ceph/src/osdc/Objecter.h

   1 // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
   2 // vim: ts=8 sw=2 smarttab
   3 /*
   4  * Ceph - scalable distributed file system
   5  *
   6  * Copyright (C) 2004-2006 Sage Weil <sage@newdream.net>
   7  *
   8  * This is free software; you can redistribute it and/or
   9  * modify it under the terms of the GNU Lesser General Public
  10  * License version 2.1, as published by the Free Software
  11  * Foundation.  See file COPYING.
  12  *
  13  */
  14
  15 #ifndef CEPH_OBJECTER_H
  16 #define CEPH_OBJECTER_H
  17
  18 #include <condition_variable>
  19 #include <list>
  20 #include <map>
  21 #include <mutex>
  22 #include <memory>
  23 #include <sstream>
  24 #include <type_traits>
  25
  26 #include <boost/thread/shared_mutex.hpp>
  27
  28 #include "include/assert.h"
  29 #include "include/buffer.h"
  30 #include "include/types.h"
  31 #include "include/rados/rados_types.hpp"
  32
  33 #include "common/admin_socket.h"
  34 #include "common/ceph_time.h"
  35 #include "common/ceph_timer.h"
  36 #include "common/Finisher.h"
  37 #include "common/shunique_lock.h"
  38 #include "common/zipkin_trace.h"
  39
  40 #include "messages/MOSDOp.h"
  41 #include "osd/OSDMap.h"
  42
  43 using namespace std;
  44
  45 class Context;
  46 class Messenger;
  47 class OSDMap;
  48 class MonClient;
  49 class Message;
  50 class Finisher;
  51
  52 class MPoolOpReply;
  53
  54 class MGetPoolStatsReply;
  55 class MStatfsReply;
  56 class MCommandReply;
  57 class MWatchNotify;
  58
  59 class PerfCounters;
  60
  61 // -----------------------------------------
  62
  63 struct ObjectOperation {
  64   vector<OSDOp> ops;
  65   int flags;
  66   int priority;
  67
  68   vector<bufferlist*> out_bl;
  69   vector<Context*> out_handler;
  70   vector<int*> out_rval;
  71
  72   ObjectOperation() : flags(0), priority(0) {}
  73   ~ObjectOperation() {
  74     while (!out_handler.empty()) {
  75       delete out_handler.back();
  76       out_handler.pop_back();
  77     }
  78   }
  79
  80   size_t size() {
  81     return ops.size();
  82   }
  83
  84   void set_last_op_flags(int flags) {
  85     assert(!ops.empty());
  86     ops.rbegin()->op.flags = flags;
  87   }
  88
  89   class C_TwoContexts;
  90   /**
  91    * Add a callback to run when this operation completes,
  92    * after any other callbacks for it.
  93    */
  94   void add_handler(Context *extra);
  95
  96   OSDOp& add_op(int op) {
  97     int s = ops.size();
  98     ops.resize(s+1);
  99     ops[s].op.op = op;
 100     out_bl.resize(s+1);
 101     out_bl[s] = NULL;
 102     out_handler.resize(s+1);
 103     out_handler[s] = NULL;
 104     out_rval.resize(s+1);
 105     out_rval[s] = NULL;
 106     return ops[s];
 107   }
 108   void add_data(int op, uint64_t off, uint64_t len, bufferlist& bl) {
 109     OSDOp& osd_op = add_op(op);
 110     osd_op.op.extent.offset = off;
 111     osd_op.op.extent.length = len;
 112     osd_op.indata.claim_append(bl);
 113   }
 114   void add_writesame(int op, uint64_t off, uint64_t write_len,
 115                      bufferlist& bl) {
 116     OSDOp& osd_op = add_op(op);
 117     osd_op.op.writesame.offset = off;
 118     osd_op.op.writesame.length = write_len;
 119     osd_op.op.writesame.data_length = bl.length();
 120     osd_op.indata.claim_append(bl);
 121   }
 122   void add_xattr(int op, const char *name, const bufferlist& data) {
 123     OSDOp& osd_op = add_op(op);
 124     osd_op.op.xattr.name_len = (name ? strlen(name) : 0);
 125     osd_op.op.xattr.value_len = data.length();
 126     if (name)
 127       osd_op.indata.append(name);
 128     osd_op.indata.append(data);
 129   }
 130   void add_xattr_cmp(int op, const char *name, uint8_t cmp_op,
 131                      uint8_t cmp_mode, const bufferlist& data) {
 132     OSDOp& osd_op = add_op(op);
 133     osd_op.op.xattr.name_len = (name ? strlen(name) : 0);
 134     osd_op.op.xattr.value_len = data.length();
 135     osd_op.op.xattr.cmp_op = cmp_op;
 136     osd_op.op.xattr.cmp_mode = cmp_mode;
 137     if (name)
 138       osd_op.indata.append(name);
 139     osd_op.indata.append(data);
 140   }
 141   void add_call(int op, const char *cname, const char *method,
 142                 bufferlist &indata,
 143                 bufferlist *outbl, Context *ctx, int *prval) {
 144     OSDOp& osd_op = add_op(op);
 145
 146     unsigned p = ops.size() - 1;
 147     out_handler[p] = ctx;
 148     out_bl[p] = outbl;
 149     out_rval[p] = prval;
 150
 151     osd_op.op.cls.class_len = strlen(cname);
 152     osd_op.op.cls.method_len = strlen(method);
 153     osd_op.op.cls.indata_len = indata.length();
 154     osd_op.indata.append(cname, osd_op.op.cls.class_len);
 155     osd_op.indata.append(method, osd_op.op.cls.method_len);
 156     osd_op.indata.append(indata);
 157   }
 158   void add_pgls(int op, uint64_t count, collection_list_handle_t cookie,
 159                 epoch_t start_epoch) {
 160     OSDOp& osd_op = add_op(op);
 161     osd_op.op.pgls.count = count;
 162     osd_op.op.pgls.start_epoch = start_epoch;
 163     ::encode(cookie, osd_op.indata);
 164   }
 165   void add_pgls_filter(int op, uint64_t count, const bufferlist& filter,
 166                        collection_list_handle_t cookie, epoch_t start_epoch) {
 167     OSDOp& osd_op = add_op(op);
 168     osd_op.op.pgls.count = count;
 169     osd_op.op.pgls.start_epoch = start_epoch;
 170     string cname = "pg";
 171     string mname = "filter";
 172     ::encode(cname, osd_op.indata);
 173     ::encode(mname, osd_op.indata);
 174     osd_op.indata.append(filter);
 175     ::encode(cookie, osd_op.indata);
 176   }
 177   void add_alloc_hint(int op, uint64_t expected_object_size,
 178                       uint64_t expected_write_size,
 179                       uint32_t flags) {
 180     OSDOp& osd_op = add_op(op);
 181     osd_op.op.alloc_hint.expected_object_size = expected_object_size;
 182     osd_op.op.alloc_hint.expected_write_size = expected_write_size;
 183     osd_op.op.alloc_hint.flags = flags;
 184   }
 185
 186   // ------
 187
 188   // pg
 189   void pg_ls(uint64_t count, bufferlist& filter,
 190              collection_list_handle_t cookie, epoch_t start_epoch) {
 191     if (filter.length() == 0)
 192       add_pgls(CEPH_OSD_OP_PGLS, count, cookie, start_epoch);
 193     else
 194       add_pgls_filter(CEPH_OSD_OP_PGLS_FILTER, count, filter, cookie,
 195                       start_epoch);
 196     flags |= CEPH_OSD_FLAG_PGOP;
 197   }
 198
 199   void pg_nls(uint64_t count, const bufferlist& filter,
 200               collection_list_handle_t cookie, epoch_t start_epoch) {
 201     if (filter.length() == 0)
 202       add_pgls(CEPH_OSD_OP_PGNLS, count, cookie, start_epoch);
 203     else
 204       add_pgls_filter(CEPH_OSD_OP_PGNLS_FILTER, count, filter, cookie,
 205                       start_epoch);
 206     flags |= CEPH_OSD_FLAG_PGOP;
 207   }
 208
 209   void scrub_ls(const librados::object_id_t& start_after,
 210                 uint64_t max_to_get,
 211                 std::vector<librados::inconsistent_obj_t> *objects,
 212                 uint32_t *interval,
 213                 int *rval);
 214   void scrub_ls(const librados::object_id_t& start_after,
 215                 uint64_t max_to_get,
 216                 std::vector<librados::inconsistent_snapset_t> *objects,
 217                 uint32_t *interval,
 218                 int *rval);
 219
 220   void create(bool excl) {
 221     OSDOp& o = add_op(CEPH_OSD_OP_CREATE);
 222     o.op.flags = (excl ? CEPH_OSD_OP_FLAG_EXCL : 0);
 223   }
 224
 225   struct C_ObjectOperation_stat : public Context {
 226     bufferlist bl;
 227     uint64_t *psize;
 228     ceph::real_time *pmtime;
 229     time_t *ptime;
 230     struct timespec *pts;
 231     int *prval;
 232     C_ObjectOperation_stat(uint64_t *ps, ceph::real_time *pm, time_t *pt, struct timespec *_pts,
 233                            int *prval)
 234       : psize(ps), pmtime(pm), ptime(pt), pts(_pts), prval(prval) {}
 235     void finish(int r) override {
 236       if (r >= 0) {
 237         bufferlist::iterator p = bl.begin();
 238         try {
 239           uint64_t size;
 240           ceph::real_time mtime;
 241           ::decode(size, p);
 242           ::decode(mtime, p);
 243           if (psize)
 244             *psize = size;
 245           if (pmtime)
 246             *pmtime = mtime;
 247           if (ptime)
 248             *ptime = ceph::real_clock::to_time_t(mtime);
 249           if (pts)
 250             *pts = ceph::real_clock::to_timespec(mtime);
 251         } catch (buffer::error& e) {
 252           if (prval)
 253             *prval = -EIO;
 254         }
 255       }
 256     }
 257   };
 258   void stat(uint64_t *psize, ceph::real_time *pmtime, int *prval) {
 259     add_op(CEPH_OSD_OP_STAT);
 260     unsigned p = ops.size() - 1;
 261     C_ObjectOperation_stat *h = new C_ObjectOperation_stat(psize, pmtime, NULL, NULL,
 262                                                            prval);
 263     out_bl[p] = &h->bl;
 264     out_handler[p] = h;
 265     out_rval[p] = prval;
 266   }
 267   void stat(uint64_t *psize, time_t *ptime, int *prval) {
 268     add_op(CEPH_OSD_OP_STAT);
 269     unsigned p = ops.size() - 1;
 270     C_ObjectOperation_stat *h = new C_ObjectOperation_stat(psize, NULL, ptime, NULL,
 271                                                            prval);
 272     out_bl[p] = &h->bl;
 273     out_handler[p] = h;
 274     out_rval[p] = prval;
 275   }
 276   void stat(uint64_t *psize, struct timespec *pts, int *prval) {
 277     add_op(CEPH_OSD_OP_STAT);
 278     unsigned p = ops.size() - 1;
 279     C_ObjectOperation_stat *h = new C_ObjectOperation_stat(psize, NULL, NULL, pts,
 280                                                            prval);
 281     out_bl[p] = &h->bl;
 282     out_handler[p] = h;
 283     out_rval[p] = prval;
 284   }
 285   // object cmpext
 286   struct C_ObjectOperation_cmpext : public Context {
 287     int *prval;
 288     C_ObjectOperation_cmpext(int *prval)
 289       : prval(prval) {}
 290
 291     void finish(int r) {
 292       if (prval)
 293         *prval = r;
 294     }
 295   };
 296
 297   void cmpext(uint64_t off, bufferlist& cmp_bl, int *prval) {
 298     add_data(CEPH_OSD_OP_CMPEXT, off, cmp_bl.length(), cmp_bl);
 299     unsigned p = ops.size() - 1;
 300     C_ObjectOperation_cmpext *h = new C_ObjectOperation_cmpext(prval);
 301     out_handler[p] = h;
 302     out_rval[p] = prval;
 303   }
 304
 305   // Used by C API
 306   void cmpext(uint64_t off, uint64_t cmp_len, const char *cmp_buf, int *prval) {
 307     bufferlist cmp_bl;
 308     cmp_bl.append(cmp_buf, cmp_len);
 309     add_data(CEPH_OSD_OP_CMPEXT, off, cmp_len, cmp_bl);
 310     unsigned p = ops.size() - 1;
 311     C_ObjectOperation_cmpext *h = new C_ObjectOperation_cmpext(prval);
 312     out_handler[p] = h;
 313     out_rval[p] = prval;
 314   }
 315
 316   void read(uint64_t off, uint64_t len, bufferlist *pbl, int *prval,
 317             Context* ctx) {
 318     bufferlist bl;
 319     add_data(CEPH_OSD_OP_READ, off, len, bl);
 320     unsigned p = ops.size() - 1;
 321     out_bl[p] = pbl;
 322     out_rval[p] = prval;
 323     out_handler[p] = ctx;
 324   }
 325
 326   struct C_ObjectOperation_sparse_read : public Context {
 327     bufferlist bl;
 328     bufferlist *data_bl;
 329     std::map<uint64_t, uint64_t> *extents;
 330     int *prval;
 331     C_ObjectOperation_sparse_read(bufferlist *data_bl,
 332                                   std::map<uint64_t, uint64_t> *extents,
 333                                   int *prval)
 334       : data_bl(data_bl), extents(extents), prval(prval) {}
 335     void finish(int r) override {
 336       bufferlist::iterator iter = bl.begin();
 337       if (r >= 0) {
 338         try {
 339           ::decode(*extents, iter);
 340           ::decode(*data_bl, iter);
 341         } catch (buffer::error& e) {
 342           if (prval)
 343             *prval = -EIO;
 344         }
 345       }
 346     }
 347   };
 348   void sparse_read(uint64_t off, uint64_t len, std::map<uint64_t,uint64_t> *m,
 349                    bufferlist *data_bl, int *prval) {
 350     bufferlist bl;
 351     add_data(CEPH_OSD_OP_SPARSE_READ, off, len, bl);
 352     unsigned p = ops.size() - 1;
 353     C_ObjectOperation_sparse_read *h =
 354       new C_ObjectOperation_sparse_read(data_bl, m, prval);
 355     out_bl[p] = &h->bl;
 356     out_handler[p] = h;
 357     out_rval[p] = prval;
 358   }
 359   void write(uint64_t off, bufferlist& bl,
 360              uint64_t truncate_size,
 361              uint32_t truncate_seq) {
 362     add_data(CEPH_OSD_OP_WRITE, off, bl.length(), bl);
 363     OSDOp& o = *ops.rbegin();
 364     o.op.extent.truncate_size = truncate_size;
 365     o.op.extent.truncate_seq = truncate_seq;
 366   }
 367   void write(uint64_t off, bufferlist& bl) {
 368     write(off, bl, 0, 0);
 369   }
 370   void write_full(bufferlist& bl) {
 371     add_data(CEPH_OSD_OP_WRITEFULL, 0, bl.length(), bl);
 372   }
 373   void writesame(uint64_t off, uint64_t write_len, bufferlist& bl) {
 374     add_writesame(CEPH_OSD_OP_WRITESAME, off, write_len, bl);
 375   }
 376   void append(bufferlist& bl) {
 377     add_data(CEPH_OSD_OP_APPEND, 0, bl.length(), bl);
 378   }
 379   void zero(uint64_t off, uint64_t len) {
 380     bufferlist bl;
 381     add_data(CEPH_OSD_OP_ZERO, off, len, bl);
 382   }
 383   void truncate(uint64_t off) {
 384     bufferlist bl;
 385     add_data(CEPH_OSD_OP_TRUNCATE, off, 0, bl);
 386   }
 387   void remove() {
 388     bufferlist bl;
 389     add_data(CEPH_OSD_OP_DELETE, 0, 0, bl);
 390   }
 391   void mapext(uint64_t off, uint64_t len) {
 392     bufferlist bl;
 393     add_data(CEPH_OSD_OP_MAPEXT, off, len, bl);
 394   }
 395   void sparse_read(uint64_t off, uint64_t len) {
 396     bufferlist bl;
 397     add_data(CEPH_OSD_OP_SPARSE_READ, off, len, bl);
 398   }
 399
 400   void checksum(uint8_t type, const bufferlist &init_value_bl,
 401                 uint64_t off, uint64_t len, size_t chunk_size,
 402                 bufferlist *pbl, int *prval, Context *ctx) {
 403     OSDOp& osd_op = add_op(CEPH_OSD_OP_CHECKSUM);
 404     osd_op.op.checksum.offset = off;
 405     osd_op.op.checksum.length = len;
 406     osd_op.op.checksum.type = type;
 407     osd_op.op.checksum.chunk_size = chunk_size;
 408     osd_op.indata.append(init_value_bl);
 409
 410     unsigned p = ops.size() - 1;
 411     out_bl[p] = pbl;
 412     out_rval[p] = prval;
 413     out_handler[p] = ctx;
 414   }
 415
 416   // object attrs
 417   void getxattr(const char *name, bufferlist *pbl, int *prval) {
 418     bufferlist bl;
 419     add_xattr(CEPH_OSD_OP_GETXATTR, name, bl);
 420     unsigned p = ops.size() - 1;
 421     out_bl[p] = pbl;
 422     out_rval[p] = prval;
 423   }
 424   struct C_ObjectOperation_decodevals : public Context {
 425     uint64_t max_entries;
 426     bufferlist bl;
 427     std::map<std::string,bufferlist> *pattrs;
 428     bool *ptruncated;
 429     int *prval;
 430     C_ObjectOperation_decodevals(uint64_t m, std::map<std::string,bufferlist> *pa,
 431                                  bool *pt, int *pr)
 432       : max_entries(m), pattrs(pa), ptruncated(pt), prval(pr) {
 433       if (ptruncated) {
 434         *ptruncated = false;
 435       }
 436     }
 437     void finish(int r) override {
 438       if (r >= 0) {
 439         bufferlist::iterator p = bl.begin();
 440         try {
 441           if (pattrs)
 442             ::decode(*pattrs, p);
 443           if (ptruncated) {
 444             std::map<std::string,bufferlist> ignore;
 445             if (!pattrs) {
 446               ::decode(ignore, p);
 447               pattrs = &ignore;
 448             }
 449             if (!p.end()) {
 450               ::decode(*ptruncated, p);
 451             } else {
 452               // the OSD did not provide this.  since old OSDs do not
 453               // enfoce omap result limits either, we can infer it from
 454               // the size of the result
 455               *ptruncated = (pattrs->size() == max_entries);
 456             }
 457           }
 458         }
 459         catch (buffer::error& e) {
 460           if (prval)
 461             *prval = -EIO;
 462         }
 463       }
 464     }
 465   };
 466   struct C_ObjectOperation_decodekeys : public Context {
 467     uint64_t max_entries;
 468     bufferlist bl;
 469     std::set<std::string> *pattrs;
 470     bool *ptruncated;
 471     int *prval;
 472     C_ObjectOperation_decodekeys(uint64_t m, std::set<std::string> *pa, bool *pt,
 473                                  int *pr)
 474       : max_entries(m), pattrs(pa), ptruncated(pt), prval(pr) {
 475       if (ptruncated) {
 476         *ptruncated = false;
 477       }
 478     }
 479     void finish(int r) override {
 480       if (r >= 0) {
 481         bufferlist::iterator p = bl.begin();
 482         try {
 483           if (pattrs)
 484             ::decode(*pattrs, p);
 485           if (ptruncated) {
 486             std::set<std::string> ignore;
 487             if (!pattrs) {
 488               ::decode(ignore, p);
 489               pattrs = &ignore;
 490             }
 491             if (!p.end()) {
 492               ::decode(*ptruncated, p);
 493             } else {
 494               // the OSD did not provide this.  since old OSDs do not
 495               // enfoce omap result limits either, we can infer it from
 496               // the size of the result
 497               *ptruncated = (pattrs->size() == max_entries);
 498             }
 499           }
 500         }
 501         catch (buffer::error& e) {
 502           if (prval)
 503             *prval = -EIO;
 504         }
 505       }
 506     }
 507   };
 508   struct C_ObjectOperation_decodewatchers : public Context {
 509     bufferlist bl;
 510     list<obj_watch_t> *pwatchers;
 511     int *prval;
 512     C_ObjectOperation_decodewatchers(list<obj_watch_t> *pw, int *pr)
 513       : pwatchers(pw), prval(pr) {}
 514     void finish(int r) override {
 515       if (r >= 0) {
 516         bufferlist::iterator p = bl.begin();
 517         try {
 518           obj_list_watch_response_t resp;
 519           ::decode(resp, p);
 520           if (pwatchers) {
 521             for (list<watch_item_t>::iterator i = resp.entries.begin() ;
 522                  i != resp.entries.end() ; ++i) {
 523               obj_watch_t ow;
 524               ostringstream sa;
 525               sa << i->addr;
 526               strncpy(ow.addr, sa.str().c_str(), 256);
 527               ow.watcher_id = i->name.num();
 528               ow.cookie = i->cookie;
 529               ow.timeout_seconds = i->timeout_seconds;
 530               pwatchers->push_back(ow);
 531             }
 532           }
 533         }
 534         catch (buffer::error& e) {
 535           if (prval)
 536             *prval = -EIO;
 537         }
 538       }
 539     }
 540   };
 541   struct C_ObjectOperation_decodesnaps : public Context {
 542     bufferlist bl;
 543     librados::snap_set_t *psnaps;
 544     int *prval;
 545     C_ObjectOperation_decodesnaps(librados::snap_set_t *ps, int *pr)
 546       : psnaps(ps), prval(pr) {}
 547     void finish(int r) override {
 548       if (r >= 0) {
 549         bufferlist::iterator p = bl.begin();
 550         try {
 551           obj_list_snap_response_t resp;
 552           ::decode(resp, p);
 553           if (psnaps) {
 554             psnaps->clones.clear();
 555             for (vector<clone_info>::iterator ci = resp.clones.begin();
 556                  ci != resp.clones.end();
 557                  ++ci) {
 558               librados::clone_info_t clone;
 559
 560               clone.cloneid = ci->cloneid;
 561               clone.snaps.reserve(ci->snaps.size());
 562               clone.snaps.insert(clone.snaps.end(), ci->snaps.begin(),
 563                                  ci->snaps.end());
 564               clone.overlap = ci->overlap;
 565               clone.size = ci->size;
 566
 567               psnaps->clones.push_back(clone);
 568             }
 569             psnaps->seq = resp.seq;
 570           }
 571         } catch (buffer::error& e) {
 572           if (prval)
 573             *prval = -EIO;
 574         }
 575       }
 576     }
 577   };
 578   void getxattrs(std::map<std::string,bufferlist> *pattrs, int *prval) {
 579     add_op(CEPH_OSD_OP_GETXATTRS);
 580     if (pattrs || prval) {
 581       unsigned p = ops.size() - 1;
 582       C_ObjectOperation_decodevals *h
 583         = new C_ObjectOperation_decodevals(0, pattrs, nullptr, prval);
 584       out_handler[p] = h;
 585       out_bl[p] = &h->bl;
 586       out_rval[p] = prval;
 587     }
 588   }
 589   void setxattr(const char *name, const bufferlist& bl) {
 590     add_xattr(CEPH_OSD_OP_SETXATTR, name, bl);
 591   }
 592   void setxattr(const char *name, const string& s) {
 593     bufferlist bl;
 594     bl.append(s);
 595     add_xattr(CEPH_OSD_OP_SETXATTR, name, bl);
 596   }
 597   void cmpxattr(const char *name, uint8_t cmp_op, uint8_t cmp_mode,
 598                 const bufferlist& bl) {
 599     add_xattr_cmp(CEPH_OSD_OP_CMPXATTR, name, cmp_op, cmp_mode, bl);
 600   }
 601   void rmxattr(const char *name) {
 602     bufferlist bl;
 603     add_xattr(CEPH_OSD_OP_RMXATTR, name, bl);
 604   }
 605   void setxattrs(map<string, bufferlist>& attrs) {
 606     bufferlist bl;
 607     ::encode(attrs, bl);
 608     add_xattr(CEPH_OSD_OP_RESETXATTRS, 0, bl.length());
 609   }
 610   void resetxattrs(const char *prefix, map<string, bufferlist>& attrs) {
 611     bufferlist bl;
 612     ::encode(attrs, bl);
 613     add_xattr(CEPH_OSD_OP_RESETXATTRS, prefix, bl);
 614   }
 615
 616   // trivialmap
 617   void tmap_update(bufferlist& bl) {
 618     add_data(CEPH_OSD_OP_TMAPUP, 0, 0, bl);
 619   }
 620   void tmap_put(bufferlist& bl) {
 621     add_data(CEPH_OSD_OP_TMAPPUT, 0, bl.length(), bl);
 622   }
 623   void tmap_get(bufferlist *pbl, int *prval) {
 624     add_op(CEPH_OSD_OP_TMAPGET);
 625     unsigned p = ops.size() - 1;
 626     out_bl[p] = pbl;
 627     out_rval[p] = prval;
 628   }
 629   void tmap_get() {
 630     add_op(CEPH_OSD_OP_TMAPGET);
 631   }
 632   void tmap_to_omap(bool nullok=false) {
 633      OSDOp& osd_op = add_op(CEPH_OSD_OP_TMAP2OMAP);
 634      if (nullok)
 635        osd_op.op.tmap2omap.flags = CEPH_OSD_TMAP2OMAP_NULLOK;
 636   }
 637
 638   // objectmap
 639   void omap_get_keys(const string &start_after,
 640                      uint64_t max_to_get,
 641                      std::set<std::string> *out_set,
 642                      bool *ptruncated,
 643                      int *prval) {
 644     OSDOp &op = add_op(CEPH_OSD_OP_OMAPGETKEYS);
 645     bufferlist bl;
 646     ::encode(start_after, bl);
 647     ::encode(max_to_get, bl);
 648     op.op.extent.offset = 0;
 649     op.op.extent.length = bl.length();
 650     op.indata.claim_append(bl);
 651     if (prval || ptruncated || out_set) {
 652       unsigned p = ops.size() - 1;
 653       C_ObjectOperation_decodekeys *h =
 654         new C_ObjectOperation_decodekeys(max_to_get, out_set, ptruncated, prval);
 655       out_handler[p] = h;
 656       out_bl[p] = &h->bl;
 657       out_rval[p] = prval;
 658     }
 659   }
 660
 661   void omap_get_vals(const string &start_after,
 662                      const string &filter_prefix,
 663                      uint64_t max_to_get,
 664                      std::map<std::string, bufferlist> *out_set,
 665                      bool *ptruncated,
 666                      int *prval) {
 667     OSDOp &op = add_op(CEPH_OSD_OP_OMAPGETVALS);
 668     bufferlist bl;
 669     ::encode(start_after, bl);
 670     ::encode(max_to_get, bl);
 671     ::encode(filter_prefix, bl);
 672     op.op.extent.offset = 0;
 673     op.op.extent.length = bl.length();
 674     op.indata.claim_append(bl);
 675     if (prval || out_set || ptruncated) {
 676       unsigned p = ops.size() - 1;
 677       C_ObjectOperation_decodevals *h =
 678         new C_ObjectOperation_decodevals(max_to_get, out_set, ptruncated, prval);
 679       out_handler[p] = h;
 680       out_bl[p] = &h->bl;
 681       out_rval[p] = prval;
 682     }
 683   }
 684
 685   void omap_get_vals_by_keys(const std::set<std::string> &to_get,
 686                             std::map<std::string, bufferlist> *out_set,
 687                             int *prval) {
 688     OSDOp &op = add_op(CEPH_OSD_OP_OMAPGETVALSBYKEYS);
 689     bufferlist bl;
 690     ::encode(to_get, bl);
 691     op.op.extent.offset = 0;
 692     op.op.extent.length = bl.length();
 693     op.indata.claim_append(bl);
 694     if (prval || out_set) {
 695       unsigned p = ops.size() - 1;
 696       C_ObjectOperation_decodevals *h =
 697         new C_ObjectOperation_decodevals(0, out_set, nullptr, prval);
 698       out_handler[p] = h;
 699       out_bl[p] = &h->bl;
 700       out_rval[p] = prval;
 701     }
 702   }
 703
 704   void omap_cmp(const std::map<std::string, pair<bufferlist,int> > &assertions,
 705                 int *prval) {
 706     OSDOp &op = add_op(CEPH_OSD_OP_OMAP_CMP);
 707     bufferlist bl;
 708     ::encode(assertions, bl);
 709     op.op.extent.offset = 0;
 710     op.op.extent.length = bl.length();
 711     op.indata.claim_append(bl);
 712     if (prval) {
 713       unsigned p = ops.size() - 1;
 714       out_rval[p] = prval;
 715     }
 716   }
 717
 718   struct C_ObjectOperation_copyget : public Context {
 719     bufferlist bl;
 720     object_copy_cursor_t *cursor;
 721     uint64_t *out_size;
 722     ceph::real_time *out_mtime;
 723     std::map<std::string,bufferlist> *out_attrs;
 724     bufferlist *out_data, *out_omap_header, *out_omap_data;
 725     vector<snapid_t> *out_snaps;
 726     snapid_t *out_snap_seq;
 727     uint32_t *out_flags;
 728     uint32_t *out_data_digest;
 729     uint32_t *out_omap_digest;
 730     mempool::osd_pglog::vector<pair<osd_reqid_t, version_t> > *out_reqids;
 731     uint64_t *out_truncate_seq;
 732     uint64_t *out_truncate_size;
 733     int *prval;
 734     C_ObjectOperation_copyget(object_copy_cursor_t *c,
 735                               uint64_t *s,
 736                               ceph::real_time *m,
 737                               std::map<std::string,bufferlist> *a,
 738                               bufferlist *d, bufferlist *oh,
 739                               bufferlist *o,
 740                               std::vector<snapid_t> *osnaps,
 741                               snapid_t *osnap_seq,
 742                               uint32_t *flags,
 743                               uint32_t *dd,
 744                               uint32_t *od,
 745                               mempool::osd_pglog::vector<pair<osd_reqid_t, version_t> > *oreqids,
 746                               uint64_t *otseq,
 747                               uint64_t *otsize,
 748                               int *r)
 749       : cursor(c),
 750         out_size(s), out_mtime(m),
 751         out_attrs(a), out_data(d), out_omap_header(oh),
 752         out_omap_data(o), out_snaps(osnaps), out_snap_seq(osnap_seq),
 753         out_flags(flags), out_data_digest(dd), out_omap_digest(od),
 754         out_reqids(oreqids),
 755         out_truncate_seq(otseq),
 756         out_truncate_size(otsize),
 757         prval(r) {}
 758     void finish(int r) override {
 759       // reqids are copied on ENOENT
 760       if (r < 0 && r != -ENOENT)
 761         return;
 762       try {
 763         bufferlist::iterator p = bl.begin();
 764         object_copy_data_t copy_reply;
 765         ::decode(copy_reply, p);
 766         if (r == -ENOENT) {
 767           if (out_reqids)
 768             *out_reqids = copy_reply.reqids;
 769           return;
 770         }
 771         if (out_size)
 772           *out_size = copy_reply.size;
 773         if (out_mtime)
 774           *out_mtime = ceph::real_clock::from_ceph_timespec(copy_reply.mtime);
 775         if (out_attrs)
 776           *out_attrs = copy_reply.attrs;
 777         if (out_data)
 778           out_data->claim_append(copy_reply.data);
 779         if (out_omap_header)
 780           out_omap_header->claim_append(copy_reply.omap_header);
 781         if (out_omap_data)
 782           *out_omap_data = copy_reply.omap_data;
 783         if (out_snaps)
 784           *out_snaps = copy_reply.snaps;
 785         if (out_snap_seq)
 786           *out_snap_seq = copy_reply.snap_seq;
 787         if (out_flags)
 788           *out_flags = copy_reply.flags;
 789         if (out_data_digest)
 790           *out_data_digest = copy_reply.data_digest;
 791         if (out_omap_digest)
 792           *out_omap_digest = copy_reply.omap_digest;
 793         if (out_reqids)
 794           *out_reqids = copy_reply.reqids;
 795         if (out_truncate_seq)
 796           *out_truncate_seq = copy_reply.truncate_seq;
 797         if (out_truncate_size)
 798           *out_truncate_size = copy_reply.truncate_size;
 799         *cursor = copy_reply.cursor;
 800       } catch (buffer::error& e) {
 801         if (prval)
 802           *prval = -EIO;
 803       }
 804     }
 805   };
 806
 807   void copy_get(object_copy_cursor_t *cursor,
 808                 uint64_t max,
 809                 uint64_t *out_size,
 810                 ceph::real_time *out_mtime,
 811                 std::map<std::string,bufferlist> *out_attrs,
 812                 bufferlist *out_data,
 813                 bufferlist *out_omap_header,
 814                 bufferlist *out_omap_data,
 815                 vector<snapid_t> *out_snaps,
 816                 snapid_t *out_snap_seq,
 817                 uint32_t *out_flags,
 818                 uint32_t *out_data_digest,
 819                 uint32_t *out_omap_digest,
 820                 mempool::osd_pglog::vector<pair<osd_reqid_t, version_t> > *out_reqids,
 821                 uint64_t *truncate_seq,
 822                 uint64_t *truncate_size,
 823                 int *prval) {
 824     OSDOp& osd_op = add_op(CEPH_OSD_OP_COPY_GET);
 825     osd_op.op.copy_get.max = max;
 826     ::encode(*cursor, osd_op.indata);
 827     ::encode(max, osd_op.indata);
 828     unsigned p = ops.size() - 1;
 829     out_rval[p] = prval;
 830     C_ObjectOperation_copyget *h =
 831       new C_ObjectOperation_copyget(cursor, out_size, out_mtime,
 832                                     out_attrs, out_data, out_omap_header,
 833                                     out_omap_data, out_snaps, out_snap_seq,
 834                                     out_flags, out_data_digest,
 835                                     out_omap_digest, out_reqids, truncate_seq,
 836                                     truncate_size, prval);
 837     out_bl[p] = &h->bl;
 838     out_handler[p] = h;
 839   }
 840
 841   void undirty() {
 842     add_op(CEPH_OSD_OP_UNDIRTY);
 843   }
 844
 845   struct C_ObjectOperation_isdirty : public Context {
 846     bufferlist bl;
 847     bool *pisdirty;
 848     int *prval;
 849     C_ObjectOperation_isdirty(bool *p, int *r)
 850       : pisdirty(p), prval(r) {}
 851     void finish(int r) override {
 852       if (r < 0)
 853         return;
 854       try {
 855         bufferlist::iterator p = bl.begin();
 856         bool isdirty;
 857         ::decode(isdirty, p);
 858         if (pisdirty)
 859           *pisdirty = isdirty;
 860       } catch (buffer::error& e) {
 861         if (prval)
 862           *prval = -EIO;
 863       }
 864     }
 865   };
 866
 867   void is_dirty(bool *pisdirty, int *prval) {
 868     add_op(CEPH_OSD_OP_ISDIRTY);
 869     unsigned p = ops.size() - 1;
 870     out_rval[p] = prval;
 871     C_ObjectOperation_isdirty *h =
 872       new C_ObjectOperation_isdirty(pisdirty, prval);
 873     out_bl[p] = &h->bl;
 874     out_handler[p] = h;
 875   }
 876
 877   struct C_ObjectOperation_hit_set_ls : public Context {
 878     bufferlist bl;
 879     std::list< std::pair<time_t, time_t> > *ptls;
 880     std::list< std::pair<ceph::real_time, ceph::real_time> > *putls;
 881     int *prval;
 882     C_ObjectOperation_hit_set_ls(std::list< std::pair<time_t, time_t> > *t,
 883                                  std::list< std::pair<ceph::real_time,
 884                                                       ceph::real_time> > *ut,
 885                                  int *r)
 886       : ptls(t), putls(ut), prval(r) {}
 887     void finish(int r) override {
 888       if (r < 0)
 889         return;
 890       try {
 891         bufferlist::iterator p = bl.begin();
 892         std::list< std::pair<ceph::real_time, ceph::real_time> > ls;
 893         ::decode(ls, p);
 894         if (ptls) {
 895           ptls->clear();
 896           for (auto p = ls.begin(); p != ls.end(); ++p)
 897             // round initial timestamp up to the next full second to
 898             // keep this a valid interval.
 899             ptls->push_back(
 900               make_pair(ceph::real_clock::to_time_t(
 901                           ceph::ceil(p->first,
 902                                      // Sadly, no time literals until C++14.
 903                                      std::chrono::seconds(1))),
 904                         ceph::real_clock::to_time_t(p->second)));
 905         }
 906         if (putls)
 907           putls->swap(ls);
 908       } catch (buffer::error& e) {
 909         r = -EIO;
 910       }
 911       if (prval)
 912         *prval = r;
 913     }
 914   };
 915
 916   /**
 917    * list available HitSets.
 918    *
 919    * We will get back a list of time intervals.  Note that the most
 920    * recent range may have an empty end timestamp if it is still
 921    * accumulating.
 922    *
 923    * @param pls [out] list of time intervals
 924    * @param prval [out] return value
 925    */
 926   void hit_set_ls(std::list< std::pair<time_t, time_t> > *pls, int *prval) {
 927     add_op(CEPH_OSD_OP_PG_HITSET_LS);
 928     unsigned p = ops.size() - 1;
 929     out_rval[p] = prval;
 930     C_ObjectOperation_hit_set_ls *h =
 931       new C_ObjectOperation_hit_set_ls(pls, NULL, prval);
 932     out_bl[p] = &h->bl;
 933     out_handler[p] = h;
 934   }
 935   void hit_set_ls(std::list<std::pair<ceph::real_time, ceph::real_time> > *pls,
 936                   int *prval) {
 937     add_op(CEPH_OSD_OP_PG_HITSET_LS);
 938     unsigned p = ops.size() - 1;
 939     out_rval[p] = prval;
 940     C_ObjectOperation_hit_set_ls *h =
 941       new C_ObjectOperation_hit_set_ls(NULL, pls, prval);
 942     out_bl[p] = &h->bl;
 943     out_handler[p] = h;
 944   }
 945
 946   /**
 947    * get HitSet
 948    *
 949    * Return an encoded HitSet that includes the provided time
 950    * interval.
 951    *
 952    * @param stamp [in] timestamp
 953    * @param pbl [out] target buffer for encoded HitSet
 954    * @param prval [out] return value
 955    */
 956   void hit_set_get(ceph::real_time stamp, bufferlist *pbl, int *prval) {
 957     OSDOp& op = add_op(CEPH_OSD_OP_PG_HITSET_GET);
 958     op.op.hit_set_get.stamp = ceph::real_clock::to_ceph_timespec(stamp);
 959     unsigned p = ops.size() - 1;
 960     out_rval[p] = prval;
 961     out_bl[p] = pbl;
 962   }
 963
 964   void omap_get_header(bufferlist *bl, int *prval) {
 965     add_op(CEPH_OSD_OP_OMAPGETHEADER);
 966     unsigned p = ops.size() - 1;
 967     out_bl[p] = bl;
 968     out_rval[p] = prval;
 969   }
 970
 971   void omap_set(const map<string, bufferlist> &map) {
 972     bufferlist bl;
 973     ::encode(map, bl);
 974     add_data(CEPH_OSD_OP_OMAPSETVALS, 0, bl.length(), bl);
 975   }
 976
 977   void omap_set_header(bufferlist &bl) {
 978     add_data(CEPH_OSD_OP_OMAPSETHEADER, 0, bl.length(), bl);
 979   }
 980
 981   void omap_clear() {
 982     add_op(CEPH_OSD_OP_OMAPCLEAR);
 983   }
 984
 985   void omap_rm_keys(const std::set<std::string> &to_remove) {
 986     bufferlist bl;
 987     ::encode(to_remove, bl);
 988     add_data(CEPH_OSD_OP_OMAPRMKEYS, 0, bl.length(), bl);
 989   }
 990
 991   // object classes
 992   void call(const char *cname, const char *method, bufferlist &indata) {
 993     add_call(CEPH_OSD_OP_CALL, cname, method, indata, NULL, NULL, NULL);
 994   }
 995
 996   void call(const char *cname, const char *method, bufferlist &indata,
 997             bufferlist *outdata, Context *ctx, int *prval) {
 998     add_call(CEPH_OSD_OP_CALL, cname, method, indata, outdata, ctx, prval);
 999   }
1000
1001   // watch/notify
1002   void watch(uint64_t cookie, __u8 op, uint32_t timeout = 0) {
1003     OSDOp& osd_op = add_op(CEPH_OSD_OP_WATCH);
1004     osd_op.op.watch.cookie = cookie;
1005     osd_op.op.watch.op = op;
1006     osd_op.op.watch.timeout = timeout;
1007   }
1008
1009   void notify(uint64_t cookie, uint32_t prot_ver, uint32_t timeout,
1010               bufferlist &bl, bufferlist *inbl) {
1011     OSDOp& osd_op = add_op(CEPH_OSD_OP_NOTIFY);
1012     osd_op.op.notify.cookie = cookie;
1013     ::encode(prot_ver, *inbl);
1014     ::encode(timeout, *inbl);
1015     ::encode(bl, *inbl);
1016     osd_op.indata.append(*inbl);
1017   }
1018
1019   void notify_ack(uint64_t notify_id, uint64_t cookie,
1020                   bufferlist& reply_bl) {
1021     OSDOp& osd_op = add_op(CEPH_OSD_OP_NOTIFY_ACK);
1022     bufferlist bl;
1023     ::encode(notify_id, bl);
1024     ::encode(cookie, bl);
1025     ::encode(reply_bl, bl);
1026     osd_op.indata.append(bl);
1027   }
1028
1029   void list_watchers(list<obj_watch_t> *out,
1030                      int *prval) {
1031     (void)add_op(CEPH_OSD_OP_LIST_WATCHERS);
1032     if (prval || out) {
1033       unsigned p = ops.size() - 1;
1034       C_ObjectOperation_decodewatchers *h =
1035         new C_ObjectOperation_decodewatchers(out, prval);
1036       out_handler[p] = h;
1037       out_bl[p] = &h->bl;
1038       out_rval[p] = prval;
1039     }
1040   }
1041
1042   void list_snaps(librados::snap_set_t *out, int *prval) {
1043     (void)add_op(CEPH_OSD_OP_LIST_SNAPS);
1044     if (prval || out) {
1045       unsigned p = ops.size() - 1;
1046       C_ObjectOperation_decodesnaps *h =
1047         new C_ObjectOperation_decodesnaps(out, prval);
1048       out_handler[p] = h;
1049       out_bl[p] = &h->bl;
1050       out_rval[p] = prval;
1051     }
1052   }
1053
1054   void assert_version(uint64_t ver) {
1055     OSDOp& osd_op = add_op(CEPH_OSD_OP_ASSERT_VER);
1056     osd_op.op.assert_ver.ver = ver;
1057   }
1058
1059   void cmpxattr(const char *name, const bufferlist& val,
1060                 int op, int mode) {
1061     add_xattr(CEPH_OSD_OP_CMPXATTR, name, val);
1062     OSDOp& o = *ops.rbegin();
1063     o.op.xattr.cmp_op = op;
1064     o.op.xattr.cmp_mode = mode;
1065   }
1066
1067   void rollback(uint64_t snapid) {
1068     OSDOp& osd_op = add_op(CEPH_OSD_OP_ROLLBACK);
1069     osd_op.op.snap.snapid = snapid;
1070   }
1071
1072   void copy_from(object_t src, snapid_t snapid, object_locator_t src_oloc,
1073                  version_t src_version, unsigned flags,
1074                  unsigned src_fadvise_flags) {
1075     OSDOp& osd_op = add_op(CEPH_OSD_OP_COPY_FROM);
1076     osd_op.op.copy_from.snapid = snapid;
1077     osd_op.op.copy_from.src_version = src_version;
1078     osd_op.op.copy_from.flags = flags;
1079     osd_op.op.copy_from.src_fadvise_flags = src_fadvise_flags;
1080     ::encode(src, osd_op.indata);
1081     ::encode(src_oloc, osd_op.indata);
1082   }
1083
1084   /**
1085    * writeback content to backing tier
1086    *
1087    * If object is marked dirty in the cache tier, write back content
1088    * to backing tier. If the object is clean this is a no-op.
1089    *
1090    * If writeback races with an update, the update will block.
1091    *
1092    * use with IGNORE_CACHE to avoid triggering promote.
1093    */
1094   void cache_flush() {
1095     add_op(CEPH_OSD_OP_CACHE_FLUSH);
1096   }
1097
1098   /**
1099    * writeback content to backing tier
1100    *
1101    * If object is marked dirty in the cache tier, write back content
1102    * to backing tier. If the object is clean this is a no-op.
1103    *
1104    * If writeback races with an update, return EAGAIN.  Requires that
1105    * the SKIPRWLOCKS flag be set.
1106    *
1107    * use with IGNORE_CACHE to avoid triggering promote.
1108    */
1109   void cache_try_flush() {
1110     add_op(CEPH_OSD_OP_CACHE_TRY_FLUSH);
1111   }
1112
1113   /**
1114    * evict object from cache tier
1115    *
1116    * If object is marked clean, remove the object from the cache tier.
1117    * Otherwise, return EBUSY.
1118    *
1119    * use with IGNORE_CACHE to avoid triggering promote.
1120    */
1121   void cache_evict() {
1122     add_op(CEPH_OSD_OP_CACHE_EVICT);
1123   }
1124
1125   /*
1126    * Extensible tier
1127    */
1128   void set_redirect(object_t tgt, snapid_t snapid, object_locator_t tgt_oloc,
1129                     version_t tgt_version) {
1130     OSDOp& osd_op = add_op(CEPH_OSD_OP_SET_REDIRECT);
1131     osd_op.op.copy_from.snapid = snapid;
1132     osd_op.op.copy_from.src_version = tgt_version;
1133     ::encode(tgt, osd_op.indata);
1134     ::encode(tgt_oloc, osd_op.indata);
1135   }
1136
1137   void set_alloc_hint(uint64_t expected_object_size,
1138                       uint64_t expected_write_size,
1139                       uint32_t flags) {
1140     add_alloc_hint(CEPH_OSD_OP_SETALLOCHINT, expected_object_size,
1141                    expected_write_size, flags);
1142
1143     // CEPH_OSD_OP_SETALLOCHINT op is advisory and therefore deemed
1144     // not worth a feature bit.  Set FAILOK per-op flag to make
1145     // sure older osds don't trip over an unsupported opcode.
1146     set_last_op_flags(CEPH_OSD_OP_FLAG_FAILOK);
1147   }
1148
1149   void dup(vector<OSDOp>& sops) {
1150     ops = sops;
1151     out_bl.resize(sops.size());
1152     out_handler.resize(sops.size());
1153     out_rval.resize(sops.size());
1154     for (uint32_t i = 0; i < sops.size(); i++) {
1155       out_bl[i] = &sops[i].outdata;
1156       out_handler[i] = NULL;
1157       out_rval[i] = &sops[i].rval;
1158     }
1159   }
1160
1161   /**
1162    * Pin/unpin an object in cache tier
1163    */
1164   void cache_pin() {
1165     add_op(CEPH_OSD_OP_CACHE_PIN);
1166   }
1167
1168   void cache_unpin() {
1169     add_op(CEPH_OSD_OP_CACHE_UNPIN);
1170   }
1171 };
1172
1173
1174 // ----------------
1175
1176
1177 class Objecter : public md_config_obs_t, public Dispatcher {
1178 public:
1179   // config observer bits
1180   const char** get_tracked_conf_keys() const override;
1181   void handle_conf_change(const struct md_config_t *conf,
1182                           const std::set <std::string> &changed) override;
1183
1184 public:
1185   Messenger *messenger;
1186   MonClient *monc;
1187   Finisher *finisher;
1188   ZTracer::Endpoint trace_endpoint;
1189 private:
1190   OSDMap    *osdmap;
1191 public:
1192   using Dispatcher::cct;
1193   std::multimap<string,string> crush_location;
1194
1195   std::atomic<bool> initialized{false};
1196
1197 private:
1198   std::atomic<uint64_t> last_tid{0};
1199   std::atomic<unsigned> inflight_ops{0};
1200   std::atomic<int> client_inc{-1};
1201   uint64_t max_linger_id;
1202   std::atomic<unsigned> num_in_flight{0};
1203   std::atomic<int> global_op_flags{0}; // flags which are applied to each IO op
1204   bool keep_balanced_budget;
1205   bool honor_osdmap_full;
1206   bool osdmap_full_try;
1207
1208   // If this is true, accumulate a set of blacklisted entities
1209   // to be drained by consume_blacklist_events.
1210   bool blacklist_events_enabled;
1211   std::set<entity_addr_t> blacklist_events;
1212
1213 public:
1214   void maybe_request_map();
1215
1216   void enable_blacklist_events();
1217 private:
1218
1219   void _maybe_request_map();
1220
1221   version_t last_seen_osdmap_version;
1222   version_t last_seen_pgmap_version;
1223
1224   mutable boost::shared_mutex rwlock;
1225   using lock_guard = std::unique_lock<decltype(rwlock)>;
1226   using unique_lock = std::unique_lock<decltype(rwlock)>;
1227   using shared_lock = boost::shared_lock<decltype(rwlock)>;
1228   using shunique_lock = ceph::shunique_lock<decltype(rwlock)>;
1229   ceph::timer<ceph::mono_clock> timer;
1230
1231   PerfCounters *logger;
1232
1233   uint64_t tick_event;
1234
1235   void start_tick();
1236   void tick();
1237   void update_crush_location();
1238
1239   class RequestStateHook;
1240
1241   RequestStateHook *m_request_state_hook;
1242
1243 public:
1244   /*** track pending operations ***/
1245   // read
1246  public:
1247
1248   struct OSDSession;
1249
1250   struct op_target_t {
1251     int flags = 0;
1252
1253     epoch_t epoch = 0;  ///< latest epoch we calculated the mapping
1254
1255     object_t base_oid;
1256     object_locator_t base_oloc;
1257     object_t target_oid;
1258     object_locator_t target_oloc;
1259
1260     ///< true if we are directed at base_pgid, not base_oid
1261     bool precalc_pgid = false;
1262
1263     ///< true if we have ever mapped to a valid pool
1264     bool pool_ever_existed = false;
1265
1266     ///< explcit pg target, if any
1267     pg_t base_pgid;
1268
1269     pg_t pgid; ///< last (raw) pg we mapped to
1270     spg_t actual_pgid; ///< last (actual) spg_t we mapped to
1271     unsigned pg_num = 0; ///< last pg_num we mapped to
1272     unsigned pg_num_mask = 0; ///< last pg_num_mask we mapped to
1273     vector<int> up; ///< set of up osds for last pg we mapped to
1274     vector<int> acting; ///< set of acting osds for last pg we mapped to
1275     int up_primary = -1; ///< last up_primary we mapped to
1276     int acting_primary = -1;  ///< last acting_primary we mapped to
1277     int size = -1; ///< the size of the pool when were were last mapped
1278     int min_size = -1; ///< the min size of the pool when were were last mapped
1279     bool sort_bitwise = false; ///< whether the hobject_t sort order is bitwise
1280
1281     bool used_replica = false;
1282     bool paused = false;
1283
1284     int osd = -1;      ///< the final target osd, or -1
1285
1286     epoch_t last_force_resend = 0;
1287
1288     op_target_t(object_t oid, object_locator_t oloc, int flags)
1289       : flags(flags),
1290         base_oid(oid),
1291         base_oloc(oloc)
1292       {}
1293
1294     op_target_t(pg_t pgid)
1295       : base_oloc(pgid.pool(), pgid.ps()),
1296         precalc_pgid(true),
1297         base_pgid(pgid)
1298       {}
1299
1300     op_target_t() = default;
1301
1302     hobject_t get_hobj() {
1303       return hobject_t(target_oid,
1304                        target_oloc.key,
1305                        CEPH_NOSNAP,
1306                        target_oloc.hash >= 0 ? target_oloc.hash : pgid.ps(),
1307                        target_oloc.pool,
1308                        target_oloc.nspace);
1309     }
1310
1311     bool contained_by(const hobject_t& begin, const hobject_t& end) {
1312       hobject_t h = get_hobj();
1313       int r = cmp(h, begin);
1314       return r == 0 || (r > 0 && h < end);
1315     }
1316
1317     void dump(Formatter *f) const;
1318   };
1319
1320   struct Op : public RefCountedObject {
1321     OSDSession *session;
1322     int incarnation;
1323
1324     op_target_t target;
1325
1326     ConnectionRef con;  // for rx buffer only
1327     uint64_t features;  // explicitly specified op features
1328
1329     vector<OSDOp> ops;
1330
1331     snapid_t snapid;
1332     SnapContext snapc;
1333     ceph::real_time mtime;
1334
1335     bufferlist *outbl;
1336     vector<bufferlist*> out_bl;
1337     vector<Context*> out_handler;
1338     vector<int*> out_rval;
1339
1340     int priority;
1341     Context *onfinish;
1342     uint64_t ontimeout;
1343
1344     ceph_tid_t tid;
1345     int attempts;
1346
1347     version_t *objver;
1348     epoch_t *reply_epoch;
1349
1350     ceph::mono_time stamp;
1351
1352     epoch_t map_dne_bound;
1353
1354     bool budgeted;
1355
1356     /// true if we should resend this message on failure
1357     bool should_resend;
1358
1359     /// true if the throttle budget is get/put on a series of OPs,
1360     /// instead of per OP basis, when this flag is set, the budget is
1361     /// acquired before sending the very first OP of the series and
1362     /// released upon receiving the last OP reply.
1363     bool ctx_budgeted;
1364
1365     int *data_offset;
1366
1367     osd_reqid_t reqid; // explicitly setting reqid
1368     ZTracer::Trace trace;
1369
1370     Op(const object_t& o, const object_locator_t& ol, vector<OSDOp>& op,
1371        int f, Context *fin, version_t *ov, int *offset = NULL,
1372        ZTracer::Trace *parent_trace = nullptr) :
1373       session(NULL), incarnation(0),
1374       target(o, ol, f),
1375       con(NULL),
1376       features(CEPH_FEATURES_SUPPORTED_DEFAULT),
1377       snapid(CEPH_NOSNAP),
1378       outbl(NULL),
1379       priority(0),
1380       onfinish(fin),
1381       ontimeout(0),
1382       tid(0),
1383       attempts(0),
1384       objver(ov),
1385       reply_epoch(NULL),
1386       map_dne_bound(0),
1387       budgeted(false),
1388       should_resend(true),
1389       ctx_budgeted(false),
1390       data_offset(offset) {
1391       ops.swap(op);
1392
1393       /* initialize out_* to match op vector */
1394       out_bl.resize(ops.size());
1395       out_rval.resize(ops.size());
1396       out_handler.resize(ops.size());
1397       for (unsigned i = 0; i < ops.size(); i++) {
1398         out_bl[i] = NULL;
1399         out_handler[i] = NULL;
1400         out_rval[i] = NULL;
1401       }
1402
1403       if (target.base_oloc.key == o)
1404         target.base_oloc.key.clear();
1405
1406       if (parent_trace && parent_trace->valid()) {
1407         trace.init("op", nullptr, parent_trace);
1408         trace.event("start");
1409       }
1410     }
1411
1412     bool operator<(const Op& other) const {
1413       return tid < other.tid;
1414     }
1415
1416     bool respects_full() const {
1417       return
1418         (target.flags & (CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_RWORDERED)) &&
1419         !(target.flags & (CEPH_OSD_FLAG_FULL_TRY | CEPH_OSD_FLAG_FULL_FORCE));
1420     }
1421
1422   private:
1423     ~Op() override {
1424       while (!out_handler.empty()) {
1425         delete out_handler.back();
1426         out_handler.pop_back();
1427       }
1428       trace.event("finish");
1429     }
1430   };
1431
1432   struct C_Op_Map_Latest : public Context {
1433     Objecter *objecter;
1434     ceph_tid_t tid;
1435     version_t latest;
1436     C_Op_Map_Latest(Objecter *o, ceph_tid_t t) : objecter(o), tid(t),
1437                                                  latest(0) {}
1438     void finish(int r) override;
1439   };
1440
1441   struct C_Command_Map_Latest : public Context {
1442     Objecter *objecter;
1443     uint64_t tid;
1444     version_t latest;
1445     C_Command_Map_Latest(Objecter *o, ceph_tid_t t) :  objecter(o), tid(t),
1446                                                        latest(0) {}
1447     void finish(int r) override;
1448   };
1449
1450   struct C_Stat : public Context {
1451     bufferlist bl;
1452     uint64_t *psize;
1453     ceph::real_time *pmtime;
1454     Context *fin;
1455     C_Stat(uint64_t *ps, ceph::real_time *pm, Context *c) :
1456       psize(ps), pmtime(pm), fin(c) {}
1457     void finish(int r) override {
1458       if (r >= 0) {
1459         bufferlist::iterator p = bl.begin();
1460         uint64_t s;
1461         ceph::real_time m;
1462         ::decode(s, p);
1463         ::decode(m, p);
1464         if (psize)
1465           *psize = s;
1466         if (pmtime)
1467           *pmtime = m;
1468       }
1469       fin->complete(r);
1470     }
1471   };
1472
1473   struct C_GetAttrs : public Context {
1474     bufferlist bl;
1475     map<string,bufferlist>& attrset;
1476     Context *fin;
1477     C_GetAttrs(map<string, bufferlist>& set, Context *c) : attrset(set),
1478                                                            fin(c) {}
1479     void finish(int r) override {
1480       if (r >= 0) {
1481         bufferlist::iterator p = bl.begin();
1482         ::decode(attrset, p);
1483       }
1484       fin->complete(r);
1485     }
1486   };
1487
1488
1489   // Pools and statistics
1490   struct NListContext {
1491     collection_list_handle_t pos;
1492
1493     // these are for !sortbitwise compat only
1494     int current_pg = 0;
1495     int starting_pg_num = 0;
1496     bool sort_bitwise = false;
1497
1498     bool at_end_of_pool = false; ///< publicly visible end flag
1499
1500     int64_t pool_id = -1;
1501     int pool_snap_seq = 0;
1502     uint64_t max_entries = 0;
1503     string nspace;
1504
1505     bufferlist bl;   // raw data read to here
1506     std::list<librados::ListObjectImpl> list;
1507
1508     bufferlist filter;
1509
1510     bufferlist extra_info;
1511
1512     // The budget associated with this context, once it is set (>= 0),
1513     // the budget is not get/released on OP basis, instead the budget
1514     // is acquired before sending the first OP and released upon receiving
1515     // the last op reply.
1516     int ctx_budget = -1;
1517
1518     bool at_end() const {
1519       return at_end_of_pool;
1520     }
1521
1522     uint32_t get_pg_hash_position() const {
1523       return pos.get_hash();
1524     }
1525   };
1526
1527   struct C_NList : public Context {
1528     NListContext *list_context;
1529     Context *final_finish;
1530     Objecter *objecter;
1531     epoch_t epoch;
1532     C_NList(NListContext *lc, Context * finish, Objecter *ob) :
1533       list_context(lc), final_finish(finish), objecter(ob), epoch(0) {}
1534     void finish(int r) override {
1535       if (r >= 0) {
1536         objecter->_nlist_reply(list_context, r, final_finish, epoch);
1537       } else {
1538         final_finish->complete(r);
1539       }
1540     }
1541   };
1542
1543   struct PoolStatOp {
1544     ceph_tid_t tid;
1545     list<string> pools;
1546
1547     map<string,pool_stat_t> *pool_stats;
1548     Context *onfinish;
1549     uint64_t ontimeout;
1550
1551     ceph::mono_time last_submit;
1552   };
1553
1554   struct StatfsOp {
1555     ceph_tid_t tid;
1556     struct ceph_statfs *stats;
1557     Context *onfinish;
1558     uint64_t ontimeout;
1559
1560     ceph::mono_time last_submit;
1561   };
1562
1563   struct PoolOp {
1564     ceph_tid_t tid;
1565     int64_t pool;
1566     string name;
1567     Context *onfinish;
1568     uint64_t ontimeout;
1569     int pool_op;
1570     uint64_t auid;
1571     int16_t crush_rule;
1572     snapid_t snapid;
1573     bufferlist *blp;
1574
1575     ceph::mono_time last_submit;
1576     PoolOp() : tid(0), pool(0), onfinish(NULL), ontimeout(0), pool_op(0),
1577                auid(0), crush_rule(0), snapid(0), blp(NULL) {}
1578   };
1579
1580   // -- osd commands --
1581   struct CommandOp : public RefCountedObject {
1582     OSDSession *session = nullptr;
1583     ceph_tid_t tid = 0;
1584     vector<string> cmd;
1585     bufferlist inbl;
1586     bufferlist *poutbl = nullptr;
1587     string *prs = nullptr;
1588
1589     // target_osd == -1 means target_pg is valid
1590     const int target_osd = -1;
1591     const pg_t target_pg;
1592
1593     op_target_t target;
1594
1595     epoch_t map_dne_bound = 0;
1596     int map_check_error = 0; // error to return if map check fails
1597     const char *map_check_error_str = nullptr;
1598
1599     Context *onfinish = nullptr;
1600     uint64_t ontimeout = 0;
1601     ceph::mono_time last_submit;
1602
1603     CommandOp(
1604       int target_osd,
1605       const vector<string> &cmd,
1606       bufferlist inbl,
1607       bufferlist *poutbl,
1608       string *prs,
1609       Context *onfinish)
1610       : cmd(cmd),
1611         inbl(inbl),
1612         poutbl(poutbl),
1613         prs(prs),
1614         target_osd(target_osd),
1615         onfinish(onfinish) {}
1616
1617     CommandOp(
1618       pg_t pgid,
1619       const vector<string> &cmd,
1620       bufferlist inbl,
1621       bufferlist *poutbl,
1622       string *prs,
1623       Context *onfinish)
1624       : cmd(cmd),
1625         inbl(inbl),
1626         poutbl(poutbl),
1627         prs(prs),
1628         target_pg(pgid),
1629         target(pgid),
1630         onfinish(onfinish) {}
1631
1632   };
1633
1634   void submit_command(CommandOp *c, ceph_tid_t *ptid);
1635   int _calc_command_target(CommandOp *c, shunique_lock &sul);
1636   void _assign_command_session(CommandOp *c, shunique_lock &sul);
1637   void _send_command(CommandOp *c);
1638   int command_op_cancel(OSDSession *s, ceph_tid_t tid, int r);
1639   void _finish_command(CommandOp *c, int r, string rs);
1640   void handle_command_reply(MCommandReply *m);
1641
1642
1643   // -- lingering ops --
1644
1645   struct WatchContext {
1646     // this simply mirrors librados WatchCtx2
1647     virtual void handle_notify(uint64_t notify_id,
1648                                uint64_t cookie,
1649                                uint64_t notifier_id,
1650                                bufferlist& bl) = 0;
1651     virtual void handle_error(uint64_t cookie, int err) = 0;
1652     virtual ~WatchContext() {}
1653   };
1654
1655   struct LingerOp : public RefCountedObject {
1656     uint64_t linger_id;
1657
1658     op_target_t target;
1659
1660     snapid_t snap;
1661     SnapContext snapc;
1662     ceph::real_time mtime;
1663
1664     vector<OSDOp> ops;
1665     bufferlist inbl;
1666     bufferlist *poutbl;
1667     version_t *pobjver;
1668
1669     bool is_watch;
1670     ceph::mono_time watch_valid_thru; ///< send time for last acked ping
1671     int last_error;  ///< error from last failed ping|reconnect, if any
1672     boost::shared_mutex watch_lock;
1673     using lock_guard = std::unique_lock<decltype(watch_lock)>;
1674     using unique_lock = std::unique_lock<decltype(watch_lock)>;
1675     using shared_lock = boost::shared_lock<decltype(watch_lock)>;
1676     using shunique_lock = ceph::shunique_lock<decltype(watch_lock)>;
1677
1678     // queue of pending async operations, with the timestamp of
1679     // when they were queued.
1680     list<ceph::mono_time> watch_pending_async;
1681
1682     uint32_t register_gen;
1683     bool registered;
1684     bool canceled;
1685     Context *on_reg_commit;
1686
1687     // we trigger these from an async finisher
1688     Context *on_notify_finish;
1689     bufferlist *notify_result_bl;
1690     uint64_t notify_id;
1691
1692     WatchContext *watch_context;
1693
1694     OSDSession *session;
1695
1696     ceph_tid_t register_tid;
1697     ceph_tid_t ping_tid;
1698     epoch_t map_dne_bound;
1699
1700     void _queued_async() {
1701       // watch_lock ust be locked unique
1702       watch_pending_async.push_back(ceph::mono_clock::now());
1703     }
1704     void finished_async() {
1705       unique_lock l(watch_lock);
1706       assert(!watch_pending_async.empty());
1707       watch_pending_async.pop_front();
1708     }
1709
1710     LingerOp() : linger_id(0),
1711                  target(object_t(), object_locator_t(), 0),
1712                  snap(CEPH_NOSNAP), poutbl(NULL), pobjver(NULL),
1713                  is_watch(false), last_error(0),
1714                  register_gen(0),
1715                  registered(false),
1716                  canceled(false),
1717                  on_reg_commit(NULL),
1718                  on_notify_finish(NULL),
1719                  notify_result_bl(NULL),
1720                  notify_id(0),
1721                  watch_context(NULL),
1722                  session(NULL),
1723                  register_tid(0),
1724                  ping_tid(0),
1725                  map_dne_bound(0) {}
1726
1727     // no copy!
1728     const LingerOp &operator=(const LingerOp& r);
1729     LingerOp(const LingerOp& o);
1730
1731     uint64_t get_cookie() {
1732       return reinterpret_cast<uint64_t>(this);
1733     }
1734
1735   private:
1736     ~LingerOp() override {
1737       delete watch_context;
1738     }
1739   };
1740
1741   struct C_Linger_Commit : public Context {
1742     Objecter *objecter;
1743     LingerOp *info;
1744     bufferlist outbl;  // used for notify only
1745     C_Linger_Commit(Objecter *o, LingerOp *l) : objecter(o), info(l) {
1746       info->get();
1747     }
1748     ~C_Linger_Commit() override {
1749       info->put();
1750     }
1751     void finish(int r) override {
1752       objecter->_linger_commit(info, r, outbl);
1753     }
1754   };
1755
1756   struct C_Linger_Reconnect : public Context {
1757     Objecter *objecter;
1758     LingerOp *info;
1759     C_Linger_Reconnect(Objecter *o, LingerOp *l) : objecter(o), info(l) {
1760       info->get();
1761     }
1762     ~C_Linger_Reconnect() override {
1763       info->put();
1764     }
1765     void finish(int r) override {
1766       objecter->_linger_reconnect(info, r);
1767     }
1768   };
1769
1770   struct C_Linger_Ping : public Context {
1771     Objecter *objecter;
1772     LingerOp *info;
1773     ceph::mono_time sent;
1774     uint32_t register_gen;
1775     C_Linger_Ping(Objecter *o, LingerOp *l)
1776       : objecter(o), info(l), register_gen(info->register_gen) {
1777       info->get();
1778     }
1779     ~C_Linger_Ping() override {
1780       info->put();
1781     }
1782     void finish(int r) override {
1783       objecter->_linger_ping(info, r, sent, register_gen);
1784     }
1785   };
1786
1787   struct C_Linger_Map_Latest : public Context {
1788     Objecter *objecter;
1789     uint64_t linger_id;
1790     version_t latest;
1791     C_Linger_Map_Latest(Objecter *o, uint64_t id) :
1792       objecter(o), linger_id(id), latest(0) {}
1793     void finish(int r) override;
1794   };
1795
1796   // -- osd sessions --
1797   struct OSDBackoff {
1798     spg_t pgid;
1799     uint64_t id;
1800     hobject_t begin, end;
1801   };
1802
1803   struct OSDSession : public RefCountedObject {
1804     boost::shared_mutex lock;
1805     using lock_guard = std::lock_guard<decltype(lock)>;
1806     using unique_lock = std::unique_lock<decltype(lock)>;
1807     using shared_lock = boost::shared_lock<decltype(lock)>;
1808     using shunique_lock = ceph::shunique_lock<decltype(lock)>;
1809
1810     // pending ops
1811     map<ceph_tid_t,Op*> ops;
1812     map<uint64_t, LingerOp*> linger_ops;
1813     map<ceph_tid_t,CommandOp*> command_ops;
1814
1815     // backoffs
1816     map<spg_t,map<hobject_t,OSDBackoff>> backoffs;
1817     map<uint64_t,OSDBackoff*> backoffs_by_id;
1818
1819     int osd;
1820     int incarnation;
1821     ConnectionRef con;
1822     int num_locks;
1823     std::unique_ptr<std::mutex[]> completion_locks;
1824     using unique_completion_lock = std::unique_lock<
1825       decltype(completion_locks)::element_type>;
1826
1827
1828     OSDSession(CephContext *cct, int o) :
1829       osd(o), incarnation(0), con(NULL),
1830       num_locks(cct->_conf->objecter_completion_locks_per_session),
1831       completion_locks(new std::mutex[num_locks]) {}
1832
1833     ~OSDSession() override;
1834
1835     bool is_homeless() { return (osd == -1); }
1836
1837     unique_completion_lock get_lock(object_t& oid);
1838   };
1839   map<int,OSDSession*> osd_sessions;
1840
1841   bool osdmap_full_flag() const;
1842   bool osdmap_pool_full(const int64_t pool_id) const;
1843
1844  private:
1845
1846   /**
1847    * Test pg_pool_t::FLAG_FULL on a pool
1848    *
1849    * @return true if the pool exists and has the flag set, or
1850    *         the global full flag is set, else false
1851    */
1852   bool _osdmap_pool_full(const int64_t pool_id) const;
1853   bool _osdmap_pool_full(const pg_pool_t &p) const;
1854   void update_pool_full_map(map<int64_t, bool>& pool_full_map);
1855
1856   map<uint64_t, LingerOp*> linger_ops;
1857   // we use this just to confirm a cookie is valid before dereferencing the ptr
1858   set<LingerOp*> linger_ops_set;
1859
1860   map<ceph_tid_t,PoolStatOp*> poolstat_ops;
1861   map<ceph_tid_t,StatfsOp*> statfs_ops;
1862   map<ceph_tid_t,PoolOp*> pool_ops;
1863   std::atomic<unsigned> num_homeless_ops{0};
1864
1865   OSDSession *homeless_session;
1866
1867   // ops waiting for an osdmap with a new pool or confirmation that
1868   // the pool does not exist (may be expanded to other uses later)
1869   map<uint64_t, LingerOp*> check_latest_map_lingers;
1870   map<ceph_tid_t, Op*> check_latest_map_ops;
1871   map<ceph_tid_t, CommandOp*> check_latest_map_commands;
1872
1873   map<epoch_t,list< pair<Context*, int> > > waiting_for_map;
1874
1875   ceph::timespan mon_timeout;
1876   ceph::timespan osd_timeout;
1877
1878   MOSDOp *_prepare_osd_op(Op *op);
1879   void _send_op(Op *op, MOSDOp *m = NULL);
1880   void _send_op_account(Op *op);
1881   void _cancel_linger_op(Op *op);
1882   void finish_op(OSDSession *session, ceph_tid_t tid);
1883   void _finish_op(Op *op, int r);
1884   static bool is_pg_changed(
1885     int oldprimary,
1886     const vector<int>& oldacting,
1887     int newprimary,
1888     const vector<int>& newacting,
1889     bool any_change=false);
1890   enum recalc_op_target_result {
1891     RECALC_OP_TARGET_NO_ACTION = 0,
1892     RECALC_OP_TARGET_NEED_RESEND,
1893     RECALC_OP_TARGET_POOL_DNE,
1894     RECALC_OP_TARGET_OSD_DNE,
1895     RECALC_OP_TARGET_OSD_DOWN,
1896   };
1897   bool _osdmap_full_flag() const;
1898   bool _osdmap_has_pool_full() const;
1899
1900   bool target_should_be_paused(op_target_t *op);
1901   int _calc_target(op_target_t *t, Connection *con,
1902                    bool any_change = false);
1903   int _map_session(op_target_t *op, OSDSession **s,
1904                    shunique_lock& lc);
1905
1906   void _session_op_assign(OSDSession *s, Op *op);
1907   void _session_op_remove(OSDSession *s, Op *op);
1908   void _session_linger_op_assign(OSDSession *to, LingerOp *op);
1909   void _session_linger_op_remove(OSDSession *from, LingerOp *op);
1910   void _session_command_op_assign(OSDSession *to, CommandOp *op);
1911   void _session_command_op_remove(OSDSession *from, CommandOp *op);
1912
1913   int _assign_op_target_session(Op *op, shunique_lock& lc,
1914                                 bool src_session_locked,
1915                                 bool dst_session_locked);
1916   int _recalc_linger_op_target(LingerOp *op, shunique_lock& lc);
1917
1918   void _linger_submit(LingerOp *info, shunique_lock& sul);
1919   void _send_linger(LingerOp *info, shunique_lock& sul);
1920   void _linger_commit(LingerOp *info, int r, bufferlist& outbl);
1921   void _linger_reconnect(LingerOp *info, int r);
1922   void _send_linger_ping(LingerOp *info);
1923   void _linger_ping(LingerOp *info, int r, ceph::mono_time sent,
1924                     uint32_t register_gen);
1925   int _normalize_watch_error(int r);
1926
1927   friend class C_DoWatchError;
1928 public:
1929   void linger_callback_flush(Context *ctx) {
1930     finisher->queue(ctx);
1931   }
1932
1933 private:
1934   void _check_op_pool_dne(Op *op, unique_lock *sl);
1935   void _send_op_map_check(Op *op);
1936   void _op_cancel_map_check(Op *op);
1937   void _check_linger_pool_dne(LingerOp *op, bool *need_unregister);
1938   void _send_linger_map_check(LingerOp *op);
1939   void _linger_cancel_map_check(LingerOp *op);
1940   void _check_command_map_dne(CommandOp *op);
1941   void _send_command_map_check(CommandOp *op);
1942   void _command_cancel_map_check(CommandOp *op);
1943
1944   void kick_requests(OSDSession *session);
1945   void _kick_requests(OSDSession *session, map<uint64_t, LingerOp *>& lresend);
1946   void _linger_ops_resend(map<uint64_t, LingerOp *>& lresend, unique_lock& ul);
1947
1948   int _get_session(int osd, OSDSession **session, shunique_lock& sul);
1949   void put_session(OSDSession *s);
1950   void get_session(OSDSession *s);
1951   void _reopen_session(OSDSession *session);
1952   void close_session(OSDSession *session);
1953
1954   void _nlist_reply(NListContext *list_context, int r, Context *final_finish,
1955                    epoch_t reply_epoch);
1956
1957   void resend_mon_ops();
1958
1959   /**
1960    * handle a budget for in-flight ops
1961    * budget is taken whenever an op goes into the ops map
1962    * and returned whenever an op is removed from the map
1963    * If throttle_op needs to throttle it will unlock client_lock.
1964    */
1965   int calc_op_budget(Op *op);
1966   void _throttle_op(Op *op, shunique_lock& sul, int op_size = 0);
1967   int _take_op_budget(Op *op, shunique_lock& sul) {
1968     assert(sul && sul.mutex() == &rwlock);
1969     int op_budget = calc_op_budget(op);
1970     if (keep_balanced_budget) {
1971       _throttle_op(op, sul, op_budget);
1972     } else {
1973       op_throttle_bytes.take(op_budget);
1974       op_throttle_ops.take(1);
1975     }
1976     op->budgeted = true;
1977     return op_budget;
1978   }
1979   void put_op_budget_bytes(int op_budget) {
1980     assert(op_budget >= 0);
1981     op_throttle_bytes.put(op_budget);
1982     op_throttle_ops.put(1);
1983   }
1984   void put_op_budget(Op *op) {
1985     assert(op->budgeted);
1986     int op_budget = calc_op_budget(op);
1987     put_op_budget_bytes(op_budget);
1988   }
1989   void put_nlist_context_budget(NListContext *list_context);
1990   Throttle op_throttle_bytes, op_throttle_ops;
1991
1992  public:
1993   Objecter(CephContext *cct_, Messenger *m, MonClient *mc,
1994            Finisher *fin,
1995            double mon_timeout,
1996            double osd_timeout) :
1997     Dispatcher(cct_), messenger(m), monc(mc), finisher(fin),
1998     trace_endpoint("0.0.0.0", 0, "Objecter"),
1999     osdmap(new OSDMap),
2000     max_linger_id(0),
2001     keep_balanced_budget(false), honor_osdmap_full(true), osdmap_full_try(false),
2002     blacklist_events_enabled(false),
2003     last_seen_osdmap_version(0), last_seen_pgmap_version(0),
2004     logger(NULL), tick_event(0), m_request_state_hook(NULL),
2005     homeless_session(new OSDSession(cct, -1)),
2006     mon_timeout(ceph::make_timespan(mon_timeout)),
2007     osd_timeout(ceph::make_timespan(osd_timeout)),
2008     op_throttle_bytes(cct, "objecter_bytes",
2009                       cct->_conf->objecter_inflight_op_bytes),
2010     op_throttle_ops(cct, "objecter_ops", cct->_conf->objecter_inflight_ops),
2011     epoch_barrier(0),
2012     retry_writes_after_first_reply(cct->_conf->objecter_retry_writes_after_first_reply)
2013   { }
2014   ~Objecter() override;
2015
2016   void init();
2017   void start(const OSDMap *o = nullptr);
2018   void shutdown();
2019
2020   // These two templates replace osdmap_(get)|(put)_read. Simply wrap
2021   // whatever functionality you want to use the OSDMap in a lambda like:
2022   //
2023   // with_osdmap([](const OSDMap& o) { o.do_stuff(); });
2024   //
2025   // or
2026   //
2027   // auto t = with_osdmap([&](const OSDMap& o) { return o.lookup_stuff(x); });
2028   //
2029   // Do not call into something that will try to lock the OSDMap from
2030   // here or you will have great woe and misery.
2031
2032   template<typename Callback, typename...Args>
2033   auto with_osdmap(Callback&& cb, Args&&... args) const ->
2034     decltype(cb(*osdmap, std::forward<Args>(args)...)) {
2035     shared_lock l(rwlock);
2036     return std::forward<Callback>(cb)(*osdmap, std::forward<Args>(args)...);
2037   }
2038
2039
2040   /**
2041    * Tell the objecter to throttle outgoing ops according to its
2042    * budget (in _conf). If you do this, ops can block, in
2043    * which case it will unlock client_lock and sleep until
2044    * incoming messages reduce the used budget low enough for
2045    * the ops to continue going; then it will lock client_lock again.
2046    */
2047   void set_balanced_budget() { keep_balanced_budget = true; }
2048   void unset_balanced_budget() { keep_balanced_budget = false; }
2049
2050   void set_honor_osdmap_full() { honor_osdmap_full = true; }
2051   void unset_honor_osdmap_full() { honor_osdmap_full = false; }
2052
2053   void set_osdmap_full_try() { osdmap_full_try = true; }
2054   void unset_osdmap_full_try() { osdmap_full_try = false; }
2055
2056   void _scan_requests(OSDSession *s,
2057                       bool force_resend,
2058                       bool cluster_full,
2059                       map<int64_t, bool> *pool_full_map,
2060                       map<ceph_tid_t, Op*>& need_resend,
2061                       list<LingerOp*>& need_resend_linger,
2062                       map<ceph_tid_t, CommandOp*>& need_resend_command,
2063                       shunique_lock& sul);
2064
2065   int64_t get_object_hash_position(int64_t pool, const string& key,
2066                                    const string& ns);
2067   int64_t get_object_pg_hash_position(int64_t pool, const string& key,
2068                                       const string& ns);
2069
2070   // messages
2071  public:
2072   bool ms_dispatch(Message *m) override;
2073   bool ms_can_fast_dispatch_any() const override {
2074     return true;
2075   }
2076   bool ms_can_fast_dispatch(const Message *m) const override {
2077     switch (m->get_type()) {
2078     case CEPH_MSG_OSD_OPREPLY:
2079     case CEPH_MSG_WATCH_NOTIFY:
2080       return true;
2081     default:
2082       return false;
2083     }
2084   }
2085   void ms_fast_dispatch(Message *m) override {
2086     ms_dispatch(m);
2087   }
2088
2089   void handle_osd_op_reply(class MOSDOpReply *m);
2090   void handle_osd_backoff(class MOSDBackoff *m);
2091   void handle_watch_notify(class MWatchNotify *m);
2092   void handle_osd_map(class MOSDMap *m);
2093   void wait_for_osd_map();
2094
2095   /**
2096    * Get list of entities blacklisted since this was last called,
2097    * and reset the list.
2098    *
2099    * Uses a std::set because typical use case is to compare some
2100    * other list of clients to see which overlap with the blacklisted
2101    * addrs.
2102    *
2103    */
2104   void consume_blacklist_events(std::set<entity_addr_t> *events);
2105
2106   int pool_snap_by_name(int64_t poolid,
2107                         const char *snap_name,
2108                         snapid_t *snap) const;
2109   int pool_snap_get_info(int64_t poolid, snapid_t snap,
2110                          pool_snap_info_t *info) const;
2111   int pool_snap_list(int64_t poolid, vector<uint64_t> *snaps);
2112 private:
2113
2114   void emit_blacklist_events(const OSDMap::Incremental &inc);
2115   void emit_blacklist_events(const OSDMap &old_osd_map,
2116                              const OSDMap &new_osd_map);
2117
2118   // low-level
2119   void _op_submit(Op *op, shunique_lock& lc, ceph_tid_t *ptid);
2120   void _op_submit_with_budget(Op *op, shunique_lock& lc,
2121                               ceph_tid_t *ptid,
2122                               int *ctx_budget = NULL);
2123   inline void unregister_op(Op *op);
2124
2125   // public interface
2126 public:
2127   void op_submit(Op *op, ceph_tid_t *ptid = NULL, int *ctx_budget = NULL);
2128   bool is_active() {
2129     shared_lock l(rwlock);
2130     return !((!inflight_ops) && linger_ops.empty() &&
2131              poolstat_ops.empty() && statfs_ops.empty());
2132   }
2133
2134   /**
2135    * Output in-flight requests
2136    */
2137   void _dump_active(OSDSession *s);
2138   void _dump_active();
2139   void dump_active();
2140   void dump_requests(Formatter *fmt);
2141   void _dump_ops(const OSDSession *s, Formatter *fmt);
2142   void dump_ops(Formatter *fmt);
2143   void _dump_linger_ops(const OSDSession *s, Formatter *fmt);
2144   void dump_linger_ops(Formatter *fmt);
2145   void _dump_command_ops(const OSDSession *s, Formatter *fmt);
2146   void dump_command_ops(Formatter *fmt);
2147   void dump_pool_ops(Formatter *fmt) const;
2148   void dump_pool_stat_ops(Formatter *fmt) const;
2149   void dump_statfs_ops(Formatter *fmt) const;
2150
2151   int get_client_incarnation() const { return client_inc; }
2152   void set_client_incarnation(int inc) { client_inc = inc; }
2153
2154   bool have_map(epoch_t epoch);
2155   /// wait for epoch; true if we already have it
2156   bool wait_for_map(epoch_t epoch, Context *c, int err=0);
2157   void _wait_for_new_map(Context *c, epoch_t epoch, int err=0);
2158   void wait_for_latest_osdmap(Context *fin);
2159   void get_latest_version(epoch_t oldest, epoch_t neweset, Context *fin);
2160   void _get_latest_version(epoch_t oldest, epoch_t neweset, Context *fin);
2161
2162   /** Get the current set of global op flags */
2163   int get_global_op_flags() const { return global_op_flags; }
2164   /** Add a flag to the global op flags, not really atomic operation */
2165   void add_global_op_flags(int flag) {
2166     global_op_flags.fetch_or(flag);
2167   }
2168   /** Clear the passed flags from the global op flag set */
2169   void clear_global_op_flag(int flags) {
2170     global_op_flags.fetch_and(~flags);
2171   }
2172
2173   /// cancel an in-progress request with the given return code
2174 private:
2175   int op_cancel(OSDSession *s, ceph_tid_t tid, int r);
2176   int _op_cancel(ceph_tid_t tid, int r);
2177 public:
2178   int op_cancel(ceph_tid_t tid, int r);
2179
2180   /**
2181    * Any write op which is in progress at the start of this call shall no
2182    * longer be in progress when this call ends.  Operations started after the
2183    * start of this call may still be in progress when this call ends.
2184    *
2185    * @return the latest possible epoch in which a cancelled op could have
2186    *         existed, or -1 if nothing was cancelled.
2187    */
2188   epoch_t op_cancel_writes(int r, int64_t pool=-1);
2189
2190   // commands
2191   void osd_command(int osd, const std::vector<string>& cmd,
2192                   const bufferlist& inbl, ceph_tid_t *ptid,
2193                   bufferlist *poutbl, string *prs, Context *onfinish) {
2194     assert(osd >= 0);
2195     CommandOp *c = new CommandOp(
2196       osd,
2197       cmd,
2198       inbl,
2199       poutbl,
2200       prs,
2201       onfinish);
2202     submit_command(c, ptid);
2203   }
2204   void pg_command(pg_t pgid, vector<string>& cmd,
2205                  const bufferlist& inbl, ceph_tid_t *ptid,
2206                  bufferlist *poutbl, string *prs, Context *onfinish) {
2207     CommandOp *c = new CommandOp(
2208       pgid,
2209       cmd,
2210       inbl,
2211       poutbl,
2212       prs,
2213       onfinish);
2214     submit_command(c, ptid);
2215   }
2216
2217   // mid-level helpers
2218   Op *prepare_mutate_op(
2219     const object_t& oid, const object_locator_t& oloc,
2220     ObjectOperation& op, const SnapContext& snapc,
2221     ceph::real_time mtime, int flags,
2222     Context *oncommit, version_t *objver = NULL,
2223     osd_reqid_t reqid = osd_reqid_t(),
2224     ZTracer::Trace *parent_trace = nullptr) {
2225     Op *o = new Op(oid, oloc, op.ops, flags | global_op_flags |
2226                    CEPH_OSD_FLAG_WRITE, oncommit, objver, nullptr, parent_trace);
2227     o->priority = op.priority;
2228     o->mtime = mtime;
2229     o->snapc = snapc;
2230     o->out_rval.swap(op.out_rval);
2231     o->reqid = reqid;
2232     return o;
2233   }
2234   ceph_tid_t mutate(
2235     const object_t& oid, const object_locator_t& oloc,
2236     ObjectOperation& op, const SnapContext& snapc,
2237     ceph::real_time mtime, int flags,
2238     Context *oncommit, version_t *objver = NULL,
2239     osd_reqid_t reqid = osd_reqid_t()) {
2240     Op *o = prepare_mutate_op(oid, oloc, op, snapc, mtime, flags,
2241                               oncommit, objver, reqid);
2242     ceph_tid_t tid;
2243     op_submit(o, &tid);
2244     return tid;
2245   }
2246   Op *prepare_read_op(
2247     const object_t& oid, const object_locator_t& oloc,
2248     ObjectOperation& op,
2249     snapid_t snapid, bufferlist *pbl, int flags,
2250     Context *onack, version_t *objver = NULL,
2251     int *data_offset = NULL,
2252     uint64_t features = 0,
2253     ZTracer::Trace *parent_trace = nullptr) {
2254     Op *o = new Op(oid, oloc, op.ops, flags | global_op_flags |
2255                    CEPH_OSD_FLAG_READ, onack, objver, data_offset, parent_trace);
2256     o->priority = op.priority;
2257     o->snapid = snapid;
2258     o->outbl = pbl;
2259     if (!o->outbl && op.size() == 1 && op.out_bl[0]->length())
2260         o->outbl = op.out_bl[0];
2261     o->out_bl.swap(op.out_bl);
2262     o->out_handler.swap(op.out_handler);
2263     o->out_rval.swap(op.out_rval);
2264     return o;
2265   }
2266   ceph_tid_t read(
2267     const object_t& oid, const object_locator_t& oloc,
2268     ObjectOperation& op,
2269     snapid_t snapid, bufferlist *pbl, int flags,
2270     Context *onack, version_t *objver = NULL,
2271     int *data_offset = NULL,
2272     uint64_t features = 0) {
2273     Op *o = prepare_read_op(oid, oloc, op, snapid, pbl, flags, onack, objver,
2274                             data_offset);
2275     if (features)
2276       o->features = features;
2277     ceph_tid_t tid;
2278     op_submit(o, &tid);
2279     return tid;
2280   }
2281   Op *prepare_pg_read_op(
2282     uint32_t hash, object_locator_t oloc,
2283     ObjectOperation& op, bufferlist *pbl, int flags,
2284     Context *onack, epoch_t *reply_epoch,
2285     int *ctx_budget) {
2286     Op *o = new Op(object_t(), oloc,
2287                    op.ops,
2288                    flags | global_op_flags | CEPH_OSD_FLAG_READ |
2289                    CEPH_OSD_FLAG_IGNORE_OVERLAY,
2290                    onack, NULL);
2291     o->target.precalc_pgid = true;
2292     o->target.base_pgid = pg_t(hash, oloc.pool);
2293     o->priority = op.priority;
2294     o->snapid = CEPH_NOSNAP;
2295     o->outbl = pbl;
2296     o->out_bl.swap(op.out_bl);
2297     o->out_handler.swap(op.out_handler);
2298     o->out_rval.swap(op.out_rval);
2299     o->reply_epoch = reply_epoch;
2300     if (ctx_budget) {
2301       // budget is tracked by listing context
2302       o->ctx_budgeted = true;
2303     }
2304     return o;
2305   }
2306   ceph_tid_t pg_read(
2307     uint32_t hash, object_locator_t oloc,
2308     ObjectOperation& op, bufferlist *pbl, int flags,
2309     Context *onack, epoch_t *reply_epoch,
2310     int *ctx_budget) {
2311     Op *o = prepare_pg_read_op(hash, oloc, op, pbl, flags,
2312                                onack, reply_epoch, ctx_budget);
2313     ceph_tid_t tid;
2314     op_submit(o, &tid, ctx_budget);
2315     return tid;
2316   }
2317
2318   // caller owns a ref
2319   LingerOp *linger_register(const object_t& oid, const object_locator_t& oloc,
2320                             int flags);
2321   ceph_tid_t linger_watch(LingerOp *info,
2322                           ObjectOperation& op,
2323                           const SnapContext& snapc, ceph::real_time mtime,
2324                           bufferlist& inbl,
2325                           Context *onfinish,
2326                           version_t *objver);
2327   ceph_tid_t linger_notify(LingerOp *info,
2328                            ObjectOperation& op,
2329                            snapid_t snap, bufferlist& inbl,
2330                            bufferlist *poutbl,
2331                            Context *onack,
2332                            version_t *objver);
2333   int linger_check(LingerOp *info);
2334   void linger_cancel(LingerOp *info);  // releases a reference
2335   void _linger_cancel(LingerOp *info);
2336
2337   void _do_watch_notify(LingerOp *info, MWatchNotify *m);
2338
2339   /**
2340    * set up initial ops in the op vector, and allocate a final op slot.
2341    *
2342    * The caller is responsible for filling in the final ops_count ops.
2343    *
2344    * @param ops op vector
2345    * @param ops_count number of final ops the caller will fill in
2346    * @param extra_ops pointer to [array of] initial op[s]
2347    * @return index of final op (for caller to fill in)
2348    */
2349   int init_ops(vector<OSDOp>& ops, int ops_count, ObjectOperation *extra_ops) {
2350     int i;
2351     int extra = 0;
2352
2353     if (extra_ops)
2354       extra = extra_ops->ops.size();
2355
2356     ops.resize(ops_count + extra);
2357
2358     for (i=0; i<extra; i++) {
2359       ops[i] = extra_ops->ops[i];
2360     }
2361
2362     return i;
2363   }
2364
2365
2366   // high-level helpers
2367   Op *prepare_stat_op(
2368     const object_t& oid, const object_locator_t& oloc,
2369     snapid_t snap, uint64_t *psize, ceph::real_time *pmtime,
2370     int flags, Context *onfinish, version_t *objver = NULL,
2371     ObjectOperation *extra_ops = NULL) {
2372     vector<OSDOp> ops;
2373     int i = init_ops(ops, 1, extra_ops);
2374     ops[i].op.op = CEPH_OSD_OP_STAT;
2375     C_Stat *fin = new C_Stat(psize, pmtime, onfinish);
2376     Op *o = new Op(oid, oloc, ops, flags | global_op_flags |
2377                    CEPH_OSD_FLAG_READ, fin, objver);
2378     o->snapid = snap;
2379     o->outbl = &fin->bl;
2380     return o;
2381   }
2382   ceph_tid_t stat(
2383     const object_t& oid, const object_locator_t& oloc,
2384     snapid_t snap, uint64_t *psize, ceph::real_time *pmtime,
2385     int flags, Context *onfinish, version_t *objver = NULL,
2386     ObjectOperation *extra_ops = NULL) {
2387     Op *o = prepare_stat_op(oid, oloc, snap, psize, pmtime, flags,
2388                             onfinish, objver, extra_ops);
2389     ceph_tid_t tid;
2390     op_submit(o, &tid);
2391     return tid;
2392   }
2393
2394   Op *prepare_read_op(
2395     const object_t& oid, const object_locator_t& oloc,
2396     uint64_t off, uint64_t len, snapid_t snap, bufferlist *pbl,
2397     int flags, Context *onfinish, version_t *objver = NULL,
2398     ObjectOperation *extra_ops = NULL, int op_flags = 0,
2399     ZTracer::Trace *parent_trace = nullptr) {
2400     vector<OSDOp> ops;
2401     int i = init_ops(ops, 1, extra_ops);
2402     ops[i].op.op = CEPH_OSD_OP_READ;
2403     ops[i].op.extent.offset = off;
2404     ops[i].op.extent.length = len;
2405     ops[i].op.extent.truncate_size = 0;
2406     ops[i].op.extent.truncate_seq = 0;
2407     ops[i].op.flags = op_flags;
2408     Op *o = new Op(oid, oloc, ops, flags | global_op_flags |
2409                    CEPH_OSD_FLAG_READ, onfinish, objver, nullptr, parent_trace);
2410     o->snapid = snap;
2411     o->outbl = pbl;
2412     return o;
2413   }
2414   ceph_tid_t read(
2415     const object_t& oid, const object_locator_t& oloc,
2416     uint64_t off, uint64_t len, snapid_t snap, bufferlist *pbl,
2417     int flags, Context *onfinish, version_t *objver = NULL,
2418     ObjectOperation *extra_ops = NULL, int op_flags = 0) {
2419     Op *o = prepare_read_op(oid, oloc, off, len, snap, pbl, flags,
2420                             onfinish, objver, extra_ops, op_flags);
2421     ceph_tid_t tid;
2422     op_submit(o, &tid);
2423     return tid;
2424   }
2425
2426   Op *prepare_cmpext_op(
2427     const object_t& oid, const object_locator_t& oloc,
2428     uint64_t off, bufferlist &cmp_bl,
2429     snapid_t snap, int flags, Context *onfinish, version_t *objver = NULL,
2430     ObjectOperation *extra_ops = NULL, int op_flags = 0) {
2431     vector<OSDOp> ops;
2432     int i = init_ops(ops, 1, extra_ops);
2433     ops[i].op.op = CEPH_OSD_OP_CMPEXT;
2434     ops[i].op.extent.offset = off;
2435     ops[i].op.extent.length = cmp_bl.length();
2436     ops[i].op.extent.truncate_size = 0;
2437     ops[i].op.extent.truncate_seq = 0;
2438     ops[i].indata = cmp_bl;
2439     ops[i].op.flags = op_flags;
2440     Op *o = new Op(oid, oloc, ops, flags | global_op_flags |
2441                    CEPH_OSD_FLAG_READ, onfinish, objver);
2442     o->snapid = snap;
2443     return o;
2444   }
2445
2446   ceph_tid_t cmpext(
2447     const object_t& oid, const object_locator_t& oloc,
2448     uint64_t off, bufferlist &cmp_bl,
2449     snapid_t snap, int flags, Context *onfinish, version_t *objver = NULL,
2450     ObjectOperation *extra_ops = NULL, int op_flags = 0) {
2451     Op *o = prepare_cmpext_op(oid, oloc, off, cmp_bl, snap,
2452                               flags, onfinish, objver, extra_ops, op_flags);
2453     ceph_tid_t tid;
2454     op_submit(o, &tid);
2455     return tid;
2456   }
2457
2458   ceph_tid_t read_trunc(const object_t& oid, const object_locator_t& oloc,
2459                         uint64_t off, uint64_t len, snapid_t snap,
2460                         bufferlist *pbl, int flags, uint64_t trunc_size,
2461                         __u32 trunc_seq, Context *onfinish,
2462                         version_t *objver = NULL,
2463                         ObjectOperation *extra_ops = NULL, int op_flags = 0) {
2464     vector<OSDOp> ops;
2465     int i = init_ops(ops, 1, extra_ops);
2466     ops[i].op.op = CEPH_OSD_OP_READ;
2467     ops[i].op.extent.offset = off;
2468     ops[i].op.extent.length = len;
2469     ops[i].op.extent.truncate_size = trunc_size;
2470     ops[i].op.extent.truncate_seq = trunc_seq;
2471     ops[i].op.flags = op_flags;
2472     Op *o = new Op(oid, oloc, ops, flags | global_op_flags |
2473                    CEPH_OSD_FLAG_READ, onfinish, objver);
2474     o->snapid = snap;
2475     o->outbl = pbl;
2476     ceph_tid_t tid;
2477     op_submit(o, &tid);
2478     return tid;
2479   }
2480   ceph_tid_t mapext(const object_t& oid, const object_locator_t& oloc,
2481                     uint64_t off, uint64_t len, snapid_t snap, bufferlist *pbl,
2482                     int flags, Context *onfinish, version_t *objver = NULL,
2483                     ObjectOperation *extra_ops = NULL) {
2484     vector<OSDOp> ops;
2485     int i = init_ops(ops, 1, extra_ops);
2486     ops[i].op.op = CEPH_OSD_OP_MAPEXT;
2487     ops[i].op.extent.offset = off;
2488     ops[i].op.extent.length = len;
2489     ops[i].op.extent.truncate_size = 0;
2490     ops[i].op.extent.truncate_seq = 0;
2491     Op *o = new Op(oid, oloc, ops, flags | global_op_flags |
2492                    CEPH_OSD_FLAG_READ, onfinish, objver);
2493     o->snapid = snap;
2494     o->outbl = pbl;
2495     ceph_tid_t tid;
2496     op_submit(o, &tid);
2497     return tid;
2498   }
2499   ceph_tid_t getxattr(const object_t& oid, const object_locator_t& oloc,
2500              const char *name, snapid_t snap, bufferlist *pbl, int flags,
2501              Context *onfinish,
2502              version_t *objver = NULL, ObjectOperation *extra_ops = NULL) {
2503     vector<OSDOp> ops;
2504     int i = init_ops(ops, 1, extra_ops);
2505     ops[i].op.op = CEPH_OSD_OP_GETXATTR;
2506     ops[i].op.xattr.name_len = (name ? strlen(name) : 0);
2507     ops[i].op.xattr.value_len = 0;
2508     if (name)
2509       ops[i].indata.append(name);
2510     Op *o = new Op(oid, oloc, ops, flags | global_op_flags |
2511                    CEPH_OSD_FLAG_READ, onfinish, objver);
2512     o->snapid = snap;
2513     o->outbl = pbl;
2514     ceph_tid_t tid;
2515     op_submit(o, &tid);
2516     return tid;
2517   }
2518
2519   ceph_tid_t getxattrs(const object_t& oid, const object_locator_t& oloc,
2520                        snapid_t snap, map<string,bufferlist>& attrset,
2521                        int flags, Context *onfinish, version_t *objver = NULL,
2522                        ObjectOperation *extra_ops = NULL) {
2523     vector<OSDOp> ops;
2524     int i = init_ops(ops, 1, extra_ops);
2525     ops[i].op.op = CEPH_OSD_OP_GETXATTRS;
2526     C_GetAttrs *fin = new C_GetAttrs(attrset, onfinish);
2527     Op *o = new Op(oid, oloc, ops, flags | global_op_flags |
2528                    CEPH_OSD_FLAG_READ, fin, objver);
2529     o->snapid = snap;
2530     o->outbl = &fin->bl;
2531     ceph_tid_t tid;
2532     op_submit(o, &tid);
2533     return tid;
2534   }
2535
2536   ceph_tid_t read_full(const object_t& oid, const object_locator_t& oloc,
2537                        snapid_t snap, bufferlist *pbl, int flags,
2538                        Context *onfinish, version_t *objver = NULL,
2539                        ObjectOperation *extra_ops = NULL) {
2540     return read(oid, oloc, 0, 0, snap, pbl, flags | global_op_flags |
2541                 CEPH_OSD_FLAG_READ, onfinish, objver, extra_ops);
2542   }
2543
2544
2545   // writes
2546   ceph_tid_t _modify(const object_t& oid, const object_locator_t& oloc,
2547                      vector<OSDOp>& ops, ceph::real_time mtime,
2548                      const SnapContext& snapc, int flags,
2549                      Context *oncommit,
2550                      version_t *objver = NULL) {
2551     Op *o = new Op(oid, oloc, ops, flags | global_op_flags |
2552                    CEPH_OSD_FLAG_WRITE, oncommit, objver);
2553     o->mtime = mtime;
2554     o->snapc = snapc;
2555     ceph_tid_t tid;
2556     op_submit(o, &tid);
2557     return tid;
2558   }
2559   Op *prepare_write_op(
2560     const object_t& oid, const object_locator_t& oloc,
2561     uint64_t off, uint64_t len, const SnapContext& snapc,
2562     const bufferlist &bl, ceph::real_time mtime, int flags,
2563     Context *oncommit, version_t *objver = NULL,
2564     ObjectOperation *extra_ops = NULL, int op_flags = 0,
2565     ZTracer::Trace *parent_trace = nullptr) {
2566     vector<OSDOp> ops;
2567     int i = init_ops(ops, 1, extra_ops);
2568     ops[i].op.op = CEPH_OSD_OP_WRITE;
2569     ops[i].op.extent.offset = off;
2570     ops[i].op.extent.length = len;
2571     ops[i].op.extent.truncate_size = 0;
2572     ops[i].op.extent.truncate_seq = 0;
2573     ops[i].indata = bl;
2574     ops[i].op.flags = op_flags;
2575     Op *o = new Op(oid, oloc, ops, flags | global_op_flags |
2576                    CEPH_OSD_FLAG_WRITE, oncommit, objver,
2577                    nullptr, parent_trace);
2578     o->mtime = mtime;
2579     o->snapc = snapc;
2580     return o;
2581   }
2582   ceph_tid_t write(
2583     const object_t& oid, const object_locator_t& oloc,
2584     uint64_t off, uint64_t len, const SnapContext& snapc,
2585     const bufferlist &bl, ceph::real_time mtime, int flags,
2586     Context *oncommit, version_t *objver = NULL,
2587     ObjectOperation *extra_ops = NULL, int op_flags = 0) {
2588     Op *o = prepare_write_op(oid, oloc, off, len, snapc, bl, mtime, flags,
2589                              oncommit, objver, extra_ops, op_flags);
2590     ceph_tid_t tid;
2591     op_submit(o, &tid);
2592     return tid;
2593   }
2594   Op *prepare_append_op(
2595     const object_t& oid, const object_locator_t& oloc,
2596     uint64_t len, const SnapContext& snapc,
2597     const bufferlist &bl, ceph::real_time mtime, int flags,
2598     Context *oncommit,
2599     version_t *objver = NULL,
2600     ObjectOperation *extra_ops = NULL) {
2601     vector<OSDOp> ops;
2602     int i = init_ops(ops, 1, extra_ops);
2603     ops[i].op.op = CEPH_OSD_OP_APPEND;
2604     ops[i].op.extent.offset = 0;
2605     ops[i].op.extent.length = len;
2606     ops[i].op.extent.truncate_size = 0;
2607     ops[i].op.extent.truncate_seq = 0;
2608     ops[i].indata = bl;
2609     Op *o = new Op(oid, oloc, ops, flags | global_op_flags |
2610                    CEPH_OSD_FLAG_WRITE, oncommit, objver);
2611     o->mtime = mtime;
2612     o->snapc = snapc;
2613     return o;
2614   }
2615   ceph_tid_t append(
2616     const object_t& oid, const object_locator_t& oloc,
2617     uint64_t len, const SnapContext& snapc,
2618     const bufferlist &bl, ceph::real_time mtime, int flags,
2619     Context *oncommit,
2620     version_t *objver = NULL,
2621     ObjectOperation *extra_ops = NULL) {
2622     Op *o = prepare_append_op(oid, oloc, len, snapc, bl, mtime, flags,
2623                               oncommit, objver, extra_ops);
2624     ceph_tid_t tid;
2625     op_submit(o, &tid);
2626     return tid;
2627   }
2628   ceph_tid_t write_trunc(const object_t& oid, const object_locator_t& oloc,
2629                          uint64_t off, uint64_t len, const SnapContext& snapc,
2630                          const bufferlist &bl, ceph::real_time mtime, int flags,
2631                          uint64_t trunc_size, __u32 trunc_seq,
2632                          Context *oncommit,
2633                          version_t *objver = NULL,
2634                          ObjectOperation *extra_ops = NULL, int op_flags = 0) {
2635     vector<OSDOp> ops;
2636     int i = init_ops(ops, 1, extra_ops);
2637     ops[i].op.op = CEPH_OSD_OP_WRITE;
2638     ops[i].op.extent.offset = off;
2639     ops[i].op.extent.length = len;
2640     ops[i].op.extent.truncate_size = trunc_size;
2641     ops[i].op.extent.truncate_seq = trunc_seq;
2642     ops[i].indata = bl;
2643     ops[i].op.flags = op_flags;
2644     Op *o = new Op(oid, oloc, ops, flags | global_op_flags |
2645                    CEPH_OSD_FLAG_WRITE, oncommit, objver);
2646     o->mtime = mtime;
2647     o->snapc = snapc;
2648     ceph_tid_t tid;
2649     op_submit(o, &tid);
2650     return tid;
2651   }
2652   Op *prepare_write_full_op(
2653     const object_t& oid, const object_locator_t& oloc,
2654     const SnapContext& snapc, const bufferlist &bl,
2655     ceph::real_time mtime, int flags,
2656     Context *oncommit, version_t *objver = NULL,
2657     ObjectOperation *extra_ops = NULL, int op_flags = 0) {
2658     vector<OSDOp> ops;
2659     int i = init_ops(ops, 1, extra_ops);
2660     ops[i].op.op = CEPH_OSD_OP_WRITEFULL;
2661     ops[i].op.extent.offset = 0;
2662     ops[i].op.extent.length = bl.length();
2663     ops[i].indata = bl;
2664     ops[i].op.flags = op_flags;
2665     Op *o = new Op(oid, oloc, ops, flags | global_op_flags |
2666                    CEPH_OSD_FLAG_WRITE, oncommit, objver);
2667     o->mtime = mtime;
2668     o->snapc = snapc;
2669     return o;
2670   }
2671   ceph_tid_t write_full(
2672     const object_t& oid, const object_locator_t& oloc,
2673     const SnapContext& snapc, const bufferlist &bl,
2674     ceph::real_time mtime, int flags,
2675     Context *oncommit, version_t *objver = NULL,
2676     ObjectOperation *extra_ops = NULL, int op_flags = 0) {
2677     Op *o = prepare_write_full_op(oid, oloc, snapc, bl, mtime, flags,
2678                                   oncommit, objver, extra_ops, op_flags);
2679     ceph_tid_t tid;
2680     op_submit(o, &tid);
2681     return tid;
2682   }
2683   Op *prepare_writesame_op(
2684     const object_t& oid, const object_locator_t& oloc,
2685     uint64_t write_len, uint64_t off,
2686     const SnapContext& snapc, const bufferlist &bl,
2687     ceph::real_time mtime, int flags,
2688     Context *oncommit, version_t *objver = NULL,
2689     ObjectOperation *extra_ops = NULL, int op_flags = 0) {
2690
2691     vector<OSDOp> ops;
2692     int i = init_ops(ops, 1, extra_ops);
2693     ops[i].op.op = CEPH_OSD_OP_WRITESAME;
2694     ops[i].op.writesame.offset = off;
2695     ops[i].op.writesame.length = write_len;
2696     ops[i].op.writesame.data_length = bl.length();
2697     ops[i].indata = bl;
2698     ops[i].op.flags = op_flags;
2699     Op *o = new Op(oid, oloc, ops, flags | global_op_flags |
2700                    CEPH_OSD_FLAG_WRITE, oncommit, objver);
2701     o->mtime = mtime;
2702     o->snapc = snapc;
2703     return o;
2704   }
2705   ceph_tid_t writesame(
2706     const object_t& oid, const object_locator_t& oloc,
2707     uint64_t write_len, uint64_t off,
2708     const SnapContext& snapc, const bufferlist &bl,
2709     ceph::real_time mtime, int flags,
2710     Context *oncommit, version_t *objver = NULL,
2711     ObjectOperation *extra_ops = NULL, int op_flags = 0) {
2712
2713     Op *o = prepare_writesame_op(oid, oloc, write_len, off, snapc, bl,
2714                                  mtime, flags, oncommit, objver,
2715                                  extra_ops, op_flags);
2716
2717     ceph_tid_t tid;
2718     op_submit(o, &tid);
2719     return tid;
2720   }
2721   ceph_tid_t trunc(const object_t& oid, const object_locator_t& oloc,
2722                    const SnapContext& snapc, ceph::real_time mtime, int flags,
2723                    uint64_t trunc_size, __u32 trunc_seq,
2724                    Context *oncommit, version_t *objver = NULL,
2725                    ObjectOperation *extra_ops = NULL) {
2726     vector<OSDOp> ops;
2727     int i = init_ops(ops, 1, extra_ops);
2728     ops[i].op.op = CEPH_OSD_OP_TRUNCATE;
2729     ops[i].op.extent.offset = trunc_size;
2730     ops[i].op.extent.truncate_size = trunc_size;
2731     ops[i].op.extent.truncate_seq = trunc_seq;
2732     Op *o = new Op(oid, oloc, ops, flags | global_op_flags |
2733                    CEPH_OSD_FLAG_WRITE, oncommit, objver);
2734     o->mtime = mtime;
2735     o->snapc = snapc;
2736     ceph_tid_t tid;
2737     op_submit(o, &tid);
2738     return tid;
2739   }
2740   ceph_tid_t zero(const object_t& oid, const object_locator_t& oloc,
2741                   uint64_t off, uint64_t len, const SnapContext& snapc,
2742                   ceph::real_time mtime, int flags, Context *oncommit,
2743              version_t *objver = NULL, ObjectOperation *extra_ops = NULL) {
2744     vector<OSDOp> ops;
2745     int i = init_ops(ops, 1, extra_ops);
2746     ops[i].op.op = CEPH_OSD_OP_ZERO;
2747     ops[i].op.extent.offset = off;
2748     ops[i].op.extent.length = len;
2749     Op *o = new Op(oid, oloc, ops, flags | global_op_flags |
2750                    CEPH_OSD_FLAG_WRITE, oncommit, objver);
2751     o->mtime = mtime;
2752     o->snapc = snapc;
2753     ceph_tid_t tid;
2754     op_submit(o, &tid);
2755     return tid;
2756   }
2757   ceph_tid_t rollback_object(const object_t& oid, const object_locator_t& oloc,
2758                              const SnapContext& snapc, snapid_t snapid,
2759                              ceph::real_time mtime, Context *oncommit,
2760                              version_t *objver = NULL,
2761                              ObjectOperation *extra_ops = NULL) {
2762     vector<OSDOp> ops;
2763     int i = init_ops(ops, 1, extra_ops);
2764     ops[i].op.op = CEPH_OSD_OP_ROLLBACK;
2765     ops[i].op.snap.snapid = snapid;
2766     Op *o = new Op(oid, oloc, ops, CEPH_OSD_FLAG_WRITE, oncommit, objver);
2767     o->mtime = mtime;
2768     o->snapc = snapc;
2769     ceph_tid_t tid;
2770     op_submit(o, &tid);
2771     return tid;
2772   }
2773   ceph_tid_t create(const object_t& oid, const object_locator_t& oloc,
2774                     const SnapContext& snapc, ceph::real_time mtime, int global_flags,
2775                     int create_flags, Context *oncommit,
2776                     version_t *objver = NULL,
2777                     ObjectOperation *extra_ops = NULL) {
2778     vector<OSDOp> ops;
2779     int i = init_ops(ops, 1, extra_ops);
2780     ops[i].op.op = CEPH_OSD_OP_CREATE;
2781     ops[i].op.flags = create_flags;
2782     Op *o = new Op(oid, oloc, ops, global_flags | global_op_flags |
2783                    CEPH_OSD_FLAG_WRITE, oncommit, objver);
2784     o->mtime = mtime;
2785     o->snapc = snapc;
2786     ceph_tid_t tid;
2787     op_submit(o, &tid);
2788     return tid;
2789   }
2790   Op *prepare_remove_op(
2791     const object_t& oid, const object_locator_t& oloc,
2792     const SnapContext& snapc, ceph::real_time mtime, int flags,
2793     Context *oncommit,
2794     version_t *objver = NULL, ObjectOperation *extra_ops = NULL) {
2795     vector<OSDOp> ops;
2796     int i = init_ops(ops, 1, extra_ops);
2797     ops[i].op.op = CEPH_OSD_OP_DELETE;
2798     Op *o = new Op(oid, oloc, ops, flags | global_op_flags |
2799                    CEPH_OSD_FLAG_WRITE, oncommit, objver);
2800     o->mtime = mtime;
2801     o->snapc = snapc;
2802     return o;
2803   }
2804   ceph_tid_t remove(
2805     const object_t& oid, const object_locator_t& oloc,
2806     const SnapContext& snapc, ceph::real_time mtime, int flags,
2807     Context *oncommit,
2808     version_t *objver = NULL, ObjectOperation *extra_ops = NULL) {
2809     Op *o = prepare_remove_op(oid, oloc, snapc, mtime, flags,
2810                               oncommit, objver, extra_ops);
2811     ceph_tid_t tid;
2812     op_submit(o, &tid);
2813     return tid;
2814   }
2815
2816   ceph_tid_t setxattr(const object_t& oid, const object_locator_t& oloc,
2817               const char *name, const SnapContext& snapc, const bufferlist &bl,
2818               ceph::real_time mtime, int flags,
2819               Context *oncommit,
2820               version_t *objver = NULL, ObjectOperation *extra_ops = NULL) {
2821     vector<OSDOp> ops;
2822     int i = init_ops(ops, 1, extra_ops);
2823     ops[i].op.op = CEPH_OSD_OP_SETXATTR;
2824     ops[i].op.xattr.name_len = (name ? strlen(name) : 0);
2825     ops[i].op.xattr.value_len = bl.length();
2826     if (name)
2827       ops[i].indata.append(name);
2828     ops[i].indata.append(bl);
2829     Op *o = new Op(oid, oloc, ops, flags | global_op_flags |
2830                    CEPH_OSD_FLAG_WRITE, oncommit, objver);
2831     o->mtime = mtime;
2832     o->snapc = snapc;
2833     ceph_tid_t tid;
2834     op_submit(o, &tid);
2835     return tid;
2836   }
2837   ceph_tid_t removexattr(const object_t& oid, const object_locator_t& oloc,
2838               const char *name, const SnapContext& snapc,
2839               ceph::real_time mtime, int flags,
2840               Context *oncommit,
2841               version_t *objver = NULL, ObjectOperation *extra_ops = NULL) {
2842     vector<OSDOp> ops;
2843     int i = init_ops(ops, 1, extra_ops);
2844     ops[i].op.op = CEPH_OSD_OP_RMXATTR;
2845     ops[i].op.xattr.name_len = (name ? strlen(name) : 0);
2846     ops[i].op.xattr.value_len = 0;
2847     if (name)
2848       ops[i].indata.append(name);
2849     Op *o = new Op(oid, oloc, ops, flags | global_op_flags |
2850                    CEPH_OSD_FLAG_WRITE, oncommit, objver);
2851     o->mtime = mtime;
2852     o->snapc = snapc;
2853     ceph_tid_t tid;
2854     op_submit(o, &tid);
2855     return tid;
2856   }
2857
2858   void list_nobjects(NListContext *p, Context *onfinish);
2859   uint32_t list_nobjects_seek(NListContext *p, uint32_t pos);
2860   uint32_t list_nobjects_seek(NListContext *list_context, const hobject_t& c);
2861   void list_nobjects_get_cursor(NListContext *list_context, hobject_t *c);
2862
2863   hobject_t enumerate_objects_begin();
2864   hobject_t enumerate_objects_end();
2865   //hobject_t enumerate_objects_begin(int n, int m);
2866   void enumerate_objects(
2867     int64_t pool_id,
2868     const std::string &ns,
2869     const hobject_t &start,
2870     const hobject_t &end,
2871     const uint32_t max,
2872     const bufferlist &filter_bl,
2873     std::list<librados::ListObjectImpl> *result,
2874     hobject_t *next,
2875     Context *on_finish);
2876
2877   void _enumerate_reply(
2878       bufferlist &bl,
2879       int r,
2880       const hobject_t &end,
2881       const int64_t pool_id,
2882       int budget,
2883       epoch_t reply_epoch,
2884       std::list<librados::ListObjectImpl> *result,
2885       hobject_t *next,
2886       Context *on_finish);
2887   friend class C_EnumerateReply;
2888
2889   // -------------------------
2890   // pool ops
2891 private:
2892   void pool_op_submit(PoolOp *op);
2893   void _pool_op_submit(PoolOp *op);
2894   void _finish_pool_op(PoolOp *op, int r);
2895   void _do_delete_pool(int64_t pool, Context *onfinish);
2896 public:
2897   int create_pool_snap(int64_t pool, string& snapName, Context *onfinish);
2898   int allocate_selfmanaged_snap(int64_t pool, snapid_t *psnapid,
2899                                 Context *onfinish);
2900   int delete_pool_snap(int64_t pool, string& snapName, Context *onfinish);
2901   int delete_selfmanaged_snap(int64_t pool, snapid_t snap, Context *onfinish);
2902
2903   int create_pool(string& name, Context *onfinish, uint64_t auid=0,
2904                   int crush_rule=-1);
2905   int delete_pool(int64_t pool, Context *onfinish);
2906   int delete_pool(const string& name, Context *onfinish);
2907   int change_pool_auid(int64_t pool, Context *onfinish, uint64_t auid);
2908
2909   void handle_pool_op_reply(MPoolOpReply *m);
2910   int pool_op_cancel(ceph_tid_t tid, int r);
2911
2912   // --------------------------
2913   // pool stats
2914 private:
2915   void _poolstat_submit(PoolStatOp *op);
2916 public:
2917   void handle_get_pool_stats_reply(MGetPoolStatsReply *m);
2918   void get_pool_stats(list<string>& pools, map<string,pool_stat_t> *result,
2919                       Context *onfinish);
2920   int pool_stat_op_cancel(ceph_tid_t tid, int r);
2921   void _finish_pool_stat_op(PoolStatOp *op, int r);
2922
2923   // ---------------------------
2924   // df stats
2925 private:
2926   void _fs_stats_submit(StatfsOp *op);
2927 public:
2928   void handle_fs_stats_reply(MStatfsReply *m);
2929   void get_fs_stats(struct ceph_statfs& result, Context *onfinish);
2930   int statfs_op_cancel(ceph_tid_t tid, int r);
2931   void _finish_statfs_op(StatfsOp *op, int r);
2932
2933   // ---------------------------
2934   // some scatter/gather hackery
2935
2936   void _sg_read_finish(vector<ObjectExtent>& extents,
2937                        vector<bufferlist>& resultbl,
2938                        bufferlist *bl, Context *onfinish);
2939
2940   struct C_SGRead : public Context {
2941     Objecter *objecter;
2942     vector<ObjectExtent> extents;
2943     vector<bufferlist> resultbl;
2944     bufferlist *bl;
2945     Context *onfinish;
2946     C_SGRead(Objecter *ob,
2947              vector<ObjectExtent>& e, vector<bufferlist>& r, bufferlist *b,
2948              Context *c) :
2949       objecter(ob), bl(b), onfinish(c) {
2950       extents.swap(e);
2951       resultbl.swap(r);
2952     }
2953     void finish(int r) override {
2954       objecter->_sg_read_finish(extents, resultbl, bl, onfinish);
2955     }
2956   };
2957
2958   void sg_read_trunc(vector<ObjectExtent>& extents, snapid_t snap,
2959                      bufferlist *bl, int flags, uint64_t trunc_size,
2960                      __u32 trunc_seq, Context *onfinish, int op_flags = 0) {
2961     if (extents.size() == 1) {
2962       read_trunc(extents[0].oid, extents[0].oloc, extents[0].offset,
2963                  extents[0].length, snap, bl, flags, extents[0].truncate_size,
2964                  trunc_seq, onfinish, 0, 0, op_flags);
2965     } else {
2966       C_GatherBuilder gather(cct);
2967       vector<bufferlist> resultbl(extents.size());
2968       int i=0;
2969       for (vector<ObjectExtent>::iterator p = extents.begin();
2970            p != extents.end();
2971            ++p) {
2972         read_trunc(p->oid, p->oloc, p->offset, p->length, snap, &resultbl[i++],
2973                    flags, p->truncate_size, trunc_seq, gather.new_sub(),
2974                    0, 0, op_flags);
2975       }
2976       gather.set_finisher(new C_SGRead(this, extents, resultbl, bl, onfinish));
2977       gather.activate();
2978     }
2979   }
2980
2981   void sg_read(vector<ObjectExtent>& extents, snapid_t snap, bufferlist *bl,
2982                int flags, Context *onfinish, int op_flags = 0) {
2983     sg_read_trunc(extents, snap, bl, flags, 0, 0, onfinish, op_flags);
2984   }
2985
2986   void sg_write_trunc(vector<ObjectExtent>& extents, const SnapContext& snapc,
2987                       const bufferlist& bl, ceph::real_time mtime, int flags,
2988                       uint64_t trunc_size, __u32 trunc_seq,
2989                       Context *oncommit, int op_flags = 0) {
2990     if (extents.size() == 1) {
2991       write_trunc(extents[0].oid, extents[0].oloc, extents[0].offset,
2992                   extents[0].length, snapc, bl, mtime, flags,
2993                   extents[0].truncate_size, trunc_seq, oncommit,
2994                   0, 0, op_flags);
2995     } else {
2996       C_GatherBuilder gcom(cct, oncommit);
2997       for (vector<ObjectExtent>::iterator p = extents.begin();
2998            p != extents.end();
2999            ++p) {
3000         bufferlist cur;
3001         for (vector<pair<uint64_t,uint64_t> >::iterator bit
3002                = p->buffer_extents.begin();
3003              bit != p->buffer_extents.end();
3004              ++bit)
3005           bl.copy(bit->first, bit->second, cur);
3006         assert(cur.length() == p->length);
3007         write_trunc(p->oid, p->oloc, p->offset, p->length,
3008               snapc, cur, mtime, flags, p->truncate_size, trunc_seq,
3009               oncommit ? gcom.new_sub():0,
3010               0, 0, op_flags);
3011       }
3012       gcom.activate();
3013     }
3014   }
3015
3016   void sg_write(vector<ObjectExtent>& extents, const SnapContext& snapc,
3017                 const bufferlist& bl, ceph::real_time mtime, int flags,
3018                 Context *oncommit, int op_flags = 0) {
3019     sg_write_trunc(extents, snapc, bl, mtime, flags, 0, 0, oncommit,
3020                    op_flags);
3021   }
3022
3023   void ms_handle_connect(Connection *con) override;
3024   bool ms_handle_reset(Connection *con) override;
3025   void ms_handle_remote_reset(Connection *con) override;
3026   bool ms_handle_refused(Connection *con) override;
3027   bool ms_get_authorizer(int dest_type,
3028                          AuthAuthorizer **authorizer,
3029                          bool force_new) override;
3030
3031   void blacklist_self(bool set);
3032
3033 private:
3034   epoch_t epoch_barrier;
3035   bool retry_writes_after_first_reply;
3036 public:
3037   void set_epoch_barrier(epoch_t epoch);
3038
3039   PerfCounters *get_logger() {
3040     return logger;
3041   }
3042 };
3043
3044 #endif