Avoid leaking duplicated file descriptors in corner cases.
pg_dump's compression modules had variations on the theme of
fp = fdopen(dup(fd), mode);
if (fp == NULL)
// fail, reporting errno
which is problematic for two reasons. First, if dup() succeeds but
fdopen() fails, we'd leak the duplicated FD. That's not important
at present since the program will just exit immediately after failure
anyway; but perhaps someday we'll try to continue, making the resource
leak potentially significant. Second, if dup() fails then fdopen()
will overwrite the useful errno (perhaps EMFILE) with a misleading
value EBADF, making it difficult to understand what went wrong.
Fix both issues by testing for dup() failure before proceeding to
the next call.
These failures are sufficiently unlikely, and the consequences minor
enough, that this doesn't seem worth the effort to back-patch.
But let's fix it in HEAD.
Author: Jianghua Yang <yjhjstz@gmail.com>
Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us>
Discussion: https://postgr.es/m/62bbe34d-2315-4b42-b768-56d901aa83e1@gmail.com
Branch
------
master
Details
-------
https://git.postgresql.org/pg/commitdiff/8b02c22bb43cb480f437704dc547ea77196b7e93
Modified Files
--------------
src/bin/pg_dump/compress_gzip.c | 20 ++++++++++++++++----
src/bin/pg_dump/compress_lz4.c | 27 ++++++++++++++++++++++-----
src/bin/pg_dump/compress_none.c | 20 ++++++++++++++++----
src/bin/pg_dump/compress_zstd.c | 28 ++++++++++++++++++++++------
4 files changed, 76 insertions(+), 19 deletions(-)
