TTY Keyboard Status Request
Subsystems (15)
v3 cover TTY Keyboard Status Request
This patch series introduces TTY keyboard status request, a feature of the n_tty line discipline that reserves a character in struct termios (^T by default) and reacts to it by printing a short informational line to the terminal and sending a Unix signal to the tty's foreground process group. The processes may, in response to the signal, output a textual description of what they're doing.
The feature has been present in a similar form at least in Free/Open/NetBSD; it would be nice to have something like this in Linux as well. There is an LKML thread[1] where users have previously expressed the rationale for this.
The current implementation does not break existing kernel API in any way, since, fortunately, all the architectures supported by the kernel happen to have at least 1 free byte in the termios control character array.
Patches 1-4 implement the semantics of the new signal and VSTATUS cc; patches 5-7 implement the kerninfo line written to the terminal by n_tty.
The series should cleanly apply to tty-next and 5.7-rc3.
[1] https://lore.kernel.org/lkml/1415200663.3247743.187387481.75CE9317@webmail.messagingengine.com/
v2 <- v1: removed useless debugging bits.
v3 <- v2: made commit message clarifications sparked by previous discussion, adapted 7/7 for y2038, re-Cc linux-api@.
Discussion of v1:
https://lore.kernel.org/lkml/[email protected]/
Discussion of v2:
https://lore.kernel.org/lkml/[email protected]/
Arseny Maslennikov (7):
signal.h: Define SIGINFO on all architectures
tty: termios: Reserve space for VSTATUS in .c_cc
n_tty: Send SIGINFO to fg pgrp on status request character linux/signal.h: Ignore SIGINFO by default in new tasks
tty: Add NOKERNINFO lflag to termios
n_tty: ->ops->write: Cut core logic out to a separate function
n_tty: Provide an informational line on VSTATUS receipt
arch/alpha/include/asm/termios.h | 4 +- arch/alpha/include/uapi/asm/termbits.h | 2 + arch/arm/include/uapi/asm/signal.h | 1 + arch/h8300/include/uapi/asm/signal.h | 1 + arch/ia64/include/asm/termios.h | 4 +- arch/ia64/include/uapi/asm/signal.h | 1 + arch/ia64/include/uapi/asm/termbits.h | 2 + arch/m68k/include/uapi/asm/signal.h | 1 + arch/mips/include/asm/termios.h | 4 +- arch/mips/include/uapi/asm/signal.h | 1 + arch/mips/include/uapi/asm/termbits.h | 2 + arch/parisc/include/asm/termios.h | 4 +- arch/parisc/include/uapi/asm/signal.h | 1 + arch/parisc/include/uapi/asm/termbits.h | 2 + arch/powerpc/include/asm/termios.h | 4 +- arch/powerpc/include/uapi/asm/signal.h | 1 + arch/powerpc/include/uapi/asm/termbits.h | 2 + arch/s390/include/asm/termios.h | 4 +- arch/s390/include/uapi/asm/signal.h | 1 + arch/sparc/include/asm/termios.h | 4 +- arch/sparc/include/uapi/asm/signal.h | 2 + arch/sparc/include/uapi/asm/termbits.h | 2 + arch/x86/include/uapi/asm/signal.h | 1 + arch/xtensa/include/uapi/asm/signal.h | 1 + arch/xtensa/include/uapi/asm/termbits.h | 2 + drivers/tty/Makefile | 3 +- drivers/tty/n_tty.c | 70 ++++- drivers/tty/n_tty_status.c | 338 +++++++++++++++++++++++ include/asm-generic/termios.h | 4 +- include/linux/sched.h | 7 + include/linux/signal.h | 5 +- include/linux/tty.h | 7 +- include/uapi/asm-generic/signal.h | 1 + include/uapi/asm-generic/termbits.h | 2 + 34 files changed, 458 insertions(+), 33 deletions(-) create mode 100644 drivers/tty/n_tty_status.c
Signature · 1 line
2.26.2
PATCH v3 1/7 signal.h: Define SIGINFO on all architectures
This complementary patch defines SIGINFO as a synonym for SIGPWR on every architecture supported by the kernel.
SIGPWR looks like a nice candidate for this role, because it is defined on every supported arch; it is currently only used to inform PID 1 of power failures, and daemons that care about low-level events do not tend to have a controlling terminal.
However, on sparcs SIGPWR is a synonym for SIGLOST, a signal unique to that architecture, with a narrow set of intended uses that do not combine well with interactively requesting status. SIGLOST is not used by any kernel code at the moment.
SIGINFO is not defined as a real-time signal for the following reasons:
* SIGRTMIN and SIGRTMAX can not be changed without breaking userspace
* the lack of queuing is irrelevant for the main use-case, since the consumption rate of information by a terminal user and the practical tty keypress processing rates are much slower than SIGINFO processing by the line discipline.
--- arch/arm/include/uapi/asm/signal.h | 1 + arch/h8300/include/uapi/asm/signal.h | 1 + arch/ia64/include/uapi/asm/signal.h | 1 + arch/m68k/include/uapi/asm/signal.h | 1 + arch/mips/include/uapi/asm/signal.h | 1 + arch/parisc/include/uapi/asm/signal.h | 1 + arch/powerpc/include/uapi/asm/signal.h | 1 + arch/s390/include/uapi/asm/signal.h | 1 + arch/sparc/include/uapi/asm/signal.h | 2 ++ arch/x86/include/uapi/asm/signal.h | 1 + arch/xtensa/include/uapi/asm/signal.h | 1 + include/uapi/asm-generic/signal.h | 1 + 12 files changed, 13 insertions(+)
arch/arm/include/uapi/asm/signal.h +1 -0
@@ -50,6 +50,7 @@ typedef unsigned long sigset_t;#define SIGLOST 29 */ #define SIGPWR 30 +#define SIGINFO SIGPWR#define SIGSYS 31#define SIGUNUSED 31
arch/h8300/include/uapi/asm/signal.h +1 -0
@@ -50,6 +50,7 @@ typedef unsigned long sigset_t;#define SIGLOST 29 */ #define SIGPWR 30 +#define SIGINFO SIGPWR#define SIGSYS 31#define SIGUNUSED 31
arch/ia64/include/uapi/asm/signal.h +1 -0
@@ -45,6 +45,7 @@#define SIGLOST 29 */ #define SIGPWR 30 +#define SIGINFO SIGPWR#define SIGSYS 31/* signal 31 is no longer "unused", but the SIGUNUSED macro remains for backwards compatibility */ #define SIGUNUSED 31
arch/m68k/include/uapi/asm/signal.h +1 -0
@@ -50,6 +50,7 @@ typedef unsigned long sigset_t;#define SIGLOST 29 */ #define SIGPWR 30 +#define SIGINFO SIGPWR#define SIGSYS 31#define SIGUNUSED 31
arch/mips/include/uapi/asm/signal.h +1 -0
@@ -43,6 +43,7 @@ typedef unsigned long old_sigset_t; /* at least 32 bits */#define SIGCHLD 18/* Child status has changed (POSIX). */ #define SIGCLD SIGCHLD/* Same as SIGCHLD (System V). */ #define SIGPWR 19/* Power failure restart (System V). */ +#define SIGINFO SIGPWR/* Keyboard status request (4.2 BSD). */ #define SIGWINCH 20/* Window size change (4.3 BSD, Sun). */ #define SIGURG 21/* Urgent condition on socket (4.2 BSD). */ #define SIGIO 22/* I/O now possible (4.2 BSD). */
arch/parisc/include/uapi/asm/signal.h +1 -0
@@ -22,6 +22,7 @@#define SIGUSR2 17#define SIGCHLD 18#define SIGPWR 19 +#define SIGINFO SIGPWR#define SIGVTALRM 20#define SIGPROF 21#define SIGIO 22
arch/powerpc/include/uapi/asm/signal.h +1 -0
@@ -53,6 +53,7 @@ typedef struct {#define SIGLOST 29 */ #define SIGPWR 30 +#define SIGINFO SIGPWR#define SIGSYS 31#define SIGUNUSED 31
arch/s390/include/uapi/asm/signal.h +1 -0
@@ -58,6 +58,7 @@ typedef unsigned long sigset_t;#define SIGLOST 29 */ #define SIGPWR 30 +#define SIGINFO SIGPWR#define SIGSYS 31#define SIGUNUSED 31
arch/sparc/include/uapi/asm/signal.h +2 -0
@@ -71,6 +71,8 @@#define SIGWINCH 28#define SIGLOST 29#define SIGPWR SIGLOST +/* XXX: is it OK for SIGINFO to collide with LOST? */ +#define SIGINFO SIGPWR#define SIGUSR1 30#define SIGUSR2 31
arch/x86/include/uapi/asm/signal.h +1 -0
@@ -55,6 +55,7 @@ typedef unsigned long sigset_t;#define SIGLOST 29 */ #define SIGPWR 30 +#define SIGINFO SIGPWR#define SIGSYS 31#define SIGUNUSED 31
arch/xtensa/include/uapi/asm/signal.h +1 -0
@@ -65,6 +65,7 @@ typedef struct {#define SIGPOLL SIGIO/* #define SIGLOST 29 */ #define SIGPWR 30 +#define SIGINFO SIGPWR#define SIGSYS 31#define SIGUNUSED 31
include/uapi/asm-generic/signal.h +1 -0
@@ -43,6 +43,7 @@#define SIGLOST 29 */ #define SIGPWR 30 +#define SIGINFO SIGPWR#define SIGSYS 31#define SIGUNUSED 31
Signature · 1 line
2.26.2
PATCH v3 2/7 tty: termios: Reserve space for VSTATUS in .c_cc
In reply to Arseny MaslennikovThere was no architecture we had to increase NCCS on, so the size of struct termios did not change.
--- arch/alpha/include/asm/termios.h | 4 ++-- arch/alpha/include/uapi/asm/termbits.h | 1 + arch/ia64/include/asm/termios.h | 4 ++-- arch/ia64/include/uapi/asm/termbits.h | 1 + arch/mips/include/asm/termios.h | 4 ++-- arch/mips/include/uapi/asm/termbits.h | 1 + arch/parisc/include/asm/termios.h | 4 ++-- arch/parisc/include/uapi/asm/termbits.h | 1 + arch/powerpc/include/asm/termios.h | 4 ++-- arch/powerpc/include/uapi/asm/termbits.h | 1 + arch/s390/include/asm/termios.h | 4 ++-- arch/sparc/include/asm/termios.h | 4 ++-- arch/sparc/include/uapi/asm/termbits.h | 1 + arch/xtensa/include/uapi/asm/termbits.h | 1 + include/asm-generic/termios.h | 4 ++-- include/uapi/asm-generic/termbits.h | 1 + 16 files changed, 24 insertions(+), 16 deletions(-)
arch/alpha/include/asm/termios.h +2 -2
@@ -8,9 +8,9 @@werase=^W kill=^U reprint=^R sxtc=\0 intr=^C quit=^\ susp=^Z <OSF/1 VDSUSP> start=^Q stop=^S lnext=^V discard=^U -vmin=\1 vtime=\0 +vmin=\1 vtime=\0 status=^T */ -#define INIT_C_CC "\004\000\000\177\027\025\022\000\003\034\032\000\021\023\026\025\001\000" +#define INIT_C_CC "\004\000\000\177\027\025\022\000\003\034\032\000\021\023\026\025\001\000\024"/* * Translate a "termio" structure into a "termios". Ugh.
arch/alpha/include/uapi/asm/termbits.h +1 -0
@@ -70,6 +70,7 @@ struct ktermios {#define VDISCARD 15#define VMIN 16#define VTIME 17 +#define VSTATUS 18/* c_iflag bits */ #define IGNBRK 0000001
arch/ia64/include/asm/termios.h +2 -2
@@ -15,9 +15,9 @@eof=^D vtime=\0 vmin=\1 sxtc=\0 start=^Q stop=^S susp=^Z eol=\0 reprint=^R discard=^U werase=^W lnext=^V -eol2=\0 +eol2=\0 status=^T */ -#define INIT_C_CC "\003\034\177\025\004\0\1\0\021\023\032\0\022\017\027\026\0" +#define INIT_C_CC "\003\034\177\025\004\0\1\0\021\023\032\0\022\017\027\026\0\024"/* * Translate a "termio" structure into a "termios". Ugh.
arch/ia64/include/uapi/asm/termbits.h +1 -0
@@ -67,6 +67,7 @@ struct ktermios {#define VWERASE 14#define VLNEXT 15#define VEOL2 16 +#define VSTATUS 17/* c_iflag bits */ #define IGNBRK 0000001
arch/mips/include/asm/termios.h +2 -2
@@ -17,9 +17,9 @@* vmin=\1 vtime=\0 eol2=\0 swtc=\0 * start=^Q stop=^S susp=^Z vdsusp= * reprint=^R discard=^U werase=^W lnext=^V -* eof=^D eol=\0 +* eof=^D eol=\0 status=^T */ -#define INIT_C_CC "\003\034\177\025\1\0\0\0\021\023\032\0\022\017\027\026\004\0" +#define INIT_C_CC "\003\034\177\025\1\0\0\0\021\023\032\0\022\017\027\026\004\0\024"#include <linux/string.h>
arch/mips/include/uapi/asm/termbits.h +1 -0
@@ -78,6 +78,7 @@ struct ktermios {#define VLNEXT 15/* Literal-next character [IEXTEN]. */ #define VEOF 16/* End-of-file character [ICANON]. */ #define VEOL 17/* End-of-line character [ICANON]. */ +#define VSTATUS 18/* Status request character [ISIG]. */ /* c_iflag bits */ #define IGNBRK 0000001/* Ignore break condition. */
arch/parisc/include/asm/termios.h +2 -2
@@ -9,9 +9,9 @@eof=^D vtime=\0 vmin=\1 sxtc=\0 start=^Q stop=^S susp=^Z eol=\0 reprint=^R discard=^U werase=^W lnext=^V -eol2=\0 +eol2=\0 status=^T */ -#define INIT_C_CC "\003\034\177\025\004\0\1\0\021\023\032\0\022\017\027\026\0" +#define INIT_C_CC "\003\034\177\025\004\0\1\0\021\023\032\0\022\017\027\026\0\024"/* * Translate a "termio" structure into a "termios". Ugh.
arch/parisc/include/uapi/asm/termbits.h +1 -0
@@ -58,6 +58,7 @@ struct ktermios {#define VWERASE 14#define VLNEXT 15#define VEOL2 16 +#define VSTATUS 17/* c_iflag bits */
arch/powerpc/include/asm/termios.h +2 -2
@@ -10,8 +10,8 @@#include <uapi/asm/termios.h> -/* ^C ^\ del ^U ^D 1 0 0 0 0 ^W ^R ^Z ^Q ^S ^V ^U */ -#define INIT_C_CC "\003\034\177\025\004\001\000\000\000\000\027\022\032\021\023\026\025" +/* ^C ^\ del ^U ^D 1 0 0 0 0 ^W ^R ^Z ^Q ^S ^V ^U ^T */ +#define INIT_C_CC "\003\034\177\025\004\001\000\000\000\000\027\022\032\021\023\026\025\024"#include <asm-generic/termios-base.h>
arch/powerpc/include/uapi/asm/termbits.h +1 -0
@@ -62,6 +62,7 @@ struct ktermios {#define VSTOP 14#define VLNEXT 15#define VDISCARD 16 +#define VSTATUS 17/* c_iflag bits */ #define IGNBRK 0000001
arch/s390/include/asm/termios.h +2 -2
@@ -14,9 +14,9 @@eof=^D vtime=\0 vmin=\1 sxtc=\0 start=^Q stop=^S susp=^Z eol=\0 reprint=^R discard=^U werase=^W lnext=^V -eol2=\0 +eol2=\0 status=^T */ -#define INIT_C_CC "\003\034\177\025\004\0\1\0\021\023\032\0\022\017\027\026\0" +#define INIT_C_CC "\003\034\177\025\004\0\1\0\021\023\032\0\022\017\027\026\0\024"#define user_termios_to_kernel_termios (k, u) copy_from_user(k, u, sizeof(struct termios2))#define kernel_termios_to_user_termios (u, k) copy_to_user(u, k, sizeof(struct termios2))
arch/sparc/include/asm/termios.h +2 -2
@@ -20,9 +20,9 @@eof=^D eol=\0 eol2=\0 sxtc=\0 start=^Q stop=^S susp=^Z dsusp=^Y reprint=^R discard=^U werase=^W lnext=^V -vmin=\1 vtime=\0 +vmin=\1 vtime=\0 status=^T */ -#define INIT_C_CC "\003\034\177\025\004\000\000\000\021\023\032\031\022\025\027\026\001" +#define INIT_C_CC "\003\034\177\025\004\000\000\000\021\023\032\031\022\025\027\026\001\000\024"/* * Translate a "termio" structure into a "termios". Ugh.
arch/sparc/include/uapi/asm/termbits.h +1 -0
@@ -80,6 +80,7 @@ struct ktermios {#define VDISCARD 13#define VWERASE 14#define VLNEXT 15 +#define VSTATUS 16/* Kernel keeps vmin/vtime separated, user apps assume vmin/vtime is * shared with eof/eol
arch/xtensa/include/uapi/asm/termbits.h +1 -0
@@ -72,6 +72,7 @@ struct ktermios {#define VWERASE 14#define VLNEXT 15#define VEOL2 16 +#define VSTATUS 17/* c_iflag bits */
include/asm-generic/termios.h +2 -2
@@ -10,9 +10,9 @@eof=^D vtime=\0 vmin=\1 sxtc=\0 start=^Q stop=^S susp=^Z eol=\0 reprint=^R discard=^U werase=^W lnext=^V -eol2=\0 +eol2=\0 status=^T */ -#define INIT_C_CC "\003\034\177\025\004\0\1\0\021\023\032\0\022\017\027\026\0" +#define INIT_C_CC "\003\034\177\025\004\0\1\0\021\023\032\0\022\017\027\026\0\024"/* * Translate a "termio" structure into a "termios". Ugh.
include/uapi/asm-generic/termbits.h +1 -0
@@ -58,6 +58,7 @@ struct ktermios {#define VWERASE 14#define VLNEXT 15#define VEOL2 16 +#define VSTATUS 17/* c_iflag bits */ #define IGNBRK 0000001
Signature · 1 line
2.26.2
PATCH v3 3/7 n_tty: Send SIGINFO to fg pgrp on status request character
In reply to Arseny MaslennikovNo kerninfo line is printed yet.
No existing implementation of this on any Unix-like system echoes the status character; no existing implementation discards or flushes pending input on VSTATUS receipt. Thus we do neither.
There are existing popular TUI applications (e. g. mutt) that only turn off icanon and not iexten, but still do not expect any special treatment of the status request character — so we require all three:
isig, icanon and iexten to trigger this.
--- drivers/tty/n_tty.c | 15 +++++++++++++-- include/linux/tty.h | 1 + 2 files changed, 14 insertions(+), 2 deletions(-)
drivers/tty/n_tty.c +13 -2
@@ -79,6 +79,10 @@#define ECHO_BLOCK 256#define ECHO_DISCARD_WATERMARK N_TTY_BUF_SIZE - (ECHO_BLOCK + 32) +#define SIG_FLUSHING_MASK ( \ + rt_sigmask(SIGINT) | rt_sigmask(SIGQUIT) | \ + rt_sigmask(SIGTSTP) ) +#define SIG_FLUSHING (sig) ((1 << sig) & SIG_FLUSHING_MASK)#undef N_TTY_TRACE#ifdef N_TTY_TRACE
@@ -1122,7 +1126,7 @@ static void isig(int sig, struct tty_struct *tty) {struct n_tty_data *ldata = tty->disc_data; -if (L_NOFLSH (tty)) { +if (L_NOFLSH (tty) || (!SIG_FLUSHING (sig))) {/* signal only */ __isig (sig, tty);
@@ -1244,7 +1248,8 @@ n_tty_receive_signal_char(struct tty_struct *tty, int signal, unsigned char c)if (I_IXON (tty))start_tty (tty);if (L_ECHO (tty)) { -echo_char (c, tty); +if (c !=STATUS_CHAR (tty)) +echo_char (c, tty);commit_echoes (tty); }else process_echoes (tty);
@@ -1294,6 +1299,9 @@ n_tty_receive_char_special(struct tty_struct *tty, unsigned char c) }else if (c ==SUSP_CHAR (tty)) {n_tty_receive_signal_char (tty,SIGTSTP , c);return 0 ; + }else if (c ==STATUS_CHAR (tty)) { +n_tty_receive_signal_char (tty,SIGINFO , c); +return 0 ; } }
@@ -1848,6 +1856,9 @@ static void n_tty_set_termios(struct tty_struct *tty, struct ktermios *old)set_bit (INTR_CHAR (tty), ldata->char_map);set_bit (QUIT_CHAR (tty), ldata->char_map);set_bit (SUSP_CHAR (tty), ldata->char_map); +if (L_ICANON (tty) &&L_IEXTEN (tty)) { +set_bit (STATUS_CHAR (tty), ldata->char_map); + } }clear_bit (__DISABLED_CHAR, ldata->char_map); ldata->raw =0 ;
include/linux/tty.h +1 -0
@@ -120,6 +120,7 @@ struct tty_bufhead {#define WERASE_CHAR (tty) ((tty)->termios.c_cc[VWERASE])#define LNEXT_CHAR (tty) ((tty)->termios.c_cc[VLNEXT])#define EOL2_CHAR (tty) ((tty)->termios.c_cc[VEOL2]) +#define STATUS_CHAR (tty) ((tty)->termios.c_cc[VSTATUS])#define _I_FLAG (tty, f) ((tty)->termios.c_iflag & (f))#define _O_FLAG (tty, f) ((tty)->termios.c_oflag & (f))
Signature · 1 line
2.26.2
PATCH v3 4/7 linux/signal.h: Ignore SIGINFO by default in new tasks
In reply to Arseny MaslennikovThis matches the behaviour of other Unix-like systems that have SIGINFO and causes less harm to processes that do not install handlers for this signal, making the keyboard status character non-fatal for them.
This is implemented with the assumption that SIGINFO is defined to be equivalent to SIGPWR; still, there is no reason for PWR to result in termination of the signal recipient anyway — it does not indicate there is a fatal problem with the recipient's execution context (like e.g. FPE/ILL do), and we have TERM/KILL for explicit termination requests.
To put it another way:
The only scenario where system behaviour actually changes is when the signal recipient has default disposition for SIGPWR. If a process chose to interpret a SIGPWR as an incentive to cleanly terminate, it would supply its own handler — and this commit does not affect processes with non-default handlers.
--- include/linux/signal.h | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
include/linux/signal.h +3 -2
@@ -369,7 +369,7 @@ extern bool unhandled_signal(struct task_struct *tsk, int sig);* | SIGSYS/SIGUNUSED | coredump | * | SIGSTKFLT | terminate | * | SIGWINCH | ignore | -* | SIGPWR | terminate | +* | SIGPWR | ignore | * | SIGRTMIN-SIGRTMAX | terminate | * +--------------------+------------------+ * | non-POSIX signal | default action |
@@ -420,7 +420,8 @@ extern bool unhandled_signal(struct task_struct *tsk, int sig);#define SIG_KERNEL_IGNORE_MASK (\ rt_sigmask(SIGCONT) | rt_sigmask(SIGCHLD) | \ - rt_sigmask(SIGWINCH) | rt_sigmask(SIGURG) ) + rt_sigmask(SIGWINCH) | rt_sigmask(SIGURG) | \ + rt_sigmask(SIGINFO) )#define SIG_SPECIFIC_SICODES_MASK (\ rt_sigmask(SIGILL) | rt_sigmask(SIGFPE) | \
Signature · 1 line
2.26.2
Re: [PATCH v3 4/7] linux/signal.h: Ignore SIGINFO by default in new tasks
Quoted 28 lines · Arseny Maslennikov
This matches the behaviour of other Unix-like systems that have SIGINFO and causes less harm to processes that do not install handlers for this signal, making the keyboard status character non-fatal for them.
This is implemented with the assumption that SIGINFO is defined to be equivalent to SIGPWR; still, there is no reason for PWR to result in termination of the signal recipient anyway — it does not indicate there is a fatal problem with the recipient's execution context (like e.g. FPE/ILL do), and we have TERM/KILL for explicit termination requests.
To put it another way:
The only scenario where system behaviour actually changes is when the signal recipient has default disposition for SIGPWR. If a process chose to interpret a SIGPWR as an incentive to cleanly terminate, it would supply its own handler — and this commit does not affect processes with non-default handlers.Signed-off-by Arseny Maslennikov <[email protected]>--- include/linux/signal.h | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)include/linux/signal.h +0
-0
include/linux/signal.h extern bool unhandled_signal(struct task_struct *tsk, int sig); · line 372
Quoted 5 lines
include/linux/signal.h +1
-1@@ -369,7 +369,7 @@ extern bool unhandled_signal(struct task_struct *tsk, int sig);* | SIGSYS/SIGUNUSED | coredump | * | SIGSTKFLT | terminate | * | SIGWINCH | ignore | -* | SIGPWR | terminate | +* | SIGPWR | ignore |
You need to update signal.7 too:
https://git.kernel.org/pub/scm/docs/man-pages/man-pages.git/tree/man7/signal.7#n285
thanks,
Signature · 2 lines
js suse labs
Re: [PATCH v3 4/7] linux/signal.h: Ignore SIGINFO by default in new tasks
Quoted 24 lines · Jiri Slaby
Quoted 23 lines · Arseny Maslennikov
This matches the behaviour of other Unix-like systems that have SIGINFO and causes less harm to processes that do not install handlers for this signal, making the keyboard status character non-fatal for them.
This is implemented with the assumption that SIGINFO is defined to be equivalent to SIGPWR; still, there is no reason for PWR to result in termination of the signal recipient anyway — it does not indicate there is a fatal problem with the recipient's execution context (like e.g. FPE/ILL do), and we have TERM/KILL for explicit termination requests.
To put it another way:
The only scenario where system behaviour actually changes is when the signal recipient has default disposition for SIGPWR. If a process chose to interpret a SIGPWR as an incentive to cleanly terminate, it would supply its own handler — and this commit does not affect processes with non-default handlers.Signed-off-by Arseny Maslennikov <[email protected]>--- include/linux/signal.h | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
Quoted 13 lines
include/linux/signal.h extern bool unhandled_signal(struct task_struct *tsk, int sig); · line 372
Quoted 10 lines
include/linux/signal.h +1
-1@@ -369,7 +369,7 @@ extern bool unhandled_signal(struct task_struct *tsk, int sig);* | SIGSYS/SIGUNUSED | coredump | * | SIGSTKFLT | terminate | * | SIGWINCH | ignore | -* | SIGPWR | terminate | +* | SIGPWR | ignore | You need to update signal.7 too:
https://git.kernel.org/pub/scm/docs/man-pages/man-pages.git/tree/man7/signal.7#n285
(I fail this whole thread via b4 and it appears that a bunch of messages are missing on lore. Might just be delay though.)
How this is this not going to break userspace?
Just for a start, SIGPWR (for better or worse) was used for a long time by some sandboxing/container runtimes to shutdown a process and still is.
Christian
Greg Kroah-Hartman
maintainerRe: [PATCH v3 4/7] linux/signal.h: Ignore SIGINFO by default in new tasks
Quoted 39 lines · Christian Brauner
Quoted 24 lines · Jiri Slaby
Quoted 23 lines · Arseny Maslennikov
This matches the behaviour of other Unix-like systems that have SIGINFO and causes less harm to processes that do not install handlers for this signal, making the keyboard status character non-fatal for them.
This is implemented with the assumption that SIGINFO is defined to be equivalent to SIGPWR; still, there is no reason for PWR to result in termination of the signal recipient anyway — it does not indicate there is a fatal problem with the recipient's execution context (like e.g. FPE/ILL do), and we have TERM/KILL for explicit termination requests.
To put it another way:
The only scenario where system behaviour actually changes is when the signal recipient has default disposition for SIGPWR. If a process chose to interpret a SIGPWR as an incentive to cleanly terminate, it would supply its own handler — and this commit does not affect processes with non-default handlers.Signed-off-by Arseny Maslennikov <[email protected]>--- include/linux/signal.h | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)Quoted 13 lines
include/linux/signal.h extern bool unhandled_signal(struct task_struct *tsk, int sig); · line 372
Quoted 10 lines
include/linux/signal.h +1
-1@@ -369,7 +369,7 @@ extern bool unhandled_signal(struct task_struct *tsk, int sig);* | SIGSYS/SIGUNUSED | coredump | * | SIGSTKFLT | terminate | * | SIGWINCH | ignore | -* | SIGPWR | terminate | +* | SIGPWR | ignore | You need to update signal.7 too:
https://git.kernel.org/pub/scm/docs/man-pages/man-pages.git/tree/man7/signal.7#n285
Quoted 4 lines
(I fail this whole thread via b4 and it appears that a bunch of messages are missing on lore. Might just be delay though.)
How this is this not going to break userspace?
That's my main hesitation for taking this patchset.
Quoted 3 lines
Just for a start, SIGPWR (for better or worse) was used for a long time by some sandboxing/container runtimes to shutdown a process and still is.
That's a good reason to not do this :(
greg k-h
Re: [PATCH v3 4/7] linux/signal.h: Ignore SIGINFO by default in new tasks
In reply to Christian BraunerQuoted 39 lines · Christian Brauner
Quoted 24 lines · Jiri Slaby
Quoted 23 lines · Arseny Maslennikov
This matches the behaviour of other Unix-like systems that have SIGINFO and causes less harm to processes that do not install handlers for this signal, making the keyboard status character non-fatal for them.
This is implemented with the assumption that SIGINFO is defined to be equivalent to SIGPWR; still, there is no reason for PWR to result in termination of the signal recipient anyway — it does not indicate there is a fatal problem with the recipient's execution context (like e.g. FPE/ILL do), and we have TERM/KILL for explicit termination requests.
To put it another way:
The only scenario where system behaviour actually changes is when the signal recipient has default disposition for SIGPWR. If a process chose to interpret a SIGPWR as an incentive to cleanly terminate, it would supply its own handler — and this commit does not affect processes with non-default handlers.Signed-off-by Arseny Maslennikov <[email protected]>--- include/linux/signal.h | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)Quoted 13 lines
include/linux/signal.h extern bool unhandled_signal(struct task_struct *tsk, int sig); · line 372
Quoted 10 lines
include/linux/signal.h +1
-1@@ -369,7 +369,7 @@ extern bool unhandled_signal(struct task_struct *tsk, int sig);* | SIGSYS/SIGUNUSED | coredump | * | SIGSTKFLT | terminate | * | SIGWINCH | ignore | -* | SIGPWR | terminate | +* | SIGPWR | ignore | You need to update signal.7 too:
https://git.kernel.org/pub/scm/docs/man-pages/man-pages.git/tree/man7/signal.7#n285
Quoted 6 lines
(I fail this whole thread via b4 and it appears that a bunch of messages are missing on lore. Might just be delay though.)
How this is this not going to break userspace? Just for a start, SIGPWR (for better or worse) was used for a long time by some sandboxing/container runtimes to shutdown a process and still is.
To play Devil's advocate -- pid1 has also always had a default-ignore signal mask (which included SIGPWR), so any pid1 that obeyed SIGPWR already had a non-default signal mask (and thus wouldn't be affected by this patch).
But I do agree that this seems like a strange change to make (SIGPWR seems like a signal you don't want to ignore by default). Unfortunately the fact that it appears to always be equal to SIGINFO means that while SIGINFO (to me at least) seems like it should be a no-op, the necessary SIGPWR change makes it harder to justify IMHO.
Signature · 4 lines
Aleksa Sarai Senior Software Engineer (Containers) SUSE Linux GmbH <https://www.cyphar.com/>
Re: [PATCH v3 4/7] linux/signal.h: Ignore SIGINFO by default in new tasks
Quoted 43 lines · Aleksa Sarai
Quoted 42 lines · Christian Brauner
Quoted 24 lines · Jiri Slaby
Quoted 23 lines · Arseny Maslennikov
This matches the behaviour of other Unix-like systems that have SIGINFO and causes less harm to processes that do not install handlers for this signal, making the keyboard status character non-fatal for them.
This is implemented with the assumption that SIGINFO is defined to be equivalent to SIGPWR; still, there is no reason for PWR to result in termination of the signal recipient anyway — it does not indicate there is a fatal problem with the recipient's execution context (like e.g. FPE/ILL do), and we have TERM/KILL for explicit termination requests.
To put it another way:
The only scenario where system behaviour actually changes is when the signal recipient has default disposition for SIGPWR. If a process chose to interpret a SIGPWR as an incentive to cleanly terminate, it would supply its own handler — and this commit does not affect processes with non-default handlers.Signed-off-by Arseny Maslennikov <[email protected]>--- include/linux/signal.h | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)Quoted 13 lines
include/linux/signal.h extern bool unhandled_signal(struct task_struct *tsk, int sig); · line 372
Quoted 10 lines
include/linux/signal.h +1
-1@@ -369,7 +369,7 @@ extern bool unhandled_signal(struct task_struct *tsk, int sig);* | SIGSYS/SIGUNUSED | coredump | * | SIGSTKFLT | terminate | * | SIGWINCH | ignore | -* | SIGPWR | terminate | +* | SIGPWR | ignore | You need to update signal.7 too:
https://git.kernel.org/pub/scm/docs/man-pages/man-pages.git/tree/man7/signal.7#n285(I fail this whole thread via b4 and it appears that a bunch of messages are missing on lore. Might just be delay though.)
Quoted 8 lines
Quoted 3 lines
How this is this not going to break userspace? Just for a start, SIGPWR (for better or worse) was used for a long time by some sandboxing/container runtimes to shutdown a process and still is.
To play Devil's advocate -- pid1 has also always had a default-ignore signal mask (which included SIGPWR), so any pid1 that obeyed SIGPWR already had a non-default signal mask (and thus wouldn't be affected by this patch).
Sure, my point wasn't specifically about init systems but rather generic processes. The reason that SIGPWR was originally used was because older init systems - apart from systemd - could be shutdown with it while other programs left it set to SIG_DFL and they would terminate too. I'm not saying this is a great idea. But changing SIGPWR - if I read this right - to go from "terminate" to "ignore" will mean that any program that left SIGPWR unhandled/SIG_DFL will potentially see altered behavior. Just looking at gvfsd and in debian.codesearch shows that they explicitly set signal(SIGPWR, SIG_DFL) and I'm sure there are more.
(You also need to keep in mind that the default ignore mask applies to signals sent from _within_ the pid namespace to pid 1. They'll be delivered just fine from an ancestor pid namespace. Otherwise I'd be interested to know how you've ever shutdown one of your containers. ;))
Quoted 6 lines
But I do agree that this seems like a strange change to make (SIGPWR seems like a signal you don't want to ignore by default). Unfortunately the fact that it appears to always be equal to SIGINFO means that while SIGINFO (to me at least) seems like it should be a no-op, the necessary SIGPWR change makes it harder to justify IMHO.
PATCH v3 5/7 tty: Add NOKERNINFO lflag to termios
In reply to Arseny MaslennikovThe termios lflag is off by default.
--- arch/alpha/include/uapi/asm/termbits.h | 1 + arch/ia64/include/uapi/asm/termbits.h | 1 + arch/mips/include/uapi/asm/termbits.h | 1 + arch/parisc/include/uapi/asm/termbits.h | 1 + arch/powerpc/include/uapi/asm/termbits.h | 1 + arch/sparc/include/uapi/asm/termbits.h | 1 + arch/xtensa/include/uapi/asm/termbits.h | 1 + include/linux/tty.h | 1 + include/uapi/asm-generic/termbits.h | 1 + 9 files changed, 9 insertions(+)
arch/alpha/include/uapi/asm/termbits.h +1 -0
@@ -204,6 +204,7 @@ struct ktermios {#define PENDIN 0x20000000#define IEXTEN 0x00000400#define EXTPROC 0x10000000 +#define NOKERNINFO 0x40000000/* Values for the ACTION argument to `tcflow'. */ #define TCOOFF 0
arch/ia64/include/uapi/asm/termbits.h +1 -0
@@ -190,6 +190,7 @@ struct ktermios {#define PENDIN 0040000#define IEXTEN 0100000#define EXTPROC 0200000 +#define NOKERNINFO 0400000/* tcflow() and TCXONC use these */ #define TCOOFF 0
arch/mips/include/uapi/asm/termbits.h +1 -0
@@ -206,6 +206,7 @@ struct ktermios {#define TOSTOP 0100000/* Send SIGTTOU for background output. */ #define ITOSTOP TOSTOP#define EXTPROC 0200000/* External processing on pty */ +#define NOKERNINFO 0400000/* Disable kernel output from VSTATUS. */ /* ioctl (fd, TIOCSERGETLSR, &result) where result may be as below */ #define TIOCSER_TEMT 0x01/* Transmitter physically empty */
arch/parisc/include/uapi/asm/termbits.h +1 -0
@@ -183,6 +183,7 @@ struct ktermios {#define PENDIN 0040000#define IEXTEN 0100000#define EXTPROC 0200000 +#define NOKERNINFO 0400000/* tcflow() and TCXONC use these */ #define TCOOFF 0
arch/powerpc/include/uapi/asm/termbits.h +1 -0
@@ -192,6 +192,7 @@ struct ktermios {#define PENDIN 0x20000000#define IEXTEN 0x00000400#define EXTPROC 0x10000000 +#define NOKERNINFO 0x40000000/* Values for the ACTION argument to `tcflow'. */ #define TCOOFF 0
arch/sparc/include/uapi/asm/termbits.h +1 -0
@@ -225,6 +225,7 @@ struct ktermios {#define PENDIN 0x00004000#define IEXTEN 0x00008000#define EXTPROC 0x00010000 +#define NOKERNINFO 0x00020000/* modem lines */ #define TIOCM_LE 0x001
arch/xtensa/include/uapi/asm/termbits.h +1 -0
@@ -199,6 +199,7 @@ struct ktermios {#define PENDIN 0040000#define IEXTEN 0100000#define EXTPROC 0200000 +#define NOKERNINFO 0400000/* tcflow() and TCXONC use these */
include/linux/tty.h +1 -0
@@ -186,6 +186,7 @@ struct tty_bufhead {#define L_PENDIN (tty) _L_FLAG((tty), PENDIN)#define L_IEXTEN (tty) _L_FLAG((tty), IEXTEN)#define L_EXTPROC (tty) _L_FLAG((tty), EXTPROC) +#define L_NOKERNINFO (tty) _L_FLAG((tty), NOKERNINFO)struct device ;struct signal_struct ;
include/uapi/asm-generic/termbits.h +1 -0
@@ -181,6 +181,7 @@ struct ktermios {#define PENDIN 0040000#define IEXTEN 0100000#define EXTPROC 0200000 +#define NOKERNINFO 0400000/* tcflow() and TCXONC use these */ #define TCOOFF 0
Signature · 1 line
2.26.2
Re: [PATCH v3 5/7] tty: Add NOKERNINFO lflag to termios
Quoted 1 line · Arseny Maslennikov
The termios lflag is off by default.
This commit message is too poor. Describing the intended use would help.
thanks,
Signature · 2 lines
js suse labs
Re: [PATCH v3 5/7] tty: Add NOKERNINFO lflag to termios
Quoted 5 lines · Jiri Slaby
Quoted 1 line · Arseny Maslennikov
The termios lflag is off by default.
This commit message is too poor. Describing the intended use would help.
I described its use in the last patch of the series, where it actually gains some use.
I agree, however, that the limited explanation in this commit does not look helpful at all when looking at e.g. git blame output.
I'll resend a v4 with this commit message improved.
PATCH v3 6/7 n_tty: ->ops->write: Cut core logic out to a separate function
In reply to Arseny MaslennikovTo simplify internal re-use of the line discipline's write method, we isolate the work it does to its own function.
Since in-kernel callers might not refer to the tty through a file, the struct file* argument might make no sense, so we also stop tty_io_nonblock() from dereferencing file too early, allowing to pass NULL as the referring file here.
--- drivers/tty/n_tty.c | 33 ++++++++++++++++++++++----------- include/linux/tty.h | 2 +- 2 files changed, 23 insertions(+), 12 deletions(-)
drivers/tty/n_tty.c +22 -11
@@ -2309,22 +2309,15 @@ static ssize_t n_tty_read(struct tty_struct *tty, struct file *file,* lock themselves) */ -static ssize_t n_tty_write (struct tty_struct *tty,struct file *file, -const unsigned char *buf,size_t nr) +static ssize_t do_n_tty_write (struct tty_struct *tty,struct file *file, +const unsigned char *buf,size_t nr) {const unsigned char *b = buf;DEFINE_WAIT_FUNC (wait, woken_wake_function);int c;ssize_t retval =0 ; -/* Job control check -- must be done at start (POSIX.1 7.1.1.4). */ -if (L_TOSTOP (tty) && file->f_op->write != redirected_tty_write) { - retval =tty_check_change (tty); -if (retval) -return retval; - } - -down_read (&tty->termios_rwsem); +lockdep_assert_held_read (&tty->termios_rwsem);/* Write out any echoed characters that are still pending */ process_echoes (tty);
@@ -2392,10 +2385,28 @@ static ssize_t n_tty_write(struct tty_struct *tty, struct file *file,remove_wait_queue (&tty->write_wait, &wait);if (nr && tty->fasync)set_bit (TTY_DO_WRITE_WAKEUP , &tty->flags); -up_read (&tty->termios_rwsem);return (b - buf) ? b - buf : retval; } +static ssize_t n_tty_write (struct tty_struct *tty,struct file *file, +const unsigned char *buf,size_t nr) +{ +ssize_t retval =0 ; + +/* Job control check -- must be done at start (POSIX.1 7.1.1.4). */ +if (L_TOSTOP (tty) && file->f_op->write != redirected_tty_write) { + retval =tty_check_change (tty); +if (retval) +return retval; + } + +down_read (&tty->termios_rwsem); + retval =do_n_tty_write (tty, file, buf, nr); +up_read (&tty->termios_rwsem); + +return retval; +} +/** * n_tty_poll - poll method for N_TTY * @tty: terminal device
include/linux/tty.h +1 -1
@@ -390,7 +390,7 @@ static inline void tty_set_flow_change(struct tty_struct *tty, int val)static inline bool tty_io_nonblock (struct tty_struct *tty,struct file *file) { -return file->f_flags &O_NONBLOCK || +return (file && file->f_flags &O_NONBLOCK ) ||test_bit (TTY_LDISC_CHANGING , &tty->flags); }
Signature · 1 line
2.26.2
PATCH v3 7/7 n_tty: Provide an informational line on VSTATUS receipt
In reply to Arseny MaslennikovIf the three termios local flags isig, icanon, iexten are enabled and the local flag nokerninfo is disabled for a tty governed by the n_tty line discipline, then on receiving the keyboard status character n_tty will generate a status message and write it out to the tty before sending SIGINFO to the tty's foreground process group.
This kerninfo line contains information about the current system load as well as some properties of "the most interesting" process in the tty's current foreground process group, namely:
- its PID as seen inside its deepest PID namespace;
* the whole process group ought to be in a single PID namespace, so this is actually deterministic
- its saved command name truncated to 16 bytes (task_struct::comm);
* at the time of writing TASK_COMM_LEN == 16
- its state and some related bits, procps-style;
- for S and D: its symbolic wait channel, if available; or a short description for other process states instead;
- its user, system and real rusage time values;
- its resident set size (as well as the high watermark) in kilobytes.
The "most interesting" process is chosen as follows:
- runnables over everything
- uninterruptibles over everything else
- among 2 runnables pick the biggest utime + stime
- any unresolved ties are decided in favour of greatest PID.
While the kerninfo line is not very useful for debugging the kernel itself, since we have much more powerful debugging tools, it still gives the user behind the terminal some meaningful feedback to a VSTATUS that works even if no processes respond.
--- drivers/tty/Makefile | 3 +- drivers/tty/n_tty.c | 22 +++ drivers/tty/n_tty_status.c | 338 +++++++++++++++++++++++++++++++++++++ include/linux/sched.h | 7 + include/linux/tty.h | 3 + 5 files changed, 372 insertions(+), 1 deletion(-) create mode 100644 drivers/tty/n_tty_status.c
drivers/tty/Makefile +2 -1
@@ -1,7 +1,8 @@ # SPDX-License-Identifier: GPL-2.0 -obj-$(CONFIG_TTY) += tty_io.o n_tty.o tty_ioctl.o tty_ldisc.o \ +obj-$(CONFIG_TTY) += tty_io.o tty_ioctl.o tty_ldisc.o \ tty_buffer.o tty_port.o tty_mutex.o \ tty_ldsem.o tty_baudrate.o tty_jobctrl.o \ + n_tty.o n_tty_status.o \ n_null.o obj-$(CONFIG_LEGACY_PTYS) += pty.o obj-$(CONFIG_UNIX98_PTYS) += pty.o
drivers/tty/n_tty.c +22 -0
@@ -34,6 +34,7 @@#include <linux/signal.h> #include <linux/fcntl.h> #include <linux/sched.h> +#include <linux/kallsyms.h> #include <linux/interrupt.h> #include <linux/tty.h> #include <linux/timer.h>
@@ -79,6 +80,8 @@#define ECHO_BLOCK 256#define ECHO_DISCARD_WATERMARK N_TTY_BUF_SIZE - (ECHO_BLOCK + 32) +#define STATUS_LINE_LEN (2 * KSYM_NAME_LEN) +#define SIG_FLUSHING_MASK ( \ rt_sigmask(SIGINT) | rt_sigmask(SIGQUIT) | \ rt_sigmask(SIGTSTP) )
@@ -158,6 +161,8 @@ static inline unsigned char *echo_buf_addr(struct n_tty_data *ldata, size_t i)return &ldata->echo_buf[i & (N_TTY_BUF_SIZE -1 )]; } +static void n_tty_status_line (struct tty_struct *tty); +/* If we are not echoing the data, perhaps this is a secret so erase it */ static void zero_buffer (struct tty_struct *tty,u8 *buffer,int size) {
@@ -1300,6 +1305,8 @@ n_tty_receive_char_special(struct tty_struct *tty, unsigned char c)n_tty_receive_signal_char (tty,SIGTSTP , c);return 0 ; }else if (c ==STATUS_CHAR (tty)) { +if (!L_NOKERNINFO (tty)) +n_tty_status_line (tty);n_tty_receive_signal_char (tty,SIGINFO , c);return 0 ; }
@@ -2489,6 +2496,21 @@ static int n_tty_ioctl(struct tty_struct *tty, struct file *file, } } +static void n_tty_status_line (struct tty_struct *tty) +{ +struct n_tty_data *ldata = tty->disc_data; +char *msg, *buf; + msg = buf =kzalloc (STATUS_LINE_LEN ,GFP_KERNEL ); +tty_sprint_status_line (tty, buf +1 ,STATUS_LINE_LEN -1 ); +/* The only current caller of this takes output_lock for us. */ +if (ldata->column !=0 ) + *msg ='\n' ; +else + msg++; +do_n_tty_write (tty,NULL , msg,strlen (msg)); +kfree (buf); +} +static struct tty_ldisc_ops n_tty_ops = { .magic =TTY_LDISC_MAGIC , .name ="n_tty" ,
drivers/tty/n_tty_status.c new file +338 -0
338 lines not on this page · raw message
include/linux/sched.h +7 -0
@@ -1318,6 +1318,8 @@ static inline struct pid *task_pid(struct task_struct *task)* task_xid_nr() : global id, i.e. the id seen from the init namespace; * task_xid_vnr() : virtual id, i.e. the id seen from the pid namespace of * current. +* task_xid_vinr() : virtual inner id, i.e. the id seen from the namespace of +* the task itself; * task_xid_nr_ns() : id seen from the ns specified; * * see also pid_nr() etc in include/linux/pid.h
@@ -1339,6 +1341,11 @@ static inline pid_t task_pid_vnr(struct task_struct *tsk)return __task_pid_nr_ns (tsk,PIDTYPE_PID ,NULL ); } +static inline pid_t task_pid_vinr (struct task_struct *tsk) +{ +return __task_pid_nr_ns (tsk,PIDTYPE_PID ,ns_of_pid (task_pid (tsk))); +} +static inline pid_t task_tgid_nr (struct task_struct *tsk) {
include/linux/tty.h +3 -0
@@ -727,6 +727,9 @@ extern void __init n_tty_init(void);static inline void n_tty_init (void ) { }#endif +/* n_tty_status.c */ +extern void tty_sprint_status_line (struct tty_struct *tty,char *msg,size_t size); +/* tty_audit.c */ #ifdef CONFIG_AUDIT extern void tty_audit_add_data(struct tty_struct *tty, const void *data,
Signature · 1 line
2.26.2
Re: [PATCH v3 7/7] n_tty: Provide an informational line on VSTATUS receipt
General comments:
- care to CC scheduler and mm people?
- couldn't this share some code with fs/proc?
- I am not sure/convinced it is worth the hassle
Quoted 26 lines · Arseny Maslennikov
If the three termios local flags isig, icanon, iexten are enabled and the local flag nokerninfo is disabled for a tty governed by the
n_ttyline discipline, then on receiving the keyboard status charactern_ttywill generate a status message and write it out to the tty before sending SIGINFO to the tty's foreground process group.This kerninfo line contains information about the current system load as well as some properties of "the most interesting" process in the tty's current foreground process group, namely:
- its PID as seen inside its deepest PID namespace;
* the whole process group ought to be in a single PID namespace, so this is actually deterministic
- its saved command name truncated to 16 bytes (task_struct::comm);
* at the time of writingTASK_COMM_LEN== 16
- its state and some related bits, procps-style;
- for S and D: its symbolic wait channel, if available; or a short description for other process states instead;
- its user, system and real rusage time values;
- its resident set size (as well as the high watermark) in kilobytes.The "most interesting" process is chosen as follows:
- runnables over everything
- uninterruptibles over everything else
- among 2 runnables pick the biggest utime + stime
- any unresolved ties are decided in favour of greatest PID.
Quoted 4 lines
While the kerninfo line is not very useful for debugging the kernel itself, since we have much more powerful debugging tools, it still gives the user behind the terminal some meaningful feedback to a VSTATUS that works even if no processes respond.
Care to append an example output to the commit message?
Quoted 1 line
Signed-off-by Arseny Maslennikov <[email protected]>
...
drivers/tty/n_tty.c line 83
Quoted 8 lines
index
f72a3fd4b..905cdd985100644drivers/tty/n_tty.c +1
-0@@ -79,6 +80,8 @@#define ECHO_BLOCK 256#define ECHO_DISCARD_WATERMARK N_TTY_BUF_SIZE - (ECHO_BLOCK + 32) +#define STATUS_LINE_LEN (2 * KSYM_NAME_LEN)
It's too magic constant.
drivers/tty/n_tty.c static int n_tty_ioctl(struct tty_struct *tty, struct file *file, · line 2505
Quoted 11 lines
drivers/tty/n_tty.c +7
-0@@ -2489,6 +2496,21 @@ static int n_tty_ioctl(struct tty_struct *tty, struct file *file, } } +static void n_tty_status_line (struct tty_struct *tty) +{ +struct n_tty_data *ldata = tty->disc_data; +char *msg, *buf; + msg = buf =kzalloc (STATUS_LINE_LEN ,GFP_KERNEL ); +tty_sprint_status_line (tty, buf +1 ,STATUS_LINE_LEN -1 ); +/* The only current caller of this takes output_lock for us. */
So add a lockdep assertion?
Quoted 42 lines
drivers/tty/n_tty.c +8
-0+if (ldata->column !=0 ) + *msg ='\n' ; +else + msg++; +do_n_tty_write (tty,NULL , msg,strlen (msg)); +kfree (buf); +} +static struct tty_ldisc_ops n_tty_ops = { .magic =TTY_LDISC_MAGIC , .name ="n_tty" ,drivers/tty/n_tty_status.c new file +25
-0@@ -0,0 +1,338 @@ +// SPDX-License-Identifier: GPL-2.0+ + +#include <linux/kallsyms.h> +#include <linux/pid.h> +#include <linux/sched.h> +#include <linux/sched/signal.h> +#include <linux/sched/loadavg.h> +#include <linux/sched/cputime.h> +#include <linux/sched/mm.h> +#include <linux/slab.h> +#include <linux/tty.h> + +#define BCOMPARE (lbool, rbool) ((lbool) << 1 | (rbool)) +#define BCOMPARE_NONE 0 +#define BCOMPARE_RIGHT 1 +#define BCOMPARE_LEFT 2 +#define BCOMPARE_BOTH (BCOMPARE_LEFT | BCOMPARE_RIGHT) + +/* +* Select the most interesting task of two. +* +* The implemented approach is simple for now: +* - pick runnable +* - if no runnables, pick uninterruptible +* - if tie between runnables, pick highest utime + stime
drivers/tty/n_tty_status.c line 35
Quoted 10 lines
drivers/tty/n_tty_status.c new file +10
-0+* - if a tie is not broken by the above, pick highest pid nr. +* +* For reference, here's the one used in FreeBSD: +* - pick runnables over anything +* - if both runnables, pick highest cpu utilization +* - if no runnables, pick shortest sleep time (the scheduler +* actually takes care to maintain this statistic) +* - other ties are decided in favour of youngest process. +*/ +static struct task_struct *__better_proc_R(struct task_struct *a,
Why so weird name __better_proc_R?
Quoted 18 lines
drivers/tty/n_tty_status.c new file +18
-0+ struct task_struct *b) +{ +unsigned long flags; +u64 atime, btime, tgutime, tgstime; +struct task_struct *ret; + +if (!lock_task_sighand (a, &flags)) + goto out_a_unlocked; +thread_group_cputime_adjusted (a, &tgutime, &tgstime); + atime = tgutime + tgstime; +unlock_task_sighand (a, &flags); + +if (!lock_task_sighand (b, &flags)) + goto out_b_unlocked; +thread_group_cputime_adjusted (b, &tgutime, &tgstime); + btime = tgutime + tgstime; +unlock_task_sighand (b, &flags); +
drivers/tty/n_tty_status.c line 63
Quoted 10 lines
drivers/tty/n_tty_status.c new file +10
-0+ ret = atime > btime ? a : b; + +return ret; + +out_b_unlocked: +out_a_unlocked: +return task_pid_vinr (a) >task_pid_vinr (b) ? a : b; +} + +static struct task_struct *__better_proc(struct task_struct *a,
Again, why "__" in the name?
Quoted 6 lines
drivers/tty/n_tty_status.c new file +6
-0+ struct task_struct *b) +{ +if (!pid_alive (a)) +return b; +if (!pid_alive (b)) +return a;
drivers/tty/n_tty_status.c line 79
Quoted 10 lines
drivers/tty/n_tty_status.c new file +10
-0+ +switch (BCOMPARE (a->state ==TASK_RUNNING , + b->state ==TASK_RUNNING )) { +case BCOMPARE_LEFT : +return a; +case BCOMPARE_RIGHT : +return b; +case BCOMPARE_BOTH : +return __better_proc_R (a, b); + }
Doesn't this look saner and better:
if (a->state == TASK_RUNNING && b->state == TASK_RUNNING)
return __better_proc_R(a, b);
if (a->state == TASK_RUNNING)
return a;
if (b->state == TASK_RUNNING)
return b;
?
And one doesn't need to decrypt the defines.
drivers/tty/n_tty_status.c line 88
Quoted 8 lines
drivers/tty/n_tty_status.c new file +8
-0+switch (BCOMPARE (a->state ==TASK_UNINTERRUPTIBLE , + b->state ==TASK_UNINTERRUPTIBLE )) { +case BCOMPARE_LEFT : +return a; +case BCOMPARE_RIGHT : +return b; +case BCOMPARE_BOTH : +break ;
dtto.
drivers/tty/n_tty_status.c line 98
Quoted 10 lines
drivers/tty/n_tty_status.c new file +10
-0+ } + +/* TODO: Perhaps we should check something else... */ +return task_pid_vinr (a) >task_pid_vinr (b) ? a : b; +} + +/* Weed out NULLs. */ +static inline struct task_struct *better_proc (struct task_struct *a, +struct task_struct *b) { +return a ? (b ?__better_proc (a, b) : a) : b;
This urgently calls for ifs and not ternany operators.
Actually you don't need this helper at all. Check a and b in the same if as the respective !pid_alive above.
Quoted 254 lines
drivers/tty/n_tty_status.c new file +240
-0+} + +static int scnprint_load (char *msgp,size_t size) +{ +unsigned long la[3 ]; + +get_avenrun (la,FIXED_1 /200 ,0 ); +return scnprintf (msgp, size,"load: %lu.%02lu; " , +LOAD_INT (la[0 ]),LOAD_FRAC (la[0 ])); +} + +static int scnprint_task (char *msgp,size_t size,struct task_struct *task) +{ +char commname[TASK_COMM_LEN ]; + +get_task_comm (commname, task); +return scnprintf (msgp, size,"%d/%s:" ,task_pid_vinr (task), commname); +} + +static int scnprint_rusage (char *msgp,ssize_t size, +struct task_struct *task,struct mm_struct *mm) +{ +struct rusage ru; +struct timespec64 utime, stime; +struct timespec64 rtime; +u64 now; +int ret =0 ; +int psz =0 ; + +getrusage (task,RUSAGE_BOTH , &ru); + now =ktime_get_ns (); + + utime.tv_sec = ru.ru_utime.tv_sec; + utime.tv_nsec = ru.ru_utime.tv_usec *NSEC_PER_USEC ; + stime.tv_sec = ru.ru_stime.tv_sec; + stime.tv_nsec = ru.ru_stime.tv_usec *NSEC_PER_USEC ; + + rtime =ns_to_timespec64 (now - task->start_time); + + psz =scnprintf (msgp, size, +"%llu.%03lur %llu.%03luu %llu.%03lus" , + rtime.tv_sec, rtime.tv_nsec /NSEC_PER_MSEC , + utime.tv_sec, utime.tv_nsec /NSEC_PER_MSEC , + stime.tv_sec, stime.tv_nsec /NSEC_PER_MSEC ); + ret += psz; + msgp += psz; + size -= psz; + +if (mm) { + psz =scnprintf (msgp, size, +" %luk/%luk" , +get_mm_rss (mm) *PAGE_SIZE /1024 , +get_mm_hiwater_rss (mm) *PAGE_SIZE /1024 ); + ret += psz; + } +return ret; +} + +static int scnprint_state (char *msgp,ssize_t size, +struct task_struct *task,struct mm_struct *mm) +{ +char stat[8 ] = {0 }; +const char *state_descr ="" ; +struct task_struct *parent =NULL ; +struct task_struct *real_parent =NULL ; +unsigned long wchan =0 ; +int psz =0 ; +char symname[KSYM_NAME_LEN ]; + + stat[psz++] =task_state_to_char (task); +if (task_nice (task) <0 ) + stat[psz++] ='<' ; +else if (task_nice (task) >0 ) + stat[psz++] ='N' ; +if (mm && mm->locked_vm) + stat[psz++] ='L' ; +if (get_nr_threads (task) >1 ) + stat[psz++] ='l' ; + +switch (stat[0 ]) { +case 'R' : +if (task_curr (task)) + stat[psz++] ='!' ; +break ; +case 'S' : +case 'D' : + wchan =get_wchan (task); +if (!wchan) +break ; +if (!lookup_symbol_name (wchan, symname)) + state_descr = symname; +else +/* get_wchan() returned something +* that looks like no kernel symbol. +*/ + state_descr ="*unknown*" ; +break ; +case 'T' : + state_descr ="stopped" ; +break ; +case 't' : + state_descr ="traced" ; +break ; +case 'Z' : +rcu_read_lock (); + real_parent =rcu_dereference (task->real_parent); + parent =rcu_dereference (task->parent); + psz =sprintf (symname,"zombie; ppid=%d" , +task_tgid_nr_ns (real_parent, +ns_of_pid (task_pid (task)))); +if (parent != real_parent) +sprintf (symname + psz," reaper=%d" , +task_tgid_nr_ns (parent, +ns_of_pid (task_pid (task)))); +rcu_read_unlock (); + state_descr = symname; +break ; +case 'I' : +/* Can this even happen? */ + state_descr ="idle" ; +break ; +default : + state_descr ="unknown" ; + } + + psz =scnprintf (msgp, size,"%s" , stat); + msgp += psz; + size -= psz; +if (*state_descr) + psz +=scnprintf (msgp, size, wchan ?" [%s]" :" (%s)" , state_descr); + +return psz; +} + +/** +* tty_sprint_status_line - produce kerninfo line +* @tty: terminal device +* @msg: preallocated memory buffer +* @length: maximum line length +* +* Reports state of foreground process group in a null-terminated string +* located at @msg, @length bytes long. If @length is insufficient, +* the line gets truncated. +*/ +void tty_sprint_status_line (struct tty_struct *t,char *msg,size_t length) +{ +struct task_struct *tsk =NULL , *mit =NULL ; +struct mm_struct *mm; +struct pid *pgrp =NULL ; +char *msgp = msg; +int psz =0 ; + +if (!length) +return ; + length--;/* Make room for trailing '\n' */ + + psz =scnprint_load (msgp, length); +if (psz >0 ) { + msgp += psz; + length -= psz; + } +if (!length) + goto finalize_message; + +/* Not sure if session pid is protected by ctrl_lock +* or tasklist_lock... +*/ + pgrp = t->session; +if (pgrp ==NULL ) { + psz =scnprintf (msgp, length,"not a controlling tty" ); +if (psz >0 ) + msgp += psz; + goto finalize_message; + } + pgrp =tty_get_pgrp (t); +if (pgrp ==NULL ) { + psz =scnprintf (msgp, length,"no foreground process group" ); +if (psz >0 ) + msgp += psz; + goto finalize_message; + } + +read_lock (&tasklist_lock); + do_each_pid_task(pgrp, PIDTYPE_PGID, tsk) + { +/* Select the most interesting task. */ +if (tsk ==better_proc (mit, tsk)) + mit = tsk; + }while_each_pid_task (pgrp,PIDTYPE_PGID , tsk); +read_unlock (&tasklist_lock); + +if (!pid_alive (mit)) + goto finalize_message; + +/* Gather intel on most interesting task. */ +/* Can the mm of a foreground process turn out to be NULL? +* Definitely; for example, if it is a zombie. +*/ + mm =get_task_mm (mit); + + psz =scnprint_task (msgp, length, mit); +if (psz >0 ) { + msgp += psz; + length -= psz; + } +if (!length) + goto finalize_message; + *msgp++ =' ' ; + length--; + + psz =scnprint_state (msgp, length, mit, mm); +if (psz >0 ) { + msgp += psz; + length -= psz; + } +if (!length) + goto finalize_message; + *msgp++ =' ' ; + length--; + + psz =scnprint_rusage (msgp, length, mit, mm); +if (psz >0 ) { + msgp += psz; + length -= psz; + } +if (!length) + goto finalize_message; + *msgp++ =' ' ; + length--; + +if (!mm) + goto finalize_message; + +mmput (mm); + +finalize_message: + *msgp++ ='\n' ; +if (pgrp) +put_pid (pgrp); +}include/linux/sched.h +2
-0@@ -1318,6 +1318,8 @@ static inline struct pid *task_pid(struct task_struct *task)* task_xid_nr() : global id, i.e. the id seen from the init namespace; * task_xid_vnr() : virtual id, i.e. the id seen from the pid namespace of * current. +* task_xid_vinr() : virtual inner id, i.e. the id seen from the namespace of +* the task itself; * task_xid_nr_ns() : id seen from the ns specified; * * see also pid_nr() etc in include/linux/pid.h
include/linux/sched.h static inline pid_t task_pid_vnr(struct task_struct *tsk) · line 1348
Quoted 8 lines
include/linux/sched.h +5
-0@@ -1339,6 +1341,11 @@ static inline pid_t task_pid_vnr(struct task_struct *tsk)return __task_pid_nr_ns (tsk,PIDTYPE_PID ,NULL ); } +static inline pid_t task_pid_vinr (struct task_struct *tsk) +{ +return __task_pid_nr_ns (tsk,PIDTYPE_PID ,ns_of_pid (task_pid (tsk))); +} +
This smells like it should be done in a separate patch.
include/linux/tty.h extern void __init n_tty_init(void); · line 731
Quoted 12 lines
include/linux/sched.h +0
-0static inline pid_t task_tgid_nr (struct task_struct *tsk) {include/linux/tty.h +2
-0@@ -727,6 +727,9 @@ extern void __init n_tty_init(void);static inline void n_tty_init (void ) { }#endif +/* n_tty_status.c */ +extern void tty_sprint_status_line (struct tty_struct *tty,char *msg,size_t size);
No need for extern.
thanks,
Signature · 2 lines
js suse labs
Re: [PATCH v3 7/7] n_tty: Provide an informational line on VSTATUS receipt
Quoted 2 lines · Jiri Slaby
General comments:
- care to CC scheduler and mm people?
I did in v1, but dropped those CCs in v2 and on, being afraid to over-cc.
I'm sorry if I did the wrong thing.
Quoted 1 line
- couldn't this share some code with fs/proc?
We can't use the proc handlers directly, since they are often written to output to the proc seqfile interface, and we can't use that.
This shares code with proc at a lower level by using the same primitives to obtain info from.
Quoted 29 lines
- I am not sure/convinced it is worth the hassle
Quoted 26 lines · Arseny Maslennikov
If the three termios local flags isig, icanon, iexten are enabled and the local flag nokerninfo is disabled for a tty governed by the
n_ttyline discipline, then on receiving the keyboard status charactern_ttywill generate a status message and write it out to the tty before sending SIGINFO to the tty's foreground process group.This kerninfo line contains information about the current system load as well as some properties of "the most interesting" process in the tty's current foreground process group, namely:
- its PID as seen inside its deepest PID namespace;
* the whole process group ought to be in a single PID namespace, so this is actually deterministic
- its saved command name truncated to 16 bytes (task_struct::comm);
* at the time of writingTASK_COMM_LEN== 16
- its state and some related bits, procps-style;
- for S and D: its symbolic wait channel, if available; or a short description for other process states instead;
- its user, system and real rusage time values;
- its resident set size (as well as the high watermark) in kilobytes.The "most interesting" process is chosen as follows:
- runnables over everything
- uninterruptibles over everything else
- among 2 runnables pick the biggest utime + stime
- any unresolved ties are decided in favour of greatest PID.
Quoted 6 lines
Quoted 4 lines
While the kerninfo line is not very useful for debugging the kernel itself, since we have much more powerful debugging tools, it still gives the user behind the terminal some meaningful feedback to a VSTATUS that works even if no processes respond.
Care to append an example output to the commit message?
$ cat load: 0.32; 1108279/cat: S 0.524r 0.000u 0.000s 700k/700k load: 0.32; 1108279/cat: S 1.873r 0.000u 0.000s 700k/700k $ sha256sum /dev/zero load: 0.32; 1108298/sha256sum: R! 1.302r 1.274u 0.015s 700k/700k load: 0.32; 1108298/sha256sum: R! 2.445r 2.378u 0.055s 700k/700k load: 0.32; 1108298/sha256sum: R! 3.309r 3.226u 0.071s 700k/700k load: 0.42; 1108298/sha256sum: R! 14.547r 14.340u 0.195s 700k/700k ^C
Here I pressed ^T 2 times before supplying EOF to cat and 4 times before interrupting sha256sum.
An example with a patched sleep:
% sleep 8
load: 0.04; 66573/sleep: S [hrtimer_nanosleep] 0.92r 0.00u 0.00s 1544k/1544k
sleep: about 7 second(s) left out of the original 8
load: 0.04; 66573/sleep: S [hrtimer_nanosleep] 2.61r 0.00u 0.00s 1552k/1552k
sleep: about 5 second(s) left out of the original 8
load: 0.04; 66573/sleep: S [hrtimer_nanosleep] 3.44r 0.00u 0.00s 1552k/1552k
sleep: about 4 second(s) left out of the original 8
load: 0.04; 66573/sleep: S [hrtimer_nanosleep] 3.65r 0.00u 0.00s 1552k/1552k
sleep: about 4 second(s) left out of the original 8
load: 0.04; 66573/sleep: S [hrtimer_nanosleep] 4.30r 0.00u 0.00s 1552k/1552k
sleep: about 3 second(s) left out of the original 8
load: 0.04; 66573/sleep: S [hrtimer_nanosleep] 4.72r 0.00u 0.00s 1552k/1552k
sleep: about 3 second(s) left out of the original 8
load: 0.04; 66573/sleep: S [hrtimer_nanosleep] 5.44r 0.00u 0.00s 1552k/1552k
sleep: about 2 second(s) left out of the original 8
load: 0.04; 66573/sleep: S [hrtimer_nanosleep] 5.57r 0.00u 0.00s 1552k/1552k
sleep: about 2 second(s) left out of the original 8
load: 0.04; 66573/sleep: S [hrtimer_nanosleep] 6.02r 0.00u 0.00s 1552k/1552k
sleep: about 1 second(s) left out of the original 8
load: 0.04; 66573/sleep: S [hrtimer_nanosleep] 6.15r 0.00u 0.00s 1552k/1552k
sleep: about 1 second(s) left out of the original 8
load: 0.04; 66573/sleep: S [hrtimer_nanosleep] 6.23r 0.00u 0.00s 1552k/1552k
sleep: about 1 second(s) left out of the original 8
load: 0.04; 66573/sleep: S [hrtimer_nanosleep] 7.20r 0.00u 0.00s 1552k/1552k
sleep: about 0 second(s) left out of the original 8
Another short example:
$ cat <<EOX zombospawner.c #include <unistd.h> #include <stdio.h> #include <error.h> #include <errno.h>
int main(int argc, char* argv[]) {
char* prog;
int e;
if (argc <= 1) return -1;
fprintf(stderr, "[my pid is %d]\n", getpid());
e = fork();
if (e)
for (;;) sleep(4);
fprintf(stderr, "[and my pid is %d. Will i become a Z?]\n", getpid());
e = execv(argv[1], argv+1);
error(e, errno, "oops");
}
EOX
$ gcc zombospawner.c -o zombospawner
$ ./zombospawner xx
[my pid is 1109004]
[and my pid is 1109005. Will i become a Z?]
./zombospawner: oops: No such file or directory
load: 1.35; 1109005/zombospawner: Z (zombie; ppid=1109004) 3.443r 0.000u 0.000s
load: 1.35; 1109005/zombospawner: Z (zombie; ppid=1109004) 3.743r 0.000u 0.000s
load: 1.32; 1109005/zombospawner: Z (zombie; ppid=1109004) 5.175r 0.000u 0.000s
^CBut you're right, this should be in the commit message, since it makes little sense to me to devote to this its own Documentation/ page.
Quoted 10 lines
Quoted 2 lines
Signed-off-by Arseny Maslennikov <[email protected]>...
Quoted 7 lines
index
f72a3fd4b..905cdd985100644drivers/tty/n_tty.c +0
-0@@ -79,6 +80,8 @@#define ECHO_BLOCK 256#define ECHO_DISCARD_WATERMARK N_TTY_BUF_SIZE - (ECHO_BLOCK + 32)
Quoted 3 lines
drivers/tty/n_tty.c line 83
Quoted 1 line
drivers/tty/n_tty.c +1
-0+#define STATUS_LINE_LEN (2 * KSYM_NAME_LEN)It's too magic constant.
The line to be constructed can optionally contain a kernel symbol name, which is not longer than KSYM_NAME_LEN, and the effective size of all the other characters together is also less than KSYM_NAME_LEN — that includes:
* "load: ", which is 6 bytes;
* the load average and the ';', which takes from 5 bytes if LA < 10 to 13 bytes for a billion CPU cores (huh!) and scales logarithmically;
* the decimal PID nr, which is not longer than 9 bytes;
* the comm, which is not longer than TASK_COMM_LEN;
* the ": S ", which is 4 bytes;
* the rusage values, each of those 3 is at least 7 bytes and scales at most logarithmically with seconds elapsed.
6 + 13 + 9 + 16 (comm) + 4 + 51 is 99 < 128, and this assumes close-to worst case for 10-char-long decimals for all the times and the PID.
Instead of a kernel symbol name, the middle "wchan" field can show things like "zombie; ppid=123456 reaper=98765", which generally take much less space than KSYM_NAME_LEN (most kernel symbol names that processes sleep at also take less space).
So realistically we're not hitting the specified generous limit, which is 256 bytes. Another reason for that number is, since the memory chunk is short-lived, it makes little sense to kmalloc less.
If some time later we miraculously do hit the 256 char limit, the line will just get truncated.
Quoted 5 lines
Quoted 5 lines
drivers/tty/n_tty.c +0
-0@@ -2489,6 +2496,21 @@ static int n_tty_ioctl(struct tty_struct *tty, struct file *file, } }
Quoted 9 lines
drivers/tty/n_tty.c static int n_tty_ioctl(struct tty_struct *tty, struct file *file, · line 2505
Quoted 7 lines
drivers/tty/n_tty.c +7
-0+static void n_tty_status_line (struct tty_struct *tty) +{ +struct n_tty_data *ldata = tty->disc_data; +char *msg, *buf; + msg = buf =kzalloc (STATUS_LINE_LEN ,GFP_KERNEL ); +tty_sprint_status_line (tty, buf +1 ,STATUS_LINE_LEN -1 ); +/* The only current caller of this takes output_lock for us. */ So add a lockdep assertion?
Yes, this is a good idea. I added an assertion in a similar situation in 6/7, but hesitated here.
Quoted 47 lines
Quoted 47 lines
drivers/tty/n_tty.c +8
-0+if (ldata->column !=0 ) + *msg ='\n' ; +else + msg++; +do_n_tty_write (tty,NULL , msg,strlen (msg)); +kfree (buf); +} +static struct tty_ldisc_ops n_tty_ops = { .magic =TTY_LDISC_MAGIC , .name ="n_tty" ,drivers/tty/n_tty_status.c new file +29
-0@@ -0,0 +1,338 @@ +// SPDX-License-Identifier: GPL-2.0+ + +#include <linux/kallsyms.h> +#include <linux/pid.h> +#include <linux/sched.h> +#include <linux/sched/signal.h> +#include <linux/sched/loadavg.h> +#include <linux/sched/cputime.h> +#include <linux/sched/mm.h> +#include <linux/slab.h> +#include <linux/tty.h> + +#define BCOMPARE (lbool, rbool) ((lbool) << 1 | (rbool)) +#define BCOMPARE_NONE 0 +#define BCOMPARE_RIGHT 1 +#define BCOMPARE_LEFT 2 +#define BCOMPARE_BOTH (BCOMPARE_LEFT | BCOMPARE_RIGHT) + +/* +* Select the most interesting task of two. +* +* The implemented approach is simple for now: +* - pick runnable +* - if no runnables, pick uninterruptible +* - if tie between runnables, pick highest utime + stime +* - if a tie is not broken by the above, pick highest pid nr. +* +* For reference, here's the one used in FreeBSD: +* - pick runnables over anything
Quoted 9 lines
drivers/tty/n_tty_status.c line 35
Quoted 6 lines
drivers/tty/n_tty_status.c new file +6
-0+* - if both runnables, pick highest cpu utilization +* - if no runnables, pick shortest sleep time (the scheduler +* actually takes care to maintain this statistic) +* - other ties are decided in favour of youngest process. +*/ +static struct task_struct *__better_proc_R(struct task_struct *a,Why so weird name
__better_proc_R?
Please see below.
Quoted 22 lines
Quoted 22 lines
drivers/tty/n_tty_status.c new file +22
-0+ struct task_struct *b) +{ +unsigned long flags; +u64 atime, btime, tgutime, tgstime; +struct task_struct *ret; + +if (!lock_task_sighand (a, &flags)) + goto out_a_unlocked; +thread_group_cputime_adjusted (a, &tgutime, &tgstime); + atime = tgutime + tgstime; +unlock_task_sighand (a, &flags); + +if (!lock_task_sighand (b, &flags)) + goto out_b_unlocked; +thread_group_cputime_adjusted (b, &tgutime, &tgstime); + btime = tgutime + tgstime; +unlock_task_sighand (b, &flags); + + ret = atime > btime ? a : b; + +return ret; +
Quoted 8 lines
drivers/tty/n_tty_status.c line 63
Quoted 6 lines
drivers/tty/n_tty_status.c new file +6
-0+out_b_unlocked: +out_a_unlocked: +return task_pid_vinr (a) >task_pid_vinr (b) ? a : b; +} + +static struct task_struct *__better_proc(struct task_struct *a,Again, why "__" in the name?
Does the prefix __ have a very special meaning so it cannot be used here, in a static symbol?
I'm sorry if it does, Documentation/process/coding-style.rst does not warn against it.
I used it to generally mean "wrapped implementation detail".
Quoted 27 lines
Quoted 7 lines
drivers/tty/n_tty_status.c new file +6
-0+ struct task_struct *b) +{ +if (!pid_alive (a)) +return b; +if (!pid_alive (b)) +return a;drivers/tty/n_tty_status.c line 79
Quoted 10 lines
drivers/tty/n_tty_status.c new file +10
-0+ +switch (BCOMPARE (a->state ==TASK_RUNNING , + b->state ==TASK_RUNNING )) { +case BCOMPARE_LEFT : +return a; +case BCOMPARE_RIGHT : +return b; +case BCOMPARE_BOTH : +return __better_proc_R (a, b); + }Doesn't this look saner and better:
if (
a->state==TASK_RUNNING&&b->state==TASK_RUNNING)
return__better_proc_R(a, b);
if (a->state==TASK_RUNNING)
return a;
if (b->state==TASK_RUNNING)
return b;
Quoted 3 lines
?
And one doesn't need to decrypt the defines.
I felt like the version with ifs would read worse. No problem changing this.
Quoted 285 lines
drivers/tty/n_tty_status.c line 88
Quoted 9 lines
drivers/tty/n_tty_status.c new file +8
-0+switch (BCOMPARE (a->state ==TASK_UNINTERRUPTIBLE , + b->state ==TASK_UNINTERRUPTIBLE )) { +case BCOMPARE_LEFT : +return a; +case BCOMPARE_RIGHT : +return b; +case BCOMPARE_BOTH : +break ;dtto.
drivers/tty/n_tty_status.c line 98
Quoted 10 lines
drivers/tty/n_tty_status.c new file +10
-0+ } + +/* TODO: Perhaps we should check something else... */ +return task_pid_vinr (a) >task_pid_vinr (b) ? a : b; +} + +/* Weed out NULLs. */ +static inline struct task_struct *better_proc (struct task_struct *a, +struct task_struct *b) { +return a ? (b ?__better_proc (a, b) : a) : b;This urgently calls for ifs and not ternany operators.
Actually you don't need this helper at all. Check a and b in the same if as the respective !pid_alive above.
Quoted 257 lines
drivers/tty/n_tty_status.c new file +240
-0+} + +static int scnprint_load (char *msgp,size_t size) +{ +unsigned long la[3 ]; + +get_avenrun (la,FIXED_1 /200 ,0 ); +return scnprintf (msgp, size,"load: %lu.%02lu; " , +LOAD_INT (la[0 ]),LOAD_FRAC (la[0 ])); +} + +static int scnprint_task (char *msgp,size_t size,struct task_struct *task) +{ +char commname[TASK_COMM_LEN ]; + +get_task_comm (commname, task); +return scnprintf (msgp, size,"%d/%s:" ,task_pid_vinr (task), commname); +} + +static int scnprint_rusage (char *msgp,ssize_t size, +struct task_struct *task,struct mm_struct *mm) +{ +struct rusage ru; +struct timespec64 utime, stime; +struct timespec64 rtime; +u64 now; +int ret =0 ; +int psz =0 ; + +getrusage (task,RUSAGE_BOTH , &ru); + now =ktime_get_ns (); + + utime.tv_sec = ru.ru_utime.tv_sec; + utime.tv_nsec = ru.ru_utime.tv_usec *NSEC_PER_USEC ; + stime.tv_sec = ru.ru_stime.tv_sec; + stime.tv_nsec = ru.ru_stime.tv_usec *NSEC_PER_USEC ; + + rtime =ns_to_timespec64 (now - task->start_time); + + psz =scnprintf (msgp, size, +"%llu.%03lur %llu.%03luu %llu.%03lus" , + rtime.tv_sec, rtime.tv_nsec /NSEC_PER_MSEC , + utime.tv_sec, utime.tv_nsec /NSEC_PER_MSEC , + stime.tv_sec, stime.tv_nsec /NSEC_PER_MSEC ); + ret += psz; + msgp += psz; + size -= psz; + +if (mm) { + psz =scnprintf (msgp, size, +" %luk/%luk" , +get_mm_rss (mm) *PAGE_SIZE /1024 , +get_mm_hiwater_rss (mm) *PAGE_SIZE /1024 ); + ret += psz; + } +return ret; +} + +static int scnprint_state (char *msgp,ssize_t size, +struct task_struct *task,struct mm_struct *mm) +{ +char stat[8 ] = {0 }; +const char *state_descr ="" ; +struct task_struct *parent =NULL ; +struct task_struct *real_parent =NULL ; +unsigned long wchan =0 ; +int psz =0 ; +char symname[KSYM_NAME_LEN ]; + + stat[psz++] =task_state_to_char (task); +if (task_nice (task) <0 ) + stat[psz++] ='<' ; +else if (task_nice (task) >0 ) + stat[psz++] ='N' ; +if (mm && mm->locked_vm) + stat[psz++] ='L' ; +if (get_nr_threads (task) >1 ) + stat[psz++] ='l' ; + +switch (stat[0 ]) { +case 'R' : +if (task_curr (task)) + stat[psz++] ='!' ; +break ; +case 'S' : +case 'D' : + wchan =get_wchan (task); +if (!wchan) +break ; +if (!lookup_symbol_name (wchan, symname)) + state_descr = symname; +else +/* get_wchan() returned something +* that looks like no kernel symbol. +*/ + state_descr ="*unknown*" ; +break ; +case 'T' : + state_descr ="stopped" ; +break ; +case 't' : + state_descr ="traced" ; +break ; +case 'Z' : +rcu_read_lock (); + real_parent =rcu_dereference (task->real_parent); + parent =rcu_dereference (task->parent); + psz =sprintf (symname,"zombie; ppid=%d" , +task_tgid_nr_ns (real_parent, +ns_of_pid (task_pid (task)))); +if (parent != real_parent) +sprintf (symname + psz," reaper=%d" , +task_tgid_nr_ns (parent, +ns_of_pid (task_pid (task)))); +rcu_read_unlock (); + state_descr = symname; +break ; +case 'I' : +/* Can this even happen? */ + state_descr ="idle" ; +break ; +default : + state_descr ="unknown" ; + } + + psz =scnprintf (msgp, size,"%s" , stat); + msgp += psz; + size -= psz; +if (*state_descr) + psz +=scnprintf (msgp, size, wchan ?" [%s]" :" (%s)" , state_descr); + +return psz; +} + +/** +* tty_sprint_status_line - produce kerninfo line +* @tty: terminal device +* @msg: preallocated memory buffer +* @length: maximum line length +* +* Reports state of foreground process group in a null-terminated string +* located at @msg, @length bytes long. If @length is insufficient, +* the line gets truncated. +*/ +void tty_sprint_status_line (struct tty_struct *t,char *msg,size_t length) +{ +struct task_struct *tsk =NULL , *mit =NULL ; +struct mm_struct *mm; +struct pid *pgrp =NULL ; +char *msgp = msg; +int psz =0 ; + +if (!length) +return ; + length--;/* Make room for trailing '\n' */ + + psz =scnprint_load (msgp, length); +if (psz >0 ) { + msgp += psz; + length -= psz; + } +if (!length) + goto finalize_message; + +/* Not sure if session pid is protected by ctrl_lock +* or tasklist_lock... +*/ + pgrp = t->session; +if (pgrp ==NULL ) { + psz =scnprintf (msgp, length,"not a controlling tty" ); +if (psz >0 ) + msgp += psz; + goto finalize_message; + } + pgrp =tty_get_pgrp (t); +if (pgrp ==NULL ) { + psz =scnprintf (msgp, length,"no foreground process group" ); +if (psz >0 ) + msgp += psz; + goto finalize_message; + } + +read_lock (&tasklist_lock); + do_each_pid_task(pgrp, PIDTYPE_PGID, tsk) + { +/* Select the most interesting task. */ +if (tsk ==better_proc (mit, tsk)) + mit = tsk; + }while_each_pid_task (pgrp,PIDTYPE_PGID , tsk); +read_unlock (&tasklist_lock); + +if (!pid_alive (mit)) + goto finalize_message; + +/* Gather intel on most interesting task. */ +/* Can the mm of a foreground process turn out to be NULL? +* Definitely; for example, if it is a zombie. +*/ + mm =get_task_mm (mit); + + psz =scnprint_task (msgp, length, mit); +if (psz >0 ) { + msgp += psz; + length -= psz; + } +if (!length) + goto finalize_message; + *msgp++ =' ' ; + length--; + + psz =scnprint_state (msgp, length, mit, mm); +if (psz >0 ) { + msgp += psz; + length -= psz; + } +if (!length) + goto finalize_message; + *msgp++ =' ' ; + length--; + + psz =scnprint_rusage (msgp, length, mit, mm); +if (psz >0 ) { + msgp += psz; + length -= psz; + } +if (!length) + goto finalize_message; + *msgp++ =' ' ; + length--; + +if (!mm) + goto finalize_message; + +mmput (mm); + +finalize_message: + *msgp++ ='\n' ; +if (pgrp) +put_pid (pgrp); +}include/linux/sched.h +2
-0@@ -1318,6 +1318,8 @@ static inline struct pid *task_pid(struct task_struct *task)* task_xid_nr() : global id, i.e. the id seen from the init namespace; * task_xid_vnr() : virtual id, i.e. the id seen from the pid namespace of * current. +* task_xid_vinr() : virtual inner id, i.e. the id seen from the namespace of +* the task itself; * task_xid_nr_ns() : id seen from the ns specified; * * see also pid_nr() etc in include/linux/pid.h @@ -1339,6 +1341,11 @@ static inline pid_t task_pid_vnr(struct task_struct *tsk)return __task_pid_nr_ns (tsk,PIDTYPE_PID ,NULL ); }
Quoted 7 lines
include/linux/sched.h static inline pid_t task_pid_vnr(struct task_struct *tsk) · line 1348
Quoted 5 lines
include/linux/sched.h +5
-0+static inline pid_t task_pid_vinr (struct task_struct *tsk) +{ +return __task_pid_nr_ns (tsk,PIDTYPE_PID ,ns_of_pid (task_pid (tsk))); +} +This smells like it should be done in a separate patch.
OK, no problem.
This also solves the over-cc consideration mentioned above.
Quoted 11 lines
Quoted 11 lines
include/linux/sched.h +0
-0static inline pid_t task_tgid_nr (struct task_struct *tsk) {include/linux/tty.h +0
-0@@ -727,6 +727,9 @@ extern void __init n_tty_init(void);static inline void n_tty_init (void ) { }#endif
Quoted 4 lines
include/linux/tty.h extern void __init n_tty_init(void); · line 731
Quoted 2 lines
include/linux/tty.h +2
-0+/* n_tty_status.c */ +extern void tty_sprint_status_line (struct tty_struct *tty,char *msg,size_t size);No need for extern.
Will fix, thanks.
Quoted 5 lines
thanks,
Signature · 2 lines
js suse labs