false_positive_rate is a parameter that can be set with the bloom
opclass in BRIN, and setting it to a value of exactly 0.25 would trigger
an assertion in the first INSERT done on the index with value set.
The assertion changed here relied on BLOOM_{MIN|MAX}_FALSE_POSITIVE_RATE
that are somewhat arbitrary values, and specifying an out-of-range value
would also trigger a failure when defining such an index. So, as-is,
the assertion was just doubling on the min-max check of the reloption.
This is now enlarged to check that it is a correct percentage value,
instead, based on a suggestion by Tom Lane.
Author: Alexander Lakhin
Reviewed-by: Tom Lane, Shihao Zhong
Discussion: https://postgr.es/m/17969-a6c54de48026d694@postgresql.org
Backpatch-through: 14
array_set_element() and related functions allow an array to be
enlarged by assigning to subscripts outside the current array bounds.
While these places were careful to check that the new bounds are
allowable, they neglected to consider the risk of integer overflow
in computing the new bounds. In edge cases, we could compute new
bounds that are invalid but get past the subsequent checks,
allowing bad things to happen. Memory stomps that are potentially
exploitable for arbitrary code execution are possible, and so is
disclosure of server memory.
To fix, perform the hazardous computations using overflow-detecting
arithmetic routines, which fortunately exist in all still-supported
branches.
The test cases added for this generate (after patching) errors that
mention the value of MaxArraySize, which is platform-dependent.
Rather than introduce multiple expected-files, use psql's VERBOSITY
parameter to suppress the printing of the message text. v11 psql
lacks that parameter, so omit the tests in that branch.
Our thanks to Pedro Gallegos for reporting this problem.
Security: CVE-2023-5869
transformAggregateCall() captures the datatypes of the aggregate's
arguments immediately to construct the Aggref.aggargtypes list.
This seems reasonable because the arguments have already been
transformed --- but there is an edge case where they haven't been.
Specifically, if we have an unknown-type literal in an ANY argument
position, nothing will have been done with it earlier. But if we
also have DISTINCT, then addTargetToGroupList() converts the literal
to "text" type, resulting in the aggargtypes list not matching the
actual runtime type of the argument. The end result is that the
aggregate tries to interpret a "text" value as being of type
"unknown", that is a zero-terminated C string. If the text value
contains no zero bytes, this could result in disclosure of server
memory following the text literal value.
To fix, move the collection of the aggargtypes list to the end
of transformAggregateCall(), after DISTINCT has been handled.
This requires slightly more code, but not a great deal.
Our thanks to Jingzhou Fu for reporting this problem.
Security: CVE-2023-5868
It was always false in single-user mode, in autovacuum workers, and in
background workers. This had no specifically-identified security
consequences, but non-core code or future work might make it
security-relevant. Back-patch to v11 (all supported versions).
Jelte Fennema-Nio. Reported by Jelte Fennema-Nio.
Documentation says it cannot signal "a backend owned by a superuser".
On the contrary, it could signal background workers, including the
logical replication launcher. It could signal autovacuum workers and
the autovacuum launcher. Block all that. Signaling autovacuum workers
and those two launchers doesn't stall progress beyond what one could
achieve other ways. If a cluster uses a non-core extension with a
background worker that does not auto-restart, this could create a denial
of service with respect to that background worker. A background worker
with bugs in its code for responding to terminations or cancellations
could experience those bugs at a time the pg_signal_backend member
chooses. Back-patch to v11 (all supported versions).
Reviewed by Jelte Fennema-Nio. Reported by Hemanth Sandrana and
Mahendrakar Srinivasarao.
Security: CVE-2023-5870
This text left one with the impression that an ON SELECT rule could
be attached to a plain table, which has not been true since commit
264c06820 (meaning the text was already misleading when written,
evidently by me in 96bd67f61). However, it didn't get really bad
until b23cd185f removed the convert-a-table-to-a-view logic, which
had made it possible for scripts that thought they were attaching
ON SELECTs to tables to still work.
Rewrite into a form that makes it clear that an ON SELECT rule
is better regarded as an implementation detail of a view.
Pre-v16, point out that adding ON SELECT to a table actually
converts it to a view.
Per bug #18178 from Joshua Uyehara. Back-patch to all supported
branches.
Discussion: https://postgr.es/m/18178-05534d7064044d2d@postgresql.org
get_explain_guc_options() crashed if a string GUC marked GUC_EXPLAIN
has a NULL boot_val. Nosing around found a couple of other places
that seemed insufficiently cautious about NULL string values, although
those are likely unreachable in practice. Add some commentary
defining the expectations for NULL values of string variables,
in hopes of forestalling future additions of more such bugs.
Xing Guo, Aleksander Alekseev, Tom Lane
Discussion: https://postgr.es/m/CACpMh+AyDx5YUpPaAgzVwC1d8zfOL4JoD-uyFDnNSa1z0EsDQQ@mail.gmail.com
The test missed that custom GUCs need to be ignored from the list of
parameters that can exist in postgresql.conf.sample. This caused the
test to fail on a server where such a module is loaded, when using
EXTRA_INSTALL and TEMP_CONFIG, for instance.
Author: Anton A. Melnikov
Discussion: https://postgr.es/m/fc5509ce-5144-4dac-8d13-21793da44fc5@postgrespro.ru
Backpatch-through: 15
This part of the documentation refers to exceptions as handled by
PL/pgSQL, and using the internal error code is confusing.
Per thinko in 66bde49d96.
Reported-by: Euler Taveira, Bruce Momjian
Discussion: https://postgr.es/m/ZUEUnLevXyW7DlCs@momjian.us
Backpatch-through: 11
4b8266415 added a precheck to pgrowlocks() to ensure the given object's
pg_class.relam is HEAP_TABLE_AM_OID, however, that check was put before
another check which was checking if the given object was a partitioned
table. Since the pg_class.relam is always InvalidOid for partitioned
tables, if pgrowlocks() was called passing a partitioned table, then the
"only heap AM is supported" error would be raised instead of the intended
error about the given object being a partitioned table.
Here we simply move the pg_class.relam check to after the check that
verifies that we are in fact working with a normal (non-partitioned)
table.
Reported-by: jian he
Discussion: https://postgr.es/m/CACJufxFaSp_WguFCf0X98951zFVX+dXFnF1mxAb-G3g1HiHOow@mail.gmail.com
Backpatch-through: 12, where 4b8266415 was introduced.
pgstatindex failed with ERRCODE_DATA_CORRUPTED, of the "can't-happen"
class XX. The other functions succeeded on an empty index; they might
have malfunctioned if the failed index build left torn I/O or other
complex state. Report an ERROR in statistics functions pgstatindex,
pgstatginindex, pgstathashindex, and pgstattuple. Report DEBUG1 and
skip all index I/O in maintenance functions brin_desummarize_range,
brin_summarize_new_values, brin_summarize_range, and
gin_clean_pending_list. Back-patch to v11 (all supported versions).
Discussion: https://postgr.es/m/20231001195309.a3@google.com
This prevents false-positive reports about "the first child of leftmost
target page is not leftmost of its level", "block %u is not leftmost"
and "left link/right link pair". They appeared if amcheck ran before
VACUUM cleaned things, after a cluster exited recovery between the
first-stage and second-stage WAL records of a deletion. Back-patch to
v11 (all supported versions).
Reviewed by Peter Geoghegan.
Discussion: https://postgr.es/m/20231005025232.c7.nmisch@google.com
Formerly, the value computed by leftmostvalue_interval() was a long
way short of the minimum possible interval value. As a result, an
index scan on a GIN index on an interval column with < or <= operators
would miss large negative interval values.
Fix by setting all fields of the leftmost interval to their minimum
values, ensuring that the result is less than any other possible
interval. Since this only affects index searches, no index rebuild is
necessary.
Back-patch to all supported branches.
Dean Rasheed, reviewed by Heikki Linnakangas.
Discussion: https://postgr.es/m/CAEZATCV80%2BgOfF8ehNUUfaKBZgZMDfCfL-g1HhWGb6kC3rpDfw%40mail.gmail.com
When looping around after finding that the set-returning function
returned zero rows for the current input tuple, ExecProjectSet
neglected to reset either of the two memory contexts it's
responsible for cleaning out. Typically this wouldn't cause much
problem, because once the SRF does return at least one row, the
contexts would get reset on the next call. However, if the SRF
returns no rows for many input tuples in succession, quite a lot
of memory could be transiently consumed.
To fix, make sure we reset both contexts while looping around.
Per bug #18172 from Sergei Kornilov. Back-patch to all supported
branches.
Discussion: https://postgr.es/m/18172-9b8c5fc1d676ded3@postgresql.org
Debian recently decided to split out a bunch of "obsolete" timezone
names into a new tzdata-legacy package, which isn't installed by
default. One of these zone names is Pacific/Enderbury, and that
breaks our regression tests (on --with-system-tzdata builds)
because our default timezone abbreviations list defines PHOT as
Pacific/Enderbury.
Pacific/Enderbury got renamed to Pacific/Kanton in tzdata 2021b,
so that in distros that still have this entry it's just a symlink
to Pacific/Kanton anyway. So one answer would be to redefine PHOT
as Pacific/Kanton. However, then things would fail if the
installed tzdata predates 2021b, which is recent enough that that
seems like a real problem.
Instead, let's just remove PHOT from the default list. That seems
likely to affect nobody in the real world, because (a) it was an
abbreviation that the tzdb crew made up in the first place, with
no evidence of real-world usage, and (b) the total human population
of the Phoenix Islands is less than two dozen persons, per Wikipedia.
If anyone does use this zone abbreviation they can easily put it back
via a custom abbreviations file.
We'll keep PHOT in the Pacific.txt reference file, but change it
to Pacific/Kanton there, as that definition seems more likely to
be useful to future readers of that file.
Per report from Victor Wagner. Back-patch to all supported
branches.
Discussion: https://postgr.es/m/20231027152049.4b5c8044@wagner.wagner.home
When calculating distance for interval values, the code mostly mimicked
interval_mi, i.e. it built a new interval value for the difference.
That however does not work for sufficiently distant interval values,
when the difference overflows the interval range.
Instead, we can calculate the distance directly, without constructing
the intermediate (and unnecessary) interval value.
Backpatch to 14, where minmax-multi indexes were introduced.
Reported-by: Dean Rasheed
Reviewed-by: Ashutosh Bapat, Dean Rasheed
Backpatch-through: 14
Discussion: https://postgr.es/m/eef0ea8c-4aaa-8d0d-027f-58b1f35dd170@enterprisedb.com
Make sure that infinite values in date/timestamp columns are treated as
if in infinite distance. Infinite values should not be merged with other
values, leaving them as outliers. The code however returned distance 0
in this case, so that infinite values were merged first. While this does
not break the index (i.e. it still produces correct query results), it
may make it much less efficient.
We don't need explicit handling of infinite date/timestamp values when
calculating distances, because those values are represented as extreme
but regular values (e.g. INT64_MIN/MAX for the timestamp type).
We don't need an exact distance, just a value that is much larger than
distanced between regular values. With the added cast to double values,
we can simply subtract the values.
The regression test queries a value in the "gap" and checks the range
was properly eliminated by the BRIN index.
This only affects minmax-multi indexes on timestamp/date columns with
infinite values, which is not very common in practice. The affected
indexes may need to be rebuilt.
Backpatch to 14, where minmax-multi indexes were introduced.
Reported-by: Ashutosh Bapat
Reviewed-by: Ashutosh Bapat, Dean Rasheed
Backpatch-through: 14
Discussion: https://postgr.es/m/eef0ea8c-4aaa-8d0d-027f-58b1f35dd170@enterprisedb.com
When calculating the distance between date values, make sure to subtract
them in the right order, i.e. (larger - smaller).
The distance is used to determine which values to merge, and is expected
to be a positive value. The code unfortunately did the subtraction in
the opposite order, i.e. (smaller - larger), thus producing negative
values and merging values the most distant values first.
The resulting index is correct (i.e. produces correct results), but may
be significantly less efficient. This affects all minmax-multi indexes
on date columns.
Backpatch to 14, where minmax-multi indexes were introduced.
Reported-by: Ashutosh Bapat
Reviewed-by: Ashutosh Bapat, Dean Rasheed
Backpatch-through: 14
Discussion: https://postgr.es/m/eef0ea8c-4aaa-8d0d-027f-58b1f35dd170@enterprisedb.com
When calculating distances for timestamp values for BRIN minmax-multi
indexes, we need to be careful about overflows for extreme values. If
the value overflows into a negative value, the index may be inefficient.
The new regression test checks this for the timestamp type by adding a
table with enough values to force range compaction/merging. The values
are close to min/max, which means a risk of overflow.
Fixed by converting the int64 values to double first, before calculating
the distance. This prevents the overflow. We may lose some precision, of
course, but that's good enough. In the worst case we build a slightly
less efficient index, but for large distances this won't matter.
This only affects minmax-multi indexes on timestamp columns, with ranges
containing values sufficiently distant to cause an overflow. That seems
like a fairly rare case in practice.
Backpatch to 14, where minmax-multi indexes were introduced.
Reported-by: Ashutosh Bapat
Reviewed-by: Ashutosh Bapat, Dean Rasheed
Backpatch-through: 14
Discussion: https://postgr.es/m/eef0ea8c-4aaa-8d0d-027f-58b1f35dd170@enterprisedb.com
As far as I can see, ecpg has no notion of a "default" open
connection. You can do "CONNECT TO DEFAULT" but that just specifies
letting libpq use all its default connection parameters --- the
resulting connection is not special subsequently. In particular,
SET CONNECTION = DEFAULT and DISCONNECT DEFAULT simply act on a
connection named DEFAULT, if you've made one; they do not have
special lookup rules. But the documentation of these commands
makes it look like they do.
Simplest fix, I think, is just to remove the paras suggesting that
DEFAULT is special here.
Also, SET CONNECTION *does* have one special lookup rule, which
is that it recognizes CURRENT as an alias for the currently selected
connection. SET CONNECTION = CURRENT is a no-op, so it's pretty
useless, but nonetheless it does something different from selecting
a connection by name; so we'd better document it.
Per report from Sylvain Frandaz. Back-patch to all supported
versions.
Discussion: https://postgr.es/m/169824721149.1769274.1553568436817652238@wrigleys.postgresql.org
When an UPDATE/DELETE/MERGE's target table is an old-style
inheritance tree, it's possible for the parent to get excluded
from the plan while some children are not. (I believe this is
only possible if we can prove that a CHECK ... NO INHERIT
constraint on the parent contradicts the query WHERE clause,
so it's a very unusual case.) In such a case, ExecInitModifyTable
mistakenly concluded that the first surviving child is the target
table, leading to at least two bugs:
1. The wrong table's statement-level triggers would get fired.
2. In v16 and up, it was possible to fail with "invalid perminfoindex
0 in RTE with relid nnnn" due to the child RTE not having permissions
data included in the query plan. This was hard to reproduce reliably
because it did not occur unless the update triggered some non-HOT
index updates.
In v14 and up, this is easy to fix by defining ModifyTable.rootRelation
to be the parent RTE in plain inheritance as well as partitioned cases.
While the wrong-triggers bug also appears in older branches, the
relevant code in both the planner and executor is quite a bit
different, so it would take a good deal of effort to develop and
test a suitable patch. Given the lack of field complaints about the
trigger issue, I'll desist for now. (Patching v11 for this seems
unwise anyway, given that it will have no more releases after next
month.)
Per bug #18147 from Hans Buschmann.
Amit Langote and Tom Lane
Discussion: https://postgr.es/m/18147-6fc796538913ee88@postgresql.org
Clearly spell out the limitations of aminsert()'s indexUnchanged hinting
mechanism in the index AM documentation.
Oversight in commit 9dc718bd, which added the "logically unchanged
index" hint (which is used to trigger bottom-up index deletion).
Author: Peter Geoghegan <pg@bowt.ie>
Reported-By: Tom Lane <tgl@sss.pgh.pa.us>
Reviewed-By: Tom Lane <tgl@sss.pgh.pa.us>
Discussion: https://postgr.es/m/CAH2-WzmU_BQ=-H9L+bxTSMQBqHMjp1DSwGypvL0gKs+dTOfkKg@mail.gmail.com
Backpatch: 14-, where indexUnchanged hinting was introduced.
When scanning build farm results, it's useful to be able to see which
version is in use. For the Meson build system, this information was
already displayed.
Back-patch to all supported branches.
Discussion: https://postgr.es/m/4022690.1697852728%40sss.pgh.pa.us
When min_dynamic_shared_memory is set above 0, we try to find space in a
pre-allocated region of the main shared memory area instead of calling
dsm_impl_XXX() routines to allocate more. The dsm_pin_segment() and
dsm_unpin_segment() routines had a bug: they called dsm_impl_XXX()
routines even for main region segments. Nobody noticed before now
because those routines do nothing on Unix, but on Windows they'd fail
while attempting to duplicate an invalid Windows HANDLE. Add the
missing gating.
Back-patch to 14, where commit 84b1c63a added this feature. Fixes
pgsql-bugs bug #18165.
Reported-by: Maxime Boyer <maxime.boyer@cra-arc.gc.ca>
Tested-by: Alexander Lakhin <exclusion@gmail.com>
Discussion: https://postgr.es/m/18165-bf4f525cea6e51de%40postgresql.org
This avoids a compiler bug occurring in AIX's xlc, even in pretty
late-model revisions. Buildfarm testing has now confirmed that
only 64-bit xlc is affected. Although we are contemplating
dropping support for xlc in v17, it's still supported in the
back branches, so we need this fix.
Back-patch of code changes from HEAD commit 19fa97731.
(The test cases were already back-patched, in 4a427b82c et al.)
Discussion: https://postgr.es/m/CA+hUKGK=DOC+hE-62FKfZy=Ybt5uLkrg3zCZD-jFykM-iPn8yw@mail.gmail.com
This doco said that use of => as an operator "is deprecated".
It's been fully disallowed since 865f14a2d back in 9.5, but
evidently that commit missed updating this statement.
Do so now.
When processing a match tag, check to see if the claimed "off"
is more than the distance back to the output buffer start.
If it is, then the data is corrupt, and what's more we would
fetch from outside the buffer boundaries and potentially incur
a SIGSEGV. (Although the odds of that seem relatively low, given
that "off" can't be more than 4K.)
Back-patch to v13; before that, this function wasn't really
trying to protect against bad data.
Report and fix by Flavien Guedez.
Discussion: https://postgr.es/m/01fc0593-e31e-463d-902c-dd43174acee2@oopacity.net
Remove use of LLVMGetElementType() and provide the type of all pointers
to LLVMBuildXXX() functions when emitting IR, as required by modern LLVM
versions[1].
* For LLVM <= 14, we'll still use the old LLVMBuildXXX() functions.
* For LLVM == 15, we'll continue to do the same, explicitly opting
out of opaque pointer mode.
* For LLVM >= 16, we'll use the new LLVMBuildXXX2() functions that take
the extra type argument.
The difference is hidden behind some new IR emitting wrapper functions
l_load(), l_gep(), l_call() etc. The change is mostly mechanical,
except that at each site the correct type had to be provided.
In some places we needed to do some extra work to get functions types,
including some new wrappers for C++ APIs that are not yet exposed by in
LLVM's C API, and some new "example" functions in llvmjit_types.c
because it's no longer possible to start from the function pointer type
and ask for the function type.
Back-patch to 12, because it's a little tricker in 11 and we agreed not
to put the latest LLVM support into the upcoming final release of 11.
[1] https://llvm.org/docs/OpaquePointers.html
Reviewed-by: Dmitry Dolgov <9erthalion6@gmail.com>
Reviewed-by: Ronan Dunklau <ronan.dunklau@aiven.io>
Reviewed-by: Andres Freund <andres@anarazel.de>
Discussion: https://postgr.es/m/CA%2BhUKGKNX_%3Df%2B1C4r06WETKTq0G4Z_7q4L4Fxn5WWpMycDj9Fw%40mail.gmail.com
Per code coverage reports, we had zero regression test coverage
of these functions. That came back to bite us, as apparently
that's allowed us to miss discovering misbehavior of this code
with AIX's xlc compiler. Install relevant portions of the
test cases added in 97957fdba, 2f0472030, 19fa97731.
(Assuming the expected outcome that the xlc problem does appear
in back branches, a code fix will follow.)
Discussion: https://postgr.es/m/CA+hUKGK=DOC+hE-62FKfZy=Ybt5uLkrg3zCZD-jFykM-iPn8yw@mail.gmail.com
The SIGTERM handler for the startup process immediately calls
proc_exit() for the duration of the restore_command, i.e., a call
to system(). This system() call forks a new process to execute the
shell command, and this child process inherits the parent's signal
handlers. If both the parent and child processes receive SIGTERM,
both will attempt to call proc_exit(). This can end badly. For
example, both processes will try to remove themselves from the
PGPROC shared array.
To fix this problem, this commit adds a check in
StartupProcShutdownHandler() to see whether MyProcPid == getpid().
If they match, this is the parent process, and we can proc_exit()
like before. If they do not match, this is a child process, and we
just emit a message to STDERR (in a signal safe manner) and
_exit(), thereby skipping any problematic exit callbacks.
This commit also adds checks in proc_exit(), ProcKill(), and
AuxiliaryProcKill() that verify they are not being called within
such child processes.
Suggested-by: Andres Freund
Reviewed-by: Thomas Munro, Andres Freund
Discussion: https://postgr.es/m/Y9nGDSgIm83FHcad%40paquier.xyz
Discussion: https://postgr.es/m/20230223231503.GA743455%40nathanxps13
Backpatch-through: 11
Dropping a temp table could entail TOAST table access to clean out
toasted catalog entries, such as large pg_constraint.conbin strings
for complex CHECK constraints. If we did that via ON COMMIT DROP,
we triggered the assertion in init_toast_snapshot(), because
there was no provision for setting up a snapshot for the drop
actions. Fix that.
(I assume here that the adjacent truncation actions for ON COMMIT
DELETE ROWS don't have a similar problem: it doesn't seem like
nontransactional truncations would need to touch any toasted fields.
If that proves wrong, we could refactor a bit to have the same
snapshot acquisition cover that too.)
The test case added here does not fail before v15, because that
assertion was added in 277692220 which was not back-patched.
However, the race condition the assertion warns of surely
exists further back, so back-patch to all supported branches.
Per report from Richard Guo.
Discussion: https://postgr.es/m/CAMbWs4-x26=_QxxgdJyNbiCDzvtr2WV5ZDso_v-CukKEe6cBZw@mail.gmail.com
If SIGTERM is received within this section, the startup process
will immediately proc_exit() in the signal handler, so it is
inadvisable to include any more code than is required there (as
such code is unlikely to be compatible with doing proc_exit() in a
signal handler). This commit moves the code recently added to this
section (see 1b06d7bac9 and 7fed801135) to outside of the section.
This ensures that the startup process only calls proc_exit() in its
SIGTERM handler for the duration of the system() call, which is how
this code worked from v8.4 to v14.
Reported-by: Michael Paquier, Thomas Munro
Analyzed-by: Andres Freund
Suggested-by: Tom Lane
Reviewed-by: Michael Paquier, Robert Haas, Thomas Munro, Andres Freund
Discussion: https://postgr.es/m/Y9nGDSgIm83FHcad%40paquier.xyz
Discussion: https://postgr.es/m/20230223231503.GA743455%40nathanxps13
Backpatch-through: 15
The old documentation encourages entering single-user mode for no
reason, which is a bad plan in most cases. Instead, discourage users
from doing that, and explain the limited cases in which it may be
desirable.
The old documentation claims that running VACUUM as anyone but the
superuser can't possibly work, which is not really true, because it
might be that some other user has enough permissions to VACUUM all
the tables that matter. Weaken the language just a bit.
The old documentation claims that you can't run any commands
when near XID exhaustion, which is false because you can still
run commands that don't require an XID, like a SELECT without a
locking clause.
The old documentation doesn't clearly explain that it's a good idea
to get rid of prepared transactons, long-running transactions, and
replication slots that are preventing (M)XID horizon advancement.
Spell out the steps to do that.
Also, discourage the use of VACUUM FULL and VACUUM FREEZE in
this type of scenario.
Back-patch to v14. Much of this is good advice on all supported
versions, but before 60f1f09ff4
the chances of VACUUM failing in multi-user mode were much higher.
Alexander Alekseev, John Naylor, Robert Haas, reviewed at various
times by Peter Geoghegan, Hannu Krosing, and Andres Freund.
Discussion: http://postgr.es/m/CA+TgmoYtsUDrzaHcmjFhLzTk1VEv29mO_u-MT+XWHrBJ_4nD8A@mail.gmail.com
Some of our src/bin tools read the control file without any kind of
interlocking against concurrent writes from the server. At least ext4
and ntfs can expose partially modified contents when you do that.
For now, we'll try to tolerate this by retrying up to 10 times if the
checksum doesn't match, until we get two reads in a row with the same
bad checksum. This is not guaranteed to reach the right conclusion, but
it seems very likely to. Thanks to Tom Lane for this suggestion.
Various ideas for interlocking or atomicity were considered too
complicated, unportable or expensive given the lack of field reports,
but remain open for future reconsideration.
Back-patch as far as 12. It doesn't seem like a good idea to put a
heuristic change for a very rare problem into the final release of 11.
Reviewed-by: Anton A. Melnikov <aamelnikov@inbox.ru>
Reviewed-by: David Steele <david@pgmasters.net>
Reviewed-by: Michael Paquier <michael@paquier.xyz>
Discussion: https://postgr.es/m/20221123014224.xisi44byq3cf5psi%40awork3.anarazel.de
Commit dc7d70ea added functions that read the control file, but didn't
acquire ControlFileLock. With unlucky timing, file systems that have
weak interlocking like ext4 and ntfs could expose partially overwritten
contents, and the checksum would fail.
Back-patch to all supported releases.
Reviewed-by: David Steele <david@pgmasters.net>
Reviewed-by: Anton A. Melnikov <aamelnikov@inbox.ru>
Reviewed-by: Michael Paquier <michael@paquier.xyz>
Discussion: https://postgr.es/m/20221123014224.xisi44byq3cf5psi%40awork3.anarazel.de
Michael Paquier