Fix readlink() for non-PostgreSQL junction points on Windows.
authorThomas Munro <[email protected]>
Tue, 25 Oct 2022 02:21:42 +0000 (15:21 +1300)
committerThomas Munro <[email protected]>
Tue, 25 Oct 2022 03:19:05 +0000 (16:19 +1300)
Since commit c5cb8f3b taught stat() to follow symlinks, and since initdb
uses pg_mkdir_p(), and that examines parent directories, our humble
readlink() implementation can now be exposed to junction points not of
PostgreSQL origin.  Those might be corrupted by our naive path mangling,
which doesn't really understand NT paths in general.

Simply decline to transform paths that don't look like a drive absolute
path.  That means that readlink() returns the NT path directly when
checking a parent directory of PGDATA that happen to point to a drive
using "rooted" format.  That  works for the purposes of our stat()
emulation.

Reported-by: Roman Zharkov <[email protected]>
Reviewed-by: Roman Zharkov <[email protected]>
Discussion: https://postgr.es/m/4590c37927d7b8ee84f9855d83229018%40postgrespro.ru
Discussion: https://postgr.es/m/CA%2BhUKG%2BajSQ_8eu2AogTncOnZ5me2D-Cn66iN_-wZnRjLN%2Bicg%40mail.gmail.com

src/port/dirmod.c

index 398787360a2b642a4d8dc1f2a60b0b1af317022d..d83316d6a2ae4724317d79c7c2fbfd45a702c30d 100644 (file)
@@ -363,10 +363,21 @@ pgreadlink(const char *path, char *buf, size_t size)
    r -= 1;
 
    /*
-    * If the path starts with "\??\", which it will do in most (all?) cases,
-    * strip those out.
+    * If the path starts with "\??\" followed by a "drive absolute" path
+    * (known to Windows APIs as RtlPathTypeDriveAbsolute), then strip that
+    * prefix.  This undoes some of the transformation performed by
+    * pqsymlink(), to get back to a format that users are used to seeing.  We
+    * don't know how to transform other path types that might be encountered
+    * outside PGDATA, so we just return them directly.
     */
-   if (r > 4 && strncmp(buf, "\\??\\", 4) == 0)
+   if (r >= 7 &&
+       buf[0] == '\\' &&
+       buf[1] == '?' &&
+       buf[2] == '?' &&
+       buf[3] == '\\' &&
+       isalpha(buf[4]) &&
+       buf[5] == ':' &&
+       buf[6] == '\\')
    {
        memmove(buf, buf + 4, strlen(buf + 4) + 1);
        r -= 4;