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
25typedef int vm_fault_t;
26
27struct address_space;
28struct mem_cgroup;
29struct 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
70struct 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
212struct 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
227typedef unsigned long vm_flags_t;
228
229/*
230 * A region containing a mapping of a non-memory backed file under NOMMU
231 * conditions. These are held in a global tree and are pinned by the VMAs that
232 * map parts of them.
233 */
234struct vm_region {
235 struct rb_node vm_rb; /* link in global region tree */
236 vm_flags_t vm_flags; /* VMA vm_flags */
237 unsigned long vm_start; /* start address of region */
238 unsigned long vm_end; /* region initialised to here */
239 unsigned long vm_top; /* region allocated to here */
240 unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */
241 struct file *vm_file; /* the backing file or NULL */
242
243 int vm_usage; /* region usage count (access under nommu_region_sem) */
244 bool vm_icache_flushed : 1; /* true if the icache has been flushed for
245 * this region */
246};
247
248#ifdef CONFIG_USERFAULTFD
249#define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
250struct vm_userfaultfd_ctx {
251 struct userfaultfd_ctx *ctx;
252};
253#else /* CONFIG_USERFAULTFD */
254#define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
255struct vm_userfaultfd_ctx {};
256#endif /* CONFIG_USERFAULTFD */
257
258/*
259 * This struct defines a memory VMM memory area. There is one of these
260 * per VM-area/task. A VM area is any part of the process virtual memory
261 * space that has a special rule for the page-fault handlers (ie a shared
262 * library, the executable area etc).
263 */
264struct vm_area_struct {
265 /* The first cache line has the info for VMA tree walking. */
266
267 unsigned long vm_start; /* Our start address within vm_mm. */
268 unsigned long vm_end; /* The first byte after our end address
269 within vm_mm. */
270
271 /* linked list of VM areas per task, sorted by address */
272 struct vm_area_struct *vm_next, *vm_prev;
273
274 struct rb_node vm_rb;
275
276 /*
277 * Largest free memory gap in bytes to the left of this VMA.
278 * Either between this VMA and vma->vm_prev, or between one of the
279 * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
280 * get_unmapped_area find a free area of the right size.
281 */
282 unsigned long rb_subtree_gap;
283
284 /* Second cache line starts here. */
285
286 struct mm_struct *vm_mm; /* The address space we belong to. */
287 pgprot_t vm_page_prot; /* Access permissions of this VMA. */
288 unsigned long vm_flags; /* Flags, see mm.h. */
289
290 /*
291 * For areas with an address space and backing store,
292 * linkage into the address_space->i_mmap interval tree.
293 */
294 struct {
295 struct rb_node rb;
296 unsigned long rb_subtree_last;
297 } shared;
298
299 /*
300 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
301 * list, after a COW of one of the file pages. A MAP_SHARED vma
302 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack
303 * or brk vma (with NULL file) can only be in an anon_vma list.
304 */
305 struct list_head anon_vma_chain; /* Serialized by mmap_sem &
306 * page_table_lock */
307 struct anon_vma *anon_vma; /* Serialized by page_table_lock */
308
309 /* Function pointers to deal with this struct. */
310 const struct vm_operations_struct *vm_ops;
311
312 /* Information about our backing store: */
313 unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE
314 units */
315 struct file * vm_file; /* File we map to (can be NULL). */
316 void * vm_private_data; /* was vm_pte (shared mem) */
317
318 atomic_long_t swap_readahead_info;
319#ifndef CONFIG_MMU
320 struct vm_region *vm_region; /* NOMMU mapping region */
321#endif
322#ifdef CONFIG_NUMA
323 struct mempolicy *vm_policy; /* NUMA policy for the VMA */
324#endif
325 struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
326} __randomize_layout;
327
328struct core_thread {
329 struct task_struct *task;
330 struct core_thread *next;
331};
332
333struct core_state {
334 atomic_t nr_threads;
335 struct core_thread dumper;
336 struct completion startup;
337};
338
339struct kioctx_table;
340struct mm_struct {
341 struct {
342 struct vm_area_struct *mmap; /* list of VMAs */
343 struct rb_root mm_rb;
344 u32 vmacache_seqnum; /* per-thread vmacache */
345#ifdef CONFIG_MMU
346 unsigned long (*get_unmapped_area) (struct file *filp,
347 unsigned long addr, unsigned long len,
348 unsigned long pgoff, unsigned long flags);
349#endif
350 unsigned long mmap_base; /* base of mmap area */
351 unsigned long mmap_legacy_base; /* base of mmap area in bottom-up allocations */
352#ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
353 /* Base adresses for compatible mmap() */
354 unsigned long mmap_compat_base;
355 unsigned long mmap_compat_legacy_base;
356#endif
357 unsigned long task_size; /* size of task vm space */
358 unsigned long highest_vm_end; /* highest vma end address */
359 pgd_t * pgd;
360
361 /**
362 * @mm_users: The number of users including userspace.
363 *
364 * Use mmget()/mmget_not_zero()/mmput() to modify. When this
365 * drops to 0 (i.e. when the task exits and there are no other
366 * temporary reference holders), we also release a reference on
367 * @mm_count (which may then free the &struct mm_struct if
368 * @mm_count also drops to 0).
369 */
370 atomic_t mm_users;
371
372 /**
373 * @mm_count: The number of references to &struct mm_struct
374 * (@mm_users count as 1).
375 *
376 * Use mmgrab()/mmdrop() to modify. When this drops to 0, the
377 * &struct mm_struct is freed.
378 */
379 atomic_t mm_count;
380
381#ifdef CONFIG_MMU
382 atomic_long_t pgtables_bytes; /* PTE page table pages */
383#endif
384 int map_count; /* number of VMAs */
385
386 spinlock_t page_table_lock; /* Protects page tables and some
387 * counters
388 */
389 struct rw_semaphore mmap_sem;
390
391 struct list_head mmlist; /* List of maybe swapped mm's. These
392 * are globally strung together off
393 * init_mm.mmlist, and are protected
394 * by mmlist_lock
395 */
396
397
398 unsigned long hiwater_rss; /* High-watermark of RSS usage */
399 unsigned long hiwater_vm; /* High-water virtual memory usage */
400
401 unsigned long total_vm; /* Total pages mapped */
402 unsigned long locked_vm; /* Pages that have PG_mlocked set */
403 unsigned long pinned_vm; /* Refcount permanently increased */
404 unsigned long data_vm; /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
405 unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
406 unsigned long stack_vm; /* VM_STACK */
407 unsigned long def_flags;
408
409 spinlock_t arg_lock; /* protect the below fields */
410 unsigned long start_code, end_code, start_data, end_data;
411 unsigned long start_brk, brk, start_stack;
412 unsigned long arg_start, arg_end, env_start, env_end;
413
414 unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
415
416 /*
417 * Special counters, in some configurations protected by the
418 * page_table_lock, in other configurations by being atomic.
419 */
420 struct mm_rss_stat rss_stat;
421
422 struct linux_binfmt *binfmt;
423
424 /* Architecture-specific MM context */
425 mm_context_t context;
426
427 unsigned long flags; /* Must use atomic bitops to access */
428
429 struct core_state *core_state; /* coredumping support */
430#ifdef CONFIG_MEMBARRIER
431 atomic_t membarrier_state;
432#endif
433#ifdef CONFIG_AIO
434 spinlock_t ioctx_lock;
435 struct kioctx_table __rcu *ioctx_table;
436#endif
437#ifdef CONFIG_MEMCG
438 /*
439 * "owner" points to a task that is regarded as the canonical
440 * user/owner of this mm. All of the following must be true in
441 * order for it to be changed:
442 *
443 * current == mm->owner
444 * current->mm != mm
445 * new_owner->mm == mm
446 * new_owner->alloc_lock is held
447 */
448 struct task_struct __rcu *owner;
449#endif
450 struct user_namespace *user_ns;
451
452 /* store ref to file /proc/<pid>/exe symlink points to */
453 struct file __rcu *exe_file;
454#ifdef CONFIG_MMU_NOTIFIER
455 struct mmu_notifier_mm *mmu_notifier_mm;
456#endif
457#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
458 pgtable_t pmd_huge_pte; /* protected by page_table_lock */
459#endif
460#ifdef CONFIG_NUMA_BALANCING
461 /*
462 * numa_next_scan is the next time that the PTEs will be marked
463 * pte_numa. NUMA hinting faults will gather statistics and
464 * migrate pages to new nodes if necessary.
465 */
466 unsigned long numa_next_scan;
467
468 /* Restart point for scanning and setting pte_numa */
469 unsigned long numa_scan_offset;
470
471 /* numa_scan_seq prevents two threads setting pte_numa */
472 int numa_scan_seq;
473#endif
474 /*
475 * An operation with batched TLB flushing is going on. Anything
476 * that can move process memory needs to flush the TLB when
477 * moving a PROT_NONE or PROT_NUMA mapped page.
478 */
479 atomic_t tlb_flush_pending;
480#ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
481 /* See flush_tlb_batched_pending() */
482 bool tlb_flush_batched;
483#endif
484 struct uprobes_state uprobes_state;
485#ifdef CONFIG_HUGETLB_PAGE
486 atomic_long_t hugetlb_usage;
487#endif
488 struct work_struct async_put_work;
489
490#if IS_ENABLED(CONFIG_HMM)
491 /* HMM needs to track a few things per mm */
492 struct hmm *hmm;
493#endif
494 } __randomize_layout;
495
496 /*
497 * The mm_cpumask needs to be at the end of mm_struct, because it
498 * is dynamically sized based on nr_cpu_ids.
499 */
500 unsigned long cpu_bitmap[];
501};
502
503extern struct mm_struct init_mm;
504
505/* Pointer magic because the dynamic array size confuses some compilers. */
506static inline void mm_init_cpumask(struct mm_struct *mm)
507{
508 unsigned long cpu_bitmap = (unsigned long)mm;
509
510 cpu_bitmap += offsetof(struct mm_struct, cpu_bitmap);
511 cpumask_clear((struct cpumask *)cpu_bitmap);
512}
513
514/* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
515static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
516{
517 return (struct cpumask *)&mm->cpu_bitmap;
518}
519
520struct mmu_gather;
521extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm,
522 unsigned long start, unsigned long end);
523extern void tlb_finish_mmu(struct mmu_gather *tlb,
524 unsigned long start, unsigned long end);
525
526static inline void init_tlb_flush_pending(struct mm_struct *mm)
527{
528 atomic_set(&mm->tlb_flush_pending, 0);
529}
530
531static inline void inc_tlb_flush_pending(struct mm_struct *mm)
532{
533 atomic_inc(&mm->tlb_flush_pending);
534 /*
535 * The only time this value is relevant is when there are indeed pages
536 * to flush. And we'll only flush pages after changing them, which
537 * requires the PTL.
538 *
539 * So the ordering here is:
540 *
541 * atomic_inc(&mm->tlb_flush_pending);
542 * spin_lock(&ptl);
543 * ...
544 * set_pte_at();
545 * spin_unlock(&ptl);
546 *
547 * spin_lock(&ptl)
548 * mm_tlb_flush_pending();
549 * ....
550 * spin_unlock(&ptl);
551 *
552 * flush_tlb_range();
553 * atomic_dec(&mm->tlb_flush_pending);
554 *
555 * Where the increment if constrained by the PTL unlock, it thus
556 * ensures that the increment is visible if the PTE modification is
557 * visible. After all, if there is no PTE modification, nobody cares
558 * about TLB flushes either.
559 *
560 * This very much relies on users (mm_tlb_flush_pending() and
561 * mm_tlb_flush_nested()) only caring about _specific_ PTEs (and
562 * therefore specific PTLs), because with SPLIT_PTE_PTLOCKS and RCpc
563 * locks (PPC) the unlock of one doesn't order against the lock of
564 * another PTL.
565 *
566 * The decrement is ordered by the flush_tlb_range(), such that
567 * mm_tlb_flush_pending() will not return false unless all flushes have
568 * completed.
569 */
570}
571
572static inline void dec_tlb_flush_pending(struct mm_struct *mm)
573{
574 /*
575 * See inc_tlb_flush_pending().
576 *
577 * This cannot be smp_mb__before_atomic() because smp_mb() simply does
578 * not order against TLB invalidate completion, which is what we need.
579 *
580 * Therefore we must rely on tlb_flush_*() to guarantee order.
581 */
582 atomic_dec(&mm->tlb_flush_pending);
583}
584
585static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
586{
587 /*
588 * Must be called after having acquired the PTL; orders against that
589 * PTLs release and therefore ensures that if we observe the modified
590 * PTE we must also observe the increment from inc_tlb_flush_pending().
591 *
592 * That is, it only guarantees to return true if there is a flush
593 * pending for _this_ PTL.
594 */
595 return atomic_read(&mm->tlb_flush_pending);
596}
597
598static inline bool mm_tlb_flush_nested(struct mm_struct *mm)
599{
600 /*
601 * Similar to mm_tlb_flush_pending(), we must have acquired the PTL
602 * for which there is a TLB flush pending in order to guarantee
603 * we've seen both that PTE modification and the increment.
604 *
605 * (no requirement on actually still holding the PTL, that is irrelevant)
606 */
607 return atomic_read(&mm->tlb_flush_pending) > 1;
608}
609
610struct vm_fault;
611
612struct vm_special_mapping {
613 const char *name; /* The name, e.g. "[vdso]". */
614
615 /*
616 * If .fault is not provided, this points to a
617 * NULL-terminated array of pages that back the special mapping.
618 *
619 * This must not be NULL unless .fault is provided.
620 */
621 struct page **pages;
622
623 /*
624 * If non-NULL, then this is called to resolve page faults
625 * on the special mapping. If used, .pages is not checked.
626 */
627 vm_fault_t (*fault)(const struct vm_special_mapping *sm,
628 struct vm_area_struct *vma,
629 struct vm_fault *vmf);
630
631 int (*mremap)(const struct vm_special_mapping *sm,
632 struct vm_area_struct *new_vma);
633};
634
635enum tlb_flush_reason {
636 TLB_FLUSH_ON_TASK_SWITCH,
637 TLB_REMOTE_SHOOTDOWN,
638 TLB_LOCAL_SHOOTDOWN,
639 TLB_LOCAL_MM_SHOOTDOWN,
640 TLB_REMOTE_SEND_IPI,
641 NR_TLB_FLUSH_REASONS,
642};
643
644 /*
645 * A swap entry has to fit into a "unsigned long", as the entry is hidden
646 * in the "index" field of the swapper address space.
647 */
648typedef struct {
649 unsigned long val;
650} swp_entry_t;
651
652#endif /* _LINUX_MM_TYPES_H */
653