X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lnet%2Flnet%2Facceptor.c;h=f287ae6492c9c31aaf55a121a4fd26d18d805aff;hb=48a9ea82eb30bbbf66cce527c1205d13fbd4eb58;hp=9db3571ce579cf48c0dc29544ae2b40903af8c2f;hpb=08aa217ce49aba1ded52e0f7adb8a607035123fd;p=fs%2Flustre-release.git diff --git a/lnet/lnet/acceptor.c b/lnet/lnet/acceptor.c index 9db3571..f287ae6 100644 --- a/lnet/lnet/acceptor.c +++ b/lnet/lnet/acceptor.c @@ -15,11 +15,7 @@ * * You should have received a copy of the GNU General Public License * version 2 along with this program; If not, see - * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf - * - * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, - * CA 95054 USA or visit www.sun.com if you need additional information or - * have any questions. + * http://www.gnu.org/licenses/gpl-2.0.html * * GPL HEADER END */ @@ -27,567 +23,510 @@ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2011, 2012, Intel Corporation. + * Copyright (c) 2011, 2017, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ - * Lustre is a trademark of Sun Microsystems, Inc. */ #define DEBUG_SUBSYSTEM S_LNET -#include -#if defined(__KERNEL__) || defined(HAVE_LIBPTHREAD) +#include +#include +#include +#include static int accept_port = 988; static int accept_backlog = 127; static int accept_timeout = 5; -struct { - int pta_shutdown; - cfs_socket_t *pta_sock; -#ifdef __KERNEL__ +static struct { + int pta_shutdown; + struct socket *pta_sock; struct completion pta_signal; + struct net *pta_ns; + wait_queue_head_t pta_waitq; + atomic_t pta_ready; +#ifdef HAVE_SK_DATA_READY_ONE_ARG + void (*pta_odata)(struct sock *); #else - mt_completion_t pta_signal; + void (*pta_odata)(struct sock *, int); #endif -} lnet_acceptor_state; +} lnet_acceptor_state = { + .pta_shutdown = 1 +}; int lnet_acceptor_port(void) { - return accept_port; + return accept_port; } static inline int lnet_accept_magic(__u32 magic, __u32 constant) { - return (magic == constant || - magic == __swab32(constant)); + return (magic == constant || + magic == __swab32(constant)); } -#ifdef __KERNEL__ - -#define mt_init_completion(c) init_completion(c) -#define mt_wait_for_completion(c) wait_for_completion(c) -#define mt_complete(c) complete(c) -#define mt_fini_completion(c) fini_completion(c) - EXPORT_SYMBOL(lnet_acceptor_port); -static char *accept = "secure"; - -CFS_MODULE_PARM(accept, "s", charp, 0444, - "Accept connections (secure|all|none)"); -CFS_MODULE_PARM(accept_port, "i", int, 0444, - "Acceptor's port (same on all nodes)"); -CFS_MODULE_PARM(accept_backlog, "i", int, 0444, - "Acceptor's listen backlog"); -CFS_MODULE_PARM(accept_timeout, "i", int, 0644, - "Acceptor's timeout (seconds)"); - -static char *accept_type = NULL; +static char *accept_type = "secure"; -int -lnet_acceptor_get_tunables(void) -{ - /* Userland acceptor uses 'accept_type' instead of 'accept', due to - * conflict with 'accept(2)', but kernel acceptor still uses 'accept' - * for compatibility. Hence the trick. */ - accept_type = accept; - return 0; -} +module_param_named(accept, accept_type, charp, 0444); +MODULE_PARM_DESC(accept, "Accept connections (secure|all|none)"); +module_param(accept_port, int, 0444); +MODULE_PARM_DESC(accept_port, "Acceptor's port (same on all nodes)"); +module_param(accept_backlog, int, 0444); +MODULE_PARM_DESC(accept_backlog, "Acceptor's listen backlog"); +module_param(accept_timeout, int, 0644); +MODULE_PARM_DESC(accept_timeout, "Acceptor's timeout (seconds)"); int lnet_acceptor_timeout(void) { - return accept_timeout; + return accept_timeout; } EXPORT_SYMBOL(lnet_acceptor_timeout); void lnet_connect_console_error (int rc, lnet_nid_t peer_nid, - __u32 peer_ip, int peer_port) + struct sockaddr *sa) { - switch (rc) { - /* "normal" errors */ - case -ECONNREFUSED: - CNETERR("Connection to %s at host %u.%u.%u.%u on port %d was " - "refused: check that Lustre is running on that node.\n", - libcfs_nid2str(peer_nid), - HIPQUAD(peer_ip), peer_port); - break; - case -EHOSTUNREACH: - case -ENETUNREACH: - CNETERR("Connection to %s at host %u.%u.%u.%u " - "was unreachable: the network or that node may " - "be down, or Lustre may be misconfigured.\n", - libcfs_nid2str(peer_nid), HIPQUAD(peer_ip)); - break; - case -ETIMEDOUT: - CNETERR("Connection to %s at host %u.%u.%u.%u on " - "port %d took too long: that node may be hung " - "or experiencing high load.\n", - libcfs_nid2str(peer_nid), - HIPQUAD(peer_ip), peer_port); - break; - case -ECONNRESET: - LCONSOLE_ERROR_MSG(0x11b, "Connection to %s at host %u.%u.%u.%u" - " on port %d was reset: " - "is it running a compatible version of " - "Lustre and is %s one of its NIDs?\n", - libcfs_nid2str(peer_nid), - HIPQUAD(peer_ip), peer_port, - libcfs_nid2str(peer_nid)); - break; - case -EPROTO: - LCONSOLE_ERROR_MSG(0x11c, "Protocol error connecting to %s at " - "host %u.%u.%u.%u on port %d: is it running " - "a compatible version of Lustre?\n", - libcfs_nid2str(peer_nid), - HIPQUAD(peer_ip), peer_port); - break; - case -EADDRINUSE: - LCONSOLE_ERROR_MSG(0x11d, "No privileged ports available to " - "connect to %s at host %u.%u.%u.%u on port " - "%d\n", libcfs_nid2str(peer_nid), - HIPQUAD(peer_ip), peer_port); - break; - default: - LCONSOLE_ERROR_MSG(0x11e, "Unexpected error %d connecting to %s" - " at host %u.%u.%u.%u on port %d\n", rc, - libcfs_nid2str(peer_nid), - HIPQUAD(peer_ip), peer_port); - break; - } + switch (rc) { + /* "normal" errors */ + case -ECONNREFUSED: + CNETERR("Connection to %s at host %pISp was refused: check that Lustre is running on that node.\n", + libcfs_nid2str(peer_nid), sa); + break; + case -EHOSTUNREACH: + case -ENETUNREACH: + CNETERR("Connection to %s at host %pIS was unreachable: the network or that node may be down, or Lustre may be misconfigured.\n", + libcfs_nid2str(peer_nid), sa); + break; + case -ETIMEDOUT: + CNETERR("Connection to %s at host %pISp took too long: that node may be hung or experiencing high load.\n", + libcfs_nid2str(peer_nid), sa); + break; + case -ECONNRESET: + LCONSOLE_ERROR_MSG(0x11b, + "Connection to %s at host %pISp was reset: is it running a compatible version of Lustre and is %s one of its NIDs?\n", + libcfs_nid2str(peer_nid), sa, + libcfs_nid2str(peer_nid)); + break; + case -EPROTO: + LCONSOLE_ERROR_MSG(0x11c, + "Protocol error connecting to %s at host %pISp: is it running a compatible version of Lustre?\n", + libcfs_nid2str(peer_nid), sa); + break; + case -EADDRINUSE: + LCONSOLE_ERROR_MSG(0x11d, + "No privileged ports available to connect to %s at host %pISp\n", + libcfs_nid2str(peer_nid), sa); + break; + default: + LCONSOLE_ERROR_MSG(0x11e, + "Unexpected error %d connecting to %s at host %pISp\n", + rc, libcfs_nid2str(peer_nid), sa); + break; + } } EXPORT_SYMBOL(lnet_connect_console_error); -int -lnet_connect(cfs_socket_t **sockp, lnet_nid_t peer_nid, - __u32 local_ip, __u32 peer_ip, int peer_port) +struct socket * +lnet_connect(lnet_nid_t peer_nid, int interface, struct sockaddr *peeraddr, + struct net *ns) { - lnet_acceptor_connreq_t cr; - cfs_socket_t *sock; - int rc; - int port; - int fatal; - - CLASSERT (sizeof(cr) <= 16); /* not too big to be on the stack */ - - for (port = LNET_ACCEPTOR_MAX_RESERVED_PORT; - port >= LNET_ACCEPTOR_MIN_RESERVED_PORT; - --port) { - /* Iterate through reserved ports. */ - - rc = libcfs_sock_connect(&sock, &fatal, - local_ip, port, - peer_ip, peer_port); - if (rc != 0) { - if (fatal) - goto failed; - continue; - } - - CLASSERT (LNET_PROTO_ACCEPTOR_VERSION == 1); - - cr.acr_magic = LNET_PROTO_ACCEPTOR_MAGIC; - cr.acr_version = LNET_PROTO_ACCEPTOR_VERSION; - cr.acr_nid = peer_nid; - - if (the_lnet.ln_testprotocompat != 0) { - /* single-shot proto check */ - lnet_net_lock(LNET_LOCK_EX); - if ((the_lnet.ln_testprotocompat & 4) != 0) { - cr.acr_version++; - the_lnet.ln_testprotocompat &= ~4; - } - if ((the_lnet.ln_testprotocompat & 8) != 0) { - cr.acr_magic = LNET_PROTO_MAGIC; - the_lnet.ln_testprotocompat &= ~8; - } - lnet_net_unlock(LNET_LOCK_EX); - } - - rc = libcfs_sock_write(sock, &cr, sizeof(cr), - accept_timeout); - if (rc != 0) - goto failed_sock; - - *sockp = sock; - return 0; - } - - rc = -EADDRINUSE; - goto failed; - - failed_sock: - libcfs_sock_release(sock); - failed: - lnet_connect_console_error(rc, peer_nid, peer_ip, peer_port); - return rc; + struct lnet_acceptor_connreq cr; + struct socket *sock; + int rc; + int port; + + BUILD_BUG_ON(sizeof(cr) > 16); /* not too big to be on the stack */ + + LASSERT(peeraddr->sa_family == AF_INET || + peeraddr->sa_family == AF_INET6); + + for (port = LNET_ACCEPTOR_MAX_RESERVED_PORT; + port >= LNET_ACCEPTOR_MIN_RESERVED_PORT; + --port) { + /* Iterate through reserved ports. */ + sock = lnet_sock_connect(interface, port, peeraddr, ns); + if (IS_ERR(sock)) { + rc = PTR_ERR(sock); + if (rc == -EADDRINUSE || rc == -EADDRNOTAVAIL) + continue; + goto failed; + } + + BUILD_BUG_ON(LNET_PROTO_ACCEPTOR_VERSION != 1); + + cr.acr_magic = LNET_PROTO_ACCEPTOR_MAGIC; + cr.acr_version = LNET_PROTO_ACCEPTOR_VERSION; + cr.acr_nid = peer_nid; + + if (the_lnet.ln_testprotocompat) { + /* single-shot proto check */ + if (test_and_clear_bit(2, &the_lnet.ln_testprotocompat)) + cr.acr_version++; + if (test_and_clear_bit(3, &the_lnet.ln_testprotocompat)) + cr.acr_magic = LNET_PROTO_MAGIC; + } + + rc = lnet_sock_write(sock, &cr, sizeof(cr), + accept_timeout); + if (rc != 0) + goto failed_sock; + + return sock; + } + + rc = -EADDRINUSE; + goto failed; + +failed_sock: + sock_release(sock); +failed: + lnet_connect_console_error(rc, peer_nid, peeraddr); + return ERR_PTR(rc); } EXPORT_SYMBOL(lnet_connect); -#else /* below is multi-threaded user-space code */ - -static char *accept_type = "secure"; - -int -lnet_acceptor_get_tunables() +static int +lnet_accept(struct socket *sock, __u32 magic) { - int rc; - char *env = getenv("LNET_ACCEPT"); - - if (env != NULL) - accept_type = env; - - rc = lnet_parse_int_tunable(&accept_port, "LNET_ACCEPT_PORT"); - - if (rc != 0) - return rc; - - rc = lnet_parse_int_tunable(&accept_backlog, "LNET_ACCEPT_BACKLOG"); - - if (rc != 0) - return rc; - - rc = lnet_parse_int_tunable(&accept_timeout, "LNET_ACCEPT_TIMEOUT"); - - if (rc != 0) - return rc; - - CDEBUG(D_NET, "accept_type = %s\n", accept_type); - CDEBUG(D_NET, "accept_port = %d\n", accept_port); - CDEBUG(D_NET, "accept_backlog = %d\n", accept_backlog); - CDEBUG(D_NET, "accept_timeout = %d\n", accept_timeout); - return 0; + struct lnet_acceptor_connreq cr; + struct sockaddr_storage peer; + int rc; + int flip; + struct lnet_ni *ni; + char *str; + + LASSERT(sizeof(cr) <= 16); /* not too big for the stack */ + + rc = lnet_sock_getaddr(sock, true, &peer); + if (rc != 0) { + CERROR("Can't determine new connection's address\n"); + return rc; + } + + if (!lnet_accept_magic(magic, LNET_PROTO_ACCEPTOR_MAGIC)) { + + if (lnet_accept_magic(magic, LNET_PROTO_MAGIC)) { + /* future version compatibility! + * When LNET unifies protocols over all LNDs, the first + * thing sent will be a version query. I send back + * LNET_PROTO_ACCEPTOR_MAGIC to tell her I'm "old" */ + + memset(&cr, 0, sizeof(cr)); + cr.acr_magic = LNET_PROTO_ACCEPTOR_MAGIC; + cr.acr_version = LNET_PROTO_ACCEPTOR_VERSION; + rc = lnet_sock_write(sock, &cr, sizeof(cr), + accept_timeout); + + if (rc != 0) + CERROR("Error sending magic+version in response to LNET magic from %pIS: %d\n", + &peer, rc); + return -EPROTO; + } + + if (lnet_accept_magic(magic, LNET_PROTO_TCP_MAGIC)) + str = "'old' socknal/tcpnal"; + else + str = "unrecognised"; + + LCONSOLE_ERROR_MSG(0x11f, "Refusing connection from %pIS" + " magic %08x: %s acceptor protocol\n", + &peer, magic, str); + return -EPROTO; + } + + flip = (magic != LNET_PROTO_ACCEPTOR_MAGIC); + + rc = lnet_sock_read(sock, &cr.acr_version, + sizeof(cr.acr_version), + accept_timeout); + if (rc != 0) { + CERROR("Error %d reading connection request version from %pIS\n", + rc, &peer); + return -EIO; + } + + if (flip) + __swab32s(&cr.acr_version); + + if (cr.acr_version != LNET_PROTO_ACCEPTOR_VERSION) { + /* future version compatibility! + * An acceptor-specific protocol rev will first send a version + * query. I send back my current version to tell her I'm + * "old". */ + int peer_version = cr.acr_version; + + memset(&cr, 0, sizeof(cr)); + cr.acr_magic = LNET_PROTO_ACCEPTOR_MAGIC; + cr.acr_version = LNET_PROTO_ACCEPTOR_VERSION; + + rc = lnet_sock_write(sock, &cr, sizeof(cr), + accept_timeout); + + if (rc != 0) + CERROR("Error sending magic+version in response to version %d from %pIS: %d\n", + peer_version, &peer, rc); + return -EPROTO; + } + + rc = lnet_sock_read(sock, &cr.acr_nid, + sizeof(cr) - + offsetof(struct lnet_acceptor_connreq, acr_nid), + accept_timeout); + if (rc != 0) { + CERROR("Error %d reading connection request from %pIS\n", + rc, &peer); + return -EIO; + } + + if (flip) + __swab64s(&cr.acr_nid); + + ni = lnet_nid2ni_addref(cr.acr_nid); + if (ni == NULL || /* no matching net */ + ni->ni_nid != cr.acr_nid) { /* right NET, wrong NID! */ + if (ni != NULL) + lnet_ni_decref(ni); + LCONSOLE_ERROR_MSG(0x120, + "Refusing connection from %pIS for %s: No matching NI\n", + &peer, libcfs_nid2str(cr.acr_nid)); + return -EPERM; + } + + if (ni->ni_net->net_lnd->lnd_accept == NULL) { + /* This catches a request for the loopback LND */ + lnet_ni_decref(ni); + LCONSOLE_ERROR_MSG(0x121, + "Refusing connection from %pIS for %s: NI doesn not accept IP connections\n", + &peer, libcfs_nid2str(cr.acr_nid)); + return -EPERM; + } + + CDEBUG(D_NET, "Accept %s from %pI4h\n", + libcfs_nid2str(cr.acr_nid), &peer); + + rc = ni->ni_net->net_lnd->lnd_accept(ni, sock); + + lnet_ni_decref(ni); + return rc; } -#endif /* __KERNEL__ */ - -/* Below is the code common for both kernel and MT user-space */ - -int -lnet_accept(cfs_socket_t *sock, __u32 magic) +#ifdef HAVE_SK_DATA_READY_ONE_ARG +static void lnet_acceptor_ready(struct sock *sk) +#else +static void lnet_acceptor_ready(struct sock *sk, int len) +#endif { - lnet_acceptor_connreq_t cr; - __u32 peer_ip; - int peer_port; - int rc; - int flip; - lnet_ni_t *ni; - char *str; - - LASSERT (sizeof(cr) <= 16); /* not too big for the stack */ - - rc = libcfs_sock_getaddr(sock, 1, &peer_ip, &peer_port); - LASSERT (rc == 0); /* we succeeded before */ - - if (!lnet_accept_magic(magic, LNET_PROTO_ACCEPTOR_MAGIC)) { - - if (lnet_accept_magic(magic, LNET_PROTO_MAGIC)) { - /* future version compatibility! - * When LNET unifies protocols over all LNDs, the first - * thing sent will be a version query. I send back - * LNET_PROTO_ACCEPTOR_MAGIC to tell her I'm "old" */ - - memset (&cr, 0, sizeof(cr)); - cr.acr_magic = LNET_PROTO_ACCEPTOR_MAGIC; - cr.acr_version = LNET_PROTO_ACCEPTOR_VERSION; - rc = libcfs_sock_write(sock, &cr, sizeof(cr), - accept_timeout); - - if (rc != 0) - CERROR("Error sending magic+version in response" - "to LNET magic from %u.%u.%u.%u: %d\n", - HIPQUAD(peer_ip), rc); - return -EPROTO; - } - - if (magic == le32_to_cpu(LNET_PROTO_TCP_MAGIC)) - str = "'old' socknal/tcpnal"; - else if (lnet_accept_magic(magic, LNET_PROTO_RA_MAGIC)) - str = "'old' ranal"; - else - str = "unrecognised"; - - LCONSOLE_ERROR_MSG(0x11f, "Refusing connection from %u.%u.%u.%u" - " magic %08x: %s acceptor protocol\n", - HIPQUAD(peer_ip), magic, str); - return -EPROTO; - } - - flip = (magic != LNET_PROTO_ACCEPTOR_MAGIC); - - rc = libcfs_sock_read(sock, &cr.acr_version, - sizeof(cr.acr_version), - accept_timeout); - if (rc != 0) { - CERROR("Error %d reading connection request version from " - "%u.%u.%u.%u\n", rc, HIPQUAD(peer_ip)); - return -EIO; - } - - if (flip) - __swab32s(&cr.acr_version); - - if (cr.acr_version != LNET_PROTO_ACCEPTOR_VERSION) { - /* future version compatibility! - * An acceptor-specific protocol rev will first send a version - * query. I send back my current version to tell her I'm - * "old". */ - int peer_version = cr.acr_version; - - memset (&cr, 0, sizeof(cr)); - cr.acr_magic = LNET_PROTO_ACCEPTOR_MAGIC; - cr.acr_version = LNET_PROTO_ACCEPTOR_VERSION; - - rc = libcfs_sock_write(sock, &cr, sizeof(cr), - accept_timeout); - - if (rc != 0) - CERROR("Error sending magic+version in response" - "to version %d from %u.%u.%u.%u: %d\n", - peer_version, HIPQUAD(peer_ip), rc); - return -EPROTO; - } - - rc = libcfs_sock_read(sock, &cr.acr_nid, - sizeof(cr) - - offsetof(lnet_acceptor_connreq_t, acr_nid), - accept_timeout); - if (rc != 0) { - CERROR("Error %d reading connection request from " - "%u.%u.%u.%u\n", rc, HIPQUAD(peer_ip)); - return -EIO; - } - - if (flip) - __swab64s(&cr.acr_nid); - - ni = lnet_net2ni(LNET_NIDNET(cr.acr_nid)); - if (ni == NULL || /* no matching net */ - ni->ni_nid != cr.acr_nid) { /* right NET, wrong NID! */ - if (ni != NULL) - lnet_ni_decref(ni); - LCONSOLE_ERROR_MSG(0x120, "Refusing connection from %u.%u.%u.%u" - " for %s: No matching NI\n", - HIPQUAD(peer_ip), libcfs_nid2str(cr.acr_nid)); - return -EPERM; - } - - if (ni->ni_lnd->lnd_accept == NULL) { - /* This catches a request for the loopback LND */ - lnet_ni_decref(ni); - LCONSOLE_ERROR_MSG(0x121, "Refusing connection from %u.%u.%u.%u" - " for %s: NI doesn not accept IP connections\n", - HIPQUAD(peer_ip), libcfs_nid2str(cr.acr_nid)); - return -EPERM; - } - - CDEBUG(D_NET, "Accept %s from %u.%u.%u.%u\n", - libcfs_nid2str(cr.acr_nid), HIPQUAD(peer_ip)); - - rc = ni->ni_lnd->lnd_accept(ni, sock); - - lnet_ni_decref(ni); - return rc; + /* Ensure pta_odata has actually been set before calling it */ + rmb(); +#ifdef HAVE_SK_DATA_READY_ONE_ARG + lnet_acceptor_state.pta_odata(sk); +#else + lnet_acceptor_state.pta_odata(sk, 0); +#endif + + atomic_set(&lnet_acceptor_state.pta_ready, 1); + wake_up(&lnet_acceptor_state.pta_waitq); } -int +static int lnet_acceptor(void *arg) { - char name[16]; - cfs_socket_t *newsock; - int rc; - __u32 magic; - __u32 peer_ip; - int peer_port; - int secure = (int)((long_ptr_t)arg); - - LASSERT (lnet_acceptor_state.pta_sock == NULL); - - snprintf(name, sizeof(name), "acceptor_%03d", accept_port); - cfs_daemonize(name); - cfs_block_allsigs(); - - rc = libcfs_sock_listen(&lnet_acceptor_state.pta_sock, - 0, accept_port, accept_backlog); - if (rc != 0) { - if (rc == -EADDRINUSE) - LCONSOLE_ERROR_MSG(0x122, "Can't start acceptor on port" - " %d: port already in use\n", - accept_port); - else - LCONSOLE_ERROR_MSG(0x123, "Can't start acceptor on port " - "%d: unexpected error %d\n", - accept_port, rc); - - lnet_acceptor_state.pta_sock = NULL; - } else { - LCONSOLE(0, "Accept %s, port %d\n", accept_type, accept_port); - } - - /* set init status and unblock parent */ - lnet_acceptor_state.pta_shutdown = rc; - mt_complete(&lnet_acceptor_state.pta_signal); - - if (rc != 0) - return rc; - - while (!lnet_acceptor_state.pta_shutdown) { - - rc = libcfs_sock_accept(&newsock, lnet_acceptor_state.pta_sock); - if (rc != 0) { - if (rc != -EAGAIN) { - CWARN("Accept error %d: pausing...\n", rc); - cfs_pause(cfs_time_seconds(1)); - } - continue; - } - - /* maybe we're waken up with libcfs_sock_abort_accept() */ - if (lnet_acceptor_state.pta_shutdown) { - libcfs_sock_release(newsock); - break; - } - - rc = libcfs_sock_getaddr(newsock, 1, &peer_ip, &peer_port); - if (rc != 0) { - CERROR("Can't determine new connection's address\n"); - goto failed; - } - - if (secure && peer_port > LNET_ACCEPTOR_MAX_RESERVED_PORT) { - CERROR("Refusing connection from %u.%u.%u.%u: " - "insecure port %d\n", - HIPQUAD(peer_ip), peer_port); - goto failed; - } - - rc = libcfs_sock_read(newsock, &magic, sizeof(magic), - accept_timeout); - if (rc != 0) { - CERROR("Error %d reading connection request from " - "%u.%u.%u.%u\n", rc, HIPQUAD(peer_ip)); - goto failed; - } - - rc = lnet_accept(newsock, magic); - if (rc != 0) - goto failed; - - continue; - - failed: - libcfs_sock_release(newsock); - } - - libcfs_sock_release(lnet_acceptor_state.pta_sock); - lnet_acceptor_state.pta_sock = NULL; - - CDEBUG(D_NET, "Acceptor stopping\n"); - - /* unblock lnet_acceptor_stop() */ - mt_complete(&lnet_acceptor_state.pta_signal); - return 0; + struct socket *newsock; + int rc; + __u32 magic; + struct sockaddr_storage peer; + int secure = (int)((uintptr_t)arg); + + LASSERT(lnet_acceptor_state.pta_sock == NULL); + + lnet_acceptor_state.pta_sock = + lnet_sock_listen(accept_port, accept_backlog, + lnet_acceptor_state.pta_ns); + if (IS_ERR(lnet_acceptor_state.pta_sock)) { + rc = PTR_ERR(lnet_acceptor_state.pta_sock); + if (rc == -EADDRINUSE) + LCONSOLE_ERROR_MSG(0x122, "Can't start acceptor on port" + " %d: port already in use\n", + accept_port); + else + LCONSOLE_ERROR_MSG(0x123, "Can't start acceptor on port " + "%d: unexpected error %d\n", + accept_port, rc); + + lnet_acceptor_state.pta_sock = NULL; + } else { + rc = 0; + LCONSOLE(0, "Accept %s, port %d\n", accept_type, accept_port); + init_waitqueue_head(&lnet_acceptor_state.pta_waitq); + lnet_acceptor_state.pta_odata = + lnet_acceptor_state.pta_sock->sk->sk_data_ready; + /* ensure pta_odata gets set before there is any chance of + * lnet_accept_ready() trying to read it. + */ + wmb(); + lnet_acceptor_state.pta_sock->sk->sk_data_ready = + lnet_acceptor_ready; + atomic_set(&lnet_acceptor_state.pta_ready, 1); + } + + /* set init status and unblock parent */ + lnet_acceptor_state.pta_shutdown = rc; + complete(&lnet_acceptor_state.pta_signal); + + if (rc != 0) + return rc; + + while (!lnet_acceptor_state.pta_shutdown) { + + wait_event_idle(lnet_acceptor_state.pta_waitq, + lnet_acceptor_state.pta_shutdown || + atomic_read(&lnet_acceptor_state.pta_ready)); + if (!atomic_read(&lnet_acceptor_state.pta_ready)) + continue; + atomic_set(&lnet_acceptor_state.pta_ready, 0); + rc = kernel_accept(lnet_acceptor_state.pta_sock, &newsock, + SOCK_NONBLOCK); + if (rc != 0) { + if (rc != -EAGAIN) { + CWARN("Accept error %d: pausing...\n", rc); + schedule_timeout_uninterruptible( + cfs_time_seconds(1)); + } + continue; + } + + /* make sure we call lnet_sock_accept() again, until it fails */ + atomic_set(&lnet_acceptor_state.pta_ready, 1); + + rc = lnet_sock_getaddr(newsock, true, &peer); + if (rc != 0) { + CERROR("Can't determine new connection's address\n"); + goto failed; + } + + if (secure && + rpc_get_port((struct sockaddr *)&peer) > + LNET_ACCEPTOR_MAX_RESERVED_PORT) { + CERROR("Refusing connection from %pISp: insecure port.\n", + &peer); + goto failed; + } + + rc = lnet_sock_read(newsock, &magic, sizeof(magic), + accept_timeout); + if (rc != 0) { + CERROR("Error %d reading connection request from %pIS\n", + rc, &peer); + goto failed; + } + + rc = lnet_accept(newsock, magic); + if (rc != 0) + goto failed; + + continue; + +failed: + sock_release(newsock); + } + + lnet_acceptor_state.pta_sock->sk->sk_data_ready = + lnet_acceptor_state.pta_odata; + sock_release(lnet_acceptor_state.pta_sock); + lnet_acceptor_state.pta_sock = NULL; + + CDEBUG(D_NET, "Acceptor stopping\n"); + + /* unblock lnet_acceptor_stop() */ + complete(&lnet_acceptor_state.pta_signal); + return 0; } static inline int accept2secure(const char *acc, long *sec) { - if (!strcmp(acc, "secure")) { - *sec = 1; - return 1; - } else if (!strcmp(acc, "all")) { - *sec = 0; - return 1; - } else if (!strcmp(acc, "none")) { - return 0; - } else { - LCONSOLE_ERROR_MSG(0x124, "Can't parse 'accept=\"%s\"'\n", - acc); - return -EINVAL; - } + if (!strcmp(acc, "secure")) { + *sec = 1; + return 1; + } else if (!strcmp(acc, "all")) { + *sec = 0; + return 1; + } else if (!strcmp(acc, "none")) { + return 0; + } else { + LCONSOLE_ERROR_MSG(0x124, "Can't parse 'accept=\"%s\"'\n", + acc); + return -EINVAL; + } } int lnet_acceptor_start(void) { - int rc; - long rc2; - long secure; - - LASSERT (lnet_acceptor_state.pta_sock == NULL); - - rc = lnet_acceptor_get_tunables(); - if (rc != 0) - return rc; - -#ifndef __KERNEL__ - /* Do nothing if we're liblustre clients */ - if ((the_lnet.ln_pid & LNET_PID_USERFLAG) != 0) - return 0; -#endif - - mt_init_completion(&lnet_acceptor_state.pta_signal); - rc = accept2secure(accept_type, &secure); - if (rc <= 0) { - mt_fini_completion(&lnet_acceptor_state.pta_signal); - return rc; - } - - if (lnet_count_acceptor_nis() == 0) /* not required */ - return 0; - - rc2 = cfs_create_thread(lnet_acceptor, (void *)(ulong_ptr_t)secure, 0); - if (rc2 < 0) { - CERROR("Can't start acceptor thread: %d\n", rc); - mt_fini_completion(&lnet_acceptor_state.pta_signal); - - return -ESRCH; - } - - /* wait for acceptor to startup */ - mt_wait_for_completion(&lnet_acceptor_state.pta_signal); - - if (!lnet_acceptor_state.pta_shutdown) { - /* started OK */ - LASSERT (lnet_acceptor_state.pta_sock != NULL); - return 0; - } - - LASSERT (lnet_acceptor_state.pta_sock == NULL); - mt_fini_completion(&lnet_acceptor_state.pta_signal); - - return -ENETDOWN; + struct task_struct *task; + int rc; + long rc2; + long secure; + + /* if acceptor is already running return immediately */ + if (!lnet_acceptor_state.pta_shutdown) + return 0; + + LASSERT(lnet_acceptor_state.pta_sock == NULL); + + init_completion(&lnet_acceptor_state.pta_signal); + rc = accept2secure(accept_type, &secure); + if (rc <= 0) + return rc; + + if (lnet_count_acceptor_nets() == 0) /* not required */ + return 0; + if (current->nsproxy && current->nsproxy->net_ns) + lnet_acceptor_state.pta_ns = current->nsproxy->net_ns; + else + lnet_acceptor_state.pta_ns = &init_net; + task = kthread_run(lnet_acceptor, (void *)(uintptr_t)secure, + "acceptor_%03ld", secure); + if (IS_ERR(task)) { + rc2 = PTR_ERR(task); + CERROR("Can't start acceptor thread: %ld\n", rc2); + return -ESRCH; + } + + /* wait for acceptor to startup */ + wait_for_completion(&lnet_acceptor_state.pta_signal); + + if (!lnet_acceptor_state.pta_shutdown) { + /* started OK */ + LASSERT(lnet_acceptor_state.pta_sock != NULL); + return 0; + } + + LASSERT(lnet_acceptor_state.pta_sock == NULL); + + return -ENETDOWN; } void lnet_acceptor_stop(void) { - if (lnet_acceptor_state.pta_sock == NULL) /* not running */ - return; + if (lnet_acceptor_state.pta_shutdown) /* not running */ + return; - lnet_acceptor_state.pta_shutdown = 1; - libcfs_sock_abort_accept(lnet_acceptor_state.pta_sock); + /* If still required, return immediately */ + if (the_lnet.ln_refcount && lnet_count_acceptor_nets() > 0) + return; - /* block until acceptor signals exit */ - mt_wait_for_completion(&lnet_acceptor_state.pta_signal); + lnet_acceptor_state.pta_shutdown = 1; + wake_up(&lnet_acceptor_state.pta_waitq); - mt_fini_completion(&lnet_acceptor_state.pta_signal); -} - -#else /* single-threaded user-space */ -int -lnet_acceptor_start(void) -{ - return 0; -} - -void -lnet_acceptor_stop(void) -{ + /* block until acceptor signals exit */ + wait_for_completion(&lnet_acceptor_state.pta_signal); } -#endif /* defined(__KERNEL__) || defined(HAVE_LIBPTHREAD) */