Home | History | Annotate | Line # | Download | only in systat
bufcache.c revision 1.18
      1 /*	$NetBSD: bufcache.c,v 1.18 2004/11/23 03:11:33 simonb 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  * 3. All advertising materials mentioning features or use of this software
     19  *    must display the following acknowledgement:
     20  *        This product includes software developed by the NetBSD
     21  *        Foundation, Inc. and its contributors.
     22  * 4. Neither the name of The NetBSD Foundation nor the names of its
     23  *    contributors may be used to endorse or promote products derived
     24  *    from this software without specific prior written permission.
     25  *
     26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     36  * POSSIBILITY OF SUCH DAMAGE.
     37  */
     38 
     39 #include <sys/cdefs.h>
     40 #ifndef lint
     41 __RCSID("$NetBSD: bufcache.c,v 1.18 2004/11/23 03:11:33 simonb Exp $");
     42 #endif /* not lint */
     43 
     44 #include <sys/param.h>
     45 #include <sys/buf.h>
     46 #include <sys/mount.h>
     47 #include <sys/sysctl.h>
     48 #include <sys/vnode.h>
     49 
     50 #include <uvm/uvm_extern.h>
     51 
     52 #include <err.h>
     53 #include <errno.h>
     54 #include <inttypes.h>
     55 #include <kvm.h>
     56 #include <math.h>
     57 #include <nlist.h>
     58 #include <stdlib.h>
     59 #include <string.h>
     60 #include <unistd.h>
     61 
     62 #include <miscfs/specfs/specdev.h>
     63 
     64 #include "systat.h"
     65 #include "extern.h"
     66 
     67 #define VCACHE_SIZE	50
     68 #define	PAGEINFO_ROWS	 5
     69 
     70 struct vcache {
     71 	int vc_age;
     72 	struct vnode *vc_addr;
     73 	struct vnode vc_node;
     74 };
     75 
     76 struct ml_entry {
     77 	u_int ml_count;
     78 	u_long ml_size;
     79 	u_long ml_valid;
     80 	struct mount *ml_addr;
     81 	LIST_ENTRY(ml_entry) ml_entries;
     82 	struct mount ml_mount;
     83 };
     84 
     85 static struct nlist namelist[] = {
     86 #define	X_BUFMEM	0
     87 	{ "_bufmem" },
     88 	{ "" },
     89 };
     90 
     91 static struct vcache vcache[VCACHE_SIZE];
     92 static LIST_HEAD(mount_list, ml_entry) mount_list;
     93 
     94 static u_long bufmem;
     95 static u_int nbuf, pgwidth, kbwidth;
     96 static struct uvmexp_sysctl uvmexp;
     97 
     98 static void	vc_init(void);
     99 static void	ml_init(void);
    100 static struct 	vnode *vc_lookup(struct vnode *);
    101 static struct 	mount *ml_lookup(struct mount *, int, int);
    102 static void	fetchuvmexp(void);
    103 
    104 
    105 WINDOW *
    106 openbufcache(void)
    107 {
    108 
    109 	return (subwin(stdscr, -1, 0, 5, 0));
    110 }
    111 
    112 void
    113 closebufcache(WINDOW *w)
    114 {
    115 
    116 	if (w == NULL)
    117 		return;
    118 	wclear(w);
    119 	wrefresh(w);
    120 	delwin(w);
    121 	ml_init();		/* Clear out mount list */
    122 }
    123 
    124 void
    125 labelbufcache(void)
    126 {
    127 	int i;
    128 
    129 	for (i = 0; i <= PAGEINFO_ROWS; i++) {
    130 		wmove(wnd, i, 0);
    131 		wclrtoeol(wnd);
    132 	}
    133 	mvwaddstr(wnd, PAGEINFO_ROWS + 1, 0, "File System          Bufs used"
    134 	    "   %   kB in use   %  Bufsize kB   %  Util %");
    135 	wclrtoeol(wnd);
    136 }
    137 
    138 void
    139 showbufcache(void)
    140 {
    141 	int tbuf, i, lastrow;
    142 	double tvalid, tsize;
    143 	struct ml_entry *ml;
    144 
    145 	NREAD(X_BUFMEM, &bufmem, sizeof(bufmem));
    146 
    147 	mvwprintw(wnd, 0, 0,
    148 	    "   %*d metadata buffers using             %*ld kBytes of "
    149 	    "memory (%2.0f%%).",
    150 	    pgwidth, nbuf, kbwidth, bufmem / 1024,
    151 	    ((bufmem * 100.0) + 0.5) / getpagesize() / uvmexp.npages);
    152 	wclrtoeol(wnd);
    153 	mvwprintw(wnd, 1, 0,
    154 	    "   %*" PRIu64 " pages for cached file data using   %*"
    155 	    PRIu64 " kBytes of memory (%2.0f%%).",
    156 	    pgwidth, uvmexp.filepages,
    157 	    kbwidth, uvmexp.filepages * getpagesize() / 1024,
    158 	    (uvmexp.filepages * 100 + 0.5) / uvmexp.npages);
    159 	wclrtoeol(wnd);
    160 	mvwprintw(wnd, 2, 0,
    161 	    "   %*" PRIu64 " pages for executables using        %*"
    162 	    PRIu64 " kBytes of memory (%2.0f%%).",
    163 	    pgwidth, uvmexp.execpages,
    164 	    kbwidth, uvmexp.execpages * getpagesize() / 1024,
    165 	    (uvmexp.execpages * 100 + 0.5) / uvmexp.npages);
    166 	wclrtoeol(wnd);
    167 	mvwprintw(wnd, 3, 0,
    168 	    "   %*" PRIu64 " pages for anon (non-file) data     %*"
    169 	    PRIu64 " kBytes of memory (%2.0f%%).",
    170 	    pgwidth, uvmexp.anonpages,
    171 	    kbwidth, uvmexp.anonpages * getpagesize() / 1024,
    172 	    (uvmexp.anonpages * 100 + 0.5) / uvmexp.npages);
    173 	wclrtoeol(wnd);
    174 	mvwprintw(wnd, 4, 0,
    175 	    "   %*" PRIu64 " free pages                         %*"
    176 	    PRIu64 " kBytes of memory (%2.0f%%).",
    177 	    pgwidth, uvmexp.free,
    178 	    kbwidth, uvmexp.free * getpagesize() / 1024,
    179 	    (uvmexp.free * 100 + 0.5) / uvmexp.npages);
    180 	wclrtoeol(wnd);
    181 
    182 	if (nbuf == 0 || bufmem == 0) {
    183 		wclrtobot(wnd);
    184 		return;
    185 	}
    186 
    187 	tbuf = 0;
    188 	tvalid = tsize = 0;
    189 	lastrow = PAGEINFO_ROWS + 2;	/* Leave room for header. */
    190 	for (i = lastrow, ml = LIST_FIRST(&mount_list); ml != NULL;
    191 	    i++, ml = LIST_NEXT(ml, ml_entries)) {
    192 
    193 		int c = ml->ml_count;
    194 		double v = ml->ml_valid;
    195 		double s = ml->ml_size;
    196 
    197 		/* Display in window if enough room. */
    198 		if (i < getmaxy(wnd) - 2) {
    199 			mvwprintw(wnd, i, 0, "%-20.20s", ml->ml_addr == NULL ?
    200 			    "NULL" : ml->ml_mount.mnt_stat.f_mntonname);
    201 			wprintw(wnd,
    202 			    "    %6d %3d    %8ld %3.0f    %8ld %3.0f     %3.0f",
    203 			    c, (100 * c) / nbuf,
    204 			    (long)(v/1024), 100 * v / bufmem,
    205 			    (long)(s/1024), 100 * s / bufmem,
    206 			    100 * v / s);
    207 			wclrtoeol(wnd);
    208 			lastrow = i;
    209 		}
    210 
    211 		/* Update statistics. */
    212 		tbuf += c;
    213 		tvalid += v;
    214 		tsize += s;
    215 	}
    216 
    217 	wclrtobot(wnd);
    218 	mvwprintw(wnd, lastrow + 2, 0,
    219 	    "%-20s    %6d %3d    %8ld %3.0f    %8ld %3.0f     %3.0f",
    220 	    "Total:", tbuf, (100 * tbuf) / nbuf,
    221 	    (long)(tvalid/1024), 100 * tvalid / bufmem,
    222 	    (long)(tsize/1024), 100 * tsize / bufmem,
    223 	    tsize != 0 ? ((100 * tvalid) / tsize) : 0);
    224 }
    225 
    226 int
    227 initbufcache(void)
    228 {
    229 	if (namelist[0].n_type == 0) {
    230 		if (kvm_nlist(kd, namelist)) {
    231 			nlisterr(namelist);
    232 			return(0);
    233 		}
    234 	}
    235 
    236 	fetchuvmexp();
    237 	pgwidth = (int)(floor(log10((double)uvmexp.npages)) + 1);
    238 	kbwidth = (int)(floor(log10(uvmexp.npages * getpagesize() / 1024.0)) +
    239 	    1);
    240 
    241 	return(1);
    242 }
    243 
    244 static void
    245 fetchuvmexp(void)
    246 {
    247 	int mib[2];
    248 	size_t size;
    249 
    250 	/* Re-read pages used for vnodes & executables */
    251 	size = sizeof(uvmexp);
    252 	mib[0] = CTL_VM;
    253 	mib[1] = VM_UVMEXP2;
    254 	if (sysctl(mib, 2, &uvmexp, &size, NULL, 0) < 0) {
    255 		error("can't get uvmexp: %s\n", strerror(errno));
    256 		memset(&uvmexp, 0, sizeof(uvmexp));
    257 	}
    258 }
    259 
    260 void
    261 fetchbufcache(void)
    262 {
    263 	int count;
    264 	struct buf_sysctl *bp, *buffers;
    265 	struct vnode *vn;
    266 	struct mount *mt;
    267 	struct ml_entry *ml;
    268 	int mib[6];
    269 	size_t size;
    270 	int extraslop = 0;
    271 
    272 	/* Re-read pages used for vnodes & executables */
    273 	fetchuvmexp();
    274 
    275 	/* Initialise vnode cache and mount list. */
    276 	vc_init();
    277 	ml_init();
    278 
    279 	/* Get metadata buffers */
    280 	size = 0;
    281 	buffers = NULL;
    282 	mib[0] = CTL_KERN;
    283 	mib[1] = KERN_BUF;
    284 	mib[2] = KERN_BUF_ALL;
    285 	mib[3] = KERN_BUF_ALL;
    286 	mib[4] = (int)sizeof(struct buf_sysctl);
    287 	mib[5] = INT_MAX; /* we want them all */
    288 again:
    289 	if (sysctl(mib, 6, NULL, &size, NULL, 0) < 0) {
    290 		error("can't get buffers size: %s\n", strerror(errno));
    291 		return;
    292 	}
    293 	if (size == 0)
    294 		return;
    295 
    296 	size += extraslop * sizeof(struct buf_sysctl);
    297 	buffers = malloc(size);
    298 	if (buffers == NULL) {
    299 		error("can't allocate buffers: %s\n", strerror(errno));
    300 		return;
    301 	}
    302 	if (sysctl(mib, 6, buffers, &size, NULL, 0) < 0) {
    303 		free(buffers);
    304 		if (extraslop == 0) {
    305 			extraslop = 100;
    306 			goto again;
    307 		}
    308 		error("can't get buffers: %s\n", strerror(errno));
    309 		return;
    310 	}
    311 
    312 	nbuf = size / sizeof(struct buf_sysctl);
    313 	for (bp = buffers; bp < buffers + nbuf; bp++) {
    314 		if (UINT64TOPTR(bp->b_vp) != NULL) {
    315 			struct mount *mp;
    316 			vn = vc_lookup(UINT64TOPTR(bp->b_vp));
    317 			if (vn == NULL)
    318 				break;
    319 
    320 			mp = vn->v_mount;
    321 			/*
    322 			 * References to mounted-on vnodes should be
    323 			 * counted towards the mounted filesystem.
    324 			 */
    325 			if (vn->v_type == VBLK && vn->v_specinfo != NULL) {
    326 				struct specinfo sp;
    327 				if (!KREAD(vn->v_specinfo, &sp, sizeof(sp)))
    328 					continue;
    329 				if (sp.si_mountpoint)
    330 					mp = sp.si_mountpoint;
    331 			}
    332 			if (mp != NULL)
    333 				mt = ml_lookup(mp,
    334 				    bp->b_bufsize,
    335 				    bp->b_bcount);
    336 		}
    337 	}
    338 
    339 	/* simple sort - there's not that many entries */
    340 	do {
    341 		if ((ml = LIST_FIRST(&mount_list)) == NULL ||
    342 		    LIST_NEXT(ml, ml_entries) == NULL)
    343 			break;
    344 
    345 		count = 0;
    346 		for (ml = LIST_FIRST(&mount_list); ml != NULL;
    347 		    ml = LIST_NEXT(ml, ml_entries)) {
    348 			if (LIST_NEXT(ml, ml_entries) == NULL)
    349 				break;
    350 			if (ml->ml_count < LIST_NEXT(ml, ml_entries)->ml_count) {
    351 				ml = LIST_NEXT(ml, ml_entries);
    352 				LIST_REMOVE(ml, ml_entries);
    353 				LIST_INSERT_HEAD(&mount_list, ml, ml_entries);
    354 				count++;
    355 			}
    356 		}
    357 	} while (count != 0);
    358 
    359 	free(buffers);
    360 }
    361 
    362 static void
    363 vc_init(void)
    364 {
    365 	int i;
    366 
    367 	/* vc_addr == NULL for unused cache entry. */
    368 	for (i = 0; i < VCACHE_SIZE; i++)
    369 		vcache[i].vc_addr = NULL;
    370 }
    371 
    372 static void
    373 ml_init(void)
    374 {
    375 	struct ml_entry *ml;
    376 
    377 	/* Throw out the current mount list and start again. */
    378 	while ((ml = LIST_FIRST(&mount_list)) != NULL) {
    379 		LIST_REMOVE(ml, ml_entries);
    380 		free(ml);
    381 	}
    382 }
    383 
    384 
    385 static struct vnode *
    386 vc_lookup(struct vnode *vaddr)
    387 {
    388 	struct vnode *ret;
    389 	size_t i, oldest;
    390 
    391 	ret = NULL;
    392 	oldest = 0;
    393 	for (i = 0; i < VCACHE_SIZE; i++) {
    394 		if (vcache[i].vc_addr == NULL)
    395 			break;
    396 		vcache[i].vc_age++;
    397 		if (vcache[i].vc_age < vcache[oldest].vc_age)
    398 			oldest = i;
    399 		if (vcache[i].vc_addr == vaddr) {
    400 			vcache[i].vc_age = 0;
    401 			ret = &vcache[i].vc_node;
    402 		}
    403 	}
    404 
    405 	/* Find an entry in the cache? */
    406 	if (ret != NULL)
    407 		return(ret);
    408 
    409 	/* Go past the end of the cache? */
    410 	if  (i >= VCACHE_SIZE)
    411 		i = oldest;
    412 
    413 	/* Read in new vnode and reset age counter. */
    414 	if (KREAD(vaddr, &vcache[i].vc_node, sizeof(struct vnode)) == 0)
    415 		return NULL;
    416 	vcache[i].vc_addr = vaddr;
    417 	vcache[i].vc_age = 0;
    418 
    419 	return(&vcache[i].vc_node);
    420 }
    421 
    422 static struct mount *
    423 ml_lookup(struct mount *maddr, int size, int valid)
    424 {
    425 	struct ml_entry *ml;
    426 
    427 	for (ml = LIST_FIRST(&mount_list); ml != NULL;
    428 	    ml = LIST_NEXT(ml, ml_entries))
    429 		if (ml->ml_addr == maddr) {
    430 			ml->ml_count++;
    431 			ml->ml_size += size;
    432 			ml->ml_valid += valid;
    433 			if (ml->ml_addr == NULL)
    434 				return(NULL);
    435 			else
    436 				return(&ml->ml_mount);
    437 		}
    438 
    439 	if ((ml = malloc(sizeof(struct ml_entry))) == NULL) {
    440 		error("out of memory");
    441 		die(0);
    442 	}
    443 	LIST_INSERT_HEAD(&mount_list, ml, ml_entries);
    444 	ml->ml_count = 1;
    445 	ml->ml_size = size;
    446 	ml->ml_valid = valid;
    447 	ml->ml_addr = maddr;
    448 	if (maddr == NULL)
    449 		return(NULL);
    450 
    451 	KREAD(maddr, &ml->ml_mount, sizeof(struct mount));
    452 	return(&ml->ml_mount);
    453 }
    454