bufcache.c revision 1.26 1 /* $NetBSD: bufcache.c,v 1.26 2013/10/18 20:47:07 christos Exp $ */
2
3 /*-
4 * Copyright (c) 1999 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Simon Burge.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29 * POSSIBILITY OF SUCH DAMAGE.
30 */
31
32 #include <sys/cdefs.h>
33 #ifndef lint
34 __RCSID("$NetBSD: bufcache.c,v 1.26 2013/10/18 20:47:07 christos Exp $");
35 #endif /* not lint */
36
37 #include <sys/param.h>
38 #include <sys/buf.h>
39 #include <sys/mount.h>
40 #include <sys/sysctl.h>
41 #include <sys/vnode.h>
42
43 #include <uvm/uvm_extern.h>
44
45 #include <err.h>
46 #include <errno.h>
47 #include <inttypes.h>
48 #include <math.h>
49 #include <stdlib.h>
50 #include <string.h>
51 #include <unistd.h>
52 #include <stdbool.h>
53
54 #include <miscfs/specfs/specdev.h>
55
56 #include "systat.h"
57 #include "extern.h"
58
59 #define VCACHE_SIZE 50
60 #define PAGEINFO_ROWS 5
61
62 struct vcache {
63 int vc_age;
64 struct vnode *vc_addr;
65 struct vnode vc_node;
66 };
67
68 struct ml_entry {
69 u_int ml_count;
70 u_long ml_size;
71 u_long ml_valid;
72 struct mount *ml_addr;
73 LIST_ENTRY(ml_entry) ml_entries;
74 struct mount ml_mount;
75 };
76
77 static struct vcache vcache[VCACHE_SIZE];
78 static LIST_HEAD(mount_list, ml_entry) mount_list;
79
80 static uint64_t bufmem;
81 static u_int nbuf, pgwidth, kbwidth;
82 static struct uvmexp_sysctl uvmexp;
83
84 static void vc_init(void);
85 static void ml_init(void);
86 static struct vnode *vc_lookup(struct vnode *);
87 static struct mount *ml_lookup(struct mount *, int, int);
88 static void fetchuvmexp(void);
89
90
91 WINDOW *
92 openbufcache(void)
93 {
94
95 return (subwin(stdscr, -1, 0, 5, 0));
96 }
97
98 void
99 closebufcache(WINDOW *w)
100 {
101
102 if (w == NULL)
103 return;
104 wclear(w);
105 wrefresh(w);
106 delwin(w);
107 ml_init(); /* Clear out mount list */
108 }
109
110 void
111 labelbufcache(void)
112 {
113 int i;
114
115 for (i = 0; i <= PAGEINFO_ROWS; i++) {
116 wmove(wnd, i, 0);
117 wclrtoeol(wnd);
118 }
119 mvwaddstr(wnd, PAGEINFO_ROWS + 1, 0, "File System Bufs used"
120 " % kB in use % Bufsize kB % Util %");
121 wclrtoeol(wnd);
122 }
123
124 void
125 showbufcache(void)
126 {
127 int tbuf, i, lastrow;
128 double tvalid, tsize;
129 struct ml_entry *ml;
130 size_t len;
131
132 len = sizeof(bufmem);
133 if (sysctlbyname("vm.bufmem", &bufmem, &len, NULL, 0))
134 error("can't get \"vm.bufmmem\": %s", strerror(errno));
135
136 mvwprintw(wnd, 0, 0,
137 " %*d metadata buffers using %*"PRIu64" kBytes of "
138 "memory (%2.0f%%).",
139 pgwidth, nbuf, kbwidth, bufmem / 1024,
140 ((bufmem * 100.0) + 0.5) / getpagesize() / uvmexp.npages);
141 wclrtoeol(wnd);
142 mvwprintw(wnd, 1, 0,
143 " %*" PRIu64 " pages for cached file data using %*"
144 PRIu64 " kBytes of memory (%2.0f%%).",
145 pgwidth, uvmexp.filepages,
146 kbwidth, uvmexp.filepages * getpagesize() / 1024,
147 (uvmexp.filepages * 100 + 0.5) / uvmexp.npages);
148 wclrtoeol(wnd);
149 mvwprintw(wnd, 2, 0,
150 " %*" PRIu64 " pages for executables using %*"
151 PRIu64 " kBytes of memory (%2.0f%%).",
152 pgwidth, uvmexp.execpages,
153 kbwidth, uvmexp.execpages * getpagesize() / 1024,
154 (uvmexp.execpages * 100 + 0.5) / uvmexp.npages);
155 wclrtoeol(wnd);
156 mvwprintw(wnd, 3, 0,
157 " %*" PRIu64 " pages for anon (non-file) data %*"
158 PRIu64 " kBytes of memory (%2.0f%%).",
159 pgwidth, uvmexp.anonpages,
160 kbwidth, uvmexp.anonpages * getpagesize() / 1024,
161 (uvmexp.anonpages * 100 + 0.5) / uvmexp.npages);
162 wclrtoeol(wnd);
163 mvwprintw(wnd, 4, 0,
164 " %*" PRIu64 " free pages %*"
165 PRIu64 " kBytes of memory (%2.0f%%).",
166 pgwidth, uvmexp.free,
167 kbwidth, uvmexp.free * getpagesize() / 1024,
168 (uvmexp.free * 100 + 0.5) / uvmexp.npages);
169 wclrtoeol(wnd);
170
171 if (nbuf == 0 || bufmem == 0) {
172 wclrtobot(wnd);
173 return;
174 }
175
176 tbuf = 0;
177 tvalid = tsize = 0;
178 lastrow = PAGEINFO_ROWS + 2; /* Leave room for header. */
179 for (i = lastrow, ml = LIST_FIRST(&mount_list); ml != NULL;
180 i++, ml = LIST_NEXT(ml, ml_entries)) {
181
182 int cnt = ml->ml_count;
183 double v = ml->ml_valid;
184 double s = ml->ml_size;
185
186 /* Display in window if enough room. */
187 if (i < getmaxy(wnd) - 2) {
188 mvwprintw(wnd, i, 0, "%-20.20s", ml->ml_addr == NULL ?
189 "NULL" : ml->ml_mount.mnt_stat.f_mntonname);
190 wprintw(wnd,
191 " %6d %3d %8ld %3.0f %8ld %3.0f %3.0f",
192 cnt, (100 * cnt) / nbuf,
193 (long)(v/1024), 100 * v / bufmem,
194 (long)(s/1024), 100 * s / bufmem,
195 100 * v / s);
196 wclrtoeol(wnd);
197 lastrow = i;
198 }
199
200 /* Update statistics. */
201 tbuf += cnt;
202 tvalid += v;
203 tsize += s;
204 }
205
206 wclrtobot(wnd);
207 mvwprintw(wnd, lastrow + 2, 0,
208 "%-20s %6d %3d %8ld %3.0f %8ld %3.0f %3.0f",
209 "Total:", tbuf, (100 * tbuf) / nbuf,
210 (long)(tvalid/1024), 100 * tvalid / bufmem,
211 (long)(tsize/1024), 100 * tsize / bufmem,
212 tsize != 0 ? ((100 * tvalid) / tsize) : 0);
213 }
214
215 int
216 initbufcache(void)
217 {
218 fetchuvmexp();
219 pgwidth = (int)(floor(log10((double)uvmexp.npages)) + 1);
220 kbwidth = (int)(floor(log10(uvmexp.npages * getpagesize() / 1024.0)) +
221 1);
222
223 return(1);
224 }
225
226 static void
227 fetchuvmexp(void)
228 {
229 int mib[2];
230 size_t size;
231
232 /* Re-read pages used for vnodes & executables */
233 size = sizeof(uvmexp);
234 mib[0] = CTL_VM;
235 mib[1] = VM_UVMEXP2;
236 if (sysctl(mib, 2, &uvmexp, &size, NULL, 0) < 0) {
237 error("can't get uvmexp: %s\n", strerror(errno));
238 memset(&uvmexp, 0, sizeof(uvmexp));
239 }
240 }
241
242 void
243 fetchbufcache(void)
244 {
245 int count;
246 struct buf_sysctl *bp, *buffers;
247 struct vnode *vn;
248 struct ml_entry *ml;
249 int mib[6];
250 size_t size;
251 int extraslop = 0;
252
253 /* Re-read pages used for vnodes & executables */
254 fetchuvmexp();
255
256 /* Initialise vnode cache and mount list. */
257 vc_init();
258 ml_init();
259
260 /* Get metadata buffers */
261 size = 0;
262 buffers = NULL;
263 mib[0] = CTL_KERN;
264 mib[1] = KERN_BUF;
265 mib[2] = KERN_BUF_ALL;
266 mib[3] = KERN_BUF_ALL;
267 mib[4] = (int)sizeof(struct buf_sysctl);
268 mib[5] = INT_MAX; /* we want them all */
269 again:
270 if (sysctl(mib, 6, NULL, &size, NULL, 0) < 0) {
271 error("can't get buffers size: %s\n", strerror(errno));
272 return;
273 }
274 if (size == 0)
275 return;
276
277 size += extraslop * sizeof(struct buf_sysctl);
278 buffers = malloc(size);
279 if (buffers == NULL) {
280 error("can't allocate buffers: %s\n", strerror(errno));
281 return;
282 }
283 if (sysctl(mib, 6, buffers, &size, NULL, 0) < 0) {
284 free(buffers);
285 if (extraslop == 0) {
286 extraslop = 100;
287 goto again;
288 }
289 error("can't get buffers: %s\n", strerror(errno));
290 return;
291 }
292
293 nbuf = size / sizeof(struct buf_sysctl);
294 for (bp = buffers; bp < buffers + nbuf; bp++) {
295 if (UINT64TOPTR(bp->b_vp) != NULL) {
296 struct mount *mp;
297 vn = vc_lookup(UINT64TOPTR(bp->b_vp));
298 if (vn == NULL)
299 break;
300
301 mp = vn->v_mount;
302 /*
303 * References to mounted-on vnodes should be
304 * counted towards the mounted filesystem.
305 */
306 if (vn->v_type == VBLK && vn->v_specnode != NULL) {
307 specnode_t sn;
308 specdev_t sd;
309 if (!KREAD(vn->v_specnode, &sn, sizeof(sn)))
310 continue;
311 if (!KREAD(sn.sn_dev, &sd, sizeof(sd)))
312 continue;
313 if (sd.sd_mountpoint)
314 mp = sd.sd_mountpoint;
315 }
316 if (mp != NULL)
317 (void)ml_lookup(mp, bp->b_bufsize,
318 bp->b_bcount);
319 }
320 }
321
322 /* simple sort - there's not that many entries */
323 do {
324 if ((ml = LIST_FIRST(&mount_list)) == NULL ||
325 LIST_NEXT(ml, ml_entries) == NULL)
326 break;
327
328 count = 0;
329 for (ml = LIST_FIRST(&mount_list); ml != NULL;
330 ml = LIST_NEXT(ml, ml_entries)) {
331 if (LIST_NEXT(ml, ml_entries) == NULL)
332 break;
333 if (ml->ml_count < LIST_NEXT(ml, ml_entries)->ml_count) {
334 ml = LIST_NEXT(ml, ml_entries);
335 LIST_REMOVE(ml, ml_entries);
336 LIST_INSERT_HEAD(&mount_list, ml, ml_entries);
337 count++;
338 }
339 }
340 } while (count != 0);
341
342 free(buffers);
343 }
344
345 static void
346 vc_init(void)
347 {
348 int i;
349
350 /* vc_addr == NULL for unused cache entry. */
351 for (i = 0; i < VCACHE_SIZE; i++)
352 vcache[i].vc_addr = NULL;
353 }
354
355 static void
356 ml_init(void)
357 {
358 struct ml_entry *ml;
359
360 /* Throw out the current mount list and start again. */
361 while ((ml = LIST_FIRST(&mount_list)) != NULL) {
362 LIST_REMOVE(ml, ml_entries);
363 free(ml);
364 }
365 }
366
367
368 static struct vnode *
369 vc_lookup(struct vnode *vaddr)
370 {
371 struct vnode *ret;
372 size_t i, oldest;
373
374 ret = NULL;
375 oldest = 0;
376 for (i = 0; i < VCACHE_SIZE; i++) {
377 if (vcache[i].vc_addr == NULL)
378 break;
379 vcache[i].vc_age++;
380 if (vcache[i].vc_age < vcache[oldest].vc_age)
381 oldest = i;
382 if (vcache[i].vc_addr == vaddr) {
383 vcache[i].vc_age = 0;
384 ret = &vcache[i].vc_node;
385 }
386 }
387
388 /* Find an entry in the cache? */
389 if (ret != NULL)
390 return(ret);
391
392 /* Go past the end of the cache? */
393 if (i >= VCACHE_SIZE)
394 i = oldest;
395
396 /* Read in new vnode and reset age counter. */
397 if (KREAD(vaddr, &vcache[i].vc_node, sizeof(struct vnode)) == 0)
398 return NULL;
399 vcache[i].vc_addr = vaddr;
400 vcache[i].vc_age = 0;
401
402 return(&vcache[i].vc_node);
403 }
404
405 static struct mount *
406 ml_lookup(struct mount *maddr, int size, int valid)
407 {
408 struct ml_entry *ml;
409
410 for (ml = LIST_FIRST(&mount_list); ml != NULL;
411 ml = LIST_NEXT(ml, ml_entries))
412 if (ml->ml_addr == maddr) {
413 ml->ml_count++;
414 ml->ml_size += size;
415 ml->ml_valid += valid;
416 if (ml->ml_addr == NULL)
417 return(NULL);
418 else
419 return(&ml->ml_mount);
420 }
421
422 if ((ml = malloc(sizeof(struct ml_entry))) == NULL) {
423 error("out of memory");
424 die(0);
425 }
426 LIST_INSERT_HEAD(&mount_list, ml, ml_entries);
427 ml->ml_count = 1;
428 ml->ml_size = size;
429 ml->ml_valid = valid;
430 ml->ml_addr = maddr;
431 if (maddr == NULL)
432 return(NULL);
433
434 KREAD(maddr, &ml->ml_mount, sizeof(struct mount));
435 return(&ml->ml_mount);
436 }
437