net/ceph/osd_client.c

   1
   2 #include <linux/ceph/ceph_debug.h>
   3
   4 #include <linux/module.h>
   5 #include <linux/err.h>
   6 #include <linux/highmem.h>
   7 #include <linux/mm.h>
   8 #include <linux/pagemap.h>
   9 #include <linux/slab.h>
  10 #include <linux/uaccess.h>
  11 #ifdef CONFIG_BLOCK
  12 #include <linux/bio.h>
  13 #endif
  14
  15 #include <linux/ceph/libceph.h>
  16 #include <linux/ceph/osd_client.h>
  17 #include <linux/ceph/messenger.h>
  18 #include <linux/ceph/decode.h>
  19 #include <linux/ceph/auth.h>
  20 #include <linux/ceph/pagelist.h>
  21
  22 #define OSD_OP_FRONT_LEN        4096
  23 #define OSD_OPREPLY_FRONT_LEN   512
  24
  25 static struct kmem_cache        *ceph_osd_request_cache;
  26
  27 static const struct ceph_connection_operations osd_con_ops;
  28
  29 static void __send_queued(struct ceph_osd_client *osdc);
  30 static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd);
  31 static void __register_request(struct ceph_osd_client *osdc,
  32                                struct ceph_osd_request *req);
  33 static void __unregister_request(struct ceph_osd_client *osdc,
  34                                  struct ceph_osd_request *req);
  35 static void __unregister_linger_request(struct ceph_osd_client *osdc,
  36                                         struct ceph_osd_request *req);
  37 static void __enqueue_request(struct ceph_osd_request *req);
  38 static void __send_request(struct ceph_osd_client *osdc,
  39                            struct ceph_osd_request *req);
  40
  41 /*
  42  * Implement client access to distributed object storage cluster.
  43  *
  44  * All data objects are stored within a cluster/cloud of OSDs, or
  45  * "object storage devices."  (Note that Ceph OSDs have _nothing_ to
  46  * do with the T10 OSD extensions to SCSI.)  Ceph OSDs are simply
  47  * remote daemons serving up and coordinating consistent and safe
  48  * access to storage.
  49  *
  50  * Cluster membership and the mapping of data objects onto storage devices
  51  * are described by the osd map.
  52  *
  53  * We keep track of pending OSD requests (read, write), resubmit
  54  * requests to different OSDs when the cluster topology/data layout
  55  * change, or retry the affected requests when the communications
  56  * channel with an OSD is reset.
  57  */
  58
  59 /*
  60  * calculate the mapping of a file extent onto an object, and fill out the
  61  * request accordingly.  shorten extent as necessary if it crosses an
  62  * object boundary.
  63  *
  64  * fill osd op in request message.
  65  */
  66 static int calc_layout(struct ceph_file_layout *layout, u64 off, u64 *plen,
  67                         u64 *objnum, u64 *objoff, u64 *objlen)
  68 {
  69         u64 orig_len = *plen;
  70         int r;
  71
  72         /* object extent? */
  73         r = ceph_calc_file_object_mapping(layout, off, orig_len, objnum,
  74                                           objoff, objlen);
  75         if (r < 0)
  76                 return r;
  77         if (*objlen < orig_len) {
  78                 *plen = *objlen;
  79                 dout(" skipping last %llu, final file extent %llu~%llu\n",
  80                      orig_len - *plen, off, *plen);
  81         }
  82
  83         dout("calc_layout objnum=%llx %llu~%llu\n", *objnum, *objoff, *objlen);
  84
  85         return 0;
  86 }
  87
  88 static void ceph_osd_data_init(struct ceph_osd_data *osd_data)
  89 {
  90         memset(osd_data, 0, sizeof (*osd_data));
  91         osd_data->type = CEPH_OSD_DATA_TYPE_NONE;
  92 }
  93
  94 static void ceph_osd_data_pages_init(struct ceph_osd_data *osd_data,
  95                         struct page **pages, u64 length, u32 alignment,
  96                         bool pages_from_pool, bool own_pages)
  97 {
  98         osd_data->type = CEPH_OSD_DATA_TYPE_PAGES;
  99         osd_data->pages = pages;
 100         osd_data->length = length;
 101         osd_data->alignment = alignment;
 102         osd_data->pages_from_pool = pages_from_pool;
 103         osd_data->own_pages = own_pages;
 104 }
 105
 106 static void ceph_osd_data_pagelist_init(struct ceph_osd_data *osd_data,
 107                         struct ceph_pagelist *pagelist)
 108 {
 109         osd_data->type = CEPH_OSD_DATA_TYPE_PAGELIST;
 110         osd_data->pagelist = pagelist;
 111 }
 112
 113 #ifdef CONFIG_BLOCK
 114 static void ceph_osd_data_bio_init(struct ceph_osd_data *osd_data,
 115                         struct bio *bio, size_t bio_length)
 116 {
 117         osd_data->type = CEPH_OSD_DATA_TYPE_BIO;
 118         osd_data->bio = bio;
 119         osd_data->bio_length = bio_length;
 120 }
 121 #endif /* CONFIG_BLOCK */
 122
 123 #define osd_req_op_data(oreq, whch, typ, fld)   \
 124         ({                                              \
 125                 BUG_ON(whch >= (oreq)->r_num_ops);      \
 126                 &(oreq)->r_ops[whch].typ.fld;           \
 127         })
 128
 129 static struct ceph_osd_data *
 130 osd_req_op_raw_data_in(struct ceph_osd_request *osd_req, unsigned int which)
 131 {
 132         BUG_ON(which >= osd_req->r_num_ops);
 133
 134         return &osd_req->r_ops[which].raw_data_in;
 135 }
 136
 137 struct ceph_osd_data *
 138 osd_req_op_extent_osd_data(struct ceph_osd_request *osd_req,
 139                         unsigned int which)
 140 {
 141         return osd_req_op_data(osd_req, which, extent, osd_data);
 142 }
 143 EXPORT_SYMBOL(osd_req_op_extent_osd_data);
 144
 145 struct ceph_osd_data *
 146 osd_req_op_cls_response_data(struct ceph_osd_request *osd_req,
 147                         unsigned int which)
 148 {
 149         return osd_req_op_data(osd_req, which, cls, response_data);
 150 }
 151 EXPORT_SYMBOL(osd_req_op_cls_response_data);    /* ??? */
 152
 153 void osd_req_op_raw_data_in_pages(struct ceph_osd_request *osd_req,
 154                         unsigned int which, struct page **pages,
 155                         u64 length, u32 alignment,
 156                         bool pages_from_pool, bool own_pages)
 157 {
 158         struct ceph_osd_data *osd_data;
 159
 160         osd_data = osd_req_op_raw_data_in(osd_req, which);
 161         ceph_osd_data_pages_init(osd_data, pages, length, alignment,
 162                                 pages_from_pool, own_pages);
 163 }
 164 EXPORT_SYMBOL(osd_req_op_raw_data_in_pages);
 165
 166 void osd_req_op_extent_osd_data_pages(struct ceph_osd_request *osd_req,
 167                         unsigned int which, struct page **pages,
 168                         u64 length, u32 alignment,
 169                         bool pages_from_pool, bool own_pages)
 170 {
 171         struct ceph_osd_data *osd_data;
 172
 173         osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
 174         ceph_osd_data_pages_init(osd_data, pages, length, alignment,
 175                                 pages_from_pool, own_pages);
 176 }
 177 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pages);
 178
 179 void osd_req_op_extent_osd_data_pagelist(struct ceph_osd_request *osd_req,
 180                         unsigned int which, struct ceph_pagelist *pagelist)
 181 {
 182         struct ceph_osd_data *osd_data;
 183
 184         osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
 185         ceph_osd_data_pagelist_init(osd_data, pagelist);
 186 }
 187 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pagelist);
 188
 189 #ifdef CONFIG_BLOCK
 190 void osd_req_op_extent_osd_data_bio(struct ceph_osd_request *osd_req,
 191                         unsigned int which, struct bio *bio, size_t bio_length)
 192 {
 193         struct ceph_osd_data *osd_data;
 194
 195         osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
 196         ceph_osd_data_bio_init(osd_data, bio, bio_length);
 197 }
 198 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bio);
 199 #endif /* CONFIG_BLOCK */
 200
 201 static void osd_req_op_cls_request_info_pagelist(
 202                         struct ceph_osd_request *osd_req,
 203                         unsigned int which, struct ceph_pagelist *pagelist)
 204 {
 205         struct ceph_osd_data *osd_data;
 206
 207         osd_data = osd_req_op_data(osd_req, which, cls, request_info);
 208         ceph_osd_data_pagelist_init(osd_data, pagelist);
 209 }
 210
 211 void osd_req_op_cls_request_data_pagelist(
 212                         struct ceph_osd_request *osd_req,
 213                         unsigned int which, struct ceph_pagelist *pagelist)
 214 {
 215         struct ceph_osd_data *osd_data;
 216
 217         osd_data = osd_req_op_data(osd_req, which, cls, request_data);
 218         ceph_osd_data_pagelist_init(osd_data, pagelist);
 219 }
 220 EXPORT_SYMBOL(osd_req_op_cls_request_data_pagelist);
 221
 222 void osd_req_op_cls_request_data_pages(struct ceph_osd_request *osd_req,
 223                         unsigned int which, struct page **pages, u64 length,
 224                         u32 alignment, bool pages_from_pool, bool own_pages)
 225 {
 226         struct ceph_osd_data *osd_data;
 227
 228         osd_data = osd_req_op_data(osd_req, which, cls, request_data);
 229         ceph_osd_data_pages_init(osd_data, pages, length, alignment,
 230                                 pages_from_pool, own_pages);
 231 }
 232 EXPORT_SYMBOL(osd_req_op_cls_request_data_pages);
 233
 234 void osd_req_op_cls_response_data_pages(struct ceph_osd_request *osd_req,
 235                         unsigned int which, struct page **pages, u64 length,
 236                         u32 alignment, bool pages_from_pool, bool own_pages)
 237 {
 238         struct ceph_osd_data *osd_data;
 239
 240         osd_data = osd_req_op_data(osd_req, which, cls, response_data);
 241         ceph_osd_data_pages_init(osd_data, pages, length, alignment,
 242                                 pages_from_pool, own_pages);
 243 }
 244 EXPORT_SYMBOL(osd_req_op_cls_response_data_pages);
 245
 246 static u64 ceph_osd_data_length(struct ceph_osd_data *osd_data)
 247 {
 248         switch (osd_data->type) {
 249         case CEPH_OSD_DATA_TYPE_NONE:
 250                 return 0;
 251         case CEPH_OSD_DATA_TYPE_PAGES:
 252                 return osd_data->length;
 253         case CEPH_OSD_DATA_TYPE_PAGELIST:
 254                 return (u64)osd_data->pagelist->length;
 255 #ifdef CONFIG_BLOCK
 256         case CEPH_OSD_DATA_TYPE_BIO:
 257                 return (u64)osd_data->bio_length;
 258 #endif /* CONFIG_BLOCK */
 259         default:
 260                 WARN(true, "unrecognized data type %d\n", (int)osd_data->type);
 261                 return 0;
 262         }
 263 }
 264
 265 static void ceph_osd_data_release(struct ceph_osd_data *osd_data)
 266 {
 267         if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES && osd_data->own_pages) {
 268                 int num_pages;
 269
 270                 num_pages = calc_pages_for((u64)osd_data->alignment,
 271                                                 (u64)osd_data->length);
 272                 ceph_release_page_vector(osd_data->pages, num_pages);
 273         }
 274         ceph_osd_data_init(osd_data);
 275 }
 276
 277 static void osd_req_op_data_release(struct ceph_osd_request *osd_req,
 278                         unsigned int which)
 279 {
 280         struct ceph_osd_req_op *op;
 281
 282         BUG_ON(which >= osd_req->r_num_ops);
 283         op = &osd_req->r_ops[which];
 284
 285         switch (op->op) {
 286         case CEPH_OSD_OP_READ:
 287         case CEPH_OSD_OP_WRITE:
 288         case CEPH_OSD_OP_WRITEFULL:
 289                 ceph_osd_data_release(&op->extent.osd_data);
 290                 break;
 291         case CEPH_OSD_OP_CALL:
 292                 ceph_osd_data_release(&op->cls.request_info);
 293                 ceph_osd_data_release(&op->cls.request_data);
 294                 ceph_osd_data_release(&op->cls.response_data);
 295                 break;
 296         case CEPH_OSD_OP_SETXATTR:
 297         case CEPH_OSD_OP_CMPXATTR:
 298                 ceph_osd_data_release(&op->xattr.osd_data);
 299                 break;
 300         case CEPH_OSD_OP_STAT:
 301                 ceph_osd_data_release(&op->raw_data_in);
 302                 break;
 303         default:
 304                 break;
 305         }
 306 }
 307
 308 /*
 309  * requests
 310  */
 311 static void ceph_osdc_release_request(struct kref *kref)
 312 {
 313         struct ceph_osd_request *req = container_of(kref,
 314                                             struct ceph_osd_request, r_kref);
 315         unsigned int which;
 316
 317         dout("%s %p (r_request %p r_reply %p)\n", __func__, req,
 318              req->r_request, req->r_reply);
 319         WARN_ON(!RB_EMPTY_NODE(&req->r_node));
 320         WARN_ON(!list_empty(&req->r_req_lru_item));
 321         WARN_ON(!list_empty(&req->r_osd_item));
 322         WARN_ON(!list_empty(&req->r_linger_item));
 323         WARN_ON(!list_empty(&req->r_linger_osd_item));
 324         WARN_ON(req->r_osd);
 325
 326         if (req->r_request)
 327                 ceph_msg_put(req->r_request);
 328         if (req->r_reply) {
 329                 ceph_msg_revoke_incoming(req->r_reply);
 330                 ceph_msg_put(req->r_reply);
 331         }
 332
 333         for (which = 0; which < req->r_num_ops; which++)
 334                 osd_req_op_data_release(req, which);
 335
 336         ceph_put_snap_context(req->r_snapc);
 337         if (req->r_mempool)
 338                 mempool_free(req, req->r_osdc->req_mempool);
 339         else
 340                 kmem_cache_free(ceph_osd_request_cache, req);
 341
 342 }
 343
 344 void ceph_osdc_get_request(struct ceph_osd_request *req)
 345 {
 346         dout("%s %p (was %d)\n", __func__, req,
 347              atomic_read(&req->r_kref.refcount));
 348         kref_get(&req->r_kref);
 349 }
 350 EXPORT_SYMBOL(ceph_osdc_get_request);
 351
 352 void ceph_osdc_put_request(struct ceph_osd_request *req)
 353 {
 354         dout("%s %p (was %d)\n", __func__, req,
 355              atomic_read(&req->r_kref.refcount));
 356         kref_put(&req->r_kref, ceph_osdc_release_request);
 357 }
 358 EXPORT_SYMBOL(ceph_osdc_put_request);
 359
 360 struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
 361                                                struct ceph_snap_context *snapc,
 362                                                unsigned int num_ops,
 363                                                bool use_mempool,
 364                                                gfp_t gfp_flags)
 365 {
 366         struct ceph_osd_request *req;
 367         struct ceph_msg *msg;
 368         size_t msg_size;
 369
 370         BUILD_BUG_ON(CEPH_OSD_MAX_OP > U16_MAX);
 371         BUG_ON(num_ops > CEPH_OSD_MAX_OP);
 372
 373         msg_size = 4 + 4 + 8 + 8 + 4+8;
 374         msg_size += 2 + 4 + 8 + 4 + 4; /* oloc */
 375         msg_size += 1 + 8 + 4 + 4;     /* pg_t */
 376         msg_size += 4 + CEPH_MAX_OID_NAME_LEN; /* oid */
 377         msg_size += 2 + num_ops*sizeof(struct ceph_osd_op);
 378         msg_size += 8;  /* snapid */
 379         msg_size += 8;  /* snap_seq */
 380         msg_size += 8 * (snapc ? snapc->num_snaps : 0);  /* snaps */
 381         msg_size += 4;
 382
 383         if (use_mempool) {
 384                 req = mempool_alloc(osdc->req_mempool, gfp_flags);
 385                 memset(req, 0, sizeof(*req));
 386         } else {
 387                 req = kmem_cache_zalloc(ceph_osd_request_cache, gfp_flags);
 388         }
 389         if (req == NULL)
 390                 return NULL;
 391
 392         req->r_osdc = osdc;
 393         req->r_mempool = use_mempool;
 394         req->r_num_ops = num_ops;
 395
 396         kref_init(&req->r_kref);
 397         init_completion(&req->r_completion);
 398         init_completion(&req->r_safe_completion);
 399         RB_CLEAR_NODE(&req->r_node);
 400         INIT_LIST_HEAD(&req->r_unsafe_item);
 401         INIT_LIST_HEAD(&req->r_linger_item);
 402         INIT_LIST_HEAD(&req->r_linger_osd_item);
 403         INIT_LIST_HEAD(&req->r_req_lru_item);
 404         INIT_LIST_HEAD(&req->r_osd_item);
 405
 406         req->r_base_oloc.pool = -1;
 407         req->r_target_oloc.pool = -1;
 408
 409         /* create reply message */
 410         if (use_mempool)
 411                 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0);
 412         else
 413                 msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY,
 414                                    OSD_OPREPLY_FRONT_LEN, gfp_flags, true);
 415         if (!msg) {
 416                 ceph_osdc_put_request(req);
 417                 return NULL;
 418         }
 419         req->r_reply = msg;
 420
 421         /* create request message; allow space for oid */
 422         if (use_mempool)
 423                 msg = ceph_msgpool_get(&osdc->msgpool_op, 0);
 424         else
 425                 msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp_flags, true);
 426         if (!msg) {
 427                 ceph_osdc_put_request(req);
 428                 return NULL;
 429         }
 430
 431         memset(msg->front.iov_base, 0, msg->front.iov_len);
 432
 433         req->r_request = msg;
 434
 435         return req;
 436 }
 437 EXPORT_SYMBOL(ceph_osdc_alloc_request);
 438
 439 static bool osd_req_opcode_valid(u16 opcode)
 440 {
 441         switch (opcode) {
 442 #define GENERATE_CASE(op, opcode, str)  case CEPH_OSD_OP_##op: return true;
 443 __CEPH_FORALL_OSD_OPS(GENERATE_CASE)
 444 #undef GENERATE_CASE
 445         default:
 446                 return false;
 447         }
 448 }
 449
 450 /*
 451  * This is an osd op init function for opcodes that have no data or
 452  * other information associated with them.  It also serves as a
 453  * common init routine for all the other init functions, below.
 454  */
 455 static struct ceph_osd_req_op *
 456 _osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which,
 457                  u16 opcode, u32 flags)
 458 {
 459         struct ceph_osd_req_op *op;
 460
 461         BUG_ON(which >= osd_req->r_num_ops);
 462         BUG_ON(!osd_req_opcode_valid(opcode));
 463
 464         op = &osd_req->r_ops[which];
 465         memset(op, 0, sizeof (*op));
 466         op->op = opcode;
 467         op->flags = flags;
 468
 469         return op;
 470 }
 471
 472 void osd_req_op_init(struct ceph_osd_request *osd_req,
 473                      unsigned int which, u16 opcode, u32 flags)
 474 {
 475         (void)_osd_req_op_init(osd_req, which, opcode, flags);
 476 }
 477 EXPORT_SYMBOL(osd_req_op_init);
 478
 479 void osd_req_op_extent_init(struct ceph_osd_request *osd_req,
 480                                 unsigned int which, u16 opcode,
 481                                 u64 offset, u64 length,
 482                                 u64 truncate_size, u32 truncate_seq)
 483 {
 484         struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
 485                                                       opcode, 0);
 486         size_t payload_len = 0;
 487
 488         BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
 489                opcode != CEPH_OSD_OP_WRITEFULL && opcode != CEPH_OSD_OP_ZERO &&
 490                opcode != CEPH_OSD_OP_TRUNCATE);
 491
 492         op->extent.offset = offset;
 493         op->extent.length = length;
 494         op->extent.truncate_size = truncate_size;
 495         op->extent.truncate_seq = truncate_seq;
 496         if (opcode == CEPH_OSD_OP_WRITE || opcode == CEPH_OSD_OP_WRITEFULL)
 497                 payload_len += length;
 498
 499         op->payload_len = payload_len;
 500 }
 501 EXPORT_SYMBOL(osd_req_op_extent_init);
 502
 503 void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
 504                                 unsigned int which, u64 length)
 505 {
 506         struct ceph_osd_req_op *op;
 507         u64 previous;
 508
 509         BUG_ON(which >= osd_req->r_num_ops);
 510         op = &osd_req->r_ops[which];
 511         previous = op->extent.length;
 512
 513         if (length == previous)
 514                 return;         /* Nothing to do */
 515         BUG_ON(length > previous);
 516
 517         op->extent.length = length;
 518         op->payload_len -= previous - length;
 519 }
 520 EXPORT_SYMBOL(osd_req_op_extent_update);
 521
 522 void osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
 523                         u16 opcode, const char *class, const char *method)
 524 {
 525         struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
 526                                                       opcode, 0);
 527         struct ceph_pagelist *pagelist;
 528         size_t payload_len = 0;
 529         size_t size;
 530
 531         BUG_ON(opcode != CEPH_OSD_OP_CALL);
 532
 533         pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
 534         BUG_ON(!pagelist);
 535         ceph_pagelist_init(pagelist);
 536
 537         op->cls.class_name = class;
 538         size = strlen(class);
 539         BUG_ON(size > (size_t) U8_MAX);
 540         op->cls.class_len = size;
 541         ceph_pagelist_append(pagelist, class, size);
 542         payload_len += size;
 543
 544         op->cls.method_name = method;
 545         size = strlen(method);
 546         BUG_ON(size > (size_t) U8_MAX);
 547         op->cls.method_len = size;
 548         ceph_pagelist_append(pagelist, method, size);
 549         payload_len += size;
 550
 551         osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist);
 552
 553         op->cls.argc = 0;       /* currently unused */
 554
 555         op->payload_len = payload_len;
 556 }
 557 EXPORT_SYMBOL(osd_req_op_cls_init);
 558
 559 int osd_req_op_xattr_init(struct ceph_osd_request *osd_req, unsigned int which,
 560                           u16 opcode, const char *name, const void *value,
 561                           size_t size, u8 cmp_op, u8 cmp_mode)
 562 {
 563         struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
 564                                                       opcode, 0);
 565         struct ceph_pagelist *pagelist;
 566         size_t payload_len;
 567
 568         BUG_ON(opcode != CEPH_OSD_OP_SETXATTR && opcode != CEPH_OSD_OP_CMPXATTR);
 569
 570         pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
 571         if (!pagelist)
 572                 return -ENOMEM;
 573
 574         ceph_pagelist_init(pagelist);
 575
 576         payload_len = strlen(name);
 577         op->xattr.name_len = payload_len;
 578         ceph_pagelist_append(pagelist, name, payload_len);
 579
 580         op->xattr.value_len = size;
 581         ceph_pagelist_append(pagelist, value, size);
 582         payload_len += size;
 583
 584         op->xattr.cmp_op = cmp_op;
 585         op->xattr.cmp_mode = cmp_mode;
 586
 587         ceph_osd_data_pagelist_init(&op->xattr.osd_data, pagelist);
 588         op->payload_len = payload_len;
 589         return 0;
 590 }
 591 EXPORT_SYMBOL(osd_req_op_xattr_init);
 592
 593 void osd_req_op_watch_init(struct ceph_osd_request *osd_req,
 594                                 unsigned int which, u16 opcode,
 595                                 u64 cookie, u64 version, int flag)
 596 {
 597         struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
 598                                                       opcode, 0);
 599
 600         BUG_ON(opcode != CEPH_OSD_OP_NOTIFY_ACK && opcode != CEPH_OSD_OP_WATCH);
 601
 602         op->watch.cookie = cookie;
 603         op->watch.ver = version;
 604         if (opcode == CEPH_OSD_OP_WATCH && flag)
 605                 op->watch.flag = (u8)1;
 606 }
 607 EXPORT_SYMBOL(osd_req_op_watch_init);
 608
 609 void osd_req_op_alloc_hint_init(struct ceph_osd_request *osd_req,
 610                                 unsigned int which,
 611                                 u64 expected_object_size,
 612                                 u64 expected_write_size)
 613 {
 614         struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
 615                                                       CEPH_OSD_OP_SETALLOCHINT,
 616                                                       0);
 617
 618         op->alloc_hint.expected_object_size = expected_object_size;
 619         op->alloc_hint.expected_write_size = expected_write_size;
 620
 621         /*
 622          * CEPH_OSD_OP_SETALLOCHINT op is advisory and therefore deemed
 623          * not worth a feature bit.  Set FAILOK per-op flag to make
 624          * sure older osds don't trip over an unsupported opcode.
 625          */
 626         op->flags |= CEPH_OSD_OP_FLAG_FAILOK;
 627 }
 628 EXPORT_SYMBOL(osd_req_op_alloc_hint_init);
 629
 630 static void ceph_osdc_msg_data_add(struct ceph_msg *msg,
 631                                 struct ceph_osd_data *osd_data)
 632 {
 633         u64 length = ceph_osd_data_length(osd_data);
 634
 635         if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
 636                 BUG_ON(length > (u64) SIZE_MAX);
 637                 if (length)
 638                         ceph_msg_data_add_pages(msg, osd_data->pages,
 639                                         length, osd_data->alignment);
 640         } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
 641                 BUG_ON(!length);
 642                 ceph_msg_data_add_pagelist(msg, osd_data->pagelist);
 643 #ifdef CONFIG_BLOCK
 644         } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
 645                 ceph_msg_data_add_bio(msg, osd_data->bio, length);
 646 #endif
 647         } else {
 648                 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
 649         }
 650 }
 651
 652 static u64 osd_req_encode_op(struct ceph_osd_request *req,
 653                               struct ceph_osd_op *dst, unsigned int which)
 654 {
 655         struct ceph_osd_req_op *src;
 656         struct ceph_osd_data *osd_data;
 657         u64 request_data_len = 0;
 658         u64 data_length;
 659
 660         BUG_ON(which >= req->r_num_ops);
 661         src = &req->r_ops[which];
 662         if (WARN_ON(!osd_req_opcode_valid(src->op))) {
 663                 pr_err("unrecognized osd opcode %d\n", src->op);
 664
 665                 return 0;
 666         }
 667
 668         switch (src->op) {
 669         case CEPH_OSD_OP_STAT:
 670                 osd_data = &src->raw_data_in;
 671                 ceph_osdc_msg_data_add(req->r_reply, osd_data);
 672                 break;
 673         case CEPH_OSD_OP_READ:
 674         case CEPH_OSD_OP_WRITE:
 675         case CEPH_OSD_OP_WRITEFULL:
 676         case CEPH_OSD_OP_ZERO:
 677         case CEPH_OSD_OP_TRUNCATE:
 678                 if (src->op == CEPH_OSD_OP_WRITE ||
 679                     src->op == CEPH_OSD_OP_WRITEFULL)
 680                         request_data_len = src->extent.length;
 681                 dst->extent.offset = cpu_to_le64(src->extent.offset);
 682                 dst->extent.length = cpu_to_le64(src->extent.length);
 683                 dst->extent.truncate_size =
 684                         cpu_to_le64(src->extent.truncate_size);
 685                 dst->extent.truncate_seq =
 686                         cpu_to_le32(src->extent.truncate_seq);
 687                 osd_data = &src->extent.osd_data;
 688                 if (src->op == CEPH_OSD_OP_WRITE ||
 689                     src->op == CEPH_OSD_OP_WRITEFULL)
 690                         ceph_osdc_msg_data_add(req->r_request, osd_data);
 691                 else
 692                         ceph_osdc_msg_data_add(req->r_reply, osd_data);
 693                 break;
 694         case CEPH_OSD_OP_CALL:
 695                 dst->cls.class_len = src->cls.class_len;
 696                 dst->cls.method_len = src->cls.method_len;
 697                 osd_data = &src->cls.request_info;
 698                 ceph_osdc_msg_data_add(req->r_request, osd_data);
 699                 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGELIST);
 700                 request_data_len = osd_data->pagelist->length;
 701
 702                 osd_data = &src->cls.request_data;
 703                 data_length = ceph_osd_data_length(osd_data);
 704                 if (data_length) {
 705                         BUG_ON(osd_data->type == CEPH_OSD_DATA_TYPE_NONE);
 706                         dst->cls.indata_len = cpu_to_le32(data_length);
 707                         ceph_osdc_msg_data_add(req->r_request, osd_data);
 708                         src->payload_len += data_length;
 709                         request_data_len += data_length;
 710                 }
 711                 osd_data = &src->cls.response_data;
 712                 ceph_osdc_msg_data_add(req->r_reply, osd_data);
 713                 break;
 714         case CEPH_OSD_OP_STARTSYNC:
 715                 break;
 716         case CEPH_OSD_OP_NOTIFY_ACK:
 717         case CEPH_OSD_OP_WATCH:
 718                 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
 719                 dst->watch.ver = cpu_to_le64(src->watch.ver);
 720                 dst->watch.flag = src->watch.flag;
 721                 break;
 722         case CEPH_OSD_OP_SETALLOCHINT:
 723                 dst->alloc_hint.expected_object_size =
 724                     cpu_to_le64(src->alloc_hint.expected_object_size);
 725                 dst->alloc_hint.expected_write_size =
 726                     cpu_to_le64(src->alloc_hint.expected_write_size);
 727                 break;
 728         case CEPH_OSD_OP_SETXATTR:
 729         case CEPH_OSD_OP_CMPXATTR:
 730                 dst->xattr.name_len = cpu_to_le32(src->xattr.name_len);
 731                 dst->xattr.value_len = cpu_to_le32(src->xattr.value_len);
 732                 dst->xattr.cmp_op = src->xattr.cmp_op;
 733                 dst->xattr.cmp_mode = src->xattr.cmp_mode;
 734                 osd_data = &src->xattr.osd_data;
 735                 ceph_osdc_msg_data_add(req->r_request, osd_data);
 736                 request_data_len = osd_data->pagelist->length;
 737                 break;
 738         case CEPH_OSD_OP_CREATE:
 739         case CEPH_OSD_OP_DELETE:
 740                 break;
 741         default:
 742                 pr_err("unsupported osd opcode %s\n",
 743                         ceph_osd_op_name(src->op));
 744                 WARN_ON(1);
 745
 746                 return 0;
 747         }
 748
 749         dst->op = cpu_to_le16(src->op);
 750         dst->flags = cpu_to_le32(src->flags);
 751         dst->payload_len = cpu_to_le32(src->payload_len);
 752
 753         return request_data_len;
 754 }
 755
 756 /*
 757  * build new request AND message, calculate layout, and adjust file
 758  * extent as needed.
 759  *
 760  * if the file was recently truncated, we include information about its
 761  * old and new size so that the object can be updated appropriately.  (we
 762  * avoid synchronously deleting truncated objects because it's slow.)
 763  *
 764  * if @do_sync, include a 'startsync' command so that the osd will flush
 765  * data quickly.
 766  */
 767 struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
 768                                                struct ceph_file_layout *layout,
 769                                                struct ceph_vino vino,
 770                                                u64 off, u64 *plen,
 771                                                unsigned int which, int num_ops,
 772                                                int opcode, int flags,
 773                                                struct ceph_snap_context *snapc,
 774                                                u32 truncate_seq,
 775                                                u64 truncate_size,
 776                                                bool use_mempool)
 777 {
 778         struct ceph_osd_request *req;
 779         u64 objnum = 0;
 780         u64 objoff = 0;
 781         u64 objlen = 0;
 782         int r;
 783
 784         BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
 785                opcode != CEPH_OSD_OP_ZERO && opcode != CEPH_OSD_OP_TRUNCATE &&
 786                opcode != CEPH_OSD_OP_CREATE && opcode != CEPH_OSD_OP_DELETE);
 787
 788         req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
 789                                         GFP_NOFS);
 790         if (!req)
 791                 return ERR_PTR(-ENOMEM);
 792
 793         req->r_flags = flags;
 794
 795         /* calculate max write size */
 796         r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
 797         if (r < 0) {
 798                 ceph_osdc_put_request(req);
 799                 return ERR_PTR(r);
 800         }
 801
 802         if (opcode == CEPH_OSD_OP_CREATE || opcode == CEPH_OSD_OP_DELETE) {
 803                 osd_req_op_init(req, which, opcode, 0);
 804         } else {
 805                 u32 object_size = le32_to_cpu(layout->fl_object_size);
 806                 u32 object_base = off - objoff;
 807                 if (!(truncate_seq == 1 && truncate_size == -1ULL)) {
 808                         if (truncate_size <= object_base) {
 809                                 truncate_size = 0;
 810                         } else {
 811                                 truncate_size -= object_base;
 812                                 if (truncate_size > object_size)
 813                                         truncate_size = object_size;
 814                         }
 815                 }
 816                 osd_req_op_extent_init(req, which, opcode, objoff, objlen,
 817                                        truncate_size, truncate_seq);
 818         }
 819
 820         req->r_base_oloc.pool = ceph_file_layout_pg_pool(*layout);
 821
 822         snprintf(req->r_base_oid.name, sizeof(req->r_base_oid.name),
 823                  "%llx.%08llx", vino.ino, objnum);
 824         req->r_base_oid.name_len = strlen(req->r_base_oid.name);
 825
 826         return req;
 827 }
 828 EXPORT_SYMBOL(ceph_osdc_new_request);
 829
 830 /*
 831  * We keep osd requests in an rbtree, sorted by ->r_tid.
 832  */
 833 static void __insert_request(struct ceph_osd_client *osdc,
 834                              struct ceph_osd_request *new)
 835 {
 836         struct rb_node **p = &osdc->requests.rb_node;
 837         struct rb_node *parent = NULL;
 838         struct ceph_osd_request *req = NULL;
 839
 840         while (*p) {
 841                 parent = *p;
 842                 req = rb_entry(parent, struct ceph_osd_request, r_node);
 843                 if (new->r_tid < req->r_tid)
 844                         p = &(*p)->rb_left;
 845                 else if (new->r_tid > req->r_tid)
 846                         p = &(*p)->rb_right;
 847                 else
 848                         BUG();
 849         }
 850
 851         rb_link_node(&new->r_node, parent, p);
 852         rb_insert_color(&new->r_node, &osdc->requests);
 853 }
 854
 855 static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc,
 856                                                  u64 tid)
 857 {
 858         struct ceph_osd_request *req;
 859         struct rb_node *n = osdc->requests.rb_node;
 860
 861         while (n) {
 862                 req = rb_entry(n, struct ceph_osd_request, r_node);
 863                 if (tid < req->r_tid)
 864                         n = n->rb_left;
 865                 else if (tid > req->r_tid)
 866                         n = n->rb_right;
 867                 else
 868                         return req;
 869         }
 870         return NULL;
 871 }
 872
 873 static struct ceph_osd_request *
 874 __lookup_request_ge(struct ceph_osd_client *osdc,
 875                     u64 tid)
 876 {
 877         struct ceph_osd_request *req;
 878         struct rb_node *n = osdc->requests.rb_node;
 879
 880         while (n) {
 881                 req = rb_entry(n, struct ceph_osd_request, r_node);
 882                 if (tid < req->r_tid) {
 883                         if (!n->rb_left)
 884                                 return req;
 885                         n = n->rb_left;
 886                 } else if (tid > req->r_tid) {
 887                         n = n->rb_right;
 888                 } else {
 889                         return req;
 890                 }
 891         }
 892         return NULL;
 893 }
 894
 895 static void __kick_linger_request(struct ceph_osd_request *req)
 896 {
 897         struct ceph_osd_client *osdc = req->r_osdc;
 898         struct ceph_osd *osd = req->r_osd;
 899
 900         /*
 901          * Linger requests need to be resent with a new tid to avoid
 902          * the dup op detection logic on the OSDs.  Achieve this with
 903          * a re-register dance instead of open-coding.
 904          */
 905         ceph_osdc_get_request(req);
 906         if (!list_empty(&req->r_linger_item))
 907                 __unregister_linger_request(osdc, req);
 908         else
 909                 __unregister_request(osdc, req);
 910         __register_request(osdc, req);
 911         ceph_osdc_put_request(req);
 912
 913         /*
 914          * Unless request has been registered as both normal and
 915          * lingering, __unregister{,_linger}_request clears r_osd.
 916          * However, here we need to preserve r_osd to make sure we
 917          * requeue on the same OSD.
 918          */
 919         WARN_ON(req->r_osd || !osd);
 920         req->r_osd = osd;
 921
 922         dout("%s requeueing %p tid %llu\n", __func__, req, req->r_tid);
 923         __enqueue_request(req);
 924 }
 925
 926 /*
 927  * Resubmit requests pending on the given osd.
 928  */
 929 static void __kick_osd_requests(struct ceph_osd_client *osdc,
 930                                 struct ceph_osd *osd)
 931 {
 932         struct ceph_osd_request *req, *nreq;
 933         LIST_HEAD(resend);
 934         LIST_HEAD(resend_linger);
 935         int err;
 936
 937         dout("%s osd%d\n", __func__, osd->o_osd);
 938         err = __reset_osd(osdc, osd);
 939         if (err)
 940                 return;
 941
 942         /*
 943          * Build up a list of requests to resend by traversing the
 944          * osd's list of requests.  Requests for a given object are
 945          * sent in tid order, and that is also the order they're
 946          * kept on this list.  Therefore all requests that are in
 947          * flight will be found first, followed by all requests that
 948          * have not yet been sent.  And to resend requests while
 949          * preserving this order we will want to put any sent
 950          * requests back on the front of the osd client's unsent
 951          * list.
 952          *
 953          * So we build a separate ordered list of already-sent
 954          * requests for the affected osd and splice it onto the
 955          * front of the osd client's unsent list.  Once we've seen a
 956          * request that has not yet been sent we're done.  Those
 957          * requests are already sitting right where they belong.
 958          */
 959         list_for_each_entry(req, &osd->o_requests, r_osd_item) {
 960                 if (!req->r_sent)
 961                         break;
 962
 963                 if (!req->r_linger) {
 964                         dout("%s requeueing %p tid %llu\n", __func__, req,
 965                              req->r_tid);
 966                         list_move_tail(&req->r_req_lru_item, &resend);
 967                         req->r_flags |= CEPH_OSD_FLAG_RETRY;
 968                 } else {
 969                         list_move_tail(&req->r_req_lru_item, &resend_linger);
 970                 }
 971         }
 972         list_splice(&resend, &osdc->req_unsent);
 973
 974         /*
 975          * Both registered and not yet registered linger requests are
 976          * enqueued with a new tid on the same OSD.  We add/move them
 977          * to req_unsent/o_requests at the end to keep things in tid
 978          * order.
 979          */
 980         list_for_each_entry_safe(req, nreq, &osd->o_linger_requests,
 981                                  r_linger_osd_item) {
 982                 WARN_ON(!list_empty(&req->r_req_lru_item));
 983                 __kick_linger_request(req);
 984         }
 985
 986         list_for_each_entry_safe(req, nreq, &resend_linger, r_req_lru_item)
 987                 __kick_linger_request(req);
 988 }
 989
 990 /*
 991  * If the osd connection drops, we need to resubmit all requests.
 992  */
 993 static void osd_reset(struct ceph_connection *con)
 994 {
 995         struct ceph_osd *osd = con->private;
 996         struct ceph_osd_client *osdc;
 997
 998         if (!osd)
 999                 return;
1000         dout("osd_reset osd%d\n", osd->o_osd);
1001         osdc = osd->o_osdc;
1002         down_read(&osdc->map_sem);
1003         mutex_lock(&osdc->request_mutex);
1004         __kick_osd_requests(osdc, osd);
1005         __send_queued(osdc);
1006         mutex_unlock(&osdc->request_mutex);
1007         up_read(&osdc->map_sem);
1008 }
1009
1010 /*
1011  * Track open sessions with osds.
1012  */
1013 static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
1014 {
1015         struct ceph_osd *osd;
1016
1017         osd = kzalloc(sizeof(*osd), GFP_NOFS);
1018         if (!osd)
1019                 return NULL;
1020
1021         atomic_set(&osd->o_ref, 1);
1022         osd->o_osdc = osdc;
1023         osd->o_osd = onum;
1024         RB_CLEAR_NODE(&osd->o_node);
1025         INIT_LIST_HEAD(&osd->o_requests);
1026         INIT_LIST_HEAD(&osd->o_linger_requests);
1027         INIT_LIST_HEAD(&osd->o_osd_lru);
1028         osd->o_incarnation = 1;
1029
1030         ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
1031
1032         INIT_LIST_HEAD(&osd->o_keepalive_item);
1033         return osd;
1034 }
1035
1036 static struct ceph_osd *get_osd(struct ceph_osd *osd)
1037 {
1038         if (atomic_inc_not_zero(&osd->o_ref)) {
1039                 dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
1040                      atomic_read(&osd->o_ref));
1041                 return osd;
1042         } else {
1043                 dout("get_osd %p FAIL\n", osd);
1044                 return NULL;
1045         }
1046 }
1047
1048 static void put_osd(struct ceph_osd *osd)
1049 {
1050         dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
1051              atomic_read(&osd->o_ref) - 1);
1052         if (atomic_dec_and_test(&osd->o_ref)) {
1053                 struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth;
1054
1055                 if (osd->o_auth.authorizer)
1056                         ceph_auth_destroy_authorizer(ac, osd->o_auth.authorizer);
1057                 kfree(osd);
1058         }
1059 }
1060
1061 /*
1062  * remove an osd from our map
1063  */
1064 static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
1065 {
1066         dout("%s %p osd%d\n", __func__, osd, osd->o_osd);
1067         WARN_ON(!list_empty(&osd->o_requests));
1068         WARN_ON(!list_empty(&osd->o_linger_requests));
1069
1070         list_del_init(&osd->o_osd_lru);
1071         rb_erase(&osd->o_node, &osdc->osds);
1072         RB_CLEAR_NODE(&osd->o_node);
1073 }
1074
1075 static void remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
1076 {
1077         dout("%s %p osd%d\n", __func__, osd, osd->o_osd);
1078
1079         if (!RB_EMPTY_NODE(&osd->o_node)) {
1080                 ceph_con_close(&osd->o_con);
1081                 __remove_osd(osdc, osd);
1082                 put_osd(osd);
1083         }
1084 }
1085
1086 static void remove_all_osds(struct ceph_osd_client *osdc)
1087 {
1088         dout("%s %p\n", __func__, osdc);
1089         mutex_lock(&osdc->request_mutex);
1090         while (!RB_EMPTY_ROOT(&osdc->osds)) {
1091                 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
1092                                                 struct ceph_osd, o_node);
1093                 remove_osd(osdc, osd);
1094         }
1095         mutex_unlock(&osdc->request_mutex);
1096 }
1097
1098 static void __move_osd_to_lru(struct ceph_osd_client *osdc,
1099                               struct ceph_osd *osd)
1100 {
1101         dout("%s %p\n", __func__, osd);
1102         BUG_ON(!list_empty(&osd->o_osd_lru));
1103
1104         list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
1105         osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl;
1106 }
1107
1108 static void maybe_move_osd_to_lru(struct ceph_osd_client *osdc,
1109                                   struct ceph_osd *osd)
1110 {
1111         dout("%s %p\n", __func__, osd);
1112
1113         if (list_empty(&osd->o_requests) &&
1114             list_empty(&osd->o_linger_requests))
1115                 __move_osd_to_lru(osdc, osd);
1116 }
1117
1118 static void __remove_osd_from_lru(struct ceph_osd *osd)
1119 {
1120         dout("__remove_osd_from_lru %p\n", osd);
1121         if (!list_empty(&osd->o_osd_lru))
1122                 list_del_init(&osd->o_osd_lru);
1123 }
1124
1125 static void remove_old_osds(struct ceph_osd_client *osdc)
1126 {
1127         struct ceph_osd *osd, *nosd;
1128
1129         dout("__remove_old_osds %p\n", osdc);
1130         mutex_lock(&osdc->request_mutex);
1131         list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
1132                 if (time_before(jiffies, osd->lru_ttl))
1133                         break;
1134                 remove_osd(osdc, osd);
1135         }
1136         mutex_unlock(&osdc->request_mutex);
1137 }
1138
1139 /*
1140  * reset osd connect
1141  */
1142 static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
1143 {
1144         struct ceph_entity_addr *peer_addr;
1145
1146         dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
1147         if (list_empty(&osd->o_requests) &&
1148             list_empty(&osd->o_linger_requests)) {
1149                 remove_osd(osdc, osd);
1150                 return -ENODEV;
1151         }
1152
1153         peer_addr = &osdc->osdmap->osd_addr[osd->o_osd];
1154         if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1155                         !ceph_con_opened(&osd->o_con)) {
1156                 struct ceph_osd_request *req;
1157
1158                 dout("osd addr hasn't changed and connection never opened, "
1159                      "letting msgr retry\n");
1160                 /* touch each r_stamp for handle_timeout()'s benfit */
1161                 list_for_each_entry(req, &osd->o_requests, r_osd_item)
1162                         req->r_stamp = jiffies;
1163
1164                 return -EAGAIN;
1165         }
1166
1167         ceph_con_close(&osd->o_con);
1168         ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1169         osd->o_incarnation++;
1170
1171         return 0;
1172 }
1173
1174 static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new)
1175 {
1176         struct rb_node **p = &osdc->osds.rb_node;
1177         struct rb_node *parent = NULL;
1178         struct ceph_osd *osd = NULL;
1179
1180         dout("__insert_osd %p osd%d\n", new, new->o_osd);
1181         while (*p) {
1182                 parent = *p;
1183                 osd = rb_entry(parent, struct ceph_osd, o_node);
1184                 if (new->o_osd < osd->o_osd)
1185                         p = &(*p)->rb_left;
1186                 else if (new->o_osd > osd->o_osd)
1187                         p = &(*p)->rb_right;
1188                 else
1189                         BUG();
1190         }
1191
1192         rb_link_node(&new->o_node, parent, p);
1193         rb_insert_color(&new->o_node, &osdc->osds);
1194 }
1195
1196 static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o)
1197 {
1198         struct ceph_osd *osd;
1199         struct rb_node *n = osdc->osds.rb_node;
1200
1201         while (n) {
1202                 osd = rb_entry(n, struct ceph_osd, o_node);
1203                 if (o < osd->o_osd)
1204                         n = n->rb_left;
1205                 else if (o > osd->o_osd)
1206                         n = n->rb_right;
1207                 else
1208                         return osd;
1209         }
1210         return NULL;
1211 }
1212
1213 static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
1214 {
1215         schedule_delayed_work(&osdc->timeout_work,
1216                               osdc->client->options->osd_keepalive_timeout);
1217 }
1218
1219 static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
1220 {
1221         cancel_delayed_work(&osdc->timeout_work);
1222 }
1223
1224 /*
1225  * Register request, assign tid.  If this is the first request, set up
1226  * the timeout event.
1227  */
1228 static void __register_request(struct ceph_osd_client *osdc,
1229                                struct ceph_osd_request *req)
1230 {
1231         req->r_tid = ++osdc->last_tid;
1232         req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
1233         dout("__register_request %p tid %lld\n", req, req->r_tid);
1234         __insert_request(osdc, req);
1235         ceph_osdc_get_request(req);
1236         osdc->num_requests++;
1237         if (osdc->num_requests == 1) {
1238                 dout(" first request, scheduling timeout\n");
1239                 __schedule_osd_timeout(osdc);
1240         }
1241 }
1242
1243 /*
1244  * called under osdc->request_mutex
1245  */
1246 static void __unregister_request(struct ceph_osd_client *osdc,
1247                                  struct ceph_osd_request *req)
1248 {
1249         if (RB_EMPTY_NODE(&req->r_node)) {
1250                 dout("__unregister_request %p tid %lld not registered\n",
1251                         req, req->r_tid);
1252                 return;
1253         }
1254
1255         dout("__unregister_request %p tid %lld\n", req, req->r_tid);
1256         rb_erase(&req->r_node, &osdc->requests);
1257         RB_CLEAR_NODE(&req->r_node);
1258         osdc->num_requests--;
1259
1260         if (req->r_osd) {
1261                 /* make sure the original request isn't in flight. */
1262                 ceph_msg_revoke(req->r_request);
1263
1264                 list_del_init(&req->r_osd_item);
1265                 maybe_move_osd_to_lru(osdc, req->r_osd);
1266                 if (list_empty(&req->r_linger_osd_item))
1267                         req->r_osd = NULL;
1268         }
1269
1270         list_del_init(&req->r_req_lru_item);
1271         ceph_osdc_put_request(req);
1272
1273         if (osdc->num_requests == 0) {
1274                 dout(" no requests, canceling timeout\n");
1275                 __cancel_osd_timeout(osdc);
1276         }
1277 }
1278
1279 /*
1280  * Cancel a previously queued request message
1281  */
1282 static void __cancel_request(struct ceph_osd_request *req)
1283 {
1284         if (req->r_sent && req->r_osd) {
1285                 ceph_msg_revoke(req->r_request);
1286                 req->r_sent = 0;
1287         }
1288 }
1289
1290 static void __register_linger_request(struct ceph_osd_client *osdc,
1291                                     struct ceph_osd_request *req)
1292 {
1293         dout("%s %p tid %llu\n", __func__, req, req->r_tid);
1294         WARN_ON(!req->r_linger);
1295
1296         ceph_osdc_get_request(req);
1297         list_add_tail(&req->r_linger_item, &osdc->req_linger);
1298         if (req->r_osd)
1299                 list_add_tail(&req->r_linger_osd_item,
1300                               &req->r_osd->o_linger_requests);
1301 }
1302
1303 static void __unregister_linger_request(struct ceph_osd_client *osdc,
1304                                         struct ceph_osd_request *req)
1305 {
1306         WARN_ON(!req->r_linger);
1307
1308         if (list_empty(&req->r_linger_item)) {
1309                 dout("%s %p tid %llu not registered\n", __func__, req,
1310                      req->r_tid);
1311                 return;
1312         }
1313
1314         dout("%s %p tid %llu\n", __func__, req, req->r_tid);
1315         list_del_init(&req->r_linger_item);
1316
1317         if (req->r_osd) {
1318                 list_del_init(&req->r_linger_osd_item);
1319                 maybe_move_osd_to_lru(osdc, req->r_osd);
1320                 if (list_empty(&req->r_osd_item))
1321                         req->r_osd = NULL;
1322         }
1323         ceph_osdc_put_request(req);
1324 }
1325
1326 void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
1327                                   struct ceph_osd_request *req)
1328 {
1329         if (!req->r_linger) {
1330                 dout("set_request_linger %p\n", req);
1331                 req->r_linger = 1;
1332         }
1333 }
1334 EXPORT_SYMBOL(ceph_osdc_set_request_linger);
1335
1336 /*
1337  * Returns whether a request should be blocked from being sent
1338  * based on the current osdmap and osd_client settings.
1339  *
1340  * Caller should hold map_sem for read.
1341  */
1342 static bool __req_should_be_paused(struct ceph_osd_client *osdc,
1343                                    struct ceph_osd_request *req)
1344 {
1345         bool pauserd = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD);
1346         bool pausewr = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR) ||
1347                 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
1348         return (req->r_flags & CEPH_OSD_FLAG_READ && pauserd) ||
1349                 (req->r_flags & CEPH_OSD_FLAG_WRITE && pausewr);
1350 }
1351
1352 /*
1353  * Calculate mapping of a request to a PG.  Takes tiering into account.
1354  */
1355 static int __calc_request_pg(struct ceph_osdmap *osdmap,
1356                              struct ceph_osd_request *req,
1357                              struct ceph_pg *pg_out)
1358 {
1359         bool need_check_tiering;
1360
1361         need_check_tiering = false;
1362         if (req->r_target_oloc.pool == -1) {
1363                 req->r_target_oloc = req->r_base_oloc; /* struct */
1364                 need_check_tiering = true;
1365         }
1366         if (req->r_target_oid.name_len == 0) {
1367                 ceph_oid_copy(&req->r_target_oid, &req->r_base_oid);
1368                 need_check_tiering = true;
1369         }
1370
1371         if (need_check_tiering &&
1372             (req->r_flags & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0) {
1373                 struct ceph_pg_pool_info *pi;
1374
1375                 pi = ceph_pg_pool_by_id(osdmap, req->r_target_oloc.pool);
1376                 if (pi) {
1377                         if ((req->r_flags & CEPH_OSD_FLAG_READ) &&
1378                             pi->read_tier >= 0)
1379                                 req->r_target_oloc.pool = pi->read_tier;
1380                         if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
1381                             pi->write_tier >= 0)
1382                                 req->r_target_oloc.pool = pi->write_tier;
1383                 }
1384                 /* !pi is caught in ceph_oloc_oid_to_pg() */
1385         }
1386
1387         return ceph_oloc_oid_to_pg(osdmap, &req->r_target_oloc,
1388                                    &req->r_target_oid, pg_out);
1389 }
1390
1391 static void __enqueue_request(struct ceph_osd_request *req)
1392 {
1393         struct ceph_osd_client *osdc = req->r_osdc;
1394
1395         dout("%s %p tid %llu to osd%d\n", __func__, req, req->r_tid,
1396              req->r_osd ? req->r_osd->o_osd : -1);
1397
1398         if (req->r_osd) {
1399                 __remove_osd_from_lru(req->r_osd);
1400                 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
1401                 list_move_tail(&req->r_req_lru_item, &osdc->req_unsent);
1402         } else {
1403                 list_move_tail(&req->r_req_lru_item, &osdc->req_notarget);
1404         }
1405 }
1406
1407 /*
1408  * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
1409  * (as needed), and set the request r_osd appropriately.  If there is
1410  * no up osd, set r_osd to NULL.  Move the request to the appropriate list
1411  * (unsent, homeless) or leave on in-flight lru.
1412  *
1413  * Return 0 if unchanged, 1 if changed, or negative on error.
1414  *
1415  * Caller should hold map_sem for read and request_mutex.
1416  */
1417 static int __map_request(struct ceph_osd_client *osdc,
1418                          struct ceph_osd_request *req, int force_resend)
1419 {
1420         struct ceph_pg pgid;
1421         int acting[CEPH_PG_MAX_SIZE];
1422         int num, o;
1423         int err;
1424         bool was_paused;
1425
1426         dout("map_request %p tid %lld\n", req, req->r_tid);
1427
1428         err = __calc_request_pg(osdc->osdmap, req, &pgid);
1429         if (err) {
1430                 list_move(&req->r_req_lru_item, &osdc->req_notarget);
1431                 return err;
1432         }
1433         req->r_pgid = pgid;
1434
1435         num = ceph_calc_pg_acting(osdc->osdmap, pgid, acting, &o);
1436         if (num < 0)
1437                 num = 0;
1438
1439         was_paused = req->r_paused;
1440         req->r_paused = __req_should_be_paused(osdc, req);
1441         if (was_paused && !req->r_paused)
1442                 force_resend = 1;
1443
1444         if ((!force_resend &&
1445              req->r_osd && req->r_osd->o_osd == o &&
1446              req->r_sent >= req->r_osd->o_incarnation &&
1447              req->r_num_pg_osds == num &&
1448              memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
1449             (req->r_osd == NULL && o == -1) ||
1450             req->r_paused)
1451                 return 0;  /* no change */
1452
1453         dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n",
1454              req->r_tid, pgid.pool, pgid.seed, o,
1455              req->r_osd ? req->r_osd->o_osd : -1);
1456
1457         /* record full pg acting set */
1458         memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
1459         req->r_num_pg_osds = num;
1460
1461         if (req->r_osd) {
1462                 __cancel_request(req);
1463                 list_del_init(&req->r_osd_item);
1464                 list_del_init(&req->r_linger_osd_item);
1465                 req->r_osd = NULL;
1466         }
1467
1468         req->r_osd = __lookup_osd(osdc, o);
1469         if (!req->r_osd && o >= 0) {
1470                 err = -ENOMEM;
1471                 req->r_osd = create_osd(osdc, o);
1472                 if (!req->r_osd) {
1473                         list_move(&req->r_req_lru_item, &osdc->req_notarget);
1474                         goto out;
1475                 }
1476
1477                 dout("map_request osd %p is osd%d\n", req->r_osd, o);
1478                 __insert_osd(osdc, req->r_osd);
1479
1480                 ceph_con_open(&req->r_osd->o_con,
1481                               CEPH_ENTITY_TYPE_OSD, o,
1482                               &osdc->osdmap->osd_addr[o]);
1483         }
1484
1485         __enqueue_request(req);
1486         err = 1;   /* osd or pg changed */
1487
1488 out:
1489         return err;
1490 }
1491
1492 /*
1493  * caller should hold map_sem (for read) and request_mutex
1494  */
1495 static void __send_request(struct ceph_osd_client *osdc,
1496                            struct ceph_osd_request *req)
1497 {
1498         void *p;
1499
1500         dout("send_request %p tid %llu to osd%d flags %d pg %lld.%x\n",
1501              req, req->r_tid, req->r_osd->o_osd, req->r_flags,
1502              (unsigned long long)req->r_pgid.pool, req->r_pgid.seed);
1503
1504         /* fill in message content that changes each time we send it */
1505         put_unaligned_le32(osdc->osdmap->epoch, req->r_request_osdmap_epoch);
1506         put_unaligned_le32(req->r_flags, req->r_request_flags);
1507         put_unaligned_le64(req->r_target_oloc.pool, req->r_request_pool);
1508         p = req->r_request_pgid;
1509         ceph_encode_64(&p, req->r_pgid.pool);
1510         ceph_encode_32(&p, req->r_pgid.seed);
1511         put_unaligned_le64(1, req->r_request_attempts);  /* FIXME */
1512         memcpy(req->r_request_reassert_version, &req->r_reassert_version,
1513                sizeof(req->r_reassert_version));
1514
1515         req->r_stamp = jiffies;
1516         list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
1517
1518         ceph_msg_get(req->r_request); /* send consumes a ref */
1519
1520         req->r_sent = req->r_osd->o_incarnation;
1521
1522         ceph_con_send(&req->r_osd->o_con, req->r_request);
1523 }
1524
1525 /*
1526  * Send any requests in the queue (req_unsent).
1527  */
1528 static void __send_queued(struct ceph_osd_client *osdc)
1529 {
1530         struct ceph_osd_request *req, *tmp;
1531
1532         dout("__send_queued\n");
1533         list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item)
1534                 __send_request(osdc, req);
1535 }
1536
1537 /*
1538  * Caller should hold map_sem for read and request_mutex.
1539  */
1540 static int __ceph_osdc_start_request(struct ceph_osd_client *osdc,
1541                                      struct ceph_osd_request *req,
1542                                      bool nofail)
1543 {
1544         int rc;
1545
1546         __register_request(osdc, req);
1547         req->r_sent = 0;
1548         req->r_got_reply = 0;
1549         rc = __map_request(osdc, req, 0);
1550         if (rc < 0) {
1551                 if (nofail) {
1552                         dout("osdc_start_request failed map, "
1553                                 " will retry %lld\n", req->r_tid);
1554                         rc = 0;
1555                 } else {
1556                         __unregister_request(osdc, req);
1557                 }
1558                 return rc;
1559         }
1560
1561         if (req->r_osd == NULL) {
1562                 dout("send_request %p no up osds in pg\n", req);
1563                 ceph_monc_request_next_osdmap(&osdc->client->monc);
1564         } else {
1565                 __send_queued(osdc);
1566         }
1567
1568         return 0;
1569 }
1570
1571 /*
1572  * Timeout callback, called every N seconds when 1 or more osd
1573  * requests has been active for more than N seconds.  When this
1574  * happens, we ping all OSDs with requests who have timed out to
1575  * ensure any communications channel reset is detected.  Reset the
1576  * request timeouts another N seconds in the future as we go.
1577  * Reschedule the timeout event another N seconds in future (unless
1578  * there are no open requests).
1579  */
1580 static void handle_timeout(struct work_struct *work)
1581 {
1582         struct ceph_osd_client *osdc =
1583                 container_of(work, struct ceph_osd_client, timeout_work.work);
1584         struct ceph_options *opts = osdc->client->options;
1585         struct ceph_osd_request *req;
1586         struct ceph_osd *osd;
1587         struct list_head slow_osds;
1588         dout("timeout\n");
1589         down_read(&osdc->map_sem);
1590
1591         ceph_monc_request_next_osdmap(&osdc->client->monc);
1592
1593         mutex_lock(&osdc->request_mutex);
1594
1595         /*
1596          * ping osds that are a bit slow.  this ensures that if there
1597          * is a break in the TCP connection we will notice, and reopen
1598          * a connection with that osd (from the fault callback).
1599          */
1600         INIT_LIST_HEAD(&slow_osds);
1601         list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
1602                 if (time_before(jiffies,
1603                                 req->r_stamp + opts->osd_keepalive_timeout))
1604                         break;
1605
1606                 osd = req->r_osd;
1607                 BUG_ON(!osd);
1608                 dout(" tid %llu is slow, will send keepalive on osd%d\n",
1609                      req->r_tid, osd->o_osd);
1610                 list_move_tail(&osd->o_keepalive_item, &slow_osds);
1611         }
1612         while (!list_empty(&slow_osds)) {
1613                 osd = list_entry(slow_osds.next, struct ceph_osd,
1614                                  o_keepalive_item);
1615                 list_del_init(&osd->o_keepalive_item);
1616                 ceph_con_keepalive(&osd->o_con);
1617         }
1618
1619         __schedule_osd_timeout(osdc);
1620         __send_queued(osdc);
1621         mutex_unlock(&osdc->request_mutex);
1622         up_read(&osdc->map_sem);
1623 }
1624
1625 static void handle_osds_timeout(struct work_struct *work)
1626 {
1627         struct ceph_osd_client *osdc =
1628                 container_of(work, struct ceph_osd_client,
1629                              osds_timeout_work.work);
1630         unsigned long delay = osdc->client->options->osd_idle_ttl / 4;
1631
1632         dout("osds timeout\n");
1633         down_read(&osdc->map_sem);
1634         remove_old_osds(osdc);
1635         up_read(&osdc->map_sem);
1636
1637         schedule_delayed_work(&osdc->osds_timeout_work,
1638                               round_jiffies_relative(delay));
1639 }
1640
1641 static int ceph_oloc_decode(void **p, void *end,
1642                             struct ceph_object_locator *oloc)
1643 {
1644         u8 struct_v, struct_cv;
1645         u32 len;
1646         void *struct_end;
1647         int ret = 0;
1648
1649         ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
1650         struct_v = ceph_decode_8(p);
1651         struct_cv = ceph_decode_8(p);
1652         if (struct_v < 3) {
1653                 pr_warn("got v %d < 3 cv %d of ceph_object_locator\n",
1654                         struct_v, struct_cv);
1655                 goto e_inval;
1656         }
1657         if (struct_cv > 6) {
1658                 pr_warn("got v %d cv %d > 6 of ceph_object_locator\n",
1659                         struct_v, struct_cv);
1660                 goto e_inval;
1661         }
1662         len = ceph_decode_32(p);
1663         ceph_decode_need(p, end, len, e_inval);
1664         struct_end = *p + len;
1665
1666         oloc->pool = ceph_decode_64(p);
1667         *p += 4; /* skip preferred */
1668
1669         len = ceph_decode_32(p);
1670         if (len > 0) {
1671                 pr_warn("ceph_object_locator::key is set\n");
1672                 goto e_inval;
1673         }
1674
1675         if (struct_v >= 5) {
1676                 len = ceph_decode_32(p);
1677                 if (len > 0) {
1678                         pr_warn("ceph_object_locator::nspace is set\n");
1679                         goto e_inval;
1680                 }
1681         }
1682
1683         if (struct_v >= 6) {
1684                 s64 hash = ceph_decode_64(p);
1685                 if (hash != -1) {
1686                         pr_warn("ceph_object_locator::hash is set\n");
1687                         goto e_inval;
1688                 }
1689         }
1690
1691         /* skip the rest */
1692         *p = struct_end;
1693 out:
1694         return ret;
1695
1696 e_inval:
1697         ret = -EINVAL;
1698         goto out;
1699 }
1700
1701 static int ceph_redirect_decode(void **p, void *end,
1702                                 struct ceph_request_redirect *redir)
1703 {
1704         u8 struct_v, struct_cv;
1705         u32 len;
1706         void *struct_end;
1707         int ret;
1708
1709         ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
1710         struct_v = ceph_decode_8(p);
1711         struct_cv = ceph_decode_8(p);
1712         if (struct_cv > 1) {
1713                 pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n",
1714                         struct_v, struct_cv);
1715                 goto e_inval;
1716         }
1717         len = ceph_decode_32(p);
1718         ceph_decode_need(p, end, len, e_inval);
1719         struct_end = *p + len;
1720
1721         ret = ceph_oloc_decode(p, end, &redir->oloc);
1722         if (ret)
1723                 goto out;
1724
1725         len = ceph_decode_32(p);
1726         if (len > 0) {
1727                 pr_warn("ceph_request_redirect::object_name is set\n");
1728                 goto e_inval;
1729         }
1730
1731         len = ceph_decode_32(p);
1732         *p += len; /* skip osd_instructions */
1733
1734         /* skip the rest */
1735         *p = struct_end;
1736 out:
1737         return ret;
1738
1739 e_inval:
1740         ret = -EINVAL;
1741         goto out;
1742 }
1743
1744 static void complete_request(struct ceph_osd_request *req)
1745 {
1746         complete_all(&req->r_safe_completion);  /* fsync waiter */
1747 }
1748
1749 /*
1750  * handle osd op reply.  either call the callback if it is specified,
1751  * or do the completion to wake up the waiting thread.
1752  */
1753 static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg,
1754                          struct ceph_connection *con)
1755 {
1756         void *p, *end;
1757         struct ceph_osd_request *req;
1758         struct ceph_request_redirect redir;
1759         u64 tid;
1760         int object_len;
1761         unsigned int numops;
1762         int payload_len, flags;
1763         s32 result;
1764         s32 retry_attempt;
1765         struct ceph_pg pg;
1766         int err;
1767         u32 reassert_epoch;
1768         u64 reassert_version;
1769         u32 osdmap_epoch;
1770         int already_completed;
1771         u32 bytes;
1772         unsigned int i;
1773
1774         tid = le64_to_cpu(msg->hdr.tid);
1775         dout("handle_reply %p tid %llu\n", msg, tid);
1776
1777         p = msg->front.iov_base;
1778         end = p + msg->front.iov_len;
1779
1780         ceph_decode_need(&p, end, 4, bad);
1781         object_len = ceph_decode_32(&p);
1782         ceph_decode_need(&p, end, object_len, bad);
1783         p += object_len;
1784
1785         err = ceph_decode_pgid(&p, end, &pg);
1786         if (err)
1787                 goto bad;
1788
1789         ceph_decode_need(&p, end, 8 + 4 + 4 + 8 + 4, bad);
1790         flags = ceph_decode_64(&p);
1791         result = ceph_decode_32(&p);
1792         reassert_epoch = ceph_decode_32(&p);
1793         reassert_version = ceph_decode_64(&p);
1794         osdmap_epoch = ceph_decode_32(&p);
1795
1796         /* lookup */
1797         down_read(&osdc->map_sem);
1798         mutex_lock(&osdc->request_mutex);
1799         req = __lookup_request(osdc, tid);
1800         if (req == NULL) {
1801                 dout("handle_reply tid %llu dne\n", tid);
1802                 goto bad_mutex;
1803         }
1804         ceph_osdc_get_request(req);
1805
1806         dout("handle_reply %p tid %llu req %p result %d\n", msg, tid,
1807              req, result);
1808
1809         ceph_decode_need(&p, end, 4, bad_put);
1810         numops = ceph_decode_32(&p);
1811         if (numops > CEPH_OSD_MAX_OP)
1812                 goto bad_put;
1813         if (numops != req->r_num_ops)
1814                 goto bad_put;
1815         payload_len = 0;
1816         ceph_decode_need(&p, end, numops * sizeof(struct ceph_osd_op), bad_put);
1817         for (i = 0; i < numops; i++) {
1818                 struct ceph_osd_op *op = p;
1819                 int len;
1820
1821                 len = le32_to_cpu(op->payload_len);
1822                 req->r_reply_op_len[i] = len;
1823                 dout(" op %d has %d bytes\n", i, len);
1824                 payload_len += len;
1825                 p += sizeof(*op);
1826         }
1827         bytes = le32_to_cpu(msg->hdr.data_len);
1828         if (payload_len != bytes) {
1829                 pr_warn("sum of op payload lens %d != data_len %d\n",
1830                         payload_len, bytes);
1831                 goto bad_put;
1832         }
1833
1834         ceph_decode_need(&p, end, 4 + numops * 4, bad_put);
1835         retry_attempt = ceph_decode_32(&p);
1836         for (i = 0; i < numops; i++)
1837                 req->r_reply_op_result[i] = ceph_decode_32(&p);
1838
1839         if (le16_to_cpu(msg->hdr.version) >= 6) {
1840                 p += 8 + 4; /* skip replay_version */
1841                 p += 8; /* skip user_version */
1842
1843                 err = ceph_redirect_decode(&p, end, &redir);
1844                 if (err)
1845                         goto bad_put;
1846         } else {
1847                 redir.oloc.pool = -1;
1848         }
1849
1850         if (redir.oloc.pool != -1) {
1851                 dout("redirect pool %lld\n", redir.oloc.pool);
1852
1853                 __unregister_request(osdc, req);
1854
1855                 req->r_target_oloc = redir.oloc; /* struct */
1856
1857                 /*
1858                  * Start redirect requests with nofail=true.  If
1859                  * mapping fails, request will end up on the notarget
1860                  * list, waiting for the new osdmap (which can take
1861                  * a while), even though the original request mapped
1862                  * successfully.  In the future we might want to follow
1863                  * original request's nofail setting here.
1864                  */
1865                 err = __ceph_osdc_start_request(osdc, req, true);
1866                 BUG_ON(err);
1867
1868                 goto out_unlock;
1869         }
1870
1871         already_completed = req->r_got_reply;
1872         if (!req->r_got_reply) {
1873                 req->r_result = result;
1874                 dout("handle_reply result %d bytes %d\n", req->r_result,
1875                      bytes);
1876                 if (req->r_result == 0)
1877                         req->r_result = bytes;
1878
1879                 /* in case this is a write and we need to replay, */
1880                 req->r_reassert_version.epoch = cpu_to_le32(reassert_epoch);
1881                 req->r_reassert_version.version = cpu_to_le64(reassert_version);
1882
1883                 req->r_got_reply = 1;
1884         } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1885                 dout("handle_reply tid %llu dup ack\n", tid);
1886                 goto out_unlock;
1887         }
1888
1889         dout("handle_reply tid %llu flags %d\n", tid, flags);
1890
1891         if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1892                 __register_linger_request(osdc, req);
1893
1894         /* either this is a read, or we got the safe response */
1895         if (result < 0 ||
1896             (flags & CEPH_OSD_FLAG_ONDISK) ||
1897             ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1898                 __unregister_request(osdc, req);
1899
1900         mutex_unlock(&osdc->request_mutex);
1901         up_read(&osdc->map_sem);
1902
1903         if (!already_completed) {
1904                 if (req->r_unsafe_callback &&
1905                     result >= 0 && !(flags & CEPH_OSD_FLAG_ONDISK))
1906                         req->r_unsafe_callback(req, true);
1907                 if (req->r_callback)
1908                         req->r_callback(req, msg);
1909                 else
1910                         complete_all(&req->r_completion);
1911         }
1912
1913         if (flags & CEPH_OSD_FLAG_ONDISK) {
1914                 if (req->r_unsafe_callback && already_completed)
1915                         req->r_unsafe_callback(req, false);
1916                 complete_request(req);
1917         }
1918
1919 out:
1920         dout("req=%p req->r_linger=%d\n", req, req->r_linger);
1921         ceph_osdc_put_request(req);
1922         return;
1923 out_unlock:
1924         mutex_unlock(&osdc->request_mutex);
1925         up_read(&osdc->map_sem);
1926         goto out;
1927
1928 bad_put:
1929         req->r_result = -EIO;
1930         __unregister_request(osdc, req);
1931         if (req->r_callback)
1932                 req->r_callback(req, msg);
1933         else
1934                 complete_all(&req->r_completion);
1935         complete_request(req);
1936         ceph_osdc_put_request(req);
1937 bad_mutex:
1938         mutex_unlock(&osdc->request_mutex);
1939         up_read(&osdc->map_sem);
1940 bad:
1941         pr_err("corrupt osd_op_reply got %d %d\n",
1942                (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len));
1943         ceph_msg_dump(msg);
1944 }
1945
1946 static void reset_changed_osds(struct ceph_osd_client *osdc)
1947 {
1948         struct rb_node *p, *n;
1949
1950         dout("%s %p\n", __func__, osdc);
1951         for (p = rb_first(&osdc->osds); p; p = n) {
1952                 struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
1953
1954                 n = rb_next(p);
1955                 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1956                     memcmp(&osd->o_con.peer_addr,
1957                            ceph_osd_addr(osdc->osdmap,
1958                                          osd->o_osd),
1959                            sizeof(struct ceph_entity_addr)) != 0)
1960                         __reset_osd(osdc, osd);
1961         }
1962 }
1963
1964 /*
1965  * Requeue requests whose mapping to an OSD has changed.  If requests map to
1966  * no osd, request a new map.
1967  *
1968  * Caller should hold map_sem for read.
1969  */
1970 static void kick_requests(struct ceph_osd_client *osdc, bool force_resend,
1971                           bool force_resend_writes)
1972 {
1973         struct ceph_osd_request *req, *nreq;
1974         struct rb_node *p;
1975         int needmap = 0;
1976         int err;
1977         bool force_resend_req;
1978
1979         dout("kick_requests %s %s\n", force_resend ? " (force resend)" : "",
1980                 force_resend_writes ? " (force resend writes)" : "");
1981         mutex_lock(&osdc->request_mutex);
1982         for (p = rb_first(&osdc->requests); p; ) {
1983                 req = rb_entry(p, struct ceph_osd_request, r_node);
1984                 p = rb_next(p);
1985
1986                 /*
1987                  * For linger requests that have not yet been
1988                  * registered, move them to the linger list; they'll
1989                  * be sent to the osd in the loop below.  Unregister
1990                  * the request before re-registering it as a linger
1991                  * request to ensure the __map_request() below
1992                  * will decide it needs to be sent.
1993                  */
1994                 if (req->r_linger && list_empty(&req->r_linger_item)) {
1995                         dout("%p tid %llu restart on osd%d\n",
1996                              req, req->r_tid,
1997                              req->r_osd ? req->r_osd->o_osd : -1);
1998                         ceph_osdc_get_request(req);
1999                         __unregister_request(osdc, req);
2000                         __register_linger_request(osdc, req);
2001                         ceph_osdc_put_request(req);
2002                         continue;
2003                 }
2004
2005                 force_resend_req = force_resend ||
2006                         (force_resend_writes &&
2007                                 req->r_flags & CEPH_OSD_FLAG_WRITE);
2008                 err = __map_request(osdc, req, force_resend_req);
2009                 if (err < 0)
2010                         continue;  /* error */
2011                 if (req->r_osd == NULL) {
2012                         dout("%p tid %llu maps to no osd\n", req, req->r_tid);
2013                         needmap++;  /* request a newer map */
2014                 } else if (err > 0) {
2015                         if (!req->r_linger) {
2016                                 dout("%p tid %llu requeued on osd%d\n", req,
2017                                      req->r_tid,
2018                                      req->r_osd ? req->r_osd->o_osd : -1);
2019                                 req->r_flags |= CEPH_OSD_FLAG_RETRY;
2020                         }
2021                 }
2022         }
2023
2024         list_for_each_entry_safe(req, nreq, &osdc->req_linger,
2025                                  r_linger_item) {
2026                 dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
2027
2028                 err = __map_request(osdc, req,
2029                                     force_resend || force_resend_writes);
2030                 dout("__map_request returned %d\n", err);
2031                 if (err < 0)
2032                         continue;  /* hrm! */
2033                 if (req->r_osd == NULL || err > 0) {
2034                         if (req->r_osd == NULL) {
2035                                 dout("lingering %p tid %llu maps to no osd\n",
2036                                      req, req->r_tid);
2037                                 /*
2038                                  * A homeless lingering request makes
2039                                  * no sense, as it's job is to keep
2040                                  * a particular OSD connection open.
2041                                  * Request a newer map and kick the
2042                                  * request, knowing that it won't be
2043                                  * resent until we actually get a map
2044                                  * that can tell us where to send it.
2045                                  */
2046                                 needmap++;
2047                         }
2048
2049                         dout("kicking lingering %p tid %llu osd%d\n", req,
2050                              req->r_tid, req->r_osd ? req->r_osd->o_osd : -1);
2051                         __register_request(osdc, req);
2052                         __unregister_linger_request(osdc, req);
2053                 }
2054         }
2055         reset_changed_osds(osdc);
2056         mutex_unlock(&osdc->request_mutex);
2057
2058         if (needmap) {
2059                 dout("%d requests for down osds, need new map\n", needmap);
2060                 ceph_monc_request_next_osdmap(&osdc->client->monc);
2061         }
2062 }
2063
2064
2065 /*
2066  * Process updated osd map.
2067  *
2068  * The message contains any number of incremental and full maps, normally
2069  * indicating some sort of topology change in the cluster.  Kick requests
2070  * off to different OSDs as needed.
2071  */
2072 void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
2073 {
2074         void *p, *end, *next;
2075         u32 nr_maps, maplen;
2076         u32 epoch;
2077         struct ceph_osdmap *newmap = NULL, *oldmap;
2078         int err;
2079         struct ceph_fsid fsid;
2080         bool was_full;
2081
2082         dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
2083         p = msg->front.iov_base;
2084         end = p + msg->front.iov_len;
2085
2086         /* verify fsid */
2087         ceph_decode_need(&p, end, sizeof(fsid), bad);
2088         ceph_decode_copy(&p, &fsid, sizeof(fsid));
2089         if (ceph_check_fsid(osdc->client, &fsid) < 0)
2090                 return;
2091
2092         down_write(&osdc->map_sem);
2093
2094         was_full = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
2095
2096         /* incremental maps */
2097         ceph_decode_32_safe(&p, end, nr_maps, bad);
2098         dout(" %d inc maps\n", nr_maps);
2099         while (nr_maps > 0) {
2100                 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
2101                 epoch = ceph_decode_32(&p);
2102                 maplen = ceph_decode_32(&p);
2103                 ceph_decode_need(&p, end, maplen, bad);
2104                 next = p + maplen;
2105                 if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
2106                         dout("applying incremental map %u len %d\n",
2107                              epoch, maplen);
2108                         newmap = osdmap_apply_incremental(&p, next,
2109                                                           osdc->osdmap,
2110                                                           &osdc->client->msgr);
2111                         if (IS_ERR(newmap)) {
2112                                 err = PTR_ERR(newmap);
2113                                 goto bad;
2114                         }
2115                         BUG_ON(!newmap);
2116                         if (newmap != osdc->osdmap) {
2117                                 ceph_osdmap_destroy(osdc->osdmap);
2118                                 osdc->osdmap = newmap;
2119                         }
2120                         was_full = was_full ||
2121                                 ceph_osdmap_flag(osdc->osdmap,
2122                                                  CEPH_OSDMAP_FULL);
2123                         kick_requests(osdc, 0, was_full);
2124                 } else {
2125                         dout("ignoring incremental map %u len %d\n",
2126                              epoch, maplen);
2127                 }
2128                 p = next;
2129                 nr_maps--;
2130         }
2131         if (newmap)
2132                 goto done;
2133
2134         /* full maps */
2135         ceph_decode_32_safe(&p, end, nr_maps, bad);
2136         dout(" %d full maps\n", nr_maps);
2137         while (nr_maps) {
2138                 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
2139                 epoch = ceph_decode_32(&p);
2140                 maplen = ceph_decode_32(&p);
2141                 ceph_decode_need(&p, end, maplen, bad);
2142                 if (nr_maps > 1) {
2143                         dout("skipping non-latest full map %u len %d\n",
2144                              epoch, maplen);
2145                 } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
2146                         dout("skipping full map %u len %d, "
2147                              "older than our %u\n", epoch, maplen,
2148                              osdc->osdmap->epoch);
2149                 } else {
2150                         int skipped_map = 0;
2151
2152                         dout("taking full map %u len %d\n", epoch, maplen);
2153                         newmap = ceph_osdmap_decode(&p, p+maplen);
2154                         if (IS_ERR(newmap)) {
2155                                 err = PTR_ERR(newmap);
2156                                 goto bad;
2157                         }
2158                         BUG_ON(!newmap);
2159                         oldmap = osdc->osdmap;
2160                         osdc->osdmap = newmap;
2161                         if (oldmap) {
2162                                 if (oldmap->epoch + 1 < newmap->epoch)
2163                                         skipped_map = 1;
2164                                 ceph_osdmap_destroy(oldmap);
2165                         }
2166                         was_full = was_full ||
2167                                 ceph_osdmap_flag(osdc->osdmap,
2168                                                  CEPH_OSDMAP_FULL);
2169                         kick_requests(osdc, skipped_map, was_full);
2170                 }
2171                 p += maplen;
2172                 nr_maps--;
2173         }
2174
2175         if (!osdc->osdmap)
2176                 goto bad;
2177 done:
2178         downgrade_write(&osdc->map_sem);
2179         ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
2180
2181         /*
2182          * subscribe to subsequent osdmap updates if full to ensure
2183          * we find out when we are no longer full and stop returning
2184          * ENOSPC.
2185          */
2186         if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) ||
2187                 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD) ||
2188                 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR))
2189                 ceph_monc_request_next_osdmap(&osdc->client->monc);
2190
2191         mutex_lock(&osdc->request_mutex);
2192         __send_queued(osdc);
2193         mutex_unlock(&osdc->request_mutex);
2194         up_read(&osdc->map_sem);
2195         wake_up_all(&osdc->client->auth_wq);
2196         return;
2197
2198 bad:
2199         pr_err("osdc handle_map corrupt msg\n");
2200         ceph_msg_dump(msg);
2201         up_write(&osdc->map_sem);
2202 }
2203
2204 /*
2205  * watch/notify callback event infrastructure
2206  *
2207  * These callbacks are used both for watch and notify operations.
2208  */
2209 static void __release_event(struct kref *kref)
2210 {
2211         struct ceph_osd_event *event =
2212                 container_of(kref, struct ceph_osd_event, kref);
2213
2214         dout("__release_event %p\n", event);
2215         kfree(event);
2216 }
2217
2218 static void get_event(struct ceph_osd_event *event)
2219 {
2220         kref_get(&event->kref);
2221 }
2222
2223 void ceph_osdc_put_event(struct ceph_osd_event *event)
2224 {
2225         kref_put(&event->kref, __release_event);
2226 }
2227 EXPORT_SYMBOL(ceph_osdc_put_event);
2228
2229 static void __insert_event(struct ceph_osd_client *osdc,
2230                              struct ceph_osd_event *new)
2231 {
2232         struct rb_node **p = &osdc->event_tree.rb_node;
2233         struct rb_node *parent = NULL;
2234         struct ceph_osd_event *event = NULL;
2235
2236         while (*p) {
2237                 parent = *p;
2238                 event = rb_entry(parent, struct ceph_osd_event, node);
2239                 if (new->cookie < event->cookie)
2240                         p = &(*p)->rb_left;
2241                 else if (new->cookie > event->cookie)
2242                         p = &(*p)->rb_right;
2243                 else
2244                         BUG();
2245         }
2246
2247         rb_link_node(&new->node, parent, p);
2248         rb_insert_color(&new->node, &osdc->event_tree);
2249 }
2250
2251 static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
2252                                                 u64 cookie)
2253 {
2254         struct rb_node **p = &osdc->event_tree.rb_node;
2255         struct rb_node *parent = NULL;
2256         struct ceph_osd_event *event = NULL;
2257
2258         while (*p) {
2259                 parent = *p;
2260                 event = rb_entry(parent, struct ceph_osd_event, node);
2261                 if (cookie < event->cookie)
2262                         p = &(*p)->rb_left;
2263                 else if (cookie > event->cookie)
2264                         p = &(*p)->rb_right;
2265                 else
2266                         return event;
2267         }
2268         return NULL;
2269 }
2270
2271 static void __remove_event(struct ceph_osd_event *event)
2272 {
2273         struct ceph_osd_client *osdc = event->osdc;
2274
2275         if (!RB_EMPTY_NODE(&event->node)) {
2276                 dout("__remove_event removed %p\n", event);
2277                 rb_erase(&event->node, &osdc->event_tree);
2278                 ceph_osdc_put_event(event);
2279         } else {
2280                 dout("__remove_event didn't remove %p\n", event);
2281         }
2282 }
2283
2284 int ceph_osdc_create_event(struct ceph_osd_client *osdc,
2285                            void (*event_cb)(u64, u64, u8, void *),
2286                            void *data, struct ceph_osd_event **pevent)
2287 {
2288         struct ceph_osd_event *event;
2289
2290         event = kmalloc(sizeof(*event), GFP_NOIO);
2291         if (!event)
2292                 return -ENOMEM;
2293
2294         dout("create_event %p\n", event);
2295         event->cb = event_cb;
2296         event->one_shot = 0;
2297         event->data = data;
2298         event->osdc = osdc;
2299         INIT_LIST_HEAD(&event->osd_node);
2300         RB_CLEAR_NODE(&event->node);
2301         kref_init(&event->kref);   /* one ref for us */
2302         kref_get(&event->kref);    /* one ref for the caller */
2303
2304         spin_lock(&osdc->event_lock);
2305         event->cookie = ++osdc->event_count;
2306         __insert_event(osdc, event);
2307         spin_unlock(&osdc->event_lock);
2308
2309         *pevent = event;
2310         return 0;
2311 }
2312 EXPORT_SYMBOL(ceph_osdc_create_event);
2313
2314 void ceph_osdc_cancel_event(struct ceph_osd_event *event)
2315 {
2316         struct ceph_osd_client *osdc = event->osdc;
2317
2318         dout("cancel_event %p\n", event);
2319         spin_lock(&osdc->event_lock);
2320         __remove_event(event);
2321         spin_unlock(&osdc->event_lock);
2322         ceph_osdc_put_event(event); /* caller's */
2323 }
2324 EXPORT_SYMBOL(ceph_osdc_cancel_event);
2325
2326
2327 static void do_event_work(struct work_struct *work)
2328 {
2329         struct ceph_osd_event_work *event_work =
2330                 container_of(work, struct ceph_osd_event_work, work);
2331         struct ceph_osd_event *event = event_work->event;
2332         u64 ver = event_work->ver;
2333         u64 notify_id = event_work->notify_id;
2334         u8 opcode = event_work->opcode;
2335
2336         dout("do_event_work completing %p\n", event);
2337         event->cb(ver, notify_id, opcode, event->data);
2338         dout("do_event_work completed %p\n", event);
2339         ceph_osdc_put_event(event);
2340         kfree(event_work);
2341 }
2342
2343
2344 /*
2345  * Process osd watch notifications
2346  */
2347 static void handle_watch_notify(struct ceph_osd_client *osdc,
2348                                 struct ceph_msg *msg)
2349 {
2350         void *p, *end;
2351         u8 proto_ver;
2352         u64 cookie, ver, notify_id;
2353         u8 opcode;
2354         struct ceph_osd_event *event;
2355         struct ceph_osd_event_work *event_work;
2356
2357         p = msg->front.iov_base;
2358         end = p + msg->front.iov_len;
2359
2360         ceph_decode_8_safe(&p, end, proto_ver, bad);
2361         ceph_decode_8_safe(&p, end, opcode, bad);
2362         ceph_decode_64_safe(&p, end, cookie, bad);
2363         ceph_decode_64_safe(&p, end, ver, bad);
2364         ceph_decode_64_safe(&p, end, notify_id, bad);
2365
2366         spin_lock(&osdc->event_lock);
2367         event = __find_event(osdc, cookie);
2368         if (event) {
2369                 BUG_ON(event->one_shot);
2370                 get_event(event);
2371         }
2372         spin_unlock(&osdc->event_lock);
2373         dout("handle_watch_notify cookie %lld ver %lld event %p\n",
2374              cookie, ver, event);
2375         if (event) {
2376                 event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
2377                 if (!event_work) {
2378                         pr_err("couldn't allocate event_work\n");
2379                         ceph_osdc_put_event(event);
2380                         return;
2381                 }
2382                 INIT_WORK(&event_work->work, do_event_work);
2383                 event_work->event = event;
2384                 event_work->ver = ver;
2385                 event_work->notify_id = notify_id;
2386                 event_work->opcode = opcode;
2387
2388                 queue_work(osdc->notify_wq, &event_work->work);
2389         }
2390
2391         return;
2392
2393 bad:
2394         pr_err("osdc handle_watch_notify corrupt msg\n");
2395 }
2396
2397 /*
2398  * build new request AND message
2399  *
2400  */
2401 void ceph_osdc_build_request(struct ceph_osd_request *req, u64 off,
2402                                 struct ceph_snap_context *snapc, u64 snap_id,
2403                                 struct timespec *mtime)
2404 {
2405         struct ceph_msg *msg = req->r_request;
2406         void *p;
2407         size_t msg_size;
2408         int flags = req->r_flags;
2409         u64 data_len;
2410         unsigned int i;
2411
2412         req->r_snapid = snap_id;
2413         req->r_snapc = ceph_get_snap_context(snapc);
2414
2415         /* encode request */
2416         msg->hdr.version = cpu_to_le16(4);
2417
2418         p = msg->front.iov_base;
2419         ceph_encode_32(&p, 1);   /* client_inc  is always 1 */
2420         req->r_request_osdmap_epoch = p;
2421         p += 4;
2422         req->r_request_flags = p;
2423         p += 4;
2424         if (req->r_flags & CEPH_OSD_FLAG_WRITE)
2425                 ceph_encode_timespec(p, mtime);
2426         p += sizeof(struct ceph_timespec);
2427         req->r_request_reassert_version = p;
2428         p += sizeof(struct ceph_eversion); /* will get filled in */
2429
2430         /* oloc */
2431         ceph_encode_8(&p, 4);
2432         ceph_encode_8(&p, 4);
2433         ceph_encode_32(&p, 8 + 4 + 4);
2434         req->r_request_pool = p;
2435         p += 8;
2436         ceph_encode_32(&p, -1);  /* preferred */
2437         ceph_encode_32(&p, 0);   /* key len */
2438
2439         ceph_encode_8(&p, 1);
2440         req->r_request_pgid = p;
2441         p += 8 + 4;
2442         ceph_encode_32(&p, -1);  /* preferred */
2443
2444         /* oid */
2445         ceph_encode_32(&p, req->r_base_oid.name_len);
2446         memcpy(p, req->r_base_oid.name, req->r_base_oid.name_len);
2447         dout("oid '%.*s' len %d\n", req->r_base_oid.name_len,
2448              req->r_base_oid.name, req->r_base_oid.name_len);
2449         p += req->r_base_oid.name_len;
2450
2451         /* ops--can imply data */
2452         ceph_encode_16(&p, (u16)req->r_num_ops);
2453         data_len = 0;
2454         for (i = 0; i < req->r_num_ops; i++) {
2455                 data_len += osd_req_encode_op(req, p, i);
2456                 p += sizeof(struct ceph_osd_op);
2457         }
2458
2459         /* snaps */
2460         ceph_encode_64(&p, req->r_snapid);
2461         ceph_encode_64(&p, req->r_snapc ? req->r_snapc->seq : 0);
2462         ceph_encode_32(&p, req->r_snapc ? req->r_snapc->num_snaps : 0);
2463         if (req->r_snapc) {
2464                 for (i = 0; i < snapc->num_snaps; i++) {
2465                         ceph_encode_64(&p, req->r_snapc->snaps[i]);
2466                 }
2467         }
2468
2469         req->r_request_attempts = p;
2470         p += 4;
2471
2472         /* data */
2473         if (flags & CEPH_OSD_FLAG_WRITE) {
2474                 u16 data_off;
2475
2476                 /*
2477                  * The header "data_off" is a hint to the receiver
2478                  * allowing it to align received data into its
2479                  * buffers such that there's no need to re-copy
2480                  * it before writing it to disk (direct I/O).
2481                  */
2482                 data_off = (u16) (off & 0xffff);
2483                 req->r_request->hdr.data_off = cpu_to_le16(data_off);
2484         }
2485         req->r_request->hdr.data_len = cpu_to_le32(data_len);
2486
2487         BUG_ON(p > msg->front.iov_base + msg->front.iov_len);
2488         msg_size = p - msg->front.iov_base;
2489         msg->front.iov_len = msg_size;
2490         msg->hdr.front_len = cpu_to_le32(msg_size);
2491
2492         dout("build_request msg_size was %d\n", (int)msg_size);
2493 }
2494 EXPORT_SYMBOL(ceph_osdc_build_request);
2495
2496 /*
2497  * Register request, send initial attempt.
2498  */
2499 int ceph_osdc_start_request(struct ceph_osd_client *osdc,
2500                             struct ceph_osd_request *req,
2501                             bool nofail)
2502 {
2503         int rc;
2504
2505         down_read(&osdc->map_sem);
2506         mutex_lock(&osdc->request_mutex);
2507
2508         rc = __ceph_osdc_start_request(osdc, req, nofail);
2509
2510         mutex_unlock(&osdc->request_mutex);
2511         up_read(&osdc->map_sem);
2512
2513         return rc;
2514 }
2515 EXPORT_SYMBOL(ceph_osdc_start_request);
2516
2517 /*
2518  * Unregister a registered request.  The request is not completed (i.e.
2519  * no callbacks or wakeups) - higher layers are supposed to know what
2520  * they are canceling.
2521  */
2522 void ceph_osdc_cancel_request(struct ceph_osd_request *req)
2523 {
2524         struct ceph_osd_client *osdc = req->r_osdc;
2525
2526         mutex_lock(&osdc->request_mutex);
2527         if (req->r_linger)
2528                 __unregister_linger_request(osdc, req);
2529         __unregister_request(osdc, req);
2530         mutex_unlock(&osdc->request_mutex);
2531
2532         dout("%s %p tid %llu canceled\n", __func__, req, req->r_tid);
2533 }
2534 EXPORT_SYMBOL(ceph_osdc_cancel_request);
2535
2536 /*
2537  * wait for a request to complete
2538  */
2539 int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
2540                            struct ceph_osd_request *req)
2541 {
2542         int rc;
2543
2544         dout("%s %p tid %llu\n", __func__, req, req->r_tid);
2545
2546         rc = wait_for_completion_interruptible(&req->r_completion);
2547         if (rc < 0) {
2548                 dout("%s %p tid %llu interrupted\n", __func__, req, req->r_tid);
2549                 ceph_osdc_cancel_request(req);
2550                 complete_request(req);
2551                 return rc;
2552         }
2553
2554         dout("%s %p tid %llu result %d\n", __func__, req, req->r_tid,
2555              req->r_result);
2556         return req->r_result;
2557 }
2558 EXPORT_SYMBOL(ceph_osdc_wait_request);
2559
2560 /*
2561  * sync - wait for all in-flight requests to flush.  avoid starvation.
2562  */
2563 void ceph_osdc_sync(struct ceph_osd_client *osdc)
2564 {
2565         struct ceph_osd_request *req;
2566         u64 last_tid, next_tid = 0;
2567
2568         mutex_lock(&osdc->request_mutex);
2569         last_tid = osdc->last_tid;
2570         while (1) {
2571                 req = __lookup_request_ge(osdc, next_tid);
2572                 if (!req)
2573                         break;
2574                 if (req->r_tid > last_tid)
2575                         break;
2576
2577                 next_tid = req->r_tid + 1;
2578                 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
2579                         continue;
2580
2581                 ceph_osdc_get_request(req);
2582                 mutex_unlock(&osdc->request_mutex);
2583                 dout("sync waiting on tid %llu (last is %llu)\n",
2584                      req->r_tid, last_tid);
2585                 wait_for_completion(&req->r_safe_completion);
2586                 mutex_lock(&osdc->request_mutex);
2587                 ceph_osdc_put_request(req);
2588         }
2589         mutex_unlock(&osdc->request_mutex);
2590         dout("sync done (thru tid %llu)\n", last_tid);
2591 }
2592 EXPORT_SYMBOL(ceph_osdc_sync);
2593
2594 /*
2595  * Call all pending notify callbacks - for use after a watch is
2596  * unregistered, to make sure no more callbacks for it will be invoked
2597  */
2598 void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
2599 {
2600         flush_workqueue(osdc->notify_wq);
2601 }
2602 EXPORT_SYMBOL(ceph_osdc_flush_notifies);
2603
2604
2605 /*
2606  * init, shutdown
2607  */
2608 int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
2609 {
2610         int err;
2611
2612         dout("init\n");
2613         osdc->client = client;
2614         osdc->osdmap = NULL;
2615         init_rwsem(&osdc->map_sem);
2616         init_completion(&osdc->map_waiters);
2617         osdc->last_requested_map = 0;
2618         mutex_init(&osdc->request_mutex);
2619         osdc->last_tid = 0;
2620         osdc->osds = RB_ROOT;
2621         INIT_LIST_HEAD(&osdc->osd_lru);
2622         osdc->requests = RB_ROOT;
2623         INIT_LIST_HEAD(&osdc->req_lru);
2624         INIT_LIST_HEAD(&osdc->req_unsent);
2625         INIT_LIST_HEAD(&osdc->req_notarget);
2626         INIT_LIST_HEAD(&osdc->req_linger);
2627         osdc->num_requests = 0;
2628         INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
2629         INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
2630         spin_lock_init(&osdc->event_lock);
2631         osdc->event_tree = RB_ROOT;
2632         osdc->event_count = 0;
2633
2634         schedule_delayed_work(&osdc->osds_timeout_work,
2635             round_jiffies_relative(osdc->client->options->osd_idle_ttl));
2636
2637         err = -ENOMEM;
2638         osdc->req_mempool = mempool_create_kmalloc_pool(10,
2639                                         sizeof(struct ceph_osd_request));
2640         if (!osdc->req_mempool)
2641                 goto out;
2642
2643         err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
2644                                 OSD_OP_FRONT_LEN, 10, true,
2645                                 "osd_op");
2646         if (err < 0)
2647                 goto out_mempool;
2648         err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
2649                                 OSD_OPREPLY_FRONT_LEN, 10, true,
2650                                 "osd_op_reply");
2651         if (err < 0)
2652                 goto out_msgpool;
2653
2654         err = -ENOMEM;
2655         osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
2656         if (!osdc->notify_wq)
2657                 goto out_msgpool_reply;
2658
2659         return 0;
2660
2661 out_msgpool_reply:
2662         ceph_msgpool_destroy(&osdc->msgpool_op_reply);
2663 out_msgpool:
2664         ceph_msgpool_destroy(&osdc->msgpool_op);
2665 out_mempool:
2666         mempool_destroy(osdc->req_mempool);
2667 out:
2668         return err;
2669 }
2670
2671 void ceph_osdc_stop(struct ceph_osd_client *osdc)
2672 {
2673         flush_workqueue(osdc->notify_wq);
2674         destroy_workqueue(osdc->notify_wq);
2675         cancel_delayed_work_sync(&osdc->timeout_work);
2676         cancel_delayed_work_sync(&osdc->osds_timeout_work);
2677         if (osdc->osdmap) {
2678                 ceph_osdmap_destroy(osdc->osdmap);
2679                 osdc->osdmap = NULL;
2680         }
2681         remove_all_osds(osdc);
2682         mempool_destroy(osdc->req_mempool);
2683         ceph_msgpool_destroy(&osdc->msgpool_op);
2684         ceph_msgpool_destroy(&osdc->msgpool_op_reply);
2685 }
2686
2687 /*
2688  * Read some contiguous pages.  If we cross a stripe boundary, shorten
2689  * *plen.  Return number of bytes read, or error.
2690  */
2691 int ceph_osdc_readpages(struct ceph_osd_client *osdc,
2692                         struct ceph_vino vino, struct ceph_file_layout *layout,
2693                         u64 off, u64 *plen,
2694                         u32 truncate_seq, u64 truncate_size,
2695                         struct page **pages, int num_pages, int page_align)
2696 {
2697         struct ceph_osd_request *req;
2698         int rc = 0;
2699
2700         dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
2701              vino.snap, off, *plen);
2702         req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 0, 1,
2703                                     CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
2704                                     NULL, truncate_seq, truncate_size,
2705                                     false);
2706         if (IS_ERR(req))
2707                 return PTR_ERR(req);
2708
2709         /* it may be a short read due to an object boundary */
2710
2711         osd_req_op_extent_osd_data_pages(req, 0,
2712                                 pages, *plen, page_align, false, false);
2713
2714         dout("readpages  final extent is %llu~%llu (%llu bytes align %d)\n",
2715              off, *plen, *plen, page_align);
2716
2717         ceph_osdc_build_request(req, off, NULL, vino.snap, NULL);
2718
2719         rc = ceph_osdc_start_request(osdc, req, false);
2720         if (!rc)
2721                 rc = ceph_osdc_wait_request(osdc, req);
2722
2723         ceph_osdc_put_request(req);
2724         dout("readpages result %d\n", rc);
2725         return rc;
2726 }
2727 EXPORT_SYMBOL(ceph_osdc_readpages);
2728
2729 /*
2730  * do a synchronous write on N pages
2731  */
2732 int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
2733                          struct ceph_file_layout *layout,
2734                          struct ceph_snap_context *snapc,
2735                          u64 off, u64 len,
2736                          u32 truncate_seq, u64 truncate_size,
2737                          struct timespec *mtime,
2738                          struct page **pages, int num_pages)
2739 {
2740         struct ceph_osd_request *req;
2741         int rc = 0;
2742         int page_align = off & ~PAGE_MASK;
2743
2744         BUG_ON(vino.snap != CEPH_NOSNAP);       /* snapshots aren't writeable */
2745         req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 0, 1,
2746                                     CEPH_OSD_OP_WRITE,
2747                                     CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
2748                                     snapc, truncate_seq, truncate_size,
2749                                     true);
2750         if (IS_ERR(req))
2751                 return PTR_ERR(req);
2752
2753         /* it may be a short write due to an object boundary */
2754         osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align,
2755                                 false, false);
2756         dout("writepages %llu~%llu (%llu bytes)\n", off, len, len);
2757
2758         ceph_osdc_build_request(req, off, snapc, CEPH_NOSNAP, mtime);
2759
2760         rc = ceph_osdc_start_request(osdc, req, true);
2761         if (!rc)
2762                 rc = ceph_osdc_wait_request(osdc, req);
2763
2764         ceph_osdc_put_request(req);
2765         if (rc == 0)
2766                 rc = len;
2767         dout("writepages result %d\n", rc);
2768         return rc;
2769 }
2770 EXPORT_SYMBOL(ceph_osdc_writepages);
2771
2772 int ceph_osdc_setup(void)
2773 {
2774         BUG_ON(ceph_osd_request_cache);
2775         ceph_osd_request_cache = kmem_cache_create("ceph_osd_request",
2776                                         sizeof (struct ceph_osd_request),
2777                                         __alignof__(struct ceph_osd_request),
2778                                         0, NULL);
2779
2780         return ceph_osd_request_cache ? 0 : -ENOMEM;
2781 }
2782 EXPORT_SYMBOL(ceph_osdc_setup);
2783
2784 void ceph_osdc_cleanup(void)
2785 {
2786         BUG_ON(!ceph_osd_request_cache);
2787         kmem_cache_destroy(ceph_osd_request_cache);
2788         ceph_osd_request_cache = NULL;
2789 }
2790 EXPORT_SYMBOL(ceph_osdc_cleanup);
2791
2792 /*
2793  * handle incoming message
2794  */
2795 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
2796 {
2797         struct ceph_osd *osd = con->private;
2798         struct ceph_osd_client *osdc;
2799         int type = le16_to_cpu(msg->hdr.type);
2800
2801         if (!osd)
2802                 goto out;
2803         osdc = osd->o_osdc;
2804
2805         switch (type) {
2806         case CEPH_MSG_OSD_MAP:
2807                 ceph_osdc_handle_map(osdc, msg);
2808                 break;
2809         case CEPH_MSG_OSD_OPREPLY:
2810                 handle_reply(osdc, msg, con);
2811                 break;
2812         case CEPH_MSG_WATCH_NOTIFY:
2813                 handle_watch_notify(osdc, msg);
2814                 break;
2815
2816         default:
2817                 pr_err("received unknown message type %d %s\n", type,
2818                        ceph_msg_type_name(type));
2819         }
2820 out:
2821         ceph_msg_put(msg);
2822 }
2823
2824 /*
2825  * Lookup and return message for incoming reply.  Don't try to do
2826  * anything about a larger than preallocated data portion of the
2827  * message at the moment - for now, just skip the message.
2828  */
2829 static struct ceph_msg *get_reply(struct ceph_connection *con,
2830                                   struct ceph_msg_header *hdr,
2831                                   int *skip)
2832 {
2833         struct ceph_osd *osd = con->private;
2834         struct ceph_osd_client *osdc = osd->o_osdc;
2835         struct ceph_msg *m;
2836         struct ceph_osd_request *req;
2837         int front_len = le32_to_cpu(hdr->front_len);
2838         int data_len = le32_to_cpu(hdr->data_len);
2839         u64 tid;
2840
2841         tid = le64_to_cpu(hdr->tid);
2842         mutex_lock(&osdc->request_mutex);
2843         req = __lookup_request(osdc, tid);
2844         if (!req) {
2845                 pr_warn("%s osd%d tid %llu unknown, skipping\n",
2846                         __func__, osd->o_osd, tid);
2847                 m = NULL;
2848                 *skip = 1;
2849                 goto out;
2850         }
2851
2852         if (req->r_reply->con)
2853                 dout("%s revoking msg %p from old con %p\n", __func__,
2854                      req->r_reply, req->r_reply->con);
2855         ceph_msg_revoke_incoming(req->r_reply);
2856
2857         if (front_len > req->r_reply->front_alloc_len) {
2858                 pr_warn("%s osd%d tid %llu front %d > preallocated %d\n",
2859                         __func__, osd->o_osd, req->r_tid, front_len,
2860                         req->r_reply->front_alloc_len);
2861                 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS,
2862                                  false);
2863                 if (!m)
2864                         goto out;
2865                 ceph_msg_put(req->r_reply);
2866                 req->r_reply = m;
2867         }
2868
2869         if (data_len > req->r_reply->data_length) {
2870                 pr_warn("%s osd%d tid %llu data %d > preallocated %zu, skipping\n",
2871                         __func__, osd->o_osd, req->r_tid, data_len,
2872                         req->r_reply->data_length);
2873                 m = NULL;
2874                 *skip = 1;
2875                 goto out;
2876         }
2877
2878         m = ceph_msg_get(req->r_reply);
2879         dout("get_reply tid %lld %p\n", tid, m);
2880
2881 out:
2882         mutex_unlock(&osdc->request_mutex);
2883         return m;
2884 }
2885
2886 static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2887                                   struct ceph_msg_header *hdr,
2888                                   int *skip)
2889 {
2890         struct ceph_osd *osd = con->private;
2891         int type = le16_to_cpu(hdr->type);
2892         int front = le32_to_cpu(hdr->front_len);
2893
2894         *skip = 0;
2895         switch (type) {
2896         case CEPH_MSG_OSD_MAP:
2897         case CEPH_MSG_WATCH_NOTIFY:
2898                 return ceph_msg_new(type, front, GFP_NOFS, false);
2899         case CEPH_MSG_OSD_OPREPLY:
2900                 return get_reply(con, hdr, skip);
2901         default:
2902                 pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2903                         osd->o_osd);
2904                 *skip = 1;
2905                 return NULL;
2906         }
2907 }
2908
2909 /*
2910  * Wrappers to refcount containing ceph_osd struct
2911  */
2912 static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2913 {
2914         struct ceph_osd *osd = con->private;
2915         if (get_osd(osd))
2916                 return con;
2917         return NULL;
2918 }
2919
2920 static void put_osd_con(struct ceph_connection *con)
2921 {
2922         struct ceph_osd *osd = con->private;
2923         put_osd(osd);
2924 }
2925
2926 /*
2927  * authentication
2928  */
2929 /*
2930  * Note: returned pointer is the address of a structure that's
2931  * managed separately.  Caller must *not* attempt to free it.
2932  */
2933 static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
2934                                         int *proto, int force_new)
2935 {
2936         struct ceph_osd *o = con->private;
2937         struct ceph_osd_client *osdc = o->o_osdc;
2938         struct ceph_auth_client *ac = osdc->client->monc.auth;
2939         struct ceph_auth_handshake *auth = &o->o_auth;
2940
2941         if (force_new && auth->authorizer) {
2942                 ceph_auth_destroy_authorizer(ac, auth->authorizer);
2943                 auth->authorizer = NULL;
2944         }
2945         if (!auth->authorizer) {
2946                 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2947                                                       auth);
2948                 if (ret)
2949                         return ERR_PTR(ret);
2950         } else {
2951                 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2952                                                      auth);
2953                 if (ret)
2954                         return ERR_PTR(ret);
2955         }
2956         *proto = ac->protocol;
2957
2958         return auth;
2959 }
2960
2961
2962 static int verify_authorizer_reply(struct ceph_connection *con, int len)
2963 {
2964         struct ceph_osd *o = con->private;
2965         struct ceph_osd_client *osdc = o->o_osdc;
2966         struct ceph_auth_client *ac = osdc->client->monc.auth;
2967
2968         return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer, len);
2969 }
2970
2971 static int invalidate_authorizer(struct ceph_connection *con)
2972 {
2973         struct ceph_osd *o = con->private;
2974         struct ceph_osd_client *osdc = o->o_osdc;
2975         struct ceph_auth_client *ac = osdc->client->monc.auth;
2976
2977         ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
2978         return ceph_monc_validate_auth(&osdc->client->monc);
2979 }
2980
2981 static int sign_message(struct ceph_connection *con, struct ceph_msg *msg)
2982 {
2983         struct ceph_osd *o = con->private;
2984         struct ceph_auth_handshake *auth = &o->o_auth;
2985         return ceph_auth_sign_message(auth, msg);
2986 }
2987
2988 static int check_message_signature(struct ceph_connection *con, struct ceph_msg *msg)
2989 {
2990         struct ceph_osd *o = con->private;
2991         struct ceph_auth_handshake *auth = &o->o_auth;
2992         return ceph_auth_check_message_signature(auth, msg);
2993 }
2994
2995 static const struct ceph_connection_operations osd_con_ops = {
2996         .get = get_osd_con,
2997         .put = put_osd_con,
2998         .dispatch = dispatch,
2999         .get_authorizer = get_authorizer,
3000         .verify_authorizer_reply = verify_authorizer_reply,
3001         .invalidate_authorizer = invalidate_authorizer,
3002         .alloc_msg = alloc_msg,
3003         .sign_message = sign_message,
3004         .check_message_signature = check_message_signature,
3005         .fault = osd_reset,
3006 };