1 2 #include <linux/ceph/ceph_debug.h> 3 4 #include <linux/module.h> 5 #include <linux/err.h> 6 #include <linux/highmem.h> 7 #include <linux/mm.h> 8 #include <linux/pagemap.h> 9 #include <linux/slab.h> 10 #include <linux/uaccess.h> 11 #ifdef CONFIG_BLOCK 12 #include <linux/bio.h> 13 #endif 14 15 #include <linux/ceph/libceph.h> 16 #include <linux/ceph/osd_client.h> 17 #include <linux/ceph/messenger.h> 18 #include <linux/ceph/decode.h> 19 #include <linux/ceph/auth.h> 20 #include <linux/ceph/pagelist.h> 21 22 #define OSD_OPREPLY_FRONT_LEN 512 23 24 static struct kmem_cache *ceph_osd_request_cache; 25 26 static const struct ceph_connection_operations osd_con_ops; 27 28 /* 29 * Implement client access to distributed object storage cluster. 30 * 31 * All data objects are stored within a cluster/cloud of OSDs, or 32 * "object storage devices." (Note that Ceph OSDs have _nothing_ to 33 * do with the T10 OSD extensions to SCSI.) Ceph OSDs are simply 34 * remote daemons serving up and coordinating consistent and safe 35 * access to storage. 36 * 37 * Cluster membership and the mapping of data objects onto storage devices 38 * are described by the osd map. 39 * 40 * We keep track of pending OSD requests (read, write), resubmit 41 * requests to different OSDs when the cluster topology/data layout 42 * change, or retry the affected requests when the communications 43 * channel with an OSD is reset. 44 */ 45 46 static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req); 47 static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req); 48 static void link_linger(struct ceph_osd *osd, 49 struct ceph_osd_linger_request *lreq); 50 static void unlink_linger(struct ceph_osd *osd, 51 struct ceph_osd_linger_request *lreq); 52 53 #if 1 54 static inline bool rwsem_is_wrlocked(struct rw_semaphore *sem) 55 { 56 bool wrlocked = true; 57 58 if (unlikely(down_read_trylock(sem))) { 59 wrlocked = false; 60 up_read(sem); 61 } 62 63 return wrlocked; 64 } 65 static inline void verify_osdc_locked(struct ceph_osd_client *osdc) 66 { 67 WARN_ON(!rwsem_is_locked(&osdc->lock)); 68 } 69 static inline void verify_osdc_wrlocked(struct ceph_osd_client *osdc) 70 { 71 WARN_ON(!rwsem_is_wrlocked(&osdc->lock)); 72 } 73 static inline void verify_osd_locked(struct ceph_osd *osd) 74 { 75 struct ceph_osd_client *osdc = osd->o_osdc; 76 77 WARN_ON(!(mutex_is_locked(&osd->lock) && 78 rwsem_is_locked(&osdc->lock)) && 79 !rwsem_is_wrlocked(&osdc->lock)); 80 } 81 static inline void verify_lreq_locked(struct ceph_osd_linger_request *lreq) 82 { 83 WARN_ON(!mutex_is_locked(&lreq->lock)); 84 } 85 #else 86 static inline void verify_osdc_locked(struct ceph_osd_client *osdc) { } 87 static inline void verify_osdc_wrlocked(struct ceph_osd_client *osdc) { } 88 static inline void verify_osd_locked(struct ceph_osd *osd) { } 89 static inline void verify_lreq_locked(struct ceph_osd_linger_request *lreq) { } 90 #endif 91 92 /* 93 * calculate the mapping of a file extent onto an object, and fill out the 94 * request accordingly. shorten extent as necessary if it crosses an 95 * object boundary. 96 * 97 * fill osd op in request message. 98 */ 99 static int calc_layout(struct ceph_file_layout *layout, u64 off, u64 *plen, 100 u64 *objnum, u64 *objoff, u64 *objlen) 101 { 102 u64 orig_len = *plen; 103 int r; 104 105 /* object extent? */ 106 r = ceph_calc_file_object_mapping(layout, off, orig_len, objnum, 107 objoff, objlen); 108 if (r < 0) 109 return r; 110 if (*objlen < orig_len) { 111 *plen = *objlen; 112 dout(" skipping last %llu, final file extent %llu~%llu\n", 113 orig_len - *plen, off, *plen); 114 } 115 116 dout("calc_layout objnum=%llx %llu~%llu\n", *objnum, *objoff, *objlen); 117 118 return 0; 119 } 120 121 static void ceph_osd_data_init(struct ceph_osd_data *osd_data) 122 { 123 memset(osd_data, 0, sizeof (*osd_data)); 124 osd_data->type = CEPH_OSD_DATA_TYPE_NONE; 125 } 126 127 static void ceph_osd_data_pages_init(struct ceph_osd_data *osd_data, 128 struct page **pages, u64 length, u32 alignment, 129 bool pages_from_pool, bool own_pages) 130 { 131 osd_data->type = CEPH_OSD_DATA_TYPE_PAGES; 132 osd_data->pages = pages; 133 osd_data->length = length; 134 osd_data->alignment = alignment; 135 osd_data->pages_from_pool = pages_from_pool; 136 osd_data->own_pages = own_pages; 137 } 138 139 static void ceph_osd_data_pagelist_init(struct ceph_osd_data *osd_data, 140 struct ceph_pagelist *pagelist) 141 { 142 osd_data->type = CEPH_OSD_DATA_TYPE_PAGELIST; 143 osd_data->pagelist = pagelist; 144 } 145 146 #ifdef CONFIG_BLOCK 147 static void ceph_osd_data_bio_init(struct ceph_osd_data *osd_data, 148 struct bio *bio, size_t bio_length) 149 { 150 osd_data->type = CEPH_OSD_DATA_TYPE_BIO; 151 osd_data->bio = bio; 152 osd_data->bio_length = bio_length; 153 } 154 #endif /* CONFIG_BLOCK */ 155 156 #define osd_req_op_data(oreq, whch, typ, fld) \ 157 ({ \ 158 struct ceph_osd_request *__oreq = (oreq); \ 159 unsigned int __whch = (whch); \ 160 BUG_ON(__whch >= __oreq->r_num_ops); \ 161 &__oreq->r_ops[__whch].typ.fld; \ 162 }) 163 164 static struct ceph_osd_data * 165 osd_req_op_raw_data_in(struct ceph_osd_request *osd_req, unsigned int which) 166 { 167 BUG_ON(which >= osd_req->r_num_ops); 168 169 return &osd_req->r_ops[which].raw_data_in; 170 } 171 172 struct ceph_osd_data * 173 osd_req_op_extent_osd_data(struct ceph_osd_request *osd_req, 174 unsigned int which) 175 { 176 return osd_req_op_data(osd_req, which, extent, osd_data); 177 } 178 EXPORT_SYMBOL(osd_req_op_extent_osd_data); 179 180 void osd_req_op_raw_data_in_pages(struct ceph_osd_request *osd_req, 181 unsigned int which, struct page **pages, 182 u64 length, u32 alignment, 183 bool pages_from_pool, bool own_pages) 184 { 185 struct ceph_osd_data *osd_data; 186 187 osd_data = osd_req_op_raw_data_in(osd_req, which); 188 ceph_osd_data_pages_init(osd_data, pages, length, alignment, 189 pages_from_pool, own_pages); 190 } 191 EXPORT_SYMBOL(osd_req_op_raw_data_in_pages); 192 193 void osd_req_op_extent_osd_data_pages(struct ceph_osd_request *osd_req, 194 unsigned int which, struct page **pages, 195 u64 length, u32 alignment, 196 bool pages_from_pool, bool own_pages) 197 { 198 struct ceph_osd_data *osd_data; 199 200 osd_data = osd_req_op_data(osd_req, which, extent, osd_data); 201 ceph_osd_data_pages_init(osd_data, pages, length, alignment, 202 pages_from_pool, own_pages); 203 } 204 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pages); 205 206 void osd_req_op_extent_osd_data_pagelist(struct ceph_osd_request *osd_req, 207 unsigned int which, struct ceph_pagelist *pagelist) 208 { 209 struct ceph_osd_data *osd_data; 210 211 osd_data = osd_req_op_data(osd_req, which, extent, osd_data); 212 ceph_osd_data_pagelist_init(osd_data, pagelist); 213 } 214 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pagelist); 215 216 #ifdef CONFIG_BLOCK 217 void osd_req_op_extent_osd_data_bio(struct ceph_osd_request *osd_req, 218 unsigned int which, struct bio *bio, size_t bio_length) 219 { 220 struct ceph_osd_data *osd_data; 221 222 osd_data = osd_req_op_data(osd_req, which, extent, osd_data); 223 ceph_osd_data_bio_init(osd_data, bio, bio_length); 224 } 225 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bio); 226 #endif /* CONFIG_BLOCK */ 227 228 static void osd_req_op_cls_request_info_pagelist( 229 struct ceph_osd_request *osd_req, 230 unsigned int which, struct ceph_pagelist *pagelist) 231 { 232 struct ceph_osd_data *osd_data; 233 234 osd_data = osd_req_op_data(osd_req, which, cls, request_info); 235 ceph_osd_data_pagelist_init(osd_data, pagelist); 236 } 237 238 void osd_req_op_cls_request_data_pagelist( 239 struct ceph_osd_request *osd_req, 240 unsigned int which, struct ceph_pagelist *pagelist) 241 { 242 struct ceph_osd_data *osd_data; 243 244 osd_data = osd_req_op_data(osd_req, which, cls, request_data); 245 ceph_osd_data_pagelist_init(osd_data, pagelist); 246 osd_req->r_ops[which].cls.indata_len += pagelist->length; 247 osd_req->r_ops[which].indata_len += pagelist->length; 248 } 249 EXPORT_SYMBOL(osd_req_op_cls_request_data_pagelist); 250 251 void osd_req_op_cls_request_data_pages(struct ceph_osd_request *osd_req, 252 unsigned int which, struct page **pages, u64 length, 253 u32 alignment, bool pages_from_pool, bool own_pages) 254 { 255 struct ceph_osd_data *osd_data; 256 257 osd_data = osd_req_op_data(osd_req, which, cls, request_data); 258 ceph_osd_data_pages_init(osd_data, pages, length, alignment, 259 pages_from_pool, own_pages); 260 osd_req->r_ops[which].cls.indata_len += length; 261 osd_req->r_ops[which].indata_len += length; 262 } 263 EXPORT_SYMBOL(osd_req_op_cls_request_data_pages); 264 265 void osd_req_op_cls_response_data_pages(struct ceph_osd_request *osd_req, 266 unsigned int which, struct page **pages, u64 length, 267 u32 alignment, bool pages_from_pool, bool own_pages) 268 { 269 struct ceph_osd_data *osd_data; 270 271 osd_data = osd_req_op_data(osd_req, which, cls, response_data); 272 ceph_osd_data_pages_init(osd_data, pages, length, alignment, 273 pages_from_pool, own_pages); 274 } 275 EXPORT_SYMBOL(osd_req_op_cls_response_data_pages); 276 277 static u64 ceph_osd_data_length(struct ceph_osd_data *osd_data) 278 { 279 switch (osd_data->type) { 280 case CEPH_OSD_DATA_TYPE_NONE: 281 return 0; 282 case CEPH_OSD_DATA_TYPE_PAGES: 283 return osd_data->length; 284 case CEPH_OSD_DATA_TYPE_PAGELIST: 285 return (u64)osd_data->pagelist->length; 286 #ifdef CONFIG_BLOCK 287 case CEPH_OSD_DATA_TYPE_BIO: 288 return (u64)osd_data->bio_length; 289 #endif /* CONFIG_BLOCK */ 290 default: 291 WARN(true, "unrecognized data type %d\n", (int)osd_data->type); 292 return 0; 293 } 294 } 295 296 static void ceph_osd_data_release(struct ceph_osd_data *osd_data) 297 { 298 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES && osd_data->own_pages) { 299 int num_pages; 300 301 num_pages = calc_pages_for((u64)osd_data->alignment, 302 (u64)osd_data->length); 303 ceph_release_page_vector(osd_data->pages, num_pages); 304 } 305 ceph_osd_data_init(osd_data); 306 } 307 308 static void osd_req_op_data_release(struct ceph_osd_request *osd_req, 309 unsigned int which) 310 { 311 struct ceph_osd_req_op *op; 312 313 BUG_ON(which >= osd_req->r_num_ops); 314 op = &osd_req->r_ops[which]; 315 316 switch (op->op) { 317 case CEPH_OSD_OP_READ: 318 case CEPH_OSD_OP_WRITE: 319 case CEPH_OSD_OP_WRITEFULL: 320 ceph_osd_data_release(&op->extent.osd_data); 321 break; 322 case CEPH_OSD_OP_CALL: 323 ceph_osd_data_release(&op->cls.request_info); 324 ceph_osd_data_release(&op->cls.request_data); 325 ceph_osd_data_release(&op->cls.response_data); 326 break; 327 case CEPH_OSD_OP_SETXATTR: 328 case CEPH_OSD_OP_CMPXATTR: 329 ceph_osd_data_release(&op->xattr.osd_data); 330 break; 331 case CEPH_OSD_OP_STAT: 332 ceph_osd_data_release(&op->raw_data_in); 333 break; 334 case CEPH_OSD_OP_NOTIFY_ACK: 335 ceph_osd_data_release(&op->notify_ack.request_data); 336 break; 337 case CEPH_OSD_OP_NOTIFY: 338 ceph_osd_data_release(&op->notify.request_data); 339 ceph_osd_data_release(&op->notify.response_data); 340 break; 341 case CEPH_OSD_OP_LIST_WATCHERS: 342 ceph_osd_data_release(&op->list_watchers.response_data); 343 break; 344 default: 345 break; 346 } 347 } 348 349 /* 350 * Assumes @t is zero-initialized. 351 */ 352 static void target_init(struct ceph_osd_request_target *t) 353 { 354 ceph_oid_init(&t->base_oid); 355 ceph_oloc_init(&t->base_oloc); 356 ceph_oid_init(&t->target_oid); 357 ceph_oloc_init(&t->target_oloc); 358 359 ceph_osds_init(&t->acting); 360 ceph_osds_init(&t->up); 361 t->size = -1; 362 t->min_size = -1; 363 364 t->osd = CEPH_HOMELESS_OSD; 365 } 366 367 static void target_copy(struct ceph_osd_request_target *dest, 368 const struct ceph_osd_request_target *src) 369 { 370 ceph_oid_copy(&dest->base_oid, &src->base_oid); 371 ceph_oloc_copy(&dest->base_oloc, &src->base_oloc); 372 ceph_oid_copy(&dest->target_oid, &src->target_oid); 373 ceph_oloc_copy(&dest->target_oloc, &src->target_oloc); 374 375 dest->pgid = src->pgid; /* struct */ 376 dest->pg_num = src->pg_num; 377 dest->pg_num_mask = src->pg_num_mask; 378 ceph_osds_copy(&dest->acting, &src->acting); 379 ceph_osds_copy(&dest->up, &src->up); 380 dest->size = src->size; 381 dest->min_size = src->min_size; 382 dest->sort_bitwise = src->sort_bitwise; 383 384 dest->flags = src->flags; 385 dest->paused = src->paused; 386 387 dest->osd = src->osd; 388 } 389 390 static void target_destroy(struct ceph_osd_request_target *t) 391 { 392 ceph_oid_destroy(&t->base_oid); 393 ceph_oloc_destroy(&t->base_oloc); 394 ceph_oid_destroy(&t->target_oid); 395 ceph_oloc_destroy(&t->target_oloc); 396 } 397 398 /* 399 * requests 400 */ 401 static void request_release_checks(struct ceph_osd_request *req) 402 { 403 WARN_ON(!RB_EMPTY_NODE(&req->r_node)); 404 WARN_ON(!RB_EMPTY_NODE(&req->r_mc_node)); 405 WARN_ON(!list_empty(&req->r_unsafe_item)); 406 WARN_ON(req->r_osd); 407 } 408 409 static void ceph_osdc_release_request(struct kref *kref) 410 { 411 struct ceph_osd_request *req = container_of(kref, 412 struct ceph_osd_request, r_kref); 413 unsigned int which; 414 415 dout("%s %p (r_request %p r_reply %p)\n", __func__, req, 416 req->r_request, req->r_reply); 417 request_release_checks(req); 418 419 if (req->r_request) 420 ceph_msg_put(req->r_request); 421 if (req->r_reply) 422 ceph_msg_put(req->r_reply); 423 424 for (which = 0; which < req->r_num_ops; which++) 425 osd_req_op_data_release(req, which); 426 427 target_destroy(&req->r_t); 428 ceph_put_snap_context(req->r_snapc); 429 430 if (req->r_mempool) 431 mempool_free(req, req->r_osdc->req_mempool); 432 else if (req->r_num_ops <= CEPH_OSD_SLAB_OPS) 433 kmem_cache_free(ceph_osd_request_cache, req); 434 else 435 kfree(req); 436 } 437 438 void ceph_osdc_get_request(struct ceph_osd_request *req) 439 { 440 dout("%s %p (was %d)\n", __func__, req, 441 kref_read(&req->r_kref)); 442 kref_get(&req->r_kref); 443 } 444 EXPORT_SYMBOL(ceph_osdc_get_request); 445 446 void ceph_osdc_put_request(struct ceph_osd_request *req) 447 { 448 if (req) { 449 dout("%s %p (was %d)\n", __func__, req, 450 kref_read(&req->r_kref)); 451 kref_put(&req->r_kref, ceph_osdc_release_request); 452 } 453 } 454 EXPORT_SYMBOL(ceph_osdc_put_request); 455 456 static void request_init(struct ceph_osd_request *req) 457 { 458 /* req only, each op is zeroed in _osd_req_op_init() */ 459 memset(req, 0, sizeof(*req)); 460 461 kref_init(&req->r_kref); 462 init_completion(&req->r_completion); 463 RB_CLEAR_NODE(&req->r_node); 464 RB_CLEAR_NODE(&req->r_mc_node); 465 INIT_LIST_HEAD(&req->r_unsafe_item); 466 467 target_init(&req->r_t); 468 } 469 470 /* 471 * This is ugly, but it allows us to reuse linger registration and ping 472 * requests, keeping the structure of the code around send_linger{_ping}() 473 * reasonable. Setting up a min_nr=2 mempool for each linger request 474 * and dealing with copying ops (this blasts req only, watch op remains 475 * intact) isn't any better. 476 */ 477 static void request_reinit(struct ceph_osd_request *req) 478 { 479 struct ceph_osd_client *osdc = req->r_osdc; 480 bool mempool = req->r_mempool; 481 unsigned int num_ops = req->r_num_ops; 482 u64 snapid = req->r_snapid; 483 struct ceph_snap_context *snapc = req->r_snapc; 484 bool linger = req->r_linger; 485 struct ceph_msg *request_msg = req->r_request; 486 struct ceph_msg *reply_msg = req->r_reply; 487 488 dout("%s req %p\n", __func__, req); 489 WARN_ON(kref_read(&req->r_kref) != 1); 490 request_release_checks(req); 491 492 WARN_ON(kref_read(&request_msg->kref) != 1); 493 WARN_ON(kref_read(&reply_msg->kref) != 1); 494 target_destroy(&req->r_t); 495 496 request_init(req); 497 req->r_osdc = osdc; 498 req->r_mempool = mempool; 499 req->r_num_ops = num_ops; 500 req->r_snapid = snapid; 501 req->r_snapc = snapc; 502 req->r_linger = linger; 503 req->r_request = request_msg; 504 req->r_reply = reply_msg; 505 } 506 507 struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc, 508 struct ceph_snap_context *snapc, 509 unsigned int num_ops, 510 bool use_mempool, 511 gfp_t gfp_flags) 512 { 513 struct ceph_osd_request *req; 514 515 if (use_mempool) { 516 BUG_ON(num_ops > CEPH_OSD_SLAB_OPS); 517 req = mempool_alloc(osdc->req_mempool, gfp_flags); 518 } else if (num_ops <= CEPH_OSD_SLAB_OPS) { 519 req = kmem_cache_alloc(ceph_osd_request_cache, gfp_flags); 520 } else { 521 BUG_ON(num_ops > CEPH_OSD_MAX_OPS); 522 req = kmalloc(sizeof(*req) + num_ops * sizeof(req->r_ops[0]), 523 gfp_flags); 524 } 525 if (unlikely(!req)) 526 return NULL; 527 528 request_init(req); 529 req->r_osdc = osdc; 530 req->r_mempool = use_mempool; 531 req->r_num_ops = num_ops; 532 req->r_snapid = CEPH_NOSNAP; 533 req->r_snapc = ceph_get_snap_context(snapc); 534 535 dout("%s req %p\n", __func__, req); 536 return req; 537 } 538 EXPORT_SYMBOL(ceph_osdc_alloc_request); 539 540 static int ceph_oloc_encoding_size(struct ceph_object_locator *oloc) 541 { 542 return 8 + 4 + 4 + 4 + (oloc->pool_ns ? oloc->pool_ns->len : 0); 543 } 544 545 int ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp) 546 { 547 struct ceph_osd_client *osdc = req->r_osdc; 548 struct ceph_msg *msg; 549 int msg_size; 550 551 WARN_ON(ceph_oid_empty(&req->r_base_oid)); 552 WARN_ON(ceph_oloc_empty(&req->r_base_oloc)); 553 554 /* create request message */ 555 msg_size = 4 + 4 + 4; /* client_inc, osdmap_epoch, flags */ 556 msg_size += 4 + 4 + 4 + 8; /* mtime, reassert_version */ 557 msg_size += CEPH_ENCODING_START_BLK_LEN + 558 ceph_oloc_encoding_size(&req->r_base_oloc); /* oloc */ 559 msg_size += 1 + 8 + 4 + 4; /* pgid */ 560 msg_size += 4 + req->r_base_oid.name_len; /* oid */ 561 msg_size += 2 + req->r_num_ops * sizeof(struct ceph_osd_op); 562 msg_size += 8; /* snapid */ 563 msg_size += 8; /* snap_seq */ 564 msg_size += 4 + 8 * (req->r_snapc ? req->r_snapc->num_snaps : 0); 565 msg_size += 4; /* retry_attempt */ 566 567 if (req->r_mempool) 568 msg = ceph_msgpool_get(&osdc->msgpool_op, 0); 569 else 570 msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp, true); 571 if (!msg) 572 return -ENOMEM; 573 574 memset(msg->front.iov_base, 0, msg->front.iov_len); 575 req->r_request = msg; 576 577 /* create reply message */ 578 msg_size = OSD_OPREPLY_FRONT_LEN; 579 msg_size += req->r_base_oid.name_len; 580 msg_size += req->r_num_ops * sizeof(struct ceph_osd_op); 581 582 if (req->r_mempool) 583 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0); 584 else 585 msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, msg_size, gfp, true); 586 if (!msg) 587 return -ENOMEM; 588 589 req->r_reply = msg; 590 591 return 0; 592 } 593 EXPORT_SYMBOL(ceph_osdc_alloc_messages); 594 595 static bool osd_req_opcode_valid(u16 opcode) 596 { 597 switch (opcode) { 598 #define GENERATE_CASE(op, opcode, str) case CEPH_OSD_OP_##op: return true; 599 __CEPH_FORALL_OSD_OPS(GENERATE_CASE) 600 #undef GENERATE_CASE 601 default: 602 return false; 603 } 604 } 605 606 /* 607 * This is an osd op init function for opcodes that have no data or 608 * other information associated with them. It also serves as a 609 * common init routine for all the other init functions, below. 610 */ 611 static struct ceph_osd_req_op * 612 _osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which, 613 u16 opcode, u32 flags) 614 { 615 struct ceph_osd_req_op *op; 616 617 BUG_ON(which >= osd_req->r_num_ops); 618 BUG_ON(!osd_req_opcode_valid(opcode)); 619 620 op = &osd_req->r_ops[which]; 621 memset(op, 0, sizeof (*op)); 622 op->op = opcode; 623 op->flags = flags; 624 625 return op; 626 } 627 628 void osd_req_op_init(struct ceph_osd_request *osd_req, 629 unsigned int which, u16 opcode, u32 flags) 630 { 631 (void)_osd_req_op_init(osd_req, which, opcode, flags); 632 } 633 EXPORT_SYMBOL(osd_req_op_init); 634 635 void osd_req_op_extent_init(struct ceph_osd_request *osd_req, 636 unsigned int which, u16 opcode, 637 u64 offset, u64 length, 638 u64 truncate_size, u32 truncate_seq) 639 { 640 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, 641 opcode, 0); 642 size_t payload_len = 0; 643 644 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE && 645 opcode != CEPH_OSD_OP_WRITEFULL && opcode != CEPH_OSD_OP_ZERO && 646 opcode != CEPH_OSD_OP_TRUNCATE); 647 648 op->extent.offset = offset; 649 op->extent.length = length; 650 op->extent.truncate_size = truncate_size; 651 op->extent.truncate_seq = truncate_seq; 652 if (opcode == CEPH_OSD_OP_WRITE || opcode == CEPH_OSD_OP_WRITEFULL) 653 payload_len += length; 654 655 op->indata_len = payload_len; 656 } 657 EXPORT_SYMBOL(osd_req_op_extent_init); 658 659 void osd_req_op_extent_update(struct ceph_osd_request *osd_req, 660 unsigned int which, u64 length) 661 { 662 struct ceph_osd_req_op *op; 663 u64 previous; 664 665 BUG_ON(which >= osd_req->r_num_ops); 666 op = &osd_req->r_ops[which]; 667 previous = op->extent.length; 668 669 if (length == previous) 670 return; /* Nothing to do */ 671 BUG_ON(length > previous); 672 673 op->extent.length = length; 674 if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL) 675 op->indata_len -= previous - length; 676 } 677 EXPORT_SYMBOL(osd_req_op_extent_update); 678 679 void osd_req_op_extent_dup_last(struct ceph_osd_request *osd_req, 680 unsigned int which, u64 offset_inc) 681 { 682 struct ceph_osd_req_op *op, *prev_op; 683 684 BUG_ON(which + 1 >= osd_req->r_num_ops); 685 686 prev_op = &osd_req->r_ops[which]; 687 op = _osd_req_op_init(osd_req, which + 1, prev_op->op, prev_op->flags); 688 /* dup previous one */ 689 op->indata_len = prev_op->indata_len; 690 op->outdata_len = prev_op->outdata_len; 691 op->extent = prev_op->extent; 692 /* adjust offset */ 693 op->extent.offset += offset_inc; 694 op->extent.length -= offset_inc; 695 696 if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL) 697 op->indata_len -= offset_inc; 698 } 699 EXPORT_SYMBOL(osd_req_op_extent_dup_last); 700 701 void osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which, 702 u16 opcode, const char *class, const char *method) 703 { 704 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, 705 opcode, 0); 706 struct ceph_pagelist *pagelist; 707 size_t payload_len = 0; 708 size_t size; 709 710 BUG_ON(opcode != CEPH_OSD_OP_CALL); 711 712 pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS); 713 BUG_ON(!pagelist); 714 ceph_pagelist_init(pagelist); 715 716 op->cls.class_name = class; 717 size = strlen(class); 718 BUG_ON(size > (size_t) U8_MAX); 719 op->cls.class_len = size; 720 ceph_pagelist_append(pagelist, class, size); 721 payload_len += size; 722 723 op->cls.method_name = method; 724 size = strlen(method); 725 BUG_ON(size > (size_t) U8_MAX); 726 op->cls.method_len = size; 727 ceph_pagelist_append(pagelist, method, size); 728 payload_len += size; 729 730 osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist); 731 732 op->indata_len = payload_len; 733 } 734 EXPORT_SYMBOL(osd_req_op_cls_init); 735 736 int osd_req_op_xattr_init(struct ceph_osd_request *osd_req, unsigned int which, 737 u16 opcode, const char *name, const void *value, 738 size_t size, u8 cmp_op, u8 cmp_mode) 739 { 740 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, 741 opcode, 0); 742 struct ceph_pagelist *pagelist; 743 size_t payload_len; 744 745 BUG_ON(opcode != CEPH_OSD_OP_SETXATTR && opcode != CEPH_OSD_OP_CMPXATTR); 746 747 pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS); 748 if (!pagelist) 749 return -ENOMEM; 750 751 ceph_pagelist_init(pagelist); 752 753 payload_len = strlen(name); 754 op->xattr.name_len = payload_len; 755 ceph_pagelist_append(pagelist, name, payload_len); 756 757 op->xattr.value_len = size; 758 ceph_pagelist_append(pagelist, value, size); 759 payload_len += size; 760 761 op->xattr.cmp_op = cmp_op; 762 op->xattr.cmp_mode = cmp_mode; 763 764 ceph_osd_data_pagelist_init(&op->xattr.osd_data, pagelist); 765 op->indata_len = payload_len; 766 return 0; 767 } 768 EXPORT_SYMBOL(osd_req_op_xattr_init); 769 770 /* 771 * @watch_opcode: CEPH_OSD_WATCH_OP_* 772 */ 773 static void osd_req_op_watch_init(struct ceph_osd_request *req, int which, 774 u64 cookie, u8 watch_opcode) 775 { 776 struct ceph_osd_req_op *op; 777 778 op = _osd_req_op_init(req, which, CEPH_OSD_OP_WATCH, 0); 779 op->watch.cookie = cookie; 780 op->watch.op = watch_opcode; 781 op->watch.gen = 0; 782 } 783 784 void osd_req_op_alloc_hint_init(struct ceph_osd_request *osd_req, 785 unsigned int which, 786 u64 expected_object_size, 787 u64 expected_write_size) 788 { 789 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, 790 CEPH_OSD_OP_SETALLOCHINT, 791 0); 792 793 op->alloc_hint.expected_object_size = expected_object_size; 794 op->alloc_hint.expected_write_size = expected_write_size; 795 796 /* 797 * CEPH_OSD_OP_SETALLOCHINT op is advisory and therefore deemed 798 * not worth a feature bit. Set FAILOK per-op flag to make 799 * sure older osds don't trip over an unsupported opcode. 800 */ 801 op->flags |= CEPH_OSD_OP_FLAG_FAILOK; 802 } 803 EXPORT_SYMBOL(osd_req_op_alloc_hint_init); 804 805 static void ceph_osdc_msg_data_add(struct ceph_msg *msg, 806 struct ceph_osd_data *osd_data) 807 { 808 u64 length = ceph_osd_data_length(osd_data); 809 810 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) { 811 BUG_ON(length > (u64) SIZE_MAX); 812 if (length) 813 ceph_msg_data_add_pages(msg, osd_data->pages, 814 length, osd_data->alignment); 815 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) { 816 BUG_ON(!length); 817 ceph_msg_data_add_pagelist(msg, osd_data->pagelist); 818 #ifdef CONFIG_BLOCK 819 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) { 820 ceph_msg_data_add_bio(msg, osd_data->bio, length); 821 #endif 822 } else { 823 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE); 824 } 825 } 826 827 static u32 osd_req_encode_op(struct ceph_osd_op *dst, 828 const struct ceph_osd_req_op *src) 829 { 830 if (WARN_ON(!osd_req_opcode_valid(src->op))) { 831 pr_err("unrecognized osd opcode %d\n", src->op); 832 833 return 0; 834 } 835 836 switch (src->op) { 837 case CEPH_OSD_OP_STAT: 838 break; 839 case CEPH_OSD_OP_READ: 840 case CEPH_OSD_OP_WRITE: 841 case CEPH_OSD_OP_WRITEFULL: 842 case CEPH_OSD_OP_ZERO: 843 case CEPH_OSD_OP_TRUNCATE: 844 dst->extent.offset = cpu_to_le64(src->extent.offset); 845 dst->extent.length = cpu_to_le64(src->extent.length); 846 dst->extent.truncate_size = 847 cpu_to_le64(src->extent.truncate_size); 848 dst->extent.truncate_seq = 849 cpu_to_le32(src->extent.truncate_seq); 850 break; 851 case CEPH_OSD_OP_CALL: 852 dst->cls.class_len = src->cls.class_len; 853 dst->cls.method_len = src->cls.method_len; 854 dst->cls.indata_len = cpu_to_le32(src->cls.indata_len); 855 break; 856 case CEPH_OSD_OP_STARTSYNC: 857 break; 858 case CEPH_OSD_OP_WATCH: 859 dst->watch.cookie = cpu_to_le64(src->watch.cookie); 860 dst->watch.ver = cpu_to_le64(0); 861 dst->watch.op = src->watch.op; 862 dst->watch.gen = cpu_to_le32(src->watch.gen); 863 break; 864 case CEPH_OSD_OP_NOTIFY_ACK: 865 break; 866 case CEPH_OSD_OP_NOTIFY: 867 dst->notify.cookie = cpu_to_le64(src->notify.cookie); 868 break; 869 case CEPH_OSD_OP_LIST_WATCHERS: 870 break; 871 case CEPH_OSD_OP_SETALLOCHINT: 872 dst->alloc_hint.expected_object_size = 873 cpu_to_le64(src->alloc_hint.expected_object_size); 874 dst->alloc_hint.expected_write_size = 875 cpu_to_le64(src->alloc_hint.expected_write_size); 876 break; 877 case CEPH_OSD_OP_SETXATTR: 878 case CEPH_OSD_OP_CMPXATTR: 879 dst->xattr.name_len = cpu_to_le32(src->xattr.name_len); 880 dst->xattr.value_len = cpu_to_le32(src->xattr.value_len); 881 dst->xattr.cmp_op = src->xattr.cmp_op; 882 dst->xattr.cmp_mode = src->xattr.cmp_mode; 883 break; 884 case CEPH_OSD_OP_CREATE: 885 case CEPH_OSD_OP_DELETE: 886 break; 887 default: 888 pr_err("unsupported osd opcode %s\n", 889 ceph_osd_op_name(src->op)); 890 WARN_ON(1); 891 892 return 0; 893 } 894 895 dst->op = cpu_to_le16(src->op); 896 dst->flags = cpu_to_le32(src->flags); 897 dst->payload_len = cpu_to_le32(src->indata_len); 898 899 return src->indata_len; 900 } 901 902 /* 903 * build new request AND message, calculate layout, and adjust file 904 * extent as needed. 905 * 906 * if the file was recently truncated, we include information about its 907 * old and new size so that the object can be updated appropriately. (we 908 * avoid synchronously deleting truncated objects because it's slow.) 909 * 910 * if @do_sync, include a 'startsync' command so that the osd will flush 911 * data quickly. 912 */ 913 struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc, 914 struct ceph_file_layout *layout, 915 struct ceph_vino vino, 916 u64 off, u64 *plen, 917 unsigned int which, int num_ops, 918 int opcode, int flags, 919 struct ceph_snap_context *snapc, 920 u32 truncate_seq, 921 u64 truncate_size, 922 bool use_mempool) 923 { 924 struct ceph_osd_request *req; 925 u64 objnum = 0; 926 u64 objoff = 0; 927 u64 objlen = 0; 928 int r; 929 930 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE && 931 opcode != CEPH_OSD_OP_ZERO && opcode != CEPH_OSD_OP_TRUNCATE && 932 opcode != CEPH_OSD_OP_CREATE && opcode != CEPH_OSD_OP_DELETE); 933 934 req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool, 935 GFP_NOFS); 936 if (!req) { 937 r = -ENOMEM; 938 goto fail; 939 } 940 941 /* calculate max write size */ 942 r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen); 943 if (r) 944 goto fail; 945 946 if (opcode == CEPH_OSD_OP_CREATE || opcode == CEPH_OSD_OP_DELETE) { 947 osd_req_op_init(req, which, opcode, 0); 948 } else { 949 u32 object_size = layout->object_size; 950 u32 object_base = off - objoff; 951 if (!(truncate_seq == 1 && truncate_size == -1ULL)) { 952 if (truncate_size <= object_base) { 953 truncate_size = 0; 954 } else { 955 truncate_size -= object_base; 956 if (truncate_size > object_size) 957 truncate_size = object_size; 958 } 959 } 960 osd_req_op_extent_init(req, which, opcode, objoff, objlen, 961 truncate_size, truncate_seq); 962 } 963 964 req->r_flags = flags; 965 req->r_base_oloc.pool = layout->pool_id; 966 req->r_base_oloc.pool_ns = ceph_try_get_string(layout->pool_ns); 967 ceph_oid_printf(&req->r_base_oid, "%llx.%08llx", vino.ino, objnum); 968 969 req->r_snapid = vino.snap; 970 if (flags & CEPH_OSD_FLAG_WRITE) 971 req->r_data_offset = off; 972 973 r = ceph_osdc_alloc_messages(req, GFP_NOFS); 974 if (r) 975 goto fail; 976 977 return req; 978 979 fail: 980 ceph_osdc_put_request(req); 981 return ERR_PTR(r); 982 } 983 EXPORT_SYMBOL(ceph_osdc_new_request); 984 985 /* 986 * We keep osd requests in an rbtree, sorted by ->r_tid. 987 */ 988 DEFINE_RB_FUNCS(request, struct ceph_osd_request, r_tid, r_node) 989 DEFINE_RB_FUNCS(request_mc, struct ceph_osd_request, r_tid, r_mc_node) 990 991 static bool osd_homeless(struct ceph_osd *osd) 992 { 993 return osd->o_osd == CEPH_HOMELESS_OSD; 994 } 995 996 static bool osd_registered(struct ceph_osd *osd) 997 { 998 verify_osdc_locked(osd->o_osdc); 999 1000 return !RB_EMPTY_NODE(&osd->o_node); 1001 } 1002 1003 /* 1004 * Assumes @osd is zero-initialized. 1005 */ 1006 static void osd_init(struct ceph_osd *osd) 1007 { 1008 refcount_set(&osd->o_ref, 1); 1009 RB_CLEAR_NODE(&osd->o_node); 1010 osd->o_requests = RB_ROOT; 1011 osd->o_linger_requests = RB_ROOT; 1012 INIT_LIST_HEAD(&osd->o_osd_lru); 1013 INIT_LIST_HEAD(&osd->o_keepalive_item); 1014 osd->o_incarnation = 1; 1015 mutex_init(&osd->lock); 1016 } 1017 1018 static void osd_cleanup(struct ceph_osd *osd) 1019 { 1020 WARN_ON(!RB_EMPTY_NODE(&osd->o_node)); 1021 WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests)); 1022 WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests)); 1023 WARN_ON(!list_empty(&osd->o_osd_lru)); 1024 WARN_ON(!list_empty(&osd->o_keepalive_item)); 1025 1026 if (osd->o_auth.authorizer) { 1027 WARN_ON(osd_homeless(osd)); 1028 ceph_auth_destroy_authorizer(osd->o_auth.authorizer); 1029 } 1030 } 1031 1032 /* 1033 * Track open sessions with osds. 1034 */ 1035 static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum) 1036 { 1037 struct ceph_osd *osd; 1038 1039 WARN_ON(onum == CEPH_HOMELESS_OSD); 1040 1041 osd = kzalloc(sizeof(*osd), GFP_NOIO | __GFP_NOFAIL); 1042 osd_init(osd); 1043 osd->o_osdc = osdc; 1044 osd->o_osd = onum; 1045 1046 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr); 1047 1048 return osd; 1049 } 1050 1051 static struct ceph_osd *get_osd(struct ceph_osd *osd) 1052 { 1053 if (refcount_inc_not_zero(&osd->o_ref)) { 1054 dout("get_osd %p %d -> %d\n", osd, refcount_read(&osd->o_ref)-1, 1055 refcount_read(&osd->o_ref)); 1056 return osd; 1057 } else { 1058 dout("get_osd %p FAIL\n", osd); 1059 return NULL; 1060 } 1061 } 1062 1063 static void put_osd(struct ceph_osd *osd) 1064 { 1065 dout("put_osd %p %d -> %d\n", osd, refcount_read(&osd->o_ref), 1066 refcount_read(&osd->o_ref) - 1); 1067 if (refcount_dec_and_test(&osd->o_ref)) { 1068 osd_cleanup(osd); 1069 kfree(osd); 1070 } 1071 } 1072 1073 DEFINE_RB_FUNCS(osd, struct ceph_osd, o_osd, o_node) 1074 1075 static void __move_osd_to_lru(struct ceph_osd *osd) 1076 { 1077 struct ceph_osd_client *osdc = osd->o_osdc; 1078 1079 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd); 1080 BUG_ON(!list_empty(&osd->o_osd_lru)); 1081 1082 spin_lock(&osdc->osd_lru_lock); 1083 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru); 1084 spin_unlock(&osdc->osd_lru_lock); 1085 1086 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl; 1087 } 1088 1089 static void maybe_move_osd_to_lru(struct ceph_osd *osd) 1090 { 1091 if (RB_EMPTY_ROOT(&osd->o_requests) && 1092 RB_EMPTY_ROOT(&osd->o_linger_requests)) 1093 __move_osd_to_lru(osd); 1094 } 1095 1096 static void __remove_osd_from_lru(struct ceph_osd *osd) 1097 { 1098 struct ceph_osd_client *osdc = osd->o_osdc; 1099 1100 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd); 1101 1102 spin_lock(&osdc->osd_lru_lock); 1103 if (!list_empty(&osd->o_osd_lru)) 1104 list_del_init(&osd->o_osd_lru); 1105 spin_unlock(&osdc->osd_lru_lock); 1106 } 1107 1108 /* 1109 * Close the connection and assign any leftover requests to the 1110 * homeless session. 1111 */ 1112 static void close_osd(struct ceph_osd *osd) 1113 { 1114 struct ceph_osd_client *osdc = osd->o_osdc; 1115 struct rb_node *n; 1116 1117 verify_osdc_wrlocked(osdc); 1118 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd); 1119 1120 ceph_con_close(&osd->o_con); 1121 1122 for (n = rb_first(&osd->o_requests); n; ) { 1123 struct ceph_osd_request *req = 1124 rb_entry(n, struct ceph_osd_request, r_node); 1125 1126 n = rb_next(n); /* unlink_request() */ 1127 1128 dout(" reassigning req %p tid %llu\n", req, req->r_tid); 1129 unlink_request(osd, req); 1130 link_request(&osdc->homeless_osd, req); 1131 } 1132 for (n = rb_first(&osd->o_linger_requests); n; ) { 1133 struct ceph_osd_linger_request *lreq = 1134 rb_entry(n, struct ceph_osd_linger_request, node); 1135 1136 n = rb_next(n); /* unlink_linger() */ 1137 1138 dout(" reassigning lreq %p linger_id %llu\n", lreq, 1139 lreq->linger_id); 1140 unlink_linger(osd, lreq); 1141 link_linger(&osdc->homeless_osd, lreq); 1142 } 1143 1144 __remove_osd_from_lru(osd); 1145 erase_osd(&osdc->osds, osd); 1146 put_osd(osd); 1147 } 1148 1149 /* 1150 * reset osd connect 1151 */ 1152 static int reopen_osd(struct ceph_osd *osd) 1153 { 1154 struct ceph_entity_addr *peer_addr; 1155 1156 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd); 1157 1158 if (RB_EMPTY_ROOT(&osd->o_requests) && 1159 RB_EMPTY_ROOT(&osd->o_linger_requests)) { 1160 close_osd(osd); 1161 return -ENODEV; 1162 } 1163 1164 peer_addr = &osd->o_osdc->osdmap->osd_addr[osd->o_osd]; 1165 if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) && 1166 !ceph_con_opened(&osd->o_con)) { 1167 struct rb_node *n; 1168 1169 dout("osd addr hasn't changed and connection never opened, " 1170 "letting msgr retry\n"); 1171 /* touch each r_stamp for handle_timeout()'s benfit */ 1172 for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) { 1173 struct ceph_osd_request *req = 1174 rb_entry(n, struct ceph_osd_request, r_node); 1175 req->r_stamp = jiffies; 1176 } 1177 1178 return -EAGAIN; 1179 } 1180 1181 ceph_con_close(&osd->o_con); 1182 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr); 1183 osd->o_incarnation++; 1184 1185 return 0; 1186 } 1187 1188 static struct ceph_osd *lookup_create_osd(struct ceph_osd_client *osdc, int o, 1189 bool wrlocked) 1190 { 1191 struct ceph_osd *osd; 1192 1193 if (wrlocked) 1194 verify_osdc_wrlocked(osdc); 1195 else 1196 verify_osdc_locked(osdc); 1197 1198 if (o != CEPH_HOMELESS_OSD) 1199 osd = lookup_osd(&osdc->osds, o); 1200 else 1201 osd = &osdc->homeless_osd; 1202 if (!osd) { 1203 if (!wrlocked) 1204 return ERR_PTR(-EAGAIN); 1205 1206 osd = create_osd(osdc, o); 1207 insert_osd(&osdc->osds, osd); 1208 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, 1209 &osdc->osdmap->osd_addr[osd->o_osd]); 1210 } 1211 1212 dout("%s osdc %p osd%d -> osd %p\n", __func__, osdc, o, osd); 1213 return osd; 1214 } 1215 1216 /* 1217 * Create request <-> OSD session relation. 1218 * 1219 * @req has to be assigned a tid, @osd may be homeless. 1220 */ 1221 static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req) 1222 { 1223 verify_osd_locked(osd); 1224 WARN_ON(!req->r_tid || req->r_osd); 1225 dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd, 1226 req, req->r_tid); 1227 1228 if (!osd_homeless(osd)) 1229 __remove_osd_from_lru(osd); 1230 else 1231 atomic_inc(&osd->o_osdc->num_homeless); 1232 1233 get_osd(osd); 1234 insert_request(&osd->o_requests, req); 1235 req->r_osd = osd; 1236 } 1237 1238 static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req) 1239 { 1240 verify_osd_locked(osd); 1241 WARN_ON(req->r_osd != osd); 1242 dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd, 1243 req, req->r_tid); 1244 1245 req->r_osd = NULL; 1246 erase_request(&osd->o_requests, req); 1247 put_osd(osd); 1248 1249 if (!osd_homeless(osd)) 1250 maybe_move_osd_to_lru(osd); 1251 else 1252 atomic_dec(&osd->o_osdc->num_homeless); 1253 } 1254 1255 static bool __pool_full(struct ceph_pg_pool_info *pi) 1256 { 1257 return pi->flags & CEPH_POOL_FLAG_FULL; 1258 } 1259 1260 static bool have_pool_full(struct ceph_osd_client *osdc) 1261 { 1262 struct rb_node *n; 1263 1264 for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) { 1265 struct ceph_pg_pool_info *pi = 1266 rb_entry(n, struct ceph_pg_pool_info, node); 1267 1268 if (__pool_full(pi)) 1269 return true; 1270 } 1271 1272 return false; 1273 } 1274 1275 static bool pool_full(struct ceph_osd_client *osdc, s64 pool_id) 1276 { 1277 struct ceph_pg_pool_info *pi; 1278 1279 pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id); 1280 if (!pi) 1281 return false; 1282 1283 return __pool_full(pi); 1284 } 1285 1286 /* 1287 * Returns whether a request should be blocked from being sent 1288 * based on the current osdmap and osd_client settings. 1289 */ 1290 static bool target_should_be_paused(struct ceph_osd_client *osdc, 1291 const struct ceph_osd_request_target *t, 1292 struct ceph_pg_pool_info *pi) 1293 { 1294 bool pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD); 1295 bool pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) || 1296 ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || 1297 __pool_full(pi); 1298 1299 WARN_ON(pi->id != t->base_oloc.pool); 1300 return (t->flags & CEPH_OSD_FLAG_READ && pauserd) || 1301 (t->flags & CEPH_OSD_FLAG_WRITE && pausewr); 1302 } 1303 1304 enum calc_target_result { 1305 CALC_TARGET_NO_ACTION = 0, 1306 CALC_TARGET_NEED_RESEND, 1307 CALC_TARGET_POOL_DNE, 1308 }; 1309 1310 static enum calc_target_result calc_target(struct ceph_osd_client *osdc, 1311 struct ceph_osd_request_target *t, 1312 u32 *last_force_resend, 1313 bool any_change) 1314 { 1315 struct ceph_pg_pool_info *pi; 1316 struct ceph_pg pgid, last_pgid; 1317 struct ceph_osds up, acting; 1318 bool force_resend = false; 1319 bool need_check_tiering = false; 1320 bool need_resend = false; 1321 bool sort_bitwise = ceph_osdmap_flag(osdc, CEPH_OSDMAP_SORTBITWISE); 1322 enum calc_target_result ct_res; 1323 int ret; 1324 1325 pi = ceph_pg_pool_by_id(osdc->osdmap, t->base_oloc.pool); 1326 if (!pi) { 1327 t->osd = CEPH_HOMELESS_OSD; 1328 ct_res = CALC_TARGET_POOL_DNE; 1329 goto out; 1330 } 1331 1332 if (osdc->osdmap->epoch == pi->last_force_request_resend) { 1333 if (last_force_resend && 1334 *last_force_resend < pi->last_force_request_resend) { 1335 *last_force_resend = pi->last_force_request_resend; 1336 force_resend = true; 1337 } else if (!last_force_resend) { 1338 force_resend = true; 1339 } 1340 } 1341 if (ceph_oid_empty(&t->target_oid) || force_resend) { 1342 ceph_oid_copy(&t->target_oid, &t->base_oid); 1343 need_check_tiering = true; 1344 } 1345 if (ceph_oloc_empty(&t->target_oloc) || force_resend) { 1346 ceph_oloc_copy(&t->target_oloc, &t->base_oloc); 1347 need_check_tiering = true; 1348 } 1349 1350 if (need_check_tiering && 1351 (t->flags & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0) { 1352 if (t->flags & CEPH_OSD_FLAG_READ && pi->read_tier >= 0) 1353 t->target_oloc.pool = pi->read_tier; 1354 if (t->flags & CEPH_OSD_FLAG_WRITE && pi->write_tier >= 0) 1355 t->target_oloc.pool = pi->write_tier; 1356 } 1357 1358 ret = ceph_object_locator_to_pg(osdc->osdmap, &t->target_oid, 1359 &t->target_oloc, &pgid); 1360 if (ret) { 1361 WARN_ON(ret != -ENOENT); 1362 t->osd = CEPH_HOMELESS_OSD; 1363 ct_res = CALC_TARGET_POOL_DNE; 1364 goto out; 1365 } 1366 last_pgid.pool = pgid.pool; 1367 last_pgid.seed = ceph_stable_mod(pgid.seed, t->pg_num, t->pg_num_mask); 1368 1369 ceph_pg_to_up_acting_osds(osdc->osdmap, &pgid, &up, &acting); 1370 if (any_change && 1371 ceph_is_new_interval(&t->acting, 1372 &acting, 1373 &t->up, 1374 &up, 1375 t->size, 1376 pi->size, 1377 t->min_size, 1378 pi->min_size, 1379 t->pg_num, 1380 pi->pg_num, 1381 t->sort_bitwise, 1382 sort_bitwise, 1383 &last_pgid)) 1384 force_resend = true; 1385 1386 if (t->paused && !target_should_be_paused(osdc, t, pi)) { 1387 t->paused = false; 1388 need_resend = true; 1389 } 1390 1391 if (ceph_pg_compare(&t->pgid, &pgid) || 1392 ceph_osds_changed(&t->acting, &acting, any_change) || 1393 force_resend) { 1394 t->pgid = pgid; /* struct */ 1395 ceph_osds_copy(&t->acting, &acting); 1396 ceph_osds_copy(&t->up, &up); 1397 t->size = pi->size; 1398 t->min_size = pi->min_size; 1399 t->pg_num = pi->pg_num; 1400 t->pg_num_mask = pi->pg_num_mask; 1401 t->sort_bitwise = sort_bitwise; 1402 1403 t->osd = acting.primary; 1404 need_resend = true; 1405 } 1406 1407 ct_res = need_resend ? CALC_TARGET_NEED_RESEND : CALC_TARGET_NO_ACTION; 1408 out: 1409 dout("%s t %p -> ct_res %d osd %d\n", __func__, t, ct_res, t->osd); 1410 return ct_res; 1411 } 1412 1413 static void setup_request_data(struct ceph_osd_request *req, 1414 struct ceph_msg *msg) 1415 { 1416 u32 data_len = 0; 1417 int i; 1418 1419 if (!list_empty(&msg->data)) 1420 return; 1421 1422 WARN_ON(msg->data_length); 1423 for (i = 0; i < req->r_num_ops; i++) { 1424 struct ceph_osd_req_op *op = &req->r_ops[i]; 1425 1426 switch (op->op) { 1427 /* request */ 1428 case CEPH_OSD_OP_WRITE: 1429 case CEPH_OSD_OP_WRITEFULL: 1430 WARN_ON(op->indata_len != op->extent.length); 1431 ceph_osdc_msg_data_add(msg, &op->extent.osd_data); 1432 break; 1433 case CEPH_OSD_OP_SETXATTR: 1434 case CEPH_OSD_OP_CMPXATTR: 1435 WARN_ON(op->indata_len != op->xattr.name_len + 1436 op->xattr.value_len); 1437 ceph_osdc_msg_data_add(msg, &op->xattr.osd_data); 1438 break; 1439 case CEPH_OSD_OP_NOTIFY_ACK: 1440 ceph_osdc_msg_data_add(msg, 1441 &op->notify_ack.request_data); 1442 break; 1443 1444 /* reply */ 1445 case CEPH_OSD_OP_STAT: 1446 ceph_osdc_msg_data_add(req->r_reply, 1447 &op->raw_data_in); 1448 break; 1449 case CEPH_OSD_OP_READ: 1450 ceph_osdc_msg_data_add(req->r_reply, 1451 &op->extent.osd_data); 1452 break; 1453 case CEPH_OSD_OP_LIST_WATCHERS: 1454 ceph_osdc_msg_data_add(req->r_reply, 1455 &op->list_watchers.response_data); 1456 break; 1457 1458 /* both */ 1459 case CEPH_OSD_OP_CALL: 1460 WARN_ON(op->indata_len != op->cls.class_len + 1461 op->cls.method_len + 1462 op->cls.indata_len); 1463 ceph_osdc_msg_data_add(msg, &op->cls.request_info); 1464 /* optional, can be NONE */ 1465 ceph_osdc_msg_data_add(msg, &op->cls.request_data); 1466 /* optional, can be NONE */ 1467 ceph_osdc_msg_data_add(req->r_reply, 1468 &op->cls.response_data); 1469 break; 1470 case CEPH_OSD_OP_NOTIFY: 1471 ceph_osdc_msg_data_add(msg, 1472 &op->notify.request_data); 1473 ceph_osdc_msg_data_add(req->r_reply, 1474 &op->notify.response_data); 1475 break; 1476 } 1477 1478 data_len += op->indata_len; 1479 } 1480 1481 WARN_ON(data_len != msg->data_length); 1482 } 1483 1484 static void encode_request(struct ceph_osd_request *req, struct ceph_msg *msg) 1485 { 1486 void *p = msg->front.iov_base; 1487 void *const end = p + msg->front_alloc_len; 1488 u32 data_len = 0; 1489 int i; 1490 1491 if (req->r_flags & CEPH_OSD_FLAG_WRITE) { 1492 /* snapshots aren't writeable */ 1493 WARN_ON(req->r_snapid != CEPH_NOSNAP); 1494 } else { 1495 WARN_ON(req->r_mtime.tv_sec || req->r_mtime.tv_nsec || 1496 req->r_data_offset || req->r_snapc); 1497 } 1498 1499 setup_request_data(req, msg); 1500 1501 ceph_encode_32(&p, 1); /* client_inc, always 1 */ 1502 ceph_encode_32(&p, req->r_osdc->osdmap->epoch); 1503 ceph_encode_32(&p, req->r_flags); 1504 ceph_encode_timespec(p, &req->r_mtime); 1505 p += sizeof(struct ceph_timespec); 1506 1507 /* reassert_version */ 1508 memset(p, 0, sizeof(struct ceph_eversion)); 1509 p += sizeof(struct ceph_eversion); 1510 1511 /* oloc */ 1512 ceph_start_encoding(&p, 5, 4, 1513 ceph_oloc_encoding_size(&req->r_t.target_oloc)); 1514 ceph_encode_64(&p, req->r_t.target_oloc.pool); 1515 ceph_encode_32(&p, -1); /* preferred */ 1516 ceph_encode_32(&p, 0); /* key len */ 1517 if (req->r_t.target_oloc.pool_ns) 1518 ceph_encode_string(&p, end, req->r_t.target_oloc.pool_ns->str, 1519 req->r_t.target_oloc.pool_ns->len); 1520 else 1521 ceph_encode_32(&p, 0); 1522 1523 /* pgid */ 1524 ceph_encode_8(&p, 1); 1525 ceph_encode_64(&p, req->r_t.pgid.pool); 1526 ceph_encode_32(&p, req->r_t.pgid.seed); 1527 ceph_encode_32(&p, -1); /* preferred */ 1528 1529 /* oid */ 1530 ceph_encode_32(&p, req->r_t.target_oid.name_len); 1531 memcpy(p, req->r_t.target_oid.name, req->r_t.target_oid.name_len); 1532 p += req->r_t.target_oid.name_len; 1533 1534 /* ops, can imply data */ 1535 ceph_encode_16(&p, req->r_num_ops); 1536 for (i = 0; i < req->r_num_ops; i++) { 1537 data_len += osd_req_encode_op(p, &req->r_ops[i]); 1538 p += sizeof(struct ceph_osd_op); 1539 } 1540 1541 ceph_encode_64(&p, req->r_snapid); /* snapid */ 1542 if (req->r_snapc) { 1543 ceph_encode_64(&p, req->r_snapc->seq); 1544 ceph_encode_32(&p, req->r_snapc->num_snaps); 1545 for (i = 0; i < req->r_snapc->num_snaps; i++) 1546 ceph_encode_64(&p, req->r_snapc->snaps[i]); 1547 } else { 1548 ceph_encode_64(&p, 0); /* snap_seq */ 1549 ceph_encode_32(&p, 0); /* snaps len */ 1550 } 1551 1552 ceph_encode_32(&p, req->r_attempts); /* retry_attempt */ 1553 1554 BUG_ON(p > end); 1555 msg->front.iov_len = p - msg->front.iov_base; 1556 msg->hdr.version = cpu_to_le16(4); /* MOSDOp v4 */ 1557 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len); 1558 msg->hdr.data_len = cpu_to_le32(data_len); 1559 /* 1560 * The header "data_off" is a hint to the receiver allowing it 1561 * to align received data into its buffers such that there's no 1562 * need to re-copy it before writing it to disk (direct I/O). 1563 */ 1564 msg->hdr.data_off = cpu_to_le16(req->r_data_offset); 1565 1566 dout("%s req %p oid %s oid_len %d front %zu data %u\n", __func__, 1567 req, req->r_t.target_oid.name, req->r_t.target_oid.name_len, 1568 msg->front.iov_len, data_len); 1569 } 1570 1571 /* 1572 * @req has to be assigned a tid and registered. 1573 */ 1574 static void send_request(struct ceph_osd_request *req) 1575 { 1576 struct ceph_osd *osd = req->r_osd; 1577 1578 verify_osd_locked(osd); 1579 WARN_ON(osd->o_osd != req->r_t.osd); 1580 1581 /* 1582 * We may have a previously queued request message hanging 1583 * around. Cancel it to avoid corrupting the msgr. 1584 */ 1585 if (req->r_sent) 1586 ceph_msg_revoke(req->r_request); 1587 1588 req->r_flags |= CEPH_OSD_FLAG_KNOWN_REDIR; 1589 if (req->r_attempts) 1590 req->r_flags |= CEPH_OSD_FLAG_RETRY; 1591 else 1592 WARN_ON(req->r_flags & CEPH_OSD_FLAG_RETRY); 1593 1594 encode_request(req, req->r_request); 1595 1596 dout("%s req %p tid %llu to pg %llu.%x osd%d flags 0x%x attempt %d\n", 1597 __func__, req, req->r_tid, req->r_t.pgid.pool, req->r_t.pgid.seed, 1598 req->r_t.osd, req->r_flags, req->r_attempts); 1599 1600 req->r_t.paused = false; 1601 req->r_stamp = jiffies; 1602 req->r_attempts++; 1603 1604 req->r_sent = osd->o_incarnation; 1605 req->r_request->hdr.tid = cpu_to_le64(req->r_tid); 1606 ceph_con_send(&osd->o_con, ceph_msg_get(req->r_request)); 1607 } 1608 1609 static void maybe_request_map(struct ceph_osd_client *osdc) 1610 { 1611 bool continuous = false; 1612 1613 verify_osdc_locked(osdc); 1614 WARN_ON(!osdc->osdmap->epoch); 1615 1616 if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || 1617 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD) || 1618 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) { 1619 dout("%s osdc %p continuous\n", __func__, osdc); 1620 continuous = true; 1621 } else { 1622 dout("%s osdc %p onetime\n", __func__, osdc); 1623 } 1624 1625 if (ceph_monc_want_map(&osdc->client->monc, CEPH_SUB_OSDMAP, 1626 osdc->osdmap->epoch + 1, continuous)) 1627 ceph_monc_renew_subs(&osdc->client->monc); 1628 } 1629 1630 static void send_map_check(struct ceph_osd_request *req); 1631 1632 static void __submit_request(struct ceph_osd_request *req, bool wrlocked) 1633 { 1634 struct ceph_osd_client *osdc = req->r_osdc; 1635 struct ceph_osd *osd; 1636 enum calc_target_result ct_res; 1637 bool need_send = false; 1638 bool promoted = false; 1639 1640 WARN_ON(req->r_tid); 1641 dout("%s req %p wrlocked %d\n", __func__, req, wrlocked); 1642 1643 again: 1644 ct_res = calc_target(osdc, &req->r_t, &req->r_last_force_resend, false); 1645 if (ct_res == CALC_TARGET_POOL_DNE && !wrlocked) 1646 goto promote; 1647 1648 osd = lookup_create_osd(osdc, req->r_t.osd, wrlocked); 1649 if (IS_ERR(osd)) { 1650 WARN_ON(PTR_ERR(osd) != -EAGAIN || wrlocked); 1651 goto promote; 1652 } 1653 1654 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) && 1655 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) { 1656 dout("req %p pausewr\n", req); 1657 req->r_t.paused = true; 1658 maybe_request_map(osdc); 1659 } else if ((req->r_flags & CEPH_OSD_FLAG_READ) && 1660 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) { 1661 dout("req %p pauserd\n", req); 1662 req->r_t.paused = true; 1663 maybe_request_map(osdc); 1664 } else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) && 1665 !(req->r_flags & (CEPH_OSD_FLAG_FULL_TRY | 1666 CEPH_OSD_FLAG_FULL_FORCE)) && 1667 (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || 1668 pool_full(osdc, req->r_t.base_oloc.pool))) { 1669 dout("req %p full/pool_full\n", req); 1670 pr_warn_ratelimited("FULL or reached pool quota\n"); 1671 req->r_t.paused = true; 1672 maybe_request_map(osdc); 1673 } else if (!osd_homeless(osd)) { 1674 need_send = true; 1675 } else { 1676 maybe_request_map(osdc); 1677 } 1678 1679 mutex_lock(&osd->lock); 1680 /* 1681 * Assign the tid atomically with send_request() to protect 1682 * multiple writes to the same object from racing with each 1683 * other, resulting in out of order ops on the OSDs. 1684 */ 1685 req->r_tid = atomic64_inc_return(&osdc->last_tid); 1686 link_request(osd, req); 1687 if (need_send) 1688 send_request(req); 1689 mutex_unlock(&osd->lock); 1690 1691 if (ct_res == CALC_TARGET_POOL_DNE) 1692 send_map_check(req); 1693 1694 if (promoted) 1695 downgrade_write(&osdc->lock); 1696 return; 1697 1698 promote: 1699 up_read(&osdc->lock); 1700 down_write(&osdc->lock); 1701 wrlocked = true; 1702 promoted = true; 1703 goto again; 1704 } 1705 1706 static void account_request(struct ceph_osd_request *req) 1707 { 1708 WARN_ON(req->r_flags & (CEPH_OSD_FLAG_ACK | CEPH_OSD_FLAG_ONDISK)); 1709 WARN_ON(!(req->r_flags & (CEPH_OSD_FLAG_READ | CEPH_OSD_FLAG_WRITE))); 1710 1711 req->r_flags |= CEPH_OSD_FLAG_ONDISK; 1712 atomic_inc(&req->r_osdc->num_requests); 1713 1714 req->r_start_stamp = jiffies; 1715 } 1716 1717 static void submit_request(struct ceph_osd_request *req, bool wrlocked) 1718 { 1719 ceph_osdc_get_request(req); 1720 account_request(req); 1721 __submit_request(req, wrlocked); 1722 } 1723 1724 static void finish_request(struct ceph_osd_request *req) 1725 { 1726 struct ceph_osd_client *osdc = req->r_osdc; 1727 struct ceph_osd *osd = req->r_osd; 1728 1729 verify_osd_locked(osd); 1730 dout("%s req %p tid %llu\n", __func__, req, req->r_tid); 1731 1732 WARN_ON(lookup_request_mc(&osdc->map_checks, req->r_tid)); 1733 unlink_request(osd, req); 1734 atomic_dec(&osdc->num_requests); 1735 1736 /* 1737 * If an OSD has failed or returned and a request has been sent 1738 * twice, it's possible to get a reply and end up here while the 1739 * request message is queued for delivery. We will ignore the 1740 * reply, so not a big deal, but better to try and catch it. 1741 */ 1742 ceph_msg_revoke(req->r_request); 1743 ceph_msg_revoke_incoming(req->r_reply); 1744 } 1745 1746 static void __complete_request(struct ceph_osd_request *req) 1747 { 1748 if (req->r_callback) { 1749 dout("%s req %p tid %llu cb %pf result %d\n", __func__, req, 1750 req->r_tid, req->r_callback, req->r_result); 1751 req->r_callback(req); 1752 } 1753 } 1754 1755 /* 1756 * This is open-coded in handle_reply(). 1757 */ 1758 static void complete_request(struct ceph_osd_request *req, int err) 1759 { 1760 dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err); 1761 1762 req->r_result = err; 1763 finish_request(req); 1764 __complete_request(req); 1765 complete_all(&req->r_completion); 1766 ceph_osdc_put_request(req); 1767 } 1768 1769 static void cancel_map_check(struct ceph_osd_request *req) 1770 { 1771 struct ceph_osd_client *osdc = req->r_osdc; 1772 struct ceph_osd_request *lookup_req; 1773 1774 verify_osdc_wrlocked(osdc); 1775 1776 lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid); 1777 if (!lookup_req) 1778 return; 1779 1780 WARN_ON(lookup_req != req); 1781 erase_request_mc(&osdc->map_checks, req); 1782 ceph_osdc_put_request(req); 1783 } 1784 1785 static void cancel_request(struct ceph_osd_request *req) 1786 { 1787 dout("%s req %p tid %llu\n", __func__, req, req->r_tid); 1788 1789 cancel_map_check(req); 1790 finish_request(req); 1791 complete_all(&req->r_completion); 1792 ceph_osdc_put_request(req); 1793 } 1794 1795 static void abort_request(struct ceph_osd_request *req, int err) 1796 { 1797 dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err); 1798 1799 cancel_map_check(req); 1800 complete_request(req, err); 1801 } 1802 1803 static void check_pool_dne(struct ceph_osd_request *req) 1804 { 1805 struct ceph_osd_client *osdc = req->r_osdc; 1806 struct ceph_osdmap *map = osdc->osdmap; 1807 1808 verify_osdc_wrlocked(osdc); 1809 WARN_ON(!map->epoch); 1810 1811 if (req->r_attempts) { 1812 /* 1813 * We sent a request earlier, which means that 1814 * previously the pool existed, and now it does not 1815 * (i.e., it was deleted). 1816 */ 1817 req->r_map_dne_bound = map->epoch; 1818 dout("%s req %p tid %llu pool disappeared\n", __func__, req, 1819 req->r_tid); 1820 } else { 1821 dout("%s req %p tid %llu map_dne_bound %u have %u\n", __func__, 1822 req, req->r_tid, req->r_map_dne_bound, map->epoch); 1823 } 1824 1825 if (req->r_map_dne_bound) { 1826 if (map->epoch >= req->r_map_dne_bound) { 1827 /* we had a new enough map */ 1828 pr_info_ratelimited("tid %llu pool does not exist\n", 1829 req->r_tid); 1830 complete_request(req, -ENOENT); 1831 } 1832 } else { 1833 send_map_check(req); 1834 } 1835 } 1836 1837 static void map_check_cb(struct ceph_mon_generic_request *greq) 1838 { 1839 struct ceph_osd_client *osdc = &greq->monc->client->osdc; 1840 struct ceph_osd_request *req; 1841 u64 tid = greq->private_data; 1842 1843 WARN_ON(greq->result || !greq->u.newest); 1844 1845 down_write(&osdc->lock); 1846 req = lookup_request_mc(&osdc->map_checks, tid); 1847 if (!req) { 1848 dout("%s tid %llu dne\n", __func__, tid); 1849 goto out_unlock; 1850 } 1851 1852 dout("%s req %p tid %llu map_dne_bound %u newest %llu\n", __func__, 1853 req, req->r_tid, req->r_map_dne_bound, greq->u.newest); 1854 if (!req->r_map_dne_bound) 1855 req->r_map_dne_bound = greq->u.newest; 1856 erase_request_mc(&osdc->map_checks, req); 1857 check_pool_dne(req); 1858 1859 ceph_osdc_put_request(req); 1860 out_unlock: 1861 up_write(&osdc->lock); 1862 } 1863 1864 static void send_map_check(struct ceph_osd_request *req) 1865 { 1866 struct ceph_osd_client *osdc = req->r_osdc; 1867 struct ceph_osd_request *lookup_req; 1868 int ret; 1869 1870 verify_osdc_wrlocked(osdc); 1871 1872 lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid); 1873 if (lookup_req) { 1874 WARN_ON(lookup_req != req); 1875 return; 1876 } 1877 1878 ceph_osdc_get_request(req); 1879 insert_request_mc(&osdc->map_checks, req); 1880 ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap", 1881 map_check_cb, req->r_tid); 1882 WARN_ON(ret); 1883 } 1884 1885 /* 1886 * lingering requests, watch/notify v2 infrastructure 1887 */ 1888 static void linger_release(struct kref *kref) 1889 { 1890 struct ceph_osd_linger_request *lreq = 1891 container_of(kref, struct ceph_osd_linger_request, kref); 1892 1893 dout("%s lreq %p reg_req %p ping_req %p\n", __func__, lreq, 1894 lreq->reg_req, lreq->ping_req); 1895 WARN_ON(!RB_EMPTY_NODE(&lreq->node)); 1896 WARN_ON(!RB_EMPTY_NODE(&lreq->osdc_node)); 1897 WARN_ON(!RB_EMPTY_NODE(&lreq->mc_node)); 1898 WARN_ON(!list_empty(&lreq->scan_item)); 1899 WARN_ON(!list_empty(&lreq->pending_lworks)); 1900 WARN_ON(lreq->osd); 1901 1902 if (lreq->reg_req) 1903 ceph_osdc_put_request(lreq->reg_req); 1904 if (lreq->ping_req) 1905 ceph_osdc_put_request(lreq->ping_req); 1906 target_destroy(&lreq->t); 1907 kfree(lreq); 1908 } 1909 1910 static void linger_put(struct ceph_osd_linger_request *lreq) 1911 { 1912 if (lreq) 1913 kref_put(&lreq->kref, linger_release); 1914 } 1915 1916 static struct ceph_osd_linger_request * 1917 linger_get(struct ceph_osd_linger_request *lreq) 1918 { 1919 kref_get(&lreq->kref); 1920 return lreq; 1921 } 1922 1923 static struct ceph_osd_linger_request * 1924 linger_alloc(struct ceph_osd_client *osdc) 1925 { 1926 struct ceph_osd_linger_request *lreq; 1927 1928 lreq = kzalloc(sizeof(*lreq), GFP_NOIO); 1929 if (!lreq) 1930 return NULL; 1931 1932 kref_init(&lreq->kref); 1933 mutex_init(&lreq->lock); 1934 RB_CLEAR_NODE(&lreq->node); 1935 RB_CLEAR_NODE(&lreq->osdc_node); 1936 RB_CLEAR_NODE(&lreq->mc_node); 1937 INIT_LIST_HEAD(&lreq->scan_item); 1938 INIT_LIST_HEAD(&lreq->pending_lworks); 1939 init_completion(&lreq->reg_commit_wait); 1940 init_completion(&lreq->notify_finish_wait); 1941 1942 lreq->osdc = osdc; 1943 target_init(&lreq->t); 1944 1945 dout("%s lreq %p\n", __func__, lreq); 1946 return lreq; 1947 } 1948 1949 DEFINE_RB_INSDEL_FUNCS(linger, struct ceph_osd_linger_request, linger_id, node) 1950 DEFINE_RB_FUNCS(linger_osdc, struct ceph_osd_linger_request, linger_id, osdc_node) 1951 DEFINE_RB_FUNCS(linger_mc, struct ceph_osd_linger_request, linger_id, mc_node) 1952 1953 /* 1954 * Create linger request <-> OSD session relation. 1955 * 1956 * @lreq has to be registered, @osd may be homeless. 1957 */ 1958 static void link_linger(struct ceph_osd *osd, 1959 struct ceph_osd_linger_request *lreq) 1960 { 1961 verify_osd_locked(osd); 1962 WARN_ON(!lreq->linger_id || lreq->osd); 1963 dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd, 1964 osd->o_osd, lreq, lreq->linger_id); 1965 1966 if (!osd_homeless(osd)) 1967 __remove_osd_from_lru(osd); 1968 else 1969 atomic_inc(&osd->o_osdc->num_homeless); 1970 1971 get_osd(osd); 1972 insert_linger(&osd->o_linger_requests, lreq); 1973 lreq->osd = osd; 1974 } 1975 1976 static void unlink_linger(struct ceph_osd *osd, 1977 struct ceph_osd_linger_request *lreq) 1978 { 1979 verify_osd_locked(osd); 1980 WARN_ON(lreq->osd != osd); 1981 dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd, 1982 osd->o_osd, lreq, lreq->linger_id); 1983 1984 lreq->osd = NULL; 1985 erase_linger(&osd->o_linger_requests, lreq); 1986 put_osd(osd); 1987 1988 if (!osd_homeless(osd)) 1989 maybe_move_osd_to_lru(osd); 1990 else 1991 atomic_dec(&osd->o_osdc->num_homeless); 1992 } 1993 1994 static bool __linger_registered(struct ceph_osd_linger_request *lreq) 1995 { 1996 verify_osdc_locked(lreq->osdc); 1997 1998 return !RB_EMPTY_NODE(&lreq->osdc_node); 1999 } 2000 2001 static bool linger_registered(struct ceph_osd_linger_request *lreq) 2002 { 2003 struct ceph_osd_client *osdc = lreq->osdc; 2004 bool registered; 2005 2006 down_read(&osdc->lock); 2007 registered = __linger_registered(lreq); 2008 up_read(&osdc->lock); 2009 2010 return registered; 2011 } 2012 2013 static void linger_register(struct ceph_osd_linger_request *lreq) 2014 { 2015 struct ceph_osd_client *osdc = lreq->osdc; 2016 2017 verify_osdc_wrlocked(osdc); 2018 WARN_ON(lreq->linger_id); 2019 2020 linger_get(lreq); 2021 lreq->linger_id = ++osdc->last_linger_id; 2022 insert_linger_osdc(&osdc->linger_requests, lreq); 2023 } 2024 2025 static void linger_unregister(struct ceph_osd_linger_request *lreq) 2026 { 2027 struct ceph_osd_client *osdc = lreq->osdc; 2028 2029 verify_osdc_wrlocked(osdc); 2030 2031 erase_linger_osdc(&osdc->linger_requests, lreq); 2032 linger_put(lreq); 2033 } 2034 2035 static void cancel_linger_request(struct ceph_osd_request *req) 2036 { 2037 struct ceph_osd_linger_request *lreq = req->r_priv; 2038 2039 WARN_ON(!req->r_linger); 2040 cancel_request(req); 2041 linger_put(lreq); 2042 } 2043 2044 struct linger_work { 2045 struct work_struct work; 2046 struct ceph_osd_linger_request *lreq; 2047 struct list_head pending_item; 2048 unsigned long queued_stamp; 2049 2050 union { 2051 struct { 2052 u64 notify_id; 2053 u64 notifier_id; 2054 void *payload; /* points into @msg front */ 2055 size_t payload_len; 2056 2057 struct ceph_msg *msg; /* for ceph_msg_put() */ 2058 } notify; 2059 struct { 2060 int err; 2061 } error; 2062 }; 2063 }; 2064 2065 static struct linger_work *lwork_alloc(struct ceph_osd_linger_request *lreq, 2066 work_func_t workfn) 2067 { 2068 struct linger_work *lwork; 2069 2070 lwork = kzalloc(sizeof(*lwork), GFP_NOIO); 2071 if (!lwork) 2072 return NULL; 2073 2074 INIT_WORK(&lwork->work, workfn); 2075 INIT_LIST_HEAD(&lwork->pending_item); 2076 lwork->lreq = linger_get(lreq); 2077 2078 return lwork; 2079 } 2080 2081 static void lwork_free(struct linger_work *lwork) 2082 { 2083 struct ceph_osd_linger_request *lreq = lwork->lreq; 2084 2085 mutex_lock(&lreq->lock); 2086 list_del(&lwork->pending_item); 2087 mutex_unlock(&lreq->lock); 2088 2089 linger_put(lreq); 2090 kfree(lwork); 2091 } 2092 2093 static void lwork_queue(struct linger_work *lwork) 2094 { 2095 struct ceph_osd_linger_request *lreq = lwork->lreq; 2096 struct ceph_osd_client *osdc = lreq->osdc; 2097 2098 verify_lreq_locked(lreq); 2099 WARN_ON(!list_empty(&lwork->pending_item)); 2100 2101 lwork->queued_stamp = jiffies; 2102 list_add_tail(&lwork->pending_item, &lreq->pending_lworks); 2103 queue_work(osdc->notify_wq, &lwork->work); 2104 } 2105 2106 static void do_watch_notify(struct work_struct *w) 2107 { 2108 struct linger_work *lwork = container_of(w, struct linger_work, work); 2109 struct ceph_osd_linger_request *lreq = lwork->lreq; 2110 2111 if (!linger_registered(lreq)) { 2112 dout("%s lreq %p not registered\n", __func__, lreq); 2113 goto out; 2114 } 2115 2116 WARN_ON(!lreq->is_watch); 2117 dout("%s lreq %p notify_id %llu notifier_id %llu payload_len %zu\n", 2118 __func__, lreq, lwork->notify.notify_id, lwork->notify.notifier_id, 2119 lwork->notify.payload_len); 2120 lreq->wcb(lreq->data, lwork->notify.notify_id, lreq->linger_id, 2121 lwork->notify.notifier_id, lwork->notify.payload, 2122 lwork->notify.payload_len); 2123 2124 out: 2125 ceph_msg_put(lwork->notify.msg); 2126 lwork_free(lwork); 2127 } 2128 2129 static void do_watch_error(struct work_struct *w) 2130 { 2131 struct linger_work *lwork = container_of(w, struct linger_work, work); 2132 struct ceph_osd_linger_request *lreq = lwork->lreq; 2133 2134 if (!linger_registered(lreq)) { 2135 dout("%s lreq %p not registered\n", __func__, lreq); 2136 goto out; 2137 } 2138 2139 dout("%s lreq %p err %d\n", __func__, lreq, lwork->error.err); 2140 lreq->errcb(lreq->data, lreq->linger_id, lwork->error.err); 2141 2142 out: 2143 lwork_free(lwork); 2144 } 2145 2146 static void queue_watch_error(struct ceph_osd_linger_request *lreq) 2147 { 2148 struct linger_work *lwork; 2149 2150 lwork = lwork_alloc(lreq, do_watch_error); 2151 if (!lwork) { 2152 pr_err("failed to allocate error-lwork\n"); 2153 return; 2154 } 2155 2156 lwork->error.err = lreq->last_error; 2157 lwork_queue(lwork); 2158 } 2159 2160 static void linger_reg_commit_complete(struct ceph_osd_linger_request *lreq, 2161 int result) 2162 { 2163 if (!completion_done(&lreq->reg_commit_wait)) { 2164 lreq->reg_commit_error = (result <= 0 ? result : 0); 2165 complete_all(&lreq->reg_commit_wait); 2166 } 2167 } 2168 2169 static void linger_commit_cb(struct ceph_osd_request *req) 2170 { 2171 struct ceph_osd_linger_request *lreq = req->r_priv; 2172 2173 mutex_lock(&lreq->lock); 2174 dout("%s lreq %p linger_id %llu result %d\n", __func__, lreq, 2175 lreq->linger_id, req->r_result); 2176 linger_reg_commit_complete(lreq, req->r_result); 2177 lreq->committed = true; 2178 2179 if (!lreq->is_watch) { 2180 struct ceph_osd_data *osd_data = 2181 osd_req_op_data(req, 0, notify, response_data); 2182 void *p = page_address(osd_data->pages[0]); 2183 2184 WARN_ON(req->r_ops[0].op != CEPH_OSD_OP_NOTIFY || 2185 osd_data->type != CEPH_OSD_DATA_TYPE_PAGES); 2186 2187 /* make note of the notify_id */ 2188 if (req->r_ops[0].outdata_len >= sizeof(u64)) { 2189 lreq->notify_id = ceph_decode_64(&p); 2190 dout("lreq %p notify_id %llu\n", lreq, 2191 lreq->notify_id); 2192 } else { 2193 dout("lreq %p no notify_id\n", lreq); 2194 } 2195 } 2196 2197 mutex_unlock(&lreq->lock); 2198 linger_put(lreq); 2199 } 2200 2201 static int normalize_watch_error(int err) 2202 { 2203 /* 2204 * Translate ENOENT -> ENOTCONN so that a delete->disconnection 2205 * notification and a failure to reconnect because we raced with 2206 * the delete appear the same to the user. 2207 */ 2208 if (err == -ENOENT) 2209 err = -ENOTCONN; 2210 2211 return err; 2212 } 2213 2214 static void linger_reconnect_cb(struct ceph_osd_request *req) 2215 { 2216 struct ceph_osd_linger_request *lreq = req->r_priv; 2217 2218 mutex_lock(&lreq->lock); 2219 dout("%s lreq %p linger_id %llu result %d last_error %d\n", __func__, 2220 lreq, lreq->linger_id, req->r_result, lreq->last_error); 2221 if (req->r_result < 0) { 2222 if (!lreq->last_error) { 2223 lreq->last_error = normalize_watch_error(req->r_result); 2224 queue_watch_error(lreq); 2225 } 2226 } 2227 2228 mutex_unlock(&lreq->lock); 2229 linger_put(lreq); 2230 } 2231 2232 static void send_linger(struct ceph_osd_linger_request *lreq) 2233 { 2234 struct ceph_osd_request *req = lreq->reg_req; 2235 struct ceph_osd_req_op *op = &req->r_ops[0]; 2236 2237 verify_osdc_wrlocked(req->r_osdc); 2238 dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id); 2239 2240 if (req->r_osd) 2241 cancel_linger_request(req); 2242 2243 request_reinit(req); 2244 ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid); 2245 ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc); 2246 req->r_flags = lreq->t.flags; 2247 req->r_mtime = lreq->mtime; 2248 2249 mutex_lock(&lreq->lock); 2250 if (lreq->is_watch && lreq->committed) { 2251 WARN_ON(op->op != CEPH_OSD_OP_WATCH || 2252 op->watch.cookie != lreq->linger_id); 2253 op->watch.op = CEPH_OSD_WATCH_OP_RECONNECT; 2254 op->watch.gen = ++lreq->register_gen; 2255 dout("lreq %p reconnect register_gen %u\n", lreq, 2256 op->watch.gen); 2257 req->r_callback = linger_reconnect_cb; 2258 } else { 2259 if (!lreq->is_watch) 2260 lreq->notify_id = 0; 2261 else 2262 WARN_ON(op->watch.op != CEPH_OSD_WATCH_OP_WATCH); 2263 dout("lreq %p register\n", lreq); 2264 req->r_callback = linger_commit_cb; 2265 } 2266 mutex_unlock(&lreq->lock); 2267 2268 req->r_priv = linger_get(lreq); 2269 req->r_linger = true; 2270 2271 submit_request(req, true); 2272 } 2273 2274 static void linger_ping_cb(struct ceph_osd_request *req) 2275 { 2276 struct ceph_osd_linger_request *lreq = req->r_priv; 2277 2278 mutex_lock(&lreq->lock); 2279 dout("%s lreq %p linger_id %llu result %d ping_sent %lu last_error %d\n", 2280 __func__, lreq, lreq->linger_id, req->r_result, lreq->ping_sent, 2281 lreq->last_error); 2282 if (lreq->register_gen == req->r_ops[0].watch.gen) { 2283 if (!req->r_result) { 2284 lreq->watch_valid_thru = lreq->ping_sent; 2285 } else if (!lreq->last_error) { 2286 lreq->last_error = normalize_watch_error(req->r_result); 2287 queue_watch_error(lreq); 2288 } 2289 } else { 2290 dout("lreq %p register_gen %u ignoring old pong %u\n", lreq, 2291 lreq->register_gen, req->r_ops[0].watch.gen); 2292 } 2293 2294 mutex_unlock(&lreq->lock); 2295 linger_put(lreq); 2296 } 2297 2298 static void send_linger_ping(struct ceph_osd_linger_request *lreq) 2299 { 2300 struct ceph_osd_client *osdc = lreq->osdc; 2301 struct ceph_osd_request *req = lreq->ping_req; 2302 struct ceph_osd_req_op *op = &req->r_ops[0]; 2303 2304 if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) { 2305 dout("%s PAUSERD\n", __func__); 2306 return; 2307 } 2308 2309 lreq->ping_sent = jiffies; 2310 dout("%s lreq %p linger_id %llu ping_sent %lu register_gen %u\n", 2311 __func__, lreq, lreq->linger_id, lreq->ping_sent, 2312 lreq->register_gen); 2313 2314 if (req->r_osd) 2315 cancel_linger_request(req); 2316 2317 request_reinit(req); 2318 target_copy(&req->r_t, &lreq->t); 2319 2320 WARN_ON(op->op != CEPH_OSD_OP_WATCH || 2321 op->watch.cookie != lreq->linger_id || 2322 op->watch.op != CEPH_OSD_WATCH_OP_PING); 2323 op->watch.gen = lreq->register_gen; 2324 req->r_callback = linger_ping_cb; 2325 req->r_priv = linger_get(lreq); 2326 req->r_linger = true; 2327 2328 ceph_osdc_get_request(req); 2329 account_request(req); 2330 req->r_tid = atomic64_inc_return(&osdc->last_tid); 2331 link_request(lreq->osd, req); 2332 send_request(req); 2333 } 2334 2335 static void linger_submit(struct ceph_osd_linger_request *lreq) 2336 { 2337 struct ceph_osd_client *osdc = lreq->osdc; 2338 struct ceph_osd *osd; 2339 2340 calc_target(osdc, &lreq->t, &lreq->last_force_resend, false); 2341 osd = lookup_create_osd(osdc, lreq->t.osd, true); 2342 link_linger(osd, lreq); 2343 2344 send_linger(lreq); 2345 } 2346 2347 static void cancel_linger_map_check(struct ceph_osd_linger_request *lreq) 2348 { 2349 struct ceph_osd_client *osdc = lreq->osdc; 2350 struct ceph_osd_linger_request *lookup_lreq; 2351 2352 verify_osdc_wrlocked(osdc); 2353 2354 lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks, 2355 lreq->linger_id); 2356 if (!lookup_lreq) 2357 return; 2358 2359 WARN_ON(lookup_lreq != lreq); 2360 erase_linger_mc(&osdc->linger_map_checks, lreq); 2361 linger_put(lreq); 2362 } 2363 2364 /* 2365 * @lreq has to be both registered and linked. 2366 */ 2367 static void __linger_cancel(struct ceph_osd_linger_request *lreq) 2368 { 2369 if (lreq->is_watch && lreq->ping_req->r_osd) 2370 cancel_linger_request(lreq->ping_req); 2371 if (lreq->reg_req->r_osd) 2372 cancel_linger_request(lreq->reg_req); 2373 cancel_linger_map_check(lreq); 2374 unlink_linger(lreq->osd, lreq); 2375 linger_unregister(lreq); 2376 } 2377 2378 static void linger_cancel(struct ceph_osd_linger_request *lreq) 2379 { 2380 struct ceph_osd_client *osdc = lreq->osdc; 2381 2382 down_write(&osdc->lock); 2383 if (__linger_registered(lreq)) 2384 __linger_cancel(lreq); 2385 up_write(&osdc->lock); 2386 } 2387 2388 static void send_linger_map_check(struct ceph_osd_linger_request *lreq); 2389 2390 static void check_linger_pool_dne(struct ceph_osd_linger_request *lreq) 2391 { 2392 struct ceph_osd_client *osdc = lreq->osdc; 2393 struct ceph_osdmap *map = osdc->osdmap; 2394 2395 verify_osdc_wrlocked(osdc); 2396 WARN_ON(!map->epoch); 2397 2398 if (lreq->register_gen) { 2399 lreq->map_dne_bound = map->epoch; 2400 dout("%s lreq %p linger_id %llu pool disappeared\n", __func__, 2401 lreq, lreq->linger_id); 2402 } else { 2403 dout("%s lreq %p linger_id %llu map_dne_bound %u have %u\n", 2404 __func__, lreq, lreq->linger_id, lreq->map_dne_bound, 2405 map->epoch); 2406 } 2407 2408 if (lreq->map_dne_bound) { 2409 if (map->epoch >= lreq->map_dne_bound) { 2410 /* we had a new enough map */ 2411 pr_info("linger_id %llu pool does not exist\n", 2412 lreq->linger_id); 2413 linger_reg_commit_complete(lreq, -ENOENT); 2414 __linger_cancel(lreq); 2415 } 2416 } else { 2417 send_linger_map_check(lreq); 2418 } 2419 } 2420 2421 static void linger_map_check_cb(struct ceph_mon_generic_request *greq) 2422 { 2423 struct ceph_osd_client *osdc = &greq->monc->client->osdc; 2424 struct ceph_osd_linger_request *lreq; 2425 u64 linger_id = greq->private_data; 2426 2427 WARN_ON(greq->result || !greq->u.newest); 2428 2429 down_write(&osdc->lock); 2430 lreq = lookup_linger_mc(&osdc->linger_map_checks, linger_id); 2431 if (!lreq) { 2432 dout("%s linger_id %llu dne\n", __func__, linger_id); 2433 goto out_unlock; 2434 } 2435 2436 dout("%s lreq %p linger_id %llu map_dne_bound %u newest %llu\n", 2437 __func__, lreq, lreq->linger_id, lreq->map_dne_bound, 2438 greq->u.newest); 2439 if (!lreq->map_dne_bound) 2440 lreq->map_dne_bound = greq->u.newest; 2441 erase_linger_mc(&osdc->linger_map_checks, lreq); 2442 check_linger_pool_dne(lreq); 2443 2444 linger_put(lreq); 2445 out_unlock: 2446 up_write(&osdc->lock); 2447 } 2448 2449 static void send_linger_map_check(struct ceph_osd_linger_request *lreq) 2450 { 2451 struct ceph_osd_client *osdc = lreq->osdc; 2452 struct ceph_osd_linger_request *lookup_lreq; 2453 int ret; 2454 2455 verify_osdc_wrlocked(osdc); 2456 2457 lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks, 2458 lreq->linger_id); 2459 if (lookup_lreq) { 2460 WARN_ON(lookup_lreq != lreq); 2461 return; 2462 } 2463 2464 linger_get(lreq); 2465 insert_linger_mc(&osdc->linger_map_checks, lreq); 2466 ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap", 2467 linger_map_check_cb, lreq->linger_id); 2468 WARN_ON(ret); 2469 } 2470 2471 static int linger_reg_commit_wait(struct ceph_osd_linger_request *lreq) 2472 { 2473 int ret; 2474 2475 dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id); 2476 ret = wait_for_completion_interruptible(&lreq->reg_commit_wait); 2477 return ret ?: lreq->reg_commit_error; 2478 } 2479 2480 static int linger_notify_finish_wait(struct ceph_osd_linger_request *lreq) 2481 { 2482 int ret; 2483 2484 dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id); 2485 ret = wait_for_completion_interruptible(&lreq->notify_finish_wait); 2486 return ret ?: lreq->notify_finish_error; 2487 } 2488 2489 /* 2490 * Timeout callback, called every N seconds. When 1 or more OSD 2491 * requests has been active for more than N seconds, we send a keepalive 2492 * (tag + timestamp) to its OSD to ensure any communications channel 2493 * reset is detected. 2494 */ 2495 static void handle_timeout(struct work_struct *work) 2496 { 2497 struct ceph_osd_client *osdc = 2498 container_of(work, struct ceph_osd_client, timeout_work.work); 2499 struct ceph_options *opts = osdc->client->options; 2500 unsigned long cutoff = jiffies - opts->osd_keepalive_timeout; 2501 unsigned long expiry_cutoff = jiffies - opts->osd_request_timeout; 2502 LIST_HEAD(slow_osds); 2503 struct rb_node *n, *p; 2504 2505 dout("%s osdc %p\n", __func__, osdc); 2506 down_write(&osdc->lock); 2507 2508 /* 2509 * ping osds that are a bit slow. this ensures that if there 2510 * is a break in the TCP connection we will notice, and reopen 2511 * a connection with that osd (from the fault callback). 2512 */ 2513 for (n = rb_first(&osdc->osds); n; n = rb_next(n)) { 2514 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node); 2515 bool found = false; 2516 2517 for (p = rb_first(&osd->o_requests); p; ) { 2518 struct ceph_osd_request *req = 2519 rb_entry(p, struct ceph_osd_request, r_node); 2520 2521 p = rb_next(p); /* abort_request() */ 2522 2523 if (time_before(req->r_stamp, cutoff)) { 2524 dout(" req %p tid %llu on osd%d is laggy\n", 2525 req, req->r_tid, osd->o_osd); 2526 found = true; 2527 } 2528 if (opts->osd_request_timeout && 2529 time_before(req->r_start_stamp, expiry_cutoff)) { 2530 pr_err_ratelimited("tid %llu on osd%d timeout\n", 2531 req->r_tid, osd->o_osd); 2532 abort_request(req, -ETIMEDOUT); 2533 } 2534 } 2535 for (p = rb_first(&osd->o_linger_requests); p; p = rb_next(p)) { 2536 struct ceph_osd_linger_request *lreq = 2537 rb_entry(p, struct ceph_osd_linger_request, node); 2538 2539 dout(" lreq %p linger_id %llu is served by osd%d\n", 2540 lreq, lreq->linger_id, osd->o_osd); 2541 found = true; 2542 2543 mutex_lock(&lreq->lock); 2544 if (lreq->is_watch && lreq->committed && !lreq->last_error) 2545 send_linger_ping(lreq); 2546 mutex_unlock(&lreq->lock); 2547 } 2548 2549 if (found) 2550 list_move_tail(&osd->o_keepalive_item, &slow_osds); 2551 } 2552 2553 if (opts->osd_request_timeout) { 2554 for (p = rb_first(&osdc->homeless_osd.o_requests); p; ) { 2555 struct ceph_osd_request *req = 2556 rb_entry(p, struct ceph_osd_request, r_node); 2557 2558 p = rb_next(p); /* abort_request() */ 2559 2560 if (time_before(req->r_start_stamp, expiry_cutoff)) { 2561 pr_err_ratelimited("tid %llu on osd%d timeout\n", 2562 req->r_tid, osdc->homeless_osd.o_osd); 2563 abort_request(req, -ETIMEDOUT); 2564 } 2565 } 2566 } 2567 2568 if (atomic_read(&osdc->num_homeless) || !list_empty(&slow_osds)) 2569 maybe_request_map(osdc); 2570 2571 while (!list_empty(&slow_osds)) { 2572 struct ceph_osd *osd = list_first_entry(&slow_osds, 2573 struct ceph_osd, 2574 o_keepalive_item); 2575 list_del_init(&osd->o_keepalive_item); 2576 ceph_con_keepalive(&osd->o_con); 2577 } 2578 2579 up_write(&osdc->lock); 2580 schedule_delayed_work(&osdc->timeout_work, 2581 osdc->client->options->osd_keepalive_timeout); 2582 } 2583 2584 static void handle_osds_timeout(struct work_struct *work) 2585 { 2586 struct ceph_osd_client *osdc = 2587 container_of(work, struct ceph_osd_client, 2588 osds_timeout_work.work); 2589 unsigned long delay = osdc->client->options->osd_idle_ttl / 4; 2590 struct ceph_osd *osd, *nosd; 2591 2592 dout("%s osdc %p\n", __func__, osdc); 2593 down_write(&osdc->lock); 2594 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) { 2595 if (time_before(jiffies, osd->lru_ttl)) 2596 break; 2597 2598 WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests)); 2599 WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests)); 2600 close_osd(osd); 2601 } 2602 2603 up_write(&osdc->lock); 2604 schedule_delayed_work(&osdc->osds_timeout_work, 2605 round_jiffies_relative(delay)); 2606 } 2607 2608 static int ceph_oloc_decode(void **p, void *end, 2609 struct ceph_object_locator *oloc) 2610 { 2611 u8 struct_v, struct_cv; 2612 u32 len; 2613 void *struct_end; 2614 int ret = 0; 2615 2616 ceph_decode_need(p, end, 1 + 1 + 4, e_inval); 2617 struct_v = ceph_decode_8(p); 2618 struct_cv = ceph_decode_8(p); 2619 if (struct_v < 3) { 2620 pr_warn("got v %d < 3 cv %d of ceph_object_locator\n", 2621 struct_v, struct_cv); 2622 goto e_inval; 2623 } 2624 if (struct_cv > 6) { 2625 pr_warn("got v %d cv %d > 6 of ceph_object_locator\n", 2626 struct_v, struct_cv); 2627 goto e_inval; 2628 } 2629 len = ceph_decode_32(p); 2630 ceph_decode_need(p, end, len, e_inval); 2631 struct_end = *p + len; 2632 2633 oloc->pool = ceph_decode_64(p); 2634 *p += 4; /* skip preferred */ 2635 2636 len = ceph_decode_32(p); 2637 if (len > 0) { 2638 pr_warn("ceph_object_locator::key is set\n"); 2639 goto e_inval; 2640 } 2641 2642 if (struct_v >= 5) { 2643 bool changed = false; 2644 2645 len = ceph_decode_32(p); 2646 if (len > 0) { 2647 ceph_decode_need(p, end, len, e_inval); 2648 if (!oloc->pool_ns || 2649 ceph_compare_string(oloc->pool_ns, *p, len)) 2650 changed = true; 2651 *p += len; 2652 } else { 2653 if (oloc->pool_ns) 2654 changed = true; 2655 } 2656 if (changed) { 2657 /* redirect changes namespace */ 2658 pr_warn("ceph_object_locator::nspace is changed\n"); 2659 goto e_inval; 2660 } 2661 } 2662 2663 if (struct_v >= 6) { 2664 s64 hash = ceph_decode_64(p); 2665 if (hash != -1) { 2666 pr_warn("ceph_object_locator::hash is set\n"); 2667 goto e_inval; 2668 } 2669 } 2670 2671 /* skip the rest */ 2672 *p = struct_end; 2673 out: 2674 return ret; 2675 2676 e_inval: 2677 ret = -EINVAL; 2678 goto out; 2679 } 2680 2681 static int ceph_redirect_decode(void **p, void *end, 2682 struct ceph_request_redirect *redir) 2683 { 2684 u8 struct_v, struct_cv; 2685 u32 len; 2686 void *struct_end; 2687 int ret; 2688 2689 ceph_decode_need(p, end, 1 + 1 + 4, e_inval); 2690 struct_v = ceph_decode_8(p); 2691 struct_cv = ceph_decode_8(p); 2692 if (struct_cv > 1) { 2693 pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n", 2694 struct_v, struct_cv); 2695 goto e_inval; 2696 } 2697 len = ceph_decode_32(p); 2698 ceph_decode_need(p, end, len, e_inval); 2699 struct_end = *p + len; 2700 2701 ret = ceph_oloc_decode(p, end, &redir->oloc); 2702 if (ret) 2703 goto out; 2704 2705 len = ceph_decode_32(p); 2706 if (len > 0) { 2707 pr_warn("ceph_request_redirect::object_name is set\n"); 2708 goto e_inval; 2709 } 2710 2711 len = ceph_decode_32(p); 2712 *p += len; /* skip osd_instructions */ 2713 2714 /* skip the rest */ 2715 *p = struct_end; 2716 out: 2717 return ret; 2718 2719 e_inval: 2720 ret = -EINVAL; 2721 goto out; 2722 } 2723 2724 struct MOSDOpReply { 2725 struct ceph_pg pgid; 2726 u64 flags; 2727 int result; 2728 u32 epoch; 2729 int num_ops; 2730 u32 outdata_len[CEPH_OSD_MAX_OPS]; 2731 s32 rval[CEPH_OSD_MAX_OPS]; 2732 int retry_attempt; 2733 struct ceph_eversion replay_version; 2734 u64 user_version; 2735 struct ceph_request_redirect redirect; 2736 }; 2737 2738 static int decode_MOSDOpReply(const struct ceph_msg *msg, struct MOSDOpReply *m) 2739 { 2740 void *p = msg->front.iov_base; 2741 void *const end = p + msg->front.iov_len; 2742 u16 version = le16_to_cpu(msg->hdr.version); 2743 struct ceph_eversion bad_replay_version; 2744 u8 decode_redir; 2745 u32 len; 2746 int ret; 2747 int i; 2748 2749 ceph_decode_32_safe(&p, end, len, e_inval); 2750 ceph_decode_need(&p, end, len, e_inval); 2751 p += len; /* skip oid */ 2752 2753 ret = ceph_decode_pgid(&p, end, &m->pgid); 2754 if (ret) 2755 return ret; 2756 2757 ceph_decode_64_safe(&p, end, m->flags, e_inval); 2758 ceph_decode_32_safe(&p, end, m->result, e_inval); 2759 ceph_decode_need(&p, end, sizeof(bad_replay_version), e_inval); 2760 memcpy(&bad_replay_version, p, sizeof(bad_replay_version)); 2761 p += sizeof(bad_replay_version); 2762 ceph_decode_32_safe(&p, end, m->epoch, e_inval); 2763 2764 ceph_decode_32_safe(&p, end, m->num_ops, e_inval); 2765 if (m->num_ops > ARRAY_SIZE(m->outdata_len)) 2766 goto e_inval; 2767 2768 ceph_decode_need(&p, end, m->num_ops * sizeof(struct ceph_osd_op), 2769 e_inval); 2770 for (i = 0; i < m->num_ops; i++) { 2771 struct ceph_osd_op *op = p; 2772 2773 m->outdata_len[i] = le32_to_cpu(op->payload_len); 2774 p += sizeof(*op); 2775 } 2776 2777 ceph_decode_32_safe(&p, end, m->retry_attempt, e_inval); 2778 for (i = 0; i < m->num_ops; i++) 2779 ceph_decode_32_safe(&p, end, m->rval[i], e_inval); 2780 2781 if (version >= 5) { 2782 ceph_decode_need(&p, end, sizeof(m->replay_version), e_inval); 2783 memcpy(&m->replay_version, p, sizeof(m->replay_version)); 2784 p += sizeof(m->replay_version); 2785 ceph_decode_64_safe(&p, end, m->user_version, e_inval); 2786 } else { 2787 m->replay_version = bad_replay_version; /* struct */ 2788 m->user_version = le64_to_cpu(m->replay_version.version); 2789 } 2790 2791 if (version >= 6) { 2792 if (version >= 7) 2793 ceph_decode_8_safe(&p, end, decode_redir, e_inval); 2794 else 2795 decode_redir = 1; 2796 } else { 2797 decode_redir = 0; 2798 } 2799 2800 if (decode_redir) { 2801 ret = ceph_redirect_decode(&p, end, &m->redirect); 2802 if (ret) 2803 return ret; 2804 } else { 2805 ceph_oloc_init(&m->redirect.oloc); 2806 } 2807 2808 return 0; 2809 2810 e_inval: 2811 return -EINVAL; 2812 } 2813 2814 /* 2815 * Handle MOSDOpReply. Set ->r_result and call the callback if it is 2816 * specified. 2817 */ 2818 static void handle_reply(struct ceph_osd *osd, struct ceph_msg *msg) 2819 { 2820 struct ceph_osd_client *osdc = osd->o_osdc; 2821 struct ceph_osd_request *req; 2822 struct MOSDOpReply m; 2823 u64 tid = le64_to_cpu(msg->hdr.tid); 2824 u32 data_len = 0; 2825 int ret; 2826 int i; 2827 2828 dout("%s msg %p tid %llu\n", __func__, msg, tid); 2829 2830 down_read(&osdc->lock); 2831 if (!osd_registered(osd)) { 2832 dout("%s osd%d unknown\n", __func__, osd->o_osd); 2833 goto out_unlock_osdc; 2834 } 2835 WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num)); 2836 2837 mutex_lock(&osd->lock); 2838 req = lookup_request(&osd->o_requests, tid); 2839 if (!req) { 2840 dout("%s osd%d tid %llu unknown\n", __func__, osd->o_osd, tid); 2841 goto out_unlock_session; 2842 } 2843 2844 m.redirect.oloc.pool_ns = req->r_t.target_oloc.pool_ns; 2845 ret = decode_MOSDOpReply(msg, &m); 2846 m.redirect.oloc.pool_ns = NULL; 2847 if (ret) { 2848 pr_err("failed to decode MOSDOpReply for tid %llu: %d\n", 2849 req->r_tid, ret); 2850 ceph_msg_dump(msg); 2851 goto fail_request; 2852 } 2853 dout("%s req %p tid %llu flags 0x%llx pgid %llu.%x epoch %u attempt %d v %u'%llu uv %llu\n", 2854 __func__, req, req->r_tid, m.flags, m.pgid.pool, m.pgid.seed, 2855 m.epoch, m.retry_attempt, le32_to_cpu(m.replay_version.epoch), 2856 le64_to_cpu(m.replay_version.version), m.user_version); 2857 2858 if (m.retry_attempt >= 0) { 2859 if (m.retry_attempt != req->r_attempts - 1) { 2860 dout("req %p tid %llu retry_attempt %d != %d, ignoring\n", 2861 req, req->r_tid, m.retry_attempt, 2862 req->r_attempts - 1); 2863 goto out_unlock_session; 2864 } 2865 } else { 2866 WARN_ON(1); /* MOSDOpReply v4 is assumed */ 2867 } 2868 2869 if (!ceph_oloc_empty(&m.redirect.oloc)) { 2870 dout("req %p tid %llu redirect pool %lld\n", req, req->r_tid, 2871 m.redirect.oloc.pool); 2872 unlink_request(osd, req); 2873 mutex_unlock(&osd->lock); 2874 2875 /* 2876 * Not ceph_oloc_copy() - changing pool_ns is not 2877 * supported. 2878 */ 2879 req->r_t.target_oloc.pool = m.redirect.oloc.pool; 2880 req->r_flags |= CEPH_OSD_FLAG_REDIRECTED; 2881 req->r_tid = 0; 2882 __submit_request(req, false); 2883 goto out_unlock_osdc; 2884 } 2885 2886 if (m.num_ops != req->r_num_ops) { 2887 pr_err("num_ops %d != %d for tid %llu\n", m.num_ops, 2888 req->r_num_ops, req->r_tid); 2889 goto fail_request; 2890 } 2891 for (i = 0; i < req->r_num_ops; i++) { 2892 dout(" req %p tid %llu op %d rval %d len %u\n", req, 2893 req->r_tid, i, m.rval[i], m.outdata_len[i]); 2894 req->r_ops[i].rval = m.rval[i]; 2895 req->r_ops[i].outdata_len = m.outdata_len[i]; 2896 data_len += m.outdata_len[i]; 2897 } 2898 if (data_len != le32_to_cpu(msg->hdr.data_len)) { 2899 pr_err("sum of lens %u != %u for tid %llu\n", data_len, 2900 le32_to_cpu(msg->hdr.data_len), req->r_tid); 2901 goto fail_request; 2902 } 2903 dout("%s req %p tid %llu result %d data_len %u\n", __func__, 2904 req, req->r_tid, m.result, data_len); 2905 2906 /* 2907 * Since we only ever request ONDISK, we should only ever get 2908 * one (type of) reply back. 2909 */ 2910 WARN_ON(!(m.flags & CEPH_OSD_FLAG_ONDISK)); 2911 req->r_result = m.result ?: data_len; 2912 finish_request(req); 2913 mutex_unlock(&osd->lock); 2914 up_read(&osdc->lock); 2915 2916 __complete_request(req); 2917 complete_all(&req->r_completion); 2918 ceph_osdc_put_request(req); 2919 return; 2920 2921 fail_request: 2922 complete_request(req, -EIO); 2923 out_unlock_session: 2924 mutex_unlock(&osd->lock); 2925 out_unlock_osdc: 2926 up_read(&osdc->lock); 2927 } 2928 2929 static void set_pool_was_full(struct ceph_osd_client *osdc) 2930 { 2931 struct rb_node *n; 2932 2933 for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) { 2934 struct ceph_pg_pool_info *pi = 2935 rb_entry(n, struct ceph_pg_pool_info, node); 2936 2937 pi->was_full = __pool_full(pi); 2938 } 2939 } 2940 2941 static bool pool_cleared_full(struct ceph_osd_client *osdc, s64 pool_id) 2942 { 2943 struct ceph_pg_pool_info *pi; 2944 2945 pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id); 2946 if (!pi) 2947 return false; 2948 2949 return pi->was_full && !__pool_full(pi); 2950 } 2951 2952 static enum calc_target_result 2953 recalc_linger_target(struct ceph_osd_linger_request *lreq) 2954 { 2955 struct ceph_osd_client *osdc = lreq->osdc; 2956 enum calc_target_result ct_res; 2957 2958 ct_res = calc_target(osdc, &lreq->t, &lreq->last_force_resend, true); 2959 if (ct_res == CALC_TARGET_NEED_RESEND) { 2960 struct ceph_osd *osd; 2961 2962 osd = lookup_create_osd(osdc, lreq->t.osd, true); 2963 if (osd != lreq->osd) { 2964 unlink_linger(lreq->osd, lreq); 2965 link_linger(osd, lreq); 2966 } 2967 } 2968 2969 return ct_res; 2970 } 2971 2972 /* 2973 * Requeue requests whose mapping to an OSD has changed. 2974 */ 2975 static void scan_requests(struct ceph_osd *osd, 2976 bool force_resend, 2977 bool cleared_full, 2978 bool check_pool_cleared_full, 2979 struct rb_root *need_resend, 2980 struct list_head *need_resend_linger) 2981 { 2982 struct ceph_osd_client *osdc = osd->o_osdc; 2983 struct rb_node *n; 2984 bool force_resend_writes; 2985 2986 for (n = rb_first(&osd->o_linger_requests); n; ) { 2987 struct ceph_osd_linger_request *lreq = 2988 rb_entry(n, struct ceph_osd_linger_request, node); 2989 enum calc_target_result ct_res; 2990 2991 n = rb_next(n); /* recalc_linger_target() */ 2992 2993 dout("%s lreq %p linger_id %llu\n", __func__, lreq, 2994 lreq->linger_id); 2995 ct_res = recalc_linger_target(lreq); 2996 switch (ct_res) { 2997 case CALC_TARGET_NO_ACTION: 2998 force_resend_writes = cleared_full || 2999 (check_pool_cleared_full && 3000 pool_cleared_full(osdc, lreq->t.base_oloc.pool)); 3001 if (!force_resend && !force_resend_writes) 3002 break; 3003 3004 /* fall through */ 3005 case CALC_TARGET_NEED_RESEND: 3006 cancel_linger_map_check(lreq); 3007 /* 3008 * scan_requests() for the previous epoch(s) 3009 * may have already added it to the list, since 3010 * it's not unlinked here. 3011 */ 3012 if (list_empty(&lreq->scan_item)) 3013 list_add_tail(&lreq->scan_item, need_resend_linger); 3014 break; 3015 case CALC_TARGET_POOL_DNE: 3016 check_linger_pool_dne(lreq); 3017 break; 3018 } 3019 } 3020 3021 for (n = rb_first(&osd->o_requests); n; ) { 3022 struct ceph_osd_request *req = 3023 rb_entry(n, struct ceph_osd_request, r_node); 3024 enum calc_target_result ct_res; 3025 3026 n = rb_next(n); /* unlink_request(), check_pool_dne() */ 3027 3028 dout("%s req %p tid %llu\n", __func__, req, req->r_tid); 3029 ct_res = calc_target(osdc, &req->r_t, 3030 &req->r_last_force_resend, false); 3031 switch (ct_res) { 3032 case CALC_TARGET_NO_ACTION: 3033 force_resend_writes = cleared_full || 3034 (check_pool_cleared_full && 3035 pool_cleared_full(osdc, req->r_t.base_oloc.pool)); 3036 if (!force_resend && 3037 (!(req->r_flags & CEPH_OSD_FLAG_WRITE) || 3038 !force_resend_writes)) 3039 break; 3040 3041 /* fall through */ 3042 case CALC_TARGET_NEED_RESEND: 3043 cancel_map_check(req); 3044 unlink_request(osd, req); 3045 insert_request(need_resend, req); 3046 break; 3047 case CALC_TARGET_POOL_DNE: 3048 check_pool_dne(req); 3049 break; 3050 } 3051 } 3052 } 3053 3054 static int handle_one_map(struct ceph_osd_client *osdc, 3055 void *p, void *end, bool incremental, 3056 struct rb_root *need_resend, 3057 struct list_head *need_resend_linger) 3058 { 3059 struct ceph_osdmap *newmap; 3060 struct rb_node *n; 3061 bool skipped_map = false; 3062 bool was_full; 3063 3064 was_full = ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL); 3065 set_pool_was_full(osdc); 3066 3067 if (incremental) 3068 newmap = osdmap_apply_incremental(&p, end, osdc->osdmap); 3069 else 3070 newmap = ceph_osdmap_decode(&p, end); 3071 if (IS_ERR(newmap)) 3072 return PTR_ERR(newmap); 3073 3074 if (newmap != osdc->osdmap) { 3075 /* 3076 * Preserve ->was_full before destroying the old map. 3077 * For pools that weren't in the old map, ->was_full 3078 * should be false. 3079 */ 3080 for (n = rb_first(&newmap->pg_pools); n; n = rb_next(n)) { 3081 struct ceph_pg_pool_info *pi = 3082 rb_entry(n, struct ceph_pg_pool_info, node); 3083 struct ceph_pg_pool_info *old_pi; 3084 3085 old_pi = ceph_pg_pool_by_id(osdc->osdmap, pi->id); 3086 if (old_pi) 3087 pi->was_full = old_pi->was_full; 3088 else 3089 WARN_ON(pi->was_full); 3090 } 3091 3092 if (osdc->osdmap->epoch && 3093 osdc->osdmap->epoch + 1 < newmap->epoch) { 3094 WARN_ON(incremental); 3095 skipped_map = true; 3096 } 3097 3098 ceph_osdmap_destroy(osdc->osdmap); 3099 osdc->osdmap = newmap; 3100 } 3101 3102 was_full &= !ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL); 3103 scan_requests(&osdc->homeless_osd, skipped_map, was_full, true, 3104 need_resend, need_resend_linger); 3105 3106 for (n = rb_first(&osdc->osds); n; ) { 3107 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node); 3108 3109 n = rb_next(n); /* close_osd() */ 3110 3111 scan_requests(osd, skipped_map, was_full, true, need_resend, 3112 need_resend_linger); 3113 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) || 3114 memcmp(&osd->o_con.peer_addr, 3115 ceph_osd_addr(osdc->osdmap, osd->o_osd), 3116 sizeof(struct ceph_entity_addr))) 3117 close_osd(osd); 3118 } 3119 3120 return 0; 3121 } 3122 3123 static void kick_requests(struct ceph_osd_client *osdc, 3124 struct rb_root *need_resend, 3125 struct list_head *need_resend_linger) 3126 { 3127 struct ceph_osd_linger_request *lreq, *nlreq; 3128 struct rb_node *n; 3129 3130 for (n = rb_first(need_resend); n; ) { 3131 struct ceph_osd_request *req = 3132 rb_entry(n, struct ceph_osd_request, r_node); 3133 struct ceph_osd *osd; 3134 3135 n = rb_next(n); 3136 erase_request(need_resend, req); /* before link_request() */ 3137 3138 WARN_ON(req->r_osd); 3139 calc_target(osdc, &req->r_t, NULL, false); 3140 osd = lookup_create_osd(osdc, req->r_t.osd, true); 3141 link_request(osd, req); 3142 if (!req->r_linger) { 3143 if (!osd_homeless(osd) && !req->r_t.paused) 3144 send_request(req); 3145 } else { 3146 cancel_linger_request(req); 3147 } 3148 } 3149 3150 list_for_each_entry_safe(lreq, nlreq, need_resend_linger, scan_item) { 3151 if (!osd_homeless(lreq->osd)) 3152 send_linger(lreq); 3153 3154 list_del_init(&lreq->scan_item); 3155 } 3156 } 3157 3158 /* 3159 * Process updated osd map. 3160 * 3161 * The message contains any number of incremental and full maps, normally 3162 * indicating some sort of topology change in the cluster. Kick requests 3163 * off to different OSDs as needed. 3164 */ 3165 void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg) 3166 { 3167 void *p = msg->front.iov_base; 3168 void *const end = p + msg->front.iov_len; 3169 u32 nr_maps, maplen; 3170 u32 epoch; 3171 struct ceph_fsid fsid; 3172 struct rb_root need_resend = RB_ROOT; 3173 LIST_HEAD(need_resend_linger); 3174 bool handled_incremental = false; 3175 bool was_pauserd, was_pausewr; 3176 bool pauserd, pausewr; 3177 int err; 3178 3179 dout("%s have %u\n", __func__, osdc->osdmap->epoch); 3180 down_write(&osdc->lock); 3181 3182 /* verify fsid */ 3183 ceph_decode_need(&p, end, sizeof(fsid), bad); 3184 ceph_decode_copy(&p, &fsid, sizeof(fsid)); 3185 if (ceph_check_fsid(osdc->client, &fsid) < 0) 3186 goto bad; 3187 3188 was_pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD); 3189 was_pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) || 3190 ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || 3191 have_pool_full(osdc); 3192 3193 /* incremental maps */ 3194 ceph_decode_32_safe(&p, end, nr_maps, bad); 3195 dout(" %d inc maps\n", nr_maps); 3196 while (nr_maps > 0) { 3197 ceph_decode_need(&p, end, 2*sizeof(u32), bad); 3198 epoch = ceph_decode_32(&p); 3199 maplen = ceph_decode_32(&p); 3200 ceph_decode_need(&p, end, maplen, bad); 3201 if (osdc->osdmap->epoch && 3202 osdc->osdmap->epoch + 1 == epoch) { 3203 dout("applying incremental map %u len %d\n", 3204 epoch, maplen); 3205 err = handle_one_map(osdc, p, p + maplen, true, 3206 &need_resend, &need_resend_linger); 3207 if (err) 3208 goto bad; 3209 handled_incremental = true; 3210 } else { 3211 dout("ignoring incremental map %u len %d\n", 3212 epoch, maplen); 3213 } 3214 p += maplen; 3215 nr_maps--; 3216 } 3217 if (handled_incremental) 3218 goto done; 3219 3220 /* full maps */ 3221 ceph_decode_32_safe(&p, end, nr_maps, bad); 3222 dout(" %d full maps\n", nr_maps); 3223 while (nr_maps) { 3224 ceph_decode_need(&p, end, 2*sizeof(u32), bad); 3225 epoch = ceph_decode_32(&p); 3226 maplen = ceph_decode_32(&p); 3227 ceph_decode_need(&p, end, maplen, bad); 3228 if (nr_maps > 1) { 3229 dout("skipping non-latest full map %u len %d\n", 3230 epoch, maplen); 3231 } else if (osdc->osdmap->epoch >= epoch) { 3232 dout("skipping full map %u len %d, " 3233 "older than our %u\n", epoch, maplen, 3234 osdc->osdmap->epoch); 3235 } else { 3236 dout("taking full map %u len %d\n", epoch, maplen); 3237 err = handle_one_map(osdc, p, p + maplen, false, 3238 &need_resend, &need_resend_linger); 3239 if (err) 3240 goto bad; 3241 } 3242 p += maplen; 3243 nr_maps--; 3244 } 3245 3246 done: 3247 /* 3248 * subscribe to subsequent osdmap updates if full to ensure 3249 * we find out when we are no longer full and stop returning 3250 * ENOSPC. 3251 */ 3252 pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD); 3253 pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) || 3254 ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || 3255 have_pool_full(osdc); 3256 if (was_pauserd || was_pausewr || pauserd || pausewr) 3257 maybe_request_map(osdc); 3258 3259 kick_requests(osdc, &need_resend, &need_resend_linger); 3260 3261 ceph_monc_got_map(&osdc->client->monc, CEPH_SUB_OSDMAP, 3262 osdc->osdmap->epoch); 3263 up_write(&osdc->lock); 3264 wake_up_all(&osdc->client->auth_wq); 3265 return; 3266 3267 bad: 3268 pr_err("osdc handle_map corrupt msg\n"); 3269 ceph_msg_dump(msg); 3270 up_write(&osdc->lock); 3271 } 3272 3273 /* 3274 * Resubmit requests pending on the given osd. 3275 */ 3276 static void kick_osd_requests(struct ceph_osd *osd) 3277 { 3278 struct rb_node *n; 3279 3280 for (n = rb_first(&osd->o_requests); n; ) { 3281 struct ceph_osd_request *req = 3282 rb_entry(n, struct ceph_osd_request, r_node); 3283 3284 n = rb_next(n); /* cancel_linger_request() */ 3285 3286 if (!req->r_linger) { 3287 if (!req->r_t.paused) 3288 send_request(req); 3289 } else { 3290 cancel_linger_request(req); 3291 } 3292 } 3293 for (n = rb_first(&osd->o_linger_requests); n; n = rb_next(n)) { 3294 struct ceph_osd_linger_request *lreq = 3295 rb_entry(n, struct ceph_osd_linger_request, node); 3296 3297 send_linger(lreq); 3298 } 3299 } 3300 3301 /* 3302 * If the osd connection drops, we need to resubmit all requests. 3303 */ 3304 static void osd_fault(struct ceph_connection *con) 3305 { 3306 struct ceph_osd *osd = con->private; 3307 struct ceph_osd_client *osdc = osd->o_osdc; 3308 3309 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd); 3310 3311 down_write(&osdc->lock); 3312 if (!osd_registered(osd)) { 3313 dout("%s osd%d unknown\n", __func__, osd->o_osd); 3314 goto out_unlock; 3315 } 3316 3317 if (!reopen_osd(osd)) 3318 kick_osd_requests(osd); 3319 maybe_request_map(osdc); 3320 3321 out_unlock: 3322 up_write(&osdc->lock); 3323 } 3324 3325 /* 3326 * Process osd watch notifications 3327 */ 3328 static void handle_watch_notify(struct ceph_osd_client *osdc, 3329 struct ceph_msg *msg) 3330 { 3331 void *p = msg->front.iov_base; 3332 void *const end = p + msg->front.iov_len; 3333 struct ceph_osd_linger_request *lreq; 3334 struct linger_work *lwork; 3335 u8 proto_ver, opcode; 3336 u64 cookie, notify_id; 3337 u64 notifier_id = 0; 3338 s32 return_code = 0; 3339 void *payload = NULL; 3340 u32 payload_len = 0; 3341 3342 ceph_decode_8_safe(&p, end, proto_ver, bad); 3343 ceph_decode_8_safe(&p, end, opcode, bad); 3344 ceph_decode_64_safe(&p, end, cookie, bad); 3345 p += 8; /* skip ver */ 3346 ceph_decode_64_safe(&p, end, notify_id, bad); 3347 3348 if (proto_ver >= 1) { 3349 ceph_decode_32_safe(&p, end, payload_len, bad); 3350 ceph_decode_need(&p, end, payload_len, bad); 3351 payload = p; 3352 p += payload_len; 3353 } 3354 3355 if (le16_to_cpu(msg->hdr.version) >= 2) 3356 ceph_decode_32_safe(&p, end, return_code, bad); 3357 3358 if (le16_to_cpu(msg->hdr.version) >= 3) 3359 ceph_decode_64_safe(&p, end, notifier_id, bad); 3360 3361 down_read(&osdc->lock); 3362 lreq = lookup_linger_osdc(&osdc->linger_requests, cookie); 3363 if (!lreq) { 3364 dout("%s opcode %d cookie %llu dne\n", __func__, opcode, 3365 cookie); 3366 goto out_unlock_osdc; 3367 } 3368 3369 mutex_lock(&lreq->lock); 3370 dout("%s opcode %d cookie %llu lreq %p is_watch %d\n", __func__, 3371 opcode, cookie, lreq, lreq->is_watch); 3372 if (opcode == CEPH_WATCH_EVENT_DISCONNECT) { 3373 if (!lreq->last_error) { 3374 lreq->last_error = -ENOTCONN; 3375 queue_watch_error(lreq); 3376 } 3377 } else if (!lreq->is_watch) { 3378 /* CEPH_WATCH_EVENT_NOTIFY_COMPLETE */ 3379 if (lreq->notify_id && lreq->notify_id != notify_id) { 3380 dout("lreq %p notify_id %llu != %llu, ignoring\n", lreq, 3381 lreq->notify_id, notify_id); 3382 } else if (!completion_done(&lreq->notify_finish_wait)) { 3383 struct ceph_msg_data *data = 3384 list_first_entry_or_null(&msg->data, 3385 struct ceph_msg_data, 3386 links); 3387 3388 if (data) { 3389 if (lreq->preply_pages) { 3390 WARN_ON(data->type != 3391 CEPH_MSG_DATA_PAGES); 3392 *lreq->preply_pages = data->pages; 3393 *lreq->preply_len = data->length; 3394 } else { 3395 ceph_release_page_vector(data->pages, 3396 calc_pages_for(0, data->length)); 3397 } 3398 } 3399 lreq->notify_finish_error = return_code; 3400 complete_all(&lreq->notify_finish_wait); 3401 } 3402 } else { 3403 /* CEPH_WATCH_EVENT_NOTIFY */ 3404 lwork = lwork_alloc(lreq, do_watch_notify); 3405 if (!lwork) { 3406 pr_err("failed to allocate notify-lwork\n"); 3407 goto out_unlock_lreq; 3408 } 3409 3410 lwork->notify.notify_id = notify_id; 3411 lwork->notify.notifier_id = notifier_id; 3412 lwork->notify.payload = payload; 3413 lwork->notify.payload_len = payload_len; 3414 lwork->notify.msg = ceph_msg_get(msg); 3415 lwork_queue(lwork); 3416 } 3417 3418 out_unlock_lreq: 3419 mutex_unlock(&lreq->lock); 3420 out_unlock_osdc: 3421 up_read(&osdc->lock); 3422 return; 3423 3424 bad: 3425 pr_err("osdc handle_watch_notify corrupt msg\n"); 3426 } 3427 3428 /* 3429 * Register request, send initial attempt. 3430 */ 3431 int ceph_osdc_start_request(struct ceph_osd_client *osdc, 3432 struct ceph_osd_request *req, 3433 bool nofail) 3434 { 3435 down_read(&osdc->lock); 3436 submit_request(req, false); 3437 up_read(&osdc->lock); 3438 3439 return 0; 3440 } 3441 EXPORT_SYMBOL(ceph_osdc_start_request); 3442 3443 /* 3444 * Unregister a registered request. The request is not completed: 3445 * ->r_result isn't set and __complete_request() isn't called. 3446 */ 3447 void ceph_osdc_cancel_request(struct ceph_osd_request *req) 3448 { 3449 struct ceph_osd_client *osdc = req->r_osdc; 3450 3451 down_write(&osdc->lock); 3452 if (req->r_osd) 3453 cancel_request(req); 3454 up_write(&osdc->lock); 3455 } 3456 EXPORT_SYMBOL(ceph_osdc_cancel_request); 3457 3458 /* 3459 * @timeout: in jiffies, 0 means "wait forever" 3460 */ 3461 static int wait_request_timeout(struct ceph_osd_request *req, 3462 unsigned long timeout) 3463 { 3464 long left; 3465 3466 dout("%s req %p tid %llu\n", __func__, req, req->r_tid); 3467 left = wait_for_completion_killable_timeout(&req->r_completion, 3468 ceph_timeout_jiffies(timeout)); 3469 if (left <= 0) { 3470 left = left ?: -ETIMEDOUT; 3471 ceph_osdc_cancel_request(req); 3472 } else { 3473 left = req->r_result; /* completed */ 3474 } 3475 3476 return left; 3477 } 3478 3479 /* 3480 * wait for a request to complete 3481 */ 3482 int ceph_osdc_wait_request(struct ceph_osd_client *osdc, 3483 struct ceph_osd_request *req) 3484 { 3485 return wait_request_timeout(req, 0); 3486 } 3487 EXPORT_SYMBOL(ceph_osdc_wait_request); 3488 3489 /* 3490 * sync - wait for all in-flight requests to flush. avoid starvation. 3491 */ 3492 void ceph_osdc_sync(struct ceph_osd_client *osdc) 3493 { 3494 struct rb_node *n, *p; 3495 u64 last_tid = atomic64_read(&osdc->last_tid); 3496 3497 again: 3498 down_read(&osdc->lock); 3499 for (n = rb_first(&osdc->osds); n; n = rb_next(n)) { 3500 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node); 3501 3502 mutex_lock(&osd->lock); 3503 for (p = rb_first(&osd->o_requests); p; p = rb_next(p)) { 3504 struct ceph_osd_request *req = 3505 rb_entry(p, struct ceph_osd_request, r_node); 3506 3507 if (req->r_tid > last_tid) 3508 break; 3509 3510 if (!(req->r_flags & CEPH_OSD_FLAG_WRITE)) 3511 continue; 3512 3513 ceph_osdc_get_request(req); 3514 mutex_unlock(&osd->lock); 3515 up_read(&osdc->lock); 3516 dout("%s waiting on req %p tid %llu last_tid %llu\n", 3517 __func__, req, req->r_tid, last_tid); 3518 wait_for_completion(&req->r_completion); 3519 ceph_osdc_put_request(req); 3520 goto again; 3521 } 3522 3523 mutex_unlock(&osd->lock); 3524 } 3525 3526 up_read(&osdc->lock); 3527 dout("%s done last_tid %llu\n", __func__, last_tid); 3528 } 3529 EXPORT_SYMBOL(ceph_osdc_sync); 3530 3531 static struct ceph_osd_request * 3532 alloc_linger_request(struct ceph_osd_linger_request *lreq) 3533 { 3534 struct ceph_osd_request *req; 3535 3536 req = ceph_osdc_alloc_request(lreq->osdc, NULL, 1, false, GFP_NOIO); 3537 if (!req) 3538 return NULL; 3539 3540 ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid); 3541 ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc); 3542 3543 if (ceph_osdc_alloc_messages(req, GFP_NOIO)) { 3544 ceph_osdc_put_request(req); 3545 return NULL; 3546 } 3547 3548 return req; 3549 } 3550 3551 /* 3552 * Returns a handle, caller owns a ref. 3553 */ 3554 struct ceph_osd_linger_request * 3555 ceph_osdc_watch(struct ceph_osd_client *osdc, 3556 struct ceph_object_id *oid, 3557 struct ceph_object_locator *oloc, 3558 rados_watchcb2_t wcb, 3559 rados_watcherrcb_t errcb, 3560 void *data) 3561 { 3562 struct ceph_osd_linger_request *lreq; 3563 int ret; 3564 3565 lreq = linger_alloc(osdc); 3566 if (!lreq) 3567 return ERR_PTR(-ENOMEM); 3568 3569 lreq->is_watch = true; 3570 lreq->wcb = wcb; 3571 lreq->errcb = errcb; 3572 lreq->data = data; 3573 lreq->watch_valid_thru = jiffies; 3574 3575 ceph_oid_copy(&lreq->t.base_oid, oid); 3576 ceph_oloc_copy(&lreq->t.base_oloc, oloc); 3577 lreq->t.flags = CEPH_OSD_FLAG_WRITE; 3578 lreq->mtime = CURRENT_TIME; 3579 3580 lreq->reg_req = alloc_linger_request(lreq); 3581 if (!lreq->reg_req) { 3582 ret = -ENOMEM; 3583 goto err_put_lreq; 3584 } 3585 3586 lreq->ping_req = alloc_linger_request(lreq); 3587 if (!lreq->ping_req) { 3588 ret = -ENOMEM; 3589 goto err_put_lreq; 3590 } 3591 3592 down_write(&osdc->lock); 3593 linger_register(lreq); /* before osd_req_op_* */ 3594 osd_req_op_watch_init(lreq->reg_req, 0, lreq->linger_id, 3595 CEPH_OSD_WATCH_OP_WATCH); 3596 osd_req_op_watch_init(lreq->ping_req, 0, lreq->linger_id, 3597 CEPH_OSD_WATCH_OP_PING); 3598 linger_submit(lreq); 3599 up_write(&osdc->lock); 3600 3601 ret = linger_reg_commit_wait(lreq); 3602 if (ret) { 3603 linger_cancel(lreq); 3604 goto err_put_lreq; 3605 } 3606 3607 return lreq; 3608 3609 err_put_lreq: 3610 linger_put(lreq); 3611 return ERR_PTR(ret); 3612 } 3613 EXPORT_SYMBOL(ceph_osdc_watch); 3614 3615 /* 3616 * Releases a ref. 3617 * 3618 * Times out after mount_timeout to preserve rbd unmap behaviour 3619 * introduced in 2894e1d76974 ("rbd: timeout watch teardown on unmap 3620 * with mount_timeout"). 3621 */ 3622 int ceph_osdc_unwatch(struct ceph_osd_client *osdc, 3623 struct ceph_osd_linger_request *lreq) 3624 { 3625 struct ceph_options *opts = osdc->client->options; 3626 struct ceph_osd_request *req; 3627 int ret; 3628 3629 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO); 3630 if (!req) 3631 return -ENOMEM; 3632 3633 ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid); 3634 ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc); 3635 req->r_flags = CEPH_OSD_FLAG_WRITE; 3636 req->r_mtime = CURRENT_TIME; 3637 osd_req_op_watch_init(req, 0, lreq->linger_id, 3638 CEPH_OSD_WATCH_OP_UNWATCH); 3639 3640 ret = ceph_osdc_alloc_messages(req, GFP_NOIO); 3641 if (ret) 3642 goto out_put_req; 3643 3644 ceph_osdc_start_request(osdc, req, false); 3645 linger_cancel(lreq); 3646 linger_put(lreq); 3647 ret = wait_request_timeout(req, opts->mount_timeout); 3648 3649 out_put_req: 3650 ceph_osdc_put_request(req); 3651 return ret; 3652 } 3653 EXPORT_SYMBOL(ceph_osdc_unwatch); 3654 3655 static int osd_req_op_notify_ack_init(struct ceph_osd_request *req, int which, 3656 u64 notify_id, u64 cookie, void *payload, 3657 size_t payload_len) 3658 { 3659 struct ceph_osd_req_op *op; 3660 struct ceph_pagelist *pl; 3661 int ret; 3662 3663 op = _osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY_ACK, 0); 3664 3665 pl = kmalloc(sizeof(*pl), GFP_NOIO); 3666 if (!pl) 3667 return -ENOMEM; 3668 3669 ceph_pagelist_init(pl); 3670 ret = ceph_pagelist_encode_64(pl, notify_id); 3671 ret |= ceph_pagelist_encode_64(pl, cookie); 3672 if (payload) { 3673 ret |= ceph_pagelist_encode_32(pl, payload_len); 3674 ret |= ceph_pagelist_append(pl, payload, payload_len); 3675 } else { 3676 ret |= ceph_pagelist_encode_32(pl, 0); 3677 } 3678 if (ret) { 3679 ceph_pagelist_release(pl); 3680 return -ENOMEM; 3681 } 3682 3683 ceph_osd_data_pagelist_init(&op->notify_ack.request_data, pl); 3684 op->indata_len = pl->length; 3685 return 0; 3686 } 3687 3688 int ceph_osdc_notify_ack(struct ceph_osd_client *osdc, 3689 struct ceph_object_id *oid, 3690 struct ceph_object_locator *oloc, 3691 u64 notify_id, 3692 u64 cookie, 3693 void *payload, 3694 size_t payload_len) 3695 { 3696 struct ceph_osd_request *req; 3697 int ret; 3698 3699 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO); 3700 if (!req) 3701 return -ENOMEM; 3702 3703 ceph_oid_copy(&req->r_base_oid, oid); 3704 ceph_oloc_copy(&req->r_base_oloc, oloc); 3705 req->r_flags = CEPH_OSD_FLAG_READ; 3706 3707 ret = ceph_osdc_alloc_messages(req, GFP_NOIO); 3708 if (ret) 3709 goto out_put_req; 3710 3711 ret = osd_req_op_notify_ack_init(req, 0, notify_id, cookie, payload, 3712 payload_len); 3713 if (ret) 3714 goto out_put_req; 3715 3716 ceph_osdc_start_request(osdc, req, false); 3717 ret = ceph_osdc_wait_request(osdc, req); 3718 3719 out_put_req: 3720 ceph_osdc_put_request(req); 3721 return ret; 3722 } 3723 EXPORT_SYMBOL(ceph_osdc_notify_ack); 3724 3725 static int osd_req_op_notify_init(struct ceph_osd_request *req, int which, 3726 u64 cookie, u32 prot_ver, u32 timeout, 3727 void *payload, size_t payload_len) 3728 { 3729 struct ceph_osd_req_op *op; 3730 struct ceph_pagelist *pl; 3731 int ret; 3732 3733 op = _osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY, 0); 3734 op->notify.cookie = cookie; 3735 3736 pl = kmalloc(sizeof(*pl), GFP_NOIO); 3737 if (!pl) 3738 return -ENOMEM; 3739 3740 ceph_pagelist_init(pl); 3741 ret = ceph_pagelist_encode_32(pl, 1); /* prot_ver */ 3742 ret |= ceph_pagelist_encode_32(pl, timeout); 3743 ret |= ceph_pagelist_encode_32(pl, payload_len); 3744 ret |= ceph_pagelist_append(pl, payload, payload_len); 3745 if (ret) { 3746 ceph_pagelist_release(pl); 3747 return -ENOMEM; 3748 } 3749 3750 ceph_osd_data_pagelist_init(&op->notify.request_data, pl); 3751 op->indata_len = pl->length; 3752 return 0; 3753 } 3754 3755 /* 3756 * @timeout: in seconds 3757 * 3758 * @preply_{pages,len} are initialized both on success and error. 3759 * The caller is responsible for: 3760 * 3761 * ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len)) 3762 */ 3763 int ceph_osdc_notify(struct ceph_osd_client *osdc, 3764 struct ceph_object_id *oid, 3765 struct ceph_object_locator *oloc, 3766 void *payload, 3767 size_t payload_len, 3768 u32 timeout, 3769 struct page ***preply_pages, 3770 size_t *preply_len) 3771 { 3772 struct ceph_osd_linger_request *lreq; 3773 struct page **pages; 3774 int ret; 3775 3776 WARN_ON(!timeout); 3777 if (preply_pages) { 3778 *preply_pages = NULL; 3779 *preply_len = 0; 3780 } 3781 3782 lreq = linger_alloc(osdc); 3783 if (!lreq) 3784 return -ENOMEM; 3785 3786 lreq->preply_pages = preply_pages; 3787 lreq->preply_len = preply_len; 3788 3789 ceph_oid_copy(&lreq->t.base_oid, oid); 3790 ceph_oloc_copy(&lreq->t.base_oloc, oloc); 3791 lreq->t.flags = CEPH_OSD_FLAG_READ; 3792 3793 lreq->reg_req = alloc_linger_request(lreq); 3794 if (!lreq->reg_req) { 3795 ret = -ENOMEM; 3796 goto out_put_lreq; 3797 } 3798 3799 /* for notify_id */ 3800 pages = ceph_alloc_page_vector(1, GFP_NOIO); 3801 if (IS_ERR(pages)) { 3802 ret = PTR_ERR(pages); 3803 goto out_put_lreq; 3804 } 3805 3806 down_write(&osdc->lock); 3807 linger_register(lreq); /* before osd_req_op_* */ 3808 ret = osd_req_op_notify_init(lreq->reg_req, 0, lreq->linger_id, 1, 3809 timeout, payload, payload_len); 3810 if (ret) { 3811 linger_unregister(lreq); 3812 up_write(&osdc->lock); 3813 ceph_release_page_vector(pages, 1); 3814 goto out_put_lreq; 3815 } 3816 ceph_osd_data_pages_init(osd_req_op_data(lreq->reg_req, 0, notify, 3817 response_data), 3818 pages, PAGE_SIZE, 0, false, true); 3819 linger_submit(lreq); 3820 up_write(&osdc->lock); 3821 3822 ret = linger_reg_commit_wait(lreq); 3823 if (!ret) 3824 ret = linger_notify_finish_wait(lreq); 3825 else 3826 dout("lreq %p failed to initiate notify %d\n", lreq, ret); 3827 3828 linger_cancel(lreq); 3829 out_put_lreq: 3830 linger_put(lreq); 3831 return ret; 3832 } 3833 EXPORT_SYMBOL(ceph_osdc_notify); 3834 3835 /* 3836 * Return the number of milliseconds since the watch was last 3837 * confirmed, or an error. If there is an error, the watch is no 3838 * longer valid, and should be destroyed with ceph_osdc_unwatch(). 3839 */ 3840 int ceph_osdc_watch_check(struct ceph_osd_client *osdc, 3841 struct ceph_osd_linger_request *lreq) 3842 { 3843 unsigned long stamp, age; 3844 int ret; 3845 3846 down_read(&osdc->lock); 3847 mutex_lock(&lreq->lock); 3848 stamp = lreq->watch_valid_thru; 3849 if (!list_empty(&lreq->pending_lworks)) { 3850 struct linger_work *lwork = 3851 list_first_entry(&lreq->pending_lworks, 3852 struct linger_work, 3853 pending_item); 3854 3855 if (time_before(lwork->queued_stamp, stamp)) 3856 stamp = lwork->queued_stamp; 3857 } 3858 age = jiffies - stamp; 3859 dout("%s lreq %p linger_id %llu age %lu last_error %d\n", __func__, 3860 lreq, lreq->linger_id, age, lreq->last_error); 3861 /* we are truncating to msecs, so return a safe upper bound */ 3862 ret = lreq->last_error ?: 1 + jiffies_to_msecs(age); 3863 3864 mutex_unlock(&lreq->lock); 3865 up_read(&osdc->lock); 3866 return ret; 3867 } 3868 3869 static int decode_watcher(void **p, void *end, struct ceph_watch_item *item) 3870 { 3871 u8 struct_v; 3872 u32 struct_len; 3873 int ret; 3874 3875 ret = ceph_start_decoding(p, end, 2, "watch_item_t", 3876 &struct_v, &struct_len); 3877 if (ret) 3878 return ret; 3879 3880 ceph_decode_copy(p, &item->name, sizeof(item->name)); 3881 item->cookie = ceph_decode_64(p); 3882 *p += 4; /* skip timeout_seconds */ 3883 if (struct_v >= 2) { 3884 ceph_decode_copy(p, &item->addr, sizeof(item->addr)); 3885 ceph_decode_addr(&item->addr); 3886 } 3887 3888 dout("%s %s%llu cookie %llu addr %s\n", __func__, 3889 ENTITY_NAME(item->name), item->cookie, 3890 ceph_pr_addr(&item->addr.in_addr)); 3891 return 0; 3892 } 3893 3894 static int decode_watchers(void **p, void *end, 3895 struct ceph_watch_item **watchers, 3896 u32 *num_watchers) 3897 { 3898 u8 struct_v; 3899 u32 struct_len; 3900 int i; 3901 int ret; 3902 3903 ret = ceph_start_decoding(p, end, 1, "obj_list_watch_response_t", 3904 &struct_v, &struct_len); 3905 if (ret) 3906 return ret; 3907 3908 *num_watchers = ceph_decode_32(p); 3909 *watchers = kcalloc(*num_watchers, sizeof(**watchers), GFP_NOIO); 3910 if (!*watchers) 3911 return -ENOMEM; 3912 3913 for (i = 0; i < *num_watchers; i++) { 3914 ret = decode_watcher(p, end, *watchers + i); 3915 if (ret) { 3916 kfree(*watchers); 3917 return ret; 3918 } 3919 } 3920 3921 return 0; 3922 } 3923 3924 /* 3925 * On success, the caller is responsible for: 3926 * 3927 * kfree(watchers); 3928 */ 3929 int ceph_osdc_list_watchers(struct ceph_osd_client *osdc, 3930 struct ceph_object_id *oid, 3931 struct ceph_object_locator *oloc, 3932 struct ceph_watch_item **watchers, 3933 u32 *num_watchers) 3934 { 3935 struct ceph_osd_request *req; 3936 struct page **pages; 3937 int ret; 3938 3939 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO); 3940 if (!req) 3941 return -ENOMEM; 3942 3943 ceph_oid_copy(&req->r_base_oid, oid); 3944 ceph_oloc_copy(&req->r_base_oloc, oloc); 3945 req->r_flags = CEPH_OSD_FLAG_READ; 3946 3947 ret = ceph_osdc_alloc_messages(req, GFP_NOIO); 3948 if (ret) 3949 goto out_put_req; 3950 3951 pages = ceph_alloc_page_vector(1, GFP_NOIO); 3952 if (IS_ERR(pages)) { 3953 ret = PTR_ERR(pages); 3954 goto out_put_req; 3955 } 3956 3957 osd_req_op_init(req, 0, CEPH_OSD_OP_LIST_WATCHERS, 0); 3958 ceph_osd_data_pages_init(osd_req_op_data(req, 0, list_watchers, 3959 response_data), 3960 pages, PAGE_SIZE, 0, false, true); 3961 3962 ceph_osdc_start_request(osdc, req, false); 3963 ret = ceph_osdc_wait_request(osdc, req); 3964 if (ret >= 0) { 3965 void *p = page_address(pages[0]); 3966 void *const end = p + req->r_ops[0].outdata_len; 3967 3968 ret = decode_watchers(&p, end, watchers, num_watchers); 3969 } 3970 3971 out_put_req: 3972 ceph_osdc_put_request(req); 3973 return ret; 3974 } 3975 EXPORT_SYMBOL(ceph_osdc_list_watchers); 3976 3977 /* 3978 * Call all pending notify callbacks - for use after a watch is 3979 * unregistered, to make sure no more callbacks for it will be invoked 3980 */ 3981 void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc) 3982 { 3983 dout("%s osdc %p\n", __func__, osdc); 3984 flush_workqueue(osdc->notify_wq); 3985 } 3986 EXPORT_SYMBOL(ceph_osdc_flush_notifies); 3987 3988 void ceph_osdc_maybe_request_map(struct ceph_osd_client *osdc) 3989 { 3990 down_read(&osdc->lock); 3991 maybe_request_map(osdc); 3992 up_read(&osdc->lock); 3993 } 3994 EXPORT_SYMBOL(ceph_osdc_maybe_request_map); 3995 3996 /* 3997 * Execute an OSD class method on an object. 3998 * 3999 * @flags: CEPH_OSD_FLAG_* 4000 * @resp_len: in/out param for reply length 4001 */ 4002 int ceph_osdc_call(struct ceph_osd_client *osdc, 4003 struct ceph_object_id *oid, 4004 struct ceph_object_locator *oloc, 4005 const char *class, const char *method, 4006 unsigned int flags, 4007 struct page *req_page, size_t req_len, 4008 struct page *resp_page, size_t *resp_len) 4009 { 4010 struct ceph_osd_request *req; 4011 int ret; 4012 4013 if (req_len > PAGE_SIZE || (resp_page && *resp_len > PAGE_SIZE)) 4014 return -E2BIG; 4015 4016 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO); 4017 if (!req) 4018 return -ENOMEM; 4019 4020 ceph_oid_copy(&req->r_base_oid, oid); 4021 ceph_oloc_copy(&req->r_base_oloc, oloc); 4022 req->r_flags = flags; 4023 4024 ret = ceph_osdc_alloc_messages(req, GFP_NOIO); 4025 if (ret) 4026 goto out_put_req; 4027 4028 osd_req_op_cls_init(req, 0, CEPH_OSD_OP_CALL, class, method); 4029 if (req_page) 4030 osd_req_op_cls_request_data_pages(req, 0, &req_page, req_len, 4031 0, false, false); 4032 if (resp_page) 4033 osd_req_op_cls_response_data_pages(req, 0, &resp_page, 4034 *resp_len, 0, false, false); 4035 4036 ceph_osdc_start_request(osdc, req, false); 4037 ret = ceph_osdc_wait_request(osdc, req); 4038 if (ret >= 0) { 4039 ret = req->r_ops[0].rval; 4040 if (resp_page) 4041 *resp_len = req->r_ops[0].outdata_len; 4042 } 4043 4044 out_put_req: 4045 ceph_osdc_put_request(req); 4046 return ret; 4047 } 4048 EXPORT_SYMBOL(ceph_osdc_call); 4049 4050 /* 4051 * init, shutdown 4052 */ 4053 int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client) 4054 { 4055 int err; 4056 4057 dout("init\n"); 4058 osdc->client = client; 4059 init_rwsem(&osdc->lock); 4060 osdc->osds = RB_ROOT; 4061 INIT_LIST_HEAD(&osdc->osd_lru); 4062 spin_lock_init(&osdc->osd_lru_lock); 4063 osd_init(&osdc->homeless_osd); 4064 osdc->homeless_osd.o_osdc = osdc; 4065 osdc->homeless_osd.o_osd = CEPH_HOMELESS_OSD; 4066 osdc->last_linger_id = CEPH_LINGER_ID_START; 4067 osdc->linger_requests = RB_ROOT; 4068 osdc->map_checks = RB_ROOT; 4069 osdc->linger_map_checks = RB_ROOT; 4070 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout); 4071 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout); 4072 4073 err = -ENOMEM; 4074 osdc->osdmap = ceph_osdmap_alloc(); 4075 if (!osdc->osdmap) 4076 goto out; 4077 4078 osdc->req_mempool = mempool_create_slab_pool(10, 4079 ceph_osd_request_cache); 4080 if (!osdc->req_mempool) 4081 goto out_map; 4082 4083 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP, 4084 PAGE_SIZE, 10, true, "osd_op"); 4085 if (err < 0) 4086 goto out_mempool; 4087 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY, 4088 PAGE_SIZE, 10, true, "osd_op_reply"); 4089 if (err < 0) 4090 goto out_msgpool; 4091 4092 err = -ENOMEM; 4093 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify"); 4094 if (!osdc->notify_wq) 4095 goto out_msgpool_reply; 4096 4097 schedule_delayed_work(&osdc->timeout_work, 4098 osdc->client->options->osd_keepalive_timeout); 4099 schedule_delayed_work(&osdc->osds_timeout_work, 4100 round_jiffies_relative(osdc->client->options->osd_idle_ttl)); 4101 4102 return 0; 4103 4104 out_msgpool_reply: 4105 ceph_msgpool_destroy(&osdc->msgpool_op_reply); 4106 out_msgpool: 4107 ceph_msgpool_destroy(&osdc->msgpool_op); 4108 out_mempool: 4109 mempool_destroy(osdc->req_mempool); 4110 out_map: 4111 ceph_osdmap_destroy(osdc->osdmap); 4112 out: 4113 return err; 4114 } 4115 4116 void ceph_osdc_stop(struct ceph_osd_client *osdc) 4117 { 4118 flush_workqueue(osdc->notify_wq); 4119 destroy_workqueue(osdc->notify_wq); 4120 cancel_delayed_work_sync(&osdc->timeout_work); 4121 cancel_delayed_work_sync(&osdc->osds_timeout_work); 4122 4123 down_write(&osdc->lock); 4124 while (!RB_EMPTY_ROOT(&osdc->osds)) { 4125 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds), 4126 struct ceph_osd, o_node); 4127 close_osd(osd); 4128 } 4129 up_write(&osdc->lock); 4130 WARN_ON(refcount_read(&osdc->homeless_osd.o_ref) != 1); 4131 osd_cleanup(&osdc->homeless_osd); 4132 4133 WARN_ON(!list_empty(&osdc->osd_lru)); 4134 WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_requests)); 4135 WARN_ON(!RB_EMPTY_ROOT(&osdc->map_checks)); 4136 WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_map_checks)); 4137 WARN_ON(atomic_read(&osdc->num_requests)); 4138 WARN_ON(atomic_read(&osdc->num_homeless)); 4139 4140 ceph_osdmap_destroy(osdc->osdmap); 4141 mempool_destroy(osdc->req_mempool); 4142 ceph_msgpool_destroy(&osdc->msgpool_op); 4143 ceph_msgpool_destroy(&osdc->msgpool_op_reply); 4144 } 4145 4146 /* 4147 * Read some contiguous pages. If we cross a stripe boundary, shorten 4148 * *plen. Return number of bytes read, or error. 4149 */ 4150 int ceph_osdc_readpages(struct ceph_osd_client *osdc, 4151 struct ceph_vino vino, struct ceph_file_layout *layout, 4152 u64 off, u64 *plen, 4153 u32 truncate_seq, u64 truncate_size, 4154 struct page **pages, int num_pages, int page_align) 4155 { 4156 struct ceph_osd_request *req; 4157 int rc = 0; 4158 4159 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino, 4160 vino.snap, off, *plen); 4161 req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 0, 1, 4162 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ, 4163 NULL, truncate_seq, truncate_size, 4164 false); 4165 if (IS_ERR(req)) 4166 return PTR_ERR(req); 4167 4168 /* it may be a short read due to an object boundary */ 4169 osd_req_op_extent_osd_data_pages(req, 0, 4170 pages, *plen, page_align, false, false); 4171 4172 dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n", 4173 off, *plen, *plen, page_align); 4174 4175 rc = ceph_osdc_start_request(osdc, req, false); 4176 if (!rc) 4177 rc = ceph_osdc_wait_request(osdc, req); 4178 4179 ceph_osdc_put_request(req); 4180 dout("readpages result %d\n", rc); 4181 return rc; 4182 } 4183 EXPORT_SYMBOL(ceph_osdc_readpages); 4184 4185 /* 4186 * do a synchronous write on N pages 4187 */ 4188 int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino, 4189 struct ceph_file_layout *layout, 4190 struct ceph_snap_context *snapc, 4191 u64 off, u64 len, 4192 u32 truncate_seq, u64 truncate_size, 4193 struct timespec *mtime, 4194 struct page **pages, int num_pages) 4195 { 4196 struct ceph_osd_request *req; 4197 int rc = 0; 4198 int page_align = off & ~PAGE_MASK; 4199 4200 req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 0, 1, 4201 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE, 4202 snapc, truncate_seq, truncate_size, 4203 true); 4204 if (IS_ERR(req)) 4205 return PTR_ERR(req); 4206 4207 /* it may be a short write due to an object boundary */ 4208 osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align, 4209 false, false); 4210 dout("writepages %llu~%llu (%llu bytes)\n", off, len, len); 4211 4212 req->r_mtime = *mtime; 4213 rc = ceph_osdc_start_request(osdc, req, true); 4214 if (!rc) 4215 rc = ceph_osdc_wait_request(osdc, req); 4216 4217 ceph_osdc_put_request(req); 4218 if (rc == 0) 4219 rc = len; 4220 dout("writepages result %d\n", rc); 4221 return rc; 4222 } 4223 EXPORT_SYMBOL(ceph_osdc_writepages); 4224 4225 int ceph_osdc_setup(void) 4226 { 4227 size_t size = sizeof(struct ceph_osd_request) + 4228 CEPH_OSD_SLAB_OPS * sizeof(struct ceph_osd_req_op); 4229 4230 BUG_ON(ceph_osd_request_cache); 4231 ceph_osd_request_cache = kmem_cache_create("ceph_osd_request", size, 4232 0, 0, NULL); 4233 4234 return ceph_osd_request_cache ? 0 : -ENOMEM; 4235 } 4236 EXPORT_SYMBOL(ceph_osdc_setup); 4237 4238 void ceph_osdc_cleanup(void) 4239 { 4240 BUG_ON(!ceph_osd_request_cache); 4241 kmem_cache_destroy(ceph_osd_request_cache); 4242 ceph_osd_request_cache = NULL; 4243 } 4244 EXPORT_SYMBOL(ceph_osdc_cleanup); 4245 4246 /* 4247 * handle incoming message 4248 */ 4249 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg) 4250 { 4251 struct ceph_osd *osd = con->private; 4252 struct ceph_osd_client *osdc = osd->o_osdc; 4253 int type = le16_to_cpu(msg->hdr.type); 4254 4255 switch (type) { 4256 case CEPH_MSG_OSD_MAP: 4257 ceph_osdc_handle_map(osdc, msg); 4258 break; 4259 case CEPH_MSG_OSD_OPREPLY: 4260 handle_reply(osd, msg); 4261 break; 4262 case CEPH_MSG_WATCH_NOTIFY: 4263 handle_watch_notify(osdc, msg); 4264 break; 4265 4266 default: 4267 pr_err("received unknown message type %d %s\n", type, 4268 ceph_msg_type_name(type)); 4269 } 4270 4271 ceph_msg_put(msg); 4272 } 4273 4274 /* 4275 * Lookup and return message for incoming reply. Don't try to do 4276 * anything about a larger than preallocated data portion of the 4277 * message at the moment - for now, just skip the message. 4278 */ 4279 static struct ceph_msg *get_reply(struct ceph_connection *con, 4280 struct ceph_msg_header *hdr, 4281 int *skip) 4282 { 4283 struct ceph_osd *osd = con->private; 4284 struct ceph_osd_client *osdc = osd->o_osdc; 4285 struct ceph_msg *m = NULL; 4286 struct ceph_osd_request *req; 4287 int front_len = le32_to_cpu(hdr->front_len); 4288 int data_len = le32_to_cpu(hdr->data_len); 4289 u64 tid = le64_to_cpu(hdr->tid); 4290 4291 down_read(&osdc->lock); 4292 if (!osd_registered(osd)) { 4293 dout("%s osd%d unknown, skipping\n", __func__, osd->o_osd); 4294 *skip = 1; 4295 goto out_unlock_osdc; 4296 } 4297 WARN_ON(osd->o_osd != le64_to_cpu(hdr->src.num)); 4298 4299 mutex_lock(&osd->lock); 4300 req = lookup_request(&osd->o_requests, tid); 4301 if (!req) { 4302 dout("%s osd%d tid %llu unknown, skipping\n", __func__, 4303 osd->o_osd, tid); 4304 *skip = 1; 4305 goto out_unlock_session; 4306 } 4307 4308 ceph_msg_revoke_incoming(req->r_reply); 4309 4310 if (front_len > req->r_reply->front_alloc_len) { 4311 pr_warn("%s osd%d tid %llu front %d > preallocated %d\n", 4312 __func__, osd->o_osd, req->r_tid, front_len, 4313 req->r_reply->front_alloc_len); 4314 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS, 4315 false); 4316 if (!m) 4317 goto out_unlock_session; 4318 ceph_msg_put(req->r_reply); 4319 req->r_reply = m; 4320 } 4321 4322 if (data_len > req->r_reply->data_length) { 4323 pr_warn("%s osd%d tid %llu data %d > preallocated %zu, skipping\n", 4324 __func__, osd->o_osd, req->r_tid, data_len, 4325 req->r_reply->data_length); 4326 m = NULL; 4327 *skip = 1; 4328 goto out_unlock_session; 4329 } 4330 4331 m = ceph_msg_get(req->r_reply); 4332 dout("get_reply tid %lld %p\n", tid, m); 4333 4334 out_unlock_session: 4335 mutex_unlock(&osd->lock); 4336 out_unlock_osdc: 4337 up_read(&osdc->lock); 4338 return m; 4339 } 4340 4341 /* 4342 * TODO: switch to a msg-owned pagelist 4343 */ 4344 static struct ceph_msg *alloc_msg_with_page_vector(struct ceph_msg_header *hdr) 4345 { 4346 struct ceph_msg *m; 4347 int type = le16_to_cpu(hdr->type); 4348 u32 front_len = le32_to_cpu(hdr->front_len); 4349 u32 data_len = le32_to_cpu(hdr->data_len); 4350 4351 m = ceph_msg_new(type, front_len, GFP_NOIO, false); 4352 if (!m) 4353 return NULL; 4354 4355 if (data_len) { 4356 struct page **pages; 4357 struct ceph_osd_data osd_data; 4358 4359 pages = ceph_alloc_page_vector(calc_pages_for(0, data_len), 4360 GFP_NOIO); 4361 if (IS_ERR(pages)) { 4362 ceph_msg_put(m); 4363 return NULL; 4364 } 4365 4366 ceph_osd_data_pages_init(&osd_data, pages, data_len, 0, false, 4367 false); 4368 ceph_osdc_msg_data_add(m, &osd_data); 4369 } 4370 4371 return m; 4372 } 4373 4374 static struct ceph_msg *alloc_msg(struct ceph_connection *con, 4375 struct ceph_msg_header *hdr, 4376 int *skip) 4377 { 4378 struct ceph_osd *osd = con->private; 4379 int type = le16_to_cpu(hdr->type); 4380 4381 *skip = 0; 4382 switch (type) { 4383 case CEPH_MSG_OSD_MAP: 4384 case CEPH_MSG_WATCH_NOTIFY: 4385 return alloc_msg_with_page_vector(hdr); 4386 case CEPH_MSG_OSD_OPREPLY: 4387 return get_reply(con, hdr, skip); 4388 default: 4389 pr_warn("%s osd%d unknown msg type %d, skipping\n", __func__, 4390 osd->o_osd, type); 4391 *skip = 1; 4392 return NULL; 4393 } 4394 } 4395 4396 /* 4397 * Wrappers to refcount containing ceph_osd struct 4398 */ 4399 static struct ceph_connection *get_osd_con(struct ceph_connection *con) 4400 { 4401 struct ceph_osd *osd = con->private; 4402 if (get_osd(osd)) 4403 return con; 4404 return NULL; 4405 } 4406 4407 static void put_osd_con(struct ceph_connection *con) 4408 { 4409 struct ceph_osd *osd = con->private; 4410 put_osd(osd); 4411 } 4412 4413 /* 4414 * authentication 4415 */ 4416 /* 4417 * Note: returned pointer is the address of a structure that's 4418 * managed separately. Caller must *not* attempt to free it. 4419 */ 4420 static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con, 4421 int *proto, int force_new) 4422 { 4423 struct ceph_osd *o = con->private; 4424 struct ceph_osd_client *osdc = o->o_osdc; 4425 struct ceph_auth_client *ac = osdc->client->monc.auth; 4426 struct ceph_auth_handshake *auth = &o->o_auth; 4427 4428 if (force_new && auth->authorizer) { 4429 ceph_auth_destroy_authorizer(auth->authorizer); 4430 auth->authorizer = NULL; 4431 } 4432 if (!auth->authorizer) { 4433 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD, 4434 auth); 4435 if (ret) 4436 return ERR_PTR(ret); 4437 } else { 4438 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD, 4439 auth); 4440 if (ret) 4441 return ERR_PTR(ret); 4442 } 4443 *proto = ac->protocol; 4444 4445 return auth; 4446 } 4447 4448 4449 static int verify_authorizer_reply(struct ceph_connection *con) 4450 { 4451 struct ceph_osd *o = con->private; 4452 struct ceph_osd_client *osdc = o->o_osdc; 4453 struct ceph_auth_client *ac = osdc->client->monc.auth; 4454 4455 return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer); 4456 } 4457 4458 static int invalidate_authorizer(struct ceph_connection *con) 4459 { 4460 struct ceph_osd *o = con->private; 4461 struct ceph_osd_client *osdc = o->o_osdc; 4462 struct ceph_auth_client *ac = osdc->client->monc.auth; 4463 4464 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD); 4465 return ceph_monc_validate_auth(&osdc->client->monc); 4466 } 4467 4468 static int osd_sign_message(struct ceph_msg *msg) 4469 { 4470 struct ceph_osd *o = msg->con->private; 4471 struct ceph_auth_handshake *auth = &o->o_auth; 4472 4473 return ceph_auth_sign_message(auth, msg); 4474 } 4475 4476 static int osd_check_message_signature(struct ceph_msg *msg) 4477 { 4478 struct ceph_osd *o = msg->con->private; 4479 struct ceph_auth_handshake *auth = &o->o_auth; 4480 4481 return ceph_auth_check_message_signature(auth, msg); 4482 } 4483 4484 static const struct ceph_connection_operations osd_con_ops = { 4485 .get = get_osd_con, 4486 .put = put_osd_con, 4487 .dispatch = dispatch, 4488 .get_authorizer = get_authorizer, 4489 .verify_authorizer_reply = verify_authorizer_reply, 4490 .invalidate_authorizer = invalidate_authorizer, 4491 .alloc_msg = alloc_msg, 4492 .sign_message = osd_sign_message, 4493 .check_message_signature = osd_check_message_signature, 4494 .fault = osd_fault, 4495 }; 4496