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