1 /*- 2 * Copyright (c) 2006 Peter Wemm 3 * Copyright (c) 2015 The FreeBSD Foundation 4 * All rights reserved. 5 * 6 * This software was developed by Andrew Turner under 7 * sponsorship from the FreeBSD Foundation. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 #include <sys/cdefs.h> 33 __FBSDID("$FreeBSD$"); 34 35 #include "opt_watchdog.h" 36 37 #include "opt_watchdog.h" 38 39 #include <sys/param.h> 40 #include <sys/systm.h> 41 #include <sys/conf.h> 42 #include <sys/cons.h> 43 #include <sys/kernel.h> 44 #include <sys/kerneldump.h> 45 #include <sys/msgbuf.h> 46 #include <sys/watchdog.h> 47 48 #include <vm/vm.h> 49 #include <vm/vm_param.h> 50 #include <vm/vm_page.h> 51 #include <vm/vm_phys.h> 52 #include <vm/pmap.h> 53 54 #include <machine/md_var.h> 55 #include <machine/pte.h> 56 #include <machine/minidump.h> 57 58 CTASSERT(sizeof(struct kerneldumpheader) == 512); 59 60 /* 61 * Don't touch the first SIZEOF_METADATA bytes on the dump device. This 62 * is to protect us from metadata and to protect metadata from us. 63 */ 64 #define SIZEOF_METADATA (64*1024) 65 66 uint64_t *vm_page_dump; 67 int vm_page_dump_size; 68 69 static struct kerneldumpheader kdh; 70 static off_t dumplo; 71 72 /* Handle chunked writes. */ 73 static size_t fragsz; 74 static void *dump_va; 75 static size_t counter, progress, dumpsize; 76 77 static uint64_t tmpbuffer[PAGE_SIZE / sizeof(uint64_t)]; 78 79 CTASSERT(sizeof(*vm_page_dump) == 8); 80 81 static int 82 is_dumpable(vm_paddr_t pa) 83 { 84 vm_page_t m; 85 int i; 86 87 if ((m = vm_phys_paddr_to_vm_page(pa)) != NULL) 88 return ((m->flags & PG_NODUMP) == 0); 89 for (i = 0; dump_avail[i] != 0 || dump_avail[i + 1] != 0; i += 2) { 90 if (pa >= dump_avail[i] && pa < dump_avail[i + 1]) 91 return (1); 92 } 93 return (0); 94 } 95 96 static int 97 blk_flush(struct dumperinfo *di) 98 { 99 int error; 100 101 if (fragsz == 0) 102 return (0); 103 104 error = dump_write(di, dump_va, 0, dumplo, fragsz); 105 dumplo += fragsz; 106 fragsz = 0; 107 return (error); 108 } 109 110 static struct { 111 int min_per; 112 int max_per; 113 int visited; 114 } progress_track[10] = { 115 { 0, 10, 0}, 116 { 10, 20, 0}, 117 { 20, 30, 0}, 118 { 30, 40, 0}, 119 { 40, 50, 0}, 120 { 50, 60, 0}, 121 { 60, 70, 0}, 122 { 70, 80, 0}, 123 { 80, 90, 0}, 124 { 90, 100, 0} 125 }; 126 127 static void 128 report_progress(size_t progress, size_t dumpsize) 129 { 130 int sofar, i; 131 132 sofar = 100 - ((progress * 100) / dumpsize); 133 for (i = 0; i < nitems(progress_track); i++) { 134 if (sofar < progress_track[i].min_per || 135 sofar > progress_track[i].max_per) 136 continue; 137 if (progress_track[i].visited) 138 return; 139 progress_track[i].visited = 1; 140 printf("..%d%%", sofar); 141 return; 142 } 143 } 144 145 static int 146 blk_write(struct dumperinfo *di, char *ptr, vm_paddr_t pa, size_t sz) 147 { 148 size_t len; 149 int error, c; 150 u_int maxdumpsz; 151 152 maxdumpsz = min(di->maxiosize, MAXDUMPPGS * PAGE_SIZE); 153 if (maxdumpsz == 0) /* seatbelt */ 154 maxdumpsz = PAGE_SIZE; 155 error = 0; 156 if ((sz % PAGE_SIZE) != 0) { 157 printf("size not page aligned\n"); 158 return (EINVAL); 159 } 160 if (ptr != NULL && pa != 0) { 161 printf("cant have both va and pa!\n"); 162 return (EINVAL); 163 } 164 if ((((uintptr_t)pa) % PAGE_SIZE) != 0) { 165 printf("address not page aligned %p\n", ptr); 166 return (EINVAL); 167 } 168 if (ptr != NULL) { 169 /* 170 * If we're doing a virtual dump, flush any 171 * pre-existing pa pages. 172 */ 173 error = blk_flush(di); 174 if (error) 175 return (error); 176 } 177 while (sz) { 178 len = maxdumpsz - fragsz; 179 if (len > sz) 180 len = sz; 181 counter += len; 182 progress -= len; 183 if (counter >> 22) { 184 report_progress(progress, dumpsize); 185 counter &= (1 << 22) - 1; 186 } 187 188 wdog_kern_pat(WD_LASTVAL); 189 190 if (ptr) { 191 error = dump_write(di, ptr, 0, dumplo, len); 192 if (error) 193 return (error); 194 dumplo += len; 195 ptr += len; 196 sz -= len; 197 } else { 198 dump_va = (void *)PHYS_TO_DMAP(pa); 199 fragsz += len; 200 pa += len; 201 sz -= len; 202 error = blk_flush(di); 203 if (error) 204 return (error); 205 } 206 207 /* Check for user abort. */ 208 c = cncheckc(); 209 if (c == 0x03) 210 return (ECANCELED); 211 if (c != -1) 212 printf(" (CTRL-C to abort) "); 213 } 214 215 return (0); 216 } 217 218 int 219 minidumpsys(struct dumperinfo *di) 220 { 221 pd_entry_t *l0, *l1, *l2; 222 pt_entry_t *l3; 223 uint32_t pmapsize; 224 vm_offset_t va; 225 vm_paddr_t pa; 226 int error; 227 uint64_t bits; 228 int i, bit; 229 int retry_count; 230 struct minidumphdr mdhdr; 231 232 retry_count = 0; 233 retry: 234 retry_count++; 235 error = 0; 236 pmapsize = 0; 237 for (va = VM_MIN_KERNEL_ADDRESS; va < kernel_vm_end; va += L2_SIZE) { 238 pmapsize += PAGE_SIZE; 239 if (!pmap_get_tables(pmap_kernel(), va, &l0, &l1, &l2, &l3)) 240 continue; 241 242 /* We should always be using the l2 table for kvm */ 243 if (l2 == NULL) 244 continue; 245 246 if ((*l2 & ATTR_DESCR_MASK) == L2_BLOCK) { 247 pa = *l2 & ~ATTR_MASK; 248 for (i = 0; i < Ln_ENTRIES; i++, pa += PAGE_SIZE) { 249 if (is_dumpable(pa)) 250 dump_add_page(pa); 251 } 252 } else if ((*l2 & ATTR_DESCR_MASK) == L2_TABLE) { 253 for (i = 0; i < Ln_ENTRIES; i++) { 254 if ((l3[i] & ATTR_DESCR_MASK) != L3_PAGE) 255 continue; 256 pa = l3[i] & ~ATTR_MASK; 257 if (is_dumpable(pa)) 258 dump_add_page(pa); 259 } 260 } 261 } 262 263 /* Calculate dump size. */ 264 dumpsize = pmapsize; 265 dumpsize += round_page(msgbufp->msg_size); 266 dumpsize += round_page(vm_page_dump_size); 267 for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) { 268 bits = vm_page_dump[i]; 269 while (bits) { 270 bit = ffsl(bits) - 1; 271 pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) + 272 bit) * PAGE_SIZE; 273 /* Clear out undumpable pages now if needed */ 274 if (is_dumpable(pa)) 275 dumpsize += PAGE_SIZE; 276 else 277 dump_drop_page(pa); 278 bits &= ~(1ul << bit); 279 } 280 } 281 dumpsize += PAGE_SIZE; 282 283 /* Determine dump offset on device. */ 284 if (di->mediasize < SIZEOF_METADATA + dumpsize + di->blocksize * 2 + 285 kerneldumpcrypto_dumpkeysize(di->kdc)) { 286 error = E2BIG; 287 goto fail; 288 } 289 dumplo = di->mediaoffset + di->mediasize - dumpsize; 290 dumplo -= di->blocksize * 2; 291 dumplo -= kerneldumpcrypto_dumpkeysize(di->kdc); 292 progress = dumpsize; 293 294 /* Initialize kernel dump crypto. */ 295 error = kerneldumpcrypto_init(di->kdc); 296 if (error) 297 goto fail; 298 299 /* Initialize mdhdr */ 300 bzero(&mdhdr, sizeof(mdhdr)); 301 strcpy(mdhdr.magic, MINIDUMP_MAGIC); 302 mdhdr.version = MINIDUMP_VERSION; 303 mdhdr.msgbufsize = msgbufp->msg_size; 304 mdhdr.bitmapsize = vm_page_dump_size; 305 mdhdr.pmapsize = pmapsize; 306 mdhdr.kernbase = VM_MIN_KERNEL_ADDRESS; 307 mdhdr.dmapphys = DMAP_MIN_PHYSADDR; 308 mdhdr.dmapbase = DMAP_MIN_ADDRESS; 309 mdhdr.dmapend = DMAP_MAX_ADDRESS; 310 311 mkdumpheader(&kdh, KERNELDUMPMAGIC, KERNELDUMP_AARCH64_VERSION, 312 dumpsize, kerneldumpcrypto_dumpkeysize(di->kdc), di->blocksize); 313 314 printf("Dumping %llu out of %ju MB:", (long long)dumpsize >> 20, 315 ptoa((uintmax_t)physmem) / 1048576); 316 317 /* Dump leader */ 318 error = dump_write_header(di, &kdh, 0, dumplo); 319 if (error) 320 goto fail; 321 dumplo += di->blocksize; 322 323 /* Dump key */ 324 error = dump_write_key(di, 0, dumplo); 325 if (error) 326 goto fail; 327 dumplo += kerneldumpcrypto_dumpkeysize(di->kdc); 328 329 /* Dump my header */ 330 bzero(&tmpbuffer, sizeof(tmpbuffer)); 331 bcopy(&mdhdr, &tmpbuffer, sizeof(mdhdr)); 332 error = blk_write(di, (char *)&tmpbuffer, 0, PAGE_SIZE); 333 if (error) 334 goto fail; 335 336 /* Dump msgbuf up front */ 337 error = blk_write(di, (char *)msgbufp->msg_ptr, 0, 338 round_page(msgbufp->msg_size)); 339 if (error) 340 goto fail; 341 342 /* Dump bitmap */ 343 error = blk_write(di, (char *)vm_page_dump, 0, 344 round_page(vm_page_dump_size)); 345 if (error) 346 goto fail; 347 348 /* Dump kernel page directory pages */ 349 bzero(&tmpbuffer, sizeof(tmpbuffer)); 350 for (va = VM_MIN_KERNEL_ADDRESS; va < kernel_vm_end; va += L2_SIZE) { 351 if (!pmap_get_tables(pmap_kernel(), va, &l0, &l1, &l2, &l3)) { 352 /* We always write a page, even if it is zero */ 353 error = blk_write(di, (char *)&tmpbuffer, 0, PAGE_SIZE); 354 if (error) 355 goto fail; 356 /* flush, in case we reuse tmpbuffer in the same block*/ 357 error = blk_flush(di); 358 if (error) 359 goto fail; 360 } else if (l2 == NULL) { 361 pa = (*l1 & ~ATTR_MASK) | (va & L1_OFFSET); 362 363 /* Generate fake l3 entries based upon the l1 entry */ 364 for (i = 0; i < Ln_ENTRIES; i++) { 365 tmpbuffer[i] = pa + (i * PAGE_SIZE) | 366 ATTR_DEFAULT | L3_PAGE; 367 } 368 /* We always write a page, even if it is zero */ 369 error = blk_write(di, (char *)&tmpbuffer, 0, PAGE_SIZE); 370 if (error) 371 goto fail; 372 /* flush, in case we reuse tmpbuffer in the same block*/ 373 error = blk_flush(di); 374 if (error) 375 goto fail; 376 bzero(&tmpbuffer, sizeof(tmpbuffer)); 377 } else if ((*l2 & ATTR_DESCR_MASK) == L2_BLOCK) { 378 /* TODO: Handle an invalid L2 entry */ 379 pa = (*l2 & ~ATTR_MASK) | (va & L2_OFFSET); 380 381 /* Generate fake l3 entries based upon the l1 entry */ 382 for (i = 0; i < Ln_ENTRIES; i++) { 383 tmpbuffer[i] = pa + (i * PAGE_SIZE) | 384 ATTR_DEFAULT | L3_PAGE; 385 } 386 /* We always write a page, even if it is zero */ 387 error = blk_write(di, (char *)&tmpbuffer, 0, PAGE_SIZE); 388 if (error) 389 goto fail; 390 /* flush, in case we reuse fakepd in the same block */ 391 error = blk_flush(di); 392 if (error) 393 goto fail; 394 bzero(&tmpbuffer, sizeof(tmpbuffer)); 395 continue; 396 } else { 397 pa = *l2 & ~ATTR_MASK; 398 399 /* We always write a page, even if it is zero */ 400 error = blk_write(di, NULL, pa, PAGE_SIZE); 401 if (error) 402 goto fail; 403 } 404 } 405 406 /* Dump memory chunks */ 407 /* XXX cluster it up and use blk_dump() */ 408 for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) { 409 bits = vm_page_dump[i]; 410 while (bits) { 411 bit = ffsl(bits) - 1; 412 pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) + 413 bit) * PAGE_SIZE; 414 error = blk_write(di, 0, pa, PAGE_SIZE); 415 if (error) 416 goto fail; 417 bits &= ~(1ul << bit); 418 } 419 } 420 421 error = blk_flush(di); 422 if (error) 423 goto fail; 424 425 /* Dump trailer */ 426 error = dump_write_header(di, &kdh, 0, dumplo); 427 if (error) 428 goto fail; 429 dumplo += di->blocksize; 430 431 /* Signal completion, signoff and exit stage left. */ 432 dump_write(di, NULL, 0, 0, 0); 433 printf("\nDump complete\n"); 434 return (0); 435 436 fail: 437 if (error < 0) 438 error = -error; 439 440 printf("\n"); 441 if (error == ENOSPC) { 442 printf("Dump map grown while dumping. "); 443 if (retry_count < 5) { 444 printf("Retrying...\n"); 445 goto retry; 446 } 447 printf("Dump failed.\n"); 448 } 449 else if (error == ECANCELED) 450 printf("Dump aborted\n"); 451 else if (error == E2BIG) 452 printf("Dump failed. Partition too small.\n"); 453 else 454 printf("** DUMP FAILED (ERROR %d) **\n", error); 455 return (error); 456 } 457 458 void 459 dump_add_page(vm_paddr_t pa) 460 { 461 int idx, bit; 462 463 pa >>= PAGE_SHIFT; 464 idx = pa >> 6; /* 2^6 = 64 */ 465 bit = pa & 63; 466 atomic_set_long(&vm_page_dump[idx], 1ul << bit); 467 } 468 469 void 470 dump_drop_page(vm_paddr_t pa) 471 { 472 int idx, bit; 473 474 pa >>= PAGE_SHIFT; 475 idx = pa >> 6; /* 2^6 = 64 */ 476 bit = pa & 63; 477 atomic_clear_long(&vm_page_dump[idx], 1ul << bit); 478 } 479