public inbox for libc-alpha@sourceware.org
 help / color / mirror / Atom feed
From: Adhemerval Zanella <adhemerval.zanella@linaro.org>
To: libc-alpha@sourceware.org, "Érico Nogueira" <ericonr@disroot.org>
Subject: Re: [PATCH 3/3] linux: implement ttyname as a wrapper around ttyname_r.
Date: Tue, 4 May 2021 13:37:27 -0300	[thread overview]
Message-ID: <ecdbd908-4784-0e40-0b11-ac917b52da9d@linaro.org> (raw)
In-Reply-To: <20210504015152.31064-3-ericonr@disroot.org>



On 03/05/2021 22:51, Érico Nogueira via Libc-alpha wrote:
> Big win in binary size and avoids duplicating the logic in multiple
> places.
> 
> On x86_64, dropped from 1883206 to 1881790, a 1416 byte decrease.
> 
> Also changed logic to track if ttyname_buf has been allocated by
> checking if it's NULL instead of tracking buflen as an additional
> variable.
> ---
> 
> I was going to simply change the function to use __fd_to_filename, like
> the 2/3 patch, but this felt like a better improvement. However, I
> haven't studied the implications deeply or verified if the ttyname()
> logic behaved noticeably (or subtly) different than what ttyname_r()
> did.
> 
>  sysdeps/unix/sysv/linux/ttyname.c | 161 ++----------------------------
>  1 file changed, 9 insertions(+), 152 deletions(-)
> 
> diff --git a/sysdeps/unix/sysv/linux/ttyname.c b/sysdeps/unix/sysv/linux/ttyname.c
> index d561ac12f2..a2bb81c78b 100644
> --- a/sysdeps/unix/sysv/linux/ttyname.c
> +++ b/sysdeps/unix/sysv/linux/ttyname.c
> @@ -17,17 +17,9 @@
>  
>  #include <errno.h>
>  #include <limits.h>
> -#include <stddef.h>
> -#include <dirent.h>
> -#include <sys/types.h>
> -#include <sys/stat.h>
>  #include <termios.h>
> -#include <unistd.h>
> -#include <string.h>
>  #include <stdlib.h>
>  
> -#include <_itoa.h>
> -
>  #include "ttyname.h"
>  
>  #if 0
> @@ -35,170 +27,35 @@
>  char *__ttyname;
>  #endif

Maybe remove this dead code as well.

>  
> -static char *getttyname (const char *dev, const struct stat64 *mytty,
> -			 int save, int *dostat);
> -
> -libc_freeres_ptr (static char *getttyname_name);
> -
> -static char *
> -attribute_compat_text_section
> -getttyname (const char *dev, const struct stat64 *mytty, int save, int *dostat)
> -{
> -  static size_t namelen;
> -  struct stat64 st;
> -  DIR *dirstream;
> -  struct dirent64 *d;
> -  size_t devlen = strlen (dev) + 1;
> -
> -  dirstream = __opendir (dev);
> -  if (dirstream == NULL)
> -    {
> -      *dostat = -1;
> -      return NULL;
> -    }
> -
> -  /* Prepare for the loop.  If we already have a buffer copy the directory
> -     name we look at into it.  */
> -  if (devlen < namelen)
> -    *((char *) __mempcpy (getttyname_name, dev, devlen - 1)) = '/';
> -
> -  while ((d = __readdir64 (dirstream)) != NULL)
> -    if ((d->d_fileno == mytty->st_ino || *dostat)
> -	&& strcmp (d->d_name, "stdin")
> -	&& strcmp (d->d_name, "stdout")
> -	&& strcmp (d->d_name, "stderr"))
> -      {
> -	size_t dlen = _D_ALLOC_NAMLEN (d);
> -	if (devlen + dlen > namelen)
> -	  {
> -	    free (getttyname_name);
> -	    namelen = 2 * (devlen + dlen); /* Big enough.  */
> -	    getttyname_name = malloc (namelen);
> -	    if (! getttyname_name)
> -	      {
> -		*dostat = -1;
> -		/* Perhaps it helps to free the directory stream buffer.  */
> -		(void) __closedir (dirstream);
> -		return NULL;
> -	      }
> -	    *((char *) __mempcpy (getttyname_name, dev, devlen - 1)) = '/';
> -	  }
> -	memcpy (&getttyname_name[devlen], d->d_name, dlen);
> -	if (__stat64 (getttyname_name, &st) == 0
> -	    && is_mytty (mytty, &st))
> -	  {
> -	    (void) __closedir (dirstream);
> -#if 0
> -	    __ttyname = getttyname_name;
> -#endif
> -	    __set_errno (save);
> -	    return getttyname_name;
> -	  }
> -      }
> -
> -  (void) __closedir (dirstream);
> -  __set_errno (save);
> -  return NULL;
> -}
> -
> -
> +#define TTYNAME_BUFLEN (PATH_MAX-1)
>  /* Static buffer in `ttyname'.  */
>  libc_freeres_ptr (static char *ttyname_buf);

Although not really related to this change, since you are touching it 
I think it should use a libc_freeres_fn instead:

| static char *ttyname_buf;
|
| libc_freeres_fn (free_mem)
| { 
|  free (ttyname_buf);
| }

>  
> -
>  /* Return the pathname of the terminal FD is open on, or NULL on errors.
>     The returned storage is good only until the next call to this function.  */
>  char *
>  ttyname (int fd)
>  {
> -  static size_t buflen;
> -  char procname[30];
> -  struct stat64 st, st1;
> -  int dostat = 0;
> -  int doispty = 0;
> -  char *name;
> -  int save = errno;
> -  struct termios term;
> -
>    /* isatty check, tcgetattr is used because it sets the correct
> -     errno (EBADF resp. ENOTTY) on error.  */
> +     errno (EBADF resp. ENOTTY) on error. Fast error path to avoid the allocation  */

Double space after period and line too long.

> +  struct termios term;
>    if (__glibc_unlikely (__tcgetattr (fd, &term) < 0))
>      return NULL;
>  
> -  if (__fstat64 (fd, &st) < 0)
> -    return NULL;
> -
> -  /* We try using the /proc filesystem.  */
> -  *_fitoa_word (fd, __stpcpy (procname, "/proc/self/fd/"), 10, 0) = '\0';
> -
> -  if (buflen == 0)
> +  if (ttyname_buf == NULL)
>      {
> -      buflen = 4095;
> -      ttyname_buf = (char *) malloc (buflen + 1);
> +      ttyname_buf = malloc (TTYNAME_BUFLEN + 1);
>        if (ttyname_buf == NULL)
>  	{
> -	  buflen = 0;
>  	  return NULL;
>  	}
>      }
>  

Ok.

> -  ssize_t len = __readlink (procname, ttyname_buf, buflen);
> -  if (__glibc_likely (len != -1))
> -    {
> -      if ((size_t) len >= buflen)
> -	return NULL;
> -
> -#define UNREACHABLE_LEN strlen ("(unreachable)")
> -      if (len > UNREACHABLE_LEN
> -	  && memcmp (ttyname_buf, "(unreachable)", UNREACHABLE_LEN) == 0)
> -	{
> -	  memmove (ttyname_buf, ttyname_buf + UNREACHABLE_LEN,
> -		   len - UNREACHABLE_LEN);
> -	  len -= UNREACHABLE_LEN;
> -	}
> -
> -      /* readlink need not terminate the string.  */
> -      ttyname_buf[len] = '\0';
> -
> -      /* Verify readlink result, fall back on iterating through devices.  */
> -      if (ttyname_buf[0] == '/'
> -	  && __stat64 (ttyname_buf, &st1) == 0
> -	  && is_mytty (&st, &st1))
> -	return ttyname_buf;
> -
> -      doispty = 1;
> -    }
> -
> -  if (__stat64 ("/dev/pts", &st1) == 0 && S_ISDIR (st1.st_mode))
> -    {
> -      name = getttyname ("/dev/pts", &st, save, &dostat);
> -    }
> -  else
> -    {
> -      __set_errno (save);
> -      name = NULL;
> -    }
> -
> -  if (!name && dostat != -1)
> +  int result = ttyname_r (fd, ttyname_buf, TTYNAME_BUFLEN);
> +  if (result != 0)

Why do you need to pass the allocated buffer size minus 1? The ttyname_r 
should handle it, it already pass buflen - 1 on readlink for instance.

>      {
> -      name = getttyname ("/dev", &st, save, &dostat);
> -    }
> -
> -  if (!name && dostat != -1)
> -    {
> -      dostat = 1;
> -      name = getttyname ("/dev", &st, save, &dostat);
> -    }
> -
> -  if (!name && doispty && is_pty (&st))
> -    {
> -      /* We failed to figure out the TTY's name, but we can at least
> -         signal that we did verify that it really is a PTY slave.
> -         This happens when we have inherited the file descriptor from
> -         a different mount namespace.  */
> -      __set_errno (ENODEV);
> +      __set_errno (result);
>        return NULL;
>      }
> -
> -  return name;
> +  return ttyname_buf;
>  }
> 

With this change ttyname now returns ERANGE instead of ENAMETOOLONG if
readlink returns a link larger than the specified buffer, but I think
it should be ok.

  reply	other threads:[~2021-05-04 16:37 UTC|newest]

Thread overview: 9+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-05-04  1:51 [PATCH 1/3] misc: use _fitoa_word to implement __fd_to_filename Érico Nogueira
2021-05-04  1:51 ` [PATCH 2/3] linux: use fd_to_filename instead of _fitoa_word in ttyname_r Érico Nogueira
2021-05-04 13:18   ` Adhemerval Zanella
2021-05-04  1:51 ` [PATCH 3/3] linux: implement ttyname as a wrapper around ttyname_r Érico Nogueira
2021-05-04 16:37   ` Adhemerval Zanella [this message]
2021-05-04 16:43     ` Adhemerval Zanella
2021-05-07 17:21       ` Adhemerval Zanella
2021-05-07 19:20         ` Érico Nogueira
2021-05-04 13:18 ` [PATCH 1/3] misc: use _fitoa_word to implement __fd_to_filename Adhemerval Zanella

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=ecdbd908-4784-0e40-0b11-ac917b52da9d@linaro.org \
    --to=adhemerval.zanella@linaro.org \
    --cc=ericonr@disroot.org \
    --cc=libc-alpha@sourceware.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).