1 /* vim:expandtab:shiftwidth=2:tabstop=2:smarttab:
5 * Copyright (C) 2011 Data Differential, http://datadifferential.com/
6 * Copyright (C) 2006-2009 Brian Aker
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions are
13 * * Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
16 * * Redistributions in binary form must reproduce the above
17 * copyright notice, this list of conditions and the following disclaimer
18 * in the documentation and/or other materials provided with the
21 * * The names of its contributors may not be used to endorse or
22 * promote products derived from this software without specific prior
25 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
26 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
27 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
28 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
29 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
30 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
31 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
32 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
33 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
34 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
35 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
40 #include "libmemcached/common.h"
47 static ssize_t
io_flush(memcached_server_write_instance_st ptr
,
48 const bool with_flush
,
49 memcached_return_t
*error
);
50 static void increment_udp_message_id(memcached_server_write_instance_st ptr
);
52 static memcached_return_t
io_wait(memcached_server_write_instance_st ptr
,
53 memc_read_or_write read_or_write
)
61 if (read_or_write
== MEM_WRITE
) /* write */
64 WATCHPOINT_SET(ptr
->io_wait_count
.write
++);
68 WATCHPOINT_SET(ptr
->io_wait_count
.read
++);
72 ** We are going to block on write, but at least on Solaris we might block
73 ** on write if we haven't read anything from our input buffer..
74 ** Try to purge the input buffer if we don't do any flow control in the
75 ** application layer (just sending a lot of data etc)
76 ** The test is moved down in the purge function to avoid duplication of
79 if (read_or_write
== MEM_WRITE
)
81 memcached_return_t rc
= memcached_purge(ptr
);
82 if (rc
!= MEMCACHED_SUCCESS
&& rc
!= MEMCACHED_STORED
)
83 return MEMCACHED_FAILURE
;
87 while (--loop_max
) // While loop is for ERESTART or EINTR
89 error
= poll(&fds
, 1, ptr
->root
->poll_timeout
);
94 WATCHPOINT_IF_LABELED_NUMBER(read_or_write
&& loop_max
< 4, "read() times we had to loop, decremented down from 5", loop_max
);
95 WATCHPOINT_IF_LABELED_NUMBER(!read_or_write
&& loop_max
< 4, "write() times we had to loop, decremented down from 5", loop_max
);
97 return MEMCACHED_SUCCESS
;
98 case 0: // Timeout occured, we let the while() loop do its thing.
99 return MEMCACHED_TIMEOUT
;
101 WATCHPOINT_ERRNO(get_socket_errno());
102 switch (get_socket_errno())
104 #ifdef TARGET_OS_LINUX
110 if (fds
.revents
& POLLERR
)
113 socklen_t len
= sizeof (err
);
114 (void)getsockopt(ptr
->fd
, SOL_SOCKET
, SO_ERROR
, &err
, &len
);
115 ptr
->cached_errno
= (err
== 0) ? get_socket_errno() : err
;
119 ptr
->cached_errno
= get_socket_errno();
121 memcached_quit_server(ptr
, true);
123 return MEMCACHED_FAILURE
;
128 /* Imposssible for anything other then -1 */
129 WATCHPOINT_ASSERT(error
== -1);
130 ptr
->cached_errno
= get_socket_errno();
131 memcached_quit_server(ptr
, true);
133 return MEMCACHED_FAILURE
;
136 memcached_return_t
memcached_io_wait_for_write(memcached_server_write_instance_st ptr
)
138 return io_wait(ptr
, MEM_WRITE
);
142 * Try to fill the input buffer for a server with as much
145 * @param ptr the server to pack
147 static bool repack_input_buffer(memcached_server_write_instance_st ptr
)
149 if (ptr
->read_ptr
!= ptr
->read_buffer
)
151 /* Move all of the data to the beginning of the buffer so
152 ** that we can fit more data into the buffer...
154 memmove(ptr
->read_buffer
, ptr
->read_ptr
, ptr
->read_buffer_length
);
155 ptr
->read_ptr
= ptr
->read_buffer
;
156 ptr
->read_data_length
= ptr
->read_buffer_length
;
159 /* There is room in the buffer, try to fill it! */
160 if (ptr
->read_buffer_length
!= MEMCACHED_MAX_BUFFER
)
162 /* Just try a single read to grab what's available */
163 ssize_t nr
= recv(ptr
->fd
,
164 ptr
->read_ptr
+ ptr
->read_data_length
,
165 MEMCACHED_MAX_BUFFER
- ptr
->read_data_length
,
170 ptr
->read_data_length
+= (size_t)nr
;
171 ptr
->read_buffer_length
+= (size_t)nr
;
179 * If the we have callbacks connected to this server structure
180 * we may start process the input queue and fire the callbacks
181 * for the incomming messages. This function is _only_ called
182 * when the input buffer is full, so that we _know_ that we have
183 * at least _one_ message to process.
185 * @param ptr the server to star processing iput messages for
186 * @return true if we processed anything, false otherwise
188 static bool process_input_buffer(memcached_server_write_instance_st ptr
)
191 ** We might be able to process some of the response messages if we
192 ** have a callback set up
194 if (ptr
->root
->callbacks
!= NULL
&& ptr
->root
->flags
.use_udp
== false)
197 * We might have responses... try to read them out and fire
200 memcached_callback_st cb
= *ptr
->root
->callbacks
;
202 memcached_set_processing_input((memcached_st
*)ptr
->root
, true);
204 char buffer
[MEMCACHED_DEFAULT_COMMAND_SIZE
];
205 memcached_return_t error
;
206 memcached_st
*root
= (memcached_st
*)ptr
->root
;
207 error
= memcached_response(ptr
, buffer
, sizeof(buffer
),
210 memcached_set_processing_input(root
, false);
212 if (error
== MEMCACHED_SUCCESS
)
214 for (unsigned int x
= 0; x
< cb
.number_of_callback
; x
++)
216 error
= (*cb
.callback
[x
])(ptr
->root
, &root
->result
, cb
.context
);
217 if (error
!= MEMCACHED_SUCCESS
)
221 /* @todo what should I do with the error message??? */
223 /* @todo what should I do with other error messages?? */
230 #if 0 // Dead code, this should be removed.
231 void memcached_io_preread(memcached_st
*ptr
)
237 for (x
= 0; x
< memcached_server_count(ptr
); x
++)
239 if (memcached_server_response_count(ptr
, x
) &&
240 ptr
->hosts
[x
].read_data_length
< MEMCACHED_MAX_BUFFER
)
244 data_read
= recv(ptr
->hosts
[x
].fd
,
245 ptr
->hosts
[x
].read_ptr
+ ptr
->hosts
[x
].read_data_length
,
246 MEMCACHED_MAX_BUFFER
- ptr
->hosts
[x
].read_data_length
, 0);
247 if (data_read
== SOCKET_ERROR
)
250 ptr
->hosts
[x
].read_buffer_length
+= data_read
;
251 ptr
->hosts
[x
].read_data_length
+= data_read
;
257 memcached_return_t
memcached_io_read(memcached_server_write_instance_st ptr
,
258 void *buffer
, size_t length
, ssize_t
*nread
)
262 buffer_ptr
= static_cast<char *>(buffer
);
266 if (not ptr
->read_buffer_length
)
272 data_read
= recv(ptr
->fd
, ptr
->read_buffer
, MEMCACHED_MAX_BUFFER
, 0);
277 else if (data_read
== SOCKET_ERROR
)
279 ptr
->cached_errno
= get_socket_errno();
280 memcached_return_t rc
= MEMCACHED_ERRNO
;
281 switch (get_socket_errno())
288 #ifdef TARGET_OS_LINUX
291 if ((rc
= io_wait(ptr
, MEM_READ
)) == MEMCACHED_SUCCESS
)
297 memcached_quit_server(ptr
, true);
299 return memcached_set_error(*ptr
, rc
, MEMCACHED_AT
);
306 EOF. Any data received so far is incomplete
307 so discard it. This always reads by byte in case of TCP
308 and protocol enforcement happens at memcached_response()
309 looking for '\n'. We do not care for UDB which requests 8 bytes
310 at once. Generally, this means that connection went away. Since
311 for blocking I/O we do not return 0 and for non-blocking case
312 it will return EGAIN if data is not immediatly available.
314 WATCHPOINT_STRING("We had a zero length recv()");
315 memcached_quit_server(ptr
, true);
317 return memcached_set_error(*ptr
, MEMCACHED_UNKNOWN_READ_FAILURE
, MEMCACHED_AT
);
321 ptr
->io_bytes_sent
= 0;
322 ptr
->read_data_length
= (size_t) data_read
;
323 ptr
->read_buffer_length
= (size_t) data_read
;
324 ptr
->read_ptr
= ptr
->read_buffer
;
331 difference
= (length
> ptr
->read_buffer_length
) ? ptr
->read_buffer_length
: length
;
333 memcpy(buffer_ptr
, ptr
->read_ptr
, difference
);
334 length
-= difference
;
335 ptr
->read_ptr
+= difference
;
336 ptr
->read_buffer_length
-= difference
;
337 buffer_ptr
+= difference
;
341 *buffer_ptr
= *ptr
->read_ptr
;
343 ptr
->read_buffer_length
--;
349 ptr
->server_failure_counter
= 0;
350 *nread
= (ssize_t
)(buffer_ptr
- (char*)buffer
);
351 return MEMCACHED_SUCCESS
;
354 static ssize_t
_io_write(memcached_server_write_instance_st ptr
,
355 const void *buffer
, size_t length
, bool with_flush
)
357 size_t original_length
;
358 const char* buffer_ptr
;
360 WATCHPOINT_ASSERT(ptr
->fd
!= INVALID_SOCKET
);
362 original_length
= length
;
363 buffer_ptr
= static_cast<const char *>(buffer
);
371 if (ptr
->type
== MEMCACHED_CONNECTION_UDP
)
373 //UDP does not support partial writes
374 buffer_end
= MAX_UDP_DATAGRAM_LENGTH
;
375 should_write
= length
;
376 if (ptr
->write_buffer_offset
+ should_write
> buffer_end
)
383 buffer_end
= MEMCACHED_MAX_BUFFER
;
384 should_write
= buffer_end
- ptr
->write_buffer_offset
;
385 should_write
= (should_write
< length
) ? should_write
: length
;
388 write_ptr
= ptr
->write_buffer
+ ptr
->write_buffer_offset
;
389 memcpy(write_ptr
, buffer_ptr
, should_write
);
390 ptr
->write_buffer_offset
+= should_write
;
391 buffer_ptr
+= should_write
;
392 length
-= should_write
;
394 if (ptr
->write_buffer_offset
== buffer_end
&& ptr
->type
!= MEMCACHED_CONNECTION_UDP
)
396 memcached_return_t rc
;
399 WATCHPOINT_ASSERT(ptr
->fd
!= INVALID_SOCKET
);
400 sent_length
= io_flush(ptr
, with_flush
, &rc
);
401 if (sent_length
== -1)
406 /* If io_flush calls memcached_purge, sent_length may be 0 */
407 unlikely (sent_length
!= 0)
409 WATCHPOINT_ASSERT(sent_length
== (ssize_t
)buffer_end
);
416 memcached_return_t rc
;
417 WATCHPOINT_ASSERT(ptr
->fd
!= INVALID_SOCKET
);
418 if (io_flush(ptr
, with_flush
, &rc
) == -1)
424 return (ssize_t
) original_length
;
427 ssize_t
memcached_io_write(memcached_server_write_instance_st ptr
,
428 const void *buffer
, size_t length
, bool with_flush
)
430 return _io_write(ptr
, buffer
, length
, with_flush
);
433 ssize_t
memcached_io_writev(memcached_server_write_instance_st ptr
,
434 const struct libmemcached_io_vector_st
*vector
,
435 size_t number_of
, bool with_flush
)
439 for (size_t x
= 0; x
< number_of
; x
++, vector
++)
443 if ((returnable
= _io_write(ptr
, vector
->buffer
, vector
->length
, false)) == -1)
452 if (memcached_io_write(ptr
, NULL
, 0, true) == -1)
462 memcached_return_t
memcached_io_close(memcached_server_write_instance_st ptr
)
464 if (ptr
->fd
== INVALID_SOCKET
)
466 return MEMCACHED_SUCCESS
;
469 /* in case of death shutdown to avoid blocking at close() */
470 if (shutdown(ptr
->fd
, SHUT_RDWR
) == SOCKET_ERROR
&& get_socket_errno() != ENOTCONN
)
472 WATCHPOINT_NUMBER(ptr
->fd
);
473 WATCHPOINT_ERRNO(get_socket_errno());
474 WATCHPOINT_ASSERT(get_socket_errno());
477 if (closesocket(ptr
->fd
) == SOCKET_ERROR
)
479 WATCHPOINT_ERRNO(get_socket_errno());
482 return MEMCACHED_SUCCESS
;
485 memcached_server_write_instance_st
memcached_io_get_readable_server(memcached_st
*memc
)
487 #define MAX_SERVERS_TO_POLL 100
488 struct pollfd fds
[MAX_SERVERS_TO_POLL
];
489 unsigned int host_index
= 0;
492 x
< memcached_server_count(memc
) && host_index
< MAX_SERVERS_TO_POLL
;
495 memcached_server_write_instance_st instance
=
496 memcached_server_instance_fetch(memc
, x
);
498 if (instance
->read_buffer_length
> 0) /* I have data in the buffer */
501 if (memcached_server_response_count(instance
) > 0)
503 fds
[host_index
].events
= POLLIN
;
504 fds
[host_index
].revents
= 0;
505 fds
[host_index
].fd
= instance
->fd
;
512 /* We have 0 or 1 server with pending events.. */
513 for (uint32_t x
= 0; x
< memcached_server_count(memc
); ++x
)
515 memcached_server_write_instance_st instance
=
516 memcached_server_instance_fetch(memc
, x
);
518 if (memcached_server_response_count(instance
) > 0)
527 switch (poll(fds
, host_index
, memc
->poll_timeout
))
530 memcached_set_errno(*memc
, get_socket_errno(), MEMCACHED_AT
);
536 for (size_t x
= 0; x
< host_index
; ++x
)
538 if (fds
[x
].revents
& POLLIN
)
540 for (uint32_t y
= 0; y
< memcached_server_count(memc
); ++y
)
542 memcached_server_write_instance_st instance
=
543 memcached_server_instance_fetch(memc
, y
);
545 if (instance
->fd
== fds
[x
].fd
)
555 static ssize_t
io_flush(memcached_server_write_instance_st ptr
,
556 const bool with_flush
,
557 memcached_return_t
*error
)
560 ** We might want to purge the input buffer if we haven't consumed
561 ** any output yet... The test for the limits is the purge is inline
562 ** in the purge function to avoid duplicating the logic..
565 memcached_return_t rc
;
566 WATCHPOINT_ASSERT(ptr
->fd
!= INVALID_SOCKET
);
567 rc
= memcached_purge(ptr
);
569 if (rc
!= MEMCACHED_SUCCESS
&& rc
!= MEMCACHED_STORED
)
575 size_t return_length
;
576 char *local_write_ptr
= ptr
->write_buffer
;
577 size_t write_length
= ptr
->write_buffer_offset
;
579 *error
= MEMCACHED_SUCCESS
;
581 WATCHPOINT_ASSERT(ptr
->fd
!= INVALID_SOCKET
);
583 // UDP Sanity check, make sure that we are not sending somthing too big
584 if (ptr
->type
== MEMCACHED_CONNECTION_UDP
&& write_length
> MAX_UDP_DATAGRAM_LENGTH
)
589 if (ptr
->write_buffer_offset
== 0 || (ptr
->type
== MEMCACHED_CONNECTION_UDP
590 && ptr
->write_buffer_offset
== UDP_DATAGRAM_HEADER_LENGTH
))
593 /* Looking for memory overflows */
595 if (write_length
== MEMCACHED_MAX_BUFFER
)
596 WATCHPOINT_ASSERT(ptr
->write_buffer
== local_write_ptr
);
597 WATCHPOINT_ASSERT((ptr
->write_buffer
+ MEMCACHED_MAX_BUFFER
) >= (local_write_ptr
+ write_length
));
603 WATCHPOINT_ASSERT(ptr
->fd
!= INVALID_SOCKET
);
604 WATCHPOINT_ASSERT(write_length
> 0);
606 if (ptr
->type
== MEMCACHED_CONNECTION_UDP
)
607 increment_udp_message_id(ptr
);
609 WATCHPOINT_ASSERT(ptr
->fd
!= INVALID_SOCKET
);
612 sent_length
= send(ptr
->fd
, local_write_ptr
, write_length
, MSG_NOSIGNAL
|MSG_DONTWAIT
);
616 sent_length
= send(ptr
->fd
, local_write_ptr
, write_length
, MSG_NOSIGNAL
|MSG_DONTWAIT
|MSG_MORE
);
619 if (sent_length
== SOCKET_ERROR
)
621 ptr
->cached_errno
= get_socket_errno();
622 #if 0 // @todo I should look at why we hit this bit of code hard frequently
623 WATCHPOINT_ERRNO(get_socket_errno());
624 WATCHPOINT_NUMBER(get_socket_errno());
626 switch (get_socket_errno())
636 * We may be blocked on write because the input buffer
637 * is full. Let's check if we have room in our input
638 * buffer for more data and retry the write before
641 if (repack_input_buffer(ptr
) ||
642 process_input_buffer(ptr
))
645 memcached_return_t rc
;
646 rc
= io_wait(ptr
, MEM_WRITE
);
648 if (rc
== MEMCACHED_SUCCESS
|| rc
== MEMCACHED_TIMEOUT
)
651 memcached_quit_server(ptr
, true);
657 memcached_quit_server(ptr
, true);
658 *error
= MEMCACHED_ERRNO
;
659 WATCHPOINT_ASSERT(ptr
->fd
== -1);
664 if (ptr
->type
== MEMCACHED_CONNECTION_UDP
&&
665 (size_t)sent_length
!= write_length
)
667 memcached_quit_server(ptr
, true);
671 ptr
->io_bytes_sent
+= (uint32_t) sent_length
;
673 local_write_ptr
+= sent_length
;
674 write_length
-= (uint32_t) sent_length
;
675 return_length
+= (uint32_t) sent_length
;
678 WATCHPOINT_ASSERT(write_length
== 0);
679 // Need to study this assert() WATCHPOINT_ASSERT(return_length ==
680 // ptr->write_buffer_offset);
682 // if we are a udp server, the begining of the buffer is reserverd for
683 // the upd frame header
684 if (ptr
->type
== MEMCACHED_CONNECTION_UDP
)
685 ptr
->write_buffer_offset
= UDP_DATAGRAM_HEADER_LENGTH
;
687 ptr
->write_buffer_offset
= 0;
689 return (ssize_t
) return_length
;
693 Eventually we will just kill off the server with the problem.
695 void memcached_io_reset(memcached_server_write_instance_st ptr
)
697 memcached_quit_server(ptr
, true);
701 * Read a given number of bytes from the server and place it into a specific
702 * buffer. Reset the IO channel on this server if an error occurs.
704 memcached_return_t
memcached_safe_read(memcached_server_write_instance_st ptr
,
709 char *data
= static_cast<char *>(dta
);
711 while (offset
< size
)
714 memcached_return_t rc
= memcached_io_read(ptr
, data
+ offset
, size
- offset
,
716 if (rc
!= MEMCACHED_SUCCESS
)
719 offset
+= (size_t) nread
;
722 return MEMCACHED_SUCCESS
;
725 memcached_return_t
memcached_io_readline(memcached_server_write_instance_st ptr
,
729 bool line_complete
= false;
732 while (!line_complete
)
734 if (ptr
->read_buffer_length
== 0)
737 * We don't have any data in the buffer, so let's fill the read
738 * buffer. Call the standard read function to avoid duplicating
742 memcached_return_t rc
= memcached_io_read(ptr
, buffer_ptr
, 1, &nread
);
743 if (rc
!= MEMCACHED_SUCCESS
)
746 if (*buffer_ptr
== '\n')
753 /* Now let's look in the buffer and copy as we go! */
754 while (ptr
->read_buffer_length
&& total_nr
< size
&& !line_complete
)
756 *buffer_ptr
= *ptr
->read_ptr
;
757 if (*buffer_ptr
== '\n')
758 line_complete
= true;
759 --ptr
->read_buffer_length
;
765 if (total_nr
== size
)
766 return MEMCACHED_PROTOCOL_ERROR
;
769 return MEMCACHED_SUCCESS
;
773 * The udp request id consists of two seperate sections
775 * 2) The message number
776 * The thread id should only be set when the memcached_st struct is created
777 * and should not be changed.
779 * The message num is incremented for each new message we send, this function
780 * extracts the message number from message_id, increments it and then
781 * writes the new value back into the header
783 static void increment_udp_message_id(memcached_server_write_instance_st ptr
)
785 struct udp_datagram_header_st
*header
= (struct udp_datagram_header_st
*)ptr
->write_buffer
;
786 uint16_t cur_req
= get_udp_datagram_request_id(header
);
787 int msg_num
= get_msg_num_from_request_id(cur_req
);
788 int thread_id
= get_thread_id_from_request_id(cur_req
);
790 if (((++msg_num
) & UDP_REQUEST_ID_THREAD_MASK
) != 0)
793 header
->request_id
= htons((uint16_t) (thread_id
| msg_num
));
796 memcached_return_t
memcached_io_init_udp_header(memcached_server_write_instance_st ptr
, uint16_t thread_id
)
798 if (thread_id
> UDP_REQUEST_ID_MAX_THREAD_ID
)
799 return MEMCACHED_FAILURE
;
801 struct udp_datagram_header_st
*header
= (struct udp_datagram_header_st
*)ptr
->write_buffer
;
802 header
->request_id
= htons((uint16_t) (generate_udp_request_thread_id(thread_id
)));
803 header
->num_datagrams
= htons(1);
804 header
->sequence_number
= htons(0);
806 return MEMCACHED_SUCCESS
;