> revision-id: alexey.kopytov@sun.com-20100824103548-ikm79qlfrvggyj9h
> parent: sunny.bains@oracle.com-20100816001222-xqc447tr6jwh8c53
> committer: Alexey Kopytov <Alexey.Kopytov@Sun.com>
> branch nick: 5.1-security
> timestamp: Tue 2010-08-24 14:35:48 +0400
> message:
> Bug #55568: user variable assignments crash server when used
> within query
>
> The server could crash after materializing a derived table
> which requires a temporary table for grouping.
>
> When destroying the temporary table used to execute a query for
> a derived table, JOIN::destroy() did not clean up Item_fields
> pointing to fields in the temporary table. This led to
> dereferencing a dangling pointer when printing out the items
> tree later in the outer SELECT.
>
> The solution is an addendum to the patch for bug37362: in
> addition to cleaning up items in tmp_all_fields3, do the same
> for items in tmp_all_fields1, since now we have an example
> where this is necessary.
greedy_search optimizer_search_depth=0
The algorithm inside restore_prev_nj_state failed to
properly update the counters within the NESTED_JOIN
tree. The counter was decremented each time a table in the
node was removed from the QEP, the correct thing to do being
only to decrement it when the last table in the child node
was removed from the plan. This lead to node counters
getting negative values and the plan thus appeared
impossible. An assertion caught this.
Fixed by not recursing up the tree unless the last table in
the join nest node is removed from the plan
The problem was in an incorrect debug assertion. The expression
used in the failing assertion states that when finding
references matching ORDER BY expressions, there can be only one
reference to a single table. But that does not make any sense,
all test cases for this bug are valid examples with multiple
identical WHERE expressions referencing the same table which
are also present in the ORDER BY list.
Fixed by removing the failing assertion. We also have to take
care of the 'found' counter so that we count multiple
references only once. We rely on this fact later in
eq_ref_table().
The crash is the result of an attempt made by JOIN::optimize to evaluate
the WHERE condition when no records have been actually read.
The fix is to remove erroneous 'outer_join' variable check.
The crash happens because greedy_serach
can not determine best plan due to
wrong inner table dependences. These
dependences affects join table sorting
which performs before greedy_search starting.
In our case table which has real 'no dependences'
should be put on top of the list but it does not
happen as inner tables have no dependences as well.
The fix is to exclude RAND_TABLE_BIT mask from
condition which checks if table dependences
should be updated.
column is used for ORDER BY
Problem: filesort isn't meant for null length sort data
(e.g. char(0)), that leads to a server crash.
Fix: disregard sort order if sort data record length is 0 (nothing
to sort).
on re-execution of prepared statement
Problem: some (see eq_ref_table()) ORDER BY/GROUP BY optimization
is called before each PS execution. However, we don't properly
initialize its stucture every time before the call.
Fix: properly initialize the sturture used.
int join_read_key(JOIN_TAB*)
The eq_ref access method TABLE_REF (accessed through
JOIN_TAB) to save state and to track if this is the
first row it finds or not.
This state was not reset on subquery re-execution
causing an assert.
Fixed by resetting the state before the subquery
re-execution.
Part 2 :
There was a special optimization on the ref access method for
ORDER BY ... DESC that was set without actually looking on the type of the
selected index for ORDER BY.
Fixed the SELECT ... ORDER BY .. DESC (it uses a different code path compared
to the ASC that has been fixed with the previous fix).
field='const1' AND field='const2' in some cases
Building multiple equality predicates containing
a constant which is compared as a datetime (with a field)
we should take this fact into account and compare the
constant with another possible constatns as datetimes
as well.
E.g. for the
SELECT ... WHERE a='2001-01-01' AND a='2001-01-01 00:00:00'
we should compare '2001-01-01' with '2001-01-01 00:00:00' as
datetimes but not as strings.
memory
The server was doing a bad class typecast causing setting of
wrong value for the maximum number of items in an internal
structure used in equality propagation.
Fixed by not doing the wrong typecast and asserting the type
of the Item where it should be done.
values
We should re-set the access method functions when changing the access
method when switching to another index to avoid sorting.
Fixed by doing a little re-engineering : encapsulating all the function
assignment into a special function and calling it when flipping the
indexes.
only const tables
The problem was caused by two shortcuts in the optimizer that
are inapplicable in the ROLLUP case.
Normally in a case when only const tables are involved in a
query, DISTINCT clause can be safely optimized away since there
may be only one row produced by the join. Similarly, we don't
need to create a temporary table to resolve DISTINCT/GROUP
BY/ORDER BY. Both of these are inapplicable when the WITH
ROLLUP modifier is present.
Fixed by disabling the said optimizations for the WITH ROLLUP
case.
Bug#41756 "Strange error messages about locks from InnoDB".
In JT_EQ_REF (join_read_key()) access method,
don't try to unlock rows in the handler, unless certain that
a) they were locked
b) they are not used.
Unlocking of rows is done by the logic of the nested join loop,
and is unaware of the possible caching that the access method may
have. This could lead to double unlocking, when a row
was unlocked first after reading into the cache, and then
when taken from cache, as well as to unlocking of rows which
were actually used (but taken from cache).
Delegate part of the unlocking logic to the access method,
and in JT_EQ_REF count how many times a record was actually
used in the join. Unlock it only if it's usage count is 0.
Implemented review comments.
with temporary tables
There were two problems the test case from this bug was
triggering:
1. JOIN::rollup_init() was supposed to wrap all constant Items
into another object for queries with the WITH ROLLUP modifier
to ensure they are never considered as constants and therefore
are written into temporary tables if the optimizer chooses to
employ them for DISTINCT/GROUP BY handling.
However, JOIN::rollup_init() was called before
make_join_statistics(), so Items corresponding to fields in
const tables could not be handled as intended, which was
causing all kinds of problems later in the query execution. In
particular, create_tmp_table() assumed all constant items
except "hidden" ones to be removed earlier by remove_const()
which led to improperly initialized Field objects for the
temporary table being created. This is what was causing crashes
and valgrind errors in storage engines.
2. Even when the above problem had been fixed, the query from
the test case produced incorrect results due to some
DISTINCT/GROUP BY optimizations being performed by the
optimizer that are inapplicable in the WITH ROLLUP case.
Fixed by disabling inapplicable DISTINCT/GROUP BY optimizations
when the WITH ROLLUP modifier is present, and splitting the
const-wrapping part of JOIN::rollup_init() into a separate
method which is now invoked after make_join_statistics() when
the const tables are already known.
subquery returning multiple rows
Error handling was missing when handling subqueires in WHERE
and when assigning a SELECT result to a @variable.
This caused crash(es).
Fixed by adding error handling code to both the WHERE
condition evaluation and to assignment to an @variable.
having clause...
The fix for bug 46184 was not very complete. It was not covering
views using temporary tables and multiple tables in a FROM clause.
Fixed by reverting the fix for 46184 and making a more general
check that is checking at the right execution stage and for all
of the non-supported cases.
Now PROCEDURE ANALYZE on non-top level SELECT is also forbidden.
Updated the analyse.test and subselect.test accordingly.
Queries with nested outer joins may lead to crashes or
bad results because an internal data structure is not handled
correctly.
The optimizer uses bitmaps of nested JOINs to determine
if certain table can be placed at a certain place in the
JOIN order.
It does maintain a bitmap describing in which JOINs
last placed table is nested.
When it puts a table it makes sure the bit of every JOIN that
contains the table in question is set (because JOINs can be nested).
It does that by recursively setting the bit for the next enclosing
JOIN when this is the first table in the JOIN and recursively
resetting the bit if it's the last table in the JOIN.
When it removes a table from the join order it should do the
opposite : recursively unset the bit if it's the only remaining
table in this join and and recursively set the bit if it's removing
the last table of a JOIN.
There was an error in how the bits was set for the upper levels :
when removing a table it was setting the bit for all the enclosing
nested JOINs even if there were more tables left in the current JOIN
(which practically means that the upper nested JOINs were not affected).
Fixed by stopping the recursion at the relevant level.
line 138 when forcing a spatial index
Problem: "Spatial indexes can be involved in the search
for queries that use a function such as MBRContains()
or MBRWithin() in the WHERE clause".
Using spatial indexes for JOINs with =, <=> etc.
predicates is incorrect.
Fix: disable spatial indexes for such queries.
query
The fix for bug 46749 removed the check for OUTER_REF_TABLE_BIT
and substituted it for a check on the presence of
Item_ident::depended_from.
Removing it altogether was wrong : OUTER_REF_TABLE_BIT should
still be checked in addition to depended_from (because it's not
set in all cases and doesn't contradict to the check of depended_from).
Fixed by returning the old condition back as a compliment to the
new one.
function,file sql_base.cc
When uncacheable queries are written to a temp table the optimizer must
preserve the original JOIN structure, because it is re-using the JOIN
structure to read from the resulting temporary table.
This was done only for uncacheable sub-queries.
But top level queries can also benefit from this mechanism, specially if
they're using index access and need a reset.
Fixed by not limiting the saving of JOIN structure to subqueries
exclusively.
Added a new test file to extend the existing (large) subquery.test.
field references
This error requires a combination of factors :
1. An "impossible where" in the outermost SELECT
2. An aggregate in the outermost SELECT
3. A correlated subquery with a WHERE clause that includes an outer
field reference as a top level WHERE sargable predicate
When JOIN::optimize detects an "impossible WHERE" it will bail out
without doing the rest of the work and initializations. It will not
call make_join_statistics() as well. And make_join_statistics fills
in various structures for each table referenced.
When processing the result of the "impossible WHERE" the query must
send a single row of data if there are aggregate functions in it.
In this case the server marks all the aggregates as having received
no rows and calls the relevant Item::val_xxx() method on the SELECT
list. However if this SELECT list happens to contain a correlated
subquery this subquery is evaluated in a normal evaluation mode.
And if this correlated subquery has a reference to a field from the
outermost "impossible where" SELECT the add_key_fields will mistakenly
consider the outer field reference as a "local" field reference when
looking for sargable predicates.
But since the SELECT where the outer field reference refers to is not
completely initialized due to the "impossible WHERE" in this level
we'll get a NULL pointer reference.
Fixed by making a better condition for discovering if a field is "local"
to the SELECT level being processed.
It's not enough to look for OUTER_REF_TABLE_BIT in this case since
for outer references to constant tables the Item_field::used_tables()
will return 0 regardless of whether the field reference is from the
local SELECT or not.
In create_myisam_from_heap() mark all errors as fatal except
HA_ERR_RECORD_FILE_FULL for a HEAP table.
Not doing so could lead to problems, e.g. in a case when a
temporary MyISAM table gets overrun due to its MAX_ROWS limit
while executing INSERT/REPLACE IGNORE ... SELECT.
The SELECT execution was aborted, but the error was
converted to a warning due to IGNORE clause, so neither 'ok'
nor 'error' packet could be sent back to the client. This
condition led to hanging client when using 5.0 server, or
assertion failure in 5.1.
with gcc 4.3.2
Compiling MySQL with gcc 4.3.2 and later produces a number of
warnings, many of which are new with the recent compiler
versions.
This bug will be resolved in more than one patch to limit the
size of changesets. This is the second patch, fixing more
of the warnings.
crashes server!
The problem affects the scenario when index merge is followed by a filesort
and the sort buffer is not big enough for all the sort keys.
In this case the filesort function will read the data to the end through the
index merge quick access method (and thus closing the cursor etc),
but will leave the pointer to the quick select method in place.
It will then create a temporary file to hold the results of the filesort and
will add it as a sort output file (in sort.io_cache).
Note that filesort will copy the original 'sort' structure in an automatic
variable and restore it after it's done.
As a result at exiting filesort() we have a sort.io_cache filled in and
nothing else (as a result of close of the cursors at end of reading data
through index merge).
Now create_sort_index() will note that there is a select and will clean it up
(as it's been used already by filesort() reading the data in). While doing that
a special case in the index merge destructor will clean up the sort.io_cache,
assuming it's an output of the index merge method and is not needed anymore.
As a result the code that tries to read the data back from the filesort output
will get no data in both memory and disk and will crash.
Fixed similarly to how filesort() does it : by copying the sort.io_cache structure
to a local variable, removing the pointer to the io_cache (so that it's not freed
by QUICK_INDEX_MERGE_SELECT::~QUICK_INDEX_MERGE_SELECT) and restoring the original
structure (together with the valid pointer) after the cleanup is done.
This is a safe thing to do because all the structures are already cleaned up by
hitting the end of the index merge's read method (QUICK_INDEX_MERGE_SELECT::get_next())
and the cleanup code being written in a way that tolerates repeating cleanups.
Holding on to the temporary inno hash index latch is an optimization in
many cases, but a pessimization in some others.
Release temporary latches for those corner cases we (or rather, or customers,
thanks!) have identified, that is, when we are about to do something that
might take a really long time, like REPAIR or filesort.
EXPLAIN EXTENDED of nested query containing a error:
1054 Unknown column '...' in 'field list'
may cause a server crash.
Parse error like described above forces a call to
JOIN::destroy() on malformed subquery.
That JOIN::destroy function closes and frees temporary
tables. However, temporary fields of these tables
may be listed in st_select_lex::group_list of outer
query, and that st_select_lex may not cleanup them
properly. So, after the JOIN::destroy call that
st_select_lex::group_list may have Item_field
objects with dangling pointers to freed temporary
table Field objects. That caused a crash.
Original commentary:
Bug #37348: Crash in or immediately after JOIN::make_sum_func_list
The optimizer pulls up aggregate functions which should be aggregated in
an outer select. At some point it may substitute such a function for a field
in the temporary table. The setup_copy_fields function doesn't take this
into account and may overrun the copy_field buffer.
Fixed by filtering out the fields referenced through the specialized
reference for aggregates (Item_aggregate_ref).
Added an assertion to make sure bugs that cause similar discrepancy
don't go undetected.
connections
The problem is that tables can enter open table cache for a thread without
being properly cleaned up. This can happen if make_join_statistics() fails
to read a const table because of e.g. a deadlock. It does set a member of
TABLE structure to a value it allocates, but doesn't clean-up this setting
on error nor does it set the rest of the members in JOIN to allow for
automatic cleanup.
As a result when such an error occurs and the next statement depends re-uses
the table from the open tables cache it will get it with this
TABLE::reginfo.join_tab pointing to a memory area that's freed.
Fixed by making sure make_join_statistics() cleans up TABLE::reginfo.join_tab
on error.
- Remove bothersome warning messages. This change focuses on the warnings
that are covered by the ignore file: support-files/compiler_warnings.supp.
- Strings are guaranteed to be max uint in length
ORDER BY could cause a server crash
Dependent subqueries like
SELECT COUNT(*) FROM t1, t2 WHERE t2.b
IN (SELECT DISTINCT t2.b FROM t2 WHERE t2.b = t1.a)
caused a memory leak proportional to the
number of outer rows.
The make_simple_join() function has been modified to
JOIN class method to store join_tab_reexec and
table_reexec values in the parent join only
(make_simple_join of tmp_join may access these values
via 'this' pointer of the parent JOIN).
NOTE: this patch doesn't include standard test case (this is
"out of memory" bug). See bug #42037 page for test cases.
messed up
"ROW(...) IN (SELECT ... FROM DUAL)" always returned TRUE.
Item_in_subselect::row_value_transformer rewrites "ROW(...)
IN SELECT" conditions into the "EXISTS (SELECT ... HAVING ...)"
form.
For a subquery from the DUAL pseudotable resulting HAVING
condition is an expression on constant values, so further
transformation with optimize_cond() eliminates this HAVING
condition and resets JOIN::having to NULL.
Then JOIN::exec treated that NULL as an always-true-HAVING
and that caused a bug.
To distinguish an optimized out "HAVING TRUE" clause from
"HAVING FALSE" we already have the JOIN::having_value flag.
However, JOIN::exec() ignored JOIN::having_value as described
above as if it always set to COND_TRUE.
The JOIN::exec method has been modified to take into account
the value of the JOIN::having_value field.
The greedy optimizer tracks the current level of nested joins and the position
inside these by setting and maintaining a state that's global for the whole FROM
clause.
This state was correctly maintained inside the selection of the next partial plan
table (in best_extension_by_limited_search()).
greedy_search() also moves the current position by adding the last partial match
table when there's not enough tables in the partial plan found by
best_extension_by_limited_search().
This may require update of the global state variables that describe the current
position in the plan if the last table placed by greedy_search is not a top-level
join table.
Fixed by updating the state after placing the partial plan table in greedy_search()
in the same way this is done on entering the best_extension_by_limited_search().
Fixed the signature of the function called to update the state :
check_interleaving_with_nj
Table could be marked dependent because it is
either 1) an inner table of an outer join, or 2) it is a part of
STRAIGHT_JOIN. In case of STRAIGHT_JOIN table->maybe_null should not
be assigned. The fix is to set st_table::maybe_null to 'true' only
for those tables which are used in outer join.
