pkg_alternatives: pass if the desired symlink already exists

This can happen when opkg installs, then configures multiple
alternatives in a batch.  Symlink to the highest prio alternative will
be created in the initial configuring stage causing later symlink call
fail with EEXIST

    Configuring busybox.
    ...
    symlink("/sbin/ip-full", "/home/yousong/j/t/lede-imagebuilder-mvebu.Linux-x86_64/build_dir/target-arm_cortex-a9+vfpv3_musl_eabi/root-mvebu/sbin/ip") = 0
    ...
    Configuring ip-full.
    ...
    symlink("/sbin/ip-full", "/home/yousong/j/t/lede-imagebuilder-mvebu.Linux-x86_64/build_dir/target-arm_cortex-a9+vfpv3_musl_eabi/root-mvebu/sbin/ip") = -1 EEXIST (File exists)
    ...

While at it, "mkdir -p" dirname(path_in_dest) before symlink in case the
following symlink call may fail with ENOENT

Ref: https://github.com/openwrt/packages/issues/4567
Reported-by: Aner Andros <aa@anerandros.info>
Signed-off-by: Yousong Zhou <yszhou4tech@gmail.com>

opkg: add --no-check-certificate argument

For cases when artifacts are stored on https:// accessible
locations and you don't want to install ca-certificates
(for various reasons).

I'll admit, using SSL like this is not recommended,
but since wget (even uclient-fetch) allows the
--no-check-certificate option, it would be nice
for opkg to support setting it if needed/configured.

Signed-off-by: Alexandru Ardelean <ardeleanalex@gmail.com>

pkg_alternatives: use ERROR level for symlink failure

Signed-off-by: Yousong Zhou <yszhou4tech@gmail.com>

pkg: alternatives support

It's a list of specs of the following form seprated by commas to describe
alternatives provided by the package

    <prio>:<path>:<altpath>

where <path> will be a symbolic link to <altpath> of the highest <prio>

Size comparison on x86_64 after the change

    function                                             old     new   delta
    pkg_alternatives_update                                -     587    +587
    pkg_parse_line                                      2101    2609    +522
    .rodata                                            24594   24738    +144
    pkg_formatted_field                                 2385    2528    +143
    pkg_deinit                                           427     486     +59
    pkg_print_status                                     264     280     +16
    opkg_configure                                        59      69     +10
    xreadlink                                            120     128      +8
    opkg_remove_pkg                                     1079    1087      +8
    ------------------------------------------------------------------------------
    (add/remove: 2/0 grow/shrink: 8/0 up/down: 1483/0)           Total: 1497 bytes

Signed-off-by: Yousong Zhou <yszhou4tech@gmail.com>

libbb: xreadlink: fix memory leak on failure case

Signed-off-by: Yousong Zhou <yszhou4tech@gmail.com>

pkg_run_script: use pkg->dest in half installed case

Fixes running scripts for packages in half-installed state which would
otherwise fail with:

 * pkg_run_script: Internal error: ... has a NULL tmp_unpack_dir.
 * opkg_remove_pkg: not removing package "...", prerm script failed

Upstream bug: https://bugzilla.yoctoproject.org/show_bug.cgi?id=11245

Signed-off-by: Alejandro del Castillo <alejandro.delcastillo@ni.com>
[Jo-Philipp Wich: rebased onto opkg-lede, updated commit message]
Signed-off-by: Jo-Philipp Wich <jo@mein.io>

check_data_file_clashes_change: remove duplicated offline root in paths

The function check_data_file_clashes_change update the ownership of
files that belong to a package being replaced. It calls
pkg_get_installed_files then adds an offline root if
opkg_conf->offline_root to each path. However, pkg_get_installed_files
already returns paths with the offline root (if present).

This fixes issue 156.

Signed-off-by: Alejandro del Castillo <alejandro.delcastillo@ni.com>

buildReplaces: do not add duplicated replacees

If package A both replaces and conflicts package B, it is added
unconditionally to the replaced_by vector of package B. Add check to
only add it if it's not there to avoid inaccurate warnings of the type:

Multiple replacers for X, using first one

Signed-off-by: Alejandro del Castillo <alejandro.delcastillo@ni.com>

opkg: compare-versions doesn't need any state

compare-versions doesn't need lock files created and databases opened, so don't.

Signed-off-by: Ross Burton <ross.burton@intel.com>

opkg_cmd: fix segmentation fault in opkg_compare_versions_cmd()

Due to the fact that we're using blob buffers internally now we cannot simpyl
call parse_version() on an uninitialized struct.

Properly create the temporary packages using pkg_new() and deallocated them
using pkg_deinit() and free().

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

opkg_cmd: fix return of opkg compare-versions

The implementation of opkg compare-version was simply returning the return value
of pkg_version_satisfied(), which returns a boolean.  This results in inverted
logic in that:

$ opkg compare-version 1.0 == 1.0 || echo different
different

That's clearly not right.  As nobody had noticed this before presumably nobody
was actually using this (likely either opkg-compare-version or
opkg-compare-versions.sh) just change the return value so it's correct.

Signed-off-by: Ross Burton <ross.burton@intel.com>

pkg_src_list_push: remove unused function

Signed-off-by: Alejandro del Castillo <alejandro.delcastillo@ni.com>

pkg_src_list_pop: remove unused function

Signed-off-by: Alejandro del Castillo <alejandro.delcastillo@ni.com>

str_list_push: remove unused function

Signed-off-by: Alejandro del Castillo <alejandro.delcastillo@ni.com>

str_list_prev: remove unused function

Signed-off-by: Alejandro del Castillo <alejandro.delcastillo@ni.com>

str_list_last: remove unused function

Signed-off-by: Alejandro del Castillo <alejandro.delcastillo@ni.com>

conffile_list_pop: remove unused function

Signed-off-by: Alejandro del Castillo <alejandro.delcastillo@ni.com>

pkg_dest_list_push: remove unused functions

Signed-off-by: Alejandro del Castillo <alejandro.delcastillo@ni.com>

nv_pair_list_prev: remove unused function

Signed-off-by: Alejandro del Castillo <alejandro.delcastillo@ni.com>

nv_pair_list_last: remove unused function

Signed-off-by: Alejandro del Castillo <alejandro.delcastillo@ni.com>

conffile_list_pop: remove unused function

Signed-off-by: Alejandro del Castillo <alejandro.delcastillo@ni.com>

conffile_list_push: remove unused function

Signed-off-by: Alejandro del Castillo <alejandro.delcastillo@ni.com>

active_list_sort: remove unused function

Signed-off-by: Alejandro del Castillo <alejandro.delcastillo@ni.com>
[Jo-Philipp Wich: remove call from opkg_active_list_test]
Signed-off-by: Jo-Philipp Wich <jo@mein.io>

active_list_add_depend: remove unused function

Signed-off-by: Alejandro del Castillo <alejandro.delcastillo@ni.com>
[Jo-Philipp Wich: remove call from opkg_active_list_test]
Signed-off-by: Jo-Philipp Wich <jo@mein.io>

pkg_dependence_satisfied: remove unused function

Signed-off-by: Alejandro del Castillo <alejandro.delcastillo@ni.com>

Fix typo in pkg_hash_fetch_best_installation_candidate.

A (presumed) typo in pkg_hash_fetch_best_installation_candidate caused the
matching_pkgs vec to be sorted twice, instead of sorting matching_apkgs.
This fixes both the call and the API to match what appears to be intended.

Signed-off-by: Michael Hansen <michael.j.hansen@hp.com>
Signed-off-by: Alejandro del Castillo <alejandro.delcastillo@ni.com>

set_flags_from_control: remove function

During installs, install_data_files calls set_flags_from_control to
capture the case where fields, such as "Essential", are present in the
package control file but are missing on the Packages file. This
operation is expensive (re-parses the control file) and is catering to a
case that should not be supported on the first place.

Signed-off-by: Alejandro del Castillo <alejandro.delcastillo@ni.com>
[Jo-Philipp Wich: remove function prototype as well]
Signed-off-by: Jo-Philipp Wich <jo@mein.io>

opkg_active_list_test: mark functions static, swap strcmp() with memcmp()

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg_test: mark functions static, remove unused opkg_test()

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

pkg_parse: fix bad message level for opkg_msg() call

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

pkg_parse: remove unused variable in pkg_parse_line()

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

pkg_parse: remove const requirement from pkg_parse_line()

To simplify parsing we'd like to be able to modify the passed line buffer
inplace, so remove the const requirement from the pkg_parse_line() argument.

Convert pkg_parse_line's sole user accordingly.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

pkg_hash: remove unused variable in pkg_hash_load_feeds()

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

pkg_depends: add missing parse_replacelist() prototype

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

opkg_utils: add missing header

Include "opkg_utils.h" header file to declare function prototypes.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: remove unused progress parameters from opkg_download()

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: remove unused progress callback

The current opkg_download() implementation completely ignores the passed
progress callback function, so drop this unused code.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: remove unused variable from opkg_install_package()

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libbb: remove unused variables

Remove unused variables from deb_extract() to avoid compiler warnings.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libbb: change gzip_read() buffer argument to void *

Make gzip_read() expect a void * buffer argument to avoid signed vs unsigned
char mismatch warnings.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libbb: fix gzip_thread() warnings

 - mark function static
 - add missing return statement
 - avoid uninitialized use of "len"

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg/opkg_cmd.c: more robust PATH handling

preserve semantics of PATH when PATH is not set in environment

error and undefined behavior reported in
  https://github.com/openwrt/packages/issues/1922
when PATH not set in environment
(lighttpd executes CGI with empty base env, plus standard CGI env vars)

Signed-off-by: Glenn Strauss <gstrauss@gluelogic.com>
[Jo-Philipp Wich: avoid free() on NULL, use default from cmake cache string]
Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: specify "/bin/sh" instead of "sh"

avoid strange behavior with execvp() when PATH is not set in environment
(in which case confstr(_CS_PATH) should return something reasonable)

reproducable running openwrt 15.05 and 15.05.1 and attempting to install
a software package (e.g. libuuid) via LuCI (prior to openwrt/luci#1048).
(https://github.com/openwrt/luci/pull/1048) libuuid.postinst fails with
status 255 on 15.05 and opkg segfaults in 15.05.1.  This probably merits
further exploration.

Originally reported in https://github.com/openwrt/packages/issues/1922

Signed-off-by: Glenn Strauss <gstrauss@gluelogic.com>

unarchive: fix possible segmentation fault in deb_extract()

When a not existing or unreachable file path is passed to deb_extract(),
the wfopen() call fails, causing a jump to the cleanup: label which leads
to a call to gzip_close() on the tar_outer structure.

The tar_outer structure however contains uninitialized memory at this point,
causing gzip_close() to operate on garbage data. Depending on the nature of
the unitialized memory, this might lead to all sorts of issues, e.g. freeing
of not allocated memory or invoking fclose() on garbage pointers.

Solve this problem by initializing the tar_outer and tar_inner structures
right at the declaration.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

pkg: forward "provided_by_hand" flag in pkg_merge()

Due to the fact that we now load status files after preparing the abstract
structures describing the cli specified packages, we need to ensure that the
"provided_by_hand" flag is properly conveyed in pkg_merge(), otherwise the
cli provided package archive might get ignored in favor to a newer version
from feeds.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

pkg: mark bit fields in pkg_t unsigned

Qualify the bit field type as unsigned int to ensure that set flags are
interpreted as "1" and not "-1".

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

opkg_cmd: avoid null pointer dereference in opkg_what_depends_conflicts_cmd()

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

opkg_cmd: fix endless loop in whatprovides/whatreplaces

Fix the loop semantics in opkg_what_provides_replaces_cmd() to properly
advance the package vector pointer.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

pkg_depends: propagate virtual provided detail flag to providers

When the provided abstract package (e.g. "px5g" provided by "px5g-mbedtls") is
marked as requiring details, then mark the provider abstract package as well.

This ensures that package info for discrete provider packages is acquired and
retained if the virtual package they provide is being depended upon by another
package requiring details.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

pkg_depends: fix unsatisfied dependency installation order

Unsatisfied dependencies are not being installed in the correct order. The
algorithm is not crawling down the dependency chain first when inserting
unsatisfied dependencies, resulting in a correct installation order only for
the first layer of dependencies.

This patch changes the unsatisfied dependency insertion order to first add leaf
dependencies, then move up the chain. The result is a list of unsatisfied
dependencies ordered most-dependent-first.

An example that resulted in the incorrect installation order was:
  A -> B
  A -> C
  B -> D

Without the fix, a most-dependent-first installation order was not guaranteed
more than one layer deep, resulting in an installation order where B is
incorrectly installed before D:
  B, D, C, A

After the fix, the installation order follows most-dependent first irrespective
of the number of layers:
  D, B, C, A

Signed-off-by: Pieter Smith <pieter.smith@philips.com>
[Jo-Philipp Wich: rebased onto opkg-lede.git]
Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: fix some overly aggressive line wrapping in pkg_depends.c

Then deeply nested blocks in pkg_depends.c tripped up Lindent, causing it
to perform overly aggressive line wrapping.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: remove stray printf()

This lone printf() statement was accidentally left in the code during some
previous refactoring session.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: switch to GPL2 SHA256 implementation

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: initialize conffile list during status parsing

When parsing conffile information from status files, we need to initialize
the conffile list head upon encountering a "Conffiles:" line, otherwise
parse_conffiles() will ignore all subsequent entries due to a missing list
pointer.

Fixes missing conffile information in /usr/lib/opkg/status on a fresh
installation.

Reported-by: Hartmut Birr <e9hack@gmail.com>
Signed-off-by: Jo-Philipp Wich <jo@mein.io>

build: enable usign support

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

build: allow to pass version via CMake

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: remove OpenSSL support

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: remove Curl support

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: remove GPG support

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: remove x509 support

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: drop custom md5 implementation, unconditionally enable sha256 support

Drop custom md5 implementation and switch to the one provided by libubox,
also remove support for enabling/disabling checksum support and always build
it unconditionally.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: fix pkg_set_int(), pkg_get_int()

Fix the return values of pkg_get_int() and pkg_set_int() to properly
convert the pointers from/to integer values.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: properly fix parse_version()

A previous attempt to fix the verison/revision split caused unterminated
strings to get stored in the blob buffer, properly fix this case now.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

build: fix handling of string options

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: do not modify original version string in parse_version()

The parse_version() function might be parsing the version string of another
pkg_t instance so we must not modify it inplace to truncate the revision
part, otherwise we will clobber the original version, leading to failing
comparisations later on.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

build: add install target for opkg-cl

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

build: enable SHA256 support

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

opkg-cl: remove config.h reference

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: remove config.h references

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

build: infer VERSION from Git history

Pass short hash and commit date as VERSION string.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

build: do not unconditionally link libpthread

Some environments, e.g. musl libc based ones, do not have a dedicated
libpthread, so only link it if it exists.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

opkg-cl: add missing include to import FNM_CASEFOLD

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

build: expose LOCK_FILE option

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

build: remove outdated files

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

build: remove automake/autoconf build system

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

.gitignore: ignore test executables

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

.gitignore: ignore CMake artifacts

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

tests: fix build of opkg_extract_test

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

build: add cmake build system

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

opkg-cl: stop preloading package feeds on install

Instead of unconditionally loading all package feed lists entirely into
memory, let opkg_install_cmd() selectively parse the lists using the new
pkg_hash_load_package_details() helper.

This will result in a slightly longer runtime but dramatically reduces the
memory requirements. On an x86_64 test system, the peak heap memory
consumption during an "opkg install luci" dropped from 3.7 MB to about
360 KB, a decrease of roughly 90%.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: selectively load package feeds in opkg_info_status_cmd()

Use the new pkg_hash_load_package_details() helper to selectively load the
dependency subgraphs of the to be installed packages.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: immediately free packages not marked for detail parsing

Stop recording packages into the dependency graph which are not flagged
with SF_NEED_DETAILS.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: avoid aggregating multiple identical providers

Since we might do multiple parsing runs over the feed list files in the
future, ensure that we do not insert the abstract provider package multiple
times into the provided_by vector.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: implement pkg_hash_load_package_details() helper

The purpose of this new function is to keep parsing feed lists until
all package detail requests are satisified.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: support passing default state flags in pkg_hash_load_feeds()

Add a new argument to pkg_hash_load_feeds() which allows specifying default
state flags that should be applied to all loaded packages.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: flag package with SF_NEED_DETAIL in pkg_init_from_file()

Flag the package loaded from file as SF_NEED_DETAIL to properly fetch and
resolve dependencies.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: fix printing provides in pkg_formatted_field()

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: fix pkg_deinit() to properly free the resources in the blob buffer

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: free old provides list in pkg_merge()

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: fix merging PKG_CONFFILES in pkg_merge()

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: fix pkg_set_ptr() to properly set NULL pointers

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: allow passing common package state flags to pkg_hash_add_from_file()

Add a new parameter to pkg_hash_add_from_file() for passing package state
flags that shall be applied to all pkg_t instances parsed from the given
file and use it to mark all packages loaded from local status files with
the SF_NEED_DETAIL flag.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: propagate SF_NEED_DETAIL from abstract to discrete package

During parsing, propagate the state of the SF_NEED_DETAIL flag from the
related abstract package to the discrete pkg_t instance getting assembled.

This is required to ensure that subsequent parsing of pkg_t's dependency
lists considers the state of SF_NEED_DETAIL.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: do not clobber package flags during parsing

When parsing "Status:" lines from package control or list files, do not
overwrite previously set flags in the package structure.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: propagate SF_NEED_DETAIL to related packages

When parsing Provides, Suggests, Conflicts, Replaces and Depends lists,
propagate the parent package SF_NEED_DETAIL flag to related abstract
packages in order to ensure that the entire related dependency graph
is processed.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: directly pass type to parseDepends()

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: expose abstract_pkg_fetch_by_name()

We need this function in other places later in order to be able to fetch
abstract packages without implicitely creating them.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: mark to-be-installed non-local, non-url packages for detail parsing

When just a package name is passed to opkg install, we need to retrieve
information about this package from the feeds list so stage an abstract
package with the given name and flag it with SF_NEED_DETAIL.

Packages specified as local file paths or direct download URLs do not
need this special treatment as the required details can be extracted
from their embedded control files.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: introduce SF_NEED_DETAIL flag

Introduce a new pkt_t state_flag value SF_NEED_DETAIL which will be used
to mark packages for which we require details.

When parsing the feed lists, we'll only keep information about packages
marked with SF_NEED_DETAIL and packages related to them through Provides,
Suggests, Conflicts, Depends etc.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

libopkg: make checksum_bin2hex() and checksum_hex2bin() cope with NULL

Those functions may be called with a NULL source pointer, especially during
pkg_merge(), so allow that to keep call sites simple.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>