fs/orangefs/super.c

   1 /*
   2  * (C) 2001 Clemson University and The University of Chicago
   3  *
   4  * See COPYING in top-level directory.
   5  */
   6
   7 #include "protocol.h"
   8 #include "orangefs-kernel.h"
   9 #include "orangefs-bufmap.h"
  10
  11 #include <linux/parser.h>
  12
  13 /* a cache for orangefs-inode objects (i.e. orangefs inode private data) */
  14 static struct kmem_cache *orangefs_inode_cache;
  15
  16 /* list for storing orangefs specific superblocks in use */
  17 LIST_HEAD(orangefs_superblocks);
  18
  19 DEFINE_SPINLOCK(orangefs_superblocks_lock);
  20
  21 enum {
  22         Opt_intr,
  23         Opt_acl,
  24         Opt_local_lock,
  25
  26         Opt_err
  27 };
  28
  29 static const match_table_t tokens = {
  30         { Opt_acl,              "acl" },
  31         { Opt_intr,             "intr" },
  32         { Opt_local_lock,       "local_lock" },
  33         { Opt_err,      NULL }
  34 };
  35
  36
  37 static int parse_mount_options(struct super_block *sb, char *options,
  38                 int silent)
  39 {
  40         struct orangefs_sb_info_s *orangefs_sb = ORANGEFS_SB(sb);
  41         substring_t args[MAX_OPT_ARGS];
  42         char *p;
  43
  44         /*
  45          * Force any potential flags that might be set from the mount
  46          * to zero, ie, initialize to unset.
  47          */
  48         sb->s_flags &= ~MS_POSIXACL;
  49         orangefs_sb->flags &= ~ORANGEFS_OPT_INTR;
  50         orangefs_sb->flags &= ~ORANGEFS_OPT_LOCAL_LOCK;
  51
  52         while ((p = strsep(&options, ",")) != NULL) {
  53                 int token;
  54
  55                 if (!*p)
  56                         continue;
  57
  58                 token = match_token(p, tokens, args);
  59                 switch (token) {
  60                 case Opt_acl:
  61                         sb->s_flags |= MS_POSIXACL;
  62                         break;
  63                 case Opt_intr:
  64                         orangefs_sb->flags |= ORANGEFS_OPT_INTR;
  65                         break;
  66                 case Opt_local_lock:
  67                         orangefs_sb->flags |= ORANGEFS_OPT_LOCAL_LOCK;
  68                         break;
  69                 default:
  70                         goto fail;
  71                 }
  72         }
  73
  74         return 0;
  75 fail:
  76         if (!silent)
  77                 gossip_err("Error: mount option [%s] is not supported.\n", p);
  78         return -EINVAL;
  79 }
  80
  81 static void orangefs_inode_cache_ctor(void *req)
  82 {
  83         struct orangefs_inode_s *orangefs_inode = req;
  84
  85         inode_init_once(&orangefs_inode->vfs_inode);
  86         init_rwsem(&orangefs_inode->xattr_sem);
  87
  88         orangefs_inode->vfs_inode.i_version = 1;
  89 }
  90
  91 static struct inode *orangefs_alloc_inode(struct super_block *sb)
  92 {
  93         struct orangefs_inode_s *orangefs_inode;
  94
  95         orangefs_inode = kmem_cache_alloc(orangefs_inode_cache,
  96                                        ORANGEFS_CACHE_ALLOC_FLAGS);
  97         if (orangefs_inode == NULL) {
  98                 gossip_err("Failed to allocate orangefs_inode\n");
  99                 return NULL;
 100         }
 101
 102         /*
 103          * We want to clear everything except for rw_semaphore and the
 104          * vfs_inode.
 105          */
 106         memset(&orangefs_inode->refn.khandle, 0, 16);
 107         orangefs_inode->refn.fs_id = ORANGEFS_FS_ID_NULL;
 108         orangefs_inode->last_failed_block_index_read = 0;
 109         memset(orangefs_inode->link_target, 0, sizeof(orangefs_inode->link_target));
 110         orangefs_inode->pinode_flags = 0;
 111
 112         gossip_debug(GOSSIP_SUPER_DEBUG,
 113                      "orangefs_alloc_inode: allocated %p\n",
 114                      &orangefs_inode->vfs_inode);
 115         return &orangefs_inode->vfs_inode;
 116 }
 117
 118 static void orangefs_destroy_inode(struct inode *inode)
 119 {
 120         struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
 121
 122         gossip_debug(GOSSIP_SUPER_DEBUG,
 123                         "%s: deallocated %p destroying inode %pU\n",
 124                         __func__, orangefs_inode, get_khandle_from_ino(inode));
 125
 126         kmem_cache_free(orangefs_inode_cache, orangefs_inode);
 127 }
 128
 129 /*
 130  * NOTE: information filled in here is typically reflected in the
 131  * output of the system command 'df'
 132 */
 133 static int orangefs_statfs(struct dentry *dentry, struct kstatfs *buf)
 134 {
 135         int ret = -ENOMEM;
 136         struct orangefs_kernel_op_s *new_op = NULL;
 137         int flags = 0;
 138         struct super_block *sb = NULL;
 139
 140         sb = dentry->d_sb;
 141
 142         gossip_debug(GOSSIP_SUPER_DEBUG,
 143                      "orangefs_statfs: called on sb %p (fs_id is %d)\n",
 144                      sb,
 145                      (int)(ORANGEFS_SB(sb)->fs_id));
 146
 147         new_op = op_alloc(ORANGEFS_VFS_OP_STATFS);
 148         if (!new_op)
 149                 return ret;
 150         new_op->upcall.req.statfs.fs_id = ORANGEFS_SB(sb)->fs_id;
 151
 152         if (ORANGEFS_SB(sb)->flags & ORANGEFS_OPT_INTR)
 153                 flags = ORANGEFS_OP_INTERRUPTIBLE;
 154
 155         ret = service_operation(new_op, "orangefs_statfs", flags);
 156
 157         if (new_op->downcall.status < 0)
 158                 goto out_op_release;
 159
 160         gossip_debug(GOSSIP_SUPER_DEBUG,
 161                      "%s: got %ld blocks available | "
 162                      "%ld blocks total | %ld block size | "
 163                      "%ld files total | %ld files avail\n",
 164                      __func__,
 165                      (long)new_op->downcall.resp.statfs.blocks_avail,
 166                      (long)new_op->downcall.resp.statfs.blocks_total,
 167                      (long)new_op->downcall.resp.statfs.block_size,
 168                      (long)new_op->downcall.resp.statfs.files_total,
 169                      (long)new_op->downcall.resp.statfs.files_avail);
 170
 171         buf->f_type = sb->s_magic;
 172         memcpy(&buf->f_fsid, &ORANGEFS_SB(sb)->fs_id, sizeof(buf->f_fsid));
 173         buf->f_bsize = new_op->downcall.resp.statfs.block_size;
 174         buf->f_namelen = ORANGEFS_NAME_LEN;
 175
 176         buf->f_blocks = (sector_t) new_op->downcall.resp.statfs.blocks_total;
 177         buf->f_bfree = (sector_t) new_op->downcall.resp.statfs.blocks_avail;
 178         buf->f_bavail = (sector_t) new_op->downcall.resp.statfs.blocks_avail;
 179         buf->f_files = (sector_t) new_op->downcall.resp.statfs.files_total;
 180         buf->f_ffree = (sector_t) new_op->downcall.resp.statfs.files_avail;
 181         buf->f_frsize = sb->s_blocksize;
 182
 183 out_op_release:
 184         op_release(new_op);
 185         gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_statfs: returning %d\n", ret);
 186         return ret;
 187 }
 188
 189 /*
 190  * Remount as initiated by VFS layer.  We just need to reparse the mount
 191  * options, no need to signal pvfs2-client-core about it.
 192  */
 193 static int orangefs_remount_fs(struct super_block *sb, int *flags, char *data)
 194 {
 195         gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_remount_fs: called\n");
 196         return parse_mount_options(sb, data, 1);
 197 }
 198
 199 /*
 200  * Remount as initiated by pvfs2-client-core on restart.  This is used to
 201  * repopulate mount information left from previous pvfs2-client-core.
 202  *
 203  * the idea here is that given a valid superblock, we're
 204  * re-initializing the user space client with the initial mount
 205  * information specified when the super block was first initialized.
 206  * this is very different than the first initialization/creation of a
 207  * superblock.  we use the special service_priority_operation to make
 208  * sure that the mount gets ahead of any other pending operation that
 209  * is waiting for servicing.  this means that the pvfs2-client won't
 210  * fail to start several times for all other pending operations before
 211  * the client regains all of the mount information from us.
 212  * NOTE: this function assumes that the request_mutex is already acquired!
 213  */
 214 int orangefs_remount(struct super_block *sb)
 215 {
 216         struct orangefs_kernel_op_s *new_op;
 217         int ret = -EINVAL;
 218
 219         gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_remount: called\n");
 220
 221         new_op = op_alloc(ORANGEFS_VFS_OP_FS_MOUNT);
 222         if (!new_op)
 223                 return -ENOMEM;
 224         strncpy(new_op->upcall.req.fs_mount.orangefs_config_server,
 225                 ORANGEFS_SB(sb)->devname,
 226                 ORANGEFS_MAX_SERVER_ADDR_LEN);
 227
 228         gossip_debug(GOSSIP_SUPER_DEBUG,
 229                      "Attempting ORANGEFS Remount via host %s\n",
 230                      new_op->upcall.req.fs_mount.orangefs_config_server);
 231
 232         /*
 233          * we assume that the calling function has already acquire the
 234          * request_mutex to prevent other operations from bypassing
 235          * this one
 236          */
 237         ret = service_operation(new_op, "orangefs_remount",
 238                 ORANGEFS_OP_PRIORITY | ORANGEFS_OP_NO_SEMAPHORE);
 239         gossip_debug(GOSSIP_SUPER_DEBUG,
 240                      "orangefs_remount: mount got return value of %d\n",
 241                      ret);
 242         if (ret == 0) {
 243                 /*
 244                  * store the id assigned to this sb -- it's just a
 245                  * short-lived mapping that the system interface uses
 246                  * to map this superblock to a particular mount entry
 247                  */
 248                 ORANGEFS_SB(sb)->id = new_op->downcall.resp.fs_mount.id;
 249                 ORANGEFS_SB(sb)->mount_pending = 0;
 250         }
 251
 252         op_release(new_op);
 253         return ret;
 254 }
 255
 256 int fsid_key_table_initialize(void)
 257 {
 258         return 0;
 259 }
 260
 261 void fsid_key_table_finalize(void)
 262 {
 263 }
 264
 265 /* Called whenever the VFS dirties the inode in response to atime updates */
 266 static void orangefs_dirty_inode(struct inode *inode, int flags)
 267 {
 268         struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
 269
 270         gossip_debug(GOSSIP_SUPER_DEBUG,
 271                      "orangefs_dirty_inode: %pU\n",
 272                      get_khandle_from_ino(inode));
 273         SetAtimeFlag(orangefs_inode);
 274 }
 275
 276 static const struct super_operations orangefs_s_ops = {
 277         .alloc_inode = orangefs_alloc_inode,
 278         .destroy_inode = orangefs_destroy_inode,
 279         .dirty_inode = orangefs_dirty_inode,
 280         .drop_inode = generic_delete_inode,
 281         .statfs = orangefs_statfs,
 282         .remount_fs = orangefs_remount_fs,
 283         .show_options = generic_show_options,
 284 };
 285
 286 static struct dentry *orangefs_fh_to_dentry(struct super_block *sb,
 287                                   struct fid *fid,
 288                                   int fh_len,
 289                                   int fh_type)
 290 {
 291         struct orangefs_object_kref refn;
 292
 293         if (fh_len < 5 || fh_type > 2)
 294                 return NULL;
 295
 296         ORANGEFS_khandle_from(&(refn.khandle), fid->raw, 16);
 297         refn.fs_id = (u32) fid->raw[4];
 298         gossip_debug(GOSSIP_SUPER_DEBUG,
 299                      "fh_to_dentry: handle %pU, fs_id %d\n",
 300                      &refn.khandle,
 301                      refn.fs_id);
 302
 303         return d_obtain_alias(orangefs_iget(sb, &refn));
 304 }
 305
 306 static int orangefs_encode_fh(struct inode *inode,
 307                     __u32 *fh,
 308                     int *max_len,
 309                     struct inode *parent)
 310 {
 311         int len = parent ? 10 : 5;
 312         int type = 1;
 313         struct orangefs_object_kref refn;
 314
 315         if (*max_len < len) {
 316                 gossip_lerr("fh buffer is too small for encoding\n");
 317                 *max_len = len;
 318                 type = 255;
 319                 goto out;
 320         }
 321
 322         refn = ORANGEFS_I(inode)->refn;
 323         ORANGEFS_khandle_to(&refn.khandle, fh, 16);
 324         fh[4] = refn.fs_id;
 325
 326         gossip_debug(GOSSIP_SUPER_DEBUG,
 327                      "Encoding fh: handle %pU, fsid %u\n",
 328                      &refn.khandle,
 329                      refn.fs_id);
 330
 331
 332         if (parent) {
 333                 refn = ORANGEFS_I(parent)->refn;
 334                 ORANGEFS_khandle_to(&refn.khandle, (char *) fh + 20, 16);
 335                 fh[9] = refn.fs_id;
 336
 337                 type = 2;
 338                 gossip_debug(GOSSIP_SUPER_DEBUG,
 339                              "Encoding parent: handle %pU, fsid %u\n",
 340                              &refn.khandle,
 341                              refn.fs_id);
 342         }
 343         *max_len = len;
 344
 345 out:
 346         return type;
 347 }
 348
 349 static const struct export_operations orangefs_export_ops = {
 350         .encode_fh = orangefs_encode_fh,
 351         .fh_to_dentry = orangefs_fh_to_dentry,
 352 };
 353
 354 static int orangefs_fill_sb(struct super_block *sb,
 355                 struct orangefs_fs_mount_response *fs_mount,
 356                 void *data, int silent)
 357 {
 358         int ret = -EINVAL;
 359         struct inode *root = NULL;
 360         struct dentry *root_dentry = NULL;
 361         struct orangefs_object_kref root_object;
 362
 363         /* alloc and init our private orangefs sb info */
 364         sb->s_fs_info =
 365                 kzalloc(sizeof(struct orangefs_sb_info_s), ORANGEFS_GFP_FLAGS);
 366         if (!ORANGEFS_SB(sb))
 367                 return -ENOMEM;
 368         ORANGEFS_SB(sb)->sb = sb;
 369
 370         ORANGEFS_SB(sb)->root_khandle = fs_mount->root_khandle;
 371         ORANGEFS_SB(sb)->fs_id = fs_mount->fs_id;
 372         ORANGEFS_SB(sb)->id = fs_mount->id;
 373
 374         if (data) {
 375                 ret = parse_mount_options(sb, data, silent);
 376                 if (ret)
 377                         return ret;
 378         }
 379
 380         /* Hang the xattr handlers off the superblock */
 381         sb->s_xattr = orangefs_xattr_handlers;
 382         sb->s_magic = ORANGEFS_SUPER_MAGIC;
 383         sb->s_op = &orangefs_s_ops;
 384         sb->s_d_op = &orangefs_dentry_operations;
 385
 386         sb->s_blocksize = orangefs_bufmap_size_query();
 387         sb->s_blocksize_bits = orangefs_bufmap_shift_query();
 388         sb->s_maxbytes = MAX_LFS_FILESIZE;
 389
 390         root_object.khandle = ORANGEFS_SB(sb)->root_khandle;
 391         root_object.fs_id = ORANGEFS_SB(sb)->fs_id;
 392         gossip_debug(GOSSIP_SUPER_DEBUG,
 393                      "get inode %pU, fsid %d\n",
 394                      &root_object.khandle,
 395                      root_object.fs_id);
 396
 397         root = orangefs_iget(sb, &root_object);
 398         if (IS_ERR(root))
 399                 return PTR_ERR(root);
 400
 401         gossip_debug(GOSSIP_SUPER_DEBUG,
 402                      "Allocated root inode [%p] with mode %x\n",
 403                      root,
 404                      root->i_mode);
 405
 406         /* allocates and places root dentry in dcache */
 407         root_dentry = d_make_root(root);
 408         if (!root_dentry)
 409                 return -ENOMEM;
 410
 411         sb->s_export_op = &orangefs_export_ops;
 412         sb->s_root = root_dentry;
 413         return 0;
 414 }
 415
 416 struct dentry *orangefs_mount(struct file_system_type *fst,
 417                            int flags,
 418                            const char *devname,
 419                            void *data)
 420 {
 421         int ret = -EINVAL;
 422         struct super_block *sb = ERR_PTR(-EINVAL);
 423         struct orangefs_kernel_op_s *new_op;
 424         struct dentry *d = ERR_PTR(-EINVAL);
 425
 426         gossip_debug(GOSSIP_SUPER_DEBUG,
 427                      "orangefs_mount: called with devname %s\n",
 428                      devname);
 429
 430         if (!devname) {
 431                 gossip_err("ERROR: device name not specified.\n");
 432                 return ERR_PTR(-EINVAL);
 433         }
 434
 435         new_op = op_alloc(ORANGEFS_VFS_OP_FS_MOUNT);
 436         if (!new_op)
 437                 return ERR_PTR(-ENOMEM);
 438
 439         strncpy(new_op->upcall.req.fs_mount.orangefs_config_server,
 440                 devname,
 441                 ORANGEFS_MAX_SERVER_ADDR_LEN);
 442
 443         gossip_debug(GOSSIP_SUPER_DEBUG,
 444                      "Attempting ORANGEFS Mount via host %s\n",
 445                      new_op->upcall.req.fs_mount.orangefs_config_server);
 446
 447         ret = service_operation(new_op, "orangefs_mount", 0);
 448         gossip_debug(GOSSIP_SUPER_DEBUG,
 449                      "orangefs_mount: mount got return value of %d\n", ret);
 450         if (ret)
 451                 goto free_op;
 452
 453         if (new_op->downcall.resp.fs_mount.fs_id == ORANGEFS_FS_ID_NULL) {
 454                 gossip_err("ERROR: Retrieved null fs_id\n");
 455                 ret = -EINVAL;
 456                 goto free_op;
 457         }
 458
 459         sb = sget(fst, NULL, set_anon_super, flags, NULL);
 460
 461         if (IS_ERR(sb)) {
 462                 d = ERR_CAST(sb);
 463                 goto free_op;
 464         }
 465
 466         ret = orangefs_fill_sb(sb,
 467               &new_op->downcall.resp.fs_mount, data,
 468               flags & MS_SILENT ? 1 : 0);
 469
 470         if (ret) {
 471                 d = ERR_PTR(ret);
 472                 goto free_op;
 473         }
 474
 475         /*
 476          * on successful mount, store the devname and data
 477          * used
 478          */
 479         strncpy(ORANGEFS_SB(sb)->devname,
 480                 devname,
 481                 ORANGEFS_MAX_SERVER_ADDR_LEN);
 482
 483         /* mount_pending must be cleared */
 484         ORANGEFS_SB(sb)->mount_pending = 0;
 485
 486         /*
 487          * finally, add this sb to our list of known orangefs
 488          * sb's
 489          */
 490         add_orangefs_sb(sb);
 491         op_release(new_op);
 492         return dget(sb->s_root);
 493
 494 free_op:
 495         gossip_err("orangefs_mount: mount request failed with %d\n", ret);
 496         if (ret == -EINVAL) {
 497                 gossip_err("Ensure that all orangefs-servers have the same FS configuration files\n");
 498                 gossip_err("Look at pvfs2-client-core log file (typically /tmp/pvfs2-client.log) for more details\n");
 499         }
 500
 501         op_release(new_op);
 502
 503         return d;
 504 }
 505
 506 void orangefs_kill_sb(struct super_block *sb)
 507 {
 508         gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_kill_sb: called\n");
 509
 510         /*
 511          * issue the unmount to userspace to tell it to remove the
 512          * dynamic mount info it has for this superblock
 513          */
 514         orangefs_unmount_sb(sb);
 515
 516         /* remove the sb from our list of orangefs specific sb's */
 517         remove_orangefs_sb(sb);
 518
 519         /* provided sb cleanup */
 520         kill_anon_super(sb);
 521
 522         /* free the orangefs superblock private data */
 523         kfree(ORANGEFS_SB(sb));
 524 }
 525
 526 int orangefs_inode_cache_initialize(void)
 527 {
 528         orangefs_inode_cache = kmem_cache_create("orangefs_inode_cache",
 529                                               sizeof(struct orangefs_inode_s),
 530                                               0,
 531                                               ORANGEFS_CACHE_CREATE_FLAGS,
 532                                               orangefs_inode_cache_ctor);
 533
 534         if (!orangefs_inode_cache) {
 535                 gossip_err("Cannot create orangefs_inode_cache\n");
 536                 return -ENOMEM;
 537         }
 538         return 0;
 539 }
 540
 541 int orangefs_inode_cache_finalize(void)
 542 {
 543         kmem_cache_destroy(orangefs_inode_cache);
 544         return 0;
 545 }