Hello, > If fd >= FD_SETSIZE then you can't use select. In BUG section in man select: POSIX allows an implementation to define an upper limit, advertised via the constant FD_SETSIZE, on the range of file descriptors that can be specified in a file descriptor set. The Linux kernel imposes no fixed limit, but the glibc implementation makes fd_set a fixed- size type, with FD_SETSIZE defined as 1024, and the FD_*() macros operating according to that limit. To monitor file descriptors greater than 1023, use poll(2) instead. USB/IP uses poll(). tools/usb/usbip/src/usbipd.c#L567. http://lxr.free-electrons.com/source/tools/usb/usbip/src/usbipd.c#L567 r = ppoll(fds, nsockfd, &timeout, &sigmask); Here, nsockfd is 2 or 1 (listen sockaddr IPv4 and/or IPv6). Also I tested with following code. --- diff -uprN -X ./dontdiff_exc_linux linux-4.8-rc9-33-nofile/tools/usb/usb ip/src/usbipd.c linux-4.8-rc9-33-nofile-test/tools/usb/usbip/src/usbipd. c --- linux-4.8-rc9-33-nofile/tools/usb/usbip/src/usbipd.c 2016-09- 29 19:04:28.761410447 +0900 +++ linux-4.8-rc9-33-nofile-test/tools/usb/usbip/src/usbipd.c 2016-10- 03 10:55:53.120517533 +0900 @@ -183,6 +183,28 @@ static void addrinfo_to_text(struct addr snprintf(buf, buf_size, "%s:%s", hbuf, sbuf); } +FILE *tmps[FD_SETSIZE]; + +static void open_test_files(void) +{ + int i; + + for (i = 0; i < FD_SETSIZE; i++) { + tmps[i] = tmpfile(); + if (tmps[i] == NULL) + err("alloc tmp"); + } +} + +static void close_test_files(void) +{ + int i; + + for (i = 0; i < FD_SETSIZE; i++) { + fclose(tmps[i]); + } +} + static int listen_all_addrinfo(struct addrinfo *ai_head, int sockfdlist[], int maxsockfd) { @@ -191,6 +213,8 @@ static int listen_all_addrinfo(struct ad const size_t ai_buf_size = NI_MAXHOST + NI_MAXSERV + 2; char ai_buf[ai_buf_size]; + open_test_files(); + for (ai = ai_head; ai && nsockfd < maxsockfd; ai = ai->ai_next) { int sock; +static void close_test_files(void) +{ + int i; + + for (i = 0; i < FD_SETSIZE; i++) { + fclose(tmps[i]); + } +} + static int listen_all_addrinfo(struct addrinfo *ai_head, int sockfdlist[], int maxsockfd) { @@ -191,6 +213,8 @@ static int listen_all_addrinfo(struct ad const size_t ai_buf_size = NI_MAXHOST + NI_MAXSERV + 2; char ai_buf[ai_buf_size]; + open_test_files(); + for (ai = ai_head; ai && nsockfd < maxsockfd; ai = ai->ai_next) { int sock; Settings in /etc/security/limits.conf --- root soft nofile 65536 root hard nofile 65536 $ cat /proc/PID/limits --- Max open files 65536 65536 files USB/IP log --- usbip: debug: usbipd.c:222:[listen_all_addrinfo] opening 0.0.0.0:3240 usbip: info: socket 1027 : 1024 usbip: debug: usbip_network.c:253:[usbip_net_set_v6only] setsockopt: IPV 6_V6ONLY usbip: info: listening on 0.0.0.0:3240 usbip: debug: usbipd.c:222:[listen_all_addrinfo] opening :::3240 usbip: info: socket 1028 : 1024 usbip: info: listening on :::3240 Best Regards, n.iwata // > -----Original Message----- > From: David Laight [mailto:David.Laight@xxxxxxxxxx] > Sent: Friday, September 30, 2016 8:08 PM > To: fx IWATA NOBUO; valentina.manea.m@xxxxxxxxx; shuah.kh@xxxxxxxxxxx; > gregkh@xxxxxxxxxxxxxxxxxxx; linux-usb@xxxxxxxxxxxxxxx > Cc: fx MICHIMURA TADAO > Subject: RE: [PATCH v1 1/1] usbip: deletion of incorrect socket descriptor > checking > > From: Nobuo Iwata > > Sent: 30 September 2016 06:44 > > This patch removes checking of socket descriptor value in daemons. > > > > It was checked to be less than FD_SETSIZE(1024 usually) but it's not > > correct. > > > > To be exact, the maximum value of descriptor comes from > > rlimit(RLIMIT_NOFILE). > > > > Following kernel code determines the value : > > get_unused_fd_flags() : fs/files.c > > __alloc_fd() : fs/files.c > > expand_files() : fs/files.c > > > > The defalut (soft limit) is defines as INR_OPEN_CUR(1024) in > > include/linux/fs.h which is referenced form INIT_RLIMS in > > include/asm-generic/resource.h. The value may be modified with ulimt, > > sysctl, security configuration and etc. > > > > With the kernel code above, when socket() system call returns positive > > value, the value must be within rlimit(RLIMIT_NOFILE). No extra > > checking is needed when socket() returns positive. > > I'm guessing that the problem is that the code wants to use select() on > the socket. > The value of RLIMIT_NOFILE isn't really relevant. > If fd >= FD_SETSIZE then you can't use select. > > In reality any program that has anywhere near that number of open fd would > be better off using something other than select(). > > David -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
