| Message ID | 20220222205550.34537-1-jobol@nonadev.net |
|---|---|
| State | New |
| Headers | show |
| Series | [v3] libio: Ensure output buffer for wchars (bug 28828) | expand |
On 22/02/2022 17:55, jobol@nonadev.net wrote: > From: "jobol@nonadev.net" <jobol@nonadev.net> > > When fileops.c checks for nullity of the write pointer, > in order to ensure its allocation, before that patch, > wfileops didn't. This was leading to crashes on some cases, > as described by bug 28828. > > The minimal sequence to produce the crash was: > > #include <stdio.h> > #include <wchar.h> > int main(int ac, char **av) > { > setvbuf(stdout, NULL, _IOLBF, 0); > fgetwc(stdin); > fputwc(10, stdout); /*CRASH HERE!*/ > return 0; > } > > The line "fgetwc(stdin);" is necessary. It introduces the > bug by setting the flag _IO_CURRENTLY_PUTTING of stdout > indirectly (file wfileops.c, function _IO_wfile_underflow, line 213). > > Signed-off-by: Jose Bollo <jobol@nonadev.net> > --- > libio/Makefile | 2 +- > libio/tst-bz28828.c | 10 ++++++++++ > libio/tst-bz28828.input | 1 + > libio/wfileops.c | 3 ++- > 4 files changed, 14 insertions(+), 2 deletions(-) > create mode 100644 libio/tst-bz28828.c > create mode 100644 libio/tst-bz28828.input > > diff --git a/libio/Makefile b/libio/Makefile > index 0e5f348bea..e97387743f 100644 > --- a/libio/Makefile > +++ b/libio/Makefile > @@ -66,7 +66,7 @@ tests = tst_swprintf tst_wprintf tst_swscanf tst_wscanf tst_getwc tst_putwc \ > tst-fwrite-error tst-ftell-partial-wide tst-ftell-active-handler \ > tst-ftell-append tst-fputws tst-bz22415 tst-fgetc-after-eof \ > tst-sprintf-ub tst-sprintf-chk-ub tst-bz24051 tst-bz24153 \ > - tst-wfile-sync > + tst-wfile-sync tst-bz28828 > > tests-internal = tst-vtables tst-vtables-interposed > > diff --git a/libio/tst-bz28828.c b/libio/tst-bz28828.c > new file mode 100644 > index 0000000000..f5849d2ca6 > --- /dev/null > +++ b/libio/tst-bz28828.c > @@ -0,0 +1,10 @@ > +#include <stdio.h> > +#include <wchar.h> > +int main(int ac, char **av) > +{ > + setvbuf(stdout, NULL, _IOLBF, 0); > + fgetwc(stdin); > + fputwc(10, stdout); /*SHOUDN'T CRASH HERE!*/ > + return 0; > +} > + The tests requires a proper Copyright header and to use libsupport (so it handles test timeout, segfault, or any other error in test itself). Also, use the proper indentation. A skeleton will be something like below: --- /* Unit test for BZ#28828. Copyright (C) 2022 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 <https://www.gnu.org/licenses/>. */ #include <support/xstdio.h> #include <support/check.h> #include <wchar.h> static int do_test (void) { /* Add the test here. */ return 0; } #include <support/test-driver.c> --- Keep in mind that using fgetwc without a method to actually input some data will make the test timeout and fail. > diff --git a/libio/tst-bz28828.input b/libio/tst-bz28828.input > new file mode 100644 > index 0000000000..ce01362503 > --- /dev/null > +++ b/libio/tst-bz28828.input > @@ -0,0 +1 @@ > +hello > diff --git a/libio/wfileops.c b/libio/wfileops.c > index fb9d45b677..b59a98881f 100644 > --- a/libio/wfileops.c > +++ b/libio/wfileops.c > @@ -412,7 +412,8 @@ _IO_wfile_overflow (FILE *f, wint_t wch) > return WEOF; > } > /* If currently reading or no buffer allocated. */ > - if ((f->_flags & _IO_CURRENTLY_PUTTING) == 0) > + if ((f->_flags & _IO_CURRENTLY_PUTTING) == 0 > + || f->_wide_data->_IO_write_base == NULL) > { > /* Allocate a buffer if needed. */ > if (f->_wide_data->_IO_write_base == 0) Based on your test I was intrigued on why you need a fgetwc(stdio) to actually trigger the issue. The fgetwc(stdio) in fact changed the stdout internal flags somewhat, which then triggered the issue. Digging a little bit deeper it seems this snippet is triggering it: libio/wfileops.c 201 /* FIXME This can/should be moved to genops ?? */ 202 if (fp->_flags & (_IO_LINE_BUF | _IO_UNBUFFERED)) 203 { 204 /* We used to flush all line-buffered stream. This really isn't 205 required by any standard. My recollection is that 206 traditional Unix systems did this for stdout. stderr better 207 not be line buffered. So we do just that here 208 explicitly. --drepper */ 209 _IO_acquire_lock (stdout); 210 211 if ((stdout->_flags & (_IO_LINKED | _IO_NO_WRITES | _IO_LINE_BUF)) 212 == (_IO_LINKED | _IO_LINE_BUF)) 213 _IO_OVERFLOW (stdout, EOF); 214 215 _IO_release_lock (stdout); 216 } And the FIXME really makes me wonder if this is really ok, the libio/fileops.c does not contain such snippet. Removing this code does not trigger any regression and also fixes the issue.
On Mär 04 2022, Adhemerval Zanella via Libc-alpha wrote:
> the libio/fileops.c does not contain such snippet.
See line 488.
On 04/03/2022 16:53, Andreas Schwab wrote: > On Mär 04 2022, Adhemerval Zanella via Libc-alpha wrote: > >> the libio/fileops.c does not contain such snippet. > > See line 488. > Should we evaluate to remove it as well then?
On Mär 07 2022, Adhemerval Zanella wrote: > On 04/03/2022 16:53, Andreas Schwab wrote: >> On Mär 04 2022, Adhemerval Zanella via Libc-alpha wrote: >> >>> the libio/fileops.c does not contain such snippet. >> >> See line 488. >> > > Should we evaluate to remove it as well then? Why would we? It's specified by POSIX.
On 07/03/2022 08:29, Andreas Schwab wrote: > On Mär 07 2022, Adhemerval Zanella wrote: > >> On 04/03/2022 16:53, Andreas Schwab wrote: >>> On Mär 04 2022, Adhemerval Zanella via Libc-alpha wrote: >>> >>>> the libio/fileops.c does not contain such snippet. >>> >>> See line 488. >>> >> >> Should we evaluate to remove it as well then? > > Why would we? It's specified by POSIX. > That's no what comment specifies ("This really isn't required by any standard"). Do you have any link that describes this behavior?
On Mär 07 2022, Adhemerval Zanella wrote: > That's no what comment specifies ("This really isn't required by any > standard"). This refers to the preceding sentence. > Do you have any link that describes this behavior? https://pubs.opengroup.org/onlinepubs/9699919799/functions/V2_chap02.html#tag_15_05
On Fri, 4 Mar 2022 16:39:36 -0300 Adhemerval Zanella <adhemerval.zanella@linaro.org> wrote: Hi Adhemerval, Hi Andreas, Hi all, I'm going to submit a new version taking your skeleton as a basis. I added the file named libio/tst-bz28828.input with the understanding that given its name, the test framework will automatically use it as input. Can you tell me if it is correct or if I have to add something some where. My though about the interaction between stdin and stdout is that it is sometime needed to get an input to flush the output. Best regards José > On 22/02/2022 17:55, jobol@nonadev.net wrote: > > From: "jobol@nonadev.net" <jobol@nonadev.net> > > > > When fileops.c checks for nullity of the write pointer, > > in order to ensure its allocation, before that patch, > > wfileops didn't. This was leading to crashes on some cases, > > as described by bug 28828. > > > > The minimal sequence to produce the crash was: > > > > #include <stdio.h> > > #include <wchar.h> > > int main(int ac, char **av) > > { > > setvbuf(stdout, NULL, _IOLBF, 0); > > fgetwc(stdin); > > fputwc(10, stdout); /*CRASH HERE!*/ > > return 0; > > } > > > > The line "fgetwc(stdin);" is necessary. It introduces the > > bug by setting the flag _IO_CURRENTLY_PUTTING of stdout > > indirectly (file wfileops.c, function _IO_wfile_underflow, line > > 213). > > > > Signed-off-by: Jose Bollo <jobol@nonadev.net> > > --- > > libio/Makefile | 2 +- > > libio/tst-bz28828.c | 10 ++++++++++ > > libio/tst-bz28828.input | 1 + > > libio/wfileops.c | 3 ++- > > 4 files changed, 14 insertions(+), 2 deletions(-) > > create mode 100644 libio/tst-bz28828.c > > create mode 100644 libio/tst-bz28828.input > > > > diff --git a/libio/Makefile b/libio/Makefile > > index 0e5f348bea..e97387743f 100644 > > --- a/libio/Makefile > > +++ b/libio/Makefile > > @@ -66,7 +66,7 @@ tests = tst_swprintf tst_wprintf tst_swscanf > > tst_wscanf tst_getwc tst_putwc \ tst-fwrite-error > > tst-ftell-partial-wide tst-ftell-active-handler \ tst-ftell-append > > tst-fputws tst-bz22415 tst-fgetc-after-eof \ tst-sprintf-ub > > tst-sprintf-chk-ub tst-bz24051 tst-bz24153 \ > > - tst-wfile-sync > > + tst-wfile-sync tst-bz28828 > > > > tests-internal = tst-vtables tst-vtables-interposed > > > > diff --git a/libio/tst-bz28828.c b/libio/tst-bz28828.c > > new file mode 100644 > > index 0000000000..f5849d2ca6 > > --- /dev/null > > +++ b/libio/tst-bz28828.c > > @@ -0,0 +1,10 @@ > > +#include <stdio.h> > > +#include <wchar.h> > > +int main(int ac, char **av) > > +{ > > + setvbuf(stdout, NULL, _IOLBF, 0); > > + fgetwc(stdin); > > + fputwc(10, stdout); /*SHOUDN'T CRASH HERE!*/ > > + return 0; > > +} > > + > > The tests requires a proper Copyright header and to use libsupport > (so it handles test timeout, segfault, or any other error in test > itself). Also, use the proper indentation. A skeleton will be > something like below: > > --- > /* Unit test for BZ#28828. > Copyright (C) 2022 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 > <https://www.gnu.org/licenses/>. */ > > #include <support/xstdio.h> > #include <support/check.h> > #include <wchar.h> > > static int > do_test (void) > { > /* Add the test here. */ > > return 0; > } > > #include <support/test-driver.c> > --- > > Keep in mind that using fgetwc without a method to actually input > some data will make the test timeout and fail. > > > > diff --git a/libio/tst-bz28828.input b/libio/tst-bz28828.input > > new file mode 100644 > > index 0000000000..ce01362503 > > --- /dev/null > > +++ b/libio/tst-bz28828.input > > @@ -0,0 +1 @@ > > +hello > > diff --git a/libio/wfileops.c b/libio/wfileops.c > > index fb9d45b677..b59a98881f 100644 > > --- a/libio/wfileops.c > > +++ b/libio/wfileops.c > > @@ -412,7 +412,8 @@ _IO_wfile_overflow (FILE *f, wint_t wch) > > return WEOF; > > } > > /* If currently reading or no buffer allocated. */ > > - if ((f->_flags & _IO_CURRENTLY_PUTTING) == 0) > > + if ((f->_flags & _IO_CURRENTLY_PUTTING) == 0 > > + || f->_wide_data->_IO_write_base == NULL) > > { > > /* Allocate a buffer if needed. */ > > if (f->_wide_data->_IO_write_base == 0) > > Based on your test I was intrigued on why you need a fgetwc(stdio) to > actually trigger the issue. The fgetwc(stdio) in fact changed the > stdout internal flags somewhat, which then triggered the issue. > > Digging a little bit deeper it seems this snippet is triggering it: > > libio/wfileops.c > > 201 /* FIXME This can/should be moved to genops ?? */ > 202 if (fp->_flags & (_IO_LINE_BUF | _IO_UNBUFFERED)) > 203 { > 204 /* We used to flush all line-buffered stream. This really > isn't 205 required by any standard. My recollection is that > 206 traditional Unix systems did this for stdout. stderr > better 207 not be line buffered. So we do just that here > 208 explicitly. --drepper */ > 209 _IO_acquire_lock (stdout); > 210 > 211 if ((stdout->_flags & (_IO_LINKED | _IO_NO_WRITES | > _IO_LINE_BUF)) 212 == (_IO_LINKED | _IO_LINE_BUF)) > 213 _IO_OVERFLOW (stdout, EOF); > 214 > 215 _IO_release_lock (stdout); > 216 } > > And the FIXME really makes me wonder if this is really ok, the > libio/fileops.c does not contain such snippet. Removing this code > does not trigger any regression and also fixes the issue.
diff --git a/libio/Makefile b/libio/Makefile index 0e5f348bea..e97387743f 100644 --- a/libio/Makefile +++ b/libio/Makefile @@ -66,7 +66,7 @@ tests = tst_swprintf tst_wprintf tst_swscanf tst_wscanf tst_getwc tst_putwc \ tst-fwrite-error tst-ftell-partial-wide tst-ftell-active-handler \ tst-ftell-append tst-fputws tst-bz22415 tst-fgetc-after-eof \ tst-sprintf-ub tst-sprintf-chk-ub tst-bz24051 tst-bz24153 \ - tst-wfile-sync + tst-wfile-sync tst-bz28828 tests-internal = tst-vtables tst-vtables-interposed diff --git a/libio/tst-bz28828.c b/libio/tst-bz28828.c new file mode 100644 index 0000000000..f5849d2ca6 --- /dev/null +++ b/libio/tst-bz28828.c @@ -0,0 +1,10 @@ +#include <stdio.h> +#include <wchar.h> +int main(int ac, char **av) +{ + setvbuf(stdout, NULL, _IOLBF, 0); + fgetwc(stdin); + fputwc(10, stdout); /*SHOUDN'T CRASH HERE!*/ + return 0; +} + diff --git a/libio/tst-bz28828.input b/libio/tst-bz28828.input new file mode 100644 index 0000000000..ce01362503 --- /dev/null +++ b/libio/tst-bz28828.input @@ -0,0 +1 @@ +hello diff --git a/libio/wfileops.c b/libio/wfileops.c index fb9d45b677..b59a98881f 100644 --- a/libio/wfileops.c +++ b/libio/wfileops.c @@ -412,7 +412,8 @@ _IO_wfile_overflow (FILE *f, wint_t wch) return WEOF; } /* If currently reading or no buffer allocated. */ - if ((f->_flags & _IO_CURRENTLY_PUTTING) == 0) + if ((f->_flags & _IO_CURRENTLY_PUTTING) == 0 + || f->_wide_data->_IO_write_base == NULL) { /* Allocate a buffer if needed. */ if (f->_wide_data->_IO_write_base == 0)