** Description changed:

  Ubuntu release: 24.04 LTS (noble), seen on arm64 and amd64
  Package: bash 5.2.21-2ubuntu4
  
  This is a low-severity report for the record. No stable release update
  is requested.
  
  == What happens ==
  
  When the exit builtin runs while a top-level command is still being executed, 
bash leaves without freeing the parsed command tree of that top-level command. 
valgrind memcheck reports the tree as "definitely lost". It happens when exit 
is called
   - from inside a function, an if, a loop, or a && / || list, or
   - at top level while an EXIT trap is set.
  
  It does not happen for a top-level exit with no EXIT trap, for a script
  that runs off its end, for a top-level exit after "trap - EXIT", or for
  bash -c 'exit 0'.
  
  == Steps to reproduce ==
  
   printf '#!/bin/bash\nf() { exit 0; }\nf\n' > /tmp/b.sh && chmod +x /tmp/b.sh
   valgrind --leak-check=full --show-leak-kinds=definite /tmp/b.sh
  
  Result (frames resolved with bash-dbgsym):
  
   72 (32 direct, 40 indirect) bytes in 1 blocks are definitely lost in loss 
record 320 of 344
     at malloc
     by make_simple_command (make_cmd.c:528)
     by yyparse (parse.y:782)
     by parse_command (eval.c:348)
     by read_command (eval.c:392)
     by reader_loop (eval.c:139)
     by main (shell.c:833)
  
  Expected: "definitely lost: 0 bytes in 0 blocks".
  
  == Variants measured ==
  
  bash 5.2.21-2ubuntu4, amd64, valgrind 3.22.0, each as a two- or three-
  line script file:
  
   exit 0 at top level, no trap: 0 bytes
   f() { exit 0; } then f: 72 bytes (32 direct, 40 indirect)
   if true; then exit 0; fi: 157 bytes (32 direct, 125 indirect)
   while true; do exit 0; done: 149 bytes (32 direct, 117 indirect)
   test 1 = 1 && exit 0: 259 bytes (32 direct, 227 indirect)
   trap "echo c" EXIT then exit 0 at top level: 56 bytes (32 direct, 24 
indirect)
   trap "echo c" EXIT then script ends: 0 bytes
   trap "echo c" EXIT, trap - EXIT, exit 0: 0 bytes
  
  The size follows the size of the top-level command that was running. It
  is always exactly one block per process: 200 extra commands before the
  exit do not change the result.
  
  The same record is present with upstream bash 5.3.0 built from the
  release tarball (trap plus top-level exit: 56 bytes, make_cmd.c:457 from
  parse.y:832). The Ubuntu 5.3 packages in later series were not measured.
  
  == Cause, from the 5.2.21 source ==
  
  builtins/exit.def ends in jump_to_top_level (EXITBLTIN). reader_loop in
  eval.c handles EXITBLTIN by setting current_command = NULL and leaving
  the loop. dispose_command is not called on that path, so the tree of the
  command that contained the exit is abandoned.
  
  == Impact ==
  
  None at run time. The block is lost at the moment the shell exits and
  the kernel reclaims it. Nothing accumulates in a running process.
  
  The only effect is on leak checking. A test harness that runs shell
  scripts under valgrind --trace-children=yes and gates on "definitely
  lost: 0" gets a record against /usr/bin/bash for almost every script,
  which hides real leaks in the programs under test.
  
- == Workaround ==
- 
- A valgrind suppression for Leak records whose frames are in
- /usr/bin/bash, or --trace-children-skip=*/bash.
- 
  == Other information ==
  
  There is no upstream fix to backport: none of the official bash 5.3 patches 
001 to 020 touches this path, and there is no release newer than 5.3. The same 
stack has been reported before without a fix:
   - bug-bash, May 2017, "memory leak in bash": 
https://lists.gnu.org/archive/html/bug-bash/2017-05/msg00098.html
   - Red Hat Bugzilla 1746101 (closed EOL): 
https://bugzilla.redhat.com/show_bug.cgi?id=1746101
   - bug-bash, June 2023, "Various small leaks": 
https://lists.gnu.org/archive/html/bug-bash/2023-06/msg00081.html
  
  LP: #1860133 is a different, startup-time record and is not a duplicate.
- 
- A possible resolution short of a code change is to ship a valgrind
- suppression file for bash.

** Description changed:

  Ubuntu release: 24.04 LTS (noble), seen on arm64 and amd64
  Package: bash 5.2.21-2ubuntu4
- 
- This is a low-severity report for the record. No stable release update
- is requested.
  
  == What happens ==
  
  When the exit builtin runs while a top-level command is still being executed, 
bash leaves without freeing the parsed command tree of that top-level command. 
valgrind memcheck reports the tree as "definitely lost". It happens when exit 
is called
   - from inside a function, an if, a loop, or a && / || list, or
   - at top level while an EXIT trap is set.
  
  It does not happen for a top-level exit with no EXIT trap, for a script
  that runs off its end, for a top-level exit after "trap - EXIT", or for
  bash -c 'exit 0'.
  
  == Steps to reproduce ==
  
   printf '#!/bin/bash\nf() { exit 0; }\nf\n' > /tmp/b.sh && chmod +x /tmp/b.sh
   valgrind --leak-check=full --show-leak-kinds=definite /tmp/b.sh
  
  Result (frames resolved with bash-dbgsym):
  
   72 (32 direct, 40 indirect) bytes in 1 blocks are definitely lost in loss 
record 320 of 344
     at malloc
     by make_simple_command (make_cmd.c:528)
     by yyparse (parse.y:782)
     by parse_command (eval.c:348)
     by read_command (eval.c:392)
     by reader_loop (eval.c:139)
     by main (shell.c:833)
  
  Expected: "definitely lost: 0 bytes in 0 blocks".
  
  == Variants measured ==
  
  bash 5.2.21-2ubuntu4, amd64, valgrind 3.22.0, each as a two- or three-
  line script file:
  
   exit 0 at top level, no trap: 0 bytes
   f() { exit 0; } then f: 72 bytes (32 direct, 40 indirect)
   if true; then exit 0; fi: 157 bytes (32 direct, 125 indirect)
   while true; do exit 0; done: 149 bytes (32 direct, 117 indirect)
   test 1 = 1 && exit 0: 259 bytes (32 direct, 227 indirect)
   trap "echo c" EXIT then exit 0 at top level: 56 bytes (32 direct, 24 
indirect)
   trap "echo c" EXIT then script ends: 0 bytes
   trap "echo c" EXIT, trap - EXIT, exit 0: 0 bytes
  
  The size follows the size of the top-level command that was running. It
  is always exactly one block per process: 200 extra commands before the
  exit do not change the result.
  
  The same record is present with upstream bash 5.3.0 built from the
  release tarball (trap plus top-level exit: 56 bytes, make_cmd.c:457 from
  parse.y:832). The Ubuntu 5.3 packages in later series were not measured.
  
  == Cause, from the 5.2.21 source ==
  
  builtins/exit.def ends in jump_to_top_level (EXITBLTIN). reader_loop in
  eval.c handles EXITBLTIN by setting current_command = NULL and leaving
  the loop. dispose_command is not called on that path, so the tree of the
  command that contained the exit is abandoned.
  
  == Impact ==
  
  None at run time. The block is lost at the moment the shell exits and
  the kernel reclaims it. Nothing accumulates in a running process.
  
  The only effect is on leak checking. A test harness that runs shell
  scripts under valgrind --trace-children=yes and gates on "definitely
  lost: 0" gets a record against /usr/bin/bash for almost every script,
  which hides real leaks in the programs under test.
  
  == Other information ==
  
  There is no upstream fix to backport: none of the official bash 5.3 patches 
001 to 020 touches this path, and there is no release newer than 5.3. The same 
stack has been reported before without a fix:
   - bug-bash, May 2017, "memory leak in bash": 
https://lists.gnu.org/archive/html/bug-bash/2017-05/msg00098.html
   - Red Hat Bugzilla 1746101 (closed EOL): 
https://bugzilla.redhat.com/show_bug.cgi?id=1746101
   - bug-bash, June 2023, "Various small leaks": 
https://lists.gnu.org/archive/html/bug-bash/2023-06/msg00081.html
  
  LP: #1860133 is a different, startup-time record and is not a duplicate.

** Description changed:

  Ubuntu release: 24.04 LTS (noble), seen on arm64 and amd64
  Package: bash 5.2.21-2ubuntu4
  
  == What happens ==
  
  When the exit builtin runs while a top-level command is still being executed, 
bash leaves without freeing the parsed command tree of that top-level command. 
valgrind memcheck reports the tree as "definitely lost". It happens when exit 
is called
   - from inside a function, an if, a loop, or a && / || list, or
   - at top level while an EXIT trap is set.
  
  It does not happen for a top-level exit with no EXIT trap, for a script
  that runs off its end, for a top-level exit after "trap - EXIT", or for
  bash -c 'exit 0'.
  
  == Steps to reproduce ==
  
   printf '#!/bin/bash\nf() { exit 0; }\nf\n' > /tmp/b.sh && chmod +x /tmp/b.sh
   valgrind --leak-check=full --show-leak-kinds=definite /tmp/b.sh
  
  Result (frames resolved with bash-dbgsym):
  
   72 (32 direct, 40 indirect) bytes in 1 blocks are definitely lost in loss 
record 320 of 344
     at malloc
     by make_simple_command (make_cmd.c:528)
     by yyparse (parse.y:782)
     by parse_command (eval.c:348)
     by read_command (eval.c:392)
     by reader_loop (eval.c:139)
     by main (shell.c:833)
  
  Expected: "definitely lost: 0 bytes in 0 blocks".
  
  == Variants measured ==
  
  bash 5.2.21-2ubuntu4, amd64, valgrind 3.22.0, each as a two- or three-
  line script file:
  
   exit 0 at top level, no trap: 0 bytes
   f() { exit 0; } then f: 72 bytes (32 direct, 40 indirect)
   if true; then exit 0; fi: 157 bytes (32 direct, 125 indirect)
   while true; do exit 0; done: 149 bytes (32 direct, 117 indirect)
   test 1 = 1 && exit 0: 259 bytes (32 direct, 227 indirect)
   trap "echo c" EXIT then exit 0 at top level: 56 bytes (32 direct, 24 
indirect)
   trap "echo c" EXIT then script ends: 0 bytes
   trap "echo c" EXIT, trap - EXIT, exit 0: 0 bytes
  
  The size follows the size of the top-level command that was running. It
  is always exactly one block per process: 200 extra commands before the
  exit do not change the result.
  
  The same record is present with upstream bash 5.3.0 built from the
  release tarball (trap plus top-level exit: 56 bytes, make_cmd.c:457 from
  parse.y:832). The Ubuntu 5.3 packages in later series were not measured.
  
  == Cause, from the 5.2.21 source ==
  
  builtins/exit.def ends in jump_to_top_level (EXITBLTIN). reader_loop in
  eval.c handles EXITBLTIN by setting current_command = NULL and leaving
  the loop. dispose_command is not called on that path, so the tree of the
  command that contained the exit is abandoned.
  
  == Impact ==
  
  None at run time. The block is lost at the moment the shell exits and
  the kernel reclaims it. Nothing accumulates in a running process.
  
  The only effect is on leak checking. A test harness that runs shell
  scripts under valgrind --trace-children=yes and gates on "definitely
  lost: 0" gets a record against /usr/bin/bash for almost every script,
  which hides real leaks in the programs under test.
  
- == Other information ==
+ == Upstream ==
  
  There is no upstream fix to backport: none of the official bash 5.3 patches 
001 to 020 touches this path, and there is no release newer than 5.3. The same 
stack has been reported before without a fix:
   - bug-bash, May 2017, "memory leak in bash": 
https://lists.gnu.org/archive/html/bug-bash/2017-05/msg00098.html
   - Red Hat Bugzilla 1746101 (closed EOL): 
https://bugzilla.redhat.com/show_bug.cgi?id=1746101
   - bug-bash, June 2023, "Various small leaks": 
https://lists.gnu.org/archive/html/bug-bash/2023-06/msg00081.html
  
  LP: #1860133 is a different, startup-time record and is not a duplicate.

-- 
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2174802

Title:
  bash does not free the command being executed when exit runs inside a
  function, loop or if (valgrind: definitely lost)

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/bash/+bug/2174802/+subscriptions


-- 
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs

Reply via email to