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/sched.h>
  12 #include <linux/seq_file.h>
  13 #include <linux/slab.h>
  14 #include <linux/statfs.h>
  15 #include <linux/string.h>
  16
  17 #include "decode.h"
  18 #include "super.h"
  19 #include "mon_client.h"
  20 #include "auth.h"
  21
  22 /*
  23  * Ceph superblock operations
  24  *
  25  * Handle the basics of mounting, unmounting.
  26  */
  27
  28
  29 /*
  30  * find filename portion of a path (/foo/bar/baz -> baz)
  31  */
  32 const char *ceph_file_part(const char *s, int len)
  33 {
  34         const char *e = s + len;
  35
  36         while (e != s && *(e-1) != '/')
  37                 e--;
  38         return e;
  39 }
  40
  41
  42 /*
  43  * super ops
  44  */
  45 static void ceph_put_super(struct super_block *s)
  46 {
  47         struct ceph_client *client = ceph_sb_to_client(s);
  48
  49         dout("put_super\n");
  50         ceph_mdsc_close_sessions(&client->mdsc);
  51
  52         /*
  53          * ensure we release the bdi before put_anon_super releases
  54          * the device name.
  55          */
  56         if (s->s_bdi == &client->backing_dev_info) {
  57                 bdi_unregister(&client->backing_dev_info);
  58                 s->s_bdi = NULL;
  59         }
  60
  61         return;
  62 }
  63
  64 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
  65 {
  66         struct ceph_client *client = ceph_inode_to_client(dentry->d_inode);
  67         struct ceph_monmap *monmap = client->monc.monmap;
  68         struct ceph_statfs st;
  69         u64 fsid;
  70         int err;
  71
  72         dout("statfs\n");
  73         err = ceph_monc_do_statfs(&client->monc, &st);
  74         if (err < 0)
  75                 return err;
  76
  77         /* fill in kstatfs */
  78         buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
  79
  80         /*
  81          * express utilization in terms of large blocks to avoid
  82          * overflow on 32-bit machines.
  83          */
  84         buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
  85         buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
  86         buf->f_bfree = (le64_to_cpu(st.kb) - le64_to_cpu(st.kb_used)) >>
  87                 (CEPH_BLOCK_SHIFT-10);
  88         buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
  89
  90         buf->f_files = le64_to_cpu(st.num_objects);
  91         buf->f_ffree = -1;
  92         buf->f_namelen = PATH_MAX;
  93         buf->f_frsize = PAGE_CACHE_SIZE;
  94
  95         /* leave fsid little-endian, regardless of host endianness */
  96         fsid = *(u64 *)(&monmap->fsid) ^ *((u64 *)&monmap->fsid + 1);
  97         buf->f_fsid.val[0] = fsid & 0xffffffff;
  98         buf->f_fsid.val[1] = fsid >> 32;
  99
 100         return 0;
 101 }
 102
 103
 104 static int ceph_syncfs(struct super_block *sb, int wait)
 105 {
 106         dout("sync_fs %d\n", wait);
 107         ceph_osdc_sync(&ceph_sb_to_client(sb)->osdc);
 108         ceph_mdsc_sync(&ceph_sb_to_client(sb)->mdsc);
 109         dout("sync_fs %d done\n", wait);
 110         return 0;
 111 }
 112
 113
 114 /**
 115  * ceph_show_options - Show mount options in /proc/mounts
 116  * @m: seq_file to write to
 117  * @mnt: mount descriptor
 118  */
 119 static int ceph_show_options(struct seq_file *m, struct vfsmount *mnt)
 120 {
 121         struct ceph_client *client = ceph_sb_to_client(mnt->mnt_sb);
 122         struct ceph_mount_args *args = client->mount_args;
 123
 124         if (args->flags & CEPH_OPT_FSID)
 125                 seq_printf(m, ",fsidmajor=%llu,fsidminor%llu",
 126                            le64_to_cpu(*(__le64 *)&args->fsid.fsid[0]),
 127                            le64_to_cpu(*(__le64 *)&args->fsid.fsid[8]));
 128         if (args->flags & CEPH_OPT_NOSHARE)
 129                 seq_puts(m, ",noshare");
 130         if (args->flags & CEPH_OPT_DIRSTAT)
 131                 seq_puts(m, ",dirstat");
 132         if ((args->flags & CEPH_OPT_RBYTES) == 0)
 133                 seq_puts(m, ",norbytes");
 134         if (args->flags & CEPH_OPT_NOCRC)
 135                 seq_puts(m, ",nocrc");
 136         if (args->flags & CEPH_OPT_NOASYNCREADDIR)
 137                 seq_puts(m, ",noasyncreaddir");
 138         if (strcmp(args->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
 139                 seq_printf(m, ",snapdirname=%s", args->snapdir_name);
 140         if (args->name)
 141                 seq_printf(m, ",name=%s", args->name);
 142         if (args->secret)
 143                 seq_puts(m, ",secret=<hidden>");
 144         return 0;
 145 }
 146
 147 /*
 148  * caches
 149  */
 150 struct kmem_cache *ceph_inode_cachep;
 151 struct kmem_cache *ceph_cap_cachep;
 152 struct kmem_cache *ceph_dentry_cachep;
 153 struct kmem_cache *ceph_file_cachep;
 154
 155 static void ceph_inode_init_once(void *foo)
 156 {
 157         struct ceph_inode_info *ci = foo;
 158         inode_init_once(&ci->vfs_inode);
 159 }
 160
 161 static int default_congestion_kb(void)
 162 {
 163         int congestion_kb;
 164
 165         /*
 166          * Copied from NFS
 167          *
 168          * congestion size, scale with available memory.
 169          *
 170          *  64MB:    8192k
 171          * 128MB:   11585k
 172          * 256MB:   16384k
 173          * 512MB:   23170k
 174          *   1GB:   32768k
 175          *   2GB:   46340k
 176          *   4GB:   65536k
 177          *   8GB:   92681k
 178          *  16GB:  131072k
 179          *
 180          * This allows larger machines to have larger/more transfers.
 181          * Limit the default to 256M
 182          */
 183         congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
 184         if (congestion_kb > 256*1024)
 185                 congestion_kb = 256*1024;
 186
 187         return congestion_kb;
 188 }
 189
 190 static int __init init_caches(void)
 191 {
 192         ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
 193                                       sizeof(struct ceph_inode_info),
 194                                       __alignof__(struct ceph_inode_info),
 195                                       (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
 196                                       ceph_inode_init_once);
 197         if (ceph_inode_cachep == NULL)
 198                 return -ENOMEM;
 199
 200         ceph_cap_cachep = KMEM_CACHE(ceph_cap,
 201                                      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 202         if (ceph_cap_cachep == NULL)
 203                 goto bad_cap;
 204
 205         ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
 206                                         SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 207         if (ceph_dentry_cachep == NULL)
 208                 goto bad_dentry;
 209
 210         ceph_file_cachep = KMEM_CACHE(ceph_file_info,
 211                                       SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 212         if (ceph_file_cachep == NULL)
 213                 goto bad_file;
 214
 215         return 0;
 216
 217 bad_file:
 218         kmem_cache_destroy(ceph_dentry_cachep);
 219 bad_dentry:
 220         kmem_cache_destroy(ceph_cap_cachep);
 221 bad_cap:
 222         kmem_cache_destroy(ceph_inode_cachep);
 223         return -ENOMEM;
 224 }
 225
 226 static void destroy_caches(void)
 227 {
 228         kmem_cache_destroy(ceph_inode_cachep);
 229         kmem_cache_destroy(ceph_cap_cachep);
 230         kmem_cache_destroy(ceph_dentry_cachep);
 231         kmem_cache_destroy(ceph_file_cachep);
 232 }
 233
 234
 235 /*
 236  * ceph_umount_begin - initiate forced umount.  Tear down down the
 237  * mount, skipping steps that may hang while waiting for server(s).
 238  */
 239 static void ceph_umount_begin(struct super_block *sb)
 240 {
 241         struct ceph_client *client = ceph_sb_to_client(sb);
 242
 243         dout("ceph_umount_begin - starting forced umount\n");
 244         if (!client)
 245                 return;
 246         client->mount_state = CEPH_MOUNT_SHUTDOWN;
 247         return;
 248 }
 249
 250 static const struct super_operations ceph_super_ops = {
 251         .alloc_inode    = ceph_alloc_inode,
 252         .destroy_inode  = ceph_destroy_inode,
 253         .write_inode    = ceph_write_inode,
 254         .sync_fs        = ceph_syncfs,
 255         .put_super      = ceph_put_super,
 256         .show_options   = ceph_show_options,
 257         .statfs         = ceph_statfs,
 258         .umount_begin   = ceph_umount_begin,
 259 };
 260
 261
 262 const char *ceph_msg_type_name(int type)
 263 {
 264         switch (type) {
 265         case CEPH_MSG_SHUTDOWN: return "shutdown";
 266         case CEPH_MSG_PING: return "ping";
 267         case CEPH_MSG_AUTH: return "auth";
 268         case CEPH_MSG_AUTH_REPLY: return "auth_reply";
 269         case CEPH_MSG_MON_MAP: return "mon_map";
 270         case CEPH_MSG_MON_GET_MAP: return "mon_get_map";
 271         case CEPH_MSG_MON_SUBSCRIBE: return "mon_subscribe";
 272         case CEPH_MSG_MON_SUBSCRIBE_ACK: return "mon_subscribe_ack";
 273         case CEPH_MSG_STATFS: return "statfs";
 274         case CEPH_MSG_STATFS_REPLY: return "statfs_reply";
 275         case CEPH_MSG_MDS_MAP: return "mds_map";
 276         case CEPH_MSG_CLIENT_SESSION: return "client_session";
 277         case CEPH_MSG_CLIENT_RECONNECT: return "client_reconnect";
 278         case CEPH_MSG_CLIENT_REQUEST: return "client_request";
 279         case CEPH_MSG_CLIENT_REQUEST_FORWARD: return "client_request_forward";
 280         case CEPH_MSG_CLIENT_REPLY: return "client_reply";
 281         case CEPH_MSG_CLIENT_CAPS: return "client_caps";
 282         case CEPH_MSG_CLIENT_CAPRELEASE: return "client_cap_release";
 283         case CEPH_MSG_CLIENT_SNAP: return "client_snap";
 284         case CEPH_MSG_CLIENT_LEASE: return "client_lease";
 285         case CEPH_MSG_OSD_MAP: return "osd_map";
 286         case CEPH_MSG_OSD_OP: return "osd_op";
 287         case CEPH_MSG_OSD_OPREPLY: return "osd_opreply";
 288         default: return "unknown";
 289         }
 290 }
 291
 292
 293 /*
 294  * mount options
 295  */
 296 enum {
 297         Opt_fsidmajor,
 298         Opt_fsidminor,
 299         Opt_monport,
 300         Opt_wsize,
 301         Opt_rsize,
 302         Opt_osdtimeout,
 303         Opt_osdkeepalivetimeout,
 304         Opt_mount_timeout,
 305         Opt_osd_idle_ttl,
 306         Opt_caps_wanted_delay_min,
 307         Opt_caps_wanted_delay_max,
 308         Opt_readdir_max_entries,
 309         Opt_congestion_kb,
 310         Opt_last_int,
 311         /* int args above */
 312         Opt_snapdirname,
 313         Opt_name,
 314         Opt_secret,
 315         Opt_last_string,
 316         /* string args above */
 317         Opt_ip,
 318         Opt_noshare,
 319         Opt_dirstat,
 320         Opt_nodirstat,
 321         Opt_rbytes,
 322         Opt_norbytes,
 323         Opt_nocrc,
 324         Opt_noasyncreaddir,
 325 };
 326
 327 static match_table_t arg_tokens = {
 328         {Opt_fsidmajor, "fsidmajor=%ld"},
 329         {Opt_fsidminor, "fsidminor=%ld"},
 330         {Opt_monport, "monport=%d"},
 331         {Opt_wsize, "wsize=%d"},
 332         {Opt_rsize, "rsize=%d"},
 333         {Opt_osdtimeout, "osdtimeout=%d"},
 334         {Opt_osdkeepalivetimeout, "osdkeepalive=%d"},
 335         {Opt_mount_timeout, "mount_timeout=%d"},
 336         {Opt_osd_idle_ttl, "osd_idle_ttl=%d"},
 337         {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
 338         {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
 339         {Opt_readdir_max_entries, "readdir_max_entries=%d"},
 340         {Opt_congestion_kb, "write_congestion_kb=%d"},
 341         /* int args above */
 342         {Opt_snapdirname, "snapdirname=%s"},
 343         {Opt_name, "name=%s"},
 344         {Opt_secret, "secret=%s"},
 345         /* string args above */
 346         {Opt_ip, "ip=%s"},
 347         {Opt_noshare, "noshare"},
 348         {Opt_dirstat, "dirstat"},
 349         {Opt_nodirstat, "nodirstat"},
 350         {Opt_rbytes, "rbytes"},
 351         {Opt_norbytes, "norbytes"},
 352         {Opt_nocrc, "nocrc"},
 353         {Opt_noasyncreaddir, "noasyncreaddir"},
 354         {-1, NULL}
 355 };
 356
 357
 358 static struct ceph_mount_args *parse_mount_args(int flags, char *options,
 359                                                 const char *dev_name,
 360                                                 const char **path)
 361 {
 362         struct ceph_mount_args *args;
 363         const char *c;
 364         int err = -ENOMEM;
 365         substring_t argstr[MAX_OPT_ARGS];
 366
 367         args = kzalloc(sizeof(*args), GFP_KERNEL);
 368         if (!args)
 369                 return ERR_PTR(-ENOMEM);
 370         args->mon_addr = kcalloc(CEPH_MAX_MON, sizeof(*args->mon_addr),
 371                                  GFP_KERNEL);
 372         if (!args->mon_addr)
 373                 goto out;
 374
 375         dout("parse_mount_args %p, dev_name '%s'\n", args, dev_name);
 376
 377         /* start with defaults */
 378         args->sb_flags = flags;
 379         args->flags = CEPH_OPT_DEFAULT;
 380         args->osd_timeout = CEPH_OSD_TIMEOUT_DEFAULT;
 381         args->osd_keepalive_timeout = CEPH_OSD_KEEPALIVE_DEFAULT;
 382         args->mount_timeout = CEPH_MOUNT_TIMEOUT_DEFAULT; /* seconds */
 383         args->osd_idle_ttl = CEPH_OSD_IDLE_TTL_DEFAULT;   /* seconds */
 384         args->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
 385         args->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
 386         args->rsize = CEPH_MOUNT_RSIZE_DEFAULT;
 387         args->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
 388         args->cap_release_safety = CEPH_CAPS_PER_RELEASE * 4;
 389         args->max_readdir = 1024;
 390         args->congestion_kb = default_congestion_kb();
 391
 392         /* ip1[:port1][,ip2[:port2]...]:/subdir/in/fs */
 393         err = -EINVAL;
 394         if (!dev_name)
 395                 goto out;
 396         *path = strstr(dev_name, ":/");
 397         if (*path == NULL) {
 398                 pr_err("device name is missing path (no :/ in %s)\n",
 399                        dev_name);
 400                 goto out;
 401         }
 402
 403         /* get mon ip(s) */
 404         err = ceph_parse_ips(dev_name, *path, args->mon_addr,
 405                              CEPH_MAX_MON, &args->num_mon);
 406         if (err < 0)
 407                 goto out;
 408
 409         /* path on server */
 410         *path += 2;
 411         dout("server path '%s'\n", *path);
 412
 413         /* parse mount options */
 414         while ((c = strsep(&options, ",")) != NULL) {
 415                 int token, intval, ret;
 416                 if (!*c)
 417                         continue;
 418                 err = -EINVAL;
 419                 token = match_token((char *)c, arg_tokens, argstr);
 420                 if (token < 0) {
 421                         pr_err("bad mount option at '%s'\n", c);
 422                         goto out;
 423                 }
 424                 if (token < Opt_last_int) {
 425                         ret = match_int(&argstr[0], &intval);
 426                         if (ret < 0) {
 427                                 pr_err("bad mount option arg (not int) "
 428                                        "at '%s'\n", c);
 429                                 continue;
 430                         }
 431                         dout("got int token %d val %d\n", token, intval);
 432                 } else if (token > Opt_last_int && token < Opt_last_string) {
 433                         dout("got string token %d val %s\n", token,
 434                              argstr[0].from);
 435                 } else {
 436                         dout("got token %d\n", token);
 437                 }
 438                 switch (token) {
 439                 case Opt_fsidmajor:
 440                         *(__le64 *)&args->fsid.fsid[0] = cpu_to_le64(intval);
 441                         break;
 442                 case Opt_fsidminor:
 443                         *(__le64 *)&args->fsid.fsid[8] = cpu_to_le64(intval);
 444                         break;
 445                 case Opt_ip:
 446                         err = ceph_parse_ips(argstr[0].from,
 447                                              argstr[0].to,
 448                                              &args->my_addr,
 449                                              1, NULL);
 450                         if (err < 0)
 451                                 goto out;
 452                         args->flags |= CEPH_OPT_MYIP;
 453                         break;
 454
 455                 case Opt_snapdirname:
 456                         kfree(args->snapdir_name);
 457                         args->snapdir_name = kstrndup(argstr[0].from,
 458                                               argstr[0].to-argstr[0].from,
 459                                               GFP_KERNEL);
 460                         break;
 461                 case Opt_name:
 462                         args->name = kstrndup(argstr[0].from,
 463                                               argstr[0].to-argstr[0].from,
 464                                               GFP_KERNEL);
 465                         break;
 466                 case Opt_secret:
 467                         args->secret = kstrndup(argstr[0].from,
 468                                                 argstr[0].to-argstr[0].from,
 469                                                 GFP_KERNEL);
 470                         break;
 471
 472                         /* misc */
 473                 case Opt_wsize:
 474                         args->wsize = intval;
 475                         break;
 476                 case Opt_rsize:
 477                         args->rsize = intval;
 478                         break;
 479                 case Opt_osdtimeout:
 480                         args->osd_timeout = intval;
 481                         break;
 482                 case Opt_osdkeepalivetimeout:
 483                         args->osd_keepalive_timeout = intval;
 484                         break;
 485                 case Opt_mount_timeout:
 486                         args->mount_timeout = intval;
 487                         break;
 488                 case Opt_caps_wanted_delay_min:
 489                         args->caps_wanted_delay_min = intval;
 490                         break;
 491                 case Opt_caps_wanted_delay_max:
 492                         args->caps_wanted_delay_max = intval;
 493                         break;
 494                 case Opt_readdir_max_entries:
 495                         args->max_readdir = intval;
 496                         break;
 497                 case Opt_congestion_kb:
 498                         args->congestion_kb = intval;
 499                         break;
 500
 501                 case Opt_noshare:
 502                         args->flags |= CEPH_OPT_NOSHARE;
 503                         break;
 504
 505                 case Opt_dirstat:
 506                         args->flags |= CEPH_OPT_DIRSTAT;
 507                         break;
 508                 case Opt_nodirstat:
 509                         args->flags &= ~CEPH_OPT_DIRSTAT;
 510                         break;
 511                 case Opt_rbytes:
 512                         args->flags |= CEPH_OPT_RBYTES;
 513                         break;
 514                 case Opt_norbytes:
 515                         args->flags &= ~CEPH_OPT_RBYTES;
 516                         break;
 517                 case Opt_nocrc:
 518                         args->flags |= CEPH_OPT_NOCRC;
 519                         break;
 520                 case Opt_noasyncreaddir:
 521                         args->flags |= CEPH_OPT_NOASYNCREADDIR;
 522                         break;
 523
 524                 default:
 525                         BUG_ON(token);
 526                 }
 527         }
 528         return args;
 529
 530 out:
 531         kfree(args->mon_addr);
 532         kfree(args);
 533         return ERR_PTR(err);
 534 }
 535
 536 static void destroy_mount_args(struct ceph_mount_args *args)
 537 {
 538         dout("destroy_mount_args %p\n", args);
 539         kfree(args->snapdir_name);
 540         args->snapdir_name = NULL;
 541         kfree(args->name);
 542         args->name = NULL;
 543         kfree(args->secret);
 544         args->secret = NULL;
 545         kfree(args);
 546 }
 547
 548 /*
 549  * create a fresh client instance
 550  */
 551 static struct ceph_client *ceph_create_client(struct ceph_mount_args *args)
 552 {
 553         struct ceph_client *client;
 554         int err = -ENOMEM;
 555
 556         client = kzalloc(sizeof(*client), GFP_KERNEL);
 557         if (client == NULL)
 558                 return ERR_PTR(-ENOMEM);
 559
 560         mutex_init(&client->mount_mutex);
 561
 562         init_waitqueue_head(&client->auth_wq);
 563
 564         client->sb = NULL;
 565         client->mount_state = CEPH_MOUNT_MOUNTING;
 566         client->mount_args = args;
 567
 568         client->msgr = NULL;
 569
 570         client->auth_err = 0;
 571         atomic_long_set(&client->writeback_count, 0);
 572
 573         err = bdi_init(&client->backing_dev_info);
 574         if (err < 0)
 575                 goto fail;
 576
 577         err = -ENOMEM;
 578         client->wb_wq = create_workqueue("ceph-writeback");
 579         if (client->wb_wq == NULL)
 580                 goto fail_bdi;
 581         client->pg_inv_wq = create_singlethread_workqueue("ceph-pg-invalid");
 582         if (client->pg_inv_wq == NULL)
 583                 goto fail_wb_wq;
 584         client->trunc_wq = create_singlethread_workqueue("ceph-trunc");
 585         if (client->trunc_wq == NULL)
 586                 goto fail_pg_inv_wq;
 587
 588         /* set up mempools */
 589         err = -ENOMEM;
 590         client->wb_pagevec_pool = mempool_create_kmalloc_pool(10,
 591                               client->mount_args->wsize >> PAGE_CACHE_SHIFT);
 592         if (!client->wb_pagevec_pool)
 593                 goto fail_trunc_wq;
 594
 595         /* caps */
 596         client->min_caps = args->max_readdir;
 597         ceph_adjust_min_caps(client->min_caps);
 598
 599         /* subsystems */
 600         err = ceph_monc_init(&client->monc, client);
 601         if (err < 0)
 602                 goto fail_mempool;
 603         err = ceph_osdc_init(&client->osdc, client);
 604         if (err < 0)
 605                 goto fail_monc;
 606         err = ceph_mdsc_init(&client->mdsc, client);
 607         if (err < 0)
 608                 goto fail_osdc;
 609         return client;
 610
 611 fail_osdc:
 612         ceph_osdc_stop(&client->osdc);
 613 fail_monc:
 614         ceph_monc_stop(&client->monc);
 615 fail_mempool:
 616         mempool_destroy(client->wb_pagevec_pool);
 617 fail_trunc_wq:
 618         destroy_workqueue(client->trunc_wq);
 619 fail_pg_inv_wq:
 620         destroy_workqueue(client->pg_inv_wq);
 621 fail_wb_wq:
 622         destroy_workqueue(client->wb_wq);
 623 fail_bdi:
 624         bdi_destroy(&client->backing_dev_info);
 625 fail:
 626         kfree(client);
 627         return ERR_PTR(err);
 628 }
 629
 630 static void ceph_destroy_client(struct ceph_client *client)
 631 {
 632         dout("destroy_client %p\n", client);
 633
 634         /* unmount */
 635         ceph_mdsc_stop(&client->mdsc);
 636         ceph_monc_stop(&client->monc);
 637         ceph_osdc_stop(&client->osdc);
 638
 639         ceph_adjust_min_caps(-client->min_caps);
 640
 641         ceph_debugfs_client_cleanup(client);
 642         destroy_workqueue(client->wb_wq);
 643         destroy_workqueue(client->pg_inv_wq);
 644         destroy_workqueue(client->trunc_wq);
 645
 646         bdi_destroy(&client->backing_dev_info);
 647
 648         if (client->msgr)
 649                 ceph_messenger_destroy(client->msgr);
 650         mempool_destroy(client->wb_pagevec_pool);
 651
 652         destroy_mount_args(client->mount_args);
 653
 654         kfree(client);
 655         dout("destroy_client %p done\n", client);
 656 }
 657
 658 /*
 659  * Initially learn our fsid, or verify an fsid matches.
 660  */
 661 int ceph_check_fsid(struct ceph_client *client, struct ceph_fsid *fsid)
 662 {
 663         if (client->have_fsid) {
 664                 if (ceph_fsid_compare(&client->fsid, fsid)) {
 665                         pr_err("bad fsid, had " FSID_FORMAT " got " FSID_FORMAT,
 666                                PR_FSID(&client->fsid), PR_FSID(fsid));
 667                         return -1;
 668                 }
 669         } else {
 670                 pr_info("client%lld fsid " FSID_FORMAT "\n",
 671                         client->monc.auth->global_id, PR_FSID(fsid));
 672                 memcpy(&client->fsid, fsid, sizeof(*fsid));
 673                 ceph_debugfs_client_init(client);
 674                 client->have_fsid = true;
 675         }
 676         return 0;
 677 }
 678
 679 /*
 680  * true if we have the mon map (and have thus joined the cluster)
 681  */
 682 static int have_mon_and_osd_map(struct ceph_client *client)
 683 {
 684         return client->monc.monmap && client->monc.monmap->epoch &&
 685                client->osdc.osdmap && client->osdc.osdmap->epoch;
 686 }
 687
 688 /*
 689  * Bootstrap mount by opening the root directory.  Note the mount
 690  * @started time from caller, and time out if this takes too long.
 691  */
 692 static struct dentry *open_root_dentry(struct ceph_client *client,
 693                                        const char *path,
 694                                        unsigned long started)
 695 {
 696         struct ceph_mds_client *mdsc = &client->mdsc;
 697         struct ceph_mds_request *req = NULL;
 698         int err;
 699         struct dentry *root;
 700
 701         /* open dir */
 702         dout("open_root_inode opening '%s'\n", path);
 703         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
 704         if (IS_ERR(req))
 705                 return ERR_PTR(PTR_ERR(req));
 706         req->r_path1 = kstrdup(path, GFP_NOFS);
 707         req->r_ino1.ino = CEPH_INO_ROOT;
 708         req->r_ino1.snap = CEPH_NOSNAP;
 709         req->r_started = started;
 710         req->r_timeout = client->mount_args->mount_timeout * HZ;
 711         req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
 712         req->r_num_caps = 2;
 713         err = ceph_mdsc_do_request(mdsc, NULL, req);
 714         if (err == 0) {
 715                 dout("open_root_inode success\n");
 716                 if (ceph_ino(req->r_target_inode) == CEPH_INO_ROOT &&
 717                     client->sb->s_root == NULL)
 718                         root = d_alloc_root(req->r_target_inode);
 719                 else
 720                         root = d_obtain_alias(req->r_target_inode);
 721                 req->r_target_inode = NULL;
 722                 dout("open_root_inode success, root dentry is %p\n", root);
 723         } else {
 724                 root = ERR_PTR(err);
 725         }
 726         ceph_mdsc_put_request(req);
 727         return root;
 728 }
 729
 730 /*
 731  * mount: join the ceph cluster, and open root directory.
 732  */
 733 static int ceph_mount(struct ceph_client *client, struct vfsmount *mnt,
 734                       const char *path)
 735 {
 736         struct ceph_entity_addr *myaddr = NULL;
 737         int err;
 738         unsigned long timeout = client->mount_args->mount_timeout * HZ;
 739         unsigned long started = jiffies;  /* note the start time */
 740         struct dentry *root;
 741
 742         dout("mount start\n");
 743         mutex_lock(&client->mount_mutex);
 744
 745         /* initialize the messenger */
 746         if (client->msgr == NULL) {
 747                 if (ceph_test_opt(client, MYIP))
 748                         myaddr = &client->mount_args->my_addr;
 749                 client->msgr = ceph_messenger_create(myaddr);
 750                 if (IS_ERR(client->msgr)) {
 751                         err = PTR_ERR(client->msgr);
 752                         client->msgr = NULL;
 753                         goto out;
 754                 }
 755                 client->msgr->nocrc = ceph_test_opt(client, NOCRC);
 756         }
 757
 758         /* open session, and wait for mon, mds, and osd maps */
 759         err = ceph_monc_open_session(&client->monc);
 760         if (err < 0)
 761                 goto out;
 762
 763         while (!have_mon_and_osd_map(client)) {
 764                 err = -EIO;
 765                 if (timeout && time_after_eq(jiffies, started + timeout))
 766                         goto out;
 767
 768                 /* wait */
 769                 dout("mount waiting for mon_map\n");
 770                 err = wait_event_interruptible_timeout(client->auth_wq,
 771                        have_mon_and_osd_map(client) || (client->auth_err < 0),
 772                        timeout);
 773                 if (err == -EINTR || err == -ERESTARTSYS)
 774                         goto out;
 775                 if (client->auth_err < 0) {
 776                         err = client->auth_err;
 777                         goto out;
 778                 }
 779         }
 780
 781         dout("mount opening root\n");
 782         root = open_root_dentry(client, "", started);
 783         if (IS_ERR(root)) {
 784                 err = PTR_ERR(root);
 785                 goto out;
 786         }
 787         if (client->sb->s_root)
 788                 dput(root);
 789         else
 790                 client->sb->s_root = root;
 791
 792         if (path[0] == 0) {
 793                 dget(root);
 794         } else {
 795                 dout("mount opening base mountpoint\n");
 796                 root = open_root_dentry(client, path, started);
 797                 if (IS_ERR(root)) {
 798                         err = PTR_ERR(root);
 799                         dput(client->sb->s_root);
 800                         client->sb->s_root = NULL;
 801                         goto out;
 802                 }
 803         }
 804
 805         mnt->mnt_root = root;
 806         mnt->mnt_sb = client->sb;
 807
 808         client->mount_state = CEPH_MOUNT_MOUNTED;
 809         dout("mount success\n");
 810         err = 0;
 811
 812 out:
 813         mutex_unlock(&client->mount_mutex);
 814         return err;
 815 }
 816
 817 static int ceph_set_super(struct super_block *s, void *data)
 818 {
 819         struct ceph_client *client = data;
 820         int ret;
 821
 822         dout("set_super %p data %p\n", s, data);
 823
 824         s->s_flags = client->mount_args->sb_flags;
 825         s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
 826
 827         s->s_fs_info = client;
 828         client->sb = s;
 829
 830         s->s_op = &ceph_super_ops;
 831         s->s_export_op = &ceph_export_ops;
 832
 833         s->s_time_gran = 1000;  /* 1000 ns == 1 us */
 834
 835         ret = set_anon_super(s, NULL);  /* what is that second arg for? */
 836         if (ret != 0)
 837                 goto fail;
 838
 839         return ret;
 840
 841 fail:
 842         s->s_fs_info = NULL;
 843         client->sb = NULL;
 844         return ret;
 845 }
 846
 847 /*
 848  * share superblock if same fs AND options
 849  */
 850 static int ceph_compare_super(struct super_block *sb, void *data)
 851 {
 852         struct ceph_client *new = data;
 853         struct ceph_mount_args *args = new->mount_args;
 854         struct ceph_client *other = ceph_sb_to_client(sb);
 855         int i;
 856
 857         dout("ceph_compare_super %p\n", sb);
 858         if (args->flags & CEPH_OPT_FSID) {
 859                 if (ceph_fsid_compare(&args->fsid, &other->fsid)) {
 860                         dout("fsid doesn't match\n");
 861                         return 0;
 862                 }
 863         } else {
 864                 /* do we share (a) monitor? */
 865                 for (i = 0; i < new->monc.monmap->num_mon; i++)
 866                         if (ceph_monmap_contains(other->monc.monmap,
 867                                          &new->monc.monmap->mon_inst[i].addr))
 868                                 break;
 869                 if (i == new->monc.monmap->num_mon) {
 870                         dout("mon ip not part of monmap\n");
 871                         return 0;
 872                 }
 873                 dout("mon ip matches existing sb %p\n", sb);
 874         }
 875         if (args->sb_flags != other->mount_args->sb_flags) {
 876                 dout("flags differ\n");
 877                 return 0;
 878         }
 879         return 1;
 880 }
 881
 882 /*
 883  * construct our own bdi so we can control readahead, etc.
 884  */
 885 static int ceph_register_bdi(struct super_block *sb, struct ceph_client *client)
 886 {
 887         int err;
 888
 889         /* set ra_pages based on rsize mount option? */
 890         if (client->mount_args->rsize >= PAGE_CACHE_SIZE)
 891                 client->backing_dev_info.ra_pages =
 892                         (client->mount_args->rsize + PAGE_CACHE_SIZE - 1)
 893                         >> PAGE_SHIFT;
 894         err = bdi_register_dev(&client->backing_dev_info, sb->s_dev);
 895         if (!err)
 896                 sb->s_bdi = &client->backing_dev_info;
 897         return err;
 898 }
 899
 900 static int ceph_get_sb(struct file_system_type *fs_type,
 901                        int flags, const char *dev_name, void *data,
 902                        struct vfsmount *mnt)
 903 {
 904         struct super_block *sb;
 905         struct ceph_client *client;
 906         int err;
 907         int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
 908         const char *path = NULL;
 909         struct ceph_mount_args *args;
 910
 911         dout("ceph_get_sb\n");
 912         args = parse_mount_args(flags, data, dev_name, &path);
 913         if (IS_ERR(args)) {
 914                 err = PTR_ERR(args);
 915                 goto out_final;
 916         }
 917
 918         /* create client (which we may/may not use) */
 919         client = ceph_create_client(args);
 920         if (IS_ERR(client)) {
 921                 err = PTR_ERR(client);
 922                 goto out_final;
 923         }
 924
 925         if (client->mount_args->flags & CEPH_OPT_NOSHARE)
 926                 compare_super = NULL;
 927         sb = sget(fs_type, compare_super, ceph_set_super, client);
 928         if (IS_ERR(sb)) {
 929                 err = PTR_ERR(sb);
 930                 goto out;
 931         }
 932
 933         if (ceph_sb_to_client(sb) != client) {
 934                 ceph_destroy_client(client);
 935                 client = ceph_sb_to_client(sb);
 936                 dout("get_sb got existing client %p\n", client);
 937         } else {
 938                 dout("get_sb using new client %p\n", client);
 939                 err = ceph_register_bdi(sb, client);
 940                 if (err < 0)
 941                         goto out_splat;
 942         }
 943
 944         err = ceph_mount(client, mnt, path);
 945         if (err < 0)
 946                 goto out_splat;
 947         dout("root %p inode %p ino %llx.%llx\n", mnt->mnt_root,
 948              mnt->mnt_root->d_inode, ceph_vinop(mnt->mnt_root->d_inode));
 949         return 0;
 950
 951 out_splat:
 952         ceph_mdsc_close_sessions(&client->mdsc);
 953         up_write(&sb->s_umount);
 954         deactivate_super(sb);
 955         goto out_final;
 956
 957 out:
 958         ceph_destroy_client(client);
 959 out_final:
 960         dout("ceph_get_sb fail %d\n", err);
 961         return err;
 962 }
 963
 964 static void ceph_kill_sb(struct super_block *s)
 965 {
 966         struct ceph_client *client = ceph_sb_to_client(s);
 967         dout("kill_sb %p\n", s);
 968         ceph_mdsc_pre_umount(&client->mdsc);
 969         kill_anon_super(s);    /* will call put_super after sb is r/o */
 970         ceph_destroy_client(client);
 971 }
 972
 973 static struct file_system_type ceph_fs_type = {
 974         .owner          = THIS_MODULE,
 975         .name           = "ceph",
 976         .get_sb         = ceph_get_sb,
 977         .kill_sb        = ceph_kill_sb,
 978         .fs_flags       = FS_RENAME_DOES_D_MOVE,
 979 };
 980
 981 #define _STRINGIFY(x) #x
 982 #define STRINGIFY(x) _STRINGIFY(x)
 983
 984 static int __init init_ceph(void)
 985 {
 986         int ret = 0;
 987
 988         ret = ceph_debugfs_init();
 989         if (ret < 0)
 990                 goto out;
 991
 992         ret = ceph_msgr_init();
 993         if (ret < 0)
 994                 goto out_debugfs;
 995
 996         ret = init_caches();
 997         if (ret)
 998                 goto out_msgr;
 999
1000         ceph_caps_init();
1001
1002         ret = register_filesystem(&ceph_fs_type);
1003         if (ret)
1004                 goto out_icache;
1005
1006         pr_info("loaded (mon/mds/osd proto %d/%d/%d, osdmap %d/%d %d/%d)\n",
1007                 CEPH_MONC_PROTOCOL, CEPH_MDSC_PROTOCOL, CEPH_OSDC_PROTOCOL,
1008                 CEPH_OSDMAP_VERSION, CEPH_OSDMAP_VERSION_EXT,
1009                 CEPH_OSDMAP_INC_VERSION, CEPH_OSDMAP_INC_VERSION_EXT);
1010         return 0;
1011
1012 out_icache:
1013         destroy_caches();
1014 out_msgr:
1015         ceph_msgr_exit();
1016 out_debugfs:
1017         ceph_debugfs_cleanup();
1018 out:
1019         return ret;
1020 }
1021
1022 static void __exit exit_ceph(void)
1023 {
1024         dout("exit_ceph\n");
1025         unregister_filesystem(&ceph_fs_type);
1026         ceph_caps_finalize();
1027         destroy_caches();
1028         ceph_msgr_exit();
1029         ceph_debugfs_cleanup();
1030 }
1031
1032 module_init(init_ceph);
1033 module_exit(exit_ceph);
1034
1035 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1036 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1037 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1038 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1039 MODULE_LICENSE("GPL");