Root/
1 | /* |
2 | * linux/mm/nommu.c |
3 | * |
4 | * Replacement code for mm functions to support CPU's that don't |
5 | * have any form of memory management unit (thus no virtual memory). |
6 | * |
7 | * See Documentation/nommu-mmap.txt |
8 | * |
9 | * Copyright (c) 2004-2008 David Howells <dhowells@redhat.com> |
10 | * Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com> |
11 | * Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org> |
12 | * Copyright (c) 2002 Greg Ungerer <gerg@snapgear.com> |
13 | * Copyright (c) 2007-2010 Paul Mundt <lethal@linux-sh.org> |
14 | */ |
15 | |
16 | #include <linux/export.h> |
17 | #include <linux/mm.h> |
18 | #include <linux/vmacache.h> |
19 | #include <linux/mman.h> |
20 | #include <linux/swap.h> |
21 | #include <linux/file.h> |
22 | #include <linux/highmem.h> |
23 | #include <linux/pagemap.h> |
24 | #include <linux/slab.h> |
25 | #include <linux/vmalloc.h> |
26 | #include <linux/blkdev.h> |
27 | #include <linux/backing-dev.h> |
28 | #include <linux/compiler.h> |
29 | #include <linux/mount.h> |
30 | #include <linux/personality.h> |
31 | #include <linux/security.h> |
32 | #include <linux/syscalls.h> |
33 | #include <linux/audit.h> |
34 | #include <linux/sched/sysctl.h> |
35 | |
36 | #include <asm/uaccess.h> |
37 | #include <asm/tlb.h> |
38 | #include <asm/tlbflush.h> |
39 | #include <asm/mmu_context.h> |
40 | #include "internal.h" |
41 | |
42 | #if 0 |
43 | #define kenter(FMT, ...) \ |
44 | printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__) |
45 | #define kleave(FMT, ...) \ |
46 | printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__) |
47 | #define kdebug(FMT, ...) \ |
48 | printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__) |
49 | #else |
50 | #define kenter(FMT, ...) \ |
51 | no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__) |
52 | #define kleave(FMT, ...) \ |
53 | no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__) |
54 | #define kdebug(FMT, ...) \ |
55 | no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__) |
56 | #endif |
57 | |
58 | void *high_memory; |
59 | struct page *mem_map; |
60 | unsigned long max_mapnr; |
61 | unsigned long highest_memmap_pfn; |
62 | struct percpu_counter vm_committed_as; |
63 | int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */ |
64 | int sysctl_overcommit_ratio = 50; /* default is 50% */ |
65 | unsigned long sysctl_overcommit_kbytes __read_mostly; |
66 | int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT; |
67 | int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS; |
68 | unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */ |
69 | unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */ |
70 | int heap_stack_gap = 0; |
71 | |
72 | atomic_long_t mmap_pages_allocated; |
73 | |
74 | /* |
75 | * The global memory commitment made in the system can be a metric |
76 | * that can be used to drive ballooning decisions when Linux is hosted |
77 | * as a guest. On Hyper-V, the host implements a policy engine for dynamically |
78 | * balancing memory across competing virtual machines that are hosted. |
79 | * Several metrics drive this policy engine including the guest reported |
80 | * memory commitment. |
81 | */ |
82 | unsigned long vm_memory_committed(void) |
83 | { |
84 | return percpu_counter_read_positive(&vm_committed_as); |
85 | } |
86 | |
87 | EXPORT_SYMBOL_GPL(vm_memory_committed); |
88 | |
89 | EXPORT_SYMBOL(mem_map); |
90 | |
91 | /* list of mapped, potentially shareable regions */ |
92 | static struct kmem_cache *vm_region_jar; |
93 | struct rb_root nommu_region_tree = RB_ROOT; |
94 | DECLARE_RWSEM(nommu_region_sem); |
95 | |
96 | const struct vm_operations_struct generic_file_vm_ops = { |
97 | }; |
98 | |
99 | /* |
100 | * Return the total memory allocated for this pointer, not |
101 | * just what the caller asked for. |
102 | * |
103 | * Doesn't have to be accurate, i.e. may have races. |
104 | */ |
105 | unsigned int kobjsize(const void *objp) |
106 | { |
107 | struct page *page; |
108 | |
109 | /* |
110 | * If the object we have should not have ksize performed on it, |
111 | * return size of 0 |
112 | */ |
113 | if (!objp || !virt_addr_valid(objp)) |
114 | return 0; |
115 | |
116 | page = virt_to_head_page(objp); |
117 | |
118 | /* |
119 | * If the allocator sets PageSlab, we know the pointer came from |
120 | * kmalloc(). |
121 | */ |
122 | if (PageSlab(page)) |
123 | return ksize(objp); |
124 | |
125 | /* |
126 | * If it's not a compound page, see if we have a matching VMA |
127 | * region. This test is intentionally done in reverse order, |
128 | * so if there's no VMA, we still fall through and hand back |
129 | * PAGE_SIZE for 0-order pages. |
130 | */ |
131 | if (!PageCompound(page)) { |
132 | struct vm_area_struct *vma; |
133 | |
134 | vma = find_vma(current->mm, (unsigned long)objp); |
135 | if (vma) |
136 | return vma->vm_end - vma->vm_start; |
137 | } |
138 | |
139 | /* |
140 | * The ksize() function is only guaranteed to work for pointers |
141 | * returned by kmalloc(). So handle arbitrary pointers here. |
142 | */ |
143 | return PAGE_SIZE << compound_order(page); |
144 | } |
145 | |
146 | long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm, |
147 | unsigned long start, unsigned long nr_pages, |
148 | unsigned int foll_flags, struct page **pages, |
149 | struct vm_area_struct **vmas, int *nonblocking) |
150 | { |
151 | struct vm_area_struct *vma; |
152 | unsigned long vm_flags; |
153 | int i; |
154 | |
155 | /* calculate required read or write permissions. |
156 | * If FOLL_FORCE is set, we only require the "MAY" flags. |
157 | */ |
158 | vm_flags = (foll_flags & FOLL_WRITE) ? |
159 | (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); |
160 | vm_flags &= (foll_flags & FOLL_FORCE) ? |
161 | (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); |
162 | |
163 | for (i = 0; i < nr_pages; i++) { |
164 | vma = find_vma(mm, start); |
165 | if (!vma) |
166 | goto finish_or_fault; |
167 | |
168 | /* protect what we can, including chardevs */ |
169 | if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) || |
170 | !(vm_flags & vma->vm_flags)) |
171 | goto finish_or_fault; |
172 | |
173 | if (pages) { |
174 | pages[i] = virt_to_page(start); |
175 | if (pages[i]) |
176 | page_cache_get(pages[i]); |
177 | } |
178 | if (vmas) |
179 | vmas[i] = vma; |
180 | start = (start + PAGE_SIZE) & PAGE_MASK; |
181 | } |
182 | |
183 | return i; |
184 | |
185 | finish_or_fault: |
186 | return i ? : -EFAULT; |
187 | } |
188 | |
189 | /* |
190 | * get a list of pages in an address range belonging to the specified process |
191 | * and indicate the VMA that covers each page |
192 | * - this is potentially dodgy as we may end incrementing the page count of a |
193 | * slab page or a secondary page from a compound page |
194 | * - don't permit access to VMAs that don't support it, such as I/O mappings |
195 | */ |
196 | long get_user_pages(struct task_struct *tsk, struct mm_struct *mm, |
197 | unsigned long start, unsigned long nr_pages, |
198 | int write, int force, struct page **pages, |
199 | struct vm_area_struct **vmas) |
200 | { |
201 | int flags = 0; |
202 | |
203 | if (write) |
204 | flags |= FOLL_WRITE; |
205 | if (force) |
206 | flags |= FOLL_FORCE; |
207 | |
208 | return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas, |
209 | NULL); |
210 | } |
211 | EXPORT_SYMBOL(get_user_pages); |
212 | |
213 | /** |
214 | * follow_pfn - look up PFN at a user virtual address |
215 | * @vma: memory mapping |
216 | * @address: user virtual address |
217 | * @pfn: location to store found PFN |
218 | * |
219 | * Only IO mappings and raw PFN mappings are allowed. |
220 | * |
221 | * Returns zero and the pfn at @pfn on success, -ve otherwise. |
222 | */ |
223 | int follow_pfn(struct vm_area_struct *vma, unsigned long address, |
224 | unsigned long *pfn) |
225 | { |
226 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) |
227 | return -EINVAL; |
228 | |
229 | *pfn = address >> PAGE_SHIFT; |
230 | return 0; |
231 | } |
232 | EXPORT_SYMBOL(follow_pfn); |
233 | |
234 | LIST_HEAD(vmap_area_list); |
235 | |
236 | void vfree(const void *addr) |
237 | { |
238 | kfree(addr); |
239 | } |
240 | EXPORT_SYMBOL(vfree); |
241 | |
242 | void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) |
243 | { |
244 | /* |
245 | * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc() |
246 | * returns only a logical address. |
247 | */ |
248 | return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM); |
249 | } |
250 | EXPORT_SYMBOL(__vmalloc); |
251 | |
252 | void *vmalloc_user(unsigned long size) |
253 | { |
254 | void *ret; |
255 | |
256 | ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, |
257 | PAGE_KERNEL); |
258 | if (ret) { |
259 | struct vm_area_struct *vma; |
260 | |
261 | down_write(¤t->mm->mmap_sem); |
262 | vma = find_vma(current->mm, (unsigned long)ret); |
263 | if (vma) |
264 | vma->vm_flags |= VM_USERMAP; |
265 | up_write(¤t->mm->mmap_sem); |
266 | } |
267 | |
268 | return ret; |
269 | } |
270 | EXPORT_SYMBOL(vmalloc_user); |
271 | |
272 | struct page *vmalloc_to_page(const void *addr) |
273 | { |
274 | return virt_to_page(addr); |
275 | } |
276 | EXPORT_SYMBOL(vmalloc_to_page); |
277 | |
278 | unsigned long vmalloc_to_pfn(const void *addr) |
279 | { |
280 | return page_to_pfn(virt_to_page(addr)); |
281 | } |
282 | EXPORT_SYMBOL(vmalloc_to_pfn); |
283 | |
284 | long vread(char *buf, char *addr, unsigned long count) |
285 | { |
286 | /* Don't allow overflow */ |
287 | if ((unsigned long) buf + count < count) |
288 | count = -(unsigned long) buf; |
289 | |
290 | memcpy(buf, addr, count); |
291 | return count; |
292 | } |
293 | |
294 | long vwrite(char *buf, char *addr, unsigned long count) |
295 | { |
296 | /* Don't allow overflow */ |
297 | if ((unsigned long) addr + count < count) |
298 | count = -(unsigned long) addr; |
299 | |
300 | memcpy(addr, buf, count); |
301 | return count; |
302 | } |
303 | |
304 | /* |
305 | * vmalloc - allocate virtually continguos memory |
306 | * |
307 | * @size: allocation size |
308 | * |
309 | * Allocate enough pages to cover @size from the page level |
310 | * allocator and map them into continguos kernel virtual space. |
311 | * |
312 | * For tight control over page level allocator and protection flags |
313 | * use __vmalloc() instead. |
314 | */ |
315 | void *vmalloc(unsigned long size) |
316 | { |
317 | return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL); |
318 | } |
319 | EXPORT_SYMBOL(vmalloc); |
320 | |
321 | /* |
322 | * vzalloc - allocate virtually continguos memory with zero fill |
323 | * |
324 | * @size: allocation size |
325 | * |
326 | * Allocate enough pages to cover @size from the page level |
327 | * allocator and map them into continguos kernel virtual space. |
328 | * The memory allocated is set to zero. |
329 | * |
330 | * For tight control over page level allocator and protection flags |
331 | * use __vmalloc() instead. |
332 | */ |
333 | void *vzalloc(unsigned long size) |
334 | { |
335 | return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, |
336 | PAGE_KERNEL); |
337 | } |
338 | EXPORT_SYMBOL(vzalloc); |
339 | |
340 | /** |
341 | * vmalloc_node - allocate memory on a specific node |
342 | * @size: allocation size |
343 | * @node: numa node |
344 | * |
345 | * Allocate enough pages to cover @size from the page level |
346 | * allocator and map them into contiguous kernel virtual space. |
347 | * |
348 | * For tight control over page level allocator and protection flags |
349 | * use __vmalloc() instead. |
350 | */ |
351 | void *vmalloc_node(unsigned long size, int node) |
352 | { |
353 | return vmalloc(size); |
354 | } |
355 | EXPORT_SYMBOL(vmalloc_node); |
356 | |
357 | /** |
358 | * vzalloc_node - allocate memory on a specific node with zero fill |
359 | * @size: allocation size |
360 | * @node: numa node |
361 | * |
362 | * Allocate enough pages to cover @size from the page level |
363 | * allocator and map them into contiguous kernel virtual space. |
364 | * The memory allocated is set to zero. |
365 | * |
366 | * For tight control over page level allocator and protection flags |
367 | * use __vmalloc() instead. |
368 | */ |
369 | void *vzalloc_node(unsigned long size, int node) |
370 | { |
371 | return vzalloc(size); |
372 | } |
373 | EXPORT_SYMBOL(vzalloc_node); |
374 | |
375 | #ifndef PAGE_KERNEL_EXEC |
376 | # define PAGE_KERNEL_EXEC PAGE_KERNEL |
377 | #endif |
378 | |
379 | /** |
380 | * vmalloc_exec - allocate virtually contiguous, executable memory |
381 | * @size: allocation size |
382 | * |
383 | * Kernel-internal function to allocate enough pages to cover @size |
384 | * the page level allocator and map them into contiguous and |
385 | * executable kernel virtual space. |
386 | * |
387 | * For tight control over page level allocator and protection flags |
388 | * use __vmalloc() instead. |
389 | */ |
390 | |
391 | void *vmalloc_exec(unsigned long size) |
392 | { |
393 | return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC); |
394 | } |
395 | |
396 | /** |
397 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) |
398 | * @size: allocation size |
399 | * |
400 | * Allocate enough 32bit PA addressable pages to cover @size from the |
401 | * page level allocator and map them into continguos kernel virtual space. |
402 | */ |
403 | void *vmalloc_32(unsigned long size) |
404 | { |
405 | return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL); |
406 | } |
407 | EXPORT_SYMBOL(vmalloc_32); |
408 | |
409 | /** |
410 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory |
411 | * @size: allocation size |
412 | * |
413 | * The resulting memory area is 32bit addressable and zeroed so it can be |
414 | * mapped to userspace without leaking data. |
415 | * |
416 | * VM_USERMAP is set on the corresponding VMA so that subsequent calls to |
417 | * remap_vmalloc_range() are permissible. |
418 | */ |
419 | void *vmalloc_32_user(unsigned long size) |
420 | { |
421 | /* |
422 | * We'll have to sort out the ZONE_DMA bits for 64-bit, |
423 | * but for now this can simply use vmalloc_user() directly. |
424 | */ |
425 | return vmalloc_user(size); |
426 | } |
427 | EXPORT_SYMBOL(vmalloc_32_user); |
428 | |
429 | void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot) |
430 | { |
431 | BUG(); |
432 | return NULL; |
433 | } |
434 | EXPORT_SYMBOL(vmap); |
435 | |
436 | void vunmap(const void *addr) |
437 | { |
438 | BUG(); |
439 | } |
440 | EXPORT_SYMBOL(vunmap); |
441 | |
442 | void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) |
443 | { |
444 | BUG(); |
445 | return NULL; |
446 | } |
447 | EXPORT_SYMBOL(vm_map_ram); |
448 | |
449 | void vm_unmap_ram(const void *mem, unsigned int count) |
450 | { |
451 | BUG(); |
452 | } |
453 | EXPORT_SYMBOL(vm_unmap_ram); |
454 | |
455 | void vm_unmap_aliases(void) |
456 | { |
457 | } |
458 | EXPORT_SYMBOL_GPL(vm_unmap_aliases); |
459 | |
460 | /* |
461 | * Implement a stub for vmalloc_sync_all() if the architecture chose not to |
462 | * have one. |
463 | */ |
464 | void __weak vmalloc_sync_all(void) |
465 | { |
466 | } |
467 | |
468 | /** |
469 | * alloc_vm_area - allocate a range of kernel address space |
470 | * @size: size of the area |
471 | * |
472 | * Returns: NULL on failure, vm_struct on success |
473 | * |
474 | * This function reserves a range of kernel address space, and |
475 | * allocates pagetables to map that range. No actual mappings |
476 | * are created. If the kernel address space is not shared |
477 | * between processes, it syncs the pagetable across all |
478 | * processes. |
479 | */ |
480 | struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes) |
481 | { |
482 | BUG(); |
483 | return NULL; |
484 | } |
485 | EXPORT_SYMBOL_GPL(alloc_vm_area); |
486 | |
487 | void free_vm_area(struct vm_struct *area) |
488 | { |
489 | BUG(); |
490 | } |
491 | EXPORT_SYMBOL_GPL(free_vm_area); |
492 | |
493 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, |
494 | struct page *page) |
495 | { |
496 | return -EINVAL; |
497 | } |
498 | EXPORT_SYMBOL(vm_insert_page); |
499 | |
500 | /* |
501 | * sys_brk() for the most part doesn't need the global kernel |
502 | * lock, except when an application is doing something nasty |
503 | * like trying to un-brk an area that has already been mapped |
504 | * to a regular file. in this case, the unmapping will need |
505 | * to invoke file system routines that need the global lock. |
506 | */ |
507 | SYSCALL_DEFINE1(brk, unsigned long, brk) |
508 | { |
509 | struct mm_struct *mm = current->mm; |
510 | |
511 | if (brk < mm->start_brk || brk > mm->context.end_brk) |
512 | return mm->brk; |
513 | |
514 | if (mm->brk == brk) |
515 | return mm->brk; |
516 | |
517 | /* |
518 | * Always allow shrinking brk |
519 | */ |
520 | if (brk <= mm->brk) { |
521 | mm->brk = brk; |
522 | return brk; |
523 | } |
524 | |
525 | /* |
526 | * Ok, looks good - let it rip. |
527 | */ |
528 | flush_icache_range(mm->brk, brk); |
529 | return mm->brk = brk; |
530 | } |
531 | |
532 | /* |
533 | * initialise the VMA and region record slabs |
534 | */ |
535 | void __init mmap_init(void) |
536 | { |
537 | int ret; |
538 | |
539 | ret = percpu_counter_init(&vm_committed_as, 0); |
540 | VM_BUG_ON(ret); |
541 | vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC); |
542 | } |
543 | |
544 | /* |
545 | * validate the region tree |
546 | * - the caller must hold the region lock |
547 | */ |
548 | #ifdef CONFIG_DEBUG_NOMMU_REGIONS |
549 | static noinline void validate_nommu_regions(void) |
550 | { |
551 | struct vm_region *region, *last; |
552 | struct rb_node *p, *lastp; |
553 | |
554 | lastp = rb_first(&nommu_region_tree); |
555 | if (!lastp) |
556 | return; |
557 | |
558 | last = rb_entry(lastp, struct vm_region, vm_rb); |
559 | BUG_ON(unlikely(last->vm_end <= last->vm_start)); |
560 | BUG_ON(unlikely(last->vm_top < last->vm_end)); |
561 | |
562 | while ((p = rb_next(lastp))) { |
563 | region = rb_entry(p, struct vm_region, vm_rb); |
564 | last = rb_entry(lastp, struct vm_region, vm_rb); |
565 | |
566 | BUG_ON(unlikely(region->vm_end <= region->vm_start)); |
567 | BUG_ON(unlikely(region->vm_top < region->vm_end)); |
568 | BUG_ON(unlikely(region->vm_start < last->vm_top)); |
569 | |
570 | lastp = p; |
571 | } |
572 | } |
573 | #else |
574 | static void validate_nommu_regions(void) |
575 | { |
576 | } |
577 | #endif |
578 | |
579 | /* |
580 | * add a region into the global tree |
581 | */ |
582 | static void add_nommu_region(struct vm_region *region) |
583 | { |
584 | struct vm_region *pregion; |
585 | struct rb_node **p, *parent; |
586 | |
587 | validate_nommu_regions(); |
588 | |
589 | parent = NULL; |
590 | p = &nommu_region_tree.rb_node; |
591 | while (*p) { |
592 | parent = *p; |
593 | pregion = rb_entry(parent, struct vm_region, vm_rb); |
594 | if (region->vm_start < pregion->vm_start) |
595 | p = &(*p)->rb_left; |
596 | else if (region->vm_start > pregion->vm_start) |
597 | p = &(*p)->rb_right; |
598 | else if (pregion == region) |
599 | return; |
600 | else |
601 | BUG(); |
602 | } |
603 | |
604 | rb_link_node(®ion->vm_rb, parent, p); |
605 | rb_insert_color(®ion->vm_rb, &nommu_region_tree); |
606 | |
607 | validate_nommu_regions(); |
608 | } |
609 | |
610 | /* |
611 | * delete a region from the global tree |
612 | */ |
613 | static void delete_nommu_region(struct vm_region *region) |
614 | { |
615 | BUG_ON(!nommu_region_tree.rb_node); |
616 | |
617 | validate_nommu_regions(); |
618 | rb_erase(®ion->vm_rb, &nommu_region_tree); |
619 | validate_nommu_regions(); |
620 | } |
621 | |
622 | /* |
623 | * free a contiguous series of pages |
624 | */ |
625 | static void free_page_series(unsigned long from, unsigned long to) |
626 | { |
627 | for (; from < to; from += PAGE_SIZE) { |
628 | struct page *page = virt_to_page(from); |
629 | |
630 | kdebug("- free %lx", from); |
631 | atomic_long_dec(&mmap_pages_allocated); |
632 | if (page_count(page) != 1) |
633 | kdebug("free page %p: refcount not one: %d", |
634 | page, page_count(page)); |
635 | put_page(page); |
636 | } |
637 | } |
638 | |
639 | /* |
640 | * release a reference to a region |
641 | * - the caller must hold the region semaphore for writing, which this releases |
642 | * - the region may not have been added to the tree yet, in which case vm_top |
643 | * will equal vm_start |
644 | */ |
645 | static void __put_nommu_region(struct vm_region *region) |
646 | __releases(nommu_region_sem) |
647 | { |
648 | kenter("%p{%d}", region, region->vm_usage); |
649 | |
650 | BUG_ON(!nommu_region_tree.rb_node); |
651 | |
652 | if (--region->vm_usage == 0) { |
653 | if (region->vm_top > region->vm_start) |
654 | delete_nommu_region(region); |
655 | up_write(&nommu_region_sem); |
656 | |
657 | if (region->vm_file) |
658 | fput(region->vm_file); |
659 | |
660 | /* IO memory and memory shared directly out of the pagecache |
661 | * from ramfs/tmpfs mustn't be released here */ |
662 | if (region->vm_flags & VM_MAPPED_COPY) { |
663 | kdebug("free series"); |
664 | free_page_series(region->vm_start, region->vm_top); |
665 | } |
666 | kmem_cache_free(vm_region_jar, region); |
667 | } else { |
668 | up_write(&nommu_region_sem); |
669 | } |
670 | } |
671 | |
672 | /* |
673 | * release a reference to a region |
674 | */ |
675 | static void put_nommu_region(struct vm_region *region) |
676 | { |
677 | down_write(&nommu_region_sem); |
678 | __put_nommu_region(region); |
679 | } |
680 | |
681 | /* |
682 | * update protection on a vma |
683 | */ |
684 | static void protect_vma(struct vm_area_struct *vma, unsigned long flags) |
685 | { |
686 | #ifdef CONFIG_MPU |
687 | struct mm_struct *mm = vma->vm_mm; |
688 | long start = vma->vm_start & PAGE_MASK; |
689 | while (start < vma->vm_end) { |
690 | protect_page(mm, start, flags); |
691 | start += PAGE_SIZE; |
692 | } |
693 | update_protections(mm); |
694 | #endif |
695 | } |
696 | |
697 | /* |
698 | * add a VMA into a process's mm_struct in the appropriate place in the list |
699 | * and tree and add to the address space's page tree also if not an anonymous |
700 | * page |
701 | * - should be called with mm->mmap_sem held writelocked |
702 | */ |
703 | static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma) |
704 | { |
705 | struct vm_area_struct *pvma, *prev; |
706 | struct address_space *mapping; |
707 | struct rb_node **p, *parent, *rb_prev; |
708 | |
709 | kenter(",%p", vma); |
710 | |
711 | BUG_ON(!vma->vm_region); |
712 | |
713 | mm->map_count++; |
714 | vma->vm_mm = mm; |
715 | |
716 | protect_vma(vma, vma->vm_flags); |
717 | |
718 | /* add the VMA to the mapping */ |
719 | if (vma->vm_file) { |
720 | mapping = vma->vm_file->f_mapping; |
721 | |
722 | mutex_lock(&mapping->i_mmap_mutex); |
723 | flush_dcache_mmap_lock(mapping); |
724 | vma_interval_tree_insert(vma, &mapping->i_mmap); |
725 | flush_dcache_mmap_unlock(mapping); |
726 | mutex_unlock(&mapping->i_mmap_mutex); |
727 | } |
728 | |
729 | /* add the VMA to the tree */ |
730 | parent = rb_prev = NULL; |
731 | p = &mm->mm_rb.rb_node; |
732 | while (*p) { |
733 | parent = *p; |
734 | pvma = rb_entry(parent, struct vm_area_struct, vm_rb); |
735 | |
736 | /* sort by: start addr, end addr, VMA struct addr in that order |
737 | * (the latter is necessary as we may get identical VMAs) */ |
738 | if (vma->vm_start < pvma->vm_start) |
739 | p = &(*p)->rb_left; |
740 | else if (vma->vm_start > pvma->vm_start) { |
741 | rb_prev = parent; |
742 | p = &(*p)->rb_right; |
743 | } else if (vma->vm_end < pvma->vm_end) |
744 | p = &(*p)->rb_left; |
745 | else if (vma->vm_end > pvma->vm_end) { |
746 | rb_prev = parent; |
747 | p = &(*p)->rb_right; |
748 | } else if (vma < pvma) |
749 | p = &(*p)->rb_left; |
750 | else if (vma > pvma) { |
751 | rb_prev = parent; |
752 | p = &(*p)->rb_right; |
753 | } else |
754 | BUG(); |
755 | } |
756 | |
757 | rb_link_node(&vma->vm_rb, parent, p); |
758 | rb_insert_color(&vma->vm_rb, &mm->mm_rb); |
759 | |
760 | /* add VMA to the VMA list also */ |
761 | prev = NULL; |
762 | if (rb_prev) |
763 | prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); |
764 | |
765 | __vma_link_list(mm, vma, prev, parent); |
766 | } |
767 | |
768 | /* |
769 | * delete a VMA from its owning mm_struct and address space |
770 | */ |
771 | static void delete_vma_from_mm(struct vm_area_struct *vma) |
772 | { |
773 | int i; |
774 | struct address_space *mapping; |
775 | struct mm_struct *mm = vma->vm_mm; |
776 | struct task_struct *curr = current; |
777 | |
778 | kenter("%p", vma); |
779 | |
780 | protect_vma(vma, 0); |
781 | |
782 | mm->map_count--; |
783 | for (i = 0; i < VMACACHE_SIZE; i++) { |
784 | /* if the vma is cached, invalidate the entire cache */ |
785 | if (curr->vmacache[i] == vma) { |
786 | vmacache_invalidate(curr->mm); |
787 | break; |
788 | } |
789 | } |
790 | |
791 | /* remove the VMA from the mapping */ |
792 | if (vma->vm_file) { |
793 | mapping = vma->vm_file->f_mapping; |
794 | |
795 | mutex_lock(&mapping->i_mmap_mutex); |
796 | flush_dcache_mmap_lock(mapping); |
797 | vma_interval_tree_remove(vma, &mapping->i_mmap); |
798 | flush_dcache_mmap_unlock(mapping); |
799 | mutex_unlock(&mapping->i_mmap_mutex); |
800 | } |
801 | |
802 | /* remove from the MM's tree and list */ |
803 | rb_erase(&vma->vm_rb, &mm->mm_rb); |
804 | |
805 | if (vma->vm_prev) |
806 | vma->vm_prev->vm_next = vma->vm_next; |
807 | else |
808 | mm->mmap = vma->vm_next; |
809 | |
810 | if (vma->vm_next) |
811 | vma->vm_next->vm_prev = vma->vm_prev; |
812 | } |
813 | |
814 | /* |
815 | * destroy a VMA record |
816 | */ |
817 | static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma) |
818 | { |
819 | kenter("%p", vma); |
820 | if (vma->vm_ops && vma->vm_ops->close) |
821 | vma->vm_ops->close(vma); |
822 | if (vma->vm_file) |
823 | fput(vma->vm_file); |
824 | put_nommu_region(vma->vm_region); |
825 | kmem_cache_free(vm_area_cachep, vma); |
826 | } |
827 | |
828 | /* |
829 | * look up the first VMA in which addr resides, NULL if none |
830 | * - should be called with mm->mmap_sem at least held readlocked |
831 | */ |
832 | struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) |
833 | { |
834 | struct vm_area_struct *vma; |
835 | |
836 | /* check the cache first */ |
837 | vma = vmacache_find(mm, addr); |
838 | if (likely(vma)) |
839 | return vma; |
840 | |
841 | /* trawl the list (there may be multiple mappings in which addr |
842 | * resides) */ |
843 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
844 | if (vma->vm_start > addr) |
845 | return NULL; |
846 | if (vma->vm_end > addr) { |
847 | vmacache_update(addr, vma); |
848 | return vma; |
849 | } |
850 | } |
851 | |
852 | return NULL; |
853 | } |
854 | EXPORT_SYMBOL(find_vma); |
855 | |
856 | /* |
857 | * find a VMA |
858 | * - we don't extend stack VMAs under NOMMU conditions |
859 | */ |
860 | struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr) |
861 | { |
862 | return find_vma(mm, addr); |
863 | } |
864 | |
865 | /* |
866 | * expand a stack to a given address |
867 | * - not supported under NOMMU conditions |
868 | */ |
869 | int expand_stack(struct vm_area_struct *vma, unsigned long address) |
870 | { |
871 | return -ENOMEM; |
872 | } |
873 | |
874 | /* |
875 | * look up the first VMA exactly that exactly matches addr |
876 | * - should be called with mm->mmap_sem at least held readlocked |
877 | */ |
878 | static struct vm_area_struct *find_vma_exact(struct mm_struct *mm, |
879 | unsigned long addr, |
880 | unsigned long len) |
881 | { |
882 | struct vm_area_struct *vma; |
883 | unsigned long end = addr + len; |
884 | |
885 | /* check the cache first */ |
886 | vma = vmacache_find_exact(mm, addr, end); |
887 | if (vma) |
888 | return vma; |
889 | |
890 | /* trawl the list (there may be multiple mappings in which addr |
891 | * resides) */ |
892 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
893 | if (vma->vm_start < addr) |
894 | continue; |
895 | if (vma->vm_start > addr) |
896 | return NULL; |
897 | if (vma->vm_end == end) { |
898 | vmacache_update(addr, vma); |
899 | return vma; |
900 | } |
901 | } |
902 | |
903 | return NULL; |
904 | } |
905 | |
906 | /* |
907 | * determine whether a mapping should be permitted and, if so, what sort of |
908 | * mapping we're capable of supporting |
909 | */ |
910 | static int validate_mmap_request(struct file *file, |
911 | unsigned long addr, |
912 | unsigned long len, |
913 | unsigned long prot, |
914 | unsigned long flags, |
915 | unsigned long pgoff, |
916 | unsigned long *_capabilities) |
917 | { |
918 | unsigned long capabilities, rlen; |
919 | int ret; |
920 | |
921 | /* do the simple checks first */ |
922 | if (flags & MAP_FIXED) { |
923 | printk(KERN_DEBUG |
924 | "%d: Can't do fixed-address/overlay mmap of RAM\n", |
925 | current->pid); |
926 | return -EINVAL; |
927 | } |
928 | |
929 | if ((flags & MAP_TYPE) != MAP_PRIVATE && |
930 | (flags & MAP_TYPE) != MAP_SHARED) |
931 | return -EINVAL; |
932 | |
933 | if (!len) |
934 | return -EINVAL; |
935 | |
936 | /* Careful about overflows.. */ |
937 | rlen = PAGE_ALIGN(len); |
938 | if (!rlen || rlen > TASK_SIZE) |
939 | return -ENOMEM; |
940 | |
941 | /* offset overflow? */ |
942 | if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff) |
943 | return -EOVERFLOW; |
944 | |
945 | if (file) { |
946 | /* validate file mapping requests */ |
947 | struct address_space *mapping; |
948 | |
949 | /* files must support mmap */ |
950 | if (!file->f_op->mmap) |
951 | return -ENODEV; |
952 | |
953 | /* work out if what we've got could possibly be shared |
954 | * - we support chardevs that provide their own "memory" |
955 | * - we support files/blockdevs that are memory backed |
956 | */ |
957 | mapping = file->f_mapping; |
958 | if (!mapping) |
959 | mapping = file_inode(file)->i_mapping; |
960 | |
961 | capabilities = 0; |
962 | if (mapping && mapping->backing_dev_info) |
963 | capabilities = mapping->backing_dev_info->capabilities; |
964 | |
965 | if (!capabilities) { |
966 | /* no explicit capabilities set, so assume some |
967 | * defaults */ |
968 | switch (file_inode(file)->i_mode & S_IFMT) { |
969 | case S_IFREG: |
970 | case S_IFBLK: |
971 | capabilities = BDI_CAP_MAP_COPY; |
972 | break; |
973 | |
974 | case S_IFCHR: |
975 | capabilities = |
976 | BDI_CAP_MAP_DIRECT | |
977 | BDI_CAP_READ_MAP | |
978 | BDI_CAP_WRITE_MAP; |
979 | break; |
980 | |
981 | default: |
982 | return -EINVAL; |
983 | } |
984 | } |
985 | |
986 | /* eliminate any capabilities that we can't support on this |
987 | * device */ |
988 | if (!file->f_op->get_unmapped_area) |
989 | capabilities &= ~BDI_CAP_MAP_DIRECT; |
990 | if (!file->f_op->read) |
991 | capabilities &= ~BDI_CAP_MAP_COPY; |
992 | |
993 | /* The file shall have been opened with read permission. */ |
994 | if (!(file->f_mode & FMODE_READ)) |
995 | return -EACCES; |
996 | |
997 | if (flags & MAP_SHARED) { |
998 | /* do checks for writing, appending and locking */ |
999 | if ((prot & PROT_WRITE) && |
1000 | !(file->f_mode & FMODE_WRITE)) |
1001 | return -EACCES; |
1002 | |
1003 | if (IS_APPEND(file_inode(file)) && |
1004 | (file->f_mode & FMODE_WRITE)) |
1005 | return -EACCES; |
1006 | |
1007 | if (locks_verify_locked(file)) |
1008 | return -EAGAIN; |
1009 | |
1010 | if (!(capabilities & BDI_CAP_MAP_DIRECT)) |
1011 | return -ENODEV; |
1012 | |
1013 | /* we mustn't privatise shared mappings */ |
1014 | capabilities &= ~BDI_CAP_MAP_COPY; |
1015 | } else { |
1016 | /* we're going to read the file into private memory we |
1017 | * allocate */ |
1018 | if (!(capabilities & BDI_CAP_MAP_COPY)) |
1019 | return -ENODEV; |
1020 | |
1021 | /* we don't permit a private writable mapping to be |
1022 | * shared with the backing device */ |
1023 | if (prot & PROT_WRITE) |
1024 | capabilities &= ~BDI_CAP_MAP_DIRECT; |
1025 | } |
1026 | |
1027 | if (capabilities & BDI_CAP_MAP_DIRECT) { |
1028 | if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) || |
1029 | ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) || |
1030 | ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP)) |
1031 | ) { |
1032 | capabilities &= ~BDI_CAP_MAP_DIRECT; |
1033 | if (flags & MAP_SHARED) { |
1034 | printk(KERN_WARNING |
1035 | "MAP_SHARED not completely supported on !MMU\n"); |
1036 | return -EINVAL; |
1037 | } |
1038 | } |
1039 | } |
1040 | |
1041 | /* handle executable mappings and implied executable |
1042 | * mappings */ |
1043 | if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) { |
1044 | if (prot & PROT_EXEC) |
1045 | return -EPERM; |
1046 | } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) { |
1047 | /* handle implication of PROT_EXEC by PROT_READ */ |
1048 | if (current->personality & READ_IMPLIES_EXEC) { |
1049 | if (capabilities & BDI_CAP_EXEC_MAP) |
1050 | prot |= PROT_EXEC; |
1051 | } |
1052 | } else if ((prot & PROT_READ) && |
1053 | (prot & PROT_EXEC) && |
1054 | !(capabilities & BDI_CAP_EXEC_MAP) |
1055 | ) { |
1056 | /* backing file is not executable, try to copy */ |
1057 | capabilities &= ~BDI_CAP_MAP_DIRECT; |
1058 | } |
1059 | } else { |
1060 | /* anonymous mappings are always memory backed and can be |
1061 | * privately mapped |
1062 | */ |
1063 | capabilities = BDI_CAP_MAP_COPY; |
1064 | |
1065 | /* handle PROT_EXEC implication by PROT_READ */ |
1066 | if ((prot & PROT_READ) && |
1067 | (current->personality & READ_IMPLIES_EXEC)) |
1068 | prot |= PROT_EXEC; |
1069 | } |
1070 | |
1071 | /* allow the security API to have its say */ |
1072 | ret = security_mmap_addr(addr); |
1073 | if (ret < 0) |
1074 | return ret; |
1075 | |
1076 | /* looks okay */ |
1077 | *_capabilities = capabilities; |
1078 | return 0; |
1079 | } |
1080 | |
1081 | /* |
1082 | * we've determined that we can make the mapping, now translate what we |
1083 | * now know into VMA flags |
1084 | */ |
1085 | static unsigned long determine_vm_flags(struct file *file, |
1086 | unsigned long prot, |
1087 | unsigned long flags, |
1088 | unsigned long capabilities) |
1089 | { |
1090 | unsigned long vm_flags; |
1091 | |
1092 | vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags); |
1093 | /* vm_flags |= mm->def_flags; */ |
1094 | |
1095 | if (!(capabilities & BDI_CAP_MAP_DIRECT)) { |
1096 | /* attempt to share read-only copies of mapped file chunks */ |
1097 | vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; |
1098 | if (file && !(prot & PROT_WRITE)) |
1099 | vm_flags |= VM_MAYSHARE; |
1100 | } else { |
1101 | /* overlay a shareable mapping on the backing device or inode |
1102 | * if possible - used for chardevs, ramfs/tmpfs/shmfs and |
1103 | * romfs/cramfs */ |
1104 | vm_flags |= VM_MAYSHARE | (capabilities & BDI_CAP_VMFLAGS); |
1105 | if (flags & MAP_SHARED) |
1106 | vm_flags |= VM_SHARED; |
1107 | } |
1108 | |
1109 | /* refuse to let anyone share private mappings with this process if |
1110 | * it's being traced - otherwise breakpoints set in it may interfere |
1111 | * with another untraced process |
1112 | */ |
1113 | if ((flags & MAP_PRIVATE) && current->ptrace) |
1114 | vm_flags &= ~VM_MAYSHARE; |
1115 | |
1116 | return vm_flags; |
1117 | } |
1118 | |
1119 | /* |
1120 | * set up a shared mapping on a file (the driver or filesystem provides and |
1121 | * pins the storage) |
1122 | */ |
1123 | static int do_mmap_shared_file(struct vm_area_struct *vma) |
1124 | { |
1125 | int ret; |
1126 | |
1127 | ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); |
1128 | if (ret == 0) { |
1129 | vma->vm_region->vm_top = vma->vm_region->vm_end; |
1130 | return 0; |
1131 | } |
1132 | if (ret != -ENOSYS) |
1133 | return ret; |
1134 | |
1135 | /* getting -ENOSYS indicates that direct mmap isn't possible (as |
1136 | * opposed to tried but failed) so we can only give a suitable error as |
1137 | * it's not possible to make a private copy if MAP_SHARED was given */ |
1138 | return -ENODEV; |
1139 | } |
1140 | |
1141 | /* |
1142 | * set up a private mapping or an anonymous shared mapping |
1143 | */ |
1144 | static int do_mmap_private(struct vm_area_struct *vma, |
1145 | struct vm_region *region, |
1146 | unsigned long len, |
1147 | unsigned long capabilities) |
1148 | { |
1149 | struct page *pages; |
1150 | unsigned long total, point, n; |
1151 | void *base; |
1152 | int ret, order; |
1153 | |
1154 | /* invoke the file's mapping function so that it can keep track of |
1155 | * shared mappings on devices or memory |
1156 | * - VM_MAYSHARE will be set if it may attempt to share |
1157 | */ |
1158 | if (capabilities & BDI_CAP_MAP_DIRECT) { |
1159 | ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); |
1160 | if (ret == 0) { |
1161 | /* shouldn't return success if we're not sharing */ |
1162 | BUG_ON(!(vma->vm_flags & VM_MAYSHARE)); |
1163 | vma->vm_region->vm_top = vma->vm_region->vm_end; |
1164 | return 0; |
1165 | } |
1166 | if (ret != -ENOSYS) |
1167 | return ret; |
1168 | |
1169 | /* getting an ENOSYS error indicates that direct mmap isn't |
1170 | * possible (as opposed to tried but failed) so we'll try to |
1171 | * make a private copy of the data and map that instead */ |
1172 | } |
1173 | |
1174 | |
1175 | /* allocate some memory to hold the mapping |
1176 | * - note that this may not return a page-aligned address if the object |
1177 | * we're allocating is smaller than a page |
1178 | */ |
1179 | order = get_order(len); |
1180 | kdebug("alloc order %d for %lx", order, len); |
1181 | |
1182 | pages = alloc_pages(GFP_KERNEL, order); |
1183 | if (!pages) |
1184 | goto enomem; |
1185 | |
1186 | total = 1 << order; |
1187 | atomic_long_add(total, &mmap_pages_allocated); |
1188 | |
1189 | point = len >> PAGE_SHIFT; |
1190 | |
1191 | /* we allocated a power-of-2 sized page set, so we may want to trim off |
1192 | * the excess */ |
1193 | if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) { |
1194 | while (total > point) { |
1195 | order = ilog2(total - point); |
1196 | n = 1 << order; |
1197 | kdebug("shave %lu/%lu @%lu", n, total - point, total); |
1198 | atomic_long_sub(n, &mmap_pages_allocated); |
1199 | total -= n; |
1200 | set_page_refcounted(pages + total); |
1201 | __free_pages(pages + total, order); |
1202 | } |
1203 | } |
1204 | |
1205 | for (point = 1; point < total; point++) |
1206 | set_page_refcounted(&pages[point]); |
1207 | |
1208 | base = page_address(pages); |
1209 | region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY; |
1210 | region->vm_start = (unsigned long) base; |
1211 | region->vm_end = region->vm_start + len; |
1212 | region->vm_top = region->vm_start + (total << PAGE_SHIFT); |
1213 | |
1214 | vma->vm_start = region->vm_start; |
1215 | vma->vm_end = region->vm_start + len; |
1216 | |
1217 | if (vma->vm_file) { |
1218 | /* read the contents of a file into the copy */ |
1219 | mm_segment_t old_fs; |
1220 | loff_t fpos; |
1221 | |
1222 | fpos = vma->vm_pgoff; |
1223 | fpos <<= PAGE_SHIFT; |
1224 | |
1225 | old_fs = get_fs(); |
1226 | set_fs(KERNEL_DS); |
1227 | ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos); |
1228 | set_fs(old_fs); |
1229 | |
1230 | if (ret < 0) |
1231 | goto error_free; |
1232 | |
1233 | /* clear the last little bit */ |
1234 | if (ret < len) |
1235 | memset(base + ret, 0, len - ret); |
1236 | |
1237 | } |
1238 | |
1239 | return 0; |
1240 | |
1241 | error_free: |
1242 | free_page_series(region->vm_start, region->vm_top); |
1243 | region->vm_start = vma->vm_start = 0; |
1244 | region->vm_end = vma->vm_end = 0; |
1245 | region->vm_top = 0; |
1246 | return ret; |
1247 | |
1248 | enomem: |
1249 | printk("Allocation of length %lu from process %d (%s) failed\n", |
1250 | len, current->pid, current->comm); |
1251 | show_free_areas(0); |
1252 | return -ENOMEM; |
1253 | } |
1254 | |
1255 | /* |
1256 | * handle mapping creation for uClinux |
1257 | */ |
1258 | unsigned long do_mmap_pgoff(struct file *file, |
1259 | unsigned long addr, |
1260 | unsigned long len, |
1261 | unsigned long prot, |
1262 | unsigned long flags, |
1263 | unsigned long pgoff, |
1264 | unsigned long *populate) |
1265 | { |
1266 | struct vm_area_struct *vma; |
1267 | struct vm_region *region; |
1268 | struct rb_node *rb; |
1269 | unsigned long capabilities, vm_flags, result; |
1270 | int ret; |
1271 | |
1272 | kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff); |
1273 | |
1274 | *populate = 0; |
1275 | |
1276 | /* decide whether we should attempt the mapping, and if so what sort of |
1277 | * mapping */ |
1278 | ret = validate_mmap_request(file, addr, len, prot, flags, pgoff, |
1279 | &capabilities); |
1280 | if (ret < 0) { |
1281 | kleave(" = %d [val]", ret); |
1282 | return ret; |
1283 | } |
1284 | |
1285 | /* we ignore the address hint */ |
1286 | addr = 0; |
1287 | len = PAGE_ALIGN(len); |
1288 | |
1289 | /* we've determined that we can make the mapping, now translate what we |
1290 | * now know into VMA flags */ |
1291 | vm_flags = determine_vm_flags(file, prot, flags, capabilities); |
1292 | |
1293 | /* we're going to need to record the mapping */ |
1294 | region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL); |
1295 | if (!region) |
1296 | goto error_getting_region; |
1297 | |
1298 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); |
1299 | if (!vma) |
1300 | goto error_getting_vma; |
1301 | |
1302 | region->vm_usage = 1; |
1303 | region->vm_flags = vm_flags; |
1304 | region->vm_pgoff = pgoff; |
1305 | |
1306 | INIT_LIST_HEAD(&vma->anon_vma_chain); |
1307 | vma->vm_flags = vm_flags; |
1308 | vma->vm_pgoff = pgoff; |
1309 | |
1310 | if (file) { |
1311 | region->vm_file = get_file(file); |
1312 | vma->vm_file = get_file(file); |
1313 | } |
1314 | |
1315 | down_write(&nommu_region_sem); |
1316 | |
1317 | /* if we want to share, we need to check for regions created by other |
1318 | * mmap() calls that overlap with our proposed mapping |
1319 | * - we can only share with a superset match on most regular files |
1320 | * - shared mappings on character devices and memory backed files are |
1321 | * permitted to overlap inexactly as far as we are concerned for in |
1322 | * these cases, sharing is handled in the driver or filesystem rather |
1323 | * than here |
1324 | */ |
1325 | if (vm_flags & VM_MAYSHARE) { |
1326 | struct vm_region *pregion; |
1327 | unsigned long pglen, rpglen, pgend, rpgend, start; |
1328 | |
1329 | pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
1330 | pgend = pgoff + pglen; |
1331 | |
1332 | for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) { |
1333 | pregion = rb_entry(rb, struct vm_region, vm_rb); |
1334 | |
1335 | if (!(pregion->vm_flags & VM_MAYSHARE)) |
1336 | continue; |
1337 | |
1338 | /* search for overlapping mappings on the same file */ |
1339 | if (file_inode(pregion->vm_file) != |
1340 | file_inode(file)) |
1341 | continue; |
1342 | |
1343 | if (pregion->vm_pgoff >= pgend) |
1344 | continue; |
1345 | |
1346 | rpglen = pregion->vm_end - pregion->vm_start; |
1347 | rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT; |
1348 | rpgend = pregion->vm_pgoff + rpglen; |
1349 | if (pgoff >= rpgend) |
1350 | continue; |
1351 | |
1352 | /* handle inexactly overlapping matches between |
1353 | * mappings */ |
1354 | if ((pregion->vm_pgoff != pgoff || rpglen != pglen) && |
1355 | !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) { |
1356 | /* new mapping is not a subset of the region */ |
1357 | if (!(capabilities & BDI_CAP_MAP_DIRECT)) |
1358 | goto sharing_violation; |
1359 | continue; |
1360 | } |
1361 | |
1362 | /* we've found a region we can share */ |
1363 | pregion->vm_usage++; |
1364 | vma->vm_region = pregion; |
1365 | start = pregion->vm_start; |
1366 | start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT; |
1367 | vma->vm_start = start; |
1368 | vma->vm_end = start + len; |
1369 | |
1370 | if (pregion->vm_flags & VM_MAPPED_COPY) { |
1371 | kdebug("share copy"); |
1372 | vma->vm_flags |= VM_MAPPED_COPY; |
1373 | } else { |
1374 | kdebug("share mmap"); |
1375 | ret = do_mmap_shared_file(vma); |
1376 | if (ret < 0) { |
1377 | vma->vm_region = NULL; |
1378 | vma->vm_start = 0; |
1379 | vma->vm_end = 0; |
1380 | pregion->vm_usage--; |
1381 | pregion = NULL; |
1382 | goto error_just_free; |
1383 | } |
1384 | } |
1385 | fput(region->vm_file); |
1386 | kmem_cache_free(vm_region_jar, region); |
1387 | region = pregion; |
1388 | result = start; |
1389 | goto share; |
1390 | } |
1391 | |
1392 | /* obtain the address at which to make a shared mapping |
1393 | * - this is the hook for quasi-memory character devices to |
1394 | * tell us the location of a shared mapping |
1395 | */ |
1396 | if (capabilities & BDI_CAP_MAP_DIRECT) { |
1397 | addr = file->f_op->get_unmapped_area(file, addr, len, |
1398 | pgoff, flags); |
1399 | if (IS_ERR_VALUE(addr)) { |
1400 | ret = addr; |
1401 | if (ret != -ENOSYS) |
1402 | goto error_just_free; |
1403 | |
1404 | /* the driver refused to tell us where to site |
1405 | * the mapping so we'll have to attempt to copy |
1406 | * it */ |
1407 | ret = -ENODEV; |
1408 | if (!(capabilities & BDI_CAP_MAP_COPY)) |
1409 | goto error_just_free; |
1410 | |
1411 | capabilities &= ~BDI_CAP_MAP_DIRECT; |
1412 | } else { |
1413 | vma->vm_start = region->vm_start = addr; |
1414 | vma->vm_end = region->vm_end = addr + len; |
1415 | } |
1416 | } |
1417 | } |
1418 | |
1419 | vma->vm_region = region; |
1420 | |
1421 | /* set up the mapping |
1422 | * - the region is filled in if BDI_CAP_MAP_DIRECT is still set |
1423 | */ |
1424 | if (file && vma->vm_flags & VM_SHARED) |
1425 | ret = do_mmap_shared_file(vma); |
1426 | else |
1427 | ret = do_mmap_private(vma, region, len, capabilities); |
1428 | if (ret < 0) |
1429 | goto error_just_free; |
1430 | add_nommu_region(region); |
1431 | |
1432 | /* clear anonymous mappings that don't ask for uninitialized data */ |
1433 | if (!vma->vm_file && !(flags & MAP_UNINITIALIZED)) |
1434 | memset((void *)region->vm_start, 0, |
1435 | region->vm_end - region->vm_start); |
1436 | |
1437 | /* okay... we have a mapping; now we have to register it */ |
1438 | result = vma->vm_start; |
1439 | |
1440 | current->mm->total_vm += len >> PAGE_SHIFT; |
1441 | |
1442 | share: |
1443 | add_vma_to_mm(current->mm, vma); |
1444 | |
1445 | /* we flush the region from the icache only when the first executable |
1446 | * mapping of it is made */ |
1447 | if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) { |
1448 | flush_icache_range(region->vm_start, region->vm_end); |
1449 | region->vm_icache_flushed = true; |
1450 | } |
1451 | |
1452 | up_write(&nommu_region_sem); |
1453 | |
1454 | kleave(" = %lx", result); |
1455 | return result; |
1456 | |
1457 | error_just_free: |
1458 | up_write(&nommu_region_sem); |
1459 | error: |
1460 | if (region->vm_file) |
1461 | fput(region->vm_file); |
1462 | kmem_cache_free(vm_region_jar, region); |
1463 | if (vma->vm_file) |
1464 | fput(vma->vm_file); |
1465 | kmem_cache_free(vm_area_cachep, vma); |
1466 | kleave(" = %d", ret); |
1467 | return ret; |
1468 | |
1469 | sharing_violation: |
1470 | up_write(&nommu_region_sem); |
1471 | printk(KERN_WARNING "Attempt to share mismatched mappings\n"); |
1472 | ret = -EINVAL; |
1473 | goto error; |
1474 | |
1475 | error_getting_vma: |
1476 | kmem_cache_free(vm_region_jar, region); |
1477 | printk(KERN_WARNING "Allocation of vma for %lu byte allocation" |
1478 | " from process %d failed\n", |
1479 | len, current->pid); |
1480 | show_free_areas(0); |
1481 | return -ENOMEM; |
1482 | |
1483 | error_getting_region: |
1484 | printk(KERN_WARNING "Allocation of vm region for %lu byte allocation" |
1485 | " from process %d failed\n", |
1486 | len, current->pid); |
1487 | show_free_areas(0); |
1488 | return -ENOMEM; |
1489 | } |
1490 | |
1491 | SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, |
1492 | unsigned long, prot, unsigned long, flags, |
1493 | unsigned long, fd, unsigned long, pgoff) |
1494 | { |
1495 | struct file *file = NULL; |
1496 | unsigned long retval = -EBADF; |
1497 | |
1498 | audit_mmap_fd(fd, flags); |
1499 | if (!(flags & MAP_ANONYMOUS)) { |
1500 | file = fget(fd); |
1501 | if (!file) |
1502 | goto out; |
1503 | } |
1504 | |
1505 | flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); |
1506 | |
1507 | retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); |
1508 | |
1509 | if (file) |
1510 | fput(file); |
1511 | out: |
1512 | return retval; |
1513 | } |
1514 | |
1515 | #ifdef __ARCH_WANT_SYS_OLD_MMAP |
1516 | struct mmap_arg_struct { |
1517 | unsigned long addr; |
1518 | unsigned long len; |
1519 | unsigned long prot; |
1520 | unsigned long flags; |
1521 | unsigned long fd; |
1522 | unsigned long offset; |
1523 | }; |
1524 | |
1525 | SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) |
1526 | { |
1527 | struct mmap_arg_struct a; |
1528 | |
1529 | if (copy_from_user(&a, arg, sizeof(a))) |
1530 | return -EFAULT; |
1531 | if (a.offset & ~PAGE_MASK) |
1532 | return -EINVAL; |
1533 | |
1534 | return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, |
1535 | a.offset >> PAGE_SHIFT); |
1536 | } |
1537 | #endif /* __ARCH_WANT_SYS_OLD_MMAP */ |
1538 | |
1539 | /* |
1540 | * split a vma into two pieces at address 'addr', a new vma is allocated either |
1541 | * for the first part or the tail. |
1542 | */ |
1543 | int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, |
1544 | unsigned long addr, int new_below) |
1545 | { |
1546 | struct vm_area_struct *new; |
1547 | struct vm_region *region; |
1548 | unsigned long npages; |
1549 | |
1550 | kenter(""); |
1551 | |
1552 | /* we're only permitted to split anonymous regions (these should have |
1553 | * only a single usage on the region) */ |
1554 | if (vma->vm_file) |
1555 | return -ENOMEM; |
1556 | |
1557 | if (mm->map_count >= sysctl_max_map_count) |
1558 | return -ENOMEM; |
1559 | |
1560 | region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL); |
1561 | if (!region) |
1562 | return -ENOMEM; |
1563 | |
1564 | new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); |
1565 | if (!new) { |
1566 | kmem_cache_free(vm_region_jar, region); |
1567 | return -ENOMEM; |
1568 | } |
1569 | |
1570 | /* most fields are the same, copy all, and then fixup */ |
1571 | *new = *vma; |
1572 | *region = *vma->vm_region; |
1573 | new->vm_region = region; |
1574 | |
1575 | npages = (addr - vma->vm_start) >> PAGE_SHIFT; |
1576 | |
1577 | if (new_below) { |
1578 | region->vm_top = region->vm_end = new->vm_end = addr; |
1579 | } else { |
1580 | region->vm_start = new->vm_start = addr; |
1581 | region->vm_pgoff = new->vm_pgoff += npages; |
1582 | } |
1583 | |
1584 | if (new->vm_ops && new->vm_ops->open) |
1585 | new->vm_ops->open(new); |
1586 | |
1587 | delete_vma_from_mm(vma); |
1588 | down_write(&nommu_region_sem); |
1589 | delete_nommu_region(vma->vm_region); |
1590 | if (new_below) { |
1591 | vma->vm_region->vm_start = vma->vm_start = addr; |
1592 | vma->vm_region->vm_pgoff = vma->vm_pgoff += npages; |
1593 | } else { |
1594 | vma->vm_region->vm_end = vma->vm_end = addr; |
1595 | vma->vm_region->vm_top = addr; |
1596 | } |
1597 | add_nommu_region(vma->vm_region); |
1598 | add_nommu_region(new->vm_region); |
1599 | up_write(&nommu_region_sem); |
1600 | add_vma_to_mm(mm, vma); |
1601 | add_vma_to_mm(mm, new); |
1602 | return 0; |
1603 | } |
1604 | |
1605 | /* |
1606 | * shrink a VMA by removing the specified chunk from either the beginning or |
1607 | * the end |
1608 | */ |
1609 | static int shrink_vma(struct mm_struct *mm, |
1610 | struct vm_area_struct *vma, |
1611 | unsigned long from, unsigned long to) |
1612 | { |
1613 | struct vm_region *region; |
1614 | |
1615 | kenter(""); |
1616 | |
1617 | /* adjust the VMA's pointers, which may reposition it in the MM's tree |
1618 | * and list */ |
1619 | delete_vma_from_mm(vma); |
1620 | if (from > vma->vm_start) |
1621 | vma->vm_end = from; |
1622 | else |
1623 | vma->vm_start = to; |
1624 | add_vma_to_mm(mm, vma); |
1625 | |
1626 | /* cut the backing region down to size */ |
1627 | region = vma->vm_region; |
1628 | BUG_ON(region->vm_usage != 1); |
1629 | |
1630 | down_write(&nommu_region_sem); |
1631 | delete_nommu_region(region); |
1632 | if (from > region->vm_start) { |
1633 | to = region->vm_top; |
1634 | region->vm_top = region->vm_end = from; |
1635 | } else { |
1636 | region->vm_start = to; |
1637 | } |
1638 | add_nommu_region(region); |
1639 | up_write(&nommu_region_sem); |
1640 | |
1641 | free_page_series(from, to); |
1642 | return 0; |
1643 | } |
1644 | |
1645 | /* |
1646 | * release a mapping |
1647 | * - under NOMMU conditions the chunk to be unmapped must be backed by a single |
1648 | * VMA, though it need not cover the whole VMA |
1649 | */ |
1650 | int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) |
1651 | { |
1652 | struct vm_area_struct *vma; |
1653 | unsigned long end; |
1654 | int ret; |
1655 | |
1656 | kenter(",%lx,%zx", start, len); |
1657 | |
1658 | len = PAGE_ALIGN(len); |
1659 | if (len == 0) |
1660 | return -EINVAL; |
1661 | |
1662 | end = start + len; |
1663 | |
1664 | /* find the first potentially overlapping VMA */ |
1665 | vma = find_vma(mm, start); |
1666 | if (!vma) { |
1667 | static int limit; |
1668 | if (limit < 5) { |
1669 | printk(KERN_WARNING |
1670 | "munmap of memory not mmapped by process %d" |
1671 | " (%s): 0x%lx-0x%lx\n", |
1672 | current->pid, current->comm, |
1673 | start, start + len - 1); |
1674 | limit++; |
1675 | } |
1676 | return -EINVAL; |
1677 | } |
1678 | |
1679 | /* we're allowed to split an anonymous VMA but not a file-backed one */ |
1680 | if (vma->vm_file) { |
1681 | do { |
1682 | if (start > vma->vm_start) { |
1683 | kleave(" = -EINVAL [miss]"); |
1684 | return -EINVAL; |
1685 | } |
1686 | if (end == vma->vm_end) |
1687 | goto erase_whole_vma; |
1688 | vma = vma->vm_next; |
1689 | } while (vma); |
1690 | kleave(" = -EINVAL [split file]"); |
1691 | return -EINVAL; |
1692 | } else { |
1693 | /* the chunk must be a subset of the VMA found */ |
1694 | if (start == vma->vm_start && end == vma->vm_end) |
1695 | goto erase_whole_vma; |
1696 | if (start < vma->vm_start || end > vma->vm_end) { |
1697 | kleave(" = -EINVAL [superset]"); |
1698 | return -EINVAL; |
1699 | } |
1700 | if (start & ~PAGE_MASK) { |
1701 | kleave(" = -EINVAL [unaligned start]"); |
1702 | return -EINVAL; |
1703 | } |
1704 | if (end != vma->vm_end && end & ~PAGE_MASK) { |
1705 | kleave(" = -EINVAL [unaligned split]"); |
1706 | return -EINVAL; |
1707 | } |
1708 | if (start != vma->vm_start && end != vma->vm_end) { |
1709 | ret = split_vma(mm, vma, start, 1); |
1710 | if (ret < 0) { |
1711 | kleave(" = %d [split]", ret); |
1712 | return ret; |
1713 | } |
1714 | } |
1715 | return shrink_vma(mm, vma, start, end); |
1716 | } |
1717 | |
1718 | erase_whole_vma: |
1719 | delete_vma_from_mm(vma); |
1720 | delete_vma(mm, vma); |
1721 | kleave(" = 0"); |
1722 | return 0; |
1723 | } |
1724 | EXPORT_SYMBOL(do_munmap); |
1725 | |
1726 | int vm_munmap(unsigned long addr, size_t len) |
1727 | { |
1728 | struct mm_struct *mm = current->mm; |
1729 | int ret; |
1730 | |
1731 | down_write(&mm->mmap_sem); |
1732 | ret = do_munmap(mm, addr, len); |
1733 | up_write(&mm->mmap_sem); |
1734 | return ret; |
1735 | } |
1736 | EXPORT_SYMBOL(vm_munmap); |
1737 | |
1738 | SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) |
1739 | { |
1740 | return vm_munmap(addr, len); |
1741 | } |
1742 | |
1743 | /* |
1744 | * release all the mappings made in a process's VM space |
1745 | */ |
1746 | void exit_mmap(struct mm_struct *mm) |
1747 | { |
1748 | struct vm_area_struct *vma; |
1749 | |
1750 | if (!mm) |
1751 | return; |
1752 | |
1753 | kenter(""); |
1754 | |
1755 | mm->total_vm = 0; |
1756 | |
1757 | while ((vma = mm->mmap)) { |
1758 | mm->mmap = vma->vm_next; |
1759 | delete_vma_from_mm(vma); |
1760 | delete_vma(mm, vma); |
1761 | cond_resched(); |
1762 | } |
1763 | |
1764 | kleave(""); |
1765 | } |
1766 | |
1767 | unsigned long vm_brk(unsigned long addr, unsigned long len) |
1768 | { |
1769 | return -ENOMEM; |
1770 | } |
1771 | |
1772 | /* |
1773 | * expand (or shrink) an existing mapping, potentially moving it at the same |
1774 | * time (controlled by the MREMAP_MAYMOVE flag and available VM space) |
1775 | * |
1776 | * under NOMMU conditions, we only permit changing a mapping's size, and only |
1777 | * as long as it stays within the region allocated by do_mmap_private() and the |
1778 | * block is not shareable |
1779 | * |
1780 | * MREMAP_FIXED is not supported under NOMMU conditions |
1781 | */ |
1782 | static unsigned long do_mremap(unsigned long addr, |
1783 | unsigned long old_len, unsigned long new_len, |
1784 | unsigned long flags, unsigned long new_addr) |
1785 | { |
1786 | struct vm_area_struct *vma; |
1787 | |
1788 | /* insanity checks first */ |
1789 | old_len = PAGE_ALIGN(old_len); |
1790 | new_len = PAGE_ALIGN(new_len); |
1791 | if (old_len == 0 || new_len == 0) |
1792 | return (unsigned long) -EINVAL; |
1793 | |
1794 | if (addr & ~PAGE_MASK) |
1795 | return -EINVAL; |
1796 | |
1797 | if (flags & MREMAP_FIXED && new_addr != addr) |
1798 | return (unsigned long) -EINVAL; |
1799 | |
1800 | vma = find_vma_exact(current->mm, addr, old_len); |
1801 | if (!vma) |
1802 | return (unsigned long) -EINVAL; |
1803 | |
1804 | if (vma->vm_end != vma->vm_start + old_len) |
1805 | return (unsigned long) -EFAULT; |
1806 | |
1807 | if (vma->vm_flags & VM_MAYSHARE) |
1808 | return (unsigned long) -EPERM; |
1809 | |
1810 | if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start) |
1811 | return (unsigned long) -ENOMEM; |
1812 | |
1813 | /* all checks complete - do it */ |
1814 | vma->vm_end = vma->vm_start + new_len; |
1815 | return vma->vm_start; |
1816 | } |
1817 | |
1818 | SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len, |
1819 | unsigned long, new_len, unsigned long, flags, |
1820 | unsigned long, new_addr) |
1821 | { |
1822 | unsigned long ret; |
1823 | |
1824 | down_write(¤t->mm->mmap_sem); |
1825 | ret = do_mremap(addr, old_len, new_len, flags, new_addr); |
1826 | up_write(¤t->mm->mmap_sem); |
1827 | return ret; |
1828 | } |
1829 | |
1830 | struct page *follow_page_mask(struct vm_area_struct *vma, |
1831 | unsigned long address, unsigned int flags, |
1832 | unsigned int *page_mask) |
1833 | { |
1834 | *page_mask = 0; |
1835 | return NULL; |
1836 | } |
1837 | |
1838 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, |
1839 | unsigned long pfn, unsigned long size, pgprot_t prot) |
1840 | { |
1841 | if (addr != (pfn << PAGE_SHIFT)) |
1842 | return -EINVAL; |
1843 | |
1844 | vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; |
1845 | return 0; |
1846 | } |
1847 | EXPORT_SYMBOL(remap_pfn_range); |
1848 | |
1849 | int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) |
1850 | { |
1851 | unsigned long pfn = start >> PAGE_SHIFT; |
1852 | unsigned long vm_len = vma->vm_end - vma->vm_start; |
1853 | |
1854 | pfn += vma->vm_pgoff; |
1855 | return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); |
1856 | } |
1857 | EXPORT_SYMBOL(vm_iomap_memory); |
1858 | |
1859 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, |
1860 | unsigned long pgoff) |
1861 | { |
1862 | unsigned int size = vma->vm_end - vma->vm_start; |
1863 | |
1864 | if (!(vma->vm_flags & VM_USERMAP)) |
1865 | return -EINVAL; |
1866 | |
1867 | vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT)); |
1868 | vma->vm_end = vma->vm_start + size; |
1869 | |
1870 | return 0; |
1871 | } |
1872 | EXPORT_SYMBOL(remap_vmalloc_range); |
1873 | |
1874 | unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr, |
1875 | unsigned long len, unsigned long pgoff, unsigned long flags) |
1876 | { |
1877 | return -ENOMEM; |
1878 | } |
1879 | |
1880 | void unmap_mapping_range(struct address_space *mapping, |
1881 | loff_t const holebegin, loff_t const holelen, |
1882 | int even_cows) |
1883 | { |
1884 | } |
1885 | EXPORT_SYMBOL(unmap_mapping_range); |
1886 | |
1887 | /* |
1888 | * Check that a process has enough memory to allocate a new virtual |
1889 | * mapping. 0 means there is enough memory for the allocation to |
1890 | * succeed and -ENOMEM implies there is not. |
1891 | * |
1892 | * We currently support three overcommit policies, which are set via the |
1893 | * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting |
1894 | * |
1895 | * Strict overcommit modes added 2002 Feb 26 by Alan Cox. |
1896 | * Additional code 2002 Jul 20 by Robert Love. |
1897 | * |
1898 | * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. |
1899 | * |
1900 | * Note this is a helper function intended to be used by LSMs which |
1901 | * wish to use this logic. |
1902 | */ |
1903 | int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) |
1904 | { |
1905 | unsigned long free, allowed, reserve; |
1906 | |
1907 | vm_acct_memory(pages); |
1908 | |
1909 | /* |
1910 | * Sometimes we want to use more memory than we have |
1911 | */ |
1912 | if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) |
1913 | return 0; |
1914 | |
1915 | if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { |
1916 | free = global_page_state(NR_FREE_PAGES); |
1917 | free += global_page_state(NR_FILE_PAGES); |
1918 | |
1919 | /* |
1920 | * shmem pages shouldn't be counted as free in this |
1921 | * case, they can't be purged, only swapped out, and |
1922 | * that won't affect the overall amount of available |
1923 | * memory in the system. |
1924 | */ |
1925 | free -= global_page_state(NR_SHMEM); |
1926 | |
1927 | free += get_nr_swap_pages(); |
1928 | |
1929 | /* |
1930 | * Any slabs which are created with the |
1931 | * SLAB_RECLAIM_ACCOUNT flag claim to have contents |
1932 | * which are reclaimable, under pressure. The dentry |
1933 | * cache and most inode caches should fall into this |
1934 | */ |
1935 | free += global_page_state(NR_SLAB_RECLAIMABLE); |
1936 | |
1937 | /* |
1938 | * Leave reserved pages. The pages are not for anonymous pages. |
1939 | */ |
1940 | if (free <= totalreserve_pages) |
1941 | goto error; |
1942 | else |
1943 | free -= totalreserve_pages; |
1944 | |
1945 | /* |
1946 | * Reserve some for root |
1947 | */ |
1948 | if (!cap_sys_admin) |
1949 | free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); |
1950 | |
1951 | if (free > pages) |
1952 | return 0; |
1953 | |
1954 | goto error; |
1955 | } |
1956 | |
1957 | allowed = vm_commit_limit(); |
1958 | /* |
1959 | * Reserve some 3% for root |
1960 | */ |
1961 | if (!cap_sys_admin) |
1962 | allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); |
1963 | |
1964 | /* |
1965 | * Don't let a single process grow so big a user can't recover |
1966 | */ |
1967 | if (mm) { |
1968 | reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10); |
1969 | allowed -= min(mm->total_vm / 32, reserve); |
1970 | } |
1971 | |
1972 | if (percpu_counter_read_positive(&vm_committed_as) < allowed) |
1973 | return 0; |
1974 | |
1975 | error: |
1976 | vm_unacct_memory(pages); |
1977 | |
1978 | return -ENOMEM; |
1979 | } |
1980 | |
1981 | int in_gate_area_no_mm(unsigned long addr) |
1982 | { |
1983 | return 0; |
1984 | } |
1985 | |
1986 | int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1987 | { |
1988 | BUG(); |
1989 | return 0; |
1990 | } |
1991 | EXPORT_SYMBOL(filemap_fault); |
1992 | |
1993 | void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf) |
1994 | { |
1995 | BUG(); |
1996 | } |
1997 | EXPORT_SYMBOL(filemap_map_pages); |
1998 | |
1999 | int generic_file_remap_pages(struct vm_area_struct *vma, unsigned long addr, |
2000 | unsigned long size, pgoff_t pgoff) |
2001 | { |
2002 | BUG(); |
2003 | return 0; |
2004 | } |
2005 | EXPORT_SYMBOL(generic_file_remap_pages); |
2006 | |
2007 | static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, |
2008 | unsigned long addr, void *buf, int len, int write) |
2009 | { |
2010 | struct vm_area_struct *vma; |
2011 | |
2012 | down_read(&mm->mmap_sem); |
2013 | |
2014 | /* the access must start within one of the target process's mappings */ |
2015 | vma = find_vma(mm, addr); |
2016 | if (vma) { |
2017 | /* don't overrun this mapping */ |
2018 | if (addr + len >= vma->vm_end) |
2019 | len = vma->vm_end - addr; |
2020 | |
2021 | /* only read or write mappings where it is permitted */ |
2022 | if (write && vma->vm_flags & VM_MAYWRITE) |
2023 | copy_to_user_page(vma, NULL, addr, |
2024 | (void *) addr, buf, len); |
2025 | else if (!write && vma->vm_flags & VM_MAYREAD) |
2026 | copy_from_user_page(vma, NULL, addr, |
2027 | buf, (void *) addr, len); |
2028 | else |
2029 | len = 0; |
2030 | } else { |
2031 | len = 0; |
2032 | } |
2033 | |
2034 | up_read(&mm->mmap_sem); |
2035 | |
2036 | return len; |
2037 | } |
2038 | |
2039 | /** |
2040 | * @access_remote_vm - access another process' address space |
2041 | * @mm: the mm_struct of the target address space |
2042 | * @addr: start address to access |
2043 | * @buf: source or destination buffer |
2044 | * @len: number of bytes to transfer |
2045 | * @write: whether the access is a write |
2046 | * |
2047 | * The caller must hold a reference on @mm. |
2048 | */ |
2049 | int access_remote_vm(struct mm_struct *mm, unsigned long addr, |
2050 | void *buf, int len, int write) |
2051 | { |
2052 | return __access_remote_vm(NULL, mm, addr, buf, len, write); |
2053 | } |
2054 | |
2055 | /* |
2056 | * Access another process' address space. |
2057 | * - source/target buffer must be kernel space |
2058 | */ |
2059 | int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write) |
2060 | { |
2061 | struct mm_struct *mm; |
2062 | |
2063 | if (addr + len < addr) |
2064 | return 0; |
2065 | |
2066 | mm = get_task_mm(tsk); |
2067 | if (!mm) |
2068 | return 0; |
2069 | |
2070 | len = __access_remote_vm(tsk, mm, addr, buf, len, write); |
2071 | |
2072 | mmput(mm); |
2073 | return len; |
2074 | } |
2075 | |
2076 | /** |
2077 | * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode |
2078 | * @inode: The inode to check |
2079 | * @size: The current filesize of the inode |
2080 | * @newsize: The proposed filesize of the inode |
2081 | * |
2082 | * Check the shared mappings on an inode on behalf of a shrinking truncate to |
2083 | * make sure that that any outstanding VMAs aren't broken and then shrink the |
2084 | * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't |
2085 | * automatically grant mappings that are too large. |
2086 | */ |
2087 | int nommu_shrink_inode_mappings(struct inode *inode, size_t size, |
2088 | size_t newsize) |
2089 | { |
2090 | struct vm_area_struct *vma; |
2091 | struct vm_region *region; |
2092 | pgoff_t low, high; |
2093 | size_t r_size, r_top; |
2094 | |
2095 | low = newsize >> PAGE_SHIFT; |
2096 | high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; |
2097 | |
2098 | down_write(&nommu_region_sem); |
2099 | mutex_lock(&inode->i_mapping->i_mmap_mutex); |
2100 | |
2101 | /* search for VMAs that fall within the dead zone */ |
2102 | vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) { |
2103 | /* found one - only interested if it's shared out of the page |
2104 | * cache */ |
2105 | if (vma->vm_flags & VM_SHARED) { |
2106 | mutex_unlock(&inode->i_mapping->i_mmap_mutex); |
2107 | up_write(&nommu_region_sem); |
2108 | return -ETXTBSY; /* not quite true, but near enough */ |
2109 | } |
2110 | } |
2111 | |
2112 | /* reduce any regions that overlap the dead zone - if in existence, |
2113 | * these will be pointed to by VMAs that don't overlap the dead zone |
2114 | * |
2115 | * we don't check for any regions that start beyond the EOF as there |
2116 | * shouldn't be any |
2117 | */ |
2118 | vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, |
2119 | 0, ULONG_MAX) { |
2120 | if (!(vma->vm_flags & VM_SHARED)) |
2121 | continue; |
2122 | |
2123 | region = vma->vm_region; |
2124 | r_size = region->vm_top - region->vm_start; |
2125 | r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size; |
2126 | |
2127 | if (r_top > newsize) { |
2128 | region->vm_top -= r_top - newsize; |
2129 | if (region->vm_end > region->vm_top) |
2130 | region->vm_end = region->vm_top; |
2131 | } |
2132 | } |
2133 | |
2134 | mutex_unlock(&inode->i_mapping->i_mmap_mutex); |
2135 | up_write(&nommu_region_sem); |
2136 | return 0; |
2137 | } |
2138 | |
2139 | /* |
2140 | * Initialise sysctl_user_reserve_kbytes. |
2141 | * |
2142 | * This is intended to prevent a user from starting a single memory hogging |
2143 | * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER |
2144 | * mode. |
2145 | * |
2146 | * The default value is min(3% of free memory, 128MB) |
2147 | * 128MB is enough to recover with sshd/login, bash, and top/kill. |
2148 | */ |
2149 | static int __meminit init_user_reserve(void) |
2150 | { |
2151 | unsigned long free_kbytes; |
2152 | |
2153 | free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); |
2154 | |
2155 | sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); |
2156 | return 0; |
2157 | } |
2158 | module_init(init_user_reserve) |
2159 | |
2160 | /* |
2161 | * Initialise sysctl_admin_reserve_kbytes. |
2162 | * |
2163 | * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin |
2164 | * to log in and kill a memory hogging process. |
2165 | * |
2166 | * Systems with more than 256MB will reserve 8MB, enough to recover |
2167 | * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will |
2168 | * only reserve 3% of free pages by default. |
2169 | */ |
2170 | static int __meminit init_admin_reserve(void) |
2171 | { |
2172 | unsigned long free_kbytes; |
2173 | |
2174 | free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); |
2175 | |
2176 | sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); |
2177 | return 0; |
2178 | } |
2179 | module_init(init_admin_reserve) |
2180 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9