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 /* 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         truncate_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         /*
 248          *      If force is set do it anyway.
 249          */
 250          
 251         if (attr->ia_valid & ATTR_FORCE)
 252                 return 0;
 253 
 254         /* Make sure a caller can chown */
 255         if ((attr->ia_valid & ATTR_UID) &&
 256             (current->fsuid != inode->i_uid ||
 257              attr->ia_uid != inode->i_uid) && !fsuser())
 258                 return -EPERM;
 259 
 260         /* Make sure caller can chgrp */
 261         if ((attr->ia_valid & ATTR_GID) &&
 262             (!in_group_p(attr->ia_gid) && attr->ia_gid != inode->i_gid) &&
 263             !fsuser())
 264                 return -EPERM;
 265 
 266         /* Make sure a caller can chmod */
 267         if (attr->ia_valid & ATTR_MODE) {
 268                 if ((current->fsuid != inode->i_uid) && !fsuser())
 269                         return -EPERM;
 270                 /* Also check the setgid bit! */
 271                 if (!fsuser() && !in_group_p((attr->ia_valid & ATTR_GID) ? attr->ia_gid :
 272                                              inode->i_gid))
 273                         attr->ia_mode &= ~S_ISGID;
 274         }
 275 
 276         /* Check for setting the inode time */
 277         if ((attr->ia_valid & ATTR_ATIME_SET) &&
 278             ((current->fsuid != inode->i_uid) && !fsuser()))
 279                 return -EPERM;
 280         if ((attr->ia_valid & ATTR_MTIME_SET) &&
 281             ((current->fsuid != inode->i_uid) && !fsuser()))
 282                 return -EPERM;
 283         return 0;
 284 }
 285 
 286 /*
 287  * Set the appropriate attributes from an attribute structure into
 288  * the inode structure.
 289  */
 290 void inode_setattr(struct inode *inode, struct iattr *attr)
     /*  */
 291 {
 292         if (attr->ia_valid & ATTR_UID)
 293                 inode->i_uid = attr->ia_uid;
 294         if (attr->ia_valid & ATTR_GID)
 295                 inode->i_gid = attr->ia_gid;
 296         if (attr->ia_valid & ATTR_SIZE)
 297                 inode->i_size = attr->ia_size;
 298         if (attr->ia_valid & ATTR_ATIME)
 299                 inode->i_atime = attr->ia_atime;
 300         if (attr->ia_valid & ATTR_MTIME)
 301                 inode->i_mtime = attr->ia_mtime;
 302         if (attr->ia_valid & ATTR_CTIME)
 303                 inode->i_ctime = attr->ia_ctime;
 304         if (attr->ia_valid & ATTR_MODE) {
 305                 inode->i_mode = attr->ia_mode;
 306                 if (!fsuser() && !in_group_p(inode->i_gid))
 307                         inode->i_mode &= ~S_ISGID;
 308         }
 309         inode->i_dirt = 1;
 310 }
 311 
 312 /*
 313  * notify_change is called for inode-changing operations such as
 314  * chown, chmod, utime, and truncate.  It is guaranteed (unlike
 315  * write_inode) to be called from the context of the user requesting
 316  * the change.
 317  */
 318 
 319 int notify_change(struct inode * inode, struct iattr *attr)
     /*  */
 320 {
 321         int retval;
 322 
 323         attr->ia_ctime = CURRENT_TIME;
 324         if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME)) {
 325                 if (!(attr->ia_valid & ATTR_ATIME_SET))
 326                         attr->ia_atime = attr->ia_ctime;
 327                 if (!(attr->ia_valid & ATTR_MTIME_SET))
 328                         attr->ia_mtime = attr->ia_ctime;
 329         }
 330 
 331         if (inode->i_sb && inode->i_sb->s_op  &&
 332             inode->i_sb->s_op->notify_change) 
 333                 return inode->i_sb->s_op->notify_change(inode, attr);
 334 
 335         if ((retval = inode_change_ok(inode, attr)) != 0)
 336                 return retval;
 337 
 338         inode_setattr(inode, attr);
 339         return 0;
 340 }
 341 
 342 /*
 343  * bmap is needed for demand-loading and paging: if this function
 344  * doesn't exist for a filesystem, then those things are impossible:
 345  * executables cannot be run from the filesystem etc...
 346  *
 347  * This isn't as bad as it sounds: the read-routines might still work,
 348  * so the filesystem would be otherwise ok (for example, you might have
 349  * a DOS filesystem, which doesn't lend itself to bmap very well, but
 350  * you could still transfer files to/from the filesystem)
 351  */
 352 int bmap(struct inode * inode, int block)
     /*  */
 353 {
 354         if (inode->i_op && inode->i_op->bmap)
 355                 return inode->i_op->bmap(inode,block);
 356         return 0;
 357 }
 358 
 359 void invalidate_inodes(kdev_t dev)
     /*  */
 360 {
 361         struct inode * inode, * next;
 362         int i;
 363 
 364         next = first_inode;
 365         for(i = nr_inodes ; i > 0 ; i--) {
 366                 inode = next;
 367                 next = inode->i_next;           /* clear_inode() changes the queues.. */
 368                 if (inode->i_dev != dev)
 369                         continue;
 370                 if (inode->i_count || inode->i_dirt || inode->i_lock) {
 371                         printk("VFS: inode busy on removed device %s\n",
 372                                kdevname(dev));
 373                         continue;
 374                 }
 375                 clear_inode(inode);
 376         }
 377 }
 378 
 379 void sync_inodes(kdev_t dev)
     /*  */
 380 {
 381         int i;
 382         struct inode * inode;
 383 
 384         inode = first_inode;
 385         for(i = 0; i < nr_inodes*2; i++, inode = inode->i_next) {
 386                 if (dev && inode->i_dev != dev)
 387                         continue;
 388                 wait_on_inode(inode);
 389                 if (inode->i_dirt)
 390                         write_inode(inode);
 391         }
 392 }
 393 
 394 void iput(struct inode * inode)
     /*  */
 395 {
 396         if (!inode)
 397                 return;
 398         wait_on_inode(inode);
 399         if (!inode->i_count) {
 400                 printk("VFS: iput: trying to free free inode\n");
 401                 printk("VFS: device %s, inode %lu, mode=0%07o\n",
 402                         kdevname(inode->i_rdev), inode->i_ino, inode->i_mode);
 403                 return;
 404         }
 405         if (inode->i_pipe)
 406                 wake_up_interruptible(&PIPE_WAIT(*inode));
 407 repeat:
 408         if (inode->i_count>1) {
 409                 inode->i_count--;
 410                 return;
 411         }
 412 
 413         wake_up(&inode_wait);
 414         if (inode->i_pipe) {
 415                 unsigned long page = (unsigned long) PIPE_BASE(*inode);
 416                 PIPE_BASE(*inode) = NULL;
 417                 free_page(page);
 418         }
 419 
 420         if (inode->i_sb && inode->i_sb->s_op && inode->i_sb->s_op->put_inode) {
 421                 inode->i_sb->s_op->put_inode(inode);
 422                 if (!inode->i_nlink)
 423                         return;
 424         }
 425 
 426         if (inode->i_dirt) {
 427                 write_inode(inode);     /* we can sleep - so do again */
 428                 wait_on_inode(inode);
 429                 goto repeat;
 430         }
 431 
 432         if (IS_WRITABLE(inode)) {
 433                 if (inode->i_sb && inode->i_sb->dq_op) {
 434                         /* Here we can sleep also. Let's do it again
 435                          * Dmitry Gorodchanin 02/11/96 
 436                          */
 437                         inode->i_lock = 1;
 438                         inode->i_sb->dq_op->drop(inode);
 439                         unlock_inode(inode);
 440                         goto repeat;
 441                 }
 442         }
 443         
 444         inode->i_count--;
 445 
 446         if (inode->i_mmap) {
 447                 printk("iput: inode %lu on device %s still has mappings.\n",
 448                         inode->i_ino, kdevname(inode->i_dev));
 449                 inode->i_mmap = NULL;
 450         }
 451 
 452         nr_free_inodes++;
 453         return;
 454 }
 455 
 456 static inline unsigned long value(struct inode * inode)
     /*  */
 457 {
 458         if (inode->i_lock)  
 459                 return 1000;
 460         if (inode->i_dirt)
 461                 return 1000;
 462         return inode->i_nrpages;
 463 }
 464 
 465 struct inode * get_empty_inode(void)
     /*  */
 466 {
 467         static int ino = 0;
 468         struct inode * inode, * best;
 469         unsigned long badness = 1000;
 470         int i;
 471 
 472         if (nr_inodes < max_inodes && nr_free_inodes < (nr_inodes >> 1))
 473                 grow_inodes();
 474 repeat:
 475         inode = first_inode;
 476         best = NULL;
 477         for (i = 0; i<nr_inodes; inode = inode->i_next, i++) {
 478                 if (!inode->i_count) {
 479                         unsigned long i = value(inode);
 480                         if (i < badness) {
 481                                 best = inode;
 482                                 if ((badness = i) == 0)
 483                                         break;
 484                         }
 485                 }
 486         }
 487         if (badness)
 488                 if (nr_inodes < max_inodes) {
 489                         if (grow_inodes() == 0)
 490                                 goto repeat;
 491                 }
 492         inode = best;
 493         if (!inode) {
 494                 printk("VFS: No free inodes - contact Linus\n");
 495                 sleep_on(&inode_wait);
 496                 goto repeat;
 497         }
 498         if (inode->i_lock) {
 499                 wait_on_inode(inode);
 500                 goto repeat;
 501         }
 502         if (inode->i_dirt) {
 503                 write_inode(inode);
 504                 goto repeat;
 505         }
 506         if (inode->i_count)
 507                 goto repeat;
 508         clear_inode(inode);
 509         inode->i_count = 1;
 510         inode->i_nlink = 1;
 511         inode->i_version = ++event;
 512         inode->i_sem.count = 1;
 513         inode->i_ino = ++ino;
 514         inode->i_dev = 0;
 515         nr_free_inodes--;
 516         if (nr_free_inodes < 0) {
 517                 printk ("VFS: get_empty_inode: bad free inode count.\n");
 518                 nr_free_inodes = 0;
 519         }
 520         return inode;
 521 }
 522 
 523 struct inode * get_pipe_inode(void)
     /*  */
 524 {
 525         struct inode * inode;
 526         extern struct inode_operations pipe_inode_operations;
 527 
 528         if (!(inode = get_empty_inode()))
 529                 return NULL;
 530         if (!(PIPE_BASE(*inode) = (char*) __get_free_page(GFP_USER))) {
 531                 iput(inode);
 532                 return NULL;
 533         }
 534         inode->i_op = &pipe_inode_operations;
 535         inode->i_count = 2;     /* sum of readers/writers */
 536         PIPE_WAIT(*inode) = NULL;
 537         PIPE_START(*inode) = PIPE_LEN(*inode) = 0;
 538         PIPE_RD_OPENERS(*inode) = PIPE_WR_OPENERS(*inode) = 0;
 539         PIPE_READERS(*inode) = PIPE_WRITERS(*inode) = 1;
 540         PIPE_LOCK(*inode) = 0;
 541         inode->i_pipe = 1;
 542         inode->i_mode |= S_IFIFO | S_IRUSR | S_IWUSR;
 543         inode->i_uid = current->fsuid;
 544         inode->i_gid = current->fsgid;
 545         inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 546         inode->i_blksize = PAGE_SIZE;
 547         return inode;
 548 }
 549 
 550 struct inode *__iget(struct super_block * sb, int nr, int crossmntp)
     /*  */
 551 {
 552         static struct wait_queue * update_wait = NULL;
 553         struct inode_hash_entry * h;
 554         struct inode * inode;
 555         struct inode * empty = NULL;
 556 
 557         if (!sb)
 558                 panic("VFS: iget with sb==NULL");
 559         h = hash(sb->s_dev, nr);
 560 repeat:
 561         for (inode = h->inode; inode ; inode = inode->i_hash_next)
 562                 if (inode->i_dev == sb->s_dev && inode->i_ino == nr)
 563                         goto found_it;
 564         if (!empty) {
 565                 h->updating++;
 566                 empty = get_empty_inode();
 567                 if (!--h->updating)
 568                         wake_up(&update_wait);
 569                 if (empty)
 570                         goto repeat;
 571                 return (NULL);
 572         }
 573         inode = empty;
 574         inode->i_sb = sb;
 575         inode->i_dev = sb->s_dev;
 576         inode->i_ino = nr;
 577         inode->i_flags = sb->s_flags;
 578         put_last_free(inode);
 579         insert_inode_hash(inode);
 580         read_inode(inode);
 581         goto return_it;
 582 
 583 found_it:
 584         if (!inode->i_count)
 585                 nr_free_inodes--;
 586         inode->i_count++;
 587         wait_on_inode(inode);
 588         if (inode->i_dev != sb->s_dev || inode->i_ino != nr) {
 589                 printk("Whee.. inode changed from under us. Tell Linus\n");
 590                 iput(inode);
 591                 goto repeat;
 592         }
 593         if (crossmntp && inode->i_mount) {
 594                 struct inode * tmp = inode->i_mount;
 595                 tmp->i_count++;
 596                 iput(inode);
 597                 inode = tmp;
 598                 wait_on_inode(inode);
 599         }
 600         if (empty)
 601                 iput(empty);
 602 
 603 return_it:
 604         while (h->updating)
 605                 sleep_on(&update_wait);
 606         return inode;
 607 }
 608 
 609 /*
 610  * The "new" scheduling primitives (new as of 0.97 or so) allow this to
 611  * be done without disabling interrupts (other than in the actual queue
 612  * updating things: only a couple of 386 instructions). This should be
 613  * much better for interrupt latency.
 614  */
 615 static void __wait_on_inode(struct inode * inode)
     /*  */
 616 {
 617         struct wait_queue wait = { current, NULL };
 618 
 619         add_wait_queue(&inode->i_wait, &wait);
 620 repeat:
 621         current->state = TASK_UNINTERRUPTIBLE;
 622         if (inode->i_lock) {
 623                 schedule();
 624                 goto repeat;
 625         }
 626         remove_wait_queue(&inode->i_wait, &wait);
 627         current->state = TASK_RUNNING;
 628 }
/* */
root/fs/inode.c

DEFINITIONS
