lustre/obdfs/super.c

   1 /*
   2  * OBDFS Super operations
   3  *
   4  * This code is issued under the GNU General Public License.
   5  * See the file COPYING in this distribution
   6  *
   7  * Copryright (C) 1996 Peter J. Braam <braam@stelias.com>
   8  * Copryright (C) 1999 Stelias Computing Inc. <braam@stelias.com>
   9  * Copryright (C) 1999 Seagate Technology Inc.
  10  * Copryright (C) 2001 Mountain View Data, Inc.
  11  *
  12  */
  13
  14 #define EXPORT_SYMTAB
  15
  16 #include <linux/config.h>
  17 #include <linux/module.h>
  18 #include <linux/kernel.h>
  19 #include <linux/mm.h>
  20 #include <linux/string.h>
  21 #include <linux/stat.h>
  22 #include <linux/errno.h>
  23 #include <linux/locks.h>
  24 #include <linux/unistd.h>
  25
  26 #include <asm/system.h>
  27 #include <asm/uaccess.h>
  28
  29 #include <linux/fs.h>
  30 #include <linux/stat.h>
  31 #include <asm/uaccess.h>
  32 #include <linux/vmalloc.h>
  33 #include <asm/segment.h>
  34
  35 #include <linux/obd_support.h>
  36 #include <linux/obd_class.h>
  37 #include <linux/obdfs.h>
  38
  39 struct list_head obdfs_super_list;
  40 struct address_space_operations obdfs_aops;
  41 struct super_operations obdfs_super_operations;
  42 long obdfs_cache_count = 0;
  43 long obdfs_mutex_start = 0;
  44 long obd_memory = 0;
  45
  46 static char *obdfs_read_opt(const char *opt, char *data)
  47 {
  48         char *value;
  49         char *retval;
  50
  51         CDEBUG(D_INFO, "option: %s, data %s\n", opt, data);
  52         if ( strncmp(opt, data, strlen(opt)) )
  53                 return NULL;
  54
  55         if ( (value = strchr(data, '=')) == NULL )
  56                 return NULL;
  57
  58         value++;
  59         OBD_ALLOC(retval, char *, strlen(value) + 1);
  60         if ( !retval ) {
  61                 printk(KERN_ALERT __FUNCTION__ ": out of memory!\n");
  62                 return NULL;
  63         }
  64
  65         memcpy(retval, value, strlen(value)+1);
  66         CDEBUG(D_PSDEV, "Assigned option: %s, value %s\n", opt, retval);
  67         return retval;
  68 }
  69
  70 static void obdfs_options(char *options, char **dev, char **vers)
  71 {
  72         char *this_char;
  73
  74         if (!options)
  75                 return;
  76
  77         for (this_char = strtok (options, ",");
  78              this_char != NULL;
  79              this_char = strtok (NULL, ",")) {
  80                 CDEBUG(D_INFO, "this_char %s\n", this_char);
  81                 if ( (!*dev && (*dev = obdfs_read_opt("device", this_char)))||
  82                      (!*vers && (*vers = obdfs_read_opt("version", this_char))) )
  83                         continue;
  84
  85         }
  86 }
  87
  88 static int obdfs_getdev(char *devpath, int *dev)
  89 {
  90         struct dentry *dentry;
  91         kdev_t devno;
  92         struct nameidata nd;
  93         int error = 0;
  94
  95         ENTRY;
  96         if (path_init(devpath, LOOKUP_POSITIVE, &nd))
  97                 error = path_walk(devpath, &nd);
  98         if (error)
  99                 return error;
 100
 101         dentry = nd.dentry;
 102         if (!S_ISCHR(dentry->d_inode->i_mode))
 103                 return -ENODEV;
 104
 105         devno = dentry->d_inode->i_rdev;
 106         if ( MAJOR(devno) != OBD_PSDEV_MAJOR )
 107                 return -ENODEV;
 108
 109         if ( MINOR(devno) >= MAX_OBD_DEVICES )
 110                 return -ENODEV;
 111
 112         *dev = devno;
 113         return 0;
 114 }
 115
 116
 117 static struct super_block * obdfs_read_super(struct super_block *sb,
 118                                             void *data, int silent)
 119 {
 120         struct inode *root = 0;
 121         struct obdfs_sb_info *sbi = (struct obdfs_sb_info *)(&sb->u.generic_sbp);
 122         struct obd_device *obddev;
 123         char *device = NULL;
 124         char *version = NULL;
 125         int devno;
 126         int err;
 127         unsigned long blocksize;
 128         unsigned long blocksize_bits;
 129         unsigned long root_ino;
 130         int scratch;
 131
 132
 133         ENTRY;
 134         MOD_INC_USE_COUNT;
 135
 136         memset(sbi, 0, sizeof(*sbi));
 137
 138         CDEBUG(D_INFO, "\n");
 139         obdfs_options(data, &device, &version);
 140         if ( !device ) {
 141                 printk(__FUNCTION__ ": no device\n");
 142                 EXIT;
 143                 goto ERR;
 144         }
 145
 146         CDEBUG(D_INFO, "\n");
 147         if ( (err = obdfs_getdev(device, &devno)) ) {
 148                 printk("Cannot get devno of %s, error %d\n", device, err);
 149                 EXIT;
 150                 goto ERR;;
 151         }
 152
 153         CDEBUG(D_INFO, "\n");
 154         if ( MAJOR(devno) != OBD_PSDEV_MAJOR ) {
 155                 printk(__FUNCTION__ ": wrong major number %d!\n", MAJOR(devno));
 156                 EXIT;
 157                 goto ERR;
 158         }
 159
 160         CDEBUG(D_INFO, "\n");
 161         if ( MINOR(devno) >= MAX_OBD_DEVICES ) {
 162                 printk(__FUNCTION__ ": minor of %s too high (%d)\n",
 163                        device, MINOR(devno));
 164                 EXIT;
 165                 goto ERR;
 166         }
 167
 168         CDEBUG(D_INFO, "\n");
 169         obddev = &obd_dev[MINOR(devno)];
 170
 171         CDEBUG(D_INFO, "\n");
 172         if ( ! (obddev->obd_flags & OBD_ATTACHED) ||
 173              ! (obddev->obd_flags & OBD_SET_UP) ){
 174                 printk("device %s not attached or not set up (%d)\n",
 175                        device, MINOR(devno));
 176                 EXIT;
 177                 goto ERR;;
 178         }
 179
 180         CDEBUG(D_INFO, "\n");
 181         sbi->osi_obd = obddev;
 182         sbi->osi_ops = sbi->osi_obd->obd_type->typ_ops;
 183
 184         sbi->osi_conn.oc_dev = obddev;
 185         err = sbi->osi_ops->o_connect(&sbi->osi_conn);
 186         if ( err ) {
 187                 printk("OBDFS: cannot connect to %s\n", device);
 188                 EXIT;
 189                 goto ERR;
 190         }
 191
 192         CDEBUG(D_INFO, "\n");
 193         /* list of dirty inodes, and a mutex to hold while modifying it */
 194         INIT_LIST_HEAD(&sbi->osi_inodes);
 195         sema_init(&sbi->osi_list_mutex, 1);
 196
 197         CDEBUG(D_INFO, "\n");
 198         sbi->osi_super = sb;
 199
 200         CDEBUG(D_INFO, "\n");
 201         err = sbi->osi_ops->o_get_info(&sbi->osi_conn, strlen("blocksize"),
 202                                        "blocksize", &scratch,
 203                                        (void *)&blocksize);
 204         if ( err ) {
 205                 printk("getinfo call to drive failed (blocksize)\n");
 206                 EXIT;
 207                 goto ERR;
 208         }
 209
 210         CDEBUG(D_INFO, "\n");
 211         err = sbi->osi_ops->o_get_info(&sbi->osi_conn, strlen("blocksize_bits"),
 212                                        "blocksize_bits", &scratch,
 213                                        (void *)&blocksize_bits);
 214         if ( err ) {
 215                 printk("getinfo call to drive failed (blocksize_bits)\n");
 216                 EXIT;
 217                 goto ERR;
 218         }
 219
 220         CDEBUG(D_INFO, "\n");
 221         err = sbi->osi_ops->o_get_info(&sbi->osi_conn, strlen("root_ino"),
 222                                        "root_ino", &scratch, (void *)&root_ino);
 223         if ( err ) {
 224                 printk("getinfo call to drive failed (root_ino)\n");
 225                 EXIT;
 226                 goto ERR;
 227         }
 228
 229         CDEBUG(D_INFO, "\n");
 230         sb->s_blocksize = blocksize;
 231         sb->s_blocksize_bits = (unsigned char)blocksize_bits;
 232         sb->s_magic = OBDFS_SUPER_MAGIC;
 233         sb->s_op = &obdfs_super_operations;
 234
 235         /* XXX how to get "sb->s_flags |= MS_RDONLY" here for snapshots? */
 236
 237         /* make root inode */
 238         CDEBUG(D_INFO, "\n");
 239         root = iget(sb, root_ino);
 240         if (!root || is_bad_inode(root)) {
 241             printk("OBDFS: bad iget for root\n");
 242             sb->s_dev = 0;
 243             err = -ENOENT;
 244             EXIT;
 245             goto ERR;
 246         }
 247
 248         CDEBUG(D_INFO, "obdfs_read_super: sbdev %d, rootino: %ld, dev %s, "
 249                "minor: %d, blocksize: %ld, blocksize bits %ld\n",
 250                sb->s_dev, root->i_ino, device, MINOR(devno),
 251                blocksize, blocksize_bits);
 252         sb->s_root = d_alloc_root(root);
 253         list_add(&sbi->osi_list, &obdfs_super_list);
 254         OBD_FREE(device, strlen(device) + 1);
 255         if (version)
 256                 OBD_FREE(version, strlen(version) + 1);
 257         EXIT;
 258         return sb;
 259
 260 ERR:
 261         MOD_DEC_USE_COUNT;
 262         if (device)
 263                 OBD_FREE(device, strlen(device) + 1);
 264         if (version)
 265                 OBD_FREE(version, strlen(version) + 1);
 266         if (sbi) {
 267                 sbi->osi_super = NULL;
 268         }
 269         if (root) {
 270                 iput(root);
 271         }
 272         sb->s_dev = 0;
 273         return NULL;
 274 } /* obdfs_read_super */
 275
 276
 277 static void obdfs_put_super(struct super_block *sb)
 278 {
 279         struct obdfs_sb_info *sbi;
 280
 281         ENTRY;
 282         sb->s_dev = 0;
 283
 284         sbi = (struct obdfs_sb_info *) &sb->u.generic_sbp;
 285         obdfs_flush_reqs(&sbi->osi_inodes, ~0UL);
 286
 287         OPS(sb,disconnect)(ID(sb));
 288         list_del(&sbi->osi_list);
 289         memset(sbi, 0, sizeof(*sbi));
 290
 291         printk(KERN_INFO "OBDFS: Bye bye.\n");
 292
 293         MOD_DEC_USE_COUNT;
 294         EXIT;
 295 } /* obdfs_put_super */
 296
 297
 298 /* all filling in of inodes postponed until lookup */
 299 static void obdfs_read_inode(struct inode *inode)
 300 {
 301         struct obdo *oa;
 302         ENTRY;
 303         oa = obdo_fromid(IID(inode), inode->i_ino,
 304                          OBD_MD_FLNOTOBD | OBD_MD_FLBLOCKS);
 305         if ( IS_ERR(oa) ) {
 306                 printk(__FUNCTION__ ": obdo_fromid failed\n");
 307                 EXIT;
 308                 return /* PTR_ERR(oa) */;
 309         }
 310
 311         ODEBUG(oa);
 312         obdfs_to_inode(inode, oa);
 313         INIT_LIST_HEAD(obdfs_iplist(inode)); /* list of dirty pages on inode */
 314         INIT_LIST_HEAD(obdfs_islist(inode)); /* list of inodes in superblock */
 315
 316         obdo_free(oa);
 317         /* OIDEBUG(inode); */
 318
 319         if (S_ISREG(inode->i_mode)) {
 320                 inode->i_op = &obdfs_file_inode_operations;
 321                 inode->i_fop = &obdfs_file_operations;
 322                 inode->i_mapping->a_ops = &obdfs_aops;
 323                 EXIT;
 324         } else if (S_ISDIR(inode->i_mode)) {
 325                 inode->i_op = &obdfs_dir_inode_operations;
 326                 inode->i_fop = &obdfs_dir_operations;
 327                 EXIT;
 328         } else if (S_ISLNK(inode->i_mode)) {
 329                 if (inode->i_blocks) {
 330                         inode->i_op = &obdfs_symlink_inode_operations;
 331                         inode->i_mapping->a_ops = &obdfs_aops;
 332                 }else {
 333                         inode->i_op = &obdfs_fast_symlink_inode_operations;
 334                 }
 335                 EXIT;
 336         } else {
 337                 init_special_inode(inode, inode->i_mode,
 338                                    ((int *)obdfs_i2info(inode)->oi_inline)[0]);
 339         }
 340
 341         return;
 342 }
 343
 344 static void obdfs_write_inode(struct inode *inode, int wait)
 345 {
 346         struct obdo *oa;
 347         int err;
 348
 349         ENTRY;
 350         if (IOPS(inode, setattr) == NULL) {
 351                 printk(KERN_ERR __FUNCTION__ ": no setattr method!\n");
 352                 EXIT;
 353                 return;
 354         }
 355         oa = obdo_alloc();
 356         if ( !oa ) {
 357                 printk(__FUNCTION__ ": obdo_alloc failed\n");
 358                 EXIT;
 359                 return;
 360         }
 361
 362         oa->o_valid = OBD_MD_FLNOTOBD;
 363         obdfs_from_inode(oa, inode);
 364         err = IOPS(inode, setattr)(IID(inode), oa);
 365
 366         if ( err )
 367                 printk(__FUNCTION__ ": obd_setattr fails (%d)\n", err);
 368
 369         EXIT;
 370         obdo_free(oa);
 371 } /* obdfs_write_inode */
 372
 373
 374 /* This routine is called from iput() (for each unlink on the inode).
 375  * We can't put this call into delete_inode() since that is called only
 376  * when i_count == 0, and we need to keep a reference on the inode while
 377  * it is in the page cache, which means i_count > 0.  Catch 22.
 378  */
 379 static void obdfs_put_inode(struct inode *inode)
 380 {
 381         ENTRY;
 382         if (inode->i_nlink) {
 383                 EXIT;
 384                 return;
 385         }
 386
 387         obdfs_dequeue_pages(inode);
 388         EXIT;
 389 } /* obdfs_put_inode */
 390
 391
 392 static void obdfs_delete_inode(struct inode *inode)
 393 {
 394         struct obdo *oa;
 395         int err;
 396
 397         ENTRY;
 398         if (IOPS(inode, destroy) == NULL) {
 399                 printk(KERN_ERR __FUNCTION__ ": no destroy method!\n");
 400                 EXIT;
 401                 return;
 402         }
 403
 404         oa = obdo_alloc();
 405         if ( !oa ) {
 406                 printk(__FUNCTION__ ": obdo_alloc failed\n");
 407                 EXIT;
 408                 return;
 409         }
 410         oa->o_valid = OBD_MD_FLNOTOBD;
 411         obdfs_from_inode(oa, inode);
 412
 413         ODEBUG(oa);
 414         err = IOPS(inode, destroy)(IID(inode), oa);
 415         obdo_free(oa);
 416         clear_inode(inode);
 417         if (err) {
 418                 printk(__FUNCTION__ ": obd_destroy fails (%d)\n", err);
 419                 EXIT;
 420                 return;
 421         }
 422
 423         EXIT;
 424 } /* obdfs_delete_inode */
 425
 426
 427 int obdfs_notify_change(struct dentry *de, struct iattr *attr)
 428 {
 429         struct inode *inode = de->d_inode;
 430         struct obdo *oa;
 431         int err;
 432
 433         ENTRY;
 434         if (IOPS(inode, setattr) == NULL) {
 435                 printk(KERN_ERR __FUNCTION__ ": no setattr method!\n");
 436                 EXIT;
 437                 return -EIO;
 438         }
 439         oa = obdo_alloc();
 440         if ( !oa ) {
 441                 printk(__FUNCTION__ ": obdo_alloc failed\n");
 442                 return -ENOMEM;
 443         }
 444
 445         oa->o_id = inode->i_ino;
 446         obdo_from_iattr(oa, attr);
 447         err = IOPS(inode, setattr)(IID(inode), oa);
 448
 449         if ( err )
 450                 printk(__FUNCTION__ ": obd_setattr fails (%d)\n", err);
 451
 452         EXIT;
 453         obdo_free(oa);
 454         return err;
 455 } /* obdfs_notify_change */
 456
 457
 458 static int obdfs_statfs(struct super_block *sb, struct statfs *buf)
 459 {
 460         struct statfs tmp;
 461         int bufsize = sizeof(*buf);
 462         int err;
 463
 464         ENTRY;
 465
 466         err = OPS(sb,statfs)(ID(sb), &tmp);
 467         if ( err ) {
 468                 printk(__FUNCTION__ ": obd_statfs fails (%d)\n", err);
 469                 return err;
 470         }
 471         copy_to_user(buf, &tmp, (bufsize<sizeof(tmp)) ? bufsize : sizeof(tmp));
 472
 473         EXIT;
 474
 475         return err;
 476 }
 477
 478 /* exported operations */
 479 struct super_operations obdfs_super_operations =
 480 {
 481         read_inode: obdfs_read_inode,
 482         write_inode: obdfs_write_inode,
 483         put_inode: obdfs_put_inode,
 484         delete_inode: obdfs_delete_inode,
 485         put_super: obdfs_put_super,
 486         statfs: obdfs_statfs
 487 };
 488
 489 struct file_system_type obdfs_fs_type = {
 490    "obdfs", 0, obdfs_read_super, NULL
 491 };
 492
 493 int init_obdfs(void)
 494 {
 495         int err;
 496
 497         printk(KERN_INFO "OBDFS v0.1, braam@stelias.com\n");
 498
 499         obdfs_sysctl_init();
 500
 501         INIT_LIST_HEAD(&obdfs_super_list);
 502         err = obdfs_init_pgrqcache();
 503         if (err)
 504                 return err;
 505
 506         obdfs_flushd_init();
 507         return register_filesystem(&obdfs_fs_type);
 508 }
 509
 510 struct address_space_operations obdfs_aops = {
 511         readpage: obdfs_readpage,
 512         writepage: obdfs_writepage,
 513         sync_page: block_sync_page,
 514         prepare_write: obdfs_prepare_write,
 515         commit_write: obdfs_commit_write,
 516         bmap: NULL
 517 };
 518
 519
 520 #ifdef MODULE
 521 int init_module(void)
 522 {
 523         return init_obdfs();
 524 }
 525
 526 void cleanup_module(void)
 527 {
 528         ENTRY;
 529
 530         obdfs_flushd_cleanup();
 531         obdfs_sysctl_clean();
 532         obdfs_cleanup_pgrqcache();
 533         unregister_filesystem(&obdfs_fs_type);
 534         CDEBUG(D_MALLOC, "OBDFS mem used %ld\n", obd_memory);
 535         EXIT;
 536 }
 537
 538 #endif