[thirdparty/glibc.git] / sysdeps / unix / sysv / linux / getdents.c

/* Copyright (C) 1993-2016 Free Software Foundation, Inc.
   This file is part of the GNU C Library.

   The GNU C Library is free software; you can redistribute it and/or
   modify it under the terms of the GNU Lesser General Public
   License as published by the Free Software Foundation; either
   version 2.1 of the License, or (at your option) any later version.

   The GNU C Library is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
   Lesser General Public License for more details.

   You should have received a copy of the GNU Lesser General Public
   License along with the GNU C Library; if not, see
   <http://www.gnu.org/licenses/>.  */

#include <alloca.h>
#include <assert.h>
#include <errno.h>
#include <dirent.h>
#include <stddef.h>
#include <stdint.h>
#include <string.h>
#include <unistd.h>
#include <sys/param.h>
#include <sys/types.h>

#include <sysdep.h>
#include <sys/syscall.h>

#include <linux/posix_types.h>

#include <kernel-features.h>

/* For Linux we need a special version of this file since the
   definition of `struct dirent' is not the same for the kernel and
   the libc.  There is one additional field which might be introduced
   in the kernel structure in the future.

   Here is the kernel definition of `struct dirent' as of 2.1.20:  */

struct kernel_dirent
  {
    long int d_ino;
    __kernel_off_t d_off;
    unsigned short int d_reclen;
    char d_name[256];
  };

struct kernel_dirent64
  {
    uint64_t		d_ino;
    int64_t		d_off;
    unsigned short int	d_reclen;
    unsigned char	d_type;
    char		d_name[256];
  };

#ifndef __GETDENTS
# define __GETDENTS __getdents
#endif
#ifndef DIRENT_TYPE
# define DIRENT_TYPE struct dirent
#endif
#ifndef DIRENT_SET_DP_INO
# define DIRENT_SET_DP_INO(dp, value) (dp)->d_ino = (value)
#endif

/* The problem here is that we cannot simply read the next NBYTES
   bytes.  We need to take the additional field into account.  We use
   some heuristic.  Assuming the directory contains names with 14
   characters on average we can compute an estimated number of entries
   which fit in the buffer.  Taking this number allows us to specify a
   reasonable number of bytes to read.  If we should be wrong, we can
   reset the file descriptor.  In practice the kernel is limiting the
   amount of data returned much more then the reduced buffer size.  */
ssize_t
internal_function
__GETDENTS (int fd, char *buf, size_t nbytes)
{
  ssize_t retval;

  /* The d_ino and d_off fields in kernel_dirent and dirent must have
     the same sizes and alignments.  */
  if (sizeof (DIRENT_TYPE) == sizeof (struct dirent)
      && (sizeof (((struct kernel_dirent *) 0)->d_ino)
	  == sizeof (((struct dirent *) 0)->d_ino))
      && (sizeof (((struct kernel_dirent *) 0)->d_off)
	  == sizeof (((struct dirent *) 0)->d_off))
      && (offsetof (struct kernel_dirent, d_off)
	  == offsetof (struct dirent, d_off))
      && (offsetof (struct kernel_dirent, d_reclen)
	  == offsetof (struct dirent, d_reclen)))
    {
      retval = INLINE_SYSCALL (getdents, 3, fd, buf, nbytes);

      /* The kernel added the d_type value after the name.  Change
	 this now.  */
      if (retval != -1)
	{
	  union
	  {
	    struct kernel_dirent k;
	    struct dirent u;
	  } *kbuf = (void *) buf;

	  while ((char *) kbuf < buf + retval)
	    {
	      char d_type = *((char *) kbuf + kbuf->k.d_reclen - 1);
	      memmove (kbuf->u.d_name, kbuf->k.d_name,
		       strlen (kbuf->k.d_name) + 1);
	      kbuf->u.d_type = d_type;

	      kbuf = (void *) ((char *) kbuf + kbuf->k.d_reclen);
	    }
	}

      return retval;
    }

  off64_t last_offset = -1;

#ifdef __NR_getdents64
  {
    union
    {
      struct kernel_dirent64 k;
      DIRENT_TYPE u;
      char b[1];
    } *kbuf = (void *) buf, *outp, *inp;
    size_t kbytes = nbytes;
    if (offsetof (DIRENT_TYPE, d_name)
	< offsetof (struct kernel_dirent64, d_name)
	&& nbytes <= sizeof (DIRENT_TYPE))
      {
	kbytes = (nbytes + offsetof (struct kernel_dirent64, d_name)
		  - offsetof (DIRENT_TYPE, d_name));
	kbuf = __alloca(kbytes);
      }
    retval = INLINE_SYSCALL (getdents64, 3, fd, kbuf, kbytes);
    const size_t size_diff = (offsetof (struct kernel_dirent64, d_name)
			      - offsetof (DIRENT_TYPE, d_name));

    /* Return the error if encountered.  */
    if (retval == -1)
      return -1;

    /* If the structure returned by the kernel is identical to what we
       need, don't do any conversions.  */
    if (offsetof (DIRENT_TYPE, d_name)
	== offsetof (struct kernel_dirent64, d_name)
	&& sizeof (outp->u.d_ino) == sizeof (inp->k.d_ino)
	&& sizeof (outp->u.d_off) == sizeof (inp->k.d_off))
      return retval;

    /* These two pointers might alias the same memory buffer.
       Standard C requires that we always use the same type for them,
       so we must use the union type.  */
    inp = kbuf;
    outp = (void *) buf;

    while (&inp->b < &kbuf->b + retval)
      {
	const size_t alignment = __alignof__ (DIRENT_TYPE);
	/* Since inp->k.d_reclen is already aligned for the kernel
	   structure this may compute a value that is bigger
	   than necessary.  */
	size_t old_reclen = inp->k.d_reclen;
	size_t new_reclen = ((old_reclen - size_diff + alignment - 1)
			     & ~(alignment - 1));

	/* Copy the data out of the old structure into temporary space.
	   Then copy the name, which may overlap if BUF == KBUF.  */
	const uint64_t d_ino = inp->k.d_ino;
	const int64_t d_off = inp->k.d_off;
	const uint8_t d_type = inp->k.d_type;

	memmove (outp->u.d_name, inp->k.d_name,
		 old_reclen - offsetof (struct kernel_dirent64, d_name));

	/* Now we have copied the data from INP and access only OUTP.  */

	DIRENT_SET_DP_INO (&outp->u, d_ino);
	outp->u.d_off = d_off;
	if ((sizeof (outp->u.d_ino) != sizeof (inp->k.d_ino)
	     && outp->u.d_ino != d_ino)
	    || (sizeof (outp->u.d_off) != sizeof (inp->k.d_off)
		&& outp->u.d_off != d_off))
	  {
	    /* Overflow.  If there was at least one entry
	       before this one, return them without error,
	       otherwise signal overflow.  */
	    if (last_offset != -1)
	      {
		__lseek64 (fd, last_offset, SEEK_SET);
		return outp->b - buf;
	      }
	    __set_errno (EOVERFLOW);
	    return -1;
	  }

	last_offset = d_off;
	outp->u.d_reclen = new_reclen;
	outp->u.d_type = d_type;

	inp = (void *) inp + old_reclen;
	outp = (void *) outp + new_reclen;
      }

    return outp->b - buf;
  }
#endif
  {
    size_t red_nbytes;
    struct kernel_dirent *skdp, *kdp;
    const size_t size_diff = (offsetof (DIRENT_TYPE, d_name)
			      - offsetof (struct kernel_dirent, d_name));

    red_nbytes = MIN (nbytes
		      - ((nbytes / (offsetof (DIRENT_TYPE, d_name) + 14))
			 * size_diff),
		      nbytes - size_diff);

    skdp = kdp = __alloca (red_nbytes);

    retval = INLINE_SYSCALL (getdents, 3, fd, (char *) kdp, red_nbytes);

    if (retval == -1)
      return -1;

    DIRENT_TYPE *dp = (DIRENT_TYPE *) buf;
    while ((char *) kdp < (char *) skdp + retval)
      {
	const size_t alignment = __alignof__ (DIRENT_TYPE);
	/* Since kdp->d_reclen is already aligned for the kernel structure
	   this may compute a value that is bigger than necessary.  */
	size_t new_reclen = ((kdp->d_reclen + size_diff + alignment - 1)
			     & ~(alignment - 1));
	if ((char *) dp + new_reclen > buf + nbytes)
	  {
	    /* Our heuristic failed.  We read too many entries.  Reset
	       the stream.  */
	    assert (last_offset != -1);
	    __lseek64 (fd, last_offset, SEEK_SET);

	    if ((char *) dp == buf)
	      {
		/* The buffer the user passed in is too small to hold even
		   one entry.  */
		__set_errno (EINVAL);
		return -1;
	      }

	    break;
	  }

	last_offset = kdp->d_off;
	DIRENT_SET_DP_INO(dp, kdp->d_ino);
	dp->d_off = kdp->d_off;
	dp->d_reclen = new_reclen;
	dp->d_type = *((char *) kdp + kdp->d_reclen - 1);
	memcpy (dp->d_name, kdp->d_name,
		kdp->d_reclen - offsetof (struct kernel_dirent, d_name));

	dp = (DIRENT_TYPE *) ((char *) dp + new_reclen);
	kdp = (struct kernel_dirent *) (((char *) kdp) + kdp->d_reclen);
      }

    return (char *) dp - buf;
  }
}
Commit	Line	Data
f7a9f785	1	/* Copyright (C) 1993-2016 Free Software Foundation, Inc.
df4ef2ab UD	2	This file is part of the GNU C Library.
	3
	4	The GNU C Library is free software; you can redistribute it and/or
41bdb6e2 AJ	5	modify it under the terms of the GNU Lesser General Public
	6	License as published by the Free Software Foundation; either
	7	version 2.1 of the License, or (at your option) any later version.
df4ef2ab UD	8
	9	The GNU C Library is distributed in the hope that it will be useful,
	10	but WITHOUT ANY WARRANTY; without even the implied warranty of
	11	MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
41bdb6e2	12	Lesser General Public License for more details.
df4ef2ab	13
41bdb6e2	14	You should have received a copy of the GNU Lesser General Public
59ba27a6 PE	15	License along with the GNU C Library; if not, see
59ba27a6 PE	16	<http://www.gnu.org/licenses/>. */
df4ef2ab	17
8d57beea	18	#include <alloca.h>
4186c9f4	19	#include <assert.h>
0dee6738	20	#include <errno.h>
df4ef2ab UD	21	#include <dirent.h>
df4ef2ab UD	22	#include <stddef.h>
da0fdef0	23	#include <stdint.h>
df4ef2ab UD	24	#include <string.h>
df4ef2ab UD	25	#include <unistd.h>
8d57beea	26	#include <sys/param.h>
df4ef2ab UD	27	#include <sys/types.h>
df4ef2ab UD	28
0dee6738 UD	29	#include <sysdep.h>
	30	#include <sys/syscall.h>
	31
df4ef2ab UD	32	#include <linux/posix_types.h>
df4ef2ab UD	33
6ddd37a4	34	#include <kernel-features.h>
14860991	35
df4ef2ab UD	36	/* For Linux we need a special version of this file since the
	37	definition of `struct dirent' is not the same for the kernel and
	38	the libc. There is one additional field which might be introduced
	39	in the kernel structure in the future.
	40
0dee6738	41	Here is the kernel definition of `struct dirent' as of 2.1.20: */
df4ef2ab UD	42
	43	struct kernel_dirent
	44	{
	45	long int d_ino;
	46	__kernel_off_t d_off;
	47	unsigned short int d_reclen;
	48	char d_name[256];
	49	};
	50
14860991 UD	51	struct kernel_dirent64
14860991 UD	52	{
da0fdef0	53	uint64_t d_ino;
14860991 UD	54	int64_t d_off;
	55	unsigned short int d_reclen;
	56	unsigned char d_type;
	57	char d_name[256];
	58	};
	59
56ddf355 UD	60	#ifndef __GETDENTS
56ddf355 UD	61	# define __GETDENTS __getdents
14860991 UD	62	#endif
14860991 UD	63	#ifndef DIRENT_TYPE
56ddf355 UD	64	# define DIRENT_TYPE struct dirent
	65	#endif
	66	#ifndef DIRENT_SET_DP_INO
	67	# define DIRENT_SET_DP_INO(dp, value) (dp)->d_ino = (value)
9756dfe1	68	#endif
df4ef2ab UD	69
	70	/* The problem here is that we cannot simply read the next NBYTES
	71	bytes. We need to take the additional field into account. We use
831372e7	72	some heuristic. Assuming the directory contains names with 14
980e5832	73	characters on average we can compute an estimated number of entries
831372e7	74	which fit in the buffer. Taking this number allows us to specify a
980e5832 UD	75	reasonable number of bytes to read. If we should be wrong, we can
	76	reset the file descriptor. In practice the kernel is limiting the
	77	amount of data returned much more then the reduced buffer size. */
8d57beea	78	ssize_t
980e5832	79	internal_function
56ddf355	80	__GETDENTS (int fd, char *buf, size_t nbytes)
df4ef2ab	81	{
14860991	82	ssize_t retval;
980e5832	83
7cd195df L	84	/* The d_ino and d_off fields in kernel_dirent and dirent must have
	85	the same sizes and alignments. */
	86	if (sizeof (DIRENT_TYPE) == sizeof (struct dirent)
	87	&& (sizeof (((struct kernel_dirent *) 0)->d_ino)
	88	== sizeof (((struct dirent *) 0)->d_ino))
	89	&& (sizeof (((struct kernel_dirent *) 0)->d_off)
	90	== sizeof (((struct dirent *) 0)->d_off))
	91	&& (offsetof (struct kernel_dirent, d_off)
	92	== offsetof (struct dirent, d_off))
	93	&& (offsetof (struct kernel_dirent, d_reclen)
	94	== offsetof (struct dirent, d_reclen)))
afedc4af	95	{
a2da1673	96	retval = INLINE_SYSCALL (getdents, 3, fd, buf, nbytes);
afedc4af UD	97
	98	/* The kernel added the d_type value after the name. Change
	99	this now. */
	100	if (retval != -1)
	101	{
	102	union
	103	{
	104	struct kernel_dirent k;
	105	struct dirent u;
	106	} kbuf = (void ) buf;
	107
	108	while ((char *) kbuf < buf + retval)
	109	{
	110	char d_type = ((char ) kbuf + kbuf->k.d_reclen - 1);
	111	memmove (kbuf->u.d_name, kbuf->k.d_name,
	112	strlen (kbuf->k.d_name) + 1);
	113	kbuf->u.d_type = d_type;
	114
	115	kbuf = (void ) ((char ) kbuf + kbuf->k.d_reclen);
	116	}
	117	}
	118
	119	return retval;
	120	}
afedc4af UD	121
	122	off64_t last_offset = -1;
	123
14860991	124	#ifdef __NR_getdents64
37ad3473 JM	125	{
37ad3473 JM	126	union
df4ef2ab	127	{
37ad3473 JM	128	struct kernel_dirent64 k;
	129	DIRENT_TYPE u;
	130	char b[1];
	131	} kbuf = (void ) buf, outp, inp;
	132	size_t kbytes = nbytes;
	133	if (offsetof (DIRENT_TYPE, d_name)
	134	< offsetof (struct kernel_dirent64, d_name)
	135	&& nbytes <= sizeof (DIRENT_TYPE))
c213fe9c	136	{
37ad3473 JM	137	kbytes = (nbytes + offsetof (struct kernel_dirent64, d_name)
	138	- offsetof (DIRENT_TYPE, d_name));
	139	kbuf = __alloca(kbytes);
	140	}
	141	retval = INLINE_SYSCALL (getdents64, 3, fd, kbuf, kbytes);
	142	const size_t size_diff = (offsetof (struct kernel_dirent64, d_name)
	143	- offsetof (DIRENT_TYPE, d_name));
	144
	145	/* Return the error if encountered. */
	146	if (retval == -1)
	147	return -1;
14860991	148
37ad3473 JM	149	/* If the structure returned by the kernel is identical to what we
	150	need, don't do any conversions. */
	151	if (offsetof (DIRENT_TYPE, d_name)
	152	== offsetof (struct kernel_dirent64, d_name)
	153	&& sizeof (outp->u.d_ino) == sizeof (inp->k.d_ino)
	154	&& sizeof (outp->u.d_off) == sizeof (inp->k.d_off))
	155	return retval;
	156
	157	/* These two pointers might alias the same memory buffer.
	158	Standard C requires that we always use the same type for them,
	159	so we must use the union type. */
	160	inp = kbuf;
	161	outp = (void *) buf;
	162
	163	while (&inp->b < &kbuf->b + retval)
	164	{
	165	const size_t alignment = __alignof__ (DIRENT_TYPE);
	166	/* Since inp->k.d_reclen is already aligned for the kernel
	167	structure this may compute a value that is bigger
	168	than necessary. */
	169	size_t old_reclen = inp->k.d_reclen;
	170	size_t new_reclen = ((old_reclen - size_diff + alignment - 1)
	171	& ~(alignment - 1));
	172
	173	/* Copy the data out of the old structure into temporary space.
	174	Then copy the name, which may overlap if BUF == KBUF. */
	175	const uint64_t d_ino = inp->k.d_ino;
	176	const int64_t d_off = inp->k.d_off;
	177	const uint8_t d_type = inp->k.d_type;
	178
	179	memmove (outp->u.d_name, inp->k.d_name,
	180	old_reclen - offsetof (struct kernel_dirent64, d_name));
	181
	182	/* Now we have copied the data from INP and access only OUTP. */
	183
	184	DIRENT_SET_DP_INO (&outp->u, d_ino);
	185	outp->u.d_off = d_off;
	186	if ((sizeof (outp->u.d_ino) != sizeof (inp->k.d_ino)
	187	&& outp->u.d_ino != d_ino)
	188	\|\| (sizeof (outp->u.d_off) != sizeof (inp->k.d_off)
	189	&& outp->u.d_off != d_off))
	190	{
	191	/* Overflow. If there was at least one entry
	192	before this one, return them without error,
	193	otherwise signal overflow. */
	194	if (last_offset != -1)
	195	{
	196	__lseek64 (fd, last_offset, SEEK_SET);
	197	return outp->b - buf;
	198	}
	199	__set_errno (EOVERFLOW);
05ae4d6a	200	return -1;
37ad3473	201	}
05ae4d6a	202
37ad3473 JM	203	last_offset = d_off;
	204	outp->u.d_reclen = new_reclen;
	205	outp->u.d_type = d_type;
f11b9da6	206
37ad3473 JM	207	inp = (void *) inp + old_reclen;
	208	outp = (void *) outp + new_reclen;
	209	}
831372e7	210
37ad3473 JM	211	return outp->b - buf;
37ad3473 JM	212	}
14860991 UD	213	#endif
	214	{
	215	size_t red_nbytes;
	216	struct kernel_dirent skdp, kdp;
	217	const size_t size_diff = (offsetof (DIRENT_TYPE, d_name)
	218	- offsetof (struct kernel_dirent, d_name));
	219
	220	red_nbytes = MIN (nbytes
	221	- ((nbytes / (offsetof (DIRENT_TYPE, d_name) + 14))
	222	* size_diff),
	223	nbytes - size_diff);
	224
14860991 UD	225	skdp = kdp = __alloca (red_nbytes);
14860991 UD	226
a2da1673	227	retval = INLINE_SYSCALL (getdents, 3, fd, (char *) kdp, red_nbytes);
14860991 UD	228
	229	if (retval == -1)
	230	return -1;
	231
c213fe9c	232	DIRENT_TYPE dp = (DIRENT_TYPE ) buf;
14860991 UD	233	while ((char ) kdp < (char ) skdp + retval)
	234	{
	235	const size_t alignment = __alignof__ (DIRENT_TYPE);
	236	/* Since kdp->d_reclen is already aligned for the kernel structure
	237	this may compute a value that is bigger than necessary. */
	238	size_t new_reclen = ((kdp->d_reclen + size_diff + alignment - 1)
	239	& ~(alignment - 1));
	240	if ((char *) dp + new_reclen > buf + nbytes)
	241	{
	242	/* Our heuristic failed. We read too many entries. Reset
	243	the stream. */
	244	assert (last_offset != -1);
f9ef3e72	245	__lseek64 (fd, last_offset, SEEK_SET);
14860991 UD	246
	247	if ((char *) dp == buf)
	248	{
	249	/* The buffer the user passed in is too small to hold even
	250	one entry. */
	251	__set_errno (EINVAL);
	252	return -1;
	253	}
	254
	255	break;
	256	}
	257
	258	last_offset = kdp->d_off;
	259	DIRENT_SET_DP_INO(dp, kdp->d_ino);
	260	dp->d_off = kdp->d_off;
	261	dp->d_reclen = new_reclen;
1ac7a2c7	262	dp->d_type = ((char ) kdp + kdp->d_reclen - 1);
14860991 UD	263	memcpy (dp->d_name, kdp->d_name,
	264	kdp->d_reclen - offsetof (struct kernel_dirent, d_name));
	265
	266	dp = (DIRENT_TYPE ) ((char ) dp + new_reclen);
	267	kdp = (struct kernel_dirent ) (((char ) kdp) + kdp->d_reclen);
	268	}
df4ef2ab	269
c213fe9c RM	270	return (char *) dp - buf;
c213fe9c RM	271	}
df4ef2ab	272	}