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 /*
  31  * Ceph superblock operations
  32  *
  33  * Handle the basics of mounting, unmounting.
  34  */
  35
  36 /*
  37  * super ops
  38  */
  39 static void ceph_put_super(struct super_block *s)
  40 {
  41         struct ceph_fs_client *fsc = ceph_sb_to_client(s);
  42
  43         dout("put_super\n");
  44         ceph_mdsc_close_sessions(fsc->mdsc);
  45 }
  46
  47 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
  48 {
  49         struct ceph_fs_client *fsc = ceph_inode_to_client(d_inode(dentry));
  50         struct ceph_mon_client *monc = &fsc->client->monc;
  51         struct ceph_statfs st;
  52         u64 fsid;
  53         int err;
  54         u64 data_pool;
  55
  56         if (fsc->mdsc->mdsmap->m_num_data_pg_pools == 1) {
  57                 data_pool = fsc->mdsc->mdsmap->m_data_pg_pools[0];
  58         } else {
  59                 data_pool = CEPH_NOPOOL;
  60         }
  61
  62         dout("statfs\n");
  63         err = ceph_monc_do_statfs(monc, data_pool, &st);
  64         if (err < 0)
  65                 return err;
  66
  67         /* fill in kstatfs */
  68         buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
  69
  70         /*
  71          * express utilization in terms of large blocks to avoid
  72          * overflow on 32-bit machines.
  73          *
  74          * NOTE: for the time being, we make bsize == frsize to humor
  75          * not-yet-ancient versions of glibc that are broken.
  76          * Someday, we will probably want to report a real block
  77          * size...  whatever that may mean for a network file system!
  78          */
  79         buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
  80         buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
  81
  82         /*
  83          * By default use root quota for stats; fallback to overall filesystem
  84          * usage if using 'noquotadf' mount option or if the root dir doesn't
  85          * have max_bytes quota set.
  86          */
  87         if (ceph_test_mount_opt(fsc, NOQUOTADF) ||
  88             !ceph_quota_update_statfs(fsc, buf)) {
  89                 buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
  90                 buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
  91                 buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
  92         }
  93
  94         buf->f_files = le64_to_cpu(st.num_objects);
  95         buf->f_ffree = -1;
  96         buf->f_namelen = NAME_MAX;
  97
  98         /* Must convert the fsid, for consistent values across arches */
  99         mutex_lock(&monc->mutex);
 100         fsid = le64_to_cpu(*(__le64 *)(&monc->monmap->fsid)) ^
 101                le64_to_cpu(*((__le64 *)&monc->monmap->fsid + 1));
 102         mutex_unlock(&monc->mutex);
 103
 104         buf->f_fsid.val[0] = fsid & 0xffffffff;
 105         buf->f_fsid.val[1] = fsid >> 32;
 106
 107         return 0;
 108 }
 109
 110 static int ceph_sync_fs(struct super_block *sb, int wait)
 111 {
 112         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
 113
 114         if (!wait) {
 115                 dout("sync_fs (non-blocking)\n");
 116                 ceph_flush_dirty_caps(fsc->mdsc);
 117                 dout("sync_fs (non-blocking) done\n");
 118                 return 0;
 119         }
 120
 121         dout("sync_fs (blocking)\n");
 122         ceph_osdc_sync(&fsc->client->osdc);
 123         ceph_mdsc_sync(fsc->mdsc);
 124         dout("sync_fs (blocking) done\n");
 125         return 0;
 126 }
 127
 128 /*
 129  * mount options
 130  */
 131 enum {
 132         Opt_wsize,
 133         Opt_rsize,
 134         Opt_rasize,
 135         Opt_caps_wanted_delay_min,
 136         Opt_caps_wanted_delay_max,
 137         Opt_caps_max,
 138         Opt_readdir_max_entries,
 139         Opt_readdir_max_bytes,
 140         Opt_congestion_kb,
 141         /* int args above */
 142         Opt_snapdirname,
 143         Opt_mds_namespace,
 144         Opt_recover_session,
 145         Opt_source,
 146         /* string args above */
 147         Opt_dirstat,
 148         Opt_rbytes,
 149         Opt_asyncreaddir,
 150         Opt_dcache,
 151         Opt_ino32,
 152         Opt_fscache,
 153         Opt_poolperm,
 154         Opt_require_active_mds,
 155         Opt_acl,
 156         Opt_quotadf,
 157         Opt_copyfrom,
 158 };
 159
 160 enum ceph_recover_session_mode {
 161         ceph_recover_session_no,
 162         ceph_recover_session_clean
 163 };
 164
 165 static const struct fs_parameter_enum ceph_mount_param_enums[] = {
 166         { Opt_recover_session,  "no",           ceph_recover_session_no },
 167         { Opt_recover_session,  "clean",        ceph_recover_session_clean },
 168         {}
 169 };
 170
 171 static const struct fs_parameter_spec ceph_mount_param_specs[] = {
 172         fsparam_flag_no ("acl",                         Opt_acl),
 173         fsparam_flag_no ("asyncreaddir",                Opt_asyncreaddir),
 174         fsparam_s32     ("caps_max",                    Opt_caps_max),
 175         fsparam_u32     ("caps_wanted_delay_max",       Opt_caps_wanted_delay_max),
 176         fsparam_u32     ("caps_wanted_delay_min",       Opt_caps_wanted_delay_min),
 177         fsparam_u32     ("write_congestion_kb",         Opt_congestion_kb),
 178         fsparam_flag_no ("copyfrom",                    Opt_copyfrom),
 179         fsparam_flag_no ("dcache",                      Opt_dcache),
 180         fsparam_flag_no ("dirstat",                     Opt_dirstat),
 181         __fsparam       (fs_param_is_string, "fsc",     Opt_fscache,
 182                          fs_param_neg_with_no | fs_param_v_optional),
 183         fsparam_flag_no ("ino32",                       Opt_ino32),
 184         fsparam_string  ("mds_namespace",               Opt_mds_namespace),
 185         fsparam_flag_no ("poolperm",                    Opt_poolperm),
 186         fsparam_flag_no ("quotadf",                     Opt_quotadf),
 187         fsparam_u32     ("rasize",                      Opt_rasize),
 188         fsparam_flag_no ("rbytes",                      Opt_rbytes),
 189         fsparam_u32     ("readdir_max_bytes",           Opt_readdir_max_bytes),
 190         fsparam_u32     ("readdir_max_entries",         Opt_readdir_max_entries),
 191         fsparam_enum    ("recover_session",             Opt_recover_session),
 192         fsparam_flag_no ("require_active_mds",          Opt_require_active_mds),
 193         fsparam_u32     ("rsize",                       Opt_rsize),
 194         fsparam_string  ("snapdirname",                 Opt_snapdirname),
 195         fsparam_string  ("source",                      Opt_source),
 196         fsparam_u32     ("wsize",                       Opt_wsize),
 197         {}
 198 };
 199
 200 static const struct fs_parameter_description ceph_mount_parameters = {
 201         .name           = "ceph",
 202         .specs          = ceph_mount_param_specs,
 203         .enums          = ceph_mount_param_enums,
 204 };
 205
 206 struct ceph_parse_opts_ctx {
 207         struct ceph_options             *copts;
 208         struct ceph_mount_options       *opts;
 209 };
 210
 211 /*
 212  * Parse the source parameter.  Distinguish the server list from the path.
 213  *
 214  * The source will look like:
 215  *     <server_spec>[,<server_spec>...]:[<path>]
 216  * where
 217  *     <server_spec> is <ip>[:<port>]
 218  *     <path> is optional, but if present must begin with '/'
 219  */
 220 static int ceph_parse_source(struct fs_parameter *param, struct fs_context *fc)
 221 {
 222         struct ceph_parse_opts_ctx *pctx = fc->fs_private;
 223         struct ceph_mount_options *fsopt = pctx->opts;
 224         char *dev_name = param->string, *dev_name_end;
 225         int ret;
 226
 227         dout("%s '%s'\n", __func__, dev_name);
 228         if (!dev_name || !*dev_name)
 229                 return invalf(fc, "ceph: Empty source");
 230
 231         dev_name_end = strchr(dev_name, '/');
 232         if (dev_name_end) {
 233                 kfree(fsopt->server_path);
 234
 235                 /*
 236                  * The server_path will include the whole chars from userland
 237                  * including the leading '/'.
 238                  */
 239                 fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
 240                 if (!fsopt->server_path)
 241                         return -ENOMEM;
 242         } else {
 243                 dev_name_end = dev_name + strlen(dev_name);
 244         }
 245
 246         dev_name_end--;         /* back up to ':' separator */
 247         if (dev_name_end < dev_name || *dev_name_end != ':')
 248                 return invalf(fc, "ceph: No path or : separator in source");
 249
 250         dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
 251         if (fsopt->server_path)
 252                 dout("server path '%s'\n", fsopt->server_path);
 253
 254         ret = ceph_parse_mon_ips(param->string, dev_name_end - dev_name,
 255                                  pctx->copts, fc);
 256         if (ret)
 257                 return ret;
 258
 259         fc->source = param->string;
 260         param->string = NULL;
 261         return 0;
 262 }
 263
 264 static int ceph_parse_mount_param(struct fs_context *fc,
 265                                   struct fs_parameter *param)
 266 {
 267         struct ceph_parse_opts_ctx *pctx = fc->fs_private;
 268         struct ceph_mount_options *fsopt = pctx->opts;
 269         struct fs_parse_result result;
 270         unsigned int mode;
 271         int token, ret;
 272
 273         ret = ceph_parse_param(param, pctx->copts, fc);
 274         if (ret != -ENOPARAM)
 275                 return ret;
 276
 277         token = fs_parse(fc, &ceph_mount_parameters, param, &result);
 278         dout("%s fs_parse '%s' token %d\n", __func__, param->key, token);
 279         if (token < 0)
 280                 return token;
 281
 282         switch (token) {
 283         case Opt_snapdirname:
 284                 kfree(fsopt->snapdir_name);
 285                 fsopt->snapdir_name = param->string;
 286                 param->string = NULL;
 287                 break;
 288         case Opt_mds_namespace:
 289                 kfree(fsopt->mds_namespace);
 290                 fsopt->mds_namespace = param->string;
 291                 param->string = NULL;
 292                 break;
 293         case Opt_recover_session:
 294                 mode = result.uint_32;
 295                 if (mode == ceph_recover_session_no)
 296                         fsopt->flags &= ~CEPH_MOUNT_OPT_CLEANRECOVER;
 297                 else if (mode == ceph_recover_session_clean)
 298                         fsopt->flags |= CEPH_MOUNT_OPT_CLEANRECOVER;
 299                 else
 300                         BUG();
 301                 break;
 302         case Opt_source:
 303                 if (fc->source)
 304                         return invalf(fc, "ceph: Multiple sources specified");
 305                 return ceph_parse_source(param, fc);
 306         case Opt_wsize:
 307                 if (result.uint_32 < PAGE_SIZE ||
 308                     result.uint_32 > CEPH_MAX_WRITE_SIZE)
 309                         goto out_of_range;
 310                 fsopt->wsize = ALIGN(result.uint_32, PAGE_SIZE);
 311                 break;
 312         case Opt_rsize:
 313                 if (result.uint_32 < PAGE_SIZE ||
 314                     result.uint_32 > CEPH_MAX_READ_SIZE)
 315                         goto out_of_range;
 316                 fsopt->rsize = ALIGN(result.uint_32, PAGE_SIZE);
 317                 break;
 318         case Opt_rasize:
 319                 fsopt->rasize = ALIGN(result.uint_32, PAGE_SIZE);
 320                 break;
 321         case Opt_caps_wanted_delay_min:
 322                 if (result.uint_32 < 1)
 323                         goto out_of_range;
 324                 fsopt->caps_wanted_delay_min = result.uint_32;
 325                 break;
 326         case Opt_caps_wanted_delay_max:
 327                 if (result.uint_32 < 1)
 328                         goto out_of_range;
 329                 fsopt->caps_wanted_delay_max = result.uint_32;
 330                 break;
 331         case Opt_caps_max:
 332                 if (result.int_32 < 0)
 333                         goto out_of_range;
 334                 fsopt->caps_max = result.int_32;
 335                 break;
 336         case Opt_readdir_max_entries:
 337                 if (result.uint_32 < 1)
 338                         goto out_of_range;
 339                 fsopt->max_readdir = result.uint_32;
 340                 break;
 341         case Opt_readdir_max_bytes:
 342                 if (result.uint_32 < PAGE_SIZE && result.uint_32 != 0)
 343                         goto out_of_range;
 344                 fsopt->max_readdir_bytes = result.uint_32;
 345                 break;
 346         case Opt_congestion_kb:
 347                 if (result.uint_32 < 1024) /* at least 1M */
 348                         goto out_of_range;
 349                 fsopt->congestion_kb = result.uint_32;
 350                 break;
 351         case Opt_dirstat:
 352                 if (!result.negated)
 353                         fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
 354                 else
 355                         fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
 356                 break;
 357         case Opt_rbytes:
 358                 if (!result.negated)
 359                         fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
 360                 else
 361                         fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
 362                 break;
 363         case Opt_asyncreaddir:
 364                 if (!result.negated)
 365                         fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
 366                 else
 367                         fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
 368                 break;
 369         case Opt_dcache:
 370                 if (!result.negated)
 371                         fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
 372                 else
 373                         fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
 374                 break;
 375         case Opt_ino32:
 376                 if (!result.negated)
 377                         fsopt->flags |= CEPH_MOUNT_OPT_INO32;
 378                 else
 379                         fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
 380                 break;
 381
 382         case Opt_fscache:
 383 #ifdef CONFIG_CEPH_FSCACHE
 384                 kfree(fsopt->fscache_uniq);
 385                 fsopt->fscache_uniq = NULL;
 386                 if (result.negated) {
 387                         fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
 388                 } else {
 389                         fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
 390                         fsopt->fscache_uniq = param->string;
 391                         param->string = NULL;
 392                 }
 393                 break;
 394 #else
 395                 return invalf(fc, "ceph: fscache support is disabled");
 396 #endif
 397         case Opt_poolperm:
 398                 if (!result.negated)
 399                         fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
 400                 else
 401                         fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
 402                 break;
 403         case Opt_require_active_mds:
 404                 if (!result.negated)
 405                         fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT;
 406                 else
 407                         fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT;
 408                 break;
 409         case Opt_quotadf:
 410                 if (!result.negated)
 411                         fsopt->flags &= ~CEPH_MOUNT_OPT_NOQUOTADF;
 412                 else
 413                         fsopt->flags |= CEPH_MOUNT_OPT_NOQUOTADF;
 414                 break;
 415         case Opt_copyfrom:
 416                 if (!result.negated)
 417                         fsopt->flags &= ~CEPH_MOUNT_OPT_NOCOPYFROM;
 418                 else
 419                         fsopt->flags |= CEPH_MOUNT_OPT_NOCOPYFROM;
 420                 break;
 421         case Opt_acl:
 422                 if (!result.negated) {
 423 #ifdef CONFIG_CEPH_FS_POSIX_ACL
 424                         fc->sb_flags |= SB_POSIXACL;
 425 #else
 426                         return invalf(fc, "ceph: POSIX ACL support is disabled");
 427 #endif
 428                 } else {
 429                         fc->sb_flags &= ~SB_POSIXACL;
 430                 }
 431                 break;
 432         default:
 433                 BUG();
 434         }
 435         return 0;
 436
 437 out_of_range:
 438         return invalf(fc, "ceph: %s out of range", param->key);
 439 }
 440
 441 static void destroy_mount_options(struct ceph_mount_options *args)
 442 {
 443         dout("destroy_mount_options %p\n", args);
 444         if (!args)
 445                 return;
 446
 447         kfree(args->snapdir_name);
 448         kfree(args->mds_namespace);
 449         kfree(args->server_path);
 450         kfree(args->fscache_uniq);
 451         kfree(args);
 452 }
 453
 454 static int strcmp_null(const char *s1, const char *s2)
 455 {
 456         if (!s1 && !s2)
 457                 return 0;
 458         if (s1 && !s2)
 459                 return -1;
 460         if (!s1 && s2)
 461                 return 1;
 462         return strcmp(s1, s2);
 463 }
 464
 465 /**
 466  * path_remove_extra_slash - Remove the extra slashes in the server path
 467  * @server_path: the server path and could be NULL
 468  *
 469  * Return NULL if the path is NULL or only consists of "/", or a string
 470  * without any extra slashes including the leading slash(es) and the
 471  * slash(es) at the end of the server path, such as:
 472  * "//dir1////dir2///" --> "dir1/dir2"
 473  */
 474 static char *path_remove_extra_slash(const char *server_path)
 475 {
 476         const char *path = server_path;
 477         const char *cur, *end;
 478         char *buf, *p;
 479         int len;
 480
 481         /* if the server path is omitted */
 482         if (!path)
 483                 return NULL;
 484
 485         /* remove all the leading slashes */
 486         while (*path == '/')
 487                 path++;
 488
 489         /* if the server path only consists of slashes */
 490         if (*path == '\0')
 491                 return NULL;
 492
 493         len = strlen(path);
 494
 495         buf = kmalloc(len + 1, GFP_KERNEL);
 496         if (!buf)
 497                 return ERR_PTR(-ENOMEM);
 498
 499         end = path + len;
 500         p = buf;
 501         do {
 502                 cur = strchr(path, '/');
 503                 if (!cur)
 504                         cur = end;
 505
 506                 len = cur - path;
 507
 508                 /* including one '/' */
 509                 if (cur != end)
 510                         len += 1;
 511
 512                 memcpy(p, path, len);
 513                 p += len;
 514
 515                 while (cur <= end && *cur == '/')
 516                         cur++;
 517                 path = cur;
 518         } while (path < end);
 519
 520         *p = '\0';
 521
 522         /*
 523          * remove the last slash if there has and just to make sure that
 524          * we will get something like "dir1/dir2"
 525          */
 526         if (*(--p) == '/')
 527                 *p = '\0';
 528
 529         return buf;
 530 }
 531
 532 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
 533                                  struct ceph_options *new_opt,
 534                                  struct ceph_fs_client *fsc)
 535 {
 536         struct ceph_mount_options *fsopt1 = new_fsopt;
 537         struct ceph_mount_options *fsopt2 = fsc->mount_options;
 538         int ofs = offsetof(struct ceph_mount_options, snapdir_name);
 539         char *p1, *p2;
 540         int ret;
 541
 542         ret = memcmp(fsopt1, fsopt2, ofs);
 543         if (ret)
 544                 return ret;
 545
 546         ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
 547         if (ret)
 548                 return ret;
 549         ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
 550         if (ret)
 551                 return ret;
 552
 553         p1 = path_remove_extra_slash(fsopt1->server_path);
 554         if (IS_ERR(p1))
 555                 return PTR_ERR(p1);
 556         p2 = path_remove_extra_slash(fsopt2->server_path);
 557         if (IS_ERR(p2)) {
 558                 kfree(p1);
 559                 return PTR_ERR(p2);
 560         }
 561         ret = strcmp_null(p1, p2);
 562         kfree(p1);
 563         kfree(p2);
 564         if (ret)
 565                 return ret;
 566
 567         ret = strcmp_null(fsopt1->fscache_uniq, fsopt2->fscache_uniq);
 568         if (ret)
 569                 return ret;
 570
 571         return ceph_compare_options(new_opt, fsc->client);
 572 }
 573
 574 /**
 575  * ceph_show_options - Show mount options in /proc/mounts
 576  * @m: seq_file to write to
 577  * @root: root of that (sub)tree
 578  */
 579 static int ceph_show_options(struct seq_file *m, struct dentry *root)
 580 {
 581         struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
 582         struct ceph_mount_options *fsopt = fsc->mount_options;
 583         size_t pos;
 584         int ret;
 585
 586         /* a comma between MNT/MS and client options */
 587         seq_putc(m, ',');
 588         pos = m->count;
 589
 590         ret = ceph_print_client_options(m, fsc->client, false);
 591         if (ret)
 592                 return ret;
 593
 594         /* retract our comma if no client options */
 595         if (m->count == pos)
 596                 m->count--;
 597
 598         if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
 599                 seq_puts(m, ",dirstat");
 600         if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
 601                 seq_puts(m, ",rbytes");
 602         if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
 603                 seq_puts(m, ",noasyncreaddir");
 604         if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
 605                 seq_puts(m, ",nodcache");
 606         if (fsopt->flags & CEPH_MOUNT_OPT_INO32)
 607                 seq_puts(m, ",ino32");
 608         if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) {
 609                 seq_show_option(m, "fsc", fsopt->fscache_uniq);
 610         }
 611         if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
 612                 seq_puts(m, ",nopoolperm");
 613         if (fsopt->flags & CEPH_MOUNT_OPT_NOQUOTADF)
 614                 seq_puts(m, ",noquotadf");
 615
 616 #ifdef CONFIG_CEPH_FS_POSIX_ACL
 617         if (root->d_sb->s_flags & SB_POSIXACL)
 618                 seq_puts(m, ",acl");
 619         else
 620                 seq_puts(m, ",noacl");
 621 #endif
 622
 623         if ((fsopt->flags & CEPH_MOUNT_OPT_NOCOPYFROM) == 0)
 624                 seq_puts(m, ",copyfrom");
 625
 626         if (fsopt->mds_namespace)
 627                 seq_show_option(m, "mds_namespace", fsopt->mds_namespace);
 628
 629         if (fsopt->flags & CEPH_MOUNT_OPT_CLEANRECOVER)
 630                 seq_show_option(m, "recover_session", "clean");
 631
 632         if (fsopt->wsize != CEPH_MAX_WRITE_SIZE)
 633                 seq_printf(m, ",wsize=%u", fsopt->wsize);
 634         if (fsopt->rsize != CEPH_MAX_READ_SIZE)
 635                 seq_printf(m, ",rsize=%u", fsopt->rsize);
 636         if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
 637                 seq_printf(m, ",rasize=%u", fsopt->rasize);
 638         if (fsopt->congestion_kb != default_congestion_kb())
 639                 seq_printf(m, ",write_congestion_kb=%u", fsopt->congestion_kb);
 640         if (fsopt->caps_max)
 641                 seq_printf(m, ",caps_max=%d", fsopt->caps_max);
 642         if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
 643                 seq_printf(m, ",caps_wanted_delay_min=%u",
 644                          fsopt->caps_wanted_delay_min);
 645         if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
 646                 seq_printf(m, ",caps_wanted_delay_max=%u",
 647                            fsopt->caps_wanted_delay_max);
 648         if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
 649                 seq_printf(m, ",readdir_max_entries=%u", fsopt->max_readdir);
 650         if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
 651                 seq_printf(m, ",readdir_max_bytes=%u", fsopt->max_readdir_bytes);
 652         if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
 653                 seq_show_option(m, "snapdirname", fsopt->snapdir_name);
 654
 655         return 0;
 656 }
 657
 658 /*
 659  * handle any mon messages the standard library doesn't understand.
 660  * return error if we don't either.
 661  */
 662 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
 663 {
 664         struct ceph_fs_client *fsc = client->private;
 665         int type = le16_to_cpu(msg->hdr.type);
 666
 667         switch (type) {
 668         case CEPH_MSG_MDS_MAP:
 669                 ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
 670                 return 0;
 671         case CEPH_MSG_FS_MAP_USER:
 672                 ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
 673                 return 0;
 674         default:
 675                 return -1;
 676         }
 677 }
 678
 679 /*
 680  * create a new fs client
 681  *
 682  * Success or not, this function consumes @fsopt and @opt.
 683  */
 684 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
 685                                         struct ceph_options *opt)
 686 {
 687         struct ceph_fs_client *fsc;
 688         int page_count;
 689         size_t size;
 690         int err;
 691
 692         fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
 693         if (!fsc) {
 694                 err = -ENOMEM;
 695                 goto fail;
 696         }
 697
 698         fsc->client = ceph_create_client(opt, fsc);
 699         if (IS_ERR(fsc->client)) {
 700                 err = PTR_ERR(fsc->client);
 701                 goto fail;
 702         }
 703         opt = NULL; /* fsc->client now owns this */
 704
 705         fsc->client->extra_mon_dispatch = extra_mon_dispatch;
 706         ceph_set_opt(fsc->client, ABORT_ON_FULL);
 707
 708         if (!fsopt->mds_namespace) {
 709                 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
 710                                    0, true);
 711         } else {
 712                 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
 713                                    0, false);
 714         }
 715
 716         fsc->mount_options = fsopt;
 717
 718         fsc->sb = NULL;
 719         fsc->mount_state = CEPH_MOUNT_MOUNTING;
 720         fsc->filp_gen = 1;
 721         fsc->have_copy_from2 = true;
 722
 723         atomic_long_set(&fsc->writeback_count, 0);
 724
 725         err = -ENOMEM;
 726         /*
 727          * The number of concurrent works can be high but they don't need
 728          * to be processed in parallel, limit concurrency.
 729          */
 730         fsc->inode_wq = alloc_workqueue("ceph-inode", WQ_UNBOUND, 0);
 731         if (!fsc->inode_wq)
 732                 goto fail_client;
 733         fsc->cap_wq = alloc_workqueue("ceph-cap", 0, 1);
 734         if (!fsc->cap_wq)
 735                 goto fail_inode_wq;
 736
 737         /* set up mempools */
 738         err = -ENOMEM;
 739         page_count = fsc->mount_options->wsize >> PAGE_SHIFT;
 740         size = sizeof (struct page *) * (page_count ? page_count : 1);
 741         fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
 742         if (!fsc->wb_pagevec_pool)
 743                 goto fail_cap_wq;
 744
 745         return fsc;
 746
 747 fail_cap_wq:
 748         destroy_workqueue(fsc->cap_wq);
 749 fail_inode_wq:
 750         destroy_workqueue(fsc->inode_wq);
 751 fail_client:
 752         ceph_destroy_client(fsc->client);
 753 fail:
 754         kfree(fsc);
 755         if (opt)
 756                 ceph_destroy_options(opt);
 757         destroy_mount_options(fsopt);
 758         return ERR_PTR(err);
 759 }
 760
 761 static void flush_fs_workqueues(struct ceph_fs_client *fsc)
 762 {
 763         flush_workqueue(fsc->inode_wq);
 764         flush_workqueue(fsc->cap_wq);
 765 }
 766
 767 static void destroy_fs_client(struct ceph_fs_client *fsc)
 768 {
 769         dout("destroy_fs_client %p\n", fsc);
 770
 771         ceph_mdsc_destroy(fsc);
 772         destroy_workqueue(fsc->inode_wq);
 773         destroy_workqueue(fsc->cap_wq);
 774
 775         mempool_destroy(fsc->wb_pagevec_pool);
 776
 777         destroy_mount_options(fsc->mount_options);
 778
 779         ceph_destroy_client(fsc->client);
 780
 781         kfree(fsc);
 782         dout("destroy_fs_client %p done\n", fsc);
 783 }
 784
 785 /*
 786  * caches
 787  */
 788 struct kmem_cache *ceph_inode_cachep;
 789 struct kmem_cache *ceph_cap_cachep;
 790 struct kmem_cache *ceph_cap_flush_cachep;
 791 struct kmem_cache *ceph_dentry_cachep;
 792 struct kmem_cache *ceph_file_cachep;
 793 struct kmem_cache *ceph_dir_file_cachep;
 794
 795 static void ceph_inode_init_once(void *foo)
 796 {
 797         struct ceph_inode_info *ci = foo;
 798         inode_init_once(&ci->vfs_inode);
 799 }
 800
 801 static int __init init_caches(void)
 802 {
 803         int error = -ENOMEM;
 804
 805         ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
 806                                       sizeof(struct ceph_inode_info),
 807                                       __alignof__(struct ceph_inode_info),
 808                                       SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
 809                                       SLAB_ACCOUNT, ceph_inode_init_once);
 810         if (!ceph_inode_cachep)
 811                 return -ENOMEM;
 812
 813         ceph_cap_cachep = KMEM_CACHE(ceph_cap, SLAB_MEM_SPREAD);
 814         if (!ceph_cap_cachep)
 815                 goto bad_cap;
 816         ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
 817                                            SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 818         if (!ceph_cap_flush_cachep)
 819                 goto bad_cap_flush;
 820
 821         ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
 822                                         SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 823         if (!ceph_dentry_cachep)
 824                 goto bad_dentry;
 825
 826         ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD);
 827         if (!ceph_file_cachep)
 828                 goto bad_file;
 829
 830         ceph_dir_file_cachep = KMEM_CACHE(ceph_dir_file_info, SLAB_MEM_SPREAD);
 831         if (!ceph_dir_file_cachep)
 832                 goto bad_dir_file;
 833
 834         error = ceph_fscache_register();
 835         if (error)
 836                 goto bad_fscache;
 837
 838         return 0;
 839
 840 bad_fscache:
 841         kmem_cache_destroy(ceph_dir_file_cachep);
 842 bad_dir_file:
 843         kmem_cache_destroy(ceph_file_cachep);
 844 bad_file:
 845         kmem_cache_destroy(ceph_dentry_cachep);
 846 bad_dentry:
 847         kmem_cache_destroy(ceph_cap_flush_cachep);
 848 bad_cap_flush:
 849         kmem_cache_destroy(ceph_cap_cachep);
 850 bad_cap:
 851         kmem_cache_destroy(ceph_inode_cachep);
 852         return error;
 853 }
 854
 855 static void destroy_caches(void)
 856 {
 857         /*
 858          * Make sure all delayed rcu free inodes are flushed before we
 859          * destroy cache.
 860          */
 861         rcu_barrier();
 862
 863         kmem_cache_destroy(ceph_inode_cachep);
 864         kmem_cache_destroy(ceph_cap_cachep);
 865         kmem_cache_destroy(ceph_cap_flush_cachep);
 866         kmem_cache_destroy(ceph_dentry_cachep);
 867         kmem_cache_destroy(ceph_file_cachep);
 868         kmem_cache_destroy(ceph_dir_file_cachep);
 869
 870         ceph_fscache_unregister();
 871 }
 872
 873 /*
 874  * ceph_umount_begin - initiate forced umount.  Tear down down the
 875  * mount, skipping steps that may hang while waiting for server(s).
 876  */
 877 static void ceph_umount_begin(struct super_block *sb)
 878 {
 879         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
 880
 881         dout("ceph_umount_begin - starting forced umount\n");
 882         if (!fsc)
 883                 return;
 884         fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
 885         ceph_osdc_abort_requests(&fsc->client->osdc, -EIO);
 886         ceph_mdsc_force_umount(fsc->mdsc);
 887         fsc->filp_gen++; // invalidate open files
 888 }
 889
 890 static const struct super_operations ceph_super_ops = {
 891         .alloc_inode    = ceph_alloc_inode,
 892         .free_inode     = ceph_free_inode,
 893         .write_inode    = ceph_write_inode,
 894         .drop_inode     = generic_delete_inode,
 895         .evict_inode    = ceph_evict_inode,
 896         .sync_fs        = ceph_sync_fs,
 897         .put_super      = ceph_put_super,
 898         .show_options   = ceph_show_options,
 899         .statfs         = ceph_statfs,
 900         .umount_begin   = ceph_umount_begin,
 901 };
 902
 903 /*
 904  * Bootstrap mount by opening the root directory.  Note the mount
 905  * @started time from caller, and time out if this takes too long.
 906  */
 907 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
 908                                        const char *path,
 909                                        unsigned long started)
 910 {
 911         struct ceph_mds_client *mdsc = fsc->mdsc;
 912         struct ceph_mds_request *req = NULL;
 913         int err;
 914         struct dentry *root;
 915
 916         /* open dir */
 917         dout("open_root_inode opening '%s'\n", path);
 918         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
 919         if (IS_ERR(req))
 920                 return ERR_CAST(req);
 921         req->r_path1 = kstrdup(path, GFP_NOFS);
 922         if (!req->r_path1) {
 923                 root = ERR_PTR(-ENOMEM);
 924                 goto out;
 925         }
 926
 927         req->r_ino1.ino = CEPH_INO_ROOT;
 928         req->r_ino1.snap = CEPH_NOSNAP;
 929         req->r_started = started;
 930         req->r_timeout = fsc->client->options->mount_timeout;
 931         req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
 932         req->r_num_caps = 2;
 933         err = ceph_mdsc_do_request(mdsc, NULL, req);
 934         if (err == 0) {
 935                 struct inode *inode = req->r_target_inode;
 936                 req->r_target_inode = NULL;
 937                 dout("open_root_inode success\n");
 938                 root = d_make_root(inode);
 939                 if (!root) {
 940                         root = ERR_PTR(-ENOMEM);
 941                         goto out;
 942                 }
 943                 dout("open_root_inode success, root dentry is %p\n", root);
 944         } else {
 945                 root = ERR_PTR(err);
 946         }
 947 out:
 948         ceph_mdsc_put_request(req);
 949         return root;
 950 }
 951
 952 /*
 953  * mount: join the ceph cluster, and open root directory.
 954  */
 955 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
 956                                       struct fs_context *fc)
 957 {
 958         int err;
 959         unsigned long started = jiffies;  /* note the start time */
 960         struct dentry *root;
 961
 962         dout("mount start %p\n", fsc);
 963         mutex_lock(&fsc->client->mount_mutex);
 964
 965         if (!fsc->sb->s_root) {
 966                 const char *path, *p;
 967                 err = __ceph_open_session(fsc->client, started);
 968                 if (err < 0)
 969                         goto out;
 970
 971                 /* setup fscache */
 972                 if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
 973                         err = ceph_fscache_register_fs(fsc, fc);
 974                         if (err < 0)
 975                                 goto out;
 976                 }
 977
 978                 p = path_remove_extra_slash(fsc->mount_options->server_path);
 979                 if (IS_ERR(p)) {
 980                         err = PTR_ERR(p);
 981                         goto out;
 982                 }
 983                 /* if the server path is omitted or just consists of '/' */
 984                 if (!p)
 985                         path = "";
 986                 else
 987                         path = p;
 988                 dout("mount opening path '%s'\n", path);
 989
 990                 ceph_fs_debugfs_init(fsc);
 991
 992                 root = open_root_dentry(fsc, path, started);
 993                 kfree(p);
 994                 if (IS_ERR(root)) {
 995                         err = PTR_ERR(root);
 996                         goto out;
 997                 }
 998                 fsc->sb->s_root = dget(root);
 999         } else {
1000                 root = dget(fsc->sb->s_root);
1001         }
1002
1003         fsc->mount_state = CEPH_MOUNT_MOUNTED;
1004         dout("mount success\n");
1005         mutex_unlock(&fsc->client->mount_mutex);
1006         return root;
1007
1008 out:
1009         mutex_unlock(&fsc->client->mount_mutex);
1010         return ERR_PTR(err);
1011 }
1012
1013 static int ceph_set_super(struct super_block *s, struct fs_context *fc)
1014 {
1015         struct ceph_fs_client *fsc = s->s_fs_info;
1016         int ret;
1017
1018         dout("set_super %p\n", s);
1019
1020         s->s_maxbytes = MAX_LFS_FILESIZE;
1021
1022         s->s_xattr = ceph_xattr_handlers;
1023         fsc->sb = s;
1024         fsc->max_file_size = 1ULL << 40; /* temp value until we get mdsmap */
1025
1026         s->s_op = &ceph_super_ops;
1027         s->s_d_op = &ceph_dentry_ops;
1028         s->s_export_op = &ceph_export_ops;
1029
1030         s->s_time_gran = 1;
1031         s->s_time_min = 0;
1032         s->s_time_max = U32_MAX;
1033
1034         ret = set_anon_super_fc(s, fc);
1035         if (ret != 0)
1036                 fsc->sb = NULL;
1037         return ret;
1038 }
1039
1040 /*
1041  * share superblock if same fs AND options
1042  */
1043 static int ceph_compare_super(struct super_block *sb, struct fs_context *fc)
1044 {
1045         struct ceph_fs_client *new = fc->s_fs_info;
1046         struct ceph_mount_options *fsopt = new->mount_options;
1047         struct ceph_options *opt = new->client->options;
1048         struct ceph_fs_client *other = ceph_sb_to_client(sb);
1049
1050         dout("ceph_compare_super %p\n", sb);
1051
1052         if (compare_mount_options(fsopt, opt, other)) {
1053                 dout("monitor(s)/mount options don't match\n");
1054                 return 0;
1055         }
1056         if ((opt->flags & CEPH_OPT_FSID) &&
1057             ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
1058                 dout("fsid doesn't match\n");
1059                 return 0;
1060         }
1061         if (fc->sb_flags != (sb->s_flags & ~SB_BORN)) {
1062                 dout("flags differ\n");
1063                 return 0;
1064         }
1065         return 1;
1066 }
1067
1068 /*
1069  * construct our own bdi so we can control readahead, etc.
1070  */
1071 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1072
1073 static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
1074 {
1075         int err;
1076
1077         err = super_setup_bdi_name(sb, "ceph-%ld",
1078                                    atomic_long_inc_return(&bdi_seq));
1079         if (err)
1080                 return err;
1081
1082         /* set ra_pages based on rasize mount option? */
1083         sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
1084
1085         /* set io_pages based on max osd read size */
1086         sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
1087
1088         return 0;
1089 }
1090
1091 static int ceph_get_tree(struct fs_context *fc)
1092 {
1093         struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1094         struct super_block *sb;
1095         struct ceph_fs_client *fsc;
1096         struct dentry *res;
1097         int (*compare_super)(struct super_block *, struct fs_context *) =
1098                 ceph_compare_super;
1099         int err;
1100
1101         dout("ceph_get_tree\n");
1102
1103         if (!fc->source)
1104                 return invalf(fc, "ceph: No source");
1105
1106 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1107         fc->sb_flags |= SB_POSIXACL;
1108 #endif
1109
1110         /* create client (which we may/may not use) */
1111         fsc = create_fs_client(pctx->opts, pctx->copts);
1112         pctx->opts = NULL;
1113         pctx->copts = NULL;
1114         if (IS_ERR(fsc)) {
1115                 err = PTR_ERR(fsc);
1116                 goto out_final;
1117         }
1118
1119         err = ceph_mdsc_init(fsc);
1120         if (err < 0)
1121                 goto out;
1122
1123         if (ceph_test_opt(fsc->client, NOSHARE))
1124                 compare_super = NULL;
1125
1126         fc->s_fs_info = fsc;
1127         sb = sget_fc(fc, compare_super, ceph_set_super);
1128         fc->s_fs_info = NULL;
1129         if (IS_ERR(sb)) {
1130                 err = PTR_ERR(sb);
1131                 goto out;
1132         }
1133
1134         if (ceph_sb_to_client(sb) != fsc) {
1135                 destroy_fs_client(fsc);
1136                 fsc = ceph_sb_to_client(sb);
1137                 dout("get_sb got existing client %p\n", fsc);
1138         } else {
1139                 dout("get_sb using new client %p\n", fsc);
1140                 err = ceph_setup_bdi(sb, fsc);
1141                 if (err < 0)
1142                         goto out_splat;
1143         }
1144
1145         res = ceph_real_mount(fsc, fc);
1146         if (IS_ERR(res)) {
1147                 err = PTR_ERR(res);
1148                 goto out_splat;
1149         }
1150         dout("root %p inode %p ino %llx.%llx\n", res,
1151              d_inode(res), ceph_vinop(d_inode(res)));
1152         fc->root = fsc->sb->s_root;
1153         return 0;
1154
1155 out_splat:
1156         if (!ceph_mdsmap_is_cluster_available(fsc->mdsc->mdsmap)) {
1157                 pr_info("No mds server is up or the cluster is laggy\n");
1158                 err = -EHOSTUNREACH;
1159         }
1160
1161         ceph_mdsc_close_sessions(fsc->mdsc);
1162         deactivate_locked_super(sb);
1163         goto out_final;
1164
1165 out:
1166         destroy_fs_client(fsc);
1167 out_final:
1168         dout("ceph_get_tree fail %d\n", err);
1169         return err;
1170 }
1171
1172 static void ceph_free_fc(struct fs_context *fc)
1173 {
1174         struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1175
1176         if (pctx) {
1177                 destroy_mount_options(pctx->opts);
1178                 ceph_destroy_options(pctx->copts);
1179                 kfree(pctx);
1180         }
1181 }
1182
1183 static int ceph_reconfigure_fc(struct fs_context *fc)
1184 {
1185         sync_filesystem(fc->root->d_sb);
1186         return 0;
1187 }
1188
1189 static const struct fs_context_operations ceph_context_ops = {
1190         .free           = ceph_free_fc,
1191         .parse_param    = ceph_parse_mount_param,
1192         .get_tree       = ceph_get_tree,
1193         .reconfigure    = ceph_reconfigure_fc,
1194 };
1195
1196 /*
1197  * Set up the filesystem mount context.
1198  */
1199 static int ceph_init_fs_context(struct fs_context *fc)
1200 {
1201         struct ceph_parse_opts_ctx *pctx;
1202         struct ceph_mount_options *fsopt;
1203
1204         pctx = kzalloc(sizeof(*pctx), GFP_KERNEL);
1205         if (!pctx)
1206                 return -ENOMEM;
1207
1208         pctx->copts = ceph_alloc_options();
1209         if (!pctx->copts)
1210                 goto nomem;
1211
1212         pctx->opts = kzalloc(sizeof(*pctx->opts), GFP_KERNEL);
1213         if (!pctx->opts)
1214                 goto nomem;
1215
1216         fsopt = pctx->opts;
1217         fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
1218
1219         fsopt->wsize = CEPH_MAX_WRITE_SIZE;
1220         fsopt->rsize = CEPH_MAX_READ_SIZE;
1221         fsopt->rasize = CEPH_RASIZE_DEFAULT;
1222         fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
1223         if (!fsopt->snapdir_name)
1224                 goto nomem;
1225
1226         fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
1227         fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
1228         fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
1229         fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
1230         fsopt->congestion_kb = default_congestion_kb();
1231
1232         fc->fs_private = pctx;
1233         fc->ops = &ceph_context_ops;
1234         return 0;
1235
1236 nomem:
1237         destroy_mount_options(pctx->opts);
1238         ceph_destroy_options(pctx->copts);
1239         kfree(pctx);
1240         return -ENOMEM;
1241 }
1242
1243 static void ceph_kill_sb(struct super_block *s)
1244 {
1245         struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1246         dev_t dev = s->s_dev;
1247
1248         dout("kill_sb %p\n", s);
1249
1250         ceph_mdsc_pre_umount(fsc->mdsc);
1251         flush_fs_workqueues(fsc);
1252
1253         generic_shutdown_super(s);
1254
1255         fsc->client->extra_mon_dispatch = NULL;
1256         ceph_fs_debugfs_cleanup(fsc);
1257
1258         ceph_fscache_unregister_fs(fsc);
1259
1260         destroy_fs_client(fsc);
1261         free_anon_bdev(dev);
1262 }
1263
1264 static struct file_system_type ceph_fs_type = {
1265         .owner          = THIS_MODULE,
1266         .name           = "ceph",
1267         .init_fs_context = ceph_init_fs_context,
1268         .kill_sb        = ceph_kill_sb,
1269         .fs_flags       = FS_RENAME_DOES_D_MOVE,
1270 };
1271 MODULE_ALIAS_FS("ceph");
1272
1273 int ceph_force_reconnect(struct super_block *sb)
1274 {
1275         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1276         int err = 0;
1277
1278         ceph_umount_begin(sb);
1279
1280         /* Make sure all page caches get invalidated.
1281          * see remove_session_caps_cb() */
1282         flush_workqueue(fsc->inode_wq);
1283
1284         /* In case that we were blacklisted. This also reset
1285          * all mon/osd connections */
1286         ceph_reset_client_addr(fsc->client);
1287
1288         ceph_osdc_clear_abort_err(&fsc->client->osdc);
1289
1290         fsc->blacklisted = false;
1291         fsc->mount_state = CEPH_MOUNT_MOUNTED;
1292
1293         if (sb->s_root) {
1294                 err = __ceph_do_getattr(d_inode(sb->s_root), NULL,
1295                                         CEPH_STAT_CAP_INODE, true);
1296         }
1297         return err;
1298 }
1299
1300 static int __init init_ceph(void)
1301 {
1302         int ret = init_caches();
1303         if (ret)
1304                 goto out;
1305
1306         ceph_flock_init();
1307         ret = register_filesystem(&ceph_fs_type);
1308         if (ret)
1309                 goto out_caches;
1310
1311         pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1312
1313         return 0;
1314
1315 out_caches:
1316         destroy_caches();
1317 out:
1318         return ret;
1319 }
1320
1321 static void __exit exit_ceph(void)
1322 {
1323         dout("exit_ceph\n");
1324         unregister_filesystem(&ceph_fs_type);
1325         destroy_caches();
1326 }
1327
1328 module_init(init_ceph);
1329 module_exit(exit_ceph);
1330
1331 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1332 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1333 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1334 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1335 MODULE_LICENSE("GPL");