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         Opt_last_int,
 268         /* int args above */
 269         Opt_snapdirname,
 270         Opt_secret,
 271         Opt_last_string,
 272         /* string args above */
 273         Opt_ip,
 274         Opt_noshare,
 275         Opt_dirstat,
 276         Opt_nodirstat,
 277         Opt_rbytes,
 278         Opt_norbytes,
 279         Opt_nocrc,
 280         Opt_noasyncreaddir,
 281 };
 282
 283 static match_table_t arg_tokens = {
 284         {Opt_fsidmajor, "fsidmajor=%ld"},
 285         {Opt_fsidminor, "fsidminor=%ld"},
 286         {Opt_monport, "monport=%d"},
 287         {Opt_wsize, "wsize=%d"},
 288         {Opt_rsize, "rsize=%d"},
 289         {Opt_osdtimeout, "osdtimeout=%d"},
 290         {Opt_mount_timeout, "mount_timeout=%d"},
 291         {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
 292         {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
 293         {Opt_readdir_max_entries, "readdir_max_entries=%d"},
 294         /* int args above */
 295         {Opt_snapdirname, "snapdirname=%s"},
 296         {Opt_secret, "secret=%s"},
 297         /* string args above */
 298         {Opt_ip, "ip=%s"},
 299         {Opt_noshare, "noshare"},
 300         {Opt_dirstat, "dirstat"},
 301         {Opt_nodirstat, "nodirstat"},
 302         {Opt_rbytes, "rbytes"},
 303         {Opt_norbytes, "norbytes"},
 304         {Opt_nocrc, "nocrc"},
 305         {Opt_noasyncreaddir, "noasyncreaddir"},
 306         {-1, NULL}
 307 };
 308
 309
 310 static int parse_mount_args(struct ceph_client *client,
 311                             int flags, char *options, const char *dev_name,
 312                             const char **path)
 313 {
 314         struct ceph_mount_args *args = &client->mount_args;
 315         const char *c;
 316         int err;
 317         substring_t argstr[MAX_OPT_ARGS];
 318         int num_mon;
 319         struct ceph_entity_addr *mon_addr;
 320         int i;
 321
 322         dout("parse_mount_args dev_name '%s'\n", dev_name);
 323         memset(args, 0, sizeof(*args));
 324
 325         mon_addr = kcalloc(CEPH_MAX_MON, sizeof(*mon_addr), GFP_KERNEL);
 326         if (!mon_addr)
 327                 return -ENOMEM;
 328
 329         /* start with defaults */
 330         args->sb_flags = flags;
 331         args->flags = CEPH_OPT_DEFAULT;
 332         args->osd_timeout = 5;    /* seconds */
 333         args->mount_timeout = CEPH_MOUNT_TIMEOUT_DEFAULT; /* seconds */
 334         args->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
 335         args->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
 336         args->rsize = CEPH_MOUNT_RSIZE_DEFAULT;
 337         args->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
 338         args->cap_release_safety = CEPH_CAPS_PER_RELEASE * 4;
 339         args->max_readdir = 1024;
 340
 341         /* ip1[:port1][,ip2[:port2]...]:/subdir/in/fs */
 342         err = -EINVAL;
 343         if (!dev_name)
 344                 goto out;
 345         *path = strstr(dev_name, ":/");
 346         if (*path == NULL) {
 347                 pr_err("device name is missing path (no :/ in %s)\n",
 348                        dev_name);
 349                 goto out;
 350         }
 351
 352         /* get mon ip(s) */
 353         err = ceph_parse_ips(dev_name, *path, mon_addr,
 354                              CEPH_MAX_MON, &num_mon);
 355         if (err < 0)
 356                 goto out;
 357
 358         /* build initial monmap */
 359         err = -ENOMEM;
 360         client->monc.monmap = kzalloc(sizeof(*client->monc.monmap) +
 361                                num_mon*sizeof(client->monc.monmap->mon_inst[0]),
 362                                GFP_KERNEL);
 363         if (!client->monc.monmap)
 364                 goto out;
 365         for (i = 0; i < num_mon; i++) {
 366                 client->monc.monmap->mon_inst[i].addr = mon_addr[i];
 367                 client->monc.monmap->mon_inst[i].addr.erank = 0;
 368                 client->monc.monmap->mon_inst[i].addr.nonce = 0;
 369                 client->monc.monmap->mon_inst[i].name.type =
 370                         CEPH_ENTITY_TYPE_MON;
 371                 client->monc.monmap->mon_inst[i].name.num = cpu_to_le64(i);
 372         }
 373         client->monc.monmap->num_mon = num_mon;
 374         memset(&args->my_addr.in_addr, 0, sizeof(args->my_addr.in_addr));
 375
 376         /* path on server */
 377         *path += 2;
 378         dout("server path '%s'\n", *path);
 379
 380         /* parse mount options */
 381         while ((c = strsep(&options, ",")) != NULL) {
 382                 int token, intval, ret;
 383                 if (!*c)
 384                         continue;
 385                 err = -EINVAL;
 386                 token = match_token((char *)c, arg_tokens, argstr);
 387                 if (token < 0) {
 388                         pr_err("bad mount option at '%s'\n", c);
 389                         goto out;
 390                 }
 391                 if (token < Opt_last_int) {
 392                         ret = match_int(&argstr[0], &intval);
 393                         if (ret < 0) {
 394                                 pr_err("bad mount option arg (not int) "
 395                                        "at '%s'\n", c);
 396                                 continue;
 397                         }
 398                         dout("got int token %d val %d\n", token, intval);
 399                 } else if (token > Opt_last_int && token < Opt_last_string) {
 400                         dout("got string token %d val %s\n", token,
 401                              argstr[0].from);
 402                 } else {
 403                         dout("got token %d\n", token);
 404                 }
 405                 switch (token) {
 406                 case Opt_fsidmajor:
 407                         *(__le64 *)&args->fsid.fsid[0] = cpu_to_le64(intval);
 408                         break;
 409                 case Opt_fsidminor:
 410                         *(__le64 *)&args->fsid.fsid[8] = cpu_to_le64(intval);
 411                         break;
 412                 case Opt_ip:
 413                         err = ceph_parse_ips(argstr[0].from,
 414                                              argstr[0].to,
 415                                              &args->my_addr,
 416                                              1, NULL);
 417                         if (err < 0)
 418                                 return err;
 419                         args->flags |= CEPH_OPT_MYIP;
 420                         break;
 421
 422                 case Opt_snapdirname:
 423                         kfree(args->snapdir_name);
 424                         args->snapdir_name = kstrndup(argstr[0].from,
 425                                               argstr[0].to-argstr[0].from,
 426                                               GFP_KERNEL);
 427                         break;
 428                 case Opt_secret:
 429                         args->secret = kstrndup(argstr[0].from,
 430                                                 argstr[0].to-argstr[0].from,
 431                                                 GFP_KERNEL);
 432                         break;
 433
 434                         /* misc */
 435                 case Opt_wsize:
 436                         args->wsize = intval;
 437                         break;
 438                 case Opt_rsize:
 439                         args->rsize = intval;
 440                         break;
 441                 case Opt_osdtimeout:
 442                         args->osd_timeout = intval;
 443                         break;
 444                 case Opt_mount_timeout:
 445                         args->mount_timeout = intval;
 446                         break;
 447                 case Opt_caps_wanted_delay_min:
 448                         args->caps_wanted_delay_min = intval;
 449                         break;
 450                 case Opt_caps_wanted_delay_max:
 451                         args->caps_wanted_delay_max = intval;
 452                         break;
 453                 case Opt_readdir_max_entries:
 454                         args->max_readdir = intval;
 455                         break;
 456
 457                 case Opt_noshare:
 458                         args->flags |= CEPH_OPT_NOSHARE;
 459                         break;
 460
 461                 case Opt_dirstat:
 462                         args->flags |= CEPH_OPT_DIRSTAT;
 463                         break;
 464                 case Opt_nodirstat:
 465                         args->flags &= ~CEPH_OPT_DIRSTAT;
 466                         break;
 467                 case Opt_rbytes:
 468                         args->flags |= CEPH_OPT_RBYTES;
 469                         break;
 470                 case Opt_norbytes:
 471                         args->flags &= ~CEPH_OPT_RBYTES;
 472                         break;
 473                 case Opt_nocrc:
 474                         args->flags |= CEPH_OPT_NOCRC;
 475                         break;
 476                 case Opt_noasyncreaddir:
 477                         args->flags |= CEPH_OPT_NOASYNCREADDIR;
 478                         break;
 479
 480                 default:
 481                         BUG_ON(token);
 482                 }
 483         }
 484         err = 0;
 485
 486 out:
 487         kfree(mon_addr);
 488         return err;
 489 }
 490
 491 static void release_mount_args(struct ceph_mount_args *args)
 492 {
 493         kfree(args->snapdir_name);
 494         args->snapdir_name = NULL;
 495         kfree(args->secret);
 496         args->secret = NULL;
 497 }
 498
 499 /*
 500  * create a fresh client instance
 501  */
 502 static struct ceph_client *ceph_create_client(void)
 503 {
 504         struct ceph_client *client;
 505         int err = -ENOMEM;
 506
 507         client = kzalloc(sizeof(*client), GFP_KERNEL);
 508         if (client == NULL)
 509                 return ERR_PTR(-ENOMEM);
 510
 511         mutex_init(&client->mount_mutex);
 512
 513         init_waitqueue_head(&client->mount_wq);
 514
 515         client->sb = NULL;
 516         client->mount_state = CEPH_MOUNT_MOUNTING;
 517         client->whoami = -1;
 518
 519         client->msgr = NULL;
 520
 521         client->mount_err = 0;
 522         client->signed_ticket = NULL;
 523         client->signed_ticket_len = 0;
 524
 525         err = -ENOMEM;
 526         client->wb_wq = create_workqueue("ceph-writeback");
 527         if (client->wb_wq == NULL)
 528                 goto fail;
 529         client->pg_inv_wq = create_singlethread_workqueue("ceph-pg-invalid");
 530         if (client->pg_inv_wq == NULL)
 531                 goto fail_wb_wq;
 532         client->trunc_wq = create_singlethread_workqueue("ceph-trunc");
 533         if (client->trunc_wq == NULL)
 534                 goto fail_pg_inv_wq;
 535
 536         /* subsystems */
 537         err = ceph_monc_init(&client->monc, client);
 538         if (err < 0)
 539                 goto fail_trunc_wq;
 540         err = ceph_osdc_init(&client->osdc, client);
 541         if (err < 0)
 542                 goto fail_monc;
 543         ceph_mdsc_init(&client->mdsc, client);
 544         return client;
 545
 546 fail_monc:
 547         ceph_monc_stop(&client->monc);
 548 fail_trunc_wq:
 549         destroy_workqueue(client->trunc_wq);
 550 fail_pg_inv_wq:
 551         destroy_workqueue(client->pg_inv_wq);
 552 fail_wb_wq:
 553         destroy_workqueue(client->wb_wq);
 554 fail:
 555         kfree(client);
 556         return ERR_PTR(err);
 557 }
 558
 559 static void ceph_destroy_client(struct ceph_client *client)
 560 {
 561         dout("destroy_client %p\n", client);
 562
 563         /* unmount */
 564         ceph_mdsc_stop(&client->mdsc);
 565         ceph_monc_stop(&client->monc);
 566         ceph_osdc_stop(&client->osdc);
 567
 568         kfree(client->signed_ticket);
 569
 570         ceph_debugfs_client_cleanup(client);
 571         destroy_workqueue(client->wb_wq);
 572         destroy_workqueue(client->pg_inv_wq);
 573         destroy_workqueue(client->trunc_wq);
 574
 575         if (client->msgr)
 576                 ceph_messenger_destroy(client->msgr);
 577         if (client->wb_pagevec_pool)
 578                 mempool_destroy(client->wb_pagevec_pool);
 579
 580         release_mount_args(&client->mount_args);
 581
 582         kfree(client);
 583         dout("destroy_client %p done\n", client);
 584 }
 585
 586 /*
 587  * true if we have the mon map (and have thus joined the cluster)
 588  */
 589 static int have_mon_map(struct ceph_client *client)
 590 {
 591         return client->monc.monmap && client->monc.monmap->epoch;
 592 }
 593
 594 /*
 595  * Bootstrap mount by opening the root directory.  Note the mount
 596  * @started time from caller, and time out if this takes too long.
 597  */
 598 static struct dentry *open_root_dentry(struct ceph_client *client,
 599                                        const char *path,
 600                                        unsigned long started)
 601 {
 602         struct ceph_mds_client *mdsc = &client->mdsc;
 603         struct ceph_mds_request *req = NULL;
 604         int err;
 605         struct dentry *root;
 606
 607         /* open dir */
 608         dout("open_root_inode opening '%s'\n", path);
 609         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
 610         if (IS_ERR(req))
 611                 return ERR_PTR(PTR_ERR(req));
 612         req->r_path1 = kstrdup(path, GFP_NOFS);
 613         req->r_ino1.ino = CEPH_INO_ROOT;
 614         req->r_ino1.snap = CEPH_NOSNAP;
 615         req->r_started = started;
 616         req->r_timeout = client->mount_args.mount_timeout * HZ;
 617         req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
 618         req->r_num_caps = 2;
 619         err = ceph_mdsc_do_request(mdsc, NULL, req);
 620         if (err == 0) {
 621                 dout("open_root_inode success\n");
 622                 if (ceph_ino(req->r_target_inode) == CEPH_INO_ROOT &&
 623                     client->sb->s_root == NULL)
 624                         root = d_alloc_root(req->r_target_inode);
 625                 else
 626                         root = d_obtain_alias(req->r_target_inode);
 627                 req->r_target_inode = NULL;
 628                 dout("open_root_inode success, root dentry is %p\n", root);
 629         } else {
 630                 root = ERR_PTR(err);
 631         }
 632         ceph_mdsc_put_request(req);
 633         return root;
 634 }
 635
 636 /*
 637  * mount: join the ceph cluster, and open root directory.
 638  */
 639 static int ceph_mount(struct ceph_client *client, struct vfsmount *mnt,
 640                       const char *path)
 641 {
 642         struct ceph_entity_addr *myaddr = NULL;
 643         int err;
 644         unsigned long timeout = client->mount_args.mount_timeout * HZ;
 645         unsigned long started = jiffies;  /* note the start time */
 646         struct dentry *root;
 647
 648         dout("mount start\n");
 649         mutex_lock(&client->mount_mutex);
 650
 651         /* initialize the messenger */
 652         if (client->msgr == NULL) {
 653                 if (ceph_test_opt(client, MYIP))
 654                         myaddr = &client->mount_args.my_addr;
 655                 client->msgr = ceph_messenger_create(myaddr);
 656                 if (IS_ERR(client->msgr)) {
 657                         err = PTR_ERR(client->msgr);
 658                         client->msgr = NULL;
 659                         goto out;
 660                 }
 661                 client->msgr->nocrc = ceph_test_opt(client, NOCRC);
 662         }
 663
 664         /* send mount request, and wait for mon, mds, and osd maps */
 665         err = ceph_monc_request_mount(&client->monc);
 666         if (err < 0)
 667                 goto out;
 668
 669         while (!have_mon_map(client) && !client->mount_err) {
 670                 err = -EIO;
 671                 if (timeout && time_after_eq(jiffies, started + timeout))
 672                         goto out;
 673
 674                 /* wait */
 675                 dout("mount waiting for mount\n");
 676                 err = wait_event_interruptible_timeout(client->mount_wq,
 677                                client->mount_err || have_mon_map(client),
 678                                timeout);
 679                 if (err == -EINTR || err == -ERESTARTSYS)
 680                         goto out;
 681                 if (client->mount_err) {
 682                         err = client->mount_err;
 683                         goto out;
 684                 }
 685         }
 686
 687         dout("mount opening root\n");
 688         root = open_root_dentry(client, "", started);
 689         if (IS_ERR(root)) {
 690                 err = PTR_ERR(root);
 691                 goto out;
 692         }
 693         if (client->sb->s_root)
 694                 dput(root);
 695         else
 696                 client->sb->s_root = root;
 697
 698         if (path[0] == 0) {
 699                 dget(root);
 700         } else {
 701                 dout("mount opening base mountpoint\n");
 702                 root = open_root_dentry(client, path, started);
 703                 if (IS_ERR(root)) {
 704                         err = PTR_ERR(root);
 705                         dput(client->sb->s_root);
 706                         client->sb->s_root = NULL;
 707                         goto out;
 708                 }
 709         }
 710
 711         mnt->mnt_root = root;
 712         mnt->mnt_sb = client->sb;
 713
 714         client->mount_state = CEPH_MOUNT_MOUNTED;
 715         dout("mount success\n");
 716         err = 0;
 717
 718 out:
 719         mutex_unlock(&client->mount_mutex);
 720         return err;
 721 }
 722
 723 static int ceph_set_super(struct super_block *s, void *data)
 724 {
 725         struct ceph_client *client = data;
 726         int ret;
 727
 728         dout("set_super %p data %p\n", s, data);
 729
 730         s->s_flags = client->mount_args.sb_flags;
 731         s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
 732
 733         s->s_fs_info = client;
 734         client->sb = s;
 735
 736         s->s_op = &ceph_super_ops;
 737         s->s_export_op = &ceph_export_ops;
 738
 739         s->s_time_gran = 1000;  /* 1000 ns == 1 us */
 740
 741         ret = set_anon_super(s, NULL);  /* what is that second arg for? */
 742         if (ret != 0)
 743                 goto fail;
 744
 745         return ret;
 746
 747 fail:
 748         s->s_fs_info = NULL;
 749         client->sb = NULL;
 750         return ret;
 751 }
 752
 753 /*
 754  * share superblock if same fs AND options
 755  */
 756 static int ceph_compare_super(struct super_block *sb, void *data)
 757 {
 758         struct ceph_client *new = data;
 759         struct ceph_mount_args *args = &new->mount_args;
 760         struct ceph_client *other = ceph_sb_to_client(sb);
 761         int i;
 762
 763         dout("ceph_compare_super %p\n", sb);
 764         if (args->flags & CEPH_OPT_FSID) {
 765                 if (ceph_fsid_compare(&args->fsid, &other->fsid)) {
 766                         dout("fsid doesn't match\n");
 767                         return 0;
 768                 }
 769         } else {
 770                 /* do we share (a) monitor? */
 771                 for (i = 0; i < new->monc.monmap->num_mon; i++)
 772                         if (ceph_monmap_contains(other->monc.monmap,
 773                                          &new->monc.monmap->mon_inst[i].addr))
 774                                 break;
 775                 if (i == new->monc.monmap->num_mon) {
 776                         dout("mon ip not part of monmap\n");
 777                         return 0;
 778                 }
 779                 dout("mon ip matches existing sb %p\n", sb);
 780         }
 781         if (args->sb_flags != other->mount_args.sb_flags) {
 782                 dout("flags differ\n");
 783                 return 0;
 784         }
 785         return 1;
 786 }
 787
 788 /*
 789  * construct our own bdi so we can control readahead, etc.
 790  */
 791 static int ceph_init_bdi(struct super_block *sb, struct ceph_client *client)
 792 {
 793         int err;
 794
 795         err = bdi_init(&client->backing_dev_info);
 796         if (err < 0)
 797                 return err;
 798         sb->s_bdi = &client->backing_dev_info;
 799
 800         /* set ra_pages based on rsize mount option? */
 801         if (client->mount_args.rsize >= PAGE_CACHE_SIZE)
 802                 client->backing_dev_info.ra_pages =
 803                         (client->mount_args.rsize + PAGE_CACHE_SIZE - 1)
 804                         >> PAGE_SHIFT;
 805
 806         err = bdi_register_dev(&client->backing_dev_info, sb->s_dev);
 807         return err;
 808 }
 809
 810 static int ceph_get_sb(struct file_system_type *fs_type,
 811                        int flags, const char *dev_name, void *data,
 812                        struct vfsmount *mnt)
 813 {
 814         struct super_block *sb;
 815         struct ceph_client *client;
 816         int err;
 817         int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
 818         const char *path = 0;
 819
 820         dout("ceph_get_sb\n");
 821
 822         /* create client (which we may/may not use) */
 823         client = ceph_create_client();
 824         if (IS_ERR(client))
 825                 return PTR_ERR(client);
 826
 827         err = parse_mount_args(client, flags, data, dev_name, &path);
 828         if (err < 0)
 829                 goto out;
 830
 831         if (client->mount_args.flags & CEPH_OPT_NOSHARE)
 832                 compare_super = NULL;
 833         sb = sget(fs_type, compare_super, ceph_set_super, client);
 834         if (IS_ERR(sb)) {
 835                 err = PTR_ERR(sb);
 836                 goto out;
 837         }
 838
 839         if (ceph_client(sb) != client) {
 840                 ceph_destroy_client(client);
 841                 client = ceph_client(sb);
 842                 dout("get_sb got existing client %p\n", client);
 843         } else {
 844                 dout("get_sb using new client %p\n", client);
 845
 846                 /* set up mempools */
 847                 err = -ENOMEM;
 848                 client->wb_pagevec_pool = mempool_create_kmalloc_pool(10,
 849                               client->mount_args.wsize >> PAGE_CACHE_SHIFT);
 850                 if (!client->wb_pagevec_pool)
 851                         goto out_splat;
 852
 853                 err = ceph_init_bdi(sb, client);
 854                 if (err < 0)
 855                         goto out_splat;
 856         }
 857
 858         err = ceph_mount(client, mnt, path);
 859         if (err < 0)
 860                 goto out_splat;
 861         dout("root %p inode %p ino %llx.%llx\n", mnt->mnt_root,
 862              mnt->mnt_root->d_inode, ceph_vinop(mnt->mnt_root->d_inode));
 863         return 0;
 864
 865 out_splat:
 866         ceph_mdsc_close_sessions(&client->mdsc);
 867         up_write(&sb->s_umount);
 868         deactivate_super(sb);
 869         goto out_final;
 870
 871 out:
 872         ceph_destroy_client(client);
 873 out_final:
 874         dout("ceph_get_sb fail %d\n", err);
 875         return err;
 876 }
 877
 878 static void ceph_kill_sb(struct super_block *s)
 879 {
 880         struct ceph_client *client = ceph_sb_to_client(s);
 881         dout("kill_sb %p\n", s);
 882         ceph_mdsc_pre_umount(&client->mdsc);
 883         kill_anon_super(s);    /* will call put_super after sb is r/o */
 884         bdi_unregister(&client->backing_dev_info);
 885         bdi_destroy(&client->backing_dev_info);
 886         ceph_destroy_client(client);
 887 }
 888
 889 static struct file_system_type ceph_fs_type = {
 890         .owner          = THIS_MODULE,
 891         .name           = "ceph",
 892         .get_sb         = ceph_get_sb,
 893         .kill_sb        = ceph_kill_sb,
 894         .fs_flags       = FS_RENAME_DOES_D_MOVE,
 895 };
 896
 897 #define _STRINGIFY(x) #x
 898 #define STRINGIFY(x) _STRINGIFY(x)
 899
 900 static int __init init_ceph(void)
 901 {
 902         int ret = 0;
 903
 904         ret = ceph_debugfs_init();
 905         if (ret < 0)
 906                 goto out;
 907
 908         ret = ceph_msgr_init();
 909         if (ret < 0)
 910                 goto out_debugfs;
 911
 912         ret = init_caches();
 913         if (ret)
 914                 goto out_msgr;
 915
 916         ceph_caps_init();
 917
 918         ret = register_filesystem(&ceph_fs_type);
 919         if (ret)
 920                 goto out_icache;
 921
 922         pr_info("loaded %d.%d.%d (mon/mds/osd proto %d/%d/%d)\n",
 923                 CEPH_VERSION_MAJOR, CEPH_VERSION_MINOR, CEPH_VERSION_PATCH,
 924                 CEPH_MONC_PROTOCOL, CEPH_MDSC_PROTOCOL, CEPH_OSDC_PROTOCOL);
 925         return 0;
 926
 927 out_icache:
 928         destroy_caches();
 929 out_msgr:
 930         ceph_msgr_exit();
 931 out_debugfs:
 932         ceph_debugfs_cleanup();
 933 out:
 934         return ret;
 935 }
 936
 937 static void __exit exit_ceph(void)
 938 {
 939         dout("exit_ceph\n");
 940         unregister_filesystem(&ceph_fs_type);
 941         ceph_caps_finalize();
 942         destroy_caches();
 943         ceph_msgr_exit();
 944         ceph_debugfs_cleanup();
 945 }
 946
 947 module_init(init_ceph);
 948 module_exit(exit_ceph);
 949
 950 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
 951 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
 952 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
 953 MODULE_DESCRIPTION("Ceph filesystem for Linux");
 954 MODULE_LICENSE("GPL");