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 /*
 130  * Consumes @pages if @own_pages is true.
 131  */
 132 static void ceph_osd_data_pages_init(struct ceph_osd_data *osd_data,
 133                         struct page **pages, u64 length, u32 alignment,
 134                         bool pages_from_pool, bool own_pages)
 135 {
 136         osd_data->type = CEPH_OSD_DATA_TYPE_PAGES;
 137         osd_data->pages = pages;
 138         osd_data->length = length;
 139         osd_data->alignment = alignment;
 140         osd_data->pages_from_pool = pages_from_pool;
 141         osd_data->own_pages = own_pages;
 142 }
 143
 144 /*
 145  * Consumes a ref on @pagelist.
 146  */
 147 static void ceph_osd_data_pagelist_init(struct ceph_osd_data *osd_data,
 148                         struct ceph_pagelist *pagelist)
 149 {
 150         osd_data->type = CEPH_OSD_DATA_TYPE_PAGELIST;
 151         osd_data->pagelist = pagelist;
 152 }
 153
 154 #ifdef CONFIG_BLOCK
 155 static void ceph_osd_data_bio_init(struct ceph_osd_data *osd_data,
 156                                    struct ceph_bio_iter *bio_pos,
 157                                    u32 bio_length)
 158 {
 159         osd_data->type = CEPH_OSD_DATA_TYPE_BIO;
 160         osd_data->bio_pos = *bio_pos;
 161         osd_data->bio_length = bio_length;
 162 }
 163 #endif /* CONFIG_BLOCK */
 164
 165 static void ceph_osd_data_bvecs_init(struct ceph_osd_data *osd_data,
 166                                      struct ceph_bvec_iter *bvec_pos,
 167                                      u32 num_bvecs)
 168 {
 169         osd_data->type = CEPH_OSD_DATA_TYPE_BVECS;
 170         osd_data->bvec_pos = *bvec_pos;
 171         osd_data->num_bvecs = num_bvecs;
 172 }
 173
 174 static struct ceph_osd_data *
 175 osd_req_op_raw_data_in(struct ceph_osd_request *osd_req, unsigned int which)
 176 {
 177         BUG_ON(which >= osd_req->r_num_ops);
 178
 179         return &osd_req->r_ops[which].raw_data_in;
 180 }
 181
 182 struct ceph_osd_data *
 183 osd_req_op_extent_osd_data(struct ceph_osd_request *osd_req,
 184                         unsigned int which)
 185 {
 186         return osd_req_op_data(osd_req, which, extent, osd_data);
 187 }
 188 EXPORT_SYMBOL(osd_req_op_extent_osd_data);
 189
 190 void osd_req_op_raw_data_in_pages(struct ceph_osd_request *osd_req,
 191                         unsigned int which, struct page **pages,
 192                         u64 length, u32 alignment,
 193                         bool pages_from_pool, bool own_pages)
 194 {
 195         struct ceph_osd_data *osd_data;
 196
 197         osd_data = osd_req_op_raw_data_in(osd_req, which);
 198         ceph_osd_data_pages_init(osd_data, pages, length, alignment,
 199                                 pages_from_pool, own_pages);
 200 }
 201 EXPORT_SYMBOL(osd_req_op_raw_data_in_pages);
 202
 203 void osd_req_op_extent_osd_data_pages(struct ceph_osd_request *osd_req,
 204                         unsigned int which, struct page **pages,
 205                         u64 length, u32 alignment,
 206                         bool pages_from_pool, bool own_pages)
 207 {
 208         struct ceph_osd_data *osd_data;
 209
 210         osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
 211         ceph_osd_data_pages_init(osd_data, pages, length, alignment,
 212                                 pages_from_pool, own_pages);
 213 }
 214 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pages);
 215
 216 void osd_req_op_extent_osd_data_pagelist(struct ceph_osd_request *osd_req,
 217                         unsigned int which, struct ceph_pagelist *pagelist)
 218 {
 219         struct ceph_osd_data *osd_data;
 220
 221         osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
 222         ceph_osd_data_pagelist_init(osd_data, pagelist);
 223 }
 224 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pagelist);
 225
 226 #ifdef CONFIG_BLOCK
 227 void osd_req_op_extent_osd_data_bio(struct ceph_osd_request *osd_req,
 228                                     unsigned int which,
 229                                     struct ceph_bio_iter *bio_pos,
 230                                     u32 bio_length)
 231 {
 232         struct ceph_osd_data *osd_data;
 233
 234         osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
 235         ceph_osd_data_bio_init(osd_data, bio_pos, bio_length);
 236 }
 237 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bio);
 238 #endif /* CONFIG_BLOCK */
 239
 240 void osd_req_op_extent_osd_data_bvecs(struct ceph_osd_request *osd_req,
 241                                       unsigned int which,
 242                                       struct bio_vec *bvecs, u32 num_bvecs,
 243                                       u32 bytes)
 244 {
 245         struct ceph_osd_data *osd_data;
 246         struct ceph_bvec_iter it = {
 247                 .bvecs = bvecs,
 248                 .iter = { .bi_size = bytes },
 249         };
 250
 251         osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
 252         ceph_osd_data_bvecs_init(osd_data, &it, num_bvecs);
 253 }
 254 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bvecs);
 255
 256 void osd_req_op_extent_osd_data_bvec_pos(struct ceph_osd_request *osd_req,
 257                                          unsigned int which,
 258                                          struct ceph_bvec_iter *bvec_pos)
 259 {
 260         struct ceph_osd_data *osd_data;
 261
 262         osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
 263         ceph_osd_data_bvecs_init(osd_data, bvec_pos, 0);
 264 }
 265 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bvec_pos);
 266
 267 static void osd_req_op_cls_request_info_pagelist(
 268                         struct ceph_osd_request *osd_req,
 269                         unsigned int which, struct ceph_pagelist *pagelist)
 270 {
 271         struct ceph_osd_data *osd_data;
 272
 273         osd_data = osd_req_op_data(osd_req, which, cls, request_info);
 274         ceph_osd_data_pagelist_init(osd_data, pagelist);
 275 }
 276
 277 void osd_req_op_cls_request_data_pagelist(
 278                         struct ceph_osd_request *osd_req,
 279                         unsigned int which, struct ceph_pagelist *pagelist)
 280 {
 281         struct ceph_osd_data *osd_data;
 282
 283         osd_data = osd_req_op_data(osd_req, which, cls, request_data);
 284         ceph_osd_data_pagelist_init(osd_data, pagelist);
 285         osd_req->r_ops[which].cls.indata_len += pagelist->length;
 286         osd_req->r_ops[which].indata_len += pagelist->length;
 287 }
 288 EXPORT_SYMBOL(osd_req_op_cls_request_data_pagelist);
 289
 290 void osd_req_op_cls_request_data_pages(struct ceph_osd_request *osd_req,
 291                         unsigned int which, struct page **pages, u64 length,
 292                         u32 alignment, bool pages_from_pool, bool own_pages)
 293 {
 294         struct ceph_osd_data *osd_data;
 295
 296         osd_data = osd_req_op_data(osd_req, which, cls, request_data);
 297         ceph_osd_data_pages_init(osd_data, pages, length, alignment,
 298                                 pages_from_pool, own_pages);
 299         osd_req->r_ops[which].cls.indata_len += length;
 300         osd_req->r_ops[which].indata_len += length;
 301 }
 302 EXPORT_SYMBOL(osd_req_op_cls_request_data_pages);
 303
 304 void osd_req_op_cls_request_data_bvecs(struct ceph_osd_request *osd_req,
 305                                        unsigned int which,
 306                                        struct bio_vec *bvecs, u32 num_bvecs,
 307                                        u32 bytes)
 308 {
 309         struct ceph_osd_data *osd_data;
 310         struct ceph_bvec_iter it = {
 311                 .bvecs = bvecs,
 312                 .iter = { .bi_size = bytes },
 313         };
 314
 315         osd_data = osd_req_op_data(osd_req, which, cls, request_data);
 316         ceph_osd_data_bvecs_init(osd_data, &it, num_bvecs);
 317         osd_req->r_ops[which].cls.indata_len += bytes;
 318         osd_req->r_ops[which].indata_len += bytes;
 319 }
 320 EXPORT_SYMBOL(osd_req_op_cls_request_data_bvecs);
 321
 322 void osd_req_op_cls_response_data_pages(struct ceph_osd_request *osd_req,
 323                         unsigned int which, struct page **pages, u64 length,
 324                         u32 alignment, bool pages_from_pool, bool own_pages)
 325 {
 326         struct ceph_osd_data *osd_data;
 327
 328         osd_data = osd_req_op_data(osd_req, which, cls, response_data);
 329         ceph_osd_data_pages_init(osd_data, pages, length, alignment,
 330                                 pages_from_pool, own_pages);
 331 }
 332 EXPORT_SYMBOL(osd_req_op_cls_response_data_pages);
 333
 334 static u64 ceph_osd_data_length(struct ceph_osd_data *osd_data)
 335 {
 336         switch (osd_data->type) {
 337         case CEPH_OSD_DATA_TYPE_NONE:
 338                 return 0;
 339         case CEPH_OSD_DATA_TYPE_PAGES:
 340                 return osd_data->length;
 341         case CEPH_OSD_DATA_TYPE_PAGELIST:
 342                 return (u64)osd_data->pagelist->length;
 343 #ifdef CONFIG_BLOCK
 344         case CEPH_OSD_DATA_TYPE_BIO:
 345                 return (u64)osd_data->bio_length;
 346 #endif /* CONFIG_BLOCK */
 347         case CEPH_OSD_DATA_TYPE_BVECS:
 348                 return osd_data->bvec_pos.iter.bi_size;
 349         default:
 350                 WARN(true, "unrecognized data type %d\n", (int)osd_data->type);
 351                 return 0;
 352         }
 353 }
 354
 355 static void ceph_osd_data_release(struct ceph_osd_data *osd_data)
 356 {
 357         if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES && osd_data->own_pages) {
 358                 int num_pages;
 359
 360                 num_pages = calc_pages_for((u64)osd_data->alignment,
 361                                                 (u64)osd_data->length);
 362                 ceph_release_page_vector(osd_data->pages, num_pages);
 363         } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
 364                 ceph_pagelist_release(osd_data->pagelist);
 365         }
 366         ceph_osd_data_init(osd_data);
 367 }
 368
 369 static void osd_req_op_data_release(struct ceph_osd_request *osd_req,
 370                         unsigned int which)
 371 {
 372         struct ceph_osd_req_op *op;
 373
 374         BUG_ON(which >= osd_req->r_num_ops);
 375         op = &osd_req->r_ops[which];
 376
 377         switch (op->op) {
 378         case CEPH_OSD_OP_READ:
 379         case CEPH_OSD_OP_WRITE:
 380         case CEPH_OSD_OP_WRITEFULL:
 381                 ceph_osd_data_release(&op->extent.osd_data);
 382                 break;
 383         case CEPH_OSD_OP_CALL:
 384                 ceph_osd_data_release(&op->cls.request_info);
 385                 ceph_osd_data_release(&op->cls.request_data);
 386                 ceph_osd_data_release(&op->cls.response_data);
 387                 break;
 388         case CEPH_OSD_OP_SETXATTR:
 389         case CEPH_OSD_OP_CMPXATTR:
 390                 ceph_osd_data_release(&op->xattr.osd_data);
 391                 break;
 392         case CEPH_OSD_OP_STAT:
 393                 ceph_osd_data_release(&op->raw_data_in);
 394                 break;
 395         case CEPH_OSD_OP_NOTIFY_ACK:
 396                 ceph_osd_data_release(&op->notify_ack.request_data);
 397                 break;
 398         case CEPH_OSD_OP_NOTIFY:
 399                 ceph_osd_data_release(&op->notify.request_data);
 400                 ceph_osd_data_release(&op->notify.response_data);
 401                 break;
 402         case CEPH_OSD_OP_LIST_WATCHERS:
 403                 ceph_osd_data_release(&op->list_watchers.response_data);
 404                 break;
 405         case CEPH_OSD_OP_COPY_FROM2:
 406                 ceph_osd_data_release(&op->copy_from.osd_data);
 407                 break;
 408         default:
 409                 break;
 410         }
 411 }
 412
 413 /*
 414  * Assumes @t is zero-initialized.
 415  */
 416 static void target_init(struct ceph_osd_request_target *t)
 417 {
 418         ceph_oid_init(&t->base_oid);
 419         ceph_oloc_init(&t->base_oloc);
 420         ceph_oid_init(&t->target_oid);
 421         ceph_oloc_init(&t->target_oloc);
 422
 423         ceph_osds_init(&t->acting);
 424         ceph_osds_init(&t->up);
 425         t->size = -1;
 426         t->min_size = -1;
 427
 428         t->osd = CEPH_HOMELESS_OSD;
 429 }
 430
 431 static void target_copy(struct ceph_osd_request_target *dest,
 432                         const struct ceph_osd_request_target *src)
 433 {
 434         ceph_oid_copy(&dest->base_oid, &src->base_oid);
 435         ceph_oloc_copy(&dest->base_oloc, &src->base_oloc);
 436         ceph_oid_copy(&dest->target_oid, &src->target_oid);
 437         ceph_oloc_copy(&dest->target_oloc, &src->target_oloc);
 438
 439         dest->pgid = src->pgid; /* struct */
 440         dest->spgid = src->spgid; /* struct */
 441         dest->pg_num = src->pg_num;
 442         dest->pg_num_mask = src->pg_num_mask;
 443         ceph_osds_copy(&dest->acting, &src->acting);
 444         ceph_osds_copy(&dest->up, &src->up);
 445         dest->size = src->size;
 446         dest->min_size = src->min_size;
 447         dest->sort_bitwise = src->sort_bitwise;
 448         dest->recovery_deletes = src->recovery_deletes;
 449
 450         dest->flags = src->flags;
 451         dest->used_replica = src->used_replica;
 452         dest->paused = src->paused;
 453
 454         dest->epoch = src->epoch;
 455         dest->last_force_resend = src->last_force_resend;
 456
 457         dest->osd = src->osd;
 458 }
 459
 460 static void target_destroy(struct ceph_osd_request_target *t)
 461 {
 462         ceph_oid_destroy(&t->base_oid);
 463         ceph_oloc_destroy(&t->base_oloc);
 464         ceph_oid_destroy(&t->target_oid);
 465         ceph_oloc_destroy(&t->target_oloc);
 466 }
 467
 468 /*
 469  * requests
 470  */
 471 static void request_release_checks(struct ceph_osd_request *req)
 472 {
 473         WARN_ON(!RB_EMPTY_NODE(&req->r_node));
 474         WARN_ON(!RB_EMPTY_NODE(&req->r_mc_node));
 475         WARN_ON(!list_empty(&req->r_private_item));
 476         WARN_ON(req->r_osd);
 477 }
 478
 479 static void ceph_osdc_release_request(struct kref *kref)
 480 {
 481         struct ceph_osd_request *req = container_of(kref,
 482                                             struct ceph_osd_request, r_kref);
 483         unsigned int which;
 484
 485         dout("%s %p (r_request %p r_reply %p)\n", __func__, req,
 486              req->r_request, req->r_reply);
 487         request_release_checks(req);
 488
 489         if (req->r_request)
 490                 ceph_msg_put(req->r_request);
 491         if (req->r_reply)
 492                 ceph_msg_put(req->r_reply);
 493
 494         for (which = 0; which < req->r_num_ops; which++)
 495                 osd_req_op_data_release(req, which);
 496
 497         target_destroy(&req->r_t);
 498         ceph_put_snap_context(req->r_snapc);
 499
 500         if (req->r_mempool)
 501                 mempool_free(req, req->r_osdc->req_mempool);
 502         else if (req->r_num_ops <= CEPH_OSD_SLAB_OPS)
 503                 kmem_cache_free(ceph_osd_request_cache, req);
 504         else
 505                 kfree(req);
 506 }
 507
 508 void ceph_osdc_get_request(struct ceph_osd_request *req)
 509 {
 510         dout("%s %p (was %d)\n", __func__, req,
 511              kref_read(&req->r_kref));
 512         kref_get(&req->r_kref);
 513 }
 514 EXPORT_SYMBOL(ceph_osdc_get_request);
 515
 516 void ceph_osdc_put_request(struct ceph_osd_request *req)
 517 {
 518         if (req) {
 519                 dout("%s %p (was %d)\n", __func__, req,
 520                      kref_read(&req->r_kref));
 521                 kref_put(&req->r_kref, ceph_osdc_release_request);
 522         }
 523 }
 524 EXPORT_SYMBOL(ceph_osdc_put_request);
 525
 526 static void request_init(struct ceph_osd_request *req)
 527 {
 528         /* req only, each op is zeroed in osd_req_op_init() */
 529         memset(req, 0, sizeof(*req));
 530
 531         kref_init(&req->r_kref);
 532         init_completion(&req->r_completion);
 533         RB_CLEAR_NODE(&req->r_node);
 534         RB_CLEAR_NODE(&req->r_mc_node);
 535         INIT_LIST_HEAD(&req->r_private_item);
 536
 537         target_init(&req->r_t);
 538 }
 539
 540 /*
 541  * This is ugly, but it allows us to reuse linger registration and ping
 542  * requests, keeping the structure of the code around send_linger{_ping}()
 543  * reasonable.  Setting up a min_nr=2 mempool for each linger request
 544  * and dealing with copying ops (this blasts req only, watch op remains
 545  * intact) isn't any better.
 546  */
 547 static void request_reinit(struct ceph_osd_request *req)
 548 {
 549         struct ceph_osd_client *osdc = req->r_osdc;
 550         bool mempool = req->r_mempool;
 551         unsigned int num_ops = req->r_num_ops;
 552         u64 snapid = req->r_snapid;
 553         struct ceph_snap_context *snapc = req->r_snapc;
 554         bool linger = req->r_linger;
 555         struct ceph_msg *request_msg = req->r_request;
 556         struct ceph_msg *reply_msg = req->r_reply;
 557
 558         dout("%s req %p\n", __func__, req);
 559         WARN_ON(kref_read(&req->r_kref) != 1);
 560         request_release_checks(req);
 561
 562         WARN_ON(kref_read(&request_msg->kref) != 1);
 563         WARN_ON(kref_read(&reply_msg->kref) != 1);
 564         target_destroy(&req->r_t);
 565
 566         request_init(req);
 567         req->r_osdc = osdc;
 568         req->r_mempool = mempool;
 569         req->r_num_ops = num_ops;
 570         req->r_snapid = snapid;
 571         req->r_snapc = snapc;
 572         req->r_linger = linger;
 573         req->r_request = request_msg;
 574         req->r_reply = reply_msg;
 575 }
 576
 577 struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
 578                                                struct ceph_snap_context *snapc,
 579                                                unsigned int num_ops,
 580                                                bool use_mempool,
 581                                                gfp_t gfp_flags)
 582 {
 583         struct ceph_osd_request *req;
 584
 585         if (use_mempool) {
 586                 BUG_ON(num_ops > CEPH_OSD_SLAB_OPS);
 587                 req = mempool_alloc(osdc->req_mempool, gfp_flags);
 588         } else if (num_ops <= CEPH_OSD_SLAB_OPS) {
 589                 req = kmem_cache_alloc(ceph_osd_request_cache, gfp_flags);
 590         } else {
 591                 BUG_ON(num_ops > CEPH_OSD_MAX_OPS);
 592                 req = kmalloc(struct_size(req, r_ops, num_ops), gfp_flags);
 593         }
 594         if (unlikely(!req))
 595                 return NULL;
 596
 597         request_init(req);
 598         req->r_osdc = osdc;
 599         req->r_mempool = use_mempool;
 600         req->r_num_ops = num_ops;
 601         req->r_snapid = CEPH_NOSNAP;
 602         req->r_snapc = ceph_get_snap_context(snapc);
 603
 604         dout("%s req %p\n", __func__, req);
 605         return req;
 606 }
 607 EXPORT_SYMBOL(ceph_osdc_alloc_request);
 608
 609 static int ceph_oloc_encoding_size(const struct ceph_object_locator *oloc)
 610 {
 611         return 8 + 4 + 4 + 4 + (oloc->pool_ns ? oloc->pool_ns->len : 0);
 612 }
 613
 614 static int __ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp,
 615                                       int num_request_data_items,
 616                                       int num_reply_data_items)
 617 {
 618         struct ceph_osd_client *osdc = req->r_osdc;
 619         struct ceph_msg *msg;
 620         int msg_size;
 621
 622         WARN_ON(req->r_request || req->r_reply);
 623         WARN_ON(ceph_oid_empty(&req->r_base_oid));
 624         WARN_ON(ceph_oloc_empty(&req->r_base_oloc));
 625
 626         /* create request message */
 627         msg_size = CEPH_ENCODING_START_BLK_LEN +
 628                         CEPH_PGID_ENCODING_LEN + 1; /* spgid */
 629         msg_size += 4 + 4 + 4; /* hash, osdmap_epoch, flags */
 630         msg_size += CEPH_ENCODING_START_BLK_LEN +
 631                         sizeof(struct ceph_osd_reqid); /* reqid */
 632         msg_size += sizeof(struct ceph_blkin_trace_info); /* trace */
 633         msg_size += 4 + sizeof(struct ceph_timespec); /* client_inc, mtime */
 634         msg_size += CEPH_ENCODING_START_BLK_LEN +
 635                         ceph_oloc_encoding_size(&req->r_base_oloc); /* oloc */
 636         msg_size += 4 + req->r_base_oid.name_len; /* oid */
 637         msg_size += 2 + req->r_num_ops * sizeof(struct ceph_osd_op);
 638         msg_size += 8; /* snapid */
 639         msg_size += 8; /* snap_seq */
 640         msg_size += 4 + 8 * (req->r_snapc ? req->r_snapc->num_snaps : 0);
 641         msg_size += 4 + 8; /* retry_attempt, features */
 642
 643         if (req->r_mempool)
 644                 msg = ceph_msgpool_get(&osdc->msgpool_op, msg_size,
 645                                        num_request_data_items);
 646         else
 647                 msg = ceph_msg_new2(CEPH_MSG_OSD_OP, msg_size,
 648                                     num_request_data_items, gfp, true);
 649         if (!msg)
 650                 return -ENOMEM;
 651
 652         memset(msg->front.iov_base, 0, msg->front.iov_len);
 653         req->r_request = msg;
 654
 655         /* create reply message */
 656         msg_size = OSD_OPREPLY_FRONT_LEN;
 657         msg_size += req->r_base_oid.name_len;
 658         msg_size += req->r_num_ops * sizeof(struct ceph_osd_op);
 659
 660         if (req->r_mempool)
 661                 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, msg_size,
 662                                        num_reply_data_items);
 663         else
 664                 msg = ceph_msg_new2(CEPH_MSG_OSD_OPREPLY, msg_size,
 665                                     num_reply_data_items, gfp, true);
 666         if (!msg)
 667                 return -ENOMEM;
 668
 669         req->r_reply = msg;
 670
 671         return 0;
 672 }
 673
 674 static bool osd_req_opcode_valid(u16 opcode)
 675 {
 676         switch (opcode) {
 677 #define GENERATE_CASE(op, opcode, str)  case CEPH_OSD_OP_##op: return true;
 678 __CEPH_FORALL_OSD_OPS(GENERATE_CASE)
 679 #undef GENERATE_CASE
 680         default:
 681                 return false;
 682         }
 683 }
 684
 685 static void get_num_data_items(struct ceph_osd_request *req,
 686                                int *num_request_data_items,
 687                                int *num_reply_data_items)
 688 {
 689         struct ceph_osd_req_op *op;
 690
 691         *num_request_data_items = 0;
 692         *num_reply_data_items = 0;
 693
 694         for (op = req->r_ops; op != &req->r_ops[req->r_num_ops]; op++) {
 695                 switch (op->op) {
 696                 /* request */
 697                 case CEPH_OSD_OP_WRITE:
 698                 case CEPH_OSD_OP_WRITEFULL:
 699                 case CEPH_OSD_OP_SETXATTR:
 700                 case CEPH_OSD_OP_CMPXATTR:
 701                 case CEPH_OSD_OP_NOTIFY_ACK:
 702                 case CEPH_OSD_OP_COPY_FROM2:
 703                         *num_request_data_items += 1;
 704                         break;
 705
 706                 /* reply */
 707                 case CEPH_OSD_OP_STAT:
 708                 case CEPH_OSD_OP_READ:
 709                 case CEPH_OSD_OP_LIST_WATCHERS:
 710                         *num_reply_data_items += 1;
 711                         break;
 712
 713                 /* both */
 714                 case CEPH_OSD_OP_NOTIFY:
 715                         *num_request_data_items += 1;
 716                         *num_reply_data_items += 1;
 717                         break;
 718                 case CEPH_OSD_OP_CALL:
 719                         *num_request_data_items += 2;
 720                         *num_reply_data_items += 1;
 721                         break;
 722
 723                 default:
 724                         WARN_ON(!osd_req_opcode_valid(op->op));
 725                         break;
 726                 }
 727         }
 728 }
 729
 730 /*
 731  * oid, oloc and OSD op opcode(s) must be filled in before this function
 732  * is called.
 733  */
 734 int ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp)
 735 {
 736         int num_request_data_items, num_reply_data_items;
 737
 738         get_num_data_items(req, &num_request_data_items, &num_reply_data_items);
 739         return __ceph_osdc_alloc_messages(req, gfp, num_request_data_items,
 740                                           num_reply_data_items);
 741 }
 742 EXPORT_SYMBOL(ceph_osdc_alloc_messages);
 743
 744 /*
 745  * This is an osd op init function for opcodes that have no data or
 746  * other information associated with them.  It also serves as a
 747  * common init routine for all the other init functions, below.
 748  */
 749 struct ceph_osd_req_op *
 750 osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which,
 751                  u16 opcode, u32 flags)
 752 {
 753         struct ceph_osd_req_op *op;
 754
 755         BUG_ON(which >= osd_req->r_num_ops);
 756         BUG_ON(!osd_req_opcode_valid(opcode));
 757
 758         op = &osd_req->r_ops[which];
 759         memset(op, 0, sizeof (*op));
 760         op->op = opcode;
 761         op->flags = flags;
 762
 763         return op;
 764 }
 765 EXPORT_SYMBOL(osd_req_op_init);
 766
 767 void osd_req_op_extent_init(struct ceph_osd_request *osd_req,
 768                                 unsigned int which, u16 opcode,
 769                                 u64 offset, u64 length,
 770                                 u64 truncate_size, u32 truncate_seq)
 771 {
 772         struct ceph_osd_req_op *op = osd_req_op_init(osd_req, which,
 773                                                      opcode, 0);
 774         size_t payload_len = 0;
 775
 776         BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
 777                opcode != CEPH_OSD_OP_WRITEFULL && opcode != CEPH_OSD_OP_ZERO &&
 778                opcode != CEPH_OSD_OP_TRUNCATE);
 779
 780         op->extent.offset = offset;
 781         op->extent.length = length;
 782         op->extent.truncate_size = truncate_size;
 783         op->extent.truncate_seq = truncate_seq;
 784         if (opcode == CEPH_OSD_OP_WRITE || opcode == CEPH_OSD_OP_WRITEFULL)
 785                 payload_len += length;
 786
 787         op->indata_len = payload_len;
 788 }
 789 EXPORT_SYMBOL(osd_req_op_extent_init);
 790
 791 void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
 792                                 unsigned int which, u64 length)
 793 {
 794         struct ceph_osd_req_op *op;
 795         u64 previous;
 796
 797         BUG_ON(which >= osd_req->r_num_ops);
 798         op = &osd_req->r_ops[which];
 799         previous = op->extent.length;
 800
 801         if (length == previous)
 802                 return;         /* Nothing to do */
 803         BUG_ON(length > previous);
 804
 805         op->extent.length = length;
 806         if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL)
 807                 op->indata_len -= previous - length;
 808 }
 809 EXPORT_SYMBOL(osd_req_op_extent_update);
 810
 811 void osd_req_op_extent_dup_last(struct ceph_osd_request *osd_req,
 812                                 unsigned int which, u64 offset_inc)
 813 {
 814         struct ceph_osd_req_op *op, *prev_op;
 815
 816         BUG_ON(which + 1 >= osd_req->r_num_ops);
 817
 818         prev_op = &osd_req->r_ops[which];
 819         op = osd_req_op_init(osd_req, which + 1, prev_op->op, prev_op->flags);
 820         /* dup previous one */
 821         op->indata_len = prev_op->indata_len;
 822         op->outdata_len = prev_op->outdata_len;
 823         op->extent = prev_op->extent;
 824         /* adjust offset */
 825         op->extent.offset += offset_inc;
 826         op->extent.length -= offset_inc;
 827
 828         if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL)
 829                 op->indata_len -= offset_inc;
 830 }
 831 EXPORT_SYMBOL(osd_req_op_extent_dup_last);
 832
 833 int osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
 834                         const char *class, const char *method)
 835 {
 836         struct ceph_osd_req_op *op;
 837         struct ceph_pagelist *pagelist;
 838         size_t payload_len = 0;
 839         size_t size;
 840         int ret;
 841
 842         op = osd_req_op_init(osd_req, which, CEPH_OSD_OP_CALL, 0);
 843
 844         pagelist = ceph_pagelist_alloc(GFP_NOFS);
 845         if (!pagelist)
 846                 return -ENOMEM;
 847
 848         op->cls.class_name = class;
 849         size = strlen(class);
 850         BUG_ON(size > (size_t) U8_MAX);
 851         op->cls.class_len = size;
 852         ret = ceph_pagelist_append(pagelist, class, size);
 853         if (ret)
 854                 goto err_pagelist_free;
 855         payload_len += size;
 856
 857         op->cls.method_name = method;
 858         size = strlen(method);
 859         BUG_ON(size > (size_t) U8_MAX);
 860         op->cls.method_len = size;
 861         ret = ceph_pagelist_append(pagelist, method, size);
 862         if (ret)
 863                 goto err_pagelist_free;
 864         payload_len += size;
 865
 866         osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist);
 867         op->indata_len = payload_len;
 868         return 0;
 869
 870 err_pagelist_free:
 871         ceph_pagelist_release(pagelist);
 872         return ret;
 873 }
 874 EXPORT_SYMBOL(osd_req_op_cls_init);
 875
 876 int osd_req_op_xattr_init(struct ceph_osd_request *osd_req, unsigned int which,
 877                           u16 opcode, const char *name, const void *value,
 878                           size_t size, u8 cmp_op, u8 cmp_mode)
 879 {
 880         struct ceph_osd_req_op *op = osd_req_op_init(osd_req, which,
 881                                                      opcode, 0);
 882         struct ceph_pagelist *pagelist;
 883         size_t payload_len;
 884         int ret;
 885
 886         BUG_ON(opcode != CEPH_OSD_OP_SETXATTR && opcode != CEPH_OSD_OP_CMPXATTR);
 887
 888         pagelist = ceph_pagelist_alloc(GFP_NOFS);
 889         if (!pagelist)
 890                 return -ENOMEM;
 891
 892         payload_len = strlen(name);
 893         op->xattr.name_len = payload_len;
 894         ret = ceph_pagelist_append(pagelist, name, payload_len);
 895         if (ret)
 896                 goto err_pagelist_free;
 897
 898         op->xattr.value_len = size;
 899         ret = ceph_pagelist_append(pagelist, value, size);
 900         if (ret)
 901                 goto err_pagelist_free;
 902         payload_len += size;
 903
 904         op->xattr.cmp_op = cmp_op;
 905         op->xattr.cmp_mode = cmp_mode;
 906
 907         ceph_osd_data_pagelist_init(&op->xattr.osd_data, pagelist);
 908         op->indata_len = payload_len;
 909         return 0;
 910
 911 err_pagelist_free:
 912         ceph_pagelist_release(pagelist);
 913         return ret;
 914 }
 915 EXPORT_SYMBOL(osd_req_op_xattr_init);
 916
 917 /*
 918  * @watch_opcode: CEPH_OSD_WATCH_OP_*
 919  */
 920 static void osd_req_op_watch_init(struct ceph_osd_request *req, int which,
 921                                   u64 cookie, u8 watch_opcode)
 922 {
 923         struct ceph_osd_req_op *op;
 924
 925         op = osd_req_op_init(req, which, CEPH_OSD_OP_WATCH, 0);
 926         op->watch.cookie = cookie;
 927         op->watch.op = watch_opcode;
 928         op->watch.gen = 0;
 929 }
 930
 931 /*
 932  * @flags: CEPH_OSD_OP_ALLOC_HINT_FLAG_*
 933  */
 934 void osd_req_op_alloc_hint_init(struct ceph_osd_request *osd_req,
 935                                 unsigned int which,
 936                                 u64 expected_object_size,
 937                                 u64 expected_write_size,
 938                                 u32 flags)
 939 {
 940         struct ceph_osd_req_op *op;
 941
 942         op = osd_req_op_init(osd_req, which, CEPH_OSD_OP_SETALLOCHINT, 0);
 943         op->alloc_hint.expected_object_size = expected_object_size;
 944         op->alloc_hint.expected_write_size = expected_write_size;
 945         op->alloc_hint.flags = flags;
 946
 947         /*
 948          * CEPH_OSD_OP_SETALLOCHINT op is advisory and therefore deemed
 949          * not worth a feature bit.  Set FAILOK per-op flag to make
 950          * sure older osds don't trip over an unsupported opcode.
 951          */
 952         op->flags |= CEPH_OSD_OP_FLAG_FAILOK;
 953 }
 954 EXPORT_SYMBOL(osd_req_op_alloc_hint_init);
 955
 956 static void ceph_osdc_msg_data_add(struct ceph_msg *msg,
 957                                 struct ceph_osd_data *osd_data)
 958 {
 959         u64 length = ceph_osd_data_length(osd_data);
 960
 961         if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
 962                 BUG_ON(length > (u64) SIZE_MAX);
 963                 if (length)
 964                         ceph_msg_data_add_pages(msg, osd_data->pages,
 965                                         length, osd_data->alignment, false);
 966         } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
 967                 BUG_ON(!length);
 968                 ceph_msg_data_add_pagelist(msg, osd_data->pagelist);
 969 #ifdef CONFIG_BLOCK
 970         } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
 971                 ceph_msg_data_add_bio(msg, &osd_data->bio_pos, length);
 972 #endif
 973         } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BVECS) {
 974                 ceph_msg_data_add_bvecs(msg, &osd_data->bvec_pos);
 975         } else {
 976                 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
 977         }
 978 }
 979
 980 static u32 osd_req_encode_op(struct ceph_osd_op *dst,
 981                              const struct ceph_osd_req_op *src)
 982 {
 983         switch (src->op) {
 984         case CEPH_OSD_OP_STAT:
 985                 break;
 986         case CEPH_OSD_OP_READ:
 987         case CEPH_OSD_OP_WRITE:
 988         case CEPH_OSD_OP_WRITEFULL:
 989         case CEPH_OSD_OP_ZERO:
 990         case CEPH_OSD_OP_TRUNCATE:
 991                 dst->extent.offset = cpu_to_le64(src->extent.offset);
 992                 dst->extent.length = cpu_to_le64(src->extent.length);
 993                 dst->extent.truncate_size =
 994                         cpu_to_le64(src->extent.truncate_size);
 995                 dst->extent.truncate_seq =
 996                         cpu_to_le32(src->extent.truncate_seq);
 997                 break;
 998         case CEPH_OSD_OP_CALL:
 999                 dst->cls.class_len = src->cls.class_len;
1000                 dst->cls.method_len = src->cls.method_len;
1001                 dst->cls.indata_len = cpu_to_le32(src->cls.indata_len);
1002                 break;
1003         case CEPH_OSD_OP_WATCH:
1004                 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
1005                 dst->watch.ver = cpu_to_le64(0);
1006                 dst->watch.op = src->watch.op;
1007                 dst->watch.gen = cpu_to_le32(src->watch.gen);
1008                 break;
1009         case CEPH_OSD_OP_NOTIFY_ACK:
1010                 break;
1011         case CEPH_OSD_OP_NOTIFY:
1012                 dst->notify.cookie = cpu_to_le64(src->notify.cookie);
1013                 break;
1014         case CEPH_OSD_OP_LIST_WATCHERS:
1015                 break;
1016         case CEPH_OSD_OP_SETALLOCHINT:
1017                 dst->alloc_hint.expected_object_size =
1018                     cpu_to_le64(src->alloc_hint.expected_object_size);
1019                 dst->alloc_hint.expected_write_size =
1020                     cpu_to_le64(src->alloc_hint.expected_write_size);
1021                 dst->alloc_hint.flags = cpu_to_le32(src->alloc_hint.flags);
1022                 break;
1023         case CEPH_OSD_OP_SETXATTR:
1024         case CEPH_OSD_OP_CMPXATTR:
1025                 dst->xattr.name_len = cpu_to_le32(src->xattr.name_len);
1026                 dst->xattr.value_len = cpu_to_le32(src->xattr.value_len);
1027                 dst->xattr.cmp_op = src->xattr.cmp_op;
1028                 dst->xattr.cmp_mode = src->xattr.cmp_mode;
1029                 break;
1030         case CEPH_OSD_OP_CREATE:
1031         case CEPH_OSD_OP_DELETE:
1032                 break;
1033         case CEPH_OSD_OP_COPY_FROM2:
1034                 dst->copy_from.snapid = cpu_to_le64(src->copy_from.snapid);
1035                 dst->copy_from.src_version =
1036                         cpu_to_le64(src->copy_from.src_version);
1037                 dst->copy_from.flags = src->copy_from.flags;
1038                 dst->copy_from.src_fadvise_flags =
1039                         cpu_to_le32(src->copy_from.src_fadvise_flags);
1040                 break;
1041         default:
1042                 pr_err("unsupported osd opcode %s\n",
1043                         ceph_osd_op_name(src->op));
1044                 WARN_ON(1);
1045
1046                 return 0;
1047         }
1048
1049         dst->op = cpu_to_le16(src->op);
1050         dst->flags = cpu_to_le32(src->flags);
1051         dst->payload_len = cpu_to_le32(src->indata_len);
1052
1053         return src->indata_len;
1054 }
1055
1056 /*
1057  * build new request AND message, calculate layout, and adjust file
1058  * extent as needed.
1059  *
1060  * if the file was recently truncated, we include information about its
1061  * old and new size so that the object can be updated appropriately.  (we
1062  * avoid synchronously deleting truncated objects because it's slow.)
1063  */
1064 struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
1065                                                struct ceph_file_layout *layout,
1066                                                struct ceph_vino vino,
1067                                                u64 off, u64 *plen,
1068                                                unsigned int which, int num_ops,
1069                                                int opcode, int flags,
1070                                                struct ceph_snap_context *snapc,
1071                                                u32 truncate_seq,
1072                                                u64 truncate_size,
1073                                                bool use_mempool)
1074 {
1075         struct ceph_osd_request *req;
1076         u64 objnum = 0;
1077         u64 objoff = 0;
1078         u64 objlen = 0;
1079         int r;
1080
1081         BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
1082                opcode != CEPH_OSD_OP_ZERO && opcode != CEPH_OSD_OP_TRUNCATE &&
1083                opcode != CEPH_OSD_OP_CREATE && opcode != CEPH_OSD_OP_DELETE);
1084
1085         req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
1086                                         GFP_NOFS);
1087         if (!req) {
1088                 r = -ENOMEM;
1089                 goto fail;
1090         }
1091
1092         /* calculate max write size */
1093         r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
1094         if (r)
1095                 goto fail;
1096
1097         if (opcode == CEPH_OSD_OP_CREATE || opcode == CEPH_OSD_OP_DELETE) {
1098                 osd_req_op_init(req, which, opcode, 0);
1099         } else {
1100                 u32 object_size = layout->object_size;
1101                 u32 object_base = off - objoff;
1102                 if (!(truncate_seq == 1 && truncate_size == -1ULL)) {
1103                         if (truncate_size <= object_base) {
1104                                 truncate_size = 0;
1105                         } else {
1106                                 truncate_size -= object_base;
1107                                 if (truncate_size > object_size)
1108                                         truncate_size = object_size;
1109                         }
1110                 }
1111                 osd_req_op_extent_init(req, which, opcode, objoff, objlen,
1112                                        truncate_size, truncate_seq);
1113         }
1114
1115         req->r_base_oloc.pool = layout->pool_id;
1116         req->r_base_oloc.pool_ns = ceph_try_get_string(layout->pool_ns);
1117         ceph_oid_printf(&req->r_base_oid, "%llx.%08llx", vino.ino, objnum);
1118         req->r_flags = flags | osdc->client->options->read_from_replica;
1119
1120         req->r_snapid = vino.snap;
1121         if (flags & CEPH_OSD_FLAG_WRITE)
1122                 req->r_data_offset = off;
1123
1124         if (num_ops > 1)
1125                 /*
1126                  * This is a special case for ceph_writepages_start(), but it
1127                  * also covers ceph_uninline_data().  If more multi-op request
1128                  * use cases emerge, we will need a separate helper.
1129                  */
1130                 r = __ceph_osdc_alloc_messages(req, GFP_NOFS, num_ops, 0);
1131         else
1132                 r = ceph_osdc_alloc_messages(req, GFP_NOFS);
1133         if (r)
1134                 goto fail;
1135
1136         return req;
1137
1138 fail:
1139         ceph_osdc_put_request(req);
1140         return ERR_PTR(r);
1141 }
1142 EXPORT_SYMBOL(ceph_osdc_new_request);
1143
1144 /*
1145  * We keep osd requests in an rbtree, sorted by ->r_tid.
1146  */
1147 DEFINE_RB_FUNCS(request, struct ceph_osd_request, r_tid, r_node)
1148 DEFINE_RB_FUNCS(request_mc, struct ceph_osd_request, r_tid, r_mc_node)
1149
1150 /*
1151  * Call @fn on each OSD request as long as @fn returns 0.
1152  */
1153 static void for_each_request(struct ceph_osd_client *osdc,
1154                         int (*fn)(struct ceph_osd_request *req, void *arg),
1155                         void *arg)
1156 {
1157         struct rb_node *n, *p;
1158
1159         for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
1160                 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
1161
1162                 for (p = rb_first(&osd->o_requests); p; ) {
1163                         struct ceph_osd_request *req =
1164                             rb_entry(p, struct ceph_osd_request, r_node);
1165
1166                         p = rb_next(p);
1167                         if (fn(req, arg))
1168                                 return;
1169                 }
1170         }
1171
1172         for (p = rb_first(&osdc->homeless_osd.o_requests); p; ) {
1173                 struct ceph_osd_request *req =
1174                     rb_entry(p, struct ceph_osd_request, r_node);
1175
1176                 p = rb_next(p);
1177                 if (fn(req, arg))
1178                         return;
1179         }
1180 }
1181
1182 static bool osd_homeless(struct ceph_osd *osd)
1183 {
1184         return osd->o_osd == CEPH_HOMELESS_OSD;
1185 }
1186
1187 static bool osd_registered(struct ceph_osd *osd)
1188 {
1189         verify_osdc_locked(osd->o_osdc);
1190
1191         return !RB_EMPTY_NODE(&osd->o_node);
1192 }
1193
1194 /*
1195  * Assumes @osd is zero-initialized.
1196  */
1197 static void osd_init(struct ceph_osd *osd)
1198 {
1199         refcount_set(&osd->o_ref, 1);
1200         RB_CLEAR_NODE(&osd->o_node);
1201         osd->o_requests = RB_ROOT;
1202         osd->o_linger_requests = RB_ROOT;
1203         osd->o_backoff_mappings = RB_ROOT;
1204         osd->o_backoffs_by_id = RB_ROOT;
1205         INIT_LIST_HEAD(&osd->o_osd_lru);
1206         INIT_LIST_HEAD(&osd->o_keepalive_item);
1207         osd->o_incarnation = 1;
1208         mutex_init(&osd->lock);
1209 }
1210
1211 static void osd_cleanup(struct ceph_osd *osd)
1212 {
1213         WARN_ON(!RB_EMPTY_NODE(&osd->o_node));
1214         WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests));
1215         WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests));
1216         WARN_ON(!RB_EMPTY_ROOT(&osd->o_backoff_mappings));
1217         WARN_ON(!RB_EMPTY_ROOT(&osd->o_backoffs_by_id));
1218         WARN_ON(!list_empty(&osd->o_osd_lru));
1219         WARN_ON(!list_empty(&osd->o_keepalive_item));
1220
1221         if (osd->o_auth.authorizer) {
1222                 WARN_ON(osd_homeless(osd));
1223                 ceph_auth_destroy_authorizer(osd->o_auth.authorizer);
1224         }
1225 }
1226
1227 /*
1228  * Track open sessions with osds.
1229  */
1230 static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
1231 {
1232         struct ceph_osd *osd;
1233
1234         WARN_ON(onum == CEPH_HOMELESS_OSD);
1235
1236         osd = kzalloc(sizeof(*osd), GFP_NOIO | __GFP_NOFAIL);
1237         osd_init(osd);
1238         osd->o_osdc = osdc;
1239         osd->o_osd = onum;
1240
1241         ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
1242
1243         return osd;
1244 }
1245
1246 static struct ceph_osd *get_osd(struct ceph_osd *osd)
1247 {
1248         if (refcount_inc_not_zero(&osd->o_ref)) {
1249                 dout("get_osd %p %d -> %d\n", osd, refcount_read(&osd->o_ref)-1,
1250                      refcount_read(&osd->o_ref));
1251                 return osd;
1252         } else {
1253                 dout("get_osd %p FAIL\n", osd);
1254                 return NULL;
1255         }
1256 }
1257
1258 static void put_osd(struct ceph_osd *osd)
1259 {
1260         dout("put_osd %p %d -> %d\n", osd, refcount_read(&osd->o_ref),
1261              refcount_read(&osd->o_ref) - 1);
1262         if (refcount_dec_and_test(&osd->o_ref)) {
1263                 osd_cleanup(osd);
1264                 kfree(osd);
1265         }
1266 }
1267
1268 DEFINE_RB_FUNCS(osd, struct ceph_osd, o_osd, o_node)
1269
1270 static void __move_osd_to_lru(struct ceph_osd *osd)
1271 {
1272         struct ceph_osd_client *osdc = osd->o_osdc;
1273
1274         dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1275         BUG_ON(!list_empty(&osd->o_osd_lru));
1276
1277         spin_lock(&osdc->osd_lru_lock);
1278         list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
1279         spin_unlock(&osdc->osd_lru_lock);
1280
1281         osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl;
1282 }
1283
1284 static void maybe_move_osd_to_lru(struct ceph_osd *osd)
1285 {
1286         if (RB_EMPTY_ROOT(&osd->o_requests) &&
1287             RB_EMPTY_ROOT(&osd->o_linger_requests))
1288                 __move_osd_to_lru(osd);
1289 }
1290
1291 static void __remove_osd_from_lru(struct ceph_osd *osd)
1292 {
1293         struct ceph_osd_client *osdc = osd->o_osdc;
1294
1295         dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1296
1297         spin_lock(&osdc->osd_lru_lock);
1298         if (!list_empty(&osd->o_osd_lru))
1299                 list_del_init(&osd->o_osd_lru);
1300         spin_unlock(&osdc->osd_lru_lock);
1301 }
1302
1303 /*
1304  * Close the connection and assign any leftover requests to the
1305  * homeless session.
1306  */
1307 static void close_osd(struct ceph_osd *osd)
1308 {
1309         struct ceph_osd_client *osdc = osd->o_osdc;
1310         struct rb_node *n;
1311
1312         verify_osdc_wrlocked(osdc);
1313         dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1314
1315         ceph_con_close(&osd->o_con);
1316
1317         for (n = rb_first(&osd->o_requests); n; ) {
1318                 struct ceph_osd_request *req =
1319                     rb_entry(n, struct ceph_osd_request, r_node);
1320
1321                 n = rb_next(n); /* unlink_request() */
1322
1323                 dout(" reassigning req %p tid %llu\n", req, req->r_tid);
1324                 unlink_request(osd, req);
1325                 link_request(&osdc->homeless_osd, req);
1326         }
1327         for (n = rb_first(&osd->o_linger_requests); n; ) {
1328                 struct ceph_osd_linger_request *lreq =
1329                     rb_entry(n, struct ceph_osd_linger_request, node);
1330
1331                 n = rb_next(n); /* unlink_linger() */
1332
1333                 dout(" reassigning lreq %p linger_id %llu\n", lreq,
1334                      lreq->linger_id);
1335                 unlink_linger(osd, lreq);
1336                 link_linger(&osdc->homeless_osd, lreq);
1337         }
1338         clear_backoffs(osd);
1339
1340         __remove_osd_from_lru(osd);
1341         erase_osd(&osdc->osds, osd);
1342         put_osd(osd);
1343 }
1344
1345 /*
1346  * reset osd connect
1347  */
1348 static int reopen_osd(struct ceph_osd *osd)
1349 {
1350         struct ceph_entity_addr *peer_addr;
1351
1352         dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1353
1354         if (RB_EMPTY_ROOT(&osd->o_requests) &&
1355             RB_EMPTY_ROOT(&osd->o_linger_requests)) {
1356                 close_osd(osd);
1357                 return -ENODEV;
1358         }
1359
1360         peer_addr = &osd->o_osdc->osdmap->osd_addr[osd->o_osd];
1361         if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1362                         !ceph_con_opened(&osd->o_con)) {
1363                 struct rb_node *n;
1364
1365                 dout("osd addr hasn't changed and connection never opened, "
1366                      "letting msgr retry\n");
1367                 /* touch each r_stamp for handle_timeout()'s benfit */
1368                 for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) {
1369                         struct ceph_osd_request *req =
1370                             rb_entry(n, struct ceph_osd_request, r_node);
1371                         req->r_stamp = jiffies;
1372                 }
1373
1374                 return -EAGAIN;
1375         }
1376
1377         ceph_con_close(&osd->o_con);
1378         ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1379         osd->o_incarnation++;
1380
1381         return 0;
1382 }
1383
1384 static struct ceph_osd *lookup_create_osd(struct ceph_osd_client *osdc, int o,
1385                                           bool wrlocked)
1386 {
1387         struct ceph_osd *osd;
1388
1389         if (wrlocked)
1390                 verify_osdc_wrlocked(osdc);
1391         else
1392                 verify_osdc_locked(osdc);
1393
1394         if (o != CEPH_HOMELESS_OSD)
1395                 osd = lookup_osd(&osdc->osds, o);
1396         else
1397                 osd = &osdc->homeless_osd;
1398         if (!osd) {
1399                 if (!wrlocked)
1400                         return ERR_PTR(-EAGAIN);
1401
1402                 osd = create_osd(osdc, o);
1403                 insert_osd(&osdc->osds, osd);
1404                 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd,
1405                               &osdc->osdmap->osd_addr[osd->o_osd]);
1406         }
1407
1408         dout("%s osdc %p osd%d -> osd %p\n", __func__, osdc, o, osd);
1409         return osd;
1410 }
1411
1412 /*
1413  * Create request <-> OSD session relation.
1414  *
1415  * @req has to be assigned a tid, @osd may be homeless.
1416  */
1417 static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req)
1418 {
1419         verify_osd_locked(osd);
1420         WARN_ON(!req->r_tid || req->r_osd);
1421         dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd,
1422              req, req->r_tid);
1423
1424         if (!osd_homeless(osd))
1425                 __remove_osd_from_lru(osd);
1426         else
1427                 atomic_inc(&osd->o_osdc->num_homeless);
1428
1429         get_osd(osd);
1430         insert_request(&osd->o_requests, req);
1431         req->r_osd = osd;
1432 }
1433
1434 static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req)
1435 {
1436         verify_osd_locked(osd);
1437         WARN_ON(req->r_osd != osd);
1438         dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd,
1439              req, req->r_tid);
1440
1441         req->r_osd = NULL;
1442         erase_request(&osd->o_requests, req);
1443         put_osd(osd);
1444
1445         if (!osd_homeless(osd))
1446                 maybe_move_osd_to_lru(osd);
1447         else
1448                 atomic_dec(&osd->o_osdc->num_homeless);
1449 }
1450
1451 static bool __pool_full(struct ceph_pg_pool_info *pi)
1452 {
1453         return pi->flags & CEPH_POOL_FLAG_FULL;
1454 }
1455
1456 static bool have_pool_full(struct ceph_osd_client *osdc)
1457 {
1458         struct rb_node *n;
1459
1460         for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) {
1461                 struct ceph_pg_pool_info *pi =
1462                     rb_entry(n, struct ceph_pg_pool_info, node);
1463
1464                 if (__pool_full(pi))
1465                         return true;
1466         }
1467
1468         return false;
1469 }
1470
1471 static bool pool_full(struct ceph_osd_client *osdc, s64 pool_id)
1472 {
1473         struct ceph_pg_pool_info *pi;
1474
1475         pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id);
1476         if (!pi)
1477                 return false;
1478
1479         return __pool_full(pi);
1480 }
1481
1482 /*
1483  * Returns whether a request should be blocked from being sent
1484  * based on the current osdmap and osd_client settings.
1485  */
1486 static bool target_should_be_paused(struct ceph_osd_client *osdc,
1487                                     const struct ceph_osd_request_target *t,
1488                                     struct ceph_pg_pool_info *pi)
1489 {
1490         bool pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
1491         bool pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
1492                        ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
1493                        __pool_full(pi);
1494
1495         WARN_ON(pi->id != t->target_oloc.pool);
1496         return ((t->flags & CEPH_OSD_FLAG_READ) && pauserd) ||
1497                ((t->flags & CEPH_OSD_FLAG_WRITE) && pausewr) ||
1498                (osdc->osdmap->epoch < osdc->epoch_barrier);
1499 }
1500
1501 static int pick_random_replica(const struct ceph_osds *acting)
1502 {
1503         int i = prandom_u32() % acting->size;
1504
1505         dout("%s picked osd%d, primary osd%d\n", __func__,
1506              acting->osds[i], acting->primary);
1507         return i;
1508 }
1509
1510 /*
1511  * Picks the closest replica based on client's location given by
1512  * crush_location option.  Prefers the primary if the locality is
1513  * the same.
1514  */
1515 static int pick_closest_replica(struct ceph_osd_client *osdc,
1516                                 const struct ceph_osds *acting)
1517 {
1518         struct ceph_options *opt = osdc->client->options;
1519         int best_i, best_locality;
1520         int i = 0, locality;
1521
1522         do {
1523                 locality = ceph_get_crush_locality(osdc->osdmap,
1524                                                    acting->osds[i],
1525                                                    &opt->crush_locs);
1526                 if (i == 0 ||
1527                     (locality >= 0 && best_locality < 0) ||
1528                     (locality >= 0 && best_locality >= 0 &&
1529                      locality < best_locality)) {
1530                         best_i = i;
1531                         best_locality = locality;
1532                 }
1533         } while (++i < acting->size);
1534
1535         dout("%s picked osd%d with locality %d, primary osd%d\n", __func__,
1536              acting->osds[best_i], best_locality, acting->primary);
1537         return best_i;
1538 }
1539
1540 enum calc_target_result {
1541         CALC_TARGET_NO_ACTION = 0,
1542         CALC_TARGET_NEED_RESEND,
1543         CALC_TARGET_POOL_DNE,
1544 };
1545
1546 static enum calc_target_result calc_target(struct ceph_osd_client *osdc,
1547                                            struct ceph_osd_request_target *t,
1548                                            bool any_change)
1549 {
1550         struct ceph_pg_pool_info *pi;
1551         struct ceph_pg pgid, last_pgid;
1552         struct ceph_osds up, acting;
1553         bool is_read = t->flags & CEPH_OSD_FLAG_READ;
1554         bool is_write = t->flags & CEPH_OSD_FLAG_WRITE;
1555         bool force_resend = false;
1556         bool unpaused = false;
1557         bool legacy_change = false;
1558         bool split = false;
1559         bool sort_bitwise = ceph_osdmap_flag(osdc, CEPH_OSDMAP_SORTBITWISE);
1560         bool recovery_deletes = ceph_osdmap_flag(osdc,
1561                                                  CEPH_OSDMAP_RECOVERY_DELETES);
1562         enum calc_target_result ct_res;
1563
1564         t->epoch = osdc->osdmap->epoch;
1565         pi = ceph_pg_pool_by_id(osdc->osdmap, t->base_oloc.pool);
1566         if (!pi) {
1567                 t->osd = CEPH_HOMELESS_OSD;
1568                 ct_res = CALC_TARGET_POOL_DNE;
1569                 goto out;
1570         }
1571
1572         if (osdc->osdmap->epoch == pi->last_force_request_resend) {
1573                 if (t->last_force_resend < pi->last_force_request_resend) {
1574                         t->last_force_resend = pi->last_force_request_resend;
1575                         force_resend = true;
1576                 } else if (t->last_force_resend == 0) {
1577                         force_resend = true;
1578                 }
1579         }
1580
1581         /* apply tiering */
1582         ceph_oid_copy(&t->target_oid, &t->base_oid);
1583         ceph_oloc_copy(&t->target_oloc, &t->base_oloc);
1584         if ((t->flags & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0) {
1585                 if (is_read && pi->read_tier >= 0)
1586                         t->target_oloc.pool = pi->read_tier;
1587                 if (is_write && pi->write_tier >= 0)
1588                         t->target_oloc.pool = pi->write_tier;
1589
1590                 pi = ceph_pg_pool_by_id(osdc->osdmap, t->target_oloc.pool);
1591                 if (!pi) {
1592                         t->osd = CEPH_HOMELESS_OSD;
1593                         ct_res = CALC_TARGET_POOL_DNE;
1594                         goto out;
1595                 }
1596         }
1597
1598         __ceph_object_locator_to_pg(pi, &t->target_oid, &t->target_oloc, &pgid);
1599         last_pgid.pool = pgid.pool;
1600         last_pgid.seed = ceph_stable_mod(pgid.seed, t->pg_num, t->pg_num_mask);
1601
1602         ceph_pg_to_up_acting_osds(osdc->osdmap, pi, &pgid, &up, &acting);
1603         if (any_change &&
1604             ceph_is_new_interval(&t->acting,
1605                                  &acting,
1606                                  &t->up,
1607                                  &up,
1608                                  t->size,
1609                                  pi->size,
1610                                  t->min_size,
1611                                  pi->min_size,
1612                                  t->pg_num,
1613                                  pi->pg_num,
1614                                  t->sort_bitwise,
1615                                  sort_bitwise,
1616                                  t->recovery_deletes,
1617                                  recovery_deletes,
1618                                  &last_pgid))
1619                 force_resend = true;
1620
1621         if (t->paused && !target_should_be_paused(osdc, t, pi)) {
1622                 t->paused = false;
1623                 unpaused = true;
1624         }
1625         legacy_change = ceph_pg_compare(&t->pgid, &pgid) ||
1626                         ceph_osds_changed(&t->acting, &acting,
1627                                           t->used_replica || any_change);
1628         if (t->pg_num)
1629                 split = ceph_pg_is_split(&last_pgid, t->pg_num, pi->pg_num);
1630
1631         if (legacy_change || force_resend || split) {
1632                 t->pgid = pgid; /* struct */
1633                 ceph_pg_to_primary_shard(osdc->osdmap, pi, &pgid, &t->spgid);
1634                 ceph_osds_copy(&t->acting, &acting);
1635                 ceph_osds_copy(&t->up, &up);
1636                 t->size = pi->size;
1637                 t->min_size = pi->min_size;
1638                 t->pg_num = pi->pg_num;
1639                 t->pg_num_mask = pi->pg_num_mask;
1640                 t->sort_bitwise = sort_bitwise;
1641                 t->recovery_deletes = recovery_deletes;
1642
1643                 if ((t->flags & (CEPH_OSD_FLAG_BALANCE_READS |
1644                                  CEPH_OSD_FLAG_LOCALIZE_READS)) &&
1645                     !is_write && pi->type == CEPH_POOL_TYPE_REP &&
1646                     acting.size > 1) {
1647                         int pos;
1648
1649                         WARN_ON(!is_read || acting.osds[0] != acting.primary);
1650                         if (t->flags & CEPH_OSD_FLAG_BALANCE_READS) {
1651                                 pos = pick_random_replica(&acting);
1652                         } else {
1653                                 pos = pick_closest_replica(osdc, &acting);
1654                         }
1655                         t->osd = acting.osds[pos];
1656                         t->used_replica = pos > 0;
1657                 } else {
1658                         t->osd = acting.primary;
1659                         t->used_replica = false;
1660                 }
1661         }
1662
1663         if (unpaused || legacy_change || force_resend || split)
1664                 ct_res = CALC_TARGET_NEED_RESEND;
1665         else
1666                 ct_res = CALC_TARGET_NO_ACTION;
1667
1668 out:
1669         dout("%s t %p -> %d%d%d%d ct_res %d osd%d\n", __func__, t, unpaused,
1670              legacy_change, force_resend, split, ct_res, t->osd);
1671         return ct_res;
1672 }
1673
1674 static struct ceph_spg_mapping *alloc_spg_mapping(void)
1675 {
1676         struct ceph_spg_mapping *spg;
1677
1678         spg = kmalloc(sizeof(*spg), GFP_NOIO);
1679         if (!spg)
1680                 return NULL;
1681
1682         RB_CLEAR_NODE(&spg->node);
1683         spg->backoffs = RB_ROOT;
1684         return spg;
1685 }
1686
1687 static void free_spg_mapping(struct ceph_spg_mapping *spg)
1688 {
1689         WARN_ON(!RB_EMPTY_NODE(&spg->node));
1690         WARN_ON(!RB_EMPTY_ROOT(&spg->backoffs));
1691
1692         kfree(spg);
1693 }
1694
1695 /*
1696  * rbtree of ceph_spg_mapping for handling map<spg_t, ...>, similar to
1697  * ceph_pg_mapping.  Used to track OSD backoffs -- a backoff [range] is
1698  * defined only within a specific spgid; it does not pass anything to
1699  * children on split, or to another primary.
1700  */
1701 DEFINE_RB_FUNCS2(spg_mapping, struct ceph_spg_mapping, spgid, ceph_spg_compare,
1702                  RB_BYPTR, const struct ceph_spg *, node)
1703
1704 static u64 hoid_get_bitwise_key(const struct ceph_hobject_id *hoid)
1705 {
1706         return hoid->is_max ? 0x100000000ull : hoid->hash_reverse_bits;
1707 }
1708
1709 static void hoid_get_effective_key(const struct ceph_hobject_id *hoid,
1710                                    void **pkey, size_t *pkey_len)
1711 {
1712         if (hoid->key_len) {
1713                 *pkey = hoid->key;
1714                 *pkey_len = hoid->key_len;
1715         } else {
1716                 *pkey = hoid->oid;
1717                 *pkey_len = hoid->oid_len;
1718         }
1719 }
1720
1721 static int compare_names(const void *name1, size_t name1_len,
1722                          const void *name2, size_t name2_len)
1723 {
1724         int ret;
1725
1726         ret = memcmp(name1, name2, min(name1_len, name2_len));
1727         if (!ret) {
1728                 if (name1_len < name2_len)
1729                         ret = -1;
1730                 else if (name1_len > name2_len)
1731                         ret = 1;
1732         }
1733         return ret;
1734 }
1735
1736 static int hoid_compare(const struct ceph_hobject_id *lhs,
1737                         const struct ceph_hobject_id *rhs)
1738 {
1739         void *effective_key1, *effective_key2;
1740         size_t effective_key1_len, effective_key2_len;
1741         int ret;
1742
1743         if (lhs->is_max < rhs->is_max)
1744                 return -1;
1745         if (lhs->is_max > rhs->is_max)
1746                 return 1;
1747
1748         if (lhs->pool < rhs->pool)
1749                 return -1;
1750         if (lhs->pool > rhs->pool)
1751                 return 1;
1752
1753         if (hoid_get_bitwise_key(lhs) < hoid_get_bitwise_key(rhs))
1754                 return -1;
1755         if (hoid_get_bitwise_key(lhs) > hoid_get_bitwise_key(rhs))
1756                 return 1;
1757
1758         ret = compare_names(lhs->nspace, lhs->nspace_len,
1759                             rhs->nspace, rhs->nspace_len);
1760         if (ret)
1761                 return ret;
1762
1763         hoid_get_effective_key(lhs, &effective_key1, &effective_key1_len);
1764         hoid_get_effective_key(rhs, &effective_key2, &effective_key2_len);
1765         ret = compare_names(effective_key1, effective_key1_len,
1766                             effective_key2, effective_key2_len);
1767         if (ret)
1768                 return ret;
1769
1770         ret = compare_names(lhs->oid, lhs->oid_len, rhs->oid, rhs->oid_len);
1771         if (ret)
1772                 return ret;
1773
1774         if (lhs->snapid < rhs->snapid)
1775                 return -1;
1776         if (lhs->snapid > rhs->snapid)
1777                 return 1;
1778
1779         return 0;
1780 }
1781
1782 /*
1783  * For decoding ->begin and ->end of MOSDBackoff only -- no MIN/MAX
1784  * compat stuff here.
1785  *
1786  * Assumes @hoid is zero-initialized.
1787  */
1788 static int decode_hoid(void **p, void *end, struct ceph_hobject_id *hoid)
1789 {
1790         u8 struct_v;
1791         u32 struct_len;
1792         int ret;
1793
1794         ret = ceph_start_decoding(p, end, 4, "hobject_t", &struct_v,
1795                                   &struct_len);
1796         if (ret)
1797                 return ret;
1798
1799         if (struct_v < 4) {
1800                 pr_err("got struct_v %d < 4 of hobject_t\n", struct_v);
1801                 goto e_inval;
1802         }
1803
1804         hoid->key = ceph_extract_encoded_string(p, end, &hoid->key_len,
1805                                                 GFP_NOIO);
1806         if (IS_ERR(hoid->key)) {
1807                 ret = PTR_ERR(hoid->key);
1808                 hoid->key = NULL;
1809                 return ret;
1810         }
1811
1812         hoid->oid = ceph_extract_encoded_string(p, end, &hoid->oid_len,
1813                                                 GFP_NOIO);
1814         if (IS_ERR(hoid->oid)) {
1815                 ret = PTR_ERR(hoid->oid);
1816                 hoid->oid = NULL;
1817                 return ret;
1818         }
1819
1820         ceph_decode_64_safe(p, end, hoid->snapid, e_inval);
1821         ceph_decode_32_safe(p, end, hoid->hash, e_inval);
1822         ceph_decode_8_safe(p, end, hoid->is_max, e_inval);
1823
1824         hoid->nspace = ceph_extract_encoded_string(p, end, &hoid->nspace_len,
1825                                                    GFP_NOIO);
1826         if (IS_ERR(hoid->nspace)) {
1827                 ret = PTR_ERR(hoid->nspace);
1828                 hoid->nspace = NULL;
1829                 return ret;
1830         }
1831
1832         ceph_decode_64_safe(p, end, hoid->pool, e_inval);
1833
1834         ceph_hoid_build_hash_cache(hoid);
1835         return 0;
1836
1837 e_inval:
1838         return -EINVAL;
1839 }
1840
1841 static int hoid_encoding_size(const struct ceph_hobject_id *hoid)
1842 {
1843         return 8 + 4 + 1 + 8 + /* snapid, hash, is_max, pool */
1844                4 + hoid->key_len + 4 + hoid->oid_len + 4 + hoid->nspace_len;
1845 }
1846
1847 static void encode_hoid(void **p, void *end, const struct ceph_hobject_id *hoid)
1848 {
1849         ceph_start_encoding(p, 4, 3, hoid_encoding_size(hoid));
1850         ceph_encode_string(p, end, hoid->key, hoid->key_len);
1851         ceph_encode_string(p, end, hoid->oid, hoid->oid_len);
1852         ceph_encode_64(p, hoid->snapid);
1853         ceph_encode_32(p, hoid->hash);
1854         ceph_encode_8(p, hoid->is_max);
1855         ceph_encode_string(p, end, hoid->nspace, hoid->nspace_len);
1856         ceph_encode_64(p, hoid->pool);
1857 }
1858
1859 static void free_hoid(struct ceph_hobject_id *hoid)
1860 {
1861         if (hoid) {
1862                 kfree(hoid->key);
1863                 kfree(hoid->oid);
1864                 kfree(hoid->nspace);
1865                 kfree(hoid);
1866         }
1867 }
1868
1869 static struct ceph_osd_backoff *alloc_backoff(void)
1870 {
1871         struct ceph_osd_backoff *backoff;
1872
1873         backoff = kzalloc(sizeof(*backoff), GFP_NOIO);
1874         if (!backoff)
1875                 return NULL;
1876
1877         RB_CLEAR_NODE(&backoff->spg_node);
1878         RB_CLEAR_NODE(&backoff->id_node);
1879         return backoff;
1880 }
1881
1882 static void free_backoff(struct ceph_osd_backoff *backoff)
1883 {
1884         WARN_ON(!RB_EMPTY_NODE(&backoff->spg_node));
1885         WARN_ON(!RB_EMPTY_NODE(&backoff->id_node));
1886
1887         free_hoid(backoff->begin);
1888         free_hoid(backoff->end);
1889         kfree(backoff);
1890 }
1891
1892 /*
1893  * Within a specific spgid, backoffs are managed by ->begin hoid.
1894  */
1895 DEFINE_RB_INSDEL_FUNCS2(backoff, struct ceph_osd_backoff, begin, hoid_compare,
1896                         RB_BYVAL, spg_node);
1897
1898 static struct ceph_osd_backoff *lookup_containing_backoff(struct rb_root *root,
1899                                             const struct ceph_hobject_id *hoid)
1900 {
1901         struct rb_node *n = root->rb_node;
1902
1903         while (n) {
1904                 struct ceph_osd_backoff *cur =
1905                     rb_entry(n, struct ceph_osd_backoff, spg_node);
1906                 int cmp;
1907
1908                 cmp = hoid_compare(hoid, cur->begin);
1909                 if (cmp < 0) {
1910                         n = n->rb_left;
1911                 } else if (cmp > 0) {
1912                         if (hoid_compare(hoid, cur->end) < 0)
1913                                 return cur;
1914
1915                         n = n->rb_right;
1916                 } else {
1917                         return cur;
1918                 }
1919         }
1920
1921         return NULL;
1922 }
1923
1924 /*
1925  * Each backoff has a unique id within its OSD session.
1926  */
1927 DEFINE_RB_FUNCS(backoff_by_id, struct ceph_osd_backoff, id, id_node)
1928
1929 static void clear_backoffs(struct ceph_osd *osd)
1930 {
1931         while (!RB_EMPTY_ROOT(&osd->o_backoff_mappings)) {
1932                 struct ceph_spg_mapping *spg =
1933                     rb_entry(rb_first(&osd->o_backoff_mappings),
1934                              struct ceph_spg_mapping, node);
1935
1936                 while (!RB_EMPTY_ROOT(&spg->backoffs)) {
1937                         struct ceph_osd_backoff *backoff =
1938                             rb_entry(rb_first(&spg->backoffs),
1939                                      struct ceph_osd_backoff, spg_node);
1940
1941                         erase_backoff(&spg->backoffs, backoff);
1942                         erase_backoff_by_id(&osd->o_backoffs_by_id, backoff);
1943                         free_backoff(backoff);
1944                 }
1945                 erase_spg_mapping(&osd->o_backoff_mappings, spg);
1946                 free_spg_mapping(spg);
1947         }
1948 }
1949
1950 /*
1951  * Set up a temporary, non-owning view into @t.
1952  */
1953 static void hoid_fill_from_target(struct ceph_hobject_id *hoid,
1954                                   const struct ceph_osd_request_target *t)
1955 {
1956         hoid->key = NULL;
1957         hoid->key_len = 0;
1958         hoid->oid = t->target_oid.name;
1959         hoid->oid_len = t->target_oid.name_len;
1960         hoid->snapid = CEPH_NOSNAP;
1961         hoid->hash = t->pgid.seed;
1962         hoid->is_max = false;
1963         if (t->target_oloc.pool_ns) {
1964                 hoid->nspace = t->target_oloc.pool_ns->str;
1965                 hoid->nspace_len = t->target_oloc.pool_ns->len;
1966         } else {
1967                 hoid->nspace = NULL;
1968                 hoid->nspace_len = 0;
1969         }
1970         hoid->pool = t->target_oloc.pool;
1971         ceph_hoid_build_hash_cache(hoid);
1972 }
1973
1974 static bool should_plug_request(struct ceph_osd_request *req)
1975 {
1976         struct ceph_osd *osd = req->r_osd;
1977         struct ceph_spg_mapping *spg;
1978         struct ceph_osd_backoff *backoff;
1979         struct ceph_hobject_id hoid;
1980
1981         spg = lookup_spg_mapping(&osd->o_backoff_mappings, &req->r_t.spgid);
1982         if (!spg)
1983                 return false;
1984
1985         hoid_fill_from_target(&hoid, &req->r_t);
1986         backoff = lookup_containing_backoff(&spg->backoffs, &hoid);
1987         if (!backoff)
1988                 return false;
1989
1990         dout("%s req %p tid %llu backoff osd%d spgid %llu.%xs%d id %llu\n",
1991              __func__, req, req->r_tid, osd->o_osd, backoff->spgid.pgid.pool,
1992              backoff->spgid.pgid.seed, backoff->spgid.shard, backoff->id);
1993         return true;
1994 }
1995
1996 /*
1997  * Keep get_num_data_items() in sync with this function.
1998  */
1999 static void setup_request_data(struct ceph_osd_request *req)
2000 {
2001         struct ceph_msg *request_msg = req->r_request;
2002         struct ceph_msg *reply_msg = req->r_reply;
2003         struct ceph_osd_req_op *op;
2004
2005         if (req->r_request->num_data_items || req->r_reply->num_data_items)
2006                 return;
2007
2008         WARN_ON(request_msg->data_length || reply_msg->data_length);
2009         for (op = req->r_ops; op != &req->r_ops[req->r_num_ops]; op++) {
2010                 switch (op->op) {
2011                 /* request */
2012                 case CEPH_OSD_OP_WRITE:
2013                 case CEPH_OSD_OP_WRITEFULL:
2014                         WARN_ON(op->indata_len != op->extent.length);
2015                         ceph_osdc_msg_data_add(request_msg,
2016                                                &op->extent.osd_data);
2017                         break;
2018                 case CEPH_OSD_OP_SETXATTR:
2019                 case CEPH_OSD_OP_CMPXATTR:
2020                         WARN_ON(op->indata_len != op->xattr.name_len +
2021                                                   op->xattr.value_len);
2022                         ceph_osdc_msg_data_add(request_msg,
2023                                                &op->xattr.osd_data);
2024                         break;
2025                 case CEPH_OSD_OP_NOTIFY_ACK:
2026                         ceph_osdc_msg_data_add(request_msg,
2027                                                &op->notify_ack.request_data);
2028                         break;
2029                 case CEPH_OSD_OP_COPY_FROM2:
2030                         ceph_osdc_msg_data_add(request_msg,
2031                                                &op->copy_from.osd_data);
2032                         break;
2033
2034                 /* reply */
2035                 case CEPH_OSD_OP_STAT:
2036                         ceph_osdc_msg_data_add(reply_msg,
2037                                                &op->raw_data_in);
2038                         break;
2039                 case CEPH_OSD_OP_READ:
2040                         ceph_osdc_msg_data_add(reply_msg,
2041                                                &op->extent.osd_data);
2042                         break;
2043                 case CEPH_OSD_OP_LIST_WATCHERS:
2044                         ceph_osdc_msg_data_add(reply_msg,
2045                                                &op->list_watchers.response_data);
2046                         break;
2047
2048                 /* both */
2049                 case CEPH_OSD_OP_CALL:
2050                         WARN_ON(op->indata_len != op->cls.class_len +
2051                                                   op->cls.method_len +
2052                                                   op->cls.indata_len);
2053                         ceph_osdc_msg_data_add(request_msg,
2054                                                &op->cls.request_info);
2055                         /* optional, can be NONE */
2056                         ceph_osdc_msg_data_add(request_msg,
2057                                                &op->cls.request_data);
2058                         /* optional, can be NONE */
2059                         ceph_osdc_msg_data_add(reply_msg,
2060                                                &op->cls.response_data);
2061                         break;
2062                 case CEPH_OSD_OP_NOTIFY:
2063                         ceph_osdc_msg_data_add(request_msg,
2064                                                &op->notify.request_data);
2065                         ceph_osdc_msg_data_add(reply_msg,
2066                                                &op->notify.response_data);
2067                         break;
2068                 }
2069         }
2070 }
2071
2072 static void encode_pgid(void **p, const struct ceph_pg *pgid)
2073 {
2074         ceph_encode_8(p, 1);
2075         ceph_encode_64(p, pgid->pool);
2076         ceph_encode_32(p, pgid->seed);
2077         ceph_encode_32(p, -1); /* preferred */
2078 }
2079
2080 static void encode_spgid(void **p, const struct ceph_spg *spgid)
2081 {
2082         ceph_start_encoding(p, 1, 1, CEPH_PGID_ENCODING_LEN + 1);
2083         encode_pgid(p, &spgid->pgid);
2084         ceph_encode_8(p, spgid->shard);
2085 }
2086
2087 static void encode_oloc(void **p, void *end,
2088                         const struct ceph_object_locator *oloc)
2089 {
2090         ceph_start_encoding(p, 5, 4, ceph_oloc_encoding_size(oloc));
2091         ceph_encode_64(p, oloc->pool);
2092         ceph_encode_32(p, -1); /* preferred */
2093         ceph_encode_32(p, 0);  /* key len */
2094         if (oloc->pool_ns)
2095                 ceph_encode_string(p, end, oloc->pool_ns->str,
2096                                    oloc->pool_ns->len);
2097         else
2098                 ceph_encode_32(p, 0);
2099 }
2100
2101 static void encode_request_partial(struct ceph_osd_request *req,
2102                                    struct ceph_msg *msg)
2103 {
2104         void *p = msg->front.iov_base;
2105         void *const end = p + msg->front_alloc_len;
2106         u32 data_len = 0;
2107         int i;
2108
2109         if (req->r_flags & CEPH_OSD_FLAG_WRITE) {
2110                 /* snapshots aren't writeable */
2111                 WARN_ON(req->r_snapid != CEPH_NOSNAP);
2112         } else {
2113                 WARN_ON(req->r_mtime.tv_sec || req->r_mtime.tv_nsec ||
2114                         req->r_data_offset || req->r_snapc);
2115         }
2116
2117         setup_request_data(req);
2118
2119         encode_spgid(&p, &req->r_t.spgid); /* actual spg */
2120         ceph_encode_32(&p, req->r_t.pgid.seed); /* raw hash */
2121         ceph_encode_32(&p, req->r_osdc->osdmap->epoch);
2122         ceph_encode_32(&p, req->r_flags);
2123
2124         /* reqid */
2125         ceph_start_encoding(&p, 2, 2, sizeof(struct ceph_osd_reqid));
2126         memset(p, 0, sizeof(struct ceph_osd_reqid));
2127         p += sizeof(struct ceph_osd_reqid);
2128
2129         /* trace */
2130         memset(p, 0, sizeof(struct ceph_blkin_trace_info));
2131         p += sizeof(struct ceph_blkin_trace_info);
2132
2133         ceph_encode_32(&p, 0); /* client_inc, always 0 */
2134         ceph_encode_timespec64(p, &req->r_mtime);
2135         p += sizeof(struct ceph_timespec);
2136
2137         encode_oloc(&p, end, &req->r_t.target_oloc);
2138         ceph_encode_string(&p, end, req->r_t.target_oid.name,
2139                            req->r_t.target_oid.name_len);
2140
2141         /* ops, can imply data */
2142         ceph_encode_16(&p, req->r_num_ops);
2143         for (i = 0; i < req->r_num_ops; i++) {
2144                 data_len += osd_req_encode_op(p, &req->r_ops[i]);
2145                 p += sizeof(struct ceph_osd_op);
2146         }
2147
2148         ceph_encode_64(&p, req->r_snapid); /* snapid */
2149         if (req->r_snapc) {
2150                 ceph_encode_64(&p, req->r_snapc->seq);
2151                 ceph_encode_32(&p, req->r_snapc->num_snaps);
2152                 for (i = 0; i < req->r_snapc->num_snaps; i++)
2153                         ceph_encode_64(&p, req->r_snapc->snaps[i]);
2154         } else {
2155                 ceph_encode_64(&p, 0); /* snap_seq */
2156                 ceph_encode_32(&p, 0); /* snaps len */
2157         }
2158
2159         ceph_encode_32(&p, req->r_attempts); /* retry_attempt */
2160         BUG_ON(p > end - 8); /* space for features */
2161
2162         msg->hdr.version = cpu_to_le16(8); /* MOSDOp v8 */
2163         /* front_len is finalized in encode_request_finish() */
2164         msg->front.iov_len = p - msg->front.iov_base;
2165         msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2166         msg->hdr.data_len = cpu_to_le32(data_len);
2167         /*
2168          * The header "data_off" is a hint to the receiver allowing it
2169          * to align received data into its buffers such that there's no
2170          * need to re-copy it before writing it to disk (direct I/O).
2171          */
2172         msg->hdr.data_off = cpu_to_le16(req->r_data_offset);
2173
2174         dout("%s req %p msg %p oid %s oid_len %d\n", __func__, req, msg,
2175              req->r_t.target_oid.name, req->r_t.target_oid.name_len);
2176 }
2177
2178 static void encode_request_finish(struct ceph_msg *msg)
2179 {
2180         void *p = msg->front.iov_base;
2181         void *const partial_end = p + msg->front.iov_len;
2182         void *const end = p + msg->front_alloc_len;
2183
2184         if (CEPH_HAVE_FEATURE(msg->con->peer_features, RESEND_ON_SPLIT)) {
2185                 /* luminous OSD -- encode features and be done */
2186                 p = partial_end;
2187                 ceph_encode_64(&p, msg->con->peer_features);
2188         } else {
2189                 struct {
2190                         char spgid[CEPH_ENCODING_START_BLK_LEN +
2191                                    CEPH_PGID_ENCODING_LEN + 1];
2192                         __le32 hash;
2193                         __le32 epoch;
2194                         __le32 flags;
2195                         char reqid[CEPH_ENCODING_START_BLK_LEN +
2196                                    sizeof(struct ceph_osd_reqid)];
2197                         char trace[sizeof(struct ceph_blkin_trace_info)];
2198                         __le32 client_inc;
2199                         struct ceph_timespec mtime;
2200                 } __packed head;
2201                 struct ceph_pg pgid;
2202                 void *oloc, *oid, *tail;
2203                 int oloc_len, oid_len, tail_len;
2204                 int len;
2205
2206                 /*
2207                  * Pre-luminous OSD -- reencode v8 into v4 using @head
2208                  * as a temporary buffer.  Encode the raw PG; the rest
2209                  * is just a matter of moving oloc, oid and tail blobs
2210                  * around.
2211                  */
2212                 memcpy(&head, p, sizeof(head));
2213                 p += sizeof(head);
2214
2215                 oloc = p;
2216                 p += CEPH_ENCODING_START_BLK_LEN;
2217                 pgid.pool = ceph_decode_64(&p);
2218                 p += 4 + 4; /* preferred, key len */
2219                 len = ceph_decode_32(&p);
2220                 p += len;   /* nspace */
2221                 oloc_len = p - oloc;
2222
2223                 oid = p;
2224                 len = ceph_decode_32(&p);
2225                 p += len;
2226                 oid_len = p - oid;
2227
2228                 tail = p;
2229                 tail_len = partial_end - p;
2230
2231                 p = msg->front.iov_base;
2232                 ceph_encode_copy(&p, &head.client_inc, sizeof(head.client_inc));
2233                 ceph_encode_copy(&p, &head.epoch, sizeof(head.epoch));
2234                 ceph_encode_copy(&p, &head.flags, sizeof(head.flags));
2235                 ceph_encode_copy(&p, &head.mtime, sizeof(head.mtime));
2236
2237                 /* reassert_version */
2238                 memset(p, 0, sizeof(struct ceph_eversion));
2239                 p += sizeof(struct ceph_eversion);
2240
2241                 BUG_ON(p >= oloc);
2242                 memmove(p, oloc, oloc_len);
2243                 p += oloc_len;
2244
2245                 pgid.seed = le32_to_cpu(head.hash);
2246                 encode_pgid(&p, &pgid); /* raw pg */
2247
2248                 BUG_ON(p >= oid);
2249                 memmove(p, oid, oid_len);
2250                 p += oid_len;
2251
2252                 /* tail -- ops, snapid, snapc, retry_attempt */
2253                 BUG_ON(p >= tail);
2254                 memmove(p, tail, tail_len);
2255                 p += tail_len;
2256
2257                 msg->hdr.version = cpu_to_le16(4); /* MOSDOp v4 */
2258         }
2259
2260         BUG_ON(p > end);
2261         msg->front.iov_len = p - msg->front.iov_base;
2262         msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2263
2264         dout("%s msg %p tid %llu %u+%u+%u v%d\n", __func__, msg,
2265              le64_to_cpu(msg->hdr.tid), le32_to_cpu(msg->hdr.front_len),
2266              le32_to_cpu(msg->hdr.middle_len), le32_to_cpu(msg->hdr.data_len),
2267              le16_to_cpu(msg->hdr.version));
2268 }
2269
2270 /*
2271  * @req has to be assigned a tid and registered.
2272  */
2273 static void send_request(struct ceph_osd_request *req)
2274 {
2275         struct ceph_osd *osd = req->r_osd;
2276
2277         verify_osd_locked(osd);
2278         WARN_ON(osd->o_osd != req->r_t.osd);
2279
2280         /* backoff? */
2281         if (should_plug_request(req))
2282                 return;
2283
2284         /*
2285          * We may have a previously queued request message hanging
2286          * around.  Cancel it to avoid corrupting the msgr.
2287          */
2288         if (req->r_sent)
2289                 ceph_msg_revoke(req->r_request);
2290
2291         req->r_flags |= CEPH_OSD_FLAG_KNOWN_REDIR;
2292         if (req->r_attempts)
2293                 req->r_flags |= CEPH_OSD_FLAG_RETRY;
2294         else
2295                 WARN_ON(req->r_flags & CEPH_OSD_FLAG_RETRY);
2296
2297         encode_request_partial(req, req->r_request);
2298
2299         dout("%s req %p tid %llu to pgid %llu.%x spgid %llu.%xs%d osd%d e%u flags 0x%x attempt %d\n",
2300              __func__, req, req->r_tid, req->r_t.pgid.pool, req->r_t.pgid.seed,
2301              req->r_t.spgid.pgid.pool, req->r_t.spgid.pgid.seed,
2302              req->r_t.spgid.shard, osd->o_osd, req->r_t.epoch, req->r_flags,
2303              req->r_attempts);
2304
2305         req->r_t.paused = false;
2306         req->r_stamp = jiffies;
2307         req->r_attempts++;
2308
2309         req->r_sent = osd->o_incarnation;
2310         req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
2311         ceph_con_send(&osd->o_con, ceph_msg_get(req->r_request));
2312 }
2313
2314 static void maybe_request_map(struct ceph_osd_client *osdc)
2315 {
2316         bool continuous = false;
2317
2318         verify_osdc_locked(osdc);
2319         WARN_ON(!osdc->osdmap->epoch);
2320
2321         if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
2322             ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD) ||
2323             ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) {
2324                 dout("%s osdc %p continuous\n", __func__, osdc);
2325                 continuous = true;
2326         } else {
2327                 dout("%s osdc %p onetime\n", __func__, osdc);
2328         }
2329
2330         if (ceph_monc_want_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
2331                                osdc->osdmap->epoch + 1, continuous))
2332                 ceph_monc_renew_subs(&osdc->client->monc);
2333 }
2334
2335 static void complete_request(struct ceph_osd_request *req, int err);
2336 static void send_map_check(struct ceph_osd_request *req);
2337
2338 static void __submit_request(struct ceph_osd_request *req, bool wrlocked)
2339 {
2340         struct ceph_osd_client *osdc = req->r_osdc;
2341         struct ceph_osd *osd;
2342         enum calc_target_result ct_res;
2343         int err = 0;
2344         bool need_send = false;
2345         bool promoted = false;
2346
2347         WARN_ON(req->r_tid);
2348         dout("%s req %p wrlocked %d\n", __func__, req, wrlocked);
2349
2350 again:
2351         ct_res = calc_target(osdc, &req->r_t, false);
2352         if (ct_res == CALC_TARGET_POOL_DNE && !wrlocked)
2353                 goto promote;
2354
2355         osd = lookup_create_osd(osdc, req->r_t.osd, wrlocked);
2356         if (IS_ERR(osd)) {
2357                 WARN_ON(PTR_ERR(osd) != -EAGAIN || wrlocked);
2358                 goto promote;
2359         }
2360
2361         if (osdc->abort_err) {
2362                 dout("req %p abort_err %d\n", req, osdc->abort_err);
2363                 err = osdc->abort_err;
2364         } else if (osdc->osdmap->epoch < osdc->epoch_barrier) {
2365                 dout("req %p epoch %u barrier %u\n", req, osdc->osdmap->epoch,
2366                      osdc->epoch_barrier);
2367                 req->r_t.paused = true;
2368                 maybe_request_map(osdc);
2369         } else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
2370                    ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) {
2371                 dout("req %p pausewr\n", req);
2372                 req->r_t.paused = true;
2373                 maybe_request_map(osdc);
2374         } else if ((req->r_flags & CEPH_OSD_FLAG_READ) &&
2375                    ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) {
2376                 dout("req %p pauserd\n", req);
2377                 req->r_t.paused = true;
2378                 maybe_request_map(osdc);
2379         } else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
2380                    !(req->r_flags & (CEPH_OSD_FLAG_FULL_TRY |
2381                                      CEPH_OSD_FLAG_FULL_FORCE)) &&
2382                    (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
2383                     pool_full(osdc, req->r_t.base_oloc.pool))) {
2384                 dout("req %p full/pool_full\n", req);
2385                 if (ceph_test_opt(osdc->client, ABORT_ON_FULL)) {
2386                         err = -ENOSPC;
2387                 } else {
2388                         if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL))
2389                                 pr_warn_ratelimited("cluster is full (osdmap FULL)\n");
2390                         else
2391                                 pr_warn_ratelimited("pool %lld is full or reached quota\n",
2392                                                     req->r_t.base_oloc.pool);
2393                         req->r_t.paused = true;
2394                         maybe_request_map(osdc);
2395                 }
2396         } else if (!osd_homeless(osd)) {
2397                 need_send = true;
2398         } else {
2399                 maybe_request_map(osdc);
2400         }
2401
2402         mutex_lock(&osd->lock);
2403         /*
2404          * Assign the tid atomically with send_request() to protect
2405          * multiple writes to the same object from racing with each
2406          * other, resulting in out of order ops on the OSDs.
2407          */
2408         req->r_tid = atomic64_inc_return(&osdc->last_tid);
2409         link_request(osd, req);
2410         if (need_send)
2411                 send_request(req);
2412         else if (err)
2413                 complete_request(req, err);
2414         mutex_unlock(&osd->lock);
2415
2416         if (!err && ct_res == CALC_TARGET_POOL_DNE)
2417                 send_map_check(req);
2418
2419         if (promoted)
2420                 downgrade_write(&osdc->lock);
2421         return;
2422
2423 promote:
2424         up_read(&osdc->lock);
2425         down_write(&osdc->lock);
2426         wrlocked = true;
2427         promoted = true;
2428         goto again;
2429 }
2430
2431 static void account_request(struct ceph_osd_request *req)
2432 {
2433         WARN_ON(req->r_flags & (CEPH_OSD_FLAG_ACK | CEPH_OSD_FLAG_ONDISK));
2434         WARN_ON(!(req->r_flags & (CEPH_OSD_FLAG_READ | CEPH_OSD_FLAG_WRITE)));
2435
2436         req->r_flags |= CEPH_OSD_FLAG_ONDISK;
2437         atomic_inc(&req->r_osdc->num_requests);
2438
2439         req->r_start_stamp = jiffies;
2440         req->r_start_latency = ktime_get();
2441 }
2442
2443 static void submit_request(struct ceph_osd_request *req, bool wrlocked)
2444 {
2445         ceph_osdc_get_request(req);
2446         account_request(req);
2447         __submit_request(req, wrlocked);
2448 }
2449
2450 static void finish_request(struct ceph_osd_request *req)
2451 {
2452         struct ceph_osd_client *osdc = req->r_osdc;
2453
2454         WARN_ON(lookup_request_mc(&osdc->map_checks, req->r_tid));
2455         dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
2456
2457         req->r_end_latency = ktime_get();
2458
2459         if (req->r_osd)
2460                 unlink_request(req->r_osd, req);
2461         atomic_dec(&osdc->num_requests);
2462
2463         /*
2464          * If an OSD has failed or returned and a request has been sent
2465          * twice, it's possible to get a reply and end up here while the
2466          * request message is queued for delivery.  We will ignore the
2467          * reply, so not a big deal, but better to try and catch it.
2468          */
2469         ceph_msg_revoke(req->r_request);
2470         ceph_msg_revoke_incoming(req->r_reply);
2471 }
2472
2473 static void __complete_request(struct ceph_osd_request *req)
2474 {
2475         dout("%s req %p tid %llu cb %ps result %d\n", __func__, req,
2476              req->r_tid, req->r_callback, req->r_result);
2477
2478         if (req->r_callback)
2479                 req->r_callback(req);
2480         complete_all(&req->r_completion);
2481         ceph_osdc_put_request(req);
2482 }
2483
2484 static void complete_request_workfn(struct work_struct *work)
2485 {
2486         struct ceph_osd_request *req =
2487             container_of(work, struct ceph_osd_request, r_complete_work);
2488
2489         __complete_request(req);
2490 }
2491
2492 /*
2493  * This is open-coded in handle_reply().
2494  */
2495 static void complete_request(struct ceph_osd_request *req, int err)
2496 {
2497         dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err);
2498
2499         req->r_result = err;
2500         finish_request(req);
2501
2502         INIT_WORK(&req->r_complete_work, complete_request_workfn);
2503         queue_work(req->r_osdc->completion_wq, &req->r_complete_work);
2504 }
2505
2506 static void cancel_map_check(struct ceph_osd_request *req)
2507 {
2508         struct ceph_osd_client *osdc = req->r_osdc;
2509         struct ceph_osd_request *lookup_req;
2510
2511         verify_osdc_wrlocked(osdc);
2512
2513         lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid);
2514         if (!lookup_req)
2515                 return;
2516
2517         WARN_ON(lookup_req != req);
2518         erase_request_mc(&osdc->map_checks, req);
2519         ceph_osdc_put_request(req);
2520 }
2521
2522 static void cancel_request(struct ceph_osd_request *req)
2523 {
2524         dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
2525
2526         cancel_map_check(req);
2527         finish_request(req);
2528         complete_all(&req->r_completion);
2529         ceph_osdc_put_request(req);
2530 }
2531
2532 static void abort_request(struct ceph_osd_request *req, int err)
2533 {
2534         dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err);
2535
2536         cancel_map_check(req);
2537         complete_request(req, err);
2538 }
2539
2540 static int abort_fn(struct ceph_osd_request *req, void *arg)
2541 {
2542         int err = *(int *)arg;
2543
2544         abort_request(req, err);
2545         return 0; /* continue iteration */
2546 }
2547
2548 /*
2549  * Abort all in-flight requests with @err and arrange for all future
2550  * requests to be failed immediately.
2551  */
2552 void ceph_osdc_abort_requests(struct ceph_osd_client *osdc, int err)
2553 {
2554         dout("%s osdc %p err %d\n", __func__, osdc, err);
2555         down_write(&osdc->lock);
2556         for_each_request(osdc, abort_fn, &err);
2557         osdc->abort_err = err;
2558         up_write(&osdc->lock);
2559 }
2560 EXPORT_SYMBOL(ceph_osdc_abort_requests);
2561
2562 void ceph_osdc_clear_abort_err(struct ceph_osd_client *osdc)
2563 {
2564         down_write(&osdc->lock);
2565         osdc->abort_err = 0;
2566         up_write(&osdc->lock);
2567 }
2568 EXPORT_SYMBOL(ceph_osdc_clear_abort_err);
2569
2570 static void update_epoch_barrier(struct ceph_osd_client *osdc, u32 eb)
2571 {
2572         if (likely(eb > osdc->epoch_barrier)) {
2573                 dout("updating epoch_barrier from %u to %u\n",
2574                                 osdc->epoch_barrier, eb);
2575                 osdc->epoch_barrier = eb;
2576                 /* Request map if we're not to the barrier yet */
2577                 if (eb > osdc->osdmap->epoch)
2578                         maybe_request_map(osdc);
2579         }
2580 }
2581
2582 void ceph_osdc_update_epoch_barrier(struct ceph_osd_client *osdc, u32 eb)
2583 {
2584         down_read(&osdc->lock);
2585         if (unlikely(eb > osdc->epoch_barrier)) {
2586                 up_read(&osdc->lock);
2587                 down_write(&osdc->lock);
2588                 update_epoch_barrier(osdc, eb);
2589                 up_write(&osdc->lock);
2590         } else {
2591                 up_read(&osdc->lock);
2592         }
2593 }
2594 EXPORT_SYMBOL(ceph_osdc_update_epoch_barrier);
2595
2596 /*
2597  * We can end up releasing caps as a result of abort_request().
2598  * In that case, we probably want to ensure that the cap release message
2599  * has an updated epoch barrier in it, so set the epoch barrier prior to
2600  * aborting the first request.
2601  */
2602 static int abort_on_full_fn(struct ceph_osd_request *req, void *arg)
2603 {
2604         struct ceph_osd_client *osdc = req->r_osdc;
2605         bool *victims = arg;
2606
2607         if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
2608             (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
2609              pool_full(osdc, req->r_t.base_oloc.pool))) {
2610                 if (!*victims) {
2611                         update_epoch_barrier(osdc, osdc->osdmap->epoch);
2612                         *victims = true;
2613                 }
2614                 abort_request(req, -ENOSPC);
2615         }
2616
2617         return 0; /* continue iteration */
2618 }
2619
2620 /*
2621  * Drop all pending requests that are stalled waiting on a full condition to
2622  * clear, and complete them with ENOSPC as the return code. Set the
2623  * osdc->epoch_barrier to the latest map epoch that we've seen if any were
2624  * cancelled.
2625  */
2626 static void ceph_osdc_abort_on_full(struct ceph_osd_client *osdc)
2627 {
2628         bool victims = false;
2629
2630         if (ceph_test_opt(osdc->client, ABORT_ON_FULL) &&
2631             (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || have_pool_full(osdc)))
2632                 for_each_request(osdc, abort_on_full_fn, &victims);
2633 }
2634
2635 static void check_pool_dne(struct ceph_osd_request *req)
2636 {
2637         struct ceph_osd_client *osdc = req->r_osdc;
2638         struct ceph_osdmap *map = osdc->osdmap;
2639
2640         verify_osdc_wrlocked(osdc);
2641         WARN_ON(!map->epoch);
2642
2643         if (req->r_attempts) {
2644                 /*
2645                  * We sent a request earlier, which means that
2646                  * previously the pool existed, and now it does not
2647                  * (i.e., it was deleted).
2648                  */
2649                 req->r_map_dne_bound = map->epoch;
2650                 dout("%s req %p tid %llu pool disappeared\n", __func__, req,
2651                      req->r_tid);
2652         } else {
2653                 dout("%s req %p tid %llu map_dne_bound %u have %u\n", __func__,
2654                      req, req->r_tid, req->r_map_dne_bound, map->epoch);
2655         }
2656
2657         if (req->r_map_dne_bound) {
2658                 if (map->epoch >= req->r_map_dne_bound) {
2659                         /* we had a new enough map */
2660                         pr_info_ratelimited("tid %llu pool does not exist\n",
2661                                             req->r_tid);
2662                         complete_request(req, -ENOENT);
2663                 }
2664         } else {
2665                 send_map_check(req);
2666         }
2667 }
2668
2669 static void map_check_cb(struct ceph_mon_generic_request *greq)
2670 {
2671         struct ceph_osd_client *osdc = &greq->monc->client->osdc;
2672         struct ceph_osd_request *req;
2673         u64 tid = greq->private_data;
2674
2675         WARN_ON(greq->result || !greq->u.newest);
2676
2677         down_write(&osdc->lock);
2678         req = lookup_request_mc(&osdc->map_checks, tid);
2679         if (!req) {
2680                 dout("%s tid %llu dne\n", __func__, tid);
2681                 goto out_unlock;
2682         }
2683
2684         dout("%s req %p tid %llu map_dne_bound %u newest %llu\n", __func__,
2685              req, req->r_tid, req->r_map_dne_bound, greq->u.newest);
2686         if (!req->r_map_dne_bound)
2687                 req->r_map_dne_bound = greq->u.newest;
2688         erase_request_mc(&osdc->map_checks, req);
2689         check_pool_dne(req);
2690
2691         ceph_osdc_put_request(req);
2692 out_unlock:
2693         up_write(&osdc->lock);
2694 }
2695
2696 static void send_map_check(struct ceph_osd_request *req)
2697 {
2698         struct ceph_osd_client *osdc = req->r_osdc;
2699         struct ceph_osd_request *lookup_req;
2700         int ret;
2701
2702         verify_osdc_wrlocked(osdc);
2703
2704         lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid);
2705         if (lookup_req) {
2706                 WARN_ON(lookup_req != req);
2707                 return;
2708         }
2709
2710         ceph_osdc_get_request(req);
2711         insert_request_mc(&osdc->map_checks, req);
2712         ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap",
2713                                           map_check_cb, req->r_tid);
2714         WARN_ON(ret);
2715 }
2716
2717 /*
2718  * lingering requests, watch/notify v2 infrastructure
2719  */
2720 static void linger_release(struct kref *kref)
2721 {
2722         struct ceph_osd_linger_request *lreq =
2723             container_of(kref, struct ceph_osd_linger_request, kref);
2724
2725         dout("%s lreq %p reg_req %p ping_req %p\n", __func__, lreq,
2726              lreq->reg_req, lreq->ping_req);
2727         WARN_ON(!RB_EMPTY_NODE(&lreq->node));
2728         WARN_ON(!RB_EMPTY_NODE(&lreq->osdc_node));
2729         WARN_ON(!RB_EMPTY_NODE(&lreq->mc_node));
2730         WARN_ON(!list_empty(&lreq->scan_item));
2731         WARN_ON(!list_empty(&lreq->pending_lworks));
2732         WARN_ON(lreq->osd);
2733
2734         if (lreq->reg_req)
2735                 ceph_osdc_put_request(lreq->reg_req);
2736         if (lreq->ping_req)
2737                 ceph_osdc_put_request(lreq->ping_req);
2738         target_destroy(&lreq->t);
2739         kfree(lreq);
2740 }
2741
2742 static void linger_put(struct ceph_osd_linger_request *lreq)
2743 {
2744         if (lreq)
2745                 kref_put(&lreq->kref, linger_release);
2746 }
2747
2748 static struct ceph_osd_linger_request *
2749 linger_get(struct ceph_osd_linger_request *lreq)
2750 {
2751         kref_get(&lreq->kref);
2752         return lreq;
2753 }
2754
2755 static struct ceph_osd_linger_request *
2756 linger_alloc(struct ceph_osd_client *osdc)
2757 {
2758         struct ceph_osd_linger_request *lreq;
2759
2760         lreq = kzalloc(sizeof(*lreq), GFP_NOIO);
2761         if (!lreq)
2762                 return NULL;
2763
2764         kref_init(&lreq->kref);
2765         mutex_init(&lreq->lock);
2766         RB_CLEAR_NODE(&lreq->node);
2767         RB_CLEAR_NODE(&lreq->osdc_node);
2768         RB_CLEAR_NODE(&lreq->mc_node);
2769         INIT_LIST_HEAD(&lreq->scan_item);
2770         INIT_LIST_HEAD(&lreq->pending_lworks);
2771         init_completion(&lreq->reg_commit_wait);
2772         init_completion(&lreq->notify_finish_wait);
2773
2774         lreq->osdc = osdc;
2775         target_init(&lreq->t);
2776
2777         dout("%s lreq %p\n", __func__, lreq);
2778         return lreq;
2779 }
2780
2781 DEFINE_RB_INSDEL_FUNCS(linger, struct ceph_osd_linger_request, linger_id, node)
2782 DEFINE_RB_FUNCS(linger_osdc, struct ceph_osd_linger_request, linger_id, osdc_node)
2783 DEFINE_RB_FUNCS(linger_mc, struct ceph_osd_linger_request, linger_id, mc_node)
2784
2785 /*
2786  * Create linger request <-> OSD session relation.
2787  *
2788  * @lreq has to be registered, @osd may be homeless.
2789  */
2790 static void link_linger(struct ceph_osd *osd,
2791                         struct ceph_osd_linger_request *lreq)
2792 {
2793         verify_osd_locked(osd);
2794         WARN_ON(!lreq->linger_id || lreq->osd);
2795         dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd,
2796              osd->o_osd, lreq, lreq->linger_id);
2797
2798         if (!osd_homeless(osd))
2799                 __remove_osd_from_lru(osd);
2800         else
2801                 atomic_inc(&osd->o_osdc->num_homeless);
2802
2803         get_osd(osd);
2804         insert_linger(&osd->o_linger_requests, lreq);
2805         lreq->osd = osd;
2806 }
2807
2808 static void unlink_linger(struct ceph_osd *osd,
2809                           struct ceph_osd_linger_request *lreq)
2810 {
2811         verify_osd_locked(osd);
2812         WARN_ON(lreq->osd != osd);
2813         dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd,
2814              osd->o_osd, lreq, lreq->linger_id);
2815
2816         lreq->osd = NULL;
2817         erase_linger(&osd->o_linger_requests, lreq);
2818         put_osd(osd);
2819
2820         if (!osd_homeless(osd))
2821                 maybe_move_osd_to_lru(osd);
2822         else
2823                 atomic_dec(&osd->o_osdc->num_homeless);
2824 }
2825
2826 static bool __linger_registered(struct ceph_osd_linger_request *lreq)
2827 {
2828         verify_osdc_locked(lreq->osdc);
2829
2830         return !RB_EMPTY_NODE(&lreq->osdc_node);
2831 }
2832
2833 static bool linger_registered(struct ceph_osd_linger_request *lreq)
2834 {
2835         struct ceph_osd_client *osdc = lreq->osdc;
2836         bool registered;
2837
2838         down_read(&osdc->lock);
2839         registered = __linger_registered(lreq);
2840         up_read(&osdc->lock);
2841
2842         return registered;
2843 }
2844
2845 static void linger_register(struct ceph_osd_linger_request *lreq)
2846 {
2847         struct ceph_osd_client *osdc = lreq->osdc;
2848
2849         verify_osdc_wrlocked(osdc);
2850         WARN_ON(lreq->linger_id);
2851
2852         linger_get(lreq);
2853         lreq->linger_id = ++osdc->last_linger_id;
2854         insert_linger_osdc(&osdc->linger_requests, lreq);
2855 }
2856
2857 static void linger_unregister(struct ceph_osd_linger_request *lreq)
2858 {
2859         struct ceph_osd_client *osdc = lreq->osdc;
2860
2861         verify_osdc_wrlocked(osdc);
2862
2863         erase_linger_osdc(&osdc->linger_requests, lreq);
2864         linger_put(lreq);
2865 }
2866
2867 static void cancel_linger_request(struct ceph_osd_request *req)
2868 {
2869         struct ceph_osd_linger_request *lreq = req->r_priv;
2870
2871         WARN_ON(!req->r_linger);
2872         cancel_request(req);
2873         linger_put(lreq);
2874 }
2875
2876 struct linger_work {
2877         struct work_struct work;
2878         struct ceph_osd_linger_request *lreq;
2879         struct list_head pending_item;
2880         unsigned long queued_stamp;
2881
2882         union {
2883                 struct {
2884                         u64 notify_id;
2885                         u64 notifier_id;
2886                         void *payload; /* points into @msg front */
2887                         size_t payload_len;
2888
2889                         struct ceph_msg *msg; /* for ceph_msg_put() */
2890                 } notify;
2891                 struct {
2892                         int err;
2893                 } error;
2894         };
2895 };
2896
2897 static struct linger_work *lwork_alloc(struct ceph_osd_linger_request *lreq,
2898                                        work_func_t workfn)
2899 {
2900         struct linger_work *lwork;
2901
2902         lwork = kzalloc(sizeof(*lwork), GFP_NOIO);
2903         if (!lwork)
2904                 return NULL;
2905
2906         INIT_WORK(&lwork->work, workfn);
2907         INIT_LIST_HEAD(&lwork->pending_item);
2908         lwork->lreq = linger_get(lreq);
2909
2910         return lwork;
2911 }
2912
2913 static void lwork_free(struct linger_work *lwork)
2914 {
2915         struct ceph_osd_linger_request *lreq = lwork->lreq;
2916
2917         mutex_lock(&lreq->lock);
2918         list_del(&lwork->pending_item);
2919         mutex_unlock(&lreq->lock);
2920
2921         linger_put(lreq);
2922         kfree(lwork);
2923 }
2924
2925 static void lwork_queue(struct linger_work *lwork)
2926 {
2927         struct ceph_osd_linger_request *lreq = lwork->lreq;
2928         struct ceph_osd_client *osdc = lreq->osdc;
2929
2930         verify_lreq_locked(lreq);
2931         WARN_ON(!list_empty(&lwork->pending_item));
2932
2933         lwork->queued_stamp = jiffies;
2934         list_add_tail(&lwork->pending_item, &lreq->pending_lworks);
2935         queue_work(osdc->notify_wq, &lwork->work);
2936 }
2937
2938 static void do_watch_notify(struct work_struct *w)
2939 {
2940         struct linger_work *lwork = container_of(w, struct linger_work, work);
2941         struct ceph_osd_linger_request *lreq = lwork->lreq;
2942
2943         if (!linger_registered(lreq)) {
2944                 dout("%s lreq %p not registered\n", __func__, lreq);
2945                 goto out;
2946         }
2947
2948         WARN_ON(!lreq->is_watch);
2949         dout("%s lreq %p notify_id %llu notifier_id %llu payload_len %zu\n",
2950              __func__, lreq, lwork->notify.notify_id, lwork->notify.notifier_id,
2951              lwork->notify.payload_len);
2952         lreq->wcb(lreq->data, lwork->notify.notify_id, lreq->linger_id,
2953                   lwork->notify.notifier_id, lwork->notify.payload,
2954                   lwork->notify.payload_len);
2955
2956 out:
2957         ceph_msg_put(lwork->notify.msg);
2958         lwork_free(lwork);
2959 }
2960
2961 static void do_watch_error(struct work_struct *w)
2962 {
2963         struct linger_work *lwork = container_of(w, struct linger_work, work);
2964         struct ceph_osd_linger_request *lreq = lwork->lreq;
2965
2966         if (!linger_registered(lreq)) {
2967                 dout("%s lreq %p not registered\n", __func__, lreq);
2968                 goto out;
2969         }
2970
2971         dout("%s lreq %p err %d\n", __func__, lreq, lwork->error.err);
2972         lreq->errcb(lreq->data, lreq->linger_id, lwork->error.err);
2973
2974 out:
2975         lwork_free(lwork);
2976 }
2977
2978 static void queue_watch_error(struct ceph_osd_linger_request *lreq)
2979 {
2980         struct linger_work *lwork;
2981
2982         lwork = lwork_alloc(lreq, do_watch_error);
2983         if (!lwork) {
2984                 pr_err("failed to allocate error-lwork\n");
2985                 return;
2986         }
2987
2988         lwork->error.err = lreq->last_error;
2989         lwork_queue(lwork);
2990 }
2991
2992 static void linger_reg_commit_complete(struct ceph_osd_linger_request *lreq,
2993                                        int result)
2994 {
2995         if (!completion_done(&lreq->reg_commit_wait)) {
2996                 lreq->reg_commit_error = (result <= 0 ? result : 0);
2997                 complete_all(&lreq->reg_commit_wait);
2998         }
2999 }
3000
3001 static void linger_commit_cb(struct ceph_osd_request *req)
3002 {
3003         struct ceph_osd_linger_request *lreq = req->r_priv;
3004
3005         mutex_lock(&lreq->lock);
3006         dout("%s lreq %p linger_id %llu result %d\n", __func__, lreq,
3007              lreq->linger_id, req->r_result);
3008         linger_reg_commit_complete(lreq, req->r_result);
3009         lreq->committed = true;
3010
3011         if (!lreq->is_watch) {
3012                 struct ceph_osd_data *osd_data =
3013                     osd_req_op_data(req, 0, notify, response_data);
3014                 void *p = page_address(osd_data->pages[0]);
3015
3016                 WARN_ON(req->r_ops[0].op != CEPH_OSD_OP_NOTIFY ||
3017                         osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
3018
3019                 /* make note of the notify_id */
3020                 if (req->r_ops[0].outdata_len >= sizeof(u64)) {
3021                         lreq->notify_id = ceph_decode_64(&p);
3022                         dout("lreq %p notify_id %llu\n", lreq,
3023                              lreq->notify_id);
3024                 } else {
3025                         dout("lreq %p no notify_id\n", lreq);
3026                 }
3027         }
3028
3029         mutex_unlock(&lreq->lock);
3030         linger_put(lreq);
3031 }
3032
3033 static int normalize_watch_error(int err)
3034 {
3035         /*
3036          * Translate ENOENT -> ENOTCONN so that a delete->disconnection
3037          * notification and a failure to reconnect because we raced with
3038          * the delete appear the same to the user.
3039          */
3040         if (err == -ENOENT)
3041                 err = -ENOTCONN;
3042
3043         return err;
3044 }
3045
3046 static void linger_reconnect_cb(struct ceph_osd_request *req)
3047 {
3048         struct ceph_osd_linger_request *lreq = req->r_priv;
3049
3050         mutex_lock(&lreq->lock);
3051         dout("%s lreq %p linger_id %llu result %d last_error %d\n", __func__,
3052              lreq, lreq->linger_id, req->r_result, lreq->last_error);
3053         if (req->r_result < 0) {
3054                 if (!lreq->last_error) {
3055                         lreq->last_error = normalize_watch_error(req->r_result);
3056                         queue_watch_error(lreq);
3057                 }
3058         }
3059
3060         mutex_unlock(&lreq->lock);
3061         linger_put(lreq);
3062 }
3063
3064 static void send_linger(struct ceph_osd_linger_request *lreq)
3065 {
3066         struct ceph_osd_request *req = lreq->reg_req;
3067         struct ceph_osd_req_op *op = &req->r_ops[0];
3068
3069         verify_osdc_wrlocked(req->r_osdc);
3070         dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
3071
3072         if (req->r_osd)
3073                 cancel_linger_request(req);
3074
3075         request_reinit(req);
3076         target_copy(&req->r_t, &lreq->t);
3077         req->r_mtime = lreq->mtime;
3078
3079         mutex_lock(&lreq->lock);
3080         if (lreq->is_watch && lreq->committed) {
3081                 WARN_ON(op->op != CEPH_OSD_OP_WATCH ||
3082                         op->watch.cookie != lreq->linger_id);
3083                 op->watch.op = CEPH_OSD_WATCH_OP_RECONNECT;
3084                 op->watch.gen = ++lreq->register_gen;
3085                 dout("lreq %p reconnect register_gen %u\n", lreq,
3086                      op->watch.gen);
3087                 req->r_callback = linger_reconnect_cb;
3088         } else {
3089                 if (!lreq->is_watch)
3090                         lreq->notify_id = 0;
3091                 else
3092                         WARN_ON(op->watch.op != CEPH_OSD_WATCH_OP_WATCH);
3093                 dout("lreq %p register\n", lreq);
3094                 req->r_callback = linger_commit_cb;
3095         }
3096         mutex_unlock(&lreq->lock);
3097
3098         req->r_priv = linger_get(lreq);
3099         req->r_linger = true;
3100
3101         submit_request(req, true);
3102 }
3103
3104 static void linger_ping_cb(struct ceph_osd_request *req)
3105 {
3106         struct ceph_osd_linger_request *lreq = req->r_priv;
3107
3108         mutex_lock(&lreq->lock);
3109         dout("%s lreq %p linger_id %llu result %d ping_sent %lu last_error %d\n",
3110              __func__, lreq, lreq->linger_id, req->r_result, lreq->ping_sent,
3111              lreq->last_error);
3112         if (lreq->register_gen == req->r_ops[0].watch.gen) {
3113                 if (!req->r_result) {
3114                         lreq->watch_valid_thru = lreq->ping_sent;
3115                 } else if (!lreq->last_error) {
3116                         lreq->last_error = normalize_watch_error(req->r_result);
3117                         queue_watch_error(lreq);
3118                 }
3119         } else {
3120                 dout("lreq %p register_gen %u ignoring old pong %u\n", lreq,
3121                      lreq->register_gen, req->r_ops[0].watch.gen);
3122         }
3123
3124         mutex_unlock(&lreq->lock);
3125         linger_put(lreq);
3126 }
3127
3128 static void send_linger_ping(struct ceph_osd_linger_request *lreq)
3129 {
3130         struct ceph_osd_client *osdc = lreq->osdc;
3131         struct ceph_osd_request *req = lreq->ping_req;
3132         struct ceph_osd_req_op *op = &req->r_ops[0];
3133
3134         if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) {
3135                 dout("%s PAUSERD\n", __func__);
3136                 return;
3137         }
3138
3139         lreq->ping_sent = jiffies;
3140         dout("%s lreq %p linger_id %llu ping_sent %lu register_gen %u\n",
3141              __func__, lreq, lreq->linger_id, lreq->ping_sent,
3142              lreq->register_gen);
3143
3144         if (req->r_osd)
3145                 cancel_linger_request(req);
3146
3147         request_reinit(req);
3148         target_copy(&req->r_t, &lreq->t);
3149
3150         WARN_ON(op->op != CEPH_OSD_OP_WATCH ||
3151                 op->watch.cookie != lreq->linger_id ||
3152                 op->watch.op != CEPH_OSD_WATCH_OP_PING);
3153         op->watch.gen = lreq->register_gen;
3154         req->r_callback = linger_ping_cb;
3155         req->r_priv = linger_get(lreq);
3156         req->r_linger = true;
3157
3158         ceph_osdc_get_request(req);
3159         account_request(req);
3160         req->r_tid = atomic64_inc_return(&osdc->last_tid);
3161         link_request(lreq->osd, req);
3162         send_request(req);
3163 }
3164
3165 static void linger_submit(struct ceph_osd_linger_request *lreq)
3166 {
3167         struct ceph_osd_client *osdc = lreq->osdc;
3168         struct ceph_osd *osd;
3169
3170         down_write(&osdc->lock);
3171         linger_register(lreq);
3172         if (lreq->is_watch) {
3173                 lreq->reg_req->r_ops[0].watch.cookie = lreq->linger_id;
3174                 lreq->ping_req->r_ops[0].watch.cookie = lreq->linger_id;
3175         } else {
3176                 lreq->reg_req->r_ops[0].notify.cookie = lreq->linger_id;
3177         }
3178
3179         calc_target(osdc, &lreq->t, false);
3180         osd = lookup_create_osd(osdc, lreq->t.osd, true);
3181         link_linger(osd, lreq);
3182
3183         send_linger(lreq);
3184         up_write(&osdc->lock);
3185 }
3186
3187 static void cancel_linger_map_check(struct ceph_osd_linger_request *lreq)
3188 {
3189         struct ceph_osd_client *osdc = lreq->osdc;
3190         struct ceph_osd_linger_request *lookup_lreq;
3191
3192         verify_osdc_wrlocked(osdc);
3193
3194         lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks,
3195                                        lreq->linger_id);
3196         if (!lookup_lreq)
3197                 return;
3198
3199         WARN_ON(lookup_lreq != lreq);
3200         erase_linger_mc(&osdc->linger_map_checks, lreq);
3201         linger_put(lreq);
3202 }
3203
3204 /*
3205  * @lreq has to be both registered and linked.
3206  */
3207 static void __linger_cancel(struct ceph_osd_linger_request *lreq)
3208 {
3209         if (lreq->is_watch && lreq->ping_req->r_osd)
3210                 cancel_linger_request(lreq->ping_req);
3211         if (lreq->reg_req->r_osd)
3212                 cancel_linger_request(lreq->reg_req);
3213         cancel_linger_map_check(lreq);
3214         unlink_linger(lreq->osd, lreq);
3215         linger_unregister(lreq);
3216 }
3217
3218 static void linger_cancel(struct ceph_osd_linger_request *lreq)
3219 {
3220         struct ceph_osd_client *osdc = lreq->osdc;
3221
3222         down_write(&osdc->lock);
3223         if (__linger_registered(lreq))
3224                 __linger_cancel(lreq);
3225         up_write(&osdc->lock);
3226 }
3227
3228 static void send_linger_map_check(struct ceph_osd_linger_request *lreq);
3229
3230 static void check_linger_pool_dne(struct ceph_osd_linger_request *lreq)
3231 {
3232         struct ceph_osd_client *osdc = lreq->osdc;
3233         struct ceph_osdmap *map = osdc->osdmap;
3234
3235         verify_osdc_wrlocked(osdc);
3236         WARN_ON(!map->epoch);
3237
3238         if (lreq->register_gen) {
3239                 lreq->map_dne_bound = map->epoch;
3240                 dout("%s lreq %p linger_id %llu pool disappeared\n", __func__,
3241                      lreq, lreq->linger_id);
3242         } else {
3243                 dout("%s lreq %p linger_id %llu map_dne_bound %u have %u\n",
3244                      __func__, lreq, lreq->linger_id, lreq->map_dne_bound,
3245                      map->epoch);
3246         }
3247
3248         if (lreq->map_dne_bound) {
3249                 if (map->epoch >= lreq->map_dne_bound) {
3250                         /* we had a new enough map */
3251                         pr_info("linger_id %llu pool does not exist\n",
3252                                 lreq->linger_id);
3253                         linger_reg_commit_complete(lreq, -ENOENT);
3254                         __linger_cancel(lreq);
3255                 }
3256         } else {
3257                 send_linger_map_check(lreq);
3258         }
3259 }
3260
3261 static void linger_map_check_cb(struct ceph_mon_generic_request *greq)
3262 {
3263         struct ceph_osd_client *osdc = &greq->monc->client->osdc;
3264         struct ceph_osd_linger_request *lreq;
3265         u64 linger_id = greq->private_data;
3266
3267         WARN_ON(greq->result || !greq->u.newest);
3268
3269         down_write(&osdc->lock);
3270         lreq = lookup_linger_mc(&osdc->linger_map_checks, linger_id);
3271         if (!lreq) {
3272                 dout("%s linger_id %llu dne\n", __func__, linger_id);
3273                 goto out_unlock;
3274         }
3275
3276         dout("%s lreq %p linger_id %llu map_dne_bound %u newest %llu\n",
3277              __func__, lreq, lreq->linger_id, lreq->map_dne_bound,
3278              greq->u.newest);
3279         if (!lreq->map_dne_bound)
3280                 lreq->map_dne_bound = greq->u.newest;
3281         erase_linger_mc(&osdc->linger_map_checks, lreq);
3282         check_linger_pool_dne(lreq);
3283
3284         linger_put(lreq);
3285 out_unlock:
3286         up_write(&osdc->lock);
3287 }
3288
3289 static void send_linger_map_check(struct ceph_osd_linger_request *lreq)
3290 {
3291         struct ceph_osd_client *osdc = lreq->osdc;
3292         struct ceph_osd_linger_request *lookup_lreq;
3293         int ret;
3294
3295         verify_osdc_wrlocked(osdc);
3296
3297         lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks,
3298                                        lreq->linger_id);
3299         if (lookup_lreq) {
3300                 WARN_ON(lookup_lreq != lreq);
3301                 return;
3302         }
3303
3304         linger_get(lreq);
3305         insert_linger_mc(&osdc->linger_map_checks, lreq);
3306         ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap",
3307                                           linger_map_check_cb, lreq->linger_id);
3308         WARN_ON(ret);
3309 }
3310
3311 static int linger_reg_commit_wait(struct ceph_osd_linger_request *lreq)
3312 {
3313         int ret;
3314
3315         dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
3316         ret = wait_for_completion_interruptible(&lreq->reg_commit_wait);
3317         return ret ?: lreq->reg_commit_error;
3318 }
3319
3320 static int linger_notify_finish_wait(struct ceph_osd_linger_request *lreq)
3321 {
3322         int ret;
3323
3324         dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
3325         ret = wait_for_completion_interruptible(&lreq->notify_finish_wait);
3326         return ret ?: lreq->notify_finish_error;
3327 }
3328
3329 /*
3330  * Timeout callback, called every N seconds.  When 1 or more OSD
3331  * requests has been active for more than N seconds, we send a keepalive
3332  * (tag + timestamp) to its OSD to ensure any communications channel
3333  * reset is detected.
3334  */
3335 static void handle_timeout(struct work_struct *work)
3336 {
3337         struct ceph_osd_client *osdc =
3338                 container_of(work, struct ceph_osd_client, timeout_work.work);
3339         struct ceph_options *opts = osdc->client->options;
3340         unsigned long cutoff = jiffies - opts->osd_keepalive_timeout;
3341         unsigned long expiry_cutoff = jiffies - opts->osd_request_timeout;
3342         LIST_HEAD(slow_osds);
3343         struct rb_node *n, *p;
3344
3345         dout("%s osdc %p\n", __func__, osdc);
3346         down_write(&osdc->lock);
3347
3348         /*
3349          * ping osds that are a bit slow.  this ensures that if there
3350          * is a break in the TCP connection we will notice, and reopen
3351          * a connection with that osd (from the fault callback).
3352          */
3353         for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
3354                 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
3355                 bool found = false;
3356
3357                 for (p = rb_first(&osd->o_requests); p; ) {
3358                         struct ceph_osd_request *req =
3359                             rb_entry(p, struct ceph_osd_request, r_node);
3360
3361                         p = rb_next(p); /* abort_request() */
3362
3363                         if (time_before(req->r_stamp, cutoff)) {
3364                                 dout(" req %p tid %llu on osd%d is laggy\n",
3365                                      req, req->r_tid, osd->o_osd);
3366                                 found = true;
3367                         }
3368                         if (opts->osd_request_timeout &&
3369                             time_before(req->r_start_stamp, expiry_cutoff)) {
3370                                 pr_err_ratelimited("tid %llu on osd%d timeout\n",
3371                                        req->r_tid, osd->o_osd);
3372                                 abort_request(req, -ETIMEDOUT);
3373                         }
3374                 }
3375                 for (p = rb_first(&osd->o_linger_requests); p; p = rb_next(p)) {
3376                         struct ceph_osd_linger_request *lreq =
3377                             rb_entry(p, struct ceph_osd_linger_request, node);
3378
3379                         dout(" lreq %p linger_id %llu is served by osd%d\n",
3380                              lreq, lreq->linger_id, osd->o_osd);
3381                         found = true;
3382
3383                         mutex_lock(&lreq->lock);
3384                         if (lreq->is_watch && lreq->committed && !lreq->last_error)
3385                                 send_linger_ping(lreq);
3386                         mutex_unlock(&lreq->lock);
3387                 }
3388
3389                 if (found)
3390                         list_move_tail(&osd->o_keepalive_item, &slow_osds);
3391         }
3392
3393         if (opts->osd_request_timeout) {
3394                 for (p = rb_first(&osdc->homeless_osd.o_requests); p; ) {
3395                         struct ceph_osd_request *req =
3396                             rb_entry(p, struct ceph_osd_request, r_node);
3397
3398                         p = rb_next(p); /* abort_request() */
3399
3400                         if (time_before(req->r_start_stamp, expiry_cutoff)) {
3401                                 pr_err_ratelimited("tid %llu on osd%d timeout\n",
3402                                        req->r_tid, osdc->homeless_osd.o_osd);
3403                                 abort_request(req, -ETIMEDOUT);
3404                         }
3405                 }
3406         }
3407
3408         if (atomic_read(&osdc->num_homeless) || !list_empty(&slow_osds))
3409                 maybe_request_map(osdc);
3410
3411         while (!list_empty(&slow_osds)) {
3412                 struct ceph_osd *osd = list_first_entry(&slow_osds,
3413                                                         struct ceph_osd,
3414                                                         o_keepalive_item);
3415                 list_del_init(&osd->o_keepalive_item);
3416                 ceph_con_keepalive(&osd->o_con);
3417         }
3418
3419         up_write(&osdc->lock);
3420         schedule_delayed_work(&osdc->timeout_work,
3421                               osdc->client->options->osd_keepalive_timeout);
3422 }
3423
3424 static void handle_osds_timeout(struct work_struct *work)
3425 {
3426         struct ceph_osd_client *osdc =
3427                 container_of(work, struct ceph_osd_client,
3428                              osds_timeout_work.work);
3429         unsigned long delay = osdc->client->options->osd_idle_ttl / 4;
3430         struct ceph_osd *osd, *nosd;
3431
3432         dout("%s osdc %p\n", __func__, osdc);
3433         down_write(&osdc->lock);
3434         list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
3435                 if (time_before(jiffies, osd->lru_ttl))
3436                         break;
3437
3438                 WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests));
3439                 WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests));
3440                 close_osd(osd);
3441         }
3442
3443         up_write(&osdc->lock);
3444         schedule_delayed_work(&osdc->osds_timeout_work,
3445                               round_jiffies_relative(delay));
3446 }
3447
3448 static int ceph_oloc_decode(void **p, void *end,
3449                             struct ceph_object_locator *oloc)
3450 {
3451         u8 struct_v, struct_cv;
3452         u32 len;
3453         void *struct_end;
3454         int ret = 0;
3455
3456         ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
3457         struct_v = ceph_decode_8(p);
3458         struct_cv = ceph_decode_8(p);
3459         if (struct_v < 3) {
3460                 pr_warn("got v %d < 3 cv %d of ceph_object_locator\n",
3461                         struct_v, struct_cv);
3462                 goto e_inval;
3463         }
3464         if (struct_cv > 6) {
3465                 pr_warn("got v %d cv %d > 6 of ceph_object_locator\n",
3466                         struct_v, struct_cv);
3467                 goto e_inval;
3468         }
3469         len = ceph_decode_32(p);
3470         ceph_decode_need(p, end, len, e_inval);
3471         struct_end = *p + len;
3472
3473         oloc->pool = ceph_decode_64(p);
3474         *p += 4; /* skip preferred */
3475
3476         len = ceph_decode_32(p);
3477         if (len > 0) {
3478                 pr_warn("ceph_object_locator::key is set\n");
3479                 goto e_inval;
3480         }
3481
3482         if (struct_v >= 5) {
3483                 bool changed = false;
3484
3485                 len = ceph_decode_32(p);
3486                 if (len > 0) {
3487                         ceph_decode_need(p, end, len, e_inval);
3488                         if (!oloc->pool_ns ||
3489                             ceph_compare_string(oloc->pool_ns, *p, len))
3490                                 changed = true;
3491                         *p += len;
3492                 } else {
3493                         if (oloc->pool_ns)
3494                                 changed = true;
3495                 }
3496                 if (changed) {
3497                         /* redirect changes namespace */
3498                         pr_warn("ceph_object_locator::nspace is changed\n");
3499                         goto e_inval;
3500                 }
3501         }
3502
3503         if (struct_v >= 6) {
3504                 s64 hash = ceph_decode_64(p);
3505                 if (hash != -1) {
3506                         pr_warn("ceph_object_locator::hash is set\n");
3507                         goto e_inval;
3508                 }
3509         }
3510
3511         /* skip the rest */
3512         *p = struct_end;
3513 out:
3514         return ret;
3515
3516 e_inval:
3517         ret = -EINVAL;
3518         goto out;
3519 }
3520
3521 static int ceph_redirect_decode(void **p, void *end,
3522                                 struct ceph_request_redirect *redir)
3523 {
3524         u8 struct_v, struct_cv;
3525         u32 len;
3526         void *struct_end;
3527         int ret;
3528
3529         ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
3530         struct_v = ceph_decode_8(p);
3531         struct_cv = ceph_decode_8(p);
3532         if (struct_cv > 1) {
3533                 pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n",
3534                         struct_v, struct_cv);
3535                 goto e_inval;
3536         }
3537         len = ceph_decode_32(p);
3538         ceph_decode_need(p, end, len, e_inval);
3539         struct_end = *p + len;
3540
3541         ret = ceph_oloc_decode(p, end, &redir->oloc);
3542         if (ret)
3543                 goto out;
3544
3545         len = ceph_decode_32(p);
3546         if (len > 0) {
3547                 pr_warn("ceph_request_redirect::object_name is set\n");
3548                 goto e_inval;
3549         }
3550
3551         /* skip the rest */
3552         *p = struct_end;
3553 out:
3554         return ret;
3555
3556 e_inval:
3557         ret = -EINVAL;
3558         goto out;
3559 }
3560
3561 struct MOSDOpReply {
3562         struct ceph_pg pgid;
3563         u64 flags;
3564         int result;
3565         u32 epoch;
3566         int num_ops;
3567         u32 outdata_len[CEPH_OSD_MAX_OPS];
3568         s32 rval[CEPH_OSD_MAX_OPS];
3569         int retry_attempt;
3570         struct ceph_eversion replay_version;
3571         u64 user_version;
3572         struct ceph_request_redirect redirect;
3573 };
3574
3575 static int decode_MOSDOpReply(const struct ceph_msg *msg, struct MOSDOpReply *m)
3576 {
3577         void *p = msg->front.iov_base;
3578         void *const end = p + msg->front.iov_len;
3579         u16 version = le16_to_cpu(msg->hdr.version);
3580         struct ceph_eversion bad_replay_version;
3581         u8 decode_redir;
3582         u32 len;
3583         int ret;
3584         int i;
3585
3586         ceph_decode_32_safe(&p, end, len, e_inval);
3587         ceph_decode_need(&p, end, len, e_inval);
3588         p += len; /* skip oid */
3589
3590         ret = ceph_decode_pgid(&p, end, &m->pgid);
3591         if (ret)
3592                 return ret;
3593
3594         ceph_decode_64_safe(&p, end, m->flags, e_inval);
3595         ceph_decode_32_safe(&p, end, m->result, e_inval);
3596         ceph_decode_need(&p, end, sizeof(bad_replay_version), e_inval);
3597         memcpy(&bad_replay_version, p, sizeof(bad_replay_version));
3598         p += sizeof(bad_replay_version);
3599         ceph_decode_32_safe(&p, end, m->epoch, e_inval);
3600
3601         ceph_decode_32_safe(&p, end, m->num_ops, e_inval);
3602         if (m->num_ops > ARRAY_SIZE(m->outdata_len))
3603                 goto e_inval;
3604
3605         ceph_decode_need(&p, end, m->num_ops * sizeof(struct ceph_osd_op),
3606                          e_inval);
3607         for (i = 0; i < m->num_ops; i++) {
3608                 struct ceph_osd_op *op = p;
3609
3610                 m->outdata_len[i] = le32_to_cpu(op->payload_len);
3611                 p += sizeof(*op);
3612         }
3613
3614         ceph_decode_32_safe(&p, end, m->retry_attempt, e_inval);
3615         for (i = 0; i < m->num_ops; i++)
3616                 ceph_decode_32_safe(&p, end, m->rval[i], e_inval);
3617
3618         if (version >= 5) {
3619                 ceph_decode_need(&p, end, sizeof(m->replay_version), e_inval);
3620                 memcpy(&m->replay_version, p, sizeof(m->replay_version));
3621                 p += sizeof(m->replay_version);
3622                 ceph_decode_64_safe(&p, end, m->user_version, e_inval);
3623         } else {
3624                 m->replay_version = bad_replay_version; /* struct */
3625                 m->user_version = le64_to_cpu(m->replay_version.version);
3626         }
3627
3628         if (version >= 6) {
3629                 if (version >= 7)
3630                         ceph_decode_8_safe(&p, end, decode_redir, e_inval);
3631                 else
3632                         decode_redir = 1;
3633         } else {
3634                 decode_redir = 0;
3635         }
3636
3637         if (decode_redir) {
3638                 ret = ceph_redirect_decode(&p, end, &m->redirect);
3639                 if (ret)
3640                         return ret;
3641         } else {
3642                 ceph_oloc_init(&m->redirect.oloc);
3643         }
3644
3645         return 0;
3646
3647 e_inval:
3648         return -EINVAL;
3649 }
3650
3651 /*
3652  * Handle MOSDOpReply.  Set ->r_result and call the callback if it is
3653  * specified.
3654  */
3655 static void handle_reply(struct ceph_osd *osd, struct ceph_msg *msg)
3656 {
3657         struct ceph_osd_client *osdc = osd->o_osdc;
3658         struct ceph_osd_request *req;
3659         struct MOSDOpReply m;
3660         u64 tid = le64_to_cpu(msg->hdr.tid);
3661         u32 data_len = 0;
3662         int ret;
3663         int i;
3664
3665         dout("%s msg %p tid %llu\n", __func__, msg, tid);
3666
3667         down_read(&osdc->lock);
3668         if (!osd_registered(osd)) {
3669                 dout("%s osd%d unknown\n", __func__, osd->o_osd);
3670                 goto out_unlock_osdc;
3671         }
3672         WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num));
3673
3674         mutex_lock(&osd->lock);
3675         req = lookup_request(&osd->o_requests, tid);
3676         if (!req) {
3677                 dout("%s osd%d tid %llu unknown\n", __func__, osd->o_osd, tid);
3678                 goto out_unlock_session;
3679         }
3680
3681         m.redirect.oloc.pool_ns = req->r_t.target_oloc.pool_ns;
3682         ret = decode_MOSDOpReply(msg, &m);
3683         m.redirect.oloc.pool_ns = NULL;
3684         if (ret) {
3685                 pr_err("failed to decode MOSDOpReply for tid %llu: %d\n",
3686                        req->r_tid, ret);
3687                 ceph_msg_dump(msg);
3688                 goto fail_request;
3689         }
3690         dout("%s req %p tid %llu flags 0x%llx pgid %llu.%x epoch %u attempt %d v %u'%llu uv %llu\n",
3691              __func__, req, req->r_tid, m.flags, m.pgid.pool, m.pgid.seed,
3692              m.epoch, m.retry_attempt, le32_to_cpu(m.replay_version.epoch),
3693              le64_to_cpu(m.replay_version.version), m.user_version);
3694
3695         if (m.retry_attempt >= 0) {
3696                 if (m.retry_attempt != req->r_attempts - 1) {
3697                         dout("req %p tid %llu retry_attempt %d != %d, ignoring\n",
3698                              req, req->r_tid, m.retry_attempt,
3699                              req->r_attempts - 1);
3700                         goto out_unlock_session;
3701                 }
3702         } else {
3703                 WARN_ON(1); /* MOSDOpReply v4 is assumed */
3704         }
3705
3706         if (!ceph_oloc_empty(&m.redirect.oloc)) {
3707                 dout("req %p tid %llu redirect pool %lld\n", req, req->r_tid,
3708                      m.redirect.oloc.pool);
3709                 unlink_request(osd, req);
3710                 mutex_unlock(&osd->lock);
3711
3712                 /*
3713                  * Not ceph_oloc_copy() - changing pool_ns is not
3714                  * supported.
3715                  */
3716                 req->r_t.target_oloc.pool = m.redirect.oloc.pool;
3717                 req->r_flags |= CEPH_OSD_FLAG_REDIRECTED |
3718                                 CEPH_OSD_FLAG_IGNORE_OVERLAY |
3719                                 CEPH_OSD_FLAG_IGNORE_CACHE;
3720                 req->r_tid = 0;
3721                 __submit_request(req, false);
3722                 goto out_unlock_osdc;
3723         }
3724
3725         if (m.result == -EAGAIN) {
3726                 dout("req %p tid %llu EAGAIN\n", req, req->r_tid);
3727                 unlink_request(osd, req);
3728                 mutex_unlock(&osd->lock);
3729
3730                 /*
3731                  * The object is missing on the replica or not (yet)
3732                  * readable.  Clear pgid to force a resend to the primary
3733                  * via legacy_change.
3734                  */
3735                 req->r_t.pgid.pool = 0;
3736                 req->r_t.pgid.seed = 0;
3737                 WARN_ON(!req->r_t.used_replica);
3738                 req->r_flags &= ~(CEPH_OSD_FLAG_BALANCE_READS |
3739                                   CEPH_OSD_FLAG_LOCALIZE_READS);
3740                 req->r_tid = 0;
3741                 __submit_request(req, false);
3742                 goto out_unlock_osdc;
3743         }
3744
3745         if (m.num_ops != req->r_num_ops) {
3746                 pr_err("num_ops %d != %d for tid %llu\n", m.num_ops,
3747                        req->r_num_ops, req->r_tid);
3748                 goto fail_request;
3749         }
3750         for (i = 0; i < req->r_num_ops; i++) {
3751                 dout(" req %p tid %llu op %d rval %d len %u\n", req,
3752                      req->r_tid, i, m.rval[i], m.outdata_len[i]);
3753                 req->r_ops[i].rval = m.rval[i];
3754                 req->r_ops[i].outdata_len = m.outdata_len[i];
3755                 data_len += m.outdata_len[i];
3756         }
3757         if (data_len != le32_to_cpu(msg->hdr.data_len)) {
3758                 pr_err("sum of lens %u != %u for tid %llu\n", data_len,
3759                        le32_to_cpu(msg->hdr.data_len), req->r_tid);
3760                 goto fail_request;
3761         }
3762         dout("%s req %p tid %llu result %d data_len %u\n", __func__,
3763              req, req->r_tid, m.result, data_len);
3764
3765         /*
3766          * Since we only ever request ONDISK, we should only ever get
3767          * one (type of) reply back.
3768          */
3769         WARN_ON(!(m.flags & CEPH_OSD_FLAG_ONDISK));
3770         req->r_result = m.result ?: data_len;
3771         finish_request(req);
3772         mutex_unlock(&osd->lock);
3773         up_read(&osdc->lock);
3774
3775         __complete_request(req);
3776         return;
3777
3778 fail_request:
3779         complete_request(req, -EIO);
3780 out_unlock_session:
3781         mutex_unlock(&osd->lock);
3782 out_unlock_osdc:
3783         up_read(&osdc->lock);
3784 }
3785
3786 static void set_pool_was_full(struct ceph_osd_client *osdc)
3787 {
3788         struct rb_node *n;
3789
3790         for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) {
3791                 struct ceph_pg_pool_info *pi =
3792                     rb_entry(n, struct ceph_pg_pool_info, node);
3793
3794                 pi->was_full = __pool_full(pi);
3795         }
3796 }
3797
3798 static bool pool_cleared_full(struct ceph_osd_client *osdc, s64 pool_id)
3799 {
3800         struct ceph_pg_pool_info *pi;
3801
3802         pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id);
3803         if (!pi)
3804                 return false;
3805
3806         return pi->was_full && !__pool_full(pi);
3807 }
3808
3809 static enum calc_target_result
3810 recalc_linger_target(struct ceph_osd_linger_request *lreq)
3811 {
3812         struct ceph_osd_client *osdc = lreq->osdc;
3813         enum calc_target_result ct_res;
3814
3815         ct_res = calc_target(osdc, &lreq->t, true);
3816         if (ct_res == CALC_TARGET_NEED_RESEND) {
3817                 struct ceph_osd *osd;
3818
3819                 osd = lookup_create_osd(osdc, lreq->t.osd, true);
3820                 if (osd != lreq->osd) {
3821                         unlink_linger(lreq->osd, lreq);
3822                         link_linger(osd, lreq);
3823                 }
3824         }
3825
3826         return ct_res;
3827 }
3828
3829 /*
3830  * Requeue requests whose mapping to an OSD has changed.
3831  */
3832 static void scan_requests(struct ceph_osd *osd,
3833                           bool force_resend,
3834                           bool cleared_full,
3835                           bool check_pool_cleared_full,
3836                           struct rb_root *need_resend,
3837                           struct list_head *need_resend_linger)
3838 {
3839         struct ceph_osd_client *osdc = osd->o_osdc;
3840         struct rb_node *n;
3841         bool force_resend_writes;
3842
3843         for (n = rb_first(&osd->o_linger_requests); n; ) {
3844                 struct ceph_osd_linger_request *lreq =
3845                     rb_entry(n, struct ceph_osd_linger_request, node);
3846                 enum calc_target_result ct_res;
3847
3848                 n = rb_next(n); /* recalc_linger_target() */
3849
3850                 dout("%s lreq %p linger_id %llu\n", __func__, lreq,
3851                      lreq->linger_id);
3852                 ct_res = recalc_linger_target(lreq);
3853                 switch (ct_res) {
3854                 case CALC_TARGET_NO_ACTION:
3855                         force_resend_writes = cleared_full ||
3856                             (check_pool_cleared_full &&
3857                              pool_cleared_full(osdc, lreq->t.base_oloc.pool));
3858                         if (!force_resend && !force_resend_writes)
3859                                 break;
3860
3861                         fallthrough;
3862                 case CALC_TARGET_NEED_RESEND:
3863                         cancel_linger_map_check(lreq);
3864                         /*
3865                          * scan_requests() for the previous epoch(s)
3866                          * may have already added it to the list, since
3867                          * it's not unlinked here.
3868                          */
3869                         if (list_empty(&lreq->scan_item))
3870                                 list_add_tail(&lreq->scan_item, need_resend_linger);
3871                         break;
3872                 case CALC_TARGET_POOL_DNE:
3873                         list_del_init(&lreq->scan_item);
3874                         check_linger_pool_dne(lreq);
3875                         break;
3876                 }
3877         }
3878
3879         for (n = rb_first(&osd->o_requests); n; ) {
3880                 struct ceph_osd_request *req =
3881                     rb_entry(n, struct ceph_osd_request, r_node);
3882                 enum calc_target_result ct_res;
3883
3884                 n = rb_next(n); /* unlink_request(), check_pool_dne() */
3885
3886                 dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
3887                 ct_res = calc_target(osdc, &req->r_t, false);
3888                 switch (ct_res) {
3889                 case CALC_TARGET_NO_ACTION:
3890                         force_resend_writes = cleared_full ||
3891                             (check_pool_cleared_full &&
3892                              pool_cleared_full(osdc, req->r_t.base_oloc.pool));
3893                         if (!force_resend &&
3894                             (!(req->r_flags & CEPH_OSD_FLAG_WRITE) ||
3895                              !force_resend_writes))
3896                                 break;
3897
3898                         fallthrough;
3899                 case CALC_TARGET_NEED_RESEND:
3900                         cancel_map_check(req);
3901                         unlink_request(osd, req);
3902                         insert_request(need_resend, req);
3903                         break;
3904                 case CALC_TARGET_POOL_DNE:
3905                         check_pool_dne(req);
3906                         break;
3907                 }
3908         }
3909 }
3910
3911 static int handle_one_map(struct ceph_osd_client *osdc,
3912                           void *p, void *end, bool incremental,
3913                           struct rb_root *need_resend,
3914                           struct list_head *need_resend_linger)
3915 {
3916         struct ceph_osdmap *newmap;
3917         struct rb_node *n;
3918         bool skipped_map = false;
3919         bool was_full;
3920
3921         was_full = ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL);
3922         set_pool_was_full(osdc);
3923
3924         if (incremental)
3925                 newmap = osdmap_apply_incremental(&p, end,
3926                                                   ceph_msgr2(osdc->client),
3927                                                   osdc->osdmap);
3928         else
3929                 newmap = ceph_osdmap_decode(&p, end, ceph_msgr2(osdc->client));
3930         if (IS_ERR(newmap))
3931                 return PTR_ERR(newmap);
3932
3933         if (newmap != osdc->osdmap) {
3934                 /*
3935                  * Preserve ->was_full before destroying the old map.
3936                  * For pools that weren't in the old map, ->was_full
3937                  * should be false.
3938                  */
3939                 for (n = rb_first(&newmap->pg_pools); n; n = rb_next(n)) {
3940                         struct ceph_pg_pool_info *pi =
3941                             rb_entry(n, struct ceph_pg_pool_info, node);
3942                         struct ceph_pg_pool_info *old_pi;
3943
3944                         old_pi = ceph_pg_pool_by_id(osdc->osdmap, pi->id);
3945                         if (old_pi)
3946                                 pi->was_full = old_pi->was_full;
3947                         else
3948                                 WARN_ON(pi->was_full);
3949                 }
3950
3951                 if (osdc->osdmap->epoch &&
3952                     osdc->osdmap->epoch + 1 < newmap->epoch) {
3953                         WARN_ON(incremental);
3954                         skipped_map = true;
3955                 }
3956
3957                 ceph_osdmap_destroy(osdc->osdmap);
3958                 osdc->osdmap = newmap;
3959         }
3960
3961         was_full &= !ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL);
3962         scan_requests(&osdc->homeless_osd, skipped_map, was_full, true,
3963                       need_resend, need_resend_linger);
3964
3965         for (n = rb_first(&osdc->osds); n; ) {
3966                 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
3967
3968                 n = rb_next(n); /* close_osd() */
3969
3970                 scan_requests(osd, skipped_map, was_full, true, need_resend,
3971                               need_resend_linger);
3972                 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
3973                     memcmp(&osd->o_con.peer_addr,
3974                            ceph_osd_addr(osdc->osdmap, osd->o_osd),
3975                            sizeof(struct ceph_entity_addr)))
3976                         close_osd(osd);
3977         }
3978
3979         return 0;
3980 }
3981
3982 static void kick_requests(struct ceph_osd_client *osdc,
3983                           struct rb_root *need_resend,
3984                           struct list_head *need_resend_linger)
3985 {
3986         struct ceph_osd_linger_request *lreq, *nlreq;
3987         enum calc_target_result ct_res;
3988         struct rb_node *n;
3989
3990         /* make sure need_resend targets reflect latest map */
3991         for (n = rb_first(need_resend); n; ) {
3992                 struct ceph_osd_request *req =
3993                     rb_entry(n, struct ceph_osd_request, r_node);
3994
3995                 n = rb_next(n);
3996
3997                 if (req->r_t.epoch < osdc->osdmap->epoch) {
3998                         ct_res = calc_target(osdc, &req->r_t, false);
3999                         if (ct_res == CALC_TARGET_POOL_DNE) {
4000                                 erase_request(need_resend, req);
4001                                 check_pool_dne(req);
4002                         }
4003                 }
4004         }
4005
4006         for (n = rb_first(need_resend); n; ) {
4007                 struct ceph_osd_request *req =
4008                     rb_entry(n, struct ceph_osd_request, r_node);
4009                 struct ceph_osd *osd;
4010
4011                 n = rb_next(n);
4012                 erase_request(need_resend, req); /* before link_request() */
4013
4014                 osd = lookup_create_osd(osdc, req->r_t.osd, true);
4015                 link_request(osd, req);
4016                 if (!req->r_linger) {
4017                         if (!osd_homeless(osd) && !req->r_t.paused)
4018                                 send_request(req);
4019                 } else {
4020                         cancel_linger_request(req);
4021                 }
4022         }
4023
4024         list_for_each_entry_safe(lreq, nlreq, need_resend_linger, scan_item) {
4025                 if (!osd_homeless(lreq->osd))
4026                         send_linger(lreq);
4027
4028                 list_del_init(&lreq->scan_item);
4029         }
4030 }
4031
4032 /*
4033  * Process updated osd map.
4034  *
4035  * The message contains any number of incremental and full maps, normally
4036  * indicating some sort of topology change in the cluster.  Kick requests
4037  * off to different OSDs as needed.
4038  */
4039 void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
4040 {
4041         void *p = msg->front.iov_base;
4042         void *const end = p + msg->front.iov_len;
4043         u32 nr_maps, maplen;
4044         u32 epoch;
4045         struct ceph_fsid fsid;
4046         struct rb_root need_resend = RB_ROOT;
4047         LIST_HEAD(need_resend_linger);
4048         bool handled_incremental = false;
4049         bool was_pauserd, was_pausewr;
4050         bool pauserd, pausewr;
4051         int err;
4052
4053         dout("%s have %u\n", __func__, osdc->osdmap->epoch);
4054         down_write(&osdc->lock);
4055
4056         /* verify fsid */
4057         ceph_decode_need(&p, end, sizeof(fsid), bad);
4058         ceph_decode_copy(&p, &fsid, sizeof(fsid));
4059         if (ceph_check_fsid(osdc->client, &fsid) < 0)
4060                 goto bad;
4061
4062         was_pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
4063         was_pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
4064                       ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
4065                       have_pool_full(osdc);
4066
4067         /* incremental maps */
4068         ceph_decode_32_safe(&p, end, nr_maps, bad);
4069         dout(" %d inc maps\n", nr_maps);
4070         while (nr_maps > 0) {
4071                 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
4072                 epoch = ceph_decode_32(&p);
4073                 maplen = ceph_decode_32(&p);
4074                 ceph_decode_need(&p, end, maplen, bad);
4075                 if (osdc->osdmap->epoch &&
4076                     osdc->osdmap->epoch + 1 == epoch) {
4077                         dout("applying incremental map %u len %d\n",
4078                              epoch, maplen);
4079                         err = handle_one_map(osdc, p, p + maplen, true,
4080                                              &need_resend, &need_resend_linger);
4081                         if (err)
4082                                 goto bad;
4083                         handled_incremental = true;
4084                 } else {
4085                         dout("ignoring incremental map %u len %d\n",
4086                              epoch, maplen);
4087                 }
4088                 p += maplen;
4089                 nr_maps--;
4090         }
4091         if (handled_incremental)
4092                 goto done;
4093
4094         /* full maps */
4095         ceph_decode_32_safe(&p, end, nr_maps, bad);
4096         dout(" %d full maps\n", nr_maps);
4097         while (nr_maps) {
4098                 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
4099                 epoch = ceph_decode_32(&p);
4100                 maplen = ceph_decode_32(&p);
4101                 ceph_decode_need(&p, end, maplen, bad);
4102                 if (nr_maps > 1) {
4103                         dout("skipping non-latest full map %u len %d\n",
4104                              epoch, maplen);
4105                 } else if (osdc->osdmap->epoch >= epoch) {
4106                         dout("skipping full map %u len %d, "
4107                              "older than our %u\n", epoch, maplen,
4108                              osdc->osdmap->epoch);
4109                 } else {
4110                         dout("taking full map %u len %d\n", epoch, maplen);
4111                         err = handle_one_map(osdc, p, p + maplen, false,
4112                                              &need_resend, &need_resend_linger);
4113                         if (err)
4114                                 goto bad;
4115                 }
4116                 p += maplen;
4117                 nr_maps--;
4118         }
4119
4120 done:
4121         /*
4122          * subscribe to subsequent osdmap updates if full to ensure
4123          * we find out when we are no longer full and stop returning
4124          * ENOSPC.
4125          */
4126         pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
4127         pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
4128                   ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
4129                   have_pool_full(osdc);
4130         if (was_pauserd || was_pausewr || pauserd || pausewr ||
4131             osdc->osdmap->epoch < osdc->epoch_barrier)
4132                 maybe_request_map(osdc);
4133
4134         kick_requests(osdc, &need_resend, &need_resend_linger);
4135
4136         ceph_osdc_abort_on_full(osdc);
4137         ceph_monc_got_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
4138                           osdc->osdmap->epoch);
4139         up_write(&osdc->lock);
4140         wake_up_all(&osdc->client->auth_wq);
4141         return;
4142
4143 bad:
4144         pr_err("osdc handle_map corrupt msg\n");
4145         ceph_msg_dump(msg);
4146         up_write(&osdc->lock);
4147 }
4148
4149 /*
4150  * Resubmit requests pending on the given osd.
4151  */
4152 static void kick_osd_requests(struct ceph_osd *osd)
4153 {
4154         struct rb_node *n;
4155
4156         clear_backoffs(osd);
4157
4158         for (n = rb_first(&osd->o_requests); n; ) {
4159                 struct ceph_osd_request *req =
4160                     rb_entry(n, struct ceph_osd_request, r_node);
4161
4162                 n = rb_next(n); /* cancel_linger_request() */
4163
4164                 if (!req->r_linger) {
4165                         if (!req->r_t.paused)
4166                                 send_request(req);
4167                 } else {
4168                         cancel_linger_request(req);
4169                 }
4170         }
4171         for (n = rb_first(&osd->o_linger_requests); n; n = rb_next(n)) {
4172                 struct ceph_osd_linger_request *lreq =
4173                     rb_entry(n, struct ceph_osd_linger_request, node);
4174
4175                 send_linger(lreq);
4176         }
4177 }
4178
4179 /*
4180  * If the osd connection drops, we need to resubmit all requests.
4181  */
4182 static void osd_fault(struct ceph_connection *con)
4183 {
4184         struct ceph_osd *osd = con->private;
4185         struct ceph_osd_client *osdc = osd->o_osdc;
4186
4187         dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
4188
4189         down_write(&osdc->lock);
4190         if (!osd_registered(osd)) {
4191                 dout("%s osd%d unknown\n", __func__, osd->o_osd);
4192                 goto out_unlock;
4193         }
4194
4195         if (!reopen_osd(osd))
4196                 kick_osd_requests(osd);
4197         maybe_request_map(osdc);
4198
4199 out_unlock:
4200         up_write(&osdc->lock);
4201 }
4202
4203 struct MOSDBackoff {
4204         struct ceph_spg spgid;
4205         u32 map_epoch;
4206         u8 op;
4207         u64 id;
4208         struct ceph_hobject_id *begin;
4209         struct ceph_hobject_id *end;
4210 };
4211
4212 static int decode_MOSDBackoff(const struct ceph_msg *msg, struct MOSDBackoff *m)
4213 {
4214         void *p = msg->front.iov_base;
4215         void *const end = p + msg->front.iov_len;
4216         u8 struct_v;
4217         u32 struct_len;
4218         int ret;
4219
4220         ret = ceph_start_decoding(&p, end, 1, "spg_t", &struct_v, &struct_len);
4221         if (ret)
4222                 return ret;
4223
4224         ret = ceph_decode_pgid(&p, end, &m->spgid.pgid);
4225         if (ret)
4226                 return ret;
4227
4228         ceph_decode_8_safe(&p, end, m->spgid.shard, e_inval);
4229         ceph_decode_32_safe(&p, end, m->map_epoch, e_inval);
4230         ceph_decode_8_safe(&p, end, m->op, e_inval);
4231         ceph_decode_64_safe(&p, end, m->id, e_inval);
4232
4233         m->begin = kzalloc(sizeof(*m->begin), GFP_NOIO);
4234         if (!m->begin)
4235                 return -ENOMEM;
4236
4237         ret = decode_hoid(&p, end, m->begin);
4238         if (ret) {
4239                 free_hoid(m->begin);
4240                 return ret;
4241         }
4242
4243         m->end = kzalloc(sizeof(*m->end), GFP_NOIO);
4244         if (!m->end) {
4245                 free_hoid(m->begin);
4246                 return -ENOMEM;
4247         }
4248
4249         ret = decode_hoid(&p, end, m->end);
4250         if (ret) {
4251                 free_hoid(m->begin);
4252                 free_hoid(m->end);
4253                 return ret;
4254         }
4255
4256         return 0;
4257
4258 e_inval:
4259         return -EINVAL;
4260 }
4261
4262 static struct ceph_msg *create_backoff_message(
4263                                 const struct ceph_osd_backoff *backoff,
4264                                 u32 map_epoch)
4265 {
4266         struct ceph_msg *msg;
4267         void *p, *end;
4268         int msg_size;
4269
4270         msg_size = CEPH_ENCODING_START_BLK_LEN +
4271                         CEPH_PGID_ENCODING_LEN + 1; /* spgid */
4272         msg_size += 4 + 1 + 8; /* map_epoch, op, id */
4273         msg_size += CEPH_ENCODING_START_BLK_LEN +
4274                         hoid_encoding_size(backoff->begin);
4275         msg_size += CEPH_ENCODING_START_BLK_LEN +
4276                         hoid_encoding_size(backoff->end);
4277
4278         msg = ceph_msg_new(CEPH_MSG_OSD_BACKOFF, msg_size, GFP_NOIO, true);
4279         if (!msg)
4280                 return NULL;
4281
4282         p = msg->front.iov_base;
4283         end = p + msg->front_alloc_len;
4284
4285         encode_spgid(&p, &backoff->spgid);
4286         ceph_encode_32(&p, map_epoch);
4287         ceph_encode_8(&p, CEPH_OSD_BACKOFF_OP_ACK_BLOCK);
4288         ceph_encode_64(&p, backoff->id);
4289         encode_hoid(&p, end, backoff->begin);
4290         encode_hoid(&p, end, backoff->end);
4291         BUG_ON(p != end);
4292
4293         msg->front.iov_len = p - msg->front.iov_base;
4294         msg->hdr.version = cpu_to_le16(1); /* MOSDBackoff v1 */
4295         msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
4296
4297         return msg;
4298 }
4299
4300 static void handle_backoff_block(struct ceph_osd *osd, struct MOSDBackoff *m)
4301 {
4302         struct ceph_spg_mapping *spg;
4303         struct ceph_osd_backoff *backoff;
4304         struct ceph_msg *msg;
4305
4306         dout("%s osd%d spgid %llu.%xs%d id %llu\n", __func__, osd->o_osd,
4307              m->spgid.pgid.pool, m->spgid.pgid.seed, m->spgid.shard, m->id);
4308
4309         spg = lookup_spg_mapping(&osd->o_backoff_mappings, &m->spgid);
4310         if (!spg) {
4311                 spg = alloc_spg_mapping();
4312                 if (!spg) {
4313                         pr_err("%s failed to allocate spg\n", __func__);
4314                         return;
4315                 }
4316                 spg->spgid = m->spgid; /* struct */
4317                 insert_spg_mapping(&osd->o_backoff_mappings, spg);
4318         }
4319
4320         backoff = alloc_backoff();
4321         if (!backoff) {
4322                 pr_err("%s failed to allocate backoff\n", __func__);
4323                 return;
4324         }
4325         backoff->spgid = m->spgid; /* struct */
4326         backoff->id = m->id;
4327         backoff->begin = m->begin;
4328         m->begin = NULL; /* backoff now owns this */
4329         backoff->end = m->end;
4330         m->end = NULL;   /* ditto */
4331
4332         insert_backoff(&spg->backoffs, backoff);
4333         insert_backoff_by_id(&osd->o_backoffs_by_id, backoff);
4334
4335         /*
4336          * Ack with original backoff's epoch so that the OSD can
4337          * discard this if there was a PG split.
4338          */
4339         msg = create_backoff_message(backoff, m->map_epoch);
4340         if (!msg) {
4341                 pr_err("%s failed to allocate msg\n", __func__);
4342                 return;
4343         }
4344         ceph_con_send(&osd->o_con, msg);
4345 }
4346
4347 static bool target_contained_by(const struct ceph_osd_request_target *t,
4348                                 const struct ceph_hobject_id *begin,
4349                                 const struct ceph_hobject_id *end)
4350 {
4351         struct ceph_hobject_id hoid;
4352         int cmp;
4353
4354         hoid_fill_from_target(&hoid, t);
4355         cmp = hoid_compare(&hoid, begin);
4356         return !cmp || (cmp > 0 && hoid_compare(&hoid, end) < 0);
4357 }
4358
4359 static void handle_backoff_unblock(struct ceph_osd *osd,
4360                                    const struct MOSDBackoff *m)
4361 {
4362         struct ceph_spg_mapping *spg;
4363         struct ceph_osd_backoff *backoff;
4364         struct rb_node *n;
4365
4366         dout("%s osd%d spgid %llu.%xs%d id %llu\n", __func__, osd->o_osd,
4367              m->spgid.pgid.pool, m->spgid.pgid.seed, m->spgid.shard, m->id);
4368
4369         backoff = lookup_backoff_by_id(&osd->o_backoffs_by_id, m->id);
4370         if (!backoff) {
4371                 pr_err("%s osd%d spgid %llu.%xs%d id %llu backoff dne\n",
4372                        __func__, osd->o_osd, m->spgid.pgid.pool,
4373                        m->spgid.pgid.seed, m->spgid.shard, m->id);
4374                 return;
4375         }
4376
4377         if (hoid_compare(backoff->begin, m->begin) &&
4378             hoid_compare(backoff->end, m->end)) {
4379                 pr_err("%s osd%d spgid %llu.%xs%d id %llu bad range?\n",
4380                        __func__, osd->o_osd, m->spgid.pgid.pool,
4381                        m->spgid.pgid.seed, m->spgid.shard, m->id);
4382                 /* unblock it anyway... */
4383         }
4384
4385         spg = lookup_spg_mapping(&osd->o_backoff_mappings, &backoff->spgid);
4386         BUG_ON(!spg);
4387
4388         erase_backoff(&spg->backoffs, backoff);
4389         erase_backoff_by_id(&osd->o_backoffs_by_id, backoff);
4390         free_backoff(backoff);
4391
4392         if (RB_EMPTY_ROOT(&spg->backoffs)) {
4393                 erase_spg_mapping(&osd->o_backoff_mappings, spg);
4394                 free_spg_mapping(spg);
4395         }
4396
4397         for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) {
4398                 struct ceph_osd_request *req =
4399                     rb_entry(n, struct ceph_osd_request, r_node);
4400
4401                 if (!ceph_spg_compare(&req->r_t.spgid, &m->spgid)) {
4402                         /*
4403                          * Match against @m, not @backoff -- the PG may
4404                          * have split on the OSD.
4405                          */
4406                         if (target_contained_by(&req->r_t, m->begin, m->end)) {
4407                                 /*
4408                                  * If no other installed backoff applies,
4409                                  * resend.
4410                                  */
4411                                 send_request(req);
4412                         }
4413                 }
4414         }
4415 }
4416
4417 static void handle_backoff(struct ceph_osd *osd, struct ceph_msg *msg)
4418 {
4419         struct ceph_osd_client *osdc = osd->o_osdc;
4420         struct MOSDBackoff m;
4421         int ret;
4422
4423         down_read(&osdc->lock);
4424         if (!osd_registered(osd)) {
4425                 dout("%s osd%d unknown\n", __func__, osd->o_osd);
4426                 up_read(&osdc->lock);
4427                 return;
4428         }
4429         WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num));
4430
4431         mutex_lock(&osd->lock);
4432         ret = decode_MOSDBackoff(msg, &m);
4433         if (ret) {
4434                 pr_err("failed to decode MOSDBackoff: %d\n", ret);
4435                 ceph_msg_dump(msg);
4436                 goto out_unlock;
4437         }
4438
4439         switch (m.op) {
4440         case CEPH_OSD_BACKOFF_OP_BLOCK:
4441                 handle_backoff_block(osd, &m);
4442                 break;
4443         case CEPH_OSD_BACKOFF_OP_UNBLOCK:
4444                 handle_backoff_unblock(osd, &m);
4445                 break;
4446         default:
4447                 pr_err("%s osd%d unknown op %d\n", __func__, osd->o_osd, m.op);
4448         }
4449
4450         free_hoid(m.begin);
4451         free_hoid(m.end);
4452
4453 out_unlock:
4454         mutex_unlock(&osd->lock);
4455         up_read(&osdc->lock);
4456 }
4457
4458 /*
4459  * Process osd watch notifications
4460  */
4461 static void handle_watch_notify(struct ceph_osd_client *osdc,
4462                                 struct ceph_msg *msg)
4463 {
4464         void *p = msg->front.iov_base;
4465         void *const end = p + msg->front.iov_len;
4466         struct ceph_osd_linger_request *lreq;
4467         struct linger_work *lwork;
4468         u8 proto_ver, opcode;
4469         u64 cookie, notify_id;
4470         u64 notifier_id = 0;
4471         s32 return_code = 0;
4472         void *payload = NULL;
4473         u32 payload_len = 0;
4474
4475         ceph_decode_8_safe(&p, end, proto_ver, bad);
4476         ceph_decode_8_safe(&p, end, opcode, bad);
4477         ceph_decode_64_safe(&p, end, cookie, bad);
4478         p += 8; /* skip ver */
4479         ceph_decode_64_safe(&p, end, notify_id, bad);
4480
4481         if (proto_ver >= 1) {
4482                 ceph_decode_32_safe(&p, end, payload_len, bad);
4483                 ceph_decode_need(&p, end, payload_len, bad);
4484                 payload = p;
4485                 p += payload_len;
4486         }
4487
4488         if (le16_to_cpu(msg->hdr.version) >= 2)
4489                 ceph_decode_32_safe(&p, end, return_code, bad);
4490
4491         if (le16_to_cpu(msg->hdr.version) >= 3)
4492                 ceph_decode_64_safe(&p, end, notifier_id, bad);
4493
4494         down_read(&osdc->lock);
4495         lreq = lookup_linger_osdc(&osdc->linger_requests, cookie);
4496         if (!lreq) {
4497                 dout("%s opcode %d cookie %llu dne\n", __func__, opcode,
4498                      cookie);
4499                 goto out_unlock_osdc;
4500         }
4501
4502         mutex_lock(&lreq->lock);
4503         dout("%s opcode %d cookie %llu lreq %p is_watch %d\n", __func__,
4504              opcode, cookie, lreq, lreq->is_watch);
4505         if (opcode == CEPH_WATCH_EVENT_DISCONNECT) {
4506                 if (!lreq->last_error) {
4507                         lreq->last_error = -ENOTCONN;
4508                         queue_watch_error(lreq);
4509                 }
4510         } else if (!lreq->is_watch) {
4511                 /* CEPH_WATCH_EVENT_NOTIFY_COMPLETE */
4512                 if (lreq->notify_id && lreq->notify_id != notify_id) {
4513                         dout("lreq %p notify_id %llu != %llu, ignoring\n", lreq,
4514                              lreq->notify_id, notify_id);
4515                 } else if (!completion_done(&lreq->notify_finish_wait)) {
4516                         struct ceph_msg_data *data =
4517                             msg->num_data_items ? &msg->data[0] : NULL;
4518
4519                         if (data) {
4520                                 if (lreq->preply_pages) {
4521                                         WARN_ON(data->type !=
4522                                                         CEPH_MSG_DATA_PAGES);
4523                                         *lreq->preply_pages = data->pages;
4524                                         *lreq->preply_len = data->length;
4525                                         data->own_pages = false;
4526                                 }
4527                         }
4528                         lreq->notify_finish_error = return_code;
4529                         complete_all(&lreq->notify_finish_wait);
4530                 }
4531         } else {
4532                 /* CEPH_WATCH_EVENT_NOTIFY */
4533                 lwork = lwork_alloc(lreq, do_watch_notify);
4534                 if (!lwork) {
4535                         pr_err("failed to allocate notify-lwork\n");
4536                         goto out_unlock_lreq;
4537                 }
4538
4539                 lwork->notify.notify_id = notify_id;
4540                 lwork->notify.notifier_id = notifier_id;
4541                 lwork->notify.payload = payload;
4542                 lwork->notify.payload_len = payload_len;
4543                 lwork->notify.msg = ceph_msg_get(msg);
4544                 lwork_queue(lwork);
4545         }
4546
4547 out_unlock_lreq:
4548         mutex_unlock(&lreq->lock);
4549 out_unlock_osdc:
4550         up_read(&osdc->lock);
4551         return;
4552
4553 bad:
4554         pr_err("osdc handle_watch_notify corrupt msg\n");
4555 }
4556
4557 /*
4558  * Register request, send initial attempt.
4559  */
4560 int ceph_osdc_start_request(struct ceph_osd_client *osdc,
4561                             struct ceph_osd_request *req,
4562                             bool nofail)
4563 {
4564         down_read(&osdc->lock);
4565         submit_request(req, false);
4566         up_read(&osdc->lock);
4567
4568         return 0;
4569 }
4570 EXPORT_SYMBOL(ceph_osdc_start_request);
4571
4572 /*
4573  * Unregister a registered request.  The request is not completed:
4574  * ->r_result isn't set and __complete_request() isn't called.
4575  */
4576 void ceph_osdc_cancel_request(struct ceph_osd_request *req)
4577 {
4578         struct ceph_osd_client *osdc = req->r_osdc;
4579
4580         down_write(&osdc->lock);
4581         if (req->r_osd)
4582                 cancel_request(req);
4583         up_write(&osdc->lock);
4584 }
4585 EXPORT_SYMBOL(ceph_osdc_cancel_request);
4586
4587 /*
4588  * @timeout: in jiffies, 0 means "wait forever"
4589  */
4590 static int wait_request_timeout(struct ceph_osd_request *req,
4591                                 unsigned long timeout)
4592 {
4593         long left;
4594
4595         dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
4596         left = wait_for_completion_killable_timeout(&req->r_completion,
4597                                                 ceph_timeout_jiffies(timeout));
4598         if (left <= 0) {
4599                 left = left ?: -ETIMEDOUT;
4600                 ceph_osdc_cancel_request(req);
4601         } else {
4602                 left = req->r_result; /* completed */
4603         }
4604
4605         return left;
4606 }
4607
4608 /*
4609  * wait for a request to complete
4610  */
4611 int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
4612                            struct ceph_osd_request *req)
4613 {
4614         return wait_request_timeout(req, 0);
4615 }
4616 EXPORT_SYMBOL(ceph_osdc_wait_request);
4617
4618 /*
4619  * sync - wait for all in-flight requests to flush.  avoid starvation.
4620  */
4621 void ceph_osdc_sync(struct ceph_osd_client *osdc)
4622 {
4623         struct rb_node *n, *p;
4624         u64 last_tid = atomic64_read(&osdc->last_tid);
4625
4626 again:
4627         down_read(&osdc->lock);
4628         for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
4629                 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
4630
4631                 mutex_lock(&osd->lock);
4632                 for (p = rb_first(&osd->o_requests); p; p = rb_next(p)) {
4633                         struct ceph_osd_request *req =
4634                             rb_entry(p, struct ceph_osd_request, r_node);
4635
4636                         if (req->r_tid > last_tid)
4637                                 break;
4638
4639                         if (!(req->r_flags & CEPH_OSD_FLAG_WRITE))
4640                                 continue;
4641
4642                         ceph_osdc_get_request(req);
4643                         mutex_unlock(&osd->lock);
4644                         up_read(&osdc->lock);
4645                         dout("%s waiting on req %p tid %llu last_tid %llu\n",
4646                              __func__, req, req->r_tid, last_tid);
4647                         wait_for_completion(&req->r_completion);
4648                         ceph_osdc_put_request(req);
4649                         goto again;
4650                 }
4651
4652                 mutex_unlock(&osd->lock);
4653         }
4654
4655         up_read(&osdc->lock);
4656         dout("%s done last_tid %llu\n", __func__, last_tid);
4657 }
4658 EXPORT_SYMBOL(ceph_osdc_sync);
4659
4660 static struct ceph_osd_request *
4661 alloc_linger_request(struct ceph_osd_linger_request *lreq)
4662 {
4663         struct ceph_osd_request *req;
4664
4665         req = ceph_osdc_alloc_request(lreq->osdc, NULL, 1, false, GFP_NOIO);
4666         if (!req)
4667                 return NULL;
4668
4669         ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
4670         ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
4671         return req;
4672 }
4673
4674 static struct ceph_osd_request *
4675 alloc_watch_request(struct ceph_osd_linger_request *lreq, u8 watch_opcode)
4676 {
4677         struct ceph_osd_request *req;
4678
4679         req = alloc_linger_request(lreq);
4680         if (!req)
4681                 return NULL;
4682
4683         /*
4684          * Pass 0 for cookie because we don't know it yet, it will be
4685          * filled in by linger_submit().
4686          */
4687         osd_req_op_watch_init(req, 0, 0, watch_opcode);
4688
4689         if (ceph_osdc_alloc_messages(req, GFP_NOIO)) {
4690                 ceph_osdc_put_request(req);
4691                 return NULL;
4692         }
4693
4694         return req;
4695 }
4696
4697 /*
4698  * Returns a handle, caller owns a ref.
4699  */
4700 struct ceph_osd_linger_request *
4701 ceph_osdc_watch(struct ceph_osd_client *osdc,
4702                 struct ceph_object_id *oid,
4703                 struct ceph_object_locator *oloc,
4704                 rados_watchcb2_t wcb,
4705                 rados_watcherrcb_t errcb,
4706                 void *data)
4707 {
4708         struct ceph_osd_linger_request *lreq;
4709         int ret;
4710
4711         lreq = linger_alloc(osdc);
4712         if (!lreq)
4713                 return ERR_PTR(-ENOMEM);
4714
4715         lreq->is_watch = true;
4716         lreq->wcb = wcb;
4717         lreq->errcb = errcb;
4718         lreq->data = data;
4719         lreq->watch_valid_thru = jiffies;
4720
4721         ceph_oid_copy(&lreq->t.base_oid, oid);
4722         ceph_oloc_copy(&lreq->t.base_oloc, oloc);
4723         lreq->t.flags = CEPH_OSD_FLAG_WRITE;
4724         ktime_get_real_ts64(&lreq->mtime);
4725
4726         lreq->reg_req = alloc_watch_request(lreq, CEPH_OSD_WATCH_OP_WATCH);
4727         if (!lreq->reg_req) {
4728                 ret = -ENOMEM;
4729                 goto err_put_lreq;
4730         }
4731
4732         lreq->ping_req = alloc_watch_request(lreq, CEPH_OSD_WATCH_OP_PING);
4733         if (!lreq->ping_req) {
4734                 ret = -ENOMEM;
4735                 goto err_put_lreq;
4736         }
4737
4738         linger_submit(lreq);
4739         ret = linger_reg_commit_wait(lreq);
4740         if (ret) {
4741                 linger_cancel(lreq);
4742                 goto err_put_lreq;
4743         }
4744
4745         return lreq;
4746
4747 err_put_lreq:
4748         linger_put(lreq);
4749         return ERR_PTR(ret);
4750 }
4751 EXPORT_SYMBOL(ceph_osdc_watch);
4752
4753 /*
4754  * Releases a ref.
4755  *
4756  * Times out after mount_timeout to preserve rbd unmap behaviour
4757  * introduced in 2894e1d76974 ("rbd: timeout watch teardown on unmap
4758  * with mount_timeout").
4759  */
4760 int ceph_osdc_unwatch(struct ceph_osd_client *osdc,
4761                       struct ceph_osd_linger_request *lreq)
4762 {
4763         struct ceph_options *opts = osdc->client->options;
4764         struct ceph_osd_request *req;
4765         int ret;
4766
4767         req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
4768         if (!req)
4769                 return -ENOMEM;
4770
4771         ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
4772         ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
4773         req->r_flags = CEPH_OSD_FLAG_WRITE;
4774         ktime_get_real_ts64(&req->r_mtime);
4775         osd_req_op_watch_init(req, 0, lreq->linger_id,
4776                               CEPH_OSD_WATCH_OP_UNWATCH);
4777
4778         ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
4779         if (ret)
4780                 goto out_put_req;
4781
4782         ceph_osdc_start_request(osdc, req, false);
4783         linger_cancel(lreq);
4784         linger_put(lreq);
4785         ret = wait_request_timeout(req, opts->mount_timeout);
4786
4787 out_put_req:
4788         ceph_osdc_put_request(req);
4789         return ret;
4790 }
4791 EXPORT_SYMBOL(ceph_osdc_unwatch);
4792
4793 static int osd_req_op_notify_ack_init(struct ceph_osd_request *req, int which,
4794                                       u64 notify_id, u64 cookie, void *payload,
4795                                       u32 payload_len)
4796 {
4797         struct ceph_osd_req_op *op;
4798         struct ceph_pagelist *pl;
4799         int ret;
4800
4801         op = osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY_ACK, 0);
4802
4803         pl = ceph_pagelist_alloc(GFP_NOIO);
4804         if (!pl)
4805                 return -ENOMEM;
4806
4807         ret = ceph_pagelist_encode_64(pl, notify_id);
4808         ret |= ceph_pagelist_encode_64(pl, cookie);
4809         if (payload) {
4810                 ret |= ceph_pagelist_encode_32(pl, payload_len);
4811                 ret |= ceph_pagelist_append(pl, payload, payload_len);
4812         } else {
4813                 ret |= ceph_pagelist_encode_32(pl, 0);
4814         }
4815         if (ret) {
4816                 ceph_pagelist_release(pl);
4817                 return -ENOMEM;
4818         }
4819
4820         ceph_osd_data_pagelist_init(&op->notify_ack.request_data, pl);
4821         op->indata_len = pl->length;
4822         return 0;
4823 }
4824
4825 int ceph_osdc_notify_ack(struct ceph_osd_client *osdc,
4826                          struct ceph_object_id *oid,
4827                          struct ceph_object_locator *oloc,
4828                          u64 notify_id,
4829                          u64 cookie,
4830                          void *payload,
4831                          u32 payload_len)
4832 {
4833         struct ceph_osd_request *req;
4834         int ret;
4835
4836         req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
4837         if (!req)
4838                 return -ENOMEM;
4839
4840         ceph_oid_copy(&req->r_base_oid, oid);
4841         ceph_oloc_copy(&req->r_base_oloc, oloc);
4842         req->r_flags = CEPH_OSD_FLAG_READ;
4843
4844         ret = osd_req_op_notify_ack_init(req, 0, notify_id, cookie, payload,
4845                                          payload_len);
4846         if (ret)
4847                 goto out_put_req;
4848
4849         ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
4850         if (ret)
4851                 goto out_put_req;
4852
4853         ceph_osdc_start_request(osdc, req, false);
4854         ret = ceph_osdc_wait_request(osdc, req);
4855
4856 out_put_req:
4857         ceph_osdc_put_request(req);
4858         return ret;
4859 }
4860 EXPORT_SYMBOL(ceph_osdc_notify_ack);
4861
4862 static int osd_req_op_notify_init(struct ceph_osd_request *req, int which,
4863                                   u64 cookie, u32 prot_ver, u32 timeout,
4864                                   void *payload, u32 payload_len)
4865 {
4866         struct ceph_osd_req_op *op;
4867         struct ceph_pagelist *pl;
4868         int ret;
4869
4870         op = osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY, 0);
4871         op->notify.cookie = cookie;
4872
4873         pl = ceph_pagelist_alloc(GFP_NOIO);
4874         if (!pl)
4875                 return -ENOMEM;
4876
4877         ret = ceph_pagelist_encode_32(pl, 1); /* prot_ver */
4878         ret |= ceph_pagelist_encode_32(pl, timeout);
4879         ret |= ceph_pagelist_encode_32(pl, payload_len);
4880         ret |= ceph_pagelist_append(pl, payload, payload_len);
4881         if (ret) {
4882                 ceph_pagelist_release(pl);
4883                 return -ENOMEM;
4884         }
4885
4886         ceph_osd_data_pagelist_init(&op->notify.request_data, pl);
4887         op->indata_len = pl->length;
4888         return 0;
4889 }
4890
4891 /*
4892  * @timeout: in seconds
4893  *
4894  * @preply_{pages,len} are initialized both on success and error.
4895  * The caller is responsible for:
4896  *
4897  *     ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len))
4898  */
4899 int ceph_osdc_notify(struct ceph_osd_client *osdc,
4900                      struct ceph_object_id *oid,
4901                      struct ceph_object_locator *oloc,
4902                      void *payload,
4903                      u32 payload_len,
4904                      u32 timeout,
4905                      struct page ***preply_pages,
4906                      size_t *preply_len)
4907 {
4908         struct ceph_osd_linger_request *lreq;
4909         struct page **pages;
4910         int ret;
4911
4912         WARN_ON(!timeout);
4913         if (preply_pages) {
4914                 *preply_pages = NULL;
4915                 *preply_len = 0;
4916         }
4917
4918         lreq = linger_alloc(osdc);
4919         if (!lreq)
4920                 return -ENOMEM;
4921
4922         lreq->preply_pages = preply_pages;
4923         lreq->preply_len = preply_len;
4924
4925         ceph_oid_copy(&lreq->t.base_oid, oid);
4926         ceph_oloc_copy(&lreq->t.base_oloc, oloc);
4927         lreq->t.flags = CEPH_OSD_FLAG_READ;
4928
4929         lreq->reg_req = alloc_linger_request(lreq);
4930         if (!lreq->reg_req) {
4931                 ret = -ENOMEM;
4932                 goto out_put_lreq;
4933         }
4934
4935         /*
4936          * Pass 0 for cookie because we don't know it yet, it will be
4937          * filled in by linger_submit().
4938          */
4939         ret = osd_req_op_notify_init(lreq->reg_req, 0, 0, 1, timeout,
4940                                      payload, payload_len);
4941         if (ret)
4942                 goto out_put_lreq;
4943
4944         /* for notify_id */
4945         pages = ceph_alloc_page_vector(1, GFP_NOIO);
4946         if (IS_ERR(pages)) {
4947                 ret = PTR_ERR(pages);
4948                 goto out_put_lreq;
4949         }
4950         ceph_osd_data_pages_init(osd_req_op_data(lreq->reg_req, 0, notify,
4951                                                  response_data),
4952                                  pages, PAGE_SIZE, 0, false, true);
4953
4954         ret = ceph_osdc_alloc_messages(lreq->reg_req, GFP_NOIO);
4955         if (ret)
4956                 goto out_put_lreq;
4957
4958         linger_submit(lreq);
4959         ret = linger_reg_commit_wait(lreq);
4960         if (!ret)
4961                 ret = linger_notify_finish_wait(lreq);
4962         else
4963                 dout("lreq %p failed to initiate notify %d\n", lreq, ret);
4964
4965         linger_cancel(lreq);
4966 out_put_lreq:
4967         linger_put(lreq);
4968         return ret;
4969 }
4970 EXPORT_SYMBOL(ceph_osdc_notify);
4971
4972 /*
4973  * Return the number of milliseconds since the watch was last
4974  * confirmed, or an error.  If there is an error, the watch is no
4975  * longer valid, and should be destroyed with ceph_osdc_unwatch().
4976  */
4977 int ceph_osdc_watch_check(struct ceph_osd_client *osdc,
4978                           struct ceph_osd_linger_request *lreq)
4979 {
4980         unsigned long stamp, age;
4981         int ret;
4982
4983         down_read(&osdc->lock);
4984         mutex_lock(&lreq->lock);
4985         stamp = lreq->watch_valid_thru;
4986         if (!list_empty(&lreq->pending_lworks)) {
4987                 struct linger_work *lwork =
4988                     list_first_entry(&lreq->pending_lworks,
4989                                      struct linger_work,
4990                                      pending_item);
4991
4992                 if (time_before(lwork->queued_stamp, stamp))
4993                         stamp = lwork->queued_stamp;
4994         }
4995         age = jiffies - stamp;
4996         dout("%s lreq %p linger_id %llu age %lu last_error %d\n", __func__,
4997              lreq, lreq->linger_id, age, lreq->last_error);
4998         /* we are truncating to msecs, so return a safe upper bound */
4999         ret = lreq->last_error ?: 1 + jiffies_to_msecs(age);
5000
5001         mutex_unlock(&lreq->lock);
5002         up_read(&osdc->lock);
5003         return ret;
5004 }
5005
5006 static int decode_watcher(void **p, void *end, struct ceph_watch_item *item)
5007 {
5008         u8 struct_v;
5009         u32 struct_len;
5010         int ret;
5011
5012         ret = ceph_start_decoding(p, end, 2, "watch_item_t",
5013                                   &struct_v, &struct_len);
5014         if (ret)
5015                 goto bad;
5016
5017         ret = -EINVAL;
5018         ceph_decode_copy_safe(p, end, &item->name, sizeof(item->name), bad);
5019         ceph_decode_64_safe(p, end, item->cookie, bad);
5020         ceph_decode_skip_32(p, end, bad); /* skip timeout seconds */
5021
5022         if (struct_v >= 2) {
5023                 ret = ceph_decode_entity_addr(p, end, &item->addr);
5024                 if (ret)
5025                         goto bad;
5026         } else {
5027                 ret = 0;
5028         }
5029
5030         dout("%s %s%llu cookie %llu addr %s\n", __func__,
5031              ENTITY_NAME(item->name), item->cookie,
5032              ceph_pr_addr(&item->addr));
5033 bad:
5034         return ret;
5035 }
5036
5037 static int decode_watchers(void **p, void *end,
5038                            struct ceph_watch_item **watchers,
5039                            u32 *num_watchers)
5040 {
5041         u8 struct_v;
5042         u32 struct_len;
5043         int i;
5044         int ret;
5045
5046         ret = ceph_start_decoding(p, end, 1, "obj_list_watch_response_t",
5047                                   &struct_v, &struct_len);
5048         if (ret)
5049                 return ret;
5050
5051         *num_watchers = ceph_decode_32(p);
5052         *watchers = kcalloc(*num_watchers, sizeof(**watchers), GFP_NOIO);
5053         if (!*watchers)
5054                 return -ENOMEM;
5055
5056         for (i = 0; i < *num_watchers; i++) {
5057                 ret = decode_watcher(p, end, *watchers + i);
5058                 if (ret) {
5059                         kfree(*watchers);
5060                         return ret;
5061                 }
5062         }
5063
5064         return 0;
5065 }
5066
5067 /*
5068  * On success, the caller is responsible for:
5069  *
5070  *     kfree(watchers);
5071  */
5072 int ceph_osdc_list_watchers(struct ceph_osd_client *osdc,
5073                             struct ceph_object_id *oid,
5074                             struct ceph_object_locator *oloc,
5075                             struct ceph_watch_item **watchers,
5076                             u32 *num_watchers)
5077 {
5078         struct ceph_osd_request *req;
5079         struct page **pages;
5080         int ret;
5081
5082         req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
5083         if (!req)
5084                 return -ENOMEM;
5085
5086         ceph_oid_copy(&req->r_base_oid, oid);
5087         ceph_oloc_copy(&req->r_base_oloc, oloc);
5088         req->r_flags = CEPH_OSD_FLAG_READ;
5089
5090         pages = ceph_alloc_page_vector(1, GFP_NOIO);
5091         if (IS_ERR(pages)) {
5092                 ret = PTR_ERR(pages);
5093                 goto out_put_req;
5094         }
5095
5096         osd_req_op_init(req, 0, CEPH_OSD_OP_LIST_WATCHERS, 0);
5097         ceph_osd_data_pages_init(osd_req_op_data(req, 0, list_watchers,
5098                                                  response_data),
5099                                  pages, PAGE_SIZE, 0, false, true);
5100
5101         ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
5102         if (ret)
5103                 goto out_put_req;
5104
5105         ceph_osdc_start_request(osdc, req, false);
5106         ret = ceph_osdc_wait_request(osdc, req);
5107         if (ret >= 0) {
5108                 void *p = page_address(pages[0]);
5109                 void *const end = p + req->r_ops[0].outdata_len;
5110
5111                 ret = decode_watchers(&p, end, watchers, num_watchers);
5112         }
5113
5114 out_put_req:
5115         ceph_osdc_put_request(req);
5116         return ret;
5117 }
5118 EXPORT_SYMBOL(ceph_osdc_list_watchers);
5119
5120 /*
5121  * Call all pending notify callbacks - for use after a watch is
5122  * unregistered, to make sure no more callbacks for it will be invoked
5123  */
5124 void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
5125 {
5126         dout("%s osdc %p\n", __func__, osdc);
5127         flush_workqueue(osdc->notify_wq);
5128 }
5129 EXPORT_SYMBOL(ceph_osdc_flush_notifies);
5130
5131 void ceph_osdc_maybe_request_map(struct ceph_osd_client *osdc)
5132 {
5133         down_read(&osdc->lock);
5134         maybe_request_map(osdc);
5135         up_read(&osdc->lock);
5136 }
5137 EXPORT_SYMBOL(ceph_osdc_maybe_request_map);
5138
5139 /*
5140  * Execute an OSD class method on an object.
5141  *
5142  * @flags: CEPH_OSD_FLAG_*
5143  * @resp_len: in/out param for reply length
5144  */
5145 int ceph_osdc_call(struct ceph_osd_client *osdc,
5146                    struct ceph_object_id *oid,
5147                    struct ceph_object_locator *oloc,
5148                    const char *class, const char *method,
5149                    unsigned int flags,
5150                    struct page *req_page, size_t req_len,
5151                    struct page **resp_pages, size_t *resp_len)
5152 {
5153         struct ceph_osd_request *req;
5154         int ret;
5155
5156         if (req_len > PAGE_SIZE)
5157                 return -E2BIG;
5158
5159         req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
5160         if (!req)
5161                 return -ENOMEM;
5162
5163         ceph_oid_copy(&req->r_base_oid, oid);
5164         ceph_oloc_copy(&req->r_base_oloc, oloc);
5165         req->r_flags = flags;
5166
5167         ret = osd_req_op_cls_init(req, 0, class, method);
5168         if (ret)
5169                 goto out_put_req;
5170
5171         if (req_page)
5172                 osd_req_op_cls_request_data_pages(req, 0, &req_page, req_len,
5173                                                   0, false, false);
5174         if (resp_pages)
5175                 osd_req_op_cls_response_data_pages(req, 0, resp_pages,
5176                                                    *resp_len, 0, false, false);
5177
5178         ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
5179         if (ret)
5180                 goto out_put_req;
5181
5182         ceph_osdc_start_request(osdc, req, false);
5183         ret = ceph_osdc_wait_request(osdc, req);
5184         if (ret >= 0) {
5185                 ret = req->r_ops[0].rval;
5186                 if (resp_pages)
5187                         *resp_len = req->r_ops[0].outdata_len;
5188         }
5189
5190 out_put_req:
5191         ceph_osdc_put_request(req);
5192         return ret;
5193 }
5194 EXPORT_SYMBOL(ceph_osdc_call);
5195
5196 /*
5197  * reset all osd connections
5198  */
5199 void ceph_osdc_reopen_osds(struct ceph_osd_client *osdc)
5200 {
5201         struct rb_node *n;
5202
5203         down_write(&osdc->lock);
5204         for (n = rb_first(&osdc->osds); n; ) {
5205                 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
5206
5207                 n = rb_next(n);
5208                 if (!reopen_osd(osd))
5209                         kick_osd_requests(osd);
5210         }
5211         up_write(&osdc->lock);
5212 }
5213
5214 /*
5215  * init, shutdown
5216  */
5217 int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
5218 {
5219         int err;
5220
5221         dout("init\n");
5222         osdc->client = client;
5223         init_rwsem(&osdc->lock);
5224         osdc->osds = RB_ROOT;
5225         INIT_LIST_HEAD(&osdc->osd_lru);
5226         spin_lock_init(&osdc->osd_lru_lock);
5227         osd_init(&osdc->homeless_osd);
5228         osdc->homeless_osd.o_osdc = osdc;
5229         osdc->homeless_osd.o_osd = CEPH_HOMELESS_OSD;
5230         osdc->last_linger_id = CEPH_LINGER_ID_START;
5231         osdc->linger_requests = RB_ROOT;
5232         osdc->map_checks = RB_ROOT;
5233         osdc->linger_map_checks = RB_ROOT;
5234         INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
5235         INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
5236
5237         err = -ENOMEM;
5238         osdc->osdmap = ceph_osdmap_alloc();
5239         if (!osdc->osdmap)
5240                 goto out;
5241
5242         osdc->req_mempool = mempool_create_slab_pool(10,
5243                                                      ceph_osd_request_cache);
5244         if (!osdc->req_mempool)
5245                 goto out_map;
5246
5247         err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
5248                                 PAGE_SIZE, CEPH_OSD_SLAB_OPS, 10, "osd_op");
5249         if (err < 0)
5250                 goto out_mempool;
5251         err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
5252                                 PAGE_SIZE, CEPH_OSD_SLAB_OPS, 10,
5253                                 "osd_op_reply");
5254         if (err < 0)
5255                 goto out_msgpool;
5256
5257         err = -ENOMEM;
5258         osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
5259         if (!osdc->notify_wq)
5260                 goto out_msgpool_reply;
5261
5262         osdc->completion_wq = create_singlethread_workqueue("ceph-completion");
5263         if (!osdc->completion_wq)
5264                 goto out_notify_wq;
5265
5266         schedule_delayed_work(&osdc->timeout_work,
5267                               osdc->client->options->osd_keepalive_timeout);
5268         schedule_delayed_work(&osdc->osds_timeout_work,
5269             round_jiffies_relative(osdc->client->options->osd_idle_ttl));
5270
5271         return 0;
5272
5273 out_notify_wq:
5274         destroy_workqueue(osdc->notify_wq);
5275 out_msgpool_reply:
5276         ceph_msgpool_destroy(&osdc->msgpool_op_reply);
5277 out_msgpool:
5278         ceph_msgpool_destroy(&osdc->msgpool_op);
5279 out_mempool:
5280         mempool_destroy(osdc->req_mempool);
5281 out_map:
5282         ceph_osdmap_destroy(osdc->osdmap);
5283 out:
5284         return err;
5285 }
5286
5287 void ceph_osdc_stop(struct ceph_osd_client *osdc)
5288 {
5289         destroy_workqueue(osdc->completion_wq);
5290         destroy_workqueue(osdc->notify_wq);
5291         cancel_delayed_work_sync(&osdc->timeout_work);
5292         cancel_delayed_work_sync(&osdc->osds_timeout_work);
5293
5294         down_write(&osdc->lock);
5295         while (!RB_EMPTY_ROOT(&osdc->osds)) {
5296                 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
5297                                                 struct ceph_osd, o_node);
5298                 close_osd(osd);
5299         }
5300         up_write(&osdc->lock);
5301         WARN_ON(refcount_read(&osdc->homeless_osd.o_ref) != 1);
5302         osd_cleanup(&osdc->homeless_osd);
5303
5304         WARN_ON(!list_empty(&osdc->osd_lru));
5305         WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_requests));
5306         WARN_ON(!RB_EMPTY_ROOT(&osdc->map_checks));
5307         WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_map_checks));
5308         WARN_ON(atomic_read(&osdc->num_requests));
5309         WARN_ON(atomic_read(&osdc->num_homeless));
5310
5311         ceph_osdmap_destroy(osdc->osdmap);
5312         mempool_destroy(osdc->req_mempool);
5313         ceph_msgpool_destroy(&osdc->msgpool_op);
5314         ceph_msgpool_destroy(&osdc->msgpool_op_reply);
5315 }
5316
5317 int osd_req_op_copy_from_init(struct ceph_osd_request *req,
5318                               u64 src_snapid, u64 src_version,
5319                               struct ceph_object_id *src_oid,
5320                               struct ceph_object_locator *src_oloc,
5321                               u32 src_fadvise_flags,
5322                               u32 dst_fadvise_flags,
5323                               u32 truncate_seq, u64 truncate_size,
5324                               u8 copy_from_flags)
5325 {
5326         struct ceph_osd_req_op *op;
5327         struct page **pages;
5328         void *p, *end;
5329
5330         pages = ceph_alloc_page_vector(1, GFP_KERNEL);
5331         if (IS_ERR(pages))
5332                 return PTR_ERR(pages);
5333
5334         op = osd_req_op_init(req, 0, CEPH_OSD_OP_COPY_FROM2,
5335                              dst_fadvise_flags);
5336         op->copy_from.snapid = src_snapid;
5337         op->copy_from.src_version = src_version;
5338         op->copy_from.flags = copy_from_flags;
5339         op->copy_from.src_fadvise_flags = src_fadvise_flags;
5340
5341         p = page_address(pages[0]);
5342         end = p + PAGE_SIZE;
5343         ceph_encode_string(&p, end, src_oid->name, src_oid->name_len);
5344         encode_oloc(&p, end, src_oloc);
5345         ceph_encode_32(&p, truncate_seq);
5346         ceph_encode_64(&p, truncate_size);
5347         op->indata_len = PAGE_SIZE - (end - p);
5348
5349         ceph_osd_data_pages_init(&op->copy_from.osd_data, pages,
5350                                  op->indata_len, 0, false, true);
5351         return 0;
5352 }
5353 EXPORT_SYMBOL(osd_req_op_copy_from_init);
5354
5355 int __init ceph_osdc_setup(void)
5356 {
5357         size_t size = sizeof(struct ceph_osd_request) +
5358             CEPH_OSD_SLAB_OPS * sizeof(struct ceph_osd_req_op);
5359
5360         BUG_ON(ceph_osd_request_cache);
5361         ceph_osd_request_cache = kmem_cache_create("ceph_osd_request", size,
5362                                                    0, 0, NULL);
5363
5364         return ceph_osd_request_cache ? 0 : -ENOMEM;
5365 }
5366
5367 void ceph_osdc_cleanup(void)
5368 {
5369         BUG_ON(!ceph_osd_request_cache);
5370         kmem_cache_destroy(ceph_osd_request_cache);
5371         ceph_osd_request_cache = NULL;
5372 }
5373
5374 /*
5375  * handle incoming message
5376  */
5377 static void osd_dispatch(struct ceph_connection *con, struct ceph_msg *msg)
5378 {
5379         struct ceph_osd *osd = con->private;
5380         struct ceph_osd_client *osdc = osd->o_osdc;
5381         int type = le16_to_cpu(msg->hdr.type);
5382
5383         switch (type) {
5384         case CEPH_MSG_OSD_MAP:
5385                 ceph_osdc_handle_map(osdc, msg);
5386                 break;
5387         case CEPH_MSG_OSD_OPREPLY:
5388                 handle_reply(osd, msg);
5389                 break;
5390         case CEPH_MSG_OSD_BACKOFF:
5391                 handle_backoff(osd, msg);
5392                 break;
5393         case CEPH_MSG_WATCH_NOTIFY:
5394                 handle_watch_notify(osdc, msg);
5395                 break;
5396
5397         default:
5398                 pr_err("received unknown message type %d %s\n", type,
5399                        ceph_msg_type_name(type));
5400         }
5401
5402         ceph_msg_put(msg);
5403 }
5404
5405 /*
5406  * Lookup and return message for incoming reply.  Don't try to do
5407  * anything about a larger than preallocated data portion of the
5408  * message at the moment - for now, just skip the message.
5409  */
5410 static struct ceph_msg *get_reply(struct ceph_connection *con,
5411                                   struct ceph_msg_header *hdr,
5412                                   int *skip)
5413 {
5414         struct ceph_osd *osd = con->private;
5415         struct ceph_osd_client *osdc = osd->o_osdc;
5416         struct ceph_msg *m = NULL;
5417         struct ceph_osd_request *req;
5418         int front_len = le32_to_cpu(hdr->front_len);
5419         int data_len = le32_to_cpu(hdr->data_len);
5420         u64 tid = le64_to_cpu(hdr->tid);
5421
5422         down_read(&osdc->lock);
5423         if (!osd_registered(osd)) {
5424                 dout("%s osd%d unknown, skipping\n", __func__, osd->o_osd);
5425                 *skip = 1;
5426                 goto out_unlock_osdc;
5427         }
5428         WARN_ON(osd->o_osd != le64_to_cpu(hdr->src.num));
5429
5430         mutex_lock(&osd->lock);
5431         req = lookup_request(&osd->o_requests, tid);
5432         if (!req) {
5433                 dout("%s osd%d tid %llu unknown, skipping\n", __func__,
5434                      osd->o_osd, tid);
5435                 *skip = 1;
5436                 goto out_unlock_session;
5437         }
5438
5439         ceph_msg_revoke_incoming(req->r_reply);
5440
5441         if (front_len > req->r_reply->front_alloc_len) {
5442                 pr_warn("%s osd%d tid %llu front %d > preallocated %d\n",
5443                         __func__, osd->o_osd, req->r_tid, front_len,
5444                         req->r_reply->front_alloc_len);
5445                 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS,
5446                                  false);
5447                 if (!m)
5448                         goto out_unlock_session;
5449                 ceph_msg_put(req->r_reply);
5450                 req->r_reply = m;
5451         }
5452
5453         if (data_len > req->r_reply->data_length) {
5454                 pr_warn("%s osd%d tid %llu data %d > preallocated %zu, skipping\n",
5455                         __func__, osd->o_osd, req->r_tid, data_len,
5456                         req->r_reply->data_length);
5457                 m = NULL;
5458                 *skip = 1;
5459                 goto out_unlock_session;
5460         }
5461
5462         m = ceph_msg_get(req->r_reply);
5463         dout("get_reply tid %lld %p\n", tid, m);
5464
5465 out_unlock_session:
5466         mutex_unlock(&osd->lock);
5467 out_unlock_osdc:
5468         up_read(&osdc->lock);
5469         return m;
5470 }
5471
5472 static struct ceph_msg *alloc_msg_with_page_vector(struct ceph_msg_header *hdr)
5473 {
5474         struct ceph_msg *m;
5475         int type = le16_to_cpu(hdr->type);
5476         u32 front_len = le32_to_cpu(hdr->front_len);
5477         u32 data_len = le32_to_cpu(hdr->data_len);
5478
5479         m = ceph_msg_new2(type, front_len, 1, GFP_NOIO, false);
5480         if (!m)
5481                 return NULL;
5482
5483         if (data_len) {
5484                 struct page **pages;
5485
5486                 pages = ceph_alloc_page_vector(calc_pages_for(0, data_len),
5487                                                GFP_NOIO);
5488                 if (IS_ERR(pages)) {
5489                         ceph_msg_put(m);
5490                         return NULL;
5491                 }
5492
5493                 ceph_msg_data_add_pages(m, pages, data_len, 0, true);
5494         }
5495
5496         return m;
5497 }
5498
5499 static struct ceph_msg *osd_alloc_msg(struct ceph_connection *con,
5500                                       struct ceph_msg_header *hdr,
5501                                       int *skip)
5502 {
5503         struct ceph_osd *osd = con->private;
5504         int type = le16_to_cpu(hdr->type);
5505
5506         *skip = 0;
5507         switch (type) {
5508         case CEPH_MSG_OSD_MAP:
5509         case CEPH_MSG_OSD_BACKOFF:
5510         case CEPH_MSG_WATCH_NOTIFY:
5511                 return alloc_msg_with_page_vector(hdr);
5512         case CEPH_MSG_OSD_OPREPLY:
5513                 return get_reply(con, hdr, skip);
5514         default:
5515                 pr_warn("%s osd%d unknown msg type %d, skipping\n", __func__,
5516                         osd->o_osd, type);
5517                 *skip = 1;
5518                 return NULL;
5519         }
5520 }
5521
5522 /*
5523  * Wrappers to refcount containing ceph_osd struct
5524  */
5525 static struct ceph_connection *osd_get_con(struct ceph_connection *con)
5526 {
5527         struct ceph_osd *osd = con->private;
5528         if (get_osd(osd))
5529                 return con;
5530         return NULL;
5531 }
5532
5533 static void osd_put_con(struct ceph_connection *con)
5534 {
5535         struct ceph_osd *osd = con->private;
5536         put_osd(osd);
5537 }
5538
5539 /*
5540  * authentication
5541  */
5542
5543 /*
5544  * Note: returned pointer is the address of a structure that's
5545  * managed separately.  Caller must *not* attempt to free it.
5546  */
5547 static struct ceph_auth_handshake *
5548 osd_get_authorizer(struct ceph_connection *con, int *proto, int force_new)
5549 {
5550         struct ceph_osd *o = con->private;
5551         struct ceph_osd_client *osdc = o->o_osdc;
5552         struct ceph_auth_client *ac = osdc->client->monc.auth;
5553         struct ceph_auth_handshake *auth = &o->o_auth;
5554         int ret;
5555
5556         ret = __ceph_auth_get_authorizer(ac, auth, CEPH_ENTITY_TYPE_OSD,
5557                                          force_new, proto, NULL, NULL);
5558         if (ret)
5559                 return ERR_PTR(ret);
5560
5561         return auth;
5562 }
5563
5564 static int osd_add_authorizer_challenge(struct ceph_connection *con,
5565                                     void *challenge_buf, int challenge_buf_len)
5566 {
5567         struct ceph_osd *o = con->private;
5568         struct ceph_osd_client *osdc = o->o_osdc;
5569         struct ceph_auth_client *ac = osdc->client->monc.auth;
5570
5571         return ceph_auth_add_authorizer_challenge(ac, o->o_auth.authorizer,
5572                                             challenge_buf, challenge_buf_len);
5573 }
5574
5575 static int osd_verify_authorizer_reply(struct ceph_connection *con)
5576 {
5577         struct ceph_osd *o = con->private;
5578         struct ceph_osd_client *osdc = o->o_osdc;
5579         struct ceph_auth_client *ac = osdc->client->monc.auth;
5580         struct ceph_auth_handshake *auth = &o->o_auth;
5581
5582         return ceph_auth_verify_authorizer_reply(ac, auth->authorizer,
5583                 auth->authorizer_reply_buf, auth->authorizer_reply_buf_len,
5584                 NULL, NULL, NULL, NULL);
5585 }
5586
5587 static int osd_invalidate_authorizer(struct ceph_connection *con)
5588 {
5589         struct ceph_osd *o = con->private;
5590         struct ceph_osd_client *osdc = o->o_osdc;
5591         struct ceph_auth_client *ac = osdc->client->monc.auth;
5592
5593         ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
5594         return ceph_monc_validate_auth(&osdc->client->monc);
5595 }
5596
5597 static int osd_get_auth_request(struct ceph_connection *con,
5598                                 void *buf, int *buf_len,
5599                                 void **authorizer, int *authorizer_len)
5600 {
5601         struct ceph_osd *o = con->private;
5602         struct ceph_auth_client *ac = o->o_osdc->client->monc.auth;
5603         struct ceph_auth_handshake *auth = &o->o_auth;
5604         int ret;
5605
5606         ret = ceph_auth_get_authorizer(ac, auth, CEPH_ENTITY_TYPE_OSD,
5607                                        buf, buf_len);
5608         if (ret)
5609                 return ret;
5610
5611         *authorizer = auth->authorizer_buf;
5612         *authorizer_len = auth->authorizer_buf_len;
5613         return 0;
5614 }
5615
5616 static int osd_handle_auth_reply_more(struct ceph_connection *con,
5617                                       void *reply, int reply_len,
5618                                       void *buf, int *buf_len,
5619                                       void **authorizer, int *authorizer_len)
5620 {
5621         struct ceph_osd *o = con->private;
5622         struct ceph_auth_client *ac = o->o_osdc->client->monc.auth;
5623         struct ceph_auth_handshake *auth = &o->o_auth;
5624         int ret;
5625
5626         ret = ceph_auth_handle_svc_reply_more(ac, auth, reply, reply_len,
5627                                               buf, buf_len);
5628         if (ret)
5629                 return ret;
5630
5631         *authorizer = auth->authorizer_buf;
5632         *authorizer_len = auth->authorizer_buf_len;
5633         return 0;
5634 }
5635
5636 static int osd_handle_auth_done(struct ceph_connection *con,
5637                                 u64 global_id, void *reply, int reply_len,
5638                                 u8 *session_key, int *session_key_len,
5639                                 u8 *con_secret, int *con_secret_len)
5640 {
5641         struct ceph_osd *o = con->private;
5642         struct ceph_auth_client *ac = o->o_osdc->client->monc.auth;
5643         struct ceph_auth_handshake *auth = &o->o_auth;
5644
5645         return ceph_auth_handle_svc_reply_done(ac, auth, reply, reply_len,
5646                                                session_key, session_key_len,
5647                                                con_secret, con_secret_len);
5648 }
5649
5650 static int osd_handle_auth_bad_method(struct ceph_connection *con,
5651                                       int used_proto, int result,
5652                                       const int *allowed_protos, int proto_cnt,
5653                                       const int *allowed_modes, int mode_cnt)
5654 {
5655         struct ceph_osd *o = con->private;
5656         struct ceph_mon_client *monc = &o->o_osdc->client->monc;
5657         int ret;
5658
5659         if (ceph_auth_handle_bad_authorizer(monc->auth, CEPH_ENTITY_TYPE_OSD,
5660                                             used_proto, result,
5661                                             allowed_protos, proto_cnt,
5662                                             allowed_modes, mode_cnt)) {
5663                 ret = ceph_monc_validate_auth(monc);
5664                 if (ret)
5665                         return ret;
5666         }
5667
5668         return -EACCES;
5669 }
5670
5671 static void osd_reencode_message(struct ceph_msg *msg)
5672 {
5673         int type = le16_to_cpu(msg->hdr.type);
5674
5675         if (type == CEPH_MSG_OSD_OP)
5676                 encode_request_finish(msg);
5677 }
5678
5679 static int osd_sign_message(struct ceph_msg *msg)
5680 {
5681         struct ceph_osd *o = msg->con->private;
5682         struct ceph_auth_handshake *auth = &o->o_auth;
5683
5684         return ceph_auth_sign_message(auth, msg);
5685 }
5686
5687 static int osd_check_message_signature(struct ceph_msg *msg)
5688 {
5689         struct ceph_osd *o = msg->con->private;
5690         struct ceph_auth_handshake *auth = &o->o_auth;
5691
5692         return ceph_auth_check_message_signature(auth, msg);
5693 }
5694
5695 static const struct ceph_connection_operations osd_con_ops = {
5696         .get = osd_get_con,
5697         .put = osd_put_con,
5698         .alloc_msg = osd_alloc_msg,
5699         .dispatch = osd_dispatch,
5700         .fault = osd_fault,
5701         .reencode_message = osd_reencode_message,
5702         .get_authorizer = osd_get_authorizer,
5703         .add_authorizer_challenge = osd_add_authorizer_challenge,
5704         .verify_authorizer_reply = osd_verify_authorizer_reply,
5705         .invalidate_authorizer = osd_invalidate_authorizer,
5706         .sign_message = osd_sign_message,
5707         .check_message_signature = osd_check_message_signature,
5708         .get_auth_request = osd_get_auth_request,
5709         .handle_auth_reply_more = osd_handle_auth_reply_more,
5710         .handle_auth_done = osd_handle_auth_done,
5711         .handle_auth_bad_method = osd_handle_auth_bad_method,
5712 };