1 | /* |
2 | * Copyright (c) 2009-2016 Apple Inc. All rights reserved. |
3 | * |
4 | * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ |
5 | * |
6 | * This file contains Original Code and/or Modifications of Original Code |
7 | * as defined in and that are subject to the Apple Public Source License |
8 | * Version 2.0 (the 'License'). You may not use this file except in |
9 | * compliance with the License. The rights granted to you under the License |
10 | * may not be used to create, or enable the creation or redistribution of, |
11 | * unlawful or unlicensed copies of an Apple operating system, or to |
12 | * circumvent, violate, or enable the circumvention or violation of, any |
13 | * terms of an Apple operating system software license agreement. |
14 | * |
15 | * Please obtain a copy of the License at |
16 | * http://www.opensource.apple.com/apsl/ and read it before using this file. |
17 | * |
18 | * The Original Code and all software distributed under the License are |
19 | * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER |
20 | * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, |
21 | * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, |
22 | * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. |
23 | * Please see the License for the specific language governing rights and |
24 | * limitations under the License. |
25 | * |
26 | * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ |
27 | */ |
28 | #include <mach_assert.h> |
29 | |
30 | #include <sys/errno.h> |
31 | #include <i386/param.h> |
32 | #include <i386/misc_protos.h> |
33 | #include <i386/cpu_data.h> |
34 | #include <i386/machine_routines.h> |
35 | #include <i386/cpuid.h> |
36 | #include <i386/vmx.h> |
37 | #include <vm/pmap.h> |
38 | #include <vm/vm_map.h> |
39 | #include <vm/vm_kern.h> |
40 | #include <vm/vm_fault.h> |
41 | #include <san/kasan.h> |
42 | |
43 | #include <sys/kdebug.h> |
44 | |
45 | #include <kern/copyout_shim.h> |
46 | |
47 | #undef copyin |
48 | #undef copyout |
49 | |
50 | static int copyio(int, user_addr_t, char *, vm_size_t, vm_size_t *, int); |
51 | static int copyio_phys(addr64_t, addr64_t, vm_size_t, int); |
52 | |
53 | /* |
54 | * Copy sizes bigger than this value will cause a kernel panic. |
55 | * |
56 | * Yes, this is an arbitrary fixed limit, but it's almost certainly |
57 | * a programming error to be copying more than this amount between |
58 | * user and wired kernel memory in a single invocation on this |
59 | * platform. |
60 | */ |
61 | const int copysize_limit_panic = (64 * MB); |
62 | |
63 | /* |
64 | * The copy engine has the following characteristics |
65 | * - copyio() handles copies to/from user or kernel space |
66 | * - copypv() deals with physical or virtual addresses |
67 | * |
68 | * Readers familiar with the 32-bit kernel will expect Joe's thesis at this |
69 | * point describing the full glory of the copy window implementation. In K64, |
70 | * however, there is no need for windowing. Thanks to the vast shared address |
71 | * space, the kernel has direct access to userspace and to physical memory. |
72 | * |
73 | * User virtual addresses are accessible provided the user's cr3 is loaded. |
74 | * Physical addresses are accessible via the direct map and the PHYSMAP_PTOV() |
75 | * translation. |
76 | * |
77 | * Copyin/out variants all boil done to just these 2 routines in locore.s which |
78 | * provide fault-recoverable copying: |
79 | */ |
80 | extern int _bcopy(const void *, void *, vm_size_t); |
81 | extern int _bcopystr(const void *, void *, vm_size_t, vm_size_t *); |
82 | extern int _copyin_word(const char *src, uint64_t *dst, vm_size_t len); |
83 | |
84 | /* On by default, optionally disabled by boot-arg */ |
85 | extern boolean_t copyio_zalloc_check; |
86 | |
87 | /* |
88 | * Types of copies: |
89 | */ |
90 | #define COPYIN 0 /* from user virtual to kernel virtual */ |
91 | #define COPYOUT 1 /* from kernel virtual to user virtual */ |
92 | #define COPYINSTR 2 /* string variant of copyout */ |
93 | #define COPYINPHYS 3 /* from user virtual to kernel physical */ |
94 | #define COPYOUTPHYS 4 /* from kernel physical to user virtual */ |
95 | #define COPYINWORD 5 /* from user virtual to kernel virtual */ |
96 | |
97 | #if ENABLE_SMAPLOG |
98 | typedef struct { |
99 | uint64_t timestamp; |
100 | thread_t thread; |
101 | uintptr_t cr4; |
102 | uint8_t cpuid; |
103 | uint8_t smap_state; |
104 | uint8_t copyio_active; |
105 | } smaplog_entry_t; |
106 | |
107 | #define SMAPLOG_BUFFER_SIZE (50) |
108 | static smaplog_entry_t smaplog_cbuf[SMAPLOG_BUFFER_SIZE]; |
109 | static uint32_t smaplog_head = 0; |
110 | |
111 | static void |
112 | smaplog_add_entry(boolean_t enabling) |
113 | { |
114 | uint32_t index = 0; |
115 | thread_t thread = current_thread(); |
116 | |
117 | do { |
118 | index = smaplog_head; |
119 | } while (!OSCompareAndSwap(index, (index + 1) % SMAPLOG_BUFFER_SIZE, &smaplog_head)); |
120 | |
121 | assert(index < SMAPLOG_BUFFER_SIZE); |
122 | assert(smaplog_head < SMAPLOG_BUFFER_SIZE); |
123 | assert(thread); |
124 | |
125 | smaplog_cbuf[index].timestamp = mach_absolute_time(); |
126 | smaplog_cbuf[index].thread = thread; |
127 | smaplog_cbuf[index].cpuid = cpu_number(); |
128 | smaplog_cbuf[index].cr4 = get_cr4(); |
129 | smaplog_cbuf[index].smap_state = enabling; |
130 | smaplog_cbuf[index].copyio_active = (thread->machine.specFlags & CopyIOActive) ? 1 : 0; |
131 | } |
132 | #endif /* ENABLE_SMAPLOG */ |
133 | |
134 | extern boolean_t pmap_smap_enabled; |
135 | static inline void user_access_enable(void) { |
136 | if (pmap_smap_enabled) { |
137 | stac(); |
138 | #if ENABLE_SMAPLOG |
139 | smaplog_add_entry(TRUE); |
140 | #endif |
141 | } |
142 | } |
143 | static inline void user_access_disable(void) { |
144 | if (pmap_smap_enabled) { |
145 | clac(); |
146 | #if ENABLE_SMAPLOG |
147 | smaplog_add_entry(FALSE); |
148 | #endif |
149 | } |
150 | } |
151 | |
152 | #if COPYIO_TRACE_ENABLED |
153 | #define COPYIO_TRACE(x, a, b, c, d, e) KERNEL_DEBUG_CONSTANT(x, a, b, c, d, e) |
154 | #else |
155 | #define COPYIO_TRACE(x, a, b, c, d, e) do { } while(0) |
156 | #endif |
157 | |
158 | static int |
159 | copyio(int copy_type, user_addr_t user_addr, char *kernel_addr, |
160 | vm_size_t nbytes, vm_size_t *lencopied, int use_kernel_map) |
161 | { |
162 | thread_t thread = current_thread(); |
163 | pmap_t pmap; |
164 | vm_size_t bytes_copied; |
165 | int error = 0; |
166 | boolean_t istate = FALSE; |
167 | boolean_t recursive_CopyIOActive; |
168 | #if COPYIO_TRACE_ENABLED |
169 | int debug_type = 0xeff70010; |
170 | debug_type += (copy_type << 2); |
171 | #endif |
172 | vm_size_t kernel_buf_size = 0; |
173 | |
174 | if (__improbable(nbytes > copysize_limit_panic)) |
175 | panic("%s(%p, %p, %lu) - transfer too large" , __func__, |
176 | (void *)user_addr, (void *)kernel_addr, nbytes); |
177 | |
178 | COPYIO_TRACE(debug_type | DBG_FUNC_START, |
179 | user_addr, kernel_addr, nbytes, use_kernel_map, 0); |
180 | |
181 | if (__improbable(nbytes == 0)) |
182 | goto out; |
183 | |
184 | pmap = thread->map->pmap; |
185 | boolean_t nopagezero = thread->map->pmap->pagezero_accessible; |
186 | |
187 | if ((copy_type != COPYINPHYS) && (copy_type != COPYOUTPHYS)) { |
188 | if (__improbable((vm_offset_t)kernel_addr < VM_MIN_KERNEL_AND_KEXT_ADDRESS)) |
189 | panic("Invalid copy parameter, copy type: %d, kernel address: %p" , copy_type, kernel_addr); |
190 | if (__probable(copyio_zalloc_check)) { |
191 | kernel_buf_size = zone_element_size(kernel_addr, NULL); |
192 | if (__improbable(kernel_buf_size && kernel_buf_size < nbytes)) |
193 | panic("copyio: kernel buffer %p has size %lu < nbytes %lu" , kernel_addr, kernel_buf_size, nbytes); |
194 | } |
195 | } |
196 | |
197 | /* Sanity and security check for addresses to/from a user */ |
198 | |
199 | if (__improbable(((pmap != kernel_pmap) && (use_kernel_map == 0)) && |
200 | ((nbytes && (user_addr+nbytes <= user_addr)) || ((user_addr + nbytes) > vm_map_max(thread->map))))) { |
201 | error = EFAULT; |
202 | goto out; |
203 | } |
204 | |
205 | #if KASAN |
206 | if (copy_type == COPYIN || copy_type == COPYINSTR || copy_type == COPYINWORD) { |
207 | __asan_storeN((uptr)kernel_addr, nbytes); |
208 | } else if (copy_type == COPYOUT) { |
209 | __asan_loadN((uptr)kernel_addr, nbytes); |
210 | } |
211 | #endif |
212 | |
213 | /* |
214 | * If the no_shared_cr3 boot-arg is set (true), the kernel runs on |
215 | * its own pmap and cr3 rather than the user's -- so that wild accesses |
216 | * from kernel or kexts can be trapped. So, during copyin and copyout, |
217 | * we need to switch back to the user's map/cr3. The thread is flagged |
218 | * "CopyIOActive" at this time so that if the thread is pre-empted, |
219 | * we will later restore the correct cr3. |
220 | */ |
221 | recursive_CopyIOActive = thread->machine.specFlags & CopyIOActive; |
222 | |
223 | boolean_t pdswitch = no_shared_cr3 || nopagezero; |
224 | |
225 | if (__improbable(pdswitch)) { |
226 | istate = ml_set_interrupts_enabled(FALSE); |
227 | if (nopagezero && pmap_pcid_ncpus) { |
228 | pmap_pcid_activate(pmap, cpu_number(), TRUE, TRUE); |
229 | } else if (get_cr3_base() != pmap->pm_cr3) { |
230 | set_cr3_raw(pmap->pm_cr3); |
231 | } |
232 | thread->machine.specFlags |= CopyIOActive; |
233 | } else { |
234 | thread->machine.specFlags |= CopyIOActive; |
235 | } |
236 | |
237 | user_access_enable(); |
238 | |
239 | #if DEVELOPMENT || DEBUG |
240 | /* |
241 | * Ensure that we're running on the target thread's cr3. |
242 | */ |
243 | if ((pmap != kernel_pmap) && !use_kernel_map && |
244 | (get_cr3_base() != pmap->pm_cr3)) { |
245 | panic("copyio(%d,%p,%p,%ld,%p,%d) cr3 is %p expects %p" , |
246 | copy_type, (void *)user_addr, kernel_addr, nbytes, lencopied, use_kernel_map, |
247 | (void *) get_cr3_raw(), (void *) pmap->pm_cr3); |
248 | } |
249 | #endif |
250 | |
251 | if (__improbable(pdswitch)) { |
252 | (void) ml_set_interrupts_enabled(istate); |
253 | } |
254 | |
255 | COPYIO_TRACE(0xeff70044 | DBG_FUNC_NONE, user_addr, |
256 | kernel_addr, nbytes, 0, 0); |
257 | |
258 | switch (copy_type) { |
259 | |
260 | case COPYIN: |
261 | error = _bcopy((const void *) user_addr, |
262 | kernel_addr, |
263 | nbytes); |
264 | break; |
265 | |
266 | case COPYOUT: |
267 | error = _bcopy(kernel_addr, |
268 | (void *) user_addr, |
269 | nbytes); |
270 | break; |
271 | |
272 | case COPYINPHYS: |
273 | error = _bcopy((const void *) user_addr, |
274 | PHYSMAP_PTOV(kernel_addr), |
275 | nbytes); |
276 | break; |
277 | |
278 | case COPYOUTPHYS: |
279 | error = _bcopy((const void *) PHYSMAP_PTOV(kernel_addr), |
280 | (void *) user_addr, |
281 | nbytes); |
282 | break; |
283 | |
284 | case COPYINWORD: |
285 | error = _copyin_word((const void *) user_addr, |
286 | (void *) kernel_addr, |
287 | nbytes); |
288 | break; |
289 | |
290 | case COPYINSTR: |
291 | error = _bcopystr((const void *) user_addr, |
292 | kernel_addr, |
293 | (int) nbytes, |
294 | &bytes_copied); |
295 | |
296 | /* |
297 | * lencopied should be updated on success |
298 | * or ENAMETOOLONG... but not EFAULT |
299 | */ |
300 | if (error != EFAULT) |
301 | *lencopied = bytes_copied; |
302 | |
303 | if (error) { |
304 | #if KDEBUG |
305 | nbytes = *lencopied; |
306 | #endif |
307 | break; |
308 | } |
309 | if (*(kernel_addr + bytes_copied - 1) == 0) { |
310 | /* |
311 | * we found a NULL terminator... we're done |
312 | */ |
313 | #if KDEBUG |
314 | nbytes = *lencopied; |
315 | #endif |
316 | break; |
317 | } else { |
318 | /* |
319 | * no more room in the buffer and we haven't |
320 | * yet come across a NULL terminator |
321 | */ |
322 | #if KDEBUG |
323 | nbytes = *lencopied; |
324 | #endif |
325 | error = ENAMETOOLONG; |
326 | break; |
327 | } |
328 | } |
329 | |
330 | user_access_disable(); |
331 | |
332 | if (__improbable(pdswitch)) { |
333 | istate = ml_set_interrupts_enabled(FALSE); |
334 | if (!recursive_CopyIOActive && (get_cr3_raw() != kernel_pmap->pm_cr3)) { |
335 | if (nopagezero && pmap_pcid_ncpus) { |
336 | pmap_pcid_activate(pmap, cpu_number(), TRUE, FALSE); |
337 | } else { |
338 | set_cr3_raw(kernel_pmap->pm_cr3); |
339 | } |
340 | } |
341 | |
342 | if (!recursive_CopyIOActive) { |
343 | thread->machine.specFlags &= ~CopyIOActive; |
344 | } |
345 | (void) ml_set_interrupts_enabled(istate); |
346 | } else if (!recursive_CopyIOActive) { |
347 | thread->machine.specFlags &= ~CopyIOActive; |
348 | } |
349 | |
350 | out: |
351 | COPYIO_TRACE(debug_type | DBG_FUNC_END, user_addr, kernel_addr, nbytes, error, 0); |
352 | |
353 | return (error); |
354 | } |
355 | |
356 | |
357 | static int |
358 | copyio_phys(addr64_t source, addr64_t sink, vm_size_t csize, int which) |
359 | { |
360 | char *paddr; |
361 | user_addr_t vaddr; |
362 | int ctype; |
363 | |
364 | if (which & cppvPsnk) { |
365 | paddr = (char *)sink; |
366 | vaddr = (user_addr_t)source; |
367 | ctype = COPYINPHYS; |
368 | } else { |
369 | paddr = (char *)source; |
370 | vaddr = (user_addr_t)sink; |
371 | ctype = COPYOUTPHYS; |
372 | CALL_COPYOUT_SHIM_PHYS((void *)PHYSMAP_PTOV(source),sink,csize) |
373 | } |
374 | return copyio(ctype, vaddr, paddr, csize, NULL, which & cppvKmap); |
375 | } |
376 | |
377 | int |
378 | copyinmsg(const user_addr_t user_addr, char *kernel_addr, mach_msg_size_t nbytes) |
379 | { |
380 | return copyio(COPYIN, user_addr, kernel_addr, nbytes, NULL, 0); |
381 | } |
382 | |
383 | int |
384 | copyin(const user_addr_t user_addr, void *kernel_addr, vm_size_t nbytes) |
385 | { |
386 | return copyio(COPYIN, user_addr, kernel_addr, nbytes, NULL, 0); |
387 | } |
388 | |
389 | /* |
390 | * copyin_word |
391 | * Read an aligned value from userspace as a single memory transaction. |
392 | * This function supports userspace synchronization features |
393 | */ |
394 | int |
395 | copyin_word(const user_addr_t user_addr, uint64_t *kernel_addr, vm_size_t nbytes) |
396 | { |
397 | /* Verify sizes */ |
398 | if ((nbytes != 4) && (nbytes != 8)) |
399 | return EINVAL; |
400 | |
401 | /* Test alignment */ |
402 | if (user_addr & (nbytes - 1)) |
403 | return EINVAL; |
404 | return copyio(COPYINWORD, user_addr, (char *)(uintptr_t)kernel_addr, nbytes, NULL, 0); |
405 | } |
406 | |
407 | int |
408 | copyinstr(const user_addr_t user_addr, char *kernel_addr, vm_size_t nbytes, vm_size_t *lencopied) |
409 | { |
410 | *lencopied = 0; |
411 | |
412 | return copyio(COPYINSTR, user_addr, kernel_addr, nbytes, lencopied, 0); |
413 | } |
414 | |
415 | int |
416 | copyoutmsg(const char *kernel_addr, user_addr_t user_addr, mach_msg_size_t nbytes) |
417 | { |
418 | CALL_COPYOUT_SHIM_MSG(kernel_addr,user_addr,(vm_size_t)nbytes) |
419 | return copyio(COPYOUT, user_addr, (char *)(uintptr_t)kernel_addr, nbytes, NULL, 0); |
420 | } |
421 | |
422 | int |
423 | copyout(const void *kernel_addr, user_addr_t user_addr, vm_size_t nbytes) |
424 | { |
425 | CALL_COPYOUT_SHIM_NRML(kernel_addr,user_addr,nbytes) |
426 | return copyio(COPYOUT, user_addr, (char *)(uintptr_t)kernel_addr, nbytes, NULL, 0); |
427 | } |
428 | |
429 | |
430 | kern_return_t |
431 | copypv(addr64_t src64, addr64_t snk64, unsigned int size, int which) |
432 | { |
433 | unsigned int lop, csize; |
434 | int bothphys = 0; |
435 | |
436 | KERNEL_DEBUG(0xeff7004c | DBG_FUNC_START, (unsigned)src64, |
437 | (unsigned)snk64, size, which, 0); |
438 | |
439 | if ((which & (cppvPsrc | cppvPsnk)) == 0 ) /* Make sure that only one is virtual */ |
440 | panic("copypv: no more than 1 parameter may be virtual\n" ); /* Not allowed */ |
441 | |
442 | if ((which & (cppvPsrc | cppvPsnk)) == (cppvPsrc | cppvPsnk)) |
443 | bothphys = 1; /* both are physical */ |
444 | |
445 | while (size) { |
446 | |
447 | if (bothphys) { |
448 | lop = (unsigned int)(PAGE_SIZE - (snk64 & (PAGE_SIZE - 1))); /* Assume sink smallest */ |
449 | |
450 | if (lop > (unsigned int)(PAGE_SIZE - (src64 & (PAGE_SIZE - 1)))) |
451 | lop = (unsigned int)(PAGE_SIZE - (src64 & (PAGE_SIZE - 1))); /* No, source is smaller */ |
452 | } else { |
453 | /* |
454 | * only need to compute the resid for the physical page |
455 | * address... we don't care about where we start/finish in |
456 | * the virtual since we just call the normal copyin/copyout |
457 | */ |
458 | if (which & cppvPsrc) |
459 | lop = (unsigned int)(PAGE_SIZE - (src64 & (PAGE_SIZE - 1))); |
460 | else |
461 | lop = (unsigned int)(PAGE_SIZE - (snk64 & (PAGE_SIZE - 1))); |
462 | } |
463 | csize = size; /* Assume we can copy it all */ |
464 | if (lop < size) |
465 | csize = lop; /* Nope, we can't do it all */ |
466 | #if 0 |
467 | /* |
468 | * flush_dcache64 is currently a nop on the i386... |
469 | * it's used when copying to non-system memory such |
470 | * as video capture cards... on PPC there was a need |
471 | * to flush due to how we mapped this memory... not |
472 | * sure if it's needed on i386. |
473 | */ |
474 | if (which & cppvFsrc) |
475 | flush_dcache64(src64, csize, 1); /* If requested, flush source before move */ |
476 | if (which & cppvFsnk) |
477 | flush_dcache64(snk64, csize, 1); /* If requested, flush sink before move */ |
478 | #endif |
479 | if (bothphys) |
480 | bcopy_phys(src64, snk64, csize); /* Do a physical copy, virtually */ |
481 | else { |
482 | if (copyio_phys(src64, snk64, csize, which)) |
483 | return (KERN_FAILURE); |
484 | } |
485 | #if 0 |
486 | if (which & cppvFsrc) |
487 | flush_dcache64(src64, csize, 1); /* If requested, flush source after move */ |
488 | if (which & cppvFsnk) |
489 | flush_dcache64(snk64, csize, 1); /* If requested, flush sink after move */ |
490 | #endif |
491 | size -= csize; /* Calculate what is left */ |
492 | snk64 += csize; /* Bump sink to next physical address */ |
493 | src64 += csize; /* Bump source to next physical address */ |
494 | } |
495 | KERNEL_DEBUG(0xeff7004c | DBG_FUNC_END, (unsigned)src64, |
496 | (unsigned)snk64, size, which, 0); |
497 | |
498 | return KERN_SUCCESS; |
499 | } |
500 | |