pseudo_util: Fix resolving relative paths from "/"
PATCH 1/1 pseudo_util: Fix resolving relative paths from "/"
pseudo_fix_path() incorrectly assumed that 'base' never ends with a slash because it's a canonical path. However, base can be "/" - a path which is canonical and yet starts with a slash. This happens when pseudo_cwd is "/" or when we're starting from a dirfd pointing to "/". The wrong result from pseudo_fix_path() caused the database lookup to fail and made pseudo abort.
---
Hi. Here is my third patch for 'pseudo'. I included a test that demonstrates the fix. Try "make test" with old and new pseudo_util.c to see the difference. Run it with PSEUDO_DEBUG to see what happens and how it leads to an abort.
JFYI: The real fix is to check "baselen == 1". The "rootlen == 1" condition is just for symmetry. Unlike pseudo_cwd and fd_path(), which can be "/", I think pseudo_chroot will never be "/" because pseudo_client_chroot sets it to NULL instead. So rootlen should never be 1. It's just to be safe.
pseudo_util.c | 12 +++++++++--- test/test-relative-from-root.sh | 14 ++++++++++++++ 2 files changed, 23 insertions(+), 3 deletions(-) create mode 100755 test/test-relative-from-root.sh
pseudo_util.c +9 -3
@@ -792,9 +792,9 @@ static char *pathbufs[PATHBUFS] = { 0 };static int pathbuf =0 ;/* Canonicalize path. "base", if present, is an already-canonicalized -* path of baselen characters, presumed not to end in a /. path is -* the new path to be canonicalized. The tricky part is that path may -* contain symlinks, which must be resolved. +* path of baselen characters, presumed not to end in a / unless it is +* just "/". path is the new path to be canonicalized. The tricky part +* is that path may contain symlinks, which must be resolved. */ char *pseudo_fix_path (const char *base,const char *path,size_t rootlen,size_t baselen,size_t *lenp,int leave_last) {
@@ -808,6 +808,12 @@ pseudo_fix_path(const char *base, const char *path, size_t rootlen, size_t baselpseudo_diag ("can't fix empty path.\n" );return 0 ; } +if (baselen ==1 ) { + baselen =0 ; + } +if (rootlen ==1 ) { + rootlen =0 ; + } newpathlen =pseudo_path_max (); pathlen =strlen (path);/* Crazy shell code (e.g. libtool) can pass in a command pipeline as a path which exceeds the max path
test/test-relative-from-root.sh new file +14 -0
@@ -0,0 +1,14 @@ +#!/bin/bash +# pseudo had a bug that made it abort() when looking up a relative path from +# base "/", such as by openat(dirfd_of_root, "foo/bar") or when cwd is /. It +# tried to look up base+"/"+path = "//foo/bar", which is wrong. + +set -e + +touch f1 +relative_pwd=${PWD#/} + +cd / +cat "$relative_pwd/f1" + +rm "$relative_pwd/f1"
Signature · 1 line
2.25.1
Quoted 3 lines · On Wed, 24 Aug 2022 16:51:36 +0000 "Tomi Belan" <[email protected]> wrote:
pseudo_fix_path()incorrectly assumed that 'base' never ends with a slash because it's a canonical path. However, base can be "/" - a path which is canonical and yet starts with a slash. This happens whenpseudo_cwdis "/" or
^^^^^^
starts -> ends?
Quoted 4 lines
when we're starting from a dirfd pointing to "/". The wrong result from
pseudo_fix_path()caused the database lookup to fail and made pseudo abort.Signed-off-by Tomi Belan <[email protected]>
Luca
Signature · 3 lines
Luca Ceresoli, Bootlin Embedded Linux and Kernel engineering https://bootlin.com
Quoted 3 lines · Luca Ceresoli
Quoted 1 line
canonical and yet starts with a slash. This happens when
pseudo_cwdis "/" or^^^^^^starts -> ends?
Oops, yes, of course. Good catch.
Should I send a v2 patch? :)
The corrected commit message line should be:
---
canonical and yet ends with a slash. This happens when pseudo_cwd is "/" or
---
Tomi
Hi Tomi,
Quoted 9 lines · On Wed, 24 Aug 2022 23:09:51 +0200 "Tomi Belan" <[email protected]> wrote:
Quoted 3 lines · Luca Ceresoli
Quoted 1 line
canonical and yet starts with a slash. This happens when
pseudo_cwdis "/" or^^^^^^starts -> ends?
Oops, yes, of course. Good catch.
Should I send a v2 patch? :)
No need. Richard already took your patch and fixed the message. It's currently in the oe-core-next branch:
https://git.yoctoproject.org/pseudo/log/?h=oe-core-next
Signature · 3 lines
Luca Ceresoli, Bootlin Embedded Linux and Kernel engineering https://bootlin.com