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 #include <sys/_task.h> 33 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, maxid; 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(maxid); 97 if (sysctlbyname("kern.smp.maxid", &maxid, &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(maxid)) { 105 list->mtl_error = MEMSTAT_ERROR_DATAERROR; 106 return (-1); 107 } 108 109 size = sizeof(count); 110 if (sysctlbyname("vm.zone_count", &count, &size, NULL, 0) < 0) { 111 if (errno == EACCES || errno == EPERM) 112 list->mtl_error = MEMSTAT_ERROR_PERMISSION; 113 else 114 list->mtl_error = MEMSTAT_ERROR_VERSION; 115 return (-1); 116 } 117 if (size != sizeof(count)) { 118 list->mtl_error = MEMSTAT_ERROR_DATAERROR; 119 return (-1); 120 } 121 122 size = sizeof(*uthp) + count * (sizeof(*uthp) + sizeof(*upsp) * 123 (maxid + 1)); 124 125 buffer = malloc(size); 126 if (buffer == NULL) { 127 list->mtl_error = MEMSTAT_ERROR_NOMEMORY; 128 return (-1); 129 } 130 131 if (sysctlbyname("vm.zone_stats", buffer, &size, NULL, 0) < 0) { 132 /* 133 * XXXRW: ENOMEM is an ambiguous return, we should bound the 134 * number of loops, perhaps. 135 */ 136 if (errno == ENOMEM) { 137 free(buffer); 138 goto retry; 139 } 140 if (errno == EACCES || errno == EPERM) 141 list->mtl_error = MEMSTAT_ERROR_PERMISSION; 142 else 143 list->mtl_error = MEMSTAT_ERROR_VERSION; 144 free(buffer); 145 return (-1); 146 } 147 148 if (size == 0) { 149 free(buffer); 150 return (0); 151 } 152 153 if (size < sizeof(*ushp)) { 154 list->mtl_error = MEMSTAT_ERROR_VERSION; 155 free(buffer); 156 return (-1); 157 } 158 p = buffer; 159 ushp = (struct uma_stream_header *)p; 160 p += sizeof(*ushp); 161 162 if (ushp->ush_version != UMA_STREAM_VERSION) { 163 list->mtl_error = MEMSTAT_ERROR_VERSION; 164 free(buffer); 165 return (-1); 166 } 167 168 /* 169 * For the remainder of this function, we are quite trusting about 170 * the layout of structures and sizes, since we've determined we have 171 * a matching version and acceptable CPU count. 172 */ 173 maxcpus = ushp->ush_maxcpus; 174 count = ushp->ush_count; 175 for (i = 0; i < count; i++) { 176 uthp = (struct uma_type_header *)p; 177 p += sizeof(*uthp); 178 179 if (hint_dontsearch == 0) { 180 mtp = memstat_mtl_find(list, ALLOCATOR_UMA, 181 uthp->uth_name); 182 } else 183 mtp = NULL; 184 if (mtp == NULL) 185 mtp = _memstat_mt_allocate(list, ALLOCATOR_UMA, 186 uthp->uth_name, maxid + 1); 187 if (mtp == NULL) { 188 _memstat_mtl_empty(list); 189 free(buffer); 190 list->mtl_error = MEMSTAT_ERROR_NOMEMORY; 191 return (-1); 192 } 193 194 /* 195 * Reset the statistics on a current node. 196 */ 197 _memstat_mt_reset_stats(mtp, maxid + 1); 198 199 mtp->mt_numallocs = uthp->uth_allocs; 200 mtp->mt_numfrees = uthp->uth_frees; 201 mtp->mt_failures = uthp->uth_fails; 202 mtp->mt_sleeps = uthp->uth_sleeps; 203 204 for (j = 0; j < maxcpus; j++) { 205 upsp = (struct uma_percpu_stat *)p; 206 p += sizeof(*upsp); 207 208 mtp->mt_percpu_cache[j].mtp_free = 209 upsp->ups_cache_free; 210 mtp->mt_free += upsp->ups_cache_free; 211 mtp->mt_numallocs += upsp->ups_allocs; 212 mtp->mt_numfrees += upsp->ups_frees; 213 } 214 215 mtp->mt_size = uthp->uth_size; 216 mtp->mt_rsize = uthp->uth_rsize; 217 mtp->mt_memalloced = mtp->mt_numallocs * uthp->uth_size; 218 mtp->mt_memfreed = mtp->mt_numfrees * uthp->uth_size; 219 mtp->mt_bytes = mtp->mt_memalloced - mtp->mt_memfreed; 220 mtp->mt_countlimit = uthp->uth_limit; 221 mtp->mt_byteslimit = uthp->uth_limit * uthp->uth_size; 222 223 mtp->mt_count = mtp->mt_numallocs - mtp->mt_numfrees; 224 mtp->mt_zonefree = uthp->uth_zone_free; 225 226 /* 227 * UMA secondary zones share a keg with the primary zone. To 228 * avoid double-reporting of free items, report keg free 229 * items only in the primary zone. 230 */ 231 if (!(uthp->uth_zone_flags & UTH_ZONE_SECONDARY)) { 232 mtp->mt_kegfree = uthp->uth_keg_free; 233 mtp->mt_free += mtp->mt_kegfree; 234 } 235 mtp->mt_free += mtp->mt_zonefree; 236 } 237 238 free(buffer); 239 240 return (0); 241 } 242 243 static int 244 kread(kvm_t *kvm, void *kvm_pointer, void *address, size_t size, 245 size_t offset) 246 { 247 ssize_t ret; 248 249 ret = kvm_read(kvm, (unsigned long)kvm_pointer + offset, address, 250 size); 251 if (ret < 0) 252 return (MEMSTAT_ERROR_KVM); 253 if ((size_t)ret != size) 254 return (MEMSTAT_ERROR_KVM_SHORTREAD); 255 return (0); 256 } 257 258 static int 259 kread_string(kvm_t *kvm, const void *kvm_pointer, char *buffer, int buflen) 260 { 261 ssize_t ret; 262 int i; 263 264 for (i = 0; i < buflen; i++) { 265 ret = kvm_read(kvm, (unsigned long)kvm_pointer + i, 266 &(buffer[i]), sizeof(char)); 267 if (ret < 0) 268 return (MEMSTAT_ERROR_KVM); 269 if ((size_t)ret != sizeof(char)) 270 return (MEMSTAT_ERROR_KVM_SHORTREAD); 271 if (buffer[i] == '\0') 272 return (0); 273 } 274 /* Truncate. */ 275 buffer[i-1] = '\0'; 276 return (0); 277 } 278 279 static int 280 kread_symbol(kvm_t *kvm, int index, void *address, size_t size, 281 size_t offset) 282 { 283 ssize_t ret; 284 285 ret = kvm_read(kvm, namelist[index].n_value + offset, address, size); 286 if (ret < 0) 287 return (MEMSTAT_ERROR_KVM); 288 if ((size_t)ret != size) 289 return (MEMSTAT_ERROR_KVM_SHORTREAD); 290 return (0); 291 } 292 293 /* 294 * memstat_kvm_uma() is similar to memstat_sysctl_uma(), only it extracts 295 * UMA(9) statistics from a kernel core/memory file. 296 */ 297 int 298 memstat_kvm_uma(struct memory_type_list *list, void *kvm_handle) 299 { 300 LIST_HEAD(, uma_keg) uma_kegs; 301 struct memory_type *mtp; 302 struct uma_bucket *ubp, ub; 303 struct uma_cache *ucp, *ucp_array; 304 struct uma_zone *uzp, uz; 305 struct uma_keg *kzp, kz; 306 int hint_dontsearch, i, mp_maxid, ret; 307 char name[MEMTYPE_MAXNAME]; 308 cpuset_t all_cpus; 309 long cpusetsize; 310 kvm_t *kvm; 311 312 kvm = (kvm_t *)kvm_handle; 313 hint_dontsearch = LIST_EMPTY(&list->mtl_list); 314 if (kvm_nlist(kvm, namelist) != 0) { 315 list->mtl_error = MEMSTAT_ERROR_KVM; 316 return (-1); 317 } 318 if (namelist[X_UMA_KEGS].n_type == 0 || 319 namelist[X_UMA_KEGS].n_value == 0) { 320 list->mtl_error = MEMSTAT_ERROR_KVM_NOSYMBOL; 321 return (-1); 322 } 323 ret = kread_symbol(kvm, X_MP_MAXID, &mp_maxid, sizeof(mp_maxid), 0); 324 if (ret != 0) { 325 list->mtl_error = ret; 326 return (-1); 327 } 328 ret = kread_symbol(kvm, X_UMA_KEGS, &uma_kegs, sizeof(uma_kegs), 0); 329 if (ret != 0) { 330 list->mtl_error = ret; 331 return (-1); 332 } 333 cpusetsize = sysconf(_SC_CPUSET_SIZE); 334 if (cpusetsize == -1 || (u_long)cpusetsize > sizeof(cpuset_t)) { 335 list->mtl_error = MEMSTAT_ERROR_KVM_NOSYMBOL; 336 return (-1); 337 } 338 CPU_ZERO(&all_cpus); 339 ret = kread_symbol(kvm, X_ALL_CPUS, &all_cpus, cpusetsize, 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, mp_maxid + 1); 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, mp_maxid + 1); 402 mtp->mt_numallocs = uz.uz_allocs; 403 mtp->mt_numfrees = uz.uz_frees; 404 mtp->mt_failures = uz.uz_fails; 405 mtp->mt_sleeps = uz.uz_sleeps; 406 if (kz.uk_flags & UMA_ZFLAG_INTERNAL) 407 goto skip_percpu; 408 for (i = 0; i < mp_maxid + 1; i++) { 409 if (!CPU_ISSET(i, &all_cpus)) 410 continue; 411 ucp = &ucp_array[i]; 412 mtp->mt_numallocs += ucp->uc_allocs; 413 mtp->mt_numfrees += ucp->uc_frees; 414 415 if (ucp->uc_allocbucket != NULL) { 416 ret = kread(kvm, ucp->uc_allocbucket, 417 &ub, sizeof(ub), 0); 418 if (ret != 0) { 419 free(ucp_array); 420 _memstat_mtl_empty(list); 421 list->mtl_error = ret; 422 return (-1); 423 } 424 mtp->mt_free += ub.ub_cnt; 425 } 426 if (ucp->uc_freebucket != NULL) { 427 ret = kread(kvm, ucp->uc_freebucket, 428 &ub, sizeof(ub), 0); 429 if (ret != 0) { 430 free(ucp_array); 431 _memstat_mtl_empty(list); 432 list->mtl_error = ret; 433 return (-1); 434 } 435 mtp->mt_free += ub.ub_cnt; 436 } 437 } 438 skip_percpu: 439 mtp->mt_size = kz.uk_size; 440 mtp->mt_rsize = kz.uk_rsize; 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_buckets); 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