This fixes FS#15294.
The code to run a command inside a chroot was refactored from the
_alpm_runscriptlet function to _alpm_run_chroot.
Signed-off-by: Xavier Chantry <shiningxc@gmail.com>
The main purpose of this function to make our code more readable.
It frees transaction specific fields of pmpkg_t. (It is used when a package
is removed from the target list.)
Signed-off-by: Nagy Gabor <ngaba@bibl.u-szeged.hu>
Signed-off-by: Dan McGee <dan@archlinux.org>
This flag indicates that the front-end will not call alpm_trans_commit(),
so the database needn't be locked. This is the first step toward fixing
FS#8905.
If this flag is set, alpm_trans_commit() does nothing and returns with
an error (that has new code: PM_ERR_TRANS_NOT_LOCKED).
Signed-off-by: Nagy Gabor <ngaba@bibl.u-szeged.hu>
Signed-off-by: Dan McGee <dan@archlinux.org>
pmsyncpkg_t data sructure was removed:
1. pmpkg_t.reason is used instead of pmsyncpkg_t.newreason. (The target
packages come from sync repos, so we can use this field without any
problems. Upgrade transaction also uses this field to store this info.)
2. pmsyncpkg_t.removes was moved to pmpkg_t.removes.
This step requires careful programming, because we don't duplicate packages
when we add them to trans->packages. So we modify sync pkgcache when we
add this transaction-only info to our package. Hence it is important to
free this list when we remove any package from the target list
(remove_unresolvable, remove_conflicts, trans_free), otherwise this could
confuse the new sync transactions (with non-pacman GUI).
Overall, our code became ~100 line shorter, and we can call our helper
functions directly on trans->packages in sync.c, we don't need to maintain
parallel package lists.
Signed-off-by: Nagy Gabor <ngaba@bibl.u-szeged.hu>
Signed-off-by: Dan McGee <dan@archlinux.org>
This reverts commit 9558639d80.
This change was wrong, popen does require /bin/sh in a subchroot.
1) pacman -S lilo -r root
Notice no error
2) rm root/bin/sh ; pacman -S lilo -r root
Notice an error :
error: scriptlet failed to execute correctly
Actually, we already get an explicit error here, when popen is run, so there
is no need to check for bin/sh explicitely.
Besides this check was problematic in some cases. For example, bash itself
has a scriptlet, but only post_install and post_upgrade, no pre_install and
pre_upgrade. However, since bash has a scriptlet, runscriptlet will also be
called before bash is installed. It won't do anything since the scriptlet
has no pre_install function. But if we keep the check, we will still get
"error : no /bin/sh".
Conflicts:
lib/libalpm/trans.c
Signed-off-by: Xavier Chantry <shiningxc@gmail.com>
Signed-off-by: Dan McGee <dan@archlinux.org>
This is a bug I noticed 2 years ago :
http://www.nabble.com/Re%3A-logging-output-crazy-to11437357.html#a11479679
I thought I fixed it with 57d77eab32
But the bug was still here. Reading man fork, this part caught my attention:
* The child inherits copies of the parent's set of open file
descriptors. Each file descriptor in the child refers to the same open
file description (see open(2)) as the corresponding file descriptor in the
parent. This means that the two descriptors share open file status
flags, current file offset, and signal-driven I/O attributes (see the
description of F_SETOWN and F_SETSIG in fcntl(2)).
Since the open file descriptors are inherited, it is probably a good idea to
flush them before forking.
Signed-off-by: Xavier Chantry <shiningxc@gmail.com>
Signed-off-by: Dan McGee <dan@archlinux.org>
From now on -Qu is an "outdated package" filter on local database.
(This is a behaviour change.)
This patch fixes some memleaks and makes the code cleaner, for details see
my comment on FS#7884.
FS#11868 is implemented.
Signed-off-by: Nagy Gabor <ngaba@bibl.u-szeged.hu>
Signed-off-by: Dan McGee <dan@archlinux.org>
We never had a call to pclose() in here before, leaving our file descriptor
in some sort of limbo state. In addition, clean up some of the other logic
such as directly calling exit(1) on a popen() failure rather than going to
our cleanup block, and handling and respecting the exit status of the
subprocess correctly.
Signed-off-by: Dan McGee <dan@archlinux.org>
It is possible to throw EINTR from a system call such as open(), close(), or
waitpid() if custom signal handlers are set up and they are not initialized
with the SA_RESTART flag. This was noticed by Andreas Radke when ^C (SIGINT)
was given during the call to waitpid(), causing it to throw the EINTR error
and we could not accommodate it.
Simply wrap these calls in a simple loop that allows us to retry the call if
interrupted.
Signed-off-by: Dan McGee <dan@archlinux.org>
Commit 8240da6cb3 broke some alpm hierarchy
and introduced a new memleak (trans->packages was never freed in case of add
transaction, even if the transaction wasn't committed), so it is reverted
now.
We follow a different approach to reduce memory usage:
_alpm_db_add_pkgincache doesn't duplicate the whole package before adding
it to the cache, only the package name and version (INFRQ_BASE).
This method needs very small extra memory (compared to the reverted method),
and after transaction commit we use less memory than before (since the
big 'files' fields are not copied to cache), this is useful in GUIs.
Note: The old add_pkgincache was a bit broken, since pkg->origin wasn't
filled in correctly.
Signed-off-by: Nagy Gabor <ngaba@bibl.u-szeged.hu>
Acked-by: Xavier Chantry <shiningxc@gmail.com>
Signed-off-by: Dan McGee <dan@archlinux.org>
We were using the stat() system call in quite a few places when we didn't
actually need anything the stat struct returned- we were simply checking for
file existence. access() will be more efficient in those cases.
Before (strace pacman -Ss pacman):
% time seconds usecs/call calls errors syscall
------ ----------- ----------- --------- --------- ----------------
33.16 0.005987 0 19016 stat64
After:
% time seconds usecs/call calls errors syscall
------ ----------- ----------- --------- --------- ----------------
34.85 0.003863 0 12633 1 access
7.95 0.000881 0 6391 7 stat64
Signed-off-by: Dan McGee <dan@archlinux.org>
Its implementation was quite broken:
* add_loadtarget() might have silently filtered out some targets when
replacing an older version.
* This was used in sync.c to determine whether a target is implicit or not,
which is incorrect behavior. Before this patch we silently removed user
confirmed replacements; now we always warn on a replacement.
* remove001.py behavior was quite odd in adding same target 5 times to the
target list, we can change this behavior to be a failure.
Signed-off-by: Nagy Gabor <ngaba@bibl.u-szeged.hu>
[Xav: changed remove001 pactest accordingly]
Signed-off-by: Chantry Xavier <shiningxc@gmail.com>
[Dan: rewrote commit message]
Signed-off-by: Dan McGee <dan@archlinux.org>
Edit _alpm_db_add_pkgincache to not duplicate packages, because this is not
needed, is slower, and uses more memory. This made the max memory usage
during base reinstall go from 10.4MB to 9.7MB.
Signed-off-by: Chantry Xavier <shiningxc@gmail.com>
Signed-off-by: Dan McGee <dan@archlinux.org>
Fixes FS#7380: alpm crashes on passing NULL to alpm_trans_commit in
a sync operation. Adds check that data parameter is not NULL in
several functions.
Signed-off-by: Allan McRae <mcrae_allan@hotmail.com>
[Dan: fix whitespace]
Signed-off-by: Dan McGee <dan@archlinux.org>
I appears that when chrooted, the /bin/sh used by popen is that of the parent
process. This is true until the process forks once chrooted, which we do not
want to do.
As such, this actually makes things nicer. We don't need a /bin/sh in a chroot
to run install scriptlets, and don't need to check for it in the root directory
Signed-off-by: Aaron Griffin <aaronmgriffin@gmail.com>
Signed-off-by: Dan McGee <dan@archlinux.org>
The new pattern is as follows:
. /path/to/scriptlet
post_upgrade X Y
This requires less frameworking in the install scripts (the three lines that
shift and eval a function are nasty)
Signed-off-by: Aaron Griffin <aaronmgriffin@gmail.com>
Signed-off-by: Dan McGee <dan@archlinux.org>
Update the GPL boilerplate to direct people to the GNU website for a copy of
the license, as well as bump all of Judd's copyrights to 2007.
Signed-off-by: Dan McGee <dan@archlinux.org>
An alpm_list_free call was missing.
Also make use of alpm_list_free_inner in both _alpm_sync_free and
_alpm_trans_free.
Signed-off-by: Chantry Xavier <shiningxc@gmail.com>
Signed-off-by: Dan McGee <dan@archlinux.org>
The alpm_get_upgrades was exactly the same as find_replacements +
_alpm_sync_sysupgrade, except that it automatically made the eventual
replacements, without asking the user : Replace %s with %s/%s? [Y/n]
The replace question, asked in find_replacements. can now be skipped by
using a NULL trans argument, so that we get the same behavior as with
alpm_get_upgrades.
So alpm_db_get_upgrades() can now be replaced by
alpm_sync_sysupgrade(db_local, syncdbs).
Signed-off-by: Chantry Xavier <shiningxc@gmail.com>
Signed-off-by: Dan McGee <dan@archlinux.org>
dash doesn't pass positional parameters to sourced scripts, causing install
scripts to fail. Instead of sourcing the script, make it executable and
call it directly which allows positional parameters to be passed correctly.
Signed-off-by: Dan McGee <dan@archlinux.org>
This file only contained one private function : _alpm_db_whatprovides .
And the public alpm_db_whatprovides was in db.c , so I moved everything there.
Signed-off-by: Chantry Xavier <shiningxc@gmail.com>
[Dan: updated POTFILES.in as well]
Signed-off-by: Dan McGee <dan@archlinux.org>
Instead of using the often-busted REQUIREDBY entries in the pacman database,
compute them each time they are required. This should help many things:
1. Simplify the codebase
2. Prevent future database corruption
3. Ensure when we do use requiredby, it is always correct
4. Shrink the pmpkg_t memory overhead
Signed-off-by: Dan McGee <dan@archlinux.org>
I broke scriptlet logging with ad691001e2.
Readd more or less what was there before, although it still needs a lot of
work including hopefully rewriting it to a new event subsystem and having
it log to a seperate file.
Signed-off-by: Dan McGee <dan@archlinux.org>
These macros take the place of the common 4 or 5 line blocks of code we had
in most places that called malloc or calloc. This should reduce some code
duplication and make memory allocation more standard in libalpm.
Highlights:
* Note that the MALLOC macro actually uses calloc, this is just for safety
so that memory is initialized to 0. This can be easily changed in one
place.
* One malloc call was completely eliminated- it made more sense to do it
on the stack.
* The use of RET_ERR in public functions (mainly the alpm_*_new functions)
was standardized, this makes sense so pm_errno is set.
Signed-off-by: Dan McGee <dan@archlinux.org>
Commit 4853a4aad9 used the tmpdir variable
for checking the existence of /bin/sh, without resetting it.
This caused /bin/sh to be deleted during the cleanup part, as soon as a scriptlet
other than pre_upgrade or pre_install was executed.
For example, on the first post_upgrade during a -Su.
I introduced two variables : clean_tmpdir and restore_cwd, for deciding what should
be done in the cleanup part.
Signed-off-by: Chantry Xavier <shiningxc@gmail.com>
This code assumed that DBPath was under RootDir, while this is not necessarily the case :
pacman doesn't enforce anymore than DBPath is under RootDir.
So now, all scriptlets will be put somewhere in RootDir/tmp/, so that when it chroots in RootDir,
the scriptlets are still available inside the chroot.
This also removes the need of normalizing both dbpath and rootdir, in order to do computation on the paths.
Signed-off-by: Chantry Xavier <shiningxc@gmail.com>
If /bin/sh is missing in the root directory, scriptlets cannot be executed, as
we're explicitly calling it.
Signed-off-by: Aaron Griffin <aaronmgriffin@gmail.com>
My two previous hacks related to this part
(8038190c7c and
b15a5194d1) were caused by the lack of
understanding of a feature introduced a while ago:
Better control over CTRL-C interruptions -- do not leave the DB in an
inconsistent state (54008798ef).
Now I have been looking at this commit, and the added feature is indeed
interesting. The main problem I had with it is that it does a rather
unusual use of alpm_trans_release, which caused a few problems that I tried
to fix in a weird way. I think these problems were caused by the fact that
there weren't any difference between "interrupt transaction" and "release a
transaction which failed" actions from the alpm_trans_release POV. So I
decided to add a new function instead, alpm_trans_interrupt, which is
called on Ctrl+C, and which only sets trans->state to STATE_INTERRUPTED so
that remove_commit and add_commit can exit cleanly at a safe moment. This
allowed me to revert my two previous hacks as well.
Also ensure we handle SIGINT correctly in all cases- if a transaction is
not ongoing, then we can free the transaction and exit quickly.
Signed-off-by: Chantry Xavier <shiningxc@gmail.com>
Signed-off-by: Dan McGee <dan@archlinux.org>
This allows us to remove the hack in the frontend where we added a newline
to everything coming out of the pm_printf functions, and instead let the
developer put newlines where they want them. This should be the last hangover
of that auto-newline stuff.
Signed-off-by: Dan McGee <dan@archlinux.org>
A bunch of changes related to my first "real" install of pacman-git into
/usr/local and trying to use it.
* Shift some uses of free -> FREE in libalpm.
* Move stat and sanity checks of config paths into libalpm from the
config and argument parsing in pacman.c.
* Fix issue where dbpath still was not defined early enough due to its
requirement for being used in alpm_db_register. This should be rewritten
so it doesn't have this dependency, but this will work for now.
Signed-off-by: Dan McGee <dan@archlinux.org>
This patch cleans up and fix the _alpm_trans_update_depends function
and ensure that all requiredby fields are filled in case of multiple satisfiers
(previously, the handling of mutliple satisfiers in that function was inconsistent).
This makes a special case handling of requiredby in commit_single_pkg() obsolete,
and so allows cleaning that code as well.
Also fixed upgrade056 pactest because :
1) the requiredby fields were wrong, and this wouldn't happen with the fixed _alpm_trans_update_depends().
2) this is a very unusual case anyway (and handling all corner cases combined to a broken database
seems nearly impossible to achieve).
References :
http://www.archlinux.org/pipermail/pacman-dev/2007-July/008919.htmlhttp://www.archlinux.org/pipermail/pacman-dev/2007-July/008920.html
Signed-off-by: Chantry Xavier <shiningxc@gmail.com>
This fixes the failure of the requiredby004 pactest in a not so pretty way,
but it gets the job done. I purposely used the extremely long name of
PM_TRANS_TYPE_REMOVEUPGRADE to be both clear and in the hope that someone
else will figure out a better solution.
Original idea from Nagy Gabor, patch updated and cleaned for current code.
Signed-off-by: Dan McGee <dan@archlinux.org>
There is no real reason to burden our translators with these messages, as
anyone helping to debug these will probably want them in English.
Signed-off-by: Dan McGee <dan@archlinux.org>