All packets but the row data packet are read into preallocated buffer,
either on the stack or on the heap (cmd_buffer). The size of the buffer
is always checked to skip buffer overruns. Only up to the size of the
buffer is stored into the buffer but packet->header.size has the size of
all data sent. In this case network layer bails.
This patch hardenes the reads, so if packets are malformed and shorter
than they should be there will be no further reads in the buffer.
In short, detection of malformed packets.
because on 32bit SIZEOF_LONG is 4 and INT64 from MySQL is 8 everything
was converted to string, even if it was able to put it in a long.
This closes Request #50651 Native type cast returns wrong result
Fix inconsistencies
- memory allocated with spprintf, or likes, outside of mysqlnd's
allocator functions should not be freed by the mysqlnd's allocator
(a wrapper around emalloc/malloc).
- memory allocated by the mysqlnd's allocator should only be freed
by it.
- add a mode to track memory usage (malloc/free)
- memory allocated with spprintf, or likes, outside of mysqlnd's
allocator functions should not be freed by the mysqlnd's allocator
(a wrapper around emalloc/malloc).
- memory allocated by the mysqlnd's allocator should only be freed
by it.
- add a mode to track memory usage (malloc/free)
Streams API registers every stream as resource, which lands then
in EG(regular_list), however doesn't clean that when the stream is
closed. At the end this is a para-leak. At the end of the script
all memory is cleaned, however this is a problem for long runnig
scripts that open connections. For every opened and closed connection
about 150 Bytes on 32bit and 250 Bytes on 64bit will be "lost",
according to memory_get_usage().
as it is in when compiled from source and the default for mysqlnd.
SuSE for example uses /var/run/mysql/mysql.sock . Also, sql.safe_mode
(ext/mysql and ingres) needs the socket.
Fix possible crashes in mysqlnd. When packets are shorter, functions should
return error.
the wire properly (as it is done when the statement is closed).
If there is more that one result sets returned from the PS
(like a call to SP that returns at least one RSet next to the
status rset) then the line was blocked. PS Multi-Res is not supported
in any libmysql from a GA-ed MySQL.
needed to move to a new structure MYSQLND_STMT. Makes
the code cleaner and less error-prone.
Also fix PDO/MySQL which directly touch mysqlnd internals
instead of using API calls.
and thus wrong data was saved MYSQLND_FIELD::db. However, this
wasn't caught because there is no code in mysql/mysqli/pdo_mysql
which reads that field of the structure. The problem was found
during code review.
Hanging was possible on the subsequent try to acquire a mutex.
Now it is correctly implemented and if a trigger is being executed
then no other trigger will be fired, on recursive calls.
Andrey Hristov