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
  21 #include <linux/ceph/decode.h>
  22 #include <linux/ceph/mon_client.h>
  23 #include <linux/ceph/auth.h>
  24 #include <linux/ceph/debugfs.h>
  25
  26 /*
  27  * Ceph superblock operations
  28  *
  29  * Handle the basics of mounting, unmounting.
  30  */
  31
  32 /*
  33  * super ops
  34  */
  35 static void ceph_put_super(struct super_block *s)
  36 {
  37         struct ceph_fs_client *fsc = ceph_sb_to_client(s);
  38
  39         dout("put_super\n");
  40         ceph_mdsc_close_sessions(fsc->mdsc);
  41
  42         /*
  43          * ensure we release the bdi before put_anon_super releases
  44          * the device name.
  45          */
  46         if (s->s_bdi == &fsc->backing_dev_info) {
  47                 bdi_unregister(&fsc->backing_dev_info);
  48                 s->s_bdi = NULL;
  49         }
  50
  51         return;
  52 }
  53
  54 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
  55 {
  56         struct ceph_fs_client *fsc = ceph_inode_to_client(dentry->d_inode);
  57         struct ceph_monmap *monmap = fsc->client->monc.monmap;
  58         struct ceph_statfs st;
  59         u64 fsid;
  60         int err;
  61
  62         dout("statfs\n");
  63         err = ceph_monc_do_statfs(&fsc->client->monc, &st);
  64         if (err < 0)
  65                 return err;
  66
  67         /* fill in kstatfs */
  68         buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
  69
  70         /*
  71          * express utilization in terms of large blocks to avoid
  72          * overflow on 32-bit machines.
  73          *
  74          * NOTE: for the time being, we make bsize == frsize to humor
  75          * not-yet-ancient versions of glibc that are broken.
  76          * Someday, we will probably want to report a real block
  77          * size...  whatever that may mean for a network file system!
  78          */
  79         buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
  80         buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
  81         buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
  82         buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
  83         buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
  84
  85         buf->f_files = le64_to_cpu(st.num_objects);
  86         buf->f_ffree = -1;
  87         buf->f_namelen = NAME_MAX;
  88
  89         /* leave fsid little-endian, regardless of host endianness */
  90         fsid = *(u64 *)(&monmap->fsid) ^ *((u64 *)&monmap->fsid + 1);
  91         buf->f_fsid.val[0] = fsid & 0xffffffff;
  92         buf->f_fsid.val[1] = fsid >> 32;
  93
  94         return 0;
  95 }
  96
  97
  98 static int ceph_sync_fs(struct super_block *sb, int wait)
  99 {
 100         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
 101
 102         if (!wait) {
 103                 dout("sync_fs (non-blocking)\n");
 104                 ceph_flush_dirty_caps(fsc->mdsc);
 105                 dout("sync_fs (non-blocking) done\n");
 106                 return 0;
 107         }
 108
 109         dout("sync_fs (blocking)\n");
 110         ceph_osdc_sync(&fsc->client->osdc);
 111         ceph_mdsc_sync(fsc->mdsc);
 112         dout("sync_fs (blocking) done\n");
 113         return 0;
 114 }
 115
 116 /*
 117  * mount options
 118  */
 119 enum {
 120         Opt_wsize,
 121         Opt_rsize,
 122         Opt_rasize,
 123         Opt_caps_wanted_delay_min,
 124         Opt_caps_wanted_delay_max,
 125         Opt_cap_release_safety,
 126         Opt_readdir_max_entries,
 127         Opt_readdir_max_bytes,
 128         Opt_congestion_kb,
 129         Opt_last_int,
 130         /* int args above */
 131         Opt_snapdirname,
 132         Opt_last_string,
 133         /* string args above */
 134         Opt_dirstat,
 135         Opt_nodirstat,
 136         Opt_rbytes,
 137         Opt_norbytes,
 138         Opt_noasyncreaddir,
 139         Opt_ino32,
 140 };
 141
 142 static match_table_t fsopt_tokens = {
 143         {Opt_wsize, "wsize=%d"},
 144         {Opt_rsize, "rsize=%d"},
 145         {Opt_rasize, "rasize=%d"},
 146         {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
 147         {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
 148         {Opt_cap_release_safety, "cap_release_safety=%d"},
 149         {Opt_readdir_max_entries, "readdir_max_entries=%d"},
 150         {Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
 151         {Opt_congestion_kb, "write_congestion_kb=%d"},
 152         /* int args above */
 153         {Opt_snapdirname, "snapdirname=%s"},
 154         /* string args above */
 155         {Opt_dirstat, "dirstat"},
 156         {Opt_nodirstat, "nodirstat"},
 157         {Opt_rbytes, "rbytes"},
 158         {Opt_norbytes, "norbytes"},
 159         {Opt_noasyncreaddir, "noasyncreaddir"},
 160         {Opt_ino32, "ino32"},
 161         {-1, NULL}
 162 };
 163
 164 static int parse_fsopt_token(char *c, void *private)
 165 {
 166         struct ceph_mount_options *fsopt = private;
 167         substring_t argstr[MAX_OPT_ARGS];
 168         int token, intval, ret;
 169
 170         token = match_token((char *)c, fsopt_tokens, argstr);
 171         if (token < 0)
 172                 return -EINVAL;
 173
 174         if (token < Opt_last_int) {
 175                 ret = match_int(&argstr[0], &intval);
 176                 if (ret < 0) {
 177                         pr_err("bad mount option arg (not int) "
 178                                "at '%s'\n", c);
 179                         return ret;
 180                 }
 181                 dout("got int token %d val %d\n", token, intval);
 182         } else if (token > Opt_last_int && token < Opt_last_string) {
 183                 dout("got string token %d val %s\n", token,
 184                      argstr[0].from);
 185         } else {
 186                 dout("got token %d\n", token);
 187         }
 188
 189         switch (token) {
 190         case Opt_snapdirname:
 191                 kfree(fsopt->snapdir_name);
 192                 fsopt->snapdir_name = kstrndup(argstr[0].from,
 193                                                argstr[0].to-argstr[0].from,
 194                                                GFP_KERNEL);
 195                 if (!fsopt->snapdir_name)
 196                         return -ENOMEM;
 197                 break;
 198
 199                 /* misc */
 200         case Opt_wsize:
 201                 fsopt->wsize = intval;
 202                 break;
 203         case Opt_rsize:
 204                 fsopt->rsize = intval;
 205                 break;
 206         case Opt_rasize:
 207                 fsopt->rasize = intval;
 208                 break;
 209         case Opt_caps_wanted_delay_min:
 210                 fsopt->caps_wanted_delay_min = intval;
 211                 break;
 212         case Opt_caps_wanted_delay_max:
 213                 fsopt->caps_wanted_delay_max = intval;
 214                 break;
 215         case Opt_readdir_max_entries:
 216                 fsopt->max_readdir = intval;
 217                 break;
 218         case Opt_readdir_max_bytes:
 219                 fsopt->max_readdir_bytes = intval;
 220                 break;
 221         case Opt_congestion_kb:
 222                 fsopt->congestion_kb = intval;
 223                 break;
 224         case Opt_dirstat:
 225                 fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
 226                 break;
 227         case Opt_nodirstat:
 228                 fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
 229                 break;
 230         case Opt_rbytes:
 231                 fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
 232                 break;
 233         case Opt_norbytes:
 234                 fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
 235                 break;
 236         case Opt_noasyncreaddir:
 237                 fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
 238                 break;
 239         case Opt_ino32:
 240                 fsopt->flags |= CEPH_MOUNT_OPT_INO32;
 241                 break;
 242         default:
 243                 BUG_ON(token);
 244         }
 245         return 0;
 246 }
 247
 248 static void destroy_mount_options(struct ceph_mount_options *args)
 249 {
 250         dout("destroy_mount_options %p\n", args);
 251         kfree(args->snapdir_name);
 252         kfree(args);
 253 }
 254
 255 static int strcmp_null(const char *s1, const char *s2)
 256 {
 257         if (!s1 && !s2)
 258                 return 0;
 259         if (s1 && !s2)
 260                 return -1;
 261         if (!s1 && s2)
 262                 return 1;
 263         return strcmp(s1, s2);
 264 }
 265
 266 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
 267                                  struct ceph_options *new_opt,
 268                                  struct ceph_fs_client *fsc)
 269 {
 270         struct ceph_mount_options *fsopt1 = new_fsopt;
 271         struct ceph_mount_options *fsopt2 = fsc->mount_options;
 272         int ofs = offsetof(struct ceph_mount_options, snapdir_name);
 273         int ret;
 274
 275         ret = memcmp(fsopt1, fsopt2, ofs);
 276         if (ret)
 277                 return ret;
 278
 279         ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
 280         if (ret)
 281                 return ret;
 282
 283         return ceph_compare_options(new_opt, fsc->client);
 284 }
 285
 286 static int parse_mount_options(struct ceph_mount_options **pfsopt,
 287                                struct ceph_options **popt,
 288                                int flags, char *options,
 289                                const char *dev_name,
 290                                const char **path)
 291 {
 292         struct ceph_mount_options *fsopt;
 293         const char *dev_name_end;
 294         int err = -ENOMEM;
 295
 296         fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
 297         if (!fsopt)
 298                 return -ENOMEM;
 299
 300         dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
 301
 302         fsopt->sb_flags = flags;
 303         fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
 304
 305         fsopt->rsize = CEPH_RSIZE_DEFAULT;
 306         fsopt->rasize = CEPH_RASIZE_DEFAULT;
 307         fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
 308         fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
 309         fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
 310         fsopt->cap_release_safety = CEPH_CAP_RELEASE_SAFETY_DEFAULT;
 311         fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
 312         fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
 313         fsopt->congestion_kb = default_congestion_kb();
 314
 315         /* ip1[:port1][,ip2[:port2]...]:/subdir/in/fs */
 316         err = -EINVAL;
 317         if (!dev_name)
 318                 goto out;
 319         *path = strstr(dev_name, ":/");
 320         if (*path == NULL) {
 321                 pr_err("device name is missing path (no :/ in %s)\n",
 322                                 dev_name);
 323                 goto out;
 324         }
 325         dev_name_end = *path;
 326         dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
 327
 328         /* path on server */
 329         *path += 2;
 330         dout("server path '%s'\n", *path);
 331
 332         err = ceph_parse_options(popt, options, dev_name, dev_name_end,
 333                                  parse_fsopt_token, (void *)fsopt);
 334         if (err)
 335                 goto out;
 336
 337         /* success */
 338         *pfsopt = fsopt;
 339         return 0;
 340
 341 out:
 342         destroy_mount_options(fsopt);
 343         return err;
 344 }
 345
 346 /**
 347  * ceph_show_options - Show mount options in /proc/mounts
 348  * @m: seq_file to write to
 349  * @mnt: mount descriptor
 350  */
 351 static int ceph_show_options(struct seq_file *m, struct vfsmount *mnt)
 352 {
 353         struct ceph_fs_client *fsc = ceph_sb_to_client(mnt->mnt_sb);
 354         struct ceph_mount_options *fsopt = fsc->mount_options;
 355         struct ceph_options *opt = fsc->client->options;
 356
 357         if (opt->flags & CEPH_OPT_FSID)
 358                 seq_printf(m, ",fsid=%pU", &opt->fsid);
 359         if (opt->flags & CEPH_OPT_NOSHARE)
 360                 seq_puts(m, ",noshare");
 361         if (opt->flags & CEPH_OPT_NOCRC)
 362                 seq_puts(m, ",nocrc");
 363
 364         if (opt->name)
 365                 seq_printf(m, ",name=%s", opt->name);
 366         if (opt->key)
 367                 seq_puts(m, ",secret=<hidden>");
 368
 369         if (opt->mount_timeout != CEPH_MOUNT_TIMEOUT_DEFAULT)
 370                 seq_printf(m, ",mount_timeout=%d", opt->mount_timeout);
 371         if (opt->osd_idle_ttl != CEPH_OSD_IDLE_TTL_DEFAULT)
 372                 seq_printf(m, ",osd_idle_ttl=%d", opt->osd_idle_ttl);
 373         if (opt->osd_timeout != CEPH_OSD_TIMEOUT_DEFAULT)
 374                 seq_printf(m, ",osdtimeout=%d", opt->osd_timeout);
 375         if (opt->osd_keepalive_timeout != CEPH_OSD_KEEPALIVE_DEFAULT)
 376                 seq_printf(m, ",osdkeepalivetimeout=%d",
 377                            opt->osd_keepalive_timeout);
 378
 379         if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
 380                 seq_puts(m, ",dirstat");
 381         if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES) == 0)
 382                 seq_puts(m, ",norbytes");
 383         if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
 384                 seq_puts(m, ",noasyncreaddir");
 385
 386         if (fsopt->wsize)
 387                 seq_printf(m, ",wsize=%d", fsopt->wsize);
 388         if (fsopt->rsize != CEPH_RSIZE_DEFAULT)
 389                 seq_printf(m, ",rsize=%d", fsopt->rsize);
 390         if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
 391                 seq_printf(m, ",rasize=%d", fsopt->rasize);
 392         if (fsopt->congestion_kb != default_congestion_kb())
 393                 seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
 394         if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
 395                 seq_printf(m, ",caps_wanted_delay_min=%d",
 396                          fsopt->caps_wanted_delay_min);
 397         if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
 398                 seq_printf(m, ",caps_wanted_delay_max=%d",
 399                            fsopt->caps_wanted_delay_max);
 400         if (fsopt->cap_release_safety != CEPH_CAP_RELEASE_SAFETY_DEFAULT)
 401                 seq_printf(m, ",cap_release_safety=%d",
 402                            fsopt->cap_release_safety);
 403         if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
 404                 seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
 405         if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
 406                 seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
 407         if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
 408                 seq_printf(m, ",snapdirname=%s", fsopt->snapdir_name);
 409         return 0;
 410 }
 411
 412 /*
 413  * handle any mon messages the standard library doesn't understand.
 414  * return error if we don't either.
 415  */
 416 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
 417 {
 418         struct ceph_fs_client *fsc = client->private;
 419         int type = le16_to_cpu(msg->hdr.type);
 420
 421         switch (type) {
 422         case CEPH_MSG_MDS_MAP:
 423                 ceph_mdsc_handle_map(fsc->mdsc, msg);
 424                 return 0;
 425
 426         default:
 427                 return -1;
 428         }
 429 }
 430
 431 /*
 432  * create a new fs client
 433  */
 434 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
 435                                         struct ceph_options *opt)
 436 {
 437         struct ceph_fs_client *fsc;
 438         const unsigned supported_features =
 439                 CEPH_FEATURE_FLOCK |
 440                 CEPH_FEATURE_DIRLAYOUTHASH;
 441         const unsigned required_features = 0;
 442         int err = -ENOMEM;
 443
 444         fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
 445         if (!fsc)
 446                 return ERR_PTR(-ENOMEM);
 447
 448         fsc->client = ceph_create_client(opt, fsc, supported_features,
 449                                          required_features);
 450         if (IS_ERR(fsc->client)) {
 451                 err = PTR_ERR(fsc->client);
 452                 goto fail;
 453         }
 454         fsc->client->extra_mon_dispatch = extra_mon_dispatch;
 455         fsc->client->monc.want_mdsmap = 1;
 456
 457         fsc->mount_options = fsopt;
 458
 459         fsc->sb = NULL;
 460         fsc->mount_state = CEPH_MOUNT_MOUNTING;
 461
 462         atomic_long_set(&fsc->writeback_count, 0);
 463
 464         err = bdi_init(&fsc->backing_dev_info);
 465         if (err < 0)
 466                 goto fail_client;
 467
 468         err = -ENOMEM;
 469         /*
 470          * The number of concurrent works can be high but they don't need
 471          * to be processed in parallel, limit concurrency.
 472          */
 473         fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
 474         if (fsc->wb_wq == NULL)
 475                 goto fail_bdi;
 476         fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
 477         if (fsc->pg_inv_wq == NULL)
 478                 goto fail_wb_wq;
 479         fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
 480         if (fsc->trunc_wq == NULL)
 481                 goto fail_pg_inv_wq;
 482
 483         /* set up mempools */
 484         err = -ENOMEM;
 485         fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10,
 486                               fsc->mount_options->wsize >> PAGE_CACHE_SHIFT);
 487         if (!fsc->wb_pagevec_pool)
 488                 goto fail_trunc_wq;
 489
 490         /* caps */
 491         fsc->min_caps = fsopt->max_readdir;
 492
 493         return fsc;
 494
 495 fail_trunc_wq:
 496         destroy_workqueue(fsc->trunc_wq);
 497 fail_pg_inv_wq:
 498         destroy_workqueue(fsc->pg_inv_wq);
 499 fail_wb_wq:
 500         destroy_workqueue(fsc->wb_wq);
 501 fail_bdi:
 502         bdi_destroy(&fsc->backing_dev_info);
 503 fail_client:
 504         ceph_destroy_client(fsc->client);
 505 fail:
 506         kfree(fsc);
 507         return ERR_PTR(err);
 508 }
 509
 510 static void destroy_fs_client(struct ceph_fs_client *fsc)
 511 {
 512         dout("destroy_fs_client %p\n", fsc);
 513
 514         destroy_workqueue(fsc->wb_wq);
 515         destroy_workqueue(fsc->pg_inv_wq);
 516         destroy_workqueue(fsc->trunc_wq);
 517
 518         bdi_destroy(&fsc->backing_dev_info);
 519
 520         mempool_destroy(fsc->wb_pagevec_pool);
 521
 522         destroy_mount_options(fsc->mount_options);
 523
 524         ceph_fs_debugfs_cleanup(fsc);
 525
 526         ceph_destroy_client(fsc->client);
 527
 528         kfree(fsc);
 529         dout("destroy_fs_client %p done\n", fsc);
 530 }
 531
 532 /*
 533  * caches
 534  */
 535 struct kmem_cache *ceph_inode_cachep;
 536 struct kmem_cache *ceph_cap_cachep;
 537 struct kmem_cache *ceph_dentry_cachep;
 538 struct kmem_cache *ceph_file_cachep;
 539
 540 static void ceph_inode_init_once(void *foo)
 541 {
 542         struct ceph_inode_info *ci = foo;
 543         inode_init_once(&ci->vfs_inode);
 544 }
 545
 546 static int __init init_caches(void)
 547 {
 548         ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
 549                                       sizeof(struct ceph_inode_info),
 550                                       __alignof__(struct ceph_inode_info),
 551                                       (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
 552                                       ceph_inode_init_once);
 553         if (ceph_inode_cachep == NULL)
 554                 return -ENOMEM;
 555
 556         ceph_cap_cachep = KMEM_CACHE(ceph_cap,
 557                                      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 558         if (ceph_cap_cachep == NULL)
 559                 goto bad_cap;
 560
 561         ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
 562                                         SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 563         if (ceph_dentry_cachep == NULL)
 564                 goto bad_dentry;
 565
 566         ceph_file_cachep = KMEM_CACHE(ceph_file_info,
 567                                       SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 568         if (ceph_file_cachep == NULL)
 569                 goto bad_file;
 570
 571         return 0;
 572
 573 bad_file:
 574         kmem_cache_destroy(ceph_dentry_cachep);
 575 bad_dentry:
 576         kmem_cache_destroy(ceph_cap_cachep);
 577 bad_cap:
 578         kmem_cache_destroy(ceph_inode_cachep);
 579         return -ENOMEM;
 580 }
 581
 582 static void destroy_caches(void)
 583 {
 584         kmem_cache_destroy(ceph_inode_cachep);
 585         kmem_cache_destroy(ceph_cap_cachep);
 586         kmem_cache_destroy(ceph_dentry_cachep);
 587         kmem_cache_destroy(ceph_file_cachep);
 588 }
 589
 590
 591 /*
 592  * ceph_umount_begin - initiate forced umount.  Tear down down the
 593  * mount, skipping steps that may hang while waiting for server(s).
 594  */
 595 static void ceph_umount_begin(struct super_block *sb)
 596 {
 597         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
 598
 599         dout("ceph_umount_begin - starting forced umount\n");
 600         if (!fsc)
 601                 return;
 602         fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
 603         return;
 604 }
 605
 606 static const struct super_operations ceph_super_ops = {
 607         .alloc_inode    = ceph_alloc_inode,
 608         .destroy_inode  = ceph_destroy_inode,
 609         .write_inode    = ceph_write_inode,
 610         .sync_fs        = ceph_sync_fs,
 611         .put_super      = ceph_put_super,
 612         .show_options   = ceph_show_options,
 613         .statfs         = ceph_statfs,
 614         .umount_begin   = ceph_umount_begin,
 615 };
 616
 617 /*
 618  * Bootstrap mount by opening the root directory.  Note the mount
 619  * @started time from caller, and time out if this takes too long.
 620  */
 621 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
 622                                        const char *path,
 623                                        unsigned long started)
 624 {
 625         struct ceph_mds_client *mdsc = fsc->mdsc;
 626         struct ceph_mds_request *req = NULL;
 627         int err;
 628         struct dentry *root;
 629
 630         /* open dir */
 631         dout("open_root_inode opening '%s'\n", path);
 632         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
 633         if (IS_ERR(req))
 634                 return ERR_CAST(req);
 635         req->r_path1 = kstrdup(path, GFP_NOFS);
 636         req->r_ino1.ino = CEPH_INO_ROOT;
 637         req->r_ino1.snap = CEPH_NOSNAP;
 638         req->r_started = started;
 639         req->r_timeout = fsc->client->options->mount_timeout * HZ;
 640         req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
 641         req->r_num_caps = 2;
 642         err = ceph_mdsc_do_request(mdsc, NULL, req);
 643         if (err == 0) {
 644                 dout("open_root_inode success\n");
 645                 if (ceph_ino(req->r_target_inode) == CEPH_INO_ROOT &&
 646                     fsc->sb->s_root == NULL) {
 647                         root = d_alloc_root(req->r_target_inode);
 648                         ceph_init_dentry(root);
 649                 } else {
 650                         root = d_obtain_alias(req->r_target_inode);
 651                 }
 652                 req->r_target_inode = NULL;
 653                 dout("open_root_inode success, root dentry is %p\n", root);
 654         } else {
 655                 root = ERR_PTR(err);
 656         }
 657         ceph_mdsc_put_request(req);
 658         return root;
 659 }
 660
 661
 662
 663
 664 /*
 665  * mount: join the ceph cluster, and open root directory.
 666  */
 667 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
 668                       const char *path)
 669 {
 670         int err;
 671         unsigned long started = jiffies;  /* note the start time */
 672         struct dentry *root;
 673         int first = 0;   /* first vfsmount for this super_block */
 674
 675         dout("mount start\n");
 676         mutex_lock(&fsc->client->mount_mutex);
 677
 678         err = __ceph_open_session(fsc->client, started);
 679         if (err < 0)
 680                 goto out;
 681
 682         dout("mount opening root\n");
 683         root = open_root_dentry(fsc, "", started);
 684         if (IS_ERR(root)) {
 685                 err = PTR_ERR(root);
 686                 goto out;
 687         }
 688         if (fsc->sb->s_root) {
 689                 dput(root);
 690         } else {
 691                 fsc->sb->s_root = root;
 692                 first = 1;
 693
 694                 err = ceph_fs_debugfs_init(fsc);
 695                 if (err < 0)
 696                         goto fail;
 697         }
 698
 699         if (path[0] == 0) {
 700                 dget(root);
 701         } else {
 702                 dout("mount opening base mountpoint\n");
 703                 root = open_root_dentry(fsc, path, started);
 704                 if (IS_ERR(root)) {
 705                         err = PTR_ERR(root);
 706                         goto fail;
 707                 }
 708         }
 709
 710         fsc->mount_state = CEPH_MOUNT_MOUNTED;
 711         dout("mount success\n");
 712         mutex_unlock(&fsc->client->mount_mutex);
 713         return root;
 714
 715 out:
 716         mutex_unlock(&fsc->client->mount_mutex);
 717         return ERR_PTR(err);
 718
 719 fail:
 720         if (first) {
 721                 dput(fsc->sb->s_root);
 722                 fsc->sb->s_root = NULL;
 723         }
 724         goto out;
 725 }
 726
 727 static int ceph_set_super(struct super_block *s, void *data)
 728 {
 729         struct ceph_fs_client *fsc = data;
 730         int ret;
 731
 732         dout("set_super %p data %p\n", s, data);
 733
 734         s->s_flags = fsc->mount_options->sb_flags;
 735         s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
 736
 737         s->s_fs_info = fsc;
 738         fsc->sb = s;
 739
 740         s->s_op = &ceph_super_ops;
 741         s->s_export_op = &ceph_export_ops;
 742
 743         s->s_time_gran = 1000;  /* 1000 ns == 1 us */
 744
 745         ret = set_anon_super(s, NULL);  /* what is that second arg for? */
 746         if (ret != 0)
 747                 goto fail;
 748
 749         return ret;
 750
 751 fail:
 752         s->s_fs_info = NULL;
 753         fsc->sb = NULL;
 754         return ret;
 755 }
 756
 757 /*
 758  * share superblock if same fs AND options
 759  */
 760 static int ceph_compare_super(struct super_block *sb, void *data)
 761 {
 762         struct ceph_fs_client *new = data;
 763         struct ceph_mount_options *fsopt = new->mount_options;
 764         struct ceph_options *opt = new->client->options;
 765         struct ceph_fs_client *other = ceph_sb_to_client(sb);
 766
 767         dout("ceph_compare_super %p\n", sb);
 768
 769         if (compare_mount_options(fsopt, opt, other)) {
 770                 dout("monitor(s)/mount options don't match\n");
 771                 return 0;
 772         }
 773         if ((opt->flags & CEPH_OPT_FSID) &&
 774             ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
 775                 dout("fsid doesn't match\n");
 776                 return 0;
 777         }
 778         if (fsopt->sb_flags != other->mount_options->sb_flags) {
 779                 dout("flags differ\n");
 780                 return 0;
 781         }
 782         return 1;
 783 }
 784
 785 /*
 786  * construct our own bdi so we can control readahead, etc.
 787  */
 788 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
 789
 790 static int ceph_register_bdi(struct super_block *sb,
 791                              struct ceph_fs_client *fsc)
 792 {
 793         int err;
 794
 795         /* set ra_pages based on rasize mount option? */
 796         if (fsc->mount_options->rasize >= PAGE_CACHE_SIZE)
 797                 fsc->backing_dev_info.ra_pages =
 798                         (fsc->mount_options->rasize + PAGE_CACHE_SIZE - 1)
 799                         >> PAGE_SHIFT;
 800         else
 801                 fsc->backing_dev_info.ra_pages =
 802                         default_backing_dev_info.ra_pages;
 803
 804         err = bdi_register(&fsc->backing_dev_info, NULL, "ceph-%d",
 805                            atomic_long_inc_return(&bdi_seq));
 806         if (!err)
 807                 sb->s_bdi = &fsc->backing_dev_info;
 808         return err;
 809 }
 810
 811 static struct dentry *ceph_mount(struct file_system_type *fs_type,
 812                        int flags, const char *dev_name, void *data)
 813 {
 814         struct super_block *sb;
 815         struct ceph_fs_client *fsc;
 816         struct dentry *res;
 817         int err;
 818         int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
 819         const char *path = NULL;
 820         struct ceph_mount_options *fsopt = NULL;
 821         struct ceph_options *opt = NULL;
 822
 823         dout("ceph_mount\n");
 824         err = parse_mount_options(&fsopt, &opt, flags, data, dev_name, &path);
 825         if (err < 0) {
 826                 res = ERR_PTR(err);
 827                 goto out_final;
 828         }
 829
 830         /* create client (which we may/may not use) */
 831         fsc = create_fs_client(fsopt, opt);
 832         if (IS_ERR(fsc)) {
 833                 res = ERR_CAST(fsc);
 834                 destroy_mount_options(fsopt);
 835                 ceph_destroy_options(opt);
 836                 goto out_final;
 837         }
 838
 839         err = ceph_mdsc_init(fsc);
 840         if (err < 0) {
 841                 res = ERR_PTR(err);
 842                 goto out;
 843         }
 844
 845         if (ceph_test_opt(fsc->client, NOSHARE))
 846                 compare_super = NULL;
 847         sb = sget(fs_type, compare_super, ceph_set_super, fsc);
 848         if (IS_ERR(sb)) {
 849                 res = ERR_CAST(sb);
 850                 goto out;
 851         }
 852
 853         if (ceph_sb_to_client(sb) != fsc) {
 854                 ceph_mdsc_destroy(fsc);
 855                 destroy_fs_client(fsc);
 856                 fsc = ceph_sb_to_client(sb);
 857                 dout("get_sb got existing client %p\n", fsc);
 858         } else {
 859                 dout("get_sb using new client %p\n", fsc);
 860                 err = ceph_register_bdi(sb, fsc);
 861                 if (err < 0) {
 862                         res = ERR_PTR(err);
 863                         goto out_splat;
 864                 }
 865         }
 866
 867         res = ceph_real_mount(fsc, path);
 868         if (IS_ERR(res))
 869                 goto out_splat;
 870         dout("root %p inode %p ino %llx.%llx\n", res,
 871              res->d_inode, ceph_vinop(res->d_inode));
 872         return res;
 873
 874 out_splat:
 875         ceph_mdsc_close_sessions(fsc->mdsc);
 876         deactivate_locked_super(sb);
 877         goto out_final;
 878
 879 out:
 880         ceph_mdsc_destroy(fsc);
 881         destroy_fs_client(fsc);
 882 out_final:
 883         dout("ceph_mount fail %ld\n", PTR_ERR(res));
 884         return res;
 885 }
 886
 887 static void ceph_kill_sb(struct super_block *s)
 888 {
 889         struct ceph_fs_client *fsc = ceph_sb_to_client(s);
 890         dout("kill_sb %p\n", s);
 891         ceph_mdsc_pre_umount(fsc->mdsc);
 892         kill_anon_super(s);    /* will call put_super after sb is r/o */
 893         ceph_mdsc_destroy(fsc);
 894         destroy_fs_client(fsc);
 895 }
 896
 897 static struct file_system_type ceph_fs_type = {
 898         .owner          = THIS_MODULE,
 899         .name           = "ceph",
 900         .mount          = ceph_mount,
 901         .kill_sb        = ceph_kill_sb,
 902         .fs_flags       = FS_RENAME_DOES_D_MOVE,
 903 };
 904
 905 #define _STRINGIFY(x) #x
 906 #define STRINGIFY(x) _STRINGIFY(x)
 907
 908 static int __init init_ceph(void)
 909 {
 910         int ret = init_caches();
 911         if (ret)
 912                 goto out;
 913
 914         ret = register_filesystem(&ceph_fs_type);
 915         if (ret)
 916                 goto out_icache;
 917
 918         pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
 919
 920         return 0;
 921
 922 out_icache:
 923         destroy_caches();
 924 out:
 925         return ret;
 926 }
 927
 928 static void __exit exit_ceph(void)
 929 {
 930         dout("exit_ceph\n");
 931         unregister_filesystem(&ceph_fs_type);
 932         destroy_caches();
 933 }
 934
 935 module_init(init_ceph);
 936 module_exit(exit_ceph);
 937
 938 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
 939 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
 940 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
 941 MODULE_DESCRIPTION("Ceph filesystem for Linux");
 942 MODULE_LICENSE("GPL");