SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/ceph/super.c

   1
   2 #include "ceph_debug.h"
   3
   4 #include <linux/backing-dev.h>
   5 #include <linux/fs.h>
   6 #include <linux/inet.h>
   7 #include <linux/in6.h>
   8 #include <linux/module.h>
   9 #include <linux/mount.h>
  10 #include <linux/parser.h>
  11 #include <linux/rwsem.h>
  12 #include <linux/sched.h>
  13 #include <linux/seq_file.h>
  14 #include <linux/statfs.h>
  15 #include <linux/string.h>
  16 #include <linux/version.h>
  17 #include <linux/vmalloc.h>
  18
  19 #include "decode.h"
  20 #include "super.h"
  21 #include "mon_client.h"
  22 #include "auth.h"
  23
  24 /*
  25  * Ceph superblock operations
  26  *
  27  * Handle the basics of mounting, unmounting.
  28  */
  29
  30
  31 /*
  32  * find filename portion of a path (/foo/bar/baz -> baz)
  33  */
  34 const char *ceph_file_part(const char *s, int len)
  35 {
  36         const char *e = s + len;
  37
  38         while (e != s && *(e-1) != '/')
  39                 e--;
  40         return e;
  41 }
  42
  43
  44 /*
  45  * super ops
  46  */
  47 static void ceph_put_super(struct super_block *s)
  48 {
  49         struct ceph_client *cl = ceph_client(s);
  50
  51         dout("put_super\n");
  52         ceph_mdsc_close_sessions(&cl->mdsc);
  53         return;
  54 }
  55
  56 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
  57 {
  58         struct ceph_client *client = ceph_inode_to_client(dentry->d_inode);
  59         struct ceph_monmap *monmap = client->monc.monmap;
  60         struct ceph_statfs st;
  61         u64 fsid;
  62         int err;
  63
  64         dout("statfs\n");
  65         err = ceph_monc_do_statfs(&client->monc, &st);
  66         if (err < 0)
  67                 return err;
  68
  69         /* fill in kstatfs */
  70         buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
  71
  72         /*
  73          * express utilization in terms of large blocks to avoid
  74          * overflow on 32-bit machines.
  75          */
  76         buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
  77         buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
  78         buf->f_bfree = (le64_to_cpu(st.kb) - le64_to_cpu(st.kb_used)) >>
  79                 (CEPH_BLOCK_SHIFT-10);
  80         buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
  81
  82         buf->f_files = le64_to_cpu(st.num_objects);
  83         buf->f_ffree = -1;
  84         buf->f_namelen = PATH_MAX;
  85         buf->f_frsize = PAGE_CACHE_SIZE;
  86
  87         /* leave fsid little-endian, regardless of host endianness */
  88         fsid = *(u64 *)(&monmap->fsid) ^ *((u64 *)&monmap->fsid + 1);
  89         buf->f_fsid.val[0] = fsid & 0xffffffff;
  90         buf->f_fsid.val[1] = fsid >> 32;
  91
  92         return 0;
  93 }
  94
  95
  96 static int ceph_syncfs(struct super_block *sb, int wait)
  97 {
  98         dout("sync_fs %d\n", wait);
  99         ceph_osdc_sync(&ceph_client(sb)->osdc);
 100         ceph_mdsc_sync(&ceph_client(sb)->mdsc);
 101         dout("sync_fs %d done\n", wait);
 102         return 0;
 103 }
 104
 105
 106 /**
 107  * ceph_show_options - Show mount options in /proc/mounts
 108  * @m: seq_file to write to
 109  * @mnt: mount descriptor
 110  */
 111 static int ceph_show_options(struct seq_file *m, struct vfsmount *mnt)
 112 {
 113         struct ceph_client *client = ceph_sb_to_client(mnt->mnt_sb);
 114         struct ceph_mount_args *args = client->mount_args;
 115
 116         if (args->flags & CEPH_OPT_FSID)
 117                 seq_printf(m, ",fsidmajor=%llu,fsidminor%llu",
 118                            le64_to_cpu(*(__le64 *)&args->fsid.fsid[0]),
 119                            le64_to_cpu(*(__le64 *)&args->fsid.fsid[8]));
 120         if (args->flags & CEPH_OPT_NOSHARE)
 121                 seq_puts(m, ",noshare");
 122         if (args->flags & CEPH_OPT_DIRSTAT)
 123                 seq_puts(m, ",dirstat");
 124         if ((args->flags & CEPH_OPT_RBYTES) == 0)
 125                 seq_puts(m, ",norbytes");
 126         if (args->flags & CEPH_OPT_NOCRC)
 127                 seq_puts(m, ",nocrc");
 128         if (args->flags & CEPH_OPT_NOASYNCREADDIR)
 129                 seq_puts(m, ",noasyncreaddir");
 130         if (strcmp(args->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
 131                 seq_printf(m, ",snapdirname=%s", args->snapdir_name);
 132         if (args->name)
 133                 seq_printf(m, ",name=%s", args->name);
 134         if (args->secret)
 135                 seq_puts(m, ",secret=<hidden>");
 136         return 0;
 137 }
 138
 139 /*
 140  * caches
 141  */
 142 struct kmem_cache *ceph_inode_cachep;
 143 struct kmem_cache *ceph_cap_cachep;
 144 struct kmem_cache *ceph_dentry_cachep;
 145 struct kmem_cache *ceph_file_cachep;
 146
 147 static void ceph_inode_init_once(void *foo)
 148 {
 149         struct ceph_inode_info *ci = foo;
 150         inode_init_once(&ci->vfs_inode);
 151 }
 152
 153 static int default_congestion_kb(void)
 154 {
 155         int congestion_kb;
 156
 157         /*
 158          * Copied from NFS
 159          *
 160          * congestion size, scale with available memory.
 161          *
 162          *  64MB:    8192k
 163          * 128MB:   11585k
 164          * 256MB:   16384k
 165          * 512MB:   23170k
 166          *   1GB:   32768k
 167          *   2GB:   46340k
 168          *   4GB:   65536k
 169          *   8GB:   92681k
 170          *  16GB:  131072k
 171          *
 172          * This allows larger machines to have larger/more transfers.
 173          * Limit the default to 256M
 174          */
 175         congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
 176         if (congestion_kb > 256*1024)
 177                 congestion_kb = 256*1024;
 178
 179         return congestion_kb;
 180 }
 181
 182 static int __init init_caches(void)
 183 {
 184         ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
 185                                       sizeof(struct ceph_inode_info),
 186                                       __alignof__(struct ceph_inode_info),
 187                                       (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
 188                                       ceph_inode_init_once);
 189         if (ceph_inode_cachep == NULL)
 190                 return -ENOMEM;
 191
 192         ceph_cap_cachep = KMEM_CACHE(ceph_cap,
 193                                      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 194         if (ceph_cap_cachep == NULL)
 195                 goto bad_cap;
 196
 197         ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
 198                                         SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 199         if (ceph_dentry_cachep == NULL)
 200                 goto bad_dentry;
 201
 202         ceph_file_cachep = KMEM_CACHE(ceph_file_info,
 203                                       SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 204         if (ceph_file_cachep == NULL)
 205                 goto bad_file;
 206
 207         return 0;
 208
 209 bad_file:
 210         kmem_cache_destroy(ceph_dentry_cachep);
 211 bad_dentry:
 212         kmem_cache_destroy(ceph_cap_cachep);
 213 bad_cap:
 214         kmem_cache_destroy(ceph_inode_cachep);
 215         return -ENOMEM;
 216 }
 217
 218 static void destroy_caches(void)
 219 {
 220         kmem_cache_destroy(ceph_inode_cachep);
 221         kmem_cache_destroy(ceph_cap_cachep);
 222         kmem_cache_destroy(ceph_dentry_cachep);
 223         kmem_cache_destroy(ceph_file_cachep);
 224 }
 225
 226
 227 /*
 228  * ceph_umount_begin - initiate forced umount.  Tear down down the
 229  * mount, skipping steps that may hang while waiting for server(s).
 230  */
 231 static void ceph_umount_begin(struct super_block *sb)
 232 {
 233         struct ceph_client *client = ceph_sb_to_client(sb);
 234
 235         dout("ceph_umount_begin - starting forced umount\n");
 236         if (!client)
 237                 return;
 238         client->mount_state = CEPH_MOUNT_SHUTDOWN;
 239         return;
 240 }
 241
 242 static const struct super_operations ceph_super_ops = {
 243         .alloc_inode    = ceph_alloc_inode,
 244         .destroy_inode  = ceph_destroy_inode,
 245         .write_inode    = ceph_write_inode,
 246         .sync_fs        = ceph_syncfs,
 247         .put_super      = ceph_put_super,
 248         .show_options   = ceph_show_options,
 249         .statfs         = ceph_statfs,
 250         .umount_begin   = ceph_umount_begin,
 251 };
 252
 253
 254 const char *ceph_msg_type_name(int type)
 255 {
 256         switch (type) {
 257         case CEPH_MSG_SHUTDOWN: return "shutdown";
 258         case CEPH_MSG_PING: return "ping";
 259         case CEPH_MSG_AUTH: return "auth";
 260         case CEPH_MSG_AUTH_REPLY: return "auth_reply";
 261         case CEPH_MSG_MON_MAP: return "mon_map";
 262         case CEPH_MSG_MON_GET_MAP: return "mon_get_map";
 263         case CEPH_MSG_MON_SUBSCRIBE: return "mon_subscribe";
 264         case CEPH_MSG_MON_SUBSCRIBE_ACK: return "mon_subscribe_ack";
 265         case CEPH_MSG_STATFS: return "statfs";
 266         case CEPH_MSG_STATFS_REPLY: return "statfs_reply";
 267         case CEPH_MSG_MDS_MAP: return "mds_map";
 268         case CEPH_MSG_CLIENT_SESSION: return "client_session";
 269         case CEPH_MSG_CLIENT_RECONNECT: return "client_reconnect";
 270         case CEPH_MSG_CLIENT_REQUEST: return "client_request";
 271         case CEPH_MSG_CLIENT_REQUEST_FORWARD: return "client_request_forward";
 272         case CEPH_MSG_CLIENT_REPLY: return "client_reply";
 273         case CEPH_MSG_CLIENT_CAPS: return "client_caps";
 274         case CEPH_MSG_CLIENT_CAPRELEASE: return "client_cap_release";
 275         case CEPH_MSG_CLIENT_SNAP: return "client_snap";
 276         case CEPH_MSG_CLIENT_LEASE: return "client_lease";
 277         case CEPH_MSG_OSD_MAP: return "osd_map";
 278         case CEPH_MSG_OSD_OP: return "osd_op";
 279         case CEPH_MSG_OSD_OPREPLY: return "osd_opreply";
 280         default: return "unknown";
 281         }
 282 }
 283
 284
 285 /*
 286  * mount options
 287  */
 288 enum {
 289         Opt_fsidmajor,
 290         Opt_fsidminor,
 291         Opt_monport,
 292         Opt_wsize,
 293         Opt_rsize,
 294         Opt_osdtimeout,
 295         Opt_mount_timeout,
 296         Opt_osd_idle_ttl,
 297         Opt_caps_wanted_delay_min,
 298         Opt_caps_wanted_delay_max,
 299         Opt_readdir_max_entries,
 300         Opt_congestion_kb,
 301         Opt_last_int,
 302         /* int args above */
 303         Opt_snapdirname,
 304         Opt_name,
 305         Opt_secret,
 306         Opt_last_string,
 307         /* string args above */
 308         Opt_ip,
 309         Opt_noshare,
 310         Opt_dirstat,
 311         Opt_nodirstat,
 312         Opt_rbytes,
 313         Opt_norbytes,
 314         Opt_nocrc,
 315         Opt_noasyncreaddir,
 316 };
 317
 318 static match_table_t arg_tokens = {
 319         {Opt_fsidmajor, "fsidmajor=%ld"},
 320         {Opt_fsidminor, "fsidminor=%ld"},
 321         {Opt_monport, "monport=%d"},
 322         {Opt_wsize, "wsize=%d"},
 323         {Opt_rsize, "rsize=%d"},
 324         {Opt_osdtimeout, "osdtimeout=%d"},
 325         {Opt_mount_timeout, "mount_timeout=%d"},
 326         {Opt_osd_idle_ttl, "osd_idle_ttl=%d"},
 327         {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
 328         {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
 329         {Opt_readdir_max_entries, "readdir_max_entries=%d"},
 330         {Opt_congestion_kb, "write_congestion_kb=%d"},
 331         /* int args above */
 332         {Opt_snapdirname, "snapdirname=%s"},
 333         {Opt_name, "name=%s"},
 334         {Opt_secret, "secret=%s"},
 335         /* string args above */
 336         {Opt_ip, "ip=%s"},
 337         {Opt_noshare, "noshare"},
 338         {Opt_dirstat, "dirstat"},
 339         {Opt_nodirstat, "nodirstat"},
 340         {Opt_rbytes, "rbytes"},
 341         {Opt_norbytes, "norbytes"},
 342         {Opt_nocrc, "nocrc"},
 343         {Opt_noasyncreaddir, "noasyncreaddir"},
 344         {-1, NULL}
 345 };
 346
 347
 348 static struct ceph_mount_args *parse_mount_args(int flags, char *options,
 349                                                 const char *dev_name,
 350                                                 const char **path)
 351 {
 352         struct ceph_mount_args *args;
 353         const char *c;
 354         int err = -ENOMEM;
 355         substring_t argstr[MAX_OPT_ARGS];
 356
 357         args = kzalloc(sizeof(*args), GFP_KERNEL);
 358         if (!args)
 359                 return ERR_PTR(-ENOMEM);
 360         args->mon_addr = kcalloc(CEPH_MAX_MON, sizeof(*args->mon_addr),
 361                                  GFP_KERNEL);
 362         if (!args->mon_addr)
 363                 goto out;
 364
 365         dout("parse_mount_args %p, dev_name '%s'\n", args, dev_name);
 366
 367         /* start with defaults */
 368         args->sb_flags = flags;
 369         args->flags = CEPH_OPT_DEFAULT;
 370         args->osd_timeout = 5;    /* seconds */
 371         args->mount_timeout = CEPH_MOUNT_TIMEOUT_DEFAULT; /* seconds */
 372         args->osd_idle_ttl = CEPH_OSD_IDLE_TTL_DEFAULT;   /* seconds */
 373         args->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
 374         args->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
 375         args->rsize = CEPH_MOUNT_RSIZE_DEFAULT;
 376         args->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
 377         args->cap_release_safety = CEPH_CAPS_PER_RELEASE * 4;
 378         args->max_readdir = 1024;
 379         args->congestion_kb = default_congestion_kb();
 380
 381         /* ip1[:port1][,ip2[:port2]...]:/subdir/in/fs */
 382         err = -EINVAL;
 383         if (!dev_name)
 384                 goto out;
 385         *path = strstr(dev_name, ":/");
 386         if (*path == NULL) {
 387                 pr_err("device name is missing path (no :/ in %s)\n",
 388                        dev_name);
 389                 goto out;
 390         }
 391
 392         /* get mon ip(s) */
 393         err = ceph_parse_ips(dev_name, *path, args->mon_addr,
 394                              CEPH_MAX_MON, &args->num_mon);
 395         if (err < 0)
 396                 goto out;
 397
 398         /* path on server */
 399         *path += 2;
 400         dout("server path '%s'\n", *path);
 401
 402         /* parse mount options */
 403         while ((c = strsep(&options, ",")) != NULL) {
 404                 int token, intval, ret;
 405                 if (!*c)
 406                         continue;
 407                 err = -EINVAL;
 408                 token = match_token((char *)c, arg_tokens, argstr);
 409                 if (token < 0) {
 410                         pr_err("bad mount option at '%s'\n", c);
 411                         goto out;
 412                 }
 413                 if (token < Opt_last_int) {
 414                         ret = match_int(&argstr[0], &intval);
 415                         if (ret < 0) {
 416                                 pr_err("bad mount option arg (not int) "
 417                                        "at '%s'\n", c);
 418                                 continue;
 419                         }
 420                         dout("got int token %d val %d\n", token, intval);
 421                 } else if (token > Opt_last_int && token < Opt_last_string) {
 422                         dout("got string token %d val %s\n", token,
 423                              argstr[0].from);
 424                 } else {
 425                         dout("got token %d\n", token);
 426                 }
 427                 switch (token) {
 428                 case Opt_fsidmajor:
 429                         *(__le64 *)&args->fsid.fsid[0] = cpu_to_le64(intval);
 430                         break;
 431                 case Opt_fsidminor:
 432                         *(__le64 *)&args->fsid.fsid[8] = cpu_to_le64(intval);
 433                         break;
 434                 case Opt_ip:
 435                         err = ceph_parse_ips(argstr[0].from,
 436                                              argstr[0].to,
 437                                              &args->my_addr,
 438                                              1, NULL);
 439                         if (err < 0)
 440                                 goto out;
 441                         args->flags |= CEPH_OPT_MYIP;
 442                         break;
 443
 444                 case Opt_snapdirname:
 445                         kfree(args->snapdir_name);
 446                         args->snapdir_name = kstrndup(argstr[0].from,
 447                                               argstr[0].to-argstr[0].from,
 448                                               GFP_KERNEL);
 449                         break;
 450                 case Opt_name:
 451                         args->name = kstrndup(argstr[0].from,
 452                                               argstr[0].to-argstr[0].from,
 453                                               GFP_KERNEL);
 454                         break;
 455                 case Opt_secret:
 456                         args->secret = kstrndup(argstr[0].from,
 457                                                 argstr[0].to-argstr[0].from,
 458                                                 GFP_KERNEL);
 459                         break;
 460
 461                         /* misc */
 462                 case Opt_wsize:
 463                         args->wsize = intval;
 464                         break;
 465                 case Opt_rsize:
 466                         args->rsize = intval;
 467                         break;
 468                 case Opt_osdtimeout:
 469                         args->osd_timeout = intval;
 470                         break;
 471                 case Opt_mount_timeout:
 472                         args->mount_timeout = intval;
 473                         break;
 474                 case Opt_caps_wanted_delay_min:
 475                         args->caps_wanted_delay_min = intval;
 476                         break;
 477                 case Opt_caps_wanted_delay_max:
 478                         args->caps_wanted_delay_max = intval;
 479                         break;
 480                 case Opt_readdir_max_entries:
 481                         args->max_readdir = intval;
 482                         break;
 483                 case Opt_congestion_kb:
 484                         args->congestion_kb = intval;
 485                         break;
 486
 487                 case Opt_noshare:
 488                         args->flags |= CEPH_OPT_NOSHARE;
 489                         break;
 490
 491                 case Opt_dirstat:
 492                         args->flags |= CEPH_OPT_DIRSTAT;
 493                         break;
 494                 case Opt_nodirstat:
 495                         args->flags &= ~CEPH_OPT_DIRSTAT;
 496                         break;
 497                 case Opt_rbytes:
 498                         args->flags |= CEPH_OPT_RBYTES;
 499                         break;
 500                 case Opt_norbytes:
 501                         args->flags &= ~CEPH_OPT_RBYTES;
 502                         break;
 503                 case Opt_nocrc:
 504                         args->flags |= CEPH_OPT_NOCRC;
 505                         break;
 506                 case Opt_noasyncreaddir:
 507                         args->flags |= CEPH_OPT_NOASYNCREADDIR;
 508                         break;
 509
 510                 default:
 511                         BUG_ON(token);
 512                 }
 513         }
 514         return args;
 515
 516 out:
 517         kfree(args->mon_addr);
 518         kfree(args);
 519         return ERR_PTR(err);
 520 }
 521
 522 static void destroy_mount_args(struct ceph_mount_args *args)
 523 {
 524         dout("destroy_mount_args %p\n", args);
 525         kfree(args->snapdir_name);
 526         args->snapdir_name = NULL;
 527         kfree(args->name);
 528         args->name = NULL;
 529         kfree(args->secret);
 530         args->secret = NULL;
 531         kfree(args);
 532 }
 533
 534 /*
 535  * create a fresh client instance
 536  */
 537 static struct ceph_client *ceph_create_client(struct ceph_mount_args *args)
 538 {
 539         struct ceph_client *client;
 540         int err = -ENOMEM;
 541
 542         client = kzalloc(sizeof(*client), GFP_KERNEL);
 543         if (client == NULL)
 544                 return ERR_PTR(-ENOMEM);
 545
 546         mutex_init(&client->mount_mutex);
 547
 548         init_waitqueue_head(&client->auth_wq);
 549
 550         client->sb = NULL;
 551         client->mount_state = CEPH_MOUNT_MOUNTING;
 552         client->mount_args = args;
 553
 554         client->msgr = NULL;
 555
 556         client->auth_err = 0;
 557         atomic_long_set(&client->writeback_count, 0);
 558
 559         err = bdi_init(&client->backing_dev_info);
 560         if (err < 0)
 561                 goto fail;
 562
 563         err = -ENOMEM;
 564         client->wb_wq = create_workqueue("ceph-writeback");
 565         if (client->wb_wq == NULL)
 566                 goto fail_bdi;
 567         client->pg_inv_wq = create_singlethread_workqueue("ceph-pg-invalid");
 568         if (client->pg_inv_wq == NULL)
 569                 goto fail_wb_wq;
 570         client->trunc_wq = create_singlethread_workqueue("ceph-trunc");
 571         if (client->trunc_wq == NULL)
 572                 goto fail_pg_inv_wq;
 573
 574         /* set up mempools */
 575         err = -ENOMEM;
 576         client->wb_pagevec_pool = mempool_create_kmalloc_pool(10,
 577                               client->mount_args->wsize >> PAGE_CACHE_SHIFT);
 578         if (!client->wb_pagevec_pool)
 579                 goto fail_trunc_wq;
 580
 581         /* caps */
 582         client->min_caps = args->max_readdir;
 583         ceph_adjust_min_caps(client->min_caps);
 584
 585         /* subsystems */
 586         err = ceph_monc_init(&client->monc, client);
 587         if (err < 0)
 588                 goto fail_mempool;
 589         err = ceph_osdc_init(&client->osdc, client);
 590         if (err < 0)
 591                 goto fail_monc;
 592         err = ceph_mdsc_init(&client->mdsc, client);
 593         if (err < 0)
 594                 goto fail_osdc;
 595         return client;
 596
 597 fail_osdc:
 598         ceph_osdc_stop(&client->osdc);
 599 fail_monc:
 600         ceph_monc_stop(&client->monc);
 601 fail_mempool:
 602         mempool_destroy(client->wb_pagevec_pool);
 603 fail_trunc_wq:
 604         destroy_workqueue(client->trunc_wq);
 605 fail_pg_inv_wq:
 606         destroy_workqueue(client->pg_inv_wq);
 607 fail_wb_wq:
 608         destroy_workqueue(client->wb_wq);
 609 fail_bdi:
 610         bdi_destroy(&client->backing_dev_info);
 611 fail:
 612         kfree(client);
 613         return ERR_PTR(err);
 614 }
 615
 616 static void ceph_destroy_client(struct ceph_client *client)
 617 {
 618         dout("destroy_client %p\n", client);
 619
 620         /* unmount */
 621         ceph_mdsc_stop(&client->mdsc);
 622         ceph_monc_stop(&client->monc);
 623         ceph_osdc_stop(&client->osdc);
 624
 625         ceph_adjust_min_caps(-client->min_caps);
 626
 627         ceph_debugfs_client_cleanup(client);
 628         destroy_workqueue(client->wb_wq);
 629         destroy_workqueue(client->pg_inv_wq);
 630         destroy_workqueue(client->trunc_wq);
 631
 632         if (client->msgr)
 633                 ceph_messenger_destroy(client->msgr);
 634         mempool_destroy(client->wb_pagevec_pool);
 635
 636         destroy_mount_args(client->mount_args);
 637
 638         kfree(client);
 639         dout("destroy_client %p done\n", client);
 640 }
 641
 642 /*
 643  * Initially learn our fsid, or verify an fsid matches.
 644  */
 645 int ceph_check_fsid(struct ceph_client *client, struct ceph_fsid *fsid)
 646 {
 647         if (client->have_fsid) {
 648                 if (ceph_fsid_compare(&client->fsid, fsid)) {
 649                         pr_err("bad fsid, had " FSID_FORMAT " got " FSID_FORMAT,
 650                                PR_FSID(&client->fsid), PR_FSID(fsid));
 651                         return -1;
 652                 }
 653         } else {
 654                 pr_info("client%lld fsid " FSID_FORMAT "\n",
 655                         client->monc.auth->global_id, PR_FSID(fsid));
 656                 memcpy(&client->fsid, fsid, sizeof(*fsid));
 657                 ceph_debugfs_client_init(client);
 658                 client->have_fsid = true;
 659         }
 660         return 0;
 661 }
 662
 663 /*
 664  * true if we have the mon map (and have thus joined the cluster)
 665  */
 666 static int have_mon_map(struct ceph_client *client)
 667 {
 668         return client->monc.monmap && client->monc.monmap->epoch;
 669 }
 670
 671 /*
 672  * Bootstrap mount by opening the root directory.  Note the mount
 673  * @started time from caller, and time out if this takes too long.
 674  */
 675 static struct dentry *open_root_dentry(struct ceph_client *client,
 676                                        const char *path,
 677                                        unsigned long started)
 678 {
 679         struct ceph_mds_client *mdsc = &client->mdsc;
 680         struct ceph_mds_request *req = NULL;
 681         int err;
 682         struct dentry *root;
 683
 684         /* open dir */
 685         dout("open_root_inode opening '%s'\n", path);
 686         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
 687         if (IS_ERR(req))
 688                 return ERR_PTR(PTR_ERR(req));
 689         req->r_path1 = kstrdup(path, GFP_NOFS);
 690         req->r_ino1.ino = CEPH_INO_ROOT;
 691         req->r_ino1.snap = CEPH_NOSNAP;
 692         req->r_started = started;
 693         req->r_timeout = client->mount_args->mount_timeout * HZ;
 694         req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
 695         req->r_num_caps = 2;
 696         err = ceph_mdsc_do_request(mdsc, NULL, req);
 697         if (err == 0) {
 698                 dout("open_root_inode success\n");
 699                 if (ceph_ino(req->r_target_inode) == CEPH_INO_ROOT &&
 700                     client->sb->s_root == NULL)
 701                         root = d_alloc_root(req->r_target_inode);
 702                 else
 703                         root = d_obtain_alias(req->r_target_inode);
 704                 req->r_target_inode = NULL;
 705                 dout("open_root_inode success, root dentry is %p\n", root);
 706         } else {
 707                 root = ERR_PTR(err);
 708         }
 709         ceph_mdsc_put_request(req);
 710         return root;
 711 }
 712
 713 /*
 714  * mount: join the ceph cluster, and open root directory.
 715  */
 716 static int ceph_mount(struct ceph_client *client, struct vfsmount *mnt,
 717                       const char *path)
 718 {
 719         struct ceph_entity_addr *myaddr = NULL;
 720         int err;
 721         unsigned long timeout = client->mount_args->mount_timeout * HZ;
 722         unsigned long started = jiffies;  /* note the start time */
 723         struct dentry *root;
 724
 725         dout("mount start\n");
 726         mutex_lock(&client->mount_mutex);
 727
 728         /* initialize the messenger */
 729         if (client->msgr == NULL) {
 730                 if (ceph_test_opt(client, MYIP))
 731                         myaddr = &client->mount_args->my_addr;
 732                 client->msgr = ceph_messenger_create(myaddr);
 733                 if (IS_ERR(client->msgr)) {
 734                         err = PTR_ERR(client->msgr);
 735                         client->msgr = NULL;
 736                         goto out;
 737                 }
 738                 client->msgr->nocrc = ceph_test_opt(client, NOCRC);
 739         }
 740
 741         /* open session, and wait for mon, mds, and osd maps */
 742         err = ceph_monc_open_session(&client->monc);
 743         if (err < 0)
 744                 goto out;
 745
 746         while (!have_mon_map(client)) {
 747                 err = -EIO;
 748                 if (timeout && time_after_eq(jiffies, started + timeout))
 749                         goto out;
 750
 751                 /* wait */
 752                 dout("mount waiting for mon_map\n");
 753                 err = wait_event_interruptible_timeout(client->auth_wq,
 754                                have_mon_map(client) || (client->auth_err < 0),
 755                                timeout);
 756                 if (err == -EINTR || err == -ERESTARTSYS)
 757                         goto out;
 758                 if (client->auth_err < 0) {
 759                         err = client->auth_err;
 760                         goto out;
 761                 }
 762         }
 763
 764         dout("mount opening root\n");
 765         root = open_root_dentry(client, "", started);
 766         if (IS_ERR(root)) {
 767                 err = PTR_ERR(root);
 768                 goto out;
 769         }
 770         if (client->sb->s_root)
 771                 dput(root);
 772         else
 773                 client->sb->s_root = root;
 774
 775         if (path[0] == 0) {
 776                 dget(root);
 777         } else {
 778                 dout("mount opening base mountpoint\n");
 779                 root = open_root_dentry(client, path, started);
 780                 if (IS_ERR(root)) {
 781                         err = PTR_ERR(root);
 782                         dput(client->sb->s_root);
 783                         client->sb->s_root = NULL;
 784                         goto out;
 785                 }
 786         }
 787
 788         mnt->mnt_root = root;
 789         mnt->mnt_sb = client->sb;
 790
 791         client->mount_state = CEPH_MOUNT_MOUNTED;
 792         dout("mount success\n");
 793         err = 0;
 794
 795 out:
 796         mutex_unlock(&client->mount_mutex);
 797         return err;
 798 }
 799
 800 static int ceph_set_super(struct super_block *s, void *data)
 801 {
 802         struct ceph_client *client = data;
 803         int ret;
 804
 805         dout("set_super %p data %p\n", s, data);
 806
 807         s->s_flags = client->mount_args->sb_flags;
 808         s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
 809
 810         s->s_fs_info = client;
 811         client->sb = s;
 812
 813         s->s_op = &ceph_super_ops;
 814         s->s_export_op = &ceph_export_ops;
 815
 816         s->s_time_gran = 1000;  /* 1000 ns == 1 us */
 817
 818         ret = set_anon_super(s, NULL);  /* what is that second arg for? */
 819         if (ret != 0)
 820                 goto fail;
 821
 822         return ret;
 823
 824 fail:
 825         s->s_fs_info = NULL;
 826         client->sb = NULL;
 827         return ret;
 828 }
 829
 830 /*
 831  * share superblock if same fs AND options
 832  */
 833 static int ceph_compare_super(struct super_block *sb, void *data)
 834 {
 835         struct ceph_client *new = data;
 836         struct ceph_mount_args *args = new->mount_args;
 837         struct ceph_client *other = ceph_sb_to_client(sb);
 838         int i;
 839
 840         dout("ceph_compare_super %p\n", sb);
 841         if (args->flags & CEPH_OPT_FSID) {
 842                 if (ceph_fsid_compare(&args->fsid, &other->fsid)) {
 843                         dout("fsid doesn't match\n");
 844                         return 0;
 845                 }
 846         } else {
 847                 /* do we share (a) monitor? */
 848                 for (i = 0; i < new->monc.monmap->num_mon; i++)
 849                         if (ceph_monmap_contains(other->monc.monmap,
 850                                          &new->monc.monmap->mon_inst[i].addr))
 851                                 break;
 852                 if (i == new->monc.monmap->num_mon) {
 853                         dout("mon ip not part of monmap\n");
 854                         return 0;
 855                 }
 856                 dout("mon ip matches existing sb %p\n", sb);
 857         }
 858         if (args->sb_flags != other->mount_args->sb_flags) {
 859                 dout("flags differ\n");
 860                 return 0;
 861         }
 862         return 1;
 863 }
 864
 865 /*
 866  * construct our own bdi so we can control readahead, etc.
 867  */
 868 static int ceph_register_bdi(struct super_block *sb, struct ceph_client *client)
 869 {
 870         int err;
 871
 872         sb->s_bdi = &client->backing_dev_info;
 873
 874         /* set ra_pages based on rsize mount option? */
 875         if (client->mount_args->rsize >= PAGE_CACHE_SIZE)
 876                 client->backing_dev_info.ra_pages =
 877                         (client->mount_args->rsize + PAGE_CACHE_SIZE - 1)
 878                         >> PAGE_SHIFT;
 879         err = bdi_register_dev(&client->backing_dev_info, sb->s_dev);
 880         return err;
 881 }
 882
 883 static int ceph_get_sb(struct file_system_type *fs_type,
 884                        int flags, const char *dev_name, void *data,
 885                        struct vfsmount *mnt)
 886 {
 887         struct super_block *sb;
 888         struct ceph_client *client;
 889         int err;
 890         int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
 891         const char *path = NULL;
 892         struct ceph_mount_args *args;
 893
 894         dout("ceph_get_sb\n");
 895         args = parse_mount_args(flags, data, dev_name, &path);
 896         if (IS_ERR(args)) {
 897                 err = PTR_ERR(args);
 898                 goto out_final;
 899         }
 900
 901         /* create client (which we may/may not use) */
 902         client = ceph_create_client(args);
 903         if (IS_ERR(client)) {
 904                 err = PTR_ERR(client);
 905                 goto out_final;
 906         }
 907
 908         if (client->mount_args->flags & CEPH_OPT_NOSHARE)
 909                 compare_super = NULL;
 910         sb = sget(fs_type, compare_super, ceph_set_super, client);
 911         if (IS_ERR(sb)) {
 912                 err = PTR_ERR(sb);
 913                 goto out;
 914         }
 915
 916         if (ceph_client(sb) != client) {
 917                 ceph_destroy_client(client);
 918                 client = ceph_client(sb);
 919                 dout("get_sb got existing client %p\n", client);
 920         } else {
 921                 dout("get_sb using new client %p\n", client);
 922                 err = ceph_register_bdi(sb, client);
 923                 if (err < 0)
 924                         goto out_splat;
 925         }
 926
 927         err = ceph_mount(client, mnt, path);
 928         if (err < 0)
 929                 goto out_splat;
 930         dout("root %p inode %p ino %llx.%llx\n", mnt->mnt_root,
 931              mnt->mnt_root->d_inode, ceph_vinop(mnt->mnt_root->d_inode));
 932         return 0;
 933
 934 out_splat:
 935         ceph_mdsc_close_sessions(&client->mdsc);
 936         up_write(&sb->s_umount);
 937         deactivate_super(sb);
 938         goto out_final;
 939
 940 out:
 941         ceph_destroy_client(client);
 942 out_final:
 943         dout("ceph_get_sb fail %d\n", err);
 944         return err;
 945 }
 946
 947 static void ceph_kill_sb(struct super_block *s)
 948 {
 949         struct ceph_client *client = ceph_sb_to_client(s);
 950         dout("kill_sb %p\n", s);
 951         ceph_mdsc_pre_umount(&client->mdsc);
 952         kill_anon_super(s);    /* will call put_super after sb is r/o */
 953         if (s->s_bdi == &client->backing_dev_info)
 954                 bdi_unregister(&client->backing_dev_info);
 955         bdi_destroy(&client->backing_dev_info);
 956         ceph_destroy_client(client);
 957 }
 958
 959 static struct file_system_type ceph_fs_type = {
 960         .owner          = THIS_MODULE,
 961         .name           = "ceph",
 962         .get_sb         = ceph_get_sb,
 963         .kill_sb        = ceph_kill_sb,
 964         .fs_flags       = FS_RENAME_DOES_D_MOVE,
 965 };
 966
 967 #define _STRINGIFY(x) #x
 968 #define STRINGIFY(x) _STRINGIFY(x)
 969
 970 static int __init init_ceph(void)
 971 {
 972         int ret = 0;
 973
 974         ret = ceph_debugfs_init();
 975         if (ret < 0)
 976                 goto out;
 977
 978         ret = ceph_msgr_init();
 979         if (ret < 0)
 980                 goto out_debugfs;
 981
 982         ret = init_caches();
 983         if (ret)
 984                 goto out_msgr;
 985
 986         ceph_caps_init();
 987
 988         ret = register_filesystem(&ceph_fs_type);
 989         if (ret)
 990                 goto out_icache;
 991
 992         pr_info("loaded %d.%d.%d (mon/mds/osd proto %d/%d/%d)\n",
 993                 CEPH_VERSION_MAJOR, CEPH_VERSION_MINOR, CEPH_VERSION_PATCH,
 994                 CEPH_MONC_PROTOCOL, CEPH_MDSC_PROTOCOL, CEPH_OSDC_PROTOCOL);
 995         return 0;
 996
 997 out_icache:
 998         destroy_caches();
 999 out_msgr:
1000         ceph_msgr_exit();
1001 out_debugfs:
1002         ceph_debugfs_cleanup();
1003 out:
1004         return ret;
1005 }
1006
1007 static void __exit exit_ceph(void)
1008 {
1009         dout("exit_ceph\n");
1010         unregister_filesystem(&ceph_fs_type);
1011         ceph_caps_finalize();
1012         destroy_caches();
1013         ceph_msgr_exit();
1014         ceph_debugfs_cleanup();
1015 }
1016
1017 module_init(init_ceph);
1018 module_exit(exit_ceph);
1019
1020 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1021 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1022 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1023 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1024 MODULE_LICENSE("GPL");