1 /*
2 * Copyright (C) 2013 The Android Open Source Project
3 *
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
7 *
8 * http://www.apache.org/licenses/LICENSE-2.0
9 *
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
15 */
16
17 #include <errno.h>
18 #include <fcntl.h>
19 #include <getopt.h>
20 #include <inttypes.h>
21 #include <stdbool.h>
22 #include <stdint.h>
23 #include <stdlib.h>
24 #include <string.h>
25 #include <sys/types.h>
26 #include <unistd.h>
27
28 #include <pagemap/pagemap.h>
29
30 #define MAX_FILENAME 64
31
32 #define GROWTH_FACTOR 10
33
34 #define NO_PATTERN 0x100
35
36 #define PR_SORTED 1
37 #define PR_VERBOSE 2
38 #define PR_ALL 4
39
40 struct vaddr {
41 unsigned long addr;
42 size_t num_pages;
43 pid_t pid;
44 };
45
46 struct ksm_page {
47 uint64_t count;
48 uint32_t hash;
49 struct vaddr *vaddr;
50 size_t vaddr_len, vaddr_size;
51 size_t vaddr_count;
52 uint16_t pattern;
53 };
54
55 struct ksm_pages {
56 struct ksm_page *pages;
57 size_t len, size;
58 };
59
60 static void usage(char *myname);
61 static int getprocname(pid_t pid, char *buf, int len);
62 static int read_pages(struct ksm_pages *kp, pm_map_t **maps, size_t num_maps, uint8_t pr_flags);
63 static void print_pages(struct ksm_pages *kp, uint8_t pr_flags);
64 static void free_pages(struct ksm_pages *kp, uint8_t pr_flags);
65 static bool is_pattern(uint8_t *data, size_t len);
66 static int cmp_pages(const void *a, const void *b);
67 extern uint32_t hashword(const uint32_t *, size_t, int32_t);
68
main(int argc,char * argv[])69 int main(int argc, char *argv[]) {
70 pm_kernel_t *ker;
71 pm_process_t *proc;
72 pid_t *pids;
73 size_t num_procs;
74 size_t i;
75 pm_map_t **maps;
76 size_t num_maps;
77 char cmdline[256]; // this must be within the range of int
78 int error;
79 int rc = EXIT_SUCCESS;
80 uint8_t pr_flags = 0;
81 struct ksm_pages kp;
82
83 memset(&kp, 0, sizeof(kp));
84
85 opterr = 0;
86 do {
87 int c = getopt(argc, argv, "hvsa");
88 if (c == -1)
89 break;
90
91 switch (c) {
92 case 'a':
93 pr_flags |= PR_ALL;
94 break;
95 case 's':
96 pr_flags |= PR_SORTED;
97 break;
98 case 'v':
99 pr_flags |= PR_VERBOSE;
100 break;
101 case 'h':
102 usage(argv[0]);
103 exit(EXIT_SUCCESS);
104 case '?':
105 fprintf(stderr, "unknown option: %c\n", optopt);
106 usage(argv[0]);
107 exit(EXIT_FAILURE);
108 }
109 } while (1);
110
111 error = pm_kernel_create(&ker);
112 if (error) {
113 fprintf(stderr, "Error creating kernel interface -- "
114 "does this kernel have pagemap?\n");
115 exit(EXIT_FAILURE);
116 }
117
118 if (pr_flags & PR_ALL) {
119 error = pm_kernel_pids(ker, &pids, &num_procs);
120 if (error) {
121 fprintf(stderr, "Error listing processes.\n");
122 exit(EXIT_FAILURE);
123 }
124 } else {
125 if (optind != argc - 1) {
126 usage(argv[0]);
127 exit(EXIT_FAILURE);
128 }
129
130 pids = malloc(sizeof(*pids));
131 if (pids == NULL) {
132 fprintf(stderr, "Error allocating pid memory\n");
133 exit(EXIT_FAILURE);
134 }
135
136 *pids = strtoul(argv[optind], NULL, 10);
137 if (*pids == 0) {
138 fprintf(stderr, "Invalid PID\n");
139 rc = EXIT_FAILURE;
140 goto exit;
141 }
142 num_procs = 1;
143 if (getprocname(*pids, cmdline, sizeof(cmdline)) < 0) {
144 cmdline[0] = '\0';
145 }
146 printf("%s (%u):\n", cmdline, *pids);
147 }
148
149 printf("Warning: this tool only compares the KSM CRCs of pages, there is a chance of "
150 "collisions\n");
151
152 for (i = 0; i < num_procs; i++) {
153 error = pm_process_create(ker, pids[i], &proc);
154 if (error) {
155 fprintf(stderr, "warning: could not create process interface for %d\n", pids[i]);
156 rc = EXIT_FAILURE;
157 goto exit;
158 }
159
160 error = pm_process_maps(proc, &maps, &num_maps);
161 if (error) {
162 pm_process_destroy(proc);
163 fprintf(stderr, "warning: could not read process map for %d\n", pids[i]);
164 rc = EXIT_FAILURE;
165 goto exit;
166 }
167
168 if (read_pages(&kp, maps, num_maps, pr_flags) < 0) {
169 free(maps);
170 pm_process_destroy(proc);
171 rc = EXIT_FAILURE;
172 goto exit;
173 }
174
175 free(maps);
176 pm_process_destroy(proc);
177 }
178
179 if (pr_flags & PR_SORTED) {
180 qsort(kp.pages, kp.len, sizeof(*kp.pages), cmp_pages);
181 }
182 print_pages(&kp, pr_flags);
183
184 exit:
185 free_pages(&kp, pr_flags);
186 free(pids);
187 return rc;
188 }
189
read_pages(struct ksm_pages * kp,pm_map_t ** maps,size_t num_maps,uint8_t pr_flags)190 static int read_pages(struct ksm_pages *kp, pm_map_t **maps, size_t num_maps, uint8_t pr_flags) {
191 size_t i, j, k;
192 uint64_t *pagemap;
193 size_t map_len;
194 uint64_t flags;
195 pm_kernel_t *ker;
196 int error;
197 unsigned long vaddr;
198 int fd;
199 off_t off;
200 char filename[MAX_FILENAME];
201 uint32_t *data;
202 uint32_t hash;
203 int rc = 0;
204 struct ksm_page *cur_page;
205 pid_t pid;
206
207 if (num_maps == 0)
208 return 0;
209
210 pid = pm_process_pid(maps[0]->proc);
211 ker = maps[0]->proc->ker;
212 error = snprintf(filename, MAX_FILENAME, "/proc/%d/mem", pid);
213 if (error < 0 || error >= MAX_FILENAME) {
214 return -1;
215 }
216
217 data = malloc(pm_kernel_pagesize(ker));
218 if (data == NULL) {
219 fprintf(stderr, "warning: not enough memory to malloc data buffer\n");
220 return -1;
221 }
222
223 fd = open(filename, O_RDONLY);
224 if (fd < 0) {
225 fprintf(stderr, "warning: could not open %s\n", filename);
226 rc = -1;
227 goto err_open;
228 }
229
230 for (i = 0; i < num_maps; i++) {
231 error = pm_map_pagemap(maps[i], &pagemap, &map_len);
232 if (error) {
233 fprintf(stderr, "warning: could not read the pagemap of %d\n",
234 pm_process_pid(maps[i]->proc));
235 continue;
236 }
237 for (j = 0; j < map_len; j++) {
238 error = pm_kernel_flags(ker, PM_PAGEMAP_PFN(pagemap[j]), &flags);
239 if (error) {
240 fprintf(stderr, "warning: could not read flags for pfn at address 0x%016" PRIx64 "\n",
241 pagemap[i]);
242 continue;
243 }
244 if (!(flags & PM_PAGE_KSM)) {
245 continue;
246 }
247 vaddr = pm_map_start(maps[i]) + j * pm_kernel_pagesize(ker);
248 off = lseek(fd, vaddr, SEEK_SET);
249 if (off == (off_t)-1) {
250 fprintf(stderr, "warning: could not lseek to 0x%08lx\n", vaddr);
251 continue;
252 }
253 ssize_t len = read(fd, data, pm_kernel_pagesize(ker));
254 if (len != pm_kernel_pagesize(ker)) {
255 fprintf(stderr, "warning: could not read page at 0x%08lx\n", vaddr);
256 continue;
257 }
258
259 hash = hashword(data, pm_kernel_pagesize(ker) / sizeof(*data), 17);
260
261 for (k = 0; k < kp->len; k++) {
262 if (kp->pages[k].hash == hash) break;
263 }
264
265 if (k == kp->len) {
266 if (kp->len == kp->size) {
267 struct ksm_page *tmp = realloc(kp->pages,
268 (kp->size + GROWTH_FACTOR) * sizeof(*kp->pages));
269 if (tmp == NULL) {
270 fprintf(stderr, "warning: not enough memory to realloc pages struct\n");
271 free(pagemap);
272 rc = -1;
273 goto err_realloc;
274 }
275 memset(&tmp[k], 0, sizeof(tmp[k]) * GROWTH_FACTOR);
276 kp->pages = tmp;
277 kp->size += GROWTH_FACTOR;
278 }
279 rc = pm_kernel_count(ker, PM_PAGEMAP_PFN(pagemap[j]), &kp->pages[kp->len].count);
280 if (rc) {
281 fprintf(stderr, "error reading page count\n");
282 free(pagemap);
283 goto err_count;
284 }
285 kp->pages[kp->len].hash = hash;
286 kp->pages[kp->len].pattern =
287 is_pattern((uint8_t *)data, pm_kernel_pagesize(ker)) ?
288 (data[0] & 0xFF) : NO_PATTERN;
289 kp->len++;
290 }
291
292 cur_page = &kp->pages[k];
293
294 if (pr_flags & PR_VERBOSE) {
295 if (cur_page->vaddr_len > 0 &&
296 cur_page->vaddr[cur_page->vaddr_len - 1].pid == pid &&
297 cur_page->vaddr[cur_page->vaddr_len - 1].addr ==
298 vaddr - (cur_page->vaddr[cur_page->vaddr_len - 1].num_pages *
299 pm_kernel_pagesize(ker))) {
300 cur_page->vaddr[cur_page->vaddr_len - 1].num_pages++;
301 } else {
302 if (cur_page->vaddr_len == cur_page->vaddr_size) {
303 struct vaddr *tmp = realloc(cur_page->vaddr,
304 (cur_page->vaddr_size + GROWTH_FACTOR) * sizeof(*(cur_page->vaddr)));
305 if (tmp == NULL) {
306 fprintf(stderr, "warning: not enough memory to realloc vaddr array\n");
307 free(pagemap);
308 rc = -1;
309 goto err_realloc;
310 }
311 memset(&tmp[cur_page->vaddr_len], 0, sizeof(tmp[cur_page->vaddr_len]) * GROWTH_FACTOR);
312 cur_page->vaddr = tmp;
313 cur_page->vaddr_size += GROWTH_FACTOR;
314 }
315 cur_page->vaddr[cur_page->vaddr_len].addr = vaddr;
316 cur_page->vaddr[cur_page->vaddr_len].num_pages = 1;
317 cur_page->vaddr[cur_page->vaddr_len].pid = pid;
318 cur_page->vaddr_len++;
319 }
320 }
321 cur_page->vaddr_count++;
322 }
323 free(pagemap);
324 }
325 goto no_err;
326
327 err_realloc:
328 err_count:
329 if (pr_flags & PR_VERBOSE) {
330 for (i = 0; i < kp->len; i++) {
331 free(kp->pages[i].vaddr);
332 }
333 }
334 free(kp->pages);
335
336 no_err:
337 close(fd);
338 err_open:
339 free(data);
340 return rc;
341 }
342
print_pages(struct ksm_pages * kp,uint8_t pr_flags)343 static void print_pages(struct ksm_pages *kp, uint8_t pr_flags) {
344 size_t i, j, k;
345 char suffix[13];
346 int index;
347
348 for (i = 0; i < kp->len; i++) {
349 if (kp->pages[i].pattern != NO_PATTERN) {
350 printf("0x%02x byte pattern: ", kp->pages[i].pattern);
351 } else {
352 printf("KSM CRC 0x%08x:", kp->pages[i].hash);
353 }
354 printf(" %4zu page", kp->pages[i].vaddr_count);
355 if (kp->pages[i].vaddr_count > 1) {
356 printf("s");
357 }
358 if (!(pr_flags & PR_ALL)) {
359 printf(" (%" PRIu64 " reference", kp->pages[i].count);
360 if (kp->pages[i].count > 1) {
361 printf("s");
362 }
363 printf(")");
364 }
365 printf("\n");
366
367 if (pr_flags & PR_VERBOSE) {
368 j = 0;
369 while (j < kp->pages[i].vaddr_len) {
370 printf(" ");
371 for (k = 0; k < 8 && j < kp->pages[i].vaddr_len; k++, j++) {
372 printf(" 0x%08lx", kp->pages[i].vaddr[j].addr);
373
374 index = snprintf(suffix, sizeof(suffix), ":%zu",
375 kp->pages[i].vaddr[j].num_pages);
376 if (pr_flags & PR_ALL) {
377 index += snprintf(suffix + index, sizeof(suffix) - index, "[%d]",
378 kp->pages[i].vaddr[j].pid);
379 }
380 printf("%-12s", suffix);
381 }
382 printf("\n");
383 }
384 }
385 }
386 }
387
free_pages(struct ksm_pages * kp,uint8_t pr_flags)388 static void free_pages(struct ksm_pages *kp, uint8_t pr_flags) {
389 size_t i;
390
391 if (pr_flags & PR_VERBOSE) {
392 for (i = 0; i < kp->len; i++) {
393 free(kp->pages[i].vaddr);
394 }
395 }
396 free(kp->pages);
397 }
398
usage(char * myname)399 static void usage(char *myname) {
400 fprintf(stderr, "Usage: %s [-s | -v | -a | -h ] <pid>\n"
401 " -s Sort pages by usage count.\n"
402 " -v Verbose: print virtual addresses.\n"
403 " -a Display all the KSM pages in the system. Ignore the pid argument.\n"
404 " -h Display this help screen.\n",
405 myname);
406 }
407
cmp_pages(const void * a,const void * b)408 static int cmp_pages(const void *a, const void *b) {
409 const struct ksm_page *pg_a = a;
410 const struct ksm_page *pg_b = b;
411 int cmp = pg_b->vaddr_count - pg_a->vaddr_count;
412
413 return cmp ? cmp : pg_b->count - pg_a->count;
414 }
415
is_pattern(uint8_t * data,size_t len)416 static bool is_pattern(uint8_t *data, size_t len) {
417 size_t i;
418 uint8_t first_byte = data[0];
419
420 for (i = 1; i < len; i++) {
421 if (first_byte != data[i]) return false;
422 }
423
424 return true;
425 }
426
427 /*
428 * Get the process name for a given PID. Inserts the process name into buffer
429 * buf of length len. The size of the buffer must be greater than zero to get
430 * any useful output.
431 *
432 * Note that fgets(3) only declares length as an int, so our buffer size is
433 * also declared as an int.
434 *
435 * Returns 0 on success, a positive value on partial success, and -1 on
436 * failure. Other interesting values:
437 * 1 on failure to create string to examine proc cmdline entry
438 * 2 on failure to open proc cmdline entry
439 * 3 on failure to read proc cmdline entry
440 */
getprocname(pid_t pid,char * buf,int len)441 static int getprocname(pid_t pid, char *buf, int len) {
442 char *filename;
443 FILE *f;
444 int rc = 0;
445 static const char* unknown_cmdline = "<unknown>";
446
447 if (len <= 0) {
448 return -1;
449 }
450
451 if (asprintf(&filename, "/proc/%d/cmdline", (int)pid) < 0) {
452 rc = 1;
453 goto exit;
454 }
455
456 f = fopen(filename, "r");
457 if (f == NULL) {
458 rc = 2;
459 goto releasefilename;
460 }
461
462 if (fgets(buf, len, f) == NULL) {
463 rc = 3;
464 goto closefile;
465 }
466
467 closefile:
468 (void) fclose(f);
469 releasefilename:
470 free(filename);
471 exit:
472 if (rc != 0) {
473 /*
474 * The process went away before we could read its process name. Try
475 * to give the user "<unknown>" here, but otherwise they get to look
476 * at a blank.
477 */
478 if (strlcpy(buf, unknown_cmdline, (size_t)len) >= (size_t)len) {
479 rc = 4;
480 }
481 }
482
483 return rc;
484 }
485
486