brain-hackers_pepepper-mirror/linux-brain
1
0
Fork 0
mirror of https://github.com/brain-hackers/linux-brain.git synced 2024-06-09 23:36:23 +09:00
Code Issues Projects Releases Wiki Activity
248f219cb8
linux-brain/net/rxrpc/call_accept.c
David Howells 248f219cb8 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 11:10:12 +01:00

588 lines
16 KiB
C
Raw Blame History

/* incoming call handling
*
* Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
* Written by David Howells (dhowells@redhat.com)
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version
* 2 of the License, or (at your option) any later version.
*/
#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
#include <linux/module.h>
#include <linux/net.h>
#include <linux/skbuff.h>
#include <linux/errqueue.h>
#include <linux/udp.h>
#include <linux/in.h>
#include <linux/in6.h>
#include <linux/icmp.h>
#include <linux/gfp.h>
#include <linux/circ_buf.h>
#include <net/sock.h>
#include <net/af_rxrpc.h>
#include <net/ip.h>
#include "ar-internal.h"
/*
* Preallocate a single service call, connection and peer and, if possible,
* give them a user ID and attach the user's side of the ID to them.
*/
static int rxrpc_service_prealloc_one(struct rxrpc_sock *rx,
struct rxrpc_backlog *b,
rxrpc_notify_rx_t notify_rx,
rxrpc_user_attach_call_t user_attach_call,
unsigned long user_call_ID, gfp_t gfp)
{
const void *here = __builtin_return_address(0);
struct rxrpc_call *call;
int max, tmp;
unsigned int size = RXRPC_BACKLOG_MAX;
unsigned int head, tail, call_head, call_tail;
max = rx->sk.sk_max_ack_backlog;
tmp = rx->sk.sk_ack_backlog;
if (tmp >= max) {
_leave(" = -ENOBUFS [full %u]", max);
return -ENOBUFS;
}
max -= tmp;
/* We don't need more conns and peers than we have calls, but on the
* other hand, we shouldn't ever use more peers than conns or conns
* than calls.
*/
call_head = b->call_backlog_head;
call_tail = READ_ONCE(b->call_backlog_tail);
tmp = CIRC_CNT(call_head, call_tail, size);
if (tmp >= max) {
_leave(" = -ENOBUFS [enough %u]", tmp);
return -ENOBUFS;
}
max = tmp + 1;
head = b->peer_backlog_head;
tail = READ_ONCE(b->peer_backlog_tail);
if (CIRC_CNT(head, tail, size) < max) {
struct rxrpc_peer *peer = rxrpc_alloc_peer(rx->local, gfp);
if (!peer)
return -ENOMEM;
b->peer_backlog[head] = peer;
smp_store_release(&b->peer_backlog_head,
(head + 1) & (size - 1));
}
head = b->conn_backlog_head;
tail = READ_ONCE(b->conn_backlog_tail);
if (CIRC_CNT(head, tail, size) < max) {
struct rxrpc_connection *conn;
conn = rxrpc_prealloc_service_connection(gfp);
if (!conn)
return -ENOMEM;
b->conn_backlog[head] = conn;
smp_store_release(&b->conn_backlog_head,
(head + 1) & (size - 1));
}
/* Now it gets complicated, because calls get registered with the
* socket here, particularly if a user ID is preassigned by the user.
*/
call = rxrpc_alloc_call(gfp);
if (!call)
return -ENOMEM;
call->flags |= (1 << RXRPC_CALL_IS_SERVICE);
call->state = RXRPC_CALL_SERVER_PREALLOC;
trace_rxrpc_call(call, rxrpc_call_new_service,
atomic_read(&call->usage),
here, (const void *)user_call_ID);
write_lock(&rx->call_lock);
if (user_attach_call) {
struct rxrpc_call *xcall;
struct rb_node *parent, **pp;
/* Check the user ID isn't already in use */
pp = &rx->calls.rb_node;
parent = NULL;
while (*pp) {
parent = *pp;
xcall = rb_entry(parent, struct rxrpc_call, sock_node);
if (user_call_ID < call->user_call_ID)
pp = &(*pp)->rb_left;
else if (user_call_ID > call->user_call_ID)
pp = &(*pp)->rb_right;
else
goto id_in_use;
}
call->user_call_ID = user_call_ID;
call->notify_rx = notify_rx;
rxrpc_get_call(call, rxrpc_call_got);
user_attach_call(call, user_call_ID);
rxrpc_get_call(call, rxrpc_call_got_userid);
rb_link_node(&call->sock_node, parent, pp);
rb_insert_color(&call->sock_node, &rx->calls);
set_bit(RXRPC_CALL_HAS_USERID, &call->flags);
}
list_add(&call->sock_link, &rx->sock_calls);
write_unlock(&rx->call_lock);
write_lock(&rxrpc_call_lock);
list_add_tail(&call->link, &rxrpc_calls);
write_unlock(&rxrpc_call_lock);
b->call_backlog[call_head] = call;
smp_store_release(&b->call_backlog_head, (call_head + 1) & (size - 1));
_leave(" = 0 [%d -> %lx]", call->debug_id, user_call_ID);
return 0;
id_in_use:
write_unlock(&rx->call_lock);
rxrpc_cleanup_call(call);
_leave(" = -EBADSLT");
return -EBADSLT;
}
/*
* Preallocate sufficient service connections, calls and peers to cover the
* entire backlog of a socket. When a new call comes in, if we don't have
* sufficient of each available, the call gets rejected as busy or ignored.
*
* The backlog is replenished when a connection is accepted or rejected.
*/
int rxrpc_service_prealloc(struct rxrpc_sock *rx, gfp_t gfp)
{
struct rxrpc_backlog *b = rx->backlog;
if (!b) {
b = kzalloc(sizeof(struct rxrpc_backlog), gfp);
if (!b)
return -ENOMEM;
rx->backlog = b;
}
if (rx->discard_new_call)
return 0;
while (rxrpc_service_prealloc_one(rx, b, NULL, NULL, 0, gfp) == 0)
;
return 0;
}
/*
* Discard the preallocation on a service.
*/
void rxrpc_discard_prealloc(struct rxrpc_sock *rx)
{
struct rxrpc_backlog *b = rx->backlog;
unsigned int size = RXRPC_BACKLOG_MAX, head, tail;
if (!b)
return;
rx->backlog = NULL;
/* Make sure that there aren't any incoming calls in progress before we
* clear the preallocation buffers.
*/
spin_lock_bh(&rx->incoming_lock);
spin_unlock_bh(&rx->incoming_lock);
head = b->peer_backlog_head;
tail = b->peer_backlog_tail;
while (CIRC_CNT(head, tail, size) > 0) {
struct rxrpc_peer *peer = b->peer_backlog[tail];
kfree(peer);
tail = (tail + 1) & (size - 1);
}
head = b->conn_backlog_head;
tail = b->conn_backlog_tail;
while (CIRC_CNT(head, tail, size) > 0) {
struct rxrpc_connection *conn = b->conn_backlog[tail];
write_lock(&rxrpc_connection_lock);
list_del(&conn->link);
list_del(&conn->proc_link);
write_unlock(&rxrpc_connection_lock);
kfree(conn);
tail = (tail + 1) & (size - 1);
}
head = b->call_backlog_head;
tail = b->call_backlog_tail;
while (CIRC_CNT(head, tail, size) > 0) {
struct rxrpc_call *call = b->call_backlog[tail];
if (rx->discard_new_call) {
_debug("discard %lx", call->user_call_ID);
rx->discard_new_call(call, call->user_call_ID);
}
rxrpc_call_completed(call);
rxrpc_release_call(rx, call);
rxrpc_put_call(call, rxrpc_call_put);
tail = (tail + 1) & (size - 1);
}
kfree(b);
}
/*
* Allocate a new incoming call from the prealloc pool, along with a connection
* and a peer as necessary.
*/
static struct rxrpc_call *rxrpc_alloc_incoming_call(struct rxrpc_sock *rx,
struct rxrpc_local *local,
struct rxrpc_connection *conn,
struct sk_buff *skb)
{
struct rxrpc_backlog *b = rx->backlog;
struct rxrpc_peer *peer, *xpeer;
struct rxrpc_call *call;
unsigned short call_head, conn_head, peer_head;
unsigned short call_tail, conn_tail, peer_tail;
unsigned short call_count, conn_count;
/* #calls >= #conns >= #peers must hold true. */
call_head = smp_load_acquire(&b->call_backlog_head);
call_tail = b->call_backlog_tail;
call_count = CIRC_CNT(call_head, call_tail, RXRPC_BACKLOG_MAX);
conn_head = smp_load_acquire(&b->conn_backlog_head);
conn_tail = b->conn_backlog_tail;
conn_count = CIRC_CNT(conn_head, conn_tail, RXRPC_BACKLOG_MAX);
ASSERTCMP(conn_count, >=, call_count);
peer_head = smp_load_acquire(&b->peer_backlog_head);
peer_tail = b->peer_backlog_tail;
ASSERTCMP(CIRC_CNT(peer_head, peer_tail, RXRPC_BACKLOG_MAX), >=,
conn_count);
if (call_count == 0)
return NULL;
if (!conn) {
/* No connection. We're going to need a peer to start off
* with. If one doesn't yet exist, use a spare from the
* preallocation set. We dump the address into the spare in
* anticipation - and to save on stack space.
*/
xpeer = b->peer_backlog[peer_tail];
if (rxrpc_extract_addr_from_skb(&xpeer->srx, skb) < 0)
return NULL;
peer = rxrpc_lookup_incoming_peer(local, xpeer);
if (peer == xpeer) {
b->peer_backlog[peer_tail] = NULL;
smp_store_release(&b->peer_backlog_tail,
(peer_tail + 1) &
(RXRPC_BACKLOG_MAX - 1));
}
/* Now allocate and set up the connection */
conn = b->conn_backlog[conn_tail];
b->conn_backlog[conn_tail] = NULL;
smp_store_release(&b->conn_backlog_tail,
(conn_tail + 1) & (RXRPC_BACKLOG_MAX - 1));
rxrpc_get_local(local);
conn->params.local = local;
conn->params.peer = peer;
rxrpc_new_incoming_connection(conn, skb);
} else {
rxrpc_get_connection(conn);
}
/* And now we can allocate and set up a new call */
call = b->call_backlog[call_tail];
b->call_backlog[call_tail] = NULL;
smp_store_release(&b->call_backlog_tail,
(call_tail + 1) & (RXRPC_BACKLOG_MAX - 1));
call->conn = conn;
call->peer = rxrpc_get_peer(conn->params.peer);
return call;
}
/*
* Set up a new incoming call. Called in BH context with the RCU read lock
* held.
*
* If this is for a kernel service, when we allocate the call, it will have
* three refs on it: (1) the kernel service, (2) the user_call_ID tree, (3) the
* retainer ref obtained from the backlog buffer. Prealloc calls for userspace
* services only have the ref from the backlog buffer. We want to pass this
* ref to non-BH context to dispose of.
*
* If we want to report an error, we mark the skb with the packet type and
* abort code and return NULL.
*/
struct rxrpc_call *rxrpc_new_incoming_call(struct rxrpc_local *local,
struct rxrpc_connection *conn,
struct sk_buff *skb)
{
struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
struct rxrpc_sock *rx;
struct rxrpc_call *call;
_enter("");
/* Get the socket providing the service */
hlist_for_each_entry_rcu_bh(rx, &local->services, listen_link) {
if (rx->srx.srx_service == sp->hdr.serviceId)
goto found_service;
}
trace_rxrpc_abort("INV", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
RX_INVALID_OPERATION, EOPNOTSUPP);
skb->mark = RXRPC_SKB_MARK_LOCAL_ABORT;
skb->priority = RX_INVALID_OPERATION;
_leave(" = NULL [service]");
return NULL;
found_service:
spin_lock(&rx->incoming_lock);
if (rx->sk.sk_state == RXRPC_CLOSE) {
trace_rxrpc_abort("CLS", sp->hdr.cid, sp->hdr.callNumber,
sp->hdr.seq, RX_INVALID_OPERATION, ESHUTDOWN);
skb->mark = RXRPC_SKB_MARK_LOCAL_ABORT;
skb->priority = RX_INVALID_OPERATION;
_leave(" = NULL [close]");
call = NULL;
goto out;
}
call = rxrpc_alloc_incoming_call(rx, local, conn, skb);
if (!call) {
skb->mark = RXRPC_SKB_MARK_BUSY;
_leave(" = NULL [busy]");
call = NULL;
goto out;
}
/* Make the call live. */
rxrpc_incoming_call(rx, call, skb);
conn = call->conn;
if (rx->notify_new_call)
rx->notify_new_call(&rx->sk, call, call->user_call_ID);
spin_lock(&conn->state_lock);
switch (conn->state) {
case RXRPC_CONN_SERVICE_UNSECURED:
conn->state = RXRPC_CONN_SERVICE_CHALLENGING;
set_bit(RXRPC_CONN_EV_CHALLENGE, &call->conn->events);
rxrpc_queue_conn(call->conn);
break;
case RXRPC_CONN_SERVICE:
write_lock(&call->state_lock);
if (rx->discard_new_call)
call->state = RXRPC_CALL_SERVER_RECV_REQUEST;
else
call->state = RXRPC_CALL_SERVER_ACCEPTING;
write_unlock(&call->state_lock);
break;
case RXRPC_CONN_REMOTELY_ABORTED:
rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
conn->remote_abort, ECONNABORTED);
break;
case RXRPC_CONN_LOCALLY_ABORTED:
rxrpc_abort_call("CON", call, sp->hdr.seq,
conn->local_abort, ECONNABORTED);
break;
default:
BUG();
}
spin_unlock(&conn->state_lock);
if (call->state == RXRPC_CALL_SERVER_ACCEPTING)
rxrpc_notify_socket(call);
_leave(" = %p{%d}", call, call->debug_id);
out:
spin_unlock(&rx->incoming_lock);
return call;
}
/*
* handle acceptance of a call by userspace
* - assign the user call ID to the call at the front of the queue
*/
struct rxrpc_call *rxrpc_accept_call(struct rxrpc_sock *rx,
unsigned long user_call_ID,
rxrpc_notify_rx_t notify_rx)
{
struct rxrpc_call *call;
struct rb_node *parent, **pp;
int ret;
_enter(",%lx", user_call_ID);
ASSERT(!irqs_disabled());
write_lock(&rx->call_lock);
ret = -ENODATA;
if (list_empty(&rx->to_be_accepted))
goto out;
/* check the user ID isn't already in use */
pp = &rx->calls.rb_node;
parent = NULL;
while (*pp) {
parent = *pp;
call = rb_entry(parent, struct rxrpc_call, sock_node);
if (user_call_ID < call->user_call_ID)
pp = &(*pp)->rb_left;
else if (user_call_ID > call->user_call_ID)
pp = &(*pp)->rb_right;
else
goto id_in_use;
}
/* Dequeue the first call and check it's still valid. We gain
* responsibility for the queue's reference.
*/
call = list_entry(rx->to_be_accepted.next,
struct rxrpc_call, accept_link);
list_del_init(&call->accept_link);
sk_acceptq_removed(&rx->sk);
rxrpc_see_call(call);
write_lock_bh(&call->state_lock);
switch (call->state) {
case RXRPC_CALL_SERVER_ACCEPTING:
call->state = RXRPC_CALL_SERVER_RECV_REQUEST;
break;
case RXRPC_CALL_COMPLETE:
ret = call->error;
goto out_release;
default:
BUG();
}
/* formalise the acceptance */
rxrpc_get_call(call, rxrpc_call_got);
call->notify_rx = notify_rx;
call->user_call_ID = user_call_ID;
rxrpc_get_call(call, rxrpc_call_got_userid);
rb_link_node(&call->sock_node, parent, pp);
rb_insert_color(&call->sock_node, &rx->calls);
if (test_and_set_bit(RXRPC_CALL_HAS_USERID, &call->flags))
BUG();
write_unlock_bh(&call->state_lock);
write_unlock(&rx->call_lock);
rxrpc_notify_socket(call);
rxrpc_service_prealloc(rx, GFP_KERNEL);
_leave(" = %p{%d}", call, call->debug_id);
return call;
out_release:
_debug("release %p", call);
write_unlock_bh(&call->state_lock);
write_unlock(&rx->call_lock);
rxrpc_release_call(rx, call);
rxrpc_put_call(call, rxrpc_call_put);
goto out;
id_in_use:
ret = -EBADSLT;
write_unlock(&rx->call_lock);
out:
rxrpc_service_prealloc(rx, GFP_KERNEL);
_leave(" = %d", ret);
return ERR_PTR(ret);
}
/*
* Handle rejection of a call by userspace
* - reject the call at the front of the queue
*/
int rxrpc_reject_call(struct rxrpc_sock *rx)
{
struct rxrpc_call *call;
bool abort = false;
int ret;
_enter("");
ASSERT(!irqs_disabled());
write_lock(&rx->call_lock);
if (list_empty(&rx->to_be_accepted)) {
write_unlock(&rx->call_lock);
return -ENODATA;
}
/* Dequeue the first call and check it's still valid. We gain
* responsibility for the queue's reference.
*/
call = list_entry(rx->to_be_accepted.next,
struct rxrpc_call, accept_link);
list_del_init(&call->accept_link);
sk_acceptq_removed(&rx->sk);
rxrpc_see_call(call);
write_lock_bh(&call->state_lock);
switch (call->state) {
case RXRPC_CALL_SERVER_ACCEPTING:
__rxrpc_abort_call("REJ", call, 1, RX_USER_ABORT, ECONNABORTED);
abort = true;
/* fall through */
case RXRPC_CALL_COMPLETE:
ret = call->error;
goto out_discard;
default:
BUG();
}
out_discard:
write_unlock_bh(&call->state_lock);
write_unlock(&rx->call_lock);
if (abort) {
rxrpc_send_call_packet(call, RXRPC_PACKET_TYPE_ABORT);
rxrpc_release_call(rx, call);
rxrpc_put_call(call, rxrpc_call_put);
}
rxrpc_service_prealloc(rx, GFP_KERNEL);
_leave(" = %d", ret);
return ret;
}
/*
* rxrpc_kernel_charge_accept - Charge up socket with preallocated calls
* @sock: The socket on which to preallocate
* @notify_rx: Event notification function for the call
* @user_attach_call: Func to attach call to user_call_ID
* @user_call_ID: The tag to attach to the preallocated call
* @gfp: The allocation conditions.
*
* Charge up the socket with preallocated calls, each with a user ID. A
* function should be provided to effect the attachment from the user's side.
* The user is given a ref to hold on the call.
*
* Note that the call may be come connected before this function returns.
*/
int rxrpc_kernel_charge_accept(struct socket *sock,
rxrpc_notify_rx_t notify_rx,
rxrpc_user_attach_call_t user_attach_call,
unsigned long user_call_ID, gfp_t gfp)
{
struct rxrpc_sock *rx = rxrpc_sk(sock->sk);
struct rxrpc_backlog *b = rx->backlog;
if (sock->sk->sk_state == RXRPC_CLOSE)
return -ESHUTDOWN;
return rxrpc_service_prealloc_one(rx, b, notify_rx,
user_attach_call, user_call_ID,
gfp);
}
EXPORT_SYMBOL(rxrpc_kernel_charge_accept);
Reference in New Issue View Git Blame Copy Permalink