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