ashmem.c revision 11980c2ac4ccfad21a5f8ee9e12059f1e687bb40
1/* mm/ashmem.c 2** 3** Anonymous Shared Memory Subsystem, ashmem 4** 5** Copyright (C) 2008 Google, Inc. 6** 7** Robert Love <rlove@google.com> 8** 9** This software is licensed under the terms of the GNU General Public 10** License version 2, as published by the Free Software Foundation, and 11** may be copied, distributed, and modified under those terms. 12** 13** This program is distributed in the hope that it will be useful, 14** but WITHOUT ANY WARRANTY; without even the implied warranty of 15** MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 16** GNU General Public License for more details. 17*/ 18 19#include <linux/module.h> 20#include <linux/file.h> 21#include <linux/fs.h> 22#include <linux/miscdevice.h> 23#include <linux/security.h> 24#include <linux/mm.h> 25#include <linux/mman.h> 26#include <linux/uaccess.h> 27#include <linux/personality.h> 28#include <linux/bitops.h> 29#include <linux/mutex.h> 30#include <linux/shmem_fs.h> 31#include "ashmem.h" 32 33#define ASHMEM_NAME_PREFIX "dev/ashmem/" 34#define ASHMEM_NAME_PREFIX_LEN (sizeof(ASHMEM_NAME_PREFIX) - 1) 35#define ASHMEM_FULL_NAME_LEN (ASHMEM_NAME_LEN + ASHMEM_NAME_PREFIX_LEN) 36 37/* 38 * ashmem_area - anonymous shared memory area 39 * Lifecycle: From our parent file's open() until its release() 40 * Locking: Protected by `ashmem_mutex' 41 * Big Note: Mappings do NOT pin this structure; it dies on close() 42 */ 43struct ashmem_area { 44 char name[ASHMEM_FULL_NAME_LEN];/* optional name for /proc/pid/maps */ 45 struct list_head unpinned_list; /* list of all ashmem areas */ 46 struct file *file; /* the shmem-based backing file */ 47 size_t size; /* size of the mapping, in bytes */ 48 unsigned long prot_mask; /* allowed prot bits, as vm_flags */ 49}; 50 51/* 52 * ashmem_range - represents an interval of unpinned (evictable) pages 53 * Lifecycle: From unpin to pin 54 * Locking: Protected by `ashmem_mutex' 55 */ 56struct ashmem_range { 57 struct list_head lru; /* entry in LRU list */ 58 struct list_head unpinned; /* entry in its area's unpinned list */ 59 struct ashmem_area *asma; /* associated area */ 60 size_t pgstart; /* starting page, inclusive */ 61 size_t pgend; /* ending page, inclusive */ 62 unsigned int purged; /* ASHMEM_NOT or ASHMEM_WAS_PURGED */ 63}; 64 65/* LRU list of unpinned pages, protected by ashmem_mutex */ 66static LIST_HEAD(ashmem_lru_list); 67 68/* Count of pages on our LRU list, protected by ashmem_mutex */ 69static unsigned long lru_count; 70 71/* 72 * ashmem_mutex - protects the list of and each individual ashmem_area 73 * 74 * Lock Ordering: ashmex_mutex -> i_mutex -> i_alloc_sem 75 */ 76static DEFINE_MUTEX(ashmem_mutex); 77 78static struct kmem_cache *ashmem_area_cachep __read_mostly; 79static struct kmem_cache *ashmem_range_cachep __read_mostly; 80 81#define range_size(range) \ 82 ((range)->pgend - (range)->pgstart + 1) 83 84#define range_on_lru(range) \ 85 ((range)->purged == ASHMEM_NOT_PURGED) 86 87#define page_range_subsumes_range(range, start, end) \ 88 (((range)->pgstart >= (start)) && ((range)->pgend <= (end))) 89 90#define page_range_subsumed_by_range(range, start, end) \ 91 (((range)->pgstart <= (start)) && ((range)->pgend >= (end))) 92 93#define page_in_range(range, page) \ 94 (((range)->pgstart <= (page)) && ((range)->pgend >= (page))) 95 96#define page_range_in_range(range, start, end) \ 97 (page_in_range(range, start) || page_in_range(range, end) || \ 98 page_range_subsumes_range(range, start, end)) 99 100#define range_before_page(range, page) \ 101 ((range)->pgend < (page)) 102 103#define PROT_MASK (PROT_EXEC | PROT_READ | PROT_WRITE) 104 105static inline void lru_add(struct ashmem_range *range) 106{ 107 list_add_tail(&range->lru, &ashmem_lru_list); 108 lru_count += range_size(range); 109} 110 111static inline void lru_del(struct ashmem_range *range) 112{ 113 list_del(&range->lru); 114 lru_count -= range_size(range); 115} 116 117/* 118 * range_alloc - allocate and initialize a new ashmem_range structure 119 * 120 * 'asma' - associated ashmem_area 121 * 'prev_range' - the previous ashmem_range in the sorted asma->unpinned list 122 * 'purged' - initial purge value (ASMEM_NOT_PURGED or ASHMEM_WAS_PURGED) 123 * 'start' - starting page, inclusive 124 * 'end' - ending page, inclusive 125 * 126 * Caller must hold ashmem_mutex. 127 */ 128static int range_alloc(struct ashmem_area *asma, 129 struct ashmem_range *prev_range, unsigned int purged, 130 size_t start, size_t end) 131{ 132 struct ashmem_range *range; 133 134 range = kmem_cache_zalloc(ashmem_range_cachep, GFP_KERNEL); 135 if (unlikely(!range)) 136 return -ENOMEM; 137 138 range->asma = asma; 139 range->pgstart = start; 140 range->pgend = end; 141 range->purged = purged; 142 143 list_add_tail(&range->unpinned, &prev_range->unpinned); 144 145 if (range_on_lru(range)) 146 lru_add(range); 147 148 return 0; 149} 150 151static void range_del(struct ashmem_range *range) 152{ 153 list_del(&range->unpinned); 154 if (range_on_lru(range)) 155 lru_del(range); 156 kmem_cache_free(ashmem_range_cachep, range); 157} 158 159/* 160 * range_shrink - shrinks a range 161 * 162 * Caller must hold ashmem_mutex. 163 */ 164static inline void range_shrink(struct ashmem_range *range, 165 size_t start, size_t end) 166{ 167 size_t pre = range_size(range); 168 169 range->pgstart = start; 170 range->pgend = end; 171 172 if (range_on_lru(range)) 173 lru_count -= pre - range_size(range); 174} 175 176static int ashmem_open(struct inode *inode, struct file *file) 177{ 178 struct ashmem_area *asma; 179 int ret; 180 181 ret = nonseekable_open(inode, file); 182 if (unlikely(ret)) 183 return ret; 184 185 asma = kmem_cache_zalloc(ashmem_area_cachep, GFP_KERNEL); 186 if (unlikely(!asma)) 187 return -ENOMEM; 188 189 INIT_LIST_HEAD(&asma->unpinned_list); 190 memcpy(asma->name, ASHMEM_NAME_PREFIX, ASHMEM_NAME_PREFIX_LEN); 191 asma->prot_mask = PROT_MASK; 192 file->private_data = asma; 193 194 return 0; 195} 196 197static int ashmem_release(struct inode *ignored, struct file *file) 198{ 199 struct ashmem_area *asma = file->private_data; 200 struct ashmem_range *range, *next; 201 202 mutex_lock(&ashmem_mutex); 203 list_for_each_entry_safe(range, next, &asma->unpinned_list, unpinned) 204 range_del(range); 205 mutex_unlock(&ashmem_mutex); 206 207 if (asma->file) 208 fput(asma->file); 209 kmem_cache_free(ashmem_area_cachep, asma); 210 211 return 0; 212} 213 214static int ashmem_mmap(struct file *file, struct vm_area_struct *vma) 215{ 216 struct ashmem_area *asma = file->private_data; 217 int ret = 0; 218 219 mutex_lock(&ashmem_mutex); 220 221 /* user needs to SET_SIZE before mapping */ 222 if (unlikely(!asma->size)) { 223 ret = -EINVAL; 224 goto out; 225 } 226 227 /* requested protection bits must match our allowed protection mask */ 228 if (unlikely((vma->vm_flags & ~asma->prot_mask) & PROT_MASK)) { 229 ret = -EPERM; 230 goto out; 231 } 232 233 if (!asma->file) { 234 char *name = ASHMEM_NAME_DEF; 235 struct file *vmfile; 236 237 if (asma->name[ASHMEM_NAME_PREFIX_LEN] != '\0') 238 name = asma->name; 239 240 /* ... and allocate the backing shmem file */ 241 vmfile = shmem_file_setup(name, asma->size, vma->vm_flags); 242 if (unlikely(IS_ERR(vmfile))) { 243 ret = PTR_ERR(vmfile); 244 goto out; 245 } 246 asma->file = vmfile; 247 } 248 get_file(asma->file); 249 250 /* 251 * XXX - Reworked to use shmem_zero_setup() instead of 252 * shmem_set_file while we're in staging. -jstultz 253 */ 254 if (vma->vm_flags & VM_SHARED) { 255 ret = shmem_zero_setup(vma); 256 if (ret) { 257 fput(asma->file); 258 goto out; 259 } 260 } 261 262 if (vma->vm_file) 263 fput(vma->vm_file); 264 vma->vm_file = asma->file; 265 vma->vm_flags |= VM_CAN_NONLINEAR; 266 267out: 268 mutex_unlock(&ashmem_mutex); 269 return ret; 270} 271 272/* 273 * ashmem_shrink - our cache shrinker, called from mm/vmscan.c :: shrink_slab 274 * 275 * 'nr_to_scan' is the number of objects (pages) to prune, or 0 to query how 276 * many objects (pages) we have in total. 277 * 278 * 'gfp_mask' is the mask of the allocation that got us into this mess. 279 * 280 * Return value is the number of objects (pages) remaining, or -1 if we cannot 281 * proceed without risk of deadlock (due to gfp_mask). 282 * 283 * We approximate LRU via least-recently-unpinned, jettisoning unpinned partial 284 * chunks of ashmem regions LRU-wise one-at-a-time until we hit 'nr_to_scan' 285 * pages freed. 286 */ 287static int ashmem_shrink(int nr_to_scan, gfp_t gfp_mask) 288{ 289 struct ashmem_range *range, *next; 290 291 /* We might recurse into filesystem code, so bail out if necessary */ 292 if (nr_to_scan && !(gfp_mask & __GFP_FS)) 293 return -1; 294 if (!nr_to_scan) 295 return lru_count; 296 297 mutex_lock(&ashmem_mutex); 298 list_for_each_entry_safe(range, next, &ashmem_lru_list, lru) { 299 struct inode *inode = range->asma->file->f_dentry->d_inode; 300 loff_t start = range->pgstart * PAGE_SIZE; 301 loff_t end = (range->pgend + 1) * PAGE_SIZE - 1; 302 303 vmtruncate_range(inode, start, end); 304 range->purged = ASHMEM_WAS_PURGED; 305 lru_del(range); 306 307 nr_to_scan -= range_size(range); 308 if (nr_to_scan <= 0) 309 break; 310 } 311 mutex_unlock(&ashmem_mutex); 312 313 return lru_count; 314} 315 316static struct shrinker ashmem_shrinker = { 317 .shrink = ashmem_shrink, 318 .seeks = DEFAULT_SEEKS * 4, 319}; 320 321static int set_prot_mask(struct ashmem_area *asma, unsigned long prot) 322{ 323 int ret = 0; 324 325 mutex_lock(&ashmem_mutex); 326 327 /* the user can only remove, not add, protection bits */ 328 if (unlikely((asma->prot_mask & prot) != prot)) { 329 ret = -EINVAL; 330 goto out; 331 } 332 333 /* does the application expect PROT_READ to imply PROT_EXEC? */ 334 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) 335 prot |= PROT_EXEC; 336 337 asma->prot_mask = prot; 338 339out: 340 mutex_unlock(&ashmem_mutex); 341 return ret; 342} 343 344static int set_name(struct ashmem_area *asma, void __user *name) 345{ 346 int ret = 0; 347 348 mutex_lock(&ashmem_mutex); 349 350 /* cannot change an existing mapping's name */ 351 if (unlikely(asma->file)) { 352 ret = -EINVAL; 353 goto out; 354 } 355 356 if (unlikely(copy_from_user(asma->name + ASHMEM_NAME_PREFIX_LEN, 357 name, ASHMEM_NAME_LEN))) 358 ret = -EFAULT; 359 asma->name[ASHMEM_FULL_NAME_LEN-1] = '\0'; 360 361out: 362 mutex_unlock(&ashmem_mutex); 363 364 return ret; 365} 366 367static int get_name(struct ashmem_area *asma, void __user *name) 368{ 369 int ret = 0; 370 371 mutex_lock(&ashmem_mutex); 372 if (asma->name[ASHMEM_NAME_PREFIX_LEN] != '\0') { 373 size_t len; 374 375 /* 376 * Copying only `len', instead of ASHMEM_NAME_LEN, bytes 377 * prevents us from revealing one user's stack to another. 378 */ 379 len = strlen(asma->name + ASHMEM_NAME_PREFIX_LEN) + 1; 380 if (unlikely(copy_to_user(name, 381 asma->name + ASHMEM_NAME_PREFIX_LEN, len))) 382 ret = -EFAULT; 383 } else { 384 if (unlikely(copy_to_user(name, ASHMEM_NAME_DEF, 385 sizeof(ASHMEM_NAME_DEF)))) 386 ret = -EFAULT; 387 } 388 mutex_unlock(&ashmem_mutex); 389 390 return ret; 391} 392 393/* 394 * ashmem_pin - pin the given ashmem region, returning whether it was 395 * previously purged (ASHMEM_WAS_PURGED) or not (ASHMEM_NOT_PURGED). 396 * 397 * Caller must hold ashmem_mutex. 398 */ 399static int ashmem_pin(struct ashmem_area *asma, size_t pgstart, size_t pgend) 400{ 401 struct ashmem_range *range, *next; 402 int ret = ASHMEM_NOT_PURGED; 403 404 list_for_each_entry_safe(range, next, &asma->unpinned_list, unpinned) { 405 /* moved past last applicable page; we can short circuit */ 406 if (range_before_page(range, pgstart)) 407 break; 408 409 /* 410 * The user can ask us to pin pages that span multiple ranges, 411 * or to pin pages that aren't even unpinned, so this is messy. 412 * 413 * Four cases: 414 * 1. The requested range subsumes an existing range, so we 415 * just remove the entire matching range. 416 * 2. The requested range overlaps the start of an existing 417 * range, so we just update that range. 418 * 3. The requested range overlaps the end of an existing 419 * range, so we just update that range. 420 * 4. The requested range punches a hole in an existing range, 421 * so we have to update one side of the range and then 422 * create a new range for the other side. 423 */ 424 if (page_range_in_range(range, pgstart, pgend)) { 425 ret |= range->purged; 426 427 /* Case #1: Easy. Just nuke the whole thing. */ 428 if (page_range_subsumes_range(range, pgstart, pgend)) { 429 range_del(range); 430 continue; 431 } 432 433 /* Case #2: We overlap from the start, so adjust it */ 434 if (range->pgstart >= pgstart) { 435 range_shrink(range, pgend + 1, range->pgend); 436 continue; 437 } 438 439 /* Case #3: We overlap from the rear, so adjust it */ 440 if (range->pgend <= pgend) { 441 range_shrink(range, range->pgstart, pgstart-1); 442 continue; 443 } 444 445 /* 446 * Case #4: We eat a chunk out of the middle. A bit 447 * more complicated, we allocate a new range for the 448 * second half and adjust the first chunk's endpoint. 449 */ 450 range_alloc(asma, range, range->purged, 451 pgend + 1, range->pgend); 452 range_shrink(range, range->pgstart, pgstart - 1); 453 break; 454 } 455 } 456 457 return ret; 458} 459 460/* 461 * ashmem_unpin - unpin the given range of pages. Returns zero on success. 462 * 463 * Caller must hold ashmem_mutex. 464 */ 465static int ashmem_unpin(struct ashmem_area *asma, size_t pgstart, size_t pgend) 466{ 467 struct ashmem_range *range, *next; 468 unsigned int purged = ASHMEM_NOT_PURGED; 469 470restart: 471 list_for_each_entry_safe(range, next, &asma->unpinned_list, unpinned) { 472 /* short circuit: this is our insertion point */ 473 if (range_before_page(range, pgstart)) 474 break; 475 476 /* 477 * The user can ask us to unpin pages that are already entirely 478 * or partially pinned. We handle those two cases here. 479 */ 480 if (page_range_subsumed_by_range(range, pgstart, pgend)) 481 return 0; 482 if (page_range_in_range(range, pgstart, pgend)) { 483 pgstart = min_t(size_t, range->pgstart, pgstart), 484 pgend = max_t(size_t, range->pgend, pgend); 485 purged |= range->purged; 486 range_del(range); 487 goto restart; 488 } 489 } 490 491 return range_alloc(asma, range, purged, pgstart, pgend); 492} 493 494/* 495 * ashmem_get_pin_status - Returns ASHMEM_IS_UNPINNED if _any_ pages in the 496 * given interval are unpinned and ASHMEM_IS_PINNED otherwise. 497 * 498 * Caller must hold ashmem_mutex. 499 */ 500static int ashmem_get_pin_status(struct ashmem_area *asma, size_t pgstart, 501 size_t pgend) 502{ 503 struct ashmem_range *range; 504 int ret = ASHMEM_IS_PINNED; 505 506 list_for_each_entry(range, &asma->unpinned_list, unpinned) { 507 if (range_before_page(range, pgstart)) 508 break; 509 if (page_range_in_range(range, pgstart, pgend)) { 510 ret = ASHMEM_IS_UNPINNED; 511 break; 512 } 513 } 514 515 return ret; 516} 517 518static int ashmem_pin_unpin(struct ashmem_area *asma, unsigned long cmd, 519 void __user *p) 520{ 521 struct ashmem_pin pin; 522 size_t pgstart, pgend; 523 int ret = -EINVAL; 524 525 if (unlikely(!asma->file)) 526 return -EINVAL; 527 528 if (unlikely(copy_from_user(&pin, p, sizeof(pin)))) 529 return -EFAULT; 530 531 /* per custom, you can pass zero for len to mean "everything onward" */ 532 if (!pin.len) 533 pin.len = PAGE_ALIGN(asma->size) - pin.offset; 534 535 if (unlikely((pin.offset | pin.len) & ~PAGE_MASK)) 536 return -EINVAL; 537 538 if (unlikely(((__u32) -1) - pin.offset < pin.len)) 539 return -EINVAL; 540 541 if (unlikely(PAGE_ALIGN(asma->size) < pin.offset + pin.len)) 542 return -EINVAL; 543 544 pgstart = pin.offset / PAGE_SIZE; 545 pgend = pgstart + (pin.len / PAGE_SIZE) - 1; 546 547 mutex_lock(&ashmem_mutex); 548 549 switch (cmd) { 550 case ASHMEM_PIN: 551 ret = ashmem_pin(asma, pgstart, pgend); 552 break; 553 case ASHMEM_UNPIN: 554 ret = ashmem_unpin(asma, pgstart, pgend); 555 break; 556 case ASHMEM_GET_PIN_STATUS: 557 ret = ashmem_get_pin_status(asma, pgstart, pgend); 558 break; 559 } 560 561 mutex_unlock(&ashmem_mutex); 562 563 return ret; 564} 565 566static long ashmem_ioctl(struct file *file, unsigned int cmd, unsigned long arg) 567{ 568 struct ashmem_area *asma = file->private_data; 569 long ret = -ENOTTY; 570 571 switch (cmd) { 572 case ASHMEM_SET_NAME: 573 ret = set_name(asma, (void __user *) arg); 574 break; 575 case ASHMEM_GET_NAME: 576 ret = get_name(asma, (void __user *) arg); 577 break; 578 case ASHMEM_SET_SIZE: 579 ret = -EINVAL; 580 if (!asma->file) { 581 ret = 0; 582 asma->size = (size_t) arg; 583 } 584 break; 585 case ASHMEM_GET_SIZE: 586 ret = asma->size; 587 break; 588 case ASHMEM_SET_PROT_MASK: 589 ret = set_prot_mask(asma, arg); 590 break; 591 case ASHMEM_GET_PROT_MASK: 592 ret = asma->prot_mask; 593 break; 594 case ASHMEM_PIN: 595 case ASHMEM_UNPIN: 596 case ASHMEM_GET_PIN_STATUS: 597 ret = ashmem_pin_unpin(asma, cmd, (void __user *) arg); 598 break; 599 case ASHMEM_PURGE_ALL_CACHES: 600 ret = -EPERM; 601 if (capable(CAP_SYS_ADMIN)) { 602 ret = ashmem_shrink(0, GFP_KERNEL); 603 ashmem_shrink(ret, GFP_KERNEL); 604 } 605 break; 606 } 607 608 return ret; 609} 610 611static struct file_operations ashmem_fops = { 612 .owner = THIS_MODULE, 613 .open = ashmem_open, 614 .release = ashmem_release, 615 .mmap = ashmem_mmap, 616 .unlocked_ioctl = ashmem_ioctl, 617 .compat_ioctl = ashmem_ioctl, 618}; 619 620static struct miscdevice ashmem_misc = { 621 .minor = MISC_DYNAMIC_MINOR, 622 .name = "ashmem", 623 .fops = &ashmem_fops, 624}; 625 626static int __init ashmem_init(void) 627{ 628 int ret; 629 630 ashmem_area_cachep = kmem_cache_create("ashmem_area_cache", 631 sizeof(struct ashmem_area), 632 0, 0, NULL); 633 if (unlikely(!ashmem_area_cachep)) { 634 printk(KERN_ERR "ashmem: failed to create slab cache\n"); 635 return -ENOMEM; 636 } 637 638 ashmem_range_cachep = kmem_cache_create("ashmem_range_cache", 639 sizeof(struct ashmem_range), 640 0, 0, NULL); 641 if (unlikely(!ashmem_range_cachep)) { 642 printk(KERN_ERR "ashmem: failed to create slab cache\n"); 643 return -ENOMEM; 644 } 645 646 ret = misc_register(&ashmem_misc); 647 if (unlikely(ret)) { 648 printk(KERN_ERR "ashmem: failed to register misc device!\n"); 649 return ret; 650 } 651 652 register_shrinker(&ashmem_shrinker); 653 654 printk(KERN_INFO "ashmem: initialized\n"); 655 656 return 0; 657} 658 659static void __exit ashmem_exit(void) 660{ 661 int ret; 662 663 unregister_shrinker(&ashmem_shrinker); 664 665 ret = misc_deregister(&ashmem_misc); 666 if (unlikely(ret)) 667 printk(KERN_ERR "ashmem: failed to unregister misc device!\n"); 668 669 kmem_cache_destroy(ashmem_range_cachep); 670 kmem_cache_destroy(ashmem_area_cachep); 671 672 printk(KERN_INFO "ashmem: unloaded\n"); 673} 674 675module_init(ashmem_init); 676module_exit(ashmem_exit); 677 678MODULE_LICENSE("GPL"); 679