Instead of giving the memory pools names with an explicit call to
isc_mempool_setname(), add the name to isc_mempool_create() call to have
all the memory pools an unconditional name.
Instead of giving the memory context names with an explicit call to
isc_mem_setname(), add the name to isc_mem_create() call to have all the
memory contexts an unconditional name.
previously, ISC_LIST_FOREACH and ISC_LIST_FOREACH_SAFE were
two separate macros, with the _SAFE version allowing entries
to be unlinked during the loop. ISC_LIST_FOREACH is now also
safe, and the separate _SAFE macro has been removed.
similarly, the ISC_LIST_FOREACH_REV macro is now safe, and
ISC_LIST_FOREACH_REV_SAFE has also been removed.
qp-tries allocate their nodes (twigs) in chunks to reduce allocator
pressure and improve memory locality. The choice of chunk size presents
a tradeoff: larger chunks benefit qp-tries with many values (as seen
in large zones and resolvers) but waste memory in smaller use cases.
Previously, our fixed chunk size of 2^10 twigs meant that even an
empty qp-trie would consume 12KB of memory, while reducing this size
would negatively impact resolver performance.
This commit implements an adaptive chunking strategy that:
- Tracks the size of the most recently allocated chunk.
- Doubles the chunk size for each new allocation until reaching a
predefined maximum.
This approach effectively balances memory efficiency for small tries
while maintaining the performance benefits of larger chunk sizes for
bigger data structures.
This commit also splits the callback freeing qpmultis into two
phases, one that frees the underlying qptree, and one that reclaims
the qpmulti memory. In order to prevent races between the qpmulti
destructor and chunk garbage collection jobs, the second phase is
protected by reference counting.
If a call_rcu thread is running, there is a possible race condition
where the destructors run before all call_rcu callbacks have finished
running. This can happen, for example, if the call_rcu callback tries to
log something after the logging context has been torn down.
In !10394, we tried to counter this by explicitely creating a call_rcu
thread an shutting it down before running the destructors, but it is
possible for things to "slip" and end up on the default call_rcu thread.
As a quickfix, this commit moves an rcu_barrier() that was in the mem
context destructor earlier, so that it "protects" all libisc
destructors.
use the ISC_LIST_FOREACH pattern in places where lists had
been iterated using a different pattern from the typical
`for` loop: for example, `while (!ISC_LIST_EMPTY(...))` or
`while ((e = ISC_LIST_HEAD(...)) != NULL)`.
the pattern `for (x = ISC_LIST_HEAD(...); x != NULL; ISC_LIST_NEXT(...)`
has been changed to `ISC_LIST_FOREACH` throughout BIND, except in a few
cases where the change would be excessively complex.
in most cases this was a straightforward change. in some places,
however, the list element variable was referenced after the loop
ended, and the code was refactored to avoid this necessity.
also, because `ISC_LIST_FOREACH` uses typeof(list.head) to declare
the list elements, compilation failures can occur if the list object
has a `const` qualifier. some `const` qualifiers have been removed
from function parameters to avoid this problem, and where that was not
possible, `UNCONST` was used.
When allocating memory under -m trace|record, the __FILE__ pointer is
stored, so it can be printed out later in order to figure out in which
file an allocation leaked. (among others, like the line number).
However named crashes when called with -m record and using a plugin
leaking memory. The reason is that plugins are unloaded earlier than
when the leaked allocations are dumped (obviously, as it's done as late
as possible). In such circumstances, __FILE__ is dangling because the
dynamically loaded library (the plugin) is not in memory anymore.
Fix the crash by systematically copying the __FILE__ string
instead of copying the pointer. Of course, this make each allocation to
consume a bit more memory (and longer, as it needs to calculate the
length of __FILE__) but this occurs only under -m trace|record debugging
flags.
In term of unit test, because grepping in C is not fun, and because the
whole "syntax" of the dump output is tested in other tests, this simply
search for a substring in the whole buffer to make sure the expected
allocations are found.
The isc_mem API is one of the most commonly used APIs that didn't
used ISC_REFCOUNT_DECL and ISC_REFCOUNT_IMPL macros. Replace the
implementation of isc_mem_attach(), isc_mem_detach() and
isc_mem_destroy() with the respective macros.
This also removes the legacy isc_mem_destroy() functionality that would
check whether all references had been detached from the memory context
as it doesn't work reliably when using the call_rcu() API. Instead of
doing this individually, call isc_mem_checkdestroyed(stderr) from the
isc_mem_destroy() macro to keep the extra check that all contexts were
freed when the program is exiting.
Instead of relying on unreliable order of execution of the library
constructors and destructors, move them to individual binaries. The
advantage is that the execution time and order will remain constant and
will not depend on the dynamic load dependency solver.
This requires more work, but that was mitigated by a simple requirement,
any executable using libisc and libdns, must include <isc/lib.h> and
<dns/lib.h> respectively (in this particular order). In turn, these two
headers must not be included from within any library as they contain
inlined functions marked with constructor/destructor attributes.
The previous work in this area was led by the belief that we might be
calling call_rcu() from within call_rcu() callbacks. After carefully
checking all the current callback, it became evident that this is not
the case and the problem isn't enough rcu_barrier() calls, but something
entirely else.
Call the rcu_barrier() just once as that's enough and the multiple
rcu_barrier() calls will not hide the real problem anymore, so we can
find it.
As it was pointed out, the alignas() can't be used on objects larger
than `max_align_t` otherwise the compiler might miscompile the code to
use auto-vectorization on unaligned memory.
As we were only using alignas() as a way to prevent false memory
sharing, we can use manual padding in the affected structures.
Instead of crude 5x rcu_barrier() call in the isc__mem_destroy(), change
the mechanism to call rcu_barrier() until the memory use and references
stops decreasing. This should deal with any number of nested call_rcu()
levels.
Additionally, don't destroy the contextslock if the list of the contexts
isn't empty. Destroying the lock could make the late threads crash.
Because we don't use jemalloc functions directly, but only via the
libisc library, the dynamic linker might pull the jemalloc library
too late when memory has been already allocated via standard libc
allocator.
Add a workaround round isc_mem_create() that makes the dynamic linker
to pull jemalloc earlier than libc.
Previously, there were two methods of working with the overmem
condition:
1. hi/lo water callback - when the overmem condition was reached
for the first time, the water callback was called with HIWATER
mark and .is_overmem boolean was set internally. Similarly,
when the used memory went below the lo water mark, the water
callback would be called with LOWATER mark and .is_overmem
was reset. This check would be called **every** time memory
was allocated or freed.
2. isc_mem_isovermem() - a simple getter for the internal
.is_overmem flag
This commit refactors removes the first method and move the hi/lo water
checks to the isc_mem_isovermem() function, thus we now have only a
single method of checking overmem condition and the check for hi/lo
water is removed from the hot path for memory contexts that doesn't use
overmem checks.
We now depend on explicitly creating memory arenas and disabling tcache
on those, and these features are not available with jemalloc < 4.
Instead of working around these issues, make the jemalloc >= 4.0.0 hard
requirement by looking for sdallocx() symbol that's only available from
that version.
The jemalloc < 4 was only used by RHEL 7 which is not supported since
BIND 9.19+.
This commit extends the internal memory management middleware code in
BIND so that memory contexts backed by dedicated jemalloc arenas can
be created. A new function (isc_mem_create_arena()) is added for that.
Moreover, it extends the existing code so that specialised memory
contexts can be created easily, should we need that functionality for
other future purposes. We have achieved that by passing the flags to
the underlying jemalloc-related calls. See the above
isc_mem_create_arena(), which can serve as an example of this.
Having this opens up possibilities for creating memory contexts tuned
for specific needs.
Use the new isc_mem_c*() calloc-like API for allocations that are
zeroed.
In turn, this also fixes couple of incorrect usage of the ISC_MEM_ZERO
for structures that need to be zeroed explicitly.
There are few places where isc_mem_cput() is used on structures with a
flexible member (or similar).
Add new isc_mem_cget(), isc_mem_creget(), and isc_mem_cput() macros to
complement the isc_mem_callocate() (which works like calloc()).
The overflow checks are implemented as macros in the <isc/mem.h>, so
that the compiler can see that the element size is constant: it should
always be `sizeof(something)`.
Because rcu_barrier() needs to be called as many times as the number of
nested call_rcu() calls (call_rcu() calls made from call_rcu thread),
and currently there's no mechanism to detect whether there are more
call_rcu callbacks scheduled, we simply call the rcu_barrier() multiple
times. The overhead is negligible and it prevents rare assertion
failures caused by the check for memory leaks in isc__mem_destroy().
As well as clearing the fresh memory, `calloc()`-like functions must
ensure that the count and size do not overflow when multiplied.
Use `isc_mem_callocate()` in `isc__uv_calloc()`.
Memory reclamation by `call_rcu()` is asynchronous, so during shutdown
it can lose a race with the destruction of its memory context. When we
defer memory reclamation, we need to attach to the memory context to
indicate that it is still in use, but that is not enough to delay its
destruction. So, call `rcu_barrier()` in `isc_mem_destroy()` to wait
for pending RCU work to finish before proceeding to destroy the memory
context.
isc_bind9 was a global bool used to indicate whether the library
was being used internally by BIND or by an external caller. external
use is no longer supported, but the variable was retained for use
by dyndb, which needed it only when being built without libtool.
building without libtool is *also* no longer supported, so the variable
can go away.
This restores the Malloced memory counter and it's now always equal to
InUse counter. This is only for backwards compatibility reason and
there is no separate counter.
The commit also cleanups little things like structure with a single
item (summary.inuse), and shuts up a wrong cppcheck warning (the
notorious NULL check after assignment).
Instead of enforcing stronger synchronization between threads, make all
the atomic operations relaxed. We are not really interested in exact
numbers at all times - the single place where we need the exact number
is when the memory context is being destroyed. Even when there's a
overmem counter, we don't care about exact ordering or exact number.
The Lost memory counter would count the memory "lost" by external
libraries. There's really no such thing as `named` require the memory
contexts to be clean on destroy.
The stats buckets were again more useful for internal allocator, because
we would see the individual "block" caches where the allocations would
fall into. Remove the stats buckets, and if needed, we can pull more
detailed statistics out of the jemalloc.
The total memory counter had again little or no meaning when we removed
the internal memory allocator. It was just a monotonic counter that
would count add the allocation sizes but never subtracted anything, so
it would be just a "big number".
The maxinuse memory counter indicated the highest amount of
memory allocated in the past. Checking and updating this high-
water mark value every time memory was allocated had an impact
on server performance, so it has been removed. Memory size can
be monitored more efficiently via an external tool logging RSS.
The malloced and maxmalloced memory counters were mostly useless since
we removed the internal allocator blocks - it would only differ from
inuse by the memory context size itself.
This commit ensures that BIND and supplementary tools still can be
built on newer versions of DragonFly BSD. It used to be the case, but
somewhere between versions 6.2 and 6.4 the OS developers rearranged
headers and moved some function definitions around.
Before that the fact that it worked was more like a coincidence, this
time we, at least, looked at the related man pages included with the
OS.
No in depth testing has been done on this OS as we do not really
support this platform - so it is more like a goodwill act. We can,
however, use this platform for testing purposes, too. Also, we know
that the OS users do use BIND, as it is included in its ports
directory.
Building with './configure' and './configure --without-jemalloc' have
been fixed and are known to work at the time the commit is made.
I.e. print the name of the function in BIND that called the system
function that returned an error. Since it was useful for pthreads
code, it seems worthwhile doing so everywhere.
Mostly generated automatically with the following semantic patch,
except where coccinelle was confused by #ifdef in lib/isc/net.c
@@ expression list args; @@
- UNEXPECTED_ERROR(__FILE__, __LINE__, args)
+ UNEXPECTED_ERROR(args)
@@ expression list args; @@
- FATAL_ERROR(__FILE__, __LINE__, args)
+ FATAL_ERROR(args)
Replace all uses of RUNTIME_CHECK() in lib/isc/include/isc/once.h with
PTHEADS_RUNTIME_CHECK(), in order to improve error reporting for any
once-related run-time failures (by augmenting error messages with
file/line/caller information and the error string corresponding to
errno).
Previously, the isc_mem_get_aligned() and friends took alignment size as
one of the arguments. Replace the specific function with more generic
extended variant that now accepts ISC_MEM_ALIGN(alignment) for aligned
allocations and ISC_MEM_ZERO for allocations that zeroes
the (re-)allocated memory before returning the pointer to the caller.
Unlike standard free(), isc_mem_free() is not a no-op when passed a
NULL pointer. For size accounting purposes it calls sallocx(), which
crashes when passed a NULL pointer. To get more helpful diagnostics,
REQUIRE() that the pointer is not NULL so that when the programmer
makes a mistake they get a backtrace that shows what went wrong.
