memstat_uma.c revision 209215 
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: head/lib/libmemstat/memstat_uma.c 209215 2010-06-15 19:28:37Z sbruno $
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
51static 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 */
73int
74memstat_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	 */
93retry:
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		mtp->mt_sleeps = uthp->uth_sleeps;
212
213		for (j = 0; j < maxcpus; j++) {
214			upsp = (struct uma_percpu_stat *)p;
215			p += sizeof(*upsp);
216
217			mtp->mt_percpu_cache[j].mtp_free =
218			    upsp->ups_cache_free;
219			mtp->mt_free += upsp->ups_cache_free;
220			mtp->mt_numallocs += upsp->ups_allocs;
221			mtp->mt_numfrees += upsp->ups_frees;
222		}
223
224		mtp->mt_size = uthp->uth_size;
225		mtp->mt_memalloced = mtp->mt_numallocs * uthp->uth_size;
226		mtp->mt_memfreed = mtp->mt_numfrees * uthp->uth_size;
227		mtp->mt_bytes = mtp->mt_memalloced - mtp->mt_memfreed;
228		mtp->mt_countlimit = uthp->uth_limit;
229		mtp->mt_byteslimit = uthp->uth_limit * uthp->uth_size;
230
231		mtp->mt_count = mtp->mt_numallocs - mtp->mt_numfrees;
232		mtp->mt_zonefree = uthp->uth_zone_free;
233
234		/*
235		 * UMA secondary zones share a keg with the primary zone.  To
236		 * avoid double-reporting of free items, report keg free
237		 * items only in the primary zone.
238		 */
239		if (!(uthp->uth_zone_flags & UTH_ZONE_SECONDARY)) {
240			mtp->mt_kegfree = uthp->uth_keg_free;
241			mtp->mt_free += mtp->mt_kegfree;
242		}
243		mtp->mt_free += mtp->mt_zonefree;
244	}
245
246	free(buffer);
247
248	return (0);
249}
250
251static int
252kread(kvm_t *kvm, void *kvm_pointer, void *address, size_t size,
253    size_t offset)
254{
255	ssize_t ret;
256
257	ret = kvm_read(kvm, (unsigned long)kvm_pointer + offset, address,
258	    size);
259	if (ret < 0)
260		return (MEMSTAT_ERROR_KVM);
261	if ((size_t)ret != size)
262		return (MEMSTAT_ERROR_KVM_SHORTREAD);
263	return (0);
264}
265
266static int
267kread_string(kvm_t *kvm, void *kvm_pointer, char *buffer, int buflen)
268{
269	ssize_t ret;
270	int i;
271
272	for (i = 0; i < buflen; i++) {
273		ret = kvm_read(kvm, (unsigned long)kvm_pointer + i,
274		    &(buffer[i]), sizeof(char));
275		if (ret < 0)
276			return (MEMSTAT_ERROR_KVM);
277		if ((size_t)ret != sizeof(char))
278			return (MEMSTAT_ERROR_KVM_SHORTREAD);
279		if (buffer[i] == '\0')
280			return (0);
281	}
282	/* Truncate. */
283	buffer[i-1] = '\0';
284	return (0);
285}
286
287static int
288kread_symbol(kvm_t *kvm, int index, void *address, size_t size,
289    size_t offset)
290{
291	ssize_t ret;
292
293	ret = kvm_read(kvm, namelist[index].n_value + offset, address, size);
294	if (ret < 0)
295		return (MEMSTAT_ERROR_KVM);
296	if ((size_t)ret != size)
297		return (MEMSTAT_ERROR_KVM_SHORTREAD);
298	return (0);
299}
300
301/*
302 * memstat_kvm_uma() is similar to memstat_sysctl_uma(), only it extracts
303 * UMA(9) statistics from a kernel core/memory file.
304 */
305int
306memstat_kvm_uma(struct memory_type_list *list, void *kvm_handle)
307{
308	LIST_HEAD(, uma_keg) uma_kegs;
309	struct memory_type *mtp;
310	struct uma_bucket *ubp, ub;
311	struct uma_cache *ucp, *ucp_array;
312	struct uma_zone *uzp, uz;
313	struct uma_keg *kzp, kz;
314	int hint_dontsearch, i, mp_maxid, ret;
315	char name[MEMTYPE_MAXNAME];
316	__cpumask_t all_cpus;
317	kvm_t *kvm;
318
319	kvm = (kvm_t *)kvm_handle;
320	hint_dontsearch = LIST_EMPTY(&list->mtl_list);
321	if (kvm_nlist(kvm, namelist) != 0) {
322		list->mtl_error = MEMSTAT_ERROR_KVM;
323		return (-1);
324	}
325	if (namelist[X_UMA_KEGS].n_type == 0 ||
326	    namelist[X_UMA_KEGS].n_value == 0) {
327		list->mtl_error = MEMSTAT_ERROR_KVM_NOSYMBOL;
328		return (-1);
329	}
330	ret = kread_symbol(kvm, X_MP_MAXID, &mp_maxid, sizeof(mp_maxid), 0);
331	if (ret != 0) {
332		list->mtl_error = ret;
333		return (-1);
334	}
335	ret = kread_symbol(kvm, X_UMA_KEGS, &uma_kegs, sizeof(uma_kegs), 0);
336	if (ret != 0) {
337		list->mtl_error = ret;
338		return (-1);
339	}
340	ret = kread_symbol(kvm, X_ALL_CPUS, &all_cpus, sizeof(all_cpus), 0);
341	if (ret != 0) {
342		list->mtl_error = ret;
343		return (-1);
344	}
345	ucp_array = malloc(sizeof(struct uma_cache) * (mp_maxid + 1));
346	if (ucp_array == NULL) {
347		list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
348		return (-1);
349	}
350	for (kzp = LIST_FIRST(&uma_kegs); kzp != NULL; kzp =
351	    LIST_NEXT(&kz, uk_link)) {
352		ret = kread(kvm, kzp, &kz, sizeof(kz), 0);
353		if (ret != 0) {
354			free(ucp_array);
355			_memstat_mtl_empty(list);
356			list->mtl_error = ret;
357			return (-1);
358		}
359		for (uzp = LIST_FIRST(&kz.uk_zones); uzp != NULL; uzp =
360		    LIST_NEXT(&uz, uz_link)) {
361			ret = kread(kvm, uzp, &uz, sizeof(uz), 0);
362			if (ret != 0) {
363				free(ucp_array);
364				_memstat_mtl_empty(list);
365				list->mtl_error = ret;
366				return (-1);
367			}
368			ret = kread(kvm, uzp, ucp_array,
369			    sizeof(struct uma_cache) * (mp_maxid + 1),
370			    offsetof(struct uma_zone, uz_cpu[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_string(kvm, uz.uz_name, name,
378			    MEMTYPE_MAXNAME);
379			if (ret != 0) {
380				free(ucp_array);
381				_memstat_mtl_empty(list);
382				list->mtl_error = ret;
383				return (-1);
384			}
385			if (hint_dontsearch == 0) {
386				mtp = memstat_mtl_find(list, ALLOCATOR_UMA,
387				    name);
388			} else
389				mtp = NULL;
390			if (mtp == NULL)
391				mtp = _memstat_mt_allocate(list, ALLOCATOR_UMA,
392				    name);
393			if (mtp == NULL) {
394				free(ucp_array);
395				_memstat_mtl_empty(list);
396				list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
397				return (-1);
398			}
399			/*
400			 * Reset the statistics on a current node.
401			 */
402			_memstat_mt_reset_stats(mtp);
403			mtp->mt_numallocs = uz.uz_allocs;
404			mtp->mt_numfrees = uz.uz_frees;
405			mtp->mt_failures = uz.uz_fails;
406			mtp->mt_sleeps = uz.uz_sleeps;
407			if (kz.uk_flags & UMA_ZFLAG_INTERNAL)
408				goto skip_percpu;
409			for (i = 0; i < mp_maxid + 1; i++) {
410				if ((all_cpus & (1 << i)) == 0)
411					continue;
412				ucp = &ucp_array[i];
413				mtp->mt_numallocs += ucp->uc_allocs;
414				mtp->mt_numfrees += ucp->uc_frees;
415
416				if (ucp->uc_allocbucket != NULL) {
417					ret = kread(kvm, ucp->uc_allocbucket,
418					    &ub, sizeof(ub), 0);
419					if (ret != 0) {
420						free(ucp_array);
421						_memstat_mtl_empty(list);
422						list->mtl_error = ret;
423						return (-1);
424					}
425					mtp->mt_free += ub.ub_cnt;
426				}
427				if (ucp->uc_freebucket != NULL) {
428					ret = kread(kvm, ucp->uc_freebucket,
429					    &ub, sizeof(ub), 0);
430					if (ret != 0) {
431						free(ucp_array);
432						_memstat_mtl_empty(list);
433						list->mtl_error = ret;
434						return (-1);
435					}
436					mtp->mt_free += ub.ub_cnt;
437				}
438			}
439skip_percpu:
440			mtp->mt_size = kz.uk_size;
441			mtp->mt_memalloced = mtp->mt_numallocs * mtp->mt_size;
442			mtp->mt_memfreed = mtp->mt_numfrees * mtp->mt_size;
443			mtp->mt_bytes = mtp->mt_memalloced - mtp->mt_memfreed;
444			if (kz.uk_ppera > 1)
445				mtp->mt_countlimit = kz.uk_maxpages /
446				    kz.uk_ipers;
447			else
448				mtp->mt_countlimit = kz.uk_maxpages *
449				    kz.uk_ipers;
450			mtp->mt_byteslimit = mtp->mt_countlimit * mtp->mt_size;
451			mtp->mt_count = mtp->mt_numallocs - mtp->mt_numfrees;
452			for (ubp = LIST_FIRST(&uz.uz_full_bucket); ubp !=
453			    NULL; ubp = LIST_NEXT(&ub, ub_link)) {
454				ret = kread(kvm, ubp, &ub, sizeof(ub), 0);
455				mtp->mt_zonefree += ub.ub_cnt;
456			}
457			if (!((kz.uk_flags & UMA_ZONE_SECONDARY) &&
458			    LIST_FIRST(&kz.uk_zones) != uzp)) {
459				mtp->mt_kegfree = kz.uk_free;
460				mtp->mt_free += mtp->mt_kegfree;
461			}
462			mtp->mt_free += mtp->mt_zonefree;
463		}
464	}
465	free(ucp_array);
466	return (0);
467}
468