diff --git a/Misc/NEWS b/Misc/NEWS --- a/Misc/NEWS +++ b/Misc/NEWS @@ -214,6 +214,11 @@ Library ------- +- Issue #9090: When a socket with a timeout fails with EWOULDBLOCK or EAGAIN, + retry the select() loop instead of bailing out. This is because select() + can incorrectly report a socket as ready for reading (for example, if it + received some data with an invalid checksum). + - Issue #1285086: Get rid of the refcounting hack and speed up urllib.unquote(). - Issue #17368: Fix an off-by-one error in the Python JSON decoder that caused diff --git a/Modules/socketmodule.c b/Modules/socketmodule.c --- a/Modules/socketmodule.c +++ b/Modules/socketmodule.c @@ -473,6 +473,17 @@ return NULL; } +#ifdef MS_WINDOWS +#ifndef WSAEAGAIN +#define WSAEAGAIN WSAEWOULDBLOCK +#endif +#define CHECK_ERRNO(expected) \ + (WSAGetLastError() == WSA ## expected) +#else +#define CHECK_ERRNO(expected) \ + (errno == expected) +#endif + /* Convenience function to raise an error according to errno and return a NULL pointer from a function. */ @@ -661,7 +672,7 @@ after they've reacquired the interpreter lock. Returns 1 on timeout, -1 on error, 0 otherwise. */ static int -internal_select(PySocketSockObject *s, int writing) +internal_select_ex(PySocketSockObject *s, int writing, double interval) { int n; @@ -673,6 +684,10 @@ if (s->sock_fd < 0) return 0; + /* Handling this condition here simplifies the select loops */ + if (interval < 0.0) + return 1; + /* Prefer poll, if available, since you can poll() any fd * which can't be done with select(). */ #ifdef HAVE_POLL @@ -684,7 +699,7 @@ pollfd.events = writing ? POLLOUT : POLLIN; /* s->sock_timeout is in seconds, timeout in ms */ - timeout = (int)(s->sock_timeout * 1000 + 0.5); + timeout = (int)(interval * 1000 + 0.5); n = poll(&pollfd, 1, timeout); } #else @@ -692,8 +707,8 @@ /* Construct the arguments to select */ fd_set fds; struct timeval tv; - tv.tv_sec = (int)s->sock_timeout; - tv.tv_usec = (int)((s->sock_timeout - tv.tv_sec) * 1e6); + tv.tv_sec = (int)interval; + tv.tv_usec = (int)((interval - tv.tv_sec) * 1e6); FD_ZERO(&fds); FD_SET(s->sock_fd, &fds); @@ -712,6 +727,49 @@ return 0; } +static int +internal_select(PySocketSockObject *s, int writing) +{ + return internal_select_ex(s, writing, s->sock_timeout); +} + +/* + Two macros for automatic retry of select() in case of false positives + (for example, select() could indicate a socket is ready for reading + but the data then discarded by the OS because of a wrong checksum). + Here is an example of use: + + BEGIN_SELECT_LOOP(s) + Py_BEGIN_ALLOW_THREADS + timeout = internal_select_ex(s, 0, interval); + if (!timeout) + outlen = recv(s->sock_fd, cbuf, len, flags); + Py_END_ALLOW_THREADS + if (timeout == 1) { + PyErr_SetString(socket_timeout, "timed out"); + return -1; + } + END_SELECT_LOOP(s) +*/ +PyAPI_FUNC(double) _PyTime_floattime(void); /* defined in timemodule.c */ +#define BEGIN_SELECT_LOOP(s) \ + { \ + double deadline, interval = s->sock_timeout; \ + int has_timeout = s->sock_timeout > 0.0; \ + if (has_timeout) { \ + deadline = _PyTime_floattime() + s->sock_timeout; \ + } \ + while (1) { \ + errno = 0; \ + +#define END_SELECT_LOOP(s) \ + if (!has_timeout || \ + (!CHECK_ERRNO(EWOULDBLOCK) && !CHECK_ERRNO(EAGAIN))) \ + break; \ + interval = deadline - _PyTime_floattime(); \ + } \ + } \ + /* Initialize a new socket object. */ static double defaulttimeout = -1.0; /* Default timeout for new sockets */ @@ -1656,8 +1714,9 @@ if (!IS_SELECTABLE(s)) return select_error(); + BEGIN_SELECT_LOOP(s) Py_BEGIN_ALLOW_THREADS - timeout = internal_select(s, 0); + timeout = internal_select_ex(s, 0, interval); if (!timeout) newfd = accept(s->sock_fd, SAS2SA(&addrbuf), &addrlen); Py_END_ALLOW_THREADS @@ -1666,6 +1725,7 @@ PyErr_SetString(socket_timeout, "timed out"); return NULL; } + END_SELECT_LOOP(s) #ifdef MS_WINDOWS if (newfd == INVALID_SOCKET) @@ -2355,8 +2415,9 @@ } #ifndef __VMS + BEGIN_SELECT_LOOP(s) Py_BEGIN_ALLOW_THREADS - timeout = internal_select(s, 0); + timeout = internal_select_ex(s, 0, interval); if (!timeout) outlen = recv(s->sock_fd, cbuf, len, flags); Py_END_ALLOW_THREADS @@ -2365,6 +2426,7 @@ PyErr_SetString(socket_timeout, "timed out"); return -1; } + END_SELECT_LOOP(s) if (outlen < 0) { /* Note: the call to errorhandler() ALWAYS indirectly returned NULL, so ignore its return value */ @@ -2386,8 +2448,9 @@ segment = remaining; } + BEGIN_SELECT_LOOP(s) Py_BEGIN_ALLOW_THREADS - timeout = internal_select(s, 0); + timeout = internal_select_ex(s, 0, interval); if (!timeout) nread = recv(s->sock_fd, read_buf, segment, flags); Py_END_ALLOW_THREADS @@ -2396,6 +2459,8 @@ PyErr_SetString(socket_timeout, "timed out"); return -1; } + END_SELECT_LOOP(s) + if (nread < 0) { s->errorhandler(); return -1; @@ -2559,9 +2624,10 @@ return -1; } + BEGIN_SELECT_LOOP(s) Py_BEGIN_ALLOW_THREADS memset(&addrbuf, 0, addrlen); - timeout = internal_select(s, 0); + timeout = internal_select_ex(s, 0, interval); if (!timeout) { #ifndef MS_WINDOWS #if defined(PYOS_OS2) && !defined(PYCC_GCC) @@ -2582,6 +2648,7 @@ PyErr_SetString(socket_timeout, "timed out"); return -1; } + END_SELECT_LOOP(s) if (n < 0) { s->errorhandler(); return -1; @@ -2719,8 +2786,9 @@ buf = pbuf.buf; len = pbuf.len; + BEGIN_SELECT_LOOP(s) Py_BEGIN_ALLOW_THREADS - timeout = internal_select(s, 1); + timeout = internal_select_ex(s, 1, interval); if (!timeout) #ifdef __VMS n = sendsegmented(s->sock_fd, buf, len, flags); @@ -2728,13 +2796,14 @@ n = send(s->sock_fd, buf, len, flags); #endif Py_END_ALLOW_THREADS - - PyBuffer_Release(&pbuf); - if (timeout == 1) { + PyBuffer_Release(&pbuf); PyErr_SetString(socket_timeout, "timed out"); return NULL; } + END_SELECT_LOOP(s) + + PyBuffer_Release(&pbuf); if (n < 0) return s->errorhandler(); return PyInt_FromLong((long)n); @@ -2768,8 +2837,9 @@ } do { + BEGIN_SELECT_LOOP(s) Py_BEGIN_ALLOW_THREADS - timeout = internal_select(s, 1); + timeout = internal_select_ex(s, 1, interval); n = -1; if (!timeout) { #ifdef __VMS @@ -2784,6 +2854,7 @@ PyErr_SetString(socket_timeout, "timed out"); return NULL; } + END_SELECT_LOOP(s) /* PyErr_CheckSignals() might change errno */ saved_errno = errno; /* We must run our signal handlers before looping again. @@ -2863,17 +2934,20 @@ return NULL; } + BEGIN_SELECT_LOOP(s) Py_BEGIN_ALLOW_THREADS - timeout = internal_select(s, 1); + timeout = internal_select_ex(s, 1, interval); if (!timeout) n = sendto(s->sock_fd, buf, len, flags, SAS2SA(&addrbuf), addrlen); Py_END_ALLOW_THREADS - - PyBuffer_Release(&pbuf); + if (timeout == 1) { + PyBuffer_Release(&pbuf); PyErr_SetString(socket_timeout, "timed out"); return NULL; } + END_SELECT_LOOP(s) + PyBuffer_Release(&pbuf); if (n < 0) return s->errorhandler(); return PyInt_FromLong((long)n); diff --git a/Modules/timemodule.c b/Modules/timemodule.c --- a/Modules/timemodule.c +++ b/Modules/timemodule.c @@ -1055,3 +1055,10 @@ return 0; } + +/* export floattime to socketmodule.c */ +PyAPI_FUNC(double) +_PyTime_floattime(void) +{ + return floattime(); +}