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