98f3db135bcf513a9252995bbeccd1be04bc5866
[dragonfly.git] / lib / libkvm / kvm_getswapinfo.c
1 /*
2  * Copyright (c) 1999 The DragonFly Project.  All rights reserved.
3  *
4  * This code is derived from software contributed to The DragonFly Project
5  * by Matthew Dillon <dillon@backplane.com>
6  * 
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in
15  *    the documentation and/or other materials provided with the
16  *    distribution.
17  * 3. Neither the name of The DragonFly Project nor the names of its
18  *    contributors may be used to endorse or promote products derived
19  *    from this software without specific, prior written permission.
20  * 
21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  * 
34  * $FreeBSD: src/lib/libkvm/kvm_getswapinfo.c,v 1.10.2.4 2003/01/12 09:23:13 dillon Exp $
35  */
36
37 #define _KERNEL_STRUCTURES
38
39 #include <sys/param.h>
40 #include <sys/time.h>
41 #include <sys/ucred.h>
42 #include <sys/stat.h>
43 #include <sys/conf.h>
44 #include <sys/blist.h>
45 #include <sys/sysctl.h>
46 #include <vm/vm_param.h>
47
48 #include <err.h>
49 #include <fcntl.h>
50 #include <nlist.h>
51 #include <paths.h>
52 #include <stdio.h>
53 #include <stdlib.h>
54 #include <string.h>
55 #include <unistd.h>
56 #include <limits.h>
57
58 #include "kvm.h"
59 #include "kvm_private.h"
60
61 static struct nlist kvm_swap_nl[] = {
62         { "_swapblist" },       /* new radix swap list          */
63         { "_swdevt" },          /* list of swap devices and sizes */
64         { "_nswdev" },          /* number of swap devices */
65         { "_dmmax" },           /* maximum size of a swap block */
66         { "" }
67 };
68
69 #define NL_SWAPBLIST    0
70 #define NL_SWDEVT       1
71 #define NL_NSWDEV       2
72 #define NL_DMMAX        3
73
74 static int kvm_swap_nl_cached = 0;
75 static int nswdev;
76 static int unswdev;
77 static int dmmax;
78
79 static int nlist_init(kvm_t *kd);
80 static void dump_blist(kvm_t *kd);
81 static int kvm_getswapinfo_sysctl(kvm_t *kd, struct kvm_swap *swap_ary,
82                               int swap_max, int flags);
83
84 #define SVAR(var) __STRING(var) /* to force expansion */
85 #define KGET(idx, var)                                                  \
86         KGET1(idx, &var, sizeof(var), SVAR(var))
87 #define KGET1(idx, p, s, msg)                                           \
88         KGET2(kvm_swap_nl[idx].n_value, p, s, msg)
89 #define KGET2(addr, p, s, msg)                                          \
90         if (kvm_read(kd, (u_long)(addr), p, s) != s)                    \
91                 warnx("cannot read %s: %s", msg, kvm_geterr(kd))
92 #define KGETN(idx, var)                                                 \
93         KGET1N(idx, &var, sizeof(var), SVAR(var))
94 #define KGET1N(idx, p, s, msg)                                          \
95         KGET2N(kvm_swap_nl[idx].n_value, p, s, msg)
96 #define KGET2N(addr, p, s, msg)                                         \
97         ((kvm_read(kd, (u_long)(addr), p, s) == s) ? 1 : 0)
98 #define KGETRET(addr, p, s, msg)                                        \
99         if (kvm_read(kd, (u_long)(addr), p, s) != s) {                  \
100                 warnx("cannot read %s: %s", msg, kvm_geterr(kd));       \
101                 return (0);                                             \
102         }
103
104 #define GETSWDEVNAME(dev, str, flags)                                   \
105         if (dev == NODEV) {                                             \
106                 strlcpy(str, "[NFS swap]", sizeof(str));                \
107         } else {                                                        \
108                 snprintf(                                               \
109                     str, sizeof(str), "%s%s",                           \
110                     ((flags & SWIF_DEV_PREFIX) ? _PATH_DEV : ""),       \
111                     devname(dev, S_IFCHR)                               \
112                 );                                                      \
113         }
114
115 int
116 kvm_getswapinfo(
117         kvm_t *kd, 
118         struct kvm_swap *swap_ary,
119         int swap_max, 
120         int flags
121 ) {
122         int i, ti, swi;
123         swblk_t ttl;
124         struct swdevt *sw;
125         struct swdevt swinfo;
126
127         /*
128          * clear cache
129          */
130         if (kd == NULL) {
131                 kvm_swap_nl_cached = 0;
132                 return(0);
133         }
134
135         if (swap_max < 1)
136                 return (-1);
137
138         /*
139          * Use sysctl if possible
140          */
141         if (kvm_ishost(kd) && (flags & SWIF_DUMP_TREE) == 0) {
142                 ti = kvm_getswapinfo_sysctl(kd, swap_ary, swap_max, flags);
143                 if (ti >= 0)
144                         return(ti);
145         }
146
147         /*
148          * namelist
149          */
150         if (!nlist_init(kd))
151                 return (-1);
152
153         swi = unswdev;
154         if (swi >= swap_max)
155                 swi = swap_max - 1;
156
157         bzero(swap_ary, sizeof(struct kvm_swap) * (swi + 1));
158
159         KGET(NL_SWDEVT, sw);
160         for (i = ti = 0; i < nswdev; ++i) {
161                 KGET2(&sw[i], &swinfo, sizeof(swinfo), "swinfo");
162
163                 if (swinfo.sw_nblks == 0)
164                         continue;
165
166                 /*
167                  * The first dmmax is never allocated to avoid
168                  * trashing the disklabels.
169                  */
170                 ttl = swinfo.sw_nblks - dmmax;
171                 if (ttl == 0)
172                         continue;
173
174                 swap_ary[swi].ksw_total += ttl;
175                 swap_ary[swi].ksw_used += swinfo.sw_nused;
176
177                 if (ti < swi) {
178                         swap_ary[ti].ksw_total = ttl;
179                         swap_ary[ti].ksw_used = swinfo.sw_nused;
180                         swap_ary[ti].ksw_flags = swinfo.sw_flags;
181                         GETSWDEVNAME(swinfo.sw_dev, swap_ary[ti].ksw_devname,
182                             flags);
183                         ++ti;
184                 }
185         }
186
187         if (flags & SWIF_DUMP_TREE)
188                 dump_blist(kd);
189         return (swi);
190 }
191
192 static int
193 nlist_init(kvm_t *kd)
194 {
195         int i;
196         struct swdevt *sw;
197         struct swdevt swinfo;
198
199         if (kvm_swap_nl_cached)
200                 return (1);
201
202         if (kvm_nlist(kd, kvm_swap_nl) < 0)
203                 return (0);
204
205         /*
206          * required entries
207          */
208         if (kvm_swap_nl[NL_SWDEVT].n_value == 0 ||
209             kvm_swap_nl[NL_NSWDEV].n_value == 0 ||
210             kvm_swap_nl[NL_DMMAX].n_value == 0 ||
211             kvm_swap_nl[NL_SWAPBLIST].n_type == 0) {
212                 return (0);
213         }
214
215         /*
216          * get globals, type of swap
217          */
218         KGET(NL_NSWDEV, nswdev);
219         KGET(NL_DMMAX, dmmax);
220
221         /*
222          * figure out how many actual swap devices are enabled
223          */
224         KGET(NL_SWDEVT, sw);
225         for (i = unswdev = 0; i < nswdev; ++i) {
226                 KGET2(&sw[i], &swinfo, sizeof(swinfo), "swinfo");
227                 if (swinfo.sw_nblks)
228                         ++unswdev;
229
230         }
231
232         kvm_swap_nl_cached = 1;
233         return (1);
234 }
235
236 /*
237  * scanradix() - support routine for radix scanner
238  */
239
240 #define TABME   tab, tab, ""
241
242 static int
243 scanradix(
244         blmeta_t *scan, 
245         blmeta_t *scan_cache,
246         swblk_t blk,
247         int64_t radix,
248         swblk_t skip,
249         swblk_t count,
250         kvm_t *kd,
251         int dmmaxr,
252         int nswdevr,
253         int64_t *availp,
254         int tab
255 ) {
256         blmeta_t meta;
257         blmeta_t scan_array[BLIST_BMAP_RADIX];
258         int64_t avail_tmp = 0;
259         int i, im;
260         int next_skip;
261
262         if (scan_cache) {
263                 meta = *scan_cache;
264         } else if (skip == BLIST_META_RADIX) {
265                 if (kvm_read(kd, (u_long)scan, scan_array, sizeof(scan_array)) != sizeof(scan_array)) {
266                         warnx("cannot read %s: %s", "blmeta_t", kvm_geterr(kd));
267                         bzero(scan_array, sizeof(scan_array));
268                 }
269                 meta = scan_array[0];
270         } else {
271                 KGET2(scan, &meta, sizeof(meta), "blmeta_t");
272         }
273
274         /*
275          * Terminator
276          */
277         if (meta.bm_bighint == (swblk_t)-1) {
278                 printf("%*.*s(0x%06jx,%jd) Terminator\n",
279                     TABME,
280                     (intmax_t)blk,
281                     (intmax_t)radix
282                 );
283                 return(-1);
284         }
285
286         if (radix == BLIST_BMAP_RADIX) {
287                 /*
288                  * Leaf bitmap
289                  */
290                 printf("%*.*s(0x%06jx,%jd) Bitmap %016jx big=%jd\n",
291                     TABME,
292                     (intmax_t)blk,
293                     (intmax_t)radix,
294                     (intmax_t)meta.u.bmu_bitmap,
295                     (intmax_t)meta.bm_bighint
296                 );
297
298                 if (meta.u.bmu_bitmap) {
299                         for (i = 0; i < BLIST_BMAP_RADIX; ++i) {
300                                 if (meta.u.bmu_bitmap & (1 << i))
301                                         ++*availp;
302                         }
303                 }
304         } else if (meta.u.bmu_avail == radix) {
305                 /*
306                  * Meta node if all free
307                  */
308                 printf("%*.*s(0x%06jx,%jd) Submap ALL-FREE (big=%jd) {\n",
309                     TABME,
310                     (intmax_t)blk,
311                     (intmax_t)radix,
312                     (intmax_t)meta.bm_bighint
313                 );
314                 *availp += radix;
315         } else if (meta.u.bmu_avail == 0) {
316                 /*
317                  * Meta node if all used
318                  */
319                 printf("%*.*s(0x%06jx,%jd) Submap ALL-ALLOCATED (big=%jd)\n",
320                     TABME,
321                     (intmax_t)blk,
322                     (intmax_t)radix,
323                     (intmax_t)meta.bm_bighint
324                 );
325         } else {
326                 /*
327                  * Meta node if not all free
328                  */
329                 printf("%*.*s(0x%06jx,%jd) Submap avail=%jd big=%jd {\n",
330                     TABME,
331                     (intmax_t)blk,
332                     (intmax_t)radix,
333                     (intmax_t)meta.u.bmu_avail,
334                     (intmax_t)meta.bm_bighint
335                 );
336
337                 radix /= BLIST_META_RADIX;
338                 next_skip = skip / BLIST_META_RADIX;
339
340                 for (im = 1; im <= skip; im += next_skip) {
341                         int r;
342                         swblk_t vcount = (count > radix) ?
343                                         (swblk_t)radix : count;
344
345                         r = scanradix(
346                             &scan[im],
347                             ((next_skip == 1) ? &scan_array[im] : NULL),
348                             blk,
349                             radix,
350                             next_skip - 1,
351                             vcount,
352                             kd,
353                             dmmaxr,
354                             nswdevr,
355                             &avail_tmp,
356                             tab + 4
357                         );
358                         if (r < 0)
359                                 break;
360                         blk += (swblk_t)radix;
361                 }
362                 *availp += avail_tmp;
363                 if (avail_tmp == meta.u.bmu_avail)
364                         printf("%*.*s}\n", TABME);
365                 else
366                         printf("%*.*s} (AVAIL MISMATCH %jd/%jd\n",
367                                 TABME,
368                                 (intmax_t)avail_tmp,
369                                 (intmax_t)meta.u.bmu_avail);
370         }
371         return(0);
372 }
373
374 static void
375 dump_blist(kvm_t *kd)
376 {
377         struct blist *swapblist = NULL;
378         struct blist blcopy = { 0 };
379         int64_t avail = 0;
380
381         KGET(NL_SWAPBLIST, swapblist);
382
383         if (swapblist == NULL) {
384                 printf("radix tree: NULL - no swap in system\n");
385                 return;
386         }
387
388         KGET2(swapblist, &blcopy, sizeof(blcopy), "*swapblist");
389
390         printf("radix tree: %jd/%jd/%jd blocks, %jdK wired\n",
391                 (intmax_t)blcopy.bl_free,
392                 (intmax_t)blcopy.bl_blocks,
393                 (intmax_t)blcopy.bl_radix,
394                 (intmax_t)((blcopy.bl_rootblks * sizeof(blmeta_t) + 1023)/
395                     1024)
396         );
397
398         scanradix(
399             blcopy.bl_root,
400             NULL,
401             0,
402             blcopy.bl_radix,
403             blcopy.bl_skip,
404             blcopy.bl_rootblks,
405             kd,
406             dmmax,
407             nswdev,
408             &avail,
409             0
410         );
411         printf("final availability: %jd\n", (intmax_t)avail);
412 }
413
414 static
415 int
416 kvm_getswapinfo_sysctl(kvm_t *kd, struct kvm_swap *swap_ary,
417                        int swap_max, int flags)
418 {
419         size_t bytes = 0;
420         size_t ksize;
421         int ti;
422         int swi;
423         int n;
424         int i;
425         char *xswbuf;
426         struct xswdev *xsw;
427
428         if (sysctlbyname("vm.swap_info_array", NULL, &bytes, NULL, 0) < 0)
429                 return(-1);
430         if (bytes == 0)
431                 return(-1);
432
433         xswbuf = malloc(bytes);
434         if (sysctlbyname("vm.swap_info_array", xswbuf, &bytes, NULL, 0) < 0) {
435                 free(xswbuf);
436                 return(-1);
437         }
438         if (bytes == 0) {
439                 free(xswbuf);
440                 return(-1);
441         }
442
443         /*
444          * Calculate size of xsw entry returned by kernel (it can be larger
445          * than the one we have if there is a version mismatch).
446          */
447         ksize = ((struct xswdev *)xswbuf)->xsw_size;
448         n = (int)(bytes / ksize);
449
450         /*
451          * Calculate the number of live swap devices and calculate
452          * the swap_ary[] index used for the cumulative result (swi)
453          */
454         for (i = swi = 0; i < n; ++i) {
455                 xsw = (void *)((char *)xswbuf + i * ksize);
456                 if ((xsw->xsw_flags & SW_FREED) == 0)
457                         continue;
458                 ++swi;
459         }
460         if (swi >= swap_max)
461                 swi = swap_max - 1;
462
463         bzero(swap_ary, sizeof(struct kvm_swap) * (swi + 1));
464
465         /*
466          * Accumulate results.  If the provided swap_ary[] is too
467          * small will only populate up to the available entries,
468          * but we always populate the cumulative results entry.
469          */
470         for (i = ti = 0; i < n; ++i) {
471                 xsw = (void *)((char *)xswbuf + i * ksize);
472
473                 if ((xsw->xsw_flags & SW_FREED) == 0)
474                         continue;
475
476                 swap_ary[swi].ksw_total += xsw->xsw_nblks;
477                 swap_ary[swi].ksw_used += xsw->xsw_used;
478
479                 if (ti < swi) {
480                         swap_ary[ti].ksw_total = xsw->xsw_nblks;
481                         swap_ary[ti].ksw_used = xsw->xsw_used;
482                         swap_ary[ti].ksw_flags = xsw->xsw_flags;
483                         GETSWDEVNAME(xsw->xsw_dev, swap_ary[ti].ksw_devname,
484                             flags);
485                         ++ti;
486                 }
487         }
488
489         free(xswbuf);
490         return(swi);
491 }