fs/ceph/super.c

   1
   2 #include "ceph_debug.h"
   3
   4 #include <linux/backing-dev.h>
   5 #include <linux/ctype.h>
   6 #include <linux/fs.h>
   7 #include <linux/inet.h>
   8 #include <linux/in6.h>
   9 #include <linux/module.h>
  10 #include <linux/mount.h>
  11 #include <linux/parser.h>
  12 #include <linux/sched.h>
  13 #include <linux/seq_file.h>
  14 #include <linux/slab.h>
  15 #include <linux/statfs.h>
  16 #include <linux/string.h>
  17
  18 #include "decode.h"
  19 #include "super.h"
  20 #include "mon_client.h"
  21 #include "auth.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 *client = ceph_sb_to_client(s);
  49
  50         dout("put_super\n");
  51         ceph_mdsc_close_sessions(&client->mdsc);
  52
  53         /*
  54          * ensure we release the bdi before put_anon_super releases
  55          * the device name.
  56          */
  57         if (s->s_bdi == &client->backing_dev_info) {
  58                 bdi_unregister(&client->backing_dev_info);
  59                 s->s_bdi = NULL;
  60         }
  61
  62         return;
  63 }
  64
  65 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
  66 {
  67         struct ceph_client *client = ceph_inode_to_client(dentry->d_inode);
  68         struct ceph_monmap *monmap = client->monc.monmap;
  69         struct ceph_statfs st;
  70         u64 fsid;
  71         int err;
  72
  73         dout("statfs\n");
  74         err = ceph_monc_do_statfs(&client->monc, &st);
  75         if (err < 0)
  76                 return err;
  77
  78         /* fill in kstatfs */
  79         buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
  80
  81         /*
  82          * express utilization in terms of large blocks to avoid
  83          * overflow on 32-bit machines.
  84          */
  85         buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
  86         buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
  87         buf->f_bfree = (le64_to_cpu(st.kb) - le64_to_cpu(st.kb_used)) >>
  88                 (CEPH_BLOCK_SHIFT-10);
  89         buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
  90
  91         buf->f_files = le64_to_cpu(st.num_objects);
  92         buf->f_ffree = -1;
  93         buf->f_namelen = NAME_MAX;
  94         buf->f_frsize = PAGE_CACHE_SIZE;
  95
  96         /* leave fsid little-endian, regardless of host endianness */
  97         fsid = *(u64 *)(&monmap->fsid) ^ *((u64 *)&monmap->fsid + 1);
  98         buf->f_fsid.val[0] = fsid & 0xffffffff;
  99         buf->f_fsid.val[1] = fsid >> 32;
 100
 101         return 0;
 102 }
 103
 104
 105 static int ceph_sync_fs(struct super_block *sb, int wait)
 106 {
 107         struct ceph_client *client = ceph_sb_to_client(sb);
 108
 109         if (!wait) {
 110                 dout("sync_fs (non-blocking)\n");
 111                 ceph_flush_dirty_caps(&client->mdsc);
 112                 dout("sync_fs (non-blocking) done\n");
 113                 return 0;
 114         }
 115
 116         dout("sync_fs (blocking)\n");
 117         ceph_osdc_sync(&ceph_sb_to_client(sb)->osdc);
 118         ceph_mdsc_sync(&ceph_sb_to_client(sb)->mdsc);
 119         dout("sync_fs (blocking) done\n");
 120         return 0;
 121 }
 122
 123 static int default_congestion_kb(void)
 124 {
 125         int congestion_kb;
 126
 127         /*
 128          * Copied from NFS
 129          *
 130          * congestion size, scale with available memory.
 131          *
 132          *  64MB:    8192k
 133          * 128MB:   11585k
 134          * 256MB:   16384k
 135          * 512MB:   23170k
 136          *   1GB:   32768k
 137          *   2GB:   46340k
 138          *   4GB:   65536k
 139          *   8GB:   92681k
 140          *  16GB:  131072k
 141          *
 142          * This allows larger machines to have larger/more transfers.
 143          * Limit the default to 256M
 144          */
 145         congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
 146         if (congestion_kb > 256*1024)
 147                 congestion_kb = 256*1024;
 148
 149         return congestion_kb;
 150 }
 151
 152 /**
 153  * ceph_show_options - Show mount options in /proc/mounts
 154  * @m: seq_file to write to
 155  * @mnt: mount descriptor
 156  */
 157 static int ceph_show_options(struct seq_file *m, struct vfsmount *mnt)
 158 {
 159         struct ceph_client *client = ceph_sb_to_client(mnt->mnt_sb);
 160         struct ceph_mount_args *args = client->mount_args;
 161
 162         if (args->flags & CEPH_OPT_FSID)
 163                 seq_printf(m, ",fsid=%pU", &args->fsid);
 164         if (args->flags & CEPH_OPT_NOSHARE)
 165                 seq_puts(m, ",noshare");
 166         if (args->flags & CEPH_OPT_DIRSTAT)
 167                 seq_puts(m, ",dirstat");
 168         if ((args->flags & CEPH_OPT_RBYTES) == 0)
 169                 seq_puts(m, ",norbytes");
 170         if (args->flags & CEPH_OPT_NOCRC)
 171                 seq_puts(m, ",nocrc");
 172         if (args->flags & CEPH_OPT_NOASYNCREADDIR)
 173                 seq_puts(m, ",noasyncreaddir");
 174
 175         if (args->mount_timeout != CEPH_MOUNT_TIMEOUT_DEFAULT)
 176                 seq_printf(m, ",mount_timeout=%d", args->mount_timeout);
 177         if (args->osd_idle_ttl != CEPH_OSD_IDLE_TTL_DEFAULT)
 178                 seq_printf(m, ",osd_idle_ttl=%d", args->osd_idle_ttl);
 179         if (args->osd_timeout != CEPH_OSD_TIMEOUT_DEFAULT)
 180                 seq_printf(m, ",osdtimeout=%d", args->osd_timeout);
 181         if (args->osd_keepalive_timeout != CEPH_OSD_KEEPALIVE_DEFAULT)
 182                 seq_printf(m, ",osdkeepalivetimeout=%d",
 183                          args->osd_keepalive_timeout);
 184         if (args->wsize)
 185                 seq_printf(m, ",wsize=%d", args->wsize);
 186         if (args->rsize != CEPH_MOUNT_RSIZE_DEFAULT)
 187                 seq_printf(m, ",rsize=%d", args->rsize);
 188         if (args->congestion_kb != default_congestion_kb())
 189                 seq_printf(m, ",write_congestion_kb=%d", args->congestion_kb);
 190         if (args->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
 191                 seq_printf(m, ",caps_wanted_delay_min=%d",
 192                          args->caps_wanted_delay_min);
 193         if (args->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
 194                 seq_printf(m, ",caps_wanted_delay_max=%d",
 195                            args->caps_wanted_delay_max);
 196         if (args->cap_release_safety != CEPH_CAP_RELEASE_SAFETY_DEFAULT)
 197                 seq_printf(m, ",cap_release_safety=%d",
 198                            args->cap_release_safety);
 199         if (args->max_readdir != CEPH_MAX_READDIR_DEFAULT)
 200                 seq_printf(m, ",readdir_max_entries=%d", args->max_readdir);
 201         if (args->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
 202                 seq_printf(m, ",readdir_max_bytes=%d", args->max_readdir_bytes);
 203         if (strcmp(args->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
 204                 seq_printf(m, ",snapdirname=%s", args->snapdir_name);
 205         if (args->name)
 206                 seq_printf(m, ",name=%s", args->name);
 207         if (args->secret)
 208                 seq_puts(m, ",secret=<hidden>");
 209         return 0;
 210 }
 211
 212 /*
 213  * caches
 214  */
 215 struct kmem_cache *ceph_inode_cachep;
 216 struct kmem_cache *ceph_cap_cachep;
 217 struct kmem_cache *ceph_dentry_cachep;
 218 struct kmem_cache *ceph_file_cachep;
 219
 220 static void ceph_inode_init_once(void *foo)
 221 {
 222         struct ceph_inode_info *ci = foo;
 223         inode_init_once(&ci->vfs_inode);
 224 }
 225
 226 static int __init init_caches(void)
 227 {
 228         ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
 229                                       sizeof(struct ceph_inode_info),
 230                                       __alignof__(struct ceph_inode_info),
 231                                       (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
 232                                       ceph_inode_init_once);
 233         if (ceph_inode_cachep == NULL)
 234                 return -ENOMEM;
 235
 236         ceph_cap_cachep = KMEM_CACHE(ceph_cap,
 237                                      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 238         if (ceph_cap_cachep == NULL)
 239                 goto bad_cap;
 240
 241         ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
 242                                         SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 243         if (ceph_dentry_cachep == NULL)
 244                 goto bad_dentry;
 245
 246         ceph_file_cachep = KMEM_CACHE(ceph_file_info,
 247                                       SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 248         if (ceph_file_cachep == NULL)
 249                 goto bad_file;
 250
 251         return 0;
 252
 253 bad_file:
 254         kmem_cache_destroy(ceph_dentry_cachep);
 255 bad_dentry:
 256         kmem_cache_destroy(ceph_cap_cachep);
 257 bad_cap:
 258         kmem_cache_destroy(ceph_inode_cachep);
 259         return -ENOMEM;
 260 }
 261
 262 static void destroy_caches(void)
 263 {
 264         kmem_cache_destroy(ceph_inode_cachep);
 265         kmem_cache_destroy(ceph_cap_cachep);
 266         kmem_cache_destroy(ceph_dentry_cachep);
 267         kmem_cache_destroy(ceph_file_cachep);
 268 }
 269
 270
 271 /*
 272  * ceph_umount_begin - initiate forced umount.  Tear down down the
 273  * mount, skipping steps that may hang while waiting for server(s).
 274  */
 275 static void ceph_umount_begin(struct super_block *sb)
 276 {
 277         struct ceph_client *client = ceph_sb_to_client(sb);
 278
 279         dout("ceph_umount_begin - starting forced umount\n");
 280         if (!client)
 281                 return;
 282         client->mount_state = CEPH_MOUNT_SHUTDOWN;
 283         return;
 284 }
 285
 286 static const struct super_operations ceph_super_ops = {
 287         .alloc_inode    = ceph_alloc_inode,
 288         .destroy_inode  = ceph_destroy_inode,
 289         .write_inode    = ceph_write_inode,
 290         .sync_fs        = ceph_sync_fs,
 291         .put_super      = ceph_put_super,
 292         .show_options   = ceph_show_options,
 293         .statfs         = ceph_statfs,
 294         .umount_begin   = ceph_umount_begin,
 295 };
 296
 297
 298 const char *ceph_msg_type_name(int type)
 299 {
 300         switch (type) {
 301         case CEPH_MSG_SHUTDOWN: return "shutdown";
 302         case CEPH_MSG_PING: return "ping";
 303         case CEPH_MSG_AUTH: return "auth";
 304         case CEPH_MSG_AUTH_REPLY: return "auth_reply";
 305         case CEPH_MSG_MON_MAP: return "mon_map";
 306         case CEPH_MSG_MON_GET_MAP: return "mon_get_map";
 307         case CEPH_MSG_MON_SUBSCRIBE: return "mon_subscribe";
 308         case CEPH_MSG_MON_SUBSCRIBE_ACK: return "mon_subscribe_ack";
 309         case CEPH_MSG_STATFS: return "statfs";
 310         case CEPH_MSG_STATFS_REPLY: return "statfs_reply";
 311         case CEPH_MSG_MDS_MAP: return "mds_map";
 312         case CEPH_MSG_CLIENT_SESSION: return "client_session";
 313         case CEPH_MSG_CLIENT_RECONNECT: return "client_reconnect";
 314         case CEPH_MSG_CLIENT_REQUEST: return "client_request";
 315         case CEPH_MSG_CLIENT_REQUEST_FORWARD: return "client_request_forward";
 316         case CEPH_MSG_CLIENT_REPLY: return "client_reply";
 317         case CEPH_MSG_CLIENT_CAPS: return "client_caps";
 318         case CEPH_MSG_CLIENT_CAPRELEASE: return "client_cap_release";
 319         case CEPH_MSG_CLIENT_SNAP: return "client_snap";
 320         case CEPH_MSG_CLIENT_LEASE: return "client_lease";
 321         case CEPH_MSG_OSD_MAP: return "osd_map";
 322         case CEPH_MSG_OSD_OP: return "osd_op";
 323         case CEPH_MSG_OSD_OPREPLY: return "osd_opreply";
 324         default: return "unknown";
 325         }
 326 }
 327
 328
 329 /*
 330  * mount options
 331  */
 332 enum {
 333         Opt_wsize,
 334         Opt_rsize,
 335         Opt_osdtimeout,
 336         Opt_osdkeepalivetimeout,
 337         Opt_mount_timeout,
 338         Opt_osd_idle_ttl,
 339         Opt_caps_wanted_delay_min,
 340         Opt_caps_wanted_delay_max,
 341         Opt_cap_release_safety,
 342         Opt_readdir_max_entries,
 343         Opt_readdir_max_bytes,
 344         Opt_congestion_kb,
 345         Opt_last_int,
 346         /* int args above */
 347         Opt_fsid,
 348         Opt_snapdirname,
 349         Opt_name,
 350         Opt_secret,
 351         Opt_last_string,
 352         /* string args above */
 353         Opt_ip,
 354         Opt_noshare,
 355         Opt_dirstat,
 356         Opt_nodirstat,
 357         Opt_rbytes,
 358         Opt_norbytes,
 359         Opt_nocrc,
 360         Opt_noasyncreaddir,
 361 };
 362
 363 static match_table_t arg_tokens = {
 364         {Opt_wsize, "wsize=%d"},
 365         {Opt_rsize, "rsize=%d"},
 366         {Opt_osdtimeout, "osdtimeout=%d"},
 367         {Opt_osdkeepalivetimeout, "osdkeepalive=%d"},
 368         {Opt_mount_timeout, "mount_timeout=%d"},
 369         {Opt_osd_idle_ttl, "osd_idle_ttl=%d"},
 370         {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
 371         {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
 372         {Opt_cap_release_safety, "cap_release_safety=%d"},
 373         {Opt_readdir_max_entries, "readdir_max_entries=%d"},
 374         {Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
 375         {Opt_congestion_kb, "write_congestion_kb=%d"},
 376         /* int args above */
 377         {Opt_fsid, "fsid=%s"},
 378         {Opt_snapdirname, "snapdirname=%s"},
 379         {Opt_name, "name=%s"},
 380         {Opt_secret, "secret=%s"},
 381         /* string args above */
 382         {Opt_ip, "ip=%s"},
 383         {Opt_noshare, "noshare"},
 384         {Opt_dirstat, "dirstat"},
 385         {Opt_nodirstat, "nodirstat"},
 386         {Opt_rbytes, "rbytes"},
 387         {Opt_norbytes, "norbytes"},
 388         {Opt_nocrc, "nocrc"},
 389         {Opt_noasyncreaddir, "noasyncreaddir"},
 390         {-1, NULL}
 391 };
 392
 393 static int parse_fsid(const char *str, struct ceph_fsid *fsid)
 394 {
 395         int i = 0;
 396         char tmp[3];
 397         int err = -EINVAL;
 398         int d;
 399
 400         dout("parse_fsid '%s'\n", str);
 401         tmp[2] = 0;
 402         while (*str && i < 16) {
 403                 if (ispunct(*str)) {
 404                         str++;
 405                         continue;
 406                 }
 407                 if (!isxdigit(str[0]) || !isxdigit(str[1]))
 408                         break;
 409                 tmp[0] = str[0];
 410                 tmp[1] = str[1];
 411                 if (sscanf(tmp, "%x", &d) < 1)
 412                         break;
 413                 fsid->fsid[i] = d & 0xff;
 414                 i++;
 415                 str += 2;
 416         }
 417
 418         if (i == 16)
 419                 err = 0;
 420         dout("parse_fsid ret %d got fsid %pU", err, fsid);
 421         return err;
 422 }
 423
 424 static struct ceph_mount_args *parse_mount_args(int flags, char *options,
 425                                                 const char *dev_name,
 426                                                 const char **path)
 427 {
 428         struct ceph_mount_args *args;
 429         const char *c;
 430         int err = -ENOMEM;
 431         substring_t argstr[MAX_OPT_ARGS];
 432
 433         args = kzalloc(sizeof(*args), GFP_KERNEL);
 434         if (!args)
 435                 return ERR_PTR(-ENOMEM);
 436         args->mon_addr = kcalloc(CEPH_MAX_MON, sizeof(*args->mon_addr),
 437                                  GFP_KERNEL);
 438         if (!args->mon_addr)
 439                 goto out;
 440
 441         dout("parse_mount_args %p, dev_name '%s'\n", args, dev_name);
 442
 443         /* start with defaults */
 444         args->sb_flags = flags;
 445         args->flags = CEPH_OPT_DEFAULT;
 446         args->osd_timeout = CEPH_OSD_TIMEOUT_DEFAULT;
 447         args->osd_keepalive_timeout = CEPH_OSD_KEEPALIVE_DEFAULT;
 448         args->mount_timeout = CEPH_MOUNT_TIMEOUT_DEFAULT; /* seconds */
 449         args->osd_idle_ttl = CEPH_OSD_IDLE_TTL_DEFAULT;   /* seconds */
 450         args->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
 451         args->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
 452         args->rsize = CEPH_MOUNT_RSIZE_DEFAULT;
 453         args->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
 454         args->cap_release_safety = CEPH_CAP_RELEASE_SAFETY_DEFAULT;
 455         args->max_readdir = CEPH_MAX_READDIR_DEFAULT;
 456         args->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
 457         args->congestion_kb = default_congestion_kb();
 458
 459         /* ip1[:port1][,ip2[:port2]...]:/subdir/in/fs */
 460         err = -EINVAL;
 461         if (!dev_name)
 462                 goto out;
 463         *path = strstr(dev_name, ":/");
 464         if (*path == NULL) {
 465                 pr_err("device name is missing path (no :/ in %s)\n",
 466                        dev_name);
 467                 goto out;
 468         }
 469
 470         /* get mon ip(s) */
 471         err = ceph_parse_ips(dev_name, *path, args->mon_addr,
 472                              CEPH_MAX_MON, &args->num_mon);
 473         if (err < 0)
 474                 goto out;
 475
 476         /* path on server */
 477         *path += 2;
 478         dout("server path '%s'\n", *path);
 479
 480         /* parse mount options */
 481         while ((c = strsep(&options, ",")) != NULL) {
 482                 int token, intval, ret;
 483                 if (!*c)
 484                         continue;
 485                 err = -EINVAL;
 486                 token = match_token((char *)c, arg_tokens, argstr);
 487                 if (token < 0) {
 488                         pr_err("bad mount option at '%s'\n", c);
 489                         goto out;
 490                 }
 491                 if (token < Opt_last_int) {
 492                         ret = match_int(&argstr[0], &intval);
 493                         if (ret < 0) {
 494                                 pr_err("bad mount option arg (not int) "
 495                                        "at '%s'\n", c);
 496                                 continue;
 497                         }
 498                         dout("got int token %d val %d\n", token, intval);
 499                 } else if (token > Opt_last_int && token < Opt_last_string) {
 500                         dout("got string token %d val %s\n", token,
 501                              argstr[0].from);
 502                 } else {
 503                         dout("got token %d\n", token);
 504                 }
 505                 switch (token) {
 506                 case Opt_ip:
 507                         err = ceph_parse_ips(argstr[0].from,
 508                                              argstr[0].to,
 509                                              &args->my_addr,
 510                                              1, NULL);
 511                         if (err < 0)
 512                                 goto out;
 513                         args->flags |= CEPH_OPT_MYIP;
 514                         break;
 515
 516                 case Opt_fsid:
 517                         err = parse_fsid(argstr[0].from, &args->fsid);
 518                         if (err == 0)
 519                                 args->flags |= CEPH_OPT_FSID;
 520                         break;
 521                 case Opt_snapdirname:
 522                         kfree(args->snapdir_name);
 523                         args->snapdir_name = kstrndup(argstr[0].from,
 524                                               argstr[0].to-argstr[0].from,
 525                                               GFP_KERNEL);
 526                         break;
 527                 case Opt_name:
 528                         args->name = kstrndup(argstr[0].from,
 529                                               argstr[0].to-argstr[0].from,
 530                                               GFP_KERNEL);
 531                         break;
 532                 case Opt_secret:
 533                         args->secret = kstrndup(argstr[0].from,
 534                                                 argstr[0].to-argstr[0].from,
 535                                                 GFP_KERNEL);
 536                         break;
 537
 538                         /* misc */
 539                 case Opt_wsize:
 540                         args->wsize = intval;
 541                         break;
 542                 case Opt_rsize:
 543                         args->rsize = intval;
 544                         break;
 545                 case Opt_osdtimeout:
 546                         args->osd_timeout = intval;
 547                         break;
 548                 case Opt_osdkeepalivetimeout:
 549                         args->osd_keepalive_timeout = intval;
 550                         break;
 551                 case Opt_osd_idle_ttl:
 552                         args->osd_idle_ttl = intval;
 553                         break;
 554                 case Opt_mount_timeout:
 555                         args->mount_timeout = intval;
 556                         break;
 557                 case Opt_caps_wanted_delay_min:
 558                         args->caps_wanted_delay_min = intval;
 559                         break;
 560                 case Opt_caps_wanted_delay_max:
 561                         args->caps_wanted_delay_max = intval;
 562                         break;
 563                 case Opt_readdir_max_entries:
 564                         args->max_readdir = intval;
 565                         break;
 566                 case Opt_readdir_max_bytes:
 567                         args->max_readdir_bytes = intval;
 568                         break;
 569                 case Opt_congestion_kb:
 570                         args->congestion_kb = intval;
 571                         break;
 572
 573                 case Opt_noshare:
 574                         args->flags |= CEPH_OPT_NOSHARE;
 575                         break;
 576
 577                 case Opt_dirstat:
 578                         args->flags |= CEPH_OPT_DIRSTAT;
 579                         break;
 580                 case Opt_nodirstat:
 581                         args->flags &= ~CEPH_OPT_DIRSTAT;
 582                         break;
 583                 case Opt_rbytes:
 584                         args->flags |= CEPH_OPT_RBYTES;
 585                         break;
 586                 case Opt_norbytes:
 587                         args->flags &= ~CEPH_OPT_RBYTES;
 588                         break;
 589                 case Opt_nocrc:
 590                         args->flags |= CEPH_OPT_NOCRC;
 591                         break;
 592                 case Opt_noasyncreaddir:
 593                         args->flags |= CEPH_OPT_NOASYNCREADDIR;
 594                         break;
 595
 596                 default:
 597                         BUG_ON(token);
 598                 }
 599         }
 600         return args;
 601
 602 out:
 603         kfree(args->mon_addr);
 604         kfree(args);
 605         return ERR_PTR(err);
 606 }
 607
 608 static void destroy_mount_args(struct ceph_mount_args *args)
 609 {
 610         dout("destroy_mount_args %p\n", args);
 611         kfree(args->snapdir_name);
 612         args->snapdir_name = NULL;
 613         kfree(args->name);
 614         args->name = NULL;
 615         kfree(args->secret);
 616         args->secret = NULL;
 617         kfree(args);
 618 }
 619
 620 /*
 621  * create a fresh client instance
 622  */
 623 static struct ceph_client *ceph_create_client(struct ceph_mount_args *args)
 624 {
 625         struct ceph_client *client;
 626         int err = -ENOMEM;
 627
 628         client = kzalloc(sizeof(*client), GFP_KERNEL);
 629         if (client == NULL)
 630                 return ERR_PTR(-ENOMEM);
 631
 632         mutex_init(&client->mount_mutex);
 633
 634         init_waitqueue_head(&client->auth_wq);
 635
 636         client->sb = NULL;
 637         client->mount_state = CEPH_MOUNT_MOUNTING;
 638         client->mount_args = args;
 639
 640         client->msgr = NULL;
 641
 642         client->auth_err = 0;
 643         atomic_long_set(&client->writeback_count, 0);
 644
 645         err = bdi_init(&client->backing_dev_info);
 646         if (err < 0)
 647                 goto fail;
 648
 649         err = -ENOMEM;
 650         client->wb_wq = create_workqueue("ceph-writeback");
 651         if (client->wb_wq == NULL)
 652                 goto fail_bdi;
 653         client->pg_inv_wq = create_singlethread_workqueue("ceph-pg-invalid");
 654         if (client->pg_inv_wq == NULL)
 655                 goto fail_wb_wq;
 656         client->trunc_wq = create_singlethread_workqueue("ceph-trunc");
 657         if (client->trunc_wq == NULL)
 658                 goto fail_pg_inv_wq;
 659
 660         /* set up mempools */
 661         err = -ENOMEM;
 662         client->wb_pagevec_pool = mempool_create_kmalloc_pool(10,
 663                               client->mount_args->wsize >> PAGE_CACHE_SHIFT);
 664         if (!client->wb_pagevec_pool)
 665                 goto fail_trunc_wq;
 666
 667         /* caps */
 668         client->min_caps = args->max_readdir;
 669
 670         /* subsystems */
 671         err = ceph_monc_init(&client->monc, client);
 672         if (err < 0)
 673                 goto fail_mempool;
 674         err = ceph_osdc_init(&client->osdc, client);
 675         if (err < 0)
 676                 goto fail_monc;
 677         err = ceph_mdsc_init(&client->mdsc, client);
 678         if (err < 0)
 679                 goto fail_osdc;
 680         return client;
 681
 682 fail_osdc:
 683         ceph_osdc_stop(&client->osdc);
 684 fail_monc:
 685         ceph_monc_stop(&client->monc);
 686 fail_mempool:
 687         mempool_destroy(client->wb_pagevec_pool);
 688 fail_trunc_wq:
 689         destroy_workqueue(client->trunc_wq);
 690 fail_pg_inv_wq:
 691         destroy_workqueue(client->pg_inv_wq);
 692 fail_wb_wq:
 693         destroy_workqueue(client->wb_wq);
 694 fail_bdi:
 695         bdi_destroy(&client->backing_dev_info);
 696 fail:
 697         kfree(client);
 698         return ERR_PTR(err);
 699 }
 700
 701 static void ceph_destroy_client(struct ceph_client *client)
 702 {
 703         dout("destroy_client %p\n", client);
 704
 705         /* unmount */
 706         ceph_mdsc_stop(&client->mdsc);
 707         ceph_osdc_stop(&client->osdc);
 708
 709         /*
 710          * make sure mds and osd connections close out before destroying
 711          * the auth module, which is needed to free those connections'
 712          * ceph_authorizers.
 713          */
 714         ceph_msgr_flush();
 715
 716         ceph_monc_stop(&client->monc);
 717
 718         ceph_debugfs_client_cleanup(client);
 719         destroy_workqueue(client->wb_wq);
 720         destroy_workqueue(client->pg_inv_wq);
 721         destroy_workqueue(client->trunc_wq);
 722
 723         bdi_destroy(&client->backing_dev_info);
 724
 725         if (client->msgr)
 726                 ceph_messenger_destroy(client->msgr);
 727         mempool_destroy(client->wb_pagevec_pool);
 728
 729         destroy_mount_args(client->mount_args);
 730
 731         kfree(client);
 732         dout("destroy_client %p done\n", client);
 733 }
 734
 735 /*
 736  * Initially learn our fsid, or verify an fsid matches.
 737  */
 738 int ceph_check_fsid(struct ceph_client *client, struct ceph_fsid *fsid)
 739 {
 740         if (client->have_fsid) {
 741                 if (ceph_fsid_compare(&client->fsid, fsid)) {
 742                         pr_err("bad fsid, had %pU got %pU",
 743                                &client->fsid, fsid);
 744                         return -1;
 745                 }
 746         } else {
 747                 pr_info("client%lld fsid %pU\n", client->monc.auth->global_id,
 748                         fsid);
 749                 memcpy(&client->fsid, fsid, sizeof(*fsid));
 750                 ceph_debugfs_client_init(client);
 751                 client->have_fsid = true;
 752         }
 753         return 0;
 754 }
 755
 756 /*
 757  * true if we have the mon map (and have thus joined the cluster)
 758  */
 759 static int have_mon_and_osd_map(struct ceph_client *client)
 760 {
 761         return client->monc.monmap && client->monc.monmap->epoch &&
 762                client->osdc.osdmap && client->osdc.osdmap->epoch;
 763 }
 764
 765 /*
 766  * Bootstrap mount by opening the root directory.  Note the mount
 767  * @started time from caller, and time out if this takes too long.
 768  */
 769 static struct dentry *open_root_dentry(struct ceph_client *client,
 770                                        const char *path,
 771                                        unsigned long started)
 772 {
 773         struct ceph_mds_client *mdsc = &client->mdsc;
 774         struct ceph_mds_request *req = NULL;
 775         int err;
 776         struct dentry *root;
 777
 778         /* open dir */
 779         dout("open_root_inode opening '%s'\n", path);
 780         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
 781         if (IS_ERR(req))
 782                 return ERR_CAST(req);
 783         req->r_path1 = kstrdup(path, GFP_NOFS);
 784         req->r_ino1.ino = CEPH_INO_ROOT;
 785         req->r_ino1.snap = CEPH_NOSNAP;
 786         req->r_started = started;
 787         req->r_timeout = client->mount_args->mount_timeout * HZ;
 788         req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
 789         req->r_num_caps = 2;
 790         err = ceph_mdsc_do_request(mdsc, NULL, req);
 791         if (err == 0) {
 792                 dout("open_root_inode success\n");
 793                 if (ceph_ino(req->r_target_inode) == CEPH_INO_ROOT &&
 794                     client->sb->s_root == NULL)
 795                         root = d_alloc_root(req->r_target_inode);
 796                 else
 797                         root = d_obtain_alias(req->r_target_inode);
 798                 req->r_target_inode = NULL;
 799                 dout("open_root_inode success, root dentry is %p\n", root);
 800         } else {
 801                 root = ERR_PTR(err);
 802         }
 803         ceph_mdsc_put_request(req);
 804         return root;
 805 }
 806
 807 /*
 808  * mount: join the ceph cluster, and open root directory.
 809  */
 810 static int ceph_mount(struct ceph_client *client, struct vfsmount *mnt,
 811                       const char *path)
 812 {
 813         struct ceph_entity_addr *myaddr = NULL;
 814         int err;
 815         unsigned long timeout = client->mount_args->mount_timeout * HZ;
 816         unsigned long started = jiffies;  /* note the start time */
 817         struct dentry *root;
 818
 819         dout("mount start\n");
 820         mutex_lock(&client->mount_mutex);
 821
 822         /* initialize the messenger */
 823         if (client->msgr == NULL) {
 824                 if (ceph_test_opt(client, MYIP))
 825                         myaddr = &client->mount_args->my_addr;
 826                 client->msgr = ceph_messenger_create(myaddr);
 827                 if (IS_ERR(client->msgr)) {
 828                         err = PTR_ERR(client->msgr);
 829                         client->msgr = NULL;
 830                         goto out;
 831                 }
 832                 client->msgr->nocrc = ceph_test_opt(client, NOCRC);
 833         }
 834
 835         /* open session, and wait for mon, mds, and osd maps */
 836         err = ceph_monc_open_session(&client->monc);
 837         if (err < 0)
 838                 goto out;
 839
 840         while (!have_mon_and_osd_map(client)) {
 841                 err = -EIO;
 842                 if (timeout && time_after_eq(jiffies, started + timeout))
 843                         goto out;
 844
 845                 /* wait */
 846                 dout("mount waiting for mon_map\n");
 847                 err = wait_event_interruptible_timeout(client->auth_wq,
 848                        have_mon_and_osd_map(client) || (client->auth_err < 0),
 849                        timeout);
 850                 if (err == -EINTR || err == -ERESTARTSYS)
 851                         goto out;
 852                 if (client->auth_err < 0) {
 853                         err = client->auth_err;
 854                         goto out;
 855                 }
 856         }
 857
 858         dout("mount opening root\n");
 859         root = open_root_dentry(client, "", started);
 860         if (IS_ERR(root)) {
 861                 err = PTR_ERR(root);
 862                 goto out;
 863         }
 864         if (client->sb->s_root)
 865                 dput(root);
 866         else
 867                 client->sb->s_root = root;
 868
 869         if (path[0] == 0) {
 870                 dget(root);
 871         } else {
 872                 dout("mount opening base mountpoint\n");
 873                 root = open_root_dentry(client, path, started);
 874                 if (IS_ERR(root)) {
 875                         err = PTR_ERR(root);
 876                         dput(client->sb->s_root);
 877                         client->sb->s_root = NULL;
 878                         goto out;
 879                 }
 880         }
 881
 882         mnt->mnt_root = root;
 883         mnt->mnt_sb = client->sb;
 884
 885         client->mount_state = CEPH_MOUNT_MOUNTED;
 886         dout("mount success\n");
 887         err = 0;
 888
 889 out:
 890         mutex_unlock(&client->mount_mutex);
 891         return err;
 892 }
 893
 894 static int ceph_set_super(struct super_block *s, void *data)
 895 {
 896         struct ceph_client *client = data;
 897         int ret;
 898
 899         dout("set_super %p data %p\n", s, data);
 900
 901         s->s_flags = client->mount_args->sb_flags;
 902         s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
 903
 904         s->s_fs_info = client;
 905         client->sb = s;
 906
 907         s->s_op = &ceph_super_ops;
 908         s->s_export_op = &ceph_export_ops;
 909
 910         s->s_time_gran = 1000;  /* 1000 ns == 1 us */
 911
 912         ret = set_anon_super(s, NULL);  /* what is that second arg for? */
 913         if (ret != 0)
 914                 goto fail;
 915
 916         return ret;
 917
 918 fail:
 919         s->s_fs_info = NULL;
 920         client->sb = NULL;
 921         return ret;
 922 }
 923
 924 /*
 925  * share superblock if same fs AND options
 926  */
 927 static int ceph_compare_super(struct super_block *sb, void *data)
 928 {
 929         struct ceph_client *new = data;
 930         struct ceph_mount_args *args = new->mount_args;
 931         struct ceph_client *other = ceph_sb_to_client(sb);
 932         int i;
 933
 934         dout("ceph_compare_super %p\n", sb);
 935         if (args->flags & CEPH_OPT_FSID) {
 936                 if (ceph_fsid_compare(&args->fsid, &other->fsid)) {
 937                         dout("fsid doesn't match\n");
 938                         return 0;
 939                 }
 940         } else {
 941                 /* do we share (a) monitor? */
 942                 for (i = 0; i < new->monc.monmap->num_mon; i++)
 943                         if (ceph_monmap_contains(other->monc.monmap,
 944                                          &new->monc.monmap->mon_inst[i].addr))
 945                                 break;
 946                 if (i == new->monc.monmap->num_mon) {
 947                         dout("mon ip not part of monmap\n");
 948                         return 0;
 949                 }
 950                 dout("mon ip matches existing sb %p\n", sb);
 951         }
 952         if (args->sb_flags != other->mount_args->sb_flags) {
 953                 dout("flags differ\n");
 954                 return 0;
 955         }
 956         return 1;
 957 }
 958
 959 /*
 960  * construct our own bdi so we can control readahead, etc.
 961  */
 962 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
 963
 964 static int ceph_register_bdi(struct super_block *sb, struct ceph_client *client)
 965 {
 966         int err;
 967
 968         /* set ra_pages based on rsize mount option? */
 969         if (client->mount_args->rsize >= PAGE_CACHE_SIZE)
 970                 client->backing_dev_info.ra_pages =
 971                         (client->mount_args->rsize + PAGE_CACHE_SIZE - 1)
 972                         >> PAGE_SHIFT;
 973         err = bdi_register(&client->backing_dev_info, NULL, "ceph-%d",
 974                            atomic_long_inc_return(&bdi_seq));
 975         if (!err)
 976                 sb->s_bdi = &client->backing_dev_info;
 977         return err;
 978 }
 979
 980 static int ceph_get_sb(struct file_system_type *fs_type,
 981                        int flags, const char *dev_name, void *data,
 982                        struct vfsmount *mnt)
 983 {
 984         struct super_block *sb;
 985         struct ceph_client *client;
 986         int err;
 987         int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
 988         const char *path = NULL;
 989         struct ceph_mount_args *args;
 990
 991         dout("ceph_get_sb\n");
 992         args = parse_mount_args(flags, data, dev_name, &path);
 993         if (IS_ERR(args)) {
 994                 err = PTR_ERR(args);
 995                 goto out_final;
 996         }
 997
 998         /* create client (which we may/may not use) */
 999         client = ceph_create_client(args);
1000         if (IS_ERR(client)) {
1001                 err = PTR_ERR(client);
1002                 goto out_final;
1003         }
1004
1005         if (client->mount_args->flags & CEPH_OPT_NOSHARE)
1006                 compare_super = NULL;
1007         sb = sget(fs_type, compare_super, ceph_set_super, client);
1008         if (IS_ERR(sb)) {
1009                 err = PTR_ERR(sb);
1010                 goto out;
1011         }
1012
1013         if (ceph_sb_to_client(sb) != client) {
1014                 ceph_destroy_client(client);
1015                 client = ceph_sb_to_client(sb);
1016                 dout("get_sb got existing client %p\n", client);
1017         } else {
1018                 dout("get_sb using new client %p\n", client);
1019                 err = ceph_register_bdi(sb, client);
1020                 if (err < 0)
1021                         goto out_splat;
1022         }
1023
1024         err = ceph_mount(client, mnt, path);
1025         if (err < 0)
1026                 goto out_splat;
1027         dout("root %p inode %p ino %llx.%llx\n", mnt->mnt_root,
1028              mnt->mnt_root->d_inode, ceph_vinop(mnt->mnt_root->d_inode));
1029         return 0;
1030
1031 out_splat:
1032         ceph_mdsc_close_sessions(&client->mdsc);
1033         deactivate_locked_super(sb);
1034         goto out_final;
1035
1036 out:
1037         ceph_destroy_client(client);
1038 out_final:
1039         dout("ceph_get_sb fail %d\n", err);
1040         return err;
1041 }
1042
1043 static void ceph_kill_sb(struct super_block *s)
1044 {
1045         struct ceph_client *client = ceph_sb_to_client(s);
1046         dout("kill_sb %p\n", s);
1047         ceph_mdsc_pre_umount(&client->mdsc);
1048         kill_anon_super(s);    /* will call put_super after sb is r/o */
1049         ceph_destroy_client(client);
1050 }
1051
1052 static struct file_system_type ceph_fs_type = {
1053         .owner          = THIS_MODULE,
1054         .name           = "ceph",
1055         .get_sb         = ceph_get_sb,
1056         .kill_sb        = ceph_kill_sb,
1057         .fs_flags       = FS_RENAME_DOES_D_MOVE,
1058 };
1059
1060 #define _STRINGIFY(x) #x
1061 #define STRINGIFY(x) _STRINGIFY(x)
1062
1063 static int __init init_ceph(void)
1064 {
1065         int ret = 0;
1066
1067         ret = ceph_debugfs_init();
1068         if (ret < 0)
1069                 goto out;
1070
1071         ret = ceph_msgr_init();
1072         if (ret < 0)
1073                 goto out_debugfs;
1074
1075         ret = init_caches();
1076         if (ret)
1077                 goto out_msgr;
1078
1079         ret = register_filesystem(&ceph_fs_type);
1080         if (ret)
1081                 goto out_icache;
1082
1083         pr_info("loaded (mon/mds/osd proto %d/%d/%d, osdmap %d/%d %d/%d)\n",
1084                 CEPH_MONC_PROTOCOL, CEPH_MDSC_PROTOCOL, CEPH_OSDC_PROTOCOL,
1085                 CEPH_OSDMAP_VERSION, CEPH_OSDMAP_VERSION_EXT,
1086                 CEPH_OSDMAP_INC_VERSION, CEPH_OSDMAP_INC_VERSION_EXT);
1087         return 0;
1088
1089 out_icache:
1090         destroy_caches();
1091 out_msgr:
1092         ceph_msgr_exit();
1093 out_debugfs:
1094         ceph_debugfs_cleanup();
1095 out:
1096         return ret;
1097 }
1098
1099 static void __exit exit_ceph(void)
1100 {
1101         dout("exit_ceph\n");
1102         unregister_filesystem(&ceph_fs_type);
1103         destroy_caches();
1104         ceph_msgr_exit();
1105         ceph_debugfs_cleanup();
1106 }
1107
1108 module_init(init_ceph);
1109 module_exit(exit_ceph);
1110
1111 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1112 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1113 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1114 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1115 MODULE_LICENSE("GPL");