1 /* SPDX-License-Identifier: GPL-2.0+ */ 2 3 #pragma once 4 5 /* Forward declarations to avoid header cycle. */ 6 struct vm_area_struct; 7 static inline void vma_start_write(struct vm_area_struct *vma); 8 9 extern const struct vm_operations_struct vma_dummy_vm_ops; 10 extern unsigned long stack_guard_gap; 11 extern const struct vm_operations_struct vma_dummy_vm_ops; 12 extern unsigned long rlimit(unsigned int limit); 13 struct task_struct *get_current(void); 14 15 #define MMF_HAS_MDWE 28 16 #define current get_current() 17 18 /* 19 * Define the task command name length as enum, then it can be visible to 20 * BPF programs. 21 */ 22 enum { 23 TASK_COMM_LEN = 16, 24 }; 25 26 /* PARTIALLY implemented types. */ 27 struct mm_struct { 28 struct maple_tree mm_mt; 29 int map_count; /* number of VMAs */ 30 unsigned long total_vm; /* Total pages mapped */ 31 unsigned long locked_vm; /* Pages that have PG_mlocked set */ 32 unsigned long data_vm; /* VM_WRITE & ~VM_SHARED & ~VM_STACK */ 33 unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE & ~VM_STACK */ 34 unsigned long stack_vm; /* VM_STACK */ 35 36 union { 37 vm_flags_t def_flags; 38 vma_flags_t def_vma_flags; 39 }; 40 41 mm_flags_t flags; /* Must use mm_flags_* helpers to access */ 42 }; 43 struct address_space { 44 struct rb_root_cached i_mmap; 45 unsigned long flags; 46 atomic_t i_mmap_writable; 47 }; 48 struct file_operations { 49 int (*mmap)(struct file *, struct vm_area_struct *); 50 int (*mmap_prepare)(struct vm_area_desc *); 51 }; 52 struct file { 53 struct address_space *f_mapping; 54 const struct file_operations *f_op; 55 }; 56 struct anon_vma_chain { 57 struct anon_vma *anon_vma; 58 struct list_head same_vma; 59 }; 60 struct task_struct { 61 char comm[TASK_COMM_LEN]; 62 pid_t pid; 63 struct mm_struct *mm; 64 65 /* Used for emulating ABI behavior of previous Linux versions: */ 66 unsigned int personality; 67 }; 68 69 struct kref { 70 refcount_t refcount; 71 }; 72 73 struct anon_vma_name { 74 struct kref kref; 75 /* The name needs to be at the end because it is dynamically sized. */ 76 char name[]; 77 }; 78 79 /* 80 * Contains declarations that are DUPLICATED from kernel source in order to 81 * faciliate userland VMA testing. 82 * 83 * These must be kept in sync with kernel source. 84 */ 85 86 #define VMA_LOCK_OFFSET 0x40000000 87 88 typedef struct { unsigned long v; } freeptr_t; 89 90 #define VM_NONE 0x00000000 91 92 typedef int __bitwise vma_flag_t; 93 94 #define ACCESS_PRIVATE(p, member) ((p)->member) 95 96 #define DECLARE_VMA_BIT(name, bitnum) \ 97 VMA_ ## name ## _BIT = ((__force vma_flag_t)bitnum) 98 #define DECLARE_VMA_BIT_ALIAS(name, aliased) \ 99 VMA_ ## name ## _BIT = VMA_ ## aliased ## _BIT 100 enum { 101 DECLARE_VMA_BIT(READ, 0), 102 DECLARE_VMA_BIT(WRITE, 1), 103 DECLARE_VMA_BIT(EXEC, 2), 104 DECLARE_VMA_BIT(SHARED, 3), 105 /* mprotect() hardcodes VM_MAYREAD >> 4 == VM_READ, and so for r/w/x bits. */ 106 DECLARE_VMA_BIT(MAYREAD, 4), /* limits for mprotect() etc. */ 107 DECLARE_VMA_BIT(MAYWRITE, 5), 108 DECLARE_VMA_BIT(MAYEXEC, 6), 109 DECLARE_VMA_BIT(MAYSHARE, 7), 110 DECLARE_VMA_BIT(GROWSDOWN, 8), /* general info on the segment */ 111 #ifdef CONFIG_MMU 112 DECLARE_VMA_BIT(UFFD_MISSING, 9),/* missing pages tracking */ 113 #else 114 /* nommu: R/O MAP_PRIVATE mapping that might overlay a file mapping */ 115 DECLARE_VMA_BIT(MAYOVERLAY, 9), 116 #endif /* CONFIG_MMU */ 117 /* Page-ranges managed without "struct page", just pure PFN */ 118 DECLARE_VMA_BIT(PFNMAP, 10), 119 DECLARE_VMA_BIT(MAYBE_GUARD, 11), 120 DECLARE_VMA_BIT(UFFD_WP, 12), /* wrprotect pages tracking */ 121 DECLARE_VMA_BIT(LOCKED, 13), 122 DECLARE_VMA_BIT(IO, 14), /* Memory mapped I/O or similar */ 123 DECLARE_VMA_BIT(SEQ_READ, 15), /* App will access data sequentially */ 124 DECLARE_VMA_BIT(RAND_READ, 16), /* App will not benefit from clustered reads */ 125 DECLARE_VMA_BIT(DONTCOPY, 17), /* Do not copy this vma on fork */ 126 DECLARE_VMA_BIT(DONTEXPAND, 18),/* Cannot expand with mremap() */ 127 DECLARE_VMA_BIT(LOCKONFAULT, 19),/* Lock pages covered when faulted in */ 128 DECLARE_VMA_BIT(ACCOUNT, 20), /* Is a VM accounted object */ 129 DECLARE_VMA_BIT(NORESERVE, 21), /* should the VM suppress accounting */ 130 DECLARE_VMA_BIT(HUGETLB, 22), /* Huge TLB Page VM */ 131 DECLARE_VMA_BIT(SYNC, 23), /* Synchronous page faults */ 132 DECLARE_VMA_BIT(ARCH_1, 24), /* Architecture-specific flag */ 133 DECLARE_VMA_BIT(WIPEONFORK, 25),/* Wipe VMA contents in child. */ 134 DECLARE_VMA_BIT(DONTDUMP, 26), /* Do not include in the core dump */ 135 DECLARE_VMA_BIT(SOFTDIRTY, 27), /* NOT soft dirty clean area */ 136 DECLARE_VMA_BIT(MIXEDMAP, 28), /* Can contain struct page and pure PFN pages */ 137 DECLARE_VMA_BIT(HUGEPAGE, 29), /* MADV_HUGEPAGE marked this vma */ 138 DECLARE_VMA_BIT(NOHUGEPAGE, 30),/* MADV_NOHUGEPAGE marked this vma */ 139 DECLARE_VMA_BIT(MERGEABLE, 31), /* KSM may merge identical pages */ 140 /* These bits are reused, we define specific uses below. */ 141 DECLARE_VMA_BIT(HIGH_ARCH_0, 32), 142 DECLARE_VMA_BIT(HIGH_ARCH_1, 33), 143 DECLARE_VMA_BIT(HIGH_ARCH_2, 34), 144 DECLARE_VMA_BIT(HIGH_ARCH_3, 35), 145 DECLARE_VMA_BIT(HIGH_ARCH_4, 36), 146 DECLARE_VMA_BIT(HIGH_ARCH_5, 37), 147 DECLARE_VMA_BIT(HIGH_ARCH_6, 38), 148 /* 149 * This flag is used to connect VFIO to arch specific KVM code. It 150 * indicates that the memory under this VMA is safe for use with any 151 * non-cachable memory type inside KVM. Some VFIO devices, on some 152 * platforms, are thought to be unsafe and can cause machine crashes 153 * if KVM does not lock down the memory type. 154 */ 155 DECLARE_VMA_BIT(ALLOW_ANY_UNCACHED, 39), 156 #ifdef CONFIG_PPC32 157 DECLARE_VMA_BIT_ALIAS(DROPPABLE, ARCH_1), 158 #else 159 DECLARE_VMA_BIT(DROPPABLE, 40), 160 #endif 161 DECLARE_VMA_BIT(UFFD_MINOR, 41), 162 DECLARE_VMA_BIT(SEALED, 42), 163 /* Flags that reuse flags above. */ 164 DECLARE_VMA_BIT_ALIAS(PKEY_BIT0, HIGH_ARCH_0), 165 DECLARE_VMA_BIT_ALIAS(PKEY_BIT1, HIGH_ARCH_1), 166 DECLARE_VMA_BIT_ALIAS(PKEY_BIT2, HIGH_ARCH_2), 167 DECLARE_VMA_BIT_ALIAS(PKEY_BIT3, HIGH_ARCH_3), 168 DECLARE_VMA_BIT_ALIAS(PKEY_BIT4, HIGH_ARCH_4), 169 #if defined(CONFIG_X86_USER_SHADOW_STACK) 170 /* 171 * VM_SHADOW_STACK should not be set with VM_SHARED because of lack of 172 * support core mm. 173 * 174 * These VMAs will get a single end guard page. This helps userspace 175 * protect itself from attacks. A single page is enough for current 176 * shadow stack archs (x86). See the comments near alloc_shstk() in 177 * arch/x86/kernel/shstk.c for more details on the guard size. 178 */ 179 DECLARE_VMA_BIT_ALIAS(SHADOW_STACK, HIGH_ARCH_5), 180 #elif defined(CONFIG_ARM64_GCS) 181 /* 182 * arm64's Guarded Control Stack implements similar functionality and 183 * has similar constraints to shadow stacks. 184 */ 185 DECLARE_VMA_BIT_ALIAS(SHADOW_STACK, HIGH_ARCH_6), 186 #endif 187 DECLARE_VMA_BIT_ALIAS(SAO, ARCH_1), /* Strong Access Ordering (powerpc) */ 188 DECLARE_VMA_BIT_ALIAS(GROWSUP, ARCH_1), /* parisc */ 189 DECLARE_VMA_BIT_ALIAS(SPARC_ADI, ARCH_1), /* sparc64 */ 190 DECLARE_VMA_BIT_ALIAS(ARM64_BTI, ARCH_1), /* arm64 */ 191 DECLARE_VMA_BIT_ALIAS(ARCH_CLEAR, ARCH_1), /* sparc64, arm64 */ 192 DECLARE_VMA_BIT_ALIAS(MAPPED_COPY, ARCH_1), /* !CONFIG_MMU */ 193 DECLARE_VMA_BIT_ALIAS(MTE, HIGH_ARCH_4), /* arm64 */ 194 DECLARE_VMA_BIT_ALIAS(MTE_ALLOWED, HIGH_ARCH_5),/* arm64 */ 195 #ifdef CONFIG_STACK_GROWSUP 196 DECLARE_VMA_BIT_ALIAS(STACK, GROWSUP), 197 DECLARE_VMA_BIT_ALIAS(STACK_EARLY, GROWSDOWN), 198 #else 199 DECLARE_VMA_BIT_ALIAS(STACK, GROWSDOWN), 200 #endif 201 }; 202 203 #define INIT_VM_FLAG(name) BIT((__force int) VMA_ ## name ## _BIT) 204 #define VM_READ INIT_VM_FLAG(READ) 205 #define VM_WRITE INIT_VM_FLAG(WRITE) 206 #define VM_EXEC INIT_VM_FLAG(EXEC) 207 #define VM_SHARED INIT_VM_FLAG(SHARED) 208 #define VM_MAYREAD INIT_VM_FLAG(MAYREAD) 209 #define VM_MAYWRITE INIT_VM_FLAG(MAYWRITE) 210 #define VM_MAYEXEC INIT_VM_FLAG(MAYEXEC) 211 #define VM_MAYSHARE INIT_VM_FLAG(MAYSHARE) 212 #define VM_GROWSDOWN INIT_VM_FLAG(GROWSDOWN) 213 #ifdef CONFIG_MMU 214 #define VM_UFFD_MISSING INIT_VM_FLAG(UFFD_MISSING) 215 #else 216 #define VM_UFFD_MISSING VM_NONE 217 #define VM_MAYOVERLAY INIT_VM_FLAG(MAYOVERLAY) 218 #endif 219 #define VM_PFNMAP INIT_VM_FLAG(PFNMAP) 220 #define VM_MAYBE_GUARD INIT_VM_FLAG(MAYBE_GUARD) 221 #define VM_UFFD_WP INIT_VM_FLAG(UFFD_WP) 222 #define VM_LOCKED INIT_VM_FLAG(LOCKED) 223 #define VM_IO INIT_VM_FLAG(IO) 224 #define VM_SEQ_READ INIT_VM_FLAG(SEQ_READ) 225 #define VM_RAND_READ INIT_VM_FLAG(RAND_READ) 226 #define VM_DONTCOPY INIT_VM_FLAG(DONTCOPY) 227 #define VM_DONTEXPAND INIT_VM_FLAG(DONTEXPAND) 228 #define VM_LOCKONFAULT INIT_VM_FLAG(LOCKONFAULT) 229 #define VM_ACCOUNT INIT_VM_FLAG(ACCOUNT) 230 #define VM_NORESERVE INIT_VM_FLAG(NORESERVE) 231 #define VM_HUGETLB INIT_VM_FLAG(HUGETLB) 232 #define VM_SYNC INIT_VM_FLAG(SYNC) 233 #define VM_ARCH_1 INIT_VM_FLAG(ARCH_1) 234 #define VM_WIPEONFORK INIT_VM_FLAG(WIPEONFORK) 235 #define VM_DONTDUMP INIT_VM_FLAG(DONTDUMP) 236 #ifdef CONFIG_MEM_SOFT_DIRTY 237 #define VM_SOFTDIRTY INIT_VM_FLAG(SOFTDIRTY) 238 #else 239 #define VM_SOFTDIRTY VM_NONE 240 #endif 241 #define VM_MIXEDMAP INIT_VM_FLAG(MIXEDMAP) 242 #define VM_HUGEPAGE INIT_VM_FLAG(HUGEPAGE) 243 #define VM_NOHUGEPAGE INIT_VM_FLAG(NOHUGEPAGE) 244 #define VM_MERGEABLE INIT_VM_FLAG(MERGEABLE) 245 #define VM_STACK INIT_VM_FLAG(STACK) 246 #ifdef CONFIG_STACK_GROWS_UP 247 #define VM_STACK_EARLY INIT_VM_FLAG(STACK_EARLY) 248 #else 249 #define VM_STACK_EARLY VM_NONE 250 #endif 251 #ifdef CONFIG_ARCH_HAS_PKEYS 252 #define VM_PKEY_SHIFT ((__force int)VMA_HIGH_ARCH_0_BIT) 253 /* Despite the naming, these are FLAGS not bits. */ 254 #define VM_PKEY_BIT0 INIT_VM_FLAG(PKEY_BIT0) 255 #define VM_PKEY_BIT1 INIT_VM_FLAG(PKEY_BIT1) 256 #define VM_PKEY_BIT2 INIT_VM_FLAG(PKEY_BIT2) 257 #if CONFIG_ARCH_PKEY_BITS > 3 258 #define VM_PKEY_BIT3 INIT_VM_FLAG(PKEY_BIT3) 259 #else 260 #define VM_PKEY_BIT3 VM_NONE 261 #endif /* CONFIG_ARCH_PKEY_BITS > 3 */ 262 #if CONFIG_ARCH_PKEY_BITS > 4 263 #define VM_PKEY_BIT4 INIT_VM_FLAG(PKEY_BIT4) 264 #else 265 #define VM_PKEY_BIT4 VM_NONE 266 #endif /* CONFIG_ARCH_PKEY_BITS > 4 */ 267 #endif /* CONFIG_ARCH_HAS_PKEYS */ 268 #if defined(CONFIG_X86_USER_SHADOW_STACK) || defined(CONFIG_ARM64_GCS) 269 #define VM_SHADOW_STACK INIT_VM_FLAG(SHADOW_STACK) 270 #else 271 #define VM_SHADOW_STACK VM_NONE 272 #endif 273 #if defined(CONFIG_PPC64) 274 #define VM_SAO INIT_VM_FLAG(SAO) 275 #elif defined(CONFIG_PARISC) 276 #define VM_GROWSUP INIT_VM_FLAG(GROWSUP) 277 #elif defined(CONFIG_SPARC64) 278 #define VM_SPARC_ADI INIT_VM_FLAG(SPARC_ADI) 279 #define VM_ARCH_CLEAR INIT_VM_FLAG(ARCH_CLEAR) 280 #elif defined(CONFIG_ARM64) 281 #define VM_ARM64_BTI INIT_VM_FLAG(ARM64_BTI) 282 #define VM_ARCH_CLEAR INIT_VM_FLAG(ARCH_CLEAR) 283 #elif !defined(CONFIG_MMU) 284 #define VM_MAPPED_COPY INIT_VM_FLAG(MAPPED_COPY) 285 #endif 286 #ifndef VM_GROWSUP 287 #define VM_GROWSUP VM_NONE 288 #endif 289 #ifdef CONFIG_ARM64_MTE 290 #define VM_MTE INIT_VM_FLAG(MTE) 291 #define VM_MTE_ALLOWED INIT_VM_FLAG(MTE_ALLOWED) 292 #else 293 #define VM_MTE VM_NONE 294 #define VM_MTE_ALLOWED VM_NONE 295 #endif 296 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR 297 #define VM_UFFD_MINOR INIT_VM_FLAG(UFFD_MINOR) 298 #else 299 #define VM_UFFD_MINOR VM_NONE 300 #endif 301 #ifdef CONFIG_64BIT 302 #define VM_ALLOW_ANY_UNCACHED INIT_VM_FLAG(ALLOW_ANY_UNCACHED) 303 #define VM_SEALED INIT_VM_FLAG(SEALED) 304 #else 305 #define VM_ALLOW_ANY_UNCACHED VM_NONE 306 #define VM_SEALED VM_NONE 307 #endif 308 #if defined(CONFIG_64BIT) || defined(CONFIG_PPC32) 309 #define VM_DROPPABLE INIT_VM_FLAG(DROPPABLE) 310 #else 311 #define VM_DROPPABLE VM_NONE 312 #endif 313 314 /* Bits set in the VMA until the stack is in its final location */ 315 #define VM_STACK_INCOMPLETE_SETUP (VM_RAND_READ | VM_SEQ_READ | VM_STACK_EARLY) 316 317 #define TASK_EXEC ((current->personality & READ_IMPLIES_EXEC) ? VM_EXEC : 0) 318 319 /* Common data flag combinations */ 320 #define VM_DATA_FLAGS_TSK_EXEC (VM_READ | VM_WRITE | TASK_EXEC | \ 321 VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC) 322 #define VM_DATA_FLAGS_NON_EXEC (VM_READ | VM_WRITE | VM_MAYREAD | \ 323 VM_MAYWRITE | VM_MAYEXEC) 324 #define VM_DATA_FLAGS_EXEC (VM_READ | VM_WRITE | VM_EXEC | \ 325 VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC) 326 327 #ifndef VM_DATA_DEFAULT_FLAGS /* arch can override this */ 328 #define VM_DATA_DEFAULT_FLAGS VM_DATA_FLAGS_EXEC 329 #endif 330 331 #ifndef VM_STACK_DEFAULT_FLAGS /* arch can override this */ 332 #define VM_STACK_DEFAULT_FLAGS VM_DATA_DEFAULT_FLAGS 333 #endif 334 335 #define VM_STARTGAP_FLAGS (VM_GROWSDOWN | VM_SHADOW_STACK) 336 337 #define VM_STACK_FLAGS (VM_STACK | VM_STACK_DEFAULT_FLAGS | VM_ACCOUNT) 338 339 /* VMA basic access permission flags */ 340 #define VM_ACCESS_FLAGS (VM_READ | VM_WRITE | VM_EXEC) 341 #define VMA_ACCESS_FLAGS mk_vma_flags(VMA_READ_BIT, VMA_WRITE_BIT, VMA_EXEC_BIT) 342 343 /* 344 * Special vmas that are non-mergable, non-mlock()able. 345 */ 346 #define VM_SPECIAL (VM_IO | VM_DONTEXPAND | VM_PFNMAP | VM_MIXEDMAP) 347 348 #define VMA_REMAP_FLAGS mk_vma_flags(VMA_IO_BIT, VMA_PFNMAP_BIT, \ 349 VMA_DONTEXPAND_BIT, VMA_DONTDUMP_BIT) 350 351 #define DEFAULT_MAP_WINDOW ((1UL << 47) - PAGE_SIZE) 352 #define TASK_SIZE_LOW DEFAULT_MAP_WINDOW 353 #define TASK_SIZE_MAX DEFAULT_MAP_WINDOW 354 #define STACK_TOP TASK_SIZE_LOW 355 #define STACK_TOP_MAX TASK_SIZE_MAX 356 357 /* This mask represents all the VMA flag bits used by mlock */ 358 #define VM_LOCKED_MASK (VM_LOCKED | VM_LOCKONFAULT) 359 360 #define TASK_EXEC ((current->personality & READ_IMPLIES_EXEC) ? VM_EXEC : 0) 361 362 #define VM_DATA_FLAGS_TSK_EXEC (VM_READ | VM_WRITE | TASK_EXEC | \ 363 VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC) 364 365 #define RLIMIT_STACK 3 /* max stack size */ 366 #define RLIMIT_MEMLOCK 8 /* max locked-in-memory address space */ 367 368 #define CAP_IPC_LOCK 14 369 370 #ifdef CONFIG_MEM_SOFT_DIRTY 371 #define VMA_STICKY_FLAGS mk_vma_flags(VMA_SOFTDIRTY_BIT, VMA_MAYBE_GUARD_BIT) 372 #else 373 #define VMA_STICKY_FLAGS mk_vma_flags(VMA_MAYBE_GUARD_BIT) 374 #endif 375 376 #define VMA_IGNORE_MERGE_FLAGS VMA_STICKY_FLAGS 377 378 #define VM_COPY_ON_FORK (VM_PFNMAP | VM_MIXEDMAP | VM_UFFD_WP | VM_MAYBE_GUARD) 379 380 #define pgprot_val(x) ((x).pgprot) 381 #define __pgprot(x) ((pgprot_t) { (x) } ) 382 383 #define for_each_vma(__vmi, __vma) \ 384 while (((__vma) = vma_next(&(__vmi))) != NULL) 385 386 /* The MM code likes to work with exclusive end addresses */ 387 #define for_each_vma_range(__vmi, __vma, __end) \ 388 while (((__vma) = vma_find(&(__vmi), (__end))) != NULL) 389 390 #define offset_in_page(p) ((unsigned long)(p) & ~PAGE_MASK) 391 392 #define PHYS_PFN(x) ((unsigned long)((x) >> PAGE_SHIFT)) 393 394 #define test_and_set_bit(nr, addr) __test_and_set_bit(nr, addr) 395 #define test_and_clear_bit(nr, addr) __test_and_clear_bit(nr, addr) 396 397 #define AS_MM_ALL_LOCKS 2 398 399 #define swap(a, b) \ 400 do { typeof(a) __tmp = (a); (a) = (b); (b) = __tmp; } while (0) 401 402 /* 403 * Flags for bug emulation. 404 * 405 * These occupy the top three bytes. 406 */ 407 enum { 408 READ_IMPLIES_EXEC = 0x0400000, 409 }; 410 411 struct vma_iterator { 412 struct ma_state mas; 413 }; 414 415 #define VMA_ITERATOR(name, __mm, __addr) \ 416 struct vma_iterator name = { \ 417 .mas = { \ 418 .tree = &(__mm)->mm_mt, \ 419 .index = __addr, \ 420 .node = NULL, \ 421 .status = ma_start, \ 422 }, \ 423 } 424 425 #define DEFINE_MUTEX(mutexname) \ 426 struct mutex mutexname = {} 427 428 #define DECLARE_BITMAP(name, bits) \ 429 unsigned long name[BITS_TO_LONGS(bits)] 430 431 #define EMPTY_VMA_FLAGS ((vma_flags_t){ }) 432 433 #define MAPCOUNT_ELF_CORE_MARGIN (5) 434 #define DEFAULT_MAX_MAP_COUNT (USHRT_MAX - MAPCOUNT_ELF_CORE_MARGIN) 435 436 static __always_inline bool vma_flags_empty(const vma_flags_t *flags) 437 { 438 const unsigned long *bitmap = flags->__vma_flags; 439 440 return bitmap_empty(bitmap, NUM_VMA_FLAG_BITS); 441 } 442 443 /* What action should be taken after an .mmap_prepare call is complete? */ 444 enum mmap_action_type { 445 MMAP_NOTHING, /* Mapping is complete, no further action. */ 446 MMAP_REMAP_PFN, /* Remap PFN range. */ 447 MMAP_IO_REMAP_PFN, /* I/O remap PFN range. */ 448 }; 449 450 /* 451 * Describes an action an mmap_prepare hook can instruct to be taken to complete 452 * the mapping of a VMA. Specified in vm_area_desc. 453 */ 454 struct mmap_action { 455 union { 456 /* Remap range. */ 457 struct { 458 unsigned long start; 459 unsigned long start_pfn; 460 unsigned long size; 461 pgprot_t pgprot; 462 } remap; 463 }; 464 enum mmap_action_type type; 465 466 /* 467 * If specified, this hook is invoked after the selected action has been 468 * successfully completed. Note that the VMA write lock still held. 469 * 470 * The absolute minimum ought to be done here. 471 * 472 * Returns 0 on success, or an error code. 473 */ 474 int (*success_hook)(const struct vm_area_struct *vma); 475 476 /* 477 * If specified, this hook is invoked when an error occurred when 478 * attempting the selection action. 479 * 480 * The hook can return an error code in order to filter the error, but 481 * it is not valid to clear the error here. 482 */ 483 int (*error_hook)(int err); 484 485 /* 486 * This should be set in rare instances where the operation required 487 * that the rmap should not be able to access the VMA until 488 * completely set up. 489 */ 490 bool hide_from_rmap_until_complete :1; 491 }; 492 493 /* Operations which modify VMAs. */ 494 enum vma_operation { 495 VMA_OP_SPLIT, 496 VMA_OP_MERGE_UNFAULTED, 497 VMA_OP_REMAP, 498 VMA_OP_FORK, 499 }; 500 501 /* 502 * Describes a VMA that is about to be mmap()'ed. Drivers may choose to 503 * manipulate mutable fields which will cause those fields to be updated in the 504 * resultant VMA. 505 * 506 * Helper functions are not required for manipulating any field. 507 */ 508 struct vm_area_desc { 509 /* Immutable state. */ 510 const struct mm_struct *const mm; 511 struct file *const file; /* May vary from vm_file in stacked callers. */ 512 unsigned long start; 513 unsigned long end; 514 515 /* Mutable fields. Populated with initial state. */ 516 pgoff_t pgoff; 517 struct file *vm_file; 518 vma_flags_t vma_flags; 519 pgprot_t page_prot; 520 521 /* Write-only fields. */ 522 const struct vm_operations_struct *vm_ops; 523 void *private_data; 524 525 /* Take further action? */ 526 struct mmap_action action; 527 }; 528 529 struct vm_area_struct { 530 /* The first cache line has the info for VMA tree walking. */ 531 532 union { 533 struct { 534 /* VMA covers [vm_start; vm_end) addresses within mm */ 535 unsigned long vm_start; 536 unsigned long vm_end; 537 }; 538 freeptr_t vm_freeptr; /* Pointer used by SLAB_TYPESAFE_BY_RCU */ 539 }; 540 541 struct mm_struct *vm_mm; /* The address space we belong to. */ 542 pgprot_t vm_page_prot; /* Access permissions of this VMA. */ 543 544 /* 545 * Flags, see mm.h. 546 * To modify use vm_flags_{init|reset|set|clear|mod} functions. 547 */ 548 union { 549 const vm_flags_t vm_flags; 550 vma_flags_t flags; 551 }; 552 553 #ifdef CONFIG_PER_VMA_LOCK 554 /* 555 * Can only be written (using WRITE_ONCE()) while holding both: 556 * - mmap_lock (in write mode) 557 * - vm_refcnt bit at VMA_LOCK_OFFSET is set 558 * Can be read reliably while holding one of: 559 * - mmap_lock (in read or write mode) 560 * - vm_refcnt bit at VMA_LOCK_OFFSET is set or vm_refcnt > 1 561 * Can be read unreliably (using READ_ONCE()) for pessimistic bailout 562 * while holding nothing (except RCU to keep the VMA struct allocated). 563 * 564 * This sequence counter is explicitly allowed to overflow; sequence 565 * counter reuse can only lead to occasional unnecessary use of the 566 * slowpath. 567 */ 568 unsigned int vm_lock_seq; 569 #endif 570 571 /* 572 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma 573 * list, after a COW of one of the file pages. A MAP_SHARED vma 574 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack 575 * or brk vma (with NULL file) can only be in an anon_vma list. 576 */ 577 struct list_head anon_vma_chain; /* Serialized by mmap_lock & 578 * page_table_lock */ 579 struct anon_vma *anon_vma; /* Serialized by page_table_lock */ 580 581 /* Function pointers to deal with this struct. */ 582 const struct vm_operations_struct *vm_ops; 583 584 /* Information about our backing store: */ 585 unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE 586 units */ 587 struct file * vm_file; /* File we map to (can be NULL). */ 588 void * vm_private_data; /* was vm_pte (shared mem) */ 589 590 #ifdef CONFIG_SWAP 591 atomic_long_t swap_readahead_info; 592 #endif 593 #ifndef CONFIG_MMU 594 struct vm_region *vm_region; /* NOMMU mapping region */ 595 #endif 596 #ifdef CONFIG_NUMA 597 struct mempolicy *vm_policy; /* NUMA policy for the VMA */ 598 #endif 599 #ifdef CONFIG_NUMA_BALANCING 600 struct vma_numab_state *numab_state; /* NUMA Balancing state */ 601 #endif 602 #ifdef CONFIG_PER_VMA_LOCK 603 /* Unstable RCU readers are allowed to read this. */ 604 refcount_t vm_refcnt; 605 #endif 606 /* 607 * For areas with an address space and backing store, 608 * linkage into the address_space->i_mmap interval tree. 609 * 610 */ 611 struct { 612 struct rb_node rb; 613 unsigned long rb_subtree_last; 614 } shared; 615 #ifdef CONFIG_ANON_VMA_NAME 616 /* 617 * For private and shared anonymous mappings, a pointer to a null 618 * terminated string containing the name given to the vma, or NULL if 619 * unnamed. Serialized by mmap_lock. Use anon_vma_name to access. 620 */ 621 struct anon_vma_name *anon_name; 622 #endif 623 struct vm_userfaultfd_ctx vm_userfaultfd_ctx; 624 } __randomize_layout; 625 626 struct vm_operations_struct { 627 void (*open)(struct vm_area_struct * area); 628 /** 629 * @close: Called when the VMA is being removed from the MM. 630 * Context: User context. May sleep. Caller holds mmap_lock. 631 */ 632 void (*close)(struct vm_area_struct * area); 633 /* Called any time before splitting to check if it's allowed */ 634 int (*may_split)(struct vm_area_struct *area, unsigned long addr); 635 int (*mremap)(struct vm_area_struct *area); 636 /* 637 * Called by mprotect() to make driver-specific permission 638 * checks before mprotect() is finalised. The VMA must not 639 * be modified. Returns 0 if mprotect() can proceed. 640 */ 641 int (*mprotect)(struct vm_area_struct *vma, unsigned long start, 642 unsigned long end, unsigned long newflags); 643 vm_fault_t (*fault)(struct vm_fault *vmf); 644 vm_fault_t (*huge_fault)(struct vm_fault *vmf, unsigned int order); 645 vm_fault_t (*map_pages)(struct vm_fault *vmf, 646 pgoff_t start_pgoff, pgoff_t end_pgoff); 647 unsigned long (*pagesize)(struct vm_area_struct * area); 648 649 /* notification that a previously read-only page is about to become 650 * writable, if an error is returned it will cause a SIGBUS */ 651 vm_fault_t (*page_mkwrite)(struct vm_fault *vmf); 652 653 /* same as page_mkwrite when using VM_PFNMAP|VM_MIXEDMAP */ 654 vm_fault_t (*pfn_mkwrite)(struct vm_fault *vmf); 655 656 /* called by access_process_vm when get_user_pages() fails, typically 657 * for use by special VMAs. See also generic_access_phys() for a generic 658 * implementation useful for any iomem mapping. 659 */ 660 int (*access)(struct vm_area_struct *vma, unsigned long addr, 661 void *buf, int len, int write); 662 663 /* Called by the /proc/PID/maps code to ask the vma whether it 664 * has a special name. Returning non-NULL will also cause this 665 * vma to be dumped unconditionally. */ 666 const char *(*name)(struct vm_area_struct *vma); 667 668 #ifdef CONFIG_NUMA 669 /* 670 * set_policy() op must add a reference to any non-NULL @new mempolicy 671 * to hold the policy upon return. Caller should pass NULL @new to 672 * remove a policy and fall back to surrounding context--i.e. do not 673 * install a MPOL_DEFAULT policy, nor the task or system default 674 * mempolicy. 675 */ 676 int (*set_policy)(struct vm_area_struct *vma, struct mempolicy *new); 677 678 /* 679 * get_policy() op must add reference [mpol_get()] to any policy at 680 * (vma,addr) marked as MPOL_SHARED. The shared policy infrastructure 681 * in mm/mempolicy.c will do this automatically. 682 * get_policy() must NOT add a ref if the policy at (vma,addr) is not 683 * marked as MPOL_SHARED. vma policies are protected by the mmap_lock. 684 * If no [shared/vma] mempolicy exists at the addr, get_policy() op 685 * must return NULL--i.e., do not "fallback" to task or system default 686 * policy. 687 */ 688 struct mempolicy *(*get_policy)(struct vm_area_struct *vma, 689 unsigned long addr, pgoff_t *ilx); 690 #endif 691 #ifdef CONFIG_FIND_NORMAL_PAGE 692 /* 693 * Called by vm_normal_page() for special PTEs in @vma at @addr. This 694 * allows for returning a "normal" page from vm_normal_page() even 695 * though the PTE indicates that the "struct page" either does not exist 696 * or should not be touched: "special". 697 * 698 * Do not add new users: this really only works when a "normal" page 699 * was mapped, but then the PTE got changed to something weird (+ 700 * marked special) that would not make pte_pfn() identify the originally 701 * inserted page. 702 */ 703 struct page *(*find_normal_page)(struct vm_area_struct *vma, 704 unsigned long addr); 705 #endif /* CONFIG_FIND_NORMAL_PAGE */ 706 }; 707 708 struct vm_unmapped_area_info { 709 #define VM_UNMAPPED_AREA_TOPDOWN 1 710 unsigned long flags; 711 unsigned long length; 712 unsigned long low_limit; 713 unsigned long high_limit; 714 unsigned long align_mask; 715 unsigned long align_offset; 716 unsigned long start_gap; 717 }; 718 719 struct pagetable_move_control { 720 struct vm_area_struct *old; /* Source VMA. */ 721 struct vm_area_struct *new; /* Destination VMA. */ 722 unsigned long old_addr; /* Address from which the move begins. */ 723 unsigned long old_end; /* Exclusive address at which old range ends. */ 724 unsigned long new_addr; /* Address to move page tables to. */ 725 unsigned long len_in; /* Bytes to remap specified by user. */ 726 727 bool need_rmap_locks; /* Do rmap locks need to be taken? */ 728 bool for_stack; /* Is this an early temp stack being moved? */ 729 }; 730 731 #define PAGETABLE_MOVE(name, old_, new_, old_addr_, new_addr_, len_) \ 732 struct pagetable_move_control name = { \ 733 .old = old_, \ 734 .new = new_, \ 735 .old_addr = old_addr_, \ 736 .old_end = (old_addr_) + (len_), \ 737 .new_addr = new_addr_, \ 738 .len_in = len_, \ 739 } 740 741 static inline void vma_iter_invalidate(struct vma_iterator *vmi) 742 { 743 mas_pause(&vmi->mas); 744 } 745 746 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot) 747 { 748 return __pgprot(pgprot_val(oldprot) | pgprot_val(newprot)); 749 } 750 751 static inline pgprot_t vm_get_page_prot(vm_flags_t vm_flags) 752 { 753 return __pgprot(vm_flags); 754 } 755 756 static inline bool mm_flags_test(int flag, const struct mm_struct *mm) 757 { 758 return test_bit(flag, ACCESS_PRIVATE(&mm->flags, __mm_flags)); 759 } 760 761 /* 762 * Copy value to the first system word of VMA flags, non-atomically. 763 * 764 * IMPORTANT: This does not overwrite bytes past the first system word. The 765 * caller must account for this. 766 */ 767 static inline void vma_flags_overwrite_word(vma_flags_t *flags, unsigned long value) 768 { 769 *ACCESS_PRIVATE(flags, __vma_flags) = value; 770 } 771 772 /* 773 * Copy value to the first system word of VMA flags ONCE, non-atomically. 774 * 775 * IMPORTANT: This does not overwrite bytes past the first system word. The 776 * caller must account for this. 777 */ 778 static inline void vma_flags_overwrite_word_once(vma_flags_t *flags, unsigned long value) 779 { 780 unsigned long *bitmap = ACCESS_PRIVATE(flags, __vma_flags); 781 782 WRITE_ONCE(*bitmap, value); 783 } 784 785 /* Update the first system word of VMA flags setting bits, non-atomically. */ 786 static inline void vma_flags_set_word(vma_flags_t *flags, unsigned long value) 787 { 788 unsigned long *bitmap = ACCESS_PRIVATE(flags, __vma_flags); 789 790 *bitmap |= value; 791 } 792 793 /* Update the first system word of VMA flags clearing bits, non-atomically. */ 794 static inline void vma_flags_clear_word(vma_flags_t *flags, unsigned long value) 795 { 796 unsigned long *bitmap = ACCESS_PRIVATE(flags, __vma_flags); 797 798 *bitmap &= ~value; 799 } 800 801 static __always_inline void vma_flags_clear_all(vma_flags_t *flags) 802 { 803 bitmap_zero(ACCESS_PRIVATE(flags, __vma_flags), NUM_VMA_FLAG_BITS); 804 } 805 806 static __always_inline void vma_flags_set_flag(vma_flags_t *flags, 807 vma_flag_t bit) 808 { 809 unsigned long *bitmap = ACCESS_PRIVATE(flags, __vma_flags); 810 811 __set_bit((__force int)bit, bitmap); 812 } 813 814 /* Use when VMA is not part of the VMA tree and needs no locking */ 815 static inline void vm_flags_init(struct vm_area_struct *vma, 816 vm_flags_t flags) 817 { 818 vma_flags_clear_all(&vma->flags); 819 vma_flags_overwrite_word(&vma->flags, flags); 820 } 821 822 /* 823 * Use when VMA is part of the VMA tree and modifications need coordination 824 * Note: vm_flags_reset and vm_flags_reset_once do not lock the vma and 825 * it should be locked explicitly beforehand. 826 */ 827 static inline void vm_flags_reset(struct vm_area_struct *vma, 828 vm_flags_t flags) 829 { 830 vma_assert_write_locked(vma); 831 vm_flags_init(vma, flags); 832 } 833 834 static inline void vm_flags_reset_once(struct vm_area_struct *vma, 835 vm_flags_t flags) 836 { 837 vma_assert_write_locked(vma); 838 /* 839 * The user should only be interested in avoiding reordering of 840 * assignment to the first word. 841 */ 842 vma_flags_clear_all(&vma->flags); 843 vma_flags_overwrite_word_once(&vma->flags, flags); 844 } 845 846 static inline void vm_flags_set(struct vm_area_struct *vma, 847 vm_flags_t flags) 848 { 849 vma_start_write(vma); 850 vma_flags_set_word(&vma->flags, flags); 851 } 852 853 static inline void vm_flags_clear(struct vm_area_struct *vma, 854 vm_flags_t flags) 855 { 856 vma_start_write(vma); 857 vma_flags_clear_word(&vma->flags, flags); 858 } 859 860 static __always_inline vma_flags_t __mk_vma_flags(vma_flags_t flags, 861 size_t count, const vma_flag_t *bits) 862 { 863 int i; 864 865 for (i = 0; i < count; i++) 866 vma_flags_set_flag(&flags, bits[i]); 867 return flags; 868 } 869 870 #define mk_vma_flags(...) __mk_vma_flags(EMPTY_VMA_FLAGS, \ 871 COUNT_ARGS(__VA_ARGS__), (const vma_flag_t []){__VA_ARGS__}) 872 873 #define append_vma_flags(flags, ...) __mk_vma_flags(flags, \ 874 COUNT_ARGS(__VA_ARGS__), (const vma_flag_t []){__VA_ARGS__}) 875 876 static __always_inline bool vma_flags_test(const vma_flags_t *flags, 877 vma_flag_t bit) 878 { 879 const unsigned long *bitmap = flags->__vma_flags; 880 881 return test_bit((__force int)bit, bitmap); 882 } 883 884 static __always_inline vma_flags_t vma_flags_and_mask(const vma_flags_t *flags, 885 vma_flags_t to_and) 886 { 887 vma_flags_t dst; 888 unsigned long *bitmap_dst = dst.__vma_flags; 889 const unsigned long *bitmap = flags->__vma_flags; 890 const unsigned long *bitmap_to_and = to_and.__vma_flags; 891 892 bitmap_and(bitmap_dst, bitmap, bitmap_to_and, NUM_VMA_FLAG_BITS); 893 return dst; 894 } 895 896 #define vma_flags_and(flags, ...) \ 897 vma_flags_and_mask(flags, mk_vma_flags(__VA_ARGS__)) 898 899 static __always_inline bool vma_flags_test_any_mask(const vma_flags_t *flags, 900 vma_flags_t to_test) 901 { 902 const unsigned long *bitmap = flags->__vma_flags; 903 const unsigned long *bitmap_to_test = to_test.__vma_flags; 904 905 return bitmap_intersects(bitmap_to_test, bitmap, NUM_VMA_FLAG_BITS); 906 } 907 908 #define vma_flags_test_any(flags, ...) \ 909 vma_flags_test_any_mask(flags, mk_vma_flags(__VA_ARGS__)) 910 911 static __always_inline bool vma_flags_test_all_mask(const vma_flags_t *flags, 912 vma_flags_t to_test) 913 { 914 const unsigned long *bitmap = flags->__vma_flags; 915 const unsigned long *bitmap_to_test = to_test.__vma_flags; 916 917 return bitmap_subset(bitmap_to_test, bitmap, NUM_VMA_FLAG_BITS); 918 } 919 920 #define vma_flags_test_all(flags, ...) \ 921 vma_flags_test_all_mask(flags, mk_vma_flags(__VA_ARGS__)) 922 923 static __always_inline void vma_flags_set_mask(vma_flags_t *flags, vma_flags_t to_set) 924 { 925 unsigned long *bitmap = flags->__vma_flags; 926 const unsigned long *bitmap_to_set = to_set.__vma_flags; 927 928 bitmap_or(bitmap, bitmap, bitmap_to_set, NUM_VMA_FLAG_BITS); 929 } 930 931 #define vma_flags_set(flags, ...) \ 932 vma_flags_set_mask(flags, mk_vma_flags(__VA_ARGS__)) 933 934 static __always_inline void vma_flags_clear_mask(vma_flags_t *flags, vma_flags_t to_clear) 935 { 936 unsigned long *bitmap = flags->__vma_flags; 937 const unsigned long *bitmap_to_clear = to_clear.__vma_flags; 938 939 bitmap_andnot(bitmap, bitmap, bitmap_to_clear, NUM_VMA_FLAG_BITS); 940 } 941 942 #define vma_flags_clear(flags, ...) \ 943 vma_flags_clear_mask(flags, mk_vma_flags(__VA_ARGS__)) 944 945 static __always_inline vma_flags_t vma_flags_diff_pair(const vma_flags_t *flags, 946 const vma_flags_t *flags_other) 947 { 948 vma_flags_t dst; 949 const unsigned long *bitmap_other = flags_other->__vma_flags; 950 const unsigned long *bitmap = flags->__vma_flags; 951 unsigned long *bitmap_dst = dst.__vma_flags; 952 953 bitmap_xor(bitmap_dst, bitmap, bitmap_other, NUM_VMA_FLAG_BITS); 954 return dst; 955 } 956 957 static __always_inline bool vma_flags_same_pair(const vma_flags_t *flags, 958 const vma_flags_t *flags_other) 959 { 960 const unsigned long *bitmap = flags->__vma_flags; 961 const unsigned long *bitmap_other = flags_other->__vma_flags; 962 963 return bitmap_equal(bitmap, bitmap_other, NUM_VMA_FLAG_BITS); 964 } 965 966 static __always_inline bool vma_flags_same_mask(const vma_flags_t *flags, 967 vma_flags_t flags_other) 968 { 969 const unsigned long *bitmap = flags->__vma_flags; 970 const unsigned long *bitmap_other = flags_other.__vma_flags; 971 972 return bitmap_equal(bitmap, bitmap_other, NUM_VMA_FLAG_BITS); 973 } 974 975 #define vma_flags_same(flags, ...) \ 976 vma_flags_same_mask(flags, mk_vma_flags(__VA_ARGS__)) 977 978 static inline bool vma_test_all_mask(const struct vm_area_struct *vma, 979 vma_flags_t flags) 980 { 981 return vma_flags_test_all_mask(&vma->flags, flags); 982 } 983 984 #define vma_test_all(vma, ...) \ 985 vma_test_all_mask(vma, mk_vma_flags(__VA_ARGS__)) 986 987 static inline bool is_shared_maywrite_vm_flags(vm_flags_t vm_flags) 988 { 989 return (vm_flags & (VM_SHARED | VM_MAYWRITE)) == 990 (VM_SHARED | VM_MAYWRITE); 991 } 992 993 static inline void vma_set_flags_mask(struct vm_area_struct *vma, 994 vma_flags_t flags) 995 { 996 vma_flags_set_mask(&vma->flags, flags); 997 } 998 999 #define vma_set_flags(vma, ...) \ 1000 vma_set_flags_mask(vma, mk_vma_flags(__VA_ARGS__)) 1001 1002 static __always_inline bool vma_desc_test(const struct vm_area_desc *desc, 1003 vma_flag_t bit) 1004 { 1005 return vma_flags_test(&desc->vma_flags, bit); 1006 } 1007 1008 static inline bool vma_desc_test_any_mask(const struct vm_area_desc *desc, 1009 vma_flags_t flags) 1010 { 1011 return vma_flags_test_any_mask(&desc->vma_flags, flags); 1012 } 1013 1014 #define vma_desc_test_any(desc, ...) \ 1015 vma_desc_test_any_mask(desc, mk_vma_flags(__VA_ARGS__)) 1016 1017 static inline bool vma_desc_test_all_mask(const struct vm_area_desc *desc, 1018 vma_flags_t flags) 1019 { 1020 return vma_flags_test_all_mask(&desc->vma_flags, flags); 1021 } 1022 1023 #define vma_desc_test_all(desc, ...) \ 1024 vma_desc_test_all_mask(desc, mk_vma_flags(__VA_ARGS__)) 1025 1026 static inline void vma_desc_set_flags_mask(struct vm_area_desc *desc, 1027 vma_flags_t flags) 1028 { 1029 vma_flags_set_mask(&desc->vma_flags, flags); 1030 } 1031 1032 #define vma_desc_set_flags(desc, ...) \ 1033 vma_desc_set_flags_mask(desc, mk_vma_flags(__VA_ARGS__)) 1034 1035 static inline void vma_desc_clear_flags_mask(struct vm_area_desc *desc, 1036 vma_flags_t flags) 1037 { 1038 vma_flags_clear_mask(&desc->vma_flags, flags); 1039 } 1040 1041 #define vma_desc_clear_flags(desc, ...) \ 1042 vma_desc_clear_flags_mask(desc, mk_vma_flags(__VA_ARGS__)) 1043 1044 static inline bool is_shared_maywrite(const vma_flags_t *flags) 1045 { 1046 return vma_flags_test_all(flags, VMA_SHARED_BIT, VMA_MAYWRITE_BIT); 1047 } 1048 1049 static inline bool vma_is_shared_maywrite(struct vm_area_struct *vma) 1050 { 1051 return is_shared_maywrite(&vma->flags); 1052 } 1053 1054 static inline struct vm_area_struct *vma_next(struct vma_iterator *vmi) 1055 { 1056 /* 1057 * Uses mas_find() to get the first VMA when the iterator starts. 1058 * Calling mas_next() could skip the first entry. 1059 */ 1060 return mas_find(&vmi->mas, ULONG_MAX); 1061 } 1062 1063 /* 1064 * WARNING: to avoid racing with vma_mark_attached()/vma_mark_detached(), these 1065 * assertions should be made either under mmap_write_lock or when the object 1066 * has been isolated under mmap_write_lock, ensuring no competing writers. 1067 */ 1068 static inline void vma_assert_attached(struct vm_area_struct *vma) 1069 { 1070 WARN_ON_ONCE(!refcount_read(&vma->vm_refcnt)); 1071 } 1072 1073 static inline void vma_assert_detached(struct vm_area_struct *vma) 1074 { 1075 WARN_ON_ONCE(refcount_read(&vma->vm_refcnt)); 1076 } 1077 1078 static inline void vma_assert_write_locked(struct vm_area_struct *); 1079 static inline void vma_mark_attached(struct vm_area_struct *vma) 1080 { 1081 vma_assert_write_locked(vma); 1082 vma_assert_detached(vma); 1083 refcount_set_release(&vma->vm_refcnt, 1); 1084 } 1085 1086 static inline void vma_mark_detached(struct vm_area_struct *vma) 1087 { 1088 vma_assert_write_locked(vma); 1089 vma_assert_attached(vma); 1090 /* We are the only writer, so no need to use vma_refcount_put(). */ 1091 if (unlikely(!refcount_dec_and_test(&vma->vm_refcnt))) { 1092 /* 1093 * Reader must have temporarily raised vm_refcnt but it will 1094 * drop it without using the vma since vma is write-locked. 1095 */ 1096 } 1097 } 1098 1099 static inline void vma_init(struct vm_area_struct *vma, struct mm_struct *mm) 1100 { 1101 memset(vma, 0, sizeof(*vma)); 1102 vma->vm_mm = mm; 1103 vma->vm_ops = &vma_dummy_vm_ops; 1104 INIT_LIST_HEAD(&vma->anon_vma_chain); 1105 vma->vm_lock_seq = UINT_MAX; 1106 } 1107 1108 /* 1109 * These are defined in vma.h, but sadly vm_stat_account() is referenced by 1110 * kernel/fork.c, so we have to these broadly available there, and temporarily 1111 * define them here to resolve the dependency cycle. 1112 */ 1113 #define is_exec_mapping(flags) \ 1114 ((flags & (VM_EXEC | VM_WRITE | VM_STACK)) == VM_EXEC) 1115 1116 #define is_stack_mapping(flags) \ 1117 (((flags & VM_STACK) == VM_STACK) || (flags & VM_SHADOW_STACK)) 1118 1119 #define is_data_mapping(flags) \ 1120 ((flags & (VM_WRITE | VM_SHARED | VM_STACK)) == VM_WRITE) 1121 1122 static inline void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, 1123 long npages) 1124 { 1125 WRITE_ONCE(mm->total_vm, READ_ONCE(mm->total_vm)+npages); 1126 1127 if (is_exec_mapping(flags)) 1128 mm->exec_vm += npages; 1129 else if (is_stack_mapping(flags)) 1130 mm->stack_vm += npages; 1131 else if (is_data_mapping(flags)) 1132 mm->data_vm += npages; 1133 } 1134 1135 #undef is_exec_mapping 1136 #undef is_stack_mapping 1137 #undef is_data_mapping 1138 1139 static inline void vm_unacct_memory(long pages) 1140 { 1141 vm_acct_memory(-pages); 1142 } 1143 1144 static inline void mapping_allow_writable(struct address_space *mapping) 1145 { 1146 atomic_inc(&mapping->i_mmap_writable); 1147 } 1148 1149 static inline 1150 struct vm_area_struct *vma_find(struct vma_iterator *vmi, unsigned long max) 1151 { 1152 return mas_find(&vmi->mas, max - 1); 1153 } 1154 1155 static inline int vma_iter_clear_gfp(struct vma_iterator *vmi, 1156 unsigned long start, unsigned long end, gfp_t gfp) 1157 { 1158 __mas_set_range(&vmi->mas, start, end - 1); 1159 mas_store_gfp(&vmi->mas, NULL, gfp); 1160 if (unlikely(mas_is_err(&vmi->mas))) 1161 return -ENOMEM; 1162 1163 return 0; 1164 } 1165 1166 static inline void vma_set_anonymous(struct vm_area_struct *vma) 1167 { 1168 vma->vm_ops = NULL; 1169 } 1170 1171 /* Declared in vma.h. */ 1172 static inline void set_vma_from_desc(struct vm_area_struct *vma, 1173 struct vm_area_desc *desc); 1174 1175 static inline int __compat_vma_mmap(const struct file_operations *f_op, 1176 struct file *file, struct vm_area_struct *vma) 1177 { 1178 struct vm_area_desc desc = { 1179 .mm = vma->vm_mm, 1180 .file = file, 1181 .start = vma->vm_start, 1182 .end = vma->vm_end, 1183 1184 .pgoff = vma->vm_pgoff, 1185 .vm_file = vma->vm_file, 1186 .vma_flags = vma->flags, 1187 .page_prot = vma->vm_page_prot, 1188 1189 .action.type = MMAP_NOTHING, /* Default */ 1190 }; 1191 int err; 1192 1193 err = f_op->mmap_prepare(&desc); 1194 if (err) 1195 return err; 1196 1197 mmap_action_prepare(&desc.action, &desc); 1198 set_vma_from_desc(vma, &desc); 1199 return mmap_action_complete(&desc.action, vma); 1200 } 1201 1202 static inline int compat_vma_mmap(struct file *file, 1203 struct vm_area_struct *vma) 1204 { 1205 return __compat_vma_mmap(file->f_op, file, vma); 1206 } 1207 1208 1209 static inline void vma_iter_init(struct vma_iterator *vmi, 1210 struct mm_struct *mm, unsigned long addr) 1211 { 1212 mas_init(&vmi->mas, &mm->mm_mt, addr); 1213 } 1214 1215 static inline unsigned long vma_pages(struct vm_area_struct *vma) 1216 { 1217 return (vma->vm_end - vma->vm_start) >> PAGE_SHIFT; 1218 } 1219 1220 static inline void mmap_assert_locked(struct mm_struct *); 1221 static inline struct vm_area_struct *find_vma_intersection(struct mm_struct *mm, 1222 unsigned long start_addr, 1223 unsigned long end_addr) 1224 { 1225 unsigned long index = start_addr; 1226 1227 mmap_assert_locked(mm); 1228 return mt_find(&mm->mm_mt, &index, end_addr - 1); 1229 } 1230 1231 static inline 1232 struct vm_area_struct *vma_lookup(struct mm_struct *mm, unsigned long addr) 1233 { 1234 return mtree_load(&mm->mm_mt, addr); 1235 } 1236 1237 static inline struct vm_area_struct *vma_prev(struct vma_iterator *vmi) 1238 { 1239 return mas_prev(&vmi->mas, 0); 1240 } 1241 1242 static inline void vma_iter_set(struct vma_iterator *vmi, unsigned long addr) 1243 { 1244 mas_set(&vmi->mas, addr); 1245 } 1246 1247 static inline bool vma_is_anonymous(struct vm_area_struct *vma) 1248 { 1249 return !vma->vm_ops; 1250 } 1251 1252 /* Defined in vma.h, so temporarily define here to avoid circular dependency. */ 1253 #define vma_iter_load(vmi) \ 1254 mas_walk(&(vmi)->mas) 1255 1256 static inline struct vm_area_struct * 1257 find_vma_prev(struct mm_struct *mm, unsigned long addr, 1258 struct vm_area_struct **pprev) 1259 { 1260 struct vm_area_struct *vma; 1261 VMA_ITERATOR(vmi, mm, addr); 1262 1263 vma = vma_iter_load(&vmi); 1264 *pprev = vma_prev(&vmi); 1265 if (!vma) 1266 vma = vma_next(&vmi); 1267 return vma; 1268 } 1269 1270 #undef vma_iter_load 1271 1272 static inline void vma_iter_free(struct vma_iterator *vmi) 1273 { 1274 mas_destroy(&vmi->mas); 1275 } 1276 1277 static inline 1278 struct vm_area_struct *vma_iter_next_range(struct vma_iterator *vmi) 1279 { 1280 return mas_next_range(&vmi->mas, ULONG_MAX); 1281 } 1282 1283 bool vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot); 1284 1285 /* Update vma->vm_page_prot to reflect vma->vm_flags. */ 1286 static inline void vma_set_page_prot(struct vm_area_struct *vma) 1287 { 1288 vm_flags_t vm_flags = vma->vm_flags; 1289 pgprot_t vm_page_prot; 1290 1291 /* testing: we inline vm_pgprot_modify() to avoid clash with vma.h. */ 1292 vm_page_prot = pgprot_modify(vma->vm_page_prot, vm_get_page_prot(vm_flags)); 1293 1294 if (vma_wants_writenotify(vma, vm_page_prot)) { 1295 vm_flags &= ~VM_SHARED; 1296 /* testing: we inline vm_pgprot_modify() to avoid clash with vma.h. */ 1297 vm_page_prot = pgprot_modify(vm_page_prot, vm_get_page_prot(vm_flags)); 1298 } 1299 /* remove_protection_ptes reads vma->vm_page_prot without mmap_lock */ 1300 WRITE_ONCE(vma->vm_page_prot, vm_page_prot); 1301 } 1302 1303 static inline unsigned long stack_guard_start_gap(struct vm_area_struct *vma) 1304 { 1305 if (vma->vm_flags & VM_GROWSDOWN) 1306 return stack_guard_gap; 1307 1308 /* See reasoning around the VM_SHADOW_STACK definition */ 1309 if (vma->vm_flags & VM_SHADOW_STACK) 1310 return PAGE_SIZE; 1311 1312 return 0; 1313 } 1314 1315 static inline unsigned long vm_start_gap(struct vm_area_struct *vma) 1316 { 1317 unsigned long gap = stack_guard_start_gap(vma); 1318 unsigned long vm_start = vma->vm_start; 1319 1320 vm_start -= gap; 1321 if (vm_start > vma->vm_start) 1322 vm_start = 0; 1323 return vm_start; 1324 } 1325 1326 static inline unsigned long vm_end_gap(struct vm_area_struct *vma) 1327 { 1328 unsigned long vm_end = vma->vm_end; 1329 1330 if (vma->vm_flags & VM_GROWSUP) { 1331 vm_end += stack_guard_gap; 1332 if (vm_end < vma->vm_end) 1333 vm_end = -PAGE_SIZE; 1334 } 1335 return vm_end; 1336 } 1337 1338 static inline bool vma_is_accessible(struct vm_area_struct *vma) 1339 { 1340 return vma->vm_flags & VM_ACCESS_FLAGS; 1341 } 1342 1343 static inline bool mlock_future_ok(const struct mm_struct *mm, 1344 vm_flags_t vm_flags, unsigned long bytes) 1345 { 1346 unsigned long locked_pages, limit_pages; 1347 1348 if (!(vm_flags & VM_LOCKED) || capable(CAP_IPC_LOCK)) 1349 return true; 1350 1351 locked_pages = bytes >> PAGE_SHIFT; 1352 locked_pages += mm->locked_vm; 1353 1354 limit_pages = rlimit(RLIMIT_MEMLOCK); 1355 limit_pages >>= PAGE_SHIFT; 1356 1357 return locked_pages <= limit_pages; 1358 } 1359 1360 static inline bool map_deny_write_exec(unsigned long old, unsigned long new) 1361 { 1362 /* If MDWE is disabled, we have nothing to deny. */ 1363 if (mm_flags_test(MMF_HAS_MDWE, current->mm)) 1364 return false; 1365 1366 /* If the new VMA is not executable, we have nothing to deny. */ 1367 if (!(new & VM_EXEC)) 1368 return false; 1369 1370 /* Under MDWE we do not accept newly writably executable VMAs... */ 1371 if (new & VM_WRITE) 1372 return true; 1373 1374 /* ...nor previously non-executable VMAs becoming executable. */ 1375 if (!(old & VM_EXEC)) 1376 return true; 1377 1378 return false; 1379 } 1380 1381 static inline int mapping_map_writable(struct address_space *mapping) 1382 { 1383 return atomic_inc_unless_negative(&mapping->i_mmap_writable) ? 1384 0 : -EPERM; 1385 } 1386 1387 /* Did the driver provide valid mmap hook configuration? */ 1388 static inline bool can_mmap_file(struct file *file) 1389 { 1390 bool has_mmap = file->f_op->mmap; 1391 bool has_mmap_prepare = file->f_op->mmap_prepare; 1392 1393 /* Hooks are mutually exclusive. */ 1394 if (WARN_ON_ONCE(has_mmap && has_mmap_prepare)) 1395 return false; 1396 if (!has_mmap && !has_mmap_prepare) 1397 return false; 1398 1399 return true; 1400 } 1401 1402 static inline int vfs_mmap(struct file *file, struct vm_area_struct *vma) 1403 { 1404 if (file->f_op->mmap_prepare) 1405 return compat_vma_mmap(file, vma); 1406 1407 return file->f_op->mmap(file, vma); 1408 } 1409 1410 static inline int vfs_mmap_prepare(struct file *file, struct vm_area_desc *desc) 1411 { 1412 return file->f_op->mmap_prepare(desc); 1413 } 1414 1415 static inline void vma_set_file(struct vm_area_struct *vma, struct file *file) 1416 { 1417 /* Changing an anonymous vma with this is illegal */ 1418 get_file(file); 1419 swap(vma->vm_file, file); 1420 fput(file); 1421 } 1422 1423 extern int sysctl_max_map_count; 1424 static inline int get_sysctl_max_map_count(void) 1425 { 1426 return READ_ONCE(sysctl_max_map_count); 1427 } 1428 1429 #ifndef pgtable_supports_soft_dirty 1430 #define pgtable_supports_soft_dirty() IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) 1431 #endif 1432