fs/inode.c

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

DEFINITIONS
