net/ceph/osd_client.c

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