3231 | */ |
| 3232 | struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags, |
| 3233 | bool can_fail) |
| 3234 | { |
| 3235 | struct ceph_msg *m; |
| 3236 | |
| 3237 | m = kmem_cache_zalloc(ceph_msg_cache, flags); |
| 3238 | if (m == NULL) |
| 3239 | goto out; |
| 3240 | |
| 3241 | m->hdr.type = cpu_to_le16(type); |
| 3242 | m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT); |
| 3243 | m->hdr.front_len = cpu_to_le32(front_len); |
| 3244 | |
| 3245 | INIT_LIST_HEAD(&m->list_head); |
| 3246 | kref_init(&m->kref); |
| 3247 | INIT_LIST_HEAD(&m->data); |
| 3248 | |
| 3249 | /* front */ |
| 3250 | if (front_len) { |
| 3251 | m->front.iov_base = ceph_kvmalloc(front_len, flags); |
| 3252 | if (m->front.iov_base == NULL) { |
| 3253 | dout("ceph_msg_new can't allocate %d bytes\n", |
| 3254 | front_len); |
| 3255 | goto out2; |
| 3256 | } |
| 3257 | } else { |
| 3258 | m->front.iov_base = NULL; |
| 3259 | } |
| 3260 | m->front_alloc_len = m->front.iov_len = front_len; |
| 3261 | |
| 3262 | dout("ceph_msg_new %p front %d\n", m, front_len); |
| 3263 | return m; |
| 3264 | |
| 3265 | out2: |
| 3266 | ceph_msg_put(m); |
| 3267 | out: |
| 3268 | if (!can_fail) { |
| 3269 | pr_err("msg_new can't create type %d front %d\n", type, |
| 3270 | front_len); |
| 3271 | WARN_ON(1); |
| 3272 | } else { |
| 3273 | dout("msg_new can't create type %d front %d\n", type, |
| 3274 | front_len); |
| 3275 | } |
| 3276 | return NULL; |
| 3277 | } |
| 3278 | EXPORT_SYMBOL(ceph_msg_new); |
| 3279 | |
| 3280 | /* |
| 3281 | * Allocate "middle" portion of a message, if it is needed and wasn't |
| 3282 | * allocated by alloc_msg. This allows us to read a small fixed-size |
| 3283 | * per-type header in the front and then gracefully fail (i.e., |
| 3284 | * propagate the error to the caller based on info in the front) when |
| 3285 | * the middle is too large. |
| 3286 | */ |
| 3287 | static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg) |
| 3288 | { |
| 3289 | int type = le16_to_cpu(msg->hdr.type); |
| 3290 | int middle_len = le32_to_cpu(msg->hdr.middle_len); |
| 3291 | |
| 3292 | dout("alloc_middle %p type %d %s middle_len %d\n", msg, type, |
| 3293 | ceph_msg_type_name(type), middle_len); |
| 3294 | BUG_ON(!middle_len); |
| 3295 | BUG_ON(msg->middle); |
| 3296 | |
| 3297 | msg->middle = ceph_buffer_new(middle_len, GFP_NOFS); |
| 3298 | if (!msg->middle) |
| 3299 | return -ENOMEM; |
| 3300 | return 0; |
| 3301 | } |
| 3302 | |
| 3303 | /* |
| 3304 | * Allocate a message for receiving an incoming message on a |
| 3305 | * connection, and save the result in con->in_msg. Uses the |
| 3306 | * connection's private alloc_msg op if available. |
| 3307 | * |
| 3308 | * Returns 0 on success, or a negative error code. |
| 3309 | * |
| 3310 | * On success, if we set *skip = 1: |
| 3311 | * - the next message should be skipped and ignored. |
| 3312 | * - con->in_msg == NULL |
| 3313 | * or if we set *skip = 0: |
| 3314 | * - con->in_msg is non-null. |
| 3315 | * On error (ENOMEM, EAGAIN, ...), |
| 3316 | * - con->in_msg == NULL |
| 3317 | */ |
| 3318 | static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip) |
| 3319 | { |
| 3320 | struct ceph_msg_header *hdr = &con->in_hdr; |
| 3321 | int middle_len = le32_to_cpu(hdr->middle_len); |
| 3322 | struct ceph_msg *msg; |
| 3323 | int ret = 0; |
| 3324 | |
| 3325 | BUG_ON(con->in_msg != NULL); |
| 3326 | BUG_ON(!con->ops->alloc_msg); |
| 3327 | |
| 3328 | mutex_unlock(&con->mutex); |
| 3329 | msg = con->ops->alloc_msg(con, hdr, skip); |
| 3330 | mutex_lock(&con->mutex); |
| 3331 | if (con->state != CON_STATE_OPEN) { |
| 3332 | if (msg) |
| 3333 | ceph_msg_put(msg); |
| 3334 | return -EAGAIN; |
| 3335 | } |
| 3336 | if (msg) { |
| 3337 | BUG_ON(*skip); |
| 3338 | con->in_msg = msg; |
| 3339 | con->in_msg->con = con->ops->get(con); |
| 3340 | BUG_ON(con->in_msg->con == NULL); |
| 3341 | } else { |
| 3342 | /* |
| 3343 | * Null message pointer means either we should skip |
| 3344 | * this message or we couldn't allocate memory. The |
| 3345 | * former is not an error. |
| 3346 | */ |
| 3347 | if (*skip) |
| 3348 | return 0; |
| 3349 | |
| 3350 | con->error_msg = "error allocating memory for incoming message"; |
| 3351 | return -ENOMEM; |
| 3352 | } |
| 3353 | memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr)); |
| 3354 | |
| 3355 | if (middle_len && !con->in_msg->middle) { |
| 3356 | ret = ceph_alloc_middle(con, con->in_msg); |
| 3357 | if (ret < 0) { |
| 3358 | ceph_msg_put(con->in_msg); |
| 3359 | con->in_msg = NULL; |
| 3360 | } |
| 3361 | } |
| 3362 | |
| 3363 | return ret; |
| 3364 | } |
| 3365 | |
| 3366 | |
| 3367 | /* |
| 3368 | * Free a generically kmalloc'd message. |
| 3369 | */ |
| 3370 | static void ceph_msg_free(struct ceph_msg *m) |
| 3371 | { |
| 3372 | dout("%s %p\n", __func__, m); |
| 3373 | kvfree(m->front.iov_base); |
| 3374 | kmem_cache_free(ceph_msg_cache, m); |
| 3375 | } |
| 3376 | |
| 3377 | static void ceph_msg_release(struct kref *kref) |
| 3378 | { |
| 3379 | struct ceph_msg *m = container_of(kref, struct ceph_msg, kref); |
| 3380 | LIST_HEAD(data); |
| 3381 | struct list_head *links; |
| 3382 | struct list_head *next; |
| 3383 | |
| 3384 | dout("%s %p\n", __func__, m); |
| 3385 | WARN_ON(!list_empty(&m->list_head)); |
| 3386 | |
| 3387 | /* drop middle, data, if any */ |
| 3388 | if (m->middle) { |
| 3389 | ceph_buffer_put(m->middle); |
| 3390 | m->middle = NULL; |
| 3391 | } |
| 3392 | |
| 3393 | list_splice_init(&m->data, &data); |
| 3394 | list_for_each_safe(links, next, &data) { |
| 3395 | struct ceph_msg_data *data; |
| 3396 | |
| 3397 | data = list_entry(links, struct ceph_msg_data, links); |
| 3398 | list_del_init(links); |
| 3399 | ceph_msg_data_destroy(data); |
| 3400 | } |
| 3401 | m->data_length = 0; |
| 3402 | |
| 3403 | if (m->pool) |
| 3404 | ceph_msgpool_put(m->pool, m); |
| 3405 | else |
| 3406 | ceph_msg_free(m); |
| 3407 | } |
| 3408 | |
| 3409 | struct ceph_msg *ceph_msg_get(struct ceph_msg *msg) |
| 3410 | { |
| 3411 | dout("%s %p (was %d)\n", __func__, msg, |
| 3412 | atomic_read(&msg->kref.refcount)); |
| 3413 | kref_get(&msg->kref); |
| 3414 | return msg; |
| 3415 | } |
| 3416 | EXPORT_SYMBOL(ceph_msg_get); |
| 3417 | |
| 3418 | void ceph_msg_put(struct ceph_msg *msg) |
| 3419 | { |
| 3420 | dout("%s %p (was %d)\n", __func__, msg, |
| 3421 | atomic_read(&msg->kref.refcount)); |
| 3422 | kref_put(&msg->kref, ceph_msg_release); |
| 3423 | } |
| 3424 | EXPORT_SYMBOL(ceph_msg_put); |
| 3425 | |
| 3426 | void ceph_msg_dump(struct ceph_msg *msg) |
| 3427 | { |
| 3428 | pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg, |
| 3429 | msg->front_alloc_len, msg->data_length); |
| 3430 | print_hex_dump(KERN_DEBUG, "header: ", |
| 3431 | DUMP_PREFIX_OFFSET, 16, 1, |
| 3432 | &msg->hdr, sizeof(msg->hdr), true); |
| 3433 | print_hex_dump(KERN_DEBUG, " front: ", |
| 3434 | DUMP_PREFIX_OFFSET, 16, 1, |
| 3435 | msg->front.iov_base, msg->front.iov_len, true); |
| 3436 | if (msg->middle) |
| 3437 | print_hex_dump(KERN_DEBUG, "middle: ", |
| 3438 | DUMP_PREFIX_OFFSET, 16, 1, |
| 3439 | msg->middle->vec.iov_base, |
| 3440 | msg->middle->vec.iov_len, true); |
| 3441 | print_hex_dump(KERN_DEBUG, "footer: ", |
| 3442 | DUMP_PREFIX_OFFSET, 16, 1, |
| 3443 | &msg->footer, sizeof(msg->footer), true); |
| 3444 | } |
| 3445 | EXPORT_SYMBOL(ceph_msg_dump); |