fs/ceph/super.c

   1
   2 #include <linux/ceph/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 "super.h"
  19 #include "mds_client.h"
  20 #include "cache.h"
  21
  22 #include <linux/ceph/ceph_features.h>
  23 #include <linux/ceph/decode.h>
  24 #include <linux/ceph/mon_client.h>
  25 #include <linux/ceph/auth.h>
  26 #include <linux/ceph/debugfs.h>
  27
  28 /*
  29  * Ceph superblock operations
  30  *
  31  * Handle the basics of mounting, unmounting.
  32  */
  33
  34 /*
  35  * super ops
  36  */
  37 static void ceph_put_super(struct super_block *s)
  38 {
  39         struct ceph_fs_client *fsc = ceph_sb_to_client(s);
  40
  41         dout("put_super\n");
  42         ceph_mdsc_close_sessions(fsc->mdsc);
  43 }
  44
  45 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
  46 {
  47         struct ceph_fs_client *fsc = ceph_inode_to_client(d_inode(dentry));
  48         struct ceph_monmap *monmap = fsc->client->monc.monmap;
  49         struct ceph_statfs st;
  50         u64 fsid;
  51         int err;
  52
  53         dout("statfs\n");
  54         err = ceph_monc_do_statfs(&fsc->client->monc, &st);
  55         if (err < 0)
  56                 return err;
  57
  58         /* fill in kstatfs */
  59         buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
  60
  61         /*
  62          * express utilization in terms of large blocks to avoid
  63          * overflow on 32-bit machines.
  64          *
  65          * NOTE: for the time being, we make bsize == frsize to humor
  66          * not-yet-ancient versions of glibc that are broken.
  67          * Someday, we will probably want to report a real block
  68          * size...  whatever that may mean for a network file system!
  69          */
  70         buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
  71         buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
  72         buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
  73         buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
  74         buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
  75
  76         buf->f_files = le64_to_cpu(st.num_objects);
  77         buf->f_ffree = -1;
  78         buf->f_namelen = NAME_MAX;
  79
  80         /* leave fsid little-endian, regardless of host endianness */
  81         fsid = *(u64 *)(&monmap->fsid) ^ *((u64 *)&monmap->fsid + 1);
  82         buf->f_fsid.val[0] = fsid & 0xffffffff;
  83         buf->f_fsid.val[1] = fsid >> 32;
  84
  85         return 0;
  86 }
  87
  88
  89 static int ceph_sync_fs(struct super_block *sb, int wait)
  90 {
  91         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
  92
  93         if (!wait) {
  94                 dout("sync_fs (non-blocking)\n");
  95                 ceph_flush_dirty_caps(fsc->mdsc);
  96                 dout("sync_fs (non-blocking) done\n");
  97                 return 0;
  98         }
  99
 100         dout("sync_fs (blocking)\n");
 101         ceph_osdc_sync(&fsc->client->osdc);
 102         ceph_mdsc_sync(fsc->mdsc);
 103         dout("sync_fs (blocking) done\n");
 104         return 0;
 105 }
 106
 107 /*
 108  * mount options
 109  */
 110 enum {
 111         Opt_mds_namespace,
 112         Opt_wsize,
 113         Opt_rsize,
 114         Opt_rasize,
 115         Opt_caps_wanted_delay_min,
 116         Opt_caps_wanted_delay_max,
 117         Opt_cap_release_safety,
 118         Opt_readdir_max_entries,
 119         Opt_readdir_max_bytes,
 120         Opt_congestion_kb,
 121         Opt_last_int,
 122         /* int args above */
 123         Opt_snapdirname,
 124         Opt_last_string,
 125         /* string args above */
 126         Opt_dirstat,
 127         Opt_nodirstat,
 128         Opt_rbytes,
 129         Opt_norbytes,
 130         Opt_asyncreaddir,
 131         Opt_noasyncreaddir,
 132         Opt_dcache,
 133         Opt_nodcache,
 134         Opt_ino32,
 135         Opt_noino32,
 136         Opt_fscache,
 137         Opt_nofscache,
 138         Opt_poolperm,
 139         Opt_nopoolperm,
 140 #ifdef CONFIG_CEPH_FS_POSIX_ACL
 141         Opt_acl,
 142 #endif
 143         Opt_noacl,
 144 };
 145
 146 static match_table_t fsopt_tokens = {
 147         {Opt_mds_namespace, "mds_namespace=%d"},
 148         {Opt_wsize, "wsize=%d"},
 149         {Opt_rsize, "rsize=%d"},
 150         {Opt_rasize, "rasize=%d"},
 151         {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
 152         {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
 153         {Opt_cap_release_safety, "cap_release_safety=%d"},
 154         {Opt_readdir_max_entries, "readdir_max_entries=%d"},
 155         {Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
 156         {Opt_congestion_kb, "write_congestion_kb=%d"},
 157         /* int args above */
 158         {Opt_snapdirname, "snapdirname=%s"},
 159         /* string args above */
 160         {Opt_dirstat, "dirstat"},
 161         {Opt_nodirstat, "nodirstat"},
 162         {Opt_rbytes, "rbytes"},
 163         {Opt_norbytes, "norbytes"},
 164         {Opt_asyncreaddir, "asyncreaddir"},
 165         {Opt_noasyncreaddir, "noasyncreaddir"},
 166         {Opt_dcache, "dcache"},
 167         {Opt_nodcache, "nodcache"},
 168         {Opt_ino32, "ino32"},
 169         {Opt_noino32, "noino32"},
 170         {Opt_fscache, "fsc"},
 171         {Opt_nofscache, "nofsc"},
 172         {Opt_poolperm, "poolperm"},
 173         {Opt_nopoolperm, "nopoolperm"},
 174 #ifdef CONFIG_CEPH_FS_POSIX_ACL
 175         {Opt_acl, "acl"},
 176 #endif
 177         {Opt_noacl, "noacl"},
 178         {-1, NULL}
 179 };
 180
 181 static int parse_fsopt_token(char *c, void *private)
 182 {
 183         struct ceph_mount_options *fsopt = private;
 184         substring_t argstr[MAX_OPT_ARGS];
 185         int token, intval, ret;
 186
 187         token = match_token((char *)c, fsopt_tokens, argstr);
 188         if (token < 0)
 189                 return -EINVAL;
 190
 191         if (token < Opt_last_int) {
 192                 ret = match_int(&argstr[0], &intval);
 193                 if (ret < 0) {
 194                         pr_err("bad mount option arg (not int) "
 195                                "at '%s'\n", c);
 196                         return ret;
 197                 }
 198                 dout("got int token %d val %d\n", token, intval);
 199         } else if (token > Opt_last_int && token < Opt_last_string) {
 200                 dout("got string token %d val %s\n", token,
 201                      argstr[0].from);
 202         } else {
 203                 dout("got token %d\n", token);
 204         }
 205
 206         switch (token) {
 207         case Opt_snapdirname:
 208                 kfree(fsopt->snapdir_name);
 209                 fsopt->snapdir_name = kstrndup(argstr[0].from,
 210                                                argstr[0].to-argstr[0].from,
 211                                                GFP_KERNEL);
 212                 if (!fsopt->snapdir_name)
 213                         return -ENOMEM;
 214                 break;
 215
 216                 /* misc */
 217         case Opt_mds_namespace:
 218                 fsopt->mds_namespace = intval;
 219                 break;
 220         case Opt_wsize:
 221                 fsopt->wsize = intval;
 222                 break;
 223         case Opt_rsize:
 224                 fsopt->rsize = intval;
 225                 break;
 226         case Opt_rasize:
 227                 fsopt->rasize = intval;
 228                 break;
 229         case Opt_caps_wanted_delay_min:
 230                 fsopt->caps_wanted_delay_min = intval;
 231                 break;
 232         case Opt_caps_wanted_delay_max:
 233                 fsopt->caps_wanted_delay_max = intval;
 234                 break;
 235         case Opt_readdir_max_entries:
 236                 fsopt->max_readdir = intval;
 237                 break;
 238         case Opt_readdir_max_bytes:
 239                 fsopt->max_readdir_bytes = intval;
 240                 break;
 241         case Opt_congestion_kb:
 242                 fsopt->congestion_kb = intval;
 243                 break;
 244         case Opt_dirstat:
 245                 fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
 246                 break;
 247         case Opt_nodirstat:
 248                 fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
 249                 break;
 250         case Opt_rbytes:
 251                 fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
 252                 break;
 253         case Opt_norbytes:
 254                 fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
 255                 break;
 256         case Opt_asyncreaddir:
 257                 fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
 258                 break;
 259         case Opt_noasyncreaddir:
 260                 fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
 261                 break;
 262         case Opt_dcache:
 263                 fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
 264                 break;
 265         case Opt_nodcache:
 266                 fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
 267                 break;
 268         case Opt_ino32:
 269                 fsopt->flags |= CEPH_MOUNT_OPT_INO32;
 270                 break;
 271         case Opt_noino32:
 272                 fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
 273                 break;
 274         case Opt_fscache:
 275                 fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
 276                 break;
 277         case Opt_nofscache:
 278                 fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
 279                 break;
 280         case Opt_poolperm:
 281                 fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
 282                 printk ("pool perm");
 283                 break;
 284         case Opt_nopoolperm:
 285                 fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
 286                 break;
 287 #ifdef CONFIG_CEPH_FS_POSIX_ACL
 288         case Opt_acl:
 289                 fsopt->sb_flags |= MS_POSIXACL;
 290                 break;
 291 #endif
 292         case Opt_noacl:
 293                 fsopt->sb_flags &= ~MS_POSIXACL;
 294                 break;
 295         default:
 296                 BUG_ON(token);
 297         }
 298         return 0;
 299 }
 300
 301 static void destroy_mount_options(struct ceph_mount_options *args)
 302 {
 303         dout("destroy_mount_options %p\n", args);
 304         kfree(args->snapdir_name);
 305         kfree(args->server_path);
 306         kfree(args);
 307 }
 308
 309 static int strcmp_null(const char *s1, const char *s2)
 310 {
 311         if (!s1 && !s2)
 312                 return 0;
 313         if (s1 && !s2)
 314                 return -1;
 315         if (!s1 && s2)
 316                 return 1;
 317         return strcmp(s1, s2);
 318 }
 319
 320 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
 321                                  struct ceph_options *new_opt,
 322                                  struct ceph_fs_client *fsc)
 323 {
 324         struct ceph_mount_options *fsopt1 = new_fsopt;
 325         struct ceph_mount_options *fsopt2 = fsc->mount_options;
 326         int ofs = offsetof(struct ceph_mount_options, snapdir_name);
 327         int ret;
 328
 329         ret = memcmp(fsopt1, fsopt2, ofs);
 330         if (ret)
 331                 return ret;
 332
 333         ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
 334         if (ret)
 335                 return ret;
 336
 337         ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
 338         if (ret)
 339                 return ret;
 340
 341         return ceph_compare_options(new_opt, fsc->client);
 342 }
 343
 344 static int parse_mount_options(struct ceph_mount_options **pfsopt,
 345                                struct ceph_options **popt,
 346                                int flags, char *options,
 347                                const char *dev_name)
 348 {
 349         struct ceph_mount_options *fsopt;
 350         const char *dev_name_end;
 351         int err;
 352
 353         if (!dev_name || !*dev_name)
 354                 return -EINVAL;
 355
 356         fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
 357         if (!fsopt)
 358                 return -ENOMEM;
 359
 360         dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
 361
 362         fsopt->sb_flags = flags;
 363         fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
 364
 365         fsopt->rsize = CEPH_RSIZE_DEFAULT;
 366         fsopt->rasize = CEPH_RASIZE_DEFAULT;
 367         fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
 368         if (!fsopt->snapdir_name) {
 369                 err = -ENOMEM;
 370                 goto out;
 371         }
 372
 373         fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
 374         fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
 375         fsopt->cap_release_safety = CEPH_CAP_RELEASE_SAFETY_DEFAULT;
 376         fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
 377         fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
 378         fsopt->congestion_kb = default_congestion_kb();
 379         fsopt->mds_namespace = CEPH_FS_CLUSTER_ID_NONE;
 380
 381         /*
 382          * Distinguish the server list from the path in "dev_name".
 383          * Internally we do not include the leading '/' in the path.
 384          *
 385          * "dev_name" will look like:
 386          *     <server_spec>[,<server_spec>...]:[<path>]
 387          * where
 388          *     <server_spec> is <ip>[:<port>]
 389          *     <path> is optional, but if present must begin with '/'
 390          */
 391         dev_name_end = strchr(dev_name, '/');
 392         if (dev_name_end) {
 393                 fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
 394                 if (!fsopt->server_path) {
 395                         err = -ENOMEM;
 396                         goto out;
 397                 }
 398         } else {
 399                 dev_name_end = dev_name + strlen(dev_name);
 400         }
 401         err = -EINVAL;
 402         dev_name_end--;         /* back up to ':' separator */
 403         if (dev_name_end < dev_name || *dev_name_end != ':') {
 404                 pr_err("device name is missing path (no : separator in %s)\n",
 405                                 dev_name);
 406                 goto out;
 407         }
 408         dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
 409         if (fsopt->server_path)
 410                 dout("server path '%s'\n", fsopt->server_path);
 411
 412         *popt = ceph_parse_options(options, dev_name, dev_name_end,
 413                                  parse_fsopt_token, (void *)fsopt);
 414         if (IS_ERR(*popt)) {
 415                 err = PTR_ERR(*popt);
 416                 goto out;
 417         }
 418
 419         /* success */
 420         *pfsopt = fsopt;
 421         return 0;
 422
 423 out:
 424         destroy_mount_options(fsopt);
 425         return err;
 426 }
 427
 428 /**
 429  * ceph_show_options - Show mount options in /proc/mounts
 430  * @m: seq_file to write to
 431  * @root: root of that (sub)tree
 432  */
 433 static int ceph_show_options(struct seq_file *m, struct dentry *root)
 434 {
 435         struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
 436         struct ceph_mount_options *fsopt = fsc->mount_options;
 437         size_t pos;
 438         int ret;
 439
 440         /* a comma between MNT/MS and client options */
 441         seq_putc(m, ',');
 442         pos = m->count;
 443
 444         ret = ceph_print_client_options(m, fsc->client);
 445         if (ret)
 446                 return ret;
 447
 448         /* retract our comma if no client options */
 449         if (m->count == pos)
 450                 m->count--;
 451
 452         if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
 453                 seq_puts(m, ",dirstat");
 454         if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
 455                 seq_puts(m, ",rbytes");
 456         if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
 457                 seq_puts(m, ",noasyncreaddir");
 458         if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
 459                 seq_puts(m, ",nodcache");
 460         if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE)
 461                 seq_puts(m, ",fsc");
 462         if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
 463                 seq_puts(m, ",nopoolperm");
 464
 465 #ifdef CONFIG_CEPH_FS_POSIX_ACL
 466         if (fsopt->sb_flags & MS_POSIXACL)
 467                 seq_puts(m, ",acl");
 468         else
 469                 seq_puts(m, ",noacl");
 470 #endif
 471
 472         if (fsopt->mds_namespace != CEPH_FS_CLUSTER_ID_NONE)
 473                 seq_printf(m, ",mds_namespace=%d", fsopt->mds_namespace);
 474         if (fsopt->wsize)
 475                 seq_printf(m, ",wsize=%d", fsopt->wsize);
 476         if (fsopt->rsize != CEPH_RSIZE_DEFAULT)
 477                 seq_printf(m, ",rsize=%d", fsopt->rsize);
 478         if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
 479                 seq_printf(m, ",rasize=%d", fsopt->rasize);
 480         if (fsopt->congestion_kb != default_congestion_kb())
 481                 seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
 482         if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
 483                 seq_printf(m, ",caps_wanted_delay_min=%d",
 484                          fsopt->caps_wanted_delay_min);
 485         if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
 486                 seq_printf(m, ",caps_wanted_delay_max=%d",
 487                            fsopt->caps_wanted_delay_max);
 488         if (fsopt->cap_release_safety != CEPH_CAP_RELEASE_SAFETY_DEFAULT)
 489                 seq_printf(m, ",cap_release_safety=%d",
 490                            fsopt->cap_release_safety);
 491         if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
 492                 seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
 493         if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
 494                 seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
 495         if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
 496                 seq_show_option(m, "snapdirname", fsopt->snapdir_name);
 497
 498         return 0;
 499 }
 500
 501 /*
 502  * handle any mon messages the standard library doesn't understand.
 503  * return error if we don't either.
 504  */
 505 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
 506 {
 507         struct ceph_fs_client *fsc = client->private;
 508         int type = le16_to_cpu(msg->hdr.type);
 509
 510         switch (type) {
 511         case CEPH_MSG_MDS_MAP:
 512                 ceph_mdsc_handle_map(fsc->mdsc, msg);
 513                 return 0;
 514
 515         default:
 516                 return -1;
 517         }
 518 }
 519
 520 /*
 521  * create a new fs client
 522  */
 523 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
 524                                         struct ceph_options *opt)
 525 {
 526         struct ceph_fs_client *fsc;
 527         const u64 supported_features =
 528                 CEPH_FEATURE_FLOCK | CEPH_FEATURE_DIRLAYOUTHASH |
 529                 CEPH_FEATURE_MDSENC | CEPH_FEATURE_MDS_INLINE_DATA;
 530         const u64 required_features = 0;
 531         int page_count;
 532         size_t size;
 533         int err = -ENOMEM;
 534
 535         fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
 536         if (!fsc)
 537                 return ERR_PTR(-ENOMEM);
 538
 539         fsc->client = ceph_create_client(opt, fsc, supported_features,
 540                                          required_features);
 541         if (IS_ERR(fsc->client)) {
 542                 err = PTR_ERR(fsc->client);
 543                 goto fail;
 544         }
 545         fsc->client->extra_mon_dispatch = extra_mon_dispatch;
 546         fsc->client->monc.fs_cluster_id = fsopt->mds_namespace;
 547         ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP, 0, true);
 548
 549         fsc->mount_options = fsopt;
 550
 551         fsc->sb = NULL;
 552         fsc->mount_state = CEPH_MOUNT_MOUNTING;
 553
 554         atomic_long_set(&fsc->writeback_count, 0);
 555
 556         err = bdi_init(&fsc->backing_dev_info);
 557         if (err < 0)
 558                 goto fail_client;
 559
 560         err = -ENOMEM;
 561         /*
 562          * The number of concurrent works can be high but they don't need
 563          * to be processed in parallel, limit concurrency.
 564          */
 565         fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
 566         if (fsc->wb_wq == NULL)
 567                 goto fail_bdi;
 568         fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
 569         if (fsc->pg_inv_wq == NULL)
 570                 goto fail_wb_wq;
 571         fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
 572         if (fsc->trunc_wq == NULL)
 573                 goto fail_pg_inv_wq;
 574
 575         /* set up mempools */
 576         err = -ENOMEM;
 577         page_count = fsc->mount_options->wsize >> PAGE_SHIFT;
 578         size = sizeof (struct page *) * (page_count ? page_count : 1);
 579         fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
 580         if (!fsc->wb_pagevec_pool)
 581                 goto fail_trunc_wq;
 582
 583         /* setup fscache */
 584         if ((fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) &&
 585             (ceph_fscache_register_fs(fsc) != 0))
 586                 goto fail_fscache;
 587
 588         /* caps */
 589         fsc->min_caps = fsopt->max_readdir;
 590
 591         return fsc;
 592
 593 fail_fscache:
 594         ceph_fscache_unregister_fs(fsc);
 595 fail_trunc_wq:
 596         destroy_workqueue(fsc->trunc_wq);
 597 fail_pg_inv_wq:
 598         destroy_workqueue(fsc->pg_inv_wq);
 599 fail_wb_wq:
 600         destroy_workqueue(fsc->wb_wq);
 601 fail_bdi:
 602         bdi_destroy(&fsc->backing_dev_info);
 603 fail_client:
 604         ceph_destroy_client(fsc->client);
 605 fail:
 606         kfree(fsc);
 607         return ERR_PTR(err);
 608 }
 609
 610 static void destroy_fs_client(struct ceph_fs_client *fsc)
 611 {
 612         dout("destroy_fs_client %p\n", fsc);
 613
 614         ceph_fscache_unregister_fs(fsc);
 615
 616         destroy_workqueue(fsc->wb_wq);
 617         destroy_workqueue(fsc->pg_inv_wq);
 618         destroy_workqueue(fsc->trunc_wq);
 619
 620         bdi_destroy(&fsc->backing_dev_info);
 621
 622         mempool_destroy(fsc->wb_pagevec_pool);
 623
 624         destroy_mount_options(fsc->mount_options);
 625
 626         ceph_fs_debugfs_cleanup(fsc);
 627
 628         ceph_destroy_client(fsc->client);
 629
 630         kfree(fsc);
 631         dout("destroy_fs_client %p done\n", fsc);
 632 }
 633
 634 /*
 635  * caches
 636  */
 637 struct kmem_cache *ceph_inode_cachep;
 638 struct kmem_cache *ceph_cap_cachep;
 639 struct kmem_cache *ceph_cap_flush_cachep;
 640 struct kmem_cache *ceph_dentry_cachep;
 641 struct kmem_cache *ceph_file_cachep;
 642
 643 static void ceph_inode_init_once(void *foo)
 644 {
 645         struct ceph_inode_info *ci = foo;
 646         inode_init_once(&ci->vfs_inode);
 647 }
 648
 649 static int __init init_caches(void)
 650 {
 651         int error = -ENOMEM;
 652
 653         ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
 654                                       sizeof(struct ceph_inode_info),
 655                                       __alignof__(struct ceph_inode_info),
 656                                       SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
 657                                       SLAB_ACCOUNT, ceph_inode_init_once);
 658         if (ceph_inode_cachep == NULL)
 659                 return -ENOMEM;
 660
 661         ceph_cap_cachep = KMEM_CACHE(ceph_cap,
 662                                      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 663         if (ceph_cap_cachep == NULL)
 664                 goto bad_cap;
 665         ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
 666                                            SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 667         if (ceph_cap_flush_cachep == NULL)
 668                 goto bad_cap_flush;
 669
 670         ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
 671                                         SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 672         if (ceph_dentry_cachep == NULL)
 673                 goto bad_dentry;
 674
 675         ceph_file_cachep = KMEM_CACHE(ceph_file_info,
 676                                       SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 677         if (ceph_file_cachep == NULL)
 678                 goto bad_file;
 679
 680         if ((error = ceph_fscache_register()))
 681                 goto bad_file;
 682
 683         return 0;
 684 bad_file:
 685         kmem_cache_destroy(ceph_dentry_cachep);
 686 bad_dentry:
 687         kmem_cache_destroy(ceph_cap_flush_cachep);
 688 bad_cap_flush:
 689         kmem_cache_destroy(ceph_cap_cachep);
 690 bad_cap:
 691         kmem_cache_destroy(ceph_inode_cachep);
 692         return error;
 693 }
 694
 695 static void destroy_caches(void)
 696 {
 697         /*
 698          * Make sure all delayed rcu free inodes are flushed before we
 699          * destroy cache.
 700          */
 701         rcu_barrier();
 702
 703         kmem_cache_destroy(ceph_inode_cachep);
 704         kmem_cache_destroy(ceph_cap_cachep);
 705         kmem_cache_destroy(ceph_cap_flush_cachep);
 706         kmem_cache_destroy(ceph_dentry_cachep);
 707         kmem_cache_destroy(ceph_file_cachep);
 708
 709         ceph_fscache_unregister();
 710 }
 711
 712
 713 /*
 714  * ceph_umount_begin - initiate forced umount.  Tear down down the
 715  * mount, skipping steps that may hang while waiting for server(s).
 716  */
 717 static void ceph_umount_begin(struct super_block *sb)
 718 {
 719         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
 720
 721         dout("ceph_umount_begin - starting forced umount\n");
 722         if (!fsc)
 723                 return;
 724         fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
 725         ceph_mdsc_force_umount(fsc->mdsc);
 726         return;
 727 }
 728
 729 static const struct super_operations ceph_super_ops = {
 730         .alloc_inode    = ceph_alloc_inode,
 731         .destroy_inode  = ceph_destroy_inode,
 732         .write_inode    = ceph_write_inode,
 733         .drop_inode     = ceph_drop_inode,
 734         .sync_fs        = ceph_sync_fs,
 735         .put_super      = ceph_put_super,
 736         .show_options   = ceph_show_options,
 737         .statfs         = ceph_statfs,
 738         .umount_begin   = ceph_umount_begin,
 739 };
 740
 741 /*
 742  * Bootstrap mount by opening the root directory.  Note the mount
 743  * @started time from caller, and time out if this takes too long.
 744  */
 745 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
 746                                        const char *path,
 747                                        unsigned long started)
 748 {
 749         struct ceph_mds_client *mdsc = fsc->mdsc;
 750         struct ceph_mds_request *req = NULL;
 751         int err;
 752         struct dentry *root;
 753
 754         /* open dir */
 755         dout("open_root_inode opening '%s'\n", path);
 756         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
 757         if (IS_ERR(req))
 758                 return ERR_CAST(req);
 759         req->r_path1 = kstrdup(path, GFP_NOFS);
 760         if (!req->r_path1) {
 761                 root = ERR_PTR(-ENOMEM);
 762                 goto out;
 763         }
 764
 765         req->r_ino1.ino = CEPH_INO_ROOT;
 766         req->r_ino1.snap = CEPH_NOSNAP;
 767         req->r_started = started;
 768         req->r_timeout = fsc->client->options->mount_timeout;
 769         req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
 770         req->r_num_caps = 2;
 771         err = ceph_mdsc_do_request(mdsc, NULL, req);
 772         if (err == 0) {
 773                 struct inode *inode = req->r_target_inode;
 774                 req->r_target_inode = NULL;
 775                 dout("open_root_inode success\n");
 776                 if (ceph_ino(inode) == CEPH_INO_ROOT &&
 777                     fsc->sb->s_root == NULL) {
 778                         root = d_make_root(inode);
 779                         if (!root) {
 780                                 root = ERR_PTR(-ENOMEM);
 781                                 goto out;
 782                         }
 783                 } else {
 784                         root = d_obtain_root(inode);
 785                 }
 786                 ceph_init_dentry(root);
 787                 dout("open_root_inode success, root dentry is %p\n", root);
 788         } else {
 789                 root = ERR_PTR(err);
 790         }
 791 out:
 792         ceph_mdsc_put_request(req);
 793         return root;
 794 }
 795
 796
 797
 798
 799 /*
 800  * mount: join the ceph cluster, and open root directory.
 801  */
 802 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc)
 803 {
 804         int err;
 805         unsigned long started = jiffies;  /* note the start time */
 806         struct dentry *root;
 807         int first = 0;   /* first vfsmount for this super_block */
 808
 809         dout("mount start %p\n", fsc);
 810         mutex_lock(&fsc->client->mount_mutex);
 811
 812         if (!fsc->sb->s_root) {
 813                 err = __ceph_open_session(fsc->client, started);
 814                 if (err < 0)
 815                         goto out;
 816
 817                 dout("mount opening root\n");
 818                 root = open_root_dentry(fsc, "", started);
 819                 if (IS_ERR(root)) {
 820                         err = PTR_ERR(root);
 821                         goto out;
 822                 }
 823                 fsc->sb->s_root = root;
 824                 first = 1;
 825
 826                 err = ceph_fs_debugfs_init(fsc);
 827                 if (err < 0)
 828                         goto fail;
 829         }
 830
 831         if (!fsc->mount_options->server_path) {
 832                 root = fsc->sb->s_root;
 833                 dget(root);
 834         } else {
 835                 const char *path = fsc->mount_options->server_path + 1;
 836                 dout("mount opening path %s\n", path);
 837                 root = open_root_dentry(fsc, path, started);
 838                 if (IS_ERR(root)) {
 839                         err = PTR_ERR(root);
 840                         goto fail;
 841                 }
 842         }
 843
 844         fsc->mount_state = CEPH_MOUNT_MOUNTED;
 845         dout("mount success\n");
 846         mutex_unlock(&fsc->client->mount_mutex);
 847         return root;
 848
 849 fail:
 850         if (first) {
 851                 dput(fsc->sb->s_root);
 852                 fsc->sb->s_root = NULL;
 853         }
 854 out:
 855         mutex_unlock(&fsc->client->mount_mutex);
 856         return ERR_PTR(err);
 857 }
 858
 859 static int ceph_set_super(struct super_block *s, void *data)
 860 {
 861         struct ceph_fs_client *fsc = data;
 862         int ret;
 863
 864         dout("set_super %p data %p\n", s, data);
 865
 866         s->s_flags = fsc->mount_options->sb_flags;
 867         s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
 868
 869         s->s_xattr = ceph_xattr_handlers;
 870         s->s_fs_info = fsc;
 871         fsc->sb = s;
 872
 873         s->s_op = &ceph_super_ops;
 874         s->s_export_op = &ceph_export_ops;
 875
 876         s->s_time_gran = 1000;  /* 1000 ns == 1 us */
 877
 878         ret = set_anon_super(s, NULL);  /* what is that second arg for? */
 879         if (ret != 0)
 880                 goto fail;
 881
 882         return ret;
 883
 884 fail:
 885         s->s_fs_info = NULL;
 886         fsc->sb = NULL;
 887         return ret;
 888 }
 889
 890 /*
 891  * share superblock if same fs AND options
 892  */
 893 static int ceph_compare_super(struct super_block *sb, void *data)
 894 {
 895         struct ceph_fs_client *new = data;
 896         struct ceph_mount_options *fsopt = new->mount_options;
 897         struct ceph_options *opt = new->client->options;
 898         struct ceph_fs_client *other = ceph_sb_to_client(sb);
 899
 900         dout("ceph_compare_super %p\n", sb);
 901
 902         if (compare_mount_options(fsopt, opt, other)) {
 903                 dout("monitor(s)/mount options don't match\n");
 904                 return 0;
 905         }
 906         if ((opt->flags & CEPH_OPT_FSID) &&
 907             ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
 908                 dout("fsid doesn't match\n");
 909                 return 0;
 910         }
 911         if (fsopt->sb_flags != other->mount_options->sb_flags) {
 912                 dout("flags differ\n");
 913                 return 0;
 914         }
 915         return 1;
 916 }
 917
 918 /*
 919  * construct our own bdi so we can control readahead, etc.
 920  */
 921 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
 922
 923 static int ceph_register_bdi(struct super_block *sb,
 924                              struct ceph_fs_client *fsc)
 925 {
 926         int err;
 927
 928         /* set ra_pages based on rasize mount option? */
 929         if (fsc->mount_options->rasize >= PAGE_SIZE)
 930                 fsc->backing_dev_info.ra_pages =
 931                         (fsc->mount_options->rasize + PAGE_SIZE - 1)
 932                         >> PAGE_SHIFT;
 933         else
 934                 fsc->backing_dev_info.ra_pages =
 935                         VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
 936
 937         err = bdi_register(&fsc->backing_dev_info, NULL, "ceph-%ld",
 938                            atomic_long_inc_return(&bdi_seq));
 939         if (!err)
 940                 sb->s_bdi = &fsc->backing_dev_info;
 941         return err;
 942 }
 943
 944 static struct dentry *ceph_mount(struct file_system_type *fs_type,
 945                        int flags, const char *dev_name, void *data)
 946 {
 947         struct super_block *sb;
 948         struct ceph_fs_client *fsc;
 949         struct dentry *res;
 950         int err;
 951         int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
 952         struct ceph_mount_options *fsopt = NULL;
 953         struct ceph_options *opt = NULL;
 954
 955         dout("ceph_mount\n");
 956
 957 #ifdef CONFIG_CEPH_FS_POSIX_ACL
 958         flags |= MS_POSIXACL;
 959 #endif
 960         err = parse_mount_options(&fsopt, &opt, flags, data, dev_name);
 961         if (err < 0) {
 962                 res = ERR_PTR(err);
 963                 goto out_final;
 964         }
 965
 966         /* create client (which we may/may not use) */
 967         fsc = create_fs_client(fsopt, opt);
 968         if (IS_ERR(fsc)) {
 969                 res = ERR_CAST(fsc);
 970                 destroy_mount_options(fsopt);
 971                 ceph_destroy_options(opt);
 972                 goto out_final;
 973         }
 974
 975         err = ceph_mdsc_init(fsc);
 976         if (err < 0) {
 977                 res = ERR_PTR(err);
 978                 goto out;
 979         }
 980
 981         if (ceph_test_opt(fsc->client, NOSHARE))
 982                 compare_super = NULL;
 983         sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc);
 984         if (IS_ERR(sb)) {
 985                 res = ERR_CAST(sb);
 986                 goto out;
 987         }
 988
 989         if (ceph_sb_to_client(sb) != fsc) {
 990                 ceph_mdsc_destroy(fsc);
 991                 destroy_fs_client(fsc);
 992                 fsc = ceph_sb_to_client(sb);
 993                 dout("get_sb got existing client %p\n", fsc);
 994         } else {
 995                 dout("get_sb using new client %p\n", fsc);
 996                 err = ceph_register_bdi(sb, fsc);
 997                 if (err < 0) {
 998                         res = ERR_PTR(err);
 999                         goto out_splat;
1000                 }
1001         }
1002
1003         res = ceph_real_mount(fsc);
1004         if (IS_ERR(res))
1005                 goto out_splat;
1006         dout("root %p inode %p ino %llx.%llx\n", res,
1007              d_inode(res), ceph_vinop(d_inode(res)));
1008         return res;
1009
1010 out_splat:
1011         ceph_mdsc_close_sessions(fsc->mdsc);
1012         deactivate_locked_super(sb);
1013         goto out_final;
1014
1015 out:
1016         ceph_mdsc_destroy(fsc);
1017         destroy_fs_client(fsc);
1018 out_final:
1019         dout("ceph_mount fail %ld\n", PTR_ERR(res));
1020         return res;
1021 }
1022
1023 static void ceph_kill_sb(struct super_block *s)
1024 {
1025         struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1026         dev_t dev = s->s_dev;
1027
1028         dout("kill_sb %p\n", s);
1029
1030         ceph_mdsc_pre_umount(fsc->mdsc);
1031         generic_shutdown_super(s);
1032         ceph_mdsc_destroy(fsc);
1033
1034         destroy_fs_client(fsc);
1035         free_anon_bdev(dev);
1036 }
1037
1038 static struct file_system_type ceph_fs_type = {
1039         .owner          = THIS_MODULE,
1040         .name           = "ceph",
1041         .mount          = ceph_mount,
1042         .kill_sb        = ceph_kill_sb,
1043         .fs_flags       = FS_RENAME_DOES_D_MOVE,
1044 };
1045 MODULE_ALIAS_FS("ceph");
1046
1047 static int __init init_ceph(void)
1048 {
1049         int ret = init_caches();
1050         if (ret)
1051                 goto out;
1052
1053         ceph_flock_init();
1054         ceph_xattr_init();
1055         ret = register_filesystem(&ceph_fs_type);
1056         if (ret)
1057                 goto out_xattr;
1058
1059         pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1060
1061         return 0;
1062
1063 out_xattr:
1064         ceph_xattr_exit();
1065         destroy_caches();
1066 out:
1067         return ret;
1068 }
1069
1070 static void __exit exit_ceph(void)
1071 {
1072         dout("exit_ceph\n");
1073         unregister_filesystem(&ceph_fs_type);
1074         ceph_xattr_exit();
1075         destroy_caches();
1076 }
1077
1078 module_init(init_ceph);
1079 module_exit(exit_ceph);
1080
1081 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1082 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1083 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1084 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1085 MODULE_LICENSE("GPL");