diff options
author | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-17 13:42:57 +0000 |
---|---|---|
committer | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-17 13:42:57 +0000 |
commit | 550d8e8e6ccef95a119bc265101792b0475a7aa0 (patch) | |
tree | fdeb0b5ff80273f95ce61607fc3613dff0b9a235 /modules/proxy/proxy_util.c | |
parent | Adding upstream version 2.4.57. (diff) | |
download | apache2-550d8e8e6ccef95a119bc265101792b0475a7aa0.tar.xz apache2-550d8e8e6ccef95a119bc265101792b0475a7aa0.zip |
Adding upstream version 2.4.59.upstream/2.4.59upstream
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to '')
-rw-r--r-- | modules/proxy/proxy_util.c | 1098 |
1 files changed, 819 insertions, 279 deletions
diff --git a/modules/proxy/proxy_util.c b/modules/proxy/proxy_util.c index caafde0..a54a4fa 100644 --- a/modules/proxy/proxy_util.c +++ b/modules/proxy/proxy_util.c @@ -21,10 +21,14 @@ #include "apr_version.h" #include "apr_strings.h" #include "apr_hash.h" +#include "apr_atomic.h" +#include "http_core.h" #include "proxy_util.h" #include "ajp.h" #include "scgi.h" +#include "mpm_common.h" /* for ap_max_mem_free */ + #include "mod_http2.h" /* for http2_get_num_workers() */ #if APR_HAVE_UNISTD_H @@ -43,7 +47,7 @@ APLOG_USE_MODULE(proxy); /* * Opaque structure containing target server info when * using a forward proxy. - * Up to now only used in combination with HTTP CONNECT. + * Up to now only used in combination with HTTP CONNECT to ProxyRemote */ typedef struct { int use_http_connect; /* Use SSL Tunneling via HTTP CONNECT */ @@ -52,6 +56,17 @@ typedef struct { const char *proxy_auth; /* Proxy authorization */ } forward_info; +/* + * Opaque structure containing a refcounted and TTL'ed address. + */ +typedef struct proxy_address { + apr_sockaddr_t *addr; /* Remote address info */ + const char *hostname; /* Remote host name */ + apr_port_t hostport; /* Remote host port */ + apr_uint32_t refcount; /* Number of conns and/or worker using it */ + apr_uint32_t expiry; /* Expiry timestamp (seconds to proxy_start_time) */ +} proxy_address; + /* Global balancer counter */ int PROXY_DECLARE_DATA proxy_lb_workers = 0; static int lb_workers_limit = 0; @@ -60,6 +75,8 @@ const apr_strmatch_pattern PROXY_DECLARE_DATA *ap_proxy_strmatch_domain; extern apr_global_mutex_t *proxy_mutex; +static const apr_time_t *proxy_start_time; /* epoch for expiring addresses */ + static int proxy_match_ipaddr(struct dirconn_entry *This, request_rec *r); static int proxy_match_domainname(struct dirconn_entry *This, request_rec *r); static int proxy_match_hostname(struct dirconn_entry *This, request_rec *r); @@ -392,8 +409,12 @@ PROXY_DECLARE(char *) return NULL; } -PROXY_DECLARE(int) ap_proxyerror(request_rec *r, int statuscode, const char *message) +static int proxyerror_core(request_rec *r, int statuscode, const char *message, + apr_status_t rv) { + ap_log_rerror(APLOG_MARK, APLOG_ERR, rv, r, APLOGNO(00898) + "%s returned by %s", message, r->uri); + apr_table_setn(r->notes, "error-notes", apr_pstrcat(r->pool, "The proxy server could not handle the request<p>" @@ -405,11 +426,14 @@ PROXY_DECLARE(int) ap_proxyerror(request_rec *r, int statuscode, const char *mes apr_table_setn(r->notes, "verbose-error-to", "*"); r->status_line = apr_psprintf(r->pool, "%3.3u Proxy Error", statuscode); - ap_log_rerror(APLOG_MARK, APLOG_ERR, 0, r, APLOGNO(00898) "%s returned by %s", message, - r->uri); return statuscode; } +PROXY_DECLARE(int) ap_proxyerror(request_rec *r, int statuscode, const char *message) +{ + return proxyerror_core(r, statuscode, message, 0); +} + static const char * proxy_get_host_of_request(request_rec *r) { @@ -1488,43 +1512,94 @@ static void socket_cleanup(proxy_conn_rec *conn) apr_pool_clear(conn->scpool); } +static void address_cleanup(proxy_conn_rec *conn) +{ + conn->address = NULL; + conn->addr = NULL; + conn->hostname = NULL; + conn->port = 0; + conn->uds_path = NULL; + if (conn->uds_pool) { + apr_pool_clear(conn->uds_pool); + } + if (conn->sock) { + socket_cleanup(conn); + } +} + static apr_status_t conn_pool_cleanup(void *theworker) { ((proxy_worker *)theworker)->cp = NULL; return APR_SUCCESS; } -static void init_conn_pool(apr_pool_t *p, proxy_worker *worker) +static apr_pool_t *make_conn_subpool(apr_pool_t *p, const char *tag, + server_rec *s) +{ + apr_pool_t *sp = NULL; + apr_allocator_t *alloc; + apr_thread_mutex_t *mutex; + apr_status_t rv; + + rv = apr_allocator_create(&alloc); + if (rv == APR_SUCCESS) { + rv = apr_thread_mutex_create(&mutex, APR_THREAD_MUTEX_DEFAULT, p); + if (rv == APR_SUCCESS) { + apr_allocator_mutex_set(alloc, mutex); + apr_allocator_max_free_set(alloc, ap_max_mem_free); + rv = apr_pool_create_ex(&sp, p, NULL, alloc); + } + else { + apr_allocator_destroy(alloc); + } + } + if (rv != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_CRIT, rv, s, APLOGNO(10474) + "failed to create %s pool", tag); + ap_abort_on_oom(); + return NULL; /* not reached */ + } + apr_allocator_owner_set(alloc, sp); + apr_pool_tag(sp, tag); + + return sp; +} + +static void init_conn_pool(apr_pool_t *p, proxy_worker *worker, server_rec *s) { - apr_pool_t *pool; - apr_pool_t *dns_pool; proxy_conn_pool *cp; /* - * Create a connection pool's subpool. - * This pool is used for connection recycling. - * Once the worker is added it is never removed but - * it can be disabled. - */ - apr_pool_create(&pool, p); - apr_pool_tag(pool, "proxy_worker_cp"); - /* - * Create a subpool of the connection pool for worker - * scoped DNS resolutions. This is needed to avoid race - * conditions in using the connection pool by multiple - * threads during ramp up. - */ - apr_pool_create(&dns_pool, pool); - apr_pool_tag(dns_pool, "proxy_worker_dns"); - /* * Alloc from the same pool as worker. * proxy_conn_pool is permanently attached to the worker. */ cp = (proxy_conn_pool *)apr_pcalloc(p, sizeof(proxy_conn_pool)); - cp->pool = pool; - cp->dns_pool = dns_pool; worker->cp = cp; + /* + * We need a first pool (cp->pool) to maintain the connections attached to + * the worker and a second one (cp->dns_pool) to maintain the DNS addresses + * in use (TTL'ed, refcounted). New connections are created as/on a subpool + * of cp->pool and new addresses as/on a subpool of cp->dns_pool, such that + * both leaks (the subpools can be destroyed when the connections and/or + * addresses are over) and race conditions (the creation/destruction of + * subpools is protected by the parent pool's mutex) can be avoided. + * + * cp->dns_pool is created before cp->pool because when a connection on the + * latter is destroyed it might destroy an address on the former, so when + * the base pools are destroyed (e.g. child exit) we thusly make sure that + * cp->dns_pool and its subpools are still alive when cp->pool gets killed. + * + * Both cp->dns_pool and cp->pool have their own allocator/mutex too since + * acquiring connections and addresses don't need to contend. + */ + cp->dns_pool = make_conn_subpool(p, "proxy_worker_dns", s); + cp->pool = make_conn_subpool(p, "proxy_worker_cp", s); + + /* When p is cleaning up the child is exiting, signal that to e.g. avoid + * destroying the subpools explicitely in connection_destructor() when + * they have been destroyed already by the reslist cleanup. + */ apr_pool_pre_cleanup_register(p, worker, conn_pool_cleanup); } @@ -1532,41 +1607,67 @@ PROXY_DECLARE(int) ap_proxy_connection_reusable(proxy_conn_rec *conn) { proxy_worker *worker = conn->worker; - return ! (conn->close || !worker->s->is_address_reusable || worker->s->disablereuse); + return !(conn->close + || conn->forward + || worker->s->disablereuse + || !worker->s->is_address_reusable); +} + +static proxy_conn_rec *connection_make(apr_pool_t *p, proxy_worker *worker) +{ + proxy_conn_rec *conn; + + conn = apr_pcalloc(p, sizeof(proxy_conn_rec)); + conn->pool = p; + conn->worker = worker; + + /* + * Create another subpool that manages the data for the + * socket and the connection member of the proxy_conn_rec struct as we + * destroy this data more frequently than other data in the proxy_conn_rec + * struct like hostname and addr (at least in the case where we have + * keepalive connections that timed out). + * + * XXX: this is really needed only when worker->s->is_address_reusable, + * otherwise conn->scpool = conn->pool would be fine. For now we + * can't change it since it's (kind of) part of the API. + */ + apr_pool_create(&conn->scpool, p); + apr_pool_tag(conn->scpool, "proxy_conn_scpool"); + + return conn; } -static apr_status_t connection_cleanup(void *theconn) +static void connection_cleanup(void *theconn) { proxy_conn_rec *conn = (proxy_conn_rec *)theconn; proxy_worker *worker = conn->worker; - if (conn->r) { - apr_pool_destroy(conn->r->pool); - conn->r = NULL; - } - /* Sanity check: Did we already return the pooled connection? */ if (conn->inreslist) { ap_log_perror(APLOG_MARK, APLOG_ERR, 0, conn->pool, APLOGNO(00923) "Pooled connection 0x%pp for worker %s has been" " already returned to the connection pool.", conn, ap_proxy_worker_name(conn->pool, worker)); - return APR_SUCCESS; + return; + } + + if (conn->r) { + apr_pool_destroy(conn->r->pool); + conn->r = NULL; } - /* determine if the connection need to be closed */ - if (!worker->s->is_address_reusable || worker->s->disablereuse) { + /* determine if the connection should be cleared, closed or reused */ + if (!worker->s->is_address_reusable) { apr_pool_t *p = conn->pool; apr_pool_clear(p); - conn = apr_pcalloc(p, sizeof(proxy_conn_rec)); - conn->pool = p; - conn->worker = worker; - apr_pool_create(&(conn->scpool), p); - apr_pool_tag(conn->scpool, "proxy_conn_scpool"); + conn = connection_make(p, worker); } else if (conn->close - || (conn->connection - && conn->connection->keepalive == AP_CONN_CLOSE)) { + || conn->forward + || (conn->connection + && conn->connection->keepalive == AP_CONN_CLOSE) + || worker->s->disablereuse) { socket_cleanup(conn); conn->close = 0; } @@ -1582,13 +1683,9 @@ static apr_status_t connection_cleanup(void *theconn) conn->inreslist = 1; apr_reslist_release(worker->cp->res, (void *)conn); } - else - { + else { worker->cp->conn = conn; } - - /* Always return the SUCCESS */ - return APR_SUCCESS; } /* DEPRECATED */ @@ -1629,35 +1726,21 @@ PROXY_DECLARE(apr_status_t) ap_proxy_ssl_connection_cleanup(proxy_conn_rec *conn static apr_status_t connection_constructor(void **resource, void *params, apr_pool_t *pool) { - apr_pool_t *ctx; - apr_pool_t *scpool; + apr_pool_t *p; proxy_conn_rec *conn; proxy_worker *worker = (proxy_worker *)params; /* - * Create the subpool for each connection + * Create a subpool for each connection * This keeps the memory consumption constant - * when disconnecting from backend. + * when it's recycled or destroyed. */ - apr_pool_create(&ctx, pool); - apr_pool_tag(ctx, "proxy_conn_pool"); - /* - * Create another subpool that manages the data for the - * socket and the connection member of the proxy_conn_rec struct as we - * destroy this data more frequently than other data in the proxy_conn_rec - * struct like hostname and addr (at least in the case where we have - * keepalive connections that timed out). - */ - apr_pool_create(&scpool, ctx); - apr_pool_tag(scpool, "proxy_conn_scpool"); - conn = apr_pcalloc(ctx, sizeof(proxy_conn_rec)); - - conn->pool = ctx; - conn->scpool = scpool; - conn->worker = worker; + apr_pool_create(&p, pool); + apr_pool_tag(p, "proxy_conn_pool"); + conn = connection_make(p, worker); conn->inreslist = 1; - *resource = conn; + *resource = conn; return APR_SUCCESS; } @@ -1881,6 +1964,7 @@ PROXY_DECLARE(char *) ap_proxy_define_worker_ex(apr_pool_t *p, proxy_worker_shared *wshared; const char *ptr = NULL, *sockpath = NULL, *pdollars = NULL; apr_port_t port_of_scheme; + int address_not_reusable = 0; apr_uri_t uri; /* @@ -1909,12 +1993,21 @@ PROXY_DECLARE(char *) ap_proxy_define_worker_ex(apr_pool_t *p, * to fail (e.g. "ProxyPassMatch ^/(a|b)(/.*)? http://host:port$2"). * So we trim all the $n from the :port and prepend them in uri.path * afterward for apr_uri_unparse() to restore the original URL below. + * If a dollar substitution is found in the hostname[:port] part of + * the URL, reusing address and connections in the same worker is not + * possible (the current implementation of active connections cache + * handles/assumes a single origin server:port per worker only), so + * we set address_not_reusable here during parsing to take that into + * account in the worker settings below. */ #define IS_REF(x) (x[0] == '$' && apr_isdigit(x[1])) const char *pos = ap_strstr_c(ptr, "://"); if (pos) { pos += 3; while (*pos && *pos != ':' && *pos != '/') { + if (*pos == '$') { + address_not_reusable = 1; + } pos++; } if (*pos == ':') { @@ -1934,6 +2027,7 @@ PROXY_DECLARE(char *) ap_proxy_define_worker_ex(apr_pool_t *p, vec[1].iov_base = (void *)path; vec[1].iov_len = strlen(path); ptr = apr_pstrcatv(p, vec, 2, NULL); + address_not_reusable = 1; } } } @@ -2029,7 +2123,9 @@ PROXY_DECLARE(char *) ap_proxy_define_worker_ex(apr_pool_t *p, wshared->port = (uri.port) ? uri.port : port_of_scheme; wshared->flush_packets = flush_off; wshared->flush_wait = PROXY_FLUSH_WAIT; - wshared->is_address_reusable = 1; + wshared->address_ttl = (address_not_reusable) ? 0 : -1; + wshared->is_address_reusable = (address_not_reusable == 0); + wshared->disablereuse = (address_not_reusable != 0); wshared->lbfactor = 100; wshared->passes = 1; wshared->fails = 1; @@ -2038,7 +2134,30 @@ PROXY_DECLARE(char *) ap_proxy_define_worker_ex(apr_pool_t *p, wshared->hash.def = ap_proxy_hashfunc(wshared->name_ex, PROXY_HASHFUNC_DEFAULT); wshared->hash.fnv = ap_proxy_hashfunc(wshared->name_ex, PROXY_HASHFUNC_FNV); wshared->was_malloced = (mask & AP_PROXY_WORKER_IS_MALLOCED) != 0; - wshared->is_name_matchable = 0; + if (mask & AP_PROXY_WORKER_IS_MATCH) { + wshared->is_name_matchable = 1; + + /* Before AP_PROXY_WORKER_IS_MATCH (< 2.4.47), a regex worker with + * dollar substitution was never matched against any actual URL, thus + * the requests fell through the generic worker. Now if a ProyPassMatch + * matches, a worker (and its parameters) is always used to determine + * the properties of the connection with the origin server. So for + * instance the same "timeout=" will be enforced for all the requests + * matched by the same ProyPassMatch worker, which is an improvement + * compared to the global/vhost [Proxy]Timeout applied by the generic + * worker. Likewise, address and connection reuse is the default for + * a ProyPassMatch worker with no dollar substitution, just like a + * "normal" worker. However to avoid DNS and connection reuse compat + * issues, connection reuse is disabled by default if there is any + * substitution in the uri-path (an explicit enablereuse=on can still + * opt-in), and reuse is even disabled definitively for substitutions + * happening in the hostname[:port] (is_address_reusable was unset + * above so it will prevent enablereuse=on to apply anyway). + */ + if (ap_strchr_c(wshared->name, '$')) { + wshared->disablereuse = 1; + } + } if (sockpath) { if (PROXY_STRNCPY(wshared->uds_path, sockpath) != APR_SUCCESS) { return apr_psprintf(p, "worker uds path (%s) too long", sockpath); @@ -2058,20 +2177,6 @@ PROXY_DECLARE(char *) ap_proxy_define_worker_ex(apr_pool_t *p, (*worker)->balancer = balancer; (*worker)->s = wshared; - if (mask & AP_PROXY_WORKER_IS_MATCH) { - (*worker)->s->is_name_matchable = 1; - if (ap_strchr_c((*worker)->s->name_ex, '$')) { - /* Before AP_PROXY_WORKER_IS_MATCH (< 2.4.47), a regex worker - * with dollar substitution was never matched against the actual - * URL thus the request fell through the generic worker. To avoid - * dns and connection reuse compat issues, let's disable connection - * reuse by default, it can still be overwritten by an explicit - * enablereuse=on. - */ - (*worker)->s->disablereuse = 1; - } - } - return NULL; } @@ -2157,12 +2262,23 @@ PROXY_DECLARE(apr_status_t) ap_proxy_initialize_worker(proxy_worker *worker, ser if (!worker->s->retry_set) { worker->s->retry = apr_time_from_sec(PROXY_WORKER_DEFAULT_RETRY); } - /* By default address is reusable unless DisableReuse is set */ - if (worker->s->disablereuse) { + /* Consistently set address and connection reusabilty: when reuse + * is disabled by configuration, or when the address is known already + * to not be reusable for this worker (in any case, thus ignore/force + * DisableReuse). + */ + if (!worker->s->address_ttl || (!worker->s->address_ttl_set + && worker->s->disablereuse)) { worker->s->is_address_reusable = 0; } - else { - worker->s->is_address_reusable = 1; + if (!worker->s->is_address_reusable && !worker->s->disablereuse) { + /* Explicit enablereuse=on can't work in this case, warn user. */ + if (worker->s->disablereuse_set) { + ap_log_error(APLOG_MARK, APLOG_WARNING, 0, s, APLOGNO(10400) + "enablereuse/disablereuse ignored for worker %s", + ap_proxy_worker_name(p, worker)); + } + worker->s->disablereuse = 1; } /* @@ -2226,7 +2342,7 @@ PROXY_DECLARE(apr_status_t) ap_proxy_initialize_worker(proxy_worker *worker, ser } #endif if (worker->cp == NULL) - init_conn_pool(p, worker); + init_conn_pool(p, worker, s); if (worker->cp == NULL) { ap_log_error(APLOG_MARK, APLOG_ERR, 0, s, APLOGNO(00929) "can not create connection pool"); @@ -2590,6 +2706,354 @@ PROXY_DECLARE(int) ap_proxy_release_connection(const char *proxy_function, return OK; } +static APR_INLINE void proxy_address_inc(proxy_address *address) +{ + apr_uint32_t old = apr_atomic_inc32(&address->refcount); + ap_assert(old > 0 && old < APR_UINT32_MAX); +} + +static APR_INLINE void proxy_address_dec(proxy_address *address) +{ + /* Use _add32(, -1) since _dec32()'s returned value does not help */ + apr_uint32_t old = apr_atomic_add32(&address->refcount, -1); + ap_assert(old > 0); + if (old == 1) { + apr_pool_destroy(address->addr->pool); + } +} + +static apr_status_t proxy_address_cleanup(void *address) +{ + proxy_address_dec(address); + return APR_SUCCESS; +} + +static APR_INLINE proxy_address *worker_address_get(proxy_worker *worker) +{ + /* No _readptr() so let's _casptr(, NULL, NULL) instead */ + return apr_atomic_casptr((void *)&worker->address, NULL, NULL); +} + +/* XXX: Call when PROXY_THREAD_LOCK()ed only! */ +static APR_INLINE void worker_address_set(proxy_worker *worker, + proxy_address *to) +{ + proxy_address *old = apr_atomic_xchgptr((void *)&worker->address, to); + if (old && old != to) { + proxy_address_dec(old); + } +} + +static apr_status_t worker_address_resolve(proxy_worker *worker, + apr_sockaddr_t **paddr, + const char *hostname, + apr_port_t hostport, + const char *proxy_function, + request_rec *r, server_rec *s) +{ + apr_status_t rv; + apr_pool_t *pool = NULL; + + apr_pool_create(&pool, worker->cp->dns_pool); + rv = apr_sockaddr_info_get(paddr, hostname, APR_UNSPEC, + hostport, 0, pool); + if (rv != APR_SUCCESS) { + if (r && !s) { + proxyerror_core(r, HTTP_INTERNAL_SERVER_ERROR, + apr_pstrcat(pool, + "DNS lookup failure for: ", + hostname, NULL), + rv); + } + else if (r) { + ap_log_rerror(APLOG_MARK, APLOG_DEBUG, rv, r, APLOGNO(10477) + "%s: resolving worker %s address", + proxy_function, hostname); + } + else { + ap_log_error(APLOG_MARK, APLOG_DEBUG, rv, s, APLOGNO(10478) + "%s: resolving worker %s address", + proxy_function, hostname); + } + apr_pool_destroy(pool); + return rv; + } + + if (r ? APLOGrdebug(r) : APLOGdebug(s)) { + char *addrs = NULL; + apr_sockaddr_t *addr = *paddr; + for (; addr; addr = addr->next) { + addrs = apr_psprintf(pool, "%s%s%pI", + addrs ? ", " : "", + addrs ? addrs : "", + addr); + } + if (r) { + ap_log_rerror(APLOG_MARK, APLOG_DEBUG, 0, r, APLOGNO(10479) + "%s: %s resolved to %s", + proxy_function, hostname, addrs); + } + else { + ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, s, APLOGNO(10480) + "%s: %s resolved to %s", + proxy_function, hostname, addrs); + } + } + + return APR_SUCCESS; +} + +static int proxy_addrs_equal(const apr_sockaddr_t *addr1, + const apr_sockaddr_t *addr2) +{ + const apr_sockaddr_t *base2 = addr2, *pos2; + while (addr1 && addr2) { + for (pos2 = base2; pos2; pos2 = pos2->next) { + if (apr_sockaddr_equal(pos2, addr1)) { + break; + } + } + if (!pos2) { + return 0; + } + addr1 = addr1->next; + addr2 = addr2->next; + } + if (addr1 || addr2) { + return 0; + } + return 1; +} + +PROXY_DECLARE(apr_status_t) ap_proxy_determine_address(const char *proxy_function, + proxy_conn_rec *conn, + const char *hostname, + apr_port_t hostport, + unsigned int flags, + request_rec *r, + server_rec *s) +{ + proxy_worker *worker = conn->worker; + apr_status_t rv; + + /* + * Worker can have the single constant backend adress. + * The single DNS lookup is used once per worker. + * If dynamic change is needed then set the addr to NULL + * inside dynamic config to force the lookup. + * The worker's addressTTL parameter may also be configured + * to perform the DNS lookups only when the TTL expires, + * or each time if that TTL is zero. + */ + if (!worker->s->is_address_reusable) { + conn->hostname = apr_pstrdup(conn->pool, hostname); + conn->port = hostport; + + rv = apr_sockaddr_info_get(&conn->addr, hostname, APR_UNSPEC, + hostport, 0, conn->pool); + if (rv != APR_SUCCESS) { + if (r && !s) { + proxyerror_core(r, HTTP_INTERNAL_SERVER_ERROR, + apr_pstrcat(r->pool, "DNS lookup failure for: ", + hostname, NULL), rv); + } + else if (r) { + ap_log_rerror(APLOG_MARK, APLOG_DEBUG, rv, r, APLOGNO(10475) + "%s: resolving backend %s address", + proxy_function, hostname); + } + else { + ap_log_error(APLOG_MARK, APLOG_DEBUG, rv, s, APLOGNO(10476) + "%s: resolving backend %s address", + proxy_function, hostname); + } + return rv; + } + } + else { + apr_sockaddr_t *addr = NULL; + proxy_address *address = NULL; + apr_int32_t ttl = worker->s->address_ttl; + apr_uint32_t now = 0; + + if (flags & PROXY_DETERMINE_ADDRESS_CHECK) { + /* The caller wants to check if the address changed, return + * APR_EEXIST if not, otherwise fall through to update the + * worker's for everyone to switch. + */ + if (!conn->addr) { + /* Need something to compare with */ + return APR_EINVAL; + } + rv = worker_address_resolve(worker, &addr, + hostname, hostport, + proxy_function, r, s); + if (rv != APR_SUCCESS) { + return rv; + } + if (proxy_addrs_equal(conn->addr, addr)) { + apr_pool_destroy(addr->pool); + return APR_EEXIST; + } + } + + AP_DEBUG_ASSERT(ttl != 0); + if (ttl > 0) { + /* TODO: use a monotonic clock here */ + now = apr_time_sec(apr_time_now() - *proxy_start_time); + } + + /* Addresses are refcounted, destroyed when their refcount reaches 0. + * + * One ref is taken by worker->address as the worker's current/latest + * address, it's dropped when that address expires/changes (see below). + * The other refs are taken by the connections when using/switching to + * the current worker address (also below), they are dropped when the + * conns are destroyed (by the reslist though it should never happen + * if hmax is greater than the number of threads) OR for an expired + * conn->address when it's replaced by the new worker->address below. + * + * Dereferencing worker->address requires holding the worker mutex or + * some concurrent connection processing might change/destroy it at any + * time. So only conn->address is safe to dereference anywhere (unless + * NULL..) since it has at least the lifetime of the connection. + */ + if (!addr) { + address = worker_address_get(worker); + } + if (!address + || conn->address != address + || apr_atomic_read32(&address->expiry) <= now) { + PROXY_THREAD_LOCK(worker); + + /* Re-check while locked, might be a new address already */ + if (!addr) { + address = worker_address_get(worker); + } + if (!address || apr_atomic_read32(&address->expiry) <= now) { + if (!addr) { + rv = worker_address_resolve(worker, &addr, + hostname, hostport, + proxy_function, r, s); + if (rv != APR_SUCCESS) { + PROXY_THREAD_UNLOCK(worker); + return rv; + } + + /* Recompute "now" should the DNS be slow + * TODO: use a monotonic clock here + */ + now = apr_time_sec(apr_time_now() - *proxy_start_time); + } + + address = apr_pcalloc(addr->pool, sizeof(*address)); + address->hostname = apr_pstrdup(addr->pool, hostname); + address->hostport = hostport; + address->addr = addr; + + if (ttl > 0) { + /* We keep each worker's expiry date shared accross all the + * children so that they update their address at the same + * time, regardless of whether a specific child forced an + * address to expire at some point (for connect() issues). + */ + address->expiry = apr_atomic_read32(&worker->s->address_expiry); + if (address->expiry <= now) { + apr_uint32_t new_expiry = address->expiry + ttl; + while (new_expiry <= now) { + new_expiry += ttl; + } + new_expiry = apr_atomic_cas32(&worker->s->address_expiry, + new_expiry, address->expiry); + /* race lost? well the expiry should grow anyway.. */ + AP_DEBUG_ASSERT(new_expiry > now); + address->expiry = new_expiry; + } + } + else { + /* Never expires */ + address->expiry = APR_UINT32_MAX; + } + + /* One ref is for worker->address in any case */ + if (worker->address || worker->cp->addr) { + apr_atomic_set32(&address->refcount, 1); + } + else { + /* Set worker->cp->addr once for compat with third-party + * modules. This addr never changed before and can't change + * underneath users now because of some TTL configuration. + * So we take one more ref for worker->cp->addr to remain + * allocated forever (though it might not be up to date..). + * Modules should use conn->addr instead of worker->cp-addr + * to get the actual address used by each conn, determined + * at connect() time. + */ + apr_atomic_set32(&address->refcount, 2); + worker->cp->addr = address->addr; + } + + /* Publish the changes. The old worker address (if any) is no + * longer used by this worker, it will be destroyed now if the + * worker is the last user (refcount == 1) or by the last conn + * using it (refcount > 1). + */ + worker_address_set(worker, address); + } + + /* Take the ref for conn->address (before dropping the mutex so to + * let no chance for this address be killed before it's used!) + */ + proxy_address_inc(address); + + PROXY_THREAD_UNLOCK(worker); + + /* Kill any socket using the old address */ + if (conn->sock) { + if (r ? APLOGrdebug(r) : APLOGdebug(s)) { + /* XXX: this requires the old conn->addr[ess] to still + * be alive since it's not copied by apr_socket_connect() + * in ap_proxy_connect_backend(). + */ + apr_sockaddr_t *local_addr = NULL; + apr_sockaddr_t *remote_addr = NULL; + apr_socket_addr_get(&local_addr, APR_LOCAL, conn->sock); + apr_socket_addr_get(&remote_addr, APR_REMOTE, conn->sock); + if (r) { + ap_log_rerror(APLOG_MARK, APLOG_DEBUG, 0, r, APLOGNO(10481) + "%s: closing connection to %s (%pI<>%pI) on " + "address change", proxy_function, hostname, + local_addr, remote_addr); + } + else { + ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, s, APLOGNO(10482) + "%s: closing connection to %s (%pI<>%pI) on " + "address change", proxy_function, hostname, + local_addr, remote_addr); + } + } + socket_cleanup(conn); + } + + /* Kill the old address (if any) and use the new one */ + if (conn->address) { + apr_pool_cleanup_run(conn->pool, conn->address, + proxy_address_cleanup); + } + apr_pool_cleanup_register(conn->pool, address, + proxy_address_cleanup, + apr_pool_cleanup_null); + address_cleanup(conn); + conn->address = address; + conn->hostname = address->hostname; + conn->port = address->hostport; + conn->addr = address->addr; + } + } + + return APR_SUCCESS; +} + PROXY_DECLARE(int) ap_proxy_determine_connection(apr_pool_t *p, request_rec *r, proxy_server_conf *conf, @@ -2603,10 +3067,6 @@ ap_proxy_determine_connection(apr_pool_t *p, request_rec *r, int server_portstr_size) { int server_port; - apr_status_t err = APR_SUCCESS; -#if APR_HAS_THREADS - apr_status_t uerr = APR_SUCCESS; -#endif const char *uds_path; /* @@ -2626,6 +3086,12 @@ ap_proxy_determine_connection(apr_pool_t *p, request_rec *r, ap_log_rerror(APLOG_MARK, APLOG_DEBUG, 0, r, APLOGNO(00944) "connecting %s to %s:%d", *url, uri->hostname, uri->port); + /* Close a possible existing socket if we are told to do so */ + if (conn->close) { + socket_cleanup(conn); + conn->close = 0; + } + /* * allocate these out of the specified connection pool * The scheme handler decides if this is permanent or @@ -2652,143 +3118,122 @@ ap_proxy_determine_connection(apr_pool_t *p, request_rec *r, * to check host and port on the conn and be careful about * spilling the cached addr from the worker. */ - uds_path = (*worker->s->uds_path ? worker->s->uds_path : apr_table_get(r->notes, "uds_path")); + uds_path = (*worker->s->uds_path + ? worker->s->uds_path + : apr_table_get(r->notes, "uds_path")); if (uds_path) { - if (conn->uds_path == NULL) { - /* use (*conn)->pool instead of worker->cp->pool to match lifetime */ - conn->uds_path = apr_pstrdup(conn->pool, uds_path); - } - if (conn->uds_path) { - ap_log_rerror(APLOG_MARK, APLOG_DEBUG, 0, r, APLOGNO(02545) - "%s: has determined UDS as %s", - uri->scheme, conn->uds_path); - } - else { - /* should never happen */ - ap_log_rerror(APLOG_MARK, APLOG_DEBUG, 0, r, APLOGNO(02546) - "%s: cannot determine UDS (%s)", - uri->scheme, uds_path); - - } - /* - * In UDS cases, some structs are NULL. Protect from de-refs - * and provide info for logging at the same time. - */ - if (!conn->addr) { - apr_sockaddr_t *sa; - apr_sockaddr_info_get(&sa, NULL, APR_UNSPEC, 0, 0, conn->pool); - conn->addr = sa; - } - conn->hostname = "httpd-UDS"; - conn->port = 0; - } - else { - int will_reuse = worker->s->is_address_reusable && !worker->s->disablereuse; - if (!conn->hostname || !will_reuse) { - if (proxyname) { - conn->hostname = apr_pstrdup(conn->pool, proxyname); - conn->port = proxyport; - /* - * If we have a forward proxy and the protocol is HTTPS, - * then we need to prepend a HTTP CONNECT request before - * sending our actual HTTPS requests. - * Save our real backend data for using it later during HTTP CONNECT. - */ - if (conn->is_ssl) { - const char *proxy_auth; - - forward_info *forward = apr_pcalloc(conn->pool, sizeof(forward_info)); - conn->forward = forward; - forward->use_http_connect = 1; - forward->target_host = apr_pstrdup(conn->pool, uri->hostname); - forward->target_port = uri->port; - /* Do we want to pass Proxy-Authorization along? - * If we haven't used it, then YES - * If we have used it then MAYBE: RFC2616 says we MAY propagate it. - * So let's make it configurable by env. - * The logic here is the same used in mod_proxy_http. - */ - proxy_auth = apr_table_get(r->headers_in, "Proxy-Authorization"); - if (proxy_auth != NULL && - proxy_auth[0] != '\0' && - r->user == NULL && /* we haven't yet authenticated */ - apr_table_get(r->subprocess_env, "Proxy-Chain-Auth")) { - forward->proxy_auth = apr_pstrdup(conn->pool, proxy_auth); - } + if (!conn->uds_path || strcmp(conn->uds_path, uds_path) != 0) { + apr_pool_t *pool = conn->pool; + if (conn->uds_path) { + address_cleanup(conn); + if (!conn->uds_pool) { + apr_pool_create(&conn->uds_pool, worker->cp->dns_pool); } + pool = conn->uds_pool; } - else { - conn->hostname = apr_pstrdup(conn->pool, uri->hostname); - conn->port = uri->port; + /* + * In UDS cases, some structs are NULL. Protect from de-refs + * and provide info for logging at the same time. + */ +#if APR_HAVE_SOCKADDR_UN + apr_sockaddr_info_get(&conn->addr, uds_path, APR_UNIX, 0, 0, pool); + if (conn->addr && conn->addr->hostname) { + conn->uds_path = conn->addr->hostname; } - if (!will_reuse) { - /* - * Only do a lookup if we should not reuse the backend address. - * Otherwise we will look it up once for the worker. - */ - err = apr_sockaddr_info_get(&(conn->addr), - conn->hostname, APR_UNSPEC, - conn->port, 0, - conn->pool); + else { + conn->uds_path = apr_pstrdup(pool, uds_path); } - socket_cleanup(conn); - conn->close = 0; +#else + apr_sockaddr_info_get(&conn->addr, NULL, APR_UNSPEC, 0, 0, pool); + conn->uds_path = apr_pstrdup(pool, uds_path); +#endif + conn->hostname = apr_pstrdup(pool, uri->hostname); + conn->port = uri->port; } - if (will_reuse) { + ap_log_rerror(APLOG_MARK, APLOG_DEBUG, 0, r, APLOGNO(02545) + "%s: has determined UDS as %s (for %s:%hu)", + uri->scheme, conn->uds_path, conn->hostname, conn->port); + } + else { + const char *hostname = uri->hostname; + apr_port_t hostport = uri->port; + + /* Not a remote CONNECT until further notice */ + conn->forward = NULL; + + if (proxyname) { + hostname = proxyname; + hostport = proxyport; + /* - * Looking up the backend address for the worker only makes sense if - * we can reuse the address. + * If we have a remote proxy and the protocol is HTTPS, + * then we need to prepend a HTTP CONNECT request before + * sending our actual HTTPS requests. */ - if (!worker->cp->addr) { -#if APR_HAS_THREADS - if ((err = PROXY_THREAD_LOCK(worker)) != APR_SUCCESS) { - ap_log_rerror(APLOG_MARK, APLOG_ERR, err, r, APLOGNO(00945) "lock"); - return HTTP_INTERNAL_SERVER_ERROR; - } -#endif - - /* - * Recheck addr after we got the lock. This may have changed - * while waiting for the lock. + if (conn->is_ssl) { + forward_info *forward; + const char *proxy_auth; + + /* Do we want to pass Proxy-Authorization along? + * If we haven't used it, then YES + * If we have used it then MAYBE: RFC2616 says we MAY propagate it. + * So let's make it configurable by env. + * The logic here is the same used in mod_proxy_http. */ - if (!AP_VOLATILIZE_T(apr_sockaddr_t *, worker->cp->addr)) { + proxy_auth = apr_table_get(r->notes, "proxy-basic-creds"); + if (proxy_auth == NULL + && (r->user == NULL /* we haven't yet authenticated */ + || apr_table_get(r->subprocess_env, "Proxy-Chain-Auth"))) { + proxy_auth = apr_table_get(r->headers_in, "Proxy-Authorization"); + } + if (proxy_auth != NULL && proxy_auth[0] == '\0') { + proxy_auth = NULL; + } - apr_sockaddr_t *addr; + /* Reset forward info if they changed */ + if (!(forward = conn->forward) + || forward->target_port != uri->port + || ap_cstr_casecmp(forward->target_host, uri->hostname) != 0 + || (forward->proxy_auth != NULL) != (proxy_auth != NULL) + || (forward->proxy_auth != NULL && proxy_auth != NULL && + strcmp(forward->proxy_auth, proxy_auth) != 0)) { + apr_pool_t *fwd_pool = conn->pool; + if (worker->s->is_address_reusable) { + if (conn->fwd_pool) { + apr_pool_clear(conn->fwd_pool); + } + else { + apr_pool_create(&conn->fwd_pool, conn->pool); + } + fwd_pool = conn->fwd_pool; + } + forward = apr_pcalloc(fwd_pool, sizeof(forward_info)); + conn->forward = forward; /* - * Worker can have the single constant backend address. - * The single DNS lookup is used once per worker. - * If dynamic change is needed then set the addr to NULL - * inside dynamic config to force the lookup. + * Save our real backend data for using it later during HTTP CONNECT. */ - err = apr_sockaddr_info_get(&addr, - conn->hostname, APR_UNSPEC, - conn->port, 0, - worker->cp->dns_pool); - worker->cp->addr = addr; - } - conn->addr = worker->cp->addr; -#if APR_HAS_THREADS - if ((uerr = PROXY_THREAD_UNLOCK(worker)) != APR_SUCCESS) { - ap_log_rerror(APLOG_MARK, APLOG_ERR, uerr, r, APLOGNO(00946) "unlock"); + forward->use_http_connect = 1; + forward->target_host = apr_pstrdup(fwd_pool, uri->hostname); + forward->target_port = uri->port; + if (proxy_auth) { + forward->proxy_auth = apr_pstrdup(fwd_pool, proxy_auth); + } } -#endif - } - else { - conn->addr = worker->cp->addr; } } - } - /* Close a possible existing socket if we are told to do so */ - if (conn->close) { - socket_cleanup(conn); - conn->close = 0; - } - if (err != APR_SUCCESS) { - return ap_proxyerror(r, HTTP_BAD_GATEWAY, - apr_pstrcat(p, "DNS lookup failure for: ", - conn->hostname, NULL)); + if (conn->hostname + && (conn->port != hostport + || ap_cstr_casecmp(conn->hostname, hostname) != 0)) { + address_cleanup(conn); + } + + /* Resolve the connection address with the determined hostname/port */ + if (ap_proxy_determine_address(uri->scheme, conn, hostname, hostport, + 0, r, NULL)) { + return HTTP_INTERNAL_SERVER_ERROR; + } } /* Get the server port for the Via headers */ @@ -2847,7 +3292,8 @@ ap_proxy_determine_connection(apr_pool_t *p, request_rec *r, } } ap_log_rerror(APLOG_MARK, APLOG_DEBUG, 0, r, APLOGNO(00947) - "connected %s to %s:%d", *url, conn->hostname, conn->port); + "connecting %s to %pI (%s:%hu)", *url, + conn->addr, conn->hostname, conn->port); return OK; } @@ -2948,7 +3394,8 @@ static apr_status_t send_http_connect(proxy_conn_rec *backend, nbytes = apr_snprintf(buffer, sizeof(buffer), "CONNECT %s:%d HTTP/1.0" CRLF, forward->target_host, forward->target_port); - /* Add proxy authorization from the initial request if necessary */ + /* Add proxy authorization from the configuration, or initial + * request if necessary */ if (forward->proxy_auth != NULL) { nbytes += apr_snprintf(buffer + nbytes, sizeof(buffer) - nbytes, "Proxy-Authorization: %s" CRLF, @@ -3171,11 +3618,14 @@ PROXY_DECLARE(int) ap_proxy_connect_backend(const char *proxy_function, { apr_status_t rv; int loglevel; - apr_sockaddr_t *backend_addr = conn->addr; + forward_info *forward = conn->forward; + apr_sockaddr_t *backend_addr; /* the local address to use for the outgoing connection */ apr_sockaddr_t *local_addr; apr_socket_t *newsock; void *sconf = s->module_config; + int address_reusable = worker->s->is_address_reusable; + int did_dns_lookup = 0; proxy_server_conf *conf = (proxy_server_conf *) ap_get_module_config(sconf, &proxy_module); @@ -3184,6 +3634,16 @@ PROXY_DECLARE(int) ap_proxy_connect_backend(const char *proxy_function, return DECLINED; } + /* We'll set conn->addr to the address actually connect()ed, so if the + * network connection is not reused (per ap_proxy_check_connection() + * above) we need to reset conn->addr to the first resolved address + * and try to connect it first. + */ + if (conn->address && rv != APR_SUCCESS) { + conn->addr = conn->address->addr; + } + backend_addr = conn->addr; + while (rv != APR_SUCCESS && (backend_addr || conn->uds_path)) { #if APR_HAVE_SYS_UN_H if (conn->uds_path) @@ -3193,11 +3653,11 @@ PROXY_DECLARE(int) ap_proxy_connect_backend(const char *proxy_function, if (rv != APR_SUCCESS) { loglevel = APLOG_ERR; ap_log_error(APLOG_MARK, loglevel, rv, s, APLOGNO(02453) - "%s: error creating Unix domain socket for " - "target %s:%d", + "%s: error creating Unix domain socket " + "%s (%s:%hu)", proxy_function, - worker->s->hostname_ex, - (int)worker->s->port); + conn->uds_path, + conn->hostname, conn->port); break; } conn->connection = NULL; @@ -3207,21 +3667,18 @@ PROXY_DECLARE(int) ap_proxy_connect_backend(const char *proxy_function, apr_socket_close(newsock); ap_log_error(APLOG_MARK, APLOG_ERR, rv, s, APLOGNO(02454) "%s: attempt to connect to Unix domain socket " - "%s (%s:%d) failed", - proxy_function, - conn->uds_path, - worker->s->hostname_ex, - (int)worker->s->port); + "%s (%s:%hu) failed", + proxy_function, conn->uds_path, + conn->hostname, conn->port); break; } ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, s, APLOGNO(02823) "%s: connection established with Unix domain socket " - "%s (%s:%d)", + "%s (%s:%hu)", proxy_function, conn->uds_path, - worker->s->hostname_ex, - (int)worker->s->port); + conn->hostname, conn->port); } else #endif @@ -3231,12 +3688,11 @@ PROXY_DECLARE(int) ap_proxy_connect_backend(const char *proxy_function, conn->scpool)) != APR_SUCCESS) { loglevel = backend_addr->next ? APLOG_DEBUG : APLOG_ERR; ap_log_error(APLOG_MARK, loglevel, rv, s, APLOGNO(00952) - "%s: error creating fam %d socket for " - "target %s:%d", + "%s: error creating fam %d socket to %pI for " + "(%s:%hu)", proxy_function, - backend_addr->family, - worker->s->hostname_ex, - (int)worker->s->port); + backend_addr->family, backend_addr, + conn->hostname, conn->port); /* * this could be an IPv6 address from the DNS but the * local machine won't give us an IPv6 socket; hopefully the @@ -3285,9 +3741,9 @@ PROXY_DECLARE(int) ap_proxy_connect_backend(const char *proxy_function, } } ap_log_error(APLOG_MARK, APLOG_TRACE2, 0, s, - "%s: fam %d socket created to connect to %s:%d", - proxy_function, backend_addr->family, - worker->s->hostname_ex, (int)worker->s->port); + "%s: fam %d socket created for %pI (%s:%hu)", + proxy_function, backend_addr->family, backend_addr, + conn->hostname, conn->port); if (conf->source_address_set) { local_addr = apr_pmemdup(conn->scpool, conf->source_address, @@ -3309,21 +3765,45 @@ PROXY_DECLARE(int) ap_proxy_connect_backend(const char *proxy_function, apr_socket_close(newsock); loglevel = backend_addr->next ? APLOG_DEBUG : APLOG_ERR; ap_log_error(APLOG_MARK, loglevel, rv, s, APLOGNO(00957) - "%s: attempt to connect to %pI (%s:%d) failed", - proxy_function, - backend_addr, - worker->s->hostname_ex, - (int)worker->s->port); + "%s: attempt to connect to %pI (%s:%hu) failed", + proxy_function, backend_addr, + conn->hostname, conn->port); backend_addr = backend_addr->next; + /* + * If we run out of resolved IP's when connecting and if + * we cache the resolution in the worker the resolution + * might have changed. Hence try a DNS lookup to see if this + * helps. + */ + if (!backend_addr && address_reusable && !did_dns_lookup) { + /* Issue a new DNS lookup to check if the address changed, + * in which case (SUCCESS) restart the loop with the new + * one(s), otherwise leave (nothing we can do about it). + */ + if (ap_proxy_determine_address(proxy_function, conn, + conn->hostname, conn->port, + PROXY_DETERMINE_ADDRESS_CHECK, + NULL, s) == APR_SUCCESS) { + backend_addr = conn->addr; + } + + /* + * In case of an error backend_addr will be NULL which + * is enough to leave the loop. If successful we'll retry + * the new addresses only once. + */ + did_dns_lookup = 1; + } continue; } ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, s, APLOGNO(02824) - "%s: connection established with %pI (%s:%d)", - proxy_function, - backend_addr, - worker->s->hostname_ex, - (int)worker->s->port); + "%s: connection established with %pI (%s:%hu)", + proxy_function, backend_addr, + conn->hostname, conn->port); + + /* Set the actual sockaddr we are connected to */ + conn->addr = backend_addr; } /* Set a timeout on the socket */ @@ -3339,13 +3819,12 @@ PROXY_DECLARE(int) ap_proxy_connect_backend(const char *proxy_function, conn->sock = newsock; - if (!conn->uds_path && conn->forward) { - forward_info *forward = (forward_info *)conn->forward; + if (forward && forward->use_http_connect) { /* * For HTTP CONNECT we need to prepend CONNECT request before * sending our actual HTTPS requests. */ - if (forward->use_http_connect) { + { rv = send_http_connect(conn, s); /* If an error occurred, loop round and try again */ if (rv != APR_SUCCESS) { @@ -3353,12 +3832,11 @@ PROXY_DECLARE(int) ap_proxy_connect_backend(const char *proxy_function, apr_socket_close(newsock); loglevel = backend_addr->next ? APLOG_DEBUG : APLOG_ERR; ap_log_error(APLOG_MARK, loglevel, rv, s, APLOGNO(00958) - "%s: attempt to connect to %s:%d " - "via http CONNECT through %pI (%s:%d) failed", + "%s: attempt to connect to %s:%hu " + "via http CONNECT through %pI (%s:%hu) failed", proxy_function, forward->target_host, forward->target_port, - backend_addr, worker->s->hostname_ex, - (int)worker->s->port); + backend_addr, conn->hostname, conn->port); backend_addr = backend_addr->next; continue; } @@ -3378,8 +3856,8 @@ PROXY_DECLARE(int) ap_proxy_connect_backend(const char *proxy_function, worker->s->error_time = apr_time_now(); worker->s->status |= PROXY_WORKER_IN_ERROR; ap_log_error(APLOG_MARK, APLOG_ERR, 0, s, APLOGNO(00959) - "ap_proxy_connect_backend disabling worker for (%s:%d) for %" - APR_TIME_T_FMT "s", + "ap_proxy_connect_backend disabling worker for (%s:%hu) " + "for %" APR_TIME_T_FMT "s", worker->s->hostname_ex, (int)worker->s->port, apr_time_sec(worker->s->retry)); } @@ -3473,7 +3951,7 @@ static int proxy_connection_create(const char *proxy_function, * the peer reset the connection already; ap_run_create_connection() * closed the socket */ - ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, + ap_log_error(APLOG_MARK, APLOG_ERR, 0, s, APLOGNO(00960) "%s: an error occurred creating a " "new connection to %pI (%s)", proxy_function, backend_addr, conn->hostname); @@ -3908,7 +4386,7 @@ PROXY_DECLARE(int) ap_proxy_create_hdrbrgd(apr_pool_t *p, apr_bucket *e; int force10 = 0, do_100_continue = 0; conn_rec *origin = p_conn->connection; - const char *host, *val; + const char *host, *creds, *val; proxy_dir_conf *dconf = ap_get_module_config(r->per_dir_config, &proxy_module); /* @@ -4128,6 +4606,11 @@ PROXY_DECLARE(int) ap_proxy_create_hdrbrgd(apr_pool_t *p, apr_table_unset(r->headers_in, "If-None-Match"); } + creds = apr_table_get(r->notes, "proxy-basic-creds"); + if (creds) { + apr_table_mergen(r->headers_in, "Proxy-Authorization", creds); + } + /* run hook to fixup the request we are about to send */ proxy_run_fixups(r); @@ -4726,7 +5209,7 @@ PROXY_DECLARE(apr_status_t) ap_proxy_tunnel_create(proxy_tunnel_rec **ptunnel, { apr_status_t rv; conn_rec *c_i = r->connection; - apr_interval_time_t timeout = -1; + apr_interval_time_t client_timeout = -1, origin_timeout = -1; proxy_tunnel_rec *tunnel; *ptunnel = NULL; @@ -4753,9 +5236,16 @@ PROXY_DECLARE(apr_status_t) ap_proxy_tunnel_create(proxy_tunnel_rec **ptunnel, tunnel->client->bb = apr_brigade_create(c_i->pool, c_i->bucket_alloc); tunnel->client->pfd = &APR_ARRAY_PUSH(tunnel->pfds, apr_pollfd_t); tunnel->client->pfd->p = r->pool; - tunnel->client->pfd->desc_type = APR_POLL_SOCKET; - tunnel->client->pfd->desc.s = ap_get_conn_socket(c_i); + tunnel->client->pfd->desc_type = APR_NO_DESC; + rv = ap_get_pollfd_from_conn(tunnel->client->c, + tunnel->client->pfd, &client_timeout); + if (rv != APR_SUCCESS) { + return rv; + } tunnel->client->pfd->client_data = tunnel->client; + if (tunnel->client->pfd->desc_type == APR_POLL_SOCKET) { + apr_socket_opt_set(tunnel->client->pfd->desc.s, APR_SO_NONBLOCK, 1); + } tunnel->origin->c = c_o; tunnel->origin->name = "origin"; @@ -4765,18 +5255,13 @@ PROXY_DECLARE(apr_status_t) ap_proxy_tunnel_create(proxy_tunnel_rec **ptunnel, tunnel->origin->pfd->desc_type = APR_POLL_SOCKET; tunnel->origin->pfd->desc.s = ap_get_conn_socket(c_o); tunnel->origin->pfd->client_data = tunnel->origin; - - /* Defaults to the biggest timeout of both connections */ - apr_socket_timeout_get(tunnel->client->pfd->desc.s, &timeout); - apr_socket_timeout_get(tunnel->origin->pfd->desc.s, &tunnel->timeout); - if (timeout >= 0 && (tunnel->timeout < 0 || tunnel->timeout < timeout)) { - tunnel->timeout = timeout; - } - - /* We should be nonblocking from now on the sockets */ - apr_socket_opt_set(tunnel->client->pfd->desc.s, APR_SO_NONBLOCK, 1); + apr_socket_timeout_get(tunnel->origin->pfd->desc.s, &origin_timeout); apr_socket_opt_set(tunnel->origin->pfd->desc.s, APR_SO_NONBLOCK, 1); + /* Defaults to the largest timeout of both connections */ + tunnel->timeout = (client_timeout >= 0 && client_timeout > origin_timeout ? + client_timeout : origin_timeout); + /* No coalescing filters */ ap_remove_output_filter_byhandle(c_i->output_filters, "SSL/TLS Coalescing Filter"); @@ -4799,14 +5284,43 @@ PROXY_DECLARE(apr_status_t) ap_proxy_tunnel_create(proxy_tunnel_rec **ptunnel, tunnel->nohalfclose = 1; } - /* Start with POLLOUT and let ap_proxy_tunnel_run() schedule both - * directions when there are no output data pending (anymore). - */ - tunnel->client->pfd->reqevents = APR_POLLOUT | APR_POLLERR; - tunnel->origin->pfd->reqevents = APR_POLLOUT | APR_POLLERR; - if ((rv = apr_pollset_add(tunnel->pollset, tunnel->client->pfd)) - || (rv = apr_pollset_add(tunnel->pollset, tunnel->origin->pfd))) { - return rv; + if (tunnel->client->pfd->desc_type == APR_POLL_SOCKET) { + /* Both ends are sockets, the poll strategy is: + * - poll both sides POLLOUT + * - when one side is writable, remove the POLLOUT + * and add POLLIN to the other side. + * - tunnel arriving data, remove POLLIN from the source + * again and add POLLOUT to the receiving side + * - on EOF on read, remove the POLLIN from that side + * Repeat until both sides are down */ + tunnel->client->pfd->reqevents = APR_POLLOUT | APR_POLLERR; + tunnel->origin->pfd->reqevents = APR_POLLOUT | APR_POLLERR; + if ((rv = apr_pollset_add(tunnel->pollset, tunnel->origin->pfd)) || + (rv = apr_pollset_add(tunnel->pollset, tunnel->client->pfd))) { + return rv; + } + } + else if (tunnel->client->pfd->desc_type == APR_POLL_FILE) { + /* Input is a PIPE fd, the poll strategy is: + * - always POLLIN on origin + * - use socket strategy described above for client only + * otherwise the same + */ + tunnel->client->pfd->reqevents = 0; + tunnel->origin->pfd->reqevents = APR_POLLIN | APR_POLLHUP | + APR_POLLOUT | APR_POLLERR; + if ((rv = apr_pollset_add(tunnel->pollset, tunnel->origin->pfd))) { + return rv; + } + } + else { + /* input is already closed, unsual, but we know nothing about + * the tunneled protocol. */ + tunnel->client->down_in = 1; + tunnel->origin->pfd->reqevents = APR_POLLIN | APR_POLLHUP; + if ((rv = apr_pollset_add(tunnel->pollset, tunnel->origin->pfd))) { + return rv; + } } *ptunnel = tunnel; @@ -4918,7 +5432,23 @@ static int proxy_tunnel_forward(proxy_tunnel_rec *tunnel, } del_pollset(tunnel->pollset, in->pfd, APR_POLLIN); - add_pollset(tunnel->pollset, out->pfd, APR_POLLOUT); + if (out->pfd->desc_type == APR_POLL_SOCKET) { + /* if the output is a SOCKET, we can stop polling the input + * until the output signals POLLOUT again. */ + add_pollset(tunnel->pollset, out->pfd, APR_POLLOUT); + } + else { + /* We can't use POLLOUT in this direction for the only + * APR_POLL_FILE case we have so far (mod_h2's "signal" pipe), + * we assume that the client's ouput filters chain will block/flush + * if necessary (i.e. no pending data), hence that the origin + * is EOF when reaching here. This direction is over. */ + ap_assert(in->down_in && APR_STATUS_IS_EOF(rv)); + ap_log_rerror(APLOG_MARK, APLOG_TRACE3, 0, tunnel->r, + "proxy: %s: %s write shutdown", + tunnel->scheme, out->name); + out->down_out = 1; + } } return OK; @@ -5107,4 +5637,14 @@ void proxy_util_register_hooks(apr_pool_t *p) APR_REGISTER_OPTIONAL_FN(ap_proxy_retry_worker); APR_REGISTER_OPTIONAL_FN(ap_proxy_clear_connection); APR_REGISTER_OPTIONAL_FN(proxy_balancer_get_best_worker); + + { + apr_time_t *start_time = ap_retained_data_get("proxy_start_time"); + if (start_time == NULL) { + start_time = ap_retained_data_create("proxy_start_time", + sizeof(*start_time)); + *start_time = apr_time_now(); + } + proxy_start_time = start_time; + } } |