1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef _LINUX_MM_H 3 #define _LINUX_MM_H 4 5 #include <linux/errno.h> 6 7 #ifdef __KERNEL__ 8 9 #include <linux/mmdebug.h> 10 #include <linux/gfp.h> 11 #include <linux/bug.h> 12 #include <linux/list.h> 13 #include <linux/mmzone.h> 14 #include <linux/rbtree.h> 15 #include <linux/atomic.h> 16 #include <linux/debug_locks.h> 17 #include <linux/mm_types.h> 18 #include <linux/mmap_lock.h> 19 #include <linux/range.h> 20 #include <linux/pfn.h> 21 #include <linux/percpu-refcount.h> 22 #include <linux/bit_spinlock.h> 23 #include <linux/shrinker.h> 24 #include <linux/resource.h> 25 #include <linux/page_ext.h> 26 #include <linux/err.h> 27 #include <linux/page-flags.h> 28 #include <linux/page_ref.h> 29 #include <linux/memremap.h> 30 #include <linux/overflow.h> 31 #include <linux/sizes.h> 32 #include <linux/sched.h> 33 #include <linux/pgtable.h> 34 #include <linux/kasan.h> 35 36 struct mempolicy; 37 struct anon_vma; 38 struct anon_vma_chain; 39 struct user_struct; 40 struct pt_regs; 41 42 extern int sysctl_page_lock_unfairness; 43 44 void init_mm_internals(void); 45 46 #ifndef CONFIG_NUMA /* Don't use mapnrs, do it properly */ 47 extern unsigned long max_mapnr; 48 49 static inline void set_max_mapnr(unsigned long limit) 50 { 51 max_mapnr = limit; 52 } 53 #else 54 static inline void set_max_mapnr(unsigned long limit) { } 55 #endif 56 57 extern atomic_long_t _totalram_pages; 58 static inline unsigned long totalram_pages(void) 59 { 60 return (unsigned long)atomic_long_read(&_totalram_pages); 61 } 62 63 static inline void totalram_pages_inc(void) 64 { 65 atomic_long_inc(&_totalram_pages); 66 } 67 68 static inline void totalram_pages_dec(void) 69 { 70 atomic_long_dec(&_totalram_pages); 71 } 72 73 static inline void totalram_pages_add(long count) 74 { 75 atomic_long_add(count, &_totalram_pages); 76 } 77 78 extern void * high_memory; 79 extern int page_cluster; 80 81 #ifdef CONFIG_SYSCTL 82 extern int sysctl_legacy_va_layout; 83 #else 84 #define sysctl_legacy_va_layout 0 85 #endif 86 87 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS 88 extern const int mmap_rnd_bits_min; 89 extern const int mmap_rnd_bits_max; 90 extern int mmap_rnd_bits __read_mostly; 91 #endif 92 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS 93 extern const int mmap_rnd_compat_bits_min; 94 extern const int mmap_rnd_compat_bits_max; 95 extern int mmap_rnd_compat_bits __read_mostly; 96 #endif 97 98 #include <asm/page.h> 99 #include <asm/processor.h> 100 101 /* 102 * Architectures that support memory tagging (assigning tags to memory regions, 103 * embedding these tags into addresses that point to these memory regions, and 104 * checking that the memory and the pointer tags match on memory accesses) 105 * redefine this macro to strip tags from pointers. 106 * It's defined as noop for architectures that don't support memory tagging. 107 */ 108 #ifndef untagged_addr 109 #define untagged_addr(addr) (addr) 110 #endif 111 112 #ifndef __pa_symbol 113 #define __pa_symbol(x) __pa(RELOC_HIDE((unsigned long)(x), 0)) 114 #endif 115 116 #ifndef page_to_virt 117 #define page_to_virt(x) __va(PFN_PHYS(page_to_pfn(x))) 118 #endif 119 120 #ifndef lm_alias 121 #define lm_alias(x) __va(__pa_symbol(x)) 122 #endif 123 124 /* 125 * To prevent common memory management code establishing 126 * a zero page mapping on a read fault. 127 * This macro should be defined within <asm/pgtable.h>. 128 * s390 does this to prevent multiplexing of hardware bits 129 * related to the physical page in case of virtualization. 130 */ 131 #ifndef mm_forbids_zeropage 132 #define mm_forbids_zeropage(X) (0) 133 #endif 134 135 /* 136 * On some architectures it is expensive to call memset() for small sizes. 137 * If an architecture decides to implement their own version of 138 * mm_zero_struct_page they should wrap the defines below in a #ifndef and 139 * define their own version of this macro in <asm/pgtable.h> 140 */ 141 #if BITS_PER_LONG == 64 142 /* This function must be updated when the size of struct page grows above 80 143 * or reduces below 56. The idea that compiler optimizes out switch() 144 * statement, and only leaves move/store instructions. Also the compiler can 145 * combine write statements if they are both assignments and can be reordered, 146 * this can result in several of the writes here being dropped. 147 */ 148 #define mm_zero_struct_page(pp) __mm_zero_struct_page(pp) 149 static inline void __mm_zero_struct_page(struct page *page) 150 { 151 unsigned long *_pp = (void *)page; 152 153 /* Check that struct page is either 56, 64, 72, or 80 bytes */ 154 BUILD_BUG_ON(sizeof(struct page) & 7); 155 BUILD_BUG_ON(sizeof(struct page) < 56); 156 BUILD_BUG_ON(sizeof(struct page) > 80); 157 158 switch (sizeof(struct page)) { 159 case 80: 160 _pp[9] = 0; 161 fallthrough; 162 case 72: 163 _pp[8] = 0; 164 fallthrough; 165 case 64: 166 _pp[7] = 0; 167 fallthrough; 168 case 56: 169 _pp[6] = 0; 170 _pp[5] = 0; 171 _pp[4] = 0; 172 _pp[3] = 0; 173 _pp[2] = 0; 174 _pp[1] = 0; 175 _pp[0] = 0; 176 } 177 } 178 #else 179 #define mm_zero_struct_page(pp) ((void)memset((pp), 0, sizeof(struct page))) 180 #endif 181 182 /* 183 * Default maximum number of active map areas, this limits the number of vmas 184 * per mm struct. Users can overwrite this number by sysctl but there is a 185 * problem. 186 * 187 * When a program's coredump is generated as ELF format, a section is created 188 * per a vma. In ELF, the number of sections is represented in unsigned short. 189 * This means the number of sections should be smaller than 65535 at coredump. 190 * Because the kernel adds some informative sections to a image of program at 191 * generating coredump, we need some margin. The number of extra sections is 192 * 1-3 now and depends on arch. We use "5" as safe margin, here. 193 * 194 * ELF extended numbering allows more than 65535 sections, so 16-bit bound is 195 * not a hard limit any more. Although some userspace tools can be surprised by 196 * that. 197 */ 198 #define MAPCOUNT_ELF_CORE_MARGIN (5) 199 #define DEFAULT_MAX_MAP_COUNT (USHRT_MAX - MAPCOUNT_ELF_CORE_MARGIN) 200 201 extern int sysctl_max_map_count; 202 203 extern unsigned long sysctl_user_reserve_kbytes; 204 extern unsigned long sysctl_admin_reserve_kbytes; 205 206 extern int sysctl_overcommit_memory; 207 extern int sysctl_overcommit_ratio; 208 extern unsigned long sysctl_overcommit_kbytes; 209 210 int overcommit_ratio_handler(struct ctl_table *, int, void *, size_t *, 211 loff_t *); 212 int overcommit_kbytes_handler(struct ctl_table *, int, void *, size_t *, 213 loff_t *); 214 int overcommit_policy_handler(struct ctl_table *, int, void *, size_t *, 215 loff_t *); 216 217 #if defined(CONFIG_SPARSEMEM) && !defined(CONFIG_SPARSEMEM_VMEMMAP) 218 #define nth_page(page,n) pfn_to_page(page_to_pfn((page)) + (n)) 219 #else 220 #define nth_page(page,n) ((page) + (n)) 221 #endif 222 223 /* to align the pointer to the (next) page boundary */ 224 #define PAGE_ALIGN(addr) ALIGN(addr, PAGE_SIZE) 225 226 /* test whether an address (unsigned long or pointer) is aligned to PAGE_SIZE */ 227 #define PAGE_ALIGNED(addr) IS_ALIGNED((unsigned long)(addr), PAGE_SIZE) 228 229 #define lru_to_page(head) (list_entry((head)->prev, struct page, lru)) 230 231 void setup_initial_init_mm(void *start_code, void *end_code, 232 void *end_data, void *brk); 233 234 /* 235 * Linux kernel virtual memory manager primitives. 236 * The idea being to have a "virtual" mm in the same way 237 * we have a virtual fs - giving a cleaner interface to the 238 * mm details, and allowing different kinds of memory mappings 239 * (from shared memory to executable loading to arbitrary 240 * mmap() functions). 241 */ 242 243 struct vm_area_struct *vm_area_alloc(struct mm_struct *); 244 struct vm_area_struct *vm_area_dup(struct vm_area_struct *); 245 void vm_area_free(struct vm_area_struct *); 246 247 #ifndef CONFIG_MMU 248 extern struct rb_root nommu_region_tree; 249 extern struct rw_semaphore nommu_region_sem; 250 251 extern unsigned int kobjsize(const void *objp); 252 #endif 253 254 /* 255 * vm_flags in vm_area_struct, see mm_types.h. 256 * When changing, update also include/trace/events/mmflags.h 257 */ 258 #define VM_NONE 0x00000000 259 260 #define VM_READ 0x00000001 /* currently active flags */ 261 #define VM_WRITE 0x00000002 262 #define VM_EXEC 0x00000004 263 #define VM_SHARED 0x00000008 264 265 /* mprotect() hardcodes VM_MAYREAD >> 4 == VM_READ, and so for r/w/x bits. */ 266 #define VM_MAYREAD 0x00000010 /* limits for mprotect() etc */ 267 #define VM_MAYWRITE 0x00000020 268 #define VM_MAYEXEC 0x00000040 269 #define VM_MAYSHARE 0x00000080 270 271 #define VM_GROWSDOWN 0x00000100 /* general info on the segment */ 272 #define VM_UFFD_MISSING 0x00000200 /* missing pages tracking */ 273 #define VM_PFNMAP 0x00000400 /* Page-ranges managed without "struct page", just pure PFN */ 274 #define VM_UFFD_WP 0x00001000 /* wrprotect pages tracking */ 275 276 #define VM_LOCKED 0x00002000 277 #define VM_IO 0x00004000 /* Memory mapped I/O or similar */ 278 279 /* Used by sys_madvise() */ 280 #define VM_SEQ_READ 0x00008000 /* App will access data sequentially */ 281 #define VM_RAND_READ 0x00010000 /* App will not benefit from clustered reads */ 282 283 #define VM_DONTCOPY 0x00020000 /* Do not copy this vma on fork */ 284 #define VM_DONTEXPAND 0x00040000 /* Cannot expand with mremap() */ 285 #define VM_LOCKONFAULT 0x00080000 /* Lock the pages covered when they are faulted in */ 286 #define VM_ACCOUNT 0x00100000 /* Is a VM accounted object */ 287 #define VM_NORESERVE 0x00200000 /* should the VM suppress accounting */ 288 #define VM_HUGETLB 0x00400000 /* Huge TLB Page VM */ 289 #define VM_SYNC 0x00800000 /* Synchronous page faults */ 290 #define VM_ARCH_1 0x01000000 /* Architecture-specific flag */ 291 #define VM_WIPEONFORK 0x02000000 /* Wipe VMA contents in child. */ 292 #define VM_DONTDUMP 0x04000000 /* Do not include in the core dump */ 293 294 #ifdef CONFIG_MEM_SOFT_DIRTY 295 # define VM_SOFTDIRTY 0x08000000 /* Not soft dirty clean area */ 296 #else 297 # define VM_SOFTDIRTY 0 298 #endif 299 300 #define VM_MIXEDMAP 0x10000000 /* Can contain "struct page" and pure PFN pages */ 301 #define VM_HUGEPAGE 0x20000000 /* MADV_HUGEPAGE marked this vma */ 302 #define VM_NOHUGEPAGE 0x40000000 /* MADV_NOHUGEPAGE marked this vma */ 303 #define VM_MERGEABLE 0x80000000 /* KSM may merge identical pages */ 304 305 #ifdef CONFIG_ARCH_USES_HIGH_VMA_FLAGS 306 #define VM_HIGH_ARCH_BIT_0 32 /* bit only usable on 64-bit architectures */ 307 #define VM_HIGH_ARCH_BIT_1 33 /* bit only usable on 64-bit architectures */ 308 #define VM_HIGH_ARCH_BIT_2 34 /* bit only usable on 64-bit architectures */ 309 #define VM_HIGH_ARCH_BIT_3 35 /* bit only usable on 64-bit architectures */ 310 #define VM_HIGH_ARCH_BIT_4 36 /* bit only usable on 64-bit architectures */ 311 #define VM_HIGH_ARCH_0 BIT(VM_HIGH_ARCH_BIT_0) 312 #define VM_HIGH_ARCH_1 BIT(VM_HIGH_ARCH_BIT_1) 313 #define VM_HIGH_ARCH_2 BIT(VM_HIGH_ARCH_BIT_2) 314 #define VM_HIGH_ARCH_3 BIT(VM_HIGH_ARCH_BIT_3) 315 #define VM_HIGH_ARCH_4 BIT(VM_HIGH_ARCH_BIT_4) 316 #endif /* CONFIG_ARCH_USES_HIGH_VMA_FLAGS */ 317 318 #ifdef CONFIG_ARCH_HAS_PKEYS 319 # define VM_PKEY_SHIFT VM_HIGH_ARCH_BIT_0 320 # define VM_PKEY_BIT0 VM_HIGH_ARCH_0 /* A protection key is a 4-bit value */ 321 # define VM_PKEY_BIT1 VM_HIGH_ARCH_1 /* on x86 and 5-bit value on ppc64 */ 322 # define VM_PKEY_BIT2 VM_HIGH_ARCH_2 323 # define VM_PKEY_BIT3 VM_HIGH_ARCH_3 324 #ifdef CONFIG_PPC 325 # define VM_PKEY_BIT4 VM_HIGH_ARCH_4 326 #else 327 # define VM_PKEY_BIT4 0 328 #endif 329 #endif /* CONFIG_ARCH_HAS_PKEYS */ 330 331 #if defined(CONFIG_X86) 332 # define VM_PAT VM_ARCH_1 /* PAT reserves whole VMA at once (x86) */ 333 #elif defined(CONFIG_PPC) 334 # define VM_SAO VM_ARCH_1 /* Strong Access Ordering (powerpc) */ 335 #elif defined(CONFIG_PARISC) 336 # define VM_GROWSUP VM_ARCH_1 337 #elif defined(CONFIG_IA64) 338 # define VM_GROWSUP VM_ARCH_1 339 #elif defined(CONFIG_SPARC64) 340 # define VM_SPARC_ADI VM_ARCH_1 /* Uses ADI tag for access control */ 341 # define VM_ARCH_CLEAR VM_SPARC_ADI 342 #elif defined(CONFIG_ARM64) 343 # define VM_ARM64_BTI VM_ARCH_1 /* BTI guarded page, a.k.a. GP bit */ 344 # define VM_ARCH_CLEAR VM_ARM64_BTI 345 #elif !defined(CONFIG_MMU) 346 # define VM_MAPPED_COPY VM_ARCH_1 /* T if mapped copy of data (nommu mmap) */ 347 #endif 348 349 #if defined(CONFIG_ARM64_MTE) 350 # define VM_MTE VM_HIGH_ARCH_0 /* Use Tagged memory for access control */ 351 # define VM_MTE_ALLOWED VM_HIGH_ARCH_1 /* Tagged memory permitted */ 352 #else 353 # define VM_MTE VM_NONE 354 # define VM_MTE_ALLOWED VM_NONE 355 #endif 356 357 #ifndef VM_GROWSUP 358 # define VM_GROWSUP VM_NONE 359 #endif 360 361 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR 362 # define VM_UFFD_MINOR_BIT 37 363 # define VM_UFFD_MINOR BIT(VM_UFFD_MINOR_BIT) /* UFFD minor faults */ 364 #else /* !CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */ 365 # define VM_UFFD_MINOR VM_NONE 366 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */ 367 368 /* Bits set in the VMA until the stack is in its final location */ 369 #define VM_STACK_INCOMPLETE_SETUP (VM_RAND_READ | VM_SEQ_READ) 370 371 #define TASK_EXEC ((current->personality & READ_IMPLIES_EXEC) ? VM_EXEC : 0) 372 373 /* Common data flag combinations */ 374 #define VM_DATA_FLAGS_TSK_EXEC (VM_READ | VM_WRITE | TASK_EXEC | \ 375 VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC) 376 #define VM_DATA_FLAGS_NON_EXEC (VM_READ | VM_WRITE | VM_MAYREAD | \ 377 VM_MAYWRITE | VM_MAYEXEC) 378 #define VM_DATA_FLAGS_EXEC (VM_READ | VM_WRITE | VM_EXEC | \ 379 VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC) 380 381 #ifndef VM_DATA_DEFAULT_FLAGS /* arch can override this */ 382 #define VM_DATA_DEFAULT_FLAGS VM_DATA_FLAGS_EXEC 383 #endif 384 385 #ifndef VM_STACK_DEFAULT_FLAGS /* arch can override this */ 386 #define VM_STACK_DEFAULT_FLAGS VM_DATA_DEFAULT_FLAGS 387 #endif 388 389 #ifdef CONFIG_STACK_GROWSUP 390 #define VM_STACK VM_GROWSUP 391 #else 392 #define VM_STACK VM_GROWSDOWN 393 #endif 394 395 #define VM_STACK_FLAGS (VM_STACK | VM_STACK_DEFAULT_FLAGS | VM_ACCOUNT) 396 397 /* VMA basic access permission flags */ 398 #define VM_ACCESS_FLAGS (VM_READ | VM_WRITE | VM_EXEC) 399 400 401 /* 402 * Special vmas that are non-mergable, non-mlock()able. 403 */ 404 #define VM_SPECIAL (VM_IO | VM_DONTEXPAND | VM_PFNMAP | VM_MIXEDMAP) 405 406 /* This mask prevents VMA from being scanned with khugepaged */ 407 #define VM_NO_KHUGEPAGED (VM_SPECIAL | VM_HUGETLB) 408 409 /* This mask defines which mm->def_flags a process can inherit its parent */ 410 #define VM_INIT_DEF_MASK VM_NOHUGEPAGE 411 412 /* This mask is used to clear all the VMA flags used by mlock */ 413 #define VM_LOCKED_CLEAR_MASK (~(VM_LOCKED | VM_LOCKONFAULT)) 414 415 /* Arch-specific flags to clear when updating VM flags on protection change */ 416 #ifndef VM_ARCH_CLEAR 417 # define VM_ARCH_CLEAR VM_NONE 418 #endif 419 #define VM_FLAGS_CLEAR (ARCH_VM_PKEY_FLAGS | VM_ARCH_CLEAR) 420 421 /* 422 * mapping from the currently active vm_flags protection bits (the 423 * low four bits) to a page protection mask.. 424 */ 425 extern pgprot_t protection_map[16]; 426 427 /* 428 * The default fault flags that should be used by most of the 429 * arch-specific page fault handlers. 430 */ 431 #define FAULT_FLAG_DEFAULT (FAULT_FLAG_ALLOW_RETRY | \ 432 FAULT_FLAG_KILLABLE | \ 433 FAULT_FLAG_INTERRUPTIBLE) 434 435 /** 436 * fault_flag_allow_retry_first - check ALLOW_RETRY the first time 437 * @flags: Fault flags. 438 * 439 * This is mostly used for places where we want to try to avoid taking 440 * the mmap_lock for too long a time when waiting for another condition 441 * to change, in which case we can try to be polite to release the 442 * mmap_lock in the first round to avoid potential starvation of other 443 * processes that would also want the mmap_lock. 444 * 445 * Return: true if the page fault allows retry and this is the first 446 * attempt of the fault handling; false otherwise. 447 */ 448 static inline bool fault_flag_allow_retry_first(enum fault_flag flags) 449 { 450 return (flags & FAULT_FLAG_ALLOW_RETRY) && 451 (!(flags & FAULT_FLAG_TRIED)); 452 } 453 454 #define FAULT_FLAG_TRACE \ 455 { FAULT_FLAG_WRITE, "WRITE" }, \ 456 { FAULT_FLAG_MKWRITE, "MKWRITE" }, \ 457 { FAULT_FLAG_ALLOW_RETRY, "ALLOW_RETRY" }, \ 458 { FAULT_FLAG_RETRY_NOWAIT, "RETRY_NOWAIT" }, \ 459 { FAULT_FLAG_KILLABLE, "KILLABLE" }, \ 460 { FAULT_FLAG_TRIED, "TRIED" }, \ 461 { FAULT_FLAG_USER, "USER" }, \ 462 { FAULT_FLAG_REMOTE, "REMOTE" }, \ 463 { FAULT_FLAG_INSTRUCTION, "INSTRUCTION" }, \ 464 { FAULT_FLAG_INTERRUPTIBLE, "INTERRUPTIBLE" } 465 466 /* 467 * vm_fault is filled by the pagefault handler and passed to the vma's 468 * ->fault function. The vma's ->fault is responsible for returning a bitmask 469 * of VM_FAULT_xxx flags that give details about how the fault was handled. 470 * 471 * MM layer fills up gfp_mask for page allocations but fault handler might 472 * alter it if its implementation requires a different allocation context. 473 * 474 * pgoff should be used in favour of virtual_address, if possible. 475 */ 476 struct vm_fault { 477 const struct { 478 struct vm_area_struct *vma; /* Target VMA */ 479 gfp_t gfp_mask; /* gfp mask to be used for allocations */ 480 pgoff_t pgoff; /* Logical page offset based on vma */ 481 unsigned long address; /* Faulting virtual address */ 482 }; 483 enum fault_flag flags; /* FAULT_FLAG_xxx flags 484 * XXX: should really be 'const' */ 485 pmd_t *pmd; /* Pointer to pmd entry matching 486 * the 'address' */ 487 pud_t *pud; /* Pointer to pud entry matching 488 * the 'address' 489 */ 490 union { 491 pte_t orig_pte; /* Value of PTE at the time of fault */ 492 pmd_t orig_pmd; /* Value of PMD at the time of fault, 493 * used by PMD fault only. 494 */ 495 }; 496 497 struct page *cow_page; /* Page handler may use for COW fault */ 498 struct page *page; /* ->fault handlers should return a 499 * page here, unless VM_FAULT_NOPAGE 500 * is set (which is also implied by 501 * VM_FAULT_ERROR). 502 */ 503 /* These three entries are valid only while holding ptl lock */ 504 pte_t *pte; /* Pointer to pte entry matching 505 * the 'address'. NULL if the page 506 * table hasn't been allocated. 507 */ 508 spinlock_t *ptl; /* Page table lock. 509 * Protects pte page table if 'pte' 510 * is not NULL, otherwise pmd. 511 */ 512 pgtable_t prealloc_pte; /* Pre-allocated pte page table. 513 * vm_ops->map_pages() sets up a page 514 * table from atomic context. 515 * do_fault_around() pre-allocates 516 * page table to avoid allocation from 517 * atomic context. 518 */ 519 }; 520 521 /* page entry size for vm->huge_fault() */ 522 enum page_entry_size { 523 PE_SIZE_PTE = 0, 524 PE_SIZE_PMD, 525 PE_SIZE_PUD, 526 }; 527 528 /* 529 * These are the virtual MM functions - opening of an area, closing and 530 * unmapping it (needed to keep files on disk up-to-date etc), pointer 531 * to the functions called when a no-page or a wp-page exception occurs. 532 */ 533 struct vm_operations_struct { 534 void (*open)(struct vm_area_struct * area); 535 /** 536 * @close: Called when the VMA is being removed from the MM. 537 * Context: User context. May sleep. Caller holds mmap_lock. 538 */ 539 void (*close)(struct vm_area_struct * area); 540 /* Called any time before splitting to check if it's allowed */ 541 int (*may_split)(struct vm_area_struct *area, unsigned long addr); 542 int (*mremap)(struct vm_area_struct *area); 543 /* 544 * Called by mprotect() to make driver-specific permission 545 * checks before mprotect() is finalised. The VMA must not 546 * be modified. Returns 0 if eprotect() can proceed. 547 */ 548 int (*mprotect)(struct vm_area_struct *vma, unsigned long start, 549 unsigned long end, unsigned long newflags); 550 vm_fault_t (*fault)(struct vm_fault *vmf); 551 vm_fault_t (*huge_fault)(struct vm_fault *vmf, 552 enum page_entry_size pe_size); 553 vm_fault_t (*map_pages)(struct vm_fault *vmf, 554 pgoff_t start_pgoff, pgoff_t end_pgoff); 555 unsigned long (*pagesize)(struct vm_area_struct * area); 556 557 /* notification that a previously read-only page is about to become 558 * writable, if an error is returned it will cause a SIGBUS */ 559 vm_fault_t (*page_mkwrite)(struct vm_fault *vmf); 560 561 /* same as page_mkwrite when using VM_PFNMAP|VM_MIXEDMAP */ 562 vm_fault_t (*pfn_mkwrite)(struct vm_fault *vmf); 563 564 /* called by access_process_vm when get_user_pages() fails, typically 565 * for use by special VMAs. See also generic_access_phys() for a generic 566 * implementation useful for any iomem mapping. 567 */ 568 int (*access)(struct vm_area_struct *vma, unsigned long addr, 569 void *buf, int len, int write); 570 571 /* Called by the /proc/PID/maps code to ask the vma whether it 572 * has a special name. Returning non-NULL will also cause this 573 * vma to be dumped unconditionally. */ 574 const char *(*name)(struct vm_area_struct *vma); 575 576 #ifdef CONFIG_NUMA 577 /* 578 * set_policy() op must add a reference to any non-NULL @new mempolicy 579 * to hold the policy upon return. Caller should pass NULL @new to 580 * remove a policy and fall back to surrounding context--i.e. do not 581 * install a MPOL_DEFAULT policy, nor the task or system default 582 * mempolicy. 583 */ 584 int (*set_policy)(struct vm_area_struct *vma, struct mempolicy *new); 585 586 /* 587 * get_policy() op must add reference [mpol_get()] to any policy at 588 * (vma,addr) marked as MPOL_SHARED. The shared policy infrastructure 589 * in mm/mempolicy.c will do this automatically. 590 * get_policy() must NOT add a ref if the policy at (vma,addr) is not 591 * marked as MPOL_SHARED. vma policies are protected by the mmap_lock. 592 * If no [shared/vma] mempolicy exists at the addr, get_policy() op 593 * must return NULL--i.e., do not "fallback" to task or system default 594 * policy. 595 */ 596 struct mempolicy *(*get_policy)(struct vm_area_struct *vma, 597 unsigned long addr); 598 #endif 599 /* 600 * Called by vm_normal_page() for special PTEs to find the 601 * page for @addr. This is useful if the default behavior 602 * (using pte_page()) would not find the correct page. 603 */ 604 struct page *(*find_special_page)(struct vm_area_struct *vma, 605 unsigned long addr); 606 }; 607 608 static inline void vma_init(struct vm_area_struct *vma, struct mm_struct *mm) 609 { 610 static const struct vm_operations_struct dummy_vm_ops = {}; 611 612 memset(vma, 0, sizeof(*vma)); 613 vma->vm_mm = mm; 614 vma->vm_ops = &dummy_vm_ops; 615 INIT_LIST_HEAD(&vma->anon_vma_chain); 616 } 617 618 static inline void vma_set_anonymous(struct vm_area_struct *vma) 619 { 620 vma->vm_ops = NULL; 621 } 622 623 static inline bool vma_is_anonymous(struct vm_area_struct *vma) 624 { 625 return !vma->vm_ops; 626 } 627 628 static inline bool vma_is_temporary_stack(struct vm_area_struct *vma) 629 { 630 int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP); 631 632 if (!maybe_stack) 633 return false; 634 635 if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) == 636 VM_STACK_INCOMPLETE_SETUP) 637 return true; 638 639 return false; 640 } 641 642 static inline bool vma_is_foreign(struct vm_area_struct *vma) 643 { 644 if (!current->mm) 645 return true; 646 647 if (current->mm != vma->vm_mm) 648 return true; 649 650 return false; 651 } 652 653 static inline bool vma_is_accessible(struct vm_area_struct *vma) 654 { 655 return vma->vm_flags & VM_ACCESS_FLAGS; 656 } 657 658 #ifdef CONFIG_SHMEM 659 /* 660 * The vma_is_shmem is not inline because it is used only by slow 661 * paths in userfault. 662 */ 663 bool vma_is_shmem(struct vm_area_struct *vma); 664 #else 665 static inline bool vma_is_shmem(struct vm_area_struct *vma) { return false; } 666 #endif 667 668 int vma_is_stack_for_current(struct vm_area_struct *vma); 669 670 /* flush_tlb_range() takes a vma, not a mm, and can care about flags */ 671 #define TLB_FLUSH_VMA(mm,flags) { .vm_mm = (mm), .vm_flags = (flags) } 672 673 struct mmu_gather; 674 struct inode; 675 676 static inline unsigned int compound_order(struct page *page) 677 { 678 if (!PageHead(page)) 679 return 0; 680 return page[1].compound_order; 681 } 682 683 /** 684 * folio_order - The allocation order of a folio. 685 * @folio: The folio. 686 * 687 * A folio is composed of 2^order pages. See get_order() for the definition 688 * of order. 689 * 690 * Return: The order of the folio. 691 */ 692 static inline unsigned int folio_order(struct folio *folio) 693 { 694 return compound_order(&folio->page); 695 } 696 697 #include <linux/huge_mm.h> 698 699 /* 700 * Methods to modify the page usage count. 701 * 702 * What counts for a page usage: 703 * - cache mapping (page->mapping) 704 * - private data (page->private) 705 * - page mapped in a task's page tables, each mapping 706 * is counted separately 707 * 708 * Also, many kernel routines increase the page count before a critical 709 * routine so they can be sure the page doesn't go away from under them. 710 */ 711 712 /* 713 * Drop a ref, return true if the refcount fell to zero (the page has no users) 714 */ 715 static inline int put_page_testzero(struct page *page) 716 { 717 VM_BUG_ON_PAGE(page_ref_count(page) == 0, page); 718 return page_ref_dec_and_test(page); 719 } 720 721 static inline int folio_put_testzero(struct folio *folio) 722 { 723 return put_page_testzero(&folio->page); 724 } 725 726 /* 727 * Try to grab a ref unless the page has a refcount of zero, return false if 728 * that is the case. 729 * This can be called when MMU is off so it must not access 730 * any of the virtual mappings. 731 */ 732 static inline bool get_page_unless_zero(struct page *page) 733 { 734 return page_ref_add_unless(page, 1, 0); 735 } 736 737 extern int page_is_ram(unsigned long pfn); 738 739 enum { 740 REGION_INTERSECTS, 741 REGION_DISJOINT, 742 REGION_MIXED, 743 }; 744 745 int region_intersects(resource_size_t offset, size_t size, unsigned long flags, 746 unsigned long desc); 747 748 /* Support for virtually mapped pages */ 749 struct page *vmalloc_to_page(const void *addr); 750 unsigned long vmalloc_to_pfn(const void *addr); 751 752 /* 753 * Determine if an address is within the vmalloc range 754 * 755 * On nommu, vmalloc/vfree wrap through kmalloc/kfree directly, so there 756 * is no special casing required. 757 */ 758 759 #ifndef is_ioremap_addr 760 #define is_ioremap_addr(x) is_vmalloc_addr(x) 761 #endif 762 763 #ifdef CONFIG_MMU 764 extern bool is_vmalloc_addr(const void *x); 765 extern int is_vmalloc_or_module_addr(const void *x); 766 #else 767 static inline bool is_vmalloc_addr(const void *x) 768 { 769 return false; 770 } 771 static inline int is_vmalloc_or_module_addr(const void *x) 772 { 773 return 0; 774 } 775 #endif 776 777 static inline int head_compound_mapcount(struct page *head) 778 { 779 return atomic_read(compound_mapcount_ptr(head)) + 1; 780 } 781 782 /* 783 * Mapcount of compound page as a whole, does not include mapped sub-pages. 784 * 785 * Must be called only for compound pages or any their tail sub-pages. 786 */ 787 static inline int compound_mapcount(struct page *page) 788 { 789 VM_BUG_ON_PAGE(!PageCompound(page), page); 790 page = compound_head(page); 791 return head_compound_mapcount(page); 792 } 793 794 /* 795 * The atomic page->_mapcount, starts from -1: so that transitions 796 * both from it and to it can be tracked, using atomic_inc_and_test 797 * and atomic_add_negative(-1). 798 */ 799 static inline void page_mapcount_reset(struct page *page) 800 { 801 atomic_set(&(page)->_mapcount, -1); 802 } 803 804 int __page_mapcount(struct page *page); 805 806 /* 807 * Mapcount of 0-order page; when compound sub-page, includes 808 * compound_mapcount(). 809 * 810 * Result is undefined for pages which cannot be mapped into userspace. 811 * For example SLAB or special types of pages. See function page_has_type(). 812 * They use this place in struct page differently. 813 */ 814 static inline int page_mapcount(struct page *page) 815 { 816 if (unlikely(PageCompound(page))) 817 return __page_mapcount(page); 818 return atomic_read(&page->_mapcount) + 1; 819 } 820 821 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 822 int total_mapcount(struct page *page); 823 int page_trans_huge_mapcount(struct page *page); 824 #else 825 static inline int total_mapcount(struct page *page) 826 { 827 return page_mapcount(page); 828 } 829 static inline int page_trans_huge_mapcount(struct page *page) 830 { 831 return page_mapcount(page); 832 } 833 #endif 834 835 static inline struct page *virt_to_head_page(const void *x) 836 { 837 struct page *page = virt_to_page(x); 838 839 return compound_head(page); 840 } 841 842 static inline struct folio *virt_to_folio(const void *x) 843 { 844 struct page *page = virt_to_page(x); 845 846 return page_folio(page); 847 } 848 849 void __put_page(struct page *page); 850 851 void put_pages_list(struct list_head *pages); 852 853 void split_page(struct page *page, unsigned int order); 854 void folio_copy(struct folio *dst, struct folio *src); 855 856 unsigned long nr_free_buffer_pages(void); 857 858 /* 859 * Compound pages have a destructor function. Provide a 860 * prototype for that function and accessor functions. 861 * These are _only_ valid on the head of a compound page. 862 */ 863 typedef void compound_page_dtor(struct page *); 864 865 /* Keep the enum in sync with compound_page_dtors array in mm/page_alloc.c */ 866 enum compound_dtor_id { 867 NULL_COMPOUND_DTOR, 868 COMPOUND_PAGE_DTOR, 869 #ifdef CONFIG_HUGETLB_PAGE 870 HUGETLB_PAGE_DTOR, 871 #endif 872 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 873 TRANSHUGE_PAGE_DTOR, 874 #endif 875 NR_COMPOUND_DTORS, 876 }; 877 extern compound_page_dtor * const compound_page_dtors[NR_COMPOUND_DTORS]; 878 879 static inline void set_compound_page_dtor(struct page *page, 880 enum compound_dtor_id compound_dtor) 881 { 882 VM_BUG_ON_PAGE(compound_dtor >= NR_COMPOUND_DTORS, page); 883 page[1].compound_dtor = compound_dtor; 884 } 885 886 static inline void destroy_compound_page(struct page *page) 887 { 888 VM_BUG_ON_PAGE(page[1].compound_dtor >= NR_COMPOUND_DTORS, page); 889 compound_page_dtors[page[1].compound_dtor](page); 890 } 891 892 static inline bool hpage_pincount_available(struct page *page) 893 { 894 /* 895 * Can the page->hpage_pinned_refcount field be used? That field is in 896 * the 3rd page of the compound page, so the smallest (2-page) compound 897 * pages cannot support it. 898 */ 899 page = compound_head(page); 900 return PageCompound(page) && compound_order(page) > 1; 901 } 902 903 static inline int head_compound_pincount(struct page *head) 904 { 905 return atomic_read(compound_pincount_ptr(head)); 906 } 907 908 static inline int compound_pincount(struct page *page) 909 { 910 VM_BUG_ON_PAGE(!hpage_pincount_available(page), page); 911 page = compound_head(page); 912 return head_compound_pincount(page); 913 } 914 915 static inline void set_compound_order(struct page *page, unsigned int order) 916 { 917 page[1].compound_order = order; 918 page[1].compound_nr = 1U << order; 919 } 920 921 /* Returns the number of pages in this potentially compound page. */ 922 static inline unsigned long compound_nr(struct page *page) 923 { 924 if (!PageHead(page)) 925 return 1; 926 return page[1].compound_nr; 927 } 928 929 /* Returns the number of bytes in this potentially compound page. */ 930 static inline unsigned long page_size(struct page *page) 931 { 932 return PAGE_SIZE << compound_order(page); 933 } 934 935 /* Returns the number of bits needed for the number of bytes in a page */ 936 static inline unsigned int page_shift(struct page *page) 937 { 938 return PAGE_SHIFT + compound_order(page); 939 } 940 941 void free_compound_page(struct page *page); 942 943 #ifdef CONFIG_MMU 944 /* 945 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when 946 * servicing faults for write access. In the normal case, do always want 947 * pte_mkwrite. But get_user_pages can cause write faults for mappings 948 * that do not have writing enabled, when used by access_process_vm. 949 */ 950 static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma) 951 { 952 if (likely(vma->vm_flags & VM_WRITE)) 953 pte = pte_mkwrite(pte); 954 return pte; 955 } 956 957 vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page); 958 void do_set_pte(struct vm_fault *vmf, struct page *page, unsigned long addr); 959 960 vm_fault_t finish_fault(struct vm_fault *vmf); 961 vm_fault_t finish_mkwrite_fault(struct vm_fault *vmf); 962 #endif 963 964 /* 965 * Multiple processes may "see" the same page. E.g. for untouched 966 * mappings of /dev/null, all processes see the same page full of 967 * zeroes, and text pages of executables and shared libraries have 968 * only one copy in memory, at most, normally. 969 * 970 * For the non-reserved pages, page_count(page) denotes a reference count. 971 * page_count() == 0 means the page is free. page->lru is then used for 972 * freelist management in the buddy allocator. 973 * page_count() > 0 means the page has been allocated. 974 * 975 * Pages are allocated by the slab allocator in order to provide memory 976 * to kmalloc and kmem_cache_alloc. In this case, the management of the 977 * page, and the fields in 'struct page' are the responsibility of mm/slab.c 978 * unless a particular usage is carefully commented. (the responsibility of 979 * freeing the kmalloc memory is the caller's, of course). 980 * 981 * A page may be used by anyone else who does a __get_free_page(). 982 * In this case, page_count still tracks the references, and should only 983 * be used through the normal accessor functions. The top bits of page->flags 984 * and page->virtual store page management information, but all other fields 985 * are unused and could be used privately, carefully. The management of this 986 * page is the responsibility of the one who allocated it, and those who have 987 * subsequently been given references to it. 988 * 989 * The other pages (we may call them "pagecache pages") are completely 990 * managed by the Linux memory manager: I/O, buffers, swapping etc. 991 * The following discussion applies only to them. 992 * 993 * A pagecache page contains an opaque `private' member, which belongs to the 994 * page's address_space. Usually, this is the address of a circular list of 995 * the page's disk buffers. PG_private must be set to tell the VM to call 996 * into the filesystem to release these pages. 997 * 998 * A page may belong to an inode's memory mapping. In this case, page->mapping 999 * is the pointer to the inode, and page->index is the file offset of the page, 1000 * in units of PAGE_SIZE. 1001 * 1002 * If pagecache pages are not associated with an inode, they are said to be 1003 * anonymous pages. These may become associated with the swapcache, and in that 1004 * case PG_swapcache is set, and page->private is an offset into the swapcache. 1005 * 1006 * In either case (swapcache or inode backed), the pagecache itself holds one 1007 * reference to the page. Setting PG_private should also increment the 1008 * refcount. The each user mapping also has a reference to the page. 1009 * 1010 * The pagecache pages are stored in a per-mapping radix tree, which is 1011 * rooted at mapping->i_pages, and indexed by offset. 1012 * Where 2.4 and early 2.6 kernels kept dirty/clean pages in per-address_space 1013 * lists, we instead now tag pages as dirty/writeback in the radix tree. 1014 * 1015 * All pagecache pages may be subject to I/O: 1016 * - inode pages may need to be read from disk, 1017 * - inode pages which have been modified and are MAP_SHARED may need 1018 * to be written back to the inode on disk, 1019 * - anonymous pages (including MAP_PRIVATE file mappings) which have been 1020 * modified may need to be swapped out to swap space and (later) to be read 1021 * back into memory. 1022 */ 1023 1024 /* 1025 * The zone field is never updated after free_area_init_core() 1026 * sets it, so none of the operations on it need to be atomic. 1027 */ 1028 1029 /* Page flags: | [SECTION] | [NODE] | ZONE | [LAST_CPUPID] | ... | FLAGS | */ 1030 #define SECTIONS_PGOFF ((sizeof(unsigned long)*8) - SECTIONS_WIDTH) 1031 #define NODES_PGOFF (SECTIONS_PGOFF - NODES_WIDTH) 1032 #define ZONES_PGOFF (NODES_PGOFF - ZONES_WIDTH) 1033 #define LAST_CPUPID_PGOFF (ZONES_PGOFF - LAST_CPUPID_WIDTH) 1034 #define KASAN_TAG_PGOFF (LAST_CPUPID_PGOFF - KASAN_TAG_WIDTH) 1035 1036 /* 1037 * Define the bit shifts to access each section. For non-existent 1038 * sections we define the shift as 0; that plus a 0 mask ensures 1039 * the compiler will optimise away reference to them. 1040 */ 1041 #define SECTIONS_PGSHIFT (SECTIONS_PGOFF * (SECTIONS_WIDTH != 0)) 1042 #define NODES_PGSHIFT (NODES_PGOFF * (NODES_WIDTH != 0)) 1043 #define ZONES_PGSHIFT (ZONES_PGOFF * (ZONES_WIDTH != 0)) 1044 #define LAST_CPUPID_PGSHIFT (LAST_CPUPID_PGOFF * (LAST_CPUPID_WIDTH != 0)) 1045 #define KASAN_TAG_PGSHIFT (KASAN_TAG_PGOFF * (KASAN_TAG_WIDTH != 0)) 1046 1047 /* NODE:ZONE or SECTION:ZONE is used to ID a zone for the buddy allocator */ 1048 #ifdef NODE_NOT_IN_PAGE_FLAGS 1049 #define ZONEID_SHIFT (SECTIONS_SHIFT + ZONES_SHIFT) 1050 #define ZONEID_PGOFF ((SECTIONS_PGOFF < ZONES_PGOFF)? \ 1051 SECTIONS_PGOFF : ZONES_PGOFF) 1052 #else 1053 #define ZONEID_SHIFT (NODES_SHIFT + ZONES_SHIFT) 1054 #define ZONEID_PGOFF ((NODES_PGOFF < ZONES_PGOFF)? \ 1055 NODES_PGOFF : ZONES_PGOFF) 1056 #endif 1057 1058 #define ZONEID_PGSHIFT (ZONEID_PGOFF * (ZONEID_SHIFT != 0)) 1059 1060 #define ZONES_MASK ((1UL << ZONES_WIDTH) - 1) 1061 #define NODES_MASK ((1UL << NODES_WIDTH) - 1) 1062 #define SECTIONS_MASK ((1UL << SECTIONS_WIDTH) - 1) 1063 #define LAST_CPUPID_MASK ((1UL << LAST_CPUPID_SHIFT) - 1) 1064 #define KASAN_TAG_MASK ((1UL << KASAN_TAG_WIDTH) - 1) 1065 #define ZONEID_MASK ((1UL << ZONEID_SHIFT) - 1) 1066 1067 static inline enum zone_type page_zonenum(const struct page *page) 1068 { 1069 ASSERT_EXCLUSIVE_BITS(page->flags, ZONES_MASK << ZONES_PGSHIFT); 1070 return (page->flags >> ZONES_PGSHIFT) & ZONES_MASK; 1071 } 1072 1073 static inline enum zone_type folio_zonenum(const struct folio *folio) 1074 { 1075 return page_zonenum(&folio->page); 1076 } 1077 1078 #ifdef CONFIG_ZONE_DEVICE 1079 static inline bool is_zone_device_page(const struct page *page) 1080 { 1081 return page_zonenum(page) == ZONE_DEVICE; 1082 } 1083 extern void memmap_init_zone_device(struct zone *, unsigned long, 1084 unsigned long, struct dev_pagemap *); 1085 #else 1086 static inline bool is_zone_device_page(const struct page *page) 1087 { 1088 return false; 1089 } 1090 #endif 1091 1092 static inline bool is_zone_movable_page(const struct page *page) 1093 { 1094 return page_zonenum(page) == ZONE_MOVABLE; 1095 } 1096 1097 #ifdef CONFIG_DEV_PAGEMAP_OPS 1098 void free_devmap_managed_page(struct page *page); 1099 DECLARE_STATIC_KEY_FALSE(devmap_managed_key); 1100 1101 static inline bool page_is_devmap_managed(struct page *page) 1102 { 1103 if (!static_branch_unlikely(&devmap_managed_key)) 1104 return false; 1105 if (!is_zone_device_page(page)) 1106 return false; 1107 switch (page->pgmap->type) { 1108 case MEMORY_DEVICE_PRIVATE: 1109 case MEMORY_DEVICE_FS_DAX: 1110 return true; 1111 default: 1112 break; 1113 } 1114 return false; 1115 } 1116 1117 void put_devmap_managed_page(struct page *page); 1118 1119 #else /* CONFIG_DEV_PAGEMAP_OPS */ 1120 static inline bool page_is_devmap_managed(struct page *page) 1121 { 1122 return false; 1123 } 1124 1125 static inline void put_devmap_managed_page(struct page *page) 1126 { 1127 } 1128 #endif /* CONFIG_DEV_PAGEMAP_OPS */ 1129 1130 static inline bool is_device_private_page(const struct page *page) 1131 { 1132 return IS_ENABLED(CONFIG_DEV_PAGEMAP_OPS) && 1133 IS_ENABLED(CONFIG_DEVICE_PRIVATE) && 1134 is_zone_device_page(page) && 1135 page->pgmap->type == MEMORY_DEVICE_PRIVATE; 1136 } 1137 1138 static inline bool is_pci_p2pdma_page(const struct page *page) 1139 { 1140 return IS_ENABLED(CONFIG_DEV_PAGEMAP_OPS) && 1141 IS_ENABLED(CONFIG_PCI_P2PDMA) && 1142 is_zone_device_page(page) && 1143 page->pgmap->type == MEMORY_DEVICE_PCI_P2PDMA; 1144 } 1145 1146 /* 127: arbitrary random number, small enough to assemble well */ 1147 #define folio_ref_zero_or_close_to_overflow(folio) \ 1148 ((unsigned int) folio_ref_count(folio) + 127u <= 127u) 1149 1150 /** 1151 * folio_get - Increment the reference count on a folio. 1152 * @folio: The folio. 1153 * 1154 * Context: May be called in any context, as long as you know that 1155 * you have a refcount on the folio. If you do not already have one, 1156 * folio_try_get() may be the right interface for you to use. 1157 */ 1158 static inline void folio_get(struct folio *folio) 1159 { 1160 VM_BUG_ON_FOLIO(folio_ref_zero_or_close_to_overflow(folio), folio); 1161 folio_ref_inc(folio); 1162 } 1163 1164 static inline void get_page(struct page *page) 1165 { 1166 folio_get(page_folio(page)); 1167 } 1168 1169 bool __must_check try_grab_page(struct page *page, unsigned int flags); 1170 struct page *try_grab_compound_head(struct page *page, int refs, 1171 unsigned int flags); 1172 1173 1174 static inline __must_check bool try_get_page(struct page *page) 1175 { 1176 page = compound_head(page); 1177 if (WARN_ON_ONCE(page_ref_count(page) <= 0)) 1178 return false; 1179 page_ref_inc(page); 1180 return true; 1181 } 1182 1183 /** 1184 * folio_put - Decrement the reference count on a folio. 1185 * @folio: The folio. 1186 * 1187 * If the folio's reference count reaches zero, the memory will be 1188 * released back to the page allocator and may be used by another 1189 * allocation immediately. Do not access the memory or the struct folio 1190 * after calling folio_put() unless you can be sure that it wasn't the 1191 * last reference. 1192 * 1193 * Context: May be called in process or interrupt context, but not in NMI 1194 * context. May be called while holding a spinlock. 1195 */ 1196 static inline void folio_put(struct folio *folio) 1197 { 1198 if (folio_put_testzero(folio)) 1199 __put_page(&folio->page); 1200 } 1201 1202 /** 1203 * folio_put_refs - Reduce the reference count on a folio. 1204 * @folio: The folio. 1205 * @refs: The amount to subtract from the folio's reference count. 1206 * 1207 * If the folio's reference count reaches zero, the memory will be 1208 * released back to the page allocator and may be used by another 1209 * allocation immediately. Do not access the memory or the struct folio 1210 * after calling folio_put_refs() unless you can be sure that these weren't 1211 * the last references. 1212 * 1213 * Context: May be called in process or interrupt context, but not in NMI 1214 * context. May be called while holding a spinlock. 1215 */ 1216 static inline void folio_put_refs(struct folio *folio, int refs) 1217 { 1218 if (folio_ref_sub_and_test(folio, refs)) 1219 __put_page(&folio->page); 1220 } 1221 1222 static inline void put_page(struct page *page) 1223 { 1224 struct folio *folio = page_folio(page); 1225 1226 /* 1227 * For devmap managed pages we need to catch refcount transition from 1228 * 2 to 1, when refcount reach one it means the page is free and we 1229 * need to inform the device driver through callback. See 1230 * include/linux/memremap.h and HMM for details. 1231 */ 1232 if (page_is_devmap_managed(&folio->page)) { 1233 put_devmap_managed_page(&folio->page); 1234 return; 1235 } 1236 1237 folio_put(folio); 1238 } 1239 1240 /* 1241 * GUP_PIN_COUNTING_BIAS, and the associated functions that use it, overload 1242 * the page's refcount so that two separate items are tracked: the original page 1243 * reference count, and also a new count of how many pin_user_pages() calls were 1244 * made against the page. ("gup-pinned" is another term for the latter). 1245 * 1246 * With this scheme, pin_user_pages() becomes special: such pages are marked as 1247 * distinct from normal pages. As such, the unpin_user_page() call (and its 1248 * variants) must be used in order to release gup-pinned pages. 1249 * 1250 * Choice of value: 1251 * 1252 * By making GUP_PIN_COUNTING_BIAS a power of two, debugging of page reference 1253 * counts with respect to pin_user_pages() and unpin_user_page() becomes 1254 * simpler, due to the fact that adding an even power of two to the page 1255 * refcount has the effect of using only the upper N bits, for the code that 1256 * counts up using the bias value. This means that the lower bits are left for 1257 * the exclusive use of the original code that increments and decrements by one 1258 * (or at least, by much smaller values than the bias value). 1259 * 1260 * Of course, once the lower bits overflow into the upper bits (and this is 1261 * OK, because subtraction recovers the original values), then visual inspection 1262 * no longer suffices to directly view the separate counts. However, for normal 1263 * applications that don't have huge page reference counts, this won't be an 1264 * issue. 1265 * 1266 * Locking: the lockless algorithm described in page_cache_get_speculative() 1267 * and page_cache_gup_pin_speculative() provides safe operation for 1268 * get_user_pages and page_mkclean and other calls that race to set up page 1269 * table entries. 1270 */ 1271 #define GUP_PIN_COUNTING_BIAS (1U << 10) 1272 1273 void unpin_user_page(struct page *page); 1274 void unpin_user_pages_dirty_lock(struct page **pages, unsigned long npages, 1275 bool make_dirty); 1276 void unpin_user_page_range_dirty_lock(struct page *page, unsigned long npages, 1277 bool make_dirty); 1278 void unpin_user_pages(struct page **pages, unsigned long npages); 1279 1280 /** 1281 * page_maybe_dma_pinned - Report if a page is pinned for DMA. 1282 * @page: The page. 1283 * 1284 * This function checks if a page has been pinned via a call to 1285 * a function in the pin_user_pages() family. 1286 * 1287 * For non-huge pages, the return value is partially fuzzy: false is not fuzzy, 1288 * because it means "definitely not pinned for DMA", but true means "probably 1289 * pinned for DMA, but possibly a false positive due to having at least 1290 * GUP_PIN_COUNTING_BIAS worth of normal page references". 1291 * 1292 * False positives are OK, because: a) it's unlikely for a page to get that many 1293 * refcounts, and b) all the callers of this routine are expected to be able to 1294 * deal gracefully with a false positive. 1295 * 1296 * For huge pages, the result will be exactly correct. That's because we have 1297 * more tracking data available: the 3rd struct page in the compound page is 1298 * used to track the pincount (instead using of the GUP_PIN_COUNTING_BIAS 1299 * scheme). 1300 * 1301 * For more information, please see Documentation/core-api/pin_user_pages.rst. 1302 * 1303 * Return: True, if it is likely that the page has been "dma-pinned". 1304 * False, if the page is definitely not dma-pinned. 1305 */ 1306 static inline bool page_maybe_dma_pinned(struct page *page) 1307 { 1308 if (hpage_pincount_available(page)) 1309 return compound_pincount(page) > 0; 1310 1311 /* 1312 * page_ref_count() is signed. If that refcount overflows, then 1313 * page_ref_count() returns a negative value, and callers will avoid 1314 * further incrementing the refcount. 1315 * 1316 * Here, for that overflow case, use the signed bit to count a little 1317 * bit higher via unsigned math, and thus still get an accurate result. 1318 */ 1319 return ((unsigned int)page_ref_count(compound_head(page))) >= 1320 GUP_PIN_COUNTING_BIAS; 1321 } 1322 1323 static inline bool is_cow_mapping(vm_flags_t flags) 1324 { 1325 return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE; 1326 } 1327 1328 /* 1329 * This should most likely only be called during fork() to see whether we 1330 * should break the cow immediately for a page on the src mm. 1331 */ 1332 static inline bool page_needs_cow_for_dma(struct vm_area_struct *vma, 1333 struct page *page) 1334 { 1335 if (!is_cow_mapping(vma->vm_flags)) 1336 return false; 1337 1338 if (!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags)) 1339 return false; 1340 1341 return page_maybe_dma_pinned(page); 1342 } 1343 1344 #if defined(CONFIG_SPARSEMEM) && !defined(CONFIG_SPARSEMEM_VMEMMAP) 1345 #define SECTION_IN_PAGE_FLAGS 1346 #endif 1347 1348 /* 1349 * The identification function is mainly used by the buddy allocator for 1350 * determining if two pages could be buddies. We are not really identifying 1351 * the zone since we could be using the section number id if we do not have 1352 * node id available in page flags. 1353 * We only guarantee that it will return the same value for two combinable 1354 * pages in a zone. 1355 */ 1356 static inline int page_zone_id(struct page *page) 1357 { 1358 return (page->flags >> ZONEID_PGSHIFT) & ZONEID_MASK; 1359 } 1360 1361 #ifdef NODE_NOT_IN_PAGE_FLAGS 1362 extern int page_to_nid(const struct page *page); 1363 #else 1364 static inline int page_to_nid(const struct page *page) 1365 { 1366 struct page *p = (struct page *)page; 1367 1368 return (PF_POISONED_CHECK(p)->flags >> NODES_PGSHIFT) & NODES_MASK; 1369 } 1370 #endif 1371 1372 static inline int folio_nid(const struct folio *folio) 1373 { 1374 return page_to_nid(&folio->page); 1375 } 1376 1377 #ifdef CONFIG_NUMA_BALANCING 1378 static inline int cpu_pid_to_cpupid(int cpu, int pid) 1379 { 1380 return ((cpu & LAST__CPU_MASK) << LAST__PID_SHIFT) | (pid & LAST__PID_MASK); 1381 } 1382 1383 static inline int cpupid_to_pid(int cpupid) 1384 { 1385 return cpupid & LAST__PID_MASK; 1386 } 1387 1388 static inline int cpupid_to_cpu(int cpupid) 1389 { 1390 return (cpupid >> LAST__PID_SHIFT) & LAST__CPU_MASK; 1391 } 1392 1393 static inline int cpupid_to_nid(int cpupid) 1394 { 1395 return cpu_to_node(cpupid_to_cpu(cpupid)); 1396 } 1397 1398 static inline bool cpupid_pid_unset(int cpupid) 1399 { 1400 return cpupid_to_pid(cpupid) == (-1 & LAST__PID_MASK); 1401 } 1402 1403 static inline bool cpupid_cpu_unset(int cpupid) 1404 { 1405 return cpupid_to_cpu(cpupid) == (-1 & LAST__CPU_MASK); 1406 } 1407 1408 static inline bool __cpupid_match_pid(pid_t task_pid, int cpupid) 1409 { 1410 return (task_pid & LAST__PID_MASK) == cpupid_to_pid(cpupid); 1411 } 1412 1413 #define cpupid_match_pid(task, cpupid) __cpupid_match_pid(task->pid, cpupid) 1414 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS 1415 static inline int page_cpupid_xchg_last(struct page *page, int cpupid) 1416 { 1417 return xchg(&page->_last_cpupid, cpupid & LAST_CPUPID_MASK); 1418 } 1419 1420 static inline int page_cpupid_last(struct page *page) 1421 { 1422 return page->_last_cpupid; 1423 } 1424 static inline void page_cpupid_reset_last(struct page *page) 1425 { 1426 page->_last_cpupid = -1 & LAST_CPUPID_MASK; 1427 } 1428 #else 1429 static inline int page_cpupid_last(struct page *page) 1430 { 1431 return (page->flags >> LAST_CPUPID_PGSHIFT) & LAST_CPUPID_MASK; 1432 } 1433 1434 extern int page_cpupid_xchg_last(struct page *page, int cpupid); 1435 1436 static inline void page_cpupid_reset_last(struct page *page) 1437 { 1438 page->flags |= LAST_CPUPID_MASK << LAST_CPUPID_PGSHIFT; 1439 } 1440 #endif /* LAST_CPUPID_NOT_IN_PAGE_FLAGS */ 1441 #else /* !CONFIG_NUMA_BALANCING */ 1442 static inline int page_cpupid_xchg_last(struct page *page, int cpupid) 1443 { 1444 return page_to_nid(page); /* XXX */ 1445 } 1446 1447 static inline int page_cpupid_last(struct page *page) 1448 { 1449 return page_to_nid(page); /* XXX */ 1450 } 1451 1452 static inline int cpupid_to_nid(int cpupid) 1453 { 1454 return -1; 1455 } 1456 1457 static inline int cpupid_to_pid(int cpupid) 1458 { 1459 return -1; 1460 } 1461 1462 static inline int cpupid_to_cpu(int cpupid) 1463 { 1464 return -1; 1465 } 1466 1467 static inline int cpu_pid_to_cpupid(int nid, int pid) 1468 { 1469 return -1; 1470 } 1471 1472 static inline bool cpupid_pid_unset(int cpupid) 1473 { 1474 return true; 1475 } 1476 1477 static inline void page_cpupid_reset_last(struct page *page) 1478 { 1479 } 1480 1481 static inline bool cpupid_match_pid(struct task_struct *task, int cpupid) 1482 { 1483 return false; 1484 } 1485 #endif /* CONFIG_NUMA_BALANCING */ 1486 1487 #if defined(CONFIG_KASAN_SW_TAGS) || defined(CONFIG_KASAN_HW_TAGS) 1488 1489 /* 1490 * KASAN per-page tags are stored xor'ed with 0xff. This allows to avoid 1491 * setting tags for all pages to native kernel tag value 0xff, as the default 1492 * value 0x00 maps to 0xff. 1493 */ 1494 1495 static inline u8 page_kasan_tag(const struct page *page) 1496 { 1497 u8 tag = 0xff; 1498 1499 if (kasan_enabled()) { 1500 tag = (page->flags >> KASAN_TAG_PGSHIFT) & KASAN_TAG_MASK; 1501 tag ^= 0xff; 1502 } 1503 1504 return tag; 1505 } 1506 1507 static inline void page_kasan_tag_set(struct page *page, u8 tag) 1508 { 1509 if (kasan_enabled()) { 1510 tag ^= 0xff; 1511 page->flags &= ~(KASAN_TAG_MASK << KASAN_TAG_PGSHIFT); 1512 page->flags |= (tag & KASAN_TAG_MASK) << KASAN_TAG_PGSHIFT; 1513 } 1514 } 1515 1516 static inline void page_kasan_tag_reset(struct page *page) 1517 { 1518 if (kasan_enabled()) 1519 page_kasan_tag_set(page, 0xff); 1520 } 1521 1522 #else /* CONFIG_KASAN_SW_TAGS || CONFIG_KASAN_HW_TAGS */ 1523 1524 static inline u8 page_kasan_tag(const struct page *page) 1525 { 1526 return 0xff; 1527 } 1528 1529 static inline void page_kasan_tag_set(struct page *page, u8 tag) { } 1530 static inline void page_kasan_tag_reset(struct page *page) { } 1531 1532 #endif /* CONFIG_KASAN_SW_TAGS || CONFIG_KASAN_HW_TAGS */ 1533 1534 static inline struct zone *page_zone(const struct page *page) 1535 { 1536 return &NODE_DATA(page_to_nid(page))->node_zones[page_zonenum(page)]; 1537 } 1538 1539 static inline pg_data_t *page_pgdat(const struct page *page) 1540 { 1541 return NODE_DATA(page_to_nid(page)); 1542 } 1543 1544 static inline struct zone *folio_zone(const struct folio *folio) 1545 { 1546 return page_zone(&folio->page); 1547 } 1548 1549 static inline pg_data_t *folio_pgdat(const struct folio *folio) 1550 { 1551 return page_pgdat(&folio->page); 1552 } 1553 1554 #ifdef SECTION_IN_PAGE_FLAGS 1555 static inline void set_page_section(struct page *page, unsigned long section) 1556 { 1557 page->flags &= ~(SECTIONS_MASK << SECTIONS_PGSHIFT); 1558 page->flags |= (section & SECTIONS_MASK) << SECTIONS_PGSHIFT; 1559 } 1560 1561 static inline unsigned long page_to_section(const struct page *page) 1562 { 1563 return (page->flags >> SECTIONS_PGSHIFT) & SECTIONS_MASK; 1564 } 1565 #endif 1566 1567 /** 1568 * folio_pfn - Return the Page Frame Number of a folio. 1569 * @folio: The folio. 1570 * 1571 * A folio may contain multiple pages. The pages have consecutive 1572 * Page Frame Numbers. 1573 * 1574 * Return: The Page Frame Number of the first page in the folio. 1575 */ 1576 static inline unsigned long folio_pfn(struct folio *folio) 1577 { 1578 return page_to_pfn(&folio->page); 1579 } 1580 1581 /* MIGRATE_CMA and ZONE_MOVABLE do not allow pin pages */ 1582 #ifdef CONFIG_MIGRATION 1583 static inline bool is_pinnable_page(struct page *page) 1584 { 1585 return !(is_zone_movable_page(page) || is_migrate_cma_page(page)) || 1586 is_zero_pfn(page_to_pfn(page)); 1587 } 1588 #else 1589 static inline bool is_pinnable_page(struct page *page) 1590 { 1591 return true; 1592 } 1593 #endif 1594 1595 static inline void set_page_zone(struct page *page, enum zone_type zone) 1596 { 1597 page->flags &= ~(ZONES_MASK << ZONES_PGSHIFT); 1598 page->flags |= (zone & ZONES_MASK) << ZONES_PGSHIFT; 1599 } 1600 1601 static inline void set_page_node(struct page *page, unsigned long node) 1602 { 1603 page->flags &= ~(NODES_MASK << NODES_PGSHIFT); 1604 page->flags |= (node & NODES_MASK) << NODES_PGSHIFT; 1605 } 1606 1607 static inline void set_page_links(struct page *page, enum zone_type zone, 1608 unsigned long node, unsigned long pfn) 1609 { 1610 set_page_zone(page, zone); 1611 set_page_node(page, node); 1612 #ifdef SECTION_IN_PAGE_FLAGS 1613 set_page_section(page, pfn_to_section_nr(pfn)); 1614 #endif 1615 } 1616 1617 /** 1618 * folio_nr_pages - The number of pages in the folio. 1619 * @folio: The folio. 1620 * 1621 * Return: A positive power of two. 1622 */ 1623 static inline long folio_nr_pages(struct folio *folio) 1624 { 1625 return compound_nr(&folio->page); 1626 } 1627 1628 /** 1629 * folio_next - Move to the next physical folio. 1630 * @folio: The folio we're currently operating on. 1631 * 1632 * If you have physically contiguous memory which may span more than 1633 * one folio (eg a &struct bio_vec), use this function to move from one 1634 * folio to the next. Do not use it if the memory is only virtually 1635 * contiguous as the folios are almost certainly not adjacent to each 1636 * other. This is the folio equivalent to writing ``page++``. 1637 * 1638 * Context: We assume that the folios are refcounted and/or locked at a 1639 * higher level and do not adjust the reference counts. 1640 * Return: The next struct folio. 1641 */ 1642 static inline struct folio *folio_next(struct folio *folio) 1643 { 1644 return (struct folio *)folio_page(folio, folio_nr_pages(folio)); 1645 } 1646 1647 /** 1648 * folio_shift - The size of the memory described by this folio. 1649 * @folio: The folio. 1650 * 1651 * A folio represents a number of bytes which is a power-of-two in size. 1652 * This function tells you which power-of-two the folio is. See also 1653 * folio_size() and folio_order(). 1654 * 1655 * Context: The caller should have a reference on the folio to prevent 1656 * it from being split. It is not necessary for the folio to be locked. 1657 * Return: The base-2 logarithm of the size of this folio. 1658 */ 1659 static inline unsigned int folio_shift(struct folio *folio) 1660 { 1661 return PAGE_SHIFT + folio_order(folio); 1662 } 1663 1664 /** 1665 * folio_size - The number of bytes in a folio. 1666 * @folio: The folio. 1667 * 1668 * Context: The caller should have a reference on the folio to prevent 1669 * it from being split. It is not necessary for the folio to be locked. 1670 * Return: The number of bytes in this folio. 1671 */ 1672 static inline size_t folio_size(struct folio *folio) 1673 { 1674 return PAGE_SIZE << folio_order(folio); 1675 } 1676 1677 #ifndef HAVE_ARCH_MAKE_PAGE_ACCESSIBLE 1678 static inline int arch_make_page_accessible(struct page *page) 1679 { 1680 return 0; 1681 } 1682 #endif 1683 1684 #ifndef HAVE_ARCH_MAKE_FOLIO_ACCESSIBLE 1685 static inline int arch_make_folio_accessible(struct folio *folio) 1686 { 1687 int ret; 1688 long i, nr = folio_nr_pages(folio); 1689 1690 for (i = 0; i < nr; i++) { 1691 ret = arch_make_page_accessible(folio_page(folio, i)); 1692 if (ret) 1693 break; 1694 } 1695 1696 return ret; 1697 } 1698 #endif 1699 1700 /* 1701 * Some inline functions in vmstat.h depend on page_zone() 1702 */ 1703 #include <linux/vmstat.h> 1704 1705 static __always_inline void *lowmem_page_address(const struct page *page) 1706 { 1707 return page_to_virt(page); 1708 } 1709 1710 #if defined(CONFIG_HIGHMEM) && !defined(WANT_PAGE_VIRTUAL) 1711 #define HASHED_PAGE_VIRTUAL 1712 #endif 1713 1714 #if defined(WANT_PAGE_VIRTUAL) 1715 static inline void *page_address(const struct page *page) 1716 { 1717 return page->virtual; 1718 } 1719 static inline void set_page_address(struct page *page, void *address) 1720 { 1721 page->virtual = address; 1722 } 1723 #define page_address_init() do { } while(0) 1724 #endif 1725 1726 #if defined(HASHED_PAGE_VIRTUAL) 1727 void *page_address(const struct page *page); 1728 void set_page_address(struct page *page, void *virtual); 1729 void page_address_init(void); 1730 #endif 1731 1732 #if !defined(HASHED_PAGE_VIRTUAL) && !defined(WANT_PAGE_VIRTUAL) 1733 #define page_address(page) lowmem_page_address(page) 1734 #define set_page_address(page, address) do { } while(0) 1735 #define page_address_init() do { } while(0) 1736 #endif 1737 1738 static inline void *folio_address(const struct folio *folio) 1739 { 1740 return page_address(&folio->page); 1741 } 1742 1743 extern void *page_rmapping(struct page *page); 1744 extern struct anon_vma *page_anon_vma(struct page *page); 1745 extern pgoff_t __page_file_index(struct page *page); 1746 1747 /* 1748 * Return the pagecache index of the passed page. Regular pagecache pages 1749 * use ->index whereas swapcache pages use swp_offset(->private) 1750 */ 1751 static inline pgoff_t page_index(struct page *page) 1752 { 1753 if (unlikely(PageSwapCache(page))) 1754 return __page_file_index(page); 1755 return page->index; 1756 } 1757 1758 bool page_mapped(struct page *page); 1759 bool folio_mapped(struct folio *folio); 1760 1761 /* 1762 * Return true only if the page has been allocated with 1763 * ALLOC_NO_WATERMARKS and the low watermark was not 1764 * met implying that the system is under some pressure. 1765 */ 1766 static inline bool page_is_pfmemalloc(const struct page *page) 1767 { 1768 /* 1769 * lru.next has bit 1 set if the page is allocated from the 1770 * pfmemalloc reserves. Callers may simply overwrite it if 1771 * they do not need to preserve that information. 1772 */ 1773 return (uintptr_t)page->lru.next & BIT(1); 1774 } 1775 1776 /* 1777 * Only to be called by the page allocator on a freshly allocated 1778 * page. 1779 */ 1780 static inline void set_page_pfmemalloc(struct page *page) 1781 { 1782 page->lru.next = (void *)BIT(1); 1783 } 1784 1785 static inline void clear_page_pfmemalloc(struct page *page) 1786 { 1787 page->lru.next = NULL; 1788 } 1789 1790 /* 1791 * Can be called by the pagefault handler when it gets a VM_FAULT_OOM. 1792 */ 1793 extern void pagefault_out_of_memory(void); 1794 1795 #define offset_in_page(p) ((unsigned long)(p) & ~PAGE_MASK) 1796 #define offset_in_thp(page, p) ((unsigned long)(p) & (thp_size(page) - 1)) 1797 #define offset_in_folio(folio, p) ((unsigned long)(p) & (folio_size(folio) - 1)) 1798 1799 /* 1800 * Flags passed to show_mem() and show_free_areas() to suppress output in 1801 * various contexts. 1802 */ 1803 #define SHOW_MEM_FILTER_NODES (0x0001u) /* disallowed nodes */ 1804 1805 extern void show_free_areas(unsigned int flags, nodemask_t *nodemask); 1806 1807 #ifdef CONFIG_MMU 1808 extern bool can_do_mlock(void); 1809 #else 1810 static inline bool can_do_mlock(void) { return false; } 1811 #endif 1812 extern int user_shm_lock(size_t, struct ucounts *); 1813 extern void user_shm_unlock(size_t, struct ucounts *); 1814 1815 struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, 1816 pte_t pte); 1817 struct page *vm_normal_page_pmd(struct vm_area_struct *vma, unsigned long addr, 1818 pmd_t pmd); 1819 1820 void zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, 1821 unsigned long size); 1822 void zap_page_range(struct vm_area_struct *vma, unsigned long address, 1823 unsigned long size); 1824 void unmap_vmas(struct mmu_gather *tlb, struct vm_area_struct *start_vma, 1825 unsigned long start, unsigned long end); 1826 1827 struct mmu_notifier_range; 1828 1829 void free_pgd_range(struct mmu_gather *tlb, unsigned long addr, 1830 unsigned long end, unsigned long floor, unsigned long ceiling); 1831 int 1832 copy_page_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma); 1833 int follow_invalidate_pte(struct mm_struct *mm, unsigned long address, 1834 struct mmu_notifier_range *range, pte_t **ptepp, 1835 pmd_t **pmdpp, spinlock_t **ptlp); 1836 int follow_pte(struct mm_struct *mm, unsigned long address, 1837 pte_t **ptepp, spinlock_t **ptlp); 1838 int follow_pfn(struct vm_area_struct *vma, unsigned long address, 1839 unsigned long *pfn); 1840 int follow_phys(struct vm_area_struct *vma, unsigned long address, 1841 unsigned int flags, unsigned long *prot, resource_size_t *phys); 1842 int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, 1843 void *buf, int len, int write); 1844 1845 extern void truncate_pagecache(struct inode *inode, loff_t new); 1846 extern void truncate_setsize(struct inode *inode, loff_t newsize); 1847 void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to); 1848 void truncate_pagecache_range(struct inode *inode, loff_t offset, loff_t end); 1849 int generic_error_remove_page(struct address_space *mapping, struct page *page); 1850 int invalidate_inode_page(struct page *page); 1851 1852 #ifdef CONFIG_MMU 1853 extern vm_fault_t handle_mm_fault(struct vm_area_struct *vma, 1854 unsigned long address, unsigned int flags, 1855 struct pt_regs *regs); 1856 extern int fixup_user_fault(struct mm_struct *mm, 1857 unsigned long address, unsigned int fault_flags, 1858 bool *unlocked); 1859 void unmap_mapping_pages(struct address_space *mapping, 1860 pgoff_t start, pgoff_t nr, bool even_cows); 1861 void unmap_mapping_range(struct address_space *mapping, 1862 loff_t const holebegin, loff_t const holelen, int even_cows); 1863 #else 1864 static inline vm_fault_t handle_mm_fault(struct vm_area_struct *vma, 1865 unsigned long address, unsigned int flags, 1866 struct pt_regs *regs) 1867 { 1868 /* should never happen if there's no MMU */ 1869 BUG(); 1870 return VM_FAULT_SIGBUS; 1871 } 1872 static inline int fixup_user_fault(struct mm_struct *mm, unsigned long address, 1873 unsigned int fault_flags, bool *unlocked) 1874 { 1875 /* should never happen if there's no MMU */ 1876 BUG(); 1877 return -EFAULT; 1878 } 1879 static inline void unmap_mapping_pages(struct address_space *mapping, 1880 pgoff_t start, pgoff_t nr, bool even_cows) { } 1881 static inline void unmap_mapping_range(struct address_space *mapping, 1882 loff_t const holebegin, loff_t const holelen, int even_cows) { } 1883 #endif 1884 1885 static inline void unmap_shared_mapping_range(struct address_space *mapping, 1886 loff_t const holebegin, loff_t const holelen) 1887 { 1888 unmap_mapping_range(mapping, holebegin, holelen, 0); 1889 } 1890 1891 extern int access_process_vm(struct task_struct *tsk, unsigned long addr, 1892 void *buf, int len, unsigned int gup_flags); 1893 extern int access_remote_vm(struct mm_struct *mm, unsigned long addr, 1894 void *buf, int len, unsigned int gup_flags); 1895 extern int __access_remote_vm(struct mm_struct *mm, unsigned long addr, 1896 void *buf, int len, unsigned int gup_flags); 1897 1898 long get_user_pages_remote(struct mm_struct *mm, 1899 unsigned long start, unsigned long nr_pages, 1900 unsigned int gup_flags, struct page **pages, 1901 struct vm_area_struct **vmas, int *locked); 1902 long pin_user_pages_remote(struct mm_struct *mm, 1903 unsigned long start, unsigned long nr_pages, 1904 unsigned int gup_flags, struct page **pages, 1905 struct vm_area_struct **vmas, int *locked); 1906 long get_user_pages(unsigned long start, unsigned long nr_pages, 1907 unsigned int gup_flags, struct page **pages, 1908 struct vm_area_struct **vmas); 1909 long pin_user_pages(unsigned long start, unsigned long nr_pages, 1910 unsigned int gup_flags, struct page **pages, 1911 struct vm_area_struct **vmas); 1912 long get_user_pages_locked(unsigned long start, unsigned long nr_pages, 1913 unsigned int gup_flags, struct page **pages, int *locked); 1914 long pin_user_pages_locked(unsigned long start, unsigned long nr_pages, 1915 unsigned int gup_flags, struct page **pages, int *locked); 1916 long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages, 1917 struct page **pages, unsigned int gup_flags); 1918 long pin_user_pages_unlocked(unsigned long start, unsigned long nr_pages, 1919 struct page **pages, unsigned int gup_flags); 1920 1921 int get_user_pages_fast(unsigned long start, int nr_pages, 1922 unsigned int gup_flags, struct page **pages); 1923 int pin_user_pages_fast(unsigned long start, int nr_pages, 1924 unsigned int gup_flags, struct page **pages); 1925 1926 int account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc); 1927 int __account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc, 1928 struct task_struct *task, bool bypass_rlim); 1929 1930 struct kvec; 1931 int get_kernel_pages(const struct kvec *iov, int nr_pages, int write, 1932 struct page **pages); 1933 struct page *get_dump_page(unsigned long addr); 1934 1935 extern void do_invalidatepage(struct page *page, unsigned int offset, 1936 unsigned int length); 1937 1938 bool folio_mark_dirty(struct folio *folio); 1939 bool set_page_dirty(struct page *page); 1940 int set_page_dirty_lock(struct page *page); 1941 1942 int get_cmdline(struct task_struct *task, char *buffer, int buflen); 1943 1944 extern unsigned long move_page_tables(struct vm_area_struct *vma, 1945 unsigned long old_addr, struct vm_area_struct *new_vma, 1946 unsigned long new_addr, unsigned long len, 1947 bool need_rmap_locks); 1948 1949 /* 1950 * Flags used by change_protection(). For now we make it a bitmap so 1951 * that we can pass in multiple flags just like parameters. However 1952 * for now all the callers are only use one of the flags at the same 1953 * time. 1954 */ 1955 /* Whether we should allow dirty bit accounting */ 1956 #define MM_CP_DIRTY_ACCT (1UL << 0) 1957 /* Whether this protection change is for NUMA hints */ 1958 #define MM_CP_PROT_NUMA (1UL << 1) 1959 /* Whether this change is for write protecting */ 1960 #define MM_CP_UFFD_WP (1UL << 2) /* do wp */ 1961 #define MM_CP_UFFD_WP_RESOLVE (1UL << 3) /* Resolve wp */ 1962 #define MM_CP_UFFD_WP_ALL (MM_CP_UFFD_WP | \ 1963 MM_CP_UFFD_WP_RESOLVE) 1964 1965 extern unsigned long change_protection(struct vm_area_struct *vma, unsigned long start, 1966 unsigned long end, pgprot_t newprot, 1967 unsigned long cp_flags); 1968 extern int mprotect_fixup(struct vm_area_struct *vma, 1969 struct vm_area_struct **pprev, unsigned long start, 1970 unsigned long end, unsigned long newflags); 1971 1972 /* 1973 * doesn't attempt to fault and will return short. 1974 */ 1975 int get_user_pages_fast_only(unsigned long start, int nr_pages, 1976 unsigned int gup_flags, struct page **pages); 1977 int pin_user_pages_fast_only(unsigned long start, int nr_pages, 1978 unsigned int gup_flags, struct page **pages); 1979 1980 static inline bool get_user_page_fast_only(unsigned long addr, 1981 unsigned int gup_flags, struct page **pagep) 1982 { 1983 return get_user_pages_fast_only(addr, 1, gup_flags, pagep) == 1; 1984 } 1985 /* 1986 * per-process(per-mm_struct) statistics. 1987 */ 1988 static inline unsigned long get_mm_counter(struct mm_struct *mm, int member) 1989 { 1990 long val = atomic_long_read(&mm->rss_stat.count[member]); 1991 1992 #ifdef SPLIT_RSS_COUNTING 1993 /* 1994 * counter is updated in asynchronous manner and may go to minus. 1995 * But it's never be expected number for users. 1996 */ 1997 if (val < 0) 1998 val = 0; 1999 #endif 2000 return (unsigned long)val; 2001 } 2002 2003 void mm_trace_rss_stat(struct mm_struct *mm, int member, long count); 2004 2005 static inline void add_mm_counter(struct mm_struct *mm, int member, long value) 2006 { 2007 long count = atomic_long_add_return(value, &mm->rss_stat.count[member]); 2008 2009 mm_trace_rss_stat(mm, member, count); 2010 } 2011 2012 static inline void inc_mm_counter(struct mm_struct *mm, int member) 2013 { 2014 long count = atomic_long_inc_return(&mm->rss_stat.count[member]); 2015 2016 mm_trace_rss_stat(mm, member, count); 2017 } 2018 2019 static inline void dec_mm_counter(struct mm_struct *mm, int member) 2020 { 2021 long count = atomic_long_dec_return(&mm->rss_stat.count[member]); 2022 2023 mm_trace_rss_stat(mm, member, count); 2024 } 2025 2026 /* Optimized variant when page is already known not to be PageAnon */ 2027 static inline int mm_counter_file(struct page *page) 2028 { 2029 if (PageSwapBacked(page)) 2030 return MM_SHMEMPAGES; 2031 return MM_FILEPAGES; 2032 } 2033 2034 static inline int mm_counter(struct page *page) 2035 { 2036 if (PageAnon(page)) 2037 return MM_ANONPAGES; 2038 return mm_counter_file(page); 2039 } 2040 2041 static inline unsigned long get_mm_rss(struct mm_struct *mm) 2042 { 2043 return get_mm_counter(mm, MM_FILEPAGES) + 2044 get_mm_counter(mm, MM_ANONPAGES) + 2045 get_mm_counter(mm, MM_SHMEMPAGES); 2046 } 2047 2048 static inline unsigned long get_mm_hiwater_rss(struct mm_struct *mm) 2049 { 2050 return max(mm->hiwater_rss, get_mm_rss(mm)); 2051 } 2052 2053 static inline unsigned long get_mm_hiwater_vm(struct mm_struct *mm) 2054 { 2055 return max(mm->hiwater_vm, mm->total_vm); 2056 } 2057 2058 static inline void update_hiwater_rss(struct mm_struct *mm) 2059 { 2060 unsigned long _rss = get_mm_rss(mm); 2061 2062 if ((mm)->hiwater_rss < _rss) 2063 (mm)->hiwater_rss = _rss; 2064 } 2065 2066 static inline void update_hiwater_vm(struct mm_struct *mm) 2067 { 2068 if (mm->hiwater_vm < mm->total_vm) 2069 mm->hiwater_vm = mm->total_vm; 2070 } 2071 2072 static inline void reset_mm_hiwater_rss(struct mm_struct *mm) 2073 { 2074 mm->hiwater_rss = get_mm_rss(mm); 2075 } 2076 2077 static inline void setmax_mm_hiwater_rss(unsigned long *maxrss, 2078 struct mm_struct *mm) 2079 { 2080 unsigned long hiwater_rss = get_mm_hiwater_rss(mm); 2081 2082 if (*maxrss < hiwater_rss) 2083 *maxrss = hiwater_rss; 2084 } 2085 2086 #if defined(SPLIT_RSS_COUNTING) 2087 void sync_mm_rss(struct mm_struct *mm); 2088 #else 2089 static inline void sync_mm_rss(struct mm_struct *mm) 2090 { 2091 } 2092 #endif 2093 2094 #ifndef CONFIG_ARCH_HAS_PTE_SPECIAL 2095 static inline int pte_special(pte_t pte) 2096 { 2097 return 0; 2098 } 2099 2100 static inline pte_t pte_mkspecial(pte_t pte) 2101 { 2102 return pte; 2103 } 2104 #endif 2105 2106 #ifndef CONFIG_ARCH_HAS_PTE_DEVMAP 2107 static inline int pte_devmap(pte_t pte) 2108 { 2109 return 0; 2110 } 2111 #endif 2112 2113 int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot); 2114 2115 extern pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr, 2116 spinlock_t **ptl); 2117 static inline pte_t *get_locked_pte(struct mm_struct *mm, unsigned long addr, 2118 spinlock_t **ptl) 2119 { 2120 pte_t *ptep; 2121 __cond_lock(*ptl, ptep = __get_locked_pte(mm, addr, ptl)); 2122 return ptep; 2123 } 2124 2125 #ifdef __PAGETABLE_P4D_FOLDED 2126 static inline int __p4d_alloc(struct mm_struct *mm, pgd_t *pgd, 2127 unsigned long address) 2128 { 2129 return 0; 2130 } 2131 #else 2132 int __p4d_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address); 2133 #endif 2134 2135 #if defined(__PAGETABLE_PUD_FOLDED) || !defined(CONFIG_MMU) 2136 static inline int __pud_alloc(struct mm_struct *mm, p4d_t *p4d, 2137 unsigned long address) 2138 { 2139 return 0; 2140 } 2141 static inline void mm_inc_nr_puds(struct mm_struct *mm) {} 2142 static inline void mm_dec_nr_puds(struct mm_struct *mm) {} 2143 2144 #else 2145 int __pud_alloc(struct mm_struct *mm, p4d_t *p4d, unsigned long address); 2146 2147 static inline void mm_inc_nr_puds(struct mm_struct *mm) 2148 { 2149 if (mm_pud_folded(mm)) 2150 return; 2151 atomic_long_add(PTRS_PER_PUD * sizeof(pud_t), &mm->pgtables_bytes); 2152 } 2153 2154 static inline void mm_dec_nr_puds(struct mm_struct *mm) 2155 { 2156 if (mm_pud_folded(mm)) 2157 return; 2158 atomic_long_sub(PTRS_PER_PUD * sizeof(pud_t), &mm->pgtables_bytes); 2159 } 2160 #endif 2161 2162 #if defined(__PAGETABLE_PMD_FOLDED) || !defined(CONFIG_MMU) 2163 static inline int __pmd_alloc(struct mm_struct *mm, pud_t *pud, 2164 unsigned long address) 2165 { 2166 return 0; 2167 } 2168 2169 static inline void mm_inc_nr_pmds(struct mm_struct *mm) {} 2170 static inline void mm_dec_nr_pmds(struct mm_struct *mm) {} 2171 2172 #else 2173 int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address); 2174 2175 static inline void mm_inc_nr_pmds(struct mm_struct *mm) 2176 { 2177 if (mm_pmd_folded(mm)) 2178 return; 2179 atomic_long_add(PTRS_PER_PMD * sizeof(pmd_t), &mm->pgtables_bytes); 2180 } 2181 2182 static inline void mm_dec_nr_pmds(struct mm_struct *mm) 2183 { 2184 if (mm_pmd_folded(mm)) 2185 return; 2186 atomic_long_sub(PTRS_PER_PMD * sizeof(pmd_t), &mm->pgtables_bytes); 2187 } 2188 #endif 2189 2190 #ifdef CONFIG_MMU 2191 static inline void mm_pgtables_bytes_init(struct mm_struct *mm) 2192 { 2193 atomic_long_set(&mm->pgtables_bytes, 0); 2194 } 2195 2196 static inline unsigned long mm_pgtables_bytes(const struct mm_struct *mm) 2197 { 2198 return atomic_long_read(&mm->pgtables_bytes); 2199 } 2200 2201 static inline void mm_inc_nr_ptes(struct mm_struct *mm) 2202 { 2203 atomic_long_add(PTRS_PER_PTE * sizeof(pte_t), &mm->pgtables_bytes); 2204 } 2205 2206 static inline void mm_dec_nr_ptes(struct mm_struct *mm) 2207 { 2208 atomic_long_sub(PTRS_PER_PTE * sizeof(pte_t), &mm->pgtables_bytes); 2209 } 2210 #else 2211 2212 static inline void mm_pgtables_bytes_init(struct mm_struct *mm) {} 2213 static inline unsigned long mm_pgtables_bytes(const struct mm_struct *mm) 2214 { 2215 return 0; 2216 } 2217 2218 static inline void mm_inc_nr_ptes(struct mm_struct *mm) {} 2219 static inline void mm_dec_nr_ptes(struct mm_struct *mm) {} 2220 #endif 2221 2222 int __pte_alloc(struct mm_struct *mm, pmd_t *pmd); 2223 int __pte_alloc_kernel(pmd_t *pmd); 2224 2225 #if defined(CONFIG_MMU) 2226 2227 static inline p4d_t *p4d_alloc(struct mm_struct *mm, pgd_t *pgd, 2228 unsigned long address) 2229 { 2230 return (unlikely(pgd_none(*pgd)) && __p4d_alloc(mm, pgd, address)) ? 2231 NULL : p4d_offset(pgd, address); 2232 } 2233 2234 static inline pud_t *pud_alloc(struct mm_struct *mm, p4d_t *p4d, 2235 unsigned long address) 2236 { 2237 return (unlikely(p4d_none(*p4d)) && __pud_alloc(mm, p4d, address)) ? 2238 NULL : pud_offset(p4d, address); 2239 } 2240 2241 static inline pmd_t *pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) 2242 { 2243 return (unlikely(pud_none(*pud)) && __pmd_alloc(mm, pud, address))? 2244 NULL: pmd_offset(pud, address); 2245 } 2246 #endif /* CONFIG_MMU */ 2247 2248 #if USE_SPLIT_PTE_PTLOCKS 2249 #if ALLOC_SPLIT_PTLOCKS 2250 void __init ptlock_cache_init(void); 2251 extern bool ptlock_alloc(struct page *page); 2252 extern void ptlock_free(struct page *page); 2253 2254 static inline spinlock_t *ptlock_ptr(struct page *page) 2255 { 2256 return page->ptl; 2257 } 2258 #else /* ALLOC_SPLIT_PTLOCKS */ 2259 static inline void ptlock_cache_init(void) 2260 { 2261 } 2262 2263 static inline bool ptlock_alloc(struct page *page) 2264 { 2265 return true; 2266 } 2267 2268 static inline void ptlock_free(struct page *page) 2269 { 2270 } 2271 2272 static inline spinlock_t *ptlock_ptr(struct page *page) 2273 { 2274 return &page->ptl; 2275 } 2276 #endif /* ALLOC_SPLIT_PTLOCKS */ 2277 2278 static inline spinlock_t *pte_lockptr(struct mm_struct *mm, pmd_t *pmd) 2279 { 2280 return ptlock_ptr(pmd_page(*pmd)); 2281 } 2282 2283 static inline bool ptlock_init(struct page *page) 2284 { 2285 /* 2286 * prep_new_page() initialize page->private (and therefore page->ptl) 2287 * with 0. Make sure nobody took it in use in between. 2288 * 2289 * It can happen if arch try to use slab for page table allocation: 2290 * slab code uses page->slab_cache, which share storage with page->ptl. 2291 */ 2292 VM_BUG_ON_PAGE(*(unsigned long *)&page->ptl, page); 2293 if (!ptlock_alloc(page)) 2294 return false; 2295 spin_lock_init(ptlock_ptr(page)); 2296 return true; 2297 } 2298 2299 #else /* !USE_SPLIT_PTE_PTLOCKS */ 2300 /* 2301 * We use mm->page_table_lock to guard all pagetable pages of the mm. 2302 */ 2303 static inline spinlock_t *pte_lockptr(struct mm_struct *mm, pmd_t *pmd) 2304 { 2305 return &mm->page_table_lock; 2306 } 2307 static inline void ptlock_cache_init(void) {} 2308 static inline bool ptlock_init(struct page *page) { return true; } 2309 static inline void ptlock_free(struct page *page) {} 2310 #endif /* USE_SPLIT_PTE_PTLOCKS */ 2311 2312 static inline void pgtable_init(void) 2313 { 2314 ptlock_cache_init(); 2315 pgtable_cache_init(); 2316 } 2317 2318 static inline bool pgtable_pte_page_ctor(struct page *page) 2319 { 2320 if (!ptlock_init(page)) 2321 return false; 2322 __SetPageTable(page); 2323 inc_lruvec_page_state(page, NR_PAGETABLE); 2324 return true; 2325 } 2326 2327 static inline void pgtable_pte_page_dtor(struct page *page) 2328 { 2329 ptlock_free(page); 2330 __ClearPageTable(page); 2331 dec_lruvec_page_state(page, NR_PAGETABLE); 2332 } 2333 2334 #define pte_offset_map_lock(mm, pmd, address, ptlp) \ 2335 ({ \ 2336 spinlock_t *__ptl = pte_lockptr(mm, pmd); \ 2337 pte_t *__pte = pte_offset_map(pmd, address); \ 2338 *(ptlp) = __ptl; \ 2339 spin_lock(__ptl); \ 2340 __pte; \ 2341 }) 2342 2343 #define pte_unmap_unlock(pte, ptl) do { \ 2344 spin_unlock(ptl); \ 2345 pte_unmap(pte); \ 2346 } while (0) 2347 2348 #define pte_alloc(mm, pmd) (unlikely(pmd_none(*(pmd))) && __pte_alloc(mm, pmd)) 2349 2350 #define pte_alloc_map(mm, pmd, address) \ 2351 (pte_alloc(mm, pmd) ? NULL : pte_offset_map(pmd, address)) 2352 2353 #define pte_alloc_map_lock(mm, pmd, address, ptlp) \ 2354 (pte_alloc(mm, pmd) ? \ 2355 NULL : pte_offset_map_lock(mm, pmd, address, ptlp)) 2356 2357 #define pte_alloc_kernel(pmd, address) \ 2358 ((unlikely(pmd_none(*(pmd))) && __pte_alloc_kernel(pmd))? \ 2359 NULL: pte_offset_kernel(pmd, address)) 2360 2361 #if USE_SPLIT_PMD_PTLOCKS 2362 2363 static struct page *pmd_to_page(pmd_t *pmd) 2364 { 2365 unsigned long mask = ~(PTRS_PER_PMD * sizeof(pmd_t) - 1); 2366 return virt_to_page((void *)((unsigned long) pmd & mask)); 2367 } 2368 2369 static inline spinlock_t *pmd_lockptr(struct mm_struct *mm, pmd_t *pmd) 2370 { 2371 return ptlock_ptr(pmd_to_page(pmd)); 2372 } 2373 2374 static inline bool pmd_ptlock_init(struct page *page) 2375 { 2376 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 2377 page->pmd_huge_pte = NULL; 2378 #endif 2379 return ptlock_init(page); 2380 } 2381 2382 static inline void pmd_ptlock_free(struct page *page) 2383 { 2384 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 2385 VM_BUG_ON_PAGE(page->pmd_huge_pte, page); 2386 #endif 2387 ptlock_free(page); 2388 } 2389 2390 #define pmd_huge_pte(mm, pmd) (pmd_to_page(pmd)->pmd_huge_pte) 2391 2392 #else 2393 2394 static inline spinlock_t *pmd_lockptr(struct mm_struct *mm, pmd_t *pmd) 2395 { 2396 return &mm->page_table_lock; 2397 } 2398 2399 static inline bool pmd_ptlock_init(struct page *page) { return true; } 2400 static inline void pmd_ptlock_free(struct page *page) {} 2401 2402 #define pmd_huge_pte(mm, pmd) ((mm)->pmd_huge_pte) 2403 2404 #endif 2405 2406 static inline spinlock_t *pmd_lock(struct mm_struct *mm, pmd_t *pmd) 2407 { 2408 spinlock_t *ptl = pmd_lockptr(mm, pmd); 2409 spin_lock(ptl); 2410 return ptl; 2411 } 2412 2413 static inline bool pgtable_pmd_page_ctor(struct page *page) 2414 { 2415 if (!pmd_ptlock_init(page)) 2416 return false; 2417 __SetPageTable(page); 2418 inc_lruvec_page_state(page, NR_PAGETABLE); 2419 return true; 2420 } 2421 2422 static inline void pgtable_pmd_page_dtor(struct page *page) 2423 { 2424 pmd_ptlock_free(page); 2425 __ClearPageTable(page); 2426 dec_lruvec_page_state(page, NR_PAGETABLE); 2427 } 2428 2429 /* 2430 * No scalability reason to split PUD locks yet, but follow the same pattern 2431 * as the PMD locks to make it easier if we decide to. The VM should not be 2432 * considered ready to switch to split PUD locks yet; there may be places 2433 * which need to be converted from page_table_lock. 2434 */ 2435 static inline spinlock_t *pud_lockptr(struct mm_struct *mm, pud_t *pud) 2436 { 2437 return &mm->page_table_lock; 2438 } 2439 2440 static inline spinlock_t *pud_lock(struct mm_struct *mm, pud_t *pud) 2441 { 2442 spinlock_t *ptl = pud_lockptr(mm, pud); 2443 2444 spin_lock(ptl); 2445 return ptl; 2446 } 2447 2448 extern void __init pagecache_init(void); 2449 extern void __init free_area_init_memoryless_node(int nid); 2450 extern void free_initmem(void); 2451 2452 /* 2453 * Free reserved pages within range [PAGE_ALIGN(start), end & PAGE_MASK) 2454 * into the buddy system. The freed pages will be poisoned with pattern 2455 * "poison" if it's within range [0, UCHAR_MAX]. 2456 * Return pages freed into the buddy system. 2457 */ 2458 extern unsigned long free_reserved_area(void *start, void *end, 2459 int poison, const char *s); 2460 2461 extern void adjust_managed_page_count(struct page *page, long count); 2462 extern void mem_init_print_info(void); 2463 2464 extern void reserve_bootmem_region(phys_addr_t start, phys_addr_t end); 2465 2466 /* Free the reserved page into the buddy system, so it gets managed. */ 2467 static inline void free_reserved_page(struct page *page) 2468 { 2469 ClearPageReserved(page); 2470 init_page_count(page); 2471 __free_page(page); 2472 adjust_managed_page_count(page, 1); 2473 } 2474 #define free_highmem_page(page) free_reserved_page(page) 2475 2476 static inline void mark_page_reserved(struct page *page) 2477 { 2478 SetPageReserved(page); 2479 adjust_managed_page_count(page, -1); 2480 } 2481 2482 /* 2483 * Default method to free all the __init memory into the buddy system. 2484 * The freed pages will be poisoned with pattern "poison" if it's within 2485 * range [0, UCHAR_MAX]. 2486 * Return pages freed into the buddy system. 2487 */ 2488 static inline unsigned long free_initmem_default(int poison) 2489 { 2490 extern char __init_begin[], __init_end[]; 2491 2492 return free_reserved_area(&__init_begin, &__init_end, 2493 poison, "unused kernel image (initmem)"); 2494 } 2495 2496 static inline unsigned long get_num_physpages(void) 2497 { 2498 int nid; 2499 unsigned long phys_pages = 0; 2500 2501 for_each_online_node(nid) 2502 phys_pages += node_present_pages(nid); 2503 2504 return phys_pages; 2505 } 2506 2507 /* 2508 * Using memblock node mappings, an architecture may initialise its 2509 * zones, allocate the backing mem_map and account for memory holes in an 2510 * architecture independent manner. 2511 * 2512 * An architecture is expected to register range of page frames backed by 2513 * physical memory with memblock_add[_node]() before calling 2514 * free_area_init() passing in the PFN each zone ends at. At a basic 2515 * usage, an architecture is expected to do something like 2516 * 2517 * unsigned long max_zone_pfns[MAX_NR_ZONES] = {max_dma, max_normal_pfn, 2518 * max_highmem_pfn}; 2519 * for_each_valid_physical_page_range() 2520 * memblock_add_node(base, size, nid, MEMBLOCK_NONE) 2521 * free_area_init(max_zone_pfns); 2522 */ 2523 void free_area_init(unsigned long *max_zone_pfn); 2524 unsigned long node_map_pfn_alignment(void); 2525 unsigned long __absent_pages_in_range(int nid, unsigned long start_pfn, 2526 unsigned long end_pfn); 2527 extern unsigned long absent_pages_in_range(unsigned long start_pfn, 2528 unsigned long end_pfn); 2529 extern void get_pfn_range_for_nid(unsigned int nid, 2530 unsigned long *start_pfn, unsigned long *end_pfn); 2531 extern unsigned long find_min_pfn_with_active_regions(void); 2532 2533 #ifndef CONFIG_NUMA 2534 static inline int early_pfn_to_nid(unsigned long pfn) 2535 { 2536 return 0; 2537 } 2538 #else 2539 /* please see mm/page_alloc.c */ 2540 extern int __meminit early_pfn_to_nid(unsigned long pfn); 2541 #endif 2542 2543 extern void set_dma_reserve(unsigned long new_dma_reserve); 2544 extern void memmap_init_range(unsigned long, int, unsigned long, 2545 unsigned long, unsigned long, enum meminit_context, 2546 struct vmem_altmap *, int migratetype); 2547 extern void setup_per_zone_wmarks(void); 2548 extern void calculate_min_free_kbytes(void); 2549 extern int __meminit init_per_zone_wmark_min(void); 2550 extern void mem_init(void); 2551 extern void __init mmap_init(void); 2552 extern void show_mem(unsigned int flags, nodemask_t *nodemask); 2553 extern long si_mem_available(void); 2554 extern void si_meminfo(struct sysinfo * val); 2555 extern void si_meminfo_node(struct sysinfo *val, int nid); 2556 #ifdef __HAVE_ARCH_RESERVED_KERNEL_PAGES 2557 extern unsigned long arch_reserved_kernel_pages(void); 2558 #endif 2559 2560 extern __printf(3, 4) 2561 void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...); 2562 2563 extern void setup_per_cpu_pageset(void); 2564 2565 /* page_alloc.c */ 2566 extern int min_free_kbytes; 2567 extern int watermark_boost_factor; 2568 extern int watermark_scale_factor; 2569 extern bool arch_has_descending_max_zone_pfns(void); 2570 2571 /* nommu.c */ 2572 extern atomic_long_t mmap_pages_allocated; 2573 extern int nommu_shrink_inode_mappings(struct inode *, size_t, size_t); 2574 2575 /* interval_tree.c */ 2576 void vma_interval_tree_insert(struct vm_area_struct *node, 2577 struct rb_root_cached *root); 2578 void vma_interval_tree_insert_after(struct vm_area_struct *node, 2579 struct vm_area_struct *prev, 2580 struct rb_root_cached *root); 2581 void vma_interval_tree_remove(struct vm_area_struct *node, 2582 struct rb_root_cached *root); 2583 struct vm_area_struct *vma_interval_tree_iter_first(struct rb_root_cached *root, 2584 unsigned long start, unsigned long last); 2585 struct vm_area_struct *vma_interval_tree_iter_next(struct vm_area_struct *node, 2586 unsigned long start, unsigned long last); 2587 2588 #define vma_interval_tree_foreach(vma, root, start, last) \ 2589 for (vma = vma_interval_tree_iter_first(root, start, last); \ 2590 vma; vma = vma_interval_tree_iter_next(vma, start, last)) 2591 2592 void anon_vma_interval_tree_insert(struct anon_vma_chain *node, 2593 struct rb_root_cached *root); 2594 void anon_vma_interval_tree_remove(struct anon_vma_chain *node, 2595 struct rb_root_cached *root); 2596 struct anon_vma_chain * 2597 anon_vma_interval_tree_iter_first(struct rb_root_cached *root, 2598 unsigned long start, unsigned long last); 2599 struct anon_vma_chain *anon_vma_interval_tree_iter_next( 2600 struct anon_vma_chain *node, unsigned long start, unsigned long last); 2601 #ifdef CONFIG_DEBUG_VM_RB 2602 void anon_vma_interval_tree_verify(struct anon_vma_chain *node); 2603 #endif 2604 2605 #define anon_vma_interval_tree_foreach(avc, root, start, last) \ 2606 for (avc = anon_vma_interval_tree_iter_first(root, start, last); \ 2607 avc; avc = anon_vma_interval_tree_iter_next(avc, start, last)) 2608 2609 /* mmap.c */ 2610 extern int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin); 2611 extern int __vma_adjust(struct vm_area_struct *vma, unsigned long start, 2612 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert, 2613 struct vm_area_struct *expand); 2614 static inline int vma_adjust(struct vm_area_struct *vma, unsigned long start, 2615 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert) 2616 { 2617 return __vma_adjust(vma, start, end, pgoff, insert, NULL); 2618 } 2619 extern struct vm_area_struct *vma_merge(struct mm_struct *, 2620 struct vm_area_struct *prev, unsigned long addr, unsigned long end, 2621 unsigned long vm_flags, struct anon_vma *, struct file *, pgoff_t, 2622 struct mempolicy *, struct vm_userfaultfd_ctx, const char *); 2623 extern struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *); 2624 extern int __split_vma(struct mm_struct *, struct vm_area_struct *, 2625 unsigned long addr, int new_below); 2626 extern int split_vma(struct mm_struct *, struct vm_area_struct *, 2627 unsigned long addr, int new_below); 2628 extern int insert_vm_struct(struct mm_struct *, struct vm_area_struct *); 2629 extern void __vma_link_rb(struct mm_struct *, struct vm_area_struct *, 2630 struct rb_node **, struct rb_node *); 2631 extern void unlink_file_vma(struct vm_area_struct *); 2632 extern struct vm_area_struct *copy_vma(struct vm_area_struct **, 2633 unsigned long addr, unsigned long len, pgoff_t pgoff, 2634 bool *need_rmap_locks); 2635 extern void exit_mmap(struct mm_struct *); 2636 2637 static inline int check_data_rlimit(unsigned long rlim, 2638 unsigned long new, 2639 unsigned long start, 2640 unsigned long end_data, 2641 unsigned long start_data) 2642 { 2643 if (rlim < RLIM_INFINITY) { 2644 if (((new - start) + (end_data - start_data)) > rlim) 2645 return -ENOSPC; 2646 } 2647 2648 return 0; 2649 } 2650 2651 extern int mm_take_all_locks(struct mm_struct *mm); 2652 extern void mm_drop_all_locks(struct mm_struct *mm); 2653 2654 extern int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file); 2655 extern int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file); 2656 extern struct file *get_mm_exe_file(struct mm_struct *mm); 2657 extern struct file *get_task_exe_file(struct task_struct *task); 2658 2659 extern bool may_expand_vm(struct mm_struct *, vm_flags_t, unsigned long npages); 2660 extern void vm_stat_account(struct mm_struct *, vm_flags_t, long npages); 2661 2662 extern bool vma_is_special_mapping(const struct vm_area_struct *vma, 2663 const struct vm_special_mapping *sm); 2664 extern struct vm_area_struct *_install_special_mapping(struct mm_struct *mm, 2665 unsigned long addr, unsigned long len, 2666 unsigned long flags, 2667 const struct vm_special_mapping *spec); 2668 /* This is an obsolete alternative to _install_special_mapping. */ 2669 extern int install_special_mapping(struct mm_struct *mm, 2670 unsigned long addr, unsigned long len, 2671 unsigned long flags, struct page **pages); 2672 2673 unsigned long randomize_stack_top(unsigned long stack_top); 2674 2675 extern unsigned long get_unmapped_area(struct file *, unsigned long, unsigned long, unsigned long, unsigned long); 2676 2677 extern unsigned long mmap_region(struct file *file, unsigned long addr, 2678 unsigned long len, vm_flags_t vm_flags, unsigned long pgoff, 2679 struct list_head *uf); 2680 extern unsigned long do_mmap(struct file *file, unsigned long addr, 2681 unsigned long len, unsigned long prot, unsigned long flags, 2682 unsigned long pgoff, unsigned long *populate, struct list_head *uf); 2683 extern int __do_munmap(struct mm_struct *, unsigned long, size_t, 2684 struct list_head *uf, bool downgrade); 2685 extern int do_munmap(struct mm_struct *, unsigned long, size_t, 2686 struct list_head *uf); 2687 extern int do_madvise(struct mm_struct *mm, unsigned long start, size_t len_in, int behavior); 2688 2689 #ifdef CONFIG_MMU 2690 extern int __mm_populate(unsigned long addr, unsigned long len, 2691 int ignore_errors); 2692 static inline void mm_populate(unsigned long addr, unsigned long len) 2693 { 2694 /* Ignore errors */ 2695 (void) __mm_populate(addr, len, 1); 2696 } 2697 #else 2698 static inline void mm_populate(unsigned long addr, unsigned long len) {} 2699 #endif 2700 2701 /* These take the mm semaphore themselves */ 2702 extern int __must_check vm_brk(unsigned long, unsigned long); 2703 extern int __must_check vm_brk_flags(unsigned long, unsigned long, unsigned long); 2704 extern int vm_munmap(unsigned long, size_t); 2705 extern unsigned long __must_check vm_mmap(struct file *, unsigned long, 2706 unsigned long, unsigned long, 2707 unsigned long, unsigned long); 2708 2709 struct vm_unmapped_area_info { 2710 #define VM_UNMAPPED_AREA_TOPDOWN 1 2711 unsigned long flags; 2712 unsigned long length; 2713 unsigned long low_limit; 2714 unsigned long high_limit; 2715 unsigned long align_mask; 2716 unsigned long align_offset; 2717 }; 2718 2719 extern unsigned long vm_unmapped_area(struct vm_unmapped_area_info *info); 2720 2721 /* truncate.c */ 2722 extern void truncate_inode_pages(struct address_space *, loff_t); 2723 extern void truncate_inode_pages_range(struct address_space *, 2724 loff_t lstart, loff_t lend); 2725 extern void truncate_inode_pages_final(struct address_space *); 2726 2727 /* generic vm_area_ops exported for stackable file systems */ 2728 extern vm_fault_t filemap_fault(struct vm_fault *vmf); 2729 extern vm_fault_t filemap_map_pages(struct vm_fault *vmf, 2730 pgoff_t start_pgoff, pgoff_t end_pgoff); 2731 extern vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf); 2732 2733 extern unsigned long stack_guard_gap; 2734 /* Generic expand stack which grows the stack according to GROWS{UP,DOWN} */ 2735 extern int expand_stack(struct vm_area_struct *vma, unsigned long address); 2736 2737 /* CONFIG_STACK_GROWSUP still needs to grow downwards at some places */ 2738 extern int expand_downwards(struct vm_area_struct *vma, 2739 unsigned long address); 2740 #if VM_GROWSUP 2741 extern int expand_upwards(struct vm_area_struct *vma, unsigned long address); 2742 #else 2743 #define expand_upwards(vma, address) (0) 2744 #endif 2745 2746 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */ 2747 extern struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr); 2748 extern struct vm_area_struct * find_vma_prev(struct mm_struct * mm, unsigned long addr, 2749 struct vm_area_struct **pprev); 2750 2751 /** 2752 * find_vma_intersection() - Look up the first VMA which intersects the interval 2753 * @mm: The process address space. 2754 * @start_addr: The inclusive start user address. 2755 * @end_addr: The exclusive end user address. 2756 * 2757 * Returns: The first VMA within the provided range, %NULL otherwise. Assumes 2758 * start_addr < end_addr. 2759 */ 2760 static inline 2761 struct vm_area_struct *find_vma_intersection(struct mm_struct *mm, 2762 unsigned long start_addr, 2763 unsigned long end_addr) 2764 { 2765 struct vm_area_struct *vma = find_vma(mm, start_addr); 2766 2767 if (vma && end_addr <= vma->vm_start) 2768 vma = NULL; 2769 return vma; 2770 } 2771 2772 /** 2773 * vma_lookup() - Find a VMA at a specific address 2774 * @mm: The process address space. 2775 * @addr: The user address. 2776 * 2777 * Return: The vm_area_struct at the given address, %NULL otherwise. 2778 */ 2779 static inline 2780 struct vm_area_struct *vma_lookup(struct mm_struct *mm, unsigned long addr) 2781 { 2782 struct vm_area_struct *vma = find_vma(mm, addr); 2783 2784 if (vma && addr < vma->vm_start) 2785 vma = NULL; 2786 2787 return vma; 2788 } 2789 2790 static inline unsigned long vm_start_gap(struct vm_area_struct *vma) 2791 { 2792 unsigned long vm_start = vma->vm_start; 2793 2794 if (vma->vm_flags & VM_GROWSDOWN) { 2795 vm_start -= stack_guard_gap; 2796 if (vm_start > vma->vm_start) 2797 vm_start = 0; 2798 } 2799 return vm_start; 2800 } 2801 2802 static inline unsigned long vm_end_gap(struct vm_area_struct *vma) 2803 { 2804 unsigned long vm_end = vma->vm_end; 2805 2806 if (vma->vm_flags & VM_GROWSUP) { 2807 vm_end += stack_guard_gap; 2808 if (vm_end < vma->vm_end) 2809 vm_end = -PAGE_SIZE; 2810 } 2811 return vm_end; 2812 } 2813 2814 static inline unsigned long vma_pages(struct vm_area_struct *vma) 2815 { 2816 return (vma->vm_end - vma->vm_start) >> PAGE_SHIFT; 2817 } 2818 2819 /* Look up the first VMA which exactly match the interval vm_start ... vm_end */ 2820 static inline struct vm_area_struct *find_exact_vma(struct mm_struct *mm, 2821 unsigned long vm_start, unsigned long vm_end) 2822 { 2823 struct vm_area_struct *vma = find_vma(mm, vm_start); 2824 2825 if (vma && (vma->vm_start != vm_start || vma->vm_end != vm_end)) 2826 vma = NULL; 2827 2828 return vma; 2829 } 2830 2831 static inline bool range_in_vma(struct vm_area_struct *vma, 2832 unsigned long start, unsigned long end) 2833 { 2834 return (vma && vma->vm_start <= start && end <= vma->vm_end); 2835 } 2836 2837 #ifdef CONFIG_MMU 2838 pgprot_t vm_get_page_prot(unsigned long vm_flags); 2839 void vma_set_page_prot(struct vm_area_struct *vma); 2840 #else 2841 static inline pgprot_t vm_get_page_prot(unsigned long vm_flags) 2842 { 2843 return __pgprot(0); 2844 } 2845 static inline void vma_set_page_prot(struct vm_area_struct *vma) 2846 { 2847 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 2848 } 2849 #endif 2850 2851 void vma_set_file(struct vm_area_struct *vma, struct file *file); 2852 2853 #ifdef CONFIG_NUMA_BALANCING 2854 unsigned long change_prot_numa(struct vm_area_struct *vma, 2855 unsigned long start, unsigned long end); 2856 #endif 2857 2858 struct vm_area_struct *find_extend_vma(struct mm_struct *, unsigned long addr); 2859 int remap_pfn_range(struct vm_area_struct *, unsigned long addr, 2860 unsigned long pfn, unsigned long size, pgprot_t); 2861 int remap_pfn_range_notrack(struct vm_area_struct *vma, unsigned long addr, 2862 unsigned long pfn, unsigned long size, pgprot_t prot); 2863 int vm_insert_page(struct vm_area_struct *, unsigned long addr, struct page *); 2864 int vm_insert_pages(struct vm_area_struct *vma, unsigned long addr, 2865 struct page **pages, unsigned long *num); 2866 int vm_map_pages(struct vm_area_struct *vma, struct page **pages, 2867 unsigned long num); 2868 int vm_map_pages_zero(struct vm_area_struct *vma, struct page **pages, 2869 unsigned long num); 2870 vm_fault_t vmf_insert_pfn(struct vm_area_struct *vma, unsigned long addr, 2871 unsigned long pfn); 2872 vm_fault_t vmf_insert_pfn_prot(struct vm_area_struct *vma, unsigned long addr, 2873 unsigned long pfn, pgprot_t pgprot); 2874 vm_fault_t vmf_insert_mixed(struct vm_area_struct *vma, unsigned long addr, 2875 pfn_t pfn); 2876 vm_fault_t vmf_insert_mixed_prot(struct vm_area_struct *vma, unsigned long addr, 2877 pfn_t pfn, pgprot_t pgprot); 2878 vm_fault_t vmf_insert_mixed_mkwrite(struct vm_area_struct *vma, 2879 unsigned long addr, pfn_t pfn); 2880 int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len); 2881 2882 static inline vm_fault_t vmf_insert_page(struct vm_area_struct *vma, 2883 unsigned long addr, struct page *page) 2884 { 2885 int err = vm_insert_page(vma, addr, page); 2886 2887 if (err == -ENOMEM) 2888 return VM_FAULT_OOM; 2889 if (err < 0 && err != -EBUSY) 2890 return VM_FAULT_SIGBUS; 2891 2892 return VM_FAULT_NOPAGE; 2893 } 2894 2895 #ifndef io_remap_pfn_range 2896 static inline int io_remap_pfn_range(struct vm_area_struct *vma, 2897 unsigned long addr, unsigned long pfn, 2898 unsigned long size, pgprot_t prot) 2899 { 2900 return remap_pfn_range(vma, addr, pfn, size, pgprot_decrypted(prot)); 2901 } 2902 #endif 2903 2904 static inline vm_fault_t vmf_error(int err) 2905 { 2906 if (err == -ENOMEM) 2907 return VM_FAULT_OOM; 2908 return VM_FAULT_SIGBUS; 2909 } 2910 2911 struct page *follow_page(struct vm_area_struct *vma, unsigned long address, 2912 unsigned int foll_flags); 2913 2914 #define FOLL_WRITE 0x01 /* check pte is writable */ 2915 #define FOLL_TOUCH 0x02 /* mark page accessed */ 2916 #define FOLL_GET 0x04 /* do get_page on page */ 2917 #define FOLL_DUMP 0x08 /* give error on hole if it would be zero */ 2918 #define FOLL_FORCE 0x10 /* get_user_pages read/write w/o permission */ 2919 #define FOLL_NOWAIT 0x20 /* if a disk transfer is needed, start the IO 2920 * and return without waiting upon it */ 2921 #define FOLL_POPULATE 0x40 /* fault in pages (with FOLL_MLOCK) */ 2922 #define FOLL_NOFAULT 0x80 /* do not fault in pages */ 2923 #define FOLL_HWPOISON 0x100 /* check page is hwpoisoned */ 2924 #define FOLL_NUMA 0x200 /* force NUMA hinting page fault */ 2925 #define FOLL_MIGRATION 0x400 /* wait for page to replace migration entry */ 2926 #define FOLL_TRIED 0x800 /* a retry, previous pass started an IO */ 2927 #define FOLL_MLOCK 0x1000 /* lock present pages */ 2928 #define FOLL_REMOTE 0x2000 /* we are working on non-current tsk/mm */ 2929 #define FOLL_COW 0x4000 /* internal GUP flag */ 2930 #define FOLL_ANON 0x8000 /* don't do file mappings */ 2931 #define FOLL_LONGTERM 0x10000 /* mapping lifetime is indefinite: see below */ 2932 #define FOLL_SPLIT_PMD 0x20000 /* split huge pmd before returning */ 2933 #define FOLL_PIN 0x40000 /* pages must be released via unpin_user_page */ 2934 #define FOLL_FAST_ONLY 0x80000 /* gup_fast: prevent fall-back to slow gup */ 2935 2936 /* 2937 * FOLL_PIN and FOLL_LONGTERM may be used in various combinations with each 2938 * other. Here is what they mean, and how to use them: 2939 * 2940 * FOLL_LONGTERM indicates that the page will be held for an indefinite time 2941 * period _often_ under userspace control. This is in contrast to 2942 * iov_iter_get_pages(), whose usages are transient. 2943 * 2944 * FIXME: For pages which are part of a filesystem, mappings are subject to the 2945 * lifetime enforced by the filesystem and we need guarantees that longterm 2946 * users like RDMA and V4L2 only establish mappings which coordinate usage with 2947 * the filesystem. Ideas for this coordination include revoking the longterm 2948 * pin, delaying writeback, bounce buffer page writeback, etc. As FS DAX was 2949 * added after the problem with filesystems was found FS DAX VMAs are 2950 * specifically failed. Filesystem pages are still subject to bugs and use of 2951 * FOLL_LONGTERM should be avoided on those pages. 2952 * 2953 * FIXME: Also NOTE that FOLL_LONGTERM is not supported in every GUP call. 2954 * Currently only get_user_pages() and get_user_pages_fast() support this flag 2955 * and calls to get_user_pages_[un]locked are specifically not allowed. This 2956 * is due to an incompatibility with the FS DAX check and 2957 * FAULT_FLAG_ALLOW_RETRY. 2958 * 2959 * In the CMA case: long term pins in a CMA region would unnecessarily fragment 2960 * that region. And so, CMA attempts to migrate the page before pinning, when 2961 * FOLL_LONGTERM is specified. 2962 * 2963 * FOLL_PIN indicates that a special kind of tracking (not just page->_refcount, 2964 * but an additional pin counting system) will be invoked. This is intended for 2965 * anything that gets a page reference and then touches page data (for example, 2966 * Direct IO). This lets the filesystem know that some non-file-system entity is 2967 * potentially changing the pages' data. In contrast to FOLL_GET (whose pages 2968 * are released via put_page()), FOLL_PIN pages must be released, ultimately, by 2969 * a call to unpin_user_page(). 2970 * 2971 * FOLL_PIN is similar to FOLL_GET: both of these pin pages. They use different 2972 * and separate refcounting mechanisms, however, and that means that each has 2973 * its own acquire and release mechanisms: 2974 * 2975 * FOLL_GET: get_user_pages*() to acquire, and put_page() to release. 2976 * 2977 * FOLL_PIN: pin_user_pages*() to acquire, and unpin_user_pages to release. 2978 * 2979 * FOLL_PIN and FOLL_GET are mutually exclusive for a given function call. 2980 * (The underlying pages may experience both FOLL_GET-based and FOLL_PIN-based 2981 * calls applied to them, and that's perfectly OK. This is a constraint on the 2982 * callers, not on the pages.) 2983 * 2984 * FOLL_PIN should be set internally by the pin_user_pages*() APIs, never 2985 * directly by the caller. That's in order to help avoid mismatches when 2986 * releasing pages: get_user_pages*() pages must be released via put_page(), 2987 * while pin_user_pages*() pages must be released via unpin_user_page(). 2988 * 2989 * Please see Documentation/core-api/pin_user_pages.rst for more information. 2990 */ 2991 2992 static inline int vm_fault_to_errno(vm_fault_t vm_fault, int foll_flags) 2993 { 2994 if (vm_fault & VM_FAULT_OOM) 2995 return -ENOMEM; 2996 if (vm_fault & (VM_FAULT_HWPOISON | VM_FAULT_HWPOISON_LARGE)) 2997 return (foll_flags & FOLL_HWPOISON) ? -EHWPOISON : -EFAULT; 2998 if (vm_fault & (VM_FAULT_SIGBUS | VM_FAULT_SIGSEGV)) 2999 return -EFAULT; 3000 return 0; 3001 } 3002 3003 typedef int (*pte_fn_t)(pte_t *pte, unsigned long addr, void *data); 3004 extern int apply_to_page_range(struct mm_struct *mm, unsigned long address, 3005 unsigned long size, pte_fn_t fn, void *data); 3006 extern int apply_to_existing_page_range(struct mm_struct *mm, 3007 unsigned long address, unsigned long size, 3008 pte_fn_t fn, void *data); 3009 3010 extern void init_mem_debugging_and_hardening(void); 3011 #ifdef CONFIG_PAGE_POISONING 3012 extern void __kernel_poison_pages(struct page *page, int numpages); 3013 extern void __kernel_unpoison_pages(struct page *page, int numpages); 3014 extern bool _page_poisoning_enabled_early; 3015 DECLARE_STATIC_KEY_FALSE(_page_poisoning_enabled); 3016 static inline bool page_poisoning_enabled(void) 3017 { 3018 return _page_poisoning_enabled_early; 3019 } 3020 /* 3021 * For use in fast paths after init_mem_debugging() has run, or when a 3022 * false negative result is not harmful when called too early. 3023 */ 3024 static inline bool page_poisoning_enabled_static(void) 3025 { 3026 return static_branch_unlikely(&_page_poisoning_enabled); 3027 } 3028 static inline void kernel_poison_pages(struct page *page, int numpages) 3029 { 3030 if (page_poisoning_enabled_static()) 3031 __kernel_poison_pages(page, numpages); 3032 } 3033 static inline void kernel_unpoison_pages(struct page *page, int numpages) 3034 { 3035 if (page_poisoning_enabled_static()) 3036 __kernel_unpoison_pages(page, numpages); 3037 } 3038 #else 3039 static inline bool page_poisoning_enabled(void) { return false; } 3040 static inline bool page_poisoning_enabled_static(void) { return false; } 3041 static inline void __kernel_poison_pages(struct page *page, int nunmpages) { } 3042 static inline void kernel_poison_pages(struct page *page, int numpages) { } 3043 static inline void kernel_unpoison_pages(struct page *page, int numpages) { } 3044 #endif 3045 3046 DECLARE_STATIC_KEY_MAYBE(CONFIG_INIT_ON_ALLOC_DEFAULT_ON, init_on_alloc); 3047 static inline bool want_init_on_alloc(gfp_t flags) 3048 { 3049 if (static_branch_maybe(CONFIG_INIT_ON_ALLOC_DEFAULT_ON, 3050 &init_on_alloc)) 3051 return true; 3052 return flags & __GFP_ZERO; 3053 } 3054 3055 DECLARE_STATIC_KEY_MAYBE(CONFIG_INIT_ON_FREE_DEFAULT_ON, init_on_free); 3056 static inline bool want_init_on_free(void) 3057 { 3058 return static_branch_maybe(CONFIG_INIT_ON_FREE_DEFAULT_ON, 3059 &init_on_free); 3060 } 3061 3062 extern bool _debug_pagealloc_enabled_early; 3063 DECLARE_STATIC_KEY_FALSE(_debug_pagealloc_enabled); 3064 3065 static inline bool debug_pagealloc_enabled(void) 3066 { 3067 return IS_ENABLED(CONFIG_DEBUG_PAGEALLOC) && 3068 _debug_pagealloc_enabled_early; 3069 } 3070 3071 /* 3072 * For use in fast paths after init_debug_pagealloc() has run, or when a 3073 * false negative result is not harmful when called too early. 3074 */ 3075 static inline bool debug_pagealloc_enabled_static(void) 3076 { 3077 if (!IS_ENABLED(CONFIG_DEBUG_PAGEALLOC)) 3078 return false; 3079 3080 return static_branch_unlikely(&_debug_pagealloc_enabled); 3081 } 3082 3083 #ifdef CONFIG_DEBUG_PAGEALLOC 3084 /* 3085 * To support DEBUG_PAGEALLOC architecture must ensure that 3086 * __kernel_map_pages() never fails 3087 */ 3088 extern void __kernel_map_pages(struct page *page, int numpages, int enable); 3089 3090 static inline void debug_pagealloc_map_pages(struct page *page, int numpages) 3091 { 3092 if (debug_pagealloc_enabled_static()) 3093 __kernel_map_pages(page, numpages, 1); 3094 } 3095 3096 static inline void debug_pagealloc_unmap_pages(struct page *page, int numpages) 3097 { 3098 if (debug_pagealloc_enabled_static()) 3099 __kernel_map_pages(page, numpages, 0); 3100 } 3101 #else /* CONFIG_DEBUG_PAGEALLOC */ 3102 static inline void debug_pagealloc_map_pages(struct page *page, int numpages) {} 3103 static inline void debug_pagealloc_unmap_pages(struct page *page, int numpages) {} 3104 #endif /* CONFIG_DEBUG_PAGEALLOC */ 3105 3106 #ifdef __HAVE_ARCH_GATE_AREA 3107 extern struct vm_area_struct *get_gate_vma(struct mm_struct *mm); 3108 extern int in_gate_area_no_mm(unsigned long addr); 3109 extern int in_gate_area(struct mm_struct *mm, unsigned long addr); 3110 #else 3111 static inline struct vm_area_struct *get_gate_vma(struct mm_struct *mm) 3112 { 3113 return NULL; 3114 } 3115 static inline int in_gate_area_no_mm(unsigned long addr) { return 0; } 3116 static inline int in_gate_area(struct mm_struct *mm, unsigned long addr) 3117 { 3118 return 0; 3119 } 3120 #endif /* __HAVE_ARCH_GATE_AREA */ 3121 3122 extern bool process_shares_mm(struct task_struct *p, struct mm_struct *mm); 3123 3124 #ifdef CONFIG_SYSCTL 3125 extern int sysctl_drop_caches; 3126 int drop_caches_sysctl_handler(struct ctl_table *, int, void *, size_t *, 3127 loff_t *); 3128 #endif 3129 3130 void drop_slab(void); 3131 3132 #ifndef CONFIG_MMU 3133 #define randomize_va_space 0 3134 #else 3135 extern int randomize_va_space; 3136 #endif 3137 3138 const char * arch_vma_name(struct vm_area_struct *vma); 3139 #ifdef CONFIG_MMU 3140 void print_vma_addr(char *prefix, unsigned long rip); 3141 #else 3142 static inline void print_vma_addr(char *prefix, unsigned long rip) 3143 { 3144 } 3145 #endif 3146 3147 int vmemmap_remap_free(unsigned long start, unsigned long end, 3148 unsigned long reuse); 3149 int vmemmap_remap_alloc(unsigned long start, unsigned long end, 3150 unsigned long reuse, gfp_t gfp_mask); 3151 3152 void *sparse_buffer_alloc(unsigned long size); 3153 struct page * __populate_section_memmap(unsigned long pfn, 3154 unsigned long nr_pages, int nid, struct vmem_altmap *altmap); 3155 pgd_t *vmemmap_pgd_populate(unsigned long addr, int node); 3156 p4d_t *vmemmap_p4d_populate(pgd_t *pgd, unsigned long addr, int node); 3157 pud_t *vmemmap_pud_populate(p4d_t *p4d, unsigned long addr, int node); 3158 pmd_t *vmemmap_pmd_populate(pud_t *pud, unsigned long addr, int node); 3159 pte_t *vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node, 3160 struct vmem_altmap *altmap); 3161 void *vmemmap_alloc_block(unsigned long size, int node); 3162 struct vmem_altmap; 3163 void *vmemmap_alloc_block_buf(unsigned long size, int node, 3164 struct vmem_altmap *altmap); 3165 void vmemmap_verify(pte_t *, int, unsigned long, unsigned long); 3166 int vmemmap_populate_basepages(unsigned long start, unsigned long end, 3167 int node, struct vmem_altmap *altmap); 3168 int vmemmap_populate(unsigned long start, unsigned long end, int node, 3169 struct vmem_altmap *altmap); 3170 void vmemmap_populate_print_last(void); 3171 #ifdef CONFIG_MEMORY_HOTPLUG 3172 void vmemmap_free(unsigned long start, unsigned long end, 3173 struct vmem_altmap *altmap); 3174 #endif 3175 void register_page_bootmem_memmap(unsigned long section_nr, struct page *map, 3176 unsigned long nr_pages); 3177 3178 enum mf_flags { 3179 MF_COUNT_INCREASED = 1 << 0, 3180 MF_ACTION_REQUIRED = 1 << 1, 3181 MF_MUST_KILL = 1 << 2, 3182 MF_SOFT_OFFLINE = 1 << 3, 3183 MF_UNPOISON = 1 << 4, 3184 }; 3185 extern int memory_failure(unsigned long pfn, int flags); 3186 extern void memory_failure_queue(unsigned long pfn, int flags); 3187 extern void memory_failure_queue_kick(int cpu); 3188 extern int unpoison_memory(unsigned long pfn); 3189 extern int sysctl_memory_failure_early_kill; 3190 extern int sysctl_memory_failure_recovery; 3191 extern void shake_page(struct page *p); 3192 extern atomic_long_t num_poisoned_pages __read_mostly; 3193 extern int soft_offline_page(unsigned long pfn, int flags); 3194 3195 #ifndef arch_memory_failure 3196 static inline int arch_memory_failure(unsigned long pfn, int flags) 3197 { 3198 return -ENXIO; 3199 } 3200 #endif 3201 3202 #ifndef arch_is_platform_page 3203 static inline bool arch_is_platform_page(u64 paddr) 3204 { 3205 return false; 3206 } 3207 #endif 3208 3209 /* 3210 * Error handlers for various types of pages. 3211 */ 3212 enum mf_result { 3213 MF_IGNORED, /* Error: cannot be handled */ 3214 MF_FAILED, /* Error: handling failed */ 3215 MF_DELAYED, /* Will be handled later */ 3216 MF_RECOVERED, /* Successfully recovered */ 3217 }; 3218 3219 enum mf_action_page_type { 3220 MF_MSG_KERNEL, 3221 MF_MSG_KERNEL_HIGH_ORDER, 3222 MF_MSG_SLAB, 3223 MF_MSG_DIFFERENT_COMPOUND, 3224 MF_MSG_HUGE, 3225 MF_MSG_FREE_HUGE, 3226 MF_MSG_NON_PMD_HUGE, 3227 MF_MSG_UNMAP_FAILED, 3228 MF_MSG_DIRTY_SWAPCACHE, 3229 MF_MSG_CLEAN_SWAPCACHE, 3230 MF_MSG_DIRTY_MLOCKED_LRU, 3231 MF_MSG_CLEAN_MLOCKED_LRU, 3232 MF_MSG_DIRTY_UNEVICTABLE_LRU, 3233 MF_MSG_CLEAN_UNEVICTABLE_LRU, 3234 MF_MSG_DIRTY_LRU, 3235 MF_MSG_CLEAN_LRU, 3236 MF_MSG_TRUNCATED_LRU, 3237 MF_MSG_BUDDY, 3238 MF_MSG_DAX, 3239 MF_MSG_UNSPLIT_THP, 3240 MF_MSG_UNKNOWN, 3241 }; 3242 3243 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) 3244 extern void clear_huge_page(struct page *page, 3245 unsigned long addr_hint, 3246 unsigned int pages_per_huge_page); 3247 extern void copy_user_huge_page(struct page *dst, struct page *src, 3248 unsigned long addr_hint, 3249 struct vm_area_struct *vma, 3250 unsigned int pages_per_huge_page); 3251 extern long copy_huge_page_from_user(struct page *dst_page, 3252 const void __user *usr_src, 3253 unsigned int pages_per_huge_page, 3254 bool allow_pagefault); 3255 3256 /** 3257 * vma_is_special_huge - Are transhuge page-table entries considered special? 3258 * @vma: Pointer to the struct vm_area_struct to consider 3259 * 3260 * Whether transhuge page-table entries are considered "special" following 3261 * the definition in vm_normal_page(). 3262 * 3263 * Return: true if transhuge page-table entries should be considered special, 3264 * false otherwise. 3265 */ 3266 static inline bool vma_is_special_huge(const struct vm_area_struct *vma) 3267 { 3268 return vma_is_dax(vma) || (vma->vm_file && 3269 (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP))); 3270 } 3271 3272 #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */ 3273 3274 #ifdef CONFIG_DEBUG_PAGEALLOC 3275 extern unsigned int _debug_guardpage_minorder; 3276 DECLARE_STATIC_KEY_FALSE(_debug_guardpage_enabled); 3277 3278 static inline unsigned int debug_guardpage_minorder(void) 3279 { 3280 return _debug_guardpage_minorder; 3281 } 3282 3283 static inline bool debug_guardpage_enabled(void) 3284 { 3285 return static_branch_unlikely(&_debug_guardpage_enabled); 3286 } 3287 3288 static inline bool page_is_guard(struct page *page) 3289 { 3290 if (!debug_guardpage_enabled()) 3291 return false; 3292 3293 return PageGuard(page); 3294 } 3295 #else 3296 static inline unsigned int debug_guardpage_minorder(void) { return 0; } 3297 static inline bool debug_guardpage_enabled(void) { return false; } 3298 static inline bool page_is_guard(struct page *page) { return false; } 3299 #endif /* CONFIG_DEBUG_PAGEALLOC */ 3300 3301 #if MAX_NUMNODES > 1 3302 void __init setup_nr_node_ids(void); 3303 #else 3304 static inline void setup_nr_node_ids(void) {} 3305 #endif 3306 3307 extern int memcmp_pages(struct page *page1, struct page *page2); 3308 3309 static inline int pages_identical(struct page *page1, struct page *page2) 3310 { 3311 return !memcmp_pages(page1, page2); 3312 } 3313 3314 #ifdef CONFIG_MAPPING_DIRTY_HELPERS 3315 unsigned long clean_record_shared_mapping_range(struct address_space *mapping, 3316 pgoff_t first_index, pgoff_t nr, 3317 pgoff_t bitmap_pgoff, 3318 unsigned long *bitmap, 3319 pgoff_t *start, 3320 pgoff_t *end); 3321 3322 unsigned long wp_shared_mapping_range(struct address_space *mapping, 3323 pgoff_t first_index, pgoff_t nr); 3324 #endif 3325 3326 extern int sysctl_nr_trim_pages; 3327 3328 #ifdef CONFIG_PRINTK 3329 void mem_dump_obj(void *object); 3330 #else 3331 static inline void mem_dump_obj(void *object) {} 3332 #endif 3333 3334 /** 3335 * seal_check_future_write - Check for F_SEAL_FUTURE_WRITE flag and handle it 3336 * @seals: the seals to check 3337 * @vma: the vma to operate on 3338 * 3339 * Check whether F_SEAL_FUTURE_WRITE is set; if so, do proper check/handling on 3340 * the vma flags. Return 0 if check pass, or <0 for errors. 3341 */ 3342 static inline int seal_check_future_write(int seals, struct vm_area_struct *vma) 3343 { 3344 if (seals & F_SEAL_FUTURE_WRITE) { 3345 /* 3346 * New PROT_WRITE and MAP_SHARED mmaps are not allowed when 3347 * "future write" seal active. 3348 */ 3349 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_WRITE)) 3350 return -EPERM; 3351 3352 /* 3353 * Since an F_SEAL_FUTURE_WRITE sealed memfd can be mapped as 3354 * MAP_SHARED and read-only, take care to not allow mprotect to 3355 * revert protections on such mappings. Do this only for shared 3356 * mappings. For private mappings, don't need to mask 3357 * VM_MAYWRITE as we still want them to be COW-writable. 3358 */ 3359 if (vma->vm_flags & VM_SHARED) 3360 vma->vm_flags &= ~(VM_MAYWRITE); 3361 } 3362 3363 return 0; 3364 } 3365 3366 #ifdef CONFIG_ANON_VMA_NAME 3367 int madvise_set_anon_name(struct mm_struct *mm, unsigned long start, 3368 unsigned long len_in, const char *name); 3369 #else 3370 static inline int 3371 madvise_set_anon_name(struct mm_struct *mm, unsigned long start, 3372 unsigned long len_in, const char *name) { 3373 return 0; 3374 } 3375 #endif 3376 3377 #endif /* __KERNEL__ */ 3378 #endif /* _LINUX_MM_H */ 3379