root/fs/inode.c

/* */

DEFINITIONS

1. hashfn
2. insert_inode_free
3. remove_inode_free
4. insert_inode_hash
5. remove_inode_hash
6. put_last_free
7. grow_inodes
8. inode_init
9. wait_on_inode
10. lock_inode
11. unlock_inode
12. clear_inode
13. fs_may_mount
14. fs_may_umount
15. fs_may_remount_ro
16. write_inode
17. read_inode
18. inode_change_ok
19. inode_setattr
20. notify_change
21. bmap
22. invalidate_inodes
23. sync_inodes
24. iput
25. value
26. get_empty_inode
27. get_pipe_inode
28. __iget
29. __wait_on_inode

   1 /*
   2  *  linux/fs/inode.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6 
   7 #include <linux/stat.h>
   8 #include <linux/sched.h>
   9 #include <linux/kernel.h>
  10 #include <linux/mm.h>
  11 #include <linux/string.h>
  12 
  13 #include <asm/system.h>
  14 
  15 static struct inode_hash_entry {
  16         struct inode * inode;
  17         int updating;
  18 } hash_table[NR_IHASH];
  19 
  20 static struct inode * first_inode;
  21 static struct wait_queue * inode_wait = NULL;
  22 /* Keep these next two contiguous in memory for sysctl.c */
  23 int nr_inodes = 0, nr_free_inodes = 0;
  24 int max_inodes = NR_INODE;
  25 
  26 static inline int const hashfn(kdev_t dev, unsigned int i)
     /*  */
  27 {
  28         return (HASHDEV(dev) ^ i) % NR_IHASH;
  29 }
  30 
  31 static inline struct inode_hash_entry * const hash(kdev_t dev, int i)
  32 {
  33         return hash_table + hashfn(dev, i);
  34 }
  35 
  36 static inline void insert_inode_free(struct inode *inode)
     /*  */
  37 {
  38         inode->i_next = first_inode;
  39         inode->i_prev = first_inode->i_prev;
  40         inode->i_next->i_prev = inode;
  41         inode->i_prev->i_next = inode;
  42         first_inode = inode;
  43 }
  44 
  45 static inline void remove_inode_free(struct inode *inode)
     /*  */
  46 {
  47         if (first_inode == inode)
  48                 first_inode = first_inode->i_next;
  49         if (inode->i_next)
  50                 inode->i_next->i_prev = inode->i_prev;
  51         if (inode->i_prev)
  52                 inode->i_prev->i_next = inode->i_next;
  53         inode->i_next = inode->i_prev = NULL;
  54 }
  55 
  56 void insert_inode_hash(struct inode *inode)
     /*  */
  57 {
  58         struct inode_hash_entry *h;
  59         h = hash(inode->i_dev, inode->i_ino);
  60 
  61         inode->i_hash_next = h->inode;
  62         inode->i_hash_prev = NULL;
  63         if (inode->i_hash_next)
  64                 inode->i_hash_next->i_hash_prev = inode;
  65         h->inode = inode;
  66 }
  67 
  68 static inline void remove_inode_hash(struct inode *inode)
     /*  */
  69 {
  70         struct inode_hash_entry *h;
  71         h = hash(inode->i_dev, inode->i_ino);
  72 
  73         if (h->inode == inode)
  74                 h->inode = inode->i_hash_next;
  75         if (inode->i_hash_next)
  76                 inode->i_hash_next->i_hash_prev = inode->i_hash_prev;
  77         if (inode->i_hash_prev)
  78                 inode->i_hash_prev->i_hash_next = inode->i_hash_next;
  79         inode->i_hash_prev = inode->i_hash_next = NULL;
  80 }
  81 
  82 static inline void put_last_free(struct inode *inode)
     /*  */
  83 {
  84         remove_inode_free(inode);
  85         inode->i_prev = first_inode->i_prev;
  86         inode->i_prev->i_next = inode;
  87         inode->i_next = first_inode;
  88         inode->i_next->i_prev = inode;
  89 }
  90 
  91 int grow_inodes(void)
     /*  */
  92 {
  93         struct inode * inode;
  94         int i;
  95 
  96         if (!(inode = (struct inode*) get_free_page(GFP_KERNEL)))
  97                 return -ENOMEM;
  98 
  99         i=PAGE_SIZE / sizeof(struct inode);
 100         nr_inodes += i;
 101         nr_free_inodes += i;
 102 
 103         if (!first_inode)
 104                 inode->i_next = inode->i_prev = first_inode = inode++, i--;
 105 
 106         for ( ; i ; i-- )
 107                 insert_inode_free(inode++);
 108         return 0;
 109 }
 110 
 111 unsigned long inode_init(unsigned long start, unsigned long end)
     /*  */
 112 {
 113         memset(hash_table, 0, sizeof(hash_table));
 114         first_inode = NULL;
 115         return start;
 116 }
 117 
 118 static void __wait_on_inode(struct inode *);
 119 
 120 static inline void wait_on_inode(struct inode * inode)
     /*  */
 121 {
 122         if (inode->i_lock)
 123                 __wait_on_inode(inode);
 124 }
 125 
 126 static inline void lock_inode(struct inode * inode)
     /*  */
 127 {
 128         wait_on_inode(inode);
 129         inode->i_lock = 1;
 130 }
 131 
 132 static inline void unlock_inode(struct inode * inode)
     /*  */
 133 {
 134         inode->i_lock = 0;
 135         wake_up(&inode->i_wait);
 136 }
 137 
 138 /*
 139  * Note that we don't want to disturb any wait-queues when we discard
 140  * an inode.
 141  *
 142  * Argghh. Got bitten by a gcc problem with inlining: no way to tell
 143  * the compiler that the inline asm function 'memset' changes 'inode'.
 144  * I've been searching for the bug for days, and was getting desperate.
 145  * Finally looked at the assembler output... Grrr.
 146  *
 147  * The solution is the weird use of 'volatile'. Ho humm. Have to report
 148  * it to the gcc lists, and hope we can do this more cleanly some day..
 149  */
 150 void clear_inode(struct inode * inode)
     /*  */
 151 {
 152         struct wait_queue * wait;
 153 
 154         invalidate_inode_pages(inode, 0);
 155         wait_on_inode(inode);
 156         if (IS_WRITABLE(inode)) {
 157                 if (inode->i_sb && inode->i_sb->dq_op)
 158                         inode->i_sb->dq_op->drop(inode);
 159         }
 160         remove_inode_hash(inode);
 161         remove_inode_free(inode);
 162         wait = ((volatile struct inode *) inode)->i_wait;
 163         if (inode->i_count)
 164                 nr_free_inodes++;
 165         memset(inode,0,sizeof(*inode));
 166         ((volatile struct inode *) inode)->i_wait = wait;
 167         insert_inode_free(inode);
 168 }
 169 
 170 int fs_may_mount(kdev_t dev)
     /*  */
 171 {
 172         struct inode * inode, * next;
 173         int i;
 174 
 175         next = first_inode;
 176         for (i = nr_inodes ; i > 0 ; i--) {
 177                 inode = next;
 178                 next = inode->i_next;   /* clear_inode() changes the queues.. */
 179                 if (inode->i_dev != dev)
 180                         continue;
 181                 if (inode->i_count || inode->i_dirt || inode->i_lock)
 182                         return 0;
 183                 clear_inode(inode);
 184         }
 185         return 1;
 186 }
 187 
 188 int fs_may_umount(kdev_t dev, struct inode * mount_root)
     /*  */
 189 {
 190         struct inode * inode;
 191         int i;
 192 
 193         inode = first_inode;
 194         for (i=0 ; i < nr_inodes ; i++, inode = inode->i_next) {
 195                 if (inode->i_dev != dev || !inode->i_count)
 196                         continue;
 197                 if (inode == mount_root && inode->i_count == 1)
 198                         continue;
 199                 return 0;
 200         }
 201         return 1;
 202 }
 203 
 204 int fs_may_remount_ro(kdev_t dev)
     /*  */
 205 {
 206         struct file * file;
 207         int i;
 208 
 209         /* Check that no files are currently opened for writing. */
 210         for (file = first_file, i=0; i<nr_files; i++, file=file->f_next) {
 211                 if (!file->f_count || !file->f_inode ||
 212                     file->f_inode->i_dev != dev)
 213                         continue;
 214                 if (S_ISREG(file->f_inode->i_mode) && (file->f_mode & 2))
 215                         return 0;
 216         }
 217         return 1;
 218 }
 219 
 220 static void write_inode(struct inode * inode)
     /*  */
 221 {
 222         if (!inode->i_dirt)
 223                 return;
 224         wait_on_inode(inode);
 225         if (!inode->i_dirt)
 226                 return;
 227         if (!inode->i_sb || !inode->i_sb->s_op || !inode->i_sb->s_op->write_inode) {
 228                 inode->i_dirt = 0;
 229                 return;
 230         }
 231         inode->i_lock = 1;      
 232         inode->i_sb->s_op->write_inode(inode);
 233         unlock_inode(inode);
 234 }
 235 
 236 static inline void read_inode(struct inode * inode)
     /*  */
 237 {
 238         lock_inode(inode);
 239         if (inode->i_sb && inode->i_sb->s_op && inode->i_sb->s_op->read_inode)
 240                 inode->i_sb->s_op->read_inode(inode);
 241         unlock_inode(inode);
 242 }
 243 
 244 /* POSIX UID/GID verification for setting inode attributes */
 245 int inode_change_ok(struct inode *inode, struct iattr *attr)
     /*  */
 246 {
 247         /* Make sure a caller can chown */
 248         if ((attr->ia_valid & ATTR_UID) &&
 249             (current->fsuid != inode->i_uid ||
 250              attr->ia_uid != inode->i_uid) && !fsuser())
 251                 return -EPERM;
 252 
 253         /* Make sure caller can chgrp */
 254         if ((attr->ia_valid & ATTR_GID) &&
 255             (!in_group_p(attr->ia_gid) && attr->ia_gid != inode->i_gid) &&
 256             !fsuser())
 257                 return -EPERM;
 258 
 259         /* Make sure a caller can chmod */
 260         if (attr->ia_valid & ATTR_MODE) {
 261                 if ((current->fsuid != inode->i_uid) && !fsuser())
 262                         return -EPERM;
 263                 /* Also check the setgid bit! */
 264                 if (!fsuser() && !in_group_p((attr->ia_valid & ATTR_GID) ? attr->ia_gid :
 265                                              inode->i_gid))
 266                         attr->ia_mode &= ~S_ISGID;
 267         }
 268 
 269         /* Check for setting the inode time */
 270         if ((attr->ia_valid & ATTR_ATIME_SET) &&
 271             ((current->fsuid != inode->i_uid) && !fsuser()))
 272                 return -EPERM;
 273         if ((attr->ia_valid & ATTR_MTIME_SET) &&
 274             ((current->fsuid != inode->i_uid) && !fsuser()))
 275                 return -EPERM;
 276 
 277 
 278         return 0;
 279 }
 280 
 281 /*
 282  * Set the appropriate attributes from an attribute structure into
 283  * the inode structure.
 284  */
 285 void inode_setattr(struct inode *inode, struct iattr *attr)
     /*  */
 286 {
 287         if (attr->ia_valid & ATTR_UID)
 288                 inode->i_uid = attr->ia_uid;
 289         if (attr->ia_valid & ATTR_GID)
 290                 inode->i_gid = attr->ia_gid;
 291         if (attr->ia_valid & ATTR_SIZE)
 292                 inode->i_size = attr->ia_size;
 293         if (attr->ia_valid & ATTR_ATIME)
 294                 inode->i_atime = attr->ia_atime;
 295         if (attr->ia_valid & ATTR_MTIME)
 296                 inode->i_mtime = attr->ia_mtime;
 297         if (attr->ia_valid & ATTR_CTIME)
 298                 inode->i_ctime = attr->ia_ctime;
 299         if (attr->ia_valid & ATTR_MODE) {
 300                 inode->i_mode = attr->ia_mode;
 301                 if (!fsuser() && !in_group_p(inode->i_gid))
 302                         inode->i_mode &= ~S_ISGID;
 303         }
 304         inode->i_dirt = 1;
 305 }
 306 
 307 /*
 308  * notify_change is called for inode-changing operations such as
 309  * chown, chmod, utime, and truncate.  It is guaranteed (unlike
 310  * write_inode) to be called from the context of the user requesting
 311  * the change.
 312  */
 313 
 314 int notify_change(struct inode * inode, struct iattr *attr)
     /*  */
 315 {
 316         int retval;
 317 
 318         if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME)) {
 319                 unsigned long now = CURRENT_TIME;
 320                 if (!(attr->ia_valid & ATTR_ATIME_SET))
 321                         attr->ia_atime = now;
 322                 if (!(attr->ia_valid & ATTR_MTIME_SET))
 323                         attr->ia_mtime = now;
 324                 attr->ia_ctime = now;
 325         }
 326 
 327         if (inode->i_sb && inode->i_sb->s_op  &&
 328             inode->i_sb->s_op->notify_change) 
 329                 return inode->i_sb->s_op->notify_change(inode, attr);
 330 
 331         if ((retval = inode_change_ok(inode, attr)) != 0)
 332                 return retval;
 333 
 334         inode_setattr(inode, attr);
 335         return 0;
 336 }
 337 
 338 /*
 339  * bmap is needed for demand-loading and paging: if this function
 340  * doesn't exist for a filesystem, then those things are impossible:
 341  * executables cannot be run from the filesystem etc...
 342  *
 343  * This isn't as bad as it sounds: the read-routines might still work,
 344  * so the filesystem would be otherwise ok (for example, you might have
 345  * a DOS filesystem, which doesn't lend itself to bmap very well, but
 346  * you could still transfer files to/from the filesystem)
 347  */
 348 int bmap(struct inode * inode, int block)
     /*  */
 349 {
 350         if (inode->i_op && inode->i_op->bmap)
 351                 return inode->i_op->bmap(inode,block);
 352         return 0;
 353 }
 354 
 355 void invalidate_inodes(kdev_t dev)
     /*  */
 356 {
 357         struct inode * inode, * next;
 358         int i;
 359 
 360         next = first_inode;
 361         for(i = nr_inodes ; i > 0 ; i--) {
 362                 inode = next;
 363                 next = inode->i_next;           /* clear_inode() changes the queues.. */
 364                 if (inode->i_dev != dev)
 365                         continue;
 366                 if (inode->i_count || inode->i_dirt || inode->i_lock) {
 367                         printk("VFS: inode busy on removed device %s\n",
 368                                kdevname(dev));
 369                         continue;
 370                 }
 371                 clear_inode(inode);
 372         }
 373 }
 374 
 375 void sync_inodes(kdev_t dev)
     /*  */
 376 {
 377         int i;
 378         struct inode * inode;
 379 
 380         inode = first_inode;
 381         for(i = 0; i < nr_inodes*2; i++, inode = inode->i_next) {
 382                 if (dev && inode->i_dev != dev)
 383                         continue;
 384                 wait_on_inode(inode);
 385                 if (inode->i_dirt)
 386                         write_inode(inode);
 387         }
 388 }
 389 
 390 void iput(struct inode * inode)
     /*  */
 391 {
 392         if (!inode)
 393                 return;
 394         wait_on_inode(inode);
 395         if (!inode->i_count) {
 396                 printk("VFS: iput: trying to free free inode\n");
 397                 printk("VFS: device %s, inode %lu, mode=0%07o\n",
 398                         kdevname(inode->i_rdev), inode->i_ino, inode->i_mode);
 399                 return;
 400         }
 401         if (inode->i_pipe)
 402                 wake_up_interruptible(&PIPE_WAIT(*inode));
 403 repeat:
 404         if (inode->i_count>1) {
 405                 inode->i_count--;
 406                 return;
 407         }
 408 
 409         wake_up(&inode_wait);
 410         if (inode->i_pipe) {
 411                 unsigned long page = (unsigned long) PIPE_BASE(*inode);
 412                 PIPE_BASE(*inode) = NULL;
 413                 free_page(page);
 414         }
 415 
 416         if (inode->i_sb && inode->i_sb->s_op && inode->i_sb->s_op->put_inode) {
 417                 inode->i_sb->s_op->put_inode(inode);
 418                 if (!inode->i_nlink)
 419                         return;
 420         }
 421 
 422         if (inode->i_dirt) {
 423                 write_inode(inode);     /* we can sleep - so do again */
 424                 wait_on_inode(inode);
 425                 goto repeat;
 426         }
 427 
 428         inode->i_count--;
 429         if (IS_WRITABLE(inode)) {
 430                 if (inode->i_sb && inode->i_sb->dq_op)
 431                         inode->i_sb->dq_op->drop(inode);
 432         }
 433 
 434         if (inode->i_mmap) {
 435                 printk("iput: inode %lu on device %s still has mappings.\n",
 436                         inode->i_ino, kdevname(inode->i_dev));
 437                 inode->i_mmap = NULL;
 438         }
 439 
 440         nr_free_inodes++;
 441         return;
 442 }
 443 
 444 static inline unsigned long value(struct inode * inode)
     /*  */
 445 {
 446         if (inode->i_lock)  
 447                 return 1000;
 448         if (inode->i_dirt)
 449                 return 1000;
 450         return inode->i_nrpages;
 451 }
 452 
 453 struct inode * get_empty_inode(void)
     /*  */
 454 {
 455         static int ino = 0;
 456         struct inode * inode, * best;
 457         unsigned long badness = 1000;
 458         int i;
 459 
 460         if (nr_inodes < max_inodes && nr_free_inodes < (nr_inodes >> 1))
 461                 grow_inodes();
 462 repeat:
 463         inode = first_inode;
 464         best = NULL;
 465         for (i = 0; i<nr_inodes; inode = inode->i_next, i++) {
 466                 if (!inode->i_count) {
 467                         unsigned long i = value(inode);
 468                         if (i < badness) {
 469                                 best = inode;
 470                                 if ((badness = i) == 0)
 471                                         break;
 472                         }
 473                 }
 474         }
 475         if (badness)
 476                 if (nr_inodes < max_inodes) {
 477                         if (grow_inodes() == 0)
 478                                 goto repeat;
 479                 }
 480         inode = best;
 481         if (!inode) {
 482                 printk("VFS: No free inodes - contact Linus\n");
 483                 sleep_on(&inode_wait);
 484                 goto repeat;
 485         }
 486         if (inode->i_lock) {
 487                 wait_on_inode(inode);
 488                 goto repeat;
 489         }
 490         if (inode->i_dirt) {
 491                 write_inode(inode);
 492                 goto repeat;
 493         }
 494         if (inode->i_count)
 495                 goto repeat;
 496         clear_inode(inode);
 497         inode->i_count = 1;
 498         inode->i_nlink = 1;
 499         inode->i_version = ++event;
 500         inode->i_sem.count = 1;
 501         inode->i_ino = ++ino;
 502         inode->i_dev = 0;
 503         nr_free_inodes--;
 504         if (nr_free_inodes < 0) {
 505                 printk ("VFS: get_empty_inode: bad free inode count.\n");
 506                 nr_free_inodes = 0;
 507         }
 508         return inode;
 509 }
 510 
 511 struct inode * get_pipe_inode(void)
     /*  */
 512 {
 513         struct inode * inode;
 514         extern struct inode_operations pipe_inode_operations;
 515 
 516         if (!(inode = get_empty_inode()))
 517                 return NULL;
 518         if (!(PIPE_BASE(*inode) = (char*) __get_free_page(GFP_USER))) {
 519                 iput(inode);
 520                 return NULL;
 521         }
 522         inode->i_op = &pipe_inode_operations;
 523         inode->i_count = 2;     /* sum of readers/writers */
 524         PIPE_WAIT(*inode) = NULL;
 525         PIPE_START(*inode) = PIPE_LEN(*inode) = 0;
 526         PIPE_RD_OPENERS(*inode) = PIPE_WR_OPENERS(*inode) = 0;
 527         PIPE_READERS(*inode) = PIPE_WRITERS(*inode) = 1;
 528         PIPE_LOCK(*inode) = 0;
 529         inode->i_pipe = 1;
 530         inode->i_mode |= S_IFIFO | S_IRUSR | S_IWUSR;
 531         inode->i_uid = current->fsuid;
 532         inode->i_gid = current->fsgid;
 533         inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 534         inode->i_blksize = PAGE_SIZE;
 535         return inode;
 536 }
 537 
 538 struct inode *__iget(struct super_block * sb, int nr, int crossmntp)
     /*  */
 539 {
 540         static struct wait_queue * update_wait = NULL;
 541         struct inode_hash_entry * h;
 542         struct inode * inode;
 543         struct inode * empty = NULL;
 544 
 545         if (!sb)
 546                 panic("VFS: iget with sb==NULL");
 547         h = hash(sb->s_dev, nr);
 548 repeat:
 549         for (inode = h->inode; inode ; inode = inode->i_hash_next)
 550                 if (inode->i_dev == sb->s_dev && inode->i_ino == nr)
 551                         goto found_it;
 552         if (!empty) {
 553                 h->updating++;
 554                 empty = get_empty_inode();
 555                 if (!--h->updating)
 556                         wake_up(&update_wait);
 557                 if (empty)
 558                         goto repeat;
 559                 return (NULL);
 560         }
 561         inode = empty;
 562         inode->i_sb = sb;
 563         inode->i_dev = sb->s_dev;
 564         inode->i_ino = nr;
 565         inode->i_flags = sb->s_flags;
 566         put_last_free(inode);
 567         insert_inode_hash(inode);
 568         read_inode(inode);
 569         goto return_it;
 570 
 571 found_it:
 572         if (!inode->i_count)
 573                 nr_free_inodes--;
 574         inode->i_count++;
 575         wait_on_inode(inode);
 576         if (inode->i_dev != sb->s_dev || inode->i_ino != nr) {
 577                 printk("Whee.. inode changed from under us. Tell Linus\n");
 578                 iput(inode);
 579                 goto repeat;
 580         }
 581         if (crossmntp && inode->i_mount) {
 582                 struct inode * tmp = inode->i_mount;
 583                 tmp->i_count++;
 584                 iput(inode);
 585                 inode = tmp;
 586                 wait_on_inode(inode);
 587         }
 588         if (empty)
 589                 iput(empty);
 590 
 591 return_it:
 592         while (h->updating)
 593                 sleep_on(&update_wait);
 594         return inode;
 595 }
 596 
 597 /*
 598  * The "new" scheduling primitives (new as of 0.97 or so) allow this to
 599  * be done without disabling interrupts (other than in the actual queue
 600  * updating things: only a couple of 386 instructions). This should be
 601  * much better for interrupt latency.
 602  */
 603 static void __wait_on_inode(struct inode * inode)
     /*  */
 604 {
 605         struct wait_queue wait = { current, NULL };
 606 
 607         add_wait_queue(&inode->i_wait, &wait);
 608 repeat:
 609         current->state = TASK_UNINTERRUPTIBLE;
 610         if (inode->i_lock) {
 611                 schedule();
 612                 goto repeat;
 613         }
 614         remove_wait_queue(&inode->i_wait, &wait);
 615         current->state = TASK_RUNNING;
 616 }

/* */