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_mon_client *monc = &fsc->client->monc;
  49         struct ceph_statfs st;
  50         u64 fsid;
  51         int err;
  52         u64 data_pool;
  53
  54         if (fsc->mdsc->mdsmap->m_num_data_pg_pools == 1) {
  55                 data_pool = fsc->mdsc->mdsmap->m_data_pg_pools[0];
  56         } else {
  57                 data_pool = CEPH_NOPOOL;
  58         }
  59
  60         dout("statfs\n");
  61         err = ceph_monc_do_statfs(monc, data_pool, &st);
  62         if (err < 0)
  63                 return err;
  64
  65         /* fill in kstatfs */
  66         buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
  67
  68         /*
  69          * express utilization in terms of large blocks to avoid
  70          * overflow on 32-bit machines.
  71          *
  72          * NOTE: for the time being, we make bsize == frsize to humor
  73          * not-yet-ancient versions of glibc that are broken.
  74          * Someday, we will probably want to report a real block
  75          * size...  whatever that may mean for a network file system!
  76          */
  77         buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
  78         buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
  79         buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
  80         buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
  81         buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
  82
  83         buf->f_files = le64_to_cpu(st.num_objects);
  84         buf->f_ffree = -1;
  85         buf->f_namelen = NAME_MAX;
  86
  87         /* Must convert the fsid, for consistent values across arches */
  88         mutex_lock(&monc->mutex);
  89         fsid = le64_to_cpu(*(__le64 *)(&monc->monmap->fsid)) ^
  90                le64_to_cpu(*((__le64 *)&monc->monmap->fsid + 1));
  91         mutex_unlock(&monc->mutex);
  92
  93         buf->f_fsid.val[0] = fsid & 0xffffffff;
  94         buf->f_fsid.val[1] = fsid >> 32;
  95
  96         return 0;
  97 }
  98
  99
 100 static int ceph_sync_fs(struct super_block *sb, int wait)
 101 {
 102         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
 103
 104         if (!wait) {
 105                 dout("sync_fs (non-blocking)\n");
 106                 ceph_flush_dirty_caps(fsc->mdsc);
 107                 dout("sync_fs (non-blocking) done\n");
 108                 return 0;
 109         }
 110
 111         dout("sync_fs (blocking)\n");
 112         ceph_osdc_sync(&fsc->client->osdc);
 113         ceph_mdsc_sync(fsc->mdsc);
 114         dout("sync_fs (blocking) done\n");
 115         return 0;
 116 }
 117
 118 /*
 119  * mount options
 120  */
 121 enum {
 122         Opt_wsize,
 123         Opt_rsize,
 124         Opt_rasize,
 125         Opt_caps_wanted_delay_min,
 126         Opt_caps_wanted_delay_max,
 127         Opt_readdir_max_entries,
 128         Opt_readdir_max_bytes,
 129         Opt_congestion_kb,
 130         Opt_last_int,
 131         /* int args above */
 132         Opt_snapdirname,
 133         Opt_mds_namespace,
 134         Opt_fscache_uniq,
 135         Opt_last_string,
 136         /* string args above */
 137         Opt_dirstat,
 138         Opt_nodirstat,
 139         Opt_rbytes,
 140         Opt_norbytes,
 141         Opt_asyncreaddir,
 142         Opt_noasyncreaddir,
 143         Opt_dcache,
 144         Opt_nodcache,
 145         Opt_ino32,
 146         Opt_noino32,
 147         Opt_fscache,
 148         Opt_nofscache,
 149         Opt_poolperm,
 150         Opt_nopoolperm,
 151         Opt_require_active_mds,
 152         Opt_norequire_active_mds,
 153 #ifdef CONFIG_CEPH_FS_POSIX_ACL
 154         Opt_acl,
 155 #endif
 156         Opt_noacl,
 157 };
 158
 159 static match_table_t fsopt_tokens = {
 160         {Opt_wsize, "wsize=%d"},
 161         {Opt_rsize, "rsize=%d"},
 162         {Opt_rasize, "rasize=%d"},
 163         {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
 164         {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
 165         {Opt_readdir_max_entries, "readdir_max_entries=%d"},
 166         {Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
 167         {Opt_congestion_kb, "write_congestion_kb=%d"},
 168         /* int args above */
 169         {Opt_snapdirname, "snapdirname=%s"},
 170         {Opt_mds_namespace, "mds_namespace=%s"},
 171         {Opt_fscache_uniq, "fsc=%s"},
 172         /* string args above */
 173         {Opt_dirstat, "dirstat"},
 174         {Opt_nodirstat, "nodirstat"},
 175         {Opt_rbytes, "rbytes"},
 176         {Opt_norbytes, "norbytes"},
 177         {Opt_asyncreaddir, "asyncreaddir"},
 178         {Opt_noasyncreaddir, "noasyncreaddir"},
 179         {Opt_dcache, "dcache"},
 180         {Opt_nodcache, "nodcache"},
 181         {Opt_ino32, "ino32"},
 182         {Opt_noino32, "noino32"},
 183         {Opt_fscache, "fsc"},
 184         {Opt_nofscache, "nofsc"},
 185         {Opt_poolperm, "poolperm"},
 186         {Opt_nopoolperm, "nopoolperm"},
 187         {Opt_require_active_mds, "require_active_mds"},
 188         {Opt_norequire_active_mds, "norequire_active_mds"},
 189 #ifdef CONFIG_CEPH_FS_POSIX_ACL
 190         {Opt_acl, "acl"},
 191 #endif
 192         {Opt_noacl, "noacl"},
 193         {-1, NULL}
 194 };
 195
 196 static int parse_fsopt_token(char *c, void *private)
 197 {
 198         struct ceph_mount_options *fsopt = private;
 199         substring_t argstr[MAX_OPT_ARGS];
 200         int token, intval, ret;
 201
 202         token = match_token((char *)c, fsopt_tokens, argstr);
 203         if (token < 0)
 204                 return -EINVAL;
 205
 206         if (token < Opt_last_int) {
 207                 ret = match_int(&argstr[0], &intval);
 208                 if (ret < 0) {
 209                         pr_err("bad mount option arg (not int) "
 210                                "at '%s'\n", c);
 211                         return ret;
 212                 }
 213                 dout("got int token %d val %d\n", token, intval);
 214         } else if (token > Opt_last_int && token < Opt_last_string) {
 215                 dout("got string token %d val %s\n", token,
 216                      argstr[0].from);
 217         } else {
 218                 dout("got token %d\n", token);
 219         }
 220
 221         switch (token) {
 222         case Opt_snapdirname:
 223                 kfree(fsopt->snapdir_name);
 224                 fsopt->snapdir_name = kstrndup(argstr[0].from,
 225                                                argstr[0].to-argstr[0].from,
 226                                                GFP_KERNEL);
 227                 if (!fsopt->snapdir_name)
 228                         return -ENOMEM;
 229                 break;
 230         case Opt_mds_namespace:
 231                 kfree(fsopt->mds_namespace);
 232                 fsopt->mds_namespace = kstrndup(argstr[0].from,
 233                                                 argstr[0].to-argstr[0].from,
 234                                                 GFP_KERNEL);
 235                 if (!fsopt->mds_namespace)
 236                         return -ENOMEM;
 237                 break;
 238         case Opt_fscache_uniq:
 239                 kfree(fsopt->fscache_uniq);
 240                 fsopt->fscache_uniq = kstrndup(argstr[0].from,
 241                                                argstr[0].to-argstr[0].from,
 242                                                GFP_KERNEL);
 243                 if (!fsopt->fscache_uniq)
 244                         return -ENOMEM;
 245                 fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
 246                 break;
 247                 /* misc */
 248         case Opt_wsize:
 249                 if (intval < PAGE_SIZE || intval > CEPH_MAX_WRITE_SIZE)
 250                         return -EINVAL;
 251                 fsopt->wsize = ALIGN(intval, PAGE_SIZE);
 252                 break;
 253         case Opt_rsize:
 254                 if (intval < PAGE_SIZE || intval > CEPH_MAX_READ_SIZE)
 255                         return -EINVAL;
 256                 fsopt->rsize = ALIGN(intval, PAGE_SIZE);
 257                 break;
 258         case Opt_rasize:
 259                 if (intval < 0)
 260                         return -EINVAL;
 261                 fsopt->rasize = ALIGN(intval, PAGE_SIZE);
 262                 break;
 263         case Opt_caps_wanted_delay_min:
 264                 if (intval < 1)
 265                         return -EINVAL;
 266                 fsopt->caps_wanted_delay_min = intval;
 267                 break;
 268         case Opt_caps_wanted_delay_max:
 269                 if (intval < 1)
 270                         return -EINVAL;
 271                 fsopt->caps_wanted_delay_max = intval;
 272                 break;
 273         case Opt_readdir_max_entries:
 274                 if (intval < 1)
 275                         return -EINVAL;
 276                 fsopt->max_readdir = intval;
 277                 break;
 278         case Opt_readdir_max_bytes:
 279                 if (intval < PAGE_SIZE && intval != 0)
 280                         return -EINVAL;
 281                 fsopt->max_readdir_bytes = intval;
 282                 break;
 283         case Opt_congestion_kb:
 284                 if (intval < 1024) /* at least 1M */
 285                         return -EINVAL;
 286                 fsopt->congestion_kb = intval;
 287                 break;
 288         case Opt_dirstat:
 289                 fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
 290                 break;
 291         case Opt_nodirstat:
 292                 fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
 293                 break;
 294         case Opt_rbytes:
 295                 fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
 296                 break;
 297         case Opt_norbytes:
 298                 fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
 299                 break;
 300         case Opt_asyncreaddir:
 301                 fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
 302                 break;
 303         case Opt_noasyncreaddir:
 304                 fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
 305                 break;
 306         case Opt_dcache:
 307                 fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
 308                 break;
 309         case Opt_nodcache:
 310                 fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
 311                 break;
 312         case Opt_ino32:
 313                 fsopt->flags |= CEPH_MOUNT_OPT_INO32;
 314                 break;
 315         case Opt_noino32:
 316                 fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
 317                 break;
 318         case Opt_fscache:
 319                 fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
 320                 break;
 321         case Opt_nofscache:
 322                 fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
 323                 break;
 324         case Opt_poolperm:
 325                 fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
 326                 printk ("pool perm");
 327                 break;
 328         case Opt_nopoolperm:
 329                 fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
 330                 break;
 331         case Opt_require_active_mds:
 332                 fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT;
 333                 break;
 334         case Opt_norequire_active_mds:
 335                 fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT;
 336                 break;
 337 #ifdef CONFIG_CEPH_FS_POSIX_ACL
 338         case Opt_acl:
 339                 fsopt->sb_flags |= SB_POSIXACL;
 340                 break;
 341 #endif
 342         case Opt_noacl:
 343                 fsopt->sb_flags &= ~SB_POSIXACL;
 344                 break;
 345         default:
 346                 BUG_ON(token);
 347         }
 348         return 0;
 349 }
 350
 351 static void destroy_mount_options(struct ceph_mount_options *args)
 352 {
 353         dout("destroy_mount_options %p\n", args);
 354         kfree(args->snapdir_name);
 355         kfree(args->mds_namespace);
 356         kfree(args->server_path);
 357         kfree(args->fscache_uniq);
 358         kfree(args);
 359 }
 360
 361 static int strcmp_null(const char *s1, const char *s2)
 362 {
 363         if (!s1 && !s2)
 364                 return 0;
 365         if (s1 && !s2)
 366                 return -1;
 367         if (!s1 && s2)
 368                 return 1;
 369         return strcmp(s1, s2);
 370 }
 371
 372 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
 373                                  struct ceph_options *new_opt,
 374                                  struct ceph_fs_client *fsc)
 375 {
 376         struct ceph_mount_options *fsopt1 = new_fsopt;
 377         struct ceph_mount_options *fsopt2 = fsc->mount_options;
 378         int ofs = offsetof(struct ceph_mount_options, snapdir_name);
 379         int ret;
 380
 381         ret = memcmp(fsopt1, fsopt2, ofs);
 382         if (ret)
 383                 return ret;
 384
 385         ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
 386         if (ret)
 387                 return ret;
 388         ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
 389         if (ret)
 390                 return ret;
 391         ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
 392         if (ret)
 393                 return ret;
 394         ret = strcmp_null(fsopt1->fscache_uniq, fsopt2->fscache_uniq);
 395         if (ret)
 396                 return ret;
 397
 398         return ceph_compare_options(new_opt, fsc->client);
 399 }
 400
 401 static int parse_mount_options(struct ceph_mount_options **pfsopt,
 402                                struct ceph_options **popt,
 403                                int flags, char *options,
 404                                const char *dev_name)
 405 {
 406         struct ceph_mount_options *fsopt;
 407         const char *dev_name_end;
 408         int err;
 409
 410         if (!dev_name || !*dev_name)
 411                 return -EINVAL;
 412
 413         fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
 414         if (!fsopt)
 415                 return -ENOMEM;
 416
 417         dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
 418
 419         fsopt->sb_flags = flags;
 420         fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
 421
 422         fsopt->wsize = CEPH_MAX_WRITE_SIZE;
 423         fsopt->rsize = CEPH_MAX_READ_SIZE;
 424         fsopt->rasize = CEPH_RASIZE_DEFAULT;
 425         fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
 426         if (!fsopt->snapdir_name) {
 427                 err = -ENOMEM;
 428                 goto out;
 429         }
 430
 431         fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
 432         fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
 433         fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
 434         fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
 435         fsopt->congestion_kb = default_congestion_kb();
 436
 437         /*
 438          * Distinguish the server list from the path in "dev_name".
 439          * Internally we do not include the leading '/' in the path.
 440          *
 441          * "dev_name" will look like:
 442          *     <server_spec>[,<server_spec>...]:[<path>]
 443          * where
 444          *     <server_spec> is <ip>[:<port>]
 445          *     <path> is optional, but if present must begin with '/'
 446          */
 447         dev_name_end = strchr(dev_name, '/');
 448         if (dev_name_end) {
 449                 if (strlen(dev_name_end) > 1) {
 450                         fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
 451                         if (!fsopt->server_path) {
 452                                 err = -ENOMEM;
 453                                 goto out;
 454                         }
 455                 }
 456         } else {
 457                 dev_name_end = dev_name + strlen(dev_name);
 458         }
 459         err = -EINVAL;
 460         dev_name_end--;         /* back up to ':' separator */
 461         if (dev_name_end < dev_name || *dev_name_end != ':') {
 462                 pr_err("device name is missing path (no : separator in %s)\n",
 463                                 dev_name);
 464                 goto out;
 465         }
 466         dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
 467         if (fsopt->server_path)
 468                 dout("server path '%s'\n", fsopt->server_path);
 469
 470         *popt = ceph_parse_options(options, dev_name, dev_name_end,
 471                                  parse_fsopt_token, (void *)fsopt);
 472         if (IS_ERR(*popt)) {
 473                 err = PTR_ERR(*popt);
 474                 goto out;
 475         }
 476
 477         /* success */
 478         *pfsopt = fsopt;
 479         return 0;
 480
 481 out:
 482         destroy_mount_options(fsopt);
 483         return err;
 484 }
 485
 486 /**
 487  * ceph_show_options - Show mount options in /proc/mounts
 488  * @m: seq_file to write to
 489  * @root: root of that (sub)tree
 490  */
 491 static int ceph_show_options(struct seq_file *m, struct dentry *root)
 492 {
 493         struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
 494         struct ceph_mount_options *fsopt = fsc->mount_options;
 495         size_t pos;
 496         int ret;
 497
 498         /* a comma between MNT/MS and client options */
 499         seq_putc(m, ',');
 500         pos = m->count;
 501
 502         ret = ceph_print_client_options(m, fsc->client);
 503         if (ret)
 504                 return ret;
 505
 506         /* retract our comma if no client options */
 507         if (m->count == pos)
 508                 m->count--;
 509
 510         if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
 511                 seq_puts(m, ",dirstat");
 512         if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
 513                 seq_puts(m, ",rbytes");
 514         if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
 515                 seq_puts(m, ",noasyncreaddir");
 516         if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
 517                 seq_puts(m, ",nodcache");
 518         if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) {
 519                 if (fsopt->fscache_uniq)
 520                         seq_printf(m, ",fsc=%s", fsopt->fscache_uniq);
 521                 else
 522                         seq_puts(m, ",fsc");
 523         }
 524         if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
 525                 seq_puts(m, ",nopoolperm");
 526
 527 #ifdef CONFIG_CEPH_FS_POSIX_ACL
 528         if (fsopt->sb_flags & SB_POSIXACL)
 529                 seq_puts(m, ",acl");
 530         else
 531                 seq_puts(m, ",noacl");
 532 #endif
 533
 534         if (fsopt->mds_namespace)
 535                 seq_printf(m, ",mds_namespace=%s", fsopt->mds_namespace);
 536         if (fsopt->wsize)
 537                 seq_printf(m, ",wsize=%d", fsopt->wsize);
 538         if (fsopt->rsize != CEPH_MAX_READ_SIZE)
 539                 seq_printf(m, ",rsize=%d", fsopt->rsize);
 540         if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
 541                 seq_printf(m, ",rasize=%d", fsopt->rasize);
 542         if (fsopt->congestion_kb != default_congestion_kb())
 543                 seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
 544         if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
 545                 seq_printf(m, ",caps_wanted_delay_min=%d",
 546                          fsopt->caps_wanted_delay_min);
 547         if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
 548                 seq_printf(m, ",caps_wanted_delay_max=%d",
 549                            fsopt->caps_wanted_delay_max);
 550         if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
 551                 seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
 552         if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
 553                 seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
 554         if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
 555                 seq_show_option(m, "snapdirname", fsopt->snapdir_name);
 556
 557         return 0;
 558 }
 559
 560 /*
 561  * handle any mon messages the standard library doesn't understand.
 562  * return error if we don't either.
 563  */
 564 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
 565 {
 566         struct ceph_fs_client *fsc = client->private;
 567         int type = le16_to_cpu(msg->hdr.type);
 568
 569         switch (type) {
 570         case CEPH_MSG_MDS_MAP:
 571                 ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
 572                 return 0;
 573         case CEPH_MSG_FS_MAP_USER:
 574                 ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
 575                 return 0;
 576         default:
 577                 return -1;
 578         }
 579 }
 580
 581 /*
 582  * create a new fs client
 583  */
 584 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
 585                                         struct ceph_options *opt)
 586 {
 587         struct ceph_fs_client *fsc;
 588         int page_count;
 589         size_t size;
 590         int err = -ENOMEM;
 591
 592         fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
 593         if (!fsc)
 594                 return ERR_PTR(-ENOMEM);
 595
 596         fsc->client = ceph_create_client(opt, fsc);
 597         if (IS_ERR(fsc->client)) {
 598                 err = PTR_ERR(fsc->client);
 599                 goto fail;
 600         }
 601         fsc->client->extra_mon_dispatch = extra_mon_dispatch;
 602
 603         if (!fsopt->mds_namespace) {
 604                 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
 605                                    0, true);
 606         } else {
 607                 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
 608                                    0, false);
 609         }
 610
 611         fsc->mount_options = fsopt;
 612
 613         fsc->sb = NULL;
 614         fsc->mount_state = CEPH_MOUNT_MOUNTING;
 615
 616         atomic_long_set(&fsc->writeback_count, 0);
 617
 618         err = -ENOMEM;
 619         /*
 620          * The number of concurrent works can be high but they don't need
 621          * to be processed in parallel, limit concurrency.
 622          */
 623         fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
 624         if (!fsc->wb_wq)
 625                 goto fail_client;
 626         fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
 627         if (!fsc->pg_inv_wq)
 628                 goto fail_wb_wq;
 629         fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
 630         if (!fsc->trunc_wq)
 631                 goto fail_pg_inv_wq;
 632
 633         /* set up mempools */
 634         err = -ENOMEM;
 635         page_count = fsc->mount_options->wsize >> PAGE_SHIFT;
 636         size = sizeof (struct page *) * (page_count ? page_count : 1);
 637         fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
 638         if (!fsc->wb_pagevec_pool)
 639                 goto fail_trunc_wq;
 640
 641         /* caps */
 642         fsc->min_caps = fsopt->max_readdir;
 643
 644         return fsc;
 645
 646 fail_trunc_wq:
 647         destroy_workqueue(fsc->trunc_wq);
 648 fail_pg_inv_wq:
 649         destroy_workqueue(fsc->pg_inv_wq);
 650 fail_wb_wq:
 651         destroy_workqueue(fsc->wb_wq);
 652 fail_client:
 653         ceph_destroy_client(fsc->client);
 654 fail:
 655         kfree(fsc);
 656         return ERR_PTR(err);
 657 }
 658
 659 static void destroy_fs_client(struct ceph_fs_client *fsc)
 660 {
 661         dout("destroy_fs_client %p\n", fsc);
 662
 663         destroy_workqueue(fsc->wb_wq);
 664         destroy_workqueue(fsc->pg_inv_wq);
 665         destroy_workqueue(fsc->trunc_wq);
 666
 667         mempool_destroy(fsc->wb_pagevec_pool);
 668
 669         destroy_mount_options(fsc->mount_options);
 670
 671         ceph_destroy_client(fsc->client);
 672
 673         kfree(fsc);
 674         dout("destroy_fs_client %p done\n", fsc);
 675 }
 676
 677 /*
 678  * caches
 679  */
 680 struct kmem_cache *ceph_inode_cachep;
 681 struct kmem_cache *ceph_cap_cachep;
 682 struct kmem_cache *ceph_cap_flush_cachep;
 683 struct kmem_cache *ceph_dentry_cachep;
 684 struct kmem_cache *ceph_file_cachep;
 685
 686 static void ceph_inode_init_once(void *foo)
 687 {
 688         struct ceph_inode_info *ci = foo;
 689         inode_init_once(&ci->vfs_inode);
 690 }
 691
 692 static int __init init_caches(void)
 693 {
 694         int error = -ENOMEM;
 695
 696         ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
 697                                       sizeof(struct ceph_inode_info),
 698                                       __alignof__(struct ceph_inode_info),
 699                                       SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
 700                                       SLAB_ACCOUNT, ceph_inode_init_once);
 701         if (!ceph_inode_cachep)
 702                 return -ENOMEM;
 703
 704         ceph_cap_cachep = KMEM_CACHE(ceph_cap,
 705                                      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 706         if (!ceph_cap_cachep)
 707                 goto bad_cap;
 708         ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
 709                                            SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 710         if (!ceph_cap_flush_cachep)
 711                 goto bad_cap_flush;
 712
 713         ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
 714                                         SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 715         if (!ceph_dentry_cachep)
 716                 goto bad_dentry;
 717
 718         ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD);
 719         if (!ceph_file_cachep)
 720                 goto bad_file;
 721
 722         error = ceph_fscache_register();
 723         if (error)
 724                 goto bad_fscache;
 725
 726         return 0;
 727
 728 bad_fscache:
 729         kmem_cache_destroy(ceph_file_cachep);
 730 bad_file:
 731         kmem_cache_destroy(ceph_dentry_cachep);
 732 bad_dentry:
 733         kmem_cache_destroy(ceph_cap_flush_cachep);
 734 bad_cap_flush:
 735         kmem_cache_destroy(ceph_cap_cachep);
 736 bad_cap:
 737         kmem_cache_destroy(ceph_inode_cachep);
 738         return error;
 739 }
 740
 741 static void destroy_caches(void)
 742 {
 743         /*
 744          * Make sure all delayed rcu free inodes are flushed before we
 745          * destroy cache.
 746          */
 747         rcu_barrier();
 748
 749         kmem_cache_destroy(ceph_inode_cachep);
 750         kmem_cache_destroy(ceph_cap_cachep);
 751         kmem_cache_destroy(ceph_cap_flush_cachep);
 752         kmem_cache_destroy(ceph_dentry_cachep);
 753         kmem_cache_destroy(ceph_file_cachep);
 754
 755         ceph_fscache_unregister();
 756 }
 757
 758
 759 /*
 760  * ceph_umount_begin - initiate forced umount.  Tear down down the
 761  * mount, skipping steps that may hang while waiting for server(s).
 762  */
 763 static void ceph_umount_begin(struct super_block *sb)
 764 {
 765         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
 766
 767         dout("ceph_umount_begin - starting forced umount\n");
 768         if (!fsc)
 769                 return;
 770         fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
 771         ceph_mdsc_force_umount(fsc->mdsc);
 772         return;
 773 }
 774
 775 static const struct super_operations ceph_super_ops = {
 776         .alloc_inode    = ceph_alloc_inode,
 777         .destroy_inode  = ceph_destroy_inode,
 778         .write_inode    = ceph_write_inode,
 779         .drop_inode     = ceph_drop_inode,
 780         .sync_fs        = ceph_sync_fs,
 781         .put_super      = ceph_put_super,
 782         .show_options   = ceph_show_options,
 783         .statfs         = ceph_statfs,
 784         .umount_begin   = ceph_umount_begin,
 785 };
 786
 787 /*
 788  * Bootstrap mount by opening the root directory.  Note the mount
 789  * @started time from caller, and time out if this takes too long.
 790  */
 791 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
 792                                        const char *path,
 793                                        unsigned long started)
 794 {
 795         struct ceph_mds_client *mdsc = fsc->mdsc;
 796         struct ceph_mds_request *req = NULL;
 797         int err;
 798         struct dentry *root;
 799
 800         /* open dir */
 801         dout("open_root_inode opening '%s'\n", path);
 802         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
 803         if (IS_ERR(req))
 804                 return ERR_CAST(req);
 805         req->r_path1 = kstrdup(path, GFP_NOFS);
 806         if (!req->r_path1) {
 807                 root = ERR_PTR(-ENOMEM);
 808                 goto out;
 809         }
 810
 811         req->r_ino1.ino = CEPH_INO_ROOT;
 812         req->r_ino1.snap = CEPH_NOSNAP;
 813         req->r_started = started;
 814         req->r_timeout = fsc->client->options->mount_timeout;
 815         req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
 816         req->r_num_caps = 2;
 817         err = ceph_mdsc_do_request(mdsc, NULL, req);
 818         if (err == 0) {
 819                 struct inode *inode = req->r_target_inode;
 820                 req->r_target_inode = NULL;
 821                 dout("open_root_inode success\n");
 822                 root = d_make_root(inode);
 823                 if (!root) {
 824                         root = ERR_PTR(-ENOMEM);
 825                         goto out;
 826                 }
 827                 dout("open_root_inode success, root dentry is %p\n", root);
 828         } else {
 829                 root = ERR_PTR(err);
 830         }
 831 out:
 832         ceph_mdsc_put_request(req);
 833         return root;
 834 }
 835
 836
 837
 838
 839 /*
 840  * mount: join the ceph cluster, and open root directory.
 841  */
 842 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc)
 843 {
 844         int err;
 845         unsigned long started = jiffies;  /* note the start time */
 846         struct dentry *root;
 847
 848         dout("mount start %p\n", fsc);
 849         mutex_lock(&fsc->client->mount_mutex);
 850
 851         if (!fsc->sb->s_root) {
 852                 const char *path;
 853                 err = __ceph_open_session(fsc->client, started);
 854                 if (err < 0)
 855                         goto out;
 856
 857                 /* setup fscache */
 858                 if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
 859                         err = ceph_fscache_register_fs(fsc);
 860                         if (err < 0)
 861                                 goto out;
 862                 }
 863
 864                 if (!fsc->mount_options->server_path) {
 865                         path = "";
 866                         dout("mount opening path \\t\n");
 867                 } else {
 868                         path = fsc->mount_options->server_path + 1;
 869                         dout("mount opening path %s\n", path);
 870                 }
 871
 872                 err = ceph_fs_debugfs_init(fsc);
 873                 if (err < 0)
 874                         goto out;
 875
 876                 root = open_root_dentry(fsc, path, started);
 877                 if (IS_ERR(root)) {
 878                         err = PTR_ERR(root);
 879                         goto out;
 880                 }
 881                 fsc->sb->s_root = dget(root);
 882         } else {
 883                 root = dget(fsc->sb->s_root);
 884         }
 885
 886         fsc->mount_state = CEPH_MOUNT_MOUNTED;
 887         dout("mount success\n");
 888         mutex_unlock(&fsc->client->mount_mutex);
 889         return root;
 890
 891 out:
 892         mutex_unlock(&fsc->client->mount_mutex);
 893         return ERR_PTR(err);
 894 }
 895
 896 static int ceph_set_super(struct super_block *s, void *data)
 897 {
 898         struct ceph_fs_client *fsc = data;
 899         int ret;
 900
 901         dout("set_super %p data %p\n", s, data);
 902
 903         s->s_flags = fsc->mount_options->sb_flags;
 904         s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
 905
 906         s->s_xattr = ceph_xattr_handlers;
 907         s->s_fs_info = fsc;
 908         fsc->sb = s;
 909
 910         s->s_op = &ceph_super_ops;
 911         s->s_d_op = &ceph_dentry_ops;
 912         s->s_export_op = &ceph_export_ops;
 913
 914         s->s_time_gran = 1000;  /* 1000 ns == 1 us */
 915
 916         ret = set_anon_super(s, NULL);  /* what is that second arg for? */
 917         if (ret != 0)
 918                 goto fail;
 919
 920         return ret;
 921
 922 fail:
 923         s->s_fs_info = NULL;
 924         fsc->sb = NULL;
 925         return ret;
 926 }
 927
 928 /*
 929  * share superblock if same fs AND options
 930  */
 931 static int ceph_compare_super(struct super_block *sb, void *data)
 932 {
 933         struct ceph_fs_client *new = data;
 934         struct ceph_mount_options *fsopt = new->mount_options;
 935         struct ceph_options *opt = new->client->options;
 936         struct ceph_fs_client *other = ceph_sb_to_client(sb);
 937
 938         dout("ceph_compare_super %p\n", sb);
 939
 940         if (compare_mount_options(fsopt, opt, other)) {
 941                 dout("monitor(s)/mount options don't match\n");
 942                 return 0;
 943         }
 944         if ((opt->flags & CEPH_OPT_FSID) &&
 945             ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
 946                 dout("fsid doesn't match\n");
 947                 return 0;
 948         }
 949         if (fsopt->sb_flags != other->mount_options->sb_flags) {
 950                 dout("flags differ\n");
 951                 return 0;
 952         }
 953         return 1;
 954 }
 955
 956 /*
 957  * construct our own bdi so we can control readahead, etc.
 958  */
 959 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
 960
 961 static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
 962 {
 963         int err;
 964
 965         err = super_setup_bdi_name(sb, "ceph-%ld",
 966                                    atomic_long_inc_return(&bdi_seq));
 967         if (err)
 968                 return err;
 969
 970         /* set ra_pages based on rasize mount option? */
 971         sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
 972
 973         /* set io_pages based on max osd read size */
 974         sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
 975
 976         return 0;
 977 }
 978
 979 static struct dentry *ceph_mount(struct file_system_type *fs_type,
 980                        int flags, const char *dev_name, void *data)
 981 {
 982         struct super_block *sb;
 983         struct ceph_fs_client *fsc;
 984         struct dentry *res;
 985         int err;
 986         int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
 987         struct ceph_mount_options *fsopt = NULL;
 988         struct ceph_options *opt = NULL;
 989
 990         dout("ceph_mount\n");
 991
 992 #ifdef CONFIG_CEPH_FS_POSIX_ACL
 993         flags |= SB_POSIXACL;
 994 #endif
 995         err = parse_mount_options(&fsopt, &opt, flags, data, dev_name);
 996         if (err < 0) {
 997                 res = ERR_PTR(err);
 998                 goto out_final;
 999         }
1000
1001         /* create client (which we may/may not use) */
1002         fsc = create_fs_client(fsopt, opt);
1003         if (IS_ERR(fsc)) {
1004                 res = ERR_CAST(fsc);
1005                 destroy_mount_options(fsopt);
1006                 ceph_destroy_options(opt);
1007                 goto out_final;
1008         }
1009
1010         err = ceph_mdsc_init(fsc);
1011         if (err < 0) {
1012                 res = ERR_PTR(err);
1013                 goto out;
1014         }
1015
1016         if (ceph_test_opt(fsc->client, NOSHARE))
1017                 compare_super = NULL;
1018         sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc);
1019         if (IS_ERR(sb)) {
1020                 res = ERR_CAST(sb);
1021                 goto out;
1022         }
1023
1024         if (ceph_sb_to_client(sb) != fsc) {
1025                 ceph_mdsc_destroy(fsc);
1026                 destroy_fs_client(fsc);
1027                 fsc = ceph_sb_to_client(sb);
1028                 dout("get_sb got existing client %p\n", fsc);
1029         } else {
1030                 dout("get_sb using new client %p\n", fsc);
1031                 err = ceph_setup_bdi(sb, fsc);
1032                 if (err < 0) {
1033                         res = ERR_PTR(err);
1034                         goto out_splat;
1035                 }
1036         }
1037
1038         res = ceph_real_mount(fsc);
1039         if (IS_ERR(res))
1040                 goto out_splat;
1041         dout("root %p inode %p ino %llx.%llx\n", res,
1042              d_inode(res), ceph_vinop(d_inode(res)));
1043         return res;
1044
1045 out_splat:
1046         ceph_mdsc_close_sessions(fsc->mdsc);
1047         deactivate_locked_super(sb);
1048         goto out_final;
1049
1050 out:
1051         ceph_mdsc_destroy(fsc);
1052         destroy_fs_client(fsc);
1053 out_final:
1054         dout("ceph_mount fail %ld\n", PTR_ERR(res));
1055         return res;
1056 }
1057
1058 static void ceph_kill_sb(struct super_block *s)
1059 {
1060         struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1061         dev_t dev = s->s_dev;
1062
1063         dout("kill_sb %p\n", s);
1064
1065         ceph_mdsc_pre_umount(fsc->mdsc);
1066         generic_shutdown_super(s);
1067
1068         fsc->client->extra_mon_dispatch = NULL;
1069         ceph_fs_debugfs_cleanup(fsc);
1070
1071         ceph_fscache_unregister_fs(fsc);
1072
1073         ceph_mdsc_destroy(fsc);
1074
1075         destroy_fs_client(fsc);
1076         free_anon_bdev(dev);
1077 }
1078
1079 static struct file_system_type ceph_fs_type = {
1080         .owner          = THIS_MODULE,
1081         .name           = "ceph",
1082         .mount          = ceph_mount,
1083         .kill_sb        = ceph_kill_sb,
1084         .fs_flags       = FS_RENAME_DOES_D_MOVE,
1085 };
1086 MODULE_ALIAS_FS("ceph");
1087
1088 static int __init init_ceph(void)
1089 {
1090         int ret = init_caches();
1091         if (ret)
1092                 goto out;
1093
1094         ceph_flock_init();
1095         ceph_xattr_init();
1096         ret = register_filesystem(&ceph_fs_type);
1097         if (ret)
1098                 goto out_xattr;
1099
1100         pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1101
1102         return 0;
1103
1104 out_xattr:
1105         ceph_xattr_exit();
1106         destroy_caches();
1107 out:
1108         return ret;
1109 }
1110
1111 static void __exit exit_ceph(void)
1112 {
1113         dout("exit_ceph\n");
1114         unregister_filesystem(&ceph_fs_type);
1115         ceph_xattr_exit();
1116         destroy_caches();
1117 }
1118
1119 module_init(init_ceph);
1120 module_exit(exit_ceph);
1121
1122 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1123 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1124 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1125 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1126 MODULE_LICENSE("GPL");