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_mon_client *monc = &fsc->client->monc;
  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(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
  80         /*
  81          * By default use root quota for stats; fallback to overall filesystem
  82          * usage if using 'noquotadf' mount option or if the root dir doesn't
  83          * have max_bytes quota set.
  84          */
  85         if (ceph_test_mount_opt(fsc, NOQUOTADF) ||
  86             !ceph_quota_update_statfs(fsc, buf)) {
  87                 buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
  88                 buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
  89                 buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
  90         }
  91
  92         buf->f_files = le64_to_cpu(st.num_objects);
  93         buf->f_ffree = -1;
  94         buf->f_namelen = NAME_MAX;
  95
  96         /* Must convert the fsid, for consistent values across arches */
  97         mutex_lock(&monc->mutex);
  98         fsid = le64_to_cpu(*(__le64 *)(&monc->monmap->fsid)) ^
  99                le64_to_cpu(*((__le64 *)&monc->monmap->fsid + 1));
 100         mutex_unlock(&monc->mutex);
 101
 102         buf->f_fsid.val[0] = fsid & 0xffffffff;
 103         buf->f_fsid.val[1] = fsid >> 32;
 104
 105         return 0;
 106 }
 107
 108 static int ceph_sync_fs(struct super_block *sb, int wait)
 109 {
 110         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
 111
 112         if (!wait) {
 113                 dout("sync_fs (non-blocking)\n");
 114                 ceph_flush_dirty_caps(fsc->mdsc);
 115                 dout("sync_fs (non-blocking) done\n");
 116                 return 0;
 117         }
 118
 119         dout("sync_fs (blocking)\n");
 120         ceph_osdc_sync(&fsc->client->osdc);
 121         ceph_mdsc_sync(fsc->mdsc);
 122         dout("sync_fs (blocking) done\n");
 123         return 0;
 124 }
 125
 126 /*
 127  * mount options
 128  */
 129 enum {
 130         Opt_wsize,
 131         Opt_rsize,
 132         Opt_rasize,
 133         Opt_caps_wanted_delay_min,
 134         Opt_caps_wanted_delay_max,
 135         Opt_readdir_max_entries,
 136         Opt_readdir_max_bytes,
 137         Opt_congestion_kb,
 138         Opt_last_int,
 139         /* int args above */
 140         Opt_snapdirname,
 141         Opt_mds_namespace,
 142         Opt_fscache_uniq,
 143         Opt_last_string,
 144         /* string args above */
 145         Opt_dirstat,
 146         Opt_nodirstat,
 147         Opt_rbytes,
 148         Opt_norbytes,
 149         Opt_asyncreaddir,
 150         Opt_noasyncreaddir,
 151         Opt_dcache,
 152         Opt_nodcache,
 153         Opt_ino32,
 154         Opt_noino32,
 155         Opt_fscache,
 156         Opt_nofscache,
 157         Opt_poolperm,
 158         Opt_nopoolperm,
 159         Opt_require_active_mds,
 160         Opt_norequire_active_mds,
 161 #ifdef CONFIG_CEPH_FS_POSIX_ACL
 162         Opt_acl,
 163 #endif
 164         Opt_noacl,
 165         Opt_quotadf,
 166         Opt_noquotadf,
 167 };
 168
 169 static match_table_t fsopt_tokens = {
 170         {Opt_wsize, "wsize=%d"},
 171         {Opt_rsize, "rsize=%d"},
 172         {Opt_rasize, "rasize=%d"},
 173         {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
 174         {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
 175         {Opt_readdir_max_entries, "readdir_max_entries=%d"},
 176         {Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
 177         {Opt_congestion_kb, "write_congestion_kb=%d"},
 178         /* int args above */
 179         {Opt_snapdirname, "snapdirname=%s"},
 180         {Opt_mds_namespace, "mds_namespace=%s"},
 181         {Opt_fscache_uniq, "fsc=%s"},
 182         /* string args above */
 183         {Opt_dirstat, "dirstat"},
 184         {Opt_nodirstat, "nodirstat"},
 185         {Opt_rbytes, "rbytes"},
 186         {Opt_norbytes, "norbytes"},
 187         {Opt_asyncreaddir, "asyncreaddir"},
 188         {Opt_noasyncreaddir, "noasyncreaddir"},
 189         {Opt_dcache, "dcache"},
 190         {Opt_nodcache, "nodcache"},
 191         {Opt_ino32, "ino32"},
 192         {Opt_noino32, "noino32"},
 193         {Opt_fscache, "fsc"},
 194         {Opt_nofscache, "nofsc"},
 195         {Opt_poolperm, "poolperm"},
 196         {Opt_nopoolperm, "nopoolperm"},
 197         {Opt_require_active_mds, "require_active_mds"},
 198         {Opt_norequire_active_mds, "norequire_active_mds"},
 199 #ifdef CONFIG_CEPH_FS_POSIX_ACL
 200         {Opt_acl, "acl"},
 201 #endif
 202         {Opt_noacl, "noacl"},
 203         {Opt_quotadf, "quotadf"},
 204         {Opt_noquotadf, "noquotadf"},
 205         {-1, NULL}
 206 };
 207
 208 /*
 209  * Remove adjacent slashes and then the trailing slash, unless it is
 210  * the only remaining character.
 211  *
 212  * E.g. "//dir1////dir2///" --> "/dir1/dir2", "///" --> "/".
 213  */
 214 static void canonicalize_path(char *path)
 215 {
 216         int i, j = 0;
 217
 218         for (i = 0; path[i] != '\0'; i++) {
 219                 if (path[i] != '/' || j < 1 || path[j - 1] != '/')
 220                         path[j++] = path[i];
 221         }
 222
 223         if (j > 1 && path[j - 1] == '/')
 224                 j--;
 225         path[j] = '\0';
 226 }
 227
 228 static int parse_fsopt_token(char *c, void *private)
 229 {
 230         struct ceph_mount_options *fsopt = private;
 231         substring_t argstr[MAX_OPT_ARGS];
 232         int token, intval, ret;
 233
 234         token = match_token((char *)c, fsopt_tokens, argstr);
 235         if (token < 0)
 236                 return -EINVAL;
 237
 238         if (token < Opt_last_int) {
 239                 ret = match_int(&argstr[0], &intval);
 240                 if (ret < 0) {
 241                         pr_err("bad option arg (not int) at '%s'\n", c);
 242                         return ret;
 243                 }
 244                 dout("got int token %d val %d\n", token, intval);
 245         } else if (token > Opt_last_int && token < Opt_last_string) {
 246                 dout("got string token %d val %s\n", token,
 247                      argstr[0].from);
 248         } else {
 249                 dout("got token %d\n", token);
 250         }
 251
 252         switch (token) {
 253         case Opt_snapdirname:
 254                 kfree(fsopt->snapdir_name);
 255                 fsopt->snapdir_name = kstrndup(argstr[0].from,
 256                                                argstr[0].to-argstr[0].from,
 257                                                GFP_KERNEL);
 258                 if (!fsopt->snapdir_name)
 259                         return -ENOMEM;
 260                 break;
 261         case Opt_mds_namespace:
 262                 kfree(fsopt->mds_namespace);
 263                 fsopt->mds_namespace = kstrndup(argstr[0].from,
 264                                                 argstr[0].to-argstr[0].from,
 265                                                 GFP_KERNEL);
 266                 if (!fsopt->mds_namespace)
 267                         return -ENOMEM;
 268                 break;
 269         case Opt_fscache_uniq:
 270 #ifdef CONFIG_CEPH_FSCACHE
 271                 kfree(fsopt->fscache_uniq);
 272                 fsopt->fscache_uniq = kstrndup(argstr[0].from,
 273                                                argstr[0].to-argstr[0].from,
 274                                                GFP_KERNEL);
 275                 if (!fsopt->fscache_uniq)
 276                         return -ENOMEM;
 277                 fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
 278                 break;
 279 #else
 280                 pr_err("fscache support is disabled\n");
 281                 return -EINVAL;
 282 #endif
 283         case Opt_wsize:
 284                 if (intval < (int)PAGE_SIZE || intval > CEPH_MAX_WRITE_SIZE)
 285                         return -EINVAL;
 286                 fsopt->wsize = ALIGN(intval, PAGE_SIZE);
 287                 break;
 288         case Opt_rsize:
 289                 if (intval < (int)PAGE_SIZE || intval > CEPH_MAX_READ_SIZE)
 290                         return -EINVAL;
 291                 fsopt->rsize = ALIGN(intval, PAGE_SIZE);
 292                 break;
 293         case Opt_rasize:
 294                 if (intval < 0)
 295                         return -EINVAL;
 296                 fsopt->rasize = ALIGN(intval, PAGE_SIZE);
 297                 break;
 298         case Opt_caps_wanted_delay_min:
 299                 if (intval < 1)
 300                         return -EINVAL;
 301                 fsopt->caps_wanted_delay_min = intval;
 302                 break;
 303         case Opt_caps_wanted_delay_max:
 304                 if (intval < 1)
 305                         return -EINVAL;
 306                 fsopt->caps_wanted_delay_max = intval;
 307                 break;
 308         case Opt_readdir_max_entries:
 309                 if (intval < 1)
 310                         return -EINVAL;
 311                 fsopt->max_readdir = intval;
 312                 break;
 313         case Opt_readdir_max_bytes:
 314                 if (intval < (int)PAGE_SIZE && intval != 0)
 315                         return -EINVAL;
 316                 fsopt->max_readdir_bytes = intval;
 317                 break;
 318         case Opt_congestion_kb:
 319                 if (intval < 1024) /* at least 1M */
 320                         return -EINVAL;
 321                 fsopt->congestion_kb = intval;
 322                 break;
 323         case Opt_dirstat:
 324                 fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
 325                 break;
 326         case Opt_nodirstat:
 327                 fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
 328                 break;
 329         case Opt_rbytes:
 330                 fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
 331                 break;
 332         case Opt_norbytes:
 333                 fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
 334                 break;
 335         case Opt_asyncreaddir:
 336                 fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
 337                 break;
 338         case Opt_noasyncreaddir:
 339                 fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
 340                 break;
 341         case Opt_dcache:
 342                 fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
 343                 break;
 344         case Opt_nodcache:
 345                 fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
 346                 break;
 347         case Opt_ino32:
 348                 fsopt->flags |= CEPH_MOUNT_OPT_INO32;
 349                 break;
 350         case Opt_noino32:
 351                 fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
 352                 break;
 353         case Opt_fscache:
 354 #ifdef CONFIG_CEPH_FSCACHE
 355                 fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
 356                 kfree(fsopt->fscache_uniq);
 357                 fsopt->fscache_uniq = NULL;
 358                 break;
 359 #else
 360                 pr_err("fscache support is disabled\n");
 361                 return -EINVAL;
 362 #endif
 363         case Opt_nofscache:
 364                 fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
 365                 kfree(fsopt->fscache_uniq);
 366                 fsopt->fscache_uniq = NULL;
 367                 break;
 368         case Opt_poolperm:
 369                 fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
 370                 break;
 371         case Opt_nopoolperm:
 372                 fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
 373                 break;
 374         case Opt_require_active_mds:
 375                 fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT;
 376                 break;
 377         case Opt_norequire_active_mds:
 378                 fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT;
 379                 break;
 380         case Opt_quotadf:
 381                 fsopt->flags &= ~CEPH_MOUNT_OPT_NOQUOTADF;
 382                 break;
 383         case Opt_noquotadf:
 384                 fsopt->flags |= CEPH_MOUNT_OPT_NOQUOTADF;
 385                 break;
 386 #ifdef CONFIG_CEPH_FS_POSIX_ACL
 387         case Opt_acl:
 388                 fsopt->sb_flags |= SB_POSIXACL;
 389                 break;
 390 #endif
 391         case Opt_noacl:
 392                 fsopt->sb_flags &= ~SB_POSIXACL;
 393                 break;
 394         default:
 395                 BUG_ON(token);
 396         }
 397         return 0;
 398 }
 399
 400 static void destroy_mount_options(struct ceph_mount_options *args)
 401 {
 402         dout("destroy_mount_options %p\n", args);
 403         kfree(args->snapdir_name);
 404         kfree(args->mds_namespace);
 405         kfree(args->server_path);
 406         kfree(args->fscache_uniq);
 407         kfree(args);
 408 }
 409
 410 static int strcmp_null(const char *s1, const char *s2)
 411 {
 412         if (!s1 && !s2)
 413                 return 0;
 414         if (s1 && !s2)
 415                 return -1;
 416         if (!s1 && s2)
 417                 return 1;
 418         return strcmp(s1, s2);
 419 }
 420
 421 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
 422                                  struct ceph_options *new_opt,
 423                                  struct ceph_fs_client *fsc)
 424 {
 425         struct ceph_mount_options *fsopt1 = new_fsopt;
 426         struct ceph_mount_options *fsopt2 = fsc->mount_options;
 427         int ofs = offsetof(struct ceph_mount_options, snapdir_name);
 428         int ret;
 429
 430         ret = memcmp(fsopt1, fsopt2, ofs);
 431         if (ret)
 432                 return ret;
 433
 434         ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
 435         if (ret)
 436                 return ret;
 437
 438         ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
 439         if (ret)
 440                 return ret;
 441
 442         ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
 443         if (ret)
 444                 return ret;
 445
 446         ret = strcmp_null(fsopt1->fscache_uniq, fsopt2->fscache_uniq);
 447         if (ret)
 448                 return ret;
 449
 450         return ceph_compare_options(new_opt, fsc->client);
 451 }
 452
 453 static int parse_mount_options(struct ceph_mount_options **pfsopt,
 454                                struct ceph_options **popt,
 455                                int flags, char *options,
 456                                const char *dev_name)
 457 {
 458         struct ceph_mount_options *fsopt;
 459         const char *dev_name_end;
 460         int err;
 461
 462         if (!dev_name || !*dev_name)
 463                 return -EINVAL;
 464
 465         fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
 466         if (!fsopt)
 467                 return -ENOMEM;
 468
 469         dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
 470
 471         fsopt->sb_flags = flags;
 472         fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
 473
 474         fsopt->wsize = CEPH_MAX_WRITE_SIZE;
 475         fsopt->rsize = CEPH_MAX_READ_SIZE;
 476         fsopt->rasize = CEPH_RASIZE_DEFAULT;
 477         fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
 478         if (!fsopt->snapdir_name) {
 479                 err = -ENOMEM;
 480                 goto out;
 481         }
 482
 483         fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
 484         fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
 485         fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
 486         fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
 487         fsopt->congestion_kb = default_congestion_kb();
 488
 489         /*
 490          * Distinguish the server list from the path in "dev_name".
 491          * Internally we do not include the leading '/' in the path.
 492          *
 493          * "dev_name" will look like:
 494          *     <server_spec>[,<server_spec>...]:[<path>]
 495          * where
 496          *     <server_spec> is <ip>[:<port>]
 497          *     <path> is optional, but if present must begin with '/'
 498          */
 499         dev_name_end = strchr(dev_name, '/');
 500         if (dev_name_end) {
 501                 /*
 502                  * The server_path will include the whole chars from userland
 503                  * including the leading '/'.
 504                  */
 505                 fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
 506                 if (!fsopt->server_path) {
 507                         err = -ENOMEM;
 508                         goto out;
 509                 }
 510
 511                 canonicalize_path(fsopt->server_path);
 512         } else {
 513                 dev_name_end = dev_name + strlen(dev_name);
 514         }
 515         err = -EINVAL;
 516         dev_name_end--;         /* back up to ':' separator */
 517         if (dev_name_end < dev_name || *dev_name_end != ':') {
 518                 pr_err("device name is missing path (no : separator in %s)\n",
 519                                 dev_name);
 520                 goto out;
 521         }
 522         dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
 523         if (fsopt->server_path)
 524                 dout("server path '%s'\n", fsopt->server_path);
 525
 526         *popt = ceph_parse_options(options, dev_name, dev_name_end,
 527                                  parse_fsopt_token, (void *)fsopt);
 528         if (IS_ERR(*popt)) {
 529                 err = PTR_ERR(*popt);
 530                 goto out;
 531         }
 532
 533         /* success */
 534         *pfsopt = fsopt;
 535         return 0;
 536
 537 out:
 538         destroy_mount_options(fsopt);
 539         return err;
 540 }
 541
 542 /**
 543  * ceph_show_options - Show mount options in /proc/mounts
 544  * @m: seq_file to write to
 545  * @root: root of that (sub)tree
 546  */
 547 static int ceph_show_options(struct seq_file *m, struct dentry *root)
 548 {
 549         struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
 550         struct ceph_mount_options *fsopt = fsc->mount_options;
 551         size_t pos;
 552         int ret;
 553
 554         /* a comma between MNT/MS and client options */
 555         seq_putc(m, ',');
 556         pos = m->count;
 557
 558         ret = ceph_print_client_options(m, fsc->client);
 559         if (ret)
 560                 return ret;
 561
 562         /* retract our comma if no client options */
 563         if (m->count == pos)
 564                 m->count--;
 565
 566         if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
 567                 seq_puts(m, ",dirstat");
 568         if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
 569                 seq_puts(m, ",rbytes");
 570         if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
 571                 seq_puts(m, ",noasyncreaddir");
 572         if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
 573                 seq_puts(m, ",nodcache");
 574         if (fsopt->flags & CEPH_MOUNT_OPT_INO32)
 575                 seq_puts(m, ",ino32");
 576         if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) {
 577                 seq_show_option(m, "fsc", fsopt->fscache_uniq);
 578         }
 579         if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
 580                 seq_puts(m, ",nopoolperm");
 581         if (fsopt->flags & CEPH_MOUNT_OPT_NOQUOTADF)
 582                 seq_puts(m, ",noquotadf");
 583
 584 #ifdef CONFIG_CEPH_FS_POSIX_ACL
 585         if (fsopt->sb_flags & SB_POSIXACL)
 586                 seq_puts(m, ",acl");
 587         else
 588                 seq_puts(m, ",noacl");
 589 #endif
 590
 591         if (fsopt->mds_namespace)
 592                 seq_show_option(m, "mds_namespace", fsopt->mds_namespace);
 593         if (fsopt->wsize != CEPH_MAX_WRITE_SIZE)
 594                 seq_printf(m, ",wsize=%d", fsopt->wsize);
 595         if (fsopt->rsize != CEPH_MAX_READ_SIZE)
 596                 seq_printf(m, ",rsize=%d", fsopt->rsize);
 597         if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
 598                 seq_printf(m, ",rasize=%d", fsopt->rasize);
 599         if (fsopt->congestion_kb != default_congestion_kb())
 600                 seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
 601         if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
 602                 seq_printf(m, ",caps_wanted_delay_min=%d",
 603                          fsopt->caps_wanted_delay_min);
 604         if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
 605                 seq_printf(m, ",caps_wanted_delay_max=%d",
 606                            fsopt->caps_wanted_delay_max);
 607         if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
 608                 seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
 609         if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
 610                 seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
 611         if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
 612                 seq_show_option(m, "snapdirname", fsopt->snapdir_name);
 613
 614         return 0;
 615 }
 616
 617 /*
 618  * handle any mon messages the standard library doesn't understand.
 619  * return error if we don't either.
 620  */
 621 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
 622 {
 623         struct ceph_fs_client *fsc = client->private;
 624         int type = le16_to_cpu(msg->hdr.type);
 625
 626         switch (type) {
 627         case CEPH_MSG_MDS_MAP:
 628                 ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
 629                 return 0;
 630         case CEPH_MSG_FS_MAP_USER:
 631                 ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
 632                 return 0;
 633         default:
 634                 return -1;
 635         }
 636 }
 637
 638 /*
 639  * create a new fs client
 640  *
 641  * Success or not, this function consumes @fsopt and @opt.
 642  */
 643 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
 644                                         struct ceph_options *opt)
 645 {
 646         struct ceph_fs_client *fsc;
 647         int page_count;
 648         size_t size;
 649         int err;
 650
 651         fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
 652         if (!fsc) {
 653                 err = -ENOMEM;
 654                 goto fail;
 655         }
 656
 657         fsc->client = ceph_create_client(opt, fsc);
 658         if (IS_ERR(fsc->client)) {
 659                 err = PTR_ERR(fsc->client);
 660                 goto fail;
 661         }
 662         opt = NULL; /* fsc->client now owns this */
 663
 664         fsc->client->extra_mon_dispatch = extra_mon_dispatch;
 665         fsc->client->osdc.abort_on_full = true;
 666
 667         if (!fsopt->mds_namespace) {
 668                 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
 669                                    0, true);
 670         } else {
 671                 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
 672                                    0, false);
 673         }
 674
 675         fsc->mount_options = fsopt;
 676
 677         fsc->sb = NULL;
 678         fsc->mount_state = CEPH_MOUNT_MOUNTING;
 679
 680         atomic_long_set(&fsc->writeback_count, 0);
 681
 682         err = -ENOMEM;
 683         /*
 684          * The number of concurrent works can be high but they don't need
 685          * to be processed in parallel, limit concurrency.
 686          */
 687         fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
 688         if (!fsc->wb_wq)
 689                 goto fail_client;
 690         fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
 691         if (!fsc->pg_inv_wq)
 692                 goto fail_wb_wq;
 693         fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
 694         if (!fsc->trunc_wq)
 695                 goto fail_pg_inv_wq;
 696
 697         /* set up mempools */
 698         err = -ENOMEM;
 699         page_count = fsc->mount_options->wsize >> PAGE_SHIFT;
 700         size = sizeof (struct page *) * (page_count ? page_count : 1);
 701         fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
 702         if (!fsc->wb_pagevec_pool)
 703                 goto fail_trunc_wq;
 704
 705         /* caps */
 706         fsc->min_caps = fsopt->max_readdir;
 707
 708         return fsc;
 709
 710 fail_trunc_wq:
 711         destroy_workqueue(fsc->trunc_wq);
 712 fail_pg_inv_wq:
 713         destroy_workqueue(fsc->pg_inv_wq);
 714 fail_wb_wq:
 715         destroy_workqueue(fsc->wb_wq);
 716 fail_client:
 717         ceph_destroy_client(fsc->client);
 718 fail:
 719         kfree(fsc);
 720         if (opt)
 721                 ceph_destroy_options(opt);
 722         destroy_mount_options(fsopt);
 723         return ERR_PTR(err);
 724 }
 725
 726 static void flush_fs_workqueues(struct ceph_fs_client *fsc)
 727 {
 728         flush_workqueue(fsc->wb_wq);
 729         flush_workqueue(fsc->pg_inv_wq);
 730         flush_workqueue(fsc->trunc_wq);
 731 }
 732
 733 static void destroy_fs_client(struct ceph_fs_client *fsc)
 734 {
 735         dout("destroy_fs_client %p\n", fsc);
 736
 737         destroy_workqueue(fsc->wb_wq);
 738         destroy_workqueue(fsc->pg_inv_wq);
 739         destroy_workqueue(fsc->trunc_wq);
 740
 741         mempool_destroy(fsc->wb_pagevec_pool);
 742
 743         destroy_mount_options(fsc->mount_options);
 744
 745         ceph_destroy_client(fsc->client);
 746
 747         kfree(fsc);
 748         dout("destroy_fs_client %p done\n", fsc);
 749 }
 750
 751 /*
 752  * caches
 753  */
 754 struct kmem_cache *ceph_inode_cachep;
 755 struct kmem_cache *ceph_cap_cachep;
 756 struct kmem_cache *ceph_cap_flush_cachep;
 757 struct kmem_cache *ceph_dentry_cachep;
 758 struct kmem_cache *ceph_file_cachep;
 759 struct kmem_cache *ceph_dir_file_cachep;
 760
 761 static void ceph_inode_init_once(void *foo)
 762 {
 763         struct ceph_inode_info *ci = foo;
 764         inode_init_once(&ci->vfs_inode);
 765 }
 766
 767 static int __init init_caches(void)
 768 {
 769         int error = -ENOMEM;
 770
 771         ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
 772                                       sizeof(struct ceph_inode_info),
 773                                       __alignof__(struct ceph_inode_info),
 774                                       SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
 775                                       SLAB_ACCOUNT, ceph_inode_init_once);
 776         if (!ceph_inode_cachep)
 777                 return -ENOMEM;
 778
 779         ceph_cap_cachep = KMEM_CACHE(ceph_cap, SLAB_MEM_SPREAD);
 780         if (!ceph_cap_cachep)
 781                 goto bad_cap;
 782         ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
 783                                            SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 784         if (!ceph_cap_flush_cachep)
 785                 goto bad_cap_flush;
 786
 787         ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
 788                                         SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 789         if (!ceph_dentry_cachep)
 790                 goto bad_dentry;
 791
 792         ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD);
 793         if (!ceph_file_cachep)
 794                 goto bad_file;
 795
 796         ceph_dir_file_cachep = KMEM_CACHE(ceph_dir_file_info, SLAB_MEM_SPREAD);
 797         if (!ceph_dir_file_cachep)
 798                 goto bad_dir_file;
 799
 800         error = ceph_fscache_register();
 801         if (error)
 802                 goto bad_fscache;
 803
 804         return 0;
 805
 806 bad_fscache:
 807         kmem_cache_destroy(ceph_dir_file_cachep);
 808 bad_dir_file:
 809         kmem_cache_destroy(ceph_file_cachep);
 810 bad_file:
 811         kmem_cache_destroy(ceph_dentry_cachep);
 812 bad_dentry:
 813         kmem_cache_destroy(ceph_cap_flush_cachep);
 814 bad_cap_flush:
 815         kmem_cache_destroy(ceph_cap_cachep);
 816 bad_cap:
 817         kmem_cache_destroy(ceph_inode_cachep);
 818         return error;
 819 }
 820
 821 static void destroy_caches(void)
 822 {
 823         /*
 824          * Make sure all delayed rcu free inodes are flushed before we
 825          * destroy cache.
 826          */
 827         rcu_barrier();
 828
 829         kmem_cache_destroy(ceph_inode_cachep);
 830         kmem_cache_destroy(ceph_cap_cachep);
 831         kmem_cache_destroy(ceph_cap_flush_cachep);
 832         kmem_cache_destroy(ceph_dentry_cachep);
 833         kmem_cache_destroy(ceph_file_cachep);
 834         kmem_cache_destroy(ceph_dir_file_cachep);
 835
 836         ceph_fscache_unregister();
 837 }
 838
 839 /*
 840  * ceph_umount_begin - initiate forced umount.  Tear down down the
 841  * mount, skipping steps that may hang while waiting for server(s).
 842  */
 843 static void ceph_umount_begin(struct super_block *sb)
 844 {
 845         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
 846
 847         dout("ceph_umount_begin - starting forced umount\n");
 848         if (!fsc)
 849                 return;
 850         fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
 851         ceph_osdc_abort_requests(&fsc->client->osdc, -EIO);
 852         ceph_mdsc_force_umount(fsc->mdsc);
 853         return;
 854 }
 855
 856 static int ceph_remount(struct super_block *sb, int *flags, char *data)
 857 {
 858         sync_filesystem(sb);
 859         return 0;
 860 }
 861
 862 static const struct super_operations ceph_super_ops = {
 863         .alloc_inode    = ceph_alloc_inode,
 864         .destroy_inode  = ceph_destroy_inode,
 865         .write_inode    = ceph_write_inode,
 866         .drop_inode     = ceph_drop_inode,
 867         .evict_inode    = ceph_evict_inode,
 868         .sync_fs        = ceph_sync_fs,
 869         .put_super      = ceph_put_super,
 870         .remount_fs     = ceph_remount,
 871         .show_options   = ceph_show_options,
 872         .statfs         = ceph_statfs,
 873         .umount_begin   = ceph_umount_begin,
 874 };
 875
 876 /*
 877  * Bootstrap mount by opening the root directory.  Note the mount
 878  * @started time from caller, and time out if this takes too long.
 879  */
 880 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
 881                                        const char *path,
 882                                        unsigned long started)
 883 {
 884         struct ceph_mds_client *mdsc = fsc->mdsc;
 885         struct ceph_mds_request *req = NULL;
 886         int err;
 887         struct dentry *root;
 888
 889         /* open dir */
 890         dout("open_root_inode opening '%s'\n", path);
 891         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
 892         if (IS_ERR(req))
 893                 return ERR_CAST(req);
 894         req->r_path1 = kstrdup(path, GFP_NOFS);
 895         if (!req->r_path1) {
 896                 root = ERR_PTR(-ENOMEM);
 897                 goto out;
 898         }
 899
 900         req->r_ino1.ino = CEPH_INO_ROOT;
 901         req->r_ino1.snap = CEPH_NOSNAP;
 902         req->r_started = started;
 903         req->r_timeout = fsc->client->options->mount_timeout;
 904         req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
 905         req->r_num_caps = 2;
 906         err = ceph_mdsc_do_request(mdsc, NULL, req);
 907         if (err == 0) {
 908                 struct inode *inode = req->r_target_inode;
 909                 req->r_target_inode = NULL;
 910                 dout("open_root_inode success\n");
 911                 root = d_make_root(inode);
 912                 if (!root) {
 913                         root = ERR_PTR(-ENOMEM);
 914                         goto out;
 915                 }
 916                 dout("open_root_inode success, root dentry is %p\n", root);
 917         } else {
 918                 root = ERR_PTR(err);
 919         }
 920 out:
 921         ceph_mdsc_put_request(req);
 922         return root;
 923 }
 924
 925 /*
 926  * mount: join the ceph cluster, and open root directory.
 927  */
 928 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc)
 929 {
 930         int err;
 931         unsigned long started = jiffies;  /* note the start time */
 932         struct dentry *root;
 933
 934         dout("mount start %p\n", fsc);
 935         mutex_lock(&fsc->client->mount_mutex);
 936
 937         if (!fsc->sb->s_root) {
 938                 const char *path = fsc->mount_options->server_path ?
 939                                      fsc->mount_options->server_path + 1 : "";
 940
 941                 err = __ceph_open_session(fsc->client, started);
 942                 if (err < 0)
 943                         goto out;
 944
 945                 /* setup fscache */
 946                 if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
 947                         err = ceph_fscache_register_fs(fsc);
 948                         if (err < 0)
 949                                 goto out;
 950                 }
 951
 952                 dout("mount opening path '%s'\n", path);
 953
 954                 err = ceph_fs_debugfs_init(fsc);
 955                 if (err < 0)
 956                         goto out;
 957
 958                 root = open_root_dentry(fsc, path, started);
 959                 if (IS_ERR(root)) {
 960                         err = PTR_ERR(root);
 961                         goto out;
 962                 }
 963                 fsc->sb->s_root = dget(root);
 964         } else {
 965                 root = dget(fsc->sb->s_root);
 966         }
 967
 968         fsc->mount_state = CEPH_MOUNT_MOUNTED;
 969         dout("mount success\n");
 970         mutex_unlock(&fsc->client->mount_mutex);
 971         return root;
 972
 973 out:
 974         mutex_unlock(&fsc->client->mount_mutex);
 975         return ERR_PTR(err);
 976 }
 977
 978 static int ceph_set_super(struct super_block *s, void *data)
 979 {
 980         struct ceph_fs_client *fsc = data;
 981         int ret;
 982
 983         dout("set_super %p data %p\n", s, data);
 984
 985         s->s_flags = fsc->mount_options->sb_flags;
 986         s->s_maxbytes = MAX_LFS_FILESIZE;
 987
 988         s->s_xattr = ceph_xattr_handlers;
 989         s->s_fs_info = fsc;
 990         fsc->sb = s;
 991         fsc->max_file_size = 1ULL << 40; /* temp value until we get mdsmap */
 992
 993         s->s_op = &ceph_super_ops;
 994         s->s_d_op = &ceph_dentry_ops;
 995         s->s_export_op = &ceph_export_ops;
 996
 997         s->s_time_gran = 1000;  /* 1000 ns == 1 us */
 998
 999         ret = set_anon_super(s, NULL);  /* what is that second arg for? */
1000         if (ret != 0)
1001                 goto fail;
1002
1003         return ret;
1004
1005 fail:
1006         s->s_fs_info = NULL;
1007         fsc->sb = NULL;
1008         return ret;
1009 }
1010
1011 /*
1012  * share superblock if same fs AND options
1013  */
1014 static int ceph_compare_super(struct super_block *sb, void *data)
1015 {
1016         struct ceph_fs_client *new = data;
1017         struct ceph_mount_options *fsopt = new->mount_options;
1018         struct ceph_options *opt = new->client->options;
1019         struct ceph_fs_client *other = ceph_sb_to_client(sb);
1020
1021         dout("ceph_compare_super %p\n", sb);
1022
1023         if (compare_mount_options(fsopt, opt, other)) {
1024                 dout("monitor(s)/mount options don't match\n");
1025                 return 0;
1026         }
1027         if ((opt->flags & CEPH_OPT_FSID) &&
1028             ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
1029                 dout("fsid doesn't match\n");
1030                 return 0;
1031         }
1032         if (fsopt->sb_flags != other->mount_options->sb_flags) {
1033                 dout("flags differ\n");
1034                 return 0;
1035         }
1036         return 1;
1037 }
1038
1039 /*
1040  * construct our own bdi so we can control readahead, etc.
1041  */
1042 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1043
1044 static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
1045 {
1046         int err;
1047
1048         err = super_setup_bdi_name(sb, "ceph-%ld",
1049                                    atomic_long_inc_return(&bdi_seq));
1050         if (err)
1051                 return err;
1052
1053         /* set ra_pages based on rasize mount option? */
1054         sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
1055
1056         /* set io_pages based on max osd read size */
1057         sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
1058
1059         return 0;
1060 }
1061
1062 static struct dentry *ceph_mount(struct file_system_type *fs_type,
1063                        int flags, const char *dev_name, void *data)
1064 {
1065         struct super_block *sb;
1066         struct ceph_fs_client *fsc;
1067         struct dentry *res;
1068         int err;
1069         int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
1070         struct ceph_mount_options *fsopt = NULL;
1071         struct ceph_options *opt = NULL;
1072
1073         dout("ceph_mount\n");
1074
1075 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1076         flags |= SB_POSIXACL;
1077 #endif
1078         err = parse_mount_options(&fsopt, &opt, flags, data, dev_name);
1079         if (err < 0) {
1080                 res = ERR_PTR(err);
1081                 goto out_final;
1082         }
1083
1084         /* create client (which we may/may not use) */
1085         fsc = create_fs_client(fsopt, opt);
1086         if (IS_ERR(fsc)) {
1087                 res = ERR_CAST(fsc);
1088                 goto out_final;
1089         }
1090
1091         err = ceph_mdsc_init(fsc);
1092         if (err < 0) {
1093                 res = ERR_PTR(err);
1094                 goto out;
1095         }
1096
1097         if (ceph_test_opt(fsc->client, NOSHARE))
1098                 compare_super = NULL;
1099         sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc);
1100         if (IS_ERR(sb)) {
1101                 res = ERR_CAST(sb);
1102                 goto out;
1103         }
1104
1105         if (ceph_sb_to_client(sb) != fsc) {
1106                 ceph_mdsc_destroy(fsc);
1107                 destroy_fs_client(fsc);
1108                 fsc = ceph_sb_to_client(sb);
1109                 dout("get_sb got existing client %p\n", fsc);
1110         } else {
1111                 dout("get_sb using new client %p\n", fsc);
1112                 err = ceph_setup_bdi(sb, fsc);
1113                 if (err < 0) {
1114                         res = ERR_PTR(err);
1115                         goto out_splat;
1116                 }
1117         }
1118
1119         res = ceph_real_mount(fsc);
1120         if (IS_ERR(res))
1121                 goto out_splat;
1122         dout("root %p inode %p ino %llx.%llx\n", res,
1123              d_inode(res), ceph_vinop(d_inode(res)));
1124         return res;
1125
1126 out_splat:
1127         if (!ceph_mdsmap_is_cluster_available(fsc->mdsc->mdsmap)) {
1128                 pr_info("No mds server is up or the cluster is laggy\n");
1129                 err = -EHOSTUNREACH;
1130         }
1131
1132         ceph_mdsc_close_sessions(fsc->mdsc);
1133         deactivate_locked_super(sb);
1134         goto out_final;
1135
1136 out:
1137         ceph_mdsc_destroy(fsc);
1138         destroy_fs_client(fsc);
1139 out_final:
1140         dout("ceph_mount fail %ld\n", PTR_ERR(res));
1141         return res;
1142 }
1143
1144 static void ceph_kill_sb(struct super_block *s)
1145 {
1146         struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1147         dev_t dev = s->s_dev;
1148
1149         dout("kill_sb %p\n", s);
1150
1151         ceph_mdsc_pre_umount(fsc->mdsc);
1152         flush_fs_workqueues(fsc);
1153
1154         generic_shutdown_super(s);
1155
1156         fsc->client->extra_mon_dispatch = NULL;
1157         ceph_fs_debugfs_cleanup(fsc);
1158
1159         ceph_fscache_unregister_fs(fsc);
1160
1161         ceph_mdsc_destroy(fsc);
1162
1163         destroy_fs_client(fsc);
1164         free_anon_bdev(dev);
1165 }
1166
1167 static struct file_system_type ceph_fs_type = {
1168         .owner          = THIS_MODULE,
1169         .name           = "ceph",
1170         .mount          = ceph_mount,
1171         .kill_sb        = ceph_kill_sb,
1172         .fs_flags       = FS_RENAME_DOES_D_MOVE,
1173 };
1174 MODULE_ALIAS_FS("ceph");
1175
1176 static int __init init_ceph(void)
1177 {
1178         int ret = init_caches();
1179         if (ret)
1180                 goto out;
1181
1182         ceph_flock_init();
1183         ceph_xattr_init();
1184         ret = register_filesystem(&ceph_fs_type);
1185         if (ret)
1186                 goto out_xattr;
1187
1188         pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1189
1190         return 0;
1191
1192 out_xattr:
1193         ceph_xattr_exit();
1194         destroy_caches();
1195 out:
1196         return ret;
1197 }
1198
1199 static void __exit exit_ceph(void)
1200 {
1201         dout("exit_ceph\n");
1202         unregister_filesystem(&ceph_fs_type);
1203         ceph_xattr_exit();
1204         destroy_caches();
1205 }
1206
1207 module_init(init_ceph);
1208 module_exit(exit_ceph);
1209
1210 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1211 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1212 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1213 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1214 MODULE_LICENSE("GPL");