net/ceph/osd_client.c

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