SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/ceph/super.c

   1
   2 #include "ceph_debug.h"
   3
   4 #include <linux/backing-dev.h>
   5 #include <linux/fs.h>
   6 #include <linux/inet.h>
   7 #include <linux/in6.h>
   8 #include <linux/module.h>
   9 #include <linux/mount.h>
  10 #include <linux/parser.h>
  11 #include <linux/rwsem.h>
  12 #include <linux/sched.h>
  13 #include <linux/seq_file.h>
  14 #include <linux/statfs.h>
  15 #include <linux/string.h>
  16 #include <linux/version.h>
  17 #include <linux/vmalloc.h>
  18
  19 #include "decode.h"
  20 #include "super.h"
  21 #include "mon_client.h"
  22
  23 /*
  24  * Ceph superblock operations
  25  *
  26  * Handle the basics of mounting, unmounting.
  27  */
  28
  29
  30 /*
  31  * find filename portion of a path (/foo/bar/baz -> baz)
  32  */
  33 const char *ceph_file_part(const char *s, int len)
  34 {
  35         const char *e = s + len;
  36
  37         while (e != s && *(e-1) != '/')
  38                 e--;
  39         return e;
  40 }
  41
  42
  43 /*
  44  * super ops
  45  */
  46 static void ceph_put_super(struct super_block *s)
  47 {
  48         struct ceph_client *cl = ceph_client(s);
  49
  50         dout("put_super\n");
  51         ceph_mdsc_close_sessions(&cl->mdsc);
  52         return;
  53 }
  54
  55 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
  56 {
  57         struct ceph_client *client = ceph_inode_to_client(dentry->d_inode);
  58         struct ceph_monmap *monmap = client->monc.monmap;
  59         struct ceph_statfs st;
  60         u64 fsid;
  61         int err;
  62
  63         dout("statfs\n");
  64         err = ceph_monc_do_statfs(&client->monc, &st);
  65         if (err < 0)
  66                 return err;
  67
  68         /* fill in kstatfs */
  69         buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
  70
  71         /*
  72          * express utilization in terms of large blocks to avoid
  73          * overflow on 32-bit machines.
  74          */
  75         buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
  76         buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
  77         buf->f_bfree = (le64_to_cpu(st.kb) - le64_to_cpu(st.kb_used)) >>
  78                 (CEPH_BLOCK_SHIFT-10);
  79         buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
  80
  81         buf->f_files = le64_to_cpu(st.num_objects);
  82         buf->f_ffree = -1;
  83         buf->f_namelen = PATH_MAX;
  84         buf->f_frsize = PAGE_CACHE_SIZE;
  85
  86         /* leave fsid little-endian, regardless of host endianness */
  87         fsid = *(u64 *)(&monmap->fsid) ^ *((u64 *)&monmap->fsid + 1);
  88         buf->f_fsid.val[0] = fsid & 0xffffffff;
  89         buf->f_fsid.val[1] = fsid >> 32;
  90
  91         return 0;
  92 }
  93
  94
  95 static int ceph_syncfs(struct super_block *sb, int wait)
  96 {
  97         dout("sync_fs %d\n", wait);
  98         ceph_osdc_sync(&ceph_client(sb)->osdc);
  99         ceph_mdsc_sync(&ceph_client(sb)->mdsc);
 100         dout("sync_fs %d done\n", wait);
 101         return 0;
 102 }
 103
 104
 105 /**
 106  * ceph_show_options - Show mount options in /proc/mounts
 107  * @m: seq_file to write to
 108  * @mnt: mount descriptor
 109  */
 110 static int ceph_show_options(struct seq_file *m, struct vfsmount *mnt)
 111 {
 112         struct ceph_client *client = ceph_sb_to_client(mnt->mnt_sb);
 113         struct ceph_mount_args *args = client->mount_args;
 114
 115         if (args->flags & CEPH_OPT_FSID)
 116                 seq_printf(m, ",fsidmajor=%llu,fsidminor%llu",
 117                            le64_to_cpu(*(__le64 *)&args->fsid.fsid[0]),
 118                            le64_to_cpu(*(__le64 *)&args->fsid.fsid[8]));
 119         if (args->flags & CEPH_OPT_NOSHARE)
 120                 seq_puts(m, ",noshare");
 121         if (args->flags & CEPH_OPT_DIRSTAT)
 122                 seq_puts(m, ",dirstat");
 123         if ((args->flags & CEPH_OPT_RBYTES) == 0)
 124                 seq_puts(m, ",norbytes");
 125         if (args->flags & CEPH_OPT_NOCRC)
 126                 seq_puts(m, ",nocrc");
 127         if (args->flags & CEPH_OPT_NOASYNCREADDIR)
 128                 seq_puts(m, ",noasyncreaddir");
 129         if (strcmp(args->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
 130                 seq_printf(m, ",snapdirname=%s", args->snapdir_name);
 131         if (args->name)
 132                 seq_printf(m, ",name=%s", args->name);
 133         if (args->secret)
 134                 seq_puts(m, ",secret=<hidden>");
 135         return 0;
 136 }
 137
 138 /*
 139  * caches
 140  */
 141 struct kmem_cache *ceph_inode_cachep;
 142 struct kmem_cache *ceph_cap_cachep;
 143 struct kmem_cache *ceph_dentry_cachep;
 144 struct kmem_cache *ceph_file_cachep;
 145
 146 static void ceph_inode_init_once(void *foo)
 147 {
 148         struct ceph_inode_info *ci = foo;
 149         inode_init_once(&ci->vfs_inode);
 150 }
 151
 152 static int __init init_caches(void)
 153 {
 154         ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
 155                                       sizeof(struct ceph_inode_info),
 156                                       __alignof__(struct ceph_inode_info),
 157                                       (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
 158                                       ceph_inode_init_once);
 159         if (ceph_inode_cachep == NULL)
 160                 return -ENOMEM;
 161
 162         ceph_cap_cachep = KMEM_CACHE(ceph_cap,
 163                                      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 164         if (ceph_cap_cachep == NULL)
 165                 goto bad_cap;
 166
 167         ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
 168                                         SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 169         if (ceph_dentry_cachep == NULL)
 170                 goto bad_dentry;
 171
 172         ceph_file_cachep = KMEM_CACHE(ceph_file_info,
 173                                       SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 174         if (ceph_file_cachep == NULL)
 175                 goto bad_file;
 176
 177         return 0;
 178
 179 bad_file:
 180         kmem_cache_destroy(ceph_dentry_cachep);
 181 bad_dentry:
 182         kmem_cache_destroy(ceph_cap_cachep);
 183 bad_cap:
 184         kmem_cache_destroy(ceph_inode_cachep);
 185         return -ENOMEM;
 186 }
 187
 188 static void destroy_caches(void)
 189 {
 190         kmem_cache_destroy(ceph_inode_cachep);
 191         kmem_cache_destroy(ceph_cap_cachep);
 192         kmem_cache_destroy(ceph_dentry_cachep);
 193         kmem_cache_destroy(ceph_file_cachep);
 194 }
 195
 196
 197 /*
 198  * ceph_umount_begin - initiate forced umount.  Tear down down the
 199  * mount, skipping steps that may hang while waiting for server(s).
 200  */
 201 static void ceph_umount_begin(struct super_block *sb)
 202 {
 203         struct ceph_client *client = ceph_sb_to_client(sb);
 204
 205         dout("ceph_umount_begin - starting forced umount\n");
 206         if (!client)
 207                 return;
 208         client->mount_state = CEPH_MOUNT_SHUTDOWN;
 209         return;
 210 }
 211
 212 static const struct super_operations ceph_super_ops = {
 213         .alloc_inode    = ceph_alloc_inode,
 214         .destroy_inode  = ceph_destroy_inode,
 215         .write_inode    = ceph_write_inode,
 216         .sync_fs        = ceph_syncfs,
 217         .put_super      = ceph_put_super,
 218         .show_options   = ceph_show_options,
 219         .statfs         = ceph_statfs,
 220         .umount_begin   = ceph_umount_begin,
 221 };
 222
 223
 224 const char *ceph_msg_type_name(int type)
 225 {
 226         switch (type) {
 227         case CEPH_MSG_SHUTDOWN: return "shutdown";
 228         case CEPH_MSG_PING: return "ping";
 229         case CEPH_MSG_AUTH: return "auth";
 230         case CEPH_MSG_AUTH_REPLY: return "auth_reply";
 231         case CEPH_MSG_MON_MAP: return "mon_map";
 232         case CEPH_MSG_MON_GET_MAP: return "mon_get_map";
 233         case CEPH_MSG_MON_SUBSCRIBE: return "mon_subscribe";
 234         case CEPH_MSG_MON_SUBSCRIBE_ACK: return "mon_subscribe_ack";
 235         case CEPH_MSG_STATFS: return "statfs";
 236         case CEPH_MSG_STATFS_REPLY: return "statfs_reply";
 237         case CEPH_MSG_MDS_MAP: return "mds_map";
 238         case CEPH_MSG_CLIENT_SESSION: return "client_session";
 239         case CEPH_MSG_CLIENT_RECONNECT: return "client_reconnect";
 240         case CEPH_MSG_CLIENT_REQUEST: return "client_request";
 241         case CEPH_MSG_CLIENT_REQUEST_FORWARD: return "client_request_forward";
 242         case CEPH_MSG_CLIENT_REPLY: return "client_reply";
 243         case CEPH_MSG_CLIENT_CAPS: return "client_caps";
 244         case CEPH_MSG_CLIENT_CAPRELEASE: return "client_cap_release";
 245         case CEPH_MSG_CLIENT_SNAP: return "client_snap";
 246         case CEPH_MSG_CLIENT_LEASE: return "client_lease";
 247         case CEPH_MSG_OSD_MAP: return "osd_map";
 248         case CEPH_MSG_OSD_OP: return "osd_op";
 249         case CEPH_MSG_OSD_OPREPLY: return "osd_opreply";
 250         default: return "unknown";
 251         }
 252 }
 253
 254
 255 /*
 256  * mount options
 257  */
 258 enum {
 259         Opt_fsidmajor,
 260         Opt_fsidminor,
 261         Opt_monport,
 262         Opt_wsize,
 263         Opt_rsize,
 264         Opt_osdtimeout,
 265         Opt_mount_timeout,
 266         Opt_caps_wanted_delay_min,
 267         Opt_caps_wanted_delay_max,
 268         Opt_readdir_max_entries,
 269         Opt_last_int,
 270         /* int args above */
 271         Opt_snapdirname,
 272         Opt_name,
 273         Opt_secret,
 274         Opt_last_string,
 275         /* string args above */
 276         Opt_ip,
 277         Opt_noshare,
 278         Opt_dirstat,
 279         Opt_nodirstat,
 280         Opt_rbytes,
 281         Opt_norbytes,
 282         Opt_nocrc,
 283         Opt_noasyncreaddir,
 284 };
 285
 286 static match_table_t arg_tokens = {
 287         {Opt_fsidmajor, "fsidmajor=%ld"},
 288         {Opt_fsidminor, "fsidminor=%ld"},
 289         {Opt_monport, "monport=%d"},
 290         {Opt_wsize, "wsize=%d"},
 291         {Opt_rsize, "rsize=%d"},
 292         {Opt_osdtimeout, "osdtimeout=%d"},
 293         {Opt_mount_timeout, "mount_timeout=%d"},
 294         {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
 295         {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
 296         {Opt_readdir_max_entries, "readdir_max_entries=%d"},
 297         /* int args above */
 298         {Opt_snapdirname, "snapdirname=%s"},
 299         {Opt_name, "name=%s"},
 300         {Opt_secret, "secret=%s"},
 301         /* string args above */
 302         {Opt_ip, "ip=%s"},
 303         {Opt_noshare, "noshare"},
 304         {Opt_dirstat, "dirstat"},
 305         {Opt_nodirstat, "nodirstat"},
 306         {Opt_rbytes, "rbytes"},
 307         {Opt_norbytes, "norbytes"},
 308         {Opt_nocrc, "nocrc"},
 309         {Opt_noasyncreaddir, "noasyncreaddir"},
 310         {-1, NULL}
 311 };
 312
 313
 314 static struct ceph_mount_args *parse_mount_args(int flags, char *options,
 315                                                 const char *dev_name,
 316                                                 const char **path)
 317 {
 318         struct ceph_mount_args *args;
 319         const char *c;
 320         int err = -ENOMEM;
 321         substring_t argstr[MAX_OPT_ARGS];
 322
 323         args = kzalloc(sizeof(*args), GFP_KERNEL);
 324         if (!args)
 325                 return ERR_PTR(-ENOMEM);
 326         args->mon_addr = kcalloc(CEPH_MAX_MON, sizeof(*args->mon_addr),
 327                                  GFP_KERNEL);
 328         if (!args->mon_addr)
 329                 goto out;
 330
 331         dout("parse_mount_args %p, dev_name '%s'\n", args, dev_name);
 332
 333         /* start with defaults */
 334         args->sb_flags = flags;
 335         args->flags = CEPH_OPT_DEFAULT;
 336         args->osd_timeout = 5;    /* seconds */
 337         args->mount_timeout = CEPH_MOUNT_TIMEOUT_DEFAULT; /* seconds */
 338         args->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
 339         args->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
 340         args->rsize = CEPH_MOUNT_RSIZE_DEFAULT;
 341         args->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
 342         args->cap_release_safety = CEPH_CAPS_PER_RELEASE * 4;
 343         args->max_readdir = 1024;
 344
 345         /* ip1[:port1][,ip2[:port2]...]:/subdir/in/fs */
 346         err = -EINVAL;
 347         if (!dev_name)
 348                 goto out;
 349         *path = strstr(dev_name, ":/");
 350         if (*path == NULL) {
 351                 pr_err("device name is missing path (no :/ in %s)\n",
 352                        dev_name);
 353                 goto out;
 354         }
 355
 356         /* get mon ip(s) */
 357         err = ceph_parse_ips(dev_name, *path, args->mon_addr,
 358                              CEPH_MAX_MON, &args->num_mon);
 359         if (err < 0)
 360                 goto out;
 361
 362         /* path on server */
 363         *path += 2;
 364         dout("server path '%s'\n", *path);
 365
 366         /* parse mount options */
 367         while ((c = strsep(&options, ",")) != NULL) {
 368                 int token, intval, ret;
 369                 if (!*c)
 370                         continue;
 371                 err = -EINVAL;
 372                 token = match_token((char *)c, arg_tokens, argstr);
 373                 if (token < 0) {
 374                         pr_err("bad mount option at '%s'\n", c);
 375                         goto out;
 376                 }
 377                 if (token < Opt_last_int) {
 378                         ret = match_int(&argstr[0], &intval);
 379                         if (ret < 0) {
 380                                 pr_err("bad mount option arg (not int) "
 381                                        "at '%s'\n", c);
 382                                 continue;
 383                         }
 384                         dout("got int token %d val %d\n", token, intval);
 385                 } else if (token > Opt_last_int && token < Opt_last_string) {
 386                         dout("got string token %d val %s\n", token,
 387                              argstr[0].from);
 388                 } else {
 389                         dout("got token %d\n", token);
 390                 }
 391                 switch (token) {
 392                 case Opt_fsidmajor:
 393                         *(__le64 *)&args->fsid.fsid[0] = cpu_to_le64(intval);
 394                         break;
 395                 case Opt_fsidminor:
 396                         *(__le64 *)&args->fsid.fsid[8] = cpu_to_le64(intval);
 397                         break;
 398                 case Opt_ip:
 399                         err = ceph_parse_ips(argstr[0].from,
 400                                              argstr[0].to,
 401                                              &args->my_addr,
 402                                              1, NULL);
 403                         if (err < 0)
 404                                 goto out;
 405                         args->flags |= CEPH_OPT_MYIP;
 406                         break;
 407
 408                 case Opt_snapdirname:
 409                         kfree(args->snapdir_name);
 410                         args->snapdir_name = kstrndup(argstr[0].from,
 411                                               argstr[0].to-argstr[0].from,
 412                                               GFP_KERNEL);
 413                         break;
 414                 case Opt_name:
 415                         args->name = kstrndup(argstr[0].from,
 416                                               argstr[0].to-argstr[0].from,
 417                                               GFP_KERNEL);
 418                         break;
 419                 case Opt_secret:
 420                         args->secret = kstrndup(argstr[0].from,
 421                                                 argstr[0].to-argstr[0].from,
 422                                                 GFP_KERNEL);
 423                         break;
 424
 425                         /* misc */
 426                 case Opt_wsize:
 427                         args->wsize = intval;
 428                         break;
 429                 case Opt_rsize:
 430                         args->rsize = intval;
 431                         break;
 432                 case Opt_osdtimeout:
 433                         args->osd_timeout = intval;
 434                         break;
 435                 case Opt_mount_timeout:
 436                         args->mount_timeout = intval;
 437                         break;
 438                 case Opt_caps_wanted_delay_min:
 439                         args->caps_wanted_delay_min = intval;
 440                         break;
 441                 case Opt_caps_wanted_delay_max:
 442                         args->caps_wanted_delay_max = intval;
 443                         break;
 444                 case Opt_readdir_max_entries:
 445                         args->max_readdir = intval;
 446                         break;
 447
 448                 case Opt_noshare:
 449                         args->flags |= CEPH_OPT_NOSHARE;
 450                         break;
 451
 452                 case Opt_dirstat:
 453                         args->flags |= CEPH_OPT_DIRSTAT;
 454                         break;
 455                 case Opt_nodirstat:
 456                         args->flags &= ~CEPH_OPT_DIRSTAT;
 457                         break;
 458                 case Opt_rbytes:
 459                         args->flags |= CEPH_OPT_RBYTES;
 460                         break;
 461                 case Opt_norbytes:
 462                         args->flags &= ~CEPH_OPT_RBYTES;
 463                         break;
 464                 case Opt_nocrc:
 465                         args->flags |= CEPH_OPT_NOCRC;
 466                         break;
 467                 case Opt_noasyncreaddir:
 468                         args->flags |= CEPH_OPT_NOASYNCREADDIR;
 469                         break;
 470
 471                 default:
 472                         BUG_ON(token);
 473                 }
 474         }
 475         return args;
 476
 477 out:
 478         kfree(args->mon_addr);
 479         kfree(args);
 480         return ERR_PTR(err);
 481 }
 482
 483 static void destroy_mount_args(struct ceph_mount_args *args)
 484 {
 485         dout("destroy_mount_args %p\n", args);
 486         kfree(args->snapdir_name);
 487         args->snapdir_name = NULL;
 488         kfree(args->name);
 489         args->name = NULL;
 490         kfree(args->secret);
 491         args->secret = NULL;
 492         kfree(args);
 493 }
 494
 495 /*
 496  * create a fresh client instance
 497  */
 498 static struct ceph_client *ceph_create_client(struct ceph_mount_args *args)
 499 {
 500         struct ceph_client *client;
 501         int err = -ENOMEM;
 502
 503         client = kzalloc(sizeof(*client), GFP_KERNEL);
 504         if (client == NULL)
 505                 return ERR_PTR(-ENOMEM);
 506
 507         mutex_init(&client->mount_mutex);
 508
 509         init_waitqueue_head(&client->mount_wq);
 510
 511         client->sb = NULL;
 512         client->mount_state = CEPH_MOUNT_MOUNTING;
 513         client->whoami = -1;
 514         client->mount_args = args;
 515
 516         client->msgr = NULL;
 517
 518         client->mount_err = 0;
 519         client->signed_ticket = NULL;
 520         client->signed_ticket_len = 0;
 521
 522         err = bdi_init(&client->backing_dev_info);
 523         if (err < 0)
 524                 goto fail;
 525
 526         err = -ENOMEM;
 527         client->wb_wq = create_workqueue("ceph-writeback");
 528         if (client->wb_wq == NULL)
 529                 goto fail_bdi;
 530         client->pg_inv_wq = create_singlethread_workqueue("ceph-pg-invalid");
 531         if (client->pg_inv_wq == NULL)
 532                 goto fail_wb_wq;
 533         client->trunc_wq = create_singlethread_workqueue("ceph-trunc");
 534         if (client->trunc_wq == NULL)
 535                 goto fail_pg_inv_wq;
 536
 537         /* set up mempools */
 538         err = -ENOMEM;
 539         client->wb_pagevec_pool = mempool_create_kmalloc_pool(10,
 540                               client->mount_args->wsize >> PAGE_CACHE_SHIFT);
 541         if (!client->wb_pagevec_pool)
 542                 goto fail_trunc_wq;
 543
 544
 545         /* subsystems */
 546         err = ceph_monc_init(&client->monc, client);
 547         if (err < 0)
 548                 goto fail_mempool;
 549         err = ceph_osdc_init(&client->osdc, client);
 550         if (err < 0)
 551                 goto fail_monc;
 552         err = ceph_mdsc_init(&client->mdsc, client);
 553         if (err < 0)
 554                 goto fail_osdc;
 555         return client;
 556
 557 fail_osdc:
 558         ceph_osdc_stop(&client->osdc);
 559 fail_monc:
 560         ceph_monc_stop(&client->monc);
 561 fail_mempool:
 562         mempool_destroy(client->wb_pagevec_pool);
 563 fail_trunc_wq:
 564         destroy_workqueue(client->trunc_wq);
 565 fail_pg_inv_wq:
 566         destroy_workqueue(client->pg_inv_wq);
 567 fail_wb_wq:
 568         destroy_workqueue(client->wb_wq);
 569 fail_bdi:
 570         bdi_destroy(&client->backing_dev_info);
 571 fail:
 572         kfree(client);
 573         return ERR_PTR(err);
 574 }
 575
 576 static void ceph_destroy_client(struct ceph_client *client)
 577 {
 578         dout("destroy_client %p\n", client);
 579
 580         /* unmount */
 581         ceph_mdsc_stop(&client->mdsc);
 582         ceph_monc_stop(&client->monc);
 583         ceph_osdc_stop(&client->osdc);
 584
 585         kfree(client->signed_ticket);
 586
 587         ceph_debugfs_client_cleanup(client);
 588         destroy_workqueue(client->wb_wq);
 589         destroy_workqueue(client->pg_inv_wq);
 590         destroy_workqueue(client->trunc_wq);
 591
 592         if (client->msgr)
 593                 ceph_messenger_destroy(client->msgr);
 594         mempool_destroy(client->wb_pagevec_pool);
 595
 596         destroy_mount_args(client->mount_args);
 597
 598         kfree(client);
 599         dout("destroy_client %p done\n", client);
 600 }
 601
 602 /*
 603  * true if we have the mon map (and have thus joined the cluster)
 604  */
 605 static int have_mon_map(struct ceph_client *client)
 606 {
 607         return client->monc.monmap && client->monc.monmap->epoch;
 608 }
 609
 610 /*
 611  * Bootstrap mount by opening the root directory.  Note the mount
 612  * @started time from caller, and time out if this takes too long.
 613  */
 614 static struct dentry *open_root_dentry(struct ceph_client *client,
 615                                        const char *path,
 616                                        unsigned long started)
 617 {
 618         struct ceph_mds_client *mdsc = &client->mdsc;
 619         struct ceph_mds_request *req = NULL;
 620         int err;
 621         struct dentry *root;
 622
 623         /* open dir */
 624         dout("open_root_inode opening '%s'\n", path);
 625         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
 626         if (IS_ERR(req))
 627                 return ERR_PTR(PTR_ERR(req));
 628         req->r_path1 = kstrdup(path, GFP_NOFS);
 629         req->r_ino1.ino = CEPH_INO_ROOT;
 630         req->r_ino1.snap = CEPH_NOSNAP;
 631         req->r_started = started;
 632         req->r_timeout = client->mount_args->mount_timeout * HZ;
 633         req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
 634         req->r_num_caps = 2;
 635         err = ceph_mdsc_do_request(mdsc, NULL, req);
 636         if (err == 0) {
 637                 dout("open_root_inode success\n");
 638                 if (ceph_ino(req->r_target_inode) == CEPH_INO_ROOT &&
 639                     client->sb->s_root == NULL)
 640                         root = d_alloc_root(req->r_target_inode);
 641                 else
 642                         root = d_obtain_alias(req->r_target_inode);
 643                 req->r_target_inode = NULL;
 644                 dout("open_root_inode success, root dentry is %p\n", root);
 645         } else {
 646                 root = ERR_PTR(err);
 647         }
 648         ceph_mdsc_put_request(req);
 649         return root;
 650 }
 651
 652 /*
 653  * mount: join the ceph cluster, and open root directory.
 654  */
 655 static int ceph_mount(struct ceph_client *client, struct vfsmount *mnt,
 656                       const char *path)
 657 {
 658         struct ceph_entity_addr *myaddr = NULL;
 659         int err;
 660         unsigned long timeout = client->mount_args->mount_timeout * HZ;
 661         unsigned long started = jiffies;  /* note the start time */
 662         struct dentry *root;
 663
 664         dout("mount start\n");
 665         mutex_lock(&client->mount_mutex);
 666
 667         /* initialize the messenger */
 668         if (client->msgr == NULL) {
 669                 if (ceph_test_opt(client, MYIP))
 670                         myaddr = &client->mount_args->my_addr;
 671                 client->msgr = ceph_messenger_create(myaddr);
 672                 if (IS_ERR(client->msgr)) {
 673                         err = PTR_ERR(client->msgr);
 674                         client->msgr = NULL;
 675                         goto out;
 676                 }
 677                 client->msgr->nocrc = ceph_test_opt(client, NOCRC);
 678         }
 679
 680         /* open session, and wait for mon, mds, and osd maps */
 681         err = ceph_monc_open_session(&client->monc);
 682         if (err < 0)
 683                 goto out;
 684
 685         while (!have_mon_map(client)) {
 686                 err = -EIO;
 687                 if (timeout && time_after_eq(jiffies, started + timeout))
 688                         goto out;
 689
 690                 /* wait */
 691                 dout("mount waiting for mon_map\n");
 692                 err = wait_event_interruptible_timeout(client->mount_wq, /* FIXME */
 693                                have_mon_map(client),
 694                                timeout);
 695                 if (err == -EINTR || err == -ERESTARTSYS)
 696                         goto out;
 697         }
 698
 699         dout("mount opening root\n");
 700         root = open_root_dentry(client, "", started);
 701         if (IS_ERR(root)) {
 702                 err = PTR_ERR(root);
 703                 goto out;
 704         }
 705         if (client->sb->s_root)
 706                 dput(root);
 707         else
 708                 client->sb->s_root = root;
 709
 710         if (path[0] == 0) {
 711                 dget(root);
 712         } else {
 713                 dout("mount opening base mountpoint\n");
 714                 root = open_root_dentry(client, path, started);
 715                 if (IS_ERR(root)) {
 716                         err = PTR_ERR(root);
 717                         dput(client->sb->s_root);
 718                         client->sb->s_root = NULL;
 719                         goto out;
 720                 }
 721         }
 722
 723         mnt->mnt_root = root;
 724         mnt->mnt_sb = client->sb;
 725
 726         client->mount_state = CEPH_MOUNT_MOUNTED;
 727         dout("mount success\n");
 728         err = 0;
 729
 730 out:
 731         mutex_unlock(&client->mount_mutex);
 732         return err;
 733 }
 734
 735 static int ceph_set_super(struct super_block *s, void *data)
 736 {
 737         struct ceph_client *client = data;
 738         int ret;
 739
 740         dout("set_super %p data %p\n", s, data);
 741
 742         s->s_flags = client->mount_args->sb_flags;
 743         s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
 744
 745         s->s_fs_info = client;
 746         client->sb = s;
 747
 748         s->s_op = &ceph_super_ops;
 749         s->s_export_op = &ceph_export_ops;
 750
 751         s->s_time_gran = 1000;  /* 1000 ns == 1 us */
 752
 753         ret = set_anon_super(s, NULL);  /* what is that second arg for? */
 754         if (ret != 0)
 755                 goto fail;
 756
 757         return ret;
 758
 759 fail:
 760         s->s_fs_info = NULL;
 761         client->sb = NULL;
 762         return ret;
 763 }
 764
 765 /*
 766  * share superblock if same fs AND options
 767  */
 768 static int ceph_compare_super(struct super_block *sb, void *data)
 769 {
 770         struct ceph_client *new = data;
 771         struct ceph_mount_args *args = new->mount_args;
 772         struct ceph_client *other = ceph_sb_to_client(sb);
 773         int i;
 774
 775         dout("ceph_compare_super %p\n", sb);
 776         if (args->flags & CEPH_OPT_FSID) {
 777                 if (ceph_fsid_compare(&args->fsid, &other->fsid)) {
 778                         dout("fsid doesn't match\n");
 779                         return 0;
 780                 }
 781         } else {
 782                 /* do we share (a) monitor? */
 783                 for (i = 0; i < new->monc.monmap->num_mon; i++)
 784                         if (ceph_monmap_contains(other->monc.monmap,
 785                                          &new->monc.monmap->mon_inst[i].addr))
 786                                 break;
 787                 if (i == new->monc.monmap->num_mon) {
 788                         dout("mon ip not part of monmap\n");
 789                         return 0;
 790                 }
 791                 dout("mon ip matches existing sb %p\n", sb);
 792         }
 793         if (args->sb_flags != other->mount_args->sb_flags) {
 794                 dout("flags differ\n");
 795                 return 0;
 796         }
 797         return 1;
 798 }
 799
 800 /*
 801  * construct our own bdi so we can control readahead, etc.
 802  */
 803 static int ceph_register_bdi(struct super_block *sb, struct ceph_client *client)
 804 {
 805         int err;
 806
 807         sb->s_bdi = &client->backing_dev_info;
 808
 809         /* set ra_pages based on rsize mount option? */
 810         if (client->mount_args->rsize >= PAGE_CACHE_SIZE)
 811                 client->backing_dev_info.ra_pages =
 812                         (client->mount_args->rsize + PAGE_CACHE_SIZE - 1)
 813                         >> PAGE_SHIFT;
 814         err = bdi_register_dev(&client->backing_dev_info, sb->s_dev);
 815         return err;
 816 }
 817
 818 static int ceph_get_sb(struct file_system_type *fs_type,
 819                        int flags, const char *dev_name, void *data,
 820                        struct vfsmount *mnt)
 821 {
 822         struct super_block *sb;
 823         struct ceph_client *client;
 824         int err;
 825         int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
 826         const char *path = NULL;
 827         struct ceph_mount_args *args;
 828
 829         dout("ceph_get_sb\n");
 830         args = parse_mount_args(flags, data, dev_name, &path);
 831         if (IS_ERR(args)) {
 832                 err = PTR_ERR(args);
 833                 goto out_final;
 834         }
 835
 836         /* create client (which we may/may not use) */
 837         client = ceph_create_client(args);
 838         if (IS_ERR(client)) {
 839                 err = PTR_ERR(client);
 840                 goto out_final;
 841         }
 842
 843         if (client->mount_args->flags & CEPH_OPT_NOSHARE)
 844                 compare_super = NULL;
 845         sb = sget(fs_type, compare_super, ceph_set_super, client);
 846         if (IS_ERR(sb)) {
 847                 err = PTR_ERR(sb);
 848                 goto out;
 849         }
 850
 851         if (ceph_client(sb) != client) {
 852                 ceph_destroy_client(client);
 853                 client = ceph_client(sb);
 854                 dout("get_sb got existing client %p\n", client);
 855         } else {
 856                 dout("get_sb using new client %p\n", client);
 857                 err = ceph_register_bdi(sb, client);
 858                 if (err < 0)
 859                         goto out_splat;
 860         }
 861
 862         err = ceph_mount(client, mnt, path);
 863         if (err < 0)
 864                 goto out_splat;
 865         dout("root %p inode %p ino %llx.%llx\n", mnt->mnt_root,
 866              mnt->mnt_root->d_inode, ceph_vinop(mnt->mnt_root->d_inode));
 867         return 0;
 868
 869 out_splat:
 870         ceph_mdsc_close_sessions(&client->mdsc);
 871         up_write(&sb->s_umount);
 872         deactivate_super(sb);
 873         goto out_final;
 874
 875 out:
 876         ceph_destroy_client(client);
 877 out_final:
 878         dout("ceph_get_sb fail %d\n", err);
 879         return err;
 880 }
 881
 882 static void ceph_kill_sb(struct super_block *s)
 883 {
 884         struct ceph_client *client = ceph_sb_to_client(s);
 885         dout("kill_sb %p\n", s);
 886         ceph_mdsc_pre_umount(&client->mdsc);
 887         kill_anon_super(s);    /* will call put_super after sb is r/o */
 888         bdi_unregister(&client->backing_dev_info);
 889         bdi_destroy(&client->backing_dev_info);
 890         ceph_destroy_client(client);
 891 }
 892
 893 static struct file_system_type ceph_fs_type = {
 894         .owner          = THIS_MODULE,
 895         .name           = "ceph",
 896         .get_sb         = ceph_get_sb,
 897         .kill_sb        = ceph_kill_sb,
 898         .fs_flags       = FS_RENAME_DOES_D_MOVE,
 899 };
 900
 901 #define _STRINGIFY(x) #x
 902 #define STRINGIFY(x) _STRINGIFY(x)
 903
 904 static int __init init_ceph(void)
 905 {
 906         int ret = 0;
 907
 908         ret = ceph_debugfs_init();
 909         if (ret < 0)
 910                 goto out;
 911
 912         ret = ceph_msgr_init();
 913         if (ret < 0)
 914                 goto out_debugfs;
 915
 916         ret = init_caches();
 917         if (ret)
 918                 goto out_msgr;
 919
 920         ceph_caps_init();
 921
 922         ret = register_filesystem(&ceph_fs_type);
 923         if (ret)
 924                 goto out_icache;
 925
 926         pr_info("loaded %d.%d.%d (mon/mds/osd proto %d/%d/%d)\n",
 927                 CEPH_VERSION_MAJOR, CEPH_VERSION_MINOR, CEPH_VERSION_PATCH,
 928                 CEPH_MONC_PROTOCOL, CEPH_MDSC_PROTOCOL, CEPH_OSDC_PROTOCOL);
 929         return 0;
 930
 931 out_icache:
 932         destroy_caches();
 933 out_msgr:
 934         ceph_msgr_exit();
 935 out_debugfs:
 936         ceph_debugfs_cleanup();
 937 out:
 938         return ret;
 939 }
 940
 941 static void __exit exit_ceph(void)
 942 {
 943         dout("exit_ceph\n");
 944         unregister_filesystem(&ceph_fs_type);
 945         ceph_caps_finalize();
 946         destroy_caches();
 947         ceph_msgr_exit();
 948         ceph_debugfs_cleanup();
 949 }
 950
 951 module_init(init_ceph);
 952 module_exit(exit_ceph);
 953
 954 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
 955 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
 956 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
 957 MODULE_DESCRIPTION("Ceph filesystem for Linux");
 958 MODULE_LICENSE("GPL");