/linux/src/fs/namei.c

   1/*
   2 *  linux/fs/namei.c
   3 *
   4 *  Copyright (C) 1991, 1992  Linus Torvalds
   5 *
   6 *  OSKit support added by the University of Utah, 1997
   7 */
   8
   9/*
  10 * Some corrections by tytso.
  11 */
  12
  13/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
  14 * lookup logic.
  15 */
  16
  17#include <linux/mm.h>
  18#include <linux/proc_fs.h>
  19#include <linux/smp_lock.h>
  20#include <linux/quotaops.h>
  21
  22#include <asm/uaccess.h>
  23#include <asm/unaligned.h>
  24#include <asm/semaphore.h>
  25#include <asm/page.h>
  26#include <asm/pgtable.h>
  27
  28#include <asm/namei.h>
  29
  30/* This can be removed after the beta phase. */
  31#define CACHE_SUPERVISE	/* debug the correctness of dcache entries */
  32#undef DEBUG		/* some other debugging */
  33
  34
  35#define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
  36
  37#ifndef OSKIT
  38/* [Feb-1997 T. Schoebel-Theuer]
  39 * Fundamental changes in the pathname lookup mechanisms (namei)
  40 * were necessary because of omirr.  The reason is that omirr needs
  41 * to know the _real_ pathname, not the user-supplied one, in case
  42 * of symlinks (and also when transname replacements occur).
  43 *
  44 * The new code replaces the old recursive symlink resolution with
  45 * an iterative one (in case of non-nested symlink chains).  It does
  46 * this with calls to <fs>_follow_link().
  47 * As a side effect, dir_namei(), _namei() and follow_link() are now 
  48 * replaced with a single function lookup_dentry() that can handle all 
  49 * the special cases of the former code.
  50 *
  51 * With the new dcache, the pathname is stored at each inode, at least as
  52 * long as the refcount of the inode is positive.  As a side effect, the
  53 * size of the dcache depends on the inode cache and thus is dynamic.
  54 *
  55 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
  56 * resolution to correspond with current state of the code.
  57 *
  58 * Note that the symlink resolution is not *completely* iterative.
  59 * There is still a significant amount of tail- and mid- recursion in
  60 * the algorithm.  Also, note that <fs>_readlink() is not used in
  61 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
  62 * may return different results than <fs>_follow_link().  Many virtual
  63 * filesystems (including /proc) exhibit this behavior.
  64 */
  65
  66/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
  67 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
  68 * and the name already exists in form of a symlink, try to create the new
  69 * name indicated by the symlink. The old code always complained that the
  70 * name already exists, due to not following the symlink even if its target
  71 * is nonexistent.  The new semantics affects also mknod() and link() when
  72 * the name is a symlink pointing to a non-existant name.
  73 *
  74 * I don't know which semantics is the right one, since I have no access
  75 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
  76 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
  77 * "old" one. Personally, I think the new semantics is much more logical.
  78 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
  79 * file does succeed in both HP-UX and SunOs, but not in Solaris
  80 * and in the old Linux semantics.
  81 */
  82
  83/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
  84 * semantics.  See the comments in "open_namei" and "do_link" below.
  85 *
  86 * [10-Sep-98 Alan Modra] Another symlink change.
  87 */
  88
  89/* In order to reduce some races, while at the same time doing additional
  90 * checking and hopefully speeding things up, we copy filenames to the
  91 * kernel data space before using them..
  92 *
  93 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
  94 */
  95static inline int do_getname(const char *filename, char *page)
  96{
  97	int retval;
  98	unsigned long len = PAGE_SIZE;
  99
 100	if ((unsigned long) filename >= TASK_SIZE) {
 101		if (!segment_eq(get_fs(), KERNEL_DS))
 102			return -EFAULT;
 103	} else if (TASK_SIZE - (unsigned long) filename < PAGE_SIZE)
 104		len = TASK_SIZE - (unsigned long) filename;
 105
 106	retval = strncpy_from_user((char *)page, filename, len);
 107	if (retval > 0) {
 108		if (retval < len)
 109			return 0;
 110		return -ENAMETOOLONG;
 111	} else if (!retval)
 112		retval = -ENOENT;
 113	return retval;
 114}
 115#endif /* OSKIT */
 116
 117char * getname(const char * filename)
 118{
 119#ifdef OSKIT
 120	return (char *)filename;
 121#else
 122	char *tmp, *result;
 123
 124	result = ERR_PTR(-ENOMEM);
 125	tmp = __getname();
 126	if (tmp)  {
 127		int retval = do_getname(filename, tmp);
 128
 129		result = tmp;
 130		if (retval < 0) {
 131			putname(tmp);
 132			result = ERR_PTR(retval);
 133		}
 134	}
 135	return result;
 136#endif /* OSKIT */
 137}
 138
 139/*
 140 *	permission()
 141 *
 142 * is used to check for read/write/execute permissions on a file.
 143 * We use "fsuid" for this, letting us set arbitrary permissions
 144 * for filesystem access without changing the "normal" uids which
 145 * are used for other things..
 146 */
 147int permission(struct inode * inode,int mask)
 148{
 149	int mode = inode->i_mode;
 150
 151	if (inode->i_op && inode->i_op->permission)
 152		return inode->i_op->permission(inode, mask);
 153	else if ((mask & S_IWOTH) && IS_RDONLY(inode) &&
 154		 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
 155		return -EROFS; /* Nobody gets write access to a read-only fs */
 156	else if ((mask & S_IWOTH) && IS_IMMUTABLE(inode))
 157		return -EACCES; /* Nobody gets write access to an immutable file */
 158	else if (current->fsuid == inode->i_uid)
 159		mode >>= 6;
 160	else if (in_group_p(inode->i_gid))
 161		mode >>= 3;
 162	if (((mode & mask & S_IRWXO) == mask) || capable(CAP_DAC_OVERRIDE))
 163		return 0;
 164	/* read and search access */
 165	if ((mask == S_IROTH) ||
 166	    (S_ISDIR(mode)  && !(mask & ~(S_IROTH | S_IXOTH))))
 167		if (capable(CAP_DAC_READ_SEARCH))
 168			return 0;
 169	return -EACCES;
 170}
 171
 172/*
 173 * get_write_access() gets write permission for a file.
 174 * put_write_access() releases this write permission.
 175 * This is used for regular files.
 176 * We cannot support write (and maybe mmap read-write shared) accesses and
 177 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
 178 * can have the following values:
 179 * 0: no writers, no VM_DENYWRITE mappings
 180 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
 181 * > 0: (i_writecount) users are writing to the file.
 182 */
 183int get_write_access(struct inode * inode)
 184{
 185#ifdef OSKIT
 186	/* This doesn't need to do anything since i_writecount is only
 187	   so exec can do ETXTBSY. */
 188#else
 189	if (inode->i_writecount < 0)
 190		return -ETXTBSY;
 191	inode->i_writecount++;
 192#endif /* OSKIT */
 193	return 0;
 194}
 195
 196void put_write_access(struct inode * inode)
 197{
 198#ifdef OSKIT
 199	/* This doesn't need to do anything since i_writecount is only
 200	   so exec can do ETXTBSY. */
 201#else
 202	inode->i_writecount--;
 203#endif
 204}
 205
 206/*
 207 * "." and ".." are special - ".." especially so because it has to be able
 208 * to know about the current root directory and parent relationships
 209 */
 210static struct dentry * reserved_lookup(struct dentry * parent, struct qstr * name)
 211{
 212	struct dentry *result = NULL;
 213	if (name->name[0] == '.') {
 214		switch (name->len) {
 215		default:
 216			break;
 217		case 2:	
 218			if (name->name[1] != '.')
 219				break;
 220
 221			if (parent != current->fs->root)
 222				parent = parent->d_covers->d_parent;
 223			/* fallthrough */
 224		case 1:
 225			result = parent;
 226		}
 227	}
 228	return dget(result);
 229}
 230
 231/*
 232 * Internal lookup() using the new generic dcache.
 233 */
 234static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, int flags)
 235{
 236	struct dentry * dentry = d_lookup(parent, name);
 237
 238	if (dentry && dentry->d_op && dentry->d_op->d_revalidate) {
 239		if (!dentry->d_op->d_revalidate(dentry, flags) && !d_invalidate(dentry)) {
 240			dput(dentry);
 241			dentry = NULL;
 242		}
 243	}
 244	return dentry;
 245}
 246
 247/*
 248 * This is called when everything else fails, and we actually have
 249 * to go to the low-level filesystem to find out what we should do..
 250 *
 251 * We get the directory semaphore, and after getting that we also
 252 * make sure that nobody added the entry to the dcache in the meantime..
 253 */
 254static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, int flags)
 255{
 256	struct dentry * result;
 257	struct inode *dir = parent->d_inode;
 258
 259	down(&dir->i_sem);
 260	/*
 261	 * First re-do the cached lookup just in case it was created
 262	 * while we waited for the directory semaphore..
 263	 *
 264	 * FIXME! This could use version numbering or similar to
 265	 * avoid unnecessary cache lookups.
 266	 */
 267	result = cached_lookup(parent, name, flags);
 268	if (!result) {
 269		struct dentry * dentry = d_alloc(parent, name);
 270		result = ERR_PTR(-ENOMEM);
 271		if (dentry) {
 272			result = dir->i_op->lookup(dir, dentry);
 273			if (result)
 274				dput(dentry);
 275			else
 276				result = dentry;
 277		}
 278	}
 279	up(&dir->i_sem);
 280	return result;
 281}
 282
 283static struct dentry * do_follow_link(struct dentry *base, struct dentry *dentry, unsigned int follow)
 284{
 285	struct inode * inode = dentry->d_inode;
 286
 287	if ((follow & LOOKUP_FOLLOW)
 288	    && inode && inode->i_op && inode->i_op->follow_link) {
 289		if (current->link_count < 5) {
 290			struct dentry * result;
 291
 292			current->link_count++;
 293			/* This eats the base */
 294			result = inode->i_op->follow_link(dentry, base, follow);
 295			current->link_count--;
 296			dput(dentry);
 297			return result;
 298		}
 299		dput(dentry);
 300		dentry = ERR_PTR(-ELOOP);
 301	}
 302	dput(base);
 303	return dentry;
 304}
 305
 306static inline struct dentry * follow_mount(struct dentry * dentry)
 307{
 308	struct dentry * mnt = dentry->d_mounts;
 309
 310	if (mnt != dentry) {
 311		dget(mnt);
 312		dput(dentry);
 313		dentry = mnt;
 314	}
 315	return dentry;
 316}
 317
 318/*
 319 * Name resolution.
 320 *
 321 * This is the basic name resolution function, turning a pathname
 322 * into the final dentry.
 323 */
 324struct dentry * lookup_dentry(const char * name, struct dentry * base, unsigned int lookup_flags)
 325{
 326	struct dentry * dentry;
 327	struct inode *inode;
 328
 329	if (*name == '/') {
 330		if (base)
 331			dput(base);
 332		do {
 333			name++;
 334		} while (*name == '/');
 335		__prefix_lookup_dentry(name, lookup_flags);
 336		base = dget(current->fs->root);
 337	} else if (!base) {
 338		base = dget(current->fs->pwd);
 339	}
 340
 341	if (!*name)
 342		goto return_base;
 343
 344	inode = base->d_inode;
 345	lookup_flags &= LOOKUP_FOLLOW | LOOKUP_DIRECTORY | LOOKUP_SLASHOK;
 346
 347	/* At this point we know we have a real path component. */
 348	for(;;) {
 349		int err;
 350		unsigned long hash;
 351		struct qstr this;
 352		unsigned int flags;
 353		unsigned int c;
 354
 355		err = permission(inode, MAY_EXEC);
 356		dentry = ERR_PTR(err);
 357 		if (err)
 358			break;
 359
 360		this.name = name;
 361		c = *(const unsigned char *)name;
 362
 363		hash = init_name_hash();
 364		do {
 365			name++;
 366			hash = partial_name_hash(c, hash);
 367			c = *(const unsigned char *)name;
 368		} while (c && (c != '/'));
 369		this.len = name - (const char *) this.name;
 370		this.hash = end_name_hash(hash);
 371
 372		/* remove trailing slashes? */
 373		flags = lookup_flags;
 374		if (c) {
 375			char tmp;
 376
 377			flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 378			do {
 379				tmp = *++name;
 380			} while (tmp == '/');
 381			if (tmp)
 382				flags |= LOOKUP_CONTINUE;
 383		}
 384
 385		/*
 386		 * See if the low-level filesystem might want
 387		 * to use its own hash..
 388		 */
 389		if (base->d_op && base->d_op->d_hash) {
 390			int error;
 391			error = base->d_op->d_hash(base, &this);
 392			if (error < 0) {
 393				dentry = ERR_PTR(error);
 394				break;
 395			}
 396		}
 397
 398		/* This does the actual lookups.. */
 399		dentry = reserved_lookup(base, &this);
 400		if (!dentry) {
 401			dentry = cached_lookup(base, &this, flags);
 402			if (!dentry) {
 403				dentry = real_lookup(base, &this, flags);
 404				if (IS_ERR(dentry))
 405					break;
 406			}
 407		}
 408
 409		/* Check mountpoints.. */
 410		dentry = follow_mount(dentry);
 411
 412		base = do_follow_link(base, dentry, flags);
 413		if (IS_ERR(base))
 414			goto return_base;
 415
 416		inode = base->d_inode;
 417		if (flags & LOOKUP_DIRECTORY) {
 418			if (!inode)
 419				goto no_inode;
 420			dentry = ERR_PTR(-ENOTDIR); 
 421			if (!inode->i_op || !inode->i_op->lookup)
 422				break;
 423			if (flags & LOOKUP_CONTINUE)
 424				continue;
 425		}
 426return_base:
 427		return base;
 428/*
 429 * The case of a nonexisting file is special.
 430 *
 431 * In the middle of a pathname lookup (ie when
 432 * LOOKUP_CONTINUE is set), it's an obvious
 433 * error and returns ENOENT.
 434 *
 435 * At the end of a pathname lookup it's legal,
 436 * and we return a negative dentry. However, we
 437 * get here only if there were trailing slashes,
 438 * which is legal only if we know it's supposed
 439 * to be a directory (ie "mkdir"). Thus the
 440 * LOOKUP_SLASHOK flag.
 441 */
 442no_inode:
 443		dentry = ERR_PTR(-ENOENT);
 444		if (flags & LOOKUP_CONTINUE)
 445			break;
 446		if (flags & LOOKUP_SLASHOK)
 447			goto return_base;
 448		break;
 449	}
 450	dput(base);
 451	return dentry;
 452}
 453
 454/*
 455 *	namei()
 456 *
 457 * is used by most simple commands to get the inode of a specified name.
 458 * Open, link etc use their own routines, but this is enough for things
 459 * like 'chmod' etc.
 460 *
 461 * namei exists in two versions: namei/lnamei. The only difference is
 462 * that namei follows links, while lnamei does not.
 463 */
 464struct dentry * __namei(const char *pathname, unsigned int lookup_flags)
 465{
 466	char *name;
 467	struct dentry *dentry;
 468
 469	name = getname(pathname);
 470	dentry = (struct dentry *) name;
 471	if (!IS_ERR(name)) {
 472		dentry = lookup_dentry(name, NULL, lookup_flags);
 473		putname(name);
 474		if (!IS_ERR(dentry)) {
 475			if (!dentry->d_inode) {
 476				dput(dentry);
 477				dentry = ERR_PTR(-ENOENT);
 478			}
 479		}
 480	}
 481	return dentry;
 482}
 483
 484/*
 485 * It's inline, so penalty for filesystems that don't use sticky bit is
 486 * minimal.
 487 */
 488static inline int check_sticky(struct inode *dir, struct inode *inode)
 489{
 490	if (!(dir->i_mode & S_ISVTX))
 491		return 0;
 492	if (inode->i_uid == current->fsuid)
 493		return 0;
 494	if (dir->i_uid == current->fsuid)
 495		return 0;
 496	return !capable(CAP_FOWNER);
 497}
 498
 499/*
 500 *	Check whether we can remove a link victim from directory dir, check
 501 *  whether the type of victim is right.
 502 *  1. We can't do it if dir is read-only (done in permission())
 503 *  2. We should have write and exec permissions on dir
 504 *  3. We can't remove anything from append-only dir
 505 *  4. We can't do anything with immutable dir (done in permission())
 506 *  5. If the sticky bit on dir is set we should either
 507 *	a. be owner of dir, or
 508 *	b. be owner of victim, or
 509 *	c. have CAP_FOWNER capability
 510 *  6. If the victim is append-only or immutable we can't do antyhing with
 511 *     links pointing to it.
 512 *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 513 *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 514 *  9. We can't remove a root or mountpoint.
 515 */
 516static inline int may_delete(struct inode *dir,struct dentry *victim, int isdir)
 517{
 518	int error;
 519	if (!victim->d_inode || victim->d_parent->d_inode != dir)
 520		return -ENOENT;
 521	error = permission(dir,MAY_WRITE | MAY_EXEC);
 522	if (error)
 523		return error;
 524	if (IS_APPEND(dir))
 525		return -EPERM;
 526	if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
 527	    IS_IMMUTABLE(victim->d_inode))
 528		return -EPERM;
 529	if (isdir) {
 530		if (!S_ISDIR(victim->d_inode->i_mode))
 531			return -ENOTDIR;
 532		if (IS_ROOT(victim))
 533			return -EBUSY;
 534		if (victim->d_mounts != victim->d_covers)
 535			return -EBUSY;
 536	} else if (S_ISDIR(victim->d_inode->i_mode))
 537		return -EISDIR;
 538	return 0;
 539}
 540
 541/*	Check whether we can create an object with dentry child in directory
 542 *  dir.
 543 *  1. We can't do it if child already exists (open has special treatment for
 544 *     this case, but since we are inlined it's OK)
 545 *  2. We can't do it if dir is read-only (done in permission())
 546 *  3. We should have write and exec permissions on dir
 547 *  4. We can't do it if dir is immutable (done in permission())
 548 */
 549static inline int may_create(struct inode *dir, struct dentry *child) {
 550	if (child->d_inode)
 551		return -EEXIST;
 552	return permission(dir,MAY_WRITE | MAY_EXEC);
 553}
 554
 555static inline struct dentry *get_parent(struct dentry *dentry)
 556{
 557	return dget(dentry->d_parent);
 558}
 559
 560static inline void unlock_dir(struct dentry *dir)
 561{
 562	up(&dir->d_inode->i_sem);
 563	dput(dir);
 564}
 565
 566/*
 567 * We need to do a check-parent every time
 568 * after we have locked the parent - to verify
 569 * that the parent is still our parent and
 570 * that we are still hashed onto it..
 571 *
 572 * This is requied in case two processes race
 573 * on removing (or moving) the same entry: the
 574 * parent lock will serialize them, but the
 575 * other process will be too late..
 576 */
 577#define check_parent(dir, dentry) \
 578	((dir) == (dentry)->d_parent && !list_empty(&dentry->d_hash))
 579
 580/*
 581 * Locking the parent is needed to:
 582 *  - serialize directory operations
 583 *  - make sure the parent doesn't change from
 584 *    under us in the middle of an operation.
 585 *
 586 * NOTE! Right now we'd rather use a "struct inode"
 587 * for this, but as I expect things to move toward
 588 * using dentries instead for most things it is
 589 * probably better to start with the conceptually
 590 * better interface of relying on a path of dentries.
 591 */
 592static inline struct dentry *lock_parent(struct dentry *dentry)
 593{
 594	struct dentry *dir = dget(dentry->d_parent);
 595
 596	down(&dir->d_inode->i_sem);
 597	return dir;
 598}
 599
 600/*
 601 * Whee.. Deadlock country. Happily there are only two VFS
 602 * operations that do this..
 603 */
 604static inline void double_lock(struct dentry *d1, struct dentry *d2)
 605{
 606	struct semaphore *s1 = &d1->d_inode->i_sem;
 607	struct semaphore *s2 = &d2->d_inode->i_sem;
 608
 609	if (s1 != s2) {
 610		if ((unsigned long) s1 < (unsigned long) s2) {
 611			struct semaphore *tmp = s2;
 612			s2 = s1; s1 = tmp;
 613		}
 614		down(s1);
 615	}
 616	down(s2);
 617}
 618
 619static inline void double_unlock(struct dentry *d1, struct dentry *d2)
 620{
 621	struct semaphore *s1 = &d1->d_inode->i_sem;
 622	struct semaphore *s2 = &d2->d_inode->i_sem;
 623
 624	up(s1);
 625	if (s1 != s2)
 626		up(s2);
 627	dput(d1);
 628	dput(d2);
 629}
 630
 631
 632/* 
 633 * Special case: O_CREAT|O_EXCL implies O_NOFOLLOW for security
 634 * reasons.
 635 *
 636 * O_DIRECTORY translates into forcing a directory lookup.
 637 */
 638static inline int lookup_flags(unsigned int f)
 639{
 640	unsigned long retval = LOOKUP_FOLLOW;
 641
 642	if (f & O_NOFOLLOW)
 643		retval &= ~LOOKUP_FOLLOW;
 644	
 645	if ((f & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
 646		retval &= ~LOOKUP_FOLLOW;
 647	
 648	if (f & O_DIRECTORY)
 649		retval |= LOOKUP_DIRECTORY;
 650	
 651	return retval;
 652}
 653
 654/*
 655 *	open_namei()
 656 *
 657 * namei for open - this is in fact almost the whole open-routine.
 658 *
 659 * Note that the low bits of "flag" aren't the same as in the open
 660 * system call - they are 00 - no permissions needed
 661 *			  01 - read permission needed
 662 *			  10 - write permission needed
 663 *			  11 - read/write permissions needed
 664 * which is a lot more logical, and also allows the "no perm" needed
 665 * for symlinks (where the permissions are checked later).
 666 */
 667struct dentry * open_namei(const char * pathname, int flag, int mode)
 668{
 669	int acc_mode, error;
 670	struct inode *inode;
 671	struct dentry *dentry;
 672
 673	mode &= S_IALLUGO & ~current->fs->umask;
 674	mode |= S_IFREG;
 675
 676	dentry = lookup_dentry(pathname, NULL, lookup_flags(flag));
 677	if (IS_ERR(dentry))
 678		return dentry;
 679
 680	acc_mode = ACC_MODE(flag);
 681	if (flag & O_CREAT) {
 682		struct dentry *dir;
 683
 684		if (dentry->d_inode) {
 685			if (!(flag & O_EXCL))
 686				goto nocreate;
 687			error = -EEXIST;
 688			goto exit;
 689		}
 690
 691		dir = lock_parent(dentry);
 692		if (!check_parent(dir, dentry)) {
 693			/*
 694			 * Really nasty race happened. What's the 
 695			 * right error code? We had a dentry, but
 696			 * before we could use it it was removed
 697			 * by somebody else. We could just re-try
 698			 * everything, I guess.
 699			 *
 700			 * ENOENT is definitely wrong.
 701			 */
 702			error = -ENOENT;
 703			unlock_dir(dir);
 704			goto exit;
 705		}
 706
 707		/*
 708		 * Somebody might have created the file while we
 709		 * waited for the directory lock.. So we have to
 710		 * re-do the existence test.
 711		 */
 712		if (dentry->d_inode) {
 713			error = 0;
 714			if (flag & O_EXCL)
 715				error = -EEXIST;
 716		} else if ((error = may_create(dir->d_inode, dentry)) == 0) {
 717			if (!dir->d_inode->i_op || !dir->d_inode->i_op->create)
 718				error = -EACCES;
 719			else {
 720				DQUOT_INIT(dir->d_inode);
 721				error = dir->d_inode->i_op->create(dir->d_inode, dentry, mode);
 722				/* Don't check for write permission, don't truncate */
 723				acc_mode = 0;
 724				flag &= ~O_TRUNC;
 725			}
 726		}
 727		unlock_dir(dir);
 728		if (error)
 729			goto exit;
 730	}
 731
 732nocreate:
 733	error = -ENOENT;
 734	inode = dentry->d_inode;
 735	if (!inode)
 736		goto exit;
 737
 738	error = -ELOOP;
 739	if (S_ISLNK(inode->i_mode))
 740		goto exit;
 741	
 742	error = -EISDIR;
 743	if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
 744		goto exit;
 745
 746	error = permission(inode,acc_mode);
 747	if (error)
 748		goto exit;
 749
 750	/*
 751	 * FIFO's, sockets and device files are special: they don't
 752	 * actually live on the filesystem itself, and as such you
 753	 * can write to them even if the filesystem is read-only.
 754	 */
 755	if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
 756	    	flag &= ~O_TRUNC;
 757	} else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
 758		error = -EACCES;
 759		if (IS_NODEV(inode))
 760			goto exit;
 761
 762		flag &= ~O_TRUNC;
 763	} else {
 764		error = -EROFS;
 765		if (IS_RDONLY(inode) && (flag & 2))
 766			goto exit;
 767	}
 768	/*
 769	 * An append-only file must be opened in append mode for writing.
 770	 */
 771	error = -EPERM;
 772	if (IS_APPEND(inode)) {
 773		if  ((flag & FMODE_WRITE) && !(flag & O_APPEND))
 774			goto exit;
 775		if (flag & O_TRUNC)
 776			goto exit;
 777	}
 778
 779	if (flag & O_TRUNC) {
 780		error = get_write_access(inode);
 781		if (error)
 782			goto exit;
 783
 784		/*
 785		 * Refuse to truncate files with mandatory locks held on them.
 786		 */
 787		error = locks_verify_locked(inode);
 788		if (!error) {
 789			DQUOT_INIT(inode);
 790			
 791			error = do_truncate(dentry, 0);
 792		}
 793		put_write_access(inode);
 794		if (error)
 795			goto exit;
 796	} else
 797		if (flag & FMODE_WRITE)
 798			DQUOT_INIT(inode);
 799
 800	return dentry;
 801
 802exit:
 803	dput(dentry);
 804	return ERR_PTR(error);
 805}
 806
 807struct dentry * do_mknod(const char * filename, int mode, dev_t dev)
 808{
 809	int error;
 810	struct dentry *dir;
 811	struct dentry *dentry, *retval;
 812
 813	mode &= ~current->fs->umask;
 814	dentry = lookup_dentry(filename, NULL, LOOKUP_FOLLOW);
 815	if (IS_ERR(dentry))
 816		return dentry;
 817
 818	dir = lock_parent(dentry);
 819	error = -ENOENT;
 820	if (!check_parent(dir, dentry))
 821		goto exit_lock;
 822
 823	error = may_create(dir->d_inode, dentry);
 824	if (error)
 825		goto exit_lock;
 826
 827	error = -EPERM;
 828	if (!dir->d_inode->i_op || !dir->d_inode->i_op->mknod)
 829		goto exit_lock;
 830
 831	DQUOT_INIT(dir->d_inode);
 832	error = dir->d_inode->i_op->mknod(dir->d_inode, dentry, mode, dev);
 833exit_lock:
 834	retval = ERR_PTR(error);
 835	if (!error)
 836		retval = dget(dentry);
 837	unlock_dir(dir);
 838	dput(dentry);
 839	return retval;
 840}
 841
 842asmlinkage int sys_mknod(const char * filename, int mode, dev_t dev)
 843{
 844	int error;
 845	char * tmp;
 846
 847	lock_kernel();
 848	error = -EPERM;
 849	if (S_ISDIR(mode) || (!S_ISFIFO(mode) && !capable(CAP_SYS_ADMIN)))
 850		goto out;
 851	error = -EINVAL;
 852	switch (mode & S_IFMT) {
 853	case 0:
 854		mode |= S_IFREG;
 855		break;
 856	case S_IFREG: case S_IFCHR: case S_IFBLK: case S_IFIFO: case S_IFSOCK:
 857		break;
 858	default:
 859		goto out;
 860	}
 861	tmp = getname(filename);
 862	error = PTR_ERR(tmp);
 863	if (!IS_ERR(tmp)) {
 864		struct dentry * dentry = do_mknod(tmp,mode,dev);
 865		putname(tmp);
 866		error = PTR_ERR(dentry);
 867		if (!IS_ERR(dentry)) {
 868			dput(dentry);
 869			error = 0;
 870		}
 871	}
 872out:
 873	unlock_kernel();
 874	return error;
 875}
 876
 877/*
 878 * Look out: this function may change a normal dentry
 879 * into a directory dentry (different size)..
 880 */
 881#ifdef OSKIT
 882       int do_mkdir(const char * pathname, int mode)
 883#else
 884static inline int do_mkdir(const char * pathname, int mode)
 885#endif
 886{
 887	int error;
 888	struct dentry *dir;
 889	struct dentry *dentry;
 890
 891	dentry = lookup_dentry(pathname, NULL, LOOKUP_SLASHOK);
 892	error = PTR_ERR(dentry);
 893	if (IS_ERR(dentry))
 894		goto exit;
 895
 896	/*
 897	 * EEXIST is kind of a strange error code to
 898	 * return, but basically if the dentry was moved
 899	 * or unlinked while we locked the parent, we
 900	 * do know that it _did_ exist before, and as
 901	 * such it makes perfect sense.. In contrast,
 902	 * ENOENT doesn't make sense for mkdir.
 903	 */
 904	dir = lock_parent(dentry);
 905	error = -EEXIST;
 906	if (!check_parent(dir, dentry))
 907		goto exit_lock;
 908
 909	error = may_create(dir->d_inode, dentry);
 910	if (error)
 911		goto exit_lock;
 912
 913	error = -EPERM;
 914	if (!dir->d_inode->i_op || !dir->d_inode->i_op->mkdir)
 915		goto exit_lock;
 916
 917	DQUOT_INIT(dir->d_inode);
 918	mode &= 0777 & ~current->fs->umask;
 919	error = dir->d_inode->i_op->mkdir(dir->d_inode, dentry, mode);
 920
 921exit_lock:
 922	unlock_dir(dir);
 923	dput(dentry);
 924exit:
 925	return error;
 926}
 927
 928asmlinkage int sys_mkdir(const char * pathname, int mode)
 929{
 930	int error;
 931	char * tmp;
 932
 933	lock_kernel();
 934	tmp = getname(pathname);
 935	error = PTR_ERR(tmp);
 936	if (!IS_ERR(tmp)) {
 937		error = do_mkdir(tmp,mode);
 938		putname(tmp);
 939	}
 940	unlock_kernel();
 941	return error;
 942}
 943
 944int vfs_rmdir(struct inode *dir, struct dentry *dentry)
 945{
 946	int error;
 947
 948	error = may_delete(dir, dentry, 1);
 949	if (error)
 950		return error;
 951
 952	if (!dir->i_op || !dir->i_op->rmdir)
 953		return -EPERM;
 954
 955	DQUOT_INIT(dir);
 956
 957	/*
 958	 * We try to drop the dentry early: we should have
 959	 * a usage count of 2 if we're the only user of this
 960	 * dentry, and if that is true (possibly after pruning
 961	 * the dcache), then we drop the dentry now.
 962	 *
 963	 * A low-level filesystem can, if it choses, legally
 964	 * do a
 965	 *
 966	 *	if (!list_empty(&dentry->d_hash))
 967	 *		return -EBUSY;
 968	 *
 969	 * if it cannot handle the case of removing a directory
 970	 * that is still in use by something else..
 971	 */
 972	switch (dentry->d_count) {
 973	default:
 974		shrink_dcache_parent(dentry);
 975		if (dentry->d_count != 2)
 976			break;
 977	case 2:
 978		d_drop(dentry);
 979	}
 980
 981	error = dir->i_op->rmdir(dir, dentry);
 982
 983	return error;
 984}
 985
 986#ifdef OSKIT
 987       int do_rmdir(const char * name)
 988#else
 989static inline int do_rmdir(const char * name)
 990#endif
 991{
 992	int error;
 993	struct dentry *dir;
 994	struct dentry *dentry;
 995
 996	dentry = lookup_dentry(name, NULL, 0);
 997	error = PTR_ERR(dentry);
 998	if (IS_ERR(dentry))
 999		goto exit;
1000
1001	error = -ENOENT;
1002	if (!dentry->d_inode)
1003		goto exit_dput;
1004
1005	dir = dget(dentry->d_parent);
1006
1007	/*
1008	 * The dentry->d_count stuff confuses d_delete() enough to
1009	 * not kill the inode from under us while it is locked. This
1010	 * wouldn't be needed, except the dentry semaphore is really
1011	 * in the inode, not in the dentry..
1012	 */
1013	dentry->d_count++;
1014	double_lock(dir, dentry);
1015
1016	error = -ENOENT;
1017	if (check_parent(dir, dentry))
1018		error = vfs_rmdir(dir->d_inode, dentry);
1019
1020	double_unlock(dentry, dir);
1021exit_dput:
1022	dput(dentry);
1023exit:
1024	return error;
1025}
1026
1027asmlinkage int sys_rmdir(const char * pathname)
1028{
1029	int error;
1030	char * tmp;
1031
1032	lock_kernel();
1033	tmp = getname(pathname);
1034	error = PTR_ERR(tmp);
1035	if (!IS_ERR(tmp)) {
1036		error = do_rmdir(tmp);
1037		putname(tmp);
1038	}
1039	unlock_kernel();
1040	return error;
1041}
1042
1043int vfs_unlink(struct inode *dir, struct dentry *dentry)
1044{
1045	int error;
1046
1047	error = may_delete(dir, dentry, 0);
1048	if (!error) {
1049		error = -EPERM;
1050		if (dir->i_op && dir->i_op->unlink) {
1051			DQUOT_INIT(dir);
1052			error = dir->i_op->unlink(dir, dentry);
1053		}
1054	}
1055	return error;
1056}
1057
1058#ifdef OSKIT
1059       int do_unlink(const char * name)
1060#else
1061static inline int do_unlink(const char * name)
1062#endif
1063{
1064	int error;
1065	struct dentry *dir;
1066	struct dentry *dentry;
1067
1068	dentry = lookup_dentry(name, NULL, 0);
1069	error = PTR_ERR(dentry);
1070	if (IS_ERR(dentry))
1071		goto exit;
1072
1073	dir = lock_parent(dentry);
1074	error = -ENOENT;
1075	if (check_parent(dir, dentry))
1076		error = vfs_unlink(dir->d_inode, dentry);
1077
1078        unlock_dir(dir);
1079	dput(dentry);
1080exit:
1081	return error;
1082}
1083
1084asmlinkage int sys_unlink(const char * pathname)
1085{
1086	int error;
1087	char * tmp;
1088
1089	lock_kernel();
1090	tmp = getname(pathname);
1091	error = PTR_ERR(tmp);
1092	if (!IS_ERR(tmp)) {
1093		error = do_unlink(tmp);
1094		putname(tmp);
1095	}
1096	unlock_kernel();
1097	return error;
1098}
1099
1100#ifdef OSKIT
1101       int do_symlink(const char * oldname, const char * newname)
1102#else
1103static inline int do_symlink(const char * oldname, const char * newname)
1104#endif
1105{
1106	int error;
1107	struct dentry *dir;
1108	struct dentry *dentry;
1109
1110	dentry = lookup_dentry(newname, NULL, 0);
1111
1112	error = PTR_ERR(dentry);
1113	if (IS_ERR(dentry))
1114		goto exit;
1115
1116	dir = lock_parent(dentry);
1117	error = -ENOENT;
1118	if (!check_parent(dir, dentry))
1119		goto exit_lock;
1120
1121	error = may_create(dir->d_inode, dentry);
1122	if (error)
1123		goto exit_lock;
1124
1125	error = -EPERM;
1126	if (!dir->d_inode->i_op || !dir->d_inode->i_op->symlink)
1127		goto exit_lock;
1128
1129	DQUOT_INIT(dir->d_inode);
1130	error = dir->d_inode->i_op->symlink(dir->d_inode, dentry, oldname);
1131
1132exit_lock:
1133	unlock_dir(dir);
1134	dput(dentry);
1135exit:
1136	return error;
1137}
1138
1139asmlinkage int sys_symlink(const char * oldname, const char * newname)
1140{
1141	int error;
1142	char * from;
1143
1144	lock_kernel();
1145	from = getname(oldname);
1146	error = PTR_ERR(from);
1147	if (!IS_ERR(from)) {
1148		char * to;
1149		to = getname(newname);
1150		error = PTR_ERR(to);
1151		if (!IS_ERR(to)) {
1152			error = do_symlink(from,to);
1153			putname(to);
1154		}
1155		putname(from);
1156	}
1157	unlock_kernel();
1158	return error;
1159}
1160
1161#ifdef OSKIT
1162       int do_link(struct dentry *dentry, const char * newname)
1163#else
1164static inline int do_link(const char * oldname, const char * newname)
1165#endif
1166{
1167	struct dentry *old_dentry, *new_dentry, *dir;
1168	struct inode *inode;
1169	int error;
1170
1171	/*
1172	 * Hardlinks are often used in delicate situations.  We avoid
1173	 * security-related surprises by not following symlinks on the
1174	 * newname.  --KAB
1175	 *
1176	 * We don't follow them on the oldname either to be compatible
1177	 * with linux 2.0, and to avoid hard-linking to directories
1178	 * and other special files.  --ADM
1179	 */
1180#ifdef OSKIT
1181	dget(dentry);
1182	old_dentry = dentry;
1183#else
1184	old_dentry = lookup_dentry(oldname, NULL, 0);
1185#endif
1186	error = PTR_ERR(old_dentry);
1187	if (IS_ERR(old_dentry))
1188		goto exit;
1189
1190	new_dentry = lookup_dentry(newname, NULL, 0);
1191	error = PTR_ERR(new_dentry);
1192	if (IS_ERR(new_dentry))
1193		goto exit_old;
1194
1195	dir = lock_parent(new_dentry);
1196	error = -ENOENT;
1197	if (!check_parent(dir, new_dentry))
1198		goto exit_lock;
1199
1200	error = -ENOENT;
1201	inode = old_dentry->d_inode;
1202	if (!inode)
1203		goto exit_lock;
1204
1205	error = may_create(dir->d_inode, new_dentry);
1206	if (error)
1207		goto exit_lock;
1208
1209	error = -EXDEV;
1210	if (dir->d_inode->i_dev != inode->i_dev)
1211		goto exit_lock;
1212
1213	/*
1214	 * A link to an append-only or immutable file cannot be created.
1215	 */
1216	error = -EPERM;
1217	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1218		goto exit_lock;
1219
1220	error = -EPERM;
1221	if (!dir->d_inode->i_op || !dir->d_inode->i_op->link)
1222		goto exit_lock;
1223
1224	DQUOT_INIT(dir->d_inode);
1225	error = dir->d_inode->i_op->link(old_dentry, dir->d_inode, new_dentry);
1226
1227exit_lock:
1228	unlock_dir(dir);
1229	dput(new_dentry);
1230exit_old:
1231	dput(old_dentry);
1232exit:
1233	return error;
1234}
1235
1236#ifndef OSKIT
1237asmlinkage int sys_link(const char * oldname, const char * newname)
1238{
1239	int error;
1240	char * from;
1241
1242	lock_kernel();
1243	from = getname(oldname);
1244	error = PTR_ERR(from);
1245	if (!IS_ERR(from)) {
1246		char * to;
1247		to = getname(newname);
1248		error = PTR_ERR(to);
1249		if (!IS_ERR(to)) {
1250			error = do_link(from,to);
1251			putname(to);
1252		}
1253		putname(from);
1254	}
1255	unlock_kernel();
1256	return error;
1257}
1258#endif
1259
1260int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
1261	       struct inode *new_dir, struct dentry *new_dentry)
1262{
1263	int error;
1264	int need_rehash = 0;
1265
1266	if (old_dentry->d_inode == new_dentry->d_inode)
1267		return 0;
1268
1269	error = may_delete(old_dir, old_dentry, 1);
1270	if (error)
1271		return error;
1272
1273	if (new_dir->i_dev != old_dir->i_dev)
1274		return -EXDEV;
1275
1276	if (!new_dentry->d_inode)
1277		error = may_create(new_dir, new_dentry);
1278	else
1279		error = may_delete(new_dir, new_dentry, 1);
1280	if (error)
1281		return error;
1282
1283	if (!old_dir->i_op || !old_dir->i_op->rename)
1284		return -EPERM;
1285
1286	/*
1287	 * If we are going to change the parent - check write permissions,
1288	 * we'll need to flip '..'.
1289	 */
1290	if (new_dir != old_dir) {
1291		error = permission(old_dentry->d_inode, MAY_WRITE);
1292	}
1293	if (error)
1294		return error;
1295
1296	DQUOT_INIT(old_dir);
1297	DQUOT_INIT(new_dir);
1298	down(&old_dir->i_sb->s_vfs_rename_sem);
1299	error = -EINVAL;
1300	if (is_subdir(new_dentry, old_dentry))
1301		goto out_unlock;
1302	if (new_dentry->d_inode) {
1303		error = -EBUSY;
1304		if (d_invalidate(new_dentry)<0)
1305			goto out_unlock;
1306		need_rehash = 1;
1307	}
1308	error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
1309	if (need_rehash)
1310		d_rehash(new_dentry);
1311	if (!error)
1312		d_move(old_dentry,new_dentry);
1313out_unlock:
1314	up(&old_dir->i_sb->s_vfs_rename_sem);
1315	return error;
1316}
1317
1318int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
1319	       struct inode *new_dir, struct dentry *new_dentry)
1320{
1321	int error;
1322
1323	if (old_dentry->d_inode == new_dentry->d_inode)
1324		return 0;
1325
1326	error = may_delete(old_dir, old_dentry, 0);
1327	if (error)
1328		return error;
1329
1330	if (new_dir->i_dev != old_dir->i_dev)
1331		return -EXDEV;
1332
1333	if (!new_dentry->d_inode)
1334		error = may_create(new_dir, new_dentry);
1335	else
1336		error = may_delete(new_dir, new_dentry, 0);
1337	if (error)
1338		return error;
1339
1340	if (!old_dir->i_op || !old_dir->i_op->rename)
1341		return -EPERM;
1342
1343	DQUOT_INIT(old_dir);
1344	DQUOT_INIT(new_dir);
1345	error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
1346	if (error)
1347		return error;
1348	/* The following d_move() should become unconditional */
1349	if (!(old_dir->i_sb->s_flags & MS_ODD_RENAME)) {
1350		d_move(old_dentry, new_dentry);
1351	}
1352	return 0;
1353}
1354
1355int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1356	       struct inode *new_dir, struct dentry *new_dentry)
1357{
1358	if (S_ISDIR(old_dentry->d_inode->i_mode))
1359		return vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
1360	else
1361		return vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
1362}
1363
1364#ifdef OSKIT
1365       int do_rename(const char * oldname, const char * newname)
1366#else
1367static inline int do_rename(const char * oldname, const char * newname)
1368#endif
1369{
1370	int error;
1371	struct dentry * old_dir, * new_dir;
1372	struct dentry * old_dentry, *new_dentry;
1373
1374	old_dentry = lookup_dentry(oldname, NULL, 0);
1375
1376	error = PTR_ERR(old_dentry);
1377	if (IS_ERR(old_dentry))
1378		goto exit;
1379
1380	error = -ENOENT;
1381	if (!old_dentry->d_inode)
1382		goto exit_old;
1383
1384	{
1385		unsigned int flags = 0;
1386		if (S_ISDIR(old_dentry->d_inode->i_mode))
1387			flags = LOOKUP_SLASHOK;
1388		new_dentry = lookup_dentry(newname, NULL, flags);
1389	}
1390
1391	error = PTR_ERR(new_dentry);
1392	if (IS_ERR(new_dentry))
1393		goto exit_old;
1394
1395	new_dir = get_parent(new_dentry);
1396	old_dir = get_parent(old_dentry);
1397
1398	double_lock(new_dir, old_dir);
1399
1400	error = -ENOENT;
1401	if (check_parent(old_dir, old_dentry) && check_parent(new_dir, new_dentry))
1402		error = vfs_rename(old_dir->d_inode, old_dentry,
1403				   new_dir->d_inode, new_dentry);
1404
1405	double_unlock(new_dir, old_dir);
1406	dput(new_dentry);
1407exit_old:
1408	dput(old_dentry);
1409exit:
1410	return error;
1411}
1412
1413asmlinkage int sys_rename(const char * oldname, const char * newname)
1414{
1415	int error;
1416	char * from;
1417
1418	lock_kernel();
1419	from = getname(oldname);
1420	error = PTR_ERR(from);
1421	if (!IS_ERR(from)) {
1422		char * to;
1423		to = getname(newname);
1424		error = PTR_ERR(to);
1425		if (!IS_ERR(to)) {
1426			error = do_rename(from,to);
1427			putname(to);
1428		}
1429		putname(from);
1430	}
1431	unlock_kernel();
1432	return error;
1433}