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