fs/ceph/super.c

   1
   2 #include <linux/ceph/ceph_debug.h>
   3
   4 #include <linux/backing-dev.h>
   5 #include <linux/ctype.h>
   6 #include <linux/fs.h>
   7 #include <linux/inet.h>
   8 #include <linux/in6.h>
   9 #include <linux/module.h>
  10 #include <linux/mount.h>
  11 #include <linux/parser.h>
  12 #include <linux/sched.h>
  13 #include <linux/seq_file.h>
  14 #include <linux/slab.h>
  15 #include <linux/statfs.h>
  16 #include <linux/string.h>
  17
  18 #include "super.h"
  19 #include "mds_client.h"
  20 #include "cache.h"
  21
  22 #include <linux/ceph/ceph_features.h>
  23 #include <linux/ceph/decode.h>
  24 #include <linux/ceph/mon_client.h>
  25 #include <linux/ceph/auth.h>
  26 #include <linux/ceph/debugfs.h>
  27
  28 /*
  29  * Ceph superblock operations
  30  *
  31  * Handle the basics of mounting, unmounting.
  32  */
  33
  34 /*
  35  * super ops
  36  */
  37 static void ceph_put_super(struct super_block *s)
  38 {
  39         struct ceph_fs_client *fsc = ceph_sb_to_client(s);
  40
  41         dout("put_super\n");
  42         ceph_mdsc_close_sessions(fsc->mdsc);
  43 }
  44
  45 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
  46 {
  47         struct ceph_fs_client *fsc = ceph_inode_to_client(d_inode(dentry));
  48         struct ceph_monmap *monmap = fsc->client->monc.monmap;
  49         struct ceph_statfs st;
  50         u64 fsid;
  51         int err;
  52         u64 data_pool;
  53
  54         if (fsc->mdsc->mdsmap->m_num_data_pg_pools == 1) {
  55                 data_pool = fsc->mdsc->mdsmap->m_data_pg_pools[0];
  56         } else {
  57                 data_pool = CEPH_NOPOOL;
  58         }
  59
  60         dout("statfs\n");
  61         err = ceph_monc_do_statfs(&fsc->client->monc, data_pool, &st);
  62         if (err < 0)
  63                 return err;
  64
  65         /* fill in kstatfs */
  66         buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
  67
  68         /*
  69          * express utilization in terms of large blocks to avoid
  70          * overflow on 32-bit machines.
  71          *
  72          * NOTE: for the time being, we make bsize == frsize to humor
  73          * not-yet-ancient versions of glibc that are broken.
  74          * Someday, we will probably want to report a real block
  75          * size...  whatever that may mean for a network file system!
  76          */
  77         buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
  78         buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
  79         buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
  80         buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
  81         buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
  82
  83         buf->f_files = le64_to_cpu(st.num_objects);
  84         buf->f_ffree = -1;
  85         buf->f_namelen = NAME_MAX;
  86
  87         /* leave fsid little-endian, regardless of host endianness */
  88         fsid = *(u64 *)(&monmap->fsid) ^ *((u64 *)&monmap->fsid + 1);
  89         buf->f_fsid.val[0] = fsid & 0xffffffff;
  90         buf->f_fsid.val[1] = fsid >> 32;
  91
  92         return 0;
  93 }
  94
  95 static int ceph_sync_fs(struct super_block *sb, int wait)
  96 {
  97         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
  98
  99         if (!wait) {
 100                 dout("sync_fs (non-blocking)\n");
 101                 ceph_flush_dirty_caps(fsc->mdsc);
 102                 dout("sync_fs (non-blocking) done\n");
 103                 return 0;
 104         }
 105
 106         dout("sync_fs (blocking)\n");
 107         ceph_osdc_sync(&fsc->client->osdc);
 108         ceph_mdsc_sync(fsc->mdsc);
 109         dout("sync_fs (blocking) done\n");
 110         return 0;
 111 }
 112
 113 /*
 114  * mount options
 115  */
 116 enum {
 117         Opt_wsize,
 118         Opt_rsize,
 119         Opt_rasize,
 120         Opt_caps_wanted_delay_min,
 121         Opt_caps_wanted_delay_max,
 122         Opt_readdir_max_entries,
 123         Opt_readdir_max_bytes,
 124         Opt_congestion_kb,
 125         Opt_last_int,
 126         /* int args above */
 127         Opt_snapdirname,
 128         Opt_mds_namespace,
 129         Opt_fscache_uniq,
 130         Opt_last_string,
 131         /* string args above */
 132         Opt_dirstat,
 133         Opt_nodirstat,
 134         Opt_rbytes,
 135         Opt_norbytes,
 136         Opt_asyncreaddir,
 137         Opt_noasyncreaddir,
 138         Opt_dcache,
 139         Opt_nodcache,
 140         Opt_ino32,
 141         Opt_noino32,
 142         Opt_fscache,
 143         Opt_nofscache,
 144         Opt_poolperm,
 145         Opt_nopoolperm,
 146         Opt_require_active_mds,
 147         Opt_norequire_active_mds,
 148 #ifdef CONFIG_CEPH_FS_POSIX_ACL
 149         Opt_acl,
 150 #endif
 151         Opt_noacl,
 152 };
 153
 154 static match_table_t fsopt_tokens = {
 155         {Opt_wsize, "wsize=%d"},
 156         {Opt_rsize, "rsize=%d"},
 157         {Opt_rasize, "rasize=%d"},
 158         {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
 159         {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
 160         {Opt_readdir_max_entries, "readdir_max_entries=%d"},
 161         {Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
 162         {Opt_congestion_kb, "write_congestion_kb=%d"},
 163         /* int args above */
 164         {Opt_snapdirname, "snapdirname=%s"},
 165         {Opt_mds_namespace, "mds_namespace=%s"},
 166         {Opt_fscache_uniq, "fsc=%s"},
 167         /* string args above */
 168         {Opt_dirstat, "dirstat"},
 169         {Opt_nodirstat, "nodirstat"},
 170         {Opt_rbytes, "rbytes"},
 171         {Opt_norbytes, "norbytes"},
 172         {Opt_asyncreaddir, "asyncreaddir"},
 173         {Opt_noasyncreaddir, "noasyncreaddir"},
 174         {Opt_dcache, "dcache"},
 175         {Opt_nodcache, "nodcache"},
 176         {Opt_ino32, "ino32"},
 177         {Opt_noino32, "noino32"},
 178         {Opt_fscache, "fsc"},
 179         {Opt_nofscache, "nofsc"},
 180         {Opt_poolperm, "poolperm"},
 181         {Opt_nopoolperm, "nopoolperm"},
 182         {Opt_require_active_mds, "require_active_mds"},
 183         {Opt_norequire_active_mds, "norequire_active_mds"},
 184 #ifdef CONFIG_CEPH_FS_POSIX_ACL
 185         {Opt_acl, "acl"},
 186 #endif
 187         {Opt_noacl, "noacl"},
 188         {-1, NULL}
 189 };
 190
 191 /*
 192  * Remove adjacent slashes and then the trailing slash, unless it is
 193  * the only remaining character.
 194  *
 195  * E.g. "//dir1////dir2///" --> "/dir1/dir2", "///" --> "/".
 196  */
 197 static void canonicalize_path(char *path)
 198 {
 199         int i, j = 0;
 200
 201         for (i = 0; path[i] != '\0'; i++) {
 202                 if (path[i] != '/' || j < 1 || path[j - 1] != '/')
 203                         path[j++] = path[i];
 204         }
 205
 206         if (j > 1 && path[j - 1] == '/')
 207                 j--;
 208         path[j] = '\0';
 209 }
 210
 211 static int parse_fsopt_token(char *c, void *private)
 212 {
 213         struct ceph_mount_options *fsopt = private;
 214         substring_t argstr[MAX_OPT_ARGS];
 215         int token, intval, ret;
 216
 217         token = match_token((char *)c, fsopt_tokens, argstr);
 218         if (token < 0)
 219                 return -EINVAL;
 220
 221         if (token < Opt_last_int) {
 222                 ret = match_int(&argstr[0], &intval);
 223                 if (ret < 0) {
 224                         pr_err("bad mount option arg (not int) "
 225                                "at '%s'\n", c);
 226                         return ret;
 227                 }
 228                 dout("got int token %d val %d\n", token, intval);
 229         } else if (token > Opt_last_int && token < Opt_last_string) {
 230                 dout("got string token %d val %s\n", token,
 231                      argstr[0].from);
 232         } else {
 233                 dout("got token %d\n", token);
 234         }
 235
 236         switch (token) {
 237         case Opt_snapdirname:
 238                 kfree(fsopt->snapdir_name);
 239                 fsopt->snapdir_name = kstrndup(argstr[0].from,
 240                                                argstr[0].to-argstr[0].from,
 241                                                GFP_KERNEL);
 242                 if (!fsopt->snapdir_name)
 243                         return -ENOMEM;
 244                 break;
 245         case Opt_mds_namespace:
 246                 kfree(fsopt->mds_namespace);
 247                 fsopt->mds_namespace = kstrndup(argstr[0].from,
 248                                                 argstr[0].to-argstr[0].from,
 249                                                 GFP_KERNEL);
 250                 if (!fsopt->mds_namespace)
 251                         return -ENOMEM;
 252                 break;
 253         case Opt_fscache_uniq:
 254 #ifdef CONFIG_CEPH_FSCACHE
 255                 kfree(fsopt->fscache_uniq);
 256                 fsopt->fscache_uniq = kstrndup(argstr[0].from,
 257                                                argstr[0].to-argstr[0].from,
 258                                                GFP_KERNEL);
 259                 if (!fsopt->fscache_uniq)
 260                         return -ENOMEM;
 261                 fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
 262                 break;
 263 #else
 264                 pr_err("fscache support is disabled\n");
 265                 return -EINVAL;
 266 #endif
 267         case Opt_wsize:
 268                 if (intval < PAGE_SIZE || intval > CEPH_MAX_WRITE_SIZE)
 269                         return -EINVAL;
 270                 fsopt->wsize = ALIGN(intval, PAGE_SIZE);
 271                 break;
 272         case Opt_rsize:
 273                 if (intval < PAGE_SIZE || intval > CEPH_MAX_READ_SIZE)
 274                         return -EINVAL;
 275                 fsopt->rsize = ALIGN(intval, PAGE_SIZE);
 276                 break;
 277         case Opt_rasize:
 278                 if (intval < 0)
 279                         return -EINVAL;
 280                 fsopt->rasize = ALIGN(intval, PAGE_SIZE);
 281                 break;
 282         case Opt_caps_wanted_delay_min:
 283                 if (intval < 1)
 284                         return -EINVAL;
 285                 fsopt->caps_wanted_delay_min = intval;
 286                 break;
 287         case Opt_caps_wanted_delay_max:
 288                 if (intval < 1)
 289                         return -EINVAL;
 290                 fsopt->caps_wanted_delay_max = intval;
 291                 break;
 292         case Opt_readdir_max_entries:
 293                 if (intval < 1)
 294                         return -EINVAL;
 295                 fsopt->max_readdir = intval;
 296                 break;
 297         case Opt_readdir_max_bytes:
 298                 if (intval < PAGE_SIZE && intval != 0)
 299                         return -EINVAL;
 300                 fsopt->max_readdir_bytes = intval;
 301                 break;
 302         case Opt_congestion_kb:
 303                 if (intval < 1024) /* at least 1M */
 304                         return -EINVAL;
 305                 fsopt->congestion_kb = intval;
 306                 break;
 307         case Opt_dirstat:
 308                 fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
 309                 break;
 310         case Opt_nodirstat:
 311                 fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
 312                 break;
 313         case Opt_rbytes:
 314                 fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
 315                 break;
 316         case Opt_norbytes:
 317                 fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
 318                 break;
 319         case Opt_asyncreaddir:
 320                 fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
 321                 break;
 322         case Opt_noasyncreaddir:
 323                 fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
 324                 break;
 325         case Opt_dcache:
 326                 fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
 327                 break;
 328         case Opt_nodcache:
 329                 fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
 330                 break;
 331         case Opt_ino32:
 332                 fsopt->flags |= CEPH_MOUNT_OPT_INO32;
 333                 break;
 334         case Opt_noino32:
 335                 fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
 336                 break;
 337         case Opt_fscache:
 338 #ifdef CONFIG_CEPH_FSCACHE
 339                 fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
 340                 break;
 341 #else
 342                 pr_err("fscache support is disabled\n");
 343                 return -EINVAL;
 344 #endif
 345         case Opt_nofscache:
 346                 fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
 347                 break;
 348         case Opt_poolperm:
 349                 fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
 350                 printk ("pool perm");
 351                 break;
 352         case Opt_nopoolperm:
 353                 fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
 354                 break;
 355         case Opt_require_active_mds:
 356                 fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT;
 357                 break;
 358         case Opt_norequire_active_mds:
 359                 fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT;
 360                 break;
 361 #ifdef CONFIG_CEPH_FS_POSIX_ACL
 362         case Opt_acl:
 363                 fsopt->sb_flags |= MS_POSIXACL;
 364                 break;
 365 #endif
 366         case Opt_noacl:
 367                 fsopt->sb_flags &= ~MS_POSIXACL;
 368                 break;
 369         default:
 370                 BUG_ON(token);
 371         }
 372         return 0;
 373 }
 374
 375 static void destroy_mount_options(struct ceph_mount_options *args)
 376 {
 377         dout("destroy_mount_options %p\n", args);
 378         kfree(args->snapdir_name);
 379         kfree(args->mds_namespace);
 380         kfree(args->server_path);
 381         kfree(args->fscache_uniq);
 382         kfree(args);
 383 }
 384
 385 static int strcmp_null(const char *s1, const char *s2)
 386 {
 387         if (!s1 && !s2)
 388                 return 0;
 389         if (s1 && !s2)
 390                 return -1;
 391         if (!s1 && s2)
 392                 return 1;
 393         return strcmp(s1, s2);
 394 }
 395
 396 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
 397                                  struct ceph_options *new_opt,
 398                                  struct ceph_fs_client *fsc)
 399 {
 400         struct ceph_mount_options *fsopt1 = new_fsopt;
 401         struct ceph_mount_options *fsopt2 = fsc->mount_options;
 402         int ofs = offsetof(struct ceph_mount_options, snapdir_name);
 403         int ret;
 404
 405         ret = memcmp(fsopt1, fsopt2, ofs);
 406         if (ret)
 407                 return ret;
 408
 409         ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
 410         if (ret)
 411                 return ret;
 412
 413         ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
 414         if (ret)
 415                 return ret;
 416
 417         ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
 418         if (ret)
 419                 return ret;
 420
 421         ret = strcmp_null(fsopt1->fscache_uniq, fsopt2->fscache_uniq);
 422         if (ret)
 423                 return ret;
 424
 425         return ceph_compare_options(new_opt, fsc->client);
 426 }
 427
 428 static int parse_mount_options(struct ceph_mount_options **pfsopt,
 429                                struct ceph_options **popt,
 430                                int flags, char *options,
 431                                const char *dev_name)
 432 {
 433         struct ceph_mount_options *fsopt;
 434         const char *dev_name_end;
 435         int err;
 436
 437         if (!dev_name || !*dev_name)
 438                 return -EINVAL;
 439
 440         fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
 441         if (!fsopt)
 442                 return -ENOMEM;
 443
 444         dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
 445
 446         fsopt->sb_flags = flags;
 447         fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
 448
 449         fsopt->wsize = CEPH_MAX_WRITE_SIZE;
 450         fsopt->rsize = CEPH_MAX_READ_SIZE;
 451         fsopt->rasize = CEPH_RASIZE_DEFAULT;
 452         fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
 453         if (!fsopt->snapdir_name) {
 454                 err = -ENOMEM;
 455                 goto out;
 456         }
 457
 458         fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
 459         fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
 460         fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
 461         fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
 462         fsopt->congestion_kb = default_congestion_kb();
 463
 464         /*
 465          * Distinguish the server list from the path in "dev_name".
 466          * Internally we do not include the leading '/' in the path.
 467          *
 468          * "dev_name" will look like:
 469          *     <server_spec>[,<server_spec>...]:[<path>]
 470          * where
 471          *     <server_spec> is <ip>[:<port>]
 472          *     <path> is optional, but if present must begin with '/'
 473          */
 474         dev_name_end = strchr(dev_name, '/');
 475         if (dev_name_end) {
 476                 /*
 477                  * The server_path will include the whole chars from userland
 478                  * including the leading '/'.
 479                  */
 480                 fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
 481                 if (!fsopt->server_path) {
 482                         err = -ENOMEM;
 483                         goto out;
 484                 }
 485
 486                 canonicalize_path(fsopt->server_path);
 487         } else {
 488                 dev_name_end = dev_name + strlen(dev_name);
 489         }
 490         err = -EINVAL;
 491         dev_name_end--;         /* back up to ':' separator */
 492         if (dev_name_end < dev_name || *dev_name_end != ':') {
 493                 pr_err("device name is missing path (no : separator in %s)\n",
 494                                 dev_name);
 495                 goto out;
 496         }
 497         dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
 498         if (fsopt->server_path)
 499                 dout("server path '%s'\n", fsopt->server_path);
 500
 501         *popt = ceph_parse_options(options, dev_name, dev_name_end,
 502                                  parse_fsopt_token, (void *)fsopt);
 503         if (IS_ERR(*popt)) {
 504                 err = PTR_ERR(*popt);
 505                 goto out;
 506         }
 507
 508         /* success */
 509         *pfsopt = fsopt;
 510         return 0;
 511
 512 out:
 513         destroy_mount_options(fsopt);
 514         return err;
 515 }
 516
 517 /**
 518  * ceph_show_options - Show mount options in /proc/mounts
 519  * @m: seq_file to write to
 520  * @root: root of that (sub)tree
 521  */
 522 static int ceph_show_options(struct seq_file *m, struct dentry *root)
 523 {
 524         struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
 525         struct ceph_mount_options *fsopt = fsc->mount_options;
 526         size_t pos;
 527         int ret;
 528
 529         /* a comma between MNT/MS and client options */
 530         seq_putc(m, ',');
 531         pos = m->count;
 532
 533         ret = ceph_print_client_options(m, fsc->client);
 534         if (ret)
 535                 return ret;
 536
 537         /* retract our comma if no client options */
 538         if (m->count == pos)
 539                 m->count--;
 540
 541         if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
 542                 seq_puts(m, ",dirstat");
 543         if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
 544                 seq_puts(m, ",rbytes");
 545         if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
 546                 seq_puts(m, ",noasyncreaddir");
 547         if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
 548                 seq_puts(m, ",nodcache");
 549         if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) {
 550                 if (fsopt->fscache_uniq)
 551                         seq_printf(m, ",fsc=%s", fsopt->fscache_uniq);
 552                 else
 553                         seq_puts(m, ",fsc");
 554         }
 555         if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
 556                 seq_puts(m, ",nopoolperm");
 557
 558 #ifdef CONFIG_CEPH_FS_POSIX_ACL
 559         if (fsopt->sb_flags & MS_POSIXACL)
 560                 seq_puts(m, ",acl");
 561         else
 562                 seq_puts(m, ",noacl");
 563 #endif
 564
 565         if (fsopt->mds_namespace)
 566                 seq_printf(m, ",mds_namespace=%s", fsopt->mds_namespace);
 567         if (fsopt->wsize)
 568                 seq_printf(m, ",wsize=%d", fsopt->wsize);
 569         if (fsopt->rsize != CEPH_MAX_READ_SIZE)
 570                 seq_printf(m, ",rsize=%d", fsopt->rsize);
 571         if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
 572                 seq_printf(m, ",rasize=%d", fsopt->rasize);
 573         if (fsopt->congestion_kb != default_congestion_kb())
 574                 seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
 575         if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
 576                 seq_printf(m, ",caps_wanted_delay_min=%d",
 577                          fsopt->caps_wanted_delay_min);
 578         if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
 579                 seq_printf(m, ",caps_wanted_delay_max=%d",
 580                            fsopt->caps_wanted_delay_max);
 581         if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
 582                 seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
 583         if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
 584                 seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
 585         if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
 586                 seq_show_option(m, "snapdirname", fsopt->snapdir_name);
 587
 588         return 0;
 589 }
 590
 591 /*
 592  * handle any mon messages the standard library doesn't understand.
 593  * return error if we don't either.
 594  */
 595 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
 596 {
 597         struct ceph_fs_client *fsc = client->private;
 598         int type = le16_to_cpu(msg->hdr.type);
 599
 600         switch (type) {
 601         case CEPH_MSG_MDS_MAP:
 602                 ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
 603                 return 0;
 604         case CEPH_MSG_FS_MAP_USER:
 605                 ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
 606                 return 0;
 607         default:
 608                 return -1;
 609         }
 610 }
 611
 612 /*
 613  * create a new fs client
 614  */
 615 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
 616                                         struct ceph_options *opt)
 617 {
 618         struct ceph_fs_client *fsc;
 619         int page_count;
 620         size_t size;
 621         int err = -ENOMEM;
 622
 623         fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
 624         if (!fsc)
 625                 return ERR_PTR(-ENOMEM);
 626
 627         fsc->client = ceph_create_client(opt, fsc);
 628         if (IS_ERR(fsc->client)) {
 629                 err = PTR_ERR(fsc->client);
 630                 goto fail;
 631         }
 632         fsc->client->extra_mon_dispatch = extra_mon_dispatch;
 633
 634         if (!fsopt->mds_namespace) {
 635                 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
 636                                    0, true);
 637         } else {
 638                 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
 639                                    0, false);
 640         }
 641
 642         fsc->mount_options = fsopt;
 643
 644         fsc->sb = NULL;
 645         fsc->mount_state = CEPH_MOUNT_MOUNTING;
 646
 647         atomic_long_set(&fsc->writeback_count, 0);
 648
 649         err = -ENOMEM;
 650         /*
 651          * The number of concurrent works can be high but they don't need
 652          * to be processed in parallel, limit concurrency.
 653          */
 654         fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
 655         if (!fsc->wb_wq)
 656                 goto fail_client;
 657         fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
 658         if (!fsc->pg_inv_wq)
 659                 goto fail_wb_wq;
 660         fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
 661         if (!fsc->trunc_wq)
 662                 goto fail_pg_inv_wq;
 663
 664         /* set up mempools */
 665         err = -ENOMEM;
 666         page_count = fsc->mount_options->wsize >> PAGE_SHIFT;
 667         size = sizeof (struct page *) * (page_count ? page_count : 1);
 668         fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
 669         if (!fsc->wb_pagevec_pool)
 670                 goto fail_trunc_wq;
 671
 672         /* caps */
 673         fsc->min_caps = fsopt->max_readdir;
 674
 675         return fsc;
 676
 677 fail_trunc_wq:
 678         destroy_workqueue(fsc->trunc_wq);
 679 fail_pg_inv_wq:
 680         destroy_workqueue(fsc->pg_inv_wq);
 681 fail_wb_wq:
 682         destroy_workqueue(fsc->wb_wq);
 683 fail_client:
 684         ceph_destroy_client(fsc->client);
 685 fail:
 686         kfree(fsc);
 687         return ERR_PTR(err);
 688 }
 689
 690 static void destroy_fs_client(struct ceph_fs_client *fsc)
 691 {
 692         dout("destroy_fs_client %p\n", fsc);
 693
 694         destroy_workqueue(fsc->wb_wq);
 695         destroy_workqueue(fsc->pg_inv_wq);
 696         destroy_workqueue(fsc->trunc_wq);
 697
 698         mempool_destroy(fsc->wb_pagevec_pool);
 699
 700         destroy_mount_options(fsc->mount_options);
 701
 702         ceph_destroy_client(fsc->client);
 703
 704         kfree(fsc);
 705         dout("destroy_fs_client %p done\n", fsc);
 706 }
 707
 708 /*
 709  * caches
 710  */
 711 struct kmem_cache *ceph_inode_cachep;
 712 struct kmem_cache *ceph_cap_cachep;
 713 struct kmem_cache *ceph_cap_flush_cachep;
 714 struct kmem_cache *ceph_dentry_cachep;
 715 struct kmem_cache *ceph_file_cachep;
 716
 717 static void ceph_inode_init_once(void *foo)
 718 {
 719         struct ceph_inode_info *ci = foo;
 720         inode_init_once(&ci->vfs_inode);
 721 }
 722
 723 static int __init init_caches(void)
 724 {
 725         int error = -ENOMEM;
 726
 727         ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
 728                                       sizeof(struct ceph_inode_info),
 729                                       __alignof__(struct ceph_inode_info),
 730                                       SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
 731                                       SLAB_ACCOUNT, ceph_inode_init_once);
 732         if (!ceph_inode_cachep)
 733                 return -ENOMEM;
 734
 735         ceph_cap_cachep = KMEM_CACHE(ceph_cap,
 736                                      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 737         if (!ceph_cap_cachep)
 738                 goto bad_cap;
 739         ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
 740                                            SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 741         if (!ceph_cap_flush_cachep)
 742                 goto bad_cap_flush;
 743
 744         ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
 745                                         SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 746         if (!ceph_dentry_cachep)
 747                 goto bad_dentry;
 748
 749         ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD);
 750         if (!ceph_file_cachep)
 751                 goto bad_file;
 752
 753         error = ceph_fscache_register();
 754         if (error)
 755                 goto bad_fscache;
 756
 757         return 0;
 758
 759 bad_fscache:
 760         kmem_cache_destroy(ceph_file_cachep);
 761 bad_file:
 762         kmem_cache_destroy(ceph_dentry_cachep);
 763 bad_dentry:
 764         kmem_cache_destroy(ceph_cap_flush_cachep);
 765 bad_cap_flush:
 766         kmem_cache_destroy(ceph_cap_cachep);
 767 bad_cap:
 768         kmem_cache_destroy(ceph_inode_cachep);
 769         return error;
 770 }
 771
 772 static void destroy_caches(void)
 773 {
 774         /*
 775          * Make sure all delayed rcu free inodes are flushed before we
 776          * destroy cache.
 777          */
 778         rcu_barrier();
 779
 780         kmem_cache_destroy(ceph_inode_cachep);
 781         kmem_cache_destroy(ceph_cap_cachep);
 782         kmem_cache_destroy(ceph_cap_flush_cachep);
 783         kmem_cache_destroy(ceph_dentry_cachep);
 784         kmem_cache_destroy(ceph_file_cachep);
 785
 786         ceph_fscache_unregister();
 787 }
 788
 789 /*
 790  * ceph_umount_begin - initiate forced umount.  Tear down down the
 791  * mount, skipping steps that may hang while waiting for server(s).
 792  */
 793 static void ceph_umount_begin(struct super_block *sb)
 794 {
 795         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
 796
 797         dout("ceph_umount_begin - starting forced umount\n");
 798         if (!fsc)
 799                 return;
 800         fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
 801         ceph_mdsc_force_umount(fsc->mdsc);
 802         return;
 803 }
 804
 805 static int ceph_remount(struct super_block *sb, int *flags, char *data)
 806 {
 807         sync_filesystem(sb);
 808         return 0;
 809 }
 810
 811 static const struct super_operations ceph_super_ops = {
 812         .alloc_inode    = ceph_alloc_inode,
 813         .destroy_inode  = ceph_destroy_inode,
 814         .write_inode    = ceph_write_inode,
 815         .drop_inode     = ceph_drop_inode,
 816         .sync_fs        = ceph_sync_fs,
 817         .put_super      = ceph_put_super,
 818         .remount_fs     = ceph_remount,
 819         .show_options   = ceph_show_options,
 820         .statfs         = ceph_statfs,
 821         .umount_begin   = ceph_umount_begin,
 822 };
 823
 824 /*
 825  * Bootstrap mount by opening the root directory.  Note the mount
 826  * @started time from caller, and time out if this takes too long.
 827  */
 828 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
 829                                        const char *path,
 830                                        unsigned long started)
 831 {
 832         struct ceph_mds_client *mdsc = fsc->mdsc;
 833         struct ceph_mds_request *req = NULL;
 834         int err;
 835         struct dentry *root;
 836
 837         /* open dir */
 838         dout("open_root_inode opening '%s'\n", path);
 839         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
 840         if (IS_ERR(req))
 841                 return ERR_CAST(req);
 842         req->r_path1 = kstrdup(path, GFP_NOFS);
 843         if (!req->r_path1) {
 844                 root = ERR_PTR(-ENOMEM);
 845                 goto out;
 846         }
 847
 848         req->r_ino1.ino = CEPH_INO_ROOT;
 849         req->r_ino1.snap = CEPH_NOSNAP;
 850         req->r_started = started;
 851         req->r_timeout = fsc->client->options->mount_timeout;
 852         req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
 853         req->r_num_caps = 2;
 854         err = ceph_mdsc_do_request(mdsc, NULL, req);
 855         if (err == 0) {
 856                 struct inode *inode = req->r_target_inode;
 857                 req->r_target_inode = NULL;
 858                 dout("open_root_inode success\n");
 859                 root = d_make_root(inode);
 860                 if (!root) {
 861                         root = ERR_PTR(-ENOMEM);
 862                         goto out;
 863                 }
 864                 dout("open_root_inode success, root dentry is %p\n", root);
 865         } else {
 866                 root = ERR_PTR(err);
 867         }
 868 out:
 869         ceph_mdsc_put_request(req);
 870         return root;
 871 }
 872
 873 /*
 874  * mount: join the ceph cluster, and open root directory.
 875  */
 876 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc)
 877 {
 878         int err;
 879         unsigned long started = jiffies;  /* note the start time */
 880         struct dentry *root;
 881
 882         dout("mount start %p\n", fsc);
 883         mutex_lock(&fsc->client->mount_mutex);
 884
 885         if (!fsc->sb->s_root) {
 886                 const char *path = fsc->mount_options->server_path ?
 887                                      fsc->mount_options->server_path + 1 : "";
 888
 889                 err = __ceph_open_session(fsc->client, started);
 890                 if (err < 0)
 891                         goto out;
 892
 893                 /* setup fscache */
 894                 if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
 895                         err = ceph_fscache_register_fs(fsc);
 896                         if (err < 0)
 897                                 goto out;
 898                 }
 899
 900                 dout("mount opening path '%s'\n", path);
 901
 902                 err = ceph_fs_debugfs_init(fsc);
 903                 if (err < 0)
 904                         goto out;
 905
 906                 root = open_root_dentry(fsc, path, started);
 907                 if (IS_ERR(root)) {
 908                         err = PTR_ERR(root);
 909                         goto out;
 910                 }
 911                 fsc->sb->s_root = dget(root);
 912         } else {
 913                 root = dget(fsc->sb->s_root);
 914         }
 915
 916         fsc->mount_state = CEPH_MOUNT_MOUNTED;
 917         dout("mount success\n");
 918         mutex_unlock(&fsc->client->mount_mutex);
 919         return root;
 920
 921 out:
 922         mutex_unlock(&fsc->client->mount_mutex);
 923         return ERR_PTR(err);
 924 }
 925
 926 static int ceph_set_super(struct super_block *s, void *data)
 927 {
 928         struct ceph_fs_client *fsc = data;
 929         int ret;
 930
 931         dout("set_super %p data %p\n", s, data);
 932
 933         s->s_flags = fsc->mount_options->sb_flags;
 934         s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
 935
 936         s->s_xattr = ceph_xattr_handlers;
 937         s->s_fs_info = fsc;
 938         fsc->sb = s;
 939
 940         s->s_op = &ceph_super_ops;
 941         s->s_d_op = &ceph_dentry_ops;
 942         s->s_export_op = &ceph_export_ops;
 943
 944         s->s_time_gran = 1000;  /* 1000 ns == 1 us */
 945
 946         ret = set_anon_super(s, NULL);  /* what is that second arg for? */
 947         if (ret != 0)
 948                 goto fail;
 949
 950         return ret;
 951
 952 fail:
 953         s->s_fs_info = NULL;
 954         fsc->sb = NULL;
 955         return ret;
 956 }
 957
 958 /*
 959  * share superblock if same fs AND options
 960  */
 961 static int ceph_compare_super(struct super_block *sb, void *data)
 962 {
 963         struct ceph_fs_client *new = data;
 964         struct ceph_mount_options *fsopt = new->mount_options;
 965         struct ceph_options *opt = new->client->options;
 966         struct ceph_fs_client *other = ceph_sb_to_client(sb);
 967
 968         dout("ceph_compare_super %p\n", sb);
 969
 970         if (compare_mount_options(fsopt, opt, other)) {
 971                 dout("monitor(s)/mount options don't match\n");
 972                 return 0;
 973         }
 974         if ((opt->flags & CEPH_OPT_FSID) &&
 975             ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
 976                 dout("fsid doesn't match\n");
 977                 return 0;
 978         }
 979         if (fsopt->sb_flags != other->mount_options->sb_flags) {
 980                 dout("flags differ\n");
 981                 return 0;
 982         }
 983         return 1;
 984 }
 985
 986 /*
 987  * construct our own bdi so we can control readahead, etc.
 988  */
 989 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
 990
 991 static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
 992 {
 993         int err;
 994
 995         err = super_setup_bdi_name(sb, "ceph-%ld",
 996                                    atomic_long_inc_return(&bdi_seq));
 997         if (err)
 998                 return err;
 999
1000         /* set ra_pages based on rasize mount option? */
1001         sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
1002
1003         /* set io_pages based on max osd read size */
1004         sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
1005
1006         return 0;
1007 }
1008
1009 static struct dentry *ceph_mount(struct file_system_type *fs_type,
1010                        int flags, const char *dev_name, void *data)
1011 {
1012         struct super_block *sb;
1013         struct ceph_fs_client *fsc;
1014         struct dentry *res;
1015         int err;
1016         int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
1017         struct ceph_mount_options *fsopt = NULL;
1018         struct ceph_options *opt = NULL;
1019
1020         dout("ceph_mount\n");
1021
1022 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1023         flags |= MS_POSIXACL;
1024 #endif
1025         err = parse_mount_options(&fsopt, &opt, flags, data, dev_name);
1026         if (err < 0) {
1027                 res = ERR_PTR(err);
1028                 goto out_final;
1029         }
1030
1031         /* create client (which we may/may not use) */
1032         fsc = create_fs_client(fsopt, opt);
1033         if (IS_ERR(fsc)) {
1034                 res = ERR_CAST(fsc);
1035                 destroy_mount_options(fsopt);
1036                 ceph_destroy_options(opt);
1037                 goto out_final;
1038         }
1039
1040         err = ceph_mdsc_init(fsc);
1041         if (err < 0) {
1042                 res = ERR_PTR(err);
1043                 goto out;
1044         }
1045
1046         if (ceph_test_opt(fsc->client, NOSHARE))
1047                 compare_super = NULL;
1048         sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc);
1049         if (IS_ERR(sb)) {
1050                 res = ERR_CAST(sb);
1051                 goto out;
1052         }
1053
1054         if (ceph_sb_to_client(sb) != fsc) {
1055                 ceph_mdsc_destroy(fsc);
1056                 destroy_fs_client(fsc);
1057                 fsc = ceph_sb_to_client(sb);
1058                 dout("get_sb got existing client %p\n", fsc);
1059         } else {
1060                 dout("get_sb using new client %p\n", fsc);
1061                 err = ceph_setup_bdi(sb, fsc);
1062                 if (err < 0) {
1063                         res = ERR_PTR(err);
1064                         goto out_splat;
1065                 }
1066         }
1067
1068         res = ceph_real_mount(fsc);
1069         if (IS_ERR(res))
1070                 goto out_splat;
1071         dout("root %p inode %p ino %llx.%llx\n", res,
1072              d_inode(res), ceph_vinop(d_inode(res)));
1073         return res;
1074
1075 out_splat:
1076         if (!ceph_mdsmap_is_cluster_available(fsc->mdsc->mdsmap)) {
1077                 pr_info("No mds server is up or the cluster is laggy\n");
1078                 err = -EHOSTUNREACH;
1079         }
1080
1081         ceph_mdsc_close_sessions(fsc->mdsc);
1082         deactivate_locked_super(sb);
1083         goto out_final;
1084
1085 out:
1086         ceph_mdsc_destroy(fsc);
1087         destroy_fs_client(fsc);
1088 out_final:
1089         dout("ceph_mount fail %ld\n", PTR_ERR(res));
1090         return res;
1091 }
1092
1093 static void ceph_kill_sb(struct super_block *s)
1094 {
1095         struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1096         dev_t dev = s->s_dev;
1097
1098         dout("kill_sb %p\n", s);
1099
1100         ceph_mdsc_pre_umount(fsc->mdsc);
1101         generic_shutdown_super(s);
1102
1103         fsc->client->extra_mon_dispatch = NULL;
1104         ceph_fs_debugfs_cleanup(fsc);
1105
1106         ceph_fscache_unregister_fs(fsc);
1107
1108         ceph_mdsc_destroy(fsc);
1109
1110         destroy_fs_client(fsc);
1111         free_anon_bdev(dev);
1112 }
1113
1114 static struct file_system_type ceph_fs_type = {
1115         .owner          = THIS_MODULE,
1116         .name           = "ceph",
1117         .mount          = ceph_mount,
1118         .kill_sb        = ceph_kill_sb,
1119         .fs_flags       = FS_RENAME_DOES_D_MOVE,
1120 };
1121 MODULE_ALIAS_FS("ceph");
1122
1123 static int __init init_ceph(void)
1124 {
1125         int ret = init_caches();
1126         if (ret)
1127                 goto out;
1128
1129         ceph_flock_init();
1130         ceph_xattr_init();
1131         ret = register_filesystem(&ceph_fs_type);
1132         if (ret)
1133                 goto out_xattr;
1134
1135         pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1136
1137         return 0;
1138
1139 out_xattr:
1140         ceph_xattr_exit();
1141         destroy_caches();
1142 out:
1143         return ret;
1144 }
1145
1146 static void __exit exit_ceph(void)
1147 {
1148         dout("exit_ceph\n");
1149         unregister_filesystem(&ceph_fs_type);
1150         ceph_xattr_exit();
1151         destroy_caches();
1152 }
1153
1154 module_init(init_ceph);
1155 module_exit(exit_ceph);
1156
1157 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1158 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1159 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1160 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1161 MODULE_LICENSE("GPL");