[WASM] Integer Overflow on 'size' after stat'ing #22924
Comments
|
I "fixed" this by patching diff --git a/pp_sys.c b/pp_sys.c
index 5f8bb0d6ed..975e3a3877 100644
--- a/pp_sys.c
+++ b/pp_sys.c
@@ -2271,7 +2271,7 @@ PP_wrapped(pp_syswrite, 0, 1)
if (MARK >= SP) {
length = blen;
} else {
-#if Size_t_size > IVSIZE
+#if Size_t_size > IVSIZE || defined(__wasi__)
length = (Size_t)SvNVx(*++MARK);
#else
length = (Size_t)SvIVx(*++MARK);
@@ -2313,7 +2313,7 @@ PP_wrapped(pp_syswrite, 0, 1)
goto say_undef;
SP = ORIGMARK;
-#if Size_t_size > IVSIZE
+#if Size_t_size > IVSIZE || defined(__wasi__)
PUSHn(retval);
#else
PUSHi(retval);
@@ -2418,7 +2418,7 @@ PP_wrapped(pp_tell, MAXARG, 0)
RETURN;
}
-#if LSEEKSIZE > IVSIZE
+#if LSEEKSIZE > IVSIZE || defined(__wasi__)
PUSHn( (NV)do_tell(gv) );
#else
PUSHi( (IV)do_tell(gv) );
@@ -2433,7 +2433,7 @@ PP_wrapped(pp_sysseek, 3, 0)
{
dSP;
const int whence = POPi;
-#if LSEEKSIZE > IVSIZE
+#if LSEEKSIZE > IVSIZE || defined(__wasi__)
const Off_t offset = (Off_t)SvNVx(POPs);
#else
const Off_t offset = (Off_t)SvIVx(POPs);
@@ -2445,7 +2445,7 @@ PP_wrapped(pp_sysseek, 3, 0)
if (io) {
const MAGIC * const mg = SvTIED_mg((const SV *)io, PERL_MAGIC_tiedscalar);
if (mg) {
-#if LSEEKSIZE > IVSIZE
+#if LSEEKSIZE > IVSIZE || defined(__wasi__)
SV *const offset_sv = newSVnv((NV) offset);
#else
SV *const offset_sv = newSViv(offset);
@@ -2464,7 +2464,7 @@ PP_wrapped(pp_sysseek, 3, 0)
PUSHs(&PL_sv_undef);
else {
SV* const sv = sought ?
-#if LSEEKSIZE > IVSIZE
+#if LSEEKSIZE > IVSIZE || defined(__wasi__)
newSVnv((NV)sought)
#else
newSViv(sought)
@@ -2486,7 +2486,7 @@ PP_wrapped(pp_truncate, 2, 0)
/* XXX Configure probe for the length type of *truncate() needed XXX */
Off_t len;
-#if Off_t_size > IVSIZE
+#if Off_t_size > IVSIZE || defined(__wasi__)
len = (Off_t)POPn;
#else
len = (Off_t)POPi;
@@ -3267,7 +3267,7 @@ PP_wrapped(pp_stat, !(PL_op->op_flags & OPf_REF), 0)
#else
PUSHs(newSVpvs_flags("", SVs_TEMP));
#endif
-#if Off_t_size > IVSIZE
+#if Off_t_size > IVSIZE || defined(__wasi__)
mPUSHn(PL_statcache.st_size);
#else
mPUSHi(PL_statcache.st_size);
@@ -3524,7 +3524,7 @@ PP(pp_ftis)
dTARGET;
switch (op_type) {
case OP_FTSIZE:
-#if Off_t_size > IVSIZE
+#if Off_t_size > IVSIZE || defined(__wasi__)
sv_setnv(TARG, (NV)PL_statcache.st_size);
#else
sv_setiv(TARG, (IV)PL_statcache.st_size); |
|
So either #define VALUE_TO_STRING(x) #x
#define VALUE(x) VALUE_TO_STRING(x)
#define VAR_NAME_VALUE(var) #var "=" VALUE(var)
#pragma message(VAR_NAME_VALUE(Size_t_size))
#pragma message(VAR_NAME_VALUE(LSEEKSIZE))
#pragma message(VAR_NAME_VALUE(Off_t_size))
#pragma message(VAR_NAME_VALUE(IVSIZE))Or maybe it determines these correctly and |
That is exactly what I wanted to recommend you when I had only read the title, so that is weird indeed. What is the value of Also, I'm not sure you should set |
I tried with It's entirely possible in the rapid testing of builds I ended up supplying values that conflict with each other. I added some additional logging and this is what is reported when I call === File Size Configuration === For |
|
Just a wild guess, but what happens if you change that |
That also worked, though I applied the same idea to diff --git a/pp_sys.c b/pp_sys.c
index 5f8bb0d6ed..6dfb479e04 100644
--- a/pp_sys.c
+++ b/pp_sys.c
@@ -2274,7 +2274,7 @@ PP_wrapped(pp_syswrite, 0, 1)
#if Size_t_size > IVSIZE
length = (Size_t)SvNVx(*++MARK);
#else
- length = (Size_t)SvIVx(*++MARK);
+ length = (Size_t)SvUVx(*++MARK);
#endif
if ((SSize_t)length < 0) {
DIE(aTHX_ "Negative length");
@@ -2316,7 +2316,7 @@ PP_wrapped(pp_syswrite, 0, 1)
#if Size_t_size > IVSIZE
PUSHn(retval);
#else
- PUSHi(retval);
+ PUSHu(retval);
#endif
RETURN;
@@ -2421,7 +2421,7 @@ PP_wrapped(pp_tell, MAXARG, 0)
#if LSEEKSIZE > IVSIZE
PUSHn( (NV)do_tell(gv) );
#else
- PUSHi( (IV)do_tell(gv) );
+ PUSHu( (UV)do_tell(gv) );
#endif
RETURN;
}
@@ -2436,7 +2436,7 @@ PP_wrapped(pp_sysseek, 3, 0)
#if LSEEKSIZE > IVSIZE
const Off_t offset = (Off_t)SvNVx(POPs);
#else
- const Off_t offset = (Off_t)SvIVx(POPs);
+ const Off_t offset = (Off_t)SvUVx(POPs);
#endif
GV * const gv = PL_last_in_gv = MUTABLE_GV(POPs);
@@ -2448,7 +2448,7 @@ PP_wrapped(pp_sysseek, 3, 0)
#if LSEEKSIZE > IVSIZE
SV *const offset_sv = newSVnv((NV) offset);
#else
- SV *const offset_sv = newSViv(offset);
+ SV *const offset_sv = newSVuv(offset);
#endif
return tied_method2(SV_CONST(SEEK), SP, MUTABLE_SV(io), mg, offset_sv,
@@ -2467,7 +2467,7 @@ PP_wrapped(pp_sysseek, 3, 0)
#if LSEEKSIZE > IVSIZE
newSVnv((NV)sought)
#else
- newSViv(sought)
+ newSVuv(sought)
#endif
: newSVpvn(zero_but_true, ZBTLEN);
mPUSHs(sv);
@@ -2489,7 +2489,7 @@ PP_wrapped(pp_truncate, 2, 0)
#if Off_t_size > IVSIZE
len = (Off_t)POPn;
#else
- len = (Off_t)POPi;
+ len = (Off_t)POPu;
#endif
/* Checking for length < 0 is problematic as the type might or
* might not be signed: if it is not, clever compilers will moan. */
@@ -3270,7 +3270,7 @@ PP_wrapped(pp_stat, !(PL_op->op_flags & OPf_REF), 0)
#if Off_t_size > IVSIZE
mPUSHn(PL_statcache.st_size);
#else
- mPUSHi(PL_statcache.st_size);
+ mPUSHu(PL_statcache.st_size);
#endif
#ifdef BIG_TIME
mPUSHn(PL_statcache.st_atime);
@@ -3527,7 +3527,7 @@ PP(pp_ftis)
#if Off_t_size > IVSIZE
sv_setnv(TARG, (NV)PL_statcache.st_size);
#else
- sv_setiv(TARG, (IV)PL_statcache.st_size);
+ sv_setiv(TARG, (UV)PL_statcache.st_size);
#endif
break;
case OP_FTMTIME:I might have done something incorrect though, so I can look with fresh eyes later. |
|
This may be this issue. That said, that doesn't explain the the 32-bit part of the issue. My best guess is that it's because of |
The return values of |
|
Thanks @Leont. I removed diff --git a/pp_sys.c b/pp_sys.c
index 5f8bb0d6ed..ad8e46e692 100644
--- a/pp_sys.c
+++ b/pp_sys.c
@@ -3270,7 +3270,7 @@ PP_wrapped(pp_stat, !(PL_op->op_flags & OPf_REF), 0)
#if Off_t_size > IVSIZE
mPUSHn(PL_statcache.st_size);
#else
- mPUSHi(PL_statcache.st_size);
+ mPUSHu(PL_statcache.st_size);
#endif
#ifdef BIG_TIME
mPUSHn(PL_statcache.st_atime);
@@ -3527,7 +3527,7 @@ PP(pp_ftis)
#if Off_t_size > IVSIZE
sv_setnv(TARG, (NV)PL_statcache.st_size);
#else
- sv_setiv(TARG, (IV)PL_statcache.st_size);
+ sv_setuv(TARG, (UV)PL_statcache.st_size);
#endif
break;
case OP_FTMTIME: |
|
Can you look up the type of |
|
In |
The behavior we're observing is strongly suggesting the type of We're already detecting the signedness of |
I'm also curious of the cause, but |
st_dev_sign is |
I've come up against a bit of a wall while porting Perl to WASM (using wasi-sdk/wasi-libc). I've gotten everything working, except for an issue with files over 2GB - when stat'ing them, the file size comes back wrong:
The negative size (-1073741824) suggests we're hitting a 32-bit integer overflow, which is odd since the build is configured with both
use64bitint=defineanduse64bitall=define. I've double-checked thativsizeis set to 8 andlseeksizeis 8 in the config.Before opening this issue I confirmed it wasn't an upstream issue in clang or wasi-libc. I'm attaching my full perl config output below.
I understand this may not be a priority issue since wasm32-wasi isn't an official target yet, but I'd appreciate any guidance on where I can look within Perl's source to potentially patch this, or if there are any specific flags in my configuration hint file I should try.
Full Configuration
While the above configuration says 5.41.7, I've also tested on 5.40.0 and the issue persist
The only other relevant issue I could find was this one from 1999.
The text was updated successfully, but these errors were encountered: