Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | * | |
8feae131 | 9 | * Copyright (c) 2004-2008 David Howells <dhowells@redhat.com> |
1da177e4 LT |
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> | |
eb6434d9 | 13 | * Copyright (c) 2007-2009 Paul Mundt <lethal@linux-sh.org> |
1da177e4 LT |
14 | */ |
15 | ||
f2b8544f | 16 | #include <linux/module.h> |
1da177e4 LT |
17 | #include <linux/mm.h> |
18 | #include <linux/mman.h> | |
19 | #include <linux/swap.h> | |
20 | #include <linux/file.h> | |
21 | #include <linux/highmem.h> | |
22 | #include <linux/pagemap.h> | |
23 | #include <linux/slab.h> | |
24 | #include <linux/vmalloc.h> | |
fa8e26cc | 25 | #include <linux/tracehook.h> |
1da177e4 LT |
26 | #include <linux/blkdev.h> |
27 | #include <linux/backing-dev.h> | |
28 | #include <linux/mount.h> | |
29 | #include <linux/personality.h> | |
30 | #include <linux/security.h> | |
31 | #include <linux/syscalls.h> | |
32 | ||
33 | #include <asm/uaccess.h> | |
34 | #include <asm/tlb.h> | |
35 | #include <asm/tlbflush.h> | |
8feae131 DH |
36 | #include "internal.h" |
37 | ||
38 | static inline __attribute__((format(printf, 1, 2))) | |
39 | void no_printk(const char *fmt, ...) | |
40 | { | |
41 | } | |
42 | ||
43 | #if 0 | |
44 | #define kenter(FMT, ...) \ | |
45 | printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__) | |
46 | #define kleave(FMT, ...) \ | |
47 | printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__) | |
48 | #define kdebug(FMT, ...) \ | |
49 | printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__) | |
50 | #else | |
51 | #define kenter(FMT, ...) \ | |
52 | no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__) | |
53 | #define kleave(FMT, ...) \ | |
54 | no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__) | |
55 | #define kdebug(FMT, ...) \ | |
56 | no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__) | |
57 | #endif | |
1da177e4 | 58 | |
b291f000 NP |
59 | #include "internal.h" |
60 | ||
1da177e4 LT |
61 | void *high_memory; |
62 | struct page *mem_map; | |
63 | unsigned long max_mapnr; | |
64 | unsigned long num_physpages; | |
00a62ce9 | 65 | struct percpu_counter vm_committed_as; |
1da177e4 LT |
66 | int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */ |
67 | int sysctl_overcommit_ratio = 50; /* default is 50% */ | |
68 | int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT; | |
fc4d5c29 | 69 | int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS; |
1da177e4 LT |
70 | int heap_stack_gap = 0; |
71 | ||
33e5d769 | 72 | atomic_long_t mmap_pages_allocated; |
8feae131 | 73 | |
1da177e4 | 74 | EXPORT_SYMBOL(mem_map); |
6a04de6d | 75 | EXPORT_SYMBOL(num_physpages); |
1da177e4 | 76 | |
8feae131 DH |
77 | /* list of mapped, potentially shareable regions */ |
78 | static struct kmem_cache *vm_region_jar; | |
79 | struct rb_root nommu_region_tree = RB_ROOT; | |
80 | DECLARE_RWSEM(nommu_region_sem); | |
1da177e4 LT |
81 | |
82 | struct vm_operations_struct generic_file_vm_ops = { | |
83 | }; | |
84 | ||
85 | /* | |
86 | * Handle all mappings that got truncated by a "truncate()" | |
87 | * system call. | |
88 | * | |
89 | * NOTE! We have to be ready to update the memory sharing | |
90 | * between the file and the memory map for a potential last | |
91 | * incomplete page. Ugly, but necessary. | |
92 | */ | |
93 | int vmtruncate(struct inode *inode, loff_t offset) | |
94 | { | |
95 | struct address_space *mapping = inode->i_mapping; | |
96 | unsigned long limit; | |
97 | ||
98 | if (inode->i_size < offset) | |
99 | goto do_expand; | |
100 | i_size_write(inode, offset); | |
101 | ||
102 | truncate_inode_pages(mapping, offset); | |
103 | goto out_truncate; | |
104 | ||
105 | do_expand: | |
106 | limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur; | |
107 | if (limit != RLIM_INFINITY && offset > limit) | |
108 | goto out_sig; | |
109 | if (offset > inode->i_sb->s_maxbytes) | |
110 | goto out; | |
111 | i_size_write(inode, offset); | |
112 | ||
113 | out_truncate: | |
acfa4380 | 114 | if (inode->i_op->truncate) |
1da177e4 LT |
115 | inode->i_op->truncate(inode); |
116 | return 0; | |
117 | out_sig: | |
118 | send_sig(SIGXFSZ, current, 0); | |
119 | out: | |
120 | return -EFBIG; | |
121 | } | |
122 | ||
123 | EXPORT_SYMBOL(vmtruncate); | |
124 | ||
125 | /* | |
126 | * Return the total memory allocated for this pointer, not | |
127 | * just what the caller asked for. | |
128 | * | |
129 | * Doesn't have to be accurate, i.e. may have races. | |
130 | */ | |
131 | unsigned int kobjsize(const void *objp) | |
132 | { | |
133 | struct page *page; | |
134 | ||
4016a139 MH |
135 | /* |
136 | * If the object we have should not have ksize performed on it, | |
137 | * return size of 0 | |
138 | */ | |
5a1603be | 139 | if (!objp || !virt_addr_valid(objp)) |
6cfd53fc PM |
140 | return 0; |
141 | ||
142 | page = virt_to_head_page(objp); | |
6cfd53fc PM |
143 | |
144 | /* | |
145 | * If the allocator sets PageSlab, we know the pointer came from | |
146 | * kmalloc(). | |
147 | */ | |
1da177e4 LT |
148 | if (PageSlab(page)) |
149 | return ksize(objp); | |
150 | ||
ab2e83ea PM |
151 | /* |
152 | * If it's not a compound page, see if we have a matching VMA | |
153 | * region. This test is intentionally done in reverse order, | |
154 | * so if there's no VMA, we still fall through and hand back | |
155 | * PAGE_SIZE for 0-order pages. | |
156 | */ | |
157 | if (!PageCompound(page)) { | |
158 | struct vm_area_struct *vma; | |
159 | ||
160 | vma = find_vma(current->mm, (unsigned long)objp); | |
161 | if (vma) | |
162 | return vma->vm_end - vma->vm_start; | |
163 | } | |
164 | ||
6cfd53fc PM |
165 | /* |
166 | * The ksize() function is only guaranteed to work for pointers | |
5a1603be | 167 | * returned by kmalloc(). So handle arbitrary pointers here. |
6cfd53fc | 168 | */ |
5a1603be | 169 | return PAGE_SIZE << compound_order(page); |
1da177e4 LT |
170 | } |
171 | ||
b291f000 | 172 | int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm, |
9d73777e PZ |
173 | unsigned long start, int nr_pages, int flags, |
174 | struct page **pages, struct vm_area_struct **vmas) | |
1da177e4 | 175 | { |
910e46da | 176 | struct vm_area_struct *vma; |
7b4d5b8b DH |
177 | unsigned long vm_flags; |
178 | int i; | |
b291f000 NP |
179 | int write = !!(flags & GUP_FLAGS_WRITE); |
180 | int force = !!(flags & GUP_FLAGS_FORCE); | |
181 | int ignore = !!(flags & GUP_FLAGS_IGNORE_VMA_PERMISSIONS); | |
7b4d5b8b DH |
182 | |
183 | /* calculate required read or write permissions. | |
184 | * - if 'force' is set, we only require the "MAY" flags. | |
185 | */ | |
186 | vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); | |
187 | vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); | |
1da177e4 | 188 | |
9d73777e | 189 | for (i = 0; i < nr_pages; i++) { |
910e46da | 190 | vma = find_vma(mm, start); |
7b4d5b8b DH |
191 | if (!vma) |
192 | goto finish_or_fault; | |
193 | ||
194 | /* protect what we can, including chardevs */ | |
195 | if (vma->vm_flags & (VM_IO | VM_PFNMAP) || | |
b291f000 | 196 | (!ignore && !(vm_flags & vma->vm_flags))) |
7b4d5b8b | 197 | goto finish_or_fault; |
910e46da | 198 | |
1da177e4 LT |
199 | if (pages) { |
200 | pages[i] = virt_to_page(start); | |
201 | if (pages[i]) | |
202 | page_cache_get(pages[i]); | |
203 | } | |
204 | if (vmas) | |
910e46da | 205 | vmas[i] = vma; |
1da177e4 LT |
206 | start += PAGE_SIZE; |
207 | } | |
7b4d5b8b DH |
208 | |
209 | return i; | |
210 | ||
211 | finish_or_fault: | |
212 | return i ? : -EFAULT; | |
1da177e4 | 213 | } |
b291f000 NP |
214 | |
215 | ||
216 | /* | |
217 | * get a list of pages in an address range belonging to the specified process | |
218 | * and indicate the VMA that covers each page | |
219 | * - this is potentially dodgy as we may end incrementing the page count of a | |
220 | * slab page or a secondary page from a compound page | |
221 | * - don't permit access to VMAs that don't support it, such as I/O mappings | |
222 | */ | |
223 | int get_user_pages(struct task_struct *tsk, struct mm_struct *mm, | |
9d73777e | 224 | unsigned long start, int nr_pages, int write, int force, |
b291f000 NP |
225 | struct page **pages, struct vm_area_struct **vmas) |
226 | { | |
227 | int flags = 0; | |
228 | ||
229 | if (write) | |
230 | flags |= GUP_FLAGS_WRITE; | |
231 | if (force) | |
232 | flags |= GUP_FLAGS_FORCE; | |
233 | ||
9d73777e | 234 | return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas); |
b291f000 | 235 | } |
66aa2b4b GU |
236 | EXPORT_SYMBOL(get_user_pages); |
237 | ||
dfc2f91a PM |
238 | /** |
239 | * follow_pfn - look up PFN at a user virtual address | |
240 | * @vma: memory mapping | |
241 | * @address: user virtual address | |
242 | * @pfn: location to store found PFN | |
243 | * | |
244 | * Only IO mappings and raw PFN mappings are allowed. | |
245 | * | |
246 | * Returns zero and the pfn at @pfn on success, -ve otherwise. | |
247 | */ | |
248 | int follow_pfn(struct vm_area_struct *vma, unsigned long address, | |
249 | unsigned long *pfn) | |
250 | { | |
251 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) | |
252 | return -EINVAL; | |
253 | ||
254 | *pfn = address >> PAGE_SHIFT; | |
255 | return 0; | |
256 | } | |
257 | EXPORT_SYMBOL(follow_pfn); | |
258 | ||
1da177e4 LT |
259 | DEFINE_RWLOCK(vmlist_lock); |
260 | struct vm_struct *vmlist; | |
261 | ||
b3bdda02 | 262 | void vfree(const void *addr) |
1da177e4 LT |
263 | { |
264 | kfree(addr); | |
265 | } | |
b5073173 | 266 | EXPORT_SYMBOL(vfree); |
1da177e4 | 267 | |
dd0fc66f | 268 | void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) |
1da177e4 LT |
269 | { |
270 | /* | |
8518609d RD |
271 | * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc() |
272 | * returns only a logical address. | |
1da177e4 | 273 | */ |
84097518 | 274 | return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM); |
1da177e4 | 275 | } |
b5073173 | 276 | EXPORT_SYMBOL(__vmalloc); |
1da177e4 | 277 | |
f905bc44 PM |
278 | void *vmalloc_user(unsigned long size) |
279 | { | |
280 | void *ret; | |
281 | ||
282 | ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, | |
283 | PAGE_KERNEL); | |
284 | if (ret) { | |
285 | struct vm_area_struct *vma; | |
286 | ||
287 | down_write(¤t->mm->mmap_sem); | |
288 | vma = find_vma(current->mm, (unsigned long)ret); | |
289 | if (vma) | |
290 | vma->vm_flags |= VM_USERMAP; | |
291 | up_write(¤t->mm->mmap_sem); | |
292 | } | |
293 | ||
294 | return ret; | |
295 | } | |
296 | EXPORT_SYMBOL(vmalloc_user); | |
297 | ||
b3bdda02 | 298 | struct page *vmalloc_to_page(const void *addr) |
1da177e4 LT |
299 | { |
300 | return virt_to_page(addr); | |
301 | } | |
b5073173 | 302 | EXPORT_SYMBOL(vmalloc_to_page); |
1da177e4 | 303 | |
b3bdda02 | 304 | unsigned long vmalloc_to_pfn(const void *addr) |
1da177e4 LT |
305 | { |
306 | return page_to_pfn(virt_to_page(addr)); | |
307 | } | |
b5073173 | 308 | EXPORT_SYMBOL(vmalloc_to_pfn); |
1da177e4 LT |
309 | |
310 | long vread(char *buf, char *addr, unsigned long count) | |
311 | { | |
312 | memcpy(buf, addr, count); | |
313 | return count; | |
314 | } | |
315 | ||
316 | long vwrite(char *buf, char *addr, unsigned long count) | |
317 | { | |
318 | /* Don't allow overflow */ | |
319 | if ((unsigned long) addr + count < count) | |
320 | count = -(unsigned long) addr; | |
321 | ||
322 | memcpy(addr, buf, count); | |
323 | return(count); | |
324 | } | |
325 | ||
326 | /* | |
327 | * vmalloc - allocate virtually continguos memory | |
328 | * | |
329 | * @size: allocation size | |
330 | * | |
331 | * Allocate enough pages to cover @size from the page level | |
332 | * allocator and map them into continguos kernel virtual space. | |
333 | * | |
c1c8897f | 334 | * For tight control over page level allocator and protection flags |
1da177e4 LT |
335 | * use __vmalloc() instead. |
336 | */ | |
337 | void *vmalloc(unsigned long size) | |
338 | { | |
339 | return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL); | |
340 | } | |
f6138882 AM |
341 | EXPORT_SYMBOL(vmalloc); |
342 | ||
343 | void *vmalloc_node(unsigned long size, int node) | |
344 | { | |
345 | return vmalloc(size); | |
346 | } | |
347 | EXPORT_SYMBOL(vmalloc_node); | |
1da177e4 | 348 | |
1af446ed PM |
349 | #ifndef PAGE_KERNEL_EXEC |
350 | # define PAGE_KERNEL_EXEC PAGE_KERNEL | |
351 | #endif | |
352 | ||
353 | /** | |
354 | * vmalloc_exec - allocate virtually contiguous, executable memory | |
355 | * @size: allocation size | |
356 | * | |
357 | * Kernel-internal function to allocate enough pages to cover @size | |
358 | * the page level allocator and map them into contiguous and | |
359 | * executable kernel virtual space. | |
360 | * | |
361 | * For tight control over page level allocator and protection flags | |
362 | * use __vmalloc() instead. | |
363 | */ | |
364 | ||
365 | void *vmalloc_exec(unsigned long size) | |
366 | { | |
367 | return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC); | |
368 | } | |
369 | ||
b5073173 PM |
370 | /** |
371 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) | |
1da177e4 LT |
372 | * @size: allocation size |
373 | * | |
374 | * Allocate enough 32bit PA addressable pages to cover @size from the | |
375 | * page level allocator and map them into continguos kernel virtual space. | |
376 | */ | |
377 | void *vmalloc_32(unsigned long size) | |
378 | { | |
379 | return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL); | |
380 | } | |
b5073173 PM |
381 | EXPORT_SYMBOL(vmalloc_32); |
382 | ||
383 | /** | |
384 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory | |
385 | * @size: allocation size | |
386 | * | |
387 | * The resulting memory area is 32bit addressable and zeroed so it can be | |
388 | * mapped to userspace without leaking data. | |
f905bc44 PM |
389 | * |
390 | * VM_USERMAP is set on the corresponding VMA so that subsequent calls to | |
391 | * remap_vmalloc_range() are permissible. | |
b5073173 PM |
392 | */ |
393 | void *vmalloc_32_user(unsigned long size) | |
394 | { | |
f905bc44 PM |
395 | /* |
396 | * We'll have to sort out the ZONE_DMA bits for 64-bit, | |
397 | * but for now this can simply use vmalloc_user() directly. | |
398 | */ | |
399 | return vmalloc_user(size); | |
b5073173 PM |
400 | } |
401 | EXPORT_SYMBOL(vmalloc_32_user); | |
1da177e4 LT |
402 | |
403 | void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot) | |
404 | { | |
405 | BUG(); | |
406 | return NULL; | |
407 | } | |
b5073173 | 408 | EXPORT_SYMBOL(vmap); |
1da177e4 | 409 | |
b3bdda02 | 410 | void vunmap(const void *addr) |
1da177e4 LT |
411 | { |
412 | BUG(); | |
413 | } | |
b5073173 | 414 | EXPORT_SYMBOL(vunmap); |
1da177e4 | 415 | |
eb6434d9 PM |
416 | void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) |
417 | { | |
418 | BUG(); | |
419 | return NULL; | |
420 | } | |
421 | EXPORT_SYMBOL(vm_map_ram); | |
422 | ||
423 | void vm_unmap_ram(const void *mem, unsigned int count) | |
424 | { | |
425 | BUG(); | |
426 | } | |
427 | EXPORT_SYMBOL(vm_unmap_ram); | |
428 | ||
429 | void vm_unmap_aliases(void) | |
430 | { | |
431 | } | |
432 | EXPORT_SYMBOL_GPL(vm_unmap_aliases); | |
433 | ||
1eeb66a1 CH |
434 | /* |
435 | * Implement a stub for vmalloc_sync_all() if the architecture chose not to | |
436 | * have one. | |
437 | */ | |
438 | void __attribute__((weak)) vmalloc_sync_all(void) | |
439 | { | |
440 | } | |
441 | ||
b5073173 PM |
442 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, |
443 | struct page *page) | |
444 | { | |
445 | return -EINVAL; | |
446 | } | |
447 | EXPORT_SYMBOL(vm_insert_page); | |
448 | ||
1da177e4 LT |
449 | /* |
450 | * sys_brk() for the most part doesn't need the global kernel | |
451 | * lock, except when an application is doing something nasty | |
452 | * like trying to un-brk an area that has already been mapped | |
453 | * to a regular file. in this case, the unmapping will need | |
454 | * to invoke file system routines that need the global lock. | |
455 | */ | |
6a6160a7 | 456 | SYSCALL_DEFINE1(brk, unsigned long, brk) |
1da177e4 LT |
457 | { |
458 | struct mm_struct *mm = current->mm; | |
459 | ||
460 | if (brk < mm->start_brk || brk > mm->context.end_brk) | |
461 | return mm->brk; | |
462 | ||
463 | if (mm->brk == brk) | |
464 | return mm->brk; | |
465 | ||
466 | /* | |
467 | * Always allow shrinking brk | |
468 | */ | |
469 | if (brk <= mm->brk) { | |
470 | mm->brk = brk; | |
471 | return brk; | |
472 | } | |
473 | ||
474 | /* | |
475 | * Ok, looks good - let it rip. | |
476 | */ | |
477 | return mm->brk = brk; | |
478 | } | |
479 | ||
8feae131 DH |
480 | /* |
481 | * initialise the VMA and region record slabs | |
482 | */ | |
483 | void __init mmap_init(void) | |
1da177e4 | 484 | { |
00a62ce9 KM |
485 | int ret; |
486 | ||
487 | ret = percpu_counter_init(&vm_committed_as, 0); | |
488 | VM_BUG_ON(ret); | |
33e5d769 | 489 | vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC); |
1da177e4 | 490 | } |
1da177e4 | 491 | |
3034097a | 492 | /* |
8feae131 DH |
493 | * validate the region tree |
494 | * - the caller must hold the region lock | |
3034097a | 495 | */ |
8feae131 DH |
496 | #ifdef CONFIG_DEBUG_NOMMU_REGIONS |
497 | static noinline void validate_nommu_regions(void) | |
3034097a | 498 | { |
8feae131 DH |
499 | struct vm_region *region, *last; |
500 | struct rb_node *p, *lastp; | |
3034097a | 501 | |
8feae131 DH |
502 | lastp = rb_first(&nommu_region_tree); |
503 | if (!lastp) | |
504 | return; | |
505 | ||
506 | last = rb_entry(lastp, struct vm_region, vm_rb); | |
33e5d769 DH |
507 | BUG_ON(unlikely(last->vm_end <= last->vm_start)); |
508 | BUG_ON(unlikely(last->vm_top < last->vm_end)); | |
8feae131 DH |
509 | |
510 | while ((p = rb_next(lastp))) { | |
511 | region = rb_entry(p, struct vm_region, vm_rb); | |
512 | last = rb_entry(lastp, struct vm_region, vm_rb); | |
513 | ||
33e5d769 DH |
514 | BUG_ON(unlikely(region->vm_end <= region->vm_start)); |
515 | BUG_ON(unlikely(region->vm_top < region->vm_end)); | |
516 | BUG_ON(unlikely(region->vm_start < last->vm_top)); | |
3034097a | 517 | |
8feae131 DH |
518 | lastp = p; |
519 | } | |
3034097a | 520 | } |
8feae131 | 521 | #else |
33e5d769 DH |
522 | static void validate_nommu_regions(void) |
523 | { | |
524 | } | |
8feae131 | 525 | #endif |
3034097a DH |
526 | |
527 | /* | |
8feae131 | 528 | * add a region into the global tree |
3034097a | 529 | */ |
8feae131 | 530 | static void add_nommu_region(struct vm_region *region) |
3034097a | 531 | { |
8feae131 DH |
532 | struct vm_region *pregion; |
533 | struct rb_node **p, *parent; | |
3034097a | 534 | |
8feae131 DH |
535 | validate_nommu_regions(); |
536 | ||
8feae131 DH |
537 | parent = NULL; |
538 | p = &nommu_region_tree.rb_node; | |
539 | while (*p) { | |
540 | parent = *p; | |
541 | pregion = rb_entry(parent, struct vm_region, vm_rb); | |
542 | if (region->vm_start < pregion->vm_start) | |
543 | p = &(*p)->rb_left; | |
544 | else if (region->vm_start > pregion->vm_start) | |
545 | p = &(*p)->rb_right; | |
546 | else if (pregion == region) | |
547 | return; | |
548 | else | |
549 | BUG(); | |
3034097a DH |
550 | } |
551 | ||
8feae131 DH |
552 | rb_link_node(®ion->vm_rb, parent, p); |
553 | rb_insert_color(®ion->vm_rb, &nommu_region_tree); | |
3034097a | 554 | |
8feae131 | 555 | validate_nommu_regions(); |
3034097a | 556 | } |
3034097a | 557 | |
930e652a | 558 | /* |
8feae131 | 559 | * delete a region from the global tree |
930e652a | 560 | */ |
8feae131 | 561 | static void delete_nommu_region(struct vm_region *region) |
930e652a | 562 | { |
8feae131 | 563 | BUG_ON(!nommu_region_tree.rb_node); |
930e652a | 564 | |
8feae131 DH |
565 | validate_nommu_regions(); |
566 | rb_erase(®ion->vm_rb, &nommu_region_tree); | |
567 | validate_nommu_regions(); | |
57c8f63e GU |
568 | } |
569 | ||
6fa5f80b | 570 | /* |
8feae131 | 571 | * free a contiguous series of pages |
6fa5f80b | 572 | */ |
8feae131 | 573 | static void free_page_series(unsigned long from, unsigned long to) |
6fa5f80b | 574 | { |
8feae131 DH |
575 | for (; from < to; from += PAGE_SIZE) { |
576 | struct page *page = virt_to_page(from); | |
577 | ||
578 | kdebug("- free %lx", from); | |
33e5d769 | 579 | atomic_long_dec(&mmap_pages_allocated); |
8feae131 | 580 | if (page_count(page) != 1) |
33e5d769 DH |
581 | kdebug("free page %p: refcount not one: %d", |
582 | page, page_count(page)); | |
8feae131 | 583 | put_page(page); |
6fa5f80b | 584 | } |
6fa5f80b DH |
585 | } |
586 | ||
3034097a | 587 | /* |
8feae131 | 588 | * release a reference to a region |
33e5d769 | 589 | * - the caller must hold the region semaphore for writing, which this releases |
dd8632a1 | 590 | * - the region may not have been added to the tree yet, in which case vm_top |
8feae131 | 591 | * will equal vm_start |
3034097a | 592 | */ |
8feae131 DH |
593 | static void __put_nommu_region(struct vm_region *region) |
594 | __releases(nommu_region_sem) | |
1da177e4 | 595 | { |
8feae131 | 596 | kenter("%p{%d}", region, atomic_read(®ion->vm_usage)); |
1da177e4 | 597 | |
8feae131 | 598 | BUG_ON(!nommu_region_tree.rb_node); |
1da177e4 | 599 | |
8feae131 | 600 | if (atomic_dec_and_test(®ion->vm_usage)) { |
dd8632a1 | 601 | if (region->vm_top > region->vm_start) |
8feae131 DH |
602 | delete_nommu_region(region); |
603 | up_write(&nommu_region_sem); | |
604 | ||
605 | if (region->vm_file) | |
606 | fput(region->vm_file); | |
607 | ||
608 | /* IO memory and memory shared directly out of the pagecache | |
609 | * from ramfs/tmpfs mustn't be released here */ | |
610 | if (region->vm_flags & VM_MAPPED_COPY) { | |
611 | kdebug("free series"); | |
dd8632a1 | 612 | free_page_series(region->vm_start, region->vm_top); |
8feae131 DH |
613 | } |
614 | kmem_cache_free(vm_region_jar, region); | |
615 | } else { | |
616 | up_write(&nommu_region_sem); | |
1da177e4 | 617 | } |
8feae131 | 618 | } |
1da177e4 | 619 | |
8feae131 DH |
620 | /* |
621 | * release a reference to a region | |
622 | */ | |
623 | static void put_nommu_region(struct vm_region *region) | |
624 | { | |
625 | down_write(&nommu_region_sem); | |
626 | __put_nommu_region(region); | |
1da177e4 LT |
627 | } |
628 | ||
3034097a | 629 | /* |
8feae131 DH |
630 | * add a VMA into a process's mm_struct in the appropriate place in the list |
631 | * and tree and add to the address space's page tree also if not an anonymous | |
632 | * page | |
633 | * - should be called with mm->mmap_sem held writelocked | |
3034097a | 634 | */ |
8feae131 | 635 | static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma) |
1da177e4 | 636 | { |
8feae131 | 637 | struct vm_area_struct *pvma, **pp; |
1da177e4 | 638 | struct address_space *mapping; |
8feae131 DH |
639 | struct rb_node **p, *parent; |
640 | ||
641 | kenter(",%p", vma); | |
642 | ||
643 | BUG_ON(!vma->vm_region); | |
644 | ||
645 | mm->map_count++; | |
646 | vma->vm_mm = mm; | |
1da177e4 LT |
647 | |
648 | /* add the VMA to the mapping */ | |
649 | if (vma->vm_file) { | |
650 | mapping = vma->vm_file->f_mapping; | |
651 | ||
652 | flush_dcache_mmap_lock(mapping); | |
653 | vma_prio_tree_insert(vma, &mapping->i_mmap); | |
654 | flush_dcache_mmap_unlock(mapping); | |
655 | } | |
656 | ||
8feae131 DH |
657 | /* add the VMA to the tree */ |
658 | parent = NULL; | |
659 | p = &mm->mm_rb.rb_node; | |
1da177e4 LT |
660 | while (*p) { |
661 | parent = *p; | |
662 | pvma = rb_entry(parent, struct vm_area_struct, vm_rb); | |
663 | ||
8feae131 DH |
664 | /* sort by: start addr, end addr, VMA struct addr in that order |
665 | * (the latter is necessary as we may get identical VMAs) */ | |
666 | if (vma->vm_start < pvma->vm_start) | |
1da177e4 | 667 | p = &(*p)->rb_left; |
8feae131 | 668 | else if (vma->vm_start > pvma->vm_start) |
1da177e4 | 669 | p = &(*p)->rb_right; |
8feae131 DH |
670 | else if (vma->vm_end < pvma->vm_end) |
671 | p = &(*p)->rb_left; | |
672 | else if (vma->vm_end > pvma->vm_end) | |
673 | p = &(*p)->rb_right; | |
674 | else if (vma < pvma) | |
675 | p = &(*p)->rb_left; | |
676 | else if (vma > pvma) | |
677 | p = &(*p)->rb_right; | |
678 | else | |
679 | BUG(); | |
1da177e4 LT |
680 | } |
681 | ||
682 | rb_link_node(&vma->vm_rb, parent, p); | |
8feae131 DH |
683 | rb_insert_color(&vma->vm_rb, &mm->mm_rb); |
684 | ||
685 | /* add VMA to the VMA list also */ | |
686 | for (pp = &mm->mmap; (pvma = *pp); pp = &(*pp)->vm_next) { | |
687 | if (pvma->vm_start > vma->vm_start) | |
688 | break; | |
689 | if (pvma->vm_start < vma->vm_start) | |
690 | continue; | |
691 | if (pvma->vm_end < vma->vm_end) | |
692 | break; | |
693 | } | |
694 | ||
695 | vma->vm_next = *pp; | |
696 | *pp = vma; | |
1da177e4 LT |
697 | } |
698 | ||
3034097a | 699 | /* |
8feae131 | 700 | * delete a VMA from its owning mm_struct and address space |
3034097a | 701 | */ |
8feae131 | 702 | static void delete_vma_from_mm(struct vm_area_struct *vma) |
1da177e4 | 703 | { |
8feae131 | 704 | struct vm_area_struct **pp; |
1da177e4 | 705 | struct address_space *mapping; |
8feae131 DH |
706 | struct mm_struct *mm = vma->vm_mm; |
707 | ||
708 | kenter("%p", vma); | |
709 | ||
710 | mm->map_count--; | |
711 | if (mm->mmap_cache == vma) | |
712 | mm->mmap_cache = NULL; | |
1da177e4 LT |
713 | |
714 | /* remove the VMA from the mapping */ | |
715 | if (vma->vm_file) { | |
716 | mapping = vma->vm_file->f_mapping; | |
717 | ||
718 | flush_dcache_mmap_lock(mapping); | |
719 | vma_prio_tree_remove(vma, &mapping->i_mmap); | |
720 | flush_dcache_mmap_unlock(mapping); | |
721 | } | |
722 | ||
8feae131 DH |
723 | /* remove from the MM's tree and list */ |
724 | rb_erase(&vma->vm_rb, &mm->mm_rb); | |
725 | for (pp = &mm->mmap; *pp; pp = &(*pp)->vm_next) { | |
726 | if (*pp == vma) { | |
727 | *pp = vma->vm_next; | |
728 | break; | |
729 | } | |
730 | } | |
731 | ||
732 | vma->vm_mm = NULL; | |
733 | } | |
734 | ||
735 | /* | |
736 | * destroy a VMA record | |
737 | */ | |
738 | static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma) | |
739 | { | |
740 | kenter("%p", vma); | |
741 | if (vma->vm_ops && vma->vm_ops->close) | |
742 | vma->vm_ops->close(vma); | |
743 | if (vma->vm_file) { | |
744 | fput(vma->vm_file); | |
745 | if (vma->vm_flags & VM_EXECUTABLE) | |
746 | removed_exe_file_vma(mm); | |
747 | } | |
748 | put_nommu_region(vma->vm_region); | |
749 | kmem_cache_free(vm_area_cachep, vma); | |
750 | } | |
751 | ||
752 | /* | |
753 | * look up the first VMA in which addr resides, NULL if none | |
754 | * - should be called with mm->mmap_sem at least held readlocked | |
755 | */ | |
756 | struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) | |
757 | { | |
758 | struct vm_area_struct *vma; | |
759 | struct rb_node *n = mm->mm_rb.rb_node; | |
760 | ||
761 | /* check the cache first */ | |
762 | vma = mm->mmap_cache; | |
763 | if (vma && vma->vm_start <= addr && vma->vm_end > addr) | |
764 | return vma; | |
765 | ||
766 | /* trawl the tree (there may be multiple mappings in which addr | |
767 | * resides) */ | |
768 | for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) { | |
769 | vma = rb_entry(n, struct vm_area_struct, vm_rb); | |
770 | if (vma->vm_start > addr) | |
771 | return NULL; | |
772 | if (vma->vm_end > addr) { | |
773 | mm->mmap_cache = vma; | |
774 | return vma; | |
775 | } | |
776 | } | |
777 | ||
778 | return NULL; | |
779 | } | |
780 | EXPORT_SYMBOL(find_vma); | |
781 | ||
782 | /* | |
783 | * find a VMA | |
784 | * - we don't extend stack VMAs under NOMMU conditions | |
785 | */ | |
786 | struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr) | |
787 | { | |
788 | return find_vma(mm, addr); | |
789 | } | |
790 | ||
791 | /* | |
792 | * expand a stack to a given address | |
793 | * - not supported under NOMMU conditions | |
794 | */ | |
795 | int expand_stack(struct vm_area_struct *vma, unsigned long address) | |
796 | { | |
797 | return -ENOMEM; | |
798 | } | |
799 | ||
800 | /* | |
801 | * look up the first VMA exactly that exactly matches addr | |
802 | * - should be called with mm->mmap_sem at least held readlocked | |
803 | */ | |
804 | static struct vm_area_struct *find_vma_exact(struct mm_struct *mm, | |
805 | unsigned long addr, | |
806 | unsigned long len) | |
807 | { | |
808 | struct vm_area_struct *vma; | |
809 | struct rb_node *n = mm->mm_rb.rb_node; | |
810 | unsigned long end = addr + len; | |
811 | ||
812 | /* check the cache first */ | |
813 | vma = mm->mmap_cache; | |
814 | if (vma && vma->vm_start == addr && vma->vm_end == end) | |
815 | return vma; | |
816 | ||
817 | /* trawl the tree (there may be multiple mappings in which addr | |
818 | * resides) */ | |
819 | for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) { | |
820 | vma = rb_entry(n, struct vm_area_struct, vm_rb); | |
821 | if (vma->vm_start < addr) | |
822 | continue; | |
823 | if (vma->vm_start > addr) | |
824 | return NULL; | |
825 | if (vma->vm_end == end) { | |
826 | mm->mmap_cache = vma; | |
827 | return vma; | |
828 | } | |
829 | } | |
830 | ||
831 | return NULL; | |
1da177e4 LT |
832 | } |
833 | ||
834 | /* | |
835 | * determine whether a mapping should be permitted and, if so, what sort of | |
836 | * mapping we're capable of supporting | |
837 | */ | |
838 | static int validate_mmap_request(struct file *file, | |
839 | unsigned long addr, | |
840 | unsigned long len, | |
841 | unsigned long prot, | |
842 | unsigned long flags, | |
843 | unsigned long pgoff, | |
844 | unsigned long *_capabilities) | |
845 | { | |
8feae131 | 846 | unsigned long capabilities, rlen; |
1da177e4 LT |
847 | unsigned long reqprot = prot; |
848 | int ret; | |
849 | ||
850 | /* do the simple checks first */ | |
851 | if (flags & MAP_FIXED || addr) { | |
852 | printk(KERN_DEBUG | |
853 | "%d: Can't do fixed-address/overlay mmap of RAM\n", | |
854 | current->pid); | |
855 | return -EINVAL; | |
856 | } | |
857 | ||
858 | if ((flags & MAP_TYPE) != MAP_PRIVATE && | |
859 | (flags & MAP_TYPE) != MAP_SHARED) | |
860 | return -EINVAL; | |
861 | ||
f81cff0d | 862 | if (!len) |
1da177e4 LT |
863 | return -EINVAL; |
864 | ||
f81cff0d | 865 | /* Careful about overflows.. */ |
8feae131 DH |
866 | rlen = PAGE_ALIGN(len); |
867 | if (!rlen || rlen > TASK_SIZE) | |
f81cff0d MF |
868 | return -ENOMEM; |
869 | ||
1da177e4 | 870 | /* offset overflow? */ |
8feae131 | 871 | if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff) |
f81cff0d | 872 | return -EOVERFLOW; |
1da177e4 LT |
873 | |
874 | if (file) { | |
875 | /* validate file mapping requests */ | |
876 | struct address_space *mapping; | |
877 | ||
878 | /* files must support mmap */ | |
879 | if (!file->f_op || !file->f_op->mmap) | |
880 | return -ENODEV; | |
881 | ||
882 | /* work out if what we've got could possibly be shared | |
883 | * - we support chardevs that provide their own "memory" | |
884 | * - we support files/blockdevs that are memory backed | |
885 | */ | |
886 | mapping = file->f_mapping; | |
887 | if (!mapping) | |
e9536ae7 | 888 | mapping = file->f_path.dentry->d_inode->i_mapping; |
1da177e4 LT |
889 | |
890 | capabilities = 0; | |
891 | if (mapping && mapping->backing_dev_info) | |
892 | capabilities = mapping->backing_dev_info->capabilities; | |
893 | ||
894 | if (!capabilities) { | |
895 | /* no explicit capabilities set, so assume some | |
896 | * defaults */ | |
e9536ae7 | 897 | switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) { |
1da177e4 LT |
898 | case S_IFREG: |
899 | case S_IFBLK: | |
900 | capabilities = BDI_CAP_MAP_COPY; | |
901 | break; | |
902 | ||
903 | case S_IFCHR: | |
904 | capabilities = | |
905 | BDI_CAP_MAP_DIRECT | | |
906 | BDI_CAP_READ_MAP | | |
907 | BDI_CAP_WRITE_MAP; | |
908 | break; | |
909 | ||
910 | default: | |
911 | return -EINVAL; | |
912 | } | |
913 | } | |
914 | ||
915 | /* eliminate any capabilities that we can't support on this | |
916 | * device */ | |
917 | if (!file->f_op->get_unmapped_area) | |
918 | capabilities &= ~BDI_CAP_MAP_DIRECT; | |
919 | if (!file->f_op->read) | |
920 | capabilities &= ~BDI_CAP_MAP_COPY; | |
921 | ||
922 | if (flags & MAP_SHARED) { | |
923 | /* do checks for writing, appending and locking */ | |
924 | if ((prot & PROT_WRITE) && | |
925 | !(file->f_mode & FMODE_WRITE)) | |
926 | return -EACCES; | |
927 | ||
e9536ae7 | 928 | if (IS_APPEND(file->f_path.dentry->d_inode) && |
1da177e4 LT |
929 | (file->f_mode & FMODE_WRITE)) |
930 | return -EACCES; | |
931 | ||
e9536ae7 | 932 | if (locks_verify_locked(file->f_path.dentry->d_inode)) |
1da177e4 LT |
933 | return -EAGAIN; |
934 | ||
935 | if (!(capabilities & BDI_CAP_MAP_DIRECT)) | |
936 | return -ENODEV; | |
937 | ||
938 | if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) || | |
939 | ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) || | |
940 | ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP)) | |
941 | ) { | |
942 | printk("MAP_SHARED not completely supported on !MMU\n"); | |
943 | return -EINVAL; | |
944 | } | |
945 | ||
946 | /* we mustn't privatise shared mappings */ | |
947 | capabilities &= ~BDI_CAP_MAP_COPY; | |
948 | } | |
949 | else { | |
950 | /* we're going to read the file into private memory we | |
951 | * allocate */ | |
952 | if (!(capabilities & BDI_CAP_MAP_COPY)) | |
953 | return -ENODEV; | |
954 | ||
955 | /* we don't permit a private writable mapping to be | |
956 | * shared with the backing device */ | |
957 | if (prot & PROT_WRITE) | |
958 | capabilities &= ~BDI_CAP_MAP_DIRECT; | |
959 | } | |
960 | ||
961 | /* handle executable mappings and implied executable | |
962 | * mappings */ | |
e9536ae7 | 963 | if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) { |
1da177e4 LT |
964 | if (prot & PROT_EXEC) |
965 | return -EPERM; | |
966 | } | |
967 | else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) { | |
968 | /* handle implication of PROT_EXEC by PROT_READ */ | |
969 | if (current->personality & READ_IMPLIES_EXEC) { | |
970 | if (capabilities & BDI_CAP_EXEC_MAP) | |
971 | prot |= PROT_EXEC; | |
972 | } | |
973 | } | |
974 | else if ((prot & PROT_READ) && | |
975 | (prot & PROT_EXEC) && | |
976 | !(capabilities & BDI_CAP_EXEC_MAP) | |
977 | ) { | |
978 | /* backing file is not executable, try to copy */ | |
979 | capabilities &= ~BDI_CAP_MAP_DIRECT; | |
980 | } | |
981 | } | |
982 | else { | |
983 | /* anonymous mappings are always memory backed and can be | |
984 | * privately mapped | |
985 | */ | |
986 | capabilities = BDI_CAP_MAP_COPY; | |
987 | ||
988 | /* handle PROT_EXEC implication by PROT_READ */ | |
989 | if ((prot & PROT_READ) && | |
990 | (current->personality & READ_IMPLIES_EXEC)) | |
991 | prot |= PROT_EXEC; | |
992 | } | |
993 | ||
994 | /* allow the security API to have its say */ | |
ed032189 | 995 | ret = security_file_mmap(file, reqprot, prot, flags, addr, 0); |
1da177e4 LT |
996 | if (ret < 0) |
997 | return ret; | |
998 | ||
999 | /* looks okay */ | |
1000 | *_capabilities = capabilities; | |
1001 | return 0; | |
1002 | } | |
1003 | ||
1004 | /* | |
1005 | * we've determined that we can make the mapping, now translate what we | |
1006 | * now know into VMA flags | |
1007 | */ | |
1008 | static unsigned long determine_vm_flags(struct file *file, | |
1009 | unsigned long prot, | |
1010 | unsigned long flags, | |
1011 | unsigned long capabilities) | |
1012 | { | |
1013 | unsigned long vm_flags; | |
1014 | ||
1015 | vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags); | |
1016 | vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; | |
1017 | /* vm_flags |= mm->def_flags; */ | |
1018 | ||
1019 | if (!(capabilities & BDI_CAP_MAP_DIRECT)) { | |
1020 | /* attempt to share read-only copies of mapped file chunks */ | |
1021 | if (file && !(prot & PROT_WRITE)) | |
1022 | vm_flags |= VM_MAYSHARE; | |
1023 | } | |
1024 | else { | |
1025 | /* overlay a shareable mapping on the backing device or inode | |
1026 | * if possible - used for chardevs, ramfs/tmpfs/shmfs and | |
1027 | * romfs/cramfs */ | |
1028 | if (flags & MAP_SHARED) | |
1029 | vm_flags |= VM_MAYSHARE | VM_SHARED; | |
1030 | else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0) | |
1031 | vm_flags |= VM_MAYSHARE; | |
1032 | } | |
1033 | ||
1034 | /* refuse to let anyone share private mappings with this process if | |
1035 | * it's being traced - otherwise breakpoints set in it may interfere | |
1036 | * with another untraced process | |
1037 | */ | |
fa8e26cc | 1038 | if ((flags & MAP_PRIVATE) && tracehook_expect_breakpoints(current)) |
1da177e4 LT |
1039 | vm_flags &= ~VM_MAYSHARE; |
1040 | ||
1041 | return vm_flags; | |
1042 | } | |
1043 | ||
1044 | /* | |
8feae131 DH |
1045 | * set up a shared mapping on a file (the driver or filesystem provides and |
1046 | * pins the storage) | |
1da177e4 | 1047 | */ |
8feae131 | 1048 | static int do_mmap_shared_file(struct vm_area_struct *vma) |
1da177e4 LT |
1049 | { |
1050 | int ret; | |
1051 | ||
1052 | ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); | |
dd8632a1 PM |
1053 | if (ret == 0) { |
1054 | vma->vm_region->vm_top = vma->vm_region->vm_end; | |
1055 | return ret; | |
1056 | } | |
1da177e4 LT |
1057 | if (ret != -ENOSYS) |
1058 | return ret; | |
1059 | ||
1060 | /* getting an ENOSYS error indicates that direct mmap isn't | |
1061 | * possible (as opposed to tried but failed) so we'll fall | |
1062 | * through to making a private copy of the data and mapping | |
1063 | * that if we can */ | |
1064 | return -ENODEV; | |
1065 | } | |
1066 | ||
1067 | /* | |
1068 | * set up a private mapping or an anonymous shared mapping | |
1069 | */ | |
8feae131 DH |
1070 | static int do_mmap_private(struct vm_area_struct *vma, |
1071 | struct vm_region *region, | |
1072 | unsigned long len) | |
1da177e4 | 1073 | { |
8feae131 DH |
1074 | struct page *pages; |
1075 | unsigned long total, point, n, rlen; | |
1da177e4 | 1076 | void *base; |
8feae131 | 1077 | int ret, order; |
1da177e4 LT |
1078 | |
1079 | /* invoke the file's mapping function so that it can keep track of | |
1080 | * shared mappings on devices or memory | |
1081 | * - VM_MAYSHARE will be set if it may attempt to share | |
1082 | */ | |
1083 | if (vma->vm_file) { | |
1084 | ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); | |
dd8632a1 | 1085 | if (ret == 0) { |
1da177e4 | 1086 | /* shouldn't return success if we're not sharing */ |
dd8632a1 PM |
1087 | BUG_ON(!(vma->vm_flags & VM_MAYSHARE)); |
1088 | vma->vm_region->vm_top = vma->vm_region->vm_end; | |
1089 | return ret; | |
1da177e4 | 1090 | } |
dd8632a1 PM |
1091 | if (ret != -ENOSYS) |
1092 | return ret; | |
1da177e4 LT |
1093 | |
1094 | /* getting an ENOSYS error indicates that direct mmap isn't | |
1095 | * possible (as opposed to tried but failed) so we'll try to | |
1096 | * make a private copy of the data and map that instead */ | |
1097 | } | |
1098 | ||
8feae131 DH |
1099 | rlen = PAGE_ALIGN(len); |
1100 | ||
1da177e4 LT |
1101 | /* allocate some memory to hold the mapping |
1102 | * - note that this may not return a page-aligned address if the object | |
1103 | * we're allocating is smaller than a page | |
1104 | */ | |
8feae131 DH |
1105 | order = get_order(rlen); |
1106 | kdebug("alloc order %d for %lx", order, len); | |
1107 | ||
1108 | pages = alloc_pages(GFP_KERNEL, order); | |
1109 | if (!pages) | |
1da177e4 LT |
1110 | goto enomem; |
1111 | ||
8feae131 | 1112 | total = 1 << order; |
33e5d769 | 1113 | atomic_long_add(total, &mmap_pages_allocated); |
8feae131 DH |
1114 | |
1115 | point = rlen >> PAGE_SHIFT; | |
dd8632a1 PM |
1116 | |
1117 | /* we allocated a power-of-2 sized page set, so we may want to trim off | |
1118 | * the excess */ | |
1119 | if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) { | |
1120 | while (total > point) { | |
1121 | order = ilog2(total - point); | |
1122 | n = 1 << order; | |
1123 | kdebug("shave %lu/%lu @%lu", n, total - point, total); | |
33e5d769 | 1124 | atomic_long_sub(n, &mmap_pages_allocated); |
dd8632a1 PM |
1125 | total -= n; |
1126 | set_page_refcounted(pages + total); | |
1127 | __free_pages(pages + total, order); | |
1128 | } | |
8feae131 DH |
1129 | } |
1130 | ||
8feae131 DH |
1131 | for (point = 1; point < total; point++) |
1132 | set_page_refcounted(&pages[point]); | |
1da177e4 | 1133 | |
8feae131 DH |
1134 | base = page_address(pages); |
1135 | region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY; | |
1136 | region->vm_start = (unsigned long) base; | |
1137 | region->vm_end = region->vm_start + rlen; | |
dd8632a1 | 1138 | region->vm_top = region->vm_start + (total << PAGE_SHIFT); |
8feae131 DH |
1139 | |
1140 | vma->vm_start = region->vm_start; | |
1141 | vma->vm_end = region->vm_start + len; | |
1da177e4 LT |
1142 | |
1143 | if (vma->vm_file) { | |
1144 | /* read the contents of a file into the copy */ | |
1145 | mm_segment_t old_fs; | |
1146 | loff_t fpos; | |
1147 | ||
1148 | fpos = vma->vm_pgoff; | |
1149 | fpos <<= PAGE_SHIFT; | |
1150 | ||
1151 | old_fs = get_fs(); | |
1152 | set_fs(KERNEL_DS); | |
8feae131 | 1153 | ret = vma->vm_file->f_op->read(vma->vm_file, base, rlen, &fpos); |
1da177e4 LT |
1154 | set_fs(old_fs); |
1155 | ||
1156 | if (ret < 0) | |
1157 | goto error_free; | |
1158 | ||
1159 | /* clear the last little bit */ | |
8feae131 DH |
1160 | if (ret < rlen) |
1161 | memset(base + ret, 0, rlen - ret); | |
1da177e4 LT |
1162 | |
1163 | } else { | |
1164 | /* if it's an anonymous mapping, then just clear it */ | |
8feae131 | 1165 | memset(base, 0, rlen); |
1da177e4 LT |
1166 | } |
1167 | ||
1168 | return 0; | |
1169 | ||
1170 | error_free: | |
8feae131 DH |
1171 | free_page_series(region->vm_start, region->vm_end); |
1172 | region->vm_start = vma->vm_start = 0; | |
1173 | region->vm_end = vma->vm_end = 0; | |
dd8632a1 | 1174 | region->vm_top = 0; |
1da177e4 LT |
1175 | return ret; |
1176 | ||
1177 | enomem: | |
05ae6fa3 GU |
1178 | printk("Allocation of length %lu from process %d (%s) failed\n", |
1179 | len, current->pid, current->comm); | |
1da177e4 LT |
1180 | show_free_areas(); |
1181 | return -ENOMEM; | |
1182 | } | |
1183 | ||
1184 | /* | |
1185 | * handle mapping creation for uClinux | |
1186 | */ | |
1187 | unsigned long do_mmap_pgoff(struct file *file, | |
1188 | unsigned long addr, | |
1189 | unsigned long len, | |
1190 | unsigned long prot, | |
1191 | unsigned long flags, | |
1192 | unsigned long pgoff) | |
1193 | { | |
8feae131 DH |
1194 | struct vm_area_struct *vma; |
1195 | struct vm_region *region; | |
1da177e4 | 1196 | struct rb_node *rb; |
8feae131 | 1197 | unsigned long capabilities, vm_flags, result; |
1da177e4 LT |
1198 | int ret; |
1199 | ||
8feae131 DH |
1200 | kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff); |
1201 | ||
7cd94146 EP |
1202 | if (!(flags & MAP_FIXED)) |
1203 | addr = round_hint_to_min(addr); | |
1204 | ||
1da177e4 LT |
1205 | /* decide whether we should attempt the mapping, and if so what sort of |
1206 | * mapping */ | |
1207 | ret = validate_mmap_request(file, addr, len, prot, flags, pgoff, | |
1208 | &capabilities); | |
8feae131 DH |
1209 | if (ret < 0) { |
1210 | kleave(" = %d [val]", ret); | |
1da177e4 | 1211 | return ret; |
8feae131 | 1212 | } |
1da177e4 LT |
1213 | |
1214 | /* we've determined that we can make the mapping, now translate what we | |
1215 | * now know into VMA flags */ | |
1216 | vm_flags = determine_vm_flags(file, prot, flags, capabilities); | |
1217 | ||
8feae131 DH |
1218 | /* we're going to need to record the mapping */ |
1219 | region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL); | |
1220 | if (!region) | |
1221 | goto error_getting_region; | |
1222 | ||
1223 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); | |
1224 | if (!vma) | |
1225 | goto error_getting_vma; | |
1da177e4 | 1226 | |
8feae131 DH |
1227 | atomic_set(®ion->vm_usage, 1); |
1228 | region->vm_flags = vm_flags; | |
1229 | region->vm_pgoff = pgoff; | |
1230 | ||
1231 | INIT_LIST_HEAD(&vma->anon_vma_node); | |
1232 | vma->vm_flags = vm_flags; | |
1233 | vma->vm_pgoff = pgoff; | |
1da177e4 | 1234 | |
8feae131 DH |
1235 | if (file) { |
1236 | region->vm_file = file; | |
1237 | get_file(file); | |
1238 | vma->vm_file = file; | |
1239 | get_file(file); | |
1240 | if (vm_flags & VM_EXECUTABLE) { | |
1241 | added_exe_file_vma(current->mm); | |
1242 | vma->vm_mm = current->mm; | |
1243 | } | |
1244 | } | |
1245 | ||
1246 | down_write(&nommu_region_sem); | |
1247 | ||
1248 | /* if we want to share, we need to check for regions created by other | |
1da177e4 | 1249 | * mmap() calls that overlap with our proposed mapping |
8feae131 | 1250 | * - we can only share with a superset match on most regular files |
1da177e4 LT |
1251 | * - shared mappings on character devices and memory backed files are |
1252 | * permitted to overlap inexactly as far as we are concerned for in | |
1253 | * these cases, sharing is handled in the driver or filesystem rather | |
1254 | * than here | |
1255 | */ | |
1256 | if (vm_flags & VM_MAYSHARE) { | |
8feae131 DH |
1257 | struct vm_region *pregion; |
1258 | unsigned long pglen, rpglen, pgend, rpgend, start; | |
1da177e4 | 1259 | |
8feae131 DH |
1260 | pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
1261 | pgend = pgoff + pglen; | |
165b2392 | 1262 | |
8feae131 DH |
1263 | for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) { |
1264 | pregion = rb_entry(rb, struct vm_region, vm_rb); | |
1da177e4 | 1265 | |
8feae131 | 1266 | if (!(pregion->vm_flags & VM_MAYSHARE)) |
1da177e4 LT |
1267 | continue; |
1268 | ||
1269 | /* search for overlapping mappings on the same file */ | |
8feae131 DH |
1270 | if (pregion->vm_file->f_path.dentry->d_inode != |
1271 | file->f_path.dentry->d_inode) | |
1da177e4 LT |
1272 | continue; |
1273 | ||
8feae131 | 1274 | if (pregion->vm_pgoff >= pgend) |
1da177e4 LT |
1275 | continue; |
1276 | ||
8feae131 DH |
1277 | rpglen = pregion->vm_end - pregion->vm_start; |
1278 | rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1279 | rpgend = pregion->vm_pgoff + rpglen; | |
1280 | if (pgoff >= rpgend) | |
1da177e4 LT |
1281 | continue; |
1282 | ||
8feae131 DH |
1283 | /* handle inexactly overlapping matches between |
1284 | * mappings */ | |
1285 | if ((pregion->vm_pgoff != pgoff || rpglen != pglen) && | |
1286 | !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) { | |
1287 | /* new mapping is not a subset of the region */ | |
1da177e4 LT |
1288 | if (!(capabilities & BDI_CAP_MAP_DIRECT)) |
1289 | goto sharing_violation; | |
1290 | continue; | |
1291 | } | |
1292 | ||
8feae131 DH |
1293 | /* we've found a region we can share */ |
1294 | atomic_inc(&pregion->vm_usage); | |
1295 | vma->vm_region = pregion; | |
1296 | start = pregion->vm_start; | |
1297 | start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT; | |
1298 | vma->vm_start = start; | |
1299 | vma->vm_end = start + len; | |
1300 | ||
1301 | if (pregion->vm_flags & VM_MAPPED_COPY) { | |
1302 | kdebug("share copy"); | |
1303 | vma->vm_flags |= VM_MAPPED_COPY; | |
1304 | } else { | |
1305 | kdebug("share mmap"); | |
1306 | ret = do_mmap_shared_file(vma); | |
1307 | if (ret < 0) { | |
1308 | vma->vm_region = NULL; | |
1309 | vma->vm_start = 0; | |
1310 | vma->vm_end = 0; | |
1311 | atomic_dec(&pregion->vm_usage); | |
1312 | pregion = NULL; | |
1313 | goto error_just_free; | |
1314 | } | |
1315 | } | |
1316 | fput(region->vm_file); | |
1317 | kmem_cache_free(vm_region_jar, region); | |
1318 | region = pregion; | |
1319 | result = start; | |
1320 | goto share; | |
1da177e4 LT |
1321 | } |
1322 | ||
1da177e4 LT |
1323 | /* obtain the address at which to make a shared mapping |
1324 | * - this is the hook for quasi-memory character devices to | |
1325 | * tell us the location of a shared mapping | |
1326 | */ | |
1327 | if (file && file->f_op->get_unmapped_area) { | |
1328 | addr = file->f_op->get_unmapped_area(file, addr, len, | |
1329 | pgoff, flags); | |
1330 | if (IS_ERR((void *) addr)) { | |
1331 | ret = addr; | |
1332 | if (ret != (unsigned long) -ENOSYS) | |
8feae131 | 1333 | goto error_just_free; |
1da177e4 LT |
1334 | |
1335 | /* the driver refused to tell us where to site | |
1336 | * the mapping so we'll have to attempt to copy | |
1337 | * it */ | |
1338 | ret = (unsigned long) -ENODEV; | |
1339 | if (!(capabilities & BDI_CAP_MAP_COPY)) | |
8feae131 | 1340 | goto error_just_free; |
1da177e4 LT |
1341 | |
1342 | capabilities &= ~BDI_CAP_MAP_DIRECT; | |
8feae131 DH |
1343 | } else { |
1344 | vma->vm_start = region->vm_start = addr; | |
1345 | vma->vm_end = region->vm_end = addr + len; | |
1da177e4 LT |
1346 | } |
1347 | } | |
1348 | } | |
1349 | ||
8feae131 | 1350 | vma->vm_region = region; |
1da177e4 LT |
1351 | |
1352 | /* set up the mapping */ | |
1353 | if (file && vma->vm_flags & VM_SHARED) | |
8feae131 | 1354 | ret = do_mmap_shared_file(vma); |
1da177e4 | 1355 | else |
8feae131 | 1356 | ret = do_mmap_private(vma, region, len); |
1da177e4 | 1357 | if (ret < 0) |
8feae131 DH |
1358 | goto error_put_region; |
1359 | ||
1360 | add_nommu_region(region); | |
1da177e4 LT |
1361 | |
1362 | /* okay... we have a mapping; now we have to register it */ | |
8feae131 | 1363 | result = vma->vm_start; |
1da177e4 | 1364 | |
1da177e4 LT |
1365 | current->mm->total_vm += len >> PAGE_SHIFT; |
1366 | ||
8feae131 DH |
1367 | share: |
1368 | add_vma_to_mm(current->mm, vma); | |
1da177e4 | 1369 | |
8feae131 | 1370 | up_write(&nommu_region_sem); |
1da177e4 LT |
1371 | |
1372 | if (prot & PROT_EXEC) | |
8feae131 | 1373 | flush_icache_range(result, result + len); |
1da177e4 | 1374 | |
8feae131 DH |
1375 | kleave(" = %lx", result); |
1376 | return result; | |
1da177e4 | 1377 | |
8feae131 DH |
1378 | error_put_region: |
1379 | __put_nommu_region(region); | |
1da177e4 | 1380 | if (vma) { |
925d1c40 | 1381 | if (vma->vm_file) { |
3fcd03e0 | 1382 | fput(vma->vm_file); |
925d1c40 MH |
1383 | if (vma->vm_flags & VM_EXECUTABLE) |
1384 | removed_exe_file_vma(vma->vm_mm); | |
1385 | } | |
8feae131 | 1386 | kmem_cache_free(vm_area_cachep, vma); |
1da177e4 | 1387 | } |
8feae131 | 1388 | kleave(" = %d [pr]", ret); |
1da177e4 LT |
1389 | return ret; |
1390 | ||
8feae131 DH |
1391 | error_just_free: |
1392 | up_write(&nommu_region_sem); | |
1393 | error: | |
1394 | fput(region->vm_file); | |
1395 | kmem_cache_free(vm_region_jar, region); | |
1396 | fput(vma->vm_file); | |
1397 | if (vma->vm_flags & VM_EXECUTABLE) | |
1398 | removed_exe_file_vma(vma->vm_mm); | |
1399 | kmem_cache_free(vm_area_cachep, vma); | |
1400 | kleave(" = %d", ret); | |
1401 | return ret; | |
1402 | ||
1403 | sharing_violation: | |
1404 | up_write(&nommu_region_sem); | |
1405 | printk(KERN_WARNING "Attempt to share mismatched mappings\n"); | |
1406 | ret = -EINVAL; | |
1407 | goto error; | |
1da177e4 | 1408 | |
8feae131 DH |
1409 | error_getting_vma: |
1410 | kmem_cache_free(vm_region_jar, region); | |
1411 | printk(KERN_WARNING "Allocation of vma for %lu byte allocation" | |
1412 | " from process %d failed\n", | |
1da177e4 LT |
1413 | len, current->pid); |
1414 | show_free_areas(); | |
1415 | return -ENOMEM; | |
1416 | ||
8feae131 DH |
1417 | error_getting_region: |
1418 | printk(KERN_WARNING "Allocation of vm region for %lu byte allocation" | |
1419 | " from process %d failed\n", | |
1da177e4 LT |
1420 | len, current->pid); |
1421 | show_free_areas(); | |
1422 | return -ENOMEM; | |
1423 | } | |
b5073173 | 1424 | EXPORT_SYMBOL(do_mmap_pgoff); |
1da177e4 LT |
1425 | |
1426 | /* | |
8feae131 DH |
1427 | * split a vma into two pieces at address 'addr', a new vma is allocated either |
1428 | * for the first part or the tail. | |
1da177e4 | 1429 | */ |
8feae131 DH |
1430 | int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, |
1431 | unsigned long addr, int new_below) | |
1da177e4 | 1432 | { |
8feae131 DH |
1433 | struct vm_area_struct *new; |
1434 | struct vm_region *region; | |
1435 | unsigned long npages; | |
1da177e4 | 1436 | |
8feae131 | 1437 | kenter(""); |
1da177e4 | 1438 | |
8feae131 DH |
1439 | /* we're only permitted to split anonymous regions that have a single |
1440 | * owner */ | |
1441 | if (vma->vm_file || | |
1442 | atomic_read(&vma->vm_region->vm_usage) != 1) | |
1443 | return -ENOMEM; | |
1da177e4 | 1444 | |
8feae131 DH |
1445 | if (mm->map_count >= sysctl_max_map_count) |
1446 | return -ENOMEM; | |
1da177e4 | 1447 | |
8feae131 DH |
1448 | region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL); |
1449 | if (!region) | |
1450 | return -ENOMEM; | |
1da177e4 | 1451 | |
8feae131 DH |
1452 | new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); |
1453 | if (!new) { | |
1454 | kmem_cache_free(vm_region_jar, region); | |
1455 | return -ENOMEM; | |
1456 | } | |
1457 | ||
1458 | /* most fields are the same, copy all, and then fixup */ | |
1459 | *new = *vma; | |
1460 | *region = *vma->vm_region; | |
1461 | new->vm_region = region; | |
1462 | ||
1463 | npages = (addr - vma->vm_start) >> PAGE_SHIFT; | |
1464 | ||
1465 | if (new_below) { | |
dd8632a1 | 1466 | region->vm_top = region->vm_end = new->vm_end = addr; |
8feae131 DH |
1467 | } else { |
1468 | region->vm_start = new->vm_start = addr; | |
1469 | region->vm_pgoff = new->vm_pgoff += npages; | |
1da177e4 | 1470 | } |
8feae131 DH |
1471 | |
1472 | if (new->vm_ops && new->vm_ops->open) | |
1473 | new->vm_ops->open(new); | |
1474 | ||
1475 | delete_vma_from_mm(vma); | |
1476 | down_write(&nommu_region_sem); | |
1477 | delete_nommu_region(vma->vm_region); | |
1478 | if (new_below) { | |
1479 | vma->vm_region->vm_start = vma->vm_start = addr; | |
1480 | vma->vm_region->vm_pgoff = vma->vm_pgoff += npages; | |
1481 | } else { | |
1482 | vma->vm_region->vm_end = vma->vm_end = addr; | |
dd8632a1 | 1483 | vma->vm_region->vm_top = addr; |
8feae131 DH |
1484 | } |
1485 | add_nommu_region(vma->vm_region); | |
1486 | add_nommu_region(new->vm_region); | |
1487 | up_write(&nommu_region_sem); | |
1488 | add_vma_to_mm(mm, vma); | |
1489 | add_vma_to_mm(mm, new); | |
1490 | return 0; | |
1da177e4 LT |
1491 | } |
1492 | ||
3034097a | 1493 | /* |
8feae131 DH |
1494 | * shrink a VMA by removing the specified chunk from either the beginning or |
1495 | * the end | |
3034097a | 1496 | */ |
8feae131 DH |
1497 | static int shrink_vma(struct mm_struct *mm, |
1498 | struct vm_area_struct *vma, | |
1499 | unsigned long from, unsigned long to) | |
1da177e4 | 1500 | { |
8feae131 | 1501 | struct vm_region *region; |
1da177e4 | 1502 | |
8feae131 | 1503 | kenter(""); |
1da177e4 | 1504 | |
8feae131 DH |
1505 | /* adjust the VMA's pointers, which may reposition it in the MM's tree |
1506 | * and list */ | |
1507 | delete_vma_from_mm(vma); | |
1508 | if (from > vma->vm_start) | |
1509 | vma->vm_end = from; | |
1510 | else | |
1511 | vma->vm_start = to; | |
1512 | add_vma_to_mm(mm, vma); | |
1da177e4 | 1513 | |
8feae131 DH |
1514 | /* cut the backing region down to size */ |
1515 | region = vma->vm_region; | |
1516 | BUG_ON(atomic_read(®ion->vm_usage) != 1); | |
1517 | ||
1518 | down_write(&nommu_region_sem); | |
1519 | delete_nommu_region(region); | |
dd8632a1 PM |
1520 | if (from > region->vm_start) { |
1521 | to = region->vm_top; | |
1522 | region->vm_top = region->vm_end = from; | |
1523 | } else { | |
8feae131 | 1524 | region->vm_start = to; |
dd8632a1 | 1525 | } |
8feae131 DH |
1526 | add_nommu_region(region); |
1527 | up_write(&nommu_region_sem); | |
1528 | ||
1529 | free_page_series(from, to); | |
1530 | return 0; | |
1531 | } | |
1da177e4 | 1532 | |
8feae131 DH |
1533 | /* |
1534 | * release a mapping | |
1535 | * - under NOMMU conditions the chunk to be unmapped must be backed by a single | |
1536 | * VMA, though it need not cover the whole VMA | |
1537 | */ | |
1538 | int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) | |
1539 | { | |
1540 | struct vm_area_struct *vma; | |
1541 | struct rb_node *rb; | |
1542 | unsigned long end = start + len; | |
1543 | int ret; | |
1da177e4 | 1544 | |
8feae131 | 1545 | kenter(",%lx,%zx", start, len); |
1da177e4 | 1546 | |
8feae131 DH |
1547 | if (len == 0) |
1548 | return -EINVAL; | |
365e9c87 | 1549 | |
8feae131 DH |
1550 | /* find the first potentially overlapping VMA */ |
1551 | vma = find_vma(mm, start); | |
1552 | if (!vma) { | |
33e5d769 DH |
1553 | static int limit = 0; |
1554 | if (limit < 5) { | |
1555 | printk(KERN_WARNING | |
1556 | "munmap of memory not mmapped by process %d" | |
1557 | " (%s): 0x%lx-0x%lx\n", | |
1558 | current->pid, current->comm, | |
1559 | start, start + len - 1); | |
1560 | limit++; | |
1561 | } | |
8feae131 DH |
1562 | return -EINVAL; |
1563 | } | |
1da177e4 | 1564 | |
8feae131 DH |
1565 | /* we're allowed to split an anonymous VMA but not a file-backed one */ |
1566 | if (vma->vm_file) { | |
1567 | do { | |
1568 | if (start > vma->vm_start) { | |
1569 | kleave(" = -EINVAL [miss]"); | |
1570 | return -EINVAL; | |
1571 | } | |
1572 | if (end == vma->vm_end) | |
1573 | goto erase_whole_vma; | |
1574 | rb = rb_next(&vma->vm_rb); | |
1575 | vma = rb_entry(rb, struct vm_area_struct, vm_rb); | |
1576 | } while (rb); | |
1577 | kleave(" = -EINVAL [split file]"); | |
1578 | return -EINVAL; | |
1579 | } else { | |
1580 | /* the chunk must be a subset of the VMA found */ | |
1581 | if (start == vma->vm_start && end == vma->vm_end) | |
1582 | goto erase_whole_vma; | |
1583 | if (start < vma->vm_start || end > vma->vm_end) { | |
1584 | kleave(" = -EINVAL [superset]"); | |
1585 | return -EINVAL; | |
1586 | } | |
1587 | if (start & ~PAGE_MASK) { | |
1588 | kleave(" = -EINVAL [unaligned start]"); | |
1589 | return -EINVAL; | |
1590 | } | |
1591 | if (end != vma->vm_end && end & ~PAGE_MASK) { | |
1592 | kleave(" = -EINVAL [unaligned split]"); | |
1593 | return -EINVAL; | |
1594 | } | |
1595 | if (start != vma->vm_start && end != vma->vm_end) { | |
1596 | ret = split_vma(mm, vma, start, 1); | |
1597 | if (ret < 0) { | |
1598 | kleave(" = %d [split]", ret); | |
1599 | return ret; | |
1600 | } | |
1601 | } | |
1602 | return shrink_vma(mm, vma, start, end); | |
1603 | } | |
1da177e4 | 1604 | |
8feae131 DH |
1605 | erase_whole_vma: |
1606 | delete_vma_from_mm(vma); | |
1607 | delete_vma(mm, vma); | |
1608 | kleave(" = 0"); | |
1da177e4 LT |
1609 | return 0; |
1610 | } | |
b5073173 | 1611 | EXPORT_SYMBOL(do_munmap); |
1da177e4 | 1612 | |
6a6160a7 | 1613 | SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) |
3034097a DH |
1614 | { |
1615 | int ret; | |
1616 | struct mm_struct *mm = current->mm; | |
1617 | ||
1618 | down_write(&mm->mmap_sem); | |
1619 | ret = do_munmap(mm, addr, len); | |
1620 | up_write(&mm->mmap_sem); | |
1621 | return ret; | |
1622 | } | |
1623 | ||
1624 | /* | |
8feae131 | 1625 | * release all the mappings made in a process's VM space |
3034097a | 1626 | */ |
8feae131 | 1627 | void exit_mmap(struct mm_struct *mm) |
1da177e4 | 1628 | { |
8feae131 | 1629 | struct vm_area_struct *vma; |
1da177e4 | 1630 | |
8feae131 DH |
1631 | if (!mm) |
1632 | return; | |
1da177e4 | 1633 | |
8feae131 | 1634 | kenter(""); |
1da177e4 | 1635 | |
8feae131 | 1636 | mm->total_vm = 0; |
1da177e4 | 1637 | |
8feae131 DH |
1638 | while ((vma = mm->mmap)) { |
1639 | mm->mmap = vma->vm_next; | |
1640 | delete_vma_from_mm(vma); | |
1641 | delete_vma(mm, vma); | |
1da177e4 | 1642 | } |
8feae131 DH |
1643 | |
1644 | kleave(""); | |
1da177e4 LT |
1645 | } |
1646 | ||
1da177e4 LT |
1647 | unsigned long do_brk(unsigned long addr, unsigned long len) |
1648 | { | |
1649 | return -ENOMEM; | |
1650 | } | |
1651 | ||
1652 | /* | |
6fa5f80b DH |
1653 | * expand (or shrink) an existing mapping, potentially moving it at the same |
1654 | * time (controlled by the MREMAP_MAYMOVE flag and available VM space) | |
1da177e4 | 1655 | * |
6fa5f80b | 1656 | * under NOMMU conditions, we only permit changing a mapping's size, and only |
8feae131 DH |
1657 | * as long as it stays within the region allocated by do_mmap_private() and the |
1658 | * block is not shareable | |
1da177e4 | 1659 | * |
6fa5f80b | 1660 | * MREMAP_FIXED is not supported under NOMMU conditions |
1da177e4 LT |
1661 | */ |
1662 | unsigned long do_mremap(unsigned long addr, | |
1663 | unsigned long old_len, unsigned long new_len, | |
1664 | unsigned long flags, unsigned long new_addr) | |
1665 | { | |
6fa5f80b | 1666 | struct vm_area_struct *vma; |
1da177e4 LT |
1667 | |
1668 | /* insanity checks first */ | |
8feae131 | 1669 | if (old_len == 0 || new_len == 0) |
1da177e4 LT |
1670 | return (unsigned long) -EINVAL; |
1671 | ||
8feae131 DH |
1672 | if (addr & ~PAGE_MASK) |
1673 | return -EINVAL; | |
1674 | ||
1da177e4 LT |
1675 | if (flags & MREMAP_FIXED && new_addr != addr) |
1676 | return (unsigned long) -EINVAL; | |
1677 | ||
8feae131 | 1678 | vma = find_vma_exact(current->mm, addr, old_len); |
6fa5f80b DH |
1679 | if (!vma) |
1680 | return (unsigned long) -EINVAL; | |
1da177e4 | 1681 | |
6fa5f80b | 1682 | if (vma->vm_end != vma->vm_start + old_len) |
1da177e4 LT |
1683 | return (unsigned long) -EFAULT; |
1684 | ||
6fa5f80b | 1685 | if (vma->vm_flags & VM_MAYSHARE) |
1da177e4 LT |
1686 | return (unsigned long) -EPERM; |
1687 | ||
8feae131 | 1688 | if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start) |
1da177e4 LT |
1689 | return (unsigned long) -ENOMEM; |
1690 | ||
1691 | /* all checks complete - do it */ | |
6fa5f80b | 1692 | vma->vm_end = vma->vm_start + new_len; |
6fa5f80b DH |
1693 | return vma->vm_start; |
1694 | } | |
b5073173 | 1695 | EXPORT_SYMBOL(do_mremap); |
6fa5f80b | 1696 | |
6a6160a7 HC |
1697 | SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len, |
1698 | unsigned long, new_len, unsigned long, flags, | |
1699 | unsigned long, new_addr) | |
6fa5f80b DH |
1700 | { |
1701 | unsigned long ret; | |
1702 | ||
1703 | down_write(¤t->mm->mmap_sem); | |
1704 | ret = do_mremap(addr, old_len, new_len, flags, new_addr); | |
1705 | up_write(¤t->mm->mmap_sem); | |
1706 | return ret; | |
1da177e4 LT |
1707 | } |
1708 | ||
6aab341e | 1709 | struct page *follow_page(struct vm_area_struct *vma, unsigned long address, |
deceb6cd | 1710 | unsigned int foll_flags) |
1da177e4 LT |
1711 | { |
1712 | return NULL; | |
1713 | } | |
1714 | ||
1da177e4 LT |
1715 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long from, |
1716 | unsigned long to, unsigned long size, pgprot_t prot) | |
1717 | { | |
66aa2b4b GU |
1718 | vma->vm_start = vma->vm_pgoff << PAGE_SHIFT; |
1719 | return 0; | |
1da177e4 | 1720 | } |
22c4af40 | 1721 | EXPORT_SYMBOL(remap_pfn_range); |
1da177e4 | 1722 | |
f905bc44 PM |
1723 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, |
1724 | unsigned long pgoff) | |
1725 | { | |
1726 | unsigned int size = vma->vm_end - vma->vm_start; | |
1727 | ||
1728 | if (!(vma->vm_flags & VM_USERMAP)) | |
1729 | return -EINVAL; | |
1730 | ||
1731 | vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT)); | |
1732 | vma->vm_end = vma->vm_start + size; | |
1733 | ||
1734 | return 0; | |
1735 | } | |
1736 | EXPORT_SYMBOL(remap_vmalloc_range); | |
1737 | ||
1da177e4 LT |
1738 | void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page) |
1739 | { | |
1740 | } | |
1741 | ||
1742 | unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr, | |
1743 | unsigned long len, unsigned long pgoff, unsigned long flags) | |
1744 | { | |
1745 | return -ENOMEM; | |
1746 | } | |
1747 | ||
1363c3cd | 1748 | void arch_unmap_area(struct mm_struct *mm, unsigned long addr) |
1da177e4 LT |
1749 | { |
1750 | } | |
1751 | ||
1da177e4 LT |
1752 | void unmap_mapping_range(struct address_space *mapping, |
1753 | loff_t const holebegin, loff_t const holelen, | |
1754 | int even_cows) | |
1755 | { | |
1756 | } | |
22c4af40 | 1757 | EXPORT_SYMBOL(unmap_mapping_range); |
1da177e4 | 1758 | |
d56e03cd DH |
1759 | /* |
1760 | * ask for an unmapped area at which to create a mapping on a file | |
1761 | */ | |
1762 | unsigned long get_unmapped_area(struct file *file, unsigned long addr, | |
1763 | unsigned long len, unsigned long pgoff, | |
1764 | unsigned long flags) | |
1765 | { | |
1766 | unsigned long (*get_area)(struct file *, unsigned long, unsigned long, | |
1767 | unsigned long, unsigned long); | |
1768 | ||
1769 | get_area = current->mm->get_unmapped_area; | |
1770 | if (file && file->f_op && file->f_op->get_unmapped_area) | |
1771 | get_area = file->f_op->get_unmapped_area; | |
1772 | ||
1773 | if (!get_area) | |
1774 | return -ENOSYS; | |
1775 | ||
1776 | return get_area(file, addr, len, pgoff, flags); | |
1777 | } | |
d56e03cd DH |
1778 | EXPORT_SYMBOL(get_unmapped_area); |
1779 | ||
1da177e4 LT |
1780 | /* |
1781 | * Check that a process has enough memory to allocate a new virtual | |
1782 | * mapping. 0 means there is enough memory for the allocation to | |
1783 | * succeed and -ENOMEM implies there is not. | |
1784 | * | |
1785 | * We currently support three overcommit policies, which are set via the | |
1786 | * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting | |
1787 | * | |
1788 | * Strict overcommit modes added 2002 Feb 26 by Alan Cox. | |
1789 | * Additional code 2002 Jul 20 by Robert Love. | |
1790 | * | |
1791 | * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. | |
1792 | * | |
1793 | * Note this is a helper function intended to be used by LSMs which | |
1794 | * wish to use this logic. | |
1795 | */ | |
34b4e4aa | 1796 | int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) |
1da177e4 LT |
1797 | { |
1798 | unsigned long free, allowed; | |
1799 | ||
1800 | vm_acct_memory(pages); | |
1801 | ||
1802 | /* | |
1803 | * Sometimes we want to use more memory than we have | |
1804 | */ | |
1805 | if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) | |
1806 | return 0; | |
1807 | ||
1808 | if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { | |
1809 | unsigned long n; | |
1810 | ||
347ce434 | 1811 | free = global_page_state(NR_FILE_PAGES); |
1da177e4 LT |
1812 | free += nr_swap_pages; |
1813 | ||
1814 | /* | |
1815 | * Any slabs which are created with the | |
1816 | * SLAB_RECLAIM_ACCOUNT flag claim to have contents | |
1817 | * which are reclaimable, under pressure. The dentry | |
1818 | * cache and most inode caches should fall into this | |
1819 | */ | |
972d1a7b | 1820 | free += global_page_state(NR_SLAB_RECLAIMABLE); |
1da177e4 LT |
1821 | |
1822 | /* | |
1823 | * Leave the last 3% for root | |
1824 | */ | |
1825 | if (!cap_sys_admin) | |
1826 | free -= free / 32; | |
1827 | ||
1828 | if (free > pages) | |
1829 | return 0; | |
1830 | ||
1831 | /* | |
1832 | * nr_free_pages() is very expensive on large systems, | |
1833 | * only call if we're about to fail. | |
1834 | */ | |
1835 | n = nr_free_pages(); | |
d5ddc79b HA |
1836 | |
1837 | /* | |
1838 | * Leave reserved pages. The pages are not for anonymous pages. | |
1839 | */ | |
1840 | if (n <= totalreserve_pages) | |
1841 | goto error; | |
1842 | else | |
1843 | n -= totalreserve_pages; | |
1844 | ||
1845 | /* | |
1846 | * Leave the last 3% for root | |
1847 | */ | |
1da177e4 LT |
1848 | if (!cap_sys_admin) |
1849 | n -= n / 32; | |
1850 | free += n; | |
1851 | ||
1852 | if (free > pages) | |
1853 | return 0; | |
d5ddc79b HA |
1854 | |
1855 | goto error; | |
1da177e4 LT |
1856 | } |
1857 | ||
1858 | allowed = totalram_pages * sysctl_overcommit_ratio / 100; | |
1859 | /* | |
1860 | * Leave the last 3% for root | |
1861 | */ | |
1862 | if (!cap_sys_admin) | |
1863 | allowed -= allowed / 32; | |
1864 | allowed += total_swap_pages; | |
1865 | ||
1866 | /* Don't let a single process grow too big: | |
1867 | leave 3% of the size of this process for other processes */ | |
731572d3 AC |
1868 | if (mm) |
1869 | allowed -= mm->total_vm / 32; | |
1da177e4 | 1870 | |
00a62ce9 | 1871 | if (percpu_counter_read_positive(&vm_committed_as) < allowed) |
1da177e4 | 1872 | return 0; |
00a62ce9 | 1873 | |
d5ddc79b | 1874 | error: |
1da177e4 LT |
1875 | vm_unacct_memory(pages); |
1876 | ||
1877 | return -ENOMEM; | |
1878 | } | |
1879 | ||
1880 | int in_gate_area_no_task(unsigned long addr) | |
1881 | { | |
1882 | return 0; | |
1883 | } | |
b0e15190 | 1884 | |
d0217ac0 | 1885 | int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
b0e15190 DH |
1886 | { |
1887 | BUG(); | |
d0217ac0 | 1888 | return 0; |
b0e15190 | 1889 | } |
b5073173 | 1890 | EXPORT_SYMBOL(filemap_fault); |
0ec76a11 DH |
1891 | |
1892 | /* | |
1893 | * Access another process' address space. | |
1894 | * - source/target buffer must be kernel space | |
1895 | */ | |
1896 | int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write) | |
1897 | { | |
0ec76a11 DH |
1898 | struct vm_area_struct *vma; |
1899 | struct mm_struct *mm; | |
1900 | ||
1901 | if (addr + len < addr) | |
1902 | return 0; | |
1903 | ||
1904 | mm = get_task_mm(tsk); | |
1905 | if (!mm) | |
1906 | return 0; | |
1907 | ||
1908 | down_read(&mm->mmap_sem); | |
1909 | ||
1910 | /* the access must start within one of the target process's mappings */ | |
0159b141 DH |
1911 | vma = find_vma(mm, addr); |
1912 | if (vma) { | |
0ec76a11 DH |
1913 | /* don't overrun this mapping */ |
1914 | if (addr + len >= vma->vm_end) | |
1915 | len = vma->vm_end - addr; | |
1916 | ||
1917 | /* only read or write mappings where it is permitted */ | |
d00c7b99 | 1918 | if (write && vma->vm_flags & VM_MAYWRITE) |
0ec76a11 | 1919 | len -= copy_to_user((void *) addr, buf, len); |
d00c7b99 | 1920 | else if (!write && vma->vm_flags & VM_MAYREAD) |
0ec76a11 DH |
1921 | len -= copy_from_user(buf, (void *) addr, len); |
1922 | else | |
1923 | len = 0; | |
1924 | } else { | |
1925 | len = 0; | |
1926 | } | |
1927 | ||
1928 | up_read(&mm->mmap_sem); | |
1929 | mmput(mm); | |
1930 | return len; | |
1931 | } |