no-longer-needed pages at the end of a table. We thought we could throw away
pages containing HEAPTUPLE_DEAD tuples; but this is not so, because such
tuples very likely have index entries pointing at them, and we wouldn't have
removed the index entries. The problem only emerges in a somewhat unlikely
race condition: the dead tuples have to have been inserted by a transaction
that later aborted, and this has to have happened between VACUUM's initial
scan of the page and then rechecking it for empty in count_nondeletable_pages.
But that timespan will include an index-cleaning pass, so it's not all that
hard to hit. This seems to explain a couple of previously unsolved bug
reports.
Per report from bitsandbytes88 <at> hotmail.com and subsequent discussion.
This is a back patch of a patch committed yesterday to CLUSTER and REINDEX.
REINDEX only processes user indexes as of 8.1, so we needn't backpatch this any
further. (CLUSTER was backpatched separately all the way back to 7.4).
remote sessions, instead of erroring out in the middle of the operation.
This is a backpatch of a previous fix applied to CLUSTER to HEAD and 8.2, all
the way back that it is relevant to.
recover from elog(ERROR). Problem was created by introduction of hash seq
search tracking awhile back, and affects all branches that have bgwriter;
in HEAD the disease has snuck into autovacuum and walwriter too. (Not sure
that the latter two use hash_seq_search at the moment, but surely they might
someday.) Per report from Sergey Koposov.
an exclusive lock on the table at this point, which we want to release as soon
as possible. This is called in the phase of lazy vacuum where we truncate the
empty pages at the end of the table.
An alternative solution would be to lower the vacuum delay settings before
starting the truncating phase, but this doesn't work very well in autovacuum
due to the autobalancing code (which can cause other processes to change our
cost delay settings). This case could be considered in the balancing code, but
it is simpler this way.
read from the temp file didn't match the file length reported by ftello(),
the wrong variable's value was printed, and so the message made no sense.
Clean up a couple other coding infelicities while at it.
relcache entry after having heap_close'd it. This could lead to misbehavior
if a relcache flush wiped out the cache entry meanwhile. In 8.2 there is a
very real risk of CREATE INDEX CONCURRENTLY using the wrong relid for locking
and waiting purposes. I think the bug is only cosmetic in 8.0 and 8.1,
because their transgression is limited to using RelationGetRelationName(rel)
in an ereport message immediately after heap_close, and there's no way (except
with special debugging options) for a cache flush to occur in that interval.
Not quite sure that it's cosmetic in 7.4, but seems best to patch anyway.
Found by trying to run the regression tests with CLOBBER_CACHE_ALWAYS enabled.
Maybe we should try to do that on a regular basis --- it's awfully slow,
but perhaps some fast buildfarm machine could do it once in awhile.
padded encryption scheme. Formerly it would try to access res[(unsigned) -1],
which resulted in core dumps on 64-bit machines, and was certainly trouble
waiting to happen on 32-bit machines (though in at least the known case
it was harmless because that byte would be overwritten after return).
Per report from Ken Colson; fix by Marko Kreen.
byte after the last full byte of the bit array, regardless of whether that
byte was part of the valid data or not. Found by buildfarm testing.
Thanks to Stefan Kaltenbrunner for nailing down the cause.
row within one query: we were firing check triggers before all the updates
were done, leading to bogus failures. Fix by making the triggers queued by
an RI update go at the end of the outer query's trigger event list, thereby
effectively making the processing "breadth-first". This was indeed how it
worked pre-8.0, so the bug does not occur in the 7.x branches.
Per report from Pavel Stehule.
hash table is allocated in a child context of the agg node's memory
context, MemoryContextReset() will reset but *not* delete the child
context. Since ExecReScanAgg() proceeds to build a new hash table
from scratch (in a new sub-context), this results in leaking the
header for the previous memory context. Therefore, use
MemoryContextResetAndDeleteChildren() instead.
Credit: My colleague Sailesh Krishnamurthy at Truviso for isolating
the cause of the leak.
on Windows. This is yet another manifestation of the problem that Windows
returns time zone names that may be in a different encoding than we are using.
I've put a better solution in HEAD, but the back branches need a simple patch.
Per report from Hiroshi Saito.
clauses in which one side or the other references both sides of the join
cannot be removed as redundant, because that expression won't have been
constrained below the join. Per report from Sergey Burladyan.
log_min_error_statement is active and there is some problem in logging the
current query string; for example, that it's too long to include in the log
message without running out of memory. This problem has existed since the
log_min_error_statement feature was introduced. No doubt the reason it
wasn't detected long ago is that 8.2 is the first release that defaults
log_min_error_statement to less than PANIC level.
Per report from Bill Moran.
truncated relation was deleted later in the WAL sequence. Since replay
normally auto-creates a relation upon its first reference by a WAL log entry,
failure is seen only if the truncate entry happens to be the first reference
after the checkpoint we're restarting from; which is a pretty unusual case but
of course not impossible. Fix by making truncate entries auto-create like
the other ones do. Per report and test case from Dharmendra Goyal.
when handed an invalidly-encoded pattern. The previous coding could get
into an infinite loop if pg_mb2wchar_with_len() returned a zero-length
string after we'd tested for nonempty pattern; which is exactly what it
will do if the string consists only of an incomplete multibyte character.
This led to either an out-of-memory error or a backend crash depending
on platform. Per report from Wiktor Wodecki.
a MIN or MAX aggregate call into an indexscan: the initplan is being made at
the current query nesting level and so we shouldn't increment query_level.
Though usually harmless, this mistake could lead to bogus "plan should not
reference subplan's variable" failures on complex queries. Per bug report
from David Sanchez i Gregori.
referencing table does not change the tuple's FK column(s), we don't bother
to check the PK table since the constraint was presumably already valid.
However, the check is still necessary if the tuple was inserted by our own
transaction, since in that case the INSERT trigger will conclude it need not
make the check (since its version of the tuple has been deleted). We got this
right for simple cases, but not when the insert and update are in different
subtransactions of the current top-level transaction; in such cases the FK
check would never be made at all. (Hence, problem dates back to 8.0 when
subtransactions were added --- it's actually the subtransaction version of a
bug fixed in 7.3.5.) Fix, and add regression test cases. Report and fix by
Affan Salman.
been broken since forever, but was not noticed because people seldom look
at raw parse trees. AFAIK, no impact on users except that debug_print_parse
might fail; but patch it all the way back anyway. Per report from Jeff Ross.
to prevent possible escalation of privilege. Provide new SECURITY
DEFINER functions with old behavior, but initially REVOKE ALL
from public for these functions. Per list discussion and design
proposed by Tom Lane.
we don't know at that point which relation OID to tell pgstat to forget.
The code was passing the relfilenode, which is incorrect, and could possibly
cause some other relation's stats to be zeroed out. While we could try to
clean this up, it seems much simpler and more reliable to let the next
invocation of pgstat_vacuum_tabstat() fix things; which indeed is how it
worked before I introduced the buggy code into 8.1.3 and later :-(.
Problem noticed by Itagaki Takahiro, fix is per subsequent discussion.
This is a Linux kernel bug that apparently exists in every extant kernel
version: sometimes shmctl() will fail with EIDRM when EINVAL is correct.
We were assuming that EIDRM indicates a possible conflict with pre-existing
backends, and refusing to start the postmaster when this happens. Fortunately,
there does not seem to be any case where Linux can legitimately return EIDRM
(it doesn't track shmem segments in a way that would allow that), so we can
get away with just assuming that EIDRM means EINVAL on this platform.
Per reports from Michael Fuhr and Jon Lapham --- it's a bit surprising
we have not seen more reports, actually.
overruns (neither of which seem likely to be exploitable as security holes,
fortunately, since the provoker can't control the data written). One of
these is due to choosing to stomp on the output of a called function, which
is bad news in any case; make it treat the called functions' results as
read-only. Avoid some unnecessary palloc/pfree traffic too; it's not
really helpful to free small temporary objects, and again this is presuming
more than it ought to about the nature of the results of called functions.
Per report from Patrick Welche and additional code-reading by Imad.
The correct test for defined-ness is SvOK(sv), not anything involving
SvTYPE. Per bug #3415 from Matt Taylor.
Back-patch as far as 8.0; no apparent problem in 7.x.
create table foo (bar int default null default 3);
due to not thinking about the special-case handling of DEFAULT NULL.
Problem noticed while investigating bug #3396.
a buffer containing a non-zeroed page. This seems appropriate now that the
8.2 fix has been seen to save at least one user from data loss due to a
buggy kernel (per report from Jaime Silvela 7-May-07). I'd go further
back than 8.1, except that the 8.0-to-8.1 bufmgr changes are large
enough that the patch doesn't work immediately; I'm hesitant to make a
change without more extensive analysis than I have time for now.
reassembled in the syslogger before writing to the log file. This prevents
partial messages from being written, which mucks up log rotation, and
messages from different backends being interleaved, which causes garbled
logs. Backport as far as 8.0, where the syslogger was introduced.
Tom Lane and Andrew Dunstan
large inputs. Also cause it to error out immediately if the result will
overflow, instead of grinding through a lot of calculation first.
Per gripe from Jim Nasby.
Tom Lane