lp://staging/~vorlon/upstart/lp.855010

Branch merges

Propose for merging

No branches dependent on this one.

Ready for review for merging into lp://staging/upstart

James Hunt: Pending requested 2012-02-28

Diff: 71 lines (+12/-20)

3 files modified

ChangeLog (+6/-0)
init/job_process.c (+4/-18)
init/tests/test_job_process.c (+2/-2)

api

Related source package recipes

No recipes using this branch. (?)

Create packaging recipe

Related snap packages

Related bugs

Bug #855010: upstart assumes we can stop tracing processes when exec() is called, breaks 'expect daemon' for cups

Undecided

New

Link a bug report

Related blueprints

1358. By Steve Langasek on 2012-02-28

don't treat exec as a sign to stop tracing; in some unusual cases (i.e.,
cups), a process may actually re-exec itself before forking.

1357. By James Hunt on 2012-02-13

* init/log.c:
  - log_flush(): Comments.
  - log_read_watch():
    - Improved comments.
    - Don't re-attempt read on EAGAIN/EWOULDBLOCK since those errors are
      nominally impossible due to the remote end of the pty already
      having ended (LP: #926468).
    - Added a debug-mode warning when 'badly-behaved' application
      detected that leaks fds to its children, as this is generally a bug.
* init/tests/test_job_process.c:
  - test_run():
    - "with single line command writing fast and exiting": added
      waitpid() to ensure no EAGAIN.
  - test_spawn():
    - "with no such file, no shell and console log": added
      waitpid() to ensure no EAGAIN.

1356. By James Hunt on 2012-02-03

* init/job_process.c: job_process_spawn():
  - Set close-on-exec for pty_master.
  - Ensure stdio buffers flushed prior to forking to ensure no data
    leakage to child (should init be run with '--debug', or the tests
    be run with redirected output, for example).
  - Free log object if child process fails to exec(3) rather than just
    closing fd. This ensures io watch is removed correctly. (LP: #922754)
  - Only need to remap pty_master if CONSOLE_LOG in operation.
* init/tests/test_job_process:
  - fd_valid(): New helper function to determine if specified fd is valid.
  - child(): Added new 'TEST_FDS' test to ensure no fd leakage to child
    processes.
  - test_run():
    - New tests:
      - "ensure sane fds with no console, no script"
      - "ensure sane fds with no console, and script"
      - "ensure sane fds with console log, no script"
      - "ensure sane fds with console log, and script"
      - "with single-line command running an invalid command, then a 1-line post-stop script"
      - "with single-line command running an invalid command, then a 2-line post-stop script"
      - "with single-line command running an invalid command, then a post-stop command"
      - "with single-line command running an invalid command, then an invalid post-stop command"
      - "with single-line command running a valid command, then a 1-line invalid post-stop command"
    - Test "with single-line command running an invalid command":
      - now diverts stderr output for less chatty test-run experience.
      - Improved checking.
    - Test "with setuid me" now diverts stderr output for less chatty
      test-run experience.
  - test_spawn():
    - New tests:
      - "with no such file, no shell and console log"
      - "ensure sane fds with no console"
      - "ensure sane fds with console log"
* init/tests/test_log.c: test_log_destroy():
  - New test "ensure watch freed when log destroyed"

1355. By James Hunt on 2012-01-27

* init/tests/test_job_process.c: test_run(): Fixed typo in
          test "with setuid me" where uid value was being set to gid.

1354. By James Hunt on 2012-01-26

* init/job_process.c: job_process_terminated(): Free log to ensure data
  written as soon as _any_ process ends (consider respawn jobs).
* init/log.c:
  - log_destroy():
    - Improved documentation.
    - Now calls new function log_flush().
  - log_flush(): New function to ensure no lingering buffered job data
    remains. Now considers EBADF (LP: #912558).
  - log_io_reader():
    - Added missing assert for @len.
    - Simplified ENOSPC handling.
    - Ensure log->io set to NULL to allow other routines to detect it
      really has gone.
  - log_file_write(): Added @len checks.
  - log_read_watch(): New function to drain data from a watch descriptor
    (which also must consider EBADF).
* init/log.h: Added define for LOG_READ_SIZE.
* init/tests/test_job_process.c:
  - test_run():
    - Added some extra pointer checks.
    - Free class *before* checking file to ensure destructor invoked at
      correct point.
    - Added test "with single-line command running an invalid command"
      (for scenario bug 912558 exposed).
    - Added test "with single-line command writing fast and exiting".
* init/tests/test_log.c: Changed all tests to use openpty(3) rather than
  pipe(2) for semantic parity with actual code.
* util/tests/test_user_sessions.sh:
  - ensure_no_output(): Now calls check_job_output() and delete_job() to
    simplify logic.
  - delete_job(): Call get_job_file() rather than doing it long-hand.
  - check_job_output(): New function.
  - start_job(): Added allow_failure parameter.
  - test_ensure_no_unexpected_output(): New test
    "ensure command job does not create log file with invalid command".

1353. By James Hunt on 2012-01-05

* init/man/init.5: Explain that all job processes affected
  by 'setuid' and 'setgid' stanzas.

1352. By James Hunt on 2011-12-22

* init/tests/test_job_process.c: test_run():
  - Changed timeout for test feature "ensure that no log
  file written for CONSOLE_NONE".

1351. By James Hunt on 2011-12-22

init/job_process.c: job_process_spawn():
- Set child handler to default rather than explicit ignore
  to avoid test failures in environments that disallow
  ignoring SIGCHLD.

1350. By James Hunt on 2011-12-15

* init/tests/test_job_process.c: Add missing include for
  fnmatch.h.

1349. By James Hunt on 2011-12-15

* init/man/init.5: lower-case all references to system jobs
  and user jobs.