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