linux/net/rxrpc/sendmsg.c

880 lines
22 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0-or-later
/* AF_RXRPC sendmsg() implementation.
*
* Copyright (C) 2007, 2016 Red Hat, Inc. All Rights Reserved.
* Written by David Howells (dhowells@redhat.com)
*/
#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
#include <linux/net.h>
#include <linux/gfp.h>
#include <linux/skbuff.h>
#include <linux/export.h>
#include <linux/sched/signal.h>
#include <net/sock.h>
#include <net/af_rxrpc.h>
#include "ar-internal.h"
/*
* Return true if there's sufficient Tx queue space.
*/
static bool rxrpc_check_tx_space(struct rxrpc_call *call, rxrpc_seq_t *_tx_win)
{
unsigned int win_size =
min_t(unsigned int, call->tx_winsize,
call->cong_cwnd + call->cong_extra);
rxrpc_seq_t tx_win = READ_ONCE(call->tx_hard_ack);
if (_tx_win)
*_tx_win = tx_win;
return call->tx_top - tx_win < win_size;
}
/*
* Wait for space to appear in the Tx queue or a signal to occur.
*/
static int rxrpc_wait_for_tx_window_intr(struct rxrpc_sock *rx,
struct rxrpc_call *call,
long *timeo)
{
for (;;) {
set_current_state(TASK_INTERRUPTIBLE);
if (rxrpc_check_tx_space(call, NULL))
return 0;
if (call->state >= RXRPC_CALL_COMPLETE)
return call->error;
if (signal_pending(current))
return sock_intr_errno(*timeo);
trace_rxrpc_transmit(call, rxrpc_transmit_wait);
mutex_unlock(&call->user_mutex);
*timeo = schedule_timeout(*timeo);
if (mutex_lock_interruptible(&call->user_mutex) < 0)
return sock_intr_errno(*timeo);
}
}
/*
* Wait for space to appear in the Tx queue uninterruptibly, but with
* a timeout of 2*RTT if no progress was made and a signal occurred.
*/
static int rxrpc_wait_for_tx_window_waitall(struct rxrpc_sock *rx,
struct rxrpc_call *call)
{
rxrpc_seq_t tx_start, tx_win;
rxrpc: Fix the excessive initial retransmission timeout rxrpc currently uses a fixed 4s retransmission timeout until the RTT is sufficiently sampled. This can cause problems with some fileservers with calls to the cache manager in the afs filesystem being dropped from the fileserver because a packet goes missing and the retransmission timeout is greater than the call expiry timeout. Fix this by: (1) Copying the RTT/RTO calculation code from Linux's TCP implementation and altering it to fit rxrpc. (2) Altering the various users of the RTT to make use of the new SRTT value. (3) Replacing the use of rxrpc_resend_timeout to use the calculated RTO value instead (which is needed in jiffies), along with a backoff. Notes: (1) rxrpc provides RTT samples by matching the serial numbers on outgoing DATA packets that have the RXRPC_REQUEST_ACK set and PING ACK packets against the reference serial number in incoming REQUESTED ACK and PING-RESPONSE ACK packets. (2) Each packet that is transmitted on an rxrpc connection gets a new per-connection serial number, even for retransmissions, so an ACK can be cross-referenced to a specific trigger packet. This allows RTT information to be drawn from retransmitted DATA packets also. (3) rxrpc maintains the RTT/RTO state on the rxrpc_peer record rather than on an rxrpc_call because many RPC calls won't live long enough to generate more than one sample. (4) The calculated SRTT value is in units of 8ths of a microsecond rather than nanoseconds. The (S)RTT and RTO values are displayed in /proc/net/rxrpc/peers. Fixes: 17926a79320a ([AF_RXRPC]: Provide secure RxRPC sockets for use by userspace and kernel both"") Signed-off-by: David Howells <dhowells@redhat.com>
2020-05-11 21:54:34 +08:00
signed long rtt, timeout;
rxrpc: Fix the excessive initial retransmission timeout rxrpc currently uses a fixed 4s retransmission timeout until the RTT is sufficiently sampled. This can cause problems with some fileservers with calls to the cache manager in the afs filesystem being dropped from the fileserver because a packet goes missing and the retransmission timeout is greater than the call expiry timeout. Fix this by: (1) Copying the RTT/RTO calculation code from Linux's TCP implementation and altering it to fit rxrpc. (2) Altering the various users of the RTT to make use of the new SRTT value. (3) Replacing the use of rxrpc_resend_timeout to use the calculated RTO value instead (which is needed in jiffies), along with a backoff. Notes: (1) rxrpc provides RTT samples by matching the serial numbers on outgoing DATA packets that have the RXRPC_REQUEST_ACK set and PING ACK packets against the reference serial number in incoming REQUESTED ACK and PING-RESPONSE ACK packets. (2) Each packet that is transmitted on an rxrpc connection gets a new per-connection serial number, even for retransmissions, so an ACK can be cross-referenced to a specific trigger packet. This allows RTT information to be drawn from retransmitted DATA packets also. (3) rxrpc maintains the RTT/RTO state on the rxrpc_peer record rather than on an rxrpc_call because many RPC calls won't live long enough to generate more than one sample. (4) The calculated SRTT value is in units of 8ths of a microsecond rather than nanoseconds. The (S)RTT and RTO values are displayed in /proc/net/rxrpc/peers. Fixes: 17926a79320a ([AF_RXRPC]: Provide secure RxRPC sockets for use by userspace and kernel both"") Signed-off-by: David Howells <dhowells@redhat.com>
2020-05-11 21:54:34 +08:00
rtt = READ_ONCE(call->peer->srtt_us) >> 3;
rtt = usecs_to_jiffies(rtt) * 2;
if (rtt < 2)
rtt = 2;
rxrpc: Fix the excessive initial retransmission timeout rxrpc currently uses a fixed 4s retransmission timeout until the RTT is sufficiently sampled. This can cause problems with some fileservers with calls to the cache manager in the afs filesystem being dropped from the fileserver because a packet goes missing and the retransmission timeout is greater than the call expiry timeout. Fix this by: (1) Copying the RTT/RTO calculation code from Linux's TCP implementation and altering it to fit rxrpc. (2) Altering the various users of the RTT to make use of the new SRTT value. (3) Replacing the use of rxrpc_resend_timeout to use the calculated RTO value instead (which is needed in jiffies), along with a backoff. Notes: (1) rxrpc provides RTT samples by matching the serial numbers on outgoing DATA packets that have the RXRPC_REQUEST_ACK set and PING ACK packets against the reference serial number in incoming REQUESTED ACK and PING-RESPONSE ACK packets. (2) Each packet that is transmitted on an rxrpc connection gets a new per-connection serial number, even for retransmissions, so an ACK can be cross-referenced to a specific trigger packet. This allows RTT information to be drawn from retransmitted DATA packets also. (3) rxrpc maintains the RTT/RTO state on the rxrpc_peer record rather than on an rxrpc_call because many RPC calls won't live long enough to generate more than one sample. (4) The calculated SRTT value is in units of 8ths of a microsecond rather than nanoseconds. The (S)RTT and RTO values are displayed in /proc/net/rxrpc/peers. Fixes: 17926a79320a ([AF_RXRPC]: Provide secure RxRPC sockets for use by userspace and kernel both"") Signed-off-by: David Howells <dhowells@redhat.com>
2020-05-11 21:54:34 +08:00
timeout = rtt;
tx_start = READ_ONCE(call->tx_hard_ack);
for (;;) {
set_current_state(TASK_UNINTERRUPTIBLE);
tx_win = READ_ONCE(call->tx_hard_ack);
if (rxrpc_check_tx_space(call, &tx_win))
return 0;
if (call->state >= RXRPC_CALL_COMPLETE)
return call->error;
if (timeout == 0 &&
tx_win == tx_start && signal_pending(current))
return -EINTR;
if (tx_win != tx_start) {
rxrpc: Fix the excessive initial retransmission timeout rxrpc currently uses a fixed 4s retransmission timeout until the RTT is sufficiently sampled. This can cause problems with some fileservers with calls to the cache manager in the afs filesystem being dropped from the fileserver because a packet goes missing and the retransmission timeout is greater than the call expiry timeout. Fix this by: (1) Copying the RTT/RTO calculation code from Linux's TCP implementation and altering it to fit rxrpc. (2) Altering the various users of the RTT to make use of the new SRTT value. (3) Replacing the use of rxrpc_resend_timeout to use the calculated RTO value instead (which is needed in jiffies), along with a backoff. Notes: (1) rxrpc provides RTT samples by matching the serial numbers on outgoing DATA packets that have the RXRPC_REQUEST_ACK set and PING ACK packets against the reference serial number in incoming REQUESTED ACK and PING-RESPONSE ACK packets. (2) Each packet that is transmitted on an rxrpc connection gets a new per-connection serial number, even for retransmissions, so an ACK can be cross-referenced to a specific trigger packet. This allows RTT information to be drawn from retransmitted DATA packets also. (3) rxrpc maintains the RTT/RTO state on the rxrpc_peer record rather than on an rxrpc_call because many RPC calls won't live long enough to generate more than one sample. (4) The calculated SRTT value is in units of 8ths of a microsecond rather than nanoseconds. The (S)RTT and RTO values are displayed in /proc/net/rxrpc/peers. Fixes: 17926a79320a ([AF_RXRPC]: Provide secure RxRPC sockets for use by userspace and kernel both"") Signed-off-by: David Howells <dhowells@redhat.com>
2020-05-11 21:54:34 +08:00
timeout = rtt;
tx_start = tx_win;
}
trace_rxrpc_transmit(call, rxrpc_transmit_wait);
timeout = schedule_timeout(timeout);
}
}
/*
* Wait for space to appear in the Tx queue uninterruptibly.
*/
static int rxrpc_wait_for_tx_window_nonintr(struct rxrpc_sock *rx,
struct rxrpc_call *call,
long *timeo)
{
for (;;) {
set_current_state(TASK_UNINTERRUPTIBLE);
if (rxrpc_check_tx_space(call, NULL))
return 0;
if (call->state >= RXRPC_CALL_COMPLETE)
return call->error;
trace_rxrpc_transmit(call, rxrpc_transmit_wait);
*timeo = schedule_timeout(*timeo);
}
}
/*
* wait for space to appear in the transmit/ACK window
* - caller holds the socket locked
*/
static int rxrpc_wait_for_tx_window(struct rxrpc_sock *rx,
struct rxrpc_call *call,
long *timeo,
bool waitall)
{
DECLARE_WAITQUEUE(myself, current);
int ret;
rxrpc: Rewrite the data and ack handling code Rewrite the data and ack handling code such that: (1) Parsing of received ACK and ABORT packets and the distribution and the filing of DATA packets happens entirely within the data_ready context called from the UDP socket. This allows us to process and discard ACK and ABORT packets much more quickly (they're no longer stashed on a queue for a background thread to process). (2) We avoid calling skb_clone(), pskb_pull() and pskb_trim(). We instead keep track of the offset and length of the content of each packet in the sk_buff metadata. This means we don't do any allocation in the receive path. (3) Jumbo DATA packet parsing is now done in data_ready context. Rather than cloning the packet once for each subpacket and pulling/trimming it, we file the packet multiple times with an annotation for each indicating which subpacket is there. From that we can directly calculate the offset and length. (4) A call's receive queue can be accessed without taking locks (memory barriers do have to be used, though). (5) Incoming calls are set up from preallocated resources and immediately made live. They can than have packets queued upon them and ACKs generated. If insufficient resources exist, DATA packet #1 is given a BUSY reply and other DATA packets are discarded). (6) sk_buffs no longer take a ref on their parent call. To make this work, the following changes are made: (1) Each call's receive buffer is now a circular buffer of sk_buff pointers (rxtx_buffer) rather than a number of sk_buff_heads spread between the call and the socket. This permits each sk_buff to be in the buffer multiple times. The receive buffer is reused for the transmit buffer. (2) A circular buffer of annotations (rxtx_annotations) is kept parallel to the data buffer. Transmission phase annotations indicate whether a buffered packet has been ACK'd or not and whether it needs retransmission. Receive phase annotations indicate whether a slot holds a whole packet or a jumbo subpacket and, if the latter, which subpacket. They also note whether the packet has been decrypted in place. (3) DATA packet window tracking is much simplified. Each phase has just two numbers representing the window (rx_hard_ack/rx_top and tx_hard_ack/tx_top). The hard_ack number is the sequence number before base of the window, representing the last packet the other side says it has consumed. hard_ack starts from 0 and the first packet is sequence number 1. The top number is the sequence number of the highest-numbered packet residing in the buffer. Packets between hard_ack+1 and top are soft-ACK'd to indicate they've been received, but not yet consumed. Four macros, before(), before_eq(), after() and after_eq() are added to compare sequence numbers within the window. This allows for the top of the window to wrap when the hard-ack sequence number gets close to the limit. Two flags, RXRPC_CALL_RX_LAST and RXRPC_CALL_TX_LAST, are added also to indicate when rx_top and tx_top point at the packets with the LAST_PACKET bit set, indicating the end of the phase. (4) Calls are queued on the socket 'receive queue' rather than packets. This means that we don't need have to invent dummy packets to queue to indicate abnormal/terminal states and we don't have to keep metadata packets (such as ABORTs) around (5) The offset and length of a (sub)packet's content are now passed to the verify_packet security op. This is currently expected to decrypt the packet in place and validate it. However, there's now nowhere to store the revised offset and length of the actual data within the decrypted blob (there may be a header and padding to skip) because an sk_buff may represent multiple packets, so a locate_data security op is added to retrieve these details from the sk_buff content when needed. (6) recvmsg() now has to handle jumbo subpackets, where each subpacket is individually secured and needs to be individually decrypted. The code to do this is broken out into rxrpc_recvmsg_data() and shared with the kernel API. It now iterates over the call's receive buffer rather than walking the socket receive queue. Additional changes: (1) The timers are condensed to a single timer that is set for the soonest of three timeouts (delayed ACK generation, DATA retransmission and call lifespan). (2) Transmission of ACK and ABORT packets is effected immediately from process-context socket ops/kernel API calls that cause them instead of them being punted off to a background work item. The data_ready handler still has to defer to the background, though. (3) A shutdown op is added to the AF_RXRPC socket so that the AFS filesystem can shut down the socket and flush its own work items before closing the socket to deal with any in-progress service calls. Future additional changes that will need to be considered: (1) Make sure that a call doesn't hog the front of the queue by receiving data from the network as fast as userspace is consuming it to the exclusion of other calls. (2) Transmit delayed ACKs from within recvmsg() when we've consumed sufficiently more packets to avoid the background work item needing to run. Signed-off-by: David Howells <dhowells@redhat.com>
2016-09-08 18:10:12 +08:00
_enter(",{%u,%u,%u}",
call->tx_hard_ack, call->tx_top, call->tx_winsize);
add_wait_queue(&call->waitq, &myself);
switch (call->interruptibility) {
case RXRPC_INTERRUPTIBLE:
if (waitall)
ret = rxrpc_wait_for_tx_window_waitall(rx, call);
else
ret = rxrpc_wait_for_tx_window_intr(rx, call, timeo);
break;
case RXRPC_PREINTERRUPTIBLE:
case RXRPC_UNINTERRUPTIBLE:
default:
ret = rxrpc_wait_for_tx_window_nonintr(rx, call, timeo);
break;
}
remove_wait_queue(&call->waitq, &myself);
set_current_state(TASK_RUNNING);
_leave(" = %d", ret);
return ret;
}
/*
rxrpc: Rewrite the data and ack handling code Rewrite the data and ack handling code such that: (1) Parsing of received ACK and ABORT packets and the distribution and the filing of DATA packets happens entirely within the data_ready context called from the UDP socket. This allows us to process and discard ACK and ABORT packets much more quickly (they're no longer stashed on a queue for a background thread to process). (2) We avoid calling skb_clone(), pskb_pull() and pskb_trim(). We instead keep track of the offset and length of the content of each packet in the sk_buff metadata. This means we don't do any allocation in the receive path. (3) Jumbo DATA packet parsing is now done in data_ready context. Rather than cloning the packet once for each subpacket and pulling/trimming it, we file the packet multiple times with an annotation for each indicating which subpacket is there. From that we can directly calculate the offset and length. (4) A call's receive queue can be accessed without taking locks (memory barriers do have to be used, though). (5) Incoming calls are set up from preallocated resources and immediately made live. They can than have packets queued upon them and ACKs generated. If insufficient resources exist, DATA packet #1 is given a BUSY reply and other DATA packets are discarded). (6) sk_buffs no longer take a ref on their parent call. To make this work, the following changes are made: (1) Each call's receive buffer is now a circular buffer of sk_buff pointers (rxtx_buffer) rather than a number of sk_buff_heads spread between the call and the socket. This permits each sk_buff to be in the buffer multiple times. The receive buffer is reused for the transmit buffer. (2) A circular buffer of annotations (rxtx_annotations) is kept parallel to the data buffer. Transmission phase annotations indicate whether a buffered packet has been ACK'd or not and whether it needs retransmission. Receive phase annotations indicate whether a slot holds a whole packet or a jumbo subpacket and, if the latter, which subpacket. They also note whether the packet has been decrypted in place. (3) DATA packet window tracking is much simplified. Each phase has just two numbers representing the window (rx_hard_ack/rx_top and tx_hard_ack/tx_top). The hard_ack number is the sequence number before base of the window, representing the last packet the other side says it has consumed. hard_ack starts from 0 and the first packet is sequence number 1. The top number is the sequence number of the highest-numbered packet residing in the buffer. Packets between hard_ack+1 and top are soft-ACK'd to indicate they've been received, but not yet consumed. Four macros, before(), before_eq(), after() and after_eq() are added to compare sequence numbers within the window. This allows for the top of the window to wrap when the hard-ack sequence number gets close to the limit. Two flags, RXRPC_CALL_RX_LAST and RXRPC_CALL_TX_LAST, are added also to indicate when rx_top and tx_top point at the packets with the LAST_PACKET bit set, indicating the end of the phase. (4) Calls are queued on the socket 'receive queue' rather than packets. This means that we don't need have to invent dummy packets to queue to indicate abnormal/terminal states and we don't have to keep metadata packets (such as ABORTs) around (5) The offset and length of a (sub)packet's content are now passed to the verify_packet security op. This is currently expected to decrypt the packet in place and validate it. However, there's now nowhere to store the revised offset and length of the actual data within the decrypted blob (there may be a header and padding to skip) because an sk_buff may represent multiple packets, so a locate_data security op is added to retrieve these details from the sk_buff content when needed. (6) recvmsg() now has to handle jumbo subpackets, where each subpacket is individually secured and needs to be individually decrypted. The code to do this is broken out into rxrpc_recvmsg_data() and shared with the kernel API. It now iterates over the call's receive buffer rather than walking the socket receive queue. Additional changes: (1) The timers are condensed to a single timer that is set for the soonest of three timeouts (delayed ACK generation, DATA retransmission and call lifespan). (2) Transmission of ACK and ABORT packets is effected immediately from process-context socket ops/kernel API calls that cause them instead of them being punted off to a background work item. The data_ready handler still has to defer to the background, though. (3) A shutdown op is added to the AF_RXRPC socket so that the AFS filesystem can shut down the socket and flush its own work items before closing the socket to deal with any in-progress service calls. Future additional changes that will need to be considered: (1) Make sure that a call doesn't hog the front of the queue by receiving data from the network as fast as userspace is consuming it to the exclusion of other calls. (2) Transmit delayed ACKs from within recvmsg() when we've consumed sufficiently more packets to avoid the background work item needing to run. Signed-off-by: David Howells <dhowells@redhat.com>
2016-09-08 18:10:12 +08:00
* Schedule an instant Tx resend.
*/
rxrpc: Rewrite the data and ack handling code Rewrite the data and ack handling code such that: (1) Parsing of received ACK and ABORT packets and the distribution and the filing of DATA packets happens entirely within the data_ready context called from the UDP socket. This allows us to process and discard ACK and ABORT packets much more quickly (they're no longer stashed on a queue for a background thread to process). (2) We avoid calling skb_clone(), pskb_pull() and pskb_trim(). We instead keep track of the offset and length of the content of each packet in the sk_buff metadata. This means we don't do any allocation in the receive path. (3) Jumbo DATA packet parsing is now done in data_ready context. Rather than cloning the packet once for each subpacket and pulling/trimming it, we file the packet multiple times with an annotation for each indicating which subpacket is there. From that we can directly calculate the offset and length. (4) A call's receive queue can be accessed without taking locks (memory barriers do have to be used, though). (5) Incoming calls are set up from preallocated resources and immediately made live. They can than have packets queued upon them and ACKs generated. If insufficient resources exist, DATA packet #1 is given a BUSY reply and other DATA packets are discarded). (6) sk_buffs no longer take a ref on their parent call. To make this work, the following changes are made: (1) Each call's receive buffer is now a circular buffer of sk_buff pointers (rxtx_buffer) rather than a number of sk_buff_heads spread between the call and the socket. This permits each sk_buff to be in the buffer multiple times. The receive buffer is reused for the transmit buffer. (2) A circular buffer of annotations (rxtx_annotations) is kept parallel to the data buffer. Transmission phase annotations indicate whether a buffered packet has been ACK'd or not and whether it needs retransmission. Receive phase annotations indicate whether a slot holds a whole packet or a jumbo subpacket and, if the latter, which subpacket. They also note whether the packet has been decrypted in place. (3) DATA packet window tracking is much simplified. Each phase has just two numbers representing the window (rx_hard_ack/rx_top and tx_hard_ack/tx_top). The hard_ack number is the sequence number before base of the window, representing the last packet the other side says it has consumed. hard_ack starts from 0 and the first packet is sequence number 1. The top number is the sequence number of the highest-numbered packet residing in the buffer. Packets between hard_ack+1 and top are soft-ACK'd to indicate they've been received, but not yet consumed. Four macros, before(), before_eq(), after() and after_eq() are added to compare sequence numbers within the window. This allows for the top of the window to wrap when the hard-ack sequence number gets close to the limit. Two flags, RXRPC_CALL_RX_LAST and RXRPC_CALL_TX_LAST, are added also to indicate when rx_top and tx_top point at the packets with the LAST_PACKET bit set, indicating the end of the phase. (4) Calls are queued on the socket 'receive queue' rather than packets. This means that we don't need have to invent dummy packets to queue to indicate abnormal/terminal states and we don't have to keep metadata packets (such as ABORTs) around (5) The offset and length of a (sub)packet's content are now passed to the verify_packet security op. This is currently expected to decrypt the packet in place and validate it. However, there's now nowhere to store the revised offset and length of the actual data within the decrypted blob (there may be a header and padding to skip) because an sk_buff may represent multiple packets, so a locate_data security op is added to retrieve these details from the sk_buff content when needed. (6) recvmsg() now has to handle jumbo subpackets, where each subpacket is individually secured and needs to be individually decrypted. The code to do this is broken out into rxrpc_recvmsg_data() and shared with the kernel API. It now iterates over the call's receive buffer rather than walking the socket receive queue. Additional changes: (1) The timers are condensed to a single timer that is set for the soonest of three timeouts (delayed ACK generation, DATA retransmission and call lifespan). (2) Transmission of ACK and ABORT packets is effected immediately from process-context socket ops/kernel API calls that cause them instead of them being punted off to a background work item. The data_ready handler still has to defer to the background, though. (3) A shutdown op is added to the AF_RXRPC socket so that the AFS filesystem can shut down the socket and flush its own work items before closing the socket to deal with any in-progress service calls. Future additional changes that will need to be considered: (1) Make sure that a call doesn't hog the front of the queue by receiving data from the network as fast as userspace is consuming it to the exclusion of other calls. (2) Transmit delayed ACKs from within recvmsg() when we've consumed sufficiently more packets to avoid the background work item needing to run. Signed-off-by: David Howells <dhowells@redhat.com>
2016-09-08 18:10:12 +08:00
static inline void rxrpc_instant_resend(struct rxrpc_call *call, int ix)
{
rxrpc: Rewrite the data and ack handling code Rewrite the data and ack handling code such that: (1) Parsing of received ACK and ABORT packets and the distribution and the filing of DATA packets happens entirely within the data_ready context called from the UDP socket. This allows us to process and discard ACK and ABORT packets much more quickly (they're no longer stashed on a queue for a background thread to process). (2) We avoid calling skb_clone(), pskb_pull() and pskb_trim(). We instead keep track of the offset and length of the content of each packet in the sk_buff metadata. This means we don't do any allocation in the receive path. (3) Jumbo DATA packet parsing is now done in data_ready context. Rather than cloning the packet once for each subpacket and pulling/trimming it, we file the packet multiple times with an annotation for each indicating which subpacket is there. From that we can directly calculate the offset and length. (4) A call's receive queue can be accessed without taking locks (memory barriers do have to be used, though). (5) Incoming calls are set up from preallocated resources and immediately made live. They can than have packets queued upon them and ACKs generated. If insufficient resources exist, DATA packet #1 is given a BUSY reply and other DATA packets are discarded). (6) sk_buffs no longer take a ref on their parent call. To make this work, the following changes are made: (1) Each call's receive buffer is now a circular buffer of sk_buff pointers (rxtx_buffer) rather than a number of sk_buff_heads spread between the call and the socket. This permits each sk_buff to be in the buffer multiple times. The receive buffer is reused for the transmit buffer. (2) A circular buffer of annotations (rxtx_annotations) is kept parallel to the data buffer. Transmission phase annotations indicate whether a buffered packet has been ACK'd or not and whether it needs retransmission. Receive phase annotations indicate whether a slot holds a whole packet or a jumbo subpacket and, if the latter, which subpacket. They also note whether the packet has been decrypted in place. (3) DATA packet window tracking is much simplified. Each phase has just two numbers representing the window (rx_hard_ack/rx_top and tx_hard_ack/tx_top). The hard_ack number is the sequence number before base of the window, representing the last packet the other side says it has consumed. hard_ack starts from 0 and the first packet is sequence number 1. The top number is the sequence number of the highest-numbered packet residing in the buffer. Packets between hard_ack+1 and top are soft-ACK'd to indicate they've been received, but not yet consumed. Four macros, before(), before_eq(), after() and after_eq() are added to compare sequence numbers within the window. This allows for the top of the window to wrap when the hard-ack sequence number gets close to the limit. Two flags, RXRPC_CALL_RX_LAST and RXRPC_CALL_TX_LAST, are added also to indicate when rx_top and tx_top point at the packets with the LAST_PACKET bit set, indicating the end of the phase. (4) Calls are queued on the socket 'receive queue' rather than packets. This means that we don't need have to invent dummy packets to queue to indicate abnormal/terminal states and we don't have to keep metadata packets (such as ABORTs) around (5) The offset and length of a (sub)packet's content are now passed to the verify_packet security op. This is currently expected to decrypt the packet in place and validate it. However, there's now nowhere to store the revised offset and length of the actual data within the decrypted blob (there may be a header and padding to skip) because an sk_buff may represent multiple packets, so a locate_data security op is added to retrieve these details from the sk_buff content when needed. (6) recvmsg() now has to handle jumbo subpackets, where each subpacket is individually secured and needs to be individually decrypted. The code to do this is broken out into rxrpc_recvmsg_data() and shared with the kernel API. It now iterates over the call's receive buffer rather than walking the socket receive queue. Additional changes: (1) The timers are condensed to a single timer that is set for the soonest of three timeouts (delayed ACK generation, DATA retransmission and call lifespan). (2) Transmission of ACK and ABORT packets is effected immediately from process-context socket ops/kernel API calls that cause them instead of them being punted off to a background work item. The data_ready handler still has to defer to the background, though. (3) A shutdown op is added to the AF_RXRPC socket so that the AFS filesystem can shut down the socket and flush its own work items before closing the socket to deal with any in-progress service calls. Future additional changes that will need to be considered: (1) Make sure that a call doesn't hog the front of the queue by receiving data from the network as fast as userspace is consuming it to the exclusion of other calls. (2) Transmit delayed ACKs from within recvmsg() when we've consumed sufficiently more packets to avoid the background work item needing to run. Signed-off-by: David Howells <dhowells@redhat.com>
2016-09-08 18:10:12 +08:00
spin_lock_bh(&call->lock);
if (call->state < RXRPC_CALL_COMPLETE) {
call->rxtx_annotations[ix] =
(call->rxtx_annotations[ix] & RXRPC_TX_ANNO_LAST) |
RXRPC_TX_ANNO_RETRANS;
rxrpc: Rewrite the data and ack handling code Rewrite the data and ack handling code such that: (1) Parsing of received ACK and ABORT packets and the distribution and the filing of DATA packets happens entirely within the data_ready context called from the UDP socket. This allows us to process and discard ACK and ABORT packets much more quickly (they're no longer stashed on a queue for a background thread to process). (2) We avoid calling skb_clone(), pskb_pull() and pskb_trim(). We instead keep track of the offset and length of the content of each packet in the sk_buff metadata. This means we don't do any allocation in the receive path. (3) Jumbo DATA packet parsing is now done in data_ready context. Rather than cloning the packet once for each subpacket and pulling/trimming it, we file the packet multiple times with an annotation for each indicating which subpacket is there. From that we can directly calculate the offset and length. (4) A call's receive queue can be accessed without taking locks (memory barriers do have to be used, though). (5) Incoming calls are set up from preallocated resources and immediately made live. They can than have packets queued upon them and ACKs generated. If insufficient resources exist, DATA packet #1 is given a BUSY reply and other DATA packets are discarded). (6) sk_buffs no longer take a ref on their parent call. To make this work, the following changes are made: (1) Each call's receive buffer is now a circular buffer of sk_buff pointers (rxtx_buffer) rather than a number of sk_buff_heads spread between the call and the socket. This permits each sk_buff to be in the buffer multiple times. The receive buffer is reused for the transmit buffer. (2) A circular buffer of annotations (rxtx_annotations) is kept parallel to the data buffer. Transmission phase annotations indicate whether a buffered packet has been ACK'd or not and whether it needs retransmission. Receive phase annotations indicate whether a slot holds a whole packet or a jumbo subpacket and, if the latter, which subpacket. They also note whether the packet has been decrypted in place. (3) DATA packet window tracking is much simplified. Each phase has just two numbers representing the window (rx_hard_ack/rx_top and tx_hard_ack/tx_top). The hard_ack number is the sequence number before base of the window, representing the last packet the other side says it has consumed. hard_ack starts from 0 and the first packet is sequence number 1. The top number is the sequence number of the highest-numbered packet residing in the buffer. Packets between hard_ack+1 and top are soft-ACK'd to indicate they've been received, but not yet consumed. Four macros, before(), before_eq(), after() and after_eq() are added to compare sequence numbers within the window. This allows for the top of the window to wrap when the hard-ack sequence number gets close to the limit. Two flags, RXRPC_CALL_RX_LAST and RXRPC_CALL_TX_LAST, are added also to indicate when rx_top and tx_top point at the packets with the LAST_PACKET bit set, indicating the end of the phase. (4) Calls are queued on the socket 'receive queue' rather than packets. This means that we don't need have to invent dummy packets to queue to indicate abnormal/terminal states and we don't have to keep metadata packets (such as ABORTs) around (5) The offset and length of a (sub)packet's content are now passed to the verify_packet security op. This is currently expected to decrypt the packet in place and validate it. However, there's now nowhere to store the revised offset and length of the actual data within the decrypted blob (there may be a header and padding to skip) because an sk_buff may represent multiple packets, so a locate_data security op is added to retrieve these details from the sk_buff content when needed. (6) recvmsg() now has to handle jumbo subpackets, where each subpacket is individually secured and needs to be individually decrypted. The code to do this is broken out into rxrpc_recvmsg_data() and shared with the kernel API. It now iterates over the call's receive buffer rather than walking the socket receive queue. Additional changes: (1) The timers are condensed to a single timer that is set for the soonest of three timeouts (delayed ACK generation, DATA retransmission and call lifespan). (2) Transmission of ACK and ABORT packets is effected immediately from process-context socket ops/kernel API calls that cause them instead of them being punted off to a background work item. The data_ready handler still has to defer to the background, though. (3) A shutdown op is added to the AF_RXRPC socket so that the AFS filesystem can shut down the socket and flush its own work items before closing the socket to deal with any in-progress service calls. Future additional changes that will need to be considered: (1) Make sure that a call doesn't hog the front of the queue by receiving data from the network as fast as userspace is consuming it to the exclusion of other calls. (2) Transmit delayed ACKs from within recvmsg() when we've consumed sufficiently more packets to avoid the background work item needing to run. Signed-off-by: David Howells <dhowells@redhat.com>
2016-09-08 18:10:12 +08:00
if (!test_and_set_bit(RXRPC_CALL_EV_RESEND, &call->events))
rxrpc_queue_call(call);
}
rxrpc: Rewrite the data and ack handling code Rewrite the data and ack handling code such that: (1) Parsing of received ACK and ABORT packets and the distribution and the filing of DATA packets happens entirely within the data_ready context called from the UDP socket. This allows us to process and discard ACK and ABORT packets much more quickly (they're no longer stashed on a queue for a background thread to process). (2) We avoid calling skb_clone(), pskb_pull() and pskb_trim(). We instead keep track of the offset and length of the content of each packet in the sk_buff metadata. This means we don't do any allocation in the receive path. (3) Jumbo DATA packet parsing is now done in data_ready context. Rather than cloning the packet once for each subpacket and pulling/trimming it, we file the packet multiple times with an annotation for each indicating which subpacket is there. From that we can directly calculate the offset and length. (4) A call's receive queue can be accessed without taking locks (memory barriers do have to be used, though). (5) Incoming calls are set up from preallocated resources and immediately made live. They can than have packets queued upon them and ACKs generated. If insufficient resources exist, DATA packet #1 is given a BUSY reply and other DATA packets are discarded). (6) sk_buffs no longer take a ref on their parent call. To make this work, the following changes are made: (1) Each call's receive buffer is now a circular buffer of sk_buff pointers (rxtx_buffer) rather than a number of sk_buff_heads spread between the call and the socket. This permits each sk_buff to be in the buffer multiple times. The receive buffer is reused for the transmit buffer. (2) A circular buffer of annotations (rxtx_annotations) is kept parallel to the data buffer. Transmission phase annotations indicate whether a buffered packet has been ACK'd or not and whether it needs retransmission. Receive phase annotations indicate whether a slot holds a whole packet or a jumbo subpacket and, if the latter, which subpacket. They also note whether the packet has been decrypted in place. (3) DATA packet window tracking is much simplified. Each phase has just two numbers representing the window (rx_hard_ack/rx_top and tx_hard_ack/tx_top). The hard_ack number is the sequence number before base of the window, representing the last packet the other side says it has consumed. hard_ack starts from 0 and the first packet is sequence number 1. The top number is the sequence number of the highest-numbered packet residing in the buffer. Packets between hard_ack+1 and top are soft-ACK'd to indicate they've been received, but not yet consumed. Four macros, before(), before_eq(), after() and after_eq() are added to compare sequence numbers within the window. This allows for the top of the window to wrap when the hard-ack sequence number gets close to the limit. Two flags, RXRPC_CALL_RX_LAST and RXRPC_CALL_TX_LAST, are added also to indicate when rx_top and tx_top point at the packets with the LAST_PACKET bit set, indicating the end of the phase. (4) Calls are queued on the socket 'receive queue' rather than packets. This means that we don't need have to invent dummy packets to queue to indicate abnormal/terminal states and we don't have to keep metadata packets (such as ABORTs) around (5) The offset and length of a (sub)packet's content are now passed to the verify_packet security op. This is currently expected to decrypt the packet in place and validate it. However, there's now nowhere to store the revised offset and length of the actual data within the decrypted blob (there may be a header and padding to skip) because an sk_buff may represent multiple packets, so a locate_data security op is added to retrieve these details from the sk_buff content when needed. (6) recvmsg() now has to handle jumbo subpackets, where each subpacket is individually secured and needs to be individually decrypted. The code to do this is broken out into rxrpc_recvmsg_data() and shared with the kernel API. It now iterates over the call's receive buffer rather than walking the socket receive queue. Additional changes: (1) The timers are condensed to a single timer that is set for the soonest of three timeouts (delayed ACK generation, DATA retransmission and call lifespan). (2) Transmission of ACK and ABORT packets is effected immediately from process-context socket ops/kernel API calls that cause them instead of them being punted off to a background work item. The data_ready handler still has to defer to the background, though. (3) A shutdown op is added to the AF_RXRPC socket so that the AFS filesystem can shut down the socket and flush its own work items before closing the socket to deal with any in-progress service calls. Future additional changes that will need to be considered: (1) Make sure that a call doesn't hog the front of the queue by receiving data from the network as fast as userspace is consuming it to the exclusion of other calls. (2) Transmit delayed ACKs from within recvmsg() when we've consumed sufficiently more packets to avoid the background work item needing to run. Signed-off-by: David Howells <dhowells@redhat.com>
2016-09-08 18:10:12 +08:00
spin_unlock_bh(&call->lock);
}
/*
* Notify the owner of the call that the transmit phase is ended and the last
* packet has been queued.
*/
static void rxrpc_notify_end_tx(struct rxrpc_sock *rx, struct rxrpc_call *call,
rxrpc_notify_end_tx_t notify_end_tx)
{
if (notify_end_tx)
notify_end_tx(&rx->sk, call, call->user_call_ID);
}
/*
* Queue a DATA packet for transmission, set the resend timeout and send
* the packet immediately. Returns the error from rxrpc_send_data_packet()
* in case the caller wants to do something with it.
*/
static int rxrpc_queue_packet(struct rxrpc_sock *rx, struct rxrpc_call *call,
struct sk_buff *skb, bool last,
rxrpc_notify_end_tx_t notify_end_tx)
{
struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
rxrpc: Fix call timeouts Fix the rxrpc call expiration timeouts and make them settable from userspace. By analogy with other rx implementations, there should be three timeouts: (1) "Normal timeout" This is set for all calls and is triggered if we haven't received any packets from the peer in a while. It is measured from the last time we received any packet on that call. This is not reset by any connection packets (such as CHALLENGE/RESPONSE packets). If a service operation takes a long time, the server should generate PING ACKs at a duration that's substantially less than the normal timeout so is to keep both sides alive. This is set at 1/6 of normal timeout. (2) "Idle timeout" This is set only for a service call and is triggered if we stop receiving the DATA packets that comprise the request data. It is measured from the last time we received a DATA packet. (3) "Hard timeout" This can be set for a call and specified the maximum lifetime of that call. It should not be specified by default. Some operations (such as volume transfer) take a long time. Allow userspace to set/change the timeouts on a call with sendmsg, using a control message: RXRPC_SET_CALL_TIMEOUTS The data to the message is a number of 32-bit words, not all of which need be given: u32 hard_timeout; /* sec from first packet */ u32 idle_timeout; /* msec from packet Rx */ u32 normal_timeout; /* msec from data Rx */ This can be set in combination with any other sendmsg() that affects a call. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-24 18:18:41 +08:00
unsigned long now;
rxrpc: Rewrite the data and ack handling code Rewrite the data and ack handling code such that: (1) Parsing of received ACK and ABORT packets and the distribution and the filing of DATA packets happens entirely within the data_ready context called from the UDP socket. This allows us to process and discard ACK and ABORT packets much more quickly (they're no longer stashed on a queue for a background thread to process). (2) We avoid calling skb_clone(), pskb_pull() and pskb_trim(). We instead keep track of the offset and length of the content of each packet in the sk_buff metadata. This means we don't do any allocation in the receive path. (3) Jumbo DATA packet parsing is now done in data_ready context. Rather than cloning the packet once for each subpacket and pulling/trimming it, we file the packet multiple times with an annotation for each indicating which subpacket is there. From that we can directly calculate the offset and length. (4) A call's receive queue can be accessed without taking locks (memory barriers do have to be used, though). (5) Incoming calls are set up from preallocated resources and immediately made live. They can than have packets queued upon them and ACKs generated. If insufficient resources exist, DATA packet #1 is given a BUSY reply and other DATA packets are discarded). (6) sk_buffs no longer take a ref on their parent call. To make this work, the following changes are made: (1) Each call's receive buffer is now a circular buffer of sk_buff pointers (rxtx_buffer) rather than a number of sk_buff_heads spread between the call and the socket. This permits each sk_buff to be in the buffer multiple times. The receive buffer is reused for the transmit buffer. (2) A circular buffer of annotations (rxtx_annotations) is kept parallel to the data buffer. Transmission phase annotations indicate whether a buffered packet has been ACK'd or not and whether it needs retransmission. Receive phase annotations indicate whether a slot holds a whole packet or a jumbo subpacket and, if the latter, which subpacket. They also note whether the packet has been decrypted in place. (3) DATA packet window tracking is much simplified. Each phase has just two numbers representing the window (rx_hard_ack/rx_top and tx_hard_ack/tx_top). The hard_ack number is the sequence number before base of the window, representing the last packet the other side says it has consumed. hard_ack starts from 0 and the first packet is sequence number 1. The top number is the sequence number of the highest-numbered packet residing in the buffer. Packets between hard_ack+1 and top are soft-ACK'd to indicate they've been received, but not yet consumed. Four macros, before(), before_eq(), after() and after_eq() are added to compare sequence numbers within the window. This allows for the top of the window to wrap when the hard-ack sequence number gets close to the limit. Two flags, RXRPC_CALL_RX_LAST and RXRPC_CALL_TX_LAST, are added also to indicate when rx_top and tx_top point at the packets with the LAST_PACKET bit set, indicating the end of the phase. (4) Calls are queued on the socket 'receive queue' rather than packets. This means that we don't need have to invent dummy packets to queue to indicate abnormal/terminal states and we don't have to keep metadata packets (such as ABORTs) around (5) The offset and length of a (sub)packet's content are now passed to the verify_packet security op. This is currently expected to decrypt the packet in place and validate it. However, there's now nowhere to store the revised offset and length of the actual data within the decrypted blob (there may be a header and padding to skip) because an sk_buff may represent multiple packets, so a locate_data security op is added to retrieve these details from the sk_buff content when needed. (6) recvmsg() now has to handle jumbo subpackets, where each subpacket is individually secured and needs to be individually decrypted. The code to do this is broken out into rxrpc_recvmsg_data() and shared with the kernel API. It now iterates over the call's receive buffer rather than walking the socket receive queue. Additional changes: (1) The timers are condensed to a single timer that is set for the soonest of three timeouts (delayed ACK generation, DATA retransmission and call lifespan). (2) Transmission of ACK and ABORT packets is effected immediately from process-context socket ops/kernel API calls that cause them instead of them being punted off to a background work item. The data_ready handler still has to defer to the background, though. (3) A shutdown op is added to the AF_RXRPC socket so that the AFS filesystem can shut down the socket and flush its own work items before closing the socket to deal with any in-progress service calls. Future additional changes that will need to be considered: (1) Make sure that a call doesn't hog the front of the queue by receiving data from the network as fast as userspace is consuming it to the exclusion of other calls. (2) Transmit delayed ACKs from within recvmsg() when we've consumed sufficiently more packets to avoid the background work item needing to run. Signed-off-by: David Howells <dhowells@redhat.com>
2016-09-08 18:10:12 +08:00
rxrpc_seq_t seq = sp->hdr.seq;
int ret, ix;
rxrpc: Pass the last Tx packet marker in the annotation buffer When the last packet of data to be transmitted on a call is queued, tx_top is set and then the RXRPC_CALL_TX_LAST flag is set. Unfortunately, this leaves a race in the ACK processing side of things because the flag affects the interpretation of tx_top and also allows us to start receiving reply data before we've finished transmitting. To fix this, make the following changes: (1) rxrpc_queue_packet() now sets a marker in the annotation buffer instead of setting the RXRPC_CALL_TX_LAST flag. (2) rxrpc_rotate_tx_window() detects the marker and sets the flag in the same context as the routines that use it. (3) rxrpc_end_tx_phase() is simplified to just shift the call state. The Tx window must have been rotated before calling to discard the last packet. (4) rxrpc_receiving_reply() is added to handle the arrival of the first DATA packet of a reply to a client call (which is an implicit ACK of the Tx phase). (5) The last part of rxrpc_input_ack() is reordered to perform Tx rotation, then soft-ACK application and then to end the phase if we've rotated the last packet. In the event of a terminal ACK, the soft-ACK application will be skipped as nAcks should be 0. (6) rxrpc_input_ackall() now has to rotate as well as ending the phase. In addition: (7) Alter the transmit tracepoint to log the rotation of the last packet. (8) Remove the no-longer relevant queue_reqack tracepoint note. The ACK-REQUESTED packet header flag is now set as needed when we actually transmit the packet and may vary by retransmission. Signed-off-by: David Howells <dhowells@redhat.com>
2016-09-23 19:39:22 +08:00
u8 annotation = RXRPC_TX_ANNO_UNACK;
rxrpc: Rewrite the data and ack handling code Rewrite the data and ack handling code such that: (1) Parsing of received ACK and ABORT packets and the distribution and the filing of DATA packets happens entirely within the data_ready context called from the UDP socket. This allows us to process and discard ACK and ABORT packets much more quickly (they're no longer stashed on a queue for a background thread to process). (2) We avoid calling skb_clone(), pskb_pull() and pskb_trim(). We instead keep track of the offset and length of the content of each packet in the sk_buff metadata. This means we don't do any allocation in the receive path. (3) Jumbo DATA packet parsing is now done in data_ready context. Rather than cloning the packet once for each subpacket and pulling/trimming it, we file the packet multiple times with an annotation for each indicating which subpacket is there. From that we can directly calculate the offset and length. (4) A call's receive queue can be accessed without taking locks (memory barriers do have to be used, though). (5) Incoming calls are set up from preallocated resources and immediately made live. They can than have packets queued upon them and ACKs generated. If insufficient resources exist, DATA packet #1 is given a BUSY reply and other DATA packets are discarded). (6) sk_buffs no longer take a ref on their parent call. To make this work, the following changes are made: (1) Each call's receive buffer is now a circular buffer of sk_buff pointers (rxtx_buffer) rather than a number of sk_buff_heads spread between the call and the socket. This permits each sk_buff to be in the buffer multiple times. The receive buffer is reused for the transmit buffer. (2) A circular buffer of annotations (rxtx_annotations) is kept parallel to the data buffer. Transmission phase annotations indicate whether a buffered packet has been ACK'd or not and whether it needs retransmission. Receive phase annotations indicate whether a slot holds a whole packet or a jumbo subpacket and, if the latter, which subpacket. They also note whether the packet has been decrypted in place. (3) DATA packet window tracking is much simplified. Each phase has just two numbers representing the window (rx_hard_ack/rx_top and tx_hard_ack/tx_top). The hard_ack number is the sequence number before base of the window, representing the last packet the other side says it has consumed. hard_ack starts from 0 and the first packet is sequence number 1. The top number is the sequence number of the highest-numbered packet residing in the buffer. Packets between hard_ack+1 and top are soft-ACK'd to indicate they've been received, but not yet consumed. Four macros, before(), before_eq(), after() and after_eq() are added to compare sequence numbers within the window. This allows for the top of the window to wrap when the hard-ack sequence number gets close to the limit. Two flags, RXRPC_CALL_RX_LAST and RXRPC_CALL_TX_LAST, are added also to indicate when rx_top and tx_top point at the packets with the LAST_PACKET bit set, indicating the end of the phase. (4) Calls are queued on the socket 'receive queue' rather than packets. This means that we don't need have to invent dummy packets to queue to indicate abnormal/terminal states and we don't have to keep metadata packets (such as ABORTs) around (5) The offset and length of a (sub)packet's content are now passed to the verify_packet security op. This is currently expected to decrypt the packet in place and validate it. However, there's now nowhere to store the revised offset and length of the actual data within the decrypted blob (there may be a header and padding to skip) because an sk_buff may represent multiple packets, so a locate_data security op is added to retrieve these details from the sk_buff content when needed. (6) recvmsg() now has to handle jumbo subpackets, where each subpacket is individually secured and needs to be individually decrypted. The code to do this is broken out into rxrpc_recvmsg_data() and shared with the kernel API. It now iterates over the call's receive buffer rather than walking the socket receive queue. Additional changes: (1) The timers are condensed to a single timer that is set for the soonest of three timeouts (delayed ACK generation, DATA retransmission and call lifespan). (2) Transmission of ACK and ABORT packets is effected immediately from process-context socket ops/kernel API calls that cause them instead of them being punted off to a background work item. The data_ready handler still has to defer to the background, though. (3) A shutdown op is added to the AF_RXRPC socket so that the AFS filesystem can shut down the socket and flush its own work items before closing the socket to deal with any in-progress service calls. Future additional changes that will need to be considered: (1) Make sure that a call doesn't hog the front of the queue by receiving data from the network as fast as userspace is consuming it to the exclusion of other calls. (2) Transmit delayed ACKs from within recvmsg() when we've consumed sufficiently more packets to avoid the background work item needing to run. Signed-off-by: David Howells <dhowells@redhat.com>
2016-09-08 18:10:12 +08:00
_net("queue skb %p [%d]", skb, seq);
rxrpc: Rewrite the data and ack handling code Rewrite the data and ack handling code such that: (1) Parsing of received ACK and ABORT packets and the distribution and the filing of DATA packets happens entirely within the data_ready context called from the UDP socket. This allows us to process and discard ACK and ABORT packets much more quickly (they're no longer stashed on a queue for a background thread to process). (2) We avoid calling skb_clone(), pskb_pull() and pskb_trim(). We instead keep track of the offset and length of the content of each packet in the sk_buff metadata. This means we don't do any allocation in the receive path. (3) Jumbo DATA packet parsing is now done in data_ready context. Rather than cloning the packet once for each subpacket and pulling/trimming it, we file the packet multiple times with an annotation for each indicating which subpacket is there. From that we can directly calculate the offset and length. (4) A call's receive queue can be accessed without taking locks (memory barriers do have to be used, though). (5) Incoming calls are set up from preallocated resources and immediately made live. They can than have packets queued upon them and ACKs generated. If insufficient resources exist, DATA packet #1 is given a BUSY reply and other DATA packets are discarded). (6) sk_buffs no longer take a ref on their parent call. To make this work, the following changes are made: (1) Each call's receive buffer is now a circular buffer of sk_buff pointers (rxtx_buffer) rather than a number of sk_buff_heads spread between the call and the socket. This permits each sk_buff to be in the buffer multiple times. The receive buffer is reused for the transmit buffer. (2) A circular buffer of annotations (rxtx_annotations) is kept parallel to the data buffer. Transmission phase annotations indicate whether a buffered packet has been ACK'd or not and whether it needs retransmission. Receive phase annotations indicate whether a slot holds a whole packet or a jumbo subpacket and, if the latter, which subpacket. They also note whether the packet has been decrypted in place. (3) DATA packet window tracking is much simplified. Each phase has just two numbers representing the window (rx_hard_ack/rx_top and tx_hard_ack/tx_top). The hard_ack number is the sequence number before base of the window, representing the last packet the other side says it has consumed. hard_ack starts from 0 and the first packet is sequence number 1. The top number is the sequence number of the highest-numbered packet residing in the buffer. Packets between hard_ack+1 and top are soft-ACK'd to indicate they've been received, but not yet consumed. Four macros, before(), before_eq(), after() and after_eq() are added to compare sequence numbers within the window. This allows for the top of the window to wrap when the hard-ack sequence number gets close to the limit. Two flags, RXRPC_CALL_RX_LAST and RXRPC_CALL_TX_LAST, are added also to indicate when rx_top and tx_top point at the packets with the LAST_PACKET bit set, indicating the end of the phase. (4) Calls are queued on the socket 'receive queue' rather than packets. This means that we don't need have to invent dummy packets to queue to indicate abnormal/terminal states and we don't have to keep metadata packets (such as ABORTs) around (5) The offset and length of a (sub)packet's content are now passed to the verify_packet security op. This is currently expected to decrypt the packet in place and validate it. However, there's now nowhere to store the revised offset and length of the actual data within the decrypted blob (there may be a header and padding to skip) because an sk_buff may represent multiple packets, so a locate_data security op is added to retrieve these details from the sk_buff content when needed. (6) recvmsg() now has to handle jumbo subpackets, where each subpacket is individually secured and needs to be individually decrypted. The code to do this is broken out into rxrpc_recvmsg_data() and shared with the kernel API. It now iterates over the call's receive buffer rather than walking the socket receive queue. Additional changes: (1) The timers are condensed to a single timer that is set for the soonest of three timeouts (delayed ACK generation, DATA retransmission and call lifespan). (2) Transmission of ACK and ABORT packets is effected immediately from process-context socket ops/kernel API calls that cause them instead of them being punted off to a background work item. The data_ready handler still has to defer to the background, though. (3) A shutdown op is added to the AF_RXRPC socket so that the AFS filesystem can shut down the socket and flush its own work items before closing the socket to deal with any in-progress service calls. Future additional changes that will need to be considered: (1) Make sure that a call doesn't hog the front of the queue by receiving data from the network as fast as userspace is consuming it to the exclusion of other calls. (2) Transmit delayed ACKs from within recvmsg() when we've consumed sufficiently more packets to avoid the background work item needing to run. Signed-off-by: David Howells <dhowells@redhat.com>
2016-09-08 18:10:12 +08:00
ASSERTCMP(seq, ==, call->tx_top + 1);
Revert "rxrpc: Allow failed client calls to be retried" The changes introduced to allow rxrpc calls to be retried creates an issue when it comes to refcounting afs_call structs. The problem is that when rxrpc_send_data() queues the last packet for an asynchronous call, the following sequence can occur: (1) The notify_end_tx callback is invoked which causes the state in the afs_call to be changed from AFS_CALL_CL_REQUESTING or AFS_CALL_SV_REPLYING. (2) afs_deliver_to_call() can then process event notifications from rxrpc on the async_work queue. (3) Delivery of events, such as an abort from the server, can cause the afs_call state to be changed to AFS_CALL_COMPLETE on async_work. (4) For an asynchronous call, afs_process_async_call() notes that the call is complete and tried to clean up all the refs on async_work. (5) rxrpc_send_data() might return the amount of data transferred (success) or an error - which could in turn reflect a local error or a received error. Synchronising the clean up after rxrpc_kernel_send_data() returns an error with the asynchronous cleanup is then tricky to get right. Mostly revert commit c038a58ccfd6704d4d7d60ed3d6a0fca13cf13a4. The two API functions the original commit added aren't currently used. This makes rxrpc_kernel_send_data() always return successfully if it queued the data it was given. Note that this doesn't affect synchronous calls since their Rx notification function merely pokes a wait queue and does not refcounting. The asynchronous call notification function *has* to do refcounting and pass a ref over the work item to avoid the need to sync the workqueue in call cleanup. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-01-11 00:59:13 +08:00
if (last)
rxrpc: Pass the last Tx packet marker in the annotation buffer When the last packet of data to be transmitted on a call is queued, tx_top is set and then the RXRPC_CALL_TX_LAST flag is set. Unfortunately, this leaves a race in the ACK processing side of things because the flag affects the interpretation of tx_top and also allows us to start receiving reply data before we've finished transmitting. To fix this, make the following changes: (1) rxrpc_queue_packet() now sets a marker in the annotation buffer instead of setting the RXRPC_CALL_TX_LAST flag. (2) rxrpc_rotate_tx_window() detects the marker and sets the flag in the same context as the routines that use it. (3) rxrpc_end_tx_phase() is simplified to just shift the call state. The Tx window must have been rotated before calling to discard the last packet. (4) rxrpc_receiving_reply() is added to handle the arrival of the first DATA packet of a reply to a client call (which is an implicit ACK of the Tx phase). (5) The last part of rxrpc_input_ack() is reordered to perform Tx rotation, then soft-ACK application and then to end the phase if we've rotated the last packet. In the event of a terminal ACK, the soft-ACK application will be skipped as nAcks should be 0. (6) rxrpc_input_ackall() now has to rotate as well as ending the phase. In addition: (7) Alter the transmit tracepoint to log the rotation of the last packet. (8) Remove the no-longer relevant queue_reqack tracepoint note. The ACK-REQUESTED packet header flag is now set as needed when we actually transmit the packet and may vary by retransmission. Signed-off-by: David Howells <dhowells@redhat.com>
2016-09-23 19:39:22 +08:00
annotation |= RXRPC_TX_ANNO_LAST;
/* We have to set the timestamp before queueing as the retransmit
* algorithm can see the packet as soon as we queue it.
*/
skb->tstamp = ktime_get_real();
rxrpc: Rewrite the data and ack handling code Rewrite the data and ack handling code such that: (1) Parsing of received ACK and ABORT packets and the distribution and the filing of DATA packets happens entirely within the data_ready context called from the UDP socket. This allows us to process and discard ACK and ABORT packets much more quickly (they're no longer stashed on a queue for a background thread to process). (2) We avoid calling skb_clone(), pskb_pull() and pskb_trim(). We instead keep track of the offset and length of the content of each packet in the sk_buff metadata. This means we don't do any allocation in the receive path. (3) Jumbo DATA packet parsing is now done in data_ready context. Rather than cloning the packet once for each subpacket and pulling/trimming it, we file the packet multiple times with an annotation for each indicating which subpacket is there. From that we can directly calculate the offset and length. (4) A call's receive queue can be accessed without taking locks (memory barriers do have to be used, though). (5) Incoming calls are set up from preallocated resources and immediately made live. They can than have packets queued upon them and ACKs generated. If insufficient resources exist, DATA packet #1 is given a BUSY reply and other DATA packets are discarded). (6) sk_buffs no longer take a ref on their parent call. To make this work, the following changes are made: (1) Each call's receive buffer is now a circular buffer of sk_buff pointers (rxtx_buffer) rather than a number of sk_buff_heads spread between the call and the socket. This permits each sk_buff to be in the buffer multiple times. The receive buffer is reused for the transmit buffer. (2) A circular buffer of annotations (rxtx_annotations) is kept parallel to the data buffer. Transmission phase annotations indicate whether a buffered packet has been ACK'd or not and whether it needs retransmission. Receive phase annotations indicate whether a slot holds a whole packet or a jumbo subpacket and, if the latter, which subpacket. They also note whether the packet has been decrypted in place. (3) DATA packet window tracking is much simplified. Each phase has just two numbers representing the window (rx_hard_ack/rx_top and tx_hard_ack/tx_top). The hard_ack number is the sequence number before base of the window, representing the last packet the other side says it has consumed. hard_ack starts from 0 and the first packet is sequence number 1. The top number is the sequence number of the highest-numbered packet residing in the buffer. Packets between hard_ack+1 and top are soft-ACK'd to indicate they've been received, but not yet consumed. Four macros, before(), before_eq(), after() and after_eq() are added to compare sequence numbers within the window. This allows for the top of the window to wrap when the hard-ack sequence number gets close to the limit. Two flags, RXRPC_CALL_RX_LAST and RXRPC_CALL_TX_LAST, are added also to indicate when rx_top and tx_top point at the packets with the LAST_PACKET bit set, indicating the end of the phase. (4) Calls are queued on the socket 'receive queue' rather than packets. This means that we don't need have to invent dummy packets to queue to indicate abnormal/terminal states and we don't have to keep metadata packets (such as ABORTs) around (5) The offset and length of a (sub)packet's content are now passed to the verify_packet security op. This is currently expected to decrypt the packet in place and validate it. However, there's now nowhere to store the revised offset and length of the actual data within the decrypted blob (there may be a header and padding to skip) because an sk_buff may represent multiple packets, so a locate_data security op is added to retrieve these details from the sk_buff content when needed. (6) recvmsg() now has to handle jumbo subpackets, where each subpacket is individually secured and needs to be individually decrypted. The code to do this is broken out into rxrpc_recvmsg_data() and shared with the kernel API. It now iterates over the call's receive buffer rather than walking the socket receive queue. Additional changes: (1) The timers are condensed to a single timer that is set for the soonest of three timeouts (delayed ACK generation, DATA retransmission and call lifespan). (2) Transmission of ACK and ABORT packets is effected immediately from process-context socket ops/kernel API calls that cause them instead of them being punted off to a background work item. The data_ready handler still has to defer to the background, though. (3) A shutdown op is added to the AF_RXRPC socket so that the AFS filesystem can shut down the socket and flush its own work items before closing the socket to deal with any in-progress service calls. Future additional changes that will need to be considered: (1) Make sure that a call doesn't hog the front of the queue by receiving data from the network as fast as userspace is consuming it to the exclusion of other calls. (2) Transmit delayed ACKs from within recvmsg() when we've consumed sufficiently more packets to avoid the background work item needing to run. Signed-off-by: David Howells <dhowells@redhat.com>
2016-09-08 18:10:12 +08:00
ix = seq & RXRPC_RXTX_BUFF_MASK;
rxrpc_get_skb(skb, rxrpc_skb_got);
rxrpc: Pass the last Tx packet marker in the annotation buffer When the last packet of data to be transmitted on a call is queued, tx_top is set and then the RXRPC_CALL_TX_LAST flag is set. Unfortunately, this leaves a race in the ACK processing side of things because the flag affects the interpretation of tx_top and also allows us to start receiving reply data before we've finished transmitting. To fix this, make the following changes: (1) rxrpc_queue_packet() now sets a marker in the annotation buffer instead of setting the RXRPC_CALL_TX_LAST flag. (2) rxrpc_rotate_tx_window() detects the marker and sets the flag in the same context as the routines that use it. (3) rxrpc_end_tx_phase() is simplified to just shift the call state. The Tx window must have been rotated before calling to discard the last packet. (4) rxrpc_receiving_reply() is added to handle the arrival of the first DATA packet of a reply to a client call (which is an implicit ACK of the Tx phase). (5) The last part of rxrpc_input_ack() is reordered to perform Tx rotation, then soft-ACK application and then to end the phase if we've rotated the last packet. In the event of a terminal ACK, the soft-ACK application will be skipped as nAcks should be 0. (6) rxrpc_input_ackall() now has to rotate as well as ending the phase. In addition: (7) Alter the transmit tracepoint to log the rotation of the last packet. (8) Remove the no-longer relevant queue_reqack tracepoint note. The ACK-REQUESTED packet header flag is now set as needed when we actually transmit the packet and may vary by retransmission. Signed-off-by: David Howells <dhowells@redhat.com>
2016-09-23 19:39:22 +08:00
call->rxtx_annotations[ix] = annotation;
smp_wmb();
rxrpc: Rewrite the data and ack handling code Rewrite the data and ack handling code such that: (1) Parsing of received ACK and ABORT packets and the distribution and the filing of DATA packets happens entirely within the data_ready context called from the UDP socket. This allows us to process and discard ACK and ABORT packets much more quickly (they're no longer stashed on a queue for a background thread to process). (2) We avoid calling skb_clone(), pskb_pull() and pskb_trim(). We instead keep track of the offset and length of the content of each packet in the sk_buff metadata. This means we don't do any allocation in the receive path. (3) Jumbo DATA packet parsing is now done in data_ready context. Rather than cloning the packet once for each subpacket and pulling/trimming it, we file the packet multiple times with an annotation for each indicating which subpacket is there. From that we can directly calculate the offset and length. (4) A call's receive queue can be accessed without taking locks (memory barriers do have to be used, though). (5) Incoming calls are set up from preallocated resources and immediately made live. They can than have packets queued upon them and ACKs generated. If insufficient resources exist, DATA packet #1 is given a BUSY reply and other DATA packets are discarded). (6) sk_buffs no longer take a ref on their parent call. To make this work, the following changes are made: (1) Each call's receive buffer is now a circular buffer of sk_buff pointers (rxtx_buffer) rather than a number of sk_buff_heads spread between the call and the socket. This permits each sk_buff to be in the buffer multiple times. The receive buffer is reused for the transmit buffer. (2) A circular buffer of annotations (rxtx_annotations) is kept parallel to the data buffer. Transmission phase annotations indicate whether a buffered packet has been ACK'd or not and whether it needs retransmission. Receive phase annotations indicate whether a slot holds a whole packet or a jumbo subpacket and, if the latter, which subpacket. They also note whether the packet has been decrypted in place. (3) DATA packet window tracking is much simplified. Each phase has just two numbers representing the window (rx_hard_ack/rx_top and tx_hard_ack/tx_top). The hard_ack number is the sequence number before base of the window, representing the last packet the other side says it has consumed. hard_ack starts from 0 and the first packet is sequence number 1. The top number is the sequence number of the highest-numbered packet residing in the buffer. Packets between hard_ack+1 and top are soft-ACK'd to indicate they've been received, but not yet consumed. Four macros, before(), before_eq(), after() and after_eq() are added to compare sequence numbers within the window. This allows for the top of the window to wrap when the hard-ack sequence number gets close to the limit. Two flags, RXRPC_CALL_RX_LAST and RXRPC_CALL_TX_LAST, are added also to indicate when rx_top and tx_top point at the packets with the LAST_PACKET bit set, indicating the end of the phase. (4) Calls are queued on the socket 'receive queue' rather than packets. This means that we don't need have to invent dummy packets to queue to indicate abnormal/terminal states and we don't have to keep metadata packets (such as ABORTs) around (5) The offset and length of a (sub)packet's content are now passed to the verify_packet security op. This is currently expected to decrypt the packet in place and validate it. However, there's now nowhere to store the revised offset and length of the actual data within the decrypted blob (there may be a header and padding to skip) because an sk_buff may represent multiple packets, so a locate_data security op is added to retrieve these details from the sk_buff content when needed. (6) recvmsg() now has to handle jumbo subpackets, where each subpacket is individually secured and needs to be individually decrypted. The code to do this is broken out into rxrpc_recvmsg_data() and shared with the kernel API. It now iterates over the call's receive buffer rather than walking the socket receive queue. Additional changes: (1) The timers are condensed to a single timer that is set for the soonest of three timeouts (delayed ACK generation, DATA retransmission and call lifespan). (2) Transmission of ACK and ABORT packets is effected immediately from process-context socket ops/kernel API calls that cause them instead of them being punted off to a background work item. The data_ready handler still has to defer to the background, though. (3) A shutdown op is added to the AF_RXRPC socket so that the AFS filesystem can shut down the socket and flush its own work items before closing the socket to deal with any in-progress service calls. Future additional changes that will need to be considered: (1) Make sure that a call doesn't hog the front of the queue by receiving data from the network as fast as userspace is consuming it to the exclusion of other calls. (2) Transmit delayed ACKs from within recvmsg() when we've consumed sufficiently more packets to avoid the background work item needing to run. Signed-off-by: David Howells <dhowells@redhat.com>
2016-09-08 18:10:12 +08:00
call->rxtx_buffer[ix] = skb;
call->tx_top = seq;
rxrpc: Pass the last Tx packet marker in the annotation buffer When the last packet of data to be transmitted on a call is queued, tx_top is set and then the RXRPC_CALL_TX_LAST flag is set. Unfortunately, this leaves a race in the ACK processing side of things because the flag affects the interpretation of tx_top and also allows us to start receiving reply data before we've finished transmitting. To fix this, make the following changes: (1) rxrpc_queue_packet() now sets a marker in the annotation buffer instead of setting the RXRPC_CALL_TX_LAST flag. (2) rxrpc_rotate_tx_window() detects the marker and sets the flag in the same context as the routines that use it. (3) rxrpc_end_tx_phase() is simplified to just shift the call state. The Tx window must have been rotated before calling to discard the last packet. (4) rxrpc_receiving_reply() is added to handle the arrival of the first DATA packet of a reply to a client call (which is an implicit ACK of the Tx phase). (5) The last part of rxrpc_input_ack() is reordered to perform Tx rotation, then soft-ACK application and then to end the phase if we've rotated the last packet. In the event of a terminal ACK, the soft-ACK application will be skipped as nAcks should be 0. (6) rxrpc_input_ackall() now has to rotate as well as ending the phase. In addition: (7) Alter the transmit tracepoint to log the rotation of the last packet. (8) Remove the no-longer relevant queue_reqack tracepoint note. The ACK-REQUESTED packet header flag is now set as needed when we actually transmit the packet and may vary by retransmission. Signed-off-by: David Howells <dhowells@redhat.com>
2016-09-23 19:39:22 +08:00
if (last)
trace_rxrpc_transmit(call, rxrpc_transmit_queue_last);
rxrpc: Pass the last Tx packet marker in the annotation buffer When the last packet of data to be transmitted on a call is queued, tx_top is set and then the RXRPC_CALL_TX_LAST flag is set. Unfortunately, this leaves a race in the ACK processing side of things because the flag affects the interpretation of tx_top and also allows us to start receiving reply data before we've finished transmitting. To fix this, make the following changes: (1) rxrpc_queue_packet() now sets a marker in the annotation buffer instead of setting the RXRPC_CALL_TX_LAST flag. (2) rxrpc_rotate_tx_window() detects the marker and sets the flag in the same context as the routines that use it. (3) rxrpc_end_tx_phase() is simplified to just shift the call state. The Tx window must have been rotated before calling to discard the last packet. (4) rxrpc_receiving_reply() is added to handle the arrival of the first DATA packet of a reply to a client call (which is an implicit ACK of the Tx phase). (5) The last part of rxrpc_input_ack() is reordered to perform Tx rotation, then soft-ACK application and then to end the phase if we've rotated the last packet. In the event of a terminal ACK, the soft-ACK application will be skipped as nAcks should be 0. (6) rxrpc_input_ackall() now has to rotate as well as ending the phase. In addition: (7) Alter the transmit tracepoint to log the rotation of the last packet. (8) Remove the no-longer relevant queue_reqack tracepoint note. The ACK-REQUESTED packet header flag is now set as needed when we actually transmit the packet and may vary by retransmission. Signed-off-by: David Howells <dhowells@redhat.com>
2016-09-23 19:39:22 +08:00
else
trace_rxrpc_transmit(call, rxrpc_transmit_queue);
if (last || call->state == RXRPC_CALL_SERVER_ACK_REQUEST) {
_debug("________awaiting reply/ACK__________");
write_lock_bh(&call->state_lock);
switch (call->state) {
case RXRPC_CALL_CLIENT_SEND_REQUEST:
call->state = RXRPC_CALL_CLIENT_AWAIT_REPLY;
rxrpc_notify_end_tx(rx, call, notify_end_tx);
break;
case RXRPC_CALL_SERVER_ACK_REQUEST:
call->state = RXRPC_CALL_SERVER_SEND_REPLY;
rxrpc: Fix call timeouts Fix the rxrpc call expiration timeouts and make them settable from userspace. By analogy with other rx implementations, there should be three timeouts: (1) "Normal timeout" This is set for all calls and is triggered if we haven't received any packets from the peer in a while. It is measured from the last time we received any packet on that call. This is not reset by any connection packets (such as CHALLENGE/RESPONSE packets). If a service operation takes a long time, the server should generate PING ACKs at a duration that's substantially less than the normal timeout so is to keep both sides alive. This is set at 1/6 of normal timeout. (2) "Idle timeout" This is set only for a service call and is triggered if we stop receiving the DATA packets that comprise the request data. It is measured from the last time we received a DATA packet. (3) "Hard timeout" This can be set for a call and specified the maximum lifetime of that call. It should not be specified by default. Some operations (such as volume transfer) take a long time. Allow userspace to set/change the timeouts on a call with sendmsg, using a control message: RXRPC_SET_CALL_TIMEOUTS The data to the message is a number of 32-bit words, not all of which need be given: u32 hard_timeout; /* sec from first packet */ u32 idle_timeout; /* msec from packet Rx */ u32 normal_timeout; /* msec from data Rx */ This can be set in combination with any other sendmsg() that affects a call. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-24 18:18:41 +08:00
now = jiffies;
WRITE_ONCE(call->ack_at, now + MAX_JIFFY_OFFSET);
if (call->ackr_reason == RXRPC_ACK_DELAY)
call->ackr_reason = 0;
rxrpc: Fix call timeouts Fix the rxrpc call expiration timeouts and make them settable from userspace. By analogy with other rx implementations, there should be three timeouts: (1) "Normal timeout" This is set for all calls and is triggered if we haven't received any packets from the peer in a while. It is measured from the last time we received any packet on that call. This is not reset by any connection packets (such as CHALLENGE/RESPONSE packets). If a service operation takes a long time, the server should generate PING ACKs at a duration that's substantially less than the normal timeout so is to keep both sides alive. This is set at 1/6 of normal timeout. (2) "Idle timeout" This is set only for a service call and is triggered if we stop receiving the DATA packets that comprise the request data. It is measured from the last time we received a DATA packet. (3) "Hard timeout" This can be set for a call and specified the maximum lifetime of that call. It should not be specified by default. Some operations (such as volume transfer) take a long time. Allow userspace to set/change the timeouts on a call with sendmsg, using a control message: RXRPC_SET_CALL_TIMEOUTS The data to the message is a number of 32-bit words, not all of which need be given: u32 hard_timeout; /* sec from first packet */ u32 idle_timeout; /* msec from packet Rx */ u32 normal_timeout; /* msec from data Rx */ This can be set in combination with any other sendmsg() that affects a call. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-24 18:18:41 +08:00
trace_rxrpc_timer(call, rxrpc_timer_init_for_send_reply, now);
if (!last)
break;
/* Fall through */
case RXRPC_CALL_SERVER_SEND_REPLY:
call->state = RXRPC_CALL_SERVER_AWAIT_ACK;
rxrpc_notify_end_tx(rx, call, notify_end_tx);
break;
default:
break;
}
write_unlock_bh(&call->state_lock);
}
rxrpc: Rewrite the data and ack handling code Rewrite the data and ack handling code such that: (1) Parsing of received ACK and ABORT packets and the distribution and the filing of DATA packets happens entirely within the data_ready context called from the UDP socket. This allows us to process and discard ACK and ABORT packets much more quickly (they're no longer stashed on a queue for a background thread to process). (2) We avoid calling skb_clone(), pskb_pull() and pskb_trim(). We instead keep track of the offset and length of the content of each packet in the sk_buff metadata. This means we don't do any allocation in the receive path. (3) Jumbo DATA packet parsing is now done in data_ready context. Rather than cloning the packet once for each subpacket and pulling/trimming it, we file the packet multiple times with an annotation for each indicating which subpacket is there. From that we can directly calculate the offset and length. (4) A call's receive queue can be accessed without taking locks (memory barriers do have to be used, though). (5) Incoming calls are set up from preallocated resources and immediately made live. They can than have packets queued upon them and ACKs generated. If insufficient resources exist, DATA packet #1 is given a BUSY reply and other DATA packets are discarded). (6) sk_buffs no longer take a ref on their parent call. To make this work, the following changes are made: (1) Each call's receive buffer is now a circular buffer of sk_buff pointers (rxtx_buffer) rather than a number of sk_buff_heads spread between the call and the socket. This permits each sk_buff to be in the buffer multiple times. The receive buffer is reused for the transmit buffer. (2) A circular buffer of annotations (rxtx_annotations) is kept parallel to the data buffer. Transmission phase annotations indicate whether a buffered packet has been ACK'd or not and whether it needs retransmission. Receive phase annotations indicate whether a slot holds a whole packet or a jumbo subpacket and, if the latter, which subpacket. They also note whether the packet has been decrypted in place. (3) DATA packet window tracking is much simplified. Each phase has just two numbers representing the window (rx_hard_ack/rx_top and tx_hard_ack/tx_top). The hard_ack number is the sequence number before base of the window, representing the last packet the other side says it has consumed. hard_ack starts from 0 and the first packet is sequence number 1. The top number is the sequence number of the highest-numbered packet residing in the buffer. Packets between hard_ack+1 and top are soft-ACK'd to indicate they've been received, but not yet consumed. Four macros, before(), before_eq(), after() and after_eq() are added to compare sequence numbers within the window. This allows for the top of the window to wrap when the hard-ack sequence number gets close to the limit. Two flags, RXRPC_CALL_RX_LAST and RXRPC_CALL_TX_LAST, are added also to indicate when rx_top and tx_top point at the packets with the LAST_PACKET bit set, indicating the end of the phase. (4) Calls are queued on the socket 'receive queue' rather than packets. This means that we don't need have to invent dummy packets to queue to indicate abnormal/terminal states and we don't have to keep metadata packets (such as ABORTs) around (5) The offset and length of a (sub)packet's content are now passed to the verify_packet security op. This is currently expected to decrypt the packet in place and validate it. However, there's now nowhere to store the revised offset and length of the actual data within the decrypted blob (there may be a header and padding to skip) because an sk_buff may represent multiple packets, so a locate_data security op is added to retrieve these details from the sk_buff content when needed. (6) recvmsg() now has to handle jumbo subpackets, where each subpacket is individually secured and needs to be individually decrypted. The code to do this is broken out into rxrpc_recvmsg_data() and shared with the kernel API. It now iterates over the call's receive buffer rather than walking the socket receive queue. Additional changes: (1) The timers are condensed to a single timer that is set for the soonest of three timeouts (delayed ACK generation, DATA retransmission and call lifespan). (2) Transmission of ACK and ABORT packets is effected immediately from process-context socket ops/kernel API calls that cause them instead of them being punted off to a background work item. The data_ready handler still has to defer to the background, though. (3) A shutdown op is added to the AF_RXRPC socket so that the AFS filesystem can shut down the socket and flush its own work items before closing the socket to deal with any in-progress service calls. Future additional changes that will need to be considered: (1) Make sure that a call doesn't hog the front of the queue by receiving data from the network as fast as userspace is consuming it to the exclusion of other calls. (2) Transmit delayed ACKs from within recvmsg() when we've consumed sufficiently more packets to avoid the background work item needing to run. Signed-off-by: David Howells <dhowells@redhat.com>
2016-09-08 18:10:12 +08:00
if (seq == 1 && rxrpc_is_client_call(call))
rxrpc_expose_client_call(call);
ret = rxrpc_send_data_packet(call, skb, false);
if (ret < 0) {
switch (ret) {
case -ENETUNREACH:
case -EHOSTUNREACH:
case -ECONNREFUSED:
rxrpc: Fix missing notification Under some circumstances, rxrpc will fail a transmit a packet through the underlying UDP socket (ie. UDP sendmsg returns an error). This may result in a call getting stuck. In the instance being seen, where AFS tries to send a probe to the Volume Location server, tracepoints show the UDP Tx failure (in this case returing error 99 EADDRNOTAVAIL) and then nothing more: afs_make_vl_call: c=0000015d VL.GetCapabilities rxrpc_call: c=0000015d NWc u=1 sp=rxrpc_kernel_begin_call+0x106/0x170 [rxrpc] a=00000000dd89ee8a rxrpc_call: c=0000015d Gus u=2 sp=rxrpc_new_client_call+0x14f/0x580 [rxrpc] a=00000000e20e4b08 rxrpc_call: c=0000015d SEE u=2 sp=rxrpc_activate_one_channel+0x7b/0x1c0 [rxrpc] a=00000000e20e4b08 rxrpc_call: c=0000015d CON u=2 sp=rxrpc_kernel_begin_call+0x106/0x170 [rxrpc] a=00000000e20e4b08 rxrpc_tx_fail: c=0000015d r=1 ret=-99 CallDataNofrag The problem is that if the initial packet fails and the retransmission timer hasn't been started, the call is set to completed and an error is returned from rxrpc_send_data_packet() to rxrpc_queue_packet(). Though rxrpc_instant_resend() is called, this does nothing because the call is marked completed. So rxrpc_notify_socket() isn't called and the error is passed back up to rxrpc_send_data(), rxrpc_kernel_send_data() and thence to afs_make_call() and afs_vl_get_capabilities() where it is simply ignored because it is assumed that the result of a probe will be collected asynchronously. Fileserver probing is similarly affected via afs_fs_get_capabilities(). Fix this by always issuing a notification in __rxrpc_set_call_completion() if it shifts a call to the completed state, even if an error is also returned to the caller through the function return value. Also put in a little bit of optimisation to avoid taking the call state_lock and disabling softirqs if the call is already in the completed state and remove some now redundant rxrpc_notify_socket() calls. Fixes: f5c17aaeb2ae ("rxrpc: Calls should only have one terminal state") Reported-by: Gerry Seidman <gerry@auristor.com> Signed-off-by: David Howells <dhowells@redhat.com> Reviewed-by: Marc Dionne <marc.dionne@auristor.com>
2020-06-04 05:21:16 +08:00
rxrpc_set_call_completion(call, RXRPC_CALL_LOCAL_ERROR,
0, ret);
goto out;
}
_debug("need instant resend %d", ret);
rxrpc: Rewrite the data and ack handling code Rewrite the data and ack handling code such that: (1) Parsing of received ACK and ABORT packets and the distribution and the filing of DATA packets happens entirely within the data_ready context called from the UDP socket. This allows us to process and discard ACK and ABORT packets much more quickly (they're no longer stashed on a queue for a background thread to process). (2) We avoid calling skb_clone(), pskb_pull() and pskb_trim(). We instead keep track of the offset and length of the content of each packet in the sk_buff metadata. This means we don't do any allocation in the receive path. (3) Jumbo DATA packet parsing is now done in data_ready context. Rather than cloning the packet once for each subpacket and pulling/trimming it, we file the packet multiple times with an annotation for each indicating which subpacket is there. From that we can directly calculate the offset and length. (4) A call's receive queue can be accessed without taking locks (memory barriers do have to be used, though). (5) Incoming calls are set up from preallocated resources and immediately made live. They can than have packets queued upon them and ACKs generated. If insufficient resources exist, DATA packet #1 is given a BUSY reply and other DATA packets are discarded). (6) sk_buffs no longer take a ref on their parent call. To make this work, the following changes are made: (1) Each call's receive buffer is now a circular buffer of sk_buff pointers (rxtx_buffer) rather than a number of sk_buff_heads spread between the call and the socket. This permits each sk_buff to be in the buffer multiple times. The receive buffer is reused for the transmit buffer. (2) A circular buffer of annotations (rxtx_annotations) is kept parallel to the data buffer. Transmission phase annotations indicate whether a buffered packet has been ACK'd or not and whether it needs retransmission. Receive phase annotations indicate whether a slot holds a whole packet or a jumbo subpacket and, if the latter, which subpacket. They also note whether the packet has been decrypted in place. (3) DATA packet window tracking is much simplified. Each phase has just two numbers representing the window (rx_hard_ack/rx_top and tx_hard_ack/tx_top). The hard_ack number is the sequence number before base of the window, representing the last packet the other side says it has consumed. hard_ack starts from 0 and the first packet is sequence number 1. The top number is the sequence number of the highest-numbered packet residing in the buffer. Packets between hard_ack+1 and top are soft-ACK'd to indicate they've been received, but not yet consumed. Four macros, before(), before_eq(), after() and after_eq() are added to compare sequence numbers within the window. This allows for the top of the window to wrap when the hard-ack sequence number gets close to the limit. Two flags, RXRPC_CALL_RX_LAST and RXRPC_CALL_TX_LAST, are added also to indicate when rx_top and tx_top point at the packets with the LAST_PACKET bit set, indicating the end of the phase. (4) Calls are queued on the socket 'receive queue' rather than packets. This means that we don't need have to invent dummy packets to queue to indicate abnormal/terminal states and we don't have to keep metadata packets (such as ABORTs) around (5) The offset and length of a (sub)packet's content are now passed to the verify_packet security op. This is currently expected to decrypt the packet in place and validate it. However, there's now nowhere to store the revised offset and length of the actual data within the decrypted blob (there may be a header and padding to skip) because an sk_buff may represent multiple packets, so a locate_data security op is added to retrieve these details from the sk_buff content when needed. (6) recvmsg() now has to handle jumbo subpackets, where each subpacket is individually secured and needs to be individually decrypted. The code to do this is broken out into rxrpc_recvmsg_data() and shared with the kernel API. It now iterates over the call's receive buffer rather than walking the socket receive queue. Additional changes: (1) The timers are condensed to a single timer that is set for the soonest of three timeouts (delayed ACK generation, DATA retransmission and call lifespan). (2) Transmission of ACK and ABORT packets is effected immediately from process-context socket ops/kernel API calls that cause them instead of them being punted off to a background work item. The data_ready handler still has to defer to the background, though. (3) A shutdown op is added to the AF_RXRPC socket so that the AFS filesystem can shut down the socket and flush its own work items before closing the socket to deal with any in-progress service calls. Future additional changes that will need to be considered: (1) Make sure that a call doesn't hog the front of the queue by receiving data from the network as fast as userspace is consuming it to the exclusion of other calls. (2) Transmit delayed ACKs from within recvmsg() when we've consumed sufficiently more packets to avoid the background work item needing to run. Signed-off-by: David Howells <dhowells@redhat.com>
2016-09-08 18:10:12 +08:00
rxrpc_instant_resend(call, ix);
} else {
rxrpc: Fix the excessive initial retransmission timeout rxrpc currently uses a fixed 4s retransmission timeout until the RTT is sufficiently sampled. This can cause problems with some fileservers with calls to the cache manager in the afs filesystem being dropped from the fileserver because a packet goes missing and the retransmission timeout is greater than the call expiry timeout. Fix this by: (1) Copying the RTT/RTO calculation code from Linux's TCP implementation and altering it to fit rxrpc. (2) Altering the various users of the RTT to make use of the new SRTT value. (3) Replacing the use of rxrpc_resend_timeout to use the calculated RTO value instead (which is needed in jiffies), along with a backoff. Notes: (1) rxrpc provides RTT samples by matching the serial numbers on outgoing DATA packets that have the RXRPC_REQUEST_ACK set and PING ACK packets against the reference serial number in incoming REQUESTED ACK and PING-RESPONSE ACK packets. (2) Each packet that is transmitted on an rxrpc connection gets a new per-connection serial number, even for retransmissions, so an ACK can be cross-referenced to a specific trigger packet. This allows RTT information to be drawn from retransmitted DATA packets also. (3) rxrpc maintains the RTT/RTO state on the rxrpc_peer record rather than on an rxrpc_call because many RPC calls won't live long enough to generate more than one sample. (4) The calculated SRTT value is in units of 8ths of a microsecond rather than nanoseconds. The (S)RTT and RTO values are displayed in /proc/net/rxrpc/peers. Fixes: 17926a79320a ([AF_RXRPC]: Provide secure RxRPC sockets for use by userspace and kernel both"") Signed-off-by: David Howells <dhowells@redhat.com>
2020-05-11 21:54:34 +08:00
unsigned long now = jiffies;
unsigned long resend_at = now + call->peer->rto_j;
rxrpc: Fix call timeouts Fix the rxrpc call expiration timeouts and make them settable from userspace. By analogy with other rx implementations, there should be three timeouts: (1) "Normal timeout" This is set for all calls and is triggered if we haven't received any packets from the peer in a while. It is measured from the last time we received any packet on that call. This is not reset by any connection packets (such as CHALLENGE/RESPONSE packets). If a service operation takes a long time, the server should generate PING ACKs at a duration that's substantially less than the normal timeout so is to keep both sides alive. This is set at 1/6 of normal timeout. (2) "Idle timeout" This is set only for a service call and is triggered if we stop receiving the DATA packets that comprise the request data. It is measured from the last time we received a DATA packet. (3) "Hard timeout" This can be set for a call and specified the maximum lifetime of that call. It should not be specified by default. Some operations (such as volume transfer) take a long time. Allow userspace to set/change the timeouts on a call with sendmsg, using a control message: RXRPC_SET_CALL_TIMEOUTS The data to the message is a number of 32-bit words, not all of which need be given: u32 hard_timeout; /* sec from first packet */ u32 idle_timeout; /* msec from packet Rx */ u32 normal_timeout; /* msec from data Rx */ This can be set in combination with any other sendmsg() that affects a call. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-24 18:18:41 +08:00
WRITE_ONCE(call->resend_at, resend_at);
rxrpc_reduce_call_timer(call, resend_at, now,
rxrpc_timer_set_for_send);
}
out:
rxrpc_free_skb(skb, rxrpc_skb_freed);
_leave(" = %d", ret);
return ret;
}
/*
* send data through a socket
* - must be called in process context
rxrpc: Fix deadlock between call creation and sendmsg/recvmsg All the routines by which rxrpc is accessed from the outside are serialised by means of the socket lock (sendmsg, recvmsg, bind, rxrpc_kernel_begin_call(), ...) and this presents a problem: (1) If a number of calls on the same socket are in the process of connection to the same peer, a maximum of four concurrent live calls are permitted before further calls need to wait for a slot. (2) If a call is waiting for a slot, it is deep inside sendmsg() or rxrpc_kernel_begin_call() and the entry function is holding the socket lock. (3) sendmsg() and recvmsg() or the in-kernel equivalents are prevented from servicing the other calls as they need to take the socket lock to do so. (4) The socket is stuck until a call is aborted and makes its slot available to the waiter. Fix this by: (1) Provide each call with a mutex ('user_mutex') that arbitrates access by the users of rxrpc separately for each specific call. (2) Make rxrpc_sendmsg() and rxrpc_recvmsg() unlock the socket as soon as they've got a call and taken its mutex. Note that I'm returning EWOULDBLOCK from recvmsg() if MSG_DONTWAIT is set but someone else has the lock. Should I instead only return EWOULDBLOCK if there's nothing currently to be done on a socket, and sleep in this particular instance because there is something to be done, but we appear to be blocked by the interrupt handler doing its ping? (3) Make rxrpc_new_client_call() unlock the socket after allocating a new call, locking its user mutex and adding it to the socket's call tree. The call is returned locked so that sendmsg() can add data to it immediately. From the moment the call is in the socket tree, it is subject to access by sendmsg() and recvmsg() - even if it isn't connected yet. (4) Lock new service calls in the UDP data_ready handler (in rxrpc_new_incoming_call()) because they may already be in the socket's tree and the data_ready handler makes them live immediately if a user ID has already been preassigned. Note that the new call is locked before any notifications are sent that it is live, so doing mutex_trylock() *ought* to always succeed. Userspace is prevented from doing sendmsg() on calls that are in a too-early state in rxrpc_do_sendmsg(). (5) Make rxrpc_new_incoming_call() return the call with the user mutex held so that a ping can be scheduled immediately under it. Note that it might be worth moving the ping call into rxrpc_new_incoming_call() and then we can drop the mutex there. (6) Make rxrpc_accept_call() take the lock on the call it is accepting and release the socket after adding the call to the socket's tree. This is slightly tricky as we've dequeued the call by that point and have to requeue it. Note that requeuing emits a trace event. (7) Make rxrpc_kernel_send_data() and rxrpc_kernel_recv_data() take the new mutex immediately and don't bother with the socket mutex at all. This patch has the nice bonus that calls on the same socket are now to some extent parallelisable. Note that we might want to move rxrpc_service_prealloc() calls out from the socket lock and give it its own lock, so that we don't hang progress in other calls because we're waiting for the allocator. We probably also want to avoid calling rxrpc_notify_socket() from within the socket lock (rxrpc_accept_call()). Signed-off-by: David Howells <dhowells@redhat.com> Tested-by: Marc Dionne <marc.c.dionne@auristor.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-02-27 23:43:06 +08:00
* - The caller holds the call user access mutex, but not the socket lock.
*/
static int rxrpc_send_data(struct rxrpc_sock *rx,
struct rxrpc_call *call,
struct msghdr *msg, size_t len,
rxrpc_notify_end_tx_t notify_end_tx)
{
struct rxrpc_skb_priv *sp;
struct sk_buff *skb;
struct sock *sk = &rx->sk;
long timeo;
bool more;
int ret, copied;
timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
/* this should be in poll */
sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
return -EPIPE;
more = msg->msg_flags & MSG_MORE;
if (call->tx_total_len != -1) {
if (len > call->tx_total_len)
return -EMSGSIZE;
if (!more && len != call->tx_total_len)
return -EMSGSIZE;
}
skb = call->tx_pending;
call->tx_pending = NULL;
rxrpc_see_skb(skb, rxrpc_skb_seen);
copied = 0;
do {
/* Check to see if there's a ping ACK to reply to. */
if (call->ackr_reason == RXRPC_ACK_PING_RESPONSE)
rxrpc_send_ack_packet(call, false, NULL);
if (!skb) {
size_t size, chunk, max, space;
_debug("alloc");
if (!rxrpc_check_tx_space(call, NULL)) {
ret = -EAGAIN;
if (msg->msg_flags & MSG_DONTWAIT)
goto maybe_error;
ret = rxrpc_wait_for_tx_window(rx, call,
&timeo,
msg->msg_flags & MSG_WAITALL);
if (ret < 0)
goto maybe_error;
}
max = RXRPC_JUMBO_DATALEN;
max -= call->conn->security_size;
max &= ~(call->conn->size_align - 1UL);
chunk = max;
if (chunk > msg_data_left(msg) && !more)
chunk = msg_data_left(msg);
space = chunk + call->conn->size_align;
space &= ~(call->conn->size_align - 1UL);
size = space + call->conn->security_size;
_debug("SIZE: %zu/%zu/%zu", chunk, space, size);
/* create a buffer that we can retain until it's ACK'd */
skb = sock_alloc_send_skb(
sk, size, msg->msg_flags & MSG_DONTWAIT, &ret);
if (!skb)
goto maybe_error;
sp = rxrpc_skb(skb);
sp->rx_flags |= RXRPC_SKB_TX_BUFFER;
rxrpc_new_skb(skb, rxrpc_skb_new);
_debug("ALLOC SEND %p", skb);
ASSERTCMP(skb->mark, ==, 0);
_debug("HS: %u", call->conn->security_size);
skb_reserve(skb, call->conn->security_size);
skb->len += call->conn->security_size;
sp->remain = chunk;
if (sp->remain > skb_tailroom(skb))
sp->remain = skb_tailroom(skb);
_net("skb: hr %d, tr %d, hl %d, rm %d",
skb_headroom(skb),
skb_tailroom(skb),
skb_headlen(skb),
sp->remain);
skb->ip_summed = CHECKSUM_UNNECESSARY;
}
_debug("append");
sp = rxrpc_skb(skb);
/* append next segment of data to the current buffer */
if (msg_data_left(msg) > 0) {
int copy = skb_tailroom(skb);
ASSERTCMP(copy, >, 0);
if (copy > msg_data_left(msg))
copy = msg_data_left(msg);
if (copy > sp->remain)
copy = sp->remain;
_debug("add");
ret = skb_add_data(skb, &msg->msg_iter, copy);
_debug("added");
if (ret < 0)
goto efault;
sp->remain -= copy;
skb->mark += copy;
copied += copy;
if (call->tx_total_len != -1)
call->tx_total_len -= copy;
}
Revert "rxrpc: Allow failed client calls to be retried" The changes introduced to allow rxrpc calls to be retried creates an issue when it comes to refcounting afs_call structs. The problem is that when rxrpc_send_data() queues the last packet for an asynchronous call, the following sequence can occur: (1) The notify_end_tx callback is invoked which causes the state in the afs_call to be changed from AFS_CALL_CL_REQUESTING or AFS_CALL_SV_REPLYING. (2) afs_deliver_to_call() can then process event notifications from rxrpc on the async_work queue. (3) Delivery of events, such as an abort from the server, can cause the afs_call state to be changed to AFS_CALL_COMPLETE on async_work. (4) For an asynchronous call, afs_process_async_call() notes that the call is complete and tried to clean up all the refs on async_work. (5) rxrpc_send_data() might return the amount of data transferred (success) or an error - which could in turn reflect a local error or a received error. Synchronising the clean up after rxrpc_kernel_send_data() returns an error with the asynchronous cleanup is then tricky to get right. Mostly revert commit c038a58ccfd6704d4d7d60ed3d6a0fca13cf13a4. The two API functions the original commit added aren't currently used. This makes rxrpc_kernel_send_data() always return successfully if it queued the data it was given. Note that this doesn't affect synchronous calls since their Rx notification function merely pokes a wait queue and does not refcounting. The asynchronous call notification function *has* to do refcounting and pass a ref over the work item to avoid the need to sync the workqueue in call cleanup. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-01-11 00:59:13 +08:00
/* check for the far side aborting the call or a network error
* occurring */
if (call->state == RXRPC_CALL_COMPLETE)
goto call_terminated;
/* add the packet to the send queue if it's now full */
if (sp->remain <= 0 ||
(msg_data_left(msg) == 0 && !more)) {
struct rxrpc_connection *conn = call->conn;
uint32_t seq;
size_t pad;
/* pad out if we're using security */
if (conn->security_ix) {
pad = conn->security_size + skb->mark;
pad = conn->size_align - pad;
pad &= conn->size_align - 1;
_debug("pad %zu", pad);
if (pad)
skb_put_zero(skb, pad);
}
rxrpc: Rewrite the data and ack handling code Rewrite the data and ack handling code such that: (1) Parsing of received ACK and ABORT packets and the distribution and the filing of DATA packets happens entirely within the data_ready context called from the UDP socket. This allows us to process and discard ACK and ABORT packets much more quickly (they're no longer stashed on a queue for a background thread to process). (2) We avoid calling skb_clone(), pskb_pull() and pskb_trim(). We instead keep track of the offset and length of the content of each packet in the sk_buff metadata. This means we don't do any allocation in the receive path. (3) Jumbo DATA packet parsing is now done in data_ready context. Rather than cloning the packet once for each subpacket and pulling/trimming it, we file the packet multiple times with an annotation for each indicating which subpacket is there. From that we can directly calculate the offset and length. (4) A call's receive queue can be accessed without taking locks (memory barriers do have to be used, though). (5) Incoming calls are set up from preallocated resources and immediately made live. They can than have packets queued upon them and ACKs generated. If insufficient resources exist, DATA packet #1 is given a BUSY reply and other DATA packets are discarded). (6) sk_buffs no longer take a ref on their parent call. To make this work, the following changes are made: (1) Each call's receive buffer is now a circular buffer of sk_buff pointers (rxtx_buffer) rather than a number of sk_buff_heads spread between the call and the socket. This permits each sk_buff to be in the buffer multiple times. The receive buffer is reused for the transmit buffer. (2) A circular buffer of annotations (rxtx_annotations) is kept parallel to the data buffer. Transmission phase annotations indicate whether a buffered packet has been ACK'd or not and whether it needs retransmission. Receive phase annotations indicate whether a slot holds a whole packet or a jumbo subpacket and, if the latter, which subpacket. They also note whether the packet has been decrypted in place. (3) DATA packet window tracking is much simplified. Each phase has just two numbers representing the window (rx_hard_ack/rx_top and tx_hard_ack/tx_top). The hard_ack number is the sequence number before base of the window, representing the last packet the other side says it has consumed. hard_ack starts from 0 and the first packet is sequence number 1. The top number is the sequence number of the highest-numbered packet residing in the buffer. Packets between hard_ack+1 and top are soft-ACK'd to indicate they've been received, but not yet consumed. Four macros, before(), before_eq(), after() and after_eq() are added to compare sequence numbers within the window. This allows for the top of the window to wrap when the hard-ack sequence number gets close to the limit. Two flags, RXRPC_CALL_RX_LAST and RXRPC_CALL_TX_LAST, are added also to indicate when rx_top and tx_top point at the packets with the LAST_PACKET bit set, indicating the end of the phase. (4) Calls are queued on the socket 'receive queue' rather than packets. This means that we don't need have to invent dummy packets to queue to indicate abnormal/terminal states and we don't have to keep metadata packets (such as ABORTs) around (5) The offset and length of a (sub)packet's content are now passed to the verify_packet security op. This is currently expected to decrypt the packet in place and validate it. However, there's now nowhere to store the revised offset and length of the actual data within the decrypted blob (there may be a header and padding to skip) because an sk_buff may represent multiple packets, so a locate_data security op is added to retrieve these details from the sk_buff content when needed. (6) recvmsg() now has to handle jumbo subpackets, where each subpacket is individually secured and needs to be individually decrypted. The code to do this is broken out into rxrpc_recvmsg_data() and shared with the kernel API. It now iterates over the call's receive buffer rather than walking the socket receive queue. Additional changes: (1) The timers are condensed to a single timer that is set for the soonest of three timeouts (delayed ACK generation, DATA retransmission and call lifespan). (2) Transmission of ACK and ABORT packets is effected immediately from process-context socket ops/kernel API calls that cause them instead of them being punted off to a background work item. The data_ready handler still has to defer to the background, though. (3) A shutdown op is added to the AF_RXRPC socket so that the AFS filesystem can shut down the socket and flush its own work items before closing the socket to deal with any in-progress service calls. Future additional changes that will need to be considered: (1) Make sure that a call doesn't hog the front of the queue by receiving data from the network as fast as userspace is consuming it to the exclusion of other calls. (2) Transmit delayed ACKs from within recvmsg() when we've consumed sufficiently more packets to avoid the background work item needing to run. Signed-off-by: David Howells <dhowells@redhat.com>
2016-09-08 18:10:12 +08:00
seq = call->tx_top + 1;
sp->hdr.seq = seq;
sp->hdr._rsvd = 0;
sp->hdr.flags = conn->out_clientflag;
if (msg_data_left(msg) == 0 && !more)
sp->hdr.flags |= RXRPC_LAST_PACKET;
rxrpc: Rewrite the data and ack handling code Rewrite the data and ack handling code such that: (1) Parsing of received ACK and ABORT packets and the distribution and the filing of DATA packets happens entirely within the data_ready context called from the UDP socket. This allows us to process and discard ACK and ABORT packets much more quickly (they're no longer stashed on a queue for a background thread to process). (2) We avoid calling skb_clone(), pskb_pull() and pskb_trim(). We instead keep track of the offset and length of the content of each packet in the sk_buff metadata. This means we don't do any allocation in the receive path. (3) Jumbo DATA packet parsing is now done in data_ready context. Rather than cloning the packet once for each subpacket and pulling/trimming it, we file the packet multiple times with an annotation for each indicating which subpacket is there. From that we can directly calculate the offset and length. (4) A call's receive queue can be accessed without taking locks (memory barriers do have to be used, though). (5) Incoming calls are set up from preallocated resources and immediately made live. They can than have packets queued upon them and ACKs generated. If insufficient resources exist, DATA packet #1 is given a BUSY reply and other DATA packets are discarded). (6) sk_buffs no longer take a ref on their parent call. To make this work, the following changes are made: (1) Each call's receive buffer is now a circular buffer of sk_buff pointers (rxtx_buffer) rather than a number of sk_buff_heads spread between the call and the socket. This permits each sk_buff to be in the buffer multiple times. The receive buffer is reused for the transmit buffer. (2) A circular buffer of annotations (rxtx_annotations) is kept parallel to the data buffer. Transmission phase annotations indicate whether a buffered packet has been ACK'd or not and whether it needs retransmission. Receive phase annotations indicate whether a slot holds a whole packet or a jumbo subpacket and, if the latter, which subpacket. They also note whether the packet has been decrypted in place. (3) DATA packet window tracking is much simplified. Each phase has just two numbers representing the window (rx_hard_ack/rx_top and tx_hard_ack/tx_top). The hard_ack number is the sequence number before base of the window, representing the last packet the other side says it has consumed. hard_ack starts from 0 and the first packet is sequence number 1. The top number is the sequence number of the highest-numbered packet residing in the buffer. Packets between hard_ack+1 and top are soft-ACK'd to indicate they've been received, but not yet consumed. Four macros, before(), before_eq(), after() and after_eq() are added to compare sequence numbers within the window. This allows for the top of the window to wrap when the hard-ack sequence number gets close to the limit. Two flags, RXRPC_CALL_RX_LAST and RXRPC_CALL_TX_LAST, are added also to indicate when rx_top and tx_top point at the packets with the LAST_PACKET bit set, indicating the end of the phase. (4) Calls are queued on the socket 'receive queue' rather than packets. This means that we don't need have to invent dummy packets to queue to indicate abnormal/terminal states and we don't have to keep metadata packets (such as ABORTs) around (5) The offset and length of a (sub)packet's content are now passed to the verify_packet security op. This is currently expected to decrypt the packet in place and validate it. However, there's now nowhere to store the revised offset and length of the actual data within the decrypted blob (there may be a header and padding to skip) because an sk_buff may represent multiple packets, so a locate_data security op is added to retrieve these details from the sk_buff content when needed. (6) recvmsg() now has to handle jumbo subpackets, where each subpacket is individually secured and needs to be individually decrypted. The code to do this is broken out into rxrpc_recvmsg_data() and shared with the kernel API. It now iterates over the call's receive buffer rather than walking the socket receive queue. Additional changes: (1) The timers are condensed to a single timer that is set for the soonest of three timeouts (delayed ACK generation, DATA retransmission and call lifespan). (2) Transmission of ACK and ABORT packets is effected immediately from process-context socket ops/kernel API calls that cause them instead of them being punted off to a background work item. The data_ready handler still has to defer to the background, though. (3) A shutdown op is added to the AF_RXRPC socket so that the AFS filesystem can shut down the socket and flush its own work items before closing the socket to deal with any in-progress service calls. Future additional changes that will need to be considered: (1) Make sure that a call doesn't hog the front of the queue by receiving data from the network as fast as userspace is consuming it to the exclusion of other calls. (2) Transmit delayed ACKs from within recvmsg() when we've consumed sufficiently more packets to avoid the background work item needing to run. Signed-off-by: David Howells <dhowells@redhat.com>
2016-09-08 18:10:12 +08:00
else if (call->tx_top - call->tx_hard_ack <
call->tx_winsize)
sp->hdr.flags |= RXRPC_MORE_PACKETS;
ret = call->security->secure_packet(
call, skb, skb->mark, skb->head);
if (ret < 0)
goto out;
ret = rxrpc_queue_packet(rx, call, skb,
!msg_data_left(msg) && !more,
notify_end_tx);
/* Should check for failure here */
skb = NULL;
}
} while (msg_data_left(msg) > 0);
success:
ret = copied;
out:
call->tx_pending = skb;
_leave(" = %d", ret);
return ret;
Revert "rxrpc: Allow failed client calls to be retried" The changes introduced to allow rxrpc calls to be retried creates an issue when it comes to refcounting afs_call structs. The problem is that when rxrpc_send_data() queues the last packet for an asynchronous call, the following sequence can occur: (1) The notify_end_tx callback is invoked which causes the state in the afs_call to be changed from AFS_CALL_CL_REQUESTING or AFS_CALL_SV_REPLYING. (2) afs_deliver_to_call() can then process event notifications from rxrpc on the async_work queue. (3) Delivery of events, such as an abort from the server, can cause the afs_call state to be changed to AFS_CALL_COMPLETE on async_work. (4) For an asynchronous call, afs_process_async_call() notes that the call is complete and tried to clean up all the refs on async_work. (5) rxrpc_send_data() might return the amount of data transferred (success) or an error - which could in turn reflect a local error or a received error. Synchronising the clean up after rxrpc_kernel_send_data() returns an error with the asynchronous cleanup is then tricky to get right. Mostly revert commit c038a58ccfd6704d4d7d60ed3d6a0fca13cf13a4. The two API functions the original commit added aren't currently used. This makes rxrpc_kernel_send_data() always return successfully if it queued the data it was given. Note that this doesn't affect synchronous calls since their Rx notification function merely pokes a wait queue and does not refcounting. The asynchronous call notification function *has* to do refcounting and pass a ref over the work item to avoid the need to sync the workqueue in call cleanup. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-01-11 00:59:13 +08:00
call_terminated:
rxrpc_free_skb(skb, rxrpc_skb_freed);
Revert "rxrpc: Allow failed client calls to be retried" The changes introduced to allow rxrpc calls to be retried creates an issue when it comes to refcounting afs_call structs. The problem is that when rxrpc_send_data() queues the last packet for an asynchronous call, the following sequence can occur: (1) The notify_end_tx callback is invoked which causes the state in the afs_call to be changed from AFS_CALL_CL_REQUESTING or AFS_CALL_SV_REPLYING. (2) afs_deliver_to_call() can then process event notifications from rxrpc on the async_work queue. (3) Delivery of events, such as an abort from the server, can cause the afs_call state to be changed to AFS_CALL_COMPLETE on async_work. (4) For an asynchronous call, afs_process_async_call() notes that the call is complete and tried to clean up all the refs on async_work. (5) rxrpc_send_data() might return the amount of data transferred (success) or an error - which could in turn reflect a local error or a received error. Synchronising the clean up after rxrpc_kernel_send_data() returns an error with the asynchronous cleanup is then tricky to get right. Mostly revert commit c038a58ccfd6704d4d7d60ed3d6a0fca13cf13a4. The two API functions the original commit added aren't currently used. This makes rxrpc_kernel_send_data() always return successfully if it queued the data it was given. Note that this doesn't affect synchronous calls since their Rx notification function merely pokes a wait queue and does not refcounting. The asynchronous call notification function *has* to do refcounting and pass a ref over the work item to avoid the need to sync the workqueue in call cleanup. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-01-11 00:59:13 +08:00
_leave(" = %d", call->error);
return call->error;
maybe_error:
if (copied)
goto success;
goto out;
efault:
ret = -EFAULT;
goto out;
}
/*
* extract control messages from the sendmsg() control buffer
*/
static int rxrpc_sendmsg_cmsg(struct msghdr *msg, struct rxrpc_send_params *p)
{
struct cmsghdr *cmsg;
bool got_user_ID = false;
int len;
if (msg->msg_controllen == 0)
return -EINVAL;
for_each_cmsghdr(cmsg, msg) {
if (!CMSG_OK(msg, cmsg))
return -EINVAL;
len = cmsg->cmsg_len - sizeof(struct cmsghdr);
_debug("CMSG %d, %d, %d",
cmsg->cmsg_level, cmsg->cmsg_type, len);
if (cmsg->cmsg_level != SOL_RXRPC)
continue;
switch (cmsg->cmsg_type) {
case RXRPC_USER_CALL_ID:
if (msg->msg_flags & MSG_CMSG_COMPAT) {
if (len != sizeof(u32))
return -EINVAL;
p->call.user_call_ID = *(u32 *)CMSG_DATA(cmsg);
} else {
if (len != sizeof(unsigned long))
return -EINVAL;
p->call.user_call_ID = *(unsigned long *)
CMSG_DATA(cmsg);
}
got_user_ID = true;
break;
case RXRPC_ABORT:
if (p->command != RXRPC_CMD_SEND_DATA)
return -EINVAL;
p->command = RXRPC_CMD_SEND_ABORT;
if (len != sizeof(p->abort_code))
return -EINVAL;
p->abort_code = *(unsigned int *)CMSG_DATA(cmsg);
if (p->abort_code == 0)
return -EINVAL;
break;
case RXRPC_ACCEPT:
if (p->command != RXRPC_CMD_SEND_DATA)
return -EINVAL;
p->command = RXRPC_CMD_ACCEPT;
if (len != 0)
return -EINVAL;
break;
case RXRPC_EXCLUSIVE_CALL:
p->exclusive = true;
if (len != 0)
return -EINVAL;
break;
case RXRPC_UPGRADE_SERVICE:
p->upgrade = true;
if (len != 0)
return -EINVAL;
break;
case RXRPC_TX_LENGTH:
if (p->call.tx_total_len != -1 || len != sizeof(__s64))
return -EINVAL;
p->call.tx_total_len = *(__s64 *)CMSG_DATA(cmsg);
if (p->call.tx_total_len < 0)
return -EINVAL;
break;
rxrpc: Fix call timeouts Fix the rxrpc call expiration timeouts and make them settable from userspace. By analogy with other rx implementations, there should be three timeouts: (1) "Normal timeout" This is set for all calls and is triggered if we haven't received any packets from the peer in a while. It is measured from the last time we received any packet on that call. This is not reset by any connection packets (such as CHALLENGE/RESPONSE packets). If a service operation takes a long time, the server should generate PING ACKs at a duration that's substantially less than the normal timeout so is to keep both sides alive. This is set at 1/6 of normal timeout. (2) "Idle timeout" This is set only for a service call and is triggered if we stop receiving the DATA packets that comprise the request data. It is measured from the last time we received a DATA packet. (3) "Hard timeout" This can be set for a call and specified the maximum lifetime of that call. It should not be specified by default. Some operations (such as volume transfer) take a long time. Allow userspace to set/change the timeouts on a call with sendmsg, using a control message: RXRPC_SET_CALL_TIMEOUTS The data to the message is a number of 32-bit words, not all of which need be given: u32 hard_timeout; /* sec from first packet */ u32 idle_timeout; /* msec from packet Rx */ u32 normal_timeout; /* msec from data Rx */ This can be set in combination with any other sendmsg() that affects a call. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-24 18:18:41 +08:00
case RXRPC_SET_CALL_TIMEOUT:
if (len & 3 || len < 4 || len > 12)
return -EINVAL;
memcpy(&p->call.timeouts, CMSG_DATA(cmsg), len);
p->call.nr_timeouts = len / 4;
if (p->call.timeouts.hard > INT_MAX / HZ)
return -ERANGE;
if (p->call.nr_timeouts >= 2 && p->call.timeouts.idle > 60 * 60 * 1000)
return -ERANGE;
if (p->call.nr_timeouts >= 3 && p->call.timeouts.normal > 60 * 60 * 1000)
return -ERANGE;
break;
default:
return -EINVAL;
}
}
if (!got_user_ID)
return -EINVAL;
if (p->call.tx_total_len != -1 && p->command != RXRPC_CMD_SEND_DATA)
return -EINVAL;
_leave(" = 0");
return 0;
}
/*
* Create a new client call for sendmsg().
rxrpc: Fix deadlock between call creation and sendmsg/recvmsg All the routines by which rxrpc is accessed from the outside are serialised by means of the socket lock (sendmsg, recvmsg, bind, rxrpc_kernel_begin_call(), ...) and this presents a problem: (1) If a number of calls on the same socket are in the process of connection to the same peer, a maximum of four concurrent live calls are permitted before further calls need to wait for a slot. (2) If a call is waiting for a slot, it is deep inside sendmsg() or rxrpc_kernel_begin_call() and the entry function is holding the socket lock. (3) sendmsg() and recvmsg() or the in-kernel equivalents are prevented from servicing the other calls as they need to take the socket lock to do so. (4) The socket is stuck until a call is aborted and makes its slot available to the waiter. Fix this by: (1) Provide each call with a mutex ('user_mutex') that arbitrates access by the users of rxrpc separately for each specific call. (2) Make rxrpc_sendmsg() and rxrpc_recvmsg() unlock the socket as soon as they've got a call and taken its mutex. Note that I'm returning EWOULDBLOCK from recvmsg() if MSG_DONTWAIT is set but someone else has the lock. Should I instead only return EWOULDBLOCK if there's nothing currently to be done on a socket, and sleep in this particular instance because there is something to be done, but we appear to be blocked by the interrupt handler doing its ping? (3) Make rxrpc_new_client_call() unlock the socket after allocating a new call, locking its user mutex and adding it to the socket's call tree. The call is returned locked so that sendmsg() can add data to it immediately. From the moment the call is in the socket tree, it is subject to access by sendmsg() and recvmsg() - even if it isn't connected yet. (4) Lock new service calls in the UDP data_ready handler (in rxrpc_new_incoming_call()) because they may already be in the socket's tree and the data_ready handler makes them live immediately if a user ID has already been preassigned. Note that the new call is locked before any notifications are sent that it is live, so doing mutex_trylock() *ought* to always succeed. Userspace is prevented from doing sendmsg() on calls that are in a too-early state in rxrpc_do_sendmsg(). (5) Make rxrpc_new_incoming_call() return the call with the user mutex held so that a ping can be scheduled immediately under it. Note that it might be worth moving the ping call into rxrpc_new_incoming_call() and then we can drop the mutex there. (6) Make rxrpc_accept_call() take the lock on the call it is accepting and release the socket after adding the call to the socket's tree. This is slightly tricky as we've dequeued the call by that point and have to requeue it. Note that requeuing emits a trace event. (7) Make rxrpc_kernel_send_data() and rxrpc_kernel_recv_data() take the new mutex immediately and don't bother with the socket mutex at all. This patch has the nice bonus that calls on the same socket are now to some extent parallelisable. Note that we might want to move rxrpc_service_prealloc() calls out from the socket lock and give it its own lock, so that we don't hang progress in other calls because we're waiting for the allocator. We probably also want to avoid calling rxrpc_notify_socket() from within the socket lock (rxrpc_accept_call()). Signed-off-by: David Howells <dhowells@redhat.com> Tested-by: Marc Dionne <marc.c.dionne@auristor.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-02-27 23:43:06 +08:00
* - Called with the socket lock held, which it must release.
* - If it returns a call, the call's lock will need releasing by the caller.
*/
static struct rxrpc_call *
rxrpc_new_client_call_for_sendmsg(struct rxrpc_sock *rx, struct msghdr *msg,
struct rxrpc_send_params *p)
rxrpc: Fix deadlock between call creation and sendmsg/recvmsg All the routines by which rxrpc is accessed from the outside are serialised by means of the socket lock (sendmsg, recvmsg, bind, rxrpc_kernel_begin_call(), ...) and this presents a problem: (1) If a number of calls on the same socket are in the process of connection to the same peer, a maximum of four concurrent live calls are permitted before further calls need to wait for a slot. (2) If a call is waiting for a slot, it is deep inside sendmsg() or rxrpc_kernel_begin_call() and the entry function is holding the socket lock. (3) sendmsg() and recvmsg() or the in-kernel equivalents are prevented from servicing the other calls as they need to take the socket lock to do so. (4) The socket is stuck until a call is aborted and makes its slot available to the waiter. Fix this by: (1) Provide each call with a mutex ('user_mutex') that arbitrates access by the users of rxrpc separately for each specific call. (2) Make rxrpc_sendmsg() and rxrpc_recvmsg() unlock the socket as soon as they've got a call and taken its mutex. Note that I'm returning EWOULDBLOCK from recvmsg() if MSG_DONTWAIT is set but someone else has the lock. Should I instead only return EWOULDBLOCK if there's nothing currently to be done on a socket, and sleep in this particular instance because there is something to be done, but we appear to be blocked by the interrupt handler doing its ping? (3) Make rxrpc_new_client_call() unlock the socket after allocating a new call, locking its user mutex and adding it to the socket's call tree. The call is returned locked so that sendmsg() can add data to it immediately. From the moment the call is in the socket tree, it is subject to access by sendmsg() and recvmsg() - even if it isn't connected yet. (4) Lock new service calls in the UDP data_ready handler (in rxrpc_new_incoming_call()) because they may already be in the socket's tree and the data_ready handler makes them live immediately if a user ID has already been preassigned. Note that the new call is locked before any notifications are sent that it is live, so doing mutex_trylock() *ought* to always succeed. Userspace is prevented from doing sendmsg() on calls that are in a too-early state in rxrpc_do_sendmsg(). (5) Make rxrpc_new_incoming_call() return the call with the user mutex held so that a ping can be scheduled immediately under it. Note that it might be worth moving the ping call into rxrpc_new_incoming_call() and then we can drop the mutex there. (6) Make rxrpc_accept_call() take the lock on the call it is accepting and release the socket after adding the call to the socket's tree. This is slightly tricky as we've dequeued the call by that point and have to requeue it. Note that requeuing emits a trace event. (7) Make rxrpc_kernel_send_data() and rxrpc_kernel_recv_data() take the new mutex immediately and don't bother with the socket mutex at all. This patch has the nice bonus that calls on the same socket are now to some extent parallelisable. Note that we might want to move rxrpc_service_prealloc() calls out from the socket lock and give it its own lock, so that we don't hang progress in other calls because we're waiting for the allocator. We probably also want to avoid calling rxrpc_notify_socket() from within the socket lock (rxrpc_accept_call()). Signed-off-by: David Howells <dhowells@redhat.com> Tested-by: Marc Dionne <marc.c.dionne@auristor.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-02-27 23:43:06 +08:00
__releases(&rx->sk.sk_lock.slock)
__acquires(&call->user_mutex)
{
struct rxrpc_conn_parameters cp;
struct rxrpc_call *call;
struct key *key;
DECLARE_SOCKADDR(struct sockaddr_rxrpc *, srx, msg->msg_name);
_enter("");
rxrpc: Fix deadlock between call creation and sendmsg/recvmsg All the routines by which rxrpc is accessed from the outside are serialised by means of the socket lock (sendmsg, recvmsg, bind, rxrpc_kernel_begin_call(), ...) and this presents a problem: (1) If a number of calls on the same socket are in the process of connection to the same peer, a maximum of four concurrent live calls are permitted before further calls need to wait for a slot. (2) If a call is waiting for a slot, it is deep inside sendmsg() or rxrpc_kernel_begin_call() and the entry function is holding the socket lock. (3) sendmsg() and recvmsg() or the in-kernel equivalents are prevented from servicing the other calls as they need to take the socket lock to do so. (4) The socket is stuck until a call is aborted and makes its slot available to the waiter. Fix this by: (1) Provide each call with a mutex ('user_mutex') that arbitrates access by the users of rxrpc separately for each specific call. (2) Make rxrpc_sendmsg() and rxrpc_recvmsg() unlock the socket as soon as they've got a call and taken its mutex. Note that I'm returning EWOULDBLOCK from recvmsg() if MSG_DONTWAIT is set but someone else has the lock. Should I instead only return EWOULDBLOCK if there's nothing currently to be done on a socket, and sleep in this particular instance because there is something to be done, but we appear to be blocked by the interrupt handler doing its ping? (3) Make rxrpc_new_client_call() unlock the socket after allocating a new call, locking its user mutex and adding it to the socket's call tree. The call is returned locked so that sendmsg() can add data to it immediately. From the moment the call is in the socket tree, it is subject to access by sendmsg() and recvmsg() - even if it isn't connected yet. (4) Lock new service calls in the UDP data_ready handler (in rxrpc_new_incoming_call()) because they may already be in the socket's tree and the data_ready handler makes them live immediately if a user ID has already been preassigned. Note that the new call is locked before any notifications are sent that it is live, so doing mutex_trylock() *ought* to always succeed. Userspace is prevented from doing sendmsg() on calls that are in a too-early state in rxrpc_do_sendmsg(). (5) Make rxrpc_new_incoming_call() return the call with the user mutex held so that a ping can be scheduled immediately under it. Note that it might be worth moving the ping call into rxrpc_new_incoming_call() and then we can drop the mutex there. (6) Make rxrpc_accept_call() take the lock on the call it is accepting and release the socket after adding the call to the socket's tree. This is slightly tricky as we've dequeued the call by that point and have to requeue it. Note that requeuing emits a trace event. (7) Make rxrpc_kernel_send_data() and rxrpc_kernel_recv_data() take the new mutex immediately and don't bother with the socket mutex at all. This patch has the nice bonus that calls on the same socket are now to some extent parallelisable. Note that we might want to move rxrpc_service_prealloc() calls out from the socket lock and give it its own lock, so that we don't hang progress in other calls because we're waiting for the allocator. We probably also want to avoid calling rxrpc_notify_socket() from within the socket lock (rxrpc_accept_call()). Signed-off-by: David Howells <dhowells@redhat.com> Tested-by: Marc Dionne <marc.c.dionne@auristor.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-02-27 23:43:06 +08:00
if (!msg->msg_name) {
release_sock(&rx->sk);
return ERR_PTR(-EDESTADDRREQ);
rxrpc: Fix deadlock between call creation and sendmsg/recvmsg All the routines by which rxrpc is accessed from the outside are serialised by means of the socket lock (sendmsg, recvmsg, bind, rxrpc_kernel_begin_call(), ...) and this presents a problem: (1) If a number of calls on the same socket are in the process of connection to the same peer, a maximum of four concurrent live calls are permitted before further calls need to wait for a slot. (2) If a call is waiting for a slot, it is deep inside sendmsg() or rxrpc_kernel_begin_call() and the entry function is holding the socket lock. (3) sendmsg() and recvmsg() or the in-kernel equivalents are prevented from servicing the other calls as they need to take the socket lock to do so. (4) The socket is stuck until a call is aborted and makes its slot available to the waiter. Fix this by: (1) Provide each call with a mutex ('user_mutex') that arbitrates access by the users of rxrpc separately for each specific call. (2) Make rxrpc_sendmsg() and rxrpc_recvmsg() unlock the socket as soon as they've got a call and taken its mutex. Note that I'm returning EWOULDBLOCK from recvmsg() if MSG_DONTWAIT is set but someone else has the lock. Should I instead only return EWOULDBLOCK if there's nothing currently to be done on a socket, and sleep in this particular instance because there is something to be done, but we appear to be blocked by the interrupt handler doing its ping? (3) Make rxrpc_new_client_call() unlock the socket after allocating a new call, locking its user mutex and adding it to the socket's call tree. The call is returned locked so that sendmsg() can add data to it immediately. From the moment the call is in the socket tree, it is subject to access by sendmsg() and recvmsg() - even if it isn't connected yet. (4) Lock new service calls in the UDP data_ready handler (in rxrpc_new_incoming_call()) because they may already be in the socket's tree and the data_ready handler makes them live immediately if a user ID has already been preassigned. Note that the new call is locked before any notifications are sent that it is live, so doing mutex_trylock() *ought* to always succeed. Userspace is prevented from doing sendmsg() on calls that are in a too-early state in rxrpc_do_sendmsg(). (5) Make rxrpc_new_incoming_call() return the call with the user mutex held so that a ping can be scheduled immediately under it. Note that it might be worth moving the ping call into rxrpc_new_incoming_call() and then we can drop the mutex there. (6) Make rxrpc_accept_call() take the lock on the call it is accepting and release the socket after adding the call to the socket's tree. This is slightly tricky as we've dequeued the call by that point and have to requeue it. Note that requeuing emits a trace event. (7) Make rxrpc_kernel_send_data() and rxrpc_kernel_recv_data() take the new mutex immediately and don't bother with the socket mutex at all. This patch has the nice bonus that calls on the same socket are now to some extent parallelisable. Note that we might want to move rxrpc_service_prealloc() calls out from the socket lock and give it its own lock, so that we don't hang progress in other calls because we're waiting for the allocator. We probably also want to avoid calling rxrpc_notify_socket() from within the socket lock (rxrpc_accept_call()). Signed-off-by: David Howells <dhowells@redhat.com> Tested-by: Marc Dionne <marc.c.dionne@auristor.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-02-27 23:43:06 +08:00
}
key = rx->key;
if (key && !rx->key->payload.data[0])
key = NULL;
memset(&cp, 0, sizeof(cp));
cp.local = rx->local;
cp.key = rx->key;
cp.security_level = rx->min_sec_level;
cp.exclusive = rx->exclusive | p->exclusive;
cp.upgrade = p->upgrade;
cp.service_id = srx->srx_service;
call = rxrpc_new_client_call(rx, &cp, srx, &p->call, GFP_KERNEL,
atomic_inc_return(&rxrpc_debug_id));
rxrpc: Fix deadlock between call creation and sendmsg/recvmsg All the routines by which rxrpc is accessed from the outside are serialised by means of the socket lock (sendmsg, recvmsg, bind, rxrpc_kernel_begin_call(), ...) and this presents a problem: (1) If a number of calls on the same socket are in the process of connection to the same peer, a maximum of four concurrent live calls are permitted before further calls need to wait for a slot. (2) If a call is waiting for a slot, it is deep inside sendmsg() or rxrpc_kernel_begin_call() and the entry function is holding the socket lock. (3) sendmsg() and recvmsg() or the in-kernel equivalents are prevented from servicing the other calls as they need to take the socket lock to do so. (4) The socket is stuck until a call is aborted and makes its slot available to the waiter. Fix this by: (1) Provide each call with a mutex ('user_mutex') that arbitrates access by the users of rxrpc separately for each specific call. (2) Make rxrpc_sendmsg() and rxrpc_recvmsg() unlock the socket as soon as they've got a call and taken its mutex. Note that I'm returning EWOULDBLOCK from recvmsg() if MSG_DONTWAIT is set but someone else has the lock. Should I instead only return EWOULDBLOCK if there's nothing currently to be done on a socket, and sleep in this particular instance because there is something to be done, but we appear to be blocked by the interrupt handler doing its ping? (3) Make rxrpc_new_client_call() unlock the socket after allocating a new call, locking its user mutex and adding it to the socket's call tree. The call is returned locked so that sendmsg() can add data to it immediately. From the moment the call is in the socket tree, it is subject to access by sendmsg() and recvmsg() - even if it isn't connected yet. (4) Lock new service calls in the UDP data_ready handler (in rxrpc_new_incoming_call()) because they may already be in the socket's tree and the data_ready handler makes them live immediately if a user ID has already been preassigned. Note that the new call is locked before any notifications are sent that it is live, so doing mutex_trylock() *ought* to always succeed. Userspace is prevented from doing sendmsg() on calls that are in a too-early state in rxrpc_do_sendmsg(). (5) Make rxrpc_new_incoming_call() return the call with the user mutex held so that a ping can be scheduled immediately under it. Note that it might be worth moving the ping call into rxrpc_new_incoming_call() and then we can drop the mutex there. (6) Make rxrpc_accept_call() take the lock on the call it is accepting and release the socket after adding the call to the socket's tree. This is slightly tricky as we've dequeued the call by that point and have to requeue it. Note that requeuing emits a trace event. (7) Make rxrpc_kernel_send_data() and rxrpc_kernel_recv_data() take the new mutex immediately and don't bother with the socket mutex at all. This patch has the nice bonus that calls on the same socket are now to some extent parallelisable. Note that we might want to move rxrpc_service_prealloc() calls out from the socket lock and give it its own lock, so that we don't hang progress in other calls because we're waiting for the allocator. We probably also want to avoid calling rxrpc_notify_socket() from within the socket lock (rxrpc_accept_call()). Signed-off-by: David Howells <dhowells@redhat.com> Tested-by: Marc Dionne <marc.c.dionne@auristor.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-02-27 23:43:06 +08:00
/* The socket is now unlocked */
rxrpc_put_peer(cp.peer);
_leave(" = %p\n", call);
return call;
}
/*
* send a message forming part of a client call through an RxRPC socket
* - caller holds the socket locked
* - the socket may be either a client socket or a server socket
*/
int rxrpc_do_sendmsg(struct rxrpc_sock *rx, struct msghdr *msg, size_t len)
rxrpc: Fix deadlock between call creation and sendmsg/recvmsg All the routines by which rxrpc is accessed from the outside are serialised by means of the socket lock (sendmsg, recvmsg, bind, rxrpc_kernel_begin_call(), ...) and this presents a problem: (1) If a number of calls on the same socket are in the process of connection to the same peer, a maximum of four concurrent live calls are permitted before further calls need to wait for a slot. (2) If a call is waiting for a slot, it is deep inside sendmsg() or rxrpc_kernel_begin_call() and the entry function is holding the socket lock. (3) sendmsg() and recvmsg() or the in-kernel equivalents are prevented from servicing the other calls as they need to take the socket lock to do so. (4) The socket is stuck until a call is aborted and makes its slot available to the waiter. Fix this by: (1) Provide each call with a mutex ('user_mutex') that arbitrates access by the users of rxrpc separately for each specific call. (2) Make rxrpc_sendmsg() and rxrpc_recvmsg() unlock the socket as soon as they've got a call and taken its mutex. Note that I'm returning EWOULDBLOCK from recvmsg() if MSG_DONTWAIT is set but someone else has the lock. Should I instead only return EWOULDBLOCK if there's nothing currently to be done on a socket, and sleep in this particular instance because there is something to be done, but we appear to be blocked by the interrupt handler doing its ping? (3) Make rxrpc_new_client_call() unlock the socket after allocating a new call, locking its user mutex and adding it to the socket's call tree. The call is returned locked so that sendmsg() can add data to it immediately. From the moment the call is in the socket tree, it is subject to access by sendmsg() and recvmsg() - even if it isn't connected yet. (4) Lock new service calls in the UDP data_ready handler (in rxrpc_new_incoming_call()) because they may already be in the socket's tree and the data_ready handler makes them live immediately if a user ID has already been preassigned. Note that the new call is locked before any notifications are sent that it is live, so doing mutex_trylock() *ought* to always succeed. Userspace is prevented from doing sendmsg() on calls that are in a too-early state in rxrpc_do_sendmsg(). (5) Make rxrpc_new_incoming_call() return the call with the user mutex held so that a ping can be scheduled immediately under it. Note that it might be worth moving the ping call into rxrpc_new_incoming_call() and then we can drop the mutex there. (6) Make rxrpc_accept_call() take the lock on the call it is accepting and release the socket after adding the call to the socket's tree. This is slightly tricky as we've dequeued the call by that point and have to requeue it. Note that requeuing emits a trace event. (7) Make rxrpc_kernel_send_data() and rxrpc_kernel_recv_data() take the new mutex immediately and don't bother with the socket mutex at all. This patch has the nice bonus that calls on the same socket are now to some extent parallelisable. Note that we might want to move rxrpc_service_prealloc() calls out from the socket lock and give it its own lock, so that we don't hang progress in other calls because we're waiting for the allocator. We probably also want to avoid calling rxrpc_notify_socket() from within the socket lock (rxrpc_accept_call()). Signed-off-by: David Howells <dhowells@redhat.com> Tested-by: Marc Dionne <marc.c.dionne@auristor.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-02-27 23:43:06 +08:00
__releases(&rx->sk.sk_lock.slock)
__releases(&call->user_mutex)
{
enum rxrpc_call_state state;
struct rxrpc_call *call;
rxrpc: Fix call timeouts Fix the rxrpc call expiration timeouts and make them settable from userspace. By analogy with other rx implementations, there should be three timeouts: (1) "Normal timeout" This is set for all calls and is triggered if we haven't received any packets from the peer in a while. It is measured from the last time we received any packet on that call. This is not reset by any connection packets (such as CHALLENGE/RESPONSE packets). If a service operation takes a long time, the server should generate PING ACKs at a duration that's substantially less than the normal timeout so is to keep both sides alive. This is set at 1/6 of normal timeout. (2) "Idle timeout" This is set only for a service call and is triggered if we stop receiving the DATA packets that comprise the request data. It is measured from the last time we received a DATA packet. (3) "Hard timeout" This can be set for a call and specified the maximum lifetime of that call. It should not be specified by default. Some operations (such as volume transfer) take a long time. Allow userspace to set/change the timeouts on a call with sendmsg, using a control message: RXRPC_SET_CALL_TIMEOUTS The data to the message is a number of 32-bit words, not all of which need be given: u32 hard_timeout; /* sec from first packet */ u32 idle_timeout; /* msec from packet Rx */ u32 normal_timeout; /* msec from data Rx */ This can be set in combination with any other sendmsg() that affects a call. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-24 18:18:41 +08:00
unsigned long now, j;
int ret;
struct rxrpc_send_params p = {
.call.tx_total_len = -1,
.call.user_call_ID = 0,
rxrpc: Fix call timeouts Fix the rxrpc call expiration timeouts and make them settable from userspace. By analogy with other rx implementations, there should be three timeouts: (1) "Normal timeout" This is set for all calls and is triggered if we haven't received any packets from the peer in a while. It is measured from the last time we received any packet on that call. This is not reset by any connection packets (such as CHALLENGE/RESPONSE packets). If a service operation takes a long time, the server should generate PING ACKs at a duration that's substantially less than the normal timeout so is to keep both sides alive. This is set at 1/6 of normal timeout. (2) "Idle timeout" This is set only for a service call and is triggered if we stop receiving the DATA packets that comprise the request data. It is measured from the last time we received a DATA packet. (3) "Hard timeout" This can be set for a call and specified the maximum lifetime of that call. It should not be specified by default. Some operations (such as volume transfer) take a long time. Allow userspace to set/change the timeouts on a call with sendmsg, using a control message: RXRPC_SET_CALL_TIMEOUTS The data to the message is a number of 32-bit words, not all of which need be given: u32 hard_timeout; /* sec from first packet */ u32 idle_timeout; /* msec from packet Rx */ u32 normal_timeout; /* msec from data Rx */ This can be set in combination with any other sendmsg() that affects a call. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-24 18:18:41 +08:00
.call.nr_timeouts = 0,
.call.interruptibility = RXRPC_INTERRUPTIBLE,
.abort_code = 0,
.command = RXRPC_CMD_SEND_DATA,
.exclusive = false,
.upgrade = false,
};
_enter("");
ret = rxrpc_sendmsg_cmsg(msg, &p);
if (ret < 0)
rxrpc: Fix deadlock between call creation and sendmsg/recvmsg All the routines by which rxrpc is accessed from the outside are serialised by means of the socket lock (sendmsg, recvmsg, bind, rxrpc_kernel_begin_call(), ...) and this presents a problem: (1) If a number of calls on the same socket are in the process of connection to the same peer, a maximum of four concurrent live calls are permitted before further calls need to wait for a slot. (2) If a call is waiting for a slot, it is deep inside sendmsg() or rxrpc_kernel_begin_call() and the entry function is holding the socket lock. (3) sendmsg() and recvmsg() or the in-kernel equivalents are prevented from servicing the other calls as they need to take the socket lock to do so. (4) The socket is stuck until a call is aborted and makes its slot available to the waiter. Fix this by: (1) Provide each call with a mutex ('user_mutex') that arbitrates access by the users of rxrpc separately for each specific call. (2) Make rxrpc_sendmsg() and rxrpc_recvmsg() unlock the socket as soon as they've got a call and taken its mutex. Note that I'm returning EWOULDBLOCK from recvmsg() if MSG_DONTWAIT is set but someone else has the lock. Should I instead only return EWOULDBLOCK if there's nothing currently to be done on a socket, and sleep in this particular instance because there is something to be done, but we appear to be blocked by the interrupt handler doing its ping? (3) Make rxrpc_new_client_call() unlock the socket after allocating a new call, locking its user mutex and adding it to the socket's call tree. The call is returned locked so that sendmsg() can add data to it immediately. From the moment the call is in the socket tree, it is subject to access by sendmsg() and recvmsg() - even if it isn't connected yet. (4) Lock new service calls in the UDP data_ready handler (in rxrpc_new_incoming_call()) because they may already be in the socket's tree and the data_ready handler makes them live immediately if a user ID has already been preassigned. Note that the new call is locked before any notifications are sent that it is live, so doing mutex_trylock() *ought* to always succeed. Userspace is prevented from doing sendmsg() on calls that are in a too-early state in rxrpc_do_sendmsg(). (5) Make rxrpc_new_incoming_call() return the call with the user mutex held so that a ping can be scheduled immediately under it. Note that it might be worth moving the ping call into rxrpc_new_incoming_call() and then we can drop the mutex there. (6) Make rxrpc_accept_call() take the lock on the call it is accepting and release the socket after adding the call to the socket's tree. This is slightly tricky as we've dequeued the call by that point and have to requeue it. Note that requeuing emits a trace event. (7) Make rxrpc_kernel_send_data() and rxrpc_kernel_recv_data() take the new mutex immediately and don't bother with the socket mutex at all. This patch has the nice bonus that calls on the same socket are now to some extent parallelisable. Note that we might want to move rxrpc_service_prealloc() calls out from the socket lock and give it its own lock, so that we don't hang progress in other calls because we're waiting for the allocator. We probably also want to avoid calling rxrpc_notify_socket() from within the socket lock (rxrpc_accept_call()). Signed-off-by: David Howells <dhowells@redhat.com> Tested-by: Marc Dionne <marc.c.dionne@auristor.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-02-27 23:43:06 +08:00
goto error_release_sock;
if (p.command == RXRPC_CMD_ACCEPT) {
rxrpc: Fix deadlock between call creation and sendmsg/recvmsg All the routines by which rxrpc is accessed from the outside are serialised by means of the socket lock (sendmsg, recvmsg, bind, rxrpc_kernel_begin_call(), ...) and this presents a problem: (1) If a number of calls on the same socket are in the process of connection to the same peer, a maximum of four concurrent live calls are permitted before further calls need to wait for a slot. (2) If a call is waiting for a slot, it is deep inside sendmsg() or rxrpc_kernel_begin_call() and the entry function is holding the socket lock. (3) sendmsg() and recvmsg() or the in-kernel equivalents are prevented from servicing the other calls as they need to take the socket lock to do so. (4) The socket is stuck until a call is aborted and makes its slot available to the waiter. Fix this by: (1) Provide each call with a mutex ('user_mutex') that arbitrates access by the users of rxrpc separately for each specific call. (2) Make rxrpc_sendmsg() and rxrpc_recvmsg() unlock the socket as soon as they've got a call and taken its mutex. Note that I'm returning EWOULDBLOCK from recvmsg() if MSG_DONTWAIT is set but someone else has the lock. Should I instead only return EWOULDBLOCK if there's nothing currently to be done on a socket, and sleep in this particular instance because there is something to be done, but we appear to be blocked by the interrupt handler doing its ping? (3) Make rxrpc_new_client_call() unlock the socket after allocating a new call, locking its user mutex and adding it to the socket's call tree. The call is returned locked so that sendmsg() can add data to it immediately. From the moment the call is in the socket tree, it is subject to access by sendmsg() and recvmsg() - even if it isn't connected yet. (4) Lock new service calls in the UDP data_ready handler (in rxrpc_new_incoming_call()) because they may already be in the socket's tree and the data_ready handler makes them live immediately if a user ID has already been preassigned. Note that the new call is locked before any notifications are sent that it is live, so doing mutex_trylock() *ought* to always succeed. Userspace is prevented from doing sendmsg() on calls that are in a too-early state in rxrpc_do_sendmsg(). (5) Make rxrpc_new_incoming_call() return the call with the user mutex held so that a ping can be scheduled immediately under it. Note that it might be worth moving the ping call into rxrpc_new_incoming_call() and then we can drop the mutex there. (6) Make rxrpc_accept_call() take the lock on the call it is accepting and release the socket after adding the call to the socket's tree. This is slightly tricky as we've dequeued the call by that point and have to requeue it. Note that requeuing emits a trace event. (7) Make rxrpc_kernel_send_data() and rxrpc_kernel_recv_data() take the new mutex immediately and don't bother with the socket mutex at all. This patch has the nice bonus that calls on the same socket are now to some extent parallelisable. Note that we might want to move rxrpc_service_prealloc() calls out from the socket lock and give it its own lock, so that we don't hang progress in other calls because we're waiting for the allocator. We probably also want to avoid calling rxrpc_notify_socket() from within the socket lock (rxrpc_accept_call()). Signed-off-by: David Howells <dhowells@redhat.com> Tested-by: Marc Dionne <marc.c.dionne@auristor.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-02-27 23:43:06 +08:00
ret = -EINVAL;
if (rx->sk.sk_state != RXRPC_SERVER_LISTENING)
rxrpc: Fix deadlock between call creation and sendmsg/recvmsg All the routines by which rxrpc is accessed from the outside are serialised by means of the socket lock (sendmsg, recvmsg, bind, rxrpc_kernel_begin_call(), ...) and this presents a problem: (1) If a number of calls on the same socket are in the process of connection to the same peer, a maximum of four concurrent live calls are permitted before further calls need to wait for a slot. (2) If a call is waiting for a slot, it is deep inside sendmsg() or rxrpc_kernel_begin_call() and the entry function is holding the socket lock. (3) sendmsg() and recvmsg() or the in-kernel equivalents are prevented from servicing the other calls as they need to take the socket lock to do so. (4) The socket is stuck until a call is aborted and makes its slot available to the waiter. Fix this by: (1) Provide each call with a mutex ('user_mutex') that arbitrates access by the users of rxrpc separately for each specific call. (2) Make rxrpc_sendmsg() and rxrpc_recvmsg() unlock the socket as soon as they've got a call and taken its mutex. Note that I'm returning EWOULDBLOCK from recvmsg() if MSG_DONTWAIT is set but someone else has the lock. Should I instead only return EWOULDBLOCK if there's nothing currently to be done on a socket, and sleep in this particular instance because there is something to be done, but we appear to be blocked by the interrupt handler doing its ping? (3) Make rxrpc_new_client_call() unlock the socket after allocating a new call, locking its user mutex and adding it to the socket's call tree. The call is returned locked so that sendmsg() can add data to it immediately. From the moment the call is in the socket tree, it is subject to access by sendmsg() and recvmsg() - even if it isn't connected yet. (4) Lock new service calls in the UDP data_ready handler (in rxrpc_new_incoming_call()) because they may already be in the socket's tree and the data_ready handler makes them live immediately if a user ID has already been preassigned. Note that the new call is locked before any notifications are sent that it is live, so doing mutex_trylock() *ought* to always succeed. Userspace is prevented from doing sendmsg() on calls that are in a too-early state in rxrpc_do_sendmsg(). (5) Make rxrpc_new_incoming_call() return the call with the user mutex held so that a ping can be scheduled immediately under it. Note that it might be worth moving the ping call into rxrpc_new_incoming_call() and then we can drop the mutex there. (6) Make rxrpc_accept_call() take the lock on the call it is accepting and release the socket after adding the call to the socket's tree. This is slightly tricky as we've dequeued the call by that point and have to requeue it. Note that requeuing emits a trace event. (7) Make rxrpc_kernel_send_data() and rxrpc_kernel_recv_data() take the new mutex immediately and don't bother with the socket mutex at all. This patch has the nice bonus that calls on the same socket are now to some extent parallelisable. Note that we might want to move rxrpc_service_prealloc() calls out from the socket lock and give it its own lock, so that we don't hang progress in other calls because we're waiting for the allocator. We probably also want to avoid calling rxrpc_notify_socket() from within the socket lock (rxrpc_accept_call()). Signed-off-by: David Howells <dhowells@redhat.com> Tested-by: Marc Dionne <marc.c.dionne@auristor.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-02-27 23:43:06 +08:00
goto error_release_sock;
call = rxrpc_accept_call(rx, p.call.user_call_ID, NULL);
rxrpc: Fix deadlock between call creation and sendmsg/recvmsg All the routines by which rxrpc is accessed from the outside are serialised by means of the socket lock (sendmsg, recvmsg, bind, rxrpc_kernel_begin_call(), ...) and this presents a problem: (1) If a number of calls on the same socket are in the process of connection to the same peer, a maximum of four concurrent live calls are permitted before further calls need to wait for a slot. (2) If a call is waiting for a slot, it is deep inside sendmsg() or rxrpc_kernel_begin_call() and the entry function is holding the socket lock. (3) sendmsg() and recvmsg() or the in-kernel equivalents are prevented from servicing the other calls as they need to take the socket lock to do so. (4) The socket is stuck until a call is aborted and makes its slot available to the waiter. Fix this by: (1) Provide each call with a mutex ('user_mutex') that arbitrates access by the users of rxrpc separately for each specific call. (2) Make rxrpc_sendmsg() and rxrpc_recvmsg() unlock the socket as soon as they've got a call and taken its mutex. Note that I'm returning EWOULDBLOCK from recvmsg() if MSG_DONTWAIT is set but someone else has the lock. Should I instead only return EWOULDBLOCK if there's nothing currently to be done on a socket, and sleep in this particular instance because there is something to be done, but we appear to be blocked by the interrupt handler doing its ping? (3) Make rxrpc_new_client_call() unlock the socket after allocating a new call, locking its user mutex and adding it to the socket's call tree. The call is returned locked so that sendmsg() can add data to it immediately. From the moment the call is in the socket tree, it is subject to access by sendmsg() and recvmsg() - even if it isn't connected yet. (4) Lock new service calls in the UDP data_ready handler (in rxrpc_new_incoming_call()) because they may already be in the socket's tree and the data_ready handler makes them live immediately if a user ID has already been preassigned. Note that the new call is locked before any notifications are sent that it is live, so doing mutex_trylock() *ought* to always succeed. Userspace is prevented from doing sendmsg() on calls that are in a too-early state in rxrpc_do_sendmsg(). (5) Make rxrpc_new_incoming_call() return the call with the user mutex held so that a ping can be scheduled immediately under it. Note that it might be worth moving the ping call into rxrpc_new_incoming_call() and then we can drop the mutex there. (6) Make rxrpc_accept_call() take the lock on the call it is accepting and release the socket after adding the call to the socket's tree. This is slightly tricky as we've dequeued the call by that point and have to requeue it. Note that requeuing emits a trace event. (7) Make rxrpc_kernel_send_data() and rxrpc_kernel_recv_data() take the new mutex immediately and don't bother with the socket mutex at all. This patch has the nice bonus that calls on the same socket are now to some extent parallelisable. Note that we might want to move rxrpc_service_prealloc() calls out from the socket lock and give it its own lock, so that we don't hang progress in other calls because we're waiting for the allocator. We probably also want to avoid calling rxrpc_notify_socket() from within the socket lock (rxrpc_accept_call()). Signed-off-by: David Howells <dhowells@redhat.com> Tested-by: Marc Dionne <marc.c.dionne@auristor.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-02-27 23:43:06 +08:00
/* The socket is now unlocked. */
if (IS_ERR(call))
return PTR_ERR(call);
ret = 0;
goto out_put_unlock;
}
call = rxrpc_find_call_by_user_ID(rx, p.call.user_call_ID);
if (!call) {
rxrpc: Fix deadlock between call creation and sendmsg/recvmsg All the routines by which rxrpc is accessed from the outside are serialised by means of the socket lock (sendmsg, recvmsg, bind, rxrpc_kernel_begin_call(), ...) and this presents a problem: (1) If a number of calls on the same socket are in the process of connection to the same peer, a maximum of four concurrent live calls are permitted before further calls need to wait for a slot. (2) If a call is waiting for a slot, it is deep inside sendmsg() or rxrpc_kernel_begin_call() and the entry function is holding the socket lock. (3) sendmsg() and recvmsg() or the in-kernel equivalents are prevented from servicing the other calls as they need to take the socket lock to do so. (4) The socket is stuck until a call is aborted and makes its slot available to the waiter. Fix this by: (1) Provide each call with a mutex ('user_mutex') that arbitrates access by the users of rxrpc separately for each specific call. (2) Make rxrpc_sendmsg() and rxrpc_recvmsg() unlock the socket as soon as they've got a call and taken its mutex. Note that I'm returning EWOULDBLOCK from recvmsg() if MSG_DONTWAIT is set but someone else has the lock. Should I instead only return EWOULDBLOCK if there's nothing currently to be done on a socket, and sleep in this particular instance because there is something to be done, but we appear to be blocked by the interrupt handler doing its ping? (3) Make rxrpc_new_client_call() unlock the socket after allocating a new call, locking its user mutex and adding it to the socket's call tree. The call is returned locked so that sendmsg() can add data to it immediately. From the moment the call is in the socket tree, it is subject to access by sendmsg() and recvmsg() - even if it isn't connected yet. (4) Lock new service calls in the UDP data_ready handler (in rxrpc_new_incoming_call()) because they may already be in the socket's tree and the data_ready handler makes them live immediately if a user ID has already been preassigned. Note that the new call is locked before any notifications are sent that it is live, so doing mutex_trylock() *ought* to always succeed. Userspace is prevented from doing sendmsg() on calls that are in a too-early state in rxrpc_do_sendmsg(). (5) Make rxrpc_new_incoming_call() return the call with the user mutex held so that a ping can be scheduled immediately under it. Note that it might be worth moving the ping call into rxrpc_new_incoming_call() and then we can drop the mutex there. (6) Make rxrpc_accept_call() take the lock on the call it is accepting and release the socket after adding the call to the socket's tree. This is slightly tricky as we've dequeued the call by that point and have to requeue it. Note that requeuing emits a trace event. (7) Make rxrpc_kernel_send_data() and rxrpc_kernel_recv_data() take the new mutex immediately and don't bother with the socket mutex at all. This patch has the nice bonus that calls on the same socket are now to some extent parallelisable. Note that we might want to move rxrpc_service_prealloc() calls out from the socket lock and give it its own lock, so that we don't hang progress in other calls because we're waiting for the allocator. We probably also want to avoid calling rxrpc_notify_socket() from within the socket lock (rxrpc_accept_call()). Signed-off-by: David Howells <dhowells@redhat.com> Tested-by: Marc Dionne <marc.c.dionne@auristor.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-02-27 23:43:06 +08:00
ret = -EBADSLT;
if (p.command != RXRPC_CMD_SEND_DATA)
rxrpc: Fix deadlock between call creation and sendmsg/recvmsg All the routines by which rxrpc is accessed from the outside are serialised by means of the socket lock (sendmsg, recvmsg, bind, rxrpc_kernel_begin_call(), ...) and this presents a problem: (1) If a number of calls on the same socket are in the process of connection to the same peer, a maximum of four concurrent live calls are permitted before further calls need to wait for a slot. (2) If a call is waiting for a slot, it is deep inside sendmsg() or rxrpc_kernel_begin_call() and the entry function is holding the socket lock. (3) sendmsg() and recvmsg() or the in-kernel equivalents are prevented from servicing the other calls as they need to take the socket lock to do so. (4) The socket is stuck until a call is aborted and makes its slot available to the waiter. Fix this by: (1) Provide each call with a mutex ('user_mutex') that arbitrates access by the users of rxrpc separately for each specific call. (2) Make rxrpc_sendmsg() and rxrpc_recvmsg() unlock the socket as soon as they've got a call and taken its mutex. Note that I'm returning EWOULDBLOCK from recvmsg() if MSG_DONTWAIT is set but someone else has the lock. Should I instead only return EWOULDBLOCK if there's nothing currently to be done on a socket, and sleep in this particular instance because there is something to be done, but we appear to be blocked by the interrupt handler doing its ping? (3) Make rxrpc_new_client_call() unlock the socket after allocating a new call, locking its user mutex and adding it to the socket's call tree. The call is returned locked so that sendmsg() can add data to it immediately. From the moment the call is in the socket tree, it is subject to access by sendmsg() and recvmsg() - even if it isn't connected yet. (4) Lock new service calls in the UDP data_ready handler (in rxrpc_new_incoming_call()) because they may already be in the socket's tree and the data_ready handler makes them live immediately if a user ID has already been preassigned. Note that the new call is locked before any notifications are sent that it is live, so doing mutex_trylock() *ought* to always succeed. Userspace is prevented from doing sendmsg() on calls that are in a too-early state in rxrpc_do_sendmsg(). (5) Make rxrpc_new_incoming_call() return the call with the user mutex held so that a ping can be scheduled immediately under it. Note that it might be worth moving the ping call into rxrpc_new_incoming_call() and then we can drop the mutex there. (6) Make rxrpc_accept_call() take the lock on the call it is accepting and release the socket after adding the call to the socket's tree. This is slightly tricky as we've dequeued the call by that point and have to requeue it. Note that requeuing emits a trace event. (7) Make rxrpc_kernel_send_data() and rxrpc_kernel_recv_data() take the new mutex immediately and don't bother with the socket mutex at all. This patch has the nice bonus that calls on the same socket are now to some extent parallelisable. Note that we might want to move rxrpc_service_prealloc() calls out from the socket lock and give it its own lock, so that we don't hang progress in other calls because we're waiting for the allocator. We probably also want to avoid calling rxrpc_notify_socket() from within the socket lock (rxrpc_accept_call()). Signed-off-by: David Howells <dhowells@redhat.com> Tested-by: Marc Dionne <marc.c.dionne@auristor.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-02-27 23:43:06 +08:00
goto error_release_sock;
call = rxrpc_new_client_call_for_sendmsg(rx, msg, &p);
rxrpc: Fix deadlock between call creation and sendmsg/recvmsg All the routines by which rxrpc is accessed from the outside are serialised by means of the socket lock (sendmsg, recvmsg, bind, rxrpc_kernel_begin_call(), ...) and this presents a problem: (1) If a number of calls on the same socket are in the process of connection to the same peer, a maximum of four concurrent live calls are permitted before further calls need to wait for a slot. (2) If a call is waiting for a slot, it is deep inside sendmsg() or rxrpc_kernel_begin_call() and the entry function is holding the socket lock. (3) sendmsg() and recvmsg() or the in-kernel equivalents are prevented from servicing the other calls as they need to take the socket lock to do so. (4) The socket is stuck until a call is aborted and makes its slot available to the waiter. Fix this by: (1) Provide each call with a mutex ('user_mutex') that arbitrates access by the users of rxrpc separately for each specific call. (2) Make rxrpc_sendmsg() and rxrpc_recvmsg() unlock the socket as soon as they've got a call and taken its mutex. Note that I'm returning EWOULDBLOCK from recvmsg() if MSG_DONTWAIT is set but someone else has the lock. Should I instead only return EWOULDBLOCK if there's nothing currently to be done on a socket, and sleep in this particular instance because there is something to be done, but we appear to be blocked by the interrupt handler doing its ping? (3) Make rxrpc_new_client_call() unlock the socket after allocating a new call, locking its user mutex and adding it to the socket's call tree. The call is returned locked so that sendmsg() can add data to it immediately. From the moment the call is in the socket tree, it is subject to access by sendmsg() and recvmsg() - even if it isn't connected yet. (4) Lock new service calls in the UDP data_ready handler (in rxrpc_new_incoming_call()) because they may already be in the socket's tree and the data_ready handler makes them live immediately if a user ID has already been preassigned. Note that the new call is locked before any notifications are sent that it is live, so doing mutex_trylock() *ought* to always succeed. Userspace is prevented from doing sendmsg() on calls that are in a too-early state in rxrpc_do_sendmsg(). (5) Make rxrpc_new_incoming_call() return the call with the user mutex held so that a ping can be scheduled immediately under it. Note that it might be worth moving the ping call into rxrpc_new_incoming_call() and then we can drop the mutex there. (6) Make rxrpc_accept_call() take the lock on the call it is accepting and release the socket after adding the call to the socket's tree. This is slightly tricky as we've dequeued the call by that point and have to requeue it. Note that requeuing emits a trace event. (7) Make rxrpc_kernel_send_data() and rxrpc_kernel_recv_data() take the new mutex immediately and don't bother with the socket mutex at all. This patch has the nice bonus that calls on the same socket are now to some extent parallelisable. Note that we might want to move rxrpc_service_prealloc() calls out from the socket lock and give it its own lock, so that we don't hang progress in other calls because we're waiting for the allocator. We probably also want to avoid calling rxrpc_notify_socket() from within the socket lock (rxrpc_accept_call()). Signed-off-by: David Howells <dhowells@redhat.com> Tested-by: Marc Dionne <marc.c.dionne@auristor.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-02-27 23:43:06 +08:00
/* The socket is now unlocked... */
if (IS_ERR(call))
return PTR_ERR(call);
rxrpc: Fix deadlock between call creation and sendmsg/recvmsg All the routines by which rxrpc is accessed from the outside are serialised by means of the socket lock (sendmsg, recvmsg, bind, rxrpc_kernel_begin_call(), ...) and this presents a problem: (1) If a number of calls on the same socket are in the process of connection to the same peer, a maximum of four concurrent live calls are permitted before further calls need to wait for a slot. (2) If a call is waiting for a slot, it is deep inside sendmsg() or rxrpc_kernel_begin_call() and the entry function is holding the socket lock. (3) sendmsg() and recvmsg() or the in-kernel equivalents are prevented from servicing the other calls as they need to take the socket lock to do so. (4) The socket is stuck until a call is aborted and makes its slot available to the waiter. Fix this by: (1) Provide each call with a mutex ('user_mutex') that arbitrates access by the users of rxrpc separately for each specific call. (2) Make rxrpc_sendmsg() and rxrpc_recvmsg() unlock the socket as soon as they've got a call and taken its mutex. Note that I'm returning EWOULDBLOCK from recvmsg() if MSG_DONTWAIT is set but someone else has the lock. Should I instead only return EWOULDBLOCK if there's nothing currently to be done on a socket, and sleep in this particular instance because there is something to be done, but we appear to be blocked by the interrupt handler doing its ping? (3) Make rxrpc_new_client_call() unlock the socket after allocating a new call, locking its user mutex and adding it to the socket's call tree. The call is returned locked so that sendmsg() can add data to it immediately. From the moment the call is in the socket tree, it is subject to access by sendmsg() and recvmsg() - even if it isn't connected yet. (4) Lock new service calls in the UDP data_ready handler (in rxrpc_new_incoming_call()) because they may already be in the socket's tree and the data_ready handler makes them live immediately if a user ID has already been preassigned. Note that the new call is locked before any notifications are sent that it is live, so doing mutex_trylock() *ought* to always succeed. Userspace is prevented from doing sendmsg() on calls that are in a too-early state in rxrpc_do_sendmsg(). (5) Make rxrpc_new_incoming_call() return the call with the user mutex held so that a ping can be scheduled immediately under it. Note that it might be worth moving the ping call into rxrpc_new_incoming_call() and then we can drop the mutex there. (6) Make rxrpc_accept_call() take the lock on the call it is accepting and release the socket after adding the call to the socket's tree. This is slightly tricky as we've dequeued the call by that point and have to requeue it. Note that requeuing emits a trace event. (7) Make rxrpc_kernel_send_data() and rxrpc_kernel_recv_data() take the new mutex immediately and don't bother with the socket mutex at all. This patch has the nice bonus that calls on the same socket are now to some extent parallelisable. Note that we might want to move rxrpc_service_prealloc() calls out from the socket lock and give it its own lock, so that we don't hang progress in other calls because we're waiting for the allocator. We probably also want to avoid calling rxrpc_notify_socket() from within the socket lock (rxrpc_accept_call()). Signed-off-by: David Howells <dhowells@redhat.com> Tested-by: Marc Dionne <marc.c.dionne@auristor.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-02-27 23:43:06 +08:00
/* ... and we have the call lock. */
} else {
switch (READ_ONCE(call->state)) {
case RXRPC_CALL_UNINITIALISED:
case RXRPC_CALL_CLIENT_AWAIT_CONN:
case RXRPC_CALL_SERVER_PREALLOC:
case RXRPC_CALL_SERVER_SECURING:
case RXRPC_CALL_SERVER_ACCEPTING:
rxrpc_put_call(call, rxrpc_call_put);
ret = -EBUSY;
goto error_release_sock;
default:
break;
}
rxrpc: Fix deadlock between call creation and sendmsg/recvmsg All the routines by which rxrpc is accessed from the outside are serialised by means of the socket lock (sendmsg, recvmsg, bind, rxrpc_kernel_begin_call(), ...) and this presents a problem: (1) If a number of calls on the same socket are in the process of connection to the same peer, a maximum of four concurrent live calls are permitted before further calls need to wait for a slot. (2) If a call is waiting for a slot, it is deep inside sendmsg() or rxrpc_kernel_begin_call() and the entry function is holding the socket lock. (3) sendmsg() and recvmsg() or the in-kernel equivalents are prevented from servicing the other calls as they need to take the socket lock to do so. (4) The socket is stuck until a call is aborted and makes its slot available to the waiter. Fix this by: (1) Provide each call with a mutex ('user_mutex') that arbitrates access by the users of rxrpc separately for each specific call. (2) Make rxrpc_sendmsg() and rxrpc_recvmsg() unlock the socket as soon as they've got a call and taken its mutex. Note that I'm returning EWOULDBLOCK from recvmsg() if MSG_DONTWAIT is set but someone else has the lock. Should I instead only return EWOULDBLOCK if there's nothing currently to be done on a socket, and sleep in this particular instance because there is something to be done, but we appear to be blocked by the interrupt handler doing its ping? (3) Make rxrpc_new_client_call() unlock the socket after allocating a new call, locking its user mutex and adding it to the socket's call tree. The call is returned locked so that sendmsg() can add data to it immediately. From the moment the call is in the socket tree, it is subject to access by sendmsg() and recvmsg() - even if it isn't connected yet. (4) Lock new service calls in the UDP data_ready handler (in rxrpc_new_incoming_call()) because they may already be in the socket's tree and the data_ready handler makes them live immediately if a user ID has already been preassigned. Note that the new call is locked before any notifications are sent that it is live, so doing mutex_trylock() *ought* to always succeed. Userspace is prevented from doing sendmsg() on calls that are in a too-early state in rxrpc_do_sendmsg(). (5) Make rxrpc_new_incoming_call() return the call with the user mutex held so that a ping can be scheduled immediately under it. Note that it might be worth moving the ping call into rxrpc_new_incoming_call() and then we can drop the mutex there. (6) Make rxrpc_accept_call() take the lock on the call it is accepting and release the socket after adding the call to the socket's tree. This is slightly tricky as we've dequeued the call by that point and have to requeue it. Note that requeuing emits a trace event. (7) Make rxrpc_kernel_send_data() and rxrpc_kernel_recv_data() take the new mutex immediately and don't bother with the socket mutex at all. This patch has the nice bonus that calls on the same socket are now to some extent parallelisable. Note that we might want to move rxrpc_service_prealloc() calls out from the socket lock and give it its own lock, so that we don't hang progress in other calls because we're waiting for the allocator. We probably also want to avoid calling rxrpc_notify_socket() from within the socket lock (rxrpc_accept_call()). Signed-off-by: David Howells <dhowells@redhat.com> Tested-by: Marc Dionne <marc.c.dionne@auristor.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-02-27 23:43:06 +08:00
ret = mutex_lock_interruptible(&call->user_mutex);
release_sock(&rx->sk);
if (ret < 0) {
ret = -ERESTARTSYS;
goto error_put;
}
if (p.call.tx_total_len != -1) {
ret = -EINVAL;
if (call->tx_total_len != -1 ||
call->tx_pending ||
call->tx_top != 0)
goto error_put;
call->tx_total_len = p.call.tx_total_len;
}
}
rxrpc: Fix call timeouts Fix the rxrpc call expiration timeouts and make them settable from userspace. By analogy with other rx implementations, there should be three timeouts: (1) "Normal timeout" This is set for all calls and is triggered if we haven't received any packets from the peer in a while. It is measured from the last time we received any packet on that call. This is not reset by any connection packets (such as CHALLENGE/RESPONSE packets). If a service operation takes a long time, the server should generate PING ACKs at a duration that's substantially less than the normal timeout so is to keep both sides alive. This is set at 1/6 of normal timeout. (2) "Idle timeout" This is set only for a service call and is triggered if we stop receiving the DATA packets that comprise the request data. It is measured from the last time we received a DATA packet. (3) "Hard timeout" This can be set for a call and specified the maximum lifetime of that call. It should not be specified by default. Some operations (such as volume transfer) take a long time. Allow userspace to set/change the timeouts on a call with sendmsg, using a control message: RXRPC_SET_CALL_TIMEOUTS The data to the message is a number of 32-bit words, not all of which need be given: u32 hard_timeout; /* sec from first packet */ u32 idle_timeout; /* msec from packet Rx */ u32 normal_timeout; /* msec from data Rx */ This can be set in combination with any other sendmsg() that affects a call. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-24 18:18:41 +08:00
switch (p.call.nr_timeouts) {
case 3:
j = msecs_to_jiffies(p.call.timeouts.normal);
if (p.call.timeouts.normal > 0 && j == 0)
j = 1;
WRITE_ONCE(call->next_rx_timo, j);
/* Fall through */
case 2:
j = msecs_to_jiffies(p.call.timeouts.idle);
if (p.call.timeouts.idle > 0 && j == 0)
j = 1;
WRITE_ONCE(call->next_req_timo, j);
/* Fall through */
case 1:
if (p.call.timeouts.hard > 0) {
j = msecs_to_jiffies(p.call.timeouts.hard);
now = jiffies;
j += now;
WRITE_ONCE(call->expect_term_by, j);
rxrpc_reduce_call_timer(call, j, now,
rxrpc_timer_set_for_hard);
}
break;
}
state = READ_ONCE(call->state);
_debug("CALL %d USR %lx ST %d on CONN %p",
call->debug_id, call->user_call_ID, state, call->conn);
if (state >= RXRPC_CALL_COMPLETE) {
/* it's too late for this call */
ret = -ESHUTDOWN;
} else if (p.command == RXRPC_CMD_SEND_ABORT) {
ret = 0;
if (rxrpc_abort_call("CMD", call, 0, p.abort_code, -ECONNABORTED))
ret = rxrpc_send_abort_packet(call);
} else if (p.command != RXRPC_CMD_SEND_DATA) {
ret = -EINVAL;
} else if (rxrpc_is_client_call(call) &&
state != RXRPC_CALL_CLIENT_SEND_REQUEST) {
/* request phase complete for this client call */
ret = -EPROTO;
} else if (rxrpc_is_service_call(call) &&
state != RXRPC_CALL_SERVER_ACK_REQUEST &&
state != RXRPC_CALL_SERVER_SEND_REPLY) {
/* Reply phase not begun or not complete for service call. */
ret = -EPROTO;
} else {
ret = rxrpc_send_data(rx, call, msg, len, NULL);
}
out_put_unlock:
rxrpc: Fix deadlock between call creation and sendmsg/recvmsg All the routines by which rxrpc is accessed from the outside are serialised by means of the socket lock (sendmsg, recvmsg, bind, rxrpc_kernel_begin_call(), ...) and this presents a problem: (1) If a number of calls on the same socket are in the process of connection to the same peer, a maximum of four concurrent live calls are permitted before further calls need to wait for a slot. (2) If a call is waiting for a slot, it is deep inside sendmsg() or rxrpc_kernel_begin_call() and the entry function is holding the socket lock. (3) sendmsg() and recvmsg() or the in-kernel equivalents are prevented from servicing the other calls as they need to take the socket lock to do so. (4) The socket is stuck until a call is aborted and makes its slot available to the waiter. Fix this by: (1) Provide each call with a mutex ('user_mutex') that arbitrates access by the users of rxrpc separately for each specific call. (2) Make rxrpc_sendmsg() and rxrpc_recvmsg() unlock the socket as soon as they've got a call and taken its mutex. Note that I'm returning EWOULDBLOCK from recvmsg() if MSG_DONTWAIT is set but someone else has the lock. Should I instead only return EWOULDBLOCK if there's nothing currently to be done on a socket, and sleep in this particular instance because there is something to be done, but we appear to be blocked by the interrupt handler doing its ping? (3) Make rxrpc_new_client_call() unlock the socket after allocating a new call, locking its user mutex and adding it to the socket's call tree. The call is returned locked so that sendmsg() can add data to it immediately. From the moment the call is in the socket tree, it is subject to access by sendmsg() and recvmsg() - even if it isn't connected yet. (4) Lock new service calls in the UDP data_ready handler (in rxrpc_new_incoming_call()) because they may already be in the socket's tree and the data_ready handler makes them live immediately if a user ID has already been preassigned. Note that the new call is locked before any notifications are sent that it is live, so doing mutex_trylock() *ought* to always succeed. Userspace is prevented from doing sendmsg() on calls that are in a too-early state in rxrpc_do_sendmsg(). (5) Make rxrpc_new_incoming_call() return the call with the user mutex held so that a ping can be scheduled immediately under it. Note that it might be worth moving the ping call into rxrpc_new_incoming_call() and then we can drop the mutex there. (6) Make rxrpc_accept_call() take the lock on the call it is accepting and release the socket after adding the call to the socket's tree. This is slightly tricky as we've dequeued the call by that point and have to requeue it. Note that requeuing emits a trace event. (7) Make rxrpc_kernel_send_data() and rxrpc_kernel_recv_data() take the new mutex immediately and don't bother with the socket mutex at all. This patch has the nice bonus that calls on the same socket are now to some extent parallelisable. Note that we might want to move rxrpc_service_prealloc() calls out from the socket lock and give it its own lock, so that we don't hang progress in other calls because we're waiting for the allocator. We probably also want to avoid calling rxrpc_notify_socket() from within the socket lock (rxrpc_accept_call()). Signed-off-by: David Howells <dhowells@redhat.com> Tested-by: Marc Dionne <marc.c.dionne@auristor.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-02-27 23:43:06 +08:00
mutex_unlock(&call->user_mutex);
error_put:
rxrpc_put_call(call, rxrpc_call_put);
_leave(" = %d", ret);
return ret;
rxrpc: Fix deadlock between call creation and sendmsg/recvmsg All the routines by which rxrpc is accessed from the outside are serialised by means of the socket lock (sendmsg, recvmsg, bind, rxrpc_kernel_begin_call(), ...) and this presents a problem: (1) If a number of calls on the same socket are in the process of connection to the same peer, a maximum of four concurrent live calls are permitted before further calls need to wait for a slot. (2) If a call is waiting for a slot, it is deep inside sendmsg() or rxrpc_kernel_begin_call() and the entry function is holding the socket lock. (3) sendmsg() and recvmsg() or the in-kernel equivalents are prevented from servicing the other calls as they need to take the socket lock to do so. (4) The socket is stuck until a call is aborted and makes its slot available to the waiter. Fix this by: (1) Provide each call with a mutex ('user_mutex') that arbitrates access by the users of rxrpc separately for each specific call. (2) Make rxrpc_sendmsg() and rxrpc_recvmsg() unlock the socket as soon as they've got a call and taken its mutex. Note that I'm returning EWOULDBLOCK from recvmsg() if MSG_DONTWAIT is set but someone else has the lock. Should I instead only return EWOULDBLOCK if there's nothing currently to be done on a socket, and sleep in this particular instance because there is something to be done, but we appear to be blocked by the interrupt handler doing its ping? (3) Make rxrpc_new_client_call() unlock the socket after allocating a new call, locking its user mutex and adding it to the socket's call tree. The call is returned locked so that sendmsg() can add data to it immediately. From the moment the call is in the socket tree, it is subject to access by sendmsg() and recvmsg() - even if it isn't connected yet. (4) Lock new service calls in the UDP data_ready handler (in rxrpc_new_incoming_call()) because they may already be in the socket's tree and the data_ready handler makes them live immediately if a user ID has already been preassigned. Note that the new call is locked before any notifications are sent that it is live, so doing mutex_trylock() *ought* to always succeed. Userspace is prevented from doing sendmsg() on calls that are in a too-early state in rxrpc_do_sendmsg(). (5) Make rxrpc_new_incoming_call() return the call with the user mutex held so that a ping can be scheduled immediately under it. Note that it might be worth moving the ping call into rxrpc_new_incoming_call() and then we can drop the mutex there. (6) Make rxrpc_accept_call() take the lock on the call it is accepting and release the socket after adding the call to the socket's tree. This is slightly tricky as we've dequeued the call by that point and have to requeue it. Note that requeuing emits a trace event. (7) Make rxrpc_kernel_send_data() and rxrpc_kernel_recv_data() take the new mutex immediately and don't bother with the socket mutex at all. This patch has the nice bonus that calls on the same socket are now to some extent parallelisable. Note that we might want to move rxrpc_service_prealloc() calls out from the socket lock and give it its own lock, so that we don't hang progress in other calls because we're waiting for the allocator. We probably also want to avoid calling rxrpc_notify_socket() from within the socket lock (rxrpc_accept_call()). Signed-off-by: David Howells <dhowells@redhat.com> Tested-by: Marc Dionne <marc.c.dionne@auristor.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-02-27 23:43:06 +08:00
error_release_sock:
release_sock(&rx->sk);
return ret;
}
/**
* rxrpc_kernel_send_data - Allow a kernel service to send data on a call
* @sock: The socket the call is on
* @call: The call to send data through
* @msg: The data to send
* @len: The amount of data to send
* @notify_end_tx: Notification that the last packet is queued.
*
* Allow a kernel service to send data on a call. The call must be in an state
* appropriate to sending data. No control data should be supplied in @msg,
* nor should an address be supplied. MSG_MORE should be flagged if there's
* more data to come, otherwise this data will end the transmission phase.
*/
int rxrpc_kernel_send_data(struct socket *sock, struct rxrpc_call *call,
struct msghdr *msg, size_t len,
rxrpc_notify_end_tx_t notify_end_tx)
{
int ret;
_enter("{%d,%s},", call->debug_id, rxrpc_call_states[call->state]);
ASSERTCMP(msg->msg_name, ==, NULL);
ASSERTCMP(msg->msg_control, ==, NULL);
rxrpc: Fix deadlock between call creation and sendmsg/recvmsg All the routines by which rxrpc is accessed from the outside are serialised by means of the socket lock (sendmsg, recvmsg, bind, rxrpc_kernel_begin_call(), ...) and this presents a problem: (1) If a number of calls on the same socket are in the process of connection to the same peer, a maximum of four concurrent live calls are permitted before further calls need to wait for a slot. (2) If a call is waiting for a slot, it is deep inside sendmsg() or rxrpc_kernel_begin_call() and the entry function is holding the socket lock. (3) sendmsg() and recvmsg() or the in-kernel equivalents are prevented from servicing the other calls as they need to take the socket lock to do so. (4) The socket is stuck until a call is aborted and makes its slot available to the waiter. Fix this by: (1) Provide each call with a mutex ('user_mutex') that arbitrates access by the users of rxrpc separately for each specific call. (2) Make rxrpc_sendmsg() and rxrpc_recvmsg() unlock the socket as soon as they've got a call and taken its mutex. Note that I'm returning EWOULDBLOCK from recvmsg() if MSG_DONTWAIT is set but someone else has the lock. Should I instead only return EWOULDBLOCK if there's nothing currently to be done on a socket, and sleep in this particular instance because there is something to be done, but we appear to be blocked by the interrupt handler doing its ping? (3) Make rxrpc_new_client_call() unlock the socket after allocating a new call, locking its user mutex and adding it to the socket's call tree. The call is returned locked so that sendmsg() can add data to it immediately. From the moment the call is in the socket tree, it is subject to access by sendmsg() and recvmsg() - even if it isn't connected yet. (4) Lock new service calls in the UDP data_ready handler (in rxrpc_new_incoming_call()) because they may already be in the socket's tree and the data_ready handler makes them live immediately if a user ID has already been preassigned. Note that the new call is locked before any notifications are sent that it is live, so doing mutex_trylock() *ought* to always succeed. Userspace is prevented from doing sendmsg() on calls that are in a too-early state in rxrpc_do_sendmsg(). (5) Make rxrpc_new_incoming_call() return the call with the user mutex held so that a ping can be scheduled immediately under it. Note that it might be worth moving the ping call into rxrpc_new_incoming_call() and then we can drop the mutex there. (6) Make rxrpc_accept_call() take the lock on the call it is accepting and release the socket after adding the call to the socket's tree. This is slightly tricky as we've dequeued the call by that point and have to requeue it. Note that requeuing emits a trace event. (7) Make rxrpc_kernel_send_data() and rxrpc_kernel_recv_data() take the new mutex immediately and don't bother with the socket mutex at all. This patch has the nice bonus that calls on the same socket are now to some extent parallelisable. Note that we might want to move rxrpc_service_prealloc() calls out from the socket lock and give it its own lock, so that we don't hang progress in other calls because we're waiting for the allocator. We probably also want to avoid calling rxrpc_notify_socket() from within the socket lock (rxrpc_accept_call()). Signed-off-by: David Howells <dhowells@redhat.com> Tested-by: Marc Dionne <marc.c.dionne@auristor.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-02-27 23:43:06 +08:00
mutex_lock(&call->user_mutex);
_debug("CALL %d USR %lx ST %d on CONN %p",
call->debug_id, call->user_call_ID, call->state, call->conn);
switch (READ_ONCE(call->state)) {
case RXRPC_CALL_CLIENT_SEND_REQUEST:
case RXRPC_CALL_SERVER_ACK_REQUEST:
case RXRPC_CALL_SERVER_SEND_REPLY:
ret = rxrpc_send_data(rxrpc_sk(sock->sk), call, msg, len,
notify_end_tx);
break;
case RXRPC_CALL_COMPLETE:
read_lock_bh(&call->state_lock);
ret = call->error;
read_unlock_bh(&call->state_lock);
break;
default:
/* Request phase complete for this client call */
trace_rxrpc_rx_eproto(call, 0, tracepoint_string("late_send"));
ret = -EPROTO;
break;
}
rxrpc: Fix deadlock between call creation and sendmsg/recvmsg All the routines by which rxrpc is accessed from the outside are serialised by means of the socket lock (sendmsg, recvmsg, bind, rxrpc_kernel_begin_call(), ...) and this presents a problem: (1) If a number of calls on the same socket are in the process of connection to the same peer, a maximum of four concurrent live calls are permitted before further calls need to wait for a slot. (2) If a call is waiting for a slot, it is deep inside sendmsg() or rxrpc_kernel_begin_call() and the entry function is holding the socket lock. (3) sendmsg() and recvmsg() or the in-kernel equivalents are prevented from servicing the other calls as they need to take the socket lock to do so. (4) The socket is stuck until a call is aborted and makes its slot available to the waiter. Fix this by: (1) Provide each call with a mutex ('user_mutex') that arbitrates access by the users of rxrpc separately for each specific call. (2) Make rxrpc_sendmsg() and rxrpc_recvmsg() unlock the socket as soon as they've got a call and taken its mutex. Note that I'm returning EWOULDBLOCK from recvmsg() if MSG_DONTWAIT is set but someone else has the lock. Should I instead only return EWOULDBLOCK if there's nothing currently to be done on a socket, and sleep in this particular instance because there is something to be done, but we appear to be blocked by the interrupt handler doing its ping? (3) Make rxrpc_new_client_call() unlock the socket after allocating a new call, locking its user mutex and adding it to the socket's call tree. The call is returned locked so that sendmsg() can add data to it immediately. From the moment the call is in the socket tree, it is subject to access by sendmsg() and recvmsg() - even if it isn't connected yet. (4) Lock new service calls in the UDP data_ready handler (in rxrpc_new_incoming_call()) because they may already be in the socket's tree and the data_ready handler makes them live immediately if a user ID has already been preassigned. Note that the new call is locked before any notifications are sent that it is live, so doing mutex_trylock() *ought* to always succeed. Userspace is prevented from doing sendmsg() on calls that are in a too-early state in rxrpc_do_sendmsg(). (5) Make rxrpc_new_incoming_call() return the call with the user mutex held so that a ping can be scheduled immediately under it. Note that it might be worth moving the ping call into rxrpc_new_incoming_call() and then we can drop the mutex there. (6) Make rxrpc_accept_call() take the lock on the call it is accepting and release the socket after adding the call to the socket's tree. This is slightly tricky as we've dequeued the call by that point and have to requeue it. Note that requeuing emits a trace event. (7) Make rxrpc_kernel_send_data() and rxrpc_kernel_recv_data() take the new mutex immediately and don't bother with the socket mutex at all. This patch has the nice bonus that calls on the same socket are now to some extent parallelisable. Note that we might want to move rxrpc_service_prealloc() calls out from the socket lock and give it its own lock, so that we don't hang progress in other calls because we're waiting for the allocator. We probably also want to avoid calling rxrpc_notify_socket() from within the socket lock (rxrpc_accept_call()). Signed-off-by: David Howells <dhowells@redhat.com> Tested-by: Marc Dionne <marc.c.dionne@auristor.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-02-27 23:43:06 +08:00
mutex_unlock(&call->user_mutex);
_leave(" = %d", ret);
return ret;
}
EXPORT_SYMBOL(rxrpc_kernel_send_data);
/**
* rxrpc_kernel_abort_call - Allow a kernel service to abort a call
* @sock: The socket the call is on
* @call: The call to be aborted
* @abort_code: The abort code to stick into the ABORT packet
* @error: Local error value
* @why: 3-char string indicating why.
*
* Allow a kernel service to abort a call, if it's still in an abortable state
* and return true if the call was aborted, false if it was already complete.
*/
bool rxrpc_kernel_abort_call(struct socket *sock, struct rxrpc_call *call,
u32 abort_code, int error, const char *why)
{
bool aborted;
_enter("{%d},%d,%d,%s", call->debug_id, abort_code, error, why);
rxrpc: Fix deadlock between call creation and sendmsg/recvmsg All the routines by which rxrpc is accessed from the outside are serialised by means of the socket lock (sendmsg, recvmsg, bind, rxrpc_kernel_begin_call(), ...) and this presents a problem: (1) If a number of calls on the same socket are in the process of connection to the same peer, a maximum of four concurrent live calls are permitted before further calls need to wait for a slot. (2) If a call is waiting for a slot, it is deep inside sendmsg() or rxrpc_kernel_begin_call() and the entry function is holding the socket lock. (3) sendmsg() and recvmsg() or the in-kernel equivalents are prevented from servicing the other calls as they need to take the socket lock to do so. (4) The socket is stuck until a call is aborted and makes its slot available to the waiter. Fix this by: (1) Provide each call with a mutex ('user_mutex') that arbitrates access by the users of rxrpc separately for each specific call. (2) Make rxrpc_sendmsg() and rxrpc_recvmsg() unlock the socket as soon as they've got a call and taken its mutex. Note that I'm returning EWOULDBLOCK from recvmsg() if MSG_DONTWAIT is set but someone else has the lock. Should I instead only return EWOULDBLOCK if there's nothing currently to be done on a socket, and sleep in this particular instance because there is something to be done, but we appear to be blocked by the interrupt handler doing its ping? (3) Make rxrpc_new_client_call() unlock the socket after allocating a new call, locking its user mutex and adding it to the socket's call tree. The call is returned locked so that sendmsg() can add data to it immediately. From the moment the call is in the socket tree, it is subject to access by sendmsg() and recvmsg() - even if it isn't connected yet. (4) Lock new service calls in the UDP data_ready handler (in rxrpc_new_incoming_call()) because they may already be in the socket's tree and the data_ready handler makes them live immediately if a user ID has already been preassigned. Note that the new call is locked before any notifications are sent that it is live, so doing mutex_trylock() *ought* to always succeed. Userspace is prevented from doing sendmsg() on calls that are in a too-early state in rxrpc_do_sendmsg(). (5) Make rxrpc_new_incoming_call() return the call with the user mutex held so that a ping can be scheduled immediately under it. Note that it might be worth moving the ping call into rxrpc_new_incoming_call() and then we can drop the mutex there. (6) Make rxrpc_accept_call() take the lock on the call it is accepting and release the socket after adding the call to the socket's tree. This is slightly tricky as we've dequeued the call by that point and have to requeue it. Note that requeuing emits a trace event. (7) Make rxrpc_kernel_send_data() and rxrpc_kernel_recv_data() take the new mutex immediately and don't bother with the socket mutex at all. This patch has the nice bonus that calls on the same socket are now to some extent parallelisable. Note that we might want to move rxrpc_service_prealloc() calls out from the socket lock and give it its own lock, so that we don't hang progress in other calls because we're waiting for the allocator. We probably also want to avoid calling rxrpc_notify_socket() from within the socket lock (rxrpc_accept_call()). Signed-off-by: David Howells <dhowells@redhat.com> Tested-by: Marc Dionne <marc.c.dionne@auristor.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-02-27 23:43:06 +08:00
mutex_lock(&call->user_mutex);
aborted = rxrpc_abort_call(why, call, 0, abort_code, error);
if (aborted)
rxrpc_send_abort_packet(call);
rxrpc: Fix deadlock between call creation and sendmsg/recvmsg All the routines by which rxrpc is accessed from the outside are serialised by means of the socket lock (sendmsg, recvmsg, bind, rxrpc_kernel_begin_call(), ...) and this presents a problem: (1) If a number of calls on the same socket are in the process of connection to the same peer, a maximum of four concurrent live calls are permitted before further calls need to wait for a slot. (2) If a call is waiting for a slot, it is deep inside sendmsg() or rxrpc_kernel_begin_call() and the entry function is holding the socket lock. (3) sendmsg() and recvmsg() or the in-kernel equivalents are prevented from servicing the other calls as they need to take the socket lock to do so. (4) The socket is stuck until a call is aborted and makes its slot available to the waiter. Fix this by: (1) Provide each call with a mutex ('user_mutex') that arbitrates access by the users of rxrpc separately for each specific call. (2) Make rxrpc_sendmsg() and rxrpc_recvmsg() unlock the socket as soon as they've got a call and taken its mutex. Note that I'm returning EWOULDBLOCK from recvmsg() if MSG_DONTWAIT is set but someone else has the lock. Should I instead only return EWOULDBLOCK if there's nothing currently to be done on a socket, and sleep in this particular instance because there is something to be done, but we appear to be blocked by the interrupt handler doing its ping? (3) Make rxrpc_new_client_call() unlock the socket after allocating a new call, locking its user mutex and adding it to the socket's call tree. The call is returned locked so that sendmsg() can add data to it immediately. From the moment the call is in the socket tree, it is subject to access by sendmsg() and recvmsg() - even if it isn't connected yet. (4) Lock new service calls in the UDP data_ready handler (in rxrpc_new_incoming_call()) because they may already be in the socket's tree and the data_ready handler makes them live immediately if a user ID has already been preassigned. Note that the new call is locked before any notifications are sent that it is live, so doing mutex_trylock() *ought* to always succeed. Userspace is prevented from doing sendmsg() on calls that are in a too-early state in rxrpc_do_sendmsg(). (5) Make rxrpc_new_incoming_call() return the call with the user mutex held so that a ping can be scheduled immediately under it. Note that it might be worth moving the ping call into rxrpc_new_incoming_call() and then we can drop the mutex there. (6) Make rxrpc_accept_call() take the lock on the call it is accepting and release the socket after adding the call to the socket's tree. This is slightly tricky as we've dequeued the call by that point and have to requeue it. Note that requeuing emits a trace event. (7) Make rxrpc_kernel_send_data() and rxrpc_kernel_recv_data() take the new mutex immediately and don't bother with the socket mutex at all. This patch has the nice bonus that calls on the same socket are now to some extent parallelisable. Note that we might want to move rxrpc_service_prealloc() calls out from the socket lock and give it its own lock, so that we don't hang progress in other calls because we're waiting for the allocator. We probably also want to avoid calling rxrpc_notify_socket() from within the socket lock (rxrpc_accept_call()). Signed-off-by: David Howells <dhowells@redhat.com> Tested-by: Marc Dionne <marc.c.dionne@auristor.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-02-27 23:43:06 +08:00
mutex_unlock(&call->user_mutex);
return aborted;
}
EXPORT_SYMBOL(rxrpc_kernel_abort_call);
/**
* rxrpc_kernel_set_tx_length - Set the total Tx length on a call
* @sock: The socket the call is on
* @call: The call to be informed
* @tx_total_len: The amount of data to be transmitted for this call
*
* Allow a kernel service to set the total transmit length on a call. This
* allows buffer-to-packet encrypt-and-copy to be performed.
*
* This function is primarily for use for setting the reply length since the
* request length can be set when beginning the call.
*/
void rxrpc_kernel_set_tx_length(struct socket *sock, struct rxrpc_call *call,
s64 tx_total_len)
{
WARN_ON(call->tx_total_len != -1);
call->tx_total_len = tx_total_len;
}
EXPORT_SYMBOL(rxrpc_kernel_set_tx_length);