xref: /freebsd/lib/libmemstat/memstat_uma.c (revision b0b1dbdd) 
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 #include <vm/uma.h>
34 #include <vm/uma_int.h>
35 
36 #include <err.h>
37 #include <errno.h>
38 #include <kvm.h>
39 #include <nlist.h>
40 #include <stddef.h>
41 #include <stdio.h>
42 #include <stdlib.h>
43 #include <string.h>
44 #include <unistd.h>
45 
46 #include "memstat.h"
47 #include "memstat_internal.h"
48 
49 static struct nlist namelist[] = {
50 #define	X_UMA_KEGS	0
51 	{ .n_name = "_uma_kegs" },
52 #define	X_MP_MAXID	1
53 	{ .n_name = "_mp_maxid" },
54 #define	X_ALL_CPUS	2
55 	{ .n_name = "_all_cpus" },
56 	{ .n_name = "" },
57 };
58 
59 /*
60  * Extract uma(9) statistics from the running kernel, and store all memory
61  * type information in the passed list.  For each type, check the list for an
62  * existing entry with the right name/allocator -- if present, update that
63  * entry.  Otherwise, add a new entry.  On error, the entire list will be
64  * cleared, as entries will be in an inconsistent state.
65  *
66  * To reduce the level of work for a list that starts empty, we keep around a
67  * hint as to whether it was empty when we began, so we can avoid searching
68  * the list for entries to update.  Updates are O(n^2) due to searching for
69  * each entry before adding it.
70  */
71 int
72 memstat_sysctl_uma(struct memory_type_list *list, int flags)
73 {
74 	struct uma_stream_header *ushp;
75 	struct uma_type_header *uthp;
76 	struct uma_percpu_stat *upsp;
77 	struct memory_type *mtp;
78 	int count, hint_dontsearch, i, j, maxcpus, maxid;
79 	char *buffer, *p;
80 	size_t size;
81 
82 	hint_dontsearch = LIST_EMPTY(&list->mtl_list);
83 
84 	/*
85 	 * Query the number of CPUs, number of malloc types so that we can
86 	 * guess an initial buffer size.  We loop until we succeed or really
87 	 * fail.  Note that the value of maxcpus we query using sysctl is not
88 	 * the version we use when processing the real data -- that is read
89 	 * from the header.
90 	 */
91 retry:
92 	size = sizeof(maxid);
93 	if (sysctlbyname("kern.smp.maxid", &maxid, &size, NULL, 0) < 0) {
94 		if (errno == EACCES || errno == EPERM)
95 			list->mtl_error = MEMSTAT_ERROR_PERMISSION;
96 		else
97 			list->mtl_error = MEMSTAT_ERROR_DATAERROR;
98 		return (-1);
99 	}
100 	if (size != sizeof(maxid)) {
101 		list->mtl_error = MEMSTAT_ERROR_DATAERROR;
102 		return (-1);
103 	}
104 
105 	size = sizeof(count);
106 	if (sysctlbyname("vm.zone_count", &count, &size, NULL, 0) < 0) {
107 		if (errno == EACCES || errno == EPERM)
108 			list->mtl_error = MEMSTAT_ERROR_PERMISSION;
109 		else
110 			list->mtl_error = MEMSTAT_ERROR_VERSION;
111 		return (-1);
112 	}
113 	if (size != sizeof(count)) {
114 		list->mtl_error = MEMSTAT_ERROR_DATAERROR;
115 		return (-1);
116 	}
117 
118 	size = sizeof(*uthp) + count * (sizeof(*uthp) + sizeof(*upsp) *
119 	    (maxid + 1));
120 
121 	buffer = malloc(size);
122 	if (buffer == NULL) {
123 		list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
124 		return (-1);
125 	}
126 
127 	if (sysctlbyname("vm.zone_stats", buffer, &size, NULL, 0) < 0) {
128 		/*
129 		 * XXXRW: ENOMEM is an ambiguous return, we should bound the
130 		 * number of loops, perhaps.
131 		 */
132 		if (errno == ENOMEM) {
133 			free(buffer);
134 			goto retry;
135 		}
136 		if (errno == EACCES || errno == EPERM)
137 			list->mtl_error = MEMSTAT_ERROR_PERMISSION;
138 		else
139 			list->mtl_error = MEMSTAT_ERROR_VERSION;
140 		free(buffer);
141 		return (-1);
142 	}
143 
144 	if (size == 0) {
145 		free(buffer);
146 		return (0);
147 	}
148 
149 	if (size < sizeof(*ushp)) {
150 		list->mtl_error = MEMSTAT_ERROR_VERSION;
151 		free(buffer);
152 		return (-1);
153 	}
154 	p = buffer;
155 	ushp = (struct uma_stream_header *)p;
156 	p += sizeof(*ushp);
157 
158 	if (ushp->ush_version != UMA_STREAM_VERSION) {
159 		list->mtl_error = MEMSTAT_ERROR_VERSION;
160 		free(buffer);
161 		return (-1);
162 	}
163 
164 	/*
165 	 * For the remainder of this function, we are quite trusting about
166 	 * the layout of structures and sizes, since we've determined we have
167 	 * a matching version and acceptable CPU count.
168 	 */
169 	maxcpus = ushp->ush_maxcpus;
170 	count = ushp->ush_count;
171 	for (i = 0; i < count; i++) {
172 		uthp = (struct uma_type_header *)p;
173 		p += sizeof(*uthp);
174 
175 		if (hint_dontsearch == 0) {
176 			mtp = memstat_mtl_find(list, ALLOCATOR_UMA,
177 			    uthp->uth_name);
178 		} else
179 			mtp = NULL;
180 		if (mtp == NULL)
181 			mtp = _memstat_mt_allocate(list, ALLOCATOR_UMA,
182 			    uthp->uth_name, maxid + 1);
183 		if (mtp == NULL) {
184 			_memstat_mtl_empty(list);
185 			free(buffer);
186 			list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
187 			return (-1);
188 		}
189 
190 		/*
191 		 * Reset the statistics on a current node.
192 		 */
193 		_memstat_mt_reset_stats(mtp, maxid + 1);
194 
195 		mtp->mt_numallocs = uthp->uth_allocs;
196 		mtp->mt_numfrees = uthp->uth_frees;
197 		mtp->mt_failures = uthp->uth_fails;
198 		mtp->mt_sleeps = uthp->uth_sleeps;
199 
200 		for (j = 0; j < maxcpus; j++) {
201 			upsp = (struct uma_percpu_stat *)p;
202 			p += sizeof(*upsp);
203 
204 			mtp->mt_percpu_cache[j].mtp_free =
205 			    upsp->ups_cache_free;
206 			mtp->mt_free += upsp->ups_cache_free;
207 			mtp->mt_numallocs += upsp->ups_allocs;
208 			mtp->mt_numfrees += upsp->ups_frees;
209 		}
210 
211 		mtp->mt_size = uthp->uth_size;
212 		mtp->mt_rsize = uthp->uth_rsize;
213 		mtp->mt_memalloced = mtp->mt_numallocs * uthp->uth_size;
214 		mtp->mt_memfreed = mtp->mt_numfrees * uthp->uth_size;
215 		mtp->mt_bytes = mtp->mt_memalloced - mtp->mt_memfreed;
216 		mtp->mt_countlimit = uthp->uth_limit;
217 		mtp->mt_byteslimit = uthp->uth_limit * uthp->uth_size;
218 
219 		mtp->mt_count = mtp->mt_numallocs - mtp->mt_numfrees;
220 		mtp->mt_zonefree = uthp->uth_zone_free;
221 
222 		/*
223 		 * UMA secondary zones share a keg with the primary zone.  To
224 		 * avoid double-reporting of free items, report keg free
225 		 * items only in the primary zone.
226 		 */
227 		if (!(uthp->uth_zone_flags & UTH_ZONE_SECONDARY)) {
228 			mtp->mt_kegfree = uthp->uth_keg_free;
229 			mtp->mt_free += mtp->mt_kegfree;
230 		}
231 		mtp->mt_free += mtp->mt_zonefree;
232 	}
233 
234 	free(buffer);
235 
236 	return (0);
237 }
238 
239 static int
240 kread(kvm_t *kvm, void *kvm_pointer, void *address, size_t size,
241     size_t offset)
242 {
243 	ssize_t ret;
244 
245 	ret = kvm_read(kvm, (unsigned long)kvm_pointer + offset, address,
246 	    size);
247 	if (ret < 0)
248 		return (MEMSTAT_ERROR_KVM);
249 	if ((size_t)ret != size)
250 		return (MEMSTAT_ERROR_KVM_SHORTREAD);
251 	return (0);
252 }
253 
254 static int
255 kread_string(kvm_t *kvm, const void *kvm_pointer, char *buffer, int buflen)
256 {
257 	ssize_t ret;
258 	int i;
259 
260 	for (i = 0; i < buflen; i++) {
261 		ret = kvm_read(kvm, (unsigned long)kvm_pointer + i,
262 		    &(buffer[i]), sizeof(char));
263 		if (ret < 0)
264 			return (MEMSTAT_ERROR_KVM);
265 		if ((size_t)ret != sizeof(char))
266 			return (MEMSTAT_ERROR_KVM_SHORTREAD);
267 		if (buffer[i] == '\0')
268 			return (0);
269 	}
270 	/* Truncate. */
271 	buffer[i-1] = '\0';
272 	return (0);
273 }
274 
275 static int
276 kread_symbol(kvm_t *kvm, int index, void *address, size_t size,
277     size_t offset)
278 {
279 	ssize_t ret;
280 
281 	ret = kvm_read(kvm, namelist[index].n_value + offset, address, size);
282 	if (ret < 0)
283 		return (MEMSTAT_ERROR_KVM);
284 	if ((size_t)ret != size)
285 		return (MEMSTAT_ERROR_KVM_SHORTREAD);
286 	return (0);
287 }
288 
289 /*
290  * memstat_kvm_uma() is similar to memstat_sysctl_uma(), only it extracts
291  * UMA(9) statistics from a kernel core/memory file.
292  */
293 int
294 memstat_kvm_uma(struct memory_type_list *list, void *kvm_handle)
295 {
296 	LIST_HEAD(, uma_keg) uma_kegs;
297 	struct memory_type *mtp;
298 	struct uma_bucket *ubp, ub;
299 	struct uma_cache *ucp, *ucp_array;
300 	struct uma_zone *uzp, uz;
301 	struct uma_keg *kzp, kz;
302 	int hint_dontsearch, i, mp_maxid, ret;
303 	char name[MEMTYPE_MAXNAME];
304 	cpuset_t all_cpus;
305 	long cpusetsize;
306 	kvm_t *kvm;
307 
308 	kvm = (kvm_t *)kvm_handle;
309 	hint_dontsearch = LIST_EMPTY(&list->mtl_list);
310 	if (kvm_nlist(kvm, namelist) != 0) {
311 		list->mtl_error = MEMSTAT_ERROR_KVM;
312 		return (-1);
313 	}
314 	if (namelist[X_UMA_KEGS].n_type == 0 ||
315 	    namelist[X_UMA_KEGS].n_value == 0) {
316 		list->mtl_error = MEMSTAT_ERROR_KVM_NOSYMBOL;
317 		return (-1);
318 	}
319 	ret = kread_symbol(kvm, X_MP_MAXID, &mp_maxid, sizeof(mp_maxid), 0);
320 	if (ret != 0) {
321 		list->mtl_error = ret;
322 		return (-1);
323 	}
324 	ret = kread_symbol(kvm, X_UMA_KEGS, &uma_kegs, sizeof(uma_kegs), 0);
325 	if (ret != 0) {
326 		list->mtl_error = ret;
327 		return (-1);
328 	}
329 	cpusetsize = sysconf(_SC_CPUSET_SIZE);
330 	if (cpusetsize == -1 || (u_long)cpusetsize > sizeof(cpuset_t)) {
331 		list->mtl_error = MEMSTAT_ERROR_KVM_NOSYMBOL;
332 		return (-1);
333 	}
334 	CPU_ZERO(&all_cpus);
335 	ret = kread_symbol(kvm, X_ALL_CPUS, &all_cpus, cpusetsize, 0);
336 	if (ret != 0) {
337 		list->mtl_error = ret;
338 		return (-1);
339 	}
340 	ucp_array = malloc(sizeof(struct uma_cache) * (mp_maxid + 1));
341 	if (ucp_array == NULL) {
342 		list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
343 		return (-1);
344 	}
345 	for (kzp = LIST_FIRST(&uma_kegs); kzp != NULL; kzp =
346 	    LIST_NEXT(&kz, uk_link)) {
347 		ret = kread(kvm, kzp, &kz, sizeof(kz), 0);
348 		if (ret != 0) {
349 			free(ucp_array);
350 			_memstat_mtl_empty(list);
351 			list->mtl_error = ret;
352 			return (-1);
353 		}
354 		for (uzp = LIST_FIRST(&kz.uk_zones); uzp != NULL; uzp =
355 		    LIST_NEXT(&uz, uz_link)) {
356 			ret = kread(kvm, uzp, &uz, sizeof(uz), 0);
357 			if (ret != 0) {
358 				free(ucp_array);
359 				_memstat_mtl_empty(list);
360 				list->mtl_error = ret;
361 				return (-1);
362 			}
363 			ret = kread(kvm, uzp, ucp_array,
364 			    sizeof(struct uma_cache) * (mp_maxid + 1),
365 			    offsetof(struct uma_zone, uz_cpu[0]));
366 			if (ret != 0) {
367 				free(ucp_array);
368 				_memstat_mtl_empty(list);
369 				list->mtl_error = ret;
370 				return (-1);
371 			}
372 			ret = kread_string(kvm, uz.uz_name, name,
373 			    MEMTYPE_MAXNAME);
374 			if (ret != 0) {
375 				free(ucp_array);
376 				_memstat_mtl_empty(list);
377 				list->mtl_error = ret;
378 				return (-1);
379 			}
380 			if (hint_dontsearch == 0) {
381 				mtp = memstat_mtl_find(list, ALLOCATOR_UMA,
382 				    name);
383 			} else
384 				mtp = NULL;
385 			if (mtp == NULL)
386 				mtp = _memstat_mt_allocate(list, ALLOCATOR_UMA,
387 				    name, mp_maxid + 1);
388 			if (mtp == NULL) {
389 				free(ucp_array);
390 				_memstat_mtl_empty(list);
391 				list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
392 				return (-1);
393 			}
394 			/*
395 			 * Reset the statistics on a current node.
396 			 */
397 			_memstat_mt_reset_stats(mtp, mp_maxid + 1);
398 			mtp->mt_numallocs = uz.uz_allocs;
399 			mtp->mt_numfrees = uz.uz_frees;
400 			mtp->mt_failures = uz.uz_fails;
401 			mtp->mt_sleeps = uz.uz_sleeps;
402 			if (kz.uk_flags & UMA_ZFLAG_INTERNAL)
403 				goto skip_percpu;
404 			for (i = 0; i < mp_maxid + 1; i++) {
405 				if (!CPU_ISSET(i, &all_cpus))
406 					continue;
407 				ucp = &ucp_array[i];
408 				mtp->mt_numallocs += ucp->uc_allocs;
409 				mtp->mt_numfrees += ucp->uc_frees;
410 
411 				if (ucp->uc_allocbucket != NULL) {
412 					ret = kread(kvm, ucp->uc_allocbucket,
413 					    &ub, sizeof(ub), 0);
414 					if (ret != 0) {
415 						free(ucp_array);
416 						_memstat_mtl_empty(list);
417 						list->mtl_error = ret;
418 						return (-1);
419 					}
420 					mtp->mt_free += ub.ub_cnt;
421 				}
422 				if (ucp->uc_freebucket != NULL) {
423 					ret = kread(kvm, ucp->uc_freebucket,
424 					    &ub, sizeof(ub), 0);
425 					if (ret != 0) {
426 						free(ucp_array);
427 						_memstat_mtl_empty(list);
428 						list->mtl_error = ret;
429 						return (-1);
430 					}
431 					mtp->mt_free += ub.ub_cnt;
432 				}
433 			}
434 skip_percpu:
435 			mtp->mt_size = kz.uk_size;
436 			mtp->mt_rsize = kz.uk_rsize;
437 			mtp->mt_memalloced = mtp->mt_numallocs * mtp->mt_size;
438 			mtp->mt_memfreed = mtp->mt_numfrees * mtp->mt_size;
439 			mtp->mt_bytes = mtp->mt_memalloced - mtp->mt_memfreed;
440 			if (kz.uk_ppera > 1)
441 				mtp->mt_countlimit = kz.uk_maxpages /
442 				    kz.uk_ipers;
443 			else
444 				mtp->mt_countlimit = kz.uk_maxpages *
445 				    kz.uk_ipers;
446 			mtp->mt_byteslimit = mtp->mt_countlimit * mtp->mt_size;
447 			mtp->mt_count = mtp->mt_numallocs - mtp->mt_numfrees;
448 			for (ubp = LIST_FIRST(&uz.uz_buckets); ubp !=
449 			    NULL; ubp = LIST_NEXT(&ub, ub_link)) {
450 				ret = kread(kvm, ubp, &ub, sizeof(ub), 0);
451 				mtp->mt_zonefree += ub.ub_cnt;
452 			}
453 			if (!((kz.uk_flags & UMA_ZONE_SECONDARY) &&
454 			    LIST_FIRST(&kz.uk_zones) != uzp)) {
455 				mtp->mt_kegfree = kz.uk_free;
456 				mtp->mt_free += mtp->mt_kegfree;
457 			}
458 			mtp->mt_free += mtp->mt_zonefree;
459 		}
460 	}
461 	free(ucp_array);
462 	return (0);
463 }
464