xref: /freebsd/lib/libmemstat/memstat_uma.c (revision 1bee2ec756f2ea5255f9f68dd58c1ceae1a2f56c) 
1 /*-
2  * Copyright (c) 2005-2006 Robert N. M. Watson
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  *
26  * $FreeBSD$
27  */
28 
29 #include <sys/param.h>
30 #include <sys/cpuset.h>
31 #include <sys/sysctl.h>
32 
33 #define	LIBMEMSTAT	/* Cause vm_page.h not to include opt_vmpage.h */
34 #include <vm/vm.h>
35 #include <vm/vm_page.h>
36 
37 #include <vm/uma.h>
38 #include <vm/uma_int.h>
39 
40 #include <err.h>
41 #include <errno.h>
42 #include <kvm.h>
43 #include <nlist.h>
44 #include <stddef.h>
45 #include <stdio.h>
46 #include <stdlib.h>
47 #include <string.h>
48 #include <unistd.h>
49 
50 #include "memstat.h"
51 #include "memstat_internal.h"
52 
53 static struct nlist namelist[] = {
54 #define	X_UMA_KEGS	0
55 	{ .n_name = "_uma_kegs" },
56 #define	X_MP_MAXID	1
57 	{ .n_name = "_mp_maxid" },
58 #define	X_ALL_CPUS	2
59 	{ .n_name = "_all_cpus" },
60 	{ .n_name = "" },
61 };
62 
63 /*
64  * Extract uma(9) statistics from the running kernel, and store all memory
65  * type information in the passed list.  For each type, check the list for an
66  * existing entry with the right name/allocator -- if present, update that
67  * entry.  Otherwise, add a new entry.  On error, the entire list will be
68  * cleared, as entries will be in an inconsistent state.
69  *
70  * To reduce the level of work for a list that starts empty, we keep around a
71  * hint as to whether it was empty when we began, so we can avoid searching
72  * the list for entries to update.  Updates are O(n^2) due to searching for
73  * each entry before adding it.
74  */
75 int
76 memstat_sysctl_uma(struct memory_type_list *list, int flags)
77 {
78 	struct uma_stream_header *ushp;
79 	struct uma_type_header *uthp;
80 	struct uma_percpu_stat *upsp;
81 	struct memory_type *mtp;
82 	int count, hint_dontsearch, i, j, maxcpus;
83 	char *buffer, *p;
84 	size_t size;
85 
86 	hint_dontsearch = LIST_EMPTY(&list->mtl_list);
87 
88 	/*
89 	 * Query the number of CPUs, number of malloc types so that we can
90 	 * guess an initial buffer size.  We loop until we succeed or really
91 	 * fail.  Note that the value of maxcpus we query using sysctl is not
92 	 * the version we use when processing the real data -- that is read
93 	 * from the header.
94 	 */
95 retry:
96 	size = sizeof(maxcpus);
97 	if (sysctlbyname("kern.smp.maxcpus", &maxcpus, &size, NULL, 0) < 0) {
98 		if (errno == EACCES || errno == EPERM)
99 			list->mtl_error = MEMSTAT_ERROR_PERMISSION;
100 		else
101 			list->mtl_error = MEMSTAT_ERROR_DATAERROR;
102 		return (-1);
103 	}
104 	if (size != sizeof(maxcpus)) {
105 		list->mtl_error = MEMSTAT_ERROR_DATAERROR;
106 		return (-1);
107 	}
108 
109 	if (maxcpus > MEMSTAT_MAXCPU) {
110 		list->mtl_error = MEMSTAT_ERROR_TOOMANYCPUS;
111 		return (-1);
112 	}
113 
114 	size = sizeof(count);
115 	if (sysctlbyname("vm.zone_count", &count, &size, NULL, 0) < 0) {
116 		if (errno == EACCES || errno == EPERM)
117 			list->mtl_error = MEMSTAT_ERROR_PERMISSION;
118 		else
119 			list->mtl_error = MEMSTAT_ERROR_VERSION;
120 		return (-1);
121 	}
122 	if (size != sizeof(count)) {
123 		list->mtl_error = MEMSTAT_ERROR_DATAERROR;
124 		return (-1);
125 	}
126 
127 	size = sizeof(*uthp) + count * (sizeof(*uthp) + sizeof(*upsp) *
128 	    maxcpus);
129 
130 	buffer = malloc(size);
131 	if (buffer == NULL) {
132 		list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
133 		return (-1);
134 	}
135 
136 	if (sysctlbyname("vm.zone_stats", buffer, &size, NULL, 0) < 0) {
137 		/*
138 		 * XXXRW: ENOMEM is an ambiguous return, we should bound the
139 		 * number of loops, perhaps.
140 		 */
141 		if (errno == ENOMEM) {
142 			free(buffer);
143 			goto retry;
144 		}
145 		if (errno == EACCES || errno == EPERM)
146 			list->mtl_error = MEMSTAT_ERROR_PERMISSION;
147 		else
148 			list->mtl_error = MEMSTAT_ERROR_VERSION;
149 		free(buffer);
150 		return (-1);
151 	}
152 
153 	if (size == 0) {
154 		free(buffer);
155 		return (0);
156 	}
157 
158 	if (size < sizeof(*ushp)) {
159 		list->mtl_error = MEMSTAT_ERROR_VERSION;
160 		free(buffer);
161 		return (-1);
162 	}
163 	p = buffer;
164 	ushp = (struct uma_stream_header *)p;
165 	p += sizeof(*ushp);
166 
167 	if (ushp->ush_version != UMA_STREAM_VERSION) {
168 		list->mtl_error = MEMSTAT_ERROR_VERSION;
169 		free(buffer);
170 		return (-1);
171 	}
172 
173 	if (ushp->ush_maxcpus > MEMSTAT_MAXCPU) {
174 		list->mtl_error = MEMSTAT_ERROR_TOOMANYCPUS;
175 		free(buffer);
176 		return (-1);
177 	}
178 
179 	/*
180 	 * For the remainder of this function, we are quite trusting about
181 	 * the layout of structures and sizes, since we've determined we have
182 	 * a matching version and acceptable CPU count.
183 	 */
184 	maxcpus = ushp->ush_maxcpus;
185 	count = ushp->ush_count;
186 	for (i = 0; i < count; i++) {
187 		uthp = (struct uma_type_header *)p;
188 		p += sizeof(*uthp);
189 
190 		if (hint_dontsearch == 0) {
191 			mtp = memstat_mtl_find(list, ALLOCATOR_UMA,
192 			    uthp->uth_name);
193 		} else
194 			mtp = NULL;
195 		if (mtp == NULL)
196 			mtp = _memstat_mt_allocate(list, ALLOCATOR_UMA,
197 			    uthp->uth_name);
198 		if (mtp == NULL) {
199 			_memstat_mtl_empty(list);
200 			free(buffer);
201 			list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
202 			return (-1);
203 		}
204 
205 		/*
206 		 * Reset the statistics on a current node.
207 		 */
208 		_memstat_mt_reset_stats(mtp);
209 
210 		mtp->mt_numallocs = uthp->uth_allocs;
211 		mtp->mt_numfrees = uthp->uth_frees;
212 		mtp->mt_failures = uthp->uth_fails;
213 		mtp->mt_sleeps = uthp->uth_sleeps;
214 
215 		for (j = 0; j < maxcpus; j++) {
216 			upsp = (struct uma_percpu_stat *)p;
217 			p += sizeof(*upsp);
218 
219 			mtp->mt_percpu_cache[j].mtp_free =
220 			    upsp->ups_cache_free;
221 			mtp->mt_free += upsp->ups_cache_free;
222 			mtp->mt_numallocs += upsp->ups_allocs;
223 			mtp->mt_numfrees += upsp->ups_frees;
224 		}
225 
226 		mtp->mt_size = uthp->uth_size;
227 		mtp->mt_memalloced = mtp->mt_numallocs * uthp->uth_size;
228 		mtp->mt_memfreed = mtp->mt_numfrees * uthp->uth_size;
229 		mtp->mt_bytes = mtp->mt_memalloced - mtp->mt_memfreed;
230 		mtp->mt_countlimit = uthp->uth_limit;
231 		mtp->mt_byteslimit = uthp->uth_limit * uthp->uth_size;
232 
233 		mtp->mt_count = mtp->mt_numallocs - mtp->mt_numfrees;
234 		mtp->mt_zonefree = uthp->uth_zone_free;
235 
236 		/*
237 		 * UMA secondary zones share a keg with the primary zone.  To
238 		 * avoid double-reporting of free items, report keg free
239 		 * items only in the primary zone.
240 		 */
241 		if (!(uthp->uth_zone_flags & UTH_ZONE_SECONDARY)) {
242 			mtp->mt_kegfree = uthp->uth_keg_free;
243 			mtp->mt_free += mtp->mt_kegfree;
244 		}
245 		mtp->mt_free += mtp->mt_zonefree;
246 	}
247 
248 	free(buffer);
249 
250 	return (0);
251 }
252 
253 static int
254 kread(kvm_t *kvm, void *kvm_pointer, void *address, size_t size,
255     size_t offset)
256 {
257 	ssize_t ret;
258 
259 	ret = kvm_read(kvm, (unsigned long)kvm_pointer + offset, address,
260 	    size);
261 	if (ret < 0)
262 		return (MEMSTAT_ERROR_KVM);
263 	if ((size_t)ret != size)
264 		return (MEMSTAT_ERROR_KVM_SHORTREAD);
265 	return (0);
266 }
267 
268 static int
269 kread_string(kvm_t *kvm, void *kvm_pointer, char *buffer, int buflen)
270 {
271 	ssize_t ret;
272 	int i;
273 
274 	for (i = 0; i < buflen; i++) {
275 		ret = kvm_read(kvm, (unsigned long)kvm_pointer + i,
276 		    &(buffer[i]), sizeof(char));
277 		if (ret < 0)
278 			return (MEMSTAT_ERROR_KVM);
279 		if ((size_t)ret != sizeof(char))
280 			return (MEMSTAT_ERROR_KVM_SHORTREAD);
281 		if (buffer[i] == '\0')
282 			return (0);
283 	}
284 	/* Truncate. */
285 	buffer[i-1] = '\0';
286 	return (0);
287 }
288 
289 static int
290 kread_symbol(kvm_t *kvm, int index, void *address, size_t size,
291     size_t offset)
292 {
293 	ssize_t ret;
294 
295 	ret = kvm_read(kvm, namelist[index].n_value + offset, address, size);
296 	if (ret < 0)
297 		return (MEMSTAT_ERROR_KVM);
298 	if ((size_t)ret != size)
299 		return (MEMSTAT_ERROR_KVM_SHORTREAD);
300 	return (0);
301 }
302 
303 /*
304  * memstat_kvm_uma() is similar to memstat_sysctl_uma(), only it extracts
305  * UMA(9) statistics from a kernel core/memory file.
306  */
307 int
308 memstat_kvm_uma(struct memory_type_list *list, void *kvm_handle)
309 {
310 	LIST_HEAD(, uma_keg) uma_kegs;
311 	struct memory_type *mtp;
312 	struct uma_bucket *ubp, ub;
313 	struct uma_cache *ucp, *ucp_array;
314 	struct uma_zone *uzp, uz;
315 	struct uma_keg *kzp, kz;
316 	int hint_dontsearch, i, mp_maxid, ret;
317 	char name[MEMTYPE_MAXNAME];
318 	cpuset_t all_cpus;
319 	long cpusetsize;
320 	kvm_t *kvm;
321 
322 	kvm = (kvm_t *)kvm_handle;
323 	hint_dontsearch = LIST_EMPTY(&list->mtl_list);
324 	if (kvm_nlist(kvm, namelist) != 0) {
325 		list->mtl_error = MEMSTAT_ERROR_KVM;
326 		return (-1);
327 	}
328 	if (namelist[X_UMA_KEGS].n_type == 0 ||
329 	    namelist[X_UMA_KEGS].n_value == 0) {
330 		list->mtl_error = MEMSTAT_ERROR_KVM_NOSYMBOL;
331 		return (-1);
332 	}
333 	ret = kread_symbol(kvm, X_MP_MAXID, &mp_maxid, sizeof(mp_maxid), 0);
334 	if (ret != 0) {
335 		list->mtl_error = ret;
336 		return (-1);
337 	}
338 	ret = kread_symbol(kvm, X_UMA_KEGS, &uma_kegs, sizeof(uma_kegs), 0);
339 	if (ret != 0) {
340 		list->mtl_error = ret;
341 		return (-1);
342 	}
343 	cpusetsize = sysconf(_SC_CPUSET_SIZE);
344 	if (cpusetsize == -1 || (u_long)cpusetsize > sizeof(cpuset_t)) {
345 		list->mtl_error = MEMSTAT_ERROR_KVM_NOSYMBOL;
346 		return (-1);
347 	}
348 	CPU_ZERO(&all_cpus);
349 	ret = kread_symbol(kvm, X_ALL_CPUS, &all_cpus, cpusetsize, 0);
350 	if (ret != 0) {
351 		list->mtl_error = ret;
352 		return (-1);
353 	}
354 	ucp_array = malloc(sizeof(struct uma_cache) * (mp_maxid + 1));
355 	if (ucp_array == NULL) {
356 		list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
357 		return (-1);
358 	}
359 	for (kzp = LIST_FIRST(&uma_kegs); kzp != NULL; kzp =
360 	    LIST_NEXT(&kz, uk_link)) {
361 		ret = kread(kvm, kzp, &kz, sizeof(kz), 0);
362 		if (ret != 0) {
363 			free(ucp_array);
364 			_memstat_mtl_empty(list);
365 			list->mtl_error = ret;
366 			return (-1);
367 		}
368 		for (uzp = LIST_FIRST(&kz.uk_zones); uzp != NULL; uzp =
369 		    LIST_NEXT(&uz, uz_link)) {
370 			ret = kread(kvm, uzp, &uz, sizeof(uz), 0);
371 			if (ret != 0) {
372 				free(ucp_array);
373 				_memstat_mtl_empty(list);
374 				list->mtl_error = ret;
375 				return (-1);
376 			}
377 			ret = kread(kvm, uzp, ucp_array,
378 			    sizeof(struct uma_cache) * (mp_maxid + 1),
379 			    offsetof(struct uma_zone, uz_cpu[0]));
380 			if (ret != 0) {
381 				free(ucp_array);
382 				_memstat_mtl_empty(list);
383 				list->mtl_error = ret;
384 				return (-1);
385 			}
386 			ret = kread_string(kvm, uz.uz_name, name,
387 			    MEMTYPE_MAXNAME);
388 			if (ret != 0) {
389 				free(ucp_array);
390 				_memstat_mtl_empty(list);
391 				list->mtl_error = ret;
392 				return (-1);
393 			}
394 			if (hint_dontsearch == 0) {
395 				mtp = memstat_mtl_find(list, ALLOCATOR_UMA,
396 				    name);
397 			} else
398 				mtp = NULL;
399 			if (mtp == NULL)
400 				mtp = _memstat_mt_allocate(list, ALLOCATOR_UMA,
401 				    name);
402 			if (mtp == NULL) {
403 				free(ucp_array);
404 				_memstat_mtl_empty(list);
405 				list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
406 				return (-1);
407 			}
408 			/*
409 			 * Reset the statistics on a current node.
410 			 */
411 			_memstat_mt_reset_stats(mtp);
412 			mtp->mt_numallocs = uz.uz_allocs;
413 			mtp->mt_numfrees = uz.uz_frees;
414 			mtp->mt_failures = uz.uz_fails;
415 			mtp->mt_sleeps = uz.uz_sleeps;
416 			if (kz.uk_flags & UMA_ZFLAG_INTERNAL)
417 				goto skip_percpu;
418 			for (i = 0; i < mp_maxid + 1; i++) {
419 				if (!CPU_ISSET(i, &all_cpus))
420 					continue;
421 				ucp = &ucp_array[i];
422 				mtp->mt_numallocs += ucp->uc_allocs;
423 				mtp->mt_numfrees += ucp->uc_frees;
424 
425 				if (ucp->uc_allocbucket != NULL) {
426 					ret = kread(kvm, ucp->uc_allocbucket,
427 					    &ub, sizeof(ub), 0);
428 					if (ret != 0) {
429 						free(ucp_array);
430 						_memstat_mtl_empty(list);
431 						list->mtl_error = ret;
432 						return (-1);
433 					}
434 					mtp->mt_free += ub.ub_cnt;
435 				}
436 				if (ucp->uc_freebucket != NULL) {
437 					ret = kread(kvm, ucp->uc_freebucket,
438 					    &ub, sizeof(ub), 0);
439 					if (ret != 0) {
440 						free(ucp_array);
441 						_memstat_mtl_empty(list);
442 						list->mtl_error = ret;
443 						return (-1);
444 					}
445 					mtp->mt_free += ub.ub_cnt;
446 				}
447 			}
448 skip_percpu:
449 			mtp->mt_size = kz.uk_size;
450 			mtp->mt_memalloced = mtp->mt_numallocs * mtp->mt_size;
451 			mtp->mt_memfreed = mtp->mt_numfrees * mtp->mt_size;
452 			mtp->mt_bytes = mtp->mt_memalloced - mtp->mt_memfreed;
453 			if (kz.uk_ppera > 1)
454 				mtp->mt_countlimit = kz.uk_maxpages /
455 				    kz.uk_ipers;
456 			else
457 				mtp->mt_countlimit = kz.uk_maxpages *
458 				    kz.uk_ipers;
459 			mtp->mt_byteslimit = mtp->mt_countlimit * mtp->mt_size;
460 			mtp->mt_count = mtp->mt_numallocs - mtp->mt_numfrees;
461 			for (ubp = LIST_FIRST(&uz.uz_full_bucket); ubp !=
462 			    NULL; ubp = LIST_NEXT(&ub, ub_link)) {
463 				ret = kread(kvm, ubp, &ub, sizeof(ub), 0);
464 				mtp->mt_zonefree += ub.ub_cnt;
465 			}
466 			if (!((kz.uk_flags & UMA_ZONE_SECONDARY) &&
467 			    LIST_FIRST(&kz.uk_zones) != uzp)) {
468 				mtp->mt_kegfree = kz.uk_free;
469 				mtp->mt_free += mtp->mt_kegfree;
470 			}
471 			mtp->mt_free += mtp->mt_zonefree;
472 		}
473 	}
474 	free(ucp_array);
475 	return (0);
476 }
477