xref: /freebsd/lib/libmemstat/memstat_uma.c (revision 7bd6fde3) 
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/sysctl.h>
31 
32 #define	LIBMEMSTAT	/* Cause vm_page.h not to include opt_vmpage.h */
33 #include <vm/vm.h>
34 #include <vm/vm_page.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 
48 #include "memstat.h"
49 #include "memstat_internal.h"
50 
51 static struct nlist namelist[] = {
52 #define	X_UMA_KEGS	0
53 	{ .n_name = "_uma_kegs" },
54 #define	X_MP_MAXID	1
55 	{ .n_name = "_mp_maxid" },
56 #define	X_ALL_CPUS	2
57 	{ .n_name = "_all_cpus" },
58 	{ .n_name = "" },
59 };
60 
61 /*
62  * Extract uma(9) statistics from the running kernel, and store all memory
63  * type information in the passed list.  For each type, check the list for an
64  * existing entry with the right name/allocator -- if present, update that
65  * entry.  Otherwise, add a new entry.  On error, the entire list will be
66  * cleared, as entries will be in an inconsistent state.
67  *
68  * To reduce the level of work for a list that starts empty, we keep around a
69  * hint as to whether it was empty when we began, so we can avoid searching
70  * the list for entries to update.  Updates are O(n^2) due to searching for
71  * each entry before adding it.
72  */
73 int
74 memstat_sysctl_uma(struct memory_type_list *list, int flags)
75 {
76 	struct uma_stream_header *ushp;
77 	struct uma_type_header *uthp;
78 	struct uma_percpu_stat *upsp;
79 	struct memory_type *mtp;
80 	int count, hint_dontsearch, i, j, maxcpus;
81 	char *buffer, *p;
82 	size_t size;
83 
84 	hint_dontsearch = LIST_EMPTY(&list->mtl_list);
85 
86 	/*
87 	 * Query the number of CPUs, number of malloc types so that we can
88 	 * guess an initial buffer size.  We loop until we succeed or really
89 	 * fail.  Note that the value of maxcpus we query using sysctl is not
90 	 * the version we use when processing the real data -- that is read
91 	 * from the header.
92 	 */
93 retry:
94 	size = sizeof(maxcpus);
95 	if (sysctlbyname("kern.smp.maxcpus", &maxcpus, &size, NULL, 0) < 0) {
96 		if (errno == EACCES || errno == EPERM)
97 			list->mtl_error = MEMSTAT_ERROR_PERMISSION;
98 		else
99 			list->mtl_error = MEMSTAT_ERROR_DATAERROR;
100 		return (-1);
101 	}
102 	if (size != sizeof(maxcpus)) {
103 		list->mtl_error = MEMSTAT_ERROR_DATAERROR;
104 		return (-1);
105 	}
106 
107 	if (maxcpus > MEMSTAT_MAXCPU) {
108 		list->mtl_error = MEMSTAT_ERROR_TOOMANYCPUS;
109 		return (-1);
110 	}
111 
112 	size = sizeof(count);
113 	if (sysctlbyname("vm.zone_count", &count, &size, NULL, 0) < 0) {
114 		if (errno == EACCES || errno == EPERM)
115 			list->mtl_error = MEMSTAT_ERROR_PERMISSION;
116 		else
117 			list->mtl_error = MEMSTAT_ERROR_VERSION;
118 		return (-1);
119 	}
120 	if (size != sizeof(count)) {
121 		list->mtl_error = MEMSTAT_ERROR_DATAERROR;
122 		return (-1);
123 	}
124 
125 	size = sizeof(*uthp) + count * (sizeof(*uthp) + sizeof(*upsp) *
126 	    maxcpus);
127 
128 	buffer = malloc(size);
129 	if (buffer == NULL) {
130 		list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
131 		return (-1);
132 	}
133 
134 	if (sysctlbyname("vm.zone_stats", buffer, &size, NULL, 0) < 0) {
135 		/*
136 		 * XXXRW: ENOMEM is an ambiguous return, we should bound the
137 		 * number of loops, perhaps.
138 		 */
139 		if (errno == ENOMEM) {
140 			free(buffer);
141 			goto retry;
142 		}
143 		if (errno == EACCES || errno == EPERM)
144 			list->mtl_error = MEMSTAT_ERROR_PERMISSION;
145 		else
146 			list->mtl_error = MEMSTAT_ERROR_VERSION;
147 		free(buffer);
148 		return (-1);
149 	}
150 
151 	if (size == 0) {
152 		free(buffer);
153 		return (0);
154 	}
155 
156 	if (size < sizeof(*ushp)) {
157 		list->mtl_error = MEMSTAT_ERROR_VERSION;
158 		free(buffer);
159 		return (-1);
160 	}
161 	p = buffer;
162 	ushp = (struct uma_stream_header *)p;
163 	p += sizeof(*ushp);
164 
165 	if (ushp->ush_version != UMA_STREAM_VERSION) {
166 		list->mtl_error = MEMSTAT_ERROR_VERSION;
167 		free(buffer);
168 		return (-1);
169 	}
170 
171 	if (ushp->ush_maxcpus > MEMSTAT_MAXCPU) {
172 		list->mtl_error = MEMSTAT_ERROR_TOOMANYCPUS;
173 		free(buffer);
174 		return (-1);
175 	}
176 
177 	/*
178 	 * For the remainder of this function, we are quite trusting about
179 	 * the layout of structures and sizes, since we've determined we have
180 	 * a matching version and acceptable CPU count.
181 	 */
182 	maxcpus = ushp->ush_maxcpus;
183 	count = ushp->ush_count;
184 	for (i = 0; i < count; i++) {
185 		uthp = (struct uma_type_header *)p;
186 		p += sizeof(*uthp);
187 
188 		if (hint_dontsearch == 0) {
189 			mtp = memstat_mtl_find(list, ALLOCATOR_UMA,
190 			    uthp->uth_name);
191 		} else
192 			mtp = NULL;
193 		if (mtp == NULL)
194 			mtp = _memstat_mt_allocate(list, ALLOCATOR_UMA,
195 			    uthp->uth_name);
196 		if (mtp == NULL) {
197 			_memstat_mtl_empty(list);
198 			free(buffer);
199 			list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
200 			return (-1);
201 		}
202 
203 		/*
204 		 * Reset the statistics on a current node.
205 		 */
206 		_memstat_mt_reset_stats(mtp);
207 
208 		mtp->mt_numallocs = uthp->uth_allocs;
209 		mtp->mt_numfrees = uthp->uth_frees;
210 		mtp->mt_failures = uthp->uth_fails;
211 
212 		for (j = 0; j < maxcpus; j++) {
213 			upsp = (struct uma_percpu_stat *)p;
214 			p += sizeof(*upsp);
215 
216 			mtp->mt_percpu_cache[j].mtp_free =
217 			    upsp->ups_cache_free;
218 			mtp->mt_free += upsp->ups_cache_free;
219 			mtp->mt_numallocs += upsp->ups_allocs;
220 			mtp->mt_numfrees += upsp->ups_frees;
221 		}
222 
223 		mtp->mt_size = uthp->uth_size;
224 		mtp->mt_memalloced = mtp->mt_numallocs * uthp->uth_size;
225 		mtp->mt_memfreed = mtp->mt_numfrees * uthp->uth_size;
226 		mtp->mt_bytes = mtp->mt_memalloced - mtp->mt_memfreed;
227 		mtp->mt_countlimit = uthp->uth_limit;
228 		mtp->mt_byteslimit = uthp->uth_limit * uthp->uth_size;
229 
230 		mtp->mt_count = mtp->mt_numallocs - mtp->mt_numfrees;
231 		mtp->mt_zonefree = uthp->uth_zone_free;
232 
233 		/*
234 		 * UMA secondary zones share a keg with the primary zone.  To
235 		 * avoid double-reporting of free items, report keg free
236 		 * items only in the primary zone.
237 		 */
238 		if (!(uthp->uth_zone_flags & UTH_ZONE_SECONDARY)) {
239 			mtp->mt_kegfree = uthp->uth_keg_free;
240 			mtp->mt_free += mtp->mt_kegfree;
241 		}
242 		mtp->mt_free += mtp->mt_zonefree;
243 	}
244 
245 	free(buffer);
246 
247 	return (0);
248 }
249 
250 static int
251 kread(kvm_t *kvm, void *kvm_pointer, void *address, size_t size,
252     size_t offset)
253 {
254 	ssize_t ret;
255 
256 	ret = kvm_read(kvm, (unsigned long)kvm_pointer + offset, address,
257 	    size);
258 	if (ret < 0)
259 		return (MEMSTAT_ERROR_KVM);
260 	if ((size_t)ret != size)
261 		return (MEMSTAT_ERROR_KVM_SHORTREAD);
262 	return (0);
263 }
264 
265 static int
266 kread_string(kvm_t *kvm, void *kvm_pointer, char *buffer, int buflen)
267 {
268 	ssize_t ret;
269 	int i;
270 
271 	for (i = 0; i < buflen; i++) {
272 		ret = kvm_read(kvm, (unsigned long)kvm_pointer + i,
273 		    &(buffer[i]), sizeof(char));
274 		if (ret < 0)
275 			return (MEMSTAT_ERROR_KVM);
276 		if ((size_t)ret != sizeof(char))
277 			return (MEMSTAT_ERROR_KVM_SHORTREAD);
278 		if (buffer[i] == '\0')
279 			return (0);
280 	}
281 	/* Truncate. */
282 	buffer[i-1] = '\0';
283 	return (0);
284 }
285 
286 static int
287 kread_symbol(kvm_t *kvm, int index, void *address, size_t size,
288     size_t offset)
289 {
290 	ssize_t ret;
291 
292 	ret = kvm_read(kvm, namelist[index].n_value + offset, address, size);
293 	if (ret < 0)
294 		return (MEMSTAT_ERROR_KVM);
295 	if ((size_t)ret != size)
296 		return (MEMSTAT_ERROR_KVM_SHORTREAD);
297 	return (0);
298 }
299 
300 /*
301  * memstat_kvm_uma() is similar to memstat_sysctl_uma(), only it extracts
302  * UMA(9) statistics from a kernel core/memory file.
303  */
304 int
305 memstat_kvm_uma(struct memory_type_list *list, void *kvm_handle)
306 {
307 	LIST_HEAD(, uma_keg) uma_kegs;
308 	struct memory_type *mtp;
309 	struct uma_bucket *ubp, ub;
310 	struct uma_cache *ucp, *ucp_array;
311 	struct uma_zone *uzp, uz;
312 	struct uma_keg *kzp, kz;
313 	int hint_dontsearch, i, mp_maxid, ret;
314 	char name[MEMTYPE_MAXNAME];
315 	__cpumask_t all_cpus;
316 	kvm_t *kvm;
317 
318 	kvm = (kvm_t *)kvm_handle;
319 	hint_dontsearch = LIST_EMPTY(&list->mtl_list);
320 	if (kvm_nlist(kvm, namelist) != 0) {
321 		list->mtl_error = MEMSTAT_ERROR_KVM;
322 		return (-1);
323 	}
324 	if (namelist[X_UMA_KEGS].n_type == 0 ||
325 	    namelist[X_UMA_KEGS].n_value == 0) {
326 		list->mtl_error = MEMSTAT_ERROR_KVM_NOSYMBOL;
327 		return (-1);
328 	}
329 	ret = kread_symbol(kvm, X_MP_MAXID, &mp_maxid, sizeof(mp_maxid), 0);
330 	if (ret != 0) {
331 		list->mtl_error = ret;
332 		return (-1);
333 	}
334 	ret = kread_symbol(kvm, X_UMA_KEGS, &uma_kegs, sizeof(uma_kegs), 0);
335 	if (ret != 0) {
336 		list->mtl_error = ret;
337 		return (-1);
338 	}
339 	ret = kread_symbol(kvm, X_ALL_CPUS, &all_cpus, sizeof(all_cpus), 0);
340 	if (ret != 0) {
341 		list->mtl_error = ret;
342 		return (-1);
343 	}
344 	ucp_array = malloc(sizeof(struct uma_cache) * (mp_maxid + 1));
345 	if (ucp_array == NULL) {
346 		list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
347 		return (-1);
348 	}
349 	for (kzp = LIST_FIRST(&uma_kegs); kzp != NULL; kzp =
350 	    LIST_NEXT(&kz, uk_link)) {
351 		ret = kread(kvm, kzp, &kz, sizeof(kz), 0);
352 		if (ret != 0) {
353 			free(ucp_array);
354 			_memstat_mtl_empty(list);
355 			list->mtl_error = ret;
356 			return (-1);
357 		}
358 		for (uzp = LIST_FIRST(&kz.uk_zones); uzp != NULL; uzp =
359 		    LIST_NEXT(&uz, uz_link)) {
360 			ret = kread(kvm, uzp, &uz, sizeof(uz), 0);
361 			if (ret != 0) {
362 				free(ucp_array);
363 				_memstat_mtl_empty(list);
364 				list->mtl_error = ret;
365 				return (-1);
366 			}
367 			ret = kread(kvm, uzp, ucp_array,
368 			    sizeof(struct uma_cache) * (mp_maxid + 1),
369 			    offsetof(struct uma_zone, uz_cpu[0]));
370 			if (ret != 0) {
371 				free(ucp_array);
372 				_memstat_mtl_empty(list);
373 				list->mtl_error = ret;
374 				return (-1);
375 			}
376 			ret = kread_string(kvm, uz.uz_name, name,
377 			    MEMTYPE_MAXNAME);
378 			if (ret != 0) {
379 				free(ucp_array);
380 				_memstat_mtl_empty(list);
381 				list->mtl_error = ret;
382 				return (-1);
383 			}
384 			if (hint_dontsearch == 0) {
385 				mtp = memstat_mtl_find(list, ALLOCATOR_UMA,
386 				    name);
387 			} else
388 				mtp = NULL;
389 			if (mtp == NULL)
390 				mtp = _memstat_mt_allocate(list, ALLOCATOR_UMA,
391 				    name);
392 			if (mtp == NULL) {
393 				free(ucp_array);
394 				_memstat_mtl_empty(list);
395 				list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
396 				return (-1);
397 			}
398 			/*
399 			 * Reset the statistics on a current node.
400 			 */
401 			_memstat_mt_reset_stats(mtp);
402 			mtp->mt_numallocs = uz.uz_allocs;
403 			mtp->mt_numfrees = uz.uz_frees;
404 			mtp->mt_failures = uz.uz_fails;
405 			if (kz.uk_flags & UMA_ZFLAG_INTERNAL)
406 				goto skip_percpu;
407 			for (i = 0; i < mp_maxid + 1; i++) {
408 				if ((all_cpus & (1 << i)) == 0)
409 					continue;
410 				ucp = &ucp_array[i];
411 				mtp->mt_numallocs += ucp->uc_allocs;
412 				mtp->mt_numfrees += ucp->uc_frees;
413 
414 				if (ucp->uc_allocbucket != NULL) {
415 					ret = kread(kvm, ucp->uc_allocbucket,
416 					    &ub, sizeof(ub), 0);
417 					if (ret != 0) {
418 						free(ucp_array);
419 						_memstat_mtl_empty(list);
420 						list->mtl_error = ret;
421 						return (-1);
422 					}
423 					mtp->mt_free += ub.ub_cnt;
424 				}
425 				if (ucp->uc_freebucket != NULL) {
426 					ret = kread(kvm, ucp->uc_freebucket,
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 			}
437 skip_percpu:
438 			mtp->mt_size = kz.uk_size;
439 			mtp->mt_memalloced = mtp->mt_numallocs * mtp->mt_size;
440 			mtp->mt_memfreed = mtp->mt_numfrees * mtp->mt_size;
441 			mtp->mt_bytes = mtp->mt_memalloced - mtp->mt_memfreed;
442 			if (kz.uk_ppera > 1)
443 				mtp->mt_countlimit = kz.uk_maxpages /
444 				    kz.uk_ipers;
445 			else
446 				mtp->mt_countlimit = kz.uk_maxpages *
447 				    kz.uk_ipers;
448 			mtp->mt_byteslimit = mtp->mt_countlimit * mtp->mt_size;
449 			mtp->mt_count = mtp->mt_numallocs - mtp->mt_numfrees;
450 			for (ubp = LIST_FIRST(&uz.uz_full_bucket); ubp !=
451 			    NULL; ubp = LIST_NEXT(&ub, ub_link)) {
452 				ret = kread(kvm, ubp, &ub, sizeof(ub), 0);
453 				mtp->mt_zonefree += ub.ub_cnt;
454 			}
455 			if (!((kz.uk_flags & UMA_ZONE_SECONDARY) &&
456 			    LIST_FIRST(&kz.uk_zones) != uzp)) {
457 				mtp->mt_kegfree = kz.uk_free;
458 				mtp->mt_free += mtp->mt_kegfree;
459 			}
460 			mtp->mt_free += mtp->mt_zonefree;
461 		}
462 	}
463 	free(ucp_array);
464 	return (0);
465 }
466