uaccess_pt.c 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474
  1. /*
  2. * User access functions based on page table walks for enhanced
  3. * system layout without hardware support.
  4. *
  5. * Copyright IBM Corp. 2006, 2012
  6. * Author(s): Gerald Schaefer (gerald.schaefer@de.ibm.com)
  7. */
  8. #include <linux/errno.h>
  9. #include <linux/hardirq.h>
  10. #include <linux/mm.h>
  11. #include <linux/hugetlb.h>
  12. #include <asm/uaccess.h>
  13. #include <asm/futex.h>
  14. #include "uaccess.h"
  15. #ifndef CONFIG_64BIT
  16. #define AHI "ahi"
  17. #define SLR "slr"
  18. #else
  19. #define AHI "aghi"
  20. #define SLR "slgr"
  21. #endif
  22. static size_t strnlen_kernel(size_t count, const char __user *src)
  23. {
  24. register unsigned long reg0 asm("0") = 0UL;
  25. unsigned long tmp1, tmp2;
  26. asm volatile(
  27. " la %2,0(%1)\n"
  28. " la %3,0(%0,%1)\n"
  29. " "SLR" %0,%0\n"
  30. "0: srst %3,%2\n"
  31. " jo 0b\n"
  32. " la %0,1(%3)\n" /* strnlen_kernel results includes \0 */
  33. " "SLR" %0,%1\n"
  34. "1:\n"
  35. EX_TABLE(0b,1b)
  36. : "+a" (count), "+a" (src), "=a" (tmp1), "=a" (tmp2)
  37. : "d" (reg0) : "cc", "memory");
  38. return count;
  39. }
  40. static size_t copy_in_kernel(size_t count, void __user *to,
  41. const void __user *from)
  42. {
  43. unsigned long tmp1;
  44. asm volatile(
  45. " "AHI" %0,-1\n"
  46. " jo 5f\n"
  47. " bras %3,3f\n"
  48. "0:"AHI" %0,257\n"
  49. "1: mvc 0(1,%1),0(%2)\n"
  50. " la %1,1(%1)\n"
  51. " la %2,1(%2)\n"
  52. " "AHI" %0,-1\n"
  53. " jnz 1b\n"
  54. " j 5f\n"
  55. "2: mvc 0(256,%1),0(%2)\n"
  56. " la %1,256(%1)\n"
  57. " la %2,256(%2)\n"
  58. "3:"AHI" %0,-256\n"
  59. " jnm 2b\n"
  60. "4: ex %0,1b-0b(%3)\n"
  61. "5:"SLR" %0,%0\n"
  62. "6:\n"
  63. EX_TABLE(1b,6b) EX_TABLE(2b,0b) EX_TABLE(4b,0b)
  64. : "+a" (count), "+a" (to), "+a" (from), "=a" (tmp1)
  65. : : "cc", "memory");
  66. return count;
  67. }
  68. /*
  69. * Returns kernel address for user virtual address. If the returned address is
  70. * >= -4095 (IS_ERR_VALUE(x) returns true), a fault has occured and the address
  71. * contains the (negative) exception code.
  72. */
  73. #ifdef CONFIG_64BIT
  74. static unsigned long follow_table(struct mm_struct *mm,
  75. unsigned long address, int write)
  76. {
  77. unsigned long *table = (unsigned long *)__pa(mm->pgd);
  78. if (unlikely(address > mm->context.asce_limit - 1))
  79. return -0x38UL;
  80. switch (mm->context.asce_bits & _ASCE_TYPE_MASK) {
  81. case _ASCE_TYPE_REGION1:
  82. table = table + ((address >> 53) & 0x7ff);
  83. if (unlikely(*table & _REGION_ENTRY_INVALID))
  84. return -0x39UL;
  85. table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
  86. /* fallthrough */
  87. case _ASCE_TYPE_REGION2:
  88. table = table + ((address >> 42) & 0x7ff);
  89. if (unlikely(*table & _REGION_ENTRY_INVALID))
  90. return -0x3aUL;
  91. table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
  92. /* fallthrough */
  93. case _ASCE_TYPE_REGION3:
  94. table = table + ((address >> 31) & 0x7ff);
  95. if (unlikely(*table & _REGION_ENTRY_INVALID))
  96. return -0x3bUL;
  97. table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
  98. /* fallthrough */
  99. case _ASCE_TYPE_SEGMENT:
  100. table = table + ((address >> 20) & 0x7ff);
  101. if (unlikely(*table & _SEGMENT_ENTRY_INVALID))
  102. return -0x10UL;
  103. if (unlikely(*table & _SEGMENT_ENTRY_LARGE)) {
  104. if (write && (*table & _SEGMENT_ENTRY_PROTECT))
  105. return -0x04UL;
  106. return (*table & _SEGMENT_ENTRY_ORIGIN_LARGE) +
  107. (address & ~_SEGMENT_ENTRY_ORIGIN_LARGE);
  108. }
  109. table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
  110. }
  111. table = table + ((address >> 12) & 0xff);
  112. if (unlikely(*table & _PAGE_INVALID))
  113. return -0x11UL;
  114. if (write && (*table & _PAGE_PROTECT))
  115. return -0x04UL;
  116. return (*table & PAGE_MASK) + (address & ~PAGE_MASK);
  117. }
  118. #else /* CONFIG_64BIT */
  119. static unsigned long follow_table(struct mm_struct *mm,
  120. unsigned long address, int write)
  121. {
  122. unsigned long *table = (unsigned long *)__pa(mm->pgd);
  123. table = table + ((address >> 20) & 0x7ff);
  124. if (unlikely(*table & _SEGMENT_ENTRY_INVALID))
  125. return -0x10UL;
  126. table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
  127. table = table + ((address >> 12) & 0xff);
  128. if (unlikely(*table & _PAGE_INVALID))
  129. return -0x11UL;
  130. if (write && (*table & _PAGE_PROTECT))
  131. return -0x04UL;
  132. return (*table & PAGE_MASK) + (address & ~PAGE_MASK);
  133. }
  134. #endif /* CONFIG_64BIT */
  135. static __always_inline size_t __user_copy_pt(unsigned long uaddr, void *kptr,
  136. size_t n, int write_user)
  137. {
  138. struct mm_struct *mm = current->mm;
  139. unsigned long offset, done, size, kaddr;
  140. void *from, *to;
  141. done = 0;
  142. retry:
  143. spin_lock(&mm->page_table_lock);
  144. do {
  145. kaddr = follow_table(mm, uaddr, write_user);
  146. if (IS_ERR_VALUE(kaddr))
  147. goto fault;
  148. offset = uaddr & ~PAGE_MASK;
  149. size = min(n - done, PAGE_SIZE - offset);
  150. if (write_user) {
  151. to = (void *) kaddr;
  152. from = kptr + done;
  153. } else {
  154. from = (void *) kaddr;
  155. to = kptr + done;
  156. }
  157. memcpy(to, from, size);
  158. done += size;
  159. uaddr += size;
  160. } while (done < n);
  161. spin_unlock(&mm->page_table_lock);
  162. return n - done;
  163. fault:
  164. spin_unlock(&mm->page_table_lock);
  165. if (__handle_fault(uaddr, -kaddr, write_user))
  166. return n - done;
  167. goto retry;
  168. }
  169. /*
  170. * Do DAT for user address by page table walk, return kernel address.
  171. * This function needs to be called with current->mm->page_table_lock held.
  172. */
  173. static __always_inline unsigned long __dat_user_addr(unsigned long uaddr,
  174. int write)
  175. {
  176. struct mm_struct *mm = current->mm;
  177. unsigned long kaddr;
  178. int rc;
  179. retry:
  180. kaddr = follow_table(mm, uaddr, write);
  181. if (IS_ERR_VALUE(kaddr))
  182. goto fault;
  183. return kaddr;
  184. fault:
  185. spin_unlock(&mm->page_table_lock);
  186. rc = __handle_fault(uaddr, -kaddr, write);
  187. spin_lock(&mm->page_table_lock);
  188. if (!rc)
  189. goto retry;
  190. return 0;
  191. }
  192. size_t copy_from_user_pt(size_t n, const void __user *from, void *to)
  193. {
  194. size_t rc;
  195. if (segment_eq(get_fs(), KERNEL_DS))
  196. return copy_in_kernel(n, (void __user *) to, from);
  197. rc = __user_copy_pt((unsigned long) from, to, n, 0);
  198. if (unlikely(rc))
  199. memset(to + n - rc, 0, rc);
  200. return rc;
  201. }
  202. size_t copy_to_user_pt(size_t n, void __user *to, const void *from)
  203. {
  204. if (segment_eq(get_fs(), KERNEL_DS))
  205. return copy_in_kernel(n, to, (void __user *) from);
  206. return __user_copy_pt((unsigned long) to, (void *) from, n, 1);
  207. }
  208. static size_t clear_user_pt(size_t n, void __user *to)
  209. {
  210. void *zpage = (void *) empty_zero_page;
  211. long done, size, ret;
  212. done = 0;
  213. do {
  214. if (n - done > PAGE_SIZE)
  215. size = PAGE_SIZE;
  216. else
  217. size = n - done;
  218. if (segment_eq(get_fs(), KERNEL_DS))
  219. ret = copy_in_kernel(n, to, (void __user *) zpage);
  220. else
  221. ret = __user_copy_pt((unsigned long) to, zpage, size, 1);
  222. done += size;
  223. to += size;
  224. if (ret)
  225. return ret + n - done;
  226. } while (done < n);
  227. return 0;
  228. }
  229. static size_t strnlen_user_pt(size_t count, const char __user *src)
  230. {
  231. unsigned long uaddr = (unsigned long) src;
  232. struct mm_struct *mm = current->mm;
  233. unsigned long offset, done, len, kaddr;
  234. size_t len_str;
  235. if (unlikely(!count))
  236. return 0;
  237. if (segment_eq(get_fs(), KERNEL_DS))
  238. return strnlen_kernel(count, src);
  239. done = 0;
  240. retry:
  241. spin_lock(&mm->page_table_lock);
  242. do {
  243. kaddr = follow_table(mm, uaddr, 0);
  244. if (IS_ERR_VALUE(kaddr))
  245. goto fault;
  246. offset = uaddr & ~PAGE_MASK;
  247. len = min(count - done, PAGE_SIZE - offset);
  248. len_str = strnlen((char *) kaddr, len);
  249. done += len_str;
  250. uaddr += len_str;
  251. } while ((len_str == len) && (done < count));
  252. spin_unlock(&mm->page_table_lock);
  253. return done + 1;
  254. fault:
  255. spin_unlock(&mm->page_table_lock);
  256. if (__handle_fault(uaddr, -kaddr, 0))
  257. return 0;
  258. goto retry;
  259. }
  260. static size_t strncpy_from_user_pt(size_t count, const char __user *src,
  261. char *dst)
  262. {
  263. size_t done, len, offset, len_str;
  264. if (unlikely(!count))
  265. return 0;
  266. done = 0;
  267. do {
  268. offset = (size_t)src & ~PAGE_MASK;
  269. len = min(count - done, PAGE_SIZE - offset);
  270. if (segment_eq(get_fs(), KERNEL_DS)) {
  271. if (copy_in_kernel(len, (void __user *) dst, src))
  272. return -EFAULT;
  273. } else {
  274. if (__user_copy_pt((unsigned long) src, dst, len, 0))
  275. return -EFAULT;
  276. }
  277. len_str = strnlen(dst, len);
  278. done += len_str;
  279. src += len_str;
  280. dst += len_str;
  281. } while ((len_str == len) && (done < count));
  282. return done;
  283. }
  284. static size_t copy_in_user_pt(size_t n, void __user *to,
  285. const void __user *from)
  286. {
  287. struct mm_struct *mm = current->mm;
  288. unsigned long offset_max, uaddr, done, size, error_code;
  289. unsigned long uaddr_from = (unsigned long) from;
  290. unsigned long uaddr_to = (unsigned long) to;
  291. unsigned long kaddr_to, kaddr_from;
  292. int write_user;
  293. if (segment_eq(get_fs(), KERNEL_DS))
  294. return copy_in_kernel(n, to, from);
  295. done = 0;
  296. retry:
  297. spin_lock(&mm->page_table_lock);
  298. do {
  299. write_user = 0;
  300. uaddr = uaddr_from;
  301. kaddr_from = follow_table(mm, uaddr_from, 0);
  302. error_code = kaddr_from;
  303. if (IS_ERR_VALUE(error_code))
  304. goto fault;
  305. write_user = 1;
  306. uaddr = uaddr_to;
  307. kaddr_to = follow_table(mm, uaddr_to, 1);
  308. error_code = (unsigned long) kaddr_to;
  309. if (IS_ERR_VALUE(error_code))
  310. goto fault;
  311. offset_max = max(uaddr_from & ~PAGE_MASK,
  312. uaddr_to & ~PAGE_MASK);
  313. size = min(n - done, PAGE_SIZE - offset_max);
  314. memcpy((void *) kaddr_to, (void *) kaddr_from, size);
  315. done += size;
  316. uaddr_from += size;
  317. uaddr_to += size;
  318. } while (done < n);
  319. spin_unlock(&mm->page_table_lock);
  320. return n - done;
  321. fault:
  322. spin_unlock(&mm->page_table_lock);
  323. if (__handle_fault(uaddr, -error_code, write_user))
  324. return n - done;
  325. goto retry;
  326. }
  327. #define __futex_atomic_op(insn, ret, oldval, newval, uaddr, oparg) \
  328. asm volatile("0: l %1,0(%6)\n" \
  329. "1: " insn \
  330. "2: cs %1,%2,0(%6)\n" \
  331. "3: jl 1b\n" \
  332. " lhi %0,0\n" \
  333. "4:\n" \
  334. EX_TABLE(0b,4b) EX_TABLE(2b,4b) EX_TABLE(3b,4b) \
  335. : "=d" (ret), "=&d" (oldval), "=&d" (newval), \
  336. "=m" (*uaddr) \
  337. : "0" (-EFAULT), "d" (oparg), "a" (uaddr), \
  338. "m" (*uaddr) : "cc" );
  339. static int __futex_atomic_op_pt(int op, u32 __user *uaddr, int oparg, int *old)
  340. {
  341. int oldval = 0, newval, ret;
  342. switch (op) {
  343. case FUTEX_OP_SET:
  344. __futex_atomic_op("lr %2,%5\n",
  345. ret, oldval, newval, uaddr, oparg);
  346. break;
  347. case FUTEX_OP_ADD:
  348. __futex_atomic_op("lr %2,%1\nar %2,%5\n",
  349. ret, oldval, newval, uaddr, oparg);
  350. break;
  351. case FUTEX_OP_OR:
  352. __futex_atomic_op("lr %2,%1\nor %2,%5\n",
  353. ret, oldval, newval, uaddr, oparg);
  354. break;
  355. case FUTEX_OP_ANDN:
  356. __futex_atomic_op("lr %2,%1\nnr %2,%5\n",
  357. ret, oldval, newval, uaddr, oparg);
  358. break;
  359. case FUTEX_OP_XOR:
  360. __futex_atomic_op("lr %2,%1\nxr %2,%5\n",
  361. ret, oldval, newval, uaddr, oparg);
  362. break;
  363. default:
  364. ret = -ENOSYS;
  365. }
  366. if (ret == 0)
  367. *old = oldval;
  368. return ret;
  369. }
  370. int futex_atomic_op_pt(int op, u32 __user *uaddr, int oparg, int *old)
  371. {
  372. int ret;
  373. if (segment_eq(get_fs(), KERNEL_DS))
  374. return __futex_atomic_op_pt(op, uaddr, oparg, old);
  375. spin_lock(&current->mm->page_table_lock);
  376. uaddr = (u32 __force __user *)
  377. __dat_user_addr((__force unsigned long) uaddr, 1);
  378. if (!uaddr) {
  379. spin_unlock(&current->mm->page_table_lock);
  380. return -EFAULT;
  381. }
  382. get_page(virt_to_page(uaddr));
  383. spin_unlock(&current->mm->page_table_lock);
  384. ret = __futex_atomic_op_pt(op, uaddr, oparg, old);
  385. put_page(virt_to_page(uaddr));
  386. return ret;
  387. }
  388. static int __futex_atomic_cmpxchg_pt(u32 *uval, u32 __user *uaddr,
  389. u32 oldval, u32 newval)
  390. {
  391. int ret;
  392. asm volatile("0: cs %1,%4,0(%5)\n"
  393. "1: la %0,0\n"
  394. "2:\n"
  395. EX_TABLE(0b,2b) EX_TABLE(1b,2b)
  396. : "=d" (ret), "+d" (oldval), "=m" (*uaddr)
  397. : "0" (-EFAULT), "d" (newval), "a" (uaddr), "m" (*uaddr)
  398. : "cc", "memory" );
  399. *uval = oldval;
  400. return ret;
  401. }
  402. int futex_atomic_cmpxchg_pt(u32 *uval, u32 __user *uaddr,
  403. u32 oldval, u32 newval)
  404. {
  405. int ret;
  406. if (segment_eq(get_fs(), KERNEL_DS))
  407. return __futex_atomic_cmpxchg_pt(uval, uaddr, oldval, newval);
  408. spin_lock(&current->mm->page_table_lock);
  409. uaddr = (u32 __force __user *)
  410. __dat_user_addr((__force unsigned long) uaddr, 1);
  411. if (!uaddr) {
  412. spin_unlock(&current->mm->page_table_lock);
  413. return -EFAULT;
  414. }
  415. get_page(virt_to_page(uaddr));
  416. spin_unlock(&current->mm->page_table_lock);
  417. ret = __futex_atomic_cmpxchg_pt(uval, uaddr, oldval, newval);
  418. put_page(virt_to_page(uaddr));
  419. return ret;
  420. }
  421. struct uaccess_ops uaccess_pt = {
  422. .copy_from_user = copy_from_user_pt,
  423. .copy_to_user = copy_to_user_pt,
  424. .copy_in_user = copy_in_user_pt,
  425. .clear_user = clear_user_pt,
  426. .strnlen_user = strnlen_user_pt,
  427. .strncpy_from_user = strncpy_from_user_pt,
  428. .futex_atomic_op = futex_atomic_op_pt,
  429. .futex_atomic_cmpxchg = futex_atomic_cmpxchg_pt,
  430. };