net/ceph/osd_client.c

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include <linux/ceph/ceph_debug.h>
   4
   5 #include <linux/module.h>
   6 #include <linux/err.h>
   7 #include <linux/highmem.h>
   8 #include <linux/mm.h>
   9 #include <linux/pagemap.h>
  10 #include <linux/slab.h>
  11 #include <linux/uaccess.h>
  12 #ifdef CONFIG_BLOCK
  13 #include <linux/bio.h>
  14 #endif
  15
  16 #include <linux/ceph/ceph_features.h>
  17 #include <linux/ceph/libceph.h>
  18 #include <linux/ceph/osd_client.h>
  19 #include <linux/ceph/messenger.h>
  20 #include <linux/ceph/decode.h>
  21 #include <linux/ceph/auth.h>
  22 #include <linux/ceph/pagelist.h>
  23 #include <linux/ceph/striper.h>
  24
  25 #define OSD_OPREPLY_FRONT_LEN   512
  26
  27 static struct kmem_cache        *ceph_osd_request_cache;
  28
  29 static const struct ceph_connection_operations osd_con_ops;
  30
  31 /*
  32  * Implement client access to distributed object storage cluster.
  33  *
  34  * All data objects are stored within a cluster/cloud of OSDs, or
  35  * "object storage devices."  (Note that Ceph OSDs have _nothing_ to
  36  * do with the T10 OSD extensions to SCSI.)  Ceph OSDs are simply
  37  * remote daemons serving up and coordinating consistent and safe
  38  * access to storage.
  39  *
  40  * Cluster membership and the mapping of data objects onto storage devices
  41  * are described by the osd map.
  42  *
  43  * We keep track of pending OSD requests (read, write), resubmit
  44  * requests to different OSDs when the cluster topology/data layout
  45  * change, or retry the affected requests when the communications
  46  * channel with an OSD is reset.
  47  */
  48
  49 static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req);
  50 static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req);
  51 static void link_linger(struct ceph_osd *osd,
  52                         struct ceph_osd_linger_request *lreq);
  53 static void unlink_linger(struct ceph_osd *osd,
  54                           struct ceph_osd_linger_request *lreq);
  55 static void clear_backoffs(struct ceph_osd *osd);
  56
  57 #if 1
  58 static inline bool rwsem_is_wrlocked(struct rw_semaphore *sem)
  59 {
  60         bool wrlocked = true;
  61
  62         if (unlikely(down_read_trylock(sem))) {
  63                 wrlocked = false;
  64                 up_read(sem);
  65         }
  66
  67         return wrlocked;
  68 }
  69 static inline void verify_osdc_locked(struct ceph_osd_client *osdc)
  70 {
  71         WARN_ON(!rwsem_is_locked(&osdc->lock));
  72 }
  73 static inline void verify_osdc_wrlocked(struct ceph_osd_client *osdc)
  74 {
  75         WARN_ON(!rwsem_is_wrlocked(&osdc->lock));
  76 }
  77 static inline void verify_osd_locked(struct ceph_osd *osd)
  78 {
  79         struct ceph_osd_client *osdc = osd->o_osdc;
  80
  81         WARN_ON(!(mutex_is_locked(&osd->lock) &&
  82                   rwsem_is_locked(&osdc->lock)) &&
  83                 !rwsem_is_wrlocked(&osdc->lock));
  84 }
  85 static inline void verify_lreq_locked(struct ceph_osd_linger_request *lreq)
  86 {
  87         WARN_ON(!mutex_is_locked(&lreq->lock));
  88 }
  89 #else
  90 static inline void verify_osdc_locked(struct ceph_osd_client *osdc) { }
  91 static inline void verify_osdc_wrlocked(struct ceph_osd_client *osdc) { }
  92 static inline void verify_osd_locked(struct ceph_osd *osd) { }
  93 static inline void verify_lreq_locked(struct ceph_osd_linger_request *lreq) { }
  94 #endif
  95
  96 /*
  97  * calculate the mapping of a file extent onto an object, and fill out the
  98  * request accordingly.  shorten extent as necessary if it crosses an
  99  * object boundary.
 100  *
 101  * fill osd op in request message.
 102  */
 103 static int calc_layout(struct ceph_file_layout *layout, u64 off, u64 *plen,
 104                         u64 *objnum, u64 *objoff, u64 *objlen)
 105 {
 106         u64 orig_len = *plen;
 107         u32 xlen;
 108
 109         /* object extent? */
 110         ceph_calc_file_object_mapping(layout, off, orig_len, objnum,
 111                                           objoff, &xlen);
 112         *objlen = xlen;
 113         if (*objlen < orig_len) {
 114                 *plen = *objlen;
 115                 dout(" skipping last %llu, final file extent %llu~%llu\n",
 116                      orig_len - *plen, off, *plen);
 117         }
 118
 119         dout("calc_layout objnum=%llx %llu~%llu\n", *objnum, *objoff, *objlen);
 120         return 0;
 121 }
 122
 123 static void ceph_osd_data_init(struct ceph_osd_data *osd_data)
 124 {
 125         memset(osd_data, 0, sizeof (*osd_data));
 126         osd_data->type = CEPH_OSD_DATA_TYPE_NONE;
 127 }
 128
 129 /*
 130  * Consumes @pages if @own_pages is true.
 131  */
 132 static void ceph_osd_data_pages_init(struct ceph_osd_data *osd_data,
 133                         struct page **pages, u64 length, u32 alignment,
 134                         bool pages_from_pool, bool own_pages)
 135 {
 136         osd_data->type = CEPH_OSD_DATA_TYPE_PAGES;
 137         osd_data->pages = pages;
 138         osd_data->length = length;
 139         osd_data->alignment = alignment;
 140         osd_data->pages_from_pool = pages_from_pool;
 141         osd_data->own_pages = own_pages;
 142 }
 143
 144 /*
 145  * Consumes a ref on @pagelist.
 146  */
 147 static void ceph_osd_data_pagelist_init(struct ceph_osd_data *osd_data,
 148                         struct ceph_pagelist *pagelist)
 149 {
 150         osd_data->type = CEPH_OSD_DATA_TYPE_PAGELIST;
 151         osd_data->pagelist = pagelist;
 152 }
 153
 154 #ifdef CONFIG_BLOCK
 155 static void ceph_osd_data_bio_init(struct ceph_osd_data *osd_data,
 156                                    struct ceph_bio_iter *bio_pos,
 157                                    u32 bio_length)
 158 {
 159         osd_data->type = CEPH_OSD_DATA_TYPE_BIO;
 160         osd_data->bio_pos = *bio_pos;
 161         osd_data->bio_length = bio_length;
 162 }
 163 #endif /* CONFIG_BLOCK */
 164
 165 static void ceph_osd_data_bvecs_init(struct ceph_osd_data *osd_data,
 166                                      struct ceph_bvec_iter *bvec_pos,
 167                                      u32 num_bvecs)
 168 {
 169         osd_data->type = CEPH_OSD_DATA_TYPE_BVECS;
 170         osd_data->bvec_pos = *bvec_pos;
 171         osd_data->num_bvecs = num_bvecs;
 172 }
 173
 174 static struct ceph_osd_data *
 175 osd_req_op_raw_data_in(struct ceph_osd_request *osd_req, unsigned int which)
 176 {
 177         BUG_ON(which >= osd_req->r_num_ops);
 178
 179         return &osd_req->r_ops[which].raw_data_in;
 180 }
 181
 182 struct ceph_osd_data *
 183 osd_req_op_extent_osd_data(struct ceph_osd_request *osd_req,
 184                         unsigned int which)
 185 {
 186         return osd_req_op_data(osd_req, which, extent, osd_data);
 187 }
 188 EXPORT_SYMBOL(osd_req_op_extent_osd_data);
 189
 190 void osd_req_op_raw_data_in_pages(struct ceph_osd_request *osd_req,
 191                         unsigned int which, struct page **pages,
 192                         u64 length, u32 alignment,
 193                         bool pages_from_pool, bool own_pages)
 194 {
 195         struct ceph_osd_data *osd_data;
 196
 197         osd_data = osd_req_op_raw_data_in(osd_req, which);
 198         ceph_osd_data_pages_init(osd_data, pages, length, alignment,
 199                                 pages_from_pool, own_pages);
 200 }
 201 EXPORT_SYMBOL(osd_req_op_raw_data_in_pages);
 202
 203 void osd_req_op_extent_osd_data_pages(struct ceph_osd_request *osd_req,
 204                         unsigned int which, struct page **pages,
 205                         u64 length, u32 alignment,
 206                         bool pages_from_pool, bool own_pages)
 207 {
 208         struct ceph_osd_data *osd_data;
 209
 210         osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
 211         ceph_osd_data_pages_init(osd_data, pages, length, alignment,
 212                                 pages_from_pool, own_pages);
 213 }
 214 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pages);
 215
 216 void osd_req_op_extent_osd_data_pagelist(struct ceph_osd_request *osd_req,
 217                         unsigned int which, struct ceph_pagelist *pagelist)
 218 {
 219         struct ceph_osd_data *osd_data;
 220
 221         osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
 222         ceph_osd_data_pagelist_init(osd_data, pagelist);
 223 }
 224 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pagelist);
 225
 226 #ifdef CONFIG_BLOCK
 227 void osd_req_op_extent_osd_data_bio(struct ceph_osd_request *osd_req,
 228                                     unsigned int which,
 229                                     struct ceph_bio_iter *bio_pos,
 230                                     u32 bio_length)
 231 {
 232         struct ceph_osd_data *osd_data;
 233
 234         osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
 235         ceph_osd_data_bio_init(osd_data, bio_pos, bio_length);
 236 }
 237 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bio);
 238 #endif /* CONFIG_BLOCK */
 239
 240 void osd_req_op_extent_osd_data_bvecs(struct ceph_osd_request *osd_req,
 241                                       unsigned int which,
 242                                       struct bio_vec *bvecs, u32 num_bvecs,
 243                                       u32 bytes)
 244 {
 245         struct ceph_osd_data *osd_data;
 246         struct ceph_bvec_iter it = {
 247                 .bvecs = bvecs,
 248                 .iter = { .bi_size = bytes },
 249         };
 250
 251         osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
 252         ceph_osd_data_bvecs_init(osd_data, &it, num_bvecs);
 253 }
 254 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bvecs);
 255
 256 void osd_req_op_extent_osd_data_bvec_pos(struct ceph_osd_request *osd_req,
 257                                          unsigned int which,
 258                                          struct ceph_bvec_iter *bvec_pos)
 259 {
 260         struct ceph_osd_data *osd_data;
 261
 262         osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
 263         ceph_osd_data_bvecs_init(osd_data, bvec_pos, 0);
 264 }
 265 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bvec_pos);
 266
 267 static void osd_req_op_cls_request_info_pagelist(
 268                         struct ceph_osd_request *osd_req,
 269                         unsigned int which, struct ceph_pagelist *pagelist)
 270 {
 271         struct ceph_osd_data *osd_data;
 272
 273         osd_data = osd_req_op_data(osd_req, which, cls, request_info);
 274         ceph_osd_data_pagelist_init(osd_data, pagelist);
 275 }
 276
 277 void osd_req_op_cls_request_data_pagelist(
 278                         struct ceph_osd_request *osd_req,
 279                         unsigned int which, struct ceph_pagelist *pagelist)
 280 {
 281         struct ceph_osd_data *osd_data;
 282
 283         osd_data = osd_req_op_data(osd_req, which, cls, request_data);
 284         ceph_osd_data_pagelist_init(osd_data, pagelist);
 285         osd_req->r_ops[which].cls.indata_len += pagelist->length;
 286         osd_req->r_ops[which].indata_len += pagelist->length;
 287 }
 288 EXPORT_SYMBOL(osd_req_op_cls_request_data_pagelist);
 289
 290 void osd_req_op_cls_request_data_pages(struct ceph_osd_request *osd_req,
 291                         unsigned int which, struct page **pages, u64 length,
 292                         u32 alignment, bool pages_from_pool, bool own_pages)
 293 {
 294         struct ceph_osd_data *osd_data;
 295
 296         osd_data = osd_req_op_data(osd_req, which, cls, request_data);
 297         ceph_osd_data_pages_init(osd_data, pages, length, alignment,
 298                                 pages_from_pool, own_pages);
 299         osd_req->r_ops[which].cls.indata_len += length;
 300         osd_req->r_ops[which].indata_len += length;
 301 }
 302 EXPORT_SYMBOL(osd_req_op_cls_request_data_pages);
 303
 304 void osd_req_op_cls_request_data_bvecs(struct ceph_osd_request *osd_req,
 305                                        unsigned int which,
 306                                        struct bio_vec *bvecs, u32 num_bvecs,
 307                                        u32 bytes)
 308 {
 309         struct ceph_osd_data *osd_data;
 310         struct ceph_bvec_iter it = {
 311                 .bvecs = bvecs,
 312                 .iter = { .bi_size = bytes },
 313         };
 314
 315         osd_data = osd_req_op_data(osd_req, which, cls, request_data);
 316         ceph_osd_data_bvecs_init(osd_data, &it, num_bvecs);
 317         osd_req->r_ops[which].cls.indata_len += bytes;
 318         osd_req->r_ops[which].indata_len += bytes;
 319 }
 320 EXPORT_SYMBOL(osd_req_op_cls_request_data_bvecs);
 321
 322 void osd_req_op_cls_response_data_pages(struct ceph_osd_request *osd_req,
 323                         unsigned int which, struct page **pages, u64 length,
 324                         u32 alignment, bool pages_from_pool, bool own_pages)
 325 {
 326         struct ceph_osd_data *osd_data;
 327
 328         osd_data = osd_req_op_data(osd_req, which, cls, response_data);
 329         ceph_osd_data_pages_init(osd_data, pages, length, alignment,
 330                                 pages_from_pool, own_pages);
 331 }
 332 EXPORT_SYMBOL(osd_req_op_cls_response_data_pages);
 333
 334 static u64 ceph_osd_data_length(struct ceph_osd_data *osd_data)
 335 {
 336         switch (osd_data->type) {
 337         case CEPH_OSD_DATA_TYPE_NONE:
 338                 return 0;
 339         case CEPH_OSD_DATA_TYPE_PAGES:
 340                 return osd_data->length;
 341         case CEPH_OSD_DATA_TYPE_PAGELIST:
 342                 return (u64)osd_data->pagelist->length;
 343 #ifdef CONFIG_BLOCK
 344         case CEPH_OSD_DATA_TYPE_BIO:
 345                 return (u64)osd_data->bio_length;
 346 #endif /* CONFIG_BLOCK */
 347         case CEPH_OSD_DATA_TYPE_BVECS:
 348                 return osd_data->bvec_pos.iter.bi_size;
 349         default:
 350                 WARN(true, "unrecognized data type %d\n", (int)osd_data->type);
 351                 return 0;
 352         }
 353 }
 354
 355 static void ceph_osd_data_release(struct ceph_osd_data *osd_data)
 356 {
 357         if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES && osd_data->own_pages) {
 358                 int num_pages;
 359
 360                 num_pages = calc_pages_for((u64)osd_data->alignment,
 361                                                 (u64)osd_data->length);
 362                 ceph_release_page_vector(osd_data->pages, num_pages);
 363         } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
 364                 ceph_pagelist_release(osd_data->pagelist);
 365         }
 366         ceph_osd_data_init(osd_data);
 367 }
 368
 369 static void osd_req_op_data_release(struct ceph_osd_request *osd_req,
 370                         unsigned int which)
 371 {
 372         struct ceph_osd_req_op *op;
 373
 374         BUG_ON(which >= osd_req->r_num_ops);
 375         op = &osd_req->r_ops[which];
 376
 377         switch (op->op) {
 378         case CEPH_OSD_OP_READ:
 379         case CEPH_OSD_OP_WRITE:
 380         case CEPH_OSD_OP_WRITEFULL:
 381                 ceph_osd_data_release(&op->extent.osd_data);
 382                 break;
 383         case CEPH_OSD_OP_CALL:
 384                 ceph_osd_data_release(&op->cls.request_info);
 385                 ceph_osd_data_release(&op->cls.request_data);
 386                 ceph_osd_data_release(&op->cls.response_data);
 387                 break;
 388         case CEPH_OSD_OP_SETXATTR:
 389         case CEPH_OSD_OP_CMPXATTR:
 390                 ceph_osd_data_release(&op->xattr.osd_data);
 391                 break;
 392         case CEPH_OSD_OP_STAT:
 393                 ceph_osd_data_release(&op->raw_data_in);
 394                 break;
 395         case CEPH_OSD_OP_NOTIFY_ACK:
 396                 ceph_osd_data_release(&op->notify_ack.request_data);
 397                 break;
 398         case CEPH_OSD_OP_NOTIFY:
 399                 ceph_osd_data_release(&op->notify.request_data);
 400                 ceph_osd_data_release(&op->notify.response_data);
 401                 break;
 402         case CEPH_OSD_OP_LIST_WATCHERS:
 403                 ceph_osd_data_release(&op->list_watchers.response_data);
 404                 break;
 405         case CEPH_OSD_OP_COPY_FROM2:
 406                 ceph_osd_data_release(&op->copy_from.osd_data);
 407                 break;
 408         default:
 409                 break;
 410         }
 411 }
 412
 413 /*
 414  * Assumes @t is zero-initialized.
 415  */
 416 static void target_init(struct ceph_osd_request_target *t)
 417 {
 418         ceph_oid_init(&t->base_oid);
 419         ceph_oloc_init(&t->base_oloc);
 420         ceph_oid_init(&t->target_oid);
 421         ceph_oloc_init(&t->target_oloc);
 422
 423         ceph_osds_init(&t->acting);
 424         ceph_osds_init(&t->up);
 425         t->size = -1;
 426         t->min_size = -1;
 427
 428         t->osd = CEPH_HOMELESS_OSD;
 429 }
 430
 431 static void target_copy(struct ceph_osd_request_target *dest,
 432                         const struct ceph_osd_request_target *src)
 433 {
 434         ceph_oid_copy(&dest->base_oid, &src->base_oid);
 435         ceph_oloc_copy(&dest->base_oloc, &src->base_oloc);
 436         ceph_oid_copy(&dest->target_oid, &src->target_oid);
 437         ceph_oloc_copy(&dest->target_oloc, &src->target_oloc);
 438
 439         dest->pgid = src->pgid; /* struct */
 440         dest->spgid = src->spgid; /* struct */
 441         dest->pg_num = src->pg_num;
 442         dest->pg_num_mask = src->pg_num_mask;
 443         ceph_osds_copy(&dest->acting, &src->acting);
 444         ceph_osds_copy(&dest->up, &src->up);
 445         dest->size = src->size;
 446         dest->min_size = src->min_size;
 447         dest->sort_bitwise = src->sort_bitwise;
 448         dest->recovery_deletes = src->recovery_deletes;
 449
 450         dest->flags = src->flags;
 451         dest->used_replica = src->used_replica;
 452         dest->paused = src->paused;
 453
 454         dest->epoch = src->epoch;
 455         dest->last_force_resend = src->last_force_resend;
 456
 457         dest->osd = src->osd;
 458 }
 459
 460 static void target_destroy(struct ceph_osd_request_target *t)
 461 {
 462         ceph_oid_destroy(&t->base_oid);
 463         ceph_oloc_destroy(&t->base_oloc);
 464         ceph_oid_destroy(&t->target_oid);
 465         ceph_oloc_destroy(&t->target_oloc);
 466 }
 467
 468 /*
 469  * requests
 470  */
 471 static void request_release_checks(struct ceph_osd_request *req)
 472 {
 473         WARN_ON(!RB_EMPTY_NODE(&req->r_node));
 474         WARN_ON(!RB_EMPTY_NODE(&req->r_mc_node));
 475         WARN_ON(!list_empty(&req->r_private_item));
 476         WARN_ON(req->r_osd);
 477 }
 478
 479 static void ceph_osdc_release_request(struct kref *kref)
 480 {
 481         struct ceph_osd_request *req = container_of(kref,
 482                                             struct ceph_osd_request, r_kref);
 483         unsigned int which;
 484
 485         dout("%s %p (r_request %p r_reply %p)\n", __func__, req,
 486              req->r_request, req->r_reply);
 487         request_release_checks(req);
 488
 489         if (req->r_request)
 490                 ceph_msg_put(req->r_request);
 491         if (req->r_reply)
 492                 ceph_msg_put(req->r_reply);
 493
 494         for (which = 0; which < req->r_num_ops; which++)
 495                 osd_req_op_data_release(req, which);
 496
 497         target_destroy(&req->r_t);
 498         ceph_put_snap_context(req->r_snapc);
 499
 500         if (req->r_mempool)
 501                 mempool_free(req, req->r_osdc->req_mempool);
 502         else if (req->r_num_ops <= CEPH_OSD_SLAB_OPS)
 503                 kmem_cache_free(ceph_osd_request_cache, req);
 504         else
 505                 kfree(req);
 506 }
 507
 508 void ceph_osdc_get_request(struct ceph_osd_request *req)
 509 {
 510         dout("%s %p (was %d)\n", __func__, req,
 511              kref_read(&req->r_kref));
 512         kref_get(&req->r_kref);
 513 }
 514 EXPORT_SYMBOL(ceph_osdc_get_request);
 515
 516 void ceph_osdc_put_request(struct ceph_osd_request *req)
 517 {
 518         if (req) {
 519                 dout("%s %p (was %d)\n", __func__, req,
 520                      kref_read(&req->r_kref));
 521                 kref_put(&req->r_kref, ceph_osdc_release_request);
 522         }
 523 }
 524 EXPORT_SYMBOL(ceph_osdc_put_request);
 525
 526 static void request_init(struct ceph_osd_request *req)
 527 {
 528         /* req only, each op is zeroed in osd_req_op_init() */
 529         memset(req, 0, sizeof(*req));
 530
 531         kref_init(&req->r_kref);
 532         init_completion(&req->r_completion);
 533         RB_CLEAR_NODE(&req->r_node);
 534         RB_CLEAR_NODE(&req->r_mc_node);
 535         INIT_LIST_HEAD(&req->r_private_item);
 536
 537         target_init(&req->r_t);
 538 }
 539
 540 /*
 541  * This is ugly, but it allows us to reuse linger registration and ping
 542  * requests, keeping the structure of the code around send_linger{_ping}()
 543  * reasonable.  Setting up a min_nr=2 mempool for each linger request
 544  * and dealing with copying ops (this blasts req only, watch op remains
 545  * intact) isn't any better.
 546  */
 547 static void request_reinit(struct ceph_osd_request *req)
 548 {
 549         struct ceph_osd_client *osdc = req->r_osdc;
 550         bool mempool = req->r_mempool;
 551         unsigned int num_ops = req->r_num_ops;
 552         u64 snapid = req->r_snapid;
 553         struct ceph_snap_context *snapc = req->r_snapc;
 554         bool linger = req->r_linger;
 555         struct ceph_msg *request_msg = req->r_request;
 556         struct ceph_msg *reply_msg = req->r_reply;
 557
 558         dout("%s req %p\n", __func__, req);
 559         WARN_ON(kref_read(&req->r_kref) != 1);
 560         request_release_checks(req);
 561
 562         WARN_ON(kref_read(&request_msg->kref) != 1);
 563         WARN_ON(kref_read(&reply_msg->kref) != 1);
 564         target_destroy(&req->r_t);
 565
 566         request_init(req);
 567         req->r_osdc = osdc;
 568         req->r_mempool = mempool;
 569         req->r_num_ops = num_ops;
 570         req->r_snapid = snapid;
 571         req->r_snapc = snapc;
 572         req->r_linger = linger;
 573         req->r_request = request_msg;
 574         req->r_reply = reply_msg;
 575 }
 576
 577 struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
 578                                                struct ceph_snap_context *snapc,
 579                                                unsigned int num_ops,
 580                                                bool use_mempool,
 581                                                gfp_t gfp_flags)
 582 {
 583         struct ceph_osd_request *req;
 584
 585         if (use_mempool) {
 586                 BUG_ON(num_ops > CEPH_OSD_SLAB_OPS);
 587                 req = mempool_alloc(osdc->req_mempool, gfp_flags);
 588         } else if (num_ops <= CEPH_OSD_SLAB_OPS) {
 589                 req = kmem_cache_alloc(ceph_osd_request_cache, gfp_flags);
 590         } else {
 591                 BUG_ON(num_ops > CEPH_OSD_MAX_OPS);
 592                 req = kmalloc(struct_size(req, r_ops, num_ops), gfp_flags);
 593         }
 594         if (unlikely(!req))
 595                 return NULL;
 596
 597         request_init(req);
 598         req->r_osdc = osdc;
 599         req->r_mempool = use_mempool;
 600         req->r_num_ops = num_ops;
 601         req->r_snapid = CEPH_NOSNAP;
 602         req->r_snapc = ceph_get_snap_context(snapc);
 603
 604         dout("%s req %p\n", __func__, req);
 605         return req;
 606 }
 607 EXPORT_SYMBOL(ceph_osdc_alloc_request);
 608
 609 static int ceph_oloc_encoding_size(const struct ceph_object_locator *oloc)
 610 {
 611         return 8 + 4 + 4 + 4 + (oloc->pool_ns ? oloc->pool_ns->len : 0);
 612 }
 613
 614 static int __ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp,
 615                                       int num_request_data_items,
 616                                       int num_reply_data_items)
 617 {
 618         struct ceph_osd_client *osdc = req->r_osdc;
 619         struct ceph_msg *msg;
 620         int msg_size;
 621
 622         WARN_ON(req->r_request || req->r_reply);
 623         WARN_ON(ceph_oid_empty(&req->r_base_oid));
 624         WARN_ON(ceph_oloc_empty(&req->r_base_oloc));
 625
 626         /* create request message */
 627         msg_size = CEPH_ENCODING_START_BLK_LEN +
 628                         CEPH_PGID_ENCODING_LEN + 1; /* spgid */
 629         msg_size += 4 + 4 + 4; /* hash, osdmap_epoch, flags */
 630         msg_size += CEPH_ENCODING_START_BLK_LEN +
 631                         sizeof(struct ceph_osd_reqid); /* reqid */
 632         msg_size += sizeof(struct ceph_blkin_trace_info); /* trace */
 633         msg_size += 4 + sizeof(struct ceph_timespec); /* client_inc, mtime */
 634         msg_size += CEPH_ENCODING_START_BLK_LEN +
 635                         ceph_oloc_encoding_size(&req->r_base_oloc); /* oloc */
 636         msg_size += 4 + req->r_base_oid.name_len; /* oid */
 637         msg_size += 2 + req->r_num_ops * sizeof(struct ceph_osd_op);
 638         msg_size += 8; /* snapid */
 639         msg_size += 8; /* snap_seq */
 640         msg_size += 4 + 8 * (req->r_snapc ? req->r_snapc->num_snaps : 0);
 641         msg_size += 4 + 8; /* retry_attempt, features */
 642
 643         if (req->r_mempool)
 644                 msg = ceph_msgpool_get(&osdc->msgpool_op, msg_size,
 645                                        num_request_data_items);
 646         else
 647                 msg = ceph_msg_new2(CEPH_MSG_OSD_OP, msg_size,
 648                                     num_request_data_items, gfp, true);
 649         if (!msg)
 650                 return -ENOMEM;
 651
 652         memset(msg->front.iov_base, 0, msg->front.iov_len);
 653         req->r_request = msg;
 654
 655         /* create reply message */
 656         msg_size = OSD_OPREPLY_FRONT_LEN;
 657         msg_size += req->r_base_oid.name_len;
 658         msg_size += req->r_num_ops * sizeof(struct ceph_osd_op);
 659
 660         if (req->r_mempool)
 661                 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, msg_size,
 662                                        num_reply_data_items);
 663         else
 664                 msg = ceph_msg_new2(CEPH_MSG_OSD_OPREPLY, msg_size,
 665                                     num_reply_data_items, gfp, true);
 666         if (!msg)
 667                 return -ENOMEM;
 668
 669         req->r_reply = msg;
 670
 671         return 0;
 672 }
 673
 674 static bool osd_req_opcode_valid(u16 opcode)
 675 {
 676         switch (opcode) {
 677 #define GENERATE_CASE(op, opcode, str)  case CEPH_OSD_OP_##op: return true;
 678 __CEPH_FORALL_OSD_OPS(GENERATE_CASE)
 679 #undef GENERATE_CASE
 680         default:
 681                 return false;
 682         }
 683 }
 684
 685 static void get_num_data_items(struct ceph_osd_request *req,
 686                                int *num_request_data_items,
 687                                int *num_reply_data_items)
 688 {
 689         struct ceph_osd_req_op *op;
 690
 691         *num_request_data_items = 0;
 692         *num_reply_data_items = 0;
 693
 694         for (op = req->r_ops; op != &req->r_ops[req->r_num_ops]; op++) {
 695                 switch (op->op) {
 696                 /* request */
 697                 case CEPH_OSD_OP_WRITE:
 698                 case CEPH_OSD_OP_WRITEFULL:
 699                 case CEPH_OSD_OP_SETXATTR:
 700                 case CEPH_OSD_OP_CMPXATTR:
 701                 case CEPH_OSD_OP_NOTIFY_ACK:
 702                 case CEPH_OSD_OP_COPY_FROM2:
 703                         *num_request_data_items += 1;
 704                         break;
 705
 706                 /* reply */
 707                 case CEPH_OSD_OP_STAT:
 708                 case CEPH_OSD_OP_READ:
 709                 case CEPH_OSD_OP_LIST_WATCHERS:
 710                         *num_reply_data_items += 1;
 711                         break;
 712
 713                 /* both */
 714                 case CEPH_OSD_OP_NOTIFY:
 715                         *num_request_data_items += 1;
 716                         *num_reply_data_items += 1;
 717                         break;
 718                 case CEPH_OSD_OP_CALL:
 719                         *num_request_data_items += 2;
 720                         *num_reply_data_items += 1;
 721                         break;
 722
 723                 default:
 724                         WARN_ON(!osd_req_opcode_valid(op->op));
 725                         break;
 726                 }
 727         }
 728 }
 729
 730 /*
 731  * oid, oloc and OSD op opcode(s) must be filled in before this function
 732  * is called.
 733  */
 734 int ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp)
 735 {
 736         int num_request_data_items, num_reply_data_items;
 737
 738         get_num_data_items(req, &num_request_data_items, &num_reply_data_items);
 739         return __ceph_osdc_alloc_messages(req, gfp, num_request_data_items,
 740                                           num_reply_data_items);
 741 }
 742 EXPORT_SYMBOL(ceph_osdc_alloc_messages);
 743
 744 /*
 745  * This is an osd op init function for opcodes that have no data or
 746  * other information associated with them.  It also serves as a
 747  * common init routine for all the other init functions, below.
 748  */
 749 struct ceph_osd_req_op *
 750 osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which,
 751                  u16 opcode, u32 flags)
 752 {
 753         struct ceph_osd_req_op *op;
 754
 755         BUG_ON(which >= osd_req->r_num_ops);
 756         BUG_ON(!osd_req_opcode_valid(opcode));
 757
 758         op = &osd_req->r_ops[which];
 759         memset(op, 0, sizeof (*op));
 760         op->op = opcode;
 761         op->flags = flags;
 762
 763         return op;
 764 }
 765 EXPORT_SYMBOL(osd_req_op_init);
 766
 767 void osd_req_op_extent_init(struct ceph_osd_request *osd_req,
 768                                 unsigned int which, u16 opcode,
 769                                 u64 offset, u64 length,
 770                                 u64 truncate_size, u32 truncate_seq)
 771 {
 772         struct ceph_osd_req_op *op = osd_req_op_init(osd_req, which,
 773                                                      opcode, 0);
 774         size_t payload_len = 0;
 775
 776         BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
 777                opcode != CEPH_OSD_OP_WRITEFULL && opcode != CEPH_OSD_OP_ZERO &&
 778                opcode != CEPH_OSD_OP_TRUNCATE);
 779
 780         op->extent.offset = offset;
 781         op->extent.length = length;
 782         op->extent.truncate_size = truncate_size;
 783         op->extent.truncate_seq = truncate_seq;
 784         if (opcode == CEPH_OSD_OP_WRITE || opcode == CEPH_OSD_OP_WRITEFULL)
 785                 payload_len += length;
 786
 787         op->indata_len = payload_len;
 788 }
 789 EXPORT_SYMBOL(osd_req_op_extent_init);
 790
 791 void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
 792                                 unsigned int which, u64 length)
 793 {
 794         struct ceph_osd_req_op *op;
 795         u64 previous;
 796
 797         BUG_ON(which >= osd_req->r_num_ops);
 798         op = &osd_req->r_ops[which];
 799         previous = op->extent.length;
 800
 801         if (length == previous)
 802                 return;         /* Nothing to do */
 803         BUG_ON(length > previous);
 804
 805         op->extent.length = length;
 806         if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL)
 807                 op->indata_len -= previous - length;
 808 }
 809 EXPORT_SYMBOL(osd_req_op_extent_update);
 810
 811 void osd_req_op_extent_dup_last(struct ceph_osd_request *osd_req,
 812                                 unsigned int which, u64 offset_inc)
 813 {
 814         struct ceph_osd_req_op *op, *prev_op;
 815
 816         BUG_ON(which + 1 >= osd_req->r_num_ops);
 817
 818         prev_op = &osd_req->r_ops[which];
 819         op = osd_req_op_init(osd_req, which + 1, prev_op->op, prev_op->flags);
 820         /* dup previous one */
 821         op->indata_len = prev_op->indata_len;
 822         op->outdata_len = prev_op->outdata_len;
 823         op->extent = prev_op->extent;
 824         /* adjust offset */
 825         op->extent.offset += offset_inc;
 826         op->extent.length -= offset_inc;
 827
 828         if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL)
 829                 op->indata_len -= offset_inc;
 830 }
 831 EXPORT_SYMBOL(osd_req_op_extent_dup_last);
 832
 833 int osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
 834                         const char *class, const char *method)
 835 {
 836         struct ceph_osd_req_op *op;
 837         struct ceph_pagelist *pagelist;
 838         size_t payload_len = 0;
 839         size_t size;
 840         int ret;
 841
 842         op = osd_req_op_init(osd_req, which, CEPH_OSD_OP_CALL, 0);
 843
 844         pagelist = ceph_pagelist_alloc(GFP_NOFS);
 845         if (!pagelist)
 846                 return -ENOMEM;
 847
 848         op->cls.class_name = class;
 849         size = strlen(class);
 850         BUG_ON(size > (size_t) U8_MAX);
 851         op->cls.class_len = size;
 852         ret = ceph_pagelist_append(pagelist, class, size);
 853         if (ret)
 854                 goto err_pagelist_free;
 855         payload_len += size;
 856
 857         op->cls.method_name = method;
 858         size = strlen(method);
 859         BUG_ON(size > (size_t) U8_MAX);
 860         op->cls.method_len = size;
 861         ret = ceph_pagelist_append(pagelist, method, size);
 862         if (ret)
 863                 goto err_pagelist_free;
 864         payload_len += size;
 865
 866         osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist);
 867         op->indata_len = payload_len;
 868         return 0;
 869
 870 err_pagelist_free:
 871         ceph_pagelist_release(pagelist);
 872         return ret;
 873 }
 874 EXPORT_SYMBOL(osd_req_op_cls_init);
 875
 876 int osd_req_op_xattr_init(struct ceph_osd_request *osd_req, unsigned int which,
 877                           u16 opcode, const char *name, const void *value,
 878                           size_t size, u8 cmp_op, u8 cmp_mode)
 879 {
 880         struct ceph_osd_req_op *op = osd_req_op_init(osd_req, which,
 881                                                      opcode, 0);
 882         struct ceph_pagelist *pagelist;
 883         size_t payload_len;
 884         int ret;
 885
 886         BUG_ON(opcode != CEPH_OSD_OP_SETXATTR && opcode != CEPH_OSD_OP_CMPXATTR);
 887
 888         pagelist = ceph_pagelist_alloc(GFP_NOFS);
 889         if (!pagelist)
 890                 return -ENOMEM;
 891
 892         payload_len = strlen(name);
 893         op->xattr.name_len = payload_len;
 894         ret = ceph_pagelist_append(pagelist, name, payload_len);
 895         if (ret)
 896                 goto err_pagelist_free;
 897
 898         op->xattr.value_len = size;
 899         ret = ceph_pagelist_append(pagelist, value, size);
 900         if (ret)
 901                 goto err_pagelist_free;
 902         payload_len += size;
 903
 904         op->xattr.cmp_op = cmp_op;
 905         op->xattr.cmp_mode = cmp_mode;
 906
 907         ceph_osd_data_pagelist_init(&op->xattr.osd_data, pagelist);
 908         op->indata_len = payload_len;
 909         return 0;
 910
 911 err_pagelist_free:
 912         ceph_pagelist_release(pagelist);
 913         return ret;
 914 }
 915 EXPORT_SYMBOL(osd_req_op_xattr_init);
 916
 917 /*
 918  * @watch_opcode: CEPH_OSD_WATCH_OP_*
 919  */
 920 static void osd_req_op_watch_init(struct ceph_osd_request *req, int which,
 921                                   u64 cookie, u8 watch_opcode)
 922 {
 923         struct ceph_osd_req_op *op;
 924
 925         op = osd_req_op_init(req, which, CEPH_OSD_OP_WATCH, 0);
 926         op->watch.cookie = cookie;
 927         op->watch.op = watch_opcode;
 928         op->watch.gen = 0;
 929 }
 930
 931 /*
 932  * @flags: CEPH_OSD_OP_ALLOC_HINT_FLAG_*
 933  */
 934 void osd_req_op_alloc_hint_init(struct ceph_osd_request *osd_req,
 935                                 unsigned int which,
 936                                 u64 expected_object_size,
 937                                 u64 expected_write_size,
 938                                 u32 flags)
 939 {
 940         struct ceph_osd_req_op *op;
 941
 942         op = osd_req_op_init(osd_req, which, CEPH_OSD_OP_SETALLOCHINT, 0);
 943         op->alloc_hint.expected_object_size = expected_object_size;
 944         op->alloc_hint.expected_write_size = expected_write_size;
 945         op->alloc_hint.flags = flags;
 946
 947         /*
 948          * CEPH_OSD_OP_SETALLOCHINT op is advisory and therefore deemed
 949          * not worth a feature bit.  Set FAILOK per-op flag to make
 950          * sure older osds don't trip over an unsupported opcode.
 951          */
 952         op->flags |= CEPH_OSD_OP_FLAG_FAILOK;
 953 }
 954 EXPORT_SYMBOL(osd_req_op_alloc_hint_init);
 955
 956 static void ceph_osdc_msg_data_add(struct ceph_msg *msg,
 957                                 struct ceph_osd_data *osd_data)
 958 {
 959         u64 length = ceph_osd_data_length(osd_data);
 960
 961         if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
 962                 BUG_ON(length > (u64) SIZE_MAX);
 963                 if (length)
 964                         ceph_msg_data_add_pages(msg, osd_data->pages,
 965                                         length, osd_data->alignment, false);
 966         } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
 967                 BUG_ON(!length);
 968                 ceph_msg_data_add_pagelist(msg, osd_data->pagelist);
 969 #ifdef CONFIG_BLOCK
 970         } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
 971                 ceph_msg_data_add_bio(msg, &osd_data->bio_pos, length);
 972 #endif
 973         } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BVECS) {
 974                 ceph_msg_data_add_bvecs(msg, &osd_data->bvec_pos);
 975         } else {
 976                 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
 977         }
 978 }
 979
 980 static u32 osd_req_encode_op(struct ceph_osd_op *dst,
 981                              const struct ceph_osd_req_op *src)
 982 {
 983         switch (src->op) {
 984         case CEPH_OSD_OP_STAT:
 985                 break;
 986         case CEPH_OSD_OP_READ:
 987         case CEPH_OSD_OP_WRITE:
 988         case CEPH_OSD_OP_WRITEFULL:
 989         case CEPH_OSD_OP_ZERO:
 990         case CEPH_OSD_OP_TRUNCATE:
 991                 dst->extent.offset = cpu_to_le64(src->extent.offset);
 992                 dst->extent.length = cpu_to_le64(src->extent.length);
 993                 dst->extent.truncate_size =
 994                         cpu_to_le64(src->extent.truncate_size);
 995                 dst->extent.truncate_seq =
 996                         cpu_to_le32(src->extent.truncate_seq);
 997                 break;
 998         case CEPH_OSD_OP_CALL:
 999                 dst->cls.class_len = src->cls.class_len;
1000                 dst->cls.method_len = src->cls.method_len;
1001                 dst->cls.indata_len = cpu_to_le32(src->cls.indata_len);
1002                 break;
1003         case CEPH_OSD_OP_WATCH:
1004                 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
1005                 dst->watch.ver = cpu_to_le64(0);
1006                 dst->watch.op = src->watch.op;
1007                 dst->watch.gen = cpu_to_le32(src->watch.gen);
1008                 break;
1009         case CEPH_OSD_OP_NOTIFY_ACK:
1010                 break;
1011         case CEPH_OSD_OP_NOTIFY:
1012                 dst->notify.cookie = cpu_to_le64(src->notify.cookie);
1013                 break;
1014         case CEPH_OSD_OP_LIST_WATCHERS:
1015                 break;
1016         case CEPH_OSD_OP_SETALLOCHINT:
1017                 dst->alloc_hint.expected_object_size =
1018                     cpu_to_le64(src->alloc_hint.expected_object_size);
1019                 dst->alloc_hint.expected_write_size =
1020                     cpu_to_le64(src->alloc_hint.expected_write_size);
1021                 dst->alloc_hint.flags = cpu_to_le32(src->alloc_hint.flags);
1022                 break;
1023         case CEPH_OSD_OP_SETXATTR:
1024         case CEPH_OSD_OP_CMPXATTR:
1025                 dst->xattr.name_len = cpu_to_le32(src->xattr.name_len);
1026                 dst->xattr.value_len = cpu_to_le32(src->xattr.value_len);
1027                 dst->xattr.cmp_op = src->xattr.cmp_op;
1028                 dst->xattr.cmp_mode = src->xattr.cmp_mode;
1029                 break;
1030         case CEPH_OSD_OP_CREATE:
1031         case CEPH_OSD_OP_DELETE:
1032                 break;
1033         case CEPH_OSD_OP_COPY_FROM2:
1034                 dst->copy_from.snapid = cpu_to_le64(src->copy_from.snapid);
1035                 dst->copy_from.src_version =
1036                         cpu_to_le64(src->copy_from.src_version);
1037                 dst->copy_from.flags = src->copy_from.flags;
1038                 dst->copy_from.src_fadvise_flags =
1039                         cpu_to_le32(src->copy_from.src_fadvise_flags);
1040                 break;
1041         default:
1042                 pr_err("unsupported osd opcode %s\n",
1043                         ceph_osd_op_name(src->op));
1044                 WARN_ON(1);
1045
1046                 return 0;
1047         }
1048
1049         dst->op = cpu_to_le16(src->op);
1050         dst->flags = cpu_to_le32(src->flags);
1051         dst->payload_len = cpu_to_le32(src->indata_len);
1052
1053         return src->indata_len;
1054 }
1055
1056 /*
1057  * build new request AND message, calculate layout, and adjust file
1058  * extent as needed.
1059  *
1060  * if the file was recently truncated, we include information about its
1061  * old and new size so that the object can be updated appropriately.  (we
1062  * avoid synchronously deleting truncated objects because it's slow.)
1063  */
1064 struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
1065                                                struct ceph_file_layout *layout,
1066                                                struct ceph_vino vino,
1067                                                u64 off, u64 *plen,
1068                                                unsigned int which, int num_ops,
1069                                                int opcode, int flags,
1070                                                struct ceph_snap_context *snapc,
1071                                                u32 truncate_seq,
1072                                                u64 truncate_size,
1073                                                bool use_mempool)
1074 {
1075         struct ceph_osd_request *req;
1076         u64 objnum = 0;
1077         u64 objoff = 0;
1078         u64 objlen = 0;
1079         int r;
1080
1081         BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
1082                opcode != CEPH_OSD_OP_ZERO && opcode != CEPH_OSD_OP_TRUNCATE &&
1083                opcode != CEPH_OSD_OP_CREATE && opcode != CEPH_OSD_OP_DELETE);
1084
1085         req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
1086                                         GFP_NOFS);
1087         if (!req) {
1088                 r = -ENOMEM;
1089                 goto fail;
1090         }
1091
1092         /* calculate max write size */
1093         r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
1094         if (r)
1095                 goto fail;
1096
1097         if (opcode == CEPH_OSD_OP_CREATE || opcode == CEPH_OSD_OP_DELETE) {
1098                 osd_req_op_init(req, which, opcode, 0);
1099         } else {
1100                 u32 object_size = layout->object_size;
1101                 u32 object_base = off - objoff;
1102                 if (!(truncate_seq == 1 && truncate_size == -1ULL)) {
1103                         if (truncate_size <= object_base) {
1104                                 truncate_size = 0;
1105                         } else {
1106                                 truncate_size -= object_base;
1107                                 if (truncate_size > object_size)
1108                                         truncate_size = object_size;
1109                         }
1110                 }
1111                 osd_req_op_extent_init(req, which, opcode, objoff, objlen,
1112                                        truncate_size, truncate_seq);
1113         }
1114
1115         req->r_base_oloc.pool = layout->pool_id;
1116         req->r_base_oloc.pool_ns = ceph_try_get_string(layout->pool_ns);
1117         ceph_oid_printf(&req->r_base_oid, "%llx.%08llx", vino.ino, objnum);
1118         req->r_flags = flags | osdc->client->options->read_from_replica;
1119
1120         req->r_snapid = vino.snap;
1121         if (flags & CEPH_OSD_FLAG_WRITE)
1122                 req->r_data_offset = off;
1123
1124         if (num_ops > 1)
1125                 /*
1126                  * This is a special case for ceph_writepages_start(), but it
1127                  * also covers ceph_uninline_data().  If more multi-op request
1128                  * use cases emerge, we will need a separate helper.
1129                  */
1130                 r = __ceph_osdc_alloc_messages(req, GFP_NOFS, num_ops, 0);
1131         else
1132                 r = ceph_osdc_alloc_messages(req, GFP_NOFS);
1133         if (r)
1134                 goto fail;
1135
1136         return req;
1137
1138 fail:
1139         ceph_osdc_put_request(req);
1140         return ERR_PTR(r);
1141 }
1142 EXPORT_SYMBOL(ceph_osdc_new_request);
1143
1144 /*
1145  * We keep osd requests in an rbtree, sorted by ->r_tid.
1146  */
1147 DEFINE_RB_FUNCS(request, struct ceph_osd_request, r_tid, r_node)
1148 DEFINE_RB_FUNCS(request_mc, struct ceph_osd_request, r_tid, r_mc_node)
1149
1150 /*
1151  * Call @fn on each OSD request as long as @fn returns 0.
1152  */
1153 static void for_each_request(struct ceph_osd_client *osdc,
1154                         int (*fn)(struct ceph_osd_request *req, void *arg),
1155                         void *arg)
1156 {
1157         struct rb_node *n, *p;
1158
1159         for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
1160                 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
1161
1162                 for (p = rb_first(&osd->o_requests); p; ) {
1163                         struct ceph_osd_request *req =
1164                             rb_entry(p, struct ceph_osd_request, r_node);
1165
1166                         p = rb_next(p);
1167                         if (fn(req, arg))
1168                                 return;
1169                 }
1170         }
1171
1172         for (p = rb_first(&osdc->homeless_osd.o_requests); p; ) {
1173                 struct ceph_osd_request *req =
1174                     rb_entry(p, struct ceph_osd_request, r_node);
1175
1176                 p = rb_next(p);
1177                 if (fn(req, arg))
1178                         return;
1179         }
1180 }
1181
1182 static bool osd_homeless(struct ceph_osd *osd)
1183 {
1184         return osd->o_osd == CEPH_HOMELESS_OSD;
1185 }
1186
1187 static bool osd_registered(struct ceph_osd *osd)
1188 {
1189         verify_osdc_locked(osd->o_osdc);
1190
1191         return !RB_EMPTY_NODE(&osd->o_node);
1192 }
1193
1194 /*
1195  * Assumes @osd is zero-initialized.
1196  */
1197 static void osd_init(struct ceph_osd *osd)
1198 {
1199         refcount_set(&osd->o_ref, 1);
1200         RB_CLEAR_NODE(&osd->o_node);
1201         osd->o_requests = RB_ROOT;
1202         osd->o_linger_requests = RB_ROOT;
1203         osd->o_backoff_mappings = RB_ROOT;
1204         osd->o_backoffs_by_id = RB_ROOT;
1205         INIT_LIST_HEAD(&osd->o_osd_lru);
1206         INIT_LIST_HEAD(&osd->o_keepalive_item);
1207         osd->o_incarnation = 1;
1208         mutex_init(&osd->lock);
1209 }
1210
1211 static void osd_cleanup(struct ceph_osd *osd)
1212 {
1213         WARN_ON(!RB_EMPTY_NODE(&osd->o_node));
1214         WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests));
1215         WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests));
1216         WARN_ON(!RB_EMPTY_ROOT(&osd->o_backoff_mappings));
1217         WARN_ON(!RB_EMPTY_ROOT(&osd->o_backoffs_by_id));
1218         WARN_ON(!list_empty(&osd->o_osd_lru));
1219         WARN_ON(!list_empty(&osd->o_keepalive_item));
1220
1221         if (osd->o_auth.authorizer) {
1222                 WARN_ON(osd_homeless(osd));
1223                 ceph_auth_destroy_authorizer(osd->o_auth.authorizer);
1224         }
1225 }
1226
1227 /*
1228  * Track open sessions with osds.
1229  */
1230 static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
1231 {
1232         struct ceph_osd *osd;
1233
1234         WARN_ON(onum == CEPH_HOMELESS_OSD);
1235
1236         osd = kzalloc(sizeof(*osd), GFP_NOIO | __GFP_NOFAIL);
1237         osd_init(osd);
1238         osd->o_osdc = osdc;
1239         osd->o_osd = onum;
1240
1241         ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
1242
1243         return osd;
1244 }
1245
1246 static struct ceph_osd *get_osd(struct ceph_osd *osd)
1247 {
1248         if (refcount_inc_not_zero(&osd->o_ref)) {
1249                 dout("get_osd %p %d -> %d\n", osd, refcount_read(&osd->o_ref)-1,
1250                      refcount_read(&osd->o_ref));
1251                 return osd;
1252         } else {
1253                 dout("get_osd %p FAIL\n", osd);
1254                 return NULL;
1255         }
1256 }
1257
1258 static void put_osd(struct ceph_osd *osd)
1259 {
1260         dout("put_osd %p %d -> %d\n", osd, refcount_read(&osd->o_ref),
1261              refcount_read(&osd->o_ref) - 1);
1262         if (refcount_dec_and_test(&osd->o_ref)) {
1263                 osd_cleanup(osd);
1264                 kfree(osd);
1265         }
1266 }
1267
1268 DEFINE_RB_FUNCS(osd, struct ceph_osd, o_osd, o_node)
1269
1270 static void __move_osd_to_lru(struct ceph_osd *osd)
1271 {
1272         struct ceph_osd_client *osdc = osd->o_osdc;
1273
1274         dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1275         BUG_ON(!list_empty(&osd->o_osd_lru));
1276
1277         spin_lock(&osdc->osd_lru_lock);
1278         list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
1279         spin_unlock(&osdc->osd_lru_lock);
1280
1281         osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl;
1282 }
1283
1284 static void maybe_move_osd_to_lru(struct ceph_osd *osd)
1285 {
1286         if (RB_EMPTY_ROOT(&osd->o_requests) &&
1287             RB_EMPTY_ROOT(&osd->o_linger_requests))
1288                 __move_osd_to_lru(osd);
1289 }
1290
1291 static void __remove_osd_from_lru(struct ceph_osd *osd)
1292 {
1293         struct ceph_osd_client *osdc = osd->o_osdc;
1294
1295         dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1296
1297         spin_lock(&osdc->osd_lru_lock);
1298         if (!list_empty(&osd->o_osd_lru))
1299                 list_del_init(&osd->o_osd_lru);
1300         spin_unlock(&osdc->osd_lru_lock);
1301 }
1302
1303 /*
1304  * Close the connection and assign any leftover requests to the
1305  * homeless session.
1306  */
1307 static void close_osd(struct ceph_osd *osd)
1308 {
1309         struct ceph_osd_client *osdc = osd->o_osdc;
1310         struct rb_node *n;
1311
1312         verify_osdc_wrlocked(osdc);
1313         dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1314
1315         ceph_con_close(&osd->o_con);
1316
1317         for (n = rb_first(&osd->o_requests); n; ) {
1318                 struct ceph_osd_request *req =
1319                     rb_entry(n, struct ceph_osd_request, r_node);
1320
1321                 n = rb_next(n); /* unlink_request() */
1322
1323                 dout(" reassigning req %p tid %llu\n", req, req->r_tid);
1324                 unlink_request(osd, req);
1325                 link_request(&osdc->homeless_osd, req);
1326         }
1327         for (n = rb_first(&osd->o_linger_requests); n; ) {
1328                 struct ceph_osd_linger_request *lreq =
1329                     rb_entry(n, struct ceph_osd_linger_request, node);
1330
1331                 n = rb_next(n); /* unlink_linger() */
1332
1333                 dout(" reassigning lreq %p linger_id %llu\n", lreq,
1334                      lreq->linger_id);
1335                 unlink_linger(osd, lreq);
1336                 link_linger(&osdc->homeless_osd, lreq);
1337         }
1338         clear_backoffs(osd);
1339
1340         __remove_osd_from_lru(osd);
1341         erase_osd(&osdc->osds, osd);
1342         put_osd(osd);
1343 }
1344
1345 /*
1346  * reset osd connect
1347  */
1348 static int reopen_osd(struct ceph_osd *osd)
1349 {
1350         struct ceph_entity_addr *peer_addr;
1351
1352         dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1353
1354         if (RB_EMPTY_ROOT(&osd->o_requests) &&
1355             RB_EMPTY_ROOT(&osd->o_linger_requests)) {
1356                 close_osd(osd);
1357                 return -ENODEV;
1358         }
1359
1360         peer_addr = &osd->o_osdc->osdmap->osd_addr[osd->o_osd];
1361         if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1362                         !ceph_con_opened(&osd->o_con)) {
1363                 struct rb_node *n;
1364
1365                 dout("osd addr hasn't changed and connection never opened, "
1366                      "letting msgr retry\n");
1367                 /* touch each r_stamp for handle_timeout()'s benfit */
1368                 for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) {
1369                         struct ceph_osd_request *req =
1370                             rb_entry(n, struct ceph_osd_request, r_node);
1371                         req->r_stamp = jiffies;
1372                 }
1373
1374                 return -EAGAIN;
1375         }
1376
1377         ceph_con_close(&osd->o_con);
1378         ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1379         osd->o_incarnation++;
1380
1381         return 0;
1382 }
1383
1384 static struct ceph_osd *lookup_create_osd(struct ceph_osd_client *osdc, int o,
1385                                           bool wrlocked)
1386 {
1387         struct ceph_osd *osd;
1388
1389         if (wrlocked)
1390                 verify_osdc_wrlocked(osdc);
1391         else
1392                 verify_osdc_locked(osdc);
1393
1394         if (o != CEPH_HOMELESS_OSD)
1395                 osd = lookup_osd(&osdc->osds, o);
1396         else
1397                 osd = &osdc->homeless_osd;
1398         if (!osd) {
1399                 if (!wrlocked)
1400                         return ERR_PTR(-EAGAIN);
1401
1402                 osd = create_osd(osdc, o);
1403                 insert_osd(&osdc->osds, osd);
1404                 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd,
1405                               &osdc->osdmap->osd_addr[osd->o_osd]);
1406         }
1407
1408         dout("%s osdc %p osd%d -> osd %p\n", __func__, osdc, o, osd);
1409         return osd;
1410 }
1411
1412 /*
1413  * Create request <-> OSD session relation.
1414  *
1415  * @req has to be assigned a tid, @osd may be homeless.
1416  */
1417 static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req)
1418 {
1419         verify_osd_locked(osd);
1420         WARN_ON(!req->r_tid || req->r_osd);
1421         dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd,
1422              req, req->r_tid);
1423
1424         if (!osd_homeless(osd))
1425                 __remove_osd_from_lru(osd);
1426         else
1427                 atomic_inc(&osd->o_osdc->num_homeless);
1428
1429         get_osd(osd);
1430         insert_request(&osd->o_requests, req);
1431         req->r_osd = osd;
1432 }
1433
1434 static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req)
1435 {
1436         verify_osd_locked(osd);
1437         WARN_ON(req->r_osd != osd);
1438         dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd,
1439              req, req->r_tid);
1440
1441         req->r_osd = NULL;
1442         erase_request(&osd->o_requests, req);
1443         put_osd(osd);
1444
1445         if (!osd_homeless(osd))
1446                 maybe_move_osd_to_lru(osd);
1447         else
1448                 atomic_dec(&osd->o_osdc->num_homeless);
1449 }
1450
1451 static bool __pool_full(struct ceph_pg_pool_info *pi)
1452 {
1453         return pi->flags & CEPH_POOL_FLAG_FULL;
1454 }
1455
1456 static bool have_pool_full(struct ceph_osd_client *osdc)
1457 {
1458         struct rb_node *n;
1459
1460         for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) {
1461                 struct ceph_pg_pool_info *pi =
1462                     rb_entry(n, struct ceph_pg_pool_info, node);
1463
1464                 if (__pool_full(pi))
1465                         return true;
1466         }
1467
1468         return false;
1469 }
1470
1471 static bool pool_full(struct ceph_osd_client *osdc, s64 pool_id)
1472 {
1473         struct ceph_pg_pool_info *pi;
1474
1475         pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id);
1476         if (!pi)
1477                 return false;
1478
1479         return __pool_full(pi);
1480 }
1481
1482 /*
1483  * Returns whether a request should be blocked from being sent
1484  * based on the current osdmap and osd_client settings.
1485  */
1486 static bool target_should_be_paused(struct ceph_osd_client *osdc,
1487                                     const struct ceph_osd_request_target *t,
1488                                     struct ceph_pg_pool_info *pi)
1489 {
1490         bool pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
1491         bool pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
1492                        ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
1493                        __pool_full(pi);
1494
1495         WARN_ON(pi->id != t->target_oloc.pool);
1496         return ((t->flags & CEPH_OSD_FLAG_READ) && pauserd) ||
1497                ((t->flags & CEPH_OSD_FLAG_WRITE) && pausewr) ||
1498                (osdc->osdmap->epoch < osdc->epoch_barrier);
1499 }
1500
1501 static int pick_random_replica(const struct ceph_osds *acting)
1502 {
1503         int i = prandom_u32() % acting->size;
1504
1505         dout("%s picked osd%d, primary osd%d\n", __func__,
1506              acting->osds[i], acting->primary);
1507         return i;
1508 }
1509
1510 /*
1511  * Picks the closest replica based on client's location given by
1512  * crush_location option.  Prefers the primary if the locality is
1513  * the same.
1514  */
1515 static int pick_closest_replica(struct ceph_osd_client *osdc,
1516                                 const struct ceph_osds *acting)
1517 {
1518         struct ceph_options *opt = osdc->client->options;
1519         int best_i, best_locality;
1520         int i = 0, locality;
1521
1522         do {
1523                 locality = ceph_get_crush_locality(osdc->osdmap,
1524                                                    acting->osds[i],
1525                                                    &opt->crush_locs);
1526                 if (i == 0 ||
1527                     (locality >= 0 && best_locality < 0) ||
1528                     (locality >= 0 && best_locality >= 0 &&
1529                      locality < best_locality)) {
1530                         best_i = i;
1531                         best_locality = locality;
1532                 }
1533         } while (++i < acting->size);
1534
1535         dout("%s picked osd%d with locality %d, primary osd%d\n", __func__,
1536              acting->osds[best_i], best_locality, acting->primary);
1537         return best_i;
1538 }
1539
1540 enum calc_target_result {
1541         CALC_TARGET_NO_ACTION = 0,
1542         CALC_TARGET_NEED_RESEND,
1543         CALC_TARGET_POOL_DNE,
1544 };
1545
1546 static enum calc_target_result calc_target(struct ceph_osd_client *osdc,
1547                                            struct ceph_osd_request_target *t,
1548                                            bool any_change)
1549 {
1550         struct ceph_pg_pool_info *pi;
1551         struct ceph_pg pgid, last_pgid;
1552         struct ceph_osds up, acting;
1553         bool is_read = t->flags & CEPH_OSD_FLAG_READ;
1554         bool is_write = t->flags & CEPH_OSD_FLAG_WRITE;
1555         bool force_resend = false;
1556         bool unpaused = false;
1557         bool legacy_change = false;
1558         bool split = false;
1559         bool sort_bitwise = ceph_osdmap_flag(osdc, CEPH_OSDMAP_SORTBITWISE);
1560         bool recovery_deletes = ceph_osdmap_flag(osdc,
1561                                                  CEPH_OSDMAP_RECOVERY_DELETES);
1562         enum calc_target_result ct_res;
1563
1564         t->epoch = osdc->osdmap->epoch;
1565         pi = ceph_pg_pool_by_id(osdc->osdmap, t->base_oloc.pool);
1566         if (!pi) {
1567                 t->osd = CEPH_HOMELESS_OSD;
1568                 ct_res = CALC_TARGET_POOL_DNE;
1569                 goto out;
1570         }
1571
1572         if (osdc->osdmap->epoch == pi->last_force_request_resend) {
1573                 if (t->last_force_resend < pi->last_force_request_resend) {
1574                         t->last_force_resend = pi->last_force_request_resend;
1575                         force_resend = true;
1576                 } else if (t->last_force_resend == 0) {
1577                         force_resend = true;
1578                 }
1579         }
1580
1581         /* apply tiering */
1582         ceph_oid_copy(&t->target_oid, &t->base_oid);
1583         ceph_oloc_copy(&t->target_oloc, &t->base_oloc);
1584         if ((t->flags & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0) {
1585                 if (is_read && pi->read_tier >= 0)
1586                         t->target_oloc.pool = pi->read_tier;
1587                 if (is_write && pi->write_tier >= 0)
1588                         t->target_oloc.pool = pi->write_tier;
1589
1590                 pi = ceph_pg_pool_by_id(osdc->osdmap, t->target_oloc.pool);
1591                 if (!pi) {
1592                         t->osd = CEPH_HOMELESS_OSD;
1593                         ct_res = CALC_TARGET_POOL_DNE;
1594                         goto out;
1595                 }
1596         }
1597
1598         __ceph_object_locator_to_pg(pi, &t->target_oid, &t->target_oloc, &pgid);
1599         last_pgid.pool = pgid.pool;
1600         last_pgid.seed = ceph_stable_mod(pgid.seed, t->pg_num, t->pg_num_mask);
1601
1602         ceph_pg_to_up_acting_osds(osdc->osdmap, pi, &pgid, &up, &acting);
1603         if (any_change &&
1604             ceph_is_new_interval(&t->acting,
1605                                  &acting,
1606                                  &t->up,
1607                                  &up,
1608                                  t->size,
1609                                  pi->size,
1610                                  t->min_size,
1611                                  pi->min_size,
1612                                  t->pg_num,
1613                                  pi->pg_num,
1614                                  t->sort_bitwise,
1615                                  sort_bitwise,
1616                                  t->recovery_deletes,
1617                                  recovery_deletes,
1618                                  &last_pgid))
1619                 force_resend = true;
1620
1621         if (t->paused && !target_should_be_paused(osdc, t, pi)) {
1622                 t->paused = false;
1623                 unpaused = true;
1624         }
1625         legacy_change = ceph_pg_compare(&t->pgid, &pgid) ||
1626                         ceph_osds_changed(&t->acting, &acting,
1627                                           t->used_replica || any_change);
1628         if (t->pg_num)
1629                 split = ceph_pg_is_split(&last_pgid, t->pg_num, pi->pg_num);
1630
1631         if (legacy_change || force_resend || split) {
1632                 t->pgid = pgid; /* struct */
1633                 ceph_pg_to_primary_shard(osdc->osdmap, pi, &pgid, &t->spgid);
1634                 ceph_osds_copy(&t->acting, &acting);
1635                 ceph_osds_copy(&t->up, &up);
1636                 t->size = pi->size;
1637                 t->min_size = pi->min_size;
1638                 t->pg_num = pi->pg_num;
1639                 t->pg_num_mask = pi->pg_num_mask;
1640                 t->sort_bitwise = sort_bitwise;
1641                 t->recovery_deletes = recovery_deletes;
1642
1643                 if ((t->flags & (CEPH_OSD_FLAG_BALANCE_READS |
1644                                  CEPH_OSD_FLAG_LOCALIZE_READS)) &&
1645                     !is_write && pi->type == CEPH_POOL_TYPE_REP &&
1646                     acting.size > 1) {
1647                         int pos;
1648
1649                         WARN_ON(!is_read || acting.osds[0] != acting.primary);
1650                         if (t->flags & CEPH_OSD_FLAG_BALANCE_READS) {
1651                                 pos = pick_random_replica(&acting);
1652                         } else {
1653                                 pos = pick_closest_replica(osdc, &acting);
1654                         }
1655                         t->osd = acting.osds[pos];
1656                         t->used_replica = pos > 0;
1657                 } else {
1658                         t->osd = acting.primary;
1659                         t->used_replica = false;
1660                 }
1661         }
1662
1663         if (unpaused || legacy_change || force_resend || split)
1664                 ct_res = CALC_TARGET_NEED_RESEND;
1665         else
1666                 ct_res = CALC_TARGET_NO_ACTION;
1667
1668 out:
1669         dout("%s t %p -> %d%d%d%d ct_res %d osd%d\n", __func__, t, unpaused,
1670              legacy_change, force_resend, split, ct_res, t->osd);
1671         return ct_res;
1672 }
1673
1674 static struct ceph_spg_mapping *alloc_spg_mapping(void)
1675 {
1676         struct ceph_spg_mapping *spg;
1677
1678         spg = kmalloc(sizeof(*spg), GFP_NOIO);
1679         if (!spg)
1680                 return NULL;
1681
1682         RB_CLEAR_NODE(&spg->node);
1683         spg->backoffs = RB_ROOT;
1684         return spg;
1685 }
1686
1687 static void free_spg_mapping(struct ceph_spg_mapping *spg)
1688 {
1689         WARN_ON(!RB_EMPTY_NODE(&spg->node));
1690         WARN_ON(!RB_EMPTY_ROOT(&spg->backoffs));
1691
1692         kfree(spg);
1693 }
1694
1695 /*
1696  * rbtree of ceph_spg_mapping for handling map<spg_t, ...>, similar to
1697  * ceph_pg_mapping.  Used to track OSD backoffs -- a backoff [range] is
1698  * defined only within a specific spgid; it does not pass anything to
1699  * children on split, or to another primary.
1700  */
1701 DEFINE_RB_FUNCS2(spg_mapping, struct ceph_spg_mapping, spgid, ceph_spg_compare,
1702                  RB_BYPTR, const struct ceph_spg *, node)
1703
1704 static u64 hoid_get_bitwise_key(const struct ceph_hobject_id *hoid)
1705 {
1706         return hoid->is_max ? 0x100000000ull : hoid->hash_reverse_bits;
1707 }
1708
1709 static void hoid_get_effective_key(const struct ceph_hobject_id *hoid,
1710                                    void **pkey, size_t *pkey_len)
1711 {
1712         if (hoid->key_len) {
1713                 *pkey = hoid->key;
1714                 *pkey_len = hoid->key_len;
1715         } else {
1716                 *pkey = hoid->oid;
1717                 *pkey_len = hoid->oid_len;
1718         }
1719 }
1720
1721 static int compare_names(const void *name1, size_t name1_len,
1722                          const void *name2, size_t name2_len)
1723 {
1724         int ret;
1725
1726         ret = memcmp(name1, name2, min(name1_len, name2_len));
1727         if (!ret) {
1728                 if (name1_len < name2_len)
1729                         ret = -1;
1730                 else if (name1_len > name2_len)
1731                         ret = 1;
1732         }
1733         return ret;
1734 }
1735
1736 static int hoid_compare(const struct ceph_hobject_id *lhs,
1737                         const struct ceph_hobject_id *rhs)
1738 {
1739         void *effective_key1, *effective_key2;
1740         size_t effective_key1_len, effective_key2_len;
1741         int ret;
1742
1743         if (lhs->is_max < rhs->is_max)
1744                 return -1;
1745         if (lhs->is_max > rhs->is_max)
1746                 return 1;
1747
1748         if (lhs->pool < rhs->pool)
1749                 return -1;
1750         if (lhs->pool > rhs->pool)
1751                 return 1;
1752
1753         if (hoid_get_bitwise_key(lhs) < hoid_get_bitwise_key(rhs))
1754                 return -1;
1755         if (hoid_get_bitwise_key(lhs) > hoid_get_bitwise_key(rhs))
1756                 return 1;
1757
1758         ret = compare_names(lhs->nspace, lhs->nspace_len,
1759                             rhs->nspace, rhs->nspace_len);
1760         if (ret)
1761                 return ret;
1762
1763         hoid_get_effective_key(lhs, &effective_key1, &effective_key1_len);
1764         hoid_get_effective_key(rhs, &effective_key2, &effective_key2_len);
1765         ret = compare_names(effective_key1, effective_key1_len,
1766                             effective_key2, effective_key2_len);
1767         if (ret)
1768                 return ret;
1769
1770         ret = compare_names(lhs->oid, lhs->oid_len, rhs->oid, rhs->oid_len);
1771         if (ret)
1772                 return ret;
1773
1774         if (lhs->snapid < rhs->snapid)
1775                 return -1;
1776         if (lhs->snapid > rhs->snapid)
1777                 return 1;
1778
1779         return 0;
1780 }
1781
1782 /*
1783  * For decoding ->begin and ->end of MOSDBackoff only -- no MIN/MAX
1784  * compat stuff here.
1785  *
1786  * Assumes @hoid is zero-initialized.
1787  */
1788 static int decode_hoid(void **p, void *end, struct ceph_hobject_id *hoid)
1789 {
1790         u8 struct_v;
1791         u32 struct_len;
1792         int ret;
1793
1794         ret = ceph_start_decoding(p, end, 4, "hobject_t", &struct_v,
1795                                   &struct_len);
1796         if (ret)
1797                 return ret;
1798
1799         if (struct_v < 4) {
1800                 pr_err("got struct_v %d < 4 of hobject_t\n", struct_v);
1801                 goto e_inval;
1802         }
1803
1804         hoid->key = ceph_extract_encoded_string(p, end, &hoid->key_len,
1805                                                 GFP_NOIO);
1806         if (IS_ERR(hoid->key)) {
1807                 ret = PTR_ERR(hoid->key);
1808                 hoid->key = NULL;
1809                 return ret;
1810         }
1811
1812         hoid->oid = ceph_extract_encoded_string(p, end, &hoid->oid_len,
1813                                                 GFP_NOIO);
1814         if (IS_ERR(hoid->oid)) {
1815                 ret = PTR_ERR(hoid->oid);
1816                 hoid->oid = NULL;
1817                 return ret;
1818         }
1819
1820         ceph_decode_64_safe(p, end, hoid->snapid, e_inval);
1821         ceph_decode_32_safe(p, end, hoid->hash, e_inval);
1822         ceph_decode_8_safe(p, end, hoid->is_max, e_inval);
1823
1824         hoid->nspace = ceph_extract_encoded_string(p, end, &hoid->nspace_len,
1825                                                    GFP_NOIO);
1826         if (IS_ERR(hoid->nspace)) {
1827                 ret = PTR_ERR(hoid->nspace);
1828                 hoid->nspace = NULL;
1829                 return ret;
1830         }
1831
1832         ceph_decode_64_safe(p, end, hoid->pool, e_inval);
1833
1834         ceph_hoid_build_hash_cache(hoid);
1835         return 0;
1836
1837 e_inval:
1838         return -EINVAL;
1839 }
1840
1841 static int hoid_encoding_size(const struct ceph_hobject_id *hoid)
1842 {
1843         return 8 + 4 + 1 + 8 + /* snapid, hash, is_max, pool */
1844                4 + hoid->key_len + 4 + hoid->oid_len + 4 + hoid->nspace_len;
1845 }
1846
1847 static void encode_hoid(void **p, void *end, const struct ceph_hobject_id *hoid)
1848 {
1849         ceph_start_encoding(p, 4, 3, hoid_encoding_size(hoid));
1850         ceph_encode_string(p, end, hoid->key, hoid->key_len);
1851         ceph_encode_string(p, end, hoid->oid, hoid->oid_len);
1852         ceph_encode_64(p, hoid->snapid);
1853         ceph_encode_32(p, hoid->hash);
1854         ceph_encode_8(p, hoid->is_max);
1855         ceph_encode_string(p, end, hoid->nspace, hoid->nspace_len);
1856         ceph_encode_64(p, hoid->pool);
1857 }
1858
1859 static void free_hoid(struct ceph_hobject_id *hoid)
1860 {
1861         if (hoid) {
1862                 kfree(hoid->key);
1863                 kfree(hoid->oid);
1864                 kfree(hoid->nspace);
1865                 kfree(hoid);
1866         }
1867 }
1868
1869 static struct ceph_osd_backoff *alloc_backoff(void)
1870 {
1871         struct ceph_osd_backoff *backoff;
1872
1873         backoff = kzalloc(sizeof(*backoff), GFP_NOIO);
1874         if (!backoff)
1875                 return NULL;
1876
1877         RB_CLEAR_NODE(&backoff->spg_node);
1878         RB_CLEAR_NODE(&backoff->id_node);
1879         return backoff;
1880 }
1881
1882 static void free_backoff(struct ceph_osd_backoff *backoff)
1883 {
1884         WARN_ON(!RB_EMPTY_NODE(&backoff->spg_node));
1885         WARN_ON(!RB_EMPTY_NODE(&backoff->id_node));
1886
1887         free_hoid(backoff->begin);
1888         free_hoid(backoff->end);
1889         kfree(backoff);
1890 }
1891
1892 /*
1893  * Within a specific spgid, backoffs are managed by ->begin hoid.
1894  */
1895 DEFINE_RB_INSDEL_FUNCS2(backoff, struct ceph_osd_backoff, begin, hoid_compare,
1896                         RB_BYVAL, spg_node);
1897
1898 static struct ceph_osd_backoff *lookup_containing_backoff(struct rb_root *root,
1899                                             const struct ceph_hobject_id *hoid)
1900 {
1901         struct rb_node *n = root->rb_node;
1902
1903         while (n) {
1904                 struct ceph_osd_backoff *cur =
1905                     rb_entry(n, struct ceph_osd_backoff, spg_node);
1906                 int cmp;
1907
1908                 cmp = hoid_compare(hoid, cur->begin);
1909                 if (cmp < 0) {
1910                         n = n->rb_left;
1911                 } else if (cmp > 0) {
1912                         if (hoid_compare(hoid, cur->end) < 0)
1913                                 return cur;
1914
1915                         n = n->rb_right;
1916                 } else {
1917                         return cur;
1918                 }
1919         }
1920
1921         return NULL;
1922 }
1923
1924 /*
1925  * Each backoff has a unique id within its OSD session.
1926  */
1927 DEFINE_RB_FUNCS(backoff_by_id, struct ceph_osd_backoff, id, id_node)
1928
1929 static void clear_backoffs(struct ceph_osd *osd)
1930 {
1931         while (!RB_EMPTY_ROOT(&osd->o_backoff_mappings)) {
1932                 struct ceph_spg_mapping *spg =
1933                     rb_entry(rb_first(&osd->o_backoff_mappings),
1934                              struct ceph_spg_mapping, node);
1935
1936                 while (!RB_EMPTY_ROOT(&spg->backoffs)) {
1937                         struct ceph_osd_backoff *backoff =
1938                             rb_entry(rb_first(&spg->backoffs),
1939                                      struct ceph_osd_backoff, spg_node);
1940
1941                         erase_backoff(&spg->backoffs, backoff);
1942                         erase_backoff_by_id(&osd->o_backoffs_by_id, backoff);
1943                         free_backoff(backoff);
1944                 }
1945                 erase_spg_mapping(&osd->o_backoff_mappings, spg);
1946                 free_spg_mapping(spg);
1947         }
1948 }
1949
1950 /*
1951  * Set up a temporary, non-owning view into @t.
1952  */
1953 static void hoid_fill_from_target(struct ceph_hobject_id *hoid,
1954                                   const struct ceph_osd_request_target *t)
1955 {
1956         hoid->key = NULL;
1957         hoid->key_len = 0;
1958         hoid->oid = t->target_oid.name;
1959         hoid->oid_len = t->target_oid.name_len;
1960         hoid->snapid = CEPH_NOSNAP;
1961         hoid->hash = t->pgid.seed;
1962         hoid->is_max = false;
1963         if (t->target_oloc.pool_ns) {
1964                 hoid->nspace = t->target_oloc.pool_ns->str;
1965                 hoid->nspace_len = t->target_oloc.pool_ns->len;
1966         } else {
1967                 hoid->nspace = NULL;
1968                 hoid->nspace_len = 0;
1969         }
1970         hoid->pool = t->target_oloc.pool;
1971         ceph_hoid_build_hash_cache(hoid);
1972 }
1973
1974 static bool should_plug_request(struct ceph_osd_request *req)
1975 {
1976         struct ceph_osd *osd = req->r_osd;
1977         struct ceph_spg_mapping *spg;
1978         struct ceph_osd_backoff *backoff;
1979         struct ceph_hobject_id hoid;
1980
1981         spg = lookup_spg_mapping(&osd->o_backoff_mappings, &req->r_t.spgid);
1982         if (!spg)
1983                 return false;
1984
1985         hoid_fill_from_target(&hoid, &req->r_t);
1986         backoff = lookup_containing_backoff(&spg->backoffs, &hoid);
1987         if (!backoff)
1988                 return false;
1989
1990         dout("%s req %p tid %llu backoff osd%d spgid %llu.%xs%d id %llu\n",
1991              __func__, req, req->r_tid, osd->o_osd, backoff->spgid.pgid.pool,
1992              backoff->spgid.pgid.seed, backoff->spgid.shard, backoff->id);
1993         return true;
1994 }
1995
1996 /*
1997  * Keep get_num_data_items() in sync with this function.
1998  */
1999 static void setup_request_data(struct ceph_osd_request *req)
2000 {
2001         struct ceph_msg *request_msg = req->r_request;
2002         struct ceph_msg *reply_msg = req->r_reply;
2003         struct ceph_osd_req_op *op;
2004
2005         if (req->r_request->num_data_items || req->r_reply->num_data_items)
2006                 return;
2007
2008         WARN_ON(request_msg->data_length || reply_msg->data_length);
2009         for (op = req->r_ops; op != &req->r_ops[req->r_num_ops]; op++) {
2010                 switch (op->op) {
2011                 /* request */
2012                 case CEPH_OSD_OP_WRITE:
2013                 case CEPH_OSD_OP_WRITEFULL:
2014                         WARN_ON(op->indata_len != op->extent.length);
2015                         ceph_osdc_msg_data_add(request_msg,
2016                                                &op->extent.osd_data);
2017                         break;
2018                 case CEPH_OSD_OP_SETXATTR:
2019                 case CEPH_OSD_OP_CMPXATTR:
2020                         WARN_ON(op->indata_len != op->xattr.name_len +
2021                                                   op->xattr.value_len);
2022                         ceph_osdc_msg_data_add(request_msg,
2023                                                &op->xattr.osd_data);
2024                         break;
2025                 case CEPH_OSD_OP_NOTIFY_ACK:
2026                         ceph_osdc_msg_data_add(request_msg,
2027                                                &op->notify_ack.request_data);
2028                         break;
2029                 case CEPH_OSD_OP_COPY_FROM2:
2030                         ceph_osdc_msg_data_add(request_msg,
2031                                                &op->copy_from.osd_data);
2032                         break;
2033
2034                 /* reply */
2035                 case CEPH_OSD_OP_STAT:
2036                         ceph_osdc_msg_data_add(reply_msg,
2037                                                &op->raw_data_in);
2038                         break;
2039                 case CEPH_OSD_OP_READ:
2040                         ceph_osdc_msg_data_add(reply_msg,
2041                                                &op->extent.osd_data);
2042                         break;
2043                 case CEPH_OSD_OP_LIST_WATCHERS:
2044                         ceph_osdc_msg_data_add(reply_msg,
2045                                                &op->list_watchers.response_data);
2046                         break;
2047
2048                 /* both */
2049                 case CEPH_OSD_OP_CALL:
2050                         WARN_ON(op->indata_len != op->cls.class_len +
2051                                                   op->cls.method_len +
2052                                                   op->cls.indata_len);
2053                         ceph_osdc_msg_data_add(request_msg,
2054                                                &op->cls.request_info);
2055                         /* optional, can be NONE */
2056                         ceph_osdc_msg_data_add(request_msg,
2057                                                &op->cls.request_data);
2058                         /* optional, can be NONE */
2059                         ceph_osdc_msg_data_add(reply_msg,
2060                                                &op->cls.response_data);
2061                         break;
2062                 case CEPH_OSD_OP_NOTIFY:
2063                         ceph_osdc_msg_data_add(request_msg,
2064                                                &op->notify.request_data);
2065                         ceph_osdc_msg_data_add(reply_msg,
2066                                                &op->notify.response_data);
2067                         break;
2068                 }
2069         }
2070 }
2071
2072 static void encode_pgid(void **p, const struct ceph_pg *pgid)
2073 {
2074         ceph_encode_8(p, 1);
2075         ceph_encode_64(p, pgid->pool);
2076         ceph_encode_32(p, pgid->seed);
2077         ceph_encode_32(p, -1); /* preferred */
2078 }
2079
2080 static void encode_spgid(void **p, const struct ceph_spg *spgid)
2081 {
2082         ceph_start_encoding(p, 1, 1, CEPH_PGID_ENCODING_LEN + 1);
2083         encode_pgid(p, &spgid->pgid);
2084         ceph_encode_8(p, spgid->shard);
2085 }
2086
2087 static void encode_oloc(void **p, void *end,
2088                         const struct ceph_object_locator *oloc)
2089 {
2090         ceph_start_encoding(p, 5, 4, ceph_oloc_encoding_size(oloc));
2091         ceph_encode_64(p, oloc->pool);
2092         ceph_encode_32(p, -1); /* preferred */
2093         ceph_encode_32(p, 0);  /* key len */
2094         if (oloc->pool_ns)
2095                 ceph_encode_string(p, end, oloc->pool_ns->str,
2096                                    oloc->pool_ns->len);
2097         else
2098                 ceph_encode_32(p, 0);
2099 }
2100
2101 static void encode_request_partial(struct ceph_osd_request *req,
2102                                    struct ceph_msg *msg)
2103 {
2104         void *p = msg->front.iov_base;
2105         void *const end = p + msg->front_alloc_len;
2106         u32 data_len = 0;
2107         int i;
2108
2109         if (req->r_flags & CEPH_OSD_FLAG_WRITE) {
2110                 /* snapshots aren't writeable */
2111                 WARN_ON(req->r_snapid != CEPH_NOSNAP);
2112         } else {
2113                 WARN_ON(req->r_mtime.tv_sec || req->r_mtime.tv_nsec ||
2114                         req->r_data_offset || req->r_snapc);
2115         }
2116
2117         setup_request_data(req);
2118
2119         encode_spgid(&p, &req->r_t.spgid); /* actual spg */
2120         ceph_encode_32(&p, req->r_t.pgid.seed); /* raw hash */
2121         ceph_encode_32(&p, req->r_osdc->osdmap->epoch);
2122         ceph_encode_32(&p, req->r_flags);
2123
2124         /* reqid */
2125         ceph_start_encoding(&p, 2, 2, sizeof(struct ceph_osd_reqid));
2126         memset(p, 0, sizeof(struct ceph_osd_reqid));
2127         p += sizeof(struct ceph_osd_reqid);
2128
2129         /* trace */
2130         memset(p, 0, sizeof(struct ceph_blkin_trace_info));
2131         p += sizeof(struct ceph_blkin_trace_info);
2132
2133         ceph_encode_32(&p, 0); /* client_inc, always 0 */
2134         ceph_encode_timespec64(p, &req->r_mtime);
2135         p += sizeof(struct ceph_timespec);
2136
2137         encode_oloc(&p, end, &req->r_t.target_oloc);
2138         ceph_encode_string(&p, end, req->r_t.target_oid.name,
2139                            req->r_t.target_oid.name_len);
2140
2141         /* ops, can imply data */
2142         ceph_encode_16(&p, req->r_num_ops);
2143         for (i = 0; i < req->r_num_ops; i++) {
2144                 data_len += osd_req_encode_op(p, &req->r_ops[i]);
2145                 p += sizeof(struct ceph_osd_op);
2146         }
2147
2148         ceph_encode_64(&p, req->r_snapid); /* snapid */
2149         if (req->r_snapc) {
2150                 ceph_encode_64(&p, req->r_snapc->seq);
2151                 ceph_encode_32(&p, req->r_snapc->num_snaps);
2152                 for (i = 0; i < req->r_snapc->num_snaps; i++)
2153                         ceph_encode_64(&p, req->r_snapc->snaps[i]);
2154         } else {
2155                 ceph_encode_64(&p, 0); /* snap_seq */
2156                 ceph_encode_32(&p, 0); /* snaps len */
2157         }
2158
2159         ceph_encode_32(&p, req->r_attempts); /* retry_attempt */
2160         BUG_ON(p > end - 8); /* space for features */
2161
2162         msg->hdr.version = cpu_to_le16(8); /* MOSDOp v8 */
2163         /* front_len is finalized in encode_request_finish() */
2164         msg->front.iov_len = p - msg->front.iov_base;
2165         msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2166         msg->hdr.data_len = cpu_to_le32(data_len);
2167         /*
2168          * The header "data_off" is a hint to the receiver allowing it
2169          * to align received data into its buffers such that there's no
2170          * need to re-copy it before writing it to disk (direct I/O).
2171          */
2172         msg->hdr.data_off = cpu_to_le16(req->r_data_offset);
2173
2174         dout("%s req %p msg %p oid %s oid_len %d\n", __func__, req, msg,
2175              req->r_t.target_oid.name, req->r_t.target_oid.name_len);
2176 }
2177
2178 static void encode_request_finish(struct ceph_msg *msg)
2179 {
2180         void *p = msg->front.iov_base;
2181         void *const partial_end = p + msg->front.iov_len;
2182         void *const end = p + msg->front_alloc_len;
2183
2184         if (CEPH_HAVE_FEATURE(msg->con->peer_features, RESEND_ON_SPLIT)) {
2185                 /* luminous OSD -- encode features and be done */
2186                 p = partial_end;
2187                 ceph_encode_64(&p, msg->con->peer_features);
2188         } else {
2189                 struct {
2190                         char spgid[CEPH_ENCODING_START_BLK_LEN +
2191                                    CEPH_PGID_ENCODING_LEN + 1];
2192                         __le32 hash;
2193                         __le32 epoch;
2194                         __le32 flags;
2195                         char reqid[CEPH_ENCODING_START_BLK_LEN +
2196                                    sizeof(struct ceph_osd_reqid)];
2197                         char trace[sizeof(struct ceph_blkin_trace_info)];
2198                         __le32 client_inc;
2199                         struct ceph_timespec mtime;
2200                 } __packed head;
2201                 struct ceph_pg pgid;
2202                 void *oloc, *oid, *tail;
2203                 int oloc_len, oid_len, tail_len;
2204                 int len;
2205
2206                 /*
2207                  * Pre-luminous OSD -- reencode v8 into v4 using @head
2208                  * as a temporary buffer.  Encode the raw PG; the rest
2209                  * is just a matter of moving oloc, oid and tail blobs
2210                  * around.
2211                  */
2212                 memcpy(&head, p, sizeof(head));
2213                 p += sizeof(head);
2214
2215                 oloc = p;
2216                 p += CEPH_ENCODING_START_BLK_LEN;
2217                 pgid.pool = ceph_decode_64(&p);
2218                 p += 4 + 4; /* preferred, key len */
2219                 len = ceph_decode_32(&p);
2220                 p += len;   /* nspace */
2221                 oloc_len = p - oloc;
2222
2223                 oid = p;
2224                 len = ceph_decode_32(&p);
2225                 p += len;
2226                 oid_len = p - oid;
2227
2228                 tail = p;
2229                 tail_len = partial_end - p;
2230
2231                 p = msg->front.iov_base;
2232                 ceph_encode_copy(&p, &head.client_inc, sizeof(head.client_inc));
2233                 ceph_encode_copy(&p, &head.epoch, sizeof(head.epoch));
2234                 ceph_encode_copy(&p, &head.flags, sizeof(head.flags));
2235                 ceph_encode_copy(&p, &head.mtime, sizeof(head.mtime));
2236
2237                 /* reassert_version */
2238                 memset(p, 0, sizeof(struct ceph_eversion));
2239                 p += sizeof(struct ceph_eversion);
2240
2241                 BUG_ON(p >= oloc);
2242                 memmove(p, oloc, oloc_len);
2243                 p += oloc_len;
2244
2245                 pgid.seed = le32_to_cpu(head.hash);
2246                 encode_pgid(&p, &pgid); /* raw pg */
2247
2248                 BUG_ON(p >= oid);
2249                 memmove(p, oid, oid_len);
2250                 p += oid_len;
2251
2252                 /* tail -- ops, snapid, snapc, retry_attempt */
2253                 BUG_ON(p >= tail);
2254                 memmove(p, tail, tail_len);
2255                 p += tail_len;
2256
2257                 msg->hdr.version = cpu_to_le16(4); /* MOSDOp v4 */
2258         }
2259
2260         BUG_ON(p > end);
2261         msg->front.iov_len = p - msg->front.iov_base;
2262         msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2263
2264         dout("%s msg %p tid %llu %u+%u+%u v%d\n", __func__, msg,
2265              le64_to_cpu(msg->hdr.tid), le32_to_cpu(msg->hdr.front_len),
2266              le32_to_cpu(msg->hdr.middle_len), le32_to_cpu(msg->hdr.data_len),
2267              le16_to_cpu(msg->hdr.version));
2268 }
2269
2270 /*
2271  * @req has to be assigned a tid and registered.
2272  */
2273 static void send_request(struct ceph_osd_request *req)
2274 {
2275         struct ceph_osd *osd = req->r_osd;
2276
2277         verify_osd_locked(osd);
2278         WARN_ON(osd->o_osd != req->r_t.osd);
2279
2280         /* backoff? */
2281         if (should_plug_request(req))
2282                 return;
2283
2284         /*
2285          * We may have a previously queued request message hanging
2286          * around.  Cancel it to avoid corrupting the msgr.
2287          */
2288         if (req->r_sent)
2289                 ceph_msg_revoke(req->r_request);
2290
2291         req->r_flags |= CEPH_OSD_FLAG_KNOWN_REDIR;
2292         if (req->r_attempts)
2293                 req->r_flags |= CEPH_OSD_FLAG_RETRY;
2294         else
2295                 WARN_ON(req->r_flags & CEPH_OSD_FLAG_RETRY);
2296
2297         encode_request_partial(req, req->r_request);
2298
2299         dout("%s req %p tid %llu to pgid %llu.%x spgid %llu.%xs%d osd%d e%u flags 0x%x attempt %d\n",
2300              __func__, req, req->r_tid, req->r_t.pgid.pool, req->r_t.pgid.seed,
2301              req->r_t.spgid.pgid.pool, req->r_t.spgid.pgid.seed,
2302              req->r_t.spgid.shard, osd->o_osd, req->r_t.epoch, req->r_flags,
2303              req->r_attempts);
2304
2305         req->r_t.paused = false;
2306         req->r_stamp = jiffies;
2307         req->r_attempts++;
2308
2309         req->r_sent = osd->o_incarnation;
2310         req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
2311         ceph_con_send(&osd->o_con, ceph_msg_get(req->r_request));
2312 }
2313
2314 static void maybe_request_map(struct ceph_osd_client *osdc)
2315 {
2316         bool continuous = false;
2317
2318         verify_osdc_locked(osdc);
2319         WARN_ON(!osdc->osdmap->epoch);
2320
2321         if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
2322             ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD) ||
2323             ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) {
2324                 dout("%s osdc %p continuous\n", __func__, osdc);
2325                 continuous = true;
2326         } else {
2327                 dout("%s osdc %p onetime\n", __func__, osdc);
2328         }
2329
2330         if (ceph_monc_want_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
2331                                osdc->osdmap->epoch + 1, continuous))
2332                 ceph_monc_renew_subs(&osdc->client->monc);
2333 }
2334
2335 static void complete_request(struct ceph_osd_request *req, int err);
2336 static void send_map_check(struct ceph_osd_request *req);
2337
2338 static void __submit_request(struct ceph_osd_request *req, bool wrlocked)
2339 {
2340         struct ceph_osd_client *osdc = req->r_osdc;
2341         struct ceph_osd *osd;
2342         enum calc_target_result ct_res;
2343         int err = 0;
2344         bool need_send = false;
2345         bool promoted = false;
2346
2347         WARN_ON(req->r_tid);
2348         dout("%s req %p wrlocked %d\n", __func__, req, wrlocked);
2349
2350 again:
2351         ct_res = calc_target(osdc, &req->r_t, false);
2352         if (ct_res == CALC_TARGET_POOL_DNE && !wrlocked)
2353                 goto promote;
2354
2355         osd = lookup_create_osd(osdc, req->r_t.osd, wrlocked);
2356         if (IS_ERR(osd)) {
2357                 WARN_ON(PTR_ERR(osd) != -EAGAIN || wrlocked);
2358                 goto promote;
2359         }
2360
2361         if (osdc->abort_err) {
2362                 dout("req %p abort_err %d\n", req, osdc->abort_err);
2363                 err = osdc->abort_err;
2364         } else if (osdc->osdmap->epoch < osdc->epoch_barrier) {
2365                 dout("req %p epoch %u barrier %u\n", req, osdc->osdmap->epoch,
2366                      osdc->epoch_barrier);
2367                 req->r_t.paused = true;
2368                 maybe_request_map(osdc);
2369         } else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
2370                    ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) {
2371                 dout("req %p pausewr\n", req);
2372                 req->r_t.paused = true;
2373                 maybe_request_map(osdc);
2374         } else if ((req->r_flags & CEPH_OSD_FLAG_READ) &&
2375                    ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) {
2376                 dout("req %p pauserd\n", req);
2377                 req->r_t.paused = true;
2378                 maybe_request_map(osdc);
2379         } else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
2380                    !(req->r_flags & (CEPH_OSD_FLAG_FULL_TRY |
2381                                      CEPH_OSD_FLAG_FULL_FORCE)) &&
2382                    (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
2383                     pool_full(osdc, req->r_t.base_oloc.pool))) {
2384                 dout("req %p full/pool_full\n", req);
2385                 if (ceph_test_opt(osdc->client, ABORT_ON_FULL)) {
2386                         err = -ENOSPC;
2387                 } else {
2388                         pr_warn_ratelimited("FULL or reached pool quota\n");
2389                         req->r_t.paused = true;
2390                         maybe_request_map(osdc);
2391                 }
2392         } else if (!osd_homeless(osd)) {
2393                 need_send = true;
2394         } else {
2395                 maybe_request_map(osdc);
2396         }
2397
2398         mutex_lock(&osd->lock);
2399         /*
2400          * Assign the tid atomically with send_request() to protect
2401          * multiple writes to the same object from racing with each
2402          * other, resulting in out of order ops on the OSDs.
2403          */
2404         req->r_tid = atomic64_inc_return(&osdc->last_tid);
2405         link_request(osd, req);
2406         if (need_send)
2407                 send_request(req);
2408         else if (err)
2409                 complete_request(req, err);
2410         mutex_unlock(&osd->lock);
2411
2412         if (!err && ct_res == CALC_TARGET_POOL_DNE)
2413                 send_map_check(req);
2414
2415         if (promoted)
2416                 downgrade_write(&osdc->lock);
2417         return;
2418
2419 promote:
2420         up_read(&osdc->lock);
2421         down_write(&osdc->lock);
2422         wrlocked = true;
2423         promoted = true;
2424         goto again;
2425 }
2426
2427 static void account_request(struct ceph_osd_request *req)
2428 {
2429         WARN_ON(req->r_flags & (CEPH_OSD_FLAG_ACK | CEPH_OSD_FLAG_ONDISK));
2430         WARN_ON(!(req->r_flags & (CEPH_OSD_FLAG_READ | CEPH_OSD_FLAG_WRITE)));
2431
2432         req->r_flags |= CEPH_OSD_FLAG_ONDISK;
2433         atomic_inc(&req->r_osdc->num_requests);
2434
2435         req->r_start_stamp = jiffies;
2436         req->r_start_latency = ktime_get();
2437 }
2438
2439 static void submit_request(struct ceph_osd_request *req, bool wrlocked)
2440 {
2441         ceph_osdc_get_request(req);
2442         account_request(req);
2443         __submit_request(req, wrlocked);
2444 }
2445
2446 static void finish_request(struct ceph_osd_request *req)
2447 {
2448         struct ceph_osd_client *osdc = req->r_osdc;
2449
2450         WARN_ON(lookup_request_mc(&osdc->map_checks, req->r_tid));
2451         dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
2452
2453         req->r_end_latency = ktime_get();
2454
2455         if (req->r_osd)
2456                 unlink_request(req->r_osd, req);
2457         atomic_dec(&osdc->num_requests);
2458
2459         /*
2460          * If an OSD has failed or returned and a request has been sent
2461          * twice, it's possible to get a reply and end up here while the
2462          * request message is queued for delivery.  We will ignore the
2463          * reply, so not a big deal, but better to try and catch it.
2464          */
2465         ceph_msg_revoke(req->r_request);
2466         ceph_msg_revoke_incoming(req->r_reply);
2467 }
2468
2469 static void __complete_request(struct ceph_osd_request *req)
2470 {
2471         dout("%s req %p tid %llu cb %ps result %d\n", __func__, req,
2472              req->r_tid, req->r_callback, req->r_result);
2473
2474         if (req->r_callback)
2475                 req->r_callback(req);
2476         complete_all(&req->r_completion);
2477         ceph_osdc_put_request(req);
2478 }
2479
2480 static void complete_request_workfn(struct work_struct *work)
2481 {
2482         struct ceph_osd_request *req =
2483             container_of(work, struct ceph_osd_request, r_complete_work);
2484
2485         __complete_request(req);
2486 }
2487
2488 /*
2489  * This is open-coded in handle_reply().
2490  */
2491 static void complete_request(struct ceph_osd_request *req, int err)
2492 {
2493         dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err);
2494
2495         req->r_result = err;
2496         finish_request(req);
2497
2498         INIT_WORK(&req->r_complete_work, complete_request_workfn);
2499         queue_work(req->r_osdc->completion_wq, &req->r_complete_work);
2500 }
2501
2502 static void cancel_map_check(struct ceph_osd_request *req)
2503 {
2504         struct ceph_osd_client *osdc = req->r_osdc;
2505         struct ceph_osd_request *lookup_req;
2506
2507         verify_osdc_wrlocked(osdc);
2508
2509         lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid);
2510         if (!lookup_req)
2511                 return;
2512
2513         WARN_ON(lookup_req != req);
2514         erase_request_mc(&osdc->map_checks, req);
2515         ceph_osdc_put_request(req);
2516 }
2517
2518 static void cancel_request(struct ceph_osd_request *req)
2519 {
2520         dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
2521
2522         cancel_map_check(req);
2523         finish_request(req);
2524         complete_all(&req->r_completion);
2525         ceph_osdc_put_request(req);
2526 }
2527
2528 static void abort_request(struct ceph_osd_request *req, int err)
2529 {
2530         dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err);
2531
2532         cancel_map_check(req);
2533         complete_request(req, err);
2534 }
2535
2536 static int abort_fn(struct ceph_osd_request *req, void *arg)
2537 {
2538         int err = *(int *)arg;
2539
2540         abort_request(req, err);
2541         return 0; /* continue iteration */
2542 }
2543
2544 /*
2545  * Abort all in-flight requests with @err and arrange for all future
2546  * requests to be failed immediately.
2547  */
2548 void ceph_osdc_abort_requests(struct ceph_osd_client *osdc, int err)
2549 {
2550         dout("%s osdc %p err %d\n", __func__, osdc, err);
2551         down_write(&osdc->lock);
2552         for_each_request(osdc, abort_fn, &err);
2553         osdc->abort_err = err;
2554         up_write(&osdc->lock);
2555 }
2556 EXPORT_SYMBOL(ceph_osdc_abort_requests);
2557
2558 void ceph_osdc_clear_abort_err(struct ceph_osd_client *osdc)
2559 {
2560         down_write(&osdc->lock);
2561         osdc->abort_err = 0;
2562         up_write(&osdc->lock);
2563 }
2564 EXPORT_SYMBOL(ceph_osdc_clear_abort_err);
2565
2566 static void update_epoch_barrier(struct ceph_osd_client *osdc, u32 eb)
2567 {
2568         if (likely(eb > osdc->epoch_barrier)) {
2569                 dout("updating epoch_barrier from %u to %u\n",
2570                                 osdc->epoch_barrier, eb);
2571                 osdc->epoch_barrier = eb;
2572                 /* Request map if we're not to the barrier yet */
2573                 if (eb > osdc->osdmap->epoch)
2574                         maybe_request_map(osdc);
2575         }
2576 }
2577
2578 void ceph_osdc_update_epoch_barrier(struct ceph_osd_client *osdc, u32 eb)
2579 {
2580         down_read(&osdc->lock);
2581         if (unlikely(eb > osdc->epoch_barrier)) {
2582                 up_read(&osdc->lock);
2583                 down_write(&osdc->lock);
2584                 update_epoch_barrier(osdc, eb);
2585                 up_write(&osdc->lock);
2586         } else {
2587                 up_read(&osdc->lock);
2588         }
2589 }
2590 EXPORT_SYMBOL(ceph_osdc_update_epoch_barrier);
2591
2592 /*
2593  * We can end up releasing caps as a result of abort_request().
2594  * In that case, we probably want to ensure that the cap release message
2595  * has an updated epoch barrier in it, so set the epoch barrier prior to
2596  * aborting the first request.
2597  */
2598 static int abort_on_full_fn(struct ceph_osd_request *req, void *arg)
2599 {
2600         struct ceph_osd_client *osdc = req->r_osdc;
2601         bool *victims = arg;
2602
2603         if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
2604             (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
2605              pool_full(osdc, req->r_t.base_oloc.pool))) {
2606                 if (!*victims) {
2607                         update_epoch_barrier(osdc, osdc->osdmap->epoch);
2608                         *victims = true;
2609                 }
2610                 abort_request(req, -ENOSPC);
2611         }
2612
2613         return 0; /* continue iteration */
2614 }
2615
2616 /*
2617  * Drop all pending requests that are stalled waiting on a full condition to
2618  * clear, and complete them with ENOSPC as the return code. Set the
2619  * osdc->epoch_barrier to the latest map epoch that we've seen if any were
2620  * cancelled.
2621  */
2622 static void ceph_osdc_abort_on_full(struct ceph_osd_client *osdc)
2623 {
2624         bool victims = false;
2625
2626         if (ceph_test_opt(osdc->client, ABORT_ON_FULL) &&
2627             (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || have_pool_full(osdc)))
2628                 for_each_request(osdc, abort_on_full_fn, &victims);
2629 }
2630
2631 static void check_pool_dne(struct ceph_osd_request *req)
2632 {
2633         struct ceph_osd_client *osdc = req->r_osdc;
2634         struct ceph_osdmap *map = osdc->osdmap;
2635
2636         verify_osdc_wrlocked(osdc);
2637         WARN_ON(!map->epoch);
2638
2639         if (req->r_attempts) {
2640                 /*
2641                  * We sent a request earlier, which means that
2642                  * previously the pool existed, and now it does not
2643                  * (i.e., it was deleted).
2644                  */
2645                 req->r_map_dne_bound = map->epoch;
2646                 dout("%s req %p tid %llu pool disappeared\n", __func__, req,
2647                      req->r_tid);
2648         } else {
2649                 dout("%s req %p tid %llu map_dne_bound %u have %u\n", __func__,
2650                      req, req->r_tid, req->r_map_dne_bound, map->epoch);
2651         }
2652
2653         if (req->r_map_dne_bound) {
2654                 if (map->epoch >= req->r_map_dne_bound) {
2655                         /* we had a new enough map */
2656                         pr_info_ratelimited("tid %llu pool does not exist\n",
2657                                             req->r_tid);
2658                         complete_request(req, -ENOENT);
2659                 }
2660         } else {
2661                 send_map_check(req);
2662         }
2663 }
2664
2665 static void map_check_cb(struct ceph_mon_generic_request *greq)
2666 {
2667         struct ceph_osd_client *osdc = &greq->monc->client->osdc;
2668         struct ceph_osd_request *req;
2669         u64 tid = greq->private_data;
2670
2671         WARN_ON(greq->result || !greq->u.newest);
2672
2673         down_write(&osdc->lock);
2674         req = lookup_request_mc(&osdc->map_checks, tid);
2675         if (!req) {
2676                 dout("%s tid %llu dne\n", __func__, tid);
2677                 goto out_unlock;
2678         }
2679
2680         dout("%s req %p tid %llu map_dne_bound %u newest %llu\n", __func__,
2681              req, req->r_tid, req->r_map_dne_bound, greq->u.newest);
2682         if (!req->r_map_dne_bound)
2683                 req->r_map_dne_bound = greq->u.newest;
2684         erase_request_mc(&osdc->map_checks, req);
2685         check_pool_dne(req);
2686
2687         ceph_osdc_put_request(req);
2688 out_unlock:
2689         up_write(&osdc->lock);
2690 }
2691
2692 static void send_map_check(struct ceph_osd_request *req)
2693 {
2694         struct ceph_osd_client *osdc = req->r_osdc;
2695         struct ceph_osd_request *lookup_req;
2696         int ret;
2697
2698         verify_osdc_wrlocked(osdc);
2699
2700         lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid);
2701         if (lookup_req) {
2702                 WARN_ON(lookup_req != req);
2703                 return;
2704         }
2705
2706         ceph_osdc_get_request(req);
2707         insert_request_mc(&osdc->map_checks, req);
2708         ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap",
2709                                           map_check_cb, req->r_tid);
2710         WARN_ON(ret);
2711 }
2712
2713 /*
2714  * lingering requests, watch/notify v2 infrastructure
2715  */
2716 static void linger_release(struct kref *kref)
2717 {
2718         struct ceph_osd_linger_request *lreq =
2719             container_of(kref, struct ceph_osd_linger_request, kref);
2720
2721         dout("%s lreq %p reg_req %p ping_req %p\n", __func__, lreq,
2722              lreq->reg_req, lreq->ping_req);
2723         WARN_ON(!RB_EMPTY_NODE(&lreq->node));
2724         WARN_ON(!RB_EMPTY_NODE(&lreq->osdc_node));
2725         WARN_ON(!RB_EMPTY_NODE(&lreq->mc_node));
2726         WARN_ON(!list_empty(&lreq->scan_item));
2727         WARN_ON(!list_empty(&lreq->pending_lworks));
2728         WARN_ON(lreq->osd);
2729
2730         if (lreq->reg_req)
2731                 ceph_osdc_put_request(lreq->reg_req);
2732         if (lreq->ping_req)
2733                 ceph_osdc_put_request(lreq->ping_req);
2734         target_destroy(&lreq->t);
2735         kfree(lreq);
2736 }
2737
2738 static void linger_put(struct ceph_osd_linger_request *lreq)
2739 {
2740         if (lreq)
2741                 kref_put(&lreq->kref, linger_release);
2742 }
2743
2744 static struct ceph_osd_linger_request *
2745 linger_get(struct ceph_osd_linger_request *lreq)
2746 {
2747         kref_get(&lreq->kref);
2748         return lreq;
2749 }
2750
2751 static struct ceph_osd_linger_request *
2752 linger_alloc(struct ceph_osd_client *osdc)
2753 {
2754         struct ceph_osd_linger_request *lreq;
2755
2756         lreq = kzalloc(sizeof(*lreq), GFP_NOIO);
2757         if (!lreq)
2758                 return NULL;
2759
2760         kref_init(&lreq->kref);
2761         mutex_init(&lreq->lock);
2762         RB_CLEAR_NODE(&lreq->node);
2763         RB_CLEAR_NODE(&lreq->osdc_node);
2764         RB_CLEAR_NODE(&lreq->mc_node);
2765         INIT_LIST_HEAD(&lreq->scan_item);
2766         INIT_LIST_HEAD(&lreq->pending_lworks);
2767         init_completion(&lreq->reg_commit_wait);
2768         init_completion(&lreq->notify_finish_wait);
2769
2770         lreq->osdc = osdc;
2771         target_init(&lreq->t);
2772
2773         dout("%s lreq %p\n", __func__, lreq);
2774         return lreq;
2775 }
2776
2777 DEFINE_RB_INSDEL_FUNCS(linger, struct ceph_osd_linger_request, linger_id, node)
2778 DEFINE_RB_FUNCS(linger_osdc, struct ceph_osd_linger_request, linger_id, osdc_node)
2779 DEFINE_RB_FUNCS(linger_mc, struct ceph_osd_linger_request, linger_id, mc_node)
2780
2781 /*
2782  * Create linger request <-> OSD session relation.
2783  *
2784  * @lreq has to be registered, @osd may be homeless.
2785  */
2786 static void link_linger(struct ceph_osd *osd,
2787                         struct ceph_osd_linger_request *lreq)
2788 {
2789         verify_osd_locked(osd);
2790         WARN_ON(!lreq->linger_id || lreq->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         if (!osd_homeless(osd))
2795                 __remove_osd_from_lru(osd);
2796         else
2797                 atomic_inc(&osd->o_osdc->num_homeless);
2798
2799         get_osd(osd);
2800         insert_linger(&osd->o_linger_requests, lreq);
2801         lreq->osd = osd;
2802 }
2803
2804 static void unlink_linger(struct ceph_osd *osd,
2805                           struct ceph_osd_linger_request *lreq)
2806 {
2807         verify_osd_locked(osd);
2808         WARN_ON(lreq->osd != osd);
2809         dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd,
2810              osd->o_osd, lreq, lreq->linger_id);
2811
2812         lreq->osd = NULL;
2813         erase_linger(&osd->o_linger_requests, lreq);
2814         put_osd(osd);
2815
2816         if (!osd_homeless(osd))
2817                 maybe_move_osd_to_lru(osd);
2818         else
2819                 atomic_dec(&osd->o_osdc->num_homeless);
2820 }
2821
2822 static bool __linger_registered(struct ceph_osd_linger_request *lreq)
2823 {
2824         verify_osdc_locked(lreq->osdc);
2825
2826         return !RB_EMPTY_NODE(&lreq->osdc_node);
2827 }
2828
2829 static bool linger_registered(struct ceph_osd_linger_request *lreq)
2830 {
2831         struct ceph_osd_client *osdc = lreq->osdc;
2832         bool registered;
2833
2834         down_read(&osdc->lock);
2835         registered = __linger_registered(lreq);
2836         up_read(&osdc->lock);
2837
2838         return registered;
2839 }
2840
2841 static void linger_register(struct ceph_osd_linger_request *lreq)
2842 {
2843         struct ceph_osd_client *osdc = lreq->osdc;
2844
2845         verify_osdc_wrlocked(osdc);
2846         WARN_ON(lreq->linger_id);
2847
2848         linger_get(lreq);
2849         lreq->linger_id = ++osdc->last_linger_id;
2850         insert_linger_osdc(&osdc->linger_requests, lreq);
2851 }
2852
2853 static void linger_unregister(struct ceph_osd_linger_request *lreq)
2854 {
2855         struct ceph_osd_client *osdc = lreq->osdc;
2856
2857         verify_osdc_wrlocked(osdc);
2858
2859         erase_linger_osdc(&osdc->linger_requests, lreq);
2860         linger_put(lreq);
2861 }
2862
2863 static void cancel_linger_request(struct ceph_osd_request *req)
2864 {
2865         struct ceph_osd_linger_request *lreq = req->r_priv;
2866
2867         WARN_ON(!req->r_linger);
2868         cancel_request(req);
2869         linger_put(lreq);
2870 }
2871
2872 struct linger_work {
2873         struct work_struct work;
2874         struct ceph_osd_linger_request *lreq;
2875         struct list_head pending_item;
2876         unsigned long queued_stamp;
2877
2878         union {
2879                 struct {
2880                         u64 notify_id;
2881                         u64 notifier_id;
2882                         void *payload; /* points into @msg front */
2883                         size_t payload_len;
2884
2885                         struct ceph_msg *msg; /* for ceph_msg_put() */
2886                 } notify;
2887                 struct {
2888                         int err;
2889                 } error;
2890         };
2891 };
2892
2893 static struct linger_work *lwork_alloc(struct ceph_osd_linger_request *lreq,
2894                                        work_func_t workfn)
2895 {
2896         struct linger_work *lwork;
2897
2898         lwork = kzalloc(sizeof(*lwork), GFP_NOIO);
2899         if (!lwork)
2900                 return NULL;
2901
2902         INIT_WORK(&lwork->work, workfn);
2903         INIT_LIST_HEAD(&lwork->pending_item);
2904         lwork->lreq = linger_get(lreq);
2905
2906         return lwork;
2907 }
2908
2909 static void lwork_free(struct linger_work *lwork)
2910 {
2911         struct ceph_osd_linger_request *lreq = lwork->lreq;
2912
2913         mutex_lock(&lreq->lock);
2914         list_del(&lwork->pending_item);
2915         mutex_unlock(&lreq->lock);
2916
2917         linger_put(lreq);
2918         kfree(lwork);
2919 }
2920
2921 static void lwork_queue(struct linger_work *lwork)
2922 {
2923         struct ceph_osd_linger_request *lreq = lwork->lreq;
2924         struct ceph_osd_client *osdc = lreq->osdc;
2925
2926         verify_lreq_locked(lreq);
2927         WARN_ON(!list_empty(&lwork->pending_item));
2928
2929         lwork->queued_stamp = jiffies;
2930         list_add_tail(&lwork->pending_item, &lreq->pending_lworks);
2931         queue_work(osdc->notify_wq, &lwork->work);
2932 }
2933
2934 static void do_watch_notify(struct work_struct *w)
2935 {
2936         struct linger_work *lwork = container_of(w, struct linger_work, work);
2937         struct ceph_osd_linger_request *lreq = lwork->lreq;
2938
2939         if (!linger_registered(lreq)) {
2940                 dout("%s lreq %p not registered\n", __func__, lreq);
2941                 goto out;
2942         }
2943
2944         WARN_ON(!lreq->is_watch);
2945         dout("%s lreq %p notify_id %llu notifier_id %llu payload_len %zu\n",
2946              __func__, lreq, lwork->notify.notify_id, lwork->notify.notifier_id,
2947              lwork->notify.payload_len);
2948         lreq->wcb(lreq->data, lwork->notify.notify_id, lreq->linger_id,
2949                   lwork->notify.notifier_id, lwork->notify.payload,
2950                   lwork->notify.payload_len);
2951
2952 out:
2953         ceph_msg_put(lwork->notify.msg);
2954         lwork_free(lwork);
2955 }
2956
2957 static void do_watch_error(struct work_struct *w)
2958 {
2959         struct linger_work *lwork = container_of(w, struct linger_work, work);
2960         struct ceph_osd_linger_request *lreq = lwork->lreq;
2961
2962         if (!linger_registered(lreq)) {
2963                 dout("%s lreq %p not registered\n", __func__, lreq);
2964                 goto out;
2965         }
2966
2967         dout("%s lreq %p err %d\n", __func__, lreq, lwork->error.err);
2968         lreq->errcb(lreq->data, lreq->linger_id, lwork->error.err);
2969
2970 out:
2971         lwork_free(lwork);
2972 }
2973
2974 static void queue_watch_error(struct ceph_osd_linger_request *lreq)
2975 {
2976         struct linger_work *lwork;
2977
2978         lwork = lwork_alloc(lreq, do_watch_error);
2979         if (!lwork) {
2980                 pr_err("failed to allocate error-lwork\n");
2981                 return;
2982         }
2983
2984         lwork->error.err = lreq->last_error;
2985         lwork_queue(lwork);
2986 }
2987
2988 static void linger_reg_commit_complete(struct ceph_osd_linger_request *lreq,
2989                                        int result)
2990 {
2991         if (!completion_done(&lreq->reg_commit_wait)) {
2992                 lreq->reg_commit_error = (result <= 0 ? result : 0);
2993                 complete_all(&lreq->reg_commit_wait);
2994         }
2995 }
2996
2997 static void linger_commit_cb(struct ceph_osd_request *req)
2998 {
2999         struct ceph_osd_linger_request *lreq = req->r_priv;
3000
3001         mutex_lock(&lreq->lock);
3002         dout("%s lreq %p linger_id %llu result %d\n", __func__, lreq,
3003              lreq->linger_id, req->r_result);
3004         linger_reg_commit_complete(lreq, req->r_result);
3005         lreq->committed = true;
3006
3007         if (!lreq->is_watch) {
3008                 struct ceph_osd_data *osd_data =
3009                     osd_req_op_data(req, 0, notify, response_data);
3010                 void *p = page_address(osd_data->pages[0]);
3011
3012                 WARN_ON(req->r_ops[0].op != CEPH_OSD_OP_NOTIFY ||
3013                         osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
3014
3015                 /* make note of the notify_id */
3016                 if (req->r_ops[0].outdata_len >= sizeof(u64)) {
3017                         lreq->notify_id = ceph_decode_64(&p);
3018                         dout("lreq %p notify_id %llu\n", lreq,
3019                              lreq->notify_id);
3020                 } else {
3021                         dout("lreq %p no notify_id\n", lreq);
3022                 }
3023         }
3024
3025         mutex_unlock(&lreq->lock);
3026         linger_put(lreq);
3027 }
3028
3029 static int normalize_watch_error(int err)
3030 {
3031         /*
3032          * Translate ENOENT -> ENOTCONN so that a delete->disconnection
3033          * notification and a failure to reconnect because we raced with
3034          * the delete appear the same to the user.
3035          */
3036         if (err == -ENOENT)
3037                 err = -ENOTCONN;
3038
3039         return err;
3040 }
3041
3042 static void linger_reconnect_cb(struct ceph_osd_request *req)
3043 {
3044         struct ceph_osd_linger_request *lreq = req->r_priv;
3045
3046         mutex_lock(&lreq->lock);
3047         dout("%s lreq %p linger_id %llu result %d last_error %d\n", __func__,
3048              lreq, lreq->linger_id, req->r_result, lreq->last_error);
3049         if (req->r_result < 0) {
3050                 if (!lreq->last_error) {
3051                         lreq->last_error = normalize_watch_error(req->r_result);
3052                         queue_watch_error(lreq);
3053                 }
3054         }
3055
3056         mutex_unlock(&lreq->lock);
3057         linger_put(lreq);
3058 }
3059
3060 static void send_linger(struct ceph_osd_linger_request *lreq)
3061 {
3062         struct ceph_osd_request *req = lreq->reg_req;
3063         struct ceph_osd_req_op *op = &req->r_ops[0];
3064
3065         verify_osdc_wrlocked(req->r_osdc);
3066         dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
3067
3068         if (req->r_osd)
3069                 cancel_linger_request(req);
3070
3071         request_reinit(req);
3072         target_copy(&req->r_t, &lreq->t);
3073         req->r_mtime = lreq->mtime;
3074
3075         mutex_lock(&lreq->lock);
3076         if (lreq->is_watch && lreq->committed) {
3077                 WARN_ON(op->op != CEPH_OSD_OP_WATCH ||
3078                         op->watch.cookie != lreq->linger_id);
3079                 op->watch.op = CEPH_OSD_WATCH_OP_RECONNECT;
3080                 op->watch.gen = ++lreq->register_gen;
3081                 dout("lreq %p reconnect register_gen %u\n", lreq,
3082                      op->watch.gen);
3083                 req->r_callback = linger_reconnect_cb;
3084         } else {
3085                 if (!lreq->is_watch)
3086                         lreq->notify_id = 0;
3087                 else
3088                         WARN_ON(op->watch.op != CEPH_OSD_WATCH_OP_WATCH);
3089                 dout("lreq %p register\n", lreq);
3090                 req->r_callback = linger_commit_cb;
3091         }
3092         mutex_unlock(&lreq->lock);
3093
3094         req->r_priv = linger_get(lreq);
3095         req->r_linger = true;
3096
3097         submit_request(req, true);
3098 }
3099
3100 static void linger_ping_cb(struct ceph_osd_request *req)
3101 {
3102         struct ceph_osd_linger_request *lreq = req->r_priv;
3103
3104         mutex_lock(&lreq->lock);
3105         dout("%s lreq %p linger_id %llu result %d ping_sent %lu last_error %d\n",
3106              __func__, lreq, lreq->linger_id, req->r_result, lreq->ping_sent,
3107              lreq->last_error);
3108         if (lreq->register_gen == req->r_ops[0].watch.gen) {
3109                 if (!req->r_result) {
3110                         lreq->watch_valid_thru = lreq->ping_sent;
3111                 } else if (!lreq->last_error) {
3112                         lreq->last_error = normalize_watch_error(req->r_result);
3113                         queue_watch_error(lreq);
3114                 }
3115         } else {
3116                 dout("lreq %p register_gen %u ignoring old pong %u\n", lreq,
3117                      lreq->register_gen, req->r_ops[0].watch.gen);
3118         }
3119
3120         mutex_unlock(&lreq->lock);
3121         linger_put(lreq);
3122 }
3123
3124 static void send_linger_ping(struct ceph_osd_linger_request *lreq)
3125 {
3126         struct ceph_osd_client *osdc = lreq->osdc;
3127         struct ceph_osd_request *req = lreq->ping_req;
3128         struct ceph_osd_req_op *op = &req->r_ops[0];
3129
3130         if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) {
3131                 dout("%s PAUSERD\n", __func__);
3132                 return;
3133         }
3134
3135         lreq->ping_sent = jiffies;
3136         dout("%s lreq %p linger_id %llu ping_sent %lu register_gen %u\n",
3137              __func__, lreq, lreq->linger_id, lreq->ping_sent,
3138              lreq->register_gen);
3139
3140         if (req->r_osd)
3141                 cancel_linger_request(req);
3142
3143         request_reinit(req);
3144         target_copy(&req->r_t, &lreq->t);
3145
3146         WARN_ON(op->op != CEPH_OSD_OP_WATCH ||
3147                 op->watch.cookie != lreq->linger_id ||
3148                 op->watch.op != CEPH_OSD_WATCH_OP_PING);
3149         op->watch.gen = lreq->register_gen;
3150         req->r_callback = linger_ping_cb;
3151         req->r_priv = linger_get(lreq);
3152         req->r_linger = true;
3153
3154         ceph_osdc_get_request(req);
3155         account_request(req);
3156         req->r_tid = atomic64_inc_return(&osdc->last_tid);
3157         link_request(lreq->osd, req);
3158         send_request(req);
3159 }
3160
3161 static void linger_submit(struct ceph_osd_linger_request *lreq)
3162 {
3163         struct ceph_osd_client *osdc = lreq->osdc;
3164         struct ceph_osd *osd;
3165
3166         down_write(&osdc->lock);
3167         linger_register(lreq);
3168         if (lreq->is_watch) {
3169                 lreq->reg_req->r_ops[0].watch.cookie = lreq->linger_id;
3170                 lreq->ping_req->r_ops[0].watch.cookie = lreq->linger_id;
3171         } else {
3172                 lreq->reg_req->r_ops[0].notify.cookie = lreq->linger_id;
3173         }
3174
3175         calc_target(osdc, &lreq->t, false);
3176         osd = lookup_create_osd(osdc, lreq->t.osd, true);
3177         link_linger(osd, lreq);
3178
3179         send_linger(lreq);
3180         up_write(&osdc->lock);
3181 }
3182
3183 static void cancel_linger_map_check(struct ceph_osd_linger_request *lreq)
3184 {
3185         struct ceph_osd_client *osdc = lreq->osdc;
3186         struct ceph_osd_linger_request *lookup_lreq;
3187
3188         verify_osdc_wrlocked(osdc);
3189
3190         lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks,
3191                                        lreq->linger_id);
3192         if (!lookup_lreq)
3193                 return;
3194
3195         WARN_ON(lookup_lreq != lreq);
3196         erase_linger_mc(&osdc->linger_map_checks, lreq);
3197         linger_put(lreq);
3198 }
3199
3200 /*
3201  * @lreq has to be both registered and linked.
3202  */
3203 static void __linger_cancel(struct ceph_osd_linger_request *lreq)
3204 {
3205         if (lreq->is_watch && lreq->ping_req->r_osd)
3206                 cancel_linger_request(lreq->ping_req);
3207         if (lreq->reg_req->r_osd)
3208                 cancel_linger_request(lreq->reg_req);
3209         cancel_linger_map_check(lreq);
3210         unlink_linger(lreq->osd, lreq);
3211         linger_unregister(lreq);
3212 }
3213
3214 static void linger_cancel(struct ceph_osd_linger_request *lreq)
3215 {
3216         struct ceph_osd_client *osdc = lreq->osdc;
3217
3218         down_write(&osdc->lock);
3219         if (__linger_registered(lreq))
3220                 __linger_cancel(lreq);
3221         up_write(&osdc->lock);
3222 }
3223
3224 static void send_linger_map_check(struct ceph_osd_linger_request *lreq);
3225
3226 static void check_linger_pool_dne(struct ceph_osd_linger_request *lreq)
3227 {
3228         struct ceph_osd_client *osdc = lreq->osdc;
3229         struct ceph_osdmap *map = osdc->osdmap;
3230
3231         verify_osdc_wrlocked(osdc);
3232         WARN_ON(!map->epoch);
3233
3234         if (lreq->register_gen) {
3235                 lreq->map_dne_bound = map->epoch;
3236                 dout("%s lreq %p linger_id %llu pool disappeared\n", __func__,
3237                      lreq, lreq->linger_id);
3238         } else {
3239                 dout("%s lreq %p linger_id %llu map_dne_bound %u have %u\n",
3240                      __func__, lreq, lreq->linger_id, lreq->map_dne_bound,
3241                      map->epoch);
3242         }
3243
3244         if (lreq->map_dne_bound) {
3245                 if (map->epoch >= lreq->map_dne_bound) {
3246                         /* we had a new enough map */
3247                         pr_info("linger_id %llu pool does not exist\n",
3248                                 lreq->linger_id);
3249                         linger_reg_commit_complete(lreq, -ENOENT);
3250                         __linger_cancel(lreq);
3251                 }
3252         } else {
3253                 send_linger_map_check(lreq);
3254         }
3255 }
3256
3257 static void linger_map_check_cb(struct ceph_mon_generic_request *greq)
3258 {
3259         struct ceph_osd_client *osdc = &greq->monc->client->osdc;
3260         struct ceph_osd_linger_request *lreq;
3261         u64 linger_id = greq->private_data;
3262
3263         WARN_ON(greq->result || !greq->u.newest);
3264
3265         down_write(&osdc->lock);
3266         lreq = lookup_linger_mc(&osdc->linger_map_checks, linger_id);
3267         if (!lreq) {
3268                 dout("%s linger_id %llu dne\n", __func__, linger_id);
3269                 goto out_unlock;
3270         }
3271
3272         dout("%s lreq %p linger_id %llu map_dne_bound %u newest %llu\n",
3273              __func__, lreq, lreq->linger_id, lreq->map_dne_bound,
3274              greq->u.newest);
3275         if (!lreq->map_dne_bound)
3276                 lreq->map_dne_bound = greq->u.newest;
3277         erase_linger_mc(&osdc->linger_map_checks, lreq);
3278         check_linger_pool_dne(lreq);
3279
3280         linger_put(lreq);
3281 out_unlock:
3282         up_write(&osdc->lock);
3283 }
3284
3285 static void send_linger_map_check(struct ceph_osd_linger_request *lreq)
3286 {
3287         struct ceph_osd_client *osdc = lreq->osdc;
3288         struct ceph_osd_linger_request *lookup_lreq;
3289         int ret;
3290
3291         verify_osdc_wrlocked(osdc);
3292
3293         lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks,
3294                                        lreq->linger_id);
3295         if (lookup_lreq) {
3296                 WARN_ON(lookup_lreq != lreq);
3297                 return;
3298         }
3299
3300         linger_get(lreq);
3301         insert_linger_mc(&osdc->linger_map_checks, lreq);
3302         ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap",
3303                                           linger_map_check_cb, lreq->linger_id);
3304         WARN_ON(ret);
3305 }
3306
3307 static int linger_reg_commit_wait(struct ceph_osd_linger_request *lreq)
3308 {
3309         int ret;
3310
3311         dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
3312         ret = wait_for_completion_interruptible(&lreq->reg_commit_wait);
3313         return ret ?: lreq->reg_commit_error;
3314 }
3315
3316 static int linger_notify_finish_wait(struct ceph_osd_linger_request *lreq)
3317 {
3318         int ret;
3319
3320         dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
3321         ret = wait_for_completion_interruptible(&lreq->notify_finish_wait);
3322         return ret ?: lreq->notify_finish_error;
3323 }
3324
3325 /*
3326  * Timeout callback, called every N seconds.  When 1 or more OSD
3327  * requests has been active for more than N seconds, we send a keepalive
3328  * (tag + timestamp) to its OSD to ensure any communications channel
3329  * reset is detected.
3330  */
3331 static void handle_timeout(struct work_struct *work)
3332 {
3333         struct ceph_osd_client *osdc =
3334                 container_of(work, struct ceph_osd_client, timeout_work.work);
3335         struct ceph_options *opts = osdc->client->options;
3336         unsigned long cutoff = jiffies - opts->osd_keepalive_timeout;
3337         unsigned long expiry_cutoff = jiffies - opts->osd_request_timeout;
3338         LIST_HEAD(slow_osds);
3339         struct rb_node *n, *p;
3340
3341         dout("%s osdc %p\n", __func__, osdc);
3342         down_write(&osdc->lock);
3343
3344         /*
3345          * ping osds that are a bit slow.  this ensures that if there
3346          * is a break in the TCP connection we will notice, and reopen
3347          * a connection with that osd (from the fault callback).
3348          */
3349         for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
3350                 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
3351                 bool found = false;
3352
3353                 for (p = rb_first(&osd->o_requests); p; ) {
3354                         struct ceph_osd_request *req =
3355                             rb_entry(p, struct ceph_osd_request, r_node);
3356
3357                         p = rb_next(p); /* abort_request() */
3358
3359                         if (time_before(req->r_stamp, cutoff)) {
3360                                 dout(" req %p tid %llu on osd%d is laggy\n",
3361                                      req, req->r_tid, osd->o_osd);
3362                                 found = true;
3363                         }
3364                         if (opts->osd_request_timeout &&
3365                             time_before(req->r_start_stamp, expiry_cutoff)) {
3366                                 pr_err_ratelimited("tid %llu on osd%d timeout\n",
3367                                        req->r_tid, osd->o_osd);
3368                                 abort_request(req, -ETIMEDOUT);
3369                         }
3370                 }
3371                 for (p = rb_first(&osd->o_linger_requests); p; p = rb_next(p)) {
3372                         struct ceph_osd_linger_request *lreq =
3373                             rb_entry(p, struct ceph_osd_linger_request, node);
3374
3375                         dout(" lreq %p linger_id %llu is served by osd%d\n",
3376                              lreq, lreq->linger_id, osd->o_osd);
3377                         found = true;
3378
3379                         mutex_lock(&lreq->lock);
3380                         if (lreq->is_watch && lreq->committed && !lreq->last_error)
3381                                 send_linger_ping(lreq);
3382                         mutex_unlock(&lreq->lock);
3383                 }
3384
3385                 if (found)
3386                         list_move_tail(&osd->o_keepalive_item, &slow_osds);
3387         }
3388
3389         if (opts->osd_request_timeout) {
3390                 for (p = rb_first(&osdc->homeless_osd.o_requests); p; ) {
3391                         struct ceph_osd_request *req =
3392                             rb_entry(p, struct ceph_osd_request, r_node);
3393
3394                         p = rb_next(p); /* abort_request() */
3395
3396                         if (time_before(req->r_start_stamp, expiry_cutoff)) {
3397                                 pr_err_ratelimited("tid %llu on osd%d timeout\n",
3398                                        req->r_tid, osdc->homeless_osd.o_osd);
3399                                 abort_request(req, -ETIMEDOUT);
3400                         }
3401                 }
3402         }
3403
3404         if (atomic_read(&osdc->num_homeless) || !list_empty(&slow_osds))
3405                 maybe_request_map(osdc);
3406
3407         while (!list_empty(&slow_osds)) {
3408                 struct ceph_osd *osd = list_first_entry(&slow_osds,
3409                                                         struct ceph_osd,
3410                                                         o_keepalive_item);
3411                 list_del_init(&osd->o_keepalive_item);
3412                 ceph_con_keepalive(&osd->o_con);
3413         }
3414
3415         up_write(&osdc->lock);
3416         schedule_delayed_work(&osdc->timeout_work,
3417                               osdc->client->options->osd_keepalive_timeout);
3418 }
3419
3420 static void handle_osds_timeout(struct work_struct *work)
3421 {
3422         struct ceph_osd_client *osdc =
3423                 container_of(work, struct ceph_osd_client,
3424                              osds_timeout_work.work);
3425         unsigned long delay = osdc->client->options->osd_idle_ttl / 4;
3426         struct ceph_osd *osd, *nosd;
3427
3428         dout("%s osdc %p\n", __func__, osdc);
3429         down_write(&osdc->lock);
3430         list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
3431                 if (time_before(jiffies, osd->lru_ttl))
3432                         break;
3433
3434                 WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests));
3435                 WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests));
3436                 close_osd(osd);
3437         }
3438
3439         up_write(&osdc->lock);
3440         schedule_delayed_work(&osdc->osds_timeout_work,
3441                               round_jiffies_relative(delay));
3442 }
3443
3444 static int ceph_oloc_decode(void **p, void *end,
3445                             struct ceph_object_locator *oloc)
3446 {
3447         u8 struct_v, struct_cv;
3448         u32 len;
3449         void *struct_end;
3450         int ret = 0;
3451
3452         ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
3453         struct_v = ceph_decode_8(p);
3454         struct_cv = ceph_decode_8(p);
3455         if (struct_v < 3) {
3456                 pr_warn("got v %d < 3 cv %d of ceph_object_locator\n",
3457                         struct_v, struct_cv);
3458                 goto e_inval;
3459         }
3460         if (struct_cv > 6) {
3461                 pr_warn("got v %d cv %d > 6 of ceph_object_locator\n",
3462                         struct_v, struct_cv);
3463                 goto e_inval;
3464         }
3465         len = ceph_decode_32(p);
3466         ceph_decode_need(p, end, len, e_inval);
3467         struct_end = *p + len;
3468
3469         oloc->pool = ceph_decode_64(p);
3470         *p += 4; /* skip preferred */
3471
3472         len = ceph_decode_32(p);
3473         if (len > 0) {
3474                 pr_warn("ceph_object_locator::key is set\n");
3475                 goto e_inval;
3476         }
3477
3478         if (struct_v >= 5) {
3479                 bool changed = false;
3480
3481                 len = ceph_decode_32(p);
3482                 if (len > 0) {
3483                         ceph_decode_need(p, end, len, e_inval);
3484                         if (!oloc->pool_ns ||
3485                             ceph_compare_string(oloc->pool_ns, *p, len))
3486                                 changed = true;
3487                         *p += len;
3488                 } else {
3489                         if (oloc->pool_ns)
3490                                 changed = true;
3491                 }
3492                 if (changed) {
3493                         /* redirect changes namespace */
3494                         pr_warn("ceph_object_locator::nspace is changed\n");
3495                         goto e_inval;
3496                 }
3497         }
3498
3499         if (struct_v >= 6) {
3500                 s64 hash = ceph_decode_64(p);
3501                 if (hash != -1) {
3502                         pr_warn("ceph_object_locator::hash is set\n");
3503                         goto e_inval;
3504                 }
3505         }
3506
3507         /* skip the rest */
3508         *p = struct_end;
3509 out:
3510         return ret;
3511
3512 e_inval:
3513         ret = -EINVAL;
3514         goto out;
3515 }
3516
3517 static int ceph_redirect_decode(void **p, void *end,
3518                                 struct ceph_request_redirect *redir)
3519 {
3520         u8 struct_v, struct_cv;
3521         u32 len;
3522         void *struct_end;
3523         int ret;
3524
3525         ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
3526         struct_v = ceph_decode_8(p);
3527         struct_cv = ceph_decode_8(p);
3528         if (struct_cv > 1) {
3529                 pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n",
3530                         struct_v, struct_cv);
3531                 goto e_inval;
3532         }
3533         len = ceph_decode_32(p);
3534         ceph_decode_need(p, end, len, e_inval);
3535         struct_end = *p + len;
3536
3537         ret = ceph_oloc_decode(p, end, &redir->oloc);
3538         if (ret)
3539                 goto out;
3540
3541         len = ceph_decode_32(p);
3542         if (len > 0) {
3543                 pr_warn("ceph_request_redirect::object_name is set\n");
3544                 goto e_inval;
3545         }
3546
3547         /* skip the rest */
3548         *p = struct_end;
3549 out:
3550         return ret;
3551
3552 e_inval:
3553         ret = -EINVAL;
3554         goto out;
3555 }
3556
3557 struct MOSDOpReply {
3558         struct ceph_pg pgid;
3559         u64 flags;
3560         int result;
3561         u32 epoch;
3562         int num_ops;
3563         u32 outdata_len[CEPH_OSD_MAX_OPS];
3564         s32 rval[CEPH_OSD_MAX_OPS];
3565         int retry_attempt;
3566         struct ceph_eversion replay_version;
3567         u64 user_version;
3568         struct ceph_request_redirect redirect;
3569 };
3570
3571 static int decode_MOSDOpReply(const struct ceph_msg *msg, struct MOSDOpReply *m)
3572 {
3573         void *p = msg->front.iov_base;
3574         void *const end = p + msg->front.iov_len;
3575         u16 version = le16_to_cpu(msg->hdr.version);
3576         struct ceph_eversion bad_replay_version;
3577         u8 decode_redir;
3578         u32 len;
3579         int ret;
3580         int i;
3581
3582         ceph_decode_32_safe(&p, end, len, e_inval);
3583         ceph_decode_need(&p, end, len, e_inval);
3584         p += len; /* skip oid */
3585
3586         ret = ceph_decode_pgid(&p, end, &m->pgid);
3587         if (ret)
3588                 return ret;
3589
3590         ceph_decode_64_safe(&p, end, m->flags, e_inval);
3591         ceph_decode_32_safe(&p, end, m->result, e_inval);
3592         ceph_decode_need(&p, end, sizeof(bad_replay_version), e_inval);
3593         memcpy(&bad_replay_version, p, sizeof(bad_replay_version));
3594         p += sizeof(bad_replay_version);
3595         ceph_decode_32_safe(&p, end, m->epoch, e_inval);
3596
3597         ceph_decode_32_safe(&p, end, m->num_ops, e_inval);
3598         if (m->num_ops > ARRAY_SIZE(m->outdata_len))
3599                 goto e_inval;
3600
3601         ceph_decode_need(&p, end, m->num_ops * sizeof(struct ceph_osd_op),
3602                          e_inval);
3603         for (i = 0; i < m->num_ops; i++) {
3604                 struct ceph_osd_op *op = p;
3605
3606                 m->outdata_len[i] = le32_to_cpu(op->payload_len);
3607                 p += sizeof(*op);
3608         }
3609
3610         ceph_decode_32_safe(&p, end, m->retry_attempt, e_inval);
3611         for (i = 0; i < m->num_ops; i++)
3612                 ceph_decode_32_safe(&p, end, m->rval[i], e_inval);
3613
3614         if (version >= 5) {
3615                 ceph_decode_need(&p, end, sizeof(m->replay_version), e_inval);
3616                 memcpy(&m->replay_version, p, sizeof(m->replay_version));
3617                 p += sizeof(m->replay_version);
3618                 ceph_decode_64_safe(&p, end, m->user_version, e_inval);
3619         } else {
3620                 m->replay_version = bad_replay_version; /* struct */
3621                 m->user_version = le64_to_cpu(m->replay_version.version);
3622         }
3623
3624         if (version >= 6) {
3625                 if (version >= 7)
3626                         ceph_decode_8_safe(&p, end, decode_redir, e_inval);
3627                 else
3628                         decode_redir = 1;
3629         } else {
3630                 decode_redir = 0;
3631         }
3632
3633         if (decode_redir) {
3634                 ret = ceph_redirect_decode(&p, end, &m->redirect);
3635                 if (ret)
3636                         return ret;
3637         } else {
3638                 ceph_oloc_init(&m->redirect.oloc);
3639         }
3640
3641         return 0;
3642
3643 e_inval:
3644         return -EINVAL;
3645 }
3646
3647 /*
3648  * Handle MOSDOpReply.  Set ->r_result and call the callback if it is
3649  * specified.
3650  */
3651 static void handle_reply(struct ceph_osd *osd, struct ceph_msg *msg)
3652 {
3653         struct ceph_osd_client *osdc = osd->o_osdc;
3654         struct ceph_osd_request *req;
3655         struct MOSDOpReply m;
3656         u64 tid = le64_to_cpu(msg->hdr.tid);
3657         u32 data_len = 0;
3658         int ret;
3659         int i;
3660
3661         dout("%s msg %p tid %llu\n", __func__, msg, tid);
3662
3663         down_read(&osdc->lock);
3664         if (!osd_registered(osd)) {
3665                 dout("%s osd%d unknown\n", __func__, osd->o_osd);
3666                 goto out_unlock_osdc;
3667         }
3668         WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num));
3669
3670         mutex_lock(&osd->lock);
3671         req = lookup_request(&osd->o_requests, tid);
3672         if (!req) {
3673                 dout("%s osd%d tid %llu unknown\n", __func__, osd->o_osd, tid);
3674                 goto out_unlock_session;
3675         }
3676
3677         m.redirect.oloc.pool_ns = req->r_t.target_oloc.pool_ns;
3678         ret = decode_MOSDOpReply(msg, &m);
3679         m.redirect.oloc.pool_ns = NULL;
3680         if (ret) {
3681                 pr_err("failed to decode MOSDOpReply for tid %llu: %d\n",
3682                        req->r_tid, ret);
3683                 ceph_msg_dump(msg);
3684                 goto fail_request;
3685         }
3686         dout("%s req %p tid %llu flags 0x%llx pgid %llu.%x epoch %u attempt %d v %u'%llu uv %llu\n",
3687              __func__, req, req->r_tid, m.flags, m.pgid.pool, m.pgid.seed,
3688              m.epoch, m.retry_attempt, le32_to_cpu(m.replay_version.epoch),
3689              le64_to_cpu(m.replay_version.version), m.user_version);
3690
3691         if (m.retry_attempt >= 0) {
3692                 if (m.retry_attempt != req->r_attempts - 1) {
3693                         dout("req %p tid %llu retry_attempt %d != %d, ignoring\n",
3694                              req, req->r_tid, m.retry_attempt,
3695                              req->r_attempts - 1);
3696                         goto out_unlock_session;
3697                 }
3698         } else {
3699                 WARN_ON(1); /* MOSDOpReply v4 is assumed */
3700         }
3701
3702         if (!ceph_oloc_empty(&m.redirect.oloc)) {
3703                 dout("req %p tid %llu redirect pool %lld\n", req, req->r_tid,
3704                      m.redirect.oloc.pool);
3705                 unlink_request(osd, req);
3706                 mutex_unlock(&osd->lock);
3707
3708                 /*
3709                  * Not ceph_oloc_copy() - changing pool_ns is not
3710                  * supported.
3711                  */
3712                 req->r_t.target_oloc.pool = m.redirect.oloc.pool;
3713                 req->r_flags |= CEPH_OSD_FLAG_REDIRECTED |
3714                                 CEPH_OSD_FLAG_IGNORE_OVERLAY |
3715                                 CEPH_OSD_FLAG_IGNORE_CACHE;
3716                 req->r_tid = 0;
3717                 __submit_request(req, false);
3718                 goto out_unlock_osdc;
3719         }
3720
3721         if (m.result == -EAGAIN) {
3722                 dout("req %p tid %llu EAGAIN\n", req, req->r_tid);
3723                 unlink_request(osd, req);
3724                 mutex_unlock(&osd->lock);
3725
3726                 /*
3727                  * The object is missing on the replica or not (yet)
3728                  * readable.  Clear pgid to force a resend to the primary
3729                  * via legacy_change.
3730                  */
3731                 req->r_t.pgid.pool = 0;
3732                 req->r_t.pgid.seed = 0;
3733                 WARN_ON(!req->r_t.used_replica);
3734                 req->r_flags &= ~(CEPH_OSD_FLAG_BALANCE_READS |
3735                                   CEPH_OSD_FLAG_LOCALIZE_READS);
3736                 req->r_tid = 0;
3737                 __submit_request(req, false);
3738                 goto out_unlock_osdc;
3739         }
3740
3741         if (m.num_ops != req->r_num_ops) {
3742                 pr_err("num_ops %d != %d for tid %llu\n", m.num_ops,
3743                        req->r_num_ops, req->r_tid);
3744                 goto fail_request;
3745         }
3746         for (i = 0; i < req->r_num_ops; i++) {
3747                 dout(" req %p tid %llu op %d rval %d len %u\n", req,
3748                      req->r_tid, i, m.rval[i], m.outdata_len[i]);
3749                 req->r_ops[i].rval = m.rval[i];
3750                 req->r_ops[i].outdata_len = m.outdata_len[i];
3751                 data_len += m.outdata_len[i];
3752         }
3753         if (data_len != le32_to_cpu(msg->hdr.data_len)) {
3754                 pr_err("sum of lens %u != %u for tid %llu\n", data_len,
3755                        le32_to_cpu(msg->hdr.data_len), req->r_tid);
3756                 goto fail_request;
3757         }
3758         dout("%s req %p tid %llu result %d data_len %u\n", __func__,
3759              req, req->r_tid, m.result, data_len);
3760
3761         /*
3762          * Since we only ever request ONDISK, we should only ever get
3763          * one (type of) reply back.
3764          */
3765         WARN_ON(!(m.flags & CEPH_OSD_FLAG_ONDISK));
3766         req->r_result = m.result ?: data_len;
3767         finish_request(req);
3768         mutex_unlock(&osd->lock);
3769         up_read(&osdc->lock);
3770
3771         __complete_request(req);
3772         return;
3773
3774 fail_request:
3775         complete_request(req, -EIO);
3776 out_unlock_session:
3777         mutex_unlock(&osd->lock);
3778 out_unlock_osdc:
3779         up_read(&osdc->lock);
3780 }
3781
3782 static void set_pool_was_full(struct ceph_osd_client *osdc)
3783 {
3784         struct rb_node *n;
3785
3786         for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) {
3787                 struct ceph_pg_pool_info *pi =
3788                     rb_entry(n, struct ceph_pg_pool_info, node);
3789
3790                 pi->was_full = __pool_full(pi);
3791         }
3792 }
3793
3794 static bool pool_cleared_full(struct ceph_osd_client *osdc, s64 pool_id)
3795 {
3796         struct ceph_pg_pool_info *pi;
3797
3798         pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id);
3799         if (!pi)
3800                 return false;
3801
3802         return pi->was_full && !__pool_full(pi);
3803 }
3804
3805 static enum calc_target_result
3806 recalc_linger_target(struct ceph_osd_linger_request *lreq)
3807 {
3808         struct ceph_osd_client *osdc = lreq->osdc;
3809         enum calc_target_result ct_res;
3810
3811         ct_res = calc_target(osdc, &lreq->t, true);
3812         if (ct_res == CALC_TARGET_NEED_RESEND) {
3813                 struct ceph_osd *osd;
3814
3815                 osd = lookup_create_osd(osdc, lreq->t.osd, true);
3816                 if (osd != lreq->osd) {
3817                         unlink_linger(lreq->osd, lreq);
3818                         link_linger(osd, lreq);
3819                 }
3820         }
3821
3822         return ct_res;
3823 }
3824
3825 /*
3826  * Requeue requests whose mapping to an OSD has changed.
3827  */
3828 static void scan_requests(struct ceph_osd *osd,
3829                           bool force_resend,
3830                           bool cleared_full,
3831                           bool check_pool_cleared_full,
3832                           struct rb_root *need_resend,
3833                           struct list_head *need_resend_linger)
3834 {
3835         struct ceph_osd_client *osdc = osd->o_osdc;
3836         struct rb_node *n;
3837         bool force_resend_writes;
3838
3839         for (n = rb_first(&osd->o_linger_requests); n; ) {
3840                 struct ceph_osd_linger_request *lreq =
3841                     rb_entry(n, struct ceph_osd_linger_request, node);
3842                 enum calc_target_result ct_res;
3843
3844                 n = rb_next(n); /* recalc_linger_target() */
3845
3846                 dout("%s lreq %p linger_id %llu\n", __func__, lreq,
3847                      lreq->linger_id);
3848                 ct_res = recalc_linger_target(lreq);
3849                 switch (ct_res) {
3850                 case CALC_TARGET_NO_ACTION:
3851                         force_resend_writes = cleared_full ||
3852                             (check_pool_cleared_full &&
3853                              pool_cleared_full(osdc, lreq->t.base_oloc.pool));
3854                         if (!force_resend && !force_resend_writes)
3855                                 break;
3856
3857                         fallthrough;
3858                 case CALC_TARGET_NEED_RESEND:
3859                         cancel_linger_map_check(lreq);
3860                         /*
3861                          * scan_requests() for the previous epoch(s)
3862                          * may have already added it to the list, since
3863                          * it's not unlinked here.
3864                          */
3865                         if (list_empty(&lreq->scan_item))
3866                                 list_add_tail(&lreq->scan_item, need_resend_linger);
3867                         break;
3868                 case CALC_TARGET_POOL_DNE:
3869                         list_del_init(&lreq->scan_item);
3870                         check_linger_pool_dne(lreq);
3871                         break;
3872                 }
3873         }
3874
3875         for (n = rb_first(&osd->o_requests); n; ) {
3876                 struct ceph_osd_request *req =
3877                     rb_entry(n, struct ceph_osd_request, r_node);
3878                 enum calc_target_result ct_res;
3879
3880                 n = rb_next(n); /* unlink_request(), check_pool_dne() */
3881
3882                 dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
3883                 ct_res = calc_target(osdc, &req->r_t, false);
3884                 switch (ct_res) {
3885                 case CALC_TARGET_NO_ACTION:
3886                         force_resend_writes = cleared_full ||
3887                             (check_pool_cleared_full &&
3888                              pool_cleared_full(osdc, req->r_t.base_oloc.pool));
3889                         if (!force_resend &&
3890                             (!(req->r_flags & CEPH_OSD_FLAG_WRITE) ||
3891                              !force_resend_writes))
3892                                 break;
3893
3894                         fallthrough;
3895                 case CALC_TARGET_NEED_RESEND:
3896                         cancel_map_check(req);
3897                         unlink_request(osd, req);
3898                         insert_request(need_resend, req);
3899                         break;
3900                 case CALC_TARGET_POOL_DNE:
3901                         check_pool_dne(req);
3902                         break;
3903                 }
3904         }
3905 }
3906
3907 static int handle_one_map(struct ceph_osd_client *osdc,
3908                           void *p, void *end, bool incremental,
3909                           struct rb_root *need_resend,
3910                           struct list_head *need_resend_linger)
3911 {
3912         struct ceph_osdmap *newmap;
3913         struct rb_node *n;
3914         bool skipped_map = false;
3915         bool was_full;
3916
3917         was_full = ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL);
3918         set_pool_was_full(osdc);
3919
3920         if (incremental)
3921                 newmap = osdmap_apply_incremental(&p, end,
3922                                                   ceph_msgr2(osdc->client),
3923                                                   osdc->osdmap);
3924         else
3925                 newmap = ceph_osdmap_decode(&p, end, ceph_msgr2(osdc->client));
3926         if (IS_ERR(newmap))
3927                 return PTR_ERR(newmap);
3928
3929         if (newmap != osdc->osdmap) {
3930                 /*
3931                  * Preserve ->was_full before destroying the old map.
3932                  * For pools that weren't in the old map, ->was_full
3933                  * should be false.
3934                  */
3935                 for (n = rb_first(&newmap->pg_pools); n; n = rb_next(n)) {
3936                         struct ceph_pg_pool_info *pi =
3937                             rb_entry(n, struct ceph_pg_pool_info, node);
3938                         struct ceph_pg_pool_info *old_pi;
3939
3940                         old_pi = ceph_pg_pool_by_id(osdc->osdmap, pi->id);
3941                         if (old_pi)
3942                                 pi->was_full = old_pi->was_full;
3943                         else
3944                                 WARN_ON(pi->was_full);
3945                 }
3946
3947                 if (osdc->osdmap->epoch &&
3948                     osdc->osdmap->epoch + 1 < newmap->epoch) {
3949                         WARN_ON(incremental);
3950                         skipped_map = true;
3951                 }
3952
3953                 ceph_osdmap_destroy(osdc->osdmap);
3954                 osdc->osdmap = newmap;
3955         }
3956
3957         was_full &= !ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL);
3958         scan_requests(&osdc->homeless_osd, skipped_map, was_full, true,
3959                       need_resend, need_resend_linger);
3960
3961         for (n = rb_first(&osdc->osds); n; ) {
3962                 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
3963
3964                 n = rb_next(n); /* close_osd() */
3965
3966                 scan_requests(osd, skipped_map, was_full, true, need_resend,
3967                               need_resend_linger);
3968                 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
3969                     memcmp(&osd->o_con.peer_addr,
3970                            ceph_osd_addr(osdc->osdmap, osd->o_osd),
3971                            sizeof(struct ceph_entity_addr)))
3972                         close_osd(osd);
3973         }
3974
3975         return 0;
3976 }
3977
3978 static void kick_requests(struct ceph_osd_client *osdc,
3979                           struct rb_root *need_resend,
3980                           struct list_head *need_resend_linger)
3981 {
3982         struct ceph_osd_linger_request *lreq, *nlreq;
3983         enum calc_target_result ct_res;
3984         struct rb_node *n;
3985
3986         /* make sure need_resend targets reflect latest map */
3987         for (n = rb_first(need_resend); n; ) {
3988                 struct ceph_osd_request *req =
3989                     rb_entry(n, struct ceph_osd_request, r_node);
3990
3991                 n = rb_next(n);
3992
3993                 if (req->r_t.epoch < osdc->osdmap->epoch) {
3994                         ct_res = calc_target(osdc, &req->r_t, false);
3995                         if (ct_res == CALC_TARGET_POOL_DNE) {
3996                                 erase_request(need_resend, req);
3997                                 check_pool_dne(req);
3998                         }
3999                 }
4000         }
4001
4002         for (n = rb_first(need_resend); n; ) {
4003                 struct ceph_osd_request *req =
4004                     rb_entry(n, struct ceph_osd_request, r_node);
4005                 struct ceph_osd *osd;
4006
4007                 n = rb_next(n);
4008                 erase_request(need_resend, req); /* before link_request() */
4009
4010                 osd = lookup_create_osd(osdc, req->r_t.osd, true);
4011                 link_request(osd, req);
4012                 if (!req->r_linger) {
4013                         if (!osd_homeless(osd) && !req->r_t.paused)
4014                                 send_request(req);
4015                 } else {
4016                         cancel_linger_request(req);
4017                 }
4018         }
4019
4020         list_for_each_entry_safe(lreq, nlreq, need_resend_linger, scan_item) {
4021                 if (!osd_homeless(lreq->osd))
4022                         send_linger(lreq);
4023
4024                 list_del_init(&lreq->scan_item);
4025         }
4026 }
4027
4028 /*
4029  * Process updated osd map.
4030  *
4031  * The message contains any number of incremental and full maps, normally
4032  * indicating some sort of topology change in the cluster.  Kick requests
4033  * off to different OSDs as needed.
4034  */
4035 void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
4036 {
4037         void *p = msg->front.iov_base;
4038         void *const end = p + msg->front.iov_len;
4039         u32 nr_maps, maplen;
4040         u32 epoch;
4041         struct ceph_fsid fsid;
4042         struct rb_root need_resend = RB_ROOT;
4043         LIST_HEAD(need_resend_linger);
4044         bool handled_incremental = false;
4045         bool was_pauserd, was_pausewr;
4046         bool pauserd, pausewr;
4047         int err;
4048
4049         dout("%s have %u\n", __func__, osdc->osdmap->epoch);
4050         down_write(&osdc->lock);
4051
4052         /* verify fsid */
4053         ceph_decode_need(&p, end, sizeof(fsid), bad);
4054         ceph_decode_copy(&p, &fsid, sizeof(fsid));
4055         if (ceph_check_fsid(osdc->client, &fsid) < 0)
4056                 goto bad;
4057
4058         was_pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
4059         was_pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
4060                       ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
4061                       have_pool_full(osdc);
4062
4063         /* incremental maps */
4064         ceph_decode_32_safe(&p, end, nr_maps, bad);
4065         dout(" %d inc maps\n", nr_maps);
4066         while (nr_maps > 0) {
4067                 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
4068                 epoch = ceph_decode_32(&p);
4069                 maplen = ceph_decode_32(&p);
4070                 ceph_decode_need(&p, end, maplen, bad);
4071                 if (osdc->osdmap->epoch &&
4072                     osdc->osdmap->epoch + 1 == epoch) {
4073                         dout("applying incremental map %u len %d\n",
4074                              epoch, maplen);
4075                         err = handle_one_map(osdc, p, p + maplen, true,
4076                                              &need_resend, &need_resend_linger);
4077                         if (err)
4078                                 goto bad;
4079                         handled_incremental = true;
4080                 } else {
4081                         dout("ignoring incremental map %u len %d\n",
4082                              epoch, maplen);
4083                 }
4084                 p += maplen;
4085                 nr_maps--;
4086         }
4087         if (handled_incremental)
4088                 goto done;
4089
4090         /* full maps */
4091         ceph_decode_32_safe(&p, end, nr_maps, bad);
4092         dout(" %d full maps\n", nr_maps);
4093         while (nr_maps) {
4094                 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
4095                 epoch = ceph_decode_32(&p);
4096                 maplen = ceph_decode_32(&p);
4097                 ceph_decode_need(&p, end, maplen, bad);
4098                 if (nr_maps > 1) {
4099                         dout("skipping non-latest full map %u len %d\n",
4100                              epoch, maplen);
4101                 } else if (osdc->osdmap->epoch >= epoch) {
4102                         dout("skipping full map %u len %d, "
4103                              "older than our %u\n", epoch, maplen,
4104                              osdc->osdmap->epoch);
4105                 } else {
4106                         dout("taking full map %u len %d\n", epoch, maplen);
4107                         err = handle_one_map(osdc, p, p + maplen, false,
4108                                              &need_resend, &need_resend_linger);
4109                         if (err)
4110                                 goto bad;
4111                 }
4112                 p += maplen;
4113                 nr_maps--;
4114         }
4115
4116 done:
4117         /*
4118          * subscribe to subsequent osdmap updates if full to ensure
4119          * we find out when we are no longer full and stop returning
4120          * ENOSPC.
4121          */
4122         pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
4123         pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
4124                   ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
4125                   have_pool_full(osdc);
4126         if (was_pauserd || was_pausewr || pauserd || pausewr ||
4127             osdc->osdmap->epoch < osdc->epoch_barrier)
4128                 maybe_request_map(osdc);
4129
4130         kick_requests(osdc, &need_resend, &need_resend_linger);
4131
4132         ceph_osdc_abort_on_full(osdc);
4133         ceph_monc_got_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
4134                           osdc->osdmap->epoch);
4135         up_write(&osdc->lock);
4136         wake_up_all(&osdc->client->auth_wq);
4137         return;
4138
4139 bad:
4140         pr_err("osdc handle_map corrupt msg\n");
4141         ceph_msg_dump(msg);
4142         up_write(&osdc->lock);
4143 }
4144
4145 /*
4146  * Resubmit requests pending on the given osd.
4147  */
4148 static void kick_osd_requests(struct ceph_osd *osd)
4149 {
4150         struct rb_node *n;
4151
4152         clear_backoffs(osd);
4153
4154         for (n = rb_first(&osd->o_requests); n; ) {
4155                 struct ceph_osd_request *req =
4156                     rb_entry(n, struct ceph_osd_request, r_node);
4157
4158                 n = rb_next(n); /* cancel_linger_request() */
4159
4160                 if (!req->r_linger) {
4161                         if (!req->r_t.paused)
4162                                 send_request(req);
4163                 } else {
4164                         cancel_linger_request(req);
4165                 }
4166         }
4167         for (n = rb_first(&osd->o_linger_requests); n; n = rb_next(n)) {
4168                 struct ceph_osd_linger_request *lreq =
4169                     rb_entry(n, struct ceph_osd_linger_request, node);
4170
4171                 send_linger(lreq);
4172         }
4173 }
4174
4175 /*
4176  * If the osd connection drops, we need to resubmit all requests.
4177  */
4178 static void osd_fault(struct ceph_connection *con)
4179 {
4180         struct ceph_osd *osd = con->private;
4181         struct ceph_osd_client *osdc = osd->o_osdc;
4182
4183         dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
4184
4185         down_write(&osdc->lock);
4186         if (!osd_registered(osd)) {
4187                 dout("%s osd%d unknown\n", __func__, osd->o_osd);
4188                 goto out_unlock;
4189         }
4190
4191         if (!reopen_osd(osd))
4192                 kick_osd_requests(osd);
4193         maybe_request_map(osdc);
4194
4195 out_unlock:
4196         up_write(&osdc->lock);
4197 }
4198
4199 struct MOSDBackoff {
4200         struct ceph_spg spgid;
4201         u32 map_epoch;
4202         u8 op;
4203         u64 id;
4204         struct ceph_hobject_id *begin;
4205         struct ceph_hobject_id *end;
4206 };
4207
4208 static int decode_MOSDBackoff(const struct ceph_msg *msg, struct MOSDBackoff *m)
4209 {
4210         void *p = msg->front.iov_base;
4211         void *const end = p + msg->front.iov_len;
4212         u8 struct_v;
4213         u32 struct_len;
4214         int ret;
4215
4216         ret = ceph_start_decoding(&p, end, 1, "spg_t", &struct_v, &struct_len);
4217         if (ret)
4218                 return ret;
4219
4220         ret = ceph_decode_pgid(&p, end, &m->spgid.pgid);
4221         if (ret)
4222                 return ret;
4223
4224         ceph_decode_8_safe(&p, end, m->spgid.shard, e_inval);
4225         ceph_decode_32_safe(&p, end, m->map_epoch, e_inval);
4226         ceph_decode_8_safe(&p, end, m->op, e_inval);
4227         ceph_decode_64_safe(&p, end, m->id, e_inval);
4228
4229         m->begin = kzalloc(sizeof(*m->begin), GFP_NOIO);
4230         if (!m->begin)
4231                 return -ENOMEM;
4232
4233         ret = decode_hoid(&p, end, m->begin);
4234         if (ret) {
4235                 free_hoid(m->begin);
4236                 return ret;
4237         }
4238
4239         m->end = kzalloc(sizeof(*m->end), GFP_NOIO);
4240         if (!m->end) {
4241                 free_hoid(m->begin);
4242                 return -ENOMEM;
4243         }
4244
4245         ret = decode_hoid(&p, end, m->end);
4246         if (ret) {
4247                 free_hoid(m->begin);
4248                 free_hoid(m->end);
4249                 return ret;
4250         }
4251
4252         return 0;
4253
4254 e_inval:
4255         return -EINVAL;
4256 }
4257
4258 static struct ceph_msg *create_backoff_message(
4259                                 const struct ceph_osd_backoff *backoff,
4260                                 u32 map_epoch)
4261 {
4262         struct ceph_msg *msg;
4263         void *p, *end;
4264         int msg_size;
4265
4266         msg_size = CEPH_ENCODING_START_BLK_LEN +
4267                         CEPH_PGID_ENCODING_LEN + 1; /* spgid */
4268         msg_size += 4 + 1 + 8; /* map_epoch, op, id */
4269         msg_size += CEPH_ENCODING_START_BLK_LEN +
4270                         hoid_encoding_size(backoff->begin);
4271         msg_size += CEPH_ENCODING_START_BLK_LEN +
4272                         hoid_encoding_size(backoff->end);
4273
4274         msg = ceph_msg_new(CEPH_MSG_OSD_BACKOFF, msg_size, GFP_NOIO, true);
4275         if (!msg)
4276                 return NULL;
4277
4278         p = msg->front.iov_base;
4279         end = p + msg->front_alloc_len;
4280
4281         encode_spgid(&p, &backoff->spgid);
4282         ceph_encode_32(&p, map_epoch);
4283         ceph_encode_8(&p, CEPH_OSD_BACKOFF_OP_ACK_BLOCK);
4284         ceph_encode_64(&p, backoff->id);
4285         encode_hoid(&p, end, backoff->begin);
4286         encode_hoid(&p, end, backoff->end);
4287         BUG_ON(p != end);
4288
4289         msg->front.iov_len = p - msg->front.iov_base;
4290         msg->hdr.version = cpu_to_le16(1); /* MOSDBackoff v1 */
4291         msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
4292
4293         return msg;
4294 }
4295
4296 static void handle_backoff_block(struct ceph_osd *osd, struct MOSDBackoff *m)
4297 {
4298         struct ceph_spg_mapping *spg;
4299         struct ceph_osd_backoff *backoff;
4300         struct ceph_msg *msg;
4301
4302         dout("%s osd%d spgid %llu.%xs%d id %llu\n", __func__, osd->o_osd,
4303              m->spgid.pgid.pool, m->spgid.pgid.seed, m->spgid.shard, m->id);
4304
4305         spg = lookup_spg_mapping(&osd->o_backoff_mappings, &m->spgid);
4306         if (!spg) {
4307                 spg = alloc_spg_mapping();
4308                 if (!spg) {
4309                         pr_err("%s failed to allocate spg\n", __func__);
4310                         return;
4311                 }
4312                 spg->spgid = m->spgid; /* struct */
4313                 insert_spg_mapping(&osd->o_backoff_mappings, spg);
4314         }
4315
4316         backoff = alloc_backoff();
4317         if (!backoff) {
4318                 pr_err("%s failed to allocate backoff\n", __func__);
4319                 return;
4320         }
4321         backoff->spgid = m->spgid; /* struct */
4322         backoff->id = m->id;
4323         backoff->begin = m->begin;
4324         m->begin = NULL; /* backoff now owns this */
4325         backoff->end = m->end;
4326         m->end = NULL;   /* ditto */
4327
4328         insert_backoff(&spg->backoffs, backoff);
4329         insert_backoff_by_id(&osd->o_backoffs_by_id, backoff);
4330
4331         /*
4332          * Ack with original backoff's epoch so that the OSD can
4333          * discard this if there was a PG split.
4334          */
4335         msg = create_backoff_message(backoff, m->map_epoch);
4336         if (!msg) {
4337                 pr_err("%s failed to allocate msg\n", __func__);
4338                 return;
4339         }
4340         ceph_con_send(&osd->o_con, msg);
4341 }
4342
4343 static bool target_contained_by(const struct ceph_osd_request_target *t,
4344                                 const struct ceph_hobject_id *begin,
4345                                 const struct ceph_hobject_id *end)
4346 {
4347         struct ceph_hobject_id hoid;
4348         int cmp;
4349
4350         hoid_fill_from_target(&hoid, t);
4351         cmp = hoid_compare(&hoid, begin);
4352         return !cmp || (cmp > 0 && hoid_compare(&hoid, end) < 0);
4353 }
4354
4355 static void handle_backoff_unblock(struct ceph_osd *osd,
4356                                    const struct MOSDBackoff *m)
4357 {
4358         struct ceph_spg_mapping *spg;
4359         struct ceph_osd_backoff *backoff;
4360         struct rb_node *n;
4361
4362         dout("%s osd%d spgid %llu.%xs%d id %llu\n", __func__, osd->o_osd,
4363              m->spgid.pgid.pool, m->spgid.pgid.seed, m->spgid.shard, m->id);
4364
4365         backoff = lookup_backoff_by_id(&osd->o_backoffs_by_id, m->id);
4366         if (!backoff) {
4367                 pr_err("%s osd%d spgid %llu.%xs%d id %llu backoff dne\n",
4368                        __func__, osd->o_osd, m->spgid.pgid.pool,
4369                        m->spgid.pgid.seed, m->spgid.shard, m->id);
4370                 return;
4371         }
4372
4373         if (hoid_compare(backoff->begin, m->begin) &&
4374             hoid_compare(backoff->end, m->end)) {
4375                 pr_err("%s osd%d spgid %llu.%xs%d id %llu bad range?\n",
4376                        __func__, osd->o_osd, m->spgid.pgid.pool,
4377                        m->spgid.pgid.seed, m->spgid.shard, m->id);
4378                 /* unblock it anyway... */
4379         }
4380
4381         spg = lookup_spg_mapping(&osd->o_backoff_mappings, &backoff->spgid);
4382         BUG_ON(!spg);
4383
4384         erase_backoff(&spg->backoffs, backoff);
4385         erase_backoff_by_id(&osd->o_backoffs_by_id, backoff);
4386         free_backoff(backoff);
4387
4388         if (RB_EMPTY_ROOT(&spg->backoffs)) {
4389                 erase_spg_mapping(&osd->o_backoff_mappings, spg);
4390                 free_spg_mapping(spg);
4391         }
4392
4393         for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) {
4394                 struct ceph_osd_request *req =
4395                     rb_entry(n, struct ceph_osd_request, r_node);
4396
4397                 if (!ceph_spg_compare(&req->r_t.spgid, &m->spgid)) {
4398                         /*
4399                          * Match against @m, not @backoff -- the PG may
4400                          * have split on the OSD.
4401                          */
4402                         if (target_contained_by(&req->r_t, m->begin, m->end)) {
4403                                 /*
4404                                  * If no other installed backoff applies,
4405                                  * resend.
4406                                  */
4407                                 send_request(req);
4408                         }
4409                 }
4410         }
4411 }
4412
4413 static void handle_backoff(struct ceph_osd *osd, struct ceph_msg *msg)
4414 {
4415         struct ceph_osd_client *osdc = osd->o_osdc;
4416         struct MOSDBackoff m;
4417         int ret;
4418
4419         down_read(&osdc->lock);
4420         if (!osd_registered(osd)) {
4421                 dout("%s osd%d unknown\n", __func__, osd->o_osd);
4422                 up_read(&osdc->lock);
4423                 return;
4424         }
4425         WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num));
4426
4427         mutex_lock(&osd->lock);
4428         ret = decode_MOSDBackoff(msg, &m);
4429         if (ret) {
4430                 pr_err("failed to decode MOSDBackoff: %d\n", ret);
4431                 ceph_msg_dump(msg);
4432                 goto out_unlock;
4433         }
4434
4435         switch (m.op) {
4436         case CEPH_OSD_BACKOFF_OP_BLOCK:
4437                 handle_backoff_block(osd, &m);
4438                 break;
4439         case CEPH_OSD_BACKOFF_OP_UNBLOCK:
4440                 handle_backoff_unblock(osd, &m);
4441                 break;
4442         default:
4443                 pr_err("%s osd%d unknown op %d\n", __func__, osd->o_osd, m.op);
4444         }
4445
4446         free_hoid(m.begin);
4447         free_hoid(m.end);
4448
4449 out_unlock:
4450         mutex_unlock(&osd->lock);
4451         up_read(&osdc->lock);
4452 }
4453
4454 /*
4455  * Process osd watch notifications
4456  */
4457 static void handle_watch_notify(struct ceph_osd_client *osdc,
4458                                 struct ceph_msg *msg)
4459 {
4460         void *p = msg->front.iov_base;
4461         void *const end = p + msg->front.iov_len;
4462         struct ceph_osd_linger_request *lreq;
4463         struct linger_work *lwork;
4464         u8 proto_ver, opcode;
4465         u64 cookie, notify_id;
4466         u64 notifier_id = 0;
4467         s32 return_code = 0;
4468         void *payload = NULL;
4469         u32 payload_len = 0;
4470
4471         ceph_decode_8_safe(&p, end, proto_ver, bad);
4472         ceph_decode_8_safe(&p, end, opcode, bad);
4473         ceph_decode_64_safe(&p, end, cookie, bad);
4474         p += 8; /* skip ver */
4475         ceph_decode_64_safe(&p, end, notify_id, bad);
4476
4477         if (proto_ver >= 1) {
4478                 ceph_decode_32_safe(&p, end, payload_len, bad);
4479                 ceph_decode_need(&p, end, payload_len, bad);
4480                 payload = p;
4481                 p += payload_len;
4482         }
4483
4484         if (le16_to_cpu(msg->hdr.version) >= 2)
4485                 ceph_decode_32_safe(&p, end, return_code, bad);
4486
4487         if (le16_to_cpu(msg->hdr.version) >= 3)
4488                 ceph_decode_64_safe(&p, end, notifier_id, bad);
4489
4490         down_read(&osdc->lock);
4491         lreq = lookup_linger_osdc(&osdc->linger_requests, cookie);
4492         if (!lreq) {
4493                 dout("%s opcode %d cookie %llu dne\n", __func__, opcode,
4494                      cookie);
4495                 goto out_unlock_osdc;
4496         }
4497
4498         mutex_lock(&lreq->lock);
4499         dout("%s opcode %d cookie %llu lreq %p is_watch %d\n", __func__,
4500              opcode, cookie, lreq, lreq->is_watch);
4501         if (opcode == CEPH_WATCH_EVENT_DISCONNECT) {
4502                 if (!lreq->last_error) {
4503                         lreq->last_error = -ENOTCONN;
4504                         queue_watch_error(lreq);
4505                 }
4506         } else if (!lreq->is_watch) {
4507                 /* CEPH_WATCH_EVENT_NOTIFY_COMPLETE */
4508                 if (lreq->notify_id && lreq->notify_id != notify_id) {
4509                         dout("lreq %p notify_id %llu != %llu, ignoring\n", lreq,
4510                              lreq->notify_id, notify_id);
4511                 } else if (!completion_done(&lreq->notify_finish_wait)) {
4512                         struct ceph_msg_data *data =
4513                             msg->num_data_items ? &msg->data[0] : NULL;
4514
4515                         if (data) {
4516                                 if (lreq->preply_pages) {
4517                                         WARN_ON(data->type !=
4518                                                         CEPH_MSG_DATA_PAGES);
4519                                         *lreq->preply_pages = data->pages;
4520                                         *lreq->preply_len = data->length;
4521                                         data->own_pages = false;
4522                                 }
4523                         }
4524                         lreq->notify_finish_error = return_code;
4525                         complete_all(&lreq->notify_finish_wait);
4526                 }
4527         } else {
4528                 /* CEPH_WATCH_EVENT_NOTIFY */
4529                 lwork = lwork_alloc(lreq, do_watch_notify);
4530                 if (!lwork) {
4531                         pr_err("failed to allocate notify-lwork\n");
4532                         goto out_unlock_lreq;
4533                 }
4534
4535                 lwork->notify.notify_id = notify_id;
4536                 lwork->notify.notifier_id = notifier_id;
4537                 lwork->notify.payload = payload;
4538                 lwork->notify.payload_len = payload_len;
4539                 lwork->notify.msg = ceph_msg_get(msg);
4540                 lwork_queue(lwork);
4541         }
4542
4543 out_unlock_lreq:
4544         mutex_unlock(&lreq->lock);
4545 out_unlock_osdc:
4546         up_read(&osdc->lock);
4547         return;
4548
4549 bad:
4550         pr_err("osdc handle_watch_notify corrupt msg\n");
4551 }
4552
4553 /*
4554  * Register request, send initial attempt.
4555  */
4556 int ceph_osdc_start_request(struct ceph_osd_client *osdc,
4557                             struct ceph_osd_request *req,
4558                             bool nofail)
4559 {
4560         down_read(&osdc->lock);
4561         submit_request(req, false);
4562         up_read(&osdc->lock);
4563
4564         return 0;
4565 }
4566 EXPORT_SYMBOL(ceph_osdc_start_request);
4567
4568 /*
4569  * Unregister a registered request.  The request is not completed:
4570  * ->r_result isn't set and __complete_request() isn't called.
4571  */
4572 void ceph_osdc_cancel_request(struct ceph_osd_request *req)
4573 {
4574         struct ceph_osd_client *osdc = req->r_osdc;
4575
4576         down_write(&osdc->lock);
4577         if (req->r_osd)
4578                 cancel_request(req);
4579         up_write(&osdc->lock);
4580 }
4581 EXPORT_SYMBOL(ceph_osdc_cancel_request);
4582
4583 /*
4584  * @timeout: in jiffies, 0 means "wait forever"
4585  */
4586 static int wait_request_timeout(struct ceph_osd_request *req,
4587                                 unsigned long timeout)
4588 {
4589         long left;
4590
4591         dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
4592         left = wait_for_completion_killable_timeout(&req->r_completion,
4593                                                 ceph_timeout_jiffies(timeout));
4594         if (left <= 0) {
4595                 left = left ?: -ETIMEDOUT;
4596                 ceph_osdc_cancel_request(req);
4597         } else {
4598                 left = req->r_result; /* completed */
4599         }
4600
4601         return left;
4602 }
4603
4604 /*
4605  * wait for a request to complete
4606  */
4607 int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
4608                            struct ceph_osd_request *req)
4609 {
4610         return wait_request_timeout(req, 0);
4611 }
4612 EXPORT_SYMBOL(ceph_osdc_wait_request);
4613
4614 /*
4615  * sync - wait for all in-flight requests to flush.  avoid starvation.
4616  */
4617 void ceph_osdc_sync(struct ceph_osd_client *osdc)
4618 {
4619         struct rb_node *n, *p;
4620         u64 last_tid = atomic64_read(&osdc->last_tid);
4621
4622 again:
4623         down_read(&osdc->lock);
4624         for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
4625                 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
4626
4627                 mutex_lock(&osd->lock);
4628                 for (p = rb_first(&osd->o_requests); p; p = rb_next(p)) {
4629                         struct ceph_osd_request *req =
4630                             rb_entry(p, struct ceph_osd_request, r_node);
4631
4632                         if (req->r_tid > last_tid)
4633                                 break;
4634
4635                         if (!(req->r_flags & CEPH_OSD_FLAG_WRITE))
4636                                 continue;
4637
4638                         ceph_osdc_get_request(req);
4639                         mutex_unlock(&osd->lock);
4640                         up_read(&osdc->lock);
4641                         dout("%s waiting on req %p tid %llu last_tid %llu\n",
4642                              __func__, req, req->r_tid, last_tid);
4643                         wait_for_completion(&req->r_completion);
4644                         ceph_osdc_put_request(req);
4645                         goto again;
4646                 }
4647
4648                 mutex_unlock(&osd->lock);
4649         }
4650
4651         up_read(&osdc->lock);
4652         dout("%s done last_tid %llu\n", __func__, last_tid);
4653 }
4654 EXPORT_SYMBOL(ceph_osdc_sync);
4655
4656 static struct ceph_osd_request *
4657 alloc_linger_request(struct ceph_osd_linger_request *lreq)
4658 {
4659         struct ceph_osd_request *req;
4660
4661         req = ceph_osdc_alloc_request(lreq->osdc, NULL, 1, false, GFP_NOIO);
4662         if (!req)
4663                 return NULL;
4664
4665         ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
4666         ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
4667         return req;
4668 }
4669
4670 static struct ceph_osd_request *
4671 alloc_watch_request(struct ceph_osd_linger_request *lreq, u8 watch_opcode)
4672 {
4673         struct ceph_osd_request *req;
4674
4675         req = alloc_linger_request(lreq);
4676         if (!req)
4677                 return NULL;
4678
4679         /*
4680          * Pass 0 for cookie because we don't know it yet, it will be
4681          * filled in by linger_submit().
4682          */
4683         osd_req_op_watch_init(req, 0, 0, watch_opcode);
4684
4685         if (ceph_osdc_alloc_messages(req, GFP_NOIO)) {
4686                 ceph_osdc_put_request(req);
4687                 return NULL;
4688         }
4689
4690         return req;
4691 }
4692
4693 /*
4694  * Returns a handle, caller owns a ref.
4695  */
4696 struct ceph_osd_linger_request *
4697 ceph_osdc_watch(struct ceph_osd_client *osdc,
4698                 struct ceph_object_id *oid,
4699                 struct ceph_object_locator *oloc,
4700                 rados_watchcb2_t wcb,
4701                 rados_watcherrcb_t errcb,
4702                 void *data)
4703 {
4704         struct ceph_osd_linger_request *lreq;
4705         int ret;
4706
4707         lreq = linger_alloc(osdc);
4708         if (!lreq)
4709                 return ERR_PTR(-ENOMEM);
4710
4711         lreq->is_watch = true;
4712         lreq->wcb = wcb;
4713         lreq->errcb = errcb;
4714         lreq->data = data;
4715         lreq->watch_valid_thru = jiffies;
4716
4717         ceph_oid_copy(&lreq->t.base_oid, oid);
4718         ceph_oloc_copy(&lreq->t.base_oloc, oloc);
4719         lreq->t.flags = CEPH_OSD_FLAG_WRITE;
4720         ktime_get_real_ts64(&lreq->mtime);
4721
4722         lreq->reg_req = alloc_watch_request(lreq, CEPH_OSD_WATCH_OP_WATCH);
4723         if (!lreq->reg_req) {
4724                 ret = -ENOMEM;
4725                 goto err_put_lreq;
4726         }
4727
4728         lreq->ping_req = alloc_watch_request(lreq, CEPH_OSD_WATCH_OP_PING);
4729         if (!lreq->ping_req) {
4730                 ret = -ENOMEM;
4731                 goto err_put_lreq;
4732         }
4733
4734         linger_submit(lreq);
4735         ret = linger_reg_commit_wait(lreq);
4736         if (ret) {
4737                 linger_cancel(lreq);
4738                 goto err_put_lreq;
4739         }
4740
4741         return lreq;
4742
4743 err_put_lreq:
4744         linger_put(lreq);
4745         return ERR_PTR(ret);
4746 }
4747 EXPORT_SYMBOL(ceph_osdc_watch);
4748
4749 /*
4750  * Releases a ref.
4751  *
4752  * Times out after mount_timeout to preserve rbd unmap behaviour
4753  * introduced in 2894e1d76974 ("rbd: timeout watch teardown on unmap
4754  * with mount_timeout").
4755  */
4756 int ceph_osdc_unwatch(struct ceph_osd_client *osdc,
4757                       struct ceph_osd_linger_request *lreq)
4758 {
4759         struct ceph_options *opts = osdc->client->options;
4760         struct ceph_osd_request *req;
4761         int ret;
4762
4763         req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
4764         if (!req)
4765                 return -ENOMEM;
4766
4767         ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
4768         ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
4769         req->r_flags = CEPH_OSD_FLAG_WRITE;
4770         ktime_get_real_ts64(&req->r_mtime);
4771         osd_req_op_watch_init(req, 0, lreq->linger_id,
4772                               CEPH_OSD_WATCH_OP_UNWATCH);
4773
4774         ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
4775         if (ret)
4776                 goto out_put_req;
4777
4778         ceph_osdc_start_request(osdc, req, false);
4779         linger_cancel(lreq);
4780         linger_put(lreq);
4781         ret = wait_request_timeout(req, opts->mount_timeout);
4782
4783 out_put_req:
4784         ceph_osdc_put_request(req);
4785         return ret;
4786 }
4787 EXPORT_SYMBOL(ceph_osdc_unwatch);
4788
4789 static int osd_req_op_notify_ack_init(struct ceph_osd_request *req, int which,
4790                                       u64 notify_id, u64 cookie, void *payload,
4791                                       u32 payload_len)
4792 {
4793         struct ceph_osd_req_op *op;
4794         struct ceph_pagelist *pl;
4795         int ret;
4796
4797         op = osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY_ACK, 0);
4798
4799         pl = ceph_pagelist_alloc(GFP_NOIO);
4800         if (!pl)
4801                 return -ENOMEM;
4802
4803         ret = ceph_pagelist_encode_64(pl, notify_id);
4804         ret |= ceph_pagelist_encode_64(pl, cookie);
4805         if (payload) {
4806                 ret |= ceph_pagelist_encode_32(pl, payload_len);
4807                 ret |= ceph_pagelist_append(pl, payload, payload_len);
4808         } else {
4809                 ret |= ceph_pagelist_encode_32(pl, 0);
4810         }
4811         if (ret) {
4812                 ceph_pagelist_release(pl);
4813                 return -ENOMEM;
4814         }
4815
4816         ceph_osd_data_pagelist_init(&op->notify_ack.request_data, pl);
4817         op->indata_len = pl->length;
4818         return 0;
4819 }
4820
4821 int ceph_osdc_notify_ack(struct ceph_osd_client *osdc,
4822                          struct ceph_object_id *oid,
4823                          struct ceph_object_locator *oloc,
4824                          u64 notify_id,
4825                          u64 cookie,
4826                          void *payload,
4827                          u32 payload_len)
4828 {
4829         struct ceph_osd_request *req;
4830         int ret;
4831
4832         req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
4833         if (!req)
4834                 return -ENOMEM;
4835
4836         ceph_oid_copy(&req->r_base_oid, oid);
4837         ceph_oloc_copy(&req->r_base_oloc, oloc);
4838         req->r_flags = CEPH_OSD_FLAG_READ;
4839
4840         ret = osd_req_op_notify_ack_init(req, 0, notify_id, cookie, payload,
4841                                          payload_len);
4842         if (ret)
4843                 goto out_put_req;
4844
4845         ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
4846         if (ret)
4847                 goto out_put_req;
4848
4849         ceph_osdc_start_request(osdc, req, false);
4850         ret = ceph_osdc_wait_request(osdc, req);
4851
4852 out_put_req:
4853         ceph_osdc_put_request(req);
4854         return ret;
4855 }
4856 EXPORT_SYMBOL(ceph_osdc_notify_ack);
4857
4858 static int osd_req_op_notify_init(struct ceph_osd_request *req, int which,
4859                                   u64 cookie, u32 prot_ver, u32 timeout,
4860                                   void *payload, u32 payload_len)
4861 {
4862         struct ceph_osd_req_op *op;
4863         struct ceph_pagelist *pl;
4864         int ret;
4865
4866         op = osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY, 0);
4867         op->notify.cookie = cookie;
4868
4869         pl = ceph_pagelist_alloc(GFP_NOIO);
4870         if (!pl)
4871                 return -ENOMEM;
4872
4873         ret = ceph_pagelist_encode_32(pl, 1); /* prot_ver */
4874         ret |= ceph_pagelist_encode_32(pl, timeout);
4875         ret |= ceph_pagelist_encode_32(pl, payload_len);
4876         ret |= ceph_pagelist_append(pl, payload, payload_len);
4877         if (ret) {
4878                 ceph_pagelist_release(pl);
4879                 return -ENOMEM;
4880         }
4881
4882         ceph_osd_data_pagelist_init(&op->notify.request_data, pl);
4883         op->indata_len = pl->length;
4884         return 0;
4885 }
4886
4887 /*
4888  * @timeout: in seconds
4889  *
4890  * @preply_{pages,len} are initialized both on success and error.
4891  * The caller is responsible for:
4892  *
4893  *     ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len))
4894  */
4895 int ceph_osdc_notify(struct ceph_osd_client *osdc,
4896                      struct ceph_object_id *oid,
4897                      struct ceph_object_locator *oloc,
4898                      void *payload,
4899                      u32 payload_len,
4900                      u32 timeout,
4901                      struct page ***preply_pages,
4902                      size_t *preply_len)
4903 {
4904         struct ceph_osd_linger_request *lreq;
4905         struct page **pages;
4906         int ret;
4907
4908         WARN_ON(!timeout);
4909         if (preply_pages) {
4910                 *preply_pages = NULL;
4911                 *preply_len = 0;
4912         }
4913
4914         lreq = linger_alloc(osdc);
4915         if (!lreq)
4916                 return -ENOMEM;
4917
4918         lreq->preply_pages = preply_pages;
4919         lreq->preply_len = preply_len;
4920
4921         ceph_oid_copy(&lreq->t.base_oid, oid);
4922         ceph_oloc_copy(&lreq->t.base_oloc, oloc);
4923         lreq->t.flags = CEPH_OSD_FLAG_READ;
4924
4925         lreq->reg_req = alloc_linger_request(lreq);
4926         if (!lreq->reg_req) {
4927                 ret = -ENOMEM;
4928                 goto out_put_lreq;
4929         }
4930
4931         /*
4932          * Pass 0 for cookie because we don't know it yet, it will be
4933          * filled in by linger_submit().
4934          */
4935         ret = osd_req_op_notify_init(lreq->reg_req, 0, 0, 1, timeout,
4936                                      payload, payload_len);
4937         if (ret)
4938                 goto out_put_lreq;
4939
4940         /* for notify_id */
4941         pages = ceph_alloc_page_vector(1, GFP_NOIO);
4942         if (IS_ERR(pages)) {
4943                 ret = PTR_ERR(pages);
4944                 goto out_put_lreq;
4945         }
4946         ceph_osd_data_pages_init(osd_req_op_data(lreq->reg_req, 0, notify,
4947                                                  response_data),
4948                                  pages, PAGE_SIZE, 0, false, true);
4949
4950         ret = ceph_osdc_alloc_messages(lreq->reg_req, GFP_NOIO);
4951         if (ret)
4952                 goto out_put_lreq;
4953
4954         linger_submit(lreq);
4955         ret = linger_reg_commit_wait(lreq);
4956         if (!ret)
4957                 ret = linger_notify_finish_wait(lreq);
4958         else
4959                 dout("lreq %p failed to initiate notify %d\n", lreq, ret);
4960
4961         linger_cancel(lreq);
4962 out_put_lreq:
4963         linger_put(lreq);
4964         return ret;
4965 }
4966 EXPORT_SYMBOL(ceph_osdc_notify);
4967
4968 /*
4969  * Return the number of milliseconds since the watch was last
4970  * confirmed, or an error.  If there is an error, the watch is no
4971  * longer valid, and should be destroyed with ceph_osdc_unwatch().
4972  */
4973 int ceph_osdc_watch_check(struct ceph_osd_client *osdc,
4974                           struct ceph_osd_linger_request *lreq)
4975 {
4976         unsigned long stamp, age;
4977         int ret;
4978
4979         down_read(&osdc->lock);
4980         mutex_lock(&lreq->lock);
4981         stamp = lreq->watch_valid_thru;
4982         if (!list_empty(&lreq->pending_lworks)) {
4983                 struct linger_work *lwork =
4984                     list_first_entry(&lreq->pending_lworks,
4985                                      struct linger_work,
4986                                      pending_item);
4987
4988                 if (time_before(lwork->queued_stamp, stamp))
4989                         stamp = lwork->queued_stamp;
4990         }
4991         age = jiffies - stamp;
4992         dout("%s lreq %p linger_id %llu age %lu last_error %d\n", __func__,
4993              lreq, lreq->linger_id, age, lreq->last_error);
4994         /* we are truncating to msecs, so return a safe upper bound */
4995         ret = lreq->last_error ?: 1 + jiffies_to_msecs(age);
4996
4997         mutex_unlock(&lreq->lock);
4998         up_read(&osdc->lock);
4999         return ret;
5000 }
5001
5002 static int decode_watcher(void **p, void *end, struct ceph_watch_item *item)
5003 {
5004         u8 struct_v;
5005         u32 struct_len;
5006         int ret;
5007
5008         ret = ceph_start_decoding(p, end, 2, "watch_item_t",
5009                                   &struct_v, &struct_len);
5010         if (ret)
5011                 goto bad;
5012
5013         ret = -EINVAL;
5014         ceph_decode_copy_safe(p, end, &item->name, sizeof(item->name), bad);
5015         ceph_decode_64_safe(p, end, item->cookie, bad);
5016         ceph_decode_skip_32(p, end, bad); /* skip timeout seconds */
5017
5018         if (struct_v >= 2) {
5019                 ret = ceph_decode_entity_addr(p, end, &item->addr);
5020                 if (ret)
5021                         goto bad;
5022         } else {
5023                 ret = 0;
5024         }
5025
5026         dout("%s %s%llu cookie %llu addr %s\n", __func__,
5027              ENTITY_NAME(item->name), item->cookie,
5028              ceph_pr_addr(&item->addr));
5029 bad:
5030         return ret;
5031 }
5032
5033 static int decode_watchers(void **p, void *end,
5034                            struct ceph_watch_item **watchers,
5035                            u32 *num_watchers)
5036 {
5037         u8 struct_v;
5038         u32 struct_len;
5039         int i;
5040         int ret;
5041
5042         ret = ceph_start_decoding(p, end, 1, "obj_list_watch_response_t",
5043                                   &struct_v, &struct_len);
5044         if (ret)
5045                 return ret;
5046
5047         *num_watchers = ceph_decode_32(p);
5048         *watchers = kcalloc(*num_watchers, sizeof(**watchers), GFP_NOIO);
5049         if (!*watchers)
5050                 return -ENOMEM;
5051
5052         for (i = 0; i < *num_watchers; i++) {
5053                 ret = decode_watcher(p, end, *watchers + i);
5054                 if (ret) {
5055                         kfree(*watchers);
5056                         return ret;
5057                 }
5058         }
5059
5060         return 0;
5061 }
5062
5063 /*
5064  * On success, the caller is responsible for:
5065  *
5066  *     kfree(watchers);
5067  */
5068 int ceph_osdc_list_watchers(struct ceph_osd_client *osdc,
5069                             struct ceph_object_id *oid,
5070                             struct ceph_object_locator *oloc,
5071                             struct ceph_watch_item **watchers,
5072                             u32 *num_watchers)
5073 {
5074         struct ceph_osd_request *req;
5075         struct page **pages;
5076         int ret;
5077
5078         req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
5079         if (!req)
5080                 return -ENOMEM;
5081
5082         ceph_oid_copy(&req->r_base_oid, oid);
5083         ceph_oloc_copy(&req->r_base_oloc, oloc);
5084         req->r_flags = CEPH_OSD_FLAG_READ;
5085
5086         pages = ceph_alloc_page_vector(1, GFP_NOIO);
5087         if (IS_ERR(pages)) {
5088                 ret = PTR_ERR(pages);
5089                 goto out_put_req;
5090         }
5091
5092         osd_req_op_init(req, 0, CEPH_OSD_OP_LIST_WATCHERS, 0);
5093         ceph_osd_data_pages_init(osd_req_op_data(req, 0, list_watchers,
5094                                                  response_data),
5095                                  pages, PAGE_SIZE, 0, false, true);
5096
5097         ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
5098         if (ret)
5099                 goto out_put_req;
5100
5101         ceph_osdc_start_request(osdc, req, false);
5102         ret = ceph_osdc_wait_request(osdc, req);
5103         if (ret >= 0) {
5104                 void *p = page_address(pages[0]);
5105                 void *const end = p + req->r_ops[0].outdata_len;
5106
5107                 ret = decode_watchers(&p, end, watchers, num_watchers);
5108         }
5109
5110 out_put_req:
5111         ceph_osdc_put_request(req);
5112         return ret;
5113 }
5114 EXPORT_SYMBOL(ceph_osdc_list_watchers);
5115
5116 /*
5117  * Call all pending notify callbacks - for use after a watch is
5118  * unregistered, to make sure no more callbacks for it will be invoked
5119  */
5120 void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
5121 {
5122         dout("%s osdc %p\n", __func__, osdc);
5123         flush_workqueue(osdc->notify_wq);
5124 }
5125 EXPORT_SYMBOL(ceph_osdc_flush_notifies);
5126
5127 void ceph_osdc_maybe_request_map(struct ceph_osd_client *osdc)
5128 {
5129         down_read(&osdc->lock);
5130         maybe_request_map(osdc);
5131         up_read(&osdc->lock);
5132 }
5133 EXPORT_SYMBOL(ceph_osdc_maybe_request_map);
5134
5135 /*
5136  * Execute an OSD class method on an object.
5137  *
5138  * @flags: CEPH_OSD_FLAG_*
5139  * @resp_len: in/out param for reply length
5140  */
5141 int ceph_osdc_call(struct ceph_osd_client *osdc,
5142                    struct ceph_object_id *oid,
5143                    struct ceph_object_locator *oloc,
5144                    const char *class, const char *method,
5145                    unsigned int flags,
5146                    struct page *req_page, size_t req_len,
5147                    struct page **resp_pages, size_t *resp_len)
5148 {
5149         struct ceph_osd_request *req;
5150         int ret;
5151
5152         if (req_len > PAGE_SIZE)
5153                 return -E2BIG;
5154
5155         req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
5156         if (!req)
5157                 return -ENOMEM;
5158
5159         ceph_oid_copy(&req->r_base_oid, oid);
5160         ceph_oloc_copy(&req->r_base_oloc, oloc);
5161         req->r_flags = flags;
5162
5163         ret = osd_req_op_cls_init(req, 0, class, method);
5164         if (ret)
5165                 goto out_put_req;
5166
5167         if (req_page)
5168                 osd_req_op_cls_request_data_pages(req, 0, &req_page, req_len,
5169                                                   0, false, false);
5170         if (resp_pages)
5171                 osd_req_op_cls_response_data_pages(req, 0, resp_pages,
5172                                                    *resp_len, 0, false, false);
5173
5174         ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
5175         if (ret)
5176                 goto out_put_req;
5177
5178         ceph_osdc_start_request(osdc, req, false);
5179         ret = ceph_osdc_wait_request(osdc, req);
5180         if (ret >= 0) {
5181                 ret = req->r_ops[0].rval;
5182                 if (resp_pages)
5183                         *resp_len = req->r_ops[0].outdata_len;
5184         }
5185
5186 out_put_req:
5187         ceph_osdc_put_request(req);
5188         return ret;
5189 }
5190 EXPORT_SYMBOL(ceph_osdc_call);
5191
5192 /*
5193  * reset all osd connections
5194  */
5195 void ceph_osdc_reopen_osds(struct ceph_osd_client *osdc)
5196 {
5197         struct rb_node *n;
5198
5199         down_write(&osdc->lock);
5200         for (n = rb_first(&osdc->osds); n; ) {
5201                 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
5202
5203                 n = rb_next(n);
5204                 if (!reopen_osd(osd))
5205                         kick_osd_requests(osd);
5206         }
5207         up_write(&osdc->lock);
5208 }
5209
5210 /*
5211  * init, shutdown
5212  */
5213 int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
5214 {
5215         int err;
5216
5217         dout("init\n");
5218         osdc->client = client;
5219         init_rwsem(&osdc->lock);
5220         osdc->osds = RB_ROOT;
5221         INIT_LIST_HEAD(&osdc->osd_lru);
5222         spin_lock_init(&osdc->osd_lru_lock);
5223         osd_init(&osdc->homeless_osd);
5224         osdc->homeless_osd.o_osdc = osdc;
5225         osdc->homeless_osd.o_osd = CEPH_HOMELESS_OSD;
5226         osdc->last_linger_id = CEPH_LINGER_ID_START;
5227         osdc->linger_requests = RB_ROOT;
5228         osdc->map_checks = RB_ROOT;
5229         osdc->linger_map_checks = RB_ROOT;
5230         INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
5231         INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
5232
5233         err = -ENOMEM;
5234         osdc->osdmap = ceph_osdmap_alloc();
5235         if (!osdc->osdmap)
5236                 goto out;
5237
5238         osdc->req_mempool = mempool_create_slab_pool(10,
5239                                                      ceph_osd_request_cache);
5240         if (!osdc->req_mempool)
5241                 goto out_map;
5242
5243         err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
5244                                 PAGE_SIZE, CEPH_OSD_SLAB_OPS, 10, "osd_op");
5245         if (err < 0)
5246                 goto out_mempool;
5247         err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
5248                                 PAGE_SIZE, CEPH_OSD_SLAB_OPS, 10,
5249                                 "osd_op_reply");
5250         if (err < 0)
5251                 goto out_msgpool;
5252
5253         err = -ENOMEM;
5254         osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
5255         if (!osdc->notify_wq)
5256                 goto out_msgpool_reply;
5257
5258         osdc->completion_wq = create_singlethread_workqueue("ceph-completion");
5259         if (!osdc->completion_wq)
5260                 goto out_notify_wq;
5261
5262         schedule_delayed_work(&osdc->timeout_work,
5263                               osdc->client->options->osd_keepalive_timeout);
5264         schedule_delayed_work(&osdc->osds_timeout_work,
5265             round_jiffies_relative(osdc->client->options->osd_idle_ttl));
5266
5267         return 0;
5268
5269 out_notify_wq:
5270         destroy_workqueue(osdc->notify_wq);
5271 out_msgpool_reply:
5272         ceph_msgpool_destroy(&osdc->msgpool_op_reply);
5273 out_msgpool:
5274         ceph_msgpool_destroy(&osdc->msgpool_op);
5275 out_mempool:
5276         mempool_destroy(osdc->req_mempool);
5277 out_map:
5278         ceph_osdmap_destroy(osdc->osdmap);
5279 out:
5280         return err;
5281 }
5282
5283 void ceph_osdc_stop(struct ceph_osd_client *osdc)
5284 {
5285         destroy_workqueue(osdc->completion_wq);
5286         destroy_workqueue(osdc->notify_wq);
5287         cancel_delayed_work_sync(&osdc->timeout_work);
5288         cancel_delayed_work_sync(&osdc->osds_timeout_work);
5289
5290         down_write(&osdc->lock);
5291         while (!RB_EMPTY_ROOT(&osdc->osds)) {
5292                 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
5293                                                 struct ceph_osd, o_node);
5294                 close_osd(osd);
5295         }
5296         up_write(&osdc->lock);
5297         WARN_ON(refcount_read(&osdc->homeless_osd.o_ref) != 1);
5298         osd_cleanup(&osdc->homeless_osd);
5299
5300         WARN_ON(!list_empty(&osdc->osd_lru));
5301         WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_requests));
5302         WARN_ON(!RB_EMPTY_ROOT(&osdc->map_checks));
5303         WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_map_checks));
5304         WARN_ON(atomic_read(&osdc->num_requests));
5305         WARN_ON(atomic_read(&osdc->num_homeless));
5306
5307         ceph_osdmap_destroy(osdc->osdmap);
5308         mempool_destroy(osdc->req_mempool);
5309         ceph_msgpool_destroy(&osdc->msgpool_op);
5310         ceph_msgpool_destroy(&osdc->msgpool_op_reply);
5311 }
5312
5313 static int osd_req_op_copy_from_init(struct ceph_osd_request *req,
5314                                      u64 src_snapid, u64 src_version,
5315                                      struct ceph_object_id *src_oid,
5316                                      struct ceph_object_locator *src_oloc,
5317                                      u32 src_fadvise_flags,
5318                                      u32 dst_fadvise_flags,
5319                                      u32 truncate_seq, u64 truncate_size,
5320                                      u8 copy_from_flags)
5321 {
5322         struct ceph_osd_req_op *op;
5323         struct page **pages;
5324         void *p, *end;
5325
5326         pages = ceph_alloc_page_vector(1, GFP_KERNEL);
5327         if (IS_ERR(pages))
5328                 return PTR_ERR(pages);
5329
5330         op = osd_req_op_init(req, 0, CEPH_OSD_OP_COPY_FROM2,
5331                              dst_fadvise_flags);
5332         op->copy_from.snapid = src_snapid;
5333         op->copy_from.src_version = src_version;
5334         op->copy_from.flags = copy_from_flags;
5335         op->copy_from.src_fadvise_flags = src_fadvise_flags;
5336
5337         p = page_address(pages[0]);
5338         end = p + PAGE_SIZE;
5339         ceph_encode_string(&p, end, src_oid->name, src_oid->name_len);
5340         encode_oloc(&p, end, src_oloc);
5341         ceph_encode_32(&p, truncate_seq);
5342         ceph_encode_64(&p, truncate_size);
5343         op->indata_len = PAGE_SIZE - (end - p);
5344
5345         ceph_osd_data_pages_init(&op->copy_from.osd_data, pages,
5346                                  op->indata_len, 0, false, true);
5347         return 0;
5348 }
5349
5350 int ceph_osdc_copy_from(struct ceph_osd_client *osdc,
5351                         u64 src_snapid, u64 src_version,
5352                         struct ceph_object_id *src_oid,
5353                         struct ceph_object_locator *src_oloc,
5354                         u32 src_fadvise_flags,
5355                         struct ceph_object_id *dst_oid,
5356                         struct ceph_object_locator *dst_oloc,
5357                         u32 dst_fadvise_flags,
5358                         u32 truncate_seq, u64 truncate_size,
5359                         u8 copy_from_flags)
5360 {
5361         struct ceph_osd_request *req;
5362         int ret;
5363
5364         req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_KERNEL);
5365         if (!req)
5366                 return -ENOMEM;
5367
5368         req->r_flags = CEPH_OSD_FLAG_WRITE;
5369
5370         ceph_oloc_copy(&req->r_t.base_oloc, dst_oloc);
5371         ceph_oid_copy(&req->r_t.base_oid, dst_oid);
5372
5373         ret = osd_req_op_copy_from_init(req, src_snapid, src_version, src_oid,
5374                                         src_oloc, src_fadvise_flags,
5375                                         dst_fadvise_flags, truncate_seq,
5376                                         truncate_size, copy_from_flags);
5377         if (ret)
5378                 goto out;
5379
5380         ret = ceph_osdc_alloc_messages(req, GFP_KERNEL);
5381         if (ret)
5382                 goto out;
5383
5384         ceph_osdc_start_request(osdc, req, false);
5385         ret = ceph_osdc_wait_request(osdc, req);
5386
5387 out:
5388         ceph_osdc_put_request(req);
5389         return ret;
5390 }
5391 EXPORT_SYMBOL(ceph_osdc_copy_from);
5392
5393 int __init ceph_osdc_setup(void)
5394 {
5395         size_t size = sizeof(struct ceph_osd_request) +
5396             CEPH_OSD_SLAB_OPS * sizeof(struct ceph_osd_req_op);
5397
5398         BUG_ON(ceph_osd_request_cache);
5399         ceph_osd_request_cache = kmem_cache_create("ceph_osd_request", size,
5400                                                    0, 0, NULL);
5401
5402         return ceph_osd_request_cache ? 0 : -ENOMEM;
5403 }
5404
5405 void ceph_osdc_cleanup(void)
5406 {
5407         BUG_ON(!ceph_osd_request_cache);
5408         kmem_cache_destroy(ceph_osd_request_cache);
5409         ceph_osd_request_cache = NULL;
5410 }
5411
5412 /*
5413  * handle incoming message
5414  */
5415 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
5416 {
5417         struct ceph_osd *osd = con->private;
5418         struct ceph_osd_client *osdc = osd->o_osdc;
5419         int type = le16_to_cpu(msg->hdr.type);
5420
5421         switch (type) {
5422         case CEPH_MSG_OSD_MAP:
5423                 ceph_osdc_handle_map(osdc, msg);
5424                 break;
5425         case CEPH_MSG_OSD_OPREPLY:
5426                 handle_reply(osd, msg);
5427                 break;
5428         case CEPH_MSG_OSD_BACKOFF:
5429                 handle_backoff(osd, msg);
5430                 break;
5431         case CEPH_MSG_WATCH_NOTIFY:
5432                 handle_watch_notify(osdc, msg);
5433                 break;
5434
5435         default:
5436                 pr_err("received unknown message type %d %s\n", type,
5437                        ceph_msg_type_name(type));
5438         }
5439
5440         ceph_msg_put(msg);
5441 }
5442
5443 /*
5444  * Lookup and return message for incoming reply.  Don't try to do
5445  * anything about a larger than preallocated data portion of the
5446  * message at the moment - for now, just skip the message.
5447  */
5448 static struct ceph_msg *get_reply(struct ceph_connection *con,
5449                                   struct ceph_msg_header *hdr,
5450                                   int *skip)
5451 {
5452         struct ceph_osd *osd = con->private;
5453         struct ceph_osd_client *osdc = osd->o_osdc;
5454         struct ceph_msg *m = NULL;
5455         struct ceph_osd_request *req;
5456         int front_len = le32_to_cpu(hdr->front_len);
5457         int data_len = le32_to_cpu(hdr->data_len);
5458         u64 tid = le64_to_cpu(hdr->tid);
5459
5460         down_read(&osdc->lock);
5461         if (!osd_registered(osd)) {
5462                 dout("%s osd%d unknown, skipping\n", __func__, osd->o_osd);
5463                 *skip = 1;
5464                 goto out_unlock_osdc;
5465         }
5466         WARN_ON(osd->o_osd != le64_to_cpu(hdr->src.num));
5467
5468         mutex_lock(&osd->lock);
5469         req = lookup_request(&osd->o_requests, tid);
5470         if (!req) {
5471                 dout("%s osd%d tid %llu unknown, skipping\n", __func__,
5472                      osd->o_osd, tid);
5473                 *skip = 1;
5474                 goto out_unlock_session;
5475         }
5476
5477         ceph_msg_revoke_incoming(req->r_reply);
5478
5479         if (front_len > req->r_reply->front_alloc_len) {
5480                 pr_warn("%s osd%d tid %llu front %d > preallocated %d\n",
5481                         __func__, osd->o_osd, req->r_tid, front_len,
5482                         req->r_reply->front_alloc_len);
5483                 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS,
5484                                  false);
5485                 if (!m)
5486                         goto out_unlock_session;
5487                 ceph_msg_put(req->r_reply);
5488                 req->r_reply = m;
5489         }
5490
5491         if (data_len > req->r_reply->data_length) {
5492                 pr_warn("%s osd%d tid %llu data %d > preallocated %zu, skipping\n",
5493                         __func__, osd->o_osd, req->r_tid, data_len,
5494                         req->r_reply->data_length);
5495                 m = NULL;
5496                 *skip = 1;
5497                 goto out_unlock_session;
5498         }
5499
5500         m = ceph_msg_get(req->r_reply);
5501         dout("get_reply tid %lld %p\n", tid, m);
5502
5503 out_unlock_session:
5504         mutex_unlock(&osd->lock);
5505 out_unlock_osdc:
5506         up_read(&osdc->lock);
5507         return m;
5508 }
5509
5510 static struct ceph_msg *alloc_msg_with_page_vector(struct ceph_msg_header *hdr)
5511 {
5512         struct ceph_msg *m;
5513         int type = le16_to_cpu(hdr->type);
5514         u32 front_len = le32_to_cpu(hdr->front_len);
5515         u32 data_len = le32_to_cpu(hdr->data_len);
5516
5517         m = ceph_msg_new2(type, front_len, 1, GFP_NOIO, false);
5518         if (!m)
5519                 return NULL;
5520
5521         if (data_len) {
5522                 struct page **pages;
5523
5524                 pages = ceph_alloc_page_vector(calc_pages_for(0, data_len),
5525                                                GFP_NOIO);
5526                 if (IS_ERR(pages)) {
5527                         ceph_msg_put(m);
5528                         return NULL;
5529                 }
5530
5531                 ceph_msg_data_add_pages(m, pages, data_len, 0, true);
5532         }
5533
5534         return m;
5535 }
5536
5537 static struct ceph_msg *alloc_msg(struct ceph_connection *con,
5538                                   struct ceph_msg_header *hdr,
5539                                   int *skip)
5540 {
5541         struct ceph_osd *osd = con->private;
5542         int type = le16_to_cpu(hdr->type);
5543
5544         *skip = 0;
5545         switch (type) {
5546         case CEPH_MSG_OSD_MAP:
5547         case CEPH_MSG_OSD_BACKOFF:
5548         case CEPH_MSG_WATCH_NOTIFY:
5549                 return alloc_msg_with_page_vector(hdr);
5550         case CEPH_MSG_OSD_OPREPLY:
5551                 return get_reply(con, hdr, skip);
5552         default:
5553                 pr_warn("%s osd%d unknown msg type %d, skipping\n", __func__,
5554                         osd->o_osd, type);
5555                 *skip = 1;
5556                 return NULL;
5557         }
5558 }
5559
5560 /*
5561  * Wrappers to refcount containing ceph_osd struct
5562  */
5563 static struct ceph_connection *get_osd_con(struct ceph_connection *con)
5564 {
5565         struct ceph_osd *osd = con->private;
5566         if (get_osd(osd))
5567                 return con;
5568         return NULL;
5569 }
5570
5571 static void put_osd_con(struct ceph_connection *con)
5572 {
5573         struct ceph_osd *osd = con->private;
5574         put_osd(osd);
5575 }
5576
5577 /*
5578  * authentication
5579  */
5580
5581 /*
5582  * Note: returned pointer is the address of a structure that's
5583  * managed separately.  Caller must *not* attempt to free it.
5584  */
5585 static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
5586                                         int *proto, int force_new)
5587 {
5588         struct ceph_osd *o = con->private;
5589         struct ceph_osd_client *osdc = o->o_osdc;
5590         struct ceph_auth_client *ac = osdc->client->monc.auth;
5591         struct ceph_auth_handshake *auth = &o->o_auth;
5592
5593         if (force_new && auth->authorizer) {
5594                 ceph_auth_destroy_authorizer(auth->authorizer);
5595                 auth->authorizer = NULL;
5596         }
5597         if (!auth->authorizer) {
5598                 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
5599                                                       auth);
5600                 if (ret)
5601                         return ERR_PTR(ret);
5602         } else {
5603                 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
5604                                                      auth);
5605                 if (ret)
5606                         return ERR_PTR(ret);
5607         }
5608         *proto = ac->protocol;
5609
5610         return auth;
5611 }
5612
5613 static int add_authorizer_challenge(struct ceph_connection *con,
5614                                     void *challenge_buf, int challenge_buf_len)
5615 {
5616         struct ceph_osd *o = con->private;
5617         struct ceph_osd_client *osdc = o->o_osdc;
5618         struct ceph_auth_client *ac = osdc->client->monc.auth;
5619
5620         return ceph_auth_add_authorizer_challenge(ac, o->o_auth.authorizer,
5621                                             challenge_buf, challenge_buf_len);
5622 }
5623
5624 static int verify_authorizer_reply(struct ceph_connection *con)
5625 {
5626         struct ceph_osd *o = con->private;
5627         struct ceph_osd_client *osdc = o->o_osdc;
5628         struct ceph_auth_client *ac = osdc->client->monc.auth;
5629         struct ceph_auth_handshake *auth = &o->o_auth;
5630
5631         return ceph_auth_verify_authorizer_reply(ac, auth->authorizer,
5632                 auth->authorizer_reply_buf, auth->authorizer_reply_buf_len,
5633                 NULL, NULL, NULL, NULL);
5634 }
5635
5636 static int invalidate_authorizer(struct ceph_connection *con)
5637 {
5638         struct ceph_osd *o = con->private;
5639         struct ceph_osd_client *osdc = o->o_osdc;
5640         struct ceph_auth_client *ac = osdc->client->monc.auth;
5641
5642         ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
5643         return ceph_monc_validate_auth(&osdc->client->monc);
5644 }
5645
5646 static int osd_get_auth_request(struct ceph_connection *con,
5647                                 void *buf, int *buf_len,
5648                                 void **authorizer, int *authorizer_len)
5649 {
5650         struct ceph_osd *o = con->private;
5651         struct ceph_auth_client *ac = o->o_osdc->client->monc.auth;
5652         struct ceph_auth_handshake *auth = &o->o_auth;
5653         int ret;
5654
5655         ret = ceph_auth_get_authorizer(ac, auth, CEPH_ENTITY_TYPE_OSD,
5656                                        buf, buf_len);
5657         if (ret)
5658                 return ret;
5659
5660         *authorizer = auth->authorizer_buf;
5661         *authorizer_len = auth->authorizer_buf_len;
5662         return 0;
5663 }
5664
5665 static int osd_handle_auth_reply_more(struct ceph_connection *con,
5666                                       void *reply, int reply_len,
5667                                       void *buf, int *buf_len,
5668                                       void **authorizer, int *authorizer_len)
5669 {
5670         struct ceph_osd *o = con->private;
5671         struct ceph_auth_client *ac = o->o_osdc->client->monc.auth;
5672         struct ceph_auth_handshake *auth = &o->o_auth;
5673         int ret;
5674
5675         ret = ceph_auth_handle_svc_reply_more(ac, auth, reply, reply_len,
5676                                               buf, buf_len);
5677         if (ret)
5678                 return ret;
5679
5680         *authorizer = auth->authorizer_buf;
5681         *authorizer_len = auth->authorizer_buf_len;
5682         return 0;
5683 }
5684
5685 static int osd_handle_auth_done(struct ceph_connection *con,
5686                                 u64 global_id, void *reply, int reply_len,
5687                                 u8 *session_key, int *session_key_len,
5688                                 u8 *con_secret, int *con_secret_len)
5689 {
5690         struct ceph_osd *o = con->private;
5691         struct ceph_auth_client *ac = o->o_osdc->client->monc.auth;
5692         struct ceph_auth_handshake *auth = &o->o_auth;
5693
5694         return ceph_auth_handle_svc_reply_done(ac, auth, reply, reply_len,
5695                                                session_key, session_key_len,
5696                                                con_secret, con_secret_len);
5697 }
5698
5699 static int osd_handle_auth_bad_method(struct ceph_connection *con,
5700                                       int used_proto, int result,
5701                                       const int *allowed_protos, int proto_cnt,
5702                                       const int *allowed_modes, int mode_cnt)
5703 {
5704         struct ceph_osd *o = con->private;
5705         struct ceph_mon_client *monc = &o->o_osdc->client->monc;
5706         int ret;
5707
5708         if (ceph_auth_handle_bad_authorizer(monc->auth, CEPH_ENTITY_TYPE_OSD,
5709                                             used_proto, result,
5710                                             allowed_protos, proto_cnt,
5711                                             allowed_modes, mode_cnt)) {
5712                 ret = ceph_monc_validate_auth(monc);
5713                 if (ret)
5714                         return ret;
5715         }
5716
5717         return -EACCES;
5718 }
5719
5720 static void osd_reencode_message(struct ceph_msg *msg)
5721 {
5722         int type = le16_to_cpu(msg->hdr.type);
5723
5724         if (type == CEPH_MSG_OSD_OP)
5725                 encode_request_finish(msg);
5726 }
5727
5728 static int osd_sign_message(struct ceph_msg *msg)
5729 {
5730         struct ceph_osd *o = msg->con->private;
5731         struct ceph_auth_handshake *auth = &o->o_auth;
5732
5733         return ceph_auth_sign_message(auth, msg);
5734 }
5735
5736 static int osd_check_message_signature(struct ceph_msg *msg)
5737 {
5738         struct ceph_osd *o = msg->con->private;
5739         struct ceph_auth_handshake *auth = &o->o_auth;
5740
5741         return ceph_auth_check_message_signature(auth, msg);
5742 }
5743
5744 static const struct ceph_connection_operations osd_con_ops = {
5745         .get = get_osd_con,
5746         .put = put_osd_con,
5747         .dispatch = dispatch,
5748         .get_authorizer = get_authorizer,
5749         .add_authorizer_challenge = add_authorizer_challenge,
5750         .verify_authorizer_reply = verify_authorizer_reply,
5751         .invalidate_authorizer = invalidate_authorizer,
5752         .alloc_msg = alloc_msg,
5753         .reencode_message = osd_reencode_message,
5754         .sign_message = osd_sign_message,
5755         .check_message_signature = osd_check_message_signature,
5756         .fault = osd_fault,
5757         .get_auth_request = osd_get_auth_request,
5758         .handle_auth_reply_more = osd_handle_auth_reply_more,
5759         .handle_auth_done = osd_handle_auth_done,
5760         .handle_auth_bad_method = osd_handle_auth_bad_method,
5761 };