net/ceph/osd_client.c

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include <linux/ceph/ceph_debug.h>
   4
   5 #include <linux/module.h>
   6 #include <linux/err.h>
   7 #include <linux/highmem.h>
   8 #include <linux/mm.h>
   9 #include <linux/pagemap.h>
  10 #include <linux/slab.h>
  11 #include <linux/uaccess.h>
  12 #ifdef CONFIG_BLOCK
  13 #include <linux/bio.h>
  14 #endif
  15
  16 #include <linux/ceph/ceph_features.h>
  17 #include <linux/ceph/libceph.h>
  18 #include <linux/ceph/osd_client.h>
  19 #include <linux/ceph/messenger.h>
  20 #include <linux/ceph/decode.h>
  21 #include <linux/ceph/auth.h>
  22 #include <linux/ceph/pagelist.h>
  23 #include <linux/ceph/striper.h>
  24
  25 #define OSD_OPREPLY_FRONT_LEN   512
  26
  27 static struct kmem_cache        *ceph_osd_request_cache;
  28
  29 static const struct ceph_connection_operations osd_con_ops;
  30
  31 /*
  32  * Implement client access to distributed object storage cluster.
  33  *
  34  * All data objects are stored within a cluster/cloud of OSDs, or
  35  * "object storage devices."  (Note that Ceph OSDs have _nothing_ to
  36  * do with the T10 OSD extensions to SCSI.)  Ceph OSDs are simply
  37  * remote daemons serving up and coordinating consistent and safe
  38  * access to storage.
  39  *
  40  * Cluster membership and the mapping of data objects onto storage devices
  41  * are described by the osd map.
  42  *
  43  * We keep track of pending OSD requests (read, write), resubmit
  44  * requests to different OSDs when the cluster topology/data layout
  45  * change, or retry the affected requests when the communications
  46  * channel with an OSD is reset.
  47  */
  48
  49 static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req);
  50 static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req);
  51 static void link_linger(struct ceph_osd *osd,
  52                         struct ceph_osd_linger_request *lreq);
  53 static void unlink_linger(struct ceph_osd *osd,
  54                           struct ceph_osd_linger_request *lreq);
  55 static void clear_backoffs(struct ceph_osd *osd);
  56
  57 #if 1
  58 static inline bool rwsem_is_wrlocked(struct rw_semaphore *sem)
  59 {
  60         bool wrlocked = true;
  61
  62         if (unlikely(down_read_trylock(sem))) {
  63                 wrlocked = false;
  64                 up_read(sem);
  65         }
  66
  67         return wrlocked;
  68 }
  69 static inline void verify_osdc_locked(struct ceph_osd_client *osdc)
  70 {
  71         WARN_ON(!rwsem_is_locked(&osdc->lock));
  72 }
  73 static inline void verify_osdc_wrlocked(struct ceph_osd_client *osdc)
  74 {
  75         WARN_ON(!rwsem_is_wrlocked(&osdc->lock));
  76 }
  77 static inline void verify_osd_locked(struct ceph_osd *osd)
  78 {
  79         struct ceph_osd_client *osdc = osd->o_osdc;
  80
  81         WARN_ON(!(mutex_is_locked(&osd->lock) &&
  82                   rwsem_is_locked(&osdc->lock)) &&
  83                 !rwsem_is_wrlocked(&osdc->lock));
  84 }
  85 static inline void verify_lreq_locked(struct ceph_osd_linger_request *lreq)
  86 {
  87         WARN_ON(!mutex_is_locked(&lreq->lock));
  88 }
  89 #else
  90 static inline void verify_osdc_locked(struct ceph_osd_client *osdc) { }
  91 static inline void verify_osdc_wrlocked(struct ceph_osd_client *osdc) { }
  92 static inline void verify_osd_locked(struct ceph_osd *osd) { }
  93 static inline void verify_lreq_locked(struct ceph_osd_linger_request *lreq) { }
  94 #endif
  95
  96 /*
  97  * calculate the mapping of a file extent onto an object, and fill out the
  98  * request accordingly.  shorten extent as necessary if it crosses an
  99  * object boundary.
 100  *
 101  * fill osd op in request message.
 102  */
 103 static int calc_layout(struct ceph_file_layout *layout, u64 off, u64 *plen,
 104                         u64 *objnum, u64 *objoff, u64 *objlen)
 105 {
 106         u64 orig_len = *plen;
 107         u32 xlen;
 108
 109         /* object extent? */
 110         ceph_calc_file_object_mapping(layout, off, orig_len, objnum,
 111                                           objoff, &xlen);
 112         *objlen = xlen;
 113         if (*objlen < orig_len) {
 114                 *plen = *objlen;
 115                 dout(" skipping last %llu, final file extent %llu~%llu\n",
 116                      orig_len - *plen, off, *plen);
 117         }
 118
 119         dout("calc_layout objnum=%llx %llu~%llu\n", *objnum, *objoff, *objlen);
 120         return 0;
 121 }
 122
 123 static void ceph_osd_data_init(struct ceph_osd_data *osd_data)
 124 {
 125         memset(osd_data, 0, sizeof (*osd_data));
 126         osd_data->type = CEPH_OSD_DATA_TYPE_NONE;
 127 }
 128
 129 /*
 130  * Consumes @pages if @own_pages is true.
 131  */
 132 static void ceph_osd_data_pages_init(struct ceph_osd_data *osd_data,
 133                         struct page **pages, u64 length, u32 alignment,
 134                         bool pages_from_pool, bool own_pages)
 135 {
 136         osd_data->type = CEPH_OSD_DATA_TYPE_PAGES;
 137         osd_data->pages = pages;
 138         osd_data->length = length;
 139         osd_data->alignment = alignment;
 140         osd_data->pages_from_pool = pages_from_pool;
 141         osd_data->own_pages = own_pages;
 142 }
 143
 144 /*
 145  * Consumes a ref on @pagelist.
 146  */
 147 static void ceph_osd_data_pagelist_init(struct ceph_osd_data *osd_data,
 148                         struct ceph_pagelist *pagelist)
 149 {
 150         osd_data->type = CEPH_OSD_DATA_TYPE_PAGELIST;
 151         osd_data->pagelist = pagelist;
 152 }
 153
 154 #ifdef CONFIG_BLOCK
 155 static void ceph_osd_data_bio_init(struct ceph_osd_data *osd_data,
 156                                    struct ceph_bio_iter *bio_pos,
 157                                    u32 bio_length)
 158 {
 159         osd_data->type = CEPH_OSD_DATA_TYPE_BIO;
 160         osd_data->bio_pos = *bio_pos;
 161         osd_data->bio_length = bio_length;
 162 }
 163 #endif /* CONFIG_BLOCK */
 164
 165 static void ceph_osd_data_bvecs_init(struct ceph_osd_data *osd_data,
 166                                      struct ceph_bvec_iter *bvec_pos,
 167                                      u32 num_bvecs)
 168 {
 169         osd_data->type = CEPH_OSD_DATA_TYPE_BVECS;
 170         osd_data->bvec_pos = *bvec_pos;
 171         osd_data->num_bvecs = num_bvecs;
 172 }
 173
 174 static void ceph_osd_iter_init(struct ceph_osd_data *osd_data,
 175                                struct iov_iter *iter)
 176 {
 177         osd_data->type = CEPH_OSD_DATA_TYPE_ITER;
 178         osd_data->iter = *iter;
 179 }
 180
 181 static struct ceph_osd_data *
 182 osd_req_op_raw_data_in(struct ceph_osd_request *osd_req, unsigned int which)
 183 {
 184         BUG_ON(which >= osd_req->r_num_ops);
 185
 186         return &osd_req->r_ops[which].raw_data_in;
 187 }
 188
 189 struct ceph_osd_data *
 190 osd_req_op_extent_osd_data(struct ceph_osd_request *osd_req,
 191                         unsigned int which)
 192 {
 193         return osd_req_op_data(osd_req, which, extent, osd_data);
 194 }
 195 EXPORT_SYMBOL(osd_req_op_extent_osd_data);
 196
 197 void osd_req_op_raw_data_in_pages(struct ceph_osd_request *osd_req,
 198                         unsigned int which, struct page **pages,
 199                         u64 length, u32 alignment,
 200                         bool pages_from_pool, bool own_pages)
 201 {
 202         struct ceph_osd_data *osd_data;
 203
 204         osd_data = osd_req_op_raw_data_in(osd_req, which);
 205         ceph_osd_data_pages_init(osd_data, pages, length, alignment,
 206                                 pages_from_pool, own_pages);
 207 }
 208 EXPORT_SYMBOL(osd_req_op_raw_data_in_pages);
 209
 210 void osd_req_op_extent_osd_data_pages(struct ceph_osd_request *osd_req,
 211                         unsigned int which, struct page **pages,
 212                         u64 length, u32 alignment,
 213                         bool pages_from_pool, bool own_pages)
 214 {
 215         struct ceph_osd_data *osd_data;
 216
 217         osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
 218         ceph_osd_data_pages_init(osd_data, pages, length, alignment,
 219                                 pages_from_pool, own_pages);
 220 }
 221 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pages);
 222
 223 void osd_req_op_extent_osd_data_pagelist(struct ceph_osd_request *osd_req,
 224                         unsigned int which, struct ceph_pagelist *pagelist)
 225 {
 226         struct ceph_osd_data *osd_data;
 227
 228         osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
 229         ceph_osd_data_pagelist_init(osd_data, pagelist);
 230 }
 231 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pagelist);
 232
 233 #ifdef CONFIG_BLOCK
 234 void osd_req_op_extent_osd_data_bio(struct ceph_osd_request *osd_req,
 235                                     unsigned int which,
 236                                     struct ceph_bio_iter *bio_pos,
 237                                     u32 bio_length)
 238 {
 239         struct ceph_osd_data *osd_data;
 240
 241         osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
 242         ceph_osd_data_bio_init(osd_data, bio_pos, bio_length);
 243 }
 244 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bio);
 245 #endif /* CONFIG_BLOCK */
 246
 247 void osd_req_op_extent_osd_data_bvecs(struct ceph_osd_request *osd_req,
 248                                       unsigned int which,
 249                                       struct bio_vec *bvecs, u32 num_bvecs,
 250                                       u32 bytes)
 251 {
 252         struct ceph_osd_data *osd_data;
 253         struct ceph_bvec_iter it = {
 254                 .bvecs = bvecs,
 255                 .iter = { .bi_size = bytes },
 256         };
 257
 258         osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
 259         ceph_osd_data_bvecs_init(osd_data, &it, num_bvecs);
 260 }
 261 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bvecs);
 262
 263 void osd_req_op_extent_osd_data_bvec_pos(struct ceph_osd_request *osd_req,
 264                                          unsigned int which,
 265                                          struct ceph_bvec_iter *bvec_pos)
 266 {
 267         struct ceph_osd_data *osd_data;
 268
 269         osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
 270         ceph_osd_data_bvecs_init(osd_data, bvec_pos, 0);
 271 }
 272 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bvec_pos);
 273
 274 /**
 275  * osd_req_op_extent_osd_iter - Set up an operation with an iterator buffer
 276  * @osd_req: The request to set up
 277  * @which: Index of the operation in which to set the iter
 278  * @iter: The buffer iterator
 279  */
 280 void osd_req_op_extent_osd_iter(struct ceph_osd_request *osd_req,
 281                                 unsigned int which, struct iov_iter *iter)
 282 {
 283         struct ceph_osd_data *osd_data;
 284
 285         osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
 286         ceph_osd_iter_init(osd_data, iter);
 287 }
 288 EXPORT_SYMBOL(osd_req_op_extent_osd_iter);
 289
 290 static void osd_req_op_cls_request_info_pagelist(
 291                         struct ceph_osd_request *osd_req,
 292                         unsigned int which, struct ceph_pagelist *pagelist)
 293 {
 294         struct ceph_osd_data *osd_data;
 295
 296         osd_data = osd_req_op_data(osd_req, which, cls, request_info);
 297         ceph_osd_data_pagelist_init(osd_data, pagelist);
 298 }
 299
 300 void osd_req_op_cls_request_data_pagelist(
 301                         struct ceph_osd_request *osd_req,
 302                         unsigned int which, struct ceph_pagelist *pagelist)
 303 {
 304         struct ceph_osd_data *osd_data;
 305
 306         osd_data = osd_req_op_data(osd_req, which, cls, request_data);
 307         ceph_osd_data_pagelist_init(osd_data, pagelist);
 308         osd_req->r_ops[which].cls.indata_len += pagelist->length;
 309         osd_req->r_ops[which].indata_len += pagelist->length;
 310 }
 311 EXPORT_SYMBOL(osd_req_op_cls_request_data_pagelist);
 312
 313 void osd_req_op_cls_request_data_pages(struct ceph_osd_request *osd_req,
 314                         unsigned int which, struct page **pages, u64 length,
 315                         u32 alignment, bool pages_from_pool, bool own_pages)
 316 {
 317         struct ceph_osd_data *osd_data;
 318
 319         osd_data = osd_req_op_data(osd_req, which, cls, request_data);
 320         ceph_osd_data_pages_init(osd_data, pages, length, alignment,
 321                                 pages_from_pool, own_pages);
 322         osd_req->r_ops[which].cls.indata_len += length;
 323         osd_req->r_ops[which].indata_len += length;
 324 }
 325 EXPORT_SYMBOL(osd_req_op_cls_request_data_pages);
 326
 327 void osd_req_op_cls_request_data_bvecs(struct ceph_osd_request *osd_req,
 328                                        unsigned int which,
 329                                        struct bio_vec *bvecs, u32 num_bvecs,
 330                                        u32 bytes)
 331 {
 332         struct ceph_osd_data *osd_data;
 333         struct ceph_bvec_iter it = {
 334                 .bvecs = bvecs,
 335                 .iter = { .bi_size = bytes },
 336         };
 337
 338         osd_data = osd_req_op_data(osd_req, which, cls, request_data);
 339         ceph_osd_data_bvecs_init(osd_data, &it, num_bvecs);
 340         osd_req->r_ops[which].cls.indata_len += bytes;
 341         osd_req->r_ops[which].indata_len += bytes;
 342 }
 343 EXPORT_SYMBOL(osd_req_op_cls_request_data_bvecs);
 344
 345 void osd_req_op_cls_response_data_pages(struct ceph_osd_request *osd_req,
 346                         unsigned int which, struct page **pages, u64 length,
 347                         u32 alignment, bool pages_from_pool, bool own_pages)
 348 {
 349         struct ceph_osd_data *osd_data;
 350
 351         osd_data = osd_req_op_data(osd_req, which, cls, response_data);
 352         ceph_osd_data_pages_init(osd_data, pages, length, alignment,
 353                                 pages_from_pool, own_pages);
 354 }
 355 EXPORT_SYMBOL(osd_req_op_cls_response_data_pages);
 356
 357 static u64 ceph_osd_data_length(struct ceph_osd_data *osd_data)
 358 {
 359         switch (osd_data->type) {
 360         case CEPH_OSD_DATA_TYPE_NONE:
 361                 return 0;
 362         case CEPH_OSD_DATA_TYPE_PAGES:
 363                 return osd_data->length;
 364         case CEPH_OSD_DATA_TYPE_PAGELIST:
 365                 return (u64)osd_data->pagelist->length;
 366 #ifdef CONFIG_BLOCK
 367         case CEPH_OSD_DATA_TYPE_BIO:
 368                 return (u64)osd_data->bio_length;
 369 #endif /* CONFIG_BLOCK */
 370         case CEPH_OSD_DATA_TYPE_BVECS:
 371                 return osd_data->bvec_pos.iter.bi_size;
 372         case CEPH_OSD_DATA_TYPE_ITER:
 373                 return iov_iter_count(&osd_data->iter);
 374         default:
 375                 WARN(true, "unrecognized data type %d\n", (int)osd_data->type);
 376                 return 0;
 377         }
 378 }
 379
 380 static void ceph_osd_data_release(struct ceph_osd_data *osd_data)
 381 {
 382         if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES && osd_data->own_pages) {
 383                 int num_pages;
 384
 385                 num_pages = calc_pages_for((u64)osd_data->alignment,
 386                                                 (u64)osd_data->length);
 387                 ceph_release_page_vector(osd_data->pages, num_pages);
 388         } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
 389                 ceph_pagelist_release(osd_data->pagelist);
 390         }
 391         ceph_osd_data_init(osd_data);
 392 }
 393
 394 static void osd_req_op_data_release(struct ceph_osd_request *osd_req,
 395                         unsigned int which)
 396 {
 397         struct ceph_osd_req_op *op;
 398
 399         BUG_ON(which >= osd_req->r_num_ops);
 400         op = &osd_req->r_ops[which];
 401
 402         switch (op->op) {
 403         case CEPH_OSD_OP_READ:
 404         case CEPH_OSD_OP_SPARSE_READ:
 405         case CEPH_OSD_OP_WRITE:
 406         case CEPH_OSD_OP_WRITEFULL:
 407                 kfree(op->extent.sparse_ext);
 408                 ceph_osd_data_release(&op->extent.osd_data);
 409                 break;
 410         case CEPH_OSD_OP_CALL:
 411                 ceph_osd_data_release(&op->cls.request_info);
 412                 ceph_osd_data_release(&op->cls.request_data);
 413                 ceph_osd_data_release(&op->cls.response_data);
 414                 break;
 415         case CEPH_OSD_OP_SETXATTR:
 416         case CEPH_OSD_OP_CMPXATTR:
 417                 ceph_osd_data_release(&op->xattr.osd_data);
 418                 break;
 419         case CEPH_OSD_OP_STAT:
 420                 ceph_osd_data_release(&op->raw_data_in);
 421                 break;
 422         case CEPH_OSD_OP_NOTIFY_ACK:
 423                 ceph_osd_data_release(&op->notify_ack.request_data);
 424                 break;
 425         case CEPH_OSD_OP_NOTIFY:
 426                 ceph_osd_data_release(&op->notify.request_data);
 427                 ceph_osd_data_release(&op->notify.response_data);
 428                 break;
 429         case CEPH_OSD_OP_LIST_WATCHERS:
 430                 ceph_osd_data_release(&op->list_watchers.response_data);
 431                 break;
 432         case CEPH_OSD_OP_COPY_FROM2:
 433                 ceph_osd_data_release(&op->copy_from.osd_data);
 434                 break;
 435         default:
 436                 break;
 437         }
 438 }
 439
 440 /*
 441  * Assumes @t is zero-initialized.
 442  */
 443 static void target_init(struct ceph_osd_request_target *t)
 444 {
 445         ceph_oid_init(&t->base_oid);
 446         ceph_oloc_init(&t->base_oloc);
 447         ceph_oid_init(&t->target_oid);
 448         ceph_oloc_init(&t->target_oloc);
 449
 450         ceph_osds_init(&t->acting);
 451         ceph_osds_init(&t->up);
 452         t->size = -1;
 453         t->min_size = -1;
 454
 455         t->osd = CEPH_HOMELESS_OSD;
 456 }
 457
 458 static void target_copy(struct ceph_osd_request_target *dest,
 459                         const struct ceph_osd_request_target *src)
 460 {
 461         ceph_oid_copy(&dest->base_oid, &src->base_oid);
 462         ceph_oloc_copy(&dest->base_oloc, &src->base_oloc);
 463         ceph_oid_copy(&dest->target_oid, &src->target_oid);
 464         ceph_oloc_copy(&dest->target_oloc, &src->target_oloc);
 465
 466         dest->pgid = src->pgid; /* struct */
 467         dest->spgid = src->spgid; /* struct */
 468         dest->pg_num = src->pg_num;
 469         dest->pg_num_mask = src->pg_num_mask;
 470         ceph_osds_copy(&dest->acting, &src->acting);
 471         ceph_osds_copy(&dest->up, &src->up);
 472         dest->size = src->size;
 473         dest->min_size = src->min_size;
 474         dest->sort_bitwise = src->sort_bitwise;
 475         dest->recovery_deletes = src->recovery_deletes;
 476
 477         dest->flags = src->flags;
 478         dest->used_replica = src->used_replica;
 479         dest->paused = src->paused;
 480
 481         dest->epoch = src->epoch;
 482         dest->last_force_resend = src->last_force_resend;
 483
 484         dest->osd = src->osd;
 485 }
 486
 487 static void target_destroy(struct ceph_osd_request_target *t)
 488 {
 489         ceph_oid_destroy(&t->base_oid);
 490         ceph_oloc_destroy(&t->base_oloc);
 491         ceph_oid_destroy(&t->target_oid);
 492         ceph_oloc_destroy(&t->target_oloc);
 493 }
 494
 495 /*
 496  * requests
 497  */
 498 static void request_release_checks(struct ceph_osd_request *req)
 499 {
 500         WARN_ON(!RB_EMPTY_NODE(&req->r_node));
 501         WARN_ON(!RB_EMPTY_NODE(&req->r_mc_node));
 502         WARN_ON(!list_empty(&req->r_private_item));
 503         WARN_ON(req->r_osd);
 504 }
 505
 506 static void ceph_osdc_release_request(struct kref *kref)
 507 {
 508         struct ceph_osd_request *req = container_of(kref,
 509                                             struct ceph_osd_request, r_kref);
 510         unsigned int which;
 511
 512         dout("%s %p (r_request %p r_reply %p)\n", __func__, req,
 513              req->r_request, req->r_reply);
 514         request_release_checks(req);
 515
 516         if (req->r_request)
 517                 ceph_msg_put(req->r_request);
 518         if (req->r_reply)
 519                 ceph_msg_put(req->r_reply);
 520
 521         for (which = 0; which < req->r_num_ops; which++)
 522                 osd_req_op_data_release(req, which);
 523
 524         target_destroy(&req->r_t);
 525         ceph_put_snap_context(req->r_snapc);
 526
 527         if (req->r_mempool)
 528                 mempool_free(req, req->r_osdc->req_mempool);
 529         else if (req->r_num_ops <= CEPH_OSD_SLAB_OPS)
 530                 kmem_cache_free(ceph_osd_request_cache, req);
 531         else
 532                 kfree(req);
 533 }
 534
 535 void ceph_osdc_get_request(struct ceph_osd_request *req)
 536 {
 537         dout("%s %p (was %d)\n", __func__, req,
 538              kref_read(&req->r_kref));
 539         kref_get(&req->r_kref);
 540 }
 541 EXPORT_SYMBOL(ceph_osdc_get_request);
 542
 543 void ceph_osdc_put_request(struct ceph_osd_request *req)
 544 {
 545         if (req) {
 546                 dout("%s %p (was %d)\n", __func__, req,
 547                      kref_read(&req->r_kref));
 548                 kref_put(&req->r_kref, ceph_osdc_release_request);
 549         }
 550 }
 551 EXPORT_SYMBOL(ceph_osdc_put_request);
 552
 553 static void request_init(struct ceph_osd_request *req)
 554 {
 555         /* req only, each op is zeroed in osd_req_op_init() */
 556         memset(req, 0, sizeof(*req));
 557
 558         kref_init(&req->r_kref);
 559         init_completion(&req->r_completion);
 560         RB_CLEAR_NODE(&req->r_node);
 561         RB_CLEAR_NODE(&req->r_mc_node);
 562         INIT_LIST_HEAD(&req->r_private_item);
 563
 564         target_init(&req->r_t);
 565 }
 566
 567 struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
 568                                                struct ceph_snap_context *snapc,
 569                                                unsigned int num_ops,
 570                                                bool use_mempool,
 571                                                gfp_t gfp_flags)
 572 {
 573         struct ceph_osd_request *req;
 574
 575         if (use_mempool) {
 576                 BUG_ON(num_ops > CEPH_OSD_SLAB_OPS);
 577                 req = mempool_alloc(osdc->req_mempool, gfp_flags);
 578         } else if (num_ops <= CEPH_OSD_SLAB_OPS) {
 579                 req = kmem_cache_alloc(ceph_osd_request_cache, gfp_flags);
 580         } else {
 581                 BUG_ON(num_ops > CEPH_OSD_MAX_OPS);
 582                 req = kmalloc(struct_size(req, r_ops, num_ops), gfp_flags);
 583         }
 584         if (unlikely(!req))
 585                 return NULL;
 586
 587         request_init(req);
 588         req->r_osdc = osdc;
 589         req->r_mempool = use_mempool;
 590         req->r_num_ops = num_ops;
 591         req->r_snapid = CEPH_NOSNAP;
 592         req->r_snapc = ceph_get_snap_context(snapc);
 593
 594         dout("%s req %p\n", __func__, req);
 595         return req;
 596 }
 597 EXPORT_SYMBOL(ceph_osdc_alloc_request);
 598
 599 static int ceph_oloc_encoding_size(const struct ceph_object_locator *oloc)
 600 {
 601         return 8 + 4 + 4 + 4 + (oloc->pool_ns ? oloc->pool_ns->len : 0);
 602 }
 603
 604 static int __ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp,
 605                                       int num_request_data_items,
 606                                       int num_reply_data_items)
 607 {
 608         struct ceph_osd_client *osdc = req->r_osdc;
 609         struct ceph_msg *msg;
 610         int msg_size;
 611
 612         WARN_ON(req->r_request || req->r_reply);
 613         WARN_ON(ceph_oid_empty(&req->r_base_oid));
 614         WARN_ON(ceph_oloc_empty(&req->r_base_oloc));
 615
 616         /* create request message */
 617         msg_size = CEPH_ENCODING_START_BLK_LEN +
 618                         CEPH_PGID_ENCODING_LEN + 1; /* spgid */
 619         msg_size += 4 + 4 + 4; /* hash, osdmap_epoch, flags */
 620         msg_size += CEPH_ENCODING_START_BLK_LEN +
 621                         sizeof(struct ceph_osd_reqid); /* reqid */
 622         msg_size += sizeof(struct ceph_blkin_trace_info); /* trace */
 623         msg_size += 4 + sizeof(struct ceph_timespec); /* client_inc, mtime */
 624         msg_size += CEPH_ENCODING_START_BLK_LEN +
 625                         ceph_oloc_encoding_size(&req->r_base_oloc); /* oloc */
 626         msg_size += 4 + req->r_base_oid.name_len; /* oid */
 627         msg_size += 2 + req->r_num_ops * sizeof(struct ceph_osd_op);
 628         msg_size += 8; /* snapid */
 629         msg_size += 8; /* snap_seq */
 630         msg_size += 4 + 8 * (req->r_snapc ? req->r_snapc->num_snaps : 0);
 631         msg_size += 4 + 8; /* retry_attempt, features */
 632
 633         if (req->r_mempool)
 634                 msg = ceph_msgpool_get(&osdc->msgpool_op, msg_size,
 635                                        num_request_data_items);
 636         else
 637                 msg = ceph_msg_new2(CEPH_MSG_OSD_OP, msg_size,
 638                                     num_request_data_items, gfp, true);
 639         if (!msg)
 640                 return -ENOMEM;
 641
 642         memset(msg->front.iov_base, 0, msg->front.iov_len);
 643         req->r_request = msg;
 644
 645         /* create reply message */
 646         msg_size = OSD_OPREPLY_FRONT_LEN;
 647         msg_size += req->r_base_oid.name_len;
 648         msg_size += req->r_num_ops * sizeof(struct ceph_osd_op);
 649
 650         if (req->r_mempool)
 651                 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, msg_size,
 652                                        num_reply_data_items);
 653         else
 654                 msg = ceph_msg_new2(CEPH_MSG_OSD_OPREPLY, msg_size,
 655                                     num_reply_data_items, gfp, true);
 656         if (!msg)
 657                 return -ENOMEM;
 658
 659         req->r_reply = msg;
 660
 661         return 0;
 662 }
 663
 664 static bool osd_req_opcode_valid(u16 opcode)
 665 {
 666         switch (opcode) {
 667 #define GENERATE_CASE(op, opcode, str)  case CEPH_OSD_OP_##op: return true;
 668 __CEPH_FORALL_OSD_OPS(GENERATE_CASE)
 669 #undef GENERATE_CASE
 670         default:
 671                 return false;
 672         }
 673 }
 674
 675 static void get_num_data_items(struct ceph_osd_request *req,
 676                                int *num_request_data_items,
 677                                int *num_reply_data_items)
 678 {
 679         struct ceph_osd_req_op *op;
 680
 681         *num_request_data_items = 0;
 682         *num_reply_data_items = 0;
 683
 684         for (op = req->r_ops; op != &req->r_ops[req->r_num_ops]; op++) {
 685                 switch (op->op) {
 686                 /* request */
 687                 case CEPH_OSD_OP_WRITE:
 688                 case CEPH_OSD_OP_WRITEFULL:
 689                 case CEPH_OSD_OP_SETXATTR:
 690                 case CEPH_OSD_OP_CMPXATTR:
 691                 case CEPH_OSD_OP_NOTIFY_ACK:
 692                 case CEPH_OSD_OP_COPY_FROM2:
 693                         *num_request_data_items += 1;
 694                         break;
 695
 696                 /* reply */
 697                 case CEPH_OSD_OP_STAT:
 698                 case CEPH_OSD_OP_READ:
 699                 case CEPH_OSD_OP_SPARSE_READ:
 700                 case CEPH_OSD_OP_LIST_WATCHERS:
 701                         *num_reply_data_items += 1;
 702                         break;
 703
 704                 /* both */
 705                 case CEPH_OSD_OP_NOTIFY:
 706                         *num_request_data_items += 1;
 707                         *num_reply_data_items += 1;
 708                         break;
 709                 case CEPH_OSD_OP_CALL:
 710                         *num_request_data_items += 2;
 711                         *num_reply_data_items += 1;
 712                         break;
 713
 714                 default:
 715                         WARN_ON(!osd_req_opcode_valid(op->op));
 716                         break;
 717                 }
 718         }
 719 }
 720
 721 /*
 722  * oid, oloc and OSD op opcode(s) must be filled in before this function
 723  * is called.
 724  */
 725 int ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp)
 726 {
 727         int num_request_data_items, num_reply_data_items;
 728
 729         get_num_data_items(req, &num_request_data_items, &num_reply_data_items);
 730         return __ceph_osdc_alloc_messages(req, gfp, num_request_data_items,
 731                                           num_reply_data_items);
 732 }
 733 EXPORT_SYMBOL(ceph_osdc_alloc_messages);
 734
 735 /*
 736  * This is an osd op init function for opcodes that have no data or
 737  * other information associated with them.  It also serves as a
 738  * common init routine for all the other init functions, below.
 739  */
 740 struct ceph_osd_req_op *
 741 osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which,
 742                  u16 opcode, u32 flags)
 743 {
 744         struct ceph_osd_req_op *op;
 745
 746         BUG_ON(which >= osd_req->r_num_ops);
 747         BUG_ON(!osd_req_opcode_valid(opcode));
 748
 749         op = &osd_req->r_ops[which];
 750         memset(op, 0, sizeof (*op));
 751         op->op = opcode;
 752         op->flags = flags;
 753
 754         return op;
 755 }
 756 EXPORT_SYMBOL(osd_req_op_init);
 757
 758 void osd_req_op_extent_init(struct ceph_osd_request *osd_req,
 759                                 unsigned int which, u16 opcode,
 760                                 u64 offset, u64 length,
 761                                 u64 truncate_size, u32 truncate_seq)
 762 {
 763         struct ceph_osd_req_op *op = osd_req_op_init(osd_req, which,
 764                                                      opcode, 0);
 765         size_t payload_len = 0;
 766
 767         BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
 768                opcode != CEPH_OSD_OP_WRITEFULL && opcode != CEPH_OSD_OP_ZERO &&
 769                opcode != CEPH_OSD_OP_TRUNCATE && opcode != CEPH_OSD_OP_SPARSE_READ);
 770
 771         op->extent.offset = offset;
 772         op->extent.length = length;
 773         op->extent.truncate_size = truncate_size;
 774         op->extent.truncate_seq = truncate_seq;
 775         if (opcode == CEPH_OSD_OP_WRITE || opcode == CEPH_OSD_OP_WRITEFULL)
 776                 payload_len += length;
 777
 778         op->indata_len = payload_len;
 779 }
 780 EXPORT_SYMBOL(osd_req_op_extent_init);
 781
 782 void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
 783                                 unsigned int which, u64 length)
 784 {
 785         struct ceph_osd_req_op *op;
 786         u64 previous;
 787
 788         BUG_ON(which >= osd_req->r_num_ops);
 789         op = &osd_req->r_ops[which];
 790         previous = op->extent.length;
 791
 792         if (length == previous)
 793                 return;         /* Nothing to do */
 794         BUG_ON(length > previous);
 795
 796         op->extent.length = length;
 797         if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL)
 798                 op->indata_len -= previous - length;
 799 }
 800 EXPORT_SYMBOL(osd_req_op_extent_update);
 801
 802 void osd_req_op_extent_dup_last(struct ceph_osd_request *osd_req,
 803                                 unsigned int which, u64 offset_inc)
 804 {
 805         struct ceph_osd_req_op *op, *prev_op;
 806
 807         BUG_ON(which + 1 >= osd_req->r_num_ops);
 808
 809         prev_op = &osd_req->r_ops[which];
 810         op = osd_req_op_init(osd_req, which + 1, prev_op->op, prev_op->flags);
 811         /* dup previous one */
 812         op->indata_len = prev_op->indata_len;
 813         op->outdata_len = prev_op->outdata_len;
 814         op->extent = prev_op->extent;
 815         /* adjust offset */
 816         op->extent.offset += offset_inc;
 817         op->extent.length -= offset_inc;
 818
 819         if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL)
 820                 op->indata_len -= offset_inc;
 821 }
 822 EXPORT_SYMBOL(osd_req_op_extent_dup_last);
 823
 824 int osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
 825                         const char *class, const char *method)
 826 {
 827         struct ceph_osd_req_op *op;
 828         struct ceph_pagelist *pagelist;
 829         size_t payload_len = 0;
 830         size_t size;
 831         int ret;
 832
 833         op = osd_req_op_init(osd_req, which, CEPH_OSD_OP_CALL, 0);
 834
 835         pagelist = ceph_pagelist_alloc(GFP_NOFS);
 836         if (!pagelist)
 837                 return -ENOMEM;
 838
 839         op->cls.class_name = class;
 840         size = strlen(class);
 841         BUG_ON(size > (size_t) U8_MAX);
 842         op->cls.class_len = size;
 843         ret = ceph_pagelist_append(pagelist, class, size);
 844         if (ret)
 845                 goto err_pagelist_free;
 846         payload_len += size;
 847
 848         op->cls.method_name = method;
 849         size = strlen(method);
 850         BUG_ON(size > (size_t) U8_MAX);
 851         op->cls.method_len = size;
 852         ret = ceph_pagelist_append(pagelist, method, size);
 853         if (ret)
 854                 goto err_pagelist_free;
 855         payload_len += size;
 856
 857         osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist);
 858         op->indata_len = payload_len;
 859         return 0;
 860
 861 err_pagelist_free:
 862         ceph_pagelist_release(pagelist);
 863         return ret;
 864 }
 865 EXPORT_SYMBOL(osd_req_op_cls_init);
 866
 867 int osd_req_op_xattr_init(struct ceph_osd_request *osd_req, unsigned int which,
 868                           u16 opcode, const char *name, const void *value,
 869                           size_t size, u8 cmp_op, u8 cmp_mode)
 870 {
 871         struct ceph_osd_req_op *op = osd_req_op_init(osd_req, which,
 872                                                      opcode, 0);
 873         struct ceph_pagelist *pagelist;
 874         size_t payload_len;
 875         int ret;
 876
 877         BUG_ON(opcode != CEPH_OSD_OP_SETXATTR && opcode != CEPH_OSD_OP_CMPXATTR);
 878
 879         pagelist = ceph_pagelist_alloc(GFP_NOFS);
 880         if (!pagelist)
 881                 return -ENOMEM;
 882
 883         payload_len = strlen(name);
 884         op->xattr.name_len = payload_len;
 885         ret = ceph_pagelist_append(pagelist, name, payload_len);
 886         if (ret)
 887                 goto err_pagelist_free;
 888
 889         op->xattr.value_len = size;
 890         ret = ceph_pagelist_append(pagelist, value, size);
 891         if (ret)
 892                 goto err_pagelist_free;
 893         payload_len += size;
 894
 895         op->xattr.cmp_op = cmp_op;
 896         op->xattr.cmp_mode = cmp_mode;
 897
 898         ceph_osd_data_pagelist_init(&op->xattr.osd_data, pagelist);
 899         op->indata_len = payload_len;
 900         return 0;
 901
 902 err_pagelist_free:
 903         ceph_pagelist_release(pagelist);
 904         return ret;
 905 }
 906 EXPORT_SYMBOL(osd_req_op_xattr_init);
 907
 908 /*
 909  * @watch_opcode: CEPH_OSD_WATCH_OP_*
 910  */
 911 static void osd_req_op_watch_init(struct ceph_osd_request *req, int which,
 912                                   u8 watch_opcode, u64 cookie, u32 gen)
 913 {
 914         struct ceph_osd_req_op *op;
 915
 916         op = osd_req_op_init(req, which, CEPH_OSD_OP_WATCH, 0);
 917         op->watch.cookie = cookie;
 918         op->watch.op = watch_opcode;
 919         op->watch.gen = gen;
 920 }
 921
 922 /*
 923  * prot_ver, timeout and notify payload (may be empty) should already be
 924  * encoded in @request_pl
 925  */
 926 static void osd_req_op_notify_init(struct ceph_osd_request *req, int which,
 927                                    u64 cookie, struct ceph_pagelist *request_pl)
 928 {
 929         struct ceph_osd_req_op *op;
 930
 931         op = osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY, 0);
 932         op->notify.cookie = cookie;
 933
 934         ceph_osd_data_pagelist_init(&op->notify.request_data, request_pl);
 935         op->indata_len = request_pl->length;
 936 }
 937
 938 /*
 939  * @flags: CEPH_OSD_OP_ALLOC_HINT_FLAG_*
 940  */
 941 void osd_req_op_alloc_hint_init(struct ceph_osd_request *osd_req,
 942                                 unsigned int which,
 943                                 u64 expected_object_size,
 944                                 u64 expected_write_size,
 945                                 u32 flags)
 946 {
 947         struct ceph_osd_req_op *op;
 948
 949         op = osd_req_op_init(osd_req, which, CEPH_OSD_OP_SETALLOCHINT, 0);
 950         op->alloc_hint.expected_object_size = expected_object_size;
 951         op->alloc_hint.expected_write_size = expected_write_size;
 952         op->alloc_hint.flags = flags;
 953
 954         /*
 955          * CEPH_OSD_OP_SETALLOCHINT op is advisory and therefore deemed
 956          * not worth a feature bit.  Set FAILOK per-op flag to make
 957          * sure older osds don't trip over an unsupported opcode.
 958          */
 959         op->flags |= CEPH_OSD_OP_FLAG_FAILOK;
 960 }
 961 EXPORT_SYMBOL(osd_req_op_alloc_hint_init);
 962
 963 static void ceph_osdc_msg_data_add(struct ceph_msg *msg,
 964                                 struct ceph_osd_data *osd_data)
 965 {
 966         u64 length = ceph_osd_data_length(osd_data);
 967
 968         if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
 969                 BUG_ON(length > (u64) SIZE_MAX);
 970                 if (length)
 971                         ceph_msg_data_add_pages(msg, osd_data->pages,
 972                                         length, osd_data->alignment, false);
 973         } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
 974                 BUG_ON(!length);
 975                 ceph_msg_data_add_pagelist(msg, osd_data->pagelist);
 976 #ifdef CONFIG_BLOCK
 977         } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
 978                 ceph_msg_data_add_bio(msg, &osd_data->bio_pos, length);
 979 #endif
 980         } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BVECS) {
 981                 ceph_msg_data_add_bvecs(msg, &osd_data->bvec_pos);
 982         } else if (osd_data->type == CEPH_OSD_DATA_TYPE_ITER) {
 983                 ceph_msg_data_add_iter(msg, &osd_data->iter);
 984         } else {
 985                 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
 986         }
 987 }
 988
 989 static u32 osd_req_encode_op(struct ceph_osd_op *dst,
 990                              const struct ceph_osd_req_op *src)
 991 {
 992         switch (src->op) {
 993         case CEPH_OSD_OP_STAT:
 994                 break;
 995         case CEPH_OSD_OP_READ:
 996         case CEPH_OSD_OP_SPARSE_READ:
 997         case CEPH_OSD_OP_WRITE:
 998         case CEPH_OSD_OP_WRITEFULL:
 999         case CEPH_OSD_OP_ZERO:
1000         case CEPH_OSD_OP_TRUNCATE:
1001                 dst->extent.offset = cpu_to_le64(src->extent.offset);
1002                 dst->extent.length = cpu_to_le64(src->extent.length);
1003                 dst->extent.truncate_size =
1004                         cpu_to_le64(src->extent.truncate_size);
1005                 dst->extent.truncate_seq =
1006                         cpu_to_le32(src->extent.truncate_seq);
1007                 break;
1008         case CEPH_OSD_OP_CALL:
1009                 dst->cls.class_len = src->cls.class_len;
1010                 dst->cls.method_len = src->cls.method_len;
1011                 dst->cls.indata_len = cpu_to_le32(src->cls.indata_len);
1012                 break;
1013         case CEPH_OSD_OP_WATCH:
1014                 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
1015                 dst->watch.ver = cpu_to_le64(0);
1016                 dst->watch.op = src->watch.op;
1017                 dst->watch.gen = cpu_to_le32(src->watch.gen);
1018                 break;
1019         case CEPH_OSD_OP_NOTIFY_ACK:
1020                 break;
1021         case CEPH_OSD_OP_NOTIFY:
1022                 dst->notify.cookie = cpu_to_le64(src->notify.cookie);
1023                 break;
1024         case CEPH_OSD_OP_LIST_WATCHERS:
1025                 break;
1026         case CEPH_OSD_OP_SETALLOCHINT:
1027                 dst->alloc_hint.expected_object_size =
1028                     cpu_to_le64(src->alloc_hint.expected_object_size);
1029                 dst->alloc_hint.expected_write_size =
1030                     cpu_to_le64(src->alloc_hint.expected_write_size);
1031                 dst->alloc_hint.flags = cpu_to_le32(src->alloc_hint.flags);
1032                 break;
1033         case CEPH_OSD_OP_SETXATTR:
1034         case CEPH_OSD_OP_CMPXATTR:
1035                 dst->xattr.name_len = cpu_to_le32(src->xattr.name_len);
1036                 dst->xattr.value_len = cpu_to_le32(src->xattr.value_len);
1037                 dst->xattr.cmp_op = src->xattr.cmp_op;
1038                 dst->xattr.cmp_mode = src->xattr.cmp_mode;
1039                 break;
1040         case CEPH_OSD_OP_CREATE:
1041         case CEPH_OSD_OP_DELETE:
1042                 break;
1043         case CEPH_OSD_OP_COPY_FROM2:
1044                 dst->copy_from.snapid = cpu_to_le64(src->copy_from.snapid);
1045                 dst->copy_from.src_version =
1046                         cpu_to_le64(src->copy_from.src_version);
1047                 dst->copy_from.flags = src->copy_from.flags;
1048                 dst->copy_from.src_fadvise_flags =
1049                         cpu_to_le32(src->copy_from.src_fadvise_flags);
1050                 break;
1051         case CEPH_OSD_OP_ASSERT_VER:
1052                 dst->assert_ver.unused = cpu_to_le64(0);
1053                 dst->assert_ver.ver = cpu_to_le64(src->assert_ver.ver);
1054                 break;
1055         default:
1056                 pr_err("unsupported osd opcode %s\n",
1057                         ceph_osd_op_name(src->op));
1058                 WARN_ON(1);
1059
1060                 return 0;
1061         }
1062
1063         dst->op = cpu_to_le16(src->op);
1064         dst->flags = cpu_to_le32(src->flags);
1065         dst->payload_len = cpu_to_le32(src->indata_len);
1066
1067         return src->indata_len;
1068 }
1069
1070 /*
1071  * build new request AND message, calculate layout, and adjust file
1072  * extent as needed.
1073  *
1074  * if the file was recently truncated, we include information about its
1075  * old and new size so that the object can be updated appropriately.  (we
1076  * avoid synchronously deleting truncated objects because it's slow.)
1077  */
1078 struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
1079                                                struct ceph_file_layout *layout,
1080                                                struct ceph_vino vino,
1081                                                u64 off, u64 *plen,
1082                                                unsigned int which, int num_ops,
1083                                                int opcode, int flags,
1084                                                struct ceph_snap_context *snapc,
1085                                                u32 truncate_seq,
1086                                                u64 truncate_size,
1087                                                bool use_mempool)
1088 {
1089         struct ceph_osd_request *req;
1090         u64 objnum = 0;
1091         u64 objoff = 0;
1092         u64 objlen = 0;
1093         int r;
1094
1095         BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
1096                opcode != CEPH_OSD_OP_ZERO && opcode != CEPH_OSD_OP_TRUNCATE &&
1097                opcode != CEPH_OSD_OP_CREATE && opcode != CEPH_OSD_OP_DELETE &&
1098                opcode != CEPH_OSD_OP_SPARSE_READ);
1099
1100         req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
1101                                         GFP_NOFS);
1102         if (!req) {
1103                 r = -ENOMEM;
1104                 goto fail;
1105         }
1106
1107         /* calculate max write size */
1108         r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
1109         if (r)
1110                 goto fail;
1111
1112         if (opcode == CEPH_OSD_OP_CREATE || opcode == CEPH_OSD_OP_DELETE) {
1113                 osd_req_op_init(req, which, opcode, 0);
1114         } else {
1115                 u32 object_size = layout->object_size;
1116                 u32 object_base = off - objoff;
1117                 if (!(truncate_seq == 1 && truncate_size == -1ULL)) {
1118                         if (truncate_size <= object_base) {
1119                                 truncate_size = 0;
1120                         } else {
1121                                 truncate_size -= object_base;
1122                                 if (truncate_size > object_size)
1123                                         truncate_size = object_size;
1124                         }
1125                 }
1126                 osd_req_op_extent_init(req, which, opcode, objoff, objlen,
1127                                        truncate_size, truncate_seq);
1128         }
1129
1130         req->r_base_oloc.pool = layout->pool_id;
1131         req->r_base_oloc.pool_ns = ceph_try_get_string(layout->pool_ns);
1132         ceph_oid_printf(&req->r_base_oid, "%llx.%08llx", vino.ino, objnum);
1133         req->r_flags = flags | osdc->client->options->read_from_replica;
1134
1135         req->r_snapid = vino.snap;
1136         if (flags & CEPH_OSD_FLAG_WRITE)
1137                 req->r_data_offset = off;
1138
1139         if (num_ops > 1) {
1140                 int num_req_ops, num_rep_ops;
1141
1142                 /*
1143                  * If this is a multi-op write request, assume that we'll need
1144                  * request ops. If it's a multi-op read then assume we'll need
1145                  * reply ops. Anything else and call it -EINVAL.
1146                  */
1147                 if (flags & CEPH_OSD_FLAG_WRITE) {
1148                         num_req_ops = num_ops;
1149                         num_rep_ops = 0;
1150                 } else if (flags & CEPH_OSD_FLAG_READ) {
1151                         num_req_ops = 0;
1152                         num_rep_ops = num_ops;
1153                 } else {
1154                         r = -EINVAL;
1155                         goto fail;
1156                 }
1157
1158                 r = __ceph_osdc_alloc_messages(req, GFP_NOFS, num_req_ops,
1159                                                num_rep_ops);
1160         } else {
1161                 r = ceph_osdc_alloc_messages(req, GFP_NOFS);
1162         }
1163         if (r)
1164                 goto fail;
1165
1166         return req;
1167
1168 fail:
1169         ceph_osdc_put_request(req);
1170         return ERR_PTR(r);
1171 }
1172 EXPORT_SYMBOL(ceph_osdc_new_request);
1173
1174 int __ceph_alloc_sparse_ext_map(struct ceph_osd_req_op *op, int cnt)
1175 {
1176         op->extent.sparse_ext_cnt = cnt;
1177         op->extent.sparse_ext = kmalloc_array(cnt,
1178                                               sizeof(*op->extent.sparse_ext),
1179                                               GFP_NOFS);
1180         if (!op->extent.sparse_ext)
1181                 return -ENOMEM;
1182         return 0;
1183 }
1184 EXPORT_SYMBOL(__ceph_alloc_sparse_ext_map);
1185
1186 /*
1187  * We keep osd requests in an rbtree, sorted by ->r_tid.
1188  */
1189 DEFINE_RB_FUNCS(request, struct ceph_osd_request, r_tid, r_node)
1190 DEFINE_RB_FUNCS(request_mc, struct ceph_osd_request, r_tid, r_mc_node)
1191
1192 /*
1193  * Call @fn on each OSD request as long as @fn returns 0.
1194  */
1195 static void for_each_request(struct ceph_osd_client *osdc,
1196                         int (*fn)(struct ceph_osd_request *req, void *arg),
1197                         void *arg)
1198 {
1199         struct rb_node *n, *p;
1200
1201         for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
1202                 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
1203
1204                 for (p = rb_first(&osd->o_requests); p; ) {
1205                         struct ceph_osd_request *req =
1206                             rb_entry(p, struct ceph_osd_request, r_node);
1207
1208                         p = rb_next(p);
1209                         if (fn(req, arg))
1210                                 return;
1211                 }
1212         }
1213
1214         for (p = rb_first(&osdc->homeless_osd.o_requests); p; ) {
1215                 struct ceph_osd_request *req =
1216                     rb_entry(p, struct ceph_osd_request, r_node);
1217
1218                 p = rb_next(p);
1219                 if (fn(req, arg))
1220                         return;
1221         }
1222 }
1223
1224 static bool osd_homeless(struct ceph_osd *osd)
1225 {
1226         return osd->o_osd == CEPH_HOMELESS_OSD;
1227 }
1228
1229 static bool osd_registered(struct ceph_osd *osd)
1230 {
1231         verify_osdc_locked(osd->o_osdc);
1232
1233         return !RB_EMPTY_NODE(&osd->o_node);
1234 }
1235
1236 /*
1237  * Assumes @osd is zero-initialized.
1238  */
1239 static void osd_init(struct ceph_osd *osd)
1240 {
1241         refcount_set(&osd->o_ref, 1);
1242         RB_CLEAR_NODE(&osd->o_node);
1243         spin_lock_init(&osd->o_requests_lock);
1244         osd->o_requests = RB_ROOT;
1245         osd->o_linger_requests = RB_ROOT;
1246         osd->o_backoff_mappings = RB_ROOT;
1247         osd->o_backoffs_by_id = RB_ROOT;
1248         INIT_LIST_HEAD(&osd->o_osd_lru);
1249         INIT_LIST_HEAD(&osd->o_keepalive_item);
1250         osd->o_incarnation = 1;
1251         mutex_init(&osd->lock);
1252 }
1253
1254 static void ceph_init_sparse_read(struct ceph_sparse_read *sr)
1255 {
1256         kfree(sr->sr_extent);
1257         memset(sr, '\0', sizeof(*sr));
1258         sr->sr_state = CEPH_SPARSE_READ_HDR;
1259 }
1260
1261 static void osd_cleanup(struct ceph_osd *osd)
1262 {
1263         WARN_ON(!RB_EMPTY_NODE(&osd->o_node));
1264         WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests));
1265         WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests));
1266         WARN_ON(!RB_EMPTY_ROOT(&osd->o_backoff_mappings));
1267         WARN_ON(!RB_EMPTY_ROOT(&osd->o_backoffs_by_id));
1268         WARN_ON(!list_empty(&osd->o_osd_lru));
1269         WARN_ON(!list_empty(&osd->o_keepalive_item));
1270
1271         ceph_init_sparse_read(&osd->o_sparse_read);
1272
1273         if (osd->o_auth.authorizer) {
1274                 WARN_ON(osd_homeless(osd));
1275                 ceph_auth_destroy_authorizer(osd->o_auth.authorizer);
1276         }
1277 }
1278
1279 /*
1280  * Track open sessions with osds.
1281  */
1282 static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
1283 {
1284         struct ceph_osd *osd;
1285
1286         WARN_ON(onum == CEPH_HOMELESS_OSD);
1287
1288         osd = kzalloc(sizeof(*osd), GFP_NOIO | __GFP_NOFAIL);
1289         osd_init(osd);
1290         osd->o_osdc = osdc;
1291         osd->o_osd = onum;
1292         osd->o_sparse_op_idx = -1;
1293
1294         ceph_init_sparse_read(&osd->o_sparse_read);
1295
1296         ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
1297
1298         return osd;
1299 }
1300
1301 static struct ceph_osd *get_osd(struct ceph_osd *osd)
1302 {
1303         if (refcount_inc_not_zero(&osd->o_ref)) {
1304                 dout("get_osd %p %d -> %d\n", osd, refcount_read(&osd->o_ref)-1,
1305                      refcount_read(&osd->o_ref));
1306                 return osd;
1307         } else {
1308                 dout("get_osd %p FAIL\n", osd);
1309                 return NULL;
1310         }
1311 }
1312
1313 static void put_osd(struct ceph_osd *osd)
1314 {
1315         dout("put_osd %p %d -> %d\n", osd, refcount_read(&osd->o_ref),
1316              refcount_read(&osd->o_ref) - 1);
1317         if (refcount_dec_and_test(&osd->o_ref)) {
1318                 osd_cleanup(osd);
1319                 kfree(osd);
1320         }
1321 }
1322
1323 DEFINE_RB_FUNCS(osd, struct ceph_osd, o_osd, o_node)
1324
1325 static void __move_osd_to_lru(struct ceph_osd *osd)
1326 {
1327         struct ceph_osd_client *osdc = osd->o_osdc;
1328
1329         dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1330         BUG_ON(!list_empty(&osd->o_osd_lru));
1331
1332         spin_lock(&osdc->osd_lru_lock);
1333         list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
1334         spin_unlock(&osdc->osd_lru_lock);
1335
1336         osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl;
1337 }
1338
1339 static void maybe_move_osd_to_lru(struct ceph_osd *osd)
1340 {
1341         if (RB_EMPTY_ROOT(&osd->o_requests) &&
1342             RB_EMPTY_ROOT(&osd->o_linger_requests))
1343                 __move_osd_to_lru(osd);
1344 }
1345
1346 static void __remove_osd_from_lru(struct ceph_osd *osd)
1347 {
1348         struct ceph_osd_client *osdc = osd->o_osdc;
1349
1350         dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1351
1352         spin_lock(&osdc->osd_lru_lock);
1353         if (!list_empty(&osd->o_osd_lru))
1354                 list_del_init(&osd->o_osd_lru);
1355         spin_unlock(&osdc->osd_lru_lock);
1356 }
1357
1358 /*
1359  * Close the connection and assign any leftover requests to the
1360  * homeless session.
1361  */
1362 static void close_osd(struct ceph_osd *osd)
1363 {
1364         struct ceph_osd_client *osdc = osd->o_osdc;
1365         struct rb_node *n;
1366
1367         verify_osdc_wrlocked(osdc);
1368         dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1369
1370         ceph_con_close(&osd->o_con);
1371
1372         for (n = rb_first(&osd->o_requests); n; ) {
1373                 struct ceph_osd_request *req =
1374                     rb_entry(n, struct ceph_osd_request, r_node);
1375
1376                 n = rb_next(n); /* unlink_request() */
1377
1378                 dout(" reassigning req %p tid %llu\n", req, req->r_tid);
1379                 unlink_request(osd, req);
1380                 link_request(&osdc->homeless_osd, req);
1381         }
1382         for (n = rb_first(&osd->o_linger_requests); n; ) {
1383                 struct ceph_osd_linger_request *lreq =
1384                     rb_entry(n, struct ceph_osd_linger_request, node);
1385
1386                 n = rb_next(n); /* unlink_linger() */
1387
1388                 dout(" reassigning lreq %p linger_id %llu\n", lreq,
1389                      lreq->linger_id);
1390                 unlink_linger(osd, lreq);
1391                 link_linger(&osdc->homeless_osd, lreq);
1392         }
1393         clear_backoffs(osd);
1394
1395         __remove_osd_from_lru(osd);
1396         erase_osd(&osdc->osds, osd);
1397         put_osd(osd);
1398 }
1399
1400 /*
1401  * reset osd connect
1402  */
1403 static int reopen_osd(struct ceph_osd *osd)
1404 {
1405         struct ceph_entity_addr *peer_addr;
1406
1407         dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1408
1409         if (RB_EMPTY_ROOT(&osd->o_requests) &&
1410             RB_EMPTY_ROOT(&osd->o_linger_requests)) {
1411                 close_osd(osd);
1412                 return -ENODEV;
1413         }
1414
1415         peer_addr = &osd->o_osdc->osdmap->osd_addr[osd->o_osd];
1416         if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1417                         !ceph_con_opened(&osd->o_con)) {
1418                 struct rb_node *n;
1419
1420                 dout("osd addr hasn't changed and connection never opened, "
1421                      "letting msgr retry\n");
1422                 /* touch each r_stamp for handle_timeout()'s benfit */
1423                 for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) {
1424                         struct ceph_osd_request *req =
1425                             rb_entry(n, struct ceph_osd_request, r_node);
1426                         req->r_stamp = jiffies;
1427                 }
1428
1429                 return -EAGAIN;
1430         }
1431
1432         ceph_con_close(&osd->o_con);
1433         ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1434         osd->o_incarnation++;
1435
1436         return 0;
1437 }
1438
1439 static struct ceph_osd *lookup_create_osd(struct ceph_osd_client *osdc, int o,
1440                                           bool wrlocked)
1441 {
1442         struct ceph_osd *osd;
1443
1444         if (wrlocked)
1445                 verify_osdc_wrlocked(osdc);
1446         else
1447                 verify_osdc_locked(osdc);
1448
1449         if (o != CEPH_HOMELESS_OSD)
1450                 osd = lookup_osd(&osdc->osds, o);
1451         else
1452                 osd = &osdc->homeless_osd;
1453         if (!osd) {
1454                 if (!wrlocked)
1455                         return ERR_PTR(-EAGAIN);
1456
1457                 osd = create_osd(osdc, o);
1458                 insert_osd(&osdc->osds, osd);
1459                 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd,
1460                               &osdc->osdmap->osd_addr[osd->o_osd]);
1461         }
1462
1463         dout("%s osdc %p osd%d -> osd %p\n", __func__, osdc, o, osd);
1464         return osd;
1465 }
1466
1467 /*
1468  * Create request <-> OSD session relation.
1469  *
1470  * @req has to be assigned a tid, @osd may be homeless.
1471  */
1472 static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req)
1473 {
1474         verify_osd_locked(osd);
1475         WARN_ON(!req->r_tid || req->r_osd);
1476         dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd,
1477              req, req->r_tid);
1478
1479         if (!osd_homeless(osd))
1480                 __remove_osd_from_lru(osd);
1481         else
1482                 atomic_inc(&osd->o_osdc->num_homeless);
1483
1484         get_osd(osd);
1485         spin_lock(&osd->o_requests_lock);
1486         insert_request(&osd->o_requests, req);
1487         spin_unlock(&osd->o_requests_lock);
1488         req->r_osd = osd;
1489 }
1490
1491 static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req)
1492 {
1493         verify_osd_locked(osd);
1494         WARN_ON(req->r_osd != osd);
1495         dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd,
1496              req, req->r_tid);
1497
1498         req->r_osd = NULL;
1499         spin_lock(&osd->o_requests_lock);
1500         erase_request(&osd->o_requests, req);
1501         spin_unlock(&osd->o_requests_lock);
1502         put_osd(osd);
1503
1504         if (!osd_homeless(osd))
1505                 maybe_move_osd_to_lru(osd);
1506         else
1507                 atomic_dec(&osd->o_osdc->num_homeless);
1508 }
1509
1510 static bool __pool_full(struct ceph_pg_pool_info *pi)
1511 {
1512         return pi->flags & CEPH_POOL_FLAG_FULL;
1513 }
1514
1515 static bool have_pool_full(struct ceph_osd_client *osdc)
1516 {
1517         struct rb_node *n;
1518
1519         for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) {
1520                 struct ceph_pg_pool_info *pi =
1521                     rb_entry(n, struct ceph_pg_pool_info, node);
1522
1523                 if (__pool_full(pi))
1524                         return true;
1525         }
1526
1527         return false;
1528 }
1529
1530 static bool pool_full(struct ceph_osd_client *osdc, s64 pool_id)
1531 {
1532         struct ceph_pg_pool_info *pi;
1533
1534         pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id);
1535         if (!pi)
1536                 return false;
1537
1538         return __pool_full(pi);
1539 }
1540
1541 /*
1542  * Returns whether a request should be blocked from being sent
1543  * based on the current osdmap and osd_client settings.
1544  */
1545 static bool target_should_be_paused(struct ceph_osd_client *osdc,
1546                                     const struct ceph_osd_request_target *t,
1547                                     struct ceph_pg_pool_info *pi)
1548 {
1549         bool pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
1550         bool pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
1551                        ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
1552                        __pool_full(pi);
1553
1554         WARN_ON(pi->id != t->target_oloc.pool);
1555         return ((t->flags & CEPH_OSD_FLAG_READ) && pauserd) ||
1556                ((t->flags & CEPH_OSD_FLAG_WRITE) && pausewr) ||
1557                (osdc->osdmap->epoch < osdc->epoch_barrier);
1558 }
1559
1560 static int pick_random_replica(const struct ceph_osds *acting)
1561 {
1562         int i = get_random_u32_below(acting->size);
1563
1564         dout("%s picked osd%d, primary osd%d\n", __func__,
1565              acting->osds[i], acting->primary);
1566         return i;
1567 }
1568
1569 /*
1570  * Picks the closest replica based on client's location given by
1571  * crush_location option.  Prefers the primary if the locality is
1572  * the same.
1573  */
1574 static int pick_closest_replica(struct ceph_osd_client *osdc,
1575                                 const struct ceph_osds *acting)
1576 {
1577         struct ceph_options *opt = osdc->client->options;
1578         int best_i, best_locality;
1579         int i = 0, locality;
1580
1581         do {
1582                 locality = ceph_get_crush_locality(osdc->osdmap,
1583                                                    acting->osds[i],
1584                                                    &opt->crush_locs);
1585                 if (i == 0 ||
1586                     (locality >= 0 && best_locality < 0) ||
1587                     (locality >= 0 && best_locality >= 0 &&
1588                      locality < best_locality)) {
1589                         best_i = i;
1590                         best_locality = locality;
1591                 }
1592         } while (++i < acting->size);
1593
1594         dout("%s picked osd%d with locality %d, primary osd%d\n", __func__,
1595              acting->osds[best_i], best_locality, acting->primary);
1596         return best_i;
1597 }
1598
1599 enum calc_target_result {
1600         CALC_TARGET_NO_ACTION = 0,
1601         CALC_TARGET_NEED_RESEND,
1602         CALC_TARGET_POOL_DNE,
1603 };
1604
1605 static enum calc_target_result calc_target(struct ceph_osd_client *osdc,
1606                                            struct ceph_osd_request_target *t,
1607                                            bool any_change)
1608 {
1609         struct ceph_pg_pool_info *pi;
1610         struct ceph_pg pgid, last_pgid;
1611         struct ceph_osds up, acting;
1612         bool is_read = t->flags & CEPH_OSD_FLAG_READ;
1613         bool is_write = t->flags & CEPH_OSD_FLAG_WRITE;
1614         bool force_resend = false;
1615         bool unpaused = false;
1616         bool legacy_change = false;
1617         bool split = false;
1618         bool sort_bitwise = ceph_osdmap_flag(osdc, CEPH_OSDMAP_SORTBITWISE);
1619         bool recovery_deletes = ceph_osdmap_flag(osdc,
1620                                                  CEPH_OSDMAP_RECOVERY_DELETES);
1621         enum calc_target_result ct_res;
1622
1623         t->epoch = osdc->osdmap->epoch;
1624         pi = ceph_pg_pool_by_id(osdc->osdmap, t->base_oloc.pool);
1625         if (!pi) {
1626                 t->osd = CEPH_HOMELESS_OSD;
1627                 ct_res = CALC_TARGET_POOL_DNE;
1628                 goto out;
1629         }
1630
1631         if (osdc->osdmap->epoch == pi->last_force_request_resend) {
1632                 if (t->last_force_resend < pi->last_force_request_resend) {
1633                         t->last_force_resend = pi->last_force_request_resend;
1634                         force_resend = true;
1635                 } else if (t->last_force_resend == 0) {
1636                         force_resend = true;
1637                 }
1638         }
1639
1640         /* apply tiering */
1641         ceph_oid_copy(&t->target_oid, &t->base_oid);
1642         ceph_oloc_copy(&t->target_oloc, &t->base_oloc);
1643         if ((t->flags & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0) {
1644                 if (is_read && pi->read_tier >= 0)
1645                         t->target_oloc.pool = pi->read_tier;
1646                 if (is_write && pi->write_tier >= 0)
1647                         t->target_oloc.pool = pi->write_tier;
1648
1649                 pi = ceph_pg_pool_by_id(osdc->osdmap, t->target_oloc.pool);
1650                 if (!pi) {
1651                         t->osd = CEPH_HOMELESS_OSD;
1652                         ct_res = CALC_TARGET_POOL_DNE;
1653                         goto out;
1654                 }
1655         }
1656
1657         __ceph_object_locator_to_pg(pi, &t->target_oid, &t->target_oloc, &pgid);
1658         last_pgid.pool = pgid.pool;
1659         last_pgid.seed = ceph_stable_mod(pgid.seed, t->pg_num, t->pg_num_mask);
1660
1661         ceph_pg_to_up_acting_osds(osdc->osdmap, pi, &pgid, &up, &acting);
1662         if (any_change &&
1663             ceph_is_new_interval(&t->acting,
1664                                  &acting,
1665                                  &t->up,
1666                                  &up,
1667                                  t->size,
1668                                  pi->size,
1669                                  t->min_size,
1670                                  pi->min_size,
1671                                  t->pg_num,
1672                                  pi->pg_num,
1673                                  t->sort_bitwise,
1674                                  sort_bitwise,
1675                                  t->recovery_deletes,
1676                                  recovery_deletes,
1677                                  &last_pgid))
1678                 force_resend = true;
1679
1680         if (t->paused && !target_should_be_paused(osdc, t, pi)) {
1681                 t->paused = false;
1682                 unpaused = true;
1683         }
1684         legacy_change = ceph_pg_compare(&t->pgid, &pgid) ||
1685                         ceph_osds_changed(&t->acting, &acting,
1686                                           t->used_replica || any_change);
1687         if (t->pg_num)
1688                 split = ceph_pg_is_split(&last_pgid, t->pg_num, pi->pg_num);
1689
1690         if (legacy_change || force_resend || split) {
1691                 t->pgid = pgid; /* struct */
1692                 ceph_pg_to_primary_shard(osdc->osdmap, pi, &pgid, &t->spgid);
1693                 ceph_osds_copy(&t->acting, &acting);
1694                 ceph_osds_copy(&t->up, &up);
1695                 t->size = pi->size;
1696                 t->min_size = pi->min_size;
1697                 t->pg_num = pi->pg_num;
1698                 t->pg_num_mask = pi->pg_num_mask;
1699                 t->sort_bitwise = sort_bitwise;
1700                 t->recovery_deletes = recovery_deletes;
1701
1702                 if ((t->flags & (CEPH_OSD_FLAG_BALANCE_READS |
1703                                  CEPH_OSD_FLAG_LOCALIZE_READS)) &&
1704                     !is_write && pi->type == CEPH_POOL_TYPE_REP &&
1705                     acting.size > 1) {
1706                         int pos;
1707
1708                         WARN_ON(!is_read || acting.osds[0] != acting.primary);
1709                         if (t->flags & CEPH_OSD_FLAG_BALANCE_READS) {
1710                                 pos = pick_random_replica(&acting);
1711                         } else {
1712                                 pos = pick_closest_replica(osdc, &acting);
1713                         }
1714                         t->osd = acting.osds[pos];
1715                         t->used_replica = pos > 0;
1716                 } else {
1717                         t->osd = acting.primary;
1718                         t->used_replica = false;
1719                 }
1720         }
1721
1722         if (unpaused || legacy_change || force_resend || split)
1723                 ct_res = CALC_TARGET_NEED_RESEND;
1724         else
1725                 ct_res = CALC_TARGET_NO_ACTION;
1726
1727 out:
1728         dout("%s t %p -> %d%d%d%d ct_res %d osd%d\n", __func__, t, unpaused,
1729              legacy_change, force_resend, split, ct_res, t->osd);
1730         return ct_res;
1731 }
1732
1733 static struct ceph_spg_mapping *alloc_spg_mapping(void)
1734 {
1735         struct ceph_spg_mapping *spg;
1736
1737         spg = kmalloc(sizeof(*spg), GFP_NOIO);
1738         if (!spg)
1739                 return NULL;
1740
1741         RB_CLEAR_NODE(&spg->node);
1742         spg->backoffs = RB_ROOT;
1743         return spg;
1744 }
1745
1746 static void free_spg_mapping(struct ceph_spg_mapping *spg)
1747 {
1748         WARN_ON(!RB_EMPTY_NODE(&spg->node));
1749         WARN_ON(!RB_EMPTY_ROOT(&spg->backoffs));
1750
1751         kfree(spg);
1752 }
1753
1754 /*
1755  * rbtree of ceph_spg_mapping for handling map<spg_t, ...>, similar to
1756  * ceph_pg_mapping.  Used to track OSD backoffs -- a backoff [range] is
1757  * defined only within a specific spgid; it does not pass anything to
1758  * children on split, or to another primary.
1759  */
1760 DEFINE_RB_FUNCS2(spg_mapping, struct ceph_spg_mapping, spgid, ceph_spg_compare,
1761                  RB_BYPTR, const struct ceph_spg *, node)
1762
1763 static u64 hoid_get_bitwise_key(const struct ceph_hobject_id *hoid)
1764 {
1765         return hoid->is_max ? 0x100000000ull : hoid->hash_reverse_bits;
1766 }
1767
1768 static void hoid_get_effective_key(const struct ceph_hobject_id *hoid,
1769                                    void **pkey, size_t *pkey_len)
1770 {
1771         if (hoid->key_len) {
1772                 *pkey = hoid->key;
1773                 *pkey_len = hoid->key_len;
1774         } else {
1775                 *pkey = hoid->oid;
1776                 *pkey_len = hoid->oid_len;
1777         }
1778 }
1779
1780 static int compare_names(const void *name1, size_t name1_len,
1781                          const void *name2, size_t name2_len)
1782 {
1783         int ret;
1784
1785         ret = memcmp(name1, name2, min(name1_len, name2_len));
1786         if (!ret) {
1787                 if (name1_len < name2_len)
1788                         ret = -1;
1789                 else if (name1_len > name2_len)
1790                         ret = 1;
1791         }
1792         return ret;
1793 }
1794
1795 static int hoid_compare(const struct ceph_hobject_id *lhs,
1796                         const struct ceph_hobject_id *rhs)
1797 {
1798         void *effective_key1, *effective_key2;
1799         size_t effective_key1_len, effective_key2_len;
1800         int ret;
1801
1802         if (lhs->is_max < rhs->is_max)
1803                 return -1;
1804         if (lhs->is_max > rhs->is_max)
1805                 return 1;
1806
1807         if (lhs->pool < rhs->pool)
1808                 return -1;
1809         if (lhs->pool > rhs->pool)
1810                 return 1;
1811
1812         if (hoid_get_bitwise_key(lhs) < hoid_get_bitwise_key(rhs))
1813                 return -1;
1814         if (hoid_get_bitwise_key(lhs) > hoid_get_bitwise_key(rhs))
1815                 return 1;
1816
1817         ret = compare_names(lhs->nspace, lhs->nspace_len,
1818                             rhs->nspace, rhs->nspace_len);
1819         if (ret)
1820                 return ret;
1821
1822         hoid_get_effective_key(lhs, &effective_key1, &effective_key1_len);
1823         hoid_get_effective_key(rhs, &effective_key2, &effective_key2_len);
1824         ret = compare_names(effective_key1, effective_key1_len,
1825                             effective_key2, effective_key2_len);
1826         if (ret)
1827                 return ret;
1828
1829         ret = compare_names(lhs->oid, lhs->oid_len, rhs->oid, rhs->oid_len);
1830         if (ret)
1831                 return ret;
1832
1833         if (lhs->snapid < rhs->snapid)
1834                 return -1;
1835         if (lhs->snapid > rhs->snapid)
1836                 return 1;
1837
1838         return 0;
1839 }
1840
1841 /*
1842  * For decoding ->begin and ->end of MOSDBackoff only -- no MIN/MAX
1843  * compat stuff here.
1844  *
1845  * Assumes @hoid is zero-initialized.
1846  */
1847 static int decode_hoid(void **p, void *end, struct ceph_hobject_id *hoid)
1848 {
1849         u8 struct_v;
1850         u32 struct_len;
1851         int ret;
1852
1853         ret = ceph_start_decoding(p, end, 4, "hobject_t", &struct_v,
1854                                   &struct_len);
1855         if (ret)
1856                 return ret;
1857
1858         if (struct_v < 4) {
1859                 pr_err("got struct_v %d < 4 of hobject_t\n", struct_v);
1860                 goto e_inval;
1861         }
1862
1863         hoid->key = ceph_extract_encoded_string(p, end, &hoid->key_len,
1864                                                 GFP_NOIO);
1865         if (IS_ERR(hoid->key)) {
1866                 ret = PTR_ERR(hoid->key);
1867                 hoid->key = NULL;
1868                 return ret;
1869         }
1870
1871         hoid->oid = ceph_extract_encoded_string(p, end, &hoid->oid_len,
1872                                                 GFP_NOIO);
1873         if (IS_ERR(hoid->oid)) {
1874                 ret = PTR_ERR(hoid->oid);
1875                 hoid->oid = NULL;
1876                 return ret;
1877         }
1878
1879         ceph_decode_64_safe(p, end, hoid->snapid, e_inval);
1880         ceph_decode_32_safe(p, end, hoid->hash, e_inval);
1881         ceph_decode_8_safe(p, end, hoid->is_max, e_inval);
1882
1883         hoid->nspace = ceph_extract_encoded_string(p, end, &hoid->nspace_len,
1884                                                    GFP_NOIO);
1885         if (IS_ERR(hoid->nspace)) {
1886                 ret = PTR_ERR(hoid->nspace);
1887                 hoid->nspace = NULL;
1888                 return ret;
1889         }
1890
1891         ceph_decode_64_safe(p, end, hoid->pool, e_inval);
1892
1893         ceph_hoid_build_hash_cache(hoid);
1894         return 0;
1895
1896 e_inval:
1897         return -EINVAL;
1898 }
1899
1900 static int hoid_encoding_size(const struct ceph_hobject_id *hoid)
1901 {
1902         return 8 + 4 + 1 + 8 + /* snapid, hash, is_max, pool */
1903                4 + hoid->key_len + 4 + hoid->oid_len + 4 + hoid->nspace_len;
1904 }
1905
1906 static void encode_hoid(void **p, void *end, const struct ceph_hobject_id *hoid)
1907 {
1908         ceph_start_encoding(p, 4, 3, hoid_encoding_size(hoid));
1909         ceph_encode_string(p, end, hoid->key, hoid->key_len);
1910         ceph_encode_string(p, end, hoid->oid, hoid->oid_len);
1911         ceph_encode_64(p, hoid->snapid);
1912         ceph_encode_32(p, hoid->hash);
1913         ceph_encode_8(p, hoid->is_max);
1914         ceph_encode_string(p, end, hoid->nspace, hoid->nspace_len);
1915         ceph_encode_64(p, hoid->pool);
1916 }
1917
1918 static void free_hoid(struct ceph_hobject_id *hoid)
1919 {
1920         if (hoid) {
1921                 kfree(hoid->key);
1922                 kfree(hoid->oid);
1923                 kfree(hoid->nspace);
1924                 kfree(hoid);
1925         }
1926 }
1927
1928 static struct ceph_osd_backoff *alloc_backoff(void)
1929 {
1930         struct ceph_osd_backoff *backoff;
1931
1932         backoff = kzalloc(sizeof(*backoff), GFP_NOIO);
1933         if (!backoff)
1934                 return NULL;
1935
1936         RB_CLEAR_NODE(&backoff->spg_node);
1937         RB_CLEAR_NODE(&backoff->id_node);
1938         return backoff;
1939 }
1940
1941 static void free_backoff(struct ceph_osd_backoff *backoff)
1942 {
1943         WARN_ON(!RB_EMPTY_NODE(&backoff->spg_node));
1944         WARN_ON(!RB_EMPTY_NODE(&backoff->id_node));
1945
1946         free_hoid(backoff->begin);
1947         free_hoid(backoff->end);
1948         kfree(backoff);
1949 }
1950
1951 /*
1952  * Within a specific spgid, backoffs are managed by ->begin hoid.
1953  */
1954 DEFINE_RB_INSDEL_FUNCS2(backoff, struct ceph_osd_backoff, begin, hoid_compare,
1955                         RB_BYVAL, spg_node);
1956
1957 static struct ceph_osd_backoff *lookup_containing_backoff(struct rb_root *root,
1958                                             const struct ceph_hobject_id *hoid)
1959 {
1960         struct rb_node *n = root->rb_node;
1961
1962         while (n) {
1963                 struct ceph_osd_backoff *cur =
1964                     rb_entry(n, struct ceph_osd_backoff, spg_node);
1965                 int cmp;
1966
1967                 cmp = hoid_compare(hoid, cur->begin);
1968                 if (cmp < 0) {
1969                         n = n->rb_left;
1970                 } else if (cmp > 0) {
1971                         if (hoid_compare(hoid, cur->end) < 0)
1972                                 return cur;
1973
1974                         n = n->rb_right;
1975                 } else {
1976                         return cur;
1977                 }
1978         }
1979
1980         return NULL;
1981 }
1982
1983 /*
1984  * Each backoff has a unique id within its OSD session.
1985  */
1986 DEFINE_RB_FUNCS(backoff_by_id, struct ceph_osd_backoff, id, id_node)
1987
1988 static void clear_backoffs(struct ceph_osd *osd)
1989 {
1990         while (!RB_EMPTY_ROOT(&osd->o_backoff_mappings)) {
1991                 struct ceph_spg_mapping *spg =
1992                     rb_entry(rb_first(&osd->o_backoff_mappings),
1993                              struct ceph_spg_mapping, node);
1994
1995                 while (!RB_EMPTY_ROOT(&spg->backoffs)) {
1996                         struct ceph_osd_backoff *backoff =
1997                             rb_entry(rb_first(&spg->backoffs),
1998                                      struct ceph_osd_backoff, spg_node);
1999
2000                         erase_backoff(&spg->backoffs, backoff);
2001                         erase_backoff_by_id(&osd->o_backoffs_by_id, backoff);
2002                         free_backoff(backoff);
2003                 }
2004                 erase_spg_mapping(&osd->o_backoff_mappings, spg);
2005                 free_spg_mapping(spg);
2006         }
2007 }
2008
2009 /*
2010  * Set up a temporary, non-owning view into @t.
2011  */
2012 static void hoid_fill_from_target(struct ceph_hobject_id *hoid,
2013                                   const struct ceph_osd_request_target *t)
2014 {
2015         hoid->key = NULL;
2016         hoid->key_len = 0;
2017         hoid->oid = t->target_oid.name;
2018         hoid->oid_len = t->target_oid.name_len;
2019         hoid->snapid = CEPH_NOSNAP;
2020         hoid->hash = t->pgid.seed;
2021         hoid->is_max = false;
2022         if (t->target_oloc.pool_ns) {
2023                 hoid->nspace = t->target_oloc.pool_ns->str;
2024                 hoid->nspace_len = t->target_oloc.pool_ns->len;
2025         } else {
2026                 hoid->nspace = NULL;
2027                 hoid->nspace_len = 0;
2028         }
2029         hoid->pool = t->target_oloc.pool;
2030         ceph_hoid_build_hash_cache(hoid);
2031 }
2032
2033 static bool should_plug_request(struct ceph_osd_request *req)
2034 {
2035         struct ceph_osd *osd = req->r_osd;
2036         struct ceph_spg_mapping *spg;
2037         struct ceph_osd_backoff *backoff;
2038         struct ceph_hobject_id hoid;
2039
2040         spg = lookup_spg_mapping(&osd->o_backoff_mappings, &req->r_t.spgid);
2041         if (!spg)
2042                 return false;
2043
2044         hoid_fill_from_target(&hoid, &req->r_t);
2045         backoff = lookup_containing_backoff(&spg->backoffs, &hoid);
2046         if (!backoff)
2047                 return false;
2048
2049         dout("%s req %p tid %llu backoff osd%d spgid %llu.%xs%d id %llu\n",
2050              __func__, req, req->r_tid, osd->o_osd, backoff->spgid.pgid.pool,
2051              backoff->spgid.pgid.seed, backoff->spgid.shard, backoff->id);
2052         return true;
2053 }
2054
2055 /*
2056  * Keep get_num_data_items() in sync with this function.
2057  */
2058 static void setup_request_data(struct ceph_osd_request *req)
2059 {
2060         struct ceph_msg *request_msg = req->r_request;
2061         struct ceph_msg *reply_msg = req->r_reply;
2062         struct ceph_osd_req_op *op;
2063
2064         if (req->r_request->num_data_items || req->r_reply->num_data_items)
2065                 return;
2066
2067         WARN_ON(request_msg->data_length || reply_msg->data_length);
2068         for (op = req->r_ops; op != &req->r_ops[req->r_num_ops]; op++) {
2069                 switch (op->op) {
2070                 /* request */
2071                 case CEPH_OSD_OP_WRITE:
2072                 case CEPH_OSD_OP_WRITEFULL:
2073                         WARN_ON(op->indata_len != op->extent.length);
2074                         ceph_osdc_msg_data_add(request_msg,
2075                                                &op->extent.osd_data);
2076                         break;
2077                 case CEPH_OSD_OP_SETXATTR:
2078                 case CEPH_OSD_OP_CMPXATTR:
2079                         WARN_ON(op->indata_len != op->xattr.name_len +
2080                                                   op->xattr.value_len);
2081                         ceph_osdc_msg_data_add(request_msg,
2082                                                &op->xattr.osd_data);
2083                         break;
2084                 case CEPH_OSD_OP_NOTIFY_ACK:
2085                         ceph_osdc_msg_data_add(request_msg,
2086                                                &op->notify_ack.request_data);
2087                         break;
2088                 case CEPH_OSD_OP_COPY_FROM2:
2089                         ceph_osdc_msg_data_add(request_msg,
2090                                                &op->copy_from.osd_data);
2091                         break;
2092
2093                 /* reply */
2094                 case CEPH_OSD_OP_STAT:
2095                         ceph_osdc_msg_data_add(reply_msg,
2096                                                &op->raw_data_in);
2097                         break;
2098                 case CEPH_OSD_OP_READ:
2099                 case CEPH_OSD_OP_SPARSE_READ:
2100                         ceph_osdc_msg_data_add(reply_msg,
2101                                                &op->extent.osd_data);
2102                         break;
2103                 case CEPH_OSD_OP_LIST_WATCHERS:
2104                         ceph_osdc_msg_data_add(reply_msg,
2105                                                &op->list_watchers.response_data);
2106                         break;
2107
2108                 /* both */
2109                 case CEPH_OSD_OP_CALL:
2110                         WARN_ON(op->indata_len != op->cls.class_len +
2111                                                   op->cls.method_len +
2112                                                   op->cls.indata_len);
2113                         ceph_osdc_msg_data_add(request_msg,
2114                                                &op->cls.request_info);
2115                         /* optional, can be NONE */
2116                         ceph_osdc_msg_data_add(request_msg,
2117                                                &op->cls.request_data);
2118                         /* optional, can be NONE */
2119                         ceph_osdc_msg_data_add(reply_msg,
2120                                                &op->cls.response_data);
2121                         break;
2122                 case CEPH_OSD_OP_NOTIFY:
2123                         ceph_osdc_msg_data_add(request_msg,
2124                                                &op->notify.request_data);
2125                         ceph_osdc_msg_data_add(reply_msg,
2126                                                &op->notify.response_data);
2127                         break;
2128                 }
2129         }
2130 }
2131
2132 static void encode_pgid(void **p, const struct ceph_pg *pgid)
2133 {
2134         ceph_encode_8(p, 1);
2135         ceph_encode_64(p, pgid->pool);
2136         ceph_encode_32(p, pgid->seed);
2137         ceph_encode_32(p, -1); /* preferred */
2138 }
2139
2140 static void encode_spgid(void **p, const struct ceph_spg *spgid)
2141 {
2142         ceph_start_encoding(p, 1, 1, CEPH_PGID_ENCODING_LEN + 1);
2143         encode_pgid(p, &spgid->pgid);
2144         ceph_encode_8(p, spgid->shard);
2145 }
2146
2147 static void encode_oloc(void **p, void *end,
2148                         const struct ceph_object_locator *oloc)
2149 {
2150         ceph_start_encoding(p, 5, 4, ceph_oloc_encoding_size(oloc));
2151         ceph_encode_64(p, oloc->pool);
2152         ceph_encode_32(p, -1); /* preferred */
2153         ceph_encode_32(p, 0);  /* key len */
2154         if (oloc->pool_ns)
2155                 ceph_encode_string(p, end, oloc->pool_ns->str,
2156                                    oloc->pool_ns->len);
2157         else
2158                 ceph_encode_32(p, 0);
2159 }
2160
2161 static void encode_request_partial(struct ceph_osd_request *req,
2162                                    struct ceph_msg *msg)
2163 {
2164         void *p = msg->front.iov_base;
2165         void *const end = p + msg->front_alloc_len;
2166         u32 data_len = 0;
2167         int i;
2168
2169         if (req->r_flags & CEPH_OSD_FLAG_WRITE) {
2170                 /* snapshots aren't writeable */
2171                 WARN_ON(req->r_snapid != CEPH_NOSNAP);
2172         } else {
2173                 WARN_ON(req->r_mtime.tv_sec || req->r_mtime.tv_nsec ||
2174                         req->r_data_offset || req->r_snapc);
2175         }
2176
2177         setup_request_data(req);
2178
2179         encode_spgid(&p, &req->r_t.spgid); /* actual spg */
2180         ceph_encode_32(&p, req->r_t.pgid.seed); /* raw hash */
2181         ceph_encode_32(&p, req->r_osdc->osdmap->epoch);
2182         ceph_encode_32(&p, req->r_flags);
2183
2184         /* reqid */
2185         ceph_start_encoding(&p, 2, 2, sizeof(struct ceph_osd_reqid));
2186         memset(p, 0, sizeof(struct ceph_osd_reqid));
2187         p += sizeof(struct ceph_osd_reqid);
2188
2189         /* trace */
2190         memset(p, 0, sizeof(struct ceph_blkin_trace_info));
2191         p += sizeof(struct ceph_blkin_trace_info);
2192
2193         ceph_encode_32(&p, 0); /* client_inc, always 0 */
2194         ceph_encode_timespec64(p, &req->r_mtime);
2195         p += sizeof(struct ceph_timespec);
2196
2197         encode_oloc(&p, end, &req->r_t.target_oloc);
2198         ceph_encode_string(&p, end, req->r_t.target_oid.name,
2199                            req->r_t.target_oid.name_len);
2200
2201         /* ops, can imply data */
2202         ceph_encode_16(&p, req->r_num_ops);
2203         for (i = 0; i < req->r_num_ops; i++) {
2204                 data_len += osd_req_encode_op(p, &req->r_ops[i]);
2205                 p += sizeof(struct ceph_osd_op);
2206         }
2207
2208         ceph_encode_64(&p, req->r_snapid); /* snapid */
2209         if (req->r_snapc) {
2210                 ceph_encode_64(&p, req->r_snapc->seq);
2211                 ceph_encode_32(&p, req->r_snapc->num_snaps);
2212                 for (i = 0; i < req->r_snapc->num_snaps; i++)
2213                         ceph_encode_64(&p, req->r_snapc->snaps[i]);
2214         } else {
2215                 ceph_encode_64(&p, 0); /* snap_seq */
2216                 ceph_encode_32(&p, 0); /* snaps len */
2217         }
2218
2219         ceph_encode_32(&p, req->r_attempts); /* retry_attempt */
2220         BUG_ON(p > end - 8); /* space for features */
2221
2222         msg->hdr.version = cpu_to_le16(8); /* MOSDOp v8 */
2223         /* front_len is finalized in encode_request_finish() */
2224         msg->front.iov_len = p - msg->front.iov_base;
2225         msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2226         msg->hdr.data_len = cpu_to_le32(data_len);
2227         /*
2228          * The header "data_off" is a hint to the receiver allowing it
2229          * to align received data into its buffers such that there's no
2230          * need to re-copy it before writing it to disk (direct I/O).
2231          */
2232         msg->hdr.data_off = cpu_to_le16(req->r_data_offset);
2233
2234         dout("%s req %p msg %p oid %s oid_len %d\n", __func__, req, msg,
2235              req->r_t.target_oid.name, req->r_t.target_oid.name_len);
2236 }
2237
2238 static void encode_request_finish(struct ceph_msg *msg)
2239 {
2240         void *p = msg->front.iov_base;
2241         void *const partial_end = p + msg->front.iov_len;
2242         void *const end = p + msg->front_alloc_len;
2243
2244         if (CEPH_HAVE_FEATURE(msg->con->peer_features, RESEND_ON_SPLIT)) {
2245                 /* luminous OSD -- encode features and be done */
2246                 p = partial_end;
2247                 ceph_encode_64(&p, msg->con->peer_features);
2248         } else {
2249                 struct {
2250                         char spgid[CEPH_ENCODING_START_BLK_LEN +
2251                                    CEPH_PGID_ENCODING_LEN + 1];
2252                         __le32 hash;
2253                         __le32 epoch;
2254                         __le32 flags;
2255                         char reqid[CEPH_ENCODING_START_BLK_LEN +
2256                                    sizeof(struct ceph_osd_reqid)];
2257                         char trace[sizeof(struct ceph_blkin_trace_info)];
2258                         __le32 client_inc;
2259                         struct ceph_timespec mtime;
2260                 } __packed head;
2261                 struct ceph_pg pgid;
2262                 void *oloc, *oid, *tail;
2263                 int oloc_len, oid_len, tail_len;
2264                 int len;
2265
2266                 /*
2267                  * Pre-luminous OSD -- reencode v8 into v4 using @head
2268                  * as a temporary buffer.  Encode the raw PG; the rest
2269                  * is just a matter of moving oloc, oid and tail blobs
2270                  * around.
2271                  */
2272                 memcpy(&head, p, sizeof(head));
2273                 p += sizeof(head);
2274
2275                 oloc = p;
2276                 p += CEPH_ENCODING_START_BLK_LEN;
2277                 pgid.pool = ceph_decode_64(&p);
2278                 p += 4 + 4; /* preferred, key len */
2279                 len = ceph_decode_32(&p);
2280                 p += len;   /* nspace */
2281                 oloc_len = p - oloc;
2282
2283                 oid = p;
2284                 len = ceph_decode_32(&p);
2285                 p += len;
2286                 oid_len = p - oid;
2287
2288                 tail = p;
2289                 tail_len = partial_end - p;
2290
2291                 p = msg->front.iov_base;
2292                 ceph_encode_copy(&p, &head.client_inc, sizeof(head.client_inc));
2293                 ceph_encode_copy(&p, &head.epoch, sizeof(head.epoch));
2294                 ceph_encode_copy(&p, &head.flags, sizeof(head.flags));
2295                 ceph_encode_copy(&p, &head.mtime, sizeof(head.mtime));
2296
2297                 /* reassert_version */
2298                 memset(p, 0, sizeof(struct ceph_eversion));
2299                 p += sizeof(struct ceph_eversion);
2300
2301                 BUG_ON(p >= oloc);
2302                 memmove(p, oloc, oloc_len);
2303                 p += oloc_len;
2304
2305                 pgid.seed = le32_to_cpu(head.hash);
2306                 encode_pgid(&p, &pgid); /* raw pg */
2307
2308                 BUG_ON(p >= oid);
2309                 memmove(p, oid, oid_len);
2310                 p += oid_len;
2311
2312                 /* tail -- ops, snapid, snapc, retry_attempt */
2313                 BUG_ON(p >= tail);
2314                 memmove(p, tail, tail_len);
2315                 p += tail_len;
2316
2317                 msg->hdr.version = cpu_to_le16(4); /* MOSDOp v4 */
2318         }
2319
2320         BUG_ON(p > end);
2321         msg->front.iov_len = p - msg->front.iov_base;
2322         msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2323
2324         dout("%s msg %p tid %llu %u+%u+%u v%d\n", __func__, msg,
2325              le64_to_cpu(msg->hdr.tid), le32_to_cpu(msg->hdr.front_len),
2326              le32_to_cpu(msg->hdr.middle_len), le32_to_cpu(msg->hdr.data_len),
2327              le16_to_cpu(msg->hdr.version));
2328 }
2329
2330 /*
2331  * @req has to be assigned a tid and registered.
2332  */
2333 static void send_request(struct ceph_osd_request *req)
2334 {
2335         struct ceph_osd *osd = req->r_osd;
2336
2337         verify_osd_locked(osd);
2338         WARN_ON(osd->o_osd != req->r_t.osd);
2339
2340         /* backoff? */
2341         if (should_plug_request(req))
2342                 return;
2343
2344         /*
2345          * We may have a previously queued request message hanging
2346          * around.  Cancel it to avoid corrupting the msgr.
2347          */
2348         if (req->r_sent)
2349                 ceph_msg_revoke(req->r_request);
2350
2351         req->r_flags |= CEPH_OSD_FLAG_KNOWN_REDIR;
2352         if (req->r_attempts)
2353                 req->r_flags |= CEPH_OSD_FLAG_RETRY;
2354         else
2355                 WARN_ON(req->r_flags & CEPH_OSD_FLAG_RETRY);
2356
2357         encode_request_partial(req, req->r_request);
2358
2359         dout("%s req %p tid %llu to pgid %llu.%x spgid %llu.%xs%d osd%d e%u flags 0x%x attempt %d\n",
2360              __func__, req, req->r_tid, req->r_t.pgid.pool, req->r_t.pgid.seed,
2361              req->r_t.spgid.pgid.pool, req->r_t.spgid.pgid.seed,
2362              req->r_t.spgid.shard, osd->o_osd, req->r_t.epoch, req->r_flags,
2363              req->r_attempts);
2364
2365         req->r_t.paused = false;
2366         req->r_stamp = jiffies;
2367         req->r_attempts++;
2368
2369         req->r_sent = osd->o_incarnation;
2370         req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
2371         ceph_con_send(&osd->o_con, ceph_msg_get(req->r_request));
2372 }
2373
2374 static void maybe_request_map(struct ceph_osd_client *osdc)
2375 {
2376         bool continuous = false;
2377
2378         verify_osdc_locked(osdc);
2379         WARN_ON(!osdc->osdmap->epoch);
2380
2381         if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
2382             ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD) ||
2383             ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) {
2384                 dout("%s osdc %p continuous\n", __func__, osdc);
2385                 continuous = true;
2386         } else {
2387                 dout("%s osdc %p onetime\n", __func__, osdc);
2388         }
2389
2390         if (ceph_monc_want_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
2391                                osdc->osdmap->epoch + 1, continuous))
2392                 ceph_monc_renew_subs(&osdc->client->monc);
2393 }
2394
2395 static void complete_request(struct ceph_osd_request *req, int err);
2396 static void send_map_check(struct ceph_osd_request *req);
2397
2398 static void __submit_request(struct ceph_osd_request *req, bool wrlocked)
2399 {
2400         struct ceph_osd_client *osdc = req->r_osdc;
2401         struct ceph_osd *osd;
2402         enum calc_target_result ct_res;
2403         int err = 0;
2404         bool need_send = false;
2405         bool promoted = false;
2406
2407         WARN_ON(req->r_tid);
2408         dout("%s req %p wrlocked %d\n", __func__, req, wrlocked);
2409
2410 again:
2411         ct_res = calc_target(osdc, &req->r_t, false);
2412         if (ct_res == CALC_TARGET_POOL_DNE && !wrlocked)
2413                 goto promote;
2414
2415         osd = lookup_create_osd(osdc, req->r_t.osd, wrlocked);
2416         if (IS_ERR(osd)) {
2417                 WARN_ON(PTR_ERR(osd) != -EAGAIN || wrlocked);
2418                 goto promote;
2419         }
2420
2421         if (osdc->abort_err) {
2422                 dout("req %p abort_err %d\n", req, osdc->abort_err);
2423                 err = osdc->abort_err;
2424         } else if (osdc->osdmap->epoch < osdc->epoch_barrier) {
2425                 dout("req %p epoch %u barrier %u\n", req, osdc->osdmap->epoch,
2426                      osdc->epoch_barrier);
2427                 req->r_t.paused = true;
2428                 maybe_request_map(osdc);
2429         } else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
2430                    ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) {
2431                 dout("req %p pausewr\n", req);
2432                 req->r_t.paused = true;
2433                 maybe_request_map(osdc);
2434         } else if ((req->r_flags & CEPH_OSD_FLAG_READ) &&
2435                    ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) {
2436                 dout("req %p pauserd\n", req);
2437                 req->r_t.paused = true;
2438                 maybe_request_map(osdc);
2439         } else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
2440                    !(req->r_flags & (CEPH_OSD_FLAG_FULL_TRY |
2441                                      CEPH_OSD_FLAG_FULL_FORCE)) &&
2442                    (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
2443                     pool_full(osdc, req->r_t.base_oloc.pool))) {
2444                 dout("req %p full/pool_full\n", req);
2445                 if (ceph_test_opt(osdc->client, ABORT_ON_FULL)) {
2446                         err = -ENOSPC;
2447                 } else {
2448                         if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL))
2449                                 pr_warn_ratelimited("cluster is full (osdmap FULL)\n");
2450                         else
2451                                 pr_warn_ratelimited("pool %lld is full or reached quota\n",
2452                                                     req->r_t.base_oloc.pool);
2453                         req->r_t.paused = true;
2454                         maybe_request_map(osdc);
2455                 }
2456         } else if (!osd_homeless(osd)) {
2457                 need_send = true;
2458         } else {
2459                 maybe_request_map(osdc);
2460         }
2461
2462         mutex_lock(&osd->lock);
2463         /*
2464          * Assign the tid atomically with send_request() to protect
2465          * multiple writes to the same object from racing with each
2466          * other, resulting in out of order ops on the OSDs.
2467          */
2468         req->r_tid = atomic64_inc_return(&osdc->last_tid);
2469         link_request(osd, req);
2470         if (need_send)
2471                 send_request(req);
2472         else if (err)
2473                 complete_request(req, err);
2474         mutex_unlock(&osd->lock);
2475
2476         if (!err && ct_res == CALC_TARGET_POOL_DNE)
2477                 send_map_check(req);
2478
2479         if (promoted)
2480                 downgrade_write(&osdc->lock);
2481         return;
2482
2483 promote:
2484         up_read(&osdc->lock);
2485         down_write(&osdc->lock);
2486         wrlocked = true;
2487         promoted = true;
2488         goto again;
2489 }
2490
2491 static void account_request(struct ceph_osd_request *req)
2492 {
2493         WARN_ON(req->r_flags & (CEPH_OSD_FLAG_ACK | CEPH_OSD_FLAG_ONDISK));
2494         WARN_ON(!(req->r_flags & (CEPH_OSD_FLAG_READ | CEPH_OSD_FLAG_WRITE)));
2495
2496         req->r_flags |= CEPH_OSD_FLAG_ONDISK;
2497         atomic_inc(&req->r_osdc->num_requests);
2498
2499         req->r_start_stamp = jiffies;
2500         req->r_start_latency = ktime_get();
2501 }
2502
2503 static void submit_request(struct ceph_osd_request *req, bool wrlocked)
2504 {
2505         ceph_osdc_get_request(req);
2506         account_request(req);
2507         __submit_request(req, wrlocked);
2508 }
2509
2510 static void finish_request(struct ceph_osd_request *req)
2511 {
2512         struct ceph_osd_client *osdc = req->r_osdc;
2513
2514         WARN_ON(lookup_request_mc(&osdc->map_checks, req->r_tid));
2515         dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
2516
2517         req->r_end_latency = ktime_get();
2518
2519         if (req->r_osd) {
2520                 ceph_init_sparse_read(&req->r_osd->o_sparse_read);
2521                 unlink_request(req->r_osd, req);
2522         }
2523         atomic_dec(&osdc->num_requests);
2524
2525         /*
2526          * If an OSD has failed or returned and a request has been sent
2527          * twice, it's possible to get a reply and end up here while the
2528          * request message is queued for delivery.  We will ignore the
2529          * reply, so not a big deal, but better to try and catch it.
2530          */
2531         ceph_msg_revoke(req->r_request);
2532         ceph_msg_revoke_incoming(req->r_reply);
2533 }
2534
2535 static void __complete_request(struct ceph_osd_request *req)
2536 {
2537         dout("%s req %p tid %llu cb %ps result %d\n", __func__, req,
2538              req->r_tid, req->r_callback, req->r_result);
2539
2540         if (req->r_callback)
2541                 req->r_callback(req);
2542         complete_all(&req->r_completion);
2543         ceph_osdc_put_request(req);
2544 }
2545
2546 static void complete_request_workfn(struct work_struct *work)
2547 {
2548         struct ceph_osd_request *req =
2549             container_of(work, struct ceph_osd_request, r_complete_work);
2550
2551         __complete_request(req);
2552 }
2553
2554 /*
2555  * This is open-coded in handle_reply().
2556  */
2557 static void complete_request(struct ceph_osd_request *req, int err)
2558 {
2559         dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err);
2560
2561         req->r_result = err;
2562         finish_request(req);
2563
2564         INIT_WORK(&req->r_complete_work, complete_request_workfn);
2565         queue_work(req->r_osdc->completion_wq, &req->r_complete_work);
2566 }
2567
2568 static void cancel_map_check(struct ceph_osd_request *req)
2569 {
2570         struct ceph_osd_client *osdc = req->r_osdc;
2571         struct ceph_osd_request *lookup_req;
2572
2573         verify_osdc_wrlocked(osdc);
2574
2575         lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid);
2576         if (!lookup_req)
2577                 return;
2578
2579         WARN_ON(lookup_req != req);
2580         erase_request_mc(&osdc->map_checks, req);
2581         ceph_osdc_put_request(req);
2582 }
2583
2584 static void cancel_request(struct ceph_osd_request *req)
2585 {
2586         dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
2587
2588         cancel_map_check(req);
2589         finish_request(req);
2590         complete_all(&req->r_completion);
2591         ceph_osdc_put_request(req);
2592 }
2593
2594 static void abort_request(struct ceph_osd_request *req, int err)
2595 {
2596         dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err);
2597
2598         cancel_map_check(req);
2599         complete_request(req, err);
2600 }
2601
2602 static int abort_fn(struct ceph_osd_request *req, void *arg)
2603 {
2604         int err = *(int *)arg;
2605
2606         abort_request(req, err);
2607         return 0; /* continue iteration */
2608 }
2609
2610 /*
2611  * Abort all in-flight requests with @err and arrange for all future
2612  * requests to be failed immediately.
2613  */
2614 void ceph_osdc_abort_requests(struct ceph_osd_client *osdc, int err)
2615 {
2616         dout("%s osdc %p err %d\n", __func__, osdc, err);
2617         down_write(&osdc->lock);
2618         for_each_request(osdc, abort_fn, &err);
2619         osdc->abort_err = err;
2620         up_write(&osdc->lock);
2621 }
2622 EXPORT_SYMBOL(ceph_osdc_abort_requests);
2623
2624 void ceph_osdc_clear_abort_err(struct ceph_osd_client *osdc)
2625 {
2626         down_write(&osdc->lock);
2627         osdc->abort_err = 0;
2628         up_write(&osdc->lock);
2629 }
2630 EXPORT_SYMBOL(ceph_osdc_clear_abort_err);
2631
2632 static void update_epoch_barrier(struct ceph_osd_client *osdc, u32 eb)
2633 {
2634         if (likely(eb > osdc->epoch_barrier)) {
2635                 dout("updating epoch_barrier from %u to %u\n",
2636                                 osdc->epoch_barrier, eb);
2637                 osdc->epoch_barrier = eb;
2638                 /* Request map if we're not to the barrier yet */
2639                 if (eb > osdc->osdmap->epoch)
2640                         maybe_request_map(osdc);
2641         }
2642 }
2643
2644 void ceph_osdc_update_epoch_barrier(struct ceph_osd_client *osdc, u32 eb)
2645 {
2646         down_read(&osdc->lock);
2647         if (unlikely(eb > osdc->epoch_barrier)) {
2648                 up_read(&osdc->lock);
2649                 down_write(&osdc->lock);
2650                 update_epoch_barrier(osdc, eb);
2651                 up_write(&osdc->lock);
2652         } else {
2653                 up_read(&osdc->lock);
2654         }
2655 }
2656 EXPORT_SYMBOL(ceph_osdc_update_epoch_barrier);
2657
2658 /*
2659  * We can end up releasing caps as a result of abort_request().
2660  * In that case, we probably want to ensure that the cap release message
2661  * has an updated epoch barrier in it, so set the epoch barrier prior to
2662  * aborting the first request.
2663  */
2664 static int abort_on_full_fn(struct ceph_osd_request *req, void *arg)
2665 {
2666         struct ceph_osd_client *osdc = req->r_osdc;
2667         bool *victims = arg;
2668
2669         if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
2670             (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
2671              pool_full(osdc, req->r_t.base_oloc.pool))) {
2672                 if (!*victims) {
2673                         update_epoch_barrier(osdc, osdc->osdmap->epoch);
2674                         *victims = true;
2675                 }
2676                 abort_request(req, -ENOSPC);
2677         }
2678
2679         return 0; /* continue iteration */
2680 }
2681
2682 /*
2683  * Drop all pending requests that are stalled waiting on a full condition to
2684  * clear, and complete them with ENOSPC as the return code. Set the
2685  * osdc->epoch_barrier to the latest map epoch that we've seen if any were
2686  * cancelled.
2687  */
2688 static void ceph_osdc_abort_on_full(struct ceph_osd_client *osdc)
2689 {
2690         bool victims = false;
2691
2692         if (ceph_test_opt(osdc->client, ABORT_ON_FULL) &&
2693             (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || have_pool_full(osdc)))
2694                 for_each_request(osdc, abort_on_full_fn, &victims);
2695 }
2696
2697 static void check_pool_dne(struct ceph_osd_request *req)
2698 {
2699         struct ceph_osd_client *osdc = req->r_osdc;
2700         struct ceph_osdmap *map = osdc->osdmap;
2701
2702         verify_osdc_wrlocked(osdc);
2703         WARN_ON(!map->epoch);
2704
2705         if (req->r_attempts) {
2706                 /*
2707                  * We sent a request earlier, which means that
2708                  * previously the pool existed, and now it does not
2709                  * (i.e., it was deleted).
2710                  */
2711                 req->r_map_dne_bound = map->epoch;
2712                 dout("%s req %p tid %llu pool disappeared\n", __func__, req,
2713                      req->r_tid);
2714         } else {
2715                 dout("%s req %p tid %llu map_dne_bound %u have %u\n", __func__,
2716                      req, req->r_tid, req->r_map_dne_bound, map->epoch);
2717         }
2718
2719         if (req->r_map_dne_bound) {
2720                 if (map->epoch >= req->r_map_dne_bound) {
2721                         /* we had a new enough map */
2722                         pr_info_ratelimited("tid %llu pool does not exist\n",
2723                                             req->r_tid);
2724                         complete_request(req, -ENOENT);
2725                 }
2726         } else {
2727                 send_map_check(req);
2728         }
2729 }
2730
2731 static void map_check_cb(struct ceph_mon_generic_request *greq)
2732 {
2733         struct ceph_osd_client *osdc = &greq->monc->client->osdc;
2734         struct ceph_osd_request *req;
2735         u64 tid = greq->private_data;
2736
2737         WARN_ON(greq->result || !greq->u.newest);
2738
2739         down_write(&osdc->lock);
2740         req = lookup_request_mc(&osdc->map_checks, tid);
2741         if (!req) {
2742                 dout("%s tid %llu dne\n", __func__, tid);
2743                 goto out_unlock;
2744         }
2745
2746         dout("%s req %p tid %llu map_dne_bound %u newest %llu\n", __func__,
2747              req, req->r_tid, req->r_map_dne_bound, greq->u.newest);
2748         if (!req->r_map_dne_bound)
2749                 req->r_map_dne_bound = greq->u.newest;
2750         erase_request_mc(&osdc->map_checks, req);
2751         check_pool_dne(req);
2752
2753         ceph_osdc_put_request(req);
2754 out_unlock:
2755         up_write(&osdc->lock);
2756 }
2757
2758 static void send_map_check(struct ceph_osd_request *req)
2759 {
2760         struct ceph_osd_client *osdc = req->r_osdc;
2761         struct ceph_osd_request *lookup_req;
2762         int ret;
2763
2764         verify_osdc_wrlocked(osdc);
2765
2766         lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid);
2767         if (lookup_req) {
2768                 WARN_ON(lookup_req != req);
2769                 return;
2770         }
2771
2772         ceph_osdc_get_request(req);
2773         insert_request_mc(&osdc->map_checks, req);
2774         ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap",
2775                                           map_check_cb, req->r_tid);
2776         WARN_ON(ret);
2777 }
2778
2779 /*
2780  * lingering requests, watch/notify v2 infrastructure
2781  */
2782 static void linger_release(struct kref *kref)
2783 {
2784         struct ceph_osd_linger_request *lreq =
2785             container_of(kref, struct ceph_osd_linger_request, kref);
2786
2787         dout("%s lreq %p reg_req %p ping_req %p\n", __func__, lreq,
2788              lreq->reg_req, lreq->ping_req);
2789         WARN_ON(!RB_EMPTY_NODE(&lreq->node));
2790         WARN_ON(!RB_EMPTY_NODE(&lreq->osdc_node));
2791         WARN_ON(!RB_EMPTY_NODE(&lreq->mc_node));
2792         WARN_ON(!list_empty(&lreq->scan_item));
2793         WARN_ON(!list_empty(&lreq->pending_lworks));
2794         WARN_ON(lreq->osd);
2795
2796         if (lreq->request_pl)
2797                 ceph_pagelist_release(lreq->request_pl);
2798         if (lreq->notify_id_pages)
2799                 ceph_release_page_vector(lreq->notify_id_pages, 1);
2800
2801         ceph_osdc_put_request(lreq->reg_req);
2802         ceph_osdc_put_request(lreq->ping_req);
2803         target_destroy(&lreq->t);
2804         kfree(lreq);
2805 }
2806
2807 static void linger_put(struct ceph_osd_linger_request *lreq)
2808 {
2809         if (lreq)
2810                 kref_put(&lreq->kref, linger_release);
2811 }
2812
2813 static struct ceph_osd_linger_request *
2814 linger_get(struct ceph_osd_linger_request *lreq)
2815 {
2816         kref_get(&lreq->kref);
2817         return lreq;
2818 }
2819
2820 static struct ceph_osd_linger_request *
2821 linger_alloc(struct ceph_osd_client *osdc)
2822 {
2823         struct ceph_osd_linger_request *lreq;
2824
2825         lreq = kzalloc(sizeof(*lreq), GFP_NOIO);
2826         if (!lreq)
2827                 return NULL;
2828
2829         kref_init(&lreq->kref);
2830         mutex_init(&lreq->lock);
2831         RB_CLEAR_NODE(&lreq->node);
2832         RB_CLEAR_NODE(&lreq->osdc_node);
2833         RB_CLEAR_NODE(&lreq->mc_node);
2834         INIT_LIST_HEAD(&lreq->scan_item);
2835         INIT_LIST_HEAD(&lreq->pending_lworks);
2836         init_completion(&lreq->reg_commit_wait);
2837         init_completion(&lreq->notify_finish_wait);
2838
2839         lreq->osdc = osdc;
2840         target_init(&lreq->t);
2841
2842         dout("%s lreq %p\n", __func__, lreq);
2843         return lreq;
2844 }
2845
2846 DEFINE_RB_INSDEL_FUNCS(linger, struct ceph_osd_linger_request, linger_id, node)
2847 DEFINE_RB_FUNCS(linger_osdc, struct ceph_osd_linger_request, linger_id, osdc_node)
2848 DEFINE_RB_FUNCS(linger_mc, struct ceph_osd_linger_request, linger_id, mc_node)
2849
2850 /*
2851  * Create linger request <-> OSD session relation.
2852  *
2853  * @lreq has to be registered, @osd may be homeless.
2854  */
2855 static void link_linger(struct ceph_osd *osd,
2856                         struct ceph_osd_linger_request *lreq)
2857 {
2858         verify_osd_locked(osd);
2859         WARN_ON(!lreq->linger_id || lreq->osd);
2860         dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd,
2861              osd->o_osd, lreq, lreq->linger_id);
2862
2863         if (!osd_homeless(osd))
2864                 __remove_osd_from_lru(osd);
2865         else
2866                 atomic_inc(&osd->o_osdc->num_homeless);
2867
2868         get_osd(osd);
2869         insert_linger(&osd->o_linger_requests, lreq);
2870         lreq->osd = osd;
2871 }
2872
2873 static void unlink_linger(struct ceph_osd *osd,
2874                           struct ceph_osd_linger_request *lreq)
2875 {
2876         verify_osd_locked(osd);
2877         WARN_ON(lreq->osd != osd);
2878         dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd,
2879              osd->o_osd, lreq, lreq->linger_id);
2880
2881         lreq->osd = NULL;
2882         erase_linger(&osd->o_linger_requests, lreq);
2883         put_osd(osd);
2884
2885         if (!osd_homeless(osd))
2886                 maybe_move_osd_to_lru(osd);
2887         else
2888                 atomic_dec(&osd->o_osdc->num_homeless);
2889 }
2890
2891 static bool __linger_registered(struct ceph_osd_linger_request *lreq)
2892 {
2893         verify_osdc_locked(lreq->osdc);
2894
2895         return !RB_EMPTY_NODE(&lreq->osdc_node);
2896 }
2897
2898 static bool linger_registered(struct ceph_osd_linger_request *lreq)
2899 {
2900         struct ceph_osd_client *osdc = lreq->osdc;
2901         bool registered;
2902
2903         down_read(&osdc->lock);
2904         registered = __linger_registered(lreq);
2905         up_read(&osdc->lock);
2906
2907         return registered;
2908 }
2909
2910 static void linger_register(struct ceph_osd_linger_request *lreq)
2911 {
2912         struct ceph_osd_client *osdc = lreq->osdc;
2913
2914         verify_osdc_wrlocked(osdc);
2915         WARN_ON(lreq->linger_id);
2916
2917         linger_get(lreq);
2918         lreq->linger_id = ++osdc->last_linger_id;
2919         insert_linger_osdc(&osdc->linger_requests, lreq);
2920 }
2921
2922 static void linger_unregister(struct ceph_osd_linger_request *lreq)
2923 {
2924         struct ceph_osd_client *osdc = lreq->osdc;
2925
2926         verify_osdc_wrlocked(osdc);
2927
2928         erase_linger_osdc(&osdc->linger_requests, lreq);
2929         linger_put(lreq);
2930 }
2931
2932 static void cancel_linger_request(struct ceph_osd_request *req)
2933 {
2934         struct ceph_osd_linger_request *lreq = req->r_priv;
2935
2936         WARN_ON(!req->r_linger);
2937         cancel_request(req);
2938         linger_put(lreq);
2939 }
2940
2941 struct linger_work {
2942         struct work_struct work;
2943         struct ceph_osd_linger_request *lreq;
2944         struct list_head pending_item;
2945         unsigned long queued_stamp;
2946
2947         union {
2948                 struct {
2949                         u64 notify_id;
2950                         u64 notifier_id;
2951                         void *payload; /* points into @msg front */
2952                         size_t payload_len;
2953
2954                         struct ceph_msg *msg; /* for ceph_msg_put() */
2955                 } notify;
2956                 struct {
2957                         int err;
2958                 } error;
2959         };
2960 };
2961
2962 static struct linger_work *lwork_alloc(struct ceph_osd_linger_request *lreq,
2963                                        work_func_t workfn)
2964 {
2965         struct linger_work *lwork;
2966
2967         lwork = kzalloc(sizeof(*lwork), GFP_NOIO);
2968         if (!lwork)
2969                 return NULL;
2970
2971         INIT_WORK(&lwork->work, workfn);
2972         INIT_LIST_HEAD(&lwork->pending_item);
2973         lwork->lreq = linger_get(lreq);
2974
2975         return lwork;
2976 }
2977
2978 static void lwork_free(struct linger_work *lwork)
2979 {
2980         struct ceph_osd_linger_request *lreq = lwork->lreq;
2981
2982         mutex_lock(&lreq->lock);
2983         list_del(&lwork->pending_item);
2984         mutex_unlock(&lreq->lock);
2985
2986         linger_put(lreq);
2987         kfree(lwork);
2988 }
2989
2990 static void lwork_queue(struct linger_work *lwork)
2991 {
2992         struct ceph_osd_linger_request *lreq = lwork->lreq;
2993         struct ceph_osd_client *osdc = lreq->osdc;
2994
2995         verify_lreq_locked(lreq);
2996         WARN_ON(!list_empty(&lwork->pending_item));
2997
2998         lwork->queued_stamp = jiffies;
2999         list_add_tail(&lwork->pending_item, &lreq->pending_lworks);
3000         queue_work(osdc->notify_wq, &lwork->work);
3001 }
3002
3003 static void do_watch_notify(struct work_struct *w)
3004 {
3005         struct linger_work *lwork = container_of(w, struct linger_work, work);
3006         struct ceph_osd_linger_request *lreq = lwork->lreq;
3007
3008         if (!linger_registered(lreq)) {
3009                 dout("%s lreq %p not registered\n", __func__, lreq);
3010                 goto out;
3011         }
3012
3013         WARN_ON(!lreq->is_watch);
3014         dout("%s lreq %p notify_id %llu notifier_id %llu payload_len %zu\n",
3015              __func__, lreq, lwork->notify.notify_id, lwork->notify.notifier_id,
3016              lwork->notify.payload_len);
3017         lreq->wcb(lreq->data, lwork->notify.notify_id, lreq->linger_id,
3018                   lwork->notify.notifier_id, lwork->notify.payload,
3019                   lwork->notify.payload_len);
3020
3021 out:
3022         ceph_msg_put(lwork->notify.msg);
3023         lwork_free(lwork);
3024 }
3025
3026 static void do_watch_error(struct work_struct *w)
3027 {
3028         struct linger_work *lwork = container_of(w, struct linger_work, work);
3029         struct ceph_osd_linger_request *lreq = lwork->lreq;
3030
3031         if (!linger_registered(lreq)) {
3032                 dout("%s lreq %p not registered\n", __func__, lreq);
3033                 goto out;
3034         }
3035
3036         dout("%s lreq %p err %d\n", __func__, lreq, lwork->error.err);
3037         lreq->errcb(lreq->data, lreq->linger_id, lwork->error.err);
3038
3039 out:
3040         lwork_free(lwork);
3041 }
3042
3043 static void queue_watch_error(struct ceph_osd_linger_request *lreq)
3044 {
3045         struct linger_work *lwork;
3046
3047         lwork = lwork_alloc(lreq, do_watch_error);
3048         if (!lwork) {
3049                 pr_err("failed to allocate error-lwork\n");
3050                 return;
3051         }
3052
3053         lwork->error.err = lreq->last_error;
3054         lwork_queue(lwork);
3055 }
3056
3057 static void linger_reg_commit_complete(struct ceph_osd_linger_request *lreq,
3058                                        int result)
3059 {
3060         if (!completion_done(&lreq->reg_commit_wait)) {
3061                 lreq->reg_commit_error = (result <= 0 ? result : 0);
3062                 complete_all(&lreq->reg_commit_wait);
3063         }
3064 }
3065
3066 static void linger_commit_cb(struct ceph_osd_request *req)
3067 {
3068         struct ceph_osd_linger_request *lreq = req->r_priv;
3069
3070         mutex_lock(&lreq->lock);
3071         if (req != lreq->reg_req) {
3072                 dout("%s lreq %p linger_id %llu unknown req (%p != %p)\n",
3073                      __func__, lreq, lreq->linger_id, req, lreq->reg_req);
3074                 goto out;
3075         }
3076
3077         dout("%s lreq %p linger_id %llu result %d\n", __func__, lreq,
3078              lreq->linger_id, req->r_result);
3079         linger_reg_commit_complete(lreq, req->r_result);
3080         lreq->committed = true;
3081
3082         if (!lreq->is_watch) {
3083                 struct ceph_osd_data *osd_data =
3084                     osd_req_op_data(req, 0, notify, response_data);
3085                 void *p = page_address(osd_data->pages[0]);
3086
3087                 WARN_ON(req->r_ops[0].op != CEPH_OSD_OP_NOTIFY ||
3088                         osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
3089
3090                 /* make note of the notify_id */
3091                 if (req->r_ops[0].outdata_len >= sizeof(u64)) {
3092                         lreq->notify_id = ceph_decode_64(&p);
3093                         dout("lreq %p notify_id %llu\n", lreq,
3094                              lreq->notify_id);
3095                 } else {
3096                         dout("lreq %p no notify_id\n", lreq);
3097                 }
3098         }
3099
3100 out:
3101         mutex_unlock(&lreq->lock);
3102         linger_put(lreq);
3103 }
3104
3105 static int normalize_watch_error(int err)
3106 {
3107         /*
3108          * Translate ENOENT -> ENOTCONN so that a delete->disconnection
3109          * notification and a failure to reconnect because we raced with
3110          * the delete appear the same to the user.
3111          */
3112         if (err == -ENOENT)
3113                 err = -ENOTCONN;
3114
3115         return err;
3116 }
3117
3118 static void linger_reconnect_cb(struct ceph_osd_request *req)
3119 {
3120         struct ceph_osd_linger_request *lreq = req->r_priv;
3121
3122         mutex_lock(&lreq->lock);
3123         if (req != lreq->reg_req) {
3124                 dout("%s lreq %p linger_id %llu unknown req (%p != %p)\n",
3125                      __func__, lreq, lreq->linger_id, req, lreq->reg_req);
3126                 goto out;
3127         }
3128
3129         dout("%s lreq %p linger_id %llu result %d last_error %d\n", __func__,
3130              lreq, lreq->linger_id, req->r_result, lreq->last_error);
3131         if (req->r_result < 0) {
3132                 if (!lreq->last_error) {
3133                         lreq->last_error = normalize_watch_error(req->r_result);
3134                         queue_watch_error(lreq);
3135                 }
3136         }
3137
3138 out:
3139         mutex_unlock(&lreq->lock);
3140         linger_put(lreq);
3141 }
3142
3143 static void send_linger(struct ceph_osd_linger_request *lreq)
3144 {
3145         struct ceph_osd_client *osdc = lreq->osdc;
3146         struct ceph_osd_request *req;
3147         int ret;
3148
3149         verify_osdc_wrlocked(osdc);
3150         mutex_lock(&lreq->lock);
3151         dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
3152
3153         if (lreq->reg_req) {
3154                 if (lreq->reg_req->r_osd)
3155                         cancel_linger_request(lreq->reg_req);
3156                 ceph_osdc_put_request(lreq->reg_req);
3157         }
3158
3159         req = ceph_osdc_alloc_request(osdc, NULL, 1, true, GFP_NOIO);
3160         BUG_ON(!req);
3161
3162         target_copy(&req->r_t, &lreq->t);
3163         req->r_mtime = lreq->mtime;
3164
3165         if (lreq->is_watch && lreq->committed) {
3166                 osd_req_op_watch_init(req, 0, CEPH_OSD_WATCH_OP_RECONNECT,
3167                                       lreq->linger_id, ++lreq->register_gen);
3168                 dout("lreq %p reconnect register_gen %u\n", lreq,
3169                      req->r_ops[0].watch.gen);
3170                 req->r_callback = linger_reconnect_cb;
3171         } else {
3172                 if (lreq->is_watch) {
3173                         osd_req_op_watch_init(req, 0, CEPH_OSD_WATCH_OP_WATCH,
3174                                               lreq->linger_id, 0);
3175                 } else {
3176                         lreq->notify_id = 0;
3177
3178                         refcount_inc(&lreq->request_pl->refcnt);
3179                         osd_req_op_notify_init(req, 0, lreq->linger_id,
3180                                                lreq->request_pl);
3181                         ceph_osd_data_pages_init(
3182                             osd_req_op_data(req, 0, notify, response_data),
3183                             lreq->notify_id_pages, PAGE_SIZE, 0, false, false);
3184                 }
3185                 dout("lreq %p register\n", lreq);
3186                 req->r_callback = linger_commit_cb;
3187         }
3188
3189         ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
3190         BUG_ON(ret);
3191
3192         req->r_priv = linger_get(lreq);
3193         req->r_linger = true;
3194         lreq->reg_req = req;
3195         mutex_unlock(&lreq->lock);
3196
3197         submit_request(req, true);
3198 }
3199
3200 static void linger_ping_cb(struct ceph_osd_request *req)
3201 {
3202         struct ceph_osd_linger_request *lreq = req->r_priv;
3203
3204         mutex_lock(&lreq->lock);
3205         if (req != lreq->ping_req) {
3206                 dout("%s lreq %p linger_id %llu unknown req (%p != %p)\n",
3207                      __func__, lreq, lreq->linger_id, req, lreq->ping_req);
3208                 goto out;
3209         }
3210
3211         dout("%s lreq %p linger_id %llu result %d ping_sent %lu last_error %d\n",
3212              __func__, lreq, lreq->linger_id, req->r_result, lreq->ping_sent,
3213              lreq->last_error);
3214         if (lreq->register_gen == req->r_ops[0].watch.gen) {
3215                 if (!req->r_result) {
3216                         lreq->watch_valid_thru = lreq->ping_sent;
3217                 } else if (!lreq->last_error) {
3218                         lreq->last_error = normalize_watch_error(req->r_result);
3219                         queue_watch_error(lreq);
3220                 }
3221         } else {
3222                 dout("lreq %p register_gen %u ignoring old pong %u\n", lreq,
3223                      lreq->register_gen, req->r_ops[0].watch.gen);
3224         }
3225
3226 out:
3227         mutex_unlock(&lreq->lock);
3228         linger_put(lreq);
3229 }
3230
3231 static void send_linger_ping(struct ceph_osd_linger_request *lreq)
3232 {
3233         struct ceph_osd_client *osdc = lreq->osdc;
3234         struct ceph_osd_request *req;
3235         int ret;
3236
3237         if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) {
3238                 dout("%s PAUSERD\n", __func__);
3239                 return;
3240         }
3241
3242         lreq->ping_sent = jiffies;
3243         dout("%s lreq %p linger_id %llu ping_sent %lu register_gen %u\n",
3244              __func__, lreq, lreq->linger_id, lreq->ping_sent,
3245              lreq->register_gen);
3246
3247         if (lreq->ping_req) {
3248                 if (lreq->ping_req->r_osd)
3249                         cancel_linger_request(lreq->ping_req);
3250                 ceph_osdc_put_request(lreq->ping_req);
3251         }
3252
3253         req = ceph_osdc_alloc_request(osdc, NULL, 1, true, GFP_NOIO);
3254         BUG_ON(!req);
3255
3256         target_copy(&req->r_t, &lreq->t);
3257         osd_req_op_watch_init(req, 0, CEPH_OSD_WATCH_OP_PING, lreq->linger_id,
3258                               lreq->register_gen);
3259         req->r_callback = linger_ping_cb;
3260
3261         ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
3262         BUG_ON(ret);
3263
3264         req->r_priv = linger_get(lreq);
3265         req->r_linger = true;
3266         lreq->ping_req = req;
3267
3268         ceph_osdc_get_request(req);
3269         account_request(req);
3270         req->r_tid = atomic64_inc_return(&osdc->last_tid);
3271         link_request(lreq->osd, req);
3272         send_request(req);
3273 }
3274
3275 static void linger_submit(struct ceph_osd_linger_request *lreq)
3276 {
3277         struct ceph_osd_client *osdc = lreq->osdc;
3278         struct ceph_osd *osd;
3279
3280         down_write(&osdc->lock);
3281         linger_register(lreq);
3282
3283         calc_target(osdc, &lreq->t, false);
3284         osd = lookup_create_osd(osdc, lreq->t.osd, true);
3285         link_linger(osd, lreq);
3286
3287         send_linger(lreq);
3288         up_write(&osdc->lock);
3289 }
3290
3291 static void cancel_linger_map_check(struct ceph_osd_linger_request *lreq)
3292 {
3293         struct ceph_osd_client *osdc = lreq->osdc;
3294         struct ceph_osd_linger_request *lookup_lreq;
3295
3296         verify_osdc_wrlocked(osdc);
3297
3298         lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks,
3299                                        lreq->linger_id);
3300         if (!lookup_lreq)
3301                 return;
3302
3303         WARN_ON(lookup_lreq != lreq);
3304         erase_linger_mc(&osdc->linger_map_checks, lreq);
3305         linger_put(lreq);
3306 }
3307
3308 /*
3309  * @lreq has to be both registered and linked.
3310  */
3311 static void __linger_cancel(struct ceph_osd_linger_request *lreq)
3312 {
3313         if (lreq->ping_req && lreq->ping_req->r_osd)
3314                 cancel_linger_request(lreq->ping_req);
3315         if (lreq->reg_req && lreq->reg_req->r_osd)
3316                 cancel_linger_request(lreq->reg_req);
3317         cancel_linger_map_check(lreq);
3318         unlink_linger(lreq->osd, lreq);
3319         linger_unregister(lreq);
3320 }
3321
3322 static void linger_cancel(struct ceph_osd_linger_request *lreq)
3323 {
3324         struct ceph_osd_client *osdc = lreq->osdc;
3325
3326         down_write(&osdc->lock);
3327         if (__linger_registered(lreq))
3328                 __linger_cancel(lreq);
3329         up_write(&osdc->lock);
3330 }
3331
3332 static void send_linger_map_check(struct ceph_osd_linger_request *lreq);
3333
3334 static void check_linger_pool_dne(struct ceph_osd_linger_request *lreq)
3335 {
3336         struct ceph_osd_client *osdc = lreq->osdc;
3337         struct ceph_osdmap *map = osdc->osdmap;
3338
3339         verify_osdc_wrlocked(osdc);
3340         WARN_ON(!map->epoch);
3341
3342         if (lreq->register_gen) {
3343                 lreq->map_dne_bound = map->epoch;
3344                 dout("%s lreq %p linger_id %llu pool disappeared\n", __func__,
3345                      lreq, lreq->linger_id);
3346         } else {
3347                 dout("%s lreq %p linger_id %llu map_dne_bound %u have %u\n",
3348                      __func__, lreq, lreq->linger_id, lreq->map_dne_bound,
3349                      map->epoch);
3350         }
3351
3352         if (lreq->map_dne_bound) {
3353                 if (map->epoch >= lreq->map_dne_bound) {
3354                         /* we had a new enough map */
3355                         pr_info("linger_id %llu pool does not exist\n",
3356                                 lreq->linger_id);
3357                         linger_reg_commit_complete(lreq, -ENOENT);
3358                         __linger_cancel(lreq);
3359                 }
3360         } else {
3361                 send_linger_map_check(lreq);
3362         }
3363 }
3364
3365 static void linger_map_check_cb(struct ceph_mon_generic_request *greq)
3366 {
3367         struct ceph_osd_client *osdc = &greq->monc->client->osdc;
3368         struct ceph_osd_linger_request *lreq;
3369         u64 linger_id = greq->private_data;
3370
3371         WARN_ON(greq->result || !greq->u.newest);
3372
3373         down_write(&osdc->lock);
3374         lreq = lookup_linger_mc(&osdc->linger_map_checks, linger_id);
3375         if (!lreq) {
3376                 dout("%s linger_id %llu dne\n", __func__, linger_id);
3377                 goto out_unlock;
3378         }
3379
3380         dout("%s lreq %p linger_id %llu map_dne_bound %u newest %llu\n",
3381              __func__, lreq, lreq->linger_id, lreq->map_dne_bound,
3382              greq->u.newest);
3383         if (!lreq->map_dne_bound)
3384                 lreq->map_dne_bound = greq->u.newest;
3385         erase_linger_mc(&osdc->linger_map_checks, lreq);
3386         check_linger_pool_dne(lreq);
3387
3388         linger_put(lreq);
3389 out_unlock:
3390         up_write(&osdc->lock);
3391 }
3392
3393 static void send_linger_map_check(struct ceph_osd_linger_request *lreq)
3394 {
3395         struct ceph_osd_client *osdc = lreq->osdc;
3396         struct ceph_osd_linger_request *lookup_lreq;
3397         int ret;
3398
3399         verify_osdc_wrlocked(osdc);
3400
3401         lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks,
3402                                        lreq->linger_id);
3403         if (lookup_lreq) {
3404                 WARN_ON(lookup_lreq != lreq);
3405                 return;
3406         }
3407
3408         linger_get(lreq);
3409         insert_linger_mc(&osdc->linger_map_checks, lreq);
3410         ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap",
3411                                           linger_map_check_cb, lreq->linger_id);
3412         WARN_ON(ret);
3413 }
3414
3415 static int linger_reg_commit_wait(struct ceph_osd_linger_request *lreq)
3416 {
3417         int ret;
3418
3419         dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
3420         ret = wait_for_completion_killable(&lreq->reg_commit_wait);
3421         return ret ?: lreq->reg_commit_error;
3422 }
3423
3424 static int linger_notify_finish_wait(struct ceph_osd_linger_request *lreq,
3425                                      unsigned long timeout)
3426 {
3427         long left;
3428
3429         dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
3430         left = wait_for_completion_killable_timeout(&lreq->notify_finish_wait,
3431                                                 ceph_timeout_jiffies(timeout));
3432         if (left <= 0)
3433                 left = left ?: -ETIMEDOUT;
3434         else
3435                 left = lreq->notify_finish_error; /* completed */
3436
3437         return left;
3438 }
3439
3440 /*
3441  * Timeout callback, called every N seconds.  When 1 or more OSD
3442  * requests has been active for more than N seconds, we send a keepalive
3443  * (tag + timestamp) to its OSD to ensure any communications channel
3444  * reset is detected.
3445  */
3446 static void handle_timeout(struct work_struct *work)
3447 {
3448         struct ceph_osd_client *osdc =
3449                 container_of(work, struct ceph_osd_client, timeout_work.work);
3450         struct ceph_options *opts = osdc->client->options;
3451         unsigned long cutoff = jiffies - opts->osd_keepalive_timeout;
3452         unsigned long expiry_cutoff = jiffies - opts->osd_request_timeout;
3453         LIST_HEAD(slow_osds);
3454         struct rb_node *n, *p;
3455
3456         dout("%s osdc %p\n", __func__, osdc);
3457         down_write(&osdc->lock);
3458
3459         /*
3460          * ping osds that are a bit slow.  this ensures that if there
3461          * is a break in the TCP connection we will notice, and reopen
3462          * a connection with that osd (from the fault callback).
3463          */
3464         for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
3465                 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
3466                 bool found = false;
3467
3468                 for (p = rb_first(&osd->o_requests); p; ) {
3469                         struct ceph_osd_request *req =
3470                             rb_entry(p, struct ceph_osd_request, r_node);
3471
3472                         p = rb_next(p); /* abort_request() */
3473
3474                         if (time_before(req->r_stamp, cutoff)) {
3475                                 dout(" req %p tid %llu on osd%d is laggy\n",
3476                                      req, req->r_tid, osd->o_osd);
3477                                 found = true;
3478                         }
3479                         if (opts->osd_request_timeout &&
3480                             time_before(req->r_start_stamp, expiry_cutoff)) {
3481                                 pr_err_ratelimited("tid %llu on osd%d timeout\n",
3482                                        req->r_tid, osd->o_osd);
3483                                 abort_request(req, -ETIMEDOUT);
3484                         }
3485                 }
3486                 for (p = rb_first(&osd->o_linger_requests); p; p = rb_next(p)) {
3487                         struct ceph_osd_linger_request *lreq =
3488                             rb_entry(p, struct ceph_osd_linger_request, node);
3489
3490                         dout(" lreq %p linger_id %llu is served by osd%d\n",
3491                              lreq, lreq->linger_id, osd->o_osd);
3492                         found = true;
3493
3494                         mutex_lock(&lreq->lock);
3495                         if (lreq->is_watch && lreq->committed && !lreq->last_error)
3496                                 send_linger_ping(lreq);
3497                         mutex_unlock(&lreq->lock);
3498                 }
3499
3500                 if (found)
3501                         list_move_tail(&osd->o_keepalive_item, &slow_osds);
3502         }
3503
3504         if (opts->osd_request_timeout) {
3505                 for (p = rb_first(&osdc->homeless_osd.o_requests); p; ) {
3506                         struct ceph_osd_request *req =
3507                             rb_entry(p, struct ceph_osd_request, r_node);
3508
3509                         p = rb_next(p); /* abort_request() */
3510
3511                         if (time_before(req->r_start_stamp, expiry_cutoff)) {
3512                                 pr_err_ratelimited("tid %llu on osd%d timeout\n",
3513                                        req->r_tid, osdc->homeless_osd.o_osd);
3514                                 abort_request(req, -ETIMEDOUT);
3515                         }
3516                 }
3517         }
3518
3519         if (atomic_read(&osdc->num_homeless) || !list_empty(&slow_osds))
3520                 maybe_request_map(osdc);
3521
3522         while (!list_empty(&slow_osds)) {
3523                 struct ceph_osd *osd = list_first_entry(&slow_osds,
3524                                                         struct ceph_osd,
3525                                                         o_keepalive_item);
3526                 list_del_init(&osd->o_keepalive_item);
3527                 ceph_con_keepalive(&osd->o_con);
3528         }
3529
3530         up_write(&osdc->lock);
3531         schedule_delayed_work(&osdc->timeout_work,
3532                               osdc->client->options->osd_keepalive_timeout);
3533 }
3534
3535 static void handle_osds_timeout(struct work_struct *work)
3536 {
3537         struct ceph_osd_client *osdc =
3538                 container_of(work, struct ceph_osd_client,
3539                              osds_timeout_work.work);
3540         unsigned long delay = osdc->client->options->osd_idle_ttl / 4;
3541         struct ceph_osd *osd, *nosd;
3542
3543         dout("%s osdc %p\n", __func__, osdc);
3544         down_write(&osdc->lock);
3545         list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
3546                 if (time_before(jiffies, osd->lru_ttl))
3547                         break;
3548
3549                 WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests));
3550                 WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests));
3551                 close_osd(osd);
3552         }
3553
3554         up_write(&osdc->lock);
3555         schedule_delayed_work(&osdc->osds_timeout_work,
3556                               round_jiffies_relative(delay));
3557 }
3558
3559 static int ceph_oloc_decode(void **p, void *end,
3560                             struct ceph_object_locator *oloc)
3561 {
3562         u8 struct_v, struct_cv;
3563         u32 len;
3564         void *struct_end;
3565         int ret = 0;
3566
3567         ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
3568         struct_v = ceph_decode_8(p);
3569         struct_cv = ceph_decode_8(p);
3570         if (struct_v < 3) {
3571                 pr_warn("got v %d < 3 cv %d of ceph_object_locator\n",
3572                         struct_v, struct_cv);
3573                 goto e_inval;
3574         }
3575         if (struct_cv > 6) {
3576                 pr_warn("got v %d cv %d > 6 of ceph_object_locator\n",
3577                         struct_v, struct_cv);
3578                 goto e_inval;
3579         }
3580         len = ceph_decode_32(p);
3581         ceph_decode_need(p, end, len, e_inval);
3582         struct_end = *p + len;
3583
3584         oloc->pool = ceph_decode_64(p);
3585         *p += 4; /* skip preferred */
3586
3587         len = ceph_decode_32(p);
3588         if (len > 0) {
3589                 pr_warn("ceph_object_locator::key is set\n");
3590                 goto e_inval;
3591         }
3592
3593         if (struct_v >= 5) {
3594                 bool changed = false;
3595
3596                 len = ceph_decode_32(p);
3597                 if (len > 0) {
3598                         ceph_decode_need(p, end, len, e_inval);
3599                         if (!oloc->pool_ns ||
3600                             ceph_compare_string(oloc->pool_ns, *p, len))
3601                                 changed = true;
3602                         *p += len;
3603                 } else {
3604                         if (oloc->pool_ns)
3605                                 changed = true;
3606                 }
3607                 if (changed) {
3608                         /* redirect changes namespace */
3609                         pr_warn("ceph_object_locator::nspace is changed\n");
3610                         goto e_inval;
3611                 }
3612         }
3613
3614         if (struct_v >= 6) {
3615                 s64 hash = ceph_decode_64(p);
3616                 if (hash != -1) {
3617                         pr_warn("ceph_object_locator::hash is set\n");
3618                         goto e_inval;
3619                 }
3620         }
3621
3622         /* skip the rest */
3623         *p = struct_end;
3624 out:
3625         return ret;
3626
3627 e_inval:
3628         ret = -EINVAL;
3629         goto out;
3630 }
3631
3632 static int ceph_redirect_decode(void **p, void *end,
3633                                 struct ceph_request_redirect *redir)
3634 {
3635         u8 struct_v, struct_cv;
3636         u32 len;
3637         void *struct_end;
3638         int ret;
3639
3640         ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
3641         struct_v = ceph_decode_8(p);
3642         struct_cv = ceph_decode_8(p);
3643         if (struct_cv > 1) {
3644                 pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n",
3645                         struct_v, struct_cv);
3646                 goto e_inval;
3647         }
3648         len = ceph_decode_32(p);
3649         ceph_decode_need(p, end, len, e_inval);
3650         struct_end = *p + len;
3651
3652         ret = ceph_oloc_decode(p, end, &redir->oloc);
3653         if (ret)
3654                 goto out;
3655
3656         len = ceph_decode_32(p);
3657         if (len > 0) {
3658                 pr_warn("ceph_request_redirect::object_name is set\n");
3659                 goto e_inval;
3660         }
3661
3662         /* skip the rest */
3663         *p = struct_end;
3664 out:
3665         return ret;
3666
3667 e_inval:
3668         ret = -EINVAL;
3669         goto out;
3670 }
3671
3672 struct MOSDOpReply {
3673         struct ceph_pg pgid;
3674         u64 flags;
3675         int result;
3676         u32 epoch;
3677         int num_ops;
3678         u32 outdata_len[CEPH_OSD_MAX_OPS];
3679         s32 rval[CEPH_OSD_MAX_OPS];
3680         int retry_attempt;
3681         struct ceph_eversion replay_version;
3682         u64 user_version;
3683         struct ceph_request_redirect redirect;
3684 };
3685
3686 static int decode_MOSDOpReply(const struct ceph_msg *msg, struct MOSDOpReply *m)
3687 {
3688         void *p = msg->front.iov_base;
3689         void *const end = p + msg->front.iov_len;
3690         u16 version = le16_to_cpu(msg->hdr.version);
3691         struct ceph_eversion bad_replay_version;
3692         u8 decode_redir;
3693         u32 len;
3694         int ret;
3695         int i;
3696
3697         ceph_decode_32_safe(&p, end, len, e_inval);
3698         ceph_decode_need(&p, end, len, e_inval);
3699         p += len; /* skip oid */
3700
3701         ret = ceph_decode_pgid(&p, end, &m->pgid);
3702         if (ret)
3703                 return ret;
3704
3705         ceph_decode_64_safe(&p, end, m->flags, e_inval);
3706         ceph_decode_32_safe(&p, end, m->result, e_inval);
3707         ceph_decode_need(&p, end, sizeof(bad_replay_version), e_inval);
3708         memcpy(&bad_replay_version, p, sizeof(bad_replay_version));
3709         p += sizeof(bad_replay_version);
3710         ceph_decode_32_safe(&p, end, m->epoch, e_inval);
3711
3712         ceph_decode_32_safe(&p, end, m->num_ops, e_inval);
3713         if (m->num_ops > ARRAY_SIZE(m->outdata_len))
3714                 goto e_inval;
3715
3716         ceph_decode_need(&p, end, m->num_ops * sizeof(struct ceph_osd_op),
3717                          e_inval);
3718         for (i = 0; i < m->num_ops; i++) {
3719                 struct ceph_osd_op *op = p;
3720
3721                 m->outdata_len[i] = le32_to_cpu(op->payload_len);
3722                 p += sizeof(*op);
3723         }
3724
3725         ceph_decode_32_safe(&p, end, m->retry_attempt, e_inval);
3726         for (i = 0; i < m->num_ops; i++)
3727                 ceph_decode_32_safe(&p, end, m->rval[i], e_inval);
3728
3729         if (version >= 5) {
3730                 ceph_decode_need(&p, end, sizeof(m->replay_version), e_inval);
3731                 memcpy(&m->replay_version, p, sizeof(m->replay_version));
3732                 p += sizeof(m->replay_version);
3733                 ceph_decode_64_safe(&p, end, m->user_version, e_inval);
3734         } else {
3735                 m->replay_version = bad_replay_version; /* struct */
3736                 m->user_version = le64_to_cpu(m->replay_version.version);
3737         }
3738
3739         if (version >= 6) {
3740                 if (version >= 7)
3741                         ceph_decode_8_safe(&p, end, decode_redir, e_inval);
3742                 else
3743                         decode_redir = 1;
3744         } else {
3745                 decode_redir = 0;
3746         }
3747
3748         if (decode_redir) {
3749                 ret = ceph_redirect_decode(&p, end, &m->redirect);
3750                 if (ret)
3751                         return ret;
3752         } else {
3753                 ceph_oloc_init(&m->redirect.oloc);
3754         }
3755
3756         return 0;
3757
3758 e_inval:
3759         return -EINVAL;
3760 }
3761
3762 /*
3763  * Handle MOSDOpReply.  Set ->r_result and call the callback if it is
3764  * specified.
3765  */
3766 static void handle_reply(struct ceph_osd *osd, struct ceph_msg *msg)
3767 {
3768         struct ceph_osd_client *osdc = osd->o_osdc;
3769         struct ceph_osd_request *req;
3770         struct MOSDOpReply m;
3771         u64 tid = le64_to_cpu(msg->hdr.tid);
3772         u32 data_len = 0;
3773         int ret;
3774         int i;
3775
3776         dout("%s msg %p tid %llu\n", __func__, msg, tid);
3777
3778         down_read(&osdc->lock);
3779         if (!osd_registered(osd)) {
3780                 dout("%s osd%d unknown\n", __func__, osd->o_osd);
3781                 goto out_unlock_osdc;
3782         }
3783         WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num));
3784
3785         mutex_lock(&osd->lock);
3786         req = lookup_request(&osd->o_requests, tid);
3787         if (!req) {
3788                 dout("%s osd%d tid %llu unknown\n", __func__, osd->o_osd, tid);
3789                 goto out_unlock_session;
3790         }
3791
3792         m.redirect.oloc.pool_ns = req->r_t.target_oloc.pool_ns;
3793         ret = decode_MOSDOpReply(msg, &m);
3794         m.redirect.oloc.pool_ns = NULL;
3795         if (ret) {
3796                 pr_err("failed to decode MOSDOpReply for tid %llu: %d\n",
3797                        req->r_tid, ret);
3798                 ceph_msg_dump(msg);
3799                 goto fail_request;
3800         }
3801         dout("%s req %p tid %llu flags 0x%llx pgid %llu.%x epoch %u attempt %d v %u'%llu uv %llu\n",
3802              __func__, req, req->r_tid, m.flags, m.pgid.pool, m.pgid.seed,
3803              m.epoch, m.retry_attempt, le32_to_cpu(m.replay_version.epoch),
3804              le64_to_cpu(m.replay_version.version), m.user_version);
3805
3806         if (m.retry_attempt >= 0) {
3807                 if (m.retry_attempt != req->r_attempts - 1) {
3808                         dout("req %p tid %llu retry_attempt %d != %d, ignoring\n",
3809                              req, req->r_tid, m.retry_attempt,
3810                              req->r_attempts - 1);
3811                         goto out_unlock_session;
3812                 }
3813         } else {
3814                 WARN_ON(1); /* MOSDOpReply v4 is assumed */
3815         }
3816
3817         if (!ceph_oloc_empty(&m.redirect.oloc)) {
3818                 dout("req %p tid %llu redirect pool %lld\n", req, req->r_tid,
3819                      m.redirect.oloc.pool);
3820                 unlink_request(osd, req);
3821                 mutex_unlock(&osd->lock);
3822
3823                 /*
3824                  * Not ceph_oloc_copy() - changing pool_ns is not
3825                  * supported.
3826                  */
3827                 req->r_t.target_oloc.pool = m.redirect.oloc.pool;
3828                 req->r_flags |= CEPH_OSD_FLAG_REDIRECTED |
3829                                 CEPH_OSD_FLAG_IGNORE_OVERLAY |
3830                                 CEPH_OSD_FLAG_IGNORE_CACHE;
3831                 req->r_tid = 0;
3832                 __submit_request(req, false);
3833                 goto out_unlock_osdc;
3834         }
3835
3836         if (m.result == -EAGAIN) {
3837                 dout("req %p tid %llu EAGAIN\n", req, req->r_tid);
3838                 unlink_request(osd, req);
3839                 mutex_unlock(&osd->lock);
3840
3841                 /*
3842                  * The object is missing on the replica or not (yet)
3843                  * readable.  Clear pgid to force a resend to the primary
3844                  * via legacy_change.
3845                  */
3846                 req->r_t.pgid.pool = 0;
3847                 req->r_t.pgid.seed = 0;
3848                 WARN_ON(!req->r_t.used_replica);
3849                 req->r_flags &= ~(CEPH_OSD_FLAG_BALANCE_READS |
3850                                   CEPH_OSD_FLAG_LOCALIZE_READS);
3851                 req->r_tid = 0;
3852                 __submit_request(req, false);
3853                 goto out_unlock_osdc;
3854         }
3855
3856         if (m.num_ops != req->r_num_ops) {
3857                 pr_err("num_ops %d != %d for tid %llu\n", m.num_ops,
3858                        req->r_num_ops, req->r_tid);
3859                 goto fail_request;
3860         }
3861         for (i = 0; i < req->r_num_ops; i++) {
3862                 dout(" req %p tid %llu op %d rval %d len %u\n", req,
3863                      req->r_tid, i, m.rval[i], m.outdata_len[i]);
3864                 req->r_ops[i].rval = m.rval[i];
3865                 req->r_ops[i].outdata_len = m.outdata_len[i];
3866                 data_len += m.outdata_len[i];
3867         }
3868         if (data_len != le32_to_cpu(msg->hdr.data_len)) {
3869                 pr_err("sum of lens %u != %u for tid %llu\n", data_len,
3870                        le32_to_cpu(msg->hdr.data_len), req->r_tid);
3871                 goto fail_request;
3872         }
3873         dout("%s req %p tid %llu result %d data_len %u\n", __func__,
3874              req, req->r_tid, m.result, data_len);
3875
3876         /*
3877          * Since we only ever request ONDISK, we should only ever get
3878          * one (type of) reply back.
3879          */
3880         WARN_ON(!(m.flags & CEPH_OSD_FLAG_ONDISK));
3881         req->r_version = m.user_version;
3882         req->r_result = m.result ?: data_len;
3883         finish_request(req);
3884         mutex_unlock(&osd->lock);
3885         up_read(&osdc->lock);
3886
3887         __complete_request(req);
3888         return;
3889
3890 fail_request:
3891         complete_request(req, -EIO);
3892 out_unlock_session:
3893         mutex_unlock(&osd->lock);
3894 out_unlock_osdc:
3895         up_read(&osdc->lock);
3896 }
3897
3898 static void set_pool_was_full(struct ceph_osd_client *osdc)
3899 {
3900         struct rb_node *n;
3901
3902         for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) {
3903                 struct ceph_pg_pool_info *pi =
3904                     rb_entry(n, struct ceph_pg_pool_info, node);
3905
3906                 pi->was_full = __pool_full(pi);
3907         }
3908 }
3909
3910 static bool pool_cleared_full(struct ceph_osd_client *osdc, s64 pool_id)
3911 {
3912         struct ceph_pg_pool_info *pi;
3913
3914         pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id);
3915         if (!pi)
3916                 return false;
3917
3918         return pi->was_full && !__pool_full(pi);
3919 }
3920
3921 static enum calc_target_result
3922 recalc_linger_target(struct ceph_osd_linger_request *lreq)
3923 {
3924         struct ceph_osd_client *osdc = lreq->osdc;
3925         enum calc_target_result ct_res;
3926
3927         ct_res = calc_target(osdc, &lreq->t, true);
3928         if (ct_res == CALC_TARGET_NEED_RESEND) {
3929                 struct ceph_osd *osd;
3930
3931                 osd = lookup_create_osd(osdc, lreq->t.osd, true);
3932                 if (osd != lreq->osd) {
3933                         unlink_linger(lreq->osd, lreq);
3934                         link_linger(osd, lreq);
3935                 }
3936         }
3937
3938         return ct_res;
3939 }
3940
3941 /*
3942  * Requeue requests whose mapping to an OSD has changed.
3943  */
3944 static void scan_requests(struct ceph_osd *osd,
3945                           bool force_resend,
3946                           bool cleared_full,
3947                           bool check_pool_cleared_full,
3948                           struct rb_root *need_resend,
3949                           struct list_head *need_resend_linger)
3950 {
3951         struct ceph_osd_client *osdc = osd->o_osdc;
3952         struct rb_node *n;
3953         bool force_resend_writes;
3954
3955         for (n = rb_first(&osd->o_linger_requests); n; ) {
3956                 struct ceph_osd_linger_request *lreq =
3957                     rb_entry(n, struct ceph_osd_linger_request, node);
3958                 enum calc_target_result ct_res;
3959
3960                 n = rb_next(n); /* recalc_linger_target() */
3961
3962                 dout("%s lreq %p linger_id %llu\n", __func__, lreq,
3963                      lreq->linger_id);
3964                 ct_res = recalc_linger_target(lreq);
3965                 switch (ct_res) {
3966                 case CALC_TARGET_NO_ACTION:
3967                         force_resend_writes = cleared_full ||
3968                             (check_pool_cleared_full &&
3969                              pool_cleared_full(osdc, lreq->t.base_oloc.pool));
3970                         if (!force_resend && !force_resend_writes)
3971                                 break;
3972
3973                         fallthrough;
3974                 case CALC_TARGET_NEED_RESEND:
3975                         cancel_linger_map_check(lreq);
3976                         /*
3977                          * scan_requests() for the previous epoch(s)
3978                          * may have already added it to the list, since
3979                          * it's not unlinked here.
3980                          */
3981                         if (list_empty(&lreq->scan_item))
3982                                 list_add_tail(&lreq->scan_item, need_resend_linger);
3983                         break;
3984                 case CALC_TARGET_POOL_DNE:
3985                         list_del_init(&lreq->scan_item);
3986                         check_linger_pool_dne(lreq);
3987                         break;
3988                 }
3989         }
3990
3991         for (n = rb_first(&osd->o_requests); n; ) {
3992                 struct ceph_osd_request *req =
3993                     rb_entry(n, struct ceph_osd_request, r_node);
3994                 enum calc_target_result ct_res;
3995
3996                 n = rb_next(n); /* unlink_request(), check_pool_dne() */
3997
3998                 dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
3999                 ct_res = calc_target(osdc, &req->r_t, false);
4000                 switch (ct_res) {
4001                 case CALC_TARGET_NO_ACTION:
4002                         force_resend_writes = cleared_full ||
4003                             (check_pool_cleared_full &&
4004                              pool_cleared_full(osdc, req->r_t.base_oloc.pool));
4005                         if (!force_resend &&
4006                             (!(req->r_flags & CEPH_OSD_FLAG_WRITE) ||
4007                              !force_resend_writes))
4008                                 break;
4009
4010                         fallthrough;
4011                 case CALC_TARGET_NEED_RESEND:
4012                         cancel_map_check(req);
4013                         unlink_request(osd, req);
4014                         insert_request(need_resend, req);
4015                         break;
4016                 case CALC_TARGET_POOL_DNE:
4017                         check_pool_dne(req);
4018                         break;
4019                 }
4020         }
4021 }
4022
4023 static int handle_one_map(struct ceph_osd_client *osdc,
4024                           void *p, void *end, bool incremental,
4025                           struct rb_root *need_resend,
4026                           struct list_head *need_resend_linger)
4027 {
4028         struct ceph_osdmap *newmap;
4029         struct rb_node *n;
4030         bool skipped_map = false;
4031         bool was_full;
4032
4033         was_full = ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL);
4034         set_pool_was_full(osdc);
4035
4036         if (incremental)
4037                 newmap = osdmap_apply_incremental(&p, end,
4038                                                   ceph_msgr2(osdc->client),
4039                                                   osdc->osdmap);
4040         else
4041                 newmap = ceph_osdmap_decode(&p, end, ceph_msgr2(osdc->client));
4042         if (IS_ERR(newmap))
4043                 return PTR_ERR(newmap);
4044
4045         if (newmap != osdc->osdmap) {
4046                 /*
4047                  * Preserve ->was_full before destroying the old map.
4048                  * For pools that weren't in the old map, ->was_full
4049                  * should be false.
4050                  */
4051                 for (n = rb_first(&newmap->pg_pools); n; n = rb_next(n)) {
4052                         struct ceph_pg_pool_info *pi =
4053                             rb_entry(n, struct ceph_pg_pool_info, node);
4054                         struct ceph_pg_pool_info *old_pi;
4055
4056                         old_pi = ceph_pg_pool_by_id(osdc->osdmap, pi->id);
4057                         if (old_pi)
4058                                 pi->was_full = old_pi->was_full;
4059                         else
4060                                 WARN_ON(pi->was_full);
4061                 }
4062
4063                 if (osdc->osdmap->epoch &&
4064                     osdc->osdmap->epoch + 1 < newmap->epoch) {
4065                         WARN_ON(incremental);
4066                         skipped_map = true;
4067                 }
4068
4069                 ceph_osdmap_destroy(osdc->osdmap);
4070                 osdc->osdmap = newmap;
4071         }
4072
4073         was_full &= !ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL);
4074         scan_requests(&osdc->homeless_osd, skipped_map, was_full, true,
4075                       need_resend, need_resend_linger);
4076
4077         for (n = rb_first(&osdc->osds); n; ) {
4078                 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
4079
4080                 n = rb_next(n); /* close_osd() */
4081
4082                 scan_requests(osd, skipped_map, was_full, true, need_resend,
4083                               need_resend_linger);
4084                 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
4085                     memcmp(&osd->o_con.peer_addr,
4086                            ceph_osd_addr(osdc->osdmap, osd->o_osd),
4087                            sizeof(struct ceph_entity_addr)))
4088                         close_osd(osd);
4089         }
4090
4091         return 0;
4092 }
4093
4094 static void kick_requests(struct ceph_osd_client *osdc,
4095                           struct rb_root *need_resend,
4096                           struct list_head *need_resend_linger)
4097 {
4098         struct ceph_osd_linger_request *lreq, *nlreq;
4099         enum calc_target_result ct_res;
4100         struct rb_node *n;
4101
4102         /* make sure need_resend targets reflect latest map */
4103         for (n = rb_first(need_resend); n; ) {
4104                 struct ceph_osd_request *req =
4105                     rb_entry(n, struct ceph_osd_request, r_node);
4106
4107                 n = rb_next(n);
4108
4109                 if (req->r_t.epoch < osdc->osdmap->epoch) {
4110                         ct_res = calc_target(osdc, &req->r_t, false);
4111                         if (ct_res == CALC_TARGET_POOL_DNE) {
4112                                 erase_request(need_resend, req);
4113                                 check_pool_dne(req);
4114                         }
4115                 }
4116         }
4117
4118         for (n = rb_first(need_resend); n; ) {
4119                 struct ceph_osd_request *req =
4120                     rb_entry(n, struct ceph_osd_request, r_node);
4121                 struct ceph_osd *osd;
4122
4123                 n = rb_next(n);
4124                 erase_request(need_resend, req); /* before link_request() */
4125
4126                 osd = lookup_create_osd(osdc, req->r_t.osd, true);
4127                 link_request(osd, req);
4128                 if (!req->r_linger) {
4129                         if (!osd_homeless(osd) && !req->r_t.paused)
4130                                 send_request(req);
4131                 } else {
4132                         cancel_linger_request(req);
4133                 }
4134         }
4135
4136         list_for_each_entry_safe(lreq, nlreq, need_resend_linger, scan_item) {
4137                 if (!osd_homeless(lreq->osd))
4138                         send_linger(lreq);
4139
4140                 list_del_init(&lreq->scan_item);
4141         }
4142 }
4143
4144 /*
4145  * Process updated osd map.
4146  *
4147  * The message contains any number of incremental and full maps, normally
4148  * indicating some sort of topology change in the cluster.  Kick requests
4149  * off to different OSDs as needed.
4150  */
4151 void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
4152 {
4153         void *p = msg->front.iov_base;
4154         void *const end = p + msg->front.iov_len;
4155         u32 nr_maps, maplen;
4156         u32 epoch;
4157         struct ceph_fsid fsid;
4158         struct rb_root need_resend = RB_ROOT;
4159         LIST_HEAD(need_resend_linger);
4160         bool handled_incremental = false;
4161         bool was_pauserd, was_pausewr;
4162         bool pauserd, pausewr;
4163         int err;
4164
4165         dout("%s have %u\n", __func__, osdc->osdmap->epoch);
4166         down_write(&osdc->lock);
4167
4168         /* verify fsid */
4169         ceph_decode_need(&p, end, sizeof(fsid), bad);
4170         ceph_decode_copy(&p, &fsid, sizeof(fsid));
4171         if (ceph_check_fsid(osdc->client, &fsid) < 0)
4172                 goto bad;
4173
4174         was_pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
4175         was_pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
4176                       ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
4177                       have_pool_full(osdc);
4178
4179         /* incremental maps */
4180         ceph_decode_32_safe(&p, end, nr_maps, bad);
4181         dout(" %d inc maps\n", nr_maps);
4182         while (nr_maps > 0) {
4183                 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
4184                 epoch = ceph_decode_32(&p);
4185                 maplen = ceph_decode_32(&p);
4186                 ceph_decode_need(&p, end, maplen, bad);
4187                 if (osdc->osdmap->epoch &&
4188                     osdc->osdmap->epoch + 1 == epoch) {
4189                         dout("applying incremental map %u len %d\n",
4190                              epoch, maplen);
4191                         err = handle_one_map(osdc, p, p + maplen, true,
4192                                              &need_resend, &need_resend_linger);
4193                         if (err)
4194                                 goto bad;
4195                         handled_incremental = true;
4196                 } else {
4197                         dout("ignoring incremental map %u len %d\n",
4198                              epoch, maplen);
4199                 }
4200                 p += maplen;
4201                 nr_maps--;
4202         }
4203         if (handled_incremental)
4204                 goto done;
4205
4206         /* full maps */
4207         ceph_decode_32_safe(&p, end, nr_maps, bad);
4208         dout(" %d full maps\n", nr_maps);
4209         while (nr_maps) {
4210                 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
4211                 epoch = ceph_decode_32(&p);
4212                 maplen = ceph_decode_32(&p);
4213                 ceph_decode_need(&p, end, maplen, bad);
4214                 if (nr_maps > 1) {
4215                         dout("skipping non-latest full map %u len %d\n",
4216                              epoch, maplen);
4217                 } else if (osdc->osdmap->epoch >= epoch) {
4218                         dout("skipping full map %u len %d, "
4219                              "older than our %u\n", epoch, maplen,
4220                              osdc->osdmap->epoch);
4221                 } else {
4222                         dout("taking full map %u len %d\n", epoch, maplen);
4223                         err = handle_one_map(osdc, p, p + maplen, false,
4224                                              &need_resend, &need_resend_linger);
4225                         if (err)
4226                                 goto bad;
4227                 }
4228                 p += maplen;
4229                 nr_maps--;
4230         }
4231
4232 done:
4233         /*
4234          * subscribe to subsequent osdmap updates if full to ensure
4235          * we find out when we are no longer full and stop returning
4236          * ENOSPC.
4237          */
4238         pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
4239         pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
4240                   ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
4241                   have_pool_full(osdc);
4242         if (was_pauserd || was_pausewr || pauserd || pausewr ||
4243             osdc->osdmap->epoch < osdc->epoch_barrier)
4244                 maybe_request_map(osdc);
4245
4246         kick_requests(osdc, &need_resend, &need_resend_linger);
4247
4248         ceph_osdc_abort_on_full(osdc);
4249         ceph_monc_got_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
4250                           osdc->osdmap->epoch);
4251         up_write(&osdc->lock);
4252         wake_up_all(&osdc->client->auth_wq);
4253         return;
4254
4255 bad:
4256         pr_err("osdc handle_map corrupt msg\n");
4257         ceph_msg_dump(msg);
4258         up_write(&osdc->lock);
4259 }
4260
4261 /*
4262  * Resubmit requests pending on the given osd.
4263  */
4264 static void kick_osd_requests(struct ceph_osd *osd)
4265 {
4266         struct rb_node *n;
4267
4268         clear_backoffs(osd);
4269
4270         for (n = rb_first(&osd->o_requests); n; ) {
4271                 struct ceph_osd_request *req =
4272                     rb_entry(n, struct ceph_osd_request, r_node);
4273
4274                 n = rb_next(n); /* cancel_linger_request() */
4275
4276                 if (!req->r_linger) {
4277                         if (!req->r_t.paused)
4278                                 send_request(req);
4279                 } else {
4280                         cancel_linger_request(req);
4281                 }
4282         }
4283         for (n = rb_first(&osd->o_linger_requests); n; n = rb_next(n)) {
4284                 struct ceph_osd_linger_request *lreq =
4285                     rb_entry(n, struct ceph_osd_linger_request, node);
4286
4287                 send_linger(lreq);
4288         }
4289 }
4290
4291 /*
4292  * If the osd connection drops, we need to resubmit all requests.
4293  */
4294 static void osd_fault(struct ceph_connection *con)
4295 {
4296         struct ceph_osd *osd = con->private;
4297         struct ceph_osd_client *osdc = osd->o_osdc;
4298
4299         dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
4300
4301         down_write(&osdc->lock);
4302         if (!osd_registered(osd)) {
4303                 dout("%s osd%d unknown\n", __func__, osd->o_osd);
4304                 goto out_unlock;
4305         }
4306
4307         if (!reopen_osd(osd))
4308                 kick_osd_requests(osd);
4309         maybe_request_map(osdc);
4310
4311 out_unlock:
4312         up_write(&osdc->lock);
4313 }
4314
4315 struct MOSDBackoff {
4316         struct ceph_spg spgid;
4317         u32 map_epoch;
4318         u8 op;
4319         u64 id;
4320         struct ceph_hobject_id *begin;
4321         struct ceph_hobject_id *end;
4322 };
4323
4324 static int decode_MOSDBackoff(const struct ceph_msg *msg, struct MOSDBackoff *m)
4325 {
4326         void *p = msg->front.iov_base;
4327         void *const end = p + msg->front.iov_len;
4328         u8 struct_v;
4329         u32 struct_len;
4330         int ret;
4331
4332         ret = ceph_start_decoding(&p, end, 1, "spg_t", &struct_v, &struct_len);
4333         if (ret)
4334                 return ret;
4335
4336         ret = ceph_decode_pgid(&p, end, &m->spgid.pgid);
4337         if (ret)
4338                 return ret;
4339
4340         ceph_decode_8_safe(&p, end, m->spgid.shard, e_inval);
4341         ceph_decode_32_safe(&p, end, m->map_epoch, e_inval);
4342         ceph_decode_8_safe(&p, end, m->op, e_inval);
4343         ceph_decode_64_safe(&p, end, m->id, e_inval);
4344
4345         m->begin = kzalloc(sizeof(*m->begin), GFP_NOIO);
4346         if (!m->begin)
4347                 return -ENOMEM;
4348
4349         ret = decode_hoid(&p, end, m->begin);
4350         if (ret) {
4351                 free_hoid(m->begin);
4352                 return ret;
4353         }
4354
4355         m->end = kzalloc(sizeof(*m->end), GFP_NOIO);
4356         if (!m->end) {
4357                 free_hoid(m->begin);
4358                 return -ENOMEM;
4359         }
4360
4361         ret = decode_hoid(&p, end, m->end);
4362         if (ret) {
4363                 free_hoid(m->begin);
4364                 free_hoid(m->end);
4365                 return ret;
4366         }
4367
4368         return 0;
4369
4370 e_inval:
4371         return -EINVAL;
4372 }
4373
4374 static struct ceph_msg *create_backoff_message(
4375                                 const struct ceph_osd_backoff *backoff,
4376                                 u32 map_epoch)
4377 {
4378         struct ceph_msg *msg;
4379         void *p, *end;
4380         int msg_size;
4381
4382         msg_size = CEPH_ENCODING_START_BLK_LEN +
4383                         CEPH_PGID_ENCODING_LEN + 1; /* spgid */
4384         msg_size += 4 + 1 + 8; /* map_epoch, op, id */
4385         msg_size += CEPH_ENCODING_START_BLK_LEN +
4386                         hoid_encoding_size(backoff->begin);
4387         msg_size += CEPH_ENCODING_START_BLK_LEN +
4388                         hoid_encoding_size(backoff->end);
4389
4390         msg = ceph_msg_new(CEPH_MSG_OSD_BACKOFF, msg_size, GFP_NOIO, true);
4391         if (!msg)
4392                 return NULL;
4393
4394         p = msg->front.iov_base;
4395         end = p + msg->front_alloc_len;
4396
4397         encode_spgid(&p, &backoff->spgid);
4398         ceph_encode_32(&p, map_epoch);
4399         ceph_encode_8(&p, CEPH_OSD_BACKOFF_OP_ACK_BLOCK);
4400         ceph_encode_64(&p, backoff->id);
4401         encode_hoid(&p, end, backoff->begin);
4402         encode_hoid(&p, end, backoff->end);
4403         BUG_ON(p != end);
4404
4405         msg->front.iov_len = p - msg->front.iov_base;
4406         msg->hdr.version = cpu_to_le16(1); /* MOSDBackoff v1 */
4407         msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
4408
4409         return msg;
4410 }
4411
4412 static void handle_backoff_block(struct ceph_osd *osd, struct MOSDBackoff *m)
4413 {
4414         struct ceph_spg_mapping *spg;
4415         struct ceph_osd_backoff *backoff;
4416         struct ceph_msg *msg;
4417
4418         dout("%s osd%d spgid %llu.%xs%d id %llu\n", __func__, osd->o_osd,
4419              m->spgid.pgid.pool, m->spgid.pgid.seed, m->spgid.shard, m->id);
4420
4421         spg = lookup_spg_mapping(&osd->o_backoff_mappings, &m->spgid);
4422         if (!spg) {
4423                 spg = alloc_spg_mapping();
4424                 if (!spg) {
4425                         pr_err("%s failed to allocate spg\n", __func__);
4426                         return;
4427                 }
4428                 spg->spgid = m->spgid; /* struct */
4429                 insert_spg_mapping(&osd->o_backoff_mappings, spg);
4430         }
4431
4432         backoff = alloc_backoff();
4433         if (!backoff) {
4434                 pr_err("%s failed to allocate backoff\n", __func__);
4435                 return;
4436         }
4437         backoff->spgid = m->spgid; /* struct */
4438         backoff->id = m->id;
4439         backoff->begin = m->begin;
4440         m->begin = NULL; /* backoff now owns this */
4441         backoff->end = m->end;
4442         m->end = NULL;   /* ditto */
4443
4444         insert_backoff(&spg->backoffs, backoff);
4445         insert_backoff_by_id(&osd->o_backoffs_by_id, backoff);
4446
4447         /*
4448          * Ack with original backoff's epoch so that the OSD can
4449          * discard this if there was a PG split.
4450          */
4451         msg = create_backoff_message(backoff, m->map_epoch);
4452         if (!msg) {
4453                 pr_err("%s failed to allocate msg\n", __func__);
4454                 return;
4455         }
4456         ceph_con_send(&osd->o_con, msg);
4457 }
4458
4459 static bool target_contained_by(const struct ceph_osd_request_target *t,
4460                                 const struct ceph_hobject_id *begin,
4461                                 const struct ceph_hobject_id *end)
4462 {
4463         struct ceph_hobject_id hoid;
4464         int cmp;
4465
4466         hoid_fill_from_target(&hoid, t);
4467         cmp = hoid_compare(&hoid, begin);
4468         return !cmp || (cmp > 0 && hoid_compare(&hoid, end) < 0);
4469 }
4470
4471 static void handle_backoff_unblock(struct ceph_osd *osd,
4472                                    const struct MOSDBackoff *m)
4473 {
4474         struct ceph_spg_mapping *spg;
4475         struct ceph_osd_backoff *backoff;
4476         struct rb_node *n;
4477
4478         dout("%s osd%d spgid %llu.%xs%d id %llu\n", __func__, osd->o_osd,
4479              m->spgid.pgid.pool, m->spgid.pgid.seed, m->spgid.shard, m->id);
4480
4481         backoff = lookup_backoff_by_id(&osd->o_backoffs_by_id, m->id);
4482         if (!backoff) {
4483                 pr_err("%s osd%d spgid %llu.%xs%d id %llu backoff dne\n",
4484                        __func__, osd->o_osd, m->spgid.pgid.pool,
4485                        m->spgid.pgid.seed, m->spgid.shard, m->id);
4486                 return;
4487         }
4488
4489         if (hoid_compare(backoff->begin, m->begin) &&
4490             hoid_compare(backoff->end, m->end)) {
4491                 pr_err("%s osd%d spgid %llu.%xs%d id %llu bad range?\n",
4492                        __func__, osd->o_osd, m->spgid.pgid.pool,
4493                        m->spgid.pgid.seed, m->spgid.shard, m->id);
4494                 /* unblock it anyway... */
4495         }
4496
4497         spg = lookup_spg_mapping(&osd->o_backoff_mappings, &backoff->spgid);
4498         BUG_ON(!spg);
4499
4500         erase_backoff(&spg->backoffs, backoff);
4501         erase_backoff_by_id(&osd->o_backoffs_by_id, backoff);
4502         free_backoff(backoff);
4503
4504         if (RB_EMPTY_ROOT(&spg->backoffs)) {
4505                 erase_spg_mapping(&osd->o_backoff_mappings, spg);
4506                 free_spg_mapping(spg);
4507         }
4508
4509         for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) {
4510                 struct ceph_osd_request *req =
4511                     rb_entry(n, struct ceph_osd_request, r_node);
4512
4513                 if (!ceph_spg_compare(&req->r_t.spgid, &m->spgid)) {
4514                         /*
4515                          * Match against @m, not @backoff -- the PG may
4516                          * have split on the OSD.
4517                          */
4518                         if (target_contained_by(&req->r_t, m->begin, m->end)) {
4519                                 /*
4520                                  * If no other installed backoff applies,
4521                                  * resend.
4522                                  */
4523                                 send_request(req);
4524                         }
4525                 }
4526         }
4527 }
4528
4529 static void handle_backoff(struct ceph_osd *osd, struct ceph_msg *msg)
4530 {
4531         struct ceph_osd_client *osdc = osd->o_osdc;
4532         struct MOSDBackoff m;
4533         int ret;
4534
4535         down_read(&osdc->lock);
4536         if (!osd_registered(osd)) {
4537                 dout("%s osd%d unknown\n", __func__, osd->o_osd);
4538                 up_read(&osdc->lock);
4539                 return;
4540         }
4541         WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num));
4542
4543         mutex_lock(&osd->lock);
4544         ret = decode_MOSDBackoff(msg, &m);
4545         if (ret) {
4546                 pr_err("failed to decode MOSDBackoff: %d\n", ret);
4547                 ceph_msg_dump(msg);
4548                 goto out_unlock;
4549         }
4550
4551         switch (m.op) {
4552         case CEPH_OSD_BACKOFF_OP_BLOCK:
4553                 handle_backoff_block(osd, &m);
4554                 break;
4555         case CEPH_OSD_BACKOFF_OP_UNBLOCK:
4556                 handle_backoff_unblock(osd, &m);
4557                 break;
4558         default:
4559                 pr_err("%s osd%d unknown op %d\n", __func__, osd->o_osd, m.op);
4560         }
4561
4562         free_hoid(m.begin);
4563         free_hoid(m.end);
4564
4565 out_unlock:
4566         mutex_unlock(&osd->lock);
4567         up_read(&osdc->lock);
4568 }
4569
4570 /*
4571  * Process osd watch notifications
4572  */
4573 static void handle_watch_notify(struct ceph_osd_client *osdc,
4574                                 struct ceph_msg *msg)
4575 {
4576         void *p = msg->front.iov_base;
4577         void *const end = p + msg->front.iov_len;
4578         struct ceph_osd_linger_request *lreq;
4579         struct linger_work *lwork;
4580         u8 proto_ver, opcode;
4581         u64 cookie, notify_id;
4582         u64 notifier_id = 0;
4583         s32 return_code = 0;
4584         void *payload = NULL;
4585         u32 payload_len = 0;
4586
4587         ceph_decode_8_safe(&p, end, proto_ver, bad);
4588         ceph_decode_8_safe(&p, end, opcode, bad);
4589         ceph_decode_64_safe(&p, end, cookie, bad);
4590         p += 8; /* skip ver */
4591         ceph_decode_64_safe(&p, end, notify_id, bad);
4592
4593         if (proto_ver >= 1) {
4594                 ceph_decode_32_safe(&p, end, payload_len, bad);
4595                 ceph_decode_need(&p, end, payload_len, bad);
4596                 payload = p;
4597                 p += payload_len;
4598         }
4599
4600         if (le16_to_cpu(msg->hdr.version) >= 2)
4601                 ceph_decode_32_safe(&p, end, return_code, bad);
4602
4603         if (le16_to_cpu(msg->hdr.version) >= 3)
4604                 ceph_decode_64_safe(&p, end, notifier_id, bad);
4605
4606         down_read(&osdc->lock);
4607         lreq = lookup_linger_osdc(&osdc->linger_requests, cookie);
4608         if (!lreq) {
4609                 dout("%s opcode %d cookie %llu dne\n", __func__, opcode,
4610                      cookie);
4611                 goto out_unlock_osdc;
4612         }
4613
4614         mutex_lock(&lreq->lock);
4615         dout("%s opcode %d cookie %llu lreq %p is_watch %d\n", __func__,
4616              opcode, cookie, lreq, lreq->is_watch);
4617         if (opcode == CEPH_WATCH_EVENT_DISCONNECT) {
4618                 if (!lreq->last_error) {
4619                         lreq->last_error = -ENOTCONN;
4620                         queue_watch_error(lreq);
4621                 }
4622         } else if (!lreq->is_watch) {
4623                 /* CEPH_WATCH_EVENT_NOTIFY_COMPLETE */
4624                 if (lreq->notify_id && lreq->notify_id != notify_id) {
4625                         dout("lreq %p notify_id %llu != %llu, ignoring\n", lreq,
4626                              lreq->notify_id, notify_id);
4627                 } else if (!completion_done(&lreq->notify_finish_wait)) {
4628                         struct ceph_msg_data *data =
4629                             msg->num_data_items ? &msg->data[0] : NULL;
4630
4631                         if (data) {
4632                                 if (lreq->preply_pages) {
4633                                         WARN_ON(data->type !=
4634                                                         CEPH_MSG_DATA_PAGES);
4635                                         *lreq->preply_pages = data->pages;
4636                                         *lreq->preply_len = data->length;
4637                                         data->own_pages = false;
4638                                 }
4639                         }
4640                         lreq->notify_finish_error = return_code;
4641                         complete_all(&lreq->notify_finish_wait);
4642                 }
4643         } else {
4644                 /* CEPH_WATCH_EVENT_NOTIFY */
4645                 lwork = lwork_alloc(lreq, do_watch_notify);
4646                 if (!lwork) {
4647                         pr_err("failed to allocate notify-lwork\n");
4648                         goto out_unlock_lreq;
4649                 }
4650
4651                 lwork->notify.notify_id = notify_id;
4652                 lwork->notify.notifier_id = notifier_id;
4653                 lwork->notify.payload = payload;
4654                 lwork->notify.payload_len = payload_len;
4655                 lwork->notify.msg = ceph_msg_get(msg);
4656                 lwork_queue(lwork);
4657         }
4658
4659 out_unlock_lreq:
4660         mutex_unlock(&lreq->lock);
4661 out_unlock_osdc:
4662         up_read(&osdc->lock);
4663         return;
4664
4665 bad:
4666         pr_err("osdc handle_watch_notify corrupt msg\n");
4667 }
4668
4669 /*
4670  * Register request, send initial attempt.
4671  */
4672 void ceph_osdc_start_request(struct ceph_osd_client *osdc,
4673                              struct ceph_osd_request *req)
4674 {
4675         down_read(&osdc->lock);
4676         submit_request(req, false);
4677         up_read(&osdc->lock);
4678 }
4679 EXPORT_SYMBOL(ceph_osdc_start_request);
4680
4681 /*
4682  * Unregister request.  If @req was registered, it isn't completed:
4683  * r_result isn't set and __complete_request() isn't invoked.
4684  *
4685  * If @req wasn't registered, this call may have raced with
4686  * handle_reply(), in which case r_result would already be set and
4687  * __complete_request() would be getting invoked, possibly even
4688  * concurrently with this call.
4689  */
4690 void ceph_osdc_cancel_request(struct ceph_osd_request *req)
4691 {
4692         struct ceph_osd_client *osdc = req->r_osdc;
4693
4694         down_write(&osdc->lock);
4695         if (req->r_osd)
4696                 cancel_request(req);
4697         up_write(&osdc->lock);
4698 }
4699 EXPORT_SYMBOL(ceph_osdc_cancel_request);
4700
4701 /*
4702  * @timeout: in jiffies, 0 means "wait forever"
4703  */
4704 static int wait_request_timeout(struct ceph_osd_request *req,
4705                                 unsigned long timeout)
4706 {
4707         long left;
4708
4709         dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
4710         left = wait_for_completion_killable_timeout(&req->r_completion,
4711                                                 ceph_timeout_jiffies(timeout));
4712         if (left <= 0) {
4713                 left = left ?: -ETIMEDOUT;
4714                 ceph_osdc_cancel_request(req);
4715         } else {
4716                 left = req->r_result; /* completed */
4717         }
4718
4719         return left;
4720 }
4721
4722 /*
4723  * wait for a request to complete
4724  */
4725 int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
4726                            struct ceph_osd_request *req)
4727 {
4728         return wait_request_timeout(req, 0);
4729 }
4730 EXPORT_SYMBOL(ceph_osdc_wait_request);
4731
4732 /*
4733  * sync - wait for all in-flight requests to flush.  avoid starvation.
4734  */
4735 void ceph_osdc_sync(struct ceph_osd_client *osdc)
4736 {
4737         struct rb_node *n, *p;
4738         u64 last_tid = atomic64_read(&osdc->last_tid);
4739
4740 again:
4741         down_read(&osdc->lock);
4742         for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
4743                 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
4744
4745                 mutex_lock(&osd->lock);
4746                 for (p = rb_first(&osd->o_requests); p; p = rb_next(p)) {
4747                         struct ceph_osd_request *req =
4748                             rb_entry(p, struct ceph_osd_request, r_node);
4749
4750                         if (req->r_tid > last_tid)
4751                                 break;
4752
4753                         if (!(req->r_flags & CEPH_OSD_FLAG_WRITE))
4754                                 continue;
4755
4756                         ceph_osdc_get_request(req);
4757                         mutex_unlock(&osd->lock);
4758                         up_read(&osdc->lock);
4759                         dout("%s waiting on req %p tid %llu last_tid %llu\n",
4760                              __func__, req, req->r_tid, last_tid);
4761                         wait_for_completion(&req->r_completion);
4762                         ceph_osdc_put_request(req);
4763                         goto again;
4764                 }
4765
4766                 mutex_unlock(&osd->lock);
4767         }
4768
4769         up_read(&osdc->lock);
4770         dout("%s done last_tid %llu\n", __func__, last_tid);
4771 }
4772 EXPORT_SYMBOL(ceph_osdc_sync);
4773
4774 /*
4775  * Returns a handle, caller owns a ref.
4776  */
4777 struct ceph_osd_linger_request *
4778 ceph_osdc_watch(struct ceph_osd_client *osdc,
4779                 struct ceph_object_id *oid,
4780                 struct ceph_object_locator *oloc,
4781                 rados_watchcb2_t wcb,
4782                 rados_watcherrcb_t errcb,
4783                 void *data)
4784 {
4785         struct ceph_osd_linger_request *lreq;
4786         int ret;
4787
4788         lreq = linger_alloc(osdc);
4789         if (!lreq)
4790                 return ERR_PTR(-ENOMEM);
4791
4792         lreq->is_watch = true;
4793         lreq->wcb = wcb;
4794         lreq->errcb = errcb;
4795         lreq->data = data;
4796         lreq->watch_valid_thru = jiffies;
4797
4798         ceph_oid_copy(&lreq->t.base_oid, oid);
4799         ceph_oloc_copy(&lreq->t.base_oloc, oloc);
4800         lreq->t.flags = CEPH_OSD_FLAG_WRITE;
4801         ktime_get_real_ts64(&lreq->mtime);
4802
4803         linger_submit(lreq);
4804         ret = linger_reg_commit_wait(lreq);
4805         if (ret) {
4806                 linger_cancel(lreq);
4807                 goto err_put_lreq;
4808         }
4809
4810         return lreq;
4811
4812 err_put_lreq:
4813         linger_put(lreq);
4814         return ERR_PTR(ret);
4815 }
4816 EXPORT_SYMBOL(ceph_osdc_watch);
4817
4818 /*
4819  * Releases a ref.
4820  *
4821  * Times out after mount_timeout to preserve rbd unmap behaviour
4822  * introduced in 2894e1d76974 ("rbd: timeout watch teardown on unmap
4823  * with mount_timeout").
4824  */
4825 int ceph_osdc_unwatch(struct ceph_osd_client *osdc,
4826                       struct ceph_osd_linger_request *lreq)
4827 {
4828         struct ceph_options *opts = osdc->client->options;
4829         struct ceph_osd_request *req;
4830         int ret;
4831
4832         req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
4833         if (!req)
4834                 return -ENOMEM;
4835
4836         ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
4837         ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
4838         req->r_flags = CEPH_OSD_FLAG_WRITE;
4839         ktime_get_real_ts64(&req->r_mtime);
4840         osd_req_op_watch_init(req, 0, CEPH_OSD_WATCH_OP_UNWATCH,
4841                               lreq->linger_id, 0);
4842
4843         ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
4844         if (ret)
4845                 goto out_put_req;
4846
4847         ceph_osdc_start_request(osdc, req);
4848         linger_cancel(lreq);
4849         linger_put(lreq);
4850         ret = wait_request_timeout(req, opts->mount_timeout);
4851
4852 out_put_req:
4853         ceph_osdc_put_request(req);
4854         return ret;
4855 }
4856 EXPORT_SYMBOL(ceph_osdc_unwatch);
4857
4858 static int osd_req_op_notify_ack_init(struct ceph_osd_request *req, int which,
4859                                       u64 notify_id, u64 cookie, void *payload,
4860                                       u32 payload_len)
4861 {
4862         struct ceph_osd_req_op *op;
4863         struct ceph_pagelist *pl;
4864         int ret;
4865
4866         op = osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY_ACK, 0);
4867
4868         pl = ceph_pagelist_alloc(GFP_NOIO);
4869         if (!pl)
4870                 return -ENOMEM;
4871
4872         ret = ceph_pagelist_encode_64(pl, notify_id);
4873         ret |= ceph_pagelist_encode_64(pl, cookie);
4874         if (payload) {
4875                 ret |= ceph_pagelist_encode_32(pl, payload_len);
4876                 ret |= ceph_pagelist_append(pl, payload, payload_len);
4877         } else {
4878                 ret |= ceph_pagelist_encode_32(pl, 0);
4879         }
4880         if (ret) {
4881                 ceph_pagelist_release(pl);
4882                 return -ENOMEM;
4883         }
4884
4885         ceph_osd_data_pagelist_init(&op->notify_ack.request_data, pl);
4886         op->indata_len = pl->length;
4887         return 0;
4888 }
4889
4890 int ceph_osdc_notify_ack(struct ceph_osd_client *osdc,
4891                          struct ceph_object_id *oid,
4892                          struct ceph_object_locator *oloc,
4893                          u64 notify_id,
4894                          u64 cookie,
4895                          void *payload,
4896                          u32 payload_len)
4897 {
4898         struct ceph_osd_request *req;
4899         int ret;
4900
4901         req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
4902         if (!req)
4903                 return -ENOMEM;
4904
4905         ceph_oid_copy(&req->r_base_oid, oid);
4906         ceph_oloc_copy(&req->r_base_oloc, oloc);
4907         req->r_flags = CEPH_OSD_FLAG_READ;
4908
4909         ret = osd_req_op_notify_ack_init(req, 0, notify_id, cookie, payload,
4910                                          payload_len);
4911         if (ret)
4912                 goto out_put_req;
4913
4914         ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
4915         if (ret)
4916                 goto out_put_req;
4917
4918         ceph_osdc_start_request(osdc, req);
4919         ret = ceph_osdc_wait_request(osdc, req);
4920
4921 out_put_req:
4922         ceph_osdc_put_request(req);
4923         return ret;
4924 }
4925 EXPORT_SYMBOL(ceph_osdc_notify_ack);
4926
4927 /*
4928  * @timeout: in seconds
4929  *
4930  * @preply_{pages,len} are initialized both on success and error.
4931  * The caller is responsible for:
4932  *
4933  *     ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len))
4934  */
4935 int ceph_osdc_notify(struct ceph_osd_client *osdc,
4936                      struct ceph_object_id *oid,
4937                      struct ceph_object_locator *oloc,
4938                      void *payload,
4939                      u32 payload_len,
4940                      u32 timeout,
4941                      struct page ***preply_pages,
4942                      size_t *preply_len)
4943 {
4944         struct ceph_osd_linger_request *lreq;
4945         int ret;
4946
4947         WARN_ON(!timeout);
4948         if (preply_pages) {
4949                 *preply_pages = NULL;
4950                 *preply_len = 0;
4951         }
4952
4953         lreq = linger_alloc(osdc);
4954         if (!lreq)
4955                 return -ENOMEM;
4956
4957         lreq->request_pl = ceph_pagelist_alloc(GFP_NOIO);
4958         if (!lreq->request_pl) {
4959                 ret = -ENOMEM;
4960                 goto out_put_lreq;
4961         }
4962
4963         ret = ceph_pagelist_encode_32(lreq->request_pl, 1); /* prot_ver */
4964         ret |= ceph_pagelist_encode_32(lreq->request_pl, timeout);
4965         ret |= ceph_pagelist_encode_32(lreq->request_pl, payload_len);
4966         ret |= ceph_pagelist_append(lreq->request_pl, payload, payload_len);
4967         if (ret) {
4968                 ret = -ENOMEM;
4969                 goto out_put_lreq;
4970         }
4971
4972         /* for notify_id */
4973         lreq->notify_id_pages = ceph_alloc_page_vector(1, GFP_NOIO);
4974         if (IS_ERR(lreq->notify_id_pages)) {
4975                 ret = PTR_ERR(lreq->notify_id_pages);
4976                 lreq->notify_id_pages = NULL;
4977                 goto out_put_lreq;
4978         }
4979
4980         lreq->preply_pages = preply_pages;
4981         lreq->preply_len = preply_len;
4982
4983         ceph_oid_copy(&lreq->t.base_oid, oid);
4984         ceph_oloc_copy(&lreq->t.base_oloc, oloc);
4985         lreq->t.flags = CEPH_OSD_FLAG_READ;
4986
4987         linger_submit(lreq);
4988         ret = linger_reg_commit_wait(lreq);
4989         if (!ret)
4990                 ret = linger_notify_finish_wait(lreq,
4991                                  msecs_to_jiffies(2 * timeout * MSEC_PER_SEC));
4992         else
4993                 dout("lreq %p failed to initiate notify %d\n", lreq, ret);
4994
4995         linger_cancel(lreq);
4996 out_put_lreq:
4997         linger_put(lreq);
4998         return ret;
4999 }
5000 EXPORT_SYMBOL(ceph_osdc_notify);
5001
5002 /*
5003  * Return the number of milliseconds since the watch was last
5004  * confirmed, or an error.  If there is an error, the watch is no
5005  * longer valid, and should be destroyed with ceph_osdc_unwatch().
5006  */
5007 int ceph_osdc_watch_check(struct ceph_osd_client *osdc,
5008                           struct ceph_osd_linger_request *lreq)
5009 {
5010         unsigned long stamp, age;
5011         int ret;
5012
5013         down_read(&osdc->lock);
5014         mutex_lock(&lreq->lock);
5015         stamp = lreq->watch_valid_thru;
5016         if (!list_empty(&lreq->pending_lworks)) {
5017                 struct linger_work *lwork =
5018                     list_first_entry(&lreq->pending_lworks,
5019                                      struct linger_work,
5020                                      pending_item);
5021
5022                 if (time_before(lwork->queued_stamp, stamp))
5023                         stamp = lwork->queued_stamp;
5024         }
5025         age = jiffies - stamp;
5026         dout("%s lreq %p linger_id %llu age %lu last_error %d\n", __func__,
5027              lreq, lreq->linger_id, age, lreq->last_error);
5028         /* we are truncating to msecs, so return a safe upper bound */
5029         ret = lreq->last_error ?: 1 + jiffies_to_msecs(age);
5030
5031         mutex_unlock(&lreq->lock);
5032         up_read(&osdc->lock);
5033         return ret;
5034 }
5035
5036 static int decode_watcher(void **p, void *end, struct ceph_watch_item *item)
5037 {
5038         u8 struct_v;
5039         u32 struct_len;
5040         int ret;
5041
5042         ret = ceph_start_decoding(p, end, 2, "watch_item_t",
5043                                   &struct_v, &struct_len);
5044         if (ret)
5045                 goto bad;
5046
5047         ret = -EINVAL;
5048         ceph_decode_copy_safe(p, end, &item->name, sizeof(item->name), bad);
5049         ceph_decode_64_safe(p, end, item->cookie, bad);
5050         ceph_decode_skip_32(p, end, bad); /* skip timeout seconds */
5051
5052         if (struct_v >= 2) {
5053                 ret = ceph_decode_entity_addr(p, end, &item->addr);
5054                 if (ret)
5055                         goto bad;
5056         } else {
5057                 ret = 0;
5058         }
5059
5060         dout("%s %s%llu cookie %llu addr %s\n", __func__,
5061              ENTITY_NAME(item->name), item->cookie,
5062              ceph_pr_addr(&item->addr));
5063 bad:
5064         return ret;
5065 }
5066
5067 static int decode_watchers(void **p, void *end,
5068                            struct ceph_watch_item **watchers,
5069                            u32 *num_watchers)
5070 {
5071         u8 struct_v;
5072         u32 struct_len;
5073         int i;
5074         int ret;
5075
5076         ret = ceph_start_decoding(p, end, 1, "obj_list_watch_response_t",
5077                                   &struct_v, &struct_len);
5078         if (ret)
5079                 return ret;
5080
5081         *num_watchers = ceph_decode_32(p);
5082         *watchers = kcalloc(*num_watchers, sizeof(**watchers), GFP_NOIO);
5083         if (!*watchers)
5084                 return -ENOMEM;
5085
5086         for (i = 0; i < *num_watchers; i++) {
5087                 ret = decode_watcher(p, end, *watchers + i);
5088                 if (ret) {
5089                         kfree(*watchers);
5090                         return ret;
5091                 }
5092         }
5093
5094         return 0;
5095 }
5096
5097 /*
5098  * On success, the caller is responsible for:
5099  *
5100  *     kfree(watchers);
5101  */
5102 int ceph_osdc_list_watchers(struct ceph_osd_client *osdc,
5103                             struct ceph_object_id *oid,
5104                             struct ceph_object_locator *oloc,
5105                             struct ceph_watch_item **watchers,
5106                             u32 *num_watchers)
5107 {
5108         struct ceph_osd_request *req;
5109         struct page **pages;
5110         int ret;
5111
5112         req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
5113         if (!req)
5114                 return -ENOMEM;
5115
5116         ceph_oid_copy(&req->r_base_oid, oid);
5117         ceph_oloc_copy(&req->r_base_oloc, oloc);
5118         req->r_flags = CEPH_OSD_FLAG_READ;
5119
5120         pages = ceph_alloc_page_vector(1, GFP_NOIO);
5121         if (IS_ERR(pages)) {
5122                 ret = PTR_ERR(pages);
5123                 goto out_put_req;
5124         }
5125
5126         osd_req_op_init(req, 0, CEPH_OSD_OP_LIST_WATCHERS, 0);
5127         ceph_osd_data_pages_init(osd_req_op_data(req, 0, list_watchers,
5128                                                  response_data),
5129                                  pages, PAGE_SIZE, 0, false, true);
5130
5131         ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
5132         if (ret)
5133                 goto out_put_req;
5134
5135         ceph_osdc_start_request(osdc, req);
5136         ret = ceph_osdc_wait_request(osdc, req);
5137         if (ret >= 0) {
5138                 void *p = page_address(pages[0]);
5139                 void *const end = p + req->r_ops[0].outdata_len;
5140
5141                 ret = decode_watchers(&p, end, watchers, num_watchers);
5142         }
5143
5144 out_put_req:
5145         ceph_osdc_put_request(req);
5146         return ret;
5147 }
5148 EXPORT_SYMBOL(ceph_osdc_list_watchers);
5149
5150 /*
5151  * Call all pending notify callbacks - for use after a watch is
5152  * unregistered, to make sure no more callbacks for it will be invoked
5153  */
5154 void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
5155 {
5156         dout("%s osdc %p\n", __func__, osdc);
5157         flush_workqueue(osdc->notify_wq);
5158 }
5159 EXPORT_SYMBOL(ceph_osdc_flush_notifies);
5160
5161 void ceph_osdc_maybe_request_map(struct ceph_osd_client *osdc)
5162 {
5163         down_read(&osdc->lock);
5164         maybe_request_map(osdc);
5165         up_read(&osdc->lock);
5166 }
5167 EXPORT_SYMBOL(ceph_osdc_maybe_request_map);
5168
5169 /*
5170  * Execute an OSD class method on an object.
5171  *
5172  * @flags: CEPH_OSD_FLAG_*
5173  * @resp_len: in/out param for reply length
5174  */
5175 int ceph_osdc_call(struct ceph_osd_client *osdc,
5176                    struct ceph_object_id *oid,
5177                    struct ceph_object_locator *oloc,
5178                    const char *class, const char *method,
5179                    unsigned int flags,
5180                    struct page *req_page, size_t req_len,
5181                    struct page **resp_pages, size_t *resp_len)
5182 {
5183         struct ceph_osd_request *req;
5184         int ret;
5185
5186         if (req_len > PAGE_SIZE)
5187                 return -E2BIG;
5188
5189         req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
5190         if (!req)
5191                 return -ENOMEM;
5192
5193         ceph_oid_copy(&req->r_base_oid, oid);
5194         ceph_oloc_copy(&req->r_base_oloc, oloc);
5195         req->r_flags = flags;
5196
5197         ret = osd_req_op_cls_init(req, 0, class, method);
5198         if (ret)
5199                 goto out_put_req;
5200
5201         if (req_page)
5202                 osd_req_op_cls_request_data_pages(req, 0, &req_page, req_len,
5203                                                   0, false, false);
5204         if (resp_pages)
5205                 osd_req_op_cls_response_data_pages(req, 0, resp_pages,
5206                                                    *resp_len, 0, false, false);
5207
5208         ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
5209         if (ret)
5210                 goto out_put_req;
5211
5212         ceph_osdc_start_request(osdc, req);
5213         ret = ceph_osdc_wait_request(osdc, req);
5214         if (ret >= 0) {
5215                 ret = req->r_ops[0].rval;
5216                 if (resp_pages)
5217                         *resp_len = req->r_ops[0].outdata_len;
5218         }
5219
5220 out_put_req:
5221         ceph_osdc_put_request(req);
5222         return ret;
5223 }
5224 EXPORT_SYMBOL(ceph_osdc_call);
5225
5226 /*
5227  * reset all osd connections
5228  */
5229 void ceph_osdc_reopen_osds(struct ceph_osd_client *osdc)
5230 {
5231         struct rb_node *n;
5232
5233         down_write(&osdc->lock);
5234         for (n = rb_first(&osdc->osds); n; ) {
5235                 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
5236
5237                 n = rb_next(n);
5238                 if (!reopen_osd(osd))
5239                         kick_osd_requests(osd);
5240         }
5241         up_write(&osdc->lock);
5242 }
5243
5244 /*
5245  * init, shutdown
5246  */
5247 int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
5248 {
5249         int err;
5250
5251         dout("init\n");
5252         osdc->client = client;
5253         init_rwsem(&osdc->lock);
5254         osdc->osds = RB_ROOT;
5255         INIT_LIST_HEAD(&osdc->osd_lru);
5256         spin_lock_init(&osdc->osd_lru_lock);
5257         osd_init(&osdc->homeless_osd);
5258         osdc->homeless_osd.o_osdc = osdc;
5259         osdc->homeless_osd.o_osd = CEPH_HOMELESS_OSD;
5260         osdc->last_linger_id = CEPH_LINGER_ID_START;
5261         osdc->linger_requests = RB_ROOT;
5262         osdc->map_checks = RB_ROOT;
5263         osdc->linger_map_checks = RB_ROOT;
5264         INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
5265         INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
5266
5267         err = -ENOMEM;
5268         osdc->osdmap = ceph_osdmap_alloc();
5269         if (!osdc->osdmap)
5270                 goto out;
5271
5272         osdc->req_mempool = mempool_create_slab_pool(10,
5273                                                      ceph_osd_request_cache);
5274         if (!osdc->req_mempool)
5275                 goto out_map;
5276
5277         err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
5278                                 PAGE_SIZE, CEPH_OSD_SLAB_OPS, 10, "osd_op");
5279         if (err < 0)
5280                 goto out_mempool;
5281         err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
5282                                 PAGE_SIZE, CEPH_OSD_SLAB_OPS, 10,
5283                                 "osd_op_reply");
5284         if (err < 0)
5285                 goto out_msgpool;
5286
5287         err = -ENOMEM;
5288         osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
5289         if (!osdc->notify_wq)
5290                 goto out_msgpool_reply;
5291
5292         osdc->completion_wq = create_singlethread_workqueue("ceph-completion");
5293         if (!osdc->completion_wq)
5294                 goto out_notify_wq;
5295
5296         schedule_delayed_work(&osdc->timeout_work,
5297                               osdc->client->options->osd_keepalive_timeout);
5298         schedule_delayed_work(&osdc->osds_timeout_work,
5299             round_jiffies_relative(osdc->client->options->osd_idle_ttl));
5300
5301         return 0;
5302
5303 out_notify_wq:
5304         destroy_workqueue(osdc->notify_wq);
5305 out_msgpool_reply:
5306         ceph_msgpool_destroy(&osdc->msgpool_op_reply);
5307 out_msgpool:
5308         ceph_msgpool_destroy(&osdc->msgpool_op);
5309 out_mempool:
5310         mempool_destroy(osdc->req_mempool);
5311 out_map:
5312         ceph_osdmap_destroy(osdc->osdmap);
5313 out:
5314         return err;
5315 }
5316
5317 void ceph_osdc_stop(struct ceph_osd_client *osdc)
5318 {
5319         destroy_workqueue(osdc->completion_wq);
5320         destroy_workqueue(osdc->notify_wq);
5321         cancel_delayed_work_sync(&osdc->timeout_work);
5322         cancel_delayed_work_sync(&osdc->osds_timeout_work);
5323
5324         down_write(&osdc->lock);
5325         while (!RB_EMPTY_ROOT(&osdc->osds)) {
5326                 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
5327                                                 struct ceph_osd, o_node);
5328                 close_osd(osd);
5329         }
5330         up_write(&osdc->lock);
5331         WARN_ON(refcount_read(&osdc->homeless_osd.o_ref) != 1);
5332         osd_cleanup(&osdc->homeless_osd);
5333
5334         WARN_ON(!list_empty(&osdc->osd_lru));
5335         WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_requests));
5336         WARN_ON(!RB_EMPTY_ROOT(&osdc->map_checks));
5337         WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_map_checks));
5338         WARN_ON(atomic_read(&osdc->num_requests));
5339         WARN_ON(atomic_read(&osdc->num_homeless));
5340
5341         ceph_osdmap_destroy(osdc->osdmap);
5342         mempool_destroy(osdc->req_mempool);
5343         ceph_msgpool_destroy(&osdc->msgpool_op);
5344         ceph_msgpool_destroy(&osdc->msgpool_op_reply);
5345 }
5346
5347 int osd_req_op_copy_from_init(struct ceph_osd_request *req,
5348                               u64 src_snapid, u64 src_version,
5349                               struct ceph_object_id *src_oid,
5350                               struct ceph_object_locator *src_oloc,
5351                               u32 src_fadvise_flags,
5352                               u32 dst_fadvise_flags,
5353                               u32 truncate_seq, u64 truncate_size,
5354                               u8 copy_from_flags)
5355 {
5356         struct ceph_osd_req_op *op;
5357         struct page **pages;
5358         void *p, *end;
5359
5360         pages = ceph_alloc_page_vector(1, GFP_KERNEL);
5361         if (IS_ERR(pages))
5362                 return PTR_ERR(pages);
5363
5364         op = osd_req_op_init(req, 0, CEPH_OSD_OP_COPY_FROM2,
5365                              dst_fadvise_flags);
5366         op->copy_from.snapid = src_snapid;
5367         op->copy_from.src_version = src_version;
5368         op->copy_from.flags = copy_from_flags;
5369         op->copy_from.src_fadvise_flags = src_fadvise_flags;
5370
5371         p = page_address(pages[0]);
5372         end = p + PAGE_SIZE;
5373         ceph_encode_string(&p, end, src_oid->name, src_oid->name_len);
5374         encode_oloc(&p, end, src_oloc);
5375         ceph_encode_32(&p, truncate_seq);
5376         ceph_encode_64(&p, truncate_size);
5377         op->indata_len = PAGE_SIZE - (end - p);
5378
5379         ceph_osd_data_pages_init(&op->copy_from.osd_data, pages,
5380                                  op->indata_len, 0, false, true);
5381         return 0;
5382 }
5383 EXPORT_SYMBOL(osd_req_op_copy_from_init);
5384
5385 int __init ceph_osdc_setup(void)
5386 {
5387         size_t size = sizeof(struct ceph_osd_request) +
5388             CEPH_OSD_SLAB_OPS * sizeof(struct ceph_osd_req_op);
5389
5390         BUG_ON(ceph_osd_request_cache);
5391         ceph_osd_request_cache = kmem_cache_create("ceph_osd_request", size,
5392                                                    0, 0, NULL);
5393
5394         return ceph_osd_request_cache ? 0 : -ENOMEM;
5395 }
5396
5397 void ceph_osdc_cleanup(void)
5398 {
5399         BUG_ON(!ceph_osd_request_cache);
5400         kmem_cache_destroy(ceph_osd_request_cache);
5401         ceph_osd_request_cache = NULL;
5402 }
5403
5404 /*
5405  * handle incoming message
5406  */
5407 static void osd_dispatch(struct ceph_connection *con, struct ceph_msg *msg)
5408 {
5409         struct ceph_osd *osd = con->private;
5410         struct ceph_osd_client *osdc = osd->o_osdc;
5411         int type = le16_to_cpu(msg->hdr.type);
5412
5413         switch (type) {
5414         case CEPH_MSG_OSD_MAP:
5415                 ceph_osdc_handle_map(osdc, msg);
5416                 break;
5417         case CEPH_MSG_OSD_OPREPLY:
5418                 handle_reply(osd, msg);
5419                 break;
5420         case CEPH_MSG_OSD_BACKOFF:
5421                 handle_backoff(osd, msg);
5422                 break;
5423         case CEPH_MSG_WATCH_NOTIFY:
5424                 handle_watch_notify(osdc, msg);
5425                 break;
5426
5427         default:
5428                 pr_err("received unknown message type %d %s\n", type,
5429                        ceph_msg_type_name(type));
5430         }
5431
5432         ceph_msg_put(msg);
5433 }
5434
5435 /* How much sparse data was requested? */
5436 static u64 sparse_data_requested(struct ceph_osd_request *req)
5437 {
5438         u64 len = 0;
5439
5440         if (req->r_flags & CEPH_OSD_FLAG_READ) {
5441                 int i;
5442
5443                 for (i = 0; i < req->r_num_ops; ++i) {
5444                         struct ceph_osd_req_op *op = &req->r_ops[i];
5445
5446                         if (op->op == CEPH_OSD_OP_SPARSE_READ)
5447                                 len += op->extent.length;
5448                 }
5449         }
5450         return len;
5451 }
5452
5453 /*
5454  * Lookup and return message for incoming reply.  Don't try to do
5455  * anything about a larger than preallocated data portion of the
5456  * message at the moment - for now, just skip the message.
5457  */
5458 static struct ceph_msg *get_reply(struct ceph_connection *con,
5459                                   struct ceph_msg_header *hdr,
5460                                   int *skip)
5461 {
5462         struct ceph_osd *osd = con->private;
5463         struct ceph_osd_client *osdc = osd->o_osdc;
5464         struct ceph_msg *m = NULL;
5465         struct ceph_osd_request *req;
5466         int front_len = le32_to_cpu(hdr->front_len);
5467         int data_len = le32_to_cpu(hdr->data_len);
5468         u64 tid = le64_to_cpu(hdr->tid);
5469         u64 srlen;
5470
5471         down_read(&osdc->lock);
5472         if (!osd_registered(osd)) {
5473                 dout("%s osd%d unknown, skipping\n", __func__, osd->o_osd);
5474                 *skip = 1;
5475                 goto out_unlock_osdc;
5476         }
5477         WARN_ON(osd->o_osd != le64_to_cpu(hdr->src.num));
5478
5479         mutex_lock(&osd->lock);
5480         req = lookup_request(&osd->o_requests, tid);
5481         if (!req) {
5482                 dout("%s osd%d tid %llu unknown, skipping\n", __func__,
5483                      osd->o_osd, tid);
5484                 *skip = 1;
5485                 goto out_unlock_session;
5486         }
5487
5488         ceph_msg_revoke_incoming(req->r_reply);
5489
5490         if (front_len > req->r_reply->front_alloc_len) {
5491                 pr_warn("%s osd%d tid %llu front %d > preallocated %d\n",
5492                         __func__, osd->o_osd, req->r_tid, front_len,
5493                         req->r_reply->front_alloc_len);
5494                 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS,
5495                                  false);
5496                 if (!m)
5497                         goto out_unlock_session;
5498                 ceph_msg_put(req->r_reply);
5499                 req->r_reply = m;
5500         }
5501
5502         srlen = sparse_data_requested(req);
5503         if (!srlen && data_len > req->r_reply->data_length) {
5504                 pr_warn("%s osd%d tid %llu data %d > preallocated %zu, skipping\n",
5505                         __func__, osd->o_osd, req->r_tid, data_len,
5506                         req->r_reply->data_length);
5507                 m = NULL;
5508                 *skip = 1;
5509                 goto out_unlock_session;
5510         }
5511
5512         m = ceph_msg_get(req->r_reply);
5513         m->sparse_read = (bool)srlen;
5514
5515         dout("get_reply tid %lld %p\n", tid, m);
5516
5517 out_unlock_session:
5518         mutex_unlock(&osd->lock);
5519 out_unlock_osdc:
5520         up_read(&osdc->lock);
5521         return m;
5522 }
5523
5524 static struct ceph_msg *alloc_msg_with_page_vector(struct ceph_msg_header *hdr)
5525 {
5526         struct ceph_msg *m;
5527         int type = le16_to_cpu(hdr->type);
5528         u32 front_len = le32_to_cpu(hdr->front_len);
5529         u32 data_len = le32_to_cpu(hdr->data_len);
5530
5531         m = ceph_msg_new2(type, front_len, 1, GFP_NOIO, false);
5532         if (!m)
5533                 return NULL;
5534
5535         if (data_len) {
5536                 struct page **pages;
5537
5538                 pages = ceph_alloc_page_vector(calc_pages_for(0, data_len),
5539                                                GFP_NOIO);
5540                 if (IS_ERR(pages)) {
5541                         ceph_msg_put(m);
5542                         return NULL;
5543                 }
5544
5545                 ceph_msg_data_add_pages(m, pages, data_len, 0, true);
5546         }
5547
5548         return m;
5549 }
5550
5551 static struct ceph_msg *osd_alloc_msg(struct ceph_connection *con,
5552                                       struct ceph_msg_header *hdr,
5553                                       int *skip)
5554 {
5555         struct ceph_osd *osd = con->private;
5556         int type = le16_to_cpu(hdr->type);
5557
5558         *skip = 0;
5559         switch (type) {
5560         case CEPH_MSG_OSD_MAP:
5561         case CEPH_MSG_OSD_BACKOFF:
5562         case CEPH_MSG_WATCH_NOTIFY:
5563                 return alloc_msg_with_page_vector(hdr);
5564         case CEPH_MSG_OSD_OPREPLY:
5565                 return get_reply(con, hdr, skip);
5566         default:
5567                 pr_warn("%s osd%d unknown msg type %d, skipping\n", __func__,
5568                         osd->o_osd, type);
5569                 *skip = 1;
5570                 return NULL;
5571         }
5572 }
5573
5574 /*
5575  * Wrappers to refcount containing ceph_osd struct
5576  */
5577 static struct ceph_connection *osd_get_con(struct ceph_connection *con)
5578 {
5579         struct ceph_osd *osd = con->private;
5580         if (get_osd(osd))
5581                 return con;
5582         return NULL;
5583 }
5584
5585 static void osd_put_con(struct ceph_connection *con)
5586 {
5587         struct ceph_osd *osd = con->private;
5588         put_osd(osd);
5589 }
5590
5591 /*
5592  * authentication
5593  */
5594
5595 /*
5596  * Note: returned pointer is the address of a structure that's
5597  * managed separately.  Caller must *not* attempt to free it.
5598  */
5599 static struct ceph_auth_handshake *
5600 osd_get_authorizer(struct ceph_connection *con, int *proto, int force_new)
5601 {
5602         struct ceph_osd *o = con->private;
5603         struct ceph_osd_client *osdc = o->o_osdc;
5604         struct ceph_auth_client *ac = osdc->client->monc.auth;
5605         struct ceph_auth_handshake *auth = &o->o_auth;
5606         int ret;
5607
5608         ret = __ceph_auth_get_authorizer(ac, auth, CEPH_ENTITY_TYPE_OSD,
5609                                          force_new, proto, NULL, NULL);
5610         if (ret)
5611                 return ERR_PTR(ret);
5612
5613         return auth;
5614 }
5615
5616 static int osd_add_authorizer_challenge(struct ceph_connection *con,
5617                                     void *challenge_buf, int challenge_buf_len)
5618 {
5619         struct ceph_osd *o = con->private;
5620         struct ceph_osd_client *osdc = o->o_osdc;
5621         struct ceph_auth_client *ac = osdc->client->monc.auth;
5622
5623         return ceph_auth_add_authorizer_challenge(ac, o->o_auth.authorizer,
5624                                             challenge_buf, challenge_buf_len);
5625 }
5626
5627 static int osd_verify_authorizer_reply(struct ceph_connection *con)
5628 {
5629         struct ceph_osd *o = con->private;
5630         struct ceph_osd_client *osdc = o->o_osdc;
5631         struct ceph_auth_client *ac = osdc->client->monc.auth;
5632         struct ceph_auth_handshake *auth = &o->o_auth;
5633
5634         return ceph_auth_verify_authorizer_reply(ac, auth->authorizer,
5635                 auth->authorizer_reply_buf, auth->authorizer_reply_buf_len,
5636                 NULL, NULL, NULL, NULL);
5637 }
5638
5639 static int osd_invalidate_authorizer(struct ceph_connection *con)
5640 {
5641         struct ceph_osd *o = con->private;
5642         struct ceph_osd_client *osdc = o->o_osdc;
5643         struct ceph_auth_client *ac = osdc->client->monc.auth;
5644
5645         ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
5646         return ceph_monc_validate_auth(&osdc->client->monc);
5647 }
5648
5649 static int osd_get_auth_request(struct ceph_connection *con,
5650                                 void *buf, int *buf_len,
5651                                 void **authorizer, int *authorizer_len)
5652 {
5653         struct ceph_osd *o = con->private;
5654         struct ceph_auth_client *ac = o->o_osdc->client->monc.auth;
5655         struct ceph_auth_handshake *auth = &o->o_auth;
5656         int ret;
5657
5658         ret = ceph_auth_get_authorizer(ac, auth, CEPH_ENTITY_TYPE_OSD,
5659                                        buf, buf_len);
5660         if (ret)
5661                 return ret;
5662
5663         *authorizer = auth->authorizer_buf;
5664         *authorizer_len = auth->authorizer_buf_len;
5665         return 0;
5666 }
5667
5668 static int osd_handle_auth_reply_more(struct ceph_connection *con,
5669                                       void *reply, int reply_len,
5670                                       void *buf, int *buf_len,
5671                                       void **authorizer, int *authorizer_len)
5672 {
5673         struct ceph_osd *o = con->private;
5674         struct ceph_auth_client *ac = o->o_osdc->client->monc.auth;
5675         struct ceph_auth_handshake *auth = &o->o_auth;
5676         int ret;
5677
5678         ret = ceph_auth_handle_svc_reply_more(ac, auth, reply, reply_len,
5679                                               buf, buf_len);
5680         if (ret)
5681                 return ret;
5682
5683         *authorizer = auth->authorizer_buf;
5684         *authorizer_len = auth->authorizer_buf_len;
5685         return 0;
5686 }
5687
5688 static int osd_handle_auth_done(struct ceph_connection *con,
5689                                 u64 global_id, void *reply, int reply_len,
5690                                 u8 *session_key, int *session_key_len,
5691                                 u8 *con_secret, int *con_secret_len)
5692 {
5693         struct ceph_osd *o = con->private;
5694         struct ceph_auth_client *ac = o->o_osdc->client->monc.auth;
5695         struct ceph_auth_handshake *auth = &o->o_auth;
5696
5697         return ceph_auth_handle_svc_reply_done(ac, auth, reply, reply_len,
5698                                                session_key, session_key_len,
5699                                                con_secret, con_secret_len);
5700 }
5701
5702 static int osd_handle_auth_bad_method(struct ceph_connection *con,
5703                                       int used_proto, int result,
5704                                       const int *allowed_protos, int proto_cnt,
5705                                       const int *allowed_modes, int mode_cnt)
5706 {
5707         struct ceph_osd *o = con->private;
5708         struct ceph_mon_client *monc = &o->o_osdc->client->monc;
5709         int ret;
5710
5711         if (ceph_auth_handle_bad_authorizer(monc->auth, CEPH_ENTITY_TYPE_OSD,
5712                                             used_proto, result,
5713                                             allowed_protos, proto_cnt,
5714                                             allowed_modes, mode_cnt)) {
5715                 ret = ceph_monc_validate_auth(monc);
5716                 if (ret)
5717                         return ret;
5718         }
5719
5720         return -EACCES;
5721 }
5722
5723 static void osd_reencode_message(struct ceph_msg *msg)
5724 {
5725         int type = le16_to_cpu(msg->hdr.type);
5726
5727         if (type == CEPH_MSG_OSD_OP)
5728                 encode_request_finish(msg);
5729 }
5730
5731 static int osd_sign_message(struct ceph_msg *msg)
5732 {
5733         struct ceph_osd *o = msg->con->private;
5734         struct ceph_auth_handshake *auth = &o->o_auth;
5735
5736         return ceph_auth_sign_message(auth, msg);
5737 }
5738
5739 static int osd_check_message_signature(struct ceph_msg *msg)
5740 {
5741         struct ceph_osd *o = msg->con->private;
5742         struct ceph_auth_handshake *auth = &o->o_auth;
5743
5744         return ceph_auth_check_message_signature(auth, msg);
5745 }
5746
5747 static void advance_cursor(struct ceph_msg_data_cursor *cursor, size_t len,
5748                            bool zero)
5749 {
5750         while (len) {
5751                 struct page *page;
5752                 size_t poff, plen;
5753
5754                 page = ceph_msg_data_next(cursor, &poff, &plen);
5755                 if (plen > len)
5756                         plen = len;
5757                 if (zero)
5758                         zero_user_segment(page, poff, poff + plen);
5759                 len -= plen;
5760                 ceph_msg_data_advance(cursor, plen);
5761         }
5762 }
5763
5764 static int prep_next_sparse_read(struct ceph_connection *con,
5765                                  struct ceph_msg_data_cursor *cursor)
5766 {
5767         struct ceph_osd *o = con->private;
5768         struct ceph_sparse_read *sr = &o->o_sparse_read;
5769         struct ceph_osd_request *req;
5770         struct ceph_osd_req_op *op;
5771
5772         spin_lock(&o->o_requests_lock);
5773         req = lookup_request(&o->o_requests, le64_to_cpu(con->in_msg->hdr.tid));
5774         if (!req) {
5775                 spin_unlock(&o->o_requests_lock);
5776                 return -EBADR;
5777         }
5778
5779         if (o->o_sparse_op_idx < 0) {
5780                 u64 srlen = sparse_data_requested(req);
5781
5782                 dout("%s: [%d] starting new sparse read req. srlen=0x%llx\n",
5783                      __func__, o->o_osd, srlen);
5784                 ceph_msg_data_cursor_init(cursor, con->in_msg, srlen);
5785         } else {
5786                 u64 end;
5787
5788                 op = &req->r_ops[o->o_sparse_op_idx];
5789
5790                 WARN_ON_ONCE(op->extent.sparse_ext);
5791
5792                 /* hand back buffer we took earlier */
5793                 op->extent.sparse_ext = sr->sr_extent;
5794                 sr->sr_extent = NULL;
5795                 op->extent.sparse_ext_cnt = sr->sr_count;
5796                 sr->sr_ext_len = 0;
5797                 dout("%s: [%d] completed extent array len %d cursor->resid %zd\n",
5798                      __func__, o->o_osd, op->extent.sparse_ext_cnt, cursor->resid);
5799                 /* Advance to end of data for this operation */
5800                 end = ceph_sparse_ext_map_end(op);
5801                 if (end < sr->sr_req_len)
5802                         advance_cursor(cursor, sr->sr_req_len - end, false);
5803         }
5804
5805         ceph_init_sparse_read(sr);
5806
5807         /* find next op in this request (if any) */
5808         while (++o->o_sparse_op_idx < req->r_num_ops) {
5809                 op = &req->r_ops[o->o_sparse_op_idx];
5810                 if (op->op == CEPH_OSD_OP_SPARSE_READ)
5811                         goto found;
5812         }
5813
5814         /* reset for next sparse read request */
5815         spin_unlock(&o->o_requests_lock);
5816         o->o_sparse_op_idx = -1;
5817         return 0;
5818 found:
5819         sr->sr_req_off = op->extent.offset;
5820         sr->sr_req_len = op->extent.length;
5821         sr->sr_pos = sr->sr_req_off;
5822         dout("%s: [%d] new sparse read op at idx %d 0x%llx~0x%llx\n", __func__,
5823              o->o_osd, o->o_sparse_op_idx, sr->sr_req_off, sr->sr_req_len);
5824
5825         /* hand off request's sparse extent map buffer */
5826         sr->sr_ext_len = op->extent.sparse_ext_cnt;
5827         op->extent.sparse_ext_cnt = 0;
5828         sr->sr_extent = op->extent.sparse_ext;
5829         op->extent.sparse_ext = NULL;
5830
5831         spin_unlock(&o->o_requests_lock);
5832         return 1;
5833 }
5834
5835 #ifdef __BIG_ENDIAN
5836 static inline void convert_extent_map(struct ceph_sparse_read *sr)
5837 {
5838         int i;
5839
5840         for (i = 0; i < sr->sr_count; i++) {
5841                 struct ceph_sparse_extent *ext = &sr->sr_extent[i];
5842
5843                 ext->off = le64_to_cpu((__force __le64)ext->off);
5844                 ext->len = le64_to_cpu((__force __le64)ext->len);
5845         }
5846 }
5847 #else
5848 static inline void convert_extent_map(struct ceph_sparse_read *sr)
5849 {
5850 }
5851 #endif
5852
5853 #define MAX_EXTENTS 4096
5854
5855 static int osd_sparse_read(struct ceph_connection *con,
5856                            struct ceph_msg_data_cursor *cursor,
5857                            char **pbuf)
5858 {
5859         struct ceph_osd *o = con->private;
5860         struct ceph_sparse_read *sr = &o->o_sparse_read;
5861         u32 count = sr->sr_count;
5862         u64 eoff, elen;
5863         int ret;
5864
5865         switch (sr->sr_state) {
5866         case CEPH_SPARSE_READ_HDR:
5867 next_op:
5868                 ret = prep_next_sparse_read(con, cursor);
5869                 if (ret <= 0)
5870                         return ret;
5871
5872                 /* number of extents */
5873                 ret = sizeof(sr->sr_count);
5874                 *pbuf = (char *)&sr->sr_count;
5875                 sr->sr_state = CEPH_SPARSE_READ_EXTENTS;
5876                 break;
5877         case CEPH_SPARSE_READ_EXTENTS:
5878                 /* Convert sr_count to host-endian */
5879                 count = le32_to_cpu((__force __le32)sr->sr_count);
5880                 sr->sr_count = count;
5881                 dout("[%d] got %u extents\n", o->o_osd, count);
5882
5883                 if (count > 0) {
5884                         if (!sr->sr_extent || count > sr->sr_ext_len) {
5885                                 /*
5886                                  * Apply a hard cap to the number of extents.
5887                                  * If we have more, assume something is wrong.
5888                                  */
5889                                 if (count > MAX_EXTENTS) {
5890                                         dout("%s: OSD returned 0x%x extents in a single reply!\n",
5891                                              __func__, count);
5892                                         return -EREMOTEIO;
5893                                 }
5894
5895                                 /* no extent array provided, or too short */
5896                                 kfree(sr->sr_extent);
5897                                 sr->sr_extent = kmalloc_array(count,
5898                                                               sizeof(*sr->sr_extent),
5899                                                               GFP_NOIO);
5900                                 if (!sr->sr_extent)
5901                                         return -ENOMEM;
5902                                 sr->sr_ext_len = count;
5903                         }
5904                         ret = count * sizeof(*sr->sr_extent);
5905                         *pbuf = (char *)sr->sr_extent;
5906                         sr->sr_state = CEPH_SPARSE_READ_DATA_LEN;
5907                         break;
5908                 }
5909                 /* No extents? Read data len */
5910                 fallthrough;
5911         case CEPH_SPARSE_READ_DATA_LEN:
5912                 convert_extent_map(sr);
5913                 ret = sizeof(sr->sr_datalen);
5914                 *pbuf = (char *)&sr->sr_datalen;
5915                 sr->sr_state = CEPH_SPARSE_READ_DATA;
5916                 break;
5917         case CEPH_SPARSE_READ_DATA:
5918                 if (sr->sr_index >= count) {
5919                         sr->sr_state = CEPH_SPARSE_READ_HDR;
5920                         goto next_op;
5921                 }
5922
5923                 eoff = sr->sr_extent[sr->sr_index].off;
5924                 elen = sr->sr_extent[sr->sr_index].len;
5925
5926                 dout("[%d] ext %d off 0x%llx len 0x%llx\n",
5927                      o->o_osd, sr->sr_index, eoff, elen);
5928
5929                 if (elen > INT_MAX) {
5930                         dout("Sparse read extent length too long (0x%llx)\n",
5931                              elen);
5932                         return -EREMOTEIO;
5933                 }
5934
5935                 /* zero out anything from sr_pos to start of extent */
5936                 if (sr->sr_pos < eoff)
5937                         advance_cursor(cursor, eoff - sr->sr_pos, true);
5938
5939                 /* Set position to end of extent */
5940                 sr->sr_pos = eoff + elen;
5941
5942                 /* send back the new length and nullify the ptr */
5943                 cursor->sr_resid = elen;
5944                 ret = elen;
5945                 *pbuf = NULL;
5946
5947                 /* Bump the array index */
5948                 ++sr->sr_index;
5949                 break;
5950         }
5951         return ret;
5952 }
5953
5954 static const struct ceph_connection_operations osd_con_ops = {
5955         .get = osd_get_con,
5956         .put = osd_put_con,
5957         .sparse_read = osd_sparse_read,
5958         .alloc_msg = osd_alloc_msg,
5959         .dispatch = osd_dispatch,
5960         .fault = osd_fault,
5961         .reencode_message = osd_reencode_message,
5962         .get_authorizer = osd_get_authorizer,
5963         .add_authorizer_challenge = osd_add_authorizer_challenge,
5964         .verify_authorizer_reply = osd_verify_authorizer_reply,
5965         .invalidate_authorizer = osd_invalidate_authorizer,
5966         .sign_message = osd_sign_message,
5967         .check_message_signature = osd_check_message_signature,
5968         .get_auth_request = osd_get_auth_request,
5969         .handle_auth_reply_more = osd_handle_auth_reply_more,
5970         .handle_auth_done = osd_handle_auth_done,
5971         .handle_auth_bad_method = osd_handle_auth_bad_method,
5972 };