Change "read(2) will return 0" to "read(2) may return 0".
Quoting Jann Horn:
Maybe make that "may return 0" instead of "will return 0" -
reading from /proc/$pid/mem can only return 0 in the
following cases AFAICS:
1. task->mm was already gone at open() time
2. mm->mm_users has dropped to zero (the mm only has lazytlb
users; page tables and VMAs are being blown away or have
been blown away)
3. the syscall was called with length 0
When a process has gone away, normally mm->mm_users will
drop to zero, but someone else could theoretically still be
holding a reference to the mm (e.g. someone else in the
middle of accessing /proc/$pid/mem). (Such references
should normally not be very long-lived though.)
Additionally, in the unlikely case that the OOM killer just
chomped through the page tables of the target process, I
think the read will return -EIO (same error as if the
address was simply unmapped) if the address is within a
non-shared mapping. (Maybe that's something procfs could do
better...)
Reported-by: Jann Horn <jannh@google.com>
Signed-off-by: Michael Kerrisk <mtk.manpages@gmail.com>
Add some strongly worded text warning the reader about the correct
uses of seccomp user-space notification.
Reported-by: Jann Horn <jannh@google.com>
Cowritten-by: Christian Brauner <christian@brauner.io>
Cowritten-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Michael Kerrisk <mtk.manpages@gmail.com>
The APIs used by this mechanism comprise not only seccomp(2), but
also a number of ioctl(2) operations. And any useful example
demonstrating these APIs is will necessarily be rather long.
Trying to cram all of this into the seccomp(2) page would make
that page unmanageably long. Therefore, let's document this
mechanism in a separate page.
Signed-off-by: Michael Kerrisk <mtk.manpages@gmail.com>
The existing text says the structures (plural!) contain a 'struct
seccomp_data'. But this is only true for the received notification
structure (seccomp_notif). So, reword the sentence to be more
general, noting simply that the structures may evolve over time.
Add some comments to the structure definition.
Signed-off-by: Michael Kerrisk <mtk.manpages@gmail.com>
Rework the description a little, and note that the close-on-exec
flag is set for the returned file descriptor.
Signed-off-by: Michael Kerrisk <mtk.manpages@gmail.com>
I can't see a reason to include it. <fcntl.h> provides O_*
constants for 'flags', S_* constants for 'mode', and mode_t.
Probably a long time ago, some of those weren't defined in
<fcntl.h>, and both headers needed to be included, or maybe it's
a historical error.
Signed-off-by: Alejandro Colomar <alx.manpages@gmail.com>
Signed-off-by: Michael Kerrisk <mtk.manpages@gmail.com>
My previous patch intended to drop the docs for the lockdown lift
SysRq, but it missed this other section that refers to lifting it
via a keyboard - an allusion to that same SysRq.
Signed-off-by: dann frazier <dann.frazier@canonical.com>
Signed-off-by: Michael Kerrisk <mtk.manpages@gmail.com>
The patch that implemented lockdown lifting via SysRq ended up
getting dropped[*] before the feature was merged upstream. Having
the feature documented but unsupported has caused some confusion
for our users.
[*] http://archive.lwn.net:8080/linux-kernel/CACdnJuuxAM06TcnczOA6NwxhnmQUeqqm3Ma8btukZpuCS+dOqg@mail.gmail.com/
Signed-off-by: dann frazier <dann.frazier@canonical.com>
Cc: Heinrich Schuchardt <xypron.glpk@gmx.de>
Cc: David Howells <dhowells@redhat.com>
Cc: Pedro Principeza <pedro.principeza@canonical.com>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: Kyle McMartin <kyle@redhat.com>
Cc: Matthew Garrett <mjg59@google.com>
Signed-off-by: Alejandro Colomar <alx.manpages@gmail.com>
Signed-off-by: Michael Kerrisk <mtk.manpages@gmail.com>
Instead of having a monolithic 'make install', break it into
multiple targets such as 'make install-man3'. This simplifies
packaging, for example in Debian, where they break this project
into several packages: 'manpages' and 'manpages-dev', each
containing different mandirs.
The above allows for multithread installation: 'make -j'
Also, don't overwrite files that don't need to be overwritten, by
having a target for files, which makes use of make's timestamp
comparison.
This allows for much faster installation times.
For comparison, on my laptop (i7-8850H; 6C/12T):
Old Makefile:
~/src/linux/man-pages$ time sudo make >/dev/null
real 0m7.509s
user 0m5.269s
sys 0m2.614s
The times with the old makefile, varied a lot, between
5 and 10 seconds. The times after applying this patch
are much more consistent. BTW, I compared these times to
the very old Makefile of man-pages-5-09, and those were
around 3.5 s, so it was a bit of my fault to have such a
slow Makefile, when I changed the Makefile some weeks ago.
New Makefile (full clean install):
~/src/linux/man-pages$ time sudo make >/dev/null
real 0m5.160s
user 0m4.326s
sys 0m1.137s
~/src/linux/man-pages$ time sudo make -j2 >/dev/null
real 0m1.602s
user 0m2.529s
sys 0m0.289s
~/src/linux/man-pages$ time sudo make -j >/dev/null
real 0m1.398s
user 0m2.502s
sys 0m0.281s
Here we can see that 'make -j' drops times drastically,
compared to the old monolithic Makefile. Not only that,
but since when we are working with the man pages there
aren't many pages involved, times will be even better.
Here are some times with a single page changed (touched):
New Makefile (one page touched):
~/src/linux/man-pages$ touch man2/membarrier.2
~/src/linux/man-pages$ time sudo make install
- INSTALL /usr/local/share/man/man2/membarrier.2
real 0m0.988s
user 0m0.966s
sys 0m0.025s
~/src/linux/man-pages$ touch man2/membarrier.2
~/src/linux/man-pages$ time sudo make install -j
- INSTALL /usr/local/share/man/man2/membarrier.2
real 0m0.989s
user 0m0.943s
sys 0m0.049s
Also, modify the output of the make install and uninstall commands
so that a line is output for each file or directory that is
installed, similarly to the kernel's Makefile. This doesn't apply
to html targets, which haven't been changed in this commit.
Also, make sure that for each invocation of $(INSTALL_DIR), no
parents are created, (i.e., avoid `mkdir -p` behavior). The GNU
make manual states that it can create race conditions. Instead,
declare as a prerequisite for each directory its parent directory,
and let make resolve the order of creation.
Also, use ':=' instead of '=' to improve performance, by
evaluating each assignment only once.
Ensure than the shell is not called when not needed, by removing
all ";" and quotes in the commands.
See also: <https://stackoverflow.com/q/67862417/6872717>
Specify conventions and rationales used in the Makefile in a comment.
Add copyright.
Signed-off-by: Alejandro Colomar <alx.manpages@gmail.com>
Signed-off-by: Michael Kerrisk <mtk.manpages@gmail.com>