| 1 | /* |
| 2 | * kexec.c - kexec system call |
| 3 | * Copyright (C) 2002-2004 Eric Biederman <ebiederm@xmission.com> |
| 4 | * |
| 5 | * This source code is licensed under the GNU General Public License, |
| 6 | * Version 2. See the file COPYING for more details. |
| 7 | */ |
| 8 | |
| 9 | #include <linux/capability.h> |
| 10 | #include <linux/mm.h> |
| 11 | #include <linux/file.h> |
| 12 | #include <linux/slab.h> |
| 13 | #include <linux/fs.h> |
| 14 | #include <linux/kexec.h> |
| 15 | #include <linux/spinlock.h> |
| 16 | #include <linux/list.h> |
| 17 | #include <linux/highmem.h> |
| 18 | #include <linux/syscalls.h> |
| 19 | #include <linux/reboot.h> |
| 20 | #include <linux/syscalls.h> |
| 21 | #include <linux/ioport.h> |
| 22 | #include <linux/hardirq.h> |
| 23 | |
| 24 | #include <asm/page.h> |
| 25 | #include <asm/uaccess.h> |
| 26 | #include <asm/io.h> |
| 27 | #include <asm/system.h> |
| 28 | #include <asm/semaphore.h> |
| 29 | |
| 30 | /* Per cpu memory for storing cpu states in case of system crash. */ |
| 31 | note_buf_t* crash_notes; |
| 32 | |
| 33 | /* Location of the reserved area for the crash kernel */ |
| 34 | struct resource crashk_res = { |
| 35 | .name = "Crash kernel", |
| 36 | .start = 0, |
| 37 | .end = 0, |
| 38 | .flags = IORESOURCE_BUSY | IORESOURCE_MEM |
| 39 | }; |
| 40 | |
| 41 | int kexec_should_crash(struct task_struct *p) |
| 42 | { |
| 43 | if (in_interrupt() || !p->pid || is_init(p) || panic_on_oops) |
| 44 | return 1; |
| 45 | return 0; |
| 46 | } |
| 47 | |
| 48 | /* |
| 49 | * When kexec transitions to the new kernel there is a one-to-one |
| 50 | * mapping between physical and virtual addresses. On processors |
| 51 | * where you can disable the MMU this is trivial, and easy. For |
| 52 | * others it is still a simple predictable page table to setup. |
| 53 | * |
| 54 | * In that environment kexec copies the new kernel to its final |
| 55 | * resting place. This means I can only support memory whose |
| 56 | * physical address can fit in an unsigned long. In particular |
| 57 | * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled. |
| 58 | * If the assembly stub has more restrictive requirements |
| 59 | * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be |
| 60 | * defined more restrictively in <asm/kexec.h>. |
| 61 | * |
| 62 | * The code for the transition from the current kernel to the |
| 63 | * the new kernel is placed in the control_code_buffer, whose size |
| 64 | * is given by KEXEC_CONTROL_CODE_SIZE. In the best case only a single |
| 65 | * page of memory is necessary, but some architectures require more. |
| 66 | * Because this memory must be identity mapped in the transition from |
| 67 | * virtual to physical addresses it must live in the range |
| 68 | * 0 - TASK_SIZE, as only the user space mappings are arbitrarily |
| 69 | * modifiable. |
| 70 | * |
| 71 | * The assembly stub in the control code buffer is passed a linked list |
| 72 | * of descriptor pages detailing the source pages of the new kernel, |
| 73 | * and the destination addresses of those source pages. As this data |
| 74 | * structure is not used in the context of the current OS, it must |
| 75 | * be self-contained. |
| 76 | * |
| 77 | * The code has been made to work with highmem pages and will use a |
| 78 | * destination page in its final resting place (if it happens |
| 79 | * to allocate it). The end product of this is that most of the |
| 80 | * physical address space, and most of RAM can be used. |
| 81 | * |
| 82 | * Future directions include: |
| 83 | * - allocating a page table with the control code buffer identity |
| 84 | * mapped, to simplify machine_kexec and make kexec_on_panic more |
| 85 | * reliable. |
| 86 | */ |
| 87 | |
| 88 | /* |
| 89 | * KIMAGE_NO_DEST is an impossible destination address..., for |
| 90 | * allocating pages whose destination address we do not care about. |
| 91 | */ |
| 92 | #define KIMAGE_NO_DEST (-1UL) |
| 93 | |
| 94 | static int kimage_is_destination_range(struct kimage *image, |
| 95 | unsigned long start, unsigned long end); |
| 96 | static struct page *kimage_alloc_page(struct kimage *image, |
| 97 | gfp_t gfp_mask, |
| 98 | unsigned long dest); |
| 99 | |
| 100 | static int do_kimage_alloc(struct kimage **rimage, unsigned long entry, |
| 101 | unsigned long nr_segments, |
| 102 | struct kexec_segment __user *segments) |
| 103 | { |
| 104 | size_t segment_bytes; |
| 105 | struct kimage *image; |
| 106 | unsigned long i; |
| 107 | int result; |
| 108 | |
| 109 | /* Allocate a controlling structure */ |
| 110 | result = -ENOMEM; |
| 111 | image = kzalloc(sizeof(*image), GFP_KERNEL); |
| 112 | if (!image) |
| 113 | goto out; |
| 114 | |
| 115 | image->head = 0; |
| 116 | image->entry = &image->head; |
| 117 | image->last_entry = &image->head; |
| 118 | image->control_page = ~0; /* By default this does not apply */ |
| 119 | image->start = entry; |
| 120 | image->type = KEXEC_TYPE_DEFAULT; |
| 121 | |
| 122 | /* Initialize the list of control pages */ |
| 123 | INIT_LIST_HEAD(&image->control_pages); |
| 124 | |
| 125 | /* Initialize the list of destination pages */ |
| 126 | INIT_LIST_HEAD(&image->dest_pages); |
| 127 | |
| 128 | /* Initialize the list of unuseable pages */ |
| 129 | INIT_LIST_HEAD(&image->unuseable_pages); |
| 130 | |
| 131 | /* Read in the segments */ |
| 132 | image->nr_segments = nr_segments; |
| 133 | segment_bytes = nr_segments * sizeof(*segments); |
| 134 | result = copy_from_user(image->segment, segments, segment_bytes); |
| 135 | if (result) |
| 136 | goto out; |
| 137 | |
| 138 | /* |
| 139 | * Verify we have good destination addresses. The caller is |
| 140 | * responsible for making certain we don't attempt to load |
| 141 | * the new image into invalid or reserved areas of RAM. This |
| 142 | * just verifies it is an address we can use. |
| 143 | * |
| 144 | * Since the kernel does everything in page size chunks ensure |
| 145 | * the destination addreses are page aligned. Too many |
| 146 | * special cases crop of when we don't do this. The most |
| 147 | * insidious is getting overlapping destination addresses |
| 148 | * simply because addresses are changed to page size |
| 149 | * granularity. |
| 150 | */ |
| 151 | result = -EADDRNOTAVAIL; |
| 152 | for (i = 0; i < nr_segments; i++) { |
| 153 | unsigned long mstart, mend; |
| 154 | |
| 155 | mstart = image->segment[i].mem; |
| 156 | mend = mstart + image->segment[i].memsz; |
| 157 | if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK)) |
| 158 | goto out; |
| 159 | if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT) |
| 160 | goto out; |
| 161 | } |
| 162 | |
| 163 | /* Verify our destination addresses do not overlap. |
| 164 | * If we alloed overlapping destination addresses |
| 165 | * through very weird things can happen with no |
| 166 | * easy explanation as one segment stops on another. |
| 167 | */ |
| 168 | result = -EINVAL; |
| 169 | for (i = 0; i < nr_segments; i++) { |
| 170 | unsigned long mstart, mend; |
| 171 | unsigned long j; |
| 172 | |
| 173 | mstart = image->segment[i].mem; |
| 174 | mend = mstart + image->segment[i].memsz; |
| 175 | for (j = 0; j < i; j++) { |
| 176 | unsigned long pstart, pend; |
| 177 | pstart = image->segment[j].mem; |
| 178 | pend = pstart + image->segment[j].memsz; |
| 179 | /* Do the segments overlap ? */ |
| 180 | if ((mend > pstart) && (mstart < pend)) |
| 181 | goto out; |
| 182 | } |
| 183 | } |
| 184 | |
| 185 | /* Ensure our buffer sizes are strictly less than |
| 186 | * our memory sizes. This should always be the case, |
| 187 | * and it is easier to check up front than to be surprised |
| 188 | * later on. |
| 189 | */ |
| 190 | result = -EINVAL; |
| 191 | for (i = 0; i < nr_segments; i++) { |
| 192 | if (image->segment[i].bufsz > image->segment[i].memsz) |
| 193 | goto out; |
| 194 | } |
| 195 | |
| 196 | result = 0; |
| 197 | out: |
| 198 | if (result == 0) |
| 199 | *rimage = image; |
| 200 | else |
| 201 | kfree(image); |
| 202 | |
| 203 | return result; |
| 204 | |
| 205 | } |
| 206 | |
| 207 | static int kimage_normal_alloc(struct kimage **rimage, unsigned long entry, |
| 208 | unsigned long nr_segments, |
| 209 | struct kexec_segment __user *segments) |
| 210 | { |
| 211 | int result; |
| 212 | struct kimage *image; |
| 213 | |
| 214 | /* Allocate and initialize a controlling structure */ |
| 215 | image = NULL; |
| 216 | result = do_kimage_alloc(&image, entry, nr_segments, segments); |
| 217 | if (result) |
| 218 | goto out; |
| 219 | |
| 220 | *rimage = image; |
| 221 | |
| 222 | /* |
| 223 | * Find a location for the control code buffer, and add it |
| 224 | * the vector of segments so that it's pages will also be |
| 225 | * counted as destination pages. |
| 226 | */ |
| 227 | result = -ENOMEM; |
| 228 | image->control_code_page = kimage_alloc_control_pages(image, |
| 229 | get_order(KEXEC_CONTROL_CODE_SIZE)); |
| 230 | if (!image->control_code_page) { |
| 231 | printk(KERN_ERR "Could not allocate control_code_buffer\n"); |
| 232 | goto out; |
| 233 | } |
| 234 | |
| 235 | result = 0; |
| 236 | out: |
| 237 | if (result == 0) |
| 238 | *rimage = image; |
| 239 | else |
| 240 | kfree(image); |
| 241 | |
| 242 | return result; |
| 243 | } |
| 244 | |
| 245 | static int kimage_crash_alloc(struct kimage **rimage, unsigned long entry, |
| 246 | unsigned long nr_segments, |
| 247 | struct kexec_segment __user *segments) |
| 248 | { |
| 249 | int result; |
| 250 | struct kimage *image; |
| 251 | unsigned long i; |
| 252 | |
| 253 | image = NULL; |
| 254 | /* Verify we have a valid entry point */ |
| 255 | if ((entry < crashk_res.start) || (entry > crashk_res.end)) { |
| 256 | result = -EADDRNOTAVAIL; |
| 257 | goto out; |
| 258 | } |
| 259 | |
| 260 | /* Allocate and initialize a controlling structure */ |
| 261 | result = do_kimage_alloc(&image, entry, nr_segments, segments); |
| 262 | if (result) |
| 263 | goto out; |
| 264 | |
| 265 | /* Enable the special crash kernel control page |
| 266 | * allocation policy. |
| 267 | */ |
| 268 | image->control_page = crashk_res.start; |
| 269 | image->type = KEXEC_TYPE_CRASH; |
| 270 | |
| 271 | /* |
| 272 | * Verify we have good destination addresses. Normally |
| 273 | * the caller is responsible for making certain we don't |
| 274 | * attempt to load the new image into invalid or reserved |
| 275 | * areas of RAM. But crash kernels are preloaded into a |
| 276 | * reserved area of ram. We must ensure the addresses |
| 277 | * are in the reserved area otherwise preloading the |
| 278 | * kernel could corrupt things. |
| 279 | */ |
| 280 | result = -EADDRNOTAVAIL; |
| 281 | for (i = 0; i < nr_segments; i++) { |
| 282 | unsigned long mstart, mend; |
| 283 | |
| 284 | mstart = image->segment[i].mem; |
| 285 | mend = mstart + image->segment[i].memsz - 1; |
| 286 | /* Ensure we are within the crash kernel limits */ |
| 287 | if ((mstart < crashk_res.start) || (mend > crashk_res.end)) |
| 288 | goto out; |
| 289 | } |
| 290 | |
| 291 | /* |
| 292 | * Find a location for the control code buffer, and add |
| 293 | * the vector of segments so that it's pages will also be |
| 294 | * counted as destination pages. |
| 295 | */ |
| 296 | result = -ENOMEM; |
| 297 | image->control_code_page = kimage_alloc_control_pages(image, |
| 298 | get_order(KEXEC_CONTROL_CODE_SIZE)); |
| 299 | if (!image->control_code_page) { |
| 300 | printk(KERN_ERR "Could not allocate control_code_buffer\n"); |
| 301 | goto out; |
| 302 | } |
| 303 | |
| 304 | result = 0; |
| 305 | out: |
| 306 | if (result == 0) |
| 307 | *rimage = image; |
| 308 | else |
| 309 | kfree(image); |
| 310 | |
| 311 | return result; |
| 312 | } |
| 313 | |
| 314 | static int kimage_is_destination_range(struct kimage *image, |
| 315 | unsigned long start, |
| 316 | unsigned long end) |
| 317 | { |
| 318 | unsigned long i; |
| 319 | |
| 320 | for (i = 0; i < image->nr_segments; i++) { |
| 321 | unsigned long mstart, mend; |
| 322 | |
| 323 | mstart = image->segment[i].mem; |
| 324 | mend = mstart + image->segment[i].memsz; |
| 325 | if ((end > mstart) && (start < mend)) |
| 326 | return 1; |
| 327 | } |
| 328 | |
| 329 | return 0; |
| 330 | } |
| 331 | |
| 332 | static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order) |
| 333 | { |
| 334 | struct page *pages; |
| 335 | |
| 336 | pages = alloc_pages(gfp_mask, order); |
| 337 | if (pages) { |
| 338 | unsigned int count, i; |
| 339 | pages->mapping = NULL; |
| 340 | set_page_private(pages, order); |
| 341 | count = 1 << order; |
| 342 | for (i = 0; i < count; i++) |
| 343 | SetPageReserved(pages + i); |
| 344 | } |
| 345 | |
| 346 | return pages; |
| 347 | } |
| 348 | |
| 349 | static void kimage_free_pages(struct page *page) |
| 350 | { |
| 351 | unsigned int order, count, i; |
| 352 | |
| 353 | order = page_private(page); |
| 354 | count = 1 << order; |
| 355 | for (i = 0; i < count; i++) |
| 356 | ClearPageReserved(page + i); |
| 357 | __free_pages(page, order); |
| 358 | } |
| 359 | |
| 360 | static void kimage_free_page_list(struct list_head *list) |
| 361 | { |
| 362 | struct list_head *pos, *next; |
| 363 | |
| 364 | list_for_each_safe(pos, next, list) { |
| 365 | struct page *page; |
| 366 | |
| 367 | page = list_entry(pos, struct page, lru); |
| 368 | list_del(&page->lru); |
| 369 | kimage_free_pages(page); |
| 370 | } |
| 371 | } |
| 372 | |
| 373 | static struct page *kimage_alloc_normal_control_pages(struct kimage *image, |
| 374 | unsigned int order) |
| 375 | { |
| 376 | /* Control pages are special, they are the intermediaries |
| 377 | * that are needed while we copy the rest of the pages |
| 378 | * to their final resting place. As such they must |
| 379 | * not conflict with either the destination addresses |
| 380 | * or memory the kernel is already using. |
| 381 | * |
| 382 | * The only case where we really need more than one of |
| 383 | * these are for architectures where we cannot disable |
| 384 | * the MMU and must instead generate an identity mapped |
| 385 | * page table for all of the memory. |
| 386 | * |
| 387 | * At worst this runs in O(N) of the image size. |
| 388 | */ |
| 389 | struct list_head extra_pages; |
| 390 | struct page *pages; |
| 391 | unsigned int count; |
| 392 | |
| 393 | count = 1 << order; |
| 394 | INIT_LIST_HEAD(&extra_pages); |
| 395 | |
| 396 | /* Loop while I can allocate a page and the page allocated |
| 397 | * is a destination page. |
| 398 | */ |
| 399 | do { |
| 400 | unsigned long pfn, epfn, addr, eaddr; |
| 401 | |
| 402 | pages = kimage_alloc_pages(GFP_KERNEL, order); |
| 403 | if (!pages) |
| 404 | break; |
| 405 | pfn = page_to_pfn(pages); |
| 406 | epfn = pfn + count; |
| 407 | addr = pfn << PAGE_SHIFT; |
| 408 | eaddr = epfn << PAGE_SHIFT; |
| 409 | if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) || |
| 410 | kimage_is_destination_range(image, addr, eaddr)) { |
| 411 | list_add(&pages->lru, &extra_pages); |
| 412 | pages = NULL; |
| 413 | } |
| 414 | } while (!pages); |
| 415 | |
| 416 | if (pages) { |
| 417 | /* Remember the allocated page... */ |
| 418 | list_add(&pages->lru, &image->control_pages); |
| 419 | |
| 420 | /* Because the page is already in it's destination |
| 421 | * location we will never allocate another page at |
| 422 | * that address. Therefore kimage_alloc_pages |
| 423 | * will not return it (again) and we don't need |
| 424 | * to give it an entry in image->segment[]. |
| 425 | */ |
| 426 | } |
| 427 | /* Deal with the destination pages I have inadvertently allocated. |
| 428 | * |
| 429 | * Ideally I would convert multi-page allocations into single |
| 430 | * page allocations, and add everyting to image->dest_pages. |
| 431 | * |
| 432 | * For now it is simpler to just free the pages. |
| 433 | */ |
| 434 | kimage_free_page_list(&extra_pages); |
| 435 | |
| 436 | return pages; |
| 437 | } |
| 438 | |
| 439 | static struct page *kimage_alloc_crash_control_pages(struct kimage *image, |
| 440 | unsigned int order) |
| 441 | { |
| 442 | /* Control pages are special, they are the intermediaries |
| 443 | * that are needed while we copy the rest of the pages |
| 444 | * to their final resting place. As such they must |
| 445 | * not conflict with either the destination addresses |
| 446 | * or memory the kernel is already using. |
| 447 | * |
| 448 | * Control pages are also the only pags we must allocate |
| 449 | * when loading a crash kernel. All of the other pages |
| 450 | * are specified by the segments and we just memcpy |
| 451 | * into them directly. |
| 452 | * |
| 453 | * The only case where we really need more than one of |
| 454 | * these are for architectures where we cannot disable |
| 455 | * the MMU and must instead generate an identity mapped |
| 456 | * page table for all of the memory. |
| 457 | * |
| 458 | * Given the low demand this implements a very simple |
| 459 | * allocator that finds the first hole of the appropriate |
| 460 | * size in the reserved memory region, and allocates all |
| 461 | * of the memory up to and including the hole. |
| 462 | */ |
| 463 | unsigned long hole_start, hole_end, size; |
| 464 | struct page *pages; |
| 465 | |
| 466 | pages = NULL; |
| 467 | size = (1 << order) << PAGE_SHIFT; |
| 468 | hole_start = (image->control_page + (size - 1)) & ~(size - 1); |
| 469 | hole_end = hole_start + size - 1; |
| 470 | while (hole_end <= crashk_res.end) { |
| 471 | unsigned long i; |
| 472 | |
| 473 | if (hole_end > KEXEC_CONTROL_MEMORY_LIMIT) |
| 474 | break; |
| 475 | if (hole_end > crashk_res.end) |
| 476 | break; |
| 477 | /* See if I overlap any of the segments */ |
| 478 | for (i = 0; i < image->nr_segments; i++) { |
| 479 | unsigned long mstart, mend; |
| 480 | |
| 481 | mstart = image->segment[i].mem; |
| 482 | mend = mstart + image->segment[i].memsz - 1; |
| 483 | if ((hole_end >= mstart) && (hole_start <= mend)) { |
| 484 | /* Advance the hole to the end of the segment */ |
| 485 | hole_start = (mend + (size - 1)) & ~(size - 1); |
| 486 | hole_end = hole_start + size - 1; |
| 487 | break; |
| 488 | } |
| 489 | } |
| 490 | /* If I don't overlap any segments I have found my hole! */ |
| 491 | if (i == image->nr_segments) { |
| 492 | pages = pfn_to_page(hole_start >> PAGE_SHIFT); |
| 493 | break; |
| 494 | } |
| 495 | } |
| 496 | if (pages) |
| 497 | image->control_page = hole_end; |
| 498 | |
| 499 | return pages; |
| 500 | } |
| 501 | |
| 502 | |
| 503 | struct page *kimage_alloc_control_pages(struct kimage *image, |
| 504 | unsigned int order) |
| 505 | { |
| 506 | struct page *pages = NULL; |
| 507 | |
| 508 | switch (image->type) { |
| 509 | case KEXEC_TYPE_DEFAULT: |
| 510 | pages = kimage_alloc_normal_control_pages(image, order); |
| 511 | break; |
| 512 | case KEXEC_TYPE_CRASH: |
| 513 | pages = kimage_alloc_crash_control_pages(image, order); |
| 514 | break; |
| 515 | } |
| 516 | |
| 517 | return pages; |
| 518 | } |
| 519 | |
| 520 | static int kimage_add_entry(struct kimage *image, kimage_entry_t entry) |
| 521 | { |
| 522 | if (*image->entry != 0) |
| 523 | image->entry++; |
| 524 | |
| 525 | if (image->entry == image->last_entry) { |
| 526 | kimage_entry_t *ind_page; |
| 527 | struct page *page; |
| 528 | |
| 529 | page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST); |
| 530 | if (!page) |
| 531 | return -ENOMEM; |
| 532 | |
| 533 | ind_page = page_address(page); |
| 534 | *image->entry = virt_to_phys(ind_page) | IND_INDIRECTION; |
| 535 | image->entry = ind_page; |
| 536 | image->last_entry = ind_page + |
| 537 | ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1); |
| 538 | } |
| 539 | *image->entry = entry; |
| 540 | image->entry++; |
| 541 | *image->entry = 0; |
| 542 | |
| 543 | return 0; |
| 544 | } |
| 545 | |
| 546 | static int kimage_set_destination(struct kimage *image, |
| 547 | unsigned long destination) |
| 548 | { |
| 549 | int result; |
| 550 | |
| 551 | destination &= PAGE_MASK; |
| 552 | result = kimage_add_entry(image, destination | IND_DESTINATION); |
| 553 | if (result == 0) |
| 554 | image->destination = destination; |
| 555 | |
| 556 | return result; |
| 557 | } |
| 558 | |
| 559 | |
| 560 | static int kimage_add_page(struct kimage *image, unsigned long page) |
| 561 | { |
| 562 | int result; |
| 563 | |
| 564 | page &= PAGE_MASK; |
| 565 | result = kimage_add_entry(image, page | IND_SOURCE); |
| 566 | if (result == 0) |
| 567 | image->destination += PAGE_SIZE; |
| 568 | |
| 569 | return result; |
| 570 | } |
| 571 | |
| 572 | |
| 573 | static void kimage_free_extra_pages(struct kimage *image) |
| 574 | { |
| 575 | /* Walk through and free any extra destination pages I may have */ |
| 576 | kimage_free_page_list(&image->dest_pages); |
| 577 | |
| 578 | /* Walk through and free any unuseable pages I have cached */ |
| 579 | kimage_free_page_list(&image->unuseable_pages); |
| 580 | |
| 581 | } |
| 582 | static int kimage_terminate(struct kimage *image) |
| 583 | { |
| 584 | if (*image->entry != 0) |
| 585 | image->entry++; |
| 586 | |
| 587 | *image->entry = IND_DONE; |
| 588 | |
| 589 | return 0; |
| 590 | } |
| 591 | |
| 592 | #define for_each_kimage_entry(image, ptr, entry) \ |
| 593 | for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \ |
| 594 | ptr = (entry & IND_INDIRECTION)? \ |
| 595 | phys_to_virt((entry & PAGE_MASK)): ptr +1) |
| 596 | |
| 597 | static void kimage_free_entry(kimage_entry_t entry) |
| 598 | { |
| 599 | struct page *page; |
| 600 | |
| 601 | page = pfn_to_page(entry >> PAGE_SHIFT); |
| 602 | kimage_free_pages(page); |
| 603 | } |
| 604 | |
| 605 | static void kimage_free(struct kimage *image) |
| 606 | { |
| 607 | kimage_entry_t *ptr, entry; |
| 608 | kimage_entry_t ind = 0; |
| 609 | |
| 610 | if (!image) |
| 611 | return; |
| 612 | |
| 613 | kimage_free_extra_pages(image); |
| 614 | for_each_kimage_entry(image, ptr, entry) { |
| 615 | if (entry & IND_INDIRECTION) { |
| 616 | /* Free the previous indirection page */ |
| 617 | if (ind & IND_INDIRECTION) |
| 618 | kimage_free_entry(ind); |
| 619 | /* Save this indirection page until we are |
| 620 | * done with it. |
| 621 | */ |
| 622 | ind = entry; |
| 623 | } |
| 624 | else if (entry & IND_SOURCE) |
| 625 | kimage_free_entry(entry); |
| 626 | } |
| 627 | /* Free the final indirection page */ |
| 628 | if (ind & IND_INDIRECTION) |
| 629 | kimage_free_entry(ind); |
| 630 | |
| 631 | /* Handle any machine specific cleanup */ |
| 632 | machine_kexec_cleanup(image); |
| 633 | |
| 634 | /* Free the kexec control pages... */ |
| 635 | kimage_free_page_list(&image->control_pages); |
| 636 | kfree(image); |
| 637 | } |
| 638 | |
| 639 | static kimage_entry_t *kimage_dst_used(struct kimage *image, |
| 640 | unsigned long page) |
| 641 | { |
| 642 | kimage_entry_t *ptr, entry; |
| 643 | unsigned long destination = 0; |
| 644 | |
| 645 | for_each_kimage_entry(image, ptr, entry) { |
| 646 | if (entry & IND_DESTINATION) |
| 647 | destination = entry & PAGE_MASK; |
| 648 | else if (entry & IND_SOURCE) { |
| 649 | if (page == destination) |
| 650 | return ptr; |
| 651 | destination += PAGE_SIZE; |
| 652 | } |
| 653 | } |
| 654 | |
| 655 | return NULL; |
| 656 | } |
| 657 | |
| 658 | static struct page *kimage_alloc_page(struct kimage *image, |
| 659 | gfp_t gfp_mask, |
| 660 | unsigned long destination) |
| 661 | { |
| 662 | /* |
| 663 | * Here we implement safeguards to ensure that a source page |
| 664 | * is not copied to its destination page before the data on |
| 665 | * the destination page is no longer useful. |
| 666 | * |
| 667 | * To do this we maintain the invariant that a source page is |
| 668 | * either its own destination page, or it is not a |
| 669 | * destination page at all. |
| 670 | * |
| 671 | * That is slightly stronger than required, but the proof |
| 672 | * that no problems will not occur is trivial, and the |
| 673 | * implementation is simply to verify. |
| 674 | * |
| 675 | * When allocating all pages normally this algorithm will run |
| 676 | * in O(N) time, but in the worst case it will run in O(N^2) |
| 677 | * time. If the runtime is a problem the data structures can |
| 678 | * be fixed. |
| 679 | */ |
| 680 | struct page *page; |
| 681 | unsigned long addr; |
| 682 | |
| 683 | /* |
| 684 | * Walk through the list of destination pages, and see if I |
| 685 | * have a match. |
| 686 | */ |
| 687 | list_for_each_entry(page, &image->dest_pages, lru) { |
| 688 | addr = page_to_pfn(page) << PAGE_SHIFT; |
| 689 | if (addr == destination) { |
| 690 | list_del(&page->lru); |
| 691 | return page; |
| 692 | } |
| 693 | } |
| 694 | page = NULL; |
| 695 | while (1) { |
| 696 | kimage_entry_t *old; |
| 697 | |
| 698 | /* Allocate a page, if we run out of memory give up */ |
| 699 | page = kimage_alloc_pages(gfp_mask, 0); |
| 700 | if (!page) |
| 701 | return NULL; |
| 702 | /* If the page cannot be used file it away */ |
| 703 | if (page_to_pfn(page) > |
| 704 | (KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) { |
| 705 | list_add(&page->lru, &image->unuseable_pages); |
| 706 | continue; |
| 707 | } |
| 708 | addr = page_to_pfn(page) << PAGE_SHIFT; |
| 709 | |
| 710 | /* If it is the destination page we want use it */ |
| 711 | if (addr == destination) |
| 712 | break; |
| 713 | |
| 714 | /* If the page is not a destination page use it */ |
| 715 | if (!kimage_is_destination_range(image, addr, |
| 716 | addr + PAGE_SIZE)) |
| 717 | break; |
| 718 | |
| 719 | /* |
| 720 | * I know that the page is someones destination page. |
| 721 | * See if there is already a source page for this |
| 722 | * destination page. And if so swap the source pages. |
| 723 | */ |
| 724 | old = kimage_dst_used(image, addr); |
| 725 | if (old) { |
| 726 | /* If so move it */ |
| 727 | unsigned long old_addr; |
| 728 | struct page *old_page; |
| 729 | |
| 730 | old_addr = *old & PAGE_MASK; |
| 731 | old_page = pfn_to_page(old_addr >> PAGE_SHIFT); |
| 732 | copy_highpage(page, old_page); |
| 733 | *old = addr | (*old & ~PAGE_MASK); |
| 734 | |
| 735 | /* The old page I have found cannot be a |
| 736 | * destination page, so return it. |
| 737 | */ |
| 738 | addr = old_addr; |
| 739 | page = old_page; |
| 740 | break; |
| 741 | } |
| 742 | else { |
| 743 | /* Place the page on the destination list I |
| 744 | * will use it later. |
| 745 | */ |
| 746 | list_add(&page->lru, &image->dest_pages); |
| 747 | } |
| 748 | } |
| 749 | |
| 750 | return page; |
| 751 | } |
| 752 | |
| 753 | static int kimage_load_normal_segment(struct kimage *image, |
| 754 | struct kexec_segment *segment) |
| 755 | { |
| 756 | unsigned long maddr; |
| 757 | unsigned long ubytes, mbytes; |
| 758 | int result; |
| 759 | unsigned char __user *buf; |
| 760 | |
| 761 | result = 0; |
| 762 | buf = segment->buf; |
| 763 | ubytes = segment->bufsz; |
| 764 | mbytes = segment->memsz; |
| 765 | maddr = segment->mem; |
| 766 | |
| 767 | result = kimage_set_destination(image, maddr); |
| 768 | if (result < 0) |
| 769 | goto out; |
| 770 | |
| 771 | while (mbytes) { |
| 772 | struct page *page; |
| 773 | char *ptr; |
| 774 | size_t uchunk, mchunk; |
| 775 | |
| 776 | page = kimage_alloc_page(image, GFP_HIGHUSER, maddr); |
| 777 | if (page == 0) { |
| 778 | result = -ENOMEM; |
| 779 | goto out; |
| 780 | } |
| 781 | result = kimage_add_page(image, page_to_pfn(page) |
| 782 | << PAGE_SHIFT); |
| 783 | if (result < 0) |
| 784 | goto out; |
| 785 | |
| 786 | ptr = kmap(page); |
| 787 | /* Start with a clear page */ |
| 788 | memset(ptr, 0, PAGE_SIZE); |
| 789 | ptr += maddr & ~PAGE_MASK; |
| 790 | mchunk = PAGE_SIZE - (maddr & ~PAGE_MASK); |
| 791 | if (mchunk > mbytes) |
| 792 | mchunk = mbytes; |
| 793 | |
| 794 | uchunk = mchunk; |
| 795 | if (uchunk > ubytes) |
| 796 | uchunk = ubytes; |
| 797 | |
| 798 | result = copy_from_user(ptr, buf, uchunk); |
| 799 | kunmap(page); |
| 800 | if (result) { |
| 801 | result = (result < 0) ? result : -EIO; |
| 802 | goto out; |
| 803 | } |
| 804 | ubytes -= uchunk; |
| 805 | maddr += mchunk; |
| 806 | buf += mchunk; |
| 807 | mbytes -= mchunk; |
| 808 | } |
| 809 | out: |
| 810 | return result; |
| 811 | } |
| 812 | |
| 813 | static int kimage_load_crash_segment(struct kimage *image, |
| 814 | struct kexec_segment *segment) |
| 815 | { |
| 816 | /* For crash dumps kernels we simply copy the data from |
| 817 | * user space to it's destination. |
| 818 | * We do things a page at a time for the sake of kmap. |
| 819 | */ |
| 820 | unsigned long maddr; |
| 821 | unsigned long ubytes, mbytes; |
| 822 | int result; |
| 823 | unsigned char __user *buf; |
| 824 | |
| 825 | result = 0; |
| 826 | buf = segment->buf; |
| 827 | ubytes = segment->bufsz; |
| 828 | mbytes = segment->memsz; |
| 829 | maddr = segment->mem; |
| 830 | while (mbytes) { |
| 831 | struct page *page; |
| 832 | char *ptr; |
| 833 | size_t uchunk, mchunk; |
| 834 | |
| 835 | page = pfn_to_page(maddr >> PAGE_SHIFT); |
| 836 | if (page == 0) { |
| 837 | result = -ENOMEM; |
| 838 | goto out; |
| 839 | } |
| 840 | ptr = kmap(page); |
| 841 | ptr += maddr & ~PAGE_MASK; |
| 842 | mchunk = PAGE_SIZE - (maddr & ~PAGE_MASK); |
| 843 | if (mchunk > mbytes) |
| 844 | mchunk = mbytes; |
| 845 | |
| 846 | uchunk = mchunk; |
| 847 | if (uchunk > ubytes) { |
| 848 | uchunk = ubytes; |
| 849 | /* Zero the trailing part of the page */ |
| 850 | memset(ptr + uchunk, 0, mchunk - uchunk); |
| 851 | } |
| 852 | result = copy_from_user(ptr, buf, uchunk); |
| 853 | kunmap(page); |
| 854 | if (result) { |
| 855 | result = (result < 0) ? result : -EIO; |
| 856 | goto out; |
| 857 | } |
| 858 | ubytes -= uchunk; |
| 859 | maddr += mchunk; |
| 860 | buf += mchunk; |
| 861 | mbytes -= mchunk; |
| 862 | } |
| 863 | out: |
| 864 | return result; |
| 865 | } |
| 866 | |
| 867 | static int kimage_load_segment(struct kimage *image, |
| 868 | struct kexec_segment *segment) |
| 869 | { |
| 870 | int result = -ENOMEM; |
| 871 | |
| 872 | switch (image->type) { |
| 873 | case KEXEC_TYPE_DEFAULT: |
| 874 | result = kimage_load_normal_segment(image, segment); |
| 875 | break; |
| 876 | case KEXEC_TYPE_CRASH: |
| 877 | result = kimage_load_crash_segment(image, segment); |
| 878 | break; |
| 879 | } |
| 880 | |
| 881 | return result; |
| 882 | } |
| 883 | |
| 884 | /* |
| 885 | * Exec Kernel system call: for obvious reasons only root may call it. |
| 886 | * |
| 887 | * This call breaks up into three pieces. |
| 888 | * - A generic part which loads the new kernel from the current |
| 889 | * address space, and very carefully places the data in the |
| 890 | * allocated pages. |
| 891 | * |
| 892 | * - A generic part that interacts with the kernel and tells all of |
| 893 | * the devices to shut down. Preventing on-going dmas, and placing |
| 894 | * the devices in a consistent state so a later kernel can |
| 895 | * reinitialize them. |
| 896 | * |
| 897 | * - A machine specific part that includes the syscall number |
| 898 | * and the copies the image to it's final destination. And |
| 899 | * jumps into the image at entry. |
| 900 | * |
| 901 | * kexec does not sync, or unmount filesystems so if you need |
| 902 | * that to happen you need to do that yourself. |
| 903 | */ |
| 904 | struct kimage *kexec_image; |
| 905 | struct kimage *kexec_crash_image; |
| 906 | /* |
| 907 | * A home grown binary mutex. |
| 908 | * Nothing can wait so this mutex is safe to use |
| 909 | * in interrupt context :) |
| 910 | */ |
| 911 | static int kexec_lock; |
| 912 | |
| 913 | asmlinkage long sys_kexec_load(unsigned long entry, unsigned long nr_segments, |
| 914 | struct kexec_segment __user *segments, |
| 915 | unsigned long flags) |
| 916 | { |
| 917 | struct kimage **dest_image, *image; |
| 918 | int locked; |
| 919 | int result; |
| 920 | |
| 921 | /* We only trust the superuser with rebooting the system. */ |
| 922 | if (!capable(CAP_SYS_BOOT)) |
| 923 | return -EPERM; |
| 924 | |
| 925 | /* |
| 926 | * Verify we have a legal set of flags |
| 927 | * This leaves us room for future extensions. |
| 928 | */ |
| 929 | if ((flags & KEXEC_FLAGS) != (flags & ~KEXEC_ARCH_MASK)) |
| 930 | return -EINVAL; |
| 931 | |
| 932 | /* Verify we are on the appropriate architecture */ |
| 933 | if (((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH) && |
| 934 | ((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH_DEFAULT)) |
| 935 | return -EINVAL; |
| 936 | |
| 937 | /* Put an artificial cap on the number |
| 938 | * of segments passed to kexec_load. |
| 939 | */ |
| 940 | if (nr_segments > KEXEC_SEGMENT_MAX) |
| 941 | return -EINVAL; |
| 942 | |
| 943 | image = NULL; |
| 944 | result = 0; |
| 945 | |
| 946 | /* Because we write directly to the reserved memory |
| 947 | * region when loading crash kernels we need a mutex here to |
| 948 | * prevent multiple crash kernels from attempting to load |
| 949 | * simultaneously, and to prevent a crash kernel from loading |
| 950 | * over the top of a in use crash kernel. |
| 951 | * |
| 952 | * KISS: always take the mutex. |
| 953 | */ |
| 954 | locked = xchg(&kexec_lock, 1); |
| 955 | if (locked) |
| 956 | return -EBUSY; |
| 957 | |
| 958 | dest_image = &kexec_image; |
| 959 | if (flags & KEXEC_ON_CRASH) |
| 960 | dest_image = &kexec_crash_image; |
| 961 | if (nr_segments > 0) { |
| 962 | unsigned long i; |
| 963 | |
| 964 | /* Loading another kernel to reboot into */ |
| 965 | if ((flags & KEXEC_ON_CRASH) == 0) |
| 966 | result = kimage_normal_alloc(&image, entry, |
| 967 | nr_segments, segments); |
| 968 | /* Loading another kernel to switch to if this one crashes */ |
| 969 | else if (flags & KEXEC_ON_CRASH) { |
| 970 | /* Free any current crash dump kernel before |
| 971 | * we corrupt it. |
| 972 | */ |
| 973 | kimage_free(xchg(&kexec_crash_image, NULL)); |
| 974 | result = kimage_crash_alloc(&image, entry, |
| 975 | nr_segments, segments); |
| 976 | } |
| 977 | if (result) |
| 978 | goto out; |
| 979 | |
| 980 | result = machine_kexec_prepare(image); |
| 981 | if (result) |
| 982 | goto out; |
| 983 | |
| 984 | for (i = 0; i < nr_segments; i++) { |
| 985 | result = kimage_load_segment(image, &image->segment[i]); |
| 986 | if (result) |
| 987 | goto out; |
| 988 | } |
| 989 | result = kimage_terminate(image); |
| 990 | if (result) |
| 991 | goto out; |
| 992 | } |
| 993 | /* Install the new kernel, and Uninstall the old */ |
| 994 | image = xchg(dest_image, image); |
| 995 | |
| 996 | out: |
| 997 | locked = xchg(&kexec_lock, 0); /* Release the mutex */ |
| 998 | BUG_ON(!locked); |
| 999 | kimage_free(image); |
| 1000 | |
| 1001 | return result; |
| 1002 | } |
| 1003 | |
| 1004 | #ifdef CONFIG_COMPAT |
| 1005 | asmlinkage long compat_sys_kexec_load(unsigned long entry, |
| 1006 | unsigned long nr_segments, |
| 1007 | struct compat_kexec_segment __user *segments, |
| 1008 | unsigned long flags) |
| 1009 | { |
| 1010 | struct compat_kexec_segment in; |
| 1011 | struct kexec_segment out, __user *ksegments; |
| 1012 | unsigned long i, result; |
| 1013 | |
| 1014 | /* Don't allow clients that don't understand the native |
| 1015 | * architecture to do anything. |
| 1016 | */ |
| 1017 | if ((flags & KEXEC_ARCH_MASK) == KEXEC_ARCH_DEFAULT) |
| 1018 | return -EINVAL; |
| 1019 | |
| 1020 | if (nr_segments > KEXEC_SEGMENT_MAX) |
| 1021 | return -EINVAL; |
| 1022 | |
| 1023 | ksegments = compat_alloc_user_space(nr_segments * sizeof(out)); |
| 1024 | for (i=0; i < nr_segments; i++) { |
| 1025 | result = copy_from_user(&in, &segments[i], sizeof(in)); |
| 1026 | if (result) |
| 1027 | return -EFAULT; |
| 1028 | |
| 1029 | out.buf = compat_ptr(in.buf); |
| 1030 | out.bufsz = in.bufsz; |
| 1031 | out.mem = in.mem; |
| 1032 | out.memsz = in.memsz; |
| 1033 | |
| 1034 | result = copy_to_user(&ksegments[i], &out, sizeof(out)); |
| 1035 | if (result) |
| 1036 | return -EFAULT; |
| 1037 | } |
| 1038 | |
| 1039 | return sys_kexec_load(entry, nr_segments, ksegments, flags); |
| 1040 | } |
| 1041 | #endif |
| 1042 | |
| 1043 | void crash_kexec(struct pt_regs *regs) |
| 1044 | { |
| 1045 | int locked; |
| 1046 | |
| 1047 | |
| 1048 | /* Take the kexec_lock here to prevent sys_kexec_load |
| 1049 | * running on one cpu from replacing the crash kernel |
| 1050 | * we are using after a panic on a different cpu. |
| 1051 | * |
| 1052 | * If the crash kernel was not located in a fixed area |
| 1053 | * of memory the xchg(&kexec_crash_image) would be |
| 1054 | * sufficient. But since I reuse the memory... |
| 1055 | */ |
| 1056 | locked = xchg(&kexec_lock, 1); |
| 1057 | if (!locked) { |
| 1058 | if (kexec_crash_image) { |
| 1059 | struct pt_regs fixed_regs; |
| 1060 | crash_setup_regs(&fixed_regs, regs); |
| 1061 | machine_crash_shutdown(&fixed_regs); |
| 1062 | machine_kexec(kexec_crash_image); |
| 1063 | } |
| 1064 | locked = xchg(&kexec_lock, 0); |
| 1065 | BUG_ON(!locked); |
| 1066 | } |
| 1067 | } |
| 1068 | |
| 1069 | static int __init crash_notes_memory_init(void) |
| 1070 | { |
| 1071 | /* Allocate memory for saving cpu registers. */ |
| 1072 | crash_notes = alloc_percpu(note_buf_t); |
| 1073 | if (!crash_notes) { |
| 1074 | printk("Kexec: Memory allocation for saving cpu register" |
| 1075 | " states failed\n"); |
| 1076 | return -ENOMEM; |
| 1077 | } |
| 1078 | return 0; |
| 1079 | } |
| 1080 | module_init(crash_notes_memory_init) |