OS-6364 lx: USER_HZ of 1000 breaks broken code which assumes 100 HZ

Reviewed by: Patrick Mooney <patrick.mooney@joyent.com>
Approved by: Patrick Mooney <patrick.mooney@joyent.com>

11 files changed, 170 insertions, 34 deletions

diff --git a/usr/src/lib/brand/lx/lx_brand/common/lx_brand.c b/usr/src/lib/brand/lx/lx_brand/common/lx_brand.c
index 99a0e64f01..fea15349c7 100644
--- a/usr/src/lib/brand/lx/lx_brand/common/lx_brand.c
+++ b/usr/src/lib/brand/lx/lx_brand/common/lx_brand.c
@@ -73,6 +73,7 @@
 #include <sys/lx_syscall.h>
 #include <sys/lx_thread.h>
 #include <lx_auxv.h>
+#include <sys/lx_userhz.h>
 
 /*
  * There is a block comment in "uts/common/brand/lx/os/lx_brand.c" that
@@ -135,6 +136,8 @@ pid_t zoneinit_pid;		/* zone init PID */
 
 thread_key_t lx_tsd_key;
 
+uint_t lx_hz_scale;		/* USER_HZ scaling factor */
+
 int
 uucopy_unsafe(const void *src, void *dst, size_t n)
 {
@@ -626,6 +629,8 @@ lx_init(int argc, char *argv[], char *envp[])
 	bzero(&reg, sizeof (reg));
 	stack_size = 2 * sysconf(_SC_PAGESIZE);
 
+	lx_hz_scale = sysconf(_SC_CLK_TCK) / LX_USERHZ;
+
 	/*
 	 * We need to shutdown all libc stdio.  libc stdio normally goes to
 	 * file descriptors, but since we're actually part of a linux
@@ -1110,7 +1115,7 @@ static lx_syscall_handler_t lx_handlers[] = {
 	NULL,				/*  97: getrlimit */
 	NULL,				/*  98: getrusage */
 	NULL,				/*  99: sysinfo */
-	lx_times,			/* 100: times */
+	NULL,				/* 100: times */
 	NULL,				/* 101: ptrace */
 	NULL,				/* 102: getuid */
 	NULL,				/* 103: syslog */
@@ -1384,7 +1389,7 @@ static lx_syscall_handler_t lx_handlers[] = {
 	lx_rmdir,			/*  40: rmdir */
 	NULL,				/*  41: dup */
 	NULL,				/*  42: pipe */
-	lx_times,			/*  43: times */
+	NULL,				/*  43: times */
 	NULL,				/*  44: prof */
 	NULL,				/*  45: brk */
 	NULL,				/*  46: setgid16 */
diff --git a/usr/src/lib/brand/lx/lx_brand/common/signal.c b/usr/src/lib/brand/lx/lx_brand/common/signal.c
index 45de7615bc..5c754733cb 100644
--- a/usr/src/lib/brand/lx/lx_brand/common/signal.c
+++ b/usr/src/lib/brand/lx/lx_brand/common/signal.c
@@ -25,7 +25,7 @@
  */
 
 /*
- * Copyright 2016 Joyent, Inc. All rights reserved.
+ * Copyright 2017 Joyent, Inc. All rights reserved.
  */
 
 #include <sys/types.h>
@@ -40,6 +40,7 @@
 #include <sys/lx_sigstack.h>
 #include <sys/lx_syscall.h>
 #include <sys/lx_thread.h>
+#include <sys/lx_userhz.h>
 #include <sys/syscall.h>
 #include <lx_provider_impl.h>
 #include <sys/stack.h>
@@ -473,8 +474,10 @@ stol_siginfo(siginfo_t *siginfop, lx_siginfo_t *lx_siginfop)
 				lx_siginfo.lsi_status = lx_stol_status(
 				    siginfop->si_status, -1);
 			}
-			lx_siginfo.lsi_utime = siginfop->si_utime;
-			lx_siginfo.lsi_stime = siginfop->si_stime;
+			lx_siginfo.lsi_utime =
+			    HZ_TO_LX_USERHZ(siginfop->si_utime);
+			lx_siginfo.lsi_stime =
+			    HZ_TO_LX_USERHZ(siginfop->si_stime);
 			break;
 
 		case LX_SIGILL:
diff --git a/usr/src/lib/brand/lx/lx_brand/common/time.c b/usr/src/lib/brand/lx/lx_brand/common/time.c
index c810aa33f1..db63c5643c 100644
--- a/usr/src/lib/brand/lx/lx_brand/common/time.c
+++ b/usr/src/lib/brand/lx/lx_brand/common/time.c
@@ -22,7 +22,7 @@
 /*
  * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
- * Copyright 2014 Joyent, Inc.  All rights reserved.
+ * Copyright 2017 Joyent, Inc.  All rights reserved.
  */
 
 #include <errno.h>
@@ -34,25 +34,6 @@
 #include <sys/lx_misc.h>
 
 /*
- * times() - The Linux implementation avoids writing to NULL, while Illumos
- *	     returns EFAULT.
- */
-long
-lx_times(uintptr_t p1)
-{
-	clock_t ret;
-	struct tms buf, *tp = (struct tms *)p1;
-
-	ret = times(&buf);
-
-	if ((ret == -1) ||
-	    ((tp != NULL) && uucopy((void *)&buf, tp, sizeof (buf)) != 0))
-		return (-errno);
-
-	return ((ret == -1) ? -errno : ret);
-}
-
-/*
  * setitimer() - the Linux implementation can handle tv_usec values greater
  *		 than 1,000,000 where Illumos would return EINVAL.
  *
diff --git a/usr/src/lib/brand/lx/lx_brand/sys/lx_syscall.h b/usr/src/lib/brand/lx/lx_brand/sys/lx_syscall.h
index 2c8db5cd32..1ea278764f 100644
--- a/usr/src/lib/brand/lx/lx_brand/sys/lx_syscall.h
+++ b/usr/src/lib/brand/lx/lx_brand/sys/lx_syscall.h
@@ -97,7 +97,6 @@ extern long lx_setgroups16(uintptr_t, uintptr_t);
 extern long lx_query_module(uintptr_t, uintptr_t, uintptr_t, uintptr_t,
     uintptr_t);
 
-extern long lx_times(uintptr_t);
 extern long lx_setitimer(uintptr_t, uintptr_t, uintptr_t);
 
 extern long lx_clone(uintptr_t, uintptr_t, uintptr_t, uintptr_t, uintptr_t);
diff --git a/usr/src/uts/common/brand/lx/os/lx_brand.c b/usr/src/uts/common/brand/lx/os/lx_brand.c
index 3646df26e5..a691b620b3 100644
--- a/usr/src/uts/common/brand/lx/os/lx_brand.c
+++ b/usr/src/uts/common/brand/lx/os/lx_brand.c
@@ -150,6 +150,7 @@
 #include <sys/lx_futex.h>
 #include <sys/lx_brand.h>
 #include <sys/lx_types.h>
+#include <sys/lx_userhz.h>
 #include <sys/param.h>
 #include <sys/termios.h>
 #include <sys/sunddi.h>
@@ -178,6 +179,7 @@
 #include <inet/udp_impl.h>
 
 int	lx_debug = 0;
+uint_t	lx_hz_scale = 0;
 
 void	lx_init_brand_data(zone_t *, kmutex_t *);
 void	lx_free_brand_data(zone_t *);
@@ -2435,7 +2437,8 @@ lx_elfexec(struct vnode *vp, struct execa *uap, struct uarg *args,
 
 	/*
 	 * We try to keep /proc's view of the aux vector consistent with
-	 * what's on the process stack.
+	 * what's on the process stack. See the comment on the lx_times
+	 * syscall for an explanation of the hardcoded LX_USERHZ.
 	 */
 	if (args->to_model == DATAMODEL_NATIVE) {
 		auxv_t phdr_auxv[4] = {
@@ -2446,7 +2449,7 @@ lx_elfexec(struct vnode *vp, struct execa *uap, struct uarg *args,
 		};
 		phdr_auxv[0].a_un.a_val = edp.ed_phdr;
 		phdr_auxv[1].a_un.a_val = ldaddr;
-		phdr_auxv[2].a_un.a_val = hz;
+		phdr_auxv[2].a_un.a_val = LX_USERHZ;
 		phdr_auxv[3].a_un.a_val = lxpd->l_vdso;
 
 		if (copyout(&phdr_auxv, args->auxp_brand,
@@ -2573,6 +2576,10 @@ _init(void)
 {
 	int err = 0;
 
+	/* Initialize USER_HZ scaling factor */
+	ASSERT(hz >= LX_USERHZ);
+	lx_hz_scale = hz / LX_USERHZ;
+
 	lx_syscall_init();
 	lx_pid_init();
 	lx_ioctl_init();
diff --git a/usr/src/uts/common/brand/lx/os/lx_misc.c b/usr/src/uts/common/brand/lx/os/lx_misc.c
index 05203c832d..53963c0f7d 100644
--- a/usr/src/uts/common/brand/lx/os/lx_misc.c
+++ b/usr/src/uts/common/brand/lx/os/lx_misc.c
@@ -40,6 +40,7 @@
 #include <sys/lx_siginfo.h>
 #include <sys/lx_futex.h>
 #include <lx_errno.h>
+#include <sys/lx_userhz.h>
 #include <sys/cmn_err.h>
 #include <sys/siginfo.h>
 #include <sys/contract/process_impl.h>
@@ -713,6 +714,10 @@ lx_winfo(proc_t *pp, k_siginfo_t *ip, struct lx_proc_data *dat)
 	ip->si_ctid = PRCTID(pp);
 	ip->si_zoneid = pp->p_zone->zone_id;
 	ip->si_status = pp->p_wdata;
+	/*
+	 * These siginfo values are converted to USER_HZ in the user-land
+	 * brand signal code.
+	 */
 	ip->si_stime = pp->p_stime;
 	ip->si_utime = pp->p_utime;
 }
@@ -961,8 +966,8 @@ stol_ksiginfo_copyout(k_siginfo_t *sip, void *ulxsip)
 			lsi.lsi_status = lx_stol_status(sip->si_status,
 			    SIGKILL);
 		}
-		lsi.lsi_utime = sip->si_utime;
-		lsi.lsi_stime = sip->si_stime;
+		lsi.lsi_utime = HZ_TO_LX_USERHZ(sip->si_utime);
+		lsi.lsi_stime = HZ_TO_LX_USERHZ(sip->si_stime);
 		break;
 
 	case LX_SIGILL:
@@ -1009,8 +1014,8 @@ stol_ksiginfo32_copyout(k_siginfo_t *sip, void *ulxsip)
 			lsi.lsi_status = lx_stol_status(sip->si_status,
 			    SIGKILL);
 		}
-		lsi.lsi_utime = sip->si_utime;
-		lsi.lsi_stime = sip->si_stime;
+		lsi.lsi_utime = HZ_TO_LX_USERHZ(sip->si_utime);
+		lsi.lsi_stime = HZ_TO_LX_USERHZ(sip->si_stime);
 		break;
 
 	case LX_SIGILL:
diff --git a/usr/src/uts/common/brand/lx/os/lx_syscall.c b/usr/src/uts/common/brand/lx/os/lx_syscall.c
index ccabe9aa0d..d308d71661 100644
--- a/usr/src/uts/common/brand/lx/os/lx_syscall.c
+++ b/usr/src/uts/common/brand/lx/os/lx_syscall.c
@@ -562,7 +562,7 @@ lx_sysent_t lx_sysent32[] = {
 	{"rmdir",	NULL,			0,		1}, /* 40 */
 	{"dup",		lx_dup,			0,		1}, /* 41 */
 	{"pipe",	lx_pipe,		0,		1}, /* 42 */
-	{"times",	NULL,			0,		1}, /* 43 */
+	{"times",	lx_times,		0,		1}, /* 43 */
 	{"prof",	NULL,			NOSYS_OBSOLETE,	0}, /* 44 */
 	{"brk",		lx_brk,			0,		1}, /* 45 */
 	{"setgid16",	lx_setgid16,		0,		1}, /* 46 */
@@ -990,7 +990,7 @@ lx_sysent_t lx_sysent64[] = {
 	{"getrlimit",	lx_getrlimit,		0,		2}, /* 97 */
 	{"getrusage",	lx_getrusage,		0,		2}, /* 98 */
 	{"sysinfo",	lx_sysinfo64,		0,		1}, /* 99 */
-	{"times",	NULL,			0,		1}, /* 100 */
+	{"times",	lx_times,		0,		1}, /* 100 */
 	{"ptrace",	lx_ptrace,		0,		4}, /* 101 */
 	{"getuid",	lx_getuid,		0,		0}, /* 102 */
 	{"syslog",	lx_syslog,		0,		3}, /* 103 */
diff --git a/usr/src/uts/common/brand/lx/sys/lx_syscalls.h b/usr/src/uts/common/brand/lx/sys/lx_syscalls.h
index 54f11a71f2..20b0c2ed55 100644
--- a/usr/src/uts/common/brand/lx/sys/lx_syscalls.h
+++ b/usr/src/uts/common/brand/lx/sys/lx_syscalls.h
@@ -254,6 +254,7 @@ extern long lx_syslog();
 extern long lx_removexattr();
 extern long lx_tgkill();
 extern long lx_time();
+extern long lx_times();
 extern long lx_timer_create();
 extern long lx_tkill();
 extern long lx_umask();
diff --git a/usr/src/uts/common/brand/lx/sys/lx_userhz.h b/usr/src/uts/common/brand/lx/sys/lx_userhz.h
new file mode 100644
index 0000000000..7f273744c8
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/sys/lx_userhz.h
@@ -0,0 +1,62 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2017 Joyent, Inc.
+ */
+
+#ifndef _LX_USERHZ_H
+#define	_LX_USERHZ_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/*
+ * Within the kernel, Linux implements an internal hz that they refer to as a
+ * "jiffy". Linux can be built with different hz, but on modern kernels
+ * it is frequently 250. However, Linux has a separate concept for the hz
+ * that is visible outside the kernel. This is called "USER_HZ" and is the
+ * value returned by 'sysconf(_SC_CLK_TCK)'. This is almost universally set to
+ * 100hz. Some (lazy) applications just hardcode 100hz instead of checking.
+ * To accommodate these broken applications, we always work with a USER_HZ of
+ * 100 and scale accordingly. See the Linux time(7) man page for a more
+ * detailed discussion of their behavior. See the comment in our
+ * uts/common/conf/param.c for a discussion of valid native hz values.
+ *
+ * There are a few interfaces which expose a clock_t to user-land and which
+ * need to be considered for USER_HZ adjustment.
+ * 1) The times(2) syscall. This is handled correctly.
+ * 2) The waitid(2) syscall passes a siginfo_t which contains si_stime and
+ *    si_utime. Testing waitid(2) on various Linux distributions shows that the
+ *    these fields are garbage. This aligns with the Linux waitid(2) man page,
+ *    which describes the subset of the siginfo_t structure that is populated.
+ *    Neither si_stime or si_utime are listed.
+ * 3) A sigaction(2) handler can pass a siginfo_t. This is only documented to
+ *    occur when the sa_flags is SA_SIGINFO. The si_stime and si_utime are
+ *    documented to only be populated when the signal is SIGCHLD. However,
+ *    testing on Linux seems to show that these fields are not consistent
+ *    with the corresponding times(2) data for the process, even for the
+ *    SIGCHLD sigaction handler case.
+ *
+ * Although the siginfo_t si_stime and si_utime data for cases #2 and #3 is not
+ * consistent on Linux, we populate these fields correctly to be on the safe
+ * side.
+ */
+extern uint_t lx_hz_scale;
+#define	LX_USERHZ		100
+#define	HZ_TO_LX_USERHZ(x)	((x) / lx_hz_scale)
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* _LX_USERHZ_H */
diff --git a/usr/src/uts/common/brand/lx/syscall/lx_time.c b/usr/src/uts/common/brand/lx/syscall/lx_time.c
new file mode 100644
index 0000000000..b9bc8e5ab4
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/syscall/lx_time.c
@@ -0,0 +1,72 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2017, Joyent, Inc.
+ */
+
+#include <sys/types.h>
+#include <sys/systm.h>
+#include <sys/thread.h>
+#include <sys/proc.h>
+#include <sys/times.h>
+#include <sys/msacct.h>
+#include <sys/lx_userhz.h>
+
+/* See the comment on LX_USERHZ for more details. */
+#define	LX_NSEC_PER_USERHZ	(NANOSEC / LX_USERHZ)
+#define	NSEC_TO_LX_USERHZ(nsec)	((nsec) / LX_NSEC_PER_USERHZ)
+
+/*
+ * Our times(2) implementation is based on the native times(2), but with
+ * the necessary scaling to adjust to USER_HZ. Also, Linux avoids writing
+ * to a NULL tp, whereas our native code returns EFAULT.
+ */
+long
+lx_times(struct tms *tp)
+{
+	proc_t *p = curproc;
+	struct tms p_time;
+	clock_t ret_lbolt;
+
+	mutex_enter(&p->p_lock);
+	p_time.tms_utime =
+	    (clock_t)NSEC_TO_LX_USERHZ(mstate_aggr_state(p, LMS_USER));
+	p_time.tms_stime =
+	    (clock_t)NSEC_TO_LX_USERHZ(mstate_aggr_state(p, LMS_SYSTEM));
+	p_time.tms_cutime = HZ_TO_LX_USERHZ(p->p_cutime);
+	p_time.tms_cstime = HZ_TO_LX_USERHZ(p->p_cstime);
+	mutex_exit(&p->p_lock);
+
+#ifdef _SYSCALL32_IMPL
+	if (get_udatamodel() != DATAMODEL_NATIVE) {
+		struct tms32 t32;
+
+		t32.tms_utime = p_time.tms_utime;
+		t32.tms_stime = p_time.tms_stime;
+		t32.tms_cutime = p_time.tms_cutime;
+		t32.tms_cstime = p_time.tms_cstime;
+
+		if (tp != NULL && copyout(&t32, tp, sizeof (t32)) != 0)
+			return (set_errno(EFAULT));
+
+		ret_lbolt = ddi_get_lbolt();
+		return ((clock32_t)HZ_TO_LX_USERHZ(ret_lbolt));
+	} else
+#endif /* _SYSCALL32_IMPL */
+	{
+		if (tp != NULL && copyout(&p_time, tp, sizeof (p_time)) != 0)
+			return (set_errno(EFAULT));
+
+		ret_lbolt = ddi_get_lbolt();
+		return (HZ_TO_LX_USERHZ(ret_lbolt));
+	}
+}
diff --git a/usr/src/uts/intel/Makefile.files b/usr/src/uts/intel/Makefile.files
index 8f8a70d22f..f118f1f157 100644
--- a/usr/src/uts/intel/Makefile.files
+++ b/usr/src/uts/intel/Makefile.files
@@ -359,6 +359,7 @@ LX_BRAND_OBJS  =		\
 	lx_syscall.o		\
 	lx_sysinfo.o		\
 	lx_thread_area.o	\
+	lx_time.o		\
 	lx_timer.o		\
 	lx_umask.o		\
 	lx_uname.o		\