1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_MM_TYPES_H
3 #define _LINUX_MM_TYPES_H
4
5 #include <linux/mm_types_task.h>
6
7 #include <linux/auxvec.h>
8 #include <linux/list.h>
9 #include <linux/spinlock.h>
10 #include <linux/rbtree.h>
11 #include <linux/rwsem.h>
12 #include <linux/completion.h>
13 #include <linux/cpumask.h>
14 #include <linux/uprobes.h>
15 #include <linux/page-flags-layout.h>
16 #include <linux/workqueue.h>
17
18 #include <asm/mmu.h>
19
20 #ifndef AT_VECTOR_SIZE_ARCH
21 #define AT_VECTOR_SIZE_ARCH 0
22 #endif
23 #define AT_VECTOR_SIZE (2*(AT_VECTOR_SIZE_ARCH + AT_VECTOR_SIZE_BASE + 1))
24
25 typedef int vm_fault_t;
26
27 struct address_space;
28 struct mem_cgroup;
29 struct hmm;
30
31 /*
32 * Each physical page in the system has a struct page associated with
33 * it to keep track of whatever it is we are using the page for at the
34 * moment. Note that we have no way to track which tasks are using
35 * a page, though if it is a pagecache page, rmap structures can tell us
36 * who is mapping it.
37 *
38 * If you allocate the page using alloc_pages(), you can use some of the
39 * space in struct page for your own purposes. The five words in the main
40 * union are available, except for bit 0 of the first word which must be
41 * kept clear. Many users use this word to store a pointer to an object
42 * which is guaranteed to be aligned. If you use the same storage as
43 * page->mapping, you must restore it to NULL before freeing the page.
44 *
45 * If your page will not be mapped to userspace, you can also use the four
46 * bytes in the mapcount union, but you must call page_mapcount_reset()
47 * before freeing it.
48 *
49 * If you want to use the refcount field, it must be used in such a way
50 * that other CPUs temporarily incrementing and then decrementing the
51 * refcount does not cause problems. On receiving the page from
52 * alloc_pages(), the refcount will be positive.
53 *
54 * If you allocate pages of order > 0, you can use some of the fields
55 * in each subpage, but you may need to restore some of their values
56 * afterwards.
57 *
58 * SLUB uses cmpxchg_double() to atomically update its freelist and
59 * counters. That requires that freelist & counters be adjacent and
60 * double-word aligned. We align all struct pages to double-word
61 * boundaries, and ensure that 'freelist' is aligned within the
62 * struct.
63 */
64 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
65 #define _struct_page_alignment __aligned(2 * sizeof(unsigned long))
66 #else
67 #define _struct_page_alignment
68 #endif
69
70 struct page {
71 unsigned long flags; /* Atomic flags, some possibly
72 * updated asynchronously */
73 /*
74 * Five words (20/40 bytes) are available in this union.
75 * WARNING: bit 0 of the first word is used for PageTail(). That
76 * means the other users of this union MUST NOT use the bit to
77 * avoid collision and false-positive PageTail().
78 */
79 union {
80 struct { /* Page cache and anonymous pages */
81 /**
82 * @lru: Pageout list, eg. active_list protected by
83 * zone_lru_lock. Sometimes used as a generic list
84 * by the page owner.
85 */
86 struct list_head lru;
87 /* See page-flags.h for PAGE_MAPPING_FLAGS */
88 struct address_space *mapping;
89 pgoff_t index; /* Our offset within mapping. */
90 /**
91 * @private: Mapping-private opaque data.
92 * Usually used for buffer_heads if PagePrivate.
93 * Used for swp_entry_t if PageSwapCache.
94 * Indicates order in the buddy system if PageBuddy.
95 */
96 unsigned long private;
97 };
98 struct { /* slab, slob and slub */
99 union {
100 struct list_head slab_list; /* uses lru */
101 struct { /* Partial pages */
102 struct page *next;
103 #ifdef CONFIG_64BIT
104 int pages; /* Nr of pages left */
105 int pobjects; /* Approximate count */
106 #else
107 short int pages;
108 short int pobjects;
109 #endif
110 };
111 };
112 struct kmem_cache *slab_cache; /* not slob */
113 /* Double-word boundary */
114 void *freelist; /* first free object */
115 union {
116 void *s_mem; /* slab: first object */
117 unsigned long counters; /* SLUB */
118 struct { /* SLUB */
119 unsigned inuse:16;
120 unsigned objects:15;
121 unsigned frozen:1;
122 };
123 };
124 };
125 struct { /* Tail pages of compound page */
126 unsigned long compound_head; /* Bit zero is set */
127
128 /* First tail page only */
129 unsigned char compound_dtor;
130 unsigned char compound_order;
131 atomic_t compound_mapcount;
132 };
133 struct { /* Second tail page of compound page */
134 unsigned long _compound_pad_1; /* compound_head */
135 unsigned long _compound_pad_2;
136 struct list_head deferred_list;
137 };
138 struct { /* Page table pages */
139 unsigned long _pt_pad_1; /* compound_head */
140 pgtable_t pmd_huge_pte; /* protected by page->ptl */
141 unsigned long _pt_pad_2; /* mapping */
142 union {
143 struct mm_struct *pt_mm; /* x86 pgds only */
144 atomic_t pt_frag_refcount; /* powerpc */
145 };
146 #if ALLOC_SPLIT_PTLOCKS
147 spinlock_t *ptl;
148 #else
149 spinlock_t ptl;
150 #endif
151 };
152 struct { /* ZONE_DEVICE pages */
153 /** @pgmap: Points to the hosting device page map. */
154 struct dev_pagemap *pgmap;
155 unsigned long hmm_data;
156 unsigned long _zd_pad_1; /* uses mapping */
157 };
158
159 /** @rcu_head: You can use this to free a page by RCU. */
160 struct rcu_head rcu_head;
161 };
162
163 union { /* This union is 4 bytes in size. */
164 /*
165 * If the page can be mapped to userspace, encodes the number
166 * of times this page is referenced by a page table.
167 */
168 atomic_t _mapcount;
169
170 /*
171 * If the page is neither PageSlab nor mappable to userspace,
172 * the value stored here may help determine what this page
173 * is used for. See page-flags.h for a list of page types
174 * which are currently stored here.
175 */
176 unsigned int page_type;
177
178 unsigned int active; /* SLAB */
179 int units; /* SLOB */
180 };
181
182 /* Usage count. *DO NOT USE DIRECTLY*. See page_ref.h */
183 atomic_t _refcount;
184
185 #ifdef CONFIG_MEMCG
186 struct mem_cgroup *mem_cgroup;
187 #endif
188
189 /*
190 * On machines where all RAM is mapped into kernel address space,
191 * we can simply calculate the virtual address. On machines with
192 * highmem some memory is mapped into kernel virtual memory
193 * dynamically, so we need a place to store that address.
194 * Note that this field could be 16 bits on x86 ... ;)
195 *
196 * Architectures with slow multiplication can define
197 * WANT_PAGE_VIRTUAL in asm/page.h
198 */
199 #if defined(WANT_PAGE_VIRTUAL)
200 void *virtual; /* Kernel virtual address (NULL if
201 not kmapped, ie. highmem) */
202 #endif /* WANT_PAGE_VIRTUAL */
203
204 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
205 int _last_cpupid;
206 #endif
207 } _struct_page_alignment;
208
209 #define PAGE_FRAG_CACHE_MAX_SIZE __ALIGN_MASK(32768, ~PAGE_MASK)
210 #define PAGE_FRAG_CACHE_MAX_ORDER get_order(PAGE_FRAG_CACHE_MAX_SIZE)
211
212 struct page_frag_cache {
213 void * va;
214 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
215 __u16 offset;
216 __u16 size;
217 #else
218 __u32 offset;
219 #endif
220 /* we maintain a pagecount bias, so that we dont dirty cache line
221 * containing page->_refcount every time we allocate a fragment.
222 */
223 unsigned int pagecnt_bias;
224 bool pfmemalloc;
225 };
226
227 typedef unsigned long vm_flags_t;
228
compound_mapcount_ptr(struct page * page)229 static inline atomic_t *compound_mapcount_ptr(struct page *page)
230 {
231 return &page[1].compound_mapcount;
232 }
233
234 /*
235 * A region containing a mapping of a non-memory backed file under NOMMU
236 * conditions. These are held in a global tree and are pinned by the VMAs that
237 * map parts of them.
238 */
239 struct vm_region {
240 struct rb_node vm_rb; /* link in global region tree */
241 vm_flags_t vm_flags; /* VMA vm_flags */
242 unsigned long vm_start; /* start address of region */
243 unsigned long vm_end; /* region initialised to here */
244 unsigned long vm_top; /* region allocated to here */
245 unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */
246 struct file *vm_file; /* the backing file or NULL */
247
248 int vm_usage; /* region usage count (access under nommu_region_sem) */
249 bool vm_icache_flushed : 1; /* true if the icache has been flushed for
250 * this region */
251 };
252
253 #ifdef CONFIG_USERFAULTFD
254 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
255 struct vm_userfaultfd_ctx {
256 struct userfaultfd_ctx *ctx;
257 };
258 #else /* CONFIG_USERFAULTFD */
259 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
260 struct vm_userfaultfd_ctx {};
261 #endif /* CONFIG_USERFAULTFD */
262
263 /*
264 * This struct defines a memory VMM memory area. There is one of these
265 * per VM-area/task. A VM area is any part of the process virtual memory
266 * space that has a special rule for the page-fault handlers (ie a shared
267 * library, the executable area etc).
268 */
269 struct vm_area_struct {
270 /* The first cache line has the info for VMA tree walking. */
271
272 unsigned long vm_start; /* Our start address within vm_mm. */
273 unsigned long vm_end; /* The first byte after our end address
274 within vm_mm. */
275
276 /* linked list of VM areas per task, sorted by address */
277 struct vm_area_struct *vm_next, *vm_prev;
278
279 struct rb_node vm_rb;
280
281 /*
282 * Largest free memory gap in bytes to the left of this VMA.
283 * Either between this VMA and vma->vm_prev, or between one of the
284 * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
285 * get_unmapped_area find a free area of the right size.
286 */
287 unsigned long rb_subtree_gap;
288
289 /* Second cache line starts here. */
290
291 struct mm_struct *vm_mm; /* The address space we belong to. */
292 pgprot_t vm_page_prot; /* Access permissions of this VMA. */
293 unsigned long vm_flags; /* Flags, see mm.h. */
294
295 /*
296 * For areas with an address space and backing store,
297 * linkage into the address_space->i_mmap interval tree.
298 */
299 struct {
300 struct rb_node rb;
301 unsigned long rb_subtree_last;
302 } shared;
303
304 /*
305 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
306 * list, after a COW of one of the file pages. A MAP_SHARED vma
307 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack
308 * or brk vma (with NULL file) can only be in an anon_vma list.
309 */
310 struct list_head anon_vma_chain; /* Serialized by mmap_sem &
311 * page_table_lock */
312 struct anon_vma *anon_vma; /* Serialized by page_table_lock */
313
314 /* Function pointers to deal with this struct. */
315 const struct vm_operations_struct *vm_ops;
316
317 /* Information about our backing store: */
318 unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE
319 units */
320 struct file * vm_file; /* File we map to (can be NULL). */
321 void * vm_private_data; /* was vm_pte (shared mem) */
322
323 atomic_long_t swap_readahead_info;
324 #ifndef CONFIG_MMU
325 struct vm_region *vm_region; /* NOMMU mapping region */
326 #endif
327 #ifdef CONFIG_NUMA
328 struct mempolicy *vm_policy; /* NUMA policy for the VMA */
329 #endif
330 struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
331 } __randomize_layout;
332
333 struct core_thread {
334 struct task_struct *task;
335 struct core_thread *next;
336 };
337
338 struct core_state {
339 atomic_t nr_threads;
340 struct core_thread dumper;
341 struct completion startup;
342 };
343
344 struct kioctx_table;
345 struct mm_struct {
346 struct {
347 struct vm_area_struct *mmap; /* list of VMAs */
348 struct rb_root mm_rb;
349 u64 vmacache_seqnum; /* per-thread vmacache */
350 #ifdef CONFIG_MMU
351 unsigned long (*get_unmapped_area) (struct file *filp,
352 unsigned long addr, unsigned long len,
353 unsigned long pgoff, unsigned long flags);
354 #endif
355 unsigned long mmap_base; /* base of mmap area */
356 unsigned long mmap_legacy_base; /* base of mmap area in bottom-up allocations */
357 #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
358 /* Base adresses for compatible mmap() */
359 unsigned long mmap_compat_base;
360 unsigned long mmap_compat_legacy_base;
361 #endif
362 unsigned long task_size; /* size of task vm space */
363 unsigned long highest_vm_end; /* highest vma end address */
364 pgd_t * pgd;
365
366 /**
367 * @mm_users: The number of users including userspace.
368 *
369 * Use mmget()/mmget_not_zero()/mmput() to modify. When this
370 * drops to 0 (i.e. when the task exits and there are no other
371 * temporary reference holders), we also release a reference on
372 * @mm_count (which may then free the &struct mm_struct if
373 * @mm_count also drops to 0).
374 */
375 atomic_t mm_users;
376
377 /**
378 * @mm_count: The number of references to &struct mm_struct
379 * (@mm_users count as 1).
380 *
381 * Use mmgrab()/mmdrop() to modify. When this drops to 0, the
382 * &struct mm_struct is freed.
383 */
384 atomic_t mm_count;
385
386 #ifdef CONFIG_MMU
387 atomic_long_t pgtables_bytes; /* PTE page table pages */
388 #endif
389 int map_count; /* number of VMAs */
390
391 spinlock_t page_table_lock; /* Protects page tables and some
392 * counters
393 */
394 struct rw_semaphore mmap_sem;
395
396 struct list_head mmlist; /* List of maybe swapped mm's. These
397 * are globally strung together off
398 * init_mm.mmlist, and are protected
399 * by mmlist_lock
400 */
401
402
403 unsigned long hiwater_rss; /* High-watermark of RSS usage */
404 unsigned long hiwater_vm; /* High-water virtual memory usage */
405
406 unsigned long total_vm; /* Total pages mapped */
407 unsigned long locked_vm; /* Pages that have PG_mlocked set */
408 unsigned long pinned_vm; /* Refcount permanently increased */
409 unsigned long data_vm; /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
410 unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
411 unsigned long stack_vm; /* VM_STACK */
412 unsigned long def_flags;
413
414 spinlock_t arg_lock; /* protect the below fields */
415 unsigned long start_code, end_code, start_data, end_data;
416 unsigned long start_brk, brk, start_stack;
417 unsigned long arg_start, arg_end, env_start, env_end;
418
419 unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
420
421 /*
422 * Special counters, in some configurations protected by the
423 * page_table_lock, in other configurations by being atomic.
424 */
425 struct mm_rss_stat rss_stat;
426
427 struct linux_binfmt *binfmt;
428
429 /* Architecture-specific MM context */
430 mm_context_t context;
431
432 unsigned long flags; /* Must use atomic bitops to access */
433
434 struct core_state *core_state; /* coredumping support */
435 #ifdef CONFIG_MEMBARRIER
436 atomic_t membarrier_state;
437 #endif
438 #ifdef CONFIG_AIO
439 spinlock_t ioctx_lock;
440 struct kioctx_table __rcu *ioctx_table;
441 #endif
442 #ifdef CONFIG_MEMCG
443 /*
444 * "owner" points to a task that is regarded as the canonical
445 * user/owner of this mm. All of the following must be true in
446 * order for it to be changed:
447 *
448 * current == mm->owner
449 * current->mm != mm
450 * new_owner->mm == mm
451 * new_owner->alloc_lock is held
452 */
453 struct task_struct __rcu *owner;
454 #endif
455 struct user_namespace *user_ns;
456
457 /* store ref to file /proc/<pid>/exe symlink points to */
458 struct file __rcu *exe_file;
459 #ifdef CONFIG_MMU_NOTIFIER
460 struct mmu_notifier_mm *mmu_notifier_mm;
461 #endif
462 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
463 pgtable_t pmd_huge_pte; /* protected by page_table_lock */
464 #endif
465 #ifdef CONFIG_NUMA_BALANCING
466 /*
467 * numa_next_scan is the next time that the PTEs will be marked
468 * pte_numa. NUMA hinting faults will gather statistics and
469 * migrate pages to new nodes if necessary.
470 */
471 unsigned long numa_next_scan;
472
473 /* Restart point for scanning and setting pte_numa */
474 unsigned long numa_scan_offset;
475
476 /* numa_scan_seq prevents two threads setting pte_numa */
477 int numa_scan_seq;
478 #endif
479 /*
480 * An operation with batched TLB flushing is going on. Anything
481 * that can move process memory needs to flush the TLB when
482 * moving a PROT_NONE or PROT_NUMA mapped page.
483 */
484 atomic_t tlb_flush_pending;
485 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
486 /* See flush_tlb_batched_pending() */
487 bool tlb_flush_batched;
488 #endif
489 struct uprobes_state uprobes_state;
490 #ifdef CONFIG_HUGETLB_PAGE
491 atomic_long_t hugetlb_usage;
492 #endif
493 struct work_struct async_put_work;
494
495 #if IS_ENABLED(CONFIG_HMM)
496 /* HMM needs to track a few things per mm */
497 struct hmm *hmm;
498 #endif
499 } __randomize_layout;
500
501 /*
502 * The mm_cpumask needs to be at the end of mm_struct, because it
503 * is dynamically sized based on nr_cpu_ids.
504 */
505 unsigned long cpu_bitmap[];
506 };
507
508 extern struct mm_struct init_mm;
509
510 /* Pointer magic because the dynamic array size confuses some compilers. */
mm_init_cpumask(struct mm_struct * mm)511 static inline void mm_init_cpumask(struct mm_struct *mm)
512 {
513 unsigned long cpu_bitmap = (unsigned long)mm;
514
515 cpu_bitmap += offsetof(struct mm_struct, cpu_bitmap);
516 cpumask_clear((struct cpumask *)cpu_bitmap);
517 }
518
519 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
mm_cpumask(struct mm_struct * mm)520 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
521 {
522 return (struct cpumask *)&mm->cpu_bitmap;
523 }
524
525 struct mmu_gather;
526 extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm,
527 unsigned long start, unsigned long end);
528 extern void tlb_finish_mmu(struct mmu_gather *tlb,
529 unsigned long start, unsigned long end);
530
init_tlb_flush_pending(struct mm_struct * mm)531 static inline void init_tlb_flush_pending(struct mm_struct *mm)
532 {
533 atomic_set(&mm->tlb_flush_pending, 0);
534 }
535
inc_tlb_flush_pending(struct mm_struct * mm)536 static inline void inc_tlb_flush_pending(struct mm_struct *mm)
537 {
538 atomic_inc(&mm->tlb_flush_pending);
539 /*
540 * The only time this value is relevant is when there are indeed pages
541 * to flush. And we'll only flush pages after changing them, which
542 * requires the PTL.
543 *
544 * So the ordering here is:
545 *
546 * atomic_inc(&mm->tlb_flush_pending);
547 * spin_lock(&ptl);
548 * ...
549 * set_pte_at();
550 * spin_unlock(&ptl);
551 *
552 * spin_lock(&ptl)
553 * mm_tlb_flush_pending();
554 * ....
555 * spin_unlock(&ptl);
556 *
557 * flush_tlb_range();
558 * atomic_dec(&mm->tlb_flush_pending);
559 *
560 * Where the increment if constrained by the PTL unlock, it thus
561 * ensures that the increment is visible if the PTE modification is
562 * visible. After all, if there is no PTE modification, nobody cares
563 * about TLB flushes either.
564 *
565 * This very much relies on users (mm_tlb_flush_pending() and
566 * mm_tlb_flush_nested()) only caring about _specific_ PTEs (and
567 * therefore specific PTLs), because with SPLIT_PTE_PTLOCKS and RCpc
568 * locks (PPC) the unlock of one doesn't order against the lock of
569 * another PTL.
570 *
571 * The decrement is ordered by the flush_tlb_range(), such that
572 * mm_tlb_flush_pending() will not return false unless all flushes have
573 * completed.
574 */
575 }
576
dec_tlb_flush_pending(struct mm_struct * mm)577 static inline void dec_tlb_flush_pending(struct mm_struct *mm)
578 {
579 /*
580 * See inc_tlb_flush_pending().
581 *
582 * This cannot be smp_mb__before_atomic() because smp_mb() simply does
583 * not order against TLB invalidate completion, which is what we need.
584 *
585 * Therefore we must rely on tlb_flush_*() to guarantee order.
586 */
587 atomic_dec(&mm->tlb_flush_pending);
588 }
589
mm_tlb_flush_pending(struct mm_struct * mm)590 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
591 {
592 /*
593 * Must be called after having acquired the PTL; orders against that
594 * PTLs release and therefore ensures that if we observe the modified
595 * PTE we must also observe the increment from inc_tlb_flush_pending().
596 *
597 * That is, it only guarantees to return true if there is a flush
598 * pending for _this_ PTL.
599 */
600 return atomic_read(&mm->tlb_flush_pending);
601 }
602
mm_tlb_flush_nested(struct mm_struct * mm)603 static inline bool mm_tlb_flush_nested(struct mm_struct *mm)
604 {
605 /*
606 * Similar to mm_tlb_flush_pending(), we must have acquired the PTL
607 * for which there is a TLB flush pending in order to guarantee
608 * we've seen both that PTE modification and the increment.
609 *
610 * (no requirement on actually still holding the PTL, that is irrelevant)
611 */
612 return atomic_read(&mm->tlb_flush_pending) > 1;
613 }
614
615 struct vm_fault;
616
617 struct vm_special_mapping {
618 const char *name; /* The name, e.g. "[vdso]". */
619
620 /*
621 * If .fault is not provided, this points to a
622 * NULL-terminated array of pages that back the special mapping.
623 *
624 * This must not be NULL unless .fault is provided.
625 */
626 struct page **pages;
627
628 /*
629 * If non-NULL, then this is called to resolve page faults
630 * on the special mapping. If used, .pages is not checked.
631 */
632 vm_fault_t (*fault)(const struct vm_special_mapping *sm,
633 struct vm_area_struct *vma,
634 struct vm_fault *vmf);
635
636 int (*mremap)(const struct vm_special_mapping *sm,
637 struct vm_area_struct *new_vma);
638 };
639
640 enum tlb_flush_reason {
641 TLB_FLUSH_ON_TASK_SWITCH,
642 TLB_REMOTE_SHOOTDOWN,
643 TLB_LOCAL_SHOOTDOWN,
644 TLB_LOCAL_MM_SHOOTDOWN,
645 TLB_REMOTE_SEND_IPI,
646 NR_TLB_FLUSH_REASONS,
647 };
648
649 /*
650 * A swap entry has to fit into a "unsigned long", as the entry is hidden
651 * in the "index" field of the swapper address space.
652 */
653 typedef struct {
654 unsigned long val;
655 } swp_entry_t;
656
657 #endif /* _LINUX_MM_TYPES_H */
658