xref: /freebsd/lib/libmemstat/memstat_uma.c (revision 6829dae12bb055451fa467da4589c43bd03b1e64) 
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2005-2006 Robert N. M. Watson
5  * All rights reserved.
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  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  *
28  * $FreeBSD$
29  */
30 
31 #include <sys/param.h>
32 #include <sys/counter.h>
33 #include <sys/cpuset.h>
34 #include <sys/sysctl.h>
35 
36 #include <vm/uma.h>
37 #include <vm/uma_int.h>
38 
39 #include <err.h>
40 #include <errno.h>
41 #include <kvm.h>
42 #include <nlist.h>
43 #include <stddef.h>
44 #include <stdio.h>
45 #include <stdlib.h>
46 #include <string.h>
47 #include <unistd.h>
48 
49 #include "memstat.h"
50 #include "memstat_internal.h"
51 
52 static struct nlist namelist[] = {
53 #define	X_UMA_KEGS	0
54 	{ .n_name = "_uma_kegs" },
55 #define	X_MP_MAXID	1
56 	{ .n_name = "_mp_maxid" },
57 #define	X_ALL_CPUS	2
58 	{ .n_name = "_all_cpus" },
59 #define	X_VM_NDOMAINS	3
60 	{ .n_name = "_vm_ndomains" },
61 	{ .n_name = "" },
62 };
63 
64 /*
65  * Extract uma(9) statistics from the running kernel, and store all memory
66  * type information in the passed list.  For each type, check the list for an
67  * existing entry with the right name/allocator -- if present, update that
68  * entry.  Otherwise, add a new entry.  On error, the entire list will be
69  * cleared, as entries will be in an inconsistent state.
70  *
71  * To reduce the level of work for a list that starts empty, we keep around a
72  * hint as to whether it was empty when we began, so we can avoid searching
73  * the list for entries to update.  Updates are O(n^2) due to searching for
74  * each entry before adding it.
75  */
76 int
77 memstat_sysctl_uma(struct memory_type_list *list, int flags)
78 {
79 	struct uma_stream_header *ushp;
80 	struct uma_type_header *uthp;
81 	struct uma_percpu_stat *upsp;
82 	struct memory_type *mtp;
83 	int count, hint_dontsearch, i, j, maxcpus, maxid;
84 	char *buffer, *p;
85 	size_t size;
86 
87 	hint_dontsearch = LIST_EMPTY(&list->mtl_list);
88 
89 	/*
90 	 * Query the number of CPUs, number of malloc types so that we can
91 	 * guess an initial buffer size.  We loop until we succeed or really
92 	 * fail.  Note that the value of maxcpus we query using sysctl is not
93 	 * the version we use when processing the real data -- that is read
94 	 * from the header.
95 	 */
96 retry:
97 	size = sizeof(maxid);
98 	if (sysctlbyname("kern.smp.maxid", &maxid, &size, NULL, 0) < 0) {
99 		if (errno == EACCES || errno == EPERM)
100 			list->mtl_error = MEMSTAT_ERROR_PERMISSION;
101 		else
102 			list->mtl_error = MEMSTAT_ERROR_DATAERROR;
103 		return (-1);
104 	}
105 	if (size != sizeof(maxid)) {
106 		list->mtl_error = MEMSTAT_ERROR_DATAERROR;
107 		return (-1);
108 	}
109 
110 	size = sizeof(count);
111 	if (sysctlbyname("vm.zone_count", &count, &size, NULL, 0) < 0) {
112 		if (errno == EACCES || errno == EPERM)
113 			list->mtl_error = MEMSTAT_ERROR_PERMISSION;
114 		else
115 			list->mtl_error = MEMSTAT_ERROR_VERSION;
116 		return (-1);
117 	}
118 	if (size != sizeof(count)) {
119 		list->mtl_error = MEMSTAT_ERROR_DATAERROR;
120 		return (-1);
121 	}
122 
123 	size = sizeof(*uthp) + count * (sizeof(*uthp) + sizeof(*upsp) *
124 	    (maxid + 1));
125 
126 	buffer = malloc(size);
127 	if (buffer == NULL) {
128 		list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
129 		return (-1);
130 	}
131 
132 	if (sysctlbyname("vm.zone_stats", buffer, &size, NULL, 0) < 0) {
133 		/*
134 		 * XXXRW: ENOMEM is an ambiguous return, we should bound the
135 		 * number of loops, perhaps.
136 		 */
137 		if (errno == ENOMEM) {
138 			free(buffer);
139 			goto retry;
140 		}
141 		if (errno == EACCES || errno == EPERM)
142 			list->mtl_error = MEMSTAT_ERROR_PERMISSION;
143 		else
144 			list->mtl_error = MEMSTAT_ERROR_VERSION;
145 		free(buffer);
146 		return (-1);
147 	}
148 
149 	if (size == 0) {
150 		free(buffer);
151 		return (0);
152 	}
153 
154 	if (size < sizeof(*ushp)) {
155 		list->mtl_error = MEMSTAT_ERROR_VERSION;
156 		free(buffer);
157 		return (-1);
158 	}
159 	p = buffer;
160 	ushp = (struct uma_stream_header *)p;
161 	p += sizeof(*ushp);
162 
163 	if (ushp->ush_version != UMA_STREAM_VERSION) {
164 		list->mtl_error = MEMSTAT_ERROR_VERSION;
165 		free(buffer);
166 		return (-1);
167 	}
168 
169 	/*
170 	 * For the remainder of this function, we are quite trusting about
171 	 * the layout of structures and sizes, since we've determined we have
172 	 * a matching version and acceptable CPU count.
173 	 */
174 	maxcpus = ushp->ush_maxcpus;
175 	count = ushp->ush_count;
176 	for (i = 0; i < count; i++) {
177 		uthp = (struct uma_type_header *)p;
178 		p += sizeof(*uthp);
179 
180 		if (hint_dontsearch == 0) {
181 			mtp = memstat_mtl_find(list, ALLOCATOR_UMA,
182 			    uthp->uth_name);
183 		} else
184 			mtp = NULL;
185 		if (mtp == NULL)
186 			mtp = _memstat_mt_allocate(list, ALLOCATOR_UMA,
187 			    uthp->uth_name, maxid + 1);
188 		if (mtp == NULL) {
189 			_memstat_mtl_empty(list);
190 			free(buffer);
191 			list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
192 			return (-1);
193 		}
194 
195 		/*
196 		 * Reset the statistics on a current node.
197 		 */
198 		_memstat_mt_reset_stats(mtp, maxid + 1);
199 
200 		mtp->mt_numallocs = uthp->uth_allocs;
201 		mtp->mt_numfrees = uthp->uth_frees;
202 		mtp->mt_failures = uthp->uth_fails;
203 		mtp->mt_sleeps = uthp->uth_sleeps;
204 
205 		for (j = 0; j < maxcpus; j++) {
206 			upsp = (struct uma_percpu_stat *)p;
207 			p += sizeof(*upsp);
208 
209 			mtp->mt_percpu_cache[j].mtp_free =
210 			    upsp->ups_cache_free;
211 			mtp->mt_free += upsp->ups_cache_free;
212 			mtp->mt_numallocs += upsp->ups_allocs;
213 			mtp->mt_numfrees += upsp->ups_frees;
214 		}
215 
216 		/*
217 		 * Values for uth_allocs and uth_frees frees are snap.
218 		 * It may happen that kernel reports that number of frees
219 		 * is greater than number of allocs. See counter(9) for
220 		 * details.
221 		 */
222 		if (mtp->mt_numallocs < mtp->mt_numfrees)
223 			mtp->mt_numallocs = mtp->mt_numfrees;
224 
225 		mtp->mt_size = uthp->uth_size;
226 		mtp->mt_rsize = uthp->uth_rsize;
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, const 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_zone_domain uzd;
313 	struct uma_bucket *ubp, ub;
314 	struct uma_cache *ucp, *ucp_array;
315 	struct uma_zone *uzp, uz;
316 	struct uma_keg *kzp, kz;
317 	int hint_dontsearch, i, mp_maxid, ndomains, ret;
318 	char name[MEMTYPE_MAXNAME];
319 	cpuset_t all_cpus;
320 	long cpusetsize;
321 	kvm_t *kvm;
322 
323 	kvm = (kvm_t *)kvm_handle;
324 	hint_dontsearch = LIST_EMPTY(&list->mtl_list);
325 	if (kvm_nlist(kvm, namelist) != 0) {
326 		list->mtl_error = MEMSTAT_ERROR_KVM;
327 		return (-1);
328 	}
329 	if (namelist[X_UMA_KEGS].n_type == 0 ||
330 	    namelist[X_UMA_KEGS].n_value == 0) {
331 		list->mtl_error = MEMSTAT_ERROR_KVM_NOSYMBOL;
332 		return (-1);
333 	}
334 	ret = kread_symbol(kvm, X_MP_MAXID, &mp_maxid, sizeof(mp_maxid), 0);
335 	if (ret != 0) {
336 		list->mtl_error = ret;
337 		return (-1);
338 	}
339 	ret = kread_symbol(kvm, X_VM_NDOMAINS, &ndomains,
340 	    sizeof(ndomains), 0);
341 	if (ret != 0) {
342 		list->mtl_error = ret;
343 		return (-1);
344 	}
345 	ret = kread_symbol(kvm, X_UMA_KEGS, &uma_kegs, sizeof(uma_kegs), 0);
346 	if (ret != 0) {
347 		list->mtl_error = ret;
348 		return (-1);
349 	}
350 	cpusetsize = sysconf(_SC_CPUSET_SIZE);
351 	if (cpusetsize == -1 || (u_long)cpusetsize > sizeof(cpuset_t)) {
352 		list->mtl_error = MEMSTAT_ERROR_KVM_NOSYMBOL;
353 		return (-1);
354 	}
355 	CPU_ZERO(&all_cpus);
356 	ret = kread_symbol(kvm, X_ALL_CPUS, &all_cpus, cpusetsize, 0);
357 	if (ret != 0) {
358 		list->mtl_error = ret;
359 		return (-1);
360 	}
361 	ucp_array = malloc(sizeof(struct uma_cache) * (mp_maxid + 1));
362 	if (ucp_array == NULL) {
363 		list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
364 		return (-1);
365 	}
366 	for (kzp = LIST_FIRST(&uma_kegs); kzp != NULL; kzp =
367 	    LIST_NEXT(&kz, uk_link)) {
368 		ret = kread(kvm, kzp, &kz, sizeof(kz), 0);
369 		if (ret != 0) {
370 			free(ucp_array);
371 			_memstat_mtl_empty(list);
372 			list->mtl_error = ret;
373 			return (-1);
374 		}
375 		for (uzp = LIST_FIRST(&kz.uk_zones); uzp != NULL; uzp =
376 		    LIST_NEXT(&uz, uz_link)) {
377 			ret = kread(kvm, uzp, &uz, sizeof(uz), 0);
378 			if (ret != 0) {
379 				free(ucp_array);
380 				_memstat_mtl_empty(list);
381 				list->mtl_error = ret;
382 				return (-1);
383 			}
384 			ret = kread(kvm, uzp, ucp_array,
385 			    sizeof(struct uma_cache) * (mp_maxid + 1),
386 			    offsetof(struct uma_zone, uz_cpu[0]));
387 			if (ret != 0) {
388 				free(ucp_array);
389 				_memstat_mtl_empty(list);
390 				list->mtl_error = ret;
391 				return (-1);
392 			}
393 			ret = kread_string(kvm, uz.uz_name, name,
394 			    MEMTYPE_MAXNAME);
395 			if (ret != 0) {
396 				free(ucp_array);
397 				_memstat_mtl_empty(list);
398 				list->mtl_error = ret;
399 				return (-1);
400 			}
401 			if (hint_dontsearch == 0) {
402 				mtp = memstat_mtl_find(list, ALLOCATOR_UMA,
403 				    name);
404 			} else
405 				mtp = NULL;
406 			if (mtp == NULL)
407 				mtp = _memstat_mt_allocate(list, ALLOCATOR_UMA,
408 				    name, mp_maxid + 1);
409 			if (mtp == NULL) {
410 				free(ucp_array);
411 				_memstat_mtl_empty(list);
412 				list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
413 				return (-1);
414 			}
415 			/*
416 			 * Reset the statistics on a current node.
417 			 */
418 			_memstat_mt_reset_stats(mtp, mp_maxid + 1);
419 			mtp->mt_numallocs = kvm_counter_u64_fetch(kvm,
420 			    (unsigned long )uz.uz_allocs);
421 			mtp->mt_numfrees = kvm_counter_u64_fetch(kvm,
422 			    (unsigned long )uz.uz_frees);
423 			mtp->mt_failures = kvm_counter_u64_fetch(kvm,
424 			    (unsigned long )uz.uz_fails);
425 			mtp->mt_sleeps = uz.uz_sleeps;
426 			if (kz.uk_flags & UMA_ZFLAG_INTERNAL)
427 				goto skip_percpu;
428 			for (i = 0; i < mp_maxid + 1; i++) {
429 				if (!CPU_ISSET(i, &all_cpus))
430 					continue;
431 				ucp = &ucp_array[i];
432 				mtp->mt_numallocs += ucp->uc_allocs;
433 				mtp->mt_numfrees += ucp->uc_frees;
434 
435 				if (ucp->uc_allocbucket != NULL) {
436 					ret = kread(kvm, ucp->uc_allocbucket,
437 					    &ub, sizeof(ub), 0);
438 					if (ret != 0) {
439 						free(ucp_array);
440 						_memstat_mtl_empty(list);
441 						list->mtl_error = ret;
442 						return (-1);
443 					}
444 					mtp->mt_free += ub.ub_cnt;
445 				}
446 				if (ucp->uc_freebucket != NULL) {
447 					ret = kread(kvm, ucp->uc_freebucket,
448 					    &ub, sizeof(ub), 0);
449 					if (ret != 0) {
450 						free(ucp_array);
451 						_memstat_mtl_empty(list);
452 						list->mtl_error = ret;
453 						return (-1);
454 					}
455 					mtp->mt_free += ub.ub_cnt;
456 				}
457 			}
458 skip_percpu:
459 			mtp->mt_size = kz.uk_size;
460 			mtp->mt_rsize = kz.uk_rsize;
461 			mtp->mt_memalloced = mtp->mt_numallocs * mtp->mt_size;
462 			mtp->mt_memfreed = mtp->mt_numfrees * mtp->mt_size;
463 			mtp->mt_bytes = mtp->mt_memalloced - mtp->mt_memfreed;
464 			mtp->mt_countlimit = uz.uz_max_items;
465 			mtp->mt_byteslimit = mtp->mt_countlimit * mtp->mt_size;
466 			mtp->mt_count = mtp->mt_numallocs - mtp->mt_numfrees;
467 			for (i = 0; i < ndomains; i++) {
468 				ret = kread(kvm, &uz.uz_domain[i], &uzd,
469 				   sizeof(uzd), 0);
470 				for (ubp =
471 				    LIST_FIRST(&uzd.uzd_buckets);
472 				    ubp != NULL;
473 				    ubp = LIST_NEXT(&ub, ub_link)) {
474 					ret = kread(kvm, ubp, &ub,
475 					   sizeof(ub), 0);
476 					mtp->mt_zonefree += ub.ub_cnt;
477 				}
478 			}
479 			if (!((kz.uk_flags & UMA_ZONE_SECONDARY) &&
480 			    LIST_FIRST(&kz.uk_zones) != uzp)) {
481 				mtp->mt_kegfree = kz.uk_free;
482 				mtp->mt_free += mtp->mt_kegfree;
483 			}
484 			mtp->mt_free += mtp->mt_zonefree;
485 		}
486 	}
487 	free(ucp_array);
488 	return (0);
489 }
490