WIP Add resolving of $PROGRAM_NAME from /dev/fd/number #23358

michal-josef-spacek · 2025-06-05T14:40:16Z

There are situations when $PROGRAM_NAME is something /dev/fd/number

example like sudo /usr/local/bin/<tool> with checksum verification - Issue with $PROGRAM_NAME (sudo + checksum verification) #23351
→ Resolves to the tool path
example with pipe: perl <( echo '#!perl -DA'; echo 'print "$0\n"');
→ Stay with /dev/fd/number

This set of changes requires a perldelta entry, and I need help writing it.

jkeenan

These changes need to be exercised by our test suite. Such a test will have to take into account the quirks of different operating systems.

perl.c

There are situations when $PROGRAM_NAME is something /dev/fd/<number> 1) example like sudo /usr/local/bin/<tool> with checksum verification (GH#23351) 2) example with pipe: perl <( echo '#!perl -DA'; echo 'print "$0\n"');

michal-josef-spacek · 2025-06-06T09:17:34Z

There were issues in the build on Windows, I updated PR.

michal-josef-spacek · 2025-06-06T12:12:48Z

hm, issue on Windows perl.obj : error LNK2019: unresolved external symbol readlink referenced in function S_open_script

Leont · 2025-06-06T16:04:19Z

This code should probably be feature guarded.

That means Configure should check if "/proc/self/fd/%d" actually works on that platform.

bulk88 · 2025-06-10T20:02:13Z

What does this C code do when it executes on on Plan9, OS/2, Windows, and VMS?

michal-josef-spacek · 2025-06-10T20:11:57Z

What does this C code do when it executes on on Plan9, OS/2, Windows, and VMS?

The actual plan is to create some configuration option to detect /proc/self/fd/%d and add a define to the actual code. ok?

bulk88 · 2025-06-10T20:13:05Z

hm, issue on Windows perl.obj : error LNK2019: unresolved external symbol readlink referenced in function S_open_script

He tried to call readlink() on Windows. @bulk88 pulls out

P.S. readlink() exists on WinOS, but this PR/this patch, has no business knowing how to correctly spell the identifier to invoke Win32 OS's readlink() function.

Update:
JK JK this PR/this patch on Win32, also has no business knowing how to correctly spell in 7-bit ASCII the /proc and /dev directories on Windows. </joke>

I would be strongly against any patch that knows hows to reach NT Kernel's root dir called / and strip off the public api C:/ prefix. That "magic spell" is undocumented API, and has zero src code compatibility with POSIX targeted src code. All of WinNT's secret / directories are written in CamelCase/PascalCase with no spelling abbreviations 😁. Also don't ask how to locate the absolute path to the Windows Registry, or mount the Windows Registry database somewhere under /. And don't ask how get any PID's real time CPU usage and RSS and VSZ mem usage from a certain folder in the Windows Regsitry.

bulk88 · 2025-06-10T19:24:12Z

perl.c

@@ -4227,6 +4227,19 @@ S_open_script(pTHX_ const char *scriptname, bool dosearch, bool *suidscript)
                Safefree(PL_origfilename);
                PL_origfilename = (char *)scriptname;
            }
+            else {
+                char proc_fd_path[64];
+                snprintf(proc_fd_path, sizeof(proc_fd_path), "/proc/self/fd/%d", fdscript);


"/proc/self/fd/" is 14 +1 chars
"/proc/self/fd/%d" is 16+ 1 chars

"%d" max legal expansion is 21 chars according to google but plz verify it, or find perl.h's or handy.h's official macro for max base10 expansion

"%d" is signed, what does "/proc/self/fd/-13" means on Linux?

The c stack buffer should really be 16+21+1 == 38 or with sizeof(size_t) round up logic, (16+21+1)/8 == 4.75; 5*8 == 40 b/c why not?

The CC will round up C auto var char[38] to the next 4 or 8 alignment mark no matter what you do.

At this point fdscript is a non-negative integer that fits in an int, see the code above that sets it up.

bulk88 · 2025-06-10T19:28:05Z

perl.c

@@ -4227,6 +4227,19 @@ S_open_script(pTHX_ const char *scriptname, bool dosearch, bool *suidscript)
                Safefree(PL_origfilename);
                PL_origfilename = (char *)scriptname;
            }
+            else {


this else { block starting here really should be split off into a separate static, not-inline function, see my other comment below why this is needed

tonycoz · 2025-07-07T02:03:32Z

perl.c

+                snprintf(proc_fd_path, sizeof(proc_fd_path), "/proc/self/fd/%d", fdscript);
+                char target_path[MAXPATHLEN];
+                SSize_t len = readlink(proc_fd_path, target_path, sizeof(target_path) - 1);
+                if (len != -1) {


While I'm not fond of large buffers on the stack we don't have a general solution at this point beyond using an SV or Newx().

At this point the stack is close to the least consumed it will be, so I think this buffer is as reasonable as it could be.

Some sort of cheap C level path abstraction would be useful - a relatively small built-in buffer, but can keep a dynamically allocated buffer if that's too small. But for now I think this is the practical choice.

char buf128chr[128]; #define pl_alloca(_l) ((_l) < sizeof(buf128chr) ? NUM2PTR(char *,buf128chr) : (_l) < U16_MAX ? alloca(_l) : SvPVX(sv_2mortal(newSV(_l))))

is my typical recipe. I don't expect anyone to do this but me, b/c its overkill to tweak char stk buffers against -O2 disassembly, I will sometimes adjust 128 downwards or upwards to a formula like
char bufchr[256 - sub rsp, 50h - 3*sizeof(void *)];
char bufchr[256 - sub rsp, 20h - 3*sizeof(void *)]

For generic guesstimates, I'd say len 96 (sse 166=96, 167=112) covers all sane src code identifiers and consts and paths with some breathing room. 128 is the max I'd put generically, its rare to see a function with more than 8 c stk vars active (liveness) at any time on x64 on top of whatever CC keeps in Win64's 8 non-vol regs.

C99's variable length arrays feature and alloca() are synonyms in ICC/Clang/GCC machine code for Win32/Win64 but that might be OS specific and shouldn't be coded again. On paper C99's VLA spec allows de-allocation on scope leave, so else { char buf[len < 256? len : 1]; }, ptr to buf is gone/invalid/now aliased to a diff and wrong var, after the closing curly, but alloca() ptrs lasts until function return per spec. IDK if any CC on any arch actually rewinds the stack pointer at the closing curly, or recycles pieces of c stack for multiple VLA buffers in same function but diff scopes. But I wouldn't ever write code assuming a ptr to a VLA buf can outlast its scope curlys, b/c any CC can add the optimization in the future randomly.

VLA doesn't exist in C++ so alloca() is the tool to use in C++ mode.

If someone wants to be standards pedantic about alloca/vla'es, alloca/vlas have to exist internally in the C compiler and in the ABI/machine code level or its not C anymore. Recursion is part of the spec. 1960s HW where recursion or function reentry was electronically impossible, since the pre-allocated C stack and the functions machine code are adjacent on the same magnetic tape is irrelevant

alloca() can always be synthesized per ISO C spec. I can always write a dispatcher function, that takes a length, a function ptr, and va_list, then round to nearest 16/32/64, then uses a switch(){} to pick 1 of 24 or 48 functions.

Each of 48 functions takes a function ptr and a va_list as args, then declares a fixed size char buf[32_64_96_128_160];, then calls the func ptr with ptr to buf, and the va_list ptr. Pedantic compliance problem solved.

At this point the stack is close to the least consumed it will be, so I think this buffer is as reasonable as it could be.

I did some breakpoints, your right, S_open_script(pTHX_ const char *scriptname, bool dosearch, bool *suidscript) is a 1x ever early on startup call and returns to caller basically instantly, its not part of BEGIN{}/do/require/use infrastructure, which can get 3-5 runloops/yylexers deep on the C stack. I thought it was in my 1st comment. but its not after grepping, setting BPs and watching it. So a 1KB fixed len buffer is totally fine here.

jkeenan

This does not appear to be ready for merging, as it's still failing tests on Windows. Those need to be resolved, after which I would request review by @tonycoz

tonycoz · 2025-07-07T01:16:45Z

perl.c

+                char proc_fd_path[64];
+                snprintf(proc_fd_path, sizeof(proc_fd_path), "/proc/self/fd/%d", fdscript);
+                char target_path[MAXPATHLEN];
+                SSize_t len = readlink(proc_fd_path, target_path, sizeof(target_path) - 1);


This should use PerlLIO_readlink() instead of raw readlink().

The else block should be #if conditional on HAS_READLINK in addition to the /proc/self/... option you suggested.

This should use PerlLIO_readlink() instead of raw readlink().

The else block should be #if conditional on HAS_READLINK in addition to the /proc/self/... option you suggested.

At work, I have to use these constants and throw in the original string constant with STRLENs(), for calculating char stack buffers.
https://learn.microsoft.com/en-us/cpp/c-runtime-library/reference/itoa-itow?view=msvc-170#maximum-conversion-count-macros

Ive never seen any max expansion constants like the MS ones in the Perl headers for Perl_my_sprintf() and friends, regarding maximum char expansion in hex and base 10 of U8/I8/I16/I32/I64/void ptr/UV/IV.

I've looked at these briefly

perl5/perl.h

Line 2461 in 8711297

#define IV_DIG (BIT_DIGITS(IVSIZE * 8))

but I dont trust them. I won't myself use them.

macro UV_DIG has ZERO users in the repo. IV_DIG exactly ONE grep hit
https://github.com/Perl/perl5/blob/blead/sv.c#L12295

for proc_fd_path id rather see STRLENs("/proc/self/fd/%d") + 21 +1 +12_whatever_padding_34 than just a random number like 64 which always make me paranoid when I see it b/c I can't verify it visually against the actual sprintf() args a few lines below, and IDK if someone before me quickly edited the fmt str and never updated the stk buf byte length to match. If the expression includes STRLENs("/proc/self/fd/%d")+whatever, paranoia gone.

jkeenan changed the title ~~Add resoving of $PROGRAM_NAME from /dev/fd/number~~ Add resolving of $PROGRAM_NAME from /dev/fd/number Jun 5, 2025

michal-josef-spacek force-pushed the fix_program_name_in_sudo branch 2 times, most recently from 085c026 to 99313c1 Compare June 5, 2025 15:24

jkeenan requested changes Jun 5, 2025

View reviewed changes

perl.c Outdated Show resolved Hide resolved

jkeenan added defer-next-dev This PR should not be merged yet, but await the next development cycle type-core labels Jun 5, 2025

michal-josef-spacek force-pushed the fix_program_name_in_sudo branch from 99313c1 to 6241f02 Compare June 5, 2025 16:32

Add resolving of $PROGRAM_NAME from /dev/fd/<number>

da4a0e7

There are situations when $PROGRAM_NAME is something /dev/fd/<number> 1) example like sudo /usr/local/bin/<tool> with checksum verification (GH#23351) 2) example with pipe: perl <( echo '#!perl -DA'; echo 'print "$0\n"');

michal-josef-spacek force-pushed the fix_program_name_in_sudo branch from 6a4065e to da4a0e7 Compare June 6, 2025 09:15

michal-josef-spacek requested a review from jkeenan June 6, 2025 09:16

michal-josef-spacek changed the title ~~Add resolving of $PROGRAM_NAME from /dev/fd/number~~ WIP Add resolving of $PROGRAM_NAME from /dev/fd/number Jun 10, 2025

bulk88 reviewed Jun 10, 2025

View reviewed changes

jkeenan removed the defer-next-dev This PR should not be merged yet, but await the next development cycle label Jul 4, 2025

jkeenan reviewed Jul 4, 2025

View reviewed changes

tonycoz reviewed Jul 7, 2025

View reviewed changes

WIP Add resolving of $PROGRAM_NAME from /dev/fd/number #23358

Are you sure you want to change the base?

WIP Add resolving of $PROGRAM_NAME from /dev/fd/number #23358

Uh oh!

Conversation

michal-josef-spacek commented Jun 5, 2025

Uh oh!

jkeenan left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

michal-josef-spacek commented Jun 6, 2025

Uh oh!

michal-josef-spacek commented Jun 6, 2025

Uh oh!

Leont commented Jun 6, 2025

Uh oh!

bulk88 commented Jun 10, 2025

Uh oh!

michal-josef-spacek commented Jun 10, 2025

Uh oh!

bulk88 commented Jun 10, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

bulk88 Jun 10, 2025

Choose a reason for hiding this comment

Uh oh!

tonycoz Jul 7, 2025

Choose a reason for hiding this comment

Uh oh!

bulk88 Jun 10, 2025

Choose a reason for hiding this comment

Uh oh!

This comment was marked as resolved.

Uh oh!

tonycoz Jul 7, 2025

Choose a reason for hiding this comment

Uh oh!

bulk88 Jul 7, 2025

Choose a reason for hiding this comment

Uh oh!

jkeenan left a comment

Choose a reason for hiding this comment

Uh oh!

tonycoz Jul 7, 2025

Choose a reason for hiding this comment

Uh oh!

bulk88 Jul 7, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

bulk88 commented Jun 10, 2025 •

edited

Loading