?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
khugepaged: ignore pmd tables with THP mapped with ptes
[linux/fpc-iii.git] / kernel / sysctl.c
blobc810f8afdb7f79439737ee3b2528f7e2b10a233a
1 /*
2 * sysctl.c: General linux system control interface
3 *
4 * Begun 24 March 1995, Stephen Tweedie
5 * Added /proc support, Dec 1995
6 * Added bdflush entry and intvec min/max checking, 2/23/96, Tom Dyas.
7 * Added hooks for /proc/sys/net (minor, minor patch), 96/4/1, Mike Shaver.
8 * Added kernel/java-{interpreter,appletviewer}, 96/5/10, Mike Shaver.
9 * Dynamic registration fixes, Stephen Tweedie.
10 * Added kswapd-interval, ctrl-alt-del, printk stuff, 1/8/97, Chris Horn.
11 * Made sysctl support optional via CONFIG_SYSCTL, 1/10/97, Chris
12 * Horn.
13 * Added proc_doulongvec_ms_jiffies_minmax, 09/08/99, Carlos H. Bauer.
14 * Added proc_doulongvec_minmax, 09/08/99, Carlos H. Bauer.
15 * Changed linked lists to use list.h instead of lists.h, 02/24/00, Bill
16 * Wendling.
17 * The list_for_each() macro wasn't appropriate for the sysctl loop.
18 * Removed it and replaced it with older style, 03/23/00, Bill Wendling
19 */
20
21 #include <linux/module.h>
22 #include <linux/aio.h>
23 #include <linux/mm.h>
24 #include <linux/swap.h>
25 #include <linux/slab.h>
26 #include <linux/sysctl.h>
27 #include <linux/bitmap.h>
28 #include <linux/signal.h>
29 #include <linux/printk.h>
30 #include <linux/proc_fs.h>
31 #include <linux/security.h>
32 #include <linux/ctype.h>
33 #include <linux/kmemcheck.h>
34 #include <linux/kmemleak.h>
35 #include <linux/fs.h>
36 #include <linux/init.h>
37 #include <linux/kernel.h>
38 #include <linux/kobject.h>
39 #include <linux/net.h>
40 #include <linux/sysrq.h>
41 #include <linux/highuid.h>
42 #include <linux/writeback.h>
43 #include <linux/ratelimit.h>
44 #include <linux/compaction.h>
45 #include <linux/hugetlb.h>
46 #include <linux/initrd.h>
47 #include <linux/key.h>
48 #include <linux/times.h>
49 #include <linux/limits.h>
50 #include <linux/dcache.h>
51 #include <linux/dnotify.h>
52 #include <linux/syscalls.h>
53 #include <linux/vmstat.h>
54 #include <linux/nfs_fs.h>
55 #include <linux/acpi.h>
56 #include <linux/reboot.h>
57 #include <linux/ftrace.h>
58 #include <linux/perf_event.h>
59 #include <linux/kprobes.h>
60 #include <linux/pipe_fs_i.h>
61 #include <linux/oom.h>
62 #include <linux/kmod.h>
63 #include <linux/capability.h>
64 #include <linux/binfmts.h>
65 #include <linux/sched/sysctl.h>
66 #include <linux/kexec.h>
67 #include <linux/bpf.h>
68
69 #include <asm/uaccess.h>
70 #include <asm/processor.h>
71
72 #ifdef CONFIG_X86
73 #include <asm/nmi.h>
74 #include <asm/stacktrace.h>
75 #include <asm/io.h>
76 #endif
77 #ifdef CONFIG_SPARC
78 #include <asm/setup.h>
79 #endif
80 #ifdef CONFIG_BSD_PROCESS_ACCT
81 #include <linux/acct.h>
82 #endif
83 #ifdef CONFIG_RT_MUTEXES
84 #include <linux/rtmutex.h>
85 #endif
86 #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_LOCK_STAT)
87 #include <linux/lockdep.h>
88 #endif
89 #ifdef CONFIG_CHR_DEV_SG
90 #include <scsi/sg.h>
91 #endif
92
93 #ifdef CONFIG_LOCKUP_DETECTOR
94 #include <linux/nmi.h>
95 #endif
96
97 #if defined(CONFIG_SYSCTL)
98
99 /* External variables not in a header file. */
100 extern int suid_dumpable;
101 #ifdef CONFIG_COREDUMP
102 extern int core_uses_pid;
103 extern char core_pattern[];
104 extern unsigned int core_pipe_limit;
105 #endif
106 extern int pid_max;
107 extern int pid_max_min, pid_max_max;
108 extern int percpu_pagelist_fraction;
109 extern int compat_log;
110 extern int latencytop_enabled;
111 extern int sysctl_nr_open_min, sysctl_nr_open_max;
112 #ifndef CONFIG_MMU
113 extern int sysctl_nr_trim_pages;
114 #endif
115
116 /* Constants used for minimum and maximum */
117 #ifdef CONFIG_LOCKUP_DETECTOR
118 static int sixty = 60;
119 #endif
120
121 static int __maybe_unused neg_one = -1;
122
123 static int zero;
124 static int __maybe_unused one = 1;
125 static int __maybe_unused two = 2;
126 static int __maybe_unused four = 4;
127 static unsigned long one_ul = 1;
128 static int one_hundred = 100;
129 #ifdef CONFIG_PRINTK
130 static int ten_thousand = 10000;
131 #endif
132
133 /* this is needed for the proc_doulongvec_minmax of vm_dirty_bytes */
134 static unsigned long dirty_bytes_min = 2 * PAGE_SIZE;
135
136 /* this is needed for the proc_dointvec_minmax for [fs_]overflow UID and GID */
137 static int maxolduid = 65535;
138 static int minolduid;
139
140 static int ngroups_max = NGROUPS_MAX;
141 static const int cap_last_cap = CAP_LAST_CAP;
142
143 /*this is needed for proc_doulongvec_minmax of sysctl_hung_task_timeout_secs */
144 #ifdef CONFIG_DETECT_HUNG_TASK
145 static unsigned long hung_task_timeout_max = (LONG_MAX/HZ);
146 #endif
147
148 #ifdef CONFIG_INOTIFY_USER
149 #include <linux/inotify.h>
150 #endif
151 #ifdef CONFIG_SPARC
152 #endif
153
154 #ifdef __hppa__
155 extern int pwrsw_enabled;
156 #endif
157
158 #ifdef CONFIG_SYSCTL_ARCH_UNALIGN_ALLOW
159 extern int unaligned_enabled;
160 #endif
161
162 #ifdef CONFIG_IA64
163 extern int unaligned_dump_stack;
164 #endif
165
166 #ifdef CONFIG_SYSCTL_ARCH_UNALIGN_NO_WARN
167 extern int no_unaligned_warning;
168 #endif
169
170 #ifdef CONFIG_PROC_SYSCTL
171
172 #define SYSCTL_WRITES_LEGACY -1
173 #define SYSCTL_WRITES_WARN 0
174 #define SYSCTL_WRITES_STRICT 1
175
176 static int sysctl_writes_strict = SYSCTL_WRITES_WARN;
177
178 static int proc_do_cad_pid(struct ctl_table *table, int write,
179 void __user *buffer, size_t *lenp, loff_t *ppos);
180 static int proc_taint(struct ctl_table *table, int write,
181 void __user *buffer, size_t *lenp, loff_t *ppos);
182 #endif
183
184 #ifdef CONFIG_PRINTK
185 static int proc_dointvec_minmax_sysadmin(struct ctl_table *table, int write,
186 void __user *buffer, size_t *lenp, loff_t *ppos);
187 #endif
188
189 static int proc_dointvec_minmax_coredump(struct ctl_table *table, int write,
190 void __user *buffer, size_t *lenp, loff_t *ppos);
191 #ifdef CONFIG_COREDUMP
192 static int proc_dostring_coredump(struct ctl_table *table, int write,
193 void __user *buffer, size_t *lenp, loff_t *ppos);
194 #endif
195
196 #ifdef CONFIG_MAGIC_SYSRQ
197 /* Note: sysrq code uses it's own private copy */
198 static int __sysrq_enabled = CONFIG_MAGIC_SYSRQ_DEFAULT_ENABLE;
199
200 static int sysrq_sysctl_handler(struct ctl_table *table, int write,
201 void __user *buffer, size_t *lenp,
202 loff_t *ppos)
203 {
204 int error;
205
206 error = proc_dointvec(table, write, buffer, lenp, ppos);
207 if (error)
208 return error;
209
210 if (write)
211 sysrq_toggle_support(__sysrq_enabled);
212
213 return 0;
214 }
215
216 #endif
217
218 static struct ctl_table kern_table[];
219 static struct ctl_table vm_table[];
220 static struct ctl_table fs_table[];
221 static struct ctl_table debug_table[];
222 static struct ctl_table dev_table[];
223 extern struct ctl_table random_table[];
224 #ifdef CONFIG_EPOLL
225 extern struct ctl_table epoll_table[];
226 #endif
227
228 #ifdef HAVE_ARCH_PICK_MMAP_LAYOUT
229 int sysctl_legacy_va_layout;
230 #endif
231
232 /* The default sysctl tables: */
233
234 static struct ctl_table sysctl_base_table[] = {
235 {
236 .procname = "kernel",
237 .mode = 0555,
238 .child = kern_table,
239 },
240 {
241 .procname = "vm",
242 .mode = 0555,
243 .child = vm_table,
244 },
245 {
246 .procname = "fs",
247 .mode = 0555,
248 .child = fs_table,
249 },
250 {
251 .procname = "debug",
252 .mode = 0555,
253 .child = debug_table,
254 },
255 {
256 .procname = "dev",
257 .mode = 0555,
258 .child = dev_table,
259 },
260 { }
261 };
262
263 #ifdef CONFIG_SCHED_DEBUG
264 static int min_sched_granularity_ns = 100000; /* 100 usecs */
265 static int max_sched_granularity_ns = NSEC_PER_SEC; /* 1 second */
266 static int min_wakeup_granularity_ns; /* 0 usecs */
267 static int max_wakeup_granularity_ns = NSEC_PER_SEC; /* 1 second */
268 #ifdef CONFIG_SMP
269 static int min_sched_tunable_scaling = SCHED_TUNABLESCALING_NONE;
270 static int max_sched_tunable_scaling = SCHED_TUNABLESCALING_END-1;
271 #endif /* CONFIG_SMP */
272 #endif /* CONFIG_SCHED_DEBUG */
273
274 #ifdef CONFIG_COMPACTION
275 static int min_extfrag_threshold;
276 static int max_extfrag_threshold = 1000;
277 #endif
278
279 static struct ctl_table kern_table[] = {
280 {
281 .procname = "sched_child_runs_first",
282 .data = &sysctl_sched_child_runs_first,
283 .maxlen = sizeof(unsigned int),
284 .mode = 0644,
285 .proc_handler = proc_dointvec,
286 },
287 #ifdef CONFIG_SCHED_DEBUG
288 {
289 .procname = "sched_min_granularity_ns",
290 .data = &sysctl_sched_min_granularity,
291 .maxlen = sizeof(unsigned int),
292 .mode = 0644,
293 .proc_handler = sched_proc_update_handler,
294 .extra1 = &min_sched_granularity_ns,
295 .extra2 = &max_sched_granularity_ns,
296 },
297 {
298 .procname = "sched_latency_ns",
299 .data = &sysctl_sched_latency,
300 .maxlen = sizeof(unsigned int),
301 .mode = 0644,
302 .proc_handler = sched_proc_update_handler,
303 .extra1 = &min_sched_granularity_ns,
304 .extra2 = &max_sched_granularity_ns,
305 },
306 {
307 .procname = "sched_wakeup_granularity_ns",
308 .data = &sysctl_sched_wakeup_granularity,
309 .maxlen = sizeof(unsigned int),
310 .mode = 0644,
311 .proc_handler = sched_proc_update_handler,
312 .extra1 = &min_wakeup_granularity_ns,
313 .extra2 = &max_wakeup_granularity_ns,
314 },
315 #ifdef CONFIG_SMP
316 {
317 .procname = "sched_tunable_scaling",
318 .data = &sysctl_sched_tunable_scaling,
319 .maxlen = sizeof(enum sched_tunable_scaling),
320 .mode = 0644,
321 .proc_handler = sched_proc_update_handler,
322 .extra1 = &min_sched_tunable_scaling,
323 .extra2 = &max_sched_tunable_scaling,
324 },
325 {
326 .procname = "sched_migration_cost_ns",
327 .data = &sysctl_sched_migration_cost,
328 .maxlen = sizeof(unsigned int),
329 .mode = 0644,
330 .proc_handler = proc_dointvec,
331 },
332 {
333 .procname = "sched_nr_migrate",
334 .data = &sysctl_sched_nr_migrate,
335 .maxlen = sizeof(unsigned int),
336 .mode = 0644,
337 .proc_handler = proc_dointvec,
338 },
339 {
340 .procname = "sched_time_avg_ms",
341 .data = &sysctl_sched_time_avg,
342 .maxlen = sizeof(unsigned int),
343 .mode = 0644,
344 .proc_handler = proc_dointvec,
345 },
346 {
347 .procname = "sched_shares_window_ns",
348 .data = &sysctl_sched_shares_window,
349 .maxlen = sizeof(unsigned int),
350 .mode = 0644,
351 .proc_handler = proc_dointvec,
352 },
353 #endif /* CONFIG_SMP */
354 #ifdef CONFIG_NUMA_BALANCING
355 {
356 .procname = "numa_balancing_scan_delay_ms",
357 .data = &sysctl_numa_balancing_scan_delay,
358 .maxlen = sizeof(unsigned int),
359 .mode = 0644,
360 .proc_handler = proc_dointvec,
361 },
362 {
363 .procname = "numa_balancing_scan_period_min_ms",
364 .data = &sysctl_numa_balancing_scan_period_min,
365 .maxlen = sizeof(unsigned int),
366 .mode = 0644,
367 .proc_handler = proc_dointvec,
368 },
369 {
370 .procname = "numa_balancing_scan_period_max_ms",
371 .data = &sysctl_numa_balancing_scan_period_max,
372 .maxlen = sizeof(unsigned int),
373 .mode = 0644,
374 .proc_handler = proc_dointvec,
375 },
376 {
377 .procname = "numa_balancing_scan_size_mb",
378 .data = &sysctl_numa_balancing_scan_size,
379 .maxlen = sizeof(unsigned int),
380 .mode = 0644,
381 .proc_handler = proc_dointvec_minmax,
382 .extra1 = &one,
383 },
384 {
385 .procname = "numa_balancing",
386 .data = NULL, /* filled in by handler */
387 .maxlen = sizeof(unsigned int),
388 .mode = 0644,
389 .proc_handler = sysctl_numa_balancing,
390 .extra1 = &zero,
391 .extra2 = &one,
392 },
393 #endif /* CONFIG_NUMA_BALANCING */
394 #endif /* CONFIG_SCHED_DEBUG */
395 {
396 .procname = "sched_rt_period_us",
397 .data = &sysctl_sched_rt_period,
398 .maxlen = sizeof(unsigned int),
399 .mode = 0644,
400 .proc_handler = sched_rt_handler,
401 },
402 {
403 .procname = "sched_rt_runtime_us",
404 .data = &sysctl_sched_rt_runtime,
405 .maxlen = sizeof(int),
406 .mode = 0644,
407 .proc_handler = sched_rt_handler,
408 },
409 {
410 .procname = "sched_rr_timeslice_ms",
411 .data = &sched_rr_timeslice,
412 .maxlen = sizeof(int),
413 .mode = 0644,
414 .proc_handler = sched_rr_handler,
415 },
416 #ifdef CONFIG_SCHED_AUTOGROUP
417 {
418 .procname = "sched_autogroup_enabled",
419 .data = &sysctl_sched_autogroup_enabled,
420 .maxlen = sizeof(unsigned int),
421 .mode = 0644,
422 .proc_handler = proc_dointvec_minmax,
423 .extra1 = &zero,
424 .extra2 = &one,
425 },
426 #endif
427 #ifdef CONFIG_CFS_BANDWIDTH
428 {
429 .procname = "sched_cfs_bandwidth_slice_us",
430 .data = &sysctl_sched_cfs_bandwidth_slice,
431 .maxlen = sizeof(unsigned int),
432 .mode = 0644,
433 .proc_handler = proc_dointvec_minmax,
434 .extra1 = &one,
435 },
436 #endif
437 #ifdef CONFIG_PROVE_LOCKING
438 {
439 .procname = "prove_locking",
440 .data = &prove_locking,
441 .maxlen = sizeof(int),
442 .mode = 0644,
443 .proc_handler = proc_dointvec,
444 },
445 #endif
446 #ifdef CONFIG_LOCK_STAT
447 {
448 .procname = "lock_stat",
449 .data = &lock_stat,
450 .maxlen = sizeof(int),
451 .mode = 0644,
452 .proc_handler = proc_dointvec,
453 },
454 #endif
455 {
456 .procname = "panic",
457 .data = &panic_timeout,
458 .maxlen = sizeof(int),
459 .mode = 0644,
460 .proc_handler = proc_dointvec,
461 },
462 #ifdef CONFIG_COREDUMP
463 {
464 .procname = "core_uses_pid",
465 .data = &core_uses_pid,
466 .maxlen = sizeof(int),
467 .mode = 0644,
468 .proc_handler = proc_dointvec,
469 },
470 {
471 .procname = "core_pattern",
472 .data = core_pattern,
473 .maxlen = CORENAME_MAX_SIZE,
474 .mode = 0644,
475 .proc_handler = proc_dostring_coredump,
476 },
477 {
478 .procname = "core_pipe_limit",
479 .data = &core_pipe_limit,
480 .maxlen = sizeof(unsigned int),
481 .mode = 0644,
482 .proc_handler = proc_dointvec,
483 },
484 #endif
485 #ifdef CONFIG_PROC_SYSCTL
486 {
487 .procname = "tainted",
488 .maxlen = sizeof(long),
489 .mode = 0644,
490 .proc_handler = proc_taint,
491 },
492 {
493 .procname = "sysctl_writes_strict",
494 .data = &sysctl_writes_strict,
495 .maxlen = sizeof(int),
496 .mode = 0644,
497 .proc_handler = proc_dointvec_minmax,
498 .extra1 = &neg_one,
499 .extra2 = &one,
500 },
501 #endif
502 #ifdef CONFIG_LATENCYTOP
503 {
504 .procname = "latencytop",
505 .data = &latencytop_enabled,
506 .maxlen = sizeof(int),
507 .mode = 0644,
508 .proc_handler = proc_dointvec,
509 },
510 #endif
511 #ifdef CONFIG_BLK_DEV_INITRD
512 {
513 .procname = "real-root-dev",
514 .data = &real_root_dev,
515 .maxlen = sizeof(int),
516 .mode = 0644,
517 .proc_handler = proc_dointvec,
518 },
519 #endif
520 {
521 .procname = "print-fatal-signals",
522 .data = &print_fatal_signals,
523 .maxlen = sizeof(int),
524 .mode = 0644,
525 .proc_handler = proc_dointvec,
526 },
527 #ifdef CONFIG_SPARC
528 {
529 .procname = "reboot-cmd",
530 .data = reboot_command,
531 .maxlen = 256,
532 .mode = 0644,
533 .proc_handler = proc_dostring,
534 },
535 {
536 .procname = "stop-a",
537 .data = &stop_a_enabled,
538 .maxlen = sizeof (int),
539 .mode = 0644,
540 .proc_handler = proc_dointvec,
541 },
542 {
543 .procname = "scons-poweroff",
544 .data = &scons_pwroff,
545 .maxlen = sizeof (int),
546 .mode = 0644,
547 .proc_handler = proc_dointvec,
548 },
549 #endif
550 #ifdef CONFIG_SPARC64
551 {
552 .procname = "tsb-ratio",
553 .data = &sysctl_tsb_ratio,
554 .maxlen = sizeof (int),
555 .mode = 0644,
556 .proc_handler = proc_dointvec,
557 },
558 #endif
559 #ifdef __hppa__
560 {
561 .procname = "soft-power",
562 .data = &pwrsw_enabled,
563 .maxlen = sizeof (int),
564 .mode = 0644,
565 .proc_handler = proc_dointvec,
566 },
567 #endif
568 #ifdef CONFIG_SYSCTL_ARCH_UNALIGN_ALLOW
569 {
570 .procname = "unaligned-trap",
571 .data = &unaligned_enabled,
572 .maxlen = sizeof (int),
573 .mode = 0644,
574 .proc_handler = proc_dointvec,
575 },
576 #endif
577 {
578 .procname = "ctrl-alt-del",
579 .data = &C_A_D,
580 .maxlen = sizeof(int),
581 .mode = 0644,
582 .proc_handler = proc_dointvec,
583 },
584 #ifdef CONFIG_FUNCTION_TRACER
585 {
586 .procname = "ftrace_enabled",
587 .data = &ftrace_enabled,
588 .maxlen = sizeof(int),
589 .mode = 0644,
590 .proc_handler = ftrace_enable_sysctl,
591 },
592 #endif
593 #ifdef CONFIG_STACK_TRACER
594 {
595 .procname = "stack_tracer_enabled",
596 .data = &stack_tracer_enabled,
597 .maxlen = sizeof(int),
598 .mode = 0644,
599 .proc_handler = stack_trace_sysctl,
600 },
601 #endif
602 #ifdef CONFIG_TRACING
603 {
604 .procname = "ftrace_dump_on_oops",
605 .data = &ftrace_dump_on_oops,
606 .maxlen = sizeof(int),
607 .mode = 0644,
608 .proc_handler = proc_dointvec,
609 },
610 {
611 .procname = "traceoff_on_warning",
612 .data = &__disable_trace_on_warning,
613 .maxlen = sizeof(__disable_trace_on_warning),
614 .mode = 0644,
615 .proc_handler = proc_dointvec,
616 },
617 {
618 .procname = "tracepoint_printk",
619 .data = &tracepoint_printk,
620 .maxlen = sizeof(tracepoint_printk),
621 .mode = 0644,
622 .proc_handler = proc_dointvec,
623 },
624 #endif
625 #ifdef CONFIG_KEXEC_CORE
626 {
627 .procname = "kexec_load_disabled",
628 .data = &kexec_load_disabled,
629 .maxlen = sizeof(int),
630 .mode = 0644,
631 /* only handle a transition from default "0" to "1" */
632 .proc_handler = proc_dointvec_minmax,
633 .extra1 = &one,
634 .extra2 = &one,
635 },
636 #endif
637 #ifdef CONFIG_MODULES
638 {
639 .procname = "modprobe",
640 .data = &modprobe_path,
641 .maxlen = KMOD_PATH_LEN,
642 .mode = 0644,
643 .proc_handler = proc_dostring,
644 },
645 {
646 .procname = "modules_disabled",
647 .data = &modules_disabled,
648 .maxlen = sizeof(int),
649 .mode = 0644,
650 /* only handle a transition from default "0" to "1" */
651 .proc_handler = proc_dointvec_minmax,
652 .extra1 = &one,
653 .extra2 = &one,
654 },
655 #endif
656 #ifdef CONFIG_UEVENT_HELPER
657 {
658 .procname = "hotplug",
659 .data = &uevent_helper,
660 .maxlen = UEVENT_HELPER_PATH_LEN,
661 .mode = 0644,
662 .proc_handler = proc_dostring,
663 },
664 #endif
665 #ifdef CONFIG_CHR_DEV_SG
666 {
667 .procname = "sg-big-buff",
668 .data = &sg_big_buff,
669 .maxlen = sizeof (int),
670 .mode = 0444,
671 .proc_handler = proc_dointvec,
672 },
673 #endif
674 #ifdef CONFIG_BSD_PROCESS_ACCT
675 {
676 .procname = "acct",
677 .data = &acct_parm,
678 .maxlen = 3*sizeof(int),
679 .mode = 0644,
680 .proc_handler = proc_dointvec,
681 },
682 #endif
683 #ifdef CONFIG_MAGIC_SYSRQ
684 {
685 .procname = "sysrq",
686 .data = &__sysrq_enabled,
687 .maxlen = sizeof (int),
688 .mode = 0644,
689 .proc_handler = sysrq_sysctl_handler,
690 },
691 #endif
692 #ifdef CONFIG_PROC_SYSCTL
693 {
694 .procname = "cad_pid",
695 .data = NULL,
696 .maxlen = sizeof (int),
697 .mode = 0600,
698 .proc_handler = proc_do_cad_pid,
699 },
700 #endif
701 {
702 .procname = "threads-max",
703 .data = NULL,
704 .maxlen = sizeof(int),
705 .mode = 0644,
706 .proc_handler = sysctl_max_threads,
707 },
708 {
709 .procname = "random",
710 .mode = 0555,
711 .child = random_table,
712 },
713 {
714 .procname = "usermodehelper",
715 .mode = 0555,
716 .child = usermodehelper_table,
717 },
718 {
719 .procname = "overflowuid",
720 .data = &overflowuid,
721 .maxlen = sizeof(int),
722 .mode = 0644,
723 .proc_handler = proc_dointvec_minmax,
724 .extra1 = &minolduid,
725 .extra2 = &maxolduid,
726 },
727 {
728 .procname = "overflowgid",
729 .data = &overflowgid,
730 .maxlen = sizeof(int),
731 .mode = 0644,
732 .proc_handler = proc_dointvec_minmax,
733 .extra1 = &minolduid,
734 .extra2 = &maxolduid,
735 },
736 #ifdef CONFIG_S390
737 #ifdef CONFIG_MATHEMU
738 {
739 .procname = "ieee_emulation_warnings",
740 .data = &sysctl_ieee_emulation_warnings,
741 .maxlen = sizeof(int),
742 .mode = 0644,
743 .proc_handler = proc_dointvec,
744 },
745 #endif
746 {
747 .procname = "userprocess_debug",
748 .data = &show_unhandled_signals,
749 .maxlen = sizeof(int),
750 .mode = 0644,
751 .proc_handler = proc_dointvec,
752 },
753 #endif
754 {
755 .procname = "pid_max",
756 .data = &pid_max,
757 .maxlen = sizeof (int),
758 .mode = 0644,
759 .proc_handler = proc_dointvec_minmax,
760 .extra1 = &pid_max_min,
761 .extra2 = &pid_max_max,
762 },
763 {
764 .procname = "panic_on_oops",
765 .data = &panic_on_oops,
766 .maxlen = sizeof(int),
767 .mode = 0644,
768 .proc_handler = proc_dointvec,
769 },
770 #if defined CONFIG_PRINTK
771 {
772 .procname = "printk",
773 .data = &console_loglevel,
774 .maxlen = 4*sizeof(int),
775 .mode = 0644,
776 .proc_handler = proc_dointvec,
777 },
778 {
779 .procname = "printk_ratelimit",
780 .data = &printk_ratelimit_state.interval,
781 .maxlen = sizeof(int),
782 .mode = 0644,
783 .proc_handler = proc_dointvec_jiffies,
784 },
785 {
786 .procname = "printk_ratelimit_burst",
787 .data = &printk_ratelimit_state.burst,
788 .maxlen = sizeof(int),
789 .mode = 0644,
790 .proc_handler = proc_dointvec,
791 },
792 {
793 .procname = "printk_delay",
794 .data = &printk_delay_msec,
795 .maxlen = sizeof(int),
796 .mode = 0644,
797 .proc_handler = proc_dointvec_minmax,
798 .extra1 = &zero,
799 .extra2 = &ten_thousand,
800 },
801 {
802 .procname = "dmesg_restrict",
803 .data = &dmesg_restrict,
804 .maxlen = sizeof(int),
805 .mode = 0644,
806 .proc_handler = proc_dointvec_minmax_sysadmin,
807 .extra1 = &zero,
808 .extra2 = &one,
809 },
810 {
811 .procname = "kptr_restrict",
812 .data = &kptr_restrict,
813 .maxlen = sizeof(int),
814 .mode = 0644,
815 .proc_handler = proc_dointvec_minmax_sysadmin,
816 .extra1 = &zero,
817 .extra2 = &two,
818 },
819 #endif
820 {
821 .procname = "ngroups_max",
822 .data = &ngroups_max,
823 .maxlen = sizeof (int),
824 .mode = 0444,
825 .proc_handler = proc_dointvec,
826 },
827 {
828 .procname = "cap_last_cap",
829 .data = (void *)&cap_last_cap,
830 .maxlen = sizeof(int),
831 .mode = 0444,
832 .proc_handler = proc_dointvec,
833 },
834 #if defined(CONFIG_LOCKUP_DETECTOR)
835 {
836 .procname = "watchdog",
837 .data = &watchdog_user_enabled,
838 .maxlen = sizeof (int),
839 .mode = 0644,
840 .proc_handler = proc_watchdog,
841 .extra1 = &zero,
842 .extra2 = &one,
843 },
844 {
845 .procname = "watchdog_thresh",
846 .data = &watchdog_thresh,
847 .maxlen = sizeof(int),
848 .mode = 0644,
849 .proc_handler = proc_watchdog_thresh,
850 .extra1 = &zero,
851 .extra2 = &sixty,
852 },
853 {
854 .procname = "nmi_watchdog",
855 .data = &nmi_watchdog_enabled,
856 .maxlen = sizeof (int),
857 .mode = 0644,
858 .proc_handler = proc_nmi_watchdog,
859 .extra1 = &zero,
860 #if defined(CONFIG_HAVE_NMI_WATCHDOG) || defined(CONFIG_HARDLOCKUP_DETECTOR)
861 .extra2 = &one,
862 #else
863 .extra2 = &zero,
864 #endif
865 },
866 {
867 .procname = "soft_watchdog",
868 .data = &soft_watchdog_enabled,
869 .maxlen = sizeof (int),
870 .mode = 0644,
871 .proc_handler = proc_soft_watchdog,
872 .extra1 = &zero,
873 .extra2 = &one,
874 },
875 {
876 .procname = "watchdog_cpumask",
877 .data = &watchdog_cpumask_bits,
878 .maxlen = NR_CPUS,
879 .mode = 0644,
880 .proc_handler = proc_watchdog_cpumask,
881 },
882 {
883 .procname = "softlockup_panic",
884 .data = &softlockup_panic,
885 .maxlen = sizeof(int),
886 .mode = 0644,
887 .proc_handler = proc_dointvec_minmax,
888 .extra1 = &zero,
889 .extra2 = &one,
890 },
891 #ifdef CONFIG_HARDLOCKUP_DETECTOR
892 {
893 .procname = "hardlockup_panic",
894 .data = &hardlockup_panic,
895 .maxlen = sizeof(int),
896 .mode = 0644,
897 .proc_handler = proc_dointvec_minmax,
898 .extra1 = &zero,
899 .extra2 = &one,
900 },
901 #endif
902 #ifdef CONFIG_SMP
903 {
904 .procname = "softlockup_all_cpu_backtrace",
905 .data = &sysctl_softlockup_all_cpu_backtrace,
906 .maxlen = sizeof(int),
907 .mode = 0644,
908 .proc_handler = proc_dointvec_minmax,
909 .extra1 = &zero,
910 .extra2 = &one,
911 },
912 {
913 .procname = "hardlockup_all_cpu_backtrace",
914 .data = &sysctl_hardlockup_all_cpu_backtrace,
915 .maxlen = sizeof(int),
916 .mode = 0644,
917 .proc_handler = proc_dointvec_minmax,
918 .extra1 = &zero,
919 .extra2 = &one,
920 },
921 #endif /* CONFIG_SMP */
922 #endif
923 #if defined(CONFIG_X86_LOCAL_APIC) && defined(CONFIG_X86)
924 {
925 .procname = "unknown_nmi_panic",
926 .data = &unknown_nmi_panic,
927 .maxlen = sizeof (int),
928 .mode = 0644,
929 .proc_handler = proc_dointvec,
930 },
931 #endif
932 #if defined(CONFIG_X86)
933 {
934 .procname = "panic_on_unrecovered_nmi",
935 .data = &panic_on_unrecovered_nmi,
936 .maxlen = sizeof(int),
937 .mode = 0644,
938 .proc_handler = proc_dointvec,
939 },
940 {
941 .procname = "panic_on_io_nmi",
942 .data = &panic_on_io_nmi,
943 .maxlen = sizeof(int),
944 .mode = 0644,
945 .proc_handler = proc_dointvec,
946 },
947 #ifdef CONFIG_DEBUG_STACKOVERFLOW
948 {
949 .procname = "panic_on_stackoverflow",
950 .data = &sysctl_panic_on_stackoverflow,
951 .maxlen = sizeof(int),
952 .mode = 0644,
953 .proc_handler = proc_dointvec,
954 },
955 #endif
956 {
957 .procname = "bootloader_type",
958 .data = &bootloader_type,
959 .maxlen = sizeof (int),
960 .mode = 0444,
961 .proc_handler = proc_dointvec,
962 },
963 {
964 .procname = "bootloader_version",
965 .data = &bootloader_version,
966 .maxlen = sizeof (int),
967 .mode = 0444,
968 .proc_handler = proc_dointvec,
969 },
970 {
971 .procname = "kstack_depth_to_print",
972 .data = &kstack_depth_to_print,
973 .maxlen = sizeof(int),
974 .mode = 0644,
975 .proc_handler = proc_dointvec,
976 },
977 {
978 .procname = "io_delay_type",
979 .data = &io_delay_type,
980 .maxlen = sizeof(int),
981 .mode = 0644,
982 .proc_handler = proc_dointvec,
983 },
984 #endif
985 #if defined(CONFIG_MMU)
986 {
987 .procname = "randomize_va_space",
988 .data = &randomize_va_space,
989 .maxlen = sizeof(int),
990 .mode = 0644,
991 .proc_handler = proc_dointvec,
992 },
993 #endif
994 #if defined(CONFIG_S390) && defined(CONFIG_SMP)
995 {
996 .procname = "spin_retry",
997 .data = &spin_retry,
998 .maxlen = sizeof (int),
999 .mode = 0644,
1000 .proc_handler = proc_dointvec,
1001 },
1002 #endif
1003 #if defined(CONFIG_ACPI_SLEEP) && defined(CONFIG_X86)
1004 {
1005 .procname = "acpi_video_flags",
1006 .data = &acpi_realmode_flags,
1007 .maxlen = sizeof (unsigned long),
1008 .mode = 0644,
1009 .proc_handler = proc_doulongvec_minmax,
1010 },
1011 #endif
1012 #ifdef CONFIG_SYSCTL_ARCH_UNALIGN_NO_WARN
1013 {
1014 .procname = "ignore-unaligned-usertrap",
1015 .data = &no_unaligned_warning,
1016 .maxlen = sizeof (int),
1017 .mode = 0644,
1018 .proc_handler = proc_dointvec,
1019 },
1020 #endif
1021 #ifdef CONFIG_IA64
1022 {
1023 .procname = "unaligned-dump-stack",
1024 .data = &unaligned_dump_stack,
1025 .maxlen = sizeof (int),
1026 .mode = 0644,
1027 .proc_handler = proc_dointvec,
1028 },
1029 #endif
1030 #ifdef CONFIG_DETECT_HUNG_TASK
1031 {
1032 .procname = "hung_task_panic",
1033 .data = &sysctl_hung_task_panic,
1034 .maxlen = sizeof(int),
1035 .mode = 0644,
1036 .proc_handler = proc_dointvec_minmax,
1037 .extra1 = &zero,
1038 .extra2 = &one,
1039 },
1040 {
1041 .procname = "hung_task_check_count",
1042 .data = &sysctl_hung_task_check_count,
1043 .maxlen = sizeof(int),
1044 .mode = 0644,
1045 .proc_handler = proc_dointvec_minmax,
1046 .extra1 = &zero,
1047 },
1048 {
1049 .procname = "hung_task_timeout_secs",
1050 .data = &sysctl_hung_task_timeout_secs,
1051 .maxlen = sizeof(unsigned long),
1052 .mode = 0644,
1053 .proc_handler = proc_dohung_task_timeout_secs,
1054 .extra2 = &hung_task_timeout_max,
1055 },
1056 {
1057 .procname = "hung_task_warnings",
1058 .data = &sysctl_hung_task_warnings,
1059 .maxlen = sizeof(int),
1060 .mode = 0644,
1061 .proc_handler = proc_dointvec_minmax,
1062 .extra1 = &neg_one,
1063 },
1064 #endif
1065 #ifdef CONFIG_COMPAT
1066 {
1067 .procname = "compat-log",
1068 .data = &compat_log,
1069 .maxlen = sizeof (int),
1070 .mode = 0644,
1071 .proc_handler = proc_dointvec,
1072 },
1073 #endif
1074 #ifdef CONFIG_RT_MUTEXES
1075 {
1076 .procname = "max_lock_depth",
1077 .data = &max_lock_depth,
1078 .maxlen = sizeof(int),
1079 .mode = 0644,
1080 .proc_handler = proc_dointvec,
1081 },
1082 #endif
1083 {
1084 .procname = "poweroff_cmd",
1085 .data = &poweroff_cmd,
1086 .maxlen = POWEROFF_CMD_PATH_LEN,
1087 .mode = 0644,
1088 .proc_handler = proc_dostring,
1089 },
1090 #ifdef CONFIG_KEYS
1091 {
1092 .procname = "keys",
1093 .mode = 0555,
1094 .child = key_sysctls,
1095 },
1096 #endif
1097 #ifdef CONFIG_PERF_EVENTS
1098 /*
1099 * User-space scripts rely on the existence of this file
1100 * as a feature check for perf_events being enabled.
1101 *
1102 * So it's an ABI, do not remove!
1103 */
1104 {
1105 .procname = "perf_event_paranoid",
1106 .data = &sysctl_perf_event_paranoid,
1107 .maxlen = sizeof(sysctl_perf_event_paranoid),
1108 .mode = 0644,
1109 .proc_handler = proc_dointvec,
1110 },
1111 {
1112 .procname = "perf_event_mlock_kb",
1113 .data = &sysctl_perf_event_mlock,
1114 .maxlen = sizeof(sysctl_perf_event_mlock),
1115 .mode = 0644,
1116 .proc_handler = proc_dointvec,
1117 },
1118 {
1119 .procname = "perf_event_max_sample_rate",
1120 .data = &sysctl_perf_event_sample_rate,
1121 .maxlen = sizeof(sysctl_perf_event_sample_rate),
1122 .mode = 0644,
1123 .proc_handler = perf_proc_update_handler,
1124 .extra1 = &one,
1125 },
1126 {
1127 .procname = "perf_cpu_time_max_percent",
1128 .data = &sysctl_perf_cpu_time_max_percent,
1129 .maxlen = sizeof(sysctl_perf_cpu_time_max_percent),
1130 .mode = 0644,
1131 .proc_handler = perf_cpu_time_max_percent_handler,
1132 .extra1 = &zero,
1133 .extra2 = &one_hundred,
1134 },
1135 #endif
1136 #ifdef CONFIG_KMEMCHECK
1137 {
1138 .procname = "kmemcheck",
1139 .data = &kmemcheck_enabled,
1140 .maxlen = sizeof(int),
1141 .mode = 0644,
1142 .proc_handler = proc_dointvec,
1143 },
1144 #endif
1145 {
1146 .procname = "panic_on_warn",
1147 .data = &panic_on_warn,
1148 .maxlen = sizeof(int),
1149 .mode = 0644,
1150 .proc_handler = proc_dointvec_minmax,
1151 .extra1 = &zero,
1152 .extra2 = &one,
1153 },
1154 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
1155 {
1156 .procname = "timer_migration",
1157 .data = &sysctl_timer_migration,
1158 .maxlen = sizeof(unsigned int),
1159 .mode = 0644,
1160 .proc_handler = timer_migration_handler,
1161 },
1162 #endif
1163 #ifdef CONFIG_BPF_SYSCALL
1164 {
1165 .procname = "unprivileged_bpf_disabled",
1166 .data = &sysctl_unprivileged_bpf_disabled,
1167 .maxlen = sizeof(sysctl_unprivileged_bpf_disabled),
1168 .mode = 0644,
1169 /* only handle a transition from default "0" to "1" */
1170 .proc_handler = proc_dointvec_minmax,
1171 .extra1 = &one,
1172 .extra2 = &one,
1173 },
1174 #endif
1175 { }
1176 };
1177
1178 static struct ctl_table vm_table[] = {
1179 {
1180 .procname = "overcommit_memory",
1181 .data = &sysctl_overcommit_memory,
1182 .maxlen = sizeof(sysctl_overcommit_memory),
1183 .mode = 0644,
1184 .proc_handler = proc_dointvec_minmax,
1185 .extra1 = &zero,
1186 .extra2 = &two,
1187 },
1188 {
1189 .procname = "panic_on_oom",
1190 .data = &sysctl_panic_on_oom,
1191 .maxlen = sizeof(sysctl_panic_on_oom),
1192 .mode = 0644,
1193 .proc_handler = proc_dointvec_minmax,
1194 .extra1 = &zero,
1195 .extra2 = &two,
1196 },
1197 {
1198 .procname = "oom_kill_allocating_task",
1199 .data = &sysctl_oom_kill_allocating_task,
1200 .maxlen = sizeof(sysctl_oom_kill_allocating_task),
1201 .mode = 0644,
1202 .proc_handler = proc_dointvec,
1203 },
1204 {
1205 .procname = "oom_dump_tasks",
1206 .data = &sysctl_oom_dump_tasks,
1207 .maxlen = sizeof(sysctl_oom_dump_tasks),
1208 .mode = 0644,
1209 .proc_handler = proc_dointvec,
1210 },
1211 {
1212 .procname = "overcommit_ratio",
1213 .data = &sysctl_overcommit_ratio,
1214 .maxlen = sizeof(sysctl_overcommit_ratio),
1215 .mode = 0644,
1216 .proc_handler = overcommit_ratio_handler,
1217 },
1218 {
1219 .procname = "overcommit_kbytes",
1220 .data = &sysctl_overcommit_kbytes,
1221 .maxlen = sizeof(sysctl_overcommit_kbytes),
1222 .mode = 0644,
1223 .proc_handler = overcommit_kbytes_handler,
1224 },
1225 {
1226 .procname = "page-cluster",
1227 .data = &page_cluster,
1228 .maxlen = sizeof(int),
1229 .mode = 0644,
1230 .proc_handler = proc_dointvec_minmax,
1231 .extra1 = &zero,
1232 },
1233 {
1234 .procname = "dirty_background_ratio",
1235 .data = &dirty_background_ratio,
1236 .maxlen = sizeof(dirty_background_ratio),
1237 .mode = 0644,
1238 .proc_handler = dirty_background_ratio_handler,
1239 .extra1 = &zero,
1240 .extra2 = &one_hundred,
1241 },
1242 {
1243 .procname = "dirty_background_bytes",
1244 .data = &dirty_background_bytes,
1245 .maxlen = sizeof(dirty_background_bytes),
1246 .mode = 0644,
1247 .proc_handler = dirty_background_bytes_handler,
1248 .extra1 = &one_ul,
1249 },
1250 {
1251 .procname = "dirty_ratio",
1252 .data = &vm_dirty_ratio,
1253 .maxlen = sizeof(vm_dirty_ratio),
1254 .mode = 0644,
1255 .proc_handler = dirty_ratio_handler,
1256 .extra1 = &zero,
1257 .extra2 = &one_hundred,
1258 },
1259 {
1260 .procname = "dirty_bytes",
1261 .data = &vm_dirty_bytes,
1262 .maxlen = sizeof(vm_dirty_bytes),
1263 .mode = 0644,
1264 .proc_handler = dirty_bytes_handler,
1265 .extra1 = &dirty_bytes_min,
1266 },
1267 {
1268 .procname = "dirty_writeback_centisecs",
1269 .data = &dirty_writeback_interval,
1270 .maxlen = sizeof(dirty_writeback_interval),
1271 .mode = 0644,
1272 .proc_handler = dirty_writeback_centisecs_handler,
1273 },
1274 {
1275 .procname = "dirty_expire_centisecs",
1276 .data = &dirty_expire_interval,
1277 .maxlen = sizeof(dirty_expire_interval),
1278 .mode = 0644,
1279 .proc_handler = proc_dointvec_minmax,
1280 .extra1 = &zero,
1281 },
1282 {
1283 .procname = "dirtytime_expire_seconds",
1284 .data = &dirtytime_expire_interval,
1285 .maxlen = sizeof(dirty_expire_interval),
1286 .mode = 0644,
1287 .proc_handler = dirtytime_interval_handler,
1288 .extra1 = &zero,
1289 },
1290 {
1291 .procname = "nr_pdflush_threads",
1292 .mode = 0444 /* read-only */,
1293 .proc_handler = pdflush_proc_obsolete,
1294 },
1295 {
1296 .procname = "swappiness",
1297 .data = &vm_swappiness,
1298 .maxlen = sizeof(vm_swappiness),
1299 .mode = 0644,
1300 .proc_handler = proc_dointvec_minmax,
1301 .extra1 = &zero,
1302 .extra2 = &one_hundred,
1303 },
1304 #ifdef CONFIG_HUGETLB_PAGE
1305 {
1306 .procname = "nr_hugepages",
1307 .data = NULL,
1308 .maxlen = sizeof(unsigned long),
1309 .mode = 0644,
1310 .proc_handler = hugetlb_sysctl_handler,
1311 },
1312 #ifdef CONFIG_NUMA
1313 {
1314 .procname = "nr_hugepages_mempolicy",
1315 .data = NULL,
1316 .maxlen = sizeof(unsigned long),
1317 .mode = 0644,
1318 .proc_handler = &hugetlb_mempolicy_sysctl_handler,
1319 },
1320 #endif
1321 {
1322 .procname = "hugetlb_shm_group",
1323 .data = &sysctl_hugetlb_shm_group,
1324 .maxlen = sizeof(gid_t),
1325 .mode = 0644,
1326 .proc_handler = proc_dointvec,
1327 },
1328 {
1329 .procname = "hugepages_treat_as_movable",
1330 .data = &hugepages_treat_as_movable,
1331 .maxlen = sizeof(int),
1332 .mode = 0644,
1333 .proc_handler = proc_dointvec,
1334 },
1335 {
1336 .procname = "nr_overcommit_hugepages",
1337 .data = NULL,
1338 .maxlen = sizeof(unsigned long),
1339 .mode = 0644,
1340 .proc_handler = hugetlb_overcommit_handler,
1341 },
1342 #endif
1343 {
1344 .procname = "lowmem_reserve_ratio",
1345 .data = &sysctl_lowmem_reserve_ratio,
1346 .maxlen = sizeof(sysctl_lowmem_reserve_ratio),
1347 .mode = 0644,
1348 .proc_handler = lowmem_reserve_ratio_sysctl_handler,
1349 },
1350 {
1351 .procname = "drop_caches",
1352 .data = &sysctl_drop_caches,
1353 .maxlen = sizeof(int),
1354 .mode = 0644,
1355 .proc_handler = drop_caches_sysctl_handler,
1356 .extra1 = &one,
1357 .extra2 = &four,
1358 },
1359 #ifdef CONFIG_COMPACTION
1360 {
1361 .procname = "compact_memory",
1362 .data = &sysctl_compact_memory,
1363 .maxlen = sizeof(int),
1364 .mode = 0200,
1365 .proc_handler = sysctl_compaction_handler,
1366 },
1367 {
1368 .procname = "extfrag_threshold",
1369 .data = &sysctl_extfrag_threshold,
1370 .maxlen = sizeof(int),
1371 .mode = 0644,
1372 .proc_handler = sysctl_extfrag_handler,
1373 .extra1 = &min_extfrag_threshold,
1374 .extra2 = &max_extfrag_threshold,
1375 },
1376 {
1377 .procname = "compact_unevictable_allowed",
1378 .data = &sysctl_compact_unevictable_allowed,
1379 .maxlen = sizeof(int),
1380 .mode = 0644,
1381 .proc_handler = proc_dointvec,
1382 .extra1 = &zero,
1383 .extra2 = &one,
1384 },
1385
1386 #endif /* CONFIG_COMPACTION */
1387 {
1388 .procname = "min_free_kbytes",
1389 .data = &min_free_kbytes,
1390 .maxlen = sizeof(min_free_kbytes),
1391 .mode = 0644,
1392 .proc_handler = min_free_kbytes_sysctl_handler,
1393 .extra1 = &zero,
1394 },
1395 {
1396 .procname = "percpu_pagelist_fraction",
1397 .data = &percpu_pagelist_fraction,
1398 .maxlen = sizeof(percpu_pagelist_fraction),
1399 .mode = 0644,
1400 .proc_handler = percpu_pagelist_fraction_sysctl_handler,
1401 .extra1 = &zero,
1402 },
1403 #ifdef CONFIG_MMU
1404 {
1405 .procname = "max_map_count",
1406 .data = &sysctl_max_map_count,
1407 .maxlen = sizeof(sysctl_max_map_count),
1408 .mode = 0644,
1409 .proc_handler = proc_dointvec_minmax,
1410 .extra1 = &zero,
1411 },
1412 #else
1413 {
1414 .procname = "nr_trim_pages",
1415 .data = &sysctl_nr_trim_pages,
1416 .maxlen = sizeof(sysctl_nr_trim_pages),
1417 .mode = 0644,
1418 .proc_handler = proc_dointvec_minmax,
1419 .extra1 = &zero,
1420 },
1421 #endif
1422 {
1423 .procname = "laptop_mode",
1424 .data = &laptop_mode,
1425 .maxlen = sizeof(laptop_mode),
1426 .mode = 0644,
1427 .proc_handler = proc_dointvec_jiffies,
1428 },
1429 {
1430 .procname = "block_dump",
1431 .data = &block_dump,
1432 .maxlen = sizeof(block_dump),
1433 .mode = 0644,
1434 .proc_handler = proc_dointvec,
1435 .extra1 = &zero,
1436 },
1437 {
1438 .procname = "vfs_cache_pressure",
1439 .data = &sysctl_vfs_cache_pressure,
1440 .maxlen = sizeof(sysctl_vfs_cache_pressure),
1441 .mode = 0644,
1442 .proc_handler = proc_dointvec,
1443 .extra1 = &zero,
1444 },
1445 #ifdef HAVE_ARCH_PICK_MMAP_LAYOUT
1446 {
1447 .procname = "legacy_va_layout",
1448 .data = &sysctl_legacy_va_layout,
1449 .maxlen = sizeof(sysctl_legacy_va_layout),
1450 .mode = 0644,
1451 .proc_handler = proc_dointvec,
1452 .extra1 = &zero,
1453 },
1454 #endif
1455 #ifdef CONFIG_NUMA
1456 {
1457 .procname = "zone_reclaim_mode",
1458 .data = &zone_reclaim_mode,
1459 .maxlen = sizeof(zone_reclaim_mode),
1460 .mode = 0644,
1461 .proc_handler = proc_dointvec,
1462 .extra1 = &zero,
1463 },
1464 {
1465 .procname = "min_unmapped_ratio",
1466 .data = &sysctl_min_unmapped_ratio,
1467 .maxlen = sizeof(sysctl_min_unmapped_ratio),
1468 .mode = 0644,
1469 .proc_handler = sysctl_min_unmapped_ratio_sysctl_handler,
1470 .extra1 = &zero,
1471 .extra2 = &one_hundred,
1472 },
1473 {
1474 .procname = "min_slab_ratio",
1475 .data = &sysctl_min_slab_ratio,
1476 .maxlen = sizeof(sysctl_min_slab_ratio),
1477 .mode = 0644,
1478 .proc_handler = sysctl_min_slab_ratio_sysctl_handler,
1479 .extra1 = &zero,
1480 .extra2 = &one_hundred,
1481 },
1482 #endif
1483 #ifdef CONFIG_SMP
1484 {
1485 .procname = "stat_interval",
1486 .data = &sysctl_stat_interval,
1487 .maxlen = sizeof(sysctl_stat_interval),
1488 .mode = 0644,
1489 .proc_handler = proc_dointvec_jiffies,
1490 },
1491 #endif
1492 #ifdef CONFIG_MMU
1493 {
1494 .procname = "mmap_min_addr",
1495 .data = &dac_mmap_min_addr,
1496 .maxlen = sizeof(unsigned long),
1497 .mode = 0644,
1498 .proc_handler = mmap_min_addr_handler,
1499 },
1500 #endif
1501 #ifdef CONFIG_NUMA
1502 {
1503 .procname = "numa_zonelist_order",
1504 .data = &numa_zonelist_order,
1505 .maxlen = NUMA_ZONELIST_ORDER_LEN,
1506 .mode = 0644,
1507 .proc_handler = numa_zonelist_order_handler,
1508 },
1509 #endif
1510 #if (defined(CONFIG_X86_32) && !defined(CONFIG_UML))|| \
1511 (defined(CONFIG_SUPERH) && defined(CONFIG_VSYSCALL))
1512 {
1513 .procname = "vdso_enabled",
1514 #ifdef CONFIG_X86_32
1515 .data = &vdso32_enabled,
1516 .maxlen = sizeof(vdso32_enabled),
1517 #else
1518 .data = &vdso_enabled,
1519 .maxlen = sizeof(vdso_enabled),
1520 #endif
1521 .mode = 0644,
1522 .proc_handler = proc_dointvec,
1523 .extra1 = &zero,
1524 },
1525 #endif
1526 #ifdef CONFIG_HIGHMEM
1527 {
1528 .procname = "highmem_is_dirtyable",
1529 .data = &vm_highmem_is_dirtyable,
1530 .maxlen = sizeof(vm_highmem_is_dirtyable),
1531 .mode = 0644,
1532 .proc_handler = proc_dointvec_minmax,
1533 .extra1 = &zero,
1534 .extra2 = &one,
1535 },
1536 #endif
1537 #ifdef CONFIG_MEMORY_FAILURE
1538 {
1539 .procname = "memory_failure_early_kill",
1540 .data = &sysctl_memory_failure_early_kill,
1541 .maxlen = sizeof(sysctl_memory_failure_early_kill),
1542 .mode = 0644,
1543 .proc_handler = proc_dointvec_minmax,
1544 .extra1 = &zero,
1545 .extra2 = &one,
1546 },
1547 {
1548 .procname = "memory_failure_recovery",
1549 .data = &sysctl_memory_failure_recovery,
1550 .maxlen = sizeof(sysctl_memory_failure_recovery),
1551 .mode = 0644,
1552 .proc_handler = proc_dointvec_minmax,
1553 .extra1 = &zero,
1554 .extra2 = &one,
1555 },
1556 #endif
1557 {
1558 .procname = "user_reserve_kbytes",
1559 .data = &sysctl_user_reserve_kbytes,
1560 .maxlen = sizeof(sysctl_user_reserve_kbytes),
1561 .mode = 0644,
1562 .proc_handler = proc_doulongvec_minmax,
1563 },
1564 {
1565 .procname = "admin_reserve_kbytes",
1566 .data = &sysctl_admin_reserve_kbytes,
1567 .maxlen = sizeof(sysctl_admin_reserve_kbytes),
1568 .mode = 0644,
1569 .proc_handler = proc_doulongvec_minmax,
1570 },
1571 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS
1572 {
1573 .procname = "mmap_rnd_bits",
1574 .data = &mmap_rnd_bits,
1575 .maxlen = sizeof(mmap_rnd_bits),
1576 .mode = 0600,
1577 .proc_handler = proc_dointvec_minmax,
1578 .extra1 = (void *)&mmap_rnd_bits_min,
1579 .extra2 = (void *)&mmap_rnd_bits_max,
1580 },
1581 #endif
1582 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
1583 {
1584 .procname = "mmap_rnd_compat_bits",
1585 .data = &mmap_rnd_compat_bits,
1586 .maxlen = sizeof(mmap_rnd_compat_bits),
1587 .mode = 0600,
1588 .proc_handler = proc_dointvec_minmax,
1589 .extra1 = (void *)&mmap_rnd_compat_bits_min,
1590 .extra2 = (void *)&mmap_rnd_compat_bits_max,
1591 },
1592 #endif
1593 { }
1594 };
1595
1596 static struct ctl_table fs_table[] = {
1597 {
1598 .procname = "inode-nr",
1599 .data = &inodes_stat,
1600 .maxlen = 2*sizeof(long),
1601 .mode = 0444,
1602 .proc_handler = proc_nr_inodes,
1603 },
1604 {
1605 .procname = "inode-state",
1606 .data = &inodes_stat,
1607 .maxlen = 7*sizeof(long),
1608 .mode = 0444,
1609 .proc_handler = proc_nr_inodes,
1610 },
1611 {
1612 .procname = "file-nr",
1613 .data = &files_stat,
1614 .maxlen = sizeof(files_stat),
1615 .mode = 0444,
1616 .proc_handler = proc_nr_files,
1617 },
1618 {
1619 .procname = "file-max",
1620 .data = &files_stat.max_files,
1621 .maxlen = sizeof(files_stat.max_files),
1622 .mode = 0644,
1623 .proc_handler = proc_doulongvec_minmax,
1624 },
1625 {
1626 .procname = "nr_open",
1627 .data = &sysctl_nr_open,
1628 .maxlen = sizeof(int),
1629 .mode = 0644,
1630 .proc_handler = proc_dointvec_minmax,
1631 .extra1 = &sysctl_nr_open_min,
1632 .extra2 = &sysctl_nr_open_max,
1633 },
1634 {
1635 .procname = "dentry-state",
1636 .data = &dentry_stat,
1637 .maxlen = 6*sizeof(long),
1638 .mode = 0444,
1639 .proc_handler = proc_nr_dentry,
1640 },
1641 {
1642 .procname = "overflowuid",
1643 .data = &fs_overflowuid,
1644 .maxlen = sizeof(int),
1645 .mode = 0644,
1646 .proc_handler = proc_dointvec_minmax,
1647 .extra1 = &minolduid,
1648 .extra2 = &maxolduid,
1649 },
1650 {
1651 .procname = "overflowgid",
1652 .data = &fs_overflowgid,
1653 .maxlen = sizeof(int),
1654 .mode = 0644,
1655 .proc_handler = proc_dointvec_minmax,
1656 .extra1 = &minolduid,
1657 .extra2 = &maxolduid,
1658 },
1659 #ifdef CONFIG_FILE_LOCKING
1660 {
1661 .procname = "leases-enable",
1662 .data = &leases_enable,
1663 .maxlen = sizeof(int),
1664 .mode = 0644,
1665 .proc_handler = proc_dointvec,
1666 },
1667 #endif
1668 #ifdef CONFIG_DNOTIFY
1669 {
1670 .procname = "dir-notify-enable",
1671 .data = &dir_notify_enable,
1672 .maxlen = sizeof(int),
1673 .mode = 0644,
1674 .proc_handler = proc_dointvec,
1675 },
1676 #endif
1677 #ifdef CONFIG_MMU
1678 #ifdef CONFIG_FILE_LOCKING
1679 {
1680 .procname = "lease-break-time",
1681 .data = &lease_break_time,
1682 .maxlen = sizeof(int),
1683 .mode = 0644,
1684 .proc_handler = proc_dointvec,
1685 },
1686 #endif
1687 #ifdef CONFIG_AIO
1688 {
1689 .procname = "aio-nr",
1690 .data = &aio_nr,
1691 .maxlen = sizeof(aio_nr),
1692 .mode = 0444,
1693 .proc_handler = proc_doulongvec_minmax,
1694 },
1695 {
1696 .procname = "aio-max-nr",
1697 .data = &aio_max_nr,
1698 .maxlen = sizeof(aio_max_nr),
1699 .mode = 0644,
1700 .proc_handler = proc_doulongvec_minmax,
1701 },
1702 #endif /* CONFIG_AIO */
1703 #ifdef CONFIG_INOTIFY_USER
1704 {
1705 .procname = "inotify",
1706 .mode = 0555,
1707 .child = inotify_table,
1708 },
1709 #endif
1710 #ifdef CONFIG_EPOLL
1711 {
1712 .procname = "epoll",
1713 .mode = 0555,
1714 .child = epoll_table,
1715 },
1716 #endif
1717 #endif
1718 {
1719 .procname = "protected_symlinks",
1720 .data = &sysctl_protected_symlinks,
1721 .maxlen = sizeof(int),
1722 .mode = 0600,
1723 .proc_handler = proc_dointvec_minmax,
1724 .extra1 = &zero,
1725 .extra2 = &one,
1726 },
1727 {
1728 .procname = "protected_hardlinks",
1729 .data = &sysctl_protected_hardlinks,
1730 .maxlen = sizeof(int),
1731 .mode = 0600,
1732 .proc_handler = proc_dointvec_minmax,
1733 .extra1 = &zero,
1734 .extra2 = &one,
1735 },
1736 {
1737 .procname = "suid_dumpable",
1738 .data = &suid_dumpable,
1739 .maxlen = sizeof(int),
1740 .mode = 0644,
1741 .proc_handler = proc_dointvec_minmax_coredump,
1742 .extra1 = &zero,
1743 .extra2 = &two,
1744 },
1745 #if defined(CONFIG_BINFMT_MISC) || defined(CONFIG_BINFMT_MISC_MODULE)
1746 {
1747 .procname = "binfmt_misc",
1748 .mode = 0555,
1749 .child = sysctl_mount_point,
1750 },
1751 #endif
1752 {
1753 .procname = "pipe-max-size",
1754 .data = &pipe_max_size,
1755 .maxlen = sizeof(int),
1756 .mode = 0644,
1757 .proc_handler = &pipe_proc_fn,
1758 .extra1 = &pipe_min_size,
1759 },
1760 { }
1761 };
1762
1763 static struct ctl_table debug_table[] = {
1764 #ifdef CONFIG_SYSCTL_EXCEPTION_TRACE
1765 {
1766 .procname = "exception-trace",
1767 .data = &show_unhandled_signals,
1768 .maxlen = sizeof(int),
1769 .mode = 0644,
1770 .proc_handler = proc_dointvec
1771 },
1772 #endif
1773 #if defined(CONFIG_OPTPROBES)
1774 {
1775 .procname = "kprobes-optimization",
1776 .data = &sysctl_kprobes_optimization,
1777 .maxlen = sizeof(int),
1778 .mode = 0644,
1779 .proc_handler = proc_kprobes_optimization_handler,
1780 .extra1 = &zero,
1781 .extra2 = &one,
1782 },
1783 #endif
1784 { }
1785 };
1786
1787 static struct ctl_table dev_table[] = {
1788 { }
1789 };
1790
1791 int __init sysctl_init(void)
1792 {
1793 struct ctl_table_header *hdr;
1794
1795 hdr = register_sysctl_table(sysctl_base_table);
1796 kmemleak_not_leak(hdr);
1797 return 0;
1798 }
1799
1800 #endif /* CONFIG_SYSCTL */
1801
1802 /*
1803 * /proc/sys support
1804 */
1805
1806 #ifdef CONFIG_PROC_SYSCTL
1807
1808 static int _proc_do_string(char *data, int maxlen, int write,
1809 char __user *buffer,
1810 size_t *lenp, loff_t *ppos)
1811 {
1812 size_t len;
1813 char __user *p;
1814 char c;
1815
1816 if (!data || !maxlen || !*lenp) {
1817 *lenp = 0;
1818 return 0;
1819 }
1820
1821 if (write) {
1822 if (sysctl_writes_strict == SYSCTL_WRITES_STRICT) {
1823 /* Only continue writes not past the end of buffer. */
1824 len = strlen(data);
1825 if (len > maxlen - 1)
1826 len = maxlen - 1;
1827
1828 if (*ppos > len)
1829 return 0;
1830 len = *ppos;
1831 } else {
1832 /* Start writing from beginning of buffer. */
1833 len = 0;
1834 }
1835
1836 *ppos += *lenp;
1837 p = buffer;
1838 while ((p - buffer) < *lenp && len < maxlen - 1) {
1839 if (get_user(c, p++))
1840 return -EFAULT;
1841 if (c == 0 || c == '\n')
1842 break;
1843 data[len++] = c;
1844 }
1845 data[len] = 0;
1846 } else {
1847 len = strlen(data);
1848 if (len > maxlen)
1849 len = maxlen;
1850
1851 if (*ppos > len) {
1852 *lenp = 0;
1853 return 0;
1854 }
1855
1856 data += *ppos;
1857 len -= *ppos;
1858
1859 if (len > *lenp)
1860 len = *lenp;
1861 if (len)
1862 if (copy_to_user(buffer, data, len))
1863 return -EFAULT;
1864 if (len < *lenp) {
1865 if (put_user('\n', buffer + len))
1866 return -EFAULT;
1867 len++;
1868 }
1869 *lenp = len;
1870 *ppos += len;
1871 }
1872 return 0;
1873 }
1874
1875 static void warn_sysctl_write(struct ctl_table *table)
1876 {
1877 pr_warn_once("%s wrote to %s when file position was not 0!\n"
1878 "This will not be supported in the future. To silence this\n"
1879 "warning, set kernel.sysctl_writes_strict = -1\n",
1880 current->comm, table->procname);
1881 }
1882
1883 /**
1884 * proc_dostring - read a string sysctl
1885 * @table: the sysctl table
1886 * @write: %TRUE if this is a write to the sysctl file
1887 * @buffer: the user buffer
1888 * @lenp: the size of the user buffer
1889 * @ppos: file position
1890 *
1891 * Reads/writes a string from/to the user buffer. If the kernel
1892 * buffer provided is not large enough to hold the string, the
1893 * string is truncated. The copied string is %NULL-terminated.
1894 * If the string is being read by the user process, it is copied
1895 * and a newline '\n' is added. It is truncated if the buffer is
1896 * not large enough.
1897 *
1898 * Returns 0 on success.
1899 */
1900 int proc_dostring(struct ctl_table *table, int write,
1901 void __user *buffer, size_t *lenp, loff_t *ppos)
1902 {
1903 if (write && *ppos && sysctl_writes_strict == SYSCTL_WRITES_WARN)
1904 warn_sysctl_write(table);
1905
1906 return _proc_do_string((char *)(table->data), table->maxlen, write,
1907 (char __user *)buffer, lenp, ppos);
1908 }
1909
1910 static size_t proc_skip_spaces(char **buf)
1911 {
1912 size_t ret;
1913 char *tmp = skip_spaces(*buf);
1914 ret = tmp - *buf;
1915 *buf = tmp;
1916 return ret;
1917 }
1918
1919 static void proc_skip_char(char **buf, size_t *size, const char v)
1920 {
1921 while (*size) {
1922 if (**buf != v)
1923 break;
1924 (*size)--;
1925 (*buf)++;
1926 }
1927 }
1928
1929 #define TMPBUFLEN 22
1930 /**
1931 * proc_get_long - reads an ASCII formatted integer from a user buffer
1932 *
1933 * @buf: a kernel buffer
1934 * @size: size of the kernel buffer
1935 * @val: this is where the number will be stored
1936 * @neg: set to %TRUE if number is negative
1937 * @perm_tr: a vector which contains the allowed trailers
1938 * @perm_tr_len: size of the perm_tr vector
1939 * @tr: pointer to store the trailer character
1940 *
1941 * In case of success %0 is returned and @buf and @size are updated with
1942 * the amount of bytes read. If @tr is non-NULL and a trailing
1943 * character exists (size is non-zero after returning from this
1944 * function), @tr is updated with the trailing character.
1945 */
1946 static int proc_get_long(char **buf, size_t *size,
1947 unsigned long *val, bool *neg,
1948 const char *perm_tr, unsigned perm_tr_len, char *tr)
1949 {
1950 int len;
1951 char *p, tmp[TMPBUFLEN];
1952
1953 if (!*size)
1954 return -EINVAL;
1955
1956 len = *size;
1957 if (len > TMPBUFLEN - 1)
1958 len = TMPBUFLEN - 1;
1959
1960 memcpy(tmp, *buf, len);
1961
1962 tmp[len] = 0;
1963 p = tmp;
1964 if (*p == '-' && *size > 1) {
1965 *neg = true;
1966 p++;
1967 } else
1968 *neg = false;
1969 if (!isdigit(*p))
1970 return -EINVAL;
1971
1972 *val = simple_strtoul(p, &p, 0);
1973
1974 len = p - tmp;
1975
1976 /* We don't know if the next char is whitespace thus we may accept
1977 * invalid integers (e.g. 1234...a) or two integers instead of one
1978 * (e.g. 123...1). So lets not allow such large numbers. */
1979 if (len == TMPBUFLEN - 1)
1980 return -EINVAL;
1981
1982 if (len < *size && perm_tr_len && !memchr(perm_tr, *p, perm_tr_len))
1983 return -EINVAL;
1984
1985 if (tr && (len < *size))
1986 *tr = *p;
1987
1988 *buf += len;
1989 *size -= len;
1990
1991 return 0;
1992 }
1993
1994 /**
1995 * proc_put_long - converts an integer to a decimal ASCII formatted string
1996 *
1997 * @buf: the user buffer
1998 * @size: the size of the user buffer
1999 * @val: the integer to be converted
2000 * @neg: sign of the number, %TRUE for negative
2001 *
2002 * In case of success %0 is returned and @buf and @size are updated with
2003 * the amount of bytes written.
2004 */
2005 static int proc_put_long(void __user **buf, size_t *size, unsigned long val,
2006 bool neg)
2007 {
2008 int len;
2009 char tmp[TMPBUFLEN], *p = tmp;
2010
2011 sprintf(p, "%s%lu", neg ? "-" : "", val);
2012 len = strlen(tmp);
2013 if (len > *size)
2014 len = *size;
2015 if (copy_to_user(*buf, tmp, len))
2016 return -EFAULT;
2017 *size -= len;
2018 *buf += len;
2019 return 0;
2020 }
2021 #undef TMPBUFLEN
2022
2023 static int proc_put_char(void __user **buf, size_t *size, char c)
2024 {
2025 if (*size) {
2026 char __user **buffer = (char __user **)buf;
2027 if (put_user(c, *buffer))
2028 return -EFAULT;
2029 (*size)--, (*buffer)++;
2030 *buf = *buffer;
2031 }
2032 return 0;
2033 }
2034
2035 static int do_proc_dointvec_conv(bool *negp, unsigned long *lvalp,
2036 int *valp,
2037 int write, void *data)
2038 {
2039 if (write) {
2040 if (*negp) {
2041 if (*lvalp > (unsigned long) INT_MAX + 1)
2042 return -EINVAL;
2043 *valp = -*lvalp;
2044 } else {
2045 if (*lvalp > (unsigned long) INT_MAX)
2046 return -EINVAL;
2047 *valp = *lvalp;
2048 }
2049 } else {
2050 int val = *valp;
2051 if (val < 0) {
2052 *negp = true;
2053 *lvalp = -(unsigned long)val;
2054 } else {
2055 *negp = false;
2056 *lvalp = (unsigned long)val;
2057 }
2058 }
2059 return 0;
2060 }
2061
2062 static const char proc_wspace_sep[] = { ' ', '\t', '\n' };
2063
2064 static int __do_proc_dointvec(void *tbl_data, struct ctl_table *table,
2065 int write, void __user *buffer,
2066 size_t *lenp, loff_t *ppos,
2067 int (*conv)(bool *negp, unsigned long *lvalp, int *valp,
2068 int write, void *data),
2069 void *data)
2070 {
2071 int *i, vleft, first = 1, err = 0;
2072 size_t left;
2073 char *kbuf = NULL, *p;
2074
2075 if (!tbl_data || !table->maxlen || !*lenp || (*ppos && !write)) {
2076 *lenp = 0;
2077 return 0;
2078 }
2079
2080 i = (int *) tbl_data;
2081 vleft = table->maxlen / sizeof(*i);
2082 left = *lenp;
2083
2084 if (!conv)
2085 conv = do_proc_dointvec_conv;
2086
2087 if (write) {
2088 if (*ppos) {
2089 switch (sysctl_writes_strict) {
2090 case SYSCTL_WRITES_STRICT:
2091 goto out;
2092 case SYSCTL_WRITES_WARN:
2093 warn_sysctl_write(table);
2094 break;
2095 default:
2096 break;
2097 }
2098 }
2099
2100 if (left > PAGE_SIZE - 1)
2101 left = PAGE_SIZE - 1;
2102 p = kbuf = memdup_user_nul(buffer, left);
2103 if (IS_ERR(kbuf))
2104 return PTR_ERR(kbuf);
2105 }
2106
2107 for (; left && vleft--; i++, first=0) {
2108 unsigned long lval;
2109 bool neg;
2110
2111 if (write) {
2112 left -= proc_skip_spaces(&p);
2113
2114 if (!left)
2115 break;
2116 err = proc_get_long(&p, &left, &lval, &neg,
2117 proc_wspace_sep,
2118 sizeof(proc_wspace_sep), NULL);
2119 if (err)
2120 break;
2121 if (conv(&neg, &lval, i, 1, data)) {
2122 err = -EINVAL;
2123 break;
2124 }
2125 } else {
2126 if (conv(&neg, &lval, i, 0, data)) {
2127 err = -EINVAL;
2128 break;
2129 }
2130 if (!first)
2131 err = proc_put_char(&buffer, &left, '\t');
2132 if (err)
2133 break;
2134 err = proc_put_long(&buffer, &left, lval, neg);
2135 if (err)
2136 break;
2137 }
2138 }
2139
2140 if (!write && !first && left && !err)
2141 err = proc_put_char(&buffer, &left, '\n');
2142 if (write && !err && left)
2143 left -= proc_skip_spaces(&p);
2144 if (write) {
2145 kfree(kbuf);
2146 if (first)
2147 return err ? : -EINVAL;
2148 }
2149 *lenp -= left;
2150 out:
2151 *ppos += *lenp;
2152 return err;
2153 }
2154
2155 static int do_proc_dointvec(struct ctl_table *table, int write,
2156 void __user *buffer, size_t *lenp, loff_t *ppos,
2157 int (*conv)(bool *negp, unsigned long *lvalp, int *valp,
2158 int write, void *data),
2159 void *data)
2160 {
2161 return __do_proc_dointvec(table->data, table, write,
2162 buffer, lenp, ppos, conv, data);
2163 }
2164
2165 /**
2166 * proc_dointvec - read a vector of integers
2167 * @table: the sysctl table
2168 * @write: %TRUE if this is a write to the sysctl file
2169 * @buffer: the user buffer
2170 * @lenp: the size of the user buffer
2171 * @ppos: file position
2172 *
2173 * Reads/writes up to table->maxlen/sizeof(unsigned int) integer
2174 * values from/to the user buffer, treated as an ASCII string.
2175 *
2176 * Returns 0 on success.
2177 */
2178 int proc_dointvec(struct ctl_table *table, int write,
2179 void __user *buffer, size_t *lenp, loff_t *ppos)
2180 {
2181 return do_proc_dointvec(table,write,buffer,lenp,ppos,
2182 NULL,NULL);
2183 }
2184
2185 /*
2186 * Taint values can only be increased
2187 * This means we can safely use a temporary.
2188 */
2189 static int proc_taint(struct ctl_table *table, int write,
2190 void __user *buffer, size_t *lenp, loff_t *ppos)
2191 {
2192 struct ctl_table t;
2193 unsigned long tmptaint = get_taint();
2194 int err;
2195
2196 if (write && !capable(CAP_SYS_ADMIN))
2197 return -EPERM;
2198
2199 t = *table;
2200 t.data = &tmptaint;
2201 err = proc_doulongvec_minmax(&t, write, buffer, lenp, ppos);
2202 if (err < 0)
2203 return err;
2204
2205 if (write) {
2206 /*
2207 * Poor man's atomic or. Not worth adding a primitive
2208 * to everyone's atomic.h for this
2209 */
2210 int i;
2211 for (i = 0; i < BITS_PER_LONG && tmptaint >> i; i++) {
2212 if ((tmptaint >> i) & 1)
2213 add_taint(i, LOCKDEP_STILL_OK);
2214 }
2215 }
2216
2217 return err;
2218 }
2219
2220 #ifdef CONFIG_PRINTK
2221 static int proc_dointvec_minmax_sysadmin(struct ctl_table *table, int write,
2222 void __user *buffer, size_t *lenp, loff_t *ppos)
2223 {
2224 if (write && !capable(CAP_SYS_ADMIN))
2225 return -EPERM;
2226
2227 return proc_dointvec_minmax(table, write, buffer, lenp, ppos);
2228 }
2229 #endif
2230
2231 struct do_proc_dointvec_minmax_conv_param {
2232 int *min;
2233 int *max;
2234 };
2235
2236 static int do_proc_dointvec_minmax_conv(bool *negp, unsigned long *lvalp,
2237 int *valp,
2238 int write, void *data)
2239 {
2240 struct do_proc_dointvec_minmax_conv_param *param = data;
2241 if (write) {
2242 int val = *negp ? -*lvalp : *lvalp;
2243 if ((param->min && *param->min > val) ||
2244 (param->max && *param->max < val))
2245 return -EINVAL;
2246 *valp = val;
2247 } else {
2248 int val = *valp;
2249 if (val < 0) {
2250 *negp = true;
2251 *lvalp = -(unsigned long)val;
2252 } else {
2253 *negp = false;
2254 *lvalp = (unsigned long)val;
2255 }
2256 }
2257 return 0;
2258 }
2259
2260 /**
2261 * proc_dointvec_minmax - read a vector of integers with min/max values
2262 * @table: the sysctl table
2263 * @write: %TRUE if this is a write to the sysctl file
2264 * @buffer: the user buffer
2265 * @lenp: the size of the user buffer
2266 * @ppos: file position
2267 *
2268 * Reads/writes up to table->maxlen/sizeof(unsigned int) integer
2269 * values from/to the user buffer, treated as an ASCII string.
2270 *
2271 * This routine will ensure the values are within the range specified by
2272 * table->extra1 (min) and table->extra2 (max).
2273 *
2274 * Returns 0 on success.
2275 */
2276 int proc_dointvec_minmax(struct ctl_table *table, int write,
2277 void __user *buffer, size_t *lenp, loff_t *ppos)
2278 {
2279 struct do_proc_dointvec_minmax_conv_param param = {
2280 .min = (int *) table->extra1,
2281 .max = (int *) table->extra2,
2282 };
2283 return do_proc_dointvec(table, write, buffer, lenp, ppos,
2284 do_proc_dointvec_minmax_conv, &param);
2285 }
2286
2287 static void validate_coredump_safety(void)
2288 {
2289 #ifdef CONFIG_COREDUMP
2290 if (suid_dumpable == SUID_DUMP_ROOT &&
2291 core_pattern[0] != '/' && core_pattern[0] != '|') {
2292 printk(KERN_WARNING "Unsafe core_pattern used with "\
2293 "suid_dumpable=2. Pipe handler or fully qualified "\
2294 "core dump path required.\n");
2295 }
2296 #endif
2297 }
2298
2299 static int proc_dointvec_minmax_coredump(struct ctl_table *table, int write,
2300 void __user *buffer, size_t *lenp, loff_t *ppos)
2301 {
2302 int error = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
2303 if (!error)
2304 validate_coredump_safety();
2305 return error;
2306 }
2307
2308 #ifdef CONFIG_COREDUMP
2309 static int proc_dostring_coredump(struct ctl_table *table, int write,
2310 void __user *buffer, size_t *lenp, loff_t *ppos)
2311 {
2312 int error = proc_dostring(table, write, buffer, lenp, ppos);
2313 if (!error)
2314 validate_coredump_safety();
2315 return error;
2316 }
2317 #endif
2318
2319 static int __do_proc_doulongvec_minmax(void *data, struct ctl_table *table, int write,
2320 void __user *buffer,
2321 size_t *lenp, loff_t *ppos,
2322 unsigned long convmul,
2323 unsigned long convdiv)
2324 {
2325 unsigned long *i, *min, *max;
2326 int vleft, first = 1, err = 0;
2327 size_t left;
2328 char *kbuf = NULL, *p;
2329
2330 if (!data || !table->maxlen || !*lenp || (*ppos && !write)) {
2331 *lenp = 0;
2332 return 0;
2333 }
2334
2335 i = (unsigned long *) data;
2336 min = (unsigned long *) table->extra1;
2337 max = (unsigned long *) table->extra2;
2338 vleft = table->maxlen / sizeof(unsigned long);
2339 left = *lenp;
2340
2341 if (write) {
2342 if (*ppos) {
2343 switch (sysctl_writes_strict) {
2344 case SYSCTL_WRITES_STRICT:
2345 goto out;
2346 case SYSCTL_WRITES_WARN:
2347 warn_sysctl_write(table);
2348 break;
2349 default:
2350 break;
2351 }
2352 }
2353
2354 if (left > PAGE_SIZE - 1)
2355 left = PAGE_SIZE - 1;
2356 p = kbuf = memdup_user_nul(buffer, left);
2357 if (IS_ERR(kbuf))
2358 return PTR_ERR(kbuf);
2359 }
2360
2361 for (; left && vleft--; i++, first = 0) {
2362 unsigned long val;
2363
2364 if (write) {
2365 bool neg;
2366
2367 left -= proc_skip_spaces(&p);
2368
2369 err = proc_get_long(&p, &left, &val, &neg,
2370 proc_wspace_sep,
2371 sizeof(proc_wspace_sep), NULL);
2372 if (err)
2373 break;
2374 if (neg)
2375 continue;
2376 if ((min && val < *min) || (max && val > *max))
2377 continue;
2378 *i = val;
2379 } else {
2380 val = convdiv * (*i) / convmul;
2381 if (!first) {
2382 err = proc_put_char(&buffer, &left, '\t');
2383 if (err)
2384 break;
2385 }
2386 err = proc_put_long(&buffer, &left, val, false);
2387 if (err)
2388 break;
2389 }
2390 }
2391
2392 if (!write && !first && left && !err)
2393 err = proc_put_char(&buffer, &left, '\n');
2394 if (write && !err)
2395 left -= proc_skip_spaces(&p);
2396 if (write) {
2397 kfree(kbuf);
2398 if (first)
2399 return err ? : -EINVAL;
2400 }
2401 *lenp -= left;
2402 out:
2403 *ppos += *lenp;
2404 return err;
2405 }
2406
2407 static int do_proc_doulongvec_minmax(struct ctl_table *table, int write,
2408 void __user *buffer,
2409 size_t *lenp, loff_t *ppos,
2410 unsigned long convmul,
2411 unsigned long convdiv)
2412 {
2413 return __do_proc_doulongvec_minmax(table->data, table, write,
2414 buffer, lenp, ppos, convmul, convdiv);
2415 }
2416
2417 /**
2418 * proc_doulongvec_minmax - read a vector of long integers with min/max values
2419 * @table: the sysctl table
2420 * @write: %TRUE if this is a write to the sysctl file
2421 * @buffer: the user buffer
2422 * @lenp: the size of the user buffer
2423 * @ppos: file position
2424 *
2425 * Reads/writes up to table->maxlen/sizeof(unsigned long) unsigned long
2426 * values from/to the user buffer, treated as an ASCII string.
2427 *
2428 * This routine will ensure the values are within the range specified by
2429 * table->extra1 (min) and table->extra2 (max).
2430 *
2431 * Returns 0 on success.
2432 */
2433 int proc_doulongvec_minmax(struct ctl_table *table, int write,
2434 void __user *buffer, size_t *lenp, loff_t *ppos)
2435 {
2436 return do_proc_doulongvec_minmax(table, write, buffer, lenp, ppos, 1l, 1l);
2437 }
2438
2439 /**
2440 * proc_doulongvec_ms_jiffies_minmax - read a vector of millisecond values with min/max values
2441 * @table: the sysctl table
2442 * @write: %TRUE if this is a write to the sysctl file
2443 * @buffer: the user buffer
2444 * @lenp: the size of the user buffer
2445 * @ppos: file position
2446 *
2447 * Reads/writes up to table->maxlen/sizeof(unsigned long) unsigned long
2448 * values from/to the user buffer, treated as an ASCII string. The values
2449 * are treated as milliseconds, and converted to jiffies when they are stored.
2450 *
2451 * This routine will ensure the values are within the range specified by
2452 * table->extra1 (min) and table->extra2 (max).
2453 *
2454 * Returns 0 on success.
2455 */
2456 int proc_doulongvec_ms_jiffies_minmax(struct ctl_table *table, int write,
2457 void __user *buffer,
2458 size_t *lenp, loff_t *ppos)
2459 {
2460 return do_proc_doulongvec_minmax(table, write, buffer,
2461 lenp, ppos, HZ, 1000l);
2462 }
2463
2464
2465 static int do_proc_dointvec_jiffies_conv(bool *negp, unsigned long *lvalp,
2466 int *valp,
2467 int write, void *data)
2468 {
2469 if (write) {
2470 if (*lvalp > LONG_MAX / HZ)
2471 return 1;
2472 *valp = *negp ? -(*lvalp*HZ) : (*lvalp*HZ);
2473 } else {
2474 int val = *valp;
2475 unsigned long lval;
2476 if (val < 0) {
2477 *negp = true;
2478 lval = -(unsigned long)val;
2479 } else {
2480 *negp = false;
2481 lval = (unsigned long)val;
2482 }
2483 *lvalp = lval / HZ;
2484 }
2485 return 0;
2486 }
2487
2488 static int do_proc_dointvec_userhz_jiffies_conv(bool *negp, unsigned long *lvalp,
2489 int *valp,
2490 int write, void *data)
2491 {
2492 if (write) {
2493 if (USER_HZ < HZ && *lvalp > (LONG_MAX / HZ) * USER_HZ)
2494 return 1;
2495 *valp = clock_t_to_jiffies(*negp ? -*lvalp : *lvalp);
2496 } else {
2497 int val = *valp;
2498 unsigned long lval;
2499 if (val < 0) {
2500 *negp = true;
2501 lval = -(unsigned long)val;
2502 } else {
2503 *negp = false;
2504 lval = (unsigned long)val;
2505 }
2506 *lvalp = jiffies_to_clock_t(lval);
2507 }
2508 return 0;
2509 }
2510
2511 static int do_proc_dointvec_ms_jiffies_conv(bool *negp, unsigned long *lvalp,
2512 int *valp,
2513 int write, void *data)
2514 {
2515 if (write) {
2516 unsigned long jif = msecs_to_jiffies(*negp ? -*lvalp : *lvalp);
2517
2518 if (jif > INT_MAX)
2519 return 1;
2520 *valp = (int)jif;
2521 } else {
2522 int val = *valp;
2523 unsigned long lval;
2524 if (val < 0) {
2525 *negp = true;
2526 lval = -(unsigned long)val;
2527 } else {
2528 *negp = false;
2529 lval = (unsigned long)val;
2530 }
2531 *lvalp = jiffies_to_msecs(lval);
2532 }
2533 return 0;
2534 }
2535
2536 /**
2537 * proc_dointvec_jiffies - read a vector of integers as seconds
2538 * @table: the sysctl table
2539 * @write: %TRUE if this is a write to the sysctl file
2540 * @buffer: the user buffer
2541 * @lenp: the size of the user buffer
2542 * @ppos: file position
2543 *
2544 * Reads/writes up to table->maxlen/sizeof(unsigned int) integer
2545 * values from/to the user buffer, treated as an ASCII string.
2546 * The values read are assumed to be in seconds, and are converted into
2547 * jiffies.
2548 *
2549 * Returns 0 on success.
2550 */
2551 int proc_dointvec_jiffies(struct ctl_table *table, int write,
2552 void __user *buffer, size_t *lenp, loff_t *ppos)
2553 {
2554 return do_proc_dointvec(table,write,buffer,lenp,ppos,
2555 do_proc_dointvec_jiffies_conv,NULL);
2556 }
2557
2558 /**
2559 * proc_dointvec_userhz_jiffies - read a vector of integers as 1/USER_HZ seconds
2560 * @table: the sysctl table
2561 * @write: %TRUE if this is a write to the sysctl file
2562 * @buffer: the user buffer
2563 * @lenp: the size of the user buffer
2564 * @ppos: pointer to the file position
2565 *
2566 * Reads/writes up to table->maxlen/sizeof(unsigned int) integer
2567 * values from/to the user buffer, treated as an ASCII string.
2568 * The values read are assumed to be in 1/USER_HZ seconds, and
2569 * are converted into jiffies.
2570 *
2571 * Returns 0 on success.
2572 */
2573 int proc_dointvec_userhz_jiffies(struct ctl_table *table, int write,
2574 void __user *buffer, size_t *lenp, loff_t *ppos)
2575 {
2576 return do_proc_dointvec(table,write,buffer,lenp,ppos,
2577 do_proc_dointvec_userhz_jiffies_conv,NULL);
2578 }
2579
2580 /**
2581 * proc_dointvec_ms_jiffies - read a vector of integers as 1 milliseconds
2582 * @table: the sysctl table
2583 * @write: %TRUE if this is a write to the sysctl file
2584 * @buffer: the user buffer
2585 * @lenp: the size of the user buffer
2586 * @ppos: file position
2587 * @ppos: the current position in the file
2588 *
2589 * Reads/writes up to table->maxlen/sizeof(unsigned int) integer
2590 * values from/to the user buffer, treated as an ASCII string.
2591 * The values read are assumed to be in 1/1000 seconds, and
2592 * are converted into jiffies.
2593 *
2594 * Returns 0 on success.
2595 */
2596 int proc_dointvec_ms_jiffies(struct ctl_table *table, int write,
2597 void __user *buffer, size_t *lenp, loff_t *ppos)
2598 {
2599 return do_proc_dointvec(table, write, buffer, lenp, ppos,
2600 do_proc_dointvec_ms_jiffies_conv, NULL);
2601 }
2602
2603 static int proc_do_cad_pid(struct ctl_table *table, int write,
2604 void __user *buffer, size_t *lenp, loff_t *ppos)
2605 {
2606 struct pid *new_pid;
2607 pid_t tmp;
2608 int r;
2609
2610 tmp = pid_vnr(cad_pid);
2611
2612 r = __do_proc_dointvec(&tmp, table, write, buffer,
2613 lenp, ppos, NULL, NULL);
2614 if (r || !write)
2615 return r;
2616
2617 new_pid = find_get_pid(tmp);
2618 if (!new_pid)
2619 return -ESRCH;
2620
2621 put_pid(xchg(&cad_pid, new_pid));
2622 return 0;
2623 }
2624
2625 /**
2626 * proc_do_large_bitmap - read/write from/to a large bitmap
2627 * @table: the sysctl table
2628 * @write: %TRUE if this is a write to the sysctl file
2629 * @buffer: the user buffer
2630 * @lenp: the size of the user buffer
2631 * @ppos: file position
2632 *
2633 * The bitmap is stored at table->data and the bitmap length (in bits)
2634 * in table->maxlen.
2635 *
2636 * We use a range comma separated format (e.g. 1,3-4,10-10) so that
2637 * large bitmaps may be represented in a compact manner. Writing into
2638 * the file will clear the bitmap then update it with the given input.
2639 *
2640 * Returns 0 on success.
2641 */
2642 int proc_do_large_bitmap(struct ctl_table *table, int write,
2643 void __user *buffer, size_t *lenp, loff_t *ppos)
2644 {
2645 int err = 0;
2646 bool first = 1;
2647 size_t left = *lenp;
2648 unsigned long bitmap_len = table->maxlen;
2649 unsigned long *bitmap = *(unsigned long **) table->data;
2650 unsigned long *tmp_bitmap = NULL;
2651 char tr_a[] = { '-', ',', '\n' }, tr_b[] = { ',', '\n', 0 }, c;
2652
2653 if (!bitmap || !bitmap_len || !left || (*ppos && !write)) {
2654 *lenp = 0;
2655 return 0;
2656 }
2657
2658 if (write) {
2659 char *kbuf, *p;
2660
2661 if (left > PAGE_SIZE - 1)
2662 left = PAGE_SIZE - 1;
2663
2664 p = kbuf = memdup_user_nul(buffer, left);
2665 if (IS_ERR(kbuf))
2666 return PTR_ERR(kbuf);
2667
2668 tmp_bitmap = kzalloc(BITS_TO_LONGS(bitmap_len) * sizeof(unsigned long),
2669 GFP_KERNEL);
2670 if (!tmp_bitmap) {
2671 kfree(kbuf);
2672 return -ENOMEM;
2673 }
2674 proc_skip_char(&p, &left, '\n');
2675 while (!err && left) {
2676 unsigned long val_a, val_b;
2677 bool neg;
2678
2679 err = proc_get_long(&p, &left, &val_a, &neg, tr_a,
2680 sizeof(tr_a), &c);
2681 if (err)
2682 break;
2683 if (val_a >= bitmap_len || neg) {
2684 err = -EINVAL;
2685 break;
2686 }
2687
2688 val_b = val_a;
2689 if (left) {
2690 p++;
2691 left--;
2692 }
2693
2694 if (c == '-') {
2695 err = proc_get_long(&p, &left, &val_b,
2696 &neg, tr_b, sizeof(tr_b),
2697 &c);
2698 if (err)
2699 break;
2700 if (val_b >= bitmap_len || neg ||
2701 val_a > val_b) {
2702 err = -EINVAL;
2703 break;
2704 }
2705 if (left) {
2706 p++;
2707 left--;
2708 }
2709 }
2710
2711 bitmap_set(tmp_bitmap, val_a, val_b - val_a + 1);
2712 first = 0;
2713 proc_skip_char(&p, &left, '\n');
2714 }
2715 kfree(kbuf);
2716 } else {
2717 unsigned long bit_a, bit_b = 0;
2718
2719 while (left) {
2720 bit_a = find_next_bit(bitmap, bitmap_len, bit_b);
2721 if (bit_a >= bitmap_len)
2722 break;
2723 bit_b = find_next_zero_bit(bitmap, bitmap_len,
2724 bit_a + 1) - 1;
2725
2726 if (!first) {
2727 err = proc_put_char(&buffer, &left, ',');
2728 if (err)
2729 break;
2730 }
2731 err = proc_put_long(&buffer, &left, bit_a, false);
2732 if (err)
2733 break;
2734 if (bit_a != bit_b) {
2735 err = proc_put_char(&buffer, &left, '-');
2736 if (err)
2737 break;
2738 err = proc_put_long(&buffer, &left, bit_b, false);
2739 if (err)
2740 break;
2741 }
2742
2743 first = 0; bit_b++;
2744 }
2745 if (!err)
2746 err = proc_put_char(&buffer, &left, '\n');
2747 }
2748
2749 if (!err) {
2750 if (write) {
2751 if (*ppos)
2752 bitmap_or(bitmap, bitmap, tmp_bitmap, bitmap_len);
2753 else
2754 bitmap_copy(bitmap, tmp_bitmap, bitmap_len);
2755 }
2756 kfree(tmp_bitmap);
2757 *lenp -= left;
2758 *ppos += *lenp;
2759 return 0;
2760 } else {
2761 kfree(tmp_bitmap);
2762 return err;
2763 }
2764 }
2765
2766 #else /* CONFIG_PROC_SYSCTL */
2767
2768 int proc_dostring(struct ctl_table *table, int write,
2769 void __user *buffer, size_t *lenp, loff_t *ppos)
2770 {
2771 return -ENOSYS;
2772 }
2773
2774 int proc_dointvec(struct ctl_table *table, int write,
2775 void __user *buffer, size_t *lenp, loff_t *ppos)
2776 {
2777 return -ENOSYS;
2778 }
2779
2780 int proc_dointvec_minmax(struct ctl_table *table, int write,
2781 void __user *buffer, size_t *lenp, loff_t *ppos)
2782 {
2783 return -ENOSYS;
2784 }
2785
2786 int proc_dointvec_jiffies(struct ctl_table *table, int write,
2787 void __user *buffer, size_t *lenp, loff_t *ppos)
2788 {
2789 return -ENOSYS;
2790 }
2791
2792 int proc_dointvec_userhz_jiffies(struct ctl_table *table, int write,
2793 void __user *buffer, size_t *lenp, loff_t *ppos)
2794 {
2795 return -ENOSYS;
2796 }
2797
2798 int proc_dointvec_ms_jiffies(struct ctl_table *table, int write,
2799 void __user *buffer, size_t *lenp, loff_t *ppos)
2800 {
2801 return -ENOSYS;
2802 }
2803
2804 int proc_doulongvec_minmax(struct ctl_table *table, int write,
2805 void __user *buffer, size_t *lenp, loff_t *ppos)
2806 {
2807 return -ENOSYS;
2808 }
2809
2810 int proc_doulongvec_ms_jiffies_minmax(struct ctl_table *table, int write,
2811 void __user *buffer,
2812 size_t *lenp, loff_t *ppos)
2813 {
2814 return -ENOSYS;
2815 }
2816
2817
2818 #endif /* CONFIG_PROC_SYSCTL */
2819
2820 /*
2821 * No sense putting this after each symbol definition, twice,
2822 * exception granted :-)
2823 */
2824 EXPORT_SYMBOL(proc_dointvec);
2825 EXPORT_SYMBOL(proc_dointvec_jiffies);
2826 EXPORT_SYMBOL(proc_dointvec_minmax);
2827 EXPORT_SYMBOL(proc_dointvec_userhz_jiffies);
2828 EXPORT_SYMBOL(proc_dointvec_ms_jiffies);
2829 EXPORT_SYMBOL(proc_dostring);
2830 EXPORT_SYMBOL(proc_doulongvec_minmax);
2831 EXPORT_SYMBOL(proc_doulongvec_ms_jiffies_minmax);
Linux with added FPC-III support