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 /* 130 * Consumes @pages if @own_pages is true. 131 */ 132 static void ceph_osd_data_pages_init(struct ceph_osd_data *osd_data, 133 struct page **pages, u64 length, u32 alignment, 134 bool pages_from_pool, bool own_pages) 135 { 136 osd_data->type = CEPH_OSD_DATA_TYPE_PAGES; 137 osd_data->pages = pages; 138 osd_data->length = length; 139 osd_data->alignment = alignment; 140 osd_data->pages_from_pool = pages_from_pool; 141 osd_data->own_pages = own_pages; 142 } 143 144 /* 145 * Consumes a ref on @pagelist. 146 */ 147 static void ceph_osd_data_pagelist_init(struct ceph_osd_data *osd_data, 148 struct ceph_pagelist *pagelist) 149 { 150 osd_data->type = CEPH_OSD_DATA_TYPE_PAGELIST; 151 osd_data->pagelist = pagelist; 152 } 153 154 #ifdef CONFIG_BLOCK 155 static void ceph_osd_data_bio_init(struct ceph_osd_data *osd_data, 156 struct ceph_bio_iter *bio_pos, 157 u32 bio_length) 158 { 159 osd_data->type = CEPH_OSD_DATA_TYPE_BIO; 160 osd_data->bio_pos = *bio_pos; 161 osd_data->bio_length = bio_length; 162 } 163 #endif /* CONFIG_BLOCK */ 164 165 static void ceph_osd_data_bvecs_init(struct ceph_osd_data *osd_data, 166 struct ceph_bvec_iter *bvec_pos, 167 u32 num_bvecs) 168 { 169 osd_data->type = CEPH_OSD_DATA_TYPE_BVECS; 170 osd_data->bvec_pos = *bvec_pos; 171 osd_data->num_bvecs = num_bvecs; 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_bvecs(struct ceph_osd_request *osd_req, 241 unsigned int which, 242 struct bio_vec *bvecs, u32 num_bvecs, 243 u32 bytes) 244 { 245 struct ceph_osd_data *osd_data; 246 struct ceph_bvec_iter it = { 247 .bvecs = bvecs, 248 .iter = { .bi_size = bytes }, 249 }; 250 251 osd_data = osd_req_op_data(osd_req, which, extent, osd_data); 252 ceph_osd_data_bvecs_init(osd_data, &it, num_bvecs); 253 } 254 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bvecs); 255 256 void osd_req_op_extent_osd_data_bvec_pos(struct ceph_osd_request *osd_req, 257 unsigned int which, 258 struct ceph_bvec_iter *bvec_pos) 259 { 260 struct ceph_osd_data *osd_data; 261 262 osd_data = osd_req_op_data(osd_req, which, extent, osd_data); 263 ceph_osd_data_bvecs_init(osd_data, bvec_pos, 0); 264 } 265 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bvec_pos); 266 267 static void osd_req_op_cls_request_info_pagelist( 268 struct ceph_osd_request *osd_req, 269 unsigned int which, struct ceph_pagelist *pagelist) 270 { 271 struct ceph_osd_data *osd_data; 272 273 osd_data = osd_req_op_data(osd_req, which, cls, request_info); 274 ceph_osd_data_pagelist_init(osd_data, pagelist); 275 } 276 277 void osd_req_op_cls_request_data_pagelist( 278 struct ceph_osd_request *osd_req, 279 unsigned int which, struct ceph_pagelist *pagelist) 280 { 281 struct ceph_osd_data *osd_data; 282 283 osd_data = osd_req_op_data(osd_req, which, cls, request_data); 284 ceph_osd_data_pagelist_init(osd_data, pagelist); 285 osd_req->r_ops[which].cls.indata_len += pagelist->length; 286 osd_req->r_ops[which].indata_len += pagelist->length; 287 } 288 EXPORT_SYMBOL(osd_req_op_cls_request_data_pagelist); 289 290 void osd_req_op_cls_request_data_pages(struct ceph_osd_request *osd_req, 291 unsigned int which, struct page **pages, u64 length, 292 u32 alignment, bool pages_from_pool, bool own_pages) 293 { 294 struct ceph_osd_data *osd_data; 295 296 osd_data = osd_req_op_data(osd_req, which, cls, request_data); 297 ceph_osd_data_pages_init(osd_data, pages, length, alignment, 298 pages_from_pool, own_pages); 299 osd_req->r_ops[which].cls.indata_len += length; 300 osd_req->r_ops[which].indata_len += length; 301 } 302 EXPORT_SYMBOL(osd_req_op_cls_request_data_pages); 303 304 void osd_req_op_cls_request_data_bvecs(struct ceph_osd_request *osd_req, 305 unsigned int which, 306 struct bio_vec *bvecs, u32 num_bvecs, 307 u32 bytes) 308 { 309 struct ceph_osd_data *osd_data; 310 struct ceph_bvec_iter it = { 311 .bvecs = bvecs, 312 .iter = { .bi_size = bytes }, 313 }; 314 315 osd_data = osd_req_op_data(osd_req, which, cls, request_data); 316 ceph_osd_data_bvecs_init(osd_data, &it, num_bvecs); 317 osd_req->r_ops[which].cls.indata_len += bytes; 318 osd_req->r_ops[which].indata_len += bytes; 319 } 320 EXPORT_SYMBOL(osd_req_op_cls_request_data_bvecs); 321 322 void osd_req_op_cls_response_data_pages(struct ceph_osd_request *osd_req, 323 unsigned int which, struct page **pages, u64 length, 324 u32 alignment, bool pages_from_pool, bool own_pages) 325 { 326 struct ceph_osd_data *osd_data; 327 328 osd_data = osd_req_op_data(osd_req, which, cls, response_data); 329 ceph_osd_data_pages_init(osd_data, pages, length, alignment, 330 pages_from_pool, own_pages); 331 } 332 EXPORT_SYMBOL(osd_req_op_cls_response_data_pages); 333 334 static u64 ceph_osd_data_length(struct ceph_osd_data *osd_data) 335 { 336 switch (osd_data->type) { 337 case CEPH_OSD_DATA_TYPE_NONE: 338 return 0; 339 case CEPH_OSD_DATA_TYPE_PAGES: 340 return osd_data->length; 341 case CEPH_OSD_DATA_TYPE_PAGELIST: 342 return (u64)osd_data->pagelist->length; 343 #ifdef CONFIG_BLOCK 344 case CEPH_OSD_DATA_TYPE_BIO: 345 return (u64)osd_data->bio_length; 346 #endif /* CONFIG_BLOCK */ 347 case CEPH_OSD_DATA_TYPE_BVECS: 348 return osd_data->bvec_pos.iter.bi_size; 349 default: 350 WARN(true, "unrecognized data type %d\n", (int)osd_data->type); 351 return 0; 352 } 353 } 354 355 static void ceph_osd_data_release(struct ceph_osd_data *osd_data) 356 { 357 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES && osd_data->own_pages) { 358 int num_pages; 359 360 num_pages = calc_pages_for((u64)osd_data->alignment, 361 (u64)osd_data->length); 362 ceph_release_page_vector(osd_data->pages, num_pages); 363 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) { 364 ceph_pagelist_release(osd_data->pagelist); 365 } 366 ceph_osd_data_init(osd_data); 367 } 368 369 static void osd_req_op_data_release(struct ceph_osd_request *osd_req, 370 unsigned int which) 371 { 372 struct ceph_osd_req_op *op; 373 374 BUG_ON(which >= osd_req->r_num_ops); 375 op = &osd_req->r_ops[which]; 376 377 switch (op->op) { 378 case CEPH_OSD_OP_READ: 379 case CEPH_OSD_OP_WRITE: 380 case CEPH_OSD_OP_WRITEFULL: 381 ceph_osd_data_release(&op->extent.osd_data); 382 break; 383 case CEPH_OSD_OP_CALL: 384 ceph_osd_data_release(&op->cls.request_info); 385 ceph_osd_data_release(&op->cls.request_data); 386 ceph_osd_data_release(&op->cls.response_data); 387 break; 388 case CEPH_OSD_OP_SETXATTR: 389 case CEPH_OSD_OP_CMPXATTR: 390 ceph_osd_data_release(&op->xattr.osd_data); 391 break; 392 case CEPH_OSD_OP_STAT: 393 ceph_osd_data_release(&op->raw_data_in); 394 break; 395 case CEPH_OSD_OP_NOTIFY_ACK: 396 ceph_osd_data_release(&op->notify_ack.request_data); 397 break; 398 case CEPH_OSD_OP_NOTIFY: 399 ceph_osd_data_release(&op->notify.request_data); 400 ceph_osd_data_release(&op->notify.response_data); 401 break; 402 case CEPH_OSD_OP_LIST_WATCHERS: 403 ceph_osd_data_release(&op->list_watchers.response_data); 404 break; 405 case CEPH_OSD_OP_COPY_FROM2: 406 ceph_osd_data_release(&op->copy_from.osd_data); 407 break; 408 default: 409 break; 410 } 411 } 412 413 /* 414 * Assumes @t is zero-initialized. 415 */ 416 static void target_init(struct ceph_osd_request_target *t) 417 { 418 ceph_oid_init(&t->base_oid); 419 ceph_oloc_init(&t->base_oloc); 420 ceph_oid_init(&t->target_oid); 421 ceph_oloc_init(&t->target_oloc); 422 423 ceph_osds_init(&t->acting); 424 ceph_osds_init(&t->up); 425 t->size = -1; 426 t->min_size = -1; 427 428 t->osd = CEPH_HOMELESS_OSD; 429 } 430 431 static void target_copy(struct ceph_osd_request_target *dest, 432 const struct ceph_osd_request_target *src) 433 { 434 ceph_oid_copy(&dest->base_oid, &src->base_oid); 435 ceph_oloc_copy(&dest->base_oloc, &src->base_oloc); 436 ceph_oid_copy(&dest->target_oid, &src->target_oid); 437 ceph_oloc_copy(&dest->target_oloc, &src->target_oloc); 438 439 dest->pgid = src->pgid; /* struct */ 440 dest->spgid = src->spgid; /* struct */ 441 dest->pg_num = src->pg_num; 442 dest->pg_num_mask = src->pg_num_mask; 443 ceph_osds_copy(&dest->acting, &src->acting); 444 ceph_osds_copy(&dest->up, &src->up); 445 dest->size = src->size; 446 dest->min_size = src->min_size; 447 dest->sort_bitwise = src->sort_bitwise; 448 dest->recovery_deletes = src->recovery_deletes; 449 450 dest->flags = src->flags; 451 dest->used_replica = src->used_replica; 452 dest->paused = src->paused; 453 454 dest->epoch = src->epoch; 455 dest->last_force_resend = src->last_force_resend; 456 457 dest->osd = src->osd; 458 } 459 460 static void target_destroy(struct ceph_osd_request_target *t) 461 { 462 ceph_oid_destroy(&t->base_oid); 463 ceph_oloc_destroy(&t->base_oloc); 464 ceph_oid_destroy(&t->target_oid); 465 ceph_oloc_destroy(&t->target_oloc); 466 } 467 468 /* 469 * requests 470 */ 471 static void request_release_checks(struct ceph_osd_request *req) 472 { 473 WARN_ON(!RB_EMPTY_NODE(&req->r_node)); 474 WARN_ON(!RB_EMPTY_NODE(&req->r_mc_node)); 475 WARN_ON(!list_empty(&req->r_private_item)); 476 WARN_ON(req->r_osd); 477 } 478 479 static void ceph_osdc_release_request(struct kref *kref) 480 { 481 struct ceph_osd_request *req = container_of(kref, 482 struct ceph_osd_request, r_kref); 483 unsigned int which; 484 485 dout("%s %p (r_request %p r_reply %p)\n", __func__, req, 486 req->r_request, req->r_reply); 487 request_release_checks(req); 488 489 if (req->r_request) 490 ceph_msg_put(req->r_request); 491 if (req->r_reply) 492 ceph_msg_put(req->r_reply); 493 494 for (which = 0; which < req->r_num_ops; which++) 495 osd_req_op_data_release(req, which); 496 497 target_destroy(&req->r_t); 498 ceph_put_snap_context(req->r_snapc); 499 500 if (req->r_mempool) 501 mempool_free(req, req->r_osdc->req_mempool); 502 else if (req->r_num_ops <= CEPH_OSD_SLAB_OPS) 503 kmem_cache_free(ceph_osd_request_cache, req); 504 else 505 kfree(req); 506 } 507 508 void ceph_osdc_get_request(struct ceph_osd_request *req) 509 { 510 dout("%s %p (was %d)\n", __func__, req, 511 kref_read(&req->r_kref)); 512 kref_get(&req->r_kref); 513 } 514 EXPORT_SYMBOL(ceph_osdc_get_request); 515 516 void ceph_osdc_put_request(struct ceph_osd_request *req) 517 { 518 if (req) { 519 dout("%s %p (was %d)\n", __func__, req, 520 kref_read(&req->r_kref)); 521 kref_put(&req->r_kref, ceph_osdc_release_request); 522 } 523 } 524 EXPORT_SYMBOL(ceph_osdc_put_request); 525 526 static void request_init(struct ceph_osd_request *req) 527 { 528 /* req only, each op is zeroed in osd_req_op_init() */ 529 memset(req, 0, sizeof(*req)); 530 531 kref_init(&req->r_kref); 532 init_completion(&req->r_completion); 533 RB_CLEAR_NODE(&req->r_node); 534 RB_CLEAR_NODE(&req->r_mc_node); 535 INIT_LIST_HEAD(&req->r_private_item); 536 537 target_init(&req->r_t); 538 } 539 540 struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc, 541 struct ceph_snap_context *snapc, 542 unsigned int num_ops, 543 bool use_mempool, 544 gfp_t gfp_flags) 545 { 546 struct ceph_osd_request *req; 547 548 if (use_mempool) { 549 BUG_ON(num_ops > CEPH_OSD_SLAB_OPS); 550 req = mempool_alloc(osdc->req_mempool, gfp_flags); 551 } else if (num_ops <= CEPH_OSD_SLAB_OPS) { 552 req = kmem_cache_alloc(ceph_osd_request_cache, gfp_flags); 553 } else { 554 BUG_ON(num_ops > CEPH_OSD_MAX_OPS); 555 req = kmalloc(struct_size(req, r_ops, num_ops), gfp_flags); 556 } 557 if (unlikely(!req)) 558 return NULL; 559 560 request_init(req); 561 req->r_osdc = osdc; 562 req->r_mempool = use_mempool; 563 req->r_num_ops = num_ops; 564 req->r_snapid = CEPH_NOSNAP; 565 req->r_snapc = ceph_get_snap_context(snapc); 566 567 dout("%s req %p\n", __func__, req); 568 return req; 569 } 570 EXPORT_SYMBOL(ceph_osdc_alloc_request); 571 572 static int ceph_oloc_encoding_size(const struct ceph_object_locator *oloc) 573 { 574 return 8 + 4 + 4 + 4 + (oloc->pool_ns ? oloc->pool_ns->len : 0); 575 } 576 577 static int __ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp, 578 int num_request_data_items, 579 int num_reply_data_items) 580 { 581 struct ceph_osd_client *osdc = req->r_osdc; 582 struct ceph_msg *msg; 583 int msg_size; 584 585 WARN_ON(req->r_request || req->r_reply); 586 WARN_ON(ceph_oid_empty(&req->r_base_oid)); 587 WARN_ON(ceph_oloc_empty(&req->r_base_oloc)); 588 589 /* create request message */ 590 msg_size = CEPH_ENCODING_START_BLK_LEN + 591 CEPH_PGID_ENCODING_LEN + 1; /* spgid */ 592 msg_size += 4 + 4 + 4; /* hash, osdmap_epoch, flags */ 593 msg_size += CEPH_ENCODING_START_BLK_LEN + 594 sizeof(struct ceph_osd_reqid); /* reqid */ 595 msg_size += sizeof(struct ceph_blkin_trace_info); /* trace */ 596 msg_size += 4 + sizeof(struct ceph_timespec); /* client_inc, mtime */ 597 msg_size += CEPH_ENCODING_START_BLK_LEN + 598 ceph_oloc_encoding_size(&req->r_base_oloc); /* oloc */ 599 msg_size += 4 + req->r_base_oid.name_len; /* oid */ 600 msg_size += 2 + req->r_num_ops * sizeof(struct ceph_osd_op); 601 msg_size += 8; /* snapid */ 602 msg_size += 8; /* snap_seq */ 603 msg_size += 4 + 8 * (req->r_snapc ? req->r_snapc->num_snaps : 0); 604 msg_size += 4 + 8; /* retry_attempt, features */ 605 606 if (req->r_mempool) 607 msg = ceph_msgpool_get(&osdc->msgpool_op, msg_size, 608 num_request_data_items); 609 else 610 msg = ceph_msg_new2(CEPH_MSG_OSD_OP, msg_size, 611 num_request_data_items, gfp, true); 612 if (!msg) 613 return -ENOMEM; 614 615 memset(msg->front.iov_base, 0, msg->front.iov_len); 616 req->r_request = msg; 617 618 /* create reply message */ 619 msg_size = OSD_OPREPLY_FRONT_LEN; 620 msg_size += req->r_base_oid.name_len; 621 msg_size += req->r_num_ops * sizeof(struct ceph_osd_op); 622 623 if (req->r_mempool) 624 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, msg_size, 625 num_reply_data_items); 626 else 627 msg = ceph_msg_new2(CEPH_MSG_OSD_OPREPLY, msg_size, 628 num_reply_data_items, gfp, true); 629 if (!msg) 630 return -ENOMEM; 631 632 req->r_reply = msg; 633 634 return 0; 635 } 636 637 static bool osd_req_opcode_valid(u16 opcode) 638 { 639 switch (opcode) { 640 #define GENERATE_CASE(op, opcode, str) case CEPH_OSD_OP_##op: return true; 641 __CEPH_FORALL_OSD_OPS(GENERATE_CASE) 642 #undef GENERATE_CASE 643 default: 644 return false; 645 } 646 } 647 648 static void get_num_data_items(struct ceph_osd_request *req, 649 int *num_request_data_items, 650 int *num_reply_data_items) 651 { 652 struct ceph_osd_req_op *op; 653 654 *num_request_data_items = 0; 655 *num_reply_data_items = 0; 656 657 for (op = req->r_ops; op != &req->r_ops[req->r_num_ops]; op++) { 658 switch (op->op) { 659 /* request */ 660 case CEPH_OSD_OP_WRITE: 661 case CEPH_OSD_OP_WRITEFULL: 662 case CEPH_OSD_OP_SETXATTR: 663 case CEPH_OSD_OP_CMPXATTR: 664 case CEPH_OSD_OP_NOTIFY_ACK: 665 case CEPH_OSD_OP_COPY_FROM2: 666 *num_request_data_items += 1; 667 break; 668 669 /* reply */ 670 case CEPH_OSD_OP_STAT: 671 case CEPH_OSD_OP_READ: 672 case CEPH_OSD_OP_LIST_WATCHERS: 673 *num_reply_data_items += 1; 674 break; 675 676 /* both */ 677 case CEPH_OSD_OP_NOTIFY: 678 *num_request_data_items += 1; 679 *num_reply_data_items += 1; 680 break; 681 case CEPH_OSD_OP_CALL: 682 *num_request_data_items += 2; 683 *num_reply_data_items += 1; 684 break; 685 686 default: 687 WARN_ON(!osd_req_opcode_valid(op->op)); 688 break; 689 } 690 } 691 } 692 693 /* 694 * oid, oloc and OSD op opcode(s) must be filled in before this function 695 * is called. 696 */ 697 int ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp) 698 { 699 int num_request_data_items, num_reply_data_items; 700 701 get_num_data_items(req, &num_request_data_items, &num_reply_data_items); 702 return __ceph_osdc_alloc_messages(req, gfp, num_request_data_items, 703 num_reply_data_items); 704 } 705 EXPORT_SYMBOL(ceph_osdc_alloc_messages); 706 707 /* 708 * This is an osd op init function for opcodes that have no data or 709 * other information associated with them. It also serves as a 710 * common init routine for all the other init functions, below. 711 */ 712 struct ceph_osd_req_op * 713 osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which, 714 u16 opcode, u32 flags) 715 { 716 struct ceph_osd_req_op *op; 717 718 BUG_ON(which >= osd_req->r_num_ops); 719 BUG_ON(!osd_req_opcode_valid(opcode)); 720 721 op = &osd_req->r_ops[which]; 722 memset(op, 0, sizeof (*op)); 723 op->op = opcode; 724 op->flags = flags; 725 726 return op; 727 } 728 EXPORT_SYMBOL(osd_req_op_init); 729 730 void osd_req_op_extent_init(struct ceph_osd_request *osd_req, 731 unsigned int which, u16 opcode, 732 u64 offset, u64 length, 733 u64 truncate_size, u32 truncate_seq) 734 { 735 struct ceph_osd_req_op *op = osd_req_op_init(osd_req, which, 736 opcode, 0); 737 size_t payload_len = 0; 738 739 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE && 740 opcode != CEPH_OSD_OP_WRITEFULL && opcode != CEPH_OSD_OP_ZERO && 741 opcode != CEPH_OSD_OP_TRUNCATE); 742 743 op->extent.offset = offset; 744 op->extent.length = length; 745 op->extent.truncate_size = truncate_size; 746 op->extent.truncate_seq = truncate_seq; 747 if (opcode == CEPH_OSD_OP_WRITE || opcode == CEPH_OSD_OP_WRITEFULL) 748 payload_len += length; 749 750 op->indata_len = payload_len; 751 } 752 EXPORT_SYMBOL(osd_req_op_extent_init); 753 754 void osd_req_op_extent_update(struct ceph_osd_request *osd_req, 755 unsigned int which, u64 length) 756 { 757 struct ceph_osd_req_op *op; 758 u64 previous; 759 760 BUG_ON(which >= osd_req->r_num_ops); 761 op = &osd_req->r_ops[which]; 762 previous = op->extent.length; 763 764 if (length == previous) 765 return; /* Nothing to do */ 766 BUG_ON(length > previous); 767 768 op->extent.length = length; 769 if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL) 770 op->indata_len -= previous - length; 771 } 772 EXPORT_SYMBOL(osd_req_op_extent_update); 773 774 void osd_req_op_extent_dup_last(struct ceph_osd_request *osd_req, 775 unsigned int which, u64 offset_inc) 776 { 777 struct ceph_osd_req_op *op, *prev_op; 778 779 BUG_ON(which + 1 >= osd_req->r_num_ops); 780 781 prev_op = &osd_req->r_ops[which]; 782 op = osd_req_op_init(osd_req, which + 1, prev_op->op, prev_op->flags); 783 /* dup previous one */ 784 op->indata_len = prev_op->indata_len; 785 op->outdata_len = prev_op->outdata_len; 786 op->extent = prev_op->extent; 787 /* adjust offset */ 788 op->extent.offset += offset_inc; 789 op->extent.length -= offset_inc; 790 791 if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL) 792 op->indata_len -= offset_inc; 793 } 794 EXPORT_SYMBOL(osd_req_op_extent_dup_last); 795 796 int osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which, 797 const char *class, const char *method) 798 { 799 struct ceph_osd_req_op *op; 800 struct ceph_pagelist *pagelist; 801 size_t payload_len = 0; 802 size_t size; 803 int ret; 804 805 op = osd_req_op_init(osd_req, which, CEPH_OSD_OP_CALL, 0); 806 807 pagelist = ceph_pagelist_alloc(GFP_NOFS); 808 if (!pagelist) 809 return -ENOMEM; 810 811 op->cls.class_name = class; 812 size = strlen(class); 813 BUG_ON(size > (size_t) U8_MAX); 814 op->cls.class_len = size; 815 ret = ceph_pagelist_append(pagelist, class, size); 816 if (ret) 817 goto err_pagelist_free; 818 payload_len += size; 819 820 op->cls.method_name = method; 821 size = strlen(method); 822 BUG_ON(size > (size_t) U8_MAX); 823 op->cls.method_len = size; 824 ret = ceph_pagelist_append(pagelist, method, size); 825 if (ret) 826 goto err_pagelist_free; 827 payload_len += size; 828 829 osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist); 830 op->indata_len = payload_len; 831 return 0; 832 833 err_pagelist_free: 834 ceph_pagelist_release(pagelist); 835 return ret; 836 } 837 EXPORT_SYMBOL(osd_req_op_cls_init); 838 839 int osd_req_op_xattr_init(struct ceph_osd_request *osd_req, unsigned int which, 840 u16 opcode, const char *name, const void *value, 841 size_t size, u8 cmp_op, u8 cmp_mode) 842 { 843 struct ceph_osd_req_op *op = osd_req_op_init(osd_req, which, 844 opcode, 0); 845 struct ceph_pagelist *pagelist; 846 size_t payload_len; 847 int ret; 848 849 BUG_ON(opcode != CEPH_OSD_OP_SETXATTR && opcode != CEPH_OSD_OP_CMPXATTR); 850 851 pagelist = ceph_pagelist_alloc(GFP_NOFS); 852 if (!pagelist) 853 return -ENOMEM; 854 855 payload_len = strlen(name); 856 op->xattr.name_len = payload_len; 857 ret = ceph_pagelist_append(pagelist, name, payload_len); 858 if (ret) 859 goto err_pagelist_free; 860 861 op->xattr.value_len = size; 862 ret = ceph_pagelist_append(pagelist, value, size); 863 if (ret) 864 goto err_pagelist_free; 865 payload_len += size; 866 867 op->xattr.cmp_op = cmp_op; 868 op->xattr.cmp_mode = cmp_mode; 869 870 ceph_osd_data_pagelist_init(&op->xattr.osd_data, pagelist); 871 op->indata_len = payload_len; 872 return 0; 873 874 err_pagelist_free: 875 ceph_pagelist_release(pagelist); 876 return ret; 877 } 878 EXPORT_SYMBOL(osd_req_op_xattr_init); 879 880 /* 881 * @watch_opcode: CEPH_OSD_WATCH_OP_* 882 */ 883 static void osd_req_op_watch_init(struct ceph_osd_request *req, int which, 884 u8 watch_opcode, u64 cookie, u32 gen) 885 { 886 struct ceph_osd_req_op *op; 887 888 op = osd_req_op_init(req, which, CEPH_OSD_OP_WATCH, 0); 889 op->watch.cookie = cookie; 890 op->watch.op = watch_opcode; 891 op->watch.gen = gen; 892 } 893 894 /* 895 * prot_ver, timeout and notify payload (may be empty) should already be 896 * encoded in @request_pl 897 */ 898 static void osd_req_op_notify_init(struct ceph_osd_request *req, int which, 899 u64 cookie, struct ceph_pagelist *request_pl) 900 { 901 struct ceph_osd_req_op *op; 902 903 op = osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY, 0); 904 op->notify.cookie = cookie; 905 906 ceph_osd_data_pagelist_init(&op->notify.request_data, request_pl); 907 op->indata_len = request_pl->length; 908 } 909 910 /* 911 * @flags: CEPH_OSD_OP_ALLOC_HINT_FLAG_* 912 */ 913 void osd_req_op_alloc_hint_init(struct ceph_osd_request *osd_req, 914 unsigned int which, 915 u64 expected_object_size, 916 u64 expected_write_size, 917 u32 flags) 918 { 919 struct ceph_osd_req_op *op; 920 921 op = osd_req_op_init(osd_req, which, CEPH_OSD_OP_SETALLOCHINT, 0); 922 op->alloc_hint.expected_object_size = expected_object_size; 923 op->alloc_hint.expected_write_size = expected_write_size; 924 op->alloc_hint.flags = flags; 925 926 /* 927 * CEPH_OSD_OP_SETALLOCHINT op is advisory and therefore deemed 928 * not worth a feature bit. Set FAILOK per-op flag to make 929 * sure older osds don't trip over an unsupported opcode. 930 */ 931 op->flags |= CEPH_OSD_OP_FLAG_FAILOK; 932 } 933 EXPORT_SYMBOL(osd_req_op_alloc_hint_init); 934 935 static void ceph_osdc_msg_data_add(struct ceph_msg *msg, 936 struct ceph_osd_data *osd_data) 937 { 938 u64 length = ceph_osd_data_length(osd_data); 939 940 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) { 941 BUG_ON(length > (u64) SIZE_MAX); 942 if (length) 943 ceph_msg_data_add_pages(msg, osd_data->pages, 944 length, osd_data->alignment, false); 945 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) { 946 BUG_ON(!length); 947 ceph_msg_data_add_pagelist(msg, osd_data->pagelist); 948 #ifdef CONFIG_BLOCK 949 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) { 950 ceph_msg_data_add_bio(msg, &osd_data->bio_pos, length); 951 #endif 952 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BVECS) { 953 ceph_msg_data_add_bvecs(msg, &osd_data->bvec_pos); 954 } else { 955 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE); 956 } 957 } 958 959 static u32 osd_req_encode_op(struct ceph_osd_op *dst, 960 const struct ceph_osd_req_op *src) 961 { 962 switch (src->op) { 963 case CEPH_OSD_OP_STAT: 964 break; 965 case CEPH_OSD_OP_READ: 966 case CEPH_OSD_OP_WRITE: 967 case CEPH_OSD_OP_WRITEFULL: 968 case CEPH_OSD_OP_ZERO: 969 case CEPH_OSD_OP_TRUNCATE: 970 dst->extent.offset = cpu_to_le64(src->extent.offset); 971 dst->extent.length = cpu_to_le64(src->extent.length); 972 dst->extent.truncate_size = 973 cpu_to_le64(src->extent.truncate_size); 974 dst->extent.truncate_seq = 975 cpu_to_le32(src->extent.truncate_seq); 976 break; 977 case CEPH_OSD_OP_CALL: 978 dst->cls.class_len = src->cls.class_len; 979 dst->cls.method_len = src->cls.method_len; 980 dst->cls.indata_len = cpu_to_le32(src->cls.indata_len); 981 break; 982 case CEPH_OSD_OP_WATCH: 983 dst->watch.cookie = cpu_to_le64(src->watch.cookie); 984 dst->watch.ver = cpu_to_le64(0); 985 dst->watch.op = src->watch.op; 986 dst->watch.gen = cpu_to_le32(src->watch.gen); 987 break; 988 case CEPH_OSD_OP_NOTIFY_ACK: 989 break; 990 case CEPH_OSD_OP_NOTIFY: 991 dst->notify.cookie = cpu_to_le64(src->notify.cookie); 992 break; 993 case CEPH_OSD_OP_LIST_WATCHERS: 994 break; 995 case CEPH_OSD_OP_SETALLOCHINT: 996 dst->alloc_hint.expected_object_size = 997 cpu_to_le64(src->alloc_hint.expected_object_size); 998 dst->alloc_hint.expected_write_size = 999 cpu_to_le64(src->alloc_hint.expected_write_size); 1000 dst->alloc_hint.flags = cpu_to_le32(src->alloc_hint.flags); 1001 break; 1002 case CEPH_OSD_OP_SETXATTR: 1003 case CEPH_OSD_OP_CMPXATTR: 1004 dst->xattr.name_len = cpu_to_le32(src->xattr.name_len); 1005 dst->xattr.value_len = cpu_to_le32(src->xattr.value_len); 1006 dst->xattr.cmp_op = src->xattr.cmp_op; 1007 dst->xattr.cmp_mode = src->xattr.cmp_mode; 1008 break; 1009 case CEPH_OSD_OP_CREATE: 1010 case CEPH_OSD_OP_DELETE: 1011 break; 1012 case CEPH_OSD_OP_COPY_FROM2: 1013 dst->copy_from.snapid = cpu_to_le64(src->copy_from.snapid); 1014 dst->copy_from.src_version = 1015 cpu_to_le64(src->copy_from.src_version); 1016 dst->copy_from.flags = src->copy_from.flags; 1017 dst->copy_from.src_fadvise_flags = 1018 cpu_to_le32(src->copy_from.src_fadvise_flags); 1019 break; 1020 default: 1021 pr_err("unsupported osd opcode %s\n", 1022 ceph_osd_op_name(src->op)); 1023 WARN_ON(1); 1024 1025 return 0; 1026 } 1027 1028 dst->op = cpu_to_le16(src->op); 1029 dst->flags = cpu_to_le32(src->flags); 1030 dst->payload_len = cpu_to_le32(src->indata_len); 1031 1032 return src->indata_len; 1033 } 1034 1035 /* 1036 * build new request AND message, calculate layout, and adjust file 1037 * extent as needed. 1038 * 1039 * if the file was recently truncated, we include information about its 1040 * old and new size so that the object can be updated appropriately. (we 1041 * avoid synchronously deleting truncated objects because it's slow.) 1042 */ 1043 struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc, 1044 struct ceph_file_layout *layout, 1045 struct ceph_vino vino, 1046 u64 off, u64 *plen, 1047 unsigned int which, int num_ops, 1048 int opcode, int flags, 1049 struct ceph_snap_context *snapc, 1050 u32 truncate_seq, 1051 u64 truncate_size, 1052 bool use_mempool) 1053 { 1054 struct ceph_osd_request *req; 1055 u64 objnum = 0; 1056 u64 objoff = 0; 1057 u64 objlen = 0; 1058 int r; 1059 1060 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE && 1061 opcode != CEPH_OSD_OP_ZERO && opcode != CEPH_OSD_OP_TRUNCATE && 1062 opcode != CEPH_OSD_OP_CREATE && opcode != CEPH_OSD_OP_DELETE); 1063 1064 req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool, 1065 GFP_NOFS); 1066 if (!req) { 1067 r = -ENOMEM; 1068 goto fail; 1069 } 1070 1071 /* calculate max write size */ 1072 r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen); 1073 if (r) 1074 goto fail; 1075 1076 if (opcode == CEPH_OSD_OP_CREATE || opcode == CEPH_OSD_OP_DELETE) { 1077 osd_req_op_init(req, which, opcode, 0); 1078 } else { 1079 u32 object_size = layout->object_size; 1080 u32 object_base = off - objoff; 1081 if (!(truncate_seq == 1 && truncate_size == -1ULL)) { 1082 if (truncate_size <= object_base) { 1083 truncate_size = 0; 1084 } else { 1085 truncate_size -= object_base; 1086 if (truncate_size > object_size) 1087 truncate_size = object_size; 1088 } 1089 } 1090 osd_req_op_extent_init(req, which, opcode, objoff, objlen, 1091 truncate_size, truncate_seq); 1092 } 1093 1094 req->r_base_oloc.pool = layout->pool_id; 1095 req->r_base_oloc.pool_ns = ceph_try_get_string(layout->pool_ns); 1096 ceph_oid_printf(&req->r_base_oid, "%llx.%08llx", vino.ino, objnum); 1097 req->r_flags = flags | osdc->client->options->read_from_replica; 1098 1099 req->r_snapid = vino.snap; 1100 if (flags & CEPH_OSD_FLAG_WRITE) 1101 req->r_data_offset = off; 1102 1103 if (num_ops > 1) 1104 /* 1105 * This is a special case for ceph_writepages_start(), but it 1106 * also covers ceph_uninline_data(). If more multi-op request 1107 * use cases emerge, we will need a separate helper. 1108 */ 1109 r = __ceph_osdc_alloc_messages(req, GFP_NOFS, num_ops, 0); 1110 else 1111 r = ceph_osdc_alloc_messages(req, GFP_NOFS); 1112 if (r) 1113 goto fail; 1114 1115 return req; 1116 1117 fail: 1118 ceph_osdc_put_request(req); 1119 return ERR_PTR(r); 1120 } 1121 EXPORT_SYMBOL(ceph_osdc_new_request); 1122 1123 /* 1124 * We keep osd requests in an rbtree, sorted by ->r_tid. 1125 */ 1126 DEFINE_RB_FUNCS(request, struct ceph_osd_request, r_tid, r_node) 1127 DEFINE_RB_FUNCS(request_mc, struct ceph_osd_request, r_tid, r_mc_node) 1128 1129 /* 1130 * Call @fn on each OSD request as long as @fn returns 0. 1131 */ 1132 static void for_each_request(struct ceph_osd_client *osdc, 1133 int (*fn)(struct ceph_osd_request *req, void *arg), 1134 void *arg) 1135 { 1136 struct rb_node *n, *p; 1137 1138 for (n = rb_first(&osdc->osds); n; n = rb_next(n)) { 1139 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node); 1140 1141 for (p = rb_first(&osd->o_requests); p; ) { 1142 struct ceph_osd_request *req = 1143 rb_entry(p, struct ceph_osd_request, r_node); 1144 1145 p = rb_next(p); 1146 if (fn(req, arg)) 1147 return; 1148 } 1149 } 1150 1151 for (p = rb_first(&osdc->homeless_osd.o_requests); p; ) { 1152 struct ceph_osd_request *req = 1153 rb_entry(p, struct ceph_osd_request, r_node); 1154 1155 p = rb_next(p); 1156 if (fn(req, arg)) 1157 return; 1158 } 1159 } 1160 1161 static bool osd_homeless(struct ceph_osd *osd) 1162 { 1163 return osd->o_osd == CEPH_HOMELESS_OSD; 1164 } 1165 1166 static bool osd_registered(struct ceph_osd *osd) 1167 { 1168 verify_osdc_locked(osd->o_osdc); 1169 1170 return !RB_EMPTY_NODE(&osd->o_node); 1171 } 1172 1173 /* 1174 * Assumes @osd is zero-initialized. 1175 */ 1176 static void osd_init(struct ceph_osd *osd) 1177 { 1178 refcount_set(&osd->o_ref, 1); 1179 RB_CLEAR_NODE(&osd->o_node); 1180 osd->o_requests = RB_ROOT; 1181 osd->o_linger_requests = RB_ROOT; 1182 osd->o_backoff_mappings = RB_ROOT; 1183 osd->o_backoffs_by_id = RB_ROOT; 1184 INIT_LIST_HEAD(&osd->o_osd_lru); 1185 INIT_LIST_HEAD(&osd->o_keepalive_item); 1186 osd->o_incarnation = 1; 1187 mutex_init(&osd->lock); 1188 } 1189 1190 static void osd_cleanup(struct ceph_osd *osd) 1191 { 1192 WARN_ON(!RB_EMPTY_NODE(&osd->o_node)); 1193 WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests)); 1194 WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests)); 1195 WARN_ON(!RB_EMPTY_ROOT(&osd->o_backoff_mappings)); 1196 WARN_ON(!RB_EMPTY_ROOT(&osd->o_backoffs_by_id)); 1197 WARN_ON(!list_empty(&osd->o_osd_lru)); 1198 WARN_ON(!list_empty(&osd->o_keepalive_item)); 1199 1200 if (osd->o_auth.authorizer) { 1201 WARN_ON(osd_homeless(osd)); 1202 ceph_auth_destroy_authorizer(osd->o_auth.authorizer); 1203 } 1204 } 1205 1206 /* 1207 * Track open sessions with osds. 1208 */ 1209 static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum) 1210 { 1211 struct ceph_osd *osd; 1212 1213 WARN_ON(onum == CEPH_HOMELESS_OSD); 1214 1215 osd = kzalloc(sizeof(*osd), GFP_NOIO | __GFP_NOFAIL); 1216 osd_init(osd); 1217 osd->o_osdc = osdc; 1218 osd->o_osd = onum; 1219 1220 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr); 1221 1222 return osd; 1223 } 1224 1225 static struct ceph_osd *get_osd(struct ceph_osd *osd) 1226 { 1227 if (refcount_inc_not_zero(&osd->o_ref)) { 1228 dout("get_osd %p %d -> %d\n", osd, refcount_read(&osd->o_ref)-1, 1229 refcount_read(&osd->o_ref)); 1230 return osd; 1231 } else { 1232 dout("get_osd %p FAIL\n", osd); 1233 return NULL; 1234 } 1235 } 1236 1237 static void put_osd(struct ceph_osd *osd) 1238 { 1239 dout("put_osd %p %d -> %d\n", osd, refcount_read(&osd->o_ref), 1240 refcount_read(&osd->o_ref) - 1); 1241 if (refcount_dec_and_test(&osd->o_ref)) { 1242 osd_cleanup(osd); 1243 kfree(osd); 1244 } 1245 } 1246 1247 DEFINE_RB_FUNCS(osd, struct ceph_osd, o_osd, o_node) 1248 1249 static void __move_osd_to_lru(struct ceph_osd *osd) 1250 { 1251 struct ceph_osd_client *osdc = osd->o_osdc; 1252 1253 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd); 1254 BUG_ON(!list_empty(&osd->o_osd_lru)); 1255 1256 spin_lock(&osdc->osd_lru_lock); 1257 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru); 1258 spin_unlock(&osdc->osd_lru_lock); 1259 1260 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl; 1261 } 1262 1263 static void maybe_move_osd_to_lru(struct ceph_osd *osd) 1264 { 1265 if (RB_EMPTY_ROOT(&osd->o_requests) && 1266 RB_EMPTY_ROOT(&osd->o_linger_requests)) 1267 __move_osd_to_lru(osd); 1268 } 1269 1270 static void __remove_osd_from_lru(struct ceph_osd *osd) 1271 { 1272 struct ceph_osd_client *osdc = osd->o_osdc; 1273 1274 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd); 1275 1276 spin_lock(&osdc->osd_lru_lock); 1277 if (!list_empty(&osd->o_osd_lru)) 1278 list_del_init(&osd->o_osd_lru); 1279 spin_unlock(&osdc->osd_lru_lock); 1280 } 1281 1282 /* 1283 * Close the connection and assign any leftover requests to the 1284 * homeless session. 1285 */ 1286 static void close_osd(struct ceph_osd *osd) 1287 { 1288 struct ceph_osd_client *osdc = osd->o_osdc; 1289 struct rb_node *n; 1290 1291 verify_osdc_wrlocked(osdc); 1292 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd); 1293 1294 ceph_con_close(&osd->o_con); 1295 1296 for (n = rb_first(&osd->o_requests); n; ) { 1297 struct ceph_osd_request *req = 1298 rb_entry(n, struct ceph_osd_request, r_node); 1299 1300 n = rb_next(n); /* unlink_request() */ 1301 1302 dout(" reassigning req %p tid %llu\n", req, req->r_tid); 1303 unlink_request(osd, req); 1304 link_request(&osdc->homeless_osd, req); 1305 } 1306 for (n = rb_first(&osd->o_linger_requests); n; ) { 1307 struct ceph_osd_linger_request *lreq = 1308 rb_entry(n, struct ceph_osd_linger_request, node); 1309 1310 n = rb_next(n); /* unlink_linger() */ 1311 1312 dout(" reassigning lreq %p linger_id %llu\n", lreq, 1313 lreq->linger_id); 1314 unlink_linger(osd, lreq); 1315 link_linger(&osdc->homeless_osd, lreq); 1316 } 1317 clear_backoffs(osd); 1318 1319 __remove_osd_from_lru(osd); 1320 erase_osd(&osdc->osds, osd); 1321 put_osd(osd); 1322 } 1323 1324 /* 1325 * reset osd connect 1326 */ 1327 static int reopen_osd(struct ceph_osd *osd) 1328 { 1329 struct ceph_entity_addr *peer_addr; 1330 1331 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd); 1332 1333 if (RB_EMPTY_ROOT(&osd->o_requests) && 1334 RB_EMPTY_ROOT(&osd->o_linger_requests)) { 1335 close_osd(osd); 1336 return -ENODEV; 1337 } 1338 1339 peer_addr = &osd->o_osdc->osdmap->osd_addr[osd->o_osd]; 1340 if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) && 1341 !ceph_con_opened(&osd->o_con)) { 1342 struct rb_node *n; 1343 1344 dout("osd addr hasn't changed and connection never opened, " 1345 "letting msgr retry\n"); 1346 /* touch each r_stamp for handle_timeout()'s benfit */ 1347 for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) { 1348 struct ceph_osd_request *req = 1349 rb_entry(n, struct ceph_osd_request, r_node); 1350 req->r_stamp = jiffies; 1351 } 1352 1353 return -EAGAIN; 1354 } 1355 1356 ceph_con_close(&osd->o_con); 1357 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr); 1358 osd->o_incarnation++; 1359 1360 return 0; 1361 } 1362 1363 static struct ceph_osd *lookup_create_osd(struct ceph_osd_client *osdc, int o, 1364 bool wrlocked) 1365 { 1366 struct ceph_osd *osd; 1367 1368 if (wrlocked) 1369 verify_osdc_wrlocked(osdc); 1370 else 1371 verify_osdc_locked(osdc); 1372 1373 if (o != CEPH_HOMELESS_OSD) 1374 osd = lookup_osd(&osdc->osds, o); 1375 else 1376 osd = &osdc->homeless_osd; 1377 if (!osd) { 1378 if (!wrlocked) 1379 return ERR_PTR(-EAGAIN); 1380 1381 osd = create_osd(osdc, o); 1382 insert_osd(&osdc->osds, osd); 1383 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, 1384 &osdc->osdmap->osd_addr[osd->o_osd]); 1385 } 1386 1387 dout("%s osdc %p osd%d -> osd %p\n", __func__, osdc, o, osd); 1388 return osd; 1389 } 1390 1391 /* 1392 * Create request <-> OSD session relation. 1393 * 1394 * @req has to be assigned a tid, @osd may be homeless. 1395 */ 1396 static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req) 1397 { 1398 verify_osd_locked(osd); 1399 WARN_ON(!req->r_tid || req->r_osd); 1400 dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd, 1401 req, req->r_tid); 1402 1403 if (!osd_homeless(osd)) 1404 __remove_osd_from_lru(osd); 1405 else 1406 atomic_inc(&osd->o_osdc->num_homeless); 1407 1408 get_osd(osd); 1409 insert_request(&osd->o_requests, req); 1410 req->r_osd = osd; 1411 } 1412 1413 static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req) 1414 { 1415 verify_osd_locked(osd); 1416 WARN_ON(req->r_osd != osd); 1417 dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd, 1418 req, req->r_tid); 1419 1420 req->r_osd = NULL; 1421 erase_request(&osd->o_requests, req); 1422 put_osd(osd); 1423 1424 if (!osd_homeless(osd)) 1425 maybe_move_osd_to_lru(osd); 1426 else 1427 atomic_dec(&osd->o_osdc->num_homeless); 1428 } 1429 1430 static bool __pool_full(struct ceph_pg_pool_info *pi) 1431 { 1432 return pi->flags & CEPH_POOL_FLAG_FULL; 1433 } 1434 1435 static bool have_pool_full(struct ceph_osd_client *osdc) 1436 { 1437 struct rb_node *n; 1438 1439 for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) { 1440 struct ceph_pg_pool_info *pi = 1441 rb_entry(n, struct ceph_pg_pool_info, node); 1442 1443 if (__pool_full(pi)) 1444 return true; 1445 } 1446 1447 return false; 1448 } 1449 1450 static bool pool_full(struct ceph_osd_client *osdc, s64 pool_id) 1451 { 1452 struct ceph_pg_pool_info *pi; 1453 1454 pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id); 1455 if (!pi) 1456 return false; 1457 1458 return __pool_full(pi); 1459 } 1460 1461 /* 1462 * Returns whether a request should be blocked from being sent 1463 * based on the current osdmap and osd_client settings. 1464 */ 1465 static bool target_should_be_paused(struct ceph_osd_client *osdc, 1466 const struct ceph_osd_request_target *t, 1467 struct ceph_pg_pool_info *pi) 1468 { 1469 bool pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD); 1470 bool pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) || 1471 ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || 1472 __pool_full(pi); 1473 1474 WARN_ON(pi->id != t->target_oloc.pool); 1475 return ((t->flags & CEPH_OSD_FLAG_READ) && pauserd) || 1476 ((t->flags & CEPH_OSD_FLAG_WRITE) && pausewr) || 1477 (osdc->osdmap->epoch < osdc->epoch_barrier); 1478 } 1479 1480 static int pick_random_replica(const struct ceph_osds *acting) 1481 { 1482 int i = prandom_u32_max(acting->size); 1483 1484 dout("%s picked osd%d, primary osd%d\n", __func__, 1485 acting->osds[i], acting->primary); 1486 return i; 1487 } 1488 1489 /* 1490 * Picks the closest replica based on client's location given by 1491 * crush_location option. Prefers the primary if the locality is 1492 * the same. 1493 */ 1494 static int pick_closest_replica(struct ceph_osd_client *osdc, 1495 const struct ceph_osds *acting) 1496 { 1497 struct ceph_options *opt = osdc->client->options; 1498 int best_i, best_locality; 1499 int i = 0, locality; 1500 1501 do { 1502 locality = ceph_get_crush_locality(osdc->osdmap, 1503 acting->osds[i], 1504 &opt->crush_locs); 1505 if (i == 0 || 1506 (locality >= 0 && best_locality < 0) || 1507 (locality >= 0 && best_locality >= 0 && 1508 locality < best_locality)) { 1509 best_i = i; 1510 best_locality = locality; 1511 } 1512 } while (++i < acting->size); 1513 1514 dout("%s picked osd%d with locality %d, primary osd%d\n", __func__, 1515 acting->osds[best_i], best_locality, acting->primary); 1516 return best_i; 1517 } 1518 1519 enum calc_target_result { 1520 CALC_TARGET_NO_ACTION = 0, 1521 CALC_TARGET_NEED_RESEND, 1522 CALC_TARGET_POOL_DNE, 1523 }; 1524 1525 static enum calc_target_result calc_target(struct ceph_osd_client *osdc, 1526 struct ceph_osd_request_target *t, 1527 bool any_change) 1528 { 1529 struct ceph_pg_pool_info *pi; 1530 struct ceph_pg pgid, last_pgid; 1531 struct ceph_osds up, acting; 1532 bool is_read = t->flags & CEPH_OSD_FLAG_READ; 1533 bool is_write = t->flags & CEPH_OSD_FLAG_WRITE; 1534 bool force_resend = false; 1535 bool unpaused = false; 1536 bool legacy_change = false; 1537 bool split = false; 1538 bool sort_bitwise = ceph_osdmap_flag(osdc, CEPH_OSDMAP_SORTBITWISE); 1539 bool recovery_deletes = ceph_osdmap_flag(osdc, 1540 CEPH_OSDMAP_RECOVERY_DELETES); 1541 enum calc_target_result ct_res; 1542 1543 t->epoch = osdc->osdmap->epoch; 1544 pi = ceph_pg_pool_by_id(osdc->osdmap, t->base_oloc.pool); 1545 if (!pi) { 1546 t->osd = CEPH_HOMELESS_OSD; 1547 ct_res = CALC_TARGET_POOL_DNE; 1548 goto out; 1549 } 1550 1551 if (osdc->osdmap->epoch == pi->last_force_request_resend) { 1552 if (t->last_force_resend < pi->last_force_request_resend) { 1553 t->last_force_resend = pi->last_force_request_resend; 1554 force_resend = true; 1555 } else if (t->last_force_resend == 0) { 1556 force_resend = true; 1557 } 1558 } 1559 1560 /* apply tiering */ 1561 ceph_oid_copy(&t->target_oid, &t->base_oid); 1562 ceph_oloc_copy(&t->target_oloc, &t->base_oloc); 1563 if ((t->flags & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0) { 1564 if (is_read && pi->read_tier >= 0) 1565 t->target_oloc.pool = pi->read_tier; 1566 if (is_write && pi->write_tier >= 0) 1567 t->target_oloc.pool = pi->write_tier; 1568 1569 pi = ceph_pg_pool_by_id(osdc->osdmap, t->target_oloc.pool); 1570 if (!pi) { 1571 t->osd = CEPH_HOMELESS_OSD; 1572 ct_res = CALC_TARGET_POOL_DNE; 1573 goto out; 1574 } 1575 } 1576 1577 __ceph_object_locator_to_pg(pi, &t->target_oid, &t->target_oloc, &pgid); 1578 last_pgid.pool = pgid.pool; 1579 last_pgid.seed = ceph_stable_mod(pgid.seed, t->pg_num, t->pg_num_mask); 1580 1581 ceph_pg_to_up_acting_osds(osdc->osdmap, pi, &pgid, &up, &acting); 1582 if (any_change && 1583 ceph_is_new_interval(&t->acting, 1584 &acting, 1585 &t->up, 1586 &up, 1587 t->size, 1588 pi->size, 1589 t->min_size, 1590 pi->min_size, 1591 t->pg_num, 1592 pi->pg_num, 1593 t->sort_bitwise, 1594 sort_bitwise, 1595 t->recovery_deletes, 1596 recovery_deletes, 1597 &last_pgid)) 1598 force_resend = true; 1599 1600 if (t->paused && !target_should_be_paused(osdc, t, pi)) { 1601 t->paused = false; 1602 unpaused = true; 1603 } 1604 legacy_change = ceph_pg_compare(&t->pgid, &pgid) || 1605 ceph_osds_changed(&t->acting, &acting, 1606 t->used_replica || any_change); 1607 if (t->pg_num) 1608 split = ceph_pg_is_split(&last_pgid, t->pg_num, pi->pg_num); 1609 1610 if (legacy_change || force_resend || split) { 1611 t->pgid = pgid; /* struct */ 1612 ceph_pg_to_primary_shard(osdc->osdmap, pi, &pgid, &t->spgid); 1613 ceph_osds_copy(&t->acting, &acting); 1614 ceph_osds_copy(&t->up, &up); 1615 t->size = pi->size; 1616 t->min_size = pi->min_size; 1617 t->pg_num = pi->pg_num; 1618 t->pg_num_mask = pi->pg_num_mask; 1619 t->sort_bitwise = sort_bitwise; 1620 t->recovery_deletes = recovery_deletes; 1621 1622 if ((t->flags & (CEPH_OSD_FLAG_BALANCE_READS | 1623 CEPH_OSD_FLAG_LOCALIZE_READS)) && 1624 !is_write && pi->type == CEPH_POOL_TYPE_REP && 1625 acting.size > 1) { 1626 int pos; 1627 1628 WARN_ON(!is_read || acting.osds[0] != acting.primary); 1629 if (t->flags & CEPH_OSD_FLAG_BALANCE_READS) { 1630 pos = pick_random_replica(&acting); 1631 } else { 1632 pos = pick_closest_replica(osdc, &acting); 1633 } 1634 t->osd = acting.osds[pos]; 1635 t->used_replica = pos > 0; 1636 } else { 1637 t->osd = acting.primary; 1638 t->used_replica = false; 1639 } 1640 } 1641 1642 if (unpaused || legacy_change || force_resend || split) 1643 ct_res = CALC_TARGET_NEED_RESEND; 1644 else 1645 ct_res = CALC_TARGET_NO_ACTION; 1646 1647 out: 1648 dout("%s t %p -> %d%d%d%d ct_res %d osd%d\n", __func__, t, unpaused, 1649 legacy_change, force_resend, split, ct_res, t->osd); 1650 return ct_res; 1651 } 1652 1653 static struct ceph_spg_mapping *alloc_spg_mapping(void) 1654 { 1655 struct ceph_spg_mapping *spg; 1656 1657 spg = kmalloc(sizeof(*spg), GFP_NOIO); 1658 if (!spg) 1659 return NULL; 1660 1661 RB_CLEAR_NODE(&spg->node); 1662 spg->backoffs = RB_ROOT; 1663 return spg; 1664 } 1665 1666 static void free_spg_mapping(struct ceph_spg_mapping *spg) 1667 { 1668 WARN_ON(!RB_EMPTY_NODE(&spg->node)); 1669 WARN_ON(!RB_EMPTY_ROOT(&spg->backoffs)); 1670 1671 kfree(spg); 1672 } 1673 1674 /* 1675 * rbtree of ceph_spg_mapping for handling map<spg_t, ...>, similar to 1676 * ceph_pg_mapping. Used to track OSD backoffs -- a backoff [range] is 1677 * defined only within a specific spgid; it does not pass anything to 1678 * children on split, or to another primary. 1679 */ 1680 DEFINE_RB_FUNCS2(spg_mapping, struct ceph_spg_mapping, spgid, ceph_spg_compare, 1681 RB_BYPTR, const struct ceph_spg *, node) 1682 1683 static u64 hoid_get_bitwise_key(const struct ceph_hobject_id *hoid) 1684 { 1685 return hoid->is_max ? 0x100000000ull : hoid->hash_reverse_bits; 1686 } 1687 1688 static void hoid_get_effective_key(const struct ceph_hobject_id *hoid, 1689 void **pkey, size_t *pkey_len) 1690 { 1691 if (hoid->key_len) { 1692 *pkey = hoid->key; 1693 *pkey_len = hoid->key_len; 1694 } else { 1695 *pkey = hoid->oid; 1696 *pkey_len = hoid->oid_len; 1697 } 1698 } 1699 1700 static int compare_names(const void *name1, size_t name1_len, 1701 const void *name2, size_t name2_len) 1702 { 1703 int ret; 1704 1705 ret = memcmp(name1, name2, min(name1_len, name2_len)); 1706 if (!ret) { 1707 if (name1_len < name2_len) 1708 ret = -1; 1709 else if (name1_len > name2_len) 1710 ret = 1; 1711 } 1712 return ret; 1713 } 1714 1715 static int hoid_compare(const struct ceph_hobject_id *lhs, 1716 const struct ceph_hobject_id *rhs) 1717 { 1718 void *effective_key1, *effective_key2; 1719 size_t effective_key1_len, effective_key2_len; 1720 int ret; 1721 1722 if (lhs->is_max < rhs->is_max) 1723 return -1; 1724 if (lhs->is_max > rhs->is_max) 1725 return 1; 1726 1727 if (lhs->pool < rhs->pool) 1728 return -1; 1729 if (lhs->pool > rhs->pool) 1730 return 1; 1731 1732 if (hoid_get_bitwise_key(lhs) < hoid_get_bitwise_key(rhs)) 1733 return -1; 1734 if (hoid_get_bitwise_key(lhs) > hoid_get_bitwise_key(rhs)) 1735 return 1; 1736 1737 ret = compare_names(lhs->nspace, lhs->nspace_len, 1738 rhs->nspace, rhs->nspace_len); 1739 if (ret) 1740 return ret; 1741 1742 hoid_get_effective_key(lhs, &effective_key1, &effective_key1_len); 1743 hoid_get_effective_key(rhs, &effective_key2, &effective_key2_len); 1744 ret = compare_names(effective_key1, effective_key1_len, 1745 effective_key2, effective_key2_len); 1746 if (ret) 1747 return ret; 1748 1749 ret = compare_names(lhs->oid, lhs->oid_len, rhs->oid, rhs->oid_len); 1750 if (ret) 1751 return ret; 1752 1753 if (lhs->snapid < rhs->snapid) 1754 return -1; 1755 if (lhs->snapid > rhs->snapid) 1756 return 1; 1757 1758 return 0; 1759 } 1760 1761 /* 1762 * For decoding ->begin and ->end of MOSDBackoff only -- no MIN/MAX 1763 * compat stuff here. 1764 * 1765 * Assumes @hoid is zero-initialized. 1766 */ 1767 static int decode_hoid(void **p, void *end, struct ceph_hobject_id *hoid) 1768 { 1769 u8 struct_v; 1770 u32 struct_len; 1771 int ret; 1772 1773 ret = ceph_start_decoding(p, end, 4, "hobject_t", &struct_v, 1774 &struct_len); 1775 if (ret) 1776 return ret; 1777 1778 if (struct_v < 4) { 1779 pr_err("got struct_v %d < 4 of hobject_t\n", struct_v); 1780 goto e_inval; 1781 } 1782 1783 hoid->key = ceph_extract_encoded_string(p, end, &hoid->key_len, 1784 GFP_NOIO); 1785 if (IS_ERR(hoid->key)) { 1786 ret = PTR_ERR(hoid->key); 1787 hoid->key = NULL; 1788 return ret; 1789 } 1790 1791 hoid->oid = ceph_extract_encoded_string(p, end, &hoid->oid_len, 1792 GFP_NOIO); 1793 if (IS_ERR(hoid->oid)) { 1794 ret = PTR_ERR(hoid->oid); 1795 hoid->oid = NULL; 1796 return ret; 1797 } 1798 1799 ceph_decode_64_safe(p, end, hoid->snapid, e_inval); 1800 ceph_decode_32_safe(p, end, hoid->hash, e_inval); 1801 ceph_decode_8_safe(p, end, hoid->is_max, e_inval); 1802 1803 hoid->nspace = ceph_extract_encoded_string(p, end, &hoid->nspace_len, 1804 GFP_NOIO); 1805 if (IS_ERR(hoid->nspace)) { 1806 ret = PTR_ERR(hoid->nspace); 1807 hoid->nspace = NULL; 1808 return ret; 1809 } 1810 1811 ceph_decode_64_safe(p, end, hoid->pool, e_inval); 1812 1813 ceph_hoid_build_hash_cache(hoid); 1814 return 0; 1815 1816 e_inval: 1817 return -EINVAL; 1818 } 1819 1820 static int hoid_encoding_size(const struct ceph_hobject_id *hoid) 1821 { 1822 return 8 + 4 + 1 + 8 + /* snapid, hash, is_max, pool */ 1823 4 + hoid->key_len + 4 + hoid->oid_len + 4 + hoid->nspace_len; 1824 } 1825 1826 static void encode_hoid(void **p, void *end, const struct ceph_hobject_id *hoid) 1827 { 1828 ceph_start_encoding(p, 4, 3, hoid_encoding_size(hoid)); 1829 ceph_encode_string(p, end, hoid->key, hoid->key_len); 1830 ceph_encode_string(p, end, hoid->oid, hoid->oid_len); 1831 ceph_encode_64(p, hoid->snapid); 1832 ceph_encode_32(p, hoid->hash); 1833 ceph_encode_8(p, hoid->is_max); 1834 ceph_encode_string(p, end, hoid->nspace, hoid->nspace_len); 1835 ceph_encode_64(p, hoid->pool); 1836 } 1837 1838 static void free_hoid(struct ceph_hobject_id *hoid) 1839 { 1840 if (hoid) { 1841 kfree(hoid->key); 1842 kfree(hoid->oid); 1843 kfree(hoid->nspace); 1844 kfree(hoid); 1845 } 1846 } 1847 1848 static struct ceph_osd_backoff *alloc_backoff(void) 1849 { 1850 struct ceph_osd_backoff *backoff; 1851 1852 backoff = kzalloc(sizeof(*backoff), GFP_NOIO); 1853 if (!backoff) 1854 return NULL; 1855 1856 RB_CLEAR_NODE(&backoff->spg_node); 1857 RB_CLEAR_NODE(&backoff->id_node); 1858 return backoff; 1859 } 1860 1861 static void free_backoff(struct ceph_osd_backoff *backoff) 1862 { 1863 WARN_ON(!RB_EMPTY_NODE(&backoff->spg_node)); 1864 WARN_ON(!RB_EMPTY_NODE(&backoff->id_node)); 1865 1866 free_hoid(backoff->begin); 1867 free_hoid(backoff->end); 1868 kfree(backoff); 1869 } 1870 1871 /* 1872 * Within a specific spgid, backoffs are managed by ->begin hoid. 1873 */ 1874 DEFINE_RB_INSDEL_FUNCS2(backoff, struct ceph_osd_backoff, begin, hoid_compare, 1875 RB_BYVAL, spg_node); 1876 1877 static struct ceph_osd_backoff *lookup_containing_backoff(struct rb_root *root, 1878 const struct ceph_hobject_id *hoid) 1879 { 1880 struct rb_node *n = root->rb_node; 1881 1882 while (n) { 1883 struct ceph_osd_backoff *cur = 1884 rb_entry(n, struct ceph_osd_backoff, spg_node); 1885 int cmp; 1886 1887 cmp = hoid_compare(hoid, cur->begin); 1888 if (cmp < 0) { 1889 n = n->rb_left; 1890 } else if (cmp > 0) { 1891 if (hoid_compare(hoid, cur->end) < 0) 1892 return cur; 1893 1894 n = n->rb_right; 1895 } else { 1896 return cur; 1897 } 1898 } 1899 1900 return NULL; 1901 } 1902 1903 /* 1904 * Each backoff has a unique id within its OSD session. 1905 */ 1906 DEFINE_RB_FUNCS(backoff_by_id, struct ceph_osd_backoff, id, id_node) 1907 1908 static void clear_backoffs(struct ceph_osd *osd) 1909 { 1910 while (!RB_EMPTY_ROOT(&osd->o_backoff_mappings)) { 1911 struct ceph_spg_mapping *spg = 1912 rb_entry(rb_first(&osd->o_backoff_mappings), 1913 struct ceph_spg_mapping, node); 1914 1915 while (!RB_EMPTY_ROOT(&spg->backoffs)) { 1916 struct ceph_osd_backoff *backoff = 1917 rb_entry(rb_first(&spg->backoffs), 1918 struct ceph_osd_backoff, spg_node); 1919 1920 erase_backoff(&spg->backoffs, backoff); 1921 erase_backoff_by_id(&osd->o_backoffs_by_id, backoff); 1922 free_backoff(backoff); 1923 } 1924 erase_spg_mapping(&osd->o_backoff_mappings, spg); 1925 free_spg_mapping(spg); 1926 } 1927 } 1928 1929 /* 1930 * Set up a temporary, non-owning view into @t. 1931 */ 1932 static void hoid_fill_from_target(struct ceph_hobject_id *hoid, 1933 const struct ceph_osd_request_target *t) 1934 { 1935 hoid->key = NULL; 1936 hoid->key_len = 0; 1937 hoid->oid = t->target_oid.name; 1938 hoid->oid_len = t->target_oid.name_len; 1939 hoid->snapid = CEPH_NOSNAP; 1940 hoid->hash = t->pgid.seed; 1941 hoid->is_max = false; 1942 if (t->target_oloc.pool_ns) { 1943 hoid->nspace = t->target_oloc.pool_ns->str; 1944 hoid->nspace_len = t->target_oloc.pool_ns->len; 1945 } else { 1946 hoid->nspace = NULL; 1947 hoid->nspace_len = 0; 1948 } 1949 hoid->pool = t->target_oloc.pool; 1950 ceph_hoid_build_hash_cache(hoid); 1951 } 1952 1953 static bool should_plug_request(struct ceph_osd_request *req) 1954 { 1955 struct ceph_osd *osd = req->r_osd; 1956 struct ceph_spg_mapping *spg; 1957 struct ceph_osd_backoff *backoff; 1958 struct ceph_hobject_id hoid; 1959 1960 spg = lookup_spg_mapping(&osd->o_backoff_mappings, &req->r_t.spgid); 1961 if (!spg) 1962 return false; 1963 1964 hoid_fill_from_target(&hoid, &req->r_t); 1965 backoff = lookup_containing_backoff(&spg->backoffs, &hoid); 1966 if (!backoff) 1967 return false; 1968 1969 dout("%s req %p tid %llu backoff osd%d spgid %llu.%xs%d id %llu\n", 1970 __func__, req, req->r_tid, osd->o_osd, backoff->spgid.pgid.pool, 1971 backoff->spgid.pgid.seed, backoff->spgid.shard, backoff->id); 1972 return true; 1973 } 1974 1975 /* 1976 * Keep get_num_data_items() in sync with this function. 1977 */ 1978 static void setup_request_data(struct ceph_osd_request *req) 1979 { 1980 struct ceph_msg *request_msg = req->r_request; 1981 struct ceph_msg *reply_msg = req->r_reply; 1982 struct ceph_osd_req_op *op; 1983 1984 if (req->r_request->num_data_items || req->r_reply->num_data_items) 1985 return; 1986 1987 WARN_ON(request_msg->data_length || reply_msg->data_length); 1988 for (op = req->r_ops; op != &req->r_ops[req->r_num_ops]; op++) { 1989 switch (op->op) { 1990 /* request */ 1991 case CEPH_OSD_OP_WRITE: 1992 case CEPH_OSD_OP_WRITEFULL: 1993 WARN_ON(op->indata_len != op->extent.length); 1994 ceph_osdc_msg_data_add(request_msg, 1995 &op->extent.osd_data); 1996 break; 1997 case CEPH_OSD_OP_SETXATTR: 1998 case CEPH_OSD_OP_CMPXATTR: 1999 WARN_ON(op->indata_len != op->xattr.name_len + 2000 op->xattr.value_len); 2001 ceph_osdc_msg_data_add(request_msg, 2002 &op->xattr.osd_data); 2003 break; 2004 case CEPH_OSD_OP_NOTIFY_ACK: 2005 ceph_osdc_msg_data_add(request_msg, 2006 &op->notify_ack.request_data); 2007 break; 2008 case CEPH_OSD_OP_COPY_FROM2: 2009 ceph_osdc_msg_data_add(request_msg, 2010 &op->copy_from.osd_data); 2011 break; 2012 2013 /* reply */ 2014 case CEPH_OSD_OP_STAT: 2015 ceph_osdc_msg_data_add(reply_msg, 2016 &op->raw_data_in); 2017 break; 2018 case CEPH_OSD_OP_READ: 2019 ceph_osdc_msg_data_add(reply_msg, 2020 &op->extent.osd_data); 2021 break; 2022 case CEPH_OSD_OP_LIST_WATCHERS: 2023 ceph_osdc_msg_data_add(reply_msg, 2024 &op->list_watchers.response_data); 2025 break; 2026 2027 /* both */ 2028 case CEPH_OSD_OP_CALL: 2029 WARN_ON(op->indata_len != op->cls.class_len + 2030 op->cls.method_len + 2031 op->cls.indata_len); 2032 ceph_osdc_msg_data_add(request_msg, 2033 &op->cls.request_info); 2034 /* optional, can be NONE */ 2035 ceph_osdc_msg_data_add(request_msg, 2036 &op->cls.request_data); 2037 /* optional, can be NONE */ 2038 ceph_osdc_msg_data_add(reply_msg, 2039 &op->cls.response_data); 2040 break; 2041 case CEPH_OSD_OP_NOTIFY: 2042 ceph_osdc_msg_data_add(request_msg, 2043 &op->notify.request_data); 2044 ceph_osdc_msg_data_add(reply_msg, 2045 &op->notify.response_data); 2046 break; 2047 } 2048 } 2049 } 2050 2051 static void encode_pgid(void **p, const struct ceph_pg *pgid) 2052 { 2053 ceph_encode_8(p, 1); 2054 ceph_encode_64(p, pgid->pool); 2055 ceph_encode_32(p, pgid->seed); 2056 ceph_encode_32(p, -1); /* preferred */ 2057 } 2058 2059 static void encode_spgid(void **p, const struct ceph_spg *spgid) 2060 { 2061 ceph_start_encoding(p, 1, 1, CEPH_PGID_ENCODING_LEN + 1); 2062 encode_pgid(p, &spgid->pgid); 2063 ceph_encode_8(p, spgid->shard); 2064 } 2065 2066 static void encode_oloc(void **p, void *end, 2067 const struct ceph_object_locator *oloc) 2068 { 2069 ceph_start_encoding(p, 5, 4, ceph_oloc_encoding_size(oloc)); 2070 ceph_encode_64(p, oloc->pool); 2071 ceph_encode_32(p, -1); /* preferred */ 2072 ceph_encode_32(p, 0); /* key len */ 2073 if (oloc->pool_ns) 2074 ceph_encode_string(p, end, oloc->pool_ns->str, 2075 oloc->pool_ns->len); 2076 else 2077 ceph_encode_32(p, 0); 2078 } 2079 2080 static void encode_request_partial(struct ceph_osd_request *req, 2081 struct ceph_msg *msg) 2082 { 2083 void *p = msg->front.iov_base; 2084 void *const end = p + msg->front_alloc_len; 2085 u32 data_len = 0; 2086 int i; 2087 2088 if (req->r_flags & CEPH_OSD_FLAG_WRITE) { 2089 /* snapshots aren't writeable */ 2090 WARN_ON(req->r_snapid != CEPH_NOSNAP); 2091 } else { 2092 WARN_ON(req->r_mtime.tv_sec || req->r_mtime.tv_nsec || 2093 req->r_data_offset || req->r_snapc); 2094 } 2095 2096 setup_request_data(req); 2097 2098 encode_spgid(&p, &req->r_t.spgid); /* actual spg */ 2099 ceph_encode_32(&p, req->r_t.pgid.seed); /* raw hash */ 2100 ceph_encode_32(&p, req->r_osdc->osdmap->epoch); 2101 ceph_encode_32(&p, req->r_flags); 2102 2103 /* reqid */ 2104 ceph_start_encoding(&p, 2, 2, sizeof(struct ceph_osd_reqid)); 2105 memset(p, 0, sizeof(struct ceph_osd_reqid)); 2106 p += sizeof(struct ceph_osd_reqid); 2107 2108 /* trace */ 2109 memset(p, 0, sizeof(struct ceph_blkin_trace_info)); 2110 p += sizeof(struct ceph_blkin_trace_info); 2111 2112 ceph_encode_32(&p, 0); /* client_inc, always 0 */ 2113 ceph_encode_timespec64(p, &req->r_mtime); 2114 p += sizeof(struct ceph_timespec); 2115 2116 encode_oloc(&p, end, &req->r_t.target_oloc); 2117 ceph_encode_string(&p, end, req->r_t.target_oid.name, 2118 req->r_t.target_oid.name_len); 2119 2120 /* ops, can imply data */ 2121 ceph_encode_16(&p, req->r_num_ops); 2122 for (i = 0; i < req->r_num_ops; i++) { 2123 data_len += osd_req_encode_op(p, &req->r_ops[i]); 2124 p += sizeof(struct ceph_osd_op); 2125 } 2126 2127 ceph_encode_64(&p, req->r_snapid); /* snapid */ 2128 if (req->r_snapc) { 2129 ceph_encode_64(&p, req->r_snapc->seq); 2130 ceph_encode_32(&p, req->r_snapc->num_snaps); 2131 for (i = 0; i < req->r_snapc->num_snaps; i++) 2132 ceph_encode_64(&p, req->r_snapc->snaps[i]); 2133 } else { 2134 ceph_encode_64(&p, 0); /* snap_seq */ 2135 ceph_encode_32(&p, 0); /* snaps len */ 2136 } 2137 2138 ceph_encode_32(&p, req->r_attempts); /* retry_attempt */ 2139 BUG_ON(p > end - 8); /* space for features */ 2140 2141 msg->hdr.version = cpu_to_le16(8); /* MOSDOp v8 */ 2142 /* front_len is finalized in encode_request_finish() */ 2143 msg->front.iov_len = p - msg->front.iov_base; 2144 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len); 2145 msg->hdr.data_len = cpu_to_le32(data_len); 2146 /* 2147 * The header "data_off" is a hint to the receiver allowing it 2148 * to align received data into its buffers such that there's no 2149 * need to re-copy it before writing it to disk (direct I/O). 2150 */ 2151 msg->hdr.data_off = cpu_to_le16(req->r_data_offset); 2152 2153 dout("%s req %p msg %p oid %s oid_len %d\n", __func__, req, msg, 2154 req->r_t.target_oid.name, req->r_t.target_oid.name_len); 2155 } 2156 2157 static void encode_request_finish(struct ceph_msg *msg) 2158 { 2159 void *p = msg->front.iov_base; 2160 void *const partial_end = p + msg->front.iov_len; 2161 void *const end = p + msg->front_alloc_len; 2162 2163 if (CEPH_HAVE_FEATURE(msg->con->peer_features, RESEND_ON_SPLIT)) { 2164 /* luminous OSD -- encode features and be done */ 2165 p = partial_end; 2166 ceph_encode_64(&p, msg->con->peer_features); 2167 } else { 2168 struct { 2169 char spgid[CEPH_ENCODING_START_BLK_LEN + 2170 CEPH_PGID_ENCODING_LEN + 1]; 2171 __le32 hash; 2172 __le32 epoch; 2173 __le32 flags; 2174 char reqid[CEPH_ENCODING_START_BLK_LEN + 2175 sizeof(struct ceph_osd_reqid)]; 2176 char trace[sizeof(struct ceph_blkin_trace_info)]; 2177 __le32 client_inc; 2178 struct ceph_timespec mtime; 2179 } __packed head; 2180 struct ceph_pg pgid; 2181 void *oloc, *oid, *tail; 2182 int oloc_len, oid_len, tail_len; 2183 int len; 2184 2185 /* 2186 * Pre-luminous OSD -- reencode v8 into v4 using @head 2187 * as a temporary buffer. Encode the raw PG; the rest 2188 * is just a matter of moving oloc, oid and tail blobs 2189 * around. 2190 */ 2191 memcpy(&head, p, sizeof(head)); 2192 p += sizeof(head); 2193 2194 oloc = p; 2195 p += CEPH_ENCODING_START_BLK_LEN; 2196 pgid.pool = ceph_decode_64(&p); 2197 p += 4 + 4; /* preferred, key len */ 2198 len = ceph_decode_32(&p); 2199 p += len; /* nspace */ 2200 oloc_len = p - oloc; 2201 2202 oid = p; 2203 len = ceph_decode_32(&p); 2204 p += len; 2205 oid_len = p - oid; 2206 2207 tail = p; 2208 tail_len = partial_end - p; 2209 2210 p = msg->front.iov_base; 2211 ceph_encode_copy(&p, &head.client_inc, sizeof(head.client_inc)); 2212 ceph_encode_copy(&p, &head.epoch, sizeof(head.epoch)); 2213 ceph_encode_copy(&p, &head.flags, sizeof(head.flags)); 2214 ceph_encode_copy(&p, &head.mtime, sizeof(head.mtime)); 2215 2216 /* reassert_version */ 2217 memset(p, 0, sizeof(struct ceph_eversion)); 2218 p += sizeof(struct ceph_eversion); 2219 2220 BUG_ON(p >= oloc); 2221 memmove(p, oloc, oloc_len); 2222 p += oloc_len; 2223 2224 pgid.seed = le32_to_cpu(head.hash); 2225 encode_pgid(&p, &pgid); /* raw pg */ 2226 2227 BUG_ON(p >= oid); 2228 memmove(p, oid, oid_len); 2229 p += oid_len; 2230 2231 /* tail -- ops, snapid, snapc, retry_attempt */ 2232 BUG_ON(p >= tail); 2233 memmove(p, tail, tail_len); 2234 p += tail_len; 2235 2236 msg->hdr.version = cpu_to_le16(4); /* MOSDOp v4 */ 2237 } 2238 2239 BUG_ON(p > end); 2240 msg->front.iov_len = p - msg->front.iov_base; 2241 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len); 2242 2243 dout("%s msg %p tid %llu %u+%u+%u v%d\n", __func__, msg, 2244 le64_to_cpu(msg->hdr.tid), le32_to_cpu(msg->hdr.front_len), 2245 le32_to_cpu(msg->hdr.middle_len), le32_to_cpu(msg->hdr.data_len), 2246 le16_to_cpu(msg->hdr.version)); 2247 } 2248 2249 /* 2250 * @req has to be assigned a tid and registered. 2251 */ 2252 static void send_request(struct ceph_osd_request *req) 2253 { 2254 struct ceph_osd *osd = req->r_osd; 2255 2256 verify_osd_locked(osd); 2257 WARN_ON(osd->o_osd != req->r_t.osd); 2258 2259 /* backoff? */ 2260 if (should_plug_request(req)) 2261 return; 2262 2263 /* 2264 * We may have a previously queued request message hanging 2265 * around. Cancel it to avoid corrupting the msgr. 2266 */ 2267 if (req->r_sent) 2268 ceph_msg_revoke(req->r_request); 2269 2270 req->r_flags |= CEPH_OSD_FLAG_KNOWN_REDIR; 2271 if (req->r_attempts) 2272 req->r_flags |= CEPH_OSD_FLAG_RETRY; 2273 else 2274 WARN_ON(req->r_flags & CEPH_OSD_FLAG_RETRY); 2275 2276 encode_request_partial(req, req->r_request); 2277 2278 dout("%s req %p tid %llu to pgid %llu.%x spgid %llu.%xs%d osd%d e%u flags 0x%x attempt %d\n", 2279 __func__, req, req->r_tid, req->r_t.pgid.pool, req->r_t.pgid.seed, 2280 req->r_t.spgid.pgid.pool, req->r_t.spgid.pgid.seed, 2281 req->r_t.spgid.shard, osd->o_osd, req->r_t.epoch, req->r_flags, 2282 req->r_attempts); 2283 2284 req->r_t.paused = false; 2285 req->r_stamp = jiffies; 2286 req->r_attempts++; 2287 2288 req->r_sent = osd->o_incarnation; 2289 req->r_request->hdr.tid = cpu_to_le64(req->r_tid); 2290 ceph_con_send(&osd->o_con, ceph_msg_get(req->r_request)); 2291 } 2292 2293 static void maybe_request_map(struct ceph_osd_client *osdc) 2294 { 2295 bool continuous = false; 2296 2297 verify_osdc_locked(osdc); 2298 WARN_ON(!osdc->osdmap->epoch); 2299 2300 if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || 2301 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD) || 2302 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) { 2303 dout("%s osdc %p continuous\n", __func__, osdc); 2304 continuous = true; 2305 } else { 2306 dout("%s osdc %p onetime\n", __func__, osdc); 2307 } 2308 2309 if (ceph_monc_want_map(&osdc->client->monc, CEPH_SUB_OSDMAP, 2310 osdc->osdmap->epoch + 1, continuous)) 2311 ceph_monc_renew_subs(&osdc->client->monc); 2312 } 2313 2314 static void complete_request(struct ceph_osd_request *req, int err); 2315 static void send_map_check(struct ceph_osd_request *req); 2316 2317 static void __submit_request(struct ceph_osd_request *req, bool wrlocked) 2318 { 2319 struct ceph_osd_client *osdc = req->r_osdc; 2320 struct ceph_osd *osd; 2321 enum calc_target_result ct_res; 2322 int err = 0; 2323 bool need_send = false; 2324 bool promoted = false; 2325 2326 WARN_ON(req->r_tid); 2327 dout("%s req %p wrlocked %d\n", __func__, req, wrlocked); 2328 2329 again: 2330 ct_res = calc_target(osdc, &req->r_t, false); 2331 if (ct_res == CALC_TARGET_POOL_DNE && !wrlocked) 2332 goto promote; 2333 2334 osd = lookup_create_osd(osdc, req->r_t.osd, wrlocked); 2335 if (IS_ERR(osd)) { 2336 WARN_ON(PTR_ERR(osd) != -EAGAIN || wrlocked); 2337 goto promote; 2338 } 2339 2340 if (osdc->abort_err) { 2341 dout("req %p abort_err %d\n", req, osdc->abort_err); 2342 err = osdc->abort_err; 2343 } else if (osdc->osdmap->epoch < osdc->epoch_barrier) { 2344 dout("req %p epoch %u barrier %u\n", req, osdc->osdmap->epoch, 2345 osdc->epoch_barrier); 2346 req->r_t.paused = true; 2347 maybe_request_map(osdc); 2348 } else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) && 2349 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) { 2350 dout("req %p pausewr\n", req); 2351 req->r_t.paused = true; 2352 maybe_request_map(osdc); 2353 } else if ((req->r_flags & CEPH_OSD_FLAG_READ) && 2354 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) { 2355 dout("req %p pauserd\n", req); 2356 req->r_t.paused = true; 2357 maybe_request_map(osdc); 2358 } else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) && 2359 !(req->r_flags & (CEPH_OSD_FLAG_FULL_TRY | 2360 CEPH_OSD_FLAG_FULL_FORCE)) && 2361 (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || 2362 pool_full(osdc, req->r_t.base_oloc.pool))) { 2363 dout("req %p full/pool_full\n", req); 2364 if (ceph_test_opt(osdc->client, ABORT_ON_FULL)) { 2365 err = -ENOSPC; 2366 } else { 2367 if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL)) 2368 pr_warn_ratelimited("cluster is full (osdmap FULL)\n"); 2369 else 2370 pr_warn_ratelimited("pool %lld is full or reached quota\n", 2371 req->r_t.base_oloc.pool); 2372 req->r_t.paused = true; 2373 maybe_request_map(osdc); 2374 } 2375 } else if (!osd_homeless(osd)) { 2376 need_send = true; 2377 } else { 2378 maybe_request_map(osdc); 2379 } 2380 2381 mutex_lock(&osd->lock); 2382 /* 2383 * Assign the tid atomically with send_request() to protect 2384 * multiple writes to the same object from racing with each 2385 * other, resulting in out of order ops on the OSDs. 2386 */ 2387 req->r_tid = atomic64_inc_return(&osdc->last_tid); 2388 link_request(osd, req); 2389 if (need_send) 2390 send_request(req); 2391 else if (err) 2392 complete_request(req, err); 2393 mutex_unlock(&osd->lock); 2394 2395 if (!err && ct_res == CALC_TARGET_POOL_DNE) 2396 send_map_check(req); 2397 2398 if (promoted) 2399 downgrade_write(&osdc->lock); 2400 return; 2401 2402 promote: 2403 up_read(&osdc->lock); 2404 down_write(&osdc->lock); 2405 wrlocked = true; 2406 promoted = true; 2407 goto again; 2408 } 2409 2410 static void account_request(struct ceph_osd_request *req) 2411 { 2412 WARN_ON(req->r_flags & (CEPH_OSD_FLAG_ACK | CEPH_OSD_FLAG_ONDISK)); 2413 WARN_ON(!(req->r_flags & (CEPH_OSD_FLAG_READ | CEPH_OSD_FLAG_WRITE))); 2414 2415 req->r_flags |= CEPH_OSD_FLAG_ONDISK; 2416 atomic_inc(&req->r_osdc->num_requests); 2417 2418 req->r_start_stamp = jiffies; 2419 req->r_start_latency = ktime_get(); 2420 } 2421 2422 static void submit_request(struct ceph_osd_request *req, bool wrlocked) 2423 { 2424 ceph_osdc_get_request(req); 2425 account_request(req); 2426 __submit_request(req, wrlocked); 2427 } 2428 2429 static void finish_request(struct ceph_osd_request *req) 2430 { 2431 struct ceph_osd_client *osdc = req->r_osdc; 2432 2433 WARN_ON(lookup_request_mc(&osdc->map_checks, req->r_tid)); 2434 dout("%s req %p tid %llu\n", __func__, req, req->r_tid); 2435 2436 req->r_end_latency = ktime_get(); 2437 2438 if (req->r_osd) 2439 unlink_request(req->r_osd, req); 2440 atomic_dec(&osdc->num_requests); 2441 2442 /* 2443 * If an OSD has failed or returned and a request has been sent 2444 * twice, it's possible to get a reply and end up here while the 2445 * request message is queued for delivery. We will ignore the 2446 * reply, so not a big deal, but better to try and catch it. 2447 */ 2448 ceph_msg_revoke(req->r_request); 2449 ceph_msg_revoke_incoming(req->r_reply); 2450 } 2451 2452 static void __complete_request(struct ceph_osd_request *req) 2453 { 2454 dout("%s req %p tid %llu cb %ps result %d\n", __func__, req, 2455 req->r_tid, req->r_callback, req->r_result); 2456 2457 if (req->r_callback) 2458 req->r_callback(req); 2459 complete_all(&req->r_completion); 2460 ceph_osdc_put_request(req); 2461 } 2462 2463 static void complete_request_workfn(struct work_struct *work) 2464 { 2465 struct ceph_osd_request *req = 2466 container_of(work, struct ceph_osd_request, r_complete_work); 2467 2468 __complete_request(req); 2469 } 2470 2471 /* 2472 * This is open-coded in handle_reply(). 2473 */ 2474 static void complete_request(struct ceph_osd_request *req, int err) 2475 { 2476 dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err); 2477 2478 req->r_result = err; 2479 finish_request(req); 2480 2481 INIT_WORK(&req->r_complete_work, complete_request_workfn); 2482 queue_work(req->r_osdc->completion_wq, &req->r_complete_work); 2483 } 2484 2485 static void cancel_map_check(struct ceph_osd_request *req) 2486 { 2487 struct ceph_osd_client *osdc = req->r_osdc; 2488 struct ceph_osd_request *lookup_req; 2489 2490 verify_osdc_wrlocked(osdc); 2491 2492 lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid); 2493 if (!lookup_req) 2494 return; 2495 2496 WARN_ON(lookup_req != req); 2497 erase_request_mc(&osdc->map_checks, req); 2498 ceph_osdc_put_request(req); 2499 } 2500 2501 static void cancel_request(struct ceph_osd_request *req) 2502 { 2503 dout("%s req %p tid %llu\n", __func__, req, req->r_tid); 2504 2505 cancel_map_check(req); 2506 finish_request(req); 2507 complete_all(&req->r_completion); 2508 ceph_osdc_put_request(req); 2509 } 2510 2511 static void abort_request(struct ceph_osd_request *req, int err) 2512 { 2513 dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err); 2514 2515 cancel_map_check(req); 2516 complete_request(req, err); 2517 } 2518 2519 static int abort_fn(struct ceph_osd_request *req, void *arg) 2520 { 2521 int err = *(int *)arg; 2522 2523 abort_request(req, err); 2524 return 0; /* continue iteration */ 2525 } 2526 2527 /* 2528 * Abort all in-flight requests with @err and arrange for all future 2529 * requests to be failed immediately. 2530 */ 2531 void ceph_osdc_abort_requests(struct ceph_osd_client *osdc, int err) 2532 { 2533 dout("%s osdc %p err %d\n", __func__, osdc, err); 2534 down_write(&osdc->lock); 2535 for_each_request(osdc, abort_fn, &err); 2536 osdc->abort_err = err; 2537 up_write(&osdc->lock); 2538 } 2539 EXPORT_SYMBOL(ceph_osdc_abort_requests); 2540 2541 void ceph_osdc_clear_abort_err(struct ceph_osd_client *osdc) 2542 { 2543 down_write(&osdc->lock); 2544 osdc->abort_err = 0; 2545 up_write(&osdc->lock); 2546 } 2547 EXPORT_SYMBOL(ceph_osdc_clear_abort_err); 2548 2549 static void update_epoch_barrier(struct ceph_osd_client *osdc, u32 eb) 2550 { 2551 if (likely(eb > osdc->epoch_barrier)) { 2552 dout("updating epoch_barrier from %u to %u\n", 2553 osdc->epoch_barrier, eb); 2554 osdc->epoch_barrier = eb; 2555 /* Request map if we're not to the barrier yet */ 2556 if (eb > osdc->osdmap->epoch) 2557 maybe_request_map(osdc); 2558 } 2559 } 2560 2561 void ceph_osdc_update_epoch_barrier(struct ceph_osd_client *osdc, u32 eb) 2562 { 2563 down_read(&osdc->lock); 2564 if (unlikely(eb > osdc->epoch_barrier)) { 2565 up_read(&osdc->lock); 2566 down_write(&osdc->lock); 2567 update_epoch_barrier(osdc, eb); 2568 up_write(&osdc->lock); 2569 } else { 2570 up_read(&osdc->lock); 2571 } 2572 } 2573 EXPORT_SYMBOL(ceph_osdc_update_epoch_barrier); 2574 2575 /* 2576 * We can end up releasing caps as a result of abort_request(). 2577 * In that case, we probably want to ensure that the cap release message 2578 * has an updated epoch barrier in it, so set the epoch barrier prior to 2579 * aborting the first request. 2580 */ 2581 static int abort_on_full_fn(struct ceph_osd_request *req, void *arg) 2582 { 2583 struct ceph_osd_client *osdc = req->r_osdc; 2584 bool *victims = arg; 2585 2586 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) && 2587 (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || 2588 pool_full(osdc, req->r_t.base_oloc.pool))) { 2589 if (!*victims) { 2590 update_epoch_barrier(osdc, osdc->osdmap->epoch); 2591 *victims = true; 2592 } 2593 abort_request(req, -ENOSPC); 2594 } 2595 2596 return 0; /* continue iteration */ 2597 } 2598 2599 /* 2600 * Drop all pending requests that are stalled waiting on a full condition to 2601 * clear, and complete them with ENOSPC as the return code. Set the 2602 * osdc->epoch_barrier to the latest map epoch that we've seen if any were 2603 * cancelled. 2604 */ 2605 static void ceph_osdc_abort_on_full(struct ceph_osd_client *osdc) 2606 { 2607 bool victims = false; 2608 2609 if (ceph_test_opt(osdc->client, ABORT_ON_FULL) && 2610 (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || have_pool_full(osdc))) 2611 for_each_request(osdc, abort_on_full_fn, &victims); 2612 } 2613 2614 static void check_pool_dne(struct ceph_osd_request *req) 2615 { 2616 struct ceph_osd_client *osdc = req->r_osdc; 2617 struct ceph_osdmap *map = osdc->osdmap; 2618 2619 verify_osdc_wrlocked(osdc); 2620 WARN_ON(!map->epoch); 2621 2622 if (req->r_attempts) { 2623 /* 2624 * We sent a request earlier, which means that 2625 * previously the pool existed, and now it does not 2626 * (i.e., it was deleted). 2627 */ 2628 req->r_map_dne_bound = map->epoch; 2629 dout("%s req %p tid %llu pool disappeared\n", __func__, req, 2630 req->r_tid); 2631 } else { 2632 dout("%s req %p tid %llu map_dne_bound %u have %u\n", __func__, 2633 req, req->r_tid, req->r_map_dne_bound, map->epoch); 2634 } 2635 2636 if (req->r_map_dne_bound) { 2637 if (map->epoch >= req->r_map_dne_bound) { 2638 /* we had a new enough map */ 2639 pr_info_ratelimited("tid %llu pool does not exist\n", 2640 req->r_tid); 2641 complete_request(req, -ENOENT); 2642 } 2643 } else { 2644 send_map_check(req); 2645 } 2646 } 2647 2648 static void map_check_cb(struct ceph_mon_generic_request *greq) 2649 { 2650 struct ceph_osd_client *osdc = &greq->monc->client->osdc; 2651 struct ceph_osd_request *req; 2652 u64 tid = greq->private_data; 2653 2654 WARN_ON(greq->result || !greq->u.newest); 2655 2656 down_write(&osdc->lock); 2657 req = lookup_request_mc(&osdc->map_checks, tid); 2658 if (!req) { 2659 dout("%s tid %llu dne\n", __func__, tid); 2660 goto out_unlock; 2661 } 2662 2663 dout("%s req %p tid %llu map_dne_bound %u newest %llu\n", __func__, 2664 req, req->r_tid, req->r_map_dne_bound, greq->u.newest); 2665 if (!req->r_map_dne_bound) 2666 req->r_map_dne_bound = greq->u.newest; 2667 erase_request_mc(&osdc->map_checks, req); 2668 check_pool_dne(req); 2669 2670 ceph_osdc_put_request(req); 2671 out_unlock: 2672 up_write(&osdc->lock); 2673 } 2674 2675 static void send_map_check(struct ceph_osd_request *req) 2676 { 2677 struct ceph_osd_client *osdc = req->r_osdc; 2678 struct ceph_osd_request *lookup_req; 2679 int ret; 2680 2681 verify_osdc_wrlocked(osdc); 2682 2683 lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid); 2684 if (lookup_req) { 2685 WARN_ON(lookup_req != req); 2686 return; 2687 } 2688 2689 ceph_osdc_get_request(req); 2690 insert_request_mc(&osdc->map_checks, req); 2691 ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap", 2692 map_check_cb, req->r_tid); 2693 WARN_ON(ret); 2694 } 2695 2696 /* 2697 * lingering requests, watch/notify v2 infrastructure 2698 */ 2699 static void linger_release(struct kref *kref) 2700 { 2701 struct ceph_osd_linger_request *lreq = 2702 container_of(kref, struct ceph_osd_linger_request, kref); 2703 2704 dout("%s lreq %p reg_req %p ping_req %p\n", __func__, lreq, 2705 lreq->reg_req, lreq->ping_req); 2706 WARN_ON(!RB_EMPTY_NODE(&lreq->node)); 2707 WARN_ON(!RB_EMPTY_NODE(&lreq->osdc_node)); 2708 WARN_ON(!RB_EMPTY_NODE(&lreq->mc_node)); 2709 WARN_ON(!list_empty(&lreq->scan_item)); 2710 WARN_ON(!list_empty(&lreq->pending_lworks)); 2711 WARN_ON(lreq->osd); 2712 2713 if (lreq->request_pl) 2714 ceph_pagelist_release(lreq->request_pl); 2715 if (lreq->notify_id_pages) 2716 ceph_release_page_vector(lreq->notify_id_pages, 1); 2717 2718 ceph_osdc_put_request(lreq->reg_req); 2719 ceph_osdc_put_request(lreq->ping_req); 2720 target_destroy(&lreq->t); 2721 kfree(lreq); 2722 } 2723 2724 static void linger_put(struct ceph_osd_linger_request *lreq) 2725 { 2726 if (lreq) 2727 kref_put(&lreq->kref, linger_release); 2728 } 2729 2730 static struct ceph_osd_linger_request * 2731 linger_get(struct ceph_osd_linger_request *lreq) 2732 { 2733 kref_get(&lreq->kref); 2734 return lreq; 2735 } 2736 2737 static struct ceph_osd_linger_request * 2738 linger_alloc(struct ceph_osd_client *osdc) 2739 { 2740 struct ceph_osd_linger_request *lreq; 2741 2742 lreq = kzalloc(sizeof(*lreq), GFP_NOIO); 2743 if (!lreq) 2744 return NULL; 2745 2746 kref_init(&lreq->kref); 2747 mutex_init(&lreq->lock); 2748 RB_CLEAR_NODE(&lreq->node); 2749 RB_CLEAR_NODE(&lreq->osdc_node); 2750 RB_CLEAR_NODE(&lreq->mc_node); 2751 INIT_LIST_HEAD(&lreq->scan_item); 2752 INIT_LIST_HEAD(&lreq->pending_lworks); 2753 init_completion(&lreq->reg_commit_wait); 2754 init_completion(&lreq->notify_finish_wait); 2755 2756 lreq->osdc = osdc; 2757 target_init(&lreq->t); 2758 2759 dout("%s lreq %p\n", __func__, lreq); 2760 return lreq; 2761 } 2762 2763 DEFINE_RB_INSDEL_FUNCS(linger, struct ceph_osd_linger_request, linger_id, node) 2764 DEFINE_RB_FUNCS(linger_osdc, struct ceph_osd_linger_request, linger_id, osdc_node) 2765 DEFINE_RB_FUNCS(linger_mc, struct ceph_osd_linger_request, linger_id, mc_node) 2766 2767 /* 2768 * Create linger request <-> OSD session relation. 2769 * 2770 * @lreq has to be registered, @osd may be homeless. 2771 */ 2772 static void link_linger(struct ceph_osd *osd, 2773 struct ceph_osd_linger_request *lreq) 2774 { 2775 verify_osd_locked(osd); 2776 WARN_ON(!lreq->linger_id || lreq->osd); 2777 dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd, 2778 osd->o_osd, lreq, lreq->linger_id); 2779 2780 if (!osd_homeless(osd)) 2781 __remove_osd_from_lru(osd); 2782 else 2783 atomic_inc(&osd->o_osdc->num_homeless); 2784 2785 get_osd(osd); 2786 insert_linger(&osd->o_linger_requests, lreq); 2787 lreq->osd = osd; 2788 } 2789 2790 static void unlink_linger(struct ceph_osd *osd, 2791 struct ceph_osd_linger_request *lreq) 2792 { 2793 verify_osd_locked(osd); 2794 WARN_ON(lreq->osd != osd); 2795 dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd, 2796 osd->o_osd, lreq, lreq->linger_id); 2797 2798 lreq->osd = NULL; 2799 erase_linger(&osd->o_linger_requests, lreq); 2800 put_osd(osd); 2801 2802 if (!osd_homeless(osd)) 2803 maybe_move_osd_to_lru(osd); 2804 else 2805 atomic_dec(&osd->o_osdc->num_homeless); 2806 } 2807 2808 static bool __linger_registered(struct ceph_osd_linger_request *lreq) 2809 { 2810 verify_osdc_locked(lreq->osdc); 2811 2812 return !RB_EMPTY_NODE(&lreq->osdc_node); 2813 } 2814 2815 static bool linger_registered(struct ceph_osd_linger_request *lreq) 2816 { 2817 struct ceph_osd_client *osdc = lreq->osdc; 2818 bool registered; 2819 2820 down_read(&osdc->lock); 2821 registered = __linger_registered(lreq); 2822 up_read(&osdc->lock); 2823 2824 return registered; 2825 } 2826 2827 static void linger_register(struct ceph_osd_linger_request *lreq) 2828 { 2829 struct ceph_osd_client *osdc = lreq->osdc; 2830 2831 verify_osdc_wrlocked(osdc); 2832 WARN_ON(lreq->linger_id); 2833 2834 linger_get(lreq); 2835 lreq->linger_id = ++osdc->last_linger_id; 2836 insert_linger_osdc(&osdc->linger_requests, lreq); 2837 } 2838 2839 static void linger_unregister(struct ceph_osd_linger_request *lreq) 2840 { 2841 struct ceph_osd_client *osdc = lreq->osdc; 2842 2843 verify_osdc_wrlocked(osdc); 2844 2845 erase_linger_osdc(&osdc->linger_requests, lreq); 2846 linger_put(lreq); 2847 } 2848 2849 static void cancel_linger_request(struct ceph_osd_request *req) 2850 { 2851 struct ceph_osd_linger_request *lreq = req->r_priv; 2852 2853 WARN_ON(!req->r_linger); 2854 cancel_request(req); 2855 linger_put(lreq); 2856 } 2857 2858 struct linger_work { 2859 struct work_struct work; 2860 struct ceph_osd_linger_request *lreq; 2861 struct list_head pending_item; 2862 unsigned long queued_stamp; 2863 2864 union { 2865 struct { 2866 u64 notify_id; 2867 u64 notifier_id; 2868 void *payload; /* points into @msg front */ 2869 size_t payload_len; 2870 2871 struct ceph_msg *msg; /* for ceph_msg_put() */ 2872 } notify; 2873 struct { 2874 int err; 2875 } error; 2876 }; 2877 }; 2878 2879 static struct linger_work *lwork_alloc(struct ceph_osd_linger_request *lreq, 2880 work_func_t workfn) 2881 { 2882 struct linger_work *lwork; 2883 2884 lwork = kzalloc(sizeof(*lwork), GFP_NOIO); 2885 if (!lwork) 2886 return NULL; 2887 2888 INIT_WORK(&lwork->work, workfn); 2889 INIT_LIST_HEAD(&lwork->pending_item); 2890 lwork->lreq = linger_get(lreq); 2891 2892 return lwork; 2893 } 2894 2895 static void lwork_free(struct linger_work *lwork) 2896 { 2897 struct ceph_osd_linger_request *lreq = lwork->lreq; 2898 2899 mutex_lock(&lreq->lock); 2900 list_del(&lwork->pending_item); 2901 mutex_unlock(&lreq->lock); 2902 2903 linger_put(lreq); 2904 kfree(lwork); 2905 } 2906 2907 static void lwork_queue(struct linger_work *lwork) 2908 { 2909 struct ceph_osd_linger_request *lreq = lwork->lreq; 2910 struct ceph_osd_client *osdc = lreq->osdc; 2911 2912 verify_lreq_locked(lreq); 2913 WARN_ON(!list_empty(&lwork->pending_item)); 2914 2915 lwork->queued_stamp = jiffies; 2916 list_add_tail(&lwork->pending_item, &lreq->pending_lworks); 2917 queue_work(osdc->notify_wq, &lwork->work); 2918 } 2919 2920 static void do_watch_notify(struct work_struct *w) 2921 { 2922 struct linger_work *lwork = container_of(w, struct linger_work, work); 2923 struct ceph_osd_linger_request *lreq = lwork->lreq; 2924 2925 if (!linger_registered(lreq)) { 2926 dout("%s lreq %p not registered\n", __func__, lreq); 2927 goto out; 2928 } 2929 2930 WARN_ON(!lreq->is_watch); 2931 dout("%s lreq %p notify_id %llu notifier_id %llu payload_len %zu\n", 2932 __func__, lreq, lwork->notify.notify_id, lwork->notify.notifier_id, 2933 lwork->notify.payload_len); 2934 lreq->wcb(lreq->data, lwork->notify.notify_id, lreq->linger_id, 2935 lwork->notify.notifier_id, lwork->notify.payload, 2936 lwork->notify.payload_len); 2937 2938 out: 2939 ceph_msg_put(lwork->notify.msg); 2940 lwork_free(lwork); 2941 } 2942 2943 static void do_watch_error(struct work_struct *w) 2944 { 2945 struct linger_work *lwork = container_of(w, struct linger_work, work); 2946 struct ceph_osd_linger_request *lreq = lwork->lreq; 2947 2948 if (!linger_registered(lreq)) { 2949 dout("%s lreq %p not registered\n", __func__, lreq); 2950 goto out; 2951 } 2952 2953 dout("%s lreq %p err %d\n", __func__, lreq, lwork->error.err); 2954 lreq->errcb(lreq->data, lreq->linger_id, lwork->error.err); 2955 2956 out: 2957 lwork_free(lwork); 2958 } 2959 2960 static void queue_watch_error(struct ceph_osd_linger_request *lreq) 2961 { 2962 struct linger_work *lwork; 2963 2964 lwork = lwork_alloc(lreq, do_watch_error); 2965 if (!lwork) { 2966 pr_err("failed to allocate error-lwork\n"); 2967 return; 2968 } 2969 2970 lwork->error.err = lreq->last_error; 2971 lwork_queue(lwork); 2972 } 2973 2974 static void linger_reg_commit_complete(struct ceph_osd_linger_request *lreq, 2975 int result) 2976 { 2977 if (!completion_done(&lreq->reg_commit_wait)) { 2978 lreq->reg_commit_error = (result <= 0 ? result : 0); 2979 complete_all(&lreq->reg_commit_wait); 2980 } 2981 } 2982 2983 static void linger_commit_cb(struct ceph_osd_request *req) 2984 { 2985 struct ceph_osd_linger_request *lreq = req->r_priv; 2986 2987 mutex_lock(&lreq->lock); 2988 if (req != lreq->reg_req) { 2989 dout("%s lreq %p linger_id %llu unknown req (%p != %p)\n", 2990 __func__, lreq, lreq->linger_id, req, lreq->reg_req); 2991 goto out; 2992 } 2993 2994 dout("%s lreq %p linger_id %llu result %d\n", __func__, lreq, 2995 lreq->linger_id, req->r_result); 2996 linger_reg_commit_complete(lreq, req->r_result); 2997 lreq->committed = true; 2998 2999 if (!lreq->is_watch) { 3000 struct ceph_osd_data *osd_data = 3001 osd_req_op_data(req, 0, notify, response_data); 3002 void *p = page_address(osd_data->pages[0]); 3003 3004 WARN_ON(req->r_ops[0].op != CEPH_OSD_OP_NOTIFY || 3005 osd_data->type != CEPH_OSD_DATA_TYPE_PAGES); 3006 3007 /* make note of the notify_id */ 3008 if (req->r_ops[0].outdata_len >= sizeof(u64)) { 3009 lreq->notify_id = ceph_decode_64(&p); 3010 dout("lreq %p notify_id %llu\n", lreq, 3011 lreq->notify_id); 3012 } else { 3013 dout("lreq %p no notify_id\n", lreq); 3014 } 3015 } 3016 3017 out: 3018 mutex_unlock(&lreq->lock); 3019 linger_put(lreq); 3020 } 3021 3022 static int normalize_watch_error(int err) 3023 { 3024 /* 3025 * Translate ENOENT -> ENOTCONN so that a delete->disconnection 3026 * notification and a failure to reconnect because we raced with 3027 * the delete appear the same to the user. 3028 */ 3029 if (err == -ENOENT) 3030 err = -ENOTCONN; 3031 3032 return err; 3033 } 3034 3035 static void linger_reconnect_cb(struct ceph_osd_request *req) 3036 { 3037 struct ceph_osd_linger_request *lreq = req->r_priv; 3038 3039 mutex_lock(&lreq->lock); 3040 if (req != lreq->reg_req) { 3041 dout("%s lreq %p linger_id %llu unknown req (%p != %p)\n", 3042 __func__, lreq, lreq->linger_id, req, lreq->reg_req); 3043 goto out; 3044 } 3045 3046 dout("%s lreq %p linger_id %llu result %d last_error %d\n", __func__, 3047 lreq, lreq->linger_id, req->r_result, lreq->last_error); 3048 if (req->r_result < 0) { 3049 if (!lreq->last_error) { 3050 lreq->last_error = normalize_watch_error(req->r_result); 3051 queue_watch_error(lreq); 3052 } 3053 } 3054 3055 out: 3056 mutex_unlock(&lreq->lock); 3057 linger_put(lreq); 3058 } 3059 3060 static void send_linger(struct ceph_osd_linger_request *lreq) 3061 { 3062 struct ceph_osd_client *osdc = lreq->osdc; 3063 struct ceph_osd_request *req; 3064 int ret; 3065 3066 verify_osdc_wrlocked(osdc); 3067 mutex_lock(&lreq->lock); 3068 dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id); 3069 3070 if (lreq->reg_req) { 3071 if (lreq->reg_req->r_osd) 3072 cancel_linger_request(lreq->reg_req); 3073 ceph_osdc_put_request(lreq->reg_req); 3074 } 3075 3076 req = ceph_osdc_alloc_request(osdc, NULL, 1, true, GFP_NOIO); 3077 BUG_ON(!req); 3078 3079 target_copy(&req->r_t, &lreq->t); 3080 req->r_mtime = lreq->mtime; 3081 3082 if (lreq->is_watch && lreq->committed) { 3083 osd_req_op_watch_init(req, 0, CEPH_OSD_WATCH_OP_RECONNECT, 3084 lreq->linger_id, ++lreq->register_gen); 3085 dout("lreq %p reconnect register_gen %u\n", lreq, 3086 req->r_ops[0].watch.gen); 3087 req->r_callback = linger_reconnect_cb; 3088 } else { 3089 if (lreq->is_watch) { 3090 osd_req_op_watch_init(req, 0, CEPH_OSD_WATCH_OP_WATCH, 3091 lreq->linger_id, 0); 3092 } else { 3093 lreq->notify_id = 0; 3094 3095 refcount_inc(&lreq->request_pl->refcnt); 3096 osd_req_op_notify_init(req, 0, lreq->linger_id, 3097 lreq->request_pl); 3098 ceph_osd_data_pages_init( 3099 osd_req_op_data(req, 0, notify, response_data), 3100 lreq->notify_id_pages, PAGE_SIZE, 0, false, false); 3101 } 3102 dout("lreq %p register\n", lreq); 3103 req->r_callback = linger_commit_cb; 3104 } 3105 3106 ret = ceph_osdc_alloc_messages(req, GFP_NOIO); 3107 BUG_ON(ret); 3108 3109 req->r_priv = linger_get(lreq); 3110 req->r_linger = true; 3111 lreq->reg_req = req; 3112 mutex_unlock(&lreq->lock); 3113 3114 submit_request(req, true); 3115 } 3116 3117 static void linger_ping_cb(struct ceph_osd_request *req) 3118 { 3119 struct ceph_osd_linger_request *lreq = req->r_priv; 3120 3121 mutex_lock(&lreq->lock); 3122 if (req != lreq->ping_req) { 3123 dout("%s lreq %p linger_id %llu unknown req (%p != %p)\n", 3124 __func__, lreq, lreq->linger_id, req, lreq->ping_req); 3125 goto out; 3126 } 3127 3128 dout("%s lreq %p linger_id %llu result %d ping_sent %lu last_error %d\n", 3129 __func__, lreq, lreq->linger_id, req->r_result, lreq->ping_sent, 3130 lreq->last_error); 3131 if (lreq->register_gen == req->r_ops[0].watch.gen) { 3132 if (!req->r_result) { 3133 lreq->watch_valid_thru = lreq->ping_sent; 3134 } else if (!lreq->last_error) { 3135 lreq->last_error = normalize_watch_error(req->r_result); 3136 queue_watch_error(lreq); 3137 } 3138 } else { 3139 dout("lreq %p register_gen %u ignoring old pong %u\n", lreq, 3140 lreq->register_gen, req->r_ops[0].watch.gen); 3141 } 3142 3143 out: 3144 mutex_unlock(&lreq->lock); 3145 linger_put(lreq); 3146 } 3147 3148 static void send_linger_ping(struct ceph_osd_linger_request *lreq) 3149 { 3150 struct ceph_osd_client *osdc = lreq->osdc; 3151 struct ceph_osd_request *req; 3152 int ret; 3153 3154 if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) { 3155 dout("%s PAUSERD\n", __func__); 3156 return; 3157 } 3158 3159 lreq->ping_sent = jiffies; 3160 dout("%s lreq %p linger_id %llu ping_sent %lu register_gen %u\n", 3161 __func__, lreq, lreq->linger_id, lreq->ping_sent, 3162 lreq->register_gen); 3163 3164 if (lreq->ping_req) { 3165 if (lreq->ping_req->r_osd) 3166 cancel_linger_request(lreq->ping_req); 3167 ceph_osdc_put_request(lreq->ping_req); 3168 } 3169 3170 req = ceph_osdc_alloc_request(osdc, NULL, 1, true, GFP_NOIO); 3171 BUG_ON(!req); 3172 3173 target_copy(&req->r_t, &lreq->t); 3174 osd_req_op_watch_init(req, 0, CEPH_OSD_WATCH_OP_PING, lreq->linger_id, 3175 lreq->register_gen); 3176 req->r_callback = linger_ping_cb; 3177 3178 ret = ceph_osdc_alloc_messages(req, GFP_NOIO); 3179 BUG_ON(ret); 3180 3181 req->r_priv = linger_get(lreq); 3182 req->r_linger = true; 3183 lreq->ping_req = req; 3184 3185 ceph_osdc_get_request(req); 3186 account_request(req); 3187 req->r_tid = atomic64_inc_return(&osdc->last_tid); 3188 link_request(lreq->osd, req); 3189 send_request(req); 3190 } 3191 3192 static void linger_submit(struct ceph_osd_linger_request *lreq) 3193 { 3194 struct ceph_osd_client *osdc = lreq->osdc; 3195 struct ceph_osd *osd; 3196 3197 down_write(&osdc->lock); 3198 linger_register(lreq); 3199 3200 calc_target(osdc, &lreq->t, false); 3201 osd = lookup_create_osd(osdc, lreq->t.osd, true); 3202 link_linger(osd, lreq); 3203 3204 send_linger(lreq); 3205 up_write(&osdc->lock); 3206 } 3207 3208 static void cancel_linger_map_check(struct ceph_osd_linger_request *lreq) 3209 { 3210 struct ceph_osd_client *osdc = lreq->osdc; 3211 struct ceph_osd_linger_request *lookup_lreq; 3212 3213 verify_osdc_wrlocked(osdc); 3214 3215 lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks, 3216 lreq->linger_id); 3217 if (!lookup_lreq) 3218 return; 3219 3220 WARN_ON(lookup_lreq != lreq); 3221 erase_linger_mc(&osdc->linger_map_checks, lreq); 3222 linger_put(lreq); 3223 } 3224 3225 /* 3226 * @lreq has to be both registered and linked. 3227 */ 3228 static void __linger_cancel(struct ceph_osd_linger_request *lreq) 3229 { 3230 if (lreq->ping_req && lreq->ping_req->r_osd) 3231 cancel_linger_request(lreq->ping_req); 3232 if (lreq->reg_req && lreq->reg_req->r_osd) 3233 cancel_linger_request(lreq->reg_req); 3234 cancel_linger_map_check(lreq); 3235 unlink_linger(lreq->osd, lreq); 3236 linger_unregister(lreq); 3237 } 3238 3239 static void linger_cancel(struct ceph_osd_linger_request *lreq) 3240 { 3241 struct ceph_osd_client *osdc = lreq->osdc; 3242 3243 down_write(&osdc->lock); 3244 if (__linger_registered(lreq)) 3245 __linger_cancel(lreq); 3246 up_write(&osdc->lock); 3247 } 3248 3249 static void send_linger_map_check(struct ceph_osd_linger_request *lreq); 3250 3251 static void check_linger_pool_dne(struct ceph_osd_linger_request *lreq) 3252 { 3253 struct ceph_osd_client *osdc = lreq->osdc; 3254 struct ceph_osdmap *map = osdc->osdmap; 3255 3256 verify_osdc_wrlocked(osdc); 3257 WARN_ON(!map->epoch); 3258 3259 if (lreq->register_gen) { 3260 lreq->map_dne_bound = map->epoch; 3261 dout("%s lreq %p linger_id %llu pool disappeared\n", __func__, 3262 lreq, lreq->linger_id); 3263 } else { 3264 dout("%s lreq %p linger_id %llu map_dne_bound %u have %u\n", 3265 __func__, lreq, lreq->linger_id, lreq->map_dne_bound, 3266 map->epoch); 3267 } 3268 3269 if (lreq->map_dne_bound) { 3270 if (map->epoch >= lreq->map_dne_bound) { 3271 /* we had a new enough map */ 3272 pr_info("linger_id %llu pool does not exist\n", 3273 lreq->linger_id); 3274 linger_reg_commit_complete(lreq, -ENOENT); 3275 __linger_cancel(lreq); 3276 } 3277 } else { 3278 send_linger_map_check(lreq); 3279 } 3280 } 3281 3282 static void linger_map_check_cb(struct ceph_mon_generic_request *greq) 3283 { 3284 struct ceph_osd_client *osdc = &greq->monc->client->osdc; 3285 struct ceph_osd_linger_request *lreq; 3286 u64 linger_id = greq->private_data; 3287 3288 WARN_ON(greq->result || !greq->u.newest); 3289 3290 down_write(&osdc->lock); 3291 lreq = lookup_linger_mc(&osdc->linger_map_checks, linger_id); 3292 if (!lreq) { 3293 dout("%s linger_id %llu dne\n", __func__, linger_id); 3294 goto out_unlock; 3295 } 3296 3297 dout("%s lreq %p linger_id %llu map_dne_bound %u newest %llu\n", 3298 __func__, lreq, lreq->linger_id, lreq->map_dne_bound, 3299 greq->u.newest); 3300 if (!lreq->map_dne_bound) 3301 lreq->map_dne_bound = greq->u.newest; 3302 erase_linger_mc(&osdc->linger_map_checks, lreq); 3303 check_linger_pool_dne(lreq); 3304 3305 linger_put(lreq); 3306 out_unlock: 3307 up_write(&osdc->lock); 3308 } 3309 3310 static void send_linger_map_check(struct ceph_osd_linger_request *lreq) 3311 { 3312 struct ceph_osd_client *osdc = lreq->osdc; 3313 struct ceph_osd_linger_request *lookup_lreq; 3314 int ret; 3315 3316 verify_osdc_wrlocked(osdc); 3317 3318 lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks, 3319 lreq->linger_id); 3320 if (lookup_lreq) { 3321 WARN_ON(lookup_lreq != lreq); 3322 return; 3323 } 3324 3325 linger_get(lreq); 3326 insert_linger_mc(&osdc->linger_map_checks, lreq); 3327 ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap", 3328 linger_map_check_cb, lreq->linger_id); 3329 WARN_ON(ret); 3330 } 3331 3332 static int linger_reg_commit_wait(struct ceph_osd_linger_request *lreq) 3333 { 3334 int ret; 3335 3336 dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id); 3337 ret = wait_for_completion_interruptible(&lreq->reg_commit_wait); 3338 return ret ?: lreq->reg_commit_error; 3339 } 3340 3341 static int linger_notify_finish_wait(struct ceph_osd_linger_request *lreq) 3342 { 3343 int ret; 3344 3345 dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id); 3346 ret = wait_for_completion_interruptible(&lreq->notify_finish_wait); 3347 return ret ?: lreq->notify_finish_error; 3348 } 3349 3350 /* 3351 * Timeout callback, called every N seconds. When 1 or more OSD 3352 * requests has been active for more than N seconds, we send a keepalive 3353 * (tag + timestamp) to its OSD to ensure any communications channel 3354 * reset is detected. 3355 */ 3356 static void handle_timeout(struct work_struct *work) 3357 { 3358 struct ceph_osd_client *osdc = 3359 container_of(work, struct ceph_osd_client, timeout_work.work); 3360 struct ceph_options *opts = osdc->client->options; 3361 unsigned long cutoff = jiffies - opts->osd_keepalive_timeout; 3362 unsigned long expiry_cutoff = jiffies - opts->osd_request_timeout; 3363 LIST_HEAD(slow_osds); 3364 struct rb_node *n, *p; 3365 3366 dout("%s osdc %p\n", __func__, osdc); 3367 down_write(&osdc->lock); 3368 3369 /* 3370 * ping osds that are a bit slow. this ensures that if there 3371 * is a break in the TCP connection we will notice, and reopen 3372 * a connection with that osd (from the fault callback). 3373 */ 3374 for (n = rb_first(&osdc->osds); n; n = rb_next(n)) { 3375 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node); 3376 bool found = false; 3377 3378 for (p = rb_first(&osd->o_requests); p; ) { 3379 struct ceph_osd_request *req = 3380 rb_entry(p, struct ceph_osd_request, r_node); 3381 3382 p = rb_next(p); /* abort_request() */ 3383 3384 if (time_before(req->r_stamp, cutoff)) { 3385 dout(" req %p tid %llu on osd%d is laggy\n", 3386 req, req->r_tid, osd->o_osd); 3387 found = true; 3388 } 3389 if (opts->osd_request_timeout && 3390 time_before(req->r_start_stamp, expiry_cutoff)) { 3391 pr_err_ratelimited("tid %llu on osd%d timeout\n", 3392 req->r_tid, osd->o_osd); 3393 abort_request(req, -ETIMEDOUT); 3394 } 3395 } 3396 for (p = rb_first(&osd->o_linger_requests); p; p = rb_next(p)) { 3397 struct ceph_osd_linger_request *lreq = 3398 rb_entry(p, struct ceph_osd_linger_request, node); 3399 3400 dout(" lreq %p linger_id %llu is served by osd%d\n", 3401 lreq, lreq->linger_id, osd->o_osd); 3402 found = true; 3403 3404 mutex_lock(&lreq->lock); 3405 if (lreq->is_watch && lreq->committed && !lreq->last_error) 3406 send_linger_ping(lreq); 3407 mutex_unlock(&lreq->lock); 3408 } 3409 3410 if (found) 3411 list_move_tail(&osd->o_keepalive_item, &slow_osds); 3412 } 3413 3414 if (opts->osd_request_timeout) { 3415 for (p = rb_first(&osdc->homeless_osd.o_requests); p; ) { 3416 struct ceph_osd_request *req = 3417 rb_entry(p, struct ceph_osd_request, r_node); 3418 3419 p = rb_next(p); /* abort_request() */ 3420 3421 if (time_before(req->r_start_stamp, expiry_cutoff)) { 3422 pr_err_ratelimited("tid %llu on osd%d timeout\n", 3423 req->r_tid, osdc->homeless_osd.o_osd); 3424 abort_request(req, -ETIMEDOUT); 3425 } 3426 } 3427 } 3428 3429 if (atomic_read(&osdc->num_homeless) || !list_empty(&slow_osds)) 3430 maybe_request_map(osdc); 3431 3432 while (!list_empty(&slow_osds)) { 3433 struct ceph_osd *osd = list_first_entry(&slow_osds, 3434 struct ceph_osd, 3435 o_keepalive_item); 3436 list_del_init(&osd->o_keepalive_item); 3437 ceph_con_keepalive(&osd->o_con); 3438 } 3439 3440 up_write(&osdc->lock); 3441 schedule_delayed_work(&osdc->timeout_work, 3442 osdc->client->options->osd_keepalive_timeout); 3443 } 3444 3445 static void handle_osds_timeout(struct work_struct *work) 3446 { 3447 struct ceph_osd_client *osdc = 3448 container_of(work, struct ceph_osd_client, 3449 osds_timeout_work.work); 3450 unsigned long delay = osdc->client->options->osd_idle_ttl / 4; 3451 struct ceph_osd *osd, *nosd; 3452 3453 dout("%s osdc %p\n", __func__, osdc); 3454 down_write(&osdc->lock); 3455 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) { 3456 if (time_before(jiffies, osd->lru_ttl)) 3457 break; 3458 3459 WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests)); 3460 WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests)); 3461 close_osd(osd); 3462 } 3463 3464 up_write(&osdc->lock); 3465 schedule_delayed_work(&osdc->osds_timeout_work, 3466 round_jiffies_relative(delay)); 3467 } 3468 3469 static int ceph_oloc_decode(void **p, void *end, 3470 struct ceph_object_locator *oloc) 3471 { 3472 u8 struct_v, struct_cv; 3473 u32 len; 3474 void *struct_end; 3475 int ret = 0; 3476 3477 ceph_decode_need(p, end, 1 + 1 + 4, e_inval); 3478 struct_v = ceph_decode_8(p); 3479 struct_cv = ceph_decode_8(p); 3480 if (struct_v < 3) { 3481 pr_warn("got v %d < 3 cv %d of ceph_object_locator\n", 3482 struct_v, struct_cv); 3483 goto e_inval; 3484 } 3485 if (struct_cv > 6) { 3486 pr_warn("got v %d cv %d > 6 of ceph_object_locator\n", 3487 struct_v, struct_cv); 3488 goto e_inval; 3489 } 3490 len = ceph_decode_32(p); 3491 ceph_decode_need(p, end, len, e_inval); 3492 struct_end = *p + len; 3493 3494 oloc->pool = ceph_decode_64(p); 3495 *p += 4; /* skip preferred */ 3496 3497 len = ceph_decode_32(p); 3498 if (len > 0) { 3499 pr_warn("ceph_object_locator::key is set\n"); 3500 goto e_inval; 3501 } 3502 3503 if (struct_v >= 5) { 3504 bool changed = false; 3505 3506 len = ceph_decode_32(p); 3507 if (len > 0) { 3508 ceph_decode_need(p, end, len, e_inval); 3509 if (!oloc->pool_ns || 3510 ceph_compare_string(oloc->pool_ns, *p, len)) 3511 changed = true; 3512 *p += len; 3513 } else { 3514 if (oloc->pool_ns) 3515 changed = true; 3516 } 3517 if (changed) { 3518 /* redirect changes namespace */ 3519 pr_warn("ceph_object_locator::nspace is changed\n"); 3520 goto e_inval; 3521 } 3522 } 3523 3524 if (struct_v >= 6) { 3525 s64 hash = ceph_decode_64(p); 3526 if (hash != -1) { 3527 pr_warn("ceph_object_locator::hash is set\n"); 3528 goto e_inval; 3529 } 3530 } 3531 3532 /* skip the rest */ 3533 *p = struct_end; 3534 out: 3535 return ret; 3536 3537 e_inval: 3538 ret = -EINVAL; 3539 goto out; 3540 } 3541 3542 static int ceph_redirect_decode(void **p, void *end, 3543 struct ceph_request_redirect *redir) 3544 { 3545 u8 struct_v, struct_cv; 3546 u32 len; 3547 void *struct_end; 3548 int ret; 3549 3550 ceph_decode_need(p, end, 1 + 1 + 4, e_inval); 3551 struct_v = ceph_decode_8(p); 3552 struct_cv = ceph_decode_8(p); 3553 if (struct_cv > 1) { 3554 pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n", 3555 struct_v, struct_cv); 3556 goto e_inval; 3557 } 3558 len = ceph_decode_32(p); 3559 ceph_decode_need(p, end, len, e_inval); 3560 struct_end = *p + len; 3561 3562 ret = ceph_oloc_decode(p, end, &redir->oloc); 3563 if (ret) 3564 goto out; 3565 3566 len = ceph_decode_32(p); 3567 if (len > 0) { 3568 pr_warn("ceph_request_redirect::object_name is set\n"); 3569 goto e_inval; 3570 } 3571 3572 /* skip the rest */ 3573 *p = struct_end; 3574 out: 3575 return ret; 3576 3577 e_inval: 3578 ret = -EINVAL; 3579 goto out; 3580 } 3581 3582 struct MOSDOpReply { 3583 struct ceph_pg pgid; 3584 u64 flags; 3585 int result; 3586 u32 epoch; 3587 int num_ops; 3588 u32 outdata_len[CEPH_OSD_MAX_OPS]; 3589 s32 rval[CEPH_OSD_MAX_OPS]; 3590 int retry_attempt; 3591 struct ceph_eversion replay_version; 3592 u64 user_version; 3593 struct ceph_request_redirect redirect; 3594 }; 3595 3596 static int decode_MOSDOpReply(const struct ceph_msg *msg, struct MOSDOpReply *m) 3597 { 3598 void *p = msg->front.iov_base; 3599 void *const end = p + msg->front.iov_len; 3600 u16 version = le16_to_cpu(msg->hdr.version); 3601 struct ceph_eversion bad_replay_version; 3602 u8 decode_redir; 3603 u32 len; 3604 int ret; 3605 int i; 3606 3607 ceph_decode_32_safe(&p, end, len, e_inval); 3608 ceph_decode_need(&p, end, len, e_inval); 3609 p += len; /* skip oid */ 3610 3611 ret = ceph_decode_pgid(&p, end, &m->pgid); 3612 if (ret) 3613 return ret; 3614 3615 ceph_decode_64_safe(&p, end, m->flags, e_inval); 3616 ceph_decode_32_safe(&p, end, m->result, e_inval); 3617 ceph_decode_need(&p, end, sizeof(bad_replay_version), e_inval); 3618 memcpy(&bad_replay_version, p, sizeof(bad_replay_version)); 3619 p += sizeof(bad_replay_version); 3620 ceph_decode_32_safe(&p, end, m->epoch, e_inval); 3621 3622 ceph_decode_32_safe(&p, end, m->num_ops, e_inval); 3623 if (m->num_ops > ARRAY_SIZE(m->outdata_len)) 3624 goto e_inval; 3625 3626 ceph_decode_need(&p, end, m->num_ops * sizeof(struct ceph_osd_op), 3627 e_inval); 3628 for (i = 0; i < m->num_ops; i++) { 3629 struct ceph_osd_op *op = p; 3630 3631 m->outdata_len[i] = le32_to_cpu(op->payload_len); 3632 p += sizeof(*op); 3633 } 3634 3635 ceph_decode_32_safe(&p, end, m->retry_attempt, e_inval); 3636 for (i = 0; i < m->num_ops; i++) 3637 ceph_decode_32_safe(&p, end, m->rval[i], e_inval); 3638 3639 if (version >= 5) { 3640 ceph_decode_need(&p, end, sizeof(m->replay_version), e_inval); 3641 memcpy(&m->replay_version, p, sizeof(m->replay_version)); 3642 p += sizeof(m->replay_version); 3643 ceph_decode_64_safe(&p, end, m->user_version, e_inval); 3644 } else { 3645 m->replay_version = bad_replay_version; /* struct */ 3646 m->user_version = le64_to_cpu(m->replay_version.version); 3647 } 3648 3649 if (version >= 6) { 3650 if (version >= 7) 3651 ceph_decode_8_safe(&p, end, decode_redir, e_inval); 3652 else 3653 decode_redir = 1; 3654 } else { 3655 decode_redir = 0; 3656 } 3657 3658 if (decode_redir) { 3659 ret = ceph_redirect_decode(&p, end, &m->redirect); 3660 if (ret) 3661 return ret; 3662 } else { 3663 ceph_oloc_init(&m->redirect.oloc); 3664 } 3665 3666 return 0; 3667 3668 e_inval: 3669 return -EINVAL; 3670 } 3671 3672 /* 3673 * Handle MOSDOpReply. Set ->r_result and call the callback if it is 3674 * specified. 3675 */ 3676 static void handle_reply(struct ceph_osd *osd, struct ceph_msg *msg) 3677 { 3678 struct ceph_osd_client *osdc = osd->o_osdc; 3679 struct ceph_osd_request *req; 3680 struct MOSDOpReply m; 3681 u64 tid = le64_to_cpu(msg->hdr.tid); 3682 u32 data_len = 0; 3683 int ret; 3684 int i; 3685 3686 dout("%s msg %p tid %llu\n", __func__, msg, tid); 3687 3688 down_read(&osdc->lock); 3689 if (!osd_registered(osd)) { 3690 dout("%s osd%d unknown\n", __func__, osd->o_osd); 3691 goto out_unlock_osdc; 3692 } 3693 WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num)); 3694 3695 mutex_lock(&osd->lock); 3696 req = lookup_request(&osd->o_requests, tid); 3697 if (!req) { 3698 dout("%s osd%d tid %llu unknown\n", __func__, osd->o_osd, tid); 3699 goto out_unlock_session; 3700 } 3701 3702 m.redirect.oloc.pool_ns = req->r_t.target_oloc.pool_ns; 3703 ret = decode_MOSDOpReply(msg, &m); 3704 m.redirect.oloc.pool_ns = NULL; 3705 if (ret) { 3706 pr_err("failed to decode MOSDOpReply for tid %llu: %d\n", 3707 req->r_tid, ret); 3708 ceph_msg_dump(msg); 3709 goto fail_request; 3710 } 3711 dout("%s req %p tid %llu flags 0x%llx pgid %llu.%x epoch %u attempt %d v %u'%llu uv %llu\n", 3712 __func__, req, req->r_tid, m.flags, m.pgid.pool, m.pgid.seed, 3713 m.epoch, m.retry_attempt, le32_to_cpu(m.replay_version.epoch), 3714 le64_to_cpu(m.replay_version.version), m.user_version); 3715 3716 if (m.retry_attempt >= 0) { 3717 if (m.retry_attempt != req->r_attempts - 1) { 3718 dout("req %p tid %llu retry_attempt %d != %d, ignoring\n", 3719 req, req->r_tid, m.retry_attempt, 3720 req->r_attempts - 1); 3721 goto out_unlock_session; 3722 } 3723 } else { 3724 WARN_ON(1); /* MOSDOpReply v4 is assumed */ 3725 } 3726 3727 if (!ceph_oloc_empty(&m.redirect.oloc)) { 3728 dout("req %p tid %llu redirect pool %lld\n", req, req->r_tid, 3729 m.redirect.oloc.pool); 3730 unlink_request(osd, req); 3731 mutex_unlock(&osd->lock); 3732 3733 /* 3734 * Not ceph_oloc_copy() - changing pool_ns is not 3735 * supported. 3736 */ 3737 req->r_t.target_oloc.pool = m.redirect.oloc.pool; 3738 req->r_flags |= CEPH_OSD_FLAG_REDIRECTED | 3739 CEPH_OSD_FLAG_IGNORE_OVERLAY | 3740 CEPH_OSD_FLAG_IGNORE_CACHE; 3741 req->r_tid = 0; 3742 __submit_request(req, false); 3743 goto out_unlock_osdc; 3744 } 3745 3746 if (m.result == -EAGAIN) { 3747 dout("req %p tid %llu EAGAIN\n", req, req->r_tid); 3748 unlink_request(osd, req); 3749 mutex_unlock(&osd->lock); 3750 3751 /* 3752 * The object is missing on the replica or not (yet) 3753 * readable. Clear pgid to force a resend to the primary 3754 * via legacy_change. 3755 */ 3756 req->r_t.pgid.pool = 0; 3757 req->r_t.pgid.seed = 0; 3758 WARN_ON(!req->r_t.used_replica); 3759 req->r_flags &= ~(CEPH_OSD_FLAG_BALANCE_READS | 3760 CEPH_OSD_FLAG_LOCALIZE_READS); 3761 req->r_tid = 0; 3762 __submit_request(req, false); 3763 goto out_unlock_osdc; 3764 } 3765 3766 if (m.num_ops != req->r_num_ops) { 3767 pr_err("num_ops %d != %d for tid %llu\n", m.num_ops, 3768 req->r_num_ops, req->r_tid); 3769 goto fail_request; 3770 } 3771 for (i = 0; i < req->r_num_ops; i++) { 3772 dout(" req %p tid %llu op %d rval %d len %u\n", req, 3773 req->r_tid, i, m.rval[i], m.outdata_len[i]); 3774 req->r_ops[i].rval = m.rval[i]; 3775 req->r_ops[i].outdata_len = m.outdata_len[i]; 3776 data_len += m.outdata_len[i]; 3777 } 3778 if (data_len != le32_to_cpu(msg->hdr.data_len)) { 3779 pr_err("sum of lens %u != %u for tid %llu\n", data_len, 3780 le32_to_cpu(msg->hdr.data_len), req->r_tid); 3781 goto fail_request; 3782 } 3783 dout("%s req %p tid %llu result %d data_len %u\n", __func__, 3784 req, req->r_tid, m.result, data_len); 3785 3786 /* 3787 * Since we only ever request ONDISK, we should only ever get 3788 * one (type of) reply back. 3789 */ 3790 WARN_ON(!(m.flags & CEPH_OSD_FLAG_ONDISK)); 3791 req->r_result = m.result ?: data_len; 3792 finish_request(req); 3793 mutex_unlock(&osd->lock); 3794 up_read(&osdc->lock); 3795 3796 __complete_request(req); 3797 return; 3798 3799 fail_request: 3800 complete_request(req, -EIO); 3801 out_unlock_session: 3802 mutex_unlock(&osd->lock); 3803 out_unlock_osdc: 3804 up_read(&osdc->lock); 3805 } 3806 3807 static void set_pool_was_full(struct ceph_osd_client *osdc) 3808 { 3809 struct rb_node *n; 3810 3811 for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) { 3812 struct ceph_pg_pool_info *pi = 3813 rb_entry(n, struct ceph_pg_pool_info, node); 3814 3815 pi->was_full = __pool_full(pi); 3816 } 3817 } 3818 3819 static bool pool_cleared_full(struct ceph_osd_client *osdc, s64 pool_id) 3820 { 3821 struct ceph_pg_pool_info *pi; 3822 3823 pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id); 3824 if (!pi) 3825 return false; 3826 3827 return pi->was_full && !__pool_full(pi); 3828 } 3829 3830 static enum calc_target_result 3831 recalc_linger_target(struct ceph_osd_linger_request *lreq) 3832 { 3833 struct ceph_osd_client *osdc = lreq->osdc; 3834 enum calc_target_result ct_res; 3835 3836 ct_res = calc_target(osdc, &lreq->t, true); 3837 if (ct_res == CALC_TARGET_NEED_RESEND) { 3838 struct ceph_osd *osd; 3839 3840 osd = lookup_create_osd(osdc, lreq->t.osd, true); 3841 if (osd != lreq->osd) { 3842 unlink_linger(lreq->osd, lreq); 3843 link_linger(osd, lreq); 3844 } 3845 } 3846 3847 return ct_res; 3848 } 3849 3850 /* 3851 * Requeue requests whose mapping to an OSD has changed. 3852 */ 3853 static void scan_requests(struct ceph_osd *osd, 3854 bool force_resend, 3855 bool cleared_full, 3856 bool check_pool_cleared_full, 3857 struct rb_root *need_resend, 3858 struct list_head *need_resend_linger) 3859 { 3860 struct ceph_osd_client *osdc = osd->o_osdc; 3861 struct rb_node *n; 3862 bool force_resend_writes; 3863 3864 for (n = rb_first(&osd->o_linger_requests); n; ) { 3865 struct ceph_osd_linger_request *lreq = 3866 rb_entry(n, struct ceph_osd_linger_request, node); 3867 enum calc_target_result ct_res; 3868 3869 n = rb_next(n); /* recalc_linger_target() */ 3870 3871 dout("%s lreq %p linger_id %llu\n", __func__, lreq, 3872 lreq->linger_id); 3873 ct_res = recalc_linger_target(lreq); 3874 switch (ct_res) { 3875 case CALC_TARGET_NO_ACTION: 3876 force_resend_writes = cleared_full || 3877 (check_pool_cleared_full && 3878 pool_cleared_full(osdc, lreq->t.base_oloc.pool)); 3879 if (!force_resend && !force_resend_writes) 3880 break; 3881 3882 fallthrough; 3883 case CALC_TARGET_NEED_RESEND: 3884 cancel_linger_map_check(lreq); 3885 /* 3886 * scan_requests() for the previous epoch(s) 3887 * may have already added it to the list, since 3888 * it's not unlinked here. 3889 */ 3890 if (list_empty(&lreq->scan_item)) 3891 list_add_tail(&lreq->scan_item, need_resend_linger); 3892 break; 3893 case CALC_TARGET_POOL_DNE: 3894 list_del_init(&lreq->scan_item); 3895 check_linger_pool_dne(lreq); 3896 break; 3897 } 3898 } 3899 3900 for (n = rb_first(&osd->o_requests); n; ) { 3901 struct ceph_osd_request *req = 3902 rb_entry(n, struct ceph_osd_request, r_node); 3903 enum calc_target_result ct_res; 3904 3905 n = rb_next(n); /* unlink_request(), check_pool_dne() */ 3906 3907 dout("%s req %p tid %llu\n", __func__, req, req->r_tid); 3908 ct_res = calc_target(osdc, &req->r_t, false); 3909 switch (ct_res) { 3910 case CALC_TARGET_NO_ACTION: 3911 force_resend_writes = cleared_full || 3912 (check_pool_cleared_full && 3913 pool_cleared_full(osdc, req->r_t.base_oloc.pool)); 3914 if (!force_resend && 3915 (!(req->r_flags & CEPH_OSD_FLAG_WRITE) || 3916 !force_resend_writes)) 3917 break; 3918 3919 fallthrough; 3920 case CALC_TARGET_NEED_RESEND: 3921 cancel_map_check(req); 3922 unlink_request(osd, req); 3923 insert_request(need_resend, req); 3924 break; 3925 case CALC_TARGET_POOL_DNE: 3926 check_pool_dne(req); 3927 break; 3928 } 3929 } 3930 } 3931 3932 static int handle_one_map(struct ceph_osd_client *osdc, 3933 void *p, void *end, bool incremental, 3934 struct rb_root *need_resend, 3935 struct list_head *need_resend_linger) 3936 { 3937 struct ceph_osdmap *newmap; 3938 struct rb_node *n; 3939 bool skipped_map = false; 3940 bool was_full; 3941 3942 was_full = ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL); 3943 set_pool_was_full(osdc); 3944 3945 if (incremental) 3946 newmap = osdmap_apply_incremental(&p, end, 3947 ceph_msgr2(osdc->client), 3948 osdc->osdmap); 3949 else 3950 newmap = ceph_osdmap_decode(&p, end, ceph_msgr2(osdc->client)); 3951 if (IS_ERR(newmap)) 3952 return PTR_ERR(newmap); 3953 3954 if (newmap != osdc->osdmap) { 3955 /* 3956 * Preserve ->was_full before destroying the old map. 3957 * For pools that weren't in the old map, ->was_full 3958 * should be false. 3959 */ 3960 for (n = rb_first(&newmap->pg_pools); n; n = rb_next(n)) { 3961 struct ceph_pg_pool_info *pi = 3962 rb_entry(n, struct ceph_pg_pool_info, node); 3963 struct ceph_pg_pool_info *old_pi; 3964 3965 old_pi = ceph_pg_pool_by_id(osdc->osdmap, pi->id); 3966 if (old_pi) 3967 pi->was_full = old_pi->was_full; 3968 else 3969 WARN_ON(pi->was_full); 3970 } 3971 3972 if (osdc->osdmap->epoch && 3973 osdc->osdmap->epoch + 1 < newmap->epoch) { 3974 WARN_ON(incremental); 3975 skipped_map = true; 3976 } 3977 3978 ceph_osdmap_destroy(osdc->osdmap); 3979 osdc->osdmap = newmap; 3980 } 3981 3982 was_full &= !ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL); 3983 scan_requests(&osdc->homeless_osd, skipped_map, was_full, true, 3984 need_resend, need_resend_linger); 3985 3986 for (n = rb_first(&osdc->osds); n; ) { 3987 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node); 3988 3989 n = rb_next(n); /* close_osd() */ 3990 3991 scan_requests(osd, skipped_map, was_full, true, need_resend, 3992 need_resend_linger); 3993 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) || 3994 memcmp(&osd->o_con.peer_addr, 3995 ceph_osd_addr(osdc->osdmap, osd->o_osd), 3996 sizeof(struct ceph_entity_addr))) 3997 close_osd(osd); 3998 } 3999 4000 return 0; 4001 } 4002 4003 static void kick_requests(struct ceph_osd_client *osdc, 4004 struct rb_root *need_resend, 4005 struct list_head *need_resend_linger) 4006 { 4007 struct ceph_osd_linger_request *lreq, *nlreq; 4008 enum calc_target_result ct_res; 4009 struct rb_node *n; 4010 4011 /* make sure need_resend targets reflect latest map */ 4012 for (n = rb_first(need_resend); n; ) { 4013 struct ceph_osd_request *req = 4014 rb_entry(n, struct ceph_osd_request, r_node); 4015 4016 n = rb_next(n); 4017 4018 if (req->r_t.epoch < osdc->osdmap->epoch) { 4019 ct_res = calc_target(osdc, &req->r_t, false); 4020 if (ct_res == CALC_TARGET_POOL_DNE) { 4021 erase_request(need_resend, req); 4022 check_pool_dne(req); 4023 } 4024 } 4025 } 4026 4027 for (n = rb_first(need_resend); n; ) { 4028 struct ceph_osd_request *req = 4029 rb_entry(n, struct ceph_osd_request, r_node); 4030 struct ceph_osd *osd; 4031 4032 n = rb_next(n); 4033 erase_request(need_resend, req); /* before link_request() */ 4034 4035 osd = lookup_create_osd(osdc, req->r_t.osd, true); 4036 link_request(osd, req); 4037 if (!req->r_linger) { 4038 if (!osd_homeless(osd) && !req->r_t.paused) 4039 send_request(req); 4040 } else { 4041 cancel_linger_request(req); 4042 } 4043 } 4044 4045 list_for_each_entry_safe(lreq, nlreq, need_resend_linger, scan_item) { 4046 if (!osd_homeless(lreq->osd)) 4047 send_linger(lreq); 4048 4049 list_del_init(&lreq->scan_item); 4050 } 4051 } 4052 4053 /* 4054 * Process updated osd map. 4055 * 4056 * The message contains any number of incremental and full maps, normally 4057 * indicating some sort of topology change in the cluster. Kick requests 4058 * off to different OSDs as needed. 4059 */ 4060 void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg) 4061 { 4062 void *p = msg->front.iov_base; 4063 void *const end = p + msg->front.iov_len; 4064 u32 nr_maps, maplen; 4065 u32 epoch; 4066 struct ceph_fsid fsid; 4067 struct rb_root need_resend = RB_ROOT; 4068 LIST_HEAD(need_resend_linger); 4069 bool handled_incremental = false; 4070 bool was_pauserd, was_pausewr; 4071 bool pauserd, pausewr; 4072 int err; 4073 4074 dout("%s have %u\n", __func__, osdc->osdmap->epoch); 4075 down_write(&osdc->lock); 4076 4077 /* verify fsid */ 4078 ceph_decode_need(&p, end, sizeof(fsid), bad); 4079 ceph_decode_copy(&p, &fsid, sizeof(fsid)); 4080 if (ceph_check_fsid(osdc->client, &fsid) < 0) 4081 goto bad; 4082 4083 was_pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD); 4084 was_pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) || 4085 ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || 4086 have_pool_full(osdc); 4087 4088 /* incremental maps */ 4089 ceph_decode_32_safe(&p, end, nr_maps, bad); 4090 dout(" %d inc maps\n", nr_maps); 4091 while (nr_maps > 0) { 4092 ceph_decode_need(&p, end, 2*sizeof(u32), bad); 4093 epoch = ceph_decode_32(&p); 4094 maplen = ceph_decode_32(&p); 4095 ceph_decode_need(&p, end, maplen, bad); 4096 if (osdc->osdmap->epoch && 4097 osdc->osdmap->epoch + 1 == epoch) { 4098 dout("applying incremental map %u len %d\n", 4099 epoch, maplen); 4100 err = handle_one_map(osdc, p, p + maplen, true, 4101 &need_resend, &need_resend_linger); 4102 if (err) 4103 goto bad; 4104 handled_incremental = true; 4105 } else { 4106 dout("ignoring incremental map %u len %d\n", 4107 epoch, maplen); 4108 } 4109 p += maplen; 4110 nr_maps--; 4111 } 4112 if (handled_incremental) 4113 goto done; 4114 4115 /* full maps */ 4116 ceph_decode_32_safe(&p, end, nr_maps, bad); 4117 dout(" %d full maps\n", nr_maps); 4118 while (nr_maps) { 4119 ceph_decode_need(&p, end, 2*sizeof(u32), bad); 4120 epoch = ceph_decode_32(&p); 4121 maplen = ceph_decode_32(&p); 4122 ceph_decode_need(&p, end, maplen, bad); 4123 if (nr_maps > 1) { 4124 dout("skipping non-latest full map %u len %d\n", 4125 epoch, maplen); 4126 } else if (osdc->osdmap->epoch >= epoch) { 4127 dout("skipping full map %u len %d, " 4128 "older than our %u\n", epoch, maplen, 4129 osdc->osdmap->epoch); 4130 } else { 4131 dout("taking full map %u len %d\n", epoch, maplen); 4132 err = handle_one_map(osdc, p, p + maplen, false, 4133 &need_resend, &need_resend_linger); 4134 if (err) 4135 goto bad; 4136 } 4137 p += maplen; 4138 nr_maps--; 4139 } 4140 4141 done: 4142 /* 4143 * subscribe to subsequent osdmap updates if full to ensure 4144 * we find out when we are no longer full and stop returning 4145 * ENOSPC. 4146 */ 4147 pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD); 4148 pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) || 4149 ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || 4150 have_pool_full(osdc); 4151 if (was_pauserd || was_pausewr || pauserd || pausewr || 4152 osdc->osdmap->epoch < osdc->epoch_barrier) 4153 maybe_request_map(osdc); 4154 4155 kick_requests(osdc, &need_resend, &need_resend_linger); 4156 4157 ceph_osdc_abort_on_full(osdc); 4158 ceph_monc_got_map(&osdc->client->monc, CEPH_SUB_OSDMAP, 4159 osdc->osdmap->epoch); 4160 up_write(&osdc->lock); 4161 wake_up_all(&osdc->client->auth_wq); 4162 return; 4163 4164 bad: 4165 pr_err("osdc handle_map corrupt msg\n"); 4166 ceph_msg_dump(msg); 4167 up_write(&osdc->lock); 4168 } 4169 4170 /* 4171 * Resubmit requests pending on the given osd. 4172 */ 4173 static void kick_osd_requests(struct ceph_osd *osd) 4174 { 4175 struct rb_node *n; 4176 4177 clear_backoffs(osd); 4178 4179 for (n = rb_first(&osd->o_requests); n; ) { 4180 struct ceph_osd_request *req = 4181 rb_entry(n, struct ceph_osd_request, r_node); 4182 4183 n = rb_next(n); /* cancel_linger_request() */ 4184 4185 if (!req->r_linger) { 4186 if (!req->r_t.paused) 4187 send_request(req); 4188 } else { 4189 cancel_linger_request(req); 4190 } 4191 } 4192 for (n = rb_first(&osd->o_linger_requests); n; n = rb_next(n)) { 4193 struct ceph_osd_linger_request *lreq = 4194 rb_entry(n, struct ceph_osd_linger_request, node); 4195 4196 send_linger(lreq); 4197 } 4198 } 4199 4200 /* 4201 * If the osd connection drops, we need to resubmit all requests. 4202 */ 4203 static void osd_fault(struct ceph_connection *con) 4204 { 4205 struct ceph_osd *osd = con->private; 4206 struct ceph_osd_client *osdc = osd->o_osdc; 4207 4208 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd); 4209 4210 down_write(&osdc->lock); 4211 if (!osd_registered(osd)) { 4212 dout("%s osd%d unknown\n", __func__, osd->o_osd); 4213 goto out_unlock; 4214 } 4215 4216 if (!reopen_osd(osd)) 4217 kick_osd_requests(osd); 4218 maybe_request_map(osdc); 4219 4220 out_unlock: 4221 up_write(&osdc->lock); 4222 } 4223 4224 struct MOSDBackoff { 4225 struct ceph_spg spgid; 4226 u32 map_epoch; 4227 u8 op; 4228 u64 id; 4229 struct ceph_hobject_id *begin; 4230 struct ceph_hobject_id *end; 4231 }; 4232 4233 static int decode_MOSDBackoff(const struct ceph_msg *msg, struct MOSDBackoff *m) 4234 { 4235 void *p = msg->front.iov_base; 4236 void *const end = p + msg->front.iov_len; 4237 u8 struct_v; 4238 u32 struct_len; 4239 int ret; 4240 4241 ret = ceph_start_decoding(&p, end, 1, "spg_t", &struct_v, &struct_len); 4242 if (ret) 4243 return ret; 4244 4245 ret = ceph_decode_pgid(&p, end, &m->spgid.pgid); 4246 if (ret) 4247 return ret; 4248 4249 ceph_decode_8_safe(&p, end, m->spgid.shard, e_inval); 4250 ceph_decode_32_safe(&p, end, m->map_epoch, e_inval); 4251 ceph_decode_8_safe(&p, end, m->op, e_inval); 4252 ceph_decode_64_safe(&p, end, m->id, e_inval); 4253 4254 m->begin = kzalloc(sizeof(*m->begin), GFP_NOIO); 4255 if (!m->begin) 4256 return -ENOMEM; 4257 4258 ret = decode_hoid(&p, end, m->begin); 4259 if (ret) { 4260 free_hoid(m->begin); 4261 return ret; 4262 } 4263 4264 m->end = kzalloc(sizeof(*m->end), GFP_NOIO); 4265 if (!m->end) { 4266 free_hoid(m->begin); 4267 return -ENOMEM; 4268 } 4269 4270 ret = decode_hoid(&p, end, m->end); 4271 if (ret) { 4272 free_hoid(m->begin); 4273 free_hoid(m->end); 4274 return ret; 4275 } 4276 4277 return 0; 4278 4279 e_inval: 4280 return -EINVAL; 4281 } 4282 4283 static struct ceph_msg *create_backoff_message( 4284 const struct ceph_osd_backoff *backoff, 4285 u32 map_epoch) 4286 { 4287 struct ceph_msg *msg; 4288 void *p, *end; 4289 int msg_size; 4290 4291 msg_size = CEPH_ENCODING_START_BLK_LEN + 4292 CEPH_PGID_ENCODING_LEN + 1; /* spgid */ 4293 msg_size += 4 + 1 + 8; /* map_epoch, op, id */ 4294 msg_size += CEPH_ENCODING_START_BLK_LEN + 4295 hoid_encoding_size(backoff->begin); 4296 msg_size += CEPH_ENCODING_START_BLK_LEN + 4297 hoid_encoding_size(backoff->end); 4298 4299 msg = ceph_msg_new(CEPH_MSG_OSD_BACKOFF, msg_size, GFP_NOIO, true); 4300 if (!msg) 4301 return NULL; 4302 4303 p = msg->front.iov_base; 4304 end = p + msg->front_alloc_len; 4305 4306 encode_spgid(&p, &backoff->spgid); 4307 ceph_encode_32(&p, map_epoch); 4308 ceph_encode_8(&p, CEPH_OSD_BACKOFF_OP_ACK_BLOCK); 4309 ceph_encode_64(&p, backoff->id); 4310 encode_hoid(&p, end, backoff->begin); 4311 encode_hoid(&p, end, backoff->end); 4312 BUG_ON(p != end); 4313 4314 msg->front.iov_len = p - msg->front.iov_base; 4315 msg->hdr.version = cpu_to_le16(1); /* MOSDBackoff v1 */ 4316 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len); 4317 4318 return msg; 4319 } 4320 4321 static void handle_backoff_block(struct ceph_osd *osd, struct MOSDBackoff *m) 4322 { 4323 struct ceph_spg_mapping *spg; 4324 struct ceph_osd_backoff *backoff; 4325 struct ceph_msg *msg; 4326 4327 dout("%s osd%d spgid %llu.%xs%d id %llu\n", __func__, osd->o_osd, 4328 m->spgid.pgid.pool, m->spgid.pgid.seed, m->spgid.shard, m->id); 4329 4330 spg = lookup_spg_mapping(&osd->o_backoff_mappings, &m->spgid); 4331 if (!spg) { 4332 spg = alloc_spg_mapping(); 4333 if (!spg) { 4334 pr_err("%s failed to allocate spg\n", __func__); 4335 return; 4336 } 4337 spg->spgid = m->spgid; /* struct */ 4338 insert_spg_mapping(&osd->o_backoff_mappings, spg); 4339 } 4340 4341 backoff = alloc_backoff(); 4342 if (!backoff) { 4343 pr_err("%s failed to allocate backoff\n", __func__); 4344 return; 4345 } 4346 backoff->spgid = m->spgid; /* struct */ 4347 backoff->id = m->id; 4348 backoff->begin = m->begin; 4349 m->begin = NULL; /* backoff now owns this */ 4350 backoff->end = m->end; 4351 m->end = NULL; /* ditto */ 4352 4353 insert_backoff(&spg->backoffs, backoff); 4354 insert_backoff_by_id(&osd->o_backoffs_by_id, backoff); 4355 4356 /* 4357 * Ack with original backoff's epoch so that the OSD can 4358 * discard this if there was a PG split. 4359 */ 4360 msg = create_backoff_message(backoff, m->map_epoch); 4361 if (!msg) { 4362 pr_err("%s failed to allocate msg\n", __func__); 4363 return; 4364 } 4365 ceph_con_send(&osd->o_con, msg); 4366 } 4367 4368 static bool target_contained_by(const struct ceph_osd_request_target *t, 4369 const struct ceph_hobject_id *begin, 4370 const struct ceph_hobject_id *end) 4371 { 4372 struct ceph_hobject_id hoid; 4373 int cmp; 4374 4375 hoid_fill_from_target(&hoid, t); 4376 cmp = hoid_compare(&hoid, begin); 4377 return !cmp || (cmp > 0 && hoid_compare(&hoid, end) < 0); 4378 } 4379 4380 static void handle_backoff_unblock(struct ceph_osd *osd, 4381 const struct MOSDBackoff *m) 4382 { 4383 struct ceph_spg_mapping *spg; 4384 struct ceph_osd_backoff *backoff; 4385 struct rb_node *n; 4386 4387 dout("%s osd%d spgid %llu.%xs%d id %llu\n", __func__, osd->o_osd, 4388 m->spgid.pgid.pool, m->spgid.pgid.seed, m->spgid.shard, m->id); 4389 4390 backoff = lookup_backoff_by_id(&osd->o_backoffs_by_id, m->id); 4391 if (!backoff) { 4392 pr_err("%s osd%d spgid %llu.%xs%d id %llu backoff dne\n", 4393 __func__, osd->o_osd, m->spgid.pgid.pool, 4394 m->spgid.pgid.seed, m->spgid.shard, m->id); 4395 return; 4396 } 4397 4398 if (hoid_compare(backoff->begin, m->begin) && 4399 hoid_compare(backoff->end, m->end)) { 4400 pr_err("%s osd%d spgid %llu.%xs%d id %llu bad range?\n", 4401 __func__, osd->o_osd, m->spgid.pgid.pool, 4402 m->spgid.pgid.seed, m->spgid.shard, m->id); 4403 /* unblock it anyway... */ 4404 } 4405 4406 spg = lookup_spg_mapping(&osd->o_backoff_mappings, &backoff->spgid); 4407 BUG_ON(!spg); 4408 4409 erase_backoff(&spg->backoffs, backoff); 4410 erase_backoff_by_id(&osd->o_backoffs_by_id, backoff); 4411 free_backoff(backoff); 4412 4413 if (RB_EMPTY_ROOT(&spg->backoffs)) { 4414 erase_spg_mapping(&osd->o_backoff_mappings, spg); 4415 free_spg_mapping(spg); 4416 } 4417 4418 for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) { 4419 struct ceph_osd_request *req = 4420 rb_entry(n, struct ceph_osd_request, r_node); 4421 4422 if (!ceph_spg_compare(&req->r_t.spgid, &m->spgid)) { 4423 /* 4424 * Match against @m, not @backoff -- the PG may 4425 * have split on the OSD. 4426 */ 4427 if (target_contained_by(&req->r_t, m->begin, m->end)) { 4428 /* 4429 * If no other installed backoff applies, 4430 * resend. 4431 */ 4432 send_request(req); 4433 } 4434 } 4435 } 4436 } 4437 4438 static void handle_backoff(struct ceph_osd *osd, struct ceph_msg *msg) 4439 { 4440 struct ceph_osd_client *osdc = osd->o_osdc; 4441 struct MOSDBackoff m; 4442 int ret; 4443 4444 down_read(&osdc->lock); 4445 if (!osd_registered(osd)) { 4446 dout("%s osd%d unknown\n", __func__, osd->o_osd); 4447 up_read(&osdc->lock); 4448 return; 4449 } 4450 WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num)); 4451 4452 mutex_lock(&osd->lock); 4453 ret = decode_MOSDBackoff(msg, &m); 4454 if (ret) { 4455 pr_err("failed to decode MOSDBackoff: %d\n", ret); 4456 ceph_msg_dump(msg); 4457 goto out_unlock; 4458 } 4459 4460 switch (m.op) { 4461 case CEPH_OSD_BACKOFF_OP_BLOCK: 4462 handle_backoff_block(osd, &m); 4463 break; 4464 case CEPH_OSD_BACKOFF_OP_UNBLOCK: 4465 handle_backoff_unblock(osd, &m); 4466 break; 4467 default: 4468 pr_err("%s osd%d unknown op %d\n", __func__, osd->o_osd, m.op); 4469 } 4470 4471 free_hoid(m.begin); 4472 free_hoid(m.end); 4473 4474 out_unlock: 4475 mutex_unlock(&osd->lock); 4476 up_read(&osdc->lock); 4477 } 4478 4479 /* 4480 * Process osd watch notifications 4481 */ 4482 static void handle_watch_notify(struct ceph_osd_client *osdc, 4483 struct ceph_msg *msg) 4484 { 4485 void *p = msg->front.iov_base; 4486 void *const end = p + msg->front.iov_len; 4487 struct ceph_osd_linger_request *lreq; 4488 struct linger_work *lwork; 4489 u8 proto_ver, opcode; 4490 u64 cookie, notify_id; 4491 u64 notifier_id = 0; 4492 s32 return_code = 0; 4493 void *payload = NULL; 4494 u32 payload_len = 0; 4495 4496 ceph_decode_8_safe(&p, end, proto_ver, bad); 4497 ceph_decode_8_safe(&p, end, opcode, bad); 4498 ceph_decode_64_safe(&p, end, cookie, bad); 4499 p += 8; /* skip ver */ 4500 ceph_decode_64_safe(&p, end, notify_id, bad); 4501 4502 if (proto_ver >= 1) { 4503 ceph_decode_32_safe(&p, end, payload_len, bad); 4504 ceph_decode_need(&p, end, payload_len, bad); 4505 payload = p; 4506 p += payload_len; 4507 } 4508 4509 if (le16_to_cpu(msg->hdr.version) >= 2) 4510 ceph_decode_32_safe(&p, end, return_code, bad); 4511 4512 if (le16_to_cpu(msg->hdr.version) >= 3) 4513 ceph_decode_64_safe(&p, end, notifier_id, bad); 4514 4515 down_read(&osdc->lock); 4516 lreq = lookup_linger_osdc(&osdc->linger_requests, cookie); 4517 if (!lreq) { 4518 dout("%s opcode %d cookie %llu dne\n", __func__, opcode, 4519 cookie); 4520 goto out_unlock_osdc; 4521 } 4522 4523 mutex_lock(&lreq->lock); 4524 dout("%s opcode %d cookie %llu lreq %p is_watch %d\n", __func__, 4525 opcode, cookie, lreq, lreq->is_watch); 4526 if (opcode == CEPH_WATCH_EVENT_DISCONNECT) { 4527 if (!lreq->last_error) { 4528 lreq->last_error = -ENOTCONN; 4529 queue_watch_error(lreq); 4530 } 4531 } else if (!lreq->is_watch) { 4532 /* CEPH_WATCH_EVENT_NOTIFY_COMPLETE */ 4533 if (lreq->notify_id && lreq->notify_id != notify_id) { 4534 dout("lreq %p notify_id %llu != %llu, ignoring\n", lreq, 4535 lreq->notify_id, notify_id); 4536 } else if (!completion_done(&lreq->notify_finish_wait)) { 4537 struct ceph_msg_data *data = 4538 msg->num_data_items ? &msg->data[0] : NULL; 4539 4540 if (data) { 4541 if (lreq->preply_pages) { 4542 WARN_ON(data->type != 4543 CEPH_MSG_DATA_PAGES); 4544 *lreq->preply_pages = data->pages; 4545 *lreq->preply_len = data->length; 4546 data->own_pages = false; 4547 } 4548 } 4549 lreq->notify_finish_error = return_code; 4550 complete_all(&lreq->notify_finish_wait); 4551 } 4552 } else { 4553 /* CEPH_WATCH_EVENT_NOTIFY */ 4554 lwork = lwork_alloc(lreq, do_watch_notify); 4555 if (!lwork) { 4556 pr_err("failed to allocate notify-lwork\n"); 4557 goto out_unlock_lreq; 4558 } 4559 4560 lwork->notify.notify_id = notify_id; 4561 lwork->notify.notifier_id = notifier_id; 4562 lwork->notify.payload = payload; 4563 lwork->notify.payload_len = payload_len; 4564 lwork->notify.msg = ceph_msg_get(msg); 4565 lwork_queue(lwork); 4566 } 4567 4568 out_unlock_lreq: 4569 mutex_unlock(&lreq->lock); 4570 out_unlock_osdc: 4571 up_read(&osdc->lock); 4572 return; 4573 4574 bad: 4575 pr_err("osdc handle_watch_notify corrupt msg\n"); 4576 } 4577 4578 /* 4579 * Register request, send initial attempt. 4580 */ 4581 void ceph_osdc_start_request(struct ceph_osd_client *osdc, 4582 struct ceph_osd_request *req) 4583 { 4584 down_read(&osdc->lock); 4585 submit_request(req, false); 4586 up_read(&osdc->lock); 4587 } 4588 EXPORT_SYMBOL(ceph_osdc_start_request); 4589 4590 /* 4591 * Unregister request. If @req was registered, it isn't completed: 4592 * r_result isn't set and __complete_request() isn't invoked. 4593 * 4594 * If @req wasn't registered, this call may have raced with 4595 * handle_reply(), in which case r_result would already be set and 4596 * __complete_request() would be getting invoked, possibly even 4597 * concurrently with this call. 4598 */ 4599 void ceph_osdc_cancel_request(struct ceph_osd_request *req) 4600 { 4601 struct ceph_osd_client *osdc = req->r_osdc; 4602 4603 down_write(&osdc->lock); 4604 if (req->r_osd) 4605 cancel_request(req); 4606 up_write(&osdc->lock); 4607 } 4608 EXPORT_SYMBOL(ceph_osdc_cancel_request); 4609 4610 /* 4611 * @timeout: in jiffies, 0 means "wait forever" 4612 */ 4613 static int wait_request_timeout(struct ceph_osd_request *req, 4614 unsigned long timeout) 4615 { 4616 long left; 4617 4618 dout("%s req %p tid %llu\n", __func__, req, req->r_tid); 4619 left = wait_for_completion_killable_timeout(&req->r_completion, 4620 ceph_timeout_jiffies(timeout)); 4621 if (left <= 0) { 4622 left = left ?: -ETIMEDOUT; 4623 ceph_osdc_cancel_request(req); 4624 } else { 4625 left = req->r_result; /* completed */ 4626 } 4627 4628 return left; 4629 } 4630 4631 /* 4632 * wait for a request to complete 4633 */ 4634 int ceph_osdc_wait_request(struct ceph_osd_client *osdc, 4635 struct ceph_osd_request *req) 4636 { 4637 return wait_request_timeout(req, 0); 4638 } 4639 EXPORT_SYMBOL(ceph_osdc_wait_request); 4640 4641 /* 4642 * sync - wait for all in-flight requests to flush. avoid starvation. 4643 */ 4644 void ceph_osdc_sync(struct ceph_osd_client *osdc) 4645 { 4646 struct rb_node *n, *p; 4647 u64 last_tid = atomic64_read(&osdc->last_tid); 4648 4649 again: 4650 down_read(&osdc->lock); 4651 for (n = rb_first(&osdc->osds); n; n = rb_next(n)) { 4652 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node); 4653 4654 mutex_lock(&osd->lock); 4655 for (p = rb_first(&osd->o_requests); p; p = rb_next(p)) { 4656 struct ceph_osd_request *req = 4657 rb_entry(p, struct ceph_osd_request, r_node); 4658 4659 if (req->r_tid > last_tid) 4660 break; 4661 4662 if (!(req->r_flags & CEPH_OSD_FLAG_WRITE)) 4663 continue; 4664 4665 ceph_osdc_get_request(req); 4666 mutex_unlock(&osd->lock); 4667 up_read(&osdc->lock); 4668 dout("%s waiting on req %p tid %llu last_tid %llu\n", 4669 __func__, req, req->r_tid, last_tid); 4670 wait_for_completion(&req->r_completion); 4671 ceph_osdc_put_request(req); 4672 goto again; 4673 } 4674 4675 mutex_unlock(&osd->lock); 4676 } 4677 4678 up_read(&osdc->lock); 4679 dout("%s done last_tid %llu\n", __func__, last_tid); 4680 } 4681 EXPORT_SYMBOL(ceph_osdc_sync); 4682 4683 /* 4684 * Returns a handle, caller owns a ref. 4685 */ 4686 struct ceph_osd_linger_request * 4687 ceph_osdc_watch(struct ceph_osd_client *osdc, 4688 struct ceph_object_id *oid, 4689 struct ceph_object_locator *oloc, 4690 rados_watchcb2_t wcb, 4691 rados_watcherrcb_t errcb, 4692 void *data) 4693 { 4694 struct ceph_osd_linger_request *lreq; 4695 int ret; 4696 4697 lreq = linger_alloc(osdc); 4698 if (!lreq) 4699 return ERR_PTR(-ENOMEM); 4700 4701 lreq->is_watch = true; 4702 lreq->wcb = wcb; 4703 lreq->errcb = errcb; 4704 lreq->data = data; 4705 lreq->watch_valid_thru = jiffies; 4706 4707 ceph_oid_copy(&lreq->t.base_oid, oid); 4708 ceph_oloc_copy(&lreq->t.base_oloc, oloc); 4709 lreq->t.flags = CEPH_OSD_FLAG_WRITE; 4710 ktime_get_real_ts64(&lreq->mtime); 4711 4712 linger_submit(lreq); 4713 ret = linger_reg_commit_wait(lreq); 4714 if (ret) { 4715 linger_cancel(lreq); 4716 goto err_put_lreq; 4717 } 4718 4719 return lreq; 4720 4721 err_put_lreq: 4722 linger_put(lreq); 4723 return ERR_PTR(ret); 4724 } 4725 EXPORT_SYMBOL(ceph_osdc_watch); 4726 4727 /* 4728 * Releases a ref. 4729 * 4730 * Times out after mount_timeout to preserve rbd unmap behaviour 4731 * introduced in 2894e1d76974 ("rbd: timeout watch teardown on unmap 4732 * with mount_timeout"). 4733 */ 4734 int ceph_osdc_unwatch(struct ceph_osd_client *osdc, 4735 struct ceph_osd_linger_request *lreq) 4736 { 4737 struct ceph_options *opts = osdc->client->options; 4738 struct ceph_osd_request *req; 4739 int ret; 4740 4741 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO); 4742 if (!req) 4743 return -ENOMEM; 4744 4745 ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid); 4746 ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc); 4747 req->r_flags = CEPH_OSD_FLAG_WRITE; 4748 ktime_get_real_ts64(&req->r_mtime); 4749 osd_req_op_watch_init(req, 0, CEPH_OSD_WATCH_OP_UNWATCH, 4750 lreq->linger_id, 0); 4751 4752 ret = ceph_osdc_alloc_messages(req, GFP_NOIO); 4753 if (ret) 4754 goto out_put_req; 4755 4756 ceph_osdc_start_request(osdc, req); 4757 linger_cancel(lreq); 4758 linger_put(lreq); 4759 ret = wait_request_timeout(req, opts->mount_timeout); 4760 4761 out_put_req: 4762 ceph_osdc_put_request(req); 4763 return ret; 4764 } 4765 EXPORT_SYMBOL(ceph_osdc_unwatch); 4766 4767 static int osd_req_op_notify_ack_init(struct ceph_osd_request *req, int which, 4768 u64 notify_id, u64 cookie, void *payload, 4769 u32 payload_len) 4770 { 4771 struct ceph_osd_req_op *op; 4772 struct ceph_pagelist *pl; 4773 int ret; 4774 4775 op = osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY_ACK, 0); 4776 4777 pl = ceph_pagelist_alloc(GFP_NOIO); 4778 if (!pl) 4779 return -ENOMEM; 4780 4781 ret = ceph_pagelist_encode_64(pl, notify_id); 4782 ret |= ceph_pagelist_encode_64(pl, cookie); 4783 if (payload) { 4784 ret |= ceph_pagelist_encode_32(pl, payload_len); 4785 ret |= ceph_pagelist_append(pl, payload, payload_len); 4786 } else { 4787 ret |= ceph_pagelist_encode_32(pl, 0); 4788 } 4789 if (ret) { 4790 ceph_pagelist_release(pl); 4791 return -ENOMEM; 4792 } 4793 4794 ceph_osd_data_pagelist_init(&op->notify_ack.request_data, pl); 4795 op->indata_len = pl->length; 4796 return 0; 4797 } 4798 4799 int ceph_osdc_notify_ack(struct ceph_osd_client *osdc, 4800 struct ceph_object_id *oid, 4801 struct ceph_object_locator *oloc, 4802 u64 notify_id, 4803 u64 cookie, 4804 void *payload, 4805 u32 payload_len) 4806 { 4807 struct ceph_osd_request *req; 4808 int ret; 4809 4810 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO); 4811 if (!req) 4812 return -ENOMEM; 4813 4814 ceph_oid_copy(&req->r_base_oid, oid); 4815 ceph_oloc_copy(&req->r_base_oloc, oloc); 4816 req->r_flags = CEPH_OSD_FLAG_READ; 4817 4818 ret = osd_req_op_notify_ack_init(req, 0, notify_id, cookie, payload, 4819 payload_len); 4820 if (ret) 4821 goto out_put_req; 4822 4823 ret = ceph_osdc_alloc_messages(req, GFP_NOIO); 4824 if (ret) 4825 goto out_put_req; 4826 4827 ceph_osdc_start_request(osdc, req); 4828 ret = ceph_osdc_wait_request(osdc, req); 4829 4830 out_put_req: 4831 ceph_osdc_put_request(req); 4832 return ret; 4833 } 4834 EXPORT_SYMBOL(ceph_osdc_notify_ack); 4835 4836 /* 4837 * @timeout: in seconds 4838 * 4839 * @preply_{pages,len} are initialized both on success and error. 4840 * The caller is responsible for: 4841 * 4842 * ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len)) 4843 */ 4844 int ceph_osdc_notify(struct ceph_osd_client *osdc, 4845 struct ceph_object_id *oid, 4846 struct ceph_object_locator *oloc, 4847 void *payload, 4848 u32 payload_len, 4849 u32 timeout, 4850 struct page ***preply_pages, 4851 size_t *preply_len) 4852 { 4853 struct ceph_osd_linger_request *lreq; 4854 int ret; 4855 4856 WARN_ON(!timeout); 4857 if (preply_pages) { 4858 *preply_pages = NULL; 4859 *preply_len = 0; 4860 } 4861 4862 lreq = linger_alloc(osdc); 4863 if (!lreq) 4864 return -ENOMEM; 4865 4866 lreq->request_pl = ceph_pagelist_alloc(GFP_NOIO); 4867 if (!lreq->request_pl) { 4868 ret = -ENOMEM; 4869 goto out_put_lreq; 4870 } 4871 4872 ret = ceph_pagelist_encode_32(lreq->request_pl, 1); /* prot_ver */ 4873 ret |= ceph_pagelist_encode_32(lreq->request_pl, timeout); 4874 ret |= ceph_pagelist_encode_32(lreq->request_pl, payload_len); 4875 ret |= ceph_pagelist_append(lreq->request_pl, payload, payload_len); 4876 if (ret) { 4877 ret = -ENOMEM; 4878 goto out_put_lreq; 4879 } 4880 4881 /* for notify_id */ 4882 lreq->notify_id_pages = ceph_alloc_page_vector(1, GFP_NOIO); 4883 if (IS_ERR(lreq->notify_id_pages)) { 4884 ret = PTR_ERR(lreq->notify_id_pages); 4885 lreq->notify_id_pages = NULL; 4886 goto out_put_lreq; 4887 } 4888 4889 lreq->preply_pages = preply_pages; 4890 lreq->preply_len = preply_len; 4891 4892 ceph_oid_copy(&lreq->t.base_oid, oid); 4893 ceph_oloc_copy(&lreq->t.base_oloc, oloc); 4894 lreq->t.flags = CEPH_OSD_FLAG_READ; 4895 4896 linger_submit(lreq); 4897 ret = linger_reg_commit_wait(lreq); 4898 if (!ret) 4899 ret = linger_notify_finish_wait(lreq); 4900 else 4901 dout("lreq %p failed to initiate notify %d\n", lreq, ret); 4902 4903 linger_cancel(lreq); 4904 out_put_lreq: 4905 linger_put(lreq); 4906 return ret; 4907 } 4908 EXPORT_SYMBOL(ceph_osdc_notify); 4909 4910 /* 4911 * Return the number of milliseconds since the watch was last 4912 * confirmed, or an error. If there is an error, the watch is no 4913 * longer valid, and should be destroyed with ceph_osdc_unwatch(). 4914 */ 4915 int ceph_osdc_watch_check(struct ceph_osd_client *osdc, 4916 struct ceph_osd_linger_request *lreq) 4917 { 4918 unsigned long stamp, age; 4919 int ret; 4920 4921 down_read(&osdc->lock); 4922 mutex_lock(&lreq->lock); 4923 stamp = lreq->watch_valid_thru; 4924 if (!list_empty(&lreq->pending_lworks)) { 4925 struct linger_work *lwork = 4926 list_first_entry(&lreq->pending_lworks, 4927 struct linger_work, 4928 pending_item); 4929 4930 if (time_before(lwork->queued_stamp, stamp)) 4931 stamp = lwork->queued_stamp; 4932 } 4933 age = jiffies - stamp; 4934 dout("%s lreq %p linger_id %llu age %lu last_error %d\n", __func__, 4935 lreq, lreq->linger_id, age, lreq->last_error); 4936 /* we are truncating to msecs, so return a safe upper bound */ 4937 ret = lreq->last_error ?: 1 + jiffies_to_msecs(age); 4938 4939 mutex_unlock(&lreq->lock); 4940 up_read(&osdc->lock); 4941 return ret; 4942 } 4943 4944 static int decode_watcher(void **p, void *end, struct ceph_watch_item *item) 4945 { 4946 u8 struct_v; 4947 u32 struct_len; 4948 int ret; 4949 4950 ret = ceph_start_decoding(p, end, 2, "watch_item_t", 4951 &struct_v, &struct_len); 4952 if (ret) 4953 goto bad; 4954 4955 ret = -EINVAL; 4956 ceph_decode_copy_safe(p, end, &item->name, sizeof(item->name), bad); 4957 ceph_decode_64_safe(p, end, item->cookie, bad); 4958 ceph_decode_skip_32(p, end, bad); /* skip timeout seconds */ 4959 4960 if (struct_v >= 2) { 4961 ret = ceph_decode_entity_addr(p, end, &item->addr); 4962 if (ret) 4963 goto bad; 4964 } else { 4965 ret = 0; 4966 } 4967 4968 dout("%s %s%llu cookie %llu addr %s\n", __func__, 4969 ENTITY_NAME(item->name), item->cookie, 4970 ceph_pr_addr(&item->addr)); 4971 bad: 4972 return ret; 4973 } 4974 4975 static int decode_watchers(void **p, void *end, 4976 struct ceph_watch_item **watchers, 4977 u32 *num_watchers) 4978 { 4979 u8 struct_v; 4980 u32 struct_len; 4981 int i; 4982 int ret; 4983 4984 ret = ceph_start_decoding(p, end, 1, "obj_list_watch_response_t", 4985 &struct_v, &struct_len); 4986 if (ret) 4987 return ret; 4988 4989 *num_watchers = ceph_decode_32(p); 4990 *watchers = kcalloc(*num_watchers, sizeof(**watchers), GFP_NOIO); 4991 if (!*watchers) 4992 return -ENOMEM; 4993 4994 for (i = 0; i < *num_watchers; i++) { 4995 ret = decode_watcher(p, end, *watchers + i); 4996 if (ret) { 4997 kfree(*watchers); 4998 return ret; 4999 } 5000 } 5001 5002 return 0; 5003 } 5004 5005 /* 5006 * On success, the caller is responsible for: 5007 * 5008 * kfree(watchers); 5009 */ 5010 int ceph_osdc_list_watchers(struct ceph_osd_client *osdc, 5011 struct ceph_object_id *oid, 5012 struct ceph_object_locator *oloc, 5013 struct ceph_watch_item **watchers, 5014 u32 *num_watchers) 5015 { 5016 struct ceph_osd_request *req; 5017 struct page **pages; 5018 int ret; 5019 5020 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO); 5021 if (!req) 5022 return -ENOMEM; 5023 5024 ceph_oid_copy(&req->r_base_oid, oid); 5025 ceph_oloc_copy(&req->r_base_oloc, oloc); 5026 req->r_flags = CEPH_OSD_FLAG_READ; 5027 5028 pages = ceph_alloc_page_vector(1, GFP_NOIO); 5029 if (IS_ERR(pages)) { 5030 ret = PTR_ERR(pages); 5031 goto out_put_req; 5032 } 5033 5034 osd_req_op_init(req, 0, CEPH_OSD_OP_LIST_WATCHERS, 0); 5035 ceph_osd_data_pages_init(osd_req_op_data(req, 0, list_watchers, 5036 response_data), 5037 pages, PAGE_SIZE, 0, false, true); 5038 5039 ret = ceph_osdc_alloc_messages(req, GFP_NOIO); 5040 if (ret) 5041 goto out_put_req; 5042 5043 ceph_osdc_start_request(osdc, req); 5044 ret = ceph_osdc_wait_request(osdc, req); 5045 if (ret >= 0) { 5046 void *p = page_address(pages[0]); 5047 void *const end = p + req->r_ops[0].outdata_len; 5048 5049 ret = decode_watchers(&p, end, watchers, num_watchers); 5050 } 5051 5052 out_put_req: 5053 ceph_osdc_put_request(req); 5054 return ret; 5055 } 5056 EXPORT_SYMBOL(ceph_osdc_list_watchers); 5057 5058 /* 5059 * Call all pending notify callbacks - for use after a watch is 5060 * unregistered, to make sure no more callbacks for it will be invoked 5061 */ 5062 void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc) 5063 { 5064 dout("%s osdc %p\n", __func__, osdc); 5065 flush_workqueue(osdc->notify_wq); 5066 } 5067 EXPORT_SYMBOL(ceph_osdc_flush_notifies); 5068 5069 void ceph_osdc_maybe_request_map(struct ceph_osd_client *osdc) 5070 { 5071 down_read(&osdc->lock); 5072 maybe_request_map(osdc); 5073 up_read(&osdc->lock); 5074 } 5075 EXPORT_SYMBOL(ceph_osdc_maybe_request_map); 5076 5077 /* 5078 * Execute an OSD class method on an object. 5079 * 5080 * @flags: CEPH_OSD_FLAG_* 5081 * @resp_len: in/out param for reply length 5082 */ 5083 int ceph_osdc_call(struct ceph_osd_client *osdc, 5084 struct ceph_object_id *oid, 5085 struct ceph_object_locator *oloc, 5086 const char *class, const char *method, 5087 unsigned int flags, 5088 struct page *req_page, size_t req_len, 5089 struct page **resp_pages, size_t *resp_len) 5090 { 5091 struct ceph_osd_request *req; 5092 int ret; 5093 5094 if (req_len > PAGE_SIZE) 5095 return -E2BIG; 5096 5097 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO); 5098 if (!req) 5099 return -ENOMEM; 5100 5101 ceph_oid_copy(&req->r_base_oid, oid); 5102 ceph_oloc_copy(&req->r_base_oloc, oloc); 5103 req->r_flags = flags; 5104 5105 ret = osd_req_op_cls_init(req, 0, class, method); 5106 if (ret) 5107 goto out_put_req; 5108 5109 if (req_page) 5110 osd_req_op_cls_request_data_pages(req, 0, &req_page, req_len, 5111 0, false, false); 5112 if (resp_pages) 5113 osd_req_op_cls_response_data_pages(req, 0, resp_pages, 5114 *resp_len, 0, false, false); 5115 5116 ret = ceph_osdc_alloc_messages(req, GFP_NOIO); 5117 if (ret) 5118 goto out_put_req; 5119 5120 ceph_osdc_start_request(osdc, req); 5121 ret = ceph_osdc_wait_request(osdc, req); 5122 if (ret >= 0) { 5123 ret = req->r_ops[0].rval; 5124 if (resp_pages) 5125 *resp_len = req->r_ops[0].outdata_len; 5126 } 5127 5128 out_put_req: 5129 ceph_osdc_put_request(req); 5130 return ret; 5131 } 5132 EXPORT_SYMBOL(ceph_osdc_call); 5133 5134 /* 5135 * reset all osd connections 5136 */ 5137 void ceph_osdc_reopen_osds(struct ceph_osd_client *osdc) 5138 { 5139 struct rb_node *n; 5140 5141 down_write(&osdc->lock); 5142 for (n = rb_first(&osdc->osds); n; ) { 5143 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node); 5144 5145 n = rb_next(n); 5146 if (!reopen_osd(osd)) 5147 kick_osd_requests(osd); 5148 } 5149 up_write(&osdc->lock); 5150 } 5151 5152 /* 5153 * init, shutdown 5154 */ 5155 int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client) 5156 { 5157 int err; 5158 5159 dout("init\n"); 5160 osdc->client = client; 5161 init_rwsem(&osdc->lock); 5162 osdc->osds = RB_ROOT; 5163 INIT_LIST_HEAD(&osdc->osd_lru); 5164 spin_lock_init(&osdc->osd_lru_lock); 5165 osd_init(&osdc->homeless_osd); 5166 osdc->homeless_osd.o_osdc = osdc; 5167 osdc->homeless_osd.o_osd = CEPH_HOMELESS_OSD; 5168 osdc->last_linger_id = CEPH_LINGER_ID_START; 5169 osdc->linger_requests = RB_ROOT; 5170 osdc->map_checks = RB_ROOT; 5171 osdc->linger_map_checks = RB_ROOT; 5172 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout); 5173 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout); 5174 5175 err = -ENOMEM; 5176 osdc->osdmap = ceph_osdmap_alloc(); 5177 if (!osdc->osdmap) 5178 goto out; 5179 5180 osdc->req_mempool = mempool_create_slab_pool(10, 5181 ceph_osd_request_cache); 5182 if (!osdc->req_mempool) 5183 goto out_map; 5184 5185 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP, 5186 PAGE_SIZE, CEPH_OSD_SLAB_OPS, 10, "osd_op"); 5187 if (err < 0) 5188 goto out_mempool; 5189 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY, 5190 PAGE_SIZE, CEPH_OSD_SLAB_OPS, 10, 5191 "osd_op_reply"); 5192 if (err < 0) 5193 goto out_msgpool; 5194 5195 err = -ENOMEM; 5196 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify"); 5197 if (!osdc->notify_wq) 5198 goto out_msgpool_reply; 5199 5200 osdc->completion_wq = create_singlethread_workqueue("ceph-completion"); 5201 if (!osdc->completion_wq) 5202 goto out_notify_wq; 5203 5204 schedule_delayed_work(&osdc->timeout_work, 5205 osdc->client->options->osd_keepalive_timeout); 5206 schedule_delayed_work(&osdc->osds_timeout_work, 5207 round_jiffies_relative(osdc->client->options->osd_idle_ttl)); 5208 5209 return 0; 5210 5211 out_notify_wq: 5212 destroy_workqueue(osdc->notify_wq); 5213 out_msgpool_reply: 5214 ceph_msgpool_destroy(&osdc->msgpool_op_reply); 5215 out_msgpool: 5216 ceph_msgpool_destroy(&osdc->msgpool_op); 5217 out_mempool: 5218 mempool_destroy(osdc->req_mempool); 5219 out_map: 5220 ceph_osdmap_destroy(osdc->osdmap); 5221 out: 5222 return err; 5223 } 5224 5225 void ceph_osdc_stop(struct ceph_osd_client *osdc) 5226 { 5227 destroy_workqueue(osdc->completion_wq); 5228 destroy_workqueue(osdc->notify_wq); 5229 cancel_delayed_work_sync(&osdc->timeout_work); 5230 cancel_delayed_work_sync(&osdc->osds_timeout_work); 5231 5232 down_write(&osdc->lock); 5233 while (!RB_EMPTY_ROOT(&osdc->osds)) { 5234 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds), 5235 struct ceph_osd, o_node); 5236 close_osd(osd); 5237 } 5238 up_write(&osdc->lock); 5239 WARN_ON(refcount_read(&osdc->homeless_osd.o_ref) != 1); 5240 osd_cleanup(&osdc->homeless_osd); 5241 5242 WARN_ON(!list_empty(&osdc->osd_lru)); 5243 WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_requests)); 5244 WARN_ON(!RB_EMPTY_ROOT(&osdc->map_checks)); 5245 WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_map_checks)); 5246 WARN_ON(atomic_read(&osdc->num_requests)); 5247 WARN_ON(atomic_read(&osdc->num_homeless)); 5248 5249 ceph_osdmap_destroy(osdc->osdmap); 5250 mempool_destroy(osdc->req_mempool); 5251 ceph_msgpool_destroy(&osdc->msgpool_op); 5252 ceph_msgpool_destroy(&osdc->msgpool_op_reply); 5253 } 5254 5255 int osd_req_op_copy_from_init(struct ceph_osd_request *req, 5256 u64 src_snapid, u64 src_version, 5257 struct ceph_object_id *src_oid, 5258 struct ceph_object_locator *src_oloc, 5259 u32 src_fadvise_flags, 5260 u32 dst_fadvise_flags, 5261 u32 truncate_seq, u64 truncate_size, 5262 u8 copy_from_flags) 5263 { 5264 struct ceph_osd_req_op *op; 5265 struct page **pages; 5266 void *p, *end; 5267 5268 pages = ceph_alloc_page_vector(1, GFP_KERNEL); 5269 if (IS_ERR(pages)) 5270 return PTR_ERR(pages); 5271 5272 op = osd_req_op_init(req, 0, CEPH_OSD_OP_COPY_FROM2, 5273 dst_fadvise_flags); 5274 op->copy_from.snapid = src_snapid; 5275 op->copy_from.src_version = src_version; 5276 op->copy_from.flags = copy_from_flags; 5277 op->copy_from.src_fadvise_flags = src_fadvise_flags; 5278 5279 p = page_address(pages[0]); 5280 end = p + PAGE_SIZE; 5281 ceph_encode_string(&p, end, src_oid->name, src_oid->name_len); 5282 encode_oloc(&p, end, src_oloc); 5283 ceph_encode_32(&p, truncate_seq); 5284 ceph_encode_64(&p, truncate_size); 5285 op->indata_len = PAGE_SIZE - (end - p); 5286 5287 ceph_osd_data_pages_init(&op->copy_from.osd_data, pages, 5288 op->indata_len, 0, false, true); 5289 return 0; 5290 } 5291 EXPORT_SYMBOL(osd_req_op_copy_from_init); 5292 5293 int __init ceph_osdc_setup(void) 5294 { 5295 size_t size = sizeof(struct ceph_osd_request) + 5296 CEPH_OSD_SLAB_OPS * sizeof(struct ceph_osd_req_op); 5297 5298 BUG_ON(ceph_osd_request_cache); 5299 ceph_osd_request_cache = kmem_cache_create("ceph_osd_request", size, 5300 0, 0, NULL); 5301 5302 return ceph_osd_request_cache ? 0 : -ENOMEM; 5303 } 5304 5305 void ceph_osdc_cleanup(void) 5306 { 5307 BUG_ON(!ceph_osd_request_cache); 5308 kmem_cache_destroy(ceph_osd_request_cache); 5309 ceph_osd_request_cache = NULL; 5310 } 5311 5312 /* 5313 * handle incoming message 5314 */ 5315 static void osd_dispatch(struct ceph_connection *con, struct ceph_msg *msg) 5316 { 5317 struct ceph_osd *osd = con->private; 5318 struct ceph_osd_client *osdc = osd->o_osdc; 5319 int type = le16_to_cpu(msg->hdr.type); 5320 5321 switch (type) { 5322 case CEPH_MSG_OSD_MAP: 5323 ceph_osdc_handle_map(osdc, msg); 5324 break; 5325 case CEPH_MSG_OSD_OPREPLY: 5326 handle_reply(osd, msg); 5327 break; 5328 case CEPH_MSG_OSD_BACKOFF: 5329 handle_backoff(osd, msg); 5330 break; 5331 case CEPH_MSG_WATCH_NOTIFY: 5332 handle_watch_notify(osdc, msg); 5333 break; 5334 5335 default: 5336 pr_err("received unknown message type %d %s\n", type, 5337 ceph_msg_type_name(type)); 5338 } 5339 5340 ceph_msg_put(msg); 5341 } 5342 5343 /* 5344 * Lookup and return message for incoming reply. Don't try to do 5345 * anything about a larger than preallocated data portion of the 5346 * message at the moment - for now, just skip the message. 5347 */ 5348 static struct ceph_msg *get_reply(struct ceph_connection *con, 5349 struct ceph_msg_header *hdr, 5350 int *skip) 5351 { 5352 struct ceph_osd *osd = con->private; 5353 struct ceph_osd_client *osdc = osd->o_osdc; 5354 struct ceph_msg *m = NULL; 5355 struct ceph_osd_request *req; 5356 int front_len = le32_to_cpu(hdr->front_len); 5357 int data_len = le32_to_cpu(hdr->data_len); 5358 u64 tid = le64_to_cpu(hdr->tid); 5359 5360 down_read(&osdc->lock); 5361 if (!osd_registered(osd)) { 5362 dout("%s osd%d unknown, skipping\n", __func__, osd->o_osd); 5363 *skip = 1; 5364 goto out_unlock_osdc; 5365 } 5366 WARN_ON(osd->o_osd != le64_to_cpu(hdr->src.num)); 5367 5368 mutex_lock(&osd->lock); 5369 req = lookup_request(&osd->o_requests, tid); 5370 if (!req) { 5371 dout("%s osd%d tid %llu unknown, skipping\n", __func__, 5372 osd->o_osd, tid); 5373 *skip = 1; 5374 goto out_unlock_session; 5375 } 5376 5377 ceph_msg_revoke_incoming(req->r_reply); 5378 5379 if (front_len > req->r_reply->front_alloc_len) { 5380 pr_warn("%s osd%d tid %llu front %d > preallocated %d\n", 5381 __func__, osd->o_osd, req->r_tid, front_len, 5382 req->r_reply->front_alloc_len); 5383 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS, 5384 false); 5385 if (!m) 5386 goto out_unlock_session; 5387 ceph_msg_put(req->r_reply); 5388 req->r_reply = m; 5389 } 5390 5391 if (data_len > req->r_reply->data_length) { 5392 pr_warn("%s osd%d tid %llu data %d > preallocated %zu, skipping\n", 5393 __func__, osd->o_osd, req->r_tid, data_len, 5394 req->r_reply->data_length); 5395 m = NULL; 5396 *skip = 1; 5397 goto out_unlock_session; 5398 } 5399 5400 m = ceph_msg_get(req->r_reply); 5401 dout("get_reply tid %lld %p\n", tid, m); 5402 5403 out_unlock_session: 5404 mutex_unlock(&osd->lock); 5405 out_unlock_osdc: 5406 up_read(&osdc->lock); 5407 return m; 5408 } 5409 5410 static struct ceph_msg *alloc_msg_with_page_vector(struct ceph_msg_header *hdr) 5411 { 5412 struct ceph_msg *m; 5413 int type = le16_to_cpu(hdr->type); 5414 u32 front_len = le32_to_cpu(hdr->front_len); 5415 u32 data_len = le32_to_cpu(hdr->data_len); 5416 5417 m = ceph_msg_new2(type, front_len, 1, GFP_NOIO, false); 5418 if (!m) 5419 return NULL; 5420 5421 if (data_len) { 5422 struct page **pages; 5423 5424 pages = ceph_alloc_page_vector(calc_pages_for(0, data_len), 5425 GFP_NOIO); 5426 if (IS_ERR(pages)) { 5427 ceph_msg_put(m); 5428 return NULL; 5429 } 5430 5431 ceph_msg_data_add_pages(m, pages, data_len, 0, true); 5432 } 5433 5434 return m; 5435 } 5436 5437 static struct ceph_msg *osd_alloc_msg(struct ceph_connection *con, 5438 struct ceph_msg_header *hdr, 5439 int *skip) 5440 { 5441 struct ceph_osd *osd = con->private; 5442 int type = le16_to_cpu(hdr->type); 5443 5444 *skip = 0; 5445 switch (type) { 5446 case CEPH_MSG_OSD_MAP: 5447 case CEPH_MSG_OSD_BACKOFF: 5448 case CEPH_MSG_WATCH_NOTIFY: 5449 return alloc_msg_with_page_vector(hdr); 5450 case CEPH_MSG_OSD_OPREPLY: 5451 return get_reply(con, hdr, skip); 5452 default: 5453 pr_warn("%s osd%d unknown msg type %d, skipping\n", __func__, 5454 osd->o_osd, type); 5455 *skip = 1; 5456 return NULL; 5457 } 5458 } 5459 5460 /* 5461 * Wrappers to refcount containing ceph_osd struct 5462 */ 5463 static struct ceph_connection *osd_get_con(struct ceph_connection *con) 5464 { 5465 struct ceph_osd *osd = con->private; 5466 if (get_osd(osd)) 5467 return con; 5468 return NULL; 5469 } 5470 5471 static void osd_put_con(struct ceph_connection *con) 5472 { 5473 struct ceph_osd *osd = con->private; 5474 put_osd(osd); 5475 } 5476 5477 /* 5478 * authentication 5479 */ 5480 5481 /* 5482 * Note: returned pointer is the address of a structure that's 5483 * managed separately. Caller must *not* attempt to free it. 5484 */ 5485 static struct ceph_auth_handshake * 5486 osd_get_authorizer(struct ceph_connection *con, int *proto, int force_new) 5487 { 5488 struct ceph_osd *o = con->private; 5489 struct ceph_osd_client *osdc = o->o_osdc; 5490 struct ceph_auth_client *ac = osdc->client->monc.auth; 5491 struct ceph_auth_handshake *auth = &o->o_auth; 5492 int ret; 5493 5494 ret = __ceph_auth_get_authorizer(ac, auth, CEPH_ENTITY_TYPE_OSD, 5495 force_new, proto, NULL, NULL); 5496 if (ret) 5497 return ERR_PTR(ret); 5498 5499 return auth; 5500 } 5501 5502 static int osd_add_authorizer_challenge(struct ceph_connection *con, 5503 void *challenge_buf, int challenge_buf_len) 5504 { 5505 struct ceph_osd *o = con->private; 5506 struct ceph_osd_client *osdc = o->o_osdc; 5507 struct ceph_auth_client *ac = osdc->client->monc.auth; 5508 5509 return ceph_auth_add_authorizer_challenge(ac, o->o_auth.authorizer, 5510 challenge_buf, challenge_buf_len); 5511 } 5512 5513 static int osd_verify_authorizer_reply(struct ceph_connection *con) 5514 { 5515 struct ceph_osd *o = con->private; 5516 struct ceph_osd_client *osdc = o->o_osdc; 5517 struct ceph_auth_client *ac = osdc->client->monc.auth; 5518 struct ceph_auth_handshake *auth = &o->o_auth; 5519 5520 return ceph_auth_verify_authorizer_reply(ac, auth->authorizer, 5521 auth->authorizer_reply_buf, auth->authorizer_reply_buf_len, 5522 NULL, NULL, NULL, NULL); 5523 } 5524 5525 static int osd_invalidate_authorizer(struct ceph_connection *con) 5526 { 5527 struct ceph_osd *o = con->private; 5528 struct ceph_osd_client *osdc = o->o_osdc; 5529 struct ceph_auth_client *ac = osdc->client->monc.auth; 5530 5531 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD); 5532 return ceph_monc_validate_auth(&osdc->client->monc); 5533 } 5534 5535 static int osd_get_auth_request(struct ceph_connection *con, 5536 void *buf, int *buf_len, 5537 void **authorizer, int *authorizer_len) 5538 { 5539 struct ceph_osd *o = con->private; 5540 struct ceph_auth_client *ac = o->o_osdc->client->monc.auth; 5541 struct ceph_auth_handshake *auth = &o->o_auth; 5542 int ret; 5543 5544 ret = ceph_auth_get_authorizer(ac, auth, CEPH_ENTITY_TYPE_OSD, 5545 buf, buf_len); 5546 if (ret) 5547 return ret; 5548 5549 *authorizer = auth->authorizer_buf; 5550 *authorizer_len = auth->authorizer_buf_len; 5551 return 0; 5552 } 5553 5554 static int osd_handle_auth_reply_more(struct ceph_connection *con, 5555 void *reply, int reply_len, 5556 void *buf, int *buf_len, 5557 void **authorizer, int *authorizer_len) 5558 { 5559 struct ceph_osd *o = con->private; 5560 struct ceph_auth_client *ac = o->o_osdc->client->monc.auth; 5561 struct ceph_auth_handshake *auth = &o->o_auth; 5562 int ret; 5563 5564 ret = ceph_auth_handle_svc_reply_more(ac, auth, reply, reply_len, 5565 buf, buf_len); 5566 if (ret) 5567 return ret; 5568 5569 *authorizer = auth->authorizer_buf; 5570 *authorizer_len = auth->authorizer_buf_len; 5571 return 0; 5572 } 5573 5574 static int osd_handle_auth_done(struct ceph_connection *con, 5575 u64 global_id, void *reply, int reply_len, 5576 u8 *session_key, int *session_key_len, 5577 u8 *con_secret, int *con_secret_len) 5578 { 5579 struct ceph_osd *o = con->private; 5580 struct ceph_auth_client *ac = o->o_osdc->client->monc.auth; 5581 struct ceph_auth_handshake *auth = &o->o_auth; 5582 5583 return ceph_auth_handle_svc_reply_done(ac, auth, reply, reply_len, 5584 session_key, session_key_len, 5585 con_secret, con_secret_len); 5586 } 5587 5588 static int osd_handle_auth_bad_method(struct ceph_connection *con, 5589 int used_proto, int result, 5590 const int *allowed_protos, int proto_cnt, 5591 const int *allowed_modes, int mode_cnt) 5592 { 5593 struct ceph_osd *o = con->private; 5594 struct ceph_mon_client *monc = &o->o_osdc->client->monc; 5595 int ret; 5596 5597 if (ceph_auth_handle_bad_authorizer(monc->auth, CEPH_ENTITY_TYPE_OSD, 5598 used_proto, result, 5599 allowed_protos, proto_cnt, 5600 allowed_modes, mode_cnt)) { 5601 ret = ceph_monc_validate_auth(monc); 5602 if (ret) 5603 return ret; 5604 } 5605 5606 return -EACCES; 5607 } 5608 5609 static void osd_reencode_message(struct ceph_msg *msg) 5610 { 5611 int type = le16_to_cpu(msg->hdr.type); 5612 5613 if (type == CEPH_MSG_OSD_OP) 5614 encode_request_finish(msg); 5615 } 5616 5617 static int osd_sign_message(struct ceph_msg *msg) 5618 { 5619 struct ceph_osd *o = msg->con->private; 5620 struct ceph_auth_handshake *auth = &o->o_auth; 5621 5622 return ceph_auth_sign_message(auth, msg); 5623 } 5624 5625 static int osd_check_message_signature(struct ceph_msg *msg) 5626 { 5627 struct ceph_osd *o = msg->con->private; 5628 struct ceph_auth_handshake *auth = &o->o_auth; 5629 5630 return ceph_auth_check_message_signature(auth, msg); 5631 } 5632 5633 static const struct ceph_connection_operations osd_con_ops = { 5634 .get = osd_get_con, 5635 .put = osd_put_con, 5636 .alloc_msg = osd_alloc_msg, 5637 .dispatch = osd_dispatch, 5638 .fault = osd_fault, 5639 .reencode_message = osd_reencode_message, 5640 .get_authorizer = osd_get_authorizer, 5641 .add_authorizer_challenge = osd_add_authorizer_challenge, 5642 .verify_authorizer_reply = osd_verify_authorizer_reply, 5643 .invalidate_authorizer = osd_invalidate_authorizer, 5644 .sign_message = osd_sign_message, 5645 .check_message_signature = osd_check_message_signature, 5646 .get_auth_request = osd_get_auth_request, 5647 .handle_auth_reply_more = osd_handle_auth_reply_more, 5648 .handle_auth_done = osd_handle_auth_done, 5649 .handle_auth_bad_method = osd_handle_auth_bad_method, 5650 }; 5651