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              kref_read(&req->r_kref));
 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                      kref_read(&req->r_kref));
 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         RB_CLEAR_NODE(&req->r_node);
 464         RB_CLEAR_NODE(&req->r_mc_node);
 465         INIT_LIST_HEAD(&req->r_unsafe_item);
 466
 467         target_init(&req->r_t);
 468 }
 469
 470 /*
 471  * This is ugly, but it allows us to reuse linger registration and ping
 472  * requests, keeping the structure of the code around send_linger{_ping}()
 473  * reasonable.  Setting up a min_nr=2 mempool for each linger request
 474  * and dealing with copying ops (this blasts req only, watch op remains
 475  * intact) isn't any better.
 476  */
 477 static void request_reinit(struct ceph_osd_request *req)
 478 {
 479         struct ceph_osd_client *osdc = req->r_osdc;
 480         bool mempool = req->r_mempool;
 481         unsigned int num_ops = req->r_num_ops;
 482         u64 snapid = req->r_snapid;
 483         struct ceph_snap_context *snapc = req->r_snapc;
 484         bool linger = req->r_linger;
 485         struct ceph_msg *request_msg = req->r_request;
 486         struct ceph_msg *reply_msg = req->r_reply;
 487
 488         dout("%s req %p\n", __func__, req);
 489         WARN_ON(kref_read(&req->r_kref) != 1);
 490         request_release_checks(req);
 491
 492         WARN_ON(kref_read(&request_msg->kref) != 1);
 493         WARN_ON(kref_read(&reply_msg->kref) != 1);
 494         target_destroy(&req->r_t);
 495
 496         request_init(req);
 497         req->r_osdc = osdc;
 498         req->r_mempool = mempool;
 499         req->r_num_ops = num_ops;
 500         req->r_snapid = snapid;
 501         req->r_snapc = snapc;
 502         req->r_linger = linger;
 503         req->r_request = request_msg;
 504         req->r_reply = reply_msg;
 505 }
 506
 507 struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
 508                                                struct ceph_snap_context *snapc,
 509                                                unsigned int num_ops,
 510                                                bool use_mempool,
 511                                                gfp_t gfp_flags)
 512 {
 513         struct ceph_osd_request *req;
 514
 515         if (use_mempool) {
 516                 BUG_ON(num_ops > CEPH_OSD_SLAB_OPS);
 517                 req = mempool_alloc(osdc->req_mempool, gfp_flags);
 518         } else if (num_ops <= CEPH_OSD_SLAB_OPS) {
 519                 req = kmem_cache_alloc(ceph_osd_request_cache, gfp_flags);
 520         } else {
 521                 BUG_ON(num_ops > CEPH_OSD_MAX_OPS);
 522                 req = kmalloc(sizeof(*req) + num_ops * sizeof(req->r_ops[0]),
 523                               gfp_flags);
 524         }
 525         if (unlikely(!req))
 526                 return NULL;
 527
 528         request_init(req);
 529         req->r_osdc = osdc;
 530         req->r_mempool = use_mempool;
 531         req->r_num_ops = num_ops;
 532         req->r_snapid = CEPH_NOSNAP;
 533         req->r_snapc = ceph_get_snap_context(snapc);
 534
 535         dout("%s req %p\n", __func__, req);
 536         return req;
 537 }
 538 EXPORT_SYMBOL(ceph_osdc_alloc_request);
 539
 540 static int ceph_oloc_encoding_size(struct ceph_object_locator *oloc)
 541 {
 542         return 8 + 4 + 4 + 4 + (oloc->pool_ns ? oloc->pool_ns->len : 0);
 543 }
 544
 545 int ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp)
 546 {
 547         struct ceph_osd_client *osdc = req->r_osdc;
 548         struct ceph_msg *msg;
 549         int msg_size;
 550
 551         WARN_ON(ceph_oid_empty(&req->r_base_oid));
 552         WARN_ON(ceph_oloc_empty(&req->r_base_oloc));
 553
 554         /* create request message */
 555         msg_size = 4 + 4 + 4; /* client_inc, osdmap_epoch, flags */
 556         msg_size += 4 + 4 + 4 + 8; /* mtime, reassert_version */
 557         msg_size += CEPH_ENCODING_START_BLK_LEN +
 558                         ceph_oloc_encoding_size(&req->r_base_oloc); /* oloc */
 559         msg_size += 1 + 8 + 4 + 4; /* pgid */
 560         msg_size += 4 + req->r_base_oid.name_len; /* oid */
 561         msg_size += 2 + req->r_num_ops * sizeof(struct ceph_osd_op);
 562         msg_size += 8; /* snapid */
 563         msg_size += 8; /* snap_seq */
 564         msg_size += 4 + 8 * (req->r_snapc ? req->r_snapc->num_snaps : 0);
 565         msg_size += 4; /* retry_attempt */
 566
 567         if (req->r_mempool)
 568                 msg = ceph_msgpool_get(&osdc->msgpool_op, 0);
 569         else
 570                 msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp, true);
 571         if (!msg)
 572                 return -ENOMEM;
 573
 574         memset(msg->front.iov_base, 0, msg->front.iov_len);
 575         req->r_request = msg;
 576
 577         /* create reply message */
 578         msg_size = OSD_OPREPLY_FRONT_LEN;
 579         msg_size += req->r_base_oid.name_len;
 580         msg_size += req->r_num_ops * sizeof(struct ceph_osd_op);
 581
 582         if (req->r_mempool)
 583                 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0);
 584         else
 585                 msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, msg_size, gfp, true);
 586         if (!msg)
 587                 return -ENOMEM;
 588
 589         req->r_reply = msg;
 590
 591         return 0;
 592 }
 593 EXPORT_SYMBOL(ceph_osdc_alloc_messages);
 594
 595 static bool osd_req_opcode_valid(u16 opcode)
 596 {
 597         switch (opcode) {
 598 #define GENERATE_CASE(op, opcode, str)  case CEPH_OSD_OP_##op: return true;
 599 __CEPH_FORALL_OSD_OPS(GENERATE_CASE)
 600 #undef GENERATE_CASE
 601         default:
 602                 return false;
 603         }
 604 }
 605
 606 /*
 607  * This is an osd op init function for opcodes that have no data or
 608  * other information associated with them.  It also serves as a
 609  * common init routine for all the other init functions, below.
 610  */
 611 static struct ceph_osd_req_op *
 612 _osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which,
 613                  u16 opcode, u32 flags)
 614 {
 615         struct ceph_osd_req_op *op;
 616
 617         BUG_ON(which >= osd_req->r_num_ops);
 618         BUG_ON(!osd_req_opcode_valid(opcode));
 619
 620         op = &osd_req->r_ops[which];
 621         memset(op, 0, sizeof (*op));
 622         op->op = opcode;
 623         op->flags = flags;
 624
 625         return op;
 626 }
 627
 628 void osd_req_op_init(struct ceph_osd_request *osd_req,
 629                      unsigned int which, u16 opcode, u32 flags)
 630 {
 631         (void)_osd_req_op_init(osd_req, which, opcode, flags);
 632 }
 633 EXPORT_SYMBOL(osd_req_op_init);
 634
 635 void osd_req_op_extent_init(struct ceph_osd_request *osd_req,
 636                                 unsigned int which, u16 opcode,
 637                                 u64 offset, u64 length,
 638                                 u64 truncate_size, u32 truncate_seq)
 639 {
 640         struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
 641                                                       opcode, 0);
 642         size_t payload_len = 0;
 643
 644         BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
 645                opcode != CEPH_OSD_OP_WRITEFULL && opcode != CEPH_OSD_OP_ZERO &&
 646                opcode != CEPH_OSD_OP_TRUNCATE);
 647
 648         op->extent.offset = offset;
 649         op->extent.length = length;
 650         op->extent.truncate_size = truncate_size;
 651         op->extent.truncate_seq = truncate_seq;
 652         if (opcode == CEPH_OSD_OP_WRITE || opcode == CEPH_OSD_OP_WRITEFULL)
 653                 payload_len += length;
 654
 655         op->indata_len = payload_len;
 656 }
 657 EXPORT_SYMBOL(osd_req_op_extent_init);
 658
 659 void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
 660                                 unsigned int which, u64 length)
 661 {
 662         struct ceph_osd_req_op *op;
 663         u64 previous;
 664
 665         BUG_ON(which >= osd_req->r_num_ops);
 666         op = &osd_req->r_ops[which];
 667         previous = op->extent.length;
 668
 669         if (length == previous)
 670                 return;         /* Nothing to do */
 671         BUG_ON(length > previous);
 672
 673         op->extent.length = length;
 674         if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL)
 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);
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         WARN_ON(req->r_flags & (CEPH_OSD_FLAG_ACK | CEPH_OSD_FLAG_ONDISK));
1708         WARN_ON(!(req->r_flags & (CEPH_OSD_FLAG_READ | CEPH_OSD_FLAG_WRITE)));
1709
1710         req->r_flags |= CEPH_OSD_FLAG_ONDISK;
1711         atomic_inc(&req->r_osdc->num_requests);
1712
1713         req->r_start_stamp = jiffies;
1714 }
1715
1716 static void submit_request(struct ceph_osd_request *req, bool wrlocked)
1717 {
1718         ceph_osdc_get_request(req);
1719         account_request(req);
1720         __submit_request(req, wrlocked);
1721 }
1722
1723 static void finish_request(struct ceph_osd_request *req)
1724 {
1725         struct ceph_osd_client *osdc = req->r_osdc;
1726         struct ceph_osd *osd = req->r_osd;
1727
1728         verify_osd_locked(osd);
1729         dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
1730
1731         WARN_ON(lookup_request_mc(&osdc->map_checks, req->r_tid));
1732         unlink_request(osd, req);
1733         atomic_dec(&osdc->num_requests);
1734
1735         /*
1736          * If an OSD has failed or returned and a request has been sent
1737          * twice, it's possible to get a reply and end up here while the
1738          * request message is queued for delivery.  We will ignore the
1739          * reply, so not a big deal, but better to try and catch it.
1740          */
1741         ceph_msg_revoke(req->r_request);
1742         ceph_msg_revoke_incoming(req->r_reply);
1743 }
1744
1745 static void __complete_request(struct ceph_osd_request *req)
1746 {
1747         if (req->r_callback) {
1748                 dout("%s req %p tid %llu cb %pf result %d\n", __func__, req,
1749                      req->r_tid, req->r_callback, req->r_result);
1750                 req->r_callback(req);
1751         }
1752 }
1753
1754 /*
1755  * This is open-coded in handle_reply().
1756  */
1757 static void complete_request(struct ceph_osd_request *req, int err)
1758 {
1759         dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err);
1760
1761         req->r_result = err;
1762         finish_request(req);
1763         __complete_request(req);
1764         complete_all(&req->r_completion);
1765         ceph_osdc_put_request(req);
1766 }
1767
1768 static void cancel_map_check(struct ceph_osd_request *req)
1769 {
1770         struct ceph_osd_client *osdc = req->r_osdc;
1771         struct ceph_osd_request *lookup_req;
1772
1773         verify_osdc_wrlocked(osdc);
1774
1775         lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid);
1776         if (!lookup_req)
1777                 return;
1778
1779         WARN_ON(lookup_req != req);
1780         erase_request_mc(&osdc->map_checks, req);
1781         ceph_osdc_put_request(req);
1782 }
1783
1784 static void cancel_request(struct ceph_osd_request *req)
1785 {
1786         dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
1787
1788         cancel_map_check(req);
1789         finish_request(req);
1790         complete_all(&req->r_completion);
1791         ceph_osdc_put_request(req);
1792 }
1793
1794 static void abort_request(struct ceph_osd_request *req, int err)
1795 {
1796         dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err);
1797
1798         cancel_map_check(req);
1799         complete_request(req, err);
1800 }
1801
1802 static void check_pool_dne(struct ceph_osd_request *req)
1803 {
1804         struct ceph_osd_client *osdc = req->r_osdc;
1805         struct ceph_osdmap *map = osdc->osdmap;
1806
1807         verify_osdc_wrlocked(osdc);
1808         WARN_ON(!map->epoch);
1809
1810         if (req->r_attempts) {
1811                 /*
1812                  * We sent a request earlier, which means that
1813                  * previously the pool existed, and now it does not
1814                  * (i.e., it was deleted).
1815                  */
1816                 req->r_map_dne_bound = map->epoch;
1817                 dout("%s req %p tid %llu pool disappeared\n", __func__, req,
1818                      req->r_tid);
1819         } else {
1820                 dout("%s req %p tid %llu map_dne_bound %u have %u\n", __func__,
1821                      req, req->r_tid, req->r_map_dne_bound, map->epoch);
1822         }
1823
1824         if (req->r_map_dne_bound) {
1825                 if (map->epoch >= req->r_map_dne_bound) {
1826                         /* we had a new enough map */
1827                         pr_info_ratelimited("tid %llu pool does not exist\n",
1828                                             req->r_tid);
1829                         complete_request(req, -ENOENT);
1830                 }
1831         } else {
1832                 send_map_check(req);
1833         }
1834 }
1835
1836 static void map_check_cb(struct ceph_mon_generic_request *greq)
1837 {
1838         struct ceph_osd_client *osdc = &greq->monc->client->osdc;
1839         struct ceph_osd_request *req;
1840         u64 tid = greq->private_data;
1841
1842         WARN_ON(greq->result || !greq->u.newest);
1843
1844         down_write(&osdc->lock);
1845         req = lookup_request_mc(&osdc->map_checks, tid);
1846         if (!req) {
1847                 dout("%s tid %llu dne\n", __func__, tid);
1848                 goto out_unlock;
1849         }
1850
1851         dout("%s req %p tid %llu map_dne_bound %u newest %llu\n", __func__,
1852              req, req->r_tid, req->r_map_dne_bound, greq->u.newest);
1853         if (!req->r_map_dne_bound)
1854                 req->r_map_dne_bound = greq->u.newest;
1855         erase_request_mc(&osdc->map_checks, req);
1856         check_pool_dne(req);
1857
1858         ceph_osdc_put_request(req);
1859 out_unlock:
1860         up_write(&osdc->lock);
1861 }
1862
1863 static void send_map_check(struct ceph_osd_request *req)
1864 {
1865         struct ceph_osd_client *osdc = req->r_osdc;
1866         struct ceph_osd_request *lookup_req;
1867         int ret;
1868
1869         verify_osdc_wrlocked(osdc);
1870
1871         lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid);
1872         if (lookup_req) {
1873                 WARN_ON(lookup_req != req);
1874                 return;
1875         }
1876
1877         ceph_osdc_get_request(req);
1878         insert_request_mc(&osdc->map_checks, req);
1879         ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap",
1880                                           map_check_cb, req->r_tid);
1881         WARN_ON(ret);
1882 }
1883
1884 /*
1885  * lingering requests, watch/notify v2 infrastructure
1886  */
1887 static void linger_release(struct kref *kref)
1888 {
1889         struct ceph_osd_linger_request *lreq =
1890             container_of(kref, struct ceph_osd_linger_request, kref);
1891
1892         dout("%s lreq %p reg_req %p ping_req %p\n", __func__, lreq,
1893              lreq->reg_req, lreq->ping_req);
1894         WARN_ON(!RB_EMPTY_NODE(&lreq->node));
1895         WARN_ON(!RB_EMPTY_NODE(&lreq->osdc_node));
1896         WARN_ON(!RB_EMPTY_NODE(&lreq->mc_node));
1897         WARN_ON(!list_empty(&lreq->scan_item));
1898         WARN_ON(!list_empty(&lreq->pending_lworks));
1899         WARN_ON(lreq->osd);
1900
1901         if (lreq->reg_req)
1902                 ceph_osdc_put_request(lreq->reg_req);
1903         if (lreq->ping_req)
1904                 ceph_osdc_put_request(lreq->ping_req);
1905         target_destroy(&lreq->t);
1906         kfree(lreq);
1907 }
1908
1909 static void linger_put(struct ceph_osd_linger_request *lreq)
1910 {
1911         if (lreq)
1912                 kref_put(&lreq->kref, linger_release);
1913 }
1914
1915 static struct ceph_osd_linger_request *
1916 linger_get(struct ceph_osd_linger_request *lreq)
1917 {
1918         kref_get(&lreq->kref);
1919         return lreq;
1920 }
1921
1922 static struct ceph_osd_linger_request *
1923 linger_alloc(struct ceph_osd_client *osdc)
1924 {
1925         struct ceph_osd_linger_request *lreq;
1926
1927         lreq = kzalloc(sizeof(*lreq), GFP_NOIO);
1928         if (!lreq)
1929                 return NULL;
1930
1931         kref_init(&lreq->kref);
1932         mutex_init(&lreq->lock);
1933         RB_CLEAR_NODE(&lreq->node);
1934         RB_CLEAR_NODE(&lreq->osdc_node);
1935         RB_CLEAR_NODE(&lreq->mc_node);
1936         INIT_LIST_HEAD(&lreq->scan_item);
1937         INIT_LIST_HEAD(&lreq->pending_lworks);
1938         init_completion(&lreq->reg_commit_wait);
1939         init_completion(&lreq->notify_finish_wait);
1940
1941         lreq->osdc = osdc;
1942         target_init(&lreq->t);
1943
1944         dout("%s lreq %p\n", __func__, lreq);
1945         return lreq;
1946 }
1947
1948 DEFINE_RB_INSDEL_FUNCS(linger, struct ceph_osd_linger_request, linger_id, node)
1949 DEFINE_RB_FUNCS(linger_osdc, struct ceph_osd_linger_request, linger_id, osdc_node)
1950 DEFINE_RB_FUNCS(linger_mc, struct ceph_osd_linger_request, linger_id, mc_node)
1951
1952 /*
1953  * Create linger request <-> OSD session relation.
1954  *
1955  * @lreq has to be registered, @osd may be homeless.
1956  */
1957 static void link_linger(struct ceph_osd *osd,
1958                         struct ceph_osd_linger_request *lreq)
1959 {
1960         verify_osd_locked(osd);
1961         WARN_ON(!lreq->linger_id || lreq->osd);
1962         dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd,
1963              osd->o_osd, lreq, lreq->linger_id);
1964
1965         if (!osd_homeless(osd))
1966                 __remove_osd_from_lru(osd);
1967         else
1968                 atomic_inc(&osd->o_osdc->num_homeless);
1969
1970         get_osd(osd);
1971         insert_linger(&osd->o_linger_requests, lreq);
1972         lreq->osd = osd;
1973 }
1974
1975 static void unlink_linger(struct ceph_osd *osd,
1976                           struct ceph_osd_linger_request *lreq)
1977 {
1978         verify_osd_locked(osd);
1979         WARN_ON(lreq->osd != osd);
1980         dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd,
1981              osd->o_osd, lreq, lreq->linger_id);
1982
1983         lreq->osd = NULL;
1984         erase_linger(&osd->o_linger_requests, lreq);
1985         put_osd(osd);
1986
1987         if (!osd_homeless(osd))
1988                 maybe_move_osd_to_lru(osd);
1989         else
1990                 atomic_dec(&osd->o_osdc->num_homeless);
1991 }
1992
1993 static bool __linger_registered(struct ceph_osd_linger_request *lreq)
1994 {
1995         verify_osdc_locked(lreq->osdc);
1996
1997         return !RB_EMPTY_NODE(&lreq->osdc_node);
1998 }
1999
2000 static bool linger_registered(struct ceph_osd_linger_request *lreq)
2001 {
2002         struct ceph_osd_client *osdc = lreq->osdc;
2003         bool registered;
2004
2005         down_read(&osdc->lock);
2006         registered = __linger_registered(lreq);
2007         up_read(&osdc->lock);
2008
2009         return registered;
2010 }
2011
2012 static void linger_register(struct ceph_osd_linger_request *lreq)
2013 {
2014         struct ceph_osd_client *osdc = lreq->osdc;
2015
2016         verify_osdc_wrlocked(osdc);
2017         WARN_ON(lreq->linger_id);
2018
2019         linger_get(lreq);
2020         lreq->linger_id = ++osdc->last_linger_id;
2021         insert_linger_osdc(&osdc->linger_requests, lreq);
2022 }
2023
2024 static void linger_unregister(struct ceph_osd_linger_request *lreq)
2025 {
2026         struct ceph_osd_client *osdc = lreq->osdc;
2027
2028         verify_osdc_wrlocked(osdc);
2029
2030         erase_linger_osdc(&osdc->linger_requests, lreq);
2031         linger_put(lreq);
2032 }
2033
2034 static void cancel_linger_request(struct ceph_osd_request *req)
2035 {
2036         struct ceph_osd_linger_request *lreq = req->r_priv;
2037
2038         WARN_ON(!req->r_linger);
2039         cancel_request(req);
2040         linger_put(lreq);
2041 }
2042
2043 struct linger_work {
2044         struct work_struct work;
2045         struct ceph_osd_linger_request *lreq;
2046         struct list_head pending_item;
2047         unsigned long queued_stamp;
2048
2049         union {
2050                 struct {
2051                         u64 notify_id;
2052                         u64 notifier_id;
2053                         void *payload; /* points into @msg front */
2054                         size_t payload_len;
2055
2056                         struct ceph_msg *msg; /* for ceph_msg_put() */
2057                 } notify;
2058                 struct {
2059                         int err;
2060                 } error;
2061         };
2062 };
2063
2064 static struct linger_work *lwork_alloc(struct ceph_osd_linger_request *lreq,
2065                                        work_func_t workfn)
2066 {
2067         struct linger_work *lwork;
2068
2069         lwork = kzalloc(sizeof(*lwork), GFP_NOIO);
2070         if (!lwork)
2071                 return NULL;
2072
2073         INIT_WORK(&lwork->work, workfn);
2074         INIT_LIST_HEAD(&lwork->pending_item);
2075         lwork->lreq = linger_get(lreq);
2076
2077         return lwork;
2078 }
2079
2080 static void lwork_free(struct linger_work *lwork)
2081 {
2082         struct ceph_osd_linger_request *lreq = lwork->lreq;
2083
2084         mutex_lock(&lreq->lock);
2085         list_del(&lwork->pending_item);
2086         mutex_unlock(&lreq->lock);
2087
2088         linger_put(lreq);
2089         kfree(lwork);
2090 }
2091
2092 static void lwork_queue(struct linger_work *lwork)
2093 {
2094         struct ceph_osd_linger_request *lreq = lwork->lreq;
2095         struct ceph_osd_client *osdc = lreq->osdc;
2096
2097         verify_lreq_locked(lreq);
2098         WARN_ON(!list_empty(&lwork->pending_item));
2099
2100         lwork->queued_stamp = jiffies;
2101         list_add_tail(&lwork->pending_item, &lreq->pending_lworks);
2102         queue_work(osdc->notify_wq, &lwork->work);
2103 }
2104
2105 static void do_watch_notify(struct work_struct *w)
2106 {
2107         struct linger_work *lwork = container_of(w, struct linger_work, work);
2108         struct ceph_osd_linger_request *lreq = lwork->lreq;
2109
2110         if (!linger_registered(lreq)) {
2111                 dout("%s lreq %p not registered\n", __func__, lreq);
2112                 goto out;
2113         }
2114
2115         WARN_ON(!lreq->is_watch);
2116         dout("%s lreq %p notify_id %llu notifier_id %llu payload_len %zu\n",
2117              __func__, lreq, lwork->notify.notify_id, lwork->notify.notifier_id,
2118              lwork->notify.payload_len);
2119         lreq->wcb(lreq->data, lwork->notify.notify_id, lreq->linger_id,
2120                   lwork->notify.notifier_id, lwork->notify.payload,
2121                   lwork->notify.payload_len);
2122
2123 out:
2124         ceph_msg_put(lwork->notify.msg);
2125         lwork_free(lwork);
2126 }
2127
2128 static void do_watch_error(struct work_struct *w)
2129 {
2130         struct linger_work *lwork = container_of(w, struct linger_work, work);
2131         struct ceph_osd_linger_request *lreq = lwork->lreq;
2132
2133         if (!linger_registered(lreq)) {
2134                 dout("%s lreq %p not registered\n", __func__, lreq);
2135                 goto out;
2136         }
2137
2138         dout("%s lreq %p err %d\n", __func__, lreq, lwork->error.err);
2139         lreq->errcb(lreq->data, lreq->linger_id, lwork->error.err);
2140
2141 out:
2142         lwork_free(lwork);
2143 }
2144
2145 static void queue_watch_error(struct ceph_osd_linger_request *lreq)
2146 {
2147         struct linger_work *lwork;
2148
2149         lwork = lwork_alloc(lreq, do_watch_error);
2150         if (!lwork) {
2151                 pr_err("failed to allocate error-lwork\n");
2152                 return;
2153         }
2154
2155         lwork->error.err = lreq->last_error;
2156         lwork_queue(lwork);
2157 }
2158
2159 static void linger_reg_commit_complete(struct ceph_osd_linger_request *lreq,
2160                                        int result)
2161 {
2162         if (!completion_done(&lreq->reg_commit_wait)) {
2163                 lreq->reg_commit_error = (result <= 0 ? result : 0);
2164                 complete_all(&lreq->reg_commit_wait);
2165         }
2166 }
2167
2168 static void linger_commit_cb(struct ceph_osd_request *req)
2169 {
2170         struct ceph_osd_linger_request *lreq = req->r_priv;
2171
2172         mutex_lock(&lreq->lock);
2173         dout("%s lreq %p linger_id %llu result %d\n", __func__, lreq,
2174              lreq->linger_id, req->r_result);
2175         linger_reg_commit_complete(lreq, req->r_result);
2176         lreq->committed = true;
2177
2178         if (!lreq->is_watch) {
2179                 struct ceph_osd_data *osd_data =
2180                     osd_req_op_data(req, 0, notify, response_data);
2181                 void *p = page_address(osd_data->pages[0]);
2182
2183                 WARN_ON(req->r_ops[0].op != CEPH_OSD_OP_NOTIFY ||
2184                         osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
2185
2186                 /* make note of the notify_id */
2187                 if (req->r_ops[0].outdata_len >= sizeof(u64)) {
2188                         lreq->notify_id = ceph_decode_64(&p);
2189                         dout("lreq %p notify_id %llu\n", lreq,
2190                              lreq->notify_id);
2191                 } else {
2192                         dout("lreq %p no notify_id\n", lreq);
2193                 }
2194         }
2195
2196         mutex_unlock(&lreq->lock);
2197         linger_put(lreq);
2198 }
2199
2200 static int normalize_watch_error(int err)
2201 {
2202         /*
2203          * Translate ENOENT -> ENOTCONN so that a delete->disconnection
2204          * notification and a failure to reconnect because we raced with
2205          * the delete appear the same to the user.
2206          */
2207         if (err == -ENOENT)
2208                 err = -ENOTCONN;
2209
2210         return err;
2211 }
2212
2213 static void linger_reconnect_cb(struct ceph_osd_request *req)
2214 {
2215         struct ceph_osd_linger_request *lreq = req->r_priv;
2216
2217         mutex_lock(&lreq->lock);
2218         dout("%s lreq %p linger_id %llu result %d last_error %d\n", __func__,
2219              lreq, lreq->linger_id, req->r_result, lreq->last_error);
2220         if (req->r_result < 0) {
2221                 if (!lreq->last_error) {
2222                         lreq->last_error = normalize_watch_error(req->r_result);
2223                         queue_watch_error(lreq);
2224                 }
2225         }
2226
2227         mutex_unlock(&lreq->lock);
2228         linger_put(lreq);
2229 }
2230
2231 static void send_linger(struct ceph_osd_linger_request *lreq)
2232 {
2233         struct ceph_osd_request *req = lreq->reg_req;
2234         struct ceph_osd_req_op *op = &req->r_ops[0];
2235
2236         verify_osdc_wrlocked(req->r_osdc);
2237         dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
2238
2239         if (req->r_osd)
2240                 cancel_linger_request(req);
2241
2242         request_reinit(req);
2243         ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
2244         ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
2245         req->r_flags = lreq->t.flags;
2246         req->r_mtime = lreq->mtime;
2247
2248         mutex_lock(&lreq->lock);
2249         if (lreq->is_watch && lreq->committed) {
2250                 WARN_ON(op->op != CEPH_OSD_OP_WATCH ||
2251                         op->watch.cookie != lreq->linger_id);
2252                 op->watch.op = CEPH_OSD_WATCH_OP_RECONNECT;
2253                 op->watch.gen = ++lreq->register_gen;
2254                 dout("lreq %p reconnect register_gen %u\n", lreq,
2255                      op->watch.gen);
2256                 req->r_callback = linger_reconnect_cb;
2257         } else {
2258                 if (!lreq->is_watch)
2259                         lreq->notify_id = 0;
2260                 else
2261                         WARN_ON(op->watch.op != CEPH_OSD_WATCH_OP_WATCH);
2262                 dout("lreq %p register\n", lreq);
2263                 req->r_callback = linger_commit_cb;
2264         }
2265         mutex_unlock(&lreq->lock);
2266
2267         req->r_priv = linger_get(lreq);
2268         req->r_linger = true;
2269
2270         submit_request(req, true);
2271 }
2272
2273 static void linger_ping_cb(struct ceph_osd_request *req)
2274 {
2275         struct ceph_osd_linger_request *lreq = req->r_priv;
2276
2277         mutex_lock(&lreq->lock);
2278         dout("%s lreq %p linger_id %llu result %d ping_sent %lu last_error %d\n",
2279              __func__, lreq, lreq->linger_id, req->r_result, lreq->ping_sent,
2280              lreq->last_error);
2281         if (lreq->register_gen == req->r_ops[0].watch.gen) {
2282                 if (!req->r_result) {
2283                         lreq->watch_valid_thru = lreq->ping_sent;
2284                 } else if (!lreq->last_error) {
2285                         lreq->last_error = normalize_watch_error(req->r_result);
2286                         queue_watch_error(lreq);
2287                 }
2288         } else {
2289                 dout("lreq %p register_gen %u ignoring old pong %u\n", lreq,
2290                      lreq->register_gen, req->r_ops[0].watch.gen);
2291         }
2292
2293         mutex_unlock(&lreq->lock);
2294         linger_put(lreq);
2295 }
2296
2297 static void send_linger_ping(struct ceph_osd_linger_request *lreq)
2298 {
2299         struct ceph_osd_client *osdc = lreq->osdc;
2300         struct ceph_osd_request *req = lreq->ping_req;
2301         struct ceph_osd_req_op *op = &req->r_ops[0];
2302
2303         if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) {
2304                 dout("%s PAUSERD\n", __func__);
2305                 return;
2306         }
2307
2308         lreq->ping_sent = jiffies;
2309         dout("%s lreq %p linger_id %llu ping_sent %lu register_gen %u\n",
2310              __func__, lreq, lreq->linger_id, lreq->ping_sent,
2311              lreq->register_gen);
2312
2313         if (req->r_osd)
2314                 cancel_linger_request(req);
2315
2316         request_reinit(req);
2317         target_copy(&req->r_t, &lreq->t);
2318
2319         WARN_ON(op->op != CEPH_OSD_OP_WATCH ||
2320                 op->watch.cookie != lreq->linger_id ||
2321                 op->watch.op != CEPH_OSD_WATCH_OP_PING);
2322         op->watch.gen = lreq->register_gen;
2323         req->r_callback = linger_ping_cb;
2324         req->r_priv = linger_get(lreq);
2325         req->r_linger = true;
2326
2327         ceph_osdc_get_request(req);
2328         account_request(req);
2329         req->r_tid = atomic64_inc_return(&osdc->last_tid);
2330         link_request(lreq->osd, req);
2331         send_request(req);
2332 }
2333
2334 static void linger_submit(struct ceph_osd_linger_request *lreq)
2335 {
2336         struct ceph_osd_client *osdc = lreq->osdc;
2337         struct ceph_osd *osd;
2338
2339         calc_target(osdc, &lreq->t, &lreq->last_force_resend, false);
2340         osd = lookup_create_osd(osdc, lreq->t.osd, true);
2341         link_linger(osd, lreq);
2342
2343         send_linger(lreq);
2344 }
2345
2346 static void cancel_linger_map_check(struct ceph_osd_linger_request *lreq)
2347 {
2348         struct ceph_osd_client *osdc = lreq->osdc;
2349         struct ceph_osd_linger_request *lookup_lreq;
2350
2351         verify_osdc_wrlocked(osdc);
2352
2353         lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks,
2354                                        lreq->linger_id);
2355         if (!lookup_lreq)
2356                 return;
2357
2358         WARN_ON(lookup_lreq != lreq);
2359         erase_linger_mc(&osdc->linger_map_checks, lreq);
2360         linger_put(lreq);
2361 }
2362
2363 /*
2364  * @lreq has to be both registered and linked.
2365  */
2366 static void __linger_cancel(struct ceph_osd_linger_request *lreq)
2367 {
2368         if (lreq->is_watch && lreq->ping_req->r_osd)
2369                 cancel_linger_request(lreq->ping_req);
2370         if (lreq->reg_req->r_osd)
2371                 cancel_linger_request(lreq->reg_req);
2372         cancel_linger_map_check(lreq);
2373         unlink_linger(lreq->osd, lreq);
2374         linger_unregister(lreq);
2375 }
2376
2377 static void linger_cancel(struct ceph_osd_linger_request *lreq)
2378 {
2379         struct ceph_osd_client *osdc = lreq->osdc;
2380
2381         down_write(&osdc->lock);
2382         if (__linger_registered(lreq))
2383                 __linger_cancel(lreq);
2384         up_write(&osdc->lock);
2385 }
2386
2387 static void send_linger_map_check(struct ceph_osd_linger_request *lreq);
2388
2389 static void check_linger_pool_dne(struct ceph_osd_linger_request *lreq)
2390 {
2391         struct ceph_osd_client *osdc = lreq->osdc;
2392         struct ceph_osdmap *map = osdc->osdmap;
2393
2394         verify_osdc_wrlocked(osdc);
2395         WARN_ON(!map->epoch);
2396
2397         if (lreq->register_gen) {
2398                 lreq->map_dne_bound = map->epoch;
2399                 dout("%s lreq %p linger_id %llu pool disappeared\n", __func__,
2400                      lreq, lreq->linger_id);
2401         } else {
2402                 dout("%s lreq %p linger_id %llu map_dne_bound %u have %u\n",
2403                      __func__, lreq, lreq->linger_id, lreq->map_dne_bound,
2404                      map->epoch);
2405         }
2406
2407         if (lreq->map_dne_bound) {
2408                 if (map->epoch >= lreq->map_dne_bound) {
2409                         /* we had a new enough map */
2410                         pr_info("linger_id %llu pool does not exist\n",
2411                                 lreq->linger_id);
2412                         linger_reg_commit_complete(lreq, -ENOENT);
2413                         __linger_cancel(lreq);
2414                 }
2415         } else {
2416                 send_linger_map_check(lreq);
2417         }
2418 }
2419
2420 static void linger_map_check_cb(struct ceph_mon_generic_request *greq)
2421 {
2422         struct ceph_osd_client *osdc = &greq->monc->client->osdc;
2423         struct ceph_osd_linger_request *lreq;
2424         u64 linger_id = greq->private_data;
2425
2426         WARN_ON(greq->result || !greq->u.newest);
2427
2428         down_write(&osdc->lock);
2429         lreq = lookup_linger_mc(&osdc->linger_map_checks, linger_id);
2430         if (!lreq) {
2431                 dout("%s linger_id %llu dne\n", __func__, linger_id);
2432                 goto out_unlock;
2433         }
2434
2435         dout("%s lreq %p linger_id %llu map_dne_bound %u newest %llu\n",
2436              __func__, lreq, lreq->linger_id, lreq->map_dne_bound,
2437              greq->u.newest);
2438         if (!lreq->map_dne_bound)
2439                 lreq->map_dne_bound = greq->u.newest;
2440         erase_linger_mc(&osdc->linger_map_checks, lreq);
2441         check_linger_pool_dne(lreq);
2442
2443         linger_put(lreq);
2444 out_unlock:
2445         up_write(&osdc->lock);
2446 }
2447
2448 static void send_linger_map_check(struct ceph_osd_linger_request *lreq)
2449 {
2450         struct ceph_osd_client *osdc = lreq->osdc;
2451         struct ceph_osd_linger_request *lookup_lreq;
2452         int ret;
2453
2454         verify_osdc_wrlocked(osdc);
2455
2456         lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks,
2457                                        lreq->linger_id);
2458         if (lookup_lreq) {
2459                 WARN_ON(lookup_lreq != lreq);
2460                 return;
2461         }
2462
2463         linger_get(lreq);
2464         insert_linger_mc(&osdc->linger_map_checks, lreq);
2465         ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap",
2466                                           linger_map_check_cb, lreq->linger_id);
2467         WARN_ON(ret);
2468 }
2469
2470 static int linger_reg_commit_wait(struct ceph_osd_linger_request *lreq)
2471 {
2472         int ret;
2473
2474         dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
2475         ret = wait_for_completion_interruptible(&lreq->reg_commit_wait);
2476         return ret ?: lreq->reg_commit_error;
2477 }
2478
2479 static int linger_notify_finish_wait(struct ceph_osd_linger_request *lreq)
2480 {
2481         int ret;
2482
2483         dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
2484         ret = wait_for_completion_interruptible(&lreq->notify_finish_wait);
2485         return ret ?: lreq->notify_finish_error;
2486 }
2487
2488 /*
2489  * Timeout callback, called every N seconds.  When 1 or more OSD
2490  * requests has been active for more than N seconds, we send a keepalive
2491  * (tag + timestamp) to its OSD to ensure any communications channel
2492  * reset is detected.
2493  */
2494 static void handle_timeout(struct work_struct *work)
2495 {
2496         struct ceph_osd_client *osdc =
2497                 container_of(work, struct ceph_osd_client, timeout_work.work);
2498         struct ceph_options *opts = osdc->client->options;
2499         unsigned long cutoff = jiffies - opts->osd_keepalive_timeout;
2500         unsigned long expiry_cutoff = jiffies - opts->osd_request_timeout;
2501         LIST_HEAD(slow_osds);
2502         struct rb_node *n, *p;
2503
2504         dout("%s osdc %p\n", __func__, osdc);
2505         down_write(&osdc->lock);
2506
2507         /*
2508          * ping osds that are a bit slow.  this ensures that if there
2509          * is a break in the TCP connection we will notice, and reopen
2510          * a connection with that osd (from the fault callback).
2511          */
2512         for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
2513                 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
2514                 bool found = false;
2515
2516                 for (p = rb_first(&osd->o_requests); p; ) {
2517                         struct ceph_osd_request *req =
2518                             rb_entry(p, struct ceph_osd_request, r_node);
2519
2520                         p = rb_next(p); /* abort_request() */
2521
2522                         if (time_before(req->r_stamp, cutoff)) {
2523                                 dout(" req %p tid %llu on osd%d is laggy\n",
2524                                      req, req->r_tid, osd->o_osd);
2525                                 found = true;
2526                         }
2527                         if (opts->osd_request_timeout &&
2528                             time_before(req->r_start_stamp, expiry_cutoff)) {
2529                                 pr_err_ratelimited("tid %llu on osd%d timeout\n",
2530                                        req->r_tid, osd->o_osd);
2531                                 abort_request(req, -ETIMEDOUT);
2532                         }
2533                 }
2534                 for (p = rb_first(&osd->o_linger_requests); p; p = rb_next(p)) {
2535                         struct ceph_osd_linger_request *lreq =
2536                             rb_entry(p, struct ceph_osd_linger_request, node);
2537
2538                         dout(" lreq %p linger_id %llu is served by osd%d\n",
2539                              lreq, lreq->linger_id, osd->o_osd);
2540                         found = true;
2541
2542                         mutex_lock(&lreq->lock);
2543                         if (lreq->is_watch && lreq->committed && !lreq->last_error)
2544                                 send_linger_ping(lreq);
2545                         mutex_unlock(&lreq->lock);
2546                 }
2547
2548                 if (found)
2549                         list_move_tail(&osd->o_keepalive_item, &slow_osds);
2550         }
2551
2552         if (opts->osd_request_timeout) {
2553                 for (p = rb_first(&osdc->homeless_osd.o_requests); p; ) {
2554                         struct ceph_osd_request *req =
2555                             rb_entry(p, struct ceph_osd_request, r_node);
2556
2557                         p = rb_next(p); /* abort_request() */
2558
2559                         if (time_before(req->r_start_stamp, expiry_cutoff)) {
2560                                 pr_err_ratelimited("tid %llu on osd%d timeout\n",
2561                                        req->r_tid, osdc->homeless_osd.o_osd);
2562                                 abort_request(req, -ETIMEDOUT);
2563                         }
2564                 }
2565         }
2566
2567         if (atomic_read(&osdc->num_homeless) || !list_empty(&slow_osds))
2568                 maybe_request_map(osdc);
2569
2570         while (!list_empty(&slow_osds)) {
2571                 struct ceph_osd *osd = list_first_entry(&slow_osds,
2572                                                         struct ceph_osd,
2573                                                         o_keepalive_item);
2574                 list_del_init(&osd->o_keepalive_item);
2575                 ceph_con_keepalive(&osd->o_con);
2576         }
2577
2578         up_write(&osdc->lock);
2579         schedule_delayed_work(&osdc->timeout_work,
2580                               osdc->client->options->osd_keepalive_timeout);
2581 }
2582
2583 static void handle_osds_timeout(struct work_struct *work)
2584 {
2585         struct ceph_osd_client *osdc =
2586                 container_of(work, struct ceph_osd_client,
2587                              osds_timeout_work.work);
2588         unsigned long delay = osdc->client->options->osd_idle_ttl / 4;
2589         struct ceph_osd *osd, *nosd;
2590
2591         dout("%s osdc %p\n", __func__, osdc);
2592         down_write(&osdc->lock);
2593         list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
2594                 if (time_before(jiffies, osd->lru_ttl))
2595                         break;
2596
2597                 WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests));
2598                 WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests));
2599                 close_osd(osd);
2600         }
2601
2602         up_write(&osdc->lock);
2603         schedule_delayed_work(&osdc->osds_timeout_work,
2604                               round_jiffies_relative(delay));
2605 }
2606
2607 static int ceph_oloc_decode(void **p, void *end,
2608                             struct ceph_object_locator *oloc)
2609 {
2610         u8 struct_v, struct_cv;
2611         u32 len;
2612         void *struct_end;
2613         int ret = 0;
2614
2615         ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
2616         struct_v = ceph_decode_8(p);
2617         struct_cv = ceph_decode_8(p);
2618         if (struct_v < 3) {
2619                 pr_warn("got v %d < 3 cv %d of ceph_object_locator\n",
2620                         struct_v, struct_cv);
2621                 goto e_inval;
2622         }
2623         if (struct_cv > 6) {
2624                 pr_warn("got v %d cv %d > 6 of ceph_object_locator\n",
2625                         struct_v, struct_cv);
2626                 goto e_inval;
2627         }
2628         len = ceph_decode_32(p);
2629         ceph_decode_need(p, end, len, e_inval);
2630         struct_end = *p + len;
2631
2632         oloc->pool = ceph_decode_64(p);
2633         *p += 4; /* skip preferred */
2634
2635         len = ceph_decode_32(p);
2636         if (len > 0) {
2637                 pr_warn("ceph_object_locator::key is set\n");
2638                 goto e_inval;
2639         }
2640
2641         if (struct_v >= 5) {
2642                 bool changed = false;
2643
2644                 len = ceph_decode_32(p);
2645                 if (len > 0) {
2646                         ceph_decode_need(p, end, len, e_inval);
2647                         if (!oloc->pool_ns ||
2648                             ceph_compare_string(oloc->pool_ns, *p, len))
2649                                 changed = true;
2650                         *p += len;
2651                 } else {
2652                         if (oloc->pool_ns)
2653                                 changed = true;
2654                 }
2655                 if (changed) {
2656                         /* redirect changes namespace */
2657                         pr_warn("ceph_object_locator::nspace is changed\n");
2658                         goto e_inval;
2659                 }
2660         }
2661
2662         if (struct_v >= 6) {
2663                 s64 hash = ceph_decode_64(p);
2664                 if (hash != -1) {
2665                         pr_warn("ceph_object_locator::hash is set\n");
2666                         goto e_inval;
2667                 }
2668         }
2669
2670         /* skip the rest */
2671         *p = struct_end;
2672 out:
2673         return ret;
2674
2675 e_inval:
2676         ret = -EINVAL;
2677         goto out;
2678 }
2679
2680 static int ceph_redirect_decode(void **p, void *end,
2681                                 struct ceph_request_redirect *redir)
2682 {
2683         u8 struct_v, struct_cv;
2684         u32 len;
2685         void *struct_end;
2686         int ret;
2687
2688         ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
2689         struct_v = ceph_decode_8(p);
2690         struct_cv = ceph_decode_8(p);
2691         if (struct_cv > 1) {
2692                 pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n",
2693                         struct_v, struct_cv);
2694                 goto e_inval;
2695         }
2696         len = ceph_decode_32(p);
2697         ceph_decode_need(p, end, len, e_inval);
2698         struct_end = *p + len;
2699
2700         ret = ceph_oloc_decode(p, end, &redir->oloc);
2701         if (ret)
2702                 goto out;
2703
2704         len = ceph_decode_32(p);
2705         if (len > 0) {
2706                 pr_warn("ceph_request_redirect::object_name is set\n");
2707                 goto e_inval;
2708         }
2709
2710         len = ceph_decode_32(p);
2711         *p += len; /* skip osd_instructions */
2712
2713         /* skip the rest */
2714         *p = struct_end;
2715 out:
2716         return ret;
2717
2718 e_inval:
2719         ret = -EINVAL;
2720         goto out;
2721 }
2722
2723 struct MOSDOpReply {
2724         struct ceph_pg pgid;
2725         u64 flags;
2726         int result;
2727         u32 epoch;
2728         int num_ops;
2729         u32 outdata_len[CEPH_OSD_MAX_OPS];
2730         s32 rval[CEPH_OSD_MAX_OPS];
2731         int retry_attempt;
2732         struct ceph_eversion replay_version;
2733         u64 user_version;
2734         struct ceph_request_redirect redirect;
2735 };
2736
2737 static int decode_MOSDOpReply(const struct ceph_msg *msg, struct MOSDOpReply *m)
2738 {
2739         void *p = msg->front.iov_base;
2740         void *const end = p + msg->front.iov_len;
2741         u16 version = le16_to_cpu(msg->hdr.version);
2742         struct ceph_eversion bad_replay_version;
2743         u8 decode_redir;
2744         u32 len;
2745         int ret;
2746         int i;
2747
2748         ceph_decode_32_safe(&p, end, len, e_inval);
2749         ceph_decode_need(&p, end, len, e_inval);
2750         p += len; /* skip oid */
2751
2752         ret = ceph_decode_pgid(&p, end, &m->pgid);
2753         if (ret)
2754                 return ret;
2755
2756         ceph_decode_64_safe(&p, end, m->flags, e_inval);
2757         ceph_decode_32_safe(&p, end, m->result, e_inval);
2758         ceph_decode_need(&p, end, sizeof(bad_replay_version), e_inval);
2759         memcpy(&bad_replay_version, p, sizeof(bad_replay_version));
2760         p += sizeof(bad_replay_version);
2761         ceph_decode_32_safe(&p, end, m->epoch, e_inval);
2762
2763         ceph_decode_32_safe(&p, end, m->num_ops, e_inval);
2764         if (m->num_ops > ARRAY_SIZE(m->outdata_len))
2765                 goto e_inval;
2766
2767         ceph_decode_need(&p, end, m->num_ops * sizeof(struct ceph_osd_op),
2768                          e_inval);
2769         for (i = 0; i < m->num_ops; i++) {
2770                 struct ceph_osd_op *op = p;
2771
2772                 m->outdata_len[i] = le32_to_cpu(op->payload_len);
2773                 p += sizeof(*op);
2774         }
2775
2776         ceph_decode_32_safe(&p, end, m->retry_attempt, e_inval);
2777         for (i = 0; i < m->num_ops; i++)
2778                 ceph_decode_32_safe(&p, end, m->rval[i], e_inval);
2779
2780         if (version >= 5) {
2781                 ceph_decode_need(&p, end, sizeof(m->replay_version), e_inval);
2782                 memcpy(&m->replay_version, p, sizeof(m->replay_version));
2783                 p += sizeof(m->replay_version);
2784                 ceph_decode_64_safe(&p, end, m->user_version, e_inval);
2785         } else {
2786                 m->replay_version = bad_replay_version; /* struct */
2787                 m->user_version = le64_to_cpu(m->replay_version.version);
2788         }
2789
2790         if (version >= 6) {
2791                 if (version >= 7)
2792                         ceph_decode_8_safe(&p, end, decode_redir, e_inval);
2793                 else
2794                         decode_redir = 1;
2795         } else {
2796                 decode_redir = 0;
2797         }
2798
2799         if (decode_redir) {
2800                 ret = ceph_redirect_decode(&p, end, &m->redirect);
2801                 if (ret)
2802                         return ret;
2803         } else {
2804                 ceph_oloc_init(&m->redirect.oloc);
2805         }
2806
2807         return 0;
2808
2809 e_inval:
2810         return -EINVAL;
2811 }
2812
2813 /*
2814  * Handle MOSDOpReply.  Set ->r_result and call the callback if it is
2815  * specified.
2816  */
2817 static void handle_reply(struct ceph_osd *osd, struct ceph_msg *msg)
2818 {
2819         struct ceph_osd_client *osdc = osd->o_osdc;
2820         struct ceph_osd_request *req;
2821         struct MOSDOpReply m;
2822         u64 tid = le64_to_cpu(msg->hdr.tid);
2823         u32 data_len = 0;
2824         int ret;
2825         int i;
2826
2827         dout("%s msg %p tid %llu\n", __func__, msg, tid);
2828
2829         down_read(&osdc->lock);
2830         if (!osd_registered(osd)) {
2831                 dout("%s osd%d unknown\n", __func__, osd->o_osd);
2832                 goto out_unlock_osdc;
2833         }
2834         WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num));
2835
2836         mutex_lock(&osd->lock);
2837         req = lookup_request(&osd->o_requests, tid);
2838         if (!req) {
2839                 dout("%s osd%d tid %llu unknown\n", __func__, osd->o_osd, tid);
2840                 goto out_unlock_session;
2841         }
2842
2843         m.redirect.oloc.pool_ns = req->r_t.target_oloc.pool_ns;
2844         ret = decode_MOSDOpReply(msg, &m);
2845         m.redirect.oloc.pool_ns = NULL;
2846         if (ret) {
2847                 pr_err("failed to decode MOSDOpReply for tid %llu: %d\n",
2848                        req->r_tid, ret);
2849                 ceph_msg_dump(msg);
2850                 goto fail_request;
2851         }
2852         dout("%s req %p tid %llu flags 0x%llx pgid %llu.%x epoch %u attempt %d v %u'%llu uv %llu\n",
2853              __func__, req, req->r_tid, m.flags, m.pgid.pool, m.pgid.seed,
2854              m.epoch, m.retry_attempt, le32_to_cpu(m.replay_version.epoch),
2855              le64_to_cpu(m.replay_version.version), m.user_version);
2856
2857         if (m.retry_attempt >= 0) {
2858                 if (m.retry_attempt != req->r_attempts - 1) {
2859                         dout("req %p tid %llu retry_attempt %d != %d, ignoring\n",
2860                              req, req->r_tid, m.retry_attempt,
2861                              req->r_attempts - 1);
2862                         goto out_unlock_session;
2863                 }
2864         } else {
2865                 WARN_ON(1); /* MOSDOpReply v4 is assumed */
2866         }
2867
2868         if (!ceph_oloc_empty(&m.redirect.oloc)) {
2869                 dout("req %p tid %llu redirect pool %lld\n", req, req->r_tid,
2870                      m.redirect.oloc.pool);
2871                 unlink_request(osd, req);
2872                 mutex_unlock(&osd->lock);
2873
2874                 /*
2875                  * Not ceph_oloc_copy() - changing pool_ns is not
2876                  * supported.
2877                  */
2878                 req->r_t.target_oloc.pool = m.redirect.oloc.pool;
2879                 req->r_flags |= CEPH_OSD_FLAG_REDIRECTED;
2880                 req->r_tid = 0;
2881                 __submit_request(req, false);
2882                 goto out_unlock_osdc;
2883         }
2884
2885         if (m.num_ops != req->r_num_ops) {
2886                 pr_err("num_ops %d != %d for tid %llu\n", m.num_ops,
2887                        req->r_num_ops, req->r_tid);
2888                 goto fail_request;
2889         }
2890         for (i = 0; i < req->r_num_ops; i++) {
2891                 dout(" req %p tid %llu op %d rval %d len %u\n", req,
2892                      req->r_tid, i, m.rval[i], m.outdata_len[i]);
2893                 req->r_ops[i].rval = m.rval[i];
2894                 req->r_ops[i].outdata_len = m.outdata_len[i];
2895                 data_len += m.outdata_len[i];
2896         }
2897         if (data_len != le32_to_cpu(msg->hdr.data_len)) {
2898                 pr_err("sum of lens %u != %u for tid %llu\n", data_len,
2899                        le32_to_cpu(msg->hdr.data_len), req->r_tid);
2900                 goto fail_request;
2901         }
2902         dout("%s req %p tid %llu result %d data_len %u\n", __func__,
2903              req, req->r_tid, m.result, data_len);
2904
2905         /*
2906          * Since we only ever request ONDISK, we should only ever get
2907          * one (type of) reply back.
2908          */
2909         WARN_ON(!(m.flags & CEPH_OSD_FLAG_ONDISK));
2910         req->r_result = m.result ?: data_len;
2911         finish_request(req);
2912         mutex_unlock(&osd->lock);
2913         up_read(&osdc->lock);
2914
2915         __complete_request(req);
2916         complete_all(&req->r_completion);
2917         ceph_osdc_put_request(req);
2918         return;
2919
2920 fail_request:
2921         complete_request(req, -EIO);
2922 out_unlock_session:
2923         mutex_unlock(&osd->lock);
2924 out_unlock_osdc:
2925         up_read(&osdc->lock);
2926 }
2927
2928 static void set_pool_was_full(struct ceph_osd_client *osdc)
2929 {
2930         struct rb_node *n;
2931
2932         for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) {
2933                 struct ceph_pg_pool_info *pi =
2934                     rb_entry(n, struct ceph_pg_pool_info, node);
2935
2936                 pi->was_full = __pool_full(pi);
2937         }
2938 }
2939
2940 static bool pool_cleared_full(struct ceph_osd_client *osdc, s64 pool_id)
2941 {
2942         struct ceph_pg_pool_info *pi;
2943
2944         pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id);
2945         if (!pi)
2946                 return false;
2947
2948         return pi->was_full && !__pool_full(pi);
2949 }
2950
2951 static enum calc_target_result
2952 recalc_linger_target(struct ceph_osd_linger_request *lreq)
2953 {
2954         struct ceph_osd_client *osdc = lreq->osdc;
2955         enum calc_target_result ct_res;
2956
2957         ct_res = calc_target(osdc, &lreq->t, &lreq->last_force_resend, true);
2958         if (ct_res == CALC_TARGET_NEED_RESEND) {
2959                 struct ceph_osd *osd;
2960
2961                 osd = lookup_create_osd(osdc, lreq->t.osd, true);
2962                 if (osd != lreq->osd) {
2963                         unlink_linger(lreq->osd, lreq);
2964                         link_linger(osd, lreq);
2965                 }
2966         }
2967
2968         return ct_res;
2969 }
2970
2971 /*
2972  * Requeue requests whose mapping to an OSD has changed.
2973  */
2974 static void scan_requests(struct ceph_osd *osd,
2975                           bool force_resend,
2976                           bool cleared_full,
2977                           bool check_pool_cleared_full,
2978                           struct rb_root *need_resend,
2979                           struct list_head *need_resend_linger)
2980 {
2981         struct ceph_osd_client *osdc = osd->o_osdc;
2982         struct rb_node *n;
2983         bool force_resend_writes;
2984
2985         for (n = rb_first(&osd->o_linger_requests); n; ) {
2986                 struct ceph_osd_linger_request *lreq =
2987                     rb_entry(n, struct ceph_osd_linger_request, node);
2988                 enum calc_target_result ct_res;
2989
2990                 n = rb_next(n); /* recalc_linger_target() */
2991
2992                 dout("%s lreq %p linger_id %llu\n", __func__, lreq,
2993                      lreq->linger_id);
2994                 ct_res = recalc_linger_target(lreq);
2995                 switch (ct_res) {
2996                 case CALC_TARGET_NO_ACTION:
2997                         force_resend_writes = cleared_full ||
2998                             (check_pool_cleared_full &&
2999                              pool_cleared_full(osdc, lreq->t.base_oloc.pool));
3000                         if (!force_resend && !force_resend_writes)
3001                                 break;
3002
3003                         /* fall through */
3004                 case CALC_TARGET_NEED_RESEND:
3005                         cancel_linger_map_check(lreq);
3006                         /*
3007                          * scan_requests() for the previous epoch(s)
3008                          * may have already added it to the list, since
3009                          * it's not unlinked here.
3010                          */
3011                         if (list_empty(&lreq->scan_item))
3012                                 list_add_tail(&lreq->scan_item, need_resend_linger);
3013                         break;
3014                 case CALC_TARGET_POOL_DNE:
3015                         check_linger_pool_dne(lreq);
3016                         break;
3017                 }
3018         }
3019
3020         for (n = rb_first(&osd->o_requests); n; ) {
3021                 struct ceph_osd_request *req =
3022                     rb_entry(n, struct ceph_osd_request, r_node);
3023                 enum calc_target_result ct_res;
3024
3025                 n = rb_next(n); /* unlink_request(), check_pool_dne() */
3026
3027                 dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
3028                 ct_res = calc_target(osdc, &req->r_t,
3029                                      &req->r_last_force_resend, false);
3030                 switch (ct_res) {
3031                 case CALC_TARGET_NO_ACTION:
3032                         force_resend_writes = cleared_full ||
3033                             (check_pool_cleared_full &&
3034                              pool_cleared_full(osdc, req->r_t.base_oloc.pool));
3035                         if (!force_resend &&
3036                             (!(req->r_flags & CEPH_OSD_FLAG_WRITE) ||
3037                              !force_resend_writes))
3038                                 break;
3039
3040                         /* fall through */
3041                 case CALC_TARGET_NEED_RESEND:
3042                         cancel_map_check(req);
3043                         unlink_request(osd, req);
3044                         insert_request(need_resend, req);
3045                         break;
3046                 case CALC_TARGET_POOL_DNE:
3047                         check_pool_dne(req);
3048                         break;
3049                 }
3050         }
3051 }
3052
3053 static int handle_one_map(struct ceph_osd_client *osdc,
3054                           void *p, void *end, bool incremental,
3055                           struct rb_root *need_resend,
3056                           struct list_head *need_resend_linger)
3057 {
3058         struct ceph_osdmap *newmap;
3059         struct rb_node *n;
3060         bool skipped_map = false;
3061         bool was_full;
3062
3063         was_full = ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL);
3064         set_pool_was_full(osdc);
3065
3066         if (incremental)
3067                 newmap = osdmap_apply_incremental(&p, end, osdc->osdmap);
3068         else
3069                 newmap = ceph_osdmap_decode(&p, end);
3070         if (IS_ERR(newmap))
3071                 return PTR_ERR(newmap);
3072
3073         if (newmap != osdc->osdmap) {
3074                 /*
3075                  * Preserve ->was_full before destroying the old map.
3076                  * For pools that weren't in the old map, ->was_full
3077                  * should be false.
3078                  */
3079                 for (n = rb_first(&newmap->pg_pools); n; n = rb_next(n)) {
3080                         struct ceph_pg_pool_info *pi =
3081                             rb_entry(n, struct ceph_pg_pool_info, node);
3082                         struct ceph_pg_pool_info *old_pi;
3083
3084                         old_pi = ceph_pg_pool_by_id(osdc->osdmap, pi->id);
3085                         if (old_pi)
3086                                 pi->was_full = old_pi->was_full;
3087                         else
3088                                 WARN_ON(pi->was_full);
3089                 }
3090
3091                 if (osdc->osdmap->epoch &&
3092                     osdc->osdmap->epoch + 1 < newmap->epoch) {
3093                         WARN_ON(incremental);
3094                         skipped_map = true;
3095                 }
3096
3097                 ceph_osdmap_destroy(osdc->osdmap);
3098                 osdc->osdmap = newmap;
3099         }
3100
3101         was_full &= !ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL);
3102         scan_requests(&osdc->homeless_osd, skipped_map, was_full, true,
3103                       need_resend, need_resend_linger);
3104
3105         for (n = rb_first(&osdc->osds); n; ) {
3106                 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
3107
3108                 n = rb_next(n); /* close_osd() */
3109
3110                 scan_requests(osd, skipped_map, was_full, true, need_resend,
3111                               need_resend_linger);
3112                 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
3113                     memcmp(&osd->o_con.peer_addr,
3114                            ceph_osd_addr(osdc->osdmap, osd->o_osd),
3115                            sizeof(struct ceph_entity_addr)))
3116                         close_osd(osd);
3117         }
3118
3119         return 0;
3120 }
3121
3122 static void kick_requests(struct ceph_osd_client *osdc,
3123                           struct rb_root *need_resend,
3124                           struct list_head *need_resend_linger)
3125 {
3126         struct ceph_osd_linger_request *lreq, *nlreq;
3127         struct rb_node *n;
3128
3129         for (n = rb_first(need_resend); n; ) {
3130                 struct ceph_osd_request *req =
3131                     rb_entry(n, struct ceph_osd_request, r_node);
3132                 struct ceph_osd *osd;
3133
3134                 n = rb_next(n);
3135                 erase_request(need_resend, req); /* before link_request() */
3136
3137                 WARN_ON(req->r_osd);
3138                 calc_target(osdc, &req->r_t, NULL, false);
3139                 osd = lookup_create_osd(osdc, req->r_t.osd, true);
3140                 link_request(osd, req);
3141                 if (!req->r_linger) {
3142                         if (!osd_homeless(osd) && !req->r_t.paused)
3143                                 send_request(req);
3144                 } else {
3145                         cancel_linger_request(req);
3146                 }
3147         }
3148
3149         list_for_each_entry_safe(lreq, nlreq, need_resend_linger, scan_item) {
3150                 if (!osd_homeless(lreq->osd))
3151                         send_linger(lreq);
3152
3153                 list_del_init(&lreq->scan_item);
3154         }
3155 }
3156
3157 /*
3158  * Process updated osd map.
3159  *
3160  * The message contains any number of incremental and full maps, normally
3161  * indicating some sort of topology change in the cluster.  Kick requests
3162  * off to different OSDs as needed.
3163  */
3164 void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
3165 {
3166         void *p = msg->front.iov_base;
3167         void *const end = p + msg->front.iov_len;
3168         u32 nr_maps, maplen;
3169         u32 epoch;
3170         struct ceph_fsid fsid;
3171         struct rb_root need_resend = RB_ROOT;
3172         LIST_HEAD(need_resend_linger);
3173         bool handled_incremental = false;
3174         bool was_pauserd, was_pausewr;
3175         bool pauserd, pausewr;
3176         int err;
3177
3178         dout("%s have %u\n", __func__, osdc->osdmap->epoch);
3179         down_write(&osdc->lock);
3180
3181         /* verify fsid */
3182         ceph_decode_need(&p, end, sizeof(fsid), bad);
3183         ceph_decode_copy(&p, &fsid, sizeof(fsid));
3184         if (ceph_check_fsid(osdc->client, &fsid) < 0)
3185                 goto bad;
3186
3187         was_pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
3188         was_pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
3189                       ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
3190                       have_pool_full(osdc);
3191
3192         /* incremental maps */
3193         ceph_decode_32_safe(&p, end, nr_maps, bad);
3194         dout(" %d inc maps\n", nr_maps);
3195         while (nr_maps > 0) {
3196                 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
3197                 epoch = ceph_decode_32(&p);
3198                 maplen = ceph_decode_32(&p);
3199                 ceph_decode_need(&p, end, maplen, bad);
3200                 if (osdc->osdmap->epoch &&
3201                     osdc->osdmap->epoch + 1 == epoch) {
3202                         dout("applying incremental map %u len %d\n",
3203                              epoch, maplen);
3204                         err = handle_one_map(osdc, p, p + maplen, true,
3205                                              &need_resend, &need_resend_linger);
3206                         if (err)
3207                                 goto bad;
3208                         handled_incremental = true;
3209                 } else {
3210                         dout("ignoring incremental map %u len %d\n",
3211                              epoch, maplen);
3212                 }
3213                 p += maplen;
3214                 nr_maps--;
3215         }
3216         if (handled_incremental)
3217                 goto done;
3218
3219         /* full maps */
3220         ceph_decode_32_safe(&p, end, nr_maps, bad);
3221         dout(" %d full maps\n", nr_maps);
3222         while (nr_maps) {
3223                 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
3224                 epoch = ceph_decode_32(&p);
3225                 maplen = ceph_decode_32(&p);
3226                 ceph_decode_need(&p, end, maplen, bad);
3227                 if (nr_maps > 1) {
3228                         dout("skipping non-latest full map %u len %d\n",
3229                              epoch, maplen);
3230                 } else if (osdc->osdmap->epoch >= epoch) {
3231                         dout("skipping full map %u len %d, "
3232                              "older than our %u\n", epoch, maplen,
3233                              osdc->osdmap->epoch);
3234                 } else {
3235                         dout("taking full map %u len %d\n", epoch, maplen);
3236                         err = handle_one_map(osdc, p, p + maplen, false,
3237                                              &need_resend, &need_resend_linger);
3238                         if (err)
3239                                 goto bad;
3240                 }
3241                 p += maplen;
3242                 nr_maps--;
3243         }
3244
3245 done:
3246         /*
3247          * subscribe to subsequent osdmap updates if full to ensure
3248          * we find out when we are no longer full and stop returning
3249          * ENOSPC.
3250          */
3251         pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
3252         pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
3253                   ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
3254                   have_pool_full(osdc);
3255         if (was_pauserd || was_pausewr || pauserd || pausewr)
3256                 maybe_request_map(osdc);
3257
3258         kick_requests(osdc, &need_resend, &need_resend_linger);
3259
3260         ceph_monc_got_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
3261                           osdc->osdmap->epoch);
3262         up_write(&osdc->lock);
3263         wake_up_all(&osdc->client->auth_wq);
3264         return;
3265
3266 bad:
3267         pr_err("osdc handle_map corrupt msg\n");
3268         ceph_msg_dump(msg);
3269         up_write(&osdc->lock);
3270 }
3271
3272 /*
3273  * Resubmit requests pending on the given osd.
3274  */
3275 static void kick_osd_requests(struct ceph_osd *osd)
3276 {
3277         struct rb_node *n;
3278
3279         for (n = rb_first(&osd->o_requests); n; ) {
3280                 struct ceph_osd_request *req =
3281                     rb_entry(n, struct ceph_osd_request, r_node);
3282
3283                 n = rb_next(n); /* cancel_linger_request() */
3284
3285                 if (!req->r_linger) {
3286                         if (!req->r_t.paused)
3287                                 send_request(req);
3288                 } else {
3289                         cancel_linger_request(req);
3290                 }
3291         }
3292         for (n = rb_first(&osd->o_linger_requests); n; n = rb_next(n)) {
3293                 struct ceph_osd_linger_request *lreq =
3294                     rb_entry(n, struct ceph_osd_linger_request, node);
3295
3296                 send_linger(lreq);
3297         }
3298 }
3299
3300 /*
3301  * If the osd connection drops, we need to resubmit all requests.
3302  */
3303 static void osd_fault(struct ceph_connection *con)
3304 {
3305         struct ceph_osd *osd = con->private;
3306         struct ceph_osd_client *osdc = osd->o_osdc;
3307
3308         dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
3309
3310         down_write(&osdc->lock);
3311         if (!osd_registered(osd)) {
3312                 dout("%s osd%d unknown\n", __func__, osd->o_osd);
3313                 goto out_unlock;
3314         }
3315
3316         if (!reopen_osd(osd))
3317                 kick_osd_requests(osd);
3318         maybe_request_map(osdc);
3319
3320 out_unlock:
3321         up_write(&osdc->lock);
3322 }
3323
3324 /*
3325  * Process osd watch notifications
3326  */
3327 static void handle_watch_notify(struct ceph_osd_client *osdc,
3328                                 struct ceph_msg *msg)
3329 {
3330         void *p = msg->front.iov_base;
3331         void *const end = p + msg->front.iov_len;
3332         struct ceph_osd_linger_request *lreq;
3333         struct linger_work *lwork;
3334         u8 proto_ver, opcode;
3335         u64 cookie, notify_id;
3336         u64 notifier_id = 0;
3337         s32 return_code = 0;
3338         void *payload = NULL;
3339         u32 payload_len = 0;
3340
3341         ceph_decode_8_safe(&p, end, proto_ver, bad);
3342         ceph_decode_8_safe(&p, end, opcode, bad);
3343         ceph_decode_64_safe(&p, end, cookie, bad);
3344         p += 8; /* skip ver */
3345         ceph_decode_64_safe(&p, end, notify_id, bad);
3346
3347         if (proto_ver >= 1) {
3348                 ceph_decode_32_safe(&p, end, payload_len, bad);
3349                 ceph_decode_need(&p, end, payload_len, bad);
3350                 payload = p;
3351                 p += payload_len;
3352         }
3353
3354         if (le16_to_cpu(msg->hdr.version) >= 2)
3355                 ceph_decode_32_safe(&p, end, return_code, bad);
3356
3357         if (le16_to_cpu(msg->hdr.version) >= 3)
3358                 ceph_decode_64_safe(&p, end, notifier_id, bad);
3359
3360         down_read(&osdc->lock);
3361         lreq = lookup_linger_osdc(&osdc->linger_requests, cookie);
3362         if (!lreq) {
3363                 dout("%s opcode %d cookie %llu dne\n", __func__, opcode,
3364                      cookie);
3365                 goto out_unlock_osdc;
3366         }
3367
3368         mutex_lock(&lreq->lock);
3369         dout("%s opcode %d cookie %llu lreq %p is_watch %d\n", __func__,
3370              opcode, cookie, lreq, lreq->is_watch);
3371         if (opcode == CEPH_WATCH_EVENT_DISCONNECT) {
3372                 if (!lreq->last_error) {
3373                         lreq->last_error = -ENOTCONN;
3374                         queue_watch_error(lreq);
3375                 }
3376         } else if (!lreq->is_watch) {
3377                 /* CEPH_WATCH_EVENT_NOTIFY_COMPLETE */
3378                 if (lreq->notify_id && lreq->notify_id != notify_id) {
3379                         dout("lreq %p notify_id %llu != %llu, ignoring\n", lreq,
3380                              lreq->notify_id, notify_id);
3381                 } else if (!completion_done(&lreq->notify_finish_wait)) {
3382                         struct ceph_msg_data *data =
3383                             list_first_entry_or_null(&msg->data,
3384                                                      struct ceph_msg_data,
3385                                                      links);
3386
3387                         if (data) {
3388                                 if (lreq->preply_pages) {
3389                                         WARN_ON(data->type !=
3390                                                         CEPH_MSG_DATA_PAGES);
3391                                         *lreq->preply_pages = data->pages;
3392                                         *lreq->preply_len = data->length;
3393                                 } else {
3394                                         ceph_release_page_vector(data->pages,
3395                                                calc_pages_for(0, data->length));
3396                                 }
3397                         }
3398                         lreq->notify_finish_error = return_code;
3399                         complete_all(&lreq->notify_finish_wait);
3400                 }
3401         } else {
3402                 /* CEPH_WATCH_EVENT_NOTIFY */
3403                 lwork = lwork_alloc(lreq, do_watch_notify);
3404                 if (!lwork) {
3405                         pr_err("failed to allocate notify-lwork\n");
3406                         goto out_unlock_lreq;
3407                 }
3408
3409                 lwork->notify.notify_id = notify_id;
3410                 lwork->notify.notifier_id = notifier_id;
3411                 lwork->notify.payload = payload;
3412                 lwork->notify.payload_len = payload_len;
3413                 lwork->notify.msg = ceph_msg_get(msg);
3414                 lwork_queue(lwork);
3415         }
3416
3417 out_unlock_lreq:
3418         mutex_unlock(&lreq->lock);
3419 out_unlock_osdc:
3420         up_read(&osdc->lock);
3421         return;
3422
3423 bad:
3424         pr_err("osdc handle_watch_notify corrupt msg\n");
3425 }
3426
3427 /*
3428  * Register request, send initial attempt.
3429  */
3430 int ceph_osdc_start_request(struct ceph_osd_client *osdc,
3431                             struct ceph_osd_request *req,
3432                             bool nofail)
3433 {
3434         down_read(&osdc->lock);
3435         submit_request(req, false);
3436         up_read(&osdc->lock);
3437
3438         return 0;
3439 }
3440 EXPORT_SYMBOL(ceph_osdc_start_request);
3441
3442 /*
3443  * Unregister a registered request.  The request is not completed:
3444  * ->r_result isn't set and __complete_request() isn't called.
3445  */
3446 void ceph_osdc_cancel_request(struct ceph_osd_request *req)
3447 {
3448         struct ceph_osd_client *osdc = req->r_osdc;
3449
3450         down_write(&osdc->lock);
3451         if (req->r_osd)
3452                 cancel_request(req);
3453         up_write(&osdc->lock);
3454 }
3455 EXPORT_SYMBOL(ceph_osdc_cancel_request);
3456
3457 /*
3458  * @timeout: in jiffies, 0 means "wait forever"
3459  */
3460 static int wait_request_timeout(struct ceph_osd_request *req,
3461                                 unsigned long timeout)
3462 {
3463         long left;
3464
3465         dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
3466         left = wait_for_completion_killable_timeout(&req->r_completion,
3467                                                 ceph_timeout_jiffies(timeout));
3468         if (left <= 0) {
3469                 left = left ?: -ETIMEDOUT;
3470                 ceph_osdc_cancel_request(req);
3471         } else {
3472                 left = req->r_result; /* completed */
3473         }
3474
3475         return left;
3476 }
3477
3478 /*
3479  * wait for a request to complete
3480  */
3481 int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
3482                            struct ceph_osd_request *req)
3483 {
3484         return wait_request_timeout(req, 0);
3485 }
3486 EXPORT_SYMBOL(ceph_osdc_wait_request);
3487
3488 /*
3489  * sync - wait for all in-flight requests to flush.  avoid starvation.
3490  */
3491 void ceph_osdc_sync(struct ceph_osd_client *osdc)
3492 {
3493         struct rb_node *n, *p;
3494         u64 last_tid = atomic64_read(&osdc->last_tid);
3495
3496 again:
3497         down_read(&osdc->lock);
3498         for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
3499                 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
3500
3501                 mutex_lock(&osd->lock);
3502                 for (p = rb_first(&osd->o_requests); p; p = rb_next(p)) {
3503                         struct ceph_osd_request *req =
3504                             rb_entry(p, struct ceph_osd_request, r_node);
3505
3506                         if (req->r_tid > last_tid)
3507                                 break;
3508
3509                         if (!(req->r_flags & CEPH_OSD_FLAG_WRITE))
3510                                 continue;
3511
3512                         ceph_osdc_get_request(req);
3513                         mutex_unlock(&osd->lock);
3514                         up_read(&osdc->lock);
3515                         dout("%s waiting on req %p tid %llu last_tid %llu\n",
3516                              __func__, req, req->r_tid, last_tid);
3517                         wait_for_completion(&req->r_completion);
3518                         ceph_osdc_put_request(req);
3519                         goto again;
3520                 }
3521
3522                 mutex_unlock(&osd->lock);
3523         }
3524
3525         up_read(&osdc->lock);
3526         dout("%s done last_tid %llu\n", __func__, last_tid);
3527 }
3528 EXPORT_SYMBOL(ceph_osdc_sync);
3529
3530 static struct ceph_osd_request *
3531 alloc_linger_request(struct ceph_osd_linger_request *lreq)
3532 {
3533         struct ceph_osd_request *req;
3534
3535         req = ceph_osdc_alloc_request(lreq->osdc, NULL, 1, false, GFP_NOIO);
3536         if (!req)
3537                 return NULL;
3538
3539         ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
3540         ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
3541
3542         if (ceph_osdc_alloc_messages(req, GFP_NOIO)) {
3543                 ceph_osdc_put_request(req);
3544                 return NULL;
3545         }
3546
3547         return req;
3548 }
3549
3550 /*
3551  * Returns a handle, caller owns a ref.
3552  */
3553 struct ceph_osd_linger_request *
3554 ceph_osdc_watch(struct ceph_osd_client *osdc,
3555                 struct ceph_object_id *oid,
3556                 struct ceph_object_locator *oloc,
3557                 rados_watchcb2_t wcb,
3558                 rados_watcherrcb_t errcb,
3559                 void *data)
3560 {
3561         struct ceph_osd_linger_request *lreq;
3562         int ret;
3563
3564         lreq = linger_alloc(osdc);
3565         if (!lreq)
3566                 return ERR_PTR(-ENOMEM);
3567
3568         lreq->is_watch = true;
3569         lreq->wcb = wcb;
3570         lreq->errcb = errcb;
3571         lreq->data = data;
3572         lreq->watch_valid_thru = jiffies;
3573
3574         ceph_oid_copy(&lreq->t.base_oid, oid);
3575         ceph_oloc_copy(&lreq->t.base_oloc, oloc);
3576         lreq->t.flags = CEPH_OSD_FLAG_WRITE;
3577         lreq->mtime = CURRENT_TIME;
3578
3579         lreq->reg_req = alloc_linger_request(lreq);
3580         if (!lreq->reg_req) {
3581                 ret = -ENOMEM;
3582                 goto err_put_lreq;
3583         }
3584
3585         lreq->ping_req = alloc_linger_request(lreq);
3586         if (!lreq->ping_req) {
3587                 ret = -ENOMEM;
3588                 goto err_put_lreq;
3589         }
3590
3591         down_write(&osdc->lock);
3592         linger_register(lreq); /* before osd_req_op_* */
3593         osd_req_op_watch_init(lreq->reg_req, 0, lreq->linger_id,
3594                               CEPH_OSD_WATCH_OP_WATCH);
3595         osd_req_op_watch_init(lreq->ping_req, 0, lreq->linger_id,
3596                               CEPH_OSD_WATCH_OP_PING);
3597         linger_submit(lreq);
3598         up_write(&osdc->lock);
3599
3600         ret = linger_reg_commit_wait(lreq);
3601         if (ret) {
3602                 linger_cancel(lreq);
3603                 goto err_put_lreq;
3604         }
3605
3606         return lreq;
3607
3608 err_put_lreq:
3609         linger_put(lreq);
3610         return ERR_PTR(ret);
3611 }
3612 EXPORT_SYMBOL(ceph_osdc_watch);
3613
3614 /*
3615  * Releases a ref.
3616  *
3617  * Times out after mount_timeout to preserve rbd unmap behaviour
3618  * introduced in 2894e1d76974 ("rbd: timeout watch teardown on unmap
3619  * with mount_timeout").
3620  */
3621 int ceph_osdc_unwatch(struct ceph_osd_client *osdc,
3622                       struct ceph_osd_linger_request *lreq)
3623 {
3624         struct ceph_options *opts = osdc->client->options;
3625         struct ceph_osd_request *req;
3626         int ret;
3627
3628         req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
3629         if (!req)
3630                 return -ENOMEM;
3631
3632         ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
3633         ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
3634         req->r_flags = CEPH_OSD_FLAG_WRITE;
3635         req->r_mtime = CURRENT_TIME;
3636         osd_req_op_watch_init(req, 0, lreq->linger_id,
3637                               CEPH_OSD_WATCH_OP_UNWATCH);
3638
3639         ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
3640         if (ret)
3641                 goto out_put_req;
3642
3643         ceph_osdc_start_request(osdc, req, false);
3644         linger_cancel(lreq);
3645         linger_put(lreq);
3646         ret = wait_request_timeout(req, opts->mount_timeout);
3647
3648 out_put_req:
3649         ceph_osdc_put_request(req);
3650         return ret;
3651 }
3652 EXPORT_SYMBOL(ceph_osdc_unwatch);
3653
3654 static int osd_req_op_notify_ack_init(struct ceph_osd_request *req, int which,
3655                                       u64 notify_id, u64 cookie, void *payload,
3656                                       size_t payload_len)
3657 {
3658         struct ceph_osd_req_op *op;
3659         struct ceph_pagelist *pl;
3660         int ret;
3661
3662         op = _osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY_ACK, 0);
3663
3664         pl = kmalloc(sizeof(*pl), GFP_NOIO);
3665         if (!pl)
3666                 return -ENOMEM;
3667
3668         ceph_pagelist_init(pl);
3669         ret = ceph_pagelist_encode_64(pl, notify_id);
3670         ret |= ceph_pagelist_encode_64(pl, cookie);
3671         if (payload) {
3672                 ret |= ceph_pagelist_encode_32(pl, payload_len);
3673                 ret |= ceph_pagelist_append(pl, payload, payload_len);
3674         } else {
3675                 ret |= ceph_pagelist_encode_32(pl, 0);
3676         }
3677         if (ret) {
3678                 ceph_pagelist_release(pl);
3679                 return -ENOMEM;
3680         }
3681
3682         ceph_osd_data_pagelist_init(&op->notify_ack.request_data, pl);
3683         op->indata_len = pl->length;
3684         return 0;
3685 }
3686
3687 int ceph_osdc_notify_ack(struct ceph_osd_client *osdc,
3688                          struct ceph_object_id *oid,
3689                          struct ceph_object_locator *oloc,
3690                          u64 notify_id,
3691                          u64 cookie,
3692                          void *payload,
3693                          size_t payload_len)
3694 {
3695         struct ceph_osd_request *req;
3696         int ret;
3697
3698         req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
3699         if (!req)
3700                 return -ENOMEM;
3701
3702         ceph_oid_copy(&req->r_base_oid, oid);
3703         ceph_oloc_copy(&req->r_base_oloc, oloc);
3704         req->r_flags = CEPH_OSD_FLAG_READ;
3705
3706         ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
3707         if (ret)
3708                 goto out_put_req;
3709
3710         ret = osd_req_op_notify_ack_init(req, 0, notify_id, cookie, payload,
3711                                          payload_len);
3712         if (ret)
3713                 goto out_put_req;
3714
3715         ceph_osdc_start_request(osdc, req, false);
3716         ret = ceph_osdc_wait_request(osdc, req);
3717
3718 out_put_req:
3719         ceph_osdc_put_request(req);
3720         return ret;
3721 }
3722 EXPORT_SYMBOL(ceph_osdc_notify_ack);
3723
3724 static int osd_req_op_notify_init(struct ceph_osd_request *req, int which,
3725                                   u64 cookie, u32 prot_ver, u32 timeout,
3726                                   void *payload, size_t payload_len)
3727 {
3728         struct ceph_osd_req_op *op;
3729         struct ceph_pagelist *pl;
3730         int ret;
3731
3732         op = _osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY, 0);
3733         op->notify.cookie = cookie;
3734
3735         pl = kmalloc(sizeof(*pl), GFP_NOIO);
3736         if (!pl)
3737                 return -ENOMEM;
3738
3739         ceph_pagelist_init(pl);
3740         ret = ceph_pagelist_encode_32(pl, 1); /* prot_ver */
3741         ret |= ceph_pagelist_encode_32(pl, timeout);
3742         ret |= ceph_pagelist_encode_32(pl, payload_len);
3743         ret |= ceph_pagelist_append(pl, payload, payload_len);
3744         if (ret) {
3745                 ceph_pagelist_release(pl);
3746                 return -ENOMEM;
3747         }
3748
3749         ceph_osd_data_pagelist_init(&op->notify.request_data, pl);
3750         op->indata_len = pl->length;
3751         return 0;
3752 }
3753
3754 /*
3755  * @timeout: in seconds
3756  *
3757  * @preply_{pages,len} are initialized both on success and error.
3758  * The caller is responsible for:
3759  *
3760  *     ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len))
3761  */
3762 int ceph_osdc_notify(struct ceph_osd_client *osdc,
3763                      struct ceph_object_id *oid,
3764                      struct ceph_object_locator *oloc,
3765                      void *payload,
3766                      size_t payload_len,
3767                      u32 timeout,
3768                      struct page ***preply_pages,
3769                      size_t *preply_len)
3770 {
3771         struct ceph_osd_linger_request *lreq;
3772         struct page **pages;
3773         int ret;
3774
3775         WARN_ON(!timeout);
3776         if (preply_pages) {
3777                 *preply_pages = NULL;
3778                 *preply_len = 0;
3779         }
3780
3781         lreq = linger_alloc(osdc);
3782         if (!lreq)
3783                 return -ENOMEM;
3784
3785         lreq->preply_pages = preply_pages;
3786         lreq->preply_len = preply_len;
3787
3788         ceph_oid_copy(&lreq->t.base_oid, oid);
3789         ceph_oloc_copy(&lreq->t.base_oloc, oloc);
3790         lreq->t.flags = CEPH_OSD_FLAG_READ;
3791
3792         lreq->reg_req = alloc_linger_request(lreq);
3793         if (!lreq->reg_req) {
3794                 ret = -ENOMEM;
3795                 goto out_put_lreq;
3796         }
3797
3798         /* for notify_id */
3799         pages = ceph_alloc_page_vector(1, GFP_NOIO);
3800         if (IS_ERR(pages)) {
3801                 ret = PTR_ERR(pages);
3802                 goto out_put_lreq;
3803         }
3804
3805         down_write(&osdc->lock);
3806         linger_register(lreq); /* before osd_req_op_* */
3807         ret = osd_req_op_notify_init(lreq->reg_req, 0, lreq->linger_id, 1,
3808                                      timeout, payload, payload_len);
3809         if (ret) {
3810                 linger_unregister(lreq);
3811                 up_write(&osdc->lock);
3812                 ceph_release_page_vector(pages, 1);
3813                 goto out_put_lreq;
3814         }
3815         ceph_osd_data_pages_init(osd_req_op_data(lreq->reg_req, 0, notify,
3816                                                  response_data),
3817                                  pages, PAGE_SIZE, 0, false, true);
3818         linger_submit(lreq);
3819         up_write(&osdc->lock);
3820
3821         ret = linger_reg_commit_wait(lreq);
3822         if (!ret)
3823                 ret = linger_notify_finish_wait(lreq);
3824         else
3825                 dout("lreq %p failed to initiate notify %d\n", lreq, ret);
3826
3827         linger_cancel(lreq);
3828 out_put_lreq:
3829         linger_put(lreq);
3830         return ret;
3831 }
3832 EXPORT_SYMBOL(ceph_osdc_notify);
3833
3834 /*
3835  * Return the number of milliseconds since the watch was last
3836  * confirmed, or an error.  If there is an error, the watch is no
3837  * longer valid, and should be destroyed with ceph_osdc_unwatch().
3838  */
3839 int ceph_osdc_watch_check(struct ceph_osd_client *osdc,
3840                           struct ceph_osd_linger_request *lreq)
3841 {
3842         unsigned long stamp, age;
3843         int ret;
3844
3845         down_read(&osdc->lock);
3846         mutex_lock(&lreq->lock);
3847         stamp = lreq->watch_valid_thru;
3848         if (!list_empty(&lreq->pending_lworks)) {
3849                 struct linger_work *lwork =
3850                     list_first_entry(&lreq->pending_lworks,
3851                                      struct linger_work,
3852                                      pending_item);
3853
3854                 if (time_before(lwork->queued_stamp, stamp))
3855                         stamp = lwork->queued_stamp;
3856         }
3857         age = jiffies - stamp;
3858         dout("%s lreq %p linger_id %llu age %lu last_error %d\n", __func__,
3859              lreq, lreq->linger_id, age, lreq->last_error);
3860         /* we are truncating to msecs, so return a safe upper bound */
3861         ret = lreq->last_error ?: 1 + jiffies_to_msecs(age);
3862
3863         mutex_unlock(&lreq->lock);
3864         up_read(&osdc->lock);
3865         return ret;
3866 }
3867
3868 static int decode_watcher(void **p, void *end, struct ceph_watch_item *item)
3869 {
3870         u8 struct_v;
3871         u32 struct_len;
3872         int ret;
3873
3874         ret = ceph_start_decoding(p, end, 2, "watch_item_t",
3875                                   &struct_v, &struct_len);
3876         if (ret)
3877                 return ret;
3878
3879         ceph_decode_copy(p, &item->name, sizeof(item->name));
3880         item->cookie = ceph_decode_64(p);
3881         *p += 4; /* skip timeout_seconds */
3882         if (struct_v >= 2) {
3883                 ceph_decode_copy(p, &item->addr, sizeof(item->addr));
3884                 ceph_decode_addr(&item->addr);
3885         }
3886
3887         dout("%s %s%llu cookie %llu addr %s\n", __func__,
3888              ENTITY_NAME(item->name), item->cookie,
3889              ceph_pr_addr(&item->addr.in_addr));
3890         return 0;
3891 }
3892
3893 static int decode_watchers(void **p, void *end,
3894                            struct ceph_watch_item **watchers,
3895                            u32 *num_watchers)
3896 {
3897         u8 struct_v;
3898         u32 struct_len;
3899         int i;
3900         int ret;
3901
3902         ret = ceph_start_decoding(p, end, 1, "obj_list_watch_response_t",
3903                                   &struct_v, &struct_len);
3904         if (ret)
3905                 return ret;
3906
3907         *num_watchers = ceph_decode_32(p);
3908         *watchers = kcalloc(*num_watchers, sizeof(**watchers), GFP_NOIO);
3909         if (!*watchers)
3910                 return -ENOMEM;
3911
3912         for (i = 0; i < *num_watchers; i++) {
3913                 ret = decode_watcher(p, end, *watchers + i);
3914                 if (ret) {
3915                         kfree(*watchers);
3916                         return ret;
3917                 }
3918         }
3919
3920         return 0;
3921 }
3922
3923 /*
3924  * On success, the caller is responsible for:
3925  *
3926  *     kfree(watchers);
3927  */
3928 int ceph_osdc_list_watchers(struct ceph_osd_client *osdc,
3929                             struct ceph_object_id *oid,
3930                             struct ceph_object_locator *oloc,
3931                             struct ceph_watch_item **watchers,
3932                             u32 *num_watchers)
3933 {
3934         struct ceph_osd_request *req;
3935         struct page **pages;
3936         int ret;
3937
3938         req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
3939         if (!req)
3940                 return -ENOMEM;
3941
3942         ceph_oid_copy(&req->r_base_oid, oid);
3943         ceph_oloc_copy(&req->r_base_oloc, oloc);
3944         req->r_flags = CEPH_OSD_FLAG_READ;
3945
3946         ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
3947         if (ret)
3948                 goto out_put_req;
3949
3950         pages = ceph_alloc_page_vector(1, GFP_NOIO);
3951         if (IS_ERR(pages)) {
3952                 ret = PTR_ERR(pages);
3953                 goto out_put_req;
3954         }
3955
3956         osd_req_op_init(req, 0, CEPH_OSD_OP_LIST_WATCHERS, 0);
3957         ceph_osd_data_pages_init(osd_req_op_data(req, 0, list_watchers,
3958                                                  response_data),
3959                                  pages, PAGE_SIZE, 0, false, true);
3960
3961         ceph_osdc_start_request(osdc, req, false);
3962         ret = ceph_osdc_wait_request(osdc, req);
3963         if (ret >= 0) {
3964                 void *p = page_address(pages[0]);
3965                 void *const end = p + req->r_ops[0].outdata_len;
3966
3967                 ret = decode_watchers(&p, end, watchers, num_watchers);
3968         }
3969
3970 out_put_req:
3971         ceph_osdc_put_request(req);
3972         return ret;
3973 }
3974 EXPORT_SYMBOL(ceph_osdc_list_watchers);
3975
3976 /*
3977  * Call all pending notify callbacks - for use after a watch is
3978  * unregistered, to make sure no more callbacks for it will be invoked
3979  */
3980 void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
3981 {
3982         dout("%s osdc %p\n", __func__, osdc);
3983         flush_workqueue(osdc->notify_wq);
3984 }
3985 EXPORT_SYMBOL(ceph_osdc_flush_notifies);
3986
3987 void ceph_osdc_maybe_request_map(struct ceph_osd_client *osdc)
3988 {
3989         down_read(&osdc->lock);
3990         maybe_request_map(osdc);
3991         up_read(&osdc->lock);
3992 }
3993 EXPORT_SYMBOL(ceph_osdc_maybe_request_map);
3994
3995 /*
3996  * Execute an OSD class method on an object.
3997  *
3998  * @flags: CEPH_OSD_FLAG_*
3999  * @resp_len: in/out param for reply length
4000  */
4001 int ceph_osdc_call(struct ceph_osd_client *osdc,
4002                    struct ceph_object_id *oid,
4003                    struct ceph_object_locator *oloc,
4004                    const char *class, const char *method,
4005                    unsigned int flags,
4006                    struct page *req_page, size_t req_len,
4007                    struct page *resp_page, size_t *resp_len)
4008 {
4009         struct ceph_osd_request *req;
4010         int ret;
4011
4012         if (req_len > PAGE_SIZE || (resp_page && *resp_len > PAGE_SIZE))
4013                 return -E2BIG;
4014
4015         req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
4016         if (!req)
4017                 return -ENOMEM;
4018
4019         ceph_oid_copy(&req->r_base_oid, oid);
4020         ceph_oloc_copy(&req->r_base_oloc, oloc);
4021         req->r_flags = flags;
4022
4023         ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
4024         if (ret)
4025                 goto out_put_req;
4026
4027         osd_req_op_cls_init(req, 0, CEPH_OSD_OP_CALL, class, method);
4028         if (req_page)
4029                 osd_req_op_cls_request_data_pages(req, 0, &req_page, req_len,
4030                                                   0, false, false);
4031         if (resp_page)
4032                 osd_req_op_cls_response_data_pages(req, 0, &resp_page,
4033                                                    *resp_len, 0, false, false);
4034
4035         ceph_osdc_start_request(osdc, req, false);
4036         ret = ceph_osdc_wait_request(osdc, req);
4037         if (ret >= 0) {
4038                 ret = req->r_ops[0].rval;
4039                 if (resp_page)
4040                         *resp_len = req->r_ops[0].outdata_len;
4041         }
4042
4043 out_put_req:
4044         ceph_osdc_put_request(req);
4045         return ret;
4046 }
4047 EXPORT_SYMBOL(ceph_osdc_call);
4048
4049 /*
4050  * init, shutdown
4051  */
4052 int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
4053 {
4054         int err;
4055
4056         dout("init\n");
4057         osdc->client = client;
4058         init_rwsem(&osdc->lock);
4059         osdc->osds = RB_ROOT;
4060         INIT_LIST_HEAD(&osdc->osd_lru);
4061         spin_lock_init(&osdc->osd_lru_lock);
4062         osd_init(&osdc->homeless_osd);
4063         osdc->homeless_osd.o_osdc = osdc;
4064         osdc->homeless_osd.o_osd = CEPH_HOMELESS_OSD;
4065         osdc->last_linger_id = CEPH_LINGER_ID_START;
4066         osdc->linger_requests = RB_ROOT;
4067         osdc->map_checks = RB_ROOT;
4068         osdc->linger_map_checks = RB_ROOT;
4069         INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
4070         INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
4071
4072         err = -ENOMEM;
4073         osdc->osdmap = ceph_osdmap_alloc();
4074         if (!osdc->osdmap)
4075                 goto out;
4076
4077         osdc->req_mempool = mempool_create_slab_pool(10,
4078                                                      ceph_osd_request_cache);
4079         if (!osdc->req_mempool)
4080                 goto out_map;
4081
4082         err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
4083                                 PAGE_SIZE, 10, true, "osd_op");
4084         if (err < 0)
4085                 goto out_mempool;
4086         err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
4087                                 PAGE_SIZE, 10, true, "osd_op_reply");
4088         if (err < 0)
4089                 goto out_msgpool;
4090
4091         err = -ENOMEM;
4092         osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
4093         if (!osdc->notify_wq)
4094                 goto out_msgpool_reply;
4095
4096         schedule_delayed_work(&osdc->timeout_work,
4097                               osdc->client->options->osd_keepalive_timeout);
4098         schedule_delayed_work(&osdc->osds_timeout_work,
4099             round_jiffies_relative(osdc->client->options->osd_idle_ttl));
4100
4101         return 0;
4102
4103 out_msgpool_reply:
4104         ceph_msgpool_destroy(&osdc->msgpool_op_reply);
4105 out_msgpool:
4106         ceph_msgpool_destroy(&osdc->msgpool_op);
4107 out_mempool:
4108         mempool_destroy(osdc->req_mempool);
4109 out_map:
4110         ceph_osdmap_destroy(osdc->osdmap);
4111 out:
4112         return err;
4113 }
4114
4115 void ceph_osdc_stop(struct ceph_osd_client *osdc)
4116 {
4117         flush_workqueue(osdc->notify_wq);
4118         destroy_workqueue(osdc->notify_wq);
4119         cancel_delayed_work_sync(&osdc->timeout_work);
4120         cancel_delayed_work_sync(&osdc->osds_timeout_work);
4121
4122         down_write(&osdc->lock);
4123         while (!RB_EMPTY_ROOT(&osdc->osds)) {
4124                 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
4125                                                 struct ceph_osd, o_node);
4126                 close_osd(osd);
4127         }
4128         up_write(&osdc->lock);
4129         WARN_ON(atomic_read(&osdc->homeless_osd.o_ref) != 1);
4130         osd_cleanup(&osdc->homeless_osd);
4131
4132         WARN_ON(!list_empty(&osdc->osd_lru));
4133         WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_requests));
4134         WARN_ON(!RB_EMPTY_ROOT(&osdc->map_checks));
4135         WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_map_checks));
4136         WARN_ON(atomic_read(&osdc->num_requests));
4137         WARN_ON(atomic_read(&osdc->num_homeless));
4138
4139         ceph_osdmap_destroy(osdc->osdmap);
4140         mempool_destroy(osdc->req_mempool);
4141         ceph_msgpool_destroy(&osdc->msgpool_op);
4142         ceph_msgpool_destroy(&osdc->msgpool_op_reply);
4143 }
4144
4145 /*
4146  * Read some contiguous pages.  If we cross a stripe boundary, shorten
4147  * *plen.  Return number of bytes read, or error.
4148  */
4149 int ceph_osdc_readpages(struct ceph_osd_client *osdc,
4150                         struct ceph_vino vino, struct ceph_file_layout *layout,
4151                         u64 off, u64 *plen,
4152                         u32 truncate_seq, u64 truncate_size,
4153                         struct page **pages, int num_pages, int page_align)
4154 {
4155         struct ceph_osd_request *req;
4156         int rc = 0;
4157
4158         dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
4159              vino.snap, off, *plen);
4160         req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 0, 1,
4161                                     CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
4162                                     NULL, truncate_seq, truncate_size,
4163                                     false);
4164         if (IS_ERR(req))
4165                 return PTR_ERR(req);
4166
4167         /* it may be a short read due to an object boundary */
4168         osd_req_op_extent_osd_data_pages(req, 0,
4169                                 pages, *plen, page_align, false, false);
4170
4171         dout("readpages  final extent is %llu~%llu (%llu bytes align %d)\n",
4172              off, *plen, *plen, page_align);
4173
4174         rc = ceph_osdc_start_request(osdc, req, false);
4175         if (!rc)
4176                 rc = ceph_osdc_wait_request(osdc, req);
4177
4178         ceph_osdc_put_request(req);
4179         dout("readpages result %d\n", rc);
4180         return rc;
4181 }
4182 EXPORT_SYMBOL(ceph_osdc_readpages);
4183
4184 /*
4185  * do a synchronous write on N pages
4186  */
4187 int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
4188                          struct ceph_file_layout *layout,
4189                          struct ceph_snap_context *snapc,
4190                          u64 off, u64 len,
4191                          u32 truncate_seq, u64 truncate_size,
4192                          struct timespec *mtime,
4193                          struct page **pages, int num_pages)
4194 {
4195         struct ceph_osd_request *req;
4196         int rc = 0;
4197         int page_align = off & ~PAGE_MASK;
4198
4199         req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 0, 1,
4200                                     CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE,
4201                                     snapc, truncate_seq, truncate_size,
4202                                     true);
4203         if (IS_ERR(req))
4204                 return PTR_ERR(req);
4205
4206         /* it may be a short write due to an object boundary */
4207         osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align,
4208                                 false, false);
4209         dout("writepages %llu~%llu (%llu bytes)\n", off, len, len);
4210
4211         req->r_mtime = *mtime;
4212         rc = ceph_osdc_start_request(osdc, req, true);
4213         if (!rc)
4214                 rc = ceph_osdc_wait_request(osdc, req);
4215
4216         ceph_osdc_put_request(req);
4217         if (rc == 0)
4218                 rc = len;
4219         dout("writepages result %d\n", rc);
4220         return rc;
4221 }
4222 EXPORT_SYMBOL(ceph_osdc_writepages);
4223
4224 int ceph_osdc_setup(void)
4225 {
4226         size_t size = sizeof(struct ceph_osd_request) +
4227             CEPH_OSD_SLAB_OPS * sizeof(struct ceph_osd_req_op);
4228
4229         BUG_ON(ceph_osd_request_cache);
4230         ceph_osd_request_cache = kmem_cache_create("ceph_osd_request", size,
4231                                                    0, 0, NULL);
4232
4233         return ceph_osd_request_cache ? 0 : -ENOMEM;
4234 }
4235 EXPORT_SYMBOL(ceph_osdc_setup);
4236
4237 void ceph_osdc_cleanup(void)
4238 {
4239         BUG_ON(!ceph_osd_request_cache);
4240         kmem_cache_destroy(ceph_osd_request_cache);
4241         ceph_osd_request_cache = NULL;
4242 }
4243 EXPORT_SYMBOL(ceph_osdc_cleanup);
4244
4245 /*
4246  * handle incoming message
4247  */
4248 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
4249 {
4250         struct ceph_osd *osd = con->private;
4251         struct ceph_osd_client *osdc = osd->o_osdc;
4252         int type = le16_to_cpu(msg->hdr.type);
4253
4254         switch (type) {
4255         case CEPH_MSG_OSD_MAP:
4256                 ceph_osdc_handle_map(osdc, msg);
4257                 break;
4258         case CEPH_MSG_OSD_OPREPLY:
4259                 handle_reply(osd, msg);
4260                 break;
4261         case CEPH_MSG_WATCH_NOTIFY:
4262                 handle_watch_notify(osdc, msg);
4263                 break;
4264
4265         default:
4266                 pr_err("received unknown message type %d %s\n", type,
4267                        ceph_msg_type_name(type));
4268         }
4269
4270         ceph_msg_put(msg);
4271 }
4272
4273 /*
4274  * Lookup and return message for incoming reply.  Don't try to do
4275  * anything about a larger than preallocated data portion of the
4276  * message at the moment - for now, just skip the message.
4277  */
4278 static struct ceph_msg *get_reply(struct ceph_connection *con,
4279                                   struct ceph_msg_header *hdr,
4280                                   int *skip)
4281 {
4282         struct ceph_osd *osd = con->private;
4283         struct ceph_osd_client *osdc = osd->o_osdc;
4284         struct ceph_msg *m = NULL;
4285         struct ceph_osd_request *req;
4286         int front_len = le32_to_cpu(hdr->front_len);
4287         int data_len = le32_to_cpu(hdr->data_len);
4288         u64 tid = le64_to_cpu(hdr->tid);
4289
4290         down_read(&osdc->lock);
4291         if (!osd_registered(osd)) {
4292                 dout("%s osd%d unknown, skipping\n", __func__, osd->o_osd);
4293                 *skip = 1;
4294                 goto out_unlock_osdc;
4295         }
4296         WARN_ON(osd->o_osd != le64_to_cpu(hdr->src.num));
4297
4298         mutex_lock(&osd->lock);
4299         req = lookup_request(&osd->o_requests, tid);
4300         if (!req) {
4301                 dout("%s osd%d tid %llu unknown, skipping\n", __func__,
4302                      osd->o_osd, tid);
4303                 *skip = 1;
4304                 goto out_unlock_session;
4305         }
4306
4307         ceph_msg_revoke_incoming(req->r_reply);
4308
4309         if (front_len > req->r_reply->front_alloc_len) {
4310                 pr_warn("%s osd%d tid %llu front %d > preallocated %d\n",
4311                         __func__, osd->o_osd, req->r_tid, front_len,
4312                         req->r_reply->front_alloc_len);
4313                 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS,
4314                                  false);
4315                 if (!m)
4316                         goto out_unlock_session;
4317                 ceph_msg_put(req->r_reply);
4318                 req->r_reply = m;
4319         }
4320
4321         if (data_len > req->r_reply->data_length) {
4322                 pr_warn("%s osd%d tid %llu data %d > preallocated %zu, skipping\n",
4323                         __func__, osd->o_osd, req->r_tid, data_len,
4324                         req->r_reply->data_length);
4325                 m = NULL;
4326                 *skip = 1;
4327                 goto out_unlock_session;
4328         }
4329
4330         m = ceph_msg_get(req->r_reply);
4331         dout("get_reply tid %lld %p\n", tid, m);
4332
4333 out_unlock_session:
4334         mutex_unlock(&osd->lock);
4335 out_unlock_osdc:
4336         up_read(&osdc->lock);
4337         return m;
4338 }
4339
4340 /*
4341  * TODO: switch to a msg-owned pagelist
4342  */
4343 static struct ceph_msg *alloc_msg_with_page_vector(struct ceph_msg_header *hdr)
4344 {
4345         struct ceph_msg *m;
4346         int type = le16_to_cpu(hdr->type);
4347         u32 front_len = le32_to_cpu(hdr->front_len);
4348         u32 data_len = le32_to_cpu(hdr->data_len);
4349
4350         m = ceph_msg_new(type, front_len, GFP_NOIO, false);
4351         if (!m)
4352                 return NULL;
4353
4354         if (data_len) {
4355                 struct page **pages;
4356                 struct ceph_osd_data osd_data;
4357
4358                 pages = ceph_alloc_page_vector(calc_pages_for(0, data_len),
4359                                                GFP_NOIO);
4360                 if (IS_ERR(pages)) {
4361                         ceph_msg_put(m);
4362                         return NULL;
4363                 }
4364
4365                 ceph_osd_data_pages_init(&osd_data, pages, data_len, 0, false,
4366                                          false);
4367                 ceph_osdc_msg_data_add(m, &osd_data);
4368         }
4369
4370         return m;
4371 }
4372
4373 static struct ceph_msg *alloc_msg(struct ceph_connection *con,
4374                                   struct ceph_msg_header *hdr,
4375                                   int *skip)
4376 {
4377         struct ceph_osd *osd = con->private;
4378         int type = le16_to_cpu(hdr->type);
4379
4380         *skip = 0;
4381         switch (type) {
4382         case CEPH_MSG_OSD_MAP:
4383         case CEPH_MSG_WATCH_NOTIFY:
4384                 return alloc_msg_with_page_vector(hdr);
4385         case CEPH_MSG_OSD_OPREPLY:
4386                 return get_reply(con, hdr, skip);
4387         default:
4388                 pr_warn("%s osd%d unknown msg type %d, skipping\n", __func__,
4389                         osd->o_osd, type);
4390                 *skip = 1;
4391                 return NULL;
4392         }
4393 }
4394
4395 /*
4396  * Wrappers to refcount containing ceph_osd struct
4397  */
4398 static struct ceph_connection *get_osd_con(struct ceph_connection *con)
4399 {
4400         struct ceph_osd *osd = con->private;
4401         if (get_osd(osd))
4402                 return con;
4403         return NULL;
4404 }
4405
4406 static void put_osd_con(struct ceph_connection *con)
4407 {
4408         struct ceph_osd *osd = con->private;
4409         put_osd(osd);
4410 }
4411
4412 /*
4413  * authentication
4414  */
4415 /*
4416  * Note: returned pointer is the address of a structure that's
4417  * managed separately.  Caller must *not* attempt to free it.
4418  */
4419 static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
4420                                         int *proto, int force_new)
4421 {
4422         struct ceph_osd *o = con->private;
4423         struct ceph_osd_client *osdc = o->o_osdc;
4424         struct ceph_auth_client *ac = osdc->client->monc.auth;
4425         struct ceph_auth_handshake *auth = &o->o_auth;
4426
4427         if (force_new && auth->authorizer) {
4428                 ceph_auth_destroy_authorizer(auth->authorizer);
4429                 auth->authorizer = NULL;
4430         }
4431         if (!auth->authorizer) {
4432                 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
4433                                                       auth);
4434                 if (ret)
4435                         return ERR_PTR(ret);
4436         } else {
4437                 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
4438                                                      auth);
4439                 if (ret)
4440                         return ERR_PTR(ret);
4441         }
4442         *proto = ac->protocol;
4443
4444         return auth;
4445 }
4446
4447
4448 static int verify_authorizer_reply(struct ceph_connection *con)
4449 {
4450         struct ceph_osd *o = con->private;
4451         struct ceph_osd_client *osdc = o->o_osdc;
4452         struct ceph_auth_client *ac = osdc->client->monc.auth;
4453
4454         return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer);
4455 }
4456
4457 static int invalidate_authorizer(struct ceph_connection *con)
4458 {
4459         struct ceph_osd *o = con->private;
4460         struct ceph_osd_client *osdc = o->o_osdc;
4461         struct ceph_auth_client *ac = osdc->client->monc.auth;
4462
4463         ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
4464         return ceph_monc_validate_auth(&osdc->client->monc);
4465 }
4466
4467 static int osd_sign_message(struct ceph_msg *msg)
4468 {
4469         struct ceph_osd *o = msg->con->private;
4470         struct ceph_auth_handshake *auth = &o->o_auth;
4471
4472         return ceph_auth_sign_message(auth, msg);
4473 }
4474
4475 static int osd_check_message_signature(struct ceph_msg *msg)
4476 {
4477         struct ceph_osd *o = msg->con->private;
4478         struct ceph_auth_handshake *auth = &o->o_auth;
4479
4480         return ceph_auth_check_message_signature(auth, msg);
4481 }
4482
4483 static const struct ceph_connection_operations osd_con_ops = {
4484         .get = get_osd_con,
4485         .put = put_osd_con,
4486         .dispatch = dispatch,
4487         .get_authorizer = get_authorizer,
4488         .verify_authorizer_reply = verify_authorizer_reply,
4489         .invalidate_authorizer = invalidate_authorizer,
4490         .alloc_msg = alloc_msg,
4491         .sign_message = osd_sign_message,
4492         .check_message_signature = osd_check_message_signature,
4493         .fault = osd_fault,
4494 };