projects/linux·blob v6.1 · refs

495 lines · 15.6 KiB

1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef _LINUX_RMAP_H
3#define _LINUX_RMAP_H
4/*
5 * Declarations for Reverse Mapping functions in mm/rmap.c
6 */
7
8#include <linux/list.h>
9#include <linux/slab.h>
10#include <linux/mm.h>
11#include <linux/rwsem.h>
12#include <linux/memcontrol.h>
13#include <linux/highmem.h>
14#include <linux/pagemap.h>
15#include <linux/memremap.h>
16
17/*
18 * The anon_vma heads a list of private "related" vmas, to scan if
19 * an anonymous page pointing to this anon_vma needs to be unmapped:
20 * the vmas on the list will be related by forking, or by splitting.
21 *
22 * Since vmas come and go as they are split and merged (particularly
23 * in mprotect), the mapping field of an anonymous page cannot point
24 * directly to a vma: instead it points to an anon_vma, on whose list
25 * the related vmas can be easily linked or unlinked.
26 *
27 * After unlinking the last vma on the list, we must garbage collect
28 * the anon_vma object itself: we're guaranteed no page can be
29 * pointing to this anon_vma once its vma list is empty.
30 */
31struct anon_vma {
32 struct anon_vma *root; /* Root of this anon_vma tree */
33 struct rw_semaphore rwsem; /* W: modification, R: walking the list */
34 /*
35 * The refcount is taken on an anon_vma when there is no
36 * guarantee that the vma of page tables will exist for
37 * the duration of the operation. A caller that takes
38 * the reference is responsible for clearing up the
39 * anon_vma if they are the last user on release
40 */
41 atomic_t refcount;
42
43 /*
44 * Count of child anon_vmas. Equals to the count of all anon_vmas that
45 * have ->parent pointing to this one, including itself.
46 *
47 * This counter is used for making decision about reusing anon_vma
48 * instead of forking new one. See comments in function anon_vma_clone.
49 */
50 unsigned long num_children;
51 /* Count of VMAs whose ->anon_vma pointer points to this object. */
52 unsigned long num_active_vmas;
53
54 struct anon_vma *parent; /* Parent of this anon_vma */
55
56 /*
57 * NOTE: the LSB of the rb_root.rb_node is set by
58 * mm_take_all_locks() _after_ taking the above lock. So the
59 * rb_root must only be read/written after taking the above lock
60 * to be sure to see a valid next pointer. The LSB bit itself
61 * is serialized by a system wide lock only visible to
62 * mm_take_all_locks() (mm_all_locks_mutex).
63 */
64
65 /* Interval tree of private "related" vmas */
66 struct rb_root_cached rb_root;
67};
68
69/*
70 * The copy-on-write semantics of fork mean that an anon_vma
71 * can become associated with multiple processes. Furthermore,
72 * each child process will have its own anon_vma, where new
73 * pages for that process are instantiated.
74 *
75 * This structure allows us to find the anon_vmas associated
76 * with a VMA, or the VMAs associated with an anon_vma.
77 * The "same_vma" list contains the anon_vma_chains linking
78 * all the anon_vmas associated with this VMA.
79 * The "rb" field indexes on an interval tree the anon_vma_chains
80 * which link all the VMAs associated with this anon_vma.
81 */
82struct anon_vma_chain {
83 struct vm_area_struct *vma;
84 struct anon_vma *anon_vma;
85 struct list_head same_vma; /* locked by mmap_lock & page_table_lock */
86 struct rb_node rb; /* locked by anon_vma->rwsem */
87 unsigned long rb_subtree_last;
88#ifdef CONFIG_DEBUG_VM_RB
89 unsigned long cached_vma_start, cached_vma_last;
90#endif
91};
92
93enum ttu_flags {
94 TTU_SPLIT_HUGE_PMD = 0x4, /* split huge PMD if any */
95 TTU_IGNORE_MLOCK = 0x8, /* ignore mlock */
96 TTU_SYNC = 0x10, /* avoid racy checks with PVMW_SYNC */
97 TTU_IGNORE_HWPOISON = 0x20, /* corrupted page is recoverable */
98 TTU_BATCH_FLUSH = 0x40, /* Batch TLB flushes where possible
99 * and caller guarantees they will
100 * do a final flush if necessary */
101 TTU_RMAP_LOCKED = 0x80, /* do not grab rmap lock:
102 * caller holds it */
103};
104
105#ifdef CONFIG_MMU
106static inline void get_anon_vma(struct anon_vma *anon_vma)
107{
108 atomic_inc(&anon_vma->refcount);
109}
110
111void __put_anon_vma(struct anon_vma *anon_vma);
112
113static inline void put_anon_vma(struct anon_vma *anon_vma)
114{
115 if (atomic_dec_and_test(&anon_vma->refcount))
116 __put_anon_vma(anon_vma);
117}
118
119static inline void anon_vma_lock_write(struct anon_vma *anon_vma)
120{
121 down_write(&anon_vma->root->rwsem);
122}
123
124static inline void anon_vma_unlock_write(struct anon_vma *anon_vma)
125{
126 up_write(&anon_vma->root->rwsem);
127}
128
129static inline void anon_vma_lock_read(struct anon_vma *anon_vma)
130{
131 down_read(&anon_vma->root->rwsem);
132}
133
134static inline int anon_vma_trylock_read(struct anon_vma *anon_vma)
135{
136 return down_read_trylock(&anon_vma->root->rwsem);
137}
138
139static inline void anon_vma_unlock_read(struct anon_vma *anon_vma)
140{
141 up_read(&anon_vma->root->rwsem);
142}
143
144
145/*
146 * anon_vma helper functions.
147 */
148void anon_vma_init(void); /* create anon_vma_cachep */
149int __anon_vma_prepare(struct vm_area_struct *);
150void unlink_anon_vmas(struct vm_area_struct *);
151int anon_vma_clone(struct vm_area_struct *, struct vm_area_struct *);
152int anon_vma_fork(struct vm_area_struct *, struct vm_area_struct *);
153
154static inline int anon_vma_prepare(struct vm_area_struct *vma)
155{
156 if (likely(vma->anon_vma))
157 return 0;
158
159 return __anon_vma_prepare(vma);
160}
161
162static inline void anon_vma_merge(struct vm_area_struct *vma,
163 struct vm_area_struct *next)
164{
165 VM_BUG_ON_VMA(vma->anon_vma != next->anon_vma, vma);
166 unlink_anon_vmas(next);
167}
168
169struct anon_vma *folio_get_anon_vma(struct folio *folio);
170
171/* RMAP flags, currently only relevant for some anon rmap operations. */
172typedef int __bitwise rmap_t;
173
174/*
175 * No special request: if the page is a subpage of a compound page, it is
176 * mapped via a PTE. The mapped (sub)page is possibly shared between processes.
177 */
178#define RMAP_NONE ((__force rmap_t)0)
179
180/* The (sub)page is exclusive to a single process. */
181#define RMAP_EXCLUSIVE ((__force rmap_t)BIT(0))
182
183/*
184 * The compound page is not mapped via PTEs, but instead via a single PMD and
185 * should be accounted accordingly.
186 */
187#define RMAP_COMPOUND ((__force rmap_t)BIT(1))
188
189/*
190 * rmap interfaces called when adding or removing pte of page
191 */
192void page_move_anon_rmap(struct page *, struct vm_area_struct *);
193void page_add_anon_rmap(struct page *, struct vm_area_struct *,
194 unsigned long address, rmap_t flags);
195void page_add_new_anon_rmap(struct page *, struct vm_area_struct *,
196 unsigned long address);
197void page_add_file_rmap(struct page *, struct vm_area_struct *,
198 bool compound);
199void page_remove_rmap(struct page *, struct vm_area_struct *,
200 bool compound);
201
202void hugepage_add_anon_rmap(struct page *, struct vm_area_struct *,
203 unsigned long address, rmap_t flags);
204void hugepage_add_new_anon_rmap(struct page *, struct vm_area_struct *,
205 unsigned long address);
206
207static inline void __page_dup_rmap(struct page *page, bool compound)
208{
209 atomic_inc(compound ? compound_mapcount_ptr(page) : &page->_mapcount);
210}
211
212static inline void page_dup_file_rmap(struct page *page, bool compound)
213{
214 __page_dup_rmap(page, compound);
215}
216
217/**
218 * page_try_dup_anon_rmap - try duplicating a mapping of an already mapped
219 * anonymous page
220 * @page: the page to duplicate the mapping for
221 * @compound: the page is mapped as compound or as a small page
222 * @vma: the source vma
223 *
224 * The caller needs to hold the PT lock and the vma->vma_mm->write_protect_seq.
225 *
226 * Duplicating the mapping can only fail if the page may be pinned; device
227 * private pages cannot get pinned and consequently this function cannot fail.
228 *
229 * If duplicating the mapping succeeds, the page has to be mapped R/O into
230 * the parent and the child. It must *not* get mapped writable after this call.
231 *
232 * Returns 0 if duplicating the mapping succeeded. Returns -EBUSY otherwise.
233 */
234static inline int page_try_dup_anon_rmap(struct page *page, bool compound,
235 struct vm_area_struct *vma)
236{
237 VM_BUG_ON_PAGE(!PageAnon(page), page);
238
239 /*
240 * No need to check+clear for already shared pages, including KSM
241 * pages.
242 */
243 if (!PageAnonExclusive(page))
244 goto dup;
245
246 /*
247 * If this page may have been pinned by the parent process,
248 * don't allow to duplicate the mapping but instead require to e.g.,
249 * copy the page immediately for the child so that we'll always
250 * guarantee the pinned page won't be randomly replaced in the
251 * future on write faults.
252 */
253 if (likely(!is_device_private_page(page) &&
254 unlikely(page_needs_cow_for_dma(vma, page))))
255 return -EBUSY;
256
257 ClearPageAnonExclusive(page);
258 /*
259 * It's okay to share the anon page between both processes, mapping
260 * the page R/O into both processes.
261 */
262dup:
263 __page_dup_rmap(page, compound);
264 return 0;
265}
266
267/**
268 * page_try_share_anon_rmap - try marking an exclusive anonymous page possibly
269 * shared to prepare for KSM or temporary unmapping
270 * @page: the exclusive anonymous page to try marking possibly shared
271 *
272 * The caller needs to hold the PT lock and has to have the page table entry
273 * cleared/invalidated.
274 *
275 * This is similar to page_try_dup_anon_rmap(), however, not used during fork()
276 * to duplicate a mapping, but instead to prepare for KSM or temporarily
277 * unmapping a page (swap, migration) via page_remove_rmap().
278 *
279 * Marking the page shared can only fail if the page may be pinned; device
280 * private pages cannot get pinned and consequently this function cannot fail.
281 *
282 * Returns 0 if marking the page possibly shared succeeded. Returns -EBUSY
283 * otherwise.
284 */
285static inline int page_try_share_anon_rmap(struct page *page)
286{
287 VM_BUG_ON_PAGE(!PageAnon(page) || !PageAnonExclusive(page), page);
288
289 /* device private pages cannot get pinned via GUP. */
290 if (unlikely(is_device_private_page(page))) {
291 ClearPageAnonExclusive(page);
292 return 0;
293 }
294
295 /*
296 * We have to make sure that when we clear PageAnonExclusive, that
297 * the page is not pinned and that concurrent GUP-fast won't succeed in
298 * concurrently pinning the page.
299 *
300 * Conceptually, PageAnonExclusive clearing consists of:
301 * (A1) Clear PTE
302 * (A2) Check if the page is pinned; back off if so.
303 * (A3) Clear PageAnonExclusive
304 * (A4) Restore PTE (optional, but certainly not writable)
305 *
306 * When clearing PageAnonExclusive, we cannot possibly map the page
307 * writable again, because anon pages that may be shared must never
308 * be writable. So in any case, if the PTE was writable it cannot
309 * be writable anymore afterwards and there would be a PTE change. Only
310 * if the PTE wasn't writable, there might not be a PTE change.
311 *
312 * Conceptually, GUP-fast pinning of an anon page consists of:
313 * (B1) Read the PTE
314 * (B2) FOLL_WRITE: check if the PTE is not writable; back off if so.
315 * (B3) Pin the mapped page
316 * (B4) Check if the PTE changed by re-reading it; back off if so.
317 * (B5) If the original PTE is not writable, check if
318 * PageAnonExclusive is not set; back off if so.
319 *
320 * If the PTE was writable, we only have to make sure that GUP-fast
321 * observes a PTE change and properly backs off.
322 *
323 * If the PTE was not writable, we have to make sure that GUP-fast either
324 * detects a (temporary) PTE change or that PageAnonExclusive is cleared
325 * and properly backs off.
326 *
327 * Consequently, when clearing PageAnonExclusive(), we have to make
328 * sure that (A1), (A2)/(A3) and (A4) happen in the right memory
329 * order. In GUP-fast pinning code, we have to make sure that (B3),(B4)
330 * and (B5) happen in the right memory order.
331 *
332 * We assume that there might not be a memory barrier after
333 * clearing/invalidating the PTE (A1) and before restoring the PTE (A4),
334 * so we use explicit ones here.
335 */
336
337 /* Paired with the memory barrier in try_grab_folio(). */
338 if (IS_ENABLED(CONFIG_HAVE_FAST_GUP))
339 smp_mb();
340
341 if (unlikely(page_maybe_dma_pinned(page)))
342 return -EBUSY;
343 ClearPageAnonExclusive(page);
344
345 /*
346 * This is conceptually a smp_wmb() paired with the smp_rmb() in
347 * gup_must_unshare().
348 */
349 if (IS_ENABLED(CONFIG_HAVE_FAST_GUP))
350 smp_mb__after_atomic();
351 return 0;
352}
353
354/*
355 * Called from mm/vmscan.c to handle paging out
356 */
357int folio_referenced(struct folio *, int is_locked,
358 struct mem_cgroup *memcg, unsigned long *vm_flags);
359
360void try_to_migrate(struct folio *folio, enum ttu_flags flags);
361void try_to_unmap(struct folio *, enum ttu_flags flags);
362
363int make_device_exclusive_range(struct mm_struct *mm, unsigned long start,
364 unsigned long end, struct page **pages,
365 void *arg);
366
367/* Avoid racy checks */
368#define PVMW_SYNC (1 << 0)
369/* Look for migration entries rather than present PTEs */
370#define PVMW_MIGRATION (1 << 1)
371
372struct page_vma_mapped_walk {
373 unsigned long pfn;
374 unsigned long nr_pages;
375 pgoff_t pgoff;
376 struct vm_area_struct *vma;
377 unsigned long address;
378 pmd_t *pmd;
379 pte_t *pte;
380 spinlock_t *ptl;
381 unsigned int flags;
382};
383
384#define DEFINE_PAGE_VMA_WALK(name, _page, _vma, _address, _flags) \
385 struct page_vma_mapped_walk name = { \
386 .pfn = page_to_pfn(_page), \
387 .nr_pages = compound_nr(_page), \
388 .pgoff = page_to_pgoff(_page), \
389 .vma = _vma, \
390 .address = _address, \
391 .flags = _flags, \
392 }
393
394#define DEFINE_FOLIO_VMA_WALK(name, _folio, _vma, _address, _flags) \
395 struct page_vma_mapped_walk name = { \
396 .pfn = folio_pfn(_folio), \
397 .nr_pages = folio_nr_pages(_folio), \
398 .pgoff = folio_pgoff(_folio), \
399 .vma = _vma, \
400 .address = _address, \
401 .flags = _flags, \
402 }
403
404static inline void page_vma_mapped_walk_done(struct page_vma_mapped_walk *pvmw)
405{
406 /* HugeTLB pte is set to the relevant page table entry without pte_mapped. */
407 if (pvmw->pte && !is_vm_hugetlb_page(pvmw->vma))
408 pte_unmap(pvmw->pte);
409 if (pvmw->ptl)
410 spin_unlock(pvmw->ptl);
411}
412
413bool page_vma_mapped_walk(struct page_vma_mapped_walk *pvmw);
414
415/*
416 * Used by swapoff to help locate where page is expected in vma.
417 */
418unsigned long page_address_in_vma(struct page *, struct vm_area_struct *);
419
420/*
421 * Cleans the PTEs of shared mappings.
422 * (and since clean PTEs should also be readonly, write protects them too)
423 *
424 * returns the number of cleaned PTEs.
425 */
426int folio_mkclean(struct folio *);
427
428int pfn_mkclean_range(unsigned long pfn, unsigned long nr_pages, pgoff_t pgoff,
429 struct vm_area_struct *vma);
430
431void remove_migration_ptes(struct folio *src, struct folio *dst, bool locked);
432
433int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma);
434
435/*
436 * rmap_walk_control: To control rmap traversing for specific needs
437 *
438 * arg: passed to rmap_one() and invalid_vma()
439 * try_lock: bail out if the rmap lock is contended
440 * contended: indicate the rmap traversal bailed out due to lock contention
441 * rmap_one: executed on each vma where page is mapped
442 * done: for checking traversing termination condition
443 * anon_lock: for getting anon_lock by optimized way rather than default
444 * invalid_vma: for skipping uninterested vma
445 */
446struct rmap_walk_control {
447 void *arg;
448 bool try_lock;
449 bool contended;
450 /*
451 * Return false if page table scanning in rmap_walk should be stopped.
452 * Otherwise, return true.
453 */
454 bool (*rmap_one)(struct folio *folio, struct vm_area_struct *vma,
455 unsigned long addr, void *arg);
456 int (*done)(struct folio *folio);
457 struct anon_vma *(*anon_lock)(struct folio *folio,
458 struct rmap_walk_control *rwc);
459 bool (*invalid_vma)(struct vm_area_struct *vma, void *arg);
460};
461
462void rmap_walk(struct folio *folio, struct rmap_walk_control *rwc);
463void rmap_walk_locked(struct folio *folio, struct rmap_walk_control *rwc);
464struct anon_vma *folio_lock_anon_vma_read(struct folio *folio,
465 struct rmap_walk_control *rwc);
466
467#else /* !CONFIG_MMU */
468
469#define anon_vma_init() do {} while (0)
470#define anon_vma_prepare(vma) (0)
471#define anon_vma_link(vma) do {} while (0)
472
473static inline int folio_referenced(struct folio *folio, int is_locked,
474 struct mem_cgroup *memcg,
475 unsigned long *vm_flags)
476{
477 *vm_flags = 0;
478 return 0;
479}
480
481static inline void try_to_unmap(struct folio *folio, enum ttu_flags flags)
482{
483}
484
485static inline int folio_mkclean(struct folio *folio)
486{
487 return 0;
488}
489#endif /* CONFIG_MMU */
490
491static inline int page_mkclean(struct page *page)
492{
493 return folio_mkclean(page_folio(page));
494}
495#endif /* _LINUX_RMAP_H */