ftello: Need adjust file offset before switching to write

Submitted by wangjianjian (C) on Dec. 18, 2019, 10:18 a.m.

Details

Message ID 07082d7e15024b06ba261fbb4e29eac5@huawei.com
State New
Series "ftello: Need adjust file offset before switching to write"
Headers show

Commit Message

wangjianjian (C) Dec. 18, 2019, 10:18 a.m.
Consider below code flow:

FILE *fp = fopen("foobar", "w+");
fputs("hello", fp);
rewind(fp);
fgetc(fp);
ungetc('x', fp);
fputs("world", fp);
off = ftello(fp);

In this code snippet, 'off' will be 10 at last which should be 5 and
file 'foobar's content will be 'helloworld' which is not correct.

The reason is that when do read, we will try our best to get more data
to fill user-provided and the default stream buffer so that the
underlying file offset may not be consistent with stream offset.
As an example above, 'fgetc' will read the whole 5 bytes from 'foobar'
even though we just would like to read one char.

And when we switch to write, we need to adjust the file offset by
decreasing the size of already buffered but not readed data.

Signed-off-by: Wang Jianjian <wangjianjian3@huawei.com>
---
 src/stdio/__towrite.c | 6 ++++++
 1 file changed, 6 insertions(+)


BR,
Wang Jianjian

Patch hide | download patch | download mbox

diff --git a/src/stdio/__towrite.c b/src/stdio/__towrite.c
index 4c9c66ae..dde3c967 100644
--- a/src/stdio/__towrite.c
+++ b/src/stdio/__towrite.c
@@ -7,6 +7,12 @@  int __towrite(FILE *f)
 		f->flags |= F_ERR;
 		return EOF;
 	}
+
+	/* Adjust offset for already buffered but not readed data */
+	if (f->rend - f->rpos > 0 &&
+	    f->seek(f, f->rpos - f->rend, SEEK_CUR) < 0)
+		return EOF;
+
 	/* Clear read buffer (easier than summoning nasal demons) */
 	f->rpos = f->rend = 0;


Comments

Rich Felker Dec. 18, 2019, 1:46 p.m.
On Wed, Dec 18, 2019 at 10:18:16AM +0000, wangjianjian (C) wrote:
> Consider below code flow:
> 
> FILE *fp = fopen("foobar", "w+");
> fputs("hello", fp);
> rewind(fp);
> fgetc(fp);
> ungetc('x', fp);
> fputs("world", fp);
  ^^^^^^^^^^^^^^^^^^^

This line produces UB. You can't perform a write after ungetc without
an intervening successful seek (or hitting EOF).

Rich
wangjianjian (C) Dec. 19, 2019, 2:47 a.m.
I don't think this has something to do with ungetc or UB.
If don't call ungetc and replace fputs with others, like fwrite or fputc, the result of ftell or final file content is not correct.

-----邮件原件-----
发件人: Rich Felker [mailto:dalias@aerifal.cx] 代表 Rich Felker
发送时间: 2019年12月18日 21:47
收件人: wangjianjian (C) <wangjianjian3@huawei.com>
抄送: musl@lists.openwall.com; Songyunlong (Euler) <yunlong.song@huawei.com>
主题: Re: [musl] [PATCH] ftello: Need adjust file offset before switching to write

On Wed, Dec 18, 2019 at 10:18:16AM +0000, wangjianjian (C) wrote:
> Consider below code flow:

> 

> FILE *fp = fopen("foobar", "w+");

> fputs("hello", fp);

> rewind(fp);

> fgetc(fp);

> ungetc('x', fp);

> fputs("world", fp);

  ^^^^^^^^^^^^^^^^^^^

This line produces UB. You can't perform a write after ungetc without an intervening successful seek (or hitting EOF).

Rich
Rich Felker Dec. 19, 2019, 3:03 a.m.
On Thu, Dec 19, 2019 at 02:47:46AM +0000, wangjianjian (C) wrote:
> I don't think this has something to do with ungetc or UB.
> If don't call ungetc and replace fputs with others, like fwrite or
> fputc, the result of ftell or final file content is not correct.

You cannot switch between reading and writing on a stream that's open
for both without a successful seek in between, unless the read ended
with hitting EOF. You're right that if you remove the ungetc, the UB
is still there because fgetc is still there. Removing both, or putting
a seek after them but before the fputs, will make things work as
expected.

For details see 7.21.5.3 The fopen function, ¶7:

    "When a file is opened with update mode ('+' as the second or
    third character in the above list of mode argument values), both
    input and output may be performed on the associated stream.
    However, output shall not be directly followed by input without an
    intervening call to the fflush function or to a file positioning
    function (fseek, fsetpos, or rewind), and input shall not be
    directly followed by output without an intervening call to a file
    positioning function, unless the input operation encounters end-
    of-file. Opening (or creating) a text file with update mode may
    instead open (or create) a binary stream in some implementations."

http://port70.net/~nsz/c/c11/n1570.html#7.21.5.3

Rich


 -----邮件原件-----
> 发件人: Rich Felker [mailto:dalias@aerifal.cx] 代表 Rich Felker
> 发送时间: 2019年12月18日 21:47
> 收件人: wangjianjian (C) <wangjianjian3@huawei.com>
> 抄送: musl@lists.openwall.com; Songyunlong (Euler) <yunlong.song@huawei.com>
> 主题: Re: [musl] [PATCH] ftello: Need adjust file offset before switching to write
> 
> On Wed, Dec 18, 2019 at 10:18:16AM +0000, wangjianjian (C) wrote:
> > Consider below code flow:
> > 
> > FILE *fp = fopen("foobar", "w+");
> > fputs("hello", fp);
> > rewind(fp);
> > fgetc(fp);
> > ungetc('x', fp);
> > fputs("world", fp);
>   ^^^^^^^^^^^^^^^^^^^
> 
> This line produces UB. You can't perform a write after ungetc without an intervening successful seek (or hitting EOF).
> 
> Rich
wangjianjian (C) Dec. 19, 2019, 3:19 a.m.
Gotcha. Thanks for detailed explanation.

-----邮件原件-----
发件人: Rich Felker [mailto:dalias@aerifal.cx] 代表 Rich Felker
发送时间: 2019年12月19日 11:03
收件人: wangjianjian (C) <wangjianjian3@huawei.com>
抄送: musl@lists.openwall.com; Songyunlong (Euler) <yunlong.song@huawei.com>
主题: Re: 答复: [musl] [PATCH] ftello: Need adjust file offset before switching to write

On Thu, Dec 19, 2019 at 02:47:46AM +0000, wangjianjian (C) wrote:
> I don't think this has something to do with ungetc or UB.

> If don't call ungetc and replace fputs with others, like fwrite or 

> fputc, the result of ftell or final file content is not correct.


You cannot switch between reading and writing on a stream that's open for both without a successful seek in between, unless the read ended with hitting EOF. You're right that if you remove the ungetc, the UB is still there because fgetc is still there. Removing both, or putting a seek after them but before the fputs, will make things work as expected.

For details see 7.21.5.3 The fopen function, ¶7:

    "When a file is opened with update mode ('+' as the second or
    third character in the above list of mode argument values), both
    input and output may be performed on the associated stream.
    However, output shall not be directly followed by input without an
    intervening call to the fflush function or to a file positioning
    function (fseek, fsetpos, or rewind), and input shall not be
    directly followed by output without an intervening call to a file
    positioning function, unless the input operation encounters end-
    of-file. Opening (or creating) a text file with update mode may
    instead open (or create) a binary stream in some implementations."

http://port70.net/~nsz/c/c11/n1570.html#7.21.5.3

Rich


 -----邮件原件-----
> 发件人: Rich Felker [mailto:dalias@aerifal.cx] 代表 Rich Felker

> 发送时间: 2019年12月18日 21:47

> 收件人: wangjianjian (C) <wangjianjian3@huawei.com>

> 抄送: musl@lists.openwall.com; Songyunlong (Euler) 

> <yunlong.song@huawei.com>

> 主题: Re: [musl] [PATCH] ftello: Need adjust file offset before 

> switching to write

> 

> On Wed, Dec 18, 2019 at 10:18:16AM +0000, wangjianjian (C) wrote:

> > Consider below code flow:

> > 

> > FILE *fp = fopen("foobar", "w+");

> > fputs("hello", fp);

> > rewind(fp);

> > fgetc(fp);

> > ungetc('x', fp);

> > fputs("world", fp);

>   ^^^^^^^^^^^^^^^^^^^

> 

> This line produces UB. You can't perform a write after ungetc without an intervening successful seek (or hitting EOF).

> 

> Rich