2 * Copyright (c) 2002 Marcel Moolenaar
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
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.
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.
27 #include <sys/param.h>
28 #include <sys/systm.h>
32 #include <sys/kernel.h>
33 #include <sys/kerneldump.h>
34 #include <sys/malloc.h>
35 #include <sys/msgbuf.h>
37 #include <sys/watchdog.h>
40 #include <vm/vm_param.h>
41 #include <vm/vm_page.h>
42 #include <vm/vm_phys.h>
43 #include <vm/vm_dumpset.h>
46 #include <machine/dump.h>
47 #include <machine/elf.h>
48 #include <machine/md_var.h>
49 #include <machine/pcb.h>
51 CTASSERT(sizeof(struct kerneldumpheader) == 512);
53 #define MD_ALIGN(x) roundup2((off_t)(x), PAGE_SIZE)
55 /* Handle buffered writes. */
58 struct dump_pa dump_map[DUMPSYS_MD_PA_NPAIRS];
60 #if !defined(__powerpc__)
62 dumpsys_gen_pa_init(void)
66 bzero(dump_map, sizeof(dump_map));
67 for (n = 0; n < nitems(dump_map); n++) {
69 if (dump_avail[idx] == 0 && dump_avail[idx + 1] == 0)
71 dump_map[n].pa_start = dump_avail[idx];
72 dump_map[n].pa_size = dump_avail[idx + 1] - dump_avail[idx];
78 dumpsys_gen_pa_next(struct dump_pa *mdp)
82 return (&dump_map[0]);
85 if (mdp->pa_size == 0)
91 dumpsys_gen_wbinv_all(void)
97 dumpsys_gen_unmap_chunk(vm_paddr_t pa __unused, size_t chunk __unused,
104 dumpsys_gen_write_aux_headers(struct dumperinfo *di)
111 dumpsys_buf_seek(struct dumperinfo *di, size_t sz)
113 static uint8_t buf[DEV_BSIZE];
117 bzero(buf, sizeof(buf));
120 nbytes = MIN(sz, sizeof(buf));
122 error = dump_append(di, buf, nbytes);
132 dumpsys_buf_write(struct dumperinfo *di, char *ptr, size_t sz)
138 len = di->blocksize - fragsz;
141 memcpy((char *)di->blockbuf + fragsz, ptr, len);
145 if (fragsz == di->blocksize) {
146 error = dump_append(di, di->blockbuf, di->blocksize);
156 dumpsys_buf_flush(struct dumperinfo *di)
163 error = dump_append(di, di->blockbuf, di->blocksize);
168 CTASSERT(PAGE_SHIFT < 20);
169 #define PG2MB(pgs) ((pgs + (1 << (20 - PAGE_SHIFT)) - 1) >> (20 - PAGE_SHIFT))
172 dumpsys_cb_dumpdata(struct dump_pa *mdp, int seqnr, void *arg)
174 struct dumperinfo *di = (struct dumperinfo*)arg;
178 size_t counter, sz, chunk;
182 error = 0; /* catch case in which chunk size is 0 */
183 counter = 0; /* Update twiddle every 16MB */
185 pgs = mdp->pa_size / PAGE_SIZE;
187 maxdumppgs = min(di->maxiosize / PAGE_SIZE, MAXDUMPPGS);
188 if (maxdumppgs == 0) /* seatbelt */
191 printf(" chunk %d: %juMB (%ju pages)", seqnr, (uintmax_t)PG2MB(pgs),
197 if (chunk > maxdumppgs)
199 sz = chunk << PAGE_SHIFT;
202 printf(" %ju", (uintmax_t)PG2MB(pgs));
203 counter &= (1 << 24) - 1;
206 dumpsys_map_chunk(pa, chunk, &va);
207 wdog_kern_pat(WD_LASTVAL);
209 error = dump_append(di, va, sz);
210 dumpsys_unmap_chunk(pa, chunk, va);
216 /* Check for user abort. */
221 printf(" (CTRL-C to abort) ");
223 printf(" ... %s\n", (error) ? "fail" : "ok");
228 dumpsys_foreach_chunk(dumpsys_callback_t cb, void *arg)
234 mdp = dumpsys_pa_next(NULL);
235 while (mdp != NULL) {
236 error = (*cb)(mdp, seqnr++, arg);
239 mdp = dumpsys_pa_next(mdp);
244 static off_t fileofs;
247 cb_dumphdr(struct dump_pa *mdp, int seqnr, void *arg)
249 struct dumperinfo *di = (struct dumperinfo*)arg;
255 bzero(&phdr, sizeof(phdr));
256 phdr.p_type = PT_LOAD;
257 phdr.p_flags = PF_R; /* XXX */
258 phdr.p_offset = fileofs;
260 phdr.p_vaddr = (do_minidump? mdp->pa_start : ~0L);
261 phdr.p_paddr = (do_minidump? ~0L : mdp->pa_start);
263 phdr.p_vaddr = mdp->pa_start;
264 phdr.p_paddr = mdp->pa_start;
266 phdr.p_filesz = size;
268 phdr.p_align = PAGE_SIZE;
270 error = dumpsys_buf_write(di, (char*)&phdr, sizeof(phdr));
271 fileofs += phdr.p_filesz;
276 cb_size(struct dump_pa *mdp, int seqnr, void *arg)
280 sz = (uint64_t *)arg;
281 *sz += (uint64_t)mdp->pa_size;
286 dumpsys_generic(struct dumperinfo *di)
288 static struct kerneldumpheader kdh;
295 #if MINIDUMP_PAGE_TRACKING == 1
297 return (minidumpsys(di, false));
300 bzero(&ehdr, sizeof(ehdr));
301 ehdr.e_ident[EI_MAG0] = ELFMAG0;
302 ehdr.e_ident[EI_MAG1] = ELFMAG1;
303 ehdr.e_ident[EI_MAG2] = ELFMAG2;
304 ehdr.e_ident[EI_MAG3] = ELFMAG3;
305 ehdr.e_ident[EI_CLASS] = ELF_CLASS;
306 #if BYTE_ORDER == LITTLE_ENDIAN
307 ehdr.e_ident[EI_DATA] = ELFDATA2LSB;
309 ehdr.e_ident[EI_DATA] = ELFDATA2MSB;
311 ehdr.e_ident[EI_VERSION] = EV_CURRENT;
312 ehdr.e_ident[EI_OSABI] = ELFOSABI_STANDALONE; /* XXX big picture? */
313 ehdr.e_type = ET_CORE;
314 ehdr.e_machine = EM_VALUE;
315 ehdr.e_phoff = sizeof(ehdr);
317 ehdr.e_ehsize = sizeof(ehdr);
318 ehdr.e_phentsize = sizeof(Elf_Phdr);
319 ehdr.e_shentsize = sizeof(Elf_Shdr);
323 /* Calculate dump size. */
325 ehdr.e_phnum = dumpsys_foreach_chunk(cb_size, &dumpsize) +
326 DUMPSYS_NUM_AUX_HDRS;
327 hdrsz = ehdr.e_phoff + ehdr.e_phnum * ehdr.e_phentsize;
328 fileofs = MD_ALIGN(hdrsz);
330 hdrgap = fileofs - roundup2((off_t)hdrsz, di->blocksize);
332 dump_init_header(di, &kdh, KERNELDUMPMAGIC, KERNELDUMP_ARCH_VERSION,
335 error = dump_start(di, &kdh);
339 printf("Dumping %ju MB (%d chunks)\n", (uintmax_t)dumpsize >> 20,
340 ehdr.e_phnum - DUMPSYS_NUM_AUX_HDRS);
342 /* Dump ELF header */
343 error = dumpsys_buf_write(di, (char*)&ehdr, sizeof(ehdr));
347 /* Dump program headers */
348 error = dumpsys_foreach_chunk(cb_dumphdr, di);
351 error = dumpsys_write_aux_headers(di);
354 dumpsys_buf_flush(di);
357 * All headers are written using blocked I/O, so we know the
358 * current offset is (still) block aligned. Skip the alignement
359 * in the file to have the segment contents aligned at page
362 error = dumpsys_buf_seek(di, (size_t)hdrgap);
366 /* Dump memory chunks. */
367 error = dumpsys_foreach_chunk(dumpsys_cb_dumpdata, di);
371 error = dump_finish(di, &kdh);
375 printf("\nDump complete\n");
382 if (error == ECANCELED)
383 printf("\nDump aborted\n");
384 else if (error == E2BIG || error == ENOSPC)
385 printf("\nDump failed. Partition too small.\n");
387 printf("\n** DUMP FAILED (ERROR %d) **\n", error);
391 #if MINIDUMP_PAGE_TRACKING == 1
393 /* Minidump progress bar */
398 } progress_track[10] = {
411 static uint64_t dumpsys_pb_size;
412 static uint64_t dumpsys_pb_remaining;
413 static uint64_t dumpsys_pb_check;
415 /* Reset the progress bar for a dump of dumpsize. */
417 dumpsys_pb_init(uint64_t dumpsize)
421 dumpsys_pb_size = dumpsys_pb_remaining = dumpsize;
422 dumpsys_pb_check = 0;
424 for (i = 0; i < nitems(progress_track); i++)
425 progress_track[i].visited = false;
429 * Update the progress according to the delta bytes that were written out.
430 * Check and print the progress percentage.
433 dumpsys_pb_progress(size_t delta)
437 dumpsys_pb_remaining -= delta;
438 dumpsys_pb_check += delta;
441 * To save time while dumping, only loop through progress_track
444 if ((dumpsys_pb_check >> DUMPSYS_PB_CHECK_BITS) == 0)
447 dumpsys_pb_check &= (1 << DUMPSYS_PB_CHECK_BITS) - 1;
449 sofar = 100 - ((dumpsys_pb_remaining * 100) / dumpsys_pb_size);
450 for (i = 0; i < nitems(progress_track); i++) {
451 if (sofar < progress_track[i].min_per ||
452 sofar > progress_track[i].max_per)
454 if (!progress_track[i].visited) {
455 progress_track[i].visited = true;
456 printf("..%d%%", sofar);
463 minidumpsys(struct dumperinfo *di, bool livedump)
465 struct minidumpstate state;
466 struct msgbuf mb_copy;
472 KASSERT(!dumping, ("live dump invoked from incorrect context"));
475 * Before invoking cpu_minidumpsys() on the live system, we
476 * must snapshot some required global state: the message
477 * buffer, and the page dump bitset. They may be modified at
478 * any moment, so for the sake of the live dump it is best to
479 * have an unchanging snapshot to work with. Both are included
480 * as part of the dump and consumed by userspace tools.
482 * Other global state important to the minidump code is the
483 * dump_avail array and the kernel's page tables, but snapshots
484 * are not taken of these. For one, dump_avail[] is expected
485 * not to change after boot. Snapshotting the kernel page
486 * tables would involve an additional walk, so this is avoided
489 * This means live dumps are best effort, and the result may or
490 * may not be usable; there are no guarantees about the
491 * consistency of the dump's contents. Any of the following
492 * (and likely more) may affect the live dump:
494 * - Data may be modified, freed, or remapped during the
495 * course of the dump, such that the contents written out
496 * are partially or entirely unrecognizable. This means
497 * valid references may point to destroyed/mangled objects,
500 * - The dumped context of any threads that ran during the
501 * dump process may be unreliable.
503 * - The set of kernel page tables included in the dump likely
504 * won't correspond exactly to the copy of the dump bitset.
505 * This means some pages will be dumped without any way to
506 * locate them, and some pages may not have been dumped
507 * despite appearing as if they should.
509 msg_ptr = malloc(msgbufsize, M_TEMP, M_WAITOK);
510 msgbuf_duplicate(msgbufp, &mb_copy, msg_ptr);
511 state.msgbufp = &mb_copy;
513 sz = BITSET_SIZE(vm_page_dump_pages);
514 state.dump_bitset = malloc(sz, M_TEMP, M_WAITOK);
515 BIT_COPY_STORE_REL(sz, vm_page_dump, state.dump_bitset);
517 KASSERT(dumping, ("minidump invoked outside of doadump()"));
519 /* Use the globals. */
520 state.msgbufp = msgbufp;
521 state.dump_bitset = vm_page_dump;
524 error = cpu_minidumpsys(di, &state);
526 free(msg_ptr, M_TEMP);
527 free(state.dump_bitset, M_TEMP);
532 #endif /* MINIDUMP_PAGE_TRACKING == 1 */