Re: [PATCH v2] Fix offset caching for streams and use it for ftell (BZ #16680)

["Carlos O'Donell" <carlos at redhat dot com>]

On 03/11/2014 08:46 AM, Siddhesh Poyarekar wrote:
> Here's take 2:

Nobody has touched the core parts of this code in years, and I appreciate
that you have taken the initiative to do so. There is no gain though
without some pain, and we see it here as we improve the implementation.
Thanks for working through this.
 
> The ftell implementation was made conservative to ensure that
> incorrectly cached offsets never affect it.  However, this causes
> problems for append mode when a file stream is rewound.  Additionally,
> the 'clever' trick of using stat to get position for append mode files
> caused more problems than it solved and broke some old behavior.  I
> have described below the various problems that it caused and then
> finally the solution.
> 
> For a and a+ mode files, rewinding the stream should result in ftell
> returning 0 as the offset, but the stat() trick caused it to
> (incorrectly) always return the end of file.  Now I couldn't find
> anything in POSIX that specifies the stream position after rewind()
> for a file opened in 'a' mode, but for 'a+' mode it should be set to
> 0.  For 'a' mode too, it probably makes sense to keep it set to 0 in
> the interest of retaining old behavior.

Right, so I was thinking along the same lines. If you didn't switch
from write to read then there is no reason why the offset can't
be the end of the file after rewind. However, that subtle change
is a change in our implementation and I agree that it would be best
if we retained the old behaviour.

Does a `write; rewind; read' work now with the patch? Was the current
patch causing this to fail also or did the `read' start reading at 
offset 0?
 
> The initial file position for append mode files is implementation
> defined, so the implementation could either retain the current file
> position or move the position to the end of file.  The earlier ftell
> implementation would move the offset to end of file for append-only
> mode, but retain the old offset for a+ mode.  It would also cache the
> offset (this detail is important).  My patch broke this and would set
> the initial position to end of file for both append modes, thus
> breaking old behavior.  I was ignorant enough to write an incorrect
> test case for it too.
> 
> The Change:
> 
> I have now brought back the behavior of seeking to end of file for
> append-only streams, but with a slight difference.  I don't cache the
> offset though, since we would want ftell to query the current file
> position through lseek while the stream is not active.  Since the
> offset is moved to the end of file, we can rely on the file position
> reported by lseek and we don't need to resort to the stat() nonsense.

OK, but I wouldn't call it nonsense, you either stat() or lseek() and
both cost a syscall. Your patch inlines the lseek from get_file_offset
and removes the stat. That's fine.

Does this reinstitution of lseek cause any problems for the use of ftell
on inactive streams? For example is it really correct to have _IO_file_open
seek to the end of the fully flushed file in append mode? What about
other users of the fd that might expect the underlying offset to remain
the same?

> Finally, the cache is always reliable, except when there are unflished
> writes in an append mode stream (i.e. both a and a+).  In the latter
> case, it is safe to just do an lseek to SEEK_END.  The value can be
> safely cached too, since the file handle is already active at this
> point.  Incidentally, this is the only state change we affect in the
> file handle (apart from taking locks of course).

When you say unflushed writes, I assume those are writes in the
stream buffers that have not been passed to the kernel write syscall,
as opposed to writes that the kernel itself has not flushed to disk
from its own buffers.

I don't understand this part. The cached offset *could* be reliable
but you would have to take into account the unflushed writes, but we
don't so it isn't? Why else is it unreliable?
 
> I have also updated the test case to correct my impression of the
> initial file position for a+ streams to the initial behavior.  I have
> verified that this does not break any existing tests in the testsuite
> and also passes with the new tests.

Thanks.

> Siddhesh
> 
> 	[BZ #16680]
> 	* libio/fileops.c (_IO_file_open): Seek to end of file but
> 	don't cache the offset.
> 	(get_file_offset): Remove function.
> 	(do_ftell): Use cached offset when available.
> 	* libio/iofdopen.c (_IO_new_fdopen): Seek to end of file but
> 	don't cache the offset.
> 	* libio/tst-ftell-active-handler.c (do_rewind_test): New test
> 	case.
> 	(do_one_test): Call it.
> 	(do_ftell_test): Fix up expected old offset for a+ mode.
> 	* libio/wfileops.c (do_ftell_wide): Used cached offset when
> 	available.
>

OK to checkin after fixing some comment nits and answering my questions.

> ---
>  libio/fileops.c                  | 102 +++++++++++++----------------------
>  libio/iofdopen.c                 |   9 ++++
>  libio/tst-ftell-active-handler.c | 112 +++++++++++++++++++++++++++++++++++++--
>  libio/wfileops.c                 |  47 ++++++++--------
>  4 files changed, 178 insertions(+), 92 deletions(-)
> 
> diff --git a/libio/fileops.c b/libio/fileops.c
> index 2e7bc8d..cf68dbf 100644
> --- a/libio/fileops.c
> +++ b/libio/fileops.c
> @@ -232,13 +232,18 @@ _IO_file_open (fp, filename, posix_mode, prot, read_write, is32not64)
>      return NULL;
>    fp->_fileno = fdesc;
>    _IO_mask_flags (fp, read_write,_IO_NO_READS+_IO_NO_WRITES+_IO_IS_APPENDING);
> -  if ((read_write & _IO_IS_APPENDING) && (read_write & _IO_NO_READS))
> -    if (_IO_SEEKOFF (fp, (_IO_off64_t)0, _IO_seek_end, _IOS_INPUT|_IOS_OUTPUT)
> -	== _IO_pos_BAD && errno != ESPIPE)
> -      {
> -	close_not_cancel (fdesc);
> -	return NULL;
> -      }
> +  /* For append mode, send the file offset to the end of the file.  Don't
> +     update the offset cache though, since the file handle is not active.  */

OK, by using _IO_SYSSEEK directly you bypass any cache update.

> +  if ((read_write & (_IO_IS_APPENDING | _IO_NO_READS))
> +      == (_IO_IS_APPENDING | _IO_NO_READS))
> +    {
> +      _IO_off64_t new_pos = _IO_SYSSEEK (fp, 0, _IO_seek_end);
> +      if (new_pos == _IO_pos_BAD && errno != ESPIPE)
> +	{
> +	  close_not_cancel (fdesc);
> +	  return NULL;
> +	}
> +    }

OK, though this is not the minimal change here. You rewrote
the conditionals to be more compact, but AFAICT they didn't
change their meaning and no boolean coercion. So OK.

>    _IO_link_in ((struct _IO_FILE_plus *) fp);
>    return fp;
>  }
> @@ -929,43 +934,13 @@ _IO_file_sync_mmap (_IO_FILE *fp)
>    return 0;
>  }
>  
> -/* Get the current file offset using a system call.  This is the safest method
> -   to get the current file offset, since we are sure that we get the current
> -   state of the file.  Before the stream handle is activated (by using fread,
> -   fwrite, etc.), an application may alter the state of the file descriptor
> -   underlying it by calling read/write/lseek on it.  Using a cached offset at
> -   this point will result in returning the incorrect value.  Same is the case
> -   when one switches from reading in a+ mode to writing, where the buffer has
> -   not been flushed - the cached offset would reflect the reading position
> -   while the actual write position would be at the end of the file.
> -
> -   do_ftell and do_ftell_wide may resort to using the cached offset in some
> -   special cases instead of calling get_file_offset, but those cases should be
> -   thoroughly described.  */
> -_IO_off64_t
> -get_file_offset (_IO_FILE *fp)
> -{
> -  if ((fp->_flags & _IO_IS_APPENDING) == _IO_IS_APPENDING)
> -    {
> -      struct stat64 st;
> -      bool ret = (_IO_SYSSTAT (fp, &st) == 0 && S_ISREG (st.st_mode));
> -      if (ret)
> -	return st.st_size;
> -      else
> -	return EOF;
> -    }
> -  else
> -    return _IO_SYSSEEK (fp, 0, _IO_seek_cur);
> -}

OK, removing get_file_offset because we've just inlined the
_IO_SYSSEEK directly into the callers but only when we know
no writing has happened.

> -
> -
> -/* ftell{,o} implementation.  Don't modify any state of the file pointer while
> -   we try to get the current state of the stream.  */
> +/* ftell{,o} implementation.  The only time we modify the state of the stream
> +   is when we have unflushed writes.  In that case we seek to the end and
> +   record that offset in the stream object.  */
>  static _IO_off64_t
>  do_ftell (_IO_FILE *fp)
>  {
> -  _IO_off64_t result = 0;
> -  bool use_cached_offset = false;
> +  _IO_off64_t result, offset = 0;

OK.

>  
>    /* No point looking at unflushed data if we haven't allocated buffers
>       yet.  */
> @@ -974,39 +949,37 @@ do_ftell (_IO_FILE *fp)
>        bool was_writing = (fp->_IO_write_ptr > fp->_IO_write_base
>  			  || _IO_in_put_mode (fp));
>  
> +      bool append_mode = (fp->_flags & _IO_IS_APPENDING) == _IO_IS_APPENDING;
> +
> +      /* When we have unflushed writes in append mode, seek to the end of the
> +	 file and record that offset.  This is the only time we change the file
> +	 stream state and it is safe since the file handle is active.  */
> +      if (was_writing && append_mode)
> +	{
> +	  result = _IO_SYSSEEK (fp, 0, _IO_seek_end);
> +	  if (result == _IO_pos_BAD)
> +	    return EOF;
> +	  else
> +	    fp->_offset = result;
> +	}

OK.

> +
>        /* Adjust for unflushed data.  */
>        if (!was_writing)
> -	result -= fp->_IO_read_end - fp->_IO_read_ptr;
> +	offset -= fp->_IO_read_end - fp->_IO_read_ptr;

OK.

>        else
> -	result += fp->_IO_write_ptr - fp->_IO_read_end;
> -
> -      /* It is safe to use the cached offset when available if there is
> -	 unbuffered data (indicating that the file handle is active) and the
> -	 handle is not for a file open in a+ mode.  The latter condition is
> -	 because there could be a scenario where there is a switch from read
> -	 mode to write mode using an fseek to an arbitrary position.  In this
> -	 case, there would be unbuffered data due to be appended to the end of
> -	 the file, but the offset may not necessarily be the end of the
> -	 file.  It is fine to use the cached offset when the a+ stream is in
> -	 read mode though, since the offset is maintained correctly in that
> -	 case.  Note that this is not a comprehensive set of cases when the
> -	 offset is reliable.  The offset may be reliable even in some cases
> -	 where there is no unflushed input and the handle is active, but it's
> -	 just that we don't have a way to identify that condition reliably.  */
> -      use_cached_offset = (result != 0 && fp->_offset != _IO_pos_BAD
> -			   && ((fp->_flags & (_IO_IS_APPENDING | _IO_NO_READS))
> -			       == (_IO_IS_APPENDING | _IO_NO_READS)
> -			       && was_writing));
> +	offset += fp->_IO_write_ptr - fp->_IO_read_end;

OK. It always seems kind of odd that simulating the buffering effects
using the various pointers and their outcome on the offset is repeated
among all the functions. What would be nicer, and I'm just talking out loud
is a "account for buffering" function that did all of this simulation itself.

>      }
>  
> -  if (use_cached_offset)
> -    result += fp->_offset;
> +  if (fp->_offset != _IO_pos_BAD)
> +    result = fp->_offset;
>    else
> -    result += get_file_offset (fp);
> +    result = _IO_SYSSEEK (fp, 0, _IO_seek_cur);

OK.

>  
>    if (result == EOF)
>      return result;
>  
> +  result += offset;
> +

OK.

>    if (result < 0)
>      {
>        __set_errno (EINVAL);
> @@ -1016,7 +989,6 @@ do_ftell (_IO_FILE *fp)
>    return result;
>  }
>  
> -
>  _IO_off64_t
>  _IO_new_file_seekoff (fp, offset, dir, mode)
>       _IO_FILE *fp;
> diff --git a/libio/iofdopen.c b/libio/iofdopen.c
> index 3f266f7..364e175 100644
> --- a/libio/iofdopen.c
> +++ b/libio/iofdopen.c
> @@ -167,6 +167,15 @@ _IO_new_fdopen (fd, mode)
>    _IO_mask_flags (&new_f->fp.file, read_write,
>  		  _IO_NO_READS+_IO_NO_WRITES+_IO_IS_APPENDING);
>  
> +  /* For append mode, send the file offset to the end of the file.  Don't

s/send/set/g

> +     update the offset cache though, since the file handle is not active.  */
> +  if ((read_write & (_IO_IS_APPENDING | _IO_NO_READS))
> +      == (_IO_IS_APPENDING | _IO_NO_READS))
> +    {
> +      _IO_off64_t new_pos = _IO_SYSSEEK (&new_f->fp.file, 0, _IO_seek_end);
> +      if (new_pos == _IO_pos_BAD && errno != ESPIPE)
> +	return NULL;
> +    }

OK.

>    return &new_f->fp.file;
>  }
>  libc_hidden_ver (_IO_new_fdopen, _IO_fdopen)
> diff --git a/libio/tst-ftell-active-handler.c b/libio/tst-ftell-active-handler.c
> index 5d5fc26..aef9cc0 100644
> --- a/libio/tst-ftell-active-handler.c
> +++ b/libio/tst-ftell-active-handler.c
> @@ -88,6 +88,107 @@ static size_t file_len;
>  typedef int (*fputs_func_t) (const void *data, FILE *fp);
>  fputs_func_t fputs_func;
>  
> +/* Test that ftell output after a rewind is correct.  */
> +static int
> +do_rewind_test (const char *filename)
> +{
> +  int ret = 0;
> +  struct test
> +    {
> +      const char *mode;
> +      int fd_mode;
> +      size_t old_off;
> +      size_t new_off;
> +    } test_modes[] = {
> +	  {"w", O_WRONLY, 0, data_len},
> +	  {"w+", O_RDWR, 0, data_len},
> +	  {"r+", O_RDWR, 0, data_len},
> +	  /* The new offsets for 'a' and 'a+' modes have to factor in the
> +	     previous writes since they compulsorily append to the end of the

s/compulsorily/always/g

> +	     file.  */
> +	  {"a", O_WRONLY, 0, 3 * data_len},
> +	  {"a+", O_RDWR, 0, 4 * data_len},
> +    };
> +
> +  /* Empty the file before the test so that our offsets are simple to
> +     calculate.  */
> +  FILE *fp = fopen (filename, "w");
> +  if (fp == NULL)
> +    {
> +      printf ("Failed to open file for emptying\n");
> +      return 1;
> +    }
> +  fclose (fp);
> +
> +  for (int j = 0; j < 2; j++)
> +    {
> +      for (int i = 0; i < sizeof (test_modes) / sizeof (struct test); i++)
> +	{
> +	  FILE *fp;
> +	  int fd;
> +	  int fileret;
> +
> +	  printf ("\trewind: %s (file, \"%s\"): ", j == 0 ? "fdopen" : "fopen",
> +		  test_modes[i].mode);
> +
> +	  if (j == 0)
> +	    fileret = get_handles_fdopen (filename, fd, fp,
> +					  test_modes[i].fd_mode,
> +					  test_modes[i].mode);
> +	  else
> +	    fileret = get_handles_fopen (filename, fd, fp, test_modes[i].mode);
> +
> +	  if (fileret != 0)
> +	    return fileret;
> +
> +	  /* Write some content to the file, rewind and ensure that the ftell
> +	     output after the rewind is 0.  POSIX does not specify what the
> +	     behavior is when a file is rewound in 'a' mode, so we retain
> +	     current behavior, which is to keep the 0 offset.  */
> +	  size_t written = fputs_func (data, fp);
> +
> +	  if (written == EOF)
> +	    {
> +	      printf ("fputs[1] failed to write data\n");
> +	      ret |= 1;
> +	    }
> +
> +	  rewind (fp);
> +	  long offset = ftell (fp);
> +
> +	  if (offset != test_modes[i].old_off)
> +	    {
> +	      printf ("Incorrect old offset.  Expected %zu, but got %ld, ",
> +		      test_modes[i].old_off, offset);
> +	      ret |= 1;
> +	    }
> +	  else
> +	    printf ("old offset = %ld, ", offset);
> +
> +	  written = fputs_func (data, fp);
> +
> +	  if (written == EOF)
> +	    {
> +	      printf ("fputs[1] failed to write data\n");
> +	      ret |= 1;
> +	    }
> +
> +	  /* After this write, the offset in append modes should factor in the
> +	     implicit lseek to the end of file.  */
> +	  offset = ftell (fp);
> +	  if (offset != test_modes[i].new_off)
> +	    {
> +	      printf ("Incorrect new offset.  Expected %zu, but got %ld\n",
> +		      test_modes[i].new_off, offset);
> +	      ret |= 1;
> +	    }
> +	  else
> +	    printf ("new offset = %ld\n", offset);
> +	}
> +    }
> +  return ret;
> +}
> +
>  /* Test that the value of ftell is not cached when the stream handle is not
>     active.  */
>  static int
> @@ -107,11 +208,13 @@ do_ftell_test (const char *filename)
>  	  {"w", O_WRONLY, 0, data_len},
>  	  {"w+", O_RDWR, 0, data_len},
>  	  {"r+", O_RDWR, 0, data_len},
> -	  /* For 'a' and 'a+' modes, the initial file position should be the
> +	  /* For the 'a' mode, the initial file position should be the
>  	     current end of file. After the write, the offset has data_len
> -	     added to the old value.  */
> +	     added to the old value.  For a+ mode however, the initial file
> +	     position is the file position of the underlying file descriptor,
> +	     since it is initially assumed to be in read mode.  */
>  	  {"a", O_WRONLY, data_len, 2 * data_len},
> -	  {"a+", O_RDWR, 2 * data_len, 3 * data_len},
> +	  {"a+", O_RDWR, 0, 3 * data_len},

OK.

>      };
>    for (int j = 0; j < 2; j++)
>      {
> @@ -157,7 +260,7 @@ do_ftell_test (const char *filename)
>  	  if (off != test_modes[i].new_off)
>  	    {
>  	      printf ("Incorrect new offset.  Expected %zu but got %ld\n",
> -		      test_modes[i].old_off, off);
> +		      test_modes[i].new_off, off);
>  	      ret |= 1;
>  	    }
>  	  else
> @@ -322,6 +425,7 @@ do_one_test (const char *filename)
>    ret |= do_ftell_test (filename);
>    ret |= do_write_test (filename);
>    ret |= do_append_test (filename);
> +  ret |= do_rewind_test (filename);
>  
>    return ret;
>  }
> diff --git a/libio/wfileops.c b/libio/wfileops.c
> index 8b2e108..3199861 100644
> --- a/libio/wfileops.c
> +++ b/libio/wfileops.c
> @@ -597,12 +597,12 @@ done:
>  }
>  
>  /* ftell{,o} implementation for wide mode.  Don't modify any state of the file
> -   pointer while we try to get the current state of the stream.  */
> +   pointer while we try to get the current state of the stream except in one
> +   case, which is when we have unflushed writes in append mode.  */
>  static _IO_off64_t
>  do_ftell_wide (_IO_FILE *fp)
>  {
>    _IO_off64_t result, offset = 0;
> -  bool use_cached_offset = false;
>  
>    /* No point looking for offsets in the buffer if it hasn't even been
>       allocated.  */
> @@ -615,6 +615,20 @@ do_ftell_wide (_IO_FILE *fp)
>  			   > fp->_wide_data->_IO_write_base)
>  			  || _IO_in_put_mode (fp));
>  
> +      bool append_mode = (fp->_flags & _IO_IS_APPENDING) == _IO_IS_APPENDING;
> +
> +      /* When we have unflushed writes in append mode, seek to the end of the
> +	 file and record that offset.  This is the only time we change the file
> +	 stream state and it is safe since the file handle is active.  */
> +      if (was_writing && append_mode)
> +	{
> +	  result = _IO_SYSSEEK (fp, 0, _IO_seek_end);
> +	  if (result == _IO_pos_BAD)
> +	    return EOF;
> +	  else
> +	    fp->_offset = result;
> +	}
> +

OK.

>        /* XXX For wide stream with backup store it is not very
>  	 reasonable to determine the offset.  The pushed-back
>  	 character might require a state change and we need not be
> @@ -703,37 +717,24 @@ do_ftell_wide (_IO_FILE *fp)
>  	     position is fp._offset - (_IO_read_end - new_write_ptr).  */
>  	  offset -= fp->_IO_read_end - fp->_IO_write_ptr;
>  	}
> -
> -      /* It is safe to use the cached offset when available if there is
> -	 unbuffered data (indicating that the file handle is active) and
> -	 the handle is not for a file open in a+ mode.  The latter
> -	 condition is because there could be a scenario where there is a
> -	 switch from read mode to write mode using an fseek to an arbitrary
> -	 position.  In this case, there would be unbuffered data due to be
> -	 appended to the end of the file, but the offset may not
> -	 necessarily be the end of the file.  It is fine to use the cached
> -	 offset when the a+ stream is in read mode though, since the offset
> -	 is maintained correctly in that case.  Note that this is not a
> -	 comprehensive set of cases when the offset is reliable.  The
> -	 offset may be reliable even in some cases where there is no
> -	 unflushed input and the handle is active, but it's just that we
> -	 don't have a way to identify that condition reliably.  */
> -      use_cached_offset = (offset != 0 && fp->_offset != _IO_pos_BAD
> -			   && ((fp->_flags & (_IO_IS_APPENDING | _IO_NO_READS))
> -			       == (_IO_IS_APPENDING | _IO_NO_READS)
> -			       && was_writing));
>      }
>  
> -  if (use_cached_offset)
> +  if (fp->_offset != _IO_pos_BAD)
>      result = fp->_offset;
>    else
> -    result = get_file_offset (fp);
> +    result = _IO_SYSSEEK (fp, 0, _IO_seek_cur);

OK.

>  
>    if (result == EOF)
>      return result;
>  
>    result += offset;
>  
> +  if (result < 0)
> +    {
> +      __set_errno (EINVAL);
> +      return EOF;
> +    }

OK.

> +
>    return result;
>  }
>  
> 

Cheers,
Carlos.