fs/ceph/super.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2
   3 #include <linux/ceph/ceph_debug.h>
   4
   5 #include <linux/backing-dev.h>
   6 #include <linux/ctype.h>
   7 #include <linux/fs.h>
   8 #include <linux/inet.h>
   9 #include <linux/in6.h>
  10 #include <linux/module.h>
  11 #include <linux/mount.h>
  12 #include <linux/fs_context.h>
  13 #include <linux/fs_parser.h>
  14 #include <linux/sched.h>
  15 #include <linux/seq_file.h>
  16 #include <linux/slab.h>
  17 #include <linux/statfs.h>
  18 #include <linux/string.h>
  19
  20 #include "super.h"
  21 #include "mds_client.h"
  22 #include "cache.h"
  23
  24 #include <linux/ceph/ceph_features.h>
  25 #include <linux/ceph/decode.h>
  26 #include <linux/ceph/mon_client.h>
  27 #include <linux/ceph/auth.h>
  28 #include <linux/ceph/debugfs.h>
  29
  30 static DEFINE_SPINLOCK(ceph_fsc_lock);
  31 static LIST_HEAD(ceph_fsc_list);
  32
  33 /*
  34  * Ceph superblock operations
  35  *
  36  * Handle the basics of mounting, unmounting.
  37  */
  38
  39 /*
  40  * super ops
  41  */
  42 static void ceph_put_super(struct super_block *s)
  43 {
  44         struct ceph_fs_client *fsc = ceph_sb_to_client(s);
  45
  46         dout("put_super\n");
  47         ceph_mdsc_close_sessions(fsc->mdsc);
  48 }
  49
  50 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
  51 {
  52         struct ceph_fs_client *fsc = ceph_inode_to_client(d_inode(dentry));
  53         struct ceph_mon_client *monc = &fsc->client->monc;
  54         struct ceph_statfs st;
  55         int i, err;
  56         u64 data_pool;
  57
  58         if (fsc->mdsc->mdsmap->m_num_data_pg_pools == 1) {
  59                 data_pool = fsc->mdsc->mdsmap->m_data_pg_pools[0];
  60         } else {
  61                 data_pool = CEPH_NOPOOL;
  62         }
  63
  64         dout("statfs\n");
  65         err = ceph_monc_do_statfs(monc, data_pool, &st);
  66         if (err < 0)
  67                 return err;
  68
  69         /* fill in kstatfs */
  70         buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
  71
  72         /*
  73          * express utilization in terms of large blocks to avoid
  74          * overflow on 32-bit machines.
  75          *
  76          * NOTE: for the time being, we make bsize == frsize to humor
  77          * not-yet-ancient versions of glibc that are broken.
  78          * Someday, we will probably want to report a real block
  79          * size...  whatever that may mean for a network file system!
  80          */
  81         buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
  82         buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
  83
  84         /*
  85          * By default use root quota for stats; fallback to overall filesystem
  86          * usage if using 'noquotadf' mount option or if the root dir doesn't
  87          * have max_bytes quota set.
  88          */
  89         if (ceph_test_mount_opt(fsc, NOQUOTADF) ||
  90             !ceph_quota_update_statfs(fsc, buf)) {
  91                 buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
  92                 buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
  93                 buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
  94         }
  95
  96         buf->f_files = le64_to_cpu(st.num_objects);
  97         buf->f_ffree = -1;
  98         buf->f_namelen = NAME_MAX;
  99
 100         /* Must convert the fsid, for consistent values across arches */
 101         buf->f_fsid.val[0] = 0;
 102         mutex_lock(&monc->mutex);
 103         for (i = 0 ; i < sizeof(monc->monmap->fsid) / sizeof(__le32) ; ++i)
 104                 buf->f_fsid.val[0] ^= le32_to_cpu(((__le32 *)&monc->monmap->fsid)[i]);
 105         mutex_unlock(&monc->mutex);
 106
 107         /* fold the fs_cluster_id into the upper bits */
 108         buf->f_fsid.val[1] = monc->fs_cluster_id;
 109
 110         return 0;
 111 }
 112
 113 static int ceph_sync_fs(struct super_block *sb, int wait)
 114 {
 115         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
 116
 117         if (!wait) {
 118                 dout("sync_fs (non-blocking)\n");
 119                 ceph_flush_dirty_caps(fsc->mdsc);
 120                 dout("sync_fs (non-blocking) done\n");
 121                 return 0;
 122         }
 123
 124         dout("sync_fs (blocking)\n");
 125         ceph_osdc_sync(&fsc->client->osdc);
 126         ceph_mdsc_sync(fsc->mdsc);
 127         dout("sync_fs (blocking) done\n");
 128         return 0;
 129 }
 130
 131 /*
 132  * mount options
 133  */
 134 enum {
 135         Opt_wsize,
 136         Opt_rsize,
 137         Opt_rasize,
 138         Opt_caps_wanted_delay_min,
 139         Opt_caps_wanted_delay_max,
 140         Opt_caps_max,
 141         Opt_readdir_max_entries,
 142         Opt_readdir_max_bytes,
 143         Opt_congestion_kb,
 144         /* int args above */
 145         Opt_snapdirname,
 146         Opt_mds_namespace,
 147         Opt_recover_session,
 148         Opt_source,
 149         Opt_mon_addr,
 150         /* string args above */
 151         Opt_dirstat,
 152         Opt_rbytes,
 153         Opt_asyncreaddir,
 154         Opt_dcache,
 155         Opt_ino32,
 156         Opt_fscache,
 157         Opt_poolperm,
 158         Opt_require_active_mds,
 159         Opt_acl,
 160         Opt_quotadf,
 161         Opt_copyfrom,
 162         Opt_wsync,
 163 };
 164
 165 enum ceph_recover_session_mode {
 166         ceph_recover_session_no,
 167         ceph_recover_session_clean
 168 };
 169
 170 static const struct constant_table ceph_param_recover[] = {
 171         { "no",         ceph_recover_session_no },
 172         { "clean",      ceph_recover_session_clean },
 173         {}
 174 };
 175
 176 static const struct fs_parameter_spec ceph_mount_parameters[] = {
 177         fsparam_flag_no ("acl",                         Opt_acl),
 178         fsparam_flag_no ("asyncreaddir",                Opt_asyncreaddir),
 179         fsparam_s32     ("caps_max",                    Opt_caps_max),
 180         fsparam_u32     ("caps_wanted_delay_max",       Opt_caps_wanted_delay_max),
 181         fsparam_u32     ("caps_wanted_delay_min",       Opt_caps_wanted_delay_min),
 182         fsparam_u32     ("write_congestion_kb",         Opt_congestion_kb),
 183         fsparam_flag_no ("copyfrom",                    Opt_copyfrom),
 184         fsparam_flag_no ("dcache",                      Opt_dcache),
 185         fsparam_flag_no ("dirstat",                     Opt_dirstat),
 186         fsparam_flag_no ("fsc",                         Opt_fscache), // fsc|nofsc
 187         fsparam_string  ("fsc",                         Opt_fscache), // fsc=...
 188         fsparam_flag_no ("ino32",                       Opt_ino32),
 189         fsparam_string  ("mds_namespace",               Opt_mds_namespace),
 190         fsparam_flag_no ("poolperm",                    Opt_poolperm),
 191         fsparam_flag_no ("quotadf",                     Opt_quotadf),
 192         fsparam_u32     ("rasize",                      Opt_rasize),
 193         fsparam_flag_no ("rbytes",                      Opt_rbytes),
 194         fsparam_u32     ("readdir_max_bytes",           Opt_readdir_max_bytes),
 195         fsparam_u32     ("readdir_max_entries",         Opt_readdir_max_entries),
 196         fsparam_enum    ("recover_session",             Opt_recover_session, ceph_param_recover),
 197         fsparam_flag_no ("require_active_mds",          Opt_require_active_mds),
 198         fsparam_u32     ("rsize",                       Opt_rsize),
 199         fsparam_string  ("snapdirname",                 Opt_snapdirname),
 200         fsparam_string  ("source",                      Opt_source),
 201         fsparam_string  ("mon_addr",                    Opt_mon_addr),
 202         fsparam_u32     ("wsize",                       Opt_wsize),
 203         fsparam_flag_no ("wsync",                       Opt_wsync),
 204         {}
 205 };
 206
 207 struct ceph_parse_opts_ctx {
 208         struct ceph_options             *copts;
 209         struct ceph_mount_options       *opts;
 210 };
 211
 212 /*
 213  * Remove adjacent slashes and then the trailing slash, unless it is
 214  * the only remaining character.
 215  *
 216  * E.g. "//dir1////dir2///" --> "/dir1/dir2", "///" --> "/".
 217  */
 218 static void canonicalize_path(char *path)
 219 {
 220         int i, j = 0;
 221
 222         for (i = 0; path[i] != '\0'; i++) {
 223                 if (path[i] != '/' || j < 1 || path[j - 1] != '/')
 224                         path[j++] = path[i];
 225         }
 226
 227         if (j > 1 && path[j - 1] == '/')
 228                 j--;
 229         path[j] = '\0';
 230 }
 231
 232 /*
 233  * Check if the mds namespace in ceph_mount_options matches
 234  * the passed in namespace string. First time match (when
 235  * ->mds_namespace is NULL) is treated specially, since
 236  * ->mds_namespace needs to be initialized by the caller.
 237  */
 238 static int namespace_equals(struct ceph_mount_options *fsopt,
 239                             const char *namespace, size_t len)
 240 {
 241         return !(fsopt->mds_namespace &&
 242                  (strlen(fsopt->mds_namespace) != len ||
 243                   strncmp(fsopt->mds_namespace, namespace, len)));
 244 }
 245
 246 static int ceph_parse_old_source(const char *dev_name, const char *dev_name_end,
 247                                  struct fs_context *fc)
 248 {
 249         int r;
 250         struct ceph_parse_opts_ctx *pctx = fc->fs_private;
 251         struct ceph_mount_options *fsopt = pctx->opts;
 252
 253         if (*dev_name_end != ':')
 254                 return invalfc(fc, "separator ':' missing in source");
 255
 256         r = ceph_parse_mon_ips(dev_name, dev_name_end - dev_name,
 257                                pctx->copts, fc->log.log, ',');
 258         if (r)
 259                 return r;
 260
 261         fsopt->new_dev_syntax = false;
 262         return 0;
 263 }
 264
 265 static int ceph_parse_new_source(const char *dev_name, const char *dev_name_end,
 266                                  struct fs_context *fc)
 267 {
 268         size_t len;
 269         struct ceph_fsid fsid;
 270         struct ceph_parse_opts_ctx *pctx = fc->fs_private;
 271         struct ceph_mount_options *fsopt = pctx->opts;
 272         char *fsid_start, *fs_name_start;
 273
 274         if (*dev_name_end != '=') {
 275                 dout("separator '=' missing in source");
 276                 return -EINVAL;
 277         }
 278
 279         fsid_start = strchr(dev_name, '@');
 280         if (!fsid_start)
 281                 return invalfc(fc, "missing cluster fsid");
 282         ++fsid_start; /* start of cluster fsid */
 283
 284         fs_name_start = strchr(fsid_start, '.');
 285         if (!fs_name_start)
 286                 return invalfc(fc, "missing file system name");
 287
 288         if (ceph_parse_fsid(fsid_start, &fsid))
 289                 return invalfc(fc, "Invalid FSID");
 290
 291         ++fs_name_start; /* start of file system name */
 292         len = dev_name_end - fs_name_start;
 293
 294         if (!namespace_equals(fsopt, fs_name_start, len))
 295                 return invalfc(fc, "Mismatching mds_namespace");
 296         kfree(fsopt->mds_namespace);
 297         fsopt->mds_namespace = kstrndup(fs_name_start, len, GFP_KERNEL);
 298         if (!fsopt->mds_namespace)
 299                 return -ENOMEM;
 300         dout("file system (mds namespace) '%s'\n", fsopt->mds_namespace);
 301
 302         fsopt->new_dev_syntax = true;
 303         return 0;
 304 }
 305
 306 /*
 307  * Parse the source parameter for new device format. Distinguish the device
 308  * spec from the path. Try parsing new device format and fallback to old
 309  * format if needed.
 310  *
 311  * New device syntax will looks like:
 312  *     <device_spec>=/<path>
 313  * where
 314  *     <device_spec> is name@fsid.fsname
 315  *     <path> is optional, but if present must begin with '/'
 316  * (monitor addresses are passed via mount option)
 317  *
 318  * Old device syntax is:
 319  *     <server_spec>[,<server_spec>...]:[<path>]
 320  * where
 321  *     <server_spec> is <ip>[:<port>]
 322  *     <path> is optional, but if present must begin with '/'
 323  */
 324 static int ceph_parse_source(struct fs_parameter *param, struct fs_context *fc)
 325 {
 326         struct ceph_parse_opts_ctx *pctx = fc->fs_private;
 327         struct ceph_mount_options *fsopt = pctx->opts;
 328         char *dev_name = param->string, *dev_name_end;
 329         int ret;
 330
 331         dout("%s '%s'\n", __func__, dev_name);
 332         if (!dev_name || !*dev_name)
 333                 return invalfc(fc, "Empty source");
 334
 335         dev_name_end = strchr(dev_name, '/');
 336         if (dev_name_end) {
 337                 /*
 338                  * The server_path will include the whole chars from userland
 339                  * including the leading '/'.
 340                  */
 341                 kfree(fsopt->server_path);
 342                 fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
 343                 if (!fsopt->server_path)
 344                         return -ENOMEM;
 345
 346                 canonicalize_path(fsopt->server_path);
 347         } else {
 348                 dev_name_end = dev_name + strlen(dev_name);
 349         }
 350
 351         dev_name_end--;         /* back up to separator */
 352         if (dev_name_end < dev_name)
 353                 return invalfc(fc, "Path missing in source");
 354
 355         dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
 356         if (fsopt->server_path)
 357                 dout("server path '%s'\n", fsopt->server_path);
 358
 359         dout("trying new device syntax");
 360         ret = ceph_parse_new_source(dev_name, dev_name_end, fc);
 361         if (ret) {
 362                 if (ret != -EINVAL)
 363                         return ret;
 364                 dout("trying old device syntax");
 365                 ret = ceph_parse_old_source(dev_name, dev_name_end, fc);
 366                 if (ret)
 367                         return ret;
 368         }
 369
 370         fc->source = param->string;
 371         param->string = NULL;
 372         return 0;
 373 }
 374
 375 static int ceph_parse_mon_addr(struct fs_parameter *param,
 376                                struct fs_context *fc)
 377 {
 378         struct ceph_parse_opts_ctx *pctx = fc->fs_private;
 379         struct ceph_mount_options *fsopt = pctx->opts;
 380
 381         kfree(fsopt->mon_addr);
 382         fsopt->mon_addr = param->string;
 383         param->string = NULL;
 384
 385         return ceph_parse_mon_ips(fsopt->mon_addr, strlen(fsopt->mon_addr),
 386                                   pctx->copts, fc->log.log, '/');
 387 }
 388
 389 static int ceph_parse_mount_param(struct fs_context *fc,
 390                                   struct fs_parameter *param)
 391 {
 392         struct ceph_parse_opts_ctx *pctx = fc->fs_private;
 393         struct ceph_mount_options *fsopt = pctx->opts;
 394         struct fs_parse_result result;
 395         unsigned int mode;
 396         int token, ret;
 397
 398         ret = ceph_parse_param(param, pctx->copts, fc->log.log);
 399         if (ret != -ENOPARAM)
 400                 return ret;
 401
 402         token = fs_parse(fc, ceph_mount_parameters, param, &result);
 403         dout("%s fs_parse '%s' token %d\n", __func__, param->key, token);
 404         if (token < 0)
 405                 return token;
 406
 407         switch (token) {
 408         case Opt_snapdirname:
 409                 kfree(fsopt->snapdir_name);
 410                 fsopt->snapdir_name = param->string;
 411                 param->string = NULL;
 412                 break;
 413         case Opt_mds_namespace:
 414                 if (!namespace_equals(fsopt, param->string, strlen(param->string)))
 415                         return invalfc(fc, "Mismatching mds_namespace");
 416                 kfree(fsopt->mds_namespace);
 417                 fsopt->mds_namespace = param->string;
 418                 param->string = NULL;
 419                 break;
 420         case Opt_recover_session:
 421                 mode = result.uint_32;
 422                 if (mode == ceph_recover_session_no)
 423                         fsopt->flags &= ~CEPH_MOUNT_OPT_CLEANRECOVER;
 424                 else if (mode == ceph_recover_session_clean)
 425                         fsopt->flags |= CEPH_MOUNT_OPT_CLEANRECOVER;
 426                 else
 427                         BUG();
 428                 break;
 429         case Opt_source:
 430                 if (fc->source)
 431                         return invalfc(fc, "Multiple sources specified");
 432                 return ceph_parse_source(param, fc);
 433         case Opt_mon_addr:
 434                 return ceph_parse_mon_addr(param, fc);
 435         case Opt_wsize:
 436                 if (result.uint_32 < PAGE_SIZE ||
 437                     result.uint_32 > CEPH_MAX_WRITE_SIZE)
 438                         goto out_of_range;
 439                 fsopt->wsize = ALIGN(result.uint_32, PAGE_SIZE);
 440                 break;
 441         case Opt_rsize:
 442                 if (result.uint_32 < PAGE_SIZE ||
 443                     result.uint_32 > CEPH_MAX_READ_SIZE)
 444                         goto out_of_range;
 445                 fsopt->rsize = ALIGN(result.uint_32, PAGE_SIZE);
 446                 break;
 447         case Opt_rasize:
 448                 fsopt->rasize = ALIGN(result.uint_32, PAGE_SIZE);
 449                 break;
 450         case Opt_caps_wanted_delay_min:
 451                 if (result.uint_32 < 1)
 452                         goto out_of_range;
 453                 fsopt->caps_wanted_delay_min = result.uint_32;
 454                 break;
 455         case Opt_caps_wanted_delay_max:
 456                 if (result.uint_32 < 1)
 457                         goto out_of_range;
 458                 fsopt->caps_wanted_delay_max = result.uint_32;
 459                 break;
 460         case Opt_caps_max:
 461                 if (result.int_32 < 0)
 462                         goto out_of_range;
 463                 fsopt->caps_max = result.int_32;
 464                 break;
 465         case Opt_readdir_max_entries:
 466                 if (result.uint_32 < 1)
 467                         goto out_of_range;
 468                 fsopt->max_readdir = result.uint_32;
 469                 break;
 470         case Opt_readdir_max_bytes:
 471                 if (result.uint_32 < PAGE_SIZE && result.uint_32 != 0)
 472                         goto out_of_range;
 473                 fsopt->max_readdir_bytes = result.uint_32;
 474                 break;
 475         case Opt_congestion_kb:
 476                 if (result.uint_32 < 1024) /* at least 1M */
 477                         goto out_of_range;
 478                 fsopt->congestion_kb = result.uint_32;
 479                 break;
 480         case Opt_dirstat:
 481                 if (!result.negated)
 482                         fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
 483                 else
 484                         fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
 485                 break;
 486         case Opt_rbytes:
 487                 if (!result.negated)
 488                         fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
 489                 else
 490                         fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
 491                 break;
 492         case Opt_asyncreaddir:
 493                 if (!result.negated)
 494                         fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
 495                 else
 496                         fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
 497                 break;
 498         case Opt_dcache:
 499                 if (!result.negated)
 500                         fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
 501                 else
 502                         fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
 503                 break;
 504         case Opt_ino32:
 505                 if (!result.negated)
 506                         fsopt->flags |= CEPH_MOUNT_OPT_INO32;
 507                 else
 508                         fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
 509                 break;
 510
 511         case Opt_fscache:
 512 #ifdef CONFIG_CEPH_FSCACHE
 513                 kfree(fsopt->fscache_uniq);
 514                 fsopt->fscache_uniq = NULL;
 515                 if (result.negated) {
 516                         fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
 517                 } else {
 518                         fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
 519                         fsopt->fscache_uniq = param->string;
 520                         param->string = NULL;
 521                 }
 522                 break;
 523 #else
 524                 return invalfc(fc, "fscache support is disabled");
 525 #endif
 526         case Opt_poolperm:
 527                 if (!result.negated)
 528                         fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
 529                 else
 530                         fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
 531                 break;
 532         case Opt_require_active_mds:
 533                 if (!result.negated)
 534                         fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT;
 535                 else
 536                         fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT;
 537                 break;
 538         case Opt_quotadf:
 539                 if (!result.negated)
 540                         fsopt->flags &= ~CEPH_MOUNT_OPT_NOQUOTADF;
 541                 else
 542                         fsopt->flags |= CEPH_MOUNT_OPT_NOQUOTADF;
 543                 break;
 544         case Opt_copyfrom:
 545                 if (!result.negated)
 546                         fsopt->flags &= ~CEPH_MOUNT_OPT_NOCOPYFROM;
 547                 else
 548                         fsopt->flags |= CEPH_MOUNT_OPT_NOCOPYFROM;
 549                 break;
 550         case Opt_acl:
 551                 if (!result.negated) {
 552 #ifdef CONFIG_CEPH_FS_POSIX_ACL
 553                         fc->sb_flags |= SB_POSIXACL;
 554 #else
 555                         return invalfc(fc, "POSIX ACL support is disabled");
 556 #endif
 557                 } else {
 558                         fc->sb_flags &= ~SB_POSIXACL;
 559                 }
 560                 break;
 561         case Opt_wsync:
 562                 if (!result.negated)
 563                         fsopt->flags &= ~CEPH_MOUNT_OPT_ASYNC_DIROPS;
 564                 else
 565                         fsopt->flags |= CEPH_MOUNT_OPT_ASYNC_DIROPS;
 566                 break;
 567         default:
 568                 BUG();
 569         }
 570         return 0;
 571
 572 out_of_range:
 573         return invalfc(fc, "%s out of range", param->key);
 574 }
 575
 576 static void destroy_mount_options(struct ceph_mount_options *args)
 577 {
 578         dout("destroy_mount_options %p\n", args);
 579         if (!args)
 580                 return;
 581
 582         kfree(args->snapdir_name);
 583         kfree(args->mds_namespace);
 584         kfree(args->server_path);
 585         kfree(args->fscache_uniq);
 586         kfree(args->mon_addr);
 587         kfree(args);
 588 }
 589
 590 static int strcmp_null(const char *s1, const char *s2)
 591 {
 592         if (!s1 && !s2)
 593                 return 0;
 594         if (s1 && !s2)
 595                 return -1;
 596         if (!s1 && s2)
 597                 return 1;
 598         return strcmp(s1, s2);
 599 }
 600
 601 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
 602                                  struct ceph_options *new_opt,
 603                                  struct ceph_fs_client *fsc)
 604 {
 605         struct ceph_mount_options *fsopt1 = new_fsopt;
 606         struct ceph_mount_options *fsopt2 = fsc->mount_options;
 607         int ofs = offsetof(struct ceph_mount_options, snapdir_name);
 608         int ret;
 609
 610         ret = memcmp(fsopt1, fsopt2, ofs);
 611         if (ret)
 612                 return ret;
 613
 614         ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
 615         if (ret)
 616                 return ret;
 617
 618         ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
 619         if (ret)
 620                 return ret;
 621
 622         ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
 623         if (ret)
 624                 return ret;
 625
 626         ret = strcmp_null(fsopt1->fscache_uniq, fsopt2->fscache_uniq);
 627         if (ret)
 628                 return ret;
 629
 630         ret = strcmp_null(fsopt1->mon_addr, fsopt2->mon_addr);
 631         if (ret)
 632                 return ret;
 633
 634         return ceph_compare_options(new_opt, fsc->client);
 635 }
 636
 637 /**
 638  * ceph_show_options - Show mount options in /proc/mounts
 639  * @m: seq_file to write to
 640  * @root: root of that (sub)tree
 641  */
 642 static int ceph_show_options(struct seq_file *m, struct dentry *root)
 643 {
 644         struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
 645         struct ceph_mount_options *fsopt = fsc->mount_options;
 646         size_t pos;
 647         int ret;
 648
 649         /* a comma between MNT/MS and client options */
 650         seq_putc(m, ',');
 651         pos = m->count;
 652
 653         ret = ceph_print_client_options(m, fsc->client, false);
 654         if (ret)
 655                 return ret;
 656
 657         /* retract our comma if no client options */
 658         if (m->count == pos)
 659                 m->count--;
 660
 661         if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
 662                 seq_puts(m, ",dirstat");
 663         if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
 664                 seq_puts(m, ",rbytes");
 665         if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
 666                 seq_puts(m, ",noasyncreaddir");
 667         if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
 668                 seq_puts(m, ",nodcache");
 669         if (fsopt->flags & CEPH_MOUNT_OPT_INO32)
 670                 seq_puts(m, ",ino32");
 671         if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) {
 672                 seq_show_option(m, "fsc", fsopt->fscache_uniq);
 673         }
 674         if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
 675                 seq_puts(m, ",nopoolperm");
 676         if (fsopt->flags & CEPH_MOUNT_OPT_NOQUOTADF)
 677                 seq_puts(m, ",noquotadf");
 678
 679 #ifdef CONFIG_CEPH_FS_POSIX_ACL
 680         if (root->d_sb->s_flags & SB_POSIXACL)
 681                 seq_puts(m, ",acl");
 682         else
 683                 seq_puts(m, ",noacl");
 684 #endif
 685
 686         if ((fsopt->flags & CEPH_MOUNT_OPT_NOCOPYFROM) == 0)
 687                 seq_puts(m, ",copyfrom");
 688
 689         /* dump mds_namespace when old device syntax is in use */
 690         if (fsopt->mds_namespace && !fsopt->new_dev_syntax)
 691                 seq_show_option(m, "mds_namespace", fsopt->mds_namespace);
 692
 693         if (fsopt->mon_addr)
 694                 seq_printf(m, ",mon_addr=%s", fsopt->mon_addr);
 695
 696         if (fsopt->flags & CEPH_MOUNT_OPT_CLEANRECOVER)
 697                 seq_show_option(m, "recover_session", "clean");
 698
 699         if (!(fsopt->flags & CEPH_MOUNT_OPT_ASYNC_DIROPS))
 700                 seq_puts(m, ",wsync");
 701
 702         if (fsopt->wsize != CEPH_MAX_WRITE_SIZE)
 703                 seq_printf(m, ",wsize=%u", fsopt->wsize);
 704         if (fsopt->rsize != CEPH_MAX_READ_SIZE)
 705                 seq_printf(m, ",rsize=%u", fsopt->rsize);
 706         if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
 707                 seq_printf(m, ",rasize=%u", fsopt->rasize);
 708         if (fsopt->congestion_kb != default_congestion_kb())
 709                 seq_printf(m, ",write_congestion_kb=%u", fsopt->congestion_kb);
 710         if (fsopt->caps_max)
 711                 seq_printf(m, ",caps_max=%d", fsopt->caps_max);
 712         if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
 713                 seq_printf(m, ",caps_wanted_delay_min=%u",
 714                          fsopt->caps_wanted_delay_min);
 715         if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
 716                 seq_printf(m, ",caps_wanted_delay_max=%u",
 717                            fsopt->caps_wanted_delay_max);
 718         if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
 719                 seq_printf(m, ",readdir_max_entries=%u", fsopt->max_readdir);
 720         if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
 721                 seq_printf(m, ",readdir_max_bytes=%u", fsopt->max_readdir_bytes);
 722         if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
 723                 seq_show_option(m, "snapdirname", fsopt->snapdir_name);
 724
 725         return 0;
 726 }
 727
 728 /*
 729  * handle any mon messages the standard library doesn't understand.
 730  * return error if we don't either.
 731  */
 732 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
 733 {
 734         struct ceph_fs_client *fsc = client->private;
 735         int type = le16_to_cpu(msg->hdr.type);
 736
 737         switch (type) {
 738         case CEPH_MSG_MDS_MAP:
 739                 ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
 740                 return 0;
 741         case CEPH_MSG_FS_MAP_USER:
 742                 ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
 743                 return 0;
 744         default:
 745                 return -1;
 746         }
 747 }
 748
 749 /*
 750  * create a new fs client
 751  *
 752  * Success or not, this function consumes @fsopt and @opt.
 753  */
 754 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
 755                                         struct ceph_options *opt)
 756 {
 757         struct ceph_fs_client *fsc;
 758         int err;
 759
 760         fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
 761         if (!fsc) {
 762                 err = -ENOMEM;
 763                 goto fail;
 764         }
 765
 766         fsc->client = ceph_create_client(opt, fsc);
 767         if (IS_ERR(fsc->client)) {
 768                 err = PTR_ERR(fsc->client);
 769                 goto fail;
 770         }
 771         opt = NULL; /* fsc->client now owns this */
 772
 773         fsc->client->extra_mon_dispatch = extra_mon_dispatch;
 774         ceph_set_opt(fsc->client, ABORT_ON_FULL);
 775
 776         if (!fsopt->mds_namespace) {
 777                 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
 778                                    0, true);
 779         } else {
 780                 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
 781                                    0, false);
 782         }
 783
 784         fsc->mount_options = fsopt;
 785
 786         fsc->sb = NULL;
 787         fsc->mount_state = CEPH_MOUNT_MOUNTING;
 788         fsc->filp_gen = 1;
 789         fsc->have_copy_from2 = true;
 790
 791         atomic_long_set(&fsc->writeback_count, 0);
 792
 793         err = -ENOMEM;
 794         /*
 795          * The number of concurrent works can be high but they don't need
 796          * to be processed in parallel, limit concurrency.
 797          */
 798         fsc->inode_wq = alloc_workqueue("ceph-inode", WQ_UNBOUND, 0);
 799         if (!fsc->inode_wq)
 800                 goto fail_client;
 801         fsc->cap_wq = alloc_workqueue("ceph-cap", 0, 1);
 802         if (!fsc->cap_wq)
 803                 goto fail_inode_wq;
 804
 805         spin_lock(&ceph_fsc_lock);
 806         list_add_tail(&fsc->metric_wakeup, &ceph_fsc_list);
 807         spin_unlock(&ceph_fsc_lock);
 808
 809         return fsc;
 810
 811 fail_inode_wq:
 812         destroy_workqueue(fsc->inode_wq);
 813 fail_client:
 814         ceph_destroy_client(fsc->client);
 815 fail:
 816         kfree(fsc);
 817         if (opt)
 818                 ceph_destroy_options(opt);
 819         destroy_mount_options(fsopt);
 820         return ERR_PTR(err);
 821 }
 822
 823 static void flush_fs_workqueues(struct ceph_fs_client *fsc)
 824 {
 825         flush_workqueue(fsc->inode_wq);
 826         flush_workqueue(fsc->cap_wq);
 827 }
 828
 829 static void destroy_fs_client(struct ceph_fs_client *fsc)
 830 {
 831         dout("destroy_fs_client %p\n", fsc);
 832
 833         spin_lock(&ceph_fsc_lock);
 834         list_del(&fsc->metric_wakeup);
 835         spin_unlock(&ceph_fsc_lock);
 836
 837         ceph_mdsc_destroy(fsc);
 838         destroy_workqueue(fsc->inode_wq);
 839         destroy_workqueue(fsc->cap_wq);
 840
 841         destroy_mount_options(fsc->mount_options);
 842
 843         ceph_destroy_client(fsc->client);
 844
 845         kfree(fsc);
 846         dout("destroy_fs_client %p done\n", fsc);
 847 }
 848
 849 /*
 850  * caches
 851  */
 852 struct kmem_cache *ceph_inode_cachep;
 853 struct kmem_cache *ceph_cap_cachep;
 854 struct kmem_cache *ceph_cap_flush_cachep;
 855 struct kmem_cache *ceph_dentry_cachep;
 856 struct kmem_cache *ceph_file_cachep;
 857 struct kmem_cache *ceph_dir_file_cachep;
 858 struct kmem_cache *ceph_mds_request_cachep;
 859 mempool_t *ceph_wb_pagevec_pool;
 860
 861 static void ceph_inode_init_once(void *foo)
 862 {
 863         struct ceph_inode_info *ci = foo;
 864         inode_init_once(&ci->vfs_inode);
 865 }
 866
 867 static int __init init_caches(void)
 868 {
 869         int error = -ENOMEM;
 870
 871         ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
 872                                       sizeof(struct ceph_inode_info),
 873                                       __alignof__(struct ceph_inode_info),
 874                                       SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
 875                                       SLAB_ACCOUNT, ceph_inode_init_once);
 876         if (!ceph_inode_cachep)
 877                 return -ENOMEM;
 878
 879         ceph_cap_cachep = KMEM_CACHE(ceph_cap, SLAB_MEM_SPREAD);
 880         if (!ceph_cap_cachep)
 881                 goto bad_cap;
 882         ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
 883                                            SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 884         if (!ceph_cap_flush_cachep)
 885                 goto bad_cap_flush;
 886
 887         ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
 888                                         SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 889         if (!ceph_dentry_cachep)
 890                 goto bad_dentry;
 891
 892         ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD);
 893         if (!ceph_file_cachep)
 894                 goto bad_file;
 895
 896         ceph_dir_file_cachep = KMEM_CACHE(ceph_dir_file_info, SLAB_MEM_SPREAD);
 897         if (!ceph_dir_file_cachep)
 898                 goto bad_dir_file;
 899
 900         ceph_mds_request_cachep = KMEM_CACHE(ceph_mds_request, SLAB_MEM_SPREAD);
 901         if (!ceph_mds_request_cachep)
 902                 goto bad_mds_req;
 903
 904         ceph_wb_pagevec_pool = mempool_create_kmalloc_pool(10, CEPH_MAX_WRITE_SIZE >> PAGE_SHIFT);
 905         if (!ceph_wb_pagevec_pool)
 906                 goto bad_pagevec_pool;
 907
 908         error = ceph_fscache_register();
 909         if (error)
 910                 goto bad_fscache;
 911
 912         return 0;
 913
 914 bad_fscache:
 915         kmem_cache_destroy(ceph_mds_request_cachep);
 916 bad_pagevec_pool:
 917         mempool_destroy(ceph_wb_pagevec_pool);
 918 bad_mds_req:
 919         kmem_cache_destroy(ceph_dir_file_cachep);
 920 bad_dir_file:
 921         kmem_cache_destroy(ceph_file_cachep);
 922 bad_file:
 923         kmem_cache_destroy(ceph_dentry_cachep);
 924 bad_dentry:
 925         kmem_cache_destroy(ceph_cap_flush_cachep);
 926 bad_cap_flush:
 927         kmem_cache_destroy(ceph_cap_cachep);
 928 bad_cap:
 929         kmem_cache_destroy(ceph_inode_cachep);
 930         return error;
 931 }
 932
 933 static void destroy_caches(void)
 934 {
 935         /*
 936          * Make sure all delayed rcu free inodes are flushed before we
 937          * destroy cache.
 938          */
 939         rcu_barrier();
 940
 941         kmem_cache_destroy(ceph_inode_cachep);
 942         kmem_cache_destroy(ceph_cap_cachep);
 943         kmem_cache_destroy(ceph_cap_flush_cachep);
 944         kmem_cache_destroy(ceph_dentry_cachep);
 945         kmem_cache_destroy(ceph_file_cachep);
 946         kmem_cache_destroy(ceph_dir_file_cachep);
 947         kmem_cache_destroy(ceph_mds_request_cachep);
 948         mempool_destroy(ceph_wb_pagevec_pool);
 949
 950         ceph_fscache_unregister();
 951 }
 952
 953 static void __ceph_umount_begin(struct ceph_fs_client *fsc)
 954 {
 955         ceph_osdc_abort_requests(&fsc->client->osdc, -EIO);
 956         ceph_mdsc_force_umount(fsc->mdsc);
 957         fsc->filp_gen++; // invalidate open files
 958 }
 959
 960 /*
 961  * ceph_umount_begin - initiate forced umount.  Tear down the
 962  * mount, skipping steps that may hang while waiting for server(s).
 963  */
 964 void ceph_umount_begin(struct super_block *sb)
 965 {
 966         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
 967
 968         dout("ceph_umount_begin - starting forced umount\n");
 969         if (!fsc)
 970                 return;
 971         fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
 972         __ceph_umount_begin(fsc);
 973 }
 974
 975 static const struct super_operations ceph_super_ops = {
 976         .alloc_inode    = ceph_alloc_inode,
 977         .free_inode     = ceph_free_inode,
 978         .write_inode    = ceph_write_inode,
 979         .drop_inode     = generic_delete_inode,
 980         .evict_inode    = ceph_evict_inode,
 981         .sync_fs        = ceph_sync_fs,
 982         .put_super      = ceph_put_super,
 983         .show_options   = ceph_show_options,
 984         .statfs         = ceph_statfs,
 985         .umount_begin   = ceph_umount_begin,
 986 };
 987
 988 /*
 989  * Bootstrap mount by opening the root directory.  Note the mount
 990  * @started time from caller, and time out if this takes too long.
 991  */
 992 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
 993                                        const char *path,
 994                                        unsigned long started)
 995 {
 996         struct ceph_mds_client *mdsc = fsc->mdsc;
 997         struct ceph_mds_request *req = NULL;
 998         int err;
 999         struct dentry *root;
1000
1001         /* open dir */
1002         dout("open_root_inode opening '%s'\n", path);
1003         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1004         if (IS_ERR(req))
1005                 return ERR_CAST(req);
1006         req->r_path1 = kstrdup(path, GFP_NOFS);
1007         if (!req->r_path1) {
1008                 root = ERR_PTR(-ENOMEM);
1009                 goto out;
1010         }
1011
1012         req->r_ino1.ino = CEPH_INO_ROOT;
1013         req->r_ino1.snap = CEPH_NOSNAP;
1014         req->r_started = started;
1015         req->r_timeout = fsc->client->options->mount_timeout;
1016         req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
1017         req->r_num_caps = 2;
1018         err = ceph_mdsc_do_request(mdsc, NULL, req);
1019         if (err == 0) {
1020                 struct inode *inode = req->r_target_inode;
1021                 req->r_target_inode = NULL;
1022                 dout("open_root_inode success\n");
1023                 root = d_make_root(inode);
1024                 if (!root) {
1025                         root = ERR_PTR(-ENOMEM);
1026                         goto out;
1027                 }
1028                 dout("open_root_inode success, root dentry is %p\n", root);
1029         } else {
1030                 root = ERR_PTR(err);
1031         }
1032 out:
1033         ceph_mdsc_put_request(req);
1034         return root;
1035 }
1036
1037 /*
1038  * mount: join the ceph cluster, and open root directory.
1039  */
1040 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
1041                                       struct fs_context *fc)
1042 {
1043         int err;
1044         unsigned long started = jiffies;  /* note the start time */
1045         struct dentry *root;
1046
1047         dout("mount start %p\n", fsc);
1048         mutex_lock(&fsc->client->mount_mutex);
1049
1050         if (!fsc->sb->s_root) {
1051                 const char *path = fsc->mount_options->server_path ?
1052                                      fsc->mount_options->server_path + 1 : "";
1053
1054                 err = __ceph_open_session(fsc->client, started);
1055                 if (err < 0)
1056                         goto out;
1057
1058                 /* setup fscache */
1059                 if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
1060                         err = ceph_fscache_register_fs(fsc, fc);
1061                         if (err < 0)
1062                                 goto out;
1063                 }
1064
1065                 dout("mount opening path '%s'\n", path);
1066
1067                 ceph_fs_debugfs_init(fsc);
1068
1069                 root = open_root_dentry(fsc, path, started);
1070                 if (IS_ERR(root)) {
1071                         err = PTR_ERR(root);
1072                         goto out;
1073                 }
1074                 fsc->sb->s_root = dget(root);
1075         } else {
1076                 root = dget(fsc->sb->s_root);
1077         }
1078
1079         fsc->mount_state = CEPH_MOUNT_MOUNTED;
1080         dout("mount success\n");
1081         mutex_unlock(&fsc->client->mount_mutex);
1082         return root;
1083
1084 out:
1085         mutex_unlock(&fsc->client->mount_mutex);
1086         return ERR_PTR(err);
1087 }
1088
1089 static int ceph_set_super(struct super_block *s, struct fs_context *fc)
1090 {
1091         struct ceph_fs_client *fsc = s->s_fs_info;
1092         int ret;
1093
1094         dout("set_super %p\n", s);
1095
1096         s->s_maxbytes = MAX_LFS_FILESIZE;
1097
1098         s->s_xattr = ceph_xattr_handlers;
1099         fsc->sb = s;
1100         fsc->max_file_size = 1ULL << 40; /* temp value until we get mdsmap */
1101
1102         s->s_op = &ceph_super_ops;
1103         s->s_d_op = &ceph_dentry_ops;
1104         s->s_export_op = &ceph_export_ops;
1105
1106         s->s_time_gran = 1;
1107         s->s_time_min = 0;
1108         s->s_time_max = U32_MAX;
1109
1110         ret = set_anon_super_fc(s, fc);
1111         if (ret != 0)
1112                 fsc->sb = NULL;
1113         return ret;
1114 }
1115
1116 /*
1117  * share superblock if same fs AND options
1118  */
1119 static int ceph_compare_super(struct super_block *sb, struct fs_context *fc)
1120 {
1121         struct ceph_fs_client *new = fc->s_fs_info;
1122         struct ceph_mount_options *fsopt = new->mount_options;
1123         struct ceph_options *opt = new->client->options;
1124         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1125
1126         dout("ceph_compare_super %p\n", sb);
1127
1128         if (compare_mount_options(fsopt, opt, fsc)) {
1129                 dout("monitor(s)/mount options don't match\n");
1130                 return 0;
1131         }
1132         if ((opt->flags & CEPH_OPT_FSID) &&
1133             ceph_fsid_compare(&opt->fsid, &fsc->client->fsid)) {
1134                 dout("fsid doesn't match\n");
1135                 return 0;
1136         }
1137         if (fc->sb_flags != (sb->s_flags & ~SB_BORN)) {
1138                 dout("flags differ\n");
1139                 return 0;
1140         }
1141
1142         if (fsc->blocklisted && !ceph_test_mount_opt(fsc, CLEANRECOVER)) {
1143                 dout("client is blocklisted (and CLEANRECOVER is not set)\n");
1144                 return 0;
1145         }
1146
1147         if (fsc->mount_state == CEPH_MOUNT_SHUTDOWN) {
1148                 dout("client has been forcibly unmounted\n");
1149                 return 0;
1150         }
1151
1152         return 1;
1153 }
1154
1155 /*
1156  * construct our own bdi so we can control readahead, etc.
1157  */
1158 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1159
1160 static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
1161 {
1162         int err;
1163
1164         err = super_setup_bdi_name(sb, "ceph-%ld",
1165                                    atomic_long_inc_return(&bdi_seq));
1166         if (err)
1167                 return err;
1168
1169         /* set ra_pages based on rasize mount option? */
1170         sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
1171
1172         /* set io_pages based on max osd read size */
1173         sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
1174
1175         return 0;
1176 }
1177
1178 static int ceph_get_tree(struct fs_context *fc)
1179 {
1180         struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1181         struct ceph_mount_options *fsopt = pctx->opts;
1182         struct super_block *sb;
1183         struct ceph_fs_client *fsc;
1184         struct dentry *res;
1185         int (*compare_super)(struct super_block *, struct fs_context *) =
1186                 ceph_compare_super;
1187         int err;
1188
1189         dout("ceph_get_tree\n");
1190
1191         if (!fc->source)
1192                 return invalfc(fc, "No source");
1193         if (fsopt->new_dev_syntax && !fsopt->mon_addr)
1194                 return invalfc(fc, "No monitor address");
1195
1196         /* create client (which we may/may not use) */
1197         fsc = create_fs_client(pctx->opts, pctx->copts);
1198         pctx->opts = NULL;
1199         pctx->copts = NULL;
1200         if (IS_ERR(fsc)) {
1201                 err = PTR_ERR(fsc);
1202                 goto out_final;
1203         }
1204
1205         err = ceph_mdsc_init(fsc);
1206         if (err < 0)
1207                 goto out;
1208
1209         if (ceph_test_opt(fsc->client, NOSHARE))
1210                 compare_super = NULL;
1211
1212         fc->s_fs_info = fsc;
1213         sb = sget_fc(fc, compare_super, ceph_set_super);
1214         fc->s_fs_info = NULL;
1215         if (IS_ERR(sb)) {
1216                 err = PTR_ERR(sb);
1217                 goto out;
1218         }
1219
1220         if (ceph_sb_to_client(sb) != fsc) {
1221                 destroy_fs_client(fsc);
1222                 fsc = ceph_sb_to_client(sb);
1223                 dout("get_sb got existing client %p\n", fsc);
1224         } else {
1225                 dout("get_sb using new client %p\n", fsc);
1226                 err = ceph_setup_bdi(sb, fsc);
1227                 if (err < 0)
1228                         goto out_splat;
1229         }
1230
1231         res = ceph_real_mount(fsc, fc);
1232         if (IS_ERR(res)) {
1233                 err = PTR_ERR(res);
1234                 goto out_splat;
1235         }
1236         dout("root %p inode %p ino %llx.%llx\n", res,
1237              d_inode(res), ceph_vinop(d_inode(res)));
1238         fc->root = fsc->sb->s_root;
1239         return 0;
1240
1241 out_splat:
1242         if (!ceph_mdsmap_is_cluster_available(fsc->mdsc->mdsmap)) {
1243                 pr_info("No mds server is up or the cluster is laggy\n");
1244                 err = -EHOSTUNREACH;
1245         }
1246
1247         ceph_mdsc_close_sessions(fsc->mdsc);
1248         deactivate_locked_super(sb);
1249         goto out_final;
1250
1251 out:
1252         destroy_fs_client(fsc);
1253 out_final:
1254         dout("ceph_get_tree fail %d\n", err);
1255         return err;
1256 }
1257
1258 static void ceph_free_fc(struct fs_context *fc)
1259 {
1260         struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1261
1262         if (pctx) {
1263                 destroy_mount_options(pctx->opts);
1264                 ceph_destroy_options(pctx->copts);
1265                 kfree(pctx);
1266         }
1267 }
1268
1269 static int ceph_reconfigure_fc(struct fs_context *fc)
1270 {
1271         struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1272         struct ceph_mount_options *fsopt = pctx->opts;
1273         struct ceph_fs_client *fsc = ceph_sb_to_client(fc->root->d_sb);
1274
1275         if (fsopt->flags & CEPH_MOUNT_OPT_ASYNC_DIROPS)
1276                 ceph_set_mount_opt(fsc, ASYNC_DIROPS);
1277         else
1278                 ceph_clear_mount_opt(fsc, ASYNC_DIROPS);
1279
1280         sync_filesystem(fc->root->d_sb);
1281         return 0;
1282 }
1283
1284 static const struct fs_context_operations ceph_context_ops = {
1285         .free           = ceph_free_fc,
1286         .parse_param    = ceph_parse_mount_param,
1287         .get_tree       = ceph_get_tree,
1288         .reconfigure    = ceph_reconfigure_fc,
1289 };
1290
1291 /*
1292  * Set up the filesystem mount context.
1293  */
1294 static int ceph_init_fs_context(struct fs_context *fc)
1295 {
1296         struct ceph_parse_opts_ctx *pctx;
1297         struct ceph_mount_options *fsopt;
1298
1299         pctx = kzalloc(sizeof(*pctx), GFP_KERNEL);
1300         if (!pctx)
1301                 return -ENOMEM;
1302
1303         pctx->copts = ceph_alloc_options();
1304         if (!pctx->copts)
1305                 goto nomem;
1306
1307         pctx->opts = kzalloc(sizeof(*pctx->opts), GFP_KERNEL);
1308         if (!pctx->opts)
1309                 goto nomem;
1310
1311         fsopt = pctx->opts;
1312         fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
1313
1314         fsopt->wsize = CEPH_MAX_WRITE_SIZE;
1315         fsopt->rsize = CEPH_MAX_READ_SIZE;
1316         fsopt->rasize = CEPH_RASIZE_DEFAULT;
1317         fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
1318         if (!fsopt->snapdir_name)
1319                 goto nomem;
1320
1321         fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
1322         fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
1323         fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
1324         fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
1325         fsopt->congestion_kb = default_congestion_kb();
1326
1327 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1328         fc->sb_flags |= SB_POSIXACL;
1329 #endif
1330
1331         fc->fs_private = pctx;
1332         fc->ops = &ceph_context_ops;
1333         return 0;
1334
1335 nomem:
1336         destroy_mount_options(pctx->opts);
1337         ceph_destroy_options(pctx->copts);
1338         kfree(pctx);
1339         return -ENOMEM;
1340 }
1341
1342 static void ceph_kill_sb(struct super_block *s)
1343 {
1344         struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1345
1346         dout("kill_sb %p\n", s);
1347
1348         ceph_mdsc_pre_umount(fsc->mdsc);
1349         flush_fs_workqueues(fsc);
1350
1351         kill_anon_super(s);
1352
1353         fsc->client->extra_mon_dispatch = NULL;
1354         ceph_fs_debugfs_cleanup(fsc);
1355
1356         ceph_fscache_unregister_fs(fsc);
1357
1358         destroy_fs_client(fsc);
1359 }
1360
1361 static struct file_system_type ceph_fs_type = {
1362         .owner          = THIS_MODULE,
1363         .name           = "ceph",
1364         .init_fs_context = ceph_init_fs_context,
1365         .kill_sb        = ceph_kill_sb,
1366         .fs_flags       = FS_RENAME_DOES_D_MOVE,
1367 };
1368 MODULE_ALIAS_FS("ceph");
1369
1370 int ceph_force_reconnect(struct super_block *sb)
1371 {
1372         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1373         int err = 0;
1374
1375         fsc->mount_state = CEPH_MOUNT_RECOVER;
1376         __ceph_umount_begin(fsc);
1377
1378         /* Make sure all page caches get invalidated.
1379          * see remove_session_caps_cb() */
1380         flush_workqueue(fsc->inode_wq);
1381
1382         /* In case that we were blocklisted. This also reset
1383          * all mon/osd connections */
1384         ceph_reset_client_addr(fsc->client);
1385
1386         ceph_osdc_clear_abort_err(&fsc->client->osdc);
1387
1388         fsc->blocklisted = false;
1389         fsc->mount_state = CEPH_MOUNT_MOUNTED;
1390
1391         if (sb->s_root) {
1392                 err = __ceph_do_getattr(d_inode(sb->s_root), NULL,
1393                                         CEPH_STAT_CAP_INODE, true);
1394         }
1395         return err;
1396 }
1397
1398 static int __init init_ceph(void)
1399 {
1400         int ret = init_caches();
1401         if (ret)
1402                 goto out;
1403
1404         ceph_flock_init();
1405         ret = register_filesystem(&ceph_fs_type);
1406         if (ret)
1407                 goto out_caches;
1408
1409         pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1410
1411         return 0;
1412
1413 out_caches:
1414         destroy_caches();
1415 out:
1416         return ret;
1417 }
1418
1419 static void __exit exit_ceph(void)
1420 {
1421         dout("exit_ceph\n");
1422         unregister_filesystem(&ceph_fs_type);
1423         destroy_caches();
1424 }
1425
1426 static int param_set_metrics(const char *val, const struct kernel_param *kp)
1427 {
1428         struct ceph_fs_client *fsc;
1429         int ret;
1430
1431         ret = param_set_bool(val, kp);
1432         if (ret) {
1433                 pr_err("Failed to parse sending metrics switch value '%s'\n",
1434                        val);
1435                 return ret;
1436         } else if (!disable_send_metrics) {
1437                 // wake up all the mds clients
1438                 spin_lock(&ceph_fsc_lock);
1439                 list_for_each_entry(fsc, &ceph_fsc_list, metric_wakeup) {
1440                         metric_schedule_delayed(&fsc->mdsc->metric);
1441                 }
1442                 spin_unlock(&ceph_fsc_lock);
1443         }
1444
1445         return 0;
1446 }
1447
1448 static const struct kernel_param_ops param_ops_metrics = {
1449         .set = param_set_metrics,
1450         .get = param_get_bool,
1451 };
1452
1453 bool disable_send_metrics = false;
1454 module_param_cb(disable_send_metrics, &param_ops_metrics, &disable_send_metrics, 0644);
1455 MODULE_PARM_DESC(disable_send_metrics, "Enable sending perf metrics to ceph cluster (default: on)");
1456
1457 module_init(init_ceph);
1458 module_exit(exit_ceph);
1459
1460 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1461 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1462 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1463 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1464 MODULE_LICENSE("GPL");