projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge commit '33b458d276bb' into kvm-master
[deliverable/linux.git] / tools / vm / page-types.c
1 /*
2 * page-types: Tool for querying page flags
3 *
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of the GNU General Public License as published by the Free
6 * Software Foundation; version 2.
7 *
8 * This program is distributed in the hope that it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
11 * more details.
12 *
13 * You should find a copy of v2 of the GNU General Public License somewhere on
14 * your Linux system; if not, write to the Free Software Foundation, Inc., 59
15 * Temple Place, Suite 330, Boston, MA 02111-1307 USA.
16 *
17 * Copyright (C) 2009 Intel corporation
18 *
19 * Authors: Wu Fengguang <fengguang.wu@intel.com>
20 */
21
22 #define _FILE_OFFSET_BITS 64
23 #define _GNU_SOURCE
24 #include <stdio.h>
25 #include <stdlib.h>
26 #include <unistd.h>
27 #include <stdint.h>
28 #include <stdarg.h>
29 #include <string.h>
30 #include <getopt.h>
31 #include <limits.h>
32 #include <assert.h>
33 #include <ftw.h>
34 #include <time.h>
35 #include <setjmp.h>
36 #include <signal.h>
37 #include <sys/types.h>
38 #include <sys/errno.h>
39 #include <sys/fcntl.h>
40 #include <sys/mount.h>
41 #include <sys/statfs.h>
42 #include <sys/mman.h>
43 #include "../../include/uapi/linux/magic.h"
44 #include "../../include/uapi/linux/kernel-page-flags.h"
45 #include <api/fs/debugfs.h>
46
47 #ifndef MAX_PATH
48 # define MAX_PATH 256
49 #endif
50
51 #ifndef STR
52 # define _STR(x) #x
53 # define STR(x) _STR(x)
54 #endif
55
56 /*
57 * pagemap kernel ABI bits
58 */
59
60 #define PM_ENTRY_BYTES sizeof(uint64_t)
61 #define PM_STATUS_BITS 3
62 #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
63 #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
64 #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
65 #define PM_PSHIFT_BITS 6
66 #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
67 #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
68 #define __PM_PSHIFT(x) (((uint64_t) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
69 #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
70 #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
71
72 #define __PM_SOFT_DIRTY (1LL)
73 #define PM_PRESENT PM_STATUS(4LL)
74 #define PM_SWAP PM_STATUS(2LL)
75 #define PM_SOFT_DIRTY __PM_PSHIFT(__PM_SOFT_DIRTY)
76
77
78 /*
79 * kernel page flags
80 */
81
82 #define KPF_BYTES 8
83 #define PROC_KPAGEFLAGS "/proc/kpageflags"
84
85 /* [32-] kernel hacking assistances */
86 #define KPF_RESERVED 32
87 #define KPF_MLOCKED 33
88 #define KPF_MAPPEDTODISK 34
89 #define KPF_PRIVATE 35
90 #define KPF_PRIVATE_2 36
91 #define KPF_OWNER_PRIVATE 37
92 #define KPF_ARCH 38
93 #define KPF_UNCACHED 39
94 #define KPF_SOFTDIRTY 40
95
96 /* [48-] take some arbitrary free slots for expanding overloaded flags
97 * not part of kernel API
98 */
99 #define KPF_READAHEAD 48
100 #define KPF_SLOB_FREE 49
101 #define KPF_SLUB_FROZEN 50
102 #define KPF_SLUB_DEBUG 51
103
104 #define KPF_ALL_BITS ((uint64_t)~0ULL)
105 #define KPF_HACKERS_BITS (0xffffULL << 32)
106 #define KPF_OVERLOADED_BITS (0xffffULL << 48)
107 #define BIT(name) (1ULL << KPF_##name)
108 #define BITS_COMPOUND (BIT(COMPOUND_HEAD) | BIT(COMPOUND_TAIL))
109
110 static const char * const page_flag_names[] = {
111 [KPF_LOCKED] = "L:locked",
112 [KPF_ERROR] = "E:error",
113 [KPF_REFERENCED] = "R:referenced",
114 [KPF_UPTODATE] = "U:uptodate",
115 [KPF_DIRTY] = "D:dirty",
116 [KPF_LRU] = "l:lru",
117 [KPF_ACTIVE] = "A:active",
118 [KPF_SLAB] = "S:slab",
119 [KPF_WRITEBACK] = "W:writeback",
120 [KPF_RECLAIM] = "I:reclaim",
121 [KPF_BUDDY] = "B:buddy",
122
123 [KPF_MMAP] = "M:mmap",
124 [KPF_ANON] = "a:anonymous",
125 [KPF_SWAPCACHE] = "s:swapcache",
126 [KPF_SWAPBACKED] = "b:swapbacked",
127 [KPF_COMPOUND_HEAD] = "H:compound_head",
128 [KPF_COMPOUND_TAIL] = "T:compound_tail",
129 [KPF_HUGE] = "G:huge",
130 [KPF_UNEVICTABLE] = "u:unevictable",
131 [KPF_HWPOISON] = "X:hwpoison",
132 [KPF_NOPAGE] = "n:nopage",
133 [KPF_KSM] = "x:ksm",
134 [KPF_THP] = "t:thp",
135
136 [KPF_RESERVED] = "r:reserved",
137 [KPF_MLOCKED] = "m:mlocked",
138 [KPF_MAPPEDTODISK] = "d:mappedtodisk",
139 [KPF_PRIVATE] = "P:private",
140 [KPF_PRIVATE_2] = "p:private_2",
141 [KPF_OWNER_PRIVATE] = "O:owner_private",
142 [KPF_ARCH] = "h:arch",
143 [KPF_UNCACHED] = "c:uncached",
144 [KPF_SOFTDIRTY] = "f:softdirty",
145
146 [KPF_READAHEAD] = "I:readahead",
147 [KPF_SLOB_FREE] = "P:slob_free",
148 [KPF_SLUB_FROZEN] = "A:slub_frozen",
149 [KPF_SLUB_DEBUG] = "E:slub_debug",
150 };
151
152
153 static const char * const debugfs_known_mountpoints[] = {
154 "/sys/kernel/debug",
155 "/debug",
156 0,
157 };
158
159 /*
160 * data structures
161 */
162
163 static int opt_raw; /* for kernel developers */
164 static int opt_list; /* list pages (in ranges) */
165 static int opt_no_summary; /* don't show summary */
166 static pid_t opt_pid; /* process to walk */
167 const char * opt_file;
168
169 #define MAX_ADDR_RANGES 1024
170 static int nr_addr_ranges;
171 static unsigned long opt_offset[MAX_ADDR_RANGES];
172 static unsigned long opt_size[MAX_ADDR_RANGES];
173
174 #define MAX_VMAS 10240
175 static int nr_vmas;
176 static unsigned long pg_start[MAX_VMAS];
177 static unsigned long pg_end[MAX_VMAS];
178
179 #define MAX_BIT_FILTERS 64
180 static int nr_bit_filters;
181 static uint64_t opt_mask[MAX_BIT_FILTERS];
182 static uint64_t opt_bits[MAX_BIT_FILTERS];
183
184 static int page_size;
185
186 static int pagemap_fd;
187 static int kpageflags_fd;
188
189 static int opt_hwpoison;
190 static int opt_unpoison;
191
192 static char *hwpoison_debug_fs;
193 static int hwpoison_inject_fd;
194 static int hwpoison_forget_fd;
195
196 #define HASH_SHIFT 13
197 #define HASH_SIZE (1 << HASH_SHIFT)
198 #define HASH_MASK (HASH_SIZE - 1)
199 #define HASH_KEY(flags) (flags & HASH_MASK)
200
201 static unsigned long total_pages;
202 static unsigned long nr_pages[HASH_SIZE];
203 static uint64_t page_flags[HASH_SIZE];
204
205
206 /*
207 * helper functions
208 */
209
210 #define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
211
212 #define min_t(type, x, y) ({ \
213 type __min1 = (x); \
214 type __min2 = (y); \
215 __min1 < __min2 ? __min1 : __min2; })
216
217 #define max_t(type, x, y) ({ \
218 type __max1 = (x); \
219 type __max2 = (y); \
220 __max1 > __max2 ? __max1 : __max2; })
221
222 static unsigned long pages2mb(unsigned long pages)
223 {
224 return (pages * page_size) >> 20;
225 }
226
227 static void fatal(const char *x, ...)
228 {
229 va_list ap;
230
231 va_start(ap, x);
232 vfprintf(stderr, x, ap);
233 va_end(ap);
234 exit(EXIT_FAILURE);
235 }
236
237 static int checked_open(const char *pathname, int flags)
238 {
239 int fd = open(pathname, flags);
240
241 if (fd < 0) {
242 perror(pathname);
243 exit(EXIT_FAILURE);
244 }
245
246 return fd;
247 }
248
249 /*
250 * pagemap/kpageflags routines
251 */
252
253 static unsigned long do_u64_read(int fd, char *name,
254 uint64_t *buf,
255 unsigned long index,
256 unsigned long count)
257 {
258 long bytes;
259
260 if (index > ULONG_MAX / 8)
261 fatal("index overflow: %lu\n", index);
262
263 bytes = pread(fd, buf, count * 8, (off_t)index * 8);
264 if (bytes < 0) {
265 perror(name);
266 exit(EXIT_FAILURE);
267 }
268 if (bytes % 8)
269 fatal("partial read: %lu bytes\n", bytes);
270
271 return bytes / 8;
272 }
273
274 static unsigned long kpageflags_read(uint64_t *buf,
275 unsigned long index,
276 unsigned long pages)
277 {
278 return do_u64_read(kpageflags_fd, PROC_KPAGEFLAGS, buf, index, pages);
279 }
280
281 static unsigned long pagemap_read(uint64_t *buf,
282 unsigned long index,
283 unsigned long pages)
284 {
285 return do_u64_read(pagemap_fd, "/proc/pid/pagemap", buf, index, pages);
286 }
287
288 static unsigned long pagemap_pfn(uint64_t val)
289 {
290 unsigned long pfn;
291
292 if (val & PM_PRESENT)
293 pfn = PM_PFRAME(val);
294 else
295 pfn = 0;
296
297 return pfn;
298 }
299
300
301 /*
302 * page flag names
303 */
304
305 static char *page_flag_name(uint64_t flags)
306 {
307 static char buf[65];
308 int present;
309 size_t i, j;
310
311 for (i = 0, j = 0; i < ARRAY_SIZE(page_flag_names); i++) {
312 present = (flags >> i) & 1;
313 if (!page_flag_names[i]) {
314 if (present)
315 fatal("unknown flag bit %d\n", i);
316 continue;
317 }
318 buf[j++] = present ? page_flag_names[i][0] : '_';
319 }
320
321 return buf;
322 }
323
324 static char *page_flag_longname(uint64_t flags)
325 {
326 static char buf[1024];
327 size_t i, n;
328
329 for (i = 0, n = 0; i < ARRAY_SIZE(page_flag_names); i++) {
330 if (!page_flag_names[i])
331 continue;
332 if ((flags >> i) & 1)
333 n += snprintf(buf + n, sizeof(buf) - n, "%s,",
334 page_flag_names[i] + 2);
335 }
336 if (n)
337 n--;
338 buf[n] = '\0';
339
340 return buf;
341 }
342
343
344 /*
345 * page list and summary
346 */
347
348 static void show_page_range(unsigned long voffset, unsigned long offset,
349 unsigned long size, uint64_t flags)
350 {
351 static uint64_t flags0;
352 static unsigned long voff;
353 static unsigned long index;
354 static unsigned long count;
355
356 if (flags == flags0 && offset == index + count &&
357 size && voffset == voff + count) {
358 count += size;
359 return;
360 }
361
362 if (count) {
363 if (opt_pid)
364 printf("%lx\t", voff);
365 if (opt_file)
366 printf("%lu\t", voff);
367 printf("%lx\t%lx\t%s\n",
368 index, count, page_flag_name(flags0));
369 }
370
371 flags0 = flags;
372 index = offset;
373 voff = voffset;
374 count = size;
375 }
376
377 static void flush_page_range(void)
378 {
379 show_page_range(0, 0, 0, 0);
380 }
381
382 static void show_page(unsigned long voffset,
383 unsigned long offset, uint64_t flags)
384 {
385 if (opt_pid)
386 printf("%lx\t", voffset);
387 if (opt_file)
388 printf("%lu\t", voffset);
389 printf("%lx\t%s\n", offset, page_flag_name(flags));
390 }
391
392 static void show_summary(void)
393 {
394 size_t i;
395
396 printf(" flags\tpage-count MB"
397 " symbolic-flags\t\t\tlong-symbolic-flags\n");
398
399 for (i = 0; i < ARRAY_SIZE(nr_pages); i++) {
400 if (nr_pages[i])
401 printf("0x%016llx\t%10lu %8lu %s\t%s\n",
402 (unsigned long long)page_flags[i],
403 nr_pages[i],
404 pages2mb(nr_pages[i]),
405 page_flag_name(page_flags[i]),
406 page_flag_longname(page_flags[i]));
407 }
408
409 printf(" total\t%10lu %8lu\n",
410 total_pages, pages2mb(total_pages));
411 }
412
413
414 /*
415 * page flag filters
416 */
417
418 static int bit_mask_ok(uint64_t flags)
419 {
420 int i;
421
422 for (i = 0; i < nr_bit_filters; i++) {
423 if (opt_bits[i] == KPF_ALL_BITS) {
424 if ((flags & opt_mask[i]) == 0)
425 return 0;
426 } else {
427 if ((flags & opt_mask[i]) != opt_bits[i])
428 return 0;
429 }
430 }
431
432 return 1;
433 }
434
435 static uint64_t expand_overloaded_flags(uint64_t flags, uint64_t pme)
436 {
437 /* SLOB/SLUB overload several page flags */
438 if (flags & BIT(SLAB)) {
439 if (flags & BIT(PRIVATE))
440 flags ^= BIT(PRIVATE) | BIT(SLOB_FREE);
441 if (flags & BIT(ACTIVE))
442 flags ^= BIT(ACTIVE) | BIT(SLUB_FROZEN);
443 if (flags & BIT(ERROR))
444 flags ^= BIT(ERROR) | BIT(SLUB_DEBUG);
445 }
446
447 /* PG_reclaim is overloaded as PG_readahead in the read path */
448 if ((flags & (BIT(RECLAIM) | BIT(WRITEBACK))) == BIT(RECLAIM))
449 flags ^= BIT(RECLAIM) | BIT(READAHEAD);
450
451 if (pme & PM_SOFT_DIRTY)
452 flags |= BIT(SOFTDIRTY);
453
454 return flags;
455 }
456
457 static uint64_t well_known_flags(uint64_t flags)
458 {
459 /* hide flags intended only for kernel hacker */
460 flags &= ~KPF_HACKERS_BITS;
461
462 /* hide non-hugeTLB compound pages */
463 if ((flags & BITS_COMPOUND) && !(flags & BIT(HUGE)))
464 flags &= ~BITS_COMPOUND;
465
466 return flags;
467 }
468
469 static uint64_t kpageflags_flags(uint64_t flags, uint64_t pme)
470 {
471 if (opt_raw)
472 flags = expand_overloaded_flags(flags, pme);
473 else
474 flags = well_known_flags(flags);
475
476 return flags;
477 }
478
479 /*
480 * page actions
481 */
482
483 static void prepare_hwpoison_fd(void)
484 {
485 char buf[MAX_PATH + 1];
486
487 hwpoison_debug_fs = debugfs_mount(NULL);
488 if (!hwpoison_debug_fs) {
489 perror("mount debugfs");
490 exit(EXIT_FAILURE);
491 }
492
493 if (opt_hwpoison && !hwpoison_inject_fd) {
494 snprintf(buf, MAX_PATH, "%s/hwpoison/corrupt-pfn",
495 hwpoison_debug_fs);
496 hwpoison_inject_fd = checked_open(buf, O_WRONLY);
497 }
498
499 if (opt_unpoison && !hwpoison_forget_fd) {
500 snprintf(buf, MAX_PATH, "%s/hwpoison/unpoison-pfn",
501 hwpoison_debug_fs);
502 hwpoison_forget_fd = checked_open(buf, O_WRONLY);
503 }
504 }
505
506 static int hwpoison_page(unsigned long offset)
507 {
508 char buf[100];
509 int len;
510
511 len = sprintf(buf, "0x%lx\n", offset);
512 len = write(hwpoison_inject_fd, buf, len);
513 if (len < 0) {
514 perror("hwpoison inject");
515 return len;
516 }
517 return 0;
518 }
519
520 static int unpoison_page(unsigned long offset)
521 {
522 char buf[100];
523 int len;
524
525 len = sprintf(buf, "0x%lx\n", offset);
526 len = write(hwpoison_forget_fd, buf, len);
527 if (len < 0) {
528 perror("hwpoison forget");
529 return len;
530 }
531 return 0;
532 }
533
534 /*
535 * page frame walker
536 */
537
538 static size_t hash_slot(uint64_t flags)
539 {
540 size_t k = HASH_KEY(flags);
541 size_t i;
542
543 /* Explicitly reserve slot 0 for flags 0: the following logic
544 * cannot distinguish an unoccupied slot from slot (flags==0).
545 */
546 if (flags == 0)
547 return 0;
548
549 /* search through the remaining (HASH_SIZE-1) slots */
550 for (i = 1; i < ARRAY_SIZE(page_flags); i++, k++) {
551 if (!k || k >= ARRAY_SIZE(page_flags))
552 k = 1;
553 if (page_flags[k] == 0) {
554 page_flags[k] = flags;
555 return k;
556 }
557 if (page_flags[k] == flags)
558 return k;
559 }
560
561 fatal("hash table full: bump up HASH_SHIFT?\n");
562 exit(EXIT_FAILURE);
563 }
564
565 static void add_page(unsigned long voffset,
566 unsigned long offset, uint64_t flags, uint64_t pme)
567 {
568 flags = kpageflags_flags(flags, pme);
569
570 if (!bit_mask_ok(flags))
571 return;
572
573 if (opt_hwpoison)
574 hwpoison_page(offset);
575 if (opt_unpoison)
576 unpoison_page(offset);
577
578 if (opt_list == 1)
579 show_page_range(voffset, offset, 1, flags);
580 else if (opt_list == 2)
581 show_page(voffset, offset, flags);
582
583 nr_pages[hash_slot(flags)]++;
584 total_pages++;
585 }
586
587 #define KPAGEFLAGS_BATCH (64 << 10) /* 64k pages */
588 static void walk_pfn(unsigned long voffset,
589 unsigned long index,
590 unsigned long count,
591 uint64_t pme)
592 {
593 uint64_t buf[KPAGEFLAGS_BATCH];
594 unsigned long batch;
595 unsigned long pages;
596 unsigned long i;
597
598 while (count) {
599 batch = min_t(unsigned long, count, KPAGEFLAGS_BATCH);
600 pages = kpageflags_read(buf, index, batch);
601 if (pages == 0)
602 break;
603
604 for (i = 0; i < pages; i++)
605 add_page(voffset + i, index + i, buf[i], pme);
606
607 index += pages;
608 count -= pages;
609 }
610 }
611
612 #define PAGEMAP_BATCH (64 << 10)
613 static void walk_vma(unsigned long index, unsigned long count)
614 {
615 uint64_t buf[PAGEMAP_BATCH];
616 unsigned long batch;
617 unsigned long pages;
618 unsigned long pfn;
619 unsigned long i;
620
621 while (count) {
622 batch = min_t(unsigned long, count, PAGEMAP_BATCH);
623 pages = pagemap_read(buf, index, batch);
624 if (pages == 0)
625 break;
626
627 for (i = 0; i < pages; i++) {
628 pfn = pagemap_pfn(buf[i]);
629 if (pfn)
630 walk_pfn(index + i, pfn, 1, buf[i]);
631 }
632
633 index += pages;
634 count -= pages;
635 }
636 }
637
638 static void walk_task(unsigned long index, unsigned long count)
639 {
640 const unsigned long end = index + count;
641 unsigned long start;
642 int i = 0;
643
644 while (index < end) {
645
646 while (pg_end[i] <= index)
647 if (++i >= nr_vmas)
648 return;
649 if (pg_start[i] >= end)
650 return;
651
652 start = max_t(unsigned long, pg_start[i], index);
653 index = min_t(unsigned long, pg_end[i], end);
654
655 assert(start < index);
656 walk_vma(start, index - start);
657 }
658 }
659
660 static void add_addr_range(unsigned long offset, unsigned long size)
661 {
662 if (nr_addr_ranges >= MAX_ADDR_RANGES)
663 fatal("too many addr ranges\n");
664
665 opt_offset[nr_addr_ranges] = offset;
666 opt_size[nr_addr_ranges] = min_t(unsigned long, size, ULONG_MAX-offset);
667 nr_addr_ranges++;
668 }
669
670 static void walk_addr_ranges(void)
671 {
672 int i;
673
674 kpageflags_fd = checked_open(PROC_KPAGEFLAGS, O_RDONLY);
675
676 if (!nr_addr_ranges)
677 add_addr_range(0, ULONG_MAX);
678
679 for (i = 0; i < nr_addr_ranges; i++)
680 if (!opt_pid)
681 walk_pfn(opt_offset[i], opt_offset[i], opt_size[i], 0);
682 else
683 walk_task(opt_offset[i], opt_size[i]);
684
685 close(kpageflags_fd);
686 }
687
688
689 /*
690 * user interface
691 */
692
693 static const char *page_flag_type(uint64_t flag)
694 {
695 if (flag & KPF_HACKERS_BITS)
696 return "(r)";
697 if (flag & KPF_OVERLOADED_BITS)
698 return "(o)";
699 return " ";
700 }
701
702 static void usage(void)
703 {
704 size_t i, j;
705
706 printf(
707 "page-types [options]\n"
708 " -r|--raw Raw mode, for kernel developers\n"
709 " -d|--describe flags Describe flags\n"
710 " -a|--addr addr-spec Walk a range of pages\n"
711 " -b|--bits bits-spec Walk pages with specified bits\n"
712 " -p|--pid pid Walk process address space\n"
713 " -f|--file filename Walk file address space\n"
714 " -l|--list Show page details in ranges\n"
715 " -L|--list-each Show page details one by one\n"
716 " -N|--no-summary Don't show summary info\n"
717 " -X|--hwpoison hwpoison pages\n"
718 " -x|--unpoison unpoison pages\n"
719 " -h|--help Show this usage message\n"
720 "flags:\n"
721 " 0x10 bitfield format, e.g.\n"
722 " anon bit-name, e.g.\n"
723 " 0x10,anon comma-separated list, e.g.\n"
724 "addr-spec:\n"
725 " N one page at offset N (unit: pages)\n"
726 " N+M pages range from N to N+M-1\n"
727 " N,M pages range from N to M-1\n"
728 " N, pages range from N to end\n"
729 " ,M pages range from 0 to M-1\n"
730 "bits-spec:\n"
731 " bit1,bit2 (flags & (bit1|bit2)) != 0\n"
732 " bit1,bit2=bit1 (flags & (bit1|bit2)) == bit1\n"
733 " bit1,~bit2 (flags & (bit1|bit2)) == bit1\n"
734 " =bit1,bit2 flags == (bit1|bit2)\n"
735 "bit-names:\n"
736 );
737
738 for (i = 0, j = 0; i < ARRAY_SIZE(page_flag_names); i++) {
739 if (!page_flag_names[i])
740 continue;
741 printf("%16s%s", page_flag_names[i] + 2,
742 page_flag_type(1ULL << i));
743 if (++j > 3) {
744 j = 0;
745 putchar('\n');
746 }
747 }
748 printf("\n "
749 "(r) raw mode bits (o) overloaded bits\n");
750 }
751
752 static unsigned long long parse_number(const char *str)
753 {
754 unsigned long long n;
755
756 n = strtoll(str, NULL, 0);
757
758 if (n == 0 && str[0] != '0')
759 fatal("invalid name or number: %s\n", str);
760
761 return n;
762 }
763
764 static void parse_pid(const char *str)
765 {
766 FILE *file;
767 char buf[5000];
768
769 opt_pid = parse_number(str);
770
771 sprintf(buf, "/proc/%d/pagemap", opt_pid);
772 pagemap_fd = checked_open(buf, O_RDONLY);
773
774 sprintf(buf, "/proc/%d/maps", opt_pid);
775 file = fopen(buf, "r");
776 if (!file) {
777 perror(buf);
778 exit(EXIT_FAILURE);
779 }
780
781 while (fgets(buf, sizeof(buf), file) != NULL) {
782 unsigned long vm_start;
783 unsigned long vm_end;
784 unsigned long long pgoff;
785 int major, minor;
786 char r, w, x, s;
787 unsigned long ino;
788 int n;
789
790 n = sscanf(buf, "%lx-%lx %c%c%c%c %llx %x:%x %lu",
791 &vm_start,
792 &vm_end,
793 &r, &w, &x, &s,
794 &pgoff,
795 &major, &minor,
796 &ino);
797 if (n < 10) {
798 fprintf(stderr, "unexpected line: %s\n", buf);
799 continue;
800 }
801 pg_start[nr_vmas] = vm_start / page_size;
802 pg_end[nr_vmas] = vm_end / page_size;
803 if (++nr_vmas >= MAX_VMAS) {
804 fprintf(stderr, "too many VMAs\n");
805 break;
806 }
807 }
808 fclose(file);
809 }
810
811 static void show_file(const char *name, const struct stat *st)
812 {
813 unsigned long long size = st->st_size;
814 char atime[64], mtime[64];
815 long now = time(NULL);
816
817 printf("%s\tInode: %u\tSize: %llu (%llu pages)\n",
818 name, (unsigned)st->st_ino,
819 size, (size + page_size - 1) / page_size);
820
821 strftime(atime, sizeof(atime), "%c", localtime(&st->st_atime));
822 strftime(mtime, sizeof(mtime), "%c", localtime(&st->st_mtime));
823
824 printf("Modify: %s (%ld seconds ago)\nAccess: %s (%ld seconds ago)\n",
825 mtime, now - st->st_mtime,
826 atime, now - st->st_atime);
827 }
828
829 static sigjmp_buf sigbus_jmp;
830
831 static void * volatile sigbus_addr;
832
833 static void sigbus_handler(int sig, siginfo_t *info, void *ucontex)
834 {
835 (void)sig;
836 (void)ucontex;
837 sigbus_addr = info ? info->si_addr : NULL;
838 siglongjmp(sigbus_jmp, 1);
839 }
840
841 static struct sigaction sigbus_action = {
842 .sa_sigaction = sigbus_handler,
843 .sa_flags = SA_SIGINFO,
844 };
845
846 static void walk_file(const char *name, const struct stat *st)
847 {
848 uint8_t vec[PAGEMAP_BATCH];
849 uint64_t buf[PAGEMAP_BATCH], flags;
850 unsigned long nr_pages, pfn, i;
851 off_t off, end = st->st_size;
852 int fd;
853 ssize_t len;
854 void *ptr;
855 int first = 1;
856
857 fd = checked_open(name, O_RDONLY|O_NOATIME|O_NOFOLLOW);
858
859 for (off = 0; off < end; off += len) {
860 nr_pages = (end - off + page_size - 1) / page_size;
861 if (nr_pages > PAGEMAP_BATCH)
862 nr_pages = PAGEMAP_BATCH;
863 len = nr_pages * page_size;
864
865 ptr = mmap(NULL, len, PROT_READ, MAP_SHARED, fd, off);
866 if (ptr == MAP_FAILED)
867 fatal("mmap failed: %s", name);
868
869 /* determine cached pages */
870 if (mincore(ptr, len, vec))
871 fatal("mincore failed: %s", name);
872
873 /* turn off readahead */
874 if (madvise(ptr, len, MADV_RANDOM))
875 fatal("madvice failed: %s", name);
876
877 if (sigsetjmp(sigbus_jmp, 1)) {
878 end = off + sigbus_addr ? sigbus_addr - ptr : 0;
879 fprintf(stderr, "got sigbus at offset %lld: %s\n",
880 (long long)end, name);
881 goto got_sigbus;
882 }
883
884 /* populate ptes */
885 for (i = 0; i < nr_pages ; i++) {
886 if (vec[i] & 1)
887 (void)*(volatile int *)(ptr + i * page_size);
888 }
889 got_sigbus:
890
891 /* turn off harvesting reference bits */
892 if (madvise(ptr, len, MADV_SEQUENTIAL))
893 fatal("madvice failed: %s", name);
894
895 if (pagemap_read(buf, (unsigned long)ptr / page_size,
896 nr_pages) != nr_pages)
897 fatal("cannot read pagemap");
898
899 munmap(ptr, len);
900
901 for (i = 0; i < nr_pages; i++) {
902 pfn = pagemap_pfn(buf[i]);
903 if (!pfn)
904 continue;
905 if (!kpageflags_read(&flags, pfn, 1))
906 continue;
907 if (first && opt_list) {
908 first = 0;
909 flush_page_range();
910 show_file(name, st);
911 }
912 add_page(off / page_size + i, pfn, flags, buf[i]);
913 }
914 }
915
916 close(fd);
917 }
918
919 int walk_tree(const char *name, const struct stat *st, int type, struct FTW *f)
920 {
921 (void)f;
922 switch (type) {
923 case FTW_F:
924 if (S_ISREG(st->st_mode))
925 walk_file(name, st);
926 break;
927 case FTW_DNR:
928 fprintf(stderr, "cannot read dir: %s\n", name);
929 break;
930 }
931 return 0;
932 }
933
934 static void walk_page_cache(void)
935 {
936 struct stat st;
937
938 kpageflags_fd = checked_open(PROC_KPAGEFLAGS, O_RDONLY);
939 pagemap_fd = checked_open("/proc/self/pagemap", O_RDONLY);
940 sigaction(SIGBUS, &sigbus_action, NULL);
941
942 if (stat(opt_file, &st))
943 fatal("stat failed: %s\n", opt_file);
944
945 if (S_ISREG(st.st_mode)) {
946 walk_file(opt_file, &st);
947 } else if (S_ISDIR(st.st_mode)) {
948 /* do not follow symlinks and mountpoints */
949 if (nftw(opt_file, walk_tree, 64, FTW_MOUNT | FTW_PHYS) < 0)
950 fatal("nftw failed: %s\n", opt_file);
951 } else
952 fatal("unhandled file type: %s\n", opt_file);
953
954 close(kpageflags_fd);
955 close(pagemap_fd);
956 signal(SIGBUS, SIG_DFL);
957 }
958
959 static void parse_file(const char *name)
960 {
961 opt_file = name;
962 }
963
964 static void parse_addr_range(const char *optarg)
965 {
966 unsigned long offset;
967 unsigned long size;
968 char *p;
969
970 p = strchr(optarg, ',');
971 if (!p)
972 p = strchr(optarg, '+');
973
974 if (p == optarg) {
975 offset = 0;
976 size = parse_number(p + 1);
977 } else if (p) {
978 offset = parse_number(optarg);
979 if (p[1] == '\0')
980 size = ULONG_MAX;
981 else {
982 size = parse_number(p + 1);
983 if (*p == ',') {
984 if (size < offset)
985 fatal("invalid range: %lu,%lu\n",
986 offset, size);
987 size -= offset;
988 }
989 }
990 } else {
991 offset = parse_number(optarg);
992 size = 1;
993 }
994
995 add_addr_range(offset, size);
996 }
997
998 static void add_bits_filter(uint64_t mask, uint64_t bits)
999 {
1000 if (nr_bit_filters >= MAX_BIT_FILTERS)
1001 fatal("too much bit filters\n");
1002
1003 opt_mask[nr_bit_filters] = mask;
1004 opt_bits[nr_bit_filters] = bits;
1005 nr_bit_filters++;
1006 }
1007
1008 static uint64_t parse_flag_name(const char *str, int len)
1009 {
1010 size_t i;
1011
1012 if (!*str || !len)
1013 return 0;
1014
1015 if (len <= 8 && !strncmp(str, "compound", len))
1016 return BITS_COMPOUND;
1017
1018 for (i = 0; i < ARRAY_SIZE(page_flag_names); i++) {
1019 if (!page_flag_names[i])
1020 continue;
1021 if (!strncmp(str, page_flag_names[i] + 2, len))
1022 return 1ULL << i;
1023 }
1024
1025 return parse_number(str);
1026 }
1027
1028 static uint64_t parse_flag_names(const char *str, int all)
1029 {
1030 const char *p = str;
1031 uint64_t flags = 0;
1032
1033 while (1) {
1034 if (*p == ',' || *p == '=' || *p == '\0') {
1035 if ((*str != '~') || (*str == '~' && all && *++str))
1036 flags |= parse_flag_name(str, p - str);
1037 if (*p != ',')
1038 break;
1039 str = p + 1;
1040 }
1041 p++;
1042 }
1043
1044 return flags;
1045 }
1046
1047 static void parse_bits_mask(const char *optarg)
1048 {
1049 uint64_t mask;
1050 uint64_t bits;
1051 const char *p;
1052
1053 p = strchr(optarg, '=');
1054 if (p == optarg) {
1055 mask = KPF_ALL_BITS;
1056 bits = parse_flag_names(p + 1, 0);
1057 } else if (p) {
1058 mask = parse_flag_names(optarg, 0);
1059 bits = parse_flag_names(p + 1, 0);
1060 } else if (strchr(optarg, '~')) {
1061 mask = parse_flag_names(optarg, 1);
1062 bits = parse_flag_names(optarg, 0);
1063 } else {
1064 mask = parse_flag_names(optarg, 0);
1065 bits = KPF_ALL_BITS;
1066 }
1067
1068 add_bits_filter(mask, bits);
1069 }
1070
1071 static void describe_flags(const char *optarg)
1072 {
1073 uint64_t flags = parse_flag_names(optarg, 0);
1074
1075 printf("0x%016llx\t%s\t%s\n",
1076 (unsigned long long)flags,
1077 page_flag_name(flags),
1078 page_flag_longname(flags));
1079 }
1080
1081 static const struct option opts[] = {
1082 { "raw" , 0, NULL, 'r' },
1083 { "pid" , 1, NULL, 'p' },
1084 { "file" , 1, NULL, 'f' },
1085 { "addr" , 1, NULL, 'a' },
1086 { "bits" , 1, NULL, 'b' },
1087 { "describe" , 1, NULL, 'd' },
1088 { "list" , 0, NULL, 'l' },
1089 { "list-each" , 0, NULL, 'L' },
1090 { "no-summary", 0, NULL, 'N' },
1091 { "hwpoison" , 0, NULL, 'X' },
1092 { "unpoison" , 0, NULL, 'x' },
1093 { "help" , 0, NULL, 'h' },
1094 { NULL , 0, NULL, 0 }
1095 };
1096
1097 int main(int argc, char *argv[])
1098 {
1099 int c;
1100
1101 page_size = getpagesize();
1102
1103 while ((c = getopt_long(argc, argv,
1104 "rp:f:a:b:d:lLNXxh", opts, NULL)) != -1) {
1105 switch (c) {
1106 case 'r':
1107 opt_raw = 1;
1108 break;
1109 case 'p':
1110 parse_pid(optarg);
1111 break;
1112 case 'f':
1113 parse_file(optarg);
1114 break;
1115 case 'a':
1116 parse_addr_range(optarg);
1117 break;
1118 case 'b':
1119 parse_bits_mask(optarg);
1120 break;
1121 case 'd':
1122 describe_flags(optarg);
1123 exit(0);
1124 case 'l':
1125 opt_list = 1;
1126 break;
1127 case 'L':
1128 opt_list = 2;
1129 break;
1130 case 'N':
1131 opt_no_summary = 1;
1132 break;
1133 case 'X':
1134 opt_hwpoison = 1;
1135 prepare_hwpoison_fd();
1136 break;
1137 case 'x':
1138 opt_unpoison = 1;
1139 prepare_hwpoison_fd();
1140 break;
1141 case 'h':
1142 usage();
1143 exit(0);
1144 default:
1145 usage();
1146 exit(1);
1147 }
1148 }
1149
1150 if (opt_list && opt_pid)
1151 printf("voffset\t");
1152 if (opt_list && opt_file)
1153 printf("foffset\t");
1154 if (opt_list == 1)
1155 printf("offset\tlen\tflags\n");
1156 if (opt_list == 2)
1157 printf("offset\tflags\n");
1158
1159 if (opt_file)
1160 walk_page_cache();
1161 else
1162 walk_addr_ranges();
1163
1164 if (opt_list == 1)
1165 flush_page_range();
1166
1167 if (opt_no_summary)
1168 return 0;
1169
1170 if (opt_list)
1171 printf("\n\n");
1172
1173 show_summary();
1174
1175 return 0;
1176 }
Linux kernel - Deliverables
This page took 0.05656 seconds and 5 git commands to generate.