xref: /freebsd/sys/kern/subr_param.c (revision 7fdf597e96a02165cfe22ff357b857d5fa15ed8a)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1980, 1986, 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  */
36 
37 #include <sys/cdefs.h>
38 #include "opt_param.h"
39 #include "opt_msgbuf.h"
40 #include "opt_maxphys.h"
41 #include "opt_maxusers.h"
42 
43 #include <sys/param.h>
44 #include <sys/_maxphys.h>
45 #include <sys/systm.h>
46 #include <sys/buf.h>
47 #include <sys/kernel.h>
48 #include <sys/limits.h>
49 #include <sys/msgbuf.h>
50 #include <sys/sysctl.h>
51 #include <sys/proc.h>
52 #include <sys/vnode.h>
53 
54 #include <vm/vm.h>
55 #include <vm/vm_param.h>
56 #include <vm/pmap.h>
57 
58 /*
59  * System parameter formulae.
60  */
61 
62 #ifndef HZ
63 #  define	HZ 1000
64 #  ifndef HZ_VM
65 #    define	HZ_VM 100
66 #  endif
67 #else
68 #  ifndef HZ_VM
69 #    define	HZ_VM HZ
70 #  endif
71 #endif
72 #define	NPROC (20 + 16 * maxusers)
73 #ifndef NBUF
74 #define NBUF 0
75 #endif
76 #ifndef MAXFILES
77 #define	MAXFILES (40 + 32 * maxusers)
78 #endif
79 
80 static int sysctl_kern_vm_guest(SYSCTL_HANDLER_ARGS);
81 
82 int	hz;				/* system clock's frequency */
83 int	tick;				/* usec per tick (1000000 / hz) */
84 time_t	tick_seconds_max;		/* max hz * seconds an integer can hold */
85 struct bintime tick_bt;			/* bintime per tick (1s / hz) */
86 sbintime_t tick_sbt;
87 int	maxusers;			/* base tunable */
88 int	maxproc;			/* maximum # of processes */
89 int	maxprocperuid;			/* max # of procs per user */
90 int	maxfiles;			/* sys. wide open files limit */
91 int	maxfilesperproc;		/* per-proc open files limit */
92 int	msgbufsize;			/* size of kernel message buffer */
93 int	nbuf;				/* number of bcache bufs */
94 int	bio_transient_maxcnt;
95 int	ngroups_max;			/* max # groups per process */
96 int	nswbuf;
97 pid_t	pid_max = PID_MAX;
98 u_long	maxswzone;			/* max swmeta KVA storage */
99 u_long	maxbcache;			/* max buffer cache KVA storage */
100 u_long	maxpipekva;			/* Limit on pipe KVA */
101 u_long	maxphys;			/* max raw I/O transfer size */
102 int	vm_guest = VM_GUEST_NO;		/* Running as virtual machine guest? */
103 u_long	maxtsiz;			/* max text size */
104 u_long	dfldsiz;			/* initial data size limit */
105 u_long	maxdsiz;			/* max data size */
106 u_long	dflssiz;			/* initial stack size limit */
107 u_long	maxssiz;			/* max stack size */
108 u_long	sgrowsiz;			/* amount to grow stack */
109 
110 SYSCTL_INT(_kern, OID_AUTO, hz, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &hz, 0,
111     "Number of clock ticks per second");
112 SYSCTL_INT(_kern, OID_AUTO, hz_max, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, HZ_MAXIMUM,
113     "Maximum hz value supported");
114 SYSCTL_INT(_kern, OID_AUTO, hz_min, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, HZ_MINIMUM,
115     "Minimum hz value supported");
116 SYSCTL_INT(_kern, OID_AUTO, nbuf, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &nbuf, 0,
117     "Number of buffers in the buffer cache");
118 SYSCTL_INT(_kern, OID_AUTO, nswbuf, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &nswbuf, 0,
119     "Number of swap buffers");
120 SYSCTL_INT(_kern, OID_AUTO, msgbufsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &msgbufsize, 0,
121     "Size of the kernel message buffer");
122 SYSCTL_LONG(_kern, OID_AUTO, maxswzone, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &maxswzone, 0,
123     "Maximum memory for swap metadata");
124 SYSCTL_LONG(_kern, OID_AUTO, maxbcache, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &maxbcache, 0,
125     "Maximum value of vfs.maxbufspace");
126 SYSCTL_INT(_kern, OID_AUTO, bio_transient_maxcnt, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
127     &bio_transient_maxcnt, 0,
128     "Maximum number of transient BIOs mappings");
129 SYSCTL_ULONG(_kern, OID_AUTO, maxtsiz, CTLFLAG_RWTUN | CTLFLAG_NOFETCH, &maxtsiz, 0,
130     "Maximum text size");
131 SYSCTL_ULONG(_kern, OID_AUTO, dfldsiz, CTLFLAG_RWTUN | CTLFLAG_NOFETCH, &dfldsiz, 0,
132     "Initial data size limit");
133 SYSCTL_ULONG(_kern, OID_AUTO, maxdsiz, CTLFLAG_RWTUN | CTLFLAG_NOFETCH, &maxdsiz, 0,
134     "Maximum data size");
135 SYSCTL_ULONG(_kern, OID_AUTO, dflssiz, CTLFLAG_RWTUN | CTLFLAG_NOFETCH, &dflssiz, 0,
136     "Initial stack size limit");
137 SYSCTL_ULONG(_kern, OID_AUTO, maxssiz, CTLFLAG_RWTUN | CTLFLAG_NOFETCH, &maxssiz, 0,
138     "Maximum stack size");
139 SYSCTL_ULONG(_kern, OID_AUTO, sgrowsiz, CTLFLAG_RWTUN | CTLFLAG_NOFETCH, &sgrowsiz, 0,
140     "Amount to grow stack on a stack fault");
141 SYSCTL_PROC(_kern, OID_AUTO, vm_guest,
142     CTLFLAG_RD | CTLTYPE_STRING | CTLFLAG_MPSAFE, NULL, 0,
143     sysctl_kern_vm_guest, "A",
144     "Virtual machine guest detected?");
145 
146 /*
147  * The elements of this array are ordered based upon the values of the
148  * corresponding enum VM_GUEST members.
149  */
150 static const char *const vm_guest_sysctl_names[] = {
151 	[VM_GUEST_NO] = "none",
152 	[VM_GUEST_VM] = "generic",
153 	[VM_GUEST_XEN] = "xen",
154 	[VM_GUEST_HV] = "hv",
155 	[VM_GUEST_VMWARE] = "vmware",
156 	[VM_GUEST_KVM] = "kvm",
157 	[VM_GUEST_BHYVE] = "bhyve",
158 	[VM_GUEST_VBOX] = "vbox",
159 	[VM_GUEST_PARALLELS] = "parallels",
160 	[VM_GUEST_NVMM] = "nvmm",
161 };
162 _Static_assert(nitems(vm_guest_sysctl_names) == VM_GUEST_LAST,
163     "new vm guest type not added to vm_guest_sysctl_names");
164 
165 /*
166  * Boot time overrides that are not scaled against main memory
167  */
168 void
169 init_param1(void)
170 {
171 
172 	TSENTER();
173 
174 	/*
175 	 * arm64 and riscv currently hard-code the thread0 kstack size
176 	 * to KSTACK_PAGES, ignoring the tunable.
177 	 */
178 	TUNABLE_INT_FETCH("kern.kstack_pages", &kstack_pages);
179 
180 	hz = -1;
181 	TUNABLE_INT_FETCH("kern.hz", &hz);
182 	if (hz == -1)
183 		hz = vm_guest > VM_GUEST_NO ? HZ_VM : HZ;
184 
185 	/* range check the "hz" value */
186 	if (__predict_false(hz < HZ_MINIMUM))
187 		hz = HZ_MINIMUM;
188 	else if (__predict_false(hz > HZ_MAXIMUM))
189 		hz = HZ_MAXIMUM;
190 
191 	tick = 1000000 / hz;
192 	tick_sbt = SBT_1S / hz;
193 	tick_bt = sbttobt(tick_sbt);
194 	tick_seconds_max = INT_MAX / hz;
195 
196 	/*
197 	 * Arrange for ticks to wrap 10 minutes after boot to help catch
198 	 * sign problems sooner.
199 	 */
200 	ticks = INT_MAX - (hz * 10 * 60);
201 
202 	vn_lock_pair_pause_max = hz / 100;
203 	if (vn_lock_pair_pause_max == 0)
204 		vn_lock_pair_pause_max = 1;
205 
206 #ifdef VM_SWZONE_SIZE_MAX
207 	maxswzone = VM_SWZONE_SIZE_MAX;
208 #endif
209 	TUNABLE_LONG_FETCH("kern.maxswzone", &maxswzone);
210 #ifdef VM_BCACHE_SIZE_MAX
211 	maxbcache = VM_BCACHE_SIZE_MAX;
212 #endif
213 	TUNABLE_LONG_FETCH("kern.maxbcache", &maxbcache);
214 	msgbufsize = MSGBUF_SIZE;
215 	TUNABLE_INT_FETCH("kern.msgbufsize", &msgbufsize);
216 
217 	maxtsiz = MAXTSIZ;
218 	TUNABLE_ULONG_FETCH("kern.maxtsiz", &maxtsiz);
219 	dfldsiz = DFLDSIZ;
220 	TUNABLE_ULONG_FETCH("kern.dfldsiz", &dfldsiz);
221 	maxdsiz = MAXDSIZ;
222 	TUNABLE_ULONG_FETCH("kern.maxdsiz", &maxdsiz);
223 	dflssiz = DFLSSIZ;
224 	TUNABLE_ULONG_FETCH("kern.dflssiz", &dflssiz);
225 	maxssiz = MAXSSIZ;
226 	TUNABLE_ULONG_FETCH("kern.maxssiz", &maxssiz);
227 	sgrowsiz = SGROWSIZ;
228 	TUNABLE_ULONG_FETCH("kern.sgrowsiz", &sgrowsiz);
229 
230 	/*
231 	 * Let the administrator set {NGROUPS_MAX}.
232 	 *
233 	 * Values less than NGROUPS_MAX would violate POSIX/SuS (see the
234 	 * specification for <limits.h>, paragraph "Runtime Increasable
235 	 * Values").
236 	 *
237 	 * On the other hand, INT_MAX would result in an overflow for the common
238 	 * 'ngroups_max + 1' computation (to obtain the size of the internal
239 	 * groups array, its first element being reserved for the effective
240 	 * GID).  Also, the number of allocated bytes for the group array must
241 	 * not overflow on 32-bit machines.  For all these reasons, we limit the
242 	 * number of supplementary groups to some very high number that we
243 	 * expect will never be reached in all practical uses and ensures we
244 	 * avoid the problems just exposed, even if 'gid_t' was to be enlarged
245 	 * by a magnitude.
246 	 */
247 	ngroups_max = NGROUPS_MAX;
248 	TUNABLE_INT_FETCH("kern.ngroups", &ngroups_max);
249 	if (ngroups_max < NGROUPS_MAX)
250 		ngroups_max = NGROUPS_MAX;
251 	else {
252 		const int ngroups_max_max = (1 << 24) - 1;
253 
254 		if (ngroups_max > ngroups_max_max)
255 			ngroups_max = ngroups_max_max;
256 	}
257 
258 	/*
259 	 * Only allow to lower the maximal pid.
260 	 * Prevent setting up a non-bootable system if pid_max is too low.
261 	 */
262 	TUNABLE_INT_FETCH("kern.pid_max", &pid_max);
263 	if (pid_max > PID_MAX)
264 		pid_max = PID_MAX;
265 	else if (pid_max < 300)
266 		pid_max = 300;
267 
268 	TUNABLE_INT_FETCH("vfs.unmapped_buf_allowed", &unmapped_buf_allowed);
269 	TSEXIT();
270 }
271 
272 /*
273  * Boot time overrides that are scaled against main memory
274  */
275 void
276 init_param2(long physpages)
277 {
278 
279 	TSENTER();
280 	/* Base parameters */
281 	maxusers = MAXUSERS;
282 	TUNABLE_INT_FETCH("kern.maxusers", &maxusers);
283 	if (maxusers == 0) {
284 		maxusers = physpages / (2 * 1024 * 1024 / PAGE_SIZE);
285 		if (maxusers < 32)
286 			maxusers = 32;
287 #ifdef VM_MAX_AUTOTUNE_MAXUSERS
288                 if (maxusers > VM_MAX_AUTOTUNE_MAXUSERS)
289                         maxusers = VM_MAX_AUTOTUNE_MAXUSERS;
290 #endif
291                 /*
292                  * Scales down the function in which maxusers grows once
293                  * we hit 384.
294                  */
295                 if (maxusers > 384)
296                         maxusers = 384 + ((maxusers - 384) / 8);
297         }
298 
299 	/*
300 	 * The following can be overridden after boot via sysctl.  Note:
301 	 * unless overridden, these macros are ultimately based on maxusers.
302 	 * Limit maxproc so that kmap entries cannot be exhausted by
303 	 * processes.
304 	 */
305 	maxproc = NPROC;
306 	TUNABLE_INT_FETCH("kern.maxproc", &maxproc);
307 	if (maxproc > (physpages / 12))
308 		maxproc = physpages / 12;
309 	if (maxproc > pid_max)
310 		maxproc = pid_max;
311 	maxprocperuid = (maxproc * 9) / 10;
312 
313 	/*
314 	 * The default limit for maxfiles is 1/12 of the number of
315 	 * physical page but not less than 16 times maxusers.
316 	 * At most it can be 1/6 the number of physical pages.
317 	 */
318 	maxfiles = imax(MAXFILES, physpages / 8);
319 	TUNABLE_INT_FETCH("kern.maxfiles", &maxfiles);
320 	if (maxfiles > (physpages / 4))
321 		maxfiles = physpages / 4;
322 	maxfilesperproc = (maxfiles / 10) * 9;
323 	TUNABLE_INT_FETCH("kern.maxfilesperproc", &maxfilesperproc);
324 
325 	/*
326 	 * Cannot be changed after boot.
327 	 */
328 	nbuf = NBUF;
329 	TUNABLE_INT_FETCH("kern.nbuf", &nbuf);
330 	TUNABLE_INT_FETCH("kern.bio_transient_maxcnt", &bio_transient_maxcnt);
331 	maxphys = MAXPHYS;
332 	TUNABLE_ULONG_FETCH("kern.maxphys", &maxphys);
333 	if (maxphys == 0) {
334 		maxphys = MAXPHYS;
335 	} else if (__bitcountl(maxphys) != 1) {	/* power of two */
336 		if (flsl(maxphys) == NBBY * sizeof(maxphys))
337 			maxphys = MAXPHYS;
338 		else
339 			maxphys = 1UL << flsl(maxphys);
340 	}
341 	if (maxphys < PAGE_SIZE)
342 		maxphys = MAXPHYS;
343 
344 	/*
345 	 * Physical buffers are pre-allocated buffers (struct buf) that
346 	 * are used as temporary holders for I/O, such as paging I/O.
347 	 */
348 	TUNABLE_INT_FETCH("kern.nswbuf", &nswbuf);
349 
350 	/*
351 	 * The default for maxpipekva is min(1/64 of the kernel address space,
352 	 * max(1/64 of main memory, 512KB)).  See sys_pipe.c for more details.
353 	 */
354 	maxpipekva = ptoa(physpages / 64);
355 	TUNABLE_LONG_FETCH("kern.ipc.maxpipekva", &maxpipekva);
356 	if (maxpipekva < 512 * 1024)
357 		maxpipekva = 512 * 1024;
358 	if (maxpipekva > (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) / 64)
359 		maxpipekva = (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) /
360 		    64;
361 	TSEXIT();
362 }
363 
364 /*
365  * Sysctl stringifying handler for kern.vm_guest.
366  */
367 static int
368 sysctl_kern_vm_guest(SYSCTL_HANDLER_ARGS)
369 {
370 	return (SYSCTL_OUT_STR(req, vm_guest_sysctl_names[vm_guest]));
371 }
372