]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - include/linux/mm_types.h
Merge branch 'work.aio' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
[mirror_ubuntu-jammy-kernel.git] / include / linux / mm_types.h
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 struct mm_struct *pt_mm; /* x86 pgds only */
143 #if ALLOC_SPLIT_PTLOCKS
144 spinlock_t *ptl;
145 #else
146 spinlock_t ptl;
147 #endif
148 };
149 struct { /* ZONE_DEVICE pages */
150 /** @pgmap: Points to the hosting device page map. */
151 struct dev_pagemap *pgmap;
152 unsigned long hmm_data;
153 unsigned long _zd_pad_1; /* uses mapping */
154 };
155
156 /** @rcu_head: You can use this to free a page by RCU. */
157 struct rcu_head rcu_head;
158 };
159
160 union { /* This union is 4 bytes in size. */
161 /*
162 * If the page can be mapped to userspace, encodes the number
163 * of times this page is referenced by a page table.
164 */
165 atomic_t _mapcount;
166
167 /*
168 * If the page is neither PageSlab nor mappable to userspace,
169 * the value stored here may help determine what this page
170 * is used for. See page-flags.h for a list of page types
171 * which are currently stored here.
172 */
173 unsigned int page_type;
174
175 unsigned int active; /* SLAB */
176 int units; /* SLOB */
177 };
178
179 /* Usage count. *DO NOT USE DIRECTLY*. See page_ref.h */
180 atomic_t _refcount;
181
182 #ifdef CONFIG_MEMCG
183 struct mem_cgroup *mem_cgroup;
184 #endif
185
186 /*
187 * On machines where all RAM is mapped into kernel address space,
188 * we can simply calculate the virtual address. On machines with
189 * highmem some memory is mapped into kernel virtual memory
190 * dynamically, so we need a place to store that address.
191 * Note that this field could be 16 bits on x86 ... ;)
192 *
193 * Architectures with slow multiplication can define
194 * WANT_PAGE_VIRTUAL in asm/page.h
195 */
196 #if defined(WANT_PAGE_VIRTUAL)
197 void *virtual; /* Kernel virtual address (NULL if
198 not kmapped, ie. highmem) */
199 #endif /* WANT_PAGE_VIRTUAL */
200
201 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
202 int _last_cpupid;
203 #endif
204 } _struct_page_alignment;
205
206 #define PAGE_FRAG_CACHE_MAX_SIZE __ALIGN_MASK(32768, ~PAGE_MASK)
207 #define PAGE_FRAG_CACHE_MAX_ORDER get_order(PAGE_FRAG_CACHE_MAX_SIZE)
208
209 struct page_frag_cache {
210 void * va;
211 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
212 __u16 offset;
213 __u16 size;
214 #else
215 __u32 offset;
216 #endif
217 /* we maintain a pagecount bias, so that we dont dirty cache line
218 * containing page->_refcount every time we allocate a fragment.
219 */
220 unsigned int pagecnt_bias;
221 bool pfmemalloc;
222 };
223
224 typedef unsigned long vm_flags_t;
225
226 /*
227 * A region containing a mapping of a non-memory backed file under NOMMU
228 * conditions. These are held in a global tree and are pinned by the VMAs that
229 * map parts of them.
230 */
231 struct vm_region {
232 struct rb_node vm_rb; /* link in global region tree */
233 vm_flags_t vm_flags; /* VMA vm_flags */
234 unsigned long vm_start; /* start address of region */
235 unsigned long vm_end; /* region initialised to here */
236 unsigned long vm_top; /* region allocated to here */
237 unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */
238 struct file *vm_file; /* the backing file or NULL */
239
240 int vm_usage; /* region usage count (access under nommu_region_sem) */
241 bool vm_icache_flushed : 1; /* true if the icache has been flushed for
242 * this region */
243 };
244
245 #ifdef CONFIG_USERFAULTFD
246 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
247 struct vm_userfaultfd_ctx {
248 struct userfaultfd_ctx *ctx;
249 };
250 #else /* CONFIG_USERFAULTFD */
251 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
252 struct vm_userfaultfd_ctx {};
253 #endif /* CONFIG_USERFAULTFD */
254
255 /*
256 * This struct defines a memory VMM memory area. There is one of these
257 * per VM-area/task. A VM area is any part of the process virtual memory
258 * space that has a special rule for the page-fault handlers (ie a shared
259 * library, the executable area etc).
260 */
261 struct vm_area_struct {
262 /* The first cache line has the info for VMA tree walking. */
263
264 unsigned long vm_start; /* Our start address within vm_mm. */
265 unsigned long vm_end; /* The first byte after our end address
266 within vm_mm. */
267
268 /* linked list of VM areas per task, sorted by address */
269 struct vm_area_struct *vm_next, *vm_prev;
270
271 struct rb_node vm_rb;
272
273 /*
274 * Largest free memory gap in bytes to the left of this VMA.
275 * Either between this VMA and vma->vm_prev, or between one of the
276 * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
277 * get_unmapped_area find a free area of the right size.
278 */
279 unsigned long rb_subtree_gap;
280
281 /* Second cache line starts here. */
282
283 struct mm_struct *vm_mm; /* The address space we belong to. */
284 pgprot_t vm_page_prot; /* Access permissions of this VMA. */
285 unsigned long vm_flags; /* Flags, see mm.h. */
286
287 /*
288 * For areas with an address space and backing store,
289 * linkage into the address_space->i_mmap interval tree.
290 */
291 struct {
292 struct rb_node rb;
293 unsigned long rb_subtree_last;
294 } shared;
295
296 /*
297 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
298 * list, after a COW of one of the file pages. A MAP_SHARED vma
299 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack
300 * or brk vma (with NULL file) can only be in an anon_vma list.
301 */
302 struct list_head anon_vma_chain; /* Serialized by mmap_sem &
303 * page_table_lock */
304 struct anon_vma *anon_vma; /* Serialized by page_table_lock */
305
306 /* Function pointers to deal with this struct. */
307 const struct vm_operations_struct *vm_ops;
308
309 /* Information about our backing store: */
310 unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE
311 units */
312 struct file * vm_file; /* File we map to (can be NULL). */
313 void * vm_private_data; /* was vm_pte (shared mem) */
314
315 atomic_long_t swap_readahead_info;
316 #ifndef CONFIG_MMU
317 struct vm_region *vm_region; /* NOMMU mapping region */
318 #endif
319 #ifdef CONFIG_NUMA
320 struct mempolicy *vm_policy; /* NUMA policy for the VMA */
321 #endif
322 struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
323 } __randomize_layout;
324
325 struct core_thread {
326 struct task_struct *task;
327 struct core_thread *next;
328 };
329
330 struct core_state {
331 atomic_t nr_threads;
332 struct core_thread dumper;
333 struct completion startup;
334 };
335
336 struct kioctx_table;
337 struct mm_struct {
338 struct vm_area_struct *mmap; /* list of VMAs */
339 struct rb_root mm_rb;
340 u32 vmacache_seqnum; /* per-thread vmacache */
341 #ifdef CONFIG_MMU
342 unsigned long (*get_unmapped_area) (struct file *filp,
343 unsigned long addr, unsigned long len,
344 unsigned long pgoff, unsigned long flags);
345 #endif
346 unsigned long mmap_base; /* base of mmap area */
347 unsigned long mmap_legacy_base; /* base of mmap area in bottom-up allocations */
348 #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
349 /* Base adresses for compatible mmap() */
350 unsigned long mmap_compat_base;
351 unsigned long mmap_compat_legacy_base;
352 #endif
353 unsigned long task_size; /* size of task vm space */
354 unsigned long highest_vm_end; /* highest vma end address */
355 pgd_t * pgd;
356
357 /**
358 * @mm_users: The number of users including userspace.
359 *
360 * Use mmget()/mmget_not_zero()/mmput() to modify. When this drops
361 * to 0 (i.e. when the task exits and there are no other temporary
362 * reference holders), we also release a reference on @mm_count
363 * (which may then free the &struct mm_struct if @mm_count also
364 * drops to 0).
365 */
366 atomic_t mm_users;
367
368 /**
369 * @mm_count: The number of references to &struct mm_struct
370 * (@mm_users count as 1).
371 *
372 * Use mmgrab()/mmdrop() to modify. When this drops to 0, the
373 * &struct mm_struct is freed.
374 */
375 atomic_t mm_count;
376
377 #ifdef CONFIG_MMU
378 atomic_long_t pgtables_bytes; /* PTE page table pages */
379 #endif
380 int map_count; /* number of VMAs */
381
382 spinlock_t page_table_lock; /* Protects page tables and some counters */
383 struct rw_semaphore mmap_sem;
384
385 struct list_head mmlist; /* List of maybe swapped mm's. These are globally strung
386 * together off init_mm.mmlist, and are protected
387 * by mmlist_lock
388 */
389
390
391 unsigned long hiwater_rss; /* High-watermark of RSS usage */
392 unsigned long hiwater_vm; /* High-water virtual memory usage */
393
394 unsigned long total_vm; /* Total pages mapped */
395 unsigned long locked_vm; /* Pages that have PG_mlocked set */
396 unsigned long pinned_vm; /* Refcount permanently increased */
397 unsigned long data_vm; /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
398 unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
399 unsigned long stack_vm; /* VM_STACK */
400 unsigned long def_flags;
401
402 spinlock_t arg_lock; /* protect the below fields */
403 unsigned long start_code, end_code, start_data, end_data;
404 unsigned long start_brk, brk, start_stack;
405 unsigned long arg_start, arg_end, env_start, env_end;
406
407 unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
408
409 /*
410 * Special counters, in some configurations protected by the
411 * page_table_lock, in other configurations by being atomic.
412 */
413 struct mm_rss_stat rss_stat;
414
415 struct linux_binfmt *binfmt;
416
417 cpumask_var_t cpu_vm_mask_var;
418
419 /* Architecture-specific MM context */
420 mm_context_t context;
421
422 unsigned long flags; /* Must use atomic bitops to access the bits */
423
424 struct core_state *core_state; /* coredumping support */
425 #ifdef CONFIG_MEMBARRIER
426 atomic_t membarrier_state;
427 #endif
428 #ifdef CONFIG_AIO
429 spinlock_t ioctx_lock;
430 struct kioctx_table __rcu *ioctx_table;
431 #endif
432 #ifdef CONFIG_MEMCG
433 /*
434 * "owner" points to a task that is regarded as the canonical
435 * user/owner of this mm. All of the following must be true in
436 * order for it to be changed:
437 *
438 * current == mm->owner
439 * current->mm != mm
440 * new_owner->mm == mm
441 * new_owner->alloc_lock is held
442 */
443 struct task_struct __rcu *owner;
444 #endif
445 struct user_namespace *user_ns;
446
447 /* store ref to file /proc/<pid>/exe symlink points to */
448 struct file __rcu *exe_file;
449 #ifdef CONFIG_MMU_NOTIFIER
450 struct mmu_notifier_mm *mmu_notifier_mm;
451 #endif
452 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
453 pgtable_t pmd_huge_pte; /* protected by page_table_lock */
454 #endif
455 #ifdef CONFIG_CPUMASK_OFFSTACK
456 struct cpumask cpumask_allocation;
457 #endif
458 #ifdef CONFIG_NUMA_BALANCING
459 /*
460 * numa_next_scan is the next time that the PTEs will be marked
461 * pte_numa. NUMA hinting faults will gather statistics and migrate
462 * pages to new nodes if necessary.
463 */
464 unsigned long numa_next_scan;
465
466 /* Restart point for scanning and setting pte_numa */
467 unsigned long numa_scan_offset;
468
469 /* numa_scan_seq prevents two threads setting pte_numa */
470 int numa_scan_seq;
471 #endif
472 /*
473 * An operation with batched TLB flushing is going on. Anything that
474 * can move process memory needs to flush the TLB when moving a
475 * PROT_NONE or PROT_NUMA mapped page.
476 */
477 atomic_t tlb_flush_pending;
478 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
479 /* See flush_tlb_batched_pending() */
480 bool tlb_flush_batched;
481 #endif
482 struct uprobes_state uprobes_state;
483 #ifdef CONFIG_HUGETLB_PAGE
484 atomic_long_t hugetlb_usage;
485 #endif
486 struct work_struct async_put_work;
487
488 #if IS_ENABLED(CONFIG_HMM)
489 /* HMM needs to track a few things per mm */
490 struct hmm *hmm;
491 #endif
492 } __randomize_layout;
493
494 extern struct mm_struct init_mm;
495
496 static inline void mm_init_cpumask(struct mm_struct *mm)
497 {
498 #ifdef CONFIG_CPUMASK_OFFSTACK
499 mm->cpu_vm_mask_var = &mm->cpumask_allocation;
500 #endif
501 cpumask_clear(mm->cpu_vm_mask_var);
502 }
503
504 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
505 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
506 {
507 return mm->cpu_vm_mask_var;
508 }
509
510 struct mmu_gather;
511 extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm,
512 unsigned long start, unsigned long end);
513 extern void tlb_finish_mmu(struct mmu_gather *tlb,
514 unsigned long start, unsigned long end);
515
516 static inline void init_tlb_flush_pending(struct mm_struct *mm)
517 {
518 atomic_set(&mm->tlb_flush_pending, 0);
519 }
520
521 static inline void inc_tlb_flush_pending(struct mm_struct *mm)
522 {
523 atomic_inc(&mm->tlb_flush_pending);
524 /*
525 * The only time this value is relevant is when there are indeed pages
526 * to flush. And we'll only flush pages after changing them, which
527 * requires the PTL.
528 *
529 * So the ordering here is:
530 *
531 * atomic_inc(&mm->tlb_flush_pending);
532 * spin_lock(&ptl);
533 * ...
534 * set_pte_at();
535 * spin_unlock(&ptl);
536 *
537 * spin_lock(&ptl)
538 * mm_tlb_flush_pending();
539 * ....
540 * spin_unlock(&ptl);
541 *
542 * flush_tlb_range();
543 * atomic_dec(&mm->tlb_flush_pending);
544 *
545 * Where the increment if constrained by the PTL unlock, it thus
546 * ensures that the increment is visible if the PTE modification is
547 * visible. After all, if there is no PTE modification, nobody cares
548 * about TLB flushes either.
549 *
550 * This very much relies on users (mm_tlb_flush_pending() and
551 * mm_tlb_flush_nested()) only caring about _specific_ PTEs (and
552 * therefore specific PTLs), because with SPLIT_PTE_PTLOCKS and RCpc
553 * locks (PPC) the unlock of one doesn't order against the lock of
554 * another PTL.
555 *
556 * The decrement is ordered by the flush_tlb_range(), such that
557 * mm_tlb_flush_pending() will not return false unless all flushes have
558 * completed.
559 */
560 }
561
562 static inline void dec_tlb_flush_pending(struct mm_struct *mm)
563 {
564 /*
565 * See inc_tlb_flush_pending().
566 *
567 * This cannot be smp_mb__before_atomic() because smp_mb() simply does
568 * not order against TLB invalidate completion, which is what we need.
569 *
570 * Therefore we must rely on tlb_flush_*() to guarantee order.
571 */
572 atomic_dec(&mm->tlb_flush_pending);
573 }
574
575 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
576 {
577 /*
578 * Must be called after having acquired the PTL; orders against that
579 * PTLs release and therefore ensures that if we observe the modified
580 * PTE we must also observe the increment from inc_tlb_flush_pending().
581 *
582 * That is, it only guarantees to return true if there is a flush
583 * pending for _this_ PTL.
584 */
585 return atomic_read(&mm->tlb_flush_pending);
586 }
587
588 static inline bool mm_tlb_flush_nested(struct mm_struct *mm)
589 {
590 /*
591 * Similar to mm_tlb_flush_pending(), we must have acquired the PTL
592 * for which there is a TLB flush pending in order to guarantee
593 * we've seen both that PTE modification and the increment.
594 *
595 * (no requirement on actually still holding the PTL, that is irrelevant)
596 */
597 return atomic_read(&mm->tlb_flush_pending) > 1;
598 }
599
600 struct vm_fault;
601
602 struct vm_special_mapping {
603 const char *name; /* The name, e.g. "[vdso]". */
604
605 /*
606 * If .fault is not provided, this points to a
607 * NULL-terminated array of pages that back the special mapping.
608 *
609 * This must not be NULL unless .fault is provided.
610 */
611 struct page **pages;
612
613 /*
614 * If non-NULL, then this is called to resolve page faults
615 * on the special mapping. If used, .pages is not checked.
616 */
617 vm_fault_t (*fault)(const struct vm_special_mapping *sm,
618 struct vm_area_struct *vma,
619 struct vm_fault *vmf);
620
621 int (*mremap)(const struct vm_special_mapping *sm,
622 struct vm_area_struct *new_vma);
623 };
624
625 enum tlb_flush_reason {
626 TLB_FLUSH_ON_TASK_SWITCH,
627 TLB_REMOTE_SHOOTDOWN,
628 TLB_LOCAL_SHOOTDOWN,
629 TLB_LOCAL_MM_SHOOTDOWN,
630 TLB_REMOTE_SEND_IPI,
631 NR_TLB_FLUSH_REASONS,
632 };
633
634 /*
635 * A swap entry has to fit into a "unsigned long", as the entry is hidden
636 * in the "index" field of the swapper address space.
637 */
638 typedef struct {
639 unsigned long val;
640 } swp_entry_t;
641
642 #endif /* _LINUX_MM_TYPES_H */