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 static int ceph_sync_fs(struct super_block *sb, int wait)
  89 {
  90         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
  91
  92         if (!wait) {
  93                 dout("sync_fs (non-blocking)\n");
  94                 ceph_flush_dirty_caps(fsc->mdsc);
  95                 dout("sync_fs (non-blocking) done\n");
  96                 return 0;
  97         }
  98
  99         dout("sync_fs (blocking)\n");
 100         ceph_osdc_sync(&fsc->client->osdc);
 101         ceph_mdsc_sync(fsc->mdsc);
 102         dout("sync_fs (blocking) done\n");
 103         return 0;
 104 }
 105
 106 /*
 107  * mount options
 108  */
 109 enum {
 110         Opt_wsize,
 111         Opt_rsize,
 112         Opt_rasize,
 113         Opt_caps_wanted_delay_min,
 114         Opt_caps_wanted_delay_max,
 115         Opt_cap_release_safety,
 116         Opt_readdir_max_entries,
 117         Opt_readdir_max_bytes,
 118         Opt_congestion_kb,
 119         Opt_last_int,
 120         /* int args above */
 121         Opt_snapdirname,
 122         Opt_mds_namespace,
 123         Opt_last_string,
 124         /* string args above */
 125         Opt_dirstat,
 126         Opt_nodirstat,
 127         Opt_rbytes,
 128         Opt_norbytes,
 129         Opt_asyncreaddir,
 130         Opt_noasyncreaddir,
 131         Opt_dcache,
 132         Opt_nodcache,
 133         Opt_ino32,
 134         Opt_noino32,
 135         Opt_fscache,
 136         Opt_nofscache,
 137         Opt_poolperm,
 138         Opt_nopoolperm,
 139 #ifdef CONFIG_CEPH_FS_POSIX_ACL
 140         Opt_acl,
 141 #endif
 142         Opt_noacl,
 143 };
 144
 145 static match_table_t fsopt_tokens = {
 146         {Opt_wsize, "wsize=%d"},
 147         {Opt_rsize, "rsize=%d"},
 148         {Opt_rasize, "rasize=%d"},
 149         {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
 150         {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
 151         {Opt_cap_release_safety, "cap_release_safety=%d"},
 152         {Opt_readdir_max_entries, "readdir_max_entries=%d"},
 153         {Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
 154         {Opt_congestion_kb, "write_congestion_kb=%d"},
 155         /* int args above */
 156         {Opt_snapdirname, "snapdirname=%s"},
 157         {Opt_mds_namespace, "mds_namespace=%s"},
 158         /* string args above */
 159         {Opt_dirstat, "dirstat"},
 160         {Opt_nodirstat, "nodirstat"},
 161         {Opt_rbytes, "rbytes"},
 162         {Opt_norbytes, "norbytes"},
 163         {Opt_asyncreaddir, "asyncreaddir"},
 164         {Opt_noasyncreaddir, "noasyncreaddir"},
 165         {Opt_dcache, "dcache"},
 166         {Opt_nodcache, "nodcache"},
 167         {Opt_ino32, "ino32"},
 168         {Opt_noino32, "noino32"},
 169         {Opt_fscache, "fsc"},
 170         {Opt_nofscache, "nofsc"},
 171         {Opt_poolperm, "poolperm"},
 172         {Opt_nopoolperm, "nopoolperm"},
 173 #ifdef CONFIG_CEPH_FS_POSIX_ACL
 174         {Opt_acl, "acl"},
 175 #endif
 176         {Opt_noacl, "noacl"},
 177         {-1, NULL}
 178 };
 179
 180 /*
 181  * Remove adjacent slashes and then the trailing slash, unless it is
 182  * the only remaining character.
 183  *
 184  * E.g. "//dir1////dir2///" --> "/dir1/dir2", "///" --> "/".
 185  */
 186 static void canonicalize_path(char *path)
 187 {
 188         int i, j = 0;
 189
 190         for (i = 0; path[i] != '\0'; i++) {
 191                 if (path[i] != '/' || j < 1 || path[j - 1] != '/')
 192                         path[j++] = path[i];
 193         }
 194
 195         if (j > 1 && path[j - 1] == '/')
 196                 j--;
 197         path[j] = '\0';
 198 }
 199
 200 static int parse_fsopt_token(char *c, void *private)
 201 {
 202         struct ceph_mount_options *fsopt = private;
 203         substring_t argstr[MAX_OPT_ARGS];
 204         int token, intval, ret;
 205
 206         token = match_token((char *)c, fsopt_tokens, argstr);
 207         if (token < 0)
 208                 return -EINVAL;
 209
 210         if (token < Opt_last_int) {
 211                 ret = match_int(&argstr[0], &intval);
 212                 if (ret < 0) {
 213                         pr_err("bad mount option arg (not int) "
 214                                "at '%s'\n", c);
 215                         return ret;
 216                 }
 217                 dout("got int token %d val %d\n", token, intval);
 218         } else if (token > Opt_last_int && token < Opt_last_string) {
 219                 dout("got string token %d val %s\n", token,
 220                      argstr[0].from);
 221         } else {
 222                 dout("got token %d\n", token);
 223         }
 224
 225         switch (token) {
 226         case Opt_snapdirname:
 227                 kfree(fsopt->snapdir_name);
 228                 fsopt->snapdir_name = kstrndup(argstr[0].from,
 229                                                argstr[0].to-argstr[0].from,
 230                                                GFP_KERNEL);
 231                 if (!fsopt->snapdir_name)
 232                         return -ENOMEM;
 233                 break;
 234         case Opt_mds_namespace:
 235                 fsopt->mds_namespace = kstrndup(argstr[0].from,
 236                                                 argstr[0].to-argstr[0].from,
 237                                                 GFP_KERNEL);
 238                 if (!fsopt->mds_namespace)
 239                         return -ENOMEM;
 240                 break;
 241                 /* misc */
 242         case Opt_wsize:
 243                 fsopt->wsize = intval;
 244                 break;
 245         case Opt_rsize:
 246                 fsopt->rsize = intval;
 247                 break;
 248         case Opt_rasize:
 249                 fsopt->rasize = intval;
 250                 break;
 251         case Opt_caps_wanted_delay_min:
 252                 fsopt->caps_wanted_delay_min = intval;
 253                 break;
 254         case Opt_caps_wanted_delay_max:
 255                 fsopt->caps_wanted_delay_max = intval;
 256                 break;
 257         case Opt_readdir_max_entries:
 258                 fsopt->max_readdir = intval;
 259                 break;
 260         case Opt_readdir_max_bytes:
 261                 fsopt->max_readdir_bytes = intval;
 262                 break;
 263         case Opt_congestion_kb:
 264                 fsopt->congestion_kb = intval;
 265                 break;
 266         case Opt_dirstat:
 267                 fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
 268                 break;
 269         case Opt_nodirstat:
 270                 fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
 271                 break;
 272         case Opt_rbytes:
 273                 fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
 274                 break;
 275         case Opt_norbytes:
 276                 fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
 277                 break;
 278         case Opt_asyncreaddir:
 279                 fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
 280                 break;
 281         case Opt_noasyncreaddir:
 282                 fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
 283                 break;
 284         case Opt_dcache:
 285                 fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
 286                 break;
 287         case Opt_nodcache:
 288                 fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
 289                 break;
 290         case Opt_ino32:
 291                 fsopt->flags |= CEPH_MOUNT_OPT_INO32;
 292                 break;
 293         case Opt_noino32:
 294                 fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
 295                 break;
 296         case Opt_fscache:
 297                 fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
 298                 break;
 299         case Opt_nofscache:
 300                 fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
 301                 break;
 302         case Opt_poolperm:
 303                 fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
 304                 printk ("pool perm");
 305                 break;
 306         case Opt_nopoolperm:
 307                 fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
 308                 break;
 309 #ifdef CONFIG_CEPH_FS_POSIX_ACL
 310         case Opt_acl:
 311                 fsopt->sb_flags |= MS_POSIXACL;
 312                 break;
 313 #endif
 314         case Opt_noacl:
 315                 fsopt->sb_flags &= ~MS_POSIXACL;
 316                 break;
 317         default:
 318                 BUG_ON(token);
 319         }
 320         return 0;
 321 }
 322
 323 static void destroy_mount_options(struct ceph_mount_options *args)
 324 {
 325         dout("destroy_mount_options %p\n", args);
 326         kfree(args->snapdir_name);
 327         kfree(args->mds_namespace);
 328         kfree(args->server_path);
 329         kfree(args);
 330 }
 331
 332 static int strcmp_null(const char *s1, const char *s2)
 333 {
 334         if (!s1 && !s2)
 335                 return 0;
 336         if (s1 && !s2)
 337                 return -1;
 338         if (!s1 && s2)
 339                 return 1;
 340         return strcmp(s1, s2);
 341 }
 342
 343 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
 344                                  struct ceph_options *new_opt,
 345                                  struct ceph_fs_client *fsc)
 346 {
 347         struct ceph_mount_options *fsopt1 = new_fsopt;
 348         struct ceph_mount_options *fsopt2 = fsc->mount_options;
 349         int ofs = offsetof(struct ceph_mount_options, snapdir_name);
 350         int ret;
 351
 352         ret = memcmp(fsopt1, fsopt2, ofs);
 353         if (ret)
 354                 return ret;
 355
 356         ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
 357         if (ret)
 358                 return ret;
 359
 360         ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
 361         if (ret)
 362                 return ret;
 363
 364         ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
 365         if (ret)
 366                 return ret;
 367
 368         return ceph_compare_options(new_opt, fsc->client);
 369 }
 370
 371 static int parse_mount_options(struct ceph_mount_options **pfsopt,
 372                                struct ceph_options **popt,
 373                                int flags, char *options,
 374                                const char *dev_name)
 375 {
 376         struct ceph_mount_options *fsopt;
 377         const char *dev_name_end;
 378         int err;
 379
 380         if (!dev_name || !*dev_name)
 381                 return -EINVAL;
 382
 383         fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
 384         if (!fsopt)
 385                 return -ENOMEM;
 386
 387         dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
 388
 389         fsopt->sb_flags = flags;
 390         fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
 391
 392         fsopt->rsize = CEPH_RSIZE_DEFAULT;
 393         fsopt->rasize = CEPH_RASIZE_DEFAULT;
 394         fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
 395         if (!fsopt->snapdir_name) {
 396                 err = -ENOMEM;
 397                 goto out;
 398         }
 399
 400         fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
 401         fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
 402         fsopt->cap_release_safety = CEPH_CAP_RELEASE_SAFETY_DEFAULT;
 403         fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
 404         fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
 405         fsopt->congestion_kb = default_congestion_kb();
 406
 407         /*
 408          * Distinguish the server list from the path in "dev_name".
 409          * Internally we do not include the leading '/' in the path.
 410          *
 411          * "dev_name" will look like:
 412          *     <server_spec>[,<server_spec>...]:[<path>]
 413          * where
 414          *     <server_spec> is <ip>[:<port>]
 415          *     <path> is optional, but if present must begin with '/'
 416          */
 417         dev_name_end = strchr(dev_name, '/');
 418         if (dev_name_end) {
 419                 /*
 420                  * The server_path will include the whole chars from userland
 421                  * including the leading '/'.
 422                  */
 423                 fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
 424                 if (!fsopt->server_path) {
 425                         err = -ENOMEM;
 426                         goto out;
 427                 }
 428
 429                 canonicalize_path(fsopt->server_path);
 430         } else {
 431                 dev_name_end = dev_name + strlen(dev_name);
 432         }
 433         err = -EINVAL;
 434         dev_name_end--;         /* back up to ':' separator */
 435         if (dev_name_end < dev_name || *dev_name_end != ':') {
 436                 pr_err("device name is missing path (no : separator in %s)\n",
 437                                 dev_name);
 438                 goto out;
 439         }
 440         dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
 441         if (fsopt->server_path)
 442                 dout("server path '%s'\n", fsopt->server_path);
 443
 444         *popt = ceph_parse_options(options, dev_name, dev_name_end,
 445                                  parse_fsopt_token, (void *)fsopt);
 446         if (IS_ERR(*popt)) {
 447                 err = PTR_ERR(*popt);
 448                 goto out;
 449         }
 450
 451         /* success */
 452         *pfsopt = fsopt;
 453         return 0;
 454
 455 out:
 456         destroy_mount_options(fsopt);
 457         return err;
 458 }
 459
 460 /**
 461  * ceph_show_options - Show mount options in /proc/mounts
 462  * @m: seq_file to write to
 463  * @root: root of that (sub)tree
 464  */
 465 static int ceph_show_options(struct seq_file *m, struct dentry *root)
 466 {
 467         struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
 468         struct ceph_mount_options *fsopt = fsc->mount_options;
 469         size_t pos;
 470         int ret;
 471
 472         /* a comma between MNT/MS and client options */
 473         seq_putc(m, ',');
 474         pos = m->count;
 475
 476         ret = ceph_print_client_options(m, fsc->client);
 477         if (ret)
 478                 return ret;
 479
 480         /* retract our comma if no client options */
 481         if (m->count == pos)
 482                 m->count--;
 483
 484         if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
 485                 seq_puts(m, ",dirstat");
 486         if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
 487                 seq_puts(m, ",rbytes");
 488         if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
 489                 seq_puts(m, ",noasyncreaddir");
 490         if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
 491                 seq_puts(m, ",nodcache");
 492         if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE)
 493                 seq_puts(m, ",fsc");
 494         if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
 495                 seq_puts(m, ",nopoolperm");
 496
 497 #ifdef CONFIG_CEPH_FS_POSIX_ACL
 498         if (fsopt->sb_flags & MS_POSIXACL)
 499                 seq_puts(m, ",acl");
 500         else
 501                 seq_puts(m, ",noacl");
 502 #endif
 503
 504         if (fsopt->mds_namespace)
 505                 seq_printf(m, ",mds_namespace=%s", fsopt->mds_namespace);
 506         if (fsopt->wsize)
 507                 seq_printf(m, ",wsize=%d", fsopt->wsize);
 508         if (fsopt->rsize != CEPH_RSIZE_DEFAULT)
 509                 seq_printf(m, ",rsize=%d", fsopt->rsize);
 510         if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
 511                 seq_printf(m, ",rasize=%d", fsopt->rasize);
 512         if (fsopt->congestion_kb != default_congestion_kb())
 513                 seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
 514         if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
 515                 seq_printf(m, ",caps_wanted_delay_min=%d",
 516                          fsopt->caps_wanted_delay_min);
 517         if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
 518                 seq_printf(m, ",caps_wanted_delay_max=%d",
 519                            fsopt->caps_wanted_delay_max);
 520         if (fsopt->cap_release_safety != CEPH_CAP_RELEASE_SAFETY_DEFAULT)
 521                 seq_printf(m, ",cap_release_safety=%d",
 522                            fsopt->cap_release_safety);
 523         if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
 524                 seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
 525         if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
 526                 seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
 527         if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
 528                 seq_show_option(m, "snapdirname", fsopt->snapdir_name);
 529
 530         return 0;
 531 }
 532
 533 /*
 534  * handle any mon messages the standard library doesn't understand.
 535  * return error if we don't either.
 536  */
 537 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
 538 {
 539         struct ceph_fs_client *fsc = client->private;
 540         int type = le16_to_cpu(msg->hdr.type);
 541
 542         switch (type) {
 543         case CEPH_MSG_MDS_MAP:
 544                 ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
 545                 return 0;
 546         case CEPH_MSG_FS_MAP_USER:
 547                 ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
 548                 return 0;
 549         default:
 550                 return -1;
 551         }
 552 }
 553
 554 /*
 555  * create a new fs client
 556  */
 557 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
 558                                         struct ceph_options *opt)
 559 {
 560         struct ceph_fs_client *fsc;
 561         const u64 supported_features =
 562                 CEPH_FEATURE_FLOCK | CEPH_FEATURE_DIRLAYOUTHASH |
 563                 CEPH_FEATURE_MDSENC | CEPH_FEATURE_MDS_INLINE_DATA;
 564         const u64 required_features = 0;
 565         int page_count;
 566         size_t size;
 567         int err = -ENOMEM;
 568
 569         fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
 570         if (!fsc)
 571                 return ERR_PTR(-ENOMEM);
 572
 573         fsc->client = ceph_create_client(opt, fsc, supported_features,
 574                                          required_features);
 575         if (IS_ERR(fsc->client)) {
 576                 err = PTR_ERR(fsc->client);
 577                 goto fail;
 578         }
 579         fsc->client->extra_mon_dispatch = extra_mon_dispatch;
 580
 581         if (fsopt->mds_namespace == NULL) {
 582                 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
 583                                    0, true);
 584         } else {
 585                 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
 586                                    0, false);
 587         }
 588
 589         fsc->mount_options = fsopt;
 590
 591         fsc->sb = NULL;
 592         fsc->mount_state = CEPH_MOUNT_MOUNTING;
 593
 594         atomic_long_set(&fsc->writeback_count, 0);
 595
 596         err = bdi_init(&fsc->backing_dev_info);
 597         if (err < 0)
 598                 goto fail_client;
 599
 600         err = -ENOMEM;
 601         /*
 602          * The number of concurrent works can be high but they don't need
 603          * to be processed in parallel, limit concurrency.
 604          */
 605         fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
 606         if (fsc->wb_wq == NULL)
 607                 goto fail_bdi;
 608         fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
 609         if (fsc->pg_inv_wq == NULL)
 610                 goto fail_wb_wq;
 611         fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
 612         if (fsc->trunc_wq == NULL)
 613                 goto fail_pg_inv_wq;
 614
 615         /* set up mempools */
 616         err = -ENOMEM;
 617         page_count = fsc->mount_options->wsize >> PAGE_SHIFT;
 618         size = sizeof (struct page *) * (page_count ? page_count : 1);
 619         fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
 620         if (!fsc->wb_pagevec_pool)
 621                 goto fail_trunc_wq;
 622
 623         /* setup fscache */
 624         if ((fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) &&
 625             (ceph_fscache_register_fs(fsc) != 0))
 626                 goto fail_fscache;
 627
 628         /* caps */
 629         fsc->min_caps = fsopt->max_readdir;
 630
 631         return fsc;
 632
 633 fail_fscache:
 634         ceph_fscache_unregister_fs(fsc);
 635 fail_trunc_wq:
 636         destroy_workqueue(fsc->trunc_wq);
 637 fail_pg_inv_wq:
 638         destroy_workqueue(fsc->pg_inv_wq);
 639 fail_wb_wq:
 640         destroy_workqueue(fsc->wb_wq);
 641 fail_bdi:
 642         bdi_destroy(&fsc->backing_dev_info);
 643 fail_client:
 644         ceph_destroy_client(fsc->client);
 645 fail:
 646         kfree(fsc);
 647         return ERR_PTR(err);
 648 }
 649
 650 static void destroy_fs_client(struct ceph_fs_client *fsc)
 651 {
 652         dout("destroy_fs_client %p\n", fsc);
 653
 654         ceph_fscache_unregister_fs(fsc);
 655
 656         destroy_workqueue(fsc->wb_wq);
 657         destroy_workqueue(fsc->pg_inv_wq);
 658         destroy_workqueue(fsc->trunc_wq);
 659
 660         bdi_destroy(&fsc->backing_dev_info);
 661
 662         mempool_destroy(fsc->wb_pagevec_pool);
 663
 664         destroy_mount_options(fsc->mount_options);
 665
 666         ceph_fs_debugfs_cleanup(fsc);
 667
 668         ceph_destroy_client(fsc->client);
 669
 670         kfree(fsc);
 671         dout("destroy_fs_client %p done\n", fsc);
 672 }
 673
 674 /*
 675  * caches
 676  */
 677 struct kmem_cache *ceph_inode_cachep;
 678 struct kmem_cache *ceph_cap_cachep;
 679 struct kmem_cache *ceph_cap_flush_cachep;
 680 struct kmem_cache *ceph_dentry_cachep;
 681 struct kmem_cache *ceph_file_cachep;
 682
 683 static void ceph_inode_init_once(void *foo)
 684 {
 685         struct ceph_inode_info *ci = foo;
 686         inode_init_once(&ci->vfs_inode);
 687 }
 688
 689 static int __init init_caches(void)
 690 {
 691         int error = -ENOMEM;
 692
 693         ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
 694                                       sizeof(struct ceph_inode_info),
 695                                       __alignof__(struct ceph_inode_info),
 696                                       SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
 697                                       SLAB_ACCOUNT, ceph_inode_init_once);
 698         if (ceph_inode_cachep == NULL)
 699                 return -ENOMEM;
 700
 701         ceph_cap_cachep = KMEM_CACHE(ceph_cap,
 702                                      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 703         if (ceph_cap_cachep == NULL)
 704                 goto bad_cap;
 705         ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
 706                                            SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 707         if (ceph_cap_flush_cachep == NULL)
 708                 goto bad_cap_flush;
 709
 710         ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
 711                                         SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 712         if (ceph_dentry_cachep == NULL)
 713                 goto bad_dentry;
 714
 715         ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD);
 716
 717         if (ceph_file_cachep == NULL)
 718                 goto bad_file;
 719
 720         if ((error = ceph_fscache_register()))
 721                 goto bad_file;
 722
 723         return 0;
 724 bad_file:
 725         kmem_cache_destroy(ceph_dentry_cachep);
 726 bad_dentry:
 727         kmem_cache_destroy(ceph_cap_flush_cachep);
 728 bad_cap_flush:
 729         kmem_cache_destroy(ceph_cap_cachep);
 730 bad_cap:
 731         kmem_cache_destroy(ceph_inode_cachep);
 732         return error;
 733 }
 734
 735 static void destroy_caches(void)
 736 {
 737         /*
 738          * Make sure all delayed rcu free inodes are flushed before we
 739          * destroy cache.
 740          */
 741         rcu_barrier();
 742
 743         kmem_cache_destroy(ceph_inode_cachep);
 744         kmem_cache_destroy(ceph_cap_cachep);
 745         kmem_cache_destroy(ceph_cap_flush_cachep);
 746         kmem_cache_destroy(ceph_dentry_cachep);
 747         kmem_cache_destroy(ceph_file_cachep);
 748
 749         ceph_fscache_unregister();
 750 }
 751
 752 /*
 753  * ceph_umount_begin - initiate forced umount.  Tear down down the
 754  * mount, skipping steps that may hang while waiting for server(s).
 755  */
 756 static void ceph_umount_begin(struct super_block *sb)
 757 {
 758         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
 759
 760         dout("ceph_umount_begin - starting forced umount\n");
 761         if (!fsc)
 762                 return;
 763         fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
 764         ceph_mdsc_force_umount(fsc->mdsc);
 765         return;
 766 }
 767
 768 static int ceph_remount(struct super_block *sb, int *flags, char *data)
 769 {
 770         sync_filesystem(sb);
 771         return 0;
 772 }
 773
 774 static const struct super_operations ceph_super_ops = {
 775         .alloc_inode    = ceph_alloc_inode,
 776         .destroy_inode  = ceph_destroy_inode,
 777         .write_inode    = ceph_write_inode,
 778         .drop_inode     = ceph_drop_inode,
 779         .evict_inode    = ceph_evict_inode,
 780         .sync_fs        = ceph_sync_fs,
 781         .put_super      = ceph_put_super,
 782         .remount_fs     = ceph_remount,
 783         .show_options   = ceph_show_options,
 784         .statfs         = ceph_statfs,
 785         .umount_begin   = ceph_umount_begin,
 786 };
 787
 788 /*
 789  * Bootstrap mount by opening the root directory.  Note the mount
 790  * @started time from caller, and time out if this takes too long.
 791  */
 792 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
 793                                        const char *path,
 794                                        unsigned long started)
 795 {
 796         struct ceph_mds_client *mdsc = fsc->mdsc;
 797         struct ceph_mds_request *req = NULL;
 798         int err;
 799         struct dentry *root;
 800
 801         /* open dir */
 802         dout("open_root_inode opening '%s'\n", path);
 803         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
 804         if (IS_ERR(req))
 805                 return ERR_CAST(req);
 806         req->r_path1 = kstrdup(path, GFP_NOFS);
 807         if (!req->r_path1) {
 808                 root = ERR_PTR(-ENOMEM);
 809                 goto out;
 810         }
 811
 812         req->r_ino1.ino = CEPH_INO_ROOT;
 813         req->r_ino1.snap = CEPH_NOSNAP;
 814         req->r_started = started;
 815         req->r_timeout = fsc->client->options->mount_timeout;
 816         req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
 817         req->r_num_caps = 2;
 818         err = ceph_mdsc_do_request(mdsc, NULL, req);
 819         if (err == 0) {
 820                 struct inode *inode = req->r_target_inode;
 821                 req->r_target_inode = NULL;
 822                 dout("open_root_inode success\n");
 823                 root = d_make_root(inode);
 824                 if (!root) {
 825                         root = ERR_PTR(-ENOMEM);
 826                         goto out;
 827                 }
 828                 ceph_init_dentry(root);
 829                 dout("open_root_inode success, root dentry is %p\n", root);
 830         } else {
 831                 root = ERR_PTR(err);
 832         }
 833 out:
 834         ceph_mdsc_put_request(req);
 835         return root;
 836 }
 837
 838 /*
 839  * mount: join the ceph cluster, and open root directory.
 840  */
 841 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc)
 842 {
 843         int err;
 844         unsigned long started = jiffies;  /* note the start time */
 845         struct dentry *root;
 846
 847         dout("mount start %p\n", fsc);
 848         mutex_lock(&fsc->client->mount_mutex);
 849
 850         if (!fsc->sb->s_root) {
 851                 const char *path = fsc->mount_options->server_path ?
 852                                      fsc->mount_options->server_path + 1 : "";
 853
 854                 err = __ceph_open_session(fsc->client, started);
 855                 if (err < 0)
 856                         goto out;
 857
 858                 dout("mount opening path '%s'\n", path);
 859
 860                 err = ceph_fs_debugfs_init(fsc);
 861                 if (err < 0)
 862                         goto out;
 863
 864                 root = open_root_dentry(fsc, path, started);
 865                 if (IS_ERR(root)) {
 866                         err = PTR_ERR(root);
 867                         goto out;
 868                 }
 869                 fsc->sb->s_root = dget(root);
 870         } else {
 871                 root = dget(fsc->sb->s_root);
 872         }
 873
 874         fsc->mount_state = CEPH_MOUNT_MOUNTED;
 875         dout("mount success\n");
 876         mutex_unlock(&fsc->client->mount_mutex);
 877         return root;
 878
 879 out:
 880         mutex_unlock(&fsc->client->mount_mutex);
 881         return ERR_PTR(err);
 882 }
 883
 884 static int ceph_set_super(struct super_block *s, void *data)
 885 {
 886         struct ceph_fs_client *fsc = data;
 887         int ret;
 888
 889         dout("set_super %p data %p\n", s, data);
 890
 891         s->s_flags = fsc->mount_options->sb_flags;
 892         s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
 893
 894         s->s_xattr = ceph_xattr_handlers;
 895         s->s_fs_info = fsc;
 896         fsc->sb = s;
 897
 898         s->s_op = &ceph_super_ops;
 899         s->s_export_op = &ceph_export_ops;
 900
 901         s->s_time_gran = 1000;  /* 1000 ns == 1 us */
 902
 903         ret = set_anon_super(s, NULL);  /* what is that second arg for? */
 904         if (ret != 0)
 905                 goto fail;
 906
 907         return ret;
 908
 909 fail:
 910         s->s_fs_info = NULL;
 911         fsc->sb = NULL;
 912         return ret;
 913 }
 914
 915 /*
 916  * share superblock if same fs AND options
 917  */
 918 static int ceph_compare_super(struct super_block *sb, void *data)
 919 {
 920         struct ceph_fs_client *new = data;
 921         struct ceph_mount_options *fsopt = new->mount_options;
 922         struct ceph_options *opt = new->client->options;
 923         struct ceph_fs_client *other = ceph_sb_to_client(sb);
 924
 925         dout("ceph_compare_super %p\n", sb);
 926
 927         if (compare_mount_options(fsopt, opt, other)) {
 928                 dout("monitor(s)/mount options don't match\n");
 929                 return 0;
 930         }
 931         if ((opt->flags & CEPH_OPT_FSID) &&
 932             ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
 933                 dout("fsid doesn't match\n");
 934                 return 0;
 935         }
 936         if (fsopt->sb_flags != other->mount_options->sb_flags) {
 937                 dout("flags differ\n");
 938                 return 0;
 939         }
 940         return 1;
 941 }
 942
 943 /*
 944  * construct our own bdi so we can control readahead, etc.
 945  */
 946 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
 947
 948 static int ceph_register_bdi(struct super_block *sb,
 949                              struct ceph_fs_client *fsc)
 950 {
 951         int err;
 952
 953         /* set ra_pages based on rasize mount option? */
 954         if (fsc->mount_options->rasize >= PAGE_SIZE)
 955                 fsc->backing_dev_info.ra_pages =
 956                         (fsc->mount_options->rasize + PAGE_SIZE - 1)
 957                         >> PAGE_SHIFT;
 958         else
 959                 fsc->backing_dev_info.ra_pages =
 960                         VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
 961
 962         err = bdi_register(&fsc->backing_dev_info, NULL, "ceph-%ld",
 963                            atomic_long_inc_return(&bdi_seq));
 964         if (!err)
 965                 sb->s_bdi = &fsc->backing_dev_info;
 966         return err;
 967 }
 968
 969 static struct dentry *ceph_mount(struct file_system_type *fs_type,
 970                        int flags, const char *dev_name, void *data)
 971 {
 972         struct super_block *sb;
 973         struct ceph_fs_client *fsc;
 974         struct dentry *res;
 975         int err;
 976         int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
 977         struct ceph_mount_options *fsopt = NULL;
 978         struct ceph_options *opt = NULL;
 979
 980         dout("ceph_mount\n");
 981
 982 #ifdef CONFIG_CEPH_FS_POSIX_ACL
 983         flags |= MS_POSIXACL;
 984 #endif
 985         err = parse_mount_options(&fsopt, &opt, flags, data, dev_name);
 986         if (err < 0) {
 987                 res = ERR_PTR(err);
 988                 goto out_final;
 989         }
 990
 991         /* create client (which we may/may not use) */
 992         fsc = create_fs_client(fsopt, opt);
 993         if (IS_ERR(fsc)) {
 994                 res = ERR_CAST(fsc);
 995                 destroy_mount_options(fsopt);
 996                 ceph_destroy_options(opt);
 997                 goto out_final;
 998         }
 999
1000         err = ceph_mdsc_init(fsc);
1001         if (err < 0) {
1002                 res = ERR_PTR(err);
1003                 goto out;
1004         }
1005
1006         if (ceph_test_opt(fsc->client, NOSHARE))
1007                 compare_super = NULL;
1008         sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc);
1009         if (IS_ERR(sb)) {
1010                 res = ERR_CAST(sb);
1011                 goto out;
1012         }
1013
1014         if (ceph_sb_to_client(sb) != fsc) {
1015                 ceph_mdsc_destroy(fsc);
1016                 destroy_fs_client(fsc);
1017                 fsc = ceph_sb_to_client(sb);
1018                 dout("get_sb got existing client %p\n", fsc);
1019         } else {
1020                 dout("get_sb using new client %p\n", fsc);
1021                 err = ceph_register_bdi(sb, fsc);
1022                 if (err < 0) {
1023                         res = ERR_PTR(err);
1024                         goto out_splat;
1025                 }
1026         }
1027
1028         res = ceph_real_mount(fsc);
1029         if (IS_ERR(res))
1030                 goto out_splat;
1031         dout("root %p inode %p ino %llx.%llx\n", res,
1032              d_inode(res), ceph_vinop(d_inode(res)));
1033         return res;
1034
1035 out_splat:
1036         ceph_mdsc_close_sessions(fsc->mdsc);
1037         deactivate_locked_super(sb);
1038         goto out_final;
1039
1040 out:
1041         ceph_mdsc_destroy(fsc);
1042         destroy_fs_client(fsc);
1043 out_final:
1044         dout("ceph_mount fail %ld\n", PTR_ERR(res));
1045         return res;
1046 }
1047
1048 static void ceph_kill_sb(struct super_block *s)
1049 {
1050         struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1051         dev_t dev = s->s_dev;
1052
1053         dout("kill_sb %p\n", s);
1054
1055         ceph_mdsc_pre_umount(fsc->mdsc);
1056         generic_shutdown_super(s);
1057         ceph_mdsc_destroy(fsc);
1058
1059         destroy_fs_client(fsc);
1060         free_anon_bdev(dev);
1061 }
1062
1063 static struct file_system_type ceph_fs_type = {
1064         .owner          = THIS_MODULE,
1065         .name           = "ceph",
1066         .mount          = ceph_mount,
1067         .kill_sb        = ceph_kill_sb,
1068         .fs_flags       = FS_RENAME_DOES_D_MOVE,
1069 };
1070 MODULE_ALIAS_FS("ceph");
1071
1072 static int __init init_ceph(void)
1073 {
1074         int ret = init_caches();
1075         if (ret)
1076                 goto out;
1077
1078         ceph_flock_init();
1079         ceph_xattr_init();
1080         ret = register_filesystem(&ceph_fs_type);
1081         if (ret)
1082                 goto out_xattr;
1083
1084         pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1085
1086         return 0;
1087
1088 out_xattr:
1089         ceph_xattr_exit();
1090         destroy_caches();
1091 out:
1092         return ret;
1093 }
1094
1095 static void __exit exit_ceph(void)
1096 {
1097         dout("exit_ceph\n");
1098         unregister_filesystem(&ceph_fs_type);
1099         ceph_xattr_exit();
1100         destroy_caches();
1101 }
1102
1103 module_init(init_ceph);
1104 module_exit(exit_ceph);
1105
1106 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1107 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1108 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1109 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1110 MODULE_LICENSE("GPL");