kernel: Introduce Time Namespace
cover kernel: Introduce Time Namespace
Discussions around time namespace are there for a long time. The first attempt to implement it was in 2006 by Jeff Dike. From that time, the topic appears on and off in various discussions.
There are two main use cases for time namespaces:
1. change date and time inside a container;
2. adjust clocks for a container restored from a checkpoint.
“It seems like this might be one of the last major obstacles keeping migration from being used in production systems, given that not all containers and connections can be migrated as long as a time dependency is capable of messing it up.” (by github.com/dav-ell)
The kernel provides access to several clocks: CLOCK_REALTIME, CLOCK_MONOTONIC, CLOCK_BOOTTIME. Last two clocks are monotonous, but the start points for them are not defined and are different for each system. When a container is migrated from one node to another, all clocks have to be restored into consistent states; in other words, they have to continue running from the same points where they have been dumped.
The main idea of this patch set is adding per-namespace offsets for system clocks. When a process in a non-root time namespace requests time of a clock, a namespace offset is added to the current value of this clock and the sum is returned.
All offsets are placed on a separate page, this allows us to map it as part of VVAR into user processes and use offsets from VDSO calls.
Now offsets are implemented for CLOCK_MONOTONIC and CLOCK_BOOTTIME clocks.
v2: There are two major changes from the previous version:
* Two versions of the VDSO library to avoid a performance penalty for host tasks outside time namespace (as suggested by Andy and Thomas).
As it has been discussed on timens RFC, adding a new conditional branch if (inside_time_ns) on VDSO for all processes is undesirable. It will add a penalty for everybody as branch predictor may mispredict the jump. Also there are instruction cache lines wasted on cmp/jmp.
Those effects of introducing time namespace are very much unwanted having in mind how much work have been spent on micro-optimisation VDSO code.
Addressing those problems, there are two versions of VDSO's .so:
for host tasks (without any penalty) and for processes inside of time namespace with clk_to_ns() that subtracts offsets from host's time.
* Allow to set clock offsets for a namespace only before any processes appear in it.
Now a time namespace looks similar to a pid namespace in a way how it is
created: unshare(CLONE_NEWTIME) system call creates a new time namespace, but doesn't set it to the current process. Then all children of the process will be born in the new time namespace, or a process can use the setns() system call to join a namespace.
This scheme allows to create a new time namespaces, set clock offsets and then populate the namespace with processes.
Our performance measurements show that the price of VDSO's clock_gettime() in a child time namespace is about 8% with a hot CPU cache and about 90% with a cold CPU cache. There is no performance regression for host processes outside time namespace on those tests.
We wrote two small benchmarks. The first one gettime_perf.c calls clock_gettime() in a loop for 3 seconds. It shows us performance with a hot CPU cache (more clock_gettime() cycles - the better):
| before | CONFIG_TIME_NS=n | host | inside timens --------|------------|------------------|-------------|------------- cycles | 139887013 | 139453003 | 139899785 | 128792458 diff (%)| 100 | 99.7 | 100 | 92
The second one gettime_perf_cold.c calls rdtsc, clock_gettime(), rdtsc and shows a difference between second and first rdtsc. The binary is called in a loop 1000 times, then calculate MODE for 1000 values. It should show us performance with a cold CPU cache
(lesser tsc per cycle - the better):
| before | CONFIG_TIME_NS=n | host | inside timens --------|------------|------------------|-------------|------------- tsc | 6748 | 6718 | 6862 | 12682 diff (%)| 100 | 99.6 | 101.7 | 188
The numbers gathered on Intel(R) Core(TM) i5-6300U CPU @ 2.40GHz.
Andrei Vagin (15):
ns: Introduce Time Namespace
timens: Add timens_offsets
timens: Introduce CLOCK_MONOTONIC offsets
timens: Introduce CLOCK_BOOTTIME offset
timerfd/timens: Take into account ns clock offsets posix-timers/timens: Take into account clock offsets timens/kernel: Take into account timens clock offsets in
clock_nanosleep
x86/vdso/timens: Add offsets page in vvar
timens/fs/proc: Introduce /proc/pid/timens_offsets selftest/timens: Add a test for timerfd
selftest/timens: Add a test for clock_nanosleep() selftest/timens: Add timer offsets test
selftests: Add a simple perf test for clock_gettime() selftest/timens: Check that a right vdso is mapped after fork and exec x86/vdso: Align VDSO functions by CPU L1 cache line
Dmitry Safonov (17):
timens: Shift /proc/uptime
x86/vdso2c: Correct err messages on file opening x86/vdso2c: Convert iterator to unsigned x86/vdso/Makefile: Add vobjs32
x86/vdso: Build timens .so(s)
x86/VDSO: Build VDSO with -ffunction-sections x86/vdso2c: Optionally produce linker script for vdso entries x86/vdso: Generate vdso{,32}-timens.lds
x86/vdso2c: Sort vdso entries by addresses for linker script x86/vdso.lds: Align !timens (host's) vdso.so entries x86/vdso2c: Align LOCAL symbols between vdso{-timens,}.so x86/vdso: Initialize timens 64-bit vdso
x86/vdso: Switch image on setns()/unshare()/clone()
timens: Add align for timens_offsets selftest/timens: Add Time Namespace test for supported clocks selftest/timens: Add procfs selftest
x86/vdso: Restrict splitting VVAR VMA
MAINTAINERS | 3 + arch/Kconfig | 5 + arch/x86/Kconfig | 1 + arch/x86/entry/vdso/.gitignore | 2 + arch/x86/entry/vdso/Makefile | 61 ++- arch/x86/entry/vdso/vclock_gettime-timens.c | 6 + arch/x86/entry/vdso/vclock_gettime.c | 42 +++ arch/x86/entry/vdso/vdso-layout.lds.S | 21 +- arch/x86/entry/vdso/vdso-timens.lds.S | 7 + arch/x86/entry/vdso/vdso2c.c | 46 ++- arch/x86/entry/vdso/vdso2c.h | 52 ++- arch/x86/entry/vdso/vdso32/.gitignore | 1 + arch/x86/entry/vdso/vdso32/sigreturn.S | 2 + arch/x86/entry/vdso/vdso32/system_call.S | 2 +- .../entry/vdso/vdso32/vclock_gettime-timens.c | 6 + .../x86/entry/vdso/vdso32/vdso32-timens.lds.S | 8 + arch/x86/entry/vdso/vma.c | 110 ++++++ arch/x86/include/asm/vdso.h | 8 + fs/proc/base.c | 101 +++++ fs/proc/namespaces.c | 4 + fs/proc/uptime.c | 3 + fs/timerfd.c | 16 +- include/linux/nsproxy.h | 2 + include/linux/proc_ns.h | 2 + include/linux/time_namespace.h | 91 +++++ include/linux/timens_offsets.h | 18 + include/linux/user_namespace.h | 1 + include/uapi/linux/sched.h | 1 + init/Kconfig | 8 + kernel/Makefile | 1 + kernel/fork.c | 3 +- kernel/nsproxy.c | 41 ++- kernel/time/hrtimer.c | 8 + kernel/time/posix-timers.c | 24 +- kernel/time/posix-timers.h | 1 + kernel/time_namespace.c | 348 ++++++++++++++++++ tools/testing/selftests/Makefile | 1 + tools/testing/selftests/timens/.gitignore | 7 + tools/testing/selftests/timens/Makefile | 12 + .../selftests/timens/clock_nanosleep.c | 99 +++++ tools/testing/selftests/timens/config | 1 + tools/testing/selftests/timens/exec.c | 91 +++++ tools/testing/selftests/timens/gettime_perf.c | 74 ++++ .../selftests/timens/gettime_perf_cold.c | 63 ++++ tools/testing/selftests/timens/log.h | 26 ++ tools/testing/selftests/timens/procfs.c | 142 +++++++ tools/testing/selftests/timens/timens.c | 191 ++++++++++ tools/testing/selftests/timens/timens.h | 63 ++++ tools/testing/selftests/timens/timer.c | 115 ++++++ tools/testing/selftests/timens/timerfd.c | 119 ++++++ 50 files changed, 2008 insertions(+), 52 deletions(-) create mode 100644 arch/x86/entry/vdso/vclock_gettime-timens.c create mode 100644 arch/x86/entry/vdso/vdso-timens.lds.S create mode 100644 arch/x86/entry/vdso/vdso32/vclock_gettime-timens.c create mode 100644 arch/x86/entry/vdso/vdso32/vdso32-timens.lds.S create mode 100644 include/linux/time_namespace.h create mode 100644 include/linux/timens_offsets.h create mode 100644 kernel/time_namespace.c create mode 100644 tools/testing/selftests/timens/.gitignore create mode 100644 tools/testing/selftests/timens/Makefile create mode 100644 tools/testing/selftests/timens/clock_nanosleep.c create mode 100644 tools/testing/selftests/timens/config create mode 100644 tools/testing/selftests/timens/exec.c create mode 100644 tools/testing/selftests/timens/gettime_perf.c create mode 100644 tools/testing/selftests/timens/gettime_perf_cold.c create mode 100644 tools/testing/selftests/timens/log.h create mode 100644 tools/testing/selftests/timens/procfs.c create mode 100644 tools/testing/selftests/timens/timens.c create mode 100644 tools/testing/selftests/timens/timens.h create mode 100644 tools/testing/selftests/timens/timer.c create mode 100644 tools/testing/selftests/timens/timerfd.c
Signature · 1 line
2.20.1
PATCH 1/32 ns: Introduce Time Namespace
From: Andrei Vagin <[email protected]>
Time Namespace isolates clock values.
The kernel provides access to several clocks CLOCK_REALTIME, CLOCK_MONOTONIC, CLOCK_BOOTTIME, etc.
CLOCK_REALTIME
System-wide clock that measures real (i.e., wall-clock) time.
CLOCK_MONOTONIC
Clock that cannot be set and represents monotonic time since some unspecified starting point.
CLOCK_BOOTTIME
Identical to CLOCK_MONOTONIC, except it also includes any time that the system is suspended.
For many users, the time namespace means the ability to changes date and time in a container (CLOCK_REALTIME).
But in a context of the checkpoint/restore functionality, monotonic and bootime clocks become interesting. Both clocks are monotonic with unspecified staring points. These clocks are widely used to measure time slices and set timers. After restoring or migrating processes, we have to guarantee that they never go backward. In an ideal case, the behavior of these clocks should be the same as for a case when a whole system is suspended. All this means that we need to be able to set CLOCK_MONOTONIC and CLOCK_BOOTTIME clocks, what can be done by adding per-namespace offsets for clocks.
A time namespace is similar to a pid namespace in a way how it is
created: unshare(CLONE_NEWTIME) system call creates a new time namespace, but doesn't set it to the current process. Then all children of the process will be born in the new time namespace, or a process can use the setns() system call to join a namespace.
This scheme allows setting clock offsets for a namespace, before any processes appear in it.
--- MAINTAINERS | 2 + fs/proc/namespaces.c | 4 + include/linux/nsproxy.h | 2 + include/linux/proc_ns.h | 2 + include/linux/time_namespace.h | 69 +++++++++++ include/linux/user_namespace.h | 1 + include/uapi/linux/sched.h | 1 + init/Kconfig | 7 ++ kernel/Makefile | 1 + kernel/fork.c | 3 +- kernel/nsproxy.c | 41 +++++-- kernel/time_namespace.c | 215 +++++++++++++++++++++++++++++++++ 12 files changed, 340 insertions(+), 8 deletions(-) create mode 100644 include/linux/time_namespace.h create mode 100644 kernel/time_namespace.c
MAINTAINERS +2 -0
@@ -12144,6 +12144,8 @@ T: git git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git timers/core S: Maintained F: fs/timerfd.c F: include/linux/timer* +F: include/linux/time_namespace.h +F: kernel/time_namespace.c F: kernel/time/*timer* POWER MANAGEMENT CORE
fs/proc/namespaces.c +4 -0
@@ -33,6 +33,10 @@ static const struct proc_ns_operations *ns_entries[] = {#ifdef CONFIG_CGROUPS &cgroupns_operations,#endif +#ifdef CONFIG_TIME_NS + &timens_operations, + &timens_for_children_operations, +#endif }; static const char *proc_ns_get_link(struct dentry *dentry,
include/linux/nsproxy.h +2 -0
@@ -35,6 +35,8 @@ struct nsproxy {struct mnt_namespace *mnt_ns;struct pid_namespace *pid_ns_for_children;struct net *net_ns; +struct time_namespace *time_ns; +struct time_namespace *time_ns_for_children;struct cgroup_namespace *cgroup_ns; };extern struct nsproxy init_nsproxy;
include/linux/proc_ns.h +2 -0
@@ -32,6 +32,8 @@ extern const struct proc_ns_operations pidns_for_children_operations;extern const struct proc_ns_operations userns_operations;extern const struct proc_ns_operations mntns_operations;extern const struct proc_ns_operations cgroupns_operations; +extern const struct proc_ns_operations timens_operations; +extern const struct proc_ns_operations timens_for_children_operations;/* * We always define these enumerators
include/linux/time_namespace.h new file +69 -0
69 lines not on this page · raw message
include/linux/user_namespace.h +1 -0
@@ -45,6 +45,7 @@ enum ucount_type {UCOUNT_NET_NAMESPACES ,UCOUNT_MNT_NAMESPACES ,UCOUNT_CGROUP_NAMESPACES , +UCOUNT_TIME_NAMESPACES , #ifdef CONFIG_INOTIFY_USERUCOUNT_INOTIFY_INSTANCES , UCOUNT_INOTIFY_WATCHES,
include/uapi/linux/sched.h +1 -0
@@ -10,6 +10,7 @@#define CLONE_FS 0x00000200/* set if fs info shared between processes */ #define CLONE_FILES 0x00000400/* set if open files shared between processes */ #define CLONE_SIGHAND 0x00000800/* set if signal handlers and blocked signals shared */ +#define CLONE_NEWTIME 0x00001000/* New time namespace */ #define CLONE_PTRACE 0x00002000/* set if we want to let tracing continue on the child too */ #define CLONE_VFORK 0x00004000/* set if the parent wants the child to wake it up on mm_release */ #define CLONE_PARENT 0x00008000/* set if we want to have the same parent as the cloner */
init/Kconfig +7 -0
@@ -965,6 +965,13 @@ config UTS_NS In this namespace tasks see different info provided with the uname() system call +config TIME_NS + bool "TIME namespace" + default y + help + In this namespace boottime and monotonic clocks can be set. + The time will keep going with the same pace. + config IPC_NS bool "IPC namespace" depends on (SYSVIPC || POSIX_MQUEUE)
kernel/Makefile +1 -0
@@ -67,6 +67,7 @@ obj-$(CONFIG_BACKTRACE_SELF_TEST) += backtracetest.o obj-$(CONFIG_COMPAT) += compat.o obj-$(CONFIG_CGROUPS) += cgroup/ obj-$(CONFIG_UTS_NS) += utsname.o +obj-$(CONFIG_TIME_NS) += time_namespace.o obj-$(CONFIG_USER_NS) += user_namespace.o obj-$(CONFIG_PID_NS) += pid_namespace.o obj-$(CONFIG_IKCONFIG) += configs.o
kernel/fork.c +2 -1
@@ -2426,7 +2426,8 @@ static int check_unshare_flags(unsigned long unshare_flags)if (unshare_flags & ~(CLONE_THREAD |CLONE_FS |CLONE_NEWNS |CLONE_SIGHAND |CLONE_VM |CLONE_FILES |CLONE_SYSVSEM |CLONE_NEWUTS |CLONE_NEWIPC |CLONE_NEWNET | -CLONE_NEWUSER |CLONE_NEWPID |CLONE_NEWCGROUP )) +CLONE_NEWUSER |CLONE_NEWPID |CLONE_NEWCGROUP | +CLONE_NEWTIME ))return -EINVAL ;/* * Not implemented, but pretend it works if there is nothing
kernel/nsproxy.c +34 -7
88 lines not on this page · raw message
kernel/time_namespace.c new file +215 -0
215 lines not on this page · raw message
Signature · 1 line
2.20.1
PATCH 2/32 timens: Add timens_offsets
In reply to Dmitry SafonovFrom: Andrei Vagin <[email protected]>
Introduce offsets for time namespace. They will contain an adjustment needed to convert clocks to/from host's.
Allocate one page for each time namespace that will be premapped into userspace among vvar pages.
--- MAINTAINERS | 1 + include/linux/time_namespace.h | 1 + include/linux/timens_offsets.h | 8 ++++++++ kernel/time_namespace.c | 14 ++++++++++++-- 4 files changed, 22 insertions(+), 2 deletions(-) create mode 100644 include/linux/timens_offsets.h
MAINTAINERS +1 -0
@@ -12145,6 +12145,7 @@ S: Maintained F: fs/timerfd.c F: include/linux/timer* F: include/linux/time_namespace.h +F: include/linux/timens_offsets.h F: kernel/time_namespace.c F: kernel/time/*timer*
include/linux/time_namespace.h +1 -0
@@ -8,6 +8,7 @@#include <linux/nsproxy.h> #include <linux/ns_common.h> #include <linux/err.h> +#include <linux/timens_offsets.h> struct user_namespace ;extern struct user_namespace init_user_ns;
include/linux/timens_offsets.h new file +8 -0
@@ -0,0 +1,8 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +#ifndef _LINUX_TIME_OFFSETS_H +#define _LINUX_TIME_OFFSETS_H + +struct timens_offsets { +}; + +#endif
kernel/time_namespace.c +12 -2
@@ -13,6 +13,7 @@#include <linux/user_namespace.h> #include <linux/proc_ns.h> #include <linux/sched/task.h> +#include <linux/mm.h> static struct ucounts *inc_time_namespaces (struct user_namespace *ns) {
@@ -46,6 +47,7 @@ static struct time_namespace *clone_time_ns(struct user_namespace *user_ns, {struct time_namespace *ns;struct ucounts *ucounts; +struct page *page;int err; err = -ENOSPC ;
@@ -58,15 +60,22 @@ static struct time_namespace *clone_time_ns(struct user_namespace *user_ns,if (!ns) goto fail_dec; + page =alloc_page (GFP_KERNEL | __GFP_ZERO); +if (!page) + goto fail_free; + ns->offsets =page_address (page); +BUILD_BUG_ON (sizeof (*ns->offsets) >PAGE_SIZE ); + err =ns_alloc_inum (&ns->ns);if (err) - goto fail_free; + goto fail_page; ns->ucounts = ucounts; ns->ns.ops = &timens_operations; ns->user_ns =get_user_ns (user_ns);return ns; - +fail_page: +free_page ((unsigned long )ns->offsets); fail_free:kfree (ns); fail_dec:
@@ -94,6 +103,7 @@ void free_time_ns(struct kref *kref)struct time_namespace *ns; ns = container_of(kref, struct time_namespace, kref); +free_page ((unsigned long )ns->offsets);dec_time_namespaces (ns->ucounts);put_user_ns (ns->user_ns);ns_free_inum (&ns->ns);
Signature · 1 line
2.20.1
PATCH 3/32 timens: Introduce CLOCK_MONOTONIC offsets
In reply to Dmitry SafonovFrom: Andrei Vagin <[email protected]>
Add monotonic time virtualisation for time namespace. Introduce timespec for monotionic clock into timens offsets and wire clock_gettime() syscall.
--- include/linux/time_namespace.h | 11 ++++++++++ include/linux/timens_offsets.h | 1 + kernel/time/posix-timers.c | 10 +++++++++ kernel/time/posix-timers.h | 1 + kernel/time_namespace.c | 38 ++++++++++++++++++++++++++++++++++ 5 files changed, 61 insertions(+)
include/linux/time_namespace.h +11 -0
@@ -41,6 +41,9 @@ static inline void put_time_ns(struct time_namespace *ns) } +extern void timens_clock_to_host (int clockid,struct timespec64 *val); +extern void timens_clock_from_host (int clockid,struct timespec64 *val); +#else static inline struct time_namespace *get_time_ns (struct time_namespace *ns) {
@@ -65,6 +68,14 @@ static inline int timens_on_fork(struct nsproxy *nsproxy, struct task_struct *tsreturn 0 ; } +static inline void timens_clock_to_host (int clockid,struct timespec64 *val) +{ +} + +static inline void timens_clock_from_host (int clockid,struct timespec64 *val) +{ +} +#endif #endif /* _LINUX_TIMENS_H */
include/linux/timens_offsets.h +1 -0
@@ -3,6 +3,7 @@#define _LINUX_TIME_OFFSETS_Hstruct timens_offsets { +struct timespec64 monotonic_time_offset; };#endif
kernel/time/posix-timers.c +10 -0
@@ -30,6 +30,7 @@#include <linux/hashtable.h> #include <linux/compat.h> #include <linux/nospec.h> +#include <linux/time_namespace.h> #include "timekeeping.h" #include "posix-timers.h"
@@ -1041,6 +1042,9 @@ SYSCALL_DEFINE2(clock_gettime, const clockid_t, which_clock, error = kc->clock_get (which_clock, &kernel_tp); +if (!error && kc->clock_timens_adjust) +timens_clock_from_host (which_clock, &kernel_tp); +if (!error &&put_timespec64 (&kernel_tp, tp)) error = -EFAULT ;
@@ -1117,6 +1121,9 @@ COMPAT_SYSCALL_DEFINE2(clock_gettime, clockid_t, which_clock, err = kc->clock_get (which_clock, &ts); +if (!err && kc->clock_timens_adjust) +timens_clock_from_host (which_clock, &ts); +if (!err &&put_old_timespec32 (&ts, tp)) err = -EFAULT ;
@@ -1259,6 +1266,7 @@ static const struct k_clock clock_realtime = {static const struct k_clock clock_monotonic = { .clock_getres = posix_get_hrtimer_res, .clock_get = posix_ktime_get_ts, + .clock_timens_adjust = true, .nsleep = common_nsleep, .timer_create = common_timer_create, .timer_set = common_timer_set,
@@ -1274,6 +1282,7 @@ static const struct k_clock clock_monotonic = {static const struct k_clock clock_monotonic_raw = { .clock_getres = posix_get_hrtimer_res, .clock_get = posix_get_monotonic_raw, + .clock_timens_adjust = true, };static const struct k_clock clock_realtime_coarse = {
@@ -1284,6 +1293,7 @@ static const struct k_clock clock_realtime_coarse = {static const struct k_clock clock_monotonic_coarse = { .clock_getres = posix_get_coarse_res, .clock_get = posix_get_monotonic_coarse, + .clock_timens_adjust = true, };static const struct k_clock clock_tai = {
kernel/time/posix-timers.h +1 -0
@@ -24,6 +24,7 @@ struct k_clock {int (*timer_try_to_cancel)(struct k_itimer *timr);void (*timer_arm)(struct k_itimer *timr,ktime_t expires,bool absolute,bool sigev_none); +bool clock_timens_adjust; };extern const struct k_clock clock_posix_cpu;
kernel/time_namespace.c +38 -0
@@ -191,6 +191,44 @@ static struct user_namespace *timens_owner(struct ns_common *ns)return to_time_ns (ns)->user_ns; } +static void clock_timens_fixup (int clockid,struct timespec64 *val,bool to_ns) +{ +struct timens_offsets *ns_offsets = current->nsproxy->time_ns->offsets; +struct timespec64 *offsets =NULL ; + +if (!ns_offsets) +return ; + +if (val->tv_sec ==0 && val->tv_nsec ==0 ) +return ; + +switch (clockid) { +case CLOCK_MONOTONIC : +case CLOCK_MONOTONIC_RAW : +case CLOCK_MONOTONIC_COARSE : + offsets = &ns_offsets->monotonic_time_offset; +break ; + } + +if (!offsets) +return ; + +if (to_ns) + *val =timespec64_add (*val, *offsets); +else + *val =timespec64_sub (*val, *offsets); +} + +void timens_clock_to_host (int clockid,struct timespec64 *val) +{ +clock_timens_fixup (clockid, val, false); +} + +void timens_clock_from_host (int clockid,struct timespec64 *val) +{ +clock_timens_fixup (clockid, val, true); +} +const struct proc_ns_operations timens_operations = { .name ="time" , .type =CLONE_NEWTIME ,
Signature · 1 line
2.20.1
PATCH 4/32 timens: Introduce CLOCK_BOOTTIME offset
In reply to Dmitry SafonovFrom: Andrei Vagin <[email protected]>
Adds boottime virtualisation for time namespace. Introduce timespec for boottime clock into timens offsets and wire clock_gettime() syscall.
--- include/linux/timens_offsets.h | 1 + kernel/time/posix-timers.c | 1 + kernel/time_namespace.c | 3 +++ 3 files changed, 5 insertions(+)
include/linux/timens_offsets.h +1 -0
@@ -4,6 +4,7 @@struct timens_offsets {struct timespec64 monotonic_time_offset; +struct timespec64 monotonic_boottime_offset; };#endif
kernel/time/posix-timers.c +1 -0
@@ -1314,6 +1314,7 @@ static const struct k_clock clock_tai = {static const struct k_clock clock_boottime = { .clock_getres = posix_get_hrtimer_res, .clock_get = posix_get_boottime, + .clock_timens_adjust = true, .nsleep = common_nsleep, .timer_create = common_timer_create, .timer_set = common_timer_set,
kernel/time_namespace.c +3 -0
@@ -208,6 +208,9 @@ static void clock_timens_fixup(int clockid, struct timespec64 *val, bool to_ns)case CLOCK_MONOTONIC_COARSE : offsets = &ns_offsets->monotonic_time_offset;break ; +case CLOCK_BOOTTIME : + offsets = &ns_offsets->monotonic_boottime_offset; +break ; }if (!offsets)
Signature · 1 line
2.20.1
PATCH 5/32 timerfd/timens: Take into account ns clock offsets
In reply to Dmitry SafonovFrom: Andrei Vagin <[email protected]>
Make timerfd respect timens offsets.
Provide two helpers timens_clock_to_host() timens_clock_from_host() that are useful to wire up timens to different kernel subsystems. Following patches will use timens_clock_from_host(), added here for completeness.
--- fs/timerfd.c | 16 +++++++++++----- 1 file changed, 11 insertions(+), 5 deletions(-)
fs/timerfd.c +11 -5
@@ -26,6 +26,7 @@#include <linux/syscalls.h> #include <linux/compat.h> #include <linux/rcupdate.h> +#include <linux/time_namespace.h> struct timerfd_ctx { union {
@@ -433,22 +434,27 @@ SYSCALL_DEFINE2(timerfd_create, int, clockid, int, flags) }static int do_timerfd_settime (int ufd,int flags, -const struct itimerspec64 *new, +struct itimerspec64 *new,struct itimerspec64 *old) {struct fd f;struct timerfd_ctx *ctx;int ret; -if ((flags & ~TFD_SETTIME_FLAGS ) || - !itimerspec64_valid (new)) -return -EINVAL ; - ret =timerfd_fget (ufd, &f);if (ret)return ret; ctx = f.file->private_data; +if (flags &TFD_TIMER_ABSTIME ) +timens_clock_to_host (ctx->clockid, &new->it_value); + +if ((flags & ~TFD_SETTIME_FLAGS ) || + !itimerspec64_valid (new)) { +fdput (f); +return -EINVAL ; + } +if (isalarm (ctx) && !capable (CAP_WAKE_ALARM )) {fdput (f);return -EPERM ;
Signature · 1 line
2.20.1
PATCH 6/32 posix-timers/timens: Take into account clock offsets
In reply to Dmitry SafonovFrom: Andrei Vagin <[email protected]>
Wire timer_settime() syscall into time namespace virtualization.
--- kernel/time/posix-timers.c | 13 +++++++++---- 1 file changed, 9 insertions(+), 4 deletions(-)
kernel/time/posix-timers.c +9 -4
@@ -854,10 +854,6 @@ static int do_timer_settime(timer_t timer_id, int flags,unsigned long flag;int error =0 ; -if (!timespec64_valid (&new_spec64->it_interval) || - !timespec64_valid (&new_spec64->it_value)) -return -EINVAL ; -if (old_spec64)memset (old_spec64,0 ,sizeof (*old_spec64)); retry:
@@ -865,6 +861,15 @@ static int do_timer_settime(timer_t timer_id, int flags,if (!timr)return -EINVAL ; +if (flags &TIMER_ABSTIME ) +timens_clock_to_host (timr->it_clock, &new_spec64->it_value); + +if (!timespec64_valid (&new_spec64->it_interval) || + !timespec64_valid (&new_spec64->it_value)) { +unlock_timer (timr, flag); +return -EINVAL ; + } + kc = timr->kclock;if (WARN_ON_ONCE (!kc || !kc->timer_set)) error = -EINVAL ;
Signature · 1 line
2.20.1
PATCH 7/32 timens/kernel: Take into account timens clock offsets in clock_nanosleep
In reply to Dmitry SafonovFrom: Andrei Vagin <[email protected]>
Wire up clock_nanosleep() to timens offsets.
--- kernel/time/hrtimer.c | 8 ++++++++ 1 file changed, 8 insertions(+)
kernel/time/hrtimer.c +8 -0
@@ -42,6 +42,7 @@#include <linux/timer.h> #include <linux/freezer.h> #include <linux/compat.h> +#include <linux/time_namespace.h> #include <linux/uaccess.h>
@@ -1721,9 +1722,16 @@ long hrtimer_nanosleep(const struct timespec64 *rqtp, {struct restart_block *restart;struct hrtimer_sleeper t; +struct timespec64 tp;int ret =0 ;u64 slack; +if (!(mode &HRTIMER_MODE_REL )) { + tp = *rqtp; + rqtp = &tp; +timens_clock_to_host (clockid, &tp); + } + slack = current->timer_slack_ns;if (dl_task (current) ||rt_task (current)) slack =0 ;
Signature · 1 line
2.20.1
PATCH 8/32 timens: Shift /proc/uptime
In reply to Dmitry SafonovRespect boottime inside time namespace for /proc/uptime
--- fs/proc/uptime.c | 3 +++ 1 file changed, 3 insertions(+)
fs/proc/uptime.c +3 -0
@@ -5,6 +5,7 @@#include <linux/sched.h> #include <linux/seq_file.h> #include <linux/time.h> +#include <linux/time_namespace.h> #include <linux/kernel_stat.h> static int uptime_proc_show (struct seq_file *m,void *v)
@@ -20,6 +21,8 @@ static int uptime_proc_show(struct seq_file *m, void *v) nsec += (__force u64 )kcpustat_cpu (i).cpustat[CPUTIME_IDLE ];ktime_get_boottime_ts64 (&uptime); +timens_clock_from_host (CLOCK_BOOTTIME , &uptime); + idle.tv_sec =div_u64_rem (nsec,NSEC_PER_SEC , &rem); idle.tv_nsec = rem; seq_printf(m, "%lu.%02lu %lu.%02lu\n",
Signature · 1 line
2.20.1
PATCH 9/32 x86/vdso2c: Correct err messages on file opening
In reply to Dmitry Safonoverr() message in main() is misleading: it should print outfilename, which is argv[3], not argv[2].
Correct error messages to be more precise about what failed and for which file.
--- arch/x86/entry/vdso/vdso2c.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
arch/x86/entry/vdso/vdso2c.c +2 -2
@@ -187,7 +187,7 @@ static void map_input(const char *name, void **addr, size_t *len, int prot)int fd =open (name,O_RDONLY );if (fd ==-1 ) -err (1 ,"%s" , name); +err (1 ,"open(%s)" , name); tmp_len =lseek (fd,0 ,SEEK_END );if (tmp_len == (off_t)-1 )
@@ -240,7 +240,7 @@ int main(int argc, char **argv) outfilename = argv[3 ]; outfile =fopen (outfilename,"w" );if (!outfile) -err (1 ,"%s" , argv[2 ]); +err (1 ,"fopen(%s)" , outfilename);go (raw_addr, raw_len, stripped_addr, stripped_len, outfile, name);
Signature · 1 line
2.20.1
PATCH 10/32 x86/vdso2c: Convert iterator to unsigned
In reply to Dmitry Safonovi and j are used everywhere with unsigned types.
Cleanup and prettify the code a bit.
Introduce syms_nr for readability and as a preparation for allocating an array of vDSO entries that will be needed for creating two vdso .so's:
one for host tasks and another for processes inside time namespace.
--- arch/x86/entry/vdso/vdso2c.h | 9 ++++----- 1 file changed, 4 insertions(+), 5 deletions(-)
arch/x86/entry/vdso/vdso2c.h +4 -5
@@ -13,7 +13,7 @@ static void BITSFUNC(go)(void *raw_addr, size_t raw_len,unsigned long load_size =-1 ;/* Work around bogus warning */ unsigned long mapping_size;ELF (Ehdr) *hdr = (ELF (Ehdr) *)raw_addr; -int i; +unsigned int i, syms_nr;unsigned long j;ELF (Shdr) *symtab_hdr =NULL , *strtab_hdr, *secstrings_hdr, *alt_sec =NULL ;
@@ -86,11 +86,10 @@ static void BITSFUNC(go)(void *raw_addr, size_t raw_len, strtab_hdr = raw_addr +GET_LE (&hdr->e_shoff) +GET_LE (&hdr->e_shentsize) *GET_LE (&symtab_hdr->sh_link); + syms_nr =GET_LE (&symtab_hdr->sh_size) /GET_LE (&symtab_hdr->sh_entsize);/* Walk the symbol table */ -for (i =0 ; - i <GET_LE (&symtab_hdr->sh_size) /GET_LE (&symtab_hdr->sh_entsize); - i++) { -int k; +for (i =0 ; i < syms_nr; i++) { +unsigned int k;ELF (Sym) *sym = raw_addr +GET_LE (&symtab_hdr->sh_offset) +GET_LE (&symtab_hdr->sh_entsize) * i; const char *name = raw_addr + GET_LE(&strtab_hdr->sh_offset) +
Signature · 1 line
2.20.1
PATCH 11/32 x86/vdso/Makefile: Add vobjs32
In reply to Dmitry SafonovAs for 64-bit vdso objects, handle ia32/i386 objects in array. This is a preparation ground to avoid code duplication on introduction of timens vdso.
--- arch/x86/entry/vdso/Makefile | 15 +++++---------- 1 file changed, 5 insertions(+), 10 deletions(-)
arch/x86/entry/vdso/Makefile +5 -10
@@ -18,6 +18,8 @@ VDSO32-$(CONFIG_IA32_EMULATION) := y # files to link into the vdso vobjs-y := vdso-note.o vclock_gettime.o vgetcpu.o +vobjs32-y := vdso32/note.o vdso32/system_call.o vdso32/sigreturn.o +vobjs32-y += vdso32/vclock_gettime.o # files to link into kernel obj-y += vma.o
@@ -31,10 +33,12 @@ vdso_img-$(VDSO32-y) += 32 obj-$(VDSO32-y) += vdso32-setup.o vobjs := $(foreach F,$(vobjs-y),$(obj)/$F) +vobjs32 := $(foreach F,$(vobjs32-y),$(obj)/$F) $(obj)/vdso.o: $(obj)/vdso.so targets += vdso.lds $(vobjs-y) +targets += vdso32/vdso32.lds $(vobjs32-y) # Build the vDSO image C files and link them in. vdso_img_objs := $(vdso_img-y:%=vdso-image-%.o)
@@ -125,10 +129,6 @@ $(obj)/vdsox32.so.dbg: $(obj)/vdsox32.lds $(vobjx32s) FORCE CPPFLAGS_vdso32.lds = $(CPPFLAGS_vdso.lds) VDSO_LDFLAGS_vdso32.lds = -m elf_i386 -soname linux-gate.so.1 -targets += vdso32/vdso32.lds -targets += vdso32/note.o vdso32/system_call.o vdso32/sigreturn.o -targets += vdso32/vclock_gettime.o - KBUILD_AFLAGS_32 := $(filter-out -m64,$(KBUILD_AFLAGS)) -DBUILD_VDSO $(obj)/vdso32.so.dbg: KBUILD_AFLAGS = $(KBUILD_AFLAGS_32) $(obj)/vdso32.so.dbg: asflags-$(CONFIG_X86_64) += -m32
@@ -153,12 +153,7 @@ endif $(obj)/vdso32.so.dbg: KBUILD_CFLAGS = $(KBUILD_CFLAGS_32) -$(obj)/vdso32.so.dbg: FORCE \ - $(obj)/vdso32/vdso32.lds \ - $(obj)/vdso32/vclock_gettime.o \ - $(obj)/vdso32/note.o \ - $(obj)/vdso32/system_call.o \ - $(obj)/vdso32/sigreturn.o +$(obj)/vdso32.so.dbg: $(obj)/vdso32/vdso32.lds $(vobjs32) FORCE $(call if_changed,vdso) #
Signature · 1 line
2.20.1
PATCH 12/32 x86/vdso/timens: Add offsets page in vvar
In reply to Dmitry SafonovFrom: Andrei Vagin <[email protected]>
As modern applications fetch time from VDSO without entering the kernel, it's needed to provide offsets for userspace code inside time namespace.
A page for timens offsets is allocated on time namespace construction. Put that page into VVAR for tasks inside timens and zero page for host processes.
As VDSO code is already optimized as much as possible in terms of speed, any new if-condition in VDSO code is undesirable; the goal is to provide two .so(s), as was originally suggested by Andy and Thomas:
- for host tasks with optimized-out clk_to_ns() without any penalty
- for processes inside timens with clk_to_ns()
For this purpose, define clk_to_ns() under BUILD_VDSO_TIME_NS, which will be enabled in the makefile for timens.so in following patches.
VDSO mappings are platform-specific, add Kconfig dependency for arch.
--- arch/Kconfig | 5 ++++ arch/x86/Kconfig | 1 + arch/x86/entry/vdso/vclock_gettime.c | 42 +++++++++++++++++++++++++++ arch/x86/entry/vdso/vdso-layout.lds.S | 9 +++++- arch/x86/entry/vdso/vdso2c.c | 3 ++ arch/x86/entry/vdso/vma.c | 12 ++++++++ arch/x86/include/asm/vdso.h | 1 + init/Kconfig | 1 + 8 files changed, 73 insertions(+), 1 deletion(-)
arch/Kconfig +5 -0
@@ -704,6 +704,11 @@ config HAVE_ARCH_HASH config ISA_BUS_API def_bool ISA +config ARCH_HAS_VDSO_TIME_NS + bool + help + VDSO can add time-ns offsets without entering kernel. + # # ABI hall of shame #
arch/x86/Kconfig +1 -0
@@ -70,6 +70,7 @@ config X86 select ARCH_HAS_STRICT_MODULE_RWX select ARCH_HAS_SYNC_CORE_BEFORE_USERMODE select ARCH_HAS_UBSAN_SANITIZE_ALL + select ARCH_HAS_VDSO_TIME_NS select ARCH_HAS_ZONE_DEVICE if X86_64 select ARCH_HAVE_NMI_SAFE_CMPXCHG select ARCH_MIGHT_HAVE_ACPI_PDC if ACPI
arch/x86/entry/vdso/vclock_gettime.c +42 -0
72 lines not on this page · raw message
arch/x86/entry/vdso/vdso-layout.lds.S +8 -1
@@ -7,6 +7,12 @@ * This script controls its layout. */ +#ifdef CONFIG_TIME_NS +# define TIMENS_SZ PAGE_SIZE +#else +# define TIMENS_SZ 0 +#endif + SECTIONS { /*
@@ -16,7 +22,7 @@ SECTIONS * segment. */ - vvar_start = . - 3 * PAGE_SIZE; + vvar_start = . - (3 * PAGE_SIZE + TIMENS_SZ); vvar_page = vvar_start; /* Place all vvars at the offsets in asm/vvar.h. */
@@ -28,6 +34,7 @@ SECTIONS pvclock_page = vvar_start + PAGE_SIZE; hvclock_page = vvar_start + 2 * PAGE_SIZE; + timens_page = vvar_start + 3 * PAGE_SIZE; . = SIZEOF_HEADERS;
arch/x86/entry/vdso/vdso2c.c +3 -0
@@ -76,6 +76,7 @@ enum { sym_hpet_page, sym_pvclock_page, sym_hvclock_page, + sym_timens_page, }; const int special_pages[] = {
@@ -83,6 +84,7 @@ const int special_pages[] = { sym_hpet_page, sym_pvclock_page, sym_hvclock_page, + sym_timens_page, }; struct vdso_sym {
@@ -96,6 +98,7 @@ struct vdso_sym required_syms[] = { [sym_hpet_page] = {"hpet_page", true}, [sym_pvclock_page] = {"pvclock_page", true}, [sym_hvclock_page] = {"hvclock_page", true}, + [sym_timens_page] = {"timens_page", true}, {"VDSO32_NOTE_MASK", true}, {"__kernel_vsyscall", true}, {"__kernel_sigreturn", true},
arch/x86/entry/vdso/vma.c +12 -0
@@ -14,6 +14,7 @@#include <linux/elf.h> #include <linux/cpu.h> #include <linux/ptrace.h> +#include <linux/time_namespace.h> #include <asm/pvclock.h> #include <asm/vgtod.h> #include <asm/proto.h>
@@ -23,6 +24,7 @@#include <asm/desc.h> #include <asm/cpufeature.h> #include <asm/mshyperv.h> +#include <asm/page.h> #if defined(CONFIG_X86_64 )unsigned int __read_mostly vdso64_enabled =1 ;
@@ -123,6 +125,16 @@ static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,if (tsc_pg &&vclock_was_used (VCLOCK_HVCLOCK ))return vmf_insert_pfn (vma, vmf->address,vmalloc_to_pfn (tsc_pg)); + }else if (sym_offset == image->sym_timens_page) { +struct time_namespace *ns = current->nsproxy->time_ns; +unsigned long pfn; + +if (!ns->offsets) + pfn =page_to_pfn (ZERO_PAGE (0 )); +else + pfn =page_to_pfn (virt_to_page (ns->offsets)); + +return vmf_insert_pfn (vma, vmf->address, pfn); }return VM_FAULT_SIGBUS ;
arch/x86/include/asm/vdso.h +1 -0
@@ -22,6 +22,7 @@ struct vdso_image {long sym_hpet_page;long sym_pvclock_page;long sym_hvclock_page; +long sym_timens_page;long sym_VDSO32_NOTE_MASK;long sym___kernel_sigreturn;long sym___kernel_rt_sigreturn;
init/Kconfig +1 -0
@@ -967,6 +967,7 @@ config UTS_NS config TIME_NS bool "TIME namespace" + depends on ARCH_HAS_VDSO_TIME_NS default y help In this namespace boottime and monotonic clocks can be set.
Signature · 1 line
2.20.1
PATCH 13/32 x86/vdso: Build timens .so(s)
In reply to Dmitry SafonovAs it has been discussed on timens RFC, adding a new conditional branch if (inside_time_ns) on VDSO for all processes is undesirable. It will add a penalty for everybody as branch predictor may mispredict the jump. Also there are instruction cache lines wasted on cmp/jmp.
Those effects of introducing time namespace are very much unwanted having in mind how much work have been spent on micro-optimisation vdso code.
Addressing those problems, build two VDSO images instead.
At this moment timens is unsupported for x32 binaries (only x86_64 and ia32). This may be added on top afterwards.
--- arch/x86/entry/vdso/Makefile | 28 +++++++++++++++++-- arch/x86/entry/vdso/vclock_gettime-timens.c | 6 ++++ .../entry/vdso/vdso32/vclock_gettime-timens.c | 6 ++++ 3 files changed, 37 insertions(+), 3 deletions(-) create mode 100644 arch/x86/entry/vdso/vclock_gettime-timens.c create mode 100644 arch/x86/entry/vdso/vdso32/vclock_gettime-timens.c
arch/x86/entry/vdso/Makefile +25 -3
@@ -17,8 +17,12 @@ VDSO32-$(CONFIG_X86_32) := y VDSO32-$(CONFIG_IA32_EMULATION) := y # files to link into the vdso -vobjs-y := vdso-note.o vclock_gettime.o vgetcpu.o +vobjs-y := vdso-note.o vgetcpu.o +vobjs-timens-y := $(vobjs-y) vclock_gettime-timens.o +vobjs-y += vclock_gettime.o + vobjs32-y := vdso32/note.o vdso32/system_call.o vdso32/sigreturn.o +vobjs32-timens-y := $(vobjs32-y) vdso32/vclock_gettime-timens.o vobjs32-y += vdso32/vclock_gettime.o # files to link into kernel
@@ -30,15 +34,21 @@ vdso_img-$(VDSO64-y) += 64 vdso_img-$(VDSOX32-y) += x32 vdso_img-$(VDSO32-y) += 32 +vdso_timens_img-$(VDSO64-y) += 64-timens +vdso_timens_img-$(VDSO32-y) += 32-timens +vdso_img-$(CONFIG_TIME_NS) += $(vdso_timens_img-y) + obj-$(VDSO32-y) += vdso32-setup.o vobjs := $(foreach F,$(vobjs-y),$(obj)/$F) vobjs32 := $(foreach F,$(vobjs32-y),$(obj)/$F) +vobjs-timens := $(foreach F,$(vobjs-timens-y),$(obj)/$F) +vobjs32-timens := $(foreach F,$(vobjs32-timens-y),$(obj)/$F) $(obj)/vdso.o: $(obj)/vdso.so -targets += vdso.lds $(vobjs-y) -targets += vdso32/vdso32.lds $(vobjs32-y) +targets += vdso.lds $(vobjs-y) $(vobjs-timens-y) +targets += vdso32/vdso32.lds $(vobjs32-y) $(vobjs32-timens-y) # Build the vDSO image C files and link them in. vdso_img_objs := $(vdso_img-y:%=vdso-image-%.o)
@@ -53,6 +63,9 @@ CPPFLAGS_vdso.lds += -P -C VDSO_LDFLAGS_vdso.lds = -m elf_x86_64 -soname linux-vdso.so.1 --no-undefined \ -z max-page-size=4096 +$(obj)/vdso64-timens.so.dbg: $(obj)/vdso.lds $(vobjs-timens) FORCE + $(call if_changed,vdso) + $(obj)/vdso64.so.dbg: $(obj)/vdso.lds $(vobjs) FORCE $(call if_changed,vdso)
@@ -81,12 +94,14 @@ endif endif $(vobjs): KBUILD_CFLAGS := $(filter-out $(GCC_PLUGINS_CFLAGS) $(RETPOLINE_CFLAGS),$(KBUILD_CFLAGS)) $(CFL) +$(vobjs-timens): KBUILD_CFLAGS := $(filter-out $(GCC_PLUGINS_CFLAGS) $(RETPOLINE_CFLAGS),$(KBUILD_CFLAGS)) $(CFL) # # vDSO code runs in userspace and -pg doesn't help with profiling anyway. # CFLAGS_REMOVE_vdso-note.o = -pg CFLAGS_REMOVE_vclock_gettime.o = -pg +CFLAGS_REMOVE_vclock_gettime-timens.o = -pg CFLAGS_REMOVE_vgetcpu.o = -pg CFLAGS_REMOVE_vvar.o = -pg
@@ -132,6 +147,8 @@ VDSO_LDFLAGS_vdso32.lds = -m elf_i386 -soname linux-gate.so.1 KBUILD_AFLAGS_32 := $(filter-out -m64,$(KBUILD_AFLAGS)) -DBUILD_VDSO $(obj)/vdso32.so.dbg: KBUILD_AFLAGS = $(KBUILD_AFLAGS_32) $(obj)/vdso32.so.dbg: asflags-$(CONFIG_X86_64) += -m32 +$(obj)/vdso32-timens.so.dbg: KBUILD_AFLAGS = $(KBUILD_AFLAGS_32) +$(obj)/vdso32-timens.so.dbg: asflags-$(CONFIG_X86_64) += -m32 KBUILD_CFLAGS_32 := $(filter-out -m64,$(KBUILD_CFLAGS)) KBUILD_CFLAGS_32 := $(filter-out -mcmodel=kernel,$(KBUILD_CFLAGS_32))
@@ -152,6 +169,10 @@ endif endif $(obj)/vdso32.so.dbg: KBUILD_CFLAGS = $(KBUILD_CFLAGS_32) +$(obj)/vdso32-timens.so.dbg: KBUILD_CFLAGS = $(KBUILD_CFLAGS_32) + +$(obj)/vdso32-timens.so.dbg: $(obj)/vdso32/vdso32.lds $(vobjs32-timens) FORCE + $(call if_changed,vdso) $(obj)/vdso32.so.dbg: $(obj)/vdso32/vdso32.lds $(vobjs32) FORCE $(call if_changed,vdso)
@@ -198,3 +219,4 @@ PHONY += vdso_install $(vdso_img_insttargets) vdso_install: $(vdso_img_insttargets) clean-files := vdso32.so vdso32.so.dbg vdso64* vdso-image-*.c vdsox32.so* +clean-files += vdso32-timens.so vdso32-timens.so.dbg
arch/x86/entry/vdso/vclock_gettime-timens.c new file +6 -0
@@ -0,0 +1,6 @@ +// SPDX-License-Identifier: GPL-2.0 + +#define BUILD_VDSO_TIME_NS + +#include "vclock_gettime.c" +
arch/x86/entry/vdso/vdso32/vclock_gettime-timens.c new file +6 -0
@@ -0,0 +1,6 @@ +// SPDX-License-Identifier: GPL-2.0 + +#define BUILD_VDSO_TIME_NS + +#include "vclock_gettime.c" +
Signature · 1 line
2.20.1
PATCH 14/32 x86/VDSO: Build VDSO with -ffunction-sections
In reply to Dmitry SafonovAs it has been discussed on timens RFC, adding a new conditional branch if (inside_time_ns) on VDSO for all processes is undesirable. It will add a penalty for everybody as branch predictor may mispredict the jump. Also there are instruction cache lines wasted on cmp/jmp.
Those effects of introducing time namespace are very much unwanted having in mind how much work have been spent on micro-optimisation vdso code.
Addressing those problems, there are two versions of VDSO's .so:
for host tasks (without any penalty) and for processes inside of time namespace with clk_to_ns() that subtracts offsets from host's time.
Unfortunately, to allow changing VDSO VMA on a running process, the entry points to VDSO should have the same offsets (addresses). That's needed as i.e. application that calls setns() may have already resolved VDSO symbols in GOT/PLT.
Compile VDSO images with -ffunction-sections so that VDSO entries can be aligned on the same addresses with linker script magic.
Put ia32 functions those are written in assembly into corresponding sections.
--- arch/x86/entry/vdso/Makefile | 2 +- arch/x86/entry/vdso/vdso32/sigreturn.S | 2 ++ arch/x86/entry/vdso/vdso32/system_call.S | 2 +- 3 files changed, 4 insertions(+), 2 deletions(-)
arch/x86/entry/vdso/Makefile +1 -1
@@ -3,7 +3,7 @@ # Building vDSO images for x86. # -KBUILD_CFLAGS += $(DISABLE_LTO) +KBUILD_CFLAGS += $(DISABLE_LTO) -ffunction-sections KASAN_SANITIZE := n UBSAN_SANITIZE := n OBJECT_FILES_NON_STANDARD := y
arch/x86/entry/vdso/vdso32/sigreturn.S +2 -0
@@ -11,6 +11,7 @@ .globl __kernel_sigreturn .type __kernel_sigreturn,@function nop /* this guy is needed for .LSTARTFDEDLSI1 below (watch for HACK) */ + .section .text.__kernel_sigreturn, "ax" ALIGN __kernel_sigreturn: .LSTART_sigreturn:
@@ -21,6 +22,7 @@ __kernel_sigreturn: nop .size __kernel_sigreturn,.-.LSTART_sigreturn + .section .text.__kernel_rt_sigreturn, "ax" .globl __kernel_rt_sigreturn .type __kernel_rt_sigreturn,@function ALIGN
arch/x86/entry/vdso/vdso32/system_call.S +1 -1
@@ -8,7 +8,7 @@ #include <asm/cpufeatures.h> #include <asm/alternative-asm.h> - .text + .section .text.__kernel_vsyscall, "ax" .globl __kernel_vsyscall .type __kernel_vsyscall,@function ALIGN
Signature · 1 line
2.20.1
PATCH 15/32 x86/vdso2c: Optionally produce linker script for vdso entries
In reply to Dmitry SafonovTwo VDSO images (with/without code for adding offsets inside timens) should be compatible by VDSO function offsets - this way kernel can remap VDSO VMA for a task without fixupping GOT/PLT.
Add an optional parameter for vdso2c to generate .entries file from vdso.so. As timens VDSO by nature is bigger in .text than VDSO for tasks outside namespace, this parameter will be used to generate .entries file from timens VDSO and include those aligns into linker script for !timens VDSO building.
--- arch/x86/entry/vdso/.gitignore | 1 + arch/x86/entry/vdso/Makefile | 7 ++++--- arch/x86/entry/vdso/vdso2c.c | 26 +++++++++++++++++++------- arch/x86/entry/vdso/vdso2c.h | 16 +++++++++++++++- 4 files changed, 39 insertions(+), 11 deletions(-)
arch/x86/entry/vdso/.gitignore +1 -0
@@ -5,3 +5,4 @@ vdso32-sysenter-syms.lds vdso32-int80-syms.lds vdso-image-*.c vdso2c +vdso*.entries
arch/x86/entry/vdso/Makefile +4 -3
@@ -47,8 +47,8 @@ vobjs32-timens := $(foreach F,$(vobjs32-timens-y),$(obj)/$F) $(obj)/vdso.o: $(obj)/vdso.so -targets += vdso.lds $(vobjs-y) $(vobjs-timens-y) -targets += vdso32/vdso32.lds $(vobjs32-y) $(vobjs32-timens-y) +targets += vdso.lds $(vobjs-y) $(vobjs-timens-y) vdso64.entries +targets += vdso32/vdso32.lds $(vobjs32-y) $(vobjs32-timens-y) vdso32.entries # Build the vDSO image C files and link them in. vdso_img_objs := $(vdso_img-y:%=vdso-image-%.o)
@@ -73,7 +73,8 @@ HOST_EXTRACFLAGS += -I$(srctree)/tools/include -I$(srctree)/include/uapi -I$(src hostprogs-y += vdso2c quiet_cmd_vdso2c = VDSO2C $@ - cmd_vdso2c = $(obj)/vdso2c $< $(<:%.dbg=%) $@ + cmd_vdso2c = $(obj)/vdso2c $< $(<:%.dbg=%) $@ \ + $(filter %.entries,$(<:%-timens.so.dbg=%.entries)) $(obj)/vdso-image-%.c: $(obj)/vdso%.so.dbg $(obj)/vdso%.so $(obj)/vdso2c FORCE $(call if_changed,vdso2c)
arch/x86/entry/vdso/vdso2c.c +19 -7
@@ -152,6 +152,7 @@ extern void bad_put_le(void);#define BITSFUNC3 (name, bits, suffix) name##bits##suffix#define BITSFUNC2 (name, bits, suffix) BITSFUNC3(name, bits, suffix)#define BITSFUNC (name) BITSFUNC2(name, ELF_BITS, ) +#define ELF_FUNC (f, x) (BITSFUNC2(ELF, ELF_BITS, _##f)(x))#define INT_BITS BITSFUNC2(int, ELF_BITS, _t)
@@ -169,16 +170,17 @@ extern void bad_put_le(void);static void go (void *raw_addr,size_t raw_len,void *stripped_addr,size_t stripped_len, -FILE *outfile,const char *name) +FILE *outfile,const char *name, +FILE *out_entries_lds) {Elf64_Ehdr *hdr = (Elf64_Ehdr *)raw_addr;if (hdr->e_ident[EI_CLASS ] ==ELFCLASS64 ) {go64 (raw_addr, raw_len, stripped_addr, stripped_len, - outfile, name); + outfile, name, out_entries_lds); }else if (hdr->e_ident[EI_CLASS ] ==ELFCLASS32 ) {go32 (raw_addr, raw_len, stripped_addr, stripped_len, - outfile, name); + outfile, name, out_entries_lds); }else {fail ("unknown ELF class\n" ); }
@@ -208,12 +210,12 @@ int main(int argc, char **argv) {size_t raw_len, stripped_len;void *raw_addr, *stripped_addr; -FILE *outfile; +FILE *outfile, *entries_lds =NULL ;char *name, *tmp;int namelen; -if (argc !=4 ) { -printf ("Usage: vdso2c RAW_INPUT STRIPPED_INPUT OUTPUT\n" ); +if (argc <4 ) { +printf ("Usage: vdso2c RAW_INPUT STRIPPED_INPUT OUTPUT [OUTPUT_ENTRIES.LDS]\n" );return 1 ; }
@@ -245,11 +247,21 @@ int main(int argc, char **argv)if (!outfile)err (1 ,"fopen(%s)" , outfilename); -go (raw_addr, raw_len, stripped_addr, stripped_len, outfile, name); +if (argc ==5 ) { + entries_lds =fopen (argv[4 ],"w" ); +if (!entries_lds) { +fclose (outfile); +err (1 ,"fopen(%s)" , argv[4 ]); + } + } + +go (raw_addr, raw_len, stripped_addr, stripped_len, outfile, name, entries_lds);munmap (raw_addr, raw_len);munmap (stripped_addr, stripped_len);fclose (outfile); +if (entries_lds) +fclose (entries_lds);return 0 ; }
arch/x86/entry/vdso/vdso2c.h +15 -1
@@ -7,7 +7,8 @@static void BITSFUNC (go )(void *raw_addr,size_t raw_len,void *stripped_addr,size_t stripped_len, -FILE *outfile,const char *name) +FILE *outfile,const char *name, +FILE *out_entries_lds) {int found_load =0 ;unsigned long load_size =-1 ;/* Work around bogus warning */
@@ -111,6 +112,19 @@ static void BITSFUNC(go)(void *raw_addr, size_t raw_len, syms[k] =GET_LE (&sym->st_value); } } + +if (!out_entries_lds) +continue ; + +if (ELF_FUNC (ST_BIND , sym->st_info) !=STB_GLOBAL ) +continue ; + +if (ELF_FUNC (ST_TYPE , sym->st_info) !=STT_FUNC ) +continue ; + +fprintf (out_entries_lds,"\t\t. = ABSOLUTE(%#lx);\n" , + (unsigned long )GET_LE (&sym->st_value)); +fprintf (out_entries_lds,"\t\t*(.text.%s*)\n" , name); }/* Validate mapping addresses. */
Signature · 1 line
2.20.1
PATCH 16/32 x86/vdso: Generate vdso{,32}-timens.lds
In reply to Dmitry SafonovAs it has been discussed on timens RFC, adding a new conditional branch if (inside_time_ns) on VDSO for all processes is undesirable. It will add a penalty for everybody as branch predictor may mispredict the jump. Also there are instruction cache lines wasted on cmp/jmp.
Those effects of introducing time namespace are very much unwanted having in mind how much work have been spent on micro-optimisation vdso code.
Addressing those problems, there are two versions of VDSO's .so:
for host tasks (without any penalty) and for processes inside of time namespace with clk_to_ns() that subtracts offsets from host's time.
Unfortunately, to allow changing VDSO VMA on a running process, the entry points to VDSO should have the same offsets (addresses). That's needed as i.e. application that calls setns() may have already resolved VDSO symbols in GOT/PLT.
Provide two linker scripts:
- *-timens.lds for building VDSO for processes inside time namespace (it has bigger functions and needs to build firstly)
- *.lds for host processes VDSO
(it has smaller functions and entry addresses should be adjusted with the linker script magic to fit with entries from timens)
--- arch/x86/entry/vdso/.gitignore | 1 + arch/x86/entry/vdso/Makefile | 18 ++++++++++++++---- arch/x86/entry/vdso/vdso-timens.lds.S | 7 +++++++ arch/x86/entry/vdso/vdso32/.gitignore | 1 + arch/x86/entry/vdso/vdso32/vdso32-timens.lds.S | 8 ++++++++ 5 files changed, 31 insertions(+), 4 deletions(-) create mode 100644 arch/x86/entry/vdso/vdso-timens.lds.S create mode 100644 arch/x86/entry/vdso/vdso32/vdso32-timens.lds.S
arch/x86/entry/vdso/.gitignore +1 -0
@@ -1,4 +1,5 @@ vdso.lds +vdso-timens.lds vdsox32.lds vdso32-syscall-syms.lds vdso32-sysenter-syms.lds
arch/x86/entry/vdso/Makefile +14 -4
@@ -44,11 +44,14 @@ vobjs := $(foreach F,$(vobjs-y),$(obj)/$F) vobjs32 := $(foreach F,$(vobjs32-y),$(obj)/$F) vobjs-timens := $(foreach F,$(vobjs-timens-y),$(obj)/$F) vobjs32-timens := $(foreach F,$(vobjs32-timens-y),$(obj)/$F) +dep-vdso.lds-$(CONFIG_TIME_NS) += $(obj)/vdso-image-64-timens.c +dep-vdso32.lds-$(CONFIG_TIME_NS) += $(obj)/vdso-image-32-timens.c $(obj)/vdso.o: $(obj)/vdso.so targets += vdso.lds $(vobjs-y) $(vobjs-timens-y) vdso64.entries targets += vdso32/vdso32.lds $(vobjs32-y) $(vobjs32-timens-y) vdso32.entries +targets += vdso-timens.lds vdso32/vdso32-timens.lds # Build the vDSO image C files and link them in. vdso_img_objs := $(vdso_img-y:%=vdso-image-%.o)
@@ -59,11 +62,13 @@ targets += $(vdso_img_cfiles) targets += $(vdso_img_sodbg) $(vdso_img-y:%=vdso%.so) CPPFLAGS_vdso.lds += -P -C +CPPFLAGS_vdso-timens.lds := $(CPPFLAGS_vdso.lds) VDSO_LDFLAGS_vdso.lds = -m elf_x86_64 -soname linux-vdso.so.1 --no-undefined \ -z max-page-size=4096 +VDSO_LDFLAGS_vdso-timens.lds := $(VDSO_LDFLAGS_vdso.lds) -$(obj)/vdso64-timens.so.dbg: $(obj)/vdso.lds $(vobjs-timens) FORCE +$(obj)/vdso64-timens.so.dbg: $(obj)/vdso-timens.lds $(vobjs-timens) FORCE $(call if_changed,vdso) $(obj)/vdso64.so.dbg: $(obj)/vdso.lds $(vobjs) FORCE
@@ -79,6 +84,9 @@ quiet_cmd_vdso2c = VDSO2C $@ $(obj)/vdso-image-%.c: $(obj)/vdso%.so.dbg $(obj)/vdso%.so $(obj)/vdso2c FORCE $(call if_changed,vdso2c) +$(obj)/vdso.lds: $(dep-vdso.lds-y) $(obj)/vdso2c FORCE +$(obj)/vdso32/vdso32.lds: $(dep-vdso32.lds-y) $(obj)/vdso2c FORCE + # # Don't omit frame pointers for ease of userspace debugging, but do # optimize sibling calls.
@@ -142,8 +150,10 @@ $(obj)/%.so: $(obj)/%.so.dbg $(obj)/vdsox32.so.dbg: $(obj)/vdsox32.lds $(vobjx32s) FORCE $(call if_changed,vdso) -CPPFLAGS_vdso32.lds = $(CPPFLAGS_vdso.lds) -VDSO_LDFLAGS_vdso32.lds = -m elf_i386 -soname linux-gate.so.1 +CPPFLAGS_vdso32.lds := $(CPPFLAGS_vdso.lds) +CPPFLAGS_vdso32-timens.lds := $(CPPFLAGS_vdso32.lds) +VDSO_LDFLAGS_vdso32.lds := -m elf_i386 -soname linux-gate.so.1 +VDSO_LDFLAGS_vdso32-timens.lds := $(VDSO_LDFLAGS_vdso32.lds) KBUILD_AFLAGS_32 := $(filter-out -m64,$(KBUILD_AFLAGS)) -DBUILD_VDSO $(obj)/vdso32.so.dbg: KBUILD_AFLAGS = $(KBUILD_AFLAGS_32)
@@ -172,7 +182,7 @@ endif $(obj)/vdso32.so.dbg: KBUILD_CFLAGS = $(KBUILD_CFLAGS_32) $(obj)/vdso32-timens.so.dbg: KBUILD_CFLAGS = $(KBUILD_CFLAGS_32) -$(obj)/vdso32-timens.so.dbg: $(obj)/vdso32/vdso32.lds $(vobjs32-timens) FORCE +$(obj)/vdso32-timens.so.dbg: $(obj)/vdso32/vdso32-timens.lds $(vobjs32-timens) FORCE $(call if_changed,vdso) $(obj)/vdso32.so.dbg: $(obj)/vdso32/vdso32.lds $(vobjs32) FORCE
arch/x86/entry/vdso/vdso-timens.lds.S new file +7 -0
@@ -0,0 +1,7 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +/* + * Linker script for 64-bit timens vDSO. + */ + +#define UNALIGNED_ENTRIES +#include "vdso.lds.S"
arch/x86/entry/vdso/vdso32/.gitignore +1 -0
@@ -1 +1,2 @@ vdso32.lds +vdso32-timens.lds
arch/x86/entry/vdso/vdso32/vdso32-timens.lds.S new file +8 -0
@@ -0,0 +1,8 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +/* + * Linker script for 32-bit timens vDSO. + */ + +#define UNALIGNED_ENTRIES + +#include "vdso32.lds.S"
Signature · 1 line
2.20.1
PATCH 17/32 x86/vdso2c: Sort vdso entries by addresses for linker script
In reply to Dmitry SafonovThere are two linker scripts for vdso .so(s):
- *-timens.lds for building vdso for processes inside time namespace (it has bigger functions and needs to build firstly)
- *.lds for host processes vdso
(it has smaller functions and entry addresses should be adjusted with the linker script magic to fit with entries from timens)
To adjust entries on host vdso, *.lds includes *.entries. Those are generated by vdso2c while parsing timens vdso.
Linker doesn't allow going back on some addresses, so sort entries to timens VDSO before writing them to .entries file.
--- arch/x86/entry/vdso/vdso2c.c | 13 +++++++++++++ arch/x86/entry/vdso/vdso2c.h | 20 +++++++++++++++++--- 2 files changed, 30 insertions(+), 3 deletions(-)
arch/x86/entry/vdso/vdso2c.c +13 -0
@@ -119,6 +119,19 @@ static void fail(const char *format, ...)va_end (ap); } +struct vdso_entry { +unsigned long addr; +const char *name; +}; + +static int entry_addr_cmp (const void *_a,const void *_b) +{ +const struct vdso_entry *a = _a; +const struct vdso_entry *b = _b; + +return (a->addr < b->addr) - (a->addr > b->addr); +} +/* * Evil macros for little-endian reads and writes */
arch/x86/entry/vdso/vdso2c.h +17 -3
@@ -21,6 +21,7 @@ static void BITSFUNC(go)(void *raw_addr, size_t raw_len,ELF (Dyn) *dyn =0 , *dyn_end =0 ;const char *secstrings;INT_BITS syms[NSYMS ] = {}; +struct vdso_entry *entries, *next_entry;ELF (Phdr) *pt = (ELF (Phdr) *)(raw_addr +GET_LE (&hdr->e_phoff));
@@ -88,6 +89,10 @@ static void BITSFUNC(go)(void *raw_addr, size_t raw_len,GET_LE (&hdr->e_shentsize) *GET_LE (&symtab_hdr->sh_link); syms_nr =GET_LE (&symtab_hdr->sh_size) /GET_LE (&symtab_hdr->sh_entsize); + entries =calloc (syms_nr,sizeof (*entries)); +if (!entries) +fail ("malloc()\n" ); + next_entry = entries;/* Walk the symbol table */ for (i =0 ; i < syms_nr; i++) {unsigned int k;
@@ -122,11 +127,20 @@ static void BITSFUNC(go)(void *raw_addr, size_t raw_len,if (ELF_FUNC (ST_TYPE , sym->st_info) !=STT_FUNC )continue ; -fprintf (out_entries_lds,"\t\t. = ABSOLUTE(%#lx);\n" , - (unsigned long )GET_LE (&sym->st_value)); -fprintf (out_entries_lds,"\t\t*(.text.%s*)\n" , name); + next_entry->addr =GET_LE (&sym->st_value); + next_entry->name = name; + next_entry++; } +qsort (entries, next_entry - entries,sizeof (*entries), entry_addr_cmp); + +while (next_entry != entries && out_entries_lds) { + next_entry--; +fprintf (out_entries_lds,"\t\t. = ABSOLUTE(%#lx);\n\t\t*(.text.%s*)\n" , + next_entry->addr, next_entry->name); + } +free (entries); +/* Validate mapping addresses. */ for (i =0 ; i <sizeof (special_pages) /sizeof (special_pages[0 ]); i++) {INT_BITS symval = syms[special_pages[i]];
Signature · 1 line
2.20.1
PATCH 18/32 x86/vdso.lds: Align !timens (host's) vdso.so entries
In reply to Dmitry SafonovAs it has been discussed on timens RFC, adding a new conditional branch if (inside_time_ns) on VDSO for all processes is undesirable. It will add a penalty for everybody as branch predictor may mispredict the jump. Also there are instruction cache lines wasted on cmp/jmp.
Those effects of introducing time namespace are very much unwanted having in mind how much work have been spent on micro-optimisation vdso code.
Addressing those problems, there are two versions of VDSO's .so:
for host tasks (without any penalty) and for processes inside of time namespace with clk_to_ns() that subtracts offsets from host's time.
Unfortunately, to allow changing VDSO VMA on a running process, the entry points to VDSO should have the same offsets (addresses). That's needed as i.e. application that calls setns() may have already resolved VDSO symbols in GOT/PLT.
Align VDSO entries for host with addresses generated from timens VDSO (which is bigger as it has code for adding offsets).
--- arch/x86/entry/vdso/vdso-layout.lds.S | 12 +++++++++++- 1 file changed, 11 insertions(+), 1 deletion(-)
arch/x86/entry/vdso/vdso-layout.lds.S +11 -1
@@ -70,7 +70,17 @@ SECTIONS * stuff that isn't used at runtime in between. */ - .text : { *(.text*) } :text =0x90909090, + .text : { +#if defined(CONFIG_TIME_NS) && !defined(UNALIGNED_ENTRIES) +#ifdef BUILD_VDSO32 +# include "vdso32.entries" +#endif +#ifdef BUILD_VDSO64 +# include "vdso64.entries" +#endif +#endif + *(.text*) + } :text =0x90909090, .altinstructions : { *(.altinstructions) } :text .altinstr_replacement : { *(.altinstr_replacement) } :text
Signature · 1 line
2.20.1
PATCH 19/32 x86/vdso2c: Align LOCAL symbols between vdso{-timens,}.so
In reply to Dmitry SafonovAlign not only VDSO entries as on timens VDSO, but also addresses of local functions. Otherwise, ld will put them after everything else into *(.text*). That will result in common VDSO size bigger than timens VDSO size (sic!).
Unfortunately, filtering by STB_WEAK doesn't work for ia32 VDSO:
by some reason gcc transforms weak symbols into local symbols in .so, i.e.:
27: 00000000 219 FUNC WEAK DEFAULT 12 clock_gettime
29: 00000000 95 FUNC WEAK DEFAULT 14 gettimeofday
32: 00000000 40 FUNC WEAK DEFAULT 16 timebecome:
20: 000006e0 219 FUNC LOCAL DEFAULT 12 clock_gettime
31: 000007c0 95 FUNC LOCAL DEFAULT 12 gettimeofday
33: 00000820 40 FUNC LOCAL DEFAULT 12 timethat results in the same align for two functions in .entries file:
. = ABSOLUTE(0x6e0); *(.text.__vdso_clock_gettime*) . = ABSOLUTE(0x6e0); *(.text.clock_gettime*)
As result, ld becomes a very sad animal and refuses to cooperate:
ld:arch/x86/entry/vdso/vdso32/vdso32.lds:339 cannot move location counter backwards (from 0000000000000762 to 00000000000006e0)
Align local functions on VDSO to timens VDSO and filter weak functions from .lds script.
--- arch/x86/entry/vdso/vdso2c.h | 15 +++++++++++++-- 1 file changed, 13 insertions(+), 2 deletions(-)
arch/x86/entry/vdso/vdso2c.h +13 -2
@@ -15,7 +15,7 @@ static void BITSFUNC(go)(void *raw_addr, size_t raw_len,unsigned long mapping_size;ELF (Ehdr) *hdr = (ELF (Ehdr) *)raw_addr;unsigned int i, syms_nr; -unsigned long j; +unsigned long j, last_entry_addr;ELF (Shdr) *symtab_hdr =NULL , *strtab_hdr, *secstrings_hdr, *alt_sec =NULL ;ELF (Dyn) *dyn =0 , *dyn_end =0 ;
@@ -121,7 +121,7 @@ static void BITSFUNC(go)(void *raw_addr, size_t raw_len,if (!out_entries_lds)continue ; -if (ELF_FUNC (ST_BIND , sym->st_info) !=STB_GLOBAL ) +if (ELF_FUNC (ST_BIND , sym->st_info) ==STB_WEAK )continue ;if (ELF_FUNC (ST_TYPE , sym->st_info) !=STT_FUNC )
@@ -134,8 +134,19 @@ static void BITSFUNC(go)(void *raw_addr, size_t raw_len,qsort (entries, next_entry - entries,sizeof (*entries), entry_addr_cmp); + last_entry_addr =-1UL ;while (next_entry != entries && out_entries_lds) { next_entry--; + +/* +* Unfortunately, WEAK symbols from objects are resoved +* into LOCAL symbols on ia32. Filter them here, as +* linker wouldn't like aligning the same symbol twice. +*/ +if (last_entry_addr == next_entry->addr) +continue ; + last_entry_addr = next_entry->addr; +fprintf (out_entries_lds,"\t\t. = ABSOLUTE(%#lx);\n\t\t*(.text.%s*)\n" , next_entry->addr, next_entry->name); }
Signature · 1 line
2.20.1
PATCH 20/32 x86/vdso: Initialize timens 64-bit vdso
In reply to Dmitry SafonovInitialize both 64-bit VDSO(s): host .so and timens one that has code for adding timens offsets.
--- arch/x86/entry/vdso/vma.c | 4 ++++ arch/x86/include/asm/vdso.h | 6 ++++++ 2 files changed, 10 insertions(+)
arch/x86/entry/vdso/vma.c +4 -0
@@ -343,6 +343,10 @@ static int __init init_vdso(void) {init_vdso_image (&vdso_image_64); +#ifdef CONFIG_TIME_NS +init_vdso_image (&vdso_image_64_timens); +#endif +#ifdef CONFIG_X86_X32_ABI init_vdso_image (&vdso_image_x32);#endif
arch/x86/include/asm/vdso.h +6 -0
@@ -32,6 +32,9 @@ struct vdso_image {#ifdef CONFIG_X86_64 extern const struct vdso_image vdso_image_64; +#ifdef CONFIG_TIME_NS +extern const struct vdso_image vdso_image_64_timens; +#endif #endif #ifdef CONFIG_X86_X32
@@ -40,6 +43,9 @@ extern const struct vdso_image vdso_image_x32;#if definedCONFIG_X86_32 || definedCONFIG_COMPAT extern const struct vdso_image vdso_image_32; +#ifdef CONFIG_TIME_NS +extern const struct vdso_image vdso_image_32_timens; +#endif #endif extern void __init init_vdso_image (const struct vdso_image *image);
Signature · 1 line
2.20.1
PATCH 21/32 x86/vdso: Switch image on setns()/unshare()/clone()
In reply to Dmitry SafonovAs it has been discussed on timens RFC, adding a new conditional branch if (inside_time_ns) on VDSO for all processes is undesirable. It will add a penalty for everybody as branch predictor may mispredict the jump. Also there are instruction cache lines wasted on cmp/jmp.
Those effects of introducing time namespace are very much unwanted having in mind how much work have been spent on micro-optimisation vdso code.
Addressing those problems, there are two versions of VDSO's .so:
for host tasks (without any penalty) and for processes inside of time namespace with clk_to_ns() that subtracts offsets from host's time.
Whenever a user does setns()/unshare() or clone() with CLONE_TIMENS, change VDSO image in mm and zap existing VVAR/VDSO page tables. They will be re-faulted with corresponding image and VVAR offsets.
--- arch/x86/entry/vdso/vma.c | 81 +++++++++++++++++++++++++++++++++++++ arch/x86/include/asm/vdso.h | 1 + kernel/time_namespace.c | 11 +++++ 3 files changed, 93 insertions(+)
arch/x86/entry/vdso/vma.c +81 -0
@@ -25,6 +25,7 @@#include <asm/cpufeature.h> #include <asm/mshyperv.h> #include <asm/page.h> +#include <asm/tlb.h> #if defined(CONFIG_X86_64 )unsigned int __read_mostly vdso64_enabled =1 ;
@@ -150,6 +151,84 @@ static const struct vm_special_mapping vvar_mapping = { .fault = vvar_fault, }; +#ifdef CONFIG_TIME_NS +static const struct vdso_image *timens_vdso (const struct vdso_image *old_img, +bool in_ns) +{ +#ifdef CONFIG_X86_X32_ABI +if (old_img == &vdso_image_x32) +return NULL ; +#endif +#if definedCONFIG_X86_32 || definedCONFIG_IA32_EMULATION +if (old_img == &vdso_image_32 || old_img == &vdso_image_32_timens) +return in_ns ? &vdso_image_32_timens : &vdso_image_32; +#endif +#ifdef CONFIG_X86_64 +if (old_img == &vdso_image_64 || old_img == &vdso_image_64_timens) +return in_ns ? &vdso_image_64_timens : &vdso_image_64; +#endif +return NULL ; +} + +static const struct vdso_image *image_to_timens (const struct vdso_image *img) +{ +bool in_ns = (current->nsproxy->time_ns != &init_time_ns); +const struct vdso_image *ns; + + ns =timens_vdso (img, in_ns); + +return ns ?: img; +} + +int vdso_join_timens (struct task_struct *task,bool inside_ns) +{ +const struct vdso_image *new_image, *old_image; +struct mm_struct *mm = task->mm; +struct vm_area_struct *vma; +int ret =0 ; + +if (down_write_killable (&mm->mmap_sem)) +return -EINTR ; + + old_image = mm->context.vdso_image; + new_image =timens_vdso (old_image, inside_ns); +if (!new_image) { + ret = -EOPNOTSUPP ; + goto out; + } + +/* Sanity checks, shouldn't happen */ +if (unlikely (old_image->size != new_image->size)) { + ret = -ENXIO ; + goto out; + } + + mm->context.vdso_image = new_image; + +for (vma = mm->mmap; vma; vma = vma->vm_next) { +unsigned long size = vma->vm_end - vma->vm_start; + +if (vma_is_special_mapping (vma, &vvar_mapping)) +zap_page_range (vma, vma->vm_start, size); +if (vma_is_special_mapping (vma, &vdso_mapping)) +zap_page_range (vma, vma->vm_start, size); + } + +out: +up_write (&mm->mmap_sem); +return ret; +} +#else /* CONFIG_TIME_NS */ +static const struct vdso_image *image_to_timens (const struct vdso_image *img) +{ +return img; +} +int vdso_join_timens (struct task_struct *task,bool inside_ns) +{ +return -ENXIO ; +} +#endif +/* * Add vdso and vvar mappings to current process. * @image - blob to map
@@ -165,6 +244,8 @@ static int map_vdso(const struct vdso_image *image, unsigned long addr)if (down_write_killable (&mm->mmap_sem))return -EINTR ; + image =image_to_timens (image); + addr =get_unmapped_area (NULL , addr, image->size - image->sym_vvar_start,0 ,0 );if (IS_ERR_VALUE (addr)) {
arch/x86/include/asm/vdso.h +1 -0
@@ -51,6 +51,7 @@ extern const struct vdso_image vdso_image_32_timens;extern void __init init_vdso_image (const struct vdso_image *image);extern int map_vdso_once (const struct vdso_image *image,unsigned long addr); +extern int vdso_join_timens (struct task_struct *task,bool inside_ns);#endif /* __ASSEMBLER__ */
kernel/time_namespace.c +11 -0
@@ -14,6 +14,7 @@#include <linux/proc_ns.h> #include <linux/sched/task.h> #include <linux/mm.h> +#include <asm/vdso.h> static struct ucounts *inc_time_namespaces (struct user_namespace *ns) {
@@ -155,11 +156,16 @@ static void timens_put(struct ns_common *ns)static int timens_install (struct nsproxy *nsproxy,struct ns_common *new) {struct time_namespace *ns =to_time_ns (new); +int ret;if (!ns_capable (ns->user_ns,CAP_SYS_ADMIN ) || !ns_capable (current_user_ns (),CAP_SYS_ADMIN ))return -EPERM ; + ret =vdso_join_timens (current, ns != &init_time_ns); +if (ret) +return ret; +get_time_ns (ns);get_time_ns (ns);put_time_ns (nsproxy->time_ns);
@@ -174,10 +180,15 @@ int timens_on_fork(struct nsproxy *nsproxy, struct task_struct *tsk) {struct ns_common *nsc = &nsproxy->time_ns_for_children->ns;struct time_namespace *ns =to_time_ns (nsc); +int ret;if (nsproxy->time_ns == nsproxy->time_ns_for_children)return 0 ; + ret =vdso_join_timens (tsk, ns != &init_time_ns); +if (ret) +return ret; +get_time_ns (ns);put_time_ns (nsproxy->time_ns); nsproxy->time_ns = ns;
Signature · 1 line
2.20.1
PATCH 22/32 timens: Add align for timens_offsets
In reply to Dmitry SafonovAlign offsets so that time namespace will work for ia32 applications on x86_64 host.
--- include/linux/timens_offsets.h | 12 ++++++++++-- 1 file changed, 10 insertions(+), 2 deletions(-)
include/linux/timens_offsets.h +10 -2
@@ -2,9 +2,17 @@#ifndef _LINUX_TIME_OFFSETS_H#define _LINUX_TIME_OFFSETS_H +/* +* Time offsets need align as they're placed on VVAR page, +* which is used by x86_64 and ia32 VDSO code. +* On ia32 offset::tv_sec (u64) has align(4), so re-align offsets +* to the same positions as 64-bit offsets. +* On 64-bit big-endian systems VDSO should convert to timespec64 +* to timespec because of a padding occurring between the fields. +*/ struct timens_offsets { -struct timespec64 monotonic_time_offset; -struct timespec64 monotonic_boottime_offset; +struct timespec64 monotonic_time_offset__aligned(8) ; +struct timespec64 monotonic_boottime_offset__aligned(8) ; };#endif
Signature · 1 line
2.20.1
PATCH 23/32 timens/fs/proc: Introduce /proc/pid/timens_offsets
In reply to Dmitry SafonovFrom: Andrei Vagin <[email protected]>
API to set time namespace offsets for children processes, i.e.:
echo "clockid off_ses off_nsec" > /proc/self/timens_offsets
--- fs/proc/base.c | 101 +++++++++++++++++++++++++++++++++ include/linux/time_namespace.h | 10 ++++ kernel/time_namespace.c | 71 +++++++++++++++++++++++ 3 files changed, 182 insertions(+)
fs/proc/base.c +101 -0
@@ -94,6 +94,7 @@#include <linux/sched/stat.h> #include <linux/flex_array.h> #include <linux/posix-timers.h> +#include <linux/time_namespace.h> #include <trace/events/oom.h> #include "internal.h" #include "fd.h"
@@ -1520,6 +1521,103 @@ static const struct file_operations proc_pid_sched_autogroup_operations = {#endif /* CONFIG_SCHED_AUTOGROUP */ +#ifdef CONFIG_TIME_NS +static int timens_offsets_show (struct seq_file *m,void *v) +{ +struct inode *inode = m->private; +struct task_struct *p; + + p =get_proc_task (inode); +if (!p) +return -ESRCH ; +proc_timens_show_offsets (p, m); + +put_task_struct (p); + +return 0 ; +} + +static ssize_t +timens_offsets_write (struct file *file,const char __user *buf, +size_t count,loff_t *ppos) +{ +struct inode *inode =file_inode (file); +struct proc_timens_offset offsets[2 ]; +char *kbuf =NULL , *pos, *next_line; +struct task_struct *p; +int ret, noffsets; + +/* Only allow < page size writes at the beginning of the file */ +if ((*ppos !=0 ) || (count >=PAGE_SIZE )) +return -EINVAL ; + +/* Slurp in the user data */ + kbuf =memdup_user_nul (buf, count); +if (IS_ERR (kbuf)) +return PTR_ERR (kbuf); + +/* Parse the user data */ + ret = -EINVAL ; + noffsets =0 ; + pos = kbuf; +for (; pos; pos = next_line) { +struct proc_timens_offset *off = &offsets[noffsets]; +int err; + +/* Find the end of line and ensure I don't look past it */ + next_line =strchr (pos,'\n' ); +if (next_line) { + *next_line ='\0' ; + next_line++; +if (*next_line =='\0' ) + next_line =NULL ; + } + + err =sscanf (pos,"%u %lld %lu" , &off->clockid, + &off->val.tv_sec, &off->val.tv_nsec); +if (err !=3 || off->val.tv_nsec >=NSEC_PER_SEC ) + goto out; +if (noffsets++ ==ARRAY_SIZE (offsets)) +break ; + } + + ret = -ESRCH ; + p =get_proc_task (inode); +if (!p) + goto out; + ret =proc_timens_set_offset (p, offsets, noffsets); +put_task_struct (p); +if (ret) + goto out; + + ret = count; +out: +kfree (kbuf); +return ret; +} + +static int timens_offsets_open (struct inode *inode,struct file *filp) +{ +int ret; + + ret =single_open (filp, timens_offsets_show,NULL ); +if (!ret) { +struct seq_file *m = filp->private_data; + + m->private = inode; + } +return ret; +} + +static const struct file_operations proc_timens_offsets_operations = { + .open = timens_offsets_open, + .read = seq_read, + .write = timens_offsets_write, + .llseek = seq_lseek, + .release = single_release, +}; +#endif /* CONFIG_TIME_NS */ +static ssize_t comm_write (struct file *file,const char __user *buf,size_t count,loff_t *offset) {
@@ -2953,6 +3051,9 @@ static const struct pid_entry tgid_base_stuff[] = {#endif #ifdef CONFIG_SCHED_AUTOGROUP REG("autogroup", S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations), +#endif +#ifdef CONFIG_TIME_NS +REG ("timens_offsets" ,S_IRUGO |S_IWUSR , proc_timens_offsets_operations), #endifREG ("comm" ,S_IRUGO |S_IWUSR , proc_pid_set_comm_operations), #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
include/linux/time_namespace.h +10 -0
@@ -41,6 +41,16 @@ static inline void put_time_ns(struct time_namespace *ns) } +extern void proc_timens_show_offsets (struct task_struct *p,struct seq_file *m); + +struct proc_timens_offset { +int clockid; +struct timespec64 val; +}; + +extern int proc_timens_set_offset (struct task_struct *p, +struct proc_timens_offset *offsets,int n); +extern void timens_clock_to_host (int clockid,struct timespec64 *val);extern void timens_clock_from_host (int clockid,struct timespec64 *val);
kernel/time_namespace.c +71 -0
83 lines not on this page · raw message
Signature · 1 line
2.20.1
PATCH 24/32 selftest/timens: Add Time Namespace test for supported clocks
In reply to Dmitry SafonovA test to check that all supported clocks work on host and inside a new time namespace. Use both ways to get time: through VDSO and by entering the kernel with implicit syscall.
Introduce a new timens directory in selftests framework for the next timens tests.
--- tools/testing/selftests/Makefile | 1 + tools/testing/selftests/timens/.gitignore | 1 + tools/testing/selftests/timens/Makefile | 5 + tools/testing/selftests/timens/config | 1 + tools/testing/selftests/timens/log.h | 26 +++ tools/testing/selftests/timens/timens.c | 191 ++++++++++++++++++++++ tools/testing/selftests/timens/timens.h | 63 +++++++ 7 files changed, 288 insertions(+) create mode 100644 tools/testing/selftests/timens/.gitignore create mode 100644 tools/testing/selftests/timens/Makefile create mode 100644 tools/testing/selftests/timens/config create mode 100644 tools/testing/selftests/timens/log.h create mode 100644 tools/testing/selftests/timens/timens.c create mode 100644 tools/testing/selftests/timens/timens.h
tools/testing/selftests/Makefile +1 -0
@@ -47,6 +47,7 @@ TARGETS += sysctl ifneq (1, $(quicktest)) TARGETS += timers endif +TARGETS += timens TARGETS += user TARGETS += vm TARGETS += x86
tools/testing/selftests/timens/.gitignore new file +1 -0
@@ -0,0 +1 @@ +timens
tools/testing/selftests/timens/Makefile new file +5 -0
@@ -0,0 +1,5 @@ +TEST_GEN_PROGS := timens + +CFLAGS := -Wall -Werror + +include ../lib.mk
tools/testing/selftests/timens/config new file +1 -0
@@ -0,0 +1 @@ +CONFIG_TIME_NS=y
tools/testing/selftests/timens/log.h new file +26 -0
@@ -0,0 +1,26 @@ +/* SPDX-License-Identifier: GPL-2.0 */ + +#ifndef __SELFTEST_TIMENS_LOG_H__ +#define __SELFTEST_TIMENS_LOG_H__ + +#define pr_msg (fmt, lvl, ...) \ + ksft_print_msg("[%s] (%s:%d)\t" fmt "\n", \ + lvl, __FILE__, __LINE__, ##__VA_ARGS__) + +#define pr_p (func, fmt, ...) func(fmt ": %m", ##__VA_ARGS__) + +#define pr_err (fmt, ...) \ + ({ \ + ksft_test_result_error(fmt, ##__VA_ARGS__); \ + -1; \ + }) + +#define pr_fail (fmt, ...) \ + ({ \ + ksft_test_result_fail(fmt, ##__VA_ARGS__); \ + -1; \ + }) + +#define pr_perror (fmt, ...) pr_p(pr_err, fmt, ##__VA_ARGS__) + +#endif
tools/testing/selftests/timens/timens.c new file +191 -0
191 lines not on this page · raw message
tools/testing/selftests/timens/timens.h new file +63 -0
63 lines not on this page · raw message
Signature · 1 line
2.20.1
PATCH 25/32 selftest/timens: Add a test for timerfd
In reply to Dmitry SafonovFrom: Andrei Vagin <[email protected]>
Check that timerfd_create() takes into account clock offsets.
--- tools/testing/selftests/timens/.gitignore | 1 + tools/testing/selftests/timens/Makefile | 2 +- tools/testing/selftests/timens/timerfd.c | 119 ++++++++++++++++++++++ 3 files changed, 121 insertions(+), 1 deletion(-) create mode 100644 tools/testing/selftests/timens/timerfd.c
tools/testing/selftests/timens/.gitignore +1 -0
@@ -1 +1,2 @@ timens +timerfd
tools/testing/selftests/timens/Makefile +1 -1
@@ -1,4 +1,4 @@ -TEST_GEN_PROGS := timens +TEST_GEN_PROGS := timens timerfd CFLAGS := -Wall -Werror
tools/testing/selftests/timens/timerfd.c new file +119 -0
119 lines not on this page · raw message
Signature · 1 line
2.20.1
PATCH 26/32 selftest/timens: Add a test for clock_nanosleep()
In reply to Dmitry SafonovFrom: Andrei Vagin <[email protected]>
Check that clock_nanosleep() takes into account clock offsets.
--- tools/testing/selftests/timens/.gitignore | 1 + tools/testing/selftests/timens/Makefile | 2 +- .../selftests/timens/clock_nanosleep.c | 99 +++++++++++++++++++ 3 files changed, 101 insertions(+), 1 deletion(-) create mode 100644 tools/testing/selftests/timens/clock_nanosleep.c
tools/testing/selftests/timens/.gitignore +1 -0
@@ -1,2 +1,3 @@ +clock_nanosleep timens timerfd
tools/testing/selftests/timens/Makefile +1 -1
@@ -1,4 +1,4 @@ -TEST_GEN_PROGS := timens timerfd +TEST_GEN_PROGS := timens timerfd clock_nanosleep CFLAGS := -Wall -Werror
tools/testing/selftests/timens/clock_nanosleep.c new file +99 -0
99 lines not on this page · raw message
Signature · 1 line
2.20.1
PATCH 27/32 selftest/timens: Add procfs selftest
In reply to Dmitry SafonovCheck that /proc/uptime is correct inside a new time namespace.
--- tools/testing/selftests/timens/.gitignore | 1 + tools/testing/selftests/timens/Makefile | 2 +- tools/testing/selftests/timens/procfs.c | 142 ++++++++++++++++++++++ 3 files changed, 144 insertions(+), 1 deletion(-) create mode 100644 tools/testing/selftests/timens/procfs.c
tools/testing/selftests/timens/.gitignore +1 -0
@@ -1,3 +1,4 @@ clock_nanosleep +procfs timens timerfd
tools/testing/selftests/timens/Makefile +1 -1
@@ -1,4 +1,4 @@ -TEST_GEN_PROGS := timens timerfd clock_nanosleep +TEST_GEN_PROGS := timens timerfd clock_nanosleep procfs CFLAGS := -Wall -Werror
tools/testing/selftests/timens/procfs.c new file +142 -0
142 lines not on this page · raw message
Signature · 1 line
2.20.1
PATCH 28/32 selftest/timens: Add timer offsets test
In reply to Dmitry SafonovFrom: Andrei Vagin <[email protected]>
Check that timer_create() takes into account clock offsets.
--- tools/testing/selftests/timens/.gitignore | 1 + tools/testing/selftests/timens/Makefile | 3 +- tools/testing/selftests/timens/timer.c | 115 ++++++++++++++++++++++ 3 files changed, 118 insertions(+), 1 deletion(-) create mode 100644 tools/testing/selftests/timens/timer.c
tools/testing/selftests/timens/.gitignore +1 -0
@@ -1,4 +1,5 @@ clock_nanosleep procfs timens +timer timerfd
tools/testing/selftests/timens/Makefile +2 -1
@@ -1,5 +1,6 @@ -TEST_GEN_PROGS := timens timerfd clock_nanosleep procfs +TEST_GEN_PROGS := timens timerfd timer clock_nanosleep procfs CFLAGS := -Wall -Werror +LDFLAGS := -lrt include ../lib.mk
tools/testing/selftests/timens/timer.c new file +115 -0
115 lines not on this page · raw message
Signature · 1 line
2.20.1
PATCH 29/32 selftests: Add a simple perf test for clock_gettime()
In reply to Dmitry SafonovFrom: Andrei Vagin <[email protected]>
--- tools/testing/selftests/timens/.gitignore | 1 + tools/testing/selftests/timens/Makefile | 8 +- tools/testing/selftests/timens/gettime_perf.c | 74 +++++++++++++++++++ .../selftests/timens/gettime_perf_cold.c | 63 ++++++++++++++++ 4 files changed, 145 insertions(+), 1 deletion(-) create mode 100644 tools/testing/selftests/timens/gettime_perf.c create mode 100644 tools/testing/selftests/timens/gettime_perf_cold.c
tools/testing/selftests/timens/.gitignore +1 -0
@@ -1,4 +1,5 @@ clock_nanosleep +gettime_perf procfs timens timer
tools/testing/selftests/timens/Makefile +7 -1
@@ -1,4 +1,10 @@ -TEST_GEN_PROGS := timens timerfd timer clock_nanosleep procfs +TEST_GEN_PROGS := timens timerfd timer clock_nanosleep procfs gettime_perf + +uname_M := $(shell uname -m 2>/dev/null || echo not) +ARCH ?= $(shell echo $(uname_M) | sed -e s/i.86/i386/) +ifeq ($(ARCH),x86_64) +TEST_GEN_PROGS += gettime_perf_cold +endif CFLAGS := -Wall -Werror LDFLAGS := -lrt
tools/testing/selftests/timens/gettime_perf.c new file +74 -0
74 lines not on this page · raw message
tools/testing/selftests/timens/gettime_perf_cold.c new file +63 -0
63 lines not on this page · raw message
Signature · 1 line
2.20.1
PATCH 30/32 selftest/timens: Check that a right vdso is mapped after fork and exec
In reply to Dmitry SafonovFrom: Andrei Vagin <[email protected]>
--- tools/testing/selftests/timens/.gitignore | 1 + tools/testing/selftests/timens/Makefile | 2 +- tools/testing/selftests/timens/exec.c | 91 +++++++++++++++++++++++ 3 files changed, 93 insertions(+), 1 deletion(-) create mode 100644 tools/testing/selftests/timens/exec.c
tools/testing/selftests/timens/.gitignore +1 -0
@@ -1,4 +1,5 @@ clock_nanosleep +exec gettime_perf procfs timens
tools/testing/selftests/timens/Makefile +1 -1
@@ -1,4 +1,4 @@ -TEST_GEN_PROGS := timens timerfd timer clock_nanosleep procfs gettime_perf +TEST_GEN_PROGS := timens timerfd timer clock_nanosleep procfs gettime_perf exec uname_M := $(shell uname -m 2>/dev/null || echo not) ARCH ?= $(shell echo $(uname_M) | sed -e s/i.86/i386/)
tools/testing/selftests/timens/exec.c new file +91 -0
91 lines not on this page · raw message
Signature · 1 line
2.20.1
PATCH 31/32 x86/vdso: Align VDSO functions by CPU L1 cache line
In reply to Dmitry SafonovFrom: Andrei Vagin <[email protected]>
After performance testing VDSO patches a noticeable 20% regression was found on gettime_perf selftest with a cold cache.
As it turns to be, before time namespaces introduction, VDSO functions were quite aligned to cache lines, but adding a new code to adjust timens offset inside namespace created a small shift and vdso functions become unaligned on cache lines.
Add align to vdso functions with gcc option to fix performance drop.
Coping the resulting numbers from cover letter:
Hot CPU cache (more gettime_perf.c cycles - the better):
| before | CONFIG_TIME_NS=n | host | inside timens --------|------------|------------------|-------------|------------- cycles | 139887013 | 139453003 | 139899785 | 128792458 diff (%)| 100 | 99.7 | 100 | 92
Cold cache (lesser tsc per gettime_perf_cold.c cycle - the better):
| before | CONFIG_TIME_NS=n | host | inside timens --------|------------|------------------|-------------|------------- tsc | 6748 | 6718 | 6862 | 12682 diff (%)| 100 | 99.6 | 101.7 | 188
Measured on Intel(R) Core(TM) i5-6300U CPU @ 2.40GHz
--- arch/x86/entry/vdso/Makefile | 1 + 1 file changed, 1 insertion(+)
arch/x86/entry/vdso/Makefile +1 -0
@@ -4,6 +4,7 @@ # KBUILD_CFLAGS += $(DISABLE_LTO) -ffunction-sections +KBUILD_CFLAGS += -falign-functions=$(CONFIG_X86_L1_CACHE_SHIFT) KASAN_SANITIZE := n UBSAN_SANITIZE := n OBJECT_FILES_NON_STANDARD := y
Signature · 1 line
2.20.1
PATCH 32/32 x86/vdso: Restrict splitting VVAR VMA
In reply to Dmitry SafonovAlthough, time namespace can work with VVAR VMA split, it seems worth to forbid splitting VVAR resulting in stricter ABI and reducing amount of corner-cases to consider while working further on VDSO.
I don't think there is any use-case for partial mremap() of vvar, but it there is - this patch can be easily dropped.
--- arch/x86/entry/vdso/vma.c | 13 +++++++++++++ 1 file changed, 13 insertions(+)
arch/x86/entry/vdso/vma.c +13 -0
@@ -87,6 +87,18 @@ static int vdso_mremap(const struct vm_special_mapping *sm,return 0 ; } +static int vvar_mremap (const struct vm_special_mapping *sm, +struct vm_area_struct *new_vma) +{ +unsigned long new_size = new_vma->vm_end - new_vma->vm_start; +const struct vdso_image *image = current->mm->context.vdso_image; + +if (new_size != -image->sym_vvar_start) +return -EINVAL ; + +return 0 ; +} +static vm_fault_t vvar_fault (const struct vm_special_mapping *sm,struct vm_area_struct *vma,struct vm_fault *vmf) {
@@ -149,6 +161,7 @@ static const struct vm_special_mapping vdso_mapping = {static const struct vm_special_mapping vvar_mapping = { .name ="[vvar]" , .fault = vvar_fault, + .mremap = vvar_mremap, };#ifdef CONFIG_TIME_NS
Signature · 1 line
2.20.1
Re: [PATCH 05/32] timerfd/timens: Take into account ns clock offsets
In reply to Dmitry SafonovQuoted 32 lines · Dmitry Safonov
From: Andrei Vagin <[email protected]>
Make timerfd respect timens offsets.
Provide two helperstimens_clock_to_host()timens_clock_from_host()that are useful to wire up timens to different kernel subsystems. Following patches will usetimens_clock_from_host(), added here for completeness.Signed-off-by Andrei Vagin <[email protected]>Co-developed-by Dmitry Safonov <[email protected]>Signed-off-by Dmitry Safonov <[email protected]>--- fs/timerfd.c | 16 +++++++++++----- 1 file changed, 11 insertions(+), 5 deletions(-)fs/timerfd.c +1
-1@@ -26,6 +26,7 @@#include <linux/syscalls.h> #include <linux/compat.h> #include <linux/rcupdate.h> +#include <linux/time_namespace.h> struct timerfd_ctx { union {@@ -433,22 +434,27 @@ SYSCALL_DEFINE2(timerfd_create, int, clockid, int, flags) } static int do_timerfd_settime(int ufd, int flags, - const struct itimerspec64 *new,
fs/timerfd.c SYSCALL_DEFINE2(timerfd_create, int, clockid, int, flags) · line 445
Quoted 10 lines
fs/timerfd.c +1
-3+struct itimerspec64 *new, struct itimerspec64 *old) { struct fd f;struct timerfd_ctx *ctx;int ret; -if ((flags & ~TFD_SETTIME_FLAGS ) || - !itimerspec64_valid (new)) -return -EINVAL ;
Please don't defer this early test of a @flags value. Otherwise if @flags is invalid you continue fget/put/clock-to-host even if result will be dropped out then.
Re: [PATCH 05/32] timerfd/timens: Take into account ns clock offsets
On Wed, Feb 06, 2019 at 11:52:03AM +0300, Cyrill Gorcunov wrote:
...
Quoted 8 lines
fs/timerfd.c SYSCALL_DEFINE2(timerfd_create, int, clockid, int, flags) · line 445
Quoted 4 lines
fs/timerfd.c +0
-3-if ((flags & ~TFD_SETTIME_FLAGS ) || - !itimerspec64_valid (new)) -return -EINVAL ;Please don't defer this early test of a @flags value. Otherwise if @flags is invalid you continue fget/put/clock-to-host even if result will be dropped out then.
Just to clarify -- this could be done on top of the series to not resend the whole bunch.
Re: [PATCH 05/32] timerfd/timens: Take into account ns clock offsets
In reply to Cyrill GorcunovQuoted 40 lines · Cyrill Gorcunov
Quoted 39 lines · Dmitry Safonov
From: Andrei Vagin <[email protected]>
Make timerfd respect timens offsets.
Provide two helperstimens_clock_to_host()timens_clock_from_host()that are useful to wire up timens to different kernel subsystems. Following patches will usetimens_clock_from_host(), added here for completeness.Signed-off-by Andrei Vagin <[email protected]>Co-developed-by Dmitry Safonov <[email protected]>Signed-off-by Dmitry Safonov <[email protected]>--- fs/timerfd.c | 16 +++++++++++----- 1 file changed, 11 insertions(+), 5 deletions(-)fs/timerfd.c +2
-1@@ -26,6 +26,7 @@#include <linux/syscalls.h> #include <linux/compat.h> #include <linux/rcupdate.h> +#include <linux/time_namespace.h> struct timerfd_ctx { union {@@ -433,22 +434,27 @@ SYSCALL_DEFINE2(timerfd_create, int, clockid, int, flags) }static int do_timerfd_settime (int ufd,int flags, -const struct itimerspec64 *new, +struct itimerspec64 *new,struct itimerspec64 *old) {struct fd f;struct timerfd_ctx *ctx;int ret;
Quoted 7 lines
fs/timerfd.c SYSCALL_DEFINE2(timerfd_create, int, clockid, int, flags) · line 445
Quoted 3 lines
fs/timerfd.c +0
-3-if ((flags & ~TFD_SETTIME_FLAGS ) || - !itimerspec64_valid (new)) -return -EINVAL ;Please don't defer this early test of a @flags value. Otherwise if @flags is invalid you continue fget/put/clock-to-host even if result will be dropped out then.
Cyrill, you are right. I moved this check together with itimerspec64_valid(). The idea was to call itimerspec64_valid() after applying clock offsets but for that, we need to know clockid.
Let's wait a bit for other comments to this patch set and then we will fix all things what will be found.
Thanks,
Andrei
Re: [PATCH 16/32] x86/vdso: Generate vdso{,32}-timens.lds
In reply to Dmitry SafonovQuoted 9 lines · Dmitry Safonov
As it has been discussed on timens RFC, adding a new conditional branch
if (inside_time_ns)on VDSO for all processes is undesirable. It will add a penalty for everybody as branch predictor may mispredict the jump. Also there are instruction cache lines wasted on cmp/jmp.Those effects of introducing time namespace are very much unwanted having in mind how much work have been spent on micro-optimisation vdso code.
Quoted 8 lines
Addressing those problems, there are two versions of VDSO's .so:
for host tasks (without any penalty) and for processes inside of time namespace withclk_to_ns()that subtracts offsets from host's time.Unfortunately, to allow changing VDSO VMA on a running process, the entry points to VDSO should have the same offsets (addresses). That's needed as i.e. application that calls
setns()may have already resolved VDSO symbols in GOT/PLT.
These (14-19, if I'm reading them right) seems to add quite a lot of complexity and fragility to the build, and other architectures would probably have to add something similar to their vdso builds.
I'm wondering why not make the rule be that a timens takes effect on next execve?
Rasmus
Re: [PATCH 16/32] x86/vdso: Generate vdso{,32}-timens.lds
Hi Rasmus,
Quoted 6 lines · Rasmus Villemoes
These (14-19, if I'm reading them right) seems to add quite a lot of complexity and fragility to the build, and other architectures would probably have to add something similar to their vdso builds.
I'm wondering why not make the rule be that a timens takes effect on next execve?
I believe, it would make setns() syscall much tricker than wanted:
At this moment the only exception is pidns which changes ns of the child and not the process-callee.
If exec() would be required to join timens - it may be a challenging problem for container systems: in order to enter it one needs to exec("/proc/self/exe") and add some new arguments/options. Furthermore, it seems to me that to enter container with this semantics, one needs to enter timens before entering mountns.
IOW, I believe, this would move complexity from kernel build time to userspace ABI. And I guess, it would require much more logic to re-create possibly nested namespaces hierarchy.
Rather I've considered using some kind of dynamic patching on vdso_init():
o static_branch - it would add some nops to !timens vdso
o something new like static_retpoline which would put RET over call to clk_to_ns(), shouldn't be a rocket since.
But in my point of view, if something can be done in compile time instead of patching code dynamically - than it reduces the complexity (lesser depends on what compiler/toolchain does).
Thanks,
Dmitry
Re: [PATCH 03/32] timens: Introduce CLOCK_MONOTONIC offsets
In reply to Dmitry Safonovkernel/time/posix-timers.c SYSCALL_DEFINE2(clock_gettime, const clockid_t, which_clock, · line 1046
Quoted 8 lines · Dmitry Safonov
#include "
timekeeping.h"
#include "posix-timers.h"kernel/time/posix-timers.c +2
-0@@ -1041,6 +1042,9 @@ SYSCALL_DEFINE2(clock_gettime, const clockid_t, which_clock, error = kc->clock_get (which_clock, &kernel_tp); +if (!error && kc->clock_timens_adjust) +timens_clock_from_host (which_clock, &kernel_tp);
Why are you adding this conditional instead of sticking the offset magic into the affected ->clock_get() implementations?
That spares you the switch() and the !offsets conditional.
kernel/time_namespace.c static struct user_namespace *timens_owner(struct ns_common *ns) · line 203
Quoted 10 lines
kernel/time_namespace.c +10
-0+static void clock_timens_fixup (int clockid,struct timespec64 *val,bool to_ns) +{ +struct timens_offsets *ns_offsets = current->nsproxy->time_ns->offsets; +struct timespec64 *offsets =NULL ; + +if (!ns_offsets) +return ; + +if (val->tv_sec ==0 && val->tv_nsec ==0 ) +return ;
I have no idea why 0/0 is special.
Quoted 9 lines
kernel/time_namespace.c +9
-0+ +switch (clockid) { +case CLOCK_MONOTONIC : +case CLOCK_MONOTONIC_RAW : +case CLOCK_MONOTONIC_COARSE : + offsets = &ns_offsets->monotonic_time_offset; +break ; + } +
kernel/time_namespace.c static struct user_namespace *timens_owner(struct ns_common *ns) · line 222
Quoted 10 lines
kernel/time_namespace.c +10
-0+if (!offsets) +return ; + +if (to_ns) + *val =timespec64_add (*val, *offsets); +else + *val =timespec64_sub (*val, *offsets); +} + +void timens_clock_to_host (int clockid,struct timespec64 *val)
Does this really need to be an out of line call? If you stick this into the clock_get() implementations then it boils down to:
static inline void timens_add_monotonic(struct timespec64 *ts)
{
struct timens_offsets *ns_offsets = current->nsproxy->time_ns->offsets;
if (ns_offsets)
*ts = timespec64_add(*ts, ns_offsets->monotonic_time_offset);
}and
static int posix_ktime_get_ts(clockid_t which_clock, struct timespec64 *tp)
{
ktime_get_ts64(tp);
timens_add_monotonic(tp);
return 0;
}Hmm?
Thanks,
tglx
Re: [PATCH 07/32] timens/kernel: Take into account timens clock offsets in clock_nanosleep
In reply to Dmitry Safonovkernel/time/hrtimer.c long hrtimer_nanosleep(const struct timespec64 *rqtp, · line 1731
Quoted 12 lines · Dmitry Safonov
kernel/time/hrtimer.c +4
-0@@ -1721,9 +1722,16 @@ long hrtimer_nanosleep(const struct timespec64 *rqtp, {struct restart_block *restart;struct hrtimer_sleeper t; +struct timespec64 tp;int ret =0 ;u64 slack; +if (!(mode &HRTIMER_MODE_REL )) { + tp = *rqtp; + rqtp = &tp;
So every invocation of hrtimer_nanosleep() gains a copy of the timespec64 even if the namespace muck is disabled.
The only relevant caller of this is common_nsleep(). So it might make sense to have common_nsleep() separated for CLOCK_MONOTONIC/BOOTTIME and handle the thing there. That again avoids the switch() to and out of line calls.
Thanks,
tglx
Re: [PATCH 03/32] timens: Introduce CLOCK_MONOTONIC offsets
In reply to Thomas GleixnerQuoted 15 lines · Thomas Gleixner
kernel/time/posix-timers.c SYSCALL_DEFINE2(clock_gettime, const clockid_t, which_clock, · line 1046
Quoted 8 lines · Dmitry Safonov
#include "
timekeeping.h"
#include "posix-timers.h"kernel/time/posix-timers.c +2
-0@@ -1041,6 +1042,9 @@ SYSCALL_DEFINE2(clock_gettime, const clockid_t, which_clock, error = kc->clock_get (which_clock, &kernel_tp); +if (!error && kc->clock_timens_adjust) +timens_clock_from_host (which_clock, &kernel_tp);Why are you adding this conditional instead of sticking the offset magic into the affected ->clock_get() implementations?
That spares you the
switch()and the !offsets conditional.
Quoted 12 lines
kernel/time_namespace.c static struct user_namespace *timens_owner(struct ns_common *ns) · line 203
Quoted 10 lines
kernel/time_namespace.c +10
-0+static void clock_timens_fixup (int clockid,struct timespec64 *val,bool to_ns) +{ +struct timens_offsets *ns_offsets = current->nsproxy->time_ns->offsets; +struct timespec64 *offsets =NULL ; + +if (!ns_offsets) +return ; + +if (val->tv_sec ==0 && val->tv_nsec ==0 ) +return ;I have no idea why 0/0 is special.
Initially this function was introduced to apply timens offsets in do_timer_settime and there it is special and means that the timer should be disarmed.
Now this functions is used in many other places and this check defenetly sould not be here.
Quoted 14 lines
Quoted 14 lines
kernel/time_namespace.c +13
-0+ +switch (clockid) { +case CLOCK_MONOTONIC : +case CLOCK_MONOTONIC_RAW : +case CLOCK_MONOTONIC_COARSE : + offsets = &ns_offsets->monotonic_time_offset; +break ; + } + +if (!offsets) +return ; + +if (to_ns)
Quoted 8 lines
kernel/time_namespace.c static struct user_namespace *timens_owner(struct ns_common *ns) · line 222
Quoted 6 lines
kernel/time_namespace.c +6
-0+ *val =timespec64_add (*val, *offsets); + else + *val =timespec64_sub (*val, *offsets); +} + +void timens_clock_to_host (int clockid,struct timespec64 *val)Does this really need to be an out of line call?
The idea was to collect all the logic about timens offsets in one place.
clock_timens_fixup() is used in all places where we need apply timens offsets (clock_gettim, posix timers, clock_nanosleep, timerfd, uptime_proc_show).
Quoted 2 lines
If you stick this into the
clock_get()implementations then it boils down to:
clock_get() is called from clock_gettime and from common_timer_get(). In common_timer_get(), we expect to get time in the root time namespace.
but I think we can handle this. For example, we can introduce a new flag CLOCL_TIMENS and
kc->clock_get(which_clock | CLOCK_TIMENS, &tp) will return time in a current time namespace.
kc->clock_get(which_clock, &tp) will return time in the root time namespace.
Quoted 11 lines
static inline void
timens_add_monotonic(struct timespec64 *ts){ struct timens_offsets *ns_offsets = current->nsproxy->time_ns->offsets; if (ns_offsets) *ts = timespec64_add(*ts, ns_offsets->monotonic_time_offset); }and
Quoted 8 lines
static int
posix_ktime_get_ts(clockid_t which_clock, struct timespec64 *tp){ ktime_get_ts64(tp); timens_add_monotonic(tp); return 0; }Hmm?
Yes, we can do this. I like this idea. This will allow us to remove timens_clock_to_host(), but I am not sure that we will be able to do something similar with timens_clock_from_host, which is used to apply offsets for timers. I need to look at the timer code again.
Thanks,
Andrei
Quoted 5 lines
Thanks,
tglx
Re: [PATCH 03/32] timens: Introduce CLOCK_MONOTONIC offsets
In reply to Thomas GleixnerQuoted 9 lines · Thomas Gleixner
Does this really need to be an out of line call? If you stick this into the
clock_get()implementations then it boils down to:static inline void
timens_add_monotonic(struct timespec64 *ts){ struct timens_offsets *ns_offsets = current->nsproxy->time_ns->offsets; if (ns_offsets) *ts = timespec64_add(*ts, ns_offsets->monotonic_time_offset);
And this needs to be a special variant of timespec64-add_safe(). timespec64_add_safe() is not sufficient, because it assumes that both values are positive, which is not the case here..
In timer_set() implementations you move the timespec_valid() check after the add. That's wrong because you really want to check the input value from user space.
Assume that the caller supplied value is valid and the adjustment brings it out of range then how should the caller understand why it it rejected?
So timespec64_add_namespace() must check for under and overflow. But doing this with timespecs is a pain. I rather suggest to rework the whole thing so hrtimer_nanosleep() takes a ktime_t expiry value and move the conversion to the call sites. Then the whole offset magic becomes:
expires = timespec64_to_ktime(rqtp); if (abstime) expires = timens_to_host_mono(expires);
and that function can nicely do the underflow and overflow detection and cap the values to 0 on underflow and KTIME_MAX on overflow.
Hmm?
Thanks,
tglx
Re: [PATCH 16/32] x86/vdso: Generate vdso{,32}-timens.lds
In reply to Rasmus VillemoesOn Thu, 7 Feb 2019, Rasmus Villemoes wrote:
Quoted 14 lines · Dmitry Safonov
Quoted 13 lines
As it has been discussed on timens RFC, adding a new conditional branch
if (inside_time_ns)on VDSO for all processes is undesirable. It will add a penalty for everybody as branch predictor may mispredict the jump. Also there are instruction cache lines wasted on cmp/jmp.Those effects of introducing time namespace are very much unwanted having in mind how much work have been spent on micro-optimisation vdso code.
Addressing those problems, there are two versions of VDSO's .so:
for host tasks (without any penalty) and for processes inside of time namespace withclk_to_ns()that subtracts offsets from host's time.
Quoted 8 lines
Quoted 4 lines
Unfortunately, to allow changing VDSO VMA on a running process, the entry points to VDSO should have the same offsets (addresses). That's needed as i.e. application that calls
setns()may have already resolved VDSO symbols in GOT/PLT.These (14-19, if I'm reading them right) seems to add quite a lot of complexity and fragility to the build, and other architectures would probably have to add something similar to their vdso builds.
Yes and we really want to avoid that. The VDSO implementations are pointlessly different accross the architectures and there is effort on the way to consolidate them:
https://lkml.kernel.org/r/[email protected]
I talked to Vincenzo earlier this week and he's working on a new version of that. The timens stuff wants to go on top of the consolidation otherwise we end up with another set of pointlessly different and differently broken VDSO variants.
Thanks,
tglx
Re: [PATCH 16/32] x86/vdso: Generate vdso{,32}-timens.lds
Quoted 31 lines · Thomas Gleixner
On Thu, 7 Feb 2019, Rasmus Villemoes wrote:
Cc + Vincenzo, WillQuoted 14 lines · Dmitry Safonov
Quoted 13 lines
As it has been discussed on timens RFC, adding a new conditional branch
if (inside_time_ns)on VDSO for all processes is undesirable. It will add a penalty for everybody as branch predictor may mispredict the jump. Also there are instruction cache lines wasted on cmp/jmp.Those effects of introducing time namespace are very much unwanted having in mind how much work have been spent on micro-optimisation vdso code.
Addressing those problems, there are two versions of VDSO's .so:
for host tasks (without any penalty) and for processes inside of time namespace withclk_to_ns()that subtracts offsets from host's time.Quoted 8 lines
Quoted 4 lines
Unfortunately, to allow changing VDSO VMA on a running process, the entry points to VDSO should have the same offsets (addresses). That's needed as i.e. application that calls
setns()may have already resolved VDSO symbols in GOT/PLT.These (14-19, if I'm reading them right) seems to add quite a lot of complexity and fragility to the build, and other architectures would probably have to add something similar to their vdso builds.
Yes and we really want to avoid that. The VDSO implementations are pointlessly different accross the architectures and there is effort on the way to consolidate them:
Quoted 6 lines
https://lkml.kernel.org/r/[email protected]
I talked to Vincenzo earlier this week and he's working on a new version of that. The timens stuff wants to go on top of the consolidation otherwise we end up with another set of pointlessly different and differently broken VDSO variants.
That looks awesome!
I've missed the tread about it, will catch the details.
Thanks much,
Dmitry
Re: [PATCH 16/32] x86/vdso: Generate vdso{,32}-timens.lds
In reply to Thomas GleixnerWhile the generic vdso patchset is in development, we decided to think about what other ways of generating two vdso libraries. In this patchset, we use a linker script, but it looks too complicated, so we decided to look at other options. Another obvious approach is the code patching technique. The main idea was to reduce the amount of arch-dependent code and Dmitry brought with the idea of three labels. Let’s look at this pseudo-code:
Int vdso_clock_gettime(clockid_t clk, struct timespec *ts) {
...l_call:
clk_to_ns(clk, ts)
l_return:
return 0;
annotate_reachable();
l_out:
nop();
return 0;
}Here we can see three labels. Without patching this code, the function will apply vdso offsets. But if we copy the code between the last two labels to the first label, we will get a version which skips vdso offsets. The patch which implements this idea will be in replies to this email. It was tested on x86_64 and with gcc as a compiler, but we suspect that there might be some issues on other architectures or with other compilers. So we would like to ask the help of the community to understand what we have to do to be sure that this code works always correctly.
The second patch implements static_branch for the vdso code. Here are only a few lines of arch-dependent code:
+static __always_inline bool timens_static_branch(void) +{
+ asm_volatile_goto("1:\n\t"
+ ".byte " __stringify(STATIC_KEY_INIT_NOP) "\n\t"
+ ".pushsection __retcall_table, \"aw\"\n\t" + "2: .word 1b - 2b, %l[l_yes] - 2b\n\t" + ".popsection\n\t" + : : : : l_yes);
+
+ return false;
+l_yes:
+ return true;
+}
This is a slightly modified version of the arch_static_branch() function. The timens code in vdso looks like this:
if (timens_static_branch()) {
clk_to_ns(clk, ts);
}The version of vdso which is compiled from sources will never execute clk_to_ns(). And then we can patch the 'no-op' in the straight-line codepath with a 'jump' instruction to the out-of-line true branch and get the timens version of the vdso library.
Now we can compare these three versions. Our opinion is that the version with three labels looks cleaner and if it will work with all compilers on all architectures, we probably have to choose it. Otherwise, we would prefer the version with static_branches, because it is simpler than the version with the linker script.
Thanks,
Andrei
Quoted 41 lines · Thomas Gleixner
On Thu, 7 Feb 2019, Rasmus Villemoes wrote:
Cc + Vincenzo, WillQuoted 14 lines · Dmitry Safonov
Quoted 13 lines
As it has been discussed on timens RFC, adding a new conditional branch
if (inside_time_ns)on VDSO for all processes is undesirable. It will add a penalty for everybody as branch predictor may mispredict the jump. Also there are instruction cache lines wasted on cmp/jmp.Those effects of introducing time namespace are very much unwanted having in mind how much work have been spent on micro-optimisation vdso code.
Addressing those problems, there are two versions of VDSO's .so:
for host tasks (without any penalty) and for processes inside of time namespace withclk_to_ns()that subtracts offsets from host's time.Quoted 8 lines
Quoted 4 lines
Unfortunately, to allow changing VDSO VMA on a running process, the entry points to VDSO should have the same offsets (addresses). That's needed as i.e. application that calls
setns()may have already resolved VDSO symbols in GOT/PLT.These (14-19, if I'm reading them right) seems to add quite a lot of complexity and fragility to the build, and other architectures would probably have to add something similar to their vdso builds.
Yes and we really want to avoid that. The VDSO implementations are pointlessly different accross the architectures and there is effort on the way to consolidate them:
https://lkml.kernel.org/r/[email protected]
I talked to Vincenzo earlier this week and he's working on a new version of that. The timens stuff wants to go on top of the consolidation otherwise we end up with another set of pointlessly different and differently broken VDSO variants.
Thanks,
tglx
RFC PATCH 1/1 x86/asm: Introduce static_retcall(s)
From: Dmitry Safonov <[email protected]>
Provide framework to overwrite tail call in a function with return.
XXX: split vdso/generic part
--- arch/x86/entry/vdso/vclock_gettime.c | 19 ++++++---- arch/x86/entry/vdso/vdso-layout.lds.S | 1 + arch/x86/entry/vdso/vdso2c.h | 11 +++++- arch/x86/entry/vdso/vma.c | 22 +++++++++++ arch/x86/include/asm/static_retcall.h | 54 +++++++++++++++++++++++++++ arch/x86/include/asm/vdso.h | 1 + 6 files changed, 99 insertions(+), 9 deletions(-) create mode 100644 arch/x86/include/asm/static_retcall.h
arch/x86/entry/vdso/vclock_gettime.c +11 -8
@@ -18,6 +18,7 @@#include <asm/msr.h> #include <asm/pvclock.h> #include <asm/mshyperv.h> +#include <asm/static_retcall.h> #include <linux/math64.h> #include <linux/time.h> #include <linux/kernel.h>
@@ -39,7 +40,7 @@ extern u8 hvclock_page __attribute__((visibility("hidden")));#endif -#ifdef BUILD_VDSO_TIME_NS +#ifdef CONFIG_TIME_NS extern u8 timens_page __attribute__((visibility("hidden")));#endif
@@ -145,9 +146,9 @@ notrace static inline u64 vgetcyc(int mode)return U64_MAX ; } +#ifdef CONFIG_TIME_NS notrace static __always_inline void clk_to_ns(clockid_t clk, struct timespec *ts) { -#ifdef BUILD_VDSO_TIME_NS struct timens_offsets *timens = (struct timens_offsets *) &timens_page;struct timespec64 *offset64;
@@ -173,9 +174,13 @@ notrace static __always_inline void clk_to_ns(clockid_t clk, struct timespec *ts ts->tv_nsec +=NSEC_PER_SEC ; ts->tv_sec--; } - -#endif } +#define _static_retcall static_retcall +#define _static_retcall_int static_retcall_int +#else +#define _static_retcall (...) +#define _static_retcall_int (...) +#endif notrace static int do_hres(clockid_t clk, struct timespec *ts) {
@@ -203,9 +208,7 @@ notrace static int do_hres(clockid_t clk, struct timespec *ts) ts->tv_sec = sec +__iter_div_u64_rem (ns,NSEC_PER_SEC , &ns); ts->tv_nsec = ns; -clk_to_ns (clk, ts); - -return 0 ; +_static_retcall_int (0 , clk_to_ns, clk, ts); } notrace static void do_coarse(clockid_t clk, struct timespec *ts)
@@ -219,7 +222,7 @@ notrace static void do_coarse(clockid_t clk, struct timespec *ts) ts->tv_nsec = base->nsec; }while (unlikely (gtod_read_retry (gtod, seq))); -clk_to_ns (clk, ts); +_static_retcall (clk_to_ns, clk, ts); } notrace int __vdso_clock_gettime(clockid_t clock, struct timespec *ts)
arch/x86/entry/vdso/vdso-layout.lds.S +1 -0
@@ -45,6 +45,7 @@ SECTIONS .gnu.version : { *(.gnu.version) } .gnu.version_d : { *(.gnu.version_d) } .gnu.version_r : { *(.gnu.version_r) } + __retcall_table : { *(__retcall_table) } :text .dynamic : { *(.dynamic) } :text :dynamic
arch/x86/entry/vdso/vdso2c.h +10 -1
@@ -16,7 +16,7 @@ static void BITSFUNC(go)(void *raw_addr, size_t raw_len,unsigned int i, syms_nr;unsigned long j;ELF (Shdr) *symtab_hdr =NULL , *strtab_hdr, *secstrings_hdr, - *alt_sec =NULL ; + *alt_sec =NULL , *retcall_sec =NULL ;ELF (Dyn) *dyn =0 , *dyn_end =0 ;const char *secstrings;INT_BITS syms[NSYMS ] = {};
@@ -78,6 +78,9 @@ static void BITSFUNC(go)(void *raw_addr, size_t raw_len,if (!strcmp (secstrings +GET_LE (&sh->sh_name),".altinstructions" )) alt_sec = sh; +if (!strcmp (secstrings +GET_LE (&sh->sh_name), +"__retcall_table" )) + retcall_sec = sh; }if (!symtab_hdr)
@@ -165,6 +168,12 @@ static void BITSFUNC(go)(void *raw_addr, size_t raw_len,fprintf (outfile,"\t.alt_len = %lu,\n" , (unsigned long )GET_LE (&alt_sec->sh_size)); } +if (retcall_sec) { +fprintf (outfile,"\t.retcall = %lu,\n" , + (unsigned long )GET_LE (&retcall_sec->sh_offset)); +fprintf (outfile,"\t.retcall_len = %lu,\n" , + (unsigned long )GET_LE (&retcall_sec->sh_size)); + }for (i =0 ; i <NSYMS ; i++) {if (required_syms[i].export && syms[i]) fprintf(outfile, "\t.sym_%s = %" PRIi64 ",\n",
arch/x86/entry/vdso/vma.c +22 -0
@@ -25,6 +25,7 @@#include <asm/cpufeature.h> #include <asm/mshyperv.h> #include <asm/page.h> +#include <asm/static_retcall.h> #include <asm/tlb.h> #if defined(CONFIG_X86_64 )
@@ -38,6 +39,25 @@ static __init int vdso_setup(char *s)__setup ("vdso=" , vdso_setup);#endif +static __init int apply_retcalls (struct retcall_entry *ent,unsigned long nr) +{ +while (nr--) { +void *call_addr = (void *)ent + ent->call; +void *ret_addr = (void *)ent + ent->ret; +size_t ret_sz = ent->out - ent->ret; + +if (WARN_ON (ret_sz >PAGE_SIZE )) + goto next; + +memcpy (call_addr, ret_addr, ret_sz); + +next: + ent++; + } + +return 0 ; +} +void __init init_vdso_image (struct vdso_image *image) {BUG_ON (image->size %PAGE_SIZE !=0 );
@@ -51,6 +71,8 @@ void __init init_vdso_image(struct vdso_image *image)return ;memcpy (image->text_timens, image->text, image->size); +apply_retcalls ((struct retcall_entry *)(image->text + image->retcall), + image->retcall_len /sizeof (struct retcall_entry ));#endif }
arch/x86/include/asm/static_retcall.h new file +54 -0
54 lines not on this page · raw message
arch/x86/include/asm/vdso.h +1 -0
@@ -16,6 +16,7 @@ struct vdso_image {unsigned long size;/* Always a multiple of PAGE_SIZE */ unsigned long alt, alt_len; +unsigned long retcall, retcall_len;long sym_vvar_start;/* Negative offset to the vvar area */
Signature · 1 line
2.20.1
RFC PATCH 1/1 vdso: introduce timens_static_branch
In reply to Andrei VaginAs it has been discussed on timens RFC, adding a new conditional branch if (inside_time_ns) on VDSO for all processes is undesirable.
Addressing those problems, there are two versions of VDSO's .so:
for host tasks (without any penalty) and for processes inside of time namespace with clk_to_ns() that subtracts offsets from host's time.
This patch introduces timens_static_branch(), which is similar with static_branch_unlikely.
The timens code in vdso looks like this:
if (timens_static_branch()) {
clk_to_ns(clk, ts);
}The version of vdso which is compiled from sources will never execute clk_to_ns(). And then we can patch the 'no-op' in the straight-line codepath with a 'jump' instruction to the out-of-line true branch and get the timens version of the vdso library.
--- arch/x86/entry/vdso/vclock_gettime.c | 21 ++++++++++++++------- arch/x86/entry/vdso/vdso-layout.lds.S | 1 + arch/x86/entry/vdso/vdso2c.h | 11 ++++++++++- arch/x86/entry/vdso/vma.c | 19 +++++++++++++++++++ arch/x86/include/asm/jump_label.h | 14 ++++++++++++++ arch/x86/include/asm/vdso.h | 1 + include/linux/jump_label.h | 5 +++++ 7 files changed, 64 insertions(+), 8 deletions(-)
arch/x86/entry/vdso/vclock_gettime.c +14 -7
@@ -18,6 +18,7 @@#include <asm/msr.h> #include <asm/pvclock.h> #include <asm/mshyperv.h> +#include <asm/jump_label.h> #include <linux/math64.h> #include <linux/time.h> #include <linux/kernel.h>
@@ -39,7 +40,7 @@ extern u8 hvclock_page __attribute__((visibility("hidden")));#endif -#ifdef BUILD_VDSO_TIME_NS +#ifdef CONFIG_TIME_NS extern u8 timens_page __attribute__((visibility("hidden")));#endif
@@ -145,9 +146,9 @@ notrace static inline u64 vgetcyc(int mode)return U64_MAX ; } +#ifdef CONFIG_TIME_NS notrace static __always_inline void clk_to_ns(clockid_t clk, struct timespec *ts) { -#ifdef BUILD_VDSO_TIME_NS struct timens_offsets *timens = (struct timens_offsets *) &timens_page;struct timespec64 *offset64;
@@ -173,9 +174,12 @@ notrace static __always_inline void clk_to_ns(clockid_t clk, struct timespec *ts ts->tv_nsec +=NSEC_PER_SEC ; ts->tv_sec--; } - -#endif } +#define _timens_static_branch_unlikely timens_static_branch_unlikely +#else +notrace static __always_inline void clk_to_ns(clockid_t clk, struct timespec *ts) {} +notrace static __always_inline bool _timens_static_branch_unlikely(void) { return false; } +#endif notrace static int do_hres(clockid_t clk, struct timespec *ts) {
@@ -203,8 +207,9 @@ notrace static int do_hres(clockid_t clk, struct timespec *ts) ts->tv_sec = sec +__iter_div_u64_rem (ns,NSEC_PER_SEC , &ns); ts->tv_nsec = ns; -clk_to_ns (clk, ts); - +if (_timens_static_branch_unlikely ()) { +clk_to_ns (clk, ts); + }return 0 ; }
@@ -219,7 +224,9 @@ notrace static void do_coarse(clockid_t clk, struct timespec *ts) ts->tv_nsec = base->nsec; }while (unlikely (gtod_read_retry (gtod, seq))); -clk_to_ns (clk, ts); +if (_timens_static_branch_unlikely ()) { +clk_to_ns (clk, ts); + } } notrace int __vdso_clock_gettime(clockid_t clock, struct timespec *ts)
arch/x86/entry/vdso/vdso-layout.lds.S +1 -0
@@ -45,6 +45,7 @@ SECTIONS .gnu.version : { *(.gnu.version) } .gnu.version_d : { *(.gnu.version_d) } .gnu.version_r : { *(.gnu.version_r) } + __jump_table : { *(__jump_table) } :text .dynamic : { *(.dynamic) } :text :dynamic
arch/x86/entry/vdso/vdso2c.h +10 -1
@@ -16,7 +16,7 @@ static void BITSFUNC(go)(void *raw_addr, size_t raw_len,unsigned int i, syms_nr;unsigned long j;ELF (Shdr) *symtab_hdr =NULL , *strtab_hdr, *secstrings_hdr, - *alt_sec =NULL ; + *alt_sec =NULL , *jump_table_sec =NULL ;ELF (Dyn) *dyn =0 , *dyn_end =0 ;const char *secstrings;INT_BITS syms[NSYMS ] = {};
@@ -78,6 +78,9 @@ static void BITSFUNC(go)(void *raw_addr, size_t raw_len,if (!strcmp (secstrings +GET_LE (&sh->sh_name),".altinstructions" )) alt_sec = sh; +if (!strcmp (secstrings +GET_LE (&sh->sh_name), +"__jump_table" )) + jump_table_sec = sh; }if (!symtab_hdr)
@@ -165,6 +168,12 @@ static void BITSFUNC(go)(void *raw_addr, size_t raw_len,fprintf (outfile,"\t.alt_len = %lu,\n" , (unsigned long )GET_LE (&alt_sec->sh_size)); } +if (jump_table_sec) { +fprintf (outfile,"\t.jump_table = %lu,\n" , + (unsigned long )GET_LE (&jump_table_sec->sh_offset)); +fprintf (outfile,"\t.jump_table_len = %lu,\n" , + (unsigned long )GET_LE (&jump_table_sec->sh_size)); + }for (i =0 ; i <NSYMS ; i++) {if (required_syms[i].export && syms[i]) fprintf(outfile, "\t.sym_%s = %" PRIi64 ",\n",
arch/x86/entry/vdso/vma.c +19 -0
@@ -15,6 +15,7 @@#include <linux/cpu.h> #include <linux/ptrace.h> #include <linux/time_namespace.h> +#include <linux/jump_label.h> #include <asm/pvclock.h> #include <asm/vgtod.h> #include <asm/proto.h>
@@ -38,6 +39,22 @@ static __init int vdso_setup(char *s)__setup ("vdso=" , vdso_setup);#endif +#ifdef CONFIG_TIME_NS +static __init int apply_jump_tables (struct vdso_jump_entry *ent,unsigned long nr) +{ +while (nr--) { +void *code_addr = (void *)ent + ent->code; +long target_addr = (long ) ent->target - (ent->code +JUMP_LABEL_NOP_SIZE ); + ((char *)code_addr)[0 ] =0xe9 ;/* JMP rel32 */ + *((long *)(code_addr +1 )) = (long )target_addr; + + ent++; + } + +return 0 ; +} +#endif +void __init init_vdso_image (struct vdso_image *image) {BUG_ON (image->size %PAGE_SIZE !=0 );
@@ -51,6 +68,8 @@ void __init init_vdso_image(struct vdso_image *image)return ;memcpy (image->text_timens, image->text, image->size); +apply_jump_tables ((struct vdso_jump_entry *)(image->text_timens + image->jump_table), + image->jump_table_len /sizeof (struct vdso_jump_entry ));#endif }
arch/x86/include/asm/jump_label.h +14 -0
@@ -51,6 +51,20 @@ static __always_inline bool arch_static_branch_jump(struct static_key *key, boolreturn true; } +static __always_inline bool timens_static_branch_unlikely(void) +{ + asm_volatile_goto("1:\n\t" + ".byte " __stringify(STATIC_KEY_INIT_NOP) "\n\t" + ".pushsection __jump_table, \"aw\"\n\t" + "2: .word 1b - 2b, %l[l_yes] - 2b\n\t" + ".popsection\n\t" + : : : : l_yes); + +return false; +l_yes: +return true; +} +#else /* __ASSEMBLY__ */ .macro STATIC_JUMP_IF_TRUE target, key, def
arch/x86/include/asm/vdso.h +1 -0
@@ -16,6 +16,7 @@ struct vdso_image {unsigned long size;/* Always a multiple of PAGE_SIZE */ unsigned long alt, alt_len; +unsigned long jump_table, jump_table_len;long sym_vvar_start;/* Negative offset to the vvar area */
include/linux/jump_label.h +5 -0
@@ -125,6 +125,11 @@ struct jump_entry {long key;// key may be far away from the core kernel under KASLR }; +struct vdso_jump_entry { +u16 code; +u16 target; +}; +static inline unsigned long jump_entry_code (const struct jump_entry *entry) {return (unsigned long )&entry->code + entry->code;
Signature · 1 line
2.20.1