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