ceph/src/rgw/rgw_file.cc

   1 // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
   2 // vim: ts=8 sw=2 smarttab
   3
   4 #include "include/compat.h"
   5 #include "include/rados/rgw_file.h"
   6
   7 #include <sys/types.h>
   8 #include <sys/stat.h>
   9
  10 #include "rgw_lib.h"
  11 #include "rgw_rados.h"
  12 #include "rgw_resolve.h"
  13 #include "rgw_op.h"
  14 #include "rgw_rest.h"
  15 #include "rgw_acl.h"
  16 #include "rgw_acl_s3.h"
  17 #include "rgw_frontend.h"
  18 #include "rgw_request.h"
  19 #include "rgw_process.h"
  20 #include "rgw_rest_user.h"
  21 #include "rgw_rest_s3.h"
  22 #include "rgw_os_lib.h"
  23 #include "rgw_auth_s3.h"
  24 #include "rgw_user.h"
  25 #include "rgw_bucket.h"
  26
  27 #include "rgw_file.h"
  28 #include "rgw_lib_frontend.h"
  29
  30 #include <atomic>
  31
  32 #define dout_subsys ceph_subsys_rgw
  33
  34 using namespace rgw;
  35
  36 namespace rgw {
  37
  38   extern RGWLib rgwlib;
  39
  40   const string RGWFileHandle::root_name = "/";
  41
  42   std::atomic<uint32_t> RGWLibFS::fs_inst_counter;
  43
  44   uint32_t RGWLibFS::write_completion_interval_s = 10;
  45
  46   ceph::timer<ceph::mono_clock> RGWLibFS::write_timer{
  47     ceph::construct_suspended};
  48
  49   inline int valid_fs_bucket_name(const string& name) {
  50     int rc = valid_s3_bucket_name(name, false /* relaxed */);
  51     if (rc != 0) {
  52       if (name.size() > 255)
  53         return -ENAMETOOLONG;
  54       return -EINVAL;
  55     }
  56     return 0;
  57   }
  58
  59   inline int valid_fs_object_name(const string& name) {
  60     int rc = valid_s3_object_name(name);
  61     if (rc != 0) {
  62       if (name.size() > 1024)
  63         return -ENAMETOOLONG;
  64       return -EINVAL;
  65     }
  66     return 0;
  67   }
  68
  69   LookupFHResult RGWLibFS::stat_bucket(RGWFileHandle* parent, const char *path,
  70                                        RGWLibFS::BucketStats& bs,
  71                                        uint32_t flags)
  72   {
  73     LookupFHResult fhr{nullptr, 0};
  74     std::string bucket_name{path};
  75     RGWStatBucketRequest req(cct, get_user(), bucket_name, bs);
  76
  77     int rc = rgwlib.get_fe()->execute_req(&req);
  78     if ((rc == 0) &&
  79         (req.get_ret() == 0) &&
  80         (req.matched())) {
  81       fhr = lookup_fh(parent, path,
  82                       (flags & RGWFileHandle::FLAG_LOCKED)|
  83                       RGWFileHandle::FLAG_CREATE|
  84                       RGWFileHandle::FLAG_BUCKET);
  85       if (get<0>(fhr)) {
  86         RGWFileHandle* rgw_fh = get<0>(fhr);
  87         if (! (flags & RGWFileHandle::FLAG_LOCKED)) {
  88           rgw_fh->mtx.lock();
  89         }
  90         rgw_fh->set_times(req.get_ctime());
  91         /* restore attributes */
  92         auto ux_key = req.get_attr(RGW_ATTR_UNIX_KEY1);
  93         auto ux_attrs = req.get_attr(RGW_ATTR_UNIX1);
  94         if (ux_key && ux_attrs) {
  95           bool old_key = rgw_fh->decode_attrs(ux_key, ux_attrs);
  96           if (old_key) {
  97             update_fhk(rgw_fh);
  98           }
  99         }
 100         if (! (flags & RGWFileHandle::FLAG_LOCKED)) {
 101           rgw_fh->mtx.unlock();
 102         }
 103       }
 104     }
 105     return fhr;
 106   }
 107
 108   LookupFHResult RGWLibFS::stat_leaf(RGWFileHandle* parent,
 109                                      const char *path,
 110                                      enum rgw_fh_type type,
 111                                      uint32_t flags)
 112   {
 113     /* find either-of <object_name>, <object_name/>, only one of
 114      * which should exist;  atomicity? */
 115     using std::get;
 116
 117     LookupFHResult fhr{nullptr, 0};
 118
 119     /* XXX the need for two round-trip operations to identify file or
 120      * directory leaf objects is unecessary--the current proposed
 121      * mechanism to avoid this is to store leaf object names with an
 122      * object locator w/o trailing slash */
 123
 124     std::string obj_path = parent->format_child_name(path, false);
 125
 126     for (auto ix : { 0, 1, 2 }) {
 127       switch (ix) {
 128       case 0:
 129       {
 130         /* type hint */
 131         if (type == RGW_FS_TYPE_DIRECTORY)
 132           continue;
 133
 134         RGWStatObjRequest req(cct, get_user(),
 135                               parent->bucket_name(), obj_path,
 136                               RGWStatObjRequest::FLAG_NONE);
 137         int rc = rgwlib.get_fe()->execute_req(&req);
 138         if ((rc == 0) &&
 139             (req.get_ret() == 0)) {
 140           fhr = lookup_fh(parent, path, RGWFileHandle::FLAG_CREATE);
 141           if (get<0>(fhr)) {
 142             RGWFileHandle* rgw_fh = get<0>(fhr);
 143             lock_guard guard(rgw_fh->mtx);
 144             rgw_fh->set_size(req.get_size());
 145             rgw_fh->set_times(req.get_mtime());
 146             /* restore attributes */
 147             auto ux_key = req.get_attr(RGW_ATTR_UNIX_KEY1);
 148             auto ux_attrs = req.get_attr(RGW_ATTR_UNIX1);
 149             if (ux_key && ux_attrs) {
 150               bool old_key = rgw_fh->decode_attrs(ux_key, ux_attrs);
 151               if (old_key) {
 152                 update_fhk(rgw_fh);
 153               }
 154             }
 155           }
 156           goto done;
 157         }
 158       }
 159       break;
 160       case 1:
 161       {
 162         /* try dir form */
 163         /* type hint */
 164         if (type == RGW_FS_TYPE_FILE)
 165           continue;
 166
 167         obj_path += "/";
 168         RGWStatObjRequest req(cct, get_user(),
 169                               parent->bucket_name(), obj_path,
 170                               RGWStatObjRequest::FLAG_NONE);
 171         int rc = rgwlib.get_fe()->execute_req(&req);
 172         if ((rc == 0) &&
 173             (req.get_ret() == 0)) {
 174           fhr = lookup_fh(parent, path, RGWFileHandle::FLAG_DIRECTORY);
 175           if (get<0>(fhr)) {
 176             RGWFileHandle* rgw_fh = get<0>(fhr);
 177             lock_guard guard(rgw_fh->mtx);
 178             rgw_fh->set_size(req.get_size());
 179             rgw_fh->set_times(req.get_mtime());
 180             /* restore attributes */
 181             auto ux_key = req.get_attr(RGW_ATTR_UNIX_KEY1);
 182             auto ux_attrs = req.get_attr(RGW_ATTR_UNIX1);
 183             if (ux_key && ux_attrs) {
 184               bool old_key = rgw_fh->decode_attrs(ux_key, ux_attrs);
 185               if (old_key) {
 186                 update_fhk(rgw_fh);
 187               }
 188             }
 189           }
 190           goto done;
 191         }
 192       }
 193       break;
 194       case 2:
 195       {
 196         std::string object_name{path};
 197         RGWStatLeafRequest req(cct, get_user(), parent, object_name);
 198         int rc = rgwlib.get_fe()->execute_req(&req);
 199         if ((rc == 0) &&
 200             (req.get_ret() == 0)) {
 201           if (req.matched) {
 202             /* we need rgw object's key name equal to file name, if
 203              * not return NULL */
 204             if ((flags & RGWFileHandle::FLAG_EXACT_MATCH) &&
 205                 !req.exact_matched) {
 206               lsubdout(get_context(), rgw, 15)
 207                 << __func__
 208                 << ": stat leaf not exact match file name = "
 209                 << path << dendl;
 210               goto done;
 211             }
 212             fhr = lookup_fh(parent, path,
 213                             RGWFileHandle::FLAG_CREATE|
 214                             ((req.is_dir) ?
 215                               RGWFileHandle::FLAG_DIRECTORY :
 216                               RGWFileHandle::FLAG_NONE));
 217             /* XXX we don't have an object--in general, there need not
 218              * be one (just a path segment in some other object).  In
 219              * actual leaf an object exists, but we'd need another round
 220              * trip to get attrs */
 221             if (get<0>(fhr)) {
 222               /* for now use the parent object's mtime */
 223               RGWFileHandle* rgw_fh = get<0>(fhr);
 224               lock_guard guard(rgw_fh->mtx);
 225               rgw_fh->set_mtime(parent->get_mtime());
 226             }
 227           }
 228         }
 229       }
 230       break;
 231       default:
 232         /* not reached */
 233         break;
 234       }
 235     }
 236   done:
 237     return fhr;
 238   } /* RGWLibFS::stat_leaf */
 239
 240   int RGWLibFS::read(RGWFileHandle* rgw_fh, uint64_t offset, size_t length,
 241                      size_t* bytes_read, void* buffer, uint32_t flags)
 242   {
 243     if (! rgw_fh->is_file())
 244       return -EINVAL;
 245
 246     if (rgw_fh->deleted())
 247       return -ESTALE;
 248
 249     RGWReadRequest req(get_context(), get_user(), rgw_fh, offset, length,
 250                        buffer);
 251
 252     int rc = rgwlib.get_fe()->execute_req(&req);
 253     if ((rc == 0) &&
 254         (req.get_ret() == 0)) {
 255       lock_guard(rgw_fh->mtx);
 256       rgw_fh->set_atime(real_clock::to_timespec(real_clock::now()));
 257       *bytes_read = req.nread;
 258     }
 259
 260     return rc;
 261   }
 262
 263   int RGWLibFS::unlink(RGWFileHandle* rgw_fh, const char* name, uint32_t flags)
 264   {
 265     int rc = 0;
 266     BucketStats bs;
 267     RGWFileHandle* parent = nullptr;
 268     RGWFileHandle* bkt_fh = nullptr;
 269
 270     if (unlikely(flags & RGWFileHandle::FLAG_UNLINK_THIS)) {
 271       /* LOCKED */
 272       parent = rgw_fh->get_parent();
 273     } else {
 274       /* atomicity */
 275       parent = rgw_fh;
 276       LookupFHResult fhr = lookup_fh(parent, name, RGWFileHandle::FLAG_LOCK);
 277       rgw_fh = get<0>(fhr);
 278       /* LOCKED */
 279     }
 280
 281     if (parent->is_root()) {
 282       /* a bucket may have an object storing Unix attributes, check
 283        * for and delete it */
 284       LookupFHResult fhr;
 285       fhr = stat_bucket(parent, name, bs, (rgw_fh) ?
 286                         RGWFileHandle::FLAG_LOCKED :
 287                         RGWFileHandle::FLAG_NONE);
 288       bkt_fh = get<0>(fhr);
 289       if (unlikely(! bkt_fh)) {
 290         /* implies !rgw_fh, so also !LOCKED */
 291         return -ENOENT;
 292       }
 293
 294       if (bs.num_entries > 1) {
 295         unref(bkt_fh); /* return stat_bucket ref */
 296         if (likely(!! rgw_fh)) { /* return lock and ref from
 297                                   * lookup_fh (or caller in the
 298                                   * special case of
 299                                   * RGWFileHandle::FLAG_UNLINK_THIS) */
 300           rgw_fh->mtx.unlock();
 301           unref(rgw_fh);
 302         }
 303         return -ENOTEMPTY;
 304       } else {
 305         /* delete object w/key "<bucket>/" (uxattrs), if any */
 306         string oname{"/"};
 307         RGWDeleteObjRequest req(cct, get_user(), bkt_fh->bucket_name(), oname);
 308         rc = rgwlib.get_fe()->execute_req(&req);
 309         /* don't care if ENOENT */
 310         unref(bkt_fh);
 311       }
 312
 313       string bname{name};
 314       RGWDeleteBucketRequest req(cct, get_user(), bname);
 315       rc = rgwlib.get_fe()->execute_req(&req);
 316       if (! rc) {
 317         rc = req.get_ret();
 318       }
 319     } else {
 320       /*
 321        * leaf object
 322        */
 323       if (! rgw_fh) {
 324         /* XXX for now, peform a hard lookup to deduce the type of
 325          * object to be deleted ("foo" vs. "foo/")--also, ensures
 326          * atomicity at this endpoint */
 327         struct rgw_file_handle *fh;
 328         rc = rgw_lookup(get_fs(), parent->get_fh(), name, &fh,
 329                         RGW_LOOKUP_FLAG_NONE);
 330         if (!! rc)
 331           return rc;
 332
 333         /* rgw_fh ref+ */
 334         rgw_fh = get_rgwfh(fh);
 335         rgw_fh->mtx.lock(); /* LOCKED */
 336       }
 337
 338       std::string oname = rgw_fh->relative_object_name();
 339       if (rgw_fh->is_dir()) {
 340         /* for the duration of our cache timer, trust positive
 341          * child cache */
 342         if (rgw_fh->has_children()) {
 343           rgw_fh->mtx.unlock();
 344           unref(rgw_fh);
 345           return(-ENOTEMPTY);
 346         }
 347         oname += "/";
 348       }
 349       RGWDeleteObjRequest req(cct, get_user(), parent->bucket_name(),
 350                               oname);
 351       rc = rgwlib.get_fe()->execute_req(&req);
 352       if (! rc) {
 353         rc = req.get_ret();
 354       }
 355     }
 356
 357     /* ENOENT when raced with other s3 gateway */
 358     if (! rc || rc == -ENOENT) {
 359       rgw_fh->flags |= RGWFileHandle::FLAG_DELETED;
 360       fh_cache.remove(rgw_fh->fh.fh_hk.object, rgw_fh,
 361                       RGWFileHandle::FHCache::FLAG_LOCK);
 362     }
 363
 364     if (! rc) {
 365       real_time t = real_clock::now();
 366       parent->set_mtime(real_clock::to_timespec(t));
 367       parent->set_ctime(real_clock::to_timespec(t));
 368     }
 369
 370     rgw_fh->mtx.unlock();
 371     unref(rgw_fh);
 372
 373     return rc;
 374   } /* RGWLibFS::unlink */
 375
 376   int RGWLibFS::rename(RGWFileHandle* src_fh, RGWFileHandle* dst_fh,
 377                        const char *_src_name, const char *_dst_name)
 378
 379   {
 380     /* XXX initial implementation: try-copy, and delete if copy
 381      * succeeds */
 382     int rc = -EINVAL;
 383
 384     real_time t;
 385
 386     std::string src_name{_src_name};
 387     std::string dst_name{_dst_name};
 388
 389     /* atomicity */
 390     LookupFHResult fhr = lookup_fh(src_fh, _src_name, RGWFileHandle::FLAG_LOCK);
 391     RGWFileHandle* rgw_fh = get<0>(fhr);
 392
 393     /* should not happen */
 394     if (! rgw_fh) {
 395       ldout(get_context(), 0) << __func__
 396                        << " BUG no such src renaming path="
 397                        << src_name
 398                        << dendl;
 399       goto out;
 400     }
 401
 402     /* forbid renaming of directories (unreasonable at scale) */
 403     if (rgw_fh->is_dir()) {
 404       ldout(get_context(), 12) << __func__
 405                         << " rejecting attempt to rename directory path="
 406                         << rgw_fh->full_object_name()
 407                         << dendl;
 408       rc = -EPERM;
 409       goto unlock;
 410     }
 411
 412     /* forbid renaming open files (violates intent, for now) */
 413     if (rgw_fh->is_open()) {
 414       ldout(get_context(), 12) << __func__
 415                         << " rejecting attempt to rename open file path="
 416                         << rgw_fh->full_object_name()
 417                         << dendl;
 418       rc = -EPERM;
 419       goto unlock;
 420     }
 421
 422     t = real_clock::now();
 423
 424     for (int ix : {0, 1}) {
 425       switch (ix) {
 426       case 0:
 427       {
 428         RGWCopyObjRequest req(cct, get_user(), src_fh, dst_fh, src_name,
 429                               dst_name);
 430         int rc = rgwlib.get_fe()->execute_req(&req);
 431         if ((rc != 0) ||
 432             ((rc = req.get_ret()) != 0)) {
 433           ldout(get_context(), 1)
 434             << __func__
 435             << " rename step 0 failed src="
 436             << src_fh->full_object_name() << " " << src_name
 437             << " dst=" << dst_fh->full_object_name()
 438             << " " << dst_name
 439             << "rc " << rc
 440             << dendl;
 441           goto unlock;
 442         }
 443         ldout(get_context(), 12)
 444           << __func__
 445           << " rename step 0 success src="
 446           << src_fh->full_object_name() << " " << src_name
 447           << " dst=" << dst_fh->full_object_name()
 448           << " " << dst_name
 449           << " rc " << rc
 450           << dendl;
 451         /* update dst change id */
 452         dst_fh->set_times(t);
 453       }
 454       break;
 455       case 1:
 456       {
 457         rc = this->unlink(rgw_fh /* LOCKED */, _src_name,
 458                           RGWFileHandle::FLAG_UNLINK_THIS);
 459         /* !LOCKED, -ref */
 460         if (! rc) {
 461           ldout(get_context(), 12)
 462             << __func__
 463             << " rename step 1 success src="
 464             << src_fh->full_object_name() << " " << src_name
 465             << " dst=" << dst_fh->full_object_name()
 466             << " " << dst_name
 467             << " rc " << rc
 468             << dendl;
 469           /* update src change id */
 470           src_fh->set_times(t);
 471         } else {
 472           ldout(get_context(), 1)
 473             << __func__
 474             << " rename step 1 failed src="
 475             << src_fh->full_object_name() << " " << src_name
 476             << " dst=" << dst_fh->full_object_name()
 477             << " " << dst_name
 478             << " rc " << rc
 479             << dendl;
 480         }
 481       }
 482       goto out;
 483       default:
 484         abort();
 485       } /* switch */
 486     } /* ix */
 487   unlock:
 488     rgw_fh->mtx.unlock(); /* !LOCKED */
 489     unref(rgw_fh); /* -ref */
 490
 491   out:
 492     return rc;
 493   } /* RGWLibFS::rename */
 494
 495   MkObjResult RGWLibFS::mkdir(RGWFileHandle* parent, const char *name,
 496                               struct stat *st, uint32_t mask, uint32_t flags)
 497   {
 498     int rc, rc2;
 499     rgw_file_handle *lfh;
 500
 501     rc = rgw_lookup(get_fs(), parent->get_fh(), name, &lfh,
 502                     RGW_LOOKUP_FLAG_NONE);
 503     if (! rc) {
 504       /* conflict! */
 505       rc = rgw_fh_rele(get_fs(), lfh, RGW_FH_RELE_FLAG_NONE);
 506       return MkObjResult{nullptr, -EEXIST};
 507     }
 508
 509     MkObjResult mkr{nullptr, -EINVAL};
 510     LookupFHResult fhr;
 511     RGWFileHandle* rgw_fh = nullptr;
 512     buffer::list ux_key, ux_attrs;
 513
 514     fhr = lookup_fh(parent, name,
 515                     RGWFileHandle::FLAG_CREATE|
 516                     RGWFileHandle::FLAG_DIRECTORY|
 517                     RGWFileHandle::FLAG_LOCK);
 518     rgw_fh = get<0>(fhr);
 519     if (rgw_fh) {
 520       rgw_fh->create_stat(st, mask);
 521       rgw_fh->set_times(real_clock::now());
 522       /* save attrs */
 523       rgw_fh->encode_attrs(ux_key, ux_attrs);
 524       if (st)
 525         rgw_fh->stat(st);
 526       get<0>(mkr) = rgw_fh;
 527     } else {
 528       get<1>(mkr) = -EIO;
 529       return mkr;
 530     }
 531
 532     if (parent->is_root()) {
 533       /* bucket */
 534       string bname{name};
 535       /* enforce S3 name restrictions */
 536       rc = valid_fs_bucket_name(bname);
 537       if (rc != 0) {
 538         rgw_fh->flags |= RGWFileHandle::FLAG_DELETED;
 539         fh_cache.remove(rgw_fh->fh.fh_hk.object, rgw_fh,
 540                         RGWFileHandle::FHCache::FLAG_LOCK);
 541         rgw_fh->mtx.unlock();
 542         unref(rgw_fh);
 543         get<0>(mkr) = nullptr;
 544         get<1>(mkr) = rc;
 545         return mkr;
 546       }
 547
 548       RGWCreateBucketRequest req(get_context(), get_user(), bname);
 549
 550       /* save attrs */
 551       req.emplace_attr(RGW_ATTR_UNIX_KEY1, std::move(ux_key));
 552       req.emplace_attr(RGW_ATTR_UNIX1, std::move(ux_attrs));
 553
 554       rc = rgwlib.get_fe()->execute_req(&req);
 555       rc2 = req.get_ret();
 556     } else {
 557       /* create an object representing the directory */
 558       buffer::list bl;
 559       string dir_name = parent->format_child_name(name, true);
 560
 561       /* need valid S3 name (characters, length <= 1024, etc) */
 562       rc = valid_fs_object_name(dir_name);
 563       if (rc != 0) {
 564         rgw_fh->flags |= RGWFileHandle::FLAG_DELETED;
 565         fh_cache.remove(rgw_fh->fh.fh_hk.object, rgw_fh,
 566                         RGWFileHandle::FHCache::FLAG_LOCK);
 567         rgw_fh->mtx.unlock();
 568         unref(rgw_fh);
 569         get<0>(mkr) = nullptr;
 570         get<1>(mkr) = rc;
 571         return mkr;
 572       }
 573
 574       RGWPutObjRequest req(get_context(), get_user(), parent->bucket_name(),
 575                           dir_name, bl);
 576
 577       /* save attrs */
 578       req.emplace_attr(RGW_ATTR_UNIX_KEY1, std::move(ux_key));
 579       req.emplace_attr(RGW_ATTR_UNIX1, std::move(ux_attrs));
 580
 581       rc = rgwlib.get_fe()->execute_req(&req);
 582       rc2 = req.get_ret();
 583     }
 584
 585     if (! ((rc == 0) &&
 586            (rc2 == 0))) {
 587       /* op failed */
 588       rgw_fh->flags |= RGWFileHandle::FLAG_DELETED;
 589       rgw_fh->mtx.unlock(); /* !LOCKED */
 590       unref(rgw_fh);
 591       get<0>(mkr) = nullptr;
 592       /* fixup rc */
 593       if (!rc)
 594         rc = rc2;
 595     } else {
 596       real_time t = real_clock::now();
 597       parent->set_mtime(real_clock::to_timespec(t));
 598       parent->set_ctime(real_clock::to_timespec(t));
 599       rgw_fh->mtx.unlock(); /* !LOCKED */
 600     }
 601
 602     get<1>(mkr) = rc;
 603
 604     return mkr;
 605   } /* RGWLibFS::mkdir */
 606
 607   MkObjResult RGWLibFS::create(RGWFileHandle* parent, const char *name,
 608                               struct stat *st, uint32_t mask, uint32_t flags)
 609   {
 610     int rc, rc2;
 611
 612     using std::get;
 613
 614     rgw_file_handle *lfh;
 615     rc = rgw_lookup(get_fs(), parent->get_fh(), name, &lfh,
 616                     RGW_LOOKUP_FLAG_NONE);
 617     if (! rc) {
 618       /* conflict! */
 619       rc = rgw_fh_rele(get_fs(), lfh, RGW_FH_RELE_FLAG_NONE);
 620       return MkObjResult{nullptr, -EEXIST};
 621     }
 622
 623     /* expand and check name */
 624     std::string obj_name = parent->format_child_name(name, false);
 625     rc = valid_fs_object_name(obj_name);
 626     if (rc != 0) {
 627       return MkObjResult{nullptr, rc};
 628     }
 629
 630     /* create it */
 631     buffer::list bl;
 632     RGWPutObjRequest req(cct, get_user(), parent->bucket_name(), obj_name, bl);
 633     MkObjResult mkr{nullptr, -EINVAL};
 634
 635     rc = rgwlib.get_fe()->execute_req(&req);
 636     rc2 = req.get_ret();
 637
 638     if ((rc == 0) &&
 639         (rc2 == 0)) {
 640       /* XXX atomicity */
 641       LookupFHResult fhr = lookup_fh(parent, name, RGWFileHandle::FLAG_CREATE |
 642                                                    RGWFileHandle::FLAG_LOCK);
 643       RGWFileHandle* rgw_fh = get<0>(fhr);
 644       if (rgw_fh) {
 645         if (get<1>(fhr) & RGWFileHandle::FLAG_CREATE) {
 646           /* fill in stat data */
 647           real_time t = real_clock::now();
 648           rgw_fh->create_stat(st, mask);
 649           rgw_fh->set_times(t);
 650
 651           parent->set_mtime(real_clock::to_timespec(t));
 652           parent->set_ctime(real_clock::to_timespec(t));
 653         }
 654         if (st)
 655           (void) rgw_fh->stat(st);
 656         get<0>(mkr) = rgw_fh;
 657         rgw_fh->mtx.unlock();
 658       } else
 659         rc = -EIO;
 660     }
 661
 662     get<1>(mkr) = rc;
 663
 664     return mkr;
 665   } /* RGWLibFS::create */
 666
 667   int RGWLibFS::getattr(RGWFileHandle* rgw_fh, struct stat* st)
 668   {
 669     switch(rgw_fh->fh.fh_type) {
 670     case RGW_FS_TYPE_FILE:
 671     {
 672       if (rgw_fh->deleted())
 673         return -ESTALE;
 674     }
 675     break;
 676     default:
 677       break;
 678     };
 679
 680     return rgw_fh->stat(st);
 681   } /* RGWLibFS::getattr */
 682
 683   int RGWLibFS::setattr(RGWFileHandle* rgw_fh, struct stat* st, uint32_t mask,
 684                         uint32_t flags)
 685   {
 686     int rc, rc2;
 687     buffer::list ux_key, ux_attrs;
 688
 689     lock_guard guard(rgw_fh->mtx);
 690
 691     switch(rgw_fh->fh.fh_type) {
 692     case RGW_FS_TYPE_FILE:
 693     {
 694       if (rgw_fh->deleted())
 695         return -ESTALE;
 696     }
 697     break;
 698     default:
 699       break;
 700     };
 701
 702     string obj_name{rgw_fh->relative_object_name()};
 703
 704     if (rgw_fh->is_dir() &&
 705         (likely(! rgw_fh->is_bucket()))) {
 706       obj_name += "/";
 707     }
 708
 709     RGWSetAttrsRequest req(cct, get_user(), rgw_fh->bucket_name(), obj_name);
 710
 711     rgw_fh->create_stat(st, mask);
 712     rgw_fh->encode_attrs(ux_key, ux_attrs);
 713
 714     /* save attrs */
 715     req.emplace_attr(RGW_ATTR_UNIX_KEY1, std::move(ux_key));
 716     req.emplace_attr(RGW_ATTR_UNIX1, std::move(ux_attrs));
 717
 718     rc = rgwlib.get_fe()->execute_req(&req);
 719     rc2 = req.get_ret();
 720
 721     if (rc == -ENOENT) {
 722       /* special case:  materialize placeholder dir */
 723       buffer::list bl;
 724       RGWPutObjRequest req(get_context(), get_user(), rgw_fh->bucket_name(),
 725                            obj_name, bl);
 726
 727       rgw_fh->encode_attrs(ux_key, ux_attrs); /* because std::moved */
 728
 729       /* save attrs */
 730       req.emplace_attr(RGW_ATTR_UNIX_KEY1, std::move(ux_key));
 731       req.emplace_attr(RGW_ATTR_UNIX1, std::move(ux_attrs));
 732
 733       rc = rgwlib.get_fe()->execute_req(&req);
 734       rc2 = req.get_ret();
 735     }
 736
 737     if ((rc != 0) || (rc2 != 0)) {
 738       return -EIO;
 739     }
 740
 741     rgw_fh->set_ctime(real_clock::to_timespec(real_clock::now()));
 742
 743     return 0;
 744   } /* RGWLibFS::setattr */
 745
 746   /* called under rgw_fh->mtx held */
 747   void RGWLibFS::update_fhk(RGWFileHandle *rgw_fh)
 748   {
 749     int rc, rc2;
 750     string obj_name{rgw_fh->relative_object_name()};
 751     buffer::list ux_key, ux_attrs;
 752
 753     if (rgw_fh->is_dir() &&
 754         (likely(! rgw_fh->is_bucket()))) {
 755       obj_name += "/";
 756     }
 757
 758     lsubdout(get_context(), rgw, 17)
 759       << __func__
 760       << " update old versioned fhk : " << obj_name
 761       << dendl;
 762
 763     RGWSetAttrsRequest req(cct, get_user(), rgw_fh->bucket_name(), obj_name);
 764
 765     rgw_fh->encode_attrs(ux_key, ux_attrs);
 766
 767     /* update ux_key only */
 768     req.emplace_attr(RGW_ATTR_UNIX_KEY1, std::move(ux_key));
 769
 770     rc = rgwlib.get_fe()->execute_req(&req);
 771     rc2 = req.get_ret();
 772
 773     if ((rc != 0) || (rc2 != 0)) {
 774       lsubdout(get_context(), rgw, 17)
 775         << __func__
 776         << " update fhk failed : " << obj_name
 777         << dendl;
 778     }
 779   } /* RGWLibFS::update_fhk */
 780
 781   void RGWLibFS::close()
 782   {
 783     state.flags |= FLAG_CLOSED;
 784
 785     class ObjUnref
 786     {
 787       RGWLibFS* fs;
 788     public:
 789       ObjUnref(RGWLibFS* fs) : fs(fs) {}
 790       void operator()(RGWFileHandle* fh) const {
 791         lsubdout(fs->get_context(), rgw, 5)
 792           << __func__
 793           << fh->name
 794           << " before ObjUnref refs=" << fh->get_refcnt()
 795           << dendl;
 796         fs->unref(fh);
 797       }
 798     };
 799
 800     /* force cache drain, forces objects to evict */
 801     fh_cache.drain(ObjUnref(this),
 802                   RGWFileHandle::FHCache::FLAG_LOCK);
 803     rgwlib.get_fe()->get_process()->unregister_fs(this);
 804     rele();
 805   } /* RGWLibFS::close */
 806
 807   inline std::ostream& operator<<(std::ostream &os, struct timespec const &ts) {
 808       os << "<timespec: tv_sec=";
 809       os << ts.tv_sec;
 810       os << "; tv_nsec=";
 811       os << ts.tv_nsec;
 812       os << ">";
 813     return os;
 814   }
 815
 816   std::ostream& operator<<(std::ostream &os, RGWLibFS::event const &ev) {
 817     os << "<event:";
 818       switch (ev.t) {
 819       case RGWLibFS::event::type::READDIR:
 820         os << "type=READDIR;";
 821         break;
 822       default:
 823         os << "type=UNKNOWN;";
 824         break;
 825       };
 826     os << "fid=" << ev.fhk.fh_hk.bucket << ":" << ev.fhk.fh_hk.object
 827        << ";ts=" << ev.ts << ">";
 828     return os;
 829   }
 830
 831   void RGWLibFS::gc()
 832   {
 833     using std::get;
 834     using directory = RGWFileHandle::directory;
 835
 836     /* dirent invalidate timeout--basically, the upper-bound on
 837      * inconsistency with the S3 namespace */
 838     auto expire_s
 839       = get_context()->_conf->rgw_nfs_namespace_expire_secs;
 840
 841     /* max events to gc in one cycle */
 842     uint32_t max_ev = get_context()->_conf->rgw_nfs_max_gc;
 843
 844     struct timespec now, expire_ts;
 845     event_vector ve;
 846     bool stop = false;
 847     std::deque<event> &events = state.events;
 848
 849     do {
 850       (void) clock_gettime(CLOCK_MONOTONIC_COARSE, &now);
 851       lsubdout(get_context(), rgw, 15)
 852         << "GC: top of expire loop"
 853         << " now=" << now
 854         << " expire_s=" << expire_s
 855         << dendl;
 856       {
 857         lock_guard guard(state.mtx); /* LOCKED */
 858         /* just return if no events */
 859         if (events.empty()) {
 860           return;
 861         }
 862         uint32_t _max_ev =
 863           (events.size() < 500) ? max_ev : (events.size() / 4);
 864         for (uint32_t ix = 0; (ix < _max_ev) && (events.size() > 0); ++ix) {
 865           event& ev = events.front();
 866           expire_ts = ev.ts;
 867           expire_ts.tv_sec += expire_s;
 868           if (expire_ts > now) {
 869             stop = true;
 870             break;
 871           }
 872           ve.push_back(ev);
 873           events.pop_front();
 874         }
 875       } /* anon */
 876       /* !LOCKED */
 877       for (auto& ev : ve) {
 878         lsubdout(get_context(), rgw, 15)
 879           << "try-expire ev: " << ev << dendl;
 880         if (likely(ev.t == event::type::READDIR)) {
 881           RGWFileHandle* rgw_fh = lookup_handle(ev.fhk.fh_hk);
 882           lsubdout(get_context(), rgw, 15)
 883             << "ev rgw_fh: " << rgw_fh << dendl;
 884           if (rgw_fh) {
 885             RGWFileHandle::directory* d;
 886             if (unlikely(! rgw_fh->is_dir())) {
 887               lsubdout(get_context(), rgw, 0)
 888                 << __func__
 889                 << " BUG non-directory found with READDIR event "
 890                 << "(" << rgw_fh->bucket_name() << ","
 891                 << rgw_fh->object_name() << ")"
 892                 << dendl;
 893               goto rele;
 894             }
 895             /* maybe clear state */
 896             d = get<directory>(&rgw_fh->variant_type);
 897             if (d) {
 898               struct timespec ev_ts = ev.ts;
 899               lock_guard guard(rgw_fh->mtx);
 900               struct timespec d_last_readdir = d->last_readdir;
 901               if (unlikely(ev_ts < d_last_readdir)) {
 902                 /* readdir cycle in progress, don't invalidate */
 903                 lsubdout(get_context(), rgw, 15)
 904                   << "GC: delay expiration for "
 905                   << rgw_fh->object_name()
 906                   << " ev.ts=" << ev_ts
 907                   << " last_readdir=" << d_last_readdir
 908                   << dendl;
 909                 continue;
 910               } else {
 911                 lsubdout(get_context(), rgw, 15)
 912                   << "GC: expiring "
 913                   << rgw_fh->object_name()
 914                   << dendl;
 915                 rgw_fh->clear_state();
 916                 rgw_fh->invalidate();
 917               }
 918             }
 919           rele:
 920             unref(rgw_fh);
 921           } /* rgw_fh */
 922         } /* event::type::READDIR */
 923       } /* ev */
 924       ve.clear();
 925     } while (! (stop || shutdown));
 926   } /* RGWLibFS::gc */
 927
 928   std::ostream& operator<<(std::ostream &os,
 929                            RGWFileHandle const &rgw_fh)
 930   {
 931     const auto& fhk = rgw_fh.get_key();
 932     const auto& fh = const_cast<RGWFileHandle&>(rgw_fh).get_fh();
 933     os << "<RGWFileHandle:";
 934     os << "addr=" << &rgw_fh << ";";
 935     switch (fh->fh_type) {
 936     case RGW_FS_TYPE_DIRECTORY:
 937         os << "type=DIRECTORY;";
 938         break;
 939     case RGW_FS_TYPE_FILE:
 940         os << "type=FILE;";
 941         break;
 942     default:
 943         os << "type=UNKNOWN;";
 944         break;
 945       };
 946     os << "fid=" << fhk.fh_hk.bucket << ":" << fhk.fh_hk.object << ";";
 947     os << "name=" << rgw_fh.object_name() << ";";
 948     os << "refcnt=" << rgw_fh.get_refcnt() << ";";
 949     os << ">";
 950     return os;
 951   }
 952
 953   RGWFileHandle::~RGWFileHandle() {
 954     /* in the non-delete case, handle may still be in handle table */
 955     if (fh_hook.is_linked()) {
 956       fs->fh_cache.remove(fh.fh_hk.object, this, FHCache::FLAG_LOCK);
 957     }
 958     /* cond-unref parent */
 959     if (parent && (! parent->is_root())) {
 960       /* safe because if parent->unref causes its deletion,
 961        * there are a) by refcnt, no other objects/paths pointing
 962        * to it and b) by the semantics of valid iteration of
 963        * fh_lru (observed, e.g., by cohort_lru<T,...>::drain())
 964        * no unsafe iterators reaching it either--n.b., this constraint
 965        * is binding oncode which may in future attempt to e.g.,
 966        * cause the eviction of objects in LRU order */
 967       (void) get_fs()->unref(parent);
 968     }
 969   }
 970
 971   void RGWFileHandle::encode_attrs(ceph::buffer::list& ux_key1,
 972                                    ceph::buffer::list& ux_attrs1)
 973   {
 974     fh_key fhk(this->fh.fh_hk);
 975     rgw::encode(fhk, ux_key1);
 976     rgw::encode(*this, ux_attrs1);
 977   } /* RGWFileHandle::encode_attrs */
 978
 979   bool RGWFileHandle::decode_attrs(const ceph::buffer::list* ux_key1,
 980                                    const ceph::buffer::list* ux_attrs1)
 981   {
 982     bool old_key = false;
 983     fh_key fhk;
 984     auto bl_iter_key1  = const_cast<buffer::list*>(ux_key1)->begin();
 985     rgw::decode(fhk, bl_iter_key1);
 986     if (fhk.version >= 2) {
 987       assert(this->fh.fh_hk == fhk.fh_hk);
 988     } else {
 989       old_key = true;
 990     }
 991
 992     auto bl_iter_unix1 = const_cast<buffer::list*>(ux_attrs1)->begin();
 993     rgw::decode(*this, bl_iter_unix1);
 994
 995     return old_key;
 996   } /* RGWFileHandle::decode_attrs */
 997
 998   bool RGWFileHandle::reclaim() {
 999     lsubdout(fs->get_context(), rgw, 17)
1000       << __func__ << " " << *this
1001       << dendl;
1002     /* remove if still in fh_cache */
1003     if (fh_hook.is_linked()) {
1004       fs->fh_cache.remove(fh.fh_hk.object, this, FHCache::FLAG_LOCK);
1005     }
1006     return true;
1007   } /* RGWFileHandle::reclaim */
1008
1009   bool RGWFileHandle::has_children() const
1010   {
1011     if (unlikely(! is_dir()))
1012       return false;
1013
1014     RGWRMdirCheck req(fs->get_context(), fs->get_user(), this);
1015     int rc = rgwlib.get_fe()->execute_req(&req);
1016     if (! rc) {
1017       return req.valid && req.has_children;
1018     }
1019
1020     return false;
1021   }
1022
1023   int RGWFileHandle::readdir(rgw_readdir_cb rcb, void *cb_arg, uint64_t *offset,
1024                              bool *eof, uint32_t flags)
1025   {
1026     using event = RGWLibFS::event;
1027     int rc = 0;
1028     struct timespec now;
1029     CephContext* cct = fs->get_context();
1030
1031     if ((*offset == 0) &&
1032         (flags & RGW_READDIR_FLAG_DOTDOT)) {
1033       /* send '.' and '..' with their NFS-defined offsets */
1034       rcb(".", cb_arg, 1, RGW_LOOKUP_FLAG_DIR);
1035       rcb("..", cb_arg, 2, RGW_LOOKUP_FLAG_DIR);
1036     }
1037
1038     lsubdout(fs->get_context(), rgw, 15)
1039       << __func__
1040       << " offset=" << *offset
1041       << dendl;
1042
1043     directory* d = get<directory>(&variant_type);
1044     if (d) {
1045       (void) clock_gettime(CLOCK_MONOTONIC_COARSE, &now); /* !LOCKED */
1046       lock_guard guard(mtx);
1047       d->last_readdir = now;
1048     }
1049
1050     if (is_root()) {
1051       RGWListBucketsRequest req(cct, fs->get_user(), this, rcb, cb_arg,
1052                                 offset);
1053       rc = rgwlib.get_fe()->execute_req(&req);
1054       if (! rc) {
1055         (void) clock_gettime(CLOCK_MONOTONIC_COARSE, &now); /* !LOCKED */
1056         lock_guard guard(mtx);
1057         state.atime = now;
1058         if (*offset == 0)
1059           set_nlink(2);
1060         inc_nlink(req.d_count);
1061         *eof = req.eof();
1062         event ev(event::type::READDIR, get_key(), state.atime);
1063         lock_guard sguard(fs->state.mtx);
1064         fs->state.push_event(ev);
1065       }
1066     } else {
1067       RGWReaddirRequest req(cct, fs->get_user(), this, rcb, cb_arg, offset);
1068       rc = rgwlib.get_fe()->execute_req(&req);
1069       if (! rc) {
1070         (void) clock_gettime(CLOCK_MONOTONIC_COARSE, &now); /* !LOCKED */
1071         lock_guard guard(mtx);
1072         state.atime = now;
1073         if (*offset == 0)
1074           set_nlink(2);
1075         inc_nlink(req.d_count);
1076         *eof = req.eof();
1077         event ev(event::type::READDIR, get_key(), state.atime);
1078         lock_guard sguard(fs->state.mtx);
1079         fs->state.push_event(ev);
1080       }
1081     }
1082
1083     lsubdout(fs->get_context(), rgw, 15)
1084       << __func__
1085       << " final link count=" << state.nlink
1086       << dendl;
1087
1088     return rc;
1089   } /* RGWFileHandle::readdir */
1090
1091   int RGWFileHandle::write(uint64_t off, size_t len, size_t *bytes_written,
1092                            void *buffer)
1093   {
1094     using std::get;
1095     using WriteCompletion = RGWLibFS::WriteCompletion;
1096
1097     lock_guard guard(mtx);
1098
1099     int rc = 0;
1100
1101     file* f = get<file>(&variant_type);
1102     if (! f)
1103       return -EISDIR;
1104
1105     if (deleted()) {
1106       lsubdout(fs->get_context(), rgw, 5)
1107         << __func__
1108         << " write attempted on deleted object "
1109         << this->object_name()
1110         << dendl;
1111       /* zap write transaction, if any */
1112       if (f->write_req) {
1113         delete f->write_req;
1114         f->write_req = nullptr;
1115       }
1116       return -ESTALE;
1117     }
1118
1119     if (! f->write_req) {
1120       /* guard--we do not support (e.g., COW-backed) partial writes */
1121       if (off != 0) {
1122         lsubdout(fs->get_context(), rgw, 5)
1123           << __func__
1124           << " " << object_name()
1125           << " non-0 initial write position " << off
1126           << dendl;
1127         return -EIO;
1128       }
1129
1130       /* start */
1131       std::string object_name = relative_object_name();
1132       f->write_req =
1133         new RGWWriteRequest(fs->get_context(), fs->get_user(), this,
1134                             bucket_name(), object_name);
1135       rc = rgwlib.get_fe()->start_req(f->write_req);
1136       if (rc < 0) {
1137         lsubdout(fs->get_context(), rgw, 5)
1138           << __func__
1139           << this->object_name()
1140           << " write start failed " << off
1141           << " (" << rc << ")"
1142           << dendl;
1143         /* zap failed write transaction */
1144         delete f->write_req;
1145         f->write_req = nullptr;
1146         return -EIO;
1147       } else {
1148         if (stateless_open())  {
1149           /* start write timer */
1150           f->write_req->timer_id =
1151             RGWLibFS::write_timer.add_event(
1152               std::chrono::seconds(RGWLibFS::write_completion_interval_s),
1153               WriteCompletion(*this));
1154         }
1155       }
1156     }
1157
1158     buffer::list bl;
1159     /* XXXX */
1160 #if 0
1161     bl.push_back(
1162       buffer::create_static(len, static_cast<char*>(buffer)));
1163 #else
1164     bl.push_back(
1165       buffer::copy(static_cast<char*>(buffer), len));
1166 #endif
1167
1168     f->write_req->put_data(off, bl);
1169     rc = f->write_req->exec_continue();
1170
1171     if (rc == 0) {
1172       size_t min_size = off + len;
1173       if (min_size > get_size())
1174         set_size(min_size);
1175       if (stateless_open()) {
1176         /* bump write timer */
1177         RGWLibFS::write_timer.adjust_event(
1178           f->write_req->timer_id, std::chrono::seconds(10));
1179       }
1180     } else {
1181       /* continuation failed (e.g., non-contiguous write position) */
1182       lsubdout(fs->get_context(), rgw, 5)
1183         << __func__
1184         << object_name()
1185         << " failed write at position " << off
1186         << " (fails write transaction) "
1187         << dendl;
1188       /* zap failed write transaction */
1189       delete f->write_req;
1190       f->write_req = nullptr;
1191       rc = -EIO;
1192     }
1193
1194     *bytes_written = (rc == 0) ? len : 0;
1195     return rc;
1196   } /* RGWFileHandle::write */
1197
1198   int RGWFileHandle::write_finish(uint32_t flags)
1199   {
1200     unique_lock guard{mtx, std::defer_lock};
1201     int rc = 0;
1202
1203     if (! (flags & FLAG_LOCKED)) {
1204       guard.lock();
1205     }
1206
1207     file* f = get<file>(&variant_type);
1208     if (f && (f->write_req)) {
1209       lsubdout(fs->get_context(), rgw, 10)
1210         << __func__
1211         << " finishing write trans on " << object_name()
1212         << dendl;
1213       rc = rgwlib.get_fe()->finish_req(f->write_req);
1214       if (! rc) {
1215         rc = f->write_req->get_ret();
1216       }
1217       delete f->write_req;
1218       f->write_req = nullptr;
1219     }
1220
1221     return rc;
1222   } /* RGWFileHandle::write_finish */
1223
1224   int RGWFileHandle::close()
1225   {
1226     lock_guard guard(mtx);
1227
1228     int rc = write_finish(FLAG_LOCKED);
1229
1230     flags &= ~FLAG_OPEN;
1231     flags &= ~FLAG_STATELESS_OPEN;
1232
1233     return rc;
1234   } /* RGWFileHandle::close */
1235
1236   RGWFileHandle::file::~file()
1237   {
1238     delete write_req;
1239   }
1240
1241   void RGWFileHandle::clear_state()
1242   {
1243     directory* d = get<directory>(&variant_type);
1244     if (d) {
1245       state.nlink = 2;
1246       d->last_marker = rgw_obj_key{};
1247     }
1248   }
1249
1250   void RGWFileHandle::invalidate() {
1251     RGWLibFS *fs = get_fs();
1252     if (fs->invalidate_cb) {
1253       fs->invalidate_cb(fs->invalidate_arg, get_key().fh_hk);
1254     }
1255   }
1256
1257   int RGWWriteRequest::exec_start() {
1258     struct req_state* s = get_state();
1259
1260     auto compression_type =
1261       get_store()->get_zone_params().get_compression_type(
1262         s->bucket_info.placement_rule);
1263
1264     /* not obviously supportable */
1265     assert(! dlo_manifest);
1266     assert(! slo_info);
1267
1268     perfcounter->inc(l_rgw_put);
1269     op_ret = -EINVAL;
1270
1271     if (s->object.empty()) {
1272       ldout(s->cct, 0) << __func__ << " called on empty object" << dendl;
1273       goto done;
1274     }
1275
1276     op_ret = get_params();
1277     if (op_ret < 0)
1278       goto done;
1279
1280     op_ret = get_system_versioning_params(s, &olh_epoch, &version_id);
1281     if (op_ret < 0) {
1282       goto done;
1283     }
1284
1285     /* user-supplied MD5 check skipped (not supplied) */
1286     /* early quota check skipped--we don't have size yet */
1287     /* skipping user-supplied etag--we might have one in future, but
1288      * like data it and other attrs would arrive after open */
1289     processor = select_processor(*static_cast<RGWObjectCtx *>(s->obj_ctx),
1290                                  &multipart);
1291     op_ret = processor->prepare(get_store(), NULL);
1292     if (op_ret < 0) {
1293       ldout(s->cct, 20) << "processor->prepare() returned ret=" << op_ret
1294                         << dendl;
1295       goto done;
1296     }
1297
1298     filter = processor;
1299     if (compression_type != "none") {
1300       plugin = Compressor::create(s->cct, compression_type);
1301     if (! plugin) {
1302       ldout(s->cct, 1) << "Cannot load plugin for rgw_compression_type "
1303                        << compression_type << dendl;
1304     } else {
1305       compressor.emplace(s->cct, plugin, filter);
1306       filter = &*compressor;
1307     }
1308   }
1309
1310   done:
1311     return op_ret;
1312   } /* exec_start */
1313
1314   int RGWWriteRequest::exec_continue()
1315   {
1316     struct req_state* s = get_state();
1317     op_ret = 0;
1318
1319     /* check guards (e.g., contig write) */
1320     if (eio)
1321       return -EIO;
1322
1323     size_t len = data.length();
1324     if (! len)
1325       return 0;
1326
1327     /* XXX we are currently synchronous--supplied data buffers cannot
1328      * be used after the caller returns  */
1329     bool need_to_wait = true;
1330     bufferlist orig_data;
1331
1332     if (need_to_wait) {
1333       orig_data = data;
1334     }
1335     hash.Update((const byte *)data.c_str(), data.length());
1336     op_ret = put_data_and_throttle(filter, data, ofs, need_to_wait);
1337     if (op_ret < 0) {
1338       if (!need_to_wait || op_ret != -EEXIST) {
1339         ldout(s->cct, 20) << "processor->thottle_data() returned ret="
1340                           << op_ret << dendl;
1341         goto done;
1342       }
1343
1344       ldout(s->cct, 5) << "NOTICE: processor->throttle_data() returned -EEXIST, need to restart write" << dendl;
1345
1346       /* restore original data */
1347       data.swap(orig_data);
1348
1349       /* restart processing with different oid suffix */
1350       dispose_processor(processor);
1351       processor = select_processor(*static_cast<RGWObjectCtx *>(s->obj_ctx),
1352                                    &multipart);
1353       filter = processor;
1354
1355       string oid_rand;
1356       char buf[33];
1357       gen_rand_alphanumeric(get_store()->ctx(), buf, sizeof(buf) - 1);
1358       oid_rand.append(buf);
1359
1360       op_ret = processor->prepare(get_store(), &oid_rand);
1361       if (op_ret < 0) {
1362         ldout(s->cct, 0) << "ERROR: processor->prepare() returned "
1363                          << op_ret << dendl;
1364         goto done;
1365       }
1366
1367       /* restore compression filter, if any */
1368       if (compressor) {
1369         compressor.emplace(s->cct, plugin, filter);
1370         filter = &*compressor;
1371       }
1372
1373       op_ret = put_data_and_throttle(filter, data, ofs, false);
1374       if (op_ret < 0) {
1375         goto done;
1376       }
1377     }
1378     bytes_written += len;
1379
1380   done:
1381     return op_ret;
1382   } /* exec_continue */
1383
1384   int RGWWriteRequest::exec_finish()
1385   {
1386     buffer::list bl, aclbl, ux_key, ux_attrs;
1387     map<string, string>::iterator iter;
1388     char calc_md5[CEPH_CRYPTO_MD5_DIGESTSIZE * 2 + 1];
1389     unsigned char m[CEPH_CRYPTO_MD5_DIGESTSIZE];
1390     struct req_state* s = get_state();
1391
1392     size_t osize = rgw_fh->get_size();
1393     struct timespec octime = rgw_fh->get_ctime();
1394     struct timespec omtime = rgw_fh->get_mtime();
1395     real_time appx_t = real_clock::now();
1396
1397     s->obj_size = ofs; // XXX check ofs
1398     perfcounter->inc(l_rgw_put_b, s->obj_size);
1399
1400     op_ret = get_store()->check_quota(s->bucket_owner.get_id(), s->bucket,
1401                                       user_quota, bucket_quota, s->obj_size);
1402     if (op_ret < 0) {
1403       goto done;
1404     }
1405
1406     op_ret = get_store()->check_bucket_shards(s->bucket_info, s->bucket, bucket_quota);
1407     if (op_ret < 0) {
1408       goto done;
1409     }
1410
1411     hash.Final(m);
1412
1413     if (compressor && compressor->is_compressed()) {
1414       bufferlist tmp;
1415       RGWCompressionInfo cs_info;
1416       cs_info.compression_type = plugin->get_type_name();
1417       cs_info.orig_size = s->obj_size;
1418       cs_info.blocks = std::move(compressor->get_compression_blocks());
1419       ::encode(cs_info, tmp);
1420       attrs[RGW_ATTR_COMPRESSION] = tmp;
1421       ldout(s->cct, 20) << "storing " << RGW_ATTR_COMPRESSION
1422                         << " with type=" << cs_info.compression_type
1423                         << ", orig_size=" << cs_info.orig_size
1424                         << ", blocks=" << cs_info.blocks.size() << dendl;
1425     }
1426
1427     buf_to_hex(m, CEPH_CRYPTO_MD5_DIGESTSIZE, calc_md5);
1428     etag = calc_md5;
1429
1430     bl.append(etag.c_str(), etag.size() + 1);
1431     emplace_attr(RGW_ATTR_ETAG, std::move(bl));
1432
1433     policy.encode(aclbl);
1434     emplace_attr(RGW_ATTR_ACL, std::move(aclbl));
1435
1436     /* unix attrs */
1437     rgw_fh->set_mtime(real_clock::to_timespec(appx_t));
1438     rgw_fh->set_ctime(real_clock::to_timespec(appx_t));
1439     rgw_fh->set_size(bytes_written);
1440     rgw_fh->encode_attrs(ux_key, ux_attrs);
1441
1442     emplace_attr(RGW_ATTR_UNIX_KEY1, std::move(ux_key));
1443     emplace_attr(RGW_ATTR_UNIX1, std::move(ux_attrs));
1444
1445     for (iter = s->generic_attrs.begin(); iter != s->generic_attrs.end();
1446          ++iter) {
1447       buffer::list& attrbl = attrs[iter->first];
1448       const string& val = iter->second;
1449       attrbl.append(val.c_str(), val.size() + 1);
1450     }
1451
1452     rgw_get_request_metadata(s->cct, s->info, attrs);
1453     encode_delete_at_attr(delete_at, attrs);
1454
1455     /* Add a custom metadata to expose the information whether an object
1456      * is an SLO or not. Appending the attribute must be performed AFTER
1457      * processing any input from user in order to prohibit overwriting. */
1458     if (unlikely(!! slo_info)) {
1459       buffer::list slo_userindicator_bl;
1460       ::encode("True", slo_userindicator_bl);
1461       emplace_attr(RGW_ATTR_SLO_UINDICATOR, std::move(slo_userindicator_bl));
1462     }
1463
1464     op_ret = processor->complete(s->obj_size, etag, &mtime, real_time(), attrs,
1465                                  (delete_at ? *delete_at : real_time()),
1466                                  if_match, if_nomatch);
1467     if (op_ret != 0) {
1468       /* revert attr updates */
1469       rgw_fh->set_mtime(omtime);
1470       rgw_fh->set_ctime(octime);
1471       rgw_fh->set_size(osize);
1472     }
1473
1474   done:
1475     dispose_processor(processor);
1476     perfcounter->tinc(l_rgw_put_lat,
1477                       (ceph_clock_now() - s->time));
1478     return op_ret;
1479   } /* exec_finish */
1480
1481 } /* namespace rgw */
1482
1483 /* librgw */
1484 extern "C" {
1485
1486 void rgwfile_version(int *major, int *minor, int *extra)
1487 {
1488   if (major)
1489     *major = LIBRGW_FILE_VER_MAJOR;
1490   if (minor)
1491     *minor = LIBRGW_FILE_VER_MINOR;
1492   if (extra)
1493     *extra = LIBRGW_FILE_VER_EXTRA;
1494 }
1495
1496 /*
1497  attach rgw namespace
1498 */
1499   int rgw_mount(librgw_t rgw, const char *uid, const char *acc_key,
1500                 const char *sec_key, struct rgw_fs **rgw_fs,
1501                 uint32_t flags)
1502 {
1503   int rc = 0;
1504
1505   /* stash access data for "mount" */
1506   RGWLibFS* new_fs = new RGWLibFS(static_cast<CephContext*>(rgw), uid, acc_key,
1507                                   sec_key);
1508   assert(new_fs);
1509
1510   rc = new_fs->authorize(rgwlib.get_store());
1511   if (rc != 0) {
1512     delete new_fs;
1513     return -EINVAL;
1514   }
1515
1516   /* register fs for shared gc */
1517   rgwlib.get_fe()->get_process()->register_fs(new_fs);
1518
1519   struct rgw_fs *fs = new_fs->get_fs();
1520   fs->rgw = rgw;
1521
1522   /* XXX we no longer assume "/" is unique, but we aren't tracking the
1523    * roots atm */
1524
1525   *rgw_fs = fs;
1526
1527   return 0;
1528 }
1529
1530 /*
1531  register invalidate callbacks
1532 */
1533 int rgw_register_invalidate(struct rgw_fs *rgw_fs, rgw_fh_callback_t cb,
1534                             void *arg, uint32_t flags)
1535
1536 {
1537   RGWLibFS *fs = static_cast<RGWLibFS*>(rgw_fs->fs_private);
1538   return fs->register_invalidate(cb, arg, flags);
1539 }
1540
1541 /*
1542  detach rgw namespace
1543 */
1544 int rgw_umount(struct rgw_fs *rgw_fs, uint32_t flags)
1545 {
1546   RGWLibFS *fs = static_cast<RGWLibFS*>(rgw_fs->fs_private);
1547   fs->close();
1548   return 0;
1549 }
1550
1551 /*
1552   get filesystem attributes
1553 */
1554 int rgw_statfs(struct rgw_fs *rgw_fs,
1555                struct rgw_file_handle *parent_fh,
1556                struct rgw_statvfs *vfs_st, uint32_t flags)
1557 {
1558   RGWLibFS *fs = static_cast<RGWLibFS*>(rgw_fs->fs_private);
1559
1560   /* XXX for now, just publish a huge capacity and
1561    * limited utiliztion */
1562   vfs_st->f_bsize = 1024*1024 /* 1M */;
1563   vfs_st->f_frsize = 1024;    /* minimal allocation unit (who cares) */
1564   vfs_st->f_blocks = UINT64_MAX;
1565   vfs_st->f_bfree = UINT64_MAX;
1566   vfs_st->f_bavail = UINT64_MAX;
1567   vfs_st->f_files = 1024; /* object count, do we have an est? */
1568   vfs_st->f_ffree = UINT64_MAX;
1569   vfs_st->f_fsid[0] = fs->get_inst();
1570   vfs_st->f_fsid[1] = fs->get_inst();
1571   vfs_st->f_flag = 0;
1572   vfs_st->f_namemax = 4096;
1573   return 0;
1574 }
1575
1576 /*
1577   generic create -- create an empty regular file
1578 */
1579 int rgw_create(struct rgw_fs *rgw_fs, struct rgw_file_handle *parent_fh,
1580                const char *name, struct stat *st, uint32_t mask,
1581                struct rgw_file_handle **fh, uint32_t posix_flags,
1582                uint32_t flags)
1583 {
1584   using std::get;
1585
1586   RGWLibFS *fs = static_cast<RGWLibFS*>(rgw_fs->fs_private);
1587   RGWFileHandle* parent = get_rgwfh(parent_fh);
1588
1589   if ((! parent) ||
1590       (parent->is_root()) ||
1591       (parent->is_file())) {
1592     /* bad parent */
1593     return -EINVAL;
1594   }
1595
1596   MkObjResult fhr = fs->create(parent, name, st, mask, flags);
1597   RGWFileHandle *nfh = get<0>(fhr); // nullptr if !success
1598
1599   if (nfh)
1600     *fh = nfh->get_fh();
1601
1602   return get<1>(fhr);
1603 } /* rgw_create */
1604
1605 /*
1606   create a new directory
1607 */
1608 int rgw_mkdir(struct rgw_fs *rgw_fs,
1609               struct rgw_file_handle *parent_fh,
1610               const char *name, struct stat *st, uint32_t mask,
1611               struct rgw_file_handle **fh, uint32_t flags)
1612 {
1613   using std::get;
1614
1615   RGWLibFS *fs = static_cast<RGWLibFS*>(rgw_fs->fs_private);
1616   RGWFileHandle* parent = get_rgwfh(parent_fh);
1617
1618   if (! parent) {
1619     /* bad parent */
1620     return -EINVAL;
1621   }
1622
1623   MkObjResult fhr = fs->mkdir(parent, name, st, mask, flags);
1624   RGWFileHandle *nfh = get<0>(fhr); // nullptr if !success
1625
1626   if (nfh)
1627     *fh = nfh->get_fh();
1628
1629   return get<1>(fhr);
1630 } /* rgw_mkdir */
1631
1632 /*
1633   rename object
1634 */
1635 int rgw_rename(struct rgw_fs *rgw_fs,
1636                struct rgw_file_handle *src, const char* src_name,
1637                struct rgw_file_handle *dst, const char* dst_name,
1638                uint32_t flags)
1639 {
1640   RGWLibFS *fs = static_cast<RGWLibFS*>(rgw_fs->fs_private);
1641
1642   RGWFileHandle* src_fh = get_rgwfh(src);
1643   RGWFileHandle* dst_fh = get_rgwfh(dst);
1644
1645   return fs->rename(src_fh, dst_fh, src_name, dst_name);
1646 }
1647
1648 /*
1649   remove file or directory
1650 */
1651 int rgw_unlink(struct rgw_fs *rgw_fs, struct rgw_file_handle *parent_fh,
1652                const char *name, uint32_t flags)
1653 {
1654   RGWLibFS *fs = static_cast<RGWLibFS*>(rgw_fs->fs_private);
1655   RGWFileHandle* parent = get_rgwfh(parent_fh);
1656
1657   return fs->unlink(parent, name);
1658 }
1659
1660 /*
1661   lookup object by name (POSIX style)
1662 */
1663 int rgw_lookup(struct rgw_fs *rgw_fs,
1664               struct rgw_file_handle *parent_fh, const char* path,
1665               struct rgw_file_handle **fh, uint32_t flags)
1666 {
1667   //CephContext* cct = static_cast<CephContext*>(rgw_fs->rgw);
1668   RGWLibFS *fs = static_cast<RGWLibFS*>(rgw_fs->fs_private);
1669
1670   RGWFileHandle* parent = get_rgwfh(parent_fh);
1671   if ((! parent) ||
1672       (! parent->is_dir())) {
1673     /* bad parent */
1674     return -EINVAL;
1675   }
1676
1677   RGWFileHandle* rgw_fh;
1678   LookupFHResult fhr;
1679
1680   if (parent->is_root()) {
1681     /* special: parent lookup--note lack of ref()! */
1682     if (unlikely((strcmp(path, "..") == 0) ||
1683                  (strcmp(path, "/") == 0))) {
1684       rgw_fh = parent;
1685     } else {
1686       RGWLibFS::BucketStats bstat;
1687       fhr = fs->stat_bucket(parent, path, bstat, RGWFileHandle::FLAG_NONE);
1688       rgw_fh = get<0>(fhr);
1689       if (! rgw_fh)
1690         return -ENOENT;
1691     }
1692   } else {
1693     /* special: after readdir--note extra ref()! */
1694     if (unlikely((strcmp(path, "..") == 0))) {
1695       rgw_fh = parent;
1696       lsubdout(fs->get_context(), rgw, 17)
1697         << __func__ << "BANG"<< *rgw_fh
1698         << dendl;
1699       fs->ref(rgw_fh);
1700     } else {
1701       /* lookup in a readdir callback */
1702       enum rgw_fh_type fh_type = fh_type_of(flags);
1703
1704       uint32_t sl_flags = (flags & RGW_LOOKUP_FLAG_RCB)
1705         ? RGWFileHandle::FLAG_NONE
1706         : RGWFileHandle::FLAG_EXACT_MATCH;
1707
1708       fhr = fs->stat_leaf(parent, path, fh_type, sl_flags);
1709       if (! get<0>(fhr)) {
1710         if (! (flags & RGW_LOOKUP_FLAG_CREATE))
1711           return -ENOENT;
1712         else
1713           fhr = fs->lookup_fh(parent, path, RGWFileHandle::FLAG_CREATE);
1714       }
1715       rgw_fh = get<0>(fhr);
1716     }
1717   } /* !root */
1718
1719   struct rgw_file_handle *rfh = rgw_fh->get_fh();
1720   *fh = rfh;
1721
1722   return 0;
1723 } /* rgw_lookup */
1724
1725 /*
1726   lookup object by handle (NFS style)
1727 */
1728 int rgw_lookup_handle(struct rgw_fs *rgw_fs, struct rgw_fh_hk *fh_hk,
1729                       struct rgw_file_handle **fh, uint32_t flags)
1730 {
1731   RGWLibFS *fs = static_cast<RGWLibFS*>(rgw_fs->fs_private);
1732
1733   RGWFileHandle* rgw_fh = fs->lookup_handle(*fh_hk);
1734   if (! rgw_fh) {
1735     /* not found */
1736     return -ENOENT;
1737   }
1738
1739   struct rgw_file_handle *rfh = rgw_fh->get_fh();
1740   *fh = rfh;
1741
1742   return 0;
1743 }
1744
1745 /*
1746  * release file handle
1747  */
1748 int rgw_fh_rele(struct rgw_fs *rgw_fs, struct rgw_file_handle *fh,
1749                 uint32_t flags)
1750 {
1751   RGWLibFS *fs = static_cast<RGWLibFS*>(rgw_fs->fs_private);
1752   RGWFileHandle* rgw_fh = get_rgwfh(fh);
1753
1754   lsubdout(fs->get_context(), rgw, 17)
1755     << __func__ << " " << *rgw_fh
1756     << dendl;
1757
1758   fs->unref(rgw_fh);
1759   return 0;
1760 }
1761
1762 /*
1763    get unix attributes for object
1764 */
1765 int rgw_getattr(struct rgw_fs *rgw_fs,
1766                 struct rgw_file_handle *fh, struct stat *st, uint32_t flags)
1767 {
1768   RGWLibFS *fs = static_cast<RGWLibFS*>(rgw_fs->fs_private);
1769   RGWFileHandle* rgw_fh = get_rgwfh(fh);
1770
1771   return fs->getattr(rgw_fh, st);
1772 }
1773
1774 /*
1775   set unix attributes for object
1776 */
1777 int rgw_setattr(struct rgw_fs *rgw_fs,
1778                 struct rgw_file_handle *fh, struct stat *st,
1779                 uint32_t mask, uint32_t flags)
1780 {
1781   RGWLibFS *fs = static_cast<RGWLibFS*>(rgw_fs->fs_private);
1782   RGWFileHandle* rgw_fh = get_rgwfh(fh);
1783
1784   return fs->setattr(rgw_fh, st, mask, flags);
1785 }
1786
1787 /*
1788    truncate file
1789 */
1790 int rgw_truncate(struct rgw_fs *rgw_fs,
1791                  struct rgw_file_handle *fh, uint64_t size, uint32_t flags)
1792 {
1793   return 0;
1794 }
1795
1796 /*
1797    open file
1798 */
1799 int rgw_open(struct rgw_fs *rgw_fs,
1800              struct rgw_file_handle *fh, uint32_t posix_flags, uint32_t flags)
1801 {
1802   RGWFileHandle* rgw_fh = get_rgwfh(fh);
1803
1804   /* XXX
1805    * need to track specific opens--at least read opens and
1806    * a write open;  we need to know when a write open is returned,
1807    * that closes a write transaction
1808    *
1809    * for now, we will support single-open only, it's preferable to
1810    * anything we can otherwise do without access to the NFS state
1811    */
1812   if (! rgw_fh->is_file())
1813     return -EISDIR;
1814
1815   return rgw_fh->open(flags);
1816 }
1817
1818 /*
1819    close file
1820 */
1821 int rgw_close(struct rgw_fs *rgw_fs,
1822               struct rgw_file_handle *fh, uint32_t flags)
1823 {
1824   RGWLibFS *fs = static_cast<RGWLibFS*>(rgw_fs->fs_private);
1825   RGWFileHandle* rgw_fh = get_rgwfh(fh);
1826   int rc = rgw_fh->close(/* XXX */);
1827
1828   if (flags & RGW_CLOSE_FLAG_RELE)
1829     fs->unref(rgw_fh);
1830
1831   return rc;
1832 }
1833
1834 int rgw_readdir(struct rgw_fs *rgw_fs,
1835                 struct rgw_file_handle *parent_fh, uint64_t *offset,
1836                 rgw_readdir_cb rcb, void *cb_arg, bool *eof,
1837                 uint32_t flags)
1838 {
1839   RGWFileHandle* parent = get_rgwfh(parent_fh);
1840   if (! parent) {
1841     /* bad parent */
1842     return -EINVAL;
1843   }
1844   int rc = parent->readdir(rcb, cb_arg, offset, eof, flags);
1845   return rc;
1846 }
1847
1848 /* project offset of dirent name */
1849 int rgw_dirent_offset(struct rgw_fs *rgw_fs,
1850                       struct rgw_file_handle *parent_fh,
1851                       const char *name, int64_t *offset,
1852                       uint32_t flags)
1853 {
1854   RGWFileHandle* parent = get_rgwfh(parent_fh);
1855   if ((! parent)) {
1856     /* bad parent */
1857     return -EINVAL;
1858   }
1859   std::string sname{name};
1860   int rc = parent->offset_of(sname, offset, flags);
1861   return rc;
1862 }
1863
1864 /*
1865    read data from file
1866 */
1867 int rgw_read(struct rgw_fs *rgw_fs,
1868              struct rgw_file_handle *fh, uint64_t offset,
1869              size_t length, size_t *bytes_read, void *buffer,
1870              uint32_t flags)
1871 {
1872   RGWLibFS *fs = static_cast<RGWLibFS*>(rgw_fs->fs_private);
1873   RGWFileHandle* rgw_fh = get_rgwfh(fh);
1874
1875   return fs->read(rgw_fh, offset, length, bytes_read, buffer, flags);
1876 }
1877
1878 /*
1879    write data to file
1880 */
1881 int rgw_write(struct rgw_fs *rgw_fs,
1882               struct rgw_file_handle *fh, uint64_t offset,
1883               size_t length, size_t *bytes_written, void *buffer,
1884               uint32_t flags)
1885 {
1886   RGWFileHandle* rgw_fh = get_rgwfh(fh);
1887   int rc;
1888
1889   *bytes_written = 0;
1890
1891   if (! rgw_fh->is_file())
1892     return -EISDIR;
1893
1894   if (! rgw_fh->is_open())
1895     return -EPERM;
1896
1897   rc = rgw_fh->write(offset, length, bytes_written, buffer);
1898
1899   return rc;
1900 }
1901
1902 /*
1903    read data from file (vector)
1904 */
1905 class RGWReadV
1906 {
1907   buffer::list bl;
1908   struct rgw_vio* vio;
1909
1910 public:
1911   RGWReadV(buffer::list& _bl, rgw_vio* _vio) : vio(_vio) {
1912     bl.claim(_bl);
1913   }
1914
1915   struct rgw_vio* get_vio() { return vio; }
1916
1917   const std::list<buffer::ptr>& buffers() { return bl.buffers(); }
1918
1919   unsigned /* XXX */ length() { return bl.length(); }
1920
1921 };
1922
1923 void rgw_readv_rele(struct rgw_uio *uio, uint32_t flags)
1924 {
1925   RGWReadV* rdv = static_cast<RGWReadV*>(uio->uio_p1);
1926   rdv->~RGWReadV();
1927   ::operator delete(rdv);
1928 }
1929
1930 int rgw_readv(struct rgw_fs *rgw_fs,
1931               struct rgw_file_handle *fh, rgw_uio *uio, uint32_t flags)
1932 {
1933 #if 0 /* XXX */
1934   CephContext* cct = static_cast<CephContext*>(rgw_fs->rgw);
1935   RGWLibFS *fs = static_cast<RGWLibFS*>(rgw_fs->fs_private);
1936   RGWFileHandle* rgw_fh = get_rgwfh(fh);
1937
1938   if (! rgw_fh->is_file())
1939     return -EINVAL;
1940
1941   int rc = 0;
1942
1943   buffer::list bl;
1944   RGWGetObjRequest req(cct, fs->get_user(), rgw_fh->bucket_name(),
1945                       rgw_fh->object_name(), uio->uio_offset, uio->uio_resid,
1946                       bl);
1947   req.do_hexdump = false;
1948
1949   rc = rgwlib.get_fe()->execute_req(&req);
1950
1951   if (! rc) {
1952     RGWReadV* rdv = static_cast<RGWReadV*>(
1953       ::operator new(sizeof(RGWReadV) +
1954                     (bl.buffers().size() * sizeof(struct rgw_vio))));
1955
1956     (void) new (rdv)
1957       RGWReadV(bl, reinterpret_cast<rgw_vio*>(rdv+sizeof(RGWReadV)));
1958
1959     uio->uio_p1 = rdv;
1960     uio->uio_cnt = rdv->buffers().size();
1961     uio->uio_resid = rdv->length();
1962     uio->uio_vio = rdv->get_vio();
1963     uio->uio_rele = rgw_readv_rele;
1964
1965     int ix = 0;
1966     auto& buffers = rdv->buffers();
1967     for (auto& bp : buffers) {
1968       rgw_vio *vio = &(uio->uio_vio[ix]);
1969       vio->vio_base = const_cast<char*>(bp.c_str());
1970       vio->vio_len = bp.length();
1971       vio->vio_u1 = nullptr;
1972       vio->vio_p1 = nullptr;
1973       ++ix;
1974     }
1975   }
1976
1977   return rc;
1978 #else
1979   return 0;
1980 #endif
1981 }
1982
1983 /*
1984    write data to file (vector)
1985 */
1986 int rgw_writev(struct rgw_fs *rgw_fs, struct rgw_file_handle *fh,
1987               rgw_uio *uio, uint32_t flags)
1988 {
1989
1990   return -ENOTSUP;
1991
1992   CephContext* cct = static_cast<CephContext*>(rgw_fs->rgw);
1993   RGWLibFS *fs = static_cast<RGWLibFS*>(rgw_fs->fs_private);
1994   RGWFileHandle* rgw_fh = get_rgwfh(fh);
1995
1996   if (! rgw_fh->is_file())
1997     return -EINVAL;
1998
1999   buffer::list bl;
2000   for (unsigned int ix = 0; ix < uio->uio_cnt; ++ix) {
2001     rgw_vio *vio = &(uio->uio_vio[ix]);
2002     bl.push_back(
2003       buffer::create_static(vio->vio_len,
2004                             static_cast<char*>(vio->vio_base)));
2005   }
2006
2007   std::string oname = rgw_fh->relative_object_name();
2008   RGWPutObjRequest req(cct, fs->get_user(), rgw_fh->bucket_name(),
2009                        oname, bl);
2010
2011   int rc = rgwlib.get_fe()->execute_req(&req);
2012
2013   /* XXX update size (in request) */
2014
2015   return rc;
2016 }
2017
2018 /*
2019    sync written data
2020 */
2021 int rgw_fsync(struct rgw_fs *rgw_fs, struct rgw_file_handle *handle,
2022               uint32_t flags)
2023 {
2024   return 0;
2025 }
2026
2027 int rgw_commit(struct rgw_fs *rgw_fs, struct rgw_file_handle *fh,
2028                uint64_t offset, uint64_t length, uint32_t flags)
2029 {
2030   RGWFileHandle* rgw_fh = get_rgwfh(fh);
2031
2032   return rgw_fh->commit(offset, length, RGWFileHandle::FLAG_NONE);
2033 }
2034
2035 } /* extern "C" */