xref: /linux/include/linux/swap.h (revision 49bda4826843be0ef97a162009a29ea3a63f3935)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_SWAP_H
3 #define _LINUX_SWAP_H
4 
5 #include <linux/spinlock.h>
6 #include <linux/linkage.h>
7 #include <linux/mmzone.h>
8 #include <linux/list.h>
9 #include <linux/memcontrol.h>
10 #include <linux/sched.h>
11 #include <linux/node.h>
12 #include <linux/fs.h>
13 #include <linux/pagemap.h>
14 #include <linux/atomic.h>
15 #include <linux/page-flags.h>
16 #include <uapi/linux/mempolicy.h>
17 #include <asm/page.h>
18 
19 #define SWAP_FLAG_PREFER	0x8000	/* set if swap priority specified */
20 #define SWAP_FLAG_PRIO_MASK	0x7fff
21 #define SWAP_FLAG_DISCARD	0x10000 /* enable discard for swap */
22 #define SWAP_FLAG_DISCARD_ONCE	0x20000 /* discard swap area at swapon-time */
23 #define SWAP_FLAG_DISCARD_PAGES 0x40000 /* discard page-clusters after use */
24 
25 #define SWAP_FLAGS_VALID	(SWAP_FLAG_PRIO_MASK | SWAP_FLAG_PREFER | \
26 				 SWAP_FLAG_DISCARD | SWAP_FLAG_DISCARD_ONCE | \
27 				 SWAP_FLAG_DISCARD_PAGES)
28 
current_is_kswapd(void)29 static inline int current_is_kswapd(void)
30 {
31 	return current->flags & PF_KSWAPD;
32 }
33 
34 /*
35  * MAX_SWAPFILES defines the maximum number of swaptypes: things which can
36  * be swapped to.  The swap type and the offset into that swap type are
37  * encoded into pte's and into pgoff_t's in the swapcache.  Using five bits
38  * for the type means that the maximum number of swapcache pages is 27 bits
39  * on 32-bit-pgoff_t architectures.  And that assumes that the architecture packs
40  * the type/offset into the pte as 5/27 as well.
41  */
42 #define MAX_SWAPFILES_SHIFT	5
43 
44 /*
45  * Use some of the swap files numbers for other purposes. This
46  * is a convenient way to hook into the VM to trigger special
47  * actions on faults.
48  */
49 
50 /*
51  * PTE markers are used to persist information onto PTEs that otherwise
52  * should be a none pte.  As its name "PTE" hints, it should only be
53  * applied to the leaves of pgtables.
54  */
55 #define SWP_PTE_MARKER_NUM 1
56 #define SWP_PTE_MARKER     (MAX_SWAPFILES + SWP_HWPOISON_NUM + \
57 			    SWP_MIGRATION_NUM + SWP_DEVICE_NUM)
58 
59 /*
60  * Unaddressable device memory support. See include/linux/hmm.h and
61  * Documentation/mm/hmm.rst. Short description is we need struct pages for
62  * device memory that is unaddressable (inaccessible) by CPU, so that we can
63  * migrate part of a process memory to device memory.
64  *
65  * When a page is migrated from CPU to device, we set the CPU page table entry
66  * to a special SWP_DEVICE_{READ|WRITE} entry.
67  *
68  * When a page is mapped by the device for exclusive access we set the CPU page
69  * table entries to a special SWP_DEVICE_EXCLUSIVE entry.
70  */
71 #ifdef CONFIG_DEVICE_PRIVATE
72 #define SWP_DEVICE_NUM 3
73 #define SWP_DEVICE_WRITE (MAX_SWAPFILES+SWP_HWPOISON_NUM+SWP_MIGRATION_NUM)
74 #define SWP_DEVICE_READ (MAX_SWAPFILES+SWP_HWPOISON_NUM+SWP_MIGRATION_NUM+1)
75 #define SWP_DEVICE_EXCLUSIVE (MAX_SWAPFILES+SWP_HWPOISON_NUM+SWP_MIGRATION_NUM+2)
76 #else
77 #define SWP_DEVICE_NUM 0
78 #endif
79 
80 /*
81  * Page migration support.
82  *
83  * SWP_MIGRATION_READ_EXCLUSIVE is only applicable to anonymous pages and
84  * indicates that the referenced (part of) an anonymous page is exclusive to
85  * a single process. For SWP_MIGRATION_WRITE, that information is implicit:
86  * (part of) an anonymous page that are mapped writable are exclusive to a
87  * single process.
88  */
89 #ifdef CONFIG_MIGRATION
90 #define SWP_MIGRATION_NUM 3
91 #define SWP_MIGRATION_READ (MAX_SWAPFILES + SWP_HWPOISON_NUM)
92 #define SWP_MIGRATION_READ_EXCLUSIVE (MAX_SWAPFILES + SWP_HWPOISON_NUM + 1)
93 #define SWP_MIGRATION_WRITE (MAX_SWAPFILES + SWP_HWPOISON_NUM + 2)
94 #else
95 #define SWP_MIGRATION_NUM 0
96 #endif
97 
98 /*
99  * Handling of hardware poisoned pages with memory corruption.
100  */
101 #ifdef CONFIG_MEMORY_FAILURE
102 #define SWP_HWPOISON_NUM 1
103 #define SWP_HWPOISON		MAX_SWAPFILES
104 #else
105 #define SWP_HWPOISON_NUM 0
106 #endif
107 
108 #define MAX_SWAPFILES \
109 	((1 << MAX_SWAPFILES_SHIFT) - SWP_DEVICE_NUM - \
110 	SWP_MIGRATION_NUM - SWP_HWPOISON_NUM - \
111 	SWP_PTE_MARKER_NUM)
112 
113 /*
114  * Magic header for a swap area. The first part of the union is
115  * what the swap magic looks like for the old (limited to 128MB)
116  * swap area format, the second part of the union adds - in the
117  * old reserved area - some extra information. Note that the first
118  * kilobyte is reserved for boot loader or disk label stuff...
119  *
120  * Having the magic at the end of the PAGE_SIZE makes detecting swap
121  * areas somewhat tricky on machines that support multiple page sizes.
122  * For 2.5 we'll probably want to move the magic to just beyond the
123  * bootbits...
124  */
125 union swap_header {
126 	struct {
127 		char reserved[PAGE_SIZE - 10];
128 		char magic[10];			/* SWAP-SPACE or SWAPSPACE2 */
129 	} magic;
130 	struct {
131 		char		bootbits[1024];	/* Space for disklabel etc. */
132 		__u32		version;
133 		__u32		last_page;
134 		__u32		nr_badpages;
135 		unsigned char	sws_uuid[16];
136 		unsigned char	sws_volume[16];
137 		__u32		padding[117];
138 		__u32		badpages[1];
139 	} info;
140 };
141 
142 /*
143  * current->reclaim_state points to one of these when a task is running
144  * memory reclaim
145  */
146 struct reclaim_state {
147 	/* pages reclaimed outside of LRU-based reclaim */
148 	unsigned long reclaimed;
149 #ifdef CONFIG_LRU_GEN
150 	/* per-thread mm walk data */
151 	struct lru_gen_mm_walk *mm_walk;
152 #endif
153 };
154 
155 /*
156  * mm_account_reclaimed_pages(): account reclaimed pages outside of LRU-based
157  * reclaim
158  * @pages: number of pages reclaimed
159  *
160  * If the current process is undergoing a reclaim operation, increment the
161  * number of reclaimed pages by @pages.
162  */
mm_account_reclaimed_pages(unsigned long pages)163 static inline void mm_account_reclaimed_pages(unsigned long pages)
164 {
165 	if (current->reclaim_state)
166 		current->reclaim_state->reclaimed += pages;
167 }
168 
169 #ifdef __KERNEL__
170 
171 struct address_space;
172 struct sysinfo;
173 struct zone;
174 
175 /*
176  * A swap extent maps a range of a swapfile's PAGE_SIZE pages onto a range of
177  * disk blocks.  A rbtree of swap extents maps the entire swapfile (Where the
178  * term `swapfile' refers to either a blockdevice or an IS_REG file). Apart
179  * from setup, they're handled identically.
180  *
181  * We always assume that blocks are of size PAGE_SIZE.
182  */
183 struct swap_extent {
184 	struct rb_node rb_node;
185 	pgoff_t start_page;
186 	pgoff_t nr_pages;
187 	sector_t start_block;
188 };
189 
190 /*
191  * Max bad pages in the new format..
192  */
193 #define MAX_SWAP_BADPAGES \
194 	((offsetof(union swap_header, magic.magic) - \
195 	  offsetof(union swap_header, info.badpages)) / sizeof(int))
196 
197 enum {
198 	SWP_USED	= (1 << 0),	/* is slot in swap_info[] used? */
199 	SWP_WRITEOK	= (1 << 1),	/* ok to write to this swap?	*/
200 	SWP_DISCARDABLE = (1 << 2),	/* blkdev support discard */
201 	SWP_DISCARDING	= (1 << 3),	/* now discarding a free cluster */
202 	SWP_SOLIDSTATE	= (1 << 4),	/* blkdev seeks are cheap */
203 	SWP_BLKDEV	= (1 << 6),	/* its a block device */
204 	SWP_ACTIVATED	= (1 << 7),	/* set after swap_activate success */
205 	SWP_AREA_DISCARD = (1 << 9),	/* single-time swap area discards */
206 	SWP_PAGE_DISCARD = (1 << 10),	/* freed swap page-cluster discards */
207 	SWP_STABLE_WRITES = (1 << 11),	/* no overwrite PG_writeback pages */
208 	SWP_SYNCHRONOUS_IO = (1 << 12),	/* synchronous IO is efficient */
209 	SWP_HIBERNATION = (1 << 13),	/* pinned for hibernation */
210 					/* add others here before... */
211 };
212 
213 #define SWAP_CLUSTER_MAX 32UL
214 #define SWAP_CLUSTER_MAX_SKIPPED (SWAP_CLUSTER_MAX << 10)
215 #define COMPACT_CLUSTER_MAX SWAP_CLUSTER_MAX
216 
217 /*
218  * The first page in the swap file is the swap header, which is always marked
219  * bad to prevent it from being allocated as an entry. This also prevents the
220  * cluster to which it belongs being marked free. Therefore 0 is safe to use as
221  * a sentinel to indicate an entry is not valid.
222  */
223 #define SWAP_ENTRY_INVALID	0
224 
225 #ifdef CONFIG_THP_SWAP
226 #define SWAP_NR_ORDERS		(PMD_ORDER + 1)
227 #else
228 #define SWAP_NR_ORDERS		1
229 #endif
230 
231 /*
232  * We keep using same cluster for rotational device so IO will be sequential.
233  * The purpose is to optimize SWAP throughput on these device.
234  */
235 struct swap_sequential_cluster {
236 	unsigned int next[SWAP_NR_ORDERS]; /* Likely next allocation offset */
237 };
238 
239 /*
240  * The in-memory structure used to track swap areas.
241  */
242 struct swap_info_struct {
243 	struct percpu_ref users;	/* indicate and keep swap device valid. */
244 	unsigned long	flags;		/* SWP_USED etc: see above */
245 	signed short	prio;		/* swap priority of this type */
246 	struct plist_node list;		/* entry in swap_active_head */
247 	signed char	type;		/* strange name for an index */
248 	unsigned int	max;		/* size of this swap device */
249 	struct swap_cluster_info *cluster_info; /* cluster info. Only for SSD */
250 	struct list_head free_clusters; /* free clusters list */
251 	struct list_head full_clusters; /* full clusters list */
252 	struct list_head nonfull_clusters[SWAP_NR_ORDERS];
253 					/* list of cluster that contains at least one free slot */
254 	struct list_head frag_clusters[SWAP_NR_ORDERS];
255 					/* list of cluster that are fragmented or contented */
256 	unsigned int pages;		/* total of usable pages of swap */
257 	atomic_long_t inuse_pages;	/* number of those currently in use */
258 	struct swap_sequential_cluster *global_cluster; /* Use one global cluster for rotating device */
259 	spinlock_t global_cluster_lock;	/* Serialize usage of global cluster */
260 	struct rb_root swap_extent_root;/* root of the swap extent rbtree */
261 	struct block_device *bdev;	/* swap device or bdev of swap file */
262 	struct file *swap_file;		/* seldom referenced */
263 	struct completion comp;		/* seldom referenced */
264 	spinlock_t lock;		/*
265 					 * protect map scan related fields like
266 					 * inuse_pages and all cluster lists.
267 					 * Other fields are only changed
268 					 * at swapon/swapoff, so are protected
269 					 * by swap_lock. changing flags need
270 					 * hold this lock and swap_lock. If
271 					 * both locks need hold, hold swap_lock
272 					 * first.
273 					 */
274 	struct work_struct discard_work; /* discard worker */
275 	struct work_struct reclaim_work; /* reclaim worker */
276 	struct list_head discard_clusters; /* discard clusters list */
277 	struct plist_node avail_list;   /* entry in swap_avail_head */
278 	const struct swap_ops *ops;
279 };
280 
page_swap_entry(struct page * page)281 static inline swp_entry_t page_swap_entry(struct page *page)
282 {
283 	struct folio *folio = page_folio(page);
284 	swp_entry_t entry = folio->swap;
285 
286 	entry.val += folio_page_idx(folio, page);
287 	return entry;
288 }
289 
290 /* linux/mm/page_alloc.c */
291 extern unsigned long totalreserve_pages;
292 
293 /* Definition of global_zone_page_state not available yet */
294 #define nr_free_pages() global_zone_page_state(NR_FREE_PAGES)
295 
296 /* linux/mm/folio.c */
297 void folio_add_lru(struct folio *folio);
298 void folio_mark_accessed(struct folio *folio);
299 void lru_add_drain_all(void);
300 
301 enum lru_cache_drained {
302 	LRU_CACHE_NOT_DRAINED,
303 	LRU_CACHE_DRAINED,
304 	LRU_CACHE_DRAINED_ALL,
305 };
306 void lru_cache_drain_for_folio(const struct folio *folio,
307 		unsigned int extra_refs, enum lru_cache_drained *drained);
308 
309 /* linux/mm/folio-compat.c */
310 void mark_page_accessed(struct page *page);
311 
312 extern atomic_t lru_disable_count;
313 
lru_cache_disabled(void)314 static inline bool lru_cache_disabled(void)
315 {
316 	return atomic_read(&lru_disable_count);
317 }
318 
319 extern unsigned long shrink_all_memory(unsigned long nr_pages);
320 long remove_mapping(struct address_space *mapping, struct folio *folio);
321 
322 #if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA)
323 extern int reclaim_register_node(struct node *node);
324 extern void reclaim_unregister_node(struct node *node);
325 
326 #else
327 
reclaim_register_node(struct node * node)328 static inline int reclaim_register_node(struct node *node)
329 {
330 	return 0;
331 }
332 
reclaim_unregister_node(struct node * node)333 static inline void reclaim_unregister_node(struct node *node)
334 {
335 }
336 #endif /* CONFIG_SYSFS && CONFIG_NUMA */
337 
338 void check_move_unevictable_folios(struct folio_batch *fbatch);
339 
340 extern void __meminit kswapd_run(int nid);
341 extern void __meminit kswapd_stop(int nid);
342 
343 #ifdef CONFIG_SWAP
344 int add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
345 		unsigned long nr_pages, sector_t start_block);
346 int generic_swapfile_activate(struct swap_info_struct *, struct file *,
347 		sector_t *);
348 
total_swapcache_pages(void)349 static inline unsigned long total_swapcache_pages(void)
350 {
351 	return global_node_page_state(NR_SWAPCACHE);
352 }
353 
354 void free_swap_cache(struct folio *folio);
355 void free_folio_and_swap_cache(struct folio *folio);
356 void free_pages_and_swap_cache(struct encoded_page **, int);
357 /* linux/mm/swapfile.c */
358 extern atomic_long_t nr_swap_pages;
359 extern long total_swap_pages;
360 extern atomic_t nr_rotate_swap;
361 
362 /* Swap 50% full? Release swapcache more aggressively.. */
vm_swap_full(void)363 static inline bool vm_swap_full(void)
364 {
365 	return atomic_long_read(&nr_swap_pages) * 2 < total_swap_pages;
366 }
367 
get_nr_swap_pages(void)368 static inline long get_nr_swap_pages(void)
369 {
370 	return atomic_long_read(&nr_swap_pages);
371 }
372 
373 extern void si_swapinfo(struct sysinfo *);
374 extern int pin_hibernation_swap_type(dev_t device, sector_t offset);
375 extern void unpin_hibernation_swap_type(int type);
376 extern int find_hibernation_swap_type(dev_t device, sector_t offset);
377 int find_first_swap(dev_t *device);
378 extern unsigned int count_swap_pages(int, int);
379 extern sector_t swapdev_block(int, pgoff_t);
380 extern int __swap_count(swp_entry_t entry);
381 extern bool swap_entry_swapped(struct swap_info_struct *si, swp_entry_t entry);
382 extern int swp_swapcount(swp_entry_t entry);
383 extern struct swap_info_struct *get_swap_device(swp_entry_t entry);
384 sector_t swap_folio_sector(struct folio *folio);
385 
386 /*
387  * If there is an existing swap slot reference (swap entry) and the caller
388  * guarantees that there is no race modification of it (e.g., PTL
389  * protecting the swap entry in page table; shmem's cmpxchg protects t
390  * he swap entry in shmem mapping), these two helpers below can be used
391  * to put/dup the entries directly.
392  *
393  * All entries must be allocated by folio_alloc_swap(). And they must have
394  * a swap count > 1. See comments of folio_*_swap helpers for more info.
395  */
396 int swap_dup_entry_direct(swp_entry_t entry);
397 void swap_put_entries_direct(swp_entry_t entry, int nr);
398 
399 /*
400  * folio_free_swap tries to free the swap entries pinned by a swap cache
401  * folio, it has to be here to be called by other components.
402  */
403 bool folio_free_swap(struct folio *folio);
404 
405 /* Allocate / free (hibernation) exclusive entries */
406 swp_entry_t swap_alloc_hibernation_slot(int type);
407 void swap_free_hibernation_slot(swp_entry_t entry);
408 
put_swap_device(struct swap_info_struct * si)409 static inline void put_swap_device(struct swap_info_struct *si)
410 {
411 	percpu_ref_put(&si->users);
412 }
413 
414 #else /* CONFIG_SWAP */
get_swap_device(swp_entry_t entry)415 static inline struct swap_info_struct *get_swap_device(swp_entry_t entry)
416 {
417 	return NULL;
418 }
419 
put_swap_device(struct swap_info_struct * si)420 static inline void put_swap_device(struct swap_info_struct *si)
421 {
422 }
423 
424 #define get_nr_swap_pages()			0L
425 #define total_swap_pages			0L
426 #define total_swapcache_pages()			0UL
427 #define vm_swap_full()				0
428 
429 #define si_swapinfo(val) \
430 	do { (val)->freeswap = (val)->totalswap = 0; } while (0)
431 #define free_folio_and_swap_cache(folio) \
432 	folio_put(folio)
433 #define free_pages_and_swap_cache(pages, nr) \
434 	release_pages((pages), (nr));
435 
free_swap_cache(struct folio * folio)436 static inline void free_swap_cache(struct folio *folio)
437 {
438 }
439 
swap_dup_entry_direct(swp_entry_t ent)440 static inline int swap_dup_entry_direct(swp_entry_t ent)
441 {
442 	return 0;
443 }
444 
swap_put_entries_direct(swp_entry_t ent,int nr)445 static inline void swap_put_entries_direct(swp_entry_t ent, int nr)
446 {
447 }
448 
__swap_count(swp_entry_t entry)449 static inline int __swap_count(swp_entry_t entry)
450 {
451 	return 0;
452 }
453 
swap_entry_swapped(struct swap_info_struct * si,swp_entry_t entry)454 static inline bool swap_entry_swapped(struct swap_info_struct *si, swp_entry_t entry)
455 {
456 	return false;
457 }
458 
swp_swapcount(swp_entry_t entry)459 static inline int swp_swapcount(swp_entry_t entry)
460 {
461 	return 0;
462 }
463 
folio_free_swap(struct folio * folio)464 static inline bool folio_free_swap(struct folio *folio)
465 {
466 	return false;
467 }
468 
add_swap_extent(struct swap_info_struct * sis,unsigned long start_page,unsigned long nr_pages,sector_t start_block)469 static inline int add_swap_extent(struct swap_info_struct *sis,
470 				  unsigned long start_page,
471 				  unsigned long nr_pages, sector_t start_block)
472 {
473 	return -EINVAL;
474 }
475 #endif /* CONFIG_SWAP */
476 #ifdef CONFIG_MEMCG
477 void lru_reparent_memcg(struct mem_cgroup *memcg, struct mem_cgroup *parent, int nid);
478 #endif
479 
480 #if defined(CONFIG_SWAP) && defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
481 void __folio_throttle_swaprate(struct folio *folio, gfp_t gfp);
folio_throttle_swaprate(struct folio * folio,gfp_t gfp)482 static inline void folio_throttle_swaprate(struct folio *folio, gfp_t gfp)
483 {
484 	if (mem_cgroup_disabled())
485 		return;
486 	__folio_throttle_swaprate(folio, gfp);
487 }
488 #else
folio_throttle_swaprate(struct folio * folio,gfp_t gfp)489 static inline void folio_throttle_swaprate(struct folio *folio, gfp_t gfp)
490 {
491 }
492 #endif
493 
494 #if defined(CONFIG_MEMCG) && defined(CONFIG_SWAP)
495 int __mem_cgroup_try_charge_swap(struct folio *folio);
mem_cgroup_try_charge_swap(struct folio * folio)496 static inline int mem_cgroup_try_charge_swap(struct folio *folio)
497 {
498 	if (mem_cgroup_disabled())
499 		return 0;
500 	return __mem_cgroup_try_charge_swap(folio);
501 }
502 
503 extern void __mem_cgroup_uncharge_swap(unsigned short id, unsigned int nr_pages);
mem_cgroup_uncharge_swap(unsigned short id,unsigned int nr_pages)504 static inline void mem_cgroup_uncharge_swap(unsigned short id, unsigned int nr_pages)
505 {
506 	if (mem_cgroup_disabled())
507 		return;
508 	__mem_cgroup_uncharge_swap(id, nr_pages);
509 }
510 
511 extern long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg);
512 extern bool mem_cgroup_swap_full(struct folio *folio);
513 #else
mem_cgroup_try_charge_swap(struct folio * folio)514 static inline int mem_cgroup_try_charge_swap(struct folio *folio)
515 {
516 	return 0;
517 }
518 
mem_cgroup_uncharge_swap(unsigned short id,unsigned int nr_pages)519 static inline void mem_cgroup_uncharge_swap(unsigned short id,
520 					    unsigned int nr_pages)
521 {
522 }
523 
mem_cgroup_get_nr_swap_pages(struct mem_cgroup * memcg)524 static inline long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg)
525 {
526 	return get_nr_swap_pages();
527 }
528 
mem_cgroup_swap_full(struct folio * folio)529 static inline bool mem_cgroup_swap_full(struct folio *folio)
530 {
531 	return vm_swap_full();
532 }
533 #endif
534 
535 /* for_each_managed_zone_pgdat - helper macro to iterate over all managed zones in a pgdat up to
536  * and including the specified highidx
537  * @zone: The current zone in the iterator
538  * @pgdat: The pgdat which node_zones are being iterated
539  * @idx: The index variable
540  * @highidx: The index of the highest zone to return
541  *
542  * This macro iterates through all managed zones up to and including the specified highidx.
543  * The zone iterator enters an invalid state after macro call and must be reinitialized
544  * before it can be used again.
545  */
546 #define for_each_managed_zone_pgdat(zone, pgdat, idx, highidx)	\
547 	for ((idx) = 0, (zone) = (pgdat)->node_zones;		\
548 	    (idx) <= (highidx);					\
549 	    (idx)++, (zone)++)					\
550 		if (!managed_zone(zone))			\
551 			continue;				\
552 		else
553 
554 #endif /* __KERNEL__*/
555 #endif /* _LINUX_SWAP_H */
556