[src/netbsd-8]: src/sys/kern Pull up following revision(s) (requested by mlel...

[martin <martin%NetBSD.org@localhost>]

details:   https://anonhg.NetBSD.org/src/rev/cd8143270141
branches:  netbsd-8
changeset: 745640:cd8143270141
user:      martin <martin%NetBSD.org@localhost>
date:      Sun Mar 08 09:47:28 2020 +0000

description:
Pull up following revision(s) (requested by mlelstv in ticket #1515):

        sys/kern/sys_select.c: revision 1.42-1.45

PR/54158: Anthony Mallet: poll(2) does not allow polling all possible fds
(hardcoded limit to 1000 + #<open-fds>). Changed to limit by the max of
the resource limit of open descriptors and the above.

Remove the slop code. Suggested by mrg@

Use the max limit (aka maxfiles or the moral equivalent of OPEN_MAX) which
makes poll(2) align with the Posix documentation (which allows EINVAL if
nfds > OPEN_MAX). From: Anthony Mallet

Add slop of 1000 and explain why.

diffstat:

 sys/kern/sys_select.c |  25 +++++++++++++++++++------
 1 files changed, 19 insertions(+), 6 deletions(-)

diffs (51 lines):

diff -r cf17f5aa1301 -r cd8143270141 sys/kern/sys_select.c
--- a/sys/kern/sys_select.c     Sun Mar 08 09:42:29 2020 +0000
+++ b/sys/kern/sys_select.c     Sun Mar 08 09:47:28 2020 +0000
@@ -1,4 +1,4 @@
-/*     $NetBSD: sys_select.c,v 1.40 2017/06/01 02:45:13 chs Exp $      */
+/*     $NetBSD: sys_select.c,v 1.40.2.1 2020/03/08 09:47:28 martin Exp $       */
 
 /*-
  * Copyright (c) 2007, 2008, 2009, 2010 The NetBSD Foundation, Inc.
@@ -84,7 +84,7 @@
  */
 
 #include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: sys_select.c,v 1.40 2017/06/01 02:45:13 chs Exp $");
+__KERNEL_RCSID(0, "$NetBSD: sys_select.c,v 1.40.2.1 2020/03/08 09:47:28 martin Exp $");
 
 #include <sys/param.h>
 #include <sys/systm.h>
@@ -488,15 +488,28 @@
        int             error;
        size_t          ni;
 
-       if (nfds > 1000 + curlwp->l_fd->fd_dt->dt_nfiles) {
+       if (nfds > curlwp->l_proc->p_rlimit[RLIMIT_NOFILE].rlim_max + 1000) {
                /*
-                * Either the user passed in a very sparse 'fds' or junk!
-                * The kmem_alloc() call below would be bad news.
-                * We could process the 'fds' array in chunks, but that
+                * Prevent userland from causing over-allocation.
+                * Raising the default limit too high can still cause
+                * a lot of memory to be allocated, but this also means
+                * that the file descriptor array will also be large.
+                *
+                * To reduce the memory requirements here, we could 
+                * process the 'fds' array in chunks, but that
                 * is a lot of code that isn't normally useful.
                 * (Or just move the copyin/out into pollscan().)
+                *
                 * Historically the code silently truncated 'fds' to
                 * dt_nfiles entries - but that does cause issues.
+                *
+                * Using the max limit equivalent to sysctl
+                * kern.maxfiles is the moral equivalent of OPEN_MAX
+                * as specified by POSIX.
+                *
+                * We add a slop of 1000 in case the resource limit was
+                * changed after opening descriptors or the same descriptor
+                * was specified more than once.
                 */
                return EINVAL;
        }