Alvaro Herrera [Mon, 5 Mar 2018 22:37:19 +0000 (19:37 -0300)]
Clone extended stats in CREATE TABLE (LIKE INCLUDING ALL)
The LIKE INCLUDING ALL clause to CREATE TABLE intuitively indicates
cloning of extended statistics on the source table, but it failed to do
so. Patch it up so that it does. Also include an INCLUDING STATISTICS
option to the LIKE clause, so that the behavior can be requested
individually, or excluded individually.
While at it, reorder the INCLUDING options, both in code and in docs, in
alphabetical order which makes more sense than feature-implementation
order that was previously used.
Backpatch this to Postgres 10, where extended statistics were
introduced, because this is seen as an oversight in a fresh feature
which is better to get consistent from the get-go instead of changing
only in pg11.
In pg11, comments on statistics objects are cloned too. In pg10 they
are not, because I (Álvaro) was too coward to change the parse node as
required to support it. Also, in pg10 I chose not to renumber the
parser symbols for the various INCLUDING options in LIKE, for the same
reason. Any corresponding user-visible changes (docs) are backpatched,
though.
Reported-by: Stephen Froehlich
Author: David Rowley Reviewed-by: Álvaro Herrera, Tomas Vondra
Discussion: https://postgr.es/m/CY1PR0601MB1927315B45667A1B679D0FD5E5EF0@CY1PR0601MB1927.namprd06.prod.outlook.com
Fujii Masao [Mon, 5 Mar 2018 17:08:18 +0000 (02:08 +0900)]
Fix pg_rewind to handle relation data files in tablespaces properly.
pg_rewind checks whether each file is a relation data file, from its path.
Previously this check logic had the bug which made pg_rewind fail to
recognize any relation data files in tablespaces. Which also caused
an assertion failure in pg_rewind.
Tom Lane [Sun, 4 Mar 2018 01:31:35 +0000 (20:31 -0500)]
Fix assorted issues in convert_to_scalar().
If convert_to_scalar is passed a pair of datatypes it can't cope with,
its former behavior was just to elog(ERROR). While this is OK so far as
the core code is concerned, there's extension code that would like to use
scalarltsel/scalargtsel/etc as selectivity estimators for operators that
work on non-core datatypes, and this behavior is a show-stopper for that
use-case. If we simply allow convert_to_scalar to return FALSE instead of
outright failing, then the main logic of scalarltsel/scalargtsel will work
fine for any operator that behaves like a scalar inequality comparison.
The lack of conversion capability will mean that we can't estimate to
better than histogram-bin-width precision, since the code will effectively
assume that the comparison constant falls at the middle of its bin. But
that's still a lot better than nothing. (Someday we should provide a way
for extension code to supply a custom version of convert_to_scalar, but
today is not that day.)
While poking at this issue, we noted that the existing code for handling
type bytea in convert_to_scalar is several bricks shy of a load.
It assumes without checking that if the comparison value is type bytea,
the bounds values are too; in the worst case this could lead to a crash.
It also fails to detoast the input values, so that the comparison result is
complete garbage if any input is toasted out-of-line, compressed, or even
just short-header. I'm not sure how often such cases actually occur ---
the bounds values, at least, are probably safe since they are elements of
an array and hence can't be toasted. But that doesn't make this code OK.
Back-patch to all supported branches, partly because author requested that,
but mostly because of the bytea bugs. The change in API for the exposed
routine convert_network_to_scalar() is theoretically a back-patch hazard,
but it seems pretty unlikely that any third-party code is calling that
function directly.
Tom Lane [Fri, 2 Mar 2018 22:40:48 +0000 (17:40 -0500)]
Fix VM buffer pin management in heap_lock_updated_tuple_rec().
Sloppy coding in this function could lead to leaking a VM buffer pin,
or to attempting to free the same pin twice. Repair. While at it,
reduce the code's tendency to free and reacquire the same page pin.
Back-patch to 9.6; before that, this routine did not concern itself
with VM pages.
Tom Lane [Fri, 2 Mar 2018 16:22:42 +0000 (11:22 -0500)]
Make gistvacuumcleanup() count the actual number of index tuples.
Previously, it just returned the heap tuple count, which might be only an
estimate, and would be completely the wrong thing if the index is partial.
Since this function scans every index page anyway to find free pages,
it's practically free to count the surviving index tuples. Let's do that
and return an accurate count.
This is easily visible as a wrong reltuples value for a partial GiST
index following VACUUM, so back-patch to all supported branches.
Tom Lane [Thu, 1 Mar 2018 21:23:30 +0000 (16:23 -0500)]
Use ereport not elog for some corrupt-HOT-chain reports.
These errors have been seen in the field in corrupted-data situations.
It seems worthwhile to report them with ERRCODE_DATA_CORRUPTED, rather
than the generic ERRCODE_INTERNAL_ERROR, for the benefit of log monitoring
and tools like amcheck. However, use errmsg_internal so that the text
strings still aren't translated; it seems unlikely to be worth
translators' time to do so.
Back-patch to 9.3, like the predecessor commit d70cf811f that introduced
these elog calls originally (replacing Asserts).
Alvaro Herrera [Thu, 1 Mar 2018 21:07:46 +0000 (18:07 -0300)]
Relax overly strict sanity check for upgraded ancient databases
Commit 4800f16a7ad0 added some sanity checks to ensure we don't
accidentally corrupt data, but in one of them we failed to consider the
effects of a database upgraded from 9.2 or earlier, where a tuple
exclusively locked prior to the upgrade has a slightly different bit
pattern. Fix that by using the macro that we fixed in commit 74ebba84aeb6 for similar situations.
Reported-by: Alexandre Garcia Reviewed-by: Andres Freund
Discussion: https://postgr.es/m/CAPYLKR6yxV4=pfW0Gwij7aPNiiPx+3ib4USVYnbuQdUtmkMaEA@mail.gmail.com
Andres suspects that this bug may have wider ranging consequences, but I
couldn't find anything.
Tom Lane [Thu, 1 Mar 2018 20:35:03 +0000 (15:35 -0500)]
Fix IOS planning when only some index columns can return an attribute.
Since 9.5, it's possible that some but not all columns of an index
support returning the indexed value for index-only scans. If the
same indexed column appears in index columns that behave both ways,
check_index_only() supposed that it'd be OK to do an index-only scan
testing that column; but that fails if we have to recheck the indexed
condition on one of the columns that doesn't support this.
In principle we could make this work by remapping the recheck expressions
to pull the value from a column that does support returning the indexed
value. But such cases are so weird and rare that, at least for now,
it doesn't seem worth the trouble. Instead, just teach check_index_only
that a value is returnable only if all the index columns containing it
are returnable, rather than any of them.
Per report from David Pereiro Lagares. Back-patch to 9.5 where the
possibility of this situation appeared.
Tom Lane [Wed, 28 Feb 2018 23:33:45 +0000 (18:33 -0500)]
Rename base64 routines to avoid conflict with Solaris built-in functions.
Solaris 11.4 has built-in functions named b64_encode and b64_decode.
Rename ours to something else to avoid the conflict (fortunately,
ours are static so the impact is limited).
One could wish for less duplication of code in this area, but that
would be a larger patch and not very suitable for back-patching.
Since this is a portability fix, we want to put it into all supported
branches.
Report and initial patch by Rainer Orth, reviewed and adjusted a bit
by Michael Paquier
Tom Lane [Wed, 28 Feb 2018 21:57:37 +0000 (16:57 -0500)]
Remove restriction on SQL block length in isolationtester scanner.
specscanner.l had a fixed limit of 1024 bytes on the length of
individual SQL stanzas in an isolation test script. People are
starting to run into that, so fix it by making the buffer resizable.
Once we allow this in HEAD, it seems inevitable that somebody will
try to back-patch a test that exceeds the old limit, so back-patch
this change as a preventive measure.
Tom Lane [Tue, 27 Feb 2018 21:46:52 +0000 (16:46 -0500)]
Fix up ecpg's configuration so it handles "long long int" in MSVC builds.
Although configure-based builds correctly define HAVE_LONG_LONG_INT when
appropriate (in both pg_config.h and ecpg_config.h), builds using the MSVC
scripts failed to do so. This currently has no impact on the backend,
since it uses that symbol nowhere; but it does prevent ecpg from
supporting "long long int". Fix that.
Also, adjust Solution.pm so that in the constructed ecpg_config.h file,
the "#if (_MSC_VER > 1200)" covers only the LONG_LONG_INT-related
#defines, not the whole file. AFAICS this was a thinko on somebody's
part: ENABLE_THREAD_SAFETY should always be defined in Windows builds,
and in branches using USE_INTEGER_DATETIMES, the setting of that shouldn't
depend on the compiler version either. If I'm wrong, I imagine the
buildfarm will say so.
Per bug #15080 from Jonathan Allen; issue diagnosed by Michael Meskes
and Andrew Gierth. Back-patch to all supported branches.
Tom Lane [Tue, 27 Feb 2018 20:56:51 +0000 (15:56 -0500)]
Use the correct tuplestore read pointer in a NamedTuplestoreScan.
Tom Kazimiers reported that transition tables don't work correctly when
they are scanned by more than one executor node. That's because commit 18ce3a4ab allocated separate read pointers for each executor node, as it
must, but failed to make them active at the appropriate times. Repair.
Tom Lane [Tue, 27 Feb 2018 20:04:21 +0000 (15:04 -0500)]
Remove regression tests' CREATE FUNCTION commands for unused C functions.
I removed these functions altogether in HEAD, in commit db3af9feb, and
it emerges that that causes trouble for cross-branch upgrade testing.
We could put back stub functions but that seems pretty silly. Instead,
back-patch a minimal subset of db3af9feb, namely just removing the
CREATE FUNCTION commands.
Tom Lane [Tue, 27 Feb 2018 18:27:38 +0000 (13:27 -0500)]
Prevent dangling-pointer access when update trigger returns old tuple.
A before-update row trigger may choose to return the "new" or "old" tuple
unmodified. ExecBRUpdateTriggers failed to consider the second
possibility, and would proceed to free the "old" tuple even if it was the
one returned, leading to subsequent access to already-deallocated memory.
In debug builds this reliably leads to an "invalid memory alloc request
size" failure; in production builds it might accidentally work, but data
corruption is also possible.
This is a very old bug. There are probably a couple of reasons it hasn't
been noticed up to now. It would be more usual to return NULL if one
wanted to suppress the update action; returning "old" is significantly less
efficient since the update will occur anyway. Also, none of the standard
PLs would ever cause this because they all returned freshly-manufactured
tuples even if they were just copying "old". But commit 4b93f5799 changed
that for plpgsql, making it possible to see the bug with a plpgsql trigger.
Still, this is certainly legal behavior for a trigger function, so it's
ExecBRUpdateTriggers's fault not plpgsql's.
It seems worth creating a test case that exercises returning "old" directly
with a C-language trigger; testing this through plpgsql seems unreliable
because its behavior might change again.
Report and fix by Rushabh Lathia; regression test case by me.
Back-patch to all supported branches.
Noah Misch [Mon, 26 Feb 2018 15:39:44 +0000 (07:39 -0800)]
Document security implications of search_path and the public schema.
The ability to create like-named objects in different schemas opens up
the potential for users to change the behavior of other users' queries,
maliciously or accidentally. When you connect to a PostgreSQL server,
you should remove from your search_path any schema for which a user
other than yourself or superusers holds the CREATE privilege. If you do
not, other users holding CREATE privilege can redefine the behavior of
your commands, causing them to perform arbitrary SQL statements under
your identity. "SET search_path = ..." and "SELECT
pg_catalog.set_config(...)" are not vulnerable to such hijacking, so one
can use either as the first command of a session. As special
exceptions, the following client applications behave as documented
regardless of search_path settings and schema privileges: clusterdb
createdb createlang createuser dropdb droplang dropuser ecpg (not
programs it generates) initdb oid2name pg_archivecleanup pg_basebackup
pg_config pg_controldata pg_ctl pg_dump pg_dumpall pg_isready
pg_receivewal pg_recvlogical pg_resetwal pg_restore pg_rewind pg_standby
pg_test_fsync pg_test_timing pg_upgrade pg_waldump reindexdb vacuumdb
vacuumlo. Not included are core client programs that run user-specified
SQL commands, namely psql and pgbench. PostgreSQL encourages non-core
client applications to do likewise.
Document this in the context of libpq connections, psql connections,
dblink connections, ECPG connections, extension packaging, and schema
usage patterns. The principal defense for applications is "SELECT
pg_catalog.set_config('search_path', '', false)", and the principal
defense for databases is "REVOKE CREATE ON SCHEMA public FROM PUBLIC".
Either one is sufficient to prevent attack. After a REVOKE, consider
auditing the public schema for objects named like pg_catalog objects.
Authors of SECURITY DEFINER functions use some of the same defenses, and
the CREATE FUNCTION reference page already covered them thoroughly.
This is a good opportunity to audit SECURITY DEFINER functions for
robust security practice.
Back-patch to 9.3 (all supported versions).
Reviewed by Michael Paquier and Jonathan S. Katz. Reported by Arseniy
Sharoglazov.
Noah Misch [Mon, 26 Feb 2018 15:39:44 +0000 (07:39 -0800)]
Empty search_path in Autovacuum and non-psql/pgbench clients.
This makes the client programs behave as documented regardless of the
connect-time search_path and regardless of user-created objects. Today,
a malicious user with CREATE permission on a search_path schema can take
control of certain of these clients' queries and invoke arbitrary SQL
functions under the client identity, often a superuser. This is
exploitable in the default configuration, where all users have CREATE
privilege on schema "public".
This changes behavior of user-defined code stored in the database, like
pg_index.indexprs and pg_extension_config_dump(). If they reach code
bearing unqualified names, "does not exist" or "no schema has been
selected to create in" errors might appear. Users may fix such errors
by schema-qualifying affected names. After upgrading, consider watching
server logs for these errors.
The --table arguments of src/bin/scripts clients have been lax; for
example, "vacuumdb -Zt pg_am\;CHECKPOINT" performed a checkpoint. That
now fails, but for now, "vacuumdb -Zt 'pg_am(amname);CHECKPOINT'" still
performs a checkpoint.
Back-patch to 9.3 (all supported versions).
Reviewed by Tom Lane, though this fix strategy was not his first choice.
Reported by Arseniy Sharoglazov.
Tom Lane [Mon, 26 Feb 2018 15:18:22 +0000 (10:18 -0500)]
Avoid using unsafe search_path settings during dump and restore.
Historically, pg_dump has "set search_path = foo, pg_catalog" when
dumping an object in schema "foo", and has also caused that setting
to be used while restoring the object. This is problematic because
functions and operators in schema "foo" could capture references meant
to refer to pg_catalog entries, both in the queries issued by pg_dump
and those issued during the subsequent restore run. That could
result in dump/restore misbehavior, or in privilege escalation if a
nefarious user installs trojan-horse functions or operators.
This patch changes pg_dump so that it does not change the search_path
dynamically. The emitted restore script sets the search_path to what
was used at dump time, and then leaves it alone thereafter. Created
objects are placed in the correct schema, regardless of the active
search_path, by dint of schema-qualifying their names in the CREATE
commands, as well as in subsequent ALTER and ALTER-like commands.
Since this change requires a change in the behavior of pg_restore
when processing an archive file made according to this new convention,
bump the archive file version number; old versions of pg_restore will
therefore refuse to process files made with new versions of pg_dump.
Peter Eisentraut [Sat, 24 Feb 2018 02:17:57 +0000 (21:17 -0500)]
Fix filtering of unsupported relations in logical replication
In the pgoutput plugin, skip changes for relations that are not
publishable, per is_publishable_class(). This concerns in particular
materialized views and information_schema tables. While those relations
cannot be part of a publication, per existing checks, they will be
considered by a FOR ALL TABLES publication. A subscription would not
actually apply changes for those relations, again per existing checks,
but trying to match incoming changes to local tables on the subscriber
would lead to errors if no matching local table exists. Skipping those
changes on the publisher avoids sending useless changes and eliminates
the error.
Bug: #15044 Reported-by: Chad Trabant <chad@iris.washington.edu> Reviewed-by: Petr Jelinek <petr.jelinek@2ndquadrant.com>
Tom Lane [Fri, 23 Feb 2018 19:38:19 +0000 (14:38 -0500)]
Allow auto_explain.log_min_duration to go up to INT_MAX.
The previous limit of INT_MAX / 1000 seems to have been cargo-culted in
from somewhere else. Or possibly the value was converted to microseconds
at some point; but in all supported releases, it's just compared to other
values, so there's no need for the restriction. This change raises the
effective limit from ~35 minutes to ~24 days, which conceivably is useful
to somebody, and anyway it's more consistent with the range of the core
log_min_duration_statement GUC.
Per complaint from Kevin Bloch. Back-patch to all supported releases.
Tom Lane [Fri, 23 Feb 2018 18:47:33 +0000 (13:47 -0500)]
Fix planner failures with overlapping mergejoin clauses in an outer join.
Given overlapping or partially redundant join clauses, for example
t1 JOIN t2 ON t1.a = t2.x AND t1.b = t2.x
the planner's EquivalenceClass machinery will ordinarily refactor the
clauses as "t1.a = t1.b AND t1.a = t2.x", so that join processing doesn't
see multiple references to the same EquivalenceClass in a list of join
equality clauses. However, if the join is outer, it's incorrect to derive
a restriction clause on the outer side from the join conditions, so the
clause refactoring does not happen and we end up with overlapping join
conditions. The code that attempted to deal with such cases had several
subtle bugs, which could result in "left and right pathkeys do not match in
mergejoin" or "outer pathkeys do not match mergeclauses" planner errors,
if the selected join plan type was a mergejoin. (It does not appear that
any actually incorrect plan could have been emitted.)
The core of the problem really was failure to recognize that the outer and
inner relations' pathkeys have different relationships to the mergeclause
list. A join's mergeclause list is constructed by reference to the outer
pathkeys, so it will always be ordered the same as the outer pathkeys, but
this cannot be presumed true for the inner pathkeys. If the inner sides of
the mergeclauses contain multiple references to the same EquivalenceClass
({t2.x} in the above example) then a simplistic rendering of the required
inner sort order is like "ORDER BY t2.x, t2.x", but the pathkey machinery
recognizes that the second sort column is redundant and throws it away.
The mergejoin planning code failed to account for that behavior properly.
One error was to try to generate cut-down versions of the mergeclause list
from cut-down versions of the inner pathkeys in the same way as the initial
construction of the mergeclause list from the outer pathkeys was done; this
could lead to choosing a mergeclause list that fails to match the outer
pathkeys. The other problem was that the pathkey cross-checking code in
create_mergejoin_plan treated the inner and outer pathkey lists
identically, whereas actually the expectations for them must be different.
That led to false "pathkeys do not match" failures in some cases, and in
principle could have led to failure to detect bogus plans in other cases,
though there is no indication that such bogus plans could be generated.
Reported by Alexander Kuzmenkov, who also reviewed this patch. This has
been broken for years (back to around 8.3 according to my testing), so
back-patch to all supported branches.
Tom Lane [Wed, 21 Feb 2018 23:40:24 +0000 (18:40 -0500)]
Repair pg_upgrade's failure to preserve relfrozenxid for matviews.
This oversight led to data corruption in matviews, manifesting as
"could not access status of transaction" before our most recent releases,
and "found xmin from before relfrozenxid" errors since then.
The proximate cause of the problem seems to have been confusion between
the task of preserving dropped-column status and the task of preserving
frozenxid status. Those are required for distinct sets of relkinds,
and the reasoning was entirely undocumented in the source code. In hopes
of forestalling future errors of the same kind, try to improve the
commentary in this area.
In passing, also improve the remarkably unhelpful comments around
pg_upgrade's set_frozenxids(). That's not actually buggy AFAICS,
but good luck figuring out what it does from the old comments.
Per report from Claudio Freire. It appears that bug #14852 from Alexey
Ermakov is an earlier report of the same issue, and there may be other
cases that we failed to identify at the time.
Patch by me based on analysis by Andres Freund. The bug dates back
to the introduction of matviews, so back-patch to all supported branches.
Tom Lane [Tue, 20 Feb 2018 16:23:34 +0000 (11:23 -0500)]
Fix pg_dump's logic for eliding sequence limits that match the defaults.
The previous coding here applied atoi() to strings that could represent
values too large to fit in an int. If the overflowed value happened to
match one of the cases it was looking for, it would drop that limit
value from the output, leading to incorrect restoration of the sequence.
Avoid the unsafe behavior, and also make the logic cleaner by explicitly
calculating the default min/max values for the appropriate kind of
sequence.
Reported and patched by Alexey Bashtanov, though I whacked his patch
around a bit. Back-patch to v10 where the faulty logic was added.
Tom Lane [Mon, 19 Feb 2018 21:00:18 +0000 (16:00 -0500)]
Fix misbehavior of CTE-used-in-a-subplan during EPQ rechecks.
An updating query that reads a CTE within an InitPlan or SubPlan could get
incorrect results if it updates rows that are concurrently being modified.
This is caused by CteScanNext supposing that nothing inside its recursive
ExecProcNode call could change which read pointer is selected in the CTE's
shared tuplestore. While that's normally true because of scoping
considerations, it can break down if an EPQ plan tree gets built during the
call, because EvalPlanQualStart builds execution trees for all subplans
whether they're going to be used during the recheck or not. And it seems
like a pretty shaky assumption anyway, so let's just reselect our own read
pointer here.
Per bug #14870 from Andrei Gorita. This has been broken since CTEs were
implemented, so back-patch to all supported branches.
Tom Lane [Thu, 15 Feb 2018 18:56:38 +0000 (13:56 -0500)]
Doc: fix minor bug in CREATE TABLE example.
One example in create_table.sgml claimed to be showing table constraint
syntax, but it was really column constraint syntax due to the omission
of a comma. This is both wrong and confusing, so fix it in all
supported branches.
Tom Lane [Wed, 14 Feb 2018 19:47:18 +0000 (14:47 -0500)]
Fix broken logic for reporting PL/Python function names in errcontext.
plpython_error_callback() reported the name of the function associated
with the topmost PL/Python execution context. This was not merely
wrong if there were nested PL/Python contexts, but it risked a core
dump if the topmost one is an inline code block rather than a named
function. That will have proname = NULL, and so we were passing a NULL
pointer to snprintf("%s"). It seems that none of the PL/Python-testing
machines in the buildfarm will dump core for that, but some platforms do,
as reported by Marina Polyakova.
Investigation finds that there actually is an existing regression test
that used to prove that the behavior was wrong, though apparently no one
had noticed that it was printing the wrong function name. It stopped
showing the problem in 9.6 when we adjusted psql to not print CONTEXT
by default for NOTICE messages. The problem is masked (if your platform
avoids the core dump) in error cases, because PL/Python will throw away
the originally generated error info in favor of a new traceback produced
at the outer level.
Repair by using ErrorContextCallback.arg to pass the correct context to
the error callback. Add a regression test illustrating correct behavior.
Back-patch to all supported branches, since they're all broken this way.
Tom Lane [Sun, 11 Feb 2018 18:24:15 +0000 (13:24 -0500)]
Fix assorted errors in pg_dump's handling of extended statistics objects.
pg_dump supposed that a stats object necessarily shares the same schema
as its underlying table, and that it doesn't have a separate owner.
These things may have been true during early development of the feature,
but they are not true as of v10 release.
Failure to track the object's schema separately turns out to have only
limited consequences, because pg_get_statisticsobjdef() always schema-
qualifies the target object name in the generated CREATE STATISTICS command
(a decision out of step with the rest of ruleutils.c, but I digress).
Therefore the restored object would be in the right schema, so that the
only problem is that the TOC entry would be mislabeled as to schema. That
could lead to wrong decisions for schema-selective restores, for example.
The ownership issue is a bit more serious: not only was the TOC entry
potentially mislabeled as to owner, but pg_dump didn't bother to issue an
ALTER OWNER command at all, so that after restore the stats object would
continue to be owned by the restoring superuser.
A final point is that decisions as to whether to dump a stats object or
not were driven by whether the underlying table was dumped or not. While
that's not wrong on its face, it won't scale nicely to the planned future
extension to cross-table statistics. Moreover, that design decision comes
out of the view of stats objects as being auxiliary to a particular table,
like a rule or trigger, which is exactly where the above problems came
from. Since we're now treating stats objects more like independent objects
in their own right, they ought to behave like standalone objects for this
purpose too. So change to using the generic selectDumpableObject() logic
for them (which presently amounts to "dump if containing schema is to be
dumped").
Along the way to fixing this, restructure so that getExtendedStatistics
collects the identity info (only) for all extended stats objects in one
query, and then for each object actually being dumped, we retrieve the
definition in dumpStatisticsExt. This is necessary to ensure that
schema-qualification in the generated CREATE STATISTICS command happens
with respect to the search path that pg_dump will now be using at restore
time (ie, the schema the stats object is in, not that of the underlying
table). It's probably also significantly faster in the typical scenario
where only a minority of tables have extended stats.
Back-patch to v10 where extended stats were introduced.
Magnus Hagander [Wed, 7 Feb 2018 10:03:55 +0000 (11:03 +0100)]
Change default git repo URL to https
Since we now support the server side handler for git over https (so
we're no longer using the "dumb protocol"), make https the primary
choice for cloning the repository, and the git protocol the secondary
choice.
In passing, also change the links to git-scm.com from http to https.
Reviewed by Stefan Kaltenbrunner and David G. Johnston
Tom Lane [Mon, 5 Feb 2018 15:58:27 +0000 (10:58 -0500)]
Ensure that all temp files made during pg_upgrade are non-world-readable.
pg_upgrade has always attempted to ensure that the transient dump files
it creates are inaccessible except to the owner. However, refactoring
in commit 76a7650c4 broke that for the file containing "pg_dumpall -g"
output; since then, that file was protected according to the process's
default umask. Since that file may contain role passwords (hopefully
encrypted, but passwords nonetheless), this is a particularly unfortunate
oversight. Prudent users of pg_upgrade on multiuser systems would
probably run it under a umask tight enough that the issue is moot, but
perhaps some users are depending only on pg_upgrade's umask changes to
protect their data.
To fix this in a future-proof way, let's just tighten the umask at
process start. There are no files pg_upgrade needs to write at a
weaker security level; and if there were, transiently relaxing the
umask around where they're created would be a safer approach.
Report and patch by Tom Lane; the idea for the fix is due to Noah Misch.
Back-patch to all supported branches.
Tom Lane [Mon, 5 Feb 2018 15:37:30 +0000 (10:37 -0500)]
Fix RelationBuildPartitionKey's processing of partition key expressions.
Failure to advance the list pointer while reading partition expressions
from a list results in invoking an input function with inappropriate data,
possibly leading to crashes or, with carefully crafted input, disclosure
of arbitrary backend memory.
Bug discovered independently by Álvaro Herrera and David Rowley.
This patch is by Álvaro but owes something to David's proposed fix.
Back-patch to v10 where the issue was introduced.
Tom Lane [Fri, 2 Feb 2018 23:26:07 +0000 (18:26 -0500)]
Be more wary about shm_toc_lookup failure.
Commit 445dbd82a basically missed the point of commit d46633506,
which was that we shouldn't allow shm_toc_lookup() failure to lead
to a core dump or assertion crash, because the odds of such a
failure should never be considered negligible. It's correct that
we can't expect the PARALLEL_KEY_ERROR_QUEUE TOC entry to be there
if we have no workers. But if we have no workers, we're not going
to do anything in this function with the lookup result anyway,
so let's just skip it. That lets the code use the easy-to-prove-safe
noError=false case, rather than anything requiring effort to review.
Investigation of 2d2d06b7e27e3177d5bef0061801c75946871db3 revealed that
identity values were not applied in some further cases, including
logical replication subscribers, VALUES RTEs, and ALTER TABLE ... ADD
COLUMN. To fix all that, apply the identity column expression in
build_column_default() instead of repeating the same logic at each call
site.
For ALTER TABLE ... ADD COLUMN ... IDENTITY, the previous coding
completely ignored that existing rows for the new column should have
values filled in from the identity sequence. The coding using
build_column_default() fails for this because the sequence ownership
isn't registered until after ALTER TABLE, and we can't do it before
because we don't have the column in the catalog yet. So we specially
remember in ColumnDef the sequence name that we decided on and build a
custom NextValueExpr using that.
Reviewed-by: Michael Paquier <michael.paquier@gmail.com>
Bruce Momjian [Wed, 31 Jan 2018 21:43:32 +0000 (16:43 -0500)]
doc: Improve pg_upgrade rsync examples to use clusterdir
Commit 9521ce4a7a1125385fb4de9689f345db594c516a from Sep 13, 2017 and
backpatched through 9.5 used rsync examples with datadir. The reporter
has pointed out, and testing has verified, that clusterdir must be used,
so update the docs accordingly.
Reported-by: Don Seiler
Discussion: https://postgr.es/m/CAHJZqBD0u9dCERpYzK6BkRv=663AmH==DFJpVC=M4Xg_rq2=CQ@mail.gmail.com
Bruce Momjian [Wed, 31 Jan 2018 21:25:21 +0000 (16:25 -0500)]
doc: mention datadir locations are actually config locations
Technically, pg_upgrade's --old-datadir and --new-datadir are
configuration directories, not necessarily data directories. This is
reflected in the 'postgres' manual page, so do the same for pg_upgrade.
Robert Haas [Wed, 31 Jan 2018 20:43:11 +0000 (15:43 -0500)]
Fix list partition constraints for partition keys of array type.
The old code generated always generated a constraint of the form
col = ANY(ARRAY[val1, val2, ...]), but that's invalid when col is an
array type. Instead, generate col = val when there's only one value,
col = val1 OR col = val2 OR ... when there are multiple values and
col is of array type, and the old form when there are multiple values
and col is not of an array type.
As a side benefit, this makes constraint exclusion able to prune
a list partition declared to accept a single Boolean value, which
didn't work before.
Robert Haas [Tue, 30 Jan 2018 19:27:38 +0000 (14:27 -0500)]
Fix test case for 'outer pathkeys do not match mergeclauses' fix.
Commit 4bbf6edfbd5d03743ff82dda2f00c738fb3208f5 added a test case,
but it turns out that the test case doesn't reliably test for the
bug, and in the context of the regression test suite did not because
ANALYZE had not been run.
Report and patch by Etsuro Fujita. I added a comment along lines
previously suggested by Tom Lane.
Andres Freund [Mon, 29 Jan 2018 19:02:09 +0000 (11:02 -0800)]
Prevent growth of simplehash tables when they're "too empty".
In cases where simplehash tables where filled with either a lot of
conflicting hash-values, or values that hash to consecutive
values (i.e. build "chains") the growth heuristics in d4c62a6b623d6eef88218158e9fa3cf974c6c7e5 could trigger rather
explosively.
To fix that, address some of the reasons (see previous commit) of why
the growth heuristics where needed, and only allow growth when the
table isn't too empty. While that means there's a few cases of bad
input that can be slower, that seems a lot better than running very
quickly out of memory.
Author: Tomas Vondra and Andres Freund, with additional input by
Thomas Munro, Tom Lane Todd A. Cook Reported-By: Todd A. Cook, Tomas Vondra, Thomas Munro
Discussion: https://postgr.es/m/20171127185700.1470.20362@wrigleys.postgresql.org
Backpatch: 10, where simplehash was introduced
Andres Freund [Mon, 29 Jan 2018 19:02:09 +0000 (11:02 -0800)]
Improve bit perturbation in TupleHashTableHash.
The changes in b81b5a96f424531b97cdd1dba97d9d1b9c9d372e did not fully
address the issue, because the bit-mixing of the IV into the final
hash-key didn't prevent clustering in the input-data survive in the
output data.
This didn't cause a lot of problems because of the additional growth
conditions added d4c62a6b623d6eef88218158e9fa3cf974c6c7e5. But as we
want to rein those in due to explosive growth in some edges, this
needs to be fixed.
Author: Andres Freund
Discussion: https://postgr.es/m/20171127185700.1470.20362@wrigleys.postgresql.org
Backpatch: 10, where simplehash was introduced
Tom Lane [Sun, 28 Jan 2018 18:39:07 +0000 (13:39 -0500)]
Add stack-overflow guards in set-operation planning.
create_plan_recurse lacked any stack depth check. This is not per
our normal coding rules, but I'd supposed it was safe because earlier
planner processing is more complex and presumably should eat more
stack. But bug #15033 from Andrew Grossman shows this isn't true,
at least not for queries having the form of a many-thousand-way
INTERSECT stack.
Further testing showed that recurse_set_operations is also capable
of being crashed in this way, since it likewise will recurse to the
bottom of a parsetree before calling any support functions that
might themselves contain any stack checks. However, its stack
consumption is only perhaps a third of create_plan_recurse's.
It's possible that this particular problem with create_plan_recurse can
only manifest in 9.6 and later, since before that we didn't build a Path
tree for set operations. But having seen this example, I now have no
faith in the proposition that create_plan_recurse doesn't need a stack
check, so back-patch to all supported branches.
Tom Lane [Sat, 27 Jan 2018 21:42:28 +0000 (16:42 -0500)]
Update time zone data files to tzdata release 2018c.
DST law changes in Brazil, Sao Tome and Principe. Historical corrections
for Bolivia, Japan, and South Sudan. The "US/Pacific-New" zone has been
removed (it was only a link to America/Los_Angeles anyway).
Tom Lane [Sat, 27 Jan 2018 18:52:24 +0000 (13:52 -0500)]
Avoid crash during EvalPlanQual recheck of an inner indexscan.
Commit 09529a70b changed nodeIndexscan.c and nodeIndexonlyscan.c to
postpone initialization of the indexscan proper until the first tuple
fetch. It overlooked the question of mark/restore behavior, which means
that if some caller attempts to mark the scan before the first tuple fetch,
you get a null pointer dereference.
The only existing user of mark/restore is nodeMergejoin.c, which (somewhat
accidentally) will never attempt to set a mark before the first inner tuple
unless the inner child node is a Material node. Hence the case can't arise
normally, so it seems sufficient to document the assumption at both ends.
However, during an EvalPlanQual recheck, ExecScanFetch doesn't call
IndexNext but just returns the jammed-in test tuple. Therefore, if we're
doing a recheck in a plan tree with a mergejoin with inner indexscan,
it's possible to reach ExecIndexMarkPos with iss_ScanDesc still null,
as reported by Guo Xiang Tan in bug #15032.
Really, when there's a test tuple supplied during an EPQ recheck, touching
the index at all is the wrong thing: rather, the behavior of mark/restore
ought to amount to saving and restoring the es_epqScanDone flag. We can
avoid finding a place to actually save the flag, for the moment, because
given the assumption that no caller will set a mark before fetching a
tuple, es_epqScanDone must always be set by the time we try to mark.
So the actual behavior change required is just to not reach the index
access if a test tuple is supplied.
The set of plan node types that need to consider this issue are those
that support EPQ test tuples (i.e., call ExecScan()) and also support
mark/restore; which is to say, IndexScan, IndexOnlyScan, and perhaps
CustomScan. It's tempting to try to fix the problem in one place by
teaching ExecMarkPos() itself about EPQ; but ExecMarkPos supports some
plan types that aren't Scans, and also it seems risky to make assumptions
about what a CustomScan wants to do here. Also, the most likely future
change here is to decide that we do need to support marks placed before
the first tuple, which would require additional work in IndexScan and
IndexOnlyScan in any case. Hence, fix the EPQ issue in nodeIndexscan.c
and nodeIndexonlyscan.c, accepting the small amount of code duplicated
thereby, and leave it to CustomScan providers to fix this bug if they
have it.
Back-patch to v10 where commit 09529a70b came in. In earlier branches,
the index_markpos() call is a waste of cycles when EPQ is active, but
no more than that, so it doesn't seem appropriate to back-patch further.
Tom Lane [Tue, 23 Jan 2018 21:50:34 +0000 (16:50 -0500)]
Teach reparameterize_path() to handle AppendPaths.
If we're inside a lateral subquery, there may be no unparameterized paths
for a particular child relation of an appendrel, in which case we *must*
be able to create similarly-parameterized paths for each other child
relation, else the planner will fail with "could not devise a query plan
for the given query". This means that there are situations where we'd
better be able to reparameterize at least one path for each child.
This calls into question the assumption in reparameterize_path() that
it can just punt if it feels like it. However, the only case that is
known broken right now is where the child is itself an appendrel so that
all its paths are AppendPaths. (I think possibly I disregarded that in
the original coding on the theory that nested appendrels would get folded
together --- but that only happens *after* reparameterize_path(), so it's
not excused from handling a child AppendPath.) Given that this code's been
like this since 9.3 when LATERAL was introduced, it seems likely we'd have
heard of other cases by now if there were a larger problem.
Per report from Elvis Pranskevichus. Back-patch to 9.3.
Robert Haas [Tue, 23 Jan 2018 16:13:42 +0000 (11:13 -0500)]
Report an ERROR if a parallel worker fails to start properly.
Commit 28724fd90d2f85a0573a8107b48abad062a86d83 fixed things so that
if a background worker fails to start due to fork() failure or because
it is terminated before startup succeeds, BGWH_STOPPED will be
reported. However, that only helps if the code that uses the
background worker machinery notices the change in status, and the code
in parallel.c did not.
To fix that, do two things. First, make sure that when a worker
exits, it triggers the leader to read from error queues. That way, if
a worker which has attached to an error queue exits uncleanly, the
leader is sure to throw some error, either the contents of the
ErrorResponse sent by the worker, or "lost connection to parallel
worker" if it exited without sending one. To cover the case where
the worker never starts up in the first place or exits before
attaching to the error queue, the ParallelContext now keeps track
of which workers have sent at least one message via the error
queue. A worker which sends no messages by the time the parallel
operation finishes will be checked to see whether it exited before
attaching to the error queue; if so, a new error message, "parallel
worker failed to initialize", will be reported. If not, we'll
continue to wait until it either starts up and exits cleanly, starts
up and exits uncleanly, or fails to start, and then take the
appropriate action.
Tom Lane [Mon, 22 Jan 2018 17:06:18 +0000 (12:06 -0500)]
Make pg_dump's ACL, sec label, and comment entries reliably identifiable.
_tocEntryRequired() expects that it can identify ACL, SECURITY LABEL,
and COMMENT TOC entries that are for large objects by seeing whether
the tag for them starts with "LARGE OBJECT ". While that works fine
for actual large objects, which are indeed tagged that way, it's
subject to false positives unless every such entry's tag starts with an
appropriate type ID. And in fact it does not work for ACLs, because
up to now we customarily tagged those entries with just the bare name
of the object. This means that an ACL for an object named
"LARGE OBJECT something" would be misclassified as data not schema,
with undesirable results in a schema-only or data-only dump ---
although pg_upgrade seems unaffected, due to the special case for
binary-upgrade mode further down in _tocEntryRequired().
We can fix this by changing all the dumpACL calls to use the label
strings already in use for comments and security labels, which do
follow the convention of starting with an object type indicator.
Well, mostly they follow it. dumpDatabase() got it wrong, using
just the bare database name for those purposes, so that a database
named "LARGE OBJECT something" would similarly be subject to having
its comment or security label dropped or included when not wanted.
Bring that into line too. (Note that up to now, database ACLs have
not been processed by pg_dump, so that this issue doesn't affect them.)
_tocEntryRequired() itself is not free of fault: it was overly liberal
about matching object tags to "LARGE OBJECT " in binary-upgrade mode.
This looks like it is probably harmless because there would be no data
component to strip anyway in that mode, but at best it's trouble
waiting to happen, so tighten that up too.
The possible misclassification of SECURITY LABEL entries for databases is
in principle a security problem, but the opportunities for actual exploits
seem too narrow to be interesting. The other cases seem like just bugs,
since an object owner can change its ACL or comment for himself, he needn't
try to trick someone else into doing it by choosing a strange name.
This has been broken since per-large-object TOC entries were introduced
in 9.0, so back-patch to all supported branches.
Bruce Momjian [Sun, 21 Jan 2018 02:47:02 +0000 (21:47 -0500)]
doc: update intermediate certificate instructions
Document how to properly create root and intermediate certificates using
v3_ca extensions and where to place intermediate certificates so they
are properly transferred to the remote side with the leaf certificate to
link to the remote root certificate. This corrects docs that used to
say that intermediate certificates must be stored with the root
certificate.
Also add instructions on how to create root, intermediate, and leaf
certificates.
Alvaro Herrera [Fri, 19 Jan 2018 13:15:08 +0000 (10:15 -0300)]
Fix StoreCatalogInheritance1 to use 32bit inhseqno
For no apparent reason, this function was using a 16bit-wide inhseqno
value, rather than the correct 32 bit width which is what is stored in
the pg_inherits catalog. This becomes evident if you try to create a
table with more than 65535 parents, because this error appears:
Needless to say, having so many parents is an uncommon situations, which
explains why this error has never been reported despite being having
been introduced with the Postgres95 1.01 sources in commit d31084e9d111:
https://git.postgresql.org/gitweb/?p=postgresql.git;a=blob;f=src/backend/commands/creatinh.c;hb=d31084e9d111#l349
Backpatch all the way back.
David Rowley noticed this while reviewing a patch of mine.
Discussion: https://postgr.es/m/CAKJS1f8Dn7swSEhOWwzZzssW7747YB=2Hi+T7uGud40dur69-g@mail.gmail.com
Tom Lane [Thu, 18 Jan 2018 16:09:44 +0000 (11:09 -0500)]
Extend configure's __int128 test to check for a known gcc bug.
On Sparc64, use of __attribute__(aligned(8)) with __int128 causes faulty
code generation in gcc versions at least through 5.5.0. We can work around
that by disabling use of __int128, so teach configure to test for the bug.
This solution doesn't fix things for the case of cross-compiling with a
buggy compiler; to support that nicely, we'd need to add a manual disable
switch. Unless more such cases turn up, it doesn't seem worth the work.
Affected users could always edit pg_config.h manually.
In passing, fix some typos in the existing configure test for __int128.
They're harmless because we only compile that code not run it, but
they're still confusing for anyone looking at it closely.
This is needed in support of commit 751804998, so back-patch to 9.5
as that was.
Robert Haas [Wed, 17 Jan 2018 21:18:39 +0000 (16:18 -0500)]
postgres_fdw: Avoid 'outer pathkeys do not match mergeclauses' error.
When pushing down a join to a foreign server, postgres_fdw constructs
an alternative plan to be used for any EvalPlanQual rechecks that
prove to be necessary. This plan is stored as the outer subplan of
the Foreign Scan implementing the pushed-down join. Previously, this
alternative plan could have a different nominal sort ordering than its
parent, which seemed OK since there will only be one tuple per base
table anyway in the case of an EvalPlanQual recheck. Actually,
though, it caused a problem if that path was used as a building block
for the EvalPlanQual recheck plan of a higher-level foreign join,
because we could end up with a merge join one of whose inputs was not
labelled with the correct sort order. Repair by injecting an extra
Sort node into the EvalPlanQual recheck plan whenever it would
otherwise fail to be sorted at least as well as its parent Foreign
Scan.
Report by Jeff Janes. Patch by me, reviewed by Tom Lane, who also
provided the test case and comment text.
Tom Lane [Fri, 12 Jan 2018 21:52:49 +0000 (16:52 -0500)]
Fix postgres_fdw to cope with duplicate GROUP BY entries.
Commit 7012b132d, which added the ability to push down aggregates and
grouping to the remote server, wasn't careful to ensure that the remote
server would have the same idea we do about which columns are the grouping
columns, in cases where there are textually identical GROUP BY expressions.
Such cases typically led to "targetlist item has multiple sortgroupref
labels" errors.
To fix this reliably, switch over to using "GROUP BY column-number" syntax
rather than "GROUP BY expression" in transmitted queries, and adjust
foreign_grouping_ok() to be more careful about duplicating the sortgroupref
labeling of the local pathtarget.
Per bug #14890 from Sean Johnston. Back-patch to v10 where the buggy code
was introduced.
Tom Lane [Fri, 12 Jan 2018 20:46:37 +0000 (15:46 -0500)]
Avoid unnecessary failure in SELECT concurrent with ALTER NO INHERIT.
If a query against an inheritance tree runs concurrently with an ALTER
TABLE that's disinheriting one of the tree members, it's possible to get
a "could not find inherited attribute" error because after obtaining lock
on the removed member, make_inh_translation_list sees that its columns
have attinhcount=0 and decides they aren't the columns it's looking for.
An ideal fix, perhaps, would avoid including such a just-removed member
table in the query at all; but there seems no way to accomplish that
without adding expensive catalog rechecks or creating a likelihood of
deadlocks. Instead, let's just drop the check on attinhcount. In this
way, a query that's included a just-disinherited child will still
succeed, which is not a completely unreasonable behavior.
This problem has existed for a long time, so back-patch to all supported
branches. Also add an isolation test verifying related behaviors.
Patch by me; the new isolation test is based on Kyotaro Horiguchi's work.
Tom Lane [Fri, 12 Jan 2018 17:24:50 +0000 (12:24 -0500)]
Fix incorrect handling of subquery pullup in the presence of grouping sets.
If we flatten a subquery whose target list contains constants or
expressions, when those output columns are used in GROUPING SET columns,
the planner was capable of doing the wrong thing by merging a pulled-up
expression into the surrounding expression during const-simplification.
Then the late processing that attempts to match subexpressions to grouping
sets would fail to match those subexpressions to grouping sets, with the
effect that they'd not go to null when expected.
To fix, wrap such subquery outputs in PlaceHolderVars, ensuring that
they preserve their separate identity throughout the planner's expression
processing. This is a bit of a band-aid, because the wrapper defeats
const-simplification even in places where it would be safe to allow.
But a nicer fix would likely be too invasive to back-patch, and the
consequences of the missed optimizations probably aren't large in most
cases.
Back-patch to 9.5 where grouping sets were introduced.
Heikki Linnakangas, with small mods and better test cases by me;
additional review by Andrew Gierth
Teodor Sigaev [Thu, 11 Jan 2018 11:42:16 +0000 (14:42 +0300)]
Fix behavior of ~> (cube, int) operator
~> (cube, int) operator was especially designed for knn-gist search.
However, it appears that knn-gist search can't work correctly with current
behavior of this operator when dataset contains cubes of variable
dimensionality. In this case, the same value of second operator argument
can point to different dimension depending on dimensionality of particular cube.
Such behavior is incompatible with gist indexing of cubes, and knn-gist doesn't
work correctly for it.
This patch changes behavior of ~> (cube, int) operator by introducing dimension
numbering where value of second argument unambiguously identifies number of
dimension. With new behavior, this operator can be correctly supported by
knn-gist. Relevant changes to cube operator class are also included.
Backpatch to v9.6 where operator was introduced.
Since behavior of ~> (cube, int) operator is changed, depending entities
must be refreshed after upgrade. Such as, expression indexes using this
operator must be reindexed, materialized views must be rebuilt, stored
procedures and client code must be revised to correctly use new behavior.
That should be mentioned in release notes.
Noticed by: Tomas Vondra
Author: Alexander Korotkov
Reviewed by: Tomas Vondra, Andrey Borodin
Discussion: https://www.postgresql.org/message-id/flat/a9657f6a-b497-36ff-e56-482a2c7e3292@2ndquadrant.com
Tom Lane [Wed, 10 Jan 2018 22:13:29 +0000 (17:13 -0500)]
Fix sample INSTR() functions in the plpgsql documentation.
These functions are stated to be Oracle-compatible, but they weren't.
Yugo Nagata noticed that while our code returns zero for a zero or
negative fourth parameter (occur_index), Oracle throws an error.
Further testing by me showed that there was also a discrepancy in the
interpretation of a negative third parameter (beg_index): Oracle thinks
that a negative beg_index indicates the last place where the target
substring can *begin*, whereas our code thinks it is the last place
where the target can *end*.
Adjust the sample code to behave like Oracle in both these respects.
Also change it to be a CDATA[] section, simplifying copying-and-pasting
out of the documentation source file. And fix minor problems in the
introductory comment, which wasn't very complete or accurate.
Back-patch to all supported branches. Although this patch only touches
documentation, we should probably call it out as a bug fix in the next
minor release notes, since users who have adopted the functions will
likely want to update their versions.
Tom Lane [Wed, 10 Jan 2018 20:50:54 +0000 (15:50 -0500)]
Remove dubious micro-optimization in ckpt_buforder_comparator().
It seems incorrect to assume that the list of CkptSortItems can never
contain duplicate page numbers: concurrent activity could result in some
page getting dropped from a low-numbered buffer and later loaded into a
high-numbered buffer while BufferSync is scanning the buffer pool.
If that happened, the comparator would give self-inconsistent results,
potentially confusing qsort(). Saving one comparison step is not worth
possibly getting the sort wrong.
So far as I can tell, nothing would actually go wrong given our current
implementation of qsort(). It might get a bit slower than expected
if there were a large number of duplicates of one value, but that's
surely a probability-epsilon case. Still, the comment is wrong,
and if we ever switched to another sort implementation it might be
less forgiving.
In passing, avoid casting away const-ness of the argument pointers;
I've not seen any compiler complaints from that, but it seems likely
that some compilers would not like it.
Back-patch to 9.6 where this code came in, just in case I've underestimated
the possible consequences.
Alvaro Herrera [Tue, 9 Jan 2018 18:54:39 +0000 (15:54 -0300)]
Change some bogus PageGetLSN calls to BufferGetLSNAtomic
As src/backend/access/transam/README says, PageGetLSN may only be called
by processes holding either exclusive lock on buffer, or a shared lock
on buffer plus buffer header lock. Therefore any place that only holds
a shared buffer lock must use BufferGetLSNAtomic instead of PageGetLSN,
which internally obtains buffer header lock prior to reading the LSN.
A few callsites failed to comply with this rule. This was detected by
running all tests under a new (not committed) assertion that verifies
PageGetLSN locking contract. All but one of the callsites that failed
the assertion are fixed by this patch. Remaining callsites were
inspected manually and determined not to need any change.
The exception (unfixed callsite) is in TestForOldSnapshot, which only
has a Page argument, making it impossible to access the corresponding
Buffer from it. Fixing that seems a much larger patch that will have to
be done separately; and that's just as well, since it was only
introduced in 9.6 and other bugs are much older.
Some of these bugs are ancient; backpatch all the way back to 9.3.
Tom Lane [Tue, 9 Jan 2018 17:34:46 +0000 (12:34 -0500)]
While waiting for a condition variable, detect postmaster death.
The general assumption for postmaster child processes is that they
should just exit(1), reasonably promptly, if the postmaster disappears.
condition_variable.c neglected this consideration and could be left
waiting forever, if the counterpart process it is waiting for has
done the right thing and exited.
We had some discussion of adjusting the WaitEventSet API to make it
harder to make this type of mistake in future; but for the moment,
and for v10, let's make this narrow fix.
Tom Lane [Tue, 9 Jan 2018 17:09:30 +0000 (12:09 -0500)]
Fix race condition during replication origin drop.
replorigin_drop() misunderstood the API for condition variables: it
had ConditionVariablePrepareToSleep and ConditionVariableCancelSleep
inside its test-and-sleep loop, rather than outside the loop as
intended. The net effect is a narrow race-condition window wherein,
if the process using a replication slot releases it immediately after
replorigin_drop() releases the ReplicationOriginLock, replorigin_drop()
would get into the condition variable's wait list too late and then
wait indefinitely for a signal that won't come.
Because there's a different CV for each replication slot, we can't
just move the ConditionVariablePrepareToSleep call to above the
test-and-sleep loop. What we can do, in the wake of commit 13db3b936,
is drop the ConditionVariablePrepareToSleep call entirely. This fix
depends on that commit because (at least in principle) the slot matching
the target replication origin might move around, so that once in a blue
moon successive loop iterations might involve different CVs. We can now
cope with such a scenario, at the cost of an extra trip through the
retry loop.
(There are ways we could fix this bug without depending on that commit,
but they're all a lot more complicated than this way.)
While at it, upgrade the rather skimpy comments in this function.
Tom Lane [Tue, 9 Jan 2018 16:39:10 +0000 (11:39 -0500)]
Allow ConditionVariable[PrepareTo]Sleep to auto-switch between CVs.
The original coding here insisted that callers manually cancel any prepared
sleep for one condition variable before starting a sleep on another one.
While that's not a huge burden today, it seems like a gotcha that will bite
us in future if the use of condition variables increases; anything we can
do to make the use of this API simpler and more robust is attractive.
Hence, allow these functions to automatically switch their attention to
a different CV when required. This is safe for the same reason it was OK
for commit aced5a92b to let a broadcast operation cancel any prepared CV
sleep: whenever we return to the other test-and-sleep loop, we will
automatically re-prepare that CV, paying at most an extra test of that
loop's exit condition.
Back-patch to v10 where condition variables were introduced. Ordinarily
we would probably not back-patch a change like this, but since it does not
invalidate any coding pattern that was legal before, it seems safe enough.
Furthermore, there's an open bug in replorigin_drop() for which the
simplest fix requires this. Even if we chose to fix that in some more
complicated way, the hazard would remain that we might back-patch some
other bug fix that requires this behavior.
Bruce Momjian [Tue, 9 Jan 2018 03:43:51 +0000 (22:43 -0500)]
pg_upgrade: prevent check on live cluster from generating error
Previously an inaccurate but harmless error was generated when running
--check on a live server before reporting the servers as compatible.
The fix is to split error reporting and exit control in the exec_prog()
API.
Reported-by: Daniel Westermann
Backpatch-through: 10
Tom Lane [Sat, 6 Jan 2018 00:42:49 +0000 (19:42 -0500)]
Reorder steps in ConditionVariablePrepareToSleep for more safety.
In the admittedly-very-unlikely case that AddWaitEventToSet fails,
ConditionVariablePrepareToSleep would error out after already having
set cv_sleep_target, which is probably bad, and after having already
set cv_wait_event_set, which is very bad. Transaction abort might or
might not clean up cv_sleep_target properly; but there is nothing
that would be aware that the WaitEventSet wasn't fully constructed,
so that all future condition variable sleeps would be broken.
We can easily guard against these hazards with slight restructuring.
Back-patch to v10 where condition_variable.c was introduced.
Tom Lane [Sat, 6 Jan 2018 00:21:30 +0000 (19:21 -0500)]
Rewrite ConditionVariableBroadcast() to avoid live-lock.
The original implementation of ConditionVariableBroadcast was, per its
self-description, "the dumbest way possible". Thomas Munro found out
it was a bit too dumb. An awakened process may immediately re-queue
itself, if the specific condition it's waiting for is not yet satisfied.
If this happens before ConditionVariableBroadcast is able to see the wait
queue as empty, then ConditionVariableBroadcast will re-awaken the same
process, repeating the cycle. Given unlucky timing this back-and-forth
can repeat indefinitely; loops lasting thousands of seconds have been
seen in testing.
To fix, add our own process to the end of the wait queue to serve as a
sentinel, and exit the broadcast loop once our process is not there
anymore. There are various special considerations described in the
comments, the principal disadvantage being that wakers can no longer
be sure whether they awakened a real waiter or just a sentinel. But in
practice nobody pays attention to the result of ConditionVariableSignal
or ConditionVariableBroadcast anyway, so that problem seems hypothetical.
Back-patch to v10 where condition_variable.c was introduced.
Alvaro Herrera [Fri, 5 Jan 2018 15:17:10 +0000 (12:17 -0300)]
Fix failure to delete spill files of aborted transactions
Logical decoding's reorderbuffer.c may spill transaction files to disk
when transactions are large. These are supposed to be removed when they
become "too old" by xid; but file removal requires the boundary LSNs of
the transaction to be known. The final_lsn is only set when we see the
commit or abort record for the transaction, but nothing sets the value
for transactions that crash, so the removal code misbehaves -- in
assertion-enabled builds, it crashes by a failed assertion.
To fix, modify the final_lsn of transactions that don't have a value
set, to the LSN of the very latest change in the transaction. This
causes the spilled files to be removed appropriately.
projects / postgresql / log