Willy Tarreau [Wed, 3 Jun 2020 18:02:28 +0000 (20:02 +0200)]
REORG: include: move sink.h to haproxy/sink{,-t}.h
The sink files could be moved with almost no change at since they
didn't rely on anything fancy. ssize_t required sys/types.h and
thread.h was needed for the locks.
Willy Tarreau [Wed, 3 Jun 2020 17:33:00 +0000 (19:33 +0200)]
REORG: include: move fd.h to haproxy/fd{,-t}.h
A few includes were missing in each file. A definition of
struct polled_mask was moved to fd-t.h. The MAX_POLLERS macro was
moved to defaults.h
Stdio used to be silently inherited from whatever path but it's needed
for list_pollers() which takes a FILE* and which can thus not be
forward-declared.
Willy Tarreau [Wed, 3 Jun 2020 16:09:46 +0000 (18:09 +0200)]
REORG: tools: split common/standard.h into haproxy/tools{,-t}.h
And also rename standard.c to tools.c. The original split between
tools.h and standard.h dates from version 1.3-dev and was mostly an
accident. This patch moves the files back to what they were expected
to be, and takes care of not changing anything else. However this
time tools.h was split between functions and types, because it contains
a small number of commonly used macros and structures (e.g. name_desc)
which in turn cause the massive list of includes of tools.h to conflict
with the callers.
They remain the ugliest files of the whole project and definitely need
to be cleaned and split apart. A few types are defined there only for
functions provided there, and some parts are even OS-specific and should
move somewhere else, such as the symbol resolution code.
Willy Tarreau [Wed, 3 Jun 2020 13:26:55 +0000 (15:26 +0200)]
REORG: include: move protocol.h to haproxy/protocol{,-t}.h
The protocol.h files are pretty low in the dependency and (sadly) used
by some files from common/. Almost nothing was changed except lifting a
few comments.
Willy Tarreau [Wed, 3 Jun 2020 12:56:08 +0000 (14:56 +0200)]
REORG: include: move common/fcgi.h to haproxy/
The file was moved almost verbatim (only stdio.h was dropped as useless).
It was not split between types and functions because it's only included
from direct C code (fcgi.c and mux_fcgi.c) as well as fcgi_app.h, included
from the same ones, which should also be remerged as a single one.
Willy Tarreau [Wed, 3 Jun 2020 07:09:57 +0000 (09:09 +0200)]
REORG: include: move hpack*.h to haproxy/ and split hpack-tbl
The various hpack files are self-contained, but hpack-tbl was one of
those showing difficulties when pools were added because that began
to add quite some dependencies. Now when built in standalone mode,
it still uses the bare minimum pool definitions and doesn't require
to know the prototypes anymore when only the structures are needed.
Thus the files were moved verbatim except for hpack-tbl which was
split between types and prototypes.
Willy Tarreau [Wed, 3 Jun 2020 06:44:35 +0000 (08:44 +0200)]
REORG: include: split common/htx.h into haproxy/htx{,-t}.h
Most of the file was a large set of HTX elements manipulation functions
and few types, so splitting them allowed to further reduce dependencies
and shrink the build time. Doing so revealed that a few files (h2.c,
mux_pt.c) needed haproxy/buf.h and were previously getting it through
htx.h. They were fixed.
Willy Tarreau [Tue, 2 Jun 2020 17:33:08 +0000 (19:33 +0200)]
REORG: include: move common/h1.h to haproxy/h1.h
The file was moved as-is. There was a wrong dependency on dynbuf.h
instead of buf.h which was addressed. There was no benefit to
splitting this between types and functions.
Willy Tarreau [Tue, 2 Jun 2020 17:25:28 +0000 (19:25 +0200)]
REORG: include: split common/http-hdr.h into haproxy/http-hdr{,-t}.h
There's only one struct and 2 inline functions. It could have been
merged into http.h but that would have added a massive dependency on
the hpack parts for nothing, so better keep it this way since hpack
is already freestanding and portable.
Willy Tarreau [Tue, 2 Jun 2020 17:11:26 +0000 (19:11 +0200)]
REORG: include: split common/http.h into haproxy/http{,-t}.h
So the enums and structs were placed into http-t.h and the functions
into http.h. This revealed that several files were dependeng on http.h
but not including it, as it was silently inherited via other files.
Willy Tarreau [Tue, 2 Jun 2020 15:46:16 +0000 (17:46 +0200)]
REORG: include: split common/xref.h into haproxy/xref{,-t}.h
The type is the only element needed by applet.h and hlua.h, while hlua.c
needs the various functions. XREF_BUSY was placed into the types as well
since it's better to have the special values there.
Willy Tarreau [Tue, 2 Jun 2020 15:32:26 +0000 (17:32 +0200)]
REORG: include: split common/regex.h into haproxy/regex{,-t}.h
Regex are essentially included for myregex_t but it turns out that
several of the C files didn't include it directly, relying on the
one included by their own .h. This has been cleanly addressed so
that only the type is included by H files which need it, and adding
the missing includes for the other ones.
Willy Tarreau [Tue, 2 Jun 2020 15:02:59 +0000 (17:02 +0200)]
REORG: include: move common/namespace.h to haproxy/namespace{,-t}.h
The type was moved out as it's used by standard.h for netns_entry.
Instead of just being a forward declaration when not used, it's an
empty struct, which makes gdb happier (the resulting stripped executable
is the same).
Willy Tarreau [Tue, 2 Jun 2020 09:28:02 +0000 (11:28 +0200)]
REORG: include: move common/buffer.h to haproxy/dynbuf{,-t}.h
The pretty confusing "buffer.h" was in fact not the place to look for
the definition of "struct buffer" but the one responsible for dynamic
buffer allocation. As such it defines the struct buffer_wait and the
few functions to allocate a buffer or wait for one.
This patch moves it renaming it to dynbuf.h. The type definition was
moved to its own file since it's included in a number of other structs.
Doing this cleanup revealed that a significant number of files used to
rely on this one to inherit struct buffer through it but didn't need
anything from this file at all.
Willy Tarreau [Tue, 2 Jun 2020 07:38:52 +0000 (09:38 +0200)]
REORG: include: move common/memory.h to haproxy/pool.h
Now the file is ready to be stored into its final destination. A few
minor reorderings were performed to keep the file properly organized,
making the various sections more visible (cache & lockless).
In addition and to stay consistent, memory.c was renamed to pool.c.
Willy Tarreau [Mon, 1 Jun 2020 17:00:28 +0000 (19:00 +0200)]
MEDIUM: memory: make local pools independent on lockless pools
Till now the local pool caches were implemented only when lockless pools
were in use. This was mainly due to the difficulties to disentangle the
code parts. However the locked pools would further benefit from the local
cache, and having this would reduce the variants in the code.
This patch does just this. It adds a new debug macro DEBUG_NO_LOCAL_POOLS
to forcefully disable local pool caches, and makes sure that the high
level functions are now strictly the same between locked and lockless
(pool_alloc(), pool_alloc_dirty(), pool_free(), pool_get_first()). The
pool index calculation was moved inside the CONFIG_HAP_LOCAL_POOLS guards.
This allowed to move them out of the giant #ifdef and to significantly
reduce the code duplication.
A quick perf test shows that with locked pools the performance increases
by roughly 10% on 8 threads and gets closer to the lockless one.
Willy Tarreau [Mon, 1 Jun 2020 16:35:24 +0000 (18:35 +0200)]
MINOR: memory: move pool-specific path of the locked pool_free() to __pool_free()
pool_free() was not identical between locked and lockless pools. The
different was the call to __pool_free() in one case versus open-coded
accesses in the other, and the poisoning brought by commit da52035a45
("MINOR: memory: also poison the area on freeing") which unfortunately
did if only for the lockless path.
Let's now have __pool_free() to work on the global pool also in the
locked case so that the code is architected similarly.
Willy Tarreau [Mon, 1 Jun 2020 16:16:57 +0000 (18:16 +0200)]
MEDIUM: memory: don't let pool_put_to_cache() free the objects itself
Just as for the allocation path, the release path was not symmetrical.
It was not logical to have pool_put_to_cache() free the objects itself,
it was pool_free's job. In addition, just because of a variable export
issue, it the insertion of the object to free back into the local cache
couldn't be inlined while it was very cheap.
This patch just slightly reorganizes this code path by making pool_free()
decide whether or not to put the object back into the cache via
pool_put_to_cache() otherwise place it back to the global pool using
__pool_free(). Then pool_put_to_cache() adds the item to the local
cache and only calls pool_evict_from_cache() if the cache is too big.
Willy Tarreau [Mon, 1 Jun 2020 15:51:05 +0000 (17:51 +0200)]
MINOR: memory: don't let __pool_get_first() pick from the cache
When building with the local cache support, we have an asymmetry in
the allocation path which is that __pool_get_first() picks from the
cache while when no cache support is used, this one directly accesses
the shared area. It looks like it was done this way only to centralize
the call to __pool_get_from_cache() but this was not a good idea as it
complicates the splitting the code.
Let's move the cache access to the upper layer so thatt __pool_get_first()
remains agnostic to the cache support.
The call tree now looks like this with the cache enabled :
pool_get_first()
__pool_get_from_cache() // if cache enabled
__pool_get_first()
pool_alloc()
pool_alloc_dirty()
__pool_get_from_cache() // if cache enabled
__pool_get_first()
__pool_refill_alloc()
Willy Tarreau [Sat, 30 May 2020 16:56:17 +0000 (18:56 +0200)]
REORG: memory: move the OS-level allocator to haproxy/pool-os.h
The memory.h file is particularly complex due to the combination of
debugging options. This patch extracts the OS-level interface and
places it into a new file: pool-os.h.
Doing this also moves pool_alloc_area() and pool_free_area() out of
the #ifndef CONFIG_HAP_LOCKLESS_POOLS, making them usable from
__pool_refill_alloc(), pool_free(), pool_flush() and pool_gc() instead
of having direct calls to malloc/free there that are hard to wrap for
debugging purposes.
Willy Tarreau [Sat, 30 May 2020 16:27:03 +0000 (18:27 +0200)]
REORG: memory: move the pool type definitions to haproxy/pool-t.h
This is the beginning of the move and cleanup of memory.h. This first
step only extracts type definitions and basic macros that are needed
by the files which reference a pool. They're moved to pool-t.h (since
"pool" is more obvious than "memory" when looking for pool-related
stuff). 3 files which didn't need to include the whole memory.h were
updated.
Willy Tarreau [Mon, 1 Jun 2020 10:35:03 +0000 (12:35 +0200)]
CLEANUP: pool: include freq_ctr.h and remove locally duplicated functions
In memory.h we had to reimplement the swrate* functions just because of
a broken circular dependency around freq_ctr.h. Now that this one is
solved, let's get rid of this copy and use the original ones instead.
Willy Tarreau [Mon, 1 Jun 2020 10:18:08 +0000 (12:18 +0200)]
REORG: include: move freq_ctr to haproxy/
types/freq_ctr.h was moved to haproxy/freq_ctr-t.h and proto/freq_ctr.h
was moved to haproxy/freq_ctr.h. Files were updated accordingly, no other
change was applied.
Willy Tarreau [Mon, 1 Jun 2020 10:09:26 +0000 (12:09 +0200)]
CLEANUP: include: remove excessive includes of common/standard.h
Some of them were simply removed as unused (possibly some leftovers
from an older cleanup session), some were turned to haproxy/bitops.h
and a few had to be added (hlua.c and stick-table.h need standard.h
for parse_time_err; htx.h requires chunk.h but used to get it through
standard.h).
Willy Tarreau [Mon, 1 Jun 2020 09:48:35 +0000 (11:48 +0200)]
REORG: include: move integer manipulation functions from standard.h to intops.h
There are quite a number of integer manipulation functions defined in
standard.h, which is one of the reasons why standard.h is included from
many places and participates to the dependencies loop.
Let's just have a new file, intops.h to place all these operations.
These are a few bitops, 32/64 bit mul/div/rotate, integer parsing and
encoding (including varints), the full avalanche hash function, and
the my_htonll/my_ntohll functions.
Willy Tarreau [Mon, 1 Jun 2020 09:05:15 +0000 (11:05 +0200)]
REORG: include: move time.h from common/ to haproxy/
This one is included almost everywhere and used to rely on a few other
.h that are not needed (unistd, stdlib, standard.h). It could possibly
make sense to split it into multiple parts to distinguish operations
performed on timers and the internal time accounting, but at this point
it does not appear much important.
Willy Tarreau [Thu, 28 May 2020 13:29:19 +0000 (15:29 +0200)]
REORG: include: split hathreads into haproxy/thread.h and haproxy/thread-t.h
This splits the hathreads.h file into types+macros and functions. Given
that most users of this file used to include it only to get the definition
of THREAD_LOCAL and MAXTHREADS, the bare minimum was placed into thread-t.h
(i.e. types and macros).
All the thread management was left to haproxy/thread.h. It's worth noting
the drop of the trailing "s" in the name, to remove the permanent confusion
that arises between this one and the system implementation (no "s") and the
makefile's option (no "s").
For consistency, src/hathreads.c was also renamed thread.c.
A number of files were updated to only include thread-t which is the one
they really needed.
Some future improvements are possible like replacing empty inlined
functions with macros for the thread-less case, as building at -O0 disables
inlining and causes these ones to be emitted. But this really is cosmetic.
Willy Tarreau [Thu, 28 May 2020 15:00:53 +0000 (17:00 +0200)]
CLEANUP: threads: remove a few needless includes of hathreads.h
A few files were including it while not needing it (anymore). Some
only required access to the atomic ops and got haproxy/atomic.h in
exchange. Others didn't need it at all. A significant number of
files still include it only for THREAD_LOCAL definition.
Willy Tarreau [Thu, 28 May 2020 13:29:33 +0000 (15:29 +0200)]
REORG: threads: extract atomic ops from hathreads.h
The hathreads.h file has quickly become a total mess because it contains
thread definitions, atomic operations and locking operations, all this for
multiple combinations of threads, debugging and architectures, and all this
done with random ordering!
This first patch extracts all the atomic ops code from hathreads.h to move
it to haproxy/atomic.h. The code there still contains several sections
based on non-thread vs thread, and GCC versions in the latter case. Each
section was arranged in the exact same order to ease finding.
The redundant HA_BARRIER() which was the same as __ha_compiler_barrier()
was dropped in favor of the latter which follows the naming convention
of all other barriers. It was only used in freq_ctr.c which was updated.
Additionally, __ha_compiler_barrier() was defined inconditionally but
used only for thread-related operations, so it was made thread-only like
HA_BARRIER() used to be. We'd still need to have two types of compiler
barriers, one for the general case (e.g. signals) and another one for
concurrency, but this was not addressed here.
Some comments were added at the beginning of each section to inform about
the use case and warn about the traps to avoid.
Some files which continue to include hathreads.h solely for atomic ops
should now be updated.
Willy Tarreau [Wed, 27 May 2020 16:01:47 +0000 (18:01 +0200)]
REORG: include: split mini-clist into haproxy/list and list-t.h
Half of the users of this include only need the type definitions and
not the manipulation macros nor the inline functions. Moves the various
types into mini-clist-t.h makes the files cleaner. The other one had all
its includes grouped at the top. A few files continued to reference it
without using it and were cleaned.
In addition it was about time that we'd rename that file, it's not
"mini" anymore and contains a bit more than just circular lists.
Willy Tarreau [Wed, 27 May 2020 15:22:10 +0000 (17:22 +0200)]
REORG: include: split buf.h into haproxy/buf-t.h and haproxy/buf.h
File buf.h is one common cause of pain in the dependencies. Many files in
the code need it to get the struct buffer definition, and a few also need
the inlined functions to manipulate a buffer, but the file used to depend
on a long chain only for BUG_ON() (addressed by last commit).
Now buf.h is split into buf-t.h which only contains the type definitions,
and buf.h for all inlined functions. Callers who don't care can continue
to use buf.h but files in types/ must only use buf-t.h. sys/types.h had
to be added to buf.h to get ssize_t as used by b_move(). It's worth noting
that ssize_t is only supposed to be a size_t supporting -1, so b_move()
ought to be rethought regarding this.
The files were moved to haproxy/ and all their users were updated
accordingly. A dependency issue was addressed on fcgi whose C file didn't
include buf.h.
Willy Tarreau [Wed, 27 May 2020 15:10:05 +0000 (17:10 +0200)]
CLEANUP: debug: drop unused function p_malloc()
This one was introduced 5 years ago for debugging and never really used.
It is the one which used to cause circular dependencies issues. Let's drop
it instead of starting to split the debug include in two.
Willy Tarreau [Wed, 27 May 2020 14:51:33 +0000 (16:51 +0200)]
REORG: include: move the BUG_ON() code to haproxy/bug.h
This one used to be stored into debug.h but the debug tools got larger
and require a lot of other includes, which can't use BUG_ON() anymore
because of this. It does not make sense and instead this macro should
be placed into the lower includes and given its omnipresence, the best
solution is to create a new bug.h with the few surrounding macros needed
to trigger bugs and place assertions anywhere.
Another benefit is that it won't be required to add include <debug.h>
anymore to use BUG_ON, it will automatically be covered by api.h. No
less than 32 occurrences were dropped.
The FSM_PRINTF macro was dropped since not used at all anymore (probably
since 1.6 or so).
Willy Tarreau [Wed, 27 May 2020 14:21:26 +0000 (16:21 +0200)]
REORG: include: move ist.h from common/ to import/
Fortunately that file wasn't made dependent upon haproxy since it was
integrated, better isolate it before it's too late. Its dependency on
api.h was the result of the change from config.h, which in turn wasn't
correct. It was changed back to stddef.h for size_t and sys/types.h for
ssize_t. The recently added reference to MAX() was changed as it was
placed only to avoid a zero length in the non-free-standing version and
was causing a build warning in the hpack encoder.
Willy Tarreau [Wed, 27 May 2020 14:26:00 +0000 (16:26 +0200)]
REORG: include: move openssl-compat.h from common/ to haproxy/
This file is to openssl what compat.h is to the libc, so it makes sense
to move it to haproxy/. It could almost be part of api.h but given the
amount of openssl stuff that gets loaded I fear it could increase the
build time.
Note that this file contains lots of inlined functions. But since it
does not depend on anything else in haproxy, it remains safe to keep
all that together.
Willy Tarreau [Wed, 27 May 2020 14:01:15 +0000 (16:01 +0200)]
CLEANUP: include: remove unused template.h
There is one "template.h" per include subdirectory to show how to create
a new file but in practice nobody knows they're here so they're useless.
Let's simply remove them.
The choice is simple: if the file only requires type definitions, it includes
api-t.h, otherwise it includes the full api.h.
In addition, in these files, explicit includes for inttypes.h and limits.h
were dropped since these are now covered by api.h and api-t.h.
No other change was performed, given that this patch is large and
affects 201 files. At least one (tools.h) was already freestanding and
didn't get the new one added.
Willy Tarreau [Wed, 27 May 2020 10:00:13 +0000 (12:00 +0200)]
REORG: include: create new file haproxy/api.h
This file includes everything that must be guaranteed to be available to
any buildable file in the project (including the contrib/ subdirs). For
now it includes <haproxy/api-t.h> so that standard integer types and
compiler macros are known, <common/initcall.h> to ease dynamic registration
of init functions, and <common/tools.h> for a few MIN/MAX macros.
version.h should probably also be added, though at the moment it doesn't
bring a great value.
All files which currently include the ones above should now switch to
haproxy/api.h or haproxy/api-t.h instead. This should also reduce build
time by having a single guard for several files at once.
Willy Tarreau [Wed, 3 Jun 2020 14:31:41 +0000 (16:31 +0200)]
REORG: include: create new file haproxy/api-t.h
This file is at the lowest level of the include tree. Its purpose is to
make sure that common types are known pretty much everywhere, particularly
in structure declarations. It will essentially cover integer types such as
uintXX_t via inttypes.h, "size_t" and "ptrdiff_t" via stddef.h, and various
type modifiers such as __maybe_unused or ALIGN() via compiler.h, compat.h
and defaults.h.
It could be enhanced later if required, for example if some macros used
to compute array sizes are needed.
Willy Tarreau [Wed, 27 May 2020 09:05:33 +0000 (11:05 +0200)]
REORG: ebtree: clean up remains of the ebtree/ directory
The only leftovers were the unused compiler.h file and the LICENSE file
which is already mentioned in each and every ebtree file header.
A few build paths were updated in the contrib/ directory not to mention
this directory anymore, and all its occurrences were dropped from the
main makefile. From now on no other include path but include/ will be
needed anymore to build any file.
Willy Tarreau [Wed, 27 May 2020 08:58:19 +0000 (10:58 +0200)]
REORG: ebtree: move the include files from ebtree to include/import/
This is where other imported components are located. All files which
used to directly include ebtree were touched to update their include
path so that "import/" is now prefixed before the ebtree-related files.
The ebtree.h file was slightly adjusted to read compiler.h from the
common/ subdirectory (this is the only change).
A build issue was encountered when eb32sctree.h is loaded before
eb32tree.h because only the former checks for the latter before
defining type u32. This was addressed by adding the reverse ifdef
in eb32tree.h.
No further cleanup was done yet in order to keep changes minimal.
Willy Tarreau [Wed, 27 May 2020 08:43:24 +0000 (10:43 +0200)]
REORG: ebtree: move the C files from ebtree/ to src/
As part of the include files cleanup, we're going to kill the ebtree
directory. For this we need to host its C files in a different location
and src/ is the right one.
BUG/MEDIUM: mworker: fix the reload with an -- option
When HAProxy is started with a '--' option, all following parameters are
considered configuration files. You can't add new options after a '--'.
The current reload system of the master-worker adds extra options at the
end of the arguments list. Which is a problem if HAProxy was started wih
'--'.
This patch fixes the issue by copying the new option at the beginning of
the arguments list instead of the end.
BUG/MINOR: init: -S can have a parameter starting with a dash
There is no reason the -S option can't take an argument which starts with
a -. This limitation must only be used for options that take a
non-finite list of parameters (-sf/-st)
This can be backported only if the previous patch which fixes
copy_argv() is backported too.
BUG/MINOR: init: -x can have a parameter starting with a dash
There is no reason the -x option can't take an argument which starts with
a -. This limitation must only be used for options that take a
non-finite list of parameters (-sf/-st)
This can be backported only if the previous patch which fixes
copy_argv() is backported too.
BUG/MEDIUM: mworker: fix the copy of options in copy_argv()
The copy_argv() function, which is used to copy and remove some of the
arguments of the command line in order to re-exec() the master process,
is poorly implemented.
The function tries to remove the -x and the -sf/-st options but without
taking into account that some of the options could take a parameter
starting with a dash.
In issue #644, haproxy starts with "-L -xfoo" which is perfectly
correct. However, the re-exec is done without "-xfoo" because the master
tries to remove the "-x" option. Indeed, the copy_argv() function does
not know how much arguments an option can have, and just assume that
everything starting with a dash is an option. So haproxy is exec() with
"-L" but without a parameter, which is wrong and leads to the exit of
the master, with usage().
To fix this issue, copy_argv() must know how much parameters an option
takes, and copy or skip the parameters correctly.
This fix is a first step but it should evolve to a cleaner way of
declaring the options to avoid deduplication of the parsing code, so we
avoid new bugs.
Should be backported with care as far as 1.8, by removing the options
that does not exists in the previous versions.
BUG/MEDIUM: contrib/prometheus-exporter: Properly set flags to dump metrics
When the metrics are dumped, in the main function promex_dump_metrics(), the
appctx flags are set before entering in a new scope, among others things to know
which metrics names and descriptions to use. But, those flags are not restored
when the dump is interrupted because of a full output buffer. If this happens
after the dump of global metrics, it may only lead to extra #TYPE and #HELP
lines. But if this happens during the dump of global metrics, the following
dumps of frontends, backends and servers metrics use names and descriptions of
global ones with the unmatching indexes. This first leads to unexisting metrics
names. For instance, "haproxy_frontend_nbproc". But also to out-of-bound
accesses to name and description arrays because there are more stats fields than
info fields.
It is easy to reproduce the bug using small buffers, setting tune.bufsize to
8192 for instance.
This patch should fix the issue #666. It must be backported as far as 2.0.
BUG/MINOR: checks: Fix test on http-check rulesets during config validity check
When checking the config validity of the http-check rulesets, the test on the
ruleset type is inverted. So a warning about ignored directive is emitted when
the config is valid and omitted when it should be reported.
BUG/MEDIUM: hlua: Lock pattern references to perform set/add/del operations
The pattern references lock must be hold to perform set/add/del
operations. Unfortunately, it is not true for the lua functions manipulating acl
and map files.
This patch should fix the issue #664. It must be backported as far as 1.8.
BUG/MEDIUM: lua: Reset analyse expiration timeout before executing a lua action
Before executing a lua action, the analyse expiration timeout of the
corresponding channel must be reset. Otherwise, when it expires, for instance
because of a call to core.msleep(), if the action yields, an expired timeout
will be used for the stream's task, leading to a loop.
This patch should fix the issue #661. It must be backported in all versions
supporting the lua.
MINOR: mux-h1/proxy: Add a proxy option to disable clear h2 upgrade
By default, HAProxy is able to implicitly upgrade an H1 client connection to an
H2 connection if the first request it receives from a given HTTP connection
matches the HTTP/2 connection preface. This way, it is possible to support H1
and H2 clients on a non-SSL connections. It could be a problem if for any
reason, the H2 upgrade is not acceptable. "option disable-h2-upgrade" may now be
used to disable it, per proxy. The main puprose of this option is to let an
admin to totally disable the H2 support for security reasons. Recently, a
critical issue in the HPACK decoder was fixed, forcing everyone to upgrade their
HAProxy version to fix the bug. It is possible to disable H2 for SSL
connections, but not on clear ones. This option would have been a viable
workaround.
This commit was simplifying the certificate chain loading with
SSL_CTX_add_extra_chain_cert() which is available in all SSL libraries.
Unfortunately this function is not compatible with the
multi-certificates bundles, which have the effect of concatenating the
chains of all certificate types instead of creating a chain for each
type (RSA, ECDSA etc.)
Willy Tarreau [Tue, 2 Jun 2020 15:17:13 +0000 (17:17 +0200)]
CLEANUP: regex: remove outdated support for regex actions
The support for reqrep and friends was removed in 2.1 but the
chain_regex() function and the "action" field in the regex struct
was still there. This patch removes them.
One point worth mentioning though. There is a check_replace_string()
function whose purpose was to validate the replacement strings passed
to reqrep. It should also be used for other replacement regex, but is
never called. Callers of exp_replace() should be checked and a call to
this function should be added to detect the error early.
Willy Tarreau [Tue, 2 Jun 2020 10:00:46 +0000 (12:00 +0200)]
BUILD: sink: address build warning on 32-bit architectures
The recently added ring section post-processing added this bening
warning on 32-bit archs:
src/sink.c: In function 'cfg_post_parse_ring':
src/sink.c:994:15: warning: format '%lu' expects argument of type 'long unsigned int', but argument 4 has type 'size_t {aka unsigned int}' [-Wformat=]
ha_warning("ring '%s' event max length '%u' exceeds size, forced to size '%lu'.\n",
^
Let's just cast b_size() to unsigned long here.
CLEANUP: ssl: remove comment from dump_crtlist_sslconf()
Since 8177ad9 ("MINOR: ssl: split config and runtime variable for
ssl-{min,max}-ver"), the dump for ssl-min-ver and ssl-max-ver is fixed,
so we can remove the comment.
MINOR: ssl: set ssl-min-ver in ambiguous configurations
Using ssl-max-ver without ssl-min-ver is ambiguous.
When the ssl-min-ver is not configured, and ssl-max-ver is set to a
value lower than the default ssl-min-ver (which is TLSv1.2 currently),
set the ssl-min-ver to the value of ssl-max-ver, and emit a warning.
Emeric Brun [Fri, 29 May 2020 23:42:45 +0000 (01:42 +0200)]
MEDIUM: ring: add new srv statement to support octet counting forward
log-proto <logproto>
The "log-proto" specifies the protocol used to forward event messages to
a server configured in a ring section. Possible values are "legacy"
and "octet-count" corresponding respectively to "Non-transparent-framing"
and "Octet counting" in rfc6587. "legacy" is the default.
Notes: a separated io_handler was created to avoid per messages test
and to prepare code to set different log protocols such as
request- response based ones.
Emeric Brun [Thu, 28 May 2020 09:13:15 +0000 (11:13 +0200)]
MEDIUM: ring: add server statement to forward messages from a ring
This patch adds new statement "server" into ring section, and the
related "timeout connect" and "timeout server".
server <name> <address> [param*]
Used to configure a syslog tcp server to forward messages from ring buffer.
This supports for all "server" parameters found in 5.2 paragraph.
Some of these parameters are irrelevant for "ring" sections.
timeout connect <timeout>
Set the maximum time to wait for a connection attempt to a server to succeed.
Arguments :
<timeout> is the timeout value specified in milliseconds by default, but
can be in any other unit if the number is suffixed by the unit,
as explained at the top of this document.
timeout server <timeout>
Set the maximum time for pending data staying into output buffer.
Arguments :
<timeout> is the timeout value specified in milliseconds by default, but
can be in any other unit if the number is suffixed by the unit,
as explained at the top of this document.
Example:
global
log ring@myring local7
ring myring
description "My local buffer"
format rfc3164
maxlen 1200
size 32764
timeout connect 5s
timeout server 10s
server mysyslogsrv 127.0.0.1:6514
Willy Tarreau [Sat, 30 May 2020 04:59:07 +0000 (06:59 +0200)]
SCRIPTS: publish-release: pass -n to gzip to remove timestamp
It just appeared that the tar.gz we put online are not reproducible
because a timestamp is put by default into the archive. Passing "-n"
to gzip is sufficient to remove this timestamp, so let's do it, and
also make the gzip command configurable for more flexibility. Now
issuing the commands multiple times finally results in the same
archives being produced.
This should be backported to supported stable branches.
Willy Tarreau [Fri, 29 May 2020 15:23:05 +0000 (17:23 +0200)]
CLEANUP: pools: use the regular lock for the flush operation on lockless pools
Commit 04f5fe87d3d introduced an rwlock in the pools to deal with the risk
that pool_flush() dereferences an area being freed, and commit 899fb8abdcd
turned it into a spinlock. The pools already contain a spinlock in case of
locked pools, so let's use the same and simplify the code by removing ifdefs.
At this point I'm really suspecting that if pool_flush() would instead
rely on __pool_get_first() to pick entries from the pool, the concurrency
problem could never happen since only one user would get a given entry at
once, thus it could not be freed by another user. It's not certain this
would be faster however because of the number of atomic ops to retrieve
one entry compared to a locked batch.
MEDIUM: ssl: use TLSv1.2 as the minimum default on bind lines
Since HAProxy 1.8, the TLS default minimum version was set to TLSv1.0 to
avoid using the deprecated SSLv3.0. Since then, the standard changed and
the recommended TLS version is now TLSv1.2.
This patch changes the minimum default version to TLSv1.2 on bind lines.
If you need to use prior TLS version, this is still possible by
using the ssl-min-ver keyword.
BUG/MEDIUM: checks: Don't add a tcpcheck ruleset twice in the shared tree
When a tcpcheck ruleset is created, it is automatically inserted in a global
tree. Unfortunately for applicative health checks (redis, mysql...), the created
ruleset is inserted a second time during the directive parsing. The leads to a
infinite loop when haproxy is stopped when we try to scan the tree to release
all tcpcheck rulesets.
Now, only the function responsible to create the tcpcheck ruleset insert it into
the tree.
Willy Tarreau [Thu, 28 May 2020 16:07:10 +0000 (18:07 +0200)]
BUG/MINOR: nameservers: fix error handling in parsing of resolv.conf
In issue #657, Coverity found a bug in the "nameserver" parser for the
resolv.conf when "parse-resolv-conf" is set. What happens is that if an
unparsable address appears on a "nameserver" line, it will destroy the
previously allocated pointer before reporting the warning, then the next
"nameserver" line will dereference it again and wlil cause a crash. If
the faulty nameserver is the last one, it will only be a memory leak.
Let's just make sure we preserve the pointer when handling the error.
The patch also fixes a typo in the warning.
The bug was introduced in 1.9 with commit 44e609bfa ("MINOR: dns:
Implement `parse-resolv-conf` directive") so the fix needs to be backported
up to 1.9 or 2.0.
MINOR: checks: I/O callback function only rely on the data layer wake callback
Most of code in event_srv_chk_io() function is inherited from the checks before
the recent refactoring. Now, it is enough to only call wake_srv_chk(). Since the
refactoring, the removed code is dead and never called. wake_srv_chk() may only
return 0 if tcpcheck_main() returns 0 and the check status is unknown
(CHK_RES_UNKNOWN). When this happens, nothing is performed in event_srv_chk_io().
BUG/MEDIUM: checks: Don't blindly subscribe for receive if waiting for connect
When an health check is waiting for a connection establishment, it subscribe for
receive or send events, depending on the next rule to be evaluated. For
subscription for send events, there is no problem. It works as expected. For
subscription for receive events, It only works for HTTP checks because the
underlying multiplexer takes care to do a receive before subscribing again,
updating the fd state. For TCP checks, the PT multiplexer only forwards
subscriptions at the transport layer. So the check itself is woken up. This
leads to a subscribe/notify loop waiting the connection establishment or a
timeout, uselessly eating CPU.
Thus, when a check is waiting for a connect, instead of blindly resubscribe for
receive events when it is woken up, we now try to receive data.
This patch should fix the issue #635. No backport needed.
HTTP_1XX, HTTP_3XX and HTTP_4XX message templates are no longer used. Only
HTTP_302 and HTTP_303 are used during configuration parsing by "errorloc" family
directives. So these templates are removed from the generic http code. And
HTTP_302 and HTTP_303 templates are moved as static strings in the function
parsing "errorloc" directives.
MINOR: http-rules: Use an action function to eval http-request auth rules
Now http-request auth rules are evaluated in a dedicated function and no longer
handled "in place" during the HTTP rules evaluation. Thus the action name
ACT_HTTP_REQ_AUTH is removed. In additionn, http_reply_40x_unauthorized() is
also removed. This part is now handled in the new action_ptr callback function.