vmalloc.c 20 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863
  1. /*
  2. * linux/mm/vmalloc.c
  3. *
  4. * Copyright (C) 1993 Linus Torvalds
  5. * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
  6. * SMP-safe vmalloc/vfree/ioremap, Tigran Aivazian <tigran@veritas.com>, May 2000
  7. * Major rework to support vmap/vunmap, Christoph Hellwig, SGI, August 2002
  8. * Numa awareness, Christoph Lameter, SGI, June 2005
  9. */
  10. #include <linux/mm.h>
  11. #include <linux/module.h>
  12. #include <linux/highmem.h>
  13. #include <linux/slab.h>
  14. #include <linux/spinlock.h>
  15. #include <linux/interrupt.h>
  16. #include <linux/vmalloc.h>
  17. #include <asm/uaccess.h>
  18. #include <asm/tlbflush.h>
  19. DEFINE_RWLOCK(vmlist_lock);
  20. struct vm_struct *vmlist;
  21. static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
  22. int node);
  23. static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
  24. {
  25. pte_t *pte;
  26. pte = pte_offset_kernel(pmd, addr);
  27. do {
  28. pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
  29. WARN_ON(!pte_none(ptent) && !pte_present(ptent));
  30. } while (pte++, addr += PAGE_SIZE, addr != end);
  31. }
  32. static inline void vunmap_pmd_range(pud_t *pud, unsigned long addr,
  33. unsigned long end)
  34. {
  35. pmd_t *pmd;
  36. unsigned long next;
  37. pmd = pmd_offset(pud, addr);
  38. do {
  39. next = pmd_addr_end(addr, end);
  40. if (pmd_none_or_clear_bad(pmd))
  41. continue;
  42. vunmap_pte_range(pmd, addr, next);
  43. } while (pmd++, addr = next, addr != end);
  44. }
  45. static inline void vunmap_pud_range(pgd_t *pgd, unsigned long addr,
  46. unsigned long end)
  47. {
  48. pud_t *pud;
  49. unsigned long next;
  50. pud = pud_offset(pgd, addr);
  51. do {
  52. next = pud_addr_end(addr, end);
  53. if (pud_none_or_clear_bad(pud))
  54. continue;
  55. vunmap_pmd_range(pud, addr, next);
  56. } while (pud++, addr = next, addr != end);
  57. }
  58. void unmap_kernel_range(unsigned long addr, unsigned long size)
  59. {
  60. pgd_t *pgd;
  61. unsigned long next;
  62. unsigned long start = addr;
  63. unsigned long end = addr + size;
  64. BUG_ON(addr >= end);
  65. pgd = pgd_offset_k(addr);
  66. flush_cache_vunmap(addr, end);
  67. do {
  68. next = pgd_addr_end(addr, end);
  69. if (pgd_none_or_clear_bad(pgd))
  70. continue;
  71. vunmap_pud_range(pgd, addr, next);
  72. } while (pgd++, addr = next, addr != end);
  73. flush_tlb_kernel_range(start, end);
  74. }
  75. static void unmap_vm_area(struct vm_struct *area)
  76. {
  77. unmap_kernel_range((unsigned long)area->addr, area->size);
  78. }
  79. static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
  80. unsigned long end, pgprot_t prot, struct page ***pages)
  81. {
  82. pte_t *pte;
  83. pte = pte_alloc_kernel(pmd, addr);
  84. if (!pte)
  85. return -ENOMEM;
  86. do {
  87. struct page *page = **pages;
  88. WARN_ON(!pte_none(*pte));
  89. if (!page)
  90. return -ENOMEM;
  91. set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
  92. (*pages)++;
  93. } while (pte++, addr += PAGE_SIZE, addr != end);
  94. return 0;
  95. }
  96. static inline int vmap_pmd_range(pud_t *pud, unsigned long addr,
  97. unsigned long end, pgprot_t prot, struct page ***pages)
  98. {
  99. pmd_t *pmd;
  100. unsigned long next;
  101. pmd = pmd_alloc(&init_mm, pud, addr);
  102. if (!pmd)
  103. return -ENOMEM;
  104. do {
  105. next = pmd_addr_end(addr, end);
  106. if (vmap_pte_range(pmd, addr, next, prot, pages))
  107. return -ENOMEM;
  108. } while (pmd++, addr = next, addr != end);
  109. return 0;
  110. }
  111. static inline int vmap_pud_range(pgd_t *pgd, unsigned long addr,
  112. unsigned long end, pgprot_t prot, struct page ***pages)
  113. {
  114. pud_t *pud;
  115. unsigned long next;
  116. pud = pud_alloc(&init_mm, pgd, addr);
  117. if (!pud)
  118. return -ENOMEM;
  119. do {
  120. next = pud_addr_end(addr, end);
  121. if (vmap_pmd_range(pud, addr, next, prot, pages))
  122. return -ENOMEM;
  123. } while (pud++, addr = next, addr != end);
  124. return 0;
  125. }
  126. int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
  127. {
  128. pgd_t *pgd;
  129. unsigned long next;
  130. unsigned long addr = (unsigned long) area->addr;
  131. unsigned long end = addr + area->size - PAGE_SIZE;
  132. int err;
  133. BUG_ON(addr >= end);
  134. pgd = pgd_offset_k(addr);
  135. do {
  136. next = pgd_addr_end(addr, end);
  137. err = vmap_pud_range(pgd, addr, next, prot, pages);
  138. if (err)
  139. break;
  140. } while (pgd++, addr = next, addr != end);
  141. flush_cache_vmap((unsigned long) area->addr, end);
  142. return err;
  143. }
  144. EXPORT_SYMBOL_GPL(map_vm_area);
  145. /*
  146. * Map a vmalloc()-space virtual address to the physical page.
  147. */
  148. struct page *vmalloc_to_page(void *vmalloc_addr)
  149. {
  150. unsigned long addr = (unsigned long) vmalloc_addr;
  151. struct page *page = NULL;
  152. pgd_t *pgd = pgd_offset_k(addr);
  153. pud_t *pud;
  154. pmd_t *pmd;
  155. pte_t *ptep, pte;
  156. if (!pgd_none(*pgd)) {
  157. pud = pud_offset(pgd, addr);
  158. if (!pud_none(*pud)) {
  159. pmd = pmd_offset(pud, addr);
  160. if (!pmd_none(*pmd)) {
  161. ptep = pte_offset_map(pmd, addr);
  162. pte = *ptep;
  163. if (pte_present(pte))
  164. page = pte_page(pte);
  165. pte_unmap(ptep);
  166. }
  167. }
  168. }
  169. return page;
  170. }
  171. EXPORT_SYMBOL(vmalloc_to_page);
  172. /*
  173. * Map a vmalloc()-space virtual address to the physical page frame number.
  174. */
  175. unsigned long vmalloc_to_pfn(void *vmalloc_addr)
  176. {
  177. return page_to_pfn(vmalloc_to_page(vmalloc_addr));
  178. }
  179. EXPORT_SYMBOL(vmalloc_to_pfn);
  180. static struct vm_struct *__get_vm_area_node(unsigned long size, unsigned long flags,
  181. unsigned long start, unsigned long end,
  182. int node, gfp_t gfp_mask)
  183. {
  184. struct vm_struct **p, *tmp, *area;
  185. unsigned long align = 1;
  186. unsigned long addr;
  187. BUG_ON(in_interrupt());
  188. if (flags & VM_IOREMAP) {
  189. int bit = fls(size);
  190. if (bit > IOREMAP_MAX_ORDER)
  191. bit = IOREMAP_MAX_ORDER;
  192. else if (bit < PAGE_SHIFT)
  193. bit = PAGE_SHIFT;
  194. align = 1ul << bit;
  195. }
  196. addr = ALIGN(start, align);
  197. size = PAGE_ALIGN(size);
  198. if (unlikely(!size))
  199. return NULL;
  200. area = kmalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
  201. if (unlikely(!area))
  202. return NULL;
  203. /*
  204. * We always allocate a guard page.
  205. */
  206. size += PAGE_SIZE;
  207. write_lock(&vmlist_lock);
  208. for (p = &vmlist; (tmp = *p) != NULL ;p = &tmp->next) {
  209. if ((unsigned long)tmp->addr < addr) {
  210. if((unsigned long)tmp->addr + tmp->size >= addr)
  211. addr = ALIGN(tmp->size +
  212. (unsigned long)tmp->addr, align);
  213. continue;
  214. }
  215. if ((size + addr) < addr)
  216. goto out;
  217. if (size + addr <= (unsigned long)tmp->addr)
  218. goto found;
  219. addr = ALIGN(tmp->size + (unsigned long)tmp->addr, align);
  220. if (addr > end - size)
  221. goto out;
  222. }
  223. found:
  224. area->next = *p;
  225. *p = area;
  226. area->flags = flags;
  227. area->addr = (void *)addr;
  228. area->size = size;
  229. area->pages = NULL;
  230. area->nr_pages = 0;
  231. area->phys_addr = 0;
  232. write_unlock(&vmlist_lock);
  233. return area;
  234. out:
  235. write_unlock(&vmlist_lock);
  236. kfree(area);
  237. if (printk_ratelimit())
  238. printk(KERN_WARNING "allocation failed: out of vmalloc space - use vmalloc=<size> to increase size.\n");
  239. return NULL;
  240. }
  241. struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
  242. unsigned long start, unsigned long end)
  243. {
  244. return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL);
  245. }
  246. EXPORT_SYMBOL_GPL(__get_vm_area);
  247. /**
  248. * get_vm_area - reserve a contiguous kernel virtual area
  249. * @size: size of the area
  250. * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
  251. *
  252. * Search an area of @size in the kernel virtual mapping area,
  253. * and reserved it for out purposes. Returns the area descriptor
  254. * on success or %NULL on failure.
  255. */
  256. struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
  257. {
  258. return __get_vm_area(size, flags, VMALLOC_START, VMALLOC_END);
  259. }
  260. struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags,
  261. int node, gfp_t gfp_mask)
  262. {
  263. return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node,
  264. gfp_mask);
  265. }
  266. /* Caller must hold vmlist_lock */
  267. static struct vm_struct *__find_vm_area(void *addr)
  268. {
  269. struct vm_struct *tmp;
  270. for (tmp = vmlist; tmp != NULL; tmp = tmp->next) {
  271. if (tmp->addr == addr)
  272. break;
  273. }
  274. return tmp;
  275. }
  276. /* Caller must hold vmlist_lock */
  277. static struct vm_struct *__remove_vm_area(void *addr)
  278. {
  279. struct vm_struct **p, *tmp;
  280. for (p = &vmlist ; (tmp = *p) != NULL ;p = &tmp->next) {
  281. if (tmp->addr == addr)
  282. goto found;
  283. }
  284. return NULL;
  285. found:
  286. unmap_vm_area(tmp);
  287. *p = tmp->next;
  288. /*
  289. * Remove the guard page.
  290. */
  291. tmp->size -= PAGE_SIZE;
  292. return tmp;
  293. }
  294. /**
  295. * remove_vm_area - find and remove a continuous kernel virtual area
  296. * @addr: base address
  297. *
  298. * Search for the kernel VM area starting at @addr, and remove it.
  299. * This function returns the found VM area, but using it is NOT safe
  300. * on SMP machines, except for its size or flags.
  301. */
  302. struct vm_struct *remove_vm_area(void *addr)
  303. {
  304. struct vm_struct *v;
  305. write_lock(&vmlist_lock);
  306. v = __remove_vm_area(addr);
  307. write_unlock(&vmlist_lock);
  308. return v;
  309. }
  310. static void __vunmap(void *addr, int deallocate_pages)
  311. {
  312. struct vm_struct *area;
  313. if (!addr)
  314. return;
  315. if ((PAGE_SIZE-1) & (unsigned long)addr) {
  316. printk(KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
  317. WARN_ON(1);
  318. return;
  319. }
  320. area = remove_vm_area(addr);
  321. if (unlikely(!area)) {
  322. printk(KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
  323. addr);
  324. WARN_ON(1);
  325. return;
  326. }
  327. debug_check_no_locks_freed(addr, area->size);
  328. if (deallocate_pages) {
  329. int i;
  330. for (i = 0; i < area->nr_pages; i++) {
  331. BUG_ON(!area->pages[i]);
  332. __free_page(area->pages[i]);
  333. }
  334. if (area->flags & VM_VPAGES)
  335. vfree(area->pages);
  336. else
  337. kfree(area->pages);
  338. }
  339. kfree(area);
  340. return;
  341. }
  342. /**
  343. * vfree - release memory allocated by vmalloc()
  344. * @addr: memory base address
  345. *
  346. * Free the virtually continuous memory area starting at @addr, as
  347. * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
  348. * NULL, no operation is performed.
  349. *
  350. * Must not be called in interrupt context.
  351. */
  352. void vfree(void *addr)
  353. {
  354. BUG_ON(in_interrupt());
  355. __vunmap(addr, 1);
  356. }
  357. EXPORT_SYMBOL(vfree);
  358. /**
  359. * vunmap - release virtual mapping obtained by vmap()
  360. * @addr: memory base address
  361. *
  362. * Free the virtually contiguous memory area starting at @addr,
  363. * which was created from the page array passed to vmap().
  364. *
  365. * Must not be called in interrupt context.
  366. */
  367. void vunmap(void *addr)
  368. {
  369. BUG_ON(in_interrupt());
  370. __vunmap(addr, 0);
  371. }
  372. EXPORT_SYMBOL(vunmap);
  373. /**
  374. * vmap - map an array of pages into virtually contiguous space
  375. * @pages: array of page pointers
  376. * @count: number of pages to map
  377. * @flags: vm_area->flags
  378. * @prot: page protection for the mapping
  379. *
  380. * Maps @count pages from @pages into contiguous kernel virtual
  381. * space.
  382. */
  383. void *vmap(struct page **pages, unsigned int count,
  384. unsigned long flags, pgprot_t prot)
  385. {
  386. struct vm_struct *area;
  387. if (count > num_physpages)
  388. return NULL;
  389. area = get_vm_area((count << PAGE_SHIFT), flags);
  390. if (!area)
  391. return NULL;
  392. if (map_vm_area(area, prot, &pages)) {
  393. vunmap(area->addr);
  394. return NULL;
  395. }
  396. return area->addr;
  397. }
  398. EXPORT_SYMBOL(vmap);
  399. void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
  400. pgprot_t prot, int node)
  401. {
  402. struct page **pages;
  403. unsigned int nr_pages, array_size, i;
  404. nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
  405. array_size = (nr_pages * sizeof(struct page *));
  406. area->nr_pages = nr_pages;
  407. /* Please note that the recursion is strictly bounded. */
  408. if (array_size > PAGE_SIZE) {
  409. pages = __vmalloc_node(array_size, gfp_mask | __GFP_ZERO,
  410. PAGE_KERNEL, node);
  411. area->flags |= VM_VPAGES;
  412. } else {
  413. pages = kmalloc_node(array_size,
  414. (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO,
  415. node);
  416. }
  417. area->pages = pages;
  418. if (!area->pages) {
  419. remove_vm_area(area->addr);
  420. kfree(area);
  421. return NULL;
  422. }
  423. for (i = 0; i < area->nr_pages; i++) {
  424. if (node < 0)
  425. area->pages[i] = alloc_page(gfp_mask);
  426. else
  427. area->pages[i] = alloc_pages_node(node, gfp_mask, 0);
  428. if (unlikely(!area->pages[i])) {
  429. /* Successfully allocated i pages, free them in __vunmap() */
  430. area->nr_pages = i;
  431. goto fail;
  432. }
  433. }
  434. if (map_vm_area(area, prot, &pages))
  435. goto fail;
  436. return area->addr;
  437. fail:
  438. vfree(area->addr);
  439. return NULL;
  440. }
  441. void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot)
  442. {
  443. return __vmalloc_area_node(area, gfp_mask, prot, -1);
  444. }
  445. /**
  446. * __vmalloc_node - allocate virtually contiguous memory
  447. * @size: allocation size
  448. * @gfp_mask: flags for the page level allocator
  449. * @prot: protection mask for the allocated pages
  450. * @node: node to use for allocation or -1
  451. *
  452. * Allocate enough pages to cover @size from the page level
  453. * allocator with @gfp_mask flags. Map them into contiguous
  454. * kernel virtual space, using a pagetable protection of @prot.
  455. */
  456. static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
  457. int node)
  458. {
  459. struct vm_struct *area;
  460. size = PAGE_ALIGN(size);
  461. if (!size || (size >> PAGE_SHIFT) > num_physpages)
  462. return NULL;
  463. area = get_vm_area_node(size, VM_ALLOC, node, gfp_mask);
  464. if (!area)
  465. return NULL;
  466. return __vmalloc_area_node(area, gfp_mask, prot, node);
  467. }
  468. void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
  469. {
  470. return __vmalloc_node(size, gfp_mask, prot, -1);
  471. }
  472. EXPORT_SYMBOL(__vmalloc);
  473. /**
  474. * vmalloc - allocate virtually contiguous memory
  475. * @size: allocation size
  476. * Allocate enough pages to cover @size from the page level
  477. * allocator and map them into contiguous kernel virtual space.
  478. *
  479. * For tight control over page level allocator and protection flags
  480. * use __vmalloc() instead.
  481. */
  482. void *vmalloc(unsigned long size)
  483. {
  484. return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
  485. }
  486. EXPORT_SYMBOL(vmalloc);
  487. /**
  488. * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
  489. * @size: allocation size
  490. *
  491. * The resulting memory area is zeroed so it can be mapped to userspace
  492. * without leaking data.
  493. */
  494. void *vmalloc_user(unsigned long size)
  495. {
  496. struct vm_struct *area;
  497. void *ret;
  498. ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL);
  499. if (ret) {
  500. write_lock(&vmlist_lock);
  501. area = __find_vm_area(ret);
  502. area->flags |= VM_USERMAP;
  503. write_unlock(&vmlist_lock);
  504. }
  505. return ret;
  506. }
  507. EXPORT_SYMBOL(vmalloc_user);
  508. /**
  509. * vmalloc_node - allocate memory on a specific node
  510. * @size: allocation size
  511. * @node: numa node
  512. *
  513. * Allocate enough pages to cover @size from the page level
  514. * allocator and map them into contiguous kernel virtual space.
  515. *
  516. * For tight control over page level allocator and protection flags
  517. * use __vmalloc() instead.
  518. */
  519. void *vmalloc_node(unsigned long size, int node)
  520. {
  521. return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, node);
  522. }
  523. EXPORT_SYMBOL(vmalloc_node);
  524. #ifndef PAGE_KERNEL_EXEC
  525. # define PAGE_KERNEL_EXEC PAGE_KERNEL
  526. #endif
  527. /**
  528. * vmalloc_exec - allocate virtually contiguous, executable memory
  529. * @size: allocation size
  530. *
  531. * Kernel-internal function to allocate enough pages to cover @size
  532. * the page level allocator and map them into contiguous and
  533. * executable kernel virtual space.
  534. *
  535. * For tight control over page level allocator and protection flags
  536. * use __vmalloc() instead.
  537. */
  538. void *vmalloc_exec(unsigned long size)
  539. {
  540. return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
  541. }
  542. #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
  543. #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
  544. #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
  545. #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
  546. #else
  547. #define GFP_VMALLOC32 GFP_KERNEL
  548. #endif
  549. /**
  550. * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
  551. * @size: allocation size
  552. *
  553. * Allocate enough 32bit PA addressable pages to cover @size from the
  554. * page level allocator and map them into contiguous kernel virtual space.
  555. */
  556. void *vmalloc_32(unsigned long size)
  557. {
  558. return __vmalloc(size, GFP_VMALLOC32, PAGE_KERNEL);
  559. }
  560. EXPORT_SYMBOL(vmalloc_32);
  561. /**
  562. * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
  563. * @size: allocation size
  564. *
  565. * The resulting memory area is 32bit addressable and zeroed so it can be
  566. * mapped to userspace without leaking data.
  567. */
  568. void *vmalloc_32_user(unsigned long size)
  569. {
  570. struct vm_struct *area;
  571. void *ret;
  572. ret = __vmalloc(size, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL);
  573. if (ret) {
  574. write_lock(&vmlist_lock);
  575. area = __find_vm_area(ret);
  576. area->flags |= VM_USERMAP;
  577. write_unlock(&vmlist_lock);
  578. }
  579. return ret;
  580. }
  581. EXPORT_SYMBOL(vmalloc_32_user);
  582. long vread(char *buf, char *addr, unsigned long count)
  583. {
  584. struct vm_struct *tmp;
  585. char *vaddr, *buf_start = buf;
  586. unsigned long n;
  587. /* Don't allow overflow */
  588. if ((unsigned long) addr + count < count)
  589. count = -(unsigned long) addr;
  590. read_lock(&vmlist_lock);
  591. for (tmp = vmlist; tmp; tmp = tmp->next) {
  592. vaddr = (char *) tmp->addr;
  593. if (addr >= vaddr + tmp->size - PAGE_SIZE)
  594. continue;
  595. while (addr < vaddr) {
  596. if (count == 0)
  597. goto finished;
  598. *buf = '\0';
  599. buf++;
  600. addr++;
  601. count--;
  602. }
  603. n = vaddr + tmp->size - PAGE_SIZE - addr;
  604. do {
  605. if (count == 0)
  606. goto finished;
  607. *buf = *addr;
  608. buf++;
  609. addr++;
  610. count--;
  611. } while (--n > 0);
  612. }
  613. finished:
  614. read_unlock(&vmlist_lock);
  615. return buf - buf_start;
  616. }
  617. long vwrite(char *buf, char *addr, unsigned long count)
  618. {
  619. struct vm_struct *tmp;
  620. char *vaddr, *buf_start = buf;
  621. unsigned long n;
  622. /* Don't allow overflow */
  623. if ((unsigned long) addr + count < count)
  624. count = -(unsigned long) addr;
  625. read_lock(&vmlist_lock);
  626. for (tmp = vmlist; tmp; tmp = tmp->next) {
  627. vaddr = (char *) tmp->addr;
  628. if (addr >= vaddr + tmp->size - PAGE_SIZE)
  629. continue;
  630. while (addr < vaddr) {
  631. if (count == 0)
  632. goto finished;
  633. buf++;
  634. addr++;
  635. count--;
  636. }
  637. n = vaddr + tmp->size - PAGE_SIZE - addr;
  638. do {
  639. if (count == 0)
  640. goto finished;
  641. *addr = *buf;
  642. buf++;
  643. addr++;
  644. count--;
  645. } while (--n > 0);
  646. }
  647. finished:
  648. read_unlock(&vmlist_lock);
  649. return buf - buf_start;
  650. }
  651. /**
  652. * remap_vmalloc_range - map vmalloc pages to userspace
  653. * @vma: vma to cover (map full range of vma)
  654. * @addr: vmalloc memory
  655. * @pgoff: number of pages into addr before first page to map
  656. * @returns: 0 for success, -Exxx on failure
  657. *
  658. * This function checks that addr is a valid vmalloc'ed area, and
  659. * that it is big enough to cover the vma. Will return failure if
  660. * that criteria isn't met.
  661. *
  662. * Similar to remap_pfn_range() (see mm/memory.c)
  663. */
  664. int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
  665. unsigned long pgoff)
  666. {
  667. struct vm_struct *area;
  668. unsigned long uaddr = vma->vm_start;
  669. unsigned long usize = vma->vm_end - vma->vm_start;
  670. int ret;
  671. if ((PAGE_SIZE-1) & (unsigned long)addr)
  672. return -EINVAL;
  673. read_lock(&vmlist_lock);
  674. area = __find_vm_area(addr);
  675. if (!area)
  676. goto out_einval_locked;
  677. if (!(area->flags & VM_USERMAP))
  678. goto out_einval_locked;
  679. if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
  680. goto out_einval_locked;
  681. read_unlock(&vmlist_lock);
  682. addr += pgoff << PAGE_SHIFT;
  683. do {
  684. struct page *page = vmalloc_to_page(addr);
  685. ret = vm_insert_page(vma, uaddr, page);
  686. if (ret)
  687. return ret;
  688. uaddr += PAGE_SIZE;
  689. addr += PAGE_SIZE;
  690. usize -= PAGE_SIZE;
  691. } while (usize > 0);
  692. /* Prevent "things" like memory migration? VM_flags need a cleanup... */
  693. vma->vm_flags |= VM_RESERVED;
  694. return ret;
  695. out_einval_locked:
  696. read_unlock(&vmlist_lock);
  697. return -EINVAL;
  698. }
  699. EXPORT_SYMBOL(remap_vmalloc_range);
  700. /*
  701. * Implement a stub for vmalloc_sync_all() if the architecture chose not to
  702. * have one.
  703. */
  704. void __attribute__((weak)) vmalloc_sync_all(void)
  705. {
  706. }
  707. static int f(pte_t *pte, struct page *pmd_page, unsigned long addr, void *data)
  708. {
  709. /* apply_to_page_range() does all the hard work. */
  710. return 0;
  711. }
  712. /**
  713. * alloc_vm_area - allocate a range of kernel address space
  714. * @size: size of the area
  715. * @returns: NULL on failure, vm_struct on success
  716. *
  717. * This function reserves a range of kernel address space, and
  718. * allocates pagetables to map that range. No actual mappings
  719. * are created. If the kernel address space is not shared
  720. * between processes, it syncs the pagetable across all
  721. * processes.
  722. */
  723. struct vm_struct *alloc_vm_area(size_t size)
  724. {
  725. struct vm_struct *area;
  726. area = get_vm_area(size, VM_IOREMAP);
  727. if (area == NULL)
  728. return NULL;
  729. /*
  730. * This ensures that page tables are constructed for this region
  731. * of kernel virtual address space and mapped into init_mm.
  732. */
  733. if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
  734. area->size, f, NULL)) {
  735. free_vm_area(area);
  736. return NULL;
  737. }
  738. /* Make sure the pagetables are constructed in process kernel
  739. mappings */
  740. vmalloc_sync_all();
  741. return area;
  742. }
  743. EXPORT_SYMBOL_GPL(alloc_vm_area);
  744. void free_vm_area(struct vm_struct *area)
  745. {
  746. struct vm_struct *ret;
  747. ret = remove_vm_area(area->addr);
  748. BUG_ON(ret != area);
  749. kfree(area);
  750. }
  751. EXPORT_SYMBOL_GPL(free_vm_area);