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