xref: /freebsd/sys/kern/kern_dump.c (revision a466cc55373fc3cf86837f09da729535b57e69a1)
1 /*-
2  * Copyright (c) 2002 Marcel Moolenaar
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  *
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
19  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25  */
26 
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
29 
30 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/conf.h>
33 #include <sys/cons.h>
34 #include <sys/kdb.h>
35 #include <sys/kernel.h>
36 #include <sys/kerneldump.h>
37 #include <sys/malloc.h>
38 #include <sys/msgbuf.h>
39 #include <sys/proc.h>
40 #include <sys/watchdog.h>
41 
42 #include <vm/vm.h>
43 #include <vm/vm_param.h>
44 #include <vm/vm_page.h>
45 #include <vm/vm_phys.h>
46 #include <vm/vm_dumpset.h>
47 #include <vm/pmap.h>
48 
49 #include <machine/dump.h>
50 #include <machine/elf.h>
51 #include <machine/md_var.h>
52 #include <machine/pcb.h>
53 
54 CTASSERT(sizeof(struct kerneldumpheader) == 512);
55 
56 #define	MD_ALIGN(x)	roundup2((off_t)(x), PAGE_SIZE)
57 
58 /* Handle buffered writes. */
59 static size_t fragsz;
60 
61 struct dump_pa dump_map[DUMPSYS_MD_PA_NPAIRS];
62 
63 #if !defined(__powerpc__)
64 void
65 dumpsys_gen_pa_init(void)
66 {
67 	int n, idx;
68 
69 	bzero(dump_map, sizeof(dump_map));
70 	for (n = 0; n < nitems(dump_map); n++) {
71 		idx = n * 2;
72 		if (dump_avail[idx] == 0 && dump_avail[idx + 1] == 0)
73 			break;
74 		dump_map[n].pa_start = dump_avail[idx];
75 		dump_map[n].pa_size = dump_avail[idx + 1] - dump_avail[idx];
76 	}
77 }
78 #endif
79 
80 struct dump_pa *
81 dumpsys_gen_pa_next(struct dump_pa *mdp)
82 {
83 
84 	if (mdp == NULL)
85 		return (&dump_map[0]);
86 
87 	mdp++;
88 	if (mdp->pa_size == 0)
89 		mdp = NULL;
90 	return (mdp);
91 }
92 
93 void
94 dumpsys_gen_wbinv_all(void)
95 {
96 
97 }
98 
99 void
100 dumpsys_gen_unmap_chunk(vm_paddr_t pa __unused, size_t chunk __unused,
101     void *va __unused)
102 {
103 
104 }
105 
106 int
107 dumpsys_gen_write_aux_headers(struct dumperinfo *di)
108 {
109 
110 	return (0);
111 }
112 
113 int
114 dumpsys_buf_seek(struct dumperinfo *di, size_t sz)
115 {
116 	static uint8_t buf[DEV_BSIZE];
117 	size_t nbytes;
118 	int error;
119 
120 	bzero(buf, sizeof(buf));
121 
122 	while (sz > 0) {
123 		nbytes = MIN(sz, sizeof(buf));
124 
125 		error = dump_append(di, buf, nbytes);
126 		if (error)
127 			return (error);
128 		sz -= nbytes;
129 	}
130 
131 	return (0);
132 }
133 
134 int
135 dumpsys_buf_write(struct dumperinfo *di, char *ptr, size_t sz)
136 {
137 	size_t len;
138 	int error;
139 
140 	while (sz) {
141 		len = di->blocksize - fragsz;
142 		if (len > sz)
143 			len = sz;
144 		memcpy((char *)di->blockbuf + fragsz, ptr, len);
145 		fragsz += len;
146 		ptr += len;
147 		sz -= len;
148 		if (fragsz == di->blocksize) {
149 			error = dump_append(di, di->blockbuf, di->blocksize);
150 			if (error)
151 				return (error);
152 			fragsz = 0;
153 		}
154 	}
155 	return (0);
156 }
157 
158 int
159 dumpsys_buf_flush(struct dumperinfo *di)
160 {
161 	int error;
162 
163 	if (fragsz == 0)
164 		return (0);
165 
166 	error = dump_append(di, di->blockbuf, di->blocksize);
167 	fragsz = 0;
168 	return (error);
169 }
170 
171 CTASSERT(PAGE_SHIFT < 20);
172 #define PG2MB(pgs) ((pgs + (1 << (20 - PAGE_SHIFT)) - 1) >> (20 - PAGE_SHIFT))
173 
174 int
175 dumpsys_cb_dumpdata(struct dump_pa *mdp, int seqnr, void *arg)
176 {
177 	struct dumperinfo *di = (struct dumperinfo*)arg;
178 	vm_paddr_t pa;
179 	void *va;
180 	uint64_t pgs;
181 	size_t counter, sz, chunk;
182 	int c, error;
183 	u_int maxdumppgs;
184 
185 	error = 0;	/* catch case in which chunk size is 0 */
186 	counter = 0;	/* Update twiddle every 16MB */
187 	va = NULL;
188 	pgs = mdp->pa_size / PAGE_SIZE;
189 	pa = mdp->pa_start;
190 	maxdumppgs = min(di->maxiosize / PAGE_SIZE, MAXDUMPPGS);
191 	if (maxdumppgs == 0)	/* seatbelt */
192 		maxdumppgs = 1;
193 
194 	printf("  chunk %d: %juMB (%ju pages)", seqnr, (uintmax_t)PG2MB(pgs),
195 	    (uintmax_t)pgs);
196 
197 	dumpsys_wbinv_all();
198 	while (pgs) {
199 		chunk = pgs;
200 		if (chunk > maxdumppgs)
201 			chunk = maxdumppgs;
202 		sz = chunk << PAGE_SHIFT;
203 		counter += sz;
204 		if (counter >> 24) {
205 			printf(" %ju", (uintmax_t)PG2MB(pgs));
206 			counter &= (1 << 24) - 1;
207 		}
208 
209 		dumpsys_map_chunk(pa, chunk, &va);
210 		wdog_kern_pat(WD_LASTVAL);
211 
212 		error = dump_append(di, va, sz);
213 		dumpsys_unmap_chunk(pa, chunk, va);
214 		if (error)
215 			break;
216 		pgs -= chunk;
217 		pa += sz;
218 
219 		/* Check for user abort. */
220 		c = cncheckc();
221 		if (c == 0x03)
222 			return (ECANCELED);
223 		if (c != -1)
224 			printf(" (CTRL-C to abort) ");
225 	}
226 	printf(" ... %s\n", (error) ? "fail" : "ok");
227 	return (error);
228 }
229 
230 int
231 dumpsys_foreach_chunk(dumpsys_callback_t cb, void *arg)
232 {
233 	struct dump_pa *mdp;
234 	int error, seqnr;
235 
236 	seqnr = 0;
237 	mdp = dumpsys_pa_next(NULL);
238 	while (mdp != NULL) {
239 		error = (*cb)(mdp, seqnr++, arg);
240 		if (error)
241 			return (-error);
242 		mdp = dumpsys_pa_next(mdp);
243 	}
244 	return (seqnr);
245 }
246 
247 static off_t fileofs;
248 
249 static int
250 cb_dumphdr(struct dump_pa *mdp, int seqnr, void *arg)
251 {
252 	struct dumperinfo *di = (struct dumperinfo*)arg;
253 	Elf_Phdr phdr;
254 	uint64_t size;
255 	int error;
256 
257 	size = mdp->pa_size;
258 	bzero(&phdr, sizeof(phdr));
259 	phdr.p_type = PT_LOAD;
260 	phdr.p_flags = PF_R;			/* XXX */
261 	phdr.p_offset = fileofs;
262 #ifdef __powerpc__
263 	phdr.p_vaddr = (do_minidump? mdp->pa_start : ~0L);
264 	phdr.p_paddr = (do_minidump? ~0L : mdp->pa_start);
265 #else
266 	phdr.p_vaddr = mdp->pa_start;
267 	phdr.p_paddr = mdp->pa_start;
268 #endif
269 	phdr.p_filesz = size;
270 	phdr.p_memsz = size;
271 	phdr.p_align = PAGE_SIZE;
272 
273 	error = dumpsys_buf_write(di, (char*)&phdr, sizeof(phdr));
274 	fileofs += phdr.p_filesz;
275 	return (error);
276 }
277 
278 static int
279 cb_size(struct dump_pa *mdp, int seqnr, void *arg)
280 {
281 	uint64_t *sz;
282 
283 	sz = (uint64_t *)arg;
284 	*sz += (uint64_t)mdp->pa_size;
285 	return (0);
286 }
287 
288 int
289 dumpsys_generic(struct dumperinfo *di)
290 {
291 	static struct kerneldumpheader kdh;
292 	Elf_Ehdr ehdr;
293 	uint64_t dumpsize;
294 	off_t hdrgap;
295 	size_t hdrsz;
296 	int error;
297 
298 #if MINIDUMP_PAGE_TRACKING == 1
299 	if (do_minidump)
300 		return (minidumpsys(di, false));
301 #endif
302 
303 	bzero(&ehdr, sizeof(ehdr));
304 	ehdr.e_ident[EI_MAG0] = ELFMAG0;
305 	ehdr.e_ident[EI_MAG1] = ELFMAG1;
306 	ehdr.e_ident[EI_MAG2] = ELFMAG2;
307 	ehdr.e_ident[EI_MAG3] = ELFMAG3;
308 	ehdr.e_ident[EI_CLASS] = ELF_CLASS;
309 #if BYTE_ORDER == LITTLE_ENDIAN
310 	ehdr.e_ident[EI_DATA] = ELFDATA2LSB;
311 #else
312 	ehdr.e_ident[EI_DATA] = ELFDATA2MSB;
313 #endif
314 	ehdr.e_ident[EI_VERSION] = EV_CURRENT;
315 	ehdr.e_ident[EI_OSABI] = ELFOSABI_STANDALONE;	/* XXX big picture? */
316 	ehdr.e_type = ET_CORE;
317 	ehdr.e_machine = EM_VALUE;
318 	ehdr.e_phoff = sizeof(ehdr);
319 	ehdr.e_flags = 0;
320 	ehdr.e_ehsize = sizeof(ehdr);
321 	ehdr.e_phentsize = sizeof(Elf_Phdr);
322 	ehdr.e_shentsize = sizeof(Elf_Shdr);
323 
324 	dumpsys_pa_init();
325 
326 	/* Calculate dump size. */
327 	dumpsize = 0L;
328 	ehdr.e_phnum = dumpsys_foreach_chunk(cb_size, &dumpsize) +
329 	    DUMPSYS_NUM_AUX_HDRS;
330 	hdrsz = ehdr.e_phoff + ehdr.e_phnum * ehdr.e_phentsize;
331 	fileofs = MD_ALIGN(hdrsz);
332 	dumpsize += fileofs;
333 	hdrgap = fileofs - roundup2((off_t)hdrsz, di->blocksize);
334 
335 	dump_init_header(di, &kdh, KERNELDUMPMAGIC, KERNELDUMP_ARCH_VERSION,
336 	    dumpsize);
337 
338 	error = dump_start(di, &kdh);
339 	if (error != 0)
340 		goto fail;
341 
342 	printf("Dumping %ju MB (%d chunks)\n", (uintmax_t)dumpsize >> 20,
343 	    ehdr.e_phnum - DUMPSYS_NUM_AUX_HDRS);
344 
345 	/* Dump ELF header */
346 	error = dumpsys_buf_write(di, (char*)&ehdr, sizeof(ehdr));
347 	if (error)
348 		goto fail;
349 
350 	/* Dump program headers */
351 	error = dumpsys_foreach_chunk(cb_dumphdr, di);
352 	if (error < 0)
353 		goto fail;
354 	error = dumpsys_write_aux_headers(di);
355 	if (error < 0)
356 		goto fail;
357 	dumpsys_buf_flush(di);
358 
359 	/*
360 	 * All headers are written using blocked I/O, so we know the
361 	 * current offset is (still) block aligned. Skip the alignement
362 	 * in the file to have the segment contents aligned at page
363 	 * boundary.
364 	 */
365 	error = dumpsys_buf_seek(di, (size_t)hdrgap);
366 	if (error)
367 		goto fail;
368 
369 	/* Dump memory chunks. */
370 	error = dumpsys_foreach_chunk(dumpsys_cb_dumpdata, di);
371 	if (error < 0)
372 		goto fail;
373 
374 	error = dump_finish(di, &kdh);
375 	if (error != 0)
376 		goto fail;
377 
378 	printf("\nDump complete\n");
379 	return (0);
380 
381  fail:
382 	if (error < 0)
383 		error = -error;
384 
385 	if (error == ECANCELED)
386 		printf("\nDump aborted\n");
387 	else if (error == E2BIG || error == ENOSPC)
388 		printf("\nDump failed. Partition too small.\n");
389 	else
390 		printf("\n** DUMP FAILED (ERROR %d) **\n", error);
391 	return (error);
392 }
393 
394 #if MINIDUMP_PAGE_TRACKING == 1
395 
396 /* Minidump progress bar */
397 static struct {
398 	const int min_per;
399 	const int max_per;
400 	bool visited;
401 } progress_track[10] = {
402 	{  0,  10, false},
403 	{ 10,  20, false},
404 	{ 20,  30, false},
405 	{ 30,  40, false},
406 	{ 40,  50, false},
407 	{ 50,  60, false},
408 	{ 60,  70, false},
409 	{ 70,  80, false},
410 	{ 80,  90, false},
411 	{ 90, 100, false}
412 };
413 
414 static uint64_t dumpsys_pb_size;
415 static uint64_t dumpsys_pb_remaining;
416 static uint64_t dumpsys_pb_check;
417 
418 /* Reset the progress bar for a dump of dumpsize. */
419 void
420 dumpsys_pb_init(uint64_t dumpsize)
421 {
422 	int i;
423 
424 	dumpsys_pb_size = dumpsys_pb_remaining = dumpsize;
425 	dumpsys_pb_check = 0;
426 
427 	for (i = 0; i < nitems(progress_track); i++)
428 		progress_track[i].visited = false;
429 }
430 
431 /*
432  * Update the progress according to the delta bytes that were written out.
433  * Check and print the progress percentage.
434  */
435 void
436 dumpsys_pb_progress(size_t delta)
437 {
438 	int sofar, i;
439 
440 	dumpsys_pb_remaining -= delta;
441 	dumpsys_pb_check += delta;
442 
443 	/*
444 	 * To save time while dumping, only loop through progress_track
445 	 * occasionally.
446 	 */
447 	if ((dumpsys_pb_check >> DUMPSYS_PB_CHECK_BITS) == 0)
448 		return;
449 	else
450 		dumpsys_pb_check &= (1 << DUMPSYS_PB_CHECK_BITS) - 1;
451 
452 	sofar = 100 - ((dumpsys_pb_remaining * 100) / dumpsys_pb_size);
453 	for (i = 0; i < nitems(progress_track); i++) {
454 		if (sofar < progress_track[i].min_per ||
455 		    sofar > progress_track[i].max_per)
456 			continue;
457 		if (!progress_track[i].visited) {
458 			progress_track[i].visited = true;
459 			printf("..%d%%", sofar);
460 		}
461 		break;
462 	}
463 }
464 
465 int
466 minidumpsys(struct dumperinfo *di, bool livedump)
467 {
468 	struct minidumpstate state;
469 	struct msgbuf mb_copy;
470 	char *msg_ptr;
471 	size_t sz;
472 	int error;
473 
474 	if (livedump) {
475 		KASSERT(!dumping, ("live dump invoked from incorrect context"));
476 
477 		/*
478 		 * Before invoking cpu_minidumpsys() on the live system, we
479 		 * must snapshot some required global state: the message
480 		 * buffer, and the page dump bitset. They may be modified at
481 		 * any moment, so for the sake of the live dump it is best to
482 		 * have an unchanging snapshot to work with. Both are included
483 		 * as part of the dump and consumed by userspace tools.
484 		 *
485 		 * Other global state important to the minidump code is the
486 		 * dump_avail array and the kernel's page tables, but snapshots
487 		 * are not taken of these. For one, dump_avail[] is expected
488 		 * not to change after boot. Snapshotting the kernel page
489 		 * tables would involve an additional walk, so this is avoided
490 		 * too.
491 		 *
492 		 * This means live dumps are best effort, and the result may or
493 		 * may not be usable; there are no guarantees about the
494 		 * consistency of the dump's contents. Any of the following
495 		 * (and likely more) may affect the live dump:
496 		 *
497 		 *  - Data may be modified, freed, or remapped during the
498 		 *    course of the dump, such that the contents written out
499 		 *    are partially or entirely unrecognizable. This means
500 		 *    valid references may point to destroyed/mangled objects,
501 		 *    and vice versa.
502 		 *
503 		 *  - The dumped context of any threads that ran during the
504 		 *    dump process may be unreliable.
505 		 *
506 		 *  - The set of kernel page tables included in the dump likely
507 		 *    won't correspond exactly to the copy of the dump bitset.
508 		 *    This means some pages will be dumped without any way to
509 		 *    locate them, and some pages may not have been dumped
510 		 *    despite appearing as if they should.
511 		 */
512 		msg_ptr = malloc(msgbufsize, M_TEMP, M_WAITOK);
513 		msgbuf_duplicate(msgbufp, &mb_copy, msg_ptr);
514 		state.msgbufp = &mb_copy;
515 
516 		sz = BITSET_SIZE(vm_page_dump_pages);
517 		state.dump_bitset = malloc(sz, M_TEMP, M_WAITOK);
518 		BIT_COPY_STORE_REL(sz, vm_page_dump, state.dump_bitset);
519 	} else {
520 		KASSERT(dumping, ("minidump invoked outside of doadump()"));
521 
522 		/* Use the globals. */
523 		state.msgbufp = msgbufp;
524 		state.dump_bitset = vm_page_dump;
525 	}
526 
527 	error = cpu_minidumpsys(di, &state);
528 	if (livedump) {
529 		free(msg_ptr, M_TEMP);
530 		free(state.dump_bitset, M_TEMP);
531 	}
532 
533 	return (error);
534 }
535 #endif /* MINIDUMP_PAGE_TRACKING == 1 */
536