1 // SPDX-License-Identifier: GPL-2.0 2 3 #include <linux/ceph/ceph_debug.h> 4 5 #include <linux/module.h> 6 #include <linux/err.h> 7 #include <linux/highmem.h> 8 #include <linux/mm.h> 9 #include <linux/pagemap.h> 10 #include <linux/slab.h> 11 #include <linux/uaccess.h> 12 #ifdef CONFIG_BLOCK 13 #include <linux/bio.h> 14 #endif 15 16 #include <linux/ceph/ceph_features.h> 17 #include <linux/ceph/libceph.h> 18 #include <linux/ceph/osd_client.h> 19 #include <linux/ceph/messenger.h> 20 #include <linux/ceph/decode.h> 21 #include <linux/ceph/auth.h> 22 #include <linux/ceph/pagelist.h> 23 #include <linux/ceph/striper.h> 24 25 #define OSD_OPREPLY_FRONT_LEN 512 26 27 static struct kmem_cache *ceph_osd_request_cache; 28 29 static const struct ceph_connection_operations osd_con_ops; 30 31 /* 32 * Implement client access to distributed object storage cluster. 33 * 34 * All data objects are stored within a cluster/cloud of OSDs, or 35 * "object storage devices." (Note that Ceph OSDs have _nothing_ to 36 * do with the T10 OSD extensions to SCSI.) Ceph OSDs are simply 37 * remote daemons serving up and coordinating consistent and safe 38 * access to storage. 39 * 40 * Cluster membership and the mapping of data objects onto storage devices 41 * are described by the osd map. 42 * 43 * We keep track of pending OSD requests (read, write), resubmit 44 * requests to different OSDs when the cluster topology/data layout 45 * change, or retry the affected requests when the communications 46 * channel with an OSD is reset. 47 */ 48 49 static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req); 50 static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req); 51 static void link_linger(struct ceph_osd *osd, 52 struct ceph_osd_linger_request *lreq); 53 static void unlink_linger(struct ceph_osd *osd, 54 struct ceph_osd_linger_request *lreq); 55 static void clear_backoffs(struct ceph_osd *osd); 56 57 #if 1 58 static inline bool rwsem_is_wrlocked(struct rw_semaphore *sem) 59 { 60 bool wrlocked = true; 61 62 if (unlikely(down_read_trylock(sem))) { 63 wrlocked = false; 64 up_read(sem); 65 } 66 67 return wrlocked; 68 } 69 static inline void verify_osdc_locked(struct ceph_osd_client *osdc) 70 { 71 WARN_ON(!rwsem_is_locked(&osdc->lock)); 72 } 73 static inline void verify_osdc_wrlocked(struct ceph_osd_client *osdc) 74 { 75 WARN_ON(!rwsem_is_wrlocked(&osdc->lock)); 76 } 77 static inline void verify_osd_locked(struct ceph_osd *osd) 78 { 79 struct ceph_osd_client *osdc = osd->o_osdc; 80 81 WARN_ON(!(mutex_is_locked(&osd->lock) && 82 rwsem_is_locked(&osdc->lock)) && 83 !rwsem_is_wrlocked(&osdc->lock)); 84 } 85 static inline void verify_lreq_locked(struct ceph_osd_linger_request *lreq) 86 { 87 WARN_ON(!mutex_is_locked(&lreq->lock)); 88 } 89 #else 90 static inline void verify_osdc_locked(struct ceph_osd_client *osdc) { } 91 static inline void verify_osdc_wrlocked(struct ceph_osd_client *osdc) { } 92 static inline void verify_osd_locked(struct ceph_osd *osd) { } 93 static inline void verify_lreq_locked(struct ceph_osd_linger_request *lreq) { } 94 #endif 95 96 /* 97 * calculate the mapping of a file extent onto an object, and fill out the 98 * request accordingly. shorten extent as necessary if it crosses an 99 * object boundary. 100 * 101 * fill osd op in request message. 102 */ 103 static int calc_layout(struct ceph_file_layout *layout, u64 off, u64 *plen, 104 u64 *objnum, u64 *objoff, u64 *objlen) 105 { 106 u64 orig_len = *plen; 107 u32 xlen; 108 109 /* object extent? */ 110 ceph_calc_file_object_mapping(layout, off, orig_len, objnum, 111 objoff, &xlen); 112 *objlen = xlen; 113 if (*objlen < orig_len) { 114 *plen = *objlen; 115 dout(" skipping last %llu, final file extent %llu~%llu\n", 116 orig_len - *plen, off, *plen); 117 } 118 119 dout("calc_layout objnum=%llx %llu~%llu\n", *objnum, *objoff, *objlen); 120 return 0; 121 } 122 123 static void ceph_osd_data_init(struct ceph_osd_data *osd_data) 124 { 125 memset(osd_data, 0, sizeof (*osd_data)); 126 osd_data->type = CEPH_OSD_DATA_TYPE_NONE; 127 } 128 129 static void ceph_osd_data_pages_init(struct ceph_osd_data *osd_data, 130 struct page **pages, u64 length, u32 alignment, 131 bool pages_from_pool, bool own_pages) 132 { 133 osd_data->type = CEPH_OSD_DATA_TYPE_PAGES; 134 osd_data->pages = pages; 135 osd_data->length = length; 136 osd_data->alignment = alignment; 137 osd_data->pages_from_pool = pages_from_pool; 138 osd_data->own_pages = own_pages; 139 } 140 141 static void ceph_osd_data_pagelist_init(struct ceph_osd_data *osd_data, 142 struct ceph_pagelist *pagelist) 143 { 144 osd_data->type = CEPH_OSD_DATA_TYPE_PAGELIST; 145 osd_data->pagelist = pagelist; 146 } 147 148 #ifdef CONFIG_BLOCK 149 static void ceph_osd_data_bio_init(struct ceph_osd_data *osd_data, 150 struct ceph_bio_iter *bio_pos, 151 u32 bio_length) 152 { 153 osd_data->type = CEPH_OSD_DATA_TYPE_BIO; 154 osd_data->bio_pos = *bio_pos; 155 osd_data->bio_length = bio_length; 156 } 157 #endif /* CONFIG_BLOCK */ 158 159 static void ceph_osd_data_bvecs_init(struct ceph_osd_data *osd_data, 160 struct ceph_bvec_iter *bvec_pos) 161 { 162 osd_data->type = CEPH_OSD_DATA_TYPE_BVECS; 163 osd_data->bvec_pos = *bvec_pos; 164 } 165 166 #define osd_req_op_data(oreq, whch, typ, fld) \ 167 ({ \ 168 struct ceph_osd_request *__oreq = (oreq); \ 169 unsigned int __whch = (whch); \ 170 BUG_ON(__whch >= __oreq->r_num_ops); \ 171 &__oreq->r_ops[__whch].typ.fld; \ 172 }) 173 174 static struct ceph_osd_data * 175 osd_req_op_raw_data_in(struct ceph_osd_request *osd_req, unsigned int which) 176 { 177 BUG_ON(which >= osd_req->r_num_ops); 178 179 return &osd_req->r_ops[which].raw_data_in; 180 } 181 182 struct ceph_osd_data * 183 osd_req_op_extent_osd_data(struct ceph_osd_request *osd_req, 184 unsigned int which) 185 { 186 return osd_req_op_data(osd_req, which, extent, osd_data); 187 } 188 EXPORT_SYMBOL(osd_req_op_extent_osd_data); 189 190 void osd_req_op_raw_data_in_pages(struct ceph_osd_request *osd_req, 191 unsigned int which, struct page **pages, 192 u64 length, u32 alignment, 193 bool pages_from_pool, bool own_pages) 194 { 195 struct ceph_osd_data *osd_data; 196 197 osd_data = osd_req_op_raw_data_in(osd_req, which); 198 ceph_osd_data_pages_init(osd_data, pages, length, alignment, 199 pages_from_pool, own_pages); 200 } 201 EXPORT_SYMBOL(osd_req_op_raw_data_in_pages); 202 203 void osd_req_op_extent_osd_data_pages(struct ceph_osd_request *osd_req, 204 unsigned int which, struct page **pages, 205 u64 length, u32 alignment, 206 bool pages_from_pool, bool own_pages) 207 { 208 struct ceph_osd_data *osd_data; 209 210 osd_data = osd_req_op_data(osd_req, which, extent, osd_data); 211 ceph_osd_data_pages_init(osd_data, pages, length, alignment, 212 pages_from_pool, own_pages); 213 } 214 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pages); 215 216 void osd_req_op_extent_osd_data_pagelist(struct ceph_osd_request *osd_req, 217 unsigned int which, struct ceph_pagelist *pagelist) 218 { 219 struct ceph_osd_data *osd_data; 220 221 osd_data = osd_req_op_data(osd_req, which, extent, osd_data); 222 ceph_osd_data_pagelist_init(osd_data, pagelist); 223 } 224 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pagelist); 225 226 #ifdef CONFIG_BLOCK 227 void osd_req_op_extent_osd_data_bio(struct ceph_osd_request *osd_req, 228 unsigned int which, 229 struct ceph_bio_iter *bio_pos, 230 u32 bio_length) 231 { 232 struct ceph_osd_data *osd_data; 233 234 osd_data = osd_req_op_data(osd_req, which, extent, osd_data); 235 ceph_osd_data_bio_init(osd_data, bio_pos, bio_length); 236 } 237 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bio); 238 #endif /* CONFIG_BLOCK */ 239 240 void osd_req_op_extent_osd_data_bvec_pos(struct ceph_osd_request *osd_req, 241 unsigned int which, 242 struct ceph_bvec_iter *bvec_pos) 243 { 244 struct ceph_osd_data *osd_data; 245 246 osd_data = osd_req_op_data(osd_req, which, extent, osd_data); 247 ceph_osd_data_bvecs_init(osd_data, bvec_pos); 248 } 249 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bvec_pos); 250 251 static void osd_req_op_cls_request_info_pagelist( 252 struct ceph_osd_request *osd_req, 253 unsigned int which, struct ceph_pagelist *pagelist) 254 { 255 struct ceph_osd_data *osd_data; 256 257 osd_data = osd_req_op_data(osd_req, which, cls, request_info); 258 ceph_osd_data_pagelist_init(osd_data, pagelist); 259 } 260 261 void osd_req_op_cls_request_data_pagelist( 262 struct ceph_osd_request *osd_req, 263 unsigned int which, struct ceph_pagelist *pagelist) 264 { 265 struct ceph_osd_data *osd_data; 266 267 osd_data = osd_req_op_data(osd_req, which, cls, request_data); 268 ceph_osd_data_pagelist_init(osd_data, pagelist); 269 osd_req->r_ops[which].cls.indata_len += pagelist->length; 270 osd_req->r_ops[which].indata_len += pagelist->length; 271 } 272 EXPORT_SYMBOL(osd_req_op_cls_request_data_pagelist); 273 274 void osd_req_op_cls_request_data_pages(struct ceph_osd_request *osd_req, 275 unsigned int which, struct page **pages, u64 length, 276 u32 alignment, bool pages_from_pool, bool own_pages) 277 { 278 struct ceph_osd_data *osd_data; 279 280 osd_data = osd_req_op_data(osd_req, which, cls, request_data); 281 ceph_osd_data_pages_init(osd_data, pages, length, alignment, 282 pages_from_pool, own_pages); 283 osd_req->r_ops[which].cls.indata_len += length; 284 osd_req->r_ops[which].indata_len += length; 285 } 286 EXPORT_SYMBOL(osd_req_op_cls_request_data_pages); 287 288 void osd_req_op_cls_request_data_bvecs(struct ceph_osd_request *osd_req, 289 unsigned int which, 290 struct bio_vec *bvecs, u32 bytes) 291 { 292 struct ceph_osd_data *osd_data; 293 struct ceph_bvec_iter it = { 294 .bvecs = bvecs, 295 .iter = { .bi_size = bytes }, 296 }; 297 298 osd_data = osd_req_op_data(osd_req, which, cls, request_data); 299 ceph_osd_data_bvecs_init(osd_data, &it); 300 osd_req->r_ops[which].cls.indata_len += bytes; 301 osd_req->r_ops[which].indata_len += bytes; 302 } 303 EXPORT_SYMBOL(osd_req_op_cls_request_data_bvecs); 304 305 void osd_req_op_cls_response_data_pages(struct ceph_osd_request *osd_req, 306 unsigned int which, struct page **pages, u64 length, 307 u32 alignment, bool pages_from_pool, bool own_pages) 308 { 309 struct ceph_osd_data *osd_data; 310 311 osd_data = osd_req_op_data(osd_req, which, cls, response_data); 312 ceph_osd_data_pages_init(osd_data, pages, length, alignment, 313 pages_from_pool, own_pages); 314 } 315 EXPORT_SYMBOL(osd_req_op_cls_response_data_pages); 316 317 static u64 ceph_osd_data_length(struct ceph_osd_data *osd_data) 318 { 319 switch (osd_data->type) { 320 case CEPH_OSD_DATA_TYPE_NONE: 321 return 0; 322 case CEPH_OSD_DATA_TYPE_PAGES: 323 return osd_data->length; 324 case CEPH_OSD_DATA_TYPE_PAGELIST: 325 return (u64)osd_data->pagelist->length; 326 #ifdef CONFIG_BLOCK 327 case CEPH_OSD_DATA_TYPE_BIO: 328 return (u64)osd_data->bio_length; 329 #endif /* CONFIG_BLOCK */ 330 case CEPH_OSD_DATA_TYPE_BVECS: 331 return osd_data->bvec_pos.iter.bi_size; 332 default: 333 WARN(true, "unrecognized data type %d\n", (int)osd_data->type); 334 return 0; 335 } 336 } 337 338 static void ceph_osd_data_release(struct ceph_osd_data *osd_data) 339 { 340 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES && osd_data->own_pages) { 341 int num_pages; 342 343 num_pages = calc_pages_for((u64)osd_data->alignment, 344 (u64)osd_data->length); 345 ceph_release_page_vector(osd_data->pages, num_pages); 346 } 347 ceph_osd_data_init(osd_data); 348 } 349 350 static void osd_req_op_data_release(struct ceph_osd_request *osd_req, 351 unsigned int which) 352 { 353 struct ceph_osd_req_op *op; 354 355 BUG_ON(which >= osd_req->r_num_ops); 356 op = &osd_req->r_ops[which]; 357 358 switch (op->op) { 359 case CEPH_OSD_OP_READ: 360 case CEPH_OSD_OP_WRITE: 361 case CEPH_OSD_OP_WRITEFULL: 362 ceph_osd_data_release(&op->extent.osd_data); 363 break; 364 case CEPH_OSD_OP_CALL: 365 ceph_osd_data_release(&op->cls.request_info); 366 ceph_osd_data_release(&op->cls.request_data); 367 ceph_osd_data_release(&op->cls.response_data); 368 break; 369 case CEPH_OSD_OP_SETXATTR: 370 case CEPH_OSD_OP_CMPXATTR: 371 ceph_osd_data_release(&op->xattr.osd_data); 372 break; 373 case CEPH_OSD_OP_STAT: 374 ceph_osd_data_release(&op->raw_data_in); 375 break; 376 case CEPH_OSD_OP_NOTIFY_ACK: 377 ceph_osd_data_release(&op->notify_ack.request_data); 378 break; 379 case CEPH_OSD_OP_NOTIFY: 380 ceph_osd_data_release(&op->notify.request_data); 381 ceph_osd_data_release(&op->notify.response_data); 382 break; 383 case CEPH_OSD_OP_LIST_WATCHERS: 384 ceph_osd_data_release(&op->list_watchers.response_data); 385 break; 386 default: 387 break; 388 } 389 } 390 391 /* 392 * Assumes @t is zero-initialized. 393 */ 394 static void target_init(struct ceph_osd_request_target *t) 395 { 396 ceph_oid_init(&t->base_oid); 397 ceph_oloc_init(&t->base_oloc); 398 ceph_oid_init(&t->target_oid); 399 ceph_oloc_init(&t->target_oloc); 400 401 ceph_osds_init(&t->acting); 402 ceph_osds_init(&t->up); 403 t->size = -1; 404 t->min_size = -1; 405 406 t->osd = CEPH_HOMELESS_OSD; 407 } 408 409 static void target_copy(struct ceph_osd_request_target *dest, 410 const struct ceph_osd_request_target *src) 411 { 412 ceph_oid_copy(&dest->base_oid, &src->base_oid); 413 ceph_oloc_copy(&dest->base_oloc, &src->base_oloc); 414 ceph_oid_copy(&dest->target_oid, &src->target_oid); 415 ceph_oloc_copy(&dest->target_oloc, &src->target_oloc); 416 417 dest->pgid = src->pgid; /* struct */ 418 dest->spgid = src->spgid; /* struct */ 419 dest->pg_num = src->pg_num; 420 dest->pg_num_mask = src->pg_num_mask; 421 ceph_osds_copy(&dest->acting, &src->acting); 422 ceph_osds_copy(&dest->up, &src->up); 423 dest->size = src->size; 424 dest->min_size = src->min_size; 425 dest->sort_bitwise = src->sort_bitwise; 426 427 dest->flags = src->flags; 428 dest->paused = src->paused; 429 430 dest->epoch = src->epoch; 431 dest->last_force_resend = src->last_force_resend; 432 433 dest->osd = src->osd; 434 } 435 436 static void target_destroy(struct ceph_osd_request_target *t) 437 { 438 ceph_oid_destroy(&t->base_oid); 439 ceph_oloc_destroy(&t->base_oloc); 440 ceph_oid_destroy(&t->target_oid); 441 ceph_oloc_destroy(&t->target_oloc); 442 } 443 444 /* 445 * requests 446 */ 447 static void request_release_checks(struct ceph_osd_request *req) 448 { 449 WARN_ON(!RB_EMPTY_NODE(&req->r_node)); 450 WARN_ON(!RB_EMPTY_NODE(&req->r_mc_node)); 451 WARN_ON(!list_empty(&req->r_unsafe_item)); 452 WARN_ON(req->r_osd); 453 } 454 455 static void ceph_osdc_release_request(struct kref *kref) 456 { 457 struct ceph_osd_request *req = container_of(kref, 458 struct ceph_osd_request, r_kref); 459 unsigned int which; 460 461 dout("%s %p (r_request %p r_reply %p)\n", __func__, req, 462 req->r_request, req->r_reply); 463 request_release_checks(req); 464 465 if (req->r_request) 466 ceph_msg_put(req->r_request); 467 if (req->r_reply) 468 ceph_msg_put(req->r_reply); 469 470 for (which = 0; which < req->r_num_ops; which++) 471 osd_req_op_data_release(req, which); 472 473 target_destroy(&req->r_t); 474 ceph_put_snap_context(req->r_snapc); 475 476 if (req->r_mempool) 477 mempool_free(req, req->r_osdc->req_mempool); 478 else if (req->r_num_ops <= CEPH_OSD_SLAB_OPS) 479 kmem_cache_free(ceph_osd_request_cache, req); 480 else 481 kfree(req); 482 } 483 484 void ceph_osdc_get_request(struct ceph_osd_request *req) 485 { 486 dout("%s %p (was %d)\n", __func__, req, 487 kref_read(&req->r_kref)); 488 kref_get(&req->r_kref); 489 } 490 EXPORT_SYMBOL(ceph_osdc_get_request); 491 492 void ceph_osdc_put_request(struct ceph_osd_request *req) 493 { 494 if (req) { 495 dout("%s %p (was %d)\n", __func__, req, 496 kref_read(&req->r_kref)); 497 kref_put(&req->r_kref, ceph_osdc_release_request); 498 } 499 } 500 EXPORT_SYMBOL(ceph_osdc_put_request); 501 502 static void request_init(struct ceph_osd_request *req) 503 { 504 /* req only, each op is zeroed in _osd_req_op_init() */ 505 memset(req, 0, sizeof(*req)); 506 507 kref_init(&req->r_kref); 508 init_completion(&req->r_completion); 509 RB_CLEAR_NODE(&req->r_node); 510 RB_CLEAR_NODE(&req->r_mc_node); 511 INIT_LIST_HEAD(&req->r_unsafe_item); 512 513 target_init(&req->r_t); 514 } 515 516 /* 517 * This is ugly, but it allows us to reuse linger registration and ping 518 * requests, keeping the structure of the code around send_linger{_ping}() 519 * reasonable. Setting up a min_nr=2 mempool for each linger request 520 * and dealing with copying ops (this blasts req only, watch op remains 521 * intact) isn't any better. 522 */ 523 static void request_reinit(struct ceph_osd_request *req) 524 { 525 struct ceph_osd_client *osdc = req->r_osdc; 526 bool mempool = req->r_mempool; 527 unsigned int num_ops = req->r_num_ops; 528 u64 snapid = req->r_snapid; 529 struct ceph_snap_context *snapc = req->r_snapc; 530 bool linger = req->r_linger; 531 struct ceph_msg *request_msg = req->r_request; 532 struct ceph_msg *reply_msg = req->r_reply; 533 534 dout("%s req %p\n", __func__, req); 535 WARN_ON(kref_read(&req->r_kref) != 1); 536 request_release_checks(req); 537 538 WARN_ON(kref_read(&request_msg->kref) != 1); 539 WARN_ON(kref_read(&reply_msg->kref) != 1); 540 target_destroy(&req->r_t); 541 542 request_init(req); 543 req->r_osdc = osdc; 544 req->r_mempool = mempool; 545 req->r_num_ops = num_ops; 546 req->r_snapid = snapid; 547 req->r_snapc = snapc; 548 req->r_linger = linger; 549 req->r_request = request_msg; 550 req->r_reply = reply_msg; 551 } 552 553 struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc, 554 struct ceph_snap_context *snapc, 555 unsigned int num_ops, 556 bool use_mempool, 557 gfp_t gfp_flags) 558 { 559 struct ceph_osd_request *req; 560 561 if (use_mempool) { 562 BUG_ON(num_ops > CEPH_OSD_SLAB_OPS); 563 req = mempool_alloc(osdc->req_mempool, gfp_flags); 564 } else if (num_ops <= CEPH_OSD_SLAB_OPS) { 565 req = kmem_cache_alloc(ceph_osd_request_cache, gfp_flags); 566 } else { 567 BUG_ON(num_ops > CEPH_OSD_MAX_OPS); 568 req = kmalloc(sizeof(*req) + num_ops * sizeof(req->r_ops[0]), 569 gfp_flags); 570 } 571 if (unlikely(!req)) 572 return NULL; 573 574 request_init(req); 575 req->r_osdc = osdc; 576 req->r_mempool = use_mempool; 577 req->r_num_ops = num_ops; 578 req->r_snapid = CEPH_NOSNAP; 579 req->r_snapc = ceph_get_snap_context(snapc); 580 581 dout("%s req %p\n", __func__, req); 582 return req; 583 } 584 EXPORT_SYMBOL(ceph_osdc_alloc_request); 585 586 static int ceph_oloc_encoding_size(const struct ceph_object_locator *oloc) 587 { 588 return 8 + 4 + 4 + 4 + (oloc->pool_ns ? oloc->pool_ns->len : 0); 589 } 590 591 int ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp) 592 { 593 struct ceph_osd_client *osdc = req->r_osdc; 594 struct ceph_msg *msg; 595 int msg_size; 596 597 WARN_ON(ceph_oid_empty(&req->r_base_oid)); 598 WARN_ON(ceph_oloc_empty(&req->r_base_oloc)); 599 600 /* create request message */ 601 msg_size = CEPH_ENCODING_START_BLK_LEN + 602 CEPH_PGID_ENCODING_LEN + 1; /* spgid */ 603 msg_size += 4 + 4 + 4; /* hash, osdmap_epoch, flags */ 604 msg_size += CEPH_ENCODING_START_BLK_LEN + 605 sizeof(struct ceph_osd_reqid); /* reqid */ 606 msg_size += sizeof(struct ceph_blkin_trace_info); /* trace */ 607 msg_size += 4 + sizeof(struct ceph_timespec); /* client_inc, mtime */ 608 msg_size += CEPH_ENCODING_START_BLK_LEN + 609 ceph_oloc_encoding_size(&req->r_base_oloc); /* oloc */ 610 msg_size += 4 + req->r_base_oid.name_len; /* oid */ 611 msg_size += 2 + req->r_num_ops * sizeof(struct ceph_osd_op); 612 msg_size += 8; /* snapid */ 613 msg_size += 8; /* snap_seq */ 614 msg_size += 4 + 8 * (req->r_snapc ? req->r_snapc->num_snaps : 0); 615 msg_size += 4 + 8; /* retry_attempt, features */ 616 617 if (req->r_mempool) 618 msg = ceph_msgpool_get(&osdc->msgpool_op, 0); 619 else 620 msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp, true); 621 if (!msg) 622 return -ENOMEM; 623 624 memset(msg->front.iov_base, 0, msg->front.iov_len); 625 req->r_request = msg; 626 627 /* create reply message */ 628 msg_size = OSD_OPREPLY_FRONT_LEN; 629 msg_size += req->r_base_oid.name_len; 630 msg_size += req->r_num_ops * sizeof(struct ceph_osd_op); 631 632 if (req->r_mempool) 633 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0); 634 else 635 msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, msg_size, gfp, true); 636 if (!msg) 637 return -ENOMEM; 638 639 req->r_reply = msg; 640 641 return 0; 642 } 643 EXPORT_SYMBOL(ceph_osdc_alloc_messages); 644 645 static bool osd_req_opcode_valid(u16 opcode) 646 { 647 switch (opcode) { 648 #define GENERATE_CASE(op, opcode, str) case CEPH_OSD_OP_##op: return true; 649 __CEPH_FORALL_OSD_OPS(GENERATE_CASE) 650 #undef GENERATE_CASE 651 default: 652 return false; 653 } 654 } 655 656 /* 657 * This is an osd op init function for opcodes that have no data or 658 * other information associated with them. It also serves as a 659 * common init routine for all the other init functions, below. 660 */ 661 static struct ceph_osd_req_op * 662 _osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which, 663 u16 opcode, u32 flags) 664 { 665 struct ceph_osd_req_op *op; 666 667 BUG_ON(which >= osd_req->r_num_ops); 668 BUG_ON(!osd_req_opcode_valid(opcode)); 669 670 op = &osd_req->r_ops[which]; 671 memset(op, 0, sizeof (*op)); 672 op->op = opcode; 673 op->flags = flags; 674 675 return op; 676 } 677 678 void osd_req_op_init(struct ceph_osd_request *osd_req, 679 unsigned int which, u16 opcode, u32 flags) 680 { 681 (void)_osd_req_op_init(osd_req, which, opcode, flags); 682 } 683 EXPORT_SYMBOL(osd_req_op_init); 684 685 void osd_req_op_extent_init(struct ceph_osd_request *osd_req, 686 unsigned int which, u16 opcode, 687 u64 offset, u64 length, 688 u64 truncate_size, u32 truncate_seq) 689 { 690 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, 691 opcode, 0); 692 size_t payload_len = 0; 693 694 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE && 695 opcode != CEPH_OSD_OP_WRITEFULL && opcode != CEPH_OSD_OP_ZERO && 696 opcode != CEPH_OSD_OP_TRUNCATE); 697 698 op->extent.offset = offset; 699 op->extent.length = length; 700 op->extent.truncate_size = truncate_size; 701 op->extent.truncate_seq = truncate_seq; 702 if (opcode == CEPH_OSD_OP_WRITE || opcode == CEPH_OSD_OP_WRITEFULL) 703 payload_len += length; 704 705 op->indata_len = payload_len; 706 } 707 EXPORT_SYMBOL(osd_req_op_extent_init); 708 709 void osd_req_op_extent_update(struct ceph_osd_request *osd_req, 710 unsigned int which, u64 length) 711 { 712 struct ceph_osd_req_op *op; 713 u64 previous; 714 715 BUG_ON(which >= osd_req->r_num_ops); 716 op = &osd_req->r_ops[which]; 717 previous = op->extent.length; 718 719 if (length == previous) 720 return; /* Nothing to do */ 721 BUG_ON(length > previous); 722 723 op->extent.length = length; 724 if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL) 725 op->indata_len -= previous - length; 726 } 727 EXPORT_SYMBOL(osd_req_op_extent_update); 728 729 void osd_req_op_extent_dup_last(struct ceph_osd_request *osd_req, 730 unsigned int which, u64 offset_inc) 731 { 732 struct ceph_osd_req_op *op, *prev_op; 733 734 BUG_ON(which + 1 >= osd_req->r_num_ops); 735 736 prev_op = &osd_req->r_ops[which]; 737 op = _osd_req_op_init(osd_req, which + 1, prev_op->op, prev_op->flags); 738 /* dup previous one */ 739 op->indata_len = prev_op->indata_len; 740 op->outdata_len = prev_op->outdata_len; 741 op->extent = prev_op->extent; 742 /* adjust offset */ 743 op->extent.offset += offset_inc; 744 op->extent.length -= offset_inc; 745 746 if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL) 747 op->indata_len -= offset_inc; 748 } 749 EXPORT_SYMBOL(osd_req_op_extent_dup_last); 750 751 void osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which, 752 u16 opcode, const char *class, const char *method) 753 { 754 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, 755 opcode, 0); 756 struct ceph_pagelist *pagelist; 757 size_t payload_len = 0; 758 size_t size; 759 760 BUG_ON(opcode != CEPH_OSD_OP_CALL); 761 762 pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS); 763 BUG_ON(!pagelist); 764 ceph_pagelist_init(pagelist); 765 766 op->cls.class_name = class; 767 size = strlen(class); 768 BUG_ON(size > (size_t) U8_MAX); 769 op->cls.class_len = size; 770 ceph_pagelist_append(pagelist, class, size); 771 payload_len += size; 772 773 op->cls.method_name = method; 774 size = strlen(method); 775 BUG_ON(size > (size_t) U8_MAX); 776 op->cls.method_len = size; 777 ceph_pagelist_append(pagelist, method, size); 778 payload_len += size; 779 780 osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist); 781 782 op->indata_len = payload_len; 783 } 784 EXPORT_SYMBOL(osd_req_op_cls_init); 785 786 int osd_req_op_xattr_init(struct ceph_osd_request *osd_req, unsigned int which, 787 u16 opcode, const char *name, const void *value, 788 size_t size, u8 cmp_op, u8 cmp_mode) 789 { 790 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, 791 opcode, 0); 792 struct ceph_pagelist *pagelist; 793 size_t payload_len; 794 795 BUG_ON(opcode != CEPH_OSD_OP_SETXATTR && opcode != CEPH_OSD_OP_CMPXATTR); 796 797 pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS); 798 if (!pagelist) 799 return -ENOMEM; 800 801 ceph_pagelist_init(pagelist); 802 803 payload_len = strlen(name); 804 op->xattr.name_len = payload_len; 805 ceph_pagelist_append(pagelist, name, payload_len); 806 807 op->xattr.value_len = size; 808 ceph_pagelist_append(pagelist, value, size); 809 payload_len += size; 810 811 op->xattr.cmp_op = cmp_op; 812 op->xattr.cmp_mode = cmp_mode; 813 814 ceph_osd_data_pagelist_init(&op->xattr.osd_data, pagelist); 815 op->indata_len = payload_len; 816 return 0; 817 } 818 EXPORT_SYMBOL(osd_req_op_xattr_init); 819 820 /* 821 * @watch_opcode: CEPH_OSD_WATCH_OP_* 822 */ 823 static void osd_req_op_watch_init(struct ceph_osd_request *req, int which, 824 u64 cookie, u8 watch_opcode) 825 { 826 struct ceph_osd_req_op *op; 827 828 op = _osd_req_op_init(req, which, CEPH_OSD_OP_WATCH, 0); 829 op->watch.cookie = cookie; 830 op->watch.op = watch_opcode; 831 op->watch.gen = 0; 832 } 833 834 void osd_req_op_alloc_hint_init(struct ceph_osd_request *osd_req, 835 unsigned int which, 836 u64 expected_object_size, 837 u64 expected_write_size) 838 { 839 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, 840 CEPH_OSD_OP_SETALLOCHINT, 841 0); 842 843 op->alloc_hint.expected_object_size = expected_object_size; 844 op->alloc_hint.expected_write_size = expected_write_size; 845 846 /* 847 * CEPH_OSD_OP_SETALLOCHINT op is advisory and therefore deemed 848 * not worth a feature bit. Set FAILOK per-op flag to make 849 * sure older osds don't trip over an unsupported opcode. 850 */ 851 op->flags |= CEPH_OSD_OP_FLAG_FAILOK; 852 } 853 EXPORT_SYMBOL(osd_req_op_alloc_hint_init); 854 855 static void ceph_osdc_msg_data_add(struct ceph_msg *msg, 856 struct ceph_osd_data *osd_data) 857 { 858 u64 length = ceph_osd_data_length(osd_data); 859 860 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) { 861 BUG_ON(length > (u64) SIZE_MAX); 862 if (length) 863 ceph_msg_data_add_pages(msg, osd_data->pages, 864 length, osd_data->alignment); 865 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) { 866 BUG_ON(!length); 867 ceph_msg_data_add_pagelist(msg, osd_data->pagelist); 868 #ifdef CONFIG_BLOCK 869 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) { 870 ceph_msg_data_add_bio(msg, &osd_data->bio_pos, length); 871 #endif 872 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BVECS) { 873 ceph_msg_data_add_bvecs(msg, &osd_data->bvec_pos); 874 } else { 875 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE); 876 } 877 } 878 879 static u32 osd_req_encode_op(struct ceph_osd_op *dst, 880 const struct ceph_osd_req_op *src) 881 { 882 if (WARN_ON(!osd_req_opcode_valid(src->op))) { 883 pr_err("unrecognized osd opcode %d\n", src->op); 884 885 return 0; 886 } 887 888 switch (src->op) { 889 case CEPH_OSD_OP_STAT: 890 break; 891 case CEPH_OSD_OP_READ: 892 case CEPH_OSD_OP_WRITE: 893 case CEPH_OSD_OP_WRITEFULL: 894 case CEPH_OSD_OP_ZERO: 895 case CEPH_OSD_OP_TRUNCATE: 896 dst->extent.offset = cpu_to_le64(src->extent.offset); 897 dst->extent.length = cpu_to_le64(src->extent.length); 898 dst->extent.truncate_size = 899 cpu_to_le64(src->extent.truncate_size); 900 dst->extent.truncate_seq = 901 cpu_to_le32(src->extent.truncate_seq); 902 break; 903 case CEPH_OSD_OP_CALL: 904 dst->cls.class_len = src->cls.class_len; 905 dst->cls.method_len = src->cls.method_len; 906 dst->cls.indata_len = cpu_to_le32(src->cls.indata_len); 907 break; 908 case CEPH_OSD_OP_WATCH: 909 dst->watch.cookie = cpu_to_le64(src->watch.cookie); 910 dst->watch.ver = cpu_to_le64(0); 911 dst->watch.op = src->watch.op; 912 dst->watch.gen = cpu_to_le32(src->watch.gen); 913 break; 914 case CEPH_OSD_OP_NOTIFY_ACK: 915 break; 916 case CEPH_OSD_OP_NOTIFY: 917 dst->notify.cookie = cpu_to_le64(src->notify.cookie); 918 break; 919 case CEPH_OSD_OP_LIST_WATCHERS: 920 break; 921 case CEPH_OSD_OP_SETALLOCHINT: 922 dst->alloc_hint.expected_object_size = 923 cpu_to_le64(src->alloc_hint.expected_object_size); 924 dst->alloc_hint.expected_write_size = 925 cpu_to_le64(src->alloc_hint.expected_write_size); 926 break; 927 case CEPH_OSD_OP_SETXATTR: 928 case CEPH_OSD_OP_CMPXATTR: 929 dst->xattr.name_len = cpu_to_le32(src->xattr.name_len); 930 dst->xattr.value_len = cpu_to_le32(src->xattr.value_len); 931 dst->xattr.cmp_op = src->xattr.cmp_op; 932 dst->xattr.cmp_mode = src->xattr.cmp_mode; 933 break; 934 case CEPH_OSD_OP_CREATE: 935 case CEPH_OSD_OP_DELETE: 936 break; 937 default: 938 pr_err("unsupported osd opcode %s\n", 939 ceph_osd_op_name(src->op)); 940 WARN_ON(1); 941 942 return 0; 943 } 944 945 dst->op = cpu_to_le16(src->op); 946 dst->flags = cpu_to_le32(src->flags); 947 dst->payload_len = cpu_to_le32(src->indata_len); 948 949 return src->indata_len; 950 } 951 952 /* 953 * build new request AND message, calculate layout, and adjust file 954 * extent as needed. 955 * 956 * if the file was recently truncated, we include information about its 957 * old and new size so that the object can be updated appropriately. (we 958 * avoid synchronously deleting truncated objects because it's slow.) 959 */ 960 struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc, 961 struct ceph_file_layout *layout, 962 struct ceph_vino vino, 963 u64 off, u64 *plen, 964 unsigned int which, int num_ops, 965 int opcode, int flags, 966 struct ceph_snap_context *snapc, 967 u32 truncate_seq, 968 u64 truncate_size, 969 bool use_mempool) 970 { 971 struct ceph_osd_request *req; 972 u64 objnum = 0; 973 u64 objoff = 0; 974 u64 objlen = 0; 975 int r; 976 977 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE && 978 opcode != CEPH_OSD_OP_ZERO && opcode != CEPH_OSD_OP_TRUNCATE && 979 opcode != CEPH_OSD_OP_CREATE && opcode != CEPH_OSD_OP_DELETE); 980 981 req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool, 982 GFP_NOFS); 983 if (!req) { 984 r = -ENOMEM; 985 goto fail; 986 } 987 988 /* calculate max write size */ 989 r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen); 990 if (r) 991 goto fail; 992 993 if (opcode == CEPH_OSD_OP_CREATE || opcode == CEPH_OSD_OP_DELETE) { 994 osd_req_op_init(req, which, opcode, 0); 995 } else { 996 u32 object_size = layout->object_size; 997 u32 object_base = off - objoff; 998 if (!(truncate_seq == 1 && truncate_size == -1ULL)) { 999 if (truncate_size <= object_base) { 1000 truncate_size = 0; 1001 } else { 1002 truncate_size -= object_base; 1003 if (truncate_size > object_size) 1004 truncate_size = object_size; 1005 } 1006 } 1007 osd_req_op_extent_init(req, which, opcode, objoff, objlen, 1008 truncate_size, truncate_seq); 1009 } 1010 1011 req->r_abort_on_full = true; 1012 req->r_flags = flags; 1013 req->r_base_oloc.pool = layout->pool_id; 1014 req->r_base_oloc.pool_ns = ceph_try_get_string(layout->pool_ns); 1015 ceph_oid_printf(&req->r_base_oid, "%llx.%08llx", vino.ino, objnum); 1016 1017 req->r_snapid = vino.snap; 1018 if (flags & CEPH_OSD_FLAG_WRITE) 1019 req->r_data_offset = off; 1020 1021 r = ceph_osdc_alloc_messages(req, GFP_NOFS); 1022 if (r) 1023 goto fail; 1024 1025 return req; 1026 1027 fail: 1028 ceph_osdc_put_request(req); 1029 return ERR_PTR(r); 1030 } 1031 EXPORT_SYMBOL(ceph_osdc_new_request); 1032 1033 /* 1034 * We keep osd requests in an rbtree, sorted by ->r_tid. 1035 */ 1036 DEFINE_RB_FUNCS(request, struct ceph_osd_request, r_tid, r_node) 1037 DEFINE_RB_FUNCS(request_mc, struct ceph_osd_request, r_tid, r_mc_node) 1038 1039 static bool osd_homeless(struct ceph_osd *osd) 1040 { 1041 return osd->o_osd == CEPH_HOMELESS_OSD; 1042 } 1043 1044 static bool osd_registered(struct ceph_osd *osd) 1045 { 1046 verify_osdc_locked(osd->o_osdc); 1047 1048 return !RB_EMPTY_NODE(&osd->o_node); 1049 } 1050 1051 /* 1052 * Assumes @osd is zero-initialized. 1053 */ 1054 static void osd_init(struct ceph_osd *osd) 1055 { 1056 refcount_set(&osd->o_ref, 1); 1057 RB_CLEAR_NODE(&osd->o_node); 1058 osd->o_requests = RB_ROOT; 1059 osd->o_linger_requests = RB_ROOT; 1060 osd->o_backoff_mappings = RB_ROOT; 1061 osd->o_backoffs_by_id = RB_ROOT; 1062 INIT_LIST_HEAD(&osd->o_osd_lru); 1063 INIT_LIST_HEAD(&osd->o_keepalive_item); 1064 osd->o_incarnation = 1; 1065 mutex_init(&osd->lock); 1066 } 1067 1068 static void osd_cleanup(struct ceph_osd *osd) 1069 { 1070 WARN_ON(!RB_EMPTY_NODE(&osd->o_node)); 1071 WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests)); 1072 WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests)); 1073 WARN_ON(!RB_EMPTY_ROOT(&osd->o_backoff_mappings)); 1074 WARN_ON(!RB_EMPTY_ROOT(&osd->o_backoffs_by_id)); 1075 WARN_ON(!list_empty(&osd->o_osd_lru)); 1076 WARN_ON(!list_empty(&osd->o_keepalive_item)); 1077 1078 if (osd->o_auth.authorizer) { 1079 WARN_ON(osd_homeless(osd)); 1080 ceph_auth_destroy_authorizer(osd->o_auth.authorizer); 1081 } 1082 } 1083 1084 /* 1085 * Track open sessions with osds. 1086 */ 1087 static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum) 1088 { 1089 struct ceph_osd *osd; 1090 1091 WARN_ON(onum == CEPH_HOMELESS_OSD); 1092 1093 osd = kzalloc(sizeof(*osd), GFP_NOIO | __GFP_NOFAIL); 1094 osd_init(osd); 1095 osd->o_osdc = osdc; 1096 osd->o_osd = onum; 1097 1098 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr); 1099 1100 return osd; 1101 } 1102 1103 static struct ceph_osd *get_osd(struct ceph_osd *osd) 1104 { 1105 if (refcount_inc_not_zero(&osd->o_ref)) { 1106 dout("get_osd %p %d -> %d\n", osd, refcount_read(&osd->o_ref)-1, 1107 refcount_read(&osd->o_ref)); 1108 return osd; 1109 } else { 1110 dout("get_osd %p FAIL\n", osd); 1111 return NULL; 1112 } 1113 } 1114 1115 static void put_osd(struct ceph_osd *osd) 1116 { 1117 dout("put_osd %p %d -> %d\n", osd, refcount_read(&osd->o_ref), 1118 refcount_read(&osd->o_ref) - 1); 1119 if (refcount_dec_and_test(&osd->o_ref)) { 1120 osd_cleanup(osd); 1121 kfree(osd); 1122 } 1123 } 1124 1125 DEFINE_RB_FUNCS(osd, struct ceph_osd, o_osd, o_node) 1126 1127 static void __move_osd_to_lru(struct ceph_osd *osd) 1128 { 1129 struct ceph_osd_client *osdc = osd->o_osdc; 1130 1131 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd); 1132 BUG_ON(!list_empty(&osd->o_osd_lru)); 1133 1134 spin_lock(&osdc->osd_lru_lock); 1135 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru); 1136 spin_unlock(&osdc->osd_lru_lock); 1137 1138 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl; 1139 } 1140 1141 static void maybe_move_osd_to_lru(struct ceph_osd *osd) 1142 { 1143 if (RB_EMPTY_ROOT(&osd->o_requests) && 1144 RB_EMPTY_ROOT(&osd->o_linger_requests)) 1145 __move_osd_to_lru(osd); 1146 } 1147 1148 static void __remove_osd_from_lru(struct ceph_osd *osd) 1149 { 1150 struct ceph_osd_client *osdc = osd->o_osdc; 1151 1152 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd); 1153 1154 spin_lock(&osdc->osd_lru_lock); 1155 if (!list_empty(&osd->o_osd_lru)) 1156 list_del_init(&osd->o_osd_lru); 1157 spin_unlock(&osdc->osd_lru_lock); 1158 } 1159 1160 /* 1161 * Close the connection and assign any leftover requests to the 1162 * homeless session. 1163 */ 1164 static void close_osd(struct ceph_osd *osd) 1165 { 1166 struct ceph_osd_client *osdc = osd->o_osdc; 1167 struct rb_node *n; 1168 1169 verify_osdc_wrlocked(osdc); 1170 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd); 1171 1172 ceph_con_close(&osd->o_con); 1173 1174 for (n = rb_first(&osd->o_requests); n; ) { 1175 struct ceph_osd_request *req = 1176 rb_entry(n, struct ceph_osd_request, r_node); 1177 1178 n = rb_next(n); /* unlink_request() */ 1179 1180 dout(" reassigning req %p tid %llu\n", req, req->r_tid); 1181 unlink_request(osd, req); 1182 link_request(&osdc->homeless_osd, req); 1183 } 1184 for (n = rb_first(&osd->o_linger_requests); n; ) { 1185 struct ceph_osd_linger_request *lreq = 1186 rb_entry(n, struct ceph_osd_linger_request, node); 1187 1188 n = rb_next(n); /* unlink_linger() */ 1189 1190 dout(" reassigning lreq %p linger_id %llu\n", lreq, 1191 lreq->linger_id); 1192 unlink_linger(osd, lreq); 1193 link_linger(&osdc->homeless_osd, lreq); 1194 } 1195 clear_backoffs(osd); 1196 1197 __remove_osd_from_lru(osd); 1198 erase_osd(&osdc->osds, osd); 1199 put_osd(osd); 1200 } 1201 1202 /* 1203 * reset osd connect 1204 */ 1205 static int reopen_osd(struct ceph_osd *osd) 1206 { 1207 struct ceph_entity_addr *peer_addr; 1208 1209 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd); 1210 1211 if (RB_EMPTY_ROOT(&osd->o_requests) && 1212 RB_EMPTY_ROOT(&osd->o_linger_requests)) { 1213 close_osd(osd); 1214 return -ENODEV; 1215 } 1216 1217 peer_addr = &osd->o_osdc->osdmap->osd_addr[osd->o_osd]; 1218 if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) && 1219 !ceph_con_opened(&osd->o_con)) { 1220 struct rb_node *n; 1221 1222 dout("osd addr hasn't changed and connection never opened, " 1223 "letting msgr retry\n"); 1224 /* touch each r_stamp for handle_timeout()'s benfit */ 1225 for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) { 1226 struct ceph_osd_request *req = 1227 rb_entry(n, struct ceph_osd_request, r_node); 1228 req->r_stamp = jiffies; 1229 } 1230 1231 return -EAGAIN; 1232 } 1233 1234 ceph_con_close(&osd->o_con); 1235 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr); 1236 osd->o_incarnation++; 1237 1238 return 0; 1239 } 1240 1241 static struct ceph_osd *lookup_create_osd(struct ceph_osd_client *osdc, int o, 1242 bool wrlocked) 1243 { 1244 struct ceph_osd *osd; 1245 1246 if (wrlocked) 1247 verify_osdc_wrlocked(osdc); 1248 else 1249 verify_osdc_locked(osdc); 1250 1251 if (o != CEPH_HOMELESS_OSD) 1252 osd = lookup_osd(&osdc->osds, o); 1253 else 1254 osd = &osdc->homeless_osd; 1255 if (!osd) { 1256 if (!wrlocked) 1257 return ERR_PTR(-EAGAIN); 1258 1259 osd = create_osd(osdc, o); 1260 insert_osd(&osdc->osds, osd); 1261 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, 1262 &osdc->osdmap->osd_addr[osd->o_osd]); 1263 } 1264 1265 dout("%s osdc %p osd%d -> osd %p\n", __func__, osdc, o, osd); 1266 return osd; 1267 } 1268 1269 /* 1270 * Create request <-> OSD session relation. 1271 * 1272 * @req has to be assigned a tid, @osd may be homeless. 1273 */ 1274 static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req) 1275 { 1276 verify_osd_locked(osd); 1277 WARN_ON(!req->r_tid || req->r_osd); 1278 dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd, 1279 req, req->r_tid); 1280 1281 if (!osd_homeless(osd)) 1282 __remove_osd_from_lru(osd); 1283 else 1284 atomic_inc(&osd->o_osdc->num_homeless); 1285 1286 get_osd(osd); 1287 insert_request(&osd->o_requests, req); 1288 req->r_osd = osd; 1289 } 1290 1291 static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req) 1292 { 1293 verify_osd_locked(osd); 1294 WARN_ON(req->r_osd != osd); 1295 dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd, 1296 req, req->r_tid); 1297 1298 req->r_osd = NULL; 1299 erase_request(&osd->o_requests, req); 1300 put_osd(osd); 1301 1302 if (!osd_homeless(osd)) 1303 maybe_move_osd_to_lru(osd); 1304 else 1305 atomic_dec(&osd->o_osdc->num_homeless); 1306 } 1307 1308 static bool __pool_full(struct ceph_pg_pool_info *pi) 1309 { 1310 return pi->flags & CEPH_POOL_FLAG_FULL; 1311 } 1312 1313 static bool have_pool_full(struct ceph_osd_client *osdc) 1314 { 1315 struct rb_node *n; 1316 1317 for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) { 1318 struct ceph_pg_pool_info *pi = 1319 rb_entry(n, struct ceph_pg_pool_info, node); 1320 1321 if (__pool_full(pi)) 1322 return true; 1323 } 1324 1325 return false; 1326 } 1327 1328 static bool pool_full(struct ceph_osd_client *osdc, s64 pool_id) 1329 { 1330 struct ceph_pg_pool_info *pi; 1331 1332 pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id); 1333 if (!pi) 1334 return false; 1335 1336 return __pool_full(pi); 1337 } 1338 1339 /* 1340 * Returns whether a request should be blocked from being sent 1341 * based on the current osdmap and osd_client settings. 1342 */ 1343 static bool target_should_be_paused(struct ceph_osd_client *osdc, 1344 const struct ceph_osd_request_target *t, 1345 struct ceph_pg_pool_info *pi) 1346 { 1347 bool pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD); 1348 bool pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) || 1349 ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || 1350 __pool_full(pi); 1351 1352 WARN_ON(pi->id != t->target_oloc.pool); 1353 return ((t->flags & CEPH_OSD_FLAG_READ) && pauserd) || 1354 ((t->flags & CEPH_OSD_FLAG_WRITE) && pausewr) || 1355 (osdc->osdmap->epoch < osdc->epoch_barrier); 1356 } 1357 1358 enum calc_target_result { 1359 CALC_TARGET_NO_ACTION = 0, 1360 CALC_TARGET_NEED_RESEND, 1361 CALC_TARGET_POOL_DNE, 1362 }; 1363 1364 static enum calc_target_result calc_target(struct ceph_osd_client *osdc, 1365 struct ceph_osd_request_target *t, 1366 struct ceph_connection *con, 1367 bool any_change) 1368 { 1369 struct ceph_pg_pool_info *pi; 1370 struct ceph_pg pgid, last_pgid; 1371 struct ceph_osds up, acting; 1372 bool force_resend = false; 1373 bool unpaused = false; 1374 bool legacy_change; 1375 bool split = false; 1376 bool sort_bitwise = ceph_osdmap_flag(osdc, CEPH_OSDMAP_SORTBITWISE); 1377 bool recovery_deletes = ceph_osdmap_flag(osdc, 1378 CEPH_OSDMAP_RECOVERY_DELETES); 1379 enum calc_target_result ct_res; 1380 int ret; 1381 1382 t->epoch = osdc->osdmap->epoch; 1383 pi = ceph_pg_pool_by_id(osdc->osdmap, t->base_oloc.pool); 1384 if (!pi) { 1385 t->osd = CEPH_HOMELESS_OSD; 1386 ct_res = CALC_TARGET_POOL_DNE; 1387 goto out; 1388 } 1389 1390 if (osdc->osdmap->epoch == pi->last_force_request_resend) { 1391 if (t->last_force_resend < pi->last_force_request_resend) { 1392 t->last_force_resend = pi->last_force_request_resend; 1393 force_resend = true; 1394 } else if (t->last_force_resend == 0) { 1395 force_resend = true; 1396 } 1397 } 1398 1399 /* apply tiering */ 1400 ceph_oid_copy(&t->target_oid, &t->base_oid); 1401 ceph_oloc_copy(&t->target_oloc, &t->base_oloc); 1402 if ((t->flags & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0) { 1403 if (t->flags & CEPH_OSD_FLAG_READ && pi->read_tier >= 0) 1404 t->target_oloc.pool = pi->read_tier; 1405 if (t->flags & CEPH_OSD_FLAG_WRITE && pi->write_tier >= 0) 1406 t->target_oloc.pool = pi->write_tier; 1407 1408 pi = ceph_pg_pool_by_id(osdc->osdmap, t->target_oloc.pool); 1409 if (!pi) { 1410 t->osd = CEPH_HOMELESS_OSD; 1411 ct_res = CALC_TARGET_POOL_DNE; 1412 goto out; 1413 } 1414 } 1415 1416 ret = __ceph_object_locator_to_pg(pi, &t->target_oid, &t->target_oloc, 1417 &pgid); 1418 if (ret) { 1419 WARN_ON(ret != -ENOENT); 1420 t->osd = CEPH_HOMELESS_OSD; 1421 ct_res = CALC_TARGET_POOL_DNE; 1422 goto out; 1423 } 1424 last_pgid.pool = pgid.pool; 1425 last_pgid.seed = ceph_stable_mod(pgid.seed, t->pg_num, t->pg_num_mask); 1426 1427 ceph_pg_to_up_acting_osds(osdc->osdmap, pi, &pgid, &up, &acting); 1428 if (any_change && 1429 ceph_is_new_interval(&t->acting, 1430 &acting, 1431 &t->up, 1432 &up, 1433 t->size, 1434 pi->size, 1435 t->min_size, 1436 pi->min_size, 1437 t->pg_num, 1438 pi->pg_num, 1439 t->sort_bitwise, 1440 sort_bitwise, 1441 t->recovery_deletes, 1442 recovery_deletes, 1443 &last_pgid)) 1444 force_resend = true; 1445 1446 if (t->paused && !target_should_be_paused(osdc, t, pi)) { 1447 t->paused = false; 1448 unpaused = true; 1449 } 1450 legacy_change = ceph_pg_compare(&t->pgid, &pgid) || 1451 ceph_osds_changed(&t->acting, &acting, any_change); 1452 if (t->pg_num) 1453 split = ceph_pg_is_split(&last_pgid, t->pg_num, pi->pg_num); 1454 1455 if (legacy_change || force_resend || split) { 1456 t->pgid = pgid; /* struct */ 1457 ceph_pg_to_primary_shard(osdc->osdmap, pi, &pgid, &t->spgid); 1458 ceph_osds_copy(&t->acting, &acting); 1459 ceph_osds_copy(&t->up, &up); 1460 t->size = pi->size; 1461 t->min_size = pi->min_size; 1462 t->pg_num = pi->pg_num; 1463 t->pg_num_mask = pi->pg_num_mask; 1464 t->sort_bitwise = sort_bitwise; 1465 t->recovery_deletes = recovery_deletes; 1466 1467 t->osd = acting.primary; 1468 } 1469 1470 if (unpaused || legacy_change || force_resend || 1471 (split && con && CEPH_HAVE_FEATURE(con->peer_features, 1472 RESEND_ON_SPLIT))) 1473 ct_res = CALC_TARGET_NEED_RESEND; 1474 else 1475 ct_res = CALC_TARGET_NO_ACTION; 1476 1477 out: 1478 dout("%s t %p -> ct_res %d osd %d\n", __func__, t, ct_res, t->osd); 1479 return ct_res; 1480 } 1481 1482 static struct ceph_spg_mapping *alloc_spg_mapping(void) 1483 { 1484 struct ceph_spg_mapping *spg; 1485 1486 spg = kmalloc(sizeof(*spg), GFP_NOIO); 1487 if (!spg) 1488 return NULL; 1489 1490 RB_CLEAR_NODE(&spg->node); 1491 spg->backoffs = RB_ROOT; 1492 return spg; 1493 } 1494 1495 static void free_spg_mapping(struct ceph_spg_mapping *spg) 1496 { 1497 WARN_ON(!RB_EMPTY_NODE(&spg->node)); 1498 WARN_ON(!RB_EMPTY_ROOT(&spg->backoffs)); 1499 1500 kfree(spg); 1501 } 1502 1503 /* 1504 * rbtree of ceph_spg_mapping for handling map<spg_t, ...>, similar to 1505 * ceph_pg_mapping. Used to track OSD backoffs -- a backoff [range] is 1506 * defined only within a specific spgid; it does not pass anything to 1507 * children on split, or to another primary. 1508 */ 1509 DEFINE_RB_FUNCS2(spg_mapping, struct ceph_spg_mapping, spgid, ceph_spg_compare, 1510 RB_BYPTR, const struct ceph_spg *, node) 1511 1512 static u64 hoid_get_bitwise_key(const struct ceph_hobject_id *hoid) 1513 { 1514 return hoid->is_max ? 0x100000000ull : hoid->hash_reverse_bits; 1515 } 1516 1517 static void hoid_get_effective_key(const struct ceph_hobject_id *hoid, 1518 void **pkey, size_t *pkey_len) 1519 { 1520 if (hoid->key_len) { 1521 *pkey = hoid->key; 1522 *pkey_len = hoid->key_len; 1523 } else { 1524 *pkey = hoid->oid; 1525 *pkey_len = hoid->oid_len; 1526 } 1527 } 1528 1529 static int compare_names(const void *name1, size_t name1_len, 1530 const void *name2, size_t name2_len) 1531 { 1532 int ret; 1533 1534 ret = memcmp(name1, name2, min(name1_len, name2_len)); 1535 if (!ret) { 1536 if (name1_len < name2_len) 1537 ret = -1; 1538 else if (name1_len > name2_len) 1539 ret = 1; 1540 } 1541 return ret; 1542 } 1543 1544 static int hoid_compare(const struct ceph_hobject_id *lhs, 1545 const struct ceph_hobject_id *rhs) 1546 { 1547 void *effective_key1, *effective_key2; 1548 size_t effective_key1_len, effective_key2_len; 1549 int ret; 1550 1551 if (lhs->is_max < rhs->is_max) 1552 return -1; 1553 if (lhs->is_max > rhs->is_max) 1554 return 1; 1555 1556 if (lhs->pool < rhs->pool) 1557 return -1; 1558 if (lhs->pool > rhs->pool) 1559 return 1; 1560 1561 if (hoid_get_bitwise_key(lhs) < hoid_get_bitwise_key(rhs)) 1562 return -1; 1563 if (hoid_get_bitwise_key(lhs) > hoid_get_bitwise_key(rhs)) 1564 return 1; 1565 1566 ret = compare_names(lhs->nspace, lhs->nspace_len, 1567 rhs->nspace, rhs->nspace_len); 1568 if (ret) 1569 return ret; 1570 1571 hoid_get_effective_key(lhs, &effective_key1, &effective_key1_len); 1572 hoid_get_effective_key(rhs, &effective_key2, &effective_key2_len); 1573 ret = compare_names(effective_key1, effective_key1_len, 1574 effective_key2, effective_key2_len); 1575 if (ret) 1576 return ret; 1577 1578 ret = compare_names(lhs->oid, lhs->oid_len, rhs->oid, rhs->oid_len); 1579 if (ret) 1580 return ret; 1581 1582 if (lhs->snapid < rhs->snapid) 1583 return -1; 1584 if (lhs->snapid > rhs->snapid) 1585 return 1; 1586 1587 return 0; 1588 } 1589 1590 /* 1591 * For decoding ->begin and ->end of MOSDBackoff only -- no MIN/MAX 1592 * compat stuff here. 1593 * 1594 * Assumes @hoid is zero-initialized. 1595 */ 1596 static int decode_hoid(void **p, void *end, struct ceph_hobject_id *hoid) 1597 { 1598 u8 struct_v; 1599 u32 struct_len; 1600 int ret; 1601 1602 ret = ceph_start_decoding(p, end, 4, "hobject_t", &struct_v, 1603 &struct_len); 1604 if (ret) 1605 return ret; 1606 1607 if (struct_v < 4) { 1608 pr_err("got struct_v %d < 4 of hobject_t\n", struct_v); 1609 goto e_inval; 1610 } 1611 1612 hoid->key = ceph_extract_encoded_string(p, end, &hoid->key_len, 1613 GFP_NOIO); 1614 if (IS_ERR(hoid->key)) { 1615 ret = PTR_ERR(hoid->key); 1616 hoid->key = NULL; 1617 return ret; 1618 } 1619 1620 hoid->oid = ceph_extract_encoded_string(p, end, &hoid->oid_len, 1621 GFP_NOIO); 1622 if (IS_ERR(hoid->oid)) { 1623 ret = PTR_ERR(hoid->oid); 1624 hoid->oid = NULL; 1625 return ret; 1626 } 1627 1628 ceph_decode_64_safe(p, end, hoid->snapid, e_inval); 1629 ceph_decode_32_safe(p, end, hoid->hash, e_inval); 1630 ceph_decode_8_safe(p, end, hoid->is_max, e_inval); 1631 1632 hoid->nspace = ceph_extract_encoded_string(p, end, &hoid->nspace_len, 1633 GFP_NOIO); 1634 if (IS_ERR(hoid->nspace)) { 1635 ret = PTR_ERR(hoid->nspace); 1636 hoid->nspace = NULL; 1637 return ret; 1638 } 1639 1640 ceph_decode_64_safe(p, end, hoid->pool, e_inval); 1641 1642 ceph_hoid_build_hash_cache(hoid); 1643 return 0; 1644 1645 e_inval: 1646 return -EINVAL; 1647 } 1648 1649 static int hoid_encoding_size(const struct ceph_hobject_id *hoid) 1650 { 1651 return 8 + 4 + 1 + 8 + /* snapid, hash, is_max, pool */ 1652 4 + hoid->key_len + 4 + hoid->oid_len + 4 + hoid->nspace_len; 1653 } 1654 1655 static void encode_hoid(void **p, void *end, const struct ceph_hobject_id *hoid) 1656 { 1657 ceph_start_encoding(p, 4, 3, hoid_encoding_size(hoid)); 1658 ceph_encode_string(p, end, hoid->key, hoid->key_len); 1659 ceph_encode_string(p, end, hoid->oid, hoid->oid_len); 1660 ceph_encode_64(p, hoid->snapid); 1661 ceph_encode_32(p, hoid->hash); 1662 ceph_encode_8(p, hoid->is_max); 1663 ceph_encode_string(p, end, hoid->nspace, hoid->nspace_len); 1664 ceph_encode_64(p, hoid->pool); 1665 } 1666 1667 static void free_hoid(struct ceph_hobject_id *hoid) 1668 { 1669 if (hoid) { 1670 kfree(hoid->key); 1671 kfree(hoid->oid); 1672 kfree(hoid->nspace); 1673 kfree(hoid); 1674 } 1675 } 1676 1677 static struct ceph_osd_backoff *alloc_backoff(void) 1678 { 1679 struct ceph_osd_backoff *backoff; 1680 1681 backoff = kzalloc(sizeof(*backoff), GFP_NOIO); 1682 if (!backoff) 1683 return NULL; 1684 1685 RB_CLEAR_NODE(&backoff->spg_node); 1686 RB_CLEAR_NODE(&backoff->id_node); 1687 return backoff; 1688 } 1689 1690 static void free_backoff(struct ceph_osd_backoff *backoff) 1691 { 1692 WARN_ON(!RB_EMPTY_NODE(&backoff->spg_node)); 1693 WARN_ON(!RB_EMPTY_NODE(&backoff->id_node)); 1694 1695 free_hoid(backoff->begin); 1696 free_hoid(backoff->end); 1697 kfree(backoff); 1698 } 1699 1700 /* 1701 * Within a specific spgid, backoffs are managed by ->begin hoid. 1702 */ 1703 DEFINE_RB_INSDEL_FUNCS2(backoff, struct ceph_osd_backoff, begin, hoid_compare, 1704 RB_BYVAL, spg_node); 1705 1706 static struct ceph_osd_backoff *lookup_containing_backoff(struct rb_root *root, 1707 const struct ceph_hobject_id *hoid) 1708 { 1709 struct rb_node *n = root->rb_node; 1710 1711 while (n) { 1712 struct ceph_osd_backoff *cur = 1713 rb_entry(n, struct ceph_osd_backoff, spg_node); 1714 int cmp; 1715 1716 cmp = hoid_compare(hoid, cur->begin); 1717 if (cmp < 0) { 1718 n = n->rb_left; 1719 } else if (cmp > 0) { 1720 if (hoid_compare(hoid, cur->end) < 0) 1721 return cur; 1722 1723 n = n->rb_right; 1724 } else { 1725 return cur; 1726 } 1727 } 1728 1729 return NULL; 1730 } 1731 1732 /* 1733 * Each backoff has a unique id within its OSD session. 1734 */ 1735 DEFINE_RB_FUNCS(backoff_by_id, struct ceph_osd_backoff, id, id_node) 1736 1737 static void clear_backoffs(struct ceph_osd *osd) 1738 { 1739 while (!RB_EMPTY_ROOT(&osd->o_backoff_mappings)) { 1740 struct ceph_spg_mapping *spg = 1741 rb_entry(rb_first(&osd->o_backoff_mappings), 1742 struct ceph_spg_mapping, node); 1743 1744 while (!RB_EMPTY_ROOT(&spg->backoffs)) { 1745 struct ceph_osd_backoff *backoff = 1746 rb_entry(rb_first(&spg->backoffs), 1747 struct ceph_osd_backoff, spg_node); 1748 1749 erase_backoff(&spg->backoffs, backoff); 1750 erase_backoff_by_id(&osd->o_backoffs_by_id, backoff); 1751 free_backoff(backoff); 1752 } 1753 erase_spg_mapping(&osd->o_backoff_mappings, spg); 1754 free_spg_mapping(spg); 1755 } 1756 } 1757 1758 /* 1759 * Set up a temporary, non-owning view into @t. 1760 */ 1761 static void hoid_fill_from_target(struct ceph_hobject_id *hoid, 1762 const struct ceph_osd_request_target *t) 1763 { 1764 hoid->key = NULL; 1765 hoid->key_len = 0; 1766 hoid->oid = t->target_oid.name; 1767 hoid->oid_len = t->target_oid.name_len; 1768 hoid->snapid = CEPH_NOSNAP; 1769 hoid->hash = t->pgid.seed; 1770 hoid->is_max = false; 1771 if (t->target_oloc.pool_ns) { 1772 hoid->nspace = t->target_oloc.pool_ns->str; 1773 hoid->nspace_len = t->target_oloc.pool_ns->len; 1774 } else { 1775 hoid->nspace = NULL; 1776 hoid->nspace_len = 0; 1777 } 1778 hoid->pool = t->target_oloc.pool; 1779 ceph_hoid_build_hash_cache(hoid); 1780 } 1781 1782 static bool should_plug_request(struct ceph_osd_request *req) 1783 { 1784 struct ceph_osd *osd = req->r_osd; 1785 struct ceph_spg_mapping *spg; 1786 struct ceph_osd_backoff *backoff; 1787 struct ceph_hobject_id hoid; 1788 1789 spg = lookup_spg_mapping(&osd->o_backoff_mappings, &req->r_t.spgid); 1790 if (!spg) 1791 return false; 1792 1793 hoid_fill_from_target(&hoid, &req->r_t); 1794 backoff = lookup_containing_backoff(&spg->backoffs, &hoid); 1795 if (!backoff) 1796 return false; 1797 1798 dout("%s req %p tid %llu backoff osd%d spgid %llu.%xs%d id %llu\n", 1799 __func__, req, req->r_tid, osd->o_osd, backoff->spgid.pgid.pool, 1800 backoff->spgid.pgid.seed, backoff->spgid.shard, backoff->id); 1801 return true; 1802 } 1803 1804 static void setup_request_data(struct ceph_osd_request *req, 1805 struct ceph_msg *msg) 1806 { 1807 u32 data_len = 0; 1808 int i; 1809 1810 if (!list_empty(&msg->data)) 1811 return; 1812 1813 WARN_ON(msg->data_length); 1814 for (i = 0; i < req->r_num_ops; i++) { 1815 struct ceph_osd_req_op *op = &req->r_ops[i]; 1816 1817 switch (op->op) { 1818 /* request */ 1819 case CEPH_OSD_OP_WRITE: 1820 case CEPH_OSD_OP_WRITEFULL: 1821 WARN_ON(op->indata_len != op->extent.length); 1822 ceph_osdc_msg_data_add(msg, &op->extent.osd_data); 1823 break; 1824 case CEPH_OSD_OP_SETXATTR: 1825 case CEPH_OSD_OP_CMPXATTR: 1826 WARN_ON(op->indata_len != op->xattr.name_len + 1827 op->xattr.value_len); 1828 ceph_osdc_msg_data_add(msg, &op->xattr.osd_data); 1829 break; 1830 case CEPH_OSD_OP_NOTIFY_ACK: 1831 ceph_osdc_msg_data_add(msg, 1832 &op->notify_ack.request_data); 1833 break; 1834 1835 /* reply */ 1836 case CEPH_OSD_OP_STAT: 1837 ceph_osdc_msg_data_add(req->r_reply, 1838 &op->raw_data_in); 1839 break; 1840 case CEPH_OSD_OP_READ: 1841 ceph_osdc_msg_data_add(req->r_reply, 1842 &op->extent.osd_data); 1843 break; 1844 case CEPH_OSD_OP_LIST_WATCHERS: 1845 ceph_osdc_msg_data_add(req->r_reply, 1846 &op->list_watchers.response_data); 1847 break; 1848 1849 /* both */ 1850 case CEPH_OSD_OP_CALL: 1851 WARN_ON(op->indata_len != op->cls.class_len + 1852 op->cls.method_len + 1853 op->cls.indata_len); 1854 ceph_osdc_msg_data_add(msg, &op->cls.request_info); 1855 /* optional, can be NONE */ 1856 ceph_osdc_msg_data_add(msg, &op->cls.request_data); 1857 /* optional, can be NONE */ 1858 ceph_osdc_msg_data_add(req->r_reply, 1859 &op->cls.response_data); 1860 break; 1861 case CEPH_OSD_OP_NOTIFY: 1862 ceph_osdc_msg_data_add(msg, 1863 &op->notify.request_data); 1864 ceph_osdc_msg_data_add(req->r_reply, 1865 &op->notify.response_data); 1866 break; 1867 } 1868 1869 data_len += op->indata_len; 1870 } 1871 1872 WARN_ON(data_len != msg->data_length); 1873 } 1874 1875 static void encode_pgid(void **p, const struct ceph_pg *pgid) 1876 { 1877 ceph_encode_8(p, 1); 1878 ceph_encode_64(p, pgid->pool); 1879 ceph_encode_32(p, pgid->seed); 1880 ceph_encode_32(p, -1); /* preferred */ 1881 } 1882 1883 static void encode_spgid(void **p, const struct ceph_spg *spgid) 1884 { 1885 ceph_start_encoding(p, 1, 1, CEPH_PGID_ENCODING_LEN + 1); 1886 encode_pgid(p, &spgid->pgid); 1887 ceph_encode_8(p, spgid->shard); 1888 } 1889 1890 static void encode_oloc(void **p, void *end, 1891 const struct ceph_object_locator *oloc) 1892 { 1893 ceph_start_encoding(p, 5, 4, ceph_oloc_encoding_size(oloc)); 1894 ceph_encode_64(p, oloc->pool); 1895 ceph_encode_32(p, -1); /* preferred */ 1896 ceph_encode_32(p, 0); /* key len */ 1897 if (oloc->pool_ns) 1898 ceph_encode_string(p, end, oloc->pool_ns->str, 1899 oloc->pool_ns->len); 1900 else 1901 ceph_encode_32(p, 0); 1902 } 1903 1904 static void encode_request_partial(struct ceph_osd_request *req, 1905 struct ceph_msg *msg) 1906 { 1907 void *p = msg->front.iov_base; 1908 void *const end = p + msg->front_alloc_len; 1909 u32 data_len = 0; 1910 int i; 1911 1912 if (req->r_flags & CEPH_OSD_FLAG_WRITE) { 1913 /* snapshots aren't writeable */ 1914 WARN_ON(req->r_snapid != CEPH_NOSNAP); 1915 } else { 1916 WARN_ON(req->r_mtime.tv_sec || req->r_mtime.tv_nsec || 1917 req->r_data_offset || req->r_snapc); 1918 } 1919 1920 setup_request_data(req, msg); 1921 1922 encode_spgid(&p, &req->r_t.spgid); /* actual spg */ 1923 ceph_encode_32(&p, req->r_t.pgid.seed); /* raw hash */ 1924 ceph_encode_32(&p, req->r_osdc->osdmap->epoch); 1925 ceph_encode_32(&p, req->r_flags); 1926 1927 /* reqid */ 1928 ceph_start_encoding(&p, 2, 2, sizeof(struct ceph_osd_reqid)); 1929 memset(p, 0, sizeof(struct ceph_osd_reqid)); 1930 p += sizeof(struct ceph_osd_reqid); 1931 1932 /* trace */ 1933 memset(p, 0, sizeof(struct ceph_blkin_trace_info)); 1934 p += sizeof(struct ceph_blkin_trace_info); 1935 1936 ceph_encode_32(&p, 0); /* client_inc, always 0 */ 1937 ceph_encode_timespec(p, &req->r_mtime); 1938 p += sizeof(struct ceph_timespec); 1939 1940 encode_oloc(&p, end, &req->r_t.target_oloc); 1941 ceph_encode_string(&p, end, req->r_t.target_oid.name, 1942 req->r_t.target_oid.name_len); 1943 1944 /* ops, can imply data */ 1945 ceph_encode_16(&p, req->r_num_ops); 1946 for (i = 0; i < req->r_num_ops; i++) { 1947 data_len += osd_req_encode_op(p, &req->r_ops[i]); 1948 p += sizeof(struct ceph_osd_op); 1949 } 1950 1951 ceph_encode_64(&p, req->r_snapid); /* snapid */ 1952 if (req->r_snapc) { 1953 ceph_encode_64(&p, req->r_snapc->seq); 1954 ceph_encode_32(&p, req->r_snapc->num_snaps); 1955 for (i = 0; i < req->r_snapc->num_snaps; i++) 1956 ceph_encode_64(&p, req->r_snapc->snaps[i]); 1957 } else { 1958 ceph_encode_64(&p, 0); /* snap_seq */ 1959 ceph_encode_32(&p, 0); /* snaps len */ 1960 } 1961 1962 ceph_encode_32(&p, req->r_attempts); /* retry_attempt */ 1963 BUG_ON(p > end - 8); /* space for features */ 1964 1965 msg->hdr.version = cpu_to_le16(8); /* MOSDOp v8 */ 1966 /* front_len is finalized in encode_request_finish() */ 1967 msg->front.iov_len = p - msg->front.iov_base; 1968 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len); 1969 msg->hdr.data_len = cpu_to_le32(data_len); 1970 /* 1971 * The header "data_off" is a hint to the receiver allowing it 1972 * to align received data into its buffers such that there's no 1973 * need to re-copy it before writing it to disk (direct I/O). 1974 */ 1975 msg->hdr.data_off = cpu_to_le16(req->r_data_offset); 1976 1977 dout("%s req %p msg %p oid %s oid_len %d\n", __func__, req, msg, 1978 req->r_t.target_oid.name, req->r_t.target_oid.name_len); 1979 } 1980 1981 static void encode_request_finish(struct ceph_msg *msg) 1982 { 1983 void *p = msg->front.iov_base; 1984 void *const partial_end = p + msg->front.iov_len; 1985 void *const end = p + msg->front_alloc_len; 1986 1987 if (CEPH_HAVE_FEATURE(msg->con->peer_features, RESEND_ON_SPLIT)) { 1988 /* luminous OSD -- encode features and be done */ 1989 p = partial_end; 1990 ceph_encode_64(&p, msg->con->peer_features); 1991 } else { 1992 struct { 1993 char spgid[CEPH_ENCODING_START_BLK_LEN + 1994 CEPH_PGID_ENCODING_LEN + 1]; 1995 __le32 hash; 1996 __le32 epoch; 1997 __le32 flags; 1998 char reqid[CEPH_ENCODING_START_BLK_LEN + 1999 sizeof(struct ceph_osd_reqid)]; 2000 char trace[sizeof(struct ceph_blkin_trace_info)]; 2001 __le32 client_inc; 2002 struct ceph_timespec mtime; 2003 } __packed head; 2004 struct ceph_pg pgid; 2005 void *oloc, *oid, *tail; 2006 int oloc_len, oid_len, tail_len; 2007 int len; 2008 2009 /* 2010 * Pre-luminous OSD -- reencode v8 into v4 using @head 2011 * as a temporary buffer. Encode the raw PG; the rest 2012 * is just a matter of moving oloc, oid and tail blobs 2013 * around. 2014 */ 2015 memcpy(&head, p, sizeof(head)); 2016 p += sizeof(head); 2017 2018 oloc = p; 2019 p += CEPH_ENCODING_START_BLK_LEN; 2020 pgid.pool = ceph_decode_64(&p); 2021 p += 4 + 4; /* preferred, key len */ 2022 len = ceph_decode_32(&p); 2023 p += len; /* nspace */ 2024 oloc_len = p - oloc; 2025 2026 oid = p; 2027 len = ceph_decode_32(&p); 2028 p += len; 2029 oid_len = p - oid; 2030 2031 tail = p; 2032 tail_len = partial_end - p; 2033 2034 p = msg->front.iov_base; 2035 ceph_encode_copy(&p, &head.client_inc, sizeof(head.client_inc)); 2036 ceph_encode_copy(&p, &head.epoch, sizeof(head.epoch)); 2037 ceph_encode_copy(&p, &head.flags, sizeof(head.flags)); 2038 ceph_encode_copy(&p, &head.mtime, sizeof(head.mtime)); 2039 2040 /* reassert_version */ 2041 memset(p, 0, sizeof(struct ceph_eversion)); 2042 p += sizeof(struct ceph_eversion); 2043 2044 BUG_ON(p >= oloc); 2045 memmove(p, oloc, oloc_len); 2046 p += oloc_len; 2047 2048 pgid.seed = le32_to_cpu(head.hash); 2049 encode_pgid(&p, &pgid); /* raw pg */ 2050 2051 BUG_ON(p >= oid); 2052 memmove(p, oid, oid_len); 2053 p += oid_len; 2054 2055 /* tail -- ops, snapid, snapc, retry_attempt */ 2056 BUG_ON(p >= tail); 2057 memmove(p, tail, tail_len); 2058 p += tail_len; 2059 2060 msg->hdr.version = cpu_to_le16(4); /* MOSDOp v4 */ 2061 } 2062 2063 BUG_ON(p > end); 2064 msg->front.iov_len = p - msg->front.iov_base; 2065 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len); 2066 2067 dout("%s msg %p tid %llu %u+%u+%u v%d\n", __func__, msg, 2068 le64_to_cpu(msg->hdr.tid), le32_to_cpu(msg->hdr.front_len), 2069 le32_to_cpu(msg->hdr.middle_len), le32_to_cpu(msg->hdr.data_len), 2070 le16_to_cpu(msg->hdr.version)); 2071 } 2072 2073 /* 2074 * @req has to be assigned a tid and registered. 2075 */ 2076 static void send_request(struct ceph_osd_request *req) 2077 { 2078 struct ceph_osd *osd = req->r_osd; 2079 2080 verify_osd_locked(osd); 2081 WARN_ON(osd->o_osd != req->r_t.osd); 2082 2083 /* backoff? */ 2084 if (should_plug_request(req)) 2085 return; 2086 2087 /* 2088 * We may have a previously queued request message hanging 2089 * around. Cancel it to avoid corrupting the msgr. 2090 */ 2091 if (req->r_sent) 2092 ceph_msg_revoke(req->r_request); 2093 2094 req->r_flags |= CEPH_OSD_FLAG_KNOWN_REDIR; 2095 if (req->r_attempts) 2096 req->r_flags |= CEPH_OSD_FLAG_RETRY; 2097 else 2098 WARN_ON(req->r_flags & CEPH_OSD_FLAG_RETRY); 2099 2100 encode_request_partial(req, req->r_request); 2101 2102 dout("%s req %p tid %llu to pgid %llu.%x spgid %llu.%xs%d osd%d e%u flags 0x%x attempt %d\n", 2103 __func__, req, req->r_tid, req->r_t.pgid.pool, req->r_t.pgid.seed, 2104 req->r_t.spgid.pgid.pool, req->r_t.spgid.pgid.seed, 2105 req->r_t.spgid.shard, osd->o_osd, req->r_t.epoch, req->r_flags, 2106 req->r_attempts); 2107 2108 req->r_t.paused = false; 2109 req->r_stamp = jiffies; 2110 req->r_attempts++; 2111 2112 req->r_sent = osd->o_incarnation; 2113 req->r_request->hdr.tid = cpu_to_le64(req->r_tid); 2114 ceph_con_send(&osd->o_con, ceph_msg_get(req->r_request)); 2115 } 2116 2117 static void maybe_request_map(struct ceph_osd_client *osdc) 2118 { 2119 bool continuous = false; 2120 2121 verify_osdc_locked(osdc); 2122 WARN_ON(!osdc->osdmap->epoch); 2123 2124 if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || 2125 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD) || 2126 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) { 2127 dout("%s osdc %p continuous\n", __func__, osdc); 2128 continuous = true; 2129 } else { 2130 dout("%s osdc %p onetime\n", __func__, osdc); 2131 } 2132 2133 if (ceph_monc_want_map(&osdc->client->monc, CEPH_SUB_OSDMAP, 2134 osdc->osdmap->epoch + 1, continuous)) 2135 ceph_monc_renew_subs(&osdc->client->monc); 2136 } 2137 2138 static void complete_request(struct ceph_osd_request *req, int err); 2139 static void send_map_check(struct ceph_osd_request *req); 2140 2141 static void __submit_request(struct ceph_osd_request *req, bool wrlocked) 2142 { 2143 struct ceph_osd_client *osdc = req->r_osdc; 2144 struct ceph_osd *osd; 2145 enum calc_target_result ct_res; 2146 bool need_send = false; 2147 bool promoted = false; 2148 bool need_abort = false; 2149 2150 WARN_ON(req->r_tid); 2151 dout("%s req %p wrlocked %d\n", __func__, req, wrlocked); 2152 2153 again: 2154 ct_res = calc_target(osdc, &req->r_t, NULL, false); 2155 if (ct_res == CALC_TARGET_POOL_DNE && !wrlocked) 2156 goto promote; 2157 2158 osd = lookup_create_osd(osdc, req->r_t.osd, wrlocked); 2159 if (IS_ERR(osd)) { 2160 WARN_ON(PTR_ERR(osd) != -EAGAIN || wrlocked); 2161 goto promote; 2162 } 2163 2164 if (osdc->osdmap->epoch < osdc->epoch_barrier) { 2165 dout("req %p epoch %u barrier %u\n", req, osdc->osdmap->epoch, 2166 osdc->epoch_barrier); 2167 req->r_t.paused = true; 2168 maybe_request_map(osdc); 2169 } else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) && 2170 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) { 2171 dout("req %p pausewr\n", req); 2172 req->r_t.paused = true; 2173 maybe_request_map(osdc); 2174 } else if ((req->r_flags & CEPH_OSD_FLAG_READ) && 2175 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) { 2176 dout("req %p pauserd\n", req); 2177 req->r_t.paused = true; 2178 maybe_request_map(osdc); 2179 } else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) && 2180 !(req->r_flags & (CEPH_OSD_FLAG_FULL_TRY | 2181 CEPH_OSD_FLAG_FULL_FORCE)) && 2182 (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || 2183 pool_full(osdc, req->r_t.base_oloc.pool))) { 2184 dout("req %p full/pool_full\n", req); 2185 pr_warn_ratelimited("FULL or reached pool quota\n"); 2186 req->r_t.paused = true; 2187 maybe_request_map(osdc); 2188 if (req->r_abort_on_full) 2189 need_abort = true; 2190 } else if (!osd_homeless(osd)) { 2191 need_send = true; 2192 } else { 2193 maybe_request_map(osdc); 2194 } 2195 2196 mutex_lock(&osd->lock); 2197 /* 2198 * Assign the tid atomically with send_request() to protect 2199 * multiple writes to the same object from racing with each 2200 * other, resulting in out of order ops on the OSDs. 2201 */ 2202 req->r_tid = atomic64_inc_return(&osdc->last_tid); 2203 link_request(osd, req); 2204 if (need_send) 2205 send_request(req); 2206 else if (need_abort) 2207 complete_request(req, -ENOSPC); 2208 mutex_unlock(&osd->lock); 2209 2210 if (ct_res == CALC_TARGET_POOL_DNE) 2211 send_map_check(req); 2212 2213 if (promoted) 2214 downgrade_write(&osdc->lock); 2215 return; 2216 2217 promote: 2218 up_read(&osdc->lock); 2219 down_write(&osdc->lock); 2220 wrlocked = true; 2221 promoted = true; 2222 goto again; 2223 } 2224 2225 static void account_request(struct ceph_osd_request *req) 2226 { 2227 WARN_ON(req->r_flags & (CEPH_OSD_FLAG_ACK | CEPH_OSD_FLAG_ONDISK)); 2228 WARN_ON(!(req->r_flags & (CEPH_OSD_FLAG_READ | CEPH_OSD_FLAG_WRITE))); 2229 2230 req->r_flags |= CEPH_OSD_FLAG_ONDISK; 2231 atomic_inc(&req->r_osdc->num_requests); 2232 2233 req->r_start_stamp = jiffies; 2234 } 2235 2236 static void submit_request(struct ceph_osd_request *req, bool wrlocked) 2237 { 2238 ceph_osdc_get_request(req); 2239 account_request(req); 2240 __submit_request(req, wrlocked); 2241 } 2242 2243 static void finish_request(struct ceph_osd_request *req) 2244 { 2245 struct ceph_osd_client *osdc = req->r_osdc; 2246 2247 WARN_ON(lookup_request_mc(&osdc->map_checks, req->r_tid)); 2248 dout("%s req %p tid %llu\n", __func__, req, req->r_tid); 2249 2250 if (req->r_osd) 2251 unlink_request(req->r_osd, req); 2252 atomic_dec(&osdc->num_requests); 2253 2254 /* 2255 * If an OSD has failed or returned and a request has been sent 2256 * twice, it's possible to get a reply and end up here while the 2257 * request message is queued for delivery. We will ignore the 2258 * reply, so not a big deal, but better to try and catch it. 2259 */ 2260 ceph_msg_revoke(req->r_request); 2261 ceph_msg_revoke_incoming(req->r_reply); 2262 } 2263 2264 static void __complete_request(struct ceph_osd_request *req) 2265 { 2266 if (req->r_callback) { 2267 dout("%s req %p tid %llu cb %pf result %d\n", __func__, req, 2268 req->r_tid, req->r_callback, req->r_result); 2269 req->r_callback(req); 2270 } 2271 } 2272 2273 /* 2274 * This is open-coded in handle_reply(). 2275 */ 2276 static void complete_request(struct ceph_osd_request *req, int err) 2277 { 2278 dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err); 2279 2280 req->r_result = err; 2281 finish_request(req); 2282 __complete_request(req); 2283 complete_all(&req->r_completion); 2284 ceph_osdc_put_request(req); 2285 } 2286 2287 static void cancel_map_check(struct ceph_osd_request *req) 2288 { 2289 struct ceph_osd_client *osdc = req->r_osdc; 2290 struct ceph_osd_request *lookup_req; 2291 2292 verify_osdc_wrlocked(osdc); 2293 2294 lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid); 2295 if (!lookup_req) 2296 return; 2297 2298 WARN_ON(lookup_req != req); 2299 erase_request_mc(&osdc->map_checks, req); 2300 ceph_osdc_put_request(req); 2301 } 2302 2303 static void cancel_request(struct ceph_osd_request *req) 2304 { 2305 dout("%s req %p tid %llu\n", __func__, req, req->r_tid); 2306 2307 cancel_map_check(req); 2308 finish_request(req); 2309 complete_all(&req->r_completion); 2310 ceph_osdc_put_request(req); 2311 } 2312 2313 static void abort_request(struct ceph_osd_request *req, int err) 2314 { 2315 dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err); 2316 2317 cancel_map_check(req); 2318 complete_request(req, err); 2319 } 2320 2321 static void update_epoch_barrier(struct ceph_osd_client *osdc, u32 eb) 2322 { 2323 if (likely(eb > osdc->epoch_barrier)) { 2324 dout("updating epoch_barrier from %u to %u\n", 2325 osdc->epoch_barrier, eb); 2326 osdc->epoch_barrier = eb; 2327 /* Request map if we're not to the barrier yet */ 2328 if (eb > osdc->osdmap->epoch) 2329 maybe_request_map(osdc); 2330 } 2331 } 2332 2333 void ceph_osdc_update_epoch_barrier(struct ceph_osd_client *osdc, u32 eb) 2334 { 2335 down_read(&osdc->lock); 2336 if (unlikely(eb > osdc->epoch_barrier)) { 2337 up_read(&osdc->lock); 2338 down_write(&osdc->lock); 2339 update_epoch_barrier(osdc, eb); 2340 up_write(&osdc->lock); 2341 } else { 2342 up_read(&osdc->lock); 2343 } 2344 } 2345 EXPORT_SYMBOL(ceph_osdc_update_epoch_barrier); 2346 2347 /* 2348 * Drop all pending requests that are stalled waiting on a full condition to 2349 * clear, and complete them with ENOSPC as the return code. Set the 2350 * osdc->epoch_barrier to the latest map epoch that we've seen if any were 2351 * cancelled. 2352 */ 2353 static void ceph_osdc_abort_on_full(struct ceph_osd_client *osdc) 2354 { 2355 struct rb_node *n; 2356 bool victims = false; 2357 2358 dout("enter abort_on_full\n"); 2359 2360 if (!ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) && !have_pool_full(osdc)) 2361 goto out; 2362 2363 /* Scan list and see if there is anything to abort */ 2364 for (n = rb_first(&osdc->osds); n; n = rb_next(n)) { 2365 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node); 2366 struct rb_node *m; 2367 2368 m = rb_first(&osd->o_requests); 2369 while (m) { 2370 struct ceph_osd_request *req = rb_entry(m, 2371 struct ceph_osd_request, r_node); 2372 m = rb_next(m); 2373 2374 if (req->r_abort_on_full) { 2375 victims = true; 2376 break; 2377 } 2378 } 2379 if (victims) 2380 break; 2381 } 2382 2383 if (!victims) 2384 goto out; 2385 2386 /* 2387 * Update the barrier to current epoch if it's behind that point, 2388 * since we know we have some calls to be aborted in the tree. 2389 */ 2390 update_epoch_barrier(osdc, osdc->osdmap->epoch); 2391 2392 for (n = rb_first(&osdc->osds); n; n = rb_next(n)) { 2393 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node); 2394 struct rb_node *m; 2395 2396 m = rb_first(&osd->o_requests); 2397 while (m) { 2398 struct ceph_osd_request *req = rb_entry(m, 2399 struct ceph_osd_request, r_node); 2400 m = rb_next(m); 2401 2402 if (req->r_abort_on_full && 2403 (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || 2404 pool_full(osdc, req->r_t.target_oloc.pool))) 2405 abort_request(req, -ENOSPC); 2406 } 2407 } 2408 out: 2409 dout("return abort_on_full barrier=%u\n", osdc->epoch_barrier); 2410 } 2411 2412 static void check_pool_dne(struct ceph_osd_request *req) 2413 { 2414 struct ceph_osd_client *osdc = req->r_osdc; 2415 struct ceph_osdmap *map = osdc->osdmap; 2416 2417 verify_osdc_wrlocked(osdc); 2418 WARN_ON(!map->epoch); 2419 2420 if (req->r_attempts) { 2421 /* 2422 * We sent a request earlier, which means that 2423 * previously the pool existed, and now it does not 2424 * (i.e., it was deleted). 2425 */ 2426 req->r_map_dne_bound = map->epoch; 2427 dout("%s req %p tid %llu pool disappeared\n", __func__, req, 2428 req->r_tid); 2429 } else { 2430 dout("%s req %p tid %llu map_dne_bound %u have %u\n", __func__, 2431 req, req->r_tid, req->r_map_dne_bound, map->epoch); 2432 } 2433 2434 if (req->r_map_dne_bound) { 2435 if (map->epoch >= req->r_map_dne_bound) { 2436 /* we had a new enough map */ 2437 pr_info_ratelimited("tid %llu pool does not exist\n", 2438 req->r_tid); 2439 complete_request(req, -ENOENT); 2440 } 2441 } else { 2442 send_map_check(req); 2443 } 2444 } 2445 2446 static void map_check_cb(struct ceph_mon_generic_request *greq) 2447 { 2448 struct ceph_osd_client *osdc = &greq->monc->client->osdc; 2449 struct ceph_osd_request *req; 2450 u64 tid = greq->private_data; 2451 2452 WARN_ON(greq->result || !greq->u.newest); 2453 2454 down_write(&osdc->lock); 2455 req = lookup_request_mc(&osdc->map_checks, tid); 2456 if (!req) { 2457 dout("%s tid %llu dne\n", __func__, tid); 2458 goto out_unlock; 2459 } 2460 2461 dout("%s req %p tid %llu map_dne_bound %u newest %llu\n", __func__, 2462 req, req->r_tid, req->r_map_dne_bound, greq->u.newest); 2463 if (!req->r_map_dne_bound) 2464 req->r_map_dne_bound = greq->u.newest; 2465 erase_request_mc(&osdc->map_checks, req); 2466 check_pool_dne(req); 2467 2468 ceph_osdc_put_request(req); 2469 out_unlock: 2470 up_write(&osdc->lock); 2471 } 2472 2473 static void send_map_check(struct ceph_osd_request *req) 2474 { 2475 struct ceph_osd_client *osdc = req->r_osdc; 2476 struct ceph_osd_request *lookup_req; 2477 int ret; 2478 2479 verify_osdc_wrlocked(osdc); 2480 2481 lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid); 2482 if (lookup_req) { 2483 WARN_ON(lookup_req != req); 2484 return; 2485 } 2486 2487 ceph_osdc_get_request(req); 2488 insert_request_mc(&osdc->map_checks, req); 2489 ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap", 2490 map_check_cb, req->r_tid); 2491 WARN_ON(ret); 2492 } 2493 2494 /* 2495 * lingering requests, watch/notify v2 infrastructure 2496 */ 2497 static void linger_release(struct kref *kref) 2498 { 2499 struct ceph_osd_linger_request *lreq = 2500 container_of(kref, struct ceph_osd_linger_request, kref); 2501 2502 dout("%s lreq %p reg_req %p ping_req %p\n", __func__, lreq, 2503 lreq->reg_req, lreq->ping_req); 2504 WARN_ON(!RB_EMPTY_NODE(&lreq->node)); 2505 WARN_ON(!RB_EMPTY_NODE(&lreq->osdc_node)); 2506 WARN_ON(!RB_EMPTY_NODE(&lreq->mc_node)); 2507 WARN_ON(!list_empty(&lreq->scan_item)); 2508 WARN_ON(!list_empty(&lreq->pending_lworks)); 2509 WARN_ON(lreq->osd); 2510 2511 if (lreq->reg_req) 2512 ceph_osdc_put_request(lreq->reg_req); 2513 if (lreq->ping_req) 2514 ceph_osdc_put_request(lreq->ping_req); 2515 target_destroy(&lreq->t); 2516 kfree(lreq); 2517 } 2518 2519 static void linger_put(struct ceph_osd_linger_request *lreq) 2520 { 2521 if (lreq) 2522 kref_put(&lreq->kref, linger_release); 2523 } 2524 2525 static struct ceph_osd_linger_request * 2526 linger_get(struct ceph_osd_linger_request *lreq) 2527 { 2528 kref_get(&lreq->kref); 2529 return lreq; 2530 } 2531 2532 static struct ceph_osd_linger_request * 2533 linger_alloc(struct ceph_osd_client *osdc) 2534 { 2535 struct ceph_osd_linger_request *lreq; 2536 2537 lreq = kzalloc(sizeof(*lreq), GFP_NOIO); 2538 if (!lreq) 2539 return NULL; 2540 2541 kref_init(&lreq->kref); 2542 mutex_init(&lreq->lock); 2543 RB_CLEAR_NODE(&lreq->node); 2544 RB_CLEAR_NODE(&lreq->osdc_node); 2545 RB_CLEAR_NODE(&lreq->mc_node); 2546 INIT_LIST_HEAD(&lreq->scan_item); 2547 INIT_LIST_HEAD(&lreq->pending_lworks); 2548 init_completion(&lreq->reg_commit_wait); 2549 init_completion(&lreq->notify_finish_wait); 2550 2551 lreq->osdc = osdc; 2552 target_init(&lreq->t); 2553 2554 dout("%s lreq %p\n", __func__, lreq); 2555 return lreq; 2556 } 2557 2558 DEFINE_RB_INSDEL_FUNCS(linger, struct ceph_osd_linger_request, linger_id, node) 2559 DEFINE_RB_FUNCS(linger_osdc, struct ceph_osd_linger_request, linger_id, osdc_node) 2560 DEFINE_RB_FUNCS(linger_mc, struct ceph_osd_linger_request, linger_id, mc_node) 2561 2562 /* 2563 * Create linger request <-> OSD session relation. 2564 * 2565 * @lreq has to be registered, @osd may be homeless. 2566 */ 2567 static void link_linger(struct ceph_osd *osd, 2568 struct ceph_osd_linger_request *lreq) 2569 { 2570 verify_osd_locked(osd); 2571 WARN_ON(!lreq->linger_id || lreq->osd); 2572 dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd, 2573 osd->o_osd, lreq, lreq->linger_id); 2574 2575 if (!osd_homeless(osd)) 2576 __remove_osd_from_lru(osd); 2577 else 2578 atomic_inc(&osd->o_osdc->num_homeless); 2579 2580 get_osd(osd); 2581 insert_linger(&osd->o_linger_requests, lreq); 2582 lreq->osd = osd; 2583 } 2584 2585 static void unlink_linger(struct ceph_osd *osd, 2586 struct ceph_osd_linger_request *lreq) 2587 { 2588 verify_osd_locked(osd); 2589 WARN_ON(lreq->osd != osd); 2590 dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd, 2591 osd->o_osd, lreq, lreq->linger_id); 2592 2593 lreq->osd = NULL; 2594 erase_linger(&osd->o_linger_requests, lreq); 2595 put_osd(osd); 2596 2597 if (!osd_homeless(osd)) 2598 maybe_move_osd_to_lru(osd); 2599 else 2600 atomic_dec(&osd->o_osdc->num_homeless); 2601 } 2602 2603 static bool __linger_registered(struct ceph_osd_linger_request *lreq) 2604 { 2605 verify_osdc_locked(lreq->osdc); 2606 2607 return !RB_EMPTY_NODE(&lreq->osdc_node); 2608 } 2609 2610 static bool linger_registered(struct ceph_osd_linger_request *lreq) 2611 { 2612 struct ceph_osd_client *osdc = lreq->osdc; 2613 bool registered; 2614 2615 down_read(&osdc->lock); 2616 registered = __linger_registered(lreq); 2617 up_read(&osdc->lock); 2618 2619 return registered; 2620 } 2621 2622 static void linger_register(struct ceph_osd_linger_request *lreq) 2623 { 2624 struct ceph_osd_client *osdc = lreq->osdc; 2625 2626 verify_osdc_wrlocked(osdc); 2627 WARN_ON(lreq->linger_id); 2628 2629 linger_get(lreq); 2630 lreq->linger_id = ++osdc->last_linger_id; 2631 insert_linger_osdc(&osdc->linger_requests, lreq); 2632 } 2633 2634 static void linger_unregister(struct ceph_osd_linger_request *lreq) 2635 { 2636 struct ceph_osd_client *osdc = lreq->osdc; 2637 2638 verify_osdc_wrlocked(osdc); 2639 2640 erase_linger_osdc(&osdc->linger_requests, lreq); 2641 linger_put(lreq); 2642 } 2643 2644 static void cancel_linger_request(struct ceph_osd_request *req) 2645 { 2646 struct ceph_osd_linger_request *lreq = req->r_priv; 2647 2648 WARN_ON(!req->r_linger); 2649 cancel_request(req); 2650 linger_put(lreq); 2651 } 2652 2653 struct linger_work { 2654 struct work_struct work; 2655 struct ceph_osd_linger_request *lreq; 2656 struct list_head pending_item; 2657 unsigned long queued_stamp; 2658 2659 union { 2660 struct { 2661 u64 notify_id; 2662 u64 notifier_id; 2663 void *payload; /* points into @msg front */ 2664 size_t payload_len; 2665 2666 struct ceph_msg *msg; /* for ceph_msg_put() */ 2667 } notify; 2668 struct { 2669 int err; 2670 } error; 2671 }; 2672 }; 2673 2674 static struct linger_work *lwork_alloc(struct ceph_osd_linger_request *lreq, 2675 work_func_t workfn) 2676 { 2677 struct linger_work *lwork; 2678 2679 lwork = kzalloc(sizeof(*lwork), GFP_NOIO); 2680 if (!lwork) 2681 return NULL; 2682 2683 INIT_WORK(&lwork->work, workfn); 2684 INIT_LIST_HEAD(&lwork->pending_item); 2685 lwork->lreq = linger_get(lreq); 2686 2687 return lwork; 2688 } 2689 2690 static void lwork_free(struct linger_work *lwork) 2691 { 2692 struct ceph_osd_linger_request *lreq = lwork->lreq; 2693 2694 mutex_lock(&lreq->lock); 2695 list_del(&lwork->pending_item); 2696 mutex_unlock(&lreq->lock); 2697 2698 linger_put(lreq); 2699 kfree(lwork); 2700 } 2701 2702 static void lwork_queue(struct linger_work *lwork) 2703 { 2704 struct ceph_osd_linger_request *lreq = lwork->lreq; 2705 struct ceph_osd_client *osdc = lreq->osdc; 2706 2707 verify_lreq_locked(lreq); 2708 WARN_ON(!list_empty(&lwork->pending_item)); 2709 2710 lwork->queued_stamp = jiffies; 2711 list_add_tail(&lwork->pending_item, &lreq->pending_lworks); 2712 queue_work(osdc->notify_wq, &lwork->work); 2713 } 2714 2715 static void do_watch_notify(struct work_struct *w) 2716 { 2717 struct linger_work *lwork = container_of(w, struct linger_work, work); 2718 struct ceph_osd_linger_request *lreq = lwork->lreq; 2719 2720 if (!linger_registered(lreq)) { 2721 dout("%s lreq %p not registered\n", __func__, lreq); 2722 goto out; 2723 } 2724 2725 WARN_ON(!lreq->is_watch); 2726 dout("%s lreq %p notify_id %llu notifier_id %llu payload_len %zu\n", 2727 __func__, lreq, lwork->notify.notify_id, lwork->notify.notifier_id, 2728 lwork->notify.payload_len); 2729 lreq->wcb(lreq->data, lwork->notify.notify_id, lreq->linger_id, 2730 lwork->notify.notifier_id, lwork->notify.payload, 2731 lwork->notify.payload_len); 2732 2733 out: 2734 ceph_msg_put(lwork->notify.msg); 2735 lwork_free(lwork); 2736 } 2737 2738 static void do_watch_error(struct work_struct *w) 2739 { 2740 struct linger_work *lwork = container_of(w, struct linger_work, work); 2741 struct ceph_osd_linger_request *lreq = lwork->lreq; 2742 2743 if (!linger_registered(lreq)) { 2744 dout("%s lreq %p not registered\n", __func__, lreq); 2745 goto out; 2746 } 2747 2748 dout("%s lreq %p err %d\n", __func__, lreq, lwork->error.err); 2749 lreq->errcb(lreq->data, lreq->linger_id, lwork->error.err); 2750 2751 out: 2752 lwork_free(lwork); 2753 } 2754 2755 static void queue_watch_error(struct ceph_osd_linger_request *lreq) 2756 { 2757 struct linger_work *lwork; 2758 2759 lwork = lwork_alloc(lreq, do_watch_error); 2760 if (!lwork) { 2761 pr_err("failed to allocate error-lwork\n"); 2762 return; 2763 } 2764 2765 lwork->error.err = lreq->last_error; 2766 lwork_queue(lwork); 2767 } 2768 2769 static void linger_reg_commit_complete(struct ceph_osd_linger_request *lreq, 2770 int result) 2771 { 2772 if (!completion_done(&lreq->reg_commit_wait)) { 2773 lreq->reg_commit_error = (result <= 0 ? result : 0); 2774 complete_all(&lreq->reg_commit_wait); 2775 } 2776 } 2777 2778 static void linger_commit_cb(struct ceph_osd_request *req) 2779 { 2780 struct ceph_osd_linger_request *lreq = req->r_priv; 2781 2782 mutex_lock(&lreq->lock); 2783 dout("%s lreq %p linger_id %llu result %d\n", __func__, lreq, 2784 lreq->linger_id, req->r_result); 2785 linger_reg_commit_complete(lreq, req->r_result); 2786 lreq->committed = true; 2787 2788 if (!lreq->is_watch) { 2789 struct ceph_osd_data *osd_data = 2790 osd_req_op_data(req, 0, notify, response_data); 2791 void *p = page_address(osd_data->pages[0]); 2792 2793 WARN_ON(req->r_ops[0].op != CEPH_OSD_OP_NOTIFY || 2794 osd_data->type != CEPH_OSD_DATA_TYPE_PAGES); 2795 2796 /* make note of the notify_id */ 2797 if (req->r_ops[0].outdata_len >= sizeof(u64)) { 2798 lreq->notify_id = ceph_decode_64(&p); 2799 dout("lreq %p notify_id %llu\n", lreq, 2800 lreq->notify_id); 2801 } else { 2802 dout("lreq %p no notify_id\n", lreq); 2803 } 2804 } 2805 2806 mutex_unlock(&lreq->lock); 2807 linger_put(lreq); 2808 } 2809 2810 static int normalize_watch_error(int err) 2811 { 2812 /* 2813 * Translate ENOENT -> ENOTCONN so that a delete->disconnection 2814 * notification and a failure to reconnect because we raced with 2815 * the delete appear the same to the user. 2816 */ 2817 if (err == -ENOENT) 2818 err = -ENOTCONN; 2819 2820 return err; 2821 } 2822 2823 static void linger_reconnect_cb(struct ceph_osd_request *req) 2824 { 2825 struct ceph_osd_linger_request *lreq = req->r_priv; 2826 2827 mutex_lock(&lreq->lock); 2828 dout("%s lreq %p linger_id %llu result %d last_error %d\n", __func__, 2829 lreq, lreq->linger_id, req->r_result, lreq->last_error); 2830 if (req->r_result < 0) { 2831 if (!lreq->last_error) { 2832 lreq->last_error = normalize_watch_error(req->r_result); 2833 queue_watch_error(lreq); 2834 } 2835 } 2836 2837 mutex_unlock(&lreq->lock); 2838 linger_put(lreq); 2839 } 2840 2841 static void send_linger(struct ceph_osd_linger_request *lreq) 2842 { 2843 struct ceph_osd_request *req = lreq->reg_req; 2844 struct ceph_osd_req_op *op = &req->r_ops[0]; 2845 2846 verify_osdc_wrlocked(req->r_osdc); 2847 dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id); 2848 2849 if (req->r_osd) 2850 cancel_linger_request(req); 2851 2852 request_reinit(req); 2853 ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid); 2854 ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc); 2855 req->r_flags = lreq->t.flags; 2856 req->r_mtime = lreq->mtime; 2857 2858 mutex_lock(&lreq->lock); 2859 if (lreq->is_watch && lreq->committed) { 2860 WARN_ON(op->op != CEPH_OSD_OP_WATCH || 2861 op->watch.cookie != lreq->linger_id); 2862 op->watch.op = CEPH_OSD_WATCH_OP_RECONNECT; 2863 op->watch.gen = ++lreq->register_gen; 2864 dout("lreq %p reconnect register_gen %u\n", lreq, 2865 op->watch.gen); 2866 req->r_callback = linger_reconnect_cb; 2867 } else { 2868 if (!lreq->is_watch) 2869 lreq->notify_id = 0; 2870 else 2871 WARN_ON(op->watch.op != CEPH_OSD_WATCH_OP_WATCH); 2872 dout("lreq %p register\n", lreq); 2873 req->r_callback = linger_commit_cb; 2874 } 2875 mutex_unlock(&lreq->lock); 2876 2877 req->r_priv = linger_get(lreq); 2878 req->r_linger = true; 2879 2880 submit_request(req, true); 2881 } 2882 2883 static void linger_ping_cb(struct ceph_osd_request *req) 2884 { 2885 struct ceph_osd_linger_request *lreq = req->r_priv; 2886 2887 mutex_lock(&lreq->lock); 2888 dout("%s lreq %p linger_id %llu result %d ping_sent %lu last_error %d\n", 2889 __func__, lreq, lreq->linger_id, req->r_result, lreq->ping_sent, 2890 lreq->last_error); 2891 if (lreq->register_gen == req->r_ops[0].watch.gen) { 2892 if (!req->r_result) { 2893 lreq->watch_valid_thru = lreq->ping_sent; 2894 } else if (!lreq->last_error) { 2895 lreq->last_error = normalize_watch_error(req->r_result); 2896 queue_watch_error(lreq); 2897 } 2898 } else { 2899 dout("lreq %p register_gen %u ignoring old pong %u\n", lreq, 2900 lreq->register_gen, req->r_ops[0].watch.gen); 2901 } 2902 2903 mutex_unlock(&lreq->lock); 2904 linger_put(lreq); 2905 } 2906 2907 static void send_linger_ping(struct ceph_osd_linger_request *lreq) 2908 { 2909 struct ceph_osd_client *osdc = lreq->osdc; 2910 struct ceph_osd_request *req = lreq->ping_req; 2911 struct ceph_osd_req_op *op = &req->r_ops[0]; 2912 2913 if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) { 2914 dout("%s PAUSERD\n", __func__); 2915 return; 2916 } 2917 2918 lreq->ping_sent = jiffies; 2919 dout("%s lreq %p linger_id %llu ping_sent %lu register_gen %u\n", 2920 __func__, lreq, lreq->linger_id, lreq->ping_sent, 2921 lreq->register_gen); 2922 2923 if (req->r_osd) 2924 cancel_linger_request(req); 2925 2926 request_reinit(req); 2927 target_copy(&req->r_t, &lreq->t); 2928 2929 WARN_ON(op->op != CEPH_OSD_OP_WATCH || 2930 op->watch.cookie != lreq->linger_id || 2931 op->watch.op != CEPH_OSD_WATCH_OP_PING); 2932 op->watch.gen = lreq->register_gen; 2933 req->r_callback = linger_ping_cb; 2934 req->r_priv = linger_get(lreq); 2935 req->r_linger = true; 2936 2937 ceph_osdc_get_request(req); 2938 account_request(req); 2939 req->r_tid = atomic64_inc_return(&osdc->last_tid); 2940 link_request(lreq->osd, req); 2941 send_request(req); 2942 } 2943 2944 static void linger_submit(struct ceph_osd_linger_request *lreq) 2945 { 2946 struct ceph_osd_client *osdc = lreq->osdc; 2947 struct ceph_osd *osd; 2948 2949 calc_target(osdc, &lreq->t, NULL, false); 2950 osd = lookup_create_osd(osdc, lreq->t.osd, true); 2951 link_linger(osd, lreq); 2952 2953 send_linger(lreq); 2954 } 2955 2956 static void cancel_linger_map_check(struct ceph_osd_linger_request *lreq) 2957 { 2958 struct ceph_osd_client *osdc = lreq->osdc; 2959 struct ceph_osd_linger_request *lookup_lreq; 2960 2961 verify_osdc_wrlocked(osdc); 2962 2963 lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks, 2964 lreq->linger_id); 2965 if (!lookup_lreq) 2966 return; 2967 2968 WARN_ON(lookup_lreq != lreq); 2969 erase_linger_mc(&osdc->linger_map_checks, lreq); 2970 linger_put(lreq); 2971 } 2972 2973 /* 2974 * @lreq has to be both registered and linked. 2975 */ 2976 static void __linger_cancel(struct ceph_osd_linger_request *lreq) 2977 { 2978 if (lreq->is_watch && lreq->ping_req->r_osd) 2979 cancel_linger_request(lreq->ping_req); 2980 if (lreq->reg_req->r_osd) 2981 cancel_linger_request(lreq->reg_req); 2982 cancel_linger_map_check(lreq); 2983 unlink_linger(lreq->osd, lreq); 2984 linger_unregister(lreq); 2985 } 2986 2987 static void linger_cancel(struct ceph_osd_linger_request *lreq) 2988 { 2989 struct ceph_osd_client *osdc = lreq->osdc; 2990 2991 down_write(&osdc->lock); 2992 if (__linger_registered(lreq)) 2993 __linger_cancel(lreq); 2994 up_write(&osdc->lock); 2995 } 2996 2997 static void send_linger_map_check(struct ceph_osd_linger_request *lreq); 2998 2999 static void check_linger_pool_dne(struct ceph_osd_linger_request *lreq) 3000 { 3001 struct ceph_osd_client *osdc = lreq->osdc; 3002 struct ceph_osdmap *map = osdc->osdmap; 3003 3004 verify_osdc_wrlocked(osdc); 3005 WARN_ON(!map->epoch); 3006 3007 if (lreq->register_gen) { 3008 lreq->map_dne_bound = map->epoch; 3009 dout("%s lreq %p linger_id %llu pool disappeared\n", __func__, 3010 lreq, lreq->linger_id); 3011 } else { 3012 dout("%s lreq %p linger_id %llu map_dne_bound %u have %u\n", 3013 __func__, lreq, lreq->linger_id, lreq->map_dne_bound, 3014 map->epoch); 3015 } 3016 3017 if (lreq->map_dne_bound) { 3018 if (map->epoch >= lreq->map_dne_bound) { 3019 /* we had a new enough map */ 3020 pr_info("linger_id %llu pool does not exist\n", 3021 lreq->linger_id); 3022 linger_reg_commit_complete(lreq, -ENOENT); 3023 __linger_cancel(lreq); 3024 } 3025 } else { 3026 send_linger_map_check(lreq); 3027 } 3028 } 3029 3030 static void linger_map_check_cb(struct ceph_mon_generic_request *greq) 3031 { 3032 struct ceph_osd_client *osdc = &greq->monc->client->osdc; 3033 struct ceph_osd_linger_request *lreq; 3034 u64 linger_id = greq->private_data; 3035 3036 WARN_ON(greq->result || !greq->u.newest); 3037 3038 down_write(&osdc->lock); 3039 lreq = lookup_linger_mc(&osdc->linger_map_checks, linger_id); 3040 if (!lreq) { 3041 dout("%s linger_id %llu dne\n", __func__, linger_id); 3042 goto out_unlock; 3043 } 3044 3045 dout("%s lreq %p linger_id %llu map_dne_bound %u newest %llu\n", 3046 __func__, lreq, lreq->linger_id, lreq->map_dne_bound, 3047 greq->u.newest); 3048 if (!lreq->map_dne_bound) 3049 lreq->map_dne_bound = greq->u.newest; 3050 erase_linger_mc(&osdc->linger_map_checks, lreq); 3051 check_linger_pool_dne(lreq); 3052 3053 linger_put(lreq); 3054 out_unlock: 3055 up_write(&osdc->lock); 3056 } 3057 3058 static void send_linger_map_check(struct ceph_osd_linger_request *lreq) 3059 { 3060 struct ceph_osd_client *osdc = lreq->osdc; 3061 struct ceph_osd_linger_request *lookup_lreq; 3062 int ret; 3063 3064 verify_osdc_wrlocked(osdc); 3065 3066 lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks, 3067 lreq->linger_id); 3068 if (lookup_lreq) { 3069 WARN_ON(lookup_lreq != lreq); 3070 return; 3071 } 3072 3073 linger_get(lreq); 3074 insert_linger_mc(&osdc->linger_map_checks, lreq); 3075 ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap", 3076 linger_map_check_cb, lreq->linger_id); 3077 WARN_ON(ret); 3078 } 3079 3080 static int linger_reg_commit_wait(struct ceph_osd_linger_request *lreq) 3081 { 3082 int ret; 3083 3084 dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id); 3085 ret = wait_for_completion_interruptible(&lreq->reg_commit_wait); 3086 return ret ?: lreq->reg_commit_error; 3087 } 3088 3089 static int linger_notify_finish_wait(struct ceph_osd_linger_request *lreq) 3090 { 3091 int ret; 3092 3093 dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id); 3094 ret = wait_for_completion_interruptible(&lreq->notify_finish_wait); 3095 return ret ?: lreq->notify_finish_error; 3096 } 3097 3098 /* 3099 * Timeout callback, called every N seconds. When 1 or more OSD 3100 * requests has been active for more than N seconds, we send a keepalive 3101 * (tag + timestamp) to its OSD to ensure any communications channel 3102 * reset is detected. 3103 */ 3104 static void handle_timeout(struct work_struct *work) 3105 { 3106 struct ceph_osd_client *osdc = 3107 container_of(work, struct ceph_osd_client, timeout_work.work); 3108 struct ceph_options *opts = osdc->client->options; 3109 unsigned long cutoff = jiffies - opts->osd_keepalive_timeout; 3110 unsigned long expiry_cutoff = jiffies - opts->osd_request_timeout; 3111 LIST_HEAD(slow_osds); 3112 struct rb_node *n, *p; 3113 3114 dout("%s osdc %p\n", __func__, osdc); 3115 down_write(&osdc->lock); 3116 3117 /* 3118 * ping osds that are a bit slow. this ensures that if there 3119 * is a break in the TCP connection we will notice, and reopen 3120 * a connection with that osd (from the fault callback). 3121 */ 3122 for (n = rb_first(&osdc->osds); n; n = rb_next(n)) { 3123 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node); 3124 bool found = false; 3125 3126 for (p = rb_first(&osd->o_requests); p; ) { 3127 struct ceph_osd_request *req = 3128 rb_entry(p, struct ceph_osd_request, r_node); 3129 3130 p = rb_next(p); /* abort_request() */ 3131 3132 if (time_before(req->r_stamp, cutoff)) { 3133 dout(" req %p tid %llu on osd%d is laggy\n", 3134 req, req->r_tid, osd->o_osd); 3135 found = true; 3136 } 3137 if (opts->osd_request_timeout && 3138 time_before(req->r_start_stamp, expiry_cutoff)) { 3139 pr_err_ratelimited("tid %llu on osd%d timeout\n", 3140 req->r_tid, osd->o_osd); 3141 abort_request(req, -ETIMEDOUT); 3142 } 3143 } 3144 for (p = rb_first(&osd->o_linger_requests); p; p = rb_next(p)) { 3145 struct ceph_osd_linger_request *lreq = 3146 rb_entry(p, struct ceph_osd_linger_request, node); 3147 3148 dout(" lreq %p linger_id %llu is served by osd%d\n", 3149 lreq, lreq->linger_id, osd->o_osd); 3150 found = true; 3151 3152 mutex_lock(&lreq->lock); 3153 if (lreq->is_watch && lreq->committed && !lreq->last_error) 3154 send_linger_ping(lreq); 3155 mutex_unlock(&lreq->lock); 3156 } 3157 3158 if (found) 3159 list_move_tail(&osd->o_keepalive_item, &slow_osds); 3160 } 3161 3162 if (opts->osd_request_timeout) { 3163 for (p = rb_first(&osdc->homeless_osd.o_requests); p; ) { 3164 struct ceph_osd_request *req = 3165 rb_entry(p, struct ceph_osd_request, r_node); 3166 3167 p = rb_next(p); /* abort_request() */ 3168 3169 if (time_before(req->r_start_stamp, expiry_cutoff)) { 3170 pr_err_ratelimited("tid %llu on osd%d timeout\n", 3171 req->r_tid, osdc->homeless_osd.o_osd); 3172 abort_request(req, -ETIMEDOUT); 3173 } 3174 } 3175 } 3176 3177 if (atomic_read(&osdc->num_homeless) || !list_empty(&slow_osds)) 3178 maybe_request_map(osdc); 3179 3180 while (!list_empty(&slow_osds)) { 3181 struct ceph_osd *osd = list_first_entry(&slow_osds, 3182 struct ceph_osd, 3183 o_keepalive_item); 3184 list_del_init(&osd->o_keepalive_item); 3185 ceph_con_keepalive(&osd->o_con); 3186 } 3187 3188 up_write(&osdc->lock); 3189 schedule_delayed_work(&osdc->timeout_work, 3190 osdc->client->options->osd_keepalive_timeout); 3191 } 3192 3193 static void handle_osds_timeout(struct work_struct *work) 3194 { 3195 struct ceph_osd_client *osdc = 3196 container_of(work, struct ceph_osd_client, 3197 osds_timeout_work.work); 3198 unsigned long delay = osdc->client->options->osd_idle_ttl / 4; 3199 struct ceph_osd *osd, *nosd; 3200 3201 dout("%s osdc %p\n", __func__, osdc); 3202 down_write(&osdc->lock); 3203 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) { 3204 if (time_before(jiffies, osd->lru_ttl)) 3205 break; 3206 3207 WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests)); 3208 WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests)); 3209 close_osd(osd); 3210 } 3211 3212 up_write(&osdc->lock); 3213 schedule_delayed_work(&osdc->osds_timeout_work, 3214 round_jiffies_relative(delay)); 3215 } 3216 3217 static int ceph_oloc_decode(void **p, void *end, 3218 struct ceph_object_locator *oloc) 3219 { 3220 u8 struct_v, struct_cv; 3221 u32 len; 3222 void *struct_end; 3223 int ret = 0; 3224 3225 ceph_decode_need(p, end, 1 + 1 + 4, e_inval); 3226 struct_v = ceph_decode_8(p); 3227 struct_cv = ceph_decode_8(p); 3228 if (struct_v < 3) { 3229 pr_warn("got v %d < 3 cv %d of ceph_object_locator\n", 3230 struct_v, struct_cv); 3231 goto e_inval; 3232 } 3233 if (struct_cv > 6) { 3234 pr_warn("got v %d cv %d > 6 of ceph_object_locator\n", 3235 struct_v, struct_cv); 3236 goto e_inval; 3237 } 3238 len = ceph_decode_32(p); 3239 ceph_decode_need(p, end, len, e_inval); 3240 struct_end = *p + len; 3241 3242 oloc->pool = ceph_decode_64(p); 3243 *p += 4; /* skip preferred */ 3244 3245 len = ceph_decode_32(p); 3246 if (len > 0) { 3247 pr_warn("ceph_object_locator::key is set\n"); 3248 goto e_inval; 3249 } 3250 3251 if (struct_v >= 5) { 3252 bool changed = false; 3253 3254 len = ceph_decode_32(p); 3255 if (len > 0) { 3256 ceph_decode_need(p, end, len, e_inval); 3257 if (!oloc->pool_ns || 3258 ceph_compare_string(oloc->pool_ns, *p, len)) 3259 changed = true; 3260 *p += len; 3261 } else { 3262 if (oloc->pool_ns) 3263 changed = true; 3264 } 3265 if (changed) { 3266 /* redirect changes namespace */ 3267 pr_warn("ceph_object_locator::nspace is changed\n"); 3268 goto e_inval; 3269 } 3270 } 3271 3272 if (struct_v >= 6) { 3273 s64 hash = ceph_decode_64(p); 3274 if (hash != -1) { 3275 pr_warn("ceph_object_locator::hash is set\n"); 3276 goto e_inval; 3277 } 3278 } 3279 3280 /* skip the rest */ 3281 *p = struct_end; 3282 out: 3283 return ret; 3284 3285 e_inval: 3286 ret = -EINVAL; 3287 goto out; 3288 } 3289 3290 static int ceph_redirect_decode(void **p, void *end, 3291 struct ceph_request_redirect *redir) 3292 { 3293 u8 struct_v, struct_cv; 3294 u32 len; 3295 void *struct_end; 3296 int ret; 3297 3298 ceph_decode_need(p, end, 1 + 1 + 4, e_inval); 3299 struct_v = ceph_decode_8(p); 3300 struct_cv = ceph_decode_8(p); 3301 if (struct_cv > 1) { 3302 pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n", 3303 struct_v, struct_cv); 3304 goto e_inval; 3305 } 3306 len = ceph_decode_32(p); 3307 ceph_decode_need(p, end, len, e_inval); 3308 struct_end = *p + len; 3309 3310 ret = ceph_oloc_decode(p, end, &redir->oloc); 3311 if (ret) 3312 goto out; 3313 3314 len = ceph_decode_32(p); 3315 if (len > 0) { 3316 pr_warn("ceph_request_redirect::object_name is set\n"); 3317 goto e_inval; 3318 } 3319 3320 len = ceph_decode_32(p); 3321 *p += len; /* skip osd_instructions */ 3322 3323 /* skip the rest */ 3324 *p = struct_end; 3325 out: 3326 return ret; 3327 3328 e_inval: 3329 ret = -EINVAL; 3330 goto out; 3331 } 3332 3333 struct MOSDOpReply { 3334 struct ceph_pg pgid; 3335 u64 flags; 3336 int result; 3337 u32 epoch; 3338 int num_ops; 3339 u32 outdata_len[CEPH_OSD_MAX_OPS]; 3340 s32 rval[CEPH_OSD_MAX_OPS]; 3341 int retry_attempt; 3342 struct ceph_eversion replay_version; 3343 u64 user_version; 3344 struct ceph_request_redirect redirect; 3345 }; 3346 3347 static int decode_MOSDOpReply(const struct ceph_msg *msg, struct MOSDOpReply *m) 3348 { 3349 void *p = msg->front.iov_base; 3350 void *const end = p + msg->front.iov_len; 3351 u16 version = le16_to_cpu(msg->hdr.version); 3352 struct ceph_eversion bad_replay_version; 3353 u8 decode_redir; 3354 u32 len; 3355 int ret; 3356 int i; 3357 3358 ceph_decode_32_safe(&p, end, len, e_inval); 3359 ceph_decode_need(&p, end, len, e_inval); 3360 p += len; /* skip oid */ 3361 3362 ret = ceph_decode_pgid(&p, end, &m->pgid); 3363 if (ret) 3364 return ret; 3365 3366 ceph_decode_64_safe(&p, end, m->flags, e_inval); 3367 ceph_decode_32_safe(&p, end, m->result, e_inval); 3368 ceph_decode_need(&p, end, sizeof(bad_replay_version), e_inval); 3369 memcpy(&bad_replay_version, p, sizeof(bad_replay_version)); 3370 p += sizeof(bad_replay_version); 3371 ceph_decode_32_safe(&p, end, m->epoch, e_inval); 3372 3373 ceph_decode_32_safe(&p, end, m->num_ops, e_inval); 3374 if (m->num_ops > ARRAY_SIZE(m->outdata_len)) 3375 goto e_inval; 3376 3377 ceph_decode_need(&p, end, m->num_ops * sizeof(struct ceph_osd_op), 3378 e_inval); 3379 for (i = 0; i < m->num_ops; i++) { 3380 struct ceph_osd_op *op = p; 3381 3382 m->outdata_len[i] = le32_to_cpu(op->payload_len); 3383 p += sizeof(*op); 3384 } 3385 3386 ceph_decode_32_safe(&p, end, m->retry_attempt, e_inval); 3387 for (i = 0; i < m->num_ops; i++) 3388 ceph_decode_32_safe(&p, end, m->rval[i], e_inval); 3389 3390 if (version >= 5) { 3391 ceph_decode_need(&p, end, sizeof(m->replay_version), e_inval); 3392 memcpy(&m->replay_version, p, sizeof(m->replay_version)); 3393 p += sizeof(m->replay_version); 3394 ceph_decode_64_safe(&p, end, m->user_version, e_inval); 3395 } else { 3396 m->replay_version = bad_replay_version; /* struct */ 3397 m->user_version = le64_to_cpu(m->replay_version.version); 3398 } 3399 3400 if (version >= 6) { 3401 if (version >= 7) 3402 ceph_decode_8_safe(&p, end, decode_redir, e_inval); 3403 else 3404 decode_redir = 1; 3405 } else { 3406 decode_redir = 0; 3407 } 3408 3409 if (decode_redir) { 3410 ret = ceph_redirect_decode(&p, end, &m->redirect); 3411 if (ret) 3412 return ret; 3413 } else { 3414 ceph_oloc_init(&m->redirect.oloc); 3415 } 3416 3417 return 0; 3418 3419 e_inval: 3420 return -EINVAL; 3421 } 3422 3423 /* 3424 * Handle MOSDOpReply. Set ->r_result and call the callback if it is 3425 * specified. 3426 */ 3427 static void handle_reply(struct ceph_osd *osd, struct ceph_msg *msg) 3428 { 3429 struct ceph_osd_client *osdc = osd->o_osdc; 3430 struct ceph_osd_request *req; 3431 struct MOSDOpReply m; 3432 u64 tid = le64_to_cpu(msg->hdr.tid); 3433 u32 data_len = 0; 3434 int ret; 3435 int i; 3436 3437 dout("%s msg %p tid %llu\n", __func__, msg, tid); 3438 3439 down_read(&osdc->lock); 3440 if (!osd_registered(osd)) { 3441 dout("%s osd%d unknown\n", __func__, osd->o_osd); 3442 goto out_unlock_osdc; 3443 } 3444 WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num)); 3445 3446 mutex_lock(&osd->lock); 3447 req = lookup_request(&osd->o_requests, tid); 3448 if (!req) { 3449 dout("%s osd%d tid %llu unknown\n", __func__, osd->o_osd, tid); 3450 goto out_unlock_session; 3451 } 3452 3453 m.redirect.oloc.pool_ns = req->r_t.target_oloc.pool_ns; 3454 ret = decode_MOSDOpReply(msg, &m); 3455 m.redirect.oloc.pool_ns = NULL; 3456 if (ret) { 3457 pr_err("failed to decode MOSDOpReply for tid %llu: %d\n", 3458 req->r_tid, ret); 3459 ceph_msg_dump(msg); 3460 goto fail_request; 3461 } 3462 dout("%s req %p tid %llu flags 0x%llx pgid %llu.%x epoch %u attempt %d v %u'%llu uv %llu\n", 3463 __func__, req, req->r_tid, m.flags, m.pgid.pool, m.pgid.seed, 3464 m.epoch, m.retry_attempt, le32_to_cpu(m.replay_version.epoch), 3465 le64_to_cpu(m.replay_version.version), m.user_version); 3466 3467 if (m.retry_attempt >= 0) { 3468 if (m.retry_attempt != req->r_attempts - 1) { 3469 dout("req %p tid %llu retry_attempt %d != %d, ignoring\n", 3470 req, req->r_tid, m.retry_attempt, 3471 req->r_attempts - 1); 3472 goto out_unlock_session; 3473 } 3474 } else { 3475 WARN_ON(1); /* MOSDOpReply v4 is assumed */ 3476 } 3477 3478 if (!ceph_oloc_empty(&m.redirect.oloc)) { 3479 dout("req %p tid %llu redirect pool %lld\n", req, req->r_tid, 3480 m.redirect.oloc.pool); 3481 unlink_request(osd, req); 3482 mutex_unlock(&osd->lock); 3483 3484 /* 3485 * Not ceph_oloc_copy() - changing pool_ns is not 3486 * supported. 3487 */ 3488 req->r_t.target_oloc.pool = m.redirect.oloc.pool; 3489 req->r_flags |= CEPH_OSD_FLAG_REDIRECTED; 3490 req->r_tid = 0; 3491 __submit_request(req, false); 3492 goto out_unlock_osdc; 3493 } 3494 3495 if (m.num_ops != req->r_num_ops) { 3496 pr_err("num_ops %d != %d for tid %llu\n", m.num_ops, 3497 req->r_num_ops, req->r_tid); 3498 goto fail_request; 3499 } 3500 for (i = 0; i < req->r_num_ops; i++) { 3501 dout(" req %p tid %llu op %d rval %d len %u\n", req, 3502 req->r_tid, i, m.rval[i], m.outdata_len[i]); 3503 req->r_ops[i].rval = m.rval[i]; 3504 req->r_ops[i].outdata_len = m.outdata_len[i]; 3505 data_len += m.outdata_len[i]; 3506 } 3507 if (data_len != le32_to_cpu(msg->hdr.data_len)) { 3508 pr_err("sum of lens %u != %u for tid %llu\n", data_len, 3509 le32_to_cpu(msg->hdr.data_len), req->r_tid); 3510 goto fail_request; 3511 } 3512 dout("%s req %p tid %llu result %d data_len %u\n", __func__, 3513 req, req->r_tid, m.result, data_len); 3514 3515 /* 3516 * Since we only ever request ONDISK, we should only ever get 3517 * one (type of) reply back. 3518 */ 3519 WARN_ON(!(m.flags & CEPH_OSD_FLAG_ONDISK)); 3520 req->r_result = m.result ?: data_len; 3521 finish_request(req); 3522 mutex_unlock(&osd->lock); 3523 up_read(&osdc->lock); 3524 3525 __complete_request(req); 3526 complete_all(&req->r_completion); 3527 ceph_osdc_put_request(req); 3528 return; 3529 3530 fail_request: 3531 complete_request(req, -EIO); 3532 out_unlock_session: 3533 mutex_unlock(&osd->lock); 3534 out_unlock_osdc: 3535 up_read(&osdc->lock); 3536 } 3537 3538 static void set_pool_was_full(struct ceph_osd_client *osdc) 3539 { 3540 struct rb_node *n; 3541 3542 for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) { 3543 struct ceph_pg_pool_info *pi = 3544 rb_entry(n, struct ceph_pg_pool_info, node); 3545 3546 pi->was_full = __pool_full(pi); 3547 } 3548 } 3549 3550 static bool pool_cleared_full(struct ceph_osd_client *osdc, s64 pool_id) 3551 { 3552 struct ceph_pg_pool_info *pi; 3553 3554 pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id); 3555 if (!pi) 3556 return false; 3557 3558 return pi->was_full && !__pool_full(pi); 3559 } 3560 3561 static enum calc_target_result 3562 recalc_linger_target(struct ceph_osd_linger_request *lreq) 3563 { 3564 struct ceph_osd_client *osdc = lreq->osdc; 3565 enum calc_target_result ct_res; 3566 3567 ct_res = calc_target(osdc, &lreq->t, NULL, true); 3568 if (ct_res == CALC_TARGET_NEED_RESEND) { 3569 struct ceph_osd *osd; 3570 3571 osd = lookup_create_osd(osdc, lreq->t.osd, true); 3572 if (osd != lreq->osd) { 3573 unlink_linger(lreq->osd, lreq); 3574 link_linger(osd, lreq); 3575 } 3576 } 3577 3578 return ct_res; 3579 } 3580 3581 /* 3582 * Requeue requests whose mapping to an OSD has changed. 3583 */ 3584 static void scan_requests(struct ceph_osd *osd, 3585 bool force_resend, 3586 bool cleared_full, 3587 bool check_pool_cleared_full, 3588 struct rb_root *need_resend, 3589 struct list_head *need_resend_linger) 3590 { 3591 struct ceph_osd_client *osdc = osd->o_osdc; 3592 struct rb_node *n; 3593 bool force_resend_writes; 3594 3595 for (n = rb_first(&osd->o_linger_requests); n; ) { 3596 struct ceph_osd_linger_request *lreq = 3597 rb_entry(n, struct ceph_osd_linger_request, node); 3598 enum calc_target_result ct_res; 3599 3600 n = rb_next(n); /* recalc_linger_target() */ 3601 3602 dout("%s lreq %p linger_id %llu\n", __func__, lreq, 3603 lreq->linger_id); 3604 ct_res = recalc_linger_target(lreq); 3605 switch (ct_res) { 3606 case CALC_TARGET_NO_ACTION: 3607 force_resend_writes = cleared_full || 3608 (check_pool_cleared_full && 3609 pool_cleared_full(osdc, lreq->t.base_oloc.pool)); 3610 if (!force_resend && !force_resend_writes) 3611 break; 3612 3613 /* fall through */ 3614 case CALC_TARGET_NEED_RESEND: 3615 cancel_linger_map_check(lreq); 3616 /* 3617 * scan_requests() for the previous epoch(s) 3618 * may have already added it to the list, since 3619 * it's not unlinked here. 3620 */ 3621 if (list_empty(&lreq->scan_item)) 3622 list_add_tail(&lreq->scan_item, need_resend_linger); 3623 break; 3624 case CALC_TARGET_POOL_DNE: 3625 list_del_init(&lreq->scan_item); 3626 check_linger_pool_dne(lreq); 3627 break; 3628 } 3629 } 3630 3631 for (n = rb_first(&osd->o_requests); n; ) { 3632 struct ceph_osd_request *req = 3633 rb_entry(n, struct ceph_osd_request, r_node); 3634 enum calc_target_result ct_res; 3635 3636 n = rb_next(n); /* unlink_request(), check_pool_dne() */ 3637 3638 dout("%s req %p tid %llu\n", __func__, req, req->r_tid); 3639 ct_res = calc_target(osdc, &req->r_t, &req->r_osd->o_con, 3640 false); 3641 switch (ct_res) { 3642 case CALC_TARGET_NO_ACTION: 3643 force_resend_writes = cleared_full || 3644 (check_pool_cleared_full && 3645 pool_cleared_full(osdc, req->r_t.base_oloc.pool)); 3646 if (!force_resend && 3647 (!(req->r_flags & CEPH_OSD_FLAG_WRITE) || 3648 !force_resend_writes)) 3649 break; 3650 3651 /* fall through */ 3652 case CALC_TARGET_NEED_RESEND: 3653 cancel_map_check(req); 3654 unlink_request(osd, req); 3655 insert_request(need_resend, req); 3656 break; 3657 case CALC_TARGET_POOL_DNE: 3658 check_pool_dne(req); 3659 break; 3660 } 3661 } 3662 } 3663 3664 static int handle_one_map(struct ceph_osd_client *osdc, 3665 void *p, void *end, bool incremental, 3666 struct rb_root *need_resend, 3667 struct list_head *need_resend_linger) 3668 { 3669 struct ceph_osdmap *newmap; 3670 struct rb_node *n; 3671 bool skipped_map = false; 3672 bool was_full; 3673 3674 was_full = ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL); 3675 set_pool_was_full(osdc); 3676 3677 if (incremental) 3678 newmap = osdmap_apply_incremental(&p, end, osdc->osdmap); 3679 else 3680 newmap = ceph_osdmap_decode(&p, end); 3681 if (IS_ERR(newmap)) 3682 return PTR_ERR(newmap); 3683 3684 if (newmap != osdc->osdmap) { 3685 /* 3686 * Preserve ->was_full before destroying the old map. 3687 * For pools that weren't in the old map, ->was_full 3688 * should be false. 3689 */ 3690 for (n = rb_first(&newmap->pg_pools); n; n = rb_next(n)) { 3691 struct ceph_pg_pool_info *pi = 3692 rb_entry(n, struct ceph_pg_pool_info, node); 3693 struct ceph_pg_pool_info *old_pi; 3694 3695 old_pi = ceph_pg_pool_by_id(osdc->osdmap, pi->id); 3696 if (old_pi) 3697 pi->was_full = old_pi->was_full; 3698 else 3699 WARN_ON(pi->was_full); 3700 } 3701 3702 if (osdc->osdmap->epoch && 3703 osdc->osdmap->epoch + 1 < newmap->epoch) { 3704 WARN_ON(incremental); 3705 skipped_map = true; 3706 } 3707 3708 ceph_osdmap_destroy(osdc->osdmap); 3709 osdc->osdmap = newmap; 3710 } 3711 3712 was_full &= !ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL); 3713 scan_requests(&osdc->homeless_osd, skipped_map, was_full, true, 3714 need_resend, need_resend_linger); 3715 3716 for (n = rb_first(&osdc->osds); n; ) { 3717 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node); 3718 3719 n = rb_next(n); /* close_osd() */ 3720 3721 scan_requests(osd, skipped_map, was_full, true, need_resend, 3722 need_resend_linger); 3723 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) || 3724 memcmp(&osd->o_con.peer_addr, 3725 ceph_osd_addr(osdc->osdmap, osd->o_osd), 3726 sizeof(struct ceph_entity_addr))) 3727 close_osd(osd); 3728 } 3729 3730 return 0; 3731 } 3732 3733 static void kick_requests(struct ceph_osd_client *osdc, 3734 struct rb_root *need_resend, 3735 struct list_head *need_resend_linger) 3736 { 3737 struct ceph_osd_linger_request *lreq, *nlreq; 3738 enum calc_target_result ct_res; 3739 struct rb_node *n; 3740 3741 /* make sure need_resend targets reflect latest map */ 3742 for (n = rb_first(need_resend); n; ) { 3743 struct ceph_osd_request *req = 3744 rb_entry(n, struct ceph_osd_request, r_node); 3745 3746 n = rb_next(n); 3747 3748 if (req->r_t.epoch < osdc->osdmap->epoch) { 3749 ct_res = calc_target(osdc, &req->r_t, NULL, false); 3750 if (ct_res == CALC_TARGET_POOL_DNE) { 3751 erase_request(need_resend, req); 3752 check_pool_dne(req); 3753 } 3754 } 3755 } 3756 3757 for (n = rb_first(need_resend); n; ) { 3758 struct ceph_osd_request *req = 3759 rb_entry(n, struct ceph_osd_request, r_node); 3760 struct ceph_osd *osd; 3761 3762 n = rb_next(n); 3763 erase_request(need_resend, req); /* before link_request() */ 3764 3765 osd = lookup_create_osd(osdc, req->r_t.osd, true); 3766 link_request(osd, req); 3767 if (!req->r_linger) { 3768 if (!osd_homeless(osd) && !req->r_t.paused) 3769 send_request(req); 3770 } else { 3771 cancel_linger_request(req); 3772 } 3773 } 3774 3775 list_for_each_entry_safe(lreq, nlreq, need_resend_linger, scan_item) { 3776 if (!osd_homeless(lreq->osd)) 3777 send_linger(lreq); 3778 3779 list_del_init(&lreq->scan_item); 3780 } 3781 } 3782 3783 /* 3784 * Process updated osd map. 3785 * 3786 * The message contains any number of incremental and full maps, normally 3787 * indicating some sort of topology change in the cluster. Kick requests 3788 * off to different OSDs as needed. 3789 */ 3790 void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg) 3791 { 3792 void *p = msg->front.iov_base; 3793 void *const end = p + msg->front.iov_len; 3794 u32 nr_maps, maplen; 3795 u32 epoch; 3796 struct ceph_fsid fsid; 3797 struct rb_root need_resend = RB_ROOT; 3798 LIST_HEAD(need_resend_linger); 3799 bool handled_incremental = false; 3800 bool was_pauserd, was_pausewr; 3801 bool pauserd, pausewr; 3802 int err; 3803 3804 dout("%s have %u\n", __func__, osdc->osdmap->epoch); 3805 down_write(&osdc->lock); 3806 3807 /* verify fsid */ 3808 ceph_decode_need(&p, end, sizeof(fsid), bad); 3809 ceph_decode_copy(&p, &fsid, sizeof(fsid)); 3810 if (ceph_check_fsid(osdc->client, &fsid) < 0) 3811 goto bad; 3812 3813 was_pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD); 3814 was_pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) || 3815 ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || 3816 have_pool_full(osdc); 3817 3818 /* incremental maps */ 3819 ceph_decode_32_safe(&p, end, nr_maps, bad); 3820 dout(" %d inc maps\n", nr_maps); 3821 while (nr_maps > 0) { 3822 ceph_decode_need(&p, end, 2*sizeof(u32), bad); 3823 epoch = ceph_decode_32(&p); 3824 maplen = ceph_decode_32(&p); 3825 ceph_decode_need(&p, end, maplen, bad); 3826 if (osdc->osdmap->epoch && 3827 osdc->osdmap->epoch + 1 == epoch) { 3828 dout("applying incremental map %u len %d\n", 3829 epoch, maplen); 3830 err = handle_one_map(osdc, p, p + maplen, true, 3831 &need_resend, &need_resend_linger); 3832 if (err) 3833 goto bad; 3834 handled_incremental = true; 3835 } else { 3836 dout("ignoring incremental map %u len %d\n", 3837 epoch, maplen); 3838 } 3839 p += maplen; 3840 nr_maps--; 3841 } 3842 if (handled_incremental) 3843 goto done; 3844 3845 /* full maps */ 3846 ceph_decode_32_safe(&p, end, nr_maps, bad); 3847 dout(" %d full maps\n", nr_maps); 3848 while (nr_maps) { 3849 ceph_decode_need(&p, end, 2*sizeof(u32), bad); 3850 epoch = ceph_decode_32(&p); 3851 maplen = ceph_decode_32(&p); 3852 ceph_decode_need(&p, end, maplen, bad); 3853 if (nr_maps > 1) { 3854 dout("skipping non-latest full map %u len %d\n", 3855 epoch, maplen); 3856 } else if (osdc->osdmap->epoch >= epoch) { 3857 dout("skipping full map %u len %d, " 3858 "older than our %u\n", epoch, maplen, 3859 osdc->osdmap->epoch); 3860 } else { 3861 dout("taking full map %u len %d\n", epoch, maplen); 3862 err = handle_one_map(osdc, p, p + maplen, false, 3863 &need_resend, &need_resend_linger); 3864 if (err) 3865 goto bad; 3866 } 3867 p += maplen; 3868 nr_maps--; 3869 } 3870 3871 done: 3872 /* 3873 * subscribe to subsequent osdmap updates if full to ensure 3874 * we find out when we are no longer full and stop returning 3875 * ENOSPC. 3876 */ 3877 pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD); 3878 pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) || 3879 ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || 3880 have_pool_full(osdc); 3881 if (was_pauserd || was_pausewr || pauserd || pausewr || 3882 osdc->osdmap->epoch < osdc->epoch_barrier) 3883 maybe_request_map(osdc); 3884 3885 kick_requests(osdc, &need_resend, &need_resend_linger); 3886 3887 ceph_osdc_abort_on_full(osdc); 3888 ceph_monc_got_map(&osdc->client->monc, CEPH_SUB_OSDMAP, 3889 osdc->osdmap->epoch); 3890 up_write(&osdc->lock); 3891 wake_up_all(&osdc->client->auth_wq); 3892 return; 3893 3894 bad: 3895 pr_err("osdc handle_map corrupt msg\n"); 3896 ceph_msg_dump(msg); 3897 up_write(&osdc->lock); 3898 } 3899 3900 /* 3901 * Resubmit requests pending on the given osd. 3902 */ 3903 static void kick_osd_requests(struct ceph_osd *osd) 3904 { 3905 struct rb_node *n; 3906 3907 clear_backoffs(osd); 3908 3909 for (n = rb_first(&osd->o_requests); n; ) { 3910 struct ceph_osd_request *req = 3911 rb_entry(n, struct ceph_osd_request, r_node); 3912 3913 n = rb_next(n); /* cancel_linger_request() */ 3914 3915 if (!req->r_linger) { 3916 if (!req->r_t.paused) 3917 send_request(req); 3918 } else { 3919 cancel_linger_request(req); 3920 } 3921 } 3922 for (n = rb_first(&osd->o_linger_requests); n; n = rb_next(n)) { 3923 struct ceph_osd_linger_request *lreq = 3924 rb_entry(n, struct ceph_osd_linger_request, node); 3925 3926 send_linger(lreq); 3927 } 3928 } 3929 3930 /* 3931 * If the osd connection drops, we need to resubmit all requests. 3932 */ 3933 static void osd_fault(struct ceph_connection *con) 3934 { 3935 struct ceph_osd *osd = con->private; 3936 struct ceph_osd_client *osdc = osd->o_osdc; 3937 3938 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd); 3939 3940 down_write(&osdc->lock); 3941 if (!osd_registered(osd)) { 3942 dout("%s osd%d unknown\n", __func__, osd->o_osd); 3943 goto out_unlock; 3944 } 3945 3946 if (!reopen_osd(osd)) 3947 kick_osd_requests(osd); 3948 maybe_request_map(osdc); 3949 3950 out_unlock: 3951 up_write(&osdc->lock); 3952 } 3953 3954 struct MOSDBackoff { 3955 struct ceph_spg spgid; 3956 u32 map_epoch; 3957 u8 op; 3958 u64 id; 3959 struct ceph_hobject_id *begin; 3960 struct ceph_hobject_id *end; 3961 }; 3962 3963 static int decode_MOSDBackoff(const struct ceph_msg *msg, struct MOSDBackoff *m) 3964 { 3965 void *p = msg->front.iov_base; 3966 void *const end = p + msg->front.iov_len; 3967 u8 struct_v; 3968 u32 struct_len; 3969 int ret; 3970 3971 ret = ceph_start_decoding(&p, end, 1, "spg_t", &struct_v, &struct_len); 3972 if (ret) 3973 return ret; 3974 3975 ret = ceph_decode_pgid(&p, end, &m->spgid.pgid); 3976 if (ret) 3977 return ret; 3978 3979 ceph_decode_8_safe(&p, end, m->spgid.shard, e_inval); 3980 ceph_decode_32_safe(&p, end, m->map_epoch, e_inval); 3981 ceph_decode_8_safe(&p, end, m->op, e_inval); 3982 ceph_decode_64_safe(&p, end, m->id, e_inval); 3983 3984 m->begin = kzalloc(sizeof(*m->begin), GFP_NOIO); 3985 if (!m->begin) 3986 return -ENOMEM; 3987 3988 ret = decode_hoid(&p, end, m->begin); 3989 if (ret) { 3990 free_hoid(m->begin); 3991 return ret; 3992 } 3993 3994 m->end = kzalloc(sizeof(*m->end), GFP_NOIO); 3995 if (!m->end) { 3996 free_hoid(m->begin); 3997 return -ENOMEM; 3998 } 3999 4000 ret = decode_hoid(&p, end, m->end); 4001 if (ret) { 4002 free_hoid(m->begin); 4003 free_hoid(m->end); 4004 return ret; 4005 } 4006 4007 return 0; 4008 4009 e_inval: 4010 return -EINVAL; 4011 } 4012 4013 static struct ceph_msg *create_backoff_message( 4014 const struct ceph_osd_backoff *backoff, 4015 u32 map_epoch) 4016 { 4017 struct ceph_msg *msg; 4018 void *p, *end; 4019 int msg_size; 4020 4021 msg_size = CEPH_ENCODING_START_BLK_LEN + 4022 CEPH_PGID_ENCODING_LEN + 1; /* spgid */ 4023 msg_size += 4 + 1 + 8; /* map_epoch, op, id */ 4024 msg_size += CEPH_ENCODING_START_BLK_LEN + 4025 hoid_encoding_size(backoff->begin); 4026 msg_size += CEPH_ENCODING_START_BLK_LEN + 4027 hoid_encoding_size(backoff->end); 4028 4029 msg = ceph_msg_new(CEPH_MSG_OSD_BACKOFF, msg_size, GFP_NOIO, true); 4030 if (!msg) 4031 return NULL; 4032 4033 p = msg->front.iov_base; 4034 end = p + msg->front_alloc_len; 4035 4036 encode_spgid(&p, &backoff->spgid); 4037 ceph_encode_32(&p, map_epoch); 4038 ceph_encode_8(&p, CEPH_OSD_BACKOFF_OP_ACK_BLOCK); 4039 ceph_encode_64(&p, backoff->id); 4040 encode_hoid(&p, end, backoff->begin); 4041 encode_hoid(&p, end, backoff->end); 4042 BUG_ON(p != end); 4043 4044 msg->front.iov_len = p - msg->front.iov_base; 4045 msg->hdr.version = cpu_to_le16(1); /* MOSDBackoff v1 */ 4046 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len); 4047 4048 return msg; 4049 } 4050 4051 static void handle_backoff_block(struct ceph_osd *osd, struct MOSDBackoff *m) 4052 { 4053 struct ceph_spg_mapping *spg; 4054 struct ceph_osd_backoff *backoff; 4055 struct ceph_msg *msg; 4056 4057 dout("%s osd%d spgid %llu.%xs%d id %llu\n", __func__, osd->o_osd, 4058 m->spgid.pgid.pool, m->spgid.pgid.seed, m->spgid.shard, m->id); 4059 4060 spg = lookup_spg_mapping(&osd->o_backoff_mappings, &m->spgid); 4061 if (!spg) { 4062 spg = alloc_spg_mapping(); 4063 if (!spg) { 4064 pr_err("%s failed to allocate spg\n", __func__); 4065 return; 4066 } 4067 spg->spgid = m->spgid; /* struct */ 4068 insert_spg_mapping(&osd->o_backoff_mappings, spg); 4069 } 4070 4071 backoff = alloc_backoff(); 4072 if (!backoff) { 4073 pr_err("%s failed to allocate backoff\n", __func__); 4074 return; 4075 } 4076 backoff->spgid = m->spgid; /* struct */ 4077 backoff->id = m->id; 4078 backoff->begin = m->begin; 4079 m->begin = NULL; /* backoff now owns this */ 4080 backoff->end = m->end; 4081 m->end = NULL; /* ditto */ 4082 4083 insert_backoff(&spg->backoffs, backoff); 4084 insert_backoff_by_id(&osd->o_backoffs_by_id, backoff); 4085 4086 /* 4087 * Ack with original backoff's epoch so that the OSD can 4088 * discard this if there was a PG split. 4089 */ 4090 msg = create_backoff_message(backoff, m->map_epoch); 4091 if (!msg) { 4092 pr_err("%s failed to allocate msg\n", __func__); 4093 return; 4094 } 4095 ceph_con_send(&osd->o_con, msg); 4096 } 4097 4098 static bool target_contained_by(const struct ceph_osd_request_target *t, 4099 const struct ceph_hobject_id *begin, 4100 const struct ceph_hobject_id *end) 4101 { 4102 struct ceph_hobject_id hoid; 4103 int cmp; 4104 4105 hoid_fill_from_target(&hoid, t); 4106 cmp = hoid_compare(&hoid, begin); 4107 return !cmp || (cmp > 0 && hoid_compare(&hoid, end) < 0); 4108 } 4109 4110 static void handle_backoff_unblock(struct ceph_osd *osd, 4111 const struct MOSDBackoff *m) 4112 { 4113 struct ceph_spg_mapping *spg; 4114 struct ceph_osd_backoff *backoff; 4115 struct rb_node *n; 4116 4117 dout("%s osd%d spgid %llu.%xs%d id %llu\n", __func__, osd->o_osd, 4118 m->spgid.pgid.pool, m->spgid.pgid.seed, m->spgid.shard, m->id); 4119 4120 backoff = lookup_backoff_by_id(&osd->o_backoffs_by_id, m->id); 4121 if (!backoff) { 4122 pr_err("%s osd%d spgid %llu.%xs%d id %llu backoff dne\n", 4123 __func__, osd->o_osd, m->spgid.pgid.pool, 4124 m->spgid.pgid.seed, m->spgid.shard, m->id); 4125 return; 4126 } 4127 4128 if (hoid_compare(backoff->begin, m->begin) && 4129 hoid_compare(backoff->end, m->end)) { 4130 pr_err("%s osd%d spgid %llu.%xs%d id %llu bad range?\n", 4131 __func__, osd->o_osd, m->spgid.pgid.pool, 4132 m->spgid.pgid.seed, m->spgid.shard, m->id); 4133 /* unblock it anyway... */ 4134 } 4135 4136 spg = lookup_spg_mapping(&osd->o_backoff_mappings, &backoff->spgid); 4137 BUG_ON(!spg); 4138 4139 erase_backoff(&spg->backoffs, backoff); 4140 erase_backoff_by_id(&osd->o_backoffs_by_id, backoff); 4141 free_backoff(backoff); 4142 4143 if (RB_EMPTY_ROOT(&spg->backoffs)) { 4144 erase_spg_mapping(&osd->o_backoff_mappings, spg); 4145 free_spg_mapping(spg); 4146 } 4147 4148 for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) { 4149 struct ceph_osd_request *req = 4150 rb_entry(n, struct ceph_osd_request, r_node); 4151 4152 if (!ceph_spg_compare(&req->r_t.spgid, &m->spgid)) { 4153 /* 4154 * Match against @m, not @backoff -- the PG may 4155 * have split on the OSD. 4156 */ 4157 if (target_contained_by(&req->r_t, m->begin, m->end)) { 4158 /* 4159 * If no other installed backoff applies, 4160 * resend. 4161 */ 4162 send_request(req); 4163 } 4164 } 4165 } 4166 } 4167 4168 static void handle_backoff(struct ceph_osd *osd, struct ceph_msg *msg) 4169 { 4170 struct ceph_osd_client *osdc = osd->o_osdc; 4171 struct MOSDBackoff m; 4172 int ret; 4173 4174 down_read(&osdc->lock); 4175 if (!osd_registered(osd)) { 4176 dout("%s osd%d unknown\n", __func__, osd->o_osd); 4177 up_read(&osdc->lock); 4178 return; 4179 } 4180 WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num)); 4181 4182 mutex_lock(&osd->lock); 4183 ret = decode_MOSDBackoff(msg, &m); 4184 if (ret) { 4185 pr_err("failed to decode MOSDBackoff: %d\n", ret); 4186 ceph_msg_dump(msg); 4187 goto out_unlock; 4188 } 4189 4190 switch (m.op) { 4191 case CEPH_OSD_BACKOFF_OP_BLOCK: 4192 handle_backoff_block(osd, &m); 4193 break; 4194 case CEPH_OSD_BACKOFF_OP_UNBLOCK: 4195 handle_backoff_unblock(osd, &m); 4196 break; 4197 default: 4198 pr_err("%s osd%d unknown op %d\n", __func__, osd->o_osd, m.op); 4199 } 4200 4201 free_hoid(m.begin); 4202 free_hoid(m.end); 4203 4204 out_unlock: 4205 mutex_unlock(&osd->lock); 4206 up_read(&osdc->lock); 4207 } 4208 4209 /* 4210 * Process osd watch notifications 4211 */ 4212 static void handle_watch_notify(struct ceph_osd_client *osdc, 4213 struct ceph_msg *msg) 4214 { 4215 void *p = msg->front.iov_base; 4216 void *const end = p + msg->front.iov_len; 4217 struct ceph_osd_linger_request *lreq; 4218 struct linger_work *lwork; 4219 u8 proto_ver, opcode; 4220 u64 cookie, notify_id; 4221 u64 notifier_id = 0; 4222 s32 return_code = 0; 4223 void *payload = NULL; 4224 u32 payload_len = 0; 4225 4226 ceph_decode_8_safe(&p, end, proto_ver, bad); 4227 ceph_decode_8_safe(&p, end, opcode, bad); 4228 ceph_decode_64_safe(&p, end, cookie, bad); 4229 p += 8; /* skip ver */ 4230 ceph_decode_64_safe(&p, end, notify_id, bad); 4231 4232 if (proto_ver >= 1) { 4233 ceph_decode_32_safe(&p, end, payload_len, bad); 4234 ceph_decode_need(&p, end, payload_len, bad); 4235 payload = p; 4236 p += payload_len; 4237 } 4238 4239 if (le16_to_cpu(msg->hdr.version) >= 2) 4240 ceph_decode_32_safe(&p, end, return_code, bad); 4241 4242 if (le16_to_cpu(msg->hdr.version) >= 3) 4243 ceph_decode_64_safe(&p, end, notifier_id, bad); 4244 4245 down_read(&osdc->lock); 4246 lreq = lookup_linger_osdc(&osdc->linger_requests, cookie); 4247 if (!lreq) { 4248 dout("%s opcode %d cookie %llu dne\n", __func__, opcode, 4249 cookie); 4250 goto out_unlock_osdc; 4251 } 4252 4253 mutex_lock(&lreq->lock); 4254 dout("%s opcode %d cookie %llu lreq %p is_watch %d\n", __func__, 4255 opcode, cookie, lreq, lreq->is_watch); 4256 if (opcode == CEPH_WATCH_EVENT_DISCONNECT) { 4257 if (!lreq->last_error) { 4258 lreq->last_error = -ENOTCONN; 4259 queue_watch_error(lreq); 4260 } 4261 } else if (!lreq->is_watch) { 4262 /* CEPH_WATCH_EVENT_NOTIFY_COMPLETE */ 4263 if (lreq->notify_id && lreq->notify_id != notify_id) { 4264 dout("lreq %p notify_id %llu != %llu, ignoring\n", lreq, 4265 lreq->notify_id, notify_id); 4266 } else if (!completion_done(&lreq->notify_finish_wait)) { 4267 struct ceph_msg_data *data = 4268 list_first_entry_or_null(&msg->data, 4269 struct ceph_msg_data, 4270 links); 4271 4272 if (data) { 4273 if (lreq->preply_pages) { 4274 WARN_ON(data->type != 4275 CEPH_MSG_DATA_PAGES); 4276 *lreq->preply_pages = data->pages; 4277 *lreq->preply_len = data->length; 4278 } else { 4279 ceph_release_page_vector(data->pages, 4280 calc_pages_for(0, data->length)); 4281 } 4282 } 4283 lreq->notify_finish_error = return_code; 4284 complete_all(&lreq->notify_finish_wait); 4285 } 4286 } else { 4287 /* CEPH_WATCH_EVENT_NOTIFY */ 4288 lwork = lwork_alloc(lreq, do_watch_notify); 4289 if (!lwork) { 4290 pr_err("failed to allocate notify-lwork\n"); 4291 goto out_unlock_lreq; 4292 } 4293 4294 lwork->notify.notify_id = notify_id; 4295 lwork->notify.notifier_id = notifier_id; 4296 lwork->notify.payload = payload; 4297 lwork->notify.payload_len = payload_len; 4298 lwork->notify.msg = ceph_msg_get(msg); 4299 lwork_queue(lwork); 4300 } 4301 4302 out_unlock_lreq: 4303 mutex_unlock(&lreq->lock); 4304 out_unlock_osdc: 4305 up_read(&osdc->lock); 4306 return; 4307 4308 bad: 4309 pr_err("osdc handle_watch_notify corrupt msg\n"); 4310 } 4311 4312 /* 4313 * Register request, send initial attempt. 4314 */ 4315 int ceph_osdc_start_request(struct ceph_osd_client *osdc, 4316 struct ceph_osd_request *req, 4317 bool nofail) 4318 { 4319 down_read(&osdc->lock); 4320 submit_request(req, false); 4321 up_read(&osdc->lock); 4322 4323 return 0; 4324 } 4325 EXPORT_SYMBOL(ceph_osdc_start_request); 4326 4327 /* 4328 * Unregister a registered request. The request is not completed: 4329 * ->r_result isn't set and __complete_request() isn't called. 4330 */ 4331 void ceph_osdc_cancel_request(struct ceph_osd_request *req) 4332 { 4333 struct ceph_osd_client *osdc = req->r_osdc; 4334 4335 down_write(&osdc->lock); 4336 if (req->r_osd) 4337 cancel_request(req); 4338 up_write(&osdc->lock); 4339 } 4340 EXPORT_SYMBOL(ceph_osdc_cancel_request); 4341 4342 /* 4343 * @timeout: in jiffies, 0 means "wait forever" 4344 */ 4345 static int wait_request_timeout(struct ceph_osd_request *req, 4346 unsigned long timeout) 4347 { 4348 long left; 4349 4350 dout("%s req %p tid %llu\n", __func__, req, req->r_tid); 4351 left = wait_for_completion_killable_timeout(&req->r_completion, 4352 ceph_timeout_jiffies(timeout)); 4353 if (left <= 0) { 4354 left = left ?: -ETIMEDOUT; 4355 ceph_osdc_cancel_request(req); 4356 } else { 4357 left = req->r_result; /* completed */ 4358 } 4359 4360 return left; 4361 } 4362 4363 /* 4364 * wait for a request to complete 4365 */ 4366 int ceph_osdc_wait_request(struct ceph_osd_client *osdc, 4367 struct ceph_osd_request *req) 4368 { 4369 return wait_request_timeout(req, 0); 4370 } 4371 EXPORT_SYMBOL(ceph_osdc_wait_request); 4372 4373 /* 4374 * sync - wait for all in-flight requests to flush. avoid starvation. 4375 */ 4376 void ceph_osdc_sync(struct ceph_osd_client *osdc) 4377 { 4378 struct rb_node *n, *p; 4379 u64 last_tid = atomic64_read(&osdc->last_tid); 4380 4381 again: 4382 down_read(&osdc->lock); 4383 for (n = rb_first(&osdc->osds); n; n = rb_next(n)) { 4384 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node); 4385 4386 mutex_lock(&osd->lock); 4387 for (p = rb_first(&osd->o_requests); p; p = rb_next(p)) { 4388 struct ceph_osd_request *req = 4389 rb_entry(p, struct ceph_osd_request, r_node); 4390 4391 if (req->r_tid > last_tid) 4392 break; 4393 4394 if (!(req->r_flags & CEPH_OSD_FLAG_WRITE)) 4395 continue; 4396 4397 ceph_osdc_get_request(req); 4398 mutex_unlock(&osd->lock); 4399 up_read(&osdc->lock); 4400 dout("%s waiting on req %p tid %llu last_tid %llu\n", 4401 __func__, req, req->r_tid, last_tid); 4402 wait_for_completion(&req->r_completion); 4403 ceph_osdc_put_request(req); 4404 goto again; 4405 } 4406 4407 mutex_unlock(&osd->lock); 4408 } 4409 4410 up_read(&osdc->lock); 4411 dout("%s done last_tid %llu\n", __func__, last_tid); 4412 } 4413 EXPORT_SYMBOL(ceph_osdc_sync); 4414 4415 static struct ceph_osd_request * 4416 alloc_linger_request(struct ceph_osd_linger_request *lreq) 4417 { 4418 struct ceph_osd_request *req; 4419 4420 req = ceph_osdc_alloc_request(lreq->osdc, NULL, 1, false, GFP_NOIO); 4421 if (!req) 4422 return NULL; 4423 4424 ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid); 4425 ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc); 4426 4427 if (ceph_osdc_alloc_messages(req, GFP_NOIO)) { 4428 ceph_osdc_put_request(req); 4429 return NULL; 4430 } 4431 4432 return req; 4433 } 4434 4435 /* 4436 * Returns a handle, caller owns a ref. 4437 */ 4438 struct ceph_osd_linger_request * 4439 ceph_osdc_watch(struct ceph_osd_client *osdc, 4440 struct ceph_object_id *oid, 4441 struct ceph_object_locator *oloc, 4442 rados_watchcb2_t wcb, 4443 rados_watcherrcb_t errcb, 4444 void *data) 4445 { 4446 struct ceph_osd_linger_request *lreq; 4447 int ret; 4448 4449 lreq = linger_alloc(osdc); 4450 if (!lreq) 4451 return ERR_PTR(-ENOMEM); 4452 4453 lreq->is_watch = true; 4454 lreq->wcb = wcb; 4455 lreq->errcb = errcb; 4456 lreq->data = data; 4457 lreq->watch_valid_thru = jiffies; 4458 4459 ceph_oid_copy(&lreq->t.base_oid, oid); 4460 ceph_oloc_copy(&lreq->t.base_oloc, oloc); 4461 lreq->t.flags = CEPH_OSD_FLAG_WRITE; 4462 ktime_get_real_ts(&lreq->mtime); 4463 4464 lreq->reg_req = alloc_linger_request(lreq); 4465 if (!lreq->reg_req) { 4466 ret = -ENOMEM; 4467 goto err_put_lreq; 4468 } 4469 4470 lreq->ping_req = alloc_linger_request(lreq); 4471 if (!lreq->ping_req) { 4472 ret = -ENOMEM; 4473 goto err_put_lreq; 4474 } 4475 4476 down_write(&osdc->lock); 4477 linger_register(lreq); /* before osd_req_op_* */ 4478 osd_req_op_watch_init(lreq->reg_req, 0, lreq->linger_id, 4479 CEPH_OSD_WATCH_OP_WATCH); 4480 osd_req_op_watch_init(lreq->ping_req, 0, lreq->linger_id, 4481 CEPH_OSD_WATCH_OP_PING); 4482 linger_submit(lreq); 4483 up_write(&osdc->lock); 4484 4485 ret = linger_reg_commit_wait(lreq); 4486 if (ret) { 4487 linger_cancel(lreq); 4488 goto err_put_lreq; 4489 } 4490 4491 return lreq; 4492 4493 err_put_lreq: 4494 linger_put(lreq); 4495 return ERR_PTR(ret); 4496 } 4497 EXPORT_SYMBOL(ceph_osdc_watch); 4498 4499 /* 4500 * Releases a ref. 4501 * 4502 * Times out after mount_timeout to preserve rbd unmap behaviour 4503 * introduced in 2894e1d76974 ("rbd: timeout watch teardown on unmap 4504 * with mount_timeout"). 4505 */ 4506 int ceph_osdc_unwatch(struct ceph_osd_client *osdc, 4507 struct ceph_osd_linger_request *lreq) 4508 { 4509 struct ceph_options *opts = osdc->client->options; 4510 struct ceph_osd_request *req; 4511 int ret; 4512 4513 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO); 4514 if (!req) 4515 return -ENOMEM; 4516 4517 ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid); 4518 ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc); 4519 req->r_flags = CEPH_OSD_FLAG_WRITE; 4520 ktime_get_real_ts(&req->r_mtime); 4521 osd_req_op_watch_init(req, 0, lreq->linger_id, 4522 CEPH_OSD_WATCH_OP_UNWATCH); 4523 4524 ret = ceph_osdc_alloc_messages(req, GFP_NOIO); 4525 if (ret) 4526 goto out_put_req; 4527 4528 ceph_osdc_start_request(osdc, req, false); 4529 linger_cancel(lreq); 4530 linger_put(lreq); 4531 ret = wait_request_timeout(req, opts->mount_timeout); 4532 4533 out_put_req: 4534 ceph_osdc_put_request(req); 4535 return ret; 4536 } 4537 EXPORT_SYMBOL(ceph_osdc_unwatch); 4538 4539 static int osd_req_op_notify_ack_init(struct ceph_osd_request *req, int which, 4540 u64 notify_id, u64 cookie, void *payload, 4541 size_t payload_len) 4542 { 4543 struct ceph_osd_req_op *op; 4544 struct ceph_pagelist *pl; 4545 int ret; 4546 4547 op = _osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY_ACK, 0); 4548 4549 pl = kmalloc(sizeof(*pl), GFP_NOIO); 4550 if (!pl) 4551 return -ENOMEM; 4552 4553 ceph_pagelist_init(pl); 4554 ret = ceph_pagelist_encode_64(pl, notify_id); 4555 ret |= ceph_pagelist_encode_64(pl, cookie); 4556 if (payload) { 4557 ret |= ceph_pagelist_encode_32(pl, payload_len); 4558 ret |= ceph_pagelist_append(pl, payload, payload_len); 4559 } else { 4560 ret |= ceph_pagelist_encode_32(pl, 0); 4561 } 4562 if (ret) { 4563 ceph_pagelist_release(pl); 4564 return -ENOMEM; 4565 } 4566 4567 ceph_osd_data_pagelist_init(&op->notify_ack.request_data, pl); 4568 op->indata_len = pl->length; 4569 return 0; 4570 } 4571 4572 int ceph_osdc_notify_ack(struct ceph_osd_client *osdc, 4573 struct ceph_object_id *oid, 4574 struct ceph_object_locator *oloc, 4575 u64 notify_id, 4576 u64 cookie, 4577 void *payload, 4578 size_t payload_len) 4579 { 4580 struct ceph_osd_request *req; 4581 int ret; 4582 4583 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO); 4584 if (!req) 4585 return -ENOMEM; 4586 4587 ceph_oid_copy(&req->r_base_oid, oid); 4588 ceph_oloc_copy(&req->r_base_oloc, oloc); 4589 req->r_flags = CEPH_OSD_FLAG_READ; 4590 4591 ret = ceph_osdc_alloc_messages(req, GFP_NOIO); 4592 if (ret) 4593 goto out_put_req; 4594 4595 ret = osd_req_op_notify_ack_init(req, 0, notify_id, cookie, payload, 4596 payload_len); 4597 if (ret) 4598 goto out_put_req; 4599 4600 ceph_osdc_start_request(osdc, req, false); 4601 ret = ceph_osdc_wait_request(osdc, req); 4602 4603 out_put_req: 4604 ceph_osdc_put_request(req); 4605 return ret; 4606 } 4607 EXPORT_SYMBOL(ceph_osdc_notify_ack); 4608 4609 static int osd_req_op_notify_init(struct ceph_osd_request *req, int which, 4610 u64 cookie, u32 prot_ver, u32 timeout, 4611 void *payload, size_t payload_len) 4612 { 4613 struct ceph_osd_req_op *op; 4614 struct ceph_pagelist *pl; 4615 int ret; 4616 4617 op = _osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY, 0); 4618 op->notify.cookie = cookie; 4619 4620 pl = kmalloc(sizeof(*pl), GFP_NOIO); 4621 if (!pl) 4622 return -ENOMEM; 4623 4624 ceph_pagelist_init(pl); 4625 ret = ceph_pagelist_encode_32(pl, 1); /* prot_ver */ 4626 ret |= ceph_pagelist_encode_32(pl, timeout); 4627 ret |= ceph_pagelist_encode_32(pl, payload_len); 4628 ret |= ceph_pagelist_append(pl, payload, payload_len); 4629 if (ret) { 4630 ceph_pagelist_release(pl); 4631 return -ENOMEM; 4632 } 4633 4634 ceph_osd_data_pagelist_init(&op->notify.request_data, pl); 4635 op->indata_len = pl->length; 4636 return 0; 4637 } 4638 4639 /* 4640 * @timeout: in seconds 4641 * 4642 * @preply_{pages,len} are initialized both on success and error. 4643 * The caller is responsible for: 4644 * 4645 * ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len)) 4646 */ 4647 int ceph_osdc_notify(struct ceph_osd_client *osdc, 4648 struct ceph_object_id *oid, 4649 struct ceph_object_locator *oloc, 4650 void *payload, 4651 size_t payload_len, 4652 u32 timeout, 4653 struct page ***preply_pages, 4654 size_t *preply_len) 4655 { 4656 struct ceph_osd_linger_request *lreq; 4657 struct page **pages; 4658 int ret; 4659 4660 WARN_ON(!timeout); 4661 if (preply_pages) { 4662 *preply_pages = NULL; 4663 *preply_len = 0; 4664 } 4665 4666 lreq = linger_alloc(osdc); 4667 if (!lreq) 4668 return -ENOMEM; 4669 4670 lreq->preply_pages = preply_pages; 4671 lreq->preply_len = preply_len; 4672 4673 ceph_oid_copy(&lreq->t.base_oid, oid); 4674 ceph_oloc_copy(&lreq->t.base_oloc, oloc); 4675 lreq->t.flags = CEPH_OSD_FLAG_READ; 4676 4677 lreq->reg_req = alloc_linger_request(lreq); 4678 if (!lreq->reg_req) { 4679 ret = -ENOMEM; 4680 goto out_put_lreq; 4681 } 4682 4683 /* for notify_id */ 4684 pages = ceph_alloc_page_vector(1, GFP_NOIO); 4685 if (IS_ERR(pages)) { 4686 ret = PTR_ERR(pages); 4687 goto out_put_lreq; 4688 } 4689 4690 down_write(&osdc->lock); 4691 linger_register(lreq); /* before osd_req_op_* */ 4692 ret = osd_req_op_notify_init(lreq->reg_req, 0, lreq->linger_id, 1, 4693 timeout, payload, payload_len); 4694 if (ret) { 4695 linger_unregister(lreq); 4696 up_write(&osdc->lock); 4697 ceph_release_page_vector(pages, 1); 4698 goto out_put_lreq; 4699 } 4700 ceph_osd_data_pages_init(osd_req_op_data(lreq->reg_req, 0, notify, 4701 response_data), 4702 pages, PAGE_SIZE, 0, false, true); 4703 linger_submit(lreq); 4704 up_write(&osdc->lock); 4705 4706 ret = linger_reg_commit_wait(lreq); 4707 if (!ret) 4708 ret = linger_notify_finish_wait(lreq); 4709 else 4710 dout("lreq %p failed to initiate notify %d\n", lreq, ret); 4711 4712 linger_cancel(lreq); 4713 out_put_lreq: 4714 linger_put(lreq); 4715 return ret; 4716 } 4717 EXPORT_SYMBOL(ceph_osdc_notify); 4718 4719 /* 4720 * Return the number of milliseconds since the watch was last 4721 * confirmed, or an error. If there is an error, the watch is no 4722 * longer valid, and should be destroyed with ceph_osdc_unwatch(). 4723 */ 4724 int ceph_osdc_watch_check(struct ceph_osd_client *osdc, 4725 struct ceph_osd_linger_request *lreq) 4726 { 4727 unsigned long stamp, age; 4728 int ret; 4729 4730 down_read(&osdc->lock); 4731 mutex_lock(&lreq->lock); 4732 stamp = lreq->watch_valid_thru; 4733 if (!list_empty(&lreq->pending_lworks)) { 4734 struct linger_work *lwork = 4735 list_first_entry(&lreq->pending_lworks, 4736 struct linger_work, 4737 pending_item); 4738 4739 if (time_before(lwork->queued_stamp, stamp)) 4740 stamp = lwork->queued_stamp; 4741 } 4742 age = jiffies - stamp; 4743 dout("%s lreq %p linger_id %llu age %lu last_error %d\n", __func__, 4744 lreq, lreq->linger_id, age, lreq->last_error); 4745 /* we are truncating to msecs, so return a safe upper bound */ 4746 ret = lreq->last_error ?: 1 + jiffies_to_msecs(age); 4747 4748 mutex_unlock(&lreq->lock); 4749 up_read(&osdc->lock); 4750 return ret; 4751 } 4752 4753 static int decode_watcher(void **p, void *end, struct ceph_watch_item *item) 4754 { 4755 u8 struct_v; 4756 u32 struct_len; 4757 int ret; 4758 4759 ret = ceph_start_decoding(p, end, 2, "watch_item_t", 4760 &struct_v, &struct_len); 4761 if (ret) 4762 return ret; 4763 4764 ceph_decode_copy(p, &item->name, sizeof(item->name)); 4765 item->cookie = ceph_decode_64(p); 4766 *p += 4; /* skip timeout_seconds */ 4767 if (struct_v >= 2) { 4768 ceph_decode_copy(p, &item->addr, sizeof(item->addr)); 4769 ceph_decode_addr(&item->addr); 4770 } 4771 4772 dout("%s %s%llu cookie %llu addr %s\n", __func__, 4773 ENTITY_NAME(item->name), item->cookie, 4774 ceph_pr_addr(&item->addr.in_addr)); 4775 return 0; 4776 } 4777 4778 static int decode_watchers(void **p, void *end, 4779 struct ceph_watch_item **watchers, 4780 u32 *num_watchers) 4781 { 4782 u8 struct_v; 4783 u32 struct_len; 4784 int i; 4785 int ret; 4786 4787 ret = ceph_start_decoding(p, end, 1, "obj_list_watch_response_t", 4788 &struct_v, &struct_len); 4789 if (ret) 4790 return ret; 4791 4792 *num_watchers = ceph_decode_32(p); 4793 *watchers = kcalloc(*num_watchers, sizeof(**watchers), GFP_NOIO); 4794 if (!*watchers) 4795 return -ENOMEM; 4796 4797 for (i = 0; i < *num_watchers; i++) { 4798 ret = decode_watcher(p, end, *watchers + i); 4799 if (ret) { 4800 kfree(*watchers); 4801 return ret; 4802 } 4803 } 4804 4805 return 0; 4806 } 4807 4808 /* 4809 * On success, the caller is responsible for: 4810 * 4811 * kfree(watchers); 4812 */ 4813 int ceph_osdc_list_watchers(struct ceph_osd_client *osdc, 4814 struct ceph_object_id *oid, 4815 struct ceph_object_locator *oloc, 4816 struct ceph_watch_item **watchers, 4817 u32 *num_watchers) 4818 { 4819 struct ceph_osd_request *req; 4820 struct page **pages; 4821 int ret; 4822 4823 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO); 4824 if (!req) 4825 return -ENOMEM; 4826 4827 ceph_oid_copy(&req->r_base_oid, oid); 4828 ceph_oloc_copy(&req->r_base_oloc, oloc); 4829 req->r_flags = CEPH_OSD_FLAG_READ; 4830 4831 ret = ceph_osdc_alloc_messages(req, GFP_NOIO); 4832 if (ret) 4833 goto out_put_req; 4834 4835 pages = ceph_alloc_page_vector(1, GFP_NOIO); 4836 if (IS_ERR(pages)) { 4837 ret = PTR_ERR(pages); 4838 goto out_put_req; 4839 } 4840 4841 osd_req_op_init(req, 0, CEPH_OSD_OP_LIST_WATCHERS, 0); 4842 ceph_osd_data_pages_init(osd_req_op_data(req, 0, list_watchers, 4843 response_data), 4844 pages, PAGE_SIZE, 0, false, true); 4845 4846 ceph_osdc_start_request(osdc, req, false); 4847 ret = ceph_osdc_wait_request(osdc, req); 4848 if (ret >= 0) { 4849 void *p = page_address(pages[0]); 4850 void *const end = p + req->r_ops[0].outdata_len; 4851 4852 ret = decode_watchers(&p, end, watchers, num_watchers); 4853 } 4854 4855 out_put_req: 4856 ceph_osdc_put_request(req); 4857 return ret; 4858 } 4859 EXPORT_SYMBOL(ceph_osdc_list_watchers); 4860 4861 /* 4862 * Call all pending notify callbacks - for use after a watch is 4863 * unregistered, to make sure no more callbacks for it will be invoked 4864 */ 4865 void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc) 4866 { 4867 dout("%s osdc %p\n", __func__, osdc); 4868 flush_workqueue(osdc->notify_wq); 4869 } 4870 EXPORT_SYMBOL(ceph_osdc_flush_notifies); 4871 4872 void ceph_osdc_maybe_request_map(struct ceph_osd_client *osdc) 4873 { 4874 down_read(&osdc->lock); 4875 maybe_request_map(osdc); 4876 up_read(&osdc->lock); 4877 } 4878 EXPORT_SYMBOL(ceph_osdc_maybe_request_map); 4879 4880 /* 4881 * Execute an OSD class method on an object. 4882 * 4883 * @flags: CEPH_OSD_FLAG_* 4884 * @resp_len: in/out param for reply length 4885 */ 4886 int ceph_osdc_call(struct ceph_osd_client *osdc, 4887 struct ceph_object_id *oid, 4888 struct ceph_object_locator *oloc, 4889 const char *class, const char *method, 4890 unsigned int flags, 4891 struct page *req_page, size_t req_len, 4892 struct page *resp_page, size_t *resp_len) 4893 { 4894 struct ceph_osd_request *req; 4895 int ret; 4896 4897 if (req_len > PAGE_SIZE || (resp_page && *resp_len > PAGE_SIZE)) 4898 return -E2BIG; 4899 4900 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO); 4901 if (!req) 4902 return -ENOMEM; 4903 4904 ceph_oid_copy(&req->r_base_oid, oid); 4905 ceph_oloc_copy(&req->r_base_oloc, oloc); 4906 req->r_flags = flags; 4907 4908 ret = ceph_osdc_alloc_messages(req, GFP_NOIO); 4909 if (ret) 4910 goto out_put_req; 4911 4912 osd_req_op_cls_init(req, 0, CEPH_OSD_OP_CALL, class, method); 4913 if (req_page) 4914 osd_req_op_cls_request_data_pages(req, 0, &req_page, req_len, 4915 0, false, false); 4916 if (resp_page) 4917 osd_req_op_cls_response_data_pages(req, 0, &resp_page, 4918 *resp_len, 0, false, false); 4919 4920 ceph_osdc_start_request(osdc, req, false); 4921 ret = ceph_osdc_wait_request(osdc, req); 4922 if (ret >= 0) { 4923 ret = req->r_ops[0].rval; 4924 if (resp_page) 4925 *resp_len = req->r_ops[0].outdata_len; 4926 } 4927 4928 out_put_req: 4929 ceph_osdc_put_request(req); 4930 return ret; 4931 } 4932 EXPORT_SYMBOL(ceph_osdc_call); 4933 4934 /* 4935 * init, shutdown 4936 */ 4937 int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client) 4938 { 4939 int err; 4940 4941 dout("init\n"); 4942 osdc->client = client; 4943 init_rwsem(&osdc->lock); 4944 osdc->osds = RB_ROOT; 4945 INIT_LIST_HEAD(&osdc->osd_lru); 4946 spin_lock_init(&osdc->osd_lru_lock); 4947 osd_init(&osdc->homeless_osd); 4948 osdc->homeless_osd.o_osdc = osdc; 4949 osdc->homeless_osd.o_osd = CEPH_HOMELESS_OSD; 4950 osdc->last_linger_id = CEPH_LINGER_ID_START; 4951 osdc->linger_requests = RB_ROOT; 4952 osdc->map_checks = RB_ROOT; 4953 osdc->linger_map_checks = RB_ROOT; 4954 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout); 4955 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout); 4956 4957 err = -ENOMEM; 4958 osdc->osdmap = ceph_osdmap_alloc(); 4959 if (!osdc->osdmap) 4960 goto out; 4961 4962 osdc->req_mempool = mempool_create_slab_pool(10, 4963 ceph_osd_request_cache); 4964 if (!osdc->req_mempool) 4965 goto out_map; 4966 4967 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP, 4968 PAGE_SIZE, 10, true, "osd_op"); 4969 if (err < 0) 4970 goto out_mempool; 4971 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY, 4972 PAGE_SIZE, 10, true, "osd_op_reply"); 4973 if (err < 0) 4974 goto out_msgpool; 4975 4976 err = -ENOMEM; 4977 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify"); 4978 if (!osdc->notify_wq) 4979 goto out_msgpool_reply; 4980 4981 schedule_delayed_work(&osdc->timeout_work, 4982 osdc->client->options->osd_keepalive_timeout); 4983 schedule_delayed_work(&osdc->osds_timeout_work, 4984 round_jiffies_relative(osdc->client->options->osd_idle_ttl)); 4985 4986 return 0; 4987 4988 out_msgpool_reply: 4989 ceph_msgpool_destroy(&osdc->msgpool_op_reply); 4990 out_msgpool: 4991 ceph_msgpool_destroy(&osdc->msgpool_op); 4992 out_mempool: 4993 mempool_destroy(osdc->req_mempool); 4994 out_map: 4995 ceph_osdmap_destroy(osdc->osdmap); 4996 out: 4997 return err; 4998 } 4999 5000 void ceph_osdc_stop(struct ceph_osd_client *osdc) 5001 { 5002 flush_workqueue(osdc->notify_wq); 5003 destroy_workqueue(osdc->notify_wq); 5004 cancel_delayed_work_sync(&osdc->timeout_work); 5005 cancel_delayed_work_sync(&osdc->osds_timeout_work); 5006 5007 down_write(&osdc->lock); 5008 while (!RB_EMPTY_ROOT(&osdc->osds)) { 5009 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds), 5010 struct ceph_osd, o_node); 5011 close_osd(osd); 5012 } 5013 up_write(&osdc->lock); 5014 WARN_ON(refcount_read(&osdc->homeless_osd.o_ref) != 1); 5015 osd_cleanup(&osdc->homeless_osd); 5016 5017 WARN_ON(!list_empty(&osdc->osd_lru)); 5018 WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_requests)); 5019 WARN_ON(!RB_EMPTY_ROOT(&osdc->map_checks)); 5020 WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_map_checks)); 5021 WARN_ON(atomic_read(&osdc->num_requests)); 5022 WARN_ON(atomic_read(&osdc->num_homeless)); 5023 5024 ceph_osdmap_destroy(osdc->osdmap); 5025 mempool_destroy(osdc->req_mempool); 5026 ceph_msgpool_destroy(&osdc->msgpool_op); 5027 ceph_msgpool_destroy(&osdc->msgpool_op_reply); 5028 } 5029 5030 /* 5031 * Read some contiguous pages. If we cross a stripe boundary, shorten 5032 * *plen. Return number of bytes read, or error. 5033 */ 5034 int ceph_osdc_readpages(struct ceph_osd_client *osdc, 5035 struct ceph_vino vino, struct ceph_file_layout *layout, 5036 u64 off, u64 *plen, 5037 u32 truncate_seq, u64 truncate_size, 5038 struct page **pages, int num_pages, int page_align) 5039 { 5040 struct ceph_osd_request *req; 5041 int rc = 0; 5042 5043 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino, 5044 vino.snap, off, *plen); 5045 req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 0, 1, 5046 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ, 5047 NULL, truncate_seq, truncate_size, 5048 false); 5049 if (IS_ERR(req)) 5050 return PTR_ERR(req); 5051 5052 /* it may be a short read due to an object boundary */ 5053 osd_req_op_extent_osd_data_pages(req, 0, 5054 pages, *plen, page_align, false, false); 5055 5056 dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n", 5057 off, *plen, *plen, page_align); 5058 5059 rc = ceph_osdc_start_request(osdc, req, false); 5060 if (!rc) 5061 rc = ceph_osdc_wait_request(osdc, req); 5062 5063 ceph_osdc_put_request(req); 5064 dout("readpages result %d\n", rc); 5065 return rc; 5066 } 5067 EXPORT_SYMBOL(ceph_osdc_readpages); 5068 5069 /* 5070 * do a synchronous write on N pages 5071 */ 5072 int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino, 5073 struct ceph_file_layout *layout, 5074 struct ceph_snap_context *snapc, 5075 u64 off, u64 len, 5076 u32 truncate_seq, u64 truncate_size, 5077 struct timespec *mtime, 5078 struct page **pages, int num_pages) 5079 { 5080 struct ceph_osd_request *req; 5081 int rc = 0; 5082 int page_align = off & ~PAGE_MASK; 5083 5084 req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 0, 1, 5085 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE, 5086 snapc, truncate_seq, truncate_size, 5087 true); 5088 if (IS_ERR(req)) 5089 return PTR_ERR(req); 5090 5091 /* it may be a short write due to an object boundary */ 5092 osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align, 5093 false, false); 5094 dout("writepages %llu~%llu (%llu bytes)\n", off, len, len); 5095 5096 req->r_mtime = *mtime; 5097 rc = ceph_osdc_start_request(osdc, req, true); 5098 if (!rc) 5099 rc = ceph_osdc_wait_request(osdc, req); 5100 5101 ceph_osdc_put_request(req); 5102 if (rc == 0) 5103 rc = len; 5104 dout("writepages result %d\n", rc); 5105 return rc; 5106 } 5107 EXPORT_SYMBOL(ceph_osdc_writepages); 5108 5109 int __init ceph_osdc_setup(void) 5110 { 5111 size_t size = sizeof(struct ceph_osd_request) + 5112 CEPH_OSD_SLAB_OPS * sizeof(struct ceph_osd_req_op); 5113 5114 BUG_ON(ceph_osd_request_cache); 5115 ceph_osd_request_cache = kmem_cache_create("ceph_osd_request", size, 5116 0, 0, NULL); 5117 5118 return ceph_osd_request_cache ? 0 : -ENOMEM; 5119 } 5120 5121 void ceph_osdc_cleanup(void) 5122 { 5123 BUG_ON(!ceph_osd_request_cache); 5124 kmem_cache_destroy(ceph_osd_request_cache); 5125 ceph_osd_request_cache = NULL; 5126 } 5127 5128 /* 5129 * handle incoming message 5130 */ 5131 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg) 5132 { 5133 struct ceph_osd *osd = con->private; 5134 struct ceph_osd_client *osdc = osd->o_osdc; 5135 int type = le16_to_cpu(msg->hdr.type); 5136 5137 switch (type) { 5138 case CEPH_MSG_OSD_MAP: 5139 ceph_osdc_handle_map(osdc, msg); 5140 break; 5141 case CEPH_MSG_OSD_OPREPLY: 5142 handle_reply(osd, msg); 5143 break; 5144 case CEPH_MSG_OSD_BACKOFF: 5145 handle_backoff(osd, msg); 5146 break; 5147 case CEPH_MSG_WATCH_NOTIFY: 5148 handle_watch_notify(osdc, msg); 5149 break; 5150 5151 default: 5152 pr_err("received unknown message type %d %s\n", type, 5153 ceph_msg_type_name(type)); 5154 } 5155 5156 ceph_msg_put(msg); 5157 } 5158 5159 /* 5160 * Lookup and return message for incoming reply. Don't try to do 5161 * anything about a larger than preallocated data portion of the 5162 * message at the moment - for now, just skip the message. 5163 */ 5164 static struct ceph_msg *get_reply(struct ceph_connection *con, 5165 struct ceph_msg_header *hdr, 5166 int *skip) 5167 { 5168 struct ceph_osd *osd = con->private; 5169 struct ceph_osd_client *osdc = osd->o_osdc; 5170 struct ceph_msg *m = NULL; 5171 struct ceph_osd_request *req; 5172 int front_len = le32_to_cpu(hdr->front_len); 5173 int data_len = le32_to_cpu(hdr->data_len); 5174 u64 tid = le64_to_cpu(hdr->tid); 5175 5176 down_read(&osdc->lock); 5177 if (!osd_registered(osd)) { 5178 dout("%s osd%d unknown, skipping\n", __func__, osd->o_osd); 5179 *skip = 1; 5180 goto out_unlock_osdc; 5181 } 5182 WARN_ON(osd->o_osd != le64_to_cpu(hdr->src.num)); 5183 5184 mutex_lock(&osd->lock); 5185 req = lookup_request(&osd->o_requests, tid); 5186 if (!req) { 5187 dout("%s osd%d tid %llu unknown, skipping\n", __func__, 5188 osd->o_osd, tid); 5189 *skip = 1; 5190 goto out_unlock_session; 5191 } 5192 5193 ceph_msg_revoke_incoming(req->r_reply); 5194 5195 if (front_len > req->r_reply->front_alloc_len) { 5196 pr_warn("%s osd%d tid %llu front %d > preallocated %d\n", 5197 __func__, osd->o_osd, req->r_tid, front_len, 5198 req->r_reply->front_alloc_len); 5199 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS, 5200 false); 5201 if (!m) 5202 goto out_unlock_session; 5203 ceph_msg_put(req->r_reply); 5204 req->r_reply = m; 5205 } 5206 5207 if (data_len > req->r_reply->data_length) { 5208 pr_warn("%s osd%d tid %llu data %d > preallocated %zu, skipping\n", 5209 __func__, osd->o_osd, req->r_tid, data_len, 5210 req->r_reply->data_length); 5211 m = NULL; 5212 *skip = 1; 5213 goto out_unlock_session; 5214 } 5215 5216 m = ceph_msg_get(req->r_reply); 5217 dout("get_reply tid %lld %p\n", tid, m); 5218 5219 out_unlock_session: 5220 mutex_unlock(&osd->lock); 5221 out_unlock_osdc: 5222 up_read(&osdc->lock); 5223 return m; 5224 } 5225 5226 /* 5227 * TODO: switch to a msg-owned pagelist 5228 */ 5229 static struct ceph_msg *alloc_msg_with_page_vector(struct ceph_msg_header *hdr) 5230 { 5231 struct ceph_msg *m; 5232 int type = le16_to_cpu(hdr->type); 5233 u32 front_len = le32_to_cpu(hdr->front_len); 5234 u32 data_len = le32_to_cpu(hdr->data_len); 5235 5236 m = ceph_msg_new(type, front_len, GFP_NOIO, false); 5237 if (!m) 5238 return NULL; 5239 5240 if (data_len) { 5241 struct page **pages; 5242 struct ceph_osd_data osd_data; 5243 5244 pages = ceph_alloc_page_vector(calc_pages_for(0, data_len), 5245 GFP_NOIO); 5246 if (IS_ERR(pages)) { 5247 ceph_msg_put(m); 5248 return NULL; 5249 } 5250 5251 ceph_osd_data_pages_init(&osd_data, pages, data_len, 0, false, 5252 false); 5253 ceph_osdc_msg_data_add(m, &osd_data); 5254 } 5255 5256 return m; 5257 } 5258 5259 static struct ceph_msg *alloc_msg(struct ceph_connection *con, 5260 struct ceph_msg_header *hdr, 5261 int *skip) 5262 { 5263 struct ceph_osd *osd = con->private; 5264 int type = le16_to_cpu(hdr->type); 5265 5266 *skip = 0; 5267 switch (type) { 5268 case CEPH_MSG_OSD_MAP: 5269 case CEPH_MSG_OSD_BACKOFF: 5270 case CEPH_MSG_WATCH_NOTIFY: 5271 return alloc_msg_with_page_vector(hdr); 5272 case CEPH_MSG_OSD_OPREPLY: 5273 return get_reply(con, hdr, skip); 5274 default: 5275 pr_warn("%s osd%d unknown msg type %d, skipping\n", __func__, 5276 osd->o_osd, type); 5277 *skip = 1; 5278 return NULL; 5279 } 5280 } 5281 5282 /* 5283 * Wrappers to refcount containing ceph_osd struct 5284 */ 5285 static struct ceph_connection *get_osd_con(struct ceph_connection *con) 5286 { 5287 struct ceph_osd *osd = con->private; 5288 if (get_osd(osd)) 5289 return con; 5290 return NULL; 5291 } 5292 5293 static void put_osd_con(struct ceph_connection *con) 5294 { 5295 struct ceph_osd *osd = con->private; 5296 put_osd(osd); 5297 } 5298 5299 /* 5300 * authentication 5301 */ 5302 /* 5303 * Note: returned pointer is the address of a structure that's 5304 * managed separately. Caller must *not* attempt to free it. 5305 */ 5306 static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con, 5307 int *proto, int force_new) 5308 { 5309 struct ceph_osd *o = con->private; 5310 struct ceph_osd_client *osdc = o->o_osdc; 5311 struct ceph_auth_client *ac = osdc->client->monc.auth; 5312 struct ceph_auth_handshake *auth = &o->o_auth; 5313 5314 if (force_new && auth->authorizer) { 5315 ceph_auth_destroy_authorizer(auth->authorizer); 5316 auth->authorizer = NULL; 5317 } 5318 if (!auth->authorizer) { 5319 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD, 5320 auth); 5321 if (ret) 5322 return ERR_PTR(ret); 5323 } else { 5324 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD, 5325 auth); 5326 if (ret) 5327 return ERR_PTR(ret); 5328 } 5329 *proto = ac->protocol; 5330 5331 return auth; 5332 } 5333 5334 5335 static int verify_authorizer_reply(struct ceph_connection *con) 5336 { 5337 struct ceph_osd *o = con->private; 5338 struct ceph_osd_client *osdc = o->o_osdc; 5339 struct ceph_auth_client *ac = osdc->client->monc.auth; 5340 5341 return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer); 5342 } 5343 5344 static int invalidate_authorizer(struct ceph_connection *con) 5345 { 5346 struct ceph_osd *o = con->private; 5347 struct ceph_osd_client *osdc = o->o_osdc; 5348 struct ceph_auth_client *ac = osdc->client->monc.auth; 5349 5350 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD); 5351 return ceph_monc_validate_auth(&osdc->client->monc); 5352 } 5353 5354 static void osd_reencode_message(struct ceph_msg *msg) 5355 { 5356 int type = le16_to_cpu(msg->hdr.type); 5357 5358 if (type == CEPH_MSG_OSD_OP) 5359 encode_request_finish(msg); 5360 } 5361 5362 static int osd_sign_message(struct ceph_msg *msg) 5363 { 5364 struct ceph_osd *o = msg->con->private; 5365 struct ceph_auth_handshake *auth = &o->o_auth; 5366 5367 return ceph_auth_sign_message(auth, msg); 5368 } 5369 5370 static int osd_check_message_signature(struct ceph_msg *msg) 5371 { 5372 struct ceph_osd *o = msg->con->private; 5373 struct ceph_auth_handshake *auth = &o->o_auth; 5374 5375 return ceph_auth_check_message_signature(auth, msg); 5376 } 5377 5378 static const struct ceph_connection_operations osd_con_ops = { 5379 .get = get_osd_con, 5380 .put = put_osd_con, 5381 .dispatch = dispatch, 5382 .get_authorizer = get_authorizer, 5383 .verify_authorizer_reply = verify_authorizer_reply, 5384 .invalidate_authorizer = invalidate_authorizer, 5385 .alloc_msg = alloc_msg, 5386 .reencode_message = osd_reencode_message, 5387 .sign_message = osd_sign_message, 5388 .check_message_signature = osd_check_message_signature, 5389 .fault = osd_fault, 5390 }; 5391