[musl] [PATCH] dynlink: fix get_lfs64() with posix_fallocate64

[musl] [PATCH] dynlink: fix get_lfs64() with posix_fallocate64

[Florian Ziesche]

Hi,

this patch increases the buffer size by one in get_lfs64() so that it
works with posix_fallocate64.
"posix_fallocate64" is 17 characters long, so 16 is one too short.

Simplified example:
before: https://compiler-explorer.com/z/4qcPhcaWr
after: https://compiler-explorer.com/z/scGvhddKW

---
diff --git a/ldso/dynlink.c b/ldso/dynlink.c
index 42687da2..8707ae1c 100644
--- a/ldso/dynlink.c
+++ b/ldso/dynlink.c
@@ -363,7 +363,7 @@ static struct symdef get_lfs64(const char *name)
         "stat\0statfs\0statvfs\0tmpfile\0truncate\0versionsort\0"
         "__fxstat\0__fxstatat\0__lxstat\0__xstat\0";
     size_t l;
-    char buf[16];
+    char buf[17];
     for (l=0; name[l]; l++) {
         if (l >= sizeof buf) goto nomatch;
         buf[l] = name[l];

Re: [musl] [PATCH] dynlink: fix get_lfs64() with posix_fallocate64

[Rich Felker]

On Fri, May 10, 2024 at 06:01:58PM +0200, Florian Ziesche wrote:
> Hi,
> 
> this patch increases the buffer size by one in get_lfs64() so that it
> works with posix_fallocate64.
> "posix_fallocate64" is 17 characters long, so 16 is one too short.
> 
> Simplified example:
> before: https://compiler-explorer.com/z/4qcPhcaWr
> after: https://compiler-explorer.com/z/scGvhddKW
> 
> ---
> diff --git a/ldso/dynlink.c b/ldso/dynlink.c
> index 42687da2..8707ae1c 100644
> --- a/ldso/dynlink.c
> +++ b/ldso/dynlink.c
> @@ -363,7 +363,7 @@ static struct symdef get_lfs64(const char *name)
>          "stat\0statfs\0statvfs\0tmpfile\0truncate\0versionsort\0"
>          "__fxstat\0__fxstatat\0__lxstat\0__xstat\0";
>      size_t l;
> -    char buf[16];
> +    char buf[17];
>      for (l=0; name[l]; l++) {
>          if (l >= sizeof buf) goto nomatch;
>          buf[l] = name[l];

Thanks for catching this. I questioned whether the 17 is sufficient,
but indeed the buffer is never nul terminated except when removing the
64, so it should be ok.

I think I'll apply your patch as a direct fix, but the whole copying
operation is unnecessary and I'll probably remove it. It works just as
well to strncmp and pass p instead of buf to find_sym, so that a
mutable copy of the name to lookup is never needed.

Rich

Re: [musl] [PATCH] dynlink: fix get_lfs64() with posix_fallocate64

[Rich Felker]

[-- Attachment #1: Type: text/plain, Size: 1532 bytes --]

On Thu, May 23, 2024 at 04:28:03PM -0400, Rich Felker wrote:
> On Fri, May 10, 2024 at 06:01:58PM +0200, Florian Ziesche wrote:
> > Hi,
> > 
> > this patch increases the buffer size by one in get_lfs64() so that it
> > works with posix_fallocate64.
> > "posix_fallocate64" is 17 characters long, so 16 is one too short.
> > 
> > Simplified example:
> > before: https://compiler-explorer.com/z/4qcPhcaWr
> > after: https://compiler-explorer.com/z/scGvhddKW
> > 
> > ---
> > diff --git a/ldso/dynlink.c b/ldso/dynlink.c
> > index 42687da2..8707ae1c 100644
> > --- a/ldso/dynlink.c
> > +++ b/ldso/dynlink.c
> > @@ -363,7 +363,7 @@ static struct symdef get_lfs64(const char *name)
> >          "stat\0statfs\0statvfs\0tmpfile\0truncate\0versionsort\0"
> >          "__fxstat\0__fxstatat\0__lxstat\0__xstat\0";
> >      size_t l;
> > -    char buf[16];
> > +    char buf[17];
> >      for (l=0; name[l]; l++) {
> >          if (l >= sizeof buf) goto nomatch;
> >          buf[l] = name[l];
> 
> Thanks for catching this. I questioned whether the 17 is sufficient,
> but indeed the buffer is never nul terminated except when removing the
> 64, so it should be ok.
> 
> I think I'll apply your patch as a direct fix, but the whole copying
> operation is unnecessary and I'll probably remove it. It works just as
> well to strncmp and pass p instead of buf to find_sym, so that a
> mutable copy of the name to lookup is never needed.

Applied. Does the attached look ok to get rid of the copy entirely? It
passed basic smoke testing here.


[-- Attachment #2: get_lfs64.diff --]
[-- Type: text/plain, Size: 919 bytes --]

diff --git a/ldso/dynlink.c b/ldso/dynlink.c
index 8707ae1c..3b57c07f 100644
--- a/ldso/dynlink.c
+++ b/ldso/dynlink.c
@@ -362,19 +362,14 @@ static struct symdef get_lfs64(const char *name)
 		"pwritev\0readdir\0scandir\0sendfile\0setrlimit\0"
 		"stat\0statfs\0statvfs\0tmpfile\0truncate\0versionsort\0"
 		"__fxstat\0__fxstatat\0__lxstat\0__xstat\0";
-	size_t l;
-	char buf[17];
-	for (l=0; name[l]; l++) {
-		if (l >= sizeof buf) goto nomatch;
-		buf[l] = name[l];
-	}
 	if (!strcmp(name, "readdir64_r"))
 		return find_sym(&ldso, "readdir_r", 1);
-	if (l<2 || name[l-2]!='6' || name[l-1]!='4')
+	size_t l = strnlen(name, 18);
+	if (l<2 || name[l-2]!='6' || name[l-1]!='4' || name[l])
 		goto nomatch;
-	buf[l-=2] = 0;
 	for (p=lfs64_list; *p; p++) {
-		if (!strcmp(buf, p)) return find_sym(&ldso, buf, 1);
+		if (!strncmp(name, p, l-2) && !p[l-2])
+			return find_sym(&ldso, p, 1);
 		while (*p) p++;
 	}
 nomatch:

Re: [musl] [PATCH] dynlink: fix get_lfs64() with posix_fallocate64

[Florian Ziesche]

[-- Attachment #1: Type: text/plain, Size: 1660 bytes --]

Yep, this works for me.

> On 22. Jun 2024, at 03:06, Rich Felker <dalias@libc.org> wrote:
> 
> On Thu, May 23, 2024 at 04:28:03PM -0400, Rich Felker wrote:
>> On Fri, May 10, 2024 at 06:01:58PM +0200, Florian Ziesche wrote:
>>> Hi,
>>> 
>>> this patch increases the buffer size by one in get_lfs64() so that it
>>> works with posix_fallocate64.
>>> "posix_fallocate64" is 17 characters long, so 16 is one too short.
>>> 
>>> Simplified example:
>>> before: https://compiler-explorer.com/z/4qcPhcaWr
>>> after: https://compiler-explorer.com/z/scGvhddKW
>>> 
>>> ---
>>> diff --git a/ldso/dynlink.c b/ldso/dynlink.c
>>> index 42687da2..8707ae1c 100644
>>> --- a/ldso/dynlink.c
>>> +++ b/ldso/dynlink.c
>>> @@ -363,7 +363,7 @@ static struct symdef get_lfs64(const char *name)
>>>         "stat\0statfs\0statvfs\0tmpfile\0truncate\0versionsort\0"
>>>         "__fxstat\0__fxstatat\0__lxstat\0__xstat\0";
>>>     size_t l;
>>> -    char buf[16];
>>> +    char buf[17];
>>>     for (l=0; name[l]; l++) {
>>>         if (l >= sizeof buf) goto nomatch;
>>>         buf[l] = name[l];
>> 
>> Thanks for catching this. I questioned whether the 17 is sufficient,
>> but indeed the buffer is never nul terminated except when removing the
>> 64, so it should be ok.
>> 
>> I think I'll apply your patch as a direct fix, but the whole copying
>> operation is unnecessary and I'll probably remove it. It works just as
>> well to strncmp and pass p instead of buf to find_sym, so that a
>> mutable copy of the name to lookup is never needed.
> 
> Applied. Does the attached look ok to get rid of the copy entirely? It
> passed basic smoke testing here.
> 
> <get_lfs64.diff>


[-- Attachment #2: Type: text/html, Size: 5420 bytes --]