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 static int default_congestion_kb(void)
 114 {
 115         int congestion_kb;
 116
 117         /*
 118          * Copied from NFS
 119          *
 120          * congestion size, scale with available memory.
 121          *
 122          *  64MB:    8192k
 123          * 128MB:   11585k
 124          * 256MB:   16384k
 125          * 512MB:   23170k
 126          *   1GB:   32768k
 127          *   2GB:   46340k
 128          *   4GB:   65536k
 129          *   8GB:   92681k
 130          *  16GB:  131072k
 131          *
 132          * This allows larger machines to have larger/more transfers.
 133          * Limit the default to 256M
 134          */
 135         congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
 136         if (congestion_kb > 256*1024)
 137                 congestion_kb = 256*1024;
 138
 139         return congestion_kb;
 140 }
 141
 142 /**
 143  * ceph_show_options - Show mount options in /proc/mounts
 144  * @m: seq_file to write to
 145  * @mnt: mount descriptor
 146  */
 147 static int ceph_show_options(struct seq_file *m, struct vfsmount *mnt)
 148 {
 149         struct ceph_client *client = ceph_sb_to_client(mnt->mnt_sb);
 150         struct ceph_mount_args *args = client->mount_args;
 151
 152         if (args->flags & CEPH_OPT_FSID)
 153                 seq_printf(m, ",fsidmajor=%llu,fsidminor%llu",
 154                            le64_to_cpu(*(__le64 *)&args->fsid.fsid[0]),
 155                            le64_to_cpu(*(__le64 *)&args->fsid.fsid[8]));
 156         if (args->flags & CEPH_OPT_NOSHARE)
 157                 seq_puts(m, ",noshare");
 158         if (args->flags & CEPH_OPT_DIRSTAT)
 159                 seq_puts(m, ",dirstat");
 160         if ((args->flags & CEPH_OPT_RBYTES) == 0)
 161                 seq_puts(m, ",norbytes");
 162         if (args->flags & CEPH_OPT_NOCRC)
 163                 seq_puts(m, ",nocrc");
 164         if (args->flags & CEPH_OPT_NOASYNCREADDIR)
 165                 seq_puts(m, ",noasyncreaddir");
 166
 167         if (args->mount_timeout != CEPH_MOUNT_TIMEOUT_DEFAULT)
 168                 seq_printf(m, ",mount_timeout=%d", args->mount_timeout);
 169         if (args->osd_idle_ttl != CEPH_OSD_IDLE_TTL_DEFAULT)
 170                 seq_printf(m, ",osd_idle_ttl=%d", args->osd_idle_ttl);
 171         if (args->osd_timeout != CEPH_OSD_TIMEOUT_DEFAULT)
 172                 seq_printf(m, ",osdtimeout=%d", args->osd_timeout);
 173         if (args->osd_keepalive_timeout != CEPH_OSD_KEEPALIVE_DEFAULT)
 174                 seq_printf(m, ",osdkeepalivetimeout=%d",
 175                          args->osd_keepalive_timeout);
 176         if (args->wsize)
 177                 seq_printf(m, ",wsize=%d", args->wsize);
 178         if (args->rsize != CEPH_MOUNT_RSIZE_DEFAULT)
 179                 seq_printf(m, ",rsize=%d", args->rsize);
 180         if (args->congestion_kb != default_congestion_kb())
 181                 seq_printf(m, ",write_congestion_kb=%d", args->congestion_kb);
 182         if (args->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
 183                 seq_printf(m, ",caps_wanted_delay_min=%d",
 184                          args->caps_wanted_delay_min);
 185         if (args->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
 186                 seq_printf(m, ",caps_wanted_delay_max=%d",
 187                            args->caps_wanted_delay_max);
 188         if (args->cap_release_safety != CEPH_CAP_RELEASE_SAFETY_DEFAULT)
 189                 seq_printf(m, ",cap_release_safety=%d",
 190                            args->cap_release_safety);
 191         if (args->max_readdir != CEPH_MAX_READDIR_DEFAULT)
 192                 seq_printf(m, ",readdir_max_entries=%d", args->max_readdir);
 193         if (strcmp(args->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
 194                 seq_printf(m, ",snapdirname=%s", args->snapdir_name);
 195         if (args->name)
 196                 seq_printf(m, ",name=%s", args->name);
 197         if (args->secret)
 198                 seq_puts(m, ",secret=<hidden>");
 199         return 0;
 200 }
 201
 202 /*
 203  * caches
 204  */
 205 struct kmem_cache *ceph_inode_cachep;
 206 struct kmem_cache *ceph_cap_cachep;
 207 struct kmem_cache *ceph_dentry_cachep;
 208 struct kmem_cache *ceph_file_cachep;
 209
 210 static void ceph_inode_init_once(void *foo)
 211 {
 212         struct ceph_inode_info *ci = foo;
 213         inode_init_once(&ci->vfs_inode);
 214 }
 215
 216 static int __init init_caches(void)
 217 {
 218         ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
 219                                       sizeof(struct ceph_inode_info),
 220                                       __alignof__(struct ceph_inode_info),
 221                                       (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
 222                                       ceph_inode_init_once);
 223         if (ceph_inode_cachep == NULL)
 224                 return -ENOMEM;
 225
 226         ceph_cap_cachep = KMEM_CACHE(ceph_cap,
 227                                      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 228         if (ceph_cap_cachep == NULL)
 229                 goto bad_cap;
 230
 231         ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
 232                                         SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 233         if (ceph_dentry_cachep == NULL)
 234                 goto bad_dentry;
 235
 236         ceph_file_cachep = KMEM_CACHE(ceph_file_info,
 237                                       SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 238         if (ceph_file_cachep == NULL)
 239                 goto bad_file;
 240
 241         return 0;
 242
 243 bad_file:
 244         kmem_cache_destroy(ceph_dentry_cachep);
 245 bad_dentry:
 246         kmem_cache_destroy(ceph_cap_cachep);
 247 bad_cap:
 248         kmem_cache_destroy(ceph_inode_cachep);
 249         return -ENOMEM;
 250 }
 251
 252 static void destroy_caches(void)
 253 {
 254         kmem_cache_destroy(ceph_inode_cachep);
 255         kmem_cache_destroy(ceph_cap_cachep);
 256         kmem_cache_destroy(ceph_dentry_cachep);
 257         kmem_cache_destroy(ceph_file_cachep);
 258 }
 259
 260
 261 /*
 262  * ceph_umount_begin - initiate forced umount.  Tear down down the
 263  * mount, skipping steps that may hang while waiting for server(s).
 264  */
 265 static void ceph_umount_begin(struct super_block *sb)
 266 {
 267         struct ceph_client *client = ceph_sb_to_client(sb);
 268
 269         dout("ceph_umount_begin - starting forced umount\n");
 270         if (!client)
 271                 return;
 272         client->mount_state = CEPH_MOUNT_SHUTDOWN;
 273         return;
 274 }
 275
 276 static const struct super_operations ceph_super_ops = {
 277         .alloc_inode    = ceph_alloc_inode,
 278         .destroy_inode  = ceph_destroy_inode,
 279         .write_inode    = ceph_write_inode,
 280         .sync_fs        = ceph_syncfs,
 281         .put_super      = ceph_put_super,
 282         .show_options   = ceph_show_options,
 283         .statfs         = ceph_statfs,
 284         .umount_begin   = ceph_umount_begin,
 285 };
 286
 287
 288 const char *ceph_msg_type_name(int type)
 289 {
 290         switch (type) {
 291         case CEPH_MSG_SHUTDOWN: return "shutdown";
 292         case CEPH_MSG_PING: return "ping";
 293         case CEPH_MSG_AUTH: return "auth";
 294         case CEPH_MSG_AUTH_REPLY: return "auth_reply";
 295         case CEPH_MSG_MON_MAP: return "mon_map";
 296         case CEPH_MSG_MON_GET_MAP: return "mon_get_map";
 297         case CEPH_MSG_MON_SUBSCRIBE: return "mon_subscribe";
 298         case CEPH_MSG_MON_SUBSCRIBE_ACK: return "mon_subscribe_ack";
 299         case CEPH_MSG_STATFS: return "statfs";
 300         case CEPH_MSG_STATFS_REPLY: return "statfs_reply";
 301         case CEPH_MSG_MDS_MAP: return "mds_map";
 302         case CEPH_MSG_CLIENT_SESSION: return "client_session";
 303         case CEPH_MSG_CLIENT_RECONNECT: return "client_reconnect";
 304         case CEPH_MSG_CLIENT_REQUEST: return "client_request";
 305         case CEPH_MSG_CLIENT_REQUEST_FORWARD: return "client_request_forward";
 306         case CEPH_MSG_CLIENT_REPLY: return "client_reply";
 307         case CEPH_MSG_CLIENT_CAPS: return "client_caps";
 308         case CEPH_MSG_CLIENT_CAPRELEASE: return "client_cap_release";
 309         case CEPH_MSG_CLIENT_SNAP: return "client_snap";
 310         case CEPH_MSG_CLIENT_LEASE: return "client_lease";
 311         case CEPH_MSG_OSD_MAP: return "osd_map";
 312         case CEPH_MSG_OSD_OP: return "osd_op";
 313         case CEPH_MSG_OSD_OPREPLY: return "osd_opreply";
 314         default: return "unknown";
 315         }
 316 }
 317
 318
 319 /*
 320  * mount options
 321  */
 322 enum {
 323         Opt_fsidmajor,
 324         Opt_fsidminor,
 325         Opt_monport,
 326         Opt_wsize,
 327         Opt_rsize,
 328         Opt_osdtimeout,
 329         Opt_osdkeepalivetimeout,
 330         Opt_mount_timeout,
 331         Opt_osd_idle_ttl,
 332         Opt_caps_wanted_delay_min,
 333         Opt_caps_wanted_delay_max,
 334         Opt_cap_release_safety,
 335         Opt_readdir_max_entries,
 336         Opt_congestion_kb,
 337         Opt_last_int,
 338         /* int args above */
 339         Opt_snapdirname,
 340         Opt_name,
 341         Opt_secret,
 342         Opt_last_string,
 343         /* string args above */
 344         Opt_ip,
 345         Opt_noshare,
 346         Opt_dirstat,
 347         Opt_nodirstat,
 348         Opt_rbytes,
 349         Opt_norbytes,
 350         Opt_nocrc,
 351         Opt_noasyncreaddir,
 352 };
 353
 354 static match_table_t arg_tokens = {
 355         {Opt_fsidmajor, "fsidmajor=%ld"},
 356         {Opt_fsidminor, "fsidminor=%ld"},
 357         {Opt_monport, "monport=%d"},
 358         {Opt_wsize, "wsize=%d"},
 359         {Opt_rsize, "rsize=%d"},
 360         {Opt_osdtimeout, "osdtimeout=%d"},
 361         {Opt_osdkeepalivetimeout, "osdkeepalive=%d"},
 362         {Opt_mount_timeout, "mount_timeout=%d"},
 363         {Opt_osd_idle_ttl, "osd_idle_ttl=%d"},
 364         {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
 365         {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
 366         {Opt_cap_release_safety, "cap_release_safety=%d"},
 367         {Opt_readdir_max_entries, "readdir_max_entries=%d"},
 368         {Opt_congestion_kb, "write_congestion_kb=%d"},
 369         /* int args above */
 370         {Opt_snapdirname, "snapdirname=%s"},
 371         {Opt_name, "name=%s"},
 372         {Opt_secret, "secret=%s"},
 373         /* string args above */
 374         {Opt_ip, "ip=%s"},
 375         {Opt_noshare, "noshare"},
 376         {Opt_dirstat, "dirstat"},
 377         {Opt_nodirstat, "nodirstat"},
 378         {Opt_rbytes, "rbytes"},
 379         {Opt_norbytes, "norbytes"},
 380         {Opt_nocrc, "nocrc"},
 381         {Opt_noasyncreaddir, "noasyncreaddir"},
 382         {-1, NULL}
 383 };
 384
 385
 386 static struct ceph_mount_args *parse_mount_args(int flags, char *options,
 387                                                 const char *dev_name,
 388                                                 const char **path)
 389 {
 390         struct ceph_mount_args *args;
 391         const char *c;
 392         int err = -ENOMEM;
 393         substring_t argstr[MAX_OPT_ARGS];
 394
 395         args = kzalloc(sizeof(*args), GFP_KERNEL);
 396         if (!args)
 397                 return ERR_PTR(-ENOMEM);
 398         args->mon_addr = kcalloc(CEPH_MAX_MON, sizeof(*args->mon_addr),
 399                                  GFP_KERNEL);
 400         if (!args->mon_addr)
 401                 goto out;
 402
 403         dout("parse_mount_args %p, dev_name '%s'\n", args, dev_name);
 404
 405         /* start with defaults */
 406         args->sb_flags = flags;
 407         args->flags = CEPH_OPT_DEFAULT;
 408         args->osd_timeout = CEPH_OSD_TIMEOUT_DEFAULT;
 409         args->osd_keepalive_timeout = CEPH_OSD_KEEPALIVE_DEFAULT;
 410         args->mount_timeout = CEPH_MOUNT_TIMEOUT_DEFAULT; /* seconds */
 411         args->osd_idle_ttl = CEPH_OSD_IDLE_TTL_DEFAULT;   /* seconds */
 412         args->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
 413         args->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
 414         args->rsize = CEPH_MOUNT_RSIZE_DEFAULT;
 415         args->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
 416         args->cap_release_safety = CEPH_CAP_RELEASE_SAFETY_DEFAULT;
 417         args->max_readdir = CEPH_MAX_READDIR_DEFAULT;
 418         args->congestion_kb = default_congestion_kb();
 419
 420         /* ip1[:port1][,ip2[:port2]...]:/subdir/in/fs */
 421         err = -EINVAL;
 422         if (!dev_name)
 423                 goto out;
 424         *path = strstr(dev_name, ":/");
 425         if (*path == NULL) {
 426                 pr_err("device name is missing path (no :/ in %s)\n",
 427                        dev_name);
 428                 goto out;
 429         }
 430
 431         /* get mon ip(s) */
 432         err = ceph_parse_ips(dev_name, *path, args->mon_addr,
 433                              CEPH_MAX_MON, &args->num_mon);
 434         if (err < 0)
 435                 goto out;
 436
 437         /* path on server */
 438         *path += 2;
 439         dout("server path '%s'\n", *path);
 440
 441         /* parse mount options */
 442         while ((c = strsep(&options, ",")) != NULL) {
 443                 int token, intval, ret;
 444                 if (!*c)
 445                         continue;
 446                 err = -EINVAL;
 447                 token = match_token((char *)c, arg_tokens, argstr);
 448                 if (token < 0) {
 449                         pr_err("bad mount option at '%s'\n", c);
 450                         goto out;
 451                 }
 452                 if (token < Opt_last_int) {
 453                         ret = match_int(&argstr[0], &intval);
 454                         if (ret < 0) {
 455                                 pr_err("bad mount option arg (not int) "
 456                                        "at '%s'\n", c);
 457                                 continue;
 458                         }
 459                         dout("got int token %d val %d\n", token, intval);
 460                 } else if (token > Opt_last_int && token < Opt_last_string) {
 461                         dout("got string token %d val %s\n", token,
 462                              argstr[0].from);
 463                 } else {
 464                         dout("got token %d\n", token);
 465                 }
 466                 switch (token) {
 467                 case Opt_fsidmajor:
 468                         *(__le64 *)&args->fsid.fsid[0] = cpu_to_le64(intval);
 469                         break;
 470                 case Opt_fsidminor:
 471                         *(__le64 *)&args->fsid.fsid[8] = cpu_to_le64(intval);
 472                         break;
 473                 case Opt_ip:
 474                         err = ceph_parse_ips(argstr[0].from,
 475                                              argstr[0].to,
 476                                              &args->my_addr,
 477                                              1, NULL);
 478                         if (err < 0)
 479                                 goto out;
 480                         args->flags |= CEPH_OPT_MYIP;
 481                         break;
 482
 483                 case Opt_snapdirname:
 484                         kfree(args->snapdir_name);
 485                         args->snapdir_name = kstrndup(argstr[0].from,
 486                                               argstr[0].to-argstr[0].from,
 487                                               GFP_KERNEL);
 488                         break;
 489                 case Opt_name:
 490                         args->name = kstrndup(argstr[0].from,
 491                                               argstr[0].to-argstr[0].from,
 492                                               GFP_KERNEL);
 493                         break;
 494                 case Opt_secret:
 495                         args->secret = kstrndup(argstr[0].from,
 496                                                 argstr[0].to-argstr[0].from,
 497                                                 GFP_KERNEL);
 498                         break;
 499
 500                         /* misc */
 501                 case Opt_wsize:
 502                         args->wsize = intval;
 503                         break;
 504                 case Opt_rsize:
 505                         args->rsize = intval;
 506                         break;
 507                 case Opt_osdtimeout:
 508                         args->osd_timeout = intval;
 509                         break;
 510                 case Opt_osdkeepalivetimeout:
 511                         args->osd_keepalive_timeout = intval;
 512                         break;
 513                 case Opt_mount_timeout:
 514                         args->mount_timeout = intval;
 515                         break;
 516                 case Opt_caps_wanted_delay_min:
 517                         args->caps_wanted_delay_min = intval;
 518                         break;
 519                 case Opt_caps_wanted_delay_max:
 520                         args->caps_wanted_delay_max = intval;
 521                         break;
 522                 case Opt_readdir_max_entries:
 523                         args->max_readdir = intval;
 524                         break;
 525                 case Opt_congestion_kb:
 526                         args->congestion_kb = intval;
 527                         break;
 528
 529                 case Opt_noshare:
 530                         args->flags |= CEPH_OPT_NOSHARE;
 531                         break;
 532
 533                 case Opt_dirstat:
 534                         args->flags |= CEPH_OPT_DIRSTAT;
 535                         break;
 536                 case Opt_nodirstat:
 537                         args->flags &= ~CEPH_OPT_DIRSTAT;
 538                         break;
 539                 case Opt_rbytes:
 540                         args->flags |= CEPH_OPT_RBYTES;
 541                         break;
 542                 case Opt_norbytes:
 543                         args->flags &= ~CEPH_OPT_RBYTES;
 544                         break;
 545                 case Opt_nocrc:
 546                         args->flags |= CEPH_OPT_NOCRC;
 547                         break;
 548                 case Opt_noasyncreaddir:
 549                         args->flags |= CEPH_OPT_NOASYNCREADDIR;
 550                         break;
 551
 552                 default:
 553                         BUG_ON(token);
 554                 }
 555         }
 556         return args;
 557
 558 out:
 559         kfree(args->mon_addr);
 560         kfree(args);
 561         return ERR_PTR(err);
 562 }
 563
 564 static void destroy_mount_args(struct ceph_mount_args *args)
 565 {
 566         dout("destroy_mount_args %p\n", args);
 567         kfree(args->snapdir_name);
 568         args->snapdir_name = NULL;
 569         kfree(args->name);
 570         args->name = NULL;
 571         kfree(args->secret);
 572         args->secret = NULL;
 573         kfree(args);
 574 }
 575
 576 /*
 577  * create a fresh client instance
 578  */
 579 static struct ceph_client *ceph_create_client(struct ceph_mount_args *args)
 580 {
 581         struct ceph_client *client;
 582         int err = -ENOMEM;
 583
 584         client = kzalloc(sizeof(*client), GFP_KERNEL);
 585         if (client == NULL)
 586                 return ERR_PTR(-ENOMEM);
 587
 588         mutex_init(&client->mount_mutex);
 589
 590         init_waitqueue_head(&client->auth_wq);
 591
 592         client->sb = NULL;
 593         client->mount_state = CEPH_MOUNT_MOUNTING;
 594         client->mount_args = args;
 595
 596         client->msgr = NULL;
 597
 598         client->auth_err = 0;
 599         atomic_long_set(&client->writeback_count, 0);
 600
 601         err = bdi_init(&client->backing_dev_info);
 602         if (err < 0)
 603                 goto fail;
 604
 605         err = -ENOMEM;
 606         client->wb_wq = create_workqueue("ceph-writeback");
 607         if (client->wb_wq == NULL)
 608                 goto fail_bdi;
 609         client->pg_inv_wq = create_singlethread_workqueue("ceph-pg-invalid");
 610         if (client->pg_inv_wq == NULL)
 611                 goto fail_wb_wq;
 612         client->trunc_wq = create_singlethread_workqueue("ceph-trunc");
 613         if (client->trunc_wq == NULL)
 614                 goto fail_pg_inv_wq;
 615
 616         /* set up mempools */
 617         err = -ENOMEM;
 618         client->wb_pagevec_pool = mempool_create_kmalloc_pool(10,
 619                               client->mount_args->wsize >> PAGE_CACHE_SHIFT);
 620         if (!client->wb_pagevec_pool)
 621                 goto fail_trunc_wq;
 622
 623         /* caps */
 624         client->min_caps = args->max_readdir;
 625         ceph_adjust_min_caps(client->min_caps);
 626
 627         /* subsystems */
 628         err = ceph_monc_init(&client->monc, client);
 629         if (err < 0)
 630                 goto fail_mempool;
 631         err = ceph_osdc_init(&client->osdc, client);
 632         if (err < 0)
 633                 goto fail_monc;
 634         err = ceph_mdsc_init(&client->mdsc, client);
 635         if (err < 0)
 636                 goto fail_osdc;
 637         return client;
 638
 639 fail_osdc:
 640         ceph_osdc_stop(&client->osdc);
 641 fail_monc:
 642         ceph_monc_stop(&client->monc);
 643 fail_mempool:
 644         mempool_destroy(client->wb_pagevec_pool);
 645 fail_trunc_wq:
 646         destroy_workqueue(client->trunc_wq);
 647 fail_pg_inv_wq:
 648         destroy_workqueue(client->pg_inv_wq);
 649 fail_wb_wq:
 650         destroy_workqueue(client->wb_wq);
 651 fail_bdi:
 652         bdi_destroy(&client->backing_dev_info);
 653 fail:
 654         kfree(client);
 655         return ERR_PTR(err);
 656 }
 657
 658 static void ceph_destroy_client(struct ceph_client *client)
 659 {
 660         dout("destroy_client %p\n", client);
 661
 662         /* unmount */
 663         ceph_mdsc_stop(&client->mdsc);
 664         ceph_monc_stop(&client->monc);
 665         ceph_osdc_stop(&client->osdc);
 666
 667         ceph_adjust_min_caps(-client->min_caps);
 668
 669         ceph_debugfs_client_cleanup(client);
 670         destroy_workqueue(client->wb_wq);
 671         destroy_workqueue(client->pg_inv_wq);
 672         destroy_workqueue(client->trunc_wq);
 673
 674         bdi_destroy(&client->backing_dev_info);
 675
 676         if (client->msgr)
 677                 ceph_messenger_destroy(client->msgr);
 678         mempool_destroy(client->wb_pagevec_pool);
 679
 680         destroy_mount_args(client->mount_args);
 681
 682         kfree(client);
 683         dout("destroy_client %p done\n", client);
 684 }
 685
 686 /*
 687  * Initially learn our fsid, or verify an fsid matches.
 688  */
 689 int ceph_check_fsid(struct ceph_client *client, struct ceph_fsid *fsid)
 690 {
 691         if (client->have_fsid) {
 692                 if (ceph_fsid_compare(&client->fsid, fsid)) {
 693                         pr_err("bad fsid, had " FSID_FORMAT " got " FSID_FORMAT,
 694                                PR_FSID(&client->fsid), PR_FSID(fsid));
 695                         return -1;
 696                 }
 697         } else {
 698                 pr_info("client%lld fsid " FSID_FORMAT "\n",
 699                         client->monc.auth->global_id, PR_FSID(fsid));
 700                 memcpy(&client->fsid, fsid, sizeof(*fsid));
 701                 ceph_debugfs_client_init(client);
 702                 client->have_fsid = true;
 703         }
 704         return 0;
 705 }
 706
 707 /*
 708  * true if we have the mon map (and have thus joined the cluster)
 709  */
 710 static int have_mon_and_osd_map(struct ceph_client *client)
 711 {
 712         return client->monc.monmap && client->monc.monmap->epoch &&
 713                client->osdc.osdmap && client->osdc.osdmap->epoch;
 714 }
 715
 716 /*
 717  * Bootstrap mount by opening the root directory.  Note the mount
 718  * @started time from caller, and time out if this takes too long.
 719  */
 720 static struct dentry *open_root_dentry(struct ceph_client *client,
 721                                        const char *path,
 722                                        unsigned long started)
 723 {
 724         struct ceph_mds_client *mdsc = &client->mdsc;
 725         struct ceph_mds_request *req = NULL;
 726         int err;
 727         struct dentry *root;
 728
 729         /* open dir */
 730         dout("open_root_inode opening '%s'\n", path);
 731         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
 732         if (IS_ERR(req))
 733                 return ERR_PTR(PTR_ERR(req));
 734         req->r_path1 = kstrdup(path, GFP_NOFS);
 735         req->r_ino1.ino = CEPH_INO_ROOT;
 736         req->r_ino1.snap = CEPH_NOSNAP;
 737         req->r_started = started;
 738         req->r_timeout = client->mount_args->mount_timeout * HZ;
 739         req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
 740         req->r_num_caps = 2;
 741         err = ceph_mdsc_do_request(mdsc, NULL, req);
 742         if (err == 0) {
 743                 dout("open_root_inode success\n");
 744                 if (ceph_ino(req->r_target_inode) == CEPH_INO_ROOT &&
 745                     client->sb->s_root == NULL)
 746                         root = d_alloc_root(req->r_target_inode);
 747                 else
 748                         root = d_obtain_alias(req->r_target_inode);
 749                 req->r_target_inode = NULL;
 750                 dout("open_root_inode success, root dentry is %p\n", root);
 751         } else {
 752                 root = ERR_PTR(err);
 753         }
 754         ceph_mdsc_put_request(req);
 755         return root;
 756 }
 757
 758 /*
 759  * mount: join the ceph cluster, and open root directory.
 760  */
 761 static int ceph_mount(struct ceph_client *client, struct vfsmount *mnt,
 762                       const char *path)
 763 {
 764         struct ceph_entity_addr *myaddr = NULL;
 765         int err;
 766         unsigned long timeout = client->mount_args->mount_timeout * HZ;
 767         unsigned long started = jiffies;  /* note the start time */
 768         struct dentry *root;
 769
 770         dout("mount start\n");
 771         mutex_lock(&client->mount_mutex);
 772
 773         /* initialize the messenger */
 774         if (client->msgr == NULL) {
 775                 if (ceph_test_opt(client, MYIP))
 776                         myaddr = &client->mount_args->my_addr;
 777                 client->msgr = ceph_messenger_create(myaddr);
 778                 if (IS_ERR(client->msgr)) {
 779                         err = PTR_ERR(client->msgr);
 780                         client->msgr = NULL;
 781                         goto out;
 782                 }
 783                 client->msgr->nocrc = ceph_test_opt(client, NOCRC);
 784         }
 785
 786         /* open session, and wait for mon, mds, and osd maps */
 787         err = ceph_monc_open_session(&client->monc);
 788         if (err < 0)
 789                 goto out;
 790
 791         while (!have_mon_and_osd_map(client)) {
 792                 err = -EIO;
 793                 if (timeout && time_after_eq(jiffies, started + timeout))
 794                         goto out;
 795
 796                 /* wait */
 797                 dout("mount waiting for mon_map\n");
 798                 err = wait_event_interruptible_timeout(client->auth_wq,
 799                        have_mon_and_osd_map(client) || (client->auth_err < 0),
 800                        timeout);
 801                 if (err == -EINTR || err == -ERESTARTSYS)
 802                         goto out;
 803                 if (client->auth_err < 0) {
 804                         err = client->auth_err;
 805                         goto out;
 806                 }
 807         }
 808
 809         dout("mount opening root\n");
 810         root = open_root_dentry(client, "", started);
 811         if (IS_ERR(root)) {
 812                 err = PTR_ERR(root);
 813                 goto out;
 814         }
 815         if (client->sb->s_root)
 816                 dput(root);
 817         else
 818                 client->sb->s_root = root;
 819
 820         if (path[0] == 0) {
 821                 dget(root);
 822         } else {
 823                 dout("mount opening base mountpoint\n");
 824                 root = open_root_dentry(client, path, started);
 825                 if (IS_ERR(root)) {
 826                         err = PTR_ERR(root);
 827                         dput(client->sb->s_root);
 828                         client->sb->s_root = NULL;
 829                         goto out;
 830                 }
 831         }
 832
 833         mnt->mnt_root = root;
 834         mnt->mnt_sb = client->sb;
 835
 836         client->mount_state = CEPH_MOUNT_MOUNTED;
 837         dout("mount success\n");
 838         err = 0;
 839
 840 out:
 841         mutex_unlock(&client->mount_mutex);
 842         return err;
 843 }
 844
 845 static int ceph_set_super(struct super_block *s, void *data)
 846 {
 847         struct ceph_client *client = data;
 848         int ret;
 849
 850         dout("set_super %p data %p\n", s, data);
 851
 852         s->s_flags = client->mount_args->sb_flags;
 853         s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
 854
 855         s->s_fs_info = client;
 856         client->sb = s;
 857
 858         s->s_op = &ceph_super_ops;
 859         s->s_export_op = &ceph_export_ops;
 860
 861         s->s_time_gran = 1000;  /* 1000 ns == 1 us */
 862
 863         ret = set_anon_super(s, NULL);  /* what is that second arg for? */
 864         if (ret != 0)
 865                 goto fail;
 866
 867         return ret;
 868
 869 fail:
 870         s->s_fs_info = NULL;
 871         client->sb = NULL;
 872         return ret;
 873 }
 874
 875 /*
 876  * share superblock if same fs AND options
 877  */
 878 static int ceph_compare_super(struct super_block *sb, void *data)
 879 {
 880         struct ceph_client *new = data;
 881         struct ceph_mount_args *args = new->mount_args;
 882         struct ceph_client *other = ceph_sb_to_client(sb);
 883         int i;
 884
 885         dout("ceph_compare_super %p\n", sb);
 886         if (args->flags & CEPH_OPT_FSID) {
 887                 if (ceph_fsid_compare(&args->fsid, &other->fsid)) {
 888                         dout("fsid doesn't match\n");
 889                         return 0;
 890                 }
 891         } else {
 892                 /* do we share (a) monitor? */
 893                 for (i = 0; i < new->monc.monmap->num_mon; i++)
 894                         if (ceph_monmap_contains(other->monc.monmap,
 895                                          &new->monc.monmap->mon_inst[i].addr))
 896                                 break;
 897                 if (i == new->monc.monmap->num_mon) {
 898                         dout("mon ip not part of monmap\n");
 899                         return 0;
 900                 }
 901                 dout("mon ip matches existing sb %p\n", sb);
 902         }
 903         if (args->sb_flags != other->mount_args->sb_flags) {
 904                 dout("flags differ\n");
 905                 return 0;
 906         }
 907         return 1;
 908 }
 909
 910 /*
 911  * construct our own bdi so we can control readahead, etc.
 912  */
 913 static int ceph_register_bdi(struct super_block *sb, struct ceph_client *client)
 914 {
 915         int err;
 916
 917         /* set ra_pages based on rsize mount option? */
 918         if (client->mount_args->rsize >= PAGE_CACHE_SIZE)
 919                 client->backing_dev_info.ra_pages =
 920                         (client->mount_args->rsize + PAGE_CACHE_SIZE - 1)
 921                         >> PAGE_SHIFT;
 922         err = bdi_register_dev(&client->backing_dev_info, sb->s_dev);
 923         if (!err)
 924                 sb->s_bdi = &client->backing_dev_info;
 925         return err;
 926 }
 927
 928 static int ceph_get_sb(struct file_system_type *fs_type,
 929                        int flags, const char *dev_name, void *data,
 930                        struct vfsmount *mnt)
 931 {
 932         struct super_block *sb;
 933         struct ceph_client *client;
 934         int err;
 935         int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
 936         const char *path = NULL;
 937         struct ceph_mount_args *args;
 938
 939         dout("ceph_get_sb\n");
 940         args = parse_mount_args(flags, data, dev_name, &path);
 941         if (IS_ERR(args)) {
 942                 err = PTR_ERR(args);
 943                 goto out_final;
 944         }
 945
 946         /* create client (which we may/may not use) */
 947         client = ceph_create_client(args);
 948         if (IS_ERR(client)) {
 949                 err = PTR_ERR(client);
 950                 goto out_final;
 951         }
 952
 953         if (client->mount_args->flags & CEPH_OPT_NOSHARE)
 954                 compare_super = NULL;
 955         sb = sget(fs_type, compare_super, ceph_set_super, client);
 956         if (IS_ERR(sb)) {
 957                 err = PTR_ERR(sb);
 958                 goto out;
 959         }
 960
 961         if (ceph_sb_to_client(sb) != client) {
 962                 ceph_destroy_client(client);
 963                 client = ceph_sb_to_client(sb);
 964                 dout("get_sb got existing client %p\n", client);
 965         } else {
 966                 dout("get_sb using new client %p\n", client);
 967                 err = ceph_register_bdi(sb, client);
 968                 if (err < 0)
 969                         goto out_splat;
 970         }
 971
 972         err = ceph_mount(client, mnt, path);
 973         if (err < 0)
 974                 goto out_splat;
 975         dout("root %p inode %p ino %llx.%llx\n", mnt->mnt_root,
 976              mnt->mnt_root->d_inode, ceph_vinop(mnt->mnt_root->d_inode));
 977         return 0;
 978
 979 out_splat:
 980         ceph_mdsc_close_sessions(&client->mdsc);
 981         up_write(&sb->s_umount);
 982         deactivate_super(sb);
 983         goto out_final;
 984
 985 out:
 986         ceph_destroy_client(client);
 987 out_final:
 988         dout("ceph_get_sb fail %d\n", err);
 989         return err;
 990 }
 991
 992 static void ceph_kill_sb(struct super_block *s)
 993 {
 994         struct ceph_client *client = ceph_sb_to_client(s);
 995         dout("kill_sb %p\n", s);
 996         ceph_mdsc_pre_umount(&client->mdsc);
 997         kill_anon_super(s);    /* will call put_super after sb is r/o */
 998         ceph_destroy_client(client);
 999 }
1000
1001 static struct file_system_type ceph_fs_type = {
1002         .owner          = THIS_MODULE,
1003         .name           = "ceph",
1004         .get_sb         = ceph_get_sb,
1005         .kill_sb        = ceph_kill_sb,
1006         .fs_flags       = FS_RENAME_DOES_D_MOVE,
1007 };
1008
1009 #define _STRINGIFY(x) #x
1010 #define STRINGIFY(x) _STRINGIFY(x)
1011
1012 static int __init init_ceph(void)
1013 {
1014         int ret = 0;
1015
1016         ret = ceph_debugfs_init();
1017         if (ret < 0)
1018                 goto out;
1019
1020         ret = ceph_msgr_init();
1021         if (ret < 0)
1022                 goto out_debugfs;
1023
1024         ret = init_caches();
1025         if (ret)
1026                 goto out_msgr;
1027
1028         ceph_caps_init();
1029
1030         ret = register_filesystem(&ceph_fs_type);
1031         if (ret)
1032                 goto out_icache;
1033
1034         pr_info("loaded (mon/mds/osd proto %d/%d/%d, osdmap %d/%d %d/%d)\n",
1035                 CEPH_MONC_PROTOCOL, CEPH_MDSC_PROTOCOL, CEPH_OSDC_PROTOCOL,
1036                 CEPH_OSDMAP_VERSION, CEPH_OSDMAP_VERSION_EXT,
1037                 CEPH_OSDMAP_INC_VERSION, CEPH_OSDMAP_INC_VERSION_EXT);
1038         return 0;
1039
1040 out_icache:
1041         destroy_caches();
1042 out_msgr:
1043         ceph_msgr_exit();
1044 out_debugfs:
1045         ceph_debugfs_cleanup();
1046 out:
1047         return ret;
1048 }
1049
1050 static void __exit exit_ceph(void)
1051 {
1052         dout("exit_ceph\n");
1053         unregister_filesystem(&ceph_fs_type);
1054         ceph_caps_finalize();
1055         destroy_caches();
1056         ceph_msgr_exit();
1057         ceph_debugfs_cleanup();
1058 }
1059
1060 module_init(init_ceph);
1061 module_exit(exit_ceph);
1062
1063 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1064 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1065 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1066 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1067 MODULE_LICENSE("GPL");